1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Linux Socket Filter - Kernel level socket filtering
4 *
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
7 *
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
9 *
10 * Authors:
11 *
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
15 *
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
18 */
19
20 #include <linux/atomic.h>
21 #include <linux/bpf_verifier.h>
22 #include <linux/module.h>
23 #include <linux/types.h>
24 #include <linux/mm.h>
25 #include <linux/fcntl.h>
26 #include <linux/socket.h>
27 #include <linux/sock_diag.h>
28 #include <linux/in.h>
29 #include <linux/inet.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_packet.h>
32 #include <linux/if_arp.h>
33 #include <linux/gfp.h>
34 #include <net/inet_common.h>
35 #include <net/ip.h>
36 #include <net/protocol.h>
37 #include <net/netlink.h>
38 #include <linux/skbuff.h>
39 #include <linux/skmsg.h>
40 #include <net/sock.h>
41 #include <net/flow_dissector.h>
42 #include <linux/errno.h>
43 #include <linux/timer.h>
44 #include <linux/uaccess.h>
45 #include <asm/unaligned.h>
46 #include <linux/filter.h>
47 #include <linux/ratelimit.h>
48 #include <linux/seccomp.h>
49 #include <linux/if_vlan.h>
50 #include <linux/bpf.h>
51 #include <linux/btf.h>
52 #include <net/sch_generic.h>
53 #include <net/cls_cgroup.h>
54 #include <net/dst_metadata.h>
55 #include <net/dst.h>
56 #include <net/sock_reuseport.h>
57 #include <net/busy_poll.h>
58 #include <net/tcp.h>
59 #include <net/xfrm.h>
60 #include <net/udp.h>
61 #include <linux/bpf_trace.h>
62 #include <net/xdp_sock.h>
63 #include <linux/inetdevice.h>
64 #include <net/inet_hashtables.h>
65 #include <net/inet6_hashtables.h>
66 #include <net/ip_fib.h>
67 #include <net/nexthop.h>
68 #include <net/flow.h>
69 #include <net/arp.h>
70 #include <net/ipv6.h>
71 #include <net/net_namespace.h>
72 #include <linux/seg6_local.h>
73 #include <net/seg6.h>
74 #include <net/seg6_local.h>
75 #include <net/lwtunnel.h>
76 #include <net/ipv6_stubs.h>
77 #include <net/bpf_sk_storage.h>
78 #include <net/transp_v6.h>
79 #include <linux/btf_ids.h>
80 #include <net/tls.h>
81 #include <net/xdp.h>
82 #include <net/mptcp.h>
83 #include <net/netfilter/nf_conntrack_bpf.h>
84 #include <net/netkit.h>
85 #include <linux/un.h>
86 #include <net/xdp_sock_drv.h>
87 #include <net/inet_dscp.h>
88
89 #include "dev.h"
90
91 /* Keep the struct bpf_fib_lookup small so that it fits into a cacheline */
92 static_assert(sizeof(struct bpf_fib_lookup) == 64, "struct bpf_fib_lookup size check");
93
94 static const struct bpf_func_proto *
95 bpf_sk_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog);
96
copy_bpf_fprog_from_user(struct sock_fprog * dst,sockptr_t src,int len)97 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
98 {
99 if (in_compat_syscall()) {
100 struct compat_sock_fprog f32;
101
102 if (len != sizeof(f32))
103 return -EINVAL;
104 if (copy_from_sockptr(&f32, src, sizeof(f32)))
105 return -EFAULT;
106 memset(dst, 0, sizeof(*dst));
107 dst->len = f32.len;
108 dst->filter = compat_ptr(f32.filter);
109 } else {
110 if (len != sizeof(*dst))
111 return -EINVAL;
112 if (copy_from_sockptr(dst, src, sizeof(*dst)))
113 return -EFAULT;
114 }
115
116 return 0;
117 }
118 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
119
120 /**
121 * sk_filter_trim_cap - run a packet through a socket filter
122 * @sk: sock associated with &sk_buff
123 * @skb: buffer to filter
124 * @cap: limit on how short the eBPF program may trim the packet
125 *
126 * Run the eBPF program and then cut skb->data to correct size returned by
127 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
128 * than pkt_len we keep whole skb->data. This is the socket level
129 * wrapper to bpf_prog_run. It returns 0 if the packet should
130 * be accepted or -EPERM if the packet should be tossed.
131 *
132 */
sk_filter_trim_cap(struct sock * sk,struct sk_buff * skb,unsigned int cap)133 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
134 {
135 int err;
136 struct sk_filter *filter;
137
138 /*
139 * If the skb was allocated from pfmemalloc reserves, only
140 * allow SOCK_MEMALLOC sockets to use it as this socket is
141 * helping free memory
142 */
143 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
144 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
145 return -ENOMEM;
146 }
147 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
148 if (err)
149 return err;
150
151 err = security_sock_rcv_skb(sk, skb);
152 if (err)
153 return err;
154
155 rcu_read_lock();
156 filter = rcu_dereference(sk->sk_filter);
157 if (filter) {
158 struct sock *save_sk = skb->sk;
159 unsigned int pkt_len;
160
161 skb->sk = sk;
162 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
163 skb->sk = save_sk;
164 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
165 }
166 rcu_read_unlock();
167
168 return err;
169 }
170 EXPORT_SYMBOL(sk_filter_trim_cap);
171
BPF_CALL_1(bpf_skb_get_pay_offset,struct sk_buff *,skb)172 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
173 {
174 return skb_get_poff(skb);
175 }
176
BPF_CALL_3(bpf_skb_get_nlattr,struct sk_buff *,skb,u32,a,u32,x)177 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
178 {
179 struct nlattr *nla;
180
181 if (skb_is_nonlinear(skb))
182 return 0;
183
184 if (skb->len < sizeof(struct nlattr))
185 return 0;
186
187 if (a > skb->len - sizeof(struct nlattr))
188 return 0;
189
190 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
191 if (nla)
192 return (void *) nla - (void *) skb->data;
193
194 return 0;
195 }
196
BPF_CALL_3(bpf_skb_get_nlattr_nest,struct sk_buff *,skb,u32,a,u32,x)197 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
198 {
199 struct nlattr *nla;
200
201 if (skb_is_nonlinear(skb))
202 return 0;
203
204 if (skb->len < sizeof(struct nlattr))
205 return 0;
206
207 if (a > skb->len - sizeof(struct nlattr))
208 return 0;
209
210 nla = (struct nlattr *) &skb->data[a];
211 if (!nla_ok(nla, skb->len - a))
212 return 0;
213
214 nla = nla_find_nested(nla, x);
215 if (nla)
216 return (void *) nla - (void *) skb->data;
217
218 return 0;
219 }
220
BPF_CALL_4(bpf_skb_load_helper_8,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)221 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
222 data, int, headlen, int, offset)
223 {
224 u8 tmp, *ptr;
225 const int len = sizeof(tmp);
226
227 if (offset >= 0) {
228 if (headlen - offset >= len)
229 return *(u8 *)(data + offset);
230 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
231 return tmp;
232 } else {
233 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
234 if (likely(ptr))
235 return *(u8 *)ptr;
236 }
237
238 return -EFAULT;
239 }
240
BPF_CALL_2(bpf_skb_load_helper_8_no_cache,const struct sk_buff *,skb,int,offset)241 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
242 int, offset)
243 {
244 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
245 offset);
246 }
247
BPF_CALL_4(bpf_skb_load_helper_16,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)248 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
249 data, int, headlen, int, offset)
250 {
251 __be16 tmp, *ptr;
252 const int len = sizeof(tmp);
253
254 if (offset >= 0) {
255 if (headlen - offset >= len)
256 return get_unaligned_be16(data + offset);
257 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
258 return be16_to_cpu(tmp);
259 } else {
260 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
261 if (likely(ptr))
262 return get_unaligned_be16(ptr);
263 }
264
265 return -EFAULT;
266 }
267
BPF_CALL_2(bpf_skb_load_helper_16_no_cache,const struct sk_buff *,skb,int,offset)268 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
269 int, offset)
270 {
271 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
272 offset);
273 }
274
BPF_CALL_4(bpf_skb_load_helper_32,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)275 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
276 data, int, headlen, int, offset)
277 {
278 __be32 tmp, *ptr;
279 const int len = sizeof(tmp);
280
281 if (likely(offset >= 0)) {
282 if (headlen - offset >= len)
283 return get_unaligned_be32(data + offset);
284 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
285 return be32_to_cpu(tmp);
286 } else {
287 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
288 if (likely(ptr))
289 return get_unaligned_be32(ptr);
290 }
291
292 return -EFAULT;
293 }
294
BPF_CALL_2(bpf_skb_load_helper_32_no_cache,const struct sk_buff *,skb,int,offset)295 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
296 int, offset)
297 {
298 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
299 offset);
300 }
301
convert_skb_access(int skb_field,int dst_reg,int src_reg,struct bpf_insn * insn_buf)302 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
303 struct bpf_insn *insn_buf)
304 {
305 struct bpf_insn *insn = insn_buf;
306
307 switch (skb_field) {
308 case SKF_AD_MARK:
309 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
310
311 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
312 offsetof(struct sk_buff, mark));
313 break;
314
315 case SKF_AD_PKTTYPE:
316 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET);
317 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
318 #ifdef __BIG_ENDIAN_BITFIELD
319 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
320 #endif
321 break;
322
323 case SKF_AD_QUEUE:
324 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
325
326 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
327 offsetof(struct sk_buff, queue_mapping));
328 break;
329
330 case SKF_AD_VLAN_TAG:
331 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
332
333 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
334 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
335 offsetof(struct sk_buff, vlan_tci));
336 break;
337 case SKF_AD_VLAN_TAG_PRESENT:
338 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_all) != 4);
339 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
340 offsetof(struct sk_buff, vlan_all));
341 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
342 *insn++ = BPF_ALU32_IMM(BPF_MOV, dst_reg, 1);
343 break;
344 }
345
346 return insn - insn_buf;
347 }
348
convert_bpf_extensions(struct sock_filter * fp,struct bpf_insn ** insnp)349 static bool convert_bpf_extensions(struct sock_filter *fp,
350 struct bpf_insn **insnp)
351 {
352 struct bpf_insn *insn = *insnp;
353 u32 cnt;
354
355 switch (fp->k) {
356 case SKF_AD_OFF + SKF_AD_PROTOCOL:
357 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
358
359 /* A = *(u16 *) (CTX + offsetof(protocol)) */
360 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
361 offsetof(struct sk_buff, protocol));
362 /* A = ntohs(A) [emitting a nop or swap16] */
363 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
364 break;
365
366 case SKF_AD_OFF + SKF_AD_PKTTYPE:
367 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
368 insn += cnt - 1;
369 break;
370
371 case SKF_AD_OFF + SKF_AD_IFINDEX:
372 case SKF_AD_OFF + SKF_AD_HATYPE:
373 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
374 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
375
376 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
377 BPF_REG_TMP, BPF_REG_CTX,
378 offsetof(struct sk_buff, dev));
379 /* if (tmp != 0) goto pc + 1 */
380 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
381 *insn++ = BPF_EXIT_INSN();
382 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
383 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
384 offsetof(struct net_device, ifindex));
385 else
386 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
387 offsetof(struct net_device, type));
388 break;
389
390 case SKF_AD_OFF + SKF_AD_MARK:
391 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
392 insn += cnt - 1;
393 break;
394
395 case SKF_AD_OFF + SKF_AD_RXHASH:
396 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
397
398 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
399 offsetof(struct sk_buff, hash));
400 break;
401
402 case SKF_AD_OFF + SKF_AD_QUEUE:
403 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
404 insn += cnt - 1;
405 break;
406
407 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
408 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
409 BPF_REG_A, BPF_REG_CTX, insn);
410 insn += cnt - 1;
411 break;
412
413 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
414 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
415 BPF_REG_A, BPF_REG_CTX, insn);
416 insn += cnt - 1;
417 break;
418
419 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
420 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
421
422 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
423 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
424 offsetof(struct sk_buff, vlan_proto));
425 /* A = ntohs(A) [emitting a nop or swap16] */
426 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
427 break;
428
429 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
430 case SKF_AD_OFF + SKF_AD_NLATTR:
431 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
432 case SKF_AD_OFF + SKF_AD_CPU:
433 case SKF_AD_OFF + SKF_AD_RANDOM:
434 /* arg1 = CTX */
435 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
436 /* arg2 = A */
437 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
438 /* arg3 = X */
439 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
440 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
441 switch (fp->k) {
442 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
443 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
444 break;
445 case SKF_AD_OFF + SKF_AD_NLATTR:
446 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
447 break;
448 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
449 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
450 break;
451 case SKF_AD_OFF + SKF_AD_CPU:
452 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
453 break;
454 case SKF_AD_OFF + SKF_AD_RANDOM:
455 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
456 bpf_user_rnd_init_once();
457 break;
458 }
459 break;
460
461 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
462 /* A ^= X */
463 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
464 break;
465
466 default:
467 /* This is just a dummy call to avoid letting the compiler
468 * evict __bpf_call_base() as an optimization. Placed here
469 * where no-one bothers.
470 */
471 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
472 return false;
473 }
474
475 *insnp = insn;
476 return true;
477 }
478
convert_bpf_ld_abs(struct sock_filter * fp,struct bpf_insn ** insnp)479 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
480 {
481 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
482 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
483 bool endian = BPF_SIZE(fp->code) == BPF_H ||
484 BPF_SIZE(fp->code) == BPF_W;
485 bool indirect = BPF_MODE(fp->code) == BPF_IND;
486 const int ip_align = NET_IP_ALIGN;
487 struct bpf_insn *insn = *insnp;
488 int offset = fp->k;
489
490 if (!indirect &&
491 ((unaligned_ok && offset >= 0) ||
492 (!unaligned_ok && offset >= 0 &&
493 offset + ip_align >= 0 &&
494 offset + ip_align % size == 0))) {
495 bool ldx_off_ok = offset <= S16_MAX;
496
497 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
498 if (offset)
499 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
500 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
501 size, 2 + endian + (!ldx_off_ok * 2));
502 if (ldx_off_ok) {
503 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
504 BPF_REG_D, offset);
505 } else {
506 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
507 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
508 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
509 BPF_REG_TMP, 0);
510 }
511 if (endian)
512 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
513 *insn++ = BPF_JMP_A(8);
514 }
515
516 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
517 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
518 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
519 if (!indirect) {
520 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
521 } else {
522 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
523 if (fp->k)
524 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
525 }
526
527 switch (BPF_SIZE(fp->code)) {
528 case BPF_B:
529 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
530 break;
531 case BPF_H:
532 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
533 break;
534 case BPF_W:
535 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
536 break;
537 default:
538 return false;
539 }
540
541 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
542 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
543 *insn = BPF_EXIT_INSN();
544
545 *insnp = insn;
546 return true;
547 }
548
549 /**
550 * bpf_convert_filter - convert filter program
551 * @prog: the user passed filter program
552 * @len: the length of the user passed filter program
553 * @new_prog: allocated 'struct bpf_prog' or NULL
554 * @new_len: pointer to store length of converted program
555 * @seen_ld_abs: bool whether we've seen ld_abs/ind
556 *
557 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
558 * style extended BPF (eBPF).
559 * Conversion workflow:
560 *
561 * 1) First pass for calculating the new program length:
562 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
563 *
564 * 2) 2nd pass to remap in two passes: 1st pass finds new
565 * jump offsets, 2nd pass remapping:
566 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
567 */
bpf_convert_filter(struct sock_filter * prog,int len,struct bpf_prog * new_prog,int * new_len,bool * seen_ld_abs)568 static int bpf_convert_filter(struct sock_filter *prog, int len,
569 struct bpf_prog *new_prog, int *new_len,
570 bool *seen_ld_abs)
571 {
572 int new_flen = 0, pass = 0, target, i, stack_off;
573 struct bpf_insn *new_insn, *first_insn = NULL;
574 struct sock_filter *fp;
575 int *addrs = NULL;
576 u8 bpf_src;
577
578 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
579 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
580
581 if (len <= 0 || len > BPF_MAXINSNS)
582 return -EINVAL;
583
584 if (new_prog) {
585 first_insn = new_prog->insnsi;
586 addrs = kcalloc(len, sizeof(*addrs),
587 GFP_KERNEL | __GFP_NOWARN);
588 if (!addrs)
589 return -ENOMEM;
590 }
591
592 do_pass:
593 new_insn = first_insn;
594 fp = prog;
595
596 /* Classic BPF related prologue emission. */
597 if (new_prog) {
598 /* Classic BPF expects A and X to be reset first. These need
599 * to be guaranteed to be the first two instructions.
600 */
601 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
602 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
603
604 /* All programs must keep CTX in callee saved BPF_REG_CTX.
605 * In eBPF case it's done by the compiler, here we need to
606 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
607 */
608 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
609 if (*seen_ld_abs) {
610 /* For packet access in classic BPF, cache skb->data
611 * in callee-saved BPF R8 and skb->len - skb->data_len
612 * (headlen) in BPF R9. Since classic BPF is read-only
613 * on CTX, we only need to cache it once.
614 */
615 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
616 BPF_REG_D, BPF_REG_CTX,
617 offsetof(struct sk_buff, data));
618 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
619 offsetof(struct sk_buff, len));
620 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
621 offsetof(struct sk_buff, data_len));
622 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
623 }
624 } else {
625 new_insn += 3;
626 }
627
628 for (i = 0; i < len; fp++, i++) {
629 struct bpf_insn tmp_insns[32] = { };
630 struct bpf_insn *insn = tmp_insns;
631
632 if (addrs)
633 addrs[i] = new_insn - first_insn;
634
635 switch (fp->code) {
636 /* All arithmetic insns and skb loads map as-is. */
637 case BPF_ALU | BPF_ADD | BPF_X:
638 case BPF_ALU | BPF_ADD | BPF_K:
639 case BPF_ALU | BPF_SUB | BPF_X:
640 case BPF_ALU | BPF_SUB | BPF_K:
641 case BPF_ALU | BPF_AND | BPF_X:
642 case BPF_ALU | BPF_AND | BPF_K:
643 case BPF_ALU | BPF_OR | BPF_X:
644 case BPF_ALU | BPF_OR | BPF_K:
645 case BPF_ALU | BPF_LSH | BPF_X:
646 case BPF_ALU | BPF_LSH | BPF_K:
647 case BPF_ALU | BPF_RSH | BPF_X:
648 case BPF_ALU | BPF_RSH | BPF_K:
649 case BPF_ALU | BPF_XOR | BPF_X:
650 case BPF_ALU | BPF_XOR | BPF_K:
651 case BPF_ALU | BPF_MUL | BPF_X:
652 case BPF_ALU | BPF_MUL | BPF_K:
653 case BPF_ALU | BPF_DIV | BPF_X:
654 case BPF_ALU | BPF_DIV | BPF_K:
655 case BPF_ALU | BPF_MOD | BPF_X:
656 case BPF_ALU | BPF_MOD | BPF_K:
657 case BPF_ALU | BPF_NEG:
658 case BPF_LD | BPF_ABS | BPF_W:
659 case BPF_LD | BPF_ABS | BPF_H:
660 case BPF_LD | BPF_ABS | BPF_B:
661 case BPF_LD | BPF_IND | BPF_W:
662 case BPF_LD | BPF_IND | BPF_H:
663 case BPF_LD | BPF_IND | BPF_B:
664 /* Check for overloaded BPF extension and
665 * directly convert it if found, otherwise
666 * just move on with mapping.
667 */
668 if (BPF_CLASS(fp->code) == BPF_LD &&
669 BPF_MODE(fp->code) == BPF_ABS &&
670 convert_bpf_extensions(fp, &insn))
671 break;
672 if (BPF_CLASS(fp->code) == BPF_LD &&
673 convert_bpf_ld_abs(fp, &insn)) {
674 *seen_ld_abs = true;
675 break;
676 }
677
678 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
679 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
680 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
681 /* Error with exception code on div/mod by 0.
682 * For cBPF programs, this was always return 0.
683 */
684 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
685 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
686 *insn++ = BPF_EXIT_INSN();
687 }
688
689 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
690 break;
691
692 /* Jump transformation cannot use BPF block macros
693 * everywhere as offset calculation and target updates
694 * require a bit more work than the rest, i.e. jump
695 * opcodes map as-is, but offsets need adjustment.
696 */
697
698 #define BPF_EMIT_JMP \
699 do { \
700 const s32 off_min = S16_MIN, off_max = S16_MAX; \
701 s32 off; \
702 \
703 if (target >= len || target < 0) \
704 goto err; \
705 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
706 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
707 off -= insn - tmp_insns; \
708 /* Reject anything not fitting into insn->off. */ \
709 if (off < off_min || off > off_max) \
710 goto err; \
711 insn->off = off; \
712 } while (0)
713
714 case BPF_JMP | BPF_JA:
715 target = i + fp->k + 1;
716 insn->code = fp->code;
717 BPF_EMIT_JMP;
718 break;
719
720 case BPF_JMP | BPF_JEQ | BPF_K:
721 case BPF_JMP | BPF_JEQ | BPF_X:
722 case BPF_JMP | BPF_JSET | BPF_K:
723 case BPF_JMP | BPF_JSET | BPF_X:
724 case BPF_JMP | BPF_JGT | BPF_K:
725 case BPF_JMP | BPF_JGT | BPF_X:
726 case BPF_JMP | BPF_JGE | BPF_K:
727 case BPF_JMP | BPF_JGE | BPF_X:
728 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
729 /* BPF immediates are signed, zero extend
730 * immediate into tmp register and use it
731 * in compare insn.
732 */
733 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
734
735 insn->dst_reg = BPF_REG_A;
736 insn->src_reg = BPF_REG_TMP;
737 bpf_src = BPF_X;
738 } else {
739 insn->dst_reg = BPF_REG_A;
740 insn->imm = fp->k;
741 bpf_src = BPF_SRC(fp->code);
742 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
743 }
744
745 /* Common case where 'jump_false' is next insn. */
746 if (fp->jf == 0) {
747 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
748 target = i + fp->jt + 1;
749 BPF_EMIT_JMP;
750 break;
751 }
752
753 /* Convert some jumps when 'jump_true' is next insn. */
754 if (fp->jt == 0) {
755 switch (BPF_OP(fp->code)) {
756 case BPF_JEQ:
757 insn->code = BPF_JMP | BPF_JNE | bpf_src;
758 break;
759 case BPF_JGT:
760 insn->code = BPF_JMP | BPF_JLE | bpf_src;
761 break;
762 case BPF_JGE:
763 insn->code = BPF_JMP | BPF_JLT | bpf_src;
764 break;
765 default:
766 goto jmp_rest;
767 }
768
769 target = i + fp->jf + 1;
770 BPF_EMIT_JMP;
771 break;
772 }
773 jmp_rest:
774 /* Other jumps are mapped into two insns: Jxx and JA. */
775 target = i + fp->jt + 1;
776 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
777 BPF_EMIT_JMP;
778 insn++;
779
780 insn->code = BPF_JMP | BPF_JA;
781 target = i + fp->jf + 1;
782 BPF_EMIT_JMP;
783 break;
784
785 /* ldxb 4 * ([14] & 0xf) is remapped into 6 insns. */
786 case BPF_LDX | BPF_MSH | BPF_B: {
787 struct sock_filter tmp = {
788 .code = BPF_LD | BPF_ABS | BPF_B,
789 .k = fp->k,
790 };
791
792 *seen_ld_abs = true;
793
794 /* X = A */
795 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
796 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
797 convert_bpf_ld_abs(&tmp, &insn);
798 insn++;
799 /* A &= 0xf */
800 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
801 /* A <<= 2 */
802 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
803 /* tmp = X */
804 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
805 /* X = A */
806 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
807 /* A = tmp */
808 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
809 break;
810 }
811 /* RET_K is remapped into 2 insns. RET_A case doesn't need an
812 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
813 */
814 case BPF_RET | BPF_A:
815 case BPF_RET | BPF_K:
816 if (BPF_RVAL(fp->code) == BPF_K)
817 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
818 0, fp->k);
819 *insn = BPF_EXIT_INSN();
820 break;
821
822 /* Store to stack. */
823 case BPF_ST:
824 case BPF_STX:
825 stack_off = fp->k * 4 + 4;
826 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
827 BPF_ST ? BPF_REG_A : BPF_REG_X,
828 -stack_off);
829 /* check_load_and_stores() verifies that classic BPF can
830 * load from stack only after write, so tracking
831 * stack_depth for ST|STX insns is enough
832 */
833 if (new_prog && new_prog->aux->stack_depth < stack_off)
834 new_prog->aux->stack_depth = stack_off;
835 break;
836
837 /* Load from stack. */
838 case BPF_LD | BPF_MEM:
839 case BPF_LDX | BPF_MEM:
840 stack_off = fp->k * 4 + 4;
841 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
842 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
843 -stack_off);
844 break;
845
846 /* A = K or X = K */
847 case BPF_LD | BPF_IMM:
848 case BPF_LDX | BPF_IMM:
849 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
850 BPF_REG_A : BPF_REG_X, fp->k);
851 break;
852
853 /* X = A */
854 case BPF_MISC | BPF_TAX:
855 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
856 break;
857
858 /* A = X */
859 case BPF_MISC | BPF_TXA:
860 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
861 break;
862
863 /* A = skb->len or X = skb->len */
864 case BPF_LD | BPF_W | BPF_LEN:
865 case BPF_LDX | BPF_W | BPF_LEN:
866 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
867 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
868 offsetof(struct sk_buff, len));
869 break;
870
871 /* Access seccomp_data fields. */
872 case BPF_LDX | BPF_ABS | BPF_W:
873 /* A = *(u32 *) (ctx + K) */
874 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
875 break;
876
877 /* Unknown instruction. */
878 default:
879 goto err;
880 }
881
882 insn++;
883 if (new_prog)
884 memcpy(new_insn, tmp_insns,
885 sizeof(*insn) * (insn - tmp_insns));
886 new_insn += insn - tmp_insns;
887 }
888
889 if (!new_prog) {
890 /* Only calculating new length. */
891 *new_len = new_insn - first_insn;
892 if (*seen_ld_abs)
893 *new_len += 4; /* Prologue bits. */
894 return 0;
895 }
896
897 pass++;
898 if (new_flen != new_insn - first_insn) {
899 new_flen = new_insn - first_insn;
900 if (pass > 2)
901 goto err;
902 goto do_pass;
903 }
904
905 kfree(addrs);
906 BUG_ON(*new_len != new_flen);
907 return 0;
908 err:
909 kfree(addrs);
910 return -EINVAL;
911 }
912
913 /* Security:
914 *
915 * As we dont want to clear mem[] array for each packet going through
916 * __bpf_prog_run(), we check that filter loaded by user never try to read
917 * a cell if not previously written, and we check all branches to be sure
918 * a malicious user doesn't try to abuse us.
919 */
check_load_and_stores(const struct sock_filter * filter,int flen)920 static int check_load_and_stores(const struct sock_filter *filter, int flen)
921 {
922 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
923 int pc, ret = 0;
924
925 BUILD_BUG_ON(BPF_MEMWORDS > 16);
926
927 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
928 if (!masks)
929 return -ENOMEM;
930
931 memset(masks, 0xff, flen * sizeof(*masks));
932
933 for (pc = 0; pc < flen; pc++) {
934 memvalid &= masks[pc];
935
936 switch (filter[pc].code) {
937 case BPF_ST:
938 case BPF_STX:
939 memvalid |= (1 << filter[pc].k);
940 break;
941 case BPF_LD | BPF_MEM:
942 case BPF_LDX | BPF_MEM:
943 if (!(memvalid & (1 << filter[pc].k))) {
944 ret = -EINVAL;
945 goto error;
946 }
947 break;
948 case BPF_JMP | BPF_JA:
949 /* A jump must set masks on target */
950 masks[pc + 1 + filter[pc].k] &= memvalid;
951 memvalid = ~0;
952 break;
953 case BPF_JMP | BPF_JEQ | BPF_K:
954 case BPF_JMP | BPF_JEQ | BPF_X:
955 case BPF_JMP | BPF_JGE | BPF_K:
956 case BPF_JMP | BPF_JGE | BPF_X:
957 case BPF_JMP | BPF_JGT | BPF_K:
958 case BPF_JMP | BPF_JGT | BPF_X:
959 case BPF_JMP | BPF_JSET | BPF_K:
960 case BPF_JMP | BPF_JSET | BPF_X:
961 /* A jump must set masks on targets */
962 masks[pc + 1 + filter[pc].jt] &= memvalid;
963 masks[pc + 1 + filter[pc].jf] &= memvalid;
964 memvalid = ~0;
965 break;
966 }
967 }
968 error:
969 kfree(masks);
970 return ret;
971 }
972
chk_code_allowed(u16 code_to_probe)973 static bool chk_code_allowed(u16 code_to_probe)
974 {
975 static const bool codes[] = {
976 /* 32 bit ALU operations */
977 [BPF_ALU | BPF_ADD | BPF_K] = true,
978 [BPF_ALU | BPF_ADD | BPF_X] = true,
979 [BPF_ALU | BPF_SUB | BPF_K] = true,
980 [BPF_ALU | BPF_SUB | BPF_X] = true,
981 [BPF_ALU | BPF_MUL | BPF_K] = true,
982 [BPF_ALU | BPF_MUL | BPF_X] = true,
983 [BPF_ALU | BPF_DIV | BPF_K] = true,
984 [BPF_ALU | BPF_DIV | BPF_X] = true,
985 [BPF_ALU | BPF_MOD | BPF_K] = true,
986 [BPF_ALU | BPF_MOD | BPF_X] = true,
987 [BPF_ALU | BPF_AND | BPF_K] = true,
988 [BPF_ALU | BPF_AND | BPF_X] = true,
989 [BPF_ALU | BPF_OR | BPF_K] = true,
990 [BPF_ALU | BPF_OR | BPF_X] = true,
991 [BPF_ALU | BPF_XOR | BPF_K] = true,
992 [BPF_ALU | BPF_XOR | BPF_X] = true,
993 [BPF_ALU | BPF_LSH | BPF_K] = true,
994 [BPF_ALU | BPF_LSH | BPF_X] = true,
995 [BPF_ALU | BPF_RSH | BPF_K] = true,
996 [BPF_ALU | BPF_RSH | BPF_X] = true,
997 [BPF_ALU | BPF_NEG] = true,
998 /* Load instructions */
999 [BPF_LD | BPF_W | BPF_ABS] = true,
1000 [BPF_LD | BPF_H | BPF_ABS] = true,
1001 [BPF_LD | BPF_B | BPF_ABS] = true,
1002 [BPF_LD | BPF_W | BPF_LEN] = true,
1003 [BPF_LD | BPF_W | BPF_IND] = true,
1004 [BPF_LD | BPF_H | BPF_IND] = true,
1005 [BPF_LD | BPF_B | BPF_IND] = true,
1006 [BPF_LD | BPF_IMM] = true,
1007 [BPF_LD | BPF_MEM] = true,
1008 [BPF_LDX | BPF_W | BPF_LEN] = true,
1009 [BPF_LDX | BPF_B | BPF_MSH] = true,
1010 [BPF_LDX | BPF_IMM] = true,
1011 [BPF_LDX | BPF_MEM] = true,
1012 /* Store instructions */
1013 [BPF_ST] = true,
1014 [BPF_STX] = true,
1015 /* Misc instructions */
1016 [BPF_MISC | BPF_TAX] = true,
1017 [BPF_MISC | BPF_TXA] = true,
1018 /* Return instructions */
1019 [BPF_RET | BPF_K] = true,
1020 [BPF_RET | BPF_A] = true,
1021 /* Jump instructions */
1022 [BPF_JMP | BPF_JA] = true,
1023 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1024 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1025 [BPF_JMP | BPF_JGE | BPF_K] = true,
1026 [BPF_JMP | BPF_JGE | BPF_X] = true,
1027 [BPF_JMP | BPF_JGT | BPF_K] = true,
1028 [BPF_JMP | BPF_JGT | BPF_X] = true,
1029 [BPF_JMP | BPF_JSET | BPF_K] = true,
1030 [BPF_JMP | BPF_JSET | BPF_X] = true,
1031 };
1032
1033 if (code_to_probe >= ARRAY_SIZE(codes))
1034 return false;
1035
1036 return codes[code_to_probe];
1037 }
1038
bpf_check_basics_ok(const struct sock_filter * filter,unsigned int flen)1039 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1040 unsigned int flen)
1041 {
1042 if (filter == NULL)
1043 return false;
1044 if (flen == 0 || flen > BPF_MAXINSNS)
1045 return false;
1046
1047 return true;
1048 }
1049
1050 /**
1051 * bpf_check_classic - verify socket filter code
1052 * @filter: filter to verify
1053 * @flen: length of filter
1054 *
1055 * Check the user's filter code. If we let some ugly
1056 * filter code slip through kaboom! The filter must contain
1057 * no references or jumps that are out of range, no illegal
1058 * instructions, and must end with a RET instruction.
1059 *
1060 * All jumps are forward as they are not signed.
1061 *
1062 * Returns 0 if the rule set is legal or -EINVAL if not.
1063 */
bpf_check_classic(const struct sock_filter * filter,unsigned int flen)1064 static int bpf_check_classic(const struct sock_filter *filter,
1065 unsigned int flen)
1066 {
1067 bool anc_found;
1068 int pc;
1069
1070 /* Check the filter code now */
1071 for (pc = 0; pc < flen; pc++) {
1072 const struct sock_filter *ftest = &filter[pc];
1073
1074 /* May we actually operate on this code? */
1075 if (!chk_code_allowed(ftest->code))
1076 return -EINVAL;
1077
1078 /* Some instructions need special checks */
1079 switch (ftest->code) {
1080 case BPF_ALU | BPF_DIV | BPF_K:
1081 case BPF_ALU | BPF_MOD | BPF_K:
1082 /* Check for division by zero */
1083 if (ftest->k == 0)
1084 return -EINVAL;
1085 break;
1086 case BPF_ALU | BPF_LSH | BPF_K:
1087 case BPF_ALU | BPF_RSH | BPF_K:
1088 if (ftest->k >= 32)
1089 return -EINVAL;
1090 break;
1091 case BPF_LD | BPF_MEM:
1092 case BPF_LDX | BPF_MEM:
1093 case BPF_ST:
1094 case BPF_STX:
1095 /* Check for invalid memory addresses */
1096 if (ftest->k >= BPF_MEMWORDS)
1097 return -EINVAL;
1098 break;
1099 case BPF_JMP | BPF_JA:
1100 /* Note, the large ftest->k might cause loops.
1101 * Compare this with conditional jumps below,
1102 * where offsets are limited. --ANK (981016)
1103 */
1104 if (ftest->k >= (unsigned int)(flen - pc - 1))
1105 return -EINVAL;
1106 break;
1107 case BPF_JMP | BPF_JEQ | BPF_K:
1108 case BPF_JMP | BPF_JEQ | BPF_X:
1109 case BPF_JMP | BPF_JGE | BPF_K:
1110 case BPF_JMP | BPF_JGE | BPF_X:
1111 case BPF_JMP | BPF_JGT | BPF_K:
1112 case BPF_JMP | BPF_JGT | BPF_X:
1113 case BPF_JMP | BPF_JSET | BPF_K:
1114 case BPF_JMP | BPF_JSET | BPF_X:
1115 /* Both conditionals must be safe */
1116 if (pc + ftest->jt + 1 >= flen ||
1117 pc + ftest->jf + 1 >= flen)
1118 return -EINVAL;
1119 break;
1120 case BPF_LD | BPF_W | BPF_ABS:
1121 case BPF_LD | BPF_H | BPF_ABS:
1122 case BPF_LD | BPF_B | BPF_ABS:
1123 anc_found = false;
1124 if (bpf_anc_helper(ftest) & BPF_ANC)
1125 anc_found = true;
1126 /* Ancillary operation unknown or unsupported */
1127 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1128 return -EINVAL;
1129 }
1130 }
1131
1132 /* Last instruction must be a RET code */
1133 switch (filter[flen - 1].code) {
1134 case BPF_RET | BPF_K:
1135 case BPF_RET | BPF_A:
1136 return check_load_and_stores(filter, flen);
1137 }
1138
1139 return -EINVAL;
1140 }
1141
bpf_prog_store_orig_filter(struct bpf_prog * fp,const struct sock_fprog * fprog)1142 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1143 const struct sock_fprog *fprog)
1144 {
1145 unsigned int fsize = bpf_classic_proglen(fprog);
1146 struct sock_fprog_kern *fkprog;
1147
1148 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1149 if (!fp->orig_prog)
1150 return -ENOMEM;
1151
1152 fkprog = fp->orig_prog;
1153 fkprog->len = fprog->len;
1154
1155 fkprog->filter = kmemdup(fp->insns, fsize,
1156 GFP_KERNEL | __GFP_NOWARN);
1157 if (!fkprog->filter) {
1158 kfree(fp->orig_prog);
1159 return -ENOMEM;
1160 }
1161
1162 return 0;
1163 }
1164
bpf_release_orig_filter(struct bpf_prog * fp)1165 static void bpf_release_orig_filter(struct bpf_prog *fp)
1166 {
1167 struct sock_fprog_kern *fprog = fp->orig_prog;
1168
1169 if (fprog) {
1170 kfree(fprog->filter);
1171 kfree(fprog);
1172 }
1173 }
1174
__bpf_prog_release(struct bpf_prog * prog)1175 static void __bpf_prog_release(struct bpf_prog *prog)
1176 {
1177 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1178 bpf_prog_put(prog);
1179 } else {
1180 bpf_release_orig_filter(prog);
1181 bpf_prog_free(prog);
1182 }
1183 }
1184
__sk_filter_release(struct sk_filter * fp)1185 static void __sk_filter_release(struct sk_filter *fp)
1186 {
1187 __bpf_prog_release(fp->prog);
1188 kfree(fp);
1189 }
1190
1191 /**
1192 * sk_filter_release_rcu - Release a socket filter by rcu_head
1193 * @rcu: rcu_head that contains the sk_filter to free
1194 */
sk_filter_release_rcu(struct rcu_head * rcu)1195 static void sk_filter_release_rcu(struct rcu_head *rcu)
1196 {
1197 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1198
1199 __sk_filter_release(fp);
1200 }
1201
1202 /**
1203 * sk_filter_release - release a socket filter
1204 * @fp: filter to remove
1205 *
1206 * Remove a filter from a socket and release its resources.
1207 */
sk_filter_release(struct sk_filter * fp)1208 static void sk_filter_release(struct sk_filter *fp)
1209 {
1210 if (refcount_dec_and_test(&fp->refcnt))
1211 call_rcu(&fp->rcu, sk_filter_release_rcu);
1212 }
1213
sk_filter_uncharge(struct sock * sk,struct sk_filter * fp)1214 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1215 {
1216 u32 filter_size = bpf_prog_size(fp->prog->len);
1217
1218 atomic_sub(filter_size, &sk->sk_omem_alloc);
1219 sk_filter_release(fp);
1220 }
1221
1222 /* try to charge the socket memory if there is space available
1223 * return true on success
1224 */
__sk_filter_charge(struct sock * sk,struct sk_filter * fp)1225 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1226 {
1227 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
1228 u32 filter_size = bpf_prog_size(fp->prog->len);
1229
1230 /* same check as in sock_kmalloc() */
1231 if (filter_size <= optmem_max &&
1232 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1233 atomic_add(filter_size, &sk->sk_omem_alloc);
1234 return true;
1235 }
1236 return false;
1237 }
1238
sk_filter_charge(struct sock * sk,struct sk_filter * fp)1239 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1240 {
1241 if (!refcount_inc_not_zero(&fp->refcnt))
1242 return false;
1243
1244 if (!__sk_filter_charge(sk, fp)) {
1245 sk_filter_release(fp);
1246 return false;
1247 }
1248 return true;
1249 }
1250
bpf_migrate_filter(struct bpf_prog * fp)1251 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1252 {
1253 struct sock_filter *old_prog;
1254 struct bpf_prog *old_fp;
1255 int err, new_len, old_len = fp->len;
1256 bool seen_ld_abs = false;
1257
1258 /* We are free to overwrite insns et al right here as it won't be used at
1259 * this point in time anymore internally after the migration to the eBPF
1260 * instruction representation.
1261 */
1262 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1263 sizeof(struct bpf_insn));
1264
1265 /* Conversion cannot happen on overlapping memory areas,
1266 * so we need to keep the user BPF around until the 2nd
1267 * pass. At this time, the user BPF is stored in fp->insns.
1268 */
1269 old_prog = kmemdup_array(fp->insns, old_len, sizeof(struct sock_filter),
1270 GFP_KERNEL | __GFP_NOWARN);
1271 if (!old_prog) {
1272 err = -ENOMEM;
1273 goto out_err;
1274 }
1275
1276 /* 1st pass: calculate the new program length. */
1277 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1278 &seen_ld_abs);
1279 if (err)
1280 goto out_err_free;
1281
1282 /* Expand fp for appending the new filter representation. */
1283 old_fp = fp;
1284 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1285 if (!fp) {
1286 /* The old_fp is still around in case we couldn't
1287 * allocate new memory, so uncharge on that one.
1288 */
1289 fp = old_fp;
1290 err = -ENOMEM;
1291 goto out_err_free;
1292 }
1293
1294 fp->len = new_len;
1295
1296 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1297 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1298 &seen_ld_abs);
1299 if (err)
1300 /* 2nd bpf_convert_filter() can fail only if it fails
1301 * to allocate memory, remapping must succeed. Note,
1302 * that at this time old_fp has already been released
1303 * by krealloc().
1304 */
1305 goto out_err_free;
1306
1307 fp = bpf_prog_select_runtime(fp, &err);
1308 if (err)
1309 goto out_err_free;
1310
1311 kfree(old_prog);
1312 return fp;
1313
1314 out_err_free:
1315 kfree(old_prog);
1316 out_err:
1317 __bpf_prog_release(fp);
1318 return ERR_PTR(err);
1319 }
1320
bpf_prepare_filter(struct bpf_prog * fp,bpf_aux_classic_check_t trans)1321 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1322 bpf_aux_classic_check_t trans)
1323 {
1324 int err;
1325
1326 fp->bpf_func = NULL;
1327 fp->jited = 0;
1328
1329 err = bpf_check_classic(fp->insns, fp->len);
1330 if (err) {
1331 __bpf_prog_release(fp);
1332 return ERR_PTR(err);
1333 }
1334
1335 /* There might be additional checks and transformations
1336 * needed on classic filters, f.e. in case of seccomp.
1337 */
1338 if (trans) {
1339 err = trans(fp->insns, fp->len);
1340 if (err) {
1341 __bpf_prog_release(fp);
1342 return ERR_PTR(err);
1343 }
1344 }
1345
1346 /* Probe if we can JIT compile the filter and if so, do
1347 * the compilation of the filter.
1348 */
1349 bpf_jit_compile(fp);
1350
1351 /* JIT compiler couldn't process this filter, so do the eBPF translation
1352 * for the optimized interpreter.
1353 */
1354 if (!fp->jited)
1355 fp = bpf_migrate_filter(fp);
1356
1357 return fp;
1358 }
1359
1360 /**
1361 * bpf_prog_create - create an unattached filter
1362 * @pfp: the unattached filter that is created
1363 * @fprog: the filter program
1364 *
1365 * Create a filter independent of any socket. We first run some
1366 * sanity checks on it to make sure it does not explode on us later.
1367 * If an error occurs or there is insufficient memory for the filter
1368 * a negative errno code is returned. On success the return is zero.
1369 */
bpf_prog_create(struct bpf_prog ** pfp,struct sock_fprog_kern * fprog)1370 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1371 {
1372 unsigned int fsize = bpf_classic_proglen(fprog);
1373 struct bpf_prog *fp;
1374
1375 /* Make sure new filter is there and in the right amounts. */
1376 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1377 return -EINVAL;
1378
1379 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1380 if (!fp)
1381 return -ENOMEM;
1382
1383 memcpy(fp->insns, fprog->filter, fsize);
1384
1385 fp->len = fprog->len;
1386 /* Since unattached filters are not copied back to user
1387 * space through sk_get_filter(), we do not need to hold
1388 * a copy here, and can spare us the work.
1389 */
1390 fp->orig_prog = NULL;
1391
1392 /* bpf_prepare_filter() already takes care of freeing
1393 * memory in case something goes wrong.
1394 */
1395 fp = bpf_prepare_filter(fp, NULL);
1396 if (IS_ERR(fp))
1397 return PTR_ERR(fp);
1398
1399 *pfp = fp;
1400 return 0;
1401 }
1402 EXPORT_SYMBOL_GPL(bpf_prog_create);
1403
1404 /**
1405 * bpf_prog_create_from_user - create an unattached filter from user buffer
1406 * @pfp: the unattached filter that is created
1407 * @fprog: the filter program
1408 * @trans: post-classic verifier transformation handler
1409 * @save_orig: save classic BPF program
1410 *
1411 * This function effectively does the same as bpf_prog_create(), only
1412 * that it builds up its insns buffer from user space provided buffer.
1413 * It also allows for passing a bpf_aux_classic_check_t handler.
1414 */
bpf_prog_create_from_user(struct bpf_prog ** pfp,struct sock_fprog * fprog,bpf_aux_classic_check_t trans,bool save_orig)1415 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1416 bpf_aux_classic_check_t trans, bool save_orig)
1417 {
1418 unsigned int fsize = bpf_classic_proglen(fprog);
1419 struct bpf_prog *fp;
1420 int err;
1421
1422 /* Make sure new filter is there and in the right amounts. */
1423 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1424 return -EINVAL;
1425
1426 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1427 if (!fp)
1428 return -ENOMEM;
1429
1430 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1431 __bpf_prog_free(fp);
1432 return -EFAULT;
1433 }
1434
1435 fp->len = fprog->len;
1436 fp->orig_prog = NULL;
1437
1438 if (save_orig) {
1439 err = bpf_prog_store_orig_filter(fp, fprog);
1440 if (err) {
1441 __bpf_prog_free(fp);
1442 return -ENOMEM;
1443 }
1444 }
1445
1446 /* bpf_prepare_filter() already takes care of freeing
1447 * memory in case something goes wrong.
1448 */
1449 fp = bpf_prepare_filter(fp, trans);
1450 if (IS_ERR(fp))
1451 return PTR_ERR(fp);
1452
1453 *pfp = fp;
1454 return 0;
1455 }
1456 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1457
bpf_prog_destroy(struct bpf_prog * fp)1458 void bpf_prog_destroy(struct bpf_prog *fp)
1459 {
1460 __bpf_prog_release(fp);
1461 }
1462 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1463
__sk_attach_prog(struct bpf_prog * prog,struct sock * sk)1464 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1465 {
1466 struct sk_filter *fp, *old_fp;
1467
1468 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1469 if (!fp)
1470 return -ENOMEM;
1471
1472 fp->prog = prog;
1473
1474 if (!__sk_filter_charge(sk, fp)) {
1475 kfree(fp);
1476 return -ENOMEM;
1477 }
1478 refcount_set(&fp->refcnt, 1);
1479
1480 old_fp = rcu_dereference_protected(sk->sk_filter,
1481 lockdep_sock_is_held(sk));
1482 rcu_assign_pointer(sk->sk_filter, fp);
1483
1484 if (old_fp)
1485 sk_filter_uncharge(sk, old_fp);
1486
1487 return 0;
1488 }
1489
1490 static
__get_filter(struct sock_fprog * fprog,struct sock * sk)1491 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1492 {
1493 unsigned int fsize = bpf_classic_proglen(fprog);
1494 struct bpf_prog *prog;
1495 int err;
1496
1497 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1498 return ERR_PTR(-EPERM);
1499
1500 /* Make sure new filter is there and in the right amounts. */
1501 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1502 return ERR_PTR(-EINVAL);
1503
1504 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1505 if (!prog)
1506 return ERR_PTR(-ENOMEM);
1507
1508 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1509 __bpf_prog_free(prog);
1510 return ERR_PTR(-EFAULT);
1511 }
1512
1513 prog->len = fprog->len;
1514
1515 err = bpf_prog_store_orig_filter(prog, fprog);
1516 if (err) {
1517 __bpf_prog_free(prog);
1518 return ERR_PTR(-ENOMEM);
1519 }
1520
1521 /* bpf_prepare_filter() already takes care of freeing
1522 * memory in case something goes wrong.
1523 */
1524 return bpf_prepare_filter(prog, NULL);
1525 }
1526
1527 /**
1528 * sk_attach_filter - attach a socket filter
1529 * @fprog: the filter program
1530 * @sk: the socket to use
1531 *
1532 * Attach the user's filter code. We first run some sanity checks on
1533 * it to make sure it does not explode on us later. If an error
1534 * occurs or there is insufficient memory for the filter a negative
1535 * errno code is returned. On success the return is zero.
1536 */
sk_attach_filter(struct sock_fprog * fprog,struct sock * sk)1537 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1538 {
1539 struct bpf_prog *prog = __get_filter(fprog, sk);
1540 int err;
1541
1542 if (IS_ERR(prog))
1543 return PTR_ERR(prog);
1544
1545 err = __sk_attach_prog(prog, sk);
1546 if (err < 0) {
1547 __bpf_prog_release(prog);
1548 return err;
1549 }
1550
1551 return 0;
1552 }
1553 EXPORT_SYMBOL_GPL(sk_attach_filter);
1554
sk_reuseport_attach_filter(struct sock_fprog * fprog,struct sock * sk)1555 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1556 {
1557 struct bpf_prog *prog = __get_filter(fprog, sk);
1558 int err, optmem_max;
1559
1560 if (IS_ERR(prog))
1561 return PTR_ERR(prog);
1562
1563 optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
1564 if (bpf_prog_size(prog->len) > optmem_max)
1565 err = -ENOMEM;
1566 else
1567 err = reuseport_attach_prog(sk, prog);
1568
1569 if (err)
1570 __bpf_prog_release(prog);
1571
1572 return err;
1573 }
1574
__get_bpf(u32 ufd,struct sock * sk)1575 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1576 {
1577 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1578 return ERR_PTR(-EPERM);
1579
1580 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1581 }
1582
sk_attach_bpf(u32 ufd,struct sock * sk)1583 int sk_attach_bpf(u32 ufd, struct sock *sk)
1584 {
1585 struct bpf_prog *prog = __get_bpf(ufd, sk);
1586 int err;
1587
1588 if (IS_ERR(prog))
1589 return PTR_ERR(prog);
1590
1591 err = __sk_attach_prog(prog, sk);
1592 if (err < 0) {
1593 bpf_prog_put(prog);
1594 return err;
1595 }
1596
1597 return 0;
1598 }
1599
sk_reuseport_attach_bpf(u32 ufd,struct sock * sk)1600 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1601 {
1602 struct bpf_prog *prog;
1603 int err, optmem_max;
1604
1605 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1606 return -EPERM;
1607
1608 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1609 if (PTR_ERR(prog) == -EINVAL)
1610 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1611 if (IS_ERR(prog))
1612 return PTR_ERR(prog);
1613
1614 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1615 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1616 * bpf prog (e.g. sockmap). It depends on the
1617 * limitation imposed by bpf_prog_load().
