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> |
28fbcfa0 | 13 | |
56f668df MKL |
14 | #include "map_in_map.h" |
15 | ||
6e71b04a | 16 | #define ARRAY_CREATE_FLAG_MASK \ |
591fe988 | 17 | (BPF_F_NUMA_NODE | BPF_F_ACCESS_MASK) |
6e71b04a | 18 | |
a10423b8 AS |
19 | static void bpf_array_free_percpu(struct bpf_array *array) |
20 | { | |
21 | int i; | |
22 | ||
32fff239 | 23 | for (i = 0; i < array->map.max_entries; i++) { |
a10423b8 | 24 | free_percpu(array->pptrs[i]); |
32fff239 ED |
25 | cond_resched(); |
26 | } | |
a10423b8 AS |
27 | } |
28 | ||
29 | static int bpf_array_alloc_percpu(struct bpf_array *array) | |
30 | { | |
31 | void __percpu *ptr; | |
32 | int i; | |
33 | ||
34 | for (i = 0; i < array->map.max_entries; i++) { | |
35 | ptr = __alloc_percpu_gfp(array->elem_size, 8, | |
36 | GFP_USER | __GFP_NOWARN); | |
37 | if (!ptr) { | |
38 | bpf_array_free_percpu(array); | |
39 | return -ENOMEM; | |
40 | } | |
41 | array->pptrs[i] = ptr; | |
32fff239 | 42 | cond_resched(); |
a10423b8 AS |
43 | } |
44 | ||
45 | return 0; | |
46 | } | |
47 | ||
28fbcfa0 | 48 | /* Called from syscall */ |
5dc4c4b7 | 49 | int array_map_alloc_check(union bpf_attr *attr) |
28fbcfa0 | 50 | { |
a10423b8 | 51 | bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY; |
96eabe7a | 52 | int numa_node = bpf_map_attr_numa_node(attr); |
28fbcfa0 AS |
53 | |
54 | /* check sanity of attributes */ | |
55 | if (attr->max_entries == 0 || attr->key_size != 4 || | |
6e71b04a CF |
56 | attr->value_size == 0 || |
57 | attr->map_flags & ~ARRAY_CREATE_FLAG_MASK || | |
591fe988 | 58 | !bpf_map_flags_access_ok(attr->map_flags) || |
96eabe7a | 59 | (percpu && numa_node != NUMA_NO_NODE)) |
ad46061f | 60 | return -EINVAL; |
28fbcfa0 | 61 | |
7984c27c | 62 | if (attr->value_size > KMALLOC_MAX_SIZE) |
01b3f521 AS |
63 | /* if value_size is bigger, the user space won't be able to |
64 | * access the elements. | |
65 | */ | |
ad46061f JK |
66 | return -E2BIG; |
67 | ||
68 | return 0; | |
69 | } | |
70 | ||
71 | static struct bpf_map *array_map_alloc(union bpf_attr *attr) | |
72 | { | |
73 | bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY; | |
9c2d63b8 | 74 | int ret, numa_node = bpf_map_attr_numa_node(attr); |
ad46061f JK |
75 | u32 elem_size, index_mask, max_entries; |
76 | bool unpriv = !capable(CAP_SYS_ADMIN); | |
9c2d63b8 | 77 | u64 cost, array_size, mask64; |
b936ca64 | 78 | struct bpf_map_memory mem; |
ad46061f | 79 | struct bpf_array *array; |
01b3f521 | 80 | |
28fbcfa0 AS |
81 | elem_size = round_up(attr->value_size, 8); |
82 | ||
b2157399 | 83 | max_entries = attr->max_entries; |
b2157399 | 84 | |
bbeb6e43 DB |
85 | /* On 32 bit archs roundup_pow_of_two() with max_entries that has |
86 | * upper most bit set in u32 space is undefined behavior due to | |
87 | * resulting 1U << 32, so do it manually here in u64 space. | |
88 | */ | |
89 | mask64 = fls_long(max_entries - 1); | |
90 | mask64 = 1ULL << mask64; | |
91 | mask64 -= 1; | |
92 | ||
93 | index_mask = mask64; | |
94 | if (unpriv) { | |
b2157399 AS |
95 | /* round up array size to nearest power of 2, |
96 | * since cpu will speculate within index_mask limits | |
97 | */ | |
98 | max_entries = index_mask + 1; | |
bbeb6e43 DB |
99 | /* Check for overflows. */ |
100 | if (max_entries < attr->max_entries) | |
101 | return ERR_PTR(-E2BIG); | |
102 | } | |
b2157399 | 103 | |
a10423b8 AS |
104 | array_size = sizeof(*array); |
105 | if (percpu) | |
b2157399 | 106 | array_size += (u64) max_entries * sizeof(void *); |
a10423b8 | 107 | else |
b2157399 | 108 | array_size += (u64) max_entries * elem_size; |
a10423b8 AS |
109 | |
110 | /* make sure there is no u32 overflow later in round_up() */ | |
9c2d63b8 | 111 | cost = array_size; |
c85d6913 | 112 | if (percpu) |
9c2d63b8 | 113 | cost += (u64)attr->max_entries * elem_size * num_possible_cpus(); |
9c2d63b8 | 114 | |
b936ca64 | 115 | ret = bpf_map_charge_init(&mem, cost); |
