Loading...
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/vmalloc.h>
15#include <linux/slab.h>
16#include <linux/mm.h>
17#include <linux/filter.h>
18#include <linux/perf_event.h>
19
20static void bpf_array_free_percpu(struct bpf_array *array)
21{
22 int i;
23
24 for (i = 0; i < array->map.max_entries; i++)
25 free_percpu(array->pptrs[i]);
26}
27
28static int bpf_array_alloc_percpu(struct bpf_array *array)
29{
30 void __percpu *ptr;
31 int i;
32
33 for (i = 0; i < array->map.max_entries; i++) {
34 ptr = __alloc_percpu_gfp(array->elem_size, 8,
35 GFP_USER | __GFP_NOWARN);
36 if (!ptr) {
37 bpf_array_free_percpu(array);
38 return -ENOMEM;
39 }
40 array->pptrs[i] = ptr;
41 }
42
43 return 0;
44}
45
46/* Called from syscall */
47static struct bpf_map *array_map_alloc(union bpf_attr *attr)
48{
49 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
50 struct bpf_array *array;
51 u64 array_size;
52 u32 elem_size;
53
54 /* check sanity of attributes */
55 if (attr->max_entries == 0 || attr->key_size != 4 ||
56 attr->value_size == 0 || attr->map_flags)
57 return ERR_PTR(-EINVAL);
58
59 if (attr->value_size >= 1 << (KMALLOC_SHIFT_MAX - 1))
60 /* if value_size is bigger, the user space won't be able to
61 * access the elements.
62 */
63 return ERR_PTR(-E2BIG);
64
65 elem_size = round_up(attr->value_size, 8);
66
67 array_size = sizeof(*array);
68 if (percpu)
69 array_size += (u64) attr->max_entries * sizeof(void *);
70 else
71 array_size += (u64) attr->max_entries * elem_size;
72
73 /* make sure there is no u32 overflow later in round_up() */
74 if (array_size >= U32_MAX - PAGE_SIZE)
75 return ERR_PTR(-ENOMEM);
76
77
78 /* allocate all map elements and zero-initialize them */
79 array = kzalloc(array_size, GFP_USER | __GFP_NOWARN);
80 if (!array) {
81 array = vzalloc(array_size);
82 if (!array)
83 return ERR_PTR(-ENOMEM);
84 }
85
86 /* copy mandatory map attributes */
87 array->map.map_type = attr->map_type;
88 array->map.key_size = attr->key_size;
89 array->map.value_size = attr->value_size;
90 array->map.max_entries = attr->max_entries;
91 array->elem_size = elem_size;
92
93 if (!percpu)
94 goto out;
95
96 array_size += (u64) attr->max_entries * elem_size * num_possible_cpus();
97
98 if (array_size >= U32_MAX - PAGE_SIZE ||
99 elem_size > PCPU_MIN_UNIT_SIZE || bpf_array_alloc_percpu(array)) {
100 kvfree(array);
101 return ERR_PTR(-ENOMEM);
102 }
103out:
104 array->map.pages = round_up(array_size, PAGE_SIZE) >> PAGE_SHIFT;
105
106 return &array->map;
107}
108
109/* Called from syscall or from eBPF program */
110static void *array_map_lookup_elem(struct bpf_map *map, void *key)
111{
112 struct bpf_array *array = container_of(map, struct bpf_array, map);
113 u32 index = *(u32 *)key;
114
115 if (unlikely(index >= array->map.max_entries))
116 return NULL;
117
118 return array->value + array->elem_size * index;
119}
120
121/* Called from eBPF program */
122static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
123{
124 struct bpf_array *array = container_of(map, struct bpf_array, map);
125 u32 index = *(u32 *)key;
126
127 if (unlikely(index >= array->map.max_entries))
128 return NULL;
129
130 return this_cpu_ptr(array->pptrs[index]);
131}
132
133int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
134{
135 struct bpf_array *array = container_of(map, struct bpf_array, map);
136 u32 index = *(u32 *)key;
137 void __percpu *pptr;
138 int cpu, off = 0;
139 u32 size;
140
141 if (unlikely(index >= array->map.max_entries))
142 return -ENOENT;
143
144 /* per_cpu areas are zero-filled and bpf programs can only
145 * access 'value_size' of them, so copying rounded areas
146 * will not leak any kernel data
147 */
148 size = round_up(map->value_size, 8);
149 rcu_read_lock();
150 pptr = array->pptrs[index];
151 for_each_possible_cpu(cpu) {
152 bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
153 off += size;
154 }
155 rcu_read_unlock();
156 return 0;
157}
158
159/* Called from syscall */
160static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
161{
162 struct bpf_array *array = container_of(map, struct bpf_array, map);
163 u32 index = *(u32 *)key;
164 u32 *next = (u32 *)next_key;
165
166 if (index >= array->map.max_entries) {
167 *next = 0;
168 return 0;
169 }
170
171 if (index == array->map.max_entries - 1)
172 return -ENOENT;
173
174 *next = index + 1;
175 return 0;
176}
177
178/* Called from syscall or from eBPF program */
179static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
180 u64 map_flags)
181{
182 struct bpf_array *array = container_of(map, struct bpf_array, map);
183 u32 index = *(u32 *)key;
184
185 if (unlikely(map_flags > BPF_EXIST))
186 /* unknown flags */
187 return -EINVAL;
188
189 if (unlikely(index >= array->map.max_entries))
190 /* all elements were pre-allocated, cannot insert a new one */
191 return -E2BIG;
192
193 if (unlikely(map_flags == BPF_NOEXIST))
194 /* all elements already exist */
195 return -EEXIST;
196
197 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
198 memcpy(this_cpu_ptr(array->pptrs[index]),
199 value, map->value_size);
200 else
201 memcpy(array->value + array->elem_size * index,
202 value, map->value_size);
203 return 0;
204}
205
206int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
207 u64 map_flags)
208{
209 struct bpf_array *array = container_of(map, struct bpf_array, map);
210 u32 index = *(u32 *)key;
211 void __percpu *pptr;
212 int cpu, off = 0;
213 u32 size;
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 /* the user space will provide round_up(value_size, 8) bytes that
228 * will be copied into per-cpu area. bpf programs can only access
229 * value_size of it. During lookup the same extra bytes will be
230 * returned or zeros which were zero-filled by percpu_alloc,
231 * so no kernel data leaks possible
232 */
233 size = round_up(map->value_size, 8);
234 rcu_read_lock();
235 pptr = array->pptrs[index];
236 for_each_possible_cpu(cpu) {
237 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
238 off += size;
239 }
240 rcu_read_unlock();
241 return 0;
242}
243
244/* Called from syscall or from eBPF program */
245static int array_map_delete_elem(struct bpf_map *map, void *key)
246{
247 return -EINVAL;
248}
249
250/* Called when map->refcnt goes to zero, either from workqueue or from syscall */
251static void array_map_free(struct bpf_map *map)
252{
253 struct bpf_array *array = container_of(map, struct bpf_array, map);
254
255 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
256 * so the programs (can be more than one that used this map) were
257 * disconnected from events. Wait for outstanding programs to complete
258 * and free the array
259 */
260 synchronize_rcu();
261
262 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
263 bpf_array_free_percpu(array);
264
265 kvfree(array);
266}
267
268static const struct bpf_map_ops array_ops = {
269 .map_alloc = array_map_alloc,
270 .map_free = array_map_free,
271 .map_get_next_key = array_map_get_next_key,
