projects/linux·blob v4.15-rc6 · refs

648 lines · 16.6 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/err.h>
15#include <linux/slab.h>
16#include <linux/mm.h>
17#include <linux/filter.h>
18#include <linux/perf_event.h>
19
20#include "map_in_map.h"
21
22#define ARRAY_CREATE_FLAG_MASK \
23 (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
24
25static void bpf_array_free_percpu(struct bpf_array *array)
26{
27 int i;
28
29 for (i = 0; i < array->map.max_entries; i++)
30 free_percpu(array->pptrs[i]);
31}
32
33static int bpf_array_alloc_percpu(struct bpf_array *array)
34{
35 void __percpu *ptr;
36 int i;
37
38 for (i = 0; i < array->map.max_entries; i++) {
39 ptr = __alloc_percpu_gfp(array->elem_size, 8,
40 GFP_USER | __GFP_NOWARN);
41 if (!ptr) {
42 bpf_array_free_percpu(array);
43 return -ENOMEM;
44 }
45 array->pptrs[i] = ptr;
46 }
47
48 return 0;
49}
50
51/* Called from syscall */
52static struct bpf_map *array_map_alloc(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 struct bpf_array *array;
57 u64 array_size;
58 u32 elem_size;
59
60 /* check sanity of attributes */
61 if (attr->max_entries == 0 || attr->key_size != 4 ||
62 attr->value_size == 0 ||
63 attr->map_flags & ~ARRAY_CREATE_FLAG_MASK ||
64 (percpu && numa_node != NUMA_NO_NODE))
65 return ERR_PTR(-EINVAL);
66
67 if (attr->value_size > KMALLOC_MAX_SIZE)
68 /* if value_size is bigger, the user space won't be able to
69 * access the elements.
70 */
71 return ERR_PTR(-E2BIG);
72
73 elem_size = round_up(attr->value_size, 8);
74
75 array_size = sizeof(*array);
76 if (percpu)
77 array_size += (u64) attr->max_entries * sizeof(void *);
78 else
79 array_size += (u64) attr->max_entries * elem_size;
80
81 /* make sure there is no u32 overflow later in round_up() */
82 if (array_size >= U32_MAX - PAGE_SIZE)
83 return ERR_PTR(-ENOMEM);
84
85 /* allocate all map elements and zero-initialize them */
86 array = bpf_map_area_alloc(array_size, numa_node);
87 if (!array)
88 return ERR_PTR(-ENOMEM);
89
90 /* copy mandatory map attributes */
91 array->map.map_type = attr->map_type;
92 array->map.key_size = attr->key_size;
93 array->map.value_size = attr->value_size;
94 array->map.max_entries = attr->max_entries;
95 array->map.map_flags = attr->map_flags;
96 array->map.numa_node = numa_node;
97 array->elem_size = elem_size;
98
99 if (!percpu)
100 goto out;
101
102 array_size += (u64) attr->max_entries * elem_size * num_possible_cpus();
103
104 if (array_size >= U32_MAX - PAGE_SIZE ||
105 bpf_array_alloc_percpu(array)) {
106 bpf_map_area_free(array);
107 return ERR_PTR(-ENOMEM);
108 }
109out:
110 array->map.pages = round_up(array_size, PAGE_SIZE) >> PAGE_SHIFT;
111
112 return &array->map;
113}
114
115/* Called from syscall or from eBPF program */
116static void *array_map_lookup_elem(struct bpf_map *map, void *key)
117{
118 struct bpf_array *array = container_of(map, struct bpf_array, map);
119 u32 index = *(u32 *)key;
120
121 if (unlikely(index >= array->map.max_entries))
122 return NULL;
123
124 return array->value + array->elem_size * index;
125}
126
127/* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
128static u32 array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
129{
130 struct bpf_insn *insn = insn_buf;
131 u32 elem_size = round_up(map->value_size, 8);
132 const int ret = BPF_REG_0;
133 const int map_ptr = BPF_REG_1;
134 const int index = BPF_REG_2;
135
136 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
137 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
138 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
139
140 if (is_power_of_2(elem_size)) {
141 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
142 } else {
143 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
144 }
145 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
146 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
147 *insn++ = BPF_MOV64_IMM(ret, 0);
148 return insn - insn_buf;
149}
150
151/* Called from eBPF program */
152static void *percpu_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 this_cpu_ptr(array->pptrs[index]);
161}
162
163int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
164{
165 struct bpf_array *array = container_of(map, struct bpf_array, map);
166 u32 index = *(u32 *)key;
167 void __percpu *pptr;
168 int cpu, off = 0;
169 u32 size;
170
171 if (unlikely(index >= array->map.max_entries))
172 return -ENOENT;
173
