1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 * Copyright (c) 2016,2017 Facebook
4 */
5 #include <linux/bpf.h>
6 #include <linux/btf.h>
7 #include <linux/err.h>
8 #include <linux/slab.h>
9 #include <linux/mm.h>
10 #include <linux/filter.h>
11 #include <linux/perf_event.h>
12 #include <uapi/linux/btf.h>
13 #include <linux/rcupdate_trace.h>
14 #include <linux/btf_ids.h>
15
16 #include "map_in_map.h"
17
18 #define ARRAY_CREATE_FLAG_MASK \
19 (BPF_F_NUMA_NODE | BPF_F_MMAPABLE | BPF_F_ACCESS_MASK | \
20 BPF_F_PRESERVE_ELEMS | BPF_F_INNER_MAP)
21
bpf_array_free_percpu(struct bpf_array * array)22 static void bpf_array_free_percpu(struct bpf_array *array)
23 {
24 int i;
25
26 for (i = 0; i < array->map.max_entries; i++) {
27 free_percpu(array->pptrs[i]);
28 cond_resched();
29 }
30 }
31
bpf_array_alloc_percpu(struct bpf_array * array)32 static int bpf_array_alloc_percpu(struct bpf_array *array)
33 {
34 void __percpu *ptr;
35 int i;
36
37 for (i = 0; i < array->map.max_entries; i++) {
38 ptr = bpf_map_alloc_percpu(&array->map, array->elem_size, 8,
39 GFP_USER | __GFP_NOWARN);
40 if (!ptr) {
41 bpf_array_free_percpu(array);
42 return -ENOMEM;
43 }
44 array->pptrs[i] = ptr;
45 cond_resched();
46 }
47
48 return 0;
49 }
50
51 /* Called from syscall */
array_map_alloc_check(union bpf_attr * attr)52 int array_map_alloc_check(union bpf_attr *attr)
53 {
54 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
55 int numa_node = bpf_map_attr_numa_node(attr);
56
57 /* check sanity of attributes */
58 if (attr->max_entries == 0 || attr->key_size != 4 ||
59 attr->value_size == 0 ||
60 attr->map_flags & ~ARRAY_CREATE_FLAG_MASK ||
61 !bpf_map_flags_access_ok(attr->map_flags) ||
62 (percpu && numa_node != NUMA_NO_NODE))
63 return -EINVAL;
64
65 if (attr->map_type != BPF_MAP_TYPE_ARRAY &&
66 attr->map_flags & (BPF_F_MMAPABLE | BPF_F_INNER_MAP))
67 return -EINVAL;
68
69 if (attr->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
70 attr->map_flags & BPF_F_PRESERVE_ELEMS)
71 return -EINVAL;
72
73 /* avoid overflow on round_up(map->value_size) */
74 if (attr->value_size > INT_MAX)
75 return -E2BIG;
76
77 return 0;
78 }
79
array_map_alloc(union bpf_attr * attr)80 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
81 {
82 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
83 int numa_node = bpf_map_attr_numa_node(attr);
84 u32 elem_size, index_mask, max_entries;
85 bool bypass_spec_v1 = bpf_bypass_spec_v1(NULL);
86 u64 array_size, mask64;
87 struct bpf_array *array;
88
89 elem_size = round_up(attr->value_size, 8);
90
91 max_entries = attr->max_entries;
92
93 /* On 32 bit archs roundup_pow_of_two() with max_entries that has
94 * upper most bit set in u32 space is undefined behavior due to
95 * resulting 1U << 32, so do it manually here in u64 space.
96 */
97 mask64 = fls_long(max_entries - 1);
98 mask64 = 1ULL << mask64;
99 mask64 -= 1;
100
101 index_mask = mask64;
102 if (!bypass_spec_v1) {
103 /* round up array size to nearest power of 2,
104 * since cpu will speculate within index_mask limits
105 */
106 max_entries = index_mask + 1;
107 /* Check for overflows. */
108 if (max_entries < attr->max_entries)
109 return ERR_PTR(-E2BIG);
110 }
111
112 array_size = sizeof(*array);
113 if (percpu) {
114 array_size += (u64) max_entries * sizeof(void *);
115 } else {
116 /* rely on vmalloc() to return page-aligned memory and
117 * ensure array->value is exactly page-aligned
118 */
119 if (attr->map_flags & BPF_F_MMAPABLE) {
120 array_size = PAGE_ALIGN(array_size);
121 array_size += PAGE_ALIGN((u64) max_entries * elem_size);
122 } else {
123 array_size += (u64) max_entries * elem_size;
124 }
125 }
126
127 /* allocate all map elements and zero-initialize them */
128 if (attr->map_flags & BPF_F_MMAPABLE) {
129 void *data;
130
131 /* kmalloc'ed memory can't be mmap'ed, use explicit vmalloc */
132 data = bpf_map_area_mmapable_alloc(array_size, numa_node);
133 if (!data)
134 return ERR_PTR(-ENOMEM);
135 array = data + PAGE_ALIGN(sizeof(struct bpf_array))
136 - offsetof(struct bpf_array, value);
137 } else {
138 array = bpf_map_area_alloc(array_size, numa_node);
139 }
140 if (!array)
141 return ERR_PTR(-ENOMEM);
142 array->index_mask = index_mask;
143 array->map.bypass_spec_v1 = bypass_spec_v1;
144
145 /* copy mandatory map attributes */
146 bpf_map_init_from_attr(&array->map, attr);
147 array->elem_size = elem_size;
148
149 if (percpu && bpf_array_alloc_percpu(array)) {
150 bpf_map_area_free(array);
151 return ERR_PTR(-ENOMEM);
152 }
153
154 return &array->map;
155 }
156
array_map_elem_ptr(struct bpf_array * array,u32 index)157 static void *array_map_elem_ptr(struct bpf_array* array, u32 index)
158 {
159 return array->value + (u64)array->elem_size * index;
160 }
161
162 /* Called from syscall or from eBPF program */
array_map_lookup_elem(struct bpf_map * map,void * key)163 static void *array_map_lookup_elem(struct bpf_map *map, void *key)
164 {
165 struct bpf_array *array = container_of(map, struct bpf_array, map);
166 u32 index = *(u32 *)key;
167
168 if (unlikely(index >= array->map.max_entries))
169 return NULL;
170
171 return array->value + (u64)array->elem_size * (index & array->index_mask);
172 }
173
array_map_direct_value_addr(const struct bpf_map * map,u64 * imm,u32 off)174 static int array_map_direct_value_addr(const struct bpf_map *map, u64 *imm,
175 u32 off)
176 {
177 struct bpf_array *array = container_of(map, struct bpf_array, map);
178
179 if (map->max_entries != 1)
180 return -ENOTSUPP;
181 if (off >= map->value_size)
182 return -EINVAL;
183
184 *imm = (unsigned long)array->value;
185 return 0;
186 }
187
array_map_direct_value_meta(const struct bpf_map * map,u64 imm,u32 * off)188 static int array_map_direct_value_meta(const struct bpf_map *map, u64 imm,
189 u32 *off)
190 {
191 struct bpf_array *array = container_of(map, struct bpf_array, map);
192 u64 base = (unsigned long)array->value;
193 u64 range = array->elem_size;
194
195 if (map->max_entries != 1)