272 .map_lookup_elem = array_map_lookup_elem,
273 .map_update_elem = array_map_update_elem,
274 .map_delete_elem = array_map_delete_elem,
275};
276
277static struct bpf_map_type_list array_type __read_mostly = {
278 .ops = &array_ops,
279 .type = BPF_MAP_TYPE_ARRAY,
280};
281
282static const struct bpf_map_ops percpu_array_ops = {
283 .map_alloc = array_map_alloc,
284 .map_free = array_map_free,
285 .map_get_next_key = array_map_get_next_key,
286 .map_lookup_elem = percpu_array_map_lookup_elem,
287 .map_update_elem = array_map_update_elem,
288 .map_delete_elem = array_map_delete_elem,
289};
290
291static struct bpf_map_type_list percpu_array_type __read_mostly = {
292 .ops = &percpu_array_ops,
293 .type = BPF_MAP_TYPE_PERCPU_ARRAY,
294};
295
296static int __init register_array_map(void)
297{
298 bpf_register_map_type(&array_type);
299 bpf_register_map_type(&percpu_array_type);
300 return 0;
301}
302late_initcall(register_array_map);
303
304static struct bpf_map *fd_array_map_alloc(union bpf_attr *attr)
305{
306 /* only file descriptors can be stored in this type of map */
307 if (attr->value_size != sizeof(u32))
308 return ERR_PTR(-EINVAL);
309 return array_map_alloc(attr);
310}
311
312static void fd_array_map_free(struct bpf_map *map)
313{
314 struct bpf_array *array = container_of(map, struct bpf_array, map);
315 int i;
316
317 synchronize_rcu();
318
319 /* make sure it's empty */
320 for (i = 0; i < array->map.max_entries; i++)
321 BUG_ON(array->ptrs[i] != NULL);
322 kvfree(array);
323}
324
325static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
326{
327 return NULL;
328}
329
330/* only called from syscall */
331static int fd_array_map_update_elem(struct bpf_map *map, void *key,
332 void *value, u64 map_flags)
333{
334 struct bpf_array *array = container_of(map, struct bpf_array, map);
335 void *new_ptr, *old_ptr;
336 u32 index = *(u32 *)key, ufd;
337
338 if (map_flags != BPF_ANY)
339 return -EINVAL;
340
341 if (index >= array->map.max_entries)
342 return -E2BIG;
343
344 ufd = *(u32 *)value;
345 new_ptr = map->ops->map_fd_get_ptr(map, ufd);
346 if (IS_ERR(new_ptr))
347 return PTR_ERR(new_ptr);
348
349 old_ptr = xchg(array->ptrs + index, new_ptr);
350 if (old_ptr)
351 map->ops->map_fd_put_ptr(old_ptr);
352
353 return 0;
354}
355
356static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
357{
358 struct bpf_array *array = container_of(map, struct bpf_array, map);
359 void *old_ptr;
360 u32 index = *(u32 *)key;
361
362 if (index >= array->map.max_entries)
363 return -E2BIG;
364
365 old_ptr = xchg(array->ptrs + index, NULL);
366 if (old_ptr) {
367 map->ops->map_fd_put_ptr(old_ptr);
368 return 0;
369 } else {
370 return -ENOENT;
371 }
372}
373
374static void *prog_fd_array_get_ptr(struct bpf_map *map, int fd)
375{
376 struct bpf_array *array = container_of(map, struct bpf_array, map);
377 struct bpf_prog *prog = bpf_prog_get(fd);
378 if (IS_ERR(prog))
379 return prog;
380
381 if (!bpf_prog_array_compatible(array, prog)) {
382 bpf_prog_put(prog);
383 return ERR_PTR(-EINVAL);
384 }
385 return prog;
386}
387
388static void prog_fd_array_put_ptr(void *ptr)
389{
390 struct bpf_prog *prog = ptr;
391
392 bpf_prog_put_rcu(prog);
393}
394
395/* decrement refcnt of all bpf_progs that are stored in this map */
396void bpf_fd_array_map_clear(struct bpf_map *map)
397{
398 struct bpf_array *array = container_of(map, struct bpf_array, map);
399 int i;
400
401 for (i = 0; i < array->map.max_entries; i++)
402 fd_array_map_delete_elem(map, &i);
403}
404
405static const struct bpf_map_ops prog_array_ops = {
406 .map_alloc = fd_array_map_alloc,
407 .map_free = fd_array_map_free,
408 .map_get_next_key = array_map_get_next_key,
409 .map_lookup_elem = fd_array_map_lookup_elem,
410 .map_update_elem = fd_array_map_update_elem,
411 .map_delete_elem = fd_array_map_delete_elem,
412 .map_fd_get_ptr = prog_fd_array_get_ptr,
413 .map_fd_put_ptr = prog_fd_array_put_ptr,
414};
415
416static struct bpf_map_type_list prog_array_type __read_mostly = {
417 .ops = &prog_array_ops,
418 .type = BPF_MAP_TYPE_PROG_ARRAY,
419};
420
421static int __init register_prog_array_map(void)
422{
423 bpf_register_map_type(&prog_array_type);
424 return 0;
425}
426late_initcall(register_prog_array_map);
427
428static void perf_event_array_map_free(struct bpf_map *map)
429{
430 bpf_fd_array_map_clear(map);
431 fd_array_map_free(map);
432}
433
434static void *perf_event_fd_array_get_ptr(struct bpf_map *map, int fd)
435{
436 struct perf_event *event;
437 const struct perf_event_attr *attr;
438 struct file *file;
439
440 file = perf_event_get(fd);
441 if (IS_ERR(file))
442 return file;
443
444 event = file->private_data;
445
446 attr = perf_event_attrs(event);
447 if (IS_ERR(attr))
448 goto err;
449
450 if (attr->inherit)
451 goto err;
452
453 if (attr->type == PERF_TYPE_RAW)
454 return file;
455
456 if (attr->type == PERF_TYPE_HARDWARE)
457 return file;
458
459 if (attr->type == PERF_TYPE_SOFTWARE &&
460 attr->config == PERF_COUNT_SW_BPF_OUTPUT)
461 return file;
462err:
463 fput(file);
464 return ERR_PTR(-EINVAL);
465}
466
467static void perf_event_fd_array_put_ptr(void *ptr)
468{
469 fput((struct file *)ptr);
470}
471
472static const struct bpf_map_ops perf_event_array_ops = {
473 .map_alloc = fd_array_map_alloc,
474 .map_free = perf_event_array_map_free,
475 .map_get_next_key = array_map_get_next_key,
476 .map_lookup_elem = fd_array_map_lookup_elem,
477 .map_update_elem = fd_array_map_update_elem,
478 .map_delete_elem = fd_array_map_delete_elem,
479 .map_fd_get_ptr = perf_event_fd_array_get_ptr,
480 .map_fd_put_ptr = perf_event_fd_array_put_ptr,
481};
482
483static struct bpf_map_type_list perf_event_array_type __read_mostly = {
484 .ops = &perf_event_array_ops,
485 .type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
486};
487
488static int __init register_perf_event_array_map(void)
489{
490 bpf_register_map_type(&perf_event_array_type);
491 return 0;
492}
493late_initcall(register_perf_event_array_map);
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
22static 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
32static 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 */
52int 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 /* percpu map value size is bound by PCPU_MIN_UNIT_SIZE */
77 if (percpu && round_up(attr->value_size, 8) > PCPU_MIN_UNIT_SIZE)
78 return -E2BIG;
79
80 return 0;
81}
82
83static struct bpf_map *array_map_alloc(union bpf_attr *attr)
84{
85 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
86 int numa_node = bpf_map_attr_numa_node(attr);
87 u32 elem_size, index_mask, max_entries;
88 bool bypass_spec_v1 = bpf_bypass_spec_v1(NULL);
89 u64 array_size, mask64;
90 struct bpf_array *array;
91
92 elem_size = round_up(attr->value_size, 8);
93
94 max_entries = attr->max_entries;
95
96 /* On 32 bit archs roundup_pow_of_two() with max_entries that has
97 * upper most bit set in u32 space is undefined behavior due to
98 * resulting 1U << 32, so do it manually here in u64 space.
