]>
Commit | Line | Data |
---|---|---|
5b497af4 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
28fbcfa0 | 2 | /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com |
81ed18ab | 3 | * Copyright (c) 2016,2017 Facebook |
28fbcfa0 AS |
4 | */ |
5 | #include <linux/bpf.h> | |
a26ca7c9 | 6 | #include <linux/btf.h> |
28fbcfa0 | 7 | #include <linux/err.h> |
28fbcfa0 AS |
8 | #include <linux/slab.h> |
9 | #include <linux/mm.h> | |
04fd61ab | 10 | #include <linux/filter.h> |
0cdf5640 | 11 | #include <linux/perf_event.h> |
a26ca7c9 | 12 | #include <uapi/linux/btf.h> |
1e6c62a8 | 13 | #include <linux/rcupdate_trace.h> |
28fbcfa0 | 14 | |
56f668df MKL |
15 | #include "map_in_map.h" |
16 | ||
6e71b04a | 17 | #define ARRAY_CREATE_FLAG_MASK \ |
792caccc | 18 | (BPF_F_NUMA_NODE | BPF_F_MMAPABLE | BPF_F_ACCESS_MASK | \ |
4a8f87e6 | 19 | BPF_F_PRESERVE_ELEMS | BPF_F_INNER_MAP) |
6e71b04a | 20 | |
a10423b8 AS |
21 | static void bpf_array_free_percpu(struct bpf_array *array) |
22 | { | |
23 | int i; | |
24 | ||
32fff239 | 25 | for (i = 0; i < array->map.max_entries; i++) { |
a10423b8 | 26 | free_percpu(array->pptrs[i]); |
32fff239 ED |
27 | cond_resched(); |
28 | } | |
a10423b8 AS |
29 | } |
30 | ||
31 | static int bpf_array_alloc_percpu(struct bpf_array *array) | |
32 | { | |
33 | void __percpu *ptr; | |
34 | int i; | |
35 | ||
36 | for (i = 0; i < array->map.max_entries; i++) { | |
6d192c79 RG |
37 | ptr = bpf_map_alloc_percpu(&array->map, array->elem_size, 8, |
38 | GFP_USER | __GFP_NOWARN); | |
a10423b8 AS |
39 | if (!ptr) { |
40 | bpf_array_free_percpu(array); | |
41 | return -ENOMEM; | |
42 | } | |
43 | array->pptrs[i] = ptr; | |
32fff239 | 44 | cond_resched(); |
a10423b8 AS |
45 | } |
46 | ||
47 | return 0; | |
48 | } | |
49 | ||
28fbcfa0 | 50 | /* Called from syscall */ |
5dc4c4b7 | 51 | int array_map_alloc_check(union bpf_attr *attr) |
28fbcfa0 | 52 | { |
a10423b8 | 53 | bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY; |
96eabe7a | 54 | int numa_node = bpf_map_attr_numa_node(attr); |
28fbcfa0 AS |
55 | |
56 | /* check sanity of attributes */ | |
57 | if (attr->max_entries == 0 || attr->key_size != 4 || | |
6e71b04a CF |
58 | attr->value_size == 0 || |
59 | attr->map_flags & ~ARRAY_CREATE_FLAG_MASK || | |
591fe988 | 60 | !bpf_map_flags_access_ok(attr->map_flags) || |
96eabe7a | 61 | (percpu && numa_node != NUMA_NO_NODE)) |
ad46061f | 62 | return -EINVAL; |
28fbcfa0 | 63 | |
fc970227 | 64 | if (attr->map_type != BPF_MAP_TYPE_ARRAY && |
4a8f87e6 | 65 | attr->map_flags & (BPF_F_MMAPABLE | BPF_F_INNER_MAP)) |
fc970227 AN |
66 | return -EINVAL; |
67 | ||
792caccc SL |
68 | if (attr->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY && |
69 | attr->map_flags & BPF_F_PRESERVE_ELEMS) | |
70 | return -EINVAL; | |
71 | ||
7984c27c | 72 | if (attr->value_size > KMALLOC_MAX_SIZE) |
01b3f521 AS |
73 | /* if value_size is bigger, the user space won't be able to |
74 | * access the elements. | |
75 | */ | |
ad46061f JK |
76 | return -E2BIG; |
77 | ||
78 | return 0; | |
79 | } | |
80 | ||
81 | static struct bpf_map *array_map_alloc(union bpf_attr *attr) | |
82 | { | |
83 | bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY; | |
1bc59756 | 84 | int numa_node = bpf_map_attr_numa_node(attr); |
ad46061f | 85 | u32 elem_size, index_mask, max_entries; |
2c78ee89 | 86 | bool bypass_spec_v1 = bpf_bypass_spec_v1(); |
1bc59756 | 87 | u64 array_size, mask64; |
ad46061f | 88 | struct bpf_array *array; |
01b3f521 | 89 | |
28fbcfa0 AS |
90 | elem_size = round_up(attr->value_size, 8); |
91 | ||
b2157399 | 92 | max_entries = attr->max_entries; |
b2157399 | 93 | |
bbeb6e43 DB |
94 | /* On 32 bit archs roundup_pow_of_two() with max_entries that has |
95 | * upper most bit set in u32 space is undefined behavior due to | |
96 | * resulting 1U << 32, so do it manually here in u64 space. | |
97 | */ | |
98 | mask64 = fls_long(max_entries - 1); | |
99 | mask64 = 1ULL << mask64; | |
100 | mask64 -= 1; | |
101 | ||
102 | index_mask = mask64; | |
2c78ee89 | 103 | if (!bypass_spec_v1) { |
b2157399 AS |
104 | /* round up array size to nearest power of 2, |
105 | * since cpu will speculate within index_mask limits | |
106 | */ | |
107 | max_entries = index_mask + 1; | |
bbeb6e43 DB |
108 | /* Check for overflows. */ |
109 | if (max_entries < attr->max_entries) | |
110 | return ERR_PTR(-E2BIG); | |
111 | } | |
b2157399 | 112 | |
a10423b8 | 113 | array_size = sizeof(*array); |
fc970227 | 114 | if (percpu) { |
b2157399 | 115 | array_size += (u64) max_entries * sizeof(void *); |
fc970227 AN |
116 | } else { |
117 | /* rely on vmalloc() to return page-aligned memory and | |
118 | * ensure array->value is exactly page-aligned | |
119 | */ | |
120 | if (attr->map_flags & BPF_F_MMAPABLE) { | |
121 | array_size = PAGE_ALIGN(array_size); | |
122 | array_size += PAGE_ALIGN((u64) max_entries * elem_size); | |
123 | } else { | |
124 | array_size += (u64) max_entries * elem_size; | |
125 | } | |
126 | } | |
a10423b8 | 127 | |
28fbcfa0 | 128 | /* allocate all map elements and zero-initialize them */ |
fc970227 AN |
129 | if (attr->map_flags & BPF_F_MMAPABLE) { |
130 | void *data; | |
131 | ||
132 | /* kmalloc'ed memory can't be mmap'ed, use explicit vmalloc */ | |
133 | data = bpf_map_area_mmapable_alloc(array_size, numa_node); | |
1bc59756 | 134 | if (!data) |
fc970227 | 135 | return ERR_PTR(-ENOMEM); |
fc970227 AN |
136 | array = data + PAGE_ALIGN(sizeof(struct bpf_array)) |
137 | - offsetof(struct bpf_array, value); | |
138 | } else { | |
139 | array = bpf_map_area_alloc(array_size, numa_node); | |
140 | } | |
1bc59756 | 141 | if (!array) |
d407bd25 | 142 | return ERR_PTR(-ENOMEM); |
b2157399 | 143 | array->index_mask = index_mask; |
2c78ee89 | 144 | array->map.bypass_spec_v1 = bypass_spec_v1; |
28fbcfa0 AS |
145 | |
146 | /* copy mandatory map attributes */ | |
32852649 | 147 | bpf_map_init_from_attr(&array->map, attr); |
28fbcfa0 AS |
148 | array->elem_size = elem_size; |
149 | ||
9c2d63b8 | 150 | if (percpu && bpf_array_alloc_percpu(array)) { |
d407bd25 | 151 | bpf_map_area_free(array); |
a10423b8 AS |
152 | return ERR_PTR(-ENOMEM); |
