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 | |
22 | static 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(pdata: array->pptrs[i]); |
28 | cond_resched(); |
29 | } |
30 | } |
31 | |
32 | static 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(map: &array->map, size: array->elem_size, align: 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 */ |
52 | int 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(access_flags: 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 | |
77 | return 0; |
78 | } |
79 | |
80 | static struct bpf_map *array_map_alloc(union bpf_attr *attr) |
81 | { |
82 | bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY; |
83 | int numa_node = bpf_map_attr_numa_node(attr); |
84 | u32 elem_size, index_mask, max_entries; |
85 | bool bypass_spec_v1 = bpf_bypass_spec_v1(); |
86 | u64 array_size, mask64; |
87 | struct bpf_array *array; |
88 | |
89 | elem_size = round_up(attr->value_size, 8); |
90 | |
91 | max_entries = attr->max_entries; |
92 | |
93 | /* On 32 bit archs roundup_pow_of_two() with max_entries that has |
94 | * upper most bit set in u32 space is undefined behavior due to |
95 | * resulting 1U << 32, so do it manually here in u64 space. |
96 | */ |
97 | mask64 = fls_long(l: max_entries - 1); |
98 | mask64 = 1ULL << mask64; |
99 | mask64 -= 1; |
100 | |
101 | index_mask = mask64; |
102 | if (!bypass_spec_v1) { |
103 | /* round up array size to nearest power of 2, |
104 | * since cpu will speculate within index_mask limits |
105 | */ |
106 | max_entries = index_mask + 1; |
107 | /* Check for overflows. */ |
108 | if (max_entries < attr->max_entries) |
109 | return ERR_PTR(error: -E2BIG); |
110 | } |
111 | |
112 | array_size = sizeof(*array); |
113 | if (percpu) { |
114 | array_size += (u64) max_entries * sizeof(void *); |
115 | } else { |
116 | /* rely on vmalloc() to return page-aligned memory and |
117 | * ensure array->value is exactly page-aligned |
118 | */ |
119 | if (attr->map_flags & BPF_F_MMAPABLE) { |
120 | array_size = PAGE_ALIGN(array_size); |
121 | array_size += PAGE_ALIGN((u64) max_entries * elem_size); |
122 | } else { |
123 | array_size += (u64) max_entries * elem_size; |
124 | } |
125 | } |
126 | |
127 | /* allocate all map elements and zero-initialize them */ |
128 | if (attr->map_flags & BPF_F_MMAPABLE) { |
129 | void *data; |
130 | |
131 | /* kmalloc'ed memory can't be mmap'ed, use explicit vmalloc */ |
132 | data = bpf_map_area_mmapable_alloc(size: array_size, numa_node); |
133 | if (!data) |
134 | return ERR_PTR(error: -ENOMEM); |
135 | array = data + PAGE_ALIGN(sizeof(struct bpf_array)) |
136 | - offsetof(struct bpf_array, value); |
137 | } else { |
138 | array = bpf_map_area_alloc(size: array_size, numa_node); |
139 | } |
140 | if (!array) |
141 | return ERR_PTR(error: -ENOMEM); |
142 | array->index_mask = index_mask; |
143 | array->map.bypass_spec_v1 = bypass_spec_v1; |
144 | |
145 | /* copy mandatory map attributes */ |
146 | bpf_map_init_from_attr(map: &array->map, attr); |
147 | array->elem_size = elem_size; |
148 | |
149 | if (percpu && bpf_array_alloc_percpu(array)) { |
150 | bpf_map_area_free(base: array); |
151 | return ERR_PTR(error: -ENOMEM); |
152 | } |
153 | |
154 | return &array->map; |
155 | } |
156 | |
157 | static void *array_map_elem_ptr(struct bpf_array* array, u32 index) |
158 | { |
159 | return array->value + (u64)array->elem_size * index; |
160 | } |
161 | |
162 | /* Called from syscall or from eBPF program */ |
163 | static void *array_map_lookup_elem(struct bpf_map *map, void *key) |
164 | { |
165 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
166 | u32 index = *(u32 *)key; |
167 | |
168 | if (unlikely(index >= array->map.max_entries)) |
169 | return NULL; |
170 | |
171 | return array->value + (u64)array->elem_size * (index & array->index_mask); |
172 | } |
173 | |
174 | static int array_map_direct_value_addr(const struct bpf_map *map, u64 *imm, |
175 | u32 off) |
176 | { |
177 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
178 | |
179 | if (map->max_entries != 1) |
180 | return -ENOTSUPP; |
181 | if (off >= map->value_size) |
182 | return -EINVAL; |
183 | |
184 | *imm = (unsigned long)array->value; |
185 | return 0; |
186 | } |
187 | |
188 | static int array_map_direct_value_meta(const struct bpf_map *map, u64 imm, |
189 | u32 *off) |
190 | { |
191 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
192 | u64 base = (unsigned long)array->value; |
193 | u64 range = array->elem_size; |
194 | |
195 | if (map->max_entries != 1) |
196 | return -ENOTSUPP; |
197 | if (imm < base || imm >= base + range) |
198 | return -ENOENT; |
199 | |
200 | *off = imm - base; |
201 | return 0; |
202 | } |
203 | |
204 | /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */ |
205 | static int array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf) |
206 | { |
207 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
208 | struct bpf_insn *insn = insn_buf; |
209 | u32 elem_size = array->elem_size; |
210 | const int ret = BPF_REG_0; |
211 | const int map_ptr = BPF_REG_1; |
212 | const int index = BPF_REG_2; |
213 | |
214 | if (map->map_flags & BPF_F_INNER_MAP) |
215 | return -EOPNOTSUPP; |
216 | |
217 | *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value)); |
218 | *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0); |
219 | if (!map->bypass_spec_v1) { |
220 | *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4); |
221 | *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask); |
222 | } else { |
223 | *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3); |
224 | } |
225 | |
226 | if (is_power_of_2(n: elem_size)) { |
227 | *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size)); |
228 | } else { |
229 | *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size); |