1618 * Hence, sysctl_optmem_max is not checked.
1619 */
1620 if ((sk->sk_type != SOCK_STREAM &&
1621 sk->sk_type != SOCK_DGRAM) ||
1622 (sk->sk_protocol != IPPROTO_UDP &&
1623 sk->sk_protocol != IPPROTO_TCP) ||
1624 (sk->sk_family != AF_INET &&
1625 sk->sk_family != AF_INET6)) {
1626 err = -ENOTSUPP;
1627 goto err_prog_put;
1628 }
1629 } else {
1630 /* BPF_PROG_TYPE_SOCKET_FILTER */
1631 optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
1632 if (bpf_prog_size(prog->len) > optmem_max) {
1633 err = -ENOMEM;
1634 goto err_prog_put;
1635 }
1636 }
1637
1638 err = reuseport_attach_prog(sk, prog);
1639 err_prog_put:
1640 if (err)
1641 bpf_prog_put(prog);
1642
1643 return err;
1644 }
1645
sk_reuseport_prog_free(struct bpf_prog * prog)1646 void sk_reuseport_prog_free(struct bpf_prog *prog)
1647 {
1648 if (!prog)
1649 return;
1650
1651 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1652 bpf_prog_put(prog);
1653 else
1654 bpf_prog_destroy(prog);
1655 }
1656
1657 struct bpf_scratchpad {
1658 union {
1659 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1660 u8 buff[MAX_BPF_STACK];
1661 };
1662 local_lock_t bh_lock;
1663 };
1664
1665 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp) = {
1666 .bh_lock = INIT_LOCAL_LOCK(bh_lock),
1667 };
1668
__bpf_try_make_writable(struct sk_buff * skb,unsigned int write_len)1669 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1670 unsigned int write_len)
1671 {
1672 #ifdef CONFIG_DEBUG_NET
1673 /* Avoid a splat in pskb_may_pull_reason() */
1674 if (write_len > INT_MAX)
1675 return -EINVAL;
1676 #endif
1677 return skb_ensure_writable(skb, write_len);
1678 }
1679
bpf_try_make_writable(struct sk_buff * skb,unsigned int write_len)1680 static inline int bpf_try_make_writable(struct sk_buff *skb,
1681 unsigned int write_len)
1682 {
1683 int err = __bpf_try_make_writable(skb, write_len);
1684
1685 bpf_compute_data_pointers(skb);
1686 return err;
1687 }
1688
bpf_try_make_head_writable(struct sk_buff * skb)1689 static int bpf_try_make_head_writable(struct sk_buff *skb)
1690 {
1691 return bpf_try_make_writable(skb, skb_headlen(skb));
1692 }
1693
bpf_push_mac_rcsum(struct sk_buff * skb)1694 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1695 {
1696 if (skb_at_tc_ingress(skb))
1697 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1698 }
1699
bpf_pull_mac_rcsum(struct sk_buff * skb)1700 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1701 {
1702 if (skb_at_tc_ingress(skb))
1703 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1704 }
1705
BPF_CALL_5(bpf_skb_store_bytes,struct sk_buff *,skb,u32,offset,const void *,from,u32,len,u64,flags)1706 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1707 const void *, from, u32, len, u64, flags)
1708 {
1709 void *ptr;
1710
1711 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1712 return -EINVAL;
1713 if (unlikely(offset > INT_MAX))
1714 return -EFAULT;
1715 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1716 return -EFAULT;
1717
1718 ptr = skb->data + offset;
1719 if (flags & BPF_F_RECOMPUTE_CSUM)
1720 __skb_postpull_rcsum(skb, ptr, len, offset);
1721
1722 memcpy(ptr, from, len);
1723
1724 if (flags & BPF_F_RECOMPUTE_CSUM)
1725 __skb_postpush_rcsum(skb, ptr, len, offset);
1726 if (flags & BPF_F_INVALIDATE_HASH)
1727 skb_clear_hash(skb);
1728
1729 return 0;
1730 }
1731
1732 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1733 .func = bpf_skb_store_bytes,
1734 .gpl_only = false,
1735 .ret_type = RET_INTEGER,
1736 .arg1_type = ARG_PTR_TO_CTX,
1737 .arg2_type = ARG_ANYTHING,
1738 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1739 .arg4_type = ARG_CONST_SIZE,
1740 .arg5_type = ARG_ANYTHING,
1741 };
1742
__bpf_skb_store_bytes(struct sk_buff * skb,u32 offset,const void * from,u32 len,u64 flags)1743 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from,
1744 u32 len, u64 flags)
1745 {
1746 return ____bpf_skb_store_bytes(skb, offset, from, len, flags);
1747 }
1748
BPF_CALL_4(bpf_skb_load_bytes,const struct sk_buff *,skb,u32,offset,void *,to,u32,len)1749 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1750 void *, to, u32, len)
1751 {
1752 void *ptr;
1753
1754 if (unlikely(offset > INT_MAX))
1755 goto err_clear;
1756
1757 ptr = skb_header_pointer(skb, offset, len, to);
1758 if (unlikely(!ptr))
1759 goto err_clear;
1760 if (ptr != to)
1761 memcpy(to, ptr, len);
1762
1763 return 0;
1764 err_clear:
1765 memset(to, 0, len);
1766 return -EFAULT;
1767 }
1768
1769 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1770 .func = bpf_skb_load_bytes,
1771 .gpl_only = false,
1772 .ret_type = RET_INTEGER,
1773 .arg1_type = ARG_PTR_TO_CTX,
1774 .arg2_type = ARG_ANYTHING,
1775 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1776 .arg4_type = ARG_CONST_SIZE,
1777 };
1778
__bpf_skb_load_bytes(const struct sk_buff * skb,u32 offset,void * to,u32 len)1779 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len)
1780 {
1781 return ____bpf_skb_load_bytes(skb, offset, to, len);
1782 }
1783
BPF_CALL_4(bpf_flow_dissector_load_bytes,const struct bpf_flow_dissector *,ctx,u32,offset,void *,to,u32,len)1784 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1785 const struct bpf_flow_dissector *, ctx, u32, offset,
1786 void *, to, u32, len)
1787 {
1788 void *ptr;
1789
1790 if (unlikely(offset > 0xffff))
1791 goto err_clear;
1792
1793 if (unlikely(!ctx->skb))
1794 goto err_clear;
1795
1796 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1797 if (unlikely(!ptr))
1798 goto err_clear;
1799 if (ptr != to)
1800 memcpy(to, ptr, len);
1801
1802 return 0;
1803 err_clear:
1804 memset(to, 0, len);
1805 return -EFAULT;
1806 }
1807
1808 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1809 .func = bpf_flow_dissector_load_bytes,
1810 .gpl_only = false,
1811 .ret_type = RET_INTEGER,
1812 .arg1_type = ARG_PTR_TO_CTX,
1813 .arg2_type = ARG_ANYTHING,
1814 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1815 .arg4_type = ARG_CONST_SIZE,
1816 };
1817
BPF_CALL_5(bpf_skb_load_bytes_relative,const struct sk_buff *,skb,u32,offset,void *,to,u32,len,u32,start_header)1818 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1819 u32, offset, void *, to, u32, len, u32, start_header)
1820 {
1821 u8 *end = skb_tail_pointer(skb);
1822 u8 *start, *ptr;
1823
1824 if (unlikely(offset > 0xffff))
1825 goto err_clear;
1826
1827 switch (start_header) {
1828 case BPF_HDR_START_MAC:
1829 if (unlikely(!skb_mac_header_was_set(skb)))
1830 goto err_clear;
1831 start = skb_mac_header(skb);
1832 break;
1833 case BPF_HDR_START_NET:
1834 start = skb_network_header(skb);
1835 break;
1836 default:
1837 goto err_clear;
1838 }
1839
1840 ptr = start + offset;
1841
1842 if (likely(ptr + len <= end)) {
1843 memcpy(to, ptr, len);
1844 return 0;
1845 }
1846
1847 err_clear:
1848 memset(to, 0, len);
1849 return -EFAULT;
1850 }
1851
1852 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1853 .func = bpf_skb_load_bytes_relative,
1854 .gpl_only = false,
1855 .ret_type = RET_INTEGER,
1856 .arg1_type = ARG_PTR_TO_CTX,
1857 .arg2_type = ARG_ANYTHING,
1858 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1859 .arg4_type = ARG_CONST_SIZE,
1860 .arg5_type = ARG_ANYTHING,
1861 };
1862
BPF_CALL_2(bpf_skb_pull_data,struct sk_buff *,skb,u32,len)1863 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1864 {
1865 /* Idea is the following: should the needed direct read/write
1866 * test fail during runtime, we can pull in more data and redo
1867 * again, since implicitly, we invalidate previous checks here.
1868 *
1869 * Or, since we know how much we need to make read/writeable,
1870 * this can be done once at the program beginning for direct
1871 * access case. By this we overcome limitations of only current
1872 * headroom being accessible.
1873 */
1874 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1875 }
1876
1877 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1878 .func = bpf_skb_pull_data,
1879 .gpl_only = false,
1880 .ret_type = RET_INTEGER,
1881 .arg1_type = ARG_PTR_TO_CTX,
1882 .arg2_type = ARG_ANYTHING,
1883 };
1884
BPF_CALL_1(bpf_sk_fullsock,struct sock *,sk)1885 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1886 {
1887 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1888 }
1889
1890 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1891 .func = bpf_sk_fullsock,
1892 .gpl_only = false,
1893 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1894 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1895 };
1896
sk_skb_try_make_writable(struct sk_buff * skb,unsigned int write_len)1897 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1898 unsigned int write_len)
1899 {
1900 return __bpf_try_make_writable(skb, write_len);
1901 }
1902
BPF_CALL_2(sk_skb_pull_data,struct sk_buff *,skb,u32,len)1903 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1904 {
1905 /* Idea is the following: should the needed direct read/write
1906 * test fail during runtime, we can pull in more data and redo
1907 * again, since implicitly, we invalidate previous checks here.
1908 *
1909 * Or, since we know how much we need to make read/writeable,
1910 * this can be done once at the program beginning for direct
1911 * access case. By this we overcome limitations of only current
1912 * headroom being accessible.
1913 */
1914 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1915 }
1916
1917 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1918 .func = sk_skb_pull_data,
1919 .gpl_only = false,
1920 .ret_type = RET_INTEGER,
1921 .arg1_type = ARG_PTR_TO_CTX,
1922 .arg2_type = ARG_ANYTHING,
1923 };
1924
BPF_CALL_5(bpf_l3_csum_replace,struct sk_buff *,skb,u32,offset,u64,from,u64,to,u64,flags)1925 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1926 u64, from, u64, to, u64, flags)
1927 {
1928 __sum16 *ptr;
1929
1930 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1931 return -EINVAL;
1932 if (unlikely(offset > 0xffff || offset & 1))
1933 return -EFAULT;
1934 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1935 return -EFAULT;
1936
1937 ptr = (__sum16 *)(skb->data + offset);
1938 switch (flags & BPF_F_HDR_FIELD_MASK) {
1939 case 0:
1940 if (unlikely(from != 0))
1941 return -EINVAL;
1942
1943 csum_replace_by_diff(ptr, to);
1944 break;
1945 case 2:
1946 csum_replace2(ptr, from, to);
1947 break;
1948 case 4:
1949 csum_replace4(ptr, from, to);
1950 break;
1951 default:
1952 return -EINVAL;
1953 }
1954
1955 return 0;
1956 }
1957
1958 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1959 .func = bpf_l3_csum_replace,
1960 .gpl_only = false,
1961 .ret_type = RET_INTEGER,
1962 .arg1_type = ARG_PTR_TO_CTX,
1963 .arg2_type = ARG_ANYTHING,
1964 .arg3_type = ARG_ANYTHING,
1965 .arg4_type = ARG_ANYTHING,
1966 .arg5_type = ARG_ANYTHING,
1967 };
1968
BPF_CALL_5(bpf_l4_csum_replace,struct sk_buff *,skb,u32,offset,u64,from,u64,to,u64,flags)1969 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1970 u64, from, u64, to, u64, flags)
1971 {
1972 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1973 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1974 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1975 __sum16 *ptr;
1976
1977 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1978 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1979 return -EINVAL;
1980 if (unlikely(offset > 0xffff || offset & 1))
1981 return -EFAULT;
1982 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1983 return -EFAULT;
1984
1985 ptr = (__sum16 *)(skb->data + offset);
1986 if (is_mmzero && !do_mforce && !*ptr)
1987 return 0;
1988
1989 switch (flags & BPF_F_HDR_FIELD_MASK) {
1990 case 0:
1991 if (unlikely(from != 0))
1992 return -EINVAL;
1993
1994 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1995 break;
1996 case 2:
1997 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1998 break;
1999 case 4:
2000 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
2001 break;
2002 default:
2003 return -EINVAL;
2004 }
2005
2006 if (is_mmzero && !*ptr)
2007 *ptr = CSUM_MANGLED_0;
2008 return 0;
2009 }
2010
2011 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
2012 .func = bpf_l4_csum_replace,
2013 .gpl_only = false,
2014 .ret_type = RET_INTEGER,
2015 .arg1_type = ARG_PTR_TO_CTX,
2016 .arg2_type = ARG_ANYTHING,
2017 .arg3_type = ARG_ANYTHING,
2018 .arg4_type = ARG_ANYTHING,
2019 .arg5_type = ARG_ANYTHING,
2020 };
2021
BPF_CALL_5(bpf_csum_diff,__be32 *,from,u32,from_size,__be32 *,to,u32,to_size,__wsum,seed)2022 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
2023 __be32 *, to, u32, to_size, __wsum, seed)
2024 {
2025 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
2026 u32 diff_size = from_size + to_size;
2027 int i, j = 0;
2028 __wsum ret;
2029
2030 /* This is quite flexible, some examples:
2031 *
2032 * from_size == 0, to_size > 0, seed := csum --> pushing data
2033 * from_size > 0, to_size == 0, seed := csum --> pulling data
2034 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2035 *
2036 * Even for diffing, from_size and to_size don't need to be equal.
2037 */
2038 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2039 diff_size > sizeof(sp->diff)))
2040 return -EINVAL;
2041
2042 local_lock_nested_bh(&bpf_sp.bh_lock);
2043 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2044 sp->diff[j] = ~from[i];
2045 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2046 sp->diff[j] = to[i];
2047
2048 ret = csum_partial(sp->diff, diff_size, seed);
2049 local_unlock_nested_bh(&bpf_sp.bh_lock);
2050 return ret;
2051 }
2052
2053 static const struct bpf_func_proto bpf_csum_diff_proto = {
2054 .func = bpf_csum_diff,
2055 .gpl_only = false,
2056 .pkt_access = true,
2057 .ret_type = RET_INTEGER,
2058 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2059 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2060 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2061 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2062 .arg5_type = ARG_ANYTHING,
2063 };
2064
BPF_CALL_2(bpf_csum_update,struct sk_buff *,skb,__wsum,csum)2065 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2066 {
2067 /* The interface is to be used in combination with bpf_csum_diff()
2068 * for direct packet writes. csum rotation for alignment as well
2069 * as emulating csum_sub() can be done from the eBPF program.
2070 */
2071 if (skb->ip_summed == CHECKSUM_COMPLETE)
2072 return (skb->csum = csum_add(skb->csum, csum));
2073
2074 return -ENOTSUPP;
2075 }
2076
2077 static const struct bpf_func_proto bpf_csum_update_proto = {
2078 .func = bpf_csum_update,
2079 .gpl_only = false,
2080 .ret_type = RET_INTEGER,
2081 .arg1_type = ARG_PTR_TO_CTX,
2082 .arg2_type = ARG_ANYTHING,
2083 };
2084
BPF_CALL_2(bpf_csum_level,struct sk_buff *,skb,u64,level)2085 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2086 {
2087 /* The interface is to be used in combination with bpf_skb_adjust_room()
2088 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2089 * is passed as flags, for example.
2090 */
2091 switch (level) {
2092 case BPF_CSUM_LEVEL_INC:
2093 __skb_incr_checksum_unnecessary(skb);
2094 break;
2095 case BPF_CSUM_LEVEL_DEC:
2096 __skb_decr_checksum_unnecessary(skb);
2097 break;
2098 case BPF_CSUM_LEVEL_RESET:
2099 __skb_reset_checksum_unnecessary(skb);
2100 break;
2101 case BPF_CSUM_LEVEL_QUERY:
2102 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2103 skb->csum_level : -EACCES;
2104 default:
2105 return -EINVAL;
2106 }
2107
2108 return 0;
2109 }
2110
2111 static const struct bpf_func_proto bpf_csum_level_proto = {
2112 .func = bpf_csum_level,
2113 .gpl_only = false,
2114 .ret_type = RET_INTEGER,
2115 .arg1_type = ARG_PTR_TO_CTX,
2116 .arg2_type = ARG_ANYTHING,
2117 };
2118
__bpf_rx_skb(struct net_device * dev,struct sk_buff * skb)2119 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2120 {
2121 return dev_forward_skb_nomtu(dev, skb);
2122 }
2123
__bpf_rx_skb_no_mac(struct net_device * dev,struct sk_buff * skb)2124 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2125 struct sk_buff *skb)
2126 {
2127 int ret = ____dev_forward_skb(dev, skb, false);
2128
2129 if (likely(!ret)) {
2130 skb->dev = dev;
2131 ret = netif_rx(skb);
2132 }
2133
2134 return ret;
2135 }
2136
__bpf_tx_skb(struct net_device * dev,struct sk_buff * skb)2137 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2138 {
2139 int ret;
2140
2141 if (dev_xmit_recursion()) {
2142 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2143 kfree_skb(skb);
2144 return -ENETDOWN;
2145 }
2146
2147 skb->dev = dev;
2148 skb_set_redirected_noclear(skb, skb_at_tc_ingress(skb));
2149 skb_clear_tstamp(skb);
2150
2151 dev_xmit_recursion_inc();
2152 ret = dev_queue_xmit(skb);
2153 dev_xmit_recursion_dec();
2154
2155 return ret;
2156 }
2157
__bpf_redirect_no_mac(struct sk_buff * skb,struct net_device * dev,u32 flags)2158 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2159 u32 flags)
2160 {
2161 unsigned int mlen = skb_network_offset(skb);
2162
2163 if (unlikely(skb->len <= mlen)) {
2164 kfree_skb(skb);
2165 return -ERANGE;
2166 }
2167
2168 if (mlen) {
2169 __skb_pull(skb, mlen);
2170
2171 /* At ingress, the mac header has already been pulled once.
2172 * At egress, skb_pospull_rcsum has to be done in case that
2173 * the skb is originated from ingress (i.e. a forwarded skb)
2174 * to ensure that rcsum starts at net header.
2175 */
2176 if (!skb_at_tc_ingress(skb))
2177 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2178 }
2179 skb_pop_mac_header(skb);
2180 skb_reset_mac_len(skb);
2181 return flags & BPF_F_INGRESS ?
2182 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2183 }
2184
__bpf_redirect_common(struct sk_buff * skb,struct net_device * dev,u32 flags)2185 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2186 u32 flags)
2187 {
2188 /* Verify that a link layer header is carried */
2189 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2190 kfree_skb(skb);
2191 return -ERANGE;
2192 }
2193
2194 bpf_push_mac_rcsum(skb);
2195 return flags & BPF_F_INGRESS ?
2196 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2197 }
2198
__bpf_redirect(struct sk_buff * skb,struct net_device * dev,u32 flags)2199 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2200 u32 flags)
2201 {
2202 if (dev_is_mac_header_xmit(dev))
2203 return __bpf_redirect_common(skb, dev, flags);
2204 else
2205 return __bpf_redirect_no_mac(skb, dev, flags);
2206 }
2207
2208 #if IS_ENABLED(CONFIG_IPV6)
bpf_out_neigh_v6(struct net * net,struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2209 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2210 struct net_device *dev, struct bpf_nh_params *nh)
2211 {
2212 u32 hh_len = LL_RESERVED_SPACE(dev);
2213 const struct in6_addr *nexthop;
2214 struct dst_entry *dst = NULL;
2215 struct neighbour *neigh;
2216
2217 if (dev_xmit_recursion()) {
2218 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2219 goto out_drop;
2220 }
2221
2222 skb->dev = dev;
2223 skb_clear_tstamp(skb);
2224
2225 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2226 skb = skb_expand_head(skb, hh_len);
2227 if (!skb)
2228 return -ENOMEM;
2229 }
2230
2231 rcu_read_lock();
2232 if (!nh) {
2233 dst = skb_dst(skb);
2234 nexthop = rt6_nexthop(dst_rt6_info(dst),
2235 &ipv6_hdr(skb)->daddr);
2236 } else {
2237 nexthop = &nh->ipv6_nh;
2238 }
2239 neigh = ip_neigh_gw6(dev, nexthop);
2240 if (likely(!IS_ERR(neigh))) {
2241 int ret;
2242
2243 sock_confirm_neigh(skb, neigh);
2244 local_bh_disable();
2245 dev_xmit_recursion_inc();
2246 ret = neigh_output(neigh, skb, false);
2247 dev_xmit_recursion_dec();
2248 local_bh_enable();
2249 rcu_read_unlock();
2250 return ret;
2251 }
2252 rcu_read_unlock_bh();
2253 if (dst)
2254 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2255 out_drop:
2256 kfree_skb(skb);
2257 return -ENETDOWN;
2258 }
2259
__bpf_redirect_neigh_v6(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2260 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2261 struct bpf_nh_params *nh)
2262 {
2263 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2264 struct net *net = dev_net(dev);
2265 int err, ret = NET_XMIT_DROP;
2266
2267 if (!nh) {
2268 struct dst_entry *dst;
2269 struct flowi6 fl6 = {
2270 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2271 .flowi6_mark = skb->mark,
2272 .flowlabel = ip6_flowinfo(ip6h),
2273 .flowi6_oif = dev->ifindex,
2274 .flowi6_proto = ip6h->nexthdr,
2275 .daddr = ip6h->daddr,
2276 .saddr = ip6h->saddr,
2277 };
2278
2279 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2280 if (IS_ERR(dst))
2281 goto out_drop;
2282
2283 skb_dst_set(skb, dst);
2284 } else if (nh->nh_family != AF_INET6) {
2285 goto out_drop;
2286 }
2287
2288 err = bpf_out_neigh_v6(net, skb, dev, nh);
2289 if (unlikely(net_xmit_eval(err)))
2290 DEV_STATS_INC(dev, tx_errors);
2291 else
2292 ret = NET_XMIT_SUCCESS;
2293 goto out_xmit;
2294 out_drop:
2295 DEV_STATS_INC(dev, tx_errors);
2296 kfree_skb(skb);
2297 out_xmit:
2298 return ret;
2299 }
2300 #else
__bpf_redirect_neigh_v6(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2301 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2302 struct bpf_nh_params *nh)
2303 {
2304 kfree_skb(skb);
2305 return NET_XMIT_DROP;
2306 }
2307 #endif /* CONFIG_IPV6 */
2308
2309 #if IS_ENABLED(CONFIG_INET)
bpf_out_neigh_v4(struct net * net,struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2310 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2311 struct net_device *dev, struct bpf_nh_params *nh)
2312 {
2313 u32 hh_len = LL_RESERVED_SPACE(dev);
2314 struct neighbour *neigh;
2315 bool is_v6gw = false;
2316
2317 if (dev_xmit_recursion()) {
2318 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2319 goto out_drop;
2320 }
2321
2322 skb->dev = dev;
2323 skb_clear_tstamp(skb);
2324
2325 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2326 skb = skb_expand_head(skb, hh_len);
2327 if (!skb)
2328 return -ENOMEM;
2329 }
2330
2331 rcu_read_lock();
2332 if (!nh) {
2333 struct rtable *rt = skb_rtable(skb);
2334
2335 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2336 } else if (nh->nh_family == AF_INET6) {
2337 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2338 is_v6gw = true;
2339 } else if (nh->nh_family == AF_INET) {
2340 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2341 } else {
2342 rcu_read_unlock();
2343 goto out_drop;
2344 }
2345
2346 if (likely(!IS_ERR(neigh))) {
2347 int ret;
2348
2349 sock_confirm_neigh(skb, neigh);
2350 local_bh_disable();
2351 dev_xmit_recursion_inc();
2352 ret = neigh_output(neigh, skb, is_v6gw);
2353 dev_xmit_recursion_dec();
2354 local_bh_enable();
2355 rcu_read_unlock();
2356 return ret;
2357 }
2358 rcu_read_unlock();
2359 out_drop:
2360 kfree_skb(skb);
2361 return -ENETDOWN;
2362 }
2363
__bpf_redirect_neigh_v4(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2364 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2365 struct bpf_nh_params *nh)
2366 {
2367 const struct iphdr *ip4h = ip_hdr(skb);
2368 struct net *net = dev_net(dev);
2369 int err, ret = NET_XMIT_DROP;
2370
2371 if (!nh) {
2372 struct flowi4 fl4 = {
2373 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2374 .flowi4_mark = skb->mark,
2375 .flowi4_tos = ip4h->tos & INET_DSCP_MASK,
2376 .flowi4_oif = dev->ifindex,
2377 .flowi4_proto = ip4h->protocol,
2378 .daddr = ip4h->daddr,
2379 .saddr = ip4h->saddr,
2380 };
2381 struct rtable *rt;
2382
2383 rt = ip_route_output_flow(net, &fl4, NULL);
2384 if (IS_ERR(rt))
2385 goto out_drop;
2386 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2387 ip_rt_put(rt);
2388 goto out_drop;
2389 }
2390
2391 skb_dst_set(skb, &rt->dst);
2392 }
2393
2394 err = bpf_out_neigh_v4(net, skb, dev, nh);
2395 if (unlikely(net_xmit_eval(err)))
2396 DEV_STATS_INC(dev, tx_errors);
2397 else
2398 ret = NET_XMIT_SUCCESS;
2399 goto out_xmit;
2400 out_drop:
2401 DEV_STATS_INC(dev, tx_errors);
2402 kfree_skb(skb);
2403 out_xmit:
2404 return ret;
2405 }
2406 #else
__bpf_redirect_neigh_v4(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2407 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2408 struct bpf_nh_params *nh)
2409 {
2410 kfree_skb(skb);
2411 return NET_XMIT_DROP;
2412 }
2413 #endif /* CONFIG_INET */
2414
__bpf_redirect_neigh(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2415 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2416 struct bpf_nh_params *nh)
2417 {
2418 struct ethhdr *ethh = eth_hdr(skb);
2419
2420 if (unlikely(skb->mac_header >= skb->network_header))
2421 goto out;
2422 bpf_push_mac_rcsum(skb);
2423 if (is_multicast_ether_addr(ethh->h_dest))
2424 goto out;
2425
2426 skb_pull(skb, sizeof(*ethh));
2427 skb_unset_mac_header(skb);
2428 skb_reset_network_header(skb);
2429
2430 if (skb->protocol == htons(ETH_P_IP))
2431 return __bpf_redirect_neigh_v4(skb, dev, nh);
2432 else if (skb->protocol == htons(ETH_P_IPV6))
2433 return __bpf_redirect_neigh_v6(skb, dev, nh);
2434 out:
2435 kfree_skb(skb);
2436 return -ENOTSUPP;
2437 }
2438
2439 /* Internal, non-exposed redirect flags. */
2440 enum {
2441 BPF_F_NEIGH = (1ULL << 1),
2442 BPF_F_PEER = (1ULL << 2),
2443 BPF_F_NEXTHOP = (1ULL << 3),
2444 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2445 };
2446
BPF_CALL_3(bpf_clone_redirect,struct sk_buff *,skb,u32,ifindex,u64,flags)2447 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2448 {
2449 struct net_device *dev;
2450 struct sk_buff *clone;
2451 int ret;
2452
2453 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2454 return -EINVAL;
2455
2456 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2457 if (unlikely(!dev))
2458 return -EINVAL;
2459
2460 clone = skb_clone(skb, GFP_ATOMIC);
2461 if (unlikely(!clone))
2462 return -ENOMEM;
2463
2464 /* For direct write, we need to keep the invariant that the skbs
2465 * we're dealing with need to be uncloned. Should uncloning fail
2466 * here, we need to free the just generated clone to unclone once
2467 * again.
2468 */
2469 ret = bpf_try_make_head_writable(skb);
2470 if (unlikely(ret)) {
2471 kfree_skb(clone);
2472 return -ENOMEM;
2473 }
2474
2475 return __bpf_redirect(clone, dev, flags);
2476 }
2477
2478 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2479 .func = bpf_clone_redirect,
2480 .gpl_only = false,
2481 .ret_type = RET_INTEGER,
2482 .arg1_type = ARG_PTR_TO_CTX,
2483 .arg2_type = ARG_ANYTHING,
2484 .arg3_type = ARG_ANYTHING,
2485 };
2486
skb_get_peer_dev(struct net_device * dev)2487 static struct net_device *skb_get_peer_dev(struct net_device *dev)
2488 {
2489 const struct net_device_ops *ops = dev->netdev_ops;
2490
2491 if (likely(ops->ndo_get_peer_dev))
2492 return INDIRECT_CALL_1(ops->ndo_get_peer_dev,
2493 netkit_peer_dev, dev);
2494 return NULL;
2495 }
2496
skb_do_redirect(struct sk_buff * skb)2497 int skb_do_redirect(struct sk_buff *skb)
2498 {
2499 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
2500 struct net *net = dev_net(skb->dev);
2501 struct net_device *dev;
2502 u32 flags = ri->flags;
2503
2504 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2505 ri->tgt_index = 0;
2506 ri->flags = 0;
2507 if (unlikely(!dev))
2508 goto out_drop;
2509 if (flags & BPF_F_PEER) {
2510 if (unlikely(!skb_at_tc_ingress(skb)))
2511 goto out_drop;
2512 dev = skb_get_peer_dev(dev);
2513 if (unlikely(!dev ||
2514 !(dev->flags & IFF_UP) ||
2515 net_eq(net, dev_net(dev))))
2516 goto out_drop;
2517 skb->dev = dev;
2518 dev_sw_netstats_rx_add(dev, skb->len);
2519 return -EAGAIN;
2520 }
2521 return flags & BPF_F_NEIGH ?
2522 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2523 &ri->nh : NULL) :
2524 __bpf_redirect(skb, dev, flags);
2525 out_drop:
2526 kfree_skb(skb);
2527 return -EINVAL;
2528 }
2529
BPF_CALL_2(bpf_redirect,u32,ifindex,u64,flags)2530 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2531 {
2532 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
2533
2534 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2535 return TC_ACT_SHOT;
2536
2537 ri->flags = flags;
2538 ri->tgt_index = ifindex;
2539
2540 return TC_ACT_REDIRECT;
2541 }
2542
2543 static const struct bpf_func_proto bpf_redirect_proto = {
2544 .func = bpf_redirect,
2545 .gpl_only = false,
2546 .ret_type = RET_INTEGER,
2547 .arg1_type = ARG_ANYTHING,
2548 .arg2_type = ARG_ANYTHING,
2549 };
2550
BPF_CALL_2(bpf_redirect_peer,u32,ifindex,u64,flags)2551 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2552 {
2553 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
2554
2555 if (unlikely(flags))
2556 return TC_ACT_SHOT;
2557
2558 ri->flags = BPF_F_PEER;
2559 ri->tgt_index = ifindex;
2560
2561 return TC_ACT_REDIRECT;
2562 }
2563
2564 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2565 .func = bpf_redirect_peer,
2566 .gpl_only = false,
2567 .ret_type = RET_INTEGER,
2568 .arg1_type = ARG_ANYTHING,
2569 .arg2_type = ARG_ANYTHING,
2570 };
2571
BPF_CALL_4(bpf_redirect_neigh,u32,ifindex,struct bpf_redir_neigh *,params,int,plen,u64,flags)2572 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2573 int, plen, u64, flags)
2574 {
2575 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
2576
2577 if (unlikely((plen && plen < sizeof(*params)) || flags))
2578 return TC_ACT_SHOT;
2579
2580 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2581 ri->tgt_index = ifindex;
2582
2583 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2584 if (plen)
2585 memcpy(&ri->nh, params, sizeof(ri->nh));
2586
2587 return TC_ACT_REDIRECT;
2588 }
2589
2590 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2591 .func = bpf_redirect_neigh,
2592 .gpl_only = false,
2593 .ret_type = RET_INTEGER,
2594 .arg1_type = ARG_ANYTHING,
2595 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2596 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2597 .arg4_type = ARG_ANYTHING,
2598 };
2599
BPF_CALL_2(bpf_msg_apply_bytes,struct sk_msg *,msg,u32,bytes)2600 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2601 {
2602 msg->apply_bytes = bytes;
2603 return 0;
2604 }
2605
2606 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2607 .func = bpf_msg_apply_bytes,
2608 .gpl_only = false,
2609 .ret_type = RET_INTEGER,
2610 .arg1_type = ARG_PTR_TO_CTX,
2611 .arg2_type = ARG_ANYTHING,
2612 };
2613
BPF_CALL_2(bpf_msg_cork_bytes,struct sk_msg *,msg,u32,bytes)2614 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2615 {
2616 msg->cork_bytes = bytes;
2617 return 0;
2618 }
2619
sk_msg_reset_curr(struct sk_msg * msg)2620 static void sk_msg_reset_curr(struct sk_msg *msg)
2621 {
2622 u32 i = msg->sg.start;
2623 u32 len = 0;
2624
2625 do {
2626 len += sk_msg_elem(msg, i)->length;
2627 sk_msg_iter_var_next(i);
2628 if (len >= msg->sg.size)
2629 break;
2630 } while (i != msg->sg.end);
2631
2632 msg->sg.curr = i;
2633 msg->sg.copybreak = 0;
2634 }
2635
2636 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2637 .func = bpf_msg_cork_bytes,
2638 .gpl_only = false,
2639 .ret_type = RET_INTEGER,
2640 .arg1_type = ARG_PTR_TO_CTX,
2641 .arg2_type = ARG_ANYTHING,
2642 };
2643
BPF_CALL_4(bpf_msg_pull_data,struct sk_msg *,msg,u32,start,u32,end,u64,flags)2644 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2645 u32, end, u64, flags)
2646 {
2647 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2648 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2649 struct scatterlist *sge;
2650 u8 *raw, *to, *from;
2651 struct page *page;
2652
2653 if (unlikely(flags || end <= start))
2654 return -EINVAL;
2655
2656 /* First find the starting scatterlist element */
2657 i = msg->sg.start;
2658 do {
2659 offset += len;
2660 len = sk_msg_elem(msg, i)->length;
2661 if (start < offset + len)
2662 break;
2663 sk_msg_iter_var_next(i);
2664 } while (i != msg->sg.end);
2665
2666 if (unlikely(start >= offset + len))
2667 return -EINVAL;
2668
2669 first_sge = i;
2670 /* The start may point into the sg element so we need to also
2671 * account for the headroom.
2672 */
2673 bytes_sg_total = start - offset + bytes;
2674 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2675 goto out;
2676
2677 /* At this point we need to linearize multiple scatterlist
2678 * elements or a single shared page. Either way we need to
2679 * copy into a linear buffer exclusively owned by BPF. Then
2680 * place the buffer in the scatterlist and fixup the original
2681 * entries by removing the entries now in the linear buffer
2682 * and shifting the remaining entries. For now we do not try
2683 * to copy partial entries to avoid complexity of running out
2684 * of sg_entry slots. The downside is reading a single byte
2685 * will copy the entire sg entry.
2686 */
2687 do {
2688 copy += sk_msg_elem(msg, i)->length;
2689 sk_msg_iter_var_next(i);
2690 if (bytes_sg_total <= copy)
2691 break;
2692 } while (i != msg->sg.end);
2693 last_sge = i;
2694
2695 if (unlikely(bytes_sg_total > copy))
2696 return -EINVAL;
2697
2698 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2699 get_order(copy));
2700 if (unlikely(!page))
2701 return -ENOMEM;
2702
2703 raw = page_address(page);
2704 i = first_sge;
2705 do {
2706 sge = sk_msg_elem(msg, i);
2707 from = sg_virt(sge);
2708 len = sge->length;
2709 to = raw + poffset;
2710
2711 memcpy(to, from, len);
2712 poffset += len;
2713 sge->length = 0;
2714 put_page(sg_page(sge));
2715
2716 sk_msg_iter_var_next(i);
2717 } while (i != last_sge);
2718
2719 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2720
2721 /* To repair sg ring we need to shift entries. If we only
2722 * had a single entry though we can just replace it and
2723 * be done. Otherwise walk the ring and shift the entries.
2724 */
2725 WARN_ON_ONCE(last_sge == first_sge);
2726 shift = last_sge > first_sge ?
2727 last_sge - first_sge - 1 :
2728 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2729 if (!shift)
2730 goto out;
2731
2732 i = first_sge;
2733 sk_msg_iter_var_next(i);
2734 do {
2735 u32 move_from;
2736
2737 if (i + shift >= NR_MSG_FRAG_IDS)
2738 move_from = i + shift - NR_MSG_FRAG_IDS;
2739 else
2740 move_from = i + shift;
2741 if (move_from == msg->sg.end)
2742 break;
2743
2744 msg->sg.data[i] = msg->sg.data[move_from];
2745 msg->sg.data[move_from].length = 0;
2746 msg->sg.data[move_from].page_link = 0;
2747 msg->sg.data[move_from].offset = 0;
2748 sk_msg_iter_var_next(i);
2749 } while (1);
2750
2751 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2752 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2753 msg->sg.end - shift;
2754 out:
2755 sk_msg_reset_curr(msg);
2756 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2757 msg->data_end = msg->data + bytes;
2758 return 0;
2759 }
2760
2761 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2762 .func = bpf_msg_pull_data,
2763 .gpl_only = false,
2764 .ret_type = RET_INTEGER,
2765 .arg1_type = ARG_PTR_TO_CTX,
2766 .arg2_type = ARG_ANYTHING,
2767 .arg3_type = ARG_ANYTHING,
2768 .arg4_type = ARG_ANYTHING,
2769 };
2770
BPF_CALL_4(bpf_msg_push_data,struct sk_msg *,msg,u32,start,u32,len,u64,flags)2771 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2772 u32, len, u64, flags)
2773 {
2774 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2775 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2776 u8 *raw, *to, *from;
2777 struct page *page;
2778
2779 if (unlikely(flags))
2780 return -EINVAL;
2781
2782 if (unlikely(len == 0))
2783 return 0;
2784
2785 /* First find the starting scatterlist element */
2786 i = msg->sg.start;
2787 do {
2788 offset += l;
2789 l = sk_msg_elem(msg, i)->length;
2790
2791 if (start < offset + l)
2792 break;
2793 sk_msg_iter_var_next(i);
2794 } while (i != msg->sg.end);
2795
2796 if (start >= offset + l)
2797 return -EINVAL;
2798
2799 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2800
2801 /* If no space available will fallback to copy, we need at
2802 * least one scatterlist elem available to push data into
2803 * when start aligns to the beginning of an element or two
2804 * when it falls inside an element. We handle the start equals
2805 * offset case because its the common case for inserting a
2806 * header.
2807 */
2808 if (!space || (space == 1 && start != offset))
2809 copy = msg->sg.data[i].length;
2810
2811 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2812 get_order(copy + len));
2813 if (unlikely(!page))
2814 return -ENOMEM;
2815
2816 if (copy) {
2817 int front, back;
2818
2819 raw = page_address(page);
2820
2821 psge = sk_msg_elem(msg, i);
2822 front = start - offset;
2823 back = psge->length - front;
2824 from = sg_virt(psge);
2825
2826 if (front)
2827 memcpy(raw, from, front);
2828
2829 if (back) {
2830 from += front;
2831 to = raw + front + len;
2832
2833 memcpy(to, from, back);
2834 }
2835
2836 put_page(sg_page(psge));
2837 } else if (start - offset) {
2838 psge = sk_msg_elem(msg, i);
2839 rsge = sk_msg_elem_cpy(msg, i);
2840
2841 psge->length = start - offset;
2842 rsge.length -= psge->length;
2843 rsge.offset += start;
2844
2845 sk_msg_iter_var_next(i);
2846 sg_unmark_end(psge);
2847 sg_unmark_end(&rsge);
2848 sk_msg_iter_next(msg, end);
2849 }
2850
2851 /* Slot(s) to place newly allocated data */
2852 new = i;
2853
2854 /* Shift one or two slots as needed */
2855 if (!copy) {
2856 sge = sk_msg_elem_cpy(msg, i);
2857
2858 sk_msg_iter_var_next(i);
2859 sg_unmark_end(&sge);
2860 sk_msg_iter_next(msg, end);
2861
2862 nsge = sk_msg_elem_cpy(msg, i);
2863 if (rsge.length) {
2864 sk_msg_iter_var_next(i);
2865 nnsge = sk_msg_elem_cpy(msg, i);
2866 }
2867
2868 while (i != msg->sg.end) {
2869 msg->sg.data[i] = sge;
2870 sge = nsge;
2871 sk_msg_iter_var_next(i);
2872 if (rsge.length) {
2873 nsge = nnsge;
2874 nnsge = sk_msg_elem_cpy(msg, i);
2875 } else {
2876 nsge = sk_msg_elem_cpy(msg, i);
2877 }
2878 }
2879 }
2880
2881 /* Place newly allocated data buffer */
2882 sk_mem_charge(msg->sk, len);
2883 msg->sg.size += len;
2884 __clear_bit(new, msg->sg.copy);
2885 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2886 if (rsge.length) {
2887 get_page(sg_page(&rsge));
2888 sk_msg_iter_var_next(new);
2889 msg->sg.data[new] = rsge;
2890 }
2891
2892 sk_msg_reset_curr(msg);
2893 sk_msg_compute_data_pointers(msg);
2894 return 0;
2895 }
2896
2897 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2898 .func = bpf_msg_push_data,
2899 .gpl_only = false,
2900 .ret_type = RET_INTEGER,
2901 .arg1_type = ARG_PTR_TO_CTX,
2902 .arg2_type = ARG_ANYTHING,
2903 .arg3_type = ARG_ANYTHING,
2904 .arg4_type = ARG_ANYTHING,
2905 };
2906
sk_msg_shift_left(struct sk_msg * msg,int i)2907 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2908 {
2909 int prev;
2910
2911 do {
2912 prev = i;
2913 sk_msg_iter_var_next(i);
2914 msg->sg.data[prev] = msg->sg.data[i];
2915 } while (i != msg->sg.end);
2916
2917 sk_msg_iter_prev(msg, end);
2918 }
2919
sk_msg_shift_right(struct sk_msg * msg,int i)2920 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2921 {
2922 struct scatterlist tmp, sge;
2923
2924 sk_msg_iter_next(msg, end);
2925 sge = sk_msg_elem_cpy(msg, i);
2926 sk_msg_iter_var_next(i);
2927 tmp = sk_msg_elem_cpy(msg, i);
2928
2929 while (i != msg->sg.end) {
2930 msg->sg.data[i] = sge;
2931 sk_msg_iter_var_next(i);
2932 sge = tmp;
2933 tmp = sk_msg_elem_cpy(msg, i);
2934 }
2935 }
2936
BPF_CALL_4(bpf_msg_pop_data,struct sk_msg *,msg,u32,start,u32,len,u64,flags)2937 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2938 u32, len, u64, flags)
2939 {
2940 u32 i = 0, l = 0, space, offset = 0;
2941 u64 last = start + len;
2942 int pop;
2943
2944 if (unlikely(flags))
2945 return -EINVAL;
2946
2947 /* First find the starting scatterlist element */
2948 i = msg->sg.start;
2949 do {
2950 offset += l;
2951 l = sk_msg_elem(msg, i)->length;
2952
2953 if (start < offset + l)
2954 break;
2955 sk_msg_iter_var_next(i);
2956 } while (i != msg->sg.end);
2957
2958 /* Bounds checks: start and pop must be inside message */
2959 if (start >= offset + l || last >= msg->sg.size)
2960 return -EINVAL;
2961
2962 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2963
2964 pop = len;
2965 /* --------------| offset
2966 * -| start |-------- len -------|
2967 *
2968 * |----- a ----|-------- pop -------|----- b ----|
2969 * |______________________________________________| length
2970 *
2971 *
2972 * a: region at front of scatter element to save
2973 * b: region at back of scatter element to save when length > A + pop
2974 * pop: region to pop from element, same as input 'pop' here will be
2975 * decremented below per iteration.
2976 *
2977 * Two top-level cases to handle when start != offset, first B is non
2978 * zero and second B is zero corresponding to when a pop includes more
2979 * than one element.
2980 *
2981 * Then if B is non-zero AND there is no space allocate space and
2982 * compact A, B regions into page. If there is space shift ring to
2983 * the right free'ing the next element in ring to place B, leaving
2984 * A untouched except to reduce length.
2985 */
2986 if (start != offset) {
2987 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2988 int a = start;
2989 int b = sge->length - pop - a;
2990
2991 sk_msg_iter_var_next(i);
2992
2993 if (pop < sge->length - a) {
2994 if (space) {
2995 sge->length = a;
2996 sk_msg_shift_right(msg, i);
2997 nsge = sk_msg_elem(msg, i);
2998 get_page(sg_page(sge));
2999 sg_set_page(nsge,
3000 sg_page(sge),
3001 b, sge->offset + pop + a);
3002 } else {
3003 struct page *page, *orig;
3004 u8 *to, *from;
3005
3006 page = alloc_pages(__GFP_NOWARN |
3007 __GFP_COMP | GFP_ATOMIC,
3008 get_order(a + b));
3009 if (unlikely(!page))
3010 return -ENOMEM;
3011
3012 sge->length = a;
3013 orig = sg_page(sge);
3014 from = sg_virt(sge);
3015 to = page_address(page);
3016 memcpy(to, from, a);
3017 memcpy(to + a, from + a + pop, b);
3018 sg_set_page(sge, page, a + b, 0);
3019 put_page(orig);
3020 }
3021 pop = 0;
3022 } else if (pop >= sge->length - a) {
3023 pop -= (sge->length - a);
3024 sge->length = a;
3025 }
3026 }
3027
3028 /* From above the current layout _must_ be as follows,
3029 *
3030 * -| offset
3031 * -| start
3032 *
3033 * |---- pop ---|---------------- b ------------|
3034 * |____________________________________________| length
3035 *
3036 * Offset and start of the current msg elem are equal because in the
3037 * previous case we handled offset != start and either consumed the
3038 * entire element and advanced to the next element OR pop == 0.
3039 *
3040 * Two cases to handle here are first pop is less than the length
3041 * leaving some remainder b above. Simply adjust the element's layout
3042 * in this case. Or pop >= length of the element so that b = 0. In this
3043 * case advance to next element decrementing pop.
3044 */
3045 while (pop) {
3046 struct scatterlist *sge = sk_msg_elem(msg, i);
3047
3048 if (pop < sge->length) {
3049 sge->length -= pop;
3050 sge->offset += pop;
3051 pop = 0;
3052 } else {
3053 pop -= sge->length;
3054 sk_msg_shift_left(msg, i);
3055 }
3056 sk_msg_iter_var_next(i);
3057 }
3058
3059 sk_mem_uncharge(msg->sk, len - pop);
3060 msg->sg.size -= (len - pop);
3061 sk_msg_reset_curr(msg);
3062 sk_msg_compute_data_pointers(msg);
3063 return 0;
3064 }
3065
3066 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3067 .func = bpf_msg_pop_data,
3068 .gpl_only = false,
3069 .ret_type = RET_INTEGER,
3070 .arg1_type = ARG_PTR_TO_CTX,
3071 .arg2_type = ARG_ANYTHING,
3072 .arg3_type = ARG_ANYTHING,
3073 .arg4_type = ARG_ANYTHING,
3074 };
3075
3076 #ifdef CONFIG_CGROUP_NET_CLASSID
BPF_CALL_0(bpf_get_cgroup_classid_curr)3077 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3078 {
3079 return __task_get_classid(current);
3080 }
3081
3082 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3083 .func = bpf_get_cgroup_classid_curr,
3084 .gpl_only = false,
3085 .ret_type = RET_INTEGER,
3086 };
3087
BPF_CALL_1(bpf_skb_cgroup_classid,const struct sk_buff *,skb)3088 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3089 {
3090 struct sock *sk = skb_to_full_sk(skb);
3091
3092 if (!sk || !sk_fullsock(sk))
3093 return 0;
3094
3095 return sock_cgroup_classid(&sk->sk_cgrp_data);
3096 }
3097
3098 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3099 .func = bpf_skb_cgroup_classid,
3100 .gpl_only = false,
3101 .ret_type = RET_INTEGER,
3102 .arg1_type = ARG_PTR_TO_CTX,
3103 };
3104 #endif
3105
BPF_CALL_1(bpf_get_cgroup_classid,const struct sk_buff *,skb)3106 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3107 {
3108 return task_get_classid(skb);
3109 }
3110
3111 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3112 .func = bpf_get_cgroup_classid,
3113 .gpl_only = false,
3114 .ret_type = RET_INTEGER,
3115 .arg1_type = ARG_PTR_TO_CTX,
3116 };
3117
BPF_CALL_1(bpf_get_route_realm,const struct sk_buff *,skb)3118 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3119 {
3120 return dst_tclassid(skb);
3121 }
3122
3123 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3124 .func = bpf_get_route_realm,
3125 .gpl_only = false,
3126 .ret_type = RET_INTEGER,
3127 .arg1_type = ARG_PTR_TO_CTX,
3128 };
3129
BPF_CALL_1(bpf_get_hash_recalc,struct sk_buff *,skb)3130 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3131 {
3132 /* If skb_clear_hash() was called due to mangling, we can
3133 * trigger SW recalculation here. Later access to hash
3134 * can then use the inline skb->hash via context directly
3135 * instead of calling this helper again.