99 */
100 mask64 = fls_long(max_entries - 1);
101 mask64 = 1ULL << mask64;
102 mask64 -= 1;
103
104 index_mask = mask64;
105 if (!bypass_spec_v1) {
106 /* round up array size to nearest power of 2,
107 * since cpu will speculate within index_mask limits
108 */
109 max_entries = index_mask + 1;
110 /* Check for overflows. */
111 if (max_entries < attr->max_entries)
112 return ERR_PTR(-E2BIG);
113 }
114
115 array_size = sizeof(*array);
116 if (percpu) {
117 array_size += (u64) max_entries * sizeof(void *);
118 } else {
119 /* rely on vmalloc() to return page-aligned memory and
120 * ensure array->value is exactly page-aligned
121 */
122 if (attr->map_flags & BPF_F_MMAPABLE) {
123 array_size = PAGE_ALIGN(array_size);
124 array_size += PAGE_ALIGN((u64) max_entries * elem_size);
125 } else {
126 array_size += (u64) max_entries * elem_size;
127 }
128 }
129
130 /* allocate all map elements and zero-initialize them */
131 if (attr->map_flags & BPF_F_MMAPABLE) {
132 void *data;
133
134 /* kmalloc'ed memory can't be mmap'ed, use explicit vmalloc */
135 data = bpf_map_area_mmapable_alloc(array_size, numa_node);
136 if (!data)
137 return ERR_PTR(-ENOMEM);
138 array = data + PAGE_ALIGN(sizeof(struct bpf_array))
139 - offsetof(struct bpf_array, value);
140 } else {
141 array = bpf_map_area_alloc(array_size, numa_node);
142 }
143 if (!array)
144 return ERR_PTR(-ENOMEM);
145 array->index_mask = index_mask;
146 array->map.bypass_spec_v1 = bypass_spec_v1;
147
148 /* copy mandatory map attributes */
149 bpf_map_init_from_attr(&array->map, attr);
150 array->elem_size = elem_size;
151
152 if (percpu && bpf_array_alloc_percpu(array)) {
153 bpf_map_area_free(array);
154 return ERR_PTR(-ENOMEM);
155 }
156
157 return &array->map;
158}
159
160static void *array_map_elem_ptr(struct bpf_array* array, u32 index)
161{
162 return array->value + (u64)array->elem_size * index;
163}
164
165/* Called from syscall or from eBPF program */
166static void *array_map_lookup_elem(struct bpf_map *map, void *key)
167{
168 struct bpf_array *array = container_of(map, struct bpf_array, map);
169 u32 index = *(u32 *)key;
170
171 if (unlikely(index >= array->map.max_entries))
172 return NULL;
173
174 return array->value + (u64)array->elem_size * (index & array->index_mask);
175}
176
177static int array_map_direct_value_addr(const struct bpf_map *map, u64 *imm,
178 u32 off)
179{
180 struct bpf_array *array = container_of(map, struct bpf_array, map);
181
182 if (map->max_entries != 1)
183 return -ENOTSUPP;
184 if (off >= map->value_size)
185 return -EINVAL;
186
187 *imm = (unsigned long)array->value;
188 return 0;
189}
190
191static int array_map_direct_value_meta(const struct bpf_map *map, u64 imm,
192 u32 *off)
193{
194 struct bpf_array *array = container_of(map, struct bpf_array, map);
195 u64 base = (unsigned long)array->value;
196 u64 range = array->elem_size;
197
198 if (map->max_entries != 1)
199 return -ENOTSUPP;
200 if (imm < base || imm >= base + range)
201 return -ENOENT;
202
203 *off = imm - base;
204 return 0;
205}
206
207/* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
208static int array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
209{
210 struct bpf_array *array = container_of(map, struct bpf_array, map);
211 struct bpf_insn *insn = insn_buf;
212 u32 elem_size = array->elem_size;
213 const int ret = BPF_REG_0;
214 const int map_ptr = BPF_REG_1;
215 const int index = BPF_REG_2;
216
217 if (map->map_flags & BPF_F_INNER_MAP)
218 return -EOPNOTSUPP;
219
220 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
221 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
222 if (!map->bypass_spec_v1) {
223 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4);
224 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
225 } else {
226 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
227 }
228
229 if (is_power_of_2(elem_size)) {
230 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
231 } else {
232 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
233 }
234 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
235 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
236 *insn++ = BPF_MOV64_IMM(ret, 0);
237 return insn - insn_buf;
238}
239
240/* Called from eBPF program */
241static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
242{
243 struct bpf_array *array = container_of(map, struct bpf_array, map);
244 u32 index = *(u32 *)key;
245
246 if (unlikely(index >= array->map.max_entries))
247 return NULL;
248
249 return this_cpu_ptr(array->pptrs[index & array->index_mask]);
250}
251
252/* emit BPF instructions equivalent to C code of percpu_array_map_lookup_elem() */
253static int percpu_array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
254{
255 struct bpf_array *array = container_of(map, struct bpf_array, map);
256 struct bpf_insn *insn = insn_buf;
257
258 if (!bpf_jit_supports_percpu_insn())
259 return -EOPNOTSUPP;
260
261 if (map->map_flags & BPF_F_INNER_MAP)
262 return -EOPNOTSUPP;
263
264 BUILD_BUG_ON(offsetof(struct bpf_array, map) != 0);
265 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, offsetof(struct bpf_array, pptrs));
266
267 *insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_0, BPF_REG_2, 0);
268 if (!map->bypass_spec_v1) {
269 *insn++ = BPF_JMP_IMM(BPF_JGE, BPF_REG_0, map->max_entries, 6);
270 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_0, array->index_mask);
271 } else {
272 *insn++ = BPF_JMP_IMM(BPF_JGE, BPF_REG_0, map->max_entries, 5);
273 }
274
275 *insn++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_0, 3);
276 *insn++ = BPF_ALU64_REG(BPF_ADD, BPF_REG_0, BPF_REG_1);
277 *insn++ = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_0, 0);
278 *insn++ = BPF_MOV64_PERCPU_REG(BPF_REG_0, BPF_REG_0);
279 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
280 *insn++ = BPF_MOV64_IMM(BPF_REG_0, 0);
281 return insn - insn_buf;
282}
283
284static void *percpu_array_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
285{
286 struct bpf_array *array = container_of(map, struct bpf_array, map);
287 u32 index = *(u32 *)key;
288
289 if (cpu >= nr_cpu_ids)
290 return NULL;
291
292 if (unlikely(index >= array->map.max_entries))
293 return NULL;
294
295 return per_cpu_ptr(array->pptrs[index & array->index_mask], cpu);
296}
297
298int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
299{
300 struct bpf_array *array = container_of(map, struct bpf_array, map);
301 u32 index = *(u32 *)key;
302 void __percpu *pptr;
303 int cpu, off = 0;
304 u32 size;
305
306 if (unlikely(index >= array->map.max_entries))
307 return -ENOENT;
308
309 /* per_cpu areas are zero-filled and bpf programs can only
310 * access 'value_size' of them, so copying rounded areas
311 * will not leak any kernel data
312 */
313 size = array->elem_size;
314 rcu_read_lock();
315 pptr = array->pptrs[index & array->index_mask];