9c2d63b8 DB |
116 | if (ret < 0) |
117 | return ERR_PTR(ret); | |
daaf427c | 118 | |
28fbcfa0 | 119 | /* allocate all map elements and zero-initialize them */ |
96eabe7a | 120 | array = bpf_map_area_alloc(array_size, numa_node); |
b936ca64 RG |
121 | if (!array) { |
122 | bpf_map_charge_finish(&mem); | |
d407bd25 | 123 | return ERR_PTR(-ENOMEM); |
b936ca64 | 124 | } |
b2157399 AS |
125 | array->index_mask = index_mask; |
126 | array->map.unpriv_array = unpriv; | |
28fbcfa0 AS |
127 | |
128 | /* copy mandatory map attributes */ | |
32852649 | 129 | bpf_map_init_from_attr(&array->map, attr); |
b936ca64 | 130 | bpf_map_charge_move(&array->map.memory, &mem); |
28fbcfa0 AS |
131 | array->elem_size = elem_size; |
132 | ||
9c2d63b8 | 133 | if (percpu && bpf_array_alloc_percpu(array)) { |
b936ca64 | 134 | bpf_map_charge_finish(&array->map.memory); |
d407bd25 | 135 | bpf_map_area_free(array); |
a10423b8 AS |
136 | return ERR_PTR(-ENOMEM); |
137 | } | |
a10423b8 | 138 | |
28fbcfa0 | 139 | return &array->map; |
28fbcfa0 AS |
140 | } |
141 | ||
142 | /* Called from syscall or from eBPF program */ | |
143 | static void *array_map_lookup_elem(struct bpf_map *map, void *key) | |
144 | { | |
145 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
146 | u32 index = *(u32 *)key; | |
147 | ||
a10423b8 | 148 | if (unlikely(index >= array->map.max_entries)) |
28fbcfa0 AS |
149 | return NULL; |
150 | ||
b2157399 | 151 | return array->value + array->elem_size * (index & array->index_mask); |
28fbcfa0 AS |
152 | } |
153 | ||
d8eca5bb DB |
154 | static int array_map_direct_value_addr(const struct bpf_map *map, u64 *imm, |
155 | u32 off) | |
156 | { | |
157 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
158 | ||
159 | if (map->max_entries != 1) | |
160 | return -ENOTSUPP; | |
161 | if (off >= map->value_size) | |
162 | return -EINVAL; | |
163 | ||
164 | *imm = (unsigned long)array->value; | |
165 | return 0; | |
166 | } | |
167 | ||
168 | static int array_map_direct_value_meta(const struct bpf_map *map, u64 imm, | |
169 | u32 *off) | |
170 | { | |
171 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
172 | u64 base = (unsigned long)array->value; | |
173 | u64 range = array->elem_size; | |
174 | ||
175 | if (map->max_entries != 1) | |
176 | return -ENOTSUPP; | |
177 | if (imm < base || imm >= base + range) | |
178 | return -ENOENT; | |
179 | ||
180 | *off = imm - base; | |
181 | return 0; | |
182 | } | |
183 | ||
81ed18ab AS |
184 | /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */ |
185 | static u32 array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf) | |
186 | { | |
b2157399 | 187 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
81ed18ab | 188 | struct bpf_insn *insn = insn_buf; |
fad73a1a | 189 | u32 elem_size = round_up(map->value_size, 8); |
81ed18ab AS |
190 | const int ret = BPF_REG_0; |
191 | const int map_ptr = BPF_REG_1; | |
192 | const int index = BPF_REG_2; | |
193 | ||
194 | *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value)); | |
195 | *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0); | |
b2157399 AS |
196 | if (map->unpriv_array) { |
197 | *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4); | |
198 | *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask); | |
199 | } else { | |
200 | *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3); | |
201 | } | |
fad73a1a MKL |
202 | |
203 | if (is_power_of_2(elem_size)) { | |
81ed18ab AS |
204 | *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size)); |
205 | } else { | |
206 | *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size); | |
207 | } | |
208 | *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr); | |
209 | *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1); | |
210 | *insn++ = BPF_MOV64_IMM(ret, 0); | |
211 | return insn - insn_buf; | |
212 | } | |
213 | ||
a10423b8 AS |
214 | /* Called from eBPF program */ |