3136 */
3137 return skb_get_hash(skb);
3138 }
3139
3140 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3141 .func = bpf_get_hash_recalc,
3142 .gpl_only = false,
3143 .ret_type = RET_INTEGER,
3144 .arg1_type = ARG_PTR_TO_CTX,
3145 };
3146
BPF_CALL_1(bpf_set_hash_invalid,struct sk_buff *,skb)3147 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3148 {
3149 /* After all direct packet write, this can be used once for
3150 * triggering a lazy recalc on next skb_get_hash() invocation.
3151 */
3152 skb_clear_hash(skb);
3153 return 0;
3154 }
3155
3156 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3157 .func = bpf_set_hash_invalid,
3158 .gpl_only = false,
3159 .ret_type = RET_INTEGER,
3160 .arg1_type = ARG_PTR_TO_CTX,
3161 };
3162
BPF_CALL_2(bpf_set_hash,struct sk_buff *,skb,u32,hash)3163 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3164 {
3165 /* Set user specified hash as L4(+), so that it gets returned
3166 * on skb_get_hash() call unless BPF prog later on triggers a
3167 * skb_clear_hash().
3168 */
3169 __skb_set_sw_hash(skb, hash, true);
3170 return 0;
3171 }
3172
3173 static const struct bpf_func_proto bpf_set_hash_proto = {
3174 .func = bpf_set_hash,
3175 .gpl_only = false,
3176 .ret_type = RET_INTEGER,
3177 .arg1_type = ARG_PTR_TO_CTX,
3178 .arg2_type = ARG_ANYTHING,
3179 };
3180
BPF_CALL_3(bpf_skb_vlan_push,struct sk_buff *,skb,__be16,vlan_proto,u16,vlan_tci)3181 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3182 u16, vlan_tci)
3183 {
3184 int ret;
3185
3186 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3187 vlan_proto != htons(ETH_P_8021AD)))
3188 vlan_proto = htons(ETH_P_8021Q);
3189
3190 bpf_push_mac_rcsum(skb);
3191 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3192 bpf_pull_mac_rcsum(skb);
3193 skb_reset_mac_len(skb);
3194
3195 bpf_compute_data_pointers(skb);
3196 return ret;
3197 }
3198
3199 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3200 .func = bpf_skb_vlan_push,
3201 .gpl_only = false,
3202 .ret_type = RET_INTEGER,
3203 .arg1_type = ARG_PTR_TO_CTX,
3204 .arg2_type = ARG_ANYTHING,
3205 .arg3_type = ARG_ANYTHING,
3206 };
3207
BPF_CALL_1(bpf_skb_vlan_pop,struct sk_buff *,skb)3208 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3209 {
3210 int ret;
3211
3212 bpf_push_mac_rcsum(skb);
3213 ret = skb_vlan_pop(skb);
3214 bpf_pull_mac_rcsum(skb);
3215
3216 bpf_compute_data_pointers(skb);
3217 return ret;
3218 }
3219
3220 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3221 .func = bpf_skb_vlan_pop,
3222 .gpl_only = false,
3223 .ret_type = RET_INTEGER,
3224 .arg1_type = ARG_PTR_TO_CTX,
3225 };
3226
bpf_skb_generic_push(struct sk_buff * skb,u32 off,u32 len)3227 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3228 {
3229 /* Caller already did skb_cow() with len as headroom,
3230 * so no need to do it here.
3231 */
3232 skb_push(skb, len);
3233 memmove(skb->data, skb->data + len, off);
3234 memset(skb->data + off, 0, len);
3235
3236 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3237 * needed here as it does not change the skb->csum
3238 * result for checksum complete when summing over
3239 * zeroed blocks.
3240 */
3241 return 0;
3242 }
3243
bpf_skb_generic_pop(struct sk_buff * skb,u32 off,u32 len)3244 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3245 {
3246 void *old_data;
3247
3248 /* skb_ensure_writable() is not needed here, as we're
3249 * already working on an uncloned skb.
3250 */
3251 if (unlikely(!pskb_may_pull(skb, off + len)))
3252 return -ENOMEM;
3253
3254 old_data = skb->data;
3255 __skb_pull(skb, len);
3256 skb_postpull_rcsum(skb, old_data + off, len);
3257 memmove(skb->data, old_data, off);
3258
3259 return 0;
3260 }
3261
bpf_skb_net_hdr_push(struct sk_buff * skb,u32 off,u32 len)3262 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3263 {
3264 bool trans_same = skb->transport_header == skb->network_header;
3265 int ret;
3266
3267 /* There's no need for __skb_push()/__skb_pull() pair to
3268 * get to the start of the mac header as we're guaranteed
3269 * to always start from here under eBPF.
3270 */
3271 ret = bpf_skb_generic_push(skb, off, len);
3272 if (likely(!ret)) {
3273 skb->mac_header -= len;
3274 skb->network_header -= len;
3275 if (trans_same)
3276 skb->transport_header = skb->network_header;
3277 }
3278
3279 return ret;
3280 }
3281
bpf_skb_net_hdr_pop(struct sk_buff * skb,u32 off,u32 len)3282 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3283 {
3284 bool trans_same = skb->transport_header == skb->network_header;
3285 int ret;
3286
3287 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3288 ret = bpf_skb_generic_pop(skb, off, len);
3289 if (likely(!ret)) {
3290 skb->mac_header += len;
3291 skb->network_header += len;
3292 if (trans_same)
3293 skb->transport_header = skb->network_header;
3294 }
3295
3296 return ret;
3297 }
3298
bpf_skb_proto_4_to_6(struct sk_buff * skb)3299 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3300 {
3301 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3302 u32 off = skb_mac_header_len(skb);
3303 int ret;
3304
3305 ret = skb_cow(skb, len_diff);
3306 if (unlikely(ret < 0))
3307 return ret;
3308
3309 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3310 if (unlikely(ret < 0))
3311 return ret;
3312
3313 if (skb_is_gso(skb)) {
3314 struct skb_shared_info *shinfo = skb_shinfo(skb);
3315
3316 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3317 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3318 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3319 shinfo->gso_type |= SKB_GSO_TCPV6;
3320 }
3321 }
3322
3323 skb->protocol = htons(ETH_P_IPV6);
3324 skb_clear_hash(skb);
3325
3326 return 0;
3327 }
3328
bpf_skb_proto_6_to_4(struct sk_buff * skb)3329 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3330 {
3331 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3332 u32 off = skb_mac_header_len(skb);
3333 int ret;
3334
3335 ret = skb_unclone(skb, GFP_ATOMIC);
3336 if (unlikely(ret < 0))
3337 return ret;
3338
3339 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3340 if (unlikely(ret < 0))
3341 return ret;
3342
3343 if (skb_is_gso(skb)) {
3344 struct skb_shared_info *shinfo = skb_shinfo(skb);
3345
3346 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3347 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3348 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3349 shinfo->gso_type |= SKB_GSO_TCPV4;
3350 }
3351 }
3352
3353 skb->protocol = htons(ETH_P_IP);
3354 skb_clear_hash(skb);
3355
3356 return 0;
3357 }
3358
bpf_skb_proto_xlat(struct sk_buff * skb,__be16 to_proto)3359 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3360 {
3361 __be16 from_proto = skb->protocol;
3362
3363 if (from_proto == htons(ETH_P_IP) &&
3364 to_proto == htons(ETH_P_IPV6))
3365 return bpf_skb_proto_4_to_6(skb);
3366
3367 if (from_proto == htons(ETH_P_IPV6) &&
3368 to_proto == htons(ETH_P_IP))
3369 return bpf_skb_proto_6_to_4(skb);
3370
3371 return -ENOTSUPP;
3372 }
3373
BPF_CALL_3(bpf_skb_change_proto,struct sk_buff *,skb,__be16,proto,u64,flags)3374 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3375 u64, flags)
3376 {
3377 int ret;
3378
3379 if (unlikely(flags))
3380 return -EINVAL;
3381
3382 /* General idea is that this helper does the basic groundwork
3383 * needed for changing the protocol, and eBPF program fills the
3384 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3385 * and other helpers, rather than passing a raw buffer here.
3386 *
3387 * The rationale is to keep this minimal and without a need to
3388 * deal with raw packet data. F.e. even if we would pass buffers
3389 * here, the program still needs to call the bpf_lX_csum_replace()
3390 * helpers anyway. Plus, this way we keep also separation of
3391 * concerns, since f.e. bpf_skb_store_bytes() should only take
3392 * care of stores.
3393 *
3394 * Currently, additional options and extension header space are
3395 * not supported, but flags register is reserved so we can adapt
3396 * that. For offloads, we mark packet as dodgy, so that headers
3397 * need to be verified first.
3398 */
3399 ret = bpf_skb_proto_xlat(skb, proto);
3400 bpf_compute_data_pointers(skb);
3401 return ret;
3402 }
3403
3404 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3405 .func = bpf_skb_change_proto,
3406 .gpl_only = false,
3407 .ret_type = RET_INTEGER,
3408 .arg1_type = ARG_PTR_TO_CTX,
3409 .arg2_type = ARG_ANYTHING,
3410 .arg3_type = ARG_ANYTHING,
3411 };
3412
BPF_CALL_2(bpf_skb_change_type,struct sk_buff *,skb,u32,pkt_type)3413 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3414 {
3415 /* We only allow a restricted subset to be changed for now. */
3416 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3417 !skb_pkt_type_ok(pkt_type)))
3418 return -EINVAL;
3419
3420 skb->pkt_type = pkt_type;
3421 return 0;
3422 }
3423
3424 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3425 .func = bpf_skb_change_type,
3426 .gpl_only = false,
3427 .ret_type = RET_INTEGER,
3428 .arg1_type = ARG_PTR_TO_CTX,
3429 .arg2_type = ARG_ANYTHING,
3430 };
3431
bpf_skb_net_base_len(const struct sk_buff * skb)3432 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3433 {
3434 switch (skb->protocol) {
3435 case htons(ETH_P_IP):
3436 return sizeof(struct iphdr);
3437 case htons(ETH_P_IPV6):
3438 return sizeof(struct ipv6hdr);
3439 default:
3440 return ~0U;
3441 }
3442 }
3443
3444 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3445 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3446
3447 #define BPF_F_ADJ_ROOM_DECAP_L3_MASK (BPF_F_ADJ_ROOM_DECAP_L3_IPV4 | \
3448 BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3449
3450 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3451 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3452 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3453 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3454 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3455 BPF_F_ADJ_ROOM_ENCAP_L2( \
3456 BPF_ADJ_ROOM_ENCAP_L2_MASK) | \
3457 BPF_F_ADJ_ROOM_DECAP_L3_MASK)
3458
bpf_skb_net_grow(struct sk_buff * skb,u32 off,u32 len_diff,u64 flags)3459 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3460 u64 flags)
3461 {
3462 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3463 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3464 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3465 unsigned int gso_type = SKB_GSO_DODGY;
3466 int ret;
3467
3468 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3469 /* udp gso_size delineates datagrams, only allow if fixed */
3470 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3471 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3472 return -ENOTSUPP;
3473 }
3474
3475 ret = skb_cow_head(skb, len_diff);
3476 if (unlikely(ret < 0))
3477 return ret;
3478
3479 if (encap) {
3480 if (skb->protocol != htons(ETH_P_IP) &&
3481 skb->protocol != htons(ETH_P_IPV6))
3482 return -ENOTSUPP;
3483
3484 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3485 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3486 return -EINVAL;
3487
3488 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3489 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3490 return -EINVAL;
3491
3492 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3493 inner_mac_len < ETH_HLEN)
3494 return -EINVAL;
3495
3496 if (skb->encapsulation)
3497 return -EALREADY;
3498
3499 mac_len = skb->network_header - skb->mac_header;
3500 inner_net = skb->network_header;
3501 if (inner_mac_len > len_diff)
3502 return -EINVAL;
3503 inner_trans = skb->transport_header;
3504 }
3505
3506 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3507 if (unlikely(ret < 0))
3508 return ret;
3509
3510 if (encap) {
3511 skb->inner_mac_header = inner_net - inner_mac_len;
3512 skb->inner_network_header = inner_net;
3513 skb->inner_transport_header = inner_trans;
3514
3515 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3516 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3517 else
3518 skb_set_inner_protocol(skb, skb->protocol);
3519
3520 skb->encapsulation = 1;
3521 skb_set_network_header(skb, mac_len);
3522
3523 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3524 gso_type |= SKB_GSO_UDP_TUNNEL;
3525 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3526 gso_type |= SKB_GSO_GRE;
3527 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3528 gso_type |= SKB_GSO_IPXIP6;
3529 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3530 gso_type |= SKB_GSO_IPXIP4;
3531
3532 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3533 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3534 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3535 sizeof(struct ipv6hdr) :
3536 sizeof(struct iphdr);
3537
3538 skb_set_transport_header(skb, mac_len + nh_len);
3539 }
3540
3541 /* Match skb->protocol to new outer l3 protocol */
3542 if (skb->protocol == htons(ETH_P_IP) &&
3543 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3544 skb->protocol = htons(ETH_P_IPV6);
3545 else if (skb->protocol == htons(ETH_P_IPV6) &&
3546 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3547 skb->protocol = htons(ETH_P_IP);
3548 }
3549
3550 if (skb_is_gso(skb)) {
3551 struct skb_shared_info *shinfo = skb_shinfo(skb);
3552
3553 /* Header must be checked, and gso_segs recomputed. */
3554 shinfo->gso_type |= gso_type;
3555 shinfo->gso_segs = 0;
3556
3557 /* Due to header growth, MSS needs to be downgraded.
3558 * There is a BUG_ON() when segmenting the frag_list with
3559 * head_frag true, so linearize the skb after downgrading
3560 * the MSS.
3561 */
3562 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) {
3563 skb_decrease_gso_size(shinfo, len_diff);
3564 if (shinfo->frag_list)
3565 return skb_linearize(skb);
3566 }
3567 }
3568
3569 return 0;
3570 }
3571
bpf_skb_net_shrink(struct sk_buff * skb,u32 off,u32 len_diff,u64 flags)3572 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3573 u64 flags)
3574 {
3575 int ret;
3576
3577 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3578 BPF_F_ADJ_ROOM_DECAP_L3_MASK |
3579 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3580 return -EINVAL;
3581
3582 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3583 /* udp gso_size delineates datagrams, only allow if fixed */
3584 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3585 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3586 return -ENOTSUPP;
3587 }
3588
3589 ret = skb_unclone(skb, GFP_ATOMIC);
3590 if (unlikely(ret < 0))
3591 return ret;
3592
3593 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3594 if (unlikely(ret < 0))
3595 return ret;
3596
3597 /* Match skb->protocol to new outer l3 protocol */
3598 if (skb->protocol == htons(ETH_P_IP) &&
3599 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3600 skb->protocol = htons(ETH_P_IPV6);
3601 else if (skb->protocol == htons(ETH_P_IPV6) &&
3602 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV4)
3603 skb->protocol = htons(ETH_P_IP);
3604
3605 if (skb_is_gso(skb)) {
3606 struct skb_shared_info *shinfo = skb_shinfo(skb);
3607
3608 /* Due to header shrink, MSS can be upgraded. */
3609 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3610 skb_increase_gso_size(shinfo, len_diff);
3611
3612 /* Header must be checked, and gso_segs recomputed. */
3613 shinfo->gso_type |= SKB_GSO_DODGY;
3614 shinfo->gso_segs = 0;
3615 }
3616
3617 return 0;
3618 }
3619
3620 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3621
BPF_CALL_4(sk_skb_adjust_room,struct sk_buff *,skb,s32,len_diff,u32,mode,u64,flags)3622 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3623 u32, mode, u64, flags)
3624 {
3625 u32 len_diff_abs = abs(len_diff);
3626 bool shrink = len_diff < 0;
3627 int ret = 0;
3628
3629 if (unlikely(flags || mode))
3630 return -EINVAL;
3631 if (unlikely(len_diff_abs > 0xfffU))
3632 return -EFAULT;
3633
3634 if (!shrink) {
3635 ret = skb_cow(skb, len_diff);
3636 if (unlikely(ret < 0))
3637 return ret;
3638 __skb_push(skb, len_diff_abs);
3639 memset(skb->data, 0, len_diff_abs);
3640 } else {
3641 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3642 return -ENOMEM;
3643 __skb_pull(skb, len_diff_abs);
3644 }
3645 if (tls_sw_has_ctx_rx(skb->sk)) {
3646 struct strp_msg *rxm = strp_msg(skb);
3647
3648 rxm->full_len += len_diff;
3649 }
3650 return ret;
3651 }
3652
3653 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3654 .func = sk_skb_adjust_room,
3655 .gpl_only = false,
3656 .ret_type = RET_INTEGER,
3657 .arg1_type = ARG_PTR_TO_CTX,
3658 .arg2_type = ARG_ANYTHING,
3659 .arg3_type = ARG_ANYTHING,
3660 .arg4_type = ARG_ANYTHING,
3661 };
3662
BPF_CALL_4(bpf_skb_adjust_room,struct sk_buff *,skb,s32,len_diff,u32,mode,u64,flags)3663 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3664 u32, mode, u64, flags)
3665 {
3666 u32 len_cur, len_diff_abs = abs(len_diff);
3667 u32 len_min = bpf_skb_net_base_len(skb);
3668 u32 len_max = BPF_SKB_MAX_LEN;
3669 __be16 proto = skb->protocol;
3670 bool shrink = len_diff < 0;
3671 u32 off;
3672 int ret;
3673
3674 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3675 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3676 return -EINVAL;
3677 if (unlikely(len_diff_abs > 0xfffU))
3678 return -EFAULT;
3679 if (unlikely(proto != htons(ETH_P_IP) &&
3680 proto != htons(ETH_P_IPV6)))
3681 return -ENOTSUPP;
3682
3683 off = skb_mac_header_len(skb);
3684 switch (mode) {
3685 case BPF_ADJ_ROOM_NET:
3686 off += bpf_skb_net_base_len(skb);
3687 break;
3688 case BPF_ADJ_ROOM_MAC:
3689 break;
3690 default:
3691 return -ENOTSUPP;
3692 }
3693
3694 if (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3695 if (!shrink)
3696 return -EINVAL;
3697
3698 switch (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3699 case BPF_F_ADJ_ROOM_DECAP_L3_IPV4:
3700 len_min = sizeof(struct iphdr);
3701 break;
3702 case BPF_F_ADJ_ROOM_DECAP_L3_IPV6:
3703 len_min = sizeof(struct ipv6hdr);
3704 break;
3705 default:
3706 return -EINVAL;
3707 }
3708 }
3709
3710 len_cur = skb->len - skb_network_offset(skb);
3711 if ((shrink && (len_diff_abs >= len_cur ||
3712 len_cur - len_diff_abs < len_min)) ||
3713 (!shrink && (skb->len + len_diff_abs > len_max &&
3714 !skb_is_gso(skb))))
3715 return -ENOTSUPP;
3716
3717 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3718 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3719 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3720 __skb_reset_checksum_unnecessary(skb);
3721
3722 bpf_compute_data_pointers(skb);
3723 return ret;
3724 }
3725
3726 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3727 .func = bpf_skb_adjust_room,
3728 .gpl_only = false,
3729 .ret_type = RET_INTEGER,
3730 .arg1_type = ARG_PTR_TO_CTX,
3731 .arg2_type = ARG_ANYTHING,
3732 .arg3_type = ARG_ANYTHING,
3733 .arg4_type = ARG_ANYTHING,
3734 };
3735
__bpf_skb_min_len(const struct sk_buff * skb)3736 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3737 {
3738 u32 min_len = skb_network_offset(skb);
3739
3740 if (skb_transport_header_was_set(skb))
3741 min_len = skb_transport_offset(skb);
3742 if (skb->ip_summed == CHECKSUM_PARTIAL)
3743 min_len = skb_checksum_start_offset(skb) +
3744 skb->csum_offset + sizeof(__sum16);
3745 return min_len;
3746 }
3747
bpf_skb_grow_rcsum(struct sk_buff * skb,unsigned int new_len)3748 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3749 {
3750 unsigned int old_len = skb->len;
3751 int ret;
3752
3753 ret = __skb_grow_rcsum(skb, new_len);
3754 if (!ret)
3755 memset(skb->data + old_len, 0, new_len - old_len);
3756 return ret;
3757 }
3758
bpf_skb_trim_rcsum(struct sk_buff * skb,unsigned int new_len)3759 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3760 {
3761 return __skb_trim_rcsum(skb, new_len);
3762 }
3763
__bpf_skb_change_tail(struct sk_buff * skb,u32 new_len,u64 flags)3764 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3765 u64 flags)
3766 {
3767 u32 max_len = BPF_SKB_MAX_LEN;
3768 u32 min_len = __bpf_skb_min_len(skb);
3769 int ret;
3770
3771 if (unlikely(flags || new_len > max_len || new_len < min_len))
3772 return -EINVAL;
3773 if (skb->encapsulation)
3774 return -ENOTSUPP;
3775
3776 /* The basic idea of this helper is that it's performing the
3777 * needed work to either grow or trim an skb, and eBPF program
3778 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3779 * bpf_lX_csum_replace() and others rather than passing a raw
3780 * buffer here. This one is a slow path helper and intended
3781 * for replies with control messages.
3782 *
3783 * Like in bpf_skb_change_proto(), we want to keep this rather
3784 * minimal and without protocol specifics so that we are able
3785 * to separate concerns as in bpf_skb_store_bytes() should only
3786 * be the one responsible for writing buffers.
3787 *
3788 * It's really expected to be a slow path operation here for
3789 * control message replies, so we're implicitly linearizing,
3790 * uncloning and drop offloads from the skb by this.
3791 */
3792 ret = __bpf_try_make_writable(skb, skb->len);
3793 if (!ret) {
3794 if (new_len > skb->len)
3795 ret = bpf_skb_grow_rcsum(skb, new_len);
3796 else if (new_len < skb->len)
3797 ret = bpf_skb_trim_rcsum(skb, new_len);
3798 if (!ret && skb_is_gso(skb))
3799 skb_gso_reset(skb);
3800 }
3801 return ret;
3802 }
3803
BPF_CALL_3(bpf_skb_change_tail,struct sk_buff *,skb,u32,new_len,u64,flags)3804 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3805 u64, flags)
3806 {
3807 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3808
3809 bpf_compute_data_pointers(skb);
3810 return ret;
3811 }
3812
3813 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3814 .func = bpf_skb_change_tail,
3815 .gpl_only = false,
3816 .ret_type = RET_INTEGER,
3817 .arg1_type = ARG_PTR_TO_CTX,
3818 .arg2_type = ARG_ANYTHING,
3819 .arg3_type = ARG_ANYTHING,
3820 };
3821
BPF_CALL_3(sk_skb_change_tail,struct sk_buff *,skb,u32,new_len,u64,flags)3822 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3823 u64, flags)
3824 {
3825 return __bpf_skb_change_tail(skb, new_len, flags);
3826 }
3827
3828 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3829 .func = sk_skb_change_tail,
3830 .gpl_only = false,
3831 .ret_type = RET_INTEGER,
3832 .arg1_type = ARG_PTR_TO_CTX,
3833 .arg2_type = ARG_ANYTHING,
3834 .arg3_type = ARG_ANYTHING,
3835 };
3836
__bpf_skb_change_head(struct sk_buff * skb,u32 head_room,u64 flags)3837 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3838 u64 flags)
3839 {
3840 u32 max_len = BPF_SKB_MAX_LEN;
3841 u32 new_len = skb->len + head_room;
3842 int ret;
3843
3844 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3845 new_len < skb->len))
3846 return -EINVAL;
3847
3848 ret = skb_cow(skb, head_room);
3849 if (likely(!ret)) {
3850 /* Idea for this helper is that we currently only
3851 * allow to expand on mac header. This means that
3852 * skb->protocol network header, etc, stay as is.
3853 * Compared to bpf_skb_change_tail(), we're more
3854 * flexible due to not needing to linearize or
3855 * reset GSO. Intention for this helper is to be
3856 * used by an L3 skb that needs to push mac header
3857 * for redirection into L2 device.
3858 */
3859 __skb_push(skb, head_room);
3860 memset(skb->data, 0, head_room);
3861 skb_reset_mac_header(skb);
3862 skb_reset_mac_len(skb);
3863 }
3864
3865 return ret;
3866 }
3867
BPF_CALL_3(bpf_skb_change_head,struct sk_buff *,skb,u32,head_room,u64,flags)3868 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3869 u64, flags)
3870 {
3871 int ret = __bpf_skb_change_head(skb, head_room, flags);
3872
3873 bpf_compute_data_pointers(skb);
3874 return ret;
3875 }
3876
3877 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3878 .func = bpf_skb_change_head,
3879 .gpl_only = false,
3880 .ret_type = RET_INTEGER,
3881 .arg1_type = ARG_PTR_TO_CTX,
3882 .arg2_type = ARG_ANYTHING,
3883 .arg3_type = ARG_ANYTHING,
3884 };
3885
BPF_CALL_3(sk_skb_change_head,struct sk_buff *,skb,u32,head_room,u64,flags)3886 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3887 u64, flags)
3888 {
3889 return __bpf_skb_change_head(skb, head_room, flags);
3890 }
3891
3892 static const struct bpf_func_proto sk_skb_change_head_proto = {
3893 .func = sk_skb_change_head,
3894 .gpl_only = false,
3895 .ret_type = RET_INTEGER,
3896 .arg1_type = ARG_PTR_TO_CTX,
3897 .arg2_type = ARG_ANYTHING,
3898 .arg3_type = ARG_ANYTHING,
3899 };
3900
BPF_CALL_1(bpf_xdp_get_buff_len,struct xdp_buff *,xdp)3901 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3902 {
3903 return xdp_get_buff_len(xdp);
3904 }
3905
3906 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3907 .func = bpf_xdp_get_buff_len,
3908 .gpl_only = false,
3909 .ret_type = RET_INTEGER,
3910 .arg1_type = ARG_PTR_TO_CTX,
3911 };
3912
3913 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3914
3915 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3916 .func = bpf_xdp_get_buff_len,
3917 .gpl_only = false,
3918 .arg1_type = ARG_PTR_TO_BTF_ID,
3919 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0],
3920 };
3921
xdp_get_metalen(const struct xdp_buff * xdp)3922 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3923 {
3924 return xdp_data_meta_unsupported(xdp) ? 0 :
3925 xdp->data - xdp->data_meta;
3926 }
3927
BPF_CALL_2(bpf_xdp_adjust_head,struct xdp_buff *,xdp,int,offset)3928 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3929 {
3930 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3931 unsigned long metalen = xdp_get_metalen(xdp);
3932 void *data_start = xdp_frame_end + metalen;
3933 void *data = xdp->data + offset;
3934
3935 if (unlikely(data < data_start ||
3936 data > xdp->data_end - ETH_HLEN))
3937 return -EINVAL;
3938
3939 if (metalen)
3940 memmove(xdp->data_meta + offset,
3941 xdp->data_meta, metalen);
3942 xdp->data_meta += offset;
3943 xdp->data = data;
3944
3945 return 0;
3946 }
3947
3948 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3949 .func = bpf_xdp_adjust_head,
3950 .gpl_only = false,
3951 .ret_type = RET_INTEGER,
3952 .arg1_type = ARG_PTR_TO_CTX,
3953 .arg2_type = ARG_ANYTHING,
3954 };
3955
bpf_xdp_copy_buf(struct xdp_buff * xdp,unsigned long off,void * buf,unsigned long len,bool flush)3956 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3957 void *buf, unsigned long len, bool flush)
3958 {
3959 unsigned long ptr_len, ptr_off = 0;
3960 skb_frag_t *next_frag, *end_frag;
3961 struct skb_shared_info *sinfo;
3962 void *src, *dst;
3963 u8 *ptr_buf;
3964
3965 if (likely(xdp->data_end - xdp->data >= off + len)) {
3966 src = flush ? buf : xdp->data + off;
3967 dst = flush ? xdp->data + off : buf;
3968 memcpy(dst, src, len);
3969 return;
3970 }
3971
3972 sinfo = xdp_get_shared_info_from_buff(xdp);
3973 end_frag = &sinfo->frags[sinfo->nr_frags];
3974 next_frag = &sinfo->frags[0];
3975
3976 ptr_len = xdp->data_end - xdp->data;
3977 ptr_buf = xdp->data;
3978
3979 while (true) {
3980 if (off < ptr_off + ptr_len) {
3981 unsigned long copy_off = off - ptr_off;
3982 unsigned long copy_len = min(len, ptr_len - copy_off);
3983
3984 src = flush ? buf : ptr_buf + copy_off;
3985 dst = flush ? ptr_buf + copy_off : buf;
3986 memcpy(dst, src, copy_len);
3987
3988 off += copy_len;
3989 len -= copy_len;
3990 buf += copy_len;
3991 }
3992
3993 if (!len || next_frag == end_frag)
3994 break;
3995
3996 ptr_off += ptr_len;
3997 ptr_buf = skb_frag_address(next_frag);
3998 ptr_len = skb_frag_size(next_frag);
3999 next_frag++;
4000 }
4001 }
4002
bpf_xdp_pointer(struct xdp_buff * xdp,u32 offset,u32 len)4003 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
4004 {
4005 u32 size = xdp->data_end - xdp->data;
4006 struct skb_shared_info *sinfo;
4007 void *addr = xdp->data;
4008 int i;
4009
4010 if (unlikely(offset > 0xffff || len > 0xffff))
4011 return ERR_PTR(-EFAULT);
4012
4013 if (unlikely(offset + len > xdp_get_buff_len(xdp)))
4014 return ERR_PTR(-EINVAL);
4015
4016 if (likely(offset < size)) /* linear area */
4017 goto out;
4018
4019 sinfo = xdp_get_shared_info_from_buff(xdp);
4020 offset -= size;
4021 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
4022 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
4023
4024 if (offset < frag_size) {
4025 addr = skb_frag_address(&sinfo->frags[i]);
4026 size = frag_size;
4027 break;
4028 }
4029 offset -= frag_size;
4030 }
4031 out:
4032 return offset + len <= size ? addr + offset : NULL;
4033 }
4034
BPF_CALL_4(bpf_xdp_load_bytes,struct xdp_buff *,xdp,u32,offset,void *,buf,u32,len)4035 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
4036 void *, buf, u32, len)
4037 {
4038 void *ptr;
4039
4040 ptr = bpf_xdp_pointer(xdp, offset, len);
4041 if (IS_ERR(ptr))
4042 return PTR_ERR(ptr);
4043
4044 if (!ptr)
4045 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
4046 else
4047 memcpy(buf, ptr, len);
4048
4049 return 0;
4050 }
4051
4052 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
4053 .func = bpf_xdp_load_bytes,
4054 .gpl_only = false,
4055 .ret_type = RET_INTEGER,
4056 .arg1_type = ARG_PTR_TO_CTX,
4057 .arg2_type = ARG_ANYTHING,
4058 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4059 .arg4_type = ARG_CONST_SIZE,
4060 };
4061
__bpf_xdp_load_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)4062 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4063 {
4064 return ____bpf_xdp_load_bytes(xdp, offset, buf, len);
4065 }
4066
BPF_CALL_4(bpf_xdp_store_bytes,struct xdp_buff *,xdp,u32,offset,void *,buf,u32,len)4067 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
4068 void *, buf, u32, len)
4069 {
4070 void *ptr;
4071
4072 ptr = bpf_xdp_pointer(xdp, offset, len);
4073 if (IS_ERR(ptr))
4074 return PTR_ERR(ptr);
4075
4076 if (!ptr)
4077 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
4078 else
4079 memcpy(ptr, buf, len);
4080
4081 return 0;
4082 }
4083
4084 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
4085 .func = bpf_xdp_store_bytes,
4086 .gpl_only = false,
4087 .ret_type = RET_INTEGER,
4088 .arg1_type = ARG_PTR_TO_CTX,
4089 .arg2_type = ARG_ANYTHING,
4090 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4091 .arg4_type = ARG_CONST_SIZE,
4092 };
4093
__bpf_xdp_store_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)4094 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4095 {
4096 return ____bpf_xdp_store_bytes(xdp, offset, buf, len);
4097 }
4098
bpf_xdp_frags_increase_tail(struct xdp_buff * xdp,int offset)4099 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
4100 {
4101 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4102 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
4103 struct xdp_rxq_info *rxq = xdp->rxq;
4104 unsigned int tailroom;
4105
4106 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
4107 return -EOPNOTSUPP;
4108
4109 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4110 if (unlikely(offset > tailroom))
4111 return -EINVAL;
4112
4113 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4114 skb_frag_size_add(frag, offset);
4115 sinfo->xdp_frags_size += offset;
4116 if (rxq->mem.type == MEM_TYPE_XSK_BUFF_POOL)
4117 xsk_buff_get_tail(xdp)->data_end += offset;
4118
4119 return 0;
4120 }
4121
bpf_xdp_shrink_data_zc(struct xdp_buff * xdp,int shrink,struct xdp_mem_info * mem_info,bool release)4122 static void bpf_xdp_shrink_data_zc(struct xdp_buff *xdp, int shrink,
4123 struct xdp_mem_info *mem_info, bool release)
4124 {
4125 struct xdp_buff *zc_frag = xsk_buff_get_tail(xdp);
4126
4127 if (release) {
4128 xsk_buff_del_tail(zc_frag);
4129 __xdp_return(NULL, mem_info, false, zc_frag);
4130 } else {
4131 zc_frag->data_end -= shrink;
4132 }
4133 }
4134
bpf_xdp_shrink_data(struct xdp_buff * xdp,skb_frag_t * frag,int shrink)4135 static bool bpf_xdp_shrink_data(struct xdp_buff *xdp, skb_frag_t *frag,
4136 int shrink)
4137 {
4138 struct xdp_mem_info *mem_info = &xdp->rxq->mem;
4139 bool release = skb_frag_size(frag) == shrink;
4140
4141 if (mem_info->type == MEM_TYPE_XSK_BUFF_POOL) {
4142 bpf_xdp_shrink_data_zc(xdp, shrink, mem_info, release);
4143 goto out;
4144 }
4145
4146 if (release) {
4147 struct page *page = skb_frag_page(frag);
4148
4149 __xdp_return(page_address(page), mem_info, false, NULL);
4150 }
4151
4152 out:
4153 return release;
4154 }
4155
bpf_xdp_frags_shrink_tail(struct xdp_buff * xdp,int offset)4156 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4157 {
4158 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4159 int i, n_frags_free = 0, len_free = 0;
4160
4161 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4162 return -EINVAL;
4163
4164 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4165 skb_frag_t *frag = &sinfo->frags[i];
4166 int shrink = min_t(int, offset, skb_frag_size(frag));
4167
4168 len_free += shrink;
4169 offset -= shrink;
4170 if (bpf_xdp_shrink_data(xdp, frag, shrink)) {
4171 n_frags_free++;
4172 } else {
4173 skb_frag_size_sub(frag, shrink);
4174 break;
4175 }
4176 }
4177 sinfo->nr_frags -= n_frags_free;
4178 sinfo->xdp_frags_size -= len_free;
4179
4180 if (unlikely(!sinfo->nr_frags)) {
4181 xdp_buff_clear_frags_flag(xdp);
4182 xdp->data_end -= offset;
4183 }
4184
4185 return 0;
4186 }
4187
BPF_CALL_2(bpf_xdp_adjust_tail,struct xdp_buff *,xdp,int,offset)4188 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4189 {
4190 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4191 void *data_end = xdp->data_end + offset;
4192
4193 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4194 if (offset < 0)
4195 return bpf_xdp_frags_shrink_tail(xdp, -offset);
4196
4197 return bpf_xdp_frags_increase_tail(xdp, offset);
4198 }
4199
4200 /* Notice that xdp_data_hard_end have reserved some tailroom */
4201 if (unlikely(data_end > data_hard_end))
4202 return -EINVAL;
4203
4204 if (unlikely(data_end < xdp->data + ETH_HLEN))
4205 return -EINVAL;
4206
4207 /* Clear memory area on grow, can contain uninit kernel memory */
4208 if (offset > 0)
4209 memset(xdp->data_end, 0, offset);
4210
4211 xdp->data_end = data_end;
4212
4213 return 0;
4214 }
4215
4216 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4217 .func = bpf_xdp_adjust_tail,
4218 .gpl_only = false,
4219 .ret_type = RET_INTEGER,
4220 .arg1_type = ARG_PTR_TO_CTX,
4221 .arg2_type = ARG_ANYTHING,
4222 };
4223
BPF_CALL_2(bpf_xdp_adjust_meta,struct xdp_buff *,xdp,int,offset)4224 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4225 {
4226 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4227 void *meta = xdp->data_meta + offset;
4228 unsigned long metalen = xdp->data - meta;
4229
4230 if (xdp_data_meta_unsupported(xdp))
4231 return -ENOTSUPP;
4232 if (unlikely(meta < xdp_frame_end ||
4233 meta > xdp->data))
4234 return -EINVAL;
4235 if (unlikely(xdp_metalen_invalid(metalen)))
4236 return -EACCES;
4237
4238 xdp->data_meta = meta;
4239
4240 return 0;
4241 }
4242
4243 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4244 .func = bpf_xdp_adjust_meta,
4245 .gpl_only = false,
4246 .ret_type = RET_INTEGER,
4247 .arg1_type = ARG_PTR_TO_CTX,
4248 .arg2_type = ARG_ANYTHING,
4249 };
4250
4251 /**
4252 * DOC: xdp redirect
4253 *
4254 * XDP_REDIRECT works by a three-step process, implemented in the functions
4255 * below:
4256 *
4257 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4258 * of the redirect and store it (along with some other metadata) in a per-CPU
4259 * struct bpf_redirect_info.
4260 *
4261 * 2. When the program returns the XDP_REDIRECT return code, the driver will
4262 * call xdp_do_redirect() which will use the information in struct
4263 * bpf_redirect_info to actually enqueue the frame into a map type-specific
4264 * bulk queue structure.
4265 *
4266 * 3. Before exiting its NAPI poll loop, the driver will call
4267 * xdp_do_flush(), which will flush all the different bulk queues,
4268 * thus completing the redirect. Note that xdp_do_flush() must be
4269 * called before napi_complete_done() in the driver, as the
4270 * XDP_REDIRECT logic relies on being inside a single NAPI instance
4271 * through to the xdp_do_flush() call for RCU protection of all
4272 * in-kernel data structures.
4273 */
4274 /*
4275 * Pointers to the map entries will be kept around for this whole sequence of
4276 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4277 * the core code; instead, the RCU protection relies on everything happening
4278 * inside a single NAPI poll sequence, which means it's between a pair of calls
4279 * to local_bh_disable()/local_bh_enable().
4280 *
4281 * The map entries are marked as __rcu and the map code makes sure to
4282 * dereference those pointers with rcu_dereference_check() in a way that works
4283 * for both sections that to hold an rcu_read_lock() and sections that are
4284 * called from NAPI without a separate rcu_read_lock(). The code below does not
4285 * use RCU annotations, but relies on those in the map code.
4286 */
xdp_do_flush(void)4287 void xdp_do_flush(void)
4288 {
4289 struct list_head *lh_map, *lh_dev, *lh_xsk;
4290
4291 bpf_net_ctx_get_all_used_flush_lists(&lh_map, &lh_dev, &lh_xsk);
4292 if (lh_dev)
4293 __dev_flush(lh_dev);
4294 if (lh_map)
4295 __cpu_map_flush(lh_map);
4296 if (lh_xsk)
4297 __xsk_map_flush(lh_xsk);
4298 }
4299 EXPORT_SYMBOL_GPL(xdp_do_flush);
4300
4301 #if defined(CONFIG_DEBUG_NET) && defined(CONFIG_BPF_SYSCALL)
xdp_do_check_flushed(struct napi_struct * napi)4302 void xdp_do_check_flushed(struct napi_struct *napi)
4303 {
4304 struct list_head *lh_map, *lh_dev, *lh_xsk;
4305 bool missed = false;
4306
4307 bpf_net_ctx_get_all_used_flush_lists(&lh_map, &lh_dev, &lh_xsk);
4308 if (lh_dev) {
4309 __dev_flush(lh_dev);
4310 missed = true;
4311 }
4312 if (lh_map) {
4313 __cpu_map_flush(lh_map);
4314 missed = true;
4315 }
4316 if (lh_xsk) {
4317 __xsk_map_flush(lh_xsk);
4318 missed = true;
4319 }
4320
4321 WARN_ONCE(missed, "Missing xdp_do_flush() invocation after NAPI by %ps\n",
4322 napi->poll);
4323 }
4324 #endif
4325
4326 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4327 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4328
xdp_master_redirect(struct xdp_buff * xdp)4329 u32 xdp_master_redirect(struct xdp_buff *xdp)
4330 {
4331 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4332 struct net_device *master, *slave;
4333
4334 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4335 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4336 if (slave && slave != xdp->rxq->dev) {
4337 /* The target device is different from the receiving device, so
4338 * redirect it to the new device.
4339 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4340 * drivers to unmap the packet from their rx ring.
4341 */
4342 ri->tgt_index = slave->ifindex;
4343 ri->map_id = INT_MAX;
4344 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4345 return XDP_REDIRECT;
4346 }
4347 return XDP_TX;
4348 }
4349 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4350
__xdp_do_redirect_xsk(struct bpf_redirect_info * ri,struct net_device * dev,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)4351 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4352 struct net_device *dev,
4353 struct xdp_buff *xdp,
4354 struct bpf_prog *xdp_prog)
4355 {
4356 enum bpf_map_type map_type = ri->map_type;
4357 void *fwd = ri->tgt_value;
4358 u32 map_id = ri->map_id;
4359 int err;
4360
4361 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4362 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4363
4364 err = __xsk_map_redirect(fwd, xdp);
4365 if (unlikely(err))
4366 goto err;
4367
4368 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4369 return 0;
4370 err:
4371 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4372 return err;
4373 }
4374
__xdp_do_redirect_frame(struct bpf_redirect_info * ri,struct net_device * dev,struct xdp_frame * xdpf,struct bpf_prog * xdp_prog)4375 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4376 struct net_device *dev,
4377 struct xdp_frame *xdpf,
4378 struct bpf_prog *xdp_prog)
4379 {
4380 enum bpf_map_type map_type = ri->map_type;
4381 void *fwd = ri->tgt_value;
4382 u32 map_id = ri->map_id;
4383 u32 flags = ri->flags;
4384 struct bpf_map *map;
4385 int err;
4386
4387 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4388 ri->flags = 0;
4389 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4390
4391 if (unlikely(!xdpf)) {
4392 err = -EOVERFLOW;
4393 goto err;
4394 }
4395
4396 switch (map_type) {
4397 case BPF_MAP_TYPE_DEVMAP:
4398 fallthrough;
4399 case BPF_MAP_TYPE_DEVMAP_HASH:
4400 if (unlikely(flags & BPF_F_BROADCAST)) {
4401 map = READ_ONCE(ri->map);
4402
4403 /* The map pointer is cleared when the map is being torn
4404 * down by dev_map_free()
4405 */
4406 if (unlikely(!map)) {
4407 err = -ENOENT;
4408 break;
4409 }
4410
4411 WRITE_ONCE(ri->map, NULL);
4412 err = dev_map_enqueue_multi(xdpf, dev, map,
4413 flags & BPF_F_EXCLUDE_INGRESS);
4414 } else {
4415 err = dev_map_enqueue(fwd, xdpf, dev);
4416 }
4417 break;
4418 case BPF_MAP_TYPE_CPUMAP:
4419 err = cpu_map_enqueue(fwd, xdpf, dev);
4420 break;
4421 case BPF_MAP_TYPE_UNSPEC:
4422 if (map_id == INT_MAX) {
4423 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4424 if (unlikely(!fwd)) {
4425 err = -EINVAL;
4426 break;
4427 }
4428 err = dev_xdp_enqueue(fwd, xdpf, dev);
4429 break;
4430 }
4431 fallthrough;
4432 default:
4433 err = -EBADRQC;
4434 }
4435
4436 if (unlikely(err))
4437 goto err;
4438
4439 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4440 return 0;
4441 err:
4442 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4443 return err;
4444 }
4445
xdp_do_redirect(struct net_device * dev,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)4446 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4447 struct bpf_prog *xdp_prog)
4448 {
4449 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4450 enum bpf_map_type map_type = ri->map_type;
4451
4452 if (map_type == BPF_MAP_TYPE_XSKMAP)
4453 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4454
4455 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4456 xdp_prog);
4457 }
4458 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4459
xdp_do_redirect_frame(struct net_device * dev,struct xdp_buff * xdp,struct xdp_frame * xdpf,struct bpf_prog * xdp_prog)4460 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4461 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4462 {
4463 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4464 enum bpf_map_type map_type = ri->map_type;
4465
4466 if (map_type == BPF_MAP_TYPE_XSKMAP)
4467 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4468
4469 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4470 }
4471 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4472
xdp_do_generic_redirect_map(struct net_device * dev,struct sk_buff * skb,struct xdp_buff * xdp,struct bpf_prog * xdp_prog,void * fwd,enum bpf_map_type map_type,u32 map_id,u32 flags)4473 static int xdp_do_generic_redirect_map(struct net_device *dev,
4474 struct sk_buff *skb,
4475 struct xdp_buff *xdp,
4476 struct bpf_prog *xdp_prog, void *fwd,
4477 enum bpf_map_type map_type, u32 map_id,
4478 u32 flags)
4479 {
4480 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4481 struct bpf_map *map;
4482 int err;
4483
4484 switch (map_type) {
4485 case BPF_MAP_TYPE_DEVMAP:
4486 fallthrough;
4487 case BPF_MAP_TYPE_DEVMAP_HASH:
4488 if (unlikely(flags & BPF_F_BROADCAST)) {
4489 map = READ_ONCE(ri->map);
4490
4491 /* The map pointer is cleared when the map is being torn
4492 * down by dev_map_free()
4493 */
4494 if (unlikely(!map)) {
4495 err = -ENOENT;
4496 break;
4497 }
4498
4499 WRITE_ONCE(ri->map, NULL);
4500 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4501 flags & BPF_F_EXCLUDE_INGRESS);
4502 } else {
4503 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4504 }
4505 if (unlikely(err))
4506 goto err;
4507 break;
4508 case BPF_MAP_TYPE_XSKMAP:
4509 err = xsk_generic_rcv(fwd, xdp);
4510 if (err)
4511 goto err;
4512 consume_skb(skb);
4513 break;
4514 case BPF_MAP_TYPE_CPUMAP:
4515 err = cpu_map_generic_redirect(fwd, skb);
4516 if (unlikely(err))
4517 goto err;
4518 break;
4519 default:
4520 err = -EBADRQC;
4521 goto err;
4522 }
4523
4524 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4525 return 0;
4526 err:
4527 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4528 return err;
4529 }
4530
xdp_do_generic_redirect(struct net_device * dev,struct sk_buff * skb,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)4531 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4532 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4533 {
4534 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4535 enum bpf_map_type map_type = ri->map_type;
4536 void *fwd = ri->tgt_value;
4537 u32 map_id = ri->map_id;
4538 u32 flags = ri->flags;
4539 int err;
4540
4541 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4542 ri->flags = 0;
4543 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4544
4545 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4546 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4547 if (unlikely(!fwd)) {
4548 err = -EINVAL;
4549 goto err;
4550 }
4551
4552 err = xdp_ok_fwd_dev(fwd, skb->len);
4553 if (unlikely(err))
4554 goto err;
4555
4556 skb->dev = fwd;
4557 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4558 generic_xdp_tx(skb, xdp_prog);
4559 return 0;
4560 }
4561
4562 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id, flags);
4563 err:
4564 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4565 return err;
4566 }
4567
BPF_CALL_2(bpf_xdp_redirect,u32,ifindex,u64,flags)4568 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4569 {
4570 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
4571
4572 if (unlikely(flags))
4573 return XDP_ABORTED;
4574
4575 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4576 * by map_idr) is used for ifindex based XDP redirect.
4577 */
4578 ri->tgt_index = ifindex;
4579 ri->map_id = INT_MAX;
4580 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4581
4582 return XDP_REDIRECT;
4583 }
4584
4585 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4586 .func = bpf_xdp_redirect,
4587 .gpl_only = false,
4588 .ret_type = RET_INTEGER,
4589 .arg1_type = ARG_ANYTHING,
4590 .arg2_type = ARG_ANYTHING,
4591 };
4592
BPF_CALL_3(bpf_xdp_redirect_map,struct bpf_map *,map,u64,key,u64,flags)4593 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key,
4594 u64, flags)
4595 {
4596 return map->ops->map_redirect(map, key, flags);
4597 }
4598
4599 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4600 .func = bpf_xdp_redirect_map,
4601 .gpl_only = false,
4602 .ret_type = RET_INTEGER,
4603 .arg1_type = ARG_CONST_MAP_PTR,
4604 .arg2_type = ARG_ANYTHING,
4605 .arg3_type = ARG_ANYTHING,
4606 };
4607
bpf_skb_copy(void * dst_buff,const void * skb,unsigned long off,unsigned long len)4608 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4609 unsigned long off, unsigned long len)
4610 {
4611 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4612
4613 if (unlikely(!ptr))
4614 return len;
4615 if (ptr != dst_buff)
4616 memcpy(dst_buff, ptr, len);
4617
4618 return 0;
4619 }
4620
BPF_CALL_5(bpf_skb_event_output,struct sk_buff *,skb,struct bpf_map *,map,u64,flags,void *,meta,u64,meta_size)4621 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4622 u64, flags, void *, meta, u64, meta_size)
4623 {
4624 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4625
4626 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4627 return -EINVAL;
4628 if (unlikely(!skb || skb_size > skb->len))
4629 return -EFAULT;
4630
4631 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4632 bpf_skb_copy);
4633 }
4634
4635 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4636 .func = bpf_skb_event_output,
4637 .gpl_only = true,
4638 .ret_type = RET_INTEGER,
4639 .arg1_type = ARG_PTR_TO_CTX,
4640 .arg2_type = ARG_CONST_MAP_PTR,
4641 .arg3_type = ARG_ANYTHING,
4642 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4643 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4644 };
4645
4646 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4647
4648 const struct bpf_func_proto bpf_skb_output_proto = {
4649 .func = bpf_skb_event_output,
4650 .gpl_only = true,
4651 .ret_type = RET_INTEGER,
4652 .arg1_type = ARG_PTR_TO_BTF_ID,
4653 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4654 .arg2_type = ARG_CONST_MAP_PTR,
4655 .arg3_type = ARG_ANYTHING,
4656 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4657 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4658 };
4659
bpf_tunnel_key_af(u64 flags)4660 static unsigned short bpf_tunnel_key_af(u64 flags)
4661 {
4662 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4663 }
4664
BPF_CALL_4(bpf_skb_get_tunnel_key,struct sk_buff *,skb,struct bpf_tunnel_key *,to,u32,size,u64,flags)4665 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4666 u32, size, u64, flags)
4667 {
4668 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4669 u8 compat[sizeof(struct bpf_tunnel_key)];
4670 void *to_orig = to;
4671 int err;
4672
4673 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4674 BPF_F_TUNINFO_FLAGS)))) {
4675 err = -EINVAL;
4676 goto err_clear;
4677 }
4678 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4679 err = -EPROTO;
4680 goto err_clear;
4681 }
4682 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4683 err = -EINVAL;
4684 switch (size) {
4685 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4686 case offsetof(struct bpf_tunnel_key, tunnel_label):
4687 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4688 goto set_compat;
4689 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4690 /* Fixup deprecated structure layouts here, so we have
4691 * a common path later on.