316 for_each_possible_cpu(cpu) {
317 copy_map_value_long(map, value + off, per_cpu_ptr(pptr, cpu));
318 check_and_init_map_value(map, value + off);
319 off += size;
320 }
321 rcu_read_unlock();
322 return 0;
323}
324
325/* Called from syscall */
326static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
327{
328 struct bpf_array *array = container_of(map, struct bpf_array, map);
329 u32 index = key ? *(u32 *)key : U32_MAX;
330 u32 *next = (u32 *)next_key;
331
332 if (index >= array->map.max_entries) {
333 *next = 0;
334 return 0;
335 }
336
337 if (index == array->map.max_entries - 1)
338 return -ENOENT;
339
340 *next = index + 1;
341 return 0;
342}
343
344/* Called from syscall or from eBPF program */
345static long array_map_update_elem(struct bpf_map *map, void *key, void *value,
346 u64 map_flags)
347{
348 struct bpf_array *array = container_of(map, struct bpf_array, map);
349 u32 index = *(u32 *)key;
350 char *val;
351
352 if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
353 /* unknown flags */
354 return -EINVAL;
355
356 if (unlikely(index >= array->map.max_entries))
357 /* all elements were pre-allocated, cannot insert a new one */
358 return -E2BIG;
359
360 if (unlikely(map_flags & BPF_NOEXIST))
361 /* all elements already exist */
362 return -EEXIST;
363
364 if (unlikely((map_flags & BPF_F_LOCK) &&
365 !btf_record_has_field(map->record, BPF_SPIN_LOCK)))
366 return -EINVAL;
367
368 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
369 val = this_cpu_ptr(array->pptrs[index & array->index_mask]);
370 copy_map_value(map, val, value);
371 bpf_obj_free_fields(array->map.record, val);
372 } else {
373 val = array->value +
374 (u64)array->elem_size * (index & array->index_mask);
375 if (map_flags & BPF_F_LOCK)
376 copy_map_value_locked(map, val, value, false);
377 else
378 copy_map_value(map, val, value);
379 bpf_obj_free_fields(array->map.record, val);
380 }
381 return 0;
382}
383
384int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
385 u64 map_flags)
386{
387 struct bpf_array *array = container_of(map, struct bpf_array, map);
388 u32 index = *(u32 *)key;
389 void __percpu *pptr;
390 int cpu, off = 0;
391 u32 size;
392
393 if (unlikely(map_flags > BPF_EXIST))
394 /* unknown flags */
395 return -EINVAL;
396
397 if (unlikely(index >= array->map.max_entries))
398 /* all elements were pre-allocated, cannot insert a new one */
399 return -E2BIG;
400
401 if (unlikely(map_flags == BPF_NOEXIST))
402 /* all elements already exist */
403 return -EEXIST;
404
405 /* the user space will provide round_up(value_size, 8) bytes that
406 * will be copied into per-cpu area. bpf programs can only access
407 * value_size of it. During lookup the same extra bytes will be
408 * returned or zeros which were zero-filled by percpu_alloc,
409 * so no kernel data leaks possible
410 */
411 size = array->elem_size;
412 rcu_read_lock();
413 pptr = array->pptrs[index & array->index_mask];
414 for_each_possible_cpu(cpu) {
415 copy_map_value_long(map, per_cpu_ptr(pptr, cpu), value + off);
416 bpf_obj_free_fields(array->map.record, per_cpu_ptr(pptr, cpu));
417 off += size;
418 }
419 rcu_read_unlock();
420 return 0;
421}
422
423/* Called from syscall or from eBPF program */
424static long array_map_delete_elem(struct bpf_map *map, void *key)
425{
426 return -EINVAL;
427}
428
429static void *array_map_vmalloc_addr(struct bpf_array *array)
430{
431 return (void *)round_down((unsigned long)array, PAGE_SIZE);
432}
433
434static void array_map_free_timers_wq(struct bpf_map *map)
435{
436 struct bpf_array *array = container_of(map, struct bpf_array, map);
437 int i;
438
439 /* We don't reset or free fields other than timer and workqueue
440 * on uref dropping to zero.
441 */
442 if (btf_record_has_field(map->record, BPF_TIMER | BPF_WORKQUEUE)) {
443 for (i = 0; i < array->map.max_entries; i++) {
444 if (btf_record_has_field(map->record, BPF_TIMER))
445 bpf_obj_free_timer(map->record, array_map_elem_ptr(array, i));
446 if (btf_record_has_field(map->record, BPF_WORKQUEUE))
447 bpf_obj_free_workqueue(map->record, array_map_elem_ptr(array, i));
448 }
449 }
450}
451
452/* Called when map->refcnt goes to zero, either from workqueue or from syscall */
453static void array_map_free(struct bpf_map *map)
454{
455 struct bpf_array *array = container_of(map, struct bpf_array, map);
456 int i;
457
458 if (!IS_ERR_OR_NULL(map->record)) {
459 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
460 for (i = 0; i < array->map.max_entries; i++) {
461 void __percpu *pptr = array->pptrs[i & array->index_mask];
462 int cpu;
463
464 for_each_possible_cpu(cpu) {
465 bpf_obj_free_fields(map->record, per_cpu_ptr(pptr, cpu));
466 cond_resched();
467 }
468 }
469 } else {
470 for (i = 0; i < array->map.max_entries; i++)
471 bpf_obj_free_fields(map->record, array_map_elem_ptr(array, i));
472 }
473 }
474
475 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
476 bpf_array_free_percpu(array);
477
478 if (array->map.map_flags & BPF_F_MMAPABLE)
479 bpf_map_area_free(array_map_vmalloc_addr(array));
480 else
481 bpf_map_area_free(array);
482}
483
484static void array_map_seq_show_elem(struct bpf_map *map, void *key,
485 struct seq_file *m)
486{
487 void *value;
488
489 rcu_read_lock();
490
491 value = array_map_lookup_elem(map, key);
492 if (!value) {
493 rcu_read_unlock();
494 return;
495 }
496
497 if (map->btf_key_type_id)
498 seq_printf(m, "%u: ", *(u32 *)key);
499 btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
500 seq_putc(m, '\n');
501
502 rcu_read_unlock();
503}
504
505static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key,
506 struct seq_file *m)
507{
508 struct bpf_array *array = container_of(map, struct bpf_array, map);
509 u32 index = *(u32 *)key;
510 void __percpu *pptr;
511 int cpu;
512
513 rcu_read_lock();
514
515 seq_printf(m, "%u: {\n", *(u32 *)key);
516 pptr = array->pptrs[index & array->index_mask];
517 for_each_possible_cpu(cpu) {
518 seq_printf(m, "\tcpu%d: ", cpu);
519 btf_type_seq_show(map->btf, map->btf_value_type_id,
520 per_cpu_ptr(pptr, cpu), m);
521 seq_putc(m, '\n');
522 }
523 seq_puts(m, "}\n");
524
525 rcu_read_unlock();
526}
527
528static int array_map_check_btf(const struct bpf_map *map,
529 const struct btf *btf,
530 const struct btf_type *key_type,
531 const struct btf_type *value_type)
532{
533 u32 int_data;
534
535 /* One exception for keyless BTF: .bss/.data/.rodata map */
536 if (btf_type_is_void(key_type)) {
537 if (map->map_type != BPF_MAP_TYPE_ARRAY ||
538 map->max_entries != 1)
539 return -EINVAL;
540
541 if (BTF_INFO_KIND(value_type->info) != BTF_KIND_DATASEC)
542 return -EINVAL;
543
544 return 0;
545 }
546
547 if (BTF_INFO_KIND(key_type->info) != BTF_KIND_INT)
548 return -EINVAL;
549
550 int_data = *(u32 *)(key_type + 1);
551 /* bpf array can only take a u32 key. This check makes sure
552 * that the btf matches the attr used during map_create.
