Loading...
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
14#include "map_in_map.h"
15
16#define ARRAY_CREATE_FLAG_MASK \
17 (BPF_F_NUMA_NODE | BPF_F_ACCESS_MASK)
18
19static void bpf_array_free_percpu(struct bpf_array *array)
20{
21 int i;
22
23 for (i = 0; i < array->map.max_entries; i++) {
24 free_percpu(array->pptrs[i]);
25 cond_resched();
26 }
27}
28
29static int bpf_array_alloc_percpu(struct bpf_array *array)
30{
31 void __percpu *ptr;
32 int i;
33
34 for (i = 0; i < array->map.max_entries; i++) {
35 ptr = __alloc_percpu_gfp(array->elem_size, 8,
36 GFP_USER | __GFP_NOWARN);
37 if (!ptr) {
38 bpf_array_free_percpu(array);
39 return -ENOMEM;
40 }
41 array->pptrs[i] = ptr;
42 cond_resched();
43 }
44
45 return 0;
46}
47
48/* Called from syscall */
49int array_map_alloc_check(union bpf_attr *attr)
50{
51 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
52 int numa_node = bpf_map_attr_numa_node(attr);
53
54 /* check sanity of attributes */
55 if (attr->max_entries == 0 || attr->key_size != 4 ||
56 attr->value_size == 0 ||
57 attr->map_flags & ~ARRAY_CREATE_FLAG_MASK ||
58 !bpf_map_flags_access_ok(attr->map_flags) ||
59 (percpu && numa_node != NUMA_NO_NODE))
60 return -EINVAL;
61
62 if (attr->value_size > KMALLOC_MAX_SIZE)
63 /* if value_size is bigger, the user space won't be able to
64 * access the elements.
65 */
66 return -E2BIG;
67
68 return 0;
69}
70
71static struct bpf_map *array_map_alloc(union bpf_attr *attr)
72{
73 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
74 int ret, numa_node = bpf_map_attr_numa_node(attr);
75 u32 elem_size, index_mask, max_entries;
76 bool unpriv = !capable(CAP_SYS_ADMIN);
77 u64 cost, array_size, mask64;
78 struct bpf_map_memory mem;
79 struct bpf_array *array;
80
81 elem_size = round_up(attr->value_size, 8);
82
83 max_entries = attr->max_entries;
84
85 /* On 32 bit archs roundup_pow_of_two() with max_entries that has
86 * upper most bit set in u32 space is undefined behavior due to
87 * resulting 1U << 32, so do it manually here in u64 space.
88 */
89 mask64 = fls_long(max_entries - 1);
90 mask64 = 1ULL << mask64;
91 mask64 -= 1;
92
93 index_mask = mask64;
94 if (unpriv) {
95 /* round up array size to nearest power of 2,
96 * since cpu will speculate within index_mask limits
97 */
98 max_entries = index_mask + 1;
99 /* Check for overflows. */
100 if (max_entries < attr->max_entries)
101 return ERR_PTR(-E2BIG);
102 }
103
104 array_size = sizeof(*array);
105 if (percpu)
106 array_size += (u64) max_entries * sizeof(void *);
107 else
108 array_size += (u64) max_entries * elem_size;
109
110 /* make sure there is no u32 overflow later in round_up() */
111 cost = array_size;
112 if (percpu)
113 cost += (u64)attr->max_entries * elem_size * num_possible_cpus();
114
115 ret = bpf_map_charge_init(&mem, cost);
116 if (ret < 0)
117 return ERR_PTR(ret);
118
119 /* allocate all map elements and zero-initialize them */
120 array = bpf_map_area_alloc(array_size, numa_node);
121 if (!array) {
122 bpf_map_charge_finish(&mem);
123 return ERR_PTR(-ENOMEM);
124 }
125 array->index_mask = index_mask;
126 array->map.unpriv_array = unpriv;
127
128 /* copy mandatory map attributes */
129 bpf_map_init_from_attr(&array->map, attr);
130 bpf_map_charge_move(&array->map.memory, &mem);
131 array->elem_size = elem_size;
132
133 if (percpu && bpf_array_alloc_percpu(array)) {
134 bpf_map_charge_finish(&array->map.memory);
135 bpf_map_area_free(array);
136 return ERR_PTR(-ENOMEM);
137 }
138
139 return &array->map;
140}
141
142/* Called from syscall or from eBPF program */
143static void *array_map_lookup_elem(struct bpf_map *map, void *key)
144{
145 struct bpf_array *array = container_of(map, struct bpf_array, map);
146 u32 index = *(u32 *)key;
147
148 if (unlikely(index >= array->map.max_entries))
149 return NULL;
150
151 return array->value + array->elem_size * (index & array->index_mask);
152}
153
154static int array_map_direct_value_addr(const struct bpf_map *map, u64 *imm,
155 u32 off)
156{
157 struct bpf_array *array = container_of(map, struct bpf_array, map);
158
159 if (map->max_entries != 1)
160 return -ENOTSUPP;
161 if (off >= map->value_size)
162 return -EINVAL;
163
164 *imm = (unsigned long)array->value;
165 return 0;
166}
167
168static int array_map_direct_value_meta(const struct bpf_map *map, u64 imm,
169 u32 *off)
170{
171 struct bpf_array *array = container_of(map, struct bpf_array, map);
172 u64 base = (unsigned long)array->value;
173 u64 range = array->elem_size;
174
175 if (map->max_entries != 1)
176 return -ENOTSUPP;
177 if (imm < base || imm >= base + range)
178 return -ENOENT;
179
180 *off = imm - base;
181 return 0;
182}
183
184/* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
185static u32 array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
186{
187 struct bpf_array *array = container_of(map, struct bpf_array, map);
188 struct bpf_insn *insn = insn_buf;
189 u32 elem_size = round_up(map->value_size, 8);
190 const int ret = BPF_REG_0;
191 const int map_ptr = BPF_REG_1;
192 const int index = BPF_REG_2;
193