215 | static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key) | |
216 | { | |
217 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
218 | u32 index = *(u32 *)key; | |
219 | ||
220 | if (unlikely(index >= array->map.max_entries)) | |
221 | return NULL; | |
222 | ||
b2157399 | 223 | return this_cpu_ptr(array->pptrs[index & array->index_mask]); |
a10423b8 AS |
224 | } |
225 | ||
15a07b33 AS |
226 | int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value) |
227 | { | |
228 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
229 | u32 index = *(u32 *)key; | |
230 | void __percpu *pptr; | |
231 | int cpu, off = 0; | |
232 | u32 size; | |
233 | ||
234 | if (unlikely(index >= array->map.max_entries)) | |
235 | return -ENOENT; | |
236 | ||
237 | /* per_cpu areas are zero-filled and bpf programs can only | |
238 | * access 'value_size' of them, so copying rounded areas | |
239 | * will not leak any kernel data | |
240 | */ | |
241 | size = round_up(map->value_size, 8); | |
242 | rcu_read_lock(); | |
b2157399 | 243 | pptr = array->pptrs[index & array->index_mask]; |
15a07b33 AS |
244 | for_each_possible_cpu(cpu) { |
245 | bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size); | |
246 | off += size; | |
247 | } | |
248 | rcu_read_unlock(); | |
249 | return 0; | |
250 | } | |
251 | ||
28fbcfa0 AS |
252 | /* Called from syscall */ |
253 | static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key) | |
254 | { | |
255 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
8fe45924 | 256 | u32 index = key ? *(u32 *)key : U32_MAX; |
28fbcfa0 AS |
257 | u32 *next = (u32 *)next_key; |
258 | ||
259 | if (index >= array->map.max_entries) { | |
260 | *next = 0; | |
261 | return 0; | |
262 | } | |
263 | ||
264 | if (index == array->map.max_entries - 1) | |
265 | return -ENOENT; | |
266 | ||
267 | *next = index + 1; | |
268 | return 0; | |
269 | } | |
270 | ||
271 | /* Called from syscall or from eBPF program */ | |
272 | static int array_map_update_elem(struct bpf_map *map, void *key, void *value, | |
273 | u64 map_flags) | |
274 | { | |
275 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
276 | u32 index = *(u32 *)key; | |
96049f3a | 277 | char *val; |
28fbcfa0 | 278 | |
96049f3a | 279 | if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST)) |
28fbcfa0 AS |
280 | /* unknown flags */ |
281 | return -EINVAL; | |
282 | ||
a10423b8 | 283 | if (unlikely(index >= array->map.max_entries)) |
28fbcfa0 AS |
284 | /* all elements were pre-allocated, cannot insert a new one */ |
285 | return -E2BIG; | |
286 | ||
96049f3a | 287 | if (unlikely(map_flags & BPF_NOEXIST)) |
daaf427c | 288 | /* all elements already exist */ |
28fbcfa0 AS |
289 | return -EEXIST; |
290 | ||
96049f3a AS |
291 | if (unlikely((map_flags & BPF_F_LOCK) && |
292 | !map_value_has_spin_lock(map))) | |
293 | return -EINVAL; | |
294 | ||
295 | if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
b2157399 | 296 | memcpy(this_cpu_ptr(array->pptrs[index & array->index_mask]), |
a10423b8 | 297 | value, map->value_size); |
96049f3a AS |
298 | } else { |
299 | val = array->value + | |
300 | array->elem_size * (index & array->index_mask); | |
301 | if (map_flags & BPF_F_LOCK) | |
302 | copy_map_value_locked(map, val, value, false); | |
303 | else | |
304 | copy_map_value(map, val, value); | |
305 | } | |
28fbcfa0 AS |
306 | return 0; |
307 | } | |
308 | ||
15a07b33 AS |
309 | int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value, |
310 | u64 map_flags) | |
311 | { | |
312 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
313 | u32 index = *(u32 *)key; | |
314 | void __percpu *pptr; | |
315 | int cpu, off = 0; | |
316 | u32 size; | |
317 | ||
318 | if (unlikely(map_flags > BPF_EXIST)) | |
319 | /* unknown flags */ | |
320 | return -EINVAL; | |
321 | ||
322 | if (unlikely(index >= array->map.max_entries)) | |