553 */
554 if (BTF_INT_BITS(int_data) != 32 || BTF_INT_OFFSET(int_data))
555 return -EINVAL;
556
557 return 0;
558}
559
560static int array_map_mmap(struct bpf_map *map, struct vm_area_struct *vma)
561{
562 struct bpf_array *array = container_of(map, struct bpf_array, map);
563 pgoff_t pgoff = PAGE_ALIGN(sizeof(*array)) >> PAGE_SHIFT;
564
565 if (!(map->map_flags & BPF_F_MMAPABLE))
566 return -EINVAL;
567
568 if (vma->vm_pgoff * PAGE_SIZE + (vma->vm_end - vma->vm_start) >
569 PAGE_ALIGN((u64)array->map.max_entries * array->elem_size))
570 return -EINVAL;
571
572 return remap_vmalloc_range(vma, array_map_vmalloc_addr(array),
573 vma->vm_pgoff + pgoff);
574}
575
576static bool array_map_meta_equal(const struct bpf_map *meta0,
577 const struct bpf_map *meta1)
578{
579 if (!bpf_map_meta_equal(meta0, meta1))
580 return false;
581 return meta0->map_flags & BPF_F_INNER_MAP ? true :
582 meta0->max_entries == meta1->max_entries;
583}
584
585struct bpf_iter_seq_array_map_info {
586 struct bpf_map *map;
587 void *percpu_value_buf;
588 u32 index;
589};
590
591static void *bpf_array_map_seq_start(struct seq_file *seq, loff_t *pos)
592{
593 struct bpf_iter_seq_array_map_info *info = seq->private;
594 struct bpf_map *map = info->map;
595 struct bpf_array *array;
596 u32 index;
597
598 if (info->index >= map->max_entries)
599 return NULL;
600
601 if (*pos == 0)
602 ++*pos;
603 array = container_of(map, struct bpf_array, map);
604 index = info->index & array->index_mask;
605 if (info->percpu_value_buf)
606 return (void *)(uintptr_t)array->pptrs[index];
607 return array_map_elem_ptr(array, index);
608}
609
610static void *bpf_array_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
611{
612 struct bpf_iter_seq_array_map_info *info = seq->private;
613 struct bpf_map *map = info->map;
614 struct bpf_array *array;
615 u32 index;
616
617 ++*pos;
618 ++info->index;
619 if (info->index >= map->max_entries)
620 return NULL;
621
622 array = container_of(map, struct bpf_array, map);
623 index = info->index & array->index_mask;
624 if (info->percpu_value_buf)
625 return (void *)(uintptr_t)array->pptrs[index];
626 return array_map_elem_ptr(array, index);
627}
628
629static int __bpf_array_map_seq_show(struct seq_file *seq, void *v)
630{
631 struct bpf_iter_seq_array_map_info *info = seq->private;
632 struct bpf_iter__bpf_map_elem ctx = {};
633 struct bpf_map *map = info->map;
634 struct bpf_array *array = container_of(map, struct bpf_array, map);
635 struct bpf_iter_meta meta;
636 struct bpf_prog *prog;
637 int off = 0, cpu = 0;
638 void __percpu *pptr;
639 u32 size;
640
641 meta.seq = seq;
642 prog = bpf_iter_get_info(&meta, v == NULL);
643 if (!prog)
644 return 0;
645
646 ctx.meta = &meta;
647 ctx.map = info->map;
648 if (v) {
649 ctx.key = &info->index;
650
651 if (!info->percpu_value_buf) {
652 ctx.value = v;
653 } else {
654 pptr = (void __percpu *)(uintptr_t)v;
655 size = array->elem_size;
656 for_each_possible_cpu(cpu) {
657 copy_map_value_long(map, info->percpu_value_buf + off,
658 per_cpu_ptr(pptr, cpu));
659 check_and_init_map_value(map, info->percpu_value_buf + off);
660 off += size;
661 }
662 ctx.value = info->percpu_value_buf;
663 }
664 }
665
666 return bpf_iter_run_prog(prog, &ctx);
667}
668
669static int bpf_array_map_seq_show(struct seq_file *seq, void *v)
670{
671 return __bpf_array_map_seq_show(seq, v);
672}
673
674static void bpf_array_map_seq_stop(struct seq_file *seq, void *v)
675{
676 if (!v)
677 (void)__bpf_array_map_seq_show(seq, NULL);
678}
679
680static int bpf_iter_init_array_map(void *priv_data,
681 struct bpf_iter_aux_info *aux)
682{
683 struct bpf_iter_seq_array_map_info *seq_info = priv_data;
684 struct bpf_map *map = aux->map;
685 struct bpf_array *array = container_of(map, struct bpf_array, map);
686 void *value_buf;
687 u32 buf_size;
688
689 if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
690 buf_size = array->elem_size * num_possible_cpus();
691 value_buf = kmalloc(buf_size, GFP_USER | __GFP_NOWARN);
692 if (!value_buf)
693 return -ENOMEM;
694
695 seq_info->percpu_value_buf = value_buf;
696 }
697
698 /* bpf_iter_attach_map() acquires a map uref, and the uref may be
699 * released before or in the middle of iterating map elements, so
700 * acquire an extra map uref for iterator.
701 */
702 bpf_map_inc_with_uref(map);
703 seq_info->map = map;
704 return 0;
705}
706
707static void bpf_iter_fini_array_map(void *priv_data)
708{
709 struct bpf_iter_seq_array_map_info *seq_info = priv_data;
710
711 bpf_map_put_with_uref(seq_info->map);
712 kfree(seq_info->percpu_value_buf);
713}
714
715static const struct seq_operations bpf_array_map_seq_ops = {
716 .start = bpf_array_map_seq_start,
717 .next = bpf_array_map_seq_next,
718 .stop = bpf_array_map_seq_stop,
719 .show = bpf_array_map_seq_show,
720};
721
722static const struct bpf_iter_seq_info iter_seq_info = {
723 .seq_ops = &bpf_array_map_seq_ops,
724 .init_seq_private = bpf_iter_init_array_map,
725 .fini_seq_private = bpf_iter_fini_array_map,
726 .seq_priv_size = sizeof(struct bpf_iter_seq_array_map_info),
727};
728
729static long bpf_for_each_array_elem(struct bpf_map *map, bpf_callback_t callback_fn,
730 void *callback_ctx, u64 flags)
731{
732 u32 i, key, num_elems = 0;
733 struct bpf_array *array;
734 bool is_percpu;
735 u64 ret = 0;
736 void *val;
737
738 if (flags != 0)
739 return -EINVAL;
740
741 is_percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
742 array = container_of(map, struct bpf_array, map);
743 if (is_percpu)
744 migrate_disable();
745 for (i = 0; i < map->max_entries; i++) {
746 if (is_percpu)
747 val = this_cpu_ptr(array->pptrs[i]);
748 else
749 val = array_map_elem_ptr(array, i);
750 num_elems++;
751 key = i;
752 ret = callback_fn((u64)(long)map, (u64)(long)&key,
753 (u64)(long)val, (u64)(long)callback_ctx, 0);
754 /* return value: 0 - continue, 1 - stop and return */
755 if (ret)
756 break;
757 }
758
759 if (is_percpu)
760 migrate_enable();
761 return num_elems;
762}
763
764static u64 array_map_mem_usage(const struct bpf_map *map)
765{
766 struct bpf_array *array = container_of(map, struct bpf_array, map);
767 bool percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
768 u32 elem_size = array->elem_size;
769 u64 entries = map->max_entries;
770 u64 usage = sizeof(*array);
771