194 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
195 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
196 if (map->unpriv_array) {
197 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4);
198 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
199 } else {
200 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
201 }
202
203 if (is_power_of_2(elem_size)) {
204 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
205 } else {
206 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
207 }
208 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
209 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
210 *insn++ = BPF_MOV64_IMM(ret, 0);
211 return insn - insn_buf;
212}
213
214/* Called from eBPF program */
215static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
216{
217 struct bpf_array *array = container_of(map, struct bpf_array, map);
218 u32 index = *(u32 *)key;
219
220 if (unlikely(index >= array->map.max_entries))
221 return NULL;
222
223 return this_cpu_ptr(array->pptrs[index & array->index_mask]);
224}
225
226int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
227{
228 struct bpf_array *array = container_of(map, struct bpf_array, map);
229 u32 index = *(u32 *)key;
230 void __percpu *pptr;
231 int cpu, off = 0;
232 u32 size;
233
234 if (unlikely(index >= array->map.max_entries))
235 return -ENOENT;
236
237 /* per_cpu areas are zero-filled and bpf programs can only
238 * access 'value_size' of them, so copying rounded areas
239 * will not leak any kernel data
240 */
241 size = round_up(map->value_size, 8);
242 rcu_read_lock();
243 pptr = array->pptrs[index & array->index_mask];
244 for_each_possible_cpu(cpu) {
245 bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
246 off += size;
247 }
248 rcu_read_unlock();
249 return 0;
250}
251
252/* Called from syscall */
253static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
254{
255 struct bpf_array *array = container_of(map, struct bpf_array, map);
256 u32 index = key ? *(u32 *)key : U32_MAX;
257 u32 *next = (u32 *)next_key;
258
259 if (index >= array->map.max_entries) {
260 *next = 0;
261 return 0;
262 }
263
264 if (index == array->map.max_entries - 1)
265 return -ENOENT;
266
267 *next = index + 1;
268 return 0;
269}
270
271/* Called from syscall or from eBPF program */
272static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
273 u64 map_flags)
274{
275 struct bpf_array *array = container_of(map, struct bpf_array, map);
276 u32 index = *(u32 *)key;
277 char *val;
278
279 if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
280 /* unknown flags */
281 return -EINVAL;
282
283 if (unlikely(index >= array->map.max_entries))
284 /* all elements were pre-allocated, cannot insert a new one */
285 return -E2BIG;
286
287 if (unlikely(map_flags & BPF_NOEXIST))
288 /* all elements already exist */
289 return -EEXIST;
290
291 if (unlikely((map_flags & BPF_F_LOCK) &&
292 !map_value_has_spin_lock(map)))
293 return -EINVAL;
294
295 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
296 memcpy(this_cpu_ptr(array->pptrs[index & array->index_mask]),
297 value, map->value_size);
298 } else {
299 val = array->value +
300 array->elem_size * (index & array->index_mask);
301 if (map_flags & BPF_F_LOCK)
302 copy_map_value_locked(map, val, value, false);
303 else
304 copy_map_value(map, val, value);
305 }
306 return 0;
307}
308
309int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
310 u64 map_flags)
311{
312 struct bpf_array *array = container_of(map, struct bpf_array, map);
313 u32 index = *(u32 *)key;
314 void __percpu *pptr;
315 int cpu, off = 0;
316 u32 size;
317
318 if (unlikely(map_flags > BPF_EXIST))
319 /* unknown flags */
320 return -EINVAL;
321
322 if (unlikely(index >= array->map.max_entries))
323 /* all elements were pre-allocated, cannot insert a new one */
324 return -E2BIG;
325
326 if (unlikely(map_flags == BPF_NOEXIST))
327 /* all elements already exist */
328 return -EEXIST;
329
330 /* the user space will provide round_up(value_size, 8) bytes that
331 * will be copied into per-cpu area. bpf programs can only access
332 * value_size of it. During lookup the same extra bytes will be
333 * returned or zeros which were zero-filled by percpu_alloc,
334 * so no kernel data leaks possible
335 */
336 size = round_up(map->value_size, 8);
337 rcu_read_lock();
338 pptr = array->pptrs[index & array->index_mask];
339 for_each_possible_cpu(cpu) {
340 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
341 off += size;
342 }
343 rcu_read_unlock();
344 return 0;
345}
346
347/* Called from syscall or from eBPF program */
348static int array_map_delete_elem(struct bpf_map *map, void *key)
349{
350 return -EINVAL;
351}
352
353/* Called when map->refcnt goes to zero, either from workqueue or from syscall */
354static void array_map_free(struct bpf_map *map)
355{
356 struct bpf_array *array = container_of(map, struct bpf_array, map);
357
358 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
359 * so the programs (can be more than one that used this map) were