4692 */
4693 if (ip_tunnel_info_af(info) != AF_INET)
4694 goto err_clear;
4695 set_compat:
4696 to = (struct bpf_tunnel_key *)compat;
4697 break;
4698 default:
4699 goto err_clear;
4700 }
4701 }
4702
4703 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4704 to->tunnel_tos = info->key.tos;
4705 to->tunnel_ttl = info->key.ttl;
4706 if (flags & BPF_F_TUNINFO_FLAGS)
4707 to->tunnel_flags = ip_tunnel_flags_to_be16(info->key.tun_flags);
4708 else
4709 to->tunnel_ext = 0;
4710
4711 if (flags & BPF_F_TUNINFO_IPV6) {
4712 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4713 sizeof(to->remote_ipv6));
4714 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4715 sizeof(to->local_ipv6));
4716 to->tunnel_label = be32_to_cpu(info->key.label);
4717 } else {
4718 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4719 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4720 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4721 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4722 to->tunnel_label = 0;
4723 }
4724
4725 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4726 memcpy(to_orig, to, size);
4727
4728 return 0;
4729 err_clear:
4730 memset(to_orig, 0, size);
4731 return err;
4732 }
4733
4734 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4735 .func = bpf_skb_get_tunnel_key,
4736 .gpl_only = false,
4737 .ret_type = RET_INTEGER,
4738 .arg1_type = ARG_PTR_TO_CTX,
4739 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4740 .arg3_type = ARG_CONST_SIZE,
4741 .arg4_type = ARG_ANYTHING,
4742 };
4743
BPF_CALL_3(bpf_skb_get_tunnel_opt,struct sk_buff *,skb,u8 *,to,u32,size)4744 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4745 {
4746 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4747 int err;
4748
4749 if (unlikely(!info ||
4750 !ip_tunnel_is_options_present(info->key.tun_flags))) {
4751 err = -ENOENT;
4752 goto err_clear;
4753 }
4754 if (unlikely(size < info->options_len)) {
4755 err = -ENOMEM;
4756 goto err_clear;
4757 }
4758
4759 ip_tunnel_info_opts_get(to, info);
4760 if (size > info->options_len)
4761 memset(to + info->options_len, 0, size - info->options_len);
4762
4763 return info->options_len;
4764 err_clear:
4765 memset(to, 0, size);
4766 return err;
4767 }
4768
4769 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4770 .func = bpf_skb_get_tunnel_opt,
4771 .gpl_only = false,
4772 .ret_type = RET_INTEGER,
4773 .arg1_type = ARG_PTR_TO_CTX,
4774 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4775 .arg3_type = ARG_CONST_SIZE,
4776 };
4777
4778 static struct metadata_dst __percpu *md_dst;
4779
BPF_CALL_4(bpf_skb_set_tunnel_key,struct sk_buff *,skb,const struct bpf_tunnel_key *,from,u32,size,u64,flags)4780 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4781 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4782 {
4783 struct metadata_dst *md = this_cpu_ptr(md_dst);
4784 u8 compat[sizeof(struct bpf_tunnel_key)];
4785 struct ip_tunnel_info *info;
4786
4787 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4788 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER |
4789 BPF_F_NO_TUNNEL_KEY)))
4790 return -EINVAL;
4791 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4792 switch (size) {
4793 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4794 case offsetof(struct bpf_tunnel_key, tunnel_label):
4795 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4796 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4797 /* Fixup deprecated structure layouts here, so we have
4798 * a common path later on.
4799 */
4800 memcpy(compat, from, size);
4801 memset(compat + size, 0, sizeof(compat) - size);
4802 from = (const struct bpf_tunnel_key *) compat;
4803 break;
4804 default:
4805 return -EINVAL;
4806 }
4807 }
4808 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4809 from->tunnel_ext))
4810 return -EINVAL;
4811
4812 skb_dst_drop(skb);
4813 dst_hold((struct dst_entry *) md);
4814 skb_dst_set(skb, (struct dst_entry *) md);
4815
4816 info = &md->u.tun_info;
4817 memset(info, 0, sizeof(*info));
4818 info->mode = IP_TUNNEL_INFO_TX;
4819
4820 __set_bit(IP_TUNNEL_NOCACHE_BIT, info->key.tun_flags);
4821 __assign_bit(IP_TUNNEL_DONT_FRAGMENT_BIT, info->key.tun_flags,
4822 flags & BPF_F_DONT_FRAGMENT);
4823 __assign_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags,
4824 !(flags & BPF_F_ZERO_CSUM_TX));
4825 __assign_bit(IP_TUNNEL_SEQ_BIT, info->key.tun_flags,
4826 flags & BPF_F_SEQ_NUMBER);
4827 __assign_bit(IP_TUNNEL_KEY_BIT, info->key.tun_flags,
4828 !(flags & BPF_F_NO_TUNNEL_KEY));
4829
4830 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4831 info->key.tos = from->tunnel_tos;
4832 info->key.ttl = from->tunnel_ttl;
4833
4834 if (flags & BPF_F_TUNINFO_IPV6) {
4835 info->mode |= IP_TUNNEL_INFO_IPV6;
4836 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4837 sizeof(from->remote_ipv6));
4838 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4839 sizeof(from->local_ipv6));
4840 info->key.label = cpu_to_be32(from->tunnel_label) &
4841 IPV6_FLOWLABEL_MASK;
4842 } else {
4843 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4844 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4845 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4846 }
4847
4848 return 0;
4849 }
4850
4851 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4852 .func = bpf_skb_set_tunnel_key,
4853 .gpl_only = false,
4854 .ret_type = RET_INTEGER,
4855 .arg1_type = ARG_PTR_TO_CTX,
4856 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4857 .arg3_type = ARG_CONST_SIZE,
4858 .arg4_type = ARG_ANYTHING,
4859 };
4860
BPF_CALL_3(bpf_skb_set_tunnel_opt,struct sk_buff *,skb,const u8 *,from,u32,size)4861 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4862 const u8 *, from, u32, size)
4863 {
4864 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4865 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4866 IP_TUNNEL_DECLARE_FLAGS(present) = { };
4867
4868 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4869 return -EINVAL;
4870 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4871 return -ENOMEM;
4872
4873 ip_tunnel_set_options_present(present);
4874 ip_tunnel_info_opts_set(info, from, size, present);
4875
4876 return 0;
4877 }
4878
4879 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4880 .func = bpf_skb_set_tunnel_opt,
4881 .gpl_only = false,
4882 .ret_type = RET_INTEGER,
4883 .arg1_type = ARG_PTR_TO_CTX,
4884 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4885 .arg3_type = ARG_CONST_SIZE,
4886 };
4887
4888 static const struct bpf_func_proto *
bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)4889 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4890 {
4891 if (!md_dst) {
4892 struct metadata_dst __percpu *tmp;
4893
4894 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4895 METADATA_IP_TUNNEL,
4896 GFP_KERNEL);
4897 if (!tmp)
4898 return NULL;
4899 if (cmpxchg(&md_dst, NULL, tmp))
4900 metadata_dst_free_percpu(tmp);
4901 }
4902
4903 switch (which) {
4904 case BPF_FUNC_skb_set_tunnel_key:
4905 return &bpf_skb_set_tunnel_key_proto;
4906 case BPF_FUNC_skb_set_tunnel_opt:
4907 return &bpf_skb_set_tunnel_opt_proto;
4908 default:
4909 return NULL;
4910 }
4911 }
4912
BPF_CALL_3(bpf_skb_under_cgroup,struct sk_buff *,skb,struct bpf_map *,map,u32,idx)4913 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4914 u32, idx)
4915 {
4916 struct bpf_array *array = container_of(map, struct bpf_array, map);
4917 struct cgroup *cgrp;
4918 struct sock *sk;
4919
4920 sk = skb_to_full_sk(skb);
4921 if (!sk || !sk_fullsock(sk))
4922 return -ENOENT;
4923 if (unlikely(idx >= array->map.max_entries))
4924 return -E2BIG;
4925
4926 cgrp = READ_ONCE(array->ptrs[idx]);
4927 if (unlikely(!cgrp))
4928 return -EAGAIN;
4929
4930 return sk_under_cgroup_hierarchy(sk, cgrp);
4931 }
4932
4933 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4934 .func = bpf_skb_under_cgroup,
4935 .gpl_only = false,
4936 .ret_type = RET_INTEGER,
4937 .arg1_type = ARG_PTR_TO_CTX,
4938 .arg2_type = ARG_CONST_MAP_PTR,
4939 .arg3_type = ARG_ANYTHING,
4940 };
4941
4942 #ifdef CONFIG_SOCK_CGROUP_DATA
__bpf_sk_cgroup_id(struct sock * sk)4943 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4944 {
4945 struct cgroup *cgrp;
4946
4947 sk = sk_to_full_sk(sk);
4948 if (!sk || !sk_fullsock(sk))
4949 return 0;
4950
4951 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4952 return cgroup_id(cgrp);
4953 }
4954
BPF_CALL_1(bpf_skb_cgroup_id,const struct sk_buff *,skb)4955 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4956 {
4957 return __bpf_sk_cgroup_id(skb->sk);
4958 }
4959
4960 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4961 .func = bpf_skb_cgroup_id,
4962 .gpl_only = false,
4963 .ret_type = RET_INTEGER,
4964 .arg1_type = ARG_PTR_TO_CTX,
4965 };
4966
__bpf_sk_ancestor_cgroup_id(struct sock * sk,int ancestor_level)4967 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4968 int ancestor_level)
4969 {
4970 struct cgroup *ancestor;
4971 struct cgroup *cgrp;
4972
4973 sk = sk_to_full_sk(sk);
4974 if (!sk || !sk_fullsock(sk))
4975 return 0;
4976
4977 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4978 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4979 if (!ancestor)
4980 return 0;
4981
4982 return cgroup_id(ancestor);
4983 }
4984
BPF_CALL_2(bpf_skb_ancestor_cgroup_id,const struct sk_buff *,skb,int,ancestor_level)4985 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4986 ancestor_level)
4987 {
4988 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4989 }
4990
4991 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4992 .func = bpf_skb_ancestor_cgroup_id,
4993 .gpl_only = false,
4994 .ret_type = RET_INTEGER,
4995 .arg1_type = ARG_PTR_TO_CTX,
4996 .arg2_type = ARG_ANYTHING,
4997 };
4998
BPF_CALL_1(bpf_sk_cgroup_id,struct sock *,sk)4999 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
5000 {
5001 return __bpf_sk_cgroup_id(sk);
5002 }
5003
5004 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
5005 .func = bpf_sk_cgroup_id,
5006 .gpl_only = false,
5007 .ret_type = RET_INTEGER,
5008 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5009 };
5010
BPF_CALL_2(bpf_sk_ancestor_cgroup_id,struct sock *,sk,int,ancestor_level)5011 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
5012 {
5013 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
5014 }
5015
5016 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
5017 .func = bpf_sk_ancestor_cgroup_id,
5018 .gpl_only = false,
5019 .ret_type = RET_INTEGER,
5020 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5021 .arg2_type = ARG_ANYTHING,
5022 };
5023 #endif
5024
bpf_xdp_copy(void * dst,const void * ctx,unsigned long off,unsigned long len)5025 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
5026 unsigned long off, unsigned long len)
5027 {
5028 struct xdp_buff *xdp = (struct xdp_buff *)ctx;
5029
5030 bpf_xdp_copy_buf(xdp, off, dst, len, false);
5031 return 0;
5032 }
5033
BPF_CALL_5(bpf_xdp_event_output,struct xdp_buff *,xdp,struct bpf_map *,map,u64,flags,void *,meta,u64,meta_size)5034 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
5035 u64, flags, void *, meta, u64, meta_size)
5036 {
5037 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
5038
5039 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
5040 return -EINVAL;
5041
5042 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
5043 return -EFAULT;
5044
5045 return bpf_event_output(map, flags, meta, meta_size, xdp,
5046 xdp_size, bpf_xdp_copy);
5047 }
5048
5049 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
5050 .func = bpf_xdp_event_output,
5051 .gpl_only = true,
5052 .ret_type = RET_INTEGER,
5053 .arg1_type = ARG_PTR_TO_CTX,
5054 .arg2_type = ARG_CONST_MAP_PTR,
5055 .arg3_type = ARG_ANYTHING,
5056 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5057 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
5058 };
5059
5060 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
5061
5062 const struct bpf_func_proto bpf_xdp_output_proto = {
5063 .func = bpf_xdp_event_output,
5064 .gpl_only = true,
5065 .ret_type = RET_INTEGER,
5066 .arg1_type = ARG_PTR_TO_BTF_ID,
5067 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
5068 .arg2_type = ARG_CONST_MAP_PTR,
5069 .arg3_type = ARG_ANYTHING,
5070 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5071 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
5072 };
5073
BPF_CALL_1(bpf_get_socket_cookie,struct sk_buff *,skb)5074 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
5075 {
5076 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
5077 }
5078
5079 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
5080 .func = bpf_get_socket_cookie,
5081 .gpl_only = false,
5082 .ret_type = RET_INTEGER,
5083 .arg1_type = ARG_PTR_TO_CTX,
5084 };
5085
BPF_CALL_1(bpf_get_socket_cookie_sock_addr,struct bpf_sock_addr_kern *,ctx)5086 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5087 {
5088 return __sock_gen_cookie(ctx->sk);
5089 }
5090
5091 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
5092 .func = bpf_get_socket_cookie_sock_addr,
5093 .gpl_only = false,
5094 .ret_type = RET_INTEGER,
5095 .arg1_type = ARG_PTR_TO_CTX,
5096 };
5097
BPF_CALL_1(bpf_get_socket_cookie_sock,struct sock *,ctx)5098 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
5099 {
5100 return __sock_gen_cookie(ctx);
5101 }
5102
5103 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
5104 .func = bpf_get_socket_cookie_sock,
5105 .gpl_only = false,
5106 .ret_type = RET_INTEGER,
5107 .arg1_type = ARG_PTR_TO_CTX,
5108 };
5109
BPF_CALL_1(bpf_get_socket_ptr_cookie,struct sock *,sk)5110 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
5111 {
5112 return sk ? sock_gen_cookie(sk) : 0;
5113 }
5114
5115 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
5116 .func = bpf_get_socket_ptr_cookie,
5117 .gpl_only = false,
5118 .ret_type = RET_INTEGER,
5119 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | PTR_MAYBE_NULL,
5120 };
5121
BPF_CALL_1(bpf_get_socket_cookie_sock_ops,struct bpf_sock_ops_kern *,ctx)5122 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5123 {
5124 return __sock_gen_cookie(ctx->sk);
5125 }
5126
5127 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
5128 .func = bpf_get_socket_cookie_sock_ops,
5129 .gpl_only = false,
5130 .ret_type = RET_INTEGER,
5131 .arg1_type = ARG_PTR_TO_CTX,
5132 };
5133
__bpf_get_netns_cookie(struct sock * sk)5134 static u64 __bpf_get_netns_cookie(struct sock *sk)
5135 {
5136 const struct net *net = sk ? sock_net(sk) : &init_net;
5137
5138 return net->net_cookie;
5139 }
5140
BPF_CALL_1(bpf_get_netns_cookie_sock,struct sock *,ctx)5141 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
5142 {
5143 return __bpf_get_netns_cookie(ctx);
5144 }
5145
5146 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
5147 .func = bpf_get_netns_cookie_sock,
5148 .gpl_only = false,
5149 .ret_type = RET_INTEGER,
5150 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5151 };
5152
BPF_CALL_1(bpf_get_netns_cookie_sock_addr,struct bpf_sock_addr_kern *,ctx)5153 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5154 {
5155 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5156 }
5157
5158 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5159 .func = bpf_get_netns_cookie_sock_addr,
5160 .gpl_only = false,
5161 .ret_type = RET_INTEGER,
5162 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5163 };
5164
BPF_CALL_1(bpf_get_netns_cookie_sock_ops,struct bpf_sock_ops_kern *,ctx)5165 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5166 {
5167 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5168 }
5169
5170 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5171 .func = bpf_get_netns_cookie_sock_ops,
5172 .gpl_only = false,
5173 .ret_type = RET_INTEGER,
5174 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5175 };
5176
BPF_CALL_1(bpf_get_netns_cookie_sk_msg,struct sk_msg *,ctx)5177 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5178 {
5179 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5180 }
5181
5182 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5183 .func = bpf_get_netns_cookie_sk_msg,
5184 .gpl_only = false,
5185 .ret_type = RET_INTEGER,
5186 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5187 };
5188
BPF_CALL_1(bpf_get_socket_uid,struct sk_buff *,skb)5189 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5190 {
5191 struct sock *sk = sk_to_full_sk(skb->sk);
5192 kuid_t kuid;
5193
5194 if (!sk || !sk_fullsock(sk))
5195 return overflowuid;
5196 kuid = sock_net_uid(sock_net(sk), sk);
5197 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5198 }
5199
5200 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5201 .func = bpf_get_socket_uid,
5202 .gpl_only = false,
5203 .ret_type = RET_INTEGER,
5204 .arg1_type = ARG_PTR_TO_CTX,
5205 };
5206
sol_socket_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5207 static int sol_socket_sockopt(struct sock *sk, int optname,
5208 char *optval, int *optlen,
5209 bool getopt)
5210 {
5211 switch (optname) {
5212 case SO_REUSEADDR:
5213 case SO_SNDBUF:
5214 case SO_RCVBUF:
5215 case SO_KEEPALIVE:
5216 case SO_PRIORITY:
5217 case SO_REUSEPORT:
5218 case SO_RCVLOWAT:
5219 case SO_MARK:
5220 case SO_MAX_PACING_RATE:
5221 case SO_BINDTOIFINDEX:
5222 case SO_TXREHASH:
5223 if (*optlen != sizeof(int))
5224 return -EINVAL;
5225 break;
5226 case SO_BINDTODEVICE:
5227 break;
5228 default:
5229 return -EINVAL;
5230 }
5231
5232 if (getopt) {
5233 if (optname == SO_BINDTODEVICE)
5234 return -EINVAL;
5235 return sk_getsockopt(sk, SOL_SOCKET, optname,
5236 KERNEL_SOCKPTR(optval),
5237 KERNEL_SOCKPTR(optlen));
5238 }
5239
5240 return sk_setsockopt(sk, SOL_SOCKET, optname,
5241 KERNEL_SOCKPTR(optval), *optlen);
5242 }
5243
bpf_sol_tcp_setsockopt(struct sock * sk,int optname,char * optval,int optlen)5244 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5245 char *optval, int optlen)
5246 {
5247 struct tcp_sock *tp = tcp_sk(sk);
5248 unsigned long timeout;
5249 int val;
5250
5251 if (optlen != sizeof(int))
5252 return -EINVAL;
5253
5254 val = *(int *)optval;
5255
5256 /* Only some options are supported */
5257 switch (optname) {
5258 case TCP_BPF_IW:
5259 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5260 return -EINVAL;
5261 tcp_snd_cwnd_set(tp, val);
5262 break;
5263 case TCP_BPF_SNDCWND_CLAMP:
5264 if (val <= 0)
5265 return -EINVAL;
5266 tp->snd_cwnd_clamp = val;
5267 tp->snd_ssthresh = val;
5268 break;
5269 case TCP_BPF_DELACK_MAX:
5270 timeout = usecs_to_jiffies(val);
5271 if (timeout > TCP_DELACK_MAX ||
5272 timeout < TCP_TIMEOUT_MIN)
5273 return -EINVAL;
5274 inet_csk(sk)->icsk_delack_max = timeout;
5275 break;
5276 case TCP_BPF_RTO_MIN:
5277 timeout = usecs_to_jiffies(val);
5278 if (timeout > TCP_RTO_MIN ||
5279 timeout < TCP_TIMEOUT_MIN)
5280 return -EINVAL;
5281 inet_csk(sk)->icsk_rto_min = timeout;
5282 break;
5283 case TCP_BPF_SOCK_OPS_CB_FLAGS:
5284 if (val & ~(BPF_SOCK_OPS_ALL_CB_FLAGS))
5285 return -EINVAL;
5286 tp->bpf_sock_ops_cb_flags = val;
5287 break;
5288 default:
5289 return -EINVAL;
5290 }
5291
5292 return 0;
5293 }
5294
sol_tcp_sockopt_congestion(struct sock * sk,char * optval,int * optlen,bool getopt)5295 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5296 int *optlen, bool getopt)
5297 {
5298 struct tcp_sock *tp;
5299 int ret;
5300
5301 if (*optlen < 2)
5302 return -EINVAL;
5303
5304 if (getopt) {
5305 if (!inet_csk(sk)->icsk_ca_ops)
5306 return -EINVAL;
5307 /* BPF expects NULL-terminated tcp-cc string */
5308 optval[--(*optlen)] = '\0';
5309 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5310 KERNEL_SOCKPTR(optval),
5311 KERNEL_SOCKPTR(optlen));
5312 }
5313
5314 /* "cdg" is the only cc that alloc a ptr
5315 * in inet_csk_ca area. The bpf-tcp-cc may
5316 * overwrite this ptr after switching to cdg.
5317 */
5318 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5319 return -ENOTSUPP;
5320
5321 /* It stops this looping
5322 *
5323 * .init => bpf_setsockopt(tcp_cc) => .init =>
5324 * bpf_setsockopt(tcp_cc)" => .init => ....
5325 *
5326 * The second bpf_setsockopt(tcp_cc) is not allowed
5327 * in order to break the loop when both .init
5328 * are the same bpf prog.
5329 *
5330 * This applies even the second bpf_setsockopt(tcp_cc)
5331 * does not cause a loop. This limits only the first
5332 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5333 * pick a fallback cc (eg. peer does not support ECN)
5334 * and the second '.init' cannot fallback to
5335 * another.
5336 */
5337 tp = tcp_sk(sk);
5338 if (tp->bpf_chg_cc_inprogress)
5339 return -EBUSY;
5340
5341 tp->bpf_chg_cc_inprogress = 1;
5342 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5343 KERNEL_SOCKPTR(optval), *optlen);
5344 tp->bpf_chg_cc_inprogress = 0;
5345 return ret;
5346 }
5347
sol_tcp_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5348 static int sol_tcp_sockopt(struct sock *sk, int optname,
5349 char *optval, int *optlen,
5350 bool getopt)
5351 {
5352 if (sk->sk_protocol != IPPROTO_TCP)
5353 return -EINVAL;
5354
5355 switch (optname) {
5356 case TCP_NODELAY:
5357 case TCP_MAXSEG:
5358 case TCP_KEEPIDLE:
5359 case TCP_KEEPINTVL:
5360 case TCP_KEEPCNT:
5361 case TCP_SYNCNT:
5362 case TCP_WINDOW_CLAMP:
5363 case TCP_THIN_LINEAR_TIMEOUTS:
5364 case TCP_USER_TIMEOUT:
5365 case TCP_NOTSENT_LOWAT:
5366 case TCP_SAVE_SYN:
5367 if (*optlen != sizeof(int))
5368 return -EINVAL;
5369 break;
5370 case TCP_CONGESTION:
5371 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5372 case TCP_SAVED_SYN:
5373 if (*optlen < 1)
5374 return -EINVAL;
5375 break;
5376 case TCP_BPF_SOCK_OPS_CB_FLAGS:
5377 if (*optlen != sizeof(int))
5378 return -EINVAL;
5379 if (getopt) {
5380 struct tcp_sock *tp = tcp_sk(sk);
5381 int cb_flags = tp->bpf_sock_ops_cb_flags;
5382
5383 memcpy(optval, &cb_flags, *optlen);
5384 return 0;
5385 }
5386 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5387 default:
5388 if (getopt)
5389 return -EINVAL;
5390 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5391 }
5392
5393 if (getopt) {
5394 if (optname == TCP_SAVED_SYN) {
5395 struct tcp_sock *tp = tcp_sk(sk);
5396
5397 if (!tp->saved_syn ||
5398 *optlen > tcp_saved_syn_len(tp->saved_syn))
5399 return -EINVAL;
5400 memcpy(optval, tp->saved_syn->data, *optlen);
5401 /* It cannot free tp->saved_syn here because it
5402 * does not know if the user space still needs it.
5403 */
5404 return 0;
5405 }
5406
5407 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5408 KERNEL_SOCKPTR(optval),
5409 KERNEL_SOCKPTR(optlen));
5410 }
5411
5412 return do_tcp_setsockopt(sk, SOL_TCP, optname,
5413 KERNEL_SOCKPTR(optval), *optlen);
5414 }
5415
sol_ip_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5416 static int sol_ip_sockopt(struct sock *sk, int optname,
5417 char *optval, int *optlen,
5418 bool getopt)
5419 {
5420 if (sk->sk_family != AF_INET)
5421 return -EINVAL;
5422
5423 switch (optname) {
5424 case IP_TOS:
5425 if (*optlen != sizeof(int))
5426 return -EINVAL;
5427 break;
5428 default:
5429 return -EINVAL;
5430 }
5431
5432 if (getopt)
5433 return do_ip_getsockopt(sk, SOL_IP, optname,
5434 KERNEL_SOCKPTR(optval),
5435 KERNEL_SOCKPTR(optlen));
5436
5437 return do_ip_setsockopt(sk, SOL_IP, optname,
5438 KERNEL_SOCKPTR(optval), *optlen);
5439 }
5440
sol_ipv6_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5441 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5442 char *optval, int *optlen,
5443 bool getopt)
5444 {
5445 if (sk->sk_family != AF_INET6)
5446 return -EINVAL;
5447
5448 switch (optname) {
5449 case IPV6_TCLASS:
5450 case IPV6_AUTOFLOWLABEL:
5451 if (*optlen != sizeof(int))
5452 return -EINVAL;
5453 break;
5454 default:
5455 return -EINVAL;
5456 }
5457
5458 if (getopt)
5459 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5460 KERNEL_SOCKPTR(optval),
5461 KERNEL_SOCKPTR(optlen));
5462
5463 return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5464 KERNEL_SOCKPTR(optval), *optlen);
5465 }
5466
__bpf_setsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5467 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5468 char *optval, int optlen)
5469 {
5470 if (!sk_fullsock(sk))
5471 return -EINVAL;
5472
5473 if (level == SOL_SOCKET)
5474 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5475 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5476 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5477 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5478 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5479 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5480 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5481
5482 return -EINVAL;
5483 }
5484
_bpf_setsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5485 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5486 char *optval, int optlen)
5487 {
5488 if (sk_fullsock(sk))
5489 sock_owned_by_me(sk);
5490 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5491 }
5492
__bpf_getsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5493 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5494 char *optval, int optlen)
5495 {
5496 int err, saved_optlen = optlen;
5497
5498 if (!sk_fullsock(sk)) {
5499 err = -EINVAL;
5500 goto done;
5501 }
5502
5503 if (level == SOL_SOCKET)
5504 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5505 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5506 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5507 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5508 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5509 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5510 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5511 else
5512 err = -EINVAL;
5513
5514 done:
5515 if (err)
5516 optlen = 0;
5517 if (optlen < saved_optlen)
5518 memset(optval + optlen, 0, saved_optlen - optlen);
5519 return err;
5520 }
5521
_bpf_getsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5522 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5523 char *optval, int optlen)
5524 {
5525 if (sk_fullsock(sk))
5526 sock_owned_by_me(sk);
5527 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5528 }
5529
BPF_CALL_5(bpf_sk_setsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5530 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5531 int, optname, char *, optval, int, optlen)
5532 {
5533 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5534 }
5535
5536 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5537 .func = bpf_sk_setsockopt,
5538 .gpl_only = false,
5539 .ret_type = RET_INTEGER,
5540 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5541 .arg2_type = ARG_ANYTHING,
5542 .arg3_type = ARG_ANYTHING,
5543 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5544 .arg5_type = ARG_CONST_SIZE,
5545 };
5546
BPF_CALL_5(bpf_sk_getsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5547 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5548 int, optname, char *, optval, int, optlen)
5549 {
5550 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5551 }
5552
5553 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5554 .func = bpf_sk_getsockopt,
5555 .gpl_only = false,
5556 .ret_type = RET_INTEGER,
5557 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5558 .arg2_type = ARG_ANYTHING,
5559 .arg3_type = ARG_ANYTHING,
5560 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5561 .arg5_type = ARG_CONST_SIZE,
5562 };
5563
BPF_CALL_5(bpf_unlocked_sk_setsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5564 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5565 int, optname, char *, optval, int, optlen)
5566 {
5567 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5568 }
5569
5570 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5571 .func = bpf_unlocked_sk_setsockopt,
5572 .gpl_only = false,
5573 .ret_type = RET_INTEGER,
5574 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5575 .arg2_type = ARG_ANYTHING,
5576 .arg3_type = ARG_ANYTHING,
5577 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5578 .arg5_type = ARG_CONST_SIZE,
5579 };
5580
BPF_CALL_5(bpf_unlocked_sk_getsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5581 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5582 int, optname, char *, optval, int, optlen)
5583 {
5584 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5585 }
5586
5587 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5588 .func = bpf_unlocked_sk_getsockopt,
5589 .gpl_only = false,
5590 .ret_type = RET_INTEGER,
5591 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5592 .arg2_type = ARG_ANYTHING,
5593 .arg3_type = ARG_ANYTHING,
5594 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5595 .arg5_type = ARG_CONST_SIZE,
5596 };
5597
BPF_CALL_5(bpf_sock_addr_setsockopt,struct bpf_sock_addr_kern *,ctx,int,level,int,optname,char *,optval,int,optlen)5598 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5599 int, level, int, optname, char *, optval, int, optlen)
5600 {
5601 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5602 }
5603
5604 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5605 .func = bpf_sock_addr_setsockopt,
5606 .gpl_only = false,
5607 .ret_type = RET_INTEGER,
5608 .arg1_type = ARG_PTR_TO_CTX,
5609 .arg2_type = ARG_ANYTHING,
5610 .arg3_type = ARG_ANYTHING,
5611 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5612 .arg5_type = ARG_CONST_SIZE,
5613 };
5614
BPF_CALL_5(bpf_sock_addr_getsockopt,struct bpf_sock_addr_kern *,ctx,int,level,int,optname,char *,optval,int,optlen)5615 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5616 int, level, int, optname, char *, optval, int, optlen)
5617 {
5618 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5619 }
5620
5621 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5622 .func = bpf_sock_addr_getsockopt,
5623 .gpl_only = false,
5624 .ret_type = RET_INTEGER,
5625 .arg1_type = ARG_PTR_TO_CTX,
5626 .arg2_type = ARG_ANYTHING,
5627 .arg3_type = ARG_ANYTHING,
5628 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5629 .arg5_type = ARG_CONST_SIZE,
5630 };
5631
BPF_CALL_5(bpf_sock_ops_setsockopt,struct bpf_sock_ops_kern *,bpf_sock,int,level,int,optname,char *,optval,int,optlen)5632 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5633 int, level, int, optname, char *, optval, int, optlen)
5634 {
5635 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5636 }
5637
5638 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5639 .func = bpf_sock_ops_setsockopt,
5640 .gpl_only = false,
5641 .ret_type = RET_INTEGER,
5642 .arg1_type = ARG_PTR_TO_CTX,
5643 .arg2_type = ARG_ANYTHING,
5644 .arg3_type = ARG_ANYTHING,
5645 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5646 .arg5_type = ARG_CONST_SIZE,
5647 };
5648
bpf_sock_ops_get_syn(struct bpf_sock_ops_kern * bpf_sock,int optname,const u8 ** start)5649 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5650 int optname, const u8 **start)
5651 {
5652 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5653 const u8 *hdr_start;
5654 int ret;
5655
5656 if (syn_skb) {
5657 /* sk is a request_sock here */
5658
5659 if (optname == TCP_BPF_SYN) {
5660 hdr_start = syn_skb->data;
5661 ret = tcp_hdrlen(syn_skb);
5662 } else if (optname == TCP_BPF_SYN_IP) {
5663 hdr_start = skb_network_header(syn_skb);
5664 ret = skb_network_header_len(syn_skb) +
5665 tcp_hdrlen(syn_skb);
5666 } else {
5667 /* optname == TCP_BPF_SYN_MAC */
5668 hdr_start = skb_mac_header(syn_skb);
5669 ret = skb_mac_header_len(syn_skb) +
5670 skb_network_header_len(syn_skb) +
5671 tcp_hdrlen(syn_skb);
5672 }
5673 } else {
5674 struct sock *sk = bpf_sock->sk;
5675 struct saved_syn *saved_syn;
5676
5677 if (sk->sk_state == TCP_NEW_SYN_RECV)
5678 /* synack retransmit. bpf_sock->syn_skb will
5679 * not be available. It has to resort to
5680 * saved_syn (if it is saved).
5681 */
5682 saved_syn = inet_reqsk(sk)->saved_syn;
5683 else
5684 saved_syn = tcp_sk(sk)->saved_syn;
5685
5686 if (!saved_syn)
5687 return -ENOENT;
5688
5689 if (optname == TCP_BPF_SYN) {
5690 hdr_start = saved_syn->data +
5691 saved_syn->mac_hdrlen +
5692 saved_syn->network_hdrlen;
5693 ret = saved_syn->tcp_hdrlen;
5694 } else if (optname == TCP_BPF_SYN_IP) {
5695 hdr_start = saved_syn->data +
5696 saved_syn->mac_hdrlen;
5697 ret = saved_syn->network_hdrlen +
5698 saved_syn->tcp_hdrlen;
5699 } else {
5700 /* optname == TCP_BPF_SYN_MAC */
5701
5702 /* TCP_SAVE_SYN may not have saved the mac hdr */
5703 if (!saved_syn->mac_hdrlen)
5704 return -ENOENT;
5705
5706 hdr_start = saved_syn->data;
5707 ret = saved_syn->mac_hdrlen +
5708 saved_syn->network_hdrlen +
5709 saved_syn->tcp_hdrlen;
5710 }
5711 }
5712
5713 *start = hdr_start;
5714 return ret;
5715 }
5716
BPF_CALL_5(bpf_sock_ops_getsockopt,struct bpf_sock_ops_kern *,bpf_sock,int,level,int,optname,char *,optval,int,optlen)5717 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5718 int, level, int, optname, char *, optval, int, optlen)
5719 {
5720 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5721 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5722 int ret, copy_len = 0;
5723 const u8 *start;
5724
5725 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5726 if (ret > 0) {
5727 copy_len = ret;
5728 if (optlen < copy_len) {
5729 copy_len = optlen;
5730 ret = -ENOSPC;
5731 }
5732
5733 memcpy(optval, start, copy_len);
5734 }
5735
5736 /* Zero out unused buffer at the end */
5737 memset(optval + copy_len, 0, optlen - copy_len);
5738
5739 return ret;
5740 }
5741
5742 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5743 }
5744
5745 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5746 .func = bpf_sock_ops_getsockopt,
5747 .gpl_only = false,
5748 .ret_type = RET_INTEGER,
5749 .arg1_type = ARG_PTR_TO_CTX,
5750 .arg2_type = ARG_ANYTHING,
5751 .arg3_type = ARG_ANYTHING,
5752 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5753 .arg5_type = ARG_CONST_SIZE,
5754 };
5755
BPF_CALL_2(bpf_sock_ops_cb_flags_set,struct bpf_sock_ops_kern *,bpf_sock,int,argval)5756 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5757 int, argval)
5758 {
5759 struct sock *sk = bpf_sock->sk;
5760 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5761
5762 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5763 return -EINVAL;
5764
5765 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5766
5767 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5768 }
5769
5770 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5771 .func = bpf_sock_ops_cb_flags_set,
5772 .gpl_only = false,
5773 .ret_type = RET_INTEGER,
5774 .arg1_type = ARG_PTR_TO_CTX,
5775 .arg2_type = ARG_ANYTHING,
5776 };
5777
5778 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5779 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5780
BPF_CALL_3(bpf_bind,struct bpf_sock_addr_kern *,ctx,struct sockaddr *,addr,int,addr_len)5781 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5782 int, addr_len)
5783 {
5784 #ifdef CONFIG_INET
5785 struct sock *sk = ctx->sk;
5786 u32 flags = BIND_FROM_BPF;
5787 int err;
5788
5789 err = -EINVAL;
5790 if (addr_len < offsetofend(struct sockaddr, sa_family))
5791 return err;
5792 if (addr->sa_family == AF_INET) {
5793 if (addr_len < sizeof(struct sockaddr_in))
5794 return err;
5795 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5796 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5797 return __inet_bind(sk, addr, addr_len, flags);
5798 #if IS_ENABLED(CONFIG_IPV6)
5799 } else if (addr->sa_family == AF_INET6) {
5800 if (addr_len < SIN6_LEN_RFC2133)
5801 return err;
5802 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5803 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5804 /* ipv6_bpf_stub cannot be NULL, since it's called from
5805 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5806 */
5807 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5808 #endif /* CONFIG_IPV6 */
5809 }
5810 #endif /* CONFIG_INET */
5811
5812 return -EAFNOSUPPORT;
5813 }
5814
5815 static const struct bpf_func_proto bpf_bind_proto = {
5816 .func = bpf_bind,
5817 .gpl_only = false,
5818 .ret_type = RET_INTEGER,
5819 .arg1_type = ARG_PTR_TO_CTX,
5820 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5821 .arg3_type = ARG_CONST_SIZE,
5822 };
5823
5824 #ifdef CONFIG_XFRM
5825
5826 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
5827 (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES))
5828
5829 struct metadata_dst __percpu *xfrm_bpf_md_dst;
5830 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst);
5831
5832 #endif
5833
BPF_CALL_5(bpf_skb_get_xfrm_state,struct sk_buff *,skb,u32,index,struct bpf_xfrm_state *,to,u32,size,u64,flags)5834 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5835 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5836 {
5837 const struct sec_path *sp = skb_sec_path(skb);
5838 const struct xfrm_state *x;
5839
5840 if (!sp || unlikely(index >= sp->len || flags))
5841 goto err_clear;
5842
5843 x = sp->xvec[index];
5844
5845 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5846 goto err_clear;
5847
5848 to->reqid = x->props.reqid;
5849 to->spi = x->id.spi;
5850 to->family = x->props.family;
5851 to->ext = 0;
5852
5853 if (to->family == AF_INET6) {
5854 memcpy(to->remote_ipv6, x->props.saddr.a6,
5855 sizeof(to->remote_ipv6));
5856 } else {
5857 to->remote_ipv4 = x->props.saddr.a4;
5858 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5859 }
5860
5861 return 0;
5862 err_clear:
5863 memset(to, 0, size);
5864 return -EINVAL;
5865 }
5866
5867 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5868 .func = bpf_skb_get_xfrm_state,
5869 .gpl_only = false,
5870 .ret_type = RET_INTEGER,
5871 .arg1_type = ARG_PTR_TO_CTX,
5872 .arg2_type = ARG_ANYTHING,
5873 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5874 .arg4_type = ARG_CONST_SIZE,
5875 .arg5_type = ARG_ANYTHING,
5876 };
5877 #endif
5878
5879 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
bpf_fib_set_fwd_params(struct bpf_fib_lookup * params,u32 mtu)5880 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, u32 mtu)
5881 {
5882 params->h_vlan_TCI = 0;
5883 params->h_vlan_proto = 0;
5884 if (mtu)
5885 params->mtu_result = mtu; /* union with tot_len */
5886
5887 return 0;
5888 }
5889 #endif
5890
5891 #if IS_ENABLED(CONFIG_INET)
bpf_ipv4_fib_lookup(struct net * net,struct bpf_fib_lookup * params,u32 flags,bool check_mtu)5892 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5893 u32 flags, bool check_mtu)
5894 {
5895 struct fib_nh_common *nhc;
5896 struct in_device *in_dev;
5897 struct neighbour *neigh;
5898 struct net_device *dev;
5899 struct fib_result res;
5900 struct flowi4 fl4;
5901 u32 mtu = 0;
5902 int err;
5903
5904 dev = dev_get_by_index_rcu(net, params->ifindex);
5905 if (unlikely(!dev))
5906 return -ENODEV;
5907
5908 /* verify forwarding is enabled on this interface */
5909 in_dev = __in_dev_get_rcu(dev);
5910 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5911 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5912
5913 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5914 fl4.flowi4_iif = 1;
5915 fl4.flowi4_oif = params->ifindex;
5916 } else {
5917 fl4.flowi4_iif = params->ifindex;
5918 fl4.flowi4_oif = 0;
5919 }
5920 fl4.flowi4_tos = params->tos & INET_DSCP_MASK;
5921 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5922 fl4.flowi4_flags = 0;
5923
5924 fl4.flowi4_proto = params->l4_protocol;
5925 fl4.daddr = params->ipv4_dst;
5926 fl4.saddr = params->ipv4_src;
5927 fl4.fl4_sport = params->sport;
5928 fl4.fl4_dport = params->dport;
5929 fl4.flowi4_multipath_hash = 0;
5930
5931 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5932 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5933 struct fib_table *tb;
5934
5935 if (flags & BPF_FIB_LOOKUP_TBID) {
5936 tbid = params->tbid;
5937 /* zero out for vlan output */
5938 params->tbid = 0;
5939 }
5940
5941 tb = fib_get_table(net, tbid);
5942 if (unlikely(!tb))
5943 return BPF_FIB_LKUP_RET_NOT_FWDED;
5944
5945 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5946 } else {
5947 if (flags & BPF_FIB_LOOKUP_MARK)
5948 fl4.flowi4_mark = params->mark;
5949 else
5950 fl4.flowi4_mark = 0;
5951 fl4.flowi4_secid = 0;
5952 fl4.flowi4_tun_key.tun_id = 0;
5953 fl4.flowi4_uid = sock_net_uid(net, NULL);
5954
5955 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5956 }
5957
5958 if (err) {
5959 /* map fib lookup errors to RTN_ type */
5960 if (err == -EINVAL)
5961 return BPF_FIB_LKUP_RET_BLACKHOLE;
5962 if (err == -EHOSTUNREACH)
5963 return BPF_FIB_LKUP_RET_UNREACHABLE;
5964 if (err == -EACCES)
5965 return BPF_FIB_LKUP_RET_PROHIBIT;
5966
5967 return BPF_FIB_LKUP_RET_NOT_FWDED;
5968 }
5969
5970 if (res.type != RTN_UNICAST)
5971 return BPF_FIB_LKUP_RET_NOT_FWDED;
5972
5973 if (fib_info_num_path(res.fi) > 1)
5974 fib_select_path(net, &res, &fl4, NULL);
5975
5976 if (check_mtu) {
5977 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5978 if (params->tot_len > mtu) {
5979 params->mtu_result = mtu; /* union with tot_len */
5980 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5981 }
5982 }
5983
5984 nhc = res.nhc;
5985
5986 /* do not handle lwt encaps right now */
5987 if (nhc->nhc_lwtstate)
5988 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5989
5990 dev = nhc->nhc_dev;
5991
5992 params->rt_metric = res.fi->fib_priority;
5993 params->ifindex = dev->ifindex;
5994
5995 if (flags & BPF_FIB_LOOKUP_SRC)
5996 params->ipv4_src = fib_result_prefsrc(net, &res);
5997
5998 /* xdp and cls_bpf programs are run in RCU-bh so
5999 * rcu_read_lock_bh is not needed here
6000 */
6001 if (likely(nhc->nhc_gw_family != AF_INET6)) {
6002 if (nhc->nhc_gw_family)
6003 params->ipv4_dst = nhc->nhc_gw.ipv4;
6004 } else {
6005 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
6006
6007 params->family = AF_INET6;
6008 *dst = nhc->nhc_gw.ipv6;
6009 }
6010
6011 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
6012 goto set_fwd_params;
6013
6014 if (likely(nhc->nhc_gw_family != AF_INET6))
6015 neigh = __ipv4_neigh_lookup_noref(dev,
6016 (__force u32)params->ipv4_dst);
6017 else
6018 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst);
6019
6020 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
6021 return BPF_FIB_LKUP_RET_NO_NEIGH;
6022 memcpy(params->dmac, neigh->ha, ETH_ALEN);
6023 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6024
6025 set_fwd_params:
6026 return bpf_fib_set_fwd_params(params, mtu);
6027 }
6028 #endif
6029
6030 #if IS_ENABLED(CONFIG_IPV6)
bpf_ipv6_fib_lookup(struct net * net,struct bpf_fib_lookup * params,u32 flags,bool check_mtu)6031 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
6032 u32 flags, bool check_mtu)
6033 {
6034 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
6035 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
6036 struct fib6_result res = {};
6037 struct neighbour *neigh;
6038 struct net_device *dev;
6039 struct inet6_dev *idev;
6040 struct flowi6 fl6;
6041 int strict = 0;
6042 int oif, err;
6043 u32 mtu = 0;
6044
6045 /* link local addresses are never forwarded */
6046 if (rt6_need_strict(dst) || rt6_need_strict(src))
6047 return BPF_FIB_LKUP_RET_NOT_FWDED;
6048
6049 dev = dev_get_by_index_rcu(net, params->ifindex);
6050 if (unlikely(!dev))
6051 return -ENODEV;
6052
6053 idev = __in6_dev_get_safely(dev);
6054 if (unlikely(!idev || !READ_ONCE(idev->cnf.forwarding)))
6055 return BPF_FIB_LKUP_RET_FWD_DISABLED;
6056
6057 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
6058 fl6.flowi6_iif = 1;
6059 oif = fl6.flowi6_oif = params->ifindex;
6060 } else {
6061 oif = fl6.flowi6_iif = params->ifindex;
6062 fl6.flowi6_oif = 0;
6063 strict = RT6_LOOKUP_F_HAS_SADDR;
6064 }
6065 fl6.flowlabel = params->flowinfo;
6066 fl6.flowi6_scope = 0;
6067 fl6.flowi6_flags = 0;
6068 fl6.mp_hash = 0;
6069
6070 fl6.flowi6_proto = params->l4_protocol;
6071 fl6.daddr = *dst;
6072 fl6.saddr = *src;
6073 fl6.fl6_sport = params->sport;
6074 fl6.fl6_dport = params->dport;
6075
6076 if (flags & BPF_FIB_LOOKUP_DIRECT) {
6077 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
6078 struct fib6_table *tb;
6079
6080 if (flags & BPF_FIB_LOOKUP_TBID) {
6081 tbid = params->tbid;
6082 /* zero out for vlan output */
6083 params->tbid = 0;
6084 }
6085
6086 tb = ipv6_stub->fib6_get_table(net, tbid);
6087 if (unlikely(!tb))
6088 return BPF_FIB_LKUP_RET_NOT_FWDED;
6089
6090 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
6091 strict);
6092 } else {
6093 if (flags & BPF_FIB_LOOKUP_MARK)
6094 fl6.flowi6_mark = params->mark;
6095 else
6096 fl6.flowi6_mark = 0;
6097 fl6.flowi6_secid = 0;
6098 fl6.flowi6_tun_key.tun_id = 0;
6099 fl6.flowi6_uid = sock_net_uid(net, NULL);
6100
6101 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
6102 }
6103
6104 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
6105 res.f6i == net->ipv6.fib6_null_entry))
6106 return BPF_FIB_LKUP_RET_NOT_FWDED;
6107
6108 switch (res.fib6_type) {
6109 /* only unicast is forwarded */
6110 case RTN_UNICAST:
6111 break;
6112 case RTN_BLACKHOLE:
6113 return BPF_FIB_LKUP_RET_BLACKHOLE;
6114 case RTN_UNREACHABLE:
6115 return BPF_FIB_LKUP_RET_UNREACHABLE;
6116 case RTN_PROHIBIT:
6117 return BPF_FIB_LKUP_RET_PROHIBIT;
6118 default:
6119 return BPF_FIB_LKUP_RET_NOT_FWDED;
6120 }
6121
6122 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
6123 fl6.flowi6_oif != 0, NULL, strict);
6124
6125 if (check_mtu) {
6126 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
6127 if (params->tot_len > mtu) {
6128 params->mtu_result = mtu; /* union with tot_len */
6129 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
6130 }
6131 }
6132
6133 if (res.nh->fib_nh_lws)
6134 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
6135
6136 if (res.nh->fib_nh_gw_family)
6137 *dst = res.nh->fib_nh_gw6;
6138
6139 dev = res.nh->fib_nh_dev;
6140 params->rt_metric = res.f6i->fib6_metric;
6141 params->ifindex = dev->ifindex;
6142
6143 if (flags & BPF_FIB_LOOKUP_SRC) {
6144 if (res.f6i->fib6_prefsrc.plen) {
6145 *src = res.f6i->fib6_prefsrc.addr;
6146 } else {
6147 err = ipv6_bpf_stub->ipv6_dev_get_saddr(net, dev,
6148 &fl6.daddr, 0,
6149 src);
6150 if (err)
6151 return BPF_FIB_LKUP_RET_NO_SRC_ADDR;
6152 }
6153 }
6154
6155 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
6156 goto set_fwd_params;
6157
6158 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
6159 * not needed here.
