projects/linux·blob v5.1 · refs

807 lines · 20.8 KiB

1/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2 * Copyright (c) 2016,2017 Facebook
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 */
13#include <linux/bpf.h>
14#include <linux/btf.h>
15#include <linux/err.h>
16#include <linux/slab.h>
17#include <linux/mm.h>
18#include <linux/filter.h>
19#include <linux/perf_event.h>
20#include <uapi/linux/btf.h>
21
22#include "map_in_map.h"
23
24#define ARRAY_CREATE_FLAG_MASK \
25 (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
26
27static void bpf_array_free_percpu(struct bpf_array *array)
28{
29 int i;
30
31 for (i = 0; i < array->map.max_entries; i++) {
32 free_percpu(array->pptrs[i]);
33 cond_resched();
34 }
35}
36
37static int bpf_array_alloc_percpu(struct bpf_array *array)
38{
39 void __percpu *ptr;
40 int i;
41
42 for (i = 0; i < array->map.max_entries; i++) {
43 ptr = __alloc_percpu_gfp(array->elem_size, 8,
44 GFP_USER | __GFP_NOWARN);
45 if (!ptr) {
46 bpf_array_free_percpu(array);
47 return -ENOMEM;
48 }
49 array->pptrs[i] = ptr;
50 cond_resched();
51 }
52
53 return 0;
54}
55
56/* Called from syscall */
57int array_map_alloc_check(union bpf_attr *attr)
58{
59 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
60 int numa_node = bpf_map_attr_numa_node(attr);
61
62 /* check sanity of attributes */
63 if (attr->max_entries == 0 || attr->key_size != 4 ||
64 attr->value_size == 0 ||
65 attr->map_flags & ~ARRAY_CREATE_FLAG_MASK ||
66 (percpu && numa_node != NUMA_NO_NODE))
67 return -EINVAL;
68
69 if (attr->value_size > KMALLOC_MAX_SIZE)
70 /* if value_size is bigger, the user space won't be able to
71 * access the elements.
72 */
73 return -E2BIG;
74
75 return 0;
76}
77
78static struct bpf_map *array_map_alloc(union bpf_attr *attr)
79{
80 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
81 int ret, numa_node = bpf_map_attr_numa_node(attr);
82 u32 elem_size, index_mask, max_entries;
83 bool unpriv = !capable(CAP_SYS_ADMIN);
84 u64 cost, array_size, mask64;
85 struct bpf_array *array;
86
87 elem_size = round_up(attr->value_size, 8);
88
89 max_entries = attr->max_entries;
90
91 /* On 32 bit archs roundup_pow_of_two() with max_entries that has
92 * upper most bit set in u32 space is undefined behavior due to
93 * resulting 1U << 32, so do it manually here in u64 space.
94 */
95 mask64 = fls_long(max_entries - 1);
96 mask64 = 1ULL << mask64;
97 mask64 -= 1;
98
99 index_mask = mask64;
100 if (unpriv) {
101 /* round up array size to nearest power of 2,
102 * since cpu will speculate within index_mask limits
103 */
104 max_entries = index_mask + 1;
105 /* Check for overflows. */
106 if (max_entries < attr->max_entries)
107 return ERR_PTR(-E2BIG);
108 }
109
110 array_size = sizeof(*array);
111 if (percpu)
112 array_size += (u64) max_entries * sizeof(void *);
113 else
114 array_size += (u64) max_entries * elem_size;
115
116 /* make sure there is no u32 overflow later in round_up() */
117 cost = array_size;
118 if (cost >= U32_MAX - PAGE_SIZE)
119 return ERR_PTR(-ENOMEM);
120 if (percpu) {
121 cost += (u64)attr->max_entries * elem_size * num_possible_cpus();
122 if (cost >= U32_MAX - PAGE_SIZE)
123 return ERR_PTR(-ENOMEM);
124 }
125 cost = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
126
127 ret = bpf_map_precharge_memlock(cost);
128 if (ret < 0)
129 return ERR_PTR(ret);
130
131 /* allocate all map elements and zero-initialize them */
132 array = bpf_map_area_alloc(array_size, numa_node);
133 if (!array)
134 return ERR_PTR(-ENOMEM);
135 array->index_mask = index_mask;
136 array->map.unpriv_array = unpriv;
137
138 /* copy mandatory map attributes */
139 bpf_map_init_from_attr(&array->map, attr);
140 array->map.pages = cost;
141 array->elem_size = elem_size;
142
143 if (percpu && bpf_array_alloc_percpu(array)) {
144 bpf_map_area_free(array);