360 * disconnected from events. Wait for outstanding programs to complete
361 * and free the array
362 */
363 synchronize_rcu();
364
365 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
366 bpf_array_free_percpu(array);
367
368 bpf_map_area_free(array);
369}
370
371static void array_map_seq_show_elem(struct bpf_map *map, void *key,
372 struct seq_file *m)
373{
374 void *value;
375
376 rcu_read_lock();
377
378 value = array_map_lookup_elem(map, key);
379 if (!value) {
380 rcu_read_unlock();
381 return;
382 }
383
384 if (map->btf_key_type_id)
385 seq_printf(m, "%u: ", *(u32 *)key);
386 btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
387 seq_puts(m, "\n");
388
389 rcu_read_unlock();
390}
391
392static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key,
393 struct seq_file *m)
394{
395 struct bpf_array *array = container_of(map, struct bpf_array, map);
396 u32 index = *(u32 *)key;
397 void __percpu *pptr;
398 int cpu;
399
400 rcu_read_lock();
401
402 seq_printf(m, "%u: {\n", *(u32 *)key);
403 pptr = array->pptrs[index & array->index_mask];
404 for_each_possible_cpu(cpu) {
405 seq_printf(m, "\tcpu%d: ", cpu);
406 btf_type_seq_show(map->btf, map->btf_value_type_id,
407 per_cpu_ptr(pptr, cpu), m);
408 seq_puts(m, "\n");
409 }
410 seq_puts(m, "}\n");
411
412 rcu_read_unlock();
413}
414
415static int array_map_check_btf(const struct bpf_map *map,
416 const struct btf *btf,
417 const struct btf_type *key_type,
418 const struct btf_type *value_type)
419{
420 u32 int_data;
421
422 /* One exception for keyless BTF: .bss/.data/.rodata map */
423 if (btf_type_is_void(key_type)) {
424 if (map->map_type != BPF_MAP_TYPE_ARRAY ||
425 map->max_entries != 1)
426 return -EINVAL;
427
428 if (BTF_INFO_KIND(value_type->info) != BTF_KIND_DATASEC)
429 return -EINVAL;
430
431 return 0;
432 }
433
434 if (BTF_INFO_KIND(key_type->info) != BTF_KIND_INT)
435 return -EINVAL;
436
437 int_data = *(u32 *)(key_type + 1);
438 /* bpf array can only take a u32 key. This check makes sure
439 * that the btf matches the attr used during map_create.
440 */
441 if (BTF_INT_BITS(int_data) != 32 || BTF_INT_OFFSET(int_data))
442 return -EINVAL;
443
444 return 0;
445}
446
447const struct bpf_map_ops array_map_ops = {
448 .map_alloc_check = array_map_alloc_check,
449 .map_alloc = array_map_alloc,
450 .map_free = array_map_free,
451 .map_get_next_key = array_map_get_next_key,
452 .map_lookup_elem = array_map_lookup_elem,
453 .map_update_elem = array_map_update_elem,
454 .map_delete_elem = array_map_delete_elem,
455 .map_gen_lookup = array_map_gen_lookup,
456 .map_direct_value_addr = array_map_direct_value_addr,
457 .map_direct_value_meta = array_map_direct_value_meta,
458 .map_seq_show_elem = array_map_seq_show_elem,
459 .map_check_btf = array_map_check_btf,
460};
461
462const struct bpf_map_ops percpu_array_map_ops = {
463 .map_alloc_check = array_map_alloc_check,
464 .map_alloc = array_map_alloc,
465 .map_free = array_map_free,
466 .map_get_next_key = array_map_get_next_key,
467 .map_lookup_elem = percpu_array_map_lookup_elem,
468 .map_update_elem = array_map_update_elem,
469 .map_delete_elem = array_map_delete_elem,
470 .map_seq_show_elem = percpu_array_map_seq_show_elem,
471 .map_check_btf = array_map_check_btf,
472};
473
474static int fd_array_map_alloc_check(union bpf_attr *attr)
475{
476 /* only file descriptors can be stored in this type of map */
477 if (attr->value_size != sizeof(u32))
478 return -EINVAL;
479 /* Program read-only/write-only not supported for special maps yet. */
480 if (attr->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG))
481 return -EINVAL;
482 return array_map_alloc_check(attr);
483}
484
485static void fd_array_map_free(struct bpf_map *map)
486{
487 struct bpf_array *array = container_of(map, struct bpf_array, map);
488 int i;
489
490 synchronize_rcu();
491
492 /* make sure it's empty */
493 for (i = 0; i < array->map.max_entries; i++)
494 BUG_ON(array->ptrs[i] != NULL);
495
496 bpf_map_area_free(array);
497}
498
499static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
500{
501 return ERR_PTR(-EOPNOTSUPP);
502}
503
504/* only called from syscall */
505int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
506{
507 void **elem, *ptr;
508 int ret = 0;
509
510 if (!map->ops->map_fd_sys_lookup_elem)
511 return -ENOTSUPP;
512
513 rcu_read_lock();
514 elem = array_map_lookup_elem(map, key);
515 if (elem && (ptr = READ_ONCE(*elem)))
516 *value = map->ops->map_fd_sys_lookup_elem(ptr);
517 else
518 ret = -ENOENT;
519 rcu_read_unlock();
520
521 return ret;
522}
523
524/* only called from syscall */
525int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
526 void *key, void *value, u64 map_flags)
527{
528 struct bpf_array *array = container_of(map, struct bpf_array, map);
529 void *new_ptr, *old_ptr;
530 u32 index = *(u32 *)key, ufd;
531
532 if (map_flags != BPF_ANY)
533 return -EINVAL;
534
535 if (index >= array->map.max_entries)
536 return -E2BIG;
537
538 ufd = *(u32 *)value;
539 new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
540 if (IS_ERR(new_ptr))
541 return PTR_ERR(new_ptr);