772 if (percpu) {
773 usage += entries * sizeof(void *);
774 usage += entries * elem_size * num_possible_cpus();
775 } else {
776 if (map->map_flags & BPF_F_MMAPABLE) {
777 usage = PAGE_ALIGN(usage);
778 usage += PAGE_ALIGN(entries * elem_size);
779 } else {
780 usage += entries * elem_size;
781 }
782 }
783 return usage;
784}
785
786BTF_ID_LIST_SINGLE(array_map_btf_ids, struct, bpf_array)
787const struct bpf_map_ops array_map_ops = {
788 .map_meta_equal = array_map_meta_equal,
789 .map_alloc_check = array_map_alloc_check,
790 .map_alloc = array_map_alloc,
791 .map_free = array_map_free,
792 .map_get_next_key = array_map_get_next_key,
793 .map_release_uref = array_map_free_timers_wq,
794 .map_lookup_elem = array_map_lookup_elem,
795 .map_update_elem = array_map_update_elem,
796 .map_delete_elem = array_map_delete_elem,
797 .map_gen_lookup = array_map_gen_lookup,
798 .map_direct_value_addr = array_map_direct_value_addr,
799 .map_direct_value_meta = array_map_direct_value_meta,
800 .map_mmap = array_map_mmap,
801 .map_seq_show_elem = array_map_seq_show_elem,
802 .map_check_btf = array_map_check_btf,
803 .map_lookup_batch = generic_map_lookup_batch,
804 .map_update_batch = generic_map_update_batch,
805 .map_set_for_each_callback_args = map_set_for_each_callback_args,
806 .map_for_each_callback = bpf_for_each_array_elem,
807 .map_mem_usage = array_map_mem_usage,
808 .map_btf_id = &array_map_btf_ids[0],
809 .iter_seq_info = &iter_seq_info,
810};
811
812const struct bpf_map_ops percpu_array_map_ops = {
813 .map_meta_equal = bpf_map_meta_equal,
814 .map_alloc_check = array_map_alloc_check,
815 .map_alloc = array_map_alloc,
816 .map_free = array_map_free,
817 .map_get_next_key = array_map_get_next_key,
818 .map_lookup_elem = percpu_array_map_lookup_elem,
819 .map_gen_lookup = percpu_array_map_gen_lookup,
820 .map_update_elem = array_map_update_elem,
821 .map_delete_elem = array_map_delete_elem,
822 .map_lookup_percpu_elem = percpu_array_map_lookup_percpu_elem,
823 .map_seq_show_elem = percpu_array_map_seq_show_elem,
824 .map_check_btf = array_map_check_btf,
825 .map_lookup_batch = generic_map_lookup_batch,
826 .map_update_batch = generic_map_update_batch,
827 .map_set_for_each_callback_args = map_set_for_each_callback_args,
828 .map_for_each_callback = bpf_for_each_array_elem,
829 .map_mem_usage = array_map_mem_usage,
830 .map_btf_id = &array_map_btf_ids[0],
831 .iter_seq_info = &iter_seq_info,
832};
833
834static int fd_array_map_alloc_check(union bpf_attr *attr)
835{
836 /* only file descriptors can be stored in this type of map */
837 if (attr->value_size != sizeof(u32))
838 return -EINVAL;
839 /* Program read-only/write-only not supported for special maps yet. */
840 if (attr->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG))
841 return -EINVAL;
842 return array_map_alloc_check(attr);
843}
844
845static void fd_array_map_free(struct bpf_map *map)
846{
847 struct bpf_array *array = container_of(map, struct bpf_array, map);
848 int i;
849
850 /* make sure it's empty */
851 for (i = 0; i < array->map.max_entries; i++)
852 BUG_ON(array->ptrs[i] != NULL);
853
854 bpf_map_area_free(array);
855}
856
857static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
858{
859 return ERR_PTR(-EOPNOTSUPP);
860}
861
862/* only called from syscall */
863int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
864{
865 void **elem, *ptr;
866 int ret = 0;
867
868 if (!map->ops->map_fd_sys_lookup_elem)
869 return -ENOTSUPP;
870
871 rcu_read_lock();
872 elem = array_map_lookup_elem(map, key);
873 if (elem && (ptr = READ_ONCE(*elem)))
874 *value = map->ops->map_fd_sys_lookup_elem(ptr);
875 else
876 ret = -ENOENT;
877 rcu_read_unlock();
878
879 return ret;
880}
881
882/* only called from syscall */
883int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
884 void *key, void *value, u64 map_flags)
885{
886 struct bpf_array *array = container_of(map, struct bpf_array, map);
887 void *new_ptr, *old_ptr;
888 u32 index = *(u32 *)key, ufd;
889
890 if (map_flags != BPF_ANY)
891 return -EINVAL;
892
893 if (index >= array->map.max_entries)
894 return -E2BIG;
895
896 ufd = *(u32 *)value;
897 new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
898 if (IS_ERR(new_ptr))
899 return PTR_ERR(new_ptr);
900
901 if (map->ops->map_poke_run) {
902 mutex_lock(&array->aux->poke_mutex);
903 old_ptr = xchg(array->ptrs + index, new_ptr);
904 map->ops->map_poke_run(map, index, old_ptr, new_ptr);
905 mutex_unlock(&array->aux->poke_mutex);
906 } else {
907 old_ptr = xchg(array->ptrs + index, new_ptr);
908 }
909
910 if (old_ptr)
911 map->ops->map_fd_put_ptr(map, old_ptr, true);
912 return 0;
913}
914
915static long __fd_array_map_delete_elem(struct bpf_map *map, void *key, bool need_defer)
916{
917 struct bpf_array *array = container_of(map, struct bpf_array, map);
918 void *old_ptr;
919 u32 index = *(u32 *)key;
920
921 if (index >= array->map.max_entries)
922 return -E2BIG;
923
924 if (map->ops->map_poke_run) {
925 mutex_lock(&array->aux->poke_mutex);
926 old_ptr = xchg(array->ptrs + index, NULL);
927 map->ops->map_poke_run(map, index, old_ptr, NULL);
928 mutex_unlock(&array->aux->poke_mutex);
929 } else {
930 old_ptr = xchg(array->ptrs + index, NULL);
931 }
932
933 if (old_ptr) {
934 map->ops->map_fd_put_ptr(map, old_ptr, need_defer);
935 return 0;
936 } else {
937 return -ENOENT;
938 }
939}
940
941static long fd_array_map_delete_elem(struct bpf_map *map, void *key)
942{
943 return __fd_array_map_delete_elem(map, key, true);
944}
945
946static void *prog_fd_array_get_ptr(struct bpf_map *map,
947 struct file *map_file, int fd)
948{
949 struct bpf_prog *prog = bpf_prog_get(fd);
950 bool is_extended;
951
952 if (IS_ERR(prog))
953 return prog;
954
955 if (prog->type == BPF_PROG_TYPE_EXT ||
956 !bpf_prog_map_compatible(map, prog)) {
957 bpf_prog_put(prog);
958 return ERR_PTR(-EINVAL);
959 }
960
961 mutex_lock(&prog->aux->ext_mutex);
962 is_extended = prog->aux->is_extended;
963 if (!is_extended)
964 prog->aux->prog_array_member_cnt++;
965 mutex_unlock(&prog->aux->ext_mutex);
966 if (is_extended) {
967 /* Extended prog can not be tail callee. It's to prevent a
968 * potential infinite loop like:
969 * tail callee prog entry -> tail callee prog subprog ->
970 * freplace prog entry --tailcall-> tail callee prog entry.