153 | } | |
a10423b8 | 154 | |
28fbcfa0 | 155 | return &array->map; |
28fbcfa0 AS |
156 | } |
157 | ||
158 | /* Called from syscall or from eBPF program */ | |
159 | static void *array_map_lookup_elem(struct bpf_map *map, void *key) | |
160 | { | |
161 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
162 | u32 index = *(u32 *)key; | |
163 | ||
a10423b8 | 164 | if (unlikely(index >= array->map.max_entries)) |
28fbcfa0 AS |
165 | return NULL; |
166 | ||
b2157399 | 167 | return array->value + array->elem_size * (index & array->index_mask); |
28fbcfa0 AS |
168 | } |
169 | ||
d8eca5bb DB |
170 | static int array_map_direct_value_addr(const struct bpf_map *map, u64 *imm, |
171 | u32 off) | |
172 | { | |
173 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
174 | ||
175 | if (map->max_entries != 1) | |
176 | return -ENOTSUPP; | |
177 | if (off >= map->value_size) | |
178 | return -EINVAL; | |
179 | ||
180 | *imm = (unsigned long)array->value; | |
181 | return 0; | |
182 | } | |
183 | ||
184 | static int array_map_direct_value_meta(const struct bpf_map *map, u64 imm, | |
185 | u32 *off) | |
186 | { | |
187 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
188 | u64 base = (unsigned long)array->value; | |
189 | u64 range = array->elem_size; | |
190 | ||
191 | if (map->max_entries != 1) | |
192 | return -ENOTSUPP; | |
193 | if (imm < base || imm >= base + range) | |
194 | return -ENOENT; | |
195 | ||
196 | *off = imm - base; | |
197 | return 0; | |
198 | } | |
199 | ||
81ed18ab | 200 | /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */ |
4a8f87e6 | 201 | static int array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf) |
81ed18ab | 202 | { |
b2157399 | 203 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
81ed18ab | 204 | struct bpf_insn *insn = insn_buf; |
fad73a1a | 205 | u32 elem_size = round_up(map->value_size, 8); |
81ed18ab AS |
206 | const int ret = BPF_REG_0; |
207 | const int map_ptr = BPF_REG_1; | |
208 | const int index = BPF_REG_2; | |
209 | ||
4a8f87e6 DB |
210 | if (map->map_flags & BPF_F_INNER_MAP) |
211 | return -EOPNOTSUPP; | |
212 | ||
81ed18ab AS |
213 | *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value)); |
214 | *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0); | |
2c78ee89 | 215 | if (!map->bypass_spec_v1) { |
b2157399 AS |
216 | *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4); |
217 | *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask); | |
218 | } else { | |
219 | *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3); | |
220 | } | |
fad73a1a MKL |
221 | |
222 | if (is_power_of_2(elem_size)) { | |
81ed18ab AS |
223 | *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size)); |
224 | } else { | |
225 | *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size); | |
226 | } | |
227 | *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr); | |
228 | *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1); | |
229 | *insn++ = BPF_MOV64_IMM(ret, 0); | |
230 | return insn - insn_buf; | |
231 | } | |
232 | ||
a10423b8 AS |
233 | /* Called from eBPF program */ |
234 | static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key) | |
235 | { | |
236 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
237 | u32 index = *(u32 *)key; | |
238 | ||
239 | if (unlikely(index >= array->map.max_entries)) | |
240 | return NULL; | |
241 | ||
b2157399 | 242 | return this_cpu_ptr(array->pptrs[index & array->index_mask]); |
a10423b8 AS |
243 | } |
244 | ||
15a07b33 AS |
245 | int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value) |
246 | { | |
247 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
248 | u32 index = *(u32 *)key; | |
249 | void __percpu *pptr; | |
250 | int cpu, off = 0; | |
251 | u32 size; | |
252 | ||
253 | if (unlikely(index >= array->map.max_entries)) | |
254 | return -ENOENT; | |
255 | ||
256 | /* per_cpu areas are zero-filled and bpf programs can only | |
257 | * access 'value_size' of them, so copying rounded areas | |
258 | * will not leak any kernel data | |
259 | */ | |
260 | size = round_up(map->value_size, 8); | |
261 | rcu_read_lock(); | |
b2157399 | 262 | pptr = array->pptrs[index & array->index_mask]; |
15a07b33 AS |
263 | for_each_possible_cpu(cpu) { |
264 | bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size); | |
265 | off += size; | |
266 | } | |
267 | rcu_read_unlock(); | |
268 | return 0; | |
269 | } | |
270 | ||
28fbcfa0 AS |
271 | /* Called from syscall */ |
272 | static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key) | |
273 | { | |
274 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
8fe45924 | 275 | u32 index = key ? *(u32 *)key : U32_MAX; |
28fbcfa0 AS |
276 | u32 *next = (u32 *)next_key; |
277 | ||
278 | if (index >= array->map.max_entries) { | |
279 | *next = 0; | |
280 | return 0; | |
281 | } | |
282 | ||
283 | if (index == array->map.max_entries - 1) | |
284 | return -ENOENT; | |
285 | ||
286 | *next = index + 1; | |
287 | return 0; | |
288 | } | |
289 | ||
290 | /* Called from syscall or from eBPF program */ | |
291 | static int array_map_update_elem(struct bpf_map *map, void *key, void *value, | |
292 | u64 map_flags) | |
293 | { | |
294 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
295 | u32 index = *(u32 *)key; | |
96049f3a | 296 | char *val; |
28fbcfa0 | 297 | |
96049f3a | 298 | if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST)) |
28fbcfa0 AS |
299 | /* unknown flags */ |
300 | return -EINVAL; | |
301 | ||
a10423b8 | 302 | if (unlikely(index >= array->map.max_entries)) |
28fbcfa0 AS |
303 | /* all elements were pre-allocated, cannot insert a new one */ |
304 | return -E2BIG; | |
305 | ||
96049f3a | 306 | if (unlikely(map_flags & BPF_NOEXIST)) |
daaf427c | 307 | /* all elements already exist */ |
28fbcfa0 AS |
308 | return -EEXIST; |
309 | ||
96049f3a AS |
310 | if (unlikely((map_flags & BPF_F_LOCK) && |
311 | !map_value_has_spin_lock(map))) | |
312 | return -EINVAL; | |
313 | ||
314 | if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
b2157399 | 315 | memcpy(this_cpu_ptr(array->pptrs[index & array->index_mask]), |