230 | } |
231 | *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr); |
232 | *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1); |
233 | *insn++ = BPF_MOV64_IMM(ret, 0); |
234 | return insn - insn_buf; |
235 | } |
236 | |
237 | /* Called from eBPF program */ |
238 | static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key) |
239 | { |
240 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
241 | u32 index = *(u32 *)key; |
242 | |
243 | if (unlikely(index >= array->map.max_entries)) |
244 | return NULL; |
245 | |
246 | return this_cpu_ptr(array->pptrs[index & array->index_mask]); |
247 | } |
248 | |
249 | static void *percpu_array_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu) |
250 | { |
251 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
252 | u32 index = *(u32 *)key; |
253 | |
254 | if (cpu >= nr_cpu_ids) |
255 | return NULL; |
256 | |
257 | if (unlikely(index >= array->map.max_entries)) |
258 | return NULL; |
259 | |
260 | return per_cpu_ptr(array->pptrs[index & array->index_mask], cpu); |
261 | } |
262 | |
263 | int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value) |
264 | { |
265 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
266 | u32 index = *(u32 *)key; |
267 | void __percpu *pptr; |
268 | int cpu, off = 0; |
269 | u32 size; |
270 | |
271 | if (unlikely(index >= array->map.max_entries)) |
272 | return -ENOENT; |
273 | |
274 | /* per_cpu areas are zero-filled and bpf programs can only |
275 | * access 'value_size' of them, so copying rounded areas |
276 | * will not leak any kernel data |
277 | */ |
278 | size = array->elem_size; |
279 | rcu_read_lock(); |
280 | pptr = array->pptrs[index & array->index_mask]; |
281 | for_each_possible_cpu(cpu) { |
282 | copy_map_value_long(map, dst: value + off, per_cpu_ptr(pptr, cpu)); |
283 | check_and_init_map_value(map, dst: value + off); |
284 | off += size; |
285 | } |
286 | rcu_read_unlock(); |
287 | return 0; |
288 | } |
289 | |
290 | /* Called from syscall */ |
291 | static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key) |
292 | { |
293 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
294 | u32 index = key ? *(u32 *)key : U32_MAX; |
295 | u32 *next = (u32 *)next_key; |
296 | |
297 | if (index >= array->map.max_entries) { |
298 | *next = 0; |
299 | return 0; |
300 | } |
301 | |
302 | if (index == array->map.max_entries - 1) |
303 | return -ENOENT; |
304 | |
305 | *next = index + 1; |
306 | return 0; |
307 | } |
308 | |
309 | /* Called from syscall or from eBPF program */ |
310 | static long array_map_update_elem(struct bpf_map *map, void *key, void *value, |
311 | u64 map_flags) |
312 | { |
313 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
314 | u32 index = *(u32 *)key; |
315 | char *val; |
316 | |
317 | if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST)) |
318 | /* unknown flags */ |
319 | return -EINVAL; |
320 | |
321 | if (unlikely(index >= array->map.max_entries)) |
322 | /* all elements were pre-allocated, cannot insert a new one */ |
323 | return -E2BIG; |
324 | |
325 | if (unlikely(map_flags & BPF_NOEXIST)) |
326 | /* all elements already exist */ |
327 | return -EEXIST; |
328 | |
329 | if (unlikely((map_flags & BPF_F_LOCK) && |
330 | !btf_record_has_field(map->record, BPF_SPIN_LOCK))) |
331 | return -EINVAL; |
332 | |
333 | if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { |
334 | val = this_cpu_ptr(array->pptrs[index & array->index_mask]); |
335 | copy_map_value(map, dst: val, src: value); |
336 | bpf_obj_free_fields(rec: array->map.record, obj: val); |
337 | } else { |
338 | val = array->value + |
339 | (u64)array->elem_size * (index & array->index_mask); |
340 | if (map_flags & BPF_F_LOCK) |
341 | copy_map_value_locked(map, dst: val, src: value, lock_src: false); |
342 | else |
343 | copy_map_value(map, dst: val, src: value); |
344 | bpf_obj_free_fields(rec: array->map.record, obj: val); |
345 | } |
346 | return 0; |
347 | } |
348 | |
349 | int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value, |
350 | u64 map_flags) |
351 | { |
352 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
353 | u32 index = *(u32 *)key; |
354 | void __percpu *pptr; |
355 | int cpu, off = 0; |
356 | u32 size; |
357 | |
358 | if (unlikely(map_flags > BPF_EXIST)) |
359 | /* unknown flags */ |
360 | return -EINVAL; |
361 | |
362 | if (unlikely(index >= array->map.max_entries)) |
363 | /* all elements were pre-allocated, cannot insert a new one */ |
364 | return -E2BIG; |
365 | |
366 | if (unlikely(map_flags == BPF_NOEXIST)) |
367 | /* all elements already exist */ |
368 | return -EEXIST; |
369 | |
370 | /* the user space will provide round_up(value_size, 8) bytes that |
371 | * will be copied into per-cpu area. bpf programs can only access |
372 | * value_size of it. During lookup the same extra bytes will be |
373 | * returned or zeros which were zero-filled by percpu_alloc, |
374 | * so no kernel data leaks possible |
375 | */ |
376 | size = array->elem_size; |
377 | rcu_read_lock(); |
378 | pptr = array->pptrs[index & array->index_mask]; |
379 | for_each_possible_cpu(cpu) { |
380 | copy_map_value_long(map, per_cpu_ptr(pptr, cpu), src: value + off); |
381 | bpf_obj_free_fields(rec: array->map.record, per_cpu_ptr(pptr, cpu)); |
382 | off += size; |
383 | } |
384 | rcu_read_unlock(); |
385 | return 0; |
386 | } |
387 | |
388 | /* Called from syscall or from eBPF program */ |
389 | static long array_map_delete_elem(struct bpf_map *map, void *key) |
390 | { |
391 | return -EINVAL; |
392 | } |
393 | |
394 | static void *array_map_vmalloc_addr(struct bpf_array *array) |
395 | { |
396 | return (void *)round_down((unsigned long)array, PAGE_SIZE); |
397 | } |
398 | |
399 | static void array_map_free_timers(struct bpf_map *map) |
400 | { |
401 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