145 return ERR_PTR(-ENOMEM);
146 }
147
148 return &array->map;
149}
150
151/* Called from syscall or from eBPF program */
152static void *array_map_lookup_elem(struct bpf_map *map, void *key)
153{
154 struct bpf_array *array = container_of(map, struct bpf_array, map);
155 u32 index = *(u32 *)key;
156
157 if (unlikely(index >= array->map.max_entries))
158 return NULL;
159
160 return array->value + array->elem_size * (index & array->index_mask);
161}
162
163/* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
164static u32 array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
165{
166 struct bpf_array *array = container_of(map, struct bpf_array, map);
167 struct bpf_insn *insn = insn_buf;
168 u32 elem_size = round_up(map->value_size, 8);
169 const int ret = BPF_REG_0;
170 const int map_ptr = BPF_REG_1;
171 const int index = BPF_REG_2;
172
173 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
174 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
175 if (map->unpriv_array) {
176 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4);
177 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
178 } else {
179 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
180 }
181
182 if (is_power_of_2(elem_size)) {
183 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
184 } else {
185 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
186 }
187 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
188 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
189 *insn++ = BPF_MOV64_IMM(ret, 0);
190 return insn - insn_buf;
191}
192
193/* Called from eBPF program */
194static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
195{
196 struct bpf_array *array = container_of(map, struct bpf_array, map);
197 u32 index = *(u32 *)key;
198
199 if (unlikely(index >= array->map.max_entries))
200 return NULL;
201
202 return this_cpu_ptr(array->pptrs[index & array->index_mask]);
203}
204
205int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
206{
207 struct bpf_array *array = container_of(map, struct bpf_array, map);
208 u32 index = *(u32 *)key;
209 void __percpu *pptr;
210 int cpu, off = 0;
211 u32 size;
212
213 if (unlikely(index >= array->map.max_entries))
214 return -ENOENT;
215
216 /* per_cpu areas are zero-filled and bpf programs can only
217 * access 'value_size' of them, so copying rounded areas
218 * will not leak any kernel data
219 */
220 size = round_up(map->value_size, 8);
221 rcu_read_lock();
222 pptr = array->pptrs[index & array->index_mask];
223 for_each_possible_cpu(cpu) {
224 bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
225 off += size;
226 }
227 rcu_read_unlock();
228 return 0;
229}
230
231/* Called from syscall */
232static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
233{
234 struct bpf_array *array = container_of(map, struct bpf_array, map);
235 u32 index = key ? *(u32 *)key : U32_MAX;
236 u32 *next = (u32 *)next_key;
237
238 if (index >= array->map.max_entries) {
239 *next = 0;
240 return 0;
241 }
242
243 if (index == array->map.max_entries - 1)
244 return -ENOENT;
245
246 *next = index + 1;
247 return 0;
248}
249
250/* Called from syscall or from eBPF program */
251static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
252 u64 map_flags)
253{
254 struct bpf_array *array = container_of(map, struct bpf_array, map);
255 u32 index = *(u32 *)key;
256 char *val;
257
258 if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
259 /* unknown flags */
260 return -EINVAL;
261
262 if (unlikely(index >= array->map.max_entries))
263 /* all elements were pre-allocated, cannot insert a new one */
264 return -E2BIG;
265
266 if (unlikely(map_flags & BPF_NOEXIST))
267 /* all elements already exist */
268 return -EEXIST;
269
270 if (unlikely((map_flags & BPF_F_LOCK) &&
271 !map_value_has_spin_lock(map)))