323 | /* all elements were pre-allocated, cannot insert a new one */ | |
324 | return -E2BIG; | |
325 | ||
326 | if (unlikely(map_flags == BPF_NOEXIST)) | |
327 | /* all elements already exist */ | |
328 | return -EEXIST; | |
329 | ||
330 | /* the user space will provide round_up(value_size, 8) bytes that | |
331 | * will be copied into per-cpu area. bpf programs can only access | |
332 | * value_size of it. During lookup the same extra bytes will be | |
333 | * returned or zeros which were zero-filled by percpu_alloc, | |
334 | * so no kernel data leaks possible | |
335 | */ | |
336 | size = round_up(map->value_size, 8); | |
337 | rcu_read_lock(); | |
b2157399 | 338 | pptr = array->pptrs[index & array->index_mask]; |
15a07b33 AS |
339 | for_each_possible_cpu(cpu) { |
340 | bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size); | |
341 | off += size; | |
342 | } | |
343 | rcu_read_unlock(); | |
344 | return 0; | |
345 | } | |
346 | ||
28fbcfa0 AS |
347 | /* Called from syscall or from eBPF program */ |
348 | static int array_map_delete_elem(struct bpf_map *map, void *key) | |
349 | { | |
350 | return -EINVAL; | |
351 | } | |
352 | ||
353 | /* Called when map->refcnt goes to zero, either from workqueue or from syscall */ | |
354 | static void array_map_free(struct bpf_map *map) | |
355 | { | |
356 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
357 | ||
358 | /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0, | |
359 | * so the programs (can be more than one that used this map) were | |
360 | * disconnected from events. Wait for outstanding programs to complete | |
361 | * and free the array | |
362 | */ | |
363 | synchronize_rcu(); | |
364 | ||
a10423b8 AS |
365 | if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) |
366 | bpf_array_free_percpu(array); | |
367 | ||
d407bd25 | 368 | bpf_map_area_free(array); |
28fbcfa0 AS |
369 | } |
370 | ||
a26ca7c9 MKL |
371 | static void array_map_seq_show_elem(struct bpf_map *map, void *key, |
372 | struct seq_file *m) | |
373 | { | |
374 | void *value; | |
375 | ||
376 | rcu_read_lock(); | |
377 | ||
378 | value = array_map_lookup_elem(map, key); | |
379 | if (!value) { | |
380 | rcu_read_unlock(); | |
381 | return; | |
382 | } | |
383 | ||
2824ecb7 DB |
384 | if (map->btf_key_type_id) |
385 | seq_printf(m, "%u: ", *(u32 *)key); | |
9b2cf328 | 386 | btf_type_seq_show(map->btf, map->btf_value_type_id, value, m); |
a26ca7c9 MKL |
387 | seq_puts(m, "\n"); |
388 | ||
389 | rcu_read_unlock(); | |
390 | } | |
391 | ||
c7b27c37 YS |
392 | static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key, |
393 | struct seq_file *m) | |
394 | { | |
395 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
396 | u32 index = *(u32 *)key; | |
397 | void __percpu *pptr; | |
398 | int cpu; | |
399 | ||
400 | rcu_read_lock(); | |
401 | ||
402 | seq_printf(m, "%u: {\n", *(u32 *)key); | |
403 | pptr = array->pptrs[index & array->index_mask]; | |
404 | for_each_possible_cpu(cpu) { | |
405 | seq_printf(m, "\tcpu%d: ", cpu); | |
406 | btf_type_seq_show(map->btf, map->btf_value_type_id, | |
407 | per_cpu_ptr(pptr, cpu), m); | |
408 | seq_puts(m, "\n"); | |
409 | } | |
410 | seq_puts(m, "}\n"); | |
411 | ||
412 | rcu_read_unlock(); | |
413 | } | |
414 | ||
e8d2bec0 | 415 | static int array_map_check_btf(const struct bpf_map *map, |
1b2b234b | 416 | const struct btf *btf, |
e8d2bec0 DB |
417 | const struct btf_type *key_type, |
418 | const struct btf_type *value_type) | |
a26ca7c9 | 419 | { |
a26ca7c9 MKL |
420 | u32 int_data; |
421 | ||
2824ecb7 DB |
422 | /* One exception for keyless BTF: .bss/.data/.rodata map */ |
423 | if (btf_type_is_void(key_type)) { | |
424 | if (map->map_type != BPF_MAP_TYPE_ARRAY || | |
425 | map->max_entries != 1) | |
426 | return -EINVAL; | |
427 | ||
428 | if (BTF_INFO_KIND(value_type->info) != BTF_KIND_DATASEC) | |
429 | return -EINVAL; | |
430 | ||
431 | return 0; | |
432 | } | |
433 | ||
