Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 */
4#include <linux/bpf.h>
5#include <linux/bpf-cgroup.h>
6#include <linux/bpf_trace.h>
7#include <linux/bpf_lirc.h>
8#include <linux/bpf_verifier.h>
9#include <linux/bsearch.h>
10#include <linux/btf.h>
11#include <linux/syscalls.h>
12#include <linux/slab.h>
13#include <linux/sched/signal.h>
14#include <linux/vmalloc.h>
15#include <linux/mmzone.h>
16#include <linux/anon_inodes.h>
17#include <linux/fdtable.h>
18#include <linux/file.h>
19#include <linux/fs.h>
20#include <linux/license.h>
21#include <linux/filter.h>
22#include <linux/kernel.h>
23#include <linux/idr.h>
24#include <linux/cred.h>
25#include <linux/timekeeping.h>
26#include <linux/ctype.h>
27#include <linux/nospec.h>
28#include <linux/audit.h>
29#include <uapi/linux/btf.h>
30#include <linux/pgtable.h>
31#include <linux/bpf_lsm.h>
32#include <linux/poll.h>
33#include <linux/sort.h>
34#include <linux/bpf-netns.h>
35#include <linux/rcupdate_trace.h>
36#include <linux/memcontrol.h>
37#include <linux/trace_events.h>
38#include <linux/tracepoint.h>
39
40#include <net/netfilter/nf_bpf_link.h>
41#include <net/netkit.h>
42#include <net/tcx.h>
43
44#define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
45 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
46 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
47#define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
48#define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
49#define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
50 IS_FD_HASH(map))
51
52#define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
53
54DEFINE_PER_CPU(int, bpf_prog_active);
55static DEFINE_IDR(prog_idr);
56static DEFINE_SPINLOCK(prog_idr_lock);
57static DEFINE_IDR(map_idr);
58static DEFINE_SPINLOCK(map_idr_lock);
59static DEFINE_IDR(link_idr);
60static DEFINE_SPINLOCK(link_idr_lock);
61
62int sysctl_unprivileged_bpf_disabled __read_mostly =
63 IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
64
65static const struct bpf_map_ops * const bpf_map_types[] = {
66#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
67#define BPF_MAP_TYPE(_id, _ops) \
68 [_id] = &_ops,
69#define BPF_LINK_TYPE(_id, _name)
70#include <linux/bpf_types.h>
71#undef BPF_PROG_TYPE
72#undef BPF_MAP_TYPE
73#undef BPF_LINK_TYPE
74};
75
76/*
77 * If we're handed a bigger struct than we know of, ensure all the unknown bits
78 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
79 * we don't know about yet.
80 *
81 * There is a ToCToU between this function call and the following
82 * copy_from_user() call. However, this is not a concern since this function is
83 * meant to be a future-proofing of bits.
84 */
85int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
86 size_t expected_size,
87 size_t actual_size)
88{
89 int res;
90
91 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
92 return -E2BIG;
93
94 if (actual_size <= expected_size)
95 return 0;
96
97 if (uaddr.is_kernel)
98 res = memchr_inv(uaddr.kernel + expected_size, 0,
99 actual_size - expected_size) == NULL;
100 else
101 res = check_zeroed_user(uaddr.user + expected_size,
102 actual_size - expected_size);
103 if (res < 0)
104 return res;
105 return res ? 0 : -E2BIG;
106}
107
108const struct bpf_map_ops bpf_map_offload_ops = {
109 .map_meta_equal = bpf_map_meta_equal,
110 .map_alloc = bpf_map_offload_map_alloc,
111 .map_free = bpf_map_offload_map_free,
112 .map_check_btf = map_check_no_btf,
113 .map_mem_usage = bpf_map_offload_map_mem_usage,
114};
115
116static void bpf_map_write_active_inc(struct bpf_map *map)
117{
118 atomic64_inc(&map->writecnt);
119}
120
121static void bpf_map_write_active_dec(struct bpf_map *map)
122{
123 atomic64_dec(&map->writecnt);
124}
125
126bool bpf_map_write_active(const struct bpf_map *map)
127{
128 return atomic64_read(&map->writecnt) != 0;
129}
130
131static u32 bpf_map_value_size(const struct bpf_map *map)
132{
133 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
134 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
135 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
136 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
137 return round_up(map->value_size, 8) * num_possible_cpus();
138 else if (IS_FD_MAP(map))
139 return sizeof(u32);
140 else
141 return map->value_size;
142}
143
144static void maybe_wait_bpf_programs(struct bpf_map *map)
145{
146 /* Wait for any running non-sleepable BPF programs to complete so that
147 * userspace, when we return to it, knows that all non-sleepable
148 * programs that could be running use the new map value. For sleepable
149 * BPF programs, synchronize_rcu_tasks_trace() should be used to wait
150 * for the completions of these programs, but considering the waiting
151 * time can be very long and userspace may think it will hang forever,
152 * so don't handle sleepable BPF programs now.
153 */
154 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
155 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
156 synchronize_rcu();
157}
158
159static void unpin_uptr_kaddr(void *kaddr)
160{
161 if (kaddr)
162 unpin_user_page(virt_to_page(kaddr));
163}
164
165static void __bpf_obj_unpin_uptrs(struct btf_record *rec, u32 cnt, void *obj)
166{
167 const struct btf_field *field;
168 void **uptr_addr;
169 int i;
170
171 for (i = 0, field = rec->fields; i < cnt; i++, field++) {
172 if (field->type != BPF_UPTR)
173 continue;
174
175 uptr_addr = obj + field->offset;
176 unpin_uptr_kaddr(*uptr_addr);
177 }
178}
179
180static void bpf_obj_unpin_uptrs(struct btf_record *rec, void *obj)
181{
182 if (!btf_record_has_field(rec, BPF_UPTR))
183 return;
184
185 __bpf_obj_unpin_uptrs(rec, rec->cnt, obj);
186}
187
188static int bpf_obj_pin_uptrs(struct btf_record *rec, void *obj)
189{
190 const struct btf_field *field;
191 const struct btf_type *t;
192 unsigned long start, end;
193 struct page *page;
194 void **uptr_addr;
195 int i, err;
196
197 if (!btf_record_has_field(rec, BPF_UPTR))
198 return 0;
199
200 for (i = 0, field = rec->fields; i < rec->cnt; i++, field++) {
201 if (field->type != BPF_UPTR)
202 continue;
203
204 uptr_addr = obj + field->offset;
205 start = *(unsigned long *)uptr_addr;
206 if (!start)
207 continue;
208
209 t = btf_type_by_id(field->kptr.btf, field->kptr.btf_id);
210 /* t->size was checked for zero before */
211 if (check_add_overflow(start, t->size - 1, &end)) {
212 err = -EFAULT;
213 goto unpin_all;
214 }
215
216 /* The uptr's struct cannot span across two pages */
217 if ((start & PAGE_MASK) != (end & PAGE_MASK)) {
218 err = -EOPNOTSUPP;
219 goto unpin_all;
220 }
221
222 err = pin_user_pages_fast(start, 1, FOLL_LONGTERM | FOLL_WRITE, &page);
223 if (err != 1)
224 goto unpin_all;
225
226 if (PageHighMem(page)) {
227 err = -EOPNOTSUPP;
228 unpin_user_page(page);
229 goto unpin_all;
230 }
231
232 *uptr_addr = page_address(page) + offset_in_page(start);
233 }
234
235 return 0;
236
237unpin_all:
238 __bpf_obj_unpin_uptrs(rec, i, obj);
239 return err;
240}
241
242static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
243 void *key, void *value, __u64 flags)
244{
245 int err;
246
247 /* Need to create a kthread, thus must support schedule */
248 if (bpf_map_is_offloaded(map)) {
249 return bpf_map_offload_update_elem(map, key, value, flags);
250 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
251 map->map_type == BPF_MAP_TYPE_ARENA ||
252 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
253 return map->ops->map_update_elem(map, key, value, flags);
254 } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
255 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
256 return sock_map_update_elem_sys(map, key, value, flags);
257 } else if (IS_FD_PROG_ARRAY(map)) {
258 return bpf_fd_array_map_update_elem(map, map_file, key, value,
259 flags);
260 }
261
262 bpf_disable_instrumentation();
263 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
264 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
265 err = bpf_percpu_hash_update(map, key, value, flags);
266 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
267 err = bpf_percpu_array_update(map, key, value, flags);
268 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
269 err = bpf_percpu_cgroup_storage_update(map, key, value,
270 flags);
271 } else if (IS_FD_ARRAY(map)) {
272 err = bpf_fd_array_map_update_elem(map, map_file, key, value,
273 flags);
274 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
275 err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
276 flags);
277 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
278 /* rcu_read_lock() is not needed */
279 err = bpf_fd_reuseport_array_update_elem(map, key, value,
280 flags);
281 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
282 map->map_type == BPF_MAP_TYPE_STACK ||
283 map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
284 err = map->ops->map_push_elem(map, value, flags);
285 } else {
286 err = bpf_obj_pin_uptrs(map->record, value);
287 if (!err) {
288 rcu_read_lock();
289 err = map->ops->map_update_elem(map, key, value, flags);
290 rcu_read_unlock();
291 if (err)
292 bpf_obj_unpin_uptrs(map->record, value);
293 }
294 }
295 bpf_enable_instrumentation();
296
297 return err;
298}
299
300static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
301 __u64 flags)
302{
303 void *ptr;
304 int err;
305
306 if (bpf_map_is_offloaded(map))
307 return bpf_map_offload_lookup_elem(map, key, value);
308
309 bpf_disable_instrumentation();
310 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
311 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
312 err = bpf_percpu_hash_copy(map, key, value);
313 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
314 err = bpf_percpu_array_copy(map, key, value);
315 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
316 err = bpf_percpu_cgroup_storage_copy(map, key, value);
317 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
318 err = bpf_stackmap_copy(map, key, value);
319 } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
320 err = bpf_fd_array_map_lookup_elem(map, key, value);
321 } else if (IS_FD_HASH(map)) {
322 err = bpf_fd_htab_map_lookup_elem(map, key, value);
323 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
324 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
325 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
326 map->map_type == BPF_MAP_TYPE_STACK ||
327 map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
328 err = map->ops->map_peek_elem(map, value);
329 } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
330 /* struct_ops map requires directly updating "value" */
331 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
332 } else {
333 rcu_read_lock();
334 if (map->ops->map_lookup_elem_sys_only)
335 ptr = map->ops->map_lookup_elem_sys_only(map, key);
336 else
337 ptr = map->ops->map_lookup_elem(map, key);
338 if (IS_ERR(ptr)) {
339 err = PTR_ERR(ptr);
340 } else if (!ptr) {
341 err = -ENOENT;
342 } else {
343 err = 0;
344 if (flags & BPF_F_LOCK)
345 /* lock 'ptr' and copy everything but lock */
346 copy_map_value_locked(map, value, ptr, true);
347 else
348 copy_map_value(map, value, ptr);
349 /* mask lock and timer, since value wasn't zero inited */
350 check_and_init_map_value(map, value);
351 }
352 rcu_read_unlock();
353 }
354
355 bpf_enable_instrumentation();
356
357 return err;
358}
359
360/* Please, do not use this function outside from the map creation path
361 * (e.g. in map update path) without taking care of setting the active
362 * memory cgroup (see at bpf_map_kmalloc_node() for example).
363 */
364static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
365{
366 /* We really just want to fail instead of triggering OOM killer
367 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
368 * which is used for lower order allocation requests.
369 *
370 * It has been observed that higher order allocation requests done by
371 * vmalloc with __GFP_NORETRY being set might fail due to not trying
372 * to reclaim memory from the page cache, thus we set
373 * __GFP_RETRY_MAYFAIL to avoid such situations.
374 */
375
376 gfp_t gfp = bpf_memcg_flags(__GFP_NOWARN | __GFP_ZERO);
377 unsigned int flags = 0;
378 unsigned long align = 1;
379 void *area;
380
381 if (size >= SIZE_MAX)
382 return NULL;
383
384 /* kmalloc()'ed memory can't be mmap()'ed */
385 if (mmapable) {
386 BUG_ON(!PAGE_ALIGNED(size));
387 align = SHMLBA;
388 flags = VM_USERMAP;
389 } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
390 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
391 numa_node);
392 if (area != NULL)
393 return area;
394 }
395
396 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
397 gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
398 flags, numa_node, __builtin_return_address(0));
399}
400
401void *bpf_map_area_alloc(u64 size, int numa_node)
402{
403 return __bpf_map_area_alloc(size, numa_node, false);
404}
405
406void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
407{
408 return __bpf_map_area_alloc(size, numa_node, true);
409}
410
411void bpf_map_area_free(void *area)
412{
413 kvfree(area);
414}
415
416static u32 bpf_map_flags_retain_permanent(u32 flags)
417{
418 /* Some map creation flags are not tied to the map object but
419 * rather to the map fd instead, so they have no meaning upon
420 * map object inspection since multiple file descriptors with
421 * different (access) properties can exist here. Thus, given
422 * this has zero meaning for the map itself, lets clear these
423 * from here.
424 */
425 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
426}
427
428void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
429{
430 map->map_type = attr->map_type;
431 map->key_size = attr->key_size;
432 map->value_size = attr->value_size;
433 map->max_entries = attr->max_entries;
434 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
435 map->numa_node = bpf_map_attr_numa_node(attr);
436 map->map_extra = attr->map_extra;
437}
438
439static int bpf_map_alloc_id(struct bpf_map *map)
440{
441 int id;
442
443 idr_preload(GFP_KERNEL);
444 spin_lock_bh(&map_idr_lock);
445 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
446 if (id > 0)
447 map->id = id;
448 spin_unlock_bh(&map_idr_lock);
449 idr_preload_end();
450
451 if (WARN_ON_ONCE(!id))
452 return -ENOSPC;
453
454 return id > 0 ? 0 : id;
455}
456
457void bpf_map_free_id(struct bpf_map *map)
458{
459 unsigned long flags;
460
461 /* Offloaded maps are removed from the IDR store when their device
462 * disappears - even if someone holds an fd to them they are unusable,
463 * the memory is gone, all ops will fail; they are simply waiting for
464 * refcnt to drop to be freed.
465 */
466 if (!map->id)
467 return;
468
469 spin_lock_irqsave(&map_idr_lock, flags);
470
471 idr_remove(&map_idr, map->id);
472 map->id = 0;
473
474 spin_unlock_irqrestore(&map_idr_lock, flags);
475}
476
477#ifdef CONFIG_MEMCG
478static void bpf_map_save_memcg(struct bpf_map *map)
479{
480 /* Currently if a map is created by a process belonging to the root
481 * memory cgroup, get_obj_cgroup_from_current() will return NULL.
482 * So we have to check map->objcg for being NULL each time it's
483 * being used.
484 */
485 if (memcg_bpf_enabled())
486 map->objcg = get_obj_cgroup_from_current();
487}
488
489static void bpf_map_release_memcg(struct bpf_map *map)
490{
491 if (map->objcg)
492 obj_cgroup_put(map->objcg);
493}
494
495static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
496{
497 if (map->objcg)
498 return get_mem_cgroup_from_objcg(map->objcg);
499
500 return root_mem_cgroup;
501}
502
503void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
504 int node)
505{
506 struct mem_cgroup *memcg, *old_memcg;
507 void *ptr;
508
509 memcg = bpf_map_get_memcg(map);
510 old_memcg = set_active_memcg(memcg);
511 ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
512 set_active_memcg(old_memcg);
513 mem_cgroup_put(memcg);
514
515 return ptr;
516}
517
518void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
519{
520 struct mem_cgroup *memcg, *old_memcg;
521 void *ptr;
522
523 memcg = bpf_map_get_memcg(map);
524 old_memcg = set_active_memcg(memcg);
525 ptr = kzalloc(size, flags | __GFP_ACCOUNT);
526 set_active_memcg(old_memcg);
527 mem_cgroup_put(memcg);
528
529 return ptr;
530}
531
532void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
533 gfp_t flags)
534{
535 struct mem_cgroup *memcg, *old_memcg;
536 void *ptr;
537
538 memcg = bpf_map_get_memcg(map);
539 old_memcg = set_active_memcg(memcg);
540 ptr = kvcalloc(n, size, flags | __GFP_ACCOUNT);
541 set_active_memcg(old_memcg);
542 mem_cgroup_put(memcg);
543
544 return ptr;
545}
546
547void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
548 size_t align, gfp_t flags)
549{
550 struct mem_cgroup *memcg, *old_memcg;
551 void __percpu *ptr;
552
553 memcg = bpf_map_get_memcg(map);
554 old_memcg = set_active_memcg(memcg);
555 ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
556 set_active_memcg(old_memcg);
557 mem_cgroup_put(memcg);
558
559 return ptr;
560}
561
562#else
563static void bpf_map_save_memcg(struct bpf_map *map)
564{
565}
566
567static void bpf_map_release_memcg(struct bpf_map *map)
568{
569}
570#endif
571
572int bpf_map_alloc_pages(const struct bpf_map *map, gfp_t gfp, int nid,
573 unsigned long nr_pages, struct page **pages)
574{
575 unsigned long i, j;
576 struct page *pg;
577 int ret = 0;
578#ifdef CONFIG_MEMCG
579 struct mem_cgroup *memcg, *old_memcg;
580
581 memcg = bpf_map_get_memcg(map);
582 old_memcg = set_active_memcg(memcg);
583#endif
584 for (i = 0; i < nr_pages; i++) {
585 pg = alloc_pages_node(nid, gfp | __GFP_ACCOUNT, 0);
586
587 if (pg) {
588 pages[i] = pg;
589 continue;
590 }
591 for (j = 0; j < i; j++)
592 __free_page(pages[j]);
593 ret = -ENOMEM;
594 break;
595 }
596
597#ifdef CONFIG_MEMCG
598 set_active_memcg(old_memcg);
599 mem_cgroup_put(memcg);
600#endif
601 return ret;
602}
603
604
605static int btf_field_cmp(const void *a, const void *b)
606{
607 const struct btf_field *f1 = a, *f2 = b;
608
609 if (f1->offset < f2->offset)
610 return -1;
611 else if (f1->offset > f2->offset)
612 return 1;
613 return 0;
614}
615
616struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
617 u32 field_mask)
618{
619 struct btf_field *field;
620
621 if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & field_mask))
622 return NULL;
623 field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
624 if (!field || !(field->type & field_mask))
625 return NULL;
626 return field;
627}
628
629void btf_record_free(struct btf_record *rec)
630{
631 int i;
632
633 if (IS_ERR_OR_NULL(rec))
634 return;
635 for (i = 0; i < rec->cnt; i++) {
636 switch (rec->fields[i].type) {
637 case BPF_KPTR_UNREF:
638 case BPF_KPTR_REF:
639 case BPF_KPTR_PERCPU:
640 case BPF_UPTR:
641 if (rec->fields[i].kptr.module)
642 module_put(rec->fields[i].kptr.module);
643 if (btf_is_kernel(rec->fields[i].kptr.btf))
644 btf_put(rec->fields[i].kptr.btf);
645 break;
646 case BPF_LIST_HEAD:
647 case BPF_LIST_NODE:
648 case BPF_RB_ROOT:
649 case BPF_RB_NODE:
650 case BPF_SPIN_LOCK:
651 case BPF_TIMER:
652 case BPF_REFCOUNT:
653 case BPF_WORKQUEUE:
654 /* Nothing to release */
655 break;
656 default:
657 WARN_ON_ONCE(1);
658 continue;
659 }
660 }
661 kfree(rec);
662}
663
664void bpf_map_free_record(struct bpf_map *map)
665{
666 btf_record_free(map->record);
667 map->record = NULL;
668}
669
670struct btf_record *btf_record_dup(const struct btf_record *rec)
671{
672 const struct btf_field *fields;
673 struct btf_record *new_rec;
674 int ret, size, i;
675
676 if (IS_ERR_OR_NULL(rec))
677 return NULL;
678 size = offsetof(struct btf_record, fields[rec->cnt]);
679 new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
680 if (!new_rec)
681 return ERR_PTR(-ENOMEM);
682 /* Do a deep copy of the btf_record */
683 fields = rec->fields;
684 new_rec->cnt = 0;
685 for (i = 0; i < rec->cnt; i++) {
686 switch (fields[i].type) {
687 case BPF_KPTR_UNREF:
688 case BPF_KPTR_REF:
689 case BPF_KPTR_PERCPU:
690 case BPF_UPTR:
691 if (btf_is_kernel(fields[i].kptr.btf))
692 btf_get(fields[i].kptr.btf);
693 if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
694 ret = -ENXIO;
695 goto free;
696 }
697 break;
698 case BPF_LIST_HEAD:
699 case BPF_LIST_NODE:
700 case BPF_RB_ROOT:
701 case BPF_RB_NODE:
702 case BPF_SPIN_LOCK:
703 case BPF_TIMER:
704 case BPF_REFCOUNT:
705 case BPF_WORKQUEUE:
706 /* Nothing to acquire */
707 break;
708 default:
709 ret = -EFAULT;
710 WARN_ON_ONCE(1);
711 goto free;
712 }
713 new_rec->cnt++;
714 }
715 return new_rec;
716free:
717 btf_record_free(new_rec);
718 return ERR_PTR(ret);
719}
720
721bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
722{
723 bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
724 int size;
725
726 if (!a_has_fields && !b_has_fields)
727 return true;
728 if (a_has_fields != b_has_fields)
729 return false;
730 if (rec_a->cnt != rec_b->cnt)
731 return false;
732 size = offsetof(struct btf_record, fields[rec_a->cnt]);
733 /* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
734 * members are zeroed out. So memcmp is safe to do without worrying
735 * about padding/unused fields.