a10423b8 | 316 | value, map->value_size); |
96049f3a AS |
317 | } else { |
318 | val = array->value + | |
319 | array->elem_size * (index & array->index_mask); | |
320 | if (map_flags & BPF_F_LOCK) | |
321 | copy_map_value_locked(map, val, value, false); | |
322 | else | |
323 | copy_map_value(map, val, value); | |
324 | } | |
28fbcfa0 AS |
325 | return 0; |
326 | } | |
327 | ||
15a07b33 AS |
328 | int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value, |
329 | u64 map_flags) | |
330 | { | |
331 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
332 | u32 index = *(u32 *)key; | |
333 | void __percpu *pptr; | |
334 | int cpu, off = 0; | |
335 | u32 size; | |
336 | ||
337 | if (unlikely(map_flags > BPF_EXIST)) | |
338 | /* unknown flags */ | |
339 | return -EINVAL; | |
340 | ||
341 | if (unlikely(index >= array->map.max_entries)) | |
342 | /* all elements were pre-allocated, cannot insert a new one */ | |
343 | return -E2BIG; | |
344 | ||
345 | if (unlikely(map_flags == BPF_NOEXIST)) | |
346 | /* all elements already exist */ | |
347 | return -EEXIST; | |
348 | ||
349 | /* the user space will provide round_up(value_size, 8) bytes that | |
350 | * will be copied into per-cpu area. bpf programs can only access | |
351 | * value_size of it. During lookup the same extra bytes will be | |
352 | * returned or zeros which were zero-filled by percpu_alloc, | |
353 | * so no kernel data leaks possible | |
354 | */ | |
355 | size = round_up(map->value_size, 8); | |
356 | rcu_read_lock(); | |
b2157399 | 357 | pptr = array->pptrs[index & array->index_mask]; |
15a07b33 AS |
358 | for_each_possible_cpu(cpu) { |
359 | bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size); | |
360 | off += size; | |
361 | } | |
362 | rcu_read_unlock(); | |
363 | return 0; | |
364 | } | |
365 | ||
28fbcfa0 AS |
366 | /* Called from syscall or from eBPF program */ |
367 | static int array_map_delete_elem(struct bpf_map *map, void *key) | |
368 | { | |
369 | return -EINVAL; | |
370 | } | |
371 | ||
fc970227 AN |
372 | static void *array_map_vmalloc_addr(struct bpf_array *array) |
373 | { | |
374 | return (void *)round_down((unsigned long)array, PAGE_SIZE); | |
375 | } | |
376 | ||
28fbcfa0 AS |
377 | /* Called when map->refcnt goes to zero, either from workqueue or from syscall */ |
378 | static void array_map_free(struct bpf_map *map) | |
379 | { | |
380 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
381 | ||
a10423b8 AS |
382 | if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) |
383 | bpf_array_free_percpu(array); | |
384 | ||
fc970227 AN |
385 | if (array->map.map_flags & BPF_F_MMAPABLE) |
386 | bpf_map_area_free(array_map_vmalloc_addr(array)); | |
387 | else | |
388 | bpf_map_area_free(array); | |
28fbcfa0 AS |
389 | } |
390 | ||
a26ca7c9 MKL |
391 | static void array_map_seq_show_elem(struct bpf_map *map, void *key, |
392 | struct seq_file *m) | |
393 | { | |
394 | void *value; | |
395 | ||
396 | rcu_read_lock(); | |
397 | ||
398 | value = array_map_lookup_elem(map, key); | |
399 | if (!value) { | |
400 | rcu_read_unlock(); | |
401 | return; | |
402 | } | |
403 | ||
2824ecb7 DB |
404 | if (map->btf_key_type_id) |
405 | seq_printf(m, "%u: ", *(u32 *)key); | |
9b2cf328 | 406 | btf_type_seq_show(map->btf, map->btf_value_type_id, value, m); |
a26ca7c9 MKL |
407 | seq_puts(m, "\n"); |
408 | ||
409 | rcu_read_unlock(); | |
410 | } | |
411 | ||
c7b27c37 YS |
412 | static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key, |
413 | struct seq_file *m) | |
414 | { | |
415 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
416 | u32 index = *(u32 *)key; | |
417 | void __percpu *pptr; | |
418 | int cpu; | |
419 | ||
420 | rcu_read_lock(); | |
421 | ||
422 | seq_printf(m, "%u: {\n", *(u32 *)key); | |
423 | pptr = array->pptrs[index & array->index_mask]; | |
424 | for_each_possible_cpu(cpu) { | |
425 | seq_printf(m, "\tcpu%d: ", cpu); | |
426 | btf_type_seq_show(map->btf, map->btf_value_type_id, | |
427 | per_cpu_ptr(pptr, cpu), m); | |
428 | seq_puts(m, "\n"); | |
429 | } | |
430 | seq_puts(m, "}\n"); | |
431 | ||
432 | rcu_read_unlock(); | |
433 | } | |
434 | ||
e8d2bec0 | 435 | static int array_map_check_btf(const struct bpf_map *map, |
1b2b234b | 436 | const struct btf *btf, |
e8d2bec0 DB |
437 | const struct btf_type *key_type, |
438 | const struct btf_type *value_type) | |
a26ca7c9 | 439 | { |
a26ca7c9 MKL |
440 | u32 int_data; |
441 | ||
2824ecb7 DB |
442 | /* One exception for keyless BTF: .bss/.data/.rodata map */ |
443 | if (btf_type_is_void(key_type)) { | |
444 | if (map->map_type != BPF_MAP_TYPE_ARRAY || | |
445 | map->max_entries != 1) | |
446 | return -EINVAL; | |
447 | ||
448 | if (BTF_INFO_KIND(value_type->info) != BTF_KIND_DATASEC) | |
449 | return -EINVAL; | |
450 | ||
451 | return 0; | |
452 | } | |
453 | ||
e8d2bec0 | 454 | if (BTF_INFO_KIND(key_type->info) != BTF_KIND_INT) |
a26ca7c9 MKL |
455 | return -EINVAL; |
456 | ||
457 | int_data = *(u32 *)(key_type + 1); | |
e8d2bec0 DB |
458 | /* bpf array can only take a u32 key. This check makes sure |
459 | * that the btf matches the attr used during map_create. | |
a26ca7c9 | 460 | */ |
e8d2bec0 | 461 | if (BTF_INT_BITS(int_data) != 32 || BTF_INT_OFFSET(int_data)) |
a26ca7c9 MKL |
462 | return -EINVAL; |
463 | ||
464 | return 0; | |
465 | } | |
466 | ||
b2e2f0e6 | 467 | static int array_map_mmap(struct bpf_map *map, struct vm_area_struct *vma) |
fc970227 AN |
468 | { |
469 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
470 | pgoff_t pgoff = PAGE_ALIGN(sizeof(*array)) >> PAGE_SHIFT; | |
471 | ||
472 | if (!(map->map_flags & BPF_F_MMAPABLE)) | |
473 | return -EINVAL; | |
474 | ||
333291ce AN |
475 | if (vma->vm_pgoff * PAGE_SIZE + (vma->vm_end - vma->vm_start) > |
476 | PAGE_ALIGN((u64)array->map.max_entries * array->elem_size)) | |
477 | return -EINVAL; | |
478 | ||
479 | return remap_vmalloc_range(vma, array_map_vmalloc_addr(array), | |
480 | vma->vm_pgoff + pgoff); | |
fc970227 AN |
481 | } |
482 | ||
134fede4 MKL |
483 | static bool array_map_meta_equal(const struct bpf_map *meta0, |
484 | const struct bpf_map *meta1) | |
