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