736 *
737 * While spin_lock, timer, and kptr have no relation to map BTF,
738 * list_head metadata is specific to map BTF, the btf and value_rec
739 * members in particular. btf is the map BTF, while value_rec points to
740 * btf_record in that map BTF.
741 *
742 * So while by default, we don't rely on the map BTF (which the records
743 * were parsed from) matching for both records, which is not backwards
744 * compatible, in case list_head is part of it, we implicitly rely on
745 * that by way of depending on memcmp succeeding for it.
746 */
747 return !memcmp(rec_a, rec_b, size);
748}
749
750void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
751{
752 if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
753 return;
754 bpf_timer_cancel_and_free(obj + rec->timer_off);
755}
756
757void bpf_obj_free_workqueue(const struct btf_record *rec, void *obj)
758{
759 if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_WORKQUEUE)))
760 return;
761 bpf_wq_cancel_and_free(obj + rec->wq_off);
762}
763
764void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
765{
766 const struct btf_field *fields;
767 int i;
768
769 if (IS_ERR_OR_NULL(rec))
770 return;
771 fields = rec->fields;
772 for (i = 0; i < rec->cnt; i++) {
773 struct btf_struct_meta *pointee_struct_meta;
774 const struct btf_field *field = &fields[i];
775 void *field_ptr = obj + field->offset;
776 void *xchgd_field;
777
778 switch (fields[i].type) {
779 case BPF_SPIN_LOCK:
780 break;
781 case BPF_TIMER:
782 bpf_timer_cancel_and_free(field_ptr);
783 break;
784 case BPF_WORKQUEUE:
785 bpf_wq_cancel_and_free(field_ptr);
786 break;
787 case BPF_KPTR_UNREF:
788 WRITE_ONCE(*(u64 *)field_ptr, 0);
789 break;
790 case BPF_KPTR_REF:
791 case BPF_KPTR_PERCPU:
792 xchgd_field = (void *)xchg((unsigned long *)field_ptr, 0);
793 if (!xchgd_field)
794 break;
795
796 if (!btf_is_kernel(field->kptr.btf)) {
797 pointee_struct_meta = btf_find_struct_meta(field->kptr.btf,
798 field->kptr.btf_id);
799 migrate_disable();
800 __bpf_obj_drop_impl(xchgd_field, pointee_struct_meta ?
801 pointee_struct_meta->record : NULL,
802 fields[i].type == BPF_KPTR_PERCPU);
803 migrate_enable();
804 } else {
805 field->kptr.dtor(xchgd_field);
806 }
807 break;
808 case BPF_UPTR:
809 /* The caller ensured that no one is using the uptr */
810 unpin_uptr_kaddr(*(void **)field_ptr);
811 break;
812 case BPF_LIST_HEAD:
813 if (WARN_ON_ONCE(rec->spin_lock_off < 0))
814 continue;
815 bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
816 break;
817 case BPF_RB_ROOT:
818 if (WARN_ON_ONCE(rec->spin_lock_off < 0))
819 continue;
820 bpf_rb_root_free(field, field_ptr, obj + rec->spin_lock_off);
821 break;
822 case BPF_LIST_NODE:
823 case BPF_RB_NODE:
824 case BPF_REFCOUNT:
825 break;
826 default:
827 WARN_ON_ONCE(1);
828 continue;
829 }
830 }
831}
832
833static void bpf_map_free(struct bpf_map *map)
834{
835 struct btf_record *rec = map->record;
836 struct btf *btf = map->btf;
837
838 /* implementation dependent freeing */
839 map->ops->map_free(map);
840 /* Delay freeing of btf_record for maps, as map_free
841 * callback usually needs access to them. It is better to do it here
842 * than require each callback to do the free itself manually.
843 *
844 * Note that the btf_record stashed in map->inner_map_meta->record was
845 * already freed using the map_free callback for map in map case which
846 * eventually calls bpf_map_free_meta, since inner_map_meta is only a
847 * template bpf_map struct used during verification.
848 */
849 btf_record_free(rec);
850 /* Delay freeing of btf for maps, as map_free callback may need
851 * struct_meta info which will be freed with btf_put().
852 */
853 btf_put(btf);
854}
855
856/* called from workqueue */
857static void bpf_map_free_deferred(struct work_struct *work)
858{
859 struct bpf_map *map = container_of(work, struct bpf_map, work);
860
861 security_bpf_map_free(map);
862 bpf_map_release_memcg(map);
863 bpf_map_free(map);
864}
865
866static void bpf_map_put_uref(struct bpf_map *map)
867{
868 if (atomic64_dec_and_test(&map->usercnt)) {
869 if (map->ops->map_release_uref)
870 map->ops->map_release_uref(map);
871 }
872}
873
874static void bpf_map_free_in_work(struct bpf_map *map)
875{
876 INIT_WORK(&map->work, bpf_map_free_deferred);
877 /* Avoid spawning kworkers, since they all might contend
878 * for the same mutex like slab_mutex.
879 */
880 queue_work(system_unbound_wq, &map->work);
881}
882
883static void bpf_map_free_rcu_gp(struct rcu_head *rcu)
884{
885 bpf_map_free_in_work(container_of(rcu, struct bpf_map, rcu));
886}
887
888static void bpf_map_free_mult_rcu_gp(struct rcu_head *rcu)
889{
890 if (rcu_trace_implies_rcu_gp())
891 bpf_map_free_rcu_gp(rcu);
892 else
893 call_rcu(rcu, bpf_map_free_rcu_gp);
894}
895
896/* decrement map refcnt and schedule it for freeing via workqueue
897 * (underlying map implementation ops->map_free() might sleep)
898 */
899void bpf_map_put(struct bpf_map *map)
900{
901 if (atomic64_dec_and_test(&map->refcnt)) {
902 /* bpf_map_free_id() must be called first */
903 bpf_map_free_id(map);
904
905 WARN_ON_ONCE(atomic64_read(&map->sleepable_refcnt));
906 if (READ_ONCE(map->free_after_mult_rcu_gp))
907 call_rcu_tasks_trace(&map->rcu, bpf_map_free_mult_rcu_gp);
908 else if (READ_ONCE(map->free_after_rcu_gp))
909 call_rcu(&map->rcu, bpf_map_free_rcu_gp);
910 else
911 bpf_map_free_in_work(map);
912 }
913}
914EXPORT_SYMBOL_GPL(bpf_map_put);
915
916void bpf_map_put_with_uref(struct bpf_map *map)
917{
918 bpf_map_put_uref(map);
919 bpf_map_put(map);
920}
921
922static int bpf_map_release(struct inode *inode, struct file *filp)
923{
924 struct bpf_map *map = filp->private_data;
925
926 if (map->ops->map_release)
927 map->ops->map_release(map, filp);
928
929 bpf_map_put_with_uref(map);
930 return 0;
931}
932
933static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
934{
935 fmode_t mode = fd_file(f)->f_mode;
936
937 /* Our file permissions may have been overridden by global
938 * map permissions facing syscall side.
939 */
940 if (READ_ONCE(map->frozen))
941 mode &= ~FMODE_CAN_WRITE;
942 return mode;
943}
944
945#ifdef CONFIG_PROC_FS
946/* Show the memory usage of a bpf map */
947static u64 bpf_map_memory_usage(const struct bpf_map *map)
948{
949 return map->ops->map_mem_usage(map);
950}
951
952static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
953{
954 struct bpf_map *map = filp->private_data;
955 u32 type = 0, jited = 0;
956
957 if (map_type_contains_progs(map)) {
958 spin_lock(&map->owner.lock);
959 type = map->owner.type;
960 jited = map->owner.jited;
961 spin_unlock(&map->owner.lock);
962 }
963
964 seq_printf(m,
965 "map_type:\t%u\n"
966 "key_size:\t%u\n"
967 "value_size:\t%u\n"
968 "max_entries:\t%u\n"
969 "map_flags:\t%#x\n"
970 "map_extra:\t%#llx\n"
971 "memlock:\t%llu\n"
972 "map_id:\t%u\n"
973 "frozen:\t%u\n",
974 map->map_type,
975 map->key_size,
976 map->value_size,
977 map->max_entries,
978 map->map_flags,
979 (unsigned long long)map->map_extra,
980 bpf_map_memory_usage(map),
981 map->id,
982 READ_ONCE(map->frozen));
983 if (type) {
984 seq_printf(m, "owner_prog_type:\t%u\n", type);
985 seq_printf(m, "owner_jited:\t%u\n", jited);
986 }
987}
988#endif
989
990static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
991 loff_t *ppos)
992{
993 /* We need this handler such that alloc_file() enables
994 * f_mode with FMODE_CAN_READ.
995 */
996 return -EINVAL;
997}
998
999static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
1000 size_t siz, loff_t *ppos)
1001{
1002 /* We need this handler such that alloc_file() enables
1003 * f_mode with FMODE_CAN_WRITE.
1004 */
1005 return -EINVAL;
1006}
1007
1008/* called for any extra memory-mapped regions (except initial) */
1009static void bpf_map_mmap_open(struct vm_area_struct *vma)
1010{
1011 struct bpf_map *map = vma->vm_file->private_data;
1012
1013 if (vma->vm_flags & VM_MAYWRITE)
1014 bpf_map_write_active_inc(map);
1015}
1016
1017/* called for all unmapped memory region (including initial) */
1018static void bpf_map_mmap_close(struct vm_area_struct *vma)
1019{
1020 struct bpf_map *map = vma->vm_file->private_data;
1021
1022 if (vma->vm_flags & VM_MAYWRITE)
1023 bpf_map_write_active_dec(map);
1024}
1025
1026static const struct vm_operations_struct bpf_map_default_vmops = {
1027 .open = bpf_map_mmap_open,
1028 .close = bpf_map_mmap_close,
1029};
1030
1031static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
1032{
1033 struct bpf_map *map = filp->private_data;
1034 int err = 0;
1035
1036 if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
1037 return -ENOTSUPP;
1038
1039 if (!(vma->vm_flags & VM_SHARED))
1040 return -EINVAL;
1041
1042 mutex_lock(&map->freeze_mutex);
1043
1044 if (vma->vm_flags & VM_WRITE) {
1045 if (map->frozen) {
1046 err = -EPERM;
1047 goto out;
1048 }
1049 /* map is meant to be read-only, so do not allow mapping as
1050 * writable, because it's possible to leak a writable page
1051 * reference and allows user-space to still modify it after
1052 * freezing, while verifier will assume contents do not change
1053 */
1054 if (map->map_flags & BPF_F_RDONLY_PROG) {
1055 err = -EACCES;
1056 goto out;
1057 }
1058 bpf_map_write_active_inc(map);
1059 }
1060out:
1061 mutex_unlock(&map->freeze_mutex);
1062 if (err)
1063 return err;
1064
1065 /* set default open/close callbacks */
1066 vma->vm_ops = &bpf_map_default_vmops;
1067 vma->vm_private_data = map;
1068 vm_flags_clear(vma, VM_MAYEXEC);
1069 /* If mapping is read-only, then disallow potentially re-mapping with
1070 * PROT_WRITE by dropping VM_MAYWRITE flag. This VM_MAYWRITE clearing
1071 * means that as far as BPF map's memory-mapped VMAs are concerned,
1072 * VM_WRITE and VM_MAYWRITE and equivalent, if one of them is set,
1073 * both should be set, so we can forget about VM_MAYWRITE and always
1074 * check just VM_WRITE
1075 */
1076 if (!(vma->vm_flags & VM_WRITE))
1077 vm_flags_clear(vma, VM_MAYWRITE);
1078
1079 err = map->ops->map_mmap(map, vma);
1080 if (err) {
1081 if (vma->vm_flags & VM_WRITE)
1082 bpf_map_write_active_dec(map);
1083 }
1084
1085 return err;
1086}
1087
1088static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
1089{
1090 struct bpf_map *map = filp->private_data;
1091
1092 if (map->ops->map_poll)
1093 return map->ops->map_poll(map, filp, pts);
1094
1095 return EPOLLERR;
1096}
1097
1098static unsigned long bpf_get_unmapped_area(struct file *filp, unsigned long addr,
1099 unsigned long len, unsigned long pgoff,
1100 unsigned long flags)
1101{
1102 struct bpf_map *map = filp->private_data;
1103
1104 if (map->ops->map_get_unmapped_area)
1105 return map->ops->map_get_unmapped_area(filp, addr, len, pgoff, flags);
1106#ifdef CONFIG_MMU
1107 return mm_get_unmapped_area(current->mm, filp, addr, len, pgoff, flags);
1108#else
1109 return addr;
1110#endif
1111}
1112
1113const struct file_operations bpf_map_fops = {
1114#ifdef CONFIG_PROC_FS
1115 .show_fdinfo = bpf_map_show_fdinfo,
1116#endif
1117 .release = bpf_map_release,
1118 .read = bpf_dummy_read,
1119 .write = bpf_dummy_write,
1120 .mmap = bpf_map_mmap,
1121 .poll = bpf_map_poll,
1122 .get_unmapped_area = bpf_get_unmapped_area,
1123};
1124
1125int bpf_map_new_fd(struct bpf_map *map, int flags)
1126{
1127 int ret;
1128
1129 ret = security_bpf_map(map, OPEN_FMODE(flags));
1130 if (ret < 0)
1131 return ret;
1132
1133 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
1134 flags | O_CLOEXEC);
1135}
1136
1137int bpf_get_file_flag(int flags)
1138{
1139 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
1140 return -EINVAL;
1141 if (flags & BPF_F_RDONLY)
1142 return O_RDONLY;
1143 if (flags & BPF_F_WRONLY)
1144 return O_WRONLY;
1145 return O_RDWR;
1146}
1147
1148/* helper macro to check that unused fields 'union bpf_attr' are zero */
1149#define CHECK_ATTR(CMD) \
1150 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
1151 sizeof(attr->CMD##_LAST_FIELD), 0, \
1152 sizeof(*attr) - \
1153 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
1154 sizeof(attr->CMD##_LAST_FIELD)) != NULL
1155
1156/* dst and src must have at least "size" number of bytes.
1157 * Return strlen on success and < 0 on error.
1158 */
1159int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
1160{
1161 const char *end = src + size;
1162 const char *orig_src = src;
1163
1164 memset(dst, 0, size);
1165 /* Copy all isalnum(), '_' and '.' chars. */
1166 while (src < end && *src) {
1167 if (!isalnum(*src) &&
1168 *src != '_' && *src != '.')
1169 return -EINVAL;
1170 *dst++ = *src++;
1171 }
1172
1173 /* No '\0' found in "size" number of bytes */
1174 if (src == end)
1175 return -EINVAL;
1176
1177 return src - orig_src;
1178}
1179
1180int map_check_no_btf(const struct bpf_map *map,
1181 const struct btf *btf,
1182 const struct btf_type *key_type,
1183 const struct btf_type *value_type)
1184{
1185 return -ENOTSUPP;
1186}
1187
1188static int map_check_btf(struct bpf_map *map, struct bpf_token *token,
1189 const struct btf *btf, u32 btf_key_id, u32 btf_value_id)
1190{
1191 const struct btf_type *key_type, *value_type;
1192 u32 key_size, value_size;
1193 int ret = 0;
1194
1195 /* Some maps allow key to be unspecified. */
1196 if (btf_key_id) {
1197 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
1198 if (!key_type || key_size != map->key_size)
1199 return -EINVAL;
1200 } else {
1201 key_type = btf_type_by_id(btf, 0);
1202 if (!map->ops->map_check_btf)
1203 return -EINVAL;
1204 }
1205
1206 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1207 if (!value_type || value_size != map->value_size)
1208 return -EINVAL;
1209
1210 map->record = btf_parse_fields(btf, value_type,
1211 BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
1212 BPF_RB_ROOT | BPF_REFCOUNT | BPF_WORKQUEUE | BPF_UPTR,
1213 map->value_size);
1214 if (!IS_ERR_OR_NULL(map->record)) {
1215 int i;
1216
1217 if (!bpf_token_capable(token, CAP_BPF)) {
1218 ret = -EPERM;
1219 goto free_map_tab;
1220 }
1221 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1222 ret = -EACCES;
1223 goto free_map_tab;
1224 }
1225 for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1226 switch (map->record->field_mask & (1 << i)) {
1227 case 0:
1228 continue;
1229 case BPF_SPIN_LOCK:
1230 if (map->map_type != BPF_MAP_TYPE_HASH &&
1231 map->map_type != BPF_MAP_TYPE_ARRAY &&
1232 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1233 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1234 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1235 map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1236 map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1237 ret = -EOPNOTSUPP;
1238 goto free_map_tab;
1239 }
1240 break;
1241 case BPF_TIMER:
1242 case BPF_WORKQUEUE:
1243 if (map->map_type != BPF_MAP_TYPE_HASH &&
1244 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1245 map->map_type != BPF_MAP_TYPE_ARRAY) {
1246 ret = -EOPNOTSUPP;
1247 goto free_map_tab;
1248 }
1249 break;
1250 case BPF_KPTR_UNREF:
1251 case BPF_KPTR_REF:
1252 case BPF_KPTR_PERCPU:
1253 case BPF_REFCOUNT:
1254 if (map->map_type != BPF_MAP_TYPE_HASH &&
1255 map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
1256 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1257 map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH &&
1258 map->map_type != BPF_MAP_TYPE_ARRAY &&
1259 map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY &&
1260 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1261 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1262 map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1263 map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1264 ret = -EOPNOTSUPP;
1265 goto free_map_tab;
1266 }
1267 break;
1268 case BPF_UPTR:
1269 if (map->map_type != BPF_MAP_TYPE_TASK_STORAGE) {
1270 ret = -EOPNOTSUPP;
1271 goto free_map_tab;
1272 }
1273 break;
1274 case BPF_LIST_HEAD:
1275 case BPF_RB_ROOT:
1276 if (map->map_type != BPF_MAP_TYPE_HASH &&
1277 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1278 map->map_type != BPF_MAP_TYPE_ARRAY) {
1279 ret = -EOPNOTSUPP;
1280 goto free_map_tab;
1281 }
1282 break;
1283 default:
1284 /* Fail if map_type checks are missing for a field type */
1285 ret = -EOPNOTSUPP;
1286 goto free_map_tab;
1287 }
1288 }
1289 }
1290
1291 ret = btf_check_and_fixup_fields(btf, map->record);
1292 if (ret < 0)
1293 goto free_map_tab;
1294
1295 if (map->ops->map_check_btf) {
1296 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1297 if (ret < 0)
1298 goto free_map_tab;
1299 }
1300
1301 return ret;
1302free_map_tab:
1303 bpf_map_free_record(map);
1304 return ret;
1305}
1306
1307static bool bpf_net_capable(void)
1308{
1309 return capable(CAP_NET_ADMIN) || capable(CAP_SYS_ADMIN);
1310}
1311
1312#define BPF_MAP_CREATE_LAST_FIELD map_token_fd
1313/* called via syscall */
1314static int map_create(union bpf_attr *attr)
1315{
1316 const struct bpf_map_ops *ops;
1317 struct bpf_token *token = NULL;
1318 int numa_node = bpf_map_attr_numa_node(attr);
1319 u32 map_type = attr->map_type;
1320 struct bpf_map *map;
1321 bool token_flag;
1322 int f_flags;
1323 int err;
1324
1325 err = CHECK_ATTR(BPF_MAP_CREATE);
1326 if (err)
1327 return -EINVAL;
1328
1329 /* check BPF_F_TOKEN_FD flag, remember if it's set, and then clear it
1330 * to avoid per-map type checks tripping on unknown flag
1331 */
1332 token_flag = attr->map_flags & BPF_F_TOKEN_FD;
1333 attr->map_flags &= ~BPF_F_TOKEN_FD;
1334
1335 if (attr->btf_vmlinux_value_type_id) {
1336 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1337 attr->btf_key_type_id || attr->btf_value_type_id)
1338 return -EINVAL;
1339 } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1340 return -EINVAL;
1341 }
1342
1343 if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1344 attr->map_type != BPF_MAP_TYPE_ARENA &&
1345 attr->map_extra != 0)
1346 return -EINVAL;
1347
1348 f_flags = bpf_get_file_flag(attr->map_flags);
1349 if (f_flags < 0)
1350 return f_flags;
1351
1352 if (numa_node != NUMA_NO_NODE &&
1353 ((unsigned int)numa_node >= nr_node_ids ||
1354 !node_online(numa_node)))
1355 return -EINVAL;
1356
1357 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1358 map_type = attr->map_type;
1359 if (map_type >= ARRAY_SIZE(bpf_map_types))
1360 return -EINVAL;
1361 map_type = array_index_nospec(map_type, ARRAY_SIZE(bpf_map_types));
1362 ops = bpf_map_types[map_type];
1363 if (!ops)
1364 return -EINVAL;
1365
1366 if (ops->map_alloc_check) {
1367 err = ops->map_alloc_check(attr);
1368 if (err)
1369 return err;
1370 }
1371 if (attr->map_ifindex)
1372 ops = &bpf_map_offload_ops;
1373 if (!ops->map_mem_usage)
1374 return -EINVAL;
1375
1376 if (token_flag) {
1377 token = bpf_token_get_from_fd(attr->map_token_fd);
1378 if (IS_ERR(token))
1379 return PTR_ERR(token);
1380
1381 /* if current token doesn't grant map creation permissions,
1382 * then we can't use this token, so ignore it and rely on
1383 * system-wide capabilities checks
1384 */
1385 if (!bpf_token_allow_cmd(token, BPF_MAP_CREATE) ||
1386 !bpf_token_allow_map_type(token, attr->map_type)) {
1387 bpf_token_put(token);
1388 token = NULL;
1389 }
1390 }
1391
1392 err = -EPERM;
1393
1394 /* Intent here is for unprivileged_bpf_disabled to block BPF map
1395 * creation for unprivileged users; other actions depend
1396 * on fd availability and access to bpffs, so are dependent on
1397 * object creation success. Even with unprivileged BPF disabled,
1398 * capability checks are still carried out.