485 | { | |
4a8f87e6 DB |
486 | if (!bpf_map_meta_equal(meta0, meta1)) |
487 | return false; | |
488 | return meta0->map_flags & BPF_F_INNER_MAP ? true : | |
489 | meta0->max_entries == meta1->max_entries; | |
134fede4 MKL |
490 | } |
491 | ||
d3cc2ab5 YS |
492 | struct bpf_iter_seq_array_map_info { |
493 | struct bpf_map *map; | |
494 | void *percpu_value_buf; | |
495 | u32 index; | |
496 | }; | |
497 | ||
498 | static void *bpf_array_map_seq_start(struct seq_file *seq, loff_t *pos) | |
499 | { | |
500 | struct bpf_iter_seq_array_map_info *info = seq->private; | |
501 | struct bpf_map *map = info->map; | |
502 | struct bpf_array *array; | |
503 | u32 index; | |
504 | ||
505 | if (info->index >= map->max_entries) | |
506 | return NULL; | |
507 | ||
508 | if (*pos == 0) | |
509 | ++*pos; | |
510 | array = container_of(map, struct bpf_array, map); | |
511 | index = info->index & array->index_mask; | |
512 | if (info->percpu_value_buf) | |
513 | return array->pptrs[index]; | |
514 | return array->value + array->elem_size * index; | |
515 | } | |
516 | ||
517 | static void *bpf_array_map_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
518 | { | |
519 | struct bpf_iter_seq_array_map_info *info = seq->private; | |
520 | struct bpf_map *map = info->map; | |
521 | struct bpf_array *array; | |
522 | u32 index; | |
523 | ||
524 | ++*pos; | |
525 | ++info->index; | |
526 | if (info->index >= map->max_entries) | |
527 | return NULL; | |
528 | ||
529 | array = container_of(map, struct bpf_array, map); | |
530 | index = info->index & array->index_mask; | |
531 | if (info->percpu_value_buf) | |
532 | return array->pptrs[index]; | |
533 | return array->value + array->elem_size * index; | |
534 | } | |
535 | ||
536 | static int __bpf_array_map_seq_show(struct seq_file *seq, void *v) | |
537 | { | |
538 | struct bpf_iter_seq_array_map_info *info = seq->private; | |
539 | struct bpf_iter__bpf_map_elem ctx = {}; | |
540 | struct bpf_map *map = info->map; | |
541 | struct bpf_iter_meta meta; | |
542 | struct bpf_prog *prog; | |
543 | int off = 0, cpu = 0; | |
544 | void __percpu **pptr; | |
545 | u32 size; | |
546 | ||
547 | meta.seq = seq; | |
548 | prog = bpf_iter_get_info(&meta, v == NULL); | |
549 | if (!prog) | |
550 | return 0; | |
551 | ||
552 | ctx.meta = &meta; | |
553 | ctx.map = info->map; | |
554 | if (v) { | |
555 | ctx.key = &info->index; | |
556 | ||
557 | if (!info->percpu_value_buf) { | |
558 | ctx.value = v; | |
559 | } else { | |
560 | pptr = v; | |
561 | size = round_up(map->value_size, 8); | |
562 | for_each_possible_cpu(cpu) { | |
563 | bpf_long_memcpy(info->percpu_value_buf + off, | |
564 | per_cpu_ptr(pptr, cpu), | |
565 | size); | |
566 | off += size; | |
567 | } | |
568 | ctx.value = info->percpu_value_buf; | |
569 | } | |
570 | } | |
571 | ||
572 | return bpf_iter_run_prog(prog, &ctx); | |
573 | } | |
574 | ||
575 | static int bpf_array_map_seq_show(struct seq_file *seq, void *v) | |
576 | { | |
577 | return __bpf_array_map_seq_show(seq, v); | |
578 | } | |
579 | ||
580 | static void bpf_array_map_seq_stop(struct seq_file *seq, void *v) | |
581 | { | |
582 | if (!v) | |
583 | (void)__bpf_array_map_seq_show(seq, NULL); | |
584 | } | |
585 | ||
586 | static int bpf_iter_init_array_map(void *priv_data, | |
587 | struct bpf_iter_aux_info *aux) | |
588 | { | |
589 | struct bpf_iter_seq_array_map_info *seq_info = priv_data; | |
590 | struct bpf_map *map = aux->map; | |
591 | void *value_buf; | |
592 | u32 buf_size; | |
593 | ||
594 | if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
595 | buf_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
596 | value_buf = kmalloc(buf_size, GFP_USER | __GFP_NOWARN); | |
597 | if (!value_buf) | |
598 | return -ENOMEM; | |
599 | ||
600 | seq_info->percpu_value_buf = value_buf; | |
601 | } | |
602 | ||
603 | seq_info->map = map; | |
604 | return 0; | |
605 | } | |
606 | ||
607 | static void bpf_iter_fini_array_map(void *priv_data) | |
608 | { | |
609 | struct bpf_iter_seq_array_map_info *seq_info = priv_data; | |
610 | ||
611 | kfree(seq_info->percpu_value_buf); | |
612 | } | |
613 | ||
614 | static const struct seq_operations bpf_array_map_seq_ops = { | |
615 | .start = bpf_array_map_seq_start, | |
616 | .next = bpf_array_map_seq_next, | |
617 | .stop = bpf_array_map_seq_stop, | |
618 | .show = bpf_array_map_seq_show, | |
619 | }; | |
620 | ||
621 | static const struct bpf_iter_seq_info iter_seq_info = { | |
622 | .seq_ops = &bpf_array_map_seq_ops, | |
623 | .init_seq_private = bpf_iter_init_array_map, | |
624 | .fini_seq_private = bpf_iter_fini_array_map, | |
625 | .seq_priv_size = sizeof(struct bpf_iter_seq_array_map_info), | |
626 | }; | |
627 | ||
41c48f3a | 628 | static int array_map_btf_id; |
40077e0c | 629 | const struct bpf_map_ops array_map_ops = { |
134fede4 | 630 | .map_meta_equal = array_map_meta_equal, |
ad46061f | 631 | .map_alloc_check = array_map_alloc_check, |
28fbcfa0 AS |
632 | .map_alloc = array_map_alloc, |
633 | .map_free = array_map_free, | |
634 | .map_get_next_key = array_map_get_next_key, | |
635 | .map_lookup_elem = array_map_lookup_elem, | |
636 | .map_update_elem = array_map_update_elem, | |
637 | .map_delete_elem = array_map_delete_elem, | |
81ed18ab | 638 | .map_gen_lookup = array_map_gen_lookup, |
d8eca5bb DB |
639 | .map_direct_value_addr = array_map_direct_value_addr, |
640 | .map_direct_value_meta = array_map_direct_value_meta, | |
fc970227 | 641 | .map_mmap = array_map_mmap, |
a26ca7c9 MKL |
642 | .map_seq_show_elem = array_map_seq_show_elem, |
643 | .map_check_btf = array_map_check_btf, | |
c60f2d28 BV |
644 | .map_lookup_batch = generic_map_lookup_batch, |
645 | .map_update_batch = generic_map_update_batch, | |
41c48f3a AI |
646 | .map_btf_name = "bpf_array", |
647 | .map_btf_id = &array_map_btf_id, | |
d3cc2ab5 | 648 | .iter_seq_info = &iter_seq_info, |
28fbcfa0 AS |
649 | }; |
650 | ||
2872e9ac | 651 | static int percpu_array_map_btf_id; |
40077e0c | 652 | const struct bpf_map_ops percpu_array_map_ops = { |
f4d05259 | 653 | .map_meta_equal = bpf_map_meta_equal, |
ad46061f | 654 | .map_alloc_check = array_map_alloc_check, |