6160 */
6161 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
6162 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
6163 return BPF_FIB_LKUP_RET_NO_NEIGH;
6164 memcpy(params->dmac, neigh->ha, ETH_ALEN);
6165 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6166
6167 set_fwd_params:
6168 return bpf_fib_set_fwd_params(params, mtu);
6169 }
6170 #endif
6171
6172 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \
6173 BPF_FIB_LOOKUP_SKIP_NEIGH | BPF_FIB_LOOKUP_TBID | \
6174 BPF_FIB_LOOKUP_SRC | BPF_FIB_LOOKUP_MARK)
6175
BPF_CALL_4(bpf_xdp_fib_lookup,struct xdp_buff *,ctx,struct bpf_fib_lookup *,params,int,plen,u32,flags)6176 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
6177 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6178 {
6179 if (plen < sizeof(*params))
6180 return -EINVAL;
6181
6182 if (flags & ~BPF_FIB_LOOKUP_MASK)
6183 return -EINVAL;
6184
6185 switch (params->family) {
6186 #if IS_ENABLED(CONFIG_INET)
6187 case AF_INET:
6188 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
6189 flags, true);
6190 #endif
6191 #if IS_ENABLED(CONFIG_IPV6)
6192 case AF_INET6:
6193 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
6194 flags, true);
6195 #endif
6196 }
6197 return -EAFNOSUPPORT;
6198 }
6199
6200 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
6201 .func = bpf_xdp_fib_lookup,
6202 .gpl_only = true,
6203 .ret_type = RET_INTEGER,
6204 .arg1_type = ARG_PTR_TO_CTX,
6205 .arg2_type = ARG_PTR_TO_MEM,
6206 .arg3_type = ARG_CONST_SIZE,
6207 .arg4_type = ARG_ANYTHING,
6208 };
6209
BPF_CALL_4(bpf_skb_fib_lookup,struct sk_buff *,skb,struct bpf_fib_lookup *,params,int,plen,u32,flags)6210 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6211 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6212 {
6213 struct net *net = dev_net(skb->dev);
6214 int rc = -EAFNOSUPPORT;
6215 bool check_mtu = false;
6216
6217 if (plen < sizeof(*params))
6218 return -EINVAL;
6219
6220 if (flags & ~BPF_FIB_LOOKUP_MASK)
6221 return -EINVAL;
6222
6223 if (params->tot_len)
6224 check_mtu = true;
6225
6226 switch (params->family) {
6227 #if IS_ENABLED(CONFIG_INET)
6228 case AF_INET:
6229 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6230 break;
6231 #endif
6232 #if IS_ENABLED(CONFIG_IPV6)
6233 case AF_INET6:
6234 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6235 break;
6236 #endif
6237 }
6238
6239 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6240 struct net_device *dev;
6241
6242 /* When tot_len isn't provided by user, check skb
6243 * against MTU of FIB lookup resulting net_device
6244 */
6245 dev = dev_get_by_index_rcu(net, params->ifindex);
6246 if (!is_skb_forwardable(dev, skb))
6247 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6248
6249 params->mtu_result = dev->mtu; /* union with tot_len */
6250 }
6251
6252 return rc;
6253 }
6254
6255 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6256 .func = bpf_skb_fib_lookup,
6257 .gpl_only = true,
6258 .ret_type = RET_INTEGER,
6259 .arg1_type = ARG_PTR_TO_CTX,
6260 .arg2_type = ARG_PTR_TO_MEM,
6261 .arg3_type = ARG_CONST_SIZE,
6262 .arg4_type = ARG_ANYTHING,
6263 };
6264
__dev_via_ifindex(struct net_device * dev_curr,u32 ifindex)6265 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6266 u32 ifindex)
6267 {
6268 struct net *netns = dev_net(dev_curr);
6269
6270 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6271 if (ifindex == 0)
6272 return dev_curr;
6273
6274 return dev_get_by_index_rcu(netns, ifindex);
6275 }
6276
BPF_CALL_5(bpf_skb_check_mtu,struct sk_buff *,skb,u32,ifindex,u32 *,mtu_len,s32,len_diff,u64,flags)6277 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6278 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6279 {
6280 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6281 struct net_device *dev = skb->dev;
6282 int skb_len, dev_len;
6283 int mtu = 0;
6284
6285 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS))) {
6286 ret = -EINVAL;
6287 goto out;
6288 }
6289
6290 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len))) {
6291 ret = -EINVAL;
6292 goto out;
6293 }
6294
6295 dev = __dev_via_ifindex(dev, ifindex);
6296 if (unlikely(!dev)) {
6297 ret = -ENODEV;
6298 goto out;
6299 }
6300
6301 mtu = READ_ONCE(dev->mtu);
6302 dev_len = mtu + dev->hard_header_len;
6303
6304 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6305 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6306
6307 skb_len += len_diff; /* minus result pass check */
6308 if (skb_len <= dev_len) {
6309 ret = BPF_MTU_CHK_RET_SUCCESS;
6310 goto out;
6311 }
6312 /* At this point, skb->len exceed MTU, but as it include length of all
6313 * segments, it can still be below MTU. The SKB can possibly get
6314 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
6315 * must choose if segs are to be MTU checked.
6316 */
6317 if (skb_is_gso(skb)) {
6318 ret = BPF_MTU_CHK_RET_SUCCESS;
6319 if (flags & BPF_MTU_CHK_SEGS &&
6320 !skb_gso_validate_network_len(skb, mtu))
6321 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6322 }
6323 out:
6324 *mtu_len = mtu;
6325 return ret;
6326 }
6327
BPF_CALL_5(bpf_xdp_check_mtu,struct xdp_buff *,xdp,u32,ifindex,u32 *,mtu_len,s32,len_diff,u64,flags)6328 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6329 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6330 {
6331 struct net_device *dev = xdp->rxq->dev;
6332 int xdp_len = xdp->data_end - xdp->data;
6333 int ret = BPF_MTU_CHK_RET_SUCCESS;
6334 int mtu = 0, dev_len;
6335
6336 /* XDP variant doesn't support multi-buffer segment check (yet) */
6337 if (unlikely(flags)) {
6338 ret = -EINVAL;
6339 goto out;
6340 }
6341
6342 dev = __dev_via_ifindex(dev, ifindex);
6343 if (unlikely(!dev)) {
6344 ret = -ENODEV;
6345 goto out;
6346 }
6347
6348 mtu = READ_ONCE(dev->mtu);
6349 dev_len = mtu + dev->hard_header_len;
6350
6351 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6352 if (*mtu_len)
6353 xdp_len = *mtu_len + dev->hard_header_len;
6354
6355 xdp_len += len_diff; /* minus result pass check */
6356 if (xdp_len > dev_len)
6357 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6358 out:
6359 *mtu_len = mtu;
6360 return ret;
6361 }
6362
6363 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6364 .func = bpf_skb_check_mtu,
6365 .gpl_only = true,
6366 .ret_type = RET_INTEGER,
6367 .arg1_type = ARG_PTR_TO_CTX,
6368 .arg2_type = ARG_ANYTHING,
6369 .arg3_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_ALIGNED,
6370 .arg3_size = sizeof(u32),
6371 .arg4_type = ARG_ANYTHING,
6372 .arg5_type = ARG_ANYTHING,
6373 };
6374
6375 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6376 .func = bpf_xdp_check_mtu,
6377 .gpl_only = true,
6378 .ret_type = RET_INTEGER,
6379 .arg1_type = ARG_PTR_TO_CTX,
6380 .arg2_type = ARG_ANYTHING,
6381 .arg3_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_ALIGNED,
6382 .arg3_size = sizeof(u32),
6383 .arg4_type = ARG_ANYTHING,
6384 .arg5_type = ARG_ANYTHING,
6385 };
6386
6387 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
bpf_push_seg6_encap(struct sk_buff * skb,u32 type,void * hdr,u32 len)6388 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6389 {
6390 int err;
6391 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6392
6393 if (!seg6_validate_srh(srh, len, false))
6394 return -EINVAL;
6395
6396 switch (type) {
6397 case BPF_LWT_ENCAP_SEG6_INLINE:
6398 if (skb->protocol != htons(ETH_P_IPV6))
6399 return -EBADMSG;
6400
6401 err = seg6_do_srh_inline(skb, srh);
6402 break;
6403 case BPF_LWT_ENCAP_SEG6:
6404 skb_reset_inner_headers(skb);
6405 skb->encapsulation = 1;
6406 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6407 break;
6408 default:
6409 return -EINVAL;
6410 }
6411
6412 bpf_compute_data_pointers(skb);
6413 if (err)
6414 return err;
6415
6416 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6417
6418 return seg6_lookup_nexthop(skb, NULL, 0);
6419 }
6420 #endif /* CONFIG_IPV6_SEG6_BPF */
6421
6422 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
bpf_push_ip_encap(struct sk_buff * skb,void * hdr,u32 len,bool ingress)6423 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6424 bool ingress)
6425 {
6426 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6427 }
6428 #endif
6429
BPF_CALL_4(bpf_lwt_in_push_encap,struct sk_buff *,skb,u32,type,void *,hdr,u32,len)6430 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6431 u32, len)
6432 {
6433 switch (type) {
6434 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6435 case BPF_LWT_ENCAP_SEG6:
6436 case BPF_LWT_ENCAP_SEG6_INLINE:
6437 return bpf_push_seg6_encap(skb, type, hdr, len);
6438 #endif
6439 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6440 case BPF_LWT_ENCAP_IP:
6441 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6442 #endif
6443 default:
6444 return -EINVAL;
6445 }
6446 }
6447
BPF_CALL_4(bpf_lwt_xmit_push_encap,struct sk_buff *,skb,u32,type,void *,hdr,u32,len)6448 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6449 void *, hdr, u32, len)
6450 {
6451 switch (type) {
6452 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6453 case BPF_LWT_ENCAP_IP:
6454 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6455 #endif
6456 default:
6457 return -EINVAL;
6458 }
6459 }
6460
6461 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6462 .func = bpf_lwt_in_push_encap,
6463 .gpl_only = false,
6464 .ret_type = RET_INTEGER,
6465 .arg1_type = ARG_PTR_TO_CTX,
6466 .arg2_type = ARG_ANYTHING,
6467 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6468 .arg4_type = ARG_CONST_SIZE
6469 };
6470
6471 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6472 .func = bpf_lwt_xmit_push_encap,
6473 .gpl_only = false,
6474 .ret_type = RET_INTEGER,
6475 .arg1_type = ARG_PTR_TO_CTX,
6476 .arg2_type = ARG_ANYTHING,
6477 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6478 .arg4_type = ARG_CONST_SIZE
6479 };
6480
6481 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
BPF_CALL_4(bpf_lwt_seg6_store_bytes,struct sk_buff *,skb,u32,offset,const void *,from,u32,len)6482 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6483 const void *, from, u32, len)
6484 {
6485 struct seg6_bpf_srh_state *srh_state =
6486 this_cpu_ptr(&seg6_bpf_srh_states);
6487 struct ipv6_sr_hdr *srh = srh_state->srh;
6488 void *srh_tlvs, *srh_end, *ptr;
6489 int srhoff = 0;
6490
6491 lockdep_assert_held(&srh_state->bh_lock);
6492 if (srh == NULL)
6493 return -EINVAL;
6494
6495 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6496 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6497
6498 ptr = skb->data + offset;
6499 if (ptr >= srh_tlvs && ptr + len <= srh_end)
6500 srh_state->valid = false;
6501 else if (ptr < (void *)&srh->flags ||
6502 ptr + len > (void *)&srh->segments)
6503 return -EFAULT;
6504
6505 if (unlikely(bpf_try_make_writable(skb, offset + len)))
6506 return -EFAULT;
6507 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6508 return -EINVAL;
6509 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6510
6511 memcpy(skb->data + offset, from, len);
6512 return 0;
6513 }
6514
6515 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6516 .func = bpf_lwt_seg6_store_bytes,
6517 .gpl_only = false,
6518 .ret_type = RET_INTEGER,
6519 .arg1_type = ARG_PTR_TO_CTX,
6520 .arg2_type = ARG_ANYTHING,
6521 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6522 .arg4_type = ARG_CONST_SIZE
6523 };
6524
bpf_update_srh_state(struct sk_buff * skb)6525 static void bpf_update_srh_state(struct sk_buff *skb)
6526 {
6527 struct seg6_bpf_srh_state *srh_state =
6528 this_cpu_ptr(&seg6_bpf_srh_states);
6529 int srhoff = 0;
6530
6531 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6532 srh_state->srh = NULL;
6533 } else {
6534 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6535 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6536 srh_state->valid = true;
6537 }
6538 }
6539
BPF_CALL_4(bpf_lwt_seg6_action,struct sk_buff *,skb,u32,action,void *,param,u32,param_len)6540 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6541 u32, action, void *, param, u32, param_len)
6542 {
6543 struct seg6_bpf_srh_state *srh_state =
6544 this_cpu_ptr(&seg6_bpf_srh_states);
6545 int hdroff = 0;
6546 int err;
6547
6548 lockdep_assert_held(&srh_state->bh_lock);
6549 switch (action) {
6550 case SEG6_LOCAL_ACTION_END_X:
6551 if (!seg6_bpf_has_valid_srh(skb))
6552 return -EBADMSG;
6553 if (param_len != sizeof(struct in6_addr))
6554 return -EINVAL;
6555 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6556 case SEG6_LOCAL_ACTION_END_T:
6557 if (!seg6_bpf_has_valid_srh(skb))
6558 return -EBADMSG;
6559 if (param_len != sizeof(int))
6560 return -EINVAL;
6561 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6562 case SEG6_LOCAL_ACTION_END_DT6:
6563 if (!seg6_bpf_has_valid_srh(skb))
6564 return -EBADMSG;
6565 if (param_len != sizeof(int))
6566 return -EINVAL;
6567
6568 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6569 return -EBADMSG;
6570 if (!pskb_pull(skb, hdroff))
6571 return -EBADMSG;
6572
6573 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6574 skb_reset_network_header(skb);
6575 skb_reset_transport_header(skb);
6576 skb->encapsulation = 0;
6577
6578 bpf_compute_data_pointers(skb);
6579 bpf_update_srh_state(skb);
6580 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6581 case SEG6_LOCAL_ACTION_END_B6:
6582 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6583 return -EBADMSG;
6584 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6585 param, param_len);
6586 if (!err)
6587 bpf_update_srh_state(skb);
6588
6589 return err;
6590 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6591 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6592 return -EBADMSG;
6593 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6594 param, param_len);
6595 if (!err)
6596 bpf_update_srh_state(skb);
6597
6598 return err;
6599 default:
6600 return -EINVAL;
6601 }
6602 }
6603
6604 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6605 .func = bpf_lwt_seg6_action,
6606 .gpl_only = false,
6607 .ret_type = RET_INTEGER,
6608 .arg1_type = ARG_PTR_TO_CTX,
6609 .arg2_type = ARG_ANYTHING,
6610 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6611 .arg4_type = ARG_CONST_SIZE
6612 };
6613
BPF_CALL_3(bpf_lwt_seg6_adjust_srh,struct sk_buff *,skb,u32,offset,s32,len)6614 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6615 s32, len)
6616 {
6617 struct seg6_bpf_srh_state *srh_state =
6618 this_cpu_ptr(&seg6_bpf_srh_states);
6619 struct ipv6_sr_hdr *srh = srh_state->srh;
6620 void *srh_end, *srh_tlvs, *ptr;
6621 struct ipv6hdr *hdr;
6622 int srhoff = 0;
6623 int ret;
6624
6625 lockdep_assert_held(&srh_state->bh_lock);
6626 if (unlikely(srh == NULL))
6627 return -EINVAL;
6628
6629 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6630 ((srh->first_segment + 1) << 4));
6631 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6632 srh_state->hdrlen);
6633 ptr = skb->data + offset;
6634
6635 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6636 return -EFAULT;
6637 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6638 return -EFAULT;
6639
6640 if (len > 0) {
6641 ret = skb_cow_head(skb, len);
6642 if (unlikely(ret < 0))
6643 return ret;
6644
6645 ret = bpf_skb_net_hdr_push(skb, offset, len);
6646 } else {
6647 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6648 }
6649
6650 bpf_compute_data_pointers(skb);
6651 if (unlikely(ret < 0))
6652 return ret;
6653
6654 hdr = (struct ipv6hdr *)skb->data;
6655 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6656
6657 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6658 return -EINVAL;
6659 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6660 srh_state->hdrlen += len;
6661 srh_state->valid = false;
6662 return 0;
6663 }
6664
6665 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6666 .func = bpf_lwt_seg6_adjust_srh,
6667 .gpl_only = false,
6668 .ret_type = RET_INTEGER,
6669 .arg1_type = ARG_PTR_TO_CTX,
6670 .arg2_type = ARG_ANYTHING,
6671 .arg3_type = ARG_ANYTHING,
6672 };
6673 #endif /* CONFIG_IPV6_SEG6_BPF */
6674
6675 #ifdef CONFIG_INET
sk_lookup(struct net * net,struct bpf_sock_tuple * tuple,int dif,int sdif,u8 family,u8 proto)6676 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6677 int dif, int sdif, u8 family, u8 proto)
6678 {
6679 struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6680 bool refcounted = false;
6681 struct sock *sk = NULL;
6682
6683 if (family == AF_INET) {
6684 __be32 src4 = tuple->ipv4.saddr;
6685 __be32 dst4 = tuple->ipv4.daddr;
6686
6687 if (proto == IPPROTO_TCP)
6688 sk = __inet_lookup(net, hinfo, NULL, 0,
6689 src4, tuple->ipv4.sport,
6690 dst4, tuple->ipv4.dport,
6691 dif, sdif, &refcounted);
6692 else
6693 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6694 dst4, tuple->ipv4.dport,
6695 dif, sdif, net->ipv4.udp_table, NULL);
6696 #if IS_ENABLED(CONFIG_IPV6)
6697 } else {
6698 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6699 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6700
6701 if (proto == IPPROTO_TCP)
6702 sk = __inet6_lookup(net, hinfo, NULL, 0,
6703 src6, tuple->ipv6.sport,
6704 dst6, ntohs(tuple->ipv6.dport),
6705 dif, sdif, &refcounted);
6706 else if (likely(ipv6_bpf_stub))
6707 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6708 src6, tuple->ipv6.sport,
6709 dst6, tuple->ipv6.dport,
6710 dif, sdif,
6711 net->ipv4.udp_table, NULL);
6712 #endif
6713 }
6714
6715 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6716 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6717 sk = NULL;
6718 }
6719 return sk;
6720 }
6721
6722 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6723 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6724 */
6725 static struct sock *
__bpf_skc_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,struct net * caller_net,u32 ifindex,u8 proto,u64 netns_id,u64 flags,int sdif)6726 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6727 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6728 u64 flags, int sdif)
6729 {
6730 struct sock *sk = NULL;
6731 struct net *net;
6732 u8 family;
6733
6734 if (len == sizeof(tuple->ipv4))
6735 family = AF_INET;
6736 else if (len == sizeof(tuple->ipv6))
6737 family = AF_INET6;
6738 else
6739 return NULL;
6740
6741 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6742 goto out;
6743
6744 if (sdif < 0) {
6745 if (family == AF_INET)
6746 sdif = inet_sdif(skb);
6747 else
6748 sdif = inet6_sdif(skb);
6749 }
6750
6751 if ((s32)netns_id < 0) {
6752 net = caller_net;
6753 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6754 } else {
6755 net = get_net_ns_by_id(caller_net, netns_id);
6756 if (unlikely(!net))
6757 goto out;
6758 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6759 put_net(net);
6760 }
6761
6762 out:
6763 return sk;
6764 }
6765
6766 static struct sock *
__bpf_sk_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,struct net * caller_net,u32 ifindex,u8 proto,u64 netns_id,u64 flags,int sdif)6767 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6768 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6769 u64 flags, int sdif)
6770 {
6771 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6772 ifindex, proto, netns_id, flags,
6773 sdif);
6774
6775 if (sk) {
6776 struct sock *sk2 = sk_to_full_sk(sk);
6777
6778 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6779 * sock refcnt is decremented to prevent a request_sock leak.
6780 */
6781 if (!sk_fullsock(sk2))
6782 sk2 = NULL;
6783 if (sk2 != sk) {
6784 sock_gen_put(sk);
6785 /* Ensure there is no need to bump sk2 refcnt */
6786 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6787 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6788 return NULL;
6789 }
6790 sk = sk2;
6791 }
6792 }
6793
6794 return sk;
6795 }
6796
6797 static struct sock *
bpf_skc_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,u8 proto,u64 netns_id,u64 flags)6798 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6799 u8 proto, u64 netns_id, u64 flags)
6800 {
6801 struct net *caller_net;
6802 int ifindex;
6803
6804 if (skb->dev) {
6805 caller_net = dev_net(skb->dev);
6806 ifindex = skb->dev->ifindex;
6807 } else {
6808 caller_net = sock_net(skb->sk);
6809 ifindex = 0;
6810 }
6811
6812 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6813 netns_id, flags, -1);
6814 }
6815
6816 static struct sock *
bpf_sk_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,u8 proto,u64 netns_id,u64 flags)6817 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6818 u8 proto, u64 netns_id, u64 flags)
6819 {
6820 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6821 flags);
6822
6823 if (sk) {
6824 struct sock *sk2 = sk_to_full_sk(sk);
6825
6826 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6827 * sock refcnt is decremented to prevent a request_sock leak.
6828 */
6829 if (!sk_fullsock(sk2))
6830 sk2 = NULL;
6831 if (sk2 != sk) {
6832 sock_gen_put(sk);
6833 /* Ensure there is no need to bump sk2 refcnt */
6834 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6835 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6836 return NULL;
6837 }
6838 sk = sk2;
6839 }
6840 }
6841
6842 return sk;
6843 }
6844
BPF_CALL_5(bpf_skc_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6845 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6846 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6847 {
6848 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6849 netns_id, flags);
6850 }
6851
6852 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6853 .func = bpf_skc_lookup_tcp,
6854 .gpl_only = false,
6855 .pkt_access = true,
6856 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6857 .arg1_type = ARG_PTR_TO_CTX,
6858 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6859 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6860 .arg4_type = ARG_ANYTHING,
6861 .arg5_type = ARG_ANYTHING,
6862 };
6863
BPF_CALL_5(bpf_sk_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6864 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6865 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6866 {
6867 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6868 netns_id, flags);
6869 }
6870
6871 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6872 .func = bpf_sk_lookup_tcp,
6873 .gpl_only = false,
6874 .pkt_access = true,
6875 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6876 .arg1_type = ARG_PTR_TO_CTX,
6877 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6878 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6879 .arg4_type = ARG_ANYTHING,
6880 .arg5_type = ARG_ANYTHING,
6881 };
6882
BPF_CALL_5(bpf_sk_lookup_udp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6883 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6884 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6885 {
6886 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6887 netns_id, flags);
6888 }
6889
6890 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6891 .func = bpf_sk_lookup_udp,
6892 .gpl_only = false,
6893 .pkt_access = true,
6894 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6895 .arg1_type = ARG_PTR_TO_CTX,
6896 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6897 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6898 .arg4_type = ARG_ANYTHING,
6899 .arg5_type = ARG_ANYTHING,
6900 };
6901
BPF_CALL_5(bpf_tc_skc_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6902 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6903 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6904 {
6905 struct net_device *dev = skb->dev;
6906 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6907 struct net *caller_net = dev_net(dev);
6908
6909 return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6910 ifindex, IPPROTO_TCP, netns_id,
6911 flags, sdif);
6912 }
6913
6914 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6915 .func = bpf_tc_skc_lookup_tcp,
6916 .gpl_only = false,
6917 .pkt_access = true,
6918 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6919 .arg1_type = ARG_PTR_TO_CTX,
6920 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6921 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6922 .arg4_type = ARG_ANYTHING,
6923 .arg5_type = ARG_ANYTHING,
6924 };
6925
BPF_CALL_5(bpf_tc_sk_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6926 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6927 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6928 {
6929 struct net_device *dev = skb->dev;
6930 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6931 struct net *caller_net = dev_net(dev);
6932
6933 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6934 ifindex, IPPROTO_TCP, netns_id,
6935 flags, sdif);
6936 }
6937
6938 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6939 .func = bpf_tc_sk_lookup_tcp,
6940 .gpl_only = false,
6941 .pkt_access = true,
6942 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6943 .arg1_type = ARG_PTR_TO_CTX,
6944 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6945 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6946 .arg4_type = ARG_ANYTHING,
6947 .arg5_type = ARG_ANYTHING,
6948 };
6949
BPF_CALL_5(bpf_tc_sk_lookup_udp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6950 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6951 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6952 {
6953 struct net_device *dev = skb->dev;
6954 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6955 struct net *caller_net = dev_net(dev);
6956
6957 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6958 ifindex, IPPROTO_UDP, netns_id,
6959 flags, sdif);
6960 }
6961
6962 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6963 .func = bpf_tc_sk_lookup_udp,
6964 .gpl_only = false,
6965 .pkt_access = true,
6966 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6967 .arg1_type = ARG_PTR_TO_CTX,
6968 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6969 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
6970 .arg4_type = ARG_ANYTHING,
6971 .arg5_type = ARG_ANYTHING,
6972 };
6973
BPF_CALL_1(bpf_sk_release,struct sock *,sk)6974 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6975 {
6976 if (sk && sk_is_refcounted(sk))
6977 sock_gen_put(sk);
6978 return 0;
6979 }
6980
6981 static const struct bpf_func_proto bpf_sk_release_proto = {
6982 .func = bpf_sk_release,
6983 .gpl_only = false,
6984 .ret_type = RET_INTEGER,
6985 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6986 };
6987
BPF_CALL_5(bpf_xdp_sk_lookup_udp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)6988 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6989 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6990 {
6991 struct net_device *dev = ctx->rxq->dev;
6992 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6993 struct net *caller_net = dev_net(dev);
6994
6995 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6996 ifindex, IPPROTO_UDP, netns_id,
6997 flags, sdif);
6998 }
6999
7000 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
7001 .func = bpf_xdp_sk_lookup_udp,
7002 .gpl_only = false,
7003 .pkt_access = true,
7004 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7005 .arg1_type = ARG_PTR_TO_CTX,
7006 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7007 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7008 .arg4_type = ARG_ANYTHING,
7009 .arg5_type = ARG_ANYTHING,
7010 };
7011
BPF_CALL_5(bpf_xdp_skc_lookup_tcp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)7012 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
7013 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
7014 {
7015 struct net_device *dev = ctx->rxq->dev;
7016 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
7017 struct net *caller_net = dev_net(dev);
7018
7019 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
7020 ifindex, IPPROTO_TCP, netns_id,
7021 flags, sdif);
7022 }
7023
7024 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
7025 .func = bpf_xdp_skc_lookup_tcp,
7026 .gpl_only = false,
7027 .pkt_access = true,
7028 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
7029 .arg1_type = ARG_PTR_TO_CTX,
7030 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7031 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7032 .arg4_type = ARG_ANYTHING,
7033 .arg5_type = ARG_ANYTHING,
7034 };
7035
BPF_CALL_5(bpf_xdp_sk_lookup_tcp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)7036 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
7037 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
7038 {
7039 struct net_device *dev = ctx->rxq->dev;
7040 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
7041 struct net *caller_net = dev_net(dev);
7042
7043 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
7044 ifindex, IPPROTO_TCP, netns_id,
7045 flags, sdif);
7046 }
7047
7048 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
7049 .func = bpf_xdp_sk_lookup_tcp,
7050 .gpl_only = false,
7051 .pkt_access = true,
7052 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7053 .arg1_type = ARG_PTR_TO_CTX,
7054 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7055 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7056 .arg4_type = ARG_ANYTHING,
7057 .arg5_type = ARG_ANYTHING,
7058 };
7059
BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp,struct bpf_sock_addr_kern *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)7060 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
7061 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7062 {
7063 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
7064 sock_net(ctx->sk), 0,
7065 IPPROTO_TCP, netns_id, flags,
7066 -1);
7067 }
7068
7069 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
7070 .func = bpf_sock_addr_skc_lookup_tcp,
7071 .gpl_only = false,
7072 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
7073 .arg1_type = ARG_PTR_TO_CTX,
7074 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7075 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7076 .arg4_type = ARG_ANYTHING,
7077 .arg5_type = ARG_ANYTHING,
7078 };
7079
BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp,struct bpf_sock_addr_kern *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)7080 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
7081 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7082 {
7083 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
7084 sock_net(ctx->sk), 0, IPPROTO_TCP,
7085 netns_id, flags, -1);
7086 }
7087
7088 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
7089 .func = bpf_sock_addr_sk_lookup_tcp,
7090 .gpl_only = false,
7091 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7092 .arg1_type = ARG_PTR_TO_CTX,
7093 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7094 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7095 .arg4_type = ARG_ANYTHING,
7096 .arg5_type = ARG_ANYTHING,
7097 };
7098
BPF_CALL_5(bpf_sock_addr_sk_lookup_udp,struct bpf_sock_addr_kern *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)7099 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
7100 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7101 {
7102 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
7103 sock_net(ctx->sk), 0, IPPROTO_UDP,
7104 netns_id, flags, -1);
7105 }
7106
7107 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
7108 .func = bpf_sock_addr_sk_lookup_udp,
7109 .gpl_only = false,
7110 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7111 .arg1_type = ARG_PTR_TO_CTX,
7112 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7113 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7114 .arg4_type = ARG_ANYTHING,
7115 .arg5_type = ARG_ANYTHING,
7116 };
7117
bpf_tcp_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)7118 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7119 struct bpf_insn_access_aux *info)
7120 {
7121 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
7122 icsk_retransmits))
7123 return false;
7124
7125 if (off % size != 0)
7126 return false;
7127
7128 switch (off) {
7129 case offsetof(struct bpf_tcp_sock, bytes_received):
7130 case offsetof(struct bpf_tcp_sock, bytes_acked):
7131 return size == sizeof(__u64);
7132 default:
7133 return size == sizeof(__u32);
7134 }
7135 }
7136
bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)7137 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
7138 const struct bpf_insn *si,
7139 struct bpf_insn *insn_buf,
7140 struct bpf_prog *prog, u32 *target_size)
7141 {
7142 struct bpf_insn *insn = insn_buf;
7143
7144 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
7145 do { \
7146 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
7147 sizeof_field(struct bpf_tcp_sock, FIELD)); \
7148 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
7149 si->dst_reg, si->src_reg, \
7150 offsetof(struct tcp_sock, FIELD)); \
7151 } while (0)
7152
7153 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
7154 do { \
7155 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
7156 FIELD) > \
7157 sizeof_field(struct bpf_tcp_sock, FIELD)); \
7158 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7159 struct inet_connection_sock, \
7160 FIELD), \
7161 si->dst_reg, si->src_reg, \
7162 offsetof( \
7163 struct inet_connection_sock, \
7164 FIELD)); \
7165 } while (0)
7166
7167 BTF_TYPE_EMIT(struct bpf_tcp_sock);
7168
7169 switch (si->off) {
7170 case offsetof(struct bpf_tcp_sock, rtt_min):
7171 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
7172 sizeof(struct minmax));
7173 BUILD_BUG_ON(sizeof(struct minmax) <
7174 sizeof(struct minmax_sample));
7175
7176 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7177 offsetof(struct tcp_sock, rtt_min) +
7178 offsetof(struct minmax_sample, v));
7179 break;
7180 case offsetof(struct bpf_tcp_sock, snd_cwnd):
7181 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
7182 break;
7183 case offsetof(struct bpf_tcp_sock, srtt_us):
7184 BPF_TCP_SOCK_GET_COMMON(srtt_us);
7185 break;
7186 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
7187 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
7188 break;
7189 case offsetof(struct bpf_tcp_sock, rcv_nxt):
7190 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
7191 break;
7192 case offsetof(struct bpf_tcp_sock, snd_nxt):
7193 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
7194 break;
7195 case offsetof(struct bpf_tcp_sock, snd_una):
7196 BPF_TCP_SOCK_GET_COMMON(snd_una);
7197 break;
7198 case offsetof(struct bpf_tcp_sock, mss_cache):
7199 BPF_TCP_SOCK_GET_COMMON(mss_cache);
7200 break;
7201 case offsetof(struct bpf_tcp_sock, ecn_flags):
7202 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
7203 break;
7204 case offsetof(struct bpf_tcp_sock, rate_delivered):
7205 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
7206 break;
7207 case offsetof(struct bpf_tcp_sock, rate_interval_us):
7208 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
7209 break;
7210 case offsetof(struct bpf_tcp_sock, packets_out):
7211 BPF_TCP_SOCK_GET_COMMON(packets_out);
7212 break;
7213 case offsetof(struct bpf_tcp_sock, retrans_out):
7214 BPF_TCP_SOCK_GET_COMMON(retrans_out);
7215 break;
7216 case offsetof(struct bpf_tcp_sock, total_retrans):
7217 BPF_TCP_SOCK_GET_COMMON(total_retrans);
7218 break;
7219 case offsetof(struct bpf_tcp_sock, segs_in):
7220 BPF_TCP_SOCK_GET_COMMON(segs_in);
7221 break;
7222 case offsetof(struct bpf_tcp_sock, data_segs_in):
7223 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
7224 break;
7225 case offsetof(struct bpf_tcp_sock, segs_out):
7226 BPF_TCP_SOCK_GET_COMMON(segs_out);
7227 break;
7228 case offsetof(struct bpf_tcp_sock, data_segs_out):
7229 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
7230 break;
7231 case offsetof(struct bpf_tcp_sock, lost_out):
7232 BPF_TCP_SOCK_GET_COMMON(lost_out);
7233 break;
7234 case offsetof(struct bpf_tcp_sock, sacked_out):
7235 BPF_TCP_SOCK_GET_COMMON(sacked_out);
7236 break;
7237 case offsetof(struct bpf_tcp_sock, bytes_received):
7238 BPF_TCP_SOCK_GET_COMMON(bytes_received);
7239 break;
7240 case offsetof(struct bpf_tcp_sock, bytes_acked):
7241 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
7242 break;
7243 case offsetof(struct bpf_tcp_sock, dsack_dups):
7244 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
7245 break;
7246 case offsetof(struct bpf_tcp_sock, delivered):
7247 BPF_TCP_SOCK_GET_COMMON(delivered);
7248 break;
7249 case offsetof(struct bpf_tcp_sock, delivered_ce):
7250 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
7251 break;
7252 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
7253 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
7254 break;
7255 }
7256
7257 return insn - insn_buf;
7258 }
7259
BPF_CALL_1(bpf_tcp_sock,struct sock *,sk)7260 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
7261 {
7262 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
7263 return (unsigned long)sk;
7264
7265 return (unsigned long)NULL;
7266 }
7267
7268 const struct bpf_func_proto bpf_tcp_sock_proto = {
7269 .func = bpf_tcp_sock,
7270 .gpl_only = false,
7271 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
7272 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7273 };
7274
BPF_CALL_1(bpf_get_listener_sock,struct sock *,sk)7275 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
7276 {
7277 sk = sk_to_full_sk(sk);
7278
7279 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
7280 return (unsigned long)sk;
7281
7282 return (unsigned long)NULL;
7283 }
7284
7285 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
7286 .func = bpf_get_listener_sock,
7287 .gpl_only = false,
7288 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7289 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7290 };
7291
BPF_CALL_1(bpf_skb_ecn_set_ce,struct sk_buff *,skb)7292 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7293 {
7294 unsigned int iphdr_len;
7295
7296 switch (skb_protocol(skb, true)) {
7297 case cpu_to_be16(ETH_P_IP):
7298 iphdr_len = sizeof(struct iphdr);
7299 break;
7300 case cpu_to_be16(ETH_P_IPV6):
7301 iphdr_len = sizeof(struct ipv6hdr);
7302 break;
7303 default:
7304 return 0;
7305 }
7306
7307 if (skb_headlen(skb) < iphdr_len)
7308 return 0;
7309
7310 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7311 return 0;
7312
7313 return INET_ECN_set_ce(skb);
7314 }
7315
bpf_xdp_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)7316 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7317 struct bpf_insn_access_aux *info)
7318 {
7319 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7320 return false;
7321
7322 if (off % size != 0)
7323 return false;
7324
7325 switch (off) {
7326 default:
7327 return size == sizeof(__u32);
7328 }
7329 }
7330
bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)7331 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7332 const struct bpf_insn *si,
7333 struct bpf_insn *insn_buf,
7334 struct bpf_prog *prog, u32 *target_size)
7335 {
7336 struct bpf_insn *insn = insn_buf;
7337
7338 #define BPF_XDP_SOCK_GET(FIELD) \
7339 do { \
7340 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
7341 sizeof_field(struct bpf_xdp_sock, FIELD)); \
7342 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7343 si->dst_reg, si->src_reg, \
7344 offsetof(struct xdp_sock, FIELD)); \
7345 } while (0)
7346
7347 switch (si->off) {
7348 case offsetof(struct bpf_xdp_sock, queue_id):
7349 BPF_XDP_SOCK_GET(queue_id);
7350 break;
7351 }
7352
7353 return insn - insn_buf;
7354 }
7355
7356 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7357 .func = bpf_skb_ecn_set_ce,
7358 .gpl_only = false,
7359 .ret_type = RET_INTEGER,
7360 .arg1_type = ARG_PTR_TO_CTX,
7361 };
7362
BPF_CALL_5(bpf_tcp_check_syncookie,struct sock *,sk,void *,iph,u32,iph_len,struct tcphdr *,th,u32,th_len)7363 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7364 struct tcphdr *, th, u32, th_len)
7365 {
7366 #ifdef CONFIG_SYN_COOKIES
7367 int ret;
7368
7369 if (unlikely(!sk || th_len < sizeof(*th)))
7370 return -EINVAL;
7371
7372 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7373 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7374 return -EINVAL;
7375
7376 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7377 return -EINVAL;
7378
7379 if (!th->ack || th->rst || th->syn)
7380 return -ENOENT;
7381
7382 if (unlikely(iph_len < sizeof(struct iphdr)))
7383 return -EINVAL;
7384
7385 if (tcp_synq_no_recent_overflow(sk))
7386 return -ENOENT;
7387
7388 /* Both struct iphdr and struct ipv6hdr have the version field at the
7389 * same offset so we can cast to the shorter header (struct iphdr).
7390 */
7391 switch (((struct iphdr *)iph)->version) {
7392 case 4:
7393 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7394 return -EINVAL;
7395
7396 ret = __cookie_v4_check((struct iphdr *)iph, th);
7397 break;
7398
7399 #if IS_BUILTIN(CONFIG_IPV6)
7400 case 6:
7401 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7402 return -EINVAL;
7403
7404 if (sk->sk_family != AF_INET6)
7405 return -EINVAL;
7406
7407 ret = __cookie_v6_check((struct ipv6hdr *)iph, th);
7408 break;
7409 #endif /* CONFIG_IPV6 */
7410
7411 default:
7412 return -EPROTONOSUPPORT;
7413 }
7414
7415 if (ret > 0)
7416 return 0;
7417
7418 return -ENOENT;
7419 #else
7420 return -ENOTSUPP;
7421 #endif
7422 }
7423
7424 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7425 .func = bpf_tcp_check_syncookie,
7426 .gpl_only = true,
7427 .pkt_access = true,
7428 .ret_type = RET_INTEGER,
7429 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7430 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7431 .arg3_type = ARG_CONST_SIZE,
7432 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7433 .arg5_type = ARG_CONST_SIZE,
7434 };
7435
BPF_CALL_5(bpf_tcp_gen_syncookie,struct sock *,sk,void *,iph,u32,iph_len,struct tcphdr *,th,u32,th_len)7436 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7437 struct tcphdr *, th, u32, th_len)
7438 {
7439 #ifdef CONFIG_SYN_COOKIES
7440 u32 cookie;
7441 u16 mss;
7442
7443 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7444 return -EINVAL;
7445
7446 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7447 return -EINVAL;
7448
7449 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7450 return -ENOENT;
7451
7452 if (!th->syn || th->ack || th->fin || th->rst)
7453 return -EINVAL;
7454
7455 if (unlikely(iph_len < sizeof(struct iphdr)))
7456 return -EINVAL;
7457
7458 /* Both struct iphdr and struct ipv6hdr have the version field at the
7459 * same offset so we can cast to the shorter header (struct iphdr).
7460 */
7461 switch (((struct iphdr *)iph)->version) {
7462 case 4:
7463 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7464 return -EINVAL;
7465
7466 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7467 break;
7468
7469 #if IS_BUILTIN(CONFIG_IPV6)
7470 case 6:
7471 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7472 return -EINVAL;
7473
7474 if (sk->sk_family != AF_INET6)
7475 return -EINVAL;
7476
7477 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7478 break;
7479 #endif /* CONFIG_IPV6 */
7480
7481 default:
7482 return -EPROTONOSUPPORT;
7483 }
7484 if (mss == 0)
7485 return -ENOENT;
7486
7487 return cookie | ((u64)mss << 32);
7488 #else
7489 return -EOPNOTSUPP;
7490 #endif /* CONFIG_SYN_COOKIES */
7491 }
7492
7493 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7494 .func = bpf_tcp_gen_syncookie,
7495 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
7496 .pkt_access = true,
7497 .ret_type = RET_INTEGER,
7498 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7499 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7500 .arg3_type = ARG_CONST_SIZE,
7501 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7502 .arg5_type = ARG_CONST_SIZE,
7503 };
7504
BPF_CALL_3(bpf_sk_assign,struct sk_buff *,skb,struct sock *,sk,u64,flags)7505 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7506 {
7507 if (!sk || flags != 0)
7508 return -EINVAL;
7509 if (!skb_at_tc_ingress(skb))
7510 return -EOPNOTSUPP;
7511 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7512 return -ENETUNREACH;
7513 if (sk_unhashed(sk))
7514 return -EOPNOTSUPP;
7515 if (sk_is_refcounted(sk) &&
7516 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7517 return -ENOENT;
7518
7519 skb_orphan(skb);
7520 skb->sk = sk;
7521 skb->destructor = sock_pfree;
7522
7523 return 0;
7524 }
7525
7526 static const struct bpf_func_proto bpf_sk_assign_proto = {
7527 .func = bpf_sk_assign,
7528 .gpl_only = false,
7529 .ret_type = RET_INTEGER,
7530 .arg1_type = ARG_PTR_TO_CTX,
7531 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7532 .arg3_type = ARG_ANYTHING,
7533 };
7534
bpf_search_tcp_opt(const u8 * op,const u8 * opend,u8 search_kind,const u8 * magic,u8 magic_len,bool * eol)7535 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7536 u8 search_kind, const u8 *magic,
7537 u8 magic_len, bool *eol)
7538 {
7539 u8 kind, kind_len;
7540
7541 *eol = false;
7542
7543 while (op < opend) {
7544 kind = op[0];
7545
7546 if (kind == TCPOPT_EOL) {
7547 *eol = true;
7548 return ERR_PTR(-ENOMSG);
7549 } else if (kind == TCPOPT_NOP) {
7550 op++;
7551 continue;
7552 }
7553
7554 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7555 /* Something is wrong in the received header.
7556 * Follow the TCP stack's tcp_parse_options()
7557 * and just bail here.
7558 */
7559 return ERR_PTR(-EFAULT);
7560
7561 kind_len = op[1];
7562 if (search_kind == kind) {
7563 if (!magic_len)
7564 return op;
7565
7566 if (magic_len > kind_len - 2)
7567 return ERR_PTR(-ENOMSG);
7568
7569 if (!memcmp(&op[2], magic, magic_len))
7570 return op;
7571 }
7572
7573 op += kind_len;
7574 }
7575
7576 return ERR_PTR(-ENOMSG);
7577 }
7578
BPF_CALL_4(bpf_sock_ops_load_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,void *,search_res,u32,len,u64,flags)7579 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7580 void *, search_res, u32, len, u64, flags)
7581 {
7582 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7583 const u8 *op, *opend, *magic, *search = search_res;
7584 u8 search_kind, search_len, copy_len, magic_len;
7585 int ret;
7586
7587 /* 2 byte is the minimal option len except TCPOPT_NOP and
7588 * TCPOPT_EOL which are useless for the bpf prog to learn
7589 * and this helper disallow loading them also.
7590 */
7591 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7592 return -EINVAL;
7593
7594 search_kind = search[0];
7595 search_len = search[1];
7596
7597 if (search_len > len || search_kind == TCPOPT_NOP ||
7598 search_kind == TCPOPT_EOL)
7599 return -EINVAL;
7600
7601 if (search_kind == TCPOPT_EXP || search_kind == 253) {
7602 /* 16 or 32 bit magic. +2 for kind and kind length */
7603 if (search_len != 4 && search_len != 6)
7604 return -EINVAL;
7605 magic = &search[2];
7606 magic_len = search_len - 2;
7607 } else {
7608 if (search_len)
7609 return -EINVAL;
7610 magic = NULL;
7611 magic_len = 0;
7612 }
7613
7614 if (load_syn) {
7615 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7616 if (ret < 0)
7617 return ret;
7618
7619 opend = op + ret;
7620 op += sizeof(struct tcphdr);
7621 } else {
7622 if (!bpf_sock->skb ||
7623 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7624 /* This bpf_sock->op cannot call this helper */
7625 return -EPERM;
7626
7627 opend = bpf_sock->skb_data_end;
7628 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7629 }
7630
7631 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7632 &eol);
7633 if (IS_ERR(op))
7634 return PTR_ERR(op);
7635
7636 copy_len = op[1];
7637 ret = copy_len;
7638 if (copy_len > len) {
7639 ret = -ENOSPC;
7640 copy_len = len;
7641 }
7642
7643 memcpy(search_res, op, copy_len);
7644 return ret;
7645 }
7646
7647 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7648 .func = bpf_sock_ops_load_hdr_opt,
7649 .gpl_only = false,
7650 .ret_type = RET_INTEGER,
7651 .arg1_type = ARG_PTR_TO_CTX,
7652 .arg2_type = ARG_PTR_TO_MEM,
7653 .arg3_type = ARG_CONST_SIZE,
7654 .arg4_type = ARG_ANYTHING,
7655 };
7656
BPF_CALL_4(bpf_sock_ops_store_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,const void *,from,u32,len,u64,flags)7657 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7658 const void *, from, u32, len, u64, flags)
7659 {
7660 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7661 const u8 *op, *new_op, *magic = NULL;
7662 struct sk_buff *skb;
7663 bool eol;
7664
7665 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7666 return -EPERM;
7667
7668 if (len < 2 || flags)
7669 return -EINVAL;
7670
7671 new_op = from;
7672 new_kind = new_op[0];
7673 new_kind_len = new_op[1];
7674
7675 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7676 new_kind == TCPOPT_EOL)
7677 return -EINVAL;
7678
7679 if (new_kind_len > bpf_sock->remaining_opt_len)
7680 return -ENOSPC;
7681
7682 /* 253 is another experimental kind */
7683 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7684 if (new_kind_len < 4)
7685 return -EINVAL;
7686 /* Match for the 2 byte magic also.
7687 * RFC 6994: the magic could be 2 or 4 bytes.
7688 * Hence, matching by 2 byte only is on the
7689 * conservative side but it is the right
7690 * thing to do for the 'search-for-duplication'
7691 * purpose.
7692 */
7693 magic = &new_op[2];
7694 magic_len = 2;
7695 }
7696
7697 /* Check for duplication */
7698 skb = bpf_sock->skb;
7699 op = skb->data + sizeof(struct tcphdr);
7700 opend = bpf_sock->skb_data_end;
7701
7702 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7703 &eol);
7704 if (!IS_ERR(op))
7705 return -EEXIST;
7706
7707 if (PTR_ERR(op) != -ENOMSG)
7708 return PTR_ERR(op);
7709
7710 if (eol)
7711 /* The option has been ended. Treat it as no more
7712 * header option can be written.