971 */
972 bpf_prog_put(prog);
973 return ERR_PTR(-EBUSY);
974 }
975
976 return prog;
977}
978
979static void prog_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
980{
981 struct bpf_prog *prog = ptr;
982
983 mutex_lock(&prog->aux->ext_mutex);
984 prog->aux->prog_array_member_cnt--;
985 mutex_unlock(&prog->aux->ext_mutex);
986 /* bpf_prog is freed after one RCU or tasks trace grace period */
987 bpf_prog_put(prog);
988}
989
990static u32 prog_fd_array_sys_lookup_elem(void *ptr)
991{
992 return ((struct bpf_prog *)ptr)->aux->id;
993}
994
995/* decrement refcnt of all bpf_progs that are stored in this map */
996static void bpf_fd_array_map_clear(struct bpf_map *map, bool need_defer)
997{
998 struct bpf_array *array = container_of(map, struct bpf_array, map);
999 int i;
1000
1001 for (i = 0; i < array->map.max_entries; i++)
1002 __fd_array_map_delete_elem(map, &i, need_defer);
1003}
1004
1005static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key,
1006 struct seq_file *m)
1007{
1008 void **elem, *ptr;
1009 u32 prog_id;
1010
1011 rcu_read_lock();
1012
1013 elem = array_map_lookup_elem(map, key);
1014 if (elem) {
1015 ptr = READ_ONCE(*elem);
1016 if (ptr) {
1017 seq_printf(m, "%u: ", *(u32 *)key);
1018 prog_id = prog_fd_array_sys_lookup_elem(ptr);
1019 btf_type_seq_show(map->btf, map->btf_value_type_id,
1020 &prog_id, m);
1021 seq_putc(m, '\n');
1022 }
1023 }
1024
1025 rcu_read_unlock();
1026}
1027
1028struct prog_poke_elem {
1029 struct list_head list;
1030 struct bpf_prog_aux *aux;
1031};
1032
1033static int prog_array_map_poke_track(struct bpf_map *map,
1034 struct bpf_prog_aux *prog_aux)
1035{
1036 struct prog_poke_elem *elem;
1037 struct bpf_array_aux *aux;
1038 int ret = 0;
1039
1040 aux = container_of(map, struct bpf_array, map)->aux;
1041 mutex_lock(&aux->poke_mutex);
1042 list_for_each_entry(elem, &aux->poke_progs, list) {
1043 if (elem->aux == prog_aux)
1044 goto out;
1045 }
1046
1047 elem = kmalloc(sizeof(*elem), GFP_KERNEL);
1048 if (!elem) {
1049 ret = -ENOMEM;
1050 goto out;
1051 }
1052
1053 INIT_LIST_HEAD(&elem->list);
1054 /* We must track the program's aux info at this point in time
1055 * since the program pointer itself may not be stable yet, see
1056 * also comment in prog_array_map_poke_run().
1057 */
1058 elem->aux = prog_aux;
1059
1060 list_add_tail(&elem->list, &aux->poke_progs);
1061out:
1062 mutex_unlock(&aux->poke_mutex);
1063 return ret;
1064}
1065
1066static void prog_array_map_poke_untrack(struct bpf_map *map,
1067 struct bpf_prog_aux *prog_aux)
1068{
1069 struct prog_poke_elem *elem, *tmp;
1070 struct bpf_array_aux *aux;
1071
1072 aux = container_of(map, struct bpf_array, map)->aux;
1073 mutex_lock(&aux->poke_mutex);
1074 list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
1075 if (elem->aux == prog_aux) {
1076 list_del_init(&elem->list);
1077 kfree(elem);
1078 break;
1079 }
1080 }
1081 mutex_unlock(&aux->poke_mutex);
1082}
1083
1084void __weak bpf_arch_poke_desc_update(struct bpf_jit_poke_descriptor *poke,
1085 struct bpf_prog *new, struct bpf_prog *old)
1086{
1087 WARN_ON_ONCE(1);
1088}
1089
1090static void prog_array_map_poke_run(struct bpf_map *map, u32 key,
1091 struct bpf_prog *old,
1092 struct bpf_prog *new)
1093{
1094 struct prog_poke_elem *elem;
1095 struct bpf_array_aux *aux;
1096
1097 aux = container_of(map, struct bpf_array, map)->aux;
1098 WARN_ON_ONCE(!mutex_is_locked(&aux->poke_mutex));
1099
1100 list_for_each_entry(elem, &aux->poke_progs, list) {
1101 struct bpf_jit_poke_descriptor *poke;
1102 int i;
1103
1104 for (i = 0; i < elem->aux->size_poke_tab; i++) {
1105 poke = &elem->aux->poke_tab[i];
1106
1107 /* Few things to be aware of:
1108 *
1109 * 1) We can only ever access aux in this context, but
1110 * not aux->prog since it might not be stable yet and
1111 * there could be danger of use after free otherwise.
1112 * 2) Initially when we start tracking aux, the program
1113 * is not JITed yet and also does not have a kallsyms
1114 * entry. We skip these as poke->tailcall_target_stable
1115 * is not active yet. The JIT will do the final fixup
1116 * before setting it stable. The various
1117 * poke->tailcall_target_stable are successively
1118 * activated, so tail call updates can arrive from here
1119 * while JIT is still finishing its final fixup for
1120 * non-activated poke entries.
1121 * 3) Also programs reaching refcount of zero while patching
1122 * is in progress is okay since we're protected under
1123 * poke_mutex and untrack the programs before the JIT
1124 * buffer is freed.
1125 */
1126 if (!READ_ONCE(poke->tailcall_target_stable))
1127 continue;
1128 if (poke->reason != BPF_POKE_REASON_TAIL_CALL)
1129 continue;
1130 if (poke->tail_call.map != map ||
1131 poke->tail_call.key != key)
1132 continue;
1133
1134 bpf_arch_poke_desc_update(poke, new, old);
1135 }
1136 }
1137}
1138
1139static void prog_array_map_clear_deferred(struct work_struct *work)
1140{
1141 struct bpf_map *map = container_of(work, struct bpf_array_aux,
1142 work)->map;
1143 bpf_fd_array_map_clear(map, true);
1144 bpf_map_put(map);
1145}
1146
1147static void prog_array_map_clear(struct bpf_map *map)
1148{
1149 struct bpf_array_aux *aux = container_of(map, struct bpf_array,
1150 map)->aux;
1151 bpf_map_inc(map);
1152 schedule_work(&aux->work);
1153}
1154
1155static struct bpf_map *prog_array_map_alloc(union bpf_attr *attr)
1156{
1157 struct bpf_array_aux *aux;
1158 struct bpf_map *map;
1159
1160 aux = kzalloc(sizeof(*aux), GFP_KERNEL_ACCOUNT);
1161 if (!aux)
1162 return ERR_PTR(-ENOMEM);
1163
1164 INIT_WORK(&aux->work, prog_array_map_clear_deferred);
1165 INIT_LIST_HEAD(&aux->poke_progs);
1166 mutex_init(&aux->poke_mutex);
1167
1168 map = array_map_alloc(attr);
1169 if (IS_ERR(map)) {
1170 kfree(aux);
1171 return map;
1172 }
1173
1174 container_of(map, struct bpf_array, map)->aux = aux;
1175 aux->map = map;
1176
1177 return map;
1178}
1179
1180static void prog_array_map_free(struct bpf_map *map)
1181{
1182 struct prog_poke_elem *elem, *tmp;
1183 struct bpf_array_aux *aux;
1184
1185 aux = container_of(map, struct bpf_array, map)->aux;
1186 list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
1187 list_del_init(&elem->list);
1188 kfree(elem);
1189 }
1190 kfree(aux);
1191 fd_array_map_free(map);
1192}
1193
1194/* prog_array->aux->{type,jited} is a runtime binding.
1195 * Doing static check alone in the verifier is not enough.
1196 * Thus, prog_array_map cannot be used as an inner_map
1197 * and map_meta_equal is not implemented.