e8d2bec0 | 434 | if (BTF_INFO_KIND(key_type->info) != BTF_KIND_INT) |
a26ca7c9 MKL |
435 | return -EINVAL; |
436 | ||
437 | int_data = *(u32 *)(key_type + 1); | |
e8d2bec0 DB |
438 | /* bpf array can only take a u32 key. This check makes sure |
439 | * that the btf matches the attr used during map_create. | |
a26ca7c9 | 440 | */ |
e8d2bec0 | 441 | if (BTF_INT_BITS(int_data) != 32 || BTF_INT_OFFSET(int_data)) |
a26ca7c9 MKL |
442 | return -EINVAL; |
443 | ||
444 | return 0; | |
445 | } | |
446 | ||
40077e0c | 447 | const struct bpf_map_ops array_map_ops = { |
ad46061f | 448 | .map_alloc_check = array_map_alloc_check, |
28fbcfa0 AS |
449 | .map_alloc = array_map_alloc, |
450 | .map_free = array_map_free, | |
451 | .map_get_next_key = array_map_get_next_key, | |
452 | .map_lookup_elem = array_map_lookup_elem, | |
453 | .map_update_elem = array_map_update_elem, | |
454 | .map_delete_elem = array_map_delete_elem, | |
81ed18ab | 455 | .map_gen_lookup = array_map_gen_lookup, |
d8eca5bb DB |
456 | .map_direct_value_addr = array_map_direct_value_addr, |
457 | .map_direct_value_meta = array_map_direct_value_meta, | |
a26ca7c9 MKL |
458 | .map_seq_show_elem = array_map_seq_show_elem, |
459 | .map_check_btf = array_map_check_btf, | |
28fbcfa0 AS |
460 | }; |
461 | ||
40077e0c | 462 | const struct bpf_map_ops percpu_array_map_ops = { |
ad46061f | 463 | .map_alloc_check = array_map_alloc_check, |
a10423b8 AS |
464 | .map_alloc = array_map_alloc, |
465 | .map_free = array_map_free, | |
466 | .map_get_next_key = array_map_get_next_key, | |
467 | .map_lookup_elem = percpu_array_map_lookup_elem, | |
468 | .map_update_elem = array_map_update_elem, | |
469 | .map_delete_elem = array_map_delete_elem, | |
c7b27c37 | 470 | .map_seq_show_elem = percpu_array_map_seq_show_elem, |
e8d2bec0 | 471 | .map_check_btf = array_map_check_btf, |
a10423b8 AS |
472 | }; |
473 | ||
ad46061f | 474 | static int fd_array_map_alloc_check(union bpf_attr *attr) |
04fd61ab | 475 | { |
2a36f0b9 | 476 | /* only file descriptors can be stored in this type of map */ |
04fd61ab | 477 | if (attr->value_size != sizeof(u32)) |
ad46061f | 478 | return -EINVAL; |
591fe988 DB |
479 | /* Program read-only/write-only not supported for special maps yet. */ |
480 | if (attr->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) | |
481 | return -EINVAL; | |
ad46061f | 482 | return array_map_alloc_check(attr); |
04fd61ab AS |
483 | } |
484 | ||
2a36f0b9 | 485 | static void fd_array_map_free(struct bpf_map *map) |
04fd61ab AS |
486 | { |
487 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
488 | int i; | |
489 | ||
490 | synchronize_rcu(); | |
491 | ||
492 | /* make sure it's empty */ | |
493 | for (i = 0; i < array->map.max_entries; i++) | |
2a36f0b9 | 494 | BUG_ON(array->ptrs[i] != NULL); |
d407bd25 DB |
495 | |
496 | bpf_map_area_free(array); | |
04fd61ab AS |
497 | } |
498 | ||
2a36f0b9 | 499 | static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key) |
04fd61ab | 500 | { |
3b4a63f6 | 501 | return ERR_PTR(-EOPNOTSUPP); |
04fd61ab AS |
502 | } |
503 | ||
14dc6f04 MKL |
504 | /* only called from syscall */ |
505 | int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value) | |
506 | { | |
507 | void **elem, *ptr; | |
508 | int ret = 0; | |
509 | ||
510 | if (!map->ops->map_fd_sys_lookup_elem) | |
511 | return -ENOTSUPP; | |
512 | ||
513 | rcu_read_lock(); | |
514 | elem = array_map_lookup_elem(map, key); | |
515 | if (elem && (ptr = READ_ONCE(*elem))) | |
516 | *value = map->ops->map_fd_sys_lookup_elem(ptr); | |
517 | else | |
518 | ret = -ENOENT; | |
519 | rcu_read_unlock(); | |
520 | ||
521 | return ret; | |
522 | } | |
523 | ||
04fd61ab | 524 | /* only called from syscall */ |
d056a788 DB |
525 | int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file, |
526 | void *key, void *value, u64 map_flags) | |
04fd61ab AS |
527 | { |
528 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
