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