402 | int i; |
403 | |
404 | /* We don't reset or free fields other than timer on uref dropping to zero. */ |
405 | if (!btf_record_has_field(rec: map->record, type: BPF_TIMER)) |
406 | return; |
407 | |
408 | for (i = 0; i < array->map.max_entries; i++) |
409 | bpf_obj_free_timer(rec: map->record, obj: array_map_elem_ptr(array, index: i)); |
410 | } |
411 | |
412 | /* Called when map->refcnt goes to zero, either from workqueue or from syscall */ |
413 | static void array_map_free(struct bpf_map *map) |
414 | { |
415 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
416 | int i; |
417 | |
418 | if (!IS_ERR_OR_NULL(ptr: map->record)) { |
419 | if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { |
420 | for (i = 0; i < array->map.max_entries; i++) { |
421 | void __percpu *pptr = array->pptrs[i & array->index_mask]; |
422 | int cpu; |
423 | |
424 | for_each_possible_cpu(cpu) { |
425 | bpf_obj_free_fields(rec: map->record, per_cpu_ptr(pptr, cpu)); |
426 | cond_resched(); |
427 | } |
428 | } |
429 | } else { |
430 | for (i = 0; i < array->map.max_entries; i++) |
431 | bpf_obj_free_fields(rec: map->record, obj: array_map_elem_ptr(array, index: i)); |
432 | } |
433 | } |
434 | |
435 | if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) |
436 | bpf_array_free_percpu(array); |
437 | |
438 | if (array->map.map_flags & BPF_F_MMAPABLE) |
439 | bpf_map_area_free(base: array_map_vmalloc_addr(array)); |
440 | else |
441 | bpf_map_area_free(base: array); |
442 | } |
443 | |
444 | static void array_map_seq_show_elem(struct bpf_map *map, void *key, |
445 | struct seq_file *m) |
446 | { |
447 | void *value; |
448 | |
449 | rcu_read_lock(); |
450 | |
451 | value = array_map_lookup_elem(map, key); |
452 | if (!value) { |
453 | rcu_read_unlock(); |
454 | return; |
455 | } |
456 | |
457 | if (map->btf_key_type_id) |
458 | seq_printf(m, fmt: "%u: " , *(u32 *)key); |
459 | btf_type_seq_show(btf: map->btf, type_id: map->btf_value_type_id, obj: value, m); |
460 | seq_puts(m, s: "\n" ); |
461 | |
462 | rcu_read_unlock(); |
463 | } |
464 | |
465 | static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key, |
466 | struct seq_file *m) |
467 | { |
468 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
469 | u32 index = *(u32 *)key; |
470 | void __percpu *pptr; |
471 | int cpu; |
472 | |
473 | rcu_read_lock(); |
474 | |
475 | seq_printf(m, fmt: "%u: {\n" , *(u32 *)key); |
476 | pptr = array->pptrs[index & array->index_mask]; |
477 | for_each_possible_cpu(cpu) { |
478 | seq_printf(m, fmt: "\tcpu%d: " , cpu); |
479 | btf_type_seq_show(btf: map->btf, type_id: map->btf_value_type_id, |
480 | per_cpu_ptr(pptr, cpu), m); |
481 | seq_puts(m, s: "\n" ); |
482 | } |
483 | seq_puts(m, s: "}\n" ); |
484 | |
485 | rcu_read_unlock(); |
486 | } |
487 | |
488 | static int array_map_check_btf(const struct bpf_map *map, |
489 | const struct btf *btf, |
490 | const struct btf_type *key_type, |
491 | const struct btf_type *value_type) |
492 | { |
493 | u32 int_data; |
494 | |
495 | /* One exception for keyless BTF: .bss/.data/.rodata map */ |
496 | if (btf_type_is_void(t: key_type)) { |
497 | if (map->map_type != BPF_MAP_TYPE_ARRAY || |
498 | map->max_entries != 1) |
499 | return -EINVAL; |
500 | |
501 | if (BTF_INFO_KIND(value_type->info) != BTF_KIND_DATASEC) |
502 | return -EINVAL; |
503 | |
504 | return 0; |
505 | } |
506 | |
507 | if (BTF_INFO_KIND(key_type->info) != BTF_KIND_INT) |
508 | return -EINVAL; |
509 | |
510 | int_data = *(u32 *)(key_type + 1); |
511 | /* bpf array can only take a u32 key. This check makes sure |
512 | * that the btf matches the attr used during map_create. |
513 | */ |
514 | if (BTF_INT_BITS(int_data) != 32 || BTF_INT_OFFSET(int_data)) |
515 | return -EINVAL; |
516 | |
517 | return 0; |
518 | } |
519 | |
520 | static int array_map_mmap(struct bpf_map *map, struct vm_area_struct *vma) |
521 | { |
522 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
523 | pgoff_t pgoff = PAGE_ALIGN(sizeof(*array)) >> PAGE_SHIFT; |
524 | |
525 | if (!(map->map_flags & BPF_F_MMAPABLE)) |
526 | return -EINVAL; |
527 | |
528 | if (vma->vm_pgoff * PAGE_SIZE + (vma->vm_end - vma->vm_start) > |
529 | PAGE_ALIGN((u64)array->map.max_entries * array->elem_size)) |
530 | return -EINVAL; |
531 | |
532 | return remap_vmalloc_range(vma, addr: array_map_vmalloc_addr(array), |
533 | pgoff: vma->vm_pgoff + pgoff); |
534 | } |
535 | |
536 | static bool array_map_meta_equal(const struct bpf_map *meta0, |
537 | const struct bpf_map *meta1) |
538 | { |
539 | if (!bpf_map_meta_equal(meta0, meta1)) |
540 | return false; |
541 | return meta0->map_flags & BPF_F_INNER_MAP ? true : |
542 | meta0->max_entries == meta1->max_entries; |
543 | } |
544 | |
545 | struct bpf_iter_seq_array_map_info { |
546 | struct bpf_map *map; |
547 | void *percpu_value_buf; |
548 | u32 index; |
549 | }; |
550 | |
551 | static void *bpf_array_map_seq_start(struct seq_file *seq, loff_t *pos) |
552 | { |
553 | struct bpf_iter_seq_array_map_info *info = seq->private; |
554 | struct bpf_map *map = info->map; |
555 | struct bpf_array *array; |
556 | u32 index; |
557 | |
558 | if (info->index >= map->max_entries) |
559 | return NULL; |
560 | |
561 | if (*pos == 0) |
562 | ++*pos; |
563 | array = container_of(map, struct bpf_array, map); |
564 | index = info->index & array->index_mask; |
565 | if (info->percpu_value_buf) |
566 | return array->pptrs[index]; |
567 | return array_map_elem_ptr(array, index); |
568 | } |
569 | |
570 | static void *bpf_array_map_seq_next(struct seq_file *seq, void *v, loff_t *pos) |
571 | { |
572 | struct bpf_iter_seq_array_map_info *info = seq->private; |
573 | struct bpf_map *map = info->map; |
574 | struct bpf_array *array; |
575 | u32 index; |
576 | |
577 | ++*pos; |
578 | ++info->index; |