1399 */
1400 if (sysctl_unprivileged_bpf_disabled && !bpf_token_capable(token, CAP_BPF))
1401 goto put_token;
1402
1403 /* check privileged map type permissions */
1404 switch (map_type) {
1405 case BPF_MAP_TYPE_ARRAY:
1406 case BPF_MAP_TYPE_PERCPU_ARRAY:
1407 case BPF_MAP_TYPE_PROG_ARRAY:
1408 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1409 case BPF_MAP_TYPE_CGROUP_ARRAY:
1410 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1411 case BPF_MAP_TYPE_HASH:
1412 case BPF_MAP_TYPE_PERCPU_HASH:
1413 case BPF_MAP_TYPE_HASH_OF_MAPS:
1414 case BPF_MAP_TYPE_RINGBUF:
1415 case BPF_MAP_TYPE_USER_RINGBUF:
1416 case BPF_MAP_TYPE_CGROUP_STORAGE:
1417 case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
1418 /* unprivileged */
1419 break;
1420 case BPF_MAP_TYPE_SK_STORAGE:
1421 case BPF_MAP_TYPE_INODE_STORAGE:
1422 case BPF_MAP_TYPE_TASK_STORAGE:
1423 case BPF_MAP_TYPE_CGRP_STORAGE:
1424 case BPF_MAP_TYPE_BLOOM_FILTER:
1425 case BPF_MAP_TYPE_LPM_TRIE:
1426 case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
1427 case BPF_MAP_TYPE_STACK_TRACE:
1428 case BPF_MAP_TYPE_QUEUE:
1429 case BPF_MAP_TYPE_STACK:
1430 case BPF_MAP_TYPE_LRU_HASH:
1431 case BPF_MAP_TYPE_LRU_PERCPU_HASH:
1432 case BPF_MAP_TYPE_STRUCT_OPS:
1433 case BPF_MAP_TYPE_CPUMAP:
1434 case BPF_MAP_TYPE_ARENA:
1435 if (!bpf_token_capable(token, CAP_BPF))
1436 goto put_token;
1437 break;
1438 case BPF_MAP_TYPE_SOCKMAP:
1439 case BPF_MAP_TYPE_SOCKHASH:
1440 case BPF_MAP_TYPE_DEVMAP:
1441 case BPF_MAP_TYPE_DEVMAP_HASH:
1442 case BPF_MAP_TYPE_XSKMAP:
1443 if (!bpf_token_capable(token, CAP_NET_ADMIN))
1444 goto put_token;
1445 break;
1446 default:
1447 WARN(1, "unsupported map type %d", map_type);
1448 goto put_token;
1449 }
1450
1451 map = ops->map_alloc(attr);
1452 if (IS_ERR(map)) {
1453 err = PTR_ERR(map);
1454 goto put_token;
1455 }
1456 map->ops = ops;
1457 map->map_type = map_type;
1458
1459 err = bpf_obj_name_cpy(map->name, attr->map_name,
1460 sizeof(attr->map_name));
1461 if (err < 0)
1462 goto free_map;
1463
1464 atomic64_set(&map->refcnt, 1);
1465 atomic64_set(&map->usercnt, 1);
1466 mutex_init(&map->freeze_mutex);
1467 spin_lock_init(&map->owner.lock);
1468
1469 if (attr->btf_key_type_id || attr->btf_value_type_id ||
1470 /* Even the map's value is a kernel's struct,
1471 * the bpf_prog.o must have BTF to begin with
1472 * to figure out the corresponding kernel's
1473 * counter part. Thus, attr->btf_fd has
1474 * to be valid also.
1475 */
1476 attr->btf_vmlinux_value_type_id) {
1477 struct btf *btf;
1478
1479 btf = btf_get_by_fd(attr->btf_fd);
1480 if (IS_ERR(btf)) {
1481 err = PTR_ERR(btf);
1482 goto free_map;
1483 }
1484 if (btf_is_kernel(btf)) {
1485 btf_put(btf);
1486 err = -EACCES;
1487 goto free_map;
1488 }
1489 map->btf = btf;
1490
1491 if (attr->btf_value_type_id) {
1492 err = map_check_btf(map, token, btf, attr->btf_key_type_id,
1493 attr->btf_value_type_id);
1494 if (err)
1495 goto free_map;
1496 }
1497
1498 map->btf_key_type_id = attr->btf_key_type_id;
1499 map->btf_value_type_id = attr->btf_value_type_id;
1500 map->btf_vmlinux_value_type_id =
1501 attr->btf_vmlinux_value_type_id;
1502 }
1503
1504 err = security_bpf_map_create(map, attr, token);
1505 if (err)
1506 goto free_map_sec;
1507
1508 err = bpf_map_alloc_id(map);
1509 if (err)
1510 goto free_map_sec;
1511
1512 bpf_map_save_memcg(map);
1513 bpf_token_put(token);
1514
1515 err = bpf_map_new_fd(map, f_flags);
1516 if (err < 0) {
1517 /* failed to allocate fd.
1518 * bpf_map_put_with_uref() is needed because the above
1519 * bpf_map_alloc_id() has published the map
1520 * to the userspace and the userspace may
1521 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1522 */
1523 bpf_map_put_with_uref(map);
1524 return err;
1525 }
1526
1527 return err;
1528
1529free_map_sec:
1530 security_bpf_map_free(map);
1531free_map:
1532 bpf_map_free(map);
1533put_token:
1534 bpf_token_put(token);
1535 return err;
1536}
1537
1538void bpf_map_inc(struct bpf_map *map)
1539{
1540 atomic64_inc(&map->refcnt);
1541}
1542EXPORT_SYMBOL_GPL(bpf_map_inc);
1543
1544void bpf_map_inc_with_uref(struct bpf_map *map)
1545{
1546 atomic64_inc(&map->refcnt);
1547 atomic64_inc(&map->usercnt);
1548}
1549EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1550
1551struct bpf_map *bpf_map_get(u32 ufd)
1552{
1553 CLASS(fd, f)(ufd);
1554 struct bpf_map *map = __bpf_map_get(f);
1555
1556 if (!IS_ERR(map))
1557 bpf_map_inc(map);
1558
1559 return map;
1560}
1561EXPORT_SYMBOL(bpf_map_get);
1562
1563struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1564{
1565 CLASS(fd, f)(ufd);
1566 struct bpf_map *map = __bpf_map_get(f);
1567
1568 if (!IS_ERR(map))
1569 bpf_map_inc_with_uref(map);
1570
1571 return map;
1572}
1573
1574/* map_idr_lock should have been held or the map should have been
1575 * protected by rcu read lock.
1576 */
1577struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1578{
1579 int refold;
1580
1581 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1582 if (!refold)
1583 return ERR_PTR(-ENOENT);
1584 if (uref)
1585 atomic64_inc(&map->usercnt);
1586
1587 return map;
1588}
1589
1590struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1591{
1592 spin_lock_bh(&map_idr_lock);
1593 map = __bpf_map_inc_not_zero(map, false);
1594 spin_unlock_bh(&map_idr_lock);
1595
1596 return map;
1597}
1598EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1599
1600int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1601{
1602 return -ENOTSUPP;
1603}
1604
1605static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1606{
1607 if (key_size)
1608 return vmemdup_user(ukey, key_size);
1609
1610 if (ukey)
1611 return ERR_PTR(-EINVAL);
1612
1613 return NULL;
1614}
1615
1616static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1617{
1618 if (key_size)
1619 return kvmemdup_bpfptr(ukey, key_size);
1620
1621 if (!bpfptr_is_null(ukey))
1622 return ERR_PTR(-EINVAL);
1623
1624 return NULL;
1625}
1626
1627/* last field in 'union bpf_attr' used by this command */
1628#define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1629
1630static int map_lookup_elem(union bpf_attr *attr)
1631{
1632 void __user *ukey = u64_to_user_ptr(attr->key);
1633 void __user *uvalue = u64_to_user_ptr(attr->value);
1634 struct bpf_map *map;
1635 void *key, *value;
1636 u32 value_size;
1637 int err;
1638
1639 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1640 return -EINVAL;
1641
1642 if (attr->flags & ~BPF_F_LOCK)
1643 return -EINVAL;
1644
1645 CLASS(fd, f)(attr->map_fd);
1646 map = __bpf_map_get(f);
1647 if (IS_ERR(map))
1648 return PTR_ERR(map);
1649 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ))
1650 return -EPERM;
1651
1652 if ((attr->flags & BPF_F_LOCK) &&
1653 !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1654 return -EINVAL;
1655
1656 key = __bpf_copy_key(ukey, map->key_size);
1657 if (IS_ERR(key))
1658 return PTR_ERR(key);
1659
1660 value_size = bpf_map_value_size(map);
1661
1662 err = -ENOMEM;
1663 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1664 if (!value)
1665 goto free_key;
1666
1667 if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1668 if (copy_from_user(value, uvalue, value_size))
1669 err = -EFAULT;
1670 else
1671 err = bpf_map_copy_value(map, key, value, attr->flags);
1672 goto free_value;
1673 }
1674
1675 err = bpf_map_copy_value(map, key, value, attr->flags);
1676 if (err)
1677 goto free_value;
1678
1679 err = -EFAULT;
1680 if (copy_to_user(uvalue, value, value_size) != 0)
1681 goto free_value;
1682
1683 err = 0;
1684
1685free_value:
1686 kvfree(value);
1687free_key:
1688 kvfree(key);
1689 return err;
1690}
1691
1692
1693#define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1694
1695static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1696{
1697 bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1698 bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1699 struct bpf_map *map;
1700 void *key, *value;
1701 u32 value_size;
1702 int err;
1703
1704 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1705 return -EINVAL;
1706
1707 CLASS(fd, f)(attr->map_fd);
1708 map = __bpf_map_get(f);
1709 if (IS_ERR(map))
1710 return PTR_ERR(map);
1711 bpf_map_write_active_inc(map);
1712 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1713 err = -EPERM;
1714 goto err_put;
1715 }
1716
1717 if ((attr->flags & BPF_F_LOCK) &&
1718 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1719 err = -EINVAL;
1720 goto err_put;
1721 }
1722
1723 key = ___bpf_copy_key(ukey, map->key_size);
1724 if (IS_ERR(key)) {
1725 err = PTR_ERR(key);
1726 goto err_put;
1727 }
1728
1729 value_size = bpf_map_value_size(map);
1730 value = kvmemdup_bpfptr(uvalue, value_size);
1731 if (IS_ERR(value)) {
1732 err = PTR_ERR(value);
1733 goto free_key;
1734 }
1735
1736 err = bpf_map_update_value(map, fd_file(f), key, value, attr->flags);
1737 if (!err)
1738 maybe_wait_bpf_programs(map);
1739
1740 kvfree(value);
1741free_key:
1742 kvfree(key);
1743err_put:
1744 bpf_map_write_active_dec(map);
1745 return err;
1746}
1747
1748#define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1749
1750static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1751{
1752 bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1753 struct bpf_map *map;
1754 void *key;
1755 int err;
1756
1757 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1758 return -EINVAL;
1759
1760 CLASS(fd, f)(attr->map_fd);
1761 map = __bpf_map_get(f);
1762 if (IS_ERR(map))
1763 return PTR_ERR(map);
1764 bpf_map_write_active_inc(map);
1765 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1766 err = -EPERM;
1767 goto err_put;
1768 }
1769
1770 key = ___bpf_copy_key(ukey, map->key_size);
1771 if (IS_ERR(key)) {
1772 err = PTR_ERR(key);
1773 goto err_put;
1774 }
1775
1776 if (bpf_map_is_offloaded(map)) {
1777 err = bpf_map_offload_delete_elem(map, key);
1778 goto out;
1779 } else if (IS_FD_PROG_ARRAY(map) ||
1780 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1781 /* These maps require sleepable context */
1782 err = map->ops->map_delete_elem(map, key);
1783 goto out;
1784 }
1785
1786 bpf_disable_instrumentation();
1787 rcu_read_lock();
1788 err = map->ops->map_delete_elem(map, key);
1789 rcu_read_unlock();
1790 bpf_enable_instrumentation();
1791 if (!err)
1792 maybe_wait_bpf_programs(map);
1793out:
1794 kvfree(key);
1795err_put:
1796 bpf_map_write_active_dec(map);
1797 return err;
1798}
1799
1800/* last field in 'union bpf_attr' used by this command */
1801#define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1802
1803static int map_get_next_key(union bpf_attr *attr)
1804{
1805 void __user *ukey = u64_to_user_ptr(attr->key);
1806 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1807 struct bpf_map *map;
1808 void *key, *next_key;
1809 int err;
1810
1811 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1812 return -EINVAL;
1813
1814 CLASS(fd, f)(attr->map_fd);
1815 map = __bpf_map_get(f);
1816 if (IS_ERR(map))
1817 return PTR_ERR(map);
1818 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ))
1819 return -EPERM;
1820
1821 if (ukey) {
1822 key = __bpf_copy_key(ukey, map->key_size);
1823 if (IS_ERR(key))
1824 return PTR_ERR(key);
1825 } else {
1826 key = NULL;
1827 }
1828
1829 err = -ENOMEM;
1830 next_key = kvmalloc(map->key_size, GFP_USER);
1831 if (!next_key)
1832 goto free_key;
1833
1834 if (bpf_map_is_offloaded(map)) {
1835 err = bpf_map_offload_get_next_key(map, key, next_key);
1836 goto out;
1837 }
1838
1839 rcu_read_lock();
1840 err = map->ops->map_get_next_key(map, key, next_key);
1841 rcu_read_unlock();
1842out:
1843 if (err)
1844 goto free_next_key;
1845
1846 err = -EFAULT;
1847 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1848 goto free_next_key;
1849
1850 err = 0;
1851
1852free_next_key:
1853 kvfree(next_key);
1854free_key:
1855 kvfree(key);
1856 return err;
1857}
1858
1859int generic_map_delete_batch(struct bpf_map *map,
1860 const union bpf_attr *attr,
1861 union bpf_attr __user *uattr)
1862{
1863 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1864 u32 cp, max_count;
1865 int err = 0;
1866 void *key;
1867
1868 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1869 return -EINVAL;
1870
1871 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1872 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1873 return -EINVAL;
1874 }
1875
1876 max_count = attr->batch.count;
1877 if (!max_count)
1878 return 0;
1879
1880 if (put_user(0, &uattr->batch.count))
1881 return -EFAULT;
1882
1883 key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1884 if (!key)
1885 return -ENOMEM;
1886
1887 for (cp = 0; cp < max_count; cp++) {
1888 err = -EFAULT;
1889 if (copy_from_user(key, keys + cp * map->key_size,
1890 map->key_size))
1891 break;
1892
1893 if (bpf_map_is_offloaded(map)) {
1894 err = bpf_map_offload_delete_elem(map, key);
1895 break;
1896 }
1897
1898 bpf_disable_instrumentation();
1899 rcu_read_lock();
1900 err = map->ops->map_delete_elem(map, key);
1901 rcu_read_unlock();
1902 bpf_enable_instrumentation();
1903 if (err)
1904 break;
1905 cond_resched();
1906 }
1907 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1908 err = -EFAULT;
1909
1910 kvfree(key);
1911
1912 return err;
1913}
1914
1915int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1916 const union bpf_attr *attr,
1917 union bpf_attr __user *uattr)
1918{
1919 void __user *values = u64_to_user_ptr(attr->batch.values);
1920 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1921 u32 value_size, cp, max_count;
1922 void *key, *value;
1923 int err = 0;
1924
1925 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1926 return -EINVAL;
1927
1928 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1929 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1930 return -EINVAL;
1931 }
1932
1933 value_size = bpf_map_value_size(map);
1934
1935 max_count = attr->batch.count;
1936 if (!max_count)
1937 return 0;
1938
1939 if (put_user(0, &uattr->batch.count))
1940 return -EFAULT;
1941
1942 key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1943 if (!key)
1944 return -ENOMEM;
1945
1946 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1947 if (!value) {
1948 kvfree(key);
1949 return -ENOMEM;
1950 }
1951
1952 for (cp = 0; cp < max_count; cp++) {
1953 err = -EFAULT;
1954 if (copy_from_user(key, keys + cp * map->key_size,
1955 map->key_size) ||
1956 copy_from_user(value, values + cp * value_size, value_size))
1957 break;
1958
1959 err = bpf_map_update_value(map, map_file, key, value,
1960 attr->batch.elem_flags);
1961
1962 if (err)
1963 break;
1964 cond_resched();
1965 }
1966
1967 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1968 err = -EFAULT;
1969
1970 kvfree(value);
1971 kvfree(key);
1972
1973 return err;
1974}
1975
1976int generic_map_lookup_batch(struct bpf_map *map,
1977 const union bpf_attr *attr,
1978 union bpf_attr __user *uattr)
1979{
1980 void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1981 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1982 void __user *values = u64_to_user_ptr(attr->batch.values);
1983 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1984 void *buf, *buf_prevkey, *prev_key, *key, *value;
1985 u32 value_size, cp, max_count;
1986 int err;
1987
1988 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1989 return -EINVAL;