542
543 old_ptr = xchg(array->ptrs + index, new_ptr);
544 if (old_ptr)
545 map->ops->map_fd_put_ptr(old_ptr);
546
547 return 0;
548}
549
550static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
551{
552 struct bpf_array *array = container_of(map, struct bpf_array, map);
553 void *old_ptr;
554 u32 index = *(u32 *)key;
555
556 if (index >= array->map.max_entries)
557 return -E2BIG;
558
559 old_ptr = xchg(array->ptrs + index, NULL);
560 if (old_ptr) {
561 map->ops->map_fd_put_ptr(old_ptr);
562 return 0;
563 } else {
564 return -ENOENT;
565 }
566}
567
568static void *prog_fd_array_get_ptr(struct bpf_map *map,
569 struct file *map_file, int fd)
570{
571 struct bpf_array *array = container_of(map, struct bpf_array, map);
572 struct bpf_prog *prog = bpf_prog_get(fd);
573
574 if (IS_ERR(prog))
575 return prog;
576
577 if (!bpf_prog_array_compatible(array, prog)) {
578 bpf_prog_put(prog);
579 return ERR_PTR(-EINVAL);
580 }
581
582 return prog;
583}
584
585static void prog_fd_array_put_ptr(void *ptr)
586{
587 bpf_prog_put(ptr);
588}
589
590static u32 prog_fd_array_sys_lookup_elem(void *ptr)
591{
592 return ((struct bpf_prog *)ptr)->aux->id;
593}
594
595/* decrement refcnt of all bpf_progs that are stored in this map */
596static void bpf_fd_array_map_clear(struct bpf_map *map)
597{
598 struct bpf_array *array = container_of(map, struct bpf_array, map);
599 int i;
600
601 for (i = 0; i < array->map.max_entries; i++)
602 fd_array_map_delete_elem(map, &i);
603}
604
605static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key,
606 struct seq_file *m)
607{
608 void **elem, *ptr;
609 u32 prog_id;
610
611 rcu_read_lock();
612
613 elem = array_map_lookup_elem(map, key);
614 if (elem) {
615 ptr = READ_ONCE(*elem);
616 if (ptr) {
617 seq_printf(m, "%u: ", *(u32 *)key);
618 prog_id = prog_fd_array_sys_lookup_elem(ptr);
619 btf_type_seq_show(map->btf, map->btf_value_type_id,
620 &prog_id, m);
621 seq_puts(m, "\n");
622 }
623 }
624
625 rcu_read_unlock();
626}
627
628const struct bpf_map_ops prog_array_map_ops = {
629 .map_alloc_check = fd_array_map_alloc_check,
630 .map_alloc = array_map_alloc,
631 .map_free = fd_array_map_free,
632 .map_get_next_key = array_map_get_next_key,
633 .map_lookup_elem = fd_array_map_lookup_elem,
634 .map_delete_elem = fd_array_map_delete_elem,
635 .map_fd_get_ptr = prog_fd_array_get_ptr,
636 .map_fd_put_ptr = prog_fd_array_put_ptr,
637 .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
638 .map_release_uref = bpf_fd_array_map_clear,
639 .map_seq_show_elem = prog_array_map_seq_show_elem,
640};
641
642static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
643 struct file *map_file)
644{
645 struct bpf_event_entry *ee;
646
647 ee = kzalloc(sizeof(*ee), GFP_ATOMIC);
648 if (ee) {
649 ee->event = perf_file->private_data;
650 ee->perf_file = perf_file;
651 ee->map_file = map_file;
652 }
653
654 return ee;
655}
656
657static void __bpf_event_entry_free(struct rcu_head *rcu)
658{
659 struct bpf_event_entry *ee;
660
661 ee = container_of(rcu, struct bpf_event_entry, rcu);
662 fput(ee->perf_file);
663 kfree(ee);
664}
665
666static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
667{
668 call_rcu(&ee->rcu, __bpf_event_entry_free);
669}
670
671static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
672 struct file *map_file, int fd)
673{
674 struct bpf_event_entry *ee;
675 struct perf_event *event;
676 struct file *perf_file;
677 u64 value;
678
679 perf_file = perf_event_get(fd);
680 if (IS_ERR(perf_file))
681 return perf_file;
682
683 ee = ERR_PTR(-EOPNOTSUPP);
684 event = perf_file->private_data;
685 if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
686 goto err_out;
687
688 ee = bpf_event_entry_gen(perf_file, map_file);
689 if (ee)
690 return ee;
691 ee = ERR_PTR(-ENOMEM);
692err_out:
693 fput(perf_file);
694 return ee;
695}
696
697static void perf_event_fd_array_put_ptr(void *ptr)
698{
699 bpf_event_entry_free_rcu(ptr);
700}
701
702static void perf_event_fd_array_release(struct bpf_map *map,
703 struct file *map_file)
704{
705 struct bpf_array *array = container_of(map, struct bpf_array, map);
706 struct bpf_event_entry *ee;
707 int i;
708
709 rcu_read_lock();
710 for (i = 0; i < array->map.max_entries; i++) {
711 ee = READ_ONCE(array->ptrs[i]);
712 if (ee && ee->map_file == map_file)
713 fd_array_map_delete_elem(map, &i);
714 }
715 rcu_read_unlock();
716}
717
718const struct bpf_map_ops perf_event_array_map_ops = {
719 .map_alloc_check = fd_array_map_alloc_check,
720 .map_alloc = array_map_alloc,
721 .map_free = fd_array_map_free,
722 .map_get_next_key = array_map_get_next_key,
723 .map_lookup_elem = fd_array_map_lookup_elem,
724 .map_delete_elem = fd_array_map_delete_elem,
725 .map_fd_get_ptr = perf_event_fd_array_get_ptr,
726 .map_fd_put_ptr = perf_event_fd_array_put_ptr,
727 .map_release = perf_event_fd_array_release,
728 .map_check_btf = map_check_no_btf,
729};
730
731#ifdef CONFIG_CGROUPS
732static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
733 struct file *map_file /* not used */,
734 int fd)
735{
736 return cgroup_get_from_fd(fd);