a10423b8 AS |
655 | .map_alloc = array_map_alloc, |
656 | .map_free = array_map_free, | |
657 | .map_get_next_key = array_map_get_next_key, | |
658 | .map_lookup_elem = percpu_array_map_lookup_elem, | |
659 | .map_update_elem = array_map_update_elem, | |
660 | .map_delete_elem = array_map_delete_elem, | |
c7b27c37 | 661 | .map_seq_show_elem = percpu_array_map_seq_show_elem, |
e8d2bec0 | 662 | .map_check_btf = array_map_check_btf, |
2872e9ac AI |
663 | .map_btf_name = "bpf_array", |
664 | .map_btf_id = &percpu_array_map_btf_id, | |
d3cc2ab5 | 665 | .iter_seq_info = &iter_seq_info, |
a10423b8 AS |
666 | }; |
667 | ||
ad46061f | 668 | static int fd_array_map_alloc_check(union bpf_attr *attr) |
04fd61ab | 669 | { |
2a36f0b9 | 670 | /* only file descriptors can be stored in this type of map */ |
04fd61ab | 671 | if (attr->value_size != sizeof(u32)) |
ad46061f | 672 | return -EINVAL; |
591fe988 DB |
673 | /* Program read-only/write-only not supported for special maps yet. */ |
674 | if (attr->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) | |
675 | return -EINVAL; | |
ad46061f | 676 | return array_map_alloc_check(attr); |
04fd61ab AS |
677 | } |
678 | ||
2a36f0b9 | 679 | static void fd_array_map_free(struct bpf_map *map) |
04fd61ab AS |
680 | { |
681 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
682 | int i; | |
683 | ||
04fd61ab AS |
684 | /* make sure it's empty */ |
685 | for (i = 0; i < array->map.max_entries; i++) | |
2a36f0b9 | 686 | BUG_ON(array->ptrs[i] != NULL); |
d407bd25 DB |
687 | |
688 | bpf_map_area_free(array); | |
04fd61ab AS |
689 | } |
690 | ||
2a36f0b9 | 691 | static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key) |
04fd61ab | 692 | { |
3b4a63f6 | 693 | return ERR_PTR(-EOPNOTSUPP); |
04fd61ab AS |
694 | } |
695 | ||
14dc6f04 MKL |
696 | /* only called from syscall */ |
697 | int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value) | |
698 | { | |
699 | void **elem, *ptr; | |
700 | int ret = 0; | |
701 | ||
702 | if (!map->ops->map_fd_sys_lookup_elem) | |
703 | return -ENOTSUPP; | |
704 | ||
705 | rcu_read_lock(); | |
706 | elem = array_map_lookup_elem(map, key); | |
707 | if (elem && (ptr = READ_ONCE(*elem))) | |
708 | *value = map->ops->map_fd_sys_lookup_elem(ptr); | |
709 | else | |
710 | ret = -ENOENT; | |
711 | rcu_read_unlock(); | |
712 | ||
713 | return ret; | |
714 | } | |
715 | ||
04fd61ab | 716 | /* only called from syscall */ |
d056a788 DB |
717 | int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file, |
718 | void *key, void *value, u64 map_flags) | |
04fd61ab AS |
719 | { |
720 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
2a36f0b9 | 721 | void *new_ptr, *old_ptr; |
04fd61ab AS |
722 | u32 index = *(u32 *)key, ufd; |
723 | ||
724 | if (map_flags != BPF_ANY) | |
725 | return -EINVAL; | |
726 | ||
727 | if (index >= array->map.max_entries) | |
728 | return -E2BIG; | |
729 | ||
730 | ufd = *(u32 *)value; | |
d056a788 | 731 | new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd); |
2a36f0b9 WN |
732 | if (IS_ERR(new_ptr)) |
733 | return PTR_ERR(new_ptr); | |
04fd61ab | 734 | |
da765a2f DB |
735 | if (map->ops->map_poke_run) { |
736 | mutex_lock(&array->aux->poke_mutex); | |
737 | old_ptr = xchg(array->ptrs + index, new_ptr); | |
738 | map->ops->map_poke_run(map, index, old_ptr, new_ptr); | |
739 | mutex_unlock(&array->aux->poke_mutex); | |
740 | } else { | |
741 | old_ptr = xchg(array->ptrs + index, new_ptr); | |
742 | } | |
743 | ||
2a36f0b9 WN |
744 | if (old_ptr) |
745 | map->ops->map_fd_put_ptr(old_ptr); | |
04fd61ab AS |
746 | return 0; |
747 | } | |
748 | ||
2a36f0b9 | 749 | static int fd_array_map_delete_elem(struct bpf_map *map, void *key) |
04fd61ab AS |
750 | { |
751 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
2a36f0b9 | 752 | void *old_ptr; |
04fd61ab AS |
753 | u32 index = *(u32 *)key; |
754 | ||
755 | if (index >= array->map.max_entries) | |
756 | return -E2BIG; | |
757 | ||
da765a2f DB |
758 | if (map->ops->map_poke_run) { |
759 | mutex_lock(&array->aux->poke_mutex); | |
760 | old_ptr = xchg(array->ptrs + index, NULL); | |
761 | map->ops->map_poke_run(map, index, old_ptr, NULL); | |
762 | mutex_unlock(&array->aux->poke_mutex); | |
763 | } else { | |
764 | old_ptr = xchg(array->ptrs + index, NULL); | |
765 | } | |
766 | ||
2a36f0b9 WN |
767 | if (old_ptr) { |
768 | map->ops->map_fd_put_ptr(old_ptr); | |
04fd61ab AS |
769 | return 0; |
770 | } else { | |
771 | return -ENOENT; | |
772 | } | |
773 | } | |
774 | ||
d056a788 DB |
775 | static void *prog_fd_array_get_ptr(struct bpf_map *map, |
776 | struct file *map_file, int fd) | |
2a36f0b9 WN |
777 | { |
778 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
779 | struct bpf_prog *prog = bpf_prog_get(fd); | |
d056a788 | 780 | |
2a36f0b9 WN |
781 | if (IS_ERR(prog)) |
782 | return prog; | |
783 | ||
784 | if (!bpf_prog_array_compatible(array, prog)) { | |
785 | bpf_prog_put(prog); | |
786 | return ERR_PTR(-EINVAL); | |
787 | } | |
d056a788 | 788 | |
2a36f0b9 WN |
789 | return prog; |
790 | } | |
791 | ||
792 | static void prog_fd_array_put_ptr(void *ptr) | |
793 | { | |
1aacde3d | 794 | bpf_prog_put(ptr); |
2a36f0b9 WN |
795 | } |
796 | ||
14dc6f04 MKL |
797 | static u32 prog_fd_array_sys_lookup_elem(void *ptr) |
798 | { | |
799 | return ((struct bpf_prog *)ptr)->aux->id; | |
800 | } | |
801 | ||
04fd61ab | 802 | /* decrement refcnt of all bpf_progs that are stored in this map */ |
ba6b8de4 | 803 | static void bpf_fd_array_map_clear(struct bpf_map *map) |
04fd61ab AS |
804 | { |
805 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
806 | int i; | |
807 | ||
808 | for (i = 0; i < array->map.max_entries; i++) | |
2a36f0b9 | 809 | fd_array_map_delete_elem(map, &i); |
04fd61ab AS |
810 | } |
811 | ||
a7c19db3 YS |
812 | static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key, |
813 | struct seq_file *m) | |
814 | { | |
815 | void **elem, *ptr; | |
816 | u32 prog_id; | |
817 | ||
818 | rcu_read_lock(); | |
819 | ||
820 | elem = array_map_lookup_elem(map, key); | |
821 | if (elem) { | |
822 | ptr = READ_ONCE(*elem); | |
823 | if (ptr) { | |