1990
1991 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1992 !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1993 return -EINVAL;
1994
1995 value_size = bpf_map_value_size(map);
1996
1997 max_count = attr->batch.count;
1998 if (!max_count)
1999 return 0;
2000
2001 if (put_user(0, &uattr->batch.count))
2002 return -EFAULT;
2003
2004 buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
2005 if (!buf_prevkey)
2006 return -ENOMEM;
2007
2008 buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
2009 if (!buf) {
2010 kvfree(buf_prevkey);
2011 return -ENOMEM;
2012 }
2013
2014 err = -EFAULT;
2015 prev_key = NULL;
2016 if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
2017 goto free_buf;
2018 key = buf;
2019 value = key + map->key_size;
2020 if (ubatch)
2021 prev_key = buf_prevkey;
2022
2023 for (cp = 0; cp < max_count;) {
2024 rcu_read_lock();
2025 err = map->ops->map_get_next_key(map, prev_key, key);
2026 rcu_read_unlock();
2027 if (err)
2028 break;
2029 err = bpf_map_copy_value(map, key, value,
2030 attr->batch.elem_flags);
2031
2032 if (err == -ENOENT)
2033 goto next_key;
2034
2035 if (err)
2036 goto free_buf;
2037
2038 if (copy_to_user(keys + cp * map->key_size, key,
2039 map->key_size)) {
2040 err = -EFAULT;
2041 goto free_buf;
2042 }
2043 if (copy_to_user(values + cp * value_size, value, value_size)) {
2044 err = -EFAULT;
2045 goto free_buf;
2046 }
2047
2048 cp++;
2049next_key:
2050 if (!prev_key)
2051 prev_key = buf_prevkey;
2052
2053 swap(prev_key, key);
2054 cond_resched();
2055 }
2056
2057 if (err == -EFAULT)
2058 goto free_buf;
2059
2060 if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
2061 (cp && copy_to_user(uobatch, prev_key, map->key_size))))
2062 err = -EFAULT;
2063
2064free_buf:
2065 kvfree(buf_prevkey);
2066 kvfree(buf);
2067 return err;
2068}
2069
2070#define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
2071
2072static int map_lookup_and_delete_elem(union bpf_attr *attr)
2073{
2074 void __user *ukey = u64_to_user_ptr(attr->key);
2075 void __user *uvalue = u64_to_user_ptr(attr->value);
2076 struct bpf_map *map;
2077 void *key, *value;
2078 u32 value_size;
2079 int err;
2080
2081 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
2082 return -EINVAL;
2083
2084 if (attr->flags & ~BPF_F_LOCK)
2085 return -EINVAL;
2086
2087 CLASS(fd, f)(attr->map_fd);
2088 map = __bpf_map_get(f);
2089 if (IS_ERR(map))
2090 return PTR_ERR(map);
2091 bpf_map_write_active_inc(map);
2092 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
2093 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
2094 err = -EPERM;
2095 goto err_put;
2096 }
2097
2098 if (attr->flags &&
2099 (map->map_type == BPF_MAP_TYPE_QUEUE ||
2100 map->map_type == BPF_MAP_TYPE_STACK)) {
2101 err = -EINVAL;
2102 goto err_put;
2103 }
2104
2105 if ((attr->flags & BPF_F_LOCK) &&
2106 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
2107 err = -EINVAL;
2108 goto err_put;
2109 }
2110
2111 key = __bpf_copy_key(ukey, map->key_size);
2112 if (IS_ERR(key)) {
2113 err = PTR_ERR(key);
2114 goto err_put;
2115 }
2116
2117 value_size = bpf_map_value_size(map);
2118
2119 err = -ENOMEM;
2120 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
2121 if (!value)
2122 goto free_key;
2123
2124 err = -ENOTSUPP;
2125 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
2126 map->map_type == BPF_MAP_TYPE_STACK) {
2127 err = map->ops->map_pop_elem(map, value);
2128 } else if (map->map_type == BPF_MAP_TYPE_HASH ||
2129 map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
2130 map->map_type == BPF_MAP_TYPE_LRU_HASH ||
2131 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
2132 if (!bpf_map_is_offloaded(map)) {
2133 bpf_disable_instrumentation();
2134 rcu_read_lock();
2135 err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
2136 rcu_read_unlock();
2137 bpf_enable_instrumentation();
2138 }
2139 }
2140
2141 if (err)
2142 goto free_value;
2143
2144 if (copy_to_user(uvalue, value, value_size) != 0) {
2145 err = -EFAULT;
2146 goto free_value;
2147 }
2148
2149 err = 0;
2150
2151free_value:
2152 kvfree(value);
2153free_key:
2154 kvfree(key);
2155err_put:
2156 bpf_map_write_active_dec(map);
2157 return err;
2158}
2159
2160#define BPF_MAP_FREEZE_LAST_FIELD map_fd
2161
2162static int map_freeze(const union bpf_attr *attr)
2163{
2164 int err = 0;
2165 struct bpf_map *map;
2166
2167 if (CHECK_ATTR(BPF_MAP_FREEZE))
2168 return -EINVAL;
2169
2170 CLASS(fd, f)(attr->map_fd);
2171 map = __bpf_map_get(f);
2172 if (IS_ERR(map))
2173 return PTR_ERR(map);
2174
2175 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record))
2176 return -ENOTSUPP;
2177
2178 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE))
2179 return -EPERM;
2180
2181 mutex_lock(&map->freeze_mutex);
2182 if (bpf_map_write_active(map)) {
2183 err = -EBUSY;
2184 goto err_put;
2185 }
2186 if (READ_ONCE(map->frozen)) {
2187 err = -EBUSY;
2188 goto err_put;
2189 }
2190
2191 WRITE_ONCE(map->frozen, true);
2192err_put:
2193 mutex_unlock(&map->freeze_mutex);
2194 return err;
2195}
2196
2197static const struct bpf_prog_ops * const bpf_prog_types[] = {
2198#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
2199 [_id] = & _name ## _prog_ops,
2200#define BPF_MAP_TYPE(_id, _ops)
2201#define BPF_LINK_TYPE(_id, _name)
2202#include <linux/bpf_types.h>
2203#undef BPF_PROG_TYPE
2204#undef BPF_MAP_TYPE
2205#undef BPF_LINK_TYPE
2206};
2207
2208static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
2209{
2210 const struct bpf_prog_ops *ops;
2211
2212 if (type >= ARRAY_SIZE(bpf_prog_types))
2213 return -EINVAL;
2214 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
2215 ops = bpf_prog_types[type];
2216 if (!ops)
2217 return -EINVAL;
2218
2219 if (!bpf_prog_is_offloaded(prog->aux))
2220 prog->aux->ops = ops;
2221 else
2222 prog->aux->ops = &bpf_offload_prog_ops;
2223 prog->type = type;
2224 return 0;
2225}
2226
2227enum bpf_audit {
2228 BPF_AUDIT_LOAD,
2229 BPF_AUDIT_UNLOAD,
2230 BPF_AUDIT_MAX,
2231};
2232
2233static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
2234 [BPF_AUDIT_LOAD] = "LOAD",
2235 [BPF_AUDIT_UNLOAD] = "UNLOAD",
2236};
2237
2238static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
2239{
2240 struct audit_context *ctx = NULL;
2241 struct audit_buffer *ab;
2242
2243 if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
2244 return;
2245 if (audit_enabled == AUDIT_OFF)
2246 return;
2247 if (!in_irq() && !irqs_disabled())
2248 ctx = audit_context();
2249 ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
2250 if (unlikely(!ab))
2251 return;
2252 audit_log_format(ab, "prog-id=%u op=%s",
2253 prog->aux->id, bpf_audit_str[op]);
2254 audit_log_end(ab);
2255}
2256
2257static int bpf_prog_alloc_id(struct bpf_prog *prog)
2258{
2259 int id;
2260
2261 idr_preload(GFP_KERNEL);
2262 spin_lock_bh(&prog_idr_lock);
2263 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2264 if (id > 0)
2265 prog->aux->id = id;
2266 spin_unlock_bh(&prog_idr_lock);
2267 idr_preload_end();
2268
2269 /* id is in [1, INT_MAX) */
2270 if (WARN_ON_ONCE(!id))
2271 return -ENOSPC;
2272
2273 return id > 0 ? 0 : id;
2274}
2275
2276void bpf_prog_free_id(struct bpf_prog *prog)
2277{
2278 unsigned long flags;
2279
2280 /* cBPF to eBPF migrations are currently not in the idr store.
2281 * Offloaded programs are removed from the store when their device
2282 * disappears - even if someone grabs an fd to them they are unusable,
2283 * simply waiting for refcnt to drop to be freed.
2284 */
2285 if (!prog->aux->id)
2286 return;
2287
2288 spin_lock_irqsave(&prog_idr_lock, flags);
2289 idr_remove(&prog_idr, prog->aux->id);
2290 prog->aux->id = 0;
2291 spin_unlock_irqrestore(&prog_idr_lock, flags);
2292}
2293
2294static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2295{
2296 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2297
2298 kvfree(aux->func_info);
2299 kfree(aux->func_info_aux);
2300 free_uid(aux->user);
2301 security_bpf_prog_free(aux->prog);
2302 bpf_prog_free(aux->prog);
2303}
2304
2305static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2306{
2307 bpf_prog_kallsyms_del_all(prog);
2308 btf_put(prog->aux->btf);
2309 module_put(prog->aux->mod);
2310 kvfree(prog->aux->jited_linfo);
2311 kvfree(prog->aux->linfo);
2312 kfree(prog->aux->kfunc_tab);
2313 if (prog->aux->attach_btf)
2314 btf_put(prog->aux->attach_btf);
2315
2316 if (deferred) {
2317 if (prog->sleepable)
2318 call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2319 else
2320 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2321 } else {
2322 __bpf_prog_put_rcu(&prog->aux->rcu);
2323 }
2324}
2325
2326static void bpf_prog_put_deferred(struct work_struct *work)
2327{
2328 struct bpf_prog_aux *aux;
2329 struct bpf_prog *prog;
2330
2331 aux = container_of(work, struct bpf_prog_aux, work);
2332 prog = aux->prog;
2333 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2334 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2335 bpf_prog_free_id(prog);
2336 __bpf_prog_put_noref(prog, true);
2337}
2338
2339static void __bpf_prog_put(struct bpf_prog *prog)
2340{
2341 struct bpf_prog_aux *aux = prog->aux;
2342
2343 if (atomic64_dec_and_test(&aux->refcnt)) {
2344 if (in_irq() || irqs_disabled()) {
2345 INIT_WORK(&aux->work, bpf_prog_put_deferred);
2346 schedule_work(&aux->work);
2347 } else {
2348 bpf_prog_put_deferred(&aux->work);
2349 }
2350 }
2351}
2352
2353void bpf_prog_put(struct bpf_prog *prog)
2354{
2355 __bpf_prog_put(prog);
2356}
2357EXPORT_SYMBOL_GPL(bpf_prog_put);
2358
2359static int bpf_prog_release(struct inode *inode, struct file *filp)
2360{
2361 struct bpf_prog *prog = filp->private_data;
2362
2363 bpf_prog_put(prog);
2364 return 0;
2365}
2366
2367struct bpf_prog_kstats {
2368 u64 nsecs;
2369 u64 cnt;
2370 u64 misses;
2371};
2372
2373void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2374{
2375 struct bpf_prog_stats *stats;
2376 unsigned int flags;
2377
2378 stats = this_cpu_ptr(prog->stats);
2379 flags = u64_stats_update_begin_irqsave(&stats->syncp);
2380 u64_stats_inc(&stats->misses);
2381 u64_stats_update_end_irqrestore(&stats->syncp, flags);
2382}
2383
2384static void bpf_prog_get_stats(const struct bpf_prog *prog,
2385 struct bpf_prog_kstats *stats)
2386{
2387 u64 nsecs = 0, cnt = 0, misses = 0;
2388 int cpu;
2389
2390 for_each_possible_cpu(cpu) {
2391 const struct bpf_prog_stats *st;
2392 unsigned int start;
2393 u64 tnsecs, tcnt, tmisses;
2394
2395 st = per_cpu_ptr(prog->stats, cpu);
2396 do {
2397 start = u64_stats_fetch_begin(&st->syncp);
2398 tnsecs = u64_stats_read(&st->nsecs);
2399 tcnt = u64_stats_read(&st->cnt);
2400 tmisses = u64_stats_read(&st->misses);
2401 } while (u64_stats_fetch_retry(&st->syncp, start));
2402 nsecs += tnsecs;
2403 cnt += tcnt;
2404 misses += tmisses;
2405 }
2406 stats->nsecs = nsecs;
2407 stats->cnt = cnt;
2408 stats->misses = misses;
2409}
2410
2411#ifdef CONFIG_PROC_FS
2412static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2413{
2414 const struct bpf_prog *prog = filp->private_data;
2415 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2416 struct bpf_prog_kstats stats;
2417
2418 bpf_prog_get_stats(prog, &stats);
2419 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2420 seq_printf(m,
2421 "prog_type:\t%u\n"
2422 "prog_jited:\t%u\n"
2423 "prog_tag:\t%s\n"
2424 "memlock:\t%llu\n"
2425 "prog_id:\t%u\n"
2426 "run_time_ns:\t%llu\n"
2427 "run_cnt:\t%llu\n"
2428 "recursion_misses:\t%llu\n"
2429 "verified_insns:\t%u\n",
2430 prog->type,
2431 prog->jited,
2432 prog_tag,
2433 prog->pages * 1ULL << PAGE_SHIFT,
2434 prog->aux->id,
2435 stats.nsecs,
2436 stats.cnt,
2437 stats.misses,
2438 prog->aux->verified_insns);
2439}
2440#endif
2441
2442const struct file_operations bpf_prog_fops = {
2443#ifdef CONFIG_PROC_FS
2444 .show_fdinfo = bpf_prog_show_fdinfo,
2445#endif
2446 .release = bpf_prog_release,
2447 .read = bpf_dummy_read,
2448 .write = bpf_dummy_write,
2449};
2450
2451int bpf_prog_new_fd(struct bpf_prog *prog)
2452{
2453 int ret;
2454
2455 ret = security_bpf_prog(prog);
2456 if (ret < 0)
2457 return ret;
2458
2459 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2460 O_RDWR | O_CLOEXEC);
2461}
2462
2463void bpf_prog_add(struct bpf_prog *prog, int i)
2464{
2465 atomic64_add(i, &prog->aux->refcnt);
2466}
2467EXPORT_SYMBOL_GPL(bpf_prog_add);
2468
2469void bpf_prog_sub(struct bpf_prog *prog, int i)
2470{
2471 /* Only to be used for undoing previous bpf_prog_add() in some
2472 * error path. We still know that another entity in our call
2473 * path holds a reference to the program, thus atomic_sub() can
2474 * be safely used in such cases!
2475 */
2476 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2477}
2478EXPORT_SYMBOL_GPL(bpf_prog_sub);
2479
2480void bpf_prog_inc(struct bpf_prog *prog)
2481{
2482 atomic64_inc(&prog->aux->refcnt);
2483}
2484EXPORT_SYMBOL_GPL(bpf_prog_inc);
2485
2486/* prog_idr_lock should have been held */
2487struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2488{
2489 int refold;
2490
2491 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2492
2493 if (!refold)
2494 return ERR_PTR(-ENOENT);
2495
2496 return prog;
2497}
2498EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2499
2500bool bpf_prog_get_ok(struct bpf_prog *prog,
2501 enum bpf_prog_type *attach_type, bool attach_drv)
2502{
2503 /* not an attachment, just a refcount inc, always allow */
2504 if (!attach_type)
2505 return true;
2506
2507 if (prog->type != *attach_type)
2508 return false;
2509 if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2510 return false;
2511
2512 return true;
2513}
2514
2515static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2516 bool attach_drv)
2517{
2518 CLASS(fd, f)(ufd);
2519 struct bpf_prog *prog;
2520
2521 if (fd_empty(f))
2522 return ERR_PTR(-EBADF);
2523 if (fd_file(f)->f_op != &bpf_prog_fops)
2524 return ERR_PTR(-EINVAL);
2525
2526 prog = fd_file(f)->private_data;
2527 if (!bpf_prog_get_ok(prog, attach_type, attach_drv))
2528 return ERR_PTR(-EINVAL);
2529
2530 bpf_prog_inc(prog);
2531 return prog;
2532}
2533
2534struct bpf_prog *bpf_prog_get(u32 ufd)
2535{
2536 return __bpf_prog_get(ufd, NULL, false);
2537}
2538
2539struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2540 bool attach_drv)
2541{
2542 return __bpf_prog_get(ufd, &type, attach_drv);
2543}
2544EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2545
2546/* Initially all BPF programs could be loaded w/o specifying
2547 * expected_attach_type. Later for some of them specifying expected_attach_type
2548 * at load time became required so that program could be validated properly.
2549 * Programs of types that are allowed to be loaded both w/ and w/o (for
2550 * backward compatibility) expected_attach_type, should have the default attach
2551 * type assigned to expected_attach_type for the latter case, so that it can be
2552 * validated later at attach time.
2553 *
2554 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2555 * prog type requires it but has some attach types that have to be backward
2556 * compatible.
2557 */
2558static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2559{
2560 switch (attr->prog_type) {
2561 case BPF_PROG_TYPE_CGROUP_SOCK:
2562 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2563 * exist so checking for non-zero is the way to go here.