272 return -EINVAL;
273
274 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
275 memcpy(this_cpu_ptr(array->pptrs[index & array->index_mask]),
276 value, map->value_size);
277 } else {
278 val = array->value +
279 array->elem_size * (index & array->index_mask);
280 if (map_flags & BPF_F_LOCK)
281 copy_map_value_locked(map, val, value, false);
282 else
283 copy_map_value(map, val, value);
284 }
285 return 0;
286}
287
288int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
289 u64 map_flags)
290{
291 struct bpf_array *array = container_of(map, struct bpf_array, map);
292 u32 index = *(u32 *)key;
293 void __percpu *pptr;
294 int cpu, off = 0;
295 u32 size;
296
297 if (unlikely(map_flags > BPF_EXIST))
298 /* unknown flags */
299 return -EINVAL;
300
301 if (unlikely(index >= array->map.max_entries))
302 /* all elements were pre-allocated, cannot insert a new one */
303 return -E2BIG;
304
305 if (unlikely(map_flags == BPF_NOEXIST))
306 /* all elements already exist */
307 return -EEXIST;
308
309 /* the user space will provide round_up(value_size, 8) bytes that
310 * will be copied into per-cpu area. bpf programs can only access
311 * value_size of it. During lookup the same extra bytes will be
312 * returned or zeros which were zero-filled by percpu_alloc,
313 * so no kernel data leaks possible
314 */
315 size = round_up(map->value_size, 8);
316 rcu_read_lock();
317 pptr = array->pptrs[index & array->index_mask];
318 for_each_possible_cpu(cpu) {
319 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
320 off += size;
321 }
322 rcu_read_unlock();
323 return 0;
324}
325
326/* Called from syscall or from eBPF program */
327static int array_map_delete_elem(struct bpf_map *map, void *key)
328{
329 return -EINVAL;
330}
331
332/* Called when map->refcnt goes to zero, either from workqueue or from syscall */
333static void array_map_free(struct bpf_map *map)
334{
335 struct bpf_array *array = container_of(map, struct bpf_array, map);
336
337 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
338 * so the programs (can be more than one that used this map) were
339 * disconnected from events. Wait for outstanding programs to complete
340 * and free the array
341 */
342 synchronize_rcu();
343
344 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
345 bpf_array_free_percpu(array);
346
347 bpf_map_area_free(array);
348}
349
350static void array_map_seq_show_elem(struct bpf_map *map, void *key,
351 struct seq_file *m)
352{
353 void *value;
354
355 rcu_read_lock();
356
357 value = array_map_lookup_elem(map, key);
358 if (!value) {
359 rcu_read_unlock();
360 return;
361 }
362
363 seq_printf(m, "%u: ", *(u32 *)key);
364 btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
365 seq_puts(m, "\n");
366
367 rcu_read_unlock();
368}
369
370static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key,
371 struct seq_file *m)
372{
373 struct bpf_array *array = container_of(map, struct bpf_array, map);
374 u32 index = *(u32 *)key;
375 void __percpu *pptr;
376 int cpu;
377
378 rcu_read_lock();
379
380 seq_printf(m, "%u: {\n", *(u32 *)key);
381 pptr = array->pptrs[index & array->index_mask];
382 for_each_possible_cpu(cpu) {
383 seq_printf(m, "\tcpu%d: ", cpu);
384 btf_type_seq_show(map->btf, map->btf_value_type_id,
385 per_cpu_ptr(pptr, cpu), m);
386 seq_puts(m, "\n");
387 }
388 seq_puts(m, "}\n");
389
390 rcu_read_unlock();
391}
392
393static int array_map_check_btf(const struct bpf_map *map,
394 const struct btf *btf,
395 const struct btf_type *key_type,
396 const struct btf_type *value_type)
397{
398 u32 int_data;
399
400 if (BTF_INFO_KIND(key_type->info) != BTF_KIND_INT)
401 return -EINVAL;
402
403 int_data = *(u32 *)(key_type + 1);
404 /* bpf array can only take a u32 key. This check makes sure
405 * that the btf matches the attr used during map_create.