7713 */
7714 return -ENOSPC;
7715
7716 /* No duplication found. Store the header option. */
7717 memcpy(opend, from, new_kind_len);
7718
7719 bpf_sock->remaining_opt_len -= new_kind_len;
7720 bpf_sock->skb_data_end += new_kind_len;
7721
7722 return 0;
7723 }
7724
7725 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7726 .func = bpf_sock_ops_store_hdr_opt,
7727 .gpl_only = false,
7728 .ret_type = RET_INTEGER,
7729 .arg1_type = ARG_PTR_TO_CTX,
7730 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7731 .arg3_type = ARG_CONST_SIZE,
7732 .arg4_type = ARG_ANYTHING,
7733 };
7734
BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,u32,len,u64,flags)7735 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7736 u32, len, u64, flags)
7737 {
7738 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7739 return -EPERM;
7740
7741 if (flags || len < 2)
7742 return -EINVAL;
7743
7744 if (len > bpf_sock->remaining_opt_len)
7745 return -ENOSPC;
7746
7747 bpf_sock->remaining_opt_len -= len;
7748
7749 return 0;
7750 }
7751
7752 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7753 .func = bpf_sock_ops_reserve_hdr_opt,
7754 .gpl_only = false,
7755 .ret_type = RET_INTEGER,
7756 .arg1_type = ARG_PTR_TO_CTX,
7757 .arg2_type = ARG_ANYTHING,
7758 .arg3_type = ARG_ANYTHING,
7759 };
7760
BPF_CALL_3(bpf_skb_set_tstamp,struct sk_buff *,skb,u64,tstamp,u32,tstamp_type)7761 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7762 u64, tstamp, u32, tstamp_type)
7763 {
7764 /* skb_clear_delivery_time() is done for inet protocol */
7765 if (skb->protocol != htons(ETH_P_IP) &&
7766 skb->protocol != htons(ETH_P_IPV6))
7767 return -EOPNOTSUPP;
7768
7769 switch (tstamp_type) {
7770 case BPF_SKB_CLOCK_REALTIME:
7771 skb->tstamp = tstamp;
7772 skb->tstamp_type = SKB_CLOCK_REALTIME;
7773 break;
7774 case BPF_SKB_CLOCK_MONOTONIC:
7775 if (!tstamp)
7776 return -EINVAL;
7777 skb->tstamp = tstamp;
7778 skb->tstamp_type = SKB_CLOCK_MONOTONIC;
7779 break;
7780 case BPF_SKB_CLOCK_TAI:
7781 if (!tstamp)
7782 return -EINVAL;
7783 skb->tstamp = tstamp;
7784 skb->tstamp_type = SKB_CLOCK_TAI;
7785 break;
7786 default:
7787 return -EINVAL;
7788 }
7789
7790 return 0;
7791 }
7792
7793 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7794 .func = bpf_skb_set_tstamp,
7795 .gpl_only = false,
7796 .ret_type = RET_INTEGER,
7797 .arg1_type = ARG_PTR_TO_CTX,
7798 .arg2_type = ARG_ANYTHING,
7799 .arg3_type = ARG_ANYTHING,
7800 };
7801
7802 #ifdef CONFIG_SYN_COOKIES
BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4,struct iphdr *,iph,struct tcphdr *,th,u32,th_len)7803 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7804 struct tcphdr *, th, u32, th_len)
7805 {
7806 u32 cookie;
7807 u16 mss;
7808
7809 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7810 return -EINVAL;
7811
7812 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7813 cookie = __cookie_v4_init_sequence(iph, th, &mss);
7814
7815 return cookie | ((u64)mss << 32);
7816 }
7817
7818 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7819 .func = bpf_tcp_raw_gen_syncookie_ipv4,
7820 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */
7821 .pkt_access = true,
7822 .ret_type = RET_INTEGER,
7823 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7824 .arg1_size = sizeof(struct iphdr),
7825 .arg2_type = ARG_PTR_TO_MEM,
7826 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7827 };
7828
BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6,struct ipv6hdr *,iph,struct tcphdr *,th,u32,th_len)7829 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7830 struct tcphdr *, th, u32, th_len)
7831 {
7832 #if IS_BUILTIN(CONFIG_IPV6)
7833 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7834 sizeof(struct ipv6hdr);
7835 u32 cookie;
7836 u16 mss;
7837
7838 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7839 return -EINVAL;
7840
7841 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7842 cookie = __cookie_v6_init_sequence(iph, th, &mss);
7843
7844 return cookie | ((u64)mss << 32);
7845 #else
7846 return -EPROTONOSUPPORT;
7847 #endif
7848 }
7849
7850 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7851 .func = bpf_tcp_raw_gen_syncookie_ipv6,
7852 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */
7853 .pkt_access = true,
7854 .ret_type = RET_INTEGER,
7855 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7856 .arg1_size = sizeof(struct ipv6hdr),
7857 .arg2_type = ARG_PTR_TO_MEM,
7858 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7859 };
7860
BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4,struct iphdr *,iph,struct tcphdr *,th)7861 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7862 struct tcphdr *, th)
7863 {
7864 if (__cookie_v4_check(iph, th) > 0)
7865 return 0;
7866
7867 return -EACCES;
7868 }
7869
7870 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7871 .func = bpf_tcp_raw_check_syncookie_ipv4,
7872 .gpl_only = true, /* __cookie_v4_check is GPL */
7873 .pkt_access = true,
7874 .ret_type = RET_INTEGER,
7875 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7876 .arg1_size = sizeof(struct iphdr),
7877 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7878 .arg2_size = sizeof(struct tcphdr),
7879 };
7880
BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6,struct ipv6hdr *,iph,struct tcphdr *,th)7881 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7882 struct tcphdr *, th)
7883 {
7884 #if IS_BUILTIN(CONFIG_IPV6)
7885 if (__cookie_v6_check(iph, th) > 0)
7886 return 0;
7887
7888 return -EACCES;
7889 #else
7890 return -EPROTONOSUPPORT;
7891 #endif
7892 }
7893
7894 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7895 .func = bpf_tcp_raw_check_syncookie_ipv6,
7896 .gpl_only = true, /* __cookie_v6_check is GPL */
7897 .pkt_access = true,
7898 .ret_type = RET_INTEGER,
7899 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7900 .arg1_size = sizeof(struct ipv6hdr),
7901 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7902 .arg2_size = sizeof(struct tcphdr),
7903 };
7904 #endif /* CONFIG_SYN_COOKIES */
7905
7906 #endif /* CONFIG_INET */
7907
bpf_helper_changes_pkt_data(void * func)7908 bool bpf_helper_changes_pkt_data(void *func)
7909 {
7910 if (func == bpf_skb_vlan_push ||
7911 func == bpf_skb_vlan_pop ||
7912 func == bpf_skb_store_bytes ||
7913 func == bpf_skb_change_proto ||
7914 func == bpf_skb_change_head ||
7915 func == sk_skb_change_head ||
7916 func == bpf_skb_change_tail ||
7917 func == sk_skb_change_tail ||
7918 func == bpf_skb_adjust_room ||
7919 func == sk_skb_adjust_room ||
7920 func == bpf_skb_pull_data ||
7921 func == sk_skb_pull_data ||
7922 func == bpf_clone_redirect ||
7923 func == bpf_l3_csum_replace ||
7924 func == bpf_l4_csum_replace ||
7925 func == bpf_xdp_adjust_head ||
7926 func == bpf_xdp_adjust_meta ||
7927 func == bpf_msg_pull_data ||
7928 func == bpf_msg_push_data ||
7929 func == bpf_msg_pop_data ||
7930 func == bpf_xdp_adjust_tail ||
7931 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7932 func == bpf_lwt_seg6_store_bytes ||
7933 func == bpf_lwt_seg6_adjust_srh ||
7934 func == bpf_lwt_seg6_action ||
7935 #endif
7936 #ifdef CONFIG_INET
7937 func == bpf_sock_ops_store_hdr_opt ||
7938 #endif
7939 func == bpf_lwt_in_push_encap ||
7940 func == bpf_lwt_xmit_push_encap)
7941 return true;
7942
7943 return false;
7944 }
7945
7946 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7947 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7948
7949 static const struct bpf_func_proto *
sock_filter_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)7950 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7951 {
7952 const struct bpf_func_proto *func_proto;
7953
7954 func_proto = cgroup_common_func_proto(func_id, prog);
7955 if (func_proto)
7956 return func_proto;
7957
7958 func_proto = cgroup_current_func_proto(func_id, prog);
7959 if (func_proto)
7960 return func_proto;
7961
7962 switch (func_id) {
7963 case BPF_FUNC_get_socket_cookie:
7964 return &bpf_get_socket_cookie_sock_proto;
7965 case BPF_FUNC_get_netns_cookie:
7966 return &bpf_get_netns_cookie_sock_proto;
7967 case BPF_FUNC_perf_event_output:
7968 return &bpf_event_output_data_proto;
7969 case BPF_FUNC_sk_storage_get:
7970 return &bpf_sk_storage_get_cg_sock_proto;
7971 case BPF_FUNC_ktime_get_coarse_ns:
7972 return &bpf_ktime_get_coarse_ns_proto;
7973 default:
7974 return bpf_base_func_proto(func_id, prog);
7975 }
7976 }
7977
7978 static const struct bpf_func_proto *
sock_addr_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)7979 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7980 {
7981 const struct bpf_func_proto *func_proto;
7982
7983 func_proto = cgroup_common_func_proto(func_id, prog);
7984 if (func_proto)
7985 return func_proto;
7986
7987 func_proto = cgroup_current_func_proto(func_id, prog);
7988 if (func_proto)
7989 return func_proto;
7990
7991 switch (func_id) {
7992 case BPF_FUNC_bind:
7993 switch (prog->expected_attach_type) {
7994 case BPF_CGROUP_INET4_CONNECT:
7995 case BPF_CGROUP_INET6_CONNECT:
7996 return &bpf_bind_proto;
7997 default:
7998 return NULL;
7999 }
8000 case BPF_FUNC_get_socket_cookie:
8001 return &bpf_get_socket_cookie_sock_addr_proto;
8002 case BPF_FUNC_get_netns_cookie:
8003 return &bpf_get_netns_cookie_sock_addr_proto;
8004 case BPF_FUNC_perf_event_output:
8005 return &bpf_event_output_data_proto;
8006 #ifdef CONFIG_INET
8007 case BPF_FUNC_sk_lookup_tcp:
8008 return &bpf_sock_addr_sk_lookup_tcp_proto;
8009 case BPF_FUNC_sk_lookup_udp:
8010 return &bpf_sock_addr_sk_lookup_udp_proto;
8011 case BPF_FUNC_sk_release:
8012 return &bpf_sk_release_proto;
8013 case BPF_FUNC_skc_lookup_tcp:
8014 return &bpf_sock_addr_skc_lookup_tcp_proto;
8015 #endif /* CONFIG_INET */
8016 case BPF_FUNC_sk_storage_get:
8017 return &bpf_sk_storage_get_proto;
8018 case BPF_FUNC_sk_storage_delete:
8019 return &bpf_sk_storage_delete_proto;
8020 case BPF_FUNC_setsockopt:
8021 switch (prog->expected_attach_type) {
8022 case BPF_CGROUP_INET4_BIND:
8023 case BPF_CGROUP_INET6_BIND:
8024 case BPF_CGROUP_INET4_CONNECT:
8025 case BPF_CGROUP_INET6_CONNECT:
8026 case BPF_CGROUP_UNIX_CONNECT:
8027 case BPF_CGROUP_UDP4_RECVMSG:
8028 case BPF_CGROUP_UDP6_RECVMSG:
8029 case BPF_CGROUP_UNIX_RECVMSG:
8030 case BPF_CGROUP_UDP4_SENDMSG:
8031 case BPF_CGROUP_UDP6_SENDMSG:
8032 case BPF_CGROUP_UNIX_SENDMSG:
8033 case BPF_CGROUP_INET4_GETPEERNAME:
8034 case BPF_CGROUP_INET6_GETPEERNAME:
8035 case BPF_CGROUP_UNIX_GETPEERNAME:
8036 case BPF_CGROUP_INET4_GETSOCKNAME:
8037 case BPF_CGROUP_INET6_GETSOCKNAME:
8038 case BPF_CGROUP_UNIX_GETSOCKNAME:
8039 return &bpf_sock_addr_setsockopt_proto;
8040 default:
8041 return NULL;
8042 }
8043 case BPF_FUNC_getsockopt:
8044 switch (prog->expected_attach_type) {
8045 case BPF_CGROUP_INET4_BIND:
8046 case BPF_CGROUP_INET6_BIND:
8047 case BPF_CGROUP_INET4_CONNECT:
8048 case BPF_CGROUP_INET6_CONNECT:
8049 case BPF_CGROUP_UNIX_CONNECT:
8050 case BPF_CGROUP_UDP4_RECVMSG:
8051 case BPF_CGROUP_UDP6_RECVMSG:
8052 case BPF_CGROUP_UNIX_RECVMSG:
8053 case BPF_CGROUP_UDP4_SENDMSG:
8054 case BPF_CGROUP_UDP6_SENDMSG:
8055 case BPF_CGROUP_UNIX_SENDMSG:
8056 case BPF_CGROUP_INET4_GETPEERNAME:
8057 case BPF_CGROUP_INET6_GETPEERNAME:
8058 case BPF_CGROUP_UNIX_GETPEERNAME:
8059 case BPF_CGROUP_INET4_GETSOCKNAME:
8060 case BPF_CGROUP_INET6_GETSOCKNAME:
8061 case BPF_CGROUP_UNIX_GETSOCKNAME:
8062 return &bpf_sock_addr_getsockopt_proto;
8063 default:
8064 return NULL;
8065 }
8066 default:
8067 return bpf_sk_base_func_proto(func_id, prog);
8068 }
8069 }
8070
8071 static const struct bpf_func_proto *
sk_filter_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8072 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8073 {
8074 switch (func_id) {
8075 case BPF_FUNC_skb_load_bytes:
8076 return &bpf_skb_load_bytes_proto;
8077 case BPF_FUNC_skb_load_bytes_relative:
8078 return &bpf_skb_load_bytes_relative_proto;
8079 case BPF_FUNC_get_socket_cookie:
8080 return &bpf_get_socket_cookie_proto;
8081 case BPF_FUNC_get_socket_uid:
8082 return &bpf_get_socket_uid_proto;
8083 case BPF_FUNC_perf_event_output:
8084 return &bpf_skb_event_output_proto;
8085 default:
8086 return bpf_sk_base_func_proto(func_id, prog);
8087 }
8088 }
8089
8090 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
8091 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
8092
8093 static const struct bpf_func_proto *
cg_skb_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8094 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8095 {
8096 const struct bpf_func_proto *func_proto;
8097
8098 func_proto = cgroup_common_func_proto(func_id, prog);
8099 if (func_proto)
8100 return func_proto;
8101
8102 switch (func_id) {
8103 case BPF_FUNC_sk_fullsock:
8104 return &bpf_sk_fullsock_proto;
8105 case BPF_FUNC_sk_storage_get:
8106 return &bpf_sk_storage_get_proto;
8107 case BPF_FUNC_sk_storage_delete:
8108 return &bpf_sk_storage_delete_proto;
8109 case BPF_FUNC_perf_event_output:
8110 return &bpf_skb_event_output_proto;
8111 #ifdef CONFIG_SOCK_CGROUP_DATA
8112 case BPF_FUNC_skb_cgroup_id:
8113 return &bpf_skb_cgroup_id_proto;
8114 case BPF_FUNC_skb_ancestor_cgroup_id:
8115 return &bpf_skb_ancestor_cgroup_id_proto;
8116 case BPF_FUNC_sk_cgroup_id:
8117 return &bpf_sk_cgroup_id_proto;
8118 case BPF_FUNC_sk_ancestor_cgroup_id:
8119 return &bpf_sk_ancestor_cgroup_id_proto;
8120 #endif
8121 #ifdef CONFIG_INET
8122 case BPF_FUNC_sk_lookup_tcp:
8123 return &bpf_sk_lookup_tcp_proto;
8124 case BPF_FUNC_sk_lookup_udp:
8125 return &bpf_sk_lookup_udp_proto;
8126 case BPF_FUNC_sk_release:
8127 return &bpf_sk_release_proto;
8128 case BPF_FUNC_skc_lookup_tcp:
8129 return &bpf_skc_lookup_tcp_proto;
8130 case BPF_FUNC_tcp_sock:
8131 return &bpf_tcp_sock_proto;
8132 case BPF_FUNC_get_listener_sock:
8133 return &bpf_get_listener_sock_proto;
8134 case BPF_FUNC_skb_ecn_set_ce:
8135 return &bpf_skb_ecn_set_ce_proto;
8136 #endif
8137 default:
8138 return sk_filter_func_proto(func_id, prog);
8139 }
8140 }
8141
8142 static const struct bpf_func_proto *
tc_cls_act_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8143 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8144 {
8145 switch (func_id) {
8146 case BPF_FUNC_skb_store_bytes:
8147 return &bpf_skb_store_bytes_proto;
8148 case BPF_FUNC_skb_load_bytes:
8149 return &bpf_skb_load_bytes_proto;
8150 case BPF_FUNC_skb_load_bytes_relative:
8151 return &bpf_skb_load_bytes_relative_proto;
8152 case BPF_FUNC_skb_pull_data:
8153 return &bpf_skb_pull_data_proto;
8154 case BPF_FUNC_csum_diff:
8155 return &bpf_csum_diff_proto;
8156 case BPF_FUNC_csum_update:
8157 return &bpf_csum_update_proto;
8158 case BPF_FUNC_csum_level:
8159 return &bpf_csum_level_proto;
8160 case BPF_FUNC_l3_csum_replace:
8161 return &bpf_l3_csum_replace_proto;
8162 case BPF_FUNC_l4_csum_replace:
8163 return &bpf_l4_csum_replace_proto;
8164 case BPF_FUNC_clone_redirect:
8165 return &bpf_clone_redirect_proto;
8166 case BPF_FUNC_get_cgroup_classid:
8167 return &bpf_get_cgroup_classid_proto;
8168 case BPF_FUNC_skb_vlan_push:
8169 return &bpf_skb_vlan_push_proto;
8170 case BPF_FUNC_skb_vlan_pop:
8171 return &bpf_skb_vlan_pop_proto;
8172 case BPF_FUNC_skb_change_proto:
8173 return &bpf_skb_change_proto_proto;
8174 case BPF_FUNC_skb_change_type:
8175 return &bpf_skb_change_type_proto;
8176 case BPF_FUNC_skb_adjust_room:
8177 return &bpf_skb_adjust_room_proto;
8178 case BPF_FUNC_skb_change_tail:
8179 return &bpf_skb_change_tail_proto;
8180 case BPF_FUNC_skb_change_head:
8181 return &bpf_skb_change_head_proto;
8182 case BPF_FUNC_skb_get_tunnel_key:
8183 return &bpf_skb_get_tunnel_key_proto;
8184 case BPF_FUNC_skb_set_tunnel_key:
8185 return bpf_get_skb_set_tunnel_proto(func_id);
8186 case BPF_FUNC_skb_get_tunnel_opt:
8187 return &bpf_skb_get_tunnel_opt_proto;
8188 case BPF_FUNC_skb_set_tunnel_opt:
8189 return bpf_get_skb_set_tunnel_proto(func_id);
8190 case BPF_FUNC_redirect:
8191 return &bpf_redirect_proto;
8192 case BPF_FUNC_redirect_neigh:
8193 return &bpf_redirect_neigh_proto;
8194 case BPF_FUNC_redirect_peer:
8195 return &bpf_redirect_peer_proto;
8196 case BPF_FUNC_get_route_realm:
8197 return &bpf_get_route_realm_proto;
8198 case BPF_FUNC_get_hash_recalc:
8199 return &bpf_get_hash_recalc_proto;
8200 case BPF_FUNC_set_hash_invalid:
8201 return &bpf_set_hash_invalid_proto;
8202 case BPF_FUNC_set_hash:
8203 return &bpf_set_hash_proto;
8204 case BPF_FUNC_perf_event_output:
8205 return &bpf_skb_event_output_proto;
8206 case BPF_FUNC_get_smp_processor_id:
8207 return &bpf_get_smp_processor_id_proto;
8208 case BPF_FUNC_skb_under_cgroup:
8209 return &bpf_skb_under_cgroup_proto;
8210 case BPF_FUNC_get_socket_cookie:
8211 return &bpf_get_socket_cookie_proto;
8212 case BPF_FUNC_get_socket_uid:
8213 return &bpf_get_socket_uid_proto;
8214 case BPF_FUNC_fib_lookup:
8215 return &bpf_skb_fib_lookup_proto;
8216 case BPF_FUNC_check_mtu:
8217 return &bpf_skb_check_mtu_proto;
8218 case BPF_FUNC_sk_fullsock:
8219 return &bpf_sk_fullsock_proto;
8220 case BPF_FUNC_sk_storage_get:
8221 return &bpf_sk_storage_get_proto;
8222 case BPF_FUNC_sk_storage_delete:
8223 return &bpf_sk_storage_delete_proto;
8224 #ifdef CONFIG_XFRM
8225 case BPF_FUNC_skb_get_xfrm_state:
8226 return &bpf_skb_get_xfrm_state_proto;
8227 #endif
8228 #ifdef CONFIG_CGROUP_NET_CLASSID
8229 case BPF_FUNC_skb_cgroup_classid:
8230 return &bpf_skb_cgroup_classid_proto;
8231 #endif
8232 #ifdef CONFIG_SOCK_CGROUP_DATA
8233 case BPF_FUNC_skb_cgroup_id:
8234 return &bpf_skb_cgroup_id_proto;
8235 case BPF_FUNC_skb_ancestor_cgroup_id:
8236 return &bpf_skb_ancestor_cgroup_id_proto;
8237 #endif
8238 #ifdef CONFIG_INET
8239 case BPF_FUNC_sk_lookup_tcp:
8240 return &bpf_tc_sk_lookup_tcp_proto;
8241 case BPF_FUNC_sk_lookup_udp:
8242 return &bpf_tc_sk_lookup_udp_proto;
8243 case BPF_FUNC_sk_release:
8244 return &bpf_sk_release_proto;
8245 case BPF_FUNC_tcp_sock:
8246 return &bpf_tcp_sock_proto;
8247 case BPF_FUNC_get_listener_sock:
8248 return &bpf_get_listener_sock_proto;
8249 case BPF_FUNC_skc_lookup_tcp:
8250 return &bpf_tc_skc_lookup_tcp_proto;
8251 case BPF_FUNC_tcp_check_syncookie:
8252 return &bpf_tcp_check_syncookie_proto;
8253 case BPF_FUNC_skb_ecn_set_ce:
8254 return &bpf_skb_ecn_set_ce_proto;
8255 case BPF_FUNC_tcp_gen_syncookie:
8256 return &bpf_tcp_gen_syncookie_proto;
8257 case BPF_FUNC_sk_assign:
8258 return &bpf_sk_assign_proto;
8259 case BPF_FUNC_skb_set_tstamp:
8260 return &bpf_skb_set_tstamp_proto;
8261 #ifdef CONFIG_SYN_COOKIES
8262 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8263 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8264 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8265 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8266 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8267 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8268 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8269 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8270 #endif
8271 #endif
8272 default:
8273 return bpf_sk_base_func_proto(func_id, prog);
8274 }
8275 }
8276
8277 static const struct bpf_func_proto *
xdp_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8278 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8279 {
8280 switch (func_id) {
8281 case BPF_FUNC_perf_event_output:
8282 return &bpf_xdp_event_output_proto;
8283 case BPF_FUNC_get_smp_processor_id:
8284 return &bpf_get_smp_processor_id_proto;
8285 case BPF_FUNC_csum_diff:
8286 return &bpf_csum_diff_proto;
8287 case BPF_FUNC_xdp_adjust_head:
8288 return &bpf_xdp_adjust_head_proto;
8289 case BPF_FUNC_xdp_adjust_meta:
8290 return &bpf_xdp_adjust_meta_proto;
8291 case BPF_FUNC_redirect:
8292 return &bpf_xdp_redirect_proto;
8293 case BPF_FUNC_redirect_map:
8294 return &bpf_xdp_redirect_map_proto;
8295 case BPF_FUNC_xdp_adjust_tail:
8296 return &bpf_xdp_adjust_tail_proto;
8297 case BPF_FUNC_xdp_get_buff_len:
8298 return &bpf_xdp_get_buff_len_proto;
8299 case BPF_FUNC_xdp_load_bytes:
8300 return &bpf_xdp_load_bytes_proto;
8301 case BPF_FUNC_xdp_store_bytes:
8302 return &bpf_xdp_store_bytes_proto;
8303 case BPF_FUNC_fib_lookup:
8304 return &bpf_xdp_fib_lookup_proto;
8305 case BPF_FUNC_check_mtu:
8306 return &bpf_xdp_check_mtu_proto;
8307 #ifdef CONFIG_INET
8308 case BPF_FUNC_sk_lookup_udp:
8309 return &bpf_xdp_sk_lookup_udp_proto;
8310 case BPF_FUNC_sk_lookup_tcp:
8311 return &bpf_xdp_sk_lookup_tcp_proto;
8312 case BPF_FUNC_sk_release:
8313 return &bpf_sk_release_proto;
8314 case BPF_FUNC_skc_lookup_tcp:
8315 return &bpf_xdp_skc_lookup_tcp_proto;
8316 case BPF_FUNC_tcp_check_syncookie:
8317 return &bpf_tcp_check_syncookie_proto;
8318 case BPF_FUNC_tcp_gen_syncookie:
8319 return &bpf_tcp_gen_syncookie_proto;
8320 #ifdef CONFIG_SYN_COOKIES
8321 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8322 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8323 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8324 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8325 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8326 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8327 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8328 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8329 #endif
8330 #endif
8331 default:
8332 return bpf_sk_base_func_proto(func_id, prog);
8333 }
8334
8335 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8336 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8337 * kfuncs are defined in two different modules, and we want to be able
8338 * to use them interchangeably with the same BTF type ID. Because modules
8339 * can't de-duplicate BTF IDs between each other, we need the type to be
8340 * referenced in the vmlinux BTF or the verifier will get confused about
8341 * the different types. So we add this dummy type reference which will
8342 * be included in vmlinux BTF, allowing both modules to refer to the
8343 * same type ID.
8344 */
8345 BTF_TYPE_EMIT(struct nf_conn___init);
8346 #endif
8347 }
8348
8349 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8350 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8351
8352 static const struct bpf_func_proto *
sock_ops_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8353 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8354 {
8355 const struct bpf_func_proto *func_proto;
8356
8357 func_proto = cgroup_common_func_proto(func_id, prog);
8358 if (func_proto)
8359 return func_proto;
8360
8361 switch (func_id) {
8362 case BPF_FUNC_setsockopt:
8363 return &bpf_sock_ops_setsockopt_proto;
8364 case BPF_FUNC_getsockopt:
8365 return &bpf_sock_ops_getsockopt_proto;
8366 case BPF_FUNC_sock_ops_cb_flags_set:
8367 return &bpf_sock_ops_cb_flags_set_proto;
8368 case BPF_FUNC_sock_map_update:
8369 return &bpf_sock_map_update_proto;
8370 case BPF_FUNC_sock_hash_update:
8371 return &bpf_sock_hash_update_proto;
8372 case BPF_FUNC_get_socket_cookie:
8373 return &bpf_get_socket_cookie_sock_ops_proto;
8374 case BPF_FUNC_perf_event_output:
8375 return &bpf_event_output_data_proto;
8376 case BPF_FUNC_sk_storage_get:
8377 return &bpf_sk_storage_get_proto;
8378 case BPF_FUNC_sk_storage_delete:
8379 return &bpf_sk_storage_delete_proto;
8380 case BPF_FUNC_get_netns_cookie:
8381 return &bpf_get_netns_cookie_sock_ops_proto;
8382 #ifdef CONFIG_INET
8383 case BPF_FUNC_load_hdr_opt:
8384 return &bpf_sock_ops_load_hdr_opt_proto;
8385 case BPF_FUNC_store_hdr_opt:
8386 return &bpf_sock_ops_store_hdr_opt_proto;
8387 case BPF_FUNC_reserve_hdr_opt:
8388 return &bpf_sock_ops_reserve_hdr_opt_proto;
8389 case BPF_FUNC_tcp_sock:
8390 return &bpf_tcp_sock_proto;
8391 #endif /* CONFIG_INET */
8392 default:
8393 return bpf_sk_base_func_proto(func_id, prog);
8394 }
8395 }
8396
8397 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8398 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8399
8400 static const struct bpf_func_proto *
sk_msg_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8401 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8402 {
8403 switch (func_id) {
8404 case BPF_FUNC_msg_redirect_map:
8405 return &bpf_msg_redirect_map_proto;
8406 case BPF_FUNC_msg_redirect_hash:
8407 return &bpf_msg_redirect_hash_proto;
8408 case BPF_FUNC_msg_apply_bytes:
8409 return &bpf_msg_apply_bytes_proto;
8410 case BPF_FUNC_msg_cork_bytes:
8411 return &bpf_msg_cork_bytes_proto;
8412 case BPF_FUNC_msg_pull_data:
8413 return &bpf_msg_pull_data_proto;
8414 case BPF_FUNC_msg_push_data:
8415 return &bpf_msg_push_data_proto;
8416 case BPF_FUNC_msg_pop_data:
8417 return &bpf_msg_pop_data_proto;
8418 case BPF_FUNC_perf_event_output:
8419 return &bpf_event_output_data_proto;
8420 case BPF_FUNC_get_current_uid_gid:
8421 return &bpf_get_current_uid_gid_proto;
8422 case BPF_FUNC_sk_storage_get:
8423 return &bpf_sk_storage_get_proto;
8424 case BPF_FUNC_sk_storage_delete:
8425 return &bpf_sk_storage_delete_proto;
8426 case BPF_FUNC_get_netns_cookie:
8427 return &bpf_get_netns_cookie_sk_msg_proto;
8428 #ifdef CONFIG_CGROUP_NET_CLASSID
8429 case BPF_FUNC_get_cgroup_classid:
8430 return &bpf_get_cgroup_classid_curr_proto;
8431 #endif
8432 default:
8433 return bpf_sk_base_func_proto(func_id, prog);
8434 }
8435 }
8436
8437 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8438 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8439
8440 static const struct bpf_func_proto *
sk_skb_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8441 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8442 {
8443 switch (func_id) {
8444 case BPF_FUNC_skb_store_bytes:
8445 return &bpf_skb_store_bytes_proto;
8446 case BPF_FUNC_skb_load_bytes:
8447 return &bpf_skb_load_bytes_proto;
8448 case BPF_FUNC_skb_pull_data:
8449 return &sk_skb_pull_data_proto;
8450 case BPF_FUNC_skb_change_tail:
8451 return &sk_skb_change_tail_proto;
8452 case BPF_FUNC_skb_change_head:
8453 return &sk_skb_change_head_proto;
8454 case BPF_FUNC_skb_adjust_room:
8455 return &sk_skb_adjust_room_proto;
8456 case BPF_FUNC_get_socket_cookie:
8457 return &bpf_get_socket_cookie_proto;
8458 case BPF_FUNC_get_socket_uid:
8459 return &bpf_get_socket_uid_proto;
8460 case BPF_FUNC_sk_redirect_map:
8461 return &bpf_sk_redirect_map_proto;
8462 case BPF_FUNC_sk_redirect_hash:
8463 return &bpf_sk_redirect_hash_proto;
8464 case BPF_FUNC_perf_event_output:
8465 return &bpf_skb_event_output_proto;
8466 #ifdef CONFIG_INET
8467 case BPF_FUNC_sk_lookup_tcp:
8468 return &bpf_sk_lookup_tcp_proto;
8469 case BPF_FUNC_sk_lookup_udp:
8470 return &bpf_sk_lookup_udp_proto;
8471 case BPF_FUNC_sk_release:
8472 return &bpf_sk_release_proto;
8473 case BPF_FUNC_skc_lookup_tcp:
8474 return &bpf_skc_lookup_tcp_proto;
8475 #endif
8476 default:
8477 return bpf_sk_base_func_proto(func_id, prog);
8478 }
8479 }
8480
8481 static const struct bpf_func_proto *
flow_dissector_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8482 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8483 {
8484 switch (func_id) {
8485 case BPF_FUNC_skb_load_bytes:
8486 return &bpf_flow_dissector_load_bytes_proto;
8487 default:
8488 return bpf_sk_base_func_proto(func_id, prog);
8489 }
8490 }
8491
8492 static const struct bpf_func_proto *
lwt_out_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8493 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8494 {
8495 switch (func_id) {
8496 case BPF_FUNC_skb_load_bytes:
8497 return &bpf_skb_load_bytes_proto;
8498 case BPF_FUNC_skb_pull_data:
8499 return &bpf_skb_pull_data_proto;
8500 case BPF_FUNC_csum_diff:
8501 return &bpf_csum_diff_proto;
8502 case BPF_FUNC_get_cgroup_classid:
8503 return &bpf_get_cgroup_classid_proto;
8504 case BPF_FUNC_get_route_realm:
8505 return &bpf_get_route_realm_proto;
8506 case BPF_FUNC_get_hash_recalc:
8507 return &bpf_get_hash_recalc_proto;
8508 case BPF_FUNC_perf_event_output:
8509 return &bpf_skb_event_output_proto;
8510 case BPF_FUNC_get_smp_processor_id:
8511 return &bpf_get_smp_processor_id_proto;
8512 case BPF_FUNC_skb_under_cgroup:
8513 return &bpf_skb_under_cgroup_proto;
8514 default:
8515 return bpf_sk_base_func_proto(func_id, prog);
8516 }
8517 }
8518
8519 static const struct bpf_func_proto *
lwt_in_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8520 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8521 {
8522 switch (func_id) {
8523 case BPF_FUNC_lwt_push_encap:
8524 return &bpf_lwt_in_push_encap_proto;
8525 default:
8526 return lwt_out_func_proto(func_id, prog);
8527 }
8528 }
8529
8530 static const struct bpf_func_proto *
lwt_xmit_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8531 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8532 {
8533 switch (func_id) {
8534 case BPF_FUNC_skb_get_tunnel_key:
8535 return &bpf_skb_get_tunnel_key_proto;
8536 case BPF_FUNC_skb_set_tunnel_key:
8537 return bpf_get_skb_set_tunnel_proto(func_id);
8538 case BPF_FUNC_skb_get_tunnel_opt:
8539 return &bpf_skb_get_tunnel_opt_proto;
8540 case BPF_FUNC_skb_set_tunnel_opt:
8541 return bpf_get_skb_set_tunnel_proto(func_id);
8542 case BPF_FUNC_redirect:
8543 return &bpf_redirect_proto;
8544 case BPF_FUNC_clone_redirect:
8545 return &bpf_clone_redirect_proto;
8546 case BPF_FUNC_skb_change_tail:
8547 return &bpf_skb_change_tail_proto;
8548 case BPF_FUNC_skb_change_head:
8549 return &bpf_skb_change_head_proto;
8550 case BPF_FUNC_skb_store_bytes:
8551 return &bpf_skb_store_bytes_proto;
8552 case BPF_FUNC_csum_update:
8553 return &bpf_csum_update_proto;
8554 case BPF_FUNC_csum_level:
8555 return &bpf_csum_level_proto;
8556 case BPF_FUNC_l3_csum_replace:
8557 return &bpf_l3_csum_replace_proto;
8558 case BPF_FUNC_l4_csum_replace:
8559 return &bpf_l4_csum_replace_proto;
8560 case BPF_FUNC_set_hash_invalid:
8561 return &bpf_set_hash_invalid_proto;
8562 case BPF_FUNC_lwt_push_encap:
8563 return &bpf_lwt_xmit_push_encap_proto;
8564 default:
8565 return lwt_out_func_proto(func_id, prog);
8566 }
8567 }
8568
8569 static const struct bpf_func_proto *
lwt_seg6local_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8570 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8571 {
8572 switch (func_id) {
8573 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8574 case BPF_FUNC_lwt_seg6_store_bytes:
8575 return &bpf_lwt_seg6_store_bytes_proto;
8576 case BPF_FUNC_lwt_seg6_action:
8577 return &bpf_lwt_seg6_action_proto;
8578 case BPF_FUNC_lwt_seg6_adjust_srh:
8579 return &bpf_lwt_seg6_adjust_srh_proto;
8580 #endif
8581 default:
8582 return lwt_out_func_proto(func_id, prog);
8583 }
8584 }
8585
bpf_skb_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8586 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8587 const struct bpf_prog *prog,
8588 struct bpf_insn_access_aux *info)
8589 {
8590 const int size_default = sizeof(__u32);
8591
8592 if (off < 0 || off >= sizeof(struct __sk_buff))
8593 return false;
8594
8595 /* The verifier guarantees that size > 0. */
8596 if (off % size != 0)
8597 return false;
8598
8599 switch (off) {
8600 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8601 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8602 return false;
8603 break;
8604 case bpf_ctx_range(struct __sk_buff, data):
8605 case bpf_ctx_range(struct __sk_buff, data_meta):
8606 case bpf_ctx_range(struct __sk_buff, data_end):
8607 if (info->is_ldsx || size != size_default)
8608 return false;
8609 break;
8610 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8611 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8612 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8613 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8614 if (size != size_default)
8615 return false;
8616 break;
8617 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8618 return false;
8619 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8620 if (type == BPF_WRITE || size != sizeof(__u64))
8621 return false;
8622 break;
8623 case bpf_ctx_range(struct __sk_buff, tstamp):
8624 if (size != sizeof(__u64))
8625 return false;
8626 break;
8627 case offsetof(struct __sk_buff, sk):
8628 if (type == BPF_WRITE || size != sizeof(__u64))
8629 return false;
8630 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8631 break;
8632 case offsetof(struct __sk_buff, tstamp_type):
8633 return false;
8634 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8635 /* Explicitly prohibit access to padding in __sk_buff. */
8636 return false;
8637 default:
8638 /* Only narrow read access allowed for now. */
8639 if (type == BPF_WRITE) {
8640 if (size != size_default)
8641 return false;
8642 } else {
8643 bpf_ctx_record_field_size(info, size_default);
8644 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8645 return false;
8646 }
8647 }
8648
8649 return true;
8650 }
8651
sk_filter_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8652 static bool sk_filter_is_valid_access(int off, int size,
8653 enum bpf_access_type type,
8654 const struct bpf_prog *prog,
8655 struct bpf_insn_access_aux *info)
8656 {
8657 switch (off) {
8658 case bpf_ctx_range(struct __sk_buff, tc_classid):
8659 case bpf_ctx_range(struct __sk_buff, data):
8660 case bpf_ctx_range(struct __sk_buff, data_meta):
8661 case bpf_ctx_range(struct __sk_buff, data_end):
8662 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8663 case bpf_ctx_range(struct __sk_buff, tstamp):
8664 case bpf_ctx_range(struct __sk_buff, wire_len):
8665 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8666 return false;
8667 }
8668
8669 if (type == BPF_WRITE) {
8670 switch (off) {
8671 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8672 break;
8673 default:
8674 return false;
8675 }
8676 }
8677
8678 return bpf_skb_is_valid_access(off, size, type, prog, info);
8679 }
8680
cg_skb_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8681 static bool cg_skb_is_valid_access(int off, int size,
8682 enum bpf_access_type type,
8683 const struct bpf_prog *prog,
8684 struct bpf_insn_access_aux *info)
8685 {
8686 switch (off) {
8687 case bpf_ctx_range(struct __sk_buff, tc_classid):
8688 case bpf_ctx_range(struct __sk_buff, data_meta):
8689 case bpf_ctx_range(struct __sk_buff, wire_len):
8690 return false;
8691 case bpf_ctx_range(struct __sk_buff, data):
8692 case bpf_ctx_range(struct __sk_buff, data_end):
8693 if (!bpf_token_capable(prog->aux->token, CAP_BPF))
8694 return false;
8695 break;
8696 }
8697
8698 if (type == BPF_WRITE) {
8699 switch (off) {
8700 case bpf_ctx_range(struct __sk_buff, mark):
8701 case bpf_ctx_range(struct __sk_buff, priority):
8702 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8703 break;
8704 case bpf_ctx_range(struct __sk_buff, tstamp):
8705 if (!bpf_token_capable(prog->aux->token, CAP_BPF))
8706 return false;
8707 break;
8708 default:
8709 return false;
8710 }
8711 }
8712
8713 switch (off) {
8714 case bpf_ctx_range(struct __sk_buff, data):
8715 info->reg_type = PTR_TO_PACKET;
8716 break;
8717 case bpf_ctx_range(struct __sk_buff, data_end):
8718 info->reg_type = PTR_TO_PACKET_END;
8719 break;
8720 }
8721
8722 return bpf_skb_is_valid_access(off, size, type, prog, info);
8723 }
8724
lwt_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8725 static bool lwt_is_valid_access(int off, int size,
8726 enum bpf_access_type type,
8727 const struct bpf_prog *prog,
8728 struct bpf_insn_access_aux *info)
8729 {
8730 switch (off) {
8731 case bpf_ctx_range(struct __sk_buff, tc_classid):
8732 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8733 case bpf_ctx_range(struct __sk_buff, data_meta):
8734 case bpf_ctx_range(struct __sk_buff, tstamp):
8735 case bpf_ctx_range(struct __sk_buff, wire_len):
8736 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8737 return false;
8738 }
8739
8740 if (type == BPF_WRITE) {
8741 switch (off) {
8742 case bpf_ctx_range(struct __sk_buff, mark):
8743 case bpf_ctx_range(struct __sk_buff, priority):
8744 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8745 break;
8746 default:
8747 return false;
8748 }
8749 }
8750
8751 switch (off) {
8752 case bpf_ctx_range(struct __sk_buff, data):
8753 info->reg_type = PTR_TO_PACKET;
8754 break;
8755 case bpf_ctx_range(struct __sk_buff, data_end):
8756 info->reg_type = PTR_TO_PACKET_END;
8757 break;
8758 }
8759
8760 return bpf_skb_is_valid_access(off, size, type, prog, info);
8761 }
8762
8763 /* Attach type specific accesses */
__sock_filter_check_attach_type(int off,enum bpf_access_type access_type,enum bpf_attach_type attach_type)8764 static bool __sock_filter_check_attach_type(int off,
8765 enum bpf_access_type access_type,
8766 enum bpf_attach_type attach_type)
8767 {
8768 switch (off) {
8769 case offsetof(struct bpf_sock, bound_dev_if):
8770 case offsetof(struct bpf_sock, mark):
8771 case offsetof(struct bpf_sock, priority):
8772 switch (attach_type) {
8773 case BPF_CGROUP_INET_SOCK_CREATE:
8774 case BPF_CGROUP_INET_SOCK_RELEASE:
8775 goto full_access;
8776 default:
8777 return false;
8778 }
8779 case bpf_ctx_range(struct bpf_sock, src_ip4):
8780 switch (attach_type) {
8781 case BPF_CGROUP_INET4_POST_BIND:
8782 goto read_only;
8783 default:
8784 return false;
8785 }
8786 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8787 switch (attach_type) {
8788 case BPF_CGROUP_INET6_POST_BIND:
8789 goto read_only;
8790 default:
8791 return false;
8792 }
8793 case bpf_ctx_range(struct bpf_sock, src_port):
8794 switch (attach_type) {
8795 case BPF_CGROUP_INET4_POST_BIND:
8796 case BPF_CGROUP_INET6_POST_BIND:
8797 goto read_only;
8798 default:
8799 return false;
8800 }
8801 }
8802 read_only:
8803 return access_type == BPF_READ;
8804 full_access:
8805 return true;
8806 }
8807
bpf_sock_common_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)8808 bool bpf_sock_common_is_valid_access(int off, int size,
8809 enum bpf_access_type type,
8810 struct bpf_insn_access_aux *info)
8811 {
8812 switch (off) {
8813 case bpf_ctx_range_till(struct bpf_sock, type, priority):
8814 return false;
8815 default:
8816 return bpf_sock_is_valid_access(off, size, type, info);
8817 }
8818 }
8819
bpf_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)8820 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8821 struct bpf_insn_access_aux *info)
8822 {
8823 const int size_default = sizeof(__u32);
8824 int field_size;
8825
8826 if (off < 0 || off >= sizeof(struct bpf_sock))
8827 return false;
8828 if (off % size != 0)
8829 return false;
8830
8831 switch (off) {
8832 case offsetof(struct bpf_sock, state):
8833 case offsetof(struct bpf_sock, family):
8834 case offsetof(struct bpf_sock, type):
8835 case offsetof(struct bpf_sock, protocol):
8836 case offsetof(struct bpf_sock, src_port):
8837 case offsetof(struct bpf_sock, rx_queue_mapping):
8838 case bpf_ctx_range(struct bpf_sock, src_ip4):
8839 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8840 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8841 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8842 bpf_ctx_record_field_size(info, size_default);
8843 return bpf_ctx_narrow_access_ok(off, size, size_default);
8844 case bpf_ctx_range(struct bpf_sock, dst_port):
8845 field_size = size == size_default ?
8846 size_default : sizeof_field(struct bpf_sock, dst_port);
8847 bpf_ctx_record_field_size(info, field_size);
8848 return bpf_ctx_narrow_access_ok(off, size, field_size);
8849 case offsetofend(struct bpf_sock, dst_port) ...
8850 offsetof(struct bpf_sock, dst_ip4) - 1:
8851 return false;
8852 }
8853
8854 return size == size_default;
8855 }
8856
sock_filter_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8857 static bool sock_filter_is_valid_access(int off, int size,
8858 enum bpf_access_type type,
8859 const struct bpf_prog *prog,
8860 struct bpf_insn_access_aux *info)
8861 {
8862 if (!bpf_sock_is_valid_access(off, size, type, info))
8863 return false;
8864 return __sock_filter_check_attach_type(off, type,
8865 prog->expected_attach_type);
8866 }
8867
bpf_noop_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)8868 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8869 const struct bpf_prog *prog)
8870 {
8871 /* Neither direct read nor direct write requires any preliminary
8872 * action.
8873 */
8874 return 0;
8875 }
8876
bpf_unclone_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog,int drop_verdict)8877 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8878 const struct bpf_prog *prog, int drop_verdict)
8879 {
8880 struct bpf_insn *insn = insn_buf;
8881
8882 if (!direct_write)
8883 return 0;
8884
8885 /* if (!skb->cloned)
8886 * goto start;
8887 *
8888 * (Fast-path, otherwise approximation that we might be
8889 * a clone, do the rest in helper.)
8890 */
8891 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8892 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8893 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8894
8895 /* ret = bpf_skb_pull_data(skb, 0); */
8896 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8897 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8898 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8899 BPF_FUNC_skb_pull_data);
8900 /* if (!ret)
8901 * goto restore;
8902 * return TC_ACT_SHOT;
8903 */
8904 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8905 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8906 *insn++ = BPF_EXIT_INSN();
8907
8908 /* restore: */
8909 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8910 /* start: */
8911 *insn++ = prog->insnsi[0];
8912
8913 return insn - insn_buf;
8914 }
8915
bpf_gen_ld_abs(const struct bpf_insn * orig,struct bpf_insn * insn_buf)8916 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8917 struct bpf_insn *insn_buf)
8918 {
8919 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8920 struct bpf_insn *insn = insn_buf;
8921
8922 if (!indirect) {
8923 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8924 } else {
8925 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8926 if (orig->imm)
8927 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8928 }
8929 /* We're guaranteed here that CTX is in R6. */
8930 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8931
8932 switch (BPF_SIZE(orig->code)) {
8933 case BPF_B:
8934 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8935 break;
8936 case BPF_H:
8937 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8938 break;
8939 case BPF_W:
8940 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8941 break;
8942 }
8943
8944 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8945 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8946 *insn++ = BPF_EXIT_INSN();
8947
8948 return insn - insn_buf;
8949 }
8950
tc_cls_act_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)8951 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8952 const struct bpf_prog *prog)
8953 {
8954 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8955 }
8956
tc_cls_act_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8957 static bool tc_cls_act_is_valid_access(int off, int size,
8958 enum bpf_access_type type,
8959 const struct bpf_prog *prog,
8960 struct bpf_insn_access_aux *info)
8961 {
8962 if (type == BPF_WRITE) {
8963 switch (off) {
8964 case bpf_ctx_range(struct __sk_buff, mark):
8965 case bpf_ctx_range(struct __sk_buff, tc_index):
8966 case bpf_ctx_range(struct __sk_buff, priority):
8967 case bpf_ctx_range(struct __sk_buff, tc_classid):
8968 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8969 case bpf_ctx_range(struct __sk_buff, tstamp):
8970 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8971 break;
8972 default:
8973 return false;
8974 }
8975 }
8976
8977 switch (off) {
8978 case bpf_ctx_range(struct __sk_buff, data):
8979 info->reg_type = PTR_TO_PACKET;
8980 break;
8981 case bpf_ctx_range(struct __sk_buff, data_meta):
8982 info->reg_type = PTR_TO_PACKET_META;
8983 break;
8984 case bpf_ctx_range(struct __sk_buff, data_end):
8985 info->reg_type = PTR_TO_PACKET_END;
8986 break;
8987 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8988 return false;
8989 case offsetof(struct __sk_buff, tstamp_type):
8990 /* The convert_ctx_access() on reading and writing
8991 * __sk_buff->tstamp depends on whether the bpf prog
8992 * has used __sk_buff->tstamp_type or not.
8993 * Thus, we need to set prog->tstamp_type_access
8994 * earlier during is_valid_access() here.
8995 */
8996 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8997 return size == sizeof(__u8);
8998 }
8999
9000 return bpf_skb_is_valid_access(off, size, type, prog, info);
9001 }
9002
9003 DEFINE_MUTEX(nf_conn_btf_access_lock);
9004 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
9005
9006 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
9007 const struct bpf_reg_state *reg,
9008 int off, int size);
9009 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
9010
tc_cls_act_btf_struct_access(struct bpf_verifier_log * log,const struct bpf_reg_state * reg,int off,int size)9011 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
9012 const struct bpf_reg_state *reg,
9013 int off, int size)
9014 {
9015 int ret = -EACCES;
9016
9017 mutex_lock(&nf_conn_btf_access_lock);
9018 if (nfct_btf_struct_access)
9019 ret = nfct_btf_struct_access(log, reg, off, size);
9020 mutex_unlock(&nf_conn_btf_access_lock);
9021
9022 return ret;
9023 }
9024
__is_valid_xdp_access(int off,int size)9025 static bool __is_valid_xdp_access(int off, int size)
9026 {
9027 if (off < 0 || off >= sizeof(struct xdp_md))
9028 return false;
9029 if (off % size != 0)
9030 return false;
9031 if (size != sizeof(__u32))
9032 return false;
9033
9034 return true;
9035 }
9036
xdp_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9037 static bool xdp_is_valid_access(int off, int size,
9038 enum bpf_access_type type,
9039 const struct bpf_prog *prog,
9040 struct bpf_insn_access_aux *info)
9041 {
9042 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
9043 switch (off) {
9044 case offsetof(struct xdp_md, egress_ifindex):
9045 return false;
9046 }
9047 }
9048
9049 if (type == BPF_WRITE) {
9050 if (bpf_prog_is_offloaded(prog->aux)) {
9051 switch (off) {
9052 case offsetof(struct xdp_md, rx_queue_index):
9053 return __is_valid_xdp_access(off, size);
9054 }
9055 }
9056 return false;
9057 } else {
9058 switch (off) {
9059 case offsetof(struct xdp_md, data_meta):
9060 case offsetof(struct xdp_md, data):
9061 case offsetof(struct xdp_md, data_end):
9062 if (info->is_ldsx)
9063 return false;
9064 }
9065 }
9066
9067 switch (off) {
9068 case offsetof(struct xdp_md, data):
9069 info->reg_type = PTR_TO_PACKET;
9070 break;
9071 case offsetof(struct xdp_md, data_meta):
9072 info->reg_type = PTR_TO_PACKET_META;
9073 break;
9074 case offsetof(struct xdp_md, data_end):
9075 info->reg_type = PTR_TO_PACKET_END;
9076 break;
9077 }
9078
9079 return __is_valid_xdp_access(off, size);
9080 }
9081
bpf_warn_invalid_xdp_action(struct net_device * dev,struct bpf_prog * prog,u32 act)9082 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
9083 {
9084 const u32 act_max = XDP_REDIRECT;
9085
9086 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
9087 act > act_max ? "Illegal" : "Driver unsupported",
9088 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
9089 }
9090 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
9091
xdp_btf_struct_access(struct bpf_verifier_log * log,const struct bpf_reg_state * reg,int off,int size)9092 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
9093 const struct bpf_reg_state *reg,
9094 int off, int size)
9095 {
9096 int ret = -EACCES;
9097
9098 mutex_lock(&nf_conn_btf_access_lock);
9099 if (nfct_btf_struct_access)
9100 ret = nfct_btf_struct_access(log, reg, off, size);
9101 mutex_unlock(&nf_conn_btf_access_lock);
9102
9103 return ret;
9104 }
9105
sock_addr_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9106 static bool sock_addr_is_valid_access(int off, int size,
9107 enum bpf_access_type type,
9108 const struct bpf_prog *prog,
9109 struct bpf_insn_access_aux *info)
9110 {
9111 const int size_default = sizeof(__u32);
9112
9113 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
9114 return false;
9115 if (off % size != 0)
9116 return false;
9117
9118 /* Disallow access to fields not belonging to the attach type's address
9119 * family.