2564 */
2565 if (!attr->expected_attach_type)
2566 attr->expected_attach_type =
2567 BPF_CGROUP_INET_SOCK_CREATE;
2568 break;
2569 case BPF_PROG_TYPE_SK_REUSEPORT:
2570 if (!attr->expected_attach_type)
2571 attr->expected_attach_type =
2572 BPF_SK_REUSEPORT_SELECT;
2573 break;
2574 }
2575}
2576
2577static int
2578bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2579 enum bpf_attach_type expected_attach_type,
2580 struct btf *attach_btf, u32 btf_id,
2581 struct bpf_prog *dst_prog)
2582{
2583 if (btf_id) {
2584 if (btf_id > BTF_MAX_TYPE)
2585 return -EINVAL;
2586
2587 if (!attach_btf && !dst_prog)
2588 return -EINVAL;
2589
2590 switch (prog_type) {
2591 case BPF_PROG_TYPE_TRACING:
2592 case BPF_PROG_TYPE_LSM:
2593 case BPF_PROG_TYPE_STRUCT_OPS:
2594 case BPF_PROG_TYPE_EXT:
2595 break;
2596 default:
2597 return -EINVAL;
2598 }
2599 }
2600
2601 if (attach_btf && (!btf_id || dst_prog))
2602 return -EINVAL;
2603
2604 if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2605 prog_type != BPF_PROG_TYPE_EXT)
2606 return -EINVAL;
2607
2608 switch (prog_type) {
2609 case BPF_PROG_TYPE_CGROUP_SOCK:
2610 switch (expected_attach_type) {
2611 case BPF_CGROUP_INET_SOCK_CREATE:
2612 case BPF_CGROUP_INET_SOCK_RELEASE:
2613 case BPF_CGROUP_INET4_POST_BIND:
2614 case BPF_CGROUP_INET6_POST_BIND:
2615 return 0;
2616 default:
2617 return -EINVAL;
2618 }
2619 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2620 switch (expected_attach_type) {
2621 case BPF_CGROUP_INET4_BIND:
2622 case BPF_CGROUP_INET6_BIND:
2623 case BPF_CGROUP_INET4_CONNECT:
2624 case BPF_CGROUP_INET6_CONNECT:
2625 case BPF_CGROUP_UNIX_CONNECT:
2626 case BPF_CGROUP_INET4_GETPEERNAME:
2627 case BPF_CGROUP_INET6_GETPEERNAME:
2628 case BPF_CGROUP_UNIX_GETPEERNAME:
2629 case BPF_CGROUP_INET4_GETSOCKNAME:
2630 case BPF_CGROUP_INET6_GETSOCKNAME:
2631 case BPF_CGROUP_UNIX_GETSOCKNAME:
2632 case BPF_CGROUP_UDP4_SENDMSG:
2633 case BPF_CGROUP_UDP6_SENDMSG:
2634 case BPF_CGROUP_UNIX_SENDMSG:
2635 case BPF_CGROUP_UDP4_RECVMSG:
2636 case BPF_CGROUP_UDP6_RECVMSG:
2637 case BPF_CGROUP_UNIX_RECVMSG:
2638 return 0;
2639 default:
2640 return -EINVAL;
2641 }
2642 case BPF_PROG_TYPE_CGROUP_SKB:
2643 switch (expected_attach_type) {
2644 case BPF_CGROUP_INET_INGRESS:
2645 case BPF_CGROUP_INET_EGRESS:
2646 return 0;
2647 default:
2648 return -EINVAL;
2649 }
2650 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2651 switch (expected_attach_type) {
2652 case BPF_CGROUP_SETSOCKOPT:
2653 case BPF_CGROUP_GETSOCKOPT:
2654 return 0;
2655 default:
2656 return -EINVAL;
2657 }
2658 case BPF_PROG_TYPE_SK_LOOKUP:
2659 if (expected_attach_type == BPF_SK_LOOKUP)
2660 return 0;
2661 return -EINVAL;
2662 case BPF_PROG_TYPE_SK_REUSEPORT:
2663 switch (expected_attach_type) {
2664 case BPF_SK_REUSEPORT_SELECT:
2665 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2666 return 0;
2667 default:
2668 return -EINVAL;
2669 }
2670 case BPF_PROG_TYPE_NETFILTER:
2671 if (expected_attach_type == BPF_NETFILTER)
2672 return 0;
2673 return -EINVAL;
2674 case BPF_PROG_TYPE_SYSCALL:
2675 case BPF_PROG_TYPE_EXT:
2676 if (expected_attach_type)
2677 return -EINVAL;
2678 fallthrough;
2679 default:
2680 return 0;
2681 }
2682}
2683
2684static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2685{
2686 switch (prog_type) {
2687 case BPF_PROG_TYPE_SCHED_CLS:
2688 case BPF_PROG_TYPE_SCHED_ACT:
2689 case BPF_PROG_TYPE_XDP:
2690 case BPF_PROG_TYPE_LWT_IN:
2691 case BPF_PROG_TYPE_LWT_OUT:
2692 case BPF_PROG_TYPE_LWT_XMIT:
2693 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2694 case BPF_PROG_TYPE_SK_SKB:
2695 case BPF_PROG_TYPE_SK_MSG:
2696 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2697 case BPF_PROG_TYPE_CGROUP_DEVICE:
2698 case BPF_PROG_TYPE_CGROUP_SOCK:
2699 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2700 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2701 case BPF_PROG_TYPE_CGROUP_SYSCTL:
2702 case BPF_PROG_TYPE_SOCK_OPS:
2703 case BPF_PROG_TYPE_EXT: /* extends any prog */
2704 case BPF_PROG_TYPE_NETFILTER:
2705 return true;
2706 case BPF_PROG_TYPE_CGROUP_SKB:
2707 /* always unpriv */
2708 case BPF_PROG_TYPE_SK_REUSEPORT:
2709 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2710 default:
2711 return false;
2712 }
2713}
2714
2715static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2716{
2717 switch (prog_type) {
2718 case BPF_PROG_TYPE_KPROBE:
2719 case BPF_PROG_TYPE_TRACEPOINT:
2720 case BPF_PROG_TYPE_PERF_EVENT:
2721 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2722 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2723 case BPF_PROG_TYPE_TRACING:
2724 case BPF_PROG_TYPE_LSM:
2725 case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2726 case BPF_PROG_TYPE_EXT: /* extends any prog */
2727 return true;
2728 default:
2729 return false;
2730 }
2731}
2732
2733/* last field in 'union bpf_attr' used by this command */
2734#define BPF_PROG_LOAD_LAST_FIELD prog_token_fd
2735
2736static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size)
2737{
2738 enum bpf_prog_type type = attr->prog_type;
2739 struct bpf_prog *prog, *dst_prog = NULL;
2740 struct btf *attach_btf = NULL;
2741 struct bpf_token *token = NULL;
2742 bool bpf_cap;
2743 int err;
2744 char license[128];
2745
2746 if (CHECK_ATTR(BPF_PROG_LOAD))
2747 return -EINVAL;
2748
2749 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2750 BPF_F_ANY_ALIGNMENT |
2751 BPF_F_TEST_STATE_FREQ |
2752 BPF_F_SLEEPABLE |
2753 BPF_F_TEST_RND_HI32 |
2754 BPF_F_XDP_HAS_FRAGS |
2755 BPF_F_XDP_DEV_BOUND_ONLY |
2756 BPF_F_TEST_REG_INVARIANTS |
2757 BPF_F_TOKEN_FD))
2758 return -EINVAL;
2759
2760 bpf_prog_load_fixup_attach_type(attr);
2761
2762 if (attr->prog_flags & BPF_F_TOKEN_FD) {
2763 token = bpf_token_get_from_fd(attr->prog_token_fd);
2764 if (IS_ERR(token))
2765 return PTR_ERR(token);
2766 /* if current token doesn't grant prog loading permissions,
2767 * then we can't use this token, so ignore it and rely on
2768 * system-wide capabilities checks
2769 */
2770 if (!bpf_token_allow_cmd(token, BPF_PROG_LOAD) ||
2771 !bpf_token_allow_prog_type(token, attr->prog_type,
2772 attr->expected_attach_type)) {
2773 bpf_token_put(token);
2774 token = NULL;
2775 }
2776 }
2777
2778 bpf_cap = bpf_token_capable(token, CAP_BPF);
2779 err = -EPERM;
2780
2781 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2782 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2783 !bpf_cap)
2784 goto put_token;
2785
2786 /* Intent here is for unprivileged_bpf_disabled to block BPF program
2787 * creation for unprivileged users; other actions depend
2788 * on fd availability and access to bpffs, so are dependent on
2789 * object creation success. Even with unprivileged BPF disabled,
2790 * capability checks are still carried out for these
2791 * and other operations.
2792 */
2793 if (sysctl_unprivileged_bpf_disabled && !bpf_cap)
2794 goto put_token;
2795
2796 if (attr->insn_cnt == 0 ||
2797 attr->insn_cnt > (bpf_cap ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS)) {
2798 err = -E2BIG;
2799 goto put_token;
2800 }
2801 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2802 type != BPF_PROG_TYPE_CGROUP_SKB &&
2803 !bpf_cap)
2804 goto put_token;
2805
2806 if (is_net_admin_prog_type(type) && !bpf_token_capable(token, CAP_NET_ADMIN))
2807 goto put_token;
2808 if (is_perfmon_prog_type(type) && !bpf_token_capable(token, CAP_PERFMON))
2809 goto put_token;
2810
2811 /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2812 * or btf, we need to check which one it is
2813 */
2814 if (attr->attach_prog_fd) {
2815 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2816 if (IS_ERR(dst_prog)) {
2817 dst_prog = NULL;
2818 attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2819 if (IS_ERR(attach_btf)) {
2820 err = -EINVAL;
2821 goto put_token;
2822 }
2823 if (!btf_is_kernel(attach_btf)) {
2824 /* attaching through specifying bpf_prog's BTF
2825 * objects directly might be supported eventually
2826 */
2827 btf_put(attach_btf);
2828 err = -ENOTSUPP;
2829 goto put_token;
2830 }
2831 }
2832 } else if (attr->attach_btf_id) {
2833 /* fall back to vmlinux BTF, if BTF type ID is specified */
2834 attach_btf = bpf_get_btf_vmlinux();
2835 if (IS_ERR(attach_btf)) {
2836 err = PTR_ERR(attach_btf);
2837 goto put_token;
2838 }
2839 if (!attach_btf) {
2840 err = -EINVAL;
2841 goto put_token;
2842 }
2843 btf_get(attach_btf);
2844 }
2845
2846 if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2847 attach_btf, attr->attach_btf_id,
2848 dst_prog)) {
2849 if (dst_prog)
2850 bpf_prog_put(dst_prog);
2851 if (attach_btf)
2852 btf_put(attach_btf);
2853 err = -EINVAL;
2854 goto put_token;
2855 }
2856
2857 /* plain bpf_prog allocation */
2858 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2859 if (!prog) {
2860 if (dst_prog)
2861 bpf_prog_put(dst_prog);
2862 if (attach_btf)
2863 btf_put(attach_btf);
2864 err = -EINVAL;
2865 goto put_token;
2866 }
2867
2868 prog->expected_attach_type = attr->expected_attach_type;
2869 prog->sleepable = !!(attr->prog_flags & BPF_F_SLEEPABLE);
2870 prog->aux->attach_btf = attach_btf;
2871 prog->aux->attach_btf_id = attr->attach_btf_id;
2872 prog->aux->dst_prog = dst_prog;
2873 prog->aux->dev_bound = !!attr->prog_ifindex;
2874 prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2875
2876 /* move token into prog->aux, reuse taken refcnt */
2877 prog->aux->token = token;
2878 token = NULL;
2879
2880 prog->aux->user = get_current_user();
2881 prog->len = attr->insn_cnt;
2882
2883 err = -EFAULT;
2884 if (copy_from_bpfptr(prog->insns,
2885 make_bpfptr(attr->insns, uattr.is_kernel),
2886 bpf_prog_insn_size(prog)) != 0)
2887 goto free_prog;
2888 /* copy eBPF program license from user space */
2889 if (strncpy_from_bpfptr(license,
2890 make_bpfptr(attr->license, uattr.is_kernel),
2891 sizeof(license) - 1) < 0)
2892 goto free_prog;
2893 license[sizeof(license) - 1] = 0;
2894
2895 /* eBPF programs must be GPL compatible to use GPL-ed functions */
2896 prog->gpl_compatible = license_is_gpl_compatible(license) ? 1 : 0;
2897
2898 prog->orig_prog = NULL;
2899 prog->jited = 0;
2900
2901 atomic64_set(&prog->aux->refcnt, 1);
2902
2903 if (bpf_prog_is_dev_bound(prog->aux)) {
2904 err = bpf_prog_dev_bound_init(prog, attr);
2905 if (err)
2906 goto free_prog;
2907 }
2908
2909 if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2910 bpf_prog_is_dev_bound(dst_prog->aux)) {
2911 err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2912 if (err)
2913 goto free_prog;
2914 }
2915
2916 /*
2917 * Bookkeeping for managing the program attachment chain.
2918 *
2919 * It might be tempting to set attach_tracing_prog flag at the attachment
2920 * time, but this will not prevent from loading bunch of tracing prog
2921 * first, then attach them one to another.
2922 *
2923 * The flag attach_tracing_prog is set for the whole program lifecycle, and
2924 * doesn't have to be cleared in bpf_tracing_link_release, since tracing
2925 * programs cannot change attachment target.
2926 */
2927 if (type == BPF_PROG_TYPE_TRACING && dst_prog &&
2928 dst_prog->type == BPF_PROG_TYPE_TRACING) {
2929 prog->aux->attach_tracing_prog = true;
2930 }
2931
2932 /* find program type: socket_filter vs tracing_filter */
2933 err = find_prog_type(type, prog);
2934 if (err < 0)
2935 goto free_prog;
2936
2937 prog->aux->load_time = ktime_get_boottime_ns();
2938 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2939 sizeof(attr->prog_name));
2940 if (err < 0)
2941 goto free_prog;
2942
2943 err = security_bpf_prog_load(prog, attr, token);
2944 if (err)
2945 goto free_prog_sec;
2946
2947 /* run eBPF verifier */
2948 err = bpf_check(&prog, attr, uattr, uattr_size);
2949 if (err < 0)
2950 goto free_used_maps;
2951
2952 prog = bpf_prog_select_runtime(prog, &err);
2953 if (err < 0)
2954 goto free_used_maps;
2955
2956 err = bpf_prog_alloc_id(prog);
2957 if (err)
2958 goto free_used_maps;
2959
2960 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2961 * effectively publicly exposed. However, retrieving via
2962 * bpf_prog_get_fd_by_id() will take another reference,
2963 * therefore it cannot be gone underneath us.
2964 *
2965 * Only for the time /after/ successful bpf_prog_new_fd()
2966 * and before returning to userspace, we might just hold
2967 * one reference and any parallel close on that fd could
2968 * rip everything out. Hence, below notifications must
2969 * happen before bpf_prog_new_fd().
2970 *
2971 * Also, any failure handling from this point onwards must
2972 * be using bpf_prog_put() given the program is exposed.
2973 */
2974 bpf_prog_kallsyms_add(prog);
2975 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2976 bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2977
2978 err = bpf_prog_new_fd(prog);
2979 if (err < 0)
2980 bpf_prog_put(prog);
2981 return err;
2982
2983free_used_maps:
2984 /* In case we have subprogs, we need to wait for a grace
2985 * period before we can tear down JIT memory since symbols
2986 * are already exposed under kallsyms.
2987 */
2988 __bpf_prog_put_noref(prog, prog->aux->real_func_cnt);
2989 return err;
2990
2991free_prog_sec:
2992 security_bpf_prog_free(prog);
2993free_prog:
2994 free_uid(prog->aux->user);
2995 if (prog->aux->attach_btf)
2996 btf_put(prog->aux->attach_btf);
2997 bpf_prog_free(prog);
2998put_token:
2999 bpf_token_put(token);
3000 return err;
3001}
3002
3003#define BPF_OBJ_LAST_FIELD path_fd
3004
3005static int bpf_obj_pin(const union bpf_attr *attr)
3006{
3007 int path_fd;
3008
3009 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags & ~BPF_F_PATH_FD)
3010 return -EINVAL;
3011
3012 /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
3013 if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
3014 return -EINVAL;
3015
3016 path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
3017 return bpf_obj_pin_user(attr->bpf_fd, path_fd,
3018 u64_to_user_ptr(attr->pathname));
3019}
3020
3021static int bpf_obj_get(const union bpf_attr *attr)
3022{
3023 int path_fd;
3024
3025 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
3026 attr->file_flags & ~(BPF_OBJ_FLAG_MASK | BPF_F_PATH_FD))
3027 return -EINVAL;
3028
3029 /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
3030 if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
3031 return -EINVAL;
3032
3033 path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
3034 return bpf_obj_get_user(path_fd, u64_to_user_ptr(attr->pathname),
3035 attr->file_flags);
3036}
3037
3038/* bpf_link_init_sleepable() allows to specify whether BPF link itself has
3039 * "sleepable" semantics, which normally would mean that BPF link's attach
3040 * hook can dereference link or link's underlying program for some time after
3041 * detachment due to RCU Tasks Trace-based lifetime protection scheme.
3042 * BPF program itself can be non-sleepable, yet, because it's transitively
3043 * reachable through BPF link, its freeing has to be delayed until after RCU
3044 * Tasks Trace GP.
3045 */
3046void bpf_link_init_sleepable(struct bpf_link *link, enum bpf_link_type type,
3047 const struct bpf_link_ops *ops, struct bpf_prog *prog,
3048 bool sleepable)
3049{
3050 WARN_ON(ops->dealloc && ops->dealloc_deferred);
3051 atomic64_set(&link->refcnt, 1);
3052 link->type = type;
3053 link->sleepable = sleepable;
3054 link->id = 0;
3055 link->ops = ops;
3056 link->prog = prog;
3057}
3058
3059void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
3060 const struct bpf_link_ops *ops, struct bpf_prog *prog)
3061{
3062 bpf_link_init_sleepable(link, type, ops, prog, false);
3063}
3064
3065static void bpf_link_free_id(int id)
3066{
3067 if (!id)
3068 return;
3069
3070 spin_lock_bh(&link_idr_lock);
3071 idr_remove(&link_idr, id);
3072 spin_unlock_bh(&link_idr_lock);
3073}
3074
3075/* Clean up bpf_link and corresponding anon_inode file and FD. After
3076 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
3077 * anon_inode's release() call. This helper marks bpf_link as
3078 * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
3079 * is not decremented, it's the responsibility of a calling code that failed
3080 * to complete bpf_link initialization.
3081 * This helper eventually calls link's dealloc callback, but does not call
3082 * link's release callback.
3083 */
3084void bpf_link_cleanup(struct bpf_link_primer *primer)
3085{
3086 primer->link->prog = NULL;
3087 bpf_link_free_id(primer->id);
3088 fput(primer->file);
3089 put_unused_fd(primer->fd);
3090}
3091
3092void bpf_link_inc(struct bpf_link *link)
3093{
3094 atomic64_inc(&link->refcnt);
3095}
3096
3097static void bpf_link_dealloc(struct bpf_link *link)
3098{
3099 /* now that we know that bpf_link itself can't be reached, put underlying BPF program */
3100 if (link->prog)
3101 bpf_prog_put(link->prog);
3102
3103 /* free bpf_link and its containing memory */
3104 if (link->ops->dealloc_deferred)
3105 link->ops->dealloc_deferred(link);
3106 else
3107 link->ops->dealloc(link);
3108}
3109
3110static void bpf_link_defer_dealloc_rcu_gp(struct rcu_head *rcu)
3111{
3112 struct bpf_link *link = container_of(rcu, struct bpf_link, rcu);
3113
3114 bpf_link_dealloc(link);
3115}
3116
3117static void bpf_link_defer_dealloc_mult_rcu_gp(struct rcu_head *rcu)
3118{
3119 if (rcu_trace_implies_rcu_gp())
3120 bpf_link_defer_dealloc_rcu_gp(rcu);
3121 else
3122 call_rcu(rcu, bpf_link_defer_dealloc_rcu_gp);
3123}
3124
3125/* bpf_link_free is guaranteed to be called from process context */
3126static void bpf_link_free(struct bpf_link *link)
3127{
3128 const struct bpf_link_ops *ops = link->ops;
3129
3130 bpf_link_free_id(link->id);
3131 /* detach BPF program, clean up used resources */
3132 if (link->prog)
3133 ops->release(link);
3134 if (ops->dealloc_deferred) {
3135 /* Schedule BPF link deallocation, which will only then
3136 * trigger putting BPF program refcount.
3137 * If underlying BPF program is sleepable or BPF link's target
3138 * attach hookpoint is sleepable or otherwise requires RCU GPs
3139 * to ensure link and its underlying BPF program is not
3140 * reachable anymore, we need to first wait for RCU tasks
3141 * trace sync, and then go through "classic" RCU grace period
3142 */
3143 if (link->sleepable || (link->prog && link->prog->sleepable))
3144 call_rcu_tasks_trace(&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp);
3145 else
3146 call_rcu(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
3147 } else if (ops->dealloc) {
3148 bpf_link_dealloc(link);
3149 }
3150}
3151
3152static void bpf_link_put_deferred(struct work_struct *work)
3153{
3154 struct bpf_link *link = container_of(work, struct bpf_link, work);
3155
3156 bpf_link_free(link);
3157}
3158
3159/* bpf_link_put might be called from atomic context. It needs to be called
3160 * from sleepable context in order to acquire sleeping locks during the process.