737}
738
739static void cgroup_fd_array_put_ptr(void *ptr)
740{
741 /* cgroup_put free cgrp after a rcu grace period */
742 cgroup_put(ptr);
743}
744
745static void cgroup_fd_array_free(struct bpf_map *map)
746{
747 bpf_fd_array_map_clear(map);
748 fd_array_map_free(map);
749}
750
751const struct bpf_map_ops cgroup_array_map_ops = {
752 .map_alloc_check = fd_array_map_alloc_check,
753 .map_alloc = array_map_alloc,
754 .map_free = cgroup_fd_array_free,
755 .map_get_next_key = array_map_get_next_key,
756 .map_lookup_elem = fd_array_map_lookup_elem,
757 .map_delete_elem = fd_array_map_delete_elem,
758 .map_fd_get_ptr = cgroup_fd_array_get_ptr,
759 .map_fd_put_ptr = cgroup_fd_array_put_ptr,
760 .map_check_btf = map_check_no_btf,
761};
762#endif
763
764static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
765{
766 struct bpf_map *map, *inner_map_meta;
767
768 inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
769 if (IS_ERR(inner_map_meta))
770 return inner_map_meta;
771
772 map = array_map_alloc(attr);
773 if (IS_ERR(map)) {
774 bpf_map_meta_free(inner_map_meta);
775 return map;
776 }
777
778 map->inner_map_meta = inner_map_meta;
779
780 return map;
781}
782
783static void array_of_map_free(struct bpf_map *map)
784{
785 /* map->inner_map_meta is only accessed by syscall which
786 * is protected by fdget/fdput.
787 */
788 bpf_map_meta_free(map->inner_map_meta);
789 bpf_fd_array_map_clear(map);
790 fd_array_map_free(map);
791}
792
793static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
794{
795 struct bpf_map **inner_map = array_map_lookup_elem(map, key);
796
797 if (!inner_map)
798 return NULL;
799
800 return READ_ONCE(*inner_map);
801}
802
803static u32 array_of_map_gen_lookup(struct bpf_map *map,
804 struct bpf_insn *insn_buf)
805{
806 struct bpf_array *array = container_of(map, struct bpf_array, map);
807 u32 elem_size = round_up(map->value_size, 8);
808 struct bpf_insn *insn = insn_buf;
809 const int ret = BPF_REG_0;
810 const int map_ptr = BPF_REG_1;
811 const int index = BPF_REG_2;
812
813 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
814 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
815 if (map->unpriv_array) {
816 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
817 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
818 } else {
819 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
820 }
821 if (is_power_of_2(elem_size))
822 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
823 else
824 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
825 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
826 *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
827 *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
828 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
829 *insn++ = BPF_MOV64_IMM(ret, 0);
830
831 return insn - insn_buf;
832}
833
834const struct bpf_map_ops array_of_maps_map_ops = {
835 .map_alloc_check = fd_array_map_alloc_check,
836 .map_alloc = array_of_map_alloc,
837 .map_free = array_of_map_free,
838 .map_get_next_key = array_map_get_next_key,
839 .map_lookup_elem = array_of_map_lookup_elem,
840 .map_delete_elem = fd_array_map_delete_elem,
841 .map_fd_get_ptr = bpf_map_fd_get_ptr,
842 .map_fd_put_ptr = bpf_map_fd_put_ptr,
843 .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
844 .map_gen_lookup = array_of_map_gen_lookup,
845 .map_check_btf = map_check_no_btf,
846};
1/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2 *
3 * This program is free software; you can redistribute it and/or
4 * modify it under the terms of version 2 of the GNU General Public
5 * License as published by the Free Software Foundation.
6 *
7 * This program is distributed in the hope that it will be useful, but
8 * WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
10 * General Public License for more details.
11 */
12#include <linux/bpf.h>
13#include <linux/err.h>
14#include <linux/slab.h>
15#include <linux/mm.h>
16#include <linux/filter.h>
17#include <linux/perf_event.h>
18
19static void bpf_array_free_percpu(struct bpf_array *array)
20{
21 int i;
22
23 for (i = 0; i < array->map.max_entries; i++)
24 free_percpu(array->pptrs[i]);
25}
26
27static int bpf_array_alloc_percpu(struct bpf_array *array)
28{
29 void __percpu *ptr;
30 int i;
31
32 for (i = 0; i < array->map.max_entries; i++) {
33 ptr = __alloc_percpu_gfp(array->elem_size, 8,
34 GFP_USER | __GFP_NOWARN);
35 if (!ptr) {
36 bpf_array_free_percpu(array);
37 return -ENOMEM;
38 }
39 array->pptrs[i] = ptr;
40 }
41
42 return 0;
43}
44
45/* Called from syscall */
46static struct bpf_map *array_map_alloc(union bpf_attr *attr)
47{
48 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
49 struct bpf_array *array;
50 u64 array_size;
51 u32 elem_size;
52
53 /* check sanity of attributes */
54 if (attr->max_entries == 0 || attr->key_size != 4 ||
55 attr->value_size == 0 || attr->map_flags)
56 return ERR_PTR(-EINVAL);
57
58 if (attr->value_size > KMALLOC_MAX_SIZE)
59 /* if value_size is bigger, the user space won't be able to
60 * access the elements.