196 return -ENOTSUPP;
197 if (imm < base || imm >= base + range)
198 return -ENOENT;
199
200 *off = imm - base;
201 return 0;
202 }
203
204 /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
array_map_gen_lookup(struct bpf_map * map,struct bpf_insn * insn_buf)205 static int array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
206 {
207 struct bpf_array *array = container_of(map, struct bpf_array, map);
208 struct bpf_insn *insn = insn_buf;
209 u32 elem_size = array->elem_size;
210 const int ret = BPF_REG_0;
211 const int map_ptr = BPF_REG_1;
212 const int index = BPF_REG_2;
213
214 if (map->map_flags & BPF_F_INNER_MAP)
215 return -EOPNOTSUPP;
216
217 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
218 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
219 if (!map->bypass_spec_v1) {
220 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4);
221 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
222 } else {
223 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
224 }
225
226 if (is_power_of_2(elem_size)) {
227 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
228 } else {
229 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
230 }
231 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
232 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
233 *insn++ = BPF_MOV64_IMM(ret, 0);
234 return insn - insn_buf;
235 }
236
237 /* Called from eBPF program */
percpu_array_map_lookup_elem(struct bpf_map * map,void * key)238 static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
239 {
240 struct bpf_array *array = container_of(map, struct bpf_array, map);
241 u32 index = *(u32 *)key;
242
243 if (unlikely(index >= array->map.max_entries))
244 return NULL;
245
246 return this_cpu_ptr(array->pptrs[index & array->index_mask]);
247 }
248
249 /* emit BPF instructions equivalent to C code of percpu_array_map_lookup_elem() */
percpu_array_map_gen_lookup(struct bpf_map * map,struct bpf_insn * insn_buf)250 static int percpu_array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
251 {
252 struct bpf_array *array = container_of(map, struct bpf_array, map);
253 struct bpf_insn *insn = insn_buf;
254
255 if (!bpf_jit_supports_percpu_insn())
256 return -EOPNOTSUPP;
257
258 if (map->map_flags & BPF_F_INNER_MAP)
259 return -EOPNOTSUPP;
260
261 BUILD_BUG_ON(offsetof(struct bpf_array, map) != 0);
262 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, offsetof(struct bpf_array, pptrs));
263
264 *insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_0, BPF_REG_2, 0);
265 if (!map->bypass_spec_v1) {
266 *insn++ = BPF_JMP_IMM(BPF_JGE, BPF_REG_0, map->max_entries, 6);
267 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_0, array->index_mask);
268 } else {
269 *insn++ = BPF_JMP_IMM(BPF_JGE, BPF_REG_0, map->max_entries, 5);
270 }
271
272 *insn++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_0, 3);
273 *insn++ = BPF_ALU64_REG(BPF_ADD, BPF_REG_0, BPF_REG_1);
274 *insn++ = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_0, 0);
275 *insn++ = BPF_MOV64_PERCPU_REG(BPF_REG_0, BPF_REG_0);
276 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
277 *insn++ = BPF_MOV64_IMM(BPF_REG_0, 0);
278 return insn - insn_buf;
279 }
280
percpu_array_map_lookup_percpu_elem(struct bpf_map * map,void * key,u32 cpu)281 static void *percpu_array_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
282 {
283 struct bpf_array *array = container_of(map, struct bpf_array, map);
284 u32 index = *(u32 *)key;
285
286 if (cpu >= nr_cpu_ids)
287 return NULL;
288
289 if (unlikely(index >= array->map.max_entries))
290 return NULL;
291
292 return per_cpu_ptr(array->pptrs[index & array->index_mask], cpu);
293 }
294
bpf_percpu_array_copy(struct bpf_map * map,void * key,void * value)295 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
296 {
297 struct bpf_array *array = container_of(map, struct bpf_array, map);
298 u32 index = *(u32 *)key;
299 void __percpu *pptr;
300 int cpu, off = 0;
301 u32 size;
302
303 if (unlikely(index >= array->map.max_entries))
304 return -ENOENT;
305
306 /* per_cpu areas are zero-filled and bpf programs can only
307 * access 'value_size' of them, so copying rounded areas
308 * will not leak any kernel data
309 */
310 size = array->elem_size;
311 rcu_read_lock();
312 pptr = array->pptrs[index & array->index_mask];
313 for_each_possible_cpu(cpu) {
314 copy_map_value_long(map, value + off, per_cpu_ptr(pptr, cpu));
315 check_and_init_map_value(map, value + off);
316 off += size;
317 }
318 rcu_read_unlock();
319 return 0;
320 }
321
322 /* Called from syscall */
array_map_get_next_key(struct bpf_map * map,void * key,void * next_key)323 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
324 {
325 struct bpf_array *array = container_of(map, struct bpf_array, map);
326 u32 index = key ? *(u32 *)key : U32_MAX;
327 u32 *next = (u32 *)next_key;
328
329 if (index >= array->map.max_entries) {
330 *next = 0;
331 return 0;
332 }
333
334 if (index == array->map.max_entries - 1)
335 return -ENOENT;
336
337 *next = index + 1;
338 return 0;
339 }
340
341 /* Called from syscall or from eBPF program */
array_map_update_elem(struct bpf_map * map,void * key,void * value,u64 map_flags)342 static long array_map_update_elem(struct bpf_map *map, void *key, void *value,
343 u64 map_flags)
344 {
345 struct bpf_array *array = container_of(map, struct bpf_array, map);
346 u32 index = *(u32 *)key;
347 char *val;
348
349 if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
350 /* unknown flags */
351 return -EINVAL;
352
353 if (unlikely(index >= array->map.max_entries))
354 /* all elements were pre-allocated, cannot insert a new one */
355 return -E2BIG;
356
357 if (unlikely(map_flags & BPF_NOEXIST))
358 /* all elements already exist */
359 return -EEXIST;
360
361 if (unlikely((map_flags & BPF_F_LOCK) &&
362 !btf_record_has_field(map->record, BPF_SPIN_LOCK)))
363 return -EINVAL;
364
365 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