406 */
407 if (BTF_INT_BITS(int_data) != 32 || BTF_INT_OFFSET(int_data))
408 return -EINVAL;
409
410 return 0;
411}
412
413const struct bpf_map_ops array_map_ops = {
414 .map_alloc_check = array_map_alloc_check,
415 .map_alloc = array_map_alloc,
416 .map_free = array_map_free,
417 .map_get_next_key = array_map_get_next_key,
418 .map_lookup_elem = array_map_lookup_elem,
419 .map_update_elem = array_map_update_elem,
420 .map_delete_elem = array_map_delete_elem,
421 .map_gen_lookup = array_map_gen_lookup,
422 .map_seq_show_elem = array_map_seq_show_elem,
423 .map_check_btf = array_map_check_btf,
424};
425
426const struct bpf_map_ops percpu_array_map_ops = {
427 .map_alloc_check = array_map_alloc_check,
428 .map_alloc = array_map_alloc,
429 .map_free = array_map_free,
430 .map_get_next_key = array_map_get_next_key,
431 .map_lookup_elem = percpu_array_map_lookup_elem,
432 .map_update_elem = array_map_update_elem,
433 .map_delete_elem = array_map_delete_elem,
434 .map_seq_show_elem = percpu_array_map_seq_show_elem,
435 .map_check_btf = array_map_check_btf,
436};
437
438static int fd_array_map_alloc_check(union bpf_attr *attr)
439{
440 /* only file descriptors can be stored in this type of map */
441 if (attr->value_size != sizeof(u32))
442 return -EINVAL;
443 return array_map_alloc_check(attr);
444}
445
446static void fd_array_map_free(struct bpf_map *map)
447{
448 struct bpf_array *array = container_of(map, struct bpf_array, map);
449 int i;
450
451 synchronize_rcu();
452
453 /* make sure it's empty */
454 for (i = 0; i < array->map.max_entries; i++)
455 BUG_ON(array->ptrs[i] != NULL);
456
457 bpf_map_area_free(array);
458}
459
460static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
461{
462 return ERR_PTR(-EOPNOTSUPP);
463}
464
465/* only called from syscall */
466int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
467{
468 void **elem, *ptr;
469 int ret = 0;
470
471 if (!map->ops->map_fd_sys_lookup_elem)
472 return -ENOTSUPP;
473
474 rcu_read_lock();
475 elem = array_map_lookup_elem(map, key);
476 if (elem && (ptr = READ_ONCE(*elem)))
477 *value = map->ops->map_fd_sys_lookup_elem(ptr);
478 else
479 ret = -ENOENT;
480 rcu_read_unlock();
481
482 return ret;
483}
484
485/* only called from syscall */
486int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
487 void *key, void *value, u64 map_flags)
488{
489 struct bpf_array *array = container_of(map, struct bpf_array, map);
490 void *new_ptr, *old_ptr;
491 u32 index = *(u32 *)key, ufd;
492
493 if (map_flags != BPF_ANY)
494 return -EINVAL;
495
496 if (index >= array->map.max_entries)
497 return -E2BIG;
498
499 ufd = *(u32 *)value;
500 new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
501 if (IS_ERR(new_ptr))
502 return PTR_ERR(new_ptr);
503
504 old_ptr = xchg(array->ptrs + index, new_ptr);
505 if (old_ptr)
506 map->ops->map_fd_put_ptr(old_ptr);
507
508 return 0;
509}
510
511static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
512{
513 struct bpf_array *array = container_of(map, struct bpf_array, map);
514 void *old_ptr;
515 u32 index = *(u32 *)key;
516
517 if (index >= array->map.max_entries)
518 return -E2BIG;
519
520 old_ptr = xchg(array->ptrs + index, NULL);
521 if (old_ptr) {
522 map->ops->map_fd_put_ptr(old_ptr);
523 return 0;
524 } else {
525 return -ENOENT;
526 }
527}
528
529static void *prog_fd_array_get_ptr(struct bpf_map *map,
530 struct file *map_file, int fd)
531{
532 struct bpf_array *array = container_of(map, struct bpf_array, map);
533 struct bpf_prog *prog = bpf_prog_get(fd);
534
535 if (IS_ERR(prog))
536 return prog;
537
538 if (!bpf_prog_array_compatible(array, prog)) {
539 bpf_prog_put(prog);
540 return ERR_PTR(-EINVAL);
541 }
542
543 return prog;
544}
545
546static void prog_fd_array_put_ptr(void *ptr)