61 */
62 return ERR_PTR(-E2BIG);
63
64 elem_size = round_up(attr->value_size, 8);
65
66 array_size = sizeof(*array);
67 if (percpu)
68 array_size += (u64) attr->max_entries * sizeof(void *);
69 else
70 array_size += (u64) attr->max_entries * elem_size;
71
72 /* make sure there is no u32 overflow later in round_up() */
73 if (array_size >= U32_MAX - PAGE_SIZE)
74 return ERR_PTR(-ENOMEM);
75
76 /* allocate all map elements and zero-initialize them */
77 array = bpf_map_area_alloc(array_size);
78 if (!array)
79 return ERR_PTR(-ENOMEM);
80
81 /* copy mandatory map attributes */
82 array->map.map_type = attr->map_type;
83 array->map.key_size = attr->key_size;
84 array->map.value_size = attr->value_size;
85 array->map.max_entries = attr->max_entries;
86 array->elem_size = elem_size;
87
88 if (!percpu)
89 goto out;
90
91 array_size += (u64) attr->max_entries * elem_size * num_possible_cpus();
92
93 if (array_size >= U32_MAX - PAGE_SIZE ||
94 elem_size > PCPU_MIN_UNIT_SIZE || bpf_array_alloc_percpu(array)) {
95 bpf_map_area_free(array);
96 return ERR_PTR(-ENOMEM);
97 }
98out:
99 array->map.pages = round_up(array_size, PAGE_SIZE) >> PAGE_SHIFT;
100
101 return &array->map;
102}
103
104/* Called from syscall or from eBPF program */
105static void *array_map_lookup_elem(struct bpf_map *map, void *key)
106{
107 struct bpf_array *array = container_of(map, struct bpf_array, map);
108 u32 index = *(u32 *)key;
109
110 if (unlikely(index >= array->map.max_entries))
111 return NULL;
112
113 return array->value + array->elem_size * index;
114}
115
116/* Called from eBPF program */
117static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
118{
119 struct bpf_array *array = container_of(map, struct bpf_array, map);
120 u32 index = *(u32 *)key;
121
122 if (unlikely(index >= array->map.max_entries))
123 return NULL;
124
125 return this_cpu_ptr(array->pptrs[index]);
126}
127
128int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
129{
130 struct bpf_array *array = container_of(map, struct bpf_array, map);
131 u32 index = *(u32 *)key;
132 void __percpu *pptr;
133 int cpu, off = 0;
134 u32 size;
135
136 if (unlikely(index >= array->map.max_entries))
137 return -ENOENT;
138
139 /* per_cpu areas are zero-filled and bpf programs can only
140 * access 'value_size' of them, so copying rounded areas
141 * will not leak any kernel data
142 */
143 size = round_up(map->value_size, 8);
144 rcu_read_lock();
145 pptr = array->pptrs[index];
146 for_each_possible_cpu(cpu) {
147 bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
148 off += size;
149 }
150 rcu_read_unlock();
151 return 0;
152}
153
154/* Called from syscall */
155static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
156{
157 struct bpf_array *array = container_of(map, struct bpf_array, map);
158 u32 index = *(u32 *)key;
159 u32 *next = (u32 *)next_key;
160
161 if (index >= array->map.max_entries) {
162 *next = 0;
163 return 0;
164 }
165
166 if (index == array->map.max_entries - 1)
167 return -ENOENT;
168
169 *next = index + 1;
170 return 0;
171}
172
173/* Called from syscall or from eBPF program */
174static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
175 u64 map_flags)
176{
177 struct bpf_array *array = container_of(map, struct bpf_array, map);
178 u32 index = *(u32 *)key;
179
180 if (unlikely(map_flags > BPF_EXIST))
181 /* unknown flags */
182 return -EINVAL;
183
184 if (unlikely(index >= array->map.max_entries))
185 /* all elements were pre-allocated, cannot insert a new one */
186 return -E2BIG;
187
188 if (unlikely(map_flags == BPF_NOEXIST))
189 /* all elements already exist */
190 return -EEXIST;
191
192 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
193 memcpy(this_cpu_ptr(array->pptrs[index]),
194 value, map->value_size);
195 else
196 memcpy(array->value + array->elem_size * index,
197 value, map->value_size);
198 return 0;
199}
200
201int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
202 u64 map_flags)
203{
204 struct bpf_array *array = container_of(map, struct bpf_array, map);
205 u32 index = *(u32 *)key;
206 void __percpu *pptr;
207 int cpu, off = 0;
208 u32 size;
209
210 if (unlikely(map_flags > BPF_EXIST))
211 /* unknown flags */
212 return -EINVAL;
213
214 if (unlikely(index >= array->map.max_entries))
215 /* all elements were pre-allocated, cannot insert a new one */
216 return -E2BIG;
217
218 if (unlikely(map_flags == BPF_NOEXIST))
219 /* all elements already exist */
220 return -EEXIST;
221
222 /* the user space will provide round_up(value_size, 8) bytes that
223 * will be copied into per-cpu area. bpf programs can only access
224 * value_size of it. During lookup the same extra bytes will be
225 * returned or zeros which were zero-filled by percpu_alloc,
226 * so no kernel data leaks possible