579 | if (info->index >= map->max_entries) |
580 | return NULL; |
581 | |
582 | array = container_of(map, struct bpf_array, map); |
583 | index = info->index & array->index_mask; |
584 | if (info->percpu_value_buf) |
585 | return array->pptrs[index]; |
586 | return array_map_elem_ptr(array, index); |
587 | } |
588 | |
589 | static int __bpf_array_map_seq_show(struct seq_file *seq, void *v) |
590 | { |
591 | struct bpf_iter_seq_array_map_info *info = seq->private; |
592 | struct bpf_iter__bpf_map_elem ctx = {}; |
593 | struct bpf_map *map = info->map; |
594 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
595 | struct bpf_iter_meta meta; |
596 | struct bpf_prog *prog; |
597 | int off = 0, cpu = 0; |
598 | void __percpu **pptr; |
599 | u32 size; |
600 | |
601 | meta.seq = seq; |
602 | prog = bpf_iter_get_info(meta: &meta, in_stop: v == NULL); |
603 | if (!prog) |
604 | return 0; |
605 | |
606 | ctx.meta = &meta; |
607 | ctx.map = info->map; |
608 | if (v) { |
609 | ctx.key = &info->index; |
610 | |
611 | if (!info->percpu_value_buf) { |
612 | ctx.value = v; |
613 | } else { |
614 | pptr = v; |
615 | size = array->elem_size; |
616 | for_each_possible_cpu(cpu) { |
617 | copy_map_value_long(map, dst: info->percpu_value_buf + off, |
618 | per_cpu_ptr(pptr, cpu)); |
619 | check_and_init_map_value(map, dst: info->percpu_value_buf + off); |
620 | off += size; |
621 | } |
622 | ctx.value = info->percpu_value_buf; |
623 | } |
624 | } |
625 | |
626 | return bpf_iter_run_prog(prog, ctx: &ctx); |
627 | } |
628 | |
629 | static int bpf_array_map_seq_show(struct seq_file *seq, void *v) |
630 | { |
631 | return __bpf_array_map_seq_show(seq, v); |
632 | } |
633 | |
634 | static void bpf_array_map_seq_stop(struct seq_file *seq, void *v) |
635 | { |
636 | if (!v) |
637 | (void)__bpf_array_map_seq_show(seq, NULL); |
638 | } |
639 | |
640 | static int bpf_iter_init_array_map(void *priv_data, |
641 | struct bpf_iter_aux_info *aux) |
642 | { |
643 | struct bpf_iter_seq_array_map_info *seq_info = priv_data; |
644 | struct bpf_map *map = aux->map; |
645 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
646 | void *value_buf; |
647 | u32 buf_size; |
648 | |
649 | if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { |
650 | buf_size = array->elem_size * num_possible_cpus(); |
651 | value_buf = kmalloc(size: buf_size, GFP_USER | __GFP_NOWARN); |
652 | if (!value_buf) |
653 | return -ENOMEM; |
654 | |
655 | seq_info->percpu_value_buf = value_buf; |
656 | } |
657 | |
658 | /* bpf_iter_attach_map() acquires a map uref, and the uref may be |
659 | * released before or in the middle of iterating map elements, so |
660 | * acquire an extra map uref for iterator. |
661 | */ |
662 | bpf_map_inc_with_uref(map); |
663 | seq_info->map = map; |
664 | return 0; |
665 | } |
666 | |
667 | static void bpf_iter_fini_array_map(void *priv_data) |
668 | { |
669 | struct bpf_iter_seq_array_map_info *seq_info = priv_data; |
670 | |
671 | bpf_map_put_with_uref(map: seq_info->map); |
672 | kfree(objp: seq_info->percpu_value_buf); |
673 | } |
674 | |
675 | static const struct seq_operations bpf_array_map_seq_ops = { |
676 | .start = bpf_array_map_seq_start, |
677 | .next = bpf_array_map_seq_next, |
678 | .stop = bpf_array_map_seq_stop, |
679 | .show = bpf_array_map_seq_show, |
680 | }; |
681 | |
682 | static const struct bpf_iter_seq_info iter_seq_info = { |
683 | .seq_ops = &bpf_array_map_seq_ops, |
684 | .init_seq_private = bpf_iter_init_array_map, |
685 | .fini_seq_private = bpf_iter_fini_array_map, |
686 | .seq_priv_size = sizeof(struct bpf_iter_seq_array_map_info), |
687 | }; |
688 | |
689 | static long bpf_for_each_array_elem(struct bpf_map *map, bpf_callback_t callback_fn, |
690 | void *callback_ctx, u64 flags) |
691 | { |
692 | u32 i, key, num_elems = 0; |
693 | struct bpf_array *array; |
694 | bool is_percpu; |
695 | u64 ret = 0; |
696 | void *val; |
697 | |
698 | if (flags != 0) |
699 | return -EINVAL; |
700 | |
701 | is_percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY; |
702 | array = container_of(map, struct bpf_array, map); |
703 | if (is_percpu) |
704 | migrate_disable(); |
705 | for (i = 0; i < map->max_entries; i++) { |
706 | if (is_percpu) |
707 | val = this_cpu_ptr(array->pptrs[i]); |
708 | else |
709 | val = array_map_elem_ptr(array, index: i); |
710 | num_elems++; |
711 | key = i; |
712 | ret = callback_fn((u64)(long)map, (u64)(long)&key, |
713 | (u64)(long)val, (u64)(long)callback_ctx, 0); |
714 | /* return value: 0 - continue, 1 - stop and return */ |
715 | if (ret) |
716 | break; |
717 | } |
718 | |
719 | if (is_percpu) |
720 | migrate_enable(); |
721 | return num_elems; |
722 | } |
723 | |
724 | static u64 array_map_mem_usage(const struct bpf_map *map) |
725 | { |
726 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
727 | bool percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY; |
728 | u32 elem_size = array->elem_size; |
729 | u64 entries = map->max_entries; |
730 | u64 usage = sizeof(*array); |
731 | |
732 | if (percpu) { |
733 | usage += entries * sizeof(void *); |
734 | usage += entries * elem_size * num_possible_cpus(); |
735 | } else { |
736 | if (map->map_flags & BPF_F_MMAPABLE) { |
737 | usage = PAGE_ALIGN(usage); |
738 | usage += PAGE_ALIGN(entries * elem_size); |
739 | } else { |
740 | usage += entries * elem_size; |
741 | } |
742 | } |
743 | return usage; |
744 | } |
745 | |
746 | BTF_ID_LIST_SINGLE(array_map_btf_ids, struct, bpf_array) |
747 | const struct bpf_map_ops array_map_ops = { |
748 | .map_meta_equal = array_map_meta_equal, |
749 | .map_alloc_check = array_map_alloc_check, |
750 | .map_alloc = array_map_alloc, |
751 | .map_free = array_map_free, |
752 | .map_get_next_key = array_map_get_next_key, |
753 | .map_release_uref = array_map_free_timers, |
754 | .map_lookup_elem = array_map_lookup_elem, |