9120 */
9121 switch (off) {
9122 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
9123 switch (prog->expected_attach_type) {
9124 case BPF_CGROUP_INET4_BIND:
9125 case BPF_CGROUP_INET4_CONNECT:
9126 case BPF_CGROUP_INET4_GETPEERNAME:
9127 case BPF_CGROUP_INET4_GETSOCKNAME:
9128 case BPF_CGROUP_UDP4_SENDMSG:
9129 case BPF_CGROUP_UDP4_RECVMSG:
9130 break;
9131 default:
9132 return false;
9133 }
9134 break;
9135 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9136 switch (prog->expected_attach_type) {
9137 case BPF_CGROUP_INET6_BIND:
9138 case BPF_CGROUP_INET6_CONNECT:
9139 case BPF_CGROUP_INET6_GETPEERNAME:
9140 case BPF_CGROUP_INET6_GETSOCKNAME:
9141 case BPF_CGROUP_UDP6_SENDMSG:
9142 case BPF_CGROUP_UDP6_RECVMSG:
9143 break;
9144 default:
9145 return false;
9146 }
9147 break;
9148 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
9149 switch (prog->expected_attach_type) {
9150 case BPF_CGROUP_UDP4_SENDMSG:
9151 break;
9152 default:
9153 return false;
9154 }
9155 break;
9156 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9157 msg_src_ip6[3]):
9158 switch (prog->expected_attach_type) {
9159 case BPF_CGROUP_UDP6_SENDMSG:
9160 break;
9161 default:
9162 return false;
9163 }
9164 break;
9165 }
9166
9167 switch (off) {
9168 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
9169 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9170 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
9171 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9172 msg_src_ip6[3]):
9173 case bpf_ctx_range(struct bpf_sock_addr, user_port):
9174 if (type == BPF_READ) {
9175 bpf_ctx_record_field_size(info, size_default);
9176
9177 if (bpf_ctx_wide_access_ok(off, size,
9178 struct bpf_sock_addr,
9179 user_ip6))
9180 return true;
9181
9182 if (bpf_ctx_wide_access_ok(off, size,
9183 struct bpf_sock_addr,
9184 msg_src_ip6))
9185 return true;
9186
9187 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
9188 return false;
9189 } else {
9190 if (bpf_ctx_wide_access_ok(off, size,
9191 struct bpf_sock_addr,
9192 user_ip6))
9193 return true;
9194
9195 if (bpf_ctx_wide_access_ok(off, size,
9196 struct bpf_sock_addr,
9197 msg_src_ip6))
9198 return true;
9199
9200 if (size != size_default)
9201 return false;
9202 }
9203 break;
9204 case offsetof(struct bpf_sock_addr, sk):
9205 if (type != BPF_READ)
9206 return false;
9207 if (size != sizeof(__u64))
9208 return false;
9209 info->reg_type = PTR_TO_SOCKET;
9210 break;
9211 default:
9212 if (type == BPF_READ) {
9213 if (size != size_default)
9214 return false;
9215 } else {
9216 return false;
9217 }
9218 }
9219
9220 return true;
9221 }
9222
sock_ops_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9223 static bool sock_ops_is_valid_access(int off, int size,
9224 enum bpf_access_type type,
9225 const struct bpf_prog *prog,
9226 struct bpf_insn_access_aux *info)
9227 {
9228 const int size_default = sizeof(__u32);
9229
9230 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
9231 return false;
9232
9233 /* The verifier guarantees that size > 0. */
9234 if (off % size != 0)
9235 return false;
9236
9237 if (type == BPF_WRITE) {
9238 switch (off) {
9239 case offsetof(struct bpf_sock_ops, reply):
9240 case offsetof(struct bpf_sock_ops, sk_txhash):
9241 if (size != size_default)
9242 return false;
9243 break;
9244 default:
9245 return false;
9246 }
9247 } else {
9248 switch (off) {
9249 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
9250 bytes_acked):
9251 if (size != sizeof(__u64))
9252 return false;
9253 break;
9254 case offsetof(struct bpf_sock_ops, sk):
9255 if (size != sizeof(__u64))
9256 return false;
9257 info->reg_type = PTR_TO_SOCKET_OR_NULL;
9258 break;
9259 case offsetof(struct bpf_sock_ops, skb_data):
9260 if (size != sizeof(__u64))
9261 return false;
9262 info->reg_type = PTR_TO_PACKET;
9263 break;
9264 case offsetof(struct bpf_sock_ops, skb_data_end):
9265 if (size != sizeof(__u64))
9266 return false;
9267 info->reg_type = PTR_TO_PACKET_END;
9268 break;
9269 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9270 bpf_ctx_record_field_size(info, size_default);
9271 return bpf_ctx_narrow_access_ok(off, size,
9272 size_default);
9273 case offsetof(struct bpf_sock_ops, skb_hwtstamp):
9274 if (size != sizeof(__u64))
9275 return false;
9276 break;
9277 default:
9278 if (size != size_default)
9279 return false;
9280 break;
9281 }
9282 }
9283
9284 return true;
9285 }
9286
sk_skb_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)9287 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
9288 const struct bpf_prog *prog)
9289 {
9290 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9291 }
9292
sk_skb_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9293 static bool sk_skb_is_valid_access(int off, int size,
9294 enum bpf_access_type type,
9295 const struct bpf_prog *prog,
9296 struct bpf_insn_access_aux *info)
9297 {
9298 switch (off) {
9299 case bpf_ctx_range(struct __sk_buff, tc_classid):
9300 case bpf_ctx_range(struct __sk_buff, data_meta):
9301 case bpf_ctx_range(struct __sk_buff, tstamp):
9302 case bpf_ctx_range(struct __sk_buff, wire_len):
9303 case bpf_ctx_range(struct __sk_buff, hwtstamp):
9304 return false;
9305 }
9306
9307 if (type == BPF_WRITE) {
9308 switch (off) {
9309 case bpf_ctx_range(struct __sk_buff, tc_index):
9310 case bpf_ctx_range(struct __sk_buff, priority):
9311 break;
9312 default:
9313 return false;
9314 }
9315 }
9316
9317 switch (off) {
9318 case bpf_ctx_range(struct __sk_buff, mark):
9319 return false;
9320 case bpf_ctx_range(struct __sk_buff, data):
9321 info->reg_type = PTR_TO_PACKET;
9322 break;
9323 case bpf_ctx_range(struct __sk_buff, data_end):
9324 info->reg_type = PTR_TO_PACKET_END;
9325 break;
9326 }
9327
9328 return bpf_skb_is_valid_access(off, size, type, prog, info);
9329 }
9330
sk_msg_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9331 static bool sk_msg_is_valid_access(int off, int size,
9332 enum bpf_access_type type,
9333 const struct bpf_prog *prog,
9334 struct bpf_insn_access_aux *info)
9335 {
9336 if (type == BPF_WRITE)
9337 return false;
9338
9339 if (off % size != 0)
9340 return false;
9341
9342 switch (off) {
9343 case offsetof(struct sk_msg_md, data):
9344 info->reg_type = PTR_TO_PACKET;
9345 if (size != sizeof(__u64))
9346 return false;
9347 break;
9348 case offsetof(struct sk_msg_md, data_end):
9349 info->reg_type = PTR_TO_PACKET_END;
9350 if (size != sizeof(__u64))
9351 return false;
9352 break;
9353 case offsetof(struct sk_msg_md, sk):
9354 if (size != sizeof(__u64))
9355 return false;
9356 info->reg_type = PTR_TO_SOCKET;
9357 break;
9358 case bpf_ctx_range(struct sk_msg_md, family):
9359 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9360 case bpf_ctx_range(struct sk_msg_md, local_ip4):
9361 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9362 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9363 case bpf_ctx_range(struct sk_msg_md, remote_port):
9364 case bpf_ctx_range(struct sk_msg_md, local_port):
9365 case bpf_ctx_range(struct sk_msg_md, size):
9366 if (size != sizeof(__u32))
9367 return false;
9368 break;
9369 default:
9370 return false;
9371 }
9372 return true;
9373 }
9374
flow_dissector_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9375 static bool flow_dissector_is_valid_access(int off, int size,
9376 enum bpf_access_type type,
9377 const struct bpf_prog *prog,
9378 struct bpf_insn_access_aux *info)
9379 {
9380 const int size_default = sizeof(__u32);
9381
9382 if (off < 0 || off >= sizeof(struct __sk_buff))
9383 return false;
9384
9385 if (type == BPF_WRITE)
9386 return false;
9387
9388 switch (off) {
9389 case bpf_ctx_range(struct __sk_buff, data):
9390 if (info->is_ldsx || size != size_default)
9391 return false;
9392 info->reg_type = PTR_TO_PACKET;
9393 return true;
9394 case bpf_ctx_range(struct __sk_buff, data_end):
9395 if (info->is_ldsx || size != size_default)
9396 return false;
9397 info->reg_type = PTR_TO_PACKET_END;
9398 return true;
9399 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9400 if (size != sizeof(__u64))
9401 return false;
9402 info->reg_type = PTR_TO_FLOW_KEYS;
9403 return true;
9404 default:
9405 return false;
9406 }
9407 }
9408
flow_dissector_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)9409 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9410 const struct bpf_insn *si,
9411 struct bpf_insn *insn_buf,
9412 struct bpf_prog *prog,
9413 u32 *target_size)
9414
9415 {
9416 struct bpf_insn *insn = insn_buf;
9417
9418 switch (si->off) {
9419 case offsetof(struct __sk_buff, data):
9420 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9421 si->dst_reg, si->src_reg,
9422 offsetof(struct bpf_flow_dissector, data));
9423 break;
9424
9425 case offsetof(struct __sk_buff, data_end):
9426 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9427 si->dst_reg, si->src_reg,
9428 offsetof(struct bpf_flow_dissector, data_end));
9429 break;
9430
9431 case offsetof(struct __sk_buff, flow_keys):
9432 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9433 si->dst_reg, si->src_reg,
9434 offsetof(struct bpf_flow_dissector, flow_keys));
9435 break;
9436 }
9437
9438 return insn - insn_buf;
9439 }
9440
bpf_convert_tstamp_type_read(const struct bpf_insn * si,struct bpf_insn * insn)9441 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9442 struct bpf_insn *insn)
9443 {
9444 __u8 value_reg = si->dst_reg;
9445 __u8 skb_reg = si->src_reg;
9446 BUILD_BUG_ON(__SKB_CLOCK_MAX != (int)BPF_SKB_CLOCK_TAI);
9447 BUILD_BUG_ON(SKB_CLOCK_REALTIME != (int)BPF_SKB_CLOCK_REALTIME);
9448 BUILD_BUG_ON(SKB_CLOCK_MONOTONIC != (int)BPF_SKB_CLOCK_MONOTONIC);
9449 BUILD_BUG_ON(SKB_CLOCK_TAI != (int)BPF_SKB_CLOCK_TAI);
9450 *insn++ = BPF_LDX_MEM(BPF_B, value_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9451 *insn++ = BPF_ALU32_IMM(BPF_AND, value_reg, SKB_TSTAMP_TYPE_MASK);
9452 #ifdef __BIG_ENDIAN_BITFIELD
9453 *insn++ = BPF_ALU32_IMM(BPF_RSH, value_reg, SKB_TSTAMP_TYPE_RSHIFT);
9454 #else
9455 BUILD_BUG_ON(!(SKB_TSTAMP_TYPE_MASK & 0x1));
9456 #endif
9457
9458 return insn;
9459 }
9460
bpf_convert_shinfo_access(__u8 dst_reg,__u8 skb_reg,struct bpf_insn * insn)9461 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg,
9462 struct bpf_insn *insn)
9463 {
9464 /* si->dst_reg = skb_shinfo(SKB); */
9465 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9466 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9467 BPF_REG_AX, skb_reg,
9468 offsetof(struct sk_buff, end));
9469 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9470 dst_reg, skb_reg,
9471 offsetof(struct sk_buff, head));
9472 *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX);
9473 #else
9474 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9475 dst_reg, skb_reg,
9476 offsetof(struct sk_buff, end));
9477 #endif
9478
9479 return insn;
9480 }
9481
bpf_convert_tstamp_read(const struct bpf_prog * prog,const struct bpf_insn * si,struct bpf_insn * insn)9482 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9483 const struct bpf_insn *si,
9484 struct bpf_insn *insn)
9485 {
9486 __u8 value_reg = si->dst_reg;
9487 __u8 skb_reg = si->src_reg;
9488
9489 #ifdef CONFIG_NET_XGRESS
9490 /* If the tstamp_type is read,
9491 * the bpf prog is aware the tstamp could have delivery time.
9492 * Thus, read skb->tstamp as is if tstamp_type_access is true.
9493 */
9494 if (!prog->tstamp_type_access) {
9495 /* AX is needed because src_reg and dst_reg could be the same */
9496 __u8 tmp_reg = BPF_REG_AX;
9497
9498 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9499 /* check if ingress mask bits is set */
9500 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9501 *insn++ = BPF_JMP_A(4);
9502 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, SKB_TSTAMP_TYPE_MASK, 1);
9503 *insn++ = BPF_JMP_A(2);
9504 /* skb->tc_at_ingress && skb->tstamp_type,
9505 * read 0 as the (rcv) timestamp.
9506 */
9507 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9508 *insn++ = BPF_JMP_A(1);
9509 }
9510 #endif
9511
9512 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9513 offsetof(struct sk_buff, tstamp));
9514 return insn;
9515 }
9516
bpf_convert_tstamp_write(const struct bpf_prog * prog,const struct bpf_insn * si,struct bpf_insn * insn)9517 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9518 const struct bpf_insn *si,
9519 struct bpf_insn *insn)
9520 {
9521 __u8 value_reg = si->src_reg;
9522 __u8 skb_reg = si->dst_reg;
9523
9524 #ifdef CONFIG_NET_XGRESS
9525 /* If the tstamp_type is read,
9526 * the bpf prog is aware the tstamp could have delivery time.
9527 * Thus, write skb->tstamp as is if tstamp_type_access is true.
9528 * Otherwise, writing at ingress will have to clear the
9529 * skb->tstamp_type bit also.
9530 */
9531 if (!prog->tstamp_type_access) {
9532 __u8 tmp_reg = BPF_REG_AX;
9533
9534 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9535 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9536 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9537 /* goto <store> */
9538 *insn++ = BPF_JMP_A(2);
9539 /* <clear>: skb->tstamp_type */
9540 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_TSTAMP_TYPE_MASK);
9541 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET);
9542 }
9543 #endif
9544
9545 /* <store>: skb->tstamp = tstamp */
9546 *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_DW | BPF_MEM,
9547 skb_reg, value_reg, offsetof(struct sk_buff, tstamp), si->imm);
9548 return insn;
9549 }
9550
9551 #define BPF_EMIT_STORE(size, si, off) \
9552 BPF_RAW_INSN(BPF_CLASS((si)->code) | (size) | BPF_MEM, \
9553 (si)->dst_reg, (si)->src_reg, (off), (si)->imm)
9554
bpf_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)9555 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9556 const struct bpf_insn *si,
9557 struct bpf_insn *insn_buf,
9558 struct bpf_prog *prog, u32 *target_size)
9559 {
9560 struct bpf_insn *insn = insn_buf;
9561 int off;
9562
9563 switch (si->off) {
9564 case offsetof(struct __sk_buff, len):
9565 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9566 bpf_target_off(struct sk_buff, len, 4,
9567 target_size));
9568 break;
9569
9570 case offsetof(struct __sk_buff, protocol):
9571 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9572 bpf_target_off(struct sk_buff, protocol, 2,
9573 target_size));
9574 break;
9575
9576 case offsetof(struct __sk_buff, vlan_proto):
9577 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9578 bpf_target_off(struct sk_buff, vlan_proto, 2,
9579 target_size));
9580 break;
9581
9582 case offsetof(struct __sk_buff, priority):
9583 if (type == BPF_WRITE)
9584 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9585 bpf_target_off(struct sk_buff, priority, 4,
9586 target_size));
9587 else
9588 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9589 bpf_target_off(struct sk_buff, priority, 4,
9590 target_size));
9591 break;
9592
9593 case offsetof(struct __sk_buff, ingress_ifindex):
9594 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9595 bpf_target_off(struct sk_buff, skb_iif, 4,
9596 target_size));
9597 break;
9598
9599 case offsetof(struct __sk_buff, ifindex):
9600 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9601 si->dst_reg, si->src_reg,
9602 offsetof(struct sk_buff, dev));
9603 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9604 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9605 bpf_target_off(struct net_device, ifindex, 4,
9606 target_size));
9607 break;
9608
9609 case offsetof(struct __sk_buff, hash):
9610 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9611 bpf_target_off(struct sk_buff, hash, 4,
9612 target_size));
9613 break;
9614
9615 case offsetof(struct __sk_buff, mark):
9616 if (type == BPF_WRITE)
9617 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9618 bpf_target_off(struct sk_buff, mark, 4,
9619 target_size));
9620 else
9621 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9622 bpf_target_off(struct sk_buff, mark, 4,
9623 target_size));
9624 break;
9625
9626 case offsetof(struct __sk_buff, pkt_type):
9627 *target_size = 1;
9628 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9629 PKT_TYPE_OFFSET);
9630 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9631 #ifdef __BIG_ENDIAN_BITFIELD
9632 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9633 #endif
9634 break;
9635
9636 case offsetof(struct __sk_buff, queue_mapping):
9637 if (type == BPF_WRITE) {
9638 u32 off = bpf_target_off(struct sk_buff, queue_mapping, 2, target_size);
9639
9640 if (BPF_CLASS(si->code) == BPF_ST && si->imm >= NO_QUEUE_MAPPING) {
9641 *insn++ = BPF_JMP_A(0); /* noop */
9642 break;
9643 }
9644
9645 if (BPF_CLASS(si->code) == BPF_STX)
9646 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9647 *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9648 } else {
9649 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9650 bpf_target_off(struct sk_buff,
9651 queue_mapping,
9652 2, target_size));
9653 }
9654 break;
9655
9656 case offsetof(struct __sk_buff, vlan_present):
9657 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9658 bpf_target_off(struct sk_buff,
9659 vlan_all, 4, target_size));
9660 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9661 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1);
9662 break;
9663
9664 case offsetof(struct __sk_buff, vlan_tci):
9665 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9666 bpf_target_off(struct sk_buff, vlan_tci, 2,
9667 target_size));
9668 break;
9669
9670 case offsetof(struct __sk_buff, cb[0]) ...
9671 offsetofend(struct __sk_buff, cb[4]) - 1:
9672 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9673 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9674 offsetof(struct qdisc_skb_cb, data)) %
9675 sizeof(__u64));
9676
9677 prog->cb_access = 1;
9678 off = si->off;
9679 off -= offsetof(struct __sk_buff, cb[0]);
9680 off += offsetof(struct sk_buff, cb);
9681 off += offsetof(struct qdisc_skb_cb, data);
9682 if (type == BPF_WRITE)
9683 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
9684 else
9685 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9686 si->src_reg, off);
9687 break;
9688
9689 case offsetof(struct __sk_buff, tc_classid):
9690 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9691
9692 off = si->off;
9693 off -= offsetof(struct __sk_buff, tc_classid);
9694 off += offsetof(struct sk_buff, cb);
9695 off += offsetof(struct qdisc_skb_cb, tc_classid);
9696 *target_size = 2;
9697 if (type == BPF_WRITE)
9698 *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9699 else
9700 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9701 si->src_reg, off);
9702 break;
9703
9704 case offsetof(struct __sk_buff, data):
9705 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9706 si->dst_reg, si->src_reg,
9707 offsetof(struct sk_buff, data));
9708 break;
9709
9710 case offsetof(struct __sk_buff, data_meta):
9711 off = si->off;
9712 off -= offsetof(struct __sk_buff, data_meta);
9713 off += offsetof(struct sk_buff, cb);
9714 off += offsetof(struct bpf_skb_data_end, data_meta);
9715 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9716 si->src_reg, off);
9717 break;
9718
9719 case offsetof(struct __sk_buff, data_end):
9720 off = si->off;
9721 off -= offsetof(struct __sk_buff, data_end);
9722 off += offsetof(struct sk_buff, cb);
9723 off += offsetof(struct bpf_skb_data_end, data_end);
9724 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9725 si->src_reg, off);
9726 break;
9727
9728 case offsetof(struct __sk_buff, tc_index):
9729 #ifdef CONFIG_NET_SCHED
9730 if (type == BPF_WRITE)
9731 *insn++ = BPF_EMIT_STORE(BPF_H, si,
9732 bpf_target_off(struct sk_buff, tc_index, 2,
9733 target_size));
9734 else
9735 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9736 bpf_target_off(struct sk_buff, tc_index, 2,
9737 target_size));
9738 #else
9739 *target_size = 2;
9740 if (type == BPF_WRITE)
9741 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9742 else
9743 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9744 #endif
9745 break;
9746
9747 case offsetof(struct __sk_buff, napi_id):
9748 #if defined(CONFIG_NET_RX_BUSY_POLL)
9749 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9750 bpf_target_off(struct sk_buff, napi_id, 4,
9751 target_size));
9752 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9753 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9754 #else
9755 *target_size = 4;
9756 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9757 #endif
9758 break;
9759 case offsetof(struct __sk_buff, family):
9760 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9761
9762 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9763 si->dst_reg, si->src_reg,
9764 offsetof(struct sk_buff, sk));
9765 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9766 bpf_target_off(struct sock_common,
9767 skc_family,
9768 2, target_size));
9769 break;
9770 case offsetof(struct __sk_buff, remote_ip4):
9771 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9772
9773 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9774 si->dst_reg, si->src_reg,
9775 offsetof(struct sk_buff, sk));
9776 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9777 bpf_target_off(struct sock_common,
9778 skc_daddr,
9779 4, target_size));
9780 break;
9781 case offsetof(struct __sk_buff, local_ip4):
9782 BUILD_BUG_ON(sizeof_field(struct sock_common,
9783 skc_rcv_saddr) != 4);
9784
9785 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9786 si->dst_reg, si->src_reg,
9787 offsetof(struct sk_buff, sk));
9788 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9789 bpf_target_off(struct sock_common,
9790 skc_rcv_saddr,
9791 4, target_size));
9792 break;
9793 case offsetof(struct __sk_buff, remote_ip6[0]) ...
9794 offsetof(struct __sk_buff, remote_ip6[3]):
9795 #if IS_ENABLED(CONFIG_IPV6)
9796 BUILD_BUG_ON(sizeof_field(struct sock_common,
9797 skc_v6_daddr.s6_addr32[0]) != 4);
9798
9799 off = si->off;
9800 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9801
9802 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9803 si->dst_reg, si->src_reg,
9804 offsetof(struct sk_buff, sk));
9805 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9806 offsetof(struct sock_common,
9807 skc_v6_daddr.s6_addr32[0]) +
9808 off);
9809 #else
9810 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9811 #endif
9812 break;
9813 case offsetof(struct __sk_buff, local_ip6[0]) ...
9814 offsetof(struct __sk_buff, local_ip6[3]):
9815 #if IS_ENABLED(CONFIG_IPV6)
9816 BUILD_BUG_ON(sizeof_field(struct sock_common,
9817 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9818
9819 off = si->off;
9820 off -= offsetof(struct __sk_buff, local_ip6[0]);
9821
9822 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9823 si->dst_reg, si->src_reg,
9824 offsetof(struct sk_buff, sk));
9825 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9826 offsetof(struct sock_common,
9827 skc_v6_rcv_saddr.s6_addr32[0]) +
9828 off);
9829 #else
9830 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9831 #endif
9832 break;
9833
9834 case offsetof(struct __sk_buff, remote_port):
9835 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9836
9837 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9838 si->dst_reg, si->src_reg,
9839 offsetof(struct sk_buff, sk));
9840 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9841 bpf_target_off(struct sock_common,
9842 skc_dport,
9843 2, target_size));
9844 #ifndef __BIG_ENDIAN_BITFIELD
9845 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9846 #endif
9847 break;
9848
9849 case offsetof(struct __sk_buff, local_port):
9850 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9851
9852 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9853 si->dst_reg, si->src_reg,
9854 offsetof(struct sk_buff, sk));
9855 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9856 bpf_target_off(struct sock_common,
9857 skc_num, 2, target_size));
9858 break;
9859
9860 case offsetof(struct __sk_buff, tstamp):
9861 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9862
9863 if (type == BPF_WRITE)
9864 insn = bpf_convert_tstamp_write(prog, si, insn);
9865 else
9866 insn = bpf_convert_tstamp_read(prog, si, insn);
9867 break;
9868
9869 case offsetof(struct __sk_buff, tstamp_type):
9870 insn = bpf_convert_tstamp_type_read(si, insn);
9871 break;
9872
9873 case offsetof(struct __sk_buff, gso_segs):
9874 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9875 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9876 si->dst_reg, si->dst_reg,
9877 bpf_target_off(struct skb_shared_info,
9878 gso_segs, 2,
9879 target_size));
9880 break;
9881 case offsetof(struct __sk_buff, gso_size):
9882 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9883 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9884 si->dst_reg, si->dst_reg,
9885 bpf_target_off(struct skb_shared_info,
9886 gso_size, 2,
9887 target_size));
9888 break;
9889 case offsetof(struct __sk_buff, wire_len):
9890 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9891
9892 off = si->off;
9893 off -= offsetof(struct __sk_buff, wire_len);
9894 off += offsetof(struct sk_buff, cb);
9895 off += offsetof(struct qdisc_skb_cb, pkt_len);
9896 *target_size = 4;
9897 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9898 break;
9899
9900 case offsetof(struct __sk_buff, sk):
9901 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9902 si->dst_reg, si->src_reg,
9903 offsetof(struct sk_buff, sk));
9904 break;
9905 case offsetof(struct __sk_buff, hwtstamp):
9906 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9907 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9908
9909 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9910 *insn++ = BPF_LDX_MEM(BPF_DW,
9911 si->dst_reg, si->dst_reg,
9912 bpf_target_off(struct skb_shared_info,
9913 hwtstamps, 8,
9914 target_size));
9915 break;
9916 }
9917
9918 return insn - insn_buf;
9919 }
9920
bpf_sock_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)9921 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9922 const struct bpf_insn *si,
9923 struct bpf_insn *insn_buf,
9924 struct bpf_prog *prog, u32 *target_size)
9925 {
9926 struct bpf_insn *insn = insn_buf;
9927 int off;
9928
9929 switch (si->off) {
9930 case offsetof(struct bpf_sock, bound_dev_if):
9931 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9932
9933 if (type == BPF_WRITE)
9934 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9935 offsetof(struct sock, sk_bound_dev_if));
9936 else
9937 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9938 offsetof(struct sock, sk_bound_dev_if));
9939 break;
9940
9941 case offsetof(struct bpf_sock, mark):
9942 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9943
9944 if (type == BPF_WRITE)
9945 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9946 offsetof(struct sock, sk_mark));
9947 else
9948 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9949 offsetof(struct sock, sk_mark));
9950 break;
9951
9952 case offsetof(struct bpf_sock, priority):
9953 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9954
9955 if (type == BPF_WRITE)
9956 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9957 offsetof(struct sock, sk_priority));
9958 else
9959 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9960 offsetof(struct sock, sk_priority));
9961 break;
9962
9963 case offsetof(struct bpf_sock, family):
9964 *insn++ = BPF_LDX_MEM(
9965 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9966 si->dst_reg, si->src_reg,
9967 bpf_target_off(struct sock_common,
9968 skc_family,
9969 sizeof_field(struct sock_common,
9970 skc_family),
9971 target_size));
9972 break;
9973
9974 case offsetof(struct bpf_sock, type):
9975 *insn++ = BPF_LDX_MEM(
9976 BPF_FIELD_SIZEOF(struct sock, sk_type),
9977 si->dst_reg, si->src_reg,
9978 bpf_target_off(struct sock, sk_type,
9979 sizeof_field(struct sock, sk_type),
9980 target_size));
9981 break;
9982
9983 case offsetof(struct bpf_sock, protocol):
9984 *insn++ = BPF_LDX_MEM(
9985 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9986 si->dst_reg, si->src_reg,
9987 bpf_target_off(struct sock, sk_protocol,
9988 sizeof_field(struct sock, sk_protocol),
9989 target_size));
9990 break;
9991
9992 case offsetof(struct bpf_sock, src_ip4):
9993 *insn++ = BPF_LDX_MEM(
9994 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9995 bpf_target_off(struct sock_common, skc_rcv_saddr,
9996 sizeof_field(struct sock_common,
9997 skc_rcv_saddr),
9998 target_size));
9999 break;
10000
10001 case offsetof(struct bpf_sock, dst_ip4):
10002 *insn++ = BPF_LDX_MEM(
10003 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
10004 bpf_target_off(struct sock_common, skc_daddr,
10005 sizeof_field(struct sock_common,
10006 skc_daddr),
10007 target_size));
10008 break;
10009
10010 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
10011 #if IS_ENABLED(CONFIG_IPV6)
10012 off = si->off;
10013 off -= offsetof(struct bpf_sock, src_ip6[0]);
10014 *insn++ = BPF_LDX_MEM(
10015 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
10016 bpf_target_off(
10017 struct sock_common,
10018 skc_v6_rcv_saddr.s6_addr32[0],
10019 sizeof_field(struct sock_common,
10020 skc_v6_rcv_saddr.s6_addr32[0]),
10021 target_size) + off);
10022 #else
10023 (void)off;
10024 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10025 #endif
10026 break;
10027
10028 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
10029 #if IS_ENABLED(CONFIG_IPV6)
10030 off = si->off;
10031 off -= offsetof(struct bpf_sock, dst_ip6[0]);
10032 *insn++ = BPF_LDX_MEM(
10033 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
10034 bpf_target_off(struct sock_common,
10035 skc_v6_daddr.s6_addr32[0],
10036 sizeof_field(struct sock_common,
10037 skc_v6_daddr.s6_addr32[0]),
10038 target_size) + off);
10039 #else
10040 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10041 *target_size = 4;
10042 #endif
10043 break;
10044
10045 case offsetof(struct bpf_sock, src_port):
10046 *insn++ = BPF_LDX_MEM(
10047 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
10048 si->dst_reg, si->src_reg,
10049 bpf_target_off(struct sock_common, skc_num,
10050 sizeof_field(struct sock_common,
10051 skc_num),
10052 target_size));
10053 break;
10054
10055 case offsetof(struct bpf_sock, dst_port):
10056 *insn++ = BPF_LDX_MEM(
10057 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
10058 si->dst_reg, si->src_reg,
10059 bpf_target_off(struct sock_common, skc_dport,
10060 sizeof_field(struct sock_common,
10061 skc_dport),
10062 target_size));
10063 break;
10064
10065 case offsetof(struct bpf_sock, state):
10066 *insn++ = BPF_LDX_MEM(
10067 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
10068 si->dst_reg, si->src_reg,
10069 bpf_target_off(struct sock_common, skc_state,
10070 sizeof_field(struct sock_common,
10071 skc_state),
10072 target_size));
10073 break;
10074 case offsetof(struct bpf_sock, rx_queue_mapping):
10075 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
10076 *insn++ = BPF_LDX_MEM(
10077 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
10078 si->dst_reg, si->src_reg,
10079 bpf_target_off(struct sock, sk_rx_queue_mapping,
10080 sizeof_field(struct sock,
10081 sk_rx_queue_mapping),
10082 target_size));
10083 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
10084 1);
10085 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
10086 #else
10087 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
10088 *target_size = 2;
10089 #endif
10090 break;
10091 }
10092
10093 return insn - insn_buf;
10094 }
10095
tc_cls_act_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10096 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
10097 const struct bpf_insn *si,
10098 struct bpf_insn *insn_buf,
10099 struct bpf_prog *prog, u32 *target_size)
10100 {
10101 struct bpf_insn *insn = insn_buf;
10102
10103 switch (si->off) {
10104 case offsetof(struct __sk_buff, ifindex):
10105 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
10106 si->dst_reg, si->src_reg,
10107 offsetof(struct sk_buff, dev));
10108 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10109 bpf_target_off(struct net_device, ifindex, 4,
10110 target_size));
10111 break;
10112 default:
10113 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10114 target_size);
10115 }
10116
10117 return insn - insn_buf;
10118 }
10119
xdp_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10120 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
10121 const struct bpf_insn *si,
10122 struct bpf_insn *insn_buf,
10123 struct bpf_prog *prog, u32 *target_size)
10124 {
10125 struct bpf_insn *insn = insn_buf;
10126
10127 switch (si->off) {
10128 case offsetof(struct xdp_md, data):
10129 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
10130 si->dst_reg, si->src_reg,
10131 offsetof(struct xdp_buff, data));
10132 break;
10133 case offsetof(struct xdp_md, data_meta):
10134 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
10135 si->dst_reg, si->src_reg,
10136 offsetof(struct xdp_buff, data_meta));
10137 break;
10138 case offsetof(struct xdp_md, data_end):
10139 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
10140 si->dst_reg, si->src_reg,
10141 offsetof(struct xdp_buff, data_end));
10142 break;
10143 case offsetof(struct xdp_md, ingress_ifindex):
10144 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
10145 si->dst_reg, si->src_reg,
10146 offsetof(struct xdp_buff, rxq));
10147 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
10148 si->dst_reg, si->dst_reg,
10149 offsetof(struct xdp_rxq_info, dev));
10150 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10151 offsetof(struct net_device, ifindex));
10152 break;
10153 case offsetof(struct xdp_md, rx_queue_index):
10154 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
10155 si->dst_reg, si->src_reg,
10156 offsetof(struct xdp_buff, rxq));
10157 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10158 offsetof(struct xdp_rxq_info,
10159 queue_index));
10160 break;
10161 case offsetof(struct xdp_md, egress_ifindex):
10162 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
10163 si->dst_reg, si->src_reg,
10164 offsetof(struct xdp_buff, txq));
10165 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
10166 si->dst_reg, si->dst_reg,
10167 offsetof(struct xdp_txq_info, dev));
10168 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10169 offsetof(struct net_device, ifindex));
10170 break;
10171 }
10172
10173 return insn - insn_buf;
10174 }
10175
10176 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
10177 * context Structure, F is Field in context structure that contains a pointer
10178 * to Nested Structure of type NS that has the field NF.
10179 *
10180 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
10181 * sure that SIZE is not greater than actual size of S.F.NF.
10182 *
10183 * If offset OFF is provided, the load happens from that offset relative to
10184 * offset of NF.
10185 */
10186 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
10187 do { \
10188 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
10189 si->src_reg, offsetof(S, F)); \
10190 *insn++ = BPF_LDX_MEM( \
10191 SIZE, si->dst_reg, si->dst_reg, \
10192 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10193 target_size) \
10194 + OFF); \
10195 } while (0)
10196
10197 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
10198 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
10199 BPF_FIELD_SIZEOF(NS, NF), 0)
10200
10201 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
10202 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
10203 *
10204 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
10205 * "register" since two registers available in convert_ctx_access are not
10206 * enough: we can't override neither SRC, since it contains value to store, nor
10207 * DST since it contains pointer to context that may be used by later
10208 * instructions. But we need a temporary place to save pointer to nested
10209 * structure whose field we want to store to.
10210 */
10211 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
10212 do { \
10213 int tmp_reg = BPF_REG_9; \
10214 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10215 --tmp_reg; \
10216 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10217 --tmp_reg; \
10218 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
10219 offsetof(S, TF)); \
10220 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
10221 si->dst_reg, offsetof(S, F)); \
10222 *insn++ = BPF_RAW_INSN(SIZE | BPF_MEM | BPF_CLASS(si->code), \
10223 tmp_reg, si->src_reg, \
10224 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10225 target_size) \
10226 + OFF, \
10227 si->imm); \
10228 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
10229 offsetof(S, TF)); \
10230 } while (0)
10231
10232 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
10233 TF) \
10234 do { \
10235 if (type == BPF_WRITE) { \
10236 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
10237 OFF, TF); \
10238 } else { \
10239 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
10240 S, NS, F, NF, SIZE, OFF); \
10241 } \
10242 } while (0)
10243
10244 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
10245 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
10246 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
10247
sock_addr_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10248 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
10249 const struct bpf_insn *si,
10250 struct bpf_insn *insn_buf,
10251 struct bpf_prog *prog, u32 *target_size)
10252 {
10253 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
10254 struct bpf_insn *insn = insn_buf;
10255
10256 switch (si->off) {
10257 case offsetof(struct bpf_sock_addr, user_family):
10258 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10259 struct sockaddr, uaddr, sa_family);
10260 break;
10261
10262 case offsetof(struct bpf_sock_addr, user_ip4):
10263 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10264 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
10265 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
10266 break;
10267
10268 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
10269 off = si->off;
10270 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
10271 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10272 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10273 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
10274 tmp_reg);
10275 break;
10276
10277 case offsetof(struct bpf_sock_addr, user_port):
10278 /* To get port we need to know sa_family first and then treat
10279 * sockaddr as either sockaddr_in or sockaddr_in6.
10280 * Though we can simplify since port field has same offset and
10281 * size in both structures.
10282 * Here we check this invariant and use just one of the
10283 * structures if it's true.
10284 */
10285 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
10286 offsetof(struct sockaddr_in6, sin6_port));
10287 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
10288 sizeof_field(struct sockaddr_in6, sin6_port));
10289 /* Account for sin6_port being smaller than user_port. */
10290 port_size = min(port_size, BPF_LDST_BYTES(si));
10291 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10292 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10293 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
10294 break;
10295
10296 case offsetof(struct bpf_sock_addr, family):
10297 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10298 struct sock, sk, sk_family);
10299 break;
10300
10301 case offsetof(struct bpf_sock_addr, type):
10302 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10303 struct sock, sk, sk_type);
10304 break;
10305
10306 case offsetof(struct bpf_sock_addr, protocol):
10307 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10308 struct sock, sk, sk_protocol);
10309 break;
10310
10311 case offsetof(struct bpf_sock_addr, msg_src_ip4):
10312 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10313 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10314 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10315 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10316 break;
10317
10318 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10319 msg_src_ip6[3]):
10320 off = si->off;
10321 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10322 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10323 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10324 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10325 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10326 break;
10327 case offsetof(struct bpf_sock_addr, sk):
10328 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10329 si->dst_reg, si->src_reg,
10330 offsetof(struct bpf_sock_addr_kern, sk));
10331 break;
10332 }
10333
10334 return insn - insn_buf;
10335 }
10336
sock_ops_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10337 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10338 const struct bpf_insn *si,
10339 struct bpf_insn *insn_buf,
10340 struct bpf_prog *prog,
10341 u32 *target_size)
10342 {
10343 struct bpf_insn *insn = insn_buf;
10344 int off;
10345
10346 /* Helper macro for adding read access to tcp_sock or sock fields. */
10347 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10348 do { \
10349 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
10350 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10351 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10352 if (si->dst_reg == reg || si->src_reg == reg) \
10353 reg--; \
10354 if (si->dst_reg == reg || si->src_reg == reg) \
10355 reg--; \
10356 if (si->dst_reg == si->src_reg) { \
10357 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10358 offsetof(struct bpf_sock_ops_kern, \
10359 temp)); \
10360 fullsock_reg = reg; \
10361 jmp += 2; \
10362 } \
10363 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10364 struct bpf_sock_ops_kern, \
10365 is_fullsock), \
10366 fullsock_reg, si->src_reg, \
10367 offsetof(struct bpf_sock_ops_kern, \
10368 is_fullsock)); \
10369 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10370 if (si->dst_reg == si->src_reg) \
10371 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10372 offsetof(struct bpf_sock_ops_kern, \
10373 temp)); \
10374 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10375 struct bpf_sock_ops_kern, sk),\
10376 si->dst_reg, si->src_reg, \
10377 offsetof(struct bpf_sock_ops_kern, sk));\
10378 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
10379 OBJ_FIELD), \
10380 si->dst_reg, si->dst_reg, \
10381 offsetof(OBJ, OBJ_FIELD)); \
10382 if (si->dst_reg == si->src_reg) { \
10383 *insn++ = BPF_JMP_A(1); \
10384 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10385 offsetof(struct bpf_sock_ops_kern, \
10386 temp)); \
10387 } \
10388 } while (0)
10389
10390 #define SOCK_OPS_GET_SK() \
10391 do { \
10392 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
10393 if (si->dst_reg == reg || si->src_reg == reg) \
10394 reg--; \
10395 if (si->dst_reg == reg || si->src_reg == reg) \
10396 reg--; \
10397 if (si->dst_reg == si->src_reg) { \
10398 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10399 offsetof(struct bpf_sock_ops_kern, \
10400 temp)); \
10401 fullsock_reg = reg; \
10402 jmp += 2; \
10403 } \
10404 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10405 struct bpf_sock_ops_kern, \
10406 is_fullsock), \
10407 fullsock_reg, si->src_reg, \
10408 offsetof(struct bpf_sock_ops_kern, \
10409 is_fullsock)); \
10410 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10411 if (si->dst_reg == si->src_reg) \
10412 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10413 offsetof(struct bpf_sock_ops_kern, \
10414 temp)); \
10415 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10416 struct bpf_sock_ops_kern, sk),\
10417 si->dst_reg, si->src_reg, \
10418 offsetof(struct bpf_sock_ops_kern, sk));\
10419 if (si->dst_reg == si->src_reg) { \
10420 *insn++ = BPF_JMP_A(1); \
10421 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10422 offsetof(struct bpf_sock_ops_kern, \
10423 temp)); \
10424 } \
10425 } while (0)
10426
10427 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10428 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10429
10430 /* Helper macro for adding write access to tcp_sock or sock fields.
10431 * The macro is called with two registers, dst_reg which contains a pointer
10432 * to ctx (context) and src_reg which contains the value that should be
10433 * stored. However, we need an additional register since we cannot overwrite
10434 * dst_reg because it may be used later in the program.
10435 * Instead we "borrow" one of the other register. We first save its value
10436 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10437 * it at the end of the macro.
10438 */
10439 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10440 do { \
10441 int reg = BPF_REG_9; \
10442 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10443 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10444 if (si->dst_reg == reg || si->src_reg == reg) \
10445 reg--; \
10446 if (si->dst_reg == reg || si->src_reg == reg) \
10447 reg--; \
10448 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
10449 offsetof(struct bpf_sock_ops_kern, \
10450 temp)); \
10451 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10452 struct bpf_sock_ops_kern, \
10453 is_fullsock), \
10454 reg, si->dst_reg, \
10455 offsetof(struct bpf_sock_ops_kern, \
10456 is_fullsock)); \
10457 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
10458 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10459 struct bpf_sock_ops_kern, sk),\
10460 reg, si->dst_reg, \
10461 offsetof(struct bpf_sock_ops_kern, sk));\
10462 *insn++ = BPF_RAW_INSN(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD) | \
10463 BPF_MEM | BPF_CLASS(si->code), \
10464 reg, si->src_reg, \
10465 offsetof(OBJ, OBJ_FIELD), \
10466 si->imm); \
10467 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
10468 offsetof(struct bpf_sock_ops_kern, \
10469 temp)); \
10470 } while (0)
10471
10472 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
10473 do { \
10474 if (TYPE == BPF_WRITE) \
10475 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10476 else \
10477 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10478 } while (0)
10479
10480 switch (si->off) {
10481 case offsetof(struct bpf_sock_ops, op):
10482 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10483 op),
10484 si->dst_reg, si->src_reg,
10485 offsetof(struct bpf_sock_ops_kern, op));
10486 break;
10487
10488 case offsetof(struct bpf_sock_ops, replylong[0]) ...
10489 offsetof(struct bpf_sock_ops, replylong[3]):
10490 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10491 sizeof_field(struct bpf_sock_ops_kern, reply));
10492 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10493 sizeof_field(struct bpf_sock_ops_kern, replylong));
10494 off = si->off;
10495 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10496 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10497 if (type == BPF_WRITE)
10498 *insn++ = BPF_EMIT_STORE(BPF_W, si, off);
10499 else
10500 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10501 off);
10502 break;
10503
10504 case offsetof(struct bpf_sock_ops, family):
10505 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10506
10507 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10508 struct bpf_sock_ops_kern, sk),
10509 si->dst_reg, si->src_reg,
10510 offsetof(struct bpf_sock_ops_kern, sk));
10511 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10512 offsetof(struct sock_common, skc_family));
10513 break;
10514
10515 case offsetof(struct bpf_sock_ops, remote_ip4):
10516 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10517
10518 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10519 struct bpf_sock_ops_kern, sk),
10520 si->dst_reg, si->src_reg,
10521 offsetof(struct bpf_sock_ops_kern, sk));
10522 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10523 offsetof(struct sock_common, skc_daddr));
10524 break;
10525
10526 case offsetof(struct bpf_sock_ops, local_ip4):
10527 BUILD_BUG_ON(sizeof_field(struct sock_common,
10528 skc_rcv_saddr) != 4);
10529
10530 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10531 struct bpf_sock_ops_kern, sk),
10532 si->dst_reg, si->src_reg,
10533 offsetof(struct bpf_sock_ops_kern, sk));
10534 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10535 offsetof(struct sock_common,
10536 skc_rcv_saddr));
10537 break;
10538
10539 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10540 offsetof(struct bpf_sock_ops, remote_ip6[3]):
10541 #if IS_ENABLED(CONFIG_IPV6)
10542 BUILD_BUG_ON(sizeof_field(struct sock_common,
10543 skc_v6_daddr.s6_addr32[0]) != 4);
10544
10545 off = si->off;
10546 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10547 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10548 struct bpf_sock_ops_kern, sk),
10549 si->dst_reg, si->src_reg,
10550 offsetof(struct bpf_sock_ops_kern, sk));
10551 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10552 offsetof(struct sock_common,
10553 skc_v6_daddr.s6_addr32[0]) +
10554 off);
10555 #else
10556 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10557 #endif
10558 break;
10559
10560 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10561 offsetof(struct bpf_sock_ops, local_ip6[3]):
10562 #if IS_ENABLED(CONFIG_IPV6)
10563 BUILD_BUG_ON(sizeof_field(struct sock_common,
10564 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10565
10566 off = si->off;
10567 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10568 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10569 struct bpf_sock_ops_kern, sk),
10570 si->dst_reg, si->src_reg,
10571 offsetof(struct bpf_sock_ops_kern, sk));
10572 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10573 offsetof(struct sock_common,
10574 skc_v6_rcv_saddr.s6_addr32[0]) +
10575 off);
10576 #else
10577 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10578 #endif
10579 break;
10580
10581 case offsetof(struct bpf_sock_ops, remote_port):
10582 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10583
10584 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10585 struct bpf_sock_ops_kern, sk),
10586 si->dst_reg, si->src_reg,
10587 offsetof(struct bpf_sock_ops_kern, sk));
10588 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10589 offsetof(struct sock_common, skc_dport));
10590 #ifndef __BIG_ENDIAN_BITFIELD
10591 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10592 #endif
10593 break;
10594
10595 case offsetof(struct bpf_sock_ops, local_port):
10596 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10597
10598 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10599 struct bpf_sock_ops_kern, sk),
10600 si->dst_reg, si->src_reg,
10601 offsetof(struct bpf_sock_ops_kern, sk));
10602 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10603 offsetof(struct sock_common, skc_num));
10604 break;
10605
10606 case offsetof(struct bpf_sock_ops, is_fullsock):
10607 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10608 struct bpf_sock_ops_kern,
10609 is_fullsock),
10610 si->dst_reg, si->src_reg,
10611 offsetof(struct bpf_sock_ops_kern,
10612 is_fullsock));
10613 break;
10614
10615 case offsetof(struct bpf_sock_ops, state):
10616 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10617
10618 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10619 struct bpf_sock_ops_kern, sk),
10620 si->dst_reg, si->src_reg,
10621 offsetof(struct bpf_sock_ops_kern, sk));
10622 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10623 offsetof(struct sock_common, skc_state));
10624 break;
10625
10626 case offsetof(struct bpf_sock_ops, rtt_min):
10627 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10628 sizeof(struct minmax));
10629 BUILD_BUG_ON(sizeof(struct minmax) <
10630 sizeof(struct minmax_sample));
10631
10632 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10633 struct bpf_sock_ops_kern, sk),
10634 si->dst_reg, si->src_reg,
10635 offsetof(struct bpf_sock_ops_kern, sk));
10636 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10637 offsetof(struct tcp_sock, rtt_min) +
10638 sizeof_field(struct minmax_sample, t));
10639 break;
10640
10641 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10642 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10643 struct tcp_sock);
10644 break;
10645
10646 case offsetof(struct bpf_sock_ops, sk_txhash):
10647 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10648 struct sock, type);
10649 break;
10650 case offsetof(struct bpf_sock_ops, snd_cwnd):
10651 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10652 break;
10653 case offsetof(struct bpf_sock_ops, srtt_us):
10654 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10655 break;
10656 case offsetof(struct bpf_sock_ops, snd_ssthresh):
10657 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10658 break;
10659 case offsetof(struct bpf_sock_ops, rcv_nxt):
10660 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10661 break;
10662 case offsetof(struct bpf_sock_ops, snd_nxt):
10663 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10664 break;
10665 case offsetof(struct bpf_sock_ops, snd_una):
10666 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10667 break;
10668 case offsetof(struct bpf_sock_ops, mss_cache):
10669 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10670 break;
10671 case offsetof(struct bpf_sock_ops, ecn_flags):
10672 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10673 break;
10674 case offsetof(struct bpf_sock_ops, rate_delivered):
10675 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10676 break;
10677 case offsetof(struct bpf_sock_ops, rate_interval_us):
10678 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10679 break;
10680 case offsetof(struct bpf_sock_ops, packets_out):
10681 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10682 break;
10683 case offsetof(struct bpf_sock_ops, retrans_out):
10684 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10685 break;
10686 case offsetof(struct bpf_sock_ops, total_retrans):
10687 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10688 break;
10689 case offsetof(struct bpf_sock_ops, segs_in):
10690 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10691 break;
10692 case offsetof(struct bpf_sock_ops, data_segs_in):
10693 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10694 break;
10695 case offsetof(struct bpf_sock_ops, segs_out):
10696 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10697 break;
10698 case offsetof(struct bpf_sock_ops, data_segs_out):
10699 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10700 break;
10701 case offsetof(struct bpf_sock_ops, lost_out):
10702 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10703 break;
10704 case offsetof(struct bpf_sock_ops, sacked_out):
10705 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10706 break;
10707 case offsetof(struct bpf_sock_ops, bytes_received):
10708 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10709 break;
10710 case offsetof(struct bpf_sock_ops, bytes_acked):
10711 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10712 break;
10713 case offsetof(struct bpf_sock_ops, sk):
10714 SOCK_OPS_GET_SK();
10715 break;
10716 case offsetof(struct bpf_sock_ops, skb_data_end):
10717 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10718 skb_data_end),
10719 si->dst_reg, si->src_reg,
10720 offsetof(struct bpf_sock_ops_kern,
10721 skb_data_end));
10722 break;
10723 case offsetof(struct bpf_sock_ops, skb_data):
10724 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10725 skb),
10726 si->dst_reg, si->src_reg,
10727 offsetof(struct bpf_sock_ops_kern,
10728 skb));
10729 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10730 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10731 si->dst_reg, si->dst_reg,
10732 offsetof(struct sk_buff, data));
10733 break;
10734 case offsetof(struct bpf_sock_ops, skb_len):
10735 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10736 skb),
10737 si->dst_reg, si->src_reg,
10738 offsetof(struct bpf_sock_ops_kern,
10739 skb));
10740 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10741 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10742 si->dst_reg, si->dst_reg,
10743 offsetof(struct sk_buff, len));
10744 break;
10745 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10746 off = offsetof(struct sk_buff, cb);
10747 off += offsetof(struct tcp_skb_cb, tcp_flags);
10748 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10749 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10750 skb),
10751 si->dst_reg, si->src_reg,
10752 offsetof(struct bpf_sock_ops_kern,
10753 skb));
10754 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10755 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10756 tcp_flags),
10757 si->dst_reg, si->dst_reg, off);
10758 break;
10759 case offsetof(struct bpf_sock_ops, skb_hwtstamp): {
10760 struct bpf_insn *jmp_on_null_skb;
10761
10762 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10763 skb),
10764 si->dst_reg, si->src_reg,
10765 offsetof(struct bpf_sock_ops_kern,
10766 skb));
10767 /* Reserve one insn to test skb == NULL */
10768 jmp_on_null_skb = insn++;
10769 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn);
10770 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
10771 bpf_target_off(struct skb_shared_info,
10772 hwtstamps, 8,
10773 target_size));
10774 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0,
10775 insn - jmp_on_null_skb - 1);
10776 break;
10777 }
10778 }
10779 return insn - insn_buf;
10780 }
10781
10782 /* data_end = skb->data + skb_headlen() */
bpf_convert_data_end_access(const struct bpf_insn * si,struct bpf_insn * insn)10783 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10784 struct bpf_insn *insn)
10785 {
10786 int reg;
10787 int temp_reg_off = offsetof(struct sk_buff, cb) +
10788 offsetof(struct sk_skb_cb, temp_reg);
10789
10790 if (si->src_reg == si->dst_reg) {
10791 /* We need an extra register, choose and save a register. */
10792 reg = BPF_REG_9;
10793 if (si->src_reg == reg || si->dst_reg == reg)
10794 reg--;
10795 if (si->src_reg == reg || si->dst_reg == reg)
10796 reg--;
10797 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10798 } else {
10799 reg = si->dst_reg;
10800 }
10801
10802 /* reg = skb->data */
10803 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10804 reg, si->src_reg,
10805 offsetof(struct sk_buff, data));
10806 /* AX = skb->len */
10807 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10808 BPF_REG_AX, si->src_reg,
10809 offsetof(struct sk_buff, len));
10810 /* reg = skb->data + skb->len */
10811 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10812 /* AX = skb->data_len */
10813 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10814 BPF_REG_AX, si->src_reg,
10815 offsetof(struct sk_buff, data_len));
10816
10817 /* reg = skb->data + skb->len - skb->data_len */
10818 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10819
10820 if (si->src_reg == si->dst_reg) {
10821 /* Restore the saved register */
10822 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10823 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10824 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10825 }
10826
10827 return insn;
10828 }
10829
sk_skb_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10830 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10831 const struct bpf_insn *si,
10832 struct bpf_insn *insn_buf,
10833 struct bpf_prog *prog, u32 *target_size)
10834 {
10835 struct bpf_insn *insn = insn_buf;
10836 int off;
10837
10838 switch (si->off) {
10839 case offsetof(struct __sk_buff, data_end):
10840 insn = bpf_convert_data_end_access(si, insn);
10841 break;
10842 case offsetof(struct __sk_buff, cb[0]) ...