824 | seq_printf(m, "%u: ", *(u32 *)key); | |
825 | prog_id = prog_fd_array_sys_lookup_elem(ptr); | |
826 | btf_type_seq_show(map->btf, map->btf_value_type_id, | |
827 | &prog_id, m); | |
828 | seq_puts(m, "\n"); | |
829 | } | |
830 | } | |
831 | ||
832 | rcu_read_unlock(); | |
833 | } | |
834 | ||
da765a2f DB |
835 | struct prog_poke_elem { |
836 | struct list_head list; | |
837 | struct bpf_prog_aux *aux; | |
838 | }; | |
839 | ||
840 | static int prog_array_map_poke_track(struct bpf_map *map, | |
841 | struct bpf_prog_aux *prog_aux) | |
842 | { | |
843 | struct prog_poke_elem *elem; | |
844 | struct bpf_array_aux *aux; | |
845 | int ret = 0; | |
846 | ||
847 | aux = container_of(map, struct bpf_array, map)->aux; | |
848 | mutex_lock(&aux->poke_mutex); | |
849 | list_for_each_entry(elem, &aux->poke_progs, list) { | |
850 | if (elem->aux == prog_aux) | |
851 | goto out; | |
852 | } | |
853 | ||
854 | elem = kmalloc(sizeof(*elem), GFP_KERNEL); | |
855 | if (!elem) { | |
856 | ret = -ENOMEM; | |
857 | goto out; | |
858 | } | |
859 | ||
860 | INIT_LIST_HEAD(&elem->list); | |
861 | /* We must track the program's aux info at this point in time | |
862 | * since the program pointer itself may not be stable yet, see | |
863 | * also comment in prog_array_map_poke_run(). | |
864 | */ | |
865 | elem->aux = prog_aux; | |
866 | ||
867 | list_add_tail(&elem->list, &aux->poke_progs); | |
868 | out: | |
869 | mutex_unlock(&aux->poke_mutex); | |
870 | return ret; | |
871 | } | |
872 | ||
873 | static void prog_array_map_poke_untrack(struct bpf_map *map, | |
874 | struct bpf_prog_aux *prog_aux) | |
875 | { | |
876 | struct prog_poke_elem *elem, *tmp; | |
877 | struct bpf_array_aux *aux; | |
878 | ||
879 | aux = container_of(map, struct bpf_array, map)->aux; | |
880 | mutex_lock(&aux->poke_mutex); | |
881 | list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) { | |
882 | if (elem->aux == prog_aux) { | |
883 | list_del_init(&elem->list); | |
884 | kfree(elem); | |
885 | break; | |
886 | } | |
887 | } | |
888 | mutex_unlock(&aux->poke_mutex); | |
889 | } | |
890 | ||
891 | static void prog_array_map_poke_run(struct bpf_map *map, u32 key, | |
892 | struct bpf_prog *old, | |
893 | struct bpf_prog *new) | |
894 | { | |
ebf7d1f5 | 895 | u8 *old_addr, *new_addr, *old_bypass_addr; |
da765a2f DB |
896 | struct prog_poke_elem *elem; |
897 | struct bpf_array_aux *aux; | |
898 | ||
da765a2f DB |
899 | aux = container_of(map, struct bpf_array, map)->aux; |
900 | WARN_ON_ONCE(!mutex_is_locked(&aux->poke_mutex)); | |
901 | ||
902 | list_for_each_entry(elem, &aux->poke_progs, list) { | |
903 | struct bpf_jit_poke_descriptor *poke; | |
904 | int i, ret; | |
905 | ||
906 | for (i = 0; i < elem->aux->size_poke_tab; i++) { | |
907 | poke = &elem->aux->poke_tab[i]; | |
908 | ||
909 | /* Few things to be aware of: | |
910 | * | |
911 | * 1) We can only ever access aux in this context, but | |
912 | * not aux->prog since it might not be stable yet and | |
913 | * there could be danger of use after free otherwise. | |
914 | * 2) Initially when we start tracking aux, the program | |
915 | * is not JITed yet and also does not have a kallsyms | |
cf71b174 MF |
916 | * entry. We skip these as poke->tailcall_target_stable |
917 | * is not active yet. The JIT will do the final fixup | |
918 | * before setting it stable. The various | |
919 | * poke->tailcall_target_stable are successively | |
920 | * activated, so tail call updates can arrive from here | |
921 | * while JIT is still finishing its final fixup for | |
922 | * non-activated poke entries. | |
da765a2f DB |
923 | * 3) On program teardown, the program's kallsym entry gets |
924 | * removed out of RCU callback, but we can only untrack | |
925 | * from sleepable context, therefore bpf_arch_text_poke() | |
926 | * might not see that this is in BPF text section and | |
927 | * bails out with -EINVAL. As these are unreachable since | |
928 | * RCU grace period already passed, we simply skip them. | |
929 | * 4) Also programs reaching refcount of zero while patching | |
930 | * is in progress is okay since we're protected under | |
931 | * poke_mutex and untrack the programs before the JIT | |
932 | * buffer is freed. When we're still in the middle of | |
933 | * patching and suddenly kallsyms entry of the program | |
934 | * gets evicted, we just skip the rest which is fine due | |
935 | * to point 3). | |
936 | * 5) Any other error happening below from bpf_arch_text_poke() | |
937 | * is a unexpected bug. | |
938 | */ | |
cf71b174 | 939 | if (!READ_ONCE(poke->tailcall_target_stable)) |
da765a2f DB |
940 | continue; |
941 | if (poke->reason != BPF_POKE_REASON_TAIL_CALL) | |
942 | continue; | |
943 | if (poke->tail_call.map != map || | |
944 | poke->tail_call.key != key) | |
945 | continue; | |
946 | ||
ebf7d1f5 MF |
947 | old_bypass_addr = old ? NULL : poke->bypass_addr; |
948 | old_addr = old ? (u8 *)old->bpf_func + poke->adj_off : NULL; | |
949 | new_addr = new ? (u8 *)new->bpf_func + poke->adj_off : NULL; | |
950 | ||
951 | if (new) { | |
952 | ret = bpf_arch_text_poke(poke->tailcall_target, | |
953 | BPF_MOD_JUMP, | |
954 | old_addr, new_addr); | |
955 | BUG_ON(ret < 0 && ret != -EINVAL); | |
956 | if (!old) { | |
957 | ret = bpf_arch_text_poke(poke->tailcall_bypass, | |
958 | BPF_MOD_JUMP, | |
959 | poke->bypass_addr, | |
960 | NULL); | |
961 | BUG_ON(ret < 0 && ret != -EINVAL); | |
962 | } | |
963 | } else { | |
964 | ret = bpf_arch_text_poke(poke->tailcall_bypass, | |
965 | BPF_MOD_JUMP, | |
966 | old_bypass_addr, | |
967 | poke->bypass_addr); | |
968 | BUG_ON(ret < 0 && ret != -EINVAL); | |
969 | /* let other CPUs finish the execution of program | |
970 | * so that it will not possible to expose them | |
971 | * to invalid nop, stack unwind, nop state | |
972 | */ | |
973 | if (!ret) | |
974 | synchronize_rcu(); | |
975 | ret = bpf_arch_text_poke(poke->tailcall_target, | |
976 | BPF_MOD_JUMP, | |
977 | old_addr, NULL); | |
978 | BUG_ON(ret < 0 && ret != -EINVAL); | |
979 | } | |
da765a2f DB |
980 | } |
981 | } | |
982 | } | |
983 | ||