755 | .map_update_elem = array_map_update_elem, |
756 | .map_delete_elem = array_map_delete_elem, |
757 | .map_gen_lookup = array_map_gen_lookup, |
758 | .map_direct_value_addr = array_map_direct_value_addr, |
759 | .map_direct_value_meta = array_map_direct_value_meta, |
760 | .map_mmap = array_map_mmap, |
761 | .map_seq_show_elem = array_map_seq_show_elem, |
762 | .map_check_btf = array_map_check_btf, |
763 | .map_lookup_batch = generic_map_lookup_batch, |
764 | .map_update_batch = generic_map_update_batch, |
765 | .map_set_for_each_callback_args = map_set_for_each_callback_args, |
766 | .map_for_each_callback = bpf_for_each_array_elem, |
767 | .map_mem_usage = array_map_mem_usage, |
768 | .map_btf_id = &array_map_btf_ids[0], |
769 | .iter_seq_info = &iter_seq_info, |
770 | }; |
771 | |
772 | const struct bpf_map_ops percpu_array_map_ops = { |
773 | .map_meta_equal = bpf_map_meta_equal, |
774 | .map_alloc_check = array_map_alloc_check, |
775 | .map_alloc = array_map_alloc, |
776 | .map_free = array_map_free, |
777 | .map_get_next_key = array_map_get_next_key, |
778 | .map_lookup_elem = percpu_array_map_lookup_elem, |
779 | .map_update_elem = array_map_update_elem, |
780 | .map_delete_elem = array_map_delete_elem, |
781 | .map_lookup_percpu_elem = percpu_array_map_lookup_percpu_elem, |
782 | .map_seq_show_elem = percpu_array_map_seq_show_elem, |
783 | .map_check_btf = array_map_check_btf, |
784 | .map_lookup_batch = generic_map_lookup_batch, |
785 | .map_update_batch = generic_map_update_batch, |
786 | .map_set_for_each_callback_args = map_set_for_each_callback_args, |
787 | .map_for_each_callback = bpf_for_each_array_elem, |
788 | .map_mem_usage = array_map_mem_usage, |
789 | .map_btf_id = &array_map_btf_ids[0], |
790 | .iter_seq_info = &iter_seq_info, |
791 | }; |
792 | |
793 | static int fd_array_map_alloc_check(union bpf_attr *attr) |
794 | { |
795 | /* only file descriptors can be stored in this type of map */ |
796 | if (attr->value_size != sizeof(u32)) |
797 | return -EINVAL; |
798 | /* Program read-only/write-only not supported for special maps yet. */ |
799 | if (attr->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) |
800 | return -EINVAL; |
801 | return array_map_alloc_check(attr); |
802 | } |
803 | |
804 | static void fd_array_map_free(struct bpf_map *map) |
805 | { |
806 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
807 | int i; |
808 | |
809 | /* make sure it's empty */ |
810 | for (i = 0; i < array->map.max_entries; i++) |
811 | BUG_ON(array->ptrs[i] != NULL); |
812 | |
813 | bpf_map_area_free(base: array); |
814 | } |
815 | |
816 | static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key) |
817 | { |
818 | return ERR_PTR(error: -EOPNOTSUPP); |
819 | } |
820 | |
821 | /* only called from syscall */ |
822 | int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value) |
823 | { |
824 | void **elem, *ptr; |
825 | int ret = 0; |
826 | |
827 | if (!map->ops->map_fd_sys_lookup_elem) |
828 | return -ENOTSUPP; |
829 | |
830 | rcu_read_lock(); |
831 | elem = array_map_lookup_elem(map, key); |
832 | if (elem && (ptr = READ_ONCE(*elem))) |
833 | *value = map->ops->map_fd_sys_lookup_elem(ptr); |
834 | else |
835 | ret = -ENOENT; |
836 | rcu_read_unlock(); |
837 | |
838 | return ret; |
839 | } |
840 | |
841 | /* only called from syscall */ |
842 | int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file, |
843 | void *key, void *value, u64 map_flags) |
844 | { |
845 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
846 | void *new_ptr, *old_ptr; |
847 | u32 index = *(u32 *)key, ufd; |
848 | |
849 | if (map_flags != BPF_ANY) |
850 | return -EINVAL; |
851 | |
852 | if (index >= array->map.max_entries) |
853 | return -E2BIG; |
854 | |
855 | ufd = *(u32 *)value; |
856 | new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd); |
857 | if (IS_ERR(ptr: new_ptr)) |
858 | return PTR_ERR(ptr: new_ptr); |
859 | |
860 | if (map->ops->map_poke_run) { |
861 | mutex_lock(&array->aux->poke_mutex); |
862 | old_ptr = xchg(array->ptrs + index, new_ptr); |
863 | map->ops->map_poke_run(map, index, old_ptr, new_ptr); |
864 | mutex_unlock(lock: &array->aux->poke_mutex); |
865 | } else { |
866 | old_ptr = xchg(array->ptrs + index, new_ptr); |
867 | } |
868 | |
869 | if (old_ptr) |
870 | map->ops->map_fd_put_ptr(old_ptr); |
871 | return 0; |
872 | } |
873 | |
874 | static long fd_array_map_delete_elem(struct bpf_map *map, void *key) |
875 | { |
876 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
877 | void *old_ptr; |
878 | u32 index = *(u32 *)key; |
879 | |
880 | if (index >= array->map.max_entries) |
881 | return -E2BIG; |
882 | |
883 | if (map->ops->map_poke_run) { |
884 | mutex_lock(&array->aux->poke_mutex); |
885 | old_ptr = xchg(array->ptrs + index, NULL); |
886 | map->ops->map_poke_run(map, index, old_ptr, NULL); |
887 | mutex_unlock(lock: &array->aux->poke_mutex); |
888 | } else { |
889 | old_ptr = xchg(array->ptrs + index, NULL); |
890 | } |
891 | |
892 | if (old_ptr) { |
893 | map->ops->map_fd_put_ptr(old_ptr); |
894 | return 0; |
895 | } else { |
896 | return -ENOENT; |
897 | } |
898 | } |
899 | |
900 | static void *prog_fd_array_get_ptr(struct bpf_map *map, |
901 | struct file *map_file, int fd) |
902 | { |
903 | struct bpf_prog *prog = bpf_prog_get(ufd: fd); |
904 | |
905 | if (IS_ERR(ptr: prog)) |
906 | return prog; |
907 | |
908 | if (!bpf_prog_map_compatible(map, fp: prog)) { |
909 | bpf_prog_put(prog); |
910 | return ERR_PTR(error: -EINVAL); |
911 | } |
912 | |
913 | return prog; |
914 | } |
915 | |
916 | static void prog_fd_array_put_ptr(void *ptr) |
917 | { |
918 | bpf_prog_put(prog: ptr); |
919 | } |
920 | |
921 | static u32 prog_fd_array_sys_lookup_elem(void *ptr) |
922 | { |
923 | return ((struct bpf_prog *)ptr)->aux->id; |
924 | } |
925 | |