366 val = this_cpu_ptr(array->pptrs[index & array->index_mask]);
367 copy_map_value(map, val, value);
368 bpf_obj_free_fields(array->map.record, val);
369 } else {
370 val = array->value +
371 (u64)array->elem_size * (index & array->index_mask);
372 if (map_flags & BPF_F_LOCK)
373 copy_map_value_locked(map, val, value, false);
374 else
375 copy_map_value(map, val, value);
376 bpf_obj_free_fields(array->map.record, val);
377 }
378 return 0;
379 }
380
bpf_percpu_array_update(struct bpf_map * map,void * key,void * value,u64 map_flags)381 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
382 u64 map_flags)
383 {
384 struct bpf_array *array = container_of(map, struct bpf_array, map);
385 u32 index = *(u32 *)key;
386 void __percpu *pptr;
387 int cpu, off = 0;
388 u32 size;
389
390 if (unlikely(map_flags > BPF_EXIST))
391 /* unknown flags */
392 return -EINVAL;
393
394 if (unlikely(index >= array->map.max_entries))
395 /* all elements were pre-allocated, cannot insert a new one */
396 return -E2BIG;
397
398 if (unlikely(map_flags == BPF_NOEXIST))
399 /* all elements already exist */
400 return -EEXIST;
401
402 /* the user space will provide round_up(value_size, 8) bytes that
403 * will be copied into per-cpu area. bpf programs can only access
404 * value_size of it. During lookup the same extra bytes will be
405 * returned or zeros which were zero-filled by percpu_alloc,
406 * so no kernel data leaks possible
407 */
408 size = array->elem_size;
409 rcu_read_lock();
410 pptr = array->pptrs[index & array->index_mask];
411 for_each_possible_cpu(cpu) {
412 copy_map_value_long(map, per_cpu_ptr(pptr, cpu), value + off);
413 bpf_obj_free_fields(array->map.record, per_cpu_ptr(pptr, cpu));
414 off += size;
415 }
416 rcu_read_unlock();
417 return 0;
418 }
419
420 /* Called from syscall or from eBPF program */
array_map_delete_elem(struct bpf_map * map,void * key)421 static long array_map_delete_elem(struct bpf_map *map, void *key)
422 {
423 return -EINVAL;
424 }
425
array_map_vmalloc_addr(struct bpf_array * array)426 static void *array_map_vmalloc_addr(struct bpf_array *array)
427 {
428 return (void *)round_down((unsigned long)array, PAGE_SIZE);
429 }
430
array_map_free_timers_wq(struct bpf_map * map)431 static void array_map_free_timers_wq(struct bpf_map *map)
432 {
433 struct bpf_array *array = container_of(map, struct bpf_array, map);
434 int i;
435
436 /* We don't reset or free fields other than timer and workqueue
437 * on uref dropping to zero.
438 */
439 if (btf_record_has_field(map->record, BPF_TIMER | BPF_WORKQUEUE)) {
440 for (i = 0; i < array->map.max_entries; i++) {
441 if (btf_record_has_field(map->record, BPF_TIMER))
442 bpf_obj_free_timer(map->record, array_map_elem_ptr(array, i));
443 if (btf_record_has_field(map->record, BPF_WORKQUEUE))
444 bpf_obj_free_workqueue(map->record, array_map_elem_ptr(array, i));
445 }
446 }
447 }
448
449 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
array_map_free(struct bpf_map * map)450 static void array_map_free(struct bpf_map *map)
451 {
452 struct bpf_array *array = container_of(map, struct bpf_array, map);
453 int i;
454
455 if (!IS_ERR_OR_NULL(map->record)) {
456 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
457 for (i = 0; i < array->map.max_entries; i++) {
458 void __percpu *pptr = array->pptrs[i & array->index_mask];
459 int cpu;
460
461 for_each_possible_cpu(cpu) {
462 bpf_obj_free_fields(map->record, per_cpu_ptr(pptr, cpu));
463 cond_resched();
464 }
465 }
466 } else {
467 for (i = 0; i < array->map.max_entries; i++)
468 bpf_obj_free_fields(map->record, array_map_elem_ptr(array, i));
469 }
470 }
471
472 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
473 bpf_array_free_percpu(array);
474
475 if (array->map.map_flags & BPF_F_MMAPABLE)
476 bpf_map_area_free(array_map_vmalloc_addr(array));
477 else
478 bpf_map_area_free(array);
479 }
480
array_map_seq_show_elem(struct bpf_map * map,void * key,struct seq_file * m)481 static void array_map_seq_show_elem(struct bpf_map *map, void *key,
482 struct seq_file *m)
483 {
484 void *value;
485
486 rcu_read_lock();
487
488 value = array_map_lookup_elem(map, key);
489 if (!value) {
490 rcu_read_unlock();
491 return;
492 }
493
494 if (map->btf_key_type_id)
495 seq_printf(m, "%u: ", *(u32 *)key);
496 btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
497 seq_puts(m, "\n");
498
499 rcu_read_unlock();
500 }
501
percpu_array_map_seq_show_elem(struct bpf_map * map,void * key,struct seq_file * m)502 static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key,
503 struct seq_file *m)
504 {
505 struct bpf_array *array = container_of(map, struct bpf_array, map);
506 u32 index = *(u32 *)key;
507 void __percpu *pptr;
508 int cpu;
509
510 rcu_read_lock();
511
512 seq_printf(m, "%u: {\n", *(u32 *)key);
513 pptr = array->pptrs[index & array->index_mask];
514 for_each_possible_cpu(cpu) {
515 seq_printf(m, "\tcpu%d: ", cpu);
516 btf_type_seq_show(map->btf, map->btf_value_type_id,
517 per_cpu_ptr(pptr, cpu), m);
518 seq_puts(m, "\n");
519 }
520 seq_puts(m, "}\n");
521
522 rcu_read_unlock();
523 }
524
array_map_check_btf(const struct bpf_map * map,const struct btf * btf,const struct btf_type * key_type,const struct btf_type * value_type)525 static int array_map_check_btf(const struct bpf_map *map,
526 const struct btf *btf,
527 const struct btf_type *key_type,
528 const struct btf_type *value_type)
529 {
530 u32 int_data;
531
532 /* One exception for keyless BTF: .bss/.data/.rodata map */
533 if (btf_type_is_void(key_type)) {
534 if (map->map_type != BPF_MAP_TYPE_ARRAY ||
535 map->max_entries != 1)
536 return -EINVAL;
537
538 if (BTF_INFO_KIND(value_type->info) != BTF_KIND_DATASEC)
539 return -EINVAL;
540
541 return 0;
542 }
543
544 if (BTF_INFO_KIND(key_type->info) != BTF_KIND_INT)
545 return -EINVAL;
546
547 int_data = *(u32 *)(key_type + 1);
548 /* bpf array can only take a u32 key. This check makes sure
549 * that the btf matches the attr used during map_create.