10843 offsetofend(struct __sk_buff, cb[4]) - 1:
10844 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10845 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10846 offsetof(struct sk_skb_cb, data)) %
10847 sizeof(__u64));
10848
10849 prog->cb_access = 1;
10850 off = si->off;
10851 off -= offsetof(struct __sk_buff, cb[0]);
10852 off += offsetof(struct sk_buff, cb);
10853 off += offsetof(struct sk_skb_cb, data);
10854 if (type == BPF_WRITE)
10855 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
10856 else
10857 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10858 si->src_reg, off);
10859 break;
10860
10861
10862 default:
10863 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10864 target_size);
10865 }
10866
10867 return insn - insn_buf;
10868 }
10869
sk_msg_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10870 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10871 const struct bpf_insn *si,
10872 struct bpf_insn *insn_buf,
10873 struct bpf_prog *prog, u32 *target_size)
10874 {
10875 struct bpf_insn *insn = insn_buf;
10876 #if IS_ENABLED(CONFIG_IPV6)
10877 int off;
10878 #endif
10879
10880 /* convert ctx uses the fact sg element is first in struct */
10881 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10882
10883 switch (si->off) {
10884 case offsetof(struct sk_msg_md, data):
10885 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10886 si->dst_reg, si->src_reg,
10887 offsetof(struct sk_msg, data));
10888 break;
10889 case offsetof(struct sk_msg_md, data_end):
10890 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10891 si->dst_reg, si->src_reg,
10892 offsetof(struct sk_msg, data_end));
10893 break;
10894 case offsetof(struct sk_msg_md, family):
10895 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10896
10897 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10898 struct sk_msg, sk),
10899 si->dst_reg, si->src_reg,
10900 offsetof(struct sk_msg, sk));
10901 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10902 offsetof(struct sock_common, skc_family));
10903 break;
10904
10905 case offsetof(struct sk_msg_md, remote_ip4):
10906 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10907
10908 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10909 struct sk_msg, sk),
10910 si->dst_reg, si->src_reg,
10911 offsetof(struct sk_msg, sk));
10912 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10913 offsetof(struct sock_common, skc_daddr));
10914 break;
10915
10916 case offsetof(struct sk_msg_md, local_ip4):
10917 BUILD_BUG_ON(sizeof_field(struct sock_common,
10918 skc_rcv_saddr) != 4);
10919
10920 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10921 struct sk_msg, sk),
10922 si->dst_reg, si->src_reg,
10923 offsetof(struct sk_msg, sk));
10924 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10925 offsetof(struct sock_common,
10926 skc_rcv_saddr));
10927 break;
10928
10929 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10930 offsetof(struct sk_msg_md, remote_ip6[3]):
10931 #if IS_ENABLED(CONFIG_IPV6)
10932 BUILD_BUG_ON(sizeof_field(struct sock_common,
10933 skc_v6_daddr.s6_addr32[0]) != 4);
10934
10935 off = si->off;
10936 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10937 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10938 struct sk_msg, sk),
10939 si->dst_reg, si->src_reg,
10940 offsetof(struct sk_msg, sk));
10941 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10942 offsetof(struct sock_common,
10943 skc_v6_daddr.s6_addr32[0]) +
10944 off);
10945 #else
10946 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10947 #endif
10948 break;
10949
10950 case offsetof(struct sk_msg_md, local_ip6[0]) ...
10951 offsetof(struct sk_msg_md, local_ip6[3]):
10952 #if IS_ENABLED(CONFIG_IPV6)
10953 BUILD_BUG_ON(sizeof_field(struct sock_common,
10954 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10955
10956 off = si->off;
10957 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10958 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10959 struct sk_msg, sk),
10960 si->dst_reg, si->src_reg,
10961 offsetof(struct sk_msg, sk));
10962 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10963 offsetof(struct sock_common,
10964 skc_v6_rcv_saddr.s6_addr32[0]) +
10965 off);
10966 #else
10967 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10968 #endif
10969 break;
10970
10971 case offsetof(struct sk_msg_md, remote_port):
10972 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10973
10974 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10975 struct sk_msg, sk),
10976 si->dst_reg, si->src_reg,
10977 offsetof(struct sk_msg, sk));
10978 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10979 offsetof(struct sock_common, skc_dport));
10980 #ifndef __BIG_ENDIAN_BITFIELD
10981 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10982 #endif
10983 break;
10984
10985 case offsetof(struct sk_msg_md, local_port):
10986 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10987
10988 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10989 struct sk_msg, sk),
10990 si->dst_reg, si->src_reg,
10991 offsetof(struct sk_msg, sk));
10992 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10993 offsetof(struct sock_common, skc_num));
10994 break;
10995
10996 case offsetof(struct sk_msg_md, size):
10997 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10998 si->dst_reg, si->src_reg,
10999 offsetof(struct sk_msg_sg, size));
11000 break;
11001
11002 case offsetof(struct sk_msg_md, sk):
11003 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
11004 si->dst_reg, si->src_reg,
11005 offsetof(struct sk_msg, sk));
11006 break;
11007 }
11008
11009 return insn - insn_buf;
11010 }
11011
11012 const struct bpf_verifier_ops sk_filter_verifier_ops = {
11013 .get_func_proto = sk_filter_func_proto,
11014 .is_valid_access = sk_filter_is_valid_access,
11015 .convert_ctx_access = bpf_convert_ctx_access,
11016 .gen_ld_abs = bpf_gen_ld_abs,
11017 };
11018
11019 const struct bpf_prog_ops sk_filter_prog_ops = {
11020 .test_run = bpf_prog_test_run_skb,
11021 };
11022
11023 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
11024 .get_func_proto = tc_cls_act_func_proto,
11025 .is_valid_access = tc_cls_act_is_valid_access,
11026 .convert_ctx_access = tc_cls_act_convert_ctx_access,
11027 .gen_prologue = tc_cls_act_prologue,
11028 .gen_ld_abs = bpf_gen_ld_abs,
11029 .btf_struct_access = tc_cls_act_btf_struct_access,
11030 };
11031
11032 const struct bpf_prog_ops tc_cls_act_prog_ops = {
11033 .test_run = bpf_prog_test_run_skb,
11034 };
11035
11036 const struct bpf_verifier_ops xdp_verifier_ops = {
11037 .get_func_proto = xdp_func_proto,
11038 .is_valid_access = xdp_is_valid_access,
11039 .convert_ctx_access = xdp_convert_ctx_access,
11040 .gen_prologue = bpf_noop_prologue,
11041 .btf_struct_access = xdp_btf_struct_access,
11042 };
11043
11044 const struct bpf_prog_ops xdp_prog_ops = {
11045 .test_run = bpf_prog_test_run_xdp,
11046 };
11047
11048 const struct bpf_verifier_ops cg_skb_verifier_ops = {
11049 .get_func_proto = cg_skb_func_proto,
11050 .is_valid_access = cg_skb_is_valid_access,
11051 .convert_ctx_access = bpf_convert_ctx_access,
11052 };
11053
11054 const struct bpf_prog_ops cg_skb_prog_ops = {
11055 .test_run = bpf_prog_test_run_skb,
11056 };
11057
11058 const struct bpf_verifier_ops lwt_in_verifier_ops = {
11059 .get_func_proto = lwt_in_func_proto,
11060 .is_valid_access = lwt_is_valid_access,
11061 .convert_ctx_access = bpf_convert_ctx_access,
11062 };
11063
11064 const struct bpf_prog_ops lwt_in_prog_ops = {
11065 .test_run = bpf_prog_test_run_skb,
11066 };
11067
11068 const struct bpf_verifier_ops lwt_out_verifier_ops = {
11069 .get_func_proto = lwt_out_func_proto,
11070 .is_valid_access = lwt_is_valid_access,
11071 .convert_ctx_access = bpf_convert_ctx_access,
11072 };
11073
11074 const struct bpf_prog_ops lwt_out_prog_ops = {
11075 .test_run = bpf_prog_test_run_skb,
11076 };
11077
11078 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
11079 .get_func_proto = lwt_xmit_func_proto,
11080 .is_valid_access = lwt_is_valid_access,
11081 .convert_ctx_access = bpf_convert_ctx_access,
11082 .gen_prologue = tc_cls_act_prologue,
11083 };
11084
11085 const struct bpf_prog_ops lwt_xmit_prog_ops = {
11086 .test_run = bpf_prog_test_run_skb,
11087 };
11088
11089 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
11090 .get_func_proto = lwt_seg6local_func_proto,
11091 .is_valid_access = lwt_is_valid_access,
11092 .convert_ctx_access = bpf_convert_ctx_access,
11093 };
11094
11095 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
11096 };
11097
11098 const struct bpf_verifier_ops cg_sock_verifier_ops = {
11099 .get_func_proto = sock_filter_func_proto,
11100 .is_valid_access = sock_filter_is_valid_access,
11101 .convert_ctx_access = bpf_sock_convert_ctx_access,
11102 };
11103
11104 const struct bpf_prog_ops cg_sock_prog_ops = {
11105 };
11106
11107 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
11108 .get_func_proto = sock_addr_func_proto,
11109 .is_valid_access = sock_addr_is_valid_access,
11110 .convert_ctx_access = sock_addr_convert_ctx_access,
11111 };
11112
11113 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
11114 };
11115
11116 const struct bpf_verifier_ops sock_ops_verifier_ops = {
11117 .get_func_proto = sock_ops_func_proto,
11118 .is_valid_access = sock_ops_is_valid_access,
11119 .convert_ctx_access = sock_ops_convert_ctx_access,
11120 };
11121
11122 const struct bpf_prog_ops sock_ops_prog_ops = {
11123 };
11124
11125 const struct bpf_verifier_ops sk_skb_verifier_ops = {
11126 .get_func_proto = sk_skb_func_proto,
11127 .is_valid_access = sk_skb_is_valid_access,
11128 .convert_ctx_access = sk_skb_convert_ctx_access,
11129 .gen_prologue = sk_skb_prologue,
11130 };
11131
11132 const struct bpf_prog_ops sk_skb_prog_ops = {
11133 };
11134
11135 const struct bpf_verifier_ops sk_msg_verifier_ops = {
11136 .get_func_proto = sk_msg_func_proto,
11137 .is_valid_access = sk_msg_is_valid_access,
11138 .convert_ctx_access = sk_msg_convert_ctx_access,
11139 .gen_prologue = bpf_noop_prologue,
11140 };
11141
11142 const struct bpf_prog_ops sk_msg_prog_ops = {
11143 };
11144
11145 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
11146 .get_func_proto = flow_dissector_func_proto,
11147 .is_valid_access = flow_dissector_is_valid_access,
11148 .convert_ctx_access = flow_dissector_convert_ctx_access,
11149 };
11150
11151 const struct bpf_prog_ops flow_dissector_prog_ops = {
11152 .test_run = bpf_prog_test_run_flow_dissector,
11153 };
11154
sk_detach_filter(struct sock * sk)11155 int sk_detach_filter(struct sock *sk)
11156 {
11157 int ret = -ENOENT;
11158 struct sk_filter *filter;
11159
11160 if (sock_flag(sk, SOCK_FILTER_LOCKED))
11161 return -EPERM;
11162
11163 filter = rcu_dereference_protected(sk->sk_filter,
11164 lockdep_sock_is_held(sk));
11165 if (filter) {
11166 RCU_INIT_POINTER(sk->sk_filter, NULL);
11167 sk_filter_uncharge(sk, filter);
11168 ret = 0;
11169 }
11170
11171 return ret;
11172 }
11173 EXPORT_SYMBOL_GPL(sk_detach_filter);
11174
sk_get_filter(struct sock * sk,sockptr_t optval,unsigned int len)11175 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
11176 {
11177 struct sock_fprog_kern *fprog;
11178 struct sk_filter *filter;
11179 int ret = 0;
11180
11181 sockopt_lock_sock(sk);
11182 filter = rcu_dereference_protected(sk->sk_filter,
11183 lockdep_sock_is_held(sk));
11184 if (!filter)
11185 goto out;
11186
11187 /* We're copying the filter that has been originally attached,
11188 * so no conversion/decode needed anymore. eBPF programs that
11189 * have no original program cannot be dumped through this.
11190 */
11191 ret = -EACCES;
11192 fprog = filter->prog->orig_prog;
11193 if (!fprog)
11194 goto out;
11195
11196 ret = fprog->len;
11197 if (!len)
11198 /* User space only enquires number of filter blocks. */
11199 goto out;
11200
11201 ret = -EINVAL;
11202 if (len < fprog->len)
11203 goto out;
11204
11205 ret = -EFAULT;
11206 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
11207 goto out;
11208
11209 /* Instead of bytes, the API requests to return the number
11210 * of filter blocks.
11211 */
11212 ret = fprog->len;
11213 out:
11214 sockopt_release_sock(sk);
11215 return ret;
11216 }
11217
11218 #ifdef CONFIG_INET
bpf_init_reuseport_kern(struct sk_reuseport_kern * reuse_kern,struct sock_reuseport * reuse,struct sock * sk,struct sk_buff * skb,struct sock * migrating_sk,u32 hash)11219 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
11220 struct sock_reuseport *reuse,
11221 struct sock *sk, struct sk_buff *skb,
11222 struct sock *migrating_sk,
11223 u32 hash)
11224 {
11225 reuse_kern->skb = skb;
11226 reuse_kern->sk = sk;
11227 reuse_kern->selected_sk = NULL;
11228 reuse_kern->migrating_sk = migrating_sk;
11229 reuse_kern->data_end = skb->data + skb_headlen(skb);
11230 reuse_kern->hash = hash;
11231 reuse_kern->reuseport_id = reuse->reuseport_id;
11232 reuse_kern->bind_inany = reuse->bind_inany;
11233 }
11234
bpf_run_sk_reuseport(struct sock_reuseport * reuse,struct sock * sk,struct bpf_prog * prog,struct sk_buff * skb,struct sock * migrating_sk,u32 hash)11235 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
11236 struct bpf_prog *prog, struct sk_buff *skb,
11237 struct sock *migrating_sk,
11238 u32 hash)
11239 {
11240 struct sk_reuseport_kern reuse_kern;
11241 enum sk_action action;
11242
11243 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
11244 action = bpf_prog_run(prog, &reuse_kern);
11245
11246 if (action == SK_PASS)
11247 return reuse_kern.selected_sk;
11248 else
11249 return ERR_PTR(-ECONNREFUSED);
11250 }
11251
BPF_CALL_4(sk_select_reuseport,struct sk_reuseport_kern *,reuse_kern,struct bpf_map *,map,void *,key,u32,flags)11252 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
11253 struct bpf_map *, map, void *, key, u32, flags)
11254 {
11255 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
11256 struct sock_reuseport *reuse;
11257 struct sock *selected_sk;
11258
11259 selected_sk = map->ops->map_lookup_elem(map, key);
11260 if (!selected_sk)
11261 return -ENOENT;
11262
11263 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
11264 if (!reuse) {
11265 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
11266 if (sk_is_refcounted(selected_sk))
11267 sock_put(selected_sk);
11268
11269 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
11270 * The only (!reuse) case here is - the sk has already been
11271 * unhashed (e.g. by close()), so treat it as -ENOENT.
11272 *
11273 * Other maps (e.g. sock_map) do not provide this guarantee and
11274 * the sk may never be in the reuseport group to begin with.
11275 */
11276 return is_sockarray ? -ENOENT : -EINVAL;
11277 }
11278
11279 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
11280 struct sock *sk = reuse_kern->sk;
11281
11282 if (sk->sk_protocol != selected_sk->sk_protocol)
11283 return -EPROTOTYPE;
11284 else if (sk->sk_family != selected_sk->sk_family)
11285 return -EAFNOSUPPORT;
11286
11287 /* Catch all. Likely bound to a different sockaddr. */
11288 return -EBADFD;
11289 }
11290
11291 reuse_kern->selected_sk = selected_sk;
11292
11293 return 0;
11294 }
11295
11296 static const struct bpf_func_proto sk_select_reuseport_proto = {
11297 .func = sk_select_reuseport,
11298 .gpl_only = false,
11299 .ret_type = RET_INTEGER,
11300 .arg1_type = ARG_PTR_TO_CTX,
11301 .arg2_type = ARG_CONST_MAP_PTR,
11302 .arg3_type = ARG_PTR_TO_MAP_KEY,
11303 .arg4_type = ARG_ANYTHING,
11304 };
11305
BPF_CALL_4(sk_reuseport_load_bytes,const struct sk_reuseport_kern *,reuse_kern,u32,offset,void *,to,u32,len)11306 BPF_CALL_4(sk_reuseport_load_bytes,
11307 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11308 void *, to, u32, len)
11309 {
11310 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11311 }
11312
11313 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11314 .func = sk_reuseport_load_bytes,
11315 .gpl_only = false,
11316 .ret_type = RET_INTEGER,
11317 .arg1_type = ARG_PTR_TO_CTX,
11318 .arg2_type = ARG_ANYTHING,
11319 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11320 .arg4_type = ARG_CONST_SIZE,
11321 };
11322
BPF_CALL_5(sk_reuseport_load_bytes_relative,const struct sk_reuseport_kern *,reuse_kern,u32,offset,void *,to,u32,len,u32,start_header)11323 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11324 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11325 void *, to, u32, len, u32, start_header)
11326 {
11327 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11328 len, start_header);
11329 }
11330
11331 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11332 .func = sk_reuseport_load_bytes_relative,
11333 .gpl_only = false,
11334 .ret_type = RET_INTEGER,
11335 .arg1_type = ARG_PTR_TO_CTX,
11336 .arg2_type = ARG_ANYTHING,
11337 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11338 .arg4_type = ARG_CONST_SIZE,
11339 .arg5_type = ARG_ANYTHING,
11340 };
11341
11342 static const struct bpf_func_proto *
sk_reuseport_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)11343 sk_reuseport_func_proto(enum bpf_func_id func_id,
11344 const struct bpf_prog *prog)
11345 {
11346 switch (func_id) {
11347 case BPF_FUNC_sk_select_reuseport:
11348 return &sk_select_reuseport_proto;
11349 case BPF_FUNC_skb_load_bytes:
11350 return &sk_reuseport_load_bytes_proto;
11351 case BPF_FUNC_skb_load_bytes_relative:
11352 return &sk_reuseport_load_bytes_relative_proto;
11353 case BPF_FUNC_get_socket_cookie:
11354 return &bpf_get_socket_ptr_cookie_proto;
11355 case BPF_FUNC_ktime_get_coarse_ns:
11356 return &bpf_ktime_get_coarse_ns_proto;
11357 default:
11358 return bpf_base_func_proto(func_id, prog);
11359 }
11360 }
11361
11362 static bool
sk_reuseport_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)11363 sk_reuseport_is_valid_access(int off, int size,
11364 enum bpf_access_type type,
11365 const struct bpf_prog *prog,
11366 struct bpf_insn_access_aux *info)
11367 {
11368 const u32 size_default = sizeof(__u32);
11369
11370 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11371 off % size || type != BPF_READ)
11372 return false;
11373
11374 switch (off) {
11375 case offsetof(struct sk_reuseport_md, data):
11376 info->reg_type = PTR_TO_PACKET;
11377 return size == sizeof(__u64);
11378
11379 case offsetof(struct sk_reuseport_md, data_end):
11380 info->reg_type = PTR_TO_PACKET_END;
11381 return size == sizeof(__u64);
11382
11383 case offsetof(struct sk_reuseport_md, hash):
11384 return size == size_default;
11385
11386 case offsetof(struct sk_reuseport_md, sk):
11387 info->reg_type = PTR_TO_SOCKET;
11388 return size == sizeof(__u64);
11389
11390 case offsetof(struct sk_reuseport_md, migrating_sk):
11391 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11392 return size == sizeof(__u64);
11393
11394 /* Fields that allow narrowing */
11395 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11396 if (size < sizeof_field(struct sk_buff, protocol))
11397 return false;
11398 fallthrough;
11399 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11400 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11401 case bpf_ctx_range(struct sk_reuseport_md, len):
11402 bpf_ctx_record_field_size(info, size_default);
11403 return bpf_ctx_narrow_access_ok(off, size, size_default);
11404
11405 default:
11406 return false;
11407 }
11408 }
11409
11410 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
11411 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11412 si->dst_reg, si->src_reg, \
11413 bpf_target_off(struct sk_reuseport_kern, F, \
11414 sizeof_field(struct sk_reuseport_kern, F), \
11415 target_size)); \
11416 })
11417
11418 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
11419 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11420 struct sk_buff, \
11421 skb, \
11422 SKB_FIELD)
11423
11424 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
11425 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11426 struct sock, \
11427 sk, \
11428 SK_FIELD)
11429
sk_reuseport_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)11430 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11431 const struct bpf_insn *si,
11432 struct bpf_insn *insn_buf,
11433 struct bpf_prog *prog,
11434 u32 *target_size)
11435 {
11436 struct bpf_insn *insn = insn_buf;
11437
11438 switch (si->off) {
11439 case offsetof(struct sk_reuseport_md, data):
11440 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11441 break;
11442
11443 case offsetof(struct sk_reuseport_md, len):
11444 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11445 break;
11446
11447 case offsetof(struct sk_reuseport_md, eth_protocol):
11448 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11449 break;
11450
11451 case offsetof(struct sk_reuseport_md, ip_protocol):
11452 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11453 break;
11454
11455 case offsetof(struct sk_reuseport_md, data_end):
11456 SK_REUSEPORT_LOAD_FIELD(data_end);
11457 break;
11458
11459 case offsetof(struct sk_reuseport_md, hash):
11460 SK_REUSEPORT_LOAD_FIELD(hash);
11461 break;
11462
11463 case offsetof(struct sk_reuseport_md, bind_inany):
11464 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11465 break;
11466
11467 case offsetof(struct sk_reuseport_md, sk):
11468 SK_REUSEPORT_LOAD_FIELD(sk);
11469 break;
11470
11471 case offsetof(struct sk_reuseport_md, migrating_sk):
11472 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11473 break;
11474 }
11475
11476 return insn - insn_buf;
11477 }
11478
11479 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11480 .get_func_proto = sk_reuseport_func_proto,
11481 .is_valid_access = sk_reuseport_is_valid_access,
11482 .convert_ctx_access = sk_reuseport_convert_ctx_access,
11483 };
11484
11485 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11486 };
11487
11488 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11489 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11490
BPF_CALL_3(bpf_sk_lookup_assign,struct bpf_sk_lookup_kern *,ctx,struct sock *,sk,u64,flags)11491 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11492 struct sock *, sk, u64, flags)
11493 {
11494 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11495 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11496 return -EINVAL;
11497 if (unlikely(sk && sk_is_refcounted(sk)))
11498 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11499 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11500 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11501 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11502 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11503
11504 /* Check if socket is suitable for packet L3/L4 protocol */
11505 if (sk && sk->sk_protocol != ctx->protocol)
11506 return -EPROTOTYPE;
11507 if (sk && sk->sk_family != ctx->family &&
11508 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11509 return -EAFNOSUPPORT;
11510
11511 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11512 return -EEXIST;
11513
11514 /* Select socket as lookup result */
11515 ctx->selected_sk = sk;
11516 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11517 return 0;
11518 }
11519
11520 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11521 .func = bpf_sk_lookup_assign,
11522 .gpl_only = false,
11523 .ret_type = RET_INTEGER,
11524 .arg1_type = ARG_PTR_TO_CTX,
11525 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
11526 .arg3_type = ARG_ANYTHING,
11527 };
11528
11529 static const struct bpf_func_proto *
sk_lookup_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)11530 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11531 {
11532 switch (func_id) {
11533 case BPF_FUNC_perf_event_output:
11534 return &bpf_event_output_data_proto;
11535 case BPF_FUNC_sk_assign:
11536 return &bpf_sk_lookup_assign_proto;
11537 case BPF_FUNC_sk_release:
11538 return &bpf_sk_release_proto;
11539 default:
11540 return bpf_sk_base_func_proto(func_id, prog);
11541 }
11542 }
11543
sk_lookup_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)11544 static bool sk_lookup_is_valid_access(int off, int size,
11545 enum bpf_access_type type,
11546 const struct bpf_prog *prog,
11547 struct bpf_insn_access_aux *info)
11548 {
11549 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11550 return false;
11551 if (off % size != 0)
11552 return false;
11553 if (type != BPF_READ)
11554 return false;
11555
11556 switch (off) {
11557 case offsetof(struct bpf_sk_lookup, sk):
11558 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11559 return size == sizeof(__u64);
11560
11561 case bpf_ctx_range(struct bpf_sk_lookup, family):
11562 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11563 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11564 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11565 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11566 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11567 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11568 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11569 bpf_ctx_record_field_size(info, sizeof(__u32));
11570 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11571
11572 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11573 /* Allow 4-byte access to 2-byte field for backward compatibility */
11574 if (size == sizeof(__u32))
11575 return true;
11576 bpf_ctx_record_field_size(info, sizeof(__be16));
11577 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11578
11579 case offsetofend(struct bpf_sk_lookup, remote_port) ...
11580 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11581 /* Allow access to zero padding for backward compatibility */
11582 bpf_ctx_record_field_size(info, sizeof(__u16));
11583 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11584
11585 default:
11586 return false;
11587 }
11588 }
11589
sk_lookup_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)11590 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11591 const struct bpf_insn *si,
11592 struct bpf_insn *insn_buf,
11593 struct bpf_prog *prog,
11594 u32 *target_size)
11595 {
11596 struct bpf_insn *insn = insn_buf;
11597
11598 switch (si->off) {
11599 case offsetof(struct bpf_sk_lookup, sk):
11600 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11601 offsetof(struct bpf_sk_lookup_kern, selected_sk));
11602 break;
11603
11604 case offsetof(struct bpf_sk_lookup, family):
11605 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11606 bpf_target_off(struct bpf_sk_lookup_kern,
11607 family, 2, target_size));
11608 break;
11609
11610 case offsetof(struct bpf_sk_lookup, protocol):
11611 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11612 bpf_target_off(struct bpf_sk_lookup_kern,
11613 protocol, 2, target_size));
11614 break;
11615
11616 case offsetof(struct bpf_sk_lookup, remote_ip4):
11617 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11618 bpf_target_off(struct bpf_sk_lookup_kern,
11619 v4.saddr, 4, target_size));
11620 break;
11621
11622 case offsetof(struct bpf_sk_lookup, local_ip4):
11623 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11624 bpf_target_off(struct bpf_sk_lookup_kern,
11625 v4.daddr, 4, target_size));
11626 break;
11627
11628 case bpf_ctx_range_till(struct bpf_sk_lookup,
11629 remote_ip6[0], remote_ip6[3]): {
11630 #if IS_ENABLED(CONFIG_IPV6)
11631 int off = si->off;
11632
11633 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11634 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11635 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11636 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11637 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11638 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11639 #else
11640 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11641 #endif
11642 break;
11643 }
11644 case bpf_ctx_range_till(struct bpf_sk_lookup,
11645 local_ip6[0], local_ip6[3]): {
11646 #if IS_ENABLED(CONFIG_IPV6)
11647 int off = si->off;
11648
11649 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11650 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11651 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11652 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11653 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11654 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11655 #else
11656 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11657 #endif
11658 break;
11659 }
11660 case offsetof(struct bpf_sk_lookup, remote_port):
11661 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11662 bpf_target_off(struct bpf_sk_lookup_kern,
11663 sport, 2, target_size));
11664 break;
11665
11666 case offsetofend(struct bpf_sk_lookup, remote_port):
11667 *target_size = 2;
11668 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11669 break;
11670
11671 case offsetof(struct bpf_sk_lookup, local_port):
11672 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11673 bpf_target_off(struct bpf_sk_lookup_kern,
11674 dport, 2, target_size));
11675 break;
11676
11677 case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11678 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11679 bpf_target_off(struct bpf_sk_lookup_kern,
11680 ingress_ifindex, 4, target_size));
11681 break;
11682 }
11683
11684 return insn - insn_buf;
11685 }
11686
11687 const struct bpf_prog_ops sk_lookup_prog_ops = {
11688 .test_run = bpf_prog_test_run_sk_lookup,
11689 };
11690
11691 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11692 .get_func_proto = sk_lookup_func_proto,
11693 .is_valid_access = sk_lookup_is_valid_access,
11694 .convert_ctx_access = sk_lookup_convert_ctx_access,
11695 };
11696
11697 #endif /* CONFIG_INET */
11698
DEFINE_BPF_DISPATCHER(xdp)11699 DEFINE_BPF_DISPATCHER(xdp)
11700
11701 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11702 {
11703 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11704 }
11705
BTF_ID_LIST_GLOBAL(btf_sock_ids,MAX_BTF_SOCK_TYPE)11706 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11707 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11708 BTF_SOCK_TYPE_xxx
11709 #undef BTF_SOCK_TYPE
11710
11711 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11712 {
11713 /* tcp6_sock type is not generated in dwarf and hence btf,
11714 * trigger an explicit type generation here.
11715 */
11716 BTF_TYPE_EMIT(struct tcp6_sock);
11717 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11718 sk->sk_family == AF_INET6)
11719 return (unsigned long)sk;
11720
11721 return (unsigned long)NULL;
11722 }
11723
11724 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11725 .func = bpf_skc_to_tcp6_sock,
11726 .gpl_only = false,
11727 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11728 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11729 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11730 };
11731
BPF_CALL_1(bpf_skc_to_tcp_sock,struct sock *,sk)11732 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11733 {
11734 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11735 return (unsigned long)sk;
11736
11737 return (unsigned long)NULL;
11738 }
11739
11740 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11741 .func = bpf_skc_to_tcp_sock,
11742 .gpl_only = false,
11743 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11744 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11745 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11746 };
11747
BPF_CALL_1(bpf_skc_to_tcp_timewait_sock,struct sock *,sk)11748 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11749 {
11750 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11751 * generated if CONFIG_INET=n. Trigger an explicit generation here.
11752 */
11753 BTF_TYPE_EMIT(struct inet_timewait_sock);
11754 BTF_TYPE_EMIT(struct tcp_timewait_sock);
11755
11756 #ifdef CONFIG_INET
11757 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11758 return (unsigned long)sk;
11759 #endif
11760
11761 #if IS_BUILTIN(CONFIG_IPV6)
11762 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11763 return (unsigned long)sk;
11764 #endif
11765
11766 return (unsigned long)NULL;
11767 }
11768
11769 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11770 .func = bpf_skc_to_tcp_timewait_sock,
11771 .gpl_only = false,
11772 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11773 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11774 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11775 };
11776
BPF_CALL_1(bpf_skc_to_tcp_request_sock,struct sock *,sk)11777 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11778 {
11779 #ifdef CONFIG_INET
11780 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11781 return (unsigned long)sk;
11782 #endif
11783
11784 #if IS_BUILTIN(CONFIG_IPV6)
11785 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11786 return (unsigned long)sk;
11787 #endif
11788
11789 return (unsigned long)NULL;
11790 }
11791
11792 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11793 .func = bpf_skc_to_tcp_request_sock,
11794 .gpl_only = false,
11795 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11796 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11797 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11798 };
11799
BPF_CALL_1(bpf_skc_to_udp6_sock,struct sock *,sk)11800 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11801 {
11802 /* udp6_sock type is not generated in dwarf and hence btf,
11803 * trigger an explicit type generation here.
11804 */
11805 BTF_TYPE_EMIT(struct udp6_sock);
11806 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11807 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11808 return (unsigned long)sk;
11809
11810 return (unsigned long)NULL;
11811 }
11812
11813 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11814 .func = bpf_skc_to_udp6_sock,
11815 .gpl_only = false,
11816 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11817 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11818 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11819 };
11820
BPF_CALL_1(bpf_skc_to_unix_sock,struct sock *,sk)11821 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11822 {
11823 /* unix_sock type is not generated in dwarf and hence btf,
11824 * trigger an explicit type generation here.
11825 */
11826 BTF_TYPE_EMIT(struct unix_sock);
11827 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11828 return (unsigned long)sk;
11829
11830 return (unsigned long)NULL;
11831 }
11832
11833 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11834 .func = bpf_skc_to_unix_sock,
11835 .gpl_only = false,
11836 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11837 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11838 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11839 };
11840
BPF_CALL_1(bpf_skc_to_mptcp_sock,struct sock *,sk)11841 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11842 {
11843 BTF_TYPE_EMIT(struct mptcp_sock);
11844 return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11845 }
11846
11847 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11848 .func = bpf_skc_to_mptcp_sock,
11849 .gpl_only = false,
11850 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11851 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
11852 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11853 };
11854
BPF_CALL_1(bpf_sock_from_file,struct file *,file)11855 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11856 {
11857 return (unsigned long)sock_from_file(file);
11858 }
11859
11860 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11861 BTF_ID(struct, socket)
11862 BTF_ID(struct, file)
11863
11864 const struct bpf_func_proto bpf_sock_from_file_proto = {
11865 .func = bpf_sock_from_file,
11866 .gpl_only = false,
11867 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11868 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
11869 .arg1_type = ARG_PTR_TO_BTF_ID,
11870 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
11871 };
11872
11873 static const struct bpf_func_proto *
bpf_sk_base_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)11874 bpf_sk_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11875 {
11876 const struct bpf_func_proto *func;
11877
11878 switch (func_id) {
11879 case BPF_FUNC_skc_to_tcp6_sock:
11880 func = &bpf_skc_to_tcp6_sock_proto;
11881 break;
11882 case BPF_FUNC_skc_to_tcp_sock:
11883 func = &bpf_skc_to_tcp_sock_proto;
11884 break;
11885 case BPF_FUNC_skc_to_tcp_timewait_sock:
11886 func = &bpf_skc_to_tcp_timewait_sock_proto;
11887 break;
11888 case BPF_FUNC_skc_to_tcp_request_sock:
11889 func = &bpf_skc_to_tcp_request_sock_proto;
11890 break;
11891 case BPF_FUNC_skc_to_udp6_sock:
11892 func = &bpf_skc_to_udp6_sock_proto;
11893 break;
11894 case BPF_FUNC_skc_to_unix_sock:
11895 func = &bpf_skc_to_unix_sock_proto;
11896 break;
11897 case BPF_FUNC_skc_to_mptcp_sock:
11898 func = &bpf_skc_to_mptcp_sock_proto;
11899 break;
11900 case BPF_FUNC_ktime_get_coarse_ns:
11901 return &bpf_ktime_get_coarse_ns_proto;
11902 default:
11903 return bpf_base_func_proto(func_id, prog);
11904 }
11905
11906 if (!bpf_token_capable(prog->aux->token, CAP_PERFMON))
11907 return NULL;
11908
11909 return func;
11910 }
11911
11912 __bpf_kfunc_start_defs();
bpf_dynptr_from_skb(struct __sk_buff * s,u64 flags,struct bpf_dynptr * ptr__uninit)11913 __bpf_kfunc int bpf_dynptr_from_skb(struct __sk_buff *s, u64 flags,
11914 struct bpf_dynptr *ptr__uninit)
11915 {
11916 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)ptr__uninit;
11917 struct sk_buff *skb = (struct sk_buff *)s;
11918
11919 if (flags) {
11920 bpf_dynptr_set_null(ptr);
11921 return -EINVAL;
11922 }
11923
11924 bpf_dynptr_init(ptr, skb, BPF_DYNPTR_TYPE_SKB, 0, skb->len);
11925
11926 return 0;
11927 }
11928
bpf_dynptr_from_xdp(struct xdp_md * x,u64 flags,struct bpf_dynptr * ptr__uninit)11929 __bpf_kfunc int bpf_dynptr_from_xdp(struct xdp_md *x, u64 flags,
11930 struct bpf_dynptr *ptr__uninit)
11931 {
11932 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)ptr__uninit;
11933 struct xdp_buff *xdp = (struct xdp_buff *)x;
11934
11935 if (flags) {
11936 bpf_dynptr_set_null(ptr);
11937 return -EINVAL;
11938 }
11939
11940 bpf_dynptr_init(ptr, xdp, BPF_DYNPTR_TYPE_XDP, 0, xdp_get_buff_len(xdp));
11941
11942 return 0;
11943 }
11944
bpf_sock_addr_set_sun_path(struct bpf_sock_addr_kern * sa_kern,const u8 * sun_path,u32 sun_path__sz)11945 __bpf_kfunc int bpf_sock_addr_set_sun_path(struct bpf_sock_addr_kern *sa_kern,
11946 const u8 *sun_path, u32 sun_path__sz)
11947 {
11948 struct sockaddr_un *un;
11949
11950 if (sa_kern->sk->sk_family != AF_UNIX)
11951 return -EINVAL;
11952
11953 /* We do not allow changing the address to unnamed or larger than the
11954 * maximum allowed address size for a unix sockaddr.
11955 */
11956 if (sun_path__sz == 0 || sun_path__sz > UNIX_PATH_MAX)
11957 return -EINVAL;
11958
11959 un = (struct sockaddr_un *)sa_kern->uaddr;
11960 memcpy(un->sun_path, sun_path, sun_path__sz);
11961 sa_kern->uaddrlen = offsetof(struct sockaddr_un, sun_path) + sun_path__sz;
11962
11963 return 0;
11964 }
11965
bpf_sk_assign_tcp_reqsk(struct __sk_buff * s,struct sock * sk,struct bpf_tcp_req_attrs * attrs,int attrs__sz)11966 __bpf_kfunc int bpf_sk_assign_tcp_reqsk(struct __sk_buff *s, struct sock *sk,
11967 struct bpf_tcp_req_attrs *attrs, int attrs__sz)
11968 {
11969 #if IS_ENABLED(CONFIG_SYN_COOKIES)
11970 struct sk_buff *skb = (struct sk_buff *)s;
11971 const struct request_sock_ops *ops;
11972 struct inet_request_sock *ireq;
11973 struct tcp_request_sock *treq;
11974 struct request_sock *req;
11975 struct net *net;
11976 __u16 min_mss;
11977 u32 tsoff = 0;
11978
11979 if (attrs__sz != sizeof(*attrs) ||
11980 attrs->reserved[0] || attrs->reserved[1] || attrs->reserved[2])
11981 return -EINVAL;
11982
11983 if (!skb_at_tc_ingress(skb))
11984 return -EINVAL;
11985
11986 net = dev_net(skb->dev);
11987 if (net != sock_net(sk))
11988 return -ENETUNREACH;
11989
11990 switch (skb->protocol) {
11991 case htons(ETH_P_IP):
11992 ops = &tcp_request_sock_ops;
11993 min_mss = 536;
11994 break;
11995 #if IS_BUILTIN(CONFIG_IPV6)
11996 case htons(ETH_P_IPV6):
11997 ops = &tcp6_request_sock_ops;
11998 min_mss = IPV6_MIN_MTU - 60;
11999 break;
12000 #endif
12001 default:
12002 return -EINVAL;
12003 }
12004
12005 if (sk->sk_type != SOCK_STREAM || sk->sk_state != TCP_LISTEN ||
12006 sk_is_mptcp(sk))
12007 return -EINVAL;
12008
12009 if (attrs->mss < min_mss)
12010 return -EINVAL;
12011
12012 if (attrs->wscale_ok) {
12013 if (!READ_ONCE(net->ipv4.sysctl_tcp_window_scaling))
12014 return -EINVAL;
12015
12016 if (attrs->snd_wscale > TCP_MAX_WSCALE ||
12017 attrs->rcv_wscale > TCP_MAX_WSCALE)
12018 return -EINVAL;
12019 }
12020
12021 if (attrs->sack_ok && !READ_ONCE(net->ipv4.sysctl_tcp_sack))
12022 return -EINVAL;
12023
12024 if (attrs->tstamp_ok) {
12025 if (!READ_ONCE(net->ipv4.sysctl_tcp_timestamps))
12026 return -EINVAL;
12027
12028 tsoff = attrs->rcv_tsecr - tcp_ns_to_ts(attrs->usec_ts_ok, tcp_clock_ns());
12029 }
12030
12031 req = inet_reqsk_alloc(ops, sk, false);
12032 if (!req)
12033 return -ENOMEM;
12034
12035 ireq = inet_rsk(req);
12036 treq = tcp_rsk(req);
12037
12038 req->rsk_listener = sk;
12039 req->syncookie = 1;
12040 req->mss = attrs->mss;
12041 req->ts_recent = attrs->rcv_tsval;
12042
12043 ireq->snd_wscale = attrs->snd_wscale;
12044 ireq->rcv_wscale = attrs->rcv_wscale;
12045 ireq->tstamp_ok = !!attrs->tstamp_ok;
12046 ireq->sack_ok = !!attrs->sack_ok;
12047 ireq->wscale_ok = !!attrs->wscale_ok;
12048 ireq->ecn_ok = !!attrs->ecn_ok;
12049
12050 treq->req_usec_ts = !!attrs->usec_ts_ok;
12051 treq->ts_off = tsoff;
12052
12053 skb_orphan(skb);
12054 skb->sk = req_to_sk(req);
12055 skb->destructor = sock_pfree;
12056
12057 return 0;
12058 #else
12059 return -EOPNOTSUPP;
12060 #endif
12061 }
12062
12063 __bpf_kfunc_end_defs();
12064
bpf_dynptr_from_skb_rdonly(struct __sk_buff * skb,u64 flags,struct bpf_dynptr * ptr__uninit)12065 int bpf_dynptr_from_skb_rdonly(struct __sk_buff *skb, u64 flags,
12066 struct bpf_dynptr *ptr__uninit)
12067 {
12068 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)ptr__uninit;
12069 int err;
12070
12071 err = bpf_dynptr_from_skb(skb, flags, ptr__uninit);
12072 if (err)
12073 return err;
12074
12075 bpf_dynptr_set_rdonly(ptr);
12076
12077 return 0;
12078 }
12079
12080 BTF_KFUNCS_START(bpf_kfunc_check_set_skb)
12081 BTF_ID_FLAGS(func, bpf_dynptr_from_skb, KF_TRUSTED_ARGS)
12082 BTF_KFUNCS_END(bpf_kfunc_check_set_skb)
12083
12084 BTF_KFUNCS_START(bpf_kfunc_check_set_xdp)
12085 BTF_ID_FLAGS(func, bpf_dynptr_from_xdp)
12086 BTF_KFUNCS_END(bpf_kfunc_check_set_xdp)
12087
12088 BTF_KFUNCS_START(bpf_kfunc_check_set_sock_addr)
12089 BTF_ID_FLAGS(func, bpf_sock_addr_set_sun_path)
12090 BTF_KFUNCS_END(bpf_kfunc_check_set_sock_addr)
12091
12092 BTF_KFUNCS_START(bpf_kfunc_check_set_tcp_reqsk)
12093 BTF_ID_FLAGS(func, bpf_sk_assign_tcp_reqsk, KF_TRUSTED_ARGS)
12094 BTF_KFUNCS_END(bpf_kfunc_check_set_tcp_reqsk)
12095
12096 static const struct btf_kfunc_id_set bpf_kfunc_set_skb = {
12097 .owner = THIS_MODULE,
12098 .set = &bpf_kfunc_check_set_skb,
12099 };
12100
12101 static const struct btf_kfunc_id_set bpf_kfunc_set_xdp = {
12102 .owner = THIS_MODULE,
12103 .set = &bpf_kfunc_check_set_xdp,
12104 };
12105
12106 static const struct btf_kfunc_id_set bpf_kfunc_set_sock_addr = {
12107 .owner = THIS_MODULE,
12108 .set = &bpf_kfunc_check_set_sock_addr,
12109 };
12110
12111 static const struct btf_kfunc_id_set bpf_kfunc_set_tcp_reqsk = {
12112 .owner = THIS_MODULE,
12113 .set = &bpf_kfunc_check_set_tcp_reqsk,
12114 };
12115
bpf_kfunc_init(void)12116 static int __init bpf_kfunc_init(void)
12117 {
12118 int ret;
12119
12120 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_skb);
12121 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_ACT, &bpf_kfunc_set_skb);
12122 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SK_SKB, &bpf_kfunc_set_skb);
12123 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SOCKET_FILTER, &bpf_kfunc_set_skb);
12124 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &bpf_kfunc_set_skb);
12125 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_OUT, &bpf_kfunc_set_skb);
12126 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_IN, &bpf_kfunc_set_skb);
12127 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_XMIT, &bpf_kfunc_set_skb);
12128 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_SEG6LOCAL, &bpf_kfunc_set_skb);
12129 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_NETFILTER, &bpf_kfunc_set_skb);
12130 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_kfunc_set_skb);
12131 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &bpf_kfunc_set_xdp);
12132 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
12133 &bpf_kfunc_set_sock_addr);
12134 return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_tcp_reqsk);
12135 }
12136 late_initcall(bpf_kfunc_init);
12137
12138 __bpf_kfunc_start_defs();
12139
12140 /* bpf_sock_destroy: Destroy the given socket with ECONNABORTED error code.
12141 *
12142 * The function expects a non-NULL pointer to a socket, and invokes the
12143 * protocol specific socket destroy handlers.
12144 *
12145 * The helper can only be called from BPF contexts that have acquired the socket
12146 * locks.
12147 *
12148 * Parameters:
12149 * @sock: Pointer to socket to be destroyed
12150 *
12151 * Return:
12152 * On error, may return EPROTONOSUPPORT, EINVAL.
12153 * EPROTONOSUPPORT if protocol specific destroy handler is not supported.
12154 * 0 otherwise
12155 */
bpf_sock_destroy(struct sock_common * sock)12156 __bpf_kfunc int bpf_sock_destroy(struct sock_common *sock)
12157 {
12158 struct sock *sk = (struct sock *)sock;
12159
12160 /* The locking semantics that allow for synchronous execution of the
12161 * destroy handlers are only supported for TCP and UDP.
12162 * Supporting protocols will need to acquire sock lock in the BPF context
12163 * prior to invoking this kfunc.
12164 */
12165 if (!sk->sk_prot->diag_destroy || (sk->sk_protocol != IPPROTO_TCP &&
12166 sk->sk_protocol != IPPROTO_UDP))
12167 return -EOPNOTSUPP;
12168
12169 return sk->sk_prot->diag_destroy(sk, ECONNABORTED);
12170 }
12171
12172 __bpf_kfunc_end_defs();
12173
12174 BTF_KFUNCS_START(bpf_sk_iter_kfunc_ids)
BTF_ID_FLAGS(func,bpf_sock_destroy,KF_TRUSTED_ARGS)12175 BTF_ID_FLAGS(func, bpf_sock_destroy, KF_TRUSTED_ARGS)
12176 BTF_KFUNCS_END(bpf_sk_iter_kfunc_ids)
12177
12178 static int tracing_iter_filter(const struct bpf_prog *prog, u32 kfunc_id)
12179 {
12180 if (btf_id_set8_contains(&bpf_sk_iter_kfunc_ids, kfunc_id) &&
12181 prog->expected_attach_type != BPF_TRACE_ITER)
12182 return -EACCES;
12183 return 0;
12184 }
12185
12186 static const struct btf_kfunc_id_set bpf_sk_iter_kfunc_set = {
12187 .owner = THIS_MODULE,
12188 .set = &bpf_sk_iter_kfunc_ids,
12189 .filter = tracing_iter_filter,
12190 };
12191
init_subsystem(void)12192 static int init_subsystem(void)
12193 {
12194 return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_sk_iter_kfunc_set);
12195 }
12196 late_initcall(init_subsystem);
12197