926 | /* decrement refcnt of all bpf_progs that are stored in this map */ |
927 | static void bpf_fd_array_map_clear(struct bpf_map *map) |
928 | { |
929 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
930 | int i; |
931 | |
932 | for (i = 0; i < array->map.max_entries; i++) |
933 | fd_array_map_delete_elem(map, key: &i); |
934 | } |
935 | |
936 | static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key, |
937 | struct seq_file *m) |
938 | { |
939 | void **elem, *ptr; |
940 | u32 prog_id; |
941 | |
942 | rcu_read_lock(); |
943 | |
944 | elem = array_map_lookup_elem(map, key); |
945 | if (elem) { |
946 | ptr = READ_ONCE(*elem); |
947 | if (ptr) { |
948 | seq_printf(m, fmt: "%u: " , *(u32 *)key); |
949 | prog_id = prog_fd_array_sys_lookup_elem(ptr); |
950 | btf_type_seq_show(btf: map->btf, type_id: map->btf_value_type_id, |
951 | obj: &prog_id, m); |
952 | seq_puts(m, s: "\n" ); |
953 | } |
954 | } |
955 | |
956 | rcu_read_unlock(); |
957 | } |
958 | |
959 | struct prog_poke_elem { |
960 | struct list_head list; |
961 | struct bpf_prog_aux *aux; |
962 | }; |
963 | |
964 | static int prog_array_map_poke_track(struct bpf_map *map, |
965 | struct bpf_prog_aux *prog_aux) |
966 | { |
967 | struct prog_poke_elem *elem; |
968 | struct bpf_array_aux *aux; |
969 | int ret = 0; |
970 | |
971 | aux = container_of(map, struct bpf_array, map)->aux; |
972 | mutex_lock(&aux->poke_mutex); |
973 | list_for_each_entry(elem, &aux->poke_progs, list) { |
974 | if (elem->aux == prog_aux) |
975 | goto out; |
976 | } |
977 | |
978 | elem = kmalloc(size: sizeof(*elem), GFP_KERNEL); |
979 | if (!elem) { |
980 | ret = -ENOMEM; |
981 | goto out; |
982 | } |
983 | |
984 | INIT_LIST_HEAD(list: &elem->list); |
985 | /* We must track the program's aux info at this point in time |
986 | * since the program pointer itself may not be stable yet, see |
987 | * also comment in prog_array_map_poke_run(). |
988 | */ |
989 | elem->aux = prog_aux; |
990 | |
991 | list_add_tail(new: &elem->list, head: &aux->poke_progs); |
992 | out: |
993 | mutex_unlock(lock: &aux->poke_mutex); |
994 | return ret; |
995 | } |
996 | |
997 | static void prog_array_map_poke_untrack(struct bpf_map *map, |
998 | struct bpf_prog_aux *prog_aux) |
999 | { |
1000 | struct prog_poke_elem *elem, *tmp; |
1001 | struct bpf_array_aux *aux; |
1002 | |
1003 | aux = container_of(map, struct bpf_array, map)->aux; |
1004 | mutex_lock(&aux->poke_mutex); |
1005 | list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) { |
1006 | if (elem->aux == prog_aux) { |
1007 | list_del_init(entry: &elem->list); |
1008 | kfree(objp: elem); |
1009 | break; |
1010 | } |
1011 | } |
1012 | mutex_unlock(lock: &aux->poke_mutex); |
1013 | } |
1014 | |
1015 | static void prog_array_map_poke_run(struct bpf_map *map, u32 key, |
1016 | struct bpf_prog *old, |
1017 | struct bpf_prog *new) |
1018 | { |
1019 | u8 *old_addr, *new_addr, *old_bypass_addr; |
1020 | struct prog_poke_elem *elem; |
1021 | struct bpf_array_aux *aux; |
1022 | |
1023 | aux = container_of(map, struct bpf_array, map)->aux; |
1024 | WARN_ON_ONCE(!mutex_is_locked(&aux->poke_mutex)); |
1025 | |
1026 | list_for_each_entry(elem, &aux->poke_progs, list) { |
1027 | struct bpf_jit_poke_descriptor *poke; |
1028 | int i, ret; |
1029 | |
1030 | for (i = 0; i < elem->aux->size_poke_tab; i++) { |
1031 | poke = &elem->aux->poke_tab[i]; |
1032 | |
1033 | /* Few things to be aware of: |
1034 | * |
1035 | * 1) We can only ever access aux in this context, but |
1036 | * not aux->prog since it might not be stable yet and |
1037 | * there could be danger of use after free otherwise. |
1038 | * 2) Initially when we start tracking aux, the program |
1039 | * is not JITed yet and also does not have a kallsyms |
1040 | * entry. We skip these as poke->tailcall_target_stable |
1041 | * is not active yet. The JIT will do the final fixup |
1042 | * before setting it stable. The various |
1043 | * poke->tailcall_target_stable are successively |
1044 | * activated, so tail call updates can arrive from here |
1045 | * while JIT is still finishing its final fixup for |
1046 | * non-activated poke entries. |
1047 | * 3) On program teardown, the program's kallsym entry gets |
1048 | * removed out of RCU callback, but we can only untrack |
1049 | * from sleepable context, therefore bpf_arch_text_poke() |
1050 | * might not see that this is in BPF text section and |
1051 | * bails out with -EINVAL. As these are unreachable since |
1052 | * RCU grace period already passed, we simply skip them. |
1053 | * 4) Also programs reaching refcount of zero while patching |
1054 | * is in progress is okay since we're protected under |
1055 | * poke_mutex and untrack the programs before the JIT |
1056 | * buffer is freed. When we're still in the middle of |
1057 | * patching and suddenly kallsyms entry of the program |
1058 | * gets evicted, we just skip the rest which is fine due |
1059 | * to point 3). |
1060 | * 5) Any other error happening below from bpf_arch_text_poke() |
1061 | * is a unexpected bug. |
1062 | */ |
1063 | if (!READ_ONCE(poke->tailcall_target_stable)) |
1064 | continue; |
1065 | if (poke->reason != BPF_POKE_REASON_TAIL_CALL) |
1066 | continue; |
1067 | if (poke->tail_call.map != map || |
1068 | poke->tail_call.key != key) |
1069 | continue; |
1070 | |
1071 | old_bypass_addr = old ? NULL : poke->bypass_addr; |
1072 | old_addr = old ? (u8 *)old->bpf_func + poke->adj_off : NULL; |
1073 | new_addr = new ? (u8 *)new->bpf_func + poke->adj_off : NULL; |
1074 | |
1075 | if (new) { |
1076 | ret = bpf_arch_text_poke(ip: poke->tailcall_target, |
1077 | t: BPF_MOD_JUMP, |
1078 | addr1: old_addr, addr2: new_addr); |
1079 | BUG_ON(ret < 0 && ret != -EINVAL); |
1080 | if (!old) { |
1081 | ret = bpf_arch_text_poke(ip: poke->tailcall_bypass, |
1082 | t: BPF_MOD_JUMP, |