550 */
551 if (BTF_INT_BITS(int_data) != 32 || BTF_INT_OFFSET(int_data))
552 return -EINVAL;
553
554 return 0;
555 }
556
array_map_mmap(struct bpf_map * map,struct vm_area_struct * vma)557 static int array_map_mmap(struct bpf_map *map, struct vm_area_struct *vma)
558 {
559 struct bpf_array *array = container_of(map, struct bpf_array, map);
560 pgoff_t pgoff = PAGE_ALIGN(sizeof(*array)) >> PAGE_SHIFT;
561
562 if (!(map->map_flags & BPF_F_MMAPABLE))
563 return -EINVAL;
564
565 if (vma->vm_pgoff * PAGE_SIZE + (vma->vm_end - vma->vm_start) >
566 PAGE_ALIGN((u64)array->map.max_entries * array->elem_size))
567 return -EINVAL;
568
569 return remap_vmalloc_range(vma, array_map_vmalloc_addr(array),
570 vma->vm_pgoff + pgoff);
571 }
572
array_map_meta_equal(const struct bpf_map * meta0,const struct bpf_map * meta1)573 static bool array_map_meta_equal(const struct bpf_map *meta0,
574 const struct bpf_map *meta1)
575 {
576 if (!bpf_map_meta_equal(meta0, meta1))
577 return false;
578 return meta0->map_flags & BPF_F_INNER_MAP ? true :
579 meta0->max_entries == meta1->max_entries;
580 }
581
582 struct bpf_iter_seq_array_map_info {
583 struct bpf_map *map;
584 void *percpu_value_buf;
585 u32 index;
586 };
587
bpf_array_map_seq_start(struct seq_file * seq,loff_t * pos)588 static void *bpf_array_map_seq_start(struct seq_file *seq, loff_t *pos)
589 {
590 struct bpf_iter_seq_array_map_info *info = seq->private;
591 struct bpf_map *map = info->map;
592 struct bpf_array *array;
593 u32 index;
594
595 if (info->index >= map->max_entries)
596 return NULL;
597
598 if (*pos == 0)
599 ++*pos;
600 array = container_of(map, struct bpf_array, map);
601 index = info->index & array->index_mask;
602 if (info->percpu_value_buf)
603 return array->pptrs[index];
604 return array_map_elem_ptr(array, index);
605 }
606
bpf_array_map_seq_next(struct seq_file * seq,void * v,loff_t * pos)607 static void *bpf_array_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
608 {
609 struct bpf_iter_seq_array_map_info *info = seq->private;
610 struct bpf_map *map = info->map;
611 struct bpf_array *array;
612 u32 index;
613
614 ++*pos;
615 ++info->index;
616 if (info->index >= map->max_entries)
617 return NULL;
618
619 array = container_of(map, struct bpf_array, map);
620 index = info->index & array->index_mask;
621 if (info->percpu_value_buf)
622 return array->pptrs[index];
623 return array_map_elem_ptr(array, index);
624 }
625
__bpf_array_map_seq_show(struct seq_file * seq,void * v)626 static int __bpf_array_map_seq_show(struct seq_file *seq, void *v)
627 {
628 struct bpf_iter_seq_array_map_info *info = seq->private;
629 struct bpf_iter__bpf_map_elem ctx = {};
630 struct bpf_map *map = info->map;
631 struct bpf_array *array = container_of(map, struct bpf_array, map);
632 struct bpf_iter_meta meta;
633 struct bpf_prog *prog;
634 int off = 0, cpu = 0;
635 void __percpu **pptr;
636 u32 size;
637
638 meta.seq = seq;
639 prog = bpf_iter_get_info(&meta, v == NULL);
640 if (!prog)
641 return 0;
642
643 ctx.meta = &meta;
644 ctx.map = info->map;
645 if (v) {
646 ctx.key = &info->index;
647
648 if (!info->percpu_value_buf) {
649 ctx.value = v;
650 } else {
651 pptr = v;
652 size = array->elem_size;
653 for_each_possible_cpu(cpu) {
654 copy_map_value_long(map, info->percpu_value_buf + off,
655 per_cpu_ptr(pptr, cpu));
656 check_and_init_map_value(map, info->percpu_value_buf + off);
657 off += size;
658 }
659 ctx.value = info->percpu_value_buf;
660 }
661 }
662
663 return bpf_iter_run_prog(prog, &ctx);
664 }
665
bpf_array_map_seq_show(struct seq_file * seq,void * v)666 static int bpf_array_map_seq_show(struct seq_file *seq, void *v)
667 {
668 return __bpf_array_map_seq_show(seq, v);
669 }
670
bpf_array_map_seq_stop(struct seq_file * seq,void * v)671 static void bpf_array_map_seq_stop(struct seq_file *seq, void *v)
672 {
673 if (!v)
674 (void)__bpf_array_map_seq_show(seq, NULL);
675 }
676
bpf_iter_init_array_map(void * priv_data,struct bpf_iter_aux_info * aux)677 static int bpf_iter_init_array_map(void *priv_data,
678 struct bpf_iter_aux_info *aux)
679 {
680 struct bpf_iter_seq_array_map_info *seq_info = priv_data;
681 struct bpf_map *map = aux->map;
682 struct bpf_array *array = container_of(map, struct bpf_array, map);
683 void *value_buf;
684 u32 buf_size;
685
686 if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
687 buf_size = array->elem_size * num_possible_cpus();
688 value_buf = kmalloc(buf_size, GFP_USER | __GFP_NOWARN);
689 if (!value_buf)
690 return -ENOMEM;
691
692 seq_info->percpu_value_buf = value_buf;
693 }
694
695 /* bpf_iter_attach_map() acquires a map uref, and the uref may be
696 * released before or in the middle of iterating map elements, so
697 * acquire an extra map uref for iterator.