3161 */
3162void bpf_link_put(struct bpf_link *link)
3163{
3164 if (!atomic64_dec_and_test(&link->refcnt))
3165 return;
3166
3167 INIT_WORK(&link->work, bpf_link_put_deferred);
3168 schedule_work(&link->work);
3169}
3170EXPORT_SYMBOL(bpf_link_put);
3171
3172static void bpf_link_put_direct(struct bpf_link *link)
3173{
3174 if (!atomic64_dec_and_test(&link->refcnt))
3175 return;
3176 bpf_link_free(link);
3177}
3178
3179static int bpf_link_release(struct inode *inode, struct file *filp)
3180{
3181 struct bpf_link *link = filp->private_data;
3182
3183 bpf_link_put_direct(link);
3184 return 0;
3185}
3186
3187#ifdef CONFIG_PROC_FS
3188#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
3189#define BPF_MAP_TYPE(_id, _ops)
3190#define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
3191static const char *bpf_link_type_strs[] = {
3192 [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
3193#include <linux/bpf_types.h>
3194};
3195#undef BPF_PROG_TYPE
3196#undef BPF_MAP_TYPE
3197#undef BPF_LINK_TYPE
3198
3199static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
3200{
3201 const struct bpf_link *link = filp->private_data;
3202 const struct bpf_prog *prog = link->prog;
3203 enum bpf_link_type type = link->type;
3204 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
3205
3206 if (type < ARRAY_SIZE(bpf_link_type_strs) && bpf_link_type_strs[type]) {
3207 seq_printf(m, "link_type:\t%s\n", bpf_link_type_strs[type]);
3208 } else {
3209 WARN_ONCE(1, "missing BPF_LINK_TYPE(...) for link type %u\n", type);
3210 seq_printf(m, "link_type:\t<%u>\n", type);
3211 }
3212 seq_printf(m, "link_id:\t%u\n", link->id);
3213
3214 if (prog) {
3215 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
3216 seq_printf(m,
3217 "prog_tag:\t%s\n"
3218 "prog_id:\t%u\n",
3219 prog_tag,
3220 prog->aux->id);
3221 }
3222 if (link->ops->show_fdinfo)
3223 link->ops->show_fdinfo(link, m);
3224}
3225#endif
3226
3227static __poll_t bpf_link_poll(struct file *file, struct poll_table_struct *pts)
3228{
3229 struct bpf_link *link = file->private_data;
3230
3231 return link->ops->poll(file, pts);
3232}
3233
3234static const struct file_operations bpf_link_fops = {
3235#ifdef CONFIG_PROC_FS
3236 .show_fdinfo = bpf_link_show_fdinfo,
3237#endif
3238 .release = bpf_link_release,
3239 .read = bpf_dummy_read,
3240 .write = bpf_dummy_write,
3241};
3242
3243static const struct file_operations bpf_link_fops_poll = {
3244#ifdef CONFIG_PROC_FS
3245 .show_fdinfo = bpf_link_show_fdinfo,
3246#endif
3247 .release = bpf_link_release,
3248 .read = bpf_dummy_read,
3249 .write = bpf_dummy_write,
3250 .poll = bpf_link_poll,
3251};
3252
3253static int bpf_link_alloc_id(struct bpf_link *link)
3254{
3255 int id;
3256
3257 idr_preload(GFP_KERNEL);
3258 spin_lock_bh(&link_idr_lock);
3259 id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
3260 spin_unlock_bh(&link_idr_lock);
3261 idr_preload_end();
3262
3263 return id;
3264}
3265
3266/* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
3267 * reserving unused FD and allocating ID from link_idr. This is to be paired
3268 * with bpf_link_settle() to install FD and ID and expose bpf_link to
3269 * user-space, if bpf_link is successfully attached. If not, bpf_link and
3270 * pre-allocated resources are to be freed with bpf_cleanup() call. All the
3271 * transient state is passed around in struct bpf_link_primer.
3272 * This is preferred way to create and initialize bpf_link, especially when
3273 * there are complicated and expensive operations in between creating bpf_link
3274 * itself and attaching it to BPF hook. By using bpf_link_prime() and
3275 * bpf_link_settle() kernel code using bpf_link doesn't have to perform
3276 * expensive (and potentially failing) roll back operations in a rare case
3277 * that file, FD, or ID can't be allocated.
3278 */
3279int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
3280{
3281 struct file *file;
3282 int fd, id;
3283
3284 fd = get_unused_fd_flags(O_CLOEXEC);
3285 if (fd < 0)
3286 return fd;
3287
3288
3289 id = bpf_link_alloc_id(link);
3290 if (id < 0) {
3291 put_unused_fd(fd);
3292 return id;
3293 }
3294
3295 file = anon_inode_getfile("bpf_link",
3296 link->ops->poll ? &bpf_link_fops_poll : &bpf_link_fops,
3297 link, O_CLOEXEC);
3298 if (IS_ERR(file)) {
3299 bpf_link_free_id(id);
3300 put_unused_fd(fd);
3301 return PTR_ERR(file);
3302 }
3303
3304 primer->link = link;
3305 primer->file = file;
3306 primer->fd = fd;
3307 primer->id = id;
3308 return 0;
3309}
3310
3311int bpf_link_settle(struct bpf_link_primer *primer)
3312{
3313 /* make bpf_link fetchable by ID */
3314 spin_lock_bh(&link_idr_lock);
3315 primer->link->id = primer->id;
3316 spin_unlock_bh(&link_idr_lock);
3317 /* make bpf_link fetchable by FD */
3318 fd_install(primer->fd, primer->file);
3319 /* pass through installed FD */
3320 return primer->fd;
3321}
3322
3323int bpf_link_new_fd(struct bpf_link *link)
3324{
3325 return anon_inode_getfd("bpf-link",
3326 link->ops->poll ? &bpf_link_fops_poll : &bpf_link_fops,
3327 link, O_CLOEXEC);
3328}
3329
3330struct bpf_link *bpf_link_get_from_fd(u32 ufd)
3331{
3332 CLASS(fd, f)(ufd);
3333 struct bpf_link *link;
3334
3335 if (fd_empty(f))
3336 return ERR_PTR(-EBADF);
3337 if (fd_file(f)->f_op != &bpf_link_fops && fd_file(f)->f_op != &bpf_link_fops_poll)
3338 return ERR_PTR(-EINVAL);
3339
3340 link = fd_file(f)->private_data;
3341 bpf_link_inc(link);
3342 return link;
3343}
3344EXPORT_SYMBOL(bpf_link_get_from_fd);
3345
3346static void bpf_tracing_link_release(struct bpf_link *link)
3347{
3348 struct bpf_tracing_link *tr_link =
3349 container_of(link, struct bpf_tracing_link, link.link);
3350
3351 WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
3352 tr_link->trampoline,
3353 tr_link->tgt_prog));
3354
3355 bpf_trampoline_put(tr_link->trampoline);
3356
3357 /* tgt_prog is NULL if target is a kernel function */
3358 if (tr_link->tgt_prog)
3359 bpf_prog_put(tr_link->tgt_prog);
3360}
3361
3362static void bpf_tracing_link_dealloc(struct bpf_link *link)
3363{
3364 struct bpf_tracing_link *tr_link =
3365 container_of(link, struct bpf_tracing_link, link.link);
3366
3367 kfree(tr_link);
3368}
3369
3370static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
3371 struct seq_file *seq)
3372{
3373 struct bpf_tracing_link *tr_link =
3374 container_of(link, struct bpf_tracing_link, link.link);
3375 u32 target_btf_id, target_obj_id;
3376
3377 bpf_trampoline_unpack_key(tr_link->trampoline->key,
3378 &target_obj_id, &target_btf_id);
3379 seq_printf(seq,
3380 "attach_type:\t%d\n"
3381 "target_obj_id:\t%u\n"
3382 "target_btf_id:\t%u\n",
3383 tr_link->attach_type,
3384 target_obj_id,
3385 target_btf_id);
3386}
3387
3388static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
3389 struct bpf_link_info *info)
3390{
3391 struct bpf_tracing_link *tr_link =
3392 container_of(link, struct bpf_tracing_link, link.link);
3393
3394 info->tracing.attach_type = tr_link->attach_type;
3395 bpf_trampoline_unpack_key(tr_link->trampoline->key,
3396 &info->tracing.target_obj_id,
3397 &info->tracing.target_btf_id);
3398
3399 return 0;
3400}
3401
3402static const struct bpf_link_ops bpf_tracing_link_lops = {
3403 .release = bpf_tracing_link_release,
3404 .dealloc = bpf_tracing_link_dealloc,
3405 .show_fdinfo = bpf_tracing_link_show_fdinfo,
3406 .fill_link_info = bpf_tracing_link_fill_link_info,
3407};
3408
3409static int bpf_tracing_prog_attach(struct bpf_prog *prog,
3410 int tgt_prog_fd,
3411 u32 btf_id,
3412 u64 bpf_cookie)
3413{
3414 struct bpf_link_primer link_primer;
3415 struct bpf_prog *tgt_prog = NULL;
3416 struct bpf_trampoline *tr = NULL;
3417 struct bpf_tracing_link *link;
3418 u64 key = 0;
3419 int err;
3420
3421 switch (prog->type) {
3422 case BPF_PROG_TYPE_TRACING:
3423 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
3424 prog->expected_attach_type != BPF_TRACE_FEXIT &&
3425 prog->expected_attach_type != BPF_MODIFY_RETURN) {
3426 err = -EINVAL;
3427 goto out_put_prog;
3428 }
3429 break;
3430 case BPF_PROG_TYPE_EXT:
3431 if (prog->expected_attach_type != 0) {
3432 err = -EINVAL;
3433 goto out_put_prog;
3434 }
3435 break;
3436 case BPF_PROG_TYPE_LSM:
3437 if (prog->expected_attach_type != BPF_LSM_MAC) {
3438 err = -EINVAL;
3439 goto out_put_prog;
3440 }
3441 break;
3442 default:
3443 err = -EINVAL;
3444 goto out_put_prog;
3445 }
3446
3447 if (!!tgt_prog_fd != !!btf_id) {
3448 err = -EINVAL;
3449 goto out_put_prog;
3450 }
3451
3452 if (tgt_prog_fd) {
3453 /*
3454 * For now we only allow new targets for BPF_PROG_TYPE_EXT. If this
3455 * part would be changed to implement the same for
3456 * BPF_PROG_TYPE_TRACING, do not forget to update the way how
3457 * attach_tracing_prog flag is set.
3458 */
3459 if (prog->type != BPF_PROG_TYPE_EXT) {
3460 err = -EINVAL;
3461 goto out_put_prog;
3462 }
3463
3464 tgt_prog = bpf_prog_get(tgt_prog_fd);
3465 if (IS_ERR(tgt_prog)) {
3466 err = PTR_ERR(tgt_prog);
3467 tgt_prog = NULL;
3468 goto out_put_prog;
3469 }
3470
3471 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3472 }
3473
3474 link = kzalloc(sizeof(*link), GFP_USER);
3475 if (!link) {
3476 err = -ENOMEM;
3477 goto out_put_prog;
3478 }
3479 bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3480 &bpf_tracing_link_lops, prog);
3481 link->attach_type = prog->expected_attach_type;
3482 link->link.cookie = bpf_cookie;
3483
3484 mutex_lock(&prog->aux->dst_mutex);
3485
3486 /* There are a few possible cases here:
3487 *
3488 * - if prog->aux->dst_trampoline is set, the program was just loaded
3489 * and not yet attached to anything, so we can use the values stored
3490 * in prog->aux
3491 *
3492 * - if prog->aux->dst_trampoline is NULL, the program has already been
3493 * attached to a target and its initial target was cleared (below)
3494 *
3495 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3496 * target_btf_id using the link_create API.
3497 *
3498 * - if tgt_prog == NULL when this function was called using the old
3499 * raw_tracepoint_open API, and we need a target from prog->aux
3500 *
3501 * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3502 * was detached and is going for re-attachment.
3503 *
3504 * - if prog->aux->dst_trampoline is NULL and tgt_prog and prog->aux->attach_btf
3505 * are NULL, then program was already attached and user did not provide
3506 * tgt_prog_fd so we have no way to find out or create trampoline
3507 */
3508 if (!prog->aux->dst_trampoline && !tgt_prog) {
3509 /*
3510 * Allow re-attach for TRACING and LSM programs. If it's
3511 * currently linked, bpf_trampoline_link_prog will fail.
3512 * EXT programs need to specify tgt_prog_fd, so they
3513 * re-attach in separate code path.
3514 */
3515 if (prog->type != BPF_PROG_TYPE_TRACING &&
3516 prog->type != BPF_PROG_TYPE_LSM) {
3517 err = -EINVAL;
3518 goto out_unlock;
3519 }
3520 /* We can allow re-attach only if we have valid attach_btf. */
3521 if (!prog->aux->attach_btf) {
3522 err = -EINVAL;
3523 goto out_unlock;
3524 }
3525 btf_id = prog->aux->attach_btf_id;
3526 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3527 }
3528
3529 if (!prog->aux->dst_trampoline ||
3530 (key && key != prog->aux->dst_trampoline->key)) {
3531 /* If there is no saved target, or the specified target is
3532 * different from the destination specified at load time, we
3533 * need a new trampoline and a check for compatibility
3534 */
3535 struct bpf_attach_target_info tgt_info = {};
3536
3537 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3538 &tgt_info);
3539 if (err)
3540 goto out_unlock;
3541
3542 if (tgt_info.tgt_mod) {
3543 module_put(prog->aux->mod);
3544 prog->aux->mod = tgt_info.tgt_mod;
3545 }
3546
3547 tr = bpf_trampoline_get(key, &tgt_info);
3548 if (!tr) {
3549 err = -ENOMEM;
3550 goto out_unlock;
3551 }
3552 } else {
3553 /* The caller didn't specify a target, or the target was the
3554 * same as the destination supplied during program load. This
3555 * means we can reuse the trampoline and reference from program
3556 * load time, and there is no need to allocate a new one. This
3557 * can only happen once for any program, as the saved values in
3558 * prog->aux are cleared below.
3559 */
3560 tr = prog->aux->dst_trampoline;
3561 tgt_prog = prog->aux->dst_prog;
3562 }
3563
3564 err = bpf_link_prime(&link->link.link, &link_primer);
3565 if (err)
3566 goto out_unlock;
3567
3568 err = bpf_trampoline_link_prog(&link->link, tr, tgt_prog);
3569 if (err) {
3570 bpf_link_cleanup(&link_primer);
3571 link = NULL;
3572 goto out_unlock;
3573 }
3574
3575 link->tgt_prog = tgt_prog;
3576 link->trampoline = tr;
3577
3578 /* Always clear the trampoline and target prog from prog->aux to make
3579 * sure the original attach destination is not kept alive after a
3580 * program is (re-)attached to another target.
3581 */
3582 if (prog->aux->dst_prog &&
3583 (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3584 /* got extra prog ref from syscall, or attaching to different prog */
3585 bpf_prog_put(prog->aux->dst_prog);
3586 if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3587 /* we allocated a new trampoline, so free the old one */
3588 bpf_trampoline_put(prog->aux->dst_trampoline);
3589
3590 prog->aux->dst_prog = NULL;
3591 prog->aux->dst_trampoline = NULL;
3592 mutex_unlock(&prog->aux->dst_mutex);
3593
3594 return bpf_link_settle(&link_primer);
3595out_unlock:
3596 if (tr && tr != prog->aux->dst_trampoline)
3597 bpf_trampoline_put(tr);
3598 mutex_unlock(&prog->aux->dst_mutex);
3599 kfree(link);
3600out_put_prog:
3601 if (tgt_prog_fd && tgt_prog)
3602 bpf_prog_put(tgt_prog);
3603 return err;
3604}
3605
3606static void bpf_raw_tp_link_release(struct bpf_link *link)
3607{
3608 struct bpf_raw_tp_link *raw_tp =
3609 container_of(link, struct bpf_raw_tp_link, link);
3610
3611 bpf_probe_unregister(raw_tp->btp, raw_tp);
3612 bpf_put_raw_tracepoint(raw_tp->btp);
3613}
3614
3615static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3616{
3617 struct bpf_raw_tp_link *raw_tp =
3618 container_of(link, struct bpf_raw_tp_link, link);
3619
3620 kfree(raw_tp);
3621}
3622
3623static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3624 struct seq_file *seq)
3625{
3626 struct bpf_raw_tp_link *raw_tp_link =
3627 container_of(link, struct bpf_raw_tp_link, link);
3628
3629 seq_printf(seq,
3630 "tp_name:\t%s\n",
3631 raw_tp_link->btp->tp->name);
3632}
3633
3634static int bpf_copy_to_user(char __user *ubuf, const char *buf, u32 ulen,
3635 u32 len)
3636{
3637 if (ulen >= len + 1) {
3638 if (copy_to_user(ubuf, buf, len + 1))
3639 return -EFAULT;
3640 } else {
3641 char zero = '\0';
3642
3643 if (copy_to_user(ubuf, buf, ulen - 1))
3644 return -EFAULT;
3645 if (put_user(zero, ubuf + ulen - 1))
3646 return -EFAULT;
3647 return -ENOSPC;
3648 }
3649
3650 return 0;
3651}
3652
3653static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3654 struct bpf_link_info *info)
3655{
3656 struct bpf_raw_tp_link *raw_tp_link =
3657 container_of(link, struct bpf_raw_tp_link, link);
3658 char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3659 const char *tp_name = raw_tp_link->btp->tp->name;
3660 u32 ulen = info->raw_tracepoint.tp_name_len;
3661 size_t tp_len = strlen(tp_name);
3662
3663 if (!ulen ^ !ubuf)
3664 return -EINVAL;
3665
3666 info->raw_tracepoint.tp_name_len = tp_len + 1;
3667
3668 if (!ubuf)
3669 return 0;
3670
3671 return bpf_copy_to_user(ubuf, tp_name, ulen, tp_len);
3672}
3673
3674static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3675 .release = bpf_raw_tp_link_release,
3676 .dealloc_deferred = bpf_raw_tp_link_dealloc,
3677 .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3678 .fill_link_info = bpf_raw_tp_link_fill_link_info,
3679};
3680
3681#ifdef CONFIG_PERF_EVENTS
3682struct bpf_perf_link {
3683 struct bpf_link link;
3684 struct file *perf_file;
3685};
3686
3687static void bpf_perf_link_release(struct bpf_link *link)
3688{
3689 struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3690 struct perf_event *event = perf_link->perf_file->private_data;
3691
3692 perf_event_free_bpf_prog(event);
3693 fput(perf_link->perf_file);
3694}
3695
3696static void bpf_perf_link_dealloc(struct bpf_link *link)
3697{
3698 struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3699
3700 kfree(perf_link);
3701}
3702
3703static int bpf_perf_link_fill_common(const struct perf_event *event,
3704 char __user *uname, u32 *ulenp,
3705 u64 *probe_offset, u64 *probe_addr,
3706 u32 *fd_type, unsigned long *missed)
3707{
3708 const char *buf;
3709 u32 prog_id, ulen;
3710 size_t len;
3711 int err;
3712
3713 ulen = *ulenp;
3714 if (!ulen ^ !uname)
3715 return -EINVAL;
3716
3717 err = bpf_get_perf_event_info(event, &prog_id, fd_type, &buf,
3718 probe_offset, probe_addr, missed);
3719 if (err)
3720 return err;
3721
3722 if (buf) {
3723 len = strlen(buf);
3724 *ulenp = len + 1;
3725 } else {
3726 *ulenp = 1;
3727 }
3728 if (!uname)
3729 return 0;
3730
3731 if (buf) {
3732 err = bpf_copy_to_user(uname, buf, ulen, len);
3733 if (err)
3734 return err;
3735 } else {
3736 char zero = '\0';
3737
3738 if (put_user(zero, uname))
3739 return -EFAULT;
3740 }
3741 return 0;
3742}
3743
3744#ifdef CONFIG_KPROBE_EVENTS
3745static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
3746 struct bpf_link_info *info)
3747{
3748 unsigned long missed;
3749 char __user *uname;
3750 u64 addr, offset;
3751 u32 ulen, type;
3752 int err;
3753
3754 uname = u64_to_user_ptr(info->perf_event.kprobe.func_name);
3755 ulen = info->perf_event.kprobe.name_len;
3756 err = bpf_perf_link_fill_common(event, uname, &ulen, &offset, &addr,
3757 &type, &missed);
3758 if (err)
3759 return err;
3760 if (type == BPF_FD_TYPE_KRETPROBE)
3761 info->perf_event.type = BPF_PERF_EVENT_KRETPROBE;
3762 else
3763 info->perf_event.type = BPF_PERF_EVENT_KPROBE;
3764 info->perf_event.kprobe.name_len = ulen;
3765 info->perf_event.kprobe.offset = offset;
3766 info->perf_event.kprobe.missed = missed;
3767 if (!kallsyms_show_value(current_cred()))
3768 addr = 0;
3769 info->perf_event.kprobe.addr = addr;
3770 info->perf_event.kprobe.cookie = event->bpf_cookie;
3771 return 0;
3772}
3773#endif
3774
3775#ifdef CONFIG_UPROBE_EVENTS
3776static int bpf_perf_link_fill_uprobe(const struct perf_event *event,
3777 struct bpf_link_info *info)
3778{
3779 char __user *uname;
3780 u64 addr, offset;
3781 u32 ulen, type;
3782 int err;
3783
3784 uname = u64_to_user_ptr(info->perf_event.uprobe.file_name);
3785 ulen = info->perf_event.uprobe.name_len;
3786 err = bpf_perf_link_fill_common(event, uname, &ulen, &offset, &addr,
3787 &type, NULL);
3788 if (err)
3789 return err;
3790
3791 if (type == BPF_FD_TYPE_URETPROBE)
3792 info->perf_event.type = BPF_PERF_EVENT_URETPROBE;
3793 else
3794 info->perf_event.type = BPF_PERF_EVENT_UPROBE;
3795 info->perf_event.uprobe.name_len = ulen;
3796 info->perf_event.uprobe.offset = offset;
3797 info->perf_event.uprobe.cookie = event->bpf_cookie;
3798 return 0;
3799}
3800#endif
3801
3802static int bpf_perf_link_fill_probe(const struct perf_event *event,
3803 struct bpf_link_info *info)
3804{
3805#ifdef CONFIG_KPROBE_EVENTS
3806 if (event->tp_event->flags & TRACE_EVENT_FL_KPROBE)
3807 return bpf_perf_link_fill_kprobe(event, info);
3808#endif
3809#ifdef CONFIG_UPROBE_EVENTS
3810 if (event->tp_event->flags & TRACE_EVENT_FL_UPROBE)
3811 return bpf_perf_link_fill_uprobe(event, info);
3812#endif
3813 return -EOPNOTSUPP;
3814}
3815
3816static int bpf_perf_link_fill_tracepoint(const struct perf_event *event,
3817 struct bpf_link_info *info)
3818{
3819 char __user *uname;
3820 u32 ulen;
3821 int err;
3822
3823 uname = u64_to_user_ptr(info->perf_event.tracepoint.tp_name);
3824 ulen = info->perf_event.tracepoint.name_len;
3825 err = bpf_perf_link_fill_common(event, uname, &ulen, NULL, NULL, NULL, NULL);
3826 if (err)
3827 return err;
3828
3829 info->perf_event.type = BPF_PERF_EVENT_TRACEPOINT;
3830 info->perf_event.tracepoint.name_len = ulen;
3831 info->perf_event.tracepoint.cookie = event->bpf_cookie;
3832 return 0;
3833}
3834
3835static int bpf_perf_link_fill_perf_event(const struct perf_event *event,
3836 struct bpf_link_info *info)
3837{
3838 info->perf_event.event.type = event->attr.type;
3839 info->perf_event.event.config = event->attr.config;
3840 info->perf_event.event.cookie = event->bpf_cookie;
3841 info->perf_event.type = BPF_PERF_EVENT_EVENT;
3842 return 0;
3843}
3844
3845static int bpf_perf_link_fill_link_info(const struct bpf_link *link,
3846 struct bpf_link_info *info)
3847{
3848 struct bpf_perf_link *perf_link;
3849 const struct perf_event *event;
3850
3851 perf_link = container_of(link, struct bpf_perf_link, link);
3852 event = perf_get_event(perf_link->perf_file);
3853 if (IS_ERR(event))
3854 return PTR_ERR(event);
3855
3856 switch (event->prog->type) {
3857 case BPF_PROG_TYPE_PERF_EVENT:
3858 return bpf_perf_link_fill_perf_event(event, info);
3859 case BPF_PROG_TYPE_TRACEPOINT:
3860 return bpf_perf_link_fill_tracepoint(event, info);
3861 case BPF_PROG_TYPE_KPROBE:
3862 return bpf_perf_link_fill_probe(event, info);
3863 default:
3864 return -EOPNOTSUPP;
3865 }
3866}
3867
3868static const struct bpf_link_ops bpf_perf_link_lops = {
3869 .release = bpf_perf_link_release,
3870 .dealloc = bpf_perf_link_dealloc,
3871 .fill_link_info = bpf_perf_link_fill_link_info,
3872};
3873
3874static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3875{
3876 struct bpf_link_primer link_primer;
3877 struct bpf_perf_link *link;
3878 struct perf_event *event;
3879 struct file *perf_file;
3880 int err;
3881
3882 if (attr->link_create.flags)
3883 return -EINVAL;
3884
3885 perf_file = perf_event_get(attr->link_create.target_fd);
3886 if (IS_ERR(perf_file))
3887 return PTR_ERR(perf_file);
3888
3889 link = kzalloc(sizeof(*link), GFP_USER);
3890 if (!link) {
3891 err = -ENOMEM;
3892 goto out_put_file;
3893 }
3894 bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3895 link->perf_file = perf_file;
3896
3897 err = bpf_link_prime(&link->link, &link_primer);
3898 if (err) {
3899 kfree(link);
3900 goto out_put_file;
3901 }
3902
3903 event = perf_file->private_data;
3904 err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3905 if (err) {
3906 bpf_link_cleanup(&link_primer);
3907 goto out_put_file;
3908 }
3909 /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3910 bpf_prog_inc(prog);
3911
3912 return bpf_link_settle(&link_primer);
3913
3914out_put_file:
3915 fput(perf_file);
3916 return err;
3917}
3918#else
3919static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3920{
3921 return -EOPNOTSUPP;
3922}
3923#endif /* CONFIG_PERF_EVENTS */
3924
3925static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3926 const char __user *user_tp_name, u64 cookie)
3927{
3928 struct bpf_link_primer link_primer;
3929 struct bpf_raw_tp_link *link;
3930 struct bpf_raw_event_map *btp;
3931 const char *tp_name;
3932 char buf[128];
3933 int err;
3934
3935 switch (prog->type) {
3936 case BPF_PROG_TYPE_TRACING:
3937 case BPF_PROG_TYPE_EXT:
3938 case BPF_PROG_TYPE_LSM:
3939 if (user_tp_name)
3940 /* The attach point for this category of programs
3941 * should be specified via btf_id during program load.