227 */
228 size = round_up(map->value_size, 8);
229 rcu_read_lock();
230 pptr = array->pptrs[index];
231 for_each_possible_cpu(cpu) {
232 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
233 off += size;
234 }
235 rcu_read_unlock();
236 return 0;
237}
238
239/* Called from syscall or from eBPF program */
240static int array_map_delete_elem(struct bpf_map *map, void *key)
241{
242 return -EINVAL;
243}
244
245/* Called when map->refcnt goes to zero, either from workqueue or from syscall */
246static void array_map_free(struct bpf_map *map)
247{
248 struct bpf_array *array = container_of(map, struct bpf_array, map);
249
250 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
251 * so the programs (can be more than one that used this map) were
252 * disconnected from events. Wait for outstanding programs to complete
253 * and free the array
254 */
255 synchronize_rcu();
256
257 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
258 bpf_array_free_percpu(array);
259
260 bpf_map_area_free(array);
261}
262
263static const struct bpf_map_ops array_ops = {
264 .map_alloc = array_map_alloc,
265 .map_free = array_map_free,
266 .map_get_next_key = array_map_get_next_key,
267 .map_lookup_elem = array_map_lookup_elem,
268 .map_update_elem = array_map_update_elem,
269 .map_delete_elem = array_map_delete_elem,
270};
271
272static struct bpf_map_type_list array_type __read_mostly = {
273 .ops = &array_ops,
274 .type = BPF_MAP_TYPE_ARRAY,
275};
276
277static const struct bpf_map_ops percpu_array_ops = {
278 .map_alloc = array_map_alloc,
279 .map_free = array_map_free,
280 .map_get_next_key = array_map_get_next_key,
281 .map_lookup_elem = percpu_array_map_lookup_elem,
282 .map_update_elem = array_map_update_elem,
283 .map_delete_elem = array_map_delete_elem,
284};
285
286static struct bpf_map_type_list percpu_array_type __read_mostly = {
287 .ops = &percpu_array_ops,
288 .type = BPF_MAP_TYPE_PERCPU_ARRAY,
289};
290
291static int __init register_array_map(void)
292{
293 bpf_register_map_type(&array_type);
294 bpf_register_map_type(&percpu_array_type);
295 return 0;
296}
297late_initcall(register_array_map);
298
299static struct bpf_map *fd_array_map_alloc(union bpf_attr *attr)
300{
301 /* only file descriptors can be stored in this type of map */
302 if (attr->value_size != sizeof(u32))
303 return ERR_PTR(-EINVAL);
304 return array_map_alloc(attr);
305}
306
307static void fd_array_map_free(struct bpf_map *map)
308{
309 struct bpf_array *array = container_of(map, struct bpf_array, map);
310 int i;
311
312 synchronize_rcu();
313
314 /* make sure it's empty */
315 for (i = 0; i < array->map.max_entries; i++)
316 BUG_ON(array->ptrs[i] != NULL);
317
318 bpf_map_area_free(array);
319}
320
321static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
322{
323 return NULL;
324}
325
326/* only called from syscall */
327int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
328 void *key, void *value, u64 map_flags)
329{
330 struct bpf_array *array = container_of(map, struct bpf_array, map);
331 void *new_ptr, *old_ptr;
332 u32 index = *(u32 *)key, ufd;
333
334 if (map_flags != BPF_ANY)
335 return -EINVAL;
336
337 if (index >= array->map.max_entries)
338 return -E2BIG;
339
340 ufd = *(u32 *)value;
341 new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
342 if (IS_ERR(new_ptr))
343 return PTR_ERR(new_ptr);
344
345 old_ptr = xchg(array->ptrs + index, new_ptr);
346 if (old_ptr)
347 map->ops->map_fd_put_ptr(old_ptr);
348
349 return 0;
350}
351
352static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
353{
354 struct bpf_array *array = container_of(map, struct bpf_array, map);
355 void *old_ptr;
356 u32 index = *(u32 *)key;
357
358 if (index >= array->map.max_entries)
359 return -E2BIG;
360
361 old_ptr = xchg(array->ptrs + index, NULL);
362 if (old_ptr) {
363 map->ops->map_fd_put_ptr(old_ptr);
364 return 0;
365 } else {
366 return -ENOENT;
367 }
368}
369
370static void *prog_fd_array_get_ptr(struct bpf_map *map,
371 struct file *map_file, int fd)
372{
373 struct bpf_array *array = container_of(map, struct bpf_array, map);
374 struct bpf_prog *prog = bpf_prog_get(fd);
375
376 if (IS_ERR(prog))
377 return prog;
378
379 if (!bpf_prog_array_compatible(array, prog)) {
380 bpf_prog_put(prog);
381 return ERR_PTR(-EINVAL);
382 }
383
384 return prog;
385}
386
387static void prog_fd_array_put_ptr(void *ptr)
388{
389 bpf_prog_put(ptr);
390}
391
392/* decrement refcnt of all bpf_progs that are stored in this map */
393void bpf_fd_array_map_clear(struct bpf_map *map)
394{
395 struct bpf_array *array = container_of(map, struct bpf_array, map);
396 int i;
397
398 for (i = 0; i < array->map.max_entries; i++)
399 fd_array_map_delete_elem(map, &i);
400}
401