547{
548 bpf_prog_put(ptr);
549}
550
551static u32 prog_fd_array_sys_lookup_elem(void *ptr)
552{
553 return ((struct bpf_prog *)ptr)->aux->id;
554}
555
556/* decrement refcnt of all bpf_progs that are stored in this map */
557static void bpf_fd_array_map_clear(struct bpf_map *map)
558{
559 struct bpf_array *array = container_of(map, struct bpf_array, map);
560 int i;
561
562 for (i = 0; i < array->map.max_entries; i++)
563 fd_array_map_delete_elem(map, &i);
564}
565
566static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key,
567 struct seq_file *m)
568{
569 void **elem, *ptr;
570 u32 prog_id;
571
572 rcu_read_lock();
573
574 elem = array_map_lookup_elem(map, key);
575 if (elem) {
576 ptr = READ_ONCE(*elem);
577 if (ptr) {
578 seq_printf(m, "%u: ", *(u32 *)key);
579 prog_id = prog_fd_array_sys_lookup_elem(ptr);
580 btf_type_seq_show(map->btf, map->btf_value_type_id,
581 &prog_id, m);
582 seq_puts(m, "\n");
583 }
584 }
585
586 rcu_read_unlock();
587}
588
589const struct bpf_map_ops prog_array_map_ops = {
590 .map_alloc_check = fd_array_map_alloc_check,
591 .map_alloc = array_map_alloc,
592 .map_free = fd_array_map_free,
593 .map_get_next_key = array_map_get_next_key,
594 .map_lookup_elem = fd_array_map_lookup_elem,
595 .map_delete_elem = fd_array_map_delete_elem,
596 .map_fd_get_ptr = prog_fd_array_get_ptr,
597 .map_fd_put_ptr = prog_fd_array_put_ptr,
598 .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
599 .map_release_uref = bpf_fd_array_map_clear,
600 .map_seq_show_elem = prog_array_map_seq_show_elem,
601};
602
603static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
604 struct file *map_file)
605{
606 struct bpf_event_entry *ee;
607
608 ee = kzalloc(sizeof(*ee), GFP_ATOMIC);
609 if (ee) {
610 ee->event = perf_file->private_data;
611 ee->perf_file = perf_file;
612 ee->map_file = map_file;
613 }
614
615 return ee;
616}
617
618static void __bpf_event_entry_free(struct rcu_head *rcu)
619{
620 struct bpf_event_entry *ee;
621
622 ee = container_of(rcu, struct bpf_event_entry, rcu);
623 fput(ee->perf_file);
624 kfree(ee);
625}
626
627static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
628{
629 call_rcu(&ee->rcu, __bpf_event_entry_free);
630}
631
632static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
633 struct file *map_file, int fd)
634{
635 struct bpf_event_entry *ee;
636 struct perf_event *event;
637 struct file *perf_file;
638 u64 value;
639
640 perf_file = perf_event_get(fd);
641 if (IS_ERR(perf_file))
642 return perf_file;
643
644 ee = ERR_PTR(-EOPNOTSUPP);
645 event = perf_file->private_data;
646 if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
647 goto err_out;
648
649 ee = bpf_event_entry_gen(perf_file, map_file);
650 if (ee)
651 return ee;
652 ee = ERR_PTR(-ENOMEM);
653err_out:
654 fput(perf_file);
655 return ee;
656}
657
658static void perf_event_fd_array_put_ptr(void *ptr)
659{
660 bpf_event_entry_free_rcu(ptr);
661}
662
663static void perf_event_fd_array_release(struct bpf_map *map,
664 struct file *map_file)
665{
666 struct bpf_array *array = container_of(map, struct bpf_array, map);
667 struct bpf_event_entry *ee;
668 int i;
669
670 rcu_read_lock();
671 for (i = 0; i < array->map.max_entries; i++) {
672 ee = READ_ONCE(array->ptrs[i]);
673 if (ee && ee->map_file == map_file)
674 fd_array_map_delete_elem(map, &i);
675 }
676 rcu_read_unlock();
677}
678
679const struct bpf_map_ops perf_event_array_map_ops = {
680 .map_alloc_check = fd_array_map_alloc_check,
681 .map_alloc = array_map_alloc,
682 .map_free = fd_array_map_free,
683 .map_get_next_key = array_map_get_next_key,
684 .map_lookup_elem = fd_array_map_lookup_elem,
685 .map_delete_elem = fd_array_map_delete_elem,
686 .map_fd_get_ptr = perf_event_fd_array_get_ptr,