2a36f0b9 | 529 | void *new_ptr, *old_ptr; |
04fd61ab AS |
530 | u32 index = *(u32 *)key, ufd; |
531 | ||
532 | if (map_flags != BPF_ANY) | |
533 | return -EINVAL; | |
534 | ||
535 | if (index >= array->map.max_entries) | |
536 | return -E2BIG; | |
537 | ||
538 | ufd = *(u32 *)value; | |
d056a788 | 539 | new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd); |
2a36f0b9 WN |
540 | if (IS_ERR(new_ptr)) |
541 | return PTR_ERR(new_ptr); | |
04fd61ab | 542 | |
2a36f0b9 WN |
543 | old_ptr = xchg(array->ptrs + index, new_ptr); |
544 | if (old_ptr) | |
545 | map->ops->map_fd_put_ptr(old_ptr); | |
04fd61ab AS |
546 | |
547 | return 0; | |
548 | } | |
549 | ||
2a36f0b9 | 550 | static int fd_array_map_delete_elem(struct bpf_map *map, void *key) |
04fd61ab AS |
551 | { |
552 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
2a36f0b9 | 553 | void *old_ptr; |
04fd61ab AS |
554 | u32 index = *(u32 *)key; |
555 | ||
556 | if (index >= array->map.max_entries) | |
557 | return -E2BIG; | |
558 | ||
2a36f0b9 WN |
559 | old_ptr = xchg(array->ptrs + index, NULL); |
560 | if (old_ptr) { | |
561 | map->ops->map_fd_put_ptr(old_ptr); | |
04fd61ab AS |
562 | return 0; |
563 | } else { | |
564 | return -ENOENT; | |
565 | } | |
566 | } | |
567 | ||
d056a788 DB |
568 | static void *prog_fd_array_get_ptr(struct bpf_map *map, |
569 | struct file *map_file, int fd) | |
2a36f0b9 WN |
570 | { |
571 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
572 | struct bpf_prog *prog = bpf_prog_get(fd); | |
d056a788 | 573 | |
2a36f0b9 WN |
574 | if (IS_ERR(prog)) |
575 | return prog; | |
576 | ||
577 | if (!bpf_prog_array_compatible(array, prog)) { | |
578 | bpf_prog_put(prog); | |
579 | return ERR_PTR(-EINVAL); | |
580 | } | |
d056a788 | 581 | |
2a36f0b9 WN |
582 | return prog; |
583 | } | |
584 | ||
585 | static void prog_fd_array_put_ptr(void *ptr) | |
586 | { | |
1aacde3d | 587 | bpf_prog_put(ptr); |
2a36f0b9 WN |
588 | } |
589 | ||
14dc6f04 MKL |
590 | static u32 prog_fd_array_sys_lookup_elem(void *ptr) |
591 | { | |
592 | return ((struct bpf_prog *)ptr)->aux->id; | |
593 | } | |
594 | ||
04fd61ab | 595 | /* decrement refcnt of all bpf_progs that are stored in this map */ |
ba6b8de4 | 596 | static void bpf_fd_array_map_clear(struct bpf_map *map) |
04fd61ab AS |
597 | { |
598 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
599 | int i; | |
600 | ||
601 | for (i = 0; i < array->map.max_entries; i++) | |
2a36f0b9 | 602 | fd_array_map_delete_elem(map, &i); |
04fd61ab AS |
603 | } |
604 | ||
a7c19db3 YS |
605 | static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key, |
606 | struct seq_file *m) | |
607 | { | |
608 | void **elem, *ptr; | |
609 | u32 prog_id; | |
610 | ||
611 | rcu_read_lock(); | |
612 | ||
613 | elem = array_map_lookup_elem(map, key); | |
614 | if (elem) { | |
615 | ptr = READ_ONCE(*elem); | |
616 | if (ptr) { | |
617 | seq_printf(m, "%u: ", *(u32 *)key); | |
618 | prog_id = prog_fd_array_sys_lookup_elem(ptr); | |
619 | btf_type_seq_show(map->btf, map->btf_value_type_id, | |
620 | &prog_id, m); | |
621 | seq_puts(m, "\n"); | |
622 | } | |
623 | } | |
624 | ||
625 | rcu_read_unlock(); | |
626 | } | |
627 | ||
40077e0c | 628 | const struct bpf_map_ops prog_array_map_ops = { |
ad46061f JK |
629 | .map_alloc_check = fd_array_map_alloc_check, |
630 | .map_alloc = array_map_alloc, | |
2a36f0b9 | 631 | .map_free = fd_array_map_free, |
04fd61ab | 632 | .map_get_next_key = array_map_get_next_key, |
2a36f0b9 | 633 | .map_lookup_elem = fd_array_map_lookup_elem, |
2a36f0b9 WN |
634 | .map_delete_elem = fd_array_map_delete_elem, |
635 | .map_fd_get_ptr = prog_fd_array_get_ptr, | |
636 | .map_fd_put_ptr = prog_fd_array_put_ptr, | |
14dc6f04 | 637 | .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem, |
ba6b8de4 | 638 | .map_release_uref = bpf_fd_array_map_clear, |
a7c19db3 | 639 | .map_seq_show_elem = prog_array_map_seq_show_elem, |