174 /* per_cpu areas are zero-filled and bpf programs can only
175 * access 'value_size' of them, so copying rounded areas
176 * will not leak any kernel data
177 */
178 size = round_up(map->value_size, 8);
179 rcu_read_lock();
180 pptr = array->pptrs[index];
181 for_each_possible_cpu(cpu) {
182 bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
183 off += size;
184 }
185 rcu_read_unlock();
186 return 0;
187}
188
189/* Called from syscall */
190static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
191{
192 struct bpf_array *array = container_of(map, struct bpf_array, map);
193 u32 index = key ? *(u32 *)key : U32_MAX;
194 u32 *next = (u32 *)next_key;
195
196 if (index >= array->map.max_entries) {
197 *next = 0;
198 return 0;
199 }
200
201 if (index == array->map.max_entries - 1)
202 return -ENOENT;
203
204 *next = index + 1;
205 return 0;
206}
207
208/* Called from syscall or from eBPF program */
209static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
210 u64 map_flags)
211{
212 struct bpf_array *array = container_of(map, struct bpf_array, map);
213 u32 index = *(u32 *)key;
214
215 if (unlikely(map_flags > BPF_EXIST))
216 /* unknown flags */
217 return -EINVAL;
218
219 if (unlikely(index >= array->map.max_entries))
220 /* all elements were pre-allocated, cannot insert a new one */
221 return -E2BIG;
222
223 if (unlikely(map_flags == BPF_NOEXIST))
224 /* all elements already exist */
225 return -EEXIST;
226
227 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
228 memcpy(this_cpu_ptr(array->pptrs[index]),
229 value, map->value_size);
230 else
231 memcpy(array->value + array->elem_size * index,
232 value, map->value_size);
233 return 0;
234}
235
236int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
237 u64 map_flags)
238{
239 struct bpf_array *array = container_of(map, struct bpf_array, map);
240 u32 index = *(u32 *)key;
241 void __percpu *pptr;
242 int cpu, off = 0;
243 u32 size;
244
245 if (unlikely(map_flags > BPF_EXIST))
246 /* unknown flags */
247 return -EINVAL;
248
249 if (unlikely(index >= array->map.max_entries))
250 /* all elements were pre-allocated, cannot insert a new one */
251 return -E2BIG;
252
253 if (unlikely(map_flags == BPF_NOEXIST))
254 /* all elements already exist */
255 return -EEXIST;
256
257 /* the user space will provide round_up(value_size, 8) bytes that
258 * will be copied into per-cpu area. bpf programs can only access
259 * value_size of it. During lookup the same extra bytes will be
260 * returned or zeros which were zero-filled by percpu_alloc,
261 * so no kernel data leaks possible
262 */
263 size = round_up(map->value_size, 8);
264 rcu_read_lock();
265 pptr = array->pptrs[index];
266 for_each_possible_cpu(cpu) {
267 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
268 off += size;
269 }
270 rcu_read_unlock();
271 return 0;
272}
273
274/* Called from syscall or from eBPF program */
275static int array_map_delete_elem(struct bpf_map *map, void *key)
276{
277 return -EINVAL;
278}
279
280/* Called when map->refcnt goes to zero, either from workqueue or from syscall */
281static void array_map_free(struct bpf_map *map)
282{
283 struct bpf_array *array = container_of(map, struct bpf_array, map);
284
285 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
286 * so the programs (can be more than one that used this map) were
287 * disconnected from events. Wait for outstanding programs to complete
288 * and free the array
289 */
290 synchronize_rcu();
291
292 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
293 bpf_array_free_percpu(array);
294
295 bpf_map_area_free(array);
296}
297
298const struct bpf_map_ops array_map_ops = {
299 .map_alloc = array_map_alloc,
300 .map_free = array_map_free,
301 .map_get_next_key = array_map_get_next_key,
302 .map_lookup_elem = array_map_lookup_elem,
303 .map_update_elem = array_map_update_elem,
304 .map_delete_elem = array_map_delete_elem,
305 .map_gen_lookup = array_map_gen_lookup,
306};
307
308const struct bpf_map_ops percpu_array_map_ops = {
309 .map_alloc = array_map_alloc,
310 .map_free = array_map_free,
311 .map_get_next_key = array_map_get_next_key,
312 .map_lookup_elem = percpu_array_map_lookup_elem,
313 .map_update_elem = array_map_update_elem,
314 .map_delete_elem = array_map_delete_elem,
315};
316
317static struct bpf_map *fd_array_map_alloc(union bpf_attr *attr)
318{
319 /* only file descriptors can be stored in this type of map */
320 if (attr->value_size != sizeof(u32))