402static const struct bpf_map_ops prog_array_ops = {
403 .map_alloc = fd_array_map_alloc,
404 .map_free = fd_array_map_free,
405 .map_get_next_key = array_map_get_next_key,
406 .map_lookup_elem = fd_array_map_lookup_elem,
407 .map_delete_elem = fd_array_map_delete_elem,
408 .map_fd_get_ptr = prog_fd_array_get_ptr,
409 .map_fd_put_ptr = prog_fd_array_put_ptr,
410};
411
412static struct bpf_map_type_list prog_array_type __read_mostly = {
413 .ops = &prog_array_ops,
414 .type = BPF_MAP_TYPE_PROG_ARRAY,
415};
416
417static int __init register_prog_array_map(void)
418{
419 bpf_register_map_type(&prog_array_type);
420 return 0;
421}
422late_initcall(register_prog_array_map);
423
424static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
425 struct file *map_file)
426{
427 struct bpf_event_entry *ee;
428
429 ee = kzalloc(sizeof(*ee), GFP_ATOMIC);
430 if (ee) {
431 ee->event = perf_file->private_data;
432 ee->perf_file = perf_file;
433 ee->map_file = map_file;
434 }
435
436 return ee;
437}
438
439static void __bpf_event_entry_free(struct rcu_head *rcu)
440{
441 struct bpf_event_entry *ee;
442
443 ee = container_of(rcu, struct bpf_event_entry, rcu);
444 fput(ee->perf_file);
445 kfree(ee);
446}
447
448static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
449{
450 call_rcu(&ee->rcu, __bpf_event_entry_free);
451}
452
453static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
454 struct file *map_file, int fd)
455{
456 const struct perf_event_attr *attr;
457 struct bpf_event_entry *ee;
458 struct perf_event *event;
459 struct file *perf_file;
460
461 perf_file = perf_event_get(fd);
462 if (IS_ERR(perf_file))
463 return perf_file;
464
465 event = perf_file->private_data;
466 ee = ERR_PTR(-EINVAL);
467
468 attr = perf_event_attrs(event);
469 if (IS_ERR(attr) || attr->inherit)
470 goto err_out;
471
472 switch (attr->type) {
473 case PERF_TYPE_SOFTWARE:
474 if (attr->config != PERF_COUNT_SW_BPF_OUTPUT)
475 goto err_out;
476 /* fall-through */
477 case PERF_TYPE_RAW:
478 case PERF_TYPE_HARDWARE:
479 ee = bpf_event_entry_gen(perf_file, map_file);
480 if (ee)
481 return ee;
482 ee = ERR_PTR(-ENOMEM);
483 /* fall-through */
484 default:
485 break;
486 }
487
488err_out:
489 fput(perf_file);
490 return ee;
491}
492
493static void perf_event_fd_array_put_ptr(void *ptr)
494{
495 bpf_event_entry_free_rcu(ptr);
496}
497
498static void perf_event_fd_array_release(struct bpf_map *map,
499 struct file *map_file)
500{
501 struct bpf_array *array = container_of(map, struct bpf_array, map);
502 struct bpf_event_entry *ee;
503 int i;
504
505 rcu_read_lock();
506 for (i = 0; i < array->map.max_entries; i++) {
507 ee = READ_ONCE(array->ptrs[i]);
508 if (ee && ee->map_file == map_file)
509 fd_array_map_delete_elem(map, &i);
510 }
511 rcu_read_unlock();
512}
513
514static const struct bpf_map_ops perf_event_array_ops = {
515 .map_alloc = fd_array_map_alloc,
516 .map_free = fd_array_map_free,
517 .map_get_next_key = array_map_get_next_key,
518 .map_lookup_elem = fd_array_map_lookup_elem,
519 .map_delete_elem = fd_array_map_delete_elem,
520 .map_fd_get_ptr = perf_event_fd_array_get_ptr,
521 .map_fd_put_ptr = perf_event_fd_array_put_ptr,
522 .map_release = perf_event_fd_array_release,
523};
524
525static struct bpf_map_type_list perf_event_array_type __read_mostly = {
526 .ops = &perf_event_array_ops,
527 .type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
528};
529
530static int __init register_perf_event_array_map(void)
531{
532 bpf_register_map_type(&perf_event_array_type);
533 return 0;
534}
535late_initcall(register_perf_event_array_map);
536
537#ifdef CONFIG_CGROUPS
538static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
539 struct file *map_file /* not used */,
540 int fd)
541{
542 return cgroup_get_from_fd(fd);
543}
544
545static void cgroup_fd_array_put_ptr(void *ptr)
546{
547 /* cgroup_put free cgrp after a rcu grace period */
548 cgroup_put(ptr);
549}
550
551static void cgroup_fd_array_free(struct bpf_map *map)
552{
553 bpf_fd_array_map_clear(map);
554 fd_array_map_free(map);
555}
556
557static const struct bpf_map_ops cgroup_array_ops = {
558 .map_alloc = fd_array_map_alloc,
559 .map_free = cgroup_fd_array_free,
560 .map_get_next_key = array_map_get_next_key,
561 .map_lookup_elem = fd_array_map_lookup_elem,
562 .map_delete_elem = fd_array_map_delete_elem,
563 .map_fd_get_ptr = cgroup_fd_array_get_ptr,
564 .map_fd_put_ptr = cgroup_fd_array_put_ptr,
565};
566
567static struct bpf_map_type_list cgroup_array_type __read_mostly = {
568 .ops = &cgroup_array_ops,
569 .type = BPF_MAP_TYPE_CGROUP_ARRAY,
570};
571
572static int __init register_cgroup_array_map(void)
573{
574 bpf_register_map_type(&cgroup_array_type);
575 return 0;
576}
577late_initcall(register_cgroup_array_map);
578#endif