3942 */
3943 return -EINVAL;
3944 if (prog->type == BPF_PROG_TYPE_TRACING &&
3945 prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3946 tp_name = prog->aux->attach_func_name;
3947 break;
3948 }
3949 return bpf_tracing_prog_attach(prog, 0, 0, 0);
3950 case BPF_PROG_TYPE_RAW_TRACEPOINT:
3951 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3952 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3953 return -EFAULT;
3954 buf[sizeof(buf) - 1] = 0;
3955 tp_name = buf;
3956 break;
3957 default:
3958 return -EINVAL;
3959 }
3960
3961 btp = bpf_get_raw_tracepoint(tp_name);
3962 if (!btp)
3963 return -ENOENT;
3964
3965 link = kzalloc(sizeof(*link), GFP_USER);
3966 if (!link) {
3967 err = -ENOMEM;
3968 goto out_put_btp;
3969 }
3970 bpf_link_init_sleepable(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3971 &bpf_raw_tp_link_lops, prog,
3972 tracepoint_is_faultable(btp->tp));
3973 link->btp = btp;
3974 link->cookie = cookie;
3975
3976 err = bpf_link_prime(&link->link, &link_primer);
3977 if (err) {
3978 kfree(link);
3979 goto out_put_btp;
3980 }
3981
3982 err = bpf_probe_register(link->btp, link);
3983 if (err) {
3984 bpf_link_cleanup(&link_primer);
3985 goto out_put_btp;
3986 }
3987
3988 return bpf_link_settle(&link_primer);
3989
3990out_put_btp:
3991 bpf_put_raw_tracepoint(btp);
3992 return err;
3993}
3994
3995#define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.cookie
3996
3997static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3998{
3999 struct bpf_prog *prog;
4000 void __user *tp_name;
4001 __u64 cookie;
4002 int fd;
4003
4004 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
4005 return -EINVAL;
4006
4007 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
4008 if (IS_ERR(prog))
4009 return PTR_ERR(prog);
4010
4011 tp_name = u64_to_user_ptr(attr->raw_tracepoint.name);
4012 cookie = attr->raw_tracepoint.cookie;
4013 fd = bpf_raw_tp_link_attach(prog, tp_name, cookie);
4014 if (fd < 0)
4015 bpf_prog_put(prog);
4016 return fd;
4017}
4018
4019static enum bpf_prog_type
4020attach_type_to_prog_type(enum bpf_attach_type attach_type)
4021{
4022 switch (attach_type) {
4023 case BPF_CGROUP_INET_INGRESS:
4024 case BPF_CGROUP_INET_EGRESS:
4025 return BPF_PROG_TYPE_CGROUP_SKB;
4026 case BPF_CGROUP_INET_SOCK_CREATE:
4027 case BPF_CGROUP_INET_SOCK_RELEASE:
4028 case BPF_CGROUP_INET4_POST_BIND:
4029 case BPF_CGROUP_INET6_POST_BIND:
4030 return BPF_PROG_TYPE_CGROUP_SOCK;
4031 case BPF_CGROUP_INET4_BIND:
4032 case BPF_CGROUP_INET6_BIND:
4033 case BPF_CGROUP_INET4_CONNECT:
4034 case BPF_CGROUP_INET6_CONNECT:
4035 case BPF_CGROUP_UNIX_CONNECT:
4036 case BPF_CGROUP_INET4_GETPEERNAME:
4037 case BPF_CGROUP_INET6_GETPEERNAME:
4038 case BPF_CGROUP_UNIX_GETPEERNAME:
4039 case BPF_CGROUP_INET4_GETSOCKNAME:
4040 case BPF_CGROUP_INET6_GETSOCKNAME:
4041 case BPF_CGROUP_UNIX_GETSOCKNAME:
4042 case BPF_CGROUP_UDP4_SENDMSG:
4043 case BPF_CGROUP_UDP6_SENDMSG:
4044 case BPF_CGROUP_UNIX_SENDMSG:
4045 case BPF_CGROUP_UDP4_RECVMSG:
4046 case BPF_CGROUP_UDP6_RECVMSG:
4047 case BPF_CGROUP_UNIX_RECVMSG:
4048 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
4049 case BPF_CGROUP_SOCK_OPS:
4050 return BPF_PROG_TYPE_SOCK_OPS;
4051 case BPF_CGROUP_DEVICE:
4052 return BPF_PROG_TYPE_CGROUP_DEVICE;
4053 case BPF_SK_MSG_VERDICT:
4054 return BPF_PROG_TYPE_SK_MSG;
4055 case BPF_SK_SKB_STREAM_PARSER:
4056 case BPF_SK_SKB_STREAM_VERDICT:
4057 case BPF_SK_SKB_VERDICT:
4058 return BPF_PROG_TYPE_SK_SKB;
4059 case BPF_LIRC_MODE2:
4060 return BPF_PROG_TYPE_LIRC_MODE2;
4061 case BPF_FLOW_DISSECTOR:
4062 return BPF_PROG_TYPE_FLOW_DISSECTOR;
4063 case BPF_CGROUP_SYSCTL:
4064 return BPF_PROG_TYPE_CGROUP_SYSCTL;
4065 case BPF_CGROUP_GETSOCKOPT:
4066 case BPF_CGROUP_SETSOCKOPT:
4067 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
4068 case BPF_TRACE_ITER:
4069 case BPF_TRACE_RAW_TP:
4070 case BPF_TRACE_FENTRY:
4071 case BPF_TRACE_FEXIT:
4072 case BPF_MODIFY_RETURN:
4073 return BPF_PROG_TYPE_TRACING;
4074 case BPF_LSM_MAC:
4075 return BPF_PROG_TYPE_LSM;
4076 case BPF_SK_LOOKUP:
4077 return BPF_PROG_TYPE_SK_LOOKUP;
4078 case BPF_XDP:
4079 return BPF_PROG_TYPE_XDP;
4080 case BPF_LSM_CGROUP:
4081 return BPF_PROG_TYPE_LSM;
4082 case BPF_TCX_INGRESS:
4083 case BPF_TCX_EGRESS:
4084 case BPF_NETKIT_PRIMARY:
4085 case BPF_NETKIT_PEER:
4086 return BPF_PROG_TYPE_SCHED_CLS;
4087 default:
4088 return BPF_PROG_TYPE_UNSPEC;
4089 }
4090}
4091
4092static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
4093 enum bpf_attach_type attach_type)
4094{
4095 enum bpf_prog_type ptype;
4096
4097 switch (prog->type) {
4098 case BPF_PROG_TYPE_CGROUP_SOCK:
4099 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4100 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4101 case BPF_PROG_TYPE_SK_LOOKUP:
4102 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
4103 case BPF_PROG_TYPE_CGROUP_SKB:
4104 if (!bpf_token_capable(prog->aux->token, CAP_NET_ADMIN))
4105 /* cg-skb progs can be loaded by unpriv user.
4106 * check permissions at attach time.
4107 */
4108 return -EPERM;
4109
4110 ptype = attach_type_to_prog_type(attach_type);
4111 if (prog->type != ptype)
4112 return -EINVAL;
4113
4114 return prog->enforce_expected_attach_type &&
4115 prog->expected_attach_type != attach_type ?
4116 -EINVAL : 0;
4117 case BPF_PROG_TYPE_EXT:
4118 return 0;
4119 case BPF_PROG_TYPE_NETFILTER:
4120 if (attach_type != BPF_NETFILTER)
4121 return -EINVAL;
4122 return 0;
4123 case BPF_PROG_TYPE_PERF_EVENT:
4124 case BPF_PROG_TYPE_TRACEPOINT:
4125 if (attach_type != BPF_PERF_EVENT)
4126 return -EINVAL;
4127 return 0;
4128 case BPF_PROG_TYPE_KPROBE:
4129 if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI &&
4130 attach_type != BPF_TRACE_KPROBE_MULTI)
4131 return -EINVAL;
4132 if (prog->expected_attach_type == BPF_TRACE_KPROBE_SESSION &&
4133 attach_type != BPF_TRACE_KPROBE_SESSION)
4134 return -EINVAL;
4135 if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI &&
4136 attach_type != BPF_TRACE_UPROBE_MULTI)
4137 return -EINVAL;
4138 if (prog->expected_attach_type == BPF_TRACE_UPROBE_SESSION &&
4139 attach_type != BPF_TRACE_UPROBE_SESSION)
4140 return -EINVAL;
4141 if (attach_type != BPF_PERF_EVENT &&
4142 attach_type != BPF_TRACE_KPROBE_MULTI &&
4143 attach_type != BPF_TRACE_KPROBE_SESSION &&
4144 attach_type != BPF_TRACE_UPROBE_MULTI &&
4145 attach_type != BPF_TRACE_UPROBE_SESSION)
4146 return -EINVAL;
4147 return 0;
4148 case BPF_PROG_TYPE_SCHED_CLS:
4149 if (attach_type != BPF_TCX_INGRESS &&
4150 attach_type != BPF_TCX_EGRESS &&
4151 attach_type != BPF_NETKIT_PRIMARY &&
4152 attach_type != BPF_NETKIT_PEER)
4153 return -EINVAL;
4154 return 0;
4155 default:
4156 ptype = attach_type_to_prog_type(attach_type);
4157 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type)
4158 return -EINVAL;
4159 return 0;
4160 }
4161}
4162
4163#define BPF_PROG_ATTACH_LAST_FIELD expected_revision
4164
4165#define BPF_F_ATTACH_MASK_BASE \
4166 (BPF_F_ALLOW_OVERRIDE | \
4167 BPF_F_ALLOW_MULTI | \
4168 BPF_F_REPLACE)
4169
4170#define BPF_F_ATTACH_MASK_MPROG \
4171 (BPF_F_REPLACE | \
4172 BPF_F_BEFORE | \
4173 BPF_F_AFTER | \
4174 BPF_F_ID | \
4175 BPF_F_LINK)
4176
4177static int bpf_prog_attach(const union bpf_attr *attr)
4178{
4179 enum bpf_prog_type ptype;
4180 struct bpf_prog *prog;
4181 int ret;
4182
4183 if (CHECK_ATTR(BPF_PROG_ATTACH))
4184 return -EINVAL;
4185
4186 ptype = attach_type_to_prog_type(attr->attach_type);
4187 if (ptype == BPF_PROG_TYPE_UNSPEC)
4188 return -EINVAL;
4189 if (bpf_mprog_supported(ptype)) {
4190 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
4191 return -EINVAL;
4192 } else {
4193 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_BASE)
4194 return -EINVAL;
4195 if (attr->relative_fd ||
4196 attr->expected_revision)
4197 return -EINVAL;
4198 }
4199
4200 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
4201 if (IS_ERR(prog))
4202 return PTR_ERR(prog);
4203
4204 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
4205 bpf_prog_put(prog);
4206 return -EINVAL;
4207 }
4208
4209 switch (ptype) {
4210 case BPF_PROG_TYPE_SK_SKB:
4211 case BPF_PROG_TYPE_SK_MSG:
4212 ret = sock_map_get_from_fd(attr, prog);
4213 break;
4214 case BPF_PROG_TYPE_LIRC_MODE2:
4215 ret = lirc_prog_attach(attr, prog);
4216 break;
4217 case BPF_PROG_TYPE_FLOW_DISSECTOR:
4218 ret = netns_bpf_prog_attach(attr, prog);
4219 break;
4220 case BPF_PROG_TYPE_CGROUP_DEVICE:
4221 case BPF_PROG_TYPE_CGROUP_SKB:
4222 case BPF_PROG_TYPE_CGROUP_SOCK:
4223 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4224 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4225 case BPF_PROG_TYPE_CGROUP_SYSCTL:
4226 case BPF_PROG_TYPE_SOCK_OPS:
4227 case BPF_PROG_TYPE_LSM:
4228 if (ptype == BPF_PROG_TYPE_LSM &&
4229 prog->expected_attach_type != BPF_LSM_CGROUP)
4230 ret = -EINVAL;
4231 else
4232 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
4233 break;
4234 case BPF_PROG_TYPE_SCHED_CLS:
4235 if (attr->attach_type == BPF_TCX_INGRESS ||
4236 attr->attach_type == BPF_TCX_EGRESS)
4237 ret = tcx_prog_attach(attr, prog);
4238 else
4239 ret = netkit_prog_attach(attr, prog);
4240 break;
4241 default:
4242 ret = -EINVAL;
4243 }
4244
4245 if (ret)
4246 bpf_prog_put(prog);
4247 return ret;
4248}
4249
4250#define BPF_PROG_DETACH_LAST_FIELD expected_revision
4251
4252static int bpf_prog_detach(const union bpf_attr *attr)
4253{
4254 struct bpf_prog *prog = NULL;
4255 enum bpf_prog_type ptype;
4256 int ret;
4257
4258 if (CHECK_ATTR(BPF_PROG_DETACH))
4259 return -EINVAL;
4260
4261 ptype = attach_type_to_prog_type(attr->attach_type);
4262 if (bpf_mprog_supported(ptype)) {
4263 if (ptype == BPF_PROG_TYPE_UNSPEC)
4264 return -EINVAL;
4265 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
4266 return -EINVAL;
4267 if (attr->attach_bpf_fd) {
4268 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
4269 if (IS_ERR(prog))
4270 return PTR_ERR(prog);
4271 }
4272 } else if (attr->attach_flags ||
4273 attr->relative_fd ||
4274 attr->expected_revision) {
4275 return -EINVAL;
4276 }
4277
4278 switch (ptype) {
4279 case BPF_PROG_TYPE_SK_MSG:
4280 case BPF_PROG_TYPE_SK_SKB:
4281 ret = sock_map_prog_detach(attr, ptype);
4282 break;
4283 case BPF_PROG_TYPE_LIRC_MODE2:
4284 ret = lirc_prog_detach(attr);
4285 break;
4286 case BPF_PROG_TYPE_FLOW_DISSECTOR:
4287 ret = netns_bpf_prog_detach(attr, ptype);
4288 break;
4289 case BPF_PROG_TYPE_CGROUP_DEVICE:
4290 case BPF_PROG_TYPE_CGROUP_SKB:
4291 case BPF_PROG_TYPE_CGROUP_SOCK:
4292 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4293 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4294 case BPF_PROG_TYPE_CGROUP_SYSCTL:
4295 case BPF_PROG_TYPE_SOCK_OPS:
4296 case BPF_PROG_TYPE_LSM:
4297 ret = cgroup_bpf_prog_detach(attr, ptype);
4298 break;
4299 case BPF_PROG_TYPE_SCHED_CLS:
4300 if (attr->attach_type == BPF_TCX_INGRESS ||
4301 attr->attach_type == BPF_TCX_EGRESS)
4302 ret = tcx_prog_detach(attr, prog);
4303 else
4304 ret = netkit_prog_detach(attr, prog);
4305 break;
4306 default:
4307 ret = -EINVAL;
4308 }
4309
4310 if (prog)
4311 bpf_prog_put(prog);
4312 return ret;
4313}
4314
4315#define BPF_PROG_QUERY_LAST_FIELD query.revision
4316
4317static int bpf_prog_query(const union bpf_attr *attr,
4318 union bpf_attr __user *uattr)
4319{
4320 if (!bpf_net_capable())
4321 return -EPERM;
4322 if (CHECK_ATTR(BPF_PROG_QUERY))
4323 return -EINVAL;
4324 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
4325 return -EINVAL;
4326
4327 switch (attr->query.attach_type) {
4328 case BPF_CGROUP_INET_INGRESS:
4329 case BPF_CGROUP_INET_EGRESS:
4330 case BPF_CGROUP_INET_SOCK_CREATE:
4331 case BPF_CGROUP_INET_SOCK_RELEASE:
4332 case BPF_CGROUP_INET4_BIND:
4333 case BPF_CGROUP_INET6_BIND:
4334 case BPF_CGROUP_INET4_POST_BIND:
4335 case BPF_CGROUP_INET6_POST_BIND:
4336 case BPF_CGROUP_INET4_CONNECT:
4337 case BPF_CGROUP_INET6_CONNECT:
4338 case BPF_CGROUP_UNIX_CONNECT:
4339 case BPF_CGROUP_INET4_GETPEERNAME:
4340 case BPF_CGROUP_INET6_GETPEERNAME:
4341 case BPF_CGROUP_UNIX_GETPEERNAME:
4342 case BPF_CGROUP_INET4_GETSOCKNAME:
4343 case BPF_CGROUP_INET6_GETSOCKNAME:
4344 case BPF_CGROUP_UNIX_GETSOCKNAME:
4345 case BPF_CGROUP_UDP4_SENDMSG:
4346 case BPF_CGROUP_UDP6_SENDMSG:
4347 case BPF_CGROUP_UNIX_SENDMSG:
4348 case BPF_CGROUP_UDP4_RECVMSG:
4349 case BPF_CGROUP_UDP6_RECVMSG:
4350 case BPF_CGROUP_UNIX_RECVMSG:
4351 case BPF_CGROUP_SOCK_OPS:
4352 case BPF_CGROUP_DEVICE:
4353 case BPF_CGROUP_SYSCTL:
4354 case BPF_CGROUP_GETSOCKOPT:
4355 case BPF_CGROUP_SETSOCKOPT:
4356 case BPF_LSM_CGROUP:
4357 return cgroup_bpf_prog_query(attr, uattr);
4358 case BPF_LIRC_MODE2:
4359 return lirc_prog_query(attr, uattr);
4360 case BPF_FLOW_DISSECTOR:
4361 case BPF_SK_LOOKUP:
4362 return netns_bpf_prog_query(attr, uattr);
4363 case BPF_SK_SKB_STREAM_PARSER:
4364 case BPF_SK_SKB_STREAM_VERDICT:
4365 case BPF_SK_MSG_VERDICT:
4366 case BPF_SK_SKB_VERDICT:
4367 return sock_map_bpf_prog_query(attr, uattr);
4368 case BPF_TCX_INGRESS:
4369 case BPF_TCX_EGRESS:
4370 return tcx_prog_query(attr, uattr);
4371 case BPF_NETKIT_PRIMARY:
4372 case BPF_NETKIT_PEER:
4373 return netkit_prog_query(attr, uattr);
4374 default:
4375 return -EINVAL;
4376 }
4377}
4378
4379#define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
4380
4381static int bpf_prog_test_run(const union bpf_attr *attr,
4382 union bpf_attr __user *uattr)
4383{
4384 struct bpf_prog *prog;
4385 int ret = -ENOTSUPP;
4386
4387 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
4388 return -EINVAL;
4389
4390 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