321 return ERR_PTR(-EINVAL);
322 return array_map_alloc(attr);
323}
324
325static void fd_array_map_free(struct bpf_map *map)
326{
327 struct bpf_array *array = container_of(map, struct bpf_array, map);
328 int i;
329
330 synchronize_rcu();
331
332 /* make sure it's empty */
333 for (i = 0; i < array->map.max_entries; i++)
334 BUG_ON(array->ptrs[i] != NULL);
335
336 bpf_map_area_free(array);
337}
338
339static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
340{
341 return NULL;
342}
343
344/* only called from syscall */
345int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
346{
347 void **elem, *ptr;
348 int ret = 0;
349
350 if (!map->ops->map_fd_sys_lookup_elem)
351 return -ENOTSUPP;
352
353 rcu_read_lock();
354 elem = array_map_lookup_elem(map, key);
355 if (elem && (ptr = READ_ONCE(*elem)))
356 *value = map->ops->map_fd_sys_lookup_elem(ptr);
357 else
358 ret = -ENOENT;
359 rcu_read_unlock();
360
361 return ret;
362}
363
364/* only called from syscall */
365int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
366 void *key, void *value, u64 map_flags)
367{
368 struct bpf_array *array = container_of(map, struct bpf_array, map);
369 void *new_ptr, *old_ptr;
370 u32 index = *(u32 *)key, ufd;
371
372 if (map_flags != BPF_ANY)
373 return -EINVAL;
374
375 if (index >= array->map.max_entries)
376 return -E2BIG;
377
378 ufd = *(u32 *)value;
379 new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
380 if (IS_ERR(new_ptr))
381 return PTR_ERR(new_ptr);
382
383 old_ptr = xchg(array->ptrs + index, new_ptr);
384 if (old_ptr)
385 map->ops->map_fd_put_ptr(old_ptr);
386
387 return 0;
388}
389
390static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
391{
392 struct bpf_array *array = container_of(map, struct bpf_array, map);
393 void *old_ptr;
394 u32 index = *(u32 *)key;
395
396 if (index >= array->map.max_entries)
397 return -E2BIG;
398
399 old_ptr = xchg(array->ptrs + index, NULL);
400 if (old_ptr) {
401 map->ops->map_fd_put_ptr(old_ptr);
402 return 0;
403 } else {
404 return -ENOENT;
405 }
406}
407
408static void *prog_fd_array_get_ptr(struct bpf_map *map,
409 struct file *map_file, int fd)
410{
411 struct bpf_array *array = container_of(map, struct bpf_array, map);
412 struct bpf_prog *prog = bpf_prog_get(fd);
413
414 if (IS_ERR(prog))
415 return prog;
416
417 if (!bpf_prog_array_compatible(array, prog)) {
418 bpf_prog_put(prog);
419 return ERR_PTR(-EINVAL);
420 }
421
422 return prog;
423}
424
425static void prog_fd_array_put_ptr(void *ptr)
426{
427 bpf_prog_put(ptr);
428}
429
430static u32 prog_fd_array_sys_lookup_elem(void *ptr)
431{
432 return ((struct bpf_prog *)ptr)->aux->id;
433}
434
435/* decrement refcnt of all bpf_progs that are stored in this map */
436void bpf_fd_array_map_clear(struct bpf_map *map)
437{
438 struct bpf_array *array = container_of(map, struct bpf_array, map);
439 int i;
440
441 for (i = 0; i < array->map.max_entries; i++)
442 fd_array_map_delete_elem(map, &i);
443}
444
445const struct bpf_map_ops prog_array_map_ops = {
446 .map_alloc = fd_array_map_alloc,
447 .map_free = fd_array_map_free,
448 .map_get_next_key = array_map_get_next_key,
449 .map_lookup_elem = fd_array_map_lookup_elem,
450 .map_delete_elem = fd_array_map_delete_elem,
451 .map_fd_get_ptr = prog_fd_array_get_ptr,
452 .map_fd_put_ptr = prog_fd_array_put_ptr,
453 .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
454};
455
456static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
457 struct file *map_file)
458{
459 struct bpf_event_entry *ee;
460
461 ee = kzalloc(sizeof(*ee), GFP_ATOMIC);
462 if (ee) {
463 ee->event = perf_file->private_data;
464 ee->perf_file = perf_file;
465 ee->map_file = map_file;
466 }
467
468 return ee;
469}
470
471static void __bpf_event_entry_free(struct rcu_head *rcu)
472{
473 struct bpf_event_entry *ee;
474
475 ee = container_of(rcu, struct bpf_event_entry, rcu);
476 fput(ee->perf_file);
477 kfree(ee);
478}
479
480static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
481{
482 call_rcu(&ee->rcu, __bpf_event_entry_free);
483}
484
485static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
486 struct file *map_file, int fd)
487{
488 struct bpf_event_entry *ee;
489 struct perf_event *event;
490 struct file *perf_file;
491 u64 value;
492
493 perf_file = perf_event_get(fd);
494 if (IS_ERR(perf_file))
495 return perf_file;
496
497 ee = ERR_PTR(-EOPNOTSUPP);
498 event = perf_file->private_data;