698 */
699 bpf_map_inc_with_uref(map);
700 seq_info->map = map;
701 return 0;
702 }
703
bpf_iter_fini_array_map(void * priv_data)704 static void bpf_iter_fini_array_map(void *priv_data)
705 {
706 struct bpf_iter_seq_array_map_info *seq_info = priv_data;
707
708 bpf_map_put_with_uref(seq_info->map);
709 kfree(seq_info->percpu_value_buf);
710 }
711
712 static const struct seq_operations bpf_array_map_seq_ops = {
713 .start = bpf_array_map_seq_start,
714 .next = bpf_array_map_seq_next,
715 .stop = bpf_array_map_seq_stop,
716 .show = bpf_array_map_seq_show,
717 };
718
719 static const struct bpf_iter_seq_info iter_seq_info = {
720 .seq_ops = &bpf_array_map_seq_ops,
721 .init_seq_private = bpf_iter_init_array_map,
722 .fini_seq_private = bpf_iter_fini_array_map,
723 .seq_priv_size = sizeof(struct bpf_iter_seq_array_map_info),
724 };
725
bpf_for_each_array_elem(struct bpf_map * map,bpf_callback_t callback_fn,void * callback_ctx,u64 flags)726 static long bpf_for_each_array_elem(struct bpf_map *map, bpf_callback_t callback_fn,
727 void *callback_ctx, u64 flags)
728 {
729 u32 i, key, num_elems = 0;
730 struct bpf_array *array;
731 bool is_percpu;
732 u64 ret = 0;
733 void *val;
734
735 if (flags != 0)
736 return -EINVAL;
737
738 is_percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
739 array = container_of(map, struct bpf_array, map);
740 if (is_percpu)
741 migrate_disable();
742 for (i = 0; i < map->max_entries; i++) {
743 if (is_percpu)
744 val = this_cpu_ptr(array->pptrs[i]);
745 else
746 val = array_map_elem_ptr(array, i);
747 num_elems++;
748 key = i;
749 ret = callback_fn((u64)(long)map, (u64)(long)&key,
750 (u64)(long)val, (u64)(long)callback_ctx, 0);
751 /* return value: 0 - continue, 1 - stop and return */
752 if (ret)
753 break;
754 }
755
756 if (is_percpu)
757 migrate_enable();
758 return num_elems;
759 }
760
array_map_mem_usage(const struct bpf_map * map)761 static u64 array_map_mem_usage(const struct bpf_map *map)
762 {
763 struct bpf_array *array = container_of(map, struct bpf_array, map);
764 bool percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
765 u32 elem_size = array->elem_size;
766 u64 entries = map->max_entries;
767 u64 usage = sizeof(*array);
768
769 if (percpu) {
770 usage += entries * sizeof(void *);
771 usage += entries * elem_size * num_possible_cpus();
772 } else {
773 if (map->map_flags & BPF_F_MMAPABLE) {
774 usage = PAGE_ALIGN(usage);
775 usage += PAGE_ALIGN(entries * elem_size);
776 } else {
777 usage += entries * elem_size;
778 }
779 }
780 return usage;
781 }
782
783 BTF_ID_LIST_SINGLE(array_map_btf_ids, struct, bpf_array)
784 const struct bpf_map_ops array_map_ops = {
785 .map_meta_equal = array_map_meta_equal,
786 .map_alloc_check = array_map_alloc_check,
787 .map_alloc = array_map_alloc,
788 .map_free = array_map_free,
789 .map_get_next_key = array_map_get_next_key,
790 .map_release_uref = array_map_free_timers_wq,
791 .map_lookup_elem = array_map_lookup_elem,
792 .map_update_elem = array_map_update_elem,
793 .map_delete_elem = array_map_delete_elem,
794 .map_gen_lookup = array_map_gen_lookup,
795 .map_direct_value_addr = array_map_direct_value_addr,
796 .map_direct_value_meta = array_map_direct_value_meta,
797 .map_mmap = array_map_mmap,
798 .map_seq_show_elem = array_map_seq_show_elem,
799 .map_check_btf = array_map_check_btf,
800 .map_lookup_batch = generic_map_lookup_batch,
801 .map_update_batch = generic_map_update_batch,
802 .map_set_for_each_callback_args = map_set_for_each_callback_args,
803 .map_for_each_callback = bpf_for_each_array_elem,
804 .map_mem_usage = array_map_mem_usage,
805 .map_btf_id = &array_map_btf_ids[0],
806 .iter_seq_info = &iter_seq_info,
807 };
808
809 const struct bpf_map_ops percpu_array_map_ops = {
810 .map_meta_equal = bpf_map_meta_equal,
811 .map_alloc_check = array_map_alloc_check,
812 .map_alloc = array_map_alloc,
813 .map_free = array_map_free,
814 .map_get_next_key = array_map_get_next_key,
815 .map_lookup_elem = percpu_array_map_lookup_elem,
816 .map_gen_lookup = percpu_array_map_gen_lookup,
817 .map_update_elem = array_map_update_elem,
818 .map_delete_elem = array_map_delete_elem,
819 .map_lookup_percpu_elem = percpu_array_map_lookup_percpu_elem,
820 .map_seq_show_elem = percpu_array_map_seq_show_elem,
821 .map_check_btf = array_map_check_btf,
822 .map_lookup_batch = generic_map_lookup_batch,
823 .map_update_batch = generic_map_update_batch,
824 .map_set_for_each_callback_args = map_set_for_each_callback_args,
825 .map_for_each_callback = bpf_for_each_array_elem,
826 .map_mem_usage = array_map_mem_usage,
827 .map_btf_id = &array_map_btf_ids[0],
828 .iter_seq_info = &iter_seq_info,
829 };
830
fd_array_map_alloc_check(union bpf_attr * attr)831 static int fd_array_map_alloc_check(union bpf_attr *attr)
832 {
833 /* only file descriptors can be stored in this type of map */
834 if (attr->value_size != sizeof(u32))
835 return -EINVAL;
836 /* Program read-only/write-only not supported for special maps yet. */
837 if (attr->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG))
838 return -EINVAL;
839 return array_map_alloc_check(attr);
840 }
841
fd_array_map_free(struct bpf_map * map)842 static void fd_array_map_free(struct bpf_map *map)
843 {
844 struct bpf_array *array = container_of(map, struct bpf_array, map);
845 int i;
846
847 /* make sure it's empty */
848 for (i = 0; i < array->map.max_entries; i++)
849 BUG_ON(array->ptrs[i] != NULL);
850
851 bpf_map_area_free(array);
852 }
853
fd_array_map_lookup_elem(struct bpf_map * map,void * key)854 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
855 {
856 return ERR_PTR(-EOPNOTSUPP);
857 }
858
859 /* only called from syscall */
bpf_fd_array_map_lookup_elem(struct bpf_map * map,void * key,u32 * value)860 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
861 {
862 void **elem, *ptr;
863 int ret = 0;
864
865 if (!map->ops->map_fd_sys_lookup_elem)
866 return -ENOTSUPP;
867
868 rcu_read_lock();
869 elem = array_map_lookup_elem(map, key);
870 if (elem && (ptr = READ_ONCE(*elem)))
871 *value = map->ops->map_fd_sys_lookup_elem(ptr);
872 else
873 ret = -ENOENT;
874 rcu_read_unlock();
875
876 return ret;
877 }
878
879 /* only called from syscall */
bpf_fd_array_map_update_elem(struct bpf_map * map,struct file * map_file,void * key,void * value,u64 map_flags)880 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
881 void *key, void *value, u64 map_flags)