04fd61ab AS |
640 | }; |
641 | ||
3b1efb19 DB |
642 | static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file, |
643 | struct file *map_file) | |
ea317b26 | 644 | { |
3b1efb19 DB |
645 | struct bpf_event_entry *ee; |
646 | ||
858d68f1 | 647 | ee = kzalloc(sizeof(*ee), GFP_ATOMIC); |
3b1efb19 DB |
648 | if (ee) { |
649 | ee->event = perf_file->private_data; | |
650 | ee->perf_file = perf_file; | |
651 | ee->map_file = map_file; | |
652 | } | |
653 | ||
654 | return ee; | |
655 | } | |
656 | ||
657 | static void __bpf_event_entry_free(struct rcu_head *rcu) | |
658 | { | |
659 | struct bpf_event_entry *ee; | |
660 | ||
661 | ee = container_of(rcu, struct bpf_event_entry, rcu); | |
662 | fput(ee->perf_file); | |
663 | kfree(ee); | |
664 | } | |
665 | ||
666 | static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee) | |
667 | { | |
668 | call_rcu(&ee->rcu, __bpf_event_entry_free); | |
ea317b26 KX |
669 | } |
670 | ||
d056a788 DB |
671 | static void *perf_event_fd_array_get_ptr(struct bpf_map *map, |
672 | struct file *map_file, int fd) | |
ea317b26 | 673 | { |
3b1efb19 DB |
674 | struct bpf_event_entry *ee; |
675 | struct perf_event *event; | |
676 | struct file *perf_file; | |
f91840a3 | 677 | u64 value; |
ea317b26 | 678 | |
3b1efb19 DB |
679 | perf_file = perf_event_get(fd); |
680 | if (IS_ERR(perf_file)) | |
681 | return perf_file; | |
e03e7ee3 | 682 | |
f91840a3 | 683 | ee = ERR_PTR(-EOPNOTSUPP); |
3b1efb19 | 684 | event = perf_file->private_data; |
97562633 | 685 | if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP) |
3b1efb19 DB |
686 | goto err_out; |
687 | ||
f91840a3 AS |
688 | ee = bpf_event_entry_gen(perf_file, map_file); |
689 | if (ee) | |
690 | return ee; | |
691 | ee = ERR_PTR(-ENOMEM); | |
3b1efb19 DB |
692 | err_out: |
693 | fput(perf_file); | |
694 | return ee; | |
ea317b26 KX |
695 | } |
696 | ||
697 | static void perf_event_fd_array_put_ptr(void *ptr) | |
698 | { | |
3b1efb19 DB |
699 | bpf_event_entry_free_rcu(ptr); |
700 | } | |
701 | ||
702 | static void perf_event_fd_array_release(struct bpf_map *map, | |
703 | struct file *map_file) | |
704 | { | |
705 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
706 | struct bpf_event_entry *ee; | |
707 | int i; | |
708 | ||
709 | rcu_read_lock(); | |
710 | for (i = 0; i < array->map.max_entries; i++) { | |
711 | ee = READ_ONCE(array->ptrs[i]); | |
712 | if (ee && ee->map_file == map_file) | |
713 | fd_array_map_delete_elem(map, &i); | |
714 | } | |
715 | rcu_read_unlock(); | |
ea317b26 KX |
716 | } |
717 | ||
40077e0c | 718 | const struct bpf_map_ops perf_event_array_map_ops = { |
ad46061f JK |
719 | .map_alloc_check = fd_array_map_alloc_check, |
720 | .map_alloc = array_map_alloc, | |
3b1efb19 | 721 | .map_free = fd_array_map_free, |
ea317b26 KX |
722 | .map_get_next_key = array_map_get_next_key, |
723 | .map_lookup_elem = fd_array_map_lookup_elem, | |
ea317b26 KX |
724 | .map_delete_elem = fd_array_map_delete_elem, |
725 | .map_fd_get_ptr = perf_event_fd_array_get_ptr, | |
726 | .map_fd_put_ptr = perf_event_fd_array_put_ptr, | |
3b1efb19 | 727 | .map_release = perf_event_fd_array_release, |
e8d2bec0 | 728 | .map_check_btf = map_check_no_btf, |
ea317b26 KX |
729 | }; |
730 | ||
60d20f91 | 731 | #ifdef CONFIG_CGROUPS |
4ed8ec52 MKL |
732 | static void *cgroup_fd_array_get_ptr(struct bpf_map *map, |
733 | struct file *map_file /* not used */, | |
734 | int fd) | |
735 | { | |
736 | return cgroup_get_from_fd(fd); | |
737 | } | |
738 | ||
739 | static void cgroup_fd_array_put_ptr(void *ptr) | |
740 | { | |
741 | /* cgroup_put free cgrp after a rcu grace period */ | |
742 | cgroup_put(ptr); | |
743 | } | |
744 | ||
745 | static void cgroup_fd_array_free(struct bpf_map *map) | |
746 | { | |
747 | bpf_fd_array_map_clear(map); | |
748 | fd_array_map_free(map); | |
749 | } | |
750 | ||
40077e0c | 751 | const struct bpf_map_ops cgroup_array_map_ops = { |
ad46061f JK |