1083 | addr1: poke->bypass_addr, |
1084 | NULL); |
1085 | BUG_ON(ret < 0 && ret != -EINVAL); |
1086 | } |
1087 | } else { |
1088 | ret = bpf_arch_text_poke(ip: poke->tailcall_bypass, |
1089 | t: BPF_MOD_JUMP, |
1090 | addr1: old_bypass_addr, |
1091 | addr2: poke->bypass_addr); |
1092 | BUG_ON(ret < 0 && ret != -EINVAL); |
1093 | /* let other CPUs finish the execution of program |
1094 | * so that it will not possible to expose them |
1095 | * to invalid nop, stack unwind, nop state |
1096 | */ |
1097 | if (!ret) |
1098 | synchronize_rcu(); |
1099 | ret = bpf_arch_text_poke(ip: poke->tailcall_target, |
1100 | t: BPF_MOD_JUMP, |
1101 | addr1: old_addr, NULL); |
1102 | BUG_ON(ret < 0 && ret != -EINVAL); |
1103 | } |
1104 | } |
1105 | } |
1106 | } |
1107 | |
1108 | static void prog_array_map_clear_deferred(struct work_struct *work) |
1109 | { |
1110 | struct bpf_map *map = container_of(work, struct bpf_array_aux, |
1111 | work)->map; |
1112 | bpf_fd_array_map_clear(map); |
1113 | bpf_map_put(map); |
1114 | } |
1115 | |
1116 | static void prog_array_map_clear(struct bpf_map *map) |
1117 | { |
1118 | struct bpf_array_aux *aux = container_of(map, struct bpf_array, |
1119 | map)->aux; |
1120 | bpf_map_inc(map); |
1121 | schedule_work(work: &aux->work); |
1122 | } |
1123 | |
1124 | static struct bpf_map *prog_array_map_alloc(union bpf_attr *attr) |
1125 | { |
1126 | struct bpf_array_aux *aux; |
1127 | struct bpf_map *map; |
1128 | |
1129 | aux = kzalloc(size: sizeof(*aux), GFP_KERNEL_ACCOUNT); |
1130 | if (!aux) |
1131 | return ERR_PTR(error: -ENOMEM); |
1132 | |
1133 | INIT_WORK(&aux->work, prog_array_map_clear_deferred); |
1134 | INIT_LIST_HEAD(list: &aux->poke_progs); |
1135 | mutex_init(&aux->poke_mutex); |
1136 | |
1137 | map = array_map_alloc(attr); |
1138 | if (IS_ERR(ptr: map)) { |
1139 | kfree(objp: aux); |
1140 | return map; |
1141 | } |
1142 | |
1143 | container_of(map, struct bpf_array, map)->aux = aux; |
1144 | aux->map = map; |
1145 | |
1146 | return map; |
1147 | } |
1148 | |
1149 | static void prog_array_map_free(struct bpf_map *map) |
1150 | { |
1151 | struct prog_poke_elem *elem, *tmp; |
1152 | struct bpf_array_aux *aux; |
1153 | |
1154 | aux = container_of(map, struct bpf_array, map)->aux; |
1155 | list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) { |
1156 | list_del_init(entry: &elem->list); |
1157 | kfree(objp: elem); |
1158 | } |
1159 | kfree(objp: aux); |
1160 | fd_array_map_free(map); |
1161 | } |
1162 | |
1163 | /* prog_array->aux->{type,jited} is a runtime binding. |
1164 | * Doing static check alone in the verifier is not enough. |
1165 | * Thus, prog_array_map cannot be used as an inner_map |
1166 | * and map_meta_equal is not implemented. |
1167 | */ |
1168 | const struct bpf_map_ops prog_array_map_ops = { |
1169 | .map_alloc_check = fd_array_map_alloc_check, |
1170 | .map_alloc = prog_array_map_alloc, |
1171 | .map_free = prog_array_map_free, |
1172 | .map_poke_track = prog_array_map_poke_track, |
1173 | .map_poke_untrack = prog_array_map_poke_untrack, |
1174 | .map_poke_run = prog_array_map_poke_run, |
1175 | .map_get_next_key = array_map_get_next_key, |
1176 | .map_lookup_elem = fd_array_map_lookup_elem, |
1177 | .map_delete_elem = fd_array_map_delete_elem, |
1178 | .map_fd_get_ptr = prog_fd_array_get_ptr, |
1179 | .map_fd_put_ptr = prog_fd_array_put_ptr, |
1180 | .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem, |
1181 | .map_release_uref = prog_array_map_clear, |
1182 | .map_seq_show_elem = prog_array_map_seq_show_elem, |
1183 | .map_mem_usage = array_map_mem_usage, |
1184 | .map_btf_id = &array_map_btf_ids[0], |
1185 | }; |
1186 | |
1187 | static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file, |
1188 | struct file *map_file) |
1189 | { |
1190 | struct bpf_event_entry *ee; |
1191 | |
1192 | ee = kzalloc(size: sizeof(*ee), GFP_ATOMIC); |
1193 | if (ee) { |
1194 | ee->event = perf_file->private_data; |
1195 | ee->perf_file = perf_file; |
1196 | ee->map_file = map_file; |
1197 | } |
1198 | |
1199 | return ee; |
1200 | } |
1201 | |
1202 | static void __bpf_event_entry_free(struct rcu_head *rcu) |
1203 | { |
1204 | struct bpf_event_entry *ee; |
1205 | |
1206 | ee = container_of(rcu, struct bpf_event_entry, rcu); |
1207 | fput(ee->perf_file); |
1208 | kfree(objp: ee); |
1209 | } |
1210 | |
1211 | static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee) |
1212 | { |
1213 | call_rcu(head: &ee->rcu, func: __bpf_event_entry_free); |
1214 | } |
1215 | |
1216 | static void *perf_event_fd_array_get_ptr(struct bpf_map *map, |
1217 | struct file *map_file, int fd) |
1218 | { |
1219 | struct bpf_event_entry *ee; |
1220 | struct perf_event *event; |
1221 | struct file *perf_file; |
1222 | u64 value; |
1223 | |
1224 | perf_file = perf_event_get(fd); |
1225 | if (IS_ERR(ptr: perf_file)) |
1226 | return perf_file; |
1227 | |
1228 | ee = ERR_PTR(error: -EOPNOTSUPP); |
1229 | event = perf_file->private_data; |
1230 | if (perf_event_read_local(event, value: &value, NULL, NULL) == -EOPNOTSUPP) |
1231 | goto err_out; |
1232 | |
1233 | ee = bpf_event_entry_gen(perf_file, map_file); |
1234 | if (ee) |
1235 | return ee; |
1236 | ee = ERR_PTR(error: -ENOMEM); |
1237 | err_out: |
1238 | fput(perf_file); |
1239 | return ee; |
1240 | } |
1241 | |
1242 | static void perf_event_fd_array_put_ptr(void *ptr) |
1243 | { |
1244 | bpf_event_entry_free_rcu(ee: ptr); |
1245 | } |
1246 | |
1247 | static void perf_event_fd_array_release(struct bpf_map *map, |
1248 | struct file *map_file) |
1249 | { |
1250 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
1251 | struct bpf_event_entry *ee; |
1252 | int i; |
1253 | |
1254 | if (map->map_flags & BPF_F_PRESERVE_ELEMS) |
1255 | return; |
1256 | |
1257 | rcu_read_lock(); |
1258 | for (i = 0; i < array->map.max_entries; i++) { |