882 {
883 struct bpf_array *array = container_of(map, struct bpf_array, map);
884 void *new_ptr, *old_ptr;
885 u32 index = *(u32 *)key, ufd;
886
887 if (map_flags != BPF_ANY)
888 return -EINVAL;
889
890 if (index >= array->map.max_entries)
891 return -E2BIG;
892
893 ufd = *(u32 *)value;
894 new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
895 if (IS_ERR(new_ptr))
896 return PTR_ERR(new_ptr);
897
898 if (map->ops->map_poke_run) {
899 mutex_lock(&array->aux->poke_mutex);
900 old_ptr = xchg(array->ptrs + index, new_ptr);
901 map->ops->map_poke_run(map, index, old_ptr, new_ptr);
902 mutex_unlock(&array->aux->poke_mutex);
903 } else {
904 old_ptr = xchg(array->ptrs + index, new_ptr);
905 }
906
907 if (old_ptr)
908 map->ops->map_fd_put_ptr(map, old_ptr, true);
909 return 0;
910 }
911
__fd_array_map_delete_elem(struct bpf_map * map,void * key,bool need_defer)912 static long __fd_array_map_delete_elem(struct bpf_map *map, void *key, bool need_defer)
913 {
914 struct bpf_array *array = container_of(map, struct bpf_array, map);
915 void *old_ptr;
916 u32 index = *(u32 *)key;
917
918 if (index >= array->map.max_entries)
919 return -E2BIG;
920
921 if (map->ops->map_poke_run) {
922 mutex_lock(&array->aux->poke_mutex);
923 old_ptr = xchg(array->ptrs + index, NULL);
924 map->ops->map_poke_run(map, index, old_ptr, NULL);
925 mutex_unlock(&array->aux->poke_mutex);
926 } else {
927 old_ptr = xchg(array->ptrs + index, NULL);
928 }
929
930 if (old_ptr) {
931 map->ops->map_fd_put_ptr(map, old_ptr, need_defer);
932 return 0;
933 } else {
934 return -ENOENT;
935 }
936 }
937
fd_array_map_delete_elem(struct bpf_map * map,void * key)938 static long fd_array_map_delete_elem(struct bpf_map *map, void *key)
939 {
940 return __fd_array_map_delete_elem(map, key, true);
941 }
942
prog_fd_array_get_ptr(struct bpf_map * map,struct file * map_file,int fd)943 static void *prog_fd_array_get_ptr(struct bpf_map *map,
944 struct file *map_file, int fd)
945 {
946 struct bpf_prog *prog = bpf_prog_get(fd);
947
948 if (IS_ERR(prog))
949 return prog;
950
951 if (!bpf_prog_map_compatible(map, prog)) {
952 bpf_prog_put(prog);
953 return ERR_PTR(-EINVAL);
954 }
955
956 return prog;
957 }
958
prog_fd_array_put_ptr(struct bpf_map * map,void * ptr,bool need_defer)959 static void prog_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
960 {
961 /* bpf_prog is freed after one RCU or tasks trace grace period */
962 bpf_prog_put(ptr);
963 }
964
prog_fd_array_sys_lookup_elem(void * ptr)965 static u32 prog_fd_array_sys_lookup_elem(void *ptr)
966 {
967 return ((struct bpf_prog *)ptr)->aux->id;
968 }
969
970 /* decrement refcnt of all bpf_progs that are stored in this map */
bpf_fd_array_map_clear(struct bpf_map * map,bool need_defer)971 static void bpf_fd_array_map_clear(struct bpf_map *map, bool need_defer)
972 {
973 struct bpf_array *array = container_of(map, struct bpf_array, map);
974 int i;
975
976 for (i = 0; i < array->map.max_entries; i++)
977 __fd_array_map_delete_elem(map, &i, need_defer);
978 }
979
prog_array_map_seq_show_elem(struct bpf_map * map,void * key,struct seq_file * m)980 static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key,
981 struct seq_file *m)
982 {
983 void **elem, *ptr;
984 u32 prog_id;
985
986 rcu_read_lock();
987
988 elem = array_map_lookup_elem(map, key);
989 if (elem) {
990 ptr = READ_ONCE(*elem);
991 if (ptr) {
992 seq_printf(m, "%u: ", *(u32 *)key);
993 prog_id = prog_fd_array_sys_lookup_elem(ptr);
994 btf_type_seq_show(map->btf, map->btf_value_type_id,
995 &prog_id, m);
996 seq_puts(m, "\n");
997 }
998 }
999
1000 rcu_read_unlock();
1001 }
1002
1003 struct prog_poke_elem {
1004 struct list_head list;
1005 struct bpf_prog_aux *aux;
1006 };
1007
prog_array_map_poke_track(struct bpf_map * map,struct bpf_prog_aux * prog_aux)1008 static int prog_array_map_poke_track(struct bpf_map *map,
1009 struct bpf_prog_aux *prog_aux)
1010 {
1011 struct prog_poke_elem *elem;
1012 struct bpf_array_aux *aux;
1013 int ret = 0;
1014
1015 aux = container_of(map, struct bpf_array, map)->aux;
1016 mutex_lock(&aux->poke_mutex);
1017 list_for_each_entry(elem, &aux->poke_progs, list) {
1018 if (elem->aux == prog_aux)
1019 goto out;
1020 }
1021
1022 elem = kmalloc(sizeof(*elem), GFP_KERNEL);
1023 if (!elem) {
1024 ret = -ENOMEM;
1025 goto out;
1026 }
1027
1028 INIT_LIST_HEAD(&elem->list);
1029 /* We must track the program's aux info at this point in time
1030 * since the program pointer itself may not be stable yet, see
1031 * also comment in prog_array_map_poke_run().
1032 */
1033 elem->aux = prog_aux;
1034
1035 list_add_tail(&elem->list, &aux->poke_progs);
1036 out:
1037 mutex_unlock(&aux->poke_mutex);
1038 return ret;
1039 }
1040
prog_array_map_poke_untrack(struct bpf_map * map,struct bpf_prog_aux * prog_aux)1041 static void prog_array_map_poke_untrack(struct bpf_map *map,
1042 struct bpf_prog_aux *prog_aux)
1043 {
1044 struct prog_poke_elem *elem, *tmp;
1045 struct bpf_array_aux *aux;
1046
1047 aux = container_of(map, struct bpf_array, map)->aux;
1048 mutex_lock(&aux->poke_mutex);
1049 list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
1050 if (elem->aux == prog_aux) {
1051 list_del_init(&elem->list);
1052 kfree(elem);
1053 break;
1054 }
1055 }
1056 mutex_unlock(&aux->poke_mutex);
1057 }
1058
bpf_arch_poke_desc_update(struct bpf_jit_poke_descriptor * poke,struct bpf_prog * new,struct bpf_prog * old)1059 void __weak bpf_arch_poke_desc_update(struct bpf_jit_poke_descriptor *poke,
1060 struct bpf_prog *new, struct bpf_prog *old)
1061 {
1062 WARN_ON_ONCE(1);
1063 }
1064
prog_array_map_poke_run(struct bpf_map * map,u32 key,struct bpf_prog * old,struct bpf_prog * new)1065 static void prog_array_map_poke_run(struct bpf_map *map, u32 key,
1066 struct bpf_prog *old,
1067 struct bpf_prog *new)
1068 {
1069 struct prog_poke_elem *elem;
1070 struct bpf_array_aux *aux;
1071
1072 aux = container_of(map, struct bpf_array, map)->aux;
1073 WARN_ON_ONCE(!mutex_is_locked(&aux->poke_mutex));
1074
1075 list_for_each_entry(elem, &aux->poke_progs, list) {
1076 struct bpf_jit_poke_descriptor *poke;
1077 int i;
1078
1079 for (i = 0; i < elem->aux->size_poke_tab; i++) {
1080 poke = &elem->aux->poke_tab[i];
1081
1082 /* Few things to be aware of:
1083 *
1084 * 1) We can only ever access aux in this context, but
1085 * not aux->prog since it might not be stable yet and
1086 * there could be danger of use after free otherwise.