4391 (!attr->test.ctx_size_in && attr->test.ctx_in))
4392 return -EINVAL;
4393
4394 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
4395 (!attr->test.ctx_size_out && attr->test.ctx_out))
4396 return -EINVAL;
4397
4398 prog = bpf_prog_get(attr->test.prog_fd);
4399 if (IS_ERR(prog))
4400 return PTR_ERR(prog);
4401
4402 if (prog->aux->ops->test_run)
4403 ret = prog->aux->ops->test_run(prog, attr, uattr);
4404
4405 bpf_prog_put(prog);
4406 return ret;
4407}
4408
4409#define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
4410
4411static int bpf_obj_get_next_id(const union bpf_attr *attr,
4412 union bpf_attr __user *uattr,
4413 struct idr *idr,
4414 spinlock_t *lock)
4415{
4416 u32 next_id = attr->start_id;
4417 int err = 0;
4418
4419 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
4420 return -EINVAL;
4421
4422 if (!capable(CAP_SYS_ADMIN))
4423 return -EPERM;
4424
4425 next_id++;
4426 spin_lock_bh(lock);
4427 if (!idr_get_next(idr, &next_id))
4428 err = -ENOENT;
4429 spin_unlock_bh(lock);
4430
4431 if (!err)
4432 err = put_user(next_id, &uattr->next_id);
4433
4434 return err;
4435}
4436
4437struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
4438{
4439 struct bpf_map *map;
4440
4441 spin_lock_bh(&map_idr_lock);
4442again:
4443 map = idr_get_next(&map_idr, id);
4444 if (map) {
4445 map = __bpf_map_inc_not_zero(map, false);
4446 if (IS_ERR(map)) {
4447 (*id)++;
4448 goto again;
4449 }
4450 }
4451 spin_unlock_bh(&map_idr_lock);
4452
4453 return map;
4454}
4455
4456struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
4457{
4458 struct bpf_prog *prog;
4459
4460 spin_lock_bh(&prog_idr_lock);
4461again:
4462 prog = idr_get_next(&prog_idr, id);
4463 if (prog) {
4464 prog = bpf_prog_inc_not_zero(prog);
4465 if (IS_ERR(prog)) {
4466 (*id)++;
4467 goto again;
4468 }
4469 }
4470 spin_unlock_bh(&prog_idr_lock);
4471
4472 return prog;
4473}
4474
4475#define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
4476
4477struct bpf_prog *bpf_prog_by_id(u32 id)
4478{
4479 struct bpf_prog *prog;
4480
4481 if (!id)
4482 return ERR_PTR(-ENOENT);
4483
4484 spin_lock_bh(&prog_idr_lock);
4485 prog = idr_find(&prog_idr, id);
4486 if (prog)
4487 prog = bpf_prog_inc_not_zero(prog);
4488 else
4489 prog = ERR_PTR(-ENOENT);
4490 spin_unlock_bh(&prog_idr_lock);
4491 return prog;
4492}
4493
4494static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
4495{
4496 struct bpf_prog *prog;
4497 u32 id = attr->prog_id;
4498 int fd;
4499
4500 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
4501 return -EINVAL;
4502
4503 if (!capable(CAP_SYS_ADMIN))
4504 return -EPERM;
4505
4506 prog = bpf_prog_by_id(id);
4507 if (IS_ERR(prog))
4508 return PTR_ERR(prog);
4509
4510 fd = bpf_prog_new_fd(prog);
4511 if (fd < 0)
4512 bpf_prog_put(prog);
4513
4514 return fd;
4515}
4516
4517#define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
4518
4519static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
4520{
4521 struct bpf_map *map;
4522 u32 id = attr->map_id;
4523 int f_flags;
4524 int fd;
4525
4526 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
4527 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
4528 return -EINVAL;
4529
4530 if (!capable(CAP_SYS_ADMIN))
4531 return -EPERM;
4532
4533 f_flags = bpf_get_file_flag(attr->open_flags);
4534 if (f_flags < 0)
4535 return f_flags;
4536
4537 spin_lock_bh(&map_idr_lock);
4538 map = idr_find(&map_idr, id);
4539 if (map)
4540 map = __bpf_map_inc_not_zero(map, true);
4541 else
4542 map = ERR_PTR(-ENOENT);
4543 spin_unlock_bh(&map_idr_lock);
4544
4545 if (IS_ERR(map))
4546 return PTR_ERR(map);
4547
4548 fd = bpf_map_new_fd(map, f_flags);
4549 if (fd < 0)
4550 bpf_map_put_with_uref(map);
4551
4552 return fd;
4553}
4554
4555static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
4556 unsigned long addr, u32 *off,
4557 u32 *type)
4558{
4559 const struct bpf_map *map;
4560 int i;
4561
4562 mutex_lock(&prog->aux->used_maps_mutex);
4563 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
4564 map = prog->aux->used_maps[i];
4565 if (map == (void *)addr) {
4566 *type = BPF_PSEUDO_MAP_FD;
4567 goto out;
4568 }
4569 if (!map->ops->map_direct_value_meta)
4570 continue;
4571 if (!map->ops->map_direct_value_meta(map, addr, off)) {
4572 *type = BPF_PSEUDO_MAP_VALUE;
4573 goto out;
4574 }
4575 }
4576 map = NULL;
4577
4578out:
4579 mutex_unlock(&prog->aux->used_maps_mutex);
4580 return map;
4581}
4582
4583static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
4584 const struct cred *f_cred)
4585{
4586 const struct bpf_map *map;
4587 struct bpf_insn *insns;
4588 u32 off, type;
4589 u64 imm;
4590 u8 code;
4591 int i;
4592
4593 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
4594 GFP_USER);
4595 if (!insns)
4596 return insns;
4597
4598 for (i = 0; i < prog->len; i++) {
4599 code = insns[i].code;
4600
4601 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
4602 insns[i].code = BPF_JMP | BPF_CALL;
4603 insns[i].imm = BPF_FUNC_tail_call;
4604 /* fall-through */
4605 }
4606 if (code == (BPF_JMP | BPF_CALL) ||
4607 code == (BPF_JMP | BPF_CALL_ARGS)) {
4608 if (code == (BPF_JMP | BPF_CALL_ARGS))
4609 insns[i].code = BPF_JMP | BPF_CALL;
4610 if (!bpf_dump_raw_ok(f_cred))
4611 insns[i].imm = 0;
4612 continue;
4613 }
4614 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
4615 insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
4616 continue;
4617 }
4618
4619 if ((BPF_CLASS(code) == BPF_LDX || BPF_CLASS(code) == BPF_STX ||
4620 BPF_CLASS(code) == BPF_ST) && BPF_MODE(code) == BPF_PROBE_MEM32) {
4621 insns[i].code = BPF_CLASS(code) | BPF_SIZE(code) | BPF_MEM;
4622 continue;
4623 }
4624
4625 if (code != (BPF_LD | BPF_IMM | BPF_DW))
4626 continue;
4627
4628 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
4629 map = bpf_map_from_imm(prog, imm, &off, &type);
4630 if (map) {
4631 insns[i].src_reg = type;
4632 insns[i].imm = map->id;
4633 insns[i + 1].imm = off;
4634 continue;
4635 }
4636 }
4637
4638 return insns;
4639}
4640
4641static int set_info_rec_size(struct bpf_prog_info *info)
4642{
4643 /*
4644 * Ensure info.*_rec_size is the same as kernel expected size
4645 *
4646 * or
4647 *
4648 * Only allow zero *_rec_size if both _rec_size and _cnt are
4649 * zero. In this case, the kernel will set the expected
4650 * _rec_size back to the info.
4651 */
4652
4653 if ((info->nr_func_info || info->func_info_rec_size) &&
4654 info->func_info_rec_size != sizeof(struct bpf_func_info))
4655 return -EINVAL;
4656
4657 if ((info->nr_line_info || info->line_info_rec_size) &&
4658 info->line_info_rec_size != sizeof(struct bpf_line_info))
4659 return -EINVAL;
4660
4661 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
4662 info->jited_line_info_rec_size != sizeof(__u64))
4663 return -EINVAL;
4664
4665 info->func_info_rec_size = sizeof(struct bpf_func_info);
4666 info->line_info_rec_size = sizeof(struct bpf_line_info);
4667 info->jited_line_info_rec_size = sizeof(__u64);
4668
4669 return 0;
4670}
4671
4672static int bpf_prog_get_info_by_fd(struct file *file,
4673 struct bpf_prog *prog,
4674 const union bpf_attr *attr,
4675 union bpf_attr __user *uattr)
4676{
4677 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4678 struct btf *attach_btf = bpf_prog_get_target_btf(prog);
4679 struct bpf_prog_info info;
4680 u32 info_len = attr->info.info_len;
4681 struct bpf_prog_kstats stats;
4682 char __user *uinsns;
4683 u32 ulen;
4684 int err;
4685
4686 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4687 if (err)
4688 return err;
4689 info_len = min_t(u32, sizeof(info), info_len);
4690
4691 memset(&info, 0, sizeof(info));
4692 if (copy_from_user(&info, uinfo, info_len))
4693 return -EFAULT;
4694
4695 info.type = prog->type;
4696 info.id = prog->aux->id;
4697 info.load_time = prog->aux->load_time;
4698 info.created_by_uid = from_kuid_munged(current_user_ns(),
4699 prog->aux->user->uid);
4700 info.gpl_compatible = prog->gpl_compatible;
4701
4702 memcpy(info.tag, prog->tag, sizeof(prog->tag));
4703 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
4704
4705 mutex_lock(&prog->aux->used_maps_mutex);
4706 ulen = info.nr_map_ids;
4707 info.nr_map_ids = prog->aux->used_map_cnt;
4708 ulen = min_t(u32, info.nr_map_ids, ulen);
4709 if (ulen) {
4710 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
4711 u32 i;
4712
4713 for (i = 0; i < ulen; i++)
4714 if (put_user(prog->aux->used_maps[i]->id,
4715 &user_map_ids[i])) {
4716 mutex_unlock(&prog->aux->used_maps_mutex);
4717 return -EFAULT;
4718 }
4719 }
4720 mutex_unlock(&prog->aux->used_maps_mutex);
4721
4722 err = set_info_rec_size(&info);
4723 if (err)
4724 return err;
4725
4726 bpf_prog_get_stats(prog, &stats);
4727 info.run_time_ns = stats.nsecs;
4728 info.run_cnt = stats.cnt;
4729 info.recursion_misses = stats.misses;
4730
4731 info.verified_insns = prog->aux->verified_insns;
4732
4733 if (!bpf_capable()) {
4734 info.jited_prog_len = 0;
4735 info.xlated_prog_len = 0;
4736 info.nr_jited_ksyms = 0;
4737 info.nr_jited_func_lens = 0;
4738 info.nr_func_info = 0;
4739 info.nr_line_info = 0;
4740 info.nr_jited_line_info = 0;
4741 goto done;
4742 }
4743
4744 ulen = info.xlated_prog_len;
4745 info.xlated_prog_len = bpf_prog_insn_size(prog);
4746 if (info.xlated_prog_len && ulen) {
4747 struct bpf_insn *insns_sanitized;
4748 bool fault;
4749
4750 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
4751 info.xlated_prog_insns = 0;
4752 goto done;
4753 }
4754 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4755 if (!insns_sanitized)
4756 return -ENOMEM;
4757 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4758 ulen = min_t(u32, info.xlated_prog_len, ulen);
4759 fault = copy_to_user(uinsns, insns_sanitized, ulen);
4760 kfree(insns_sanitized);
4761 if (fault)
4762 return -EFAULT;
4763 }
4764
4765 if (bpf_prog_is_offloaded(prog->aux)) {
4766 err = bpf_prog_offload_info_fill(&info, prog);
4767 if (err)
4768 return err;
4769 goto done;
4770 }
4771
4772 /* NOTE: the following code is supposed to be skipped for offload.
4773 * bpf_prog_offload_info_fill() is the place to fill similar fields
4774 * for offload.
4775 */
4776 ulen = info.jited_prog_len;
4777 if (prog->aux->func_cnt) {
4778 u32 i;
4779
4780 info.jited_prog_len = 0;
4781 for (i = 0; i < prog->aux->func_cnt; i++)
4782 info.jited_prog_len += prog->aux->func[i]->jited_len;
4783 } else {
4784 info.jited_prog_len = prog->jited_len;
4785 }
4786
4787 if (info.jited_prog_len && ulen) {
4788 if (bpf_dump_raw_ok(file->f_cred)) {
4789 uinsns = u64_to_user_ptr(info.jited_prog_insns);
4790 ulen = min_t(u32, info.jited_prog_len, ulen);
4791
4792 /* for multi-function programs, copy the JITed
4793 * instructions for all the functions
4794 */
4795 if (prog->aux->func_cnt) {
4796 u32 len, free, i;
4797 u8 *img;
4798
4799 free = ulen;
4800 for (i = 0; i < prog->aux->func_cnt; i++) {
4801 len = prog->aux->func[i]->jited_len;
4802 len = min_t(u32, len, free);
4803 img = (u8 *) prog->aux->func[i]->bpf_func;
4804 if (copy_to_user(uinsns, img, len))
4805 return -EFAULT;
4806 uinsns += len;
4807 free -= len;
4808 if (!free)
4809 break;
4810 }
4811 } else {
4812 if (copy_to_user(uinsns, prog->bpf_func, ulen))
4813 return -EFAULT;
4814 }
4815 } else {
4816 info.jited_prog_insns = 0;
4817 }
4818 }
4819
4820 ulen = info.nr_jited_ksyms;
4821 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4822 if (ulen) {
4823 if (bpf_dump_raw_ok(file->f_cred)) {
4824 unsigned long ksym_addr;
4825 u64 __user *user_ksyms;
4826 u32 i;
4827
4828 /* copy the address of the kernel symbol
4829 * corresponding to each function
4830 */
4831 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4832 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4833 if (prog->aux->func_cnt) {
4834 for (i = 0; i < ulen; i++) {
4835 ksym_addr = (unsigned long)
4836 prog->aux->func[i]->bpf_func;
4837 if (put_user((u64) ksym_addr,
4838 &user_ksyms[i]))
4839 return -EFAULT;
4840 }
4841 } else {
4842 ksym_addr = (unsigned long) prog->bpf_func;
4843 if (put_user((u64) ksym_addr, &user_ksyms[0]))
4844 return -EFAULT;
4845 }
4846 } else {
4847 info.jited_ksyms = 0;
4848 }
4849 }
4850
4851 ulen = info.nr_jited_func_lens;
4852 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4853 if (ulen) {
4854 if (bpf_dump_raw_ok(file->f_cred)) {
4855 u32 __user *user_lens;
4856 u32 func_len, i;
4857
4858 /* copy the JITed image lengths for each function */
4859 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4860 user_lens = u64_to_user_ptr(info.jited_func_lens);
4861 if (prog->aux->func_cnt) {
4862 for (i = 0; i < ulen; i++) {
4863 func_len =
4864 prog->aux->func[i]->jited_len;
4865 if (put_user(func_len, &user_lens[i]))
4866 return -EFAULT;
4867 }
4868 } else {
4869 func_len = prog->jited_len;
4870 if (put_user(func_len, &user_lens[0]))
4871 return -EFAULT;
4872 }
4873 } else {
4874 info.jited_func_lens = 0;
4875 }
4876 }
4877
4878 if (prog->aux->btf)
4879 info.btf_id = btf_obj_id(prog->aux->btf);
4880 info.attach_btf_id = prog->aux->attach_btf_id;
4881 if (attach_btf)
4882 info.attach_btf_obj_id = btf_obj_id(attach_btf);
4883
4884 ulen = info.nr_func_info;
4885 info.nr_func_info = prog->aux->func_info_cnt;
4886 if (info.nr_func_info && ulen) {
4887 char __user *user_finfo;
4888
4889 user_finfo = u64_to_user_ptr(info.func_info);
4890 ulen = min_t(u32, info.nr_func_info, ulen);
4891 if (copy_to_user(user_finfo, prog->aux->func_info,
4892 info.func_info_rec_size * ulen))
4893 return -EFAULT;
4894 }
4895
4896 ulen = info.nr_line_info;
4897 info