984 | static void prog_array_map_clear_deferred(struct work_struct *work) | |
985 | { | |
986 | struct bpf_map *map = container_of(work, struct bpf_array_aux, | |
987 | work)->map; | |
988 | bpf_fd_array_map_clear(map); | |
989 | bpf_map_put(map); | |
990 | } | |
991 | ||
992 | static void prog_array_map_clear(struct bpf_map *map) | |
993 | { | |
994 | struct bpf_array_aux *aux = container_of(map, struct bpf_array, | |
995 | map)->aux; | |
996 | bpf_map_inc(map); | |
997 | schedule_work(&aux->work); | |
998 | } | |
999 | ||
2beee5f5 DB |
1000 | static struct bpf_map *prog_array_map_alloc(union bpf_attr *attr) |
1001 | { | |
1002 | struct bpf_array_aux *aux; | |
1003 | struct bpf_map *map; | |
1004 | ||
6d192c79 | 1005 | aux = kzalloc(sizeof(*aux), GFP_KERNEL_ACCOUNT); |
2beee5f5 DB |
1006 | if (!aux) |
1007 | return ERR_PTR(-ENOMEM); | |
1008 | ||
da765a2f DB |
1009 | INIT_WORK(&aux->work, prog_array_map_clear_deferred); |
1010 | INIT_LIST_HEAD(&aux->poke_progs); | |
1011 | mutex_init(&aux->poke_mutex); | |
1012 | ||
2beee5f5 DB |
1013 | map = array_map_alloc(attr); |
1014 | if (IS_ERR(map)) { | |
1015 | kfree(aux); | |
1016 | return map; | |
1017 | } | |
1018 | ||
1019 | container_of(map, struct bpf_array, map)->aux = aux; | |
da765a2f DB |
1020 | aux->map = map; |
1021 | ||
2beee5f5 DB |
1022 | return map; |
1023 | } | |
1024 | ||
1025 | static void prog_array_map_free(struct bpf_map *map) | |
1026 | { | |
da765a2f | 1027 | struct prog_poke_elem *elem, *tmp; |
2beee5f5 DB |
1028 | struct bpf_array_aux *aux; |
1029 | ||
1030 | aux = container_of(map, struct bpf_array, map)->aux; | |
da765a2f DB |
1031 | list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) { |
1032 | list_del_init(&elem->list); | |
1033 | kfree(elem); | |
1034 | } | |
2beee5f5 DB |
1035 | kfree(aux); |
1036 | fd_array_map_free(map); | |
1037 | } | |
1038 | ||
f4d05259 MKL |
1039 | /* prog_array->aux->{type,jited} is a runtime binding. |
1040 | * Doing static check alone in the verifier is not enough. | |
1041 | * Thus, prog_array_map cannot be used as an inner_map | |
1042 | * and map_meta_equal is not implemented. | |
1043 | */ | |
2872e9ac | 1044 | static int prog_array_map_btf_id; |
40077e0c | 1045 | const struct bpf_map_ops prog_array_map_ops = { |
ad46061f | 1046 | .map_alloc_check = fd_array_map_alloc_check, |
2beee5f5 DB |
1047 | .map_alloc = prog_array_map_alloc, |
1048 | .map_free = prog_array_map_free, | |
da765a2f DB |
1049 | .map_poke_track = prog_array_map_poke_track, |
1050 | .map_poke_untrack = prog_array_map_poke_untrack, | |
1051 | .map_poke_run = prog_array_map_poke_run, | |
04fd61ab | 1052 | .map_get_next_key = array_map_get_next_key, |
2a36f0b9 | 1053 | .map_lookup_elem = fd_array_map_lookup_elem, |
2a36f0b9 WN |
1054 | .map_delete_elem = fd_array_map_delete_elem, |
1055 | .map_fd_get_ptr = prog_fd_array_get_ptr, | |
1056 | .map_fd_put_ptr = prog_fd_array_put_ptr, | |
14dc6f04 | 1057 | .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem, |
da765a2f | 1058 | .map_release_uref = prog_array_map_clear, |
a7c19db3 | 1059 | .map_seq_show_elem = prog_array_map_seq_show_elem, |
2872e9ac AI |
1060 | .map_btf_name = "bpf_array", |
1061 | .map_btf_id = &prog_array_map_btf_id, | |
04fd61ab AS |
1062 | }; |
1063 | ||
3b1efb19 DB |
1064 | static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file, |
1065 | struct file *map_file) | |
ea317b26 | 1066 | { |
3b1efb19 DB |
1067 | struct bpf_event_entry *ee; |
1068 | ||
858d68f1 | 1069 | ee = kzalloc(sizeof(*ee), GFP_ATOMIC); |
3b1efb19 DB |
1070 | if (ee) { |
1071 | ee->event = perf_file->private_data; | |
1072 | ee->perf_file = perf_file; | |
1073 | ee->map_file = map_file; | |
1074 | } | |
1075 | ||
1076 | return ee; | |
1077 | } | |
1078 | ||
1079 | static void __bpf_event_entry_free(struct rcu_head *rcu) | |
1080 | { | |
1081 | struct bpf_event_entry *ee; | |
1082 | ||
1083 | ee = container_of(rcu, struct bpf_event_entry, rcu); | |
1084 | fput(ee->perf_file); | |
1085 | kfree(ee); | |
1086 | } | |
1087 | ||
1088 | static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee) | |
1089 | { | |
1090 | call_rcu(&ee->rcu, __bpf_event_entry_free); | |
ea317b26 KX |
1091 | } |
1092 | ||
d056a788 DB |
1093 | static void *perf_event_fd_array_get_ptr(struct bpf_map *map, |
1094 | struct file *map_file, int fd) | |
ea317b26 | 1095 | { |
3b1efb19 DB |
1096 | struct bpf_event_entry *ee; |
1097 | struct perf_event *event; | |
1098 | struct file *perf_file; | |
f91840a3 | 1099 | u64 value; |
ea317b26 | 1100 | |
3b1efb19 DB |
1101 | perf_file = perf_event_get(fd); |
1102 | if (IS_ERR(perf_file)) | |
1103 | return perf_file; | |
e03e7ee3 | 1104 | |
f91840a3 | 1105 | ee = ERR_PTR(-EOPNOTSUPP); |
3b1efb19 | 1106 | event = perf_file->private_data; |
97562633 | 1107 | if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP) |
3b1efb19 DB |
1108 | goto err_out; |
1109 | ||
f91840a3 AS |
1110 | ee = bpf_event_entry_gen(perf_file, map_file); |
1111 | if (ee) | |
1112 | return ee; | |
1113 | ee = ERR_PTR(-ENOMEM); | |
3b1efb19 DB |
1114 | err_out: |
1115 | fput(perf_file); | |
1116 | return ee; | |
ea317b26 KX |
1117 | } |
1118 | ||
1119 | static void perf_event_fd_array_put_ptr(void *ptr) | |
1120 | { | |
3b1efb19 DB |
1121 | bpf_event_entry_free_rcu(ptr); |
1122 | } | |
1123 | ||
1124 | static void perf_event_fd_array_release(struct bpf_map *map, | |
1125 | struct file *map_file) | |
1126 | { | |
1127 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
1128 | struct bpf_event_entry *ee; | |
1129 | int i; | |
1130 | ||
792caccc SL |
1131 | if (map->map_flags & BPF_F_PRESERVE_ELEMS) |
1132 | return; | |
1133 | ||
3b1efb19 DB |
1134 | rcu_read_lock(); |
1135 | for (i = 0; i < array->map.max_entries; i++) { | |
1136 | ee = READ_ONCE(array->ptrs[i]); | |
1137 | if (ee && ee->map_file == map_file) | |
1138 | fd_array_map_delete_elem(map, &i); | |
1139 | } | |
1140 | rcu_read_unlock(); | |
ea317b26 KX |
1141 | } |
1142 | ||
792caccc SL |
1143 | static void perf_event_fd_array_map_free(struct bpf_map *map) |
1144 | { | |
1145 | if (map->map_flags & BPF_F_PRESERVE_ELEMS) | |
1146 | bpf_fd_array_map_clear(map); | |