1087 * 2) Initially when we start tracking aux, the program
1088 * is not JITed yet and also does not have a kallsyms
1089 * entry. We skip these as poke->tailcall_target_stable
1090 * is not active yet. The JIT will do the final fixup
1091 * before setting it stable. The various
1092 * poke->tailcall_target_stable are successively
1093 * activated, so tail call updates can arrive from here
1094 * while JIT is still finishing its final fixup for
1095 * non-activated poke entries.
1096 * 3) Also programs reaching refcount of zero while patching
1097 * is in progress is okay since we're protected under
1098 * poke_mutex and untrack the programs before the JIT
1099 * buffer is freed.
1100 */
1101 if (!READ_ONCE(poke->tailcall_target_stable))
1102 continue;
1103 if (poke->reason != BPF_POKE_REASON_TAIL_CALL)
1104 continue;
1105 if (poke->tail_call.map != map ||
1106 poke->tail_call.key != key)
1107 continue;
1108
1109 bpf_arch_poke_desc_update(poke, new, old);
1110 }
1111 }
1112 }
1113
prog_array_map_clear_deferred(struct work_struct * work)1114 static void prog_array_map_clear_deferred(struct work_struct *work)
1115 {
1116 struct bpf_map *map = container_of(work, struct bpf_array_aux,
1117 work)->map;
1118 bpf_fd_array_map_clear(map, true);
1119 bpf_map_put(map);
1120 }
1121
prog_array_map_clear(struct bpf_map * map)1122 static void prog_array_map_clear(struct bpf_map *map)
1123 {
1124 struct bpf_array_aux *aux = container_of(map, struct bpf_array,
1125 map)->aux;
1126 bpf_map_inc(map);
1127 schedule_work(&aux->work);
1128 }
1129
prog_array_map_alloc(union bpf_attr * attr)1130 static struct bpf_map *prog_array_map_alloc(union bpf_attr *attr)
1131 {
1132 struct bpf_array_aux *aux;
1133 struct bpf_map *map;
1134
1135 aux = kzalloc(sizeof(*aux), GFP_KERNEL_ACCOUNT);
1136 if (!aux)
1137 return ERR_PTR(-ENOMEM);
1138
1139 INIT_WORK(&aux->work, prog_array_map_clear_deferred);
1140 INIT_LIST_HEAD(&aux->poke_progs);
1141 mutex_init(&aux->poke_mutex);
1142
1143 map = array_map_alloc(attr);
1144 if (IS_ERR(map)) {
1145 kfree(aux);
1146 return map;
1147 }
1148
1149 container_of(map, struct bpf_array, map)->aux = aux;
1150 aux->map = map;
1151
1152 return map;
1153 }
1154
prog_array_map_free(struct bpf_map * map)1155 static void prog_array_map_free(struct bpf_map *map)
1156 {
1157 struct prog_poke_elem *elem, *tmp;
1158 struct bpf_array_aux *aux;
1159
1160 aux = container_of(map, struct bpf_array, map)->aux;
1161 list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
1162 list_del_init(&elem->list);
1163 kfree(elem);
1164 }
1165 kfree(aux);
1166 fd_array_map_free(map);
1167 }
1168
1169 /* prog_array->aux->{type,jited} is a runtime binding.
1170 * Doing static check alone in the verifier is not enough.
1171 * Thus, prog_array_map cannot be used as an inner_map
1172 * and map_meta_equal is not implemented.
1173 */
1174 const struct bpf_map_ops prog_array_map_ops = {
1175 .map_alloc_check = fd_array_map_alloc_check,
1176 .map_alloc = prog_array_map_alloc,
1177 .map_free = prog_array_map_free,
1178 .map_poke_track = prog_array_map_poke_track,
1179 .map_poke_untrack = prog_array_map_poke_untrack,
1180 .map_poke_run = prog_array_map_poke_run,
1181 .map_get_next_key = array_map_get_next_key,
1182 .map_lookup_elem = fd_array_map_lookup_elem,
1183 .map_delete_elem = fd_array_map_delete_elem,
1184 .map_fd_get_ptr = prog_fd_array_get_ptr,
1185 .map_fd_put_ptr = prog_fd_array_put_ptr,
1186 .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
1187 .map_release_uref = prog_array_map_clear,
1188 .map_seq_show_elem = prog_array_map_seq_show_elem,
1189 .map_mem_usage = array_map_mem_usage,
1190 .map_btf_id = &array_map_btf_ids[0],
1191 };
1192
bpf_event_entry_gen(struct file * perf_file,struct file * map_file)1193 static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
1194 struct file *map_file)
1195 {
1196 struct bpf_event_entry *ee;
1197
1198 ee = kzalloc(sizeof(*ee), GFP_KERNEL);
1199 if (ee) {
1200 ee->event = perf_file->private_data;
1201 ee->perf_file = perf_file;
1202 ee->map_file = map_file;
1203 }
1204
1205 return ee;
1206 }
1207
__bpf_event_entry_free(struct rcu_head * rcu)1208 static void __bpf_event_entry_free(struct rcu_head *rcu)
1209 {
1210 struct bpf_event_entry *ee;
1211
1212 ee = container_of(rcu, struct bpf_event_entry, rcu);
1213 fput(ee->perf_file);
1214 kfree(ee);
1215 }
1216
bpf_event_entry_free_rcu(struct bpf_event_entry * ee)1217 static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
1218 {
1219 call_rcu(&ee->rcu, __bpf_event_entry_free);
1220 }
1221
perf_event_fd_array_get_ptr(struct bpf_map * map,struct file * map_file,int fd)1222 static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
1223 struct file *map_file, int fd)
1224 {
1225 struct bpf_event_entry *ee;
1226 struct perf_event *event;
1227 struct file *perf_file;
1228 u64 value;
1229
1230 perf_file = perf_event_get(fd);
1231 if (IS_ERR(perf_file))
1232 return perf_file;
1233
1234 ee = ERR_PTR(-EOPNOTSUPP);
1235 event = perf_file->private_data;
1236 if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
1237 goto err_out;
1238
1239 ee = bpf_event_entry_gen(perf_file, map_file);
1240 if (ee)
1241 return ee;
1242 ee = ERR_PTR(-ENOMEM);
1243 err_out:
1244 fput(perf_file);
1245 return ee;
1246 }
1247
perf_event_fd_array_put_ptr(struct bpf_map * map,void * ptr,bool need_defer)1248 static void perf_event_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
1249 {
1250 /* bpf_perf_event is freed after one RCU grace period */
1251 bpf_event_entry_free_rcu(ptr);
1252 }
1253
perf_event_fd_array_release(struct bpf_map * map,struct file * map_file)1254 static void perf_event_fd_array_release(struct bpf_map *map,
1255 struct file *map_file)
1256 {
1257 struct bpf_array *array = container_of(map, struct bpf_array, map);