499 if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
500 goto err_out;
501
502 ee = bpf_event_entry_gen(perf_file, map_file);
503 if (ee)
504 return ee;
505 ee = ERR_PTR(-ENOMEM);
506err_out:
507 fput(perf_file);
508 return ee;
509}
510
511static void perf_event_fd_array_put_ptr(void *ptr)
512{
513 bpf_event_entry_free_rcu(ptr);
514}
515
516static void perf_event_fd_array_release(struct bpf_map *map,
517 struct file *map_file)
518{
519 struct bpf_array *array = container_of(map, struct bpf_array, map);
520 struct bpf_event_entry *ee;
521 int i;
522
523 rcu_read_lock();
524 for (i = 0; i < array->map.max_entries; i++) {
525 ee = READ_ONCE(array->ptrs[i]);
526 if (ee && ee->map_file == map_file)
527 fd_array_map_delete_elem(map, &i);
528 }
529 rcu_read_unlock();
530}
531
532const struct bpf_map_ops perf_event_array_map_ops = {
533 .map_alloc = fd_array_map_alloc,
534 .map_free = fd_array_map_free,
535 .map_get_next_key = array_map_get_next_key,
536 .map_lookup_elem = fd_array_map_lookup_elem,
537 .map_delete_elem = fd_array_map_delete_elem,
538 .map_fd_get_ptr = perf_event_fd_array_get_ptr,
539 .map_fd_put_ptr = perf_event_fd_array_put_ptr,
540 .map_release = perf_event_fd_array_release,
541};
542
543#ifdef CONFIG_CGROUPS
544static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
545 struct file *map_file /* not used */,
546 int fd)
547{
548 return cgroup_get_from_fd(fd);
549}
550
551static void cgroup_fd_array_put_ptr(void *ptr)
552{
553 /* cgroup_put free cgrp after a rcu grace period */
554 cgroup_put(ptr);
555}
556
557static void cgroup_fd_array_free(struct bpf_map *map)
558{
559 bpf_fd_array_map_clear(map);
560 fd_array_map_free(map);
561}
562
563const struct bpf_map_ops cgroup_array_map_ops = {
564 .map_alloc = fd_array_map_alloc,
565 .map_free = cgroup_fd_array_free,
566 .map_get_next_key = array_map_get_next_key,
567 .map_lookup_elem = fd_array_map_lookup_elem,
568 .map_delete_elem = fd_array_map_delete_elem,
569 .map_fd_get_ptr = cgroup_fd_array_get_ptr,
570 .map_fd_put_ptr = cgroup_fd_array_put_ptr,
571};
572#endif
573
574static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
575{
576 struct bpf_map *map, *inner_map_meta;
577
578 inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
579 if (IS_ERR(inner_map_meta))
580 return inner_map_meta;
581
582 map = fd_array_map_alloc(attr);
583 if (IS_ERR(map)) {
584 bpf_map_meta_free(inner_map_meta);
585 return map;
586 }
587
588 map->inner_map_meta = inner_map_meta;
589
590 return map;
591}
592
593static void array_of_map_free(struct bpf_map *map)
594{
595 /* map->inner_map_meta is only accessed by syscall which
596 * is protected by fdget/fdput.
597 */
598 bpf_map_meta_free(map->inner_map_meta);
599 bpf_fd_array_map_clear(map);
600 fd_array_map_free(map);
601}
602
603static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
604{
605 struct bpf_map **inner_map = array_map_lookup_elem(map, key);
606
607 if (!inner_map)
608 return NULL;
609
610 return READ_ONCE(*inner_map);
611}
612
613static u32 array_of_map_gen_lookup(struct bpf_map *map,
614 struct bpf_insn *insn_buf)
615{
616 u32 elem_size = round_up(map->value_size, 8);
617 struct bpf_insn *insn = insn_buf;
618 const int ret = BPF_REG_0;
619 const int map_ptr = BPF_REG_1;
620 const int index = BPF_REG_2;
621
622 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
623 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
624 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
625 if (is_power_of_2(elem_size))
626 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
627 else
628 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
629 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
630 *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
631 *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
632 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
633 *insn++ = BPF_MOV64_IMM(ret, 0);
634
635 return insn - insn_buf;
636}
637
638const struct bpf_map_ops array_of_maps_map_ops = {
639 .map_alloc = array_of_map_alloc,
640 .map_free = array_of_map_free,
641 .map_get_next_key = array_map_get_next_key,
642 .map_lookup_elem = array_of_map_lookup_elem,
643 .map_delete_elem = fd_array_map_delete_elem,
644 .map_fd_get_ptr = bpf_map_fd_get_ptr,
645 .map_fd_put_ptr = bpf_map_fd_put_ptr,
646 .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
647 .map_gen_lookup = array_of_map_gen_lookup,
648};