1259 | ee = READ_ONCE(array->ptrs[i]); |
1260 | if (ee && ee->map_file == map_file) |
1261 | fd_array_map_delete_elem(map, key: &i); |
1262 | } |
1263 | rcu_read_unlock(); |
1264 | } |
1265 | |
1266 | static void perf_event_fd_array_map_free(struct bpf_map *map) |
1267 | { |
1268 | if (map->map_flags & BPF_F_PRESERVE_ELEMS) |
1269 | bpf_fd_array_map_clear(map); |
1270 | fd_array_map_free(map); |
1271 | } |
1272 | |
1273 | const struct bpf_map_ops perf_event_array_map_ops = { |
1274 | .map_meta_equal = bpf_map_meta_equal, |
1275 | .map_alloc_check = fd_array_map_alloc_check, |
1276 | .map_alloc = array_map_alloc, |
1277 | .map_free = perf_event_fd_array_map_free, |
1278 | .map_get_next_key = array_map_get_next_key, |
1279 | .map_lookup_elem = fd_array_map_lookup_elem, |
1280 | .map_delete_elem = fd_array_map_delete_elem, |
1281 | .map_fd_get_ptr = perf_event_fd_array_get_ptr, |
1282 | .map_fd_put_ptr = perf_event_fd_array_put_ptr, |
1283 | .map_release = perf_event_fd_array_release, |
1284 | .map_check_btf = map_check_no_btf, |
1285 | .map_mem_usage = array_map_mem_usage, |
1286 | .map_btf_id = &array_map_btf_ids[0], |
1287 | }; |
1288 | |
1289 | #ifdef CONFIG_CGROUPS |
1290 | static void *cgroup_fd_array_get_ptr(struct bpf_map *map, |
1291 | struct file *map_file /* not used */, |
1292 | int fd) |
1293 | { |
1294 | return cgroup_get_from_fd(fd); |
1295 | } |
1296 | |
1297 | static void cgroup_fd_array_put_ptr(void *ptr) |
1298 | { |
1299 | /* cgroup_put free cgrp after a rcu grace period */ |
1300 | cgroup_put(cgrp: ptr); |
1301 | } |
1302 | |
1303 | static void cgroup_fd_array_free(struct bpf_map *map) |
1304 | { |
1305 | bpf_fd_array_map_clear(map); |
1306 | fd_array_map_free(map); |
1307 | } |
1308 | |
1309 | const struct bpf_map_ops cgroup_array_map_ops = { |
1310 | .map_meta_equal = bpf_map_meta_equal, |
1311 | .map_alloc_check = fd_array_map_alloc_check, |
1312 | .map_alloc = array_map_alloc, |
1313 | .map_free = cgroup_fd_array_free, |
1314 | .map_get_next_key = array_map_get_next_key, |
1315 | .map_lookup_elem = fd_array_map_lookup_elem, |
1316 | .map_delete_elem = fd_array_map_delete_elem, |
1317 | .map_fd_get_ptr = cgroup_fd_array_get_ptr, |
1318 | .map_fd_put_ptr = cgroup_fd_array_put_ptr, |
1319 | .map_check_btf = map_check_no_btf, |
1320 | .map_mem_usage = array_map_mem_usage, |
1321 | .map_btf_id = &array_map_btf_ids[0], |
1322 | }; |
1323 | #endif |
1324 | |
1325 | static struct bpf_map *array_of_map_alloc(union bpf_attr *attr) |
1326 | { |
1327 | struct bpf_map *map, *inner_map_meta; |
1328 | |
1329 | inner_map_meta = bpf_map_meta_alloc(inner_map_ufd: attr->inner_map_fd); |
1330 | if (IS_ERR(ptr: inner_map_meta)) |
1331 | return inner_map_meta; |
1332 | |
1333 | map = array_map_alloc(attr); |
1334 | if (IS_ERR(ptr: map)) { |
1335 | bpf_map_meta_free(map_meta: inner_map_meta); |
1336 | return map; |
1337 | } |
1338 | |
1339 | map->inner_map_meta = inner_map_meta; |
1340 | |
1341 | return map; |
1342 | } |
1343 | |
1344 | static void array_of_map_free(struct bpf_map *map) |
1345 | { |
1346 | /* map->inner_map_meta is only accessed by syscall which |
1347 | * is protected by fdget/fdput. |
1348 | */ |
1349 | bpf_map_meta_free(map_meta: map->inner_map_meta); |
1350 | bpf_fd_array_map_clear(map); |
1351 | fd_array_map_free(map); |
1352 | } |
1353 | |
1354 | static void *array_of_map_lookup_elem(struct bpf_map *map, void *key) |
1355 | { |
1356 | struct bpf_map **inner_map = array_map_lookup_elem(map, key); |
1357 | |
1358 | if (!inner_map) |
1359 | return NULL; |
1360 | |
1361 | return READ_ONCE(*inner_map); |
1362 | } |
1363 | |
1364 | static int array_of_map_gen_lookup(struct bpf_map *map, |
1365 | struct bpf_insn *insn_buf) |
1366 | { |
1367 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
1368 | u32 elem_size = array->elem_size; |
1369 | struct bpf_insn *insn = insn_buf; |
1370 | const int ret = BPF_REG_0; |
1371 | const int map_ptr = BPF_REG_1; |
1372 | const int index = BPF_REG_2; |
1373 | |
1374 | *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value)); |
1375 | *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0); |
1376 | if (!map->bypass_spec_v1) { |
1377 | *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6); |
1378 | *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask); |
1379 | } else { |
1380 | *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5); |
1381 | } |
1382 | if (is_power_of_2(n: elem_size)) |
1383 | *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size)); |
1384 | else |
1385 | *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size); |
1386 | *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr); |
1387 | *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0); |
1388 | *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1); |
1389 | *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1); |
1390 | *insn++ = BPF_MOV64_IMM(ret, 0); |
1391 | |
1392 | return insn - insn_buf; |
1393 | } |
1394 | |
1395 | const struct bpf_map_ops array_of_maps_map_ops = { |
1396 | .map_alloc_check = fd_array_map_alloc_check, |
1397 | .map_alloc = array_of_map_alloc, |
1398 | .map_free = array_of_map_free, |
1399 | .map_get_next_key = array_map_get_next_key, |
1400 | .map_lookup_elem = array_of_map_lookup_elem, |
1401 | .map_delete_elem = fd_array_map_delete_elem, |
1402 | .map_fd_get_ptr = bpf_map_fd_get_ptr, |
1403 | .map_fd_put_ptr = bpf_map_fd_put_ptr, |
1404 | .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem, |
1405 | .map_gen_lookup = array_of_map_gen_lookup, |
1406 | .map_lookup_batch = generic_map_lookup_batch, |
1407 | .map_update_batch = generic_map_update_batch, |
1408 | .map_check_btf = map_check_no_btf, |
1409 | .map_mem_usage = array_map_mem_usage, |
1410 | .map_btf_id = &array_map_btf_ids[0], |
1411 | }; |
1412 | |