1198 */
1199const struct bpf_map_ops prog_array_map_ops = {
1200 .map_alloc_check = fd_array_map_alloc_check,
1201 .map_alloc = prog_array_map_alloc,
1202 .map_free = prog_array_map_free,
1203 .map_poke_track = prog_array_map_poke_track,
1204 .map_poke_untrack = prog_array_map_poke_untrack,
1205 .map_poke_run = prog_array_map_poke_run,
1206 .map_get_next_key = array_map_get_next_key,
1207 .map_lookup_elem = fd_array_map_lookup_elem,
1208 .map_delete_elem = fd_array_map_delete_elem,
1209 .map_fd_get_ptr = prog_fd_array_get_ptr,
1210 .map_fd_put_ptr = prog_fd_array_put_ptr,
1211 .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
1212 .map_release_uref = prog_array_map_clear,
1213 .map_seq_show_elem = prog_array_map_seq_show_elem,
1214 .map_mem_usage = array_map_mem_usage,
1215 .map_btf_id = &array_map_btf_ids[0],
1216};
1217
1218static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
1219 struct file *map_file)
1220{
1221 struct bpf_event_entry *ee;
1222
1223 ee = kzalloc(sizeof(*ee), GFP_KERNEL);
1224 if (ee) {
1225 ee->event = perf_file->private_data;
1226 ee->perf_file = perf_file;
1227 ee->map_file = map_file;
1228 }
1229
1230 return ee;
1231}
1232
1233static void __bpf_event_entry_free(struct rcu_head *rcu)
1234{
1235 struct bpf_event_entry *ee;
1236
1237 ee = container_of(rcu, struct bpf_event_entry, rcu);
1238 fput(ee->perf_file);
1239 kfree(ee);
1240}
1241
1242static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
1243{
1244 call_rcu(&ee->rcu, __bpf_event_entry_free);
1245}
1246
1247static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
1248 struct file *map_file, int fd)
1249{
1250 struct bpf_event_entry *ee;
1251 struct perf_event *event;
1252 struct file *perf_file;
1253 u64 value;
1254
1255 perf_file = perf_event_get(fd);
1256 if (IS_ERR(perf_file))
1257 return perf_file;
1258
1259 ee = ERR_PTR(-EOPNOTSUPP);
1260 event = perf_file->private_data;
1261 if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
1262 goto err_out;
1263
1264 ee = bpf_event_entry_gen(perf_file, map_file);
1265 if (ee)
1266 return ee;
1267 ee = ERR_PTR(-ENOMEM);
1268err_out:
1269 fput(perf_file);
1270 return ee;
1271}
1272
1273static void perf_event_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
1274{
1275 /* bpf_perf_event is freed after one RCU grace period */
1276 bpf_event_entry_free_rcu(ptr);
1277}
1278
1279static void perf_event_fd_array_release(struct bpf_map *map,
1280 struct file *map_file)
1281{
1282 struct bpf_array *array = container_of(map, struct bpf_array, map);
1283 struct bpf_event_entry *ee;
1284 int i;
1285
1286 if (map->map_flags & BPF_F_PRESERVE_ELEMS)
1287 return;
1288
1289 rcu_read_lock();
1290 for (i = 0; i < array->map.max_entries; i++) {
1291 ee = READ_ONCE(array->ptrs[i]);
1292 if (ee && ee->map_file == map_file)
1293 __fd_array_map_delete_elem(map, &i, true);
1294 }
1295 rcu_read_unlock();
1296}
1297
1298static void perf_event_fd_array_map_free(struct bpf_map *map)
1299{
1300 if (map->map_flags & BPF_F_PRESERVE_ELEMS)
1301 bpf_fd_array_map_clear(map, false);
1302 fd_array_map_free(map);
1303}
1304
1305const struct bpf_map_ops perf_event_array_map_ops = {
1306 .map_meta_equal = bpf_map_meta_equal,
1307 .map_alloc_check = fd_array_map_alloc_check,
1308 .map_alloc = array_map_alloc,
1309 .map_free = perf_event_fd_array_map_free,
1310 .map_get_next_key = array_map_get_next_key,
1311 .map_lookup_elem = fd_array_map_lookup_elem,
1312 .map_delete_elem = fd_array_map_delete_elem,
1313 .map_fd_get_ptr = perf_event_fd_array_get_ptr,
1314 .map_fd_put_ptr = perf_event_fd_array_put_ptr,
1315 .map_release = perf_event_fd_array_release,
1316 .map_check_btf = map_check_no_btf,
1317 .map_mem_usage = array_map_mem_usage,
1318 .map_btf_id = &array_map_btf_ids[0],
1319};
1320
1321#ifdef CONFIG_CGROUPS
1322static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
1323 struct file *map_file /* not used */,
1324 int fd)
1325{
1326 return cgroup_get_from_fd(fd);
1327}
1328
1329static void cgroup_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
1330{
1331 /* cgroup_put free cgrp after a rcu grace period */
1332 cgroup_put(ptr);
1333}
1334
1335static void cgroup_fd_array_free(struct bpf_map *map)
1336{
1337 bpf_fd_array_map_clear(map, false);
1338 fd_array_map_free(map);
1339}
1340
1341const struct bpf_map_ops cgroup_array_map_ops = {
1342 .map_meta_equal = bpf_map_meta_equal,
1343 .map_alloc_check = fd_array_map_alloc_check,
1344 .map_alloc = array_map_alloc,
1345 .map_free = cgroup_fd_array_free,
1346 .map_get_next_key = array_map_get_next_key,
1347 .map_lookup_elem = fd_array_map_lookup_elem,
1348 .map_delete_elem = fd_array_map_delete_elem,
1349 .map_fd_get_ptr = cgroup_fd_array_get_ptr,
1350 .map_fd_put_ptr = cgroup_fd_array_put_ptr,
1351 .map_check_btf = map_check_no_btf,
1352 .map_mem_usage = array_map_mem_usage,
1353 .map_btf_id = &array_map_btf_ids[0],
1354};
1355#endif
1356
1357static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
1358{
1359 struct bpf_map *map, *inner_map_meta;
1360
1361 inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
1362 if (IS_ERR(inner_map_meta))
1363 return inner_map_meta;
1364
1365 map = array_map_alloc(attr);
1366 if (IS_ERR(map)) {
1367 bpf_map_meta_free(inner_map_meta);
1368 return map;
1369 }
1370
1371 map->inner_map_meta = inner_map_meta;
1372
1373 return map;
1374}
1375
1376static void array_of_map_free(struct bpf_map *map)
1377{
1378 /* map->inner_map_meta is only accessed by syscall which
1379 * is protected by fdget/fdput.
1380 */
1381 bpf_map_meta_free(map->inner_map_meta);
1382 bpf_fd_array_map_clear(map, false);
1383 fd_array_map_free(map);
1384}
1385
1386static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
1387{
1388 struct bpf_map **inner_map = array_map_lookup_elem(map, key);
1389
1390 if (!inner_map)
1391 return NULL;
1392
1393 return READ_ONCE(*inner_map);
1394}
1395
1396static int array_of_map_gen_lookup(struct bpf_map *map,
1397 struct bpf_insn *insn_buf)
1398{
1399 struct bpf_array *array = container_of(map, struct bpf_array, map);
1400 u32 elem_size = array->elem_size;
1401 struct bpf_insn *insn = insn_buf;
1402 const int ret = BPF_REG_0;
1403 const int map_ptr = BPF_REG_1;
1404 const int index = BPF_REG_2;
1405
1406 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
1407 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
1408 if (!map->bypass_spec_v1) {
1409 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
1410 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
1411 } else {
1412 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
1413 }
1414 if (is_power_of_2(elem_size))
1415 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
1416 else
1417 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
1418 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
1419 *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
1420 *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
1421 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
1422 *insn++ = BPF_MOV64_IMM(ret, 0);
1423
1424 return insn - insn_buf;
1425}
1426
1427const struct bpf_map_ops array_of_maps_map_ops = {
1428 .map_alloc_check = fd_array_map_alloc_check,
1429 .map_alloc = array_of_map_alloc,
1430 .map_free = array_of_map_free,
1431 .map_get_next_key = array_map_get_next_key,
1432 .map_lookup_elem = array_of_map_lookup_elem,
1433 .map_delete_elem = fd_array_map_delete_elem,
1434 .map_fd_get_ptr = bpf_map_fd_get_ptr,
1435 .map_fd_put_ptr = bpf_map_fd_put_ptr,
1436 .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
1437 .map_gen_lookup = array_of_map_gen_lookup,
1438 .map_lookup_batch = generic_map_lookup_batch,
1439 .map_update_batch = generic_map_update_batch,
1440 .map_check_btf = map_check_no_btf,
1441 .map_mem_usage = array_map_mem_usage,
1442 .map_btf_id = &array_map_btf_ids[0],
1443};