687 .map_fd_put_ptr = perf_event_fd_array_put_ptr,
688 .map_release = perf_event_fd_array_release,
689 .map_check_btf = map_check_no_btf,
690};
691
692#ifdef CONFIG_CGROUPS
693static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
694 struct file *map_file /* not used */,
695 int fd)
696{
697 return cgroup_get_from_fd(fd);
698}
699
700static void cgroup_fd_array_put_ptr(void *ptr)
701{
702 /* cgroup_put free cgrp after a rcu grace period */
703 cgroup_put(ptr);
704}
705
706static void cgroup_fd_array_free(struct bpf_map *map)
707{
708 bpf_fd_array_map_clear(map);
709 fd_array_map_free(map);
710}
711
712const struct bpf_map_ops cgroup_array_map_ops = {
713 .map_alloc_check = fd_array_map_alloc_check,
714 .map_alloc = array_map_alloc,
715 .map_free = cgroup_fd_array_free,
716 .map_get_next_key = array_map_get_next_key,
717 .map_lookup_elem = fd_array_map_lookup_elem,
718 .map_delete_elem = fd_array_map_delete_elem,
719 .map_fd_get_ptr = cgroup_fd_array_get_ptr,
720 .map_fd_put_ptr = cgroup_fd_array_put_ptr,
721 .map_check_btf = map_check_no_btf,
722};
723#endif
724
725static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
726{
727 struct bpf_map *map, *inner_map_meta;
728
729 inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
730 if (IS_ERR(inner_map_meta))
731 return inner_map_meta;
732
733 map = array_map_alloc(attr);
734 if (IS_ERR(map)) {
735 bpf_map_meta_free(inner_map_meta);
736 return map;
737 }
738
739 map->inner_map_meta = inner_map_meta;
740
741 return map;
742}
743
744static void array_of_map_free(struct bpf_map *map)
745{
746 /* map->inner_map_meta is only accessed by syscall which
747 * is protected by fdget/fdput.
748 */
749 bpf_map_meta_free(map->inner_map_meta);
750 bpf_fd_array_map_clear(map);
751 fd_array_map_free(map);
752}
753
754static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
755{
756 struct bpf_map **inner_map = array_map_lookup_elem(map, key);
757
758 if (!inner_map)
759 return NULL;
760
761 return READ_ONCE(*inner_map);
762}
763
764static u32 array_of_map_gen_lookup(struct bpf_map *map,
765 struct bpf_insn *insn_buf)
766{
767 struct bpf_array *array = container_of(map, struct bpf_array, map);
768 u32 elem_size = round_up(map->value_size, 8);
769 struct bpf_insn *insn = insn_buf;
770 const int ret = BPF_REG_0;
771 const int map_ptr = BPF_REG_1;
772 const int index = BPF_REG_2;
773
774 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
775 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
776 if (map->unpriv_array) {
777 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
778 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
779 } else {
780 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
781 }
782 if (is_power_of_2(elem_size))
783 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
784 else
785 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
786 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
787 *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
788 *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
789 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
790 *insn++ = BPF_MOV64_IMM(ret, 0);
791
792 return insn - insn_buf;
793}
794
795const struct bpf_map_ops array_of_maps_map_ops = {
796 .map_alloc_check = fd_array_map_alloc_check,
797 .map_alloc = array_of_map_alloc,
798 .map_free = array_of_map_free,
799 .map_get_next_key = array_map_get_next_key,
800 .map_lookup_elem = array_of_map_lookup_elem,
801 .map_delete_elem = fd_array_map_delete_elem,
802 .map_fd_get_ptr = bpf_map_fd_get_ptr,
803 .map_fd_put_ptr = bpf_map_fd_put_ptr,
804 .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
805 .map_gen_lookup = array_of_map_gen_lookup,
806 .map_check_btf = map_check_no_btf,
807};