1258 struct bpf_event_entry *ee;
1259 int i;
1260
1261 if (map->map_flags & BPF_F_PRESERVE_ELEMS)
1262 return;
1263
1264 rcu_read_lock();
1265 for (i = 0; i < array->map.max_entries; i++) {
1266 ee = READ_ONCE(array->ptrs[i]);
1267 if (ee && ee->map_file == map_file)
1268 __fd_array_map_delete_elem(map, &i, true);
1269 }
1270 rcu_read_unlock();
1271 }
1272
perf_event_fd_array_map_free(struct bpf_map * map)1273 static void perf_event_fd_array_map_free(struct bpf_map *map)
1274 {
1275 if (map->map_flags & BPF_F_PRESERVE_ELEMS)
1276 bpf_fd_array_map_clear(map, false);
1277 fd_array_map_free(map);
1278 }
1279
1280 const struct bpf_map_ops perf_event_array_map_ops = {
1281 .map_meta_equal = bpf_map_meta_equal,
1282 .map_alloc_check = fd_array_map_alloc_check,
1283 .map_alloc = array_map_alloc,
1284 .map_free = perf_event_fd_array_map_free,
1285 .map_get_next_key = array_map_get_next_key,
1286 .map_lookup_elem = fd_array_map_lookup_elem,
1287 .map_delete_elem = fd_array_map_delete_elem,
1288 .map_fd_get_ptr = perf_event_fd_array_get_ptr,
1289 .map_fd_put_ptr = perf_event_fd_array_put_ptr,
1290 .map_release = perf_event_fd_array_release,
1291 .map_check_btf = map_check_no_btf,
1292 .map_mem_usage = array_map_mem_usage,
1293 .map_btf_id = &array_map_btf_ids[0],
1294 };
1295
1296 #ifdef CONFIG_CGROUPS
cgroup_fd_array_get_ptr(struct bpf_map * map,struct file * map_file,int fd)1297 static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
1298 struct file *map_file /* not used */,
1299 int fd)
1300 {
1301 return cgroup_get_from_fd(fd);
1302 }
1303
cgroup_fd_array_put_ptr(struct bpf_map * map,void * ptr,bool need_defer)1304 static void cgroup_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
1305 {
1306 /* cgroup_put free cgrp after a rcu grace period */
1307 cgroup_put(ptr);
1308 }
1309
cgroup_fd_array_free(struct bpf_map * map)1310 static void cgroup_fd_array_free(struct bpf_map *map)
1311 {
1312 bpf_fd_array_map_clear(map, false);
1313 fd_array_map_free(map);
1314 }
1315
1316 const struct bpf_map_ops cgroup_array_map_ops = {
1317 .map_meta_equal = bpf_map_meta_equal,
1318 .map_alloc_check = fd_array_map_alloc_check,
1319 .map_alloc = array_map_alloc,
1320 .map_free = cgroup_fd_array_free,
1321 .map_get_next_key = array_map_get_next_key,
1322 .map_lookup_elem = fd_array_map_lookup_elem,
1323 .map_delete_elem = fd_array_map_delete_elem,
1324 .map_fd_get_ptr = cgroup_fd_array_get_ptr,
1325 .map_fd_put_ptr = cgroup_fd_array_put_ptr,
1326 .map_check_btf = map_check_no_btf,
1327 .map_mem_usage = array_map_mem_usage,
1328 .map_btf_id = &array_map_btf_ids[0],
1329 };
1330 #endif
1331
array_of_map_alloc(union bpf_attr * attr)1332 static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
1333 {
1334 struct bpf_map *map, *inner_map_meta;
1335
1336 inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
1337 if (IS_ERR(inner_map_meta))
1338 return inner_map_meta;
1339
1340 map = array_map_alloc(attr);
1341 if (IS_ERR(map)) {
1342 bpf_map_meta_free(inner_map_meta);
1343 return map;
1344 }
1345
1346 map->inner_map_meta = inner_map_meta;
1347
1348 return map;
1349 }
1350
array_of_map_free(struct bpf_map * map)1351 static void array_of_map_free(struct bpf_map *map)
1352 {
1353 /* map->inner_map_meta is only accessed by syscall which
1354 * is protected by fdget/fdput.
1355 */
1356 bpf_map_meta_free(map->inner_map_meta);
1357 bpf_fd_array_map_clear(map, false);
1358 fd_array_map_free(map);
1359 }
1360
array_of_map_lookup_elem(struct bpf_map * map,void * key)1361 static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
1362 {
1363 struct bpf_map **inner_map = array_map_lookup_elem(map, key);
1364
1365 if (!inner_map)
1366 return NULL;
1367
1368 return READ_ONCE(*inner_map);
1369 }
1370
array_of_map_gen_lookup(struct bpf_map * map,struct bpf_insn * insn_buf)1371 static int array_of_map_gen_lookup(struct bpf_map *map,
1372 struct bpf_insn *insn_buf)
1373 {
1374 struct bpf_array *array = container_of(map, struct bpf_array, map);
1375 u32 elem_size = array->elem_size;
1376 struct bpf_insn *insn = insn_buf;
1377 const int ret = BPF_REG_0;
1378 const int map_ptr = BPF_REG_1;
1379 const int index = BPF_REG_2;
1380
1381 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
1382 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
1383 if (!map->bypass_spec_v1) {
1384 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
1385 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
1386 } else {
1387 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
1388 }
1389 if (is_power_of_2(elem_size))
1390 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
1391 else
1392 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
1393 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
1394 *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
1395 *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
1396 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
1397 *insn++ = BPF_MOV64_IMM(ret, 0);
1398
1399 return insn - insn_buf;
1400 }
1401
1402 const struct bpf_map_ops array_of_maps_map_ops = {
1403 .map_alloc_check = fd_array_map_alloc_check,
1404 .map_alloc = array_of_map_alloc,
1405 .map_free = array_of_map_free,
1406 .map_get_next_key = array_map_get_next_key,
1407 .map_lookup_elem = array_of_map_lookup_elem,
1408 .map_delete_elem = fd_array_map_delete_elem,
1409 .map_fd_get_ptr = bpf_map_fd_get_ptr,
1410 .map_fd_put_ptr = bpf_map_fd_put_ptr,
1411 .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
1412 .map_gen_lookup = array_of_map_gen_lookup,
1413 .map_lookup_batch = generic_map_lookup_batch,
1414 .map_update_batch = generic_map_update_batch,
1415 .map_check_btf = map_check_no_btf,
1416 .map_mem_usage = array_map_mem_usage,
1417 .map_btf_id = &array_map_btf_ids[0],
1418 };
1419