752 | .map_alloc_check = fd_array_map_alloc_check, |
753 | .map_alloc = array_map_alloc, | |
4ed8ec52 MKL |
754 | .map_free = cgroup_fd_array_free, |
755 | .map_get_next_key = array_map_get_next_key, | |
756 | .map_lookup_elem = fd_array_map_lookup_elem, | |
757 | .map_delete_elem = fd_array_map_delete_elem, | |
758 | .map_fd_get_ptr = cgroup_fd_array_get_ptr, | |
759 | .map_fd_put_ptr = cgroup_fd_array_put_ptr, | |
e8d2bec0 | 760 | .map_check_btf = map_check_no_btf, |
4ed8ec52 | 761 | }; |
4ed8ec52 | 762 | #endif |
56f668df MKL |
763 | |
764 | static struct bpf_map *array_of_map_alloc(union bpf_attr *attr) | |
765 | { | |
766 | struct bpf_map *map, *inner_map_meta; | |
767 | ||
768 | inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd); | |
769 | if (IS_ERR(inner_map_meta)) | |
770 | return inner_map_meta; | |
771 | ||
ad46061f | 772 | map = array_map_alloc(attr); |
56f668df MKL |
773 | if (IS_ERR(map)) { |
774 | bpf_map_meta_free(inner_map_meta); | |
775 | return map; | |
776 | } | |
777 | ||
778 | map->inner_map_meta = inner_map_meta; | |
779 | ||
780 | return map; | |
781 | } | |
782 | ||
783 | static void array_of_map_free(struct bpf_map *map) | |
784 | { | |
785 | /* map->inner_map_meta is only accessed by syscall which | |
786 | * is protected by fdget/fdput. | |
787 | */ | |
788 | bpf_map_meta_free(map->inner_map_meta); | |
789 | bpf_fd_array_map_clear(map); | |
790 | fd_array_map_free(map); | |
791 | } | |
792 | ||
793 | static void *array_of_map_lookup_elem(struct bpf_map *map, void *key) | |
794 | { | |
795 | struct bpf_map **inner_map = array_map_lookup_elem(map, key); | |
796 | ||
797 | if (!inner_map) | |
798 | return NULL; | |
799 | ||
800 | return READ_ONCE(*inner_map); | |
801 | } | |
802 | ||
7b0c2a05 DB |
803 | static u32 array_of_map_gen_lookup(struct bpf_map *map, |
804 | struct bpf_insn *insn_buf) | |
805 | { | |
b2157399 | 806 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
7b0c2a05 DB |
807 | u32 elem_size = round_up(map->value_size, 8); |
808 | struct bpf_insn *insn = insn_buf; | |
809 | const int ret = BPF_REG_0; | |
810 | const int map_ptr = BPF_REG_1; | |
811 | const int index = BPF_REG_2; | |
812 | ||
813 | *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value)); | |
814 | *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0); | |
b2157399 AS |
815 | if (map->unpriv_array) { |
816 | *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6); | |
817 | *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask); | |
818 | } else { | |
819 | *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5); | |
820 | } | |
7b0c2a05 DB |
821 | if (is_power_of_2(elem_size)) |
822 | *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size)); | |
823 | else | |
824 | *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size); | |
825 | *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr); | |
826 | *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0); | |
827 | *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1); | |
828 | *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1); | |
829 | *insn++ = BPF_MOV64_IMM(ret, 0); | |
830 | ||
831 | return insn - insn_buf; | |
832 | } | |
833 | ||
40077e0c | 834 | const struct bpf_map_ops array_of_maps_map_ops = { |
ad46061f | 835 | .map_alloc_check = fd_array_map_alloc_check, |
56f668df MKL |
836 | .map_alloc = array_of_map_alloc, |
837 | .map_free = array_of_map_free, | |
838 | .map_get_next_key = array_map_get_next_key, | |
839 | .map_lookup_elem = array_of_map_lookup_elem, | |
840 | .map_delete_elem = fd_array_map_delete_elem, | |
841 | .map_fd_get_ptr = bpf_map_fd_get_ptr, | |
842 | .map_fd_put_ptr = bpf_map_fd_put_ptr, | |
14dc6f04 | 843 | .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem, |
7b0c2a05 | 844 | .map_gen_lookup = array_of_map_gen_lookup, |
e8d2bec0 | 845 | .map_check_btf = map_check_no_btf, |
56f668df | 846 | }; |