1147 | fd_array_map_free(map); | |
1148 | } | |
1149 | ||
2872e9ac | 1150 | static int perf_event_array_map_btf_id; |
40077e0c | 1151 | const struct bpf_map_ops perf_event_array_map_ops = { |
f4d05259 | 1152 | .map_meta_equal = bpf_map_meta_equal, |
ad46061f JK |
1153 | .map_alloc_check = fd_array_map_alloc_check, |
1154 | .map_alloc = array_map_alloc, | |
792caccc | 1155 | .map_free = perf_event_fd_array_map_free, |
ea317b26 KX |
1156 | .map_get_next_key = array_map_get_next_key, |
1157 | .map_lookup_elem = fd_array_map_lookup_elem, | |
ea317b26 KX |
1158 | .map_delete_elem = fd_array_map_delete_elem, |
1159 | .map_fd_get_ptr = perf_event_fd_array_get_ptr, | |
1160 | .map_fd_put_ptr = perf_event_fd_array_put_ptr, | |
3b1efb19 | 1161 | .map_release = perf_event_fd_array_release, |
e8d2bec0 | 1162 | .map_check_btf = map_check_no_btf, |
2872e9ac AI |
1163 | .map_btf_name = "bpf_array", |
1164 | .map_btf_id = &perf_event_array_map_btf_id, | |
ea317b26 KX |
1165 | }; |
1166 | ||
60d20f91 | 1167 | #ifdef CONFIG_CGROUPS |
4ed8ec52 MKL |
1168 | static void *cgroup_fd_array_get_ptr(struct bpf_map *map, |
1169 | struct file *map_file /* not used */, | |
1170 | int fd) | |
1171 | { | |
1172 | return cgroup_get_from_fd(fd); | |
1173 | } | |
1174 | ||
1175 | static void cgroup_fd_array_put_ptr(void *ptr) | |
1176 | { | |
1177 | /* cgroup_put free cgrp after a rcu grace period */ | |
1178 | cgroup_put(ptr); | |
1179 | } | |
1180 | ||
1181 | static void cgroup_fd_array_free(struct bpf_map *map) | |
1182 | { | |
1183 | bpf_fd_array_map_clear(map); | |
1184 | fd_array_map_free(map); | |
1185 | } | |
1186 | ||
2872e9ac | 1187 | static int cgroup_array_map_btf_id; |
40077e0c | 1188 | const struct bpf_map_ops cgroup_array_map_ops = { |
f4d05259 | 1189 | .map_meta_equal = bpf_map_meta_equal, |
ad46061f JK |
1190 | .map_alloc_check = fd_array_map_alloc_check, |
1191 | .map_alloc = array_map_alloc, | |
4ed8ec52 MKL |
1192 | .map_free = cgroup_fd_array_free, |
1193 | .map_get_next_key = array_map_get_next_key, | |
1194 | .map_lookup_elem = fd_array_map_lookup_elem, | |
1195 | .map_delete_elem = fd_array_map_delete_elem, | |
1196 | .map_fd_get_ptr = cgroup_fd_array_get_ptr, | |
1197 | .map_fd_put_ptr = cgroup_fd_array_put_ptr, | |
e8d2bec0 | 1198 | .map_check_btf = map_check_no_btf, |
2872e9ac AI |
1199 | .map_btf_name = "bpf_array", |
1200 | .map_btf_id = &cgroup_array_map_btf_id, | |
4ed8ec52 | 1201 | }; |
4ed8ec52 | 1202 | #endif |
56f668df MKL |
1203 | |
1204 | static struct bpf_map *array_of_map_alloc(union bpf_attr *attr) | |
1205 | { | |
1206 | struct bpf_map *map, *inner_map_meta; | |
1207 | ||
1208 | inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd); | |
1209 | if (IS_ERR(inner_map_meta)) | |
1210 | return inner_map_meta; | |
1211 | ||
ad46061f | 1212 | map = array_map_alloc(attr); |
56f668df MKL |
1213 | if (IS_ERR(map)) { |
1214 | bpf_map_meta_free(inner_map_meta); | |
1215 | return map; | |
1216 | } | |
1217 | ||
1218 | map->inner_map_meta = inner_map_meta; | |
1219 | ||
1220 | return map; | |
1221 | } | |
1222 | ||
1223 | static void array_of_map_free(struct bpf_map *map) | |
1224 | { | |
1225 | /* map->inner_map_meta is only accessed by syscall which | |
1226 | * is protected by fdget/fdput. | |
1227 | */ | |
1228 | bpf_map_meta_free(map->inner_map_meta); | |
1229 | bpf_fd_array_map_clear(map); | |
1230 | fd_array_map_free(map); | |
1231 | } | |
1232 | ||
1233 | static void *array_of_map_lookup_elem(struct bpf_map *map, void *key) | |
1234 | { | |
1235 | struct bpf_map **inner_map = array_map_lookup_elem(map, key); | |
1236 | ||
1237 | if (!inner_map) | |
1238 | return NULL; | |
1239 | ||
1240 | return READ_ONCE(*inner_map); | |
1241 | } | |
1242 | ||
4a8f87e6 | 1243 | static int array_of_map_gen_lookup(struct bpf_map *map, |
7b0c2a05 DB |
1244 | struct bpf_insn *insn_buf) |
1245 | { | |
b2157399 | 1246 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
7b0c2a05 DB |
1247 | u32 elem_size = round_up(map->value_size, 8); |
1248 | struct bpf_insn *insn = insn_buf; | |
1249 | const int ret = BPF_REG_0; | |
1250 | const int map_ptr = BPF_REG_1; | |
1251 | const int index = BPF_REG_2; | |
1252 | ||
1253 | *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value)); | |
1254 | *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0); | |
2c78ee89 | 1255 | if (!map->bypass_spec_v1) { |
b2157399 AS |
1256 | *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6); |
1257 | *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask); | |
1258 | } else { | |
1259 | *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5); | |
1260 | } | |
7b0c2a05 DB |
1261 | if (is_power_of_2(elem_size)) |
1262 | *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size)); | |
1263 | else | |
1264 | *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size); | |
1265 | *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr); | |
1266 | *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0); | |
1267 | *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1); | |
1268 | *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1); | |
1269 | *insn++ = BPF_MOV64_IMM(ret, 0); | |
1270 | ||
1271 | return insn - insn_buf; | |
1272 | } | |
1273 | ||
2872e9ac | 1274 | static int array_of_maps_map_btf_id; |
40077e0c | 1275 | const struct bpf_map_ops array_of_maps_map_ops = { |
ad46061f | 1276 | .map_alloc_check = fd_array_map_alloc_check, |
56f668df MKL |
1277 | .map_alloc = array_of_map_alloc, |
1278 | .map_free = array_of_map_free, | |
1279 | .map_get_next_key = array_map_get_next_key, | |
1280 | .map_lookup_elem = array_of_map_lookup_elem, | |
1281 | .map_delete_elem = fd_array_map_delete_elem, | |
1282 | .map_fd_get_ptr = bpf_map_fd_get_ptr, | |
1283 | .map_fd_put_ptr = bpf_map_fd_put_ptr, | |
14dc6f04 | 1284 | .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem, |
7b0c2a05 | 1285 | .map_gen_lookup = array_of_map_gen_lookup, |
e8d2bec0 | 1286 | .map_check_btf = map_check_no_btf, |
2872e9ac AI |
1287 | .map_btf_name = "bpf_array", |
1288 | .map_btf_id = &array_of_maps_map_btf_id, | |
56f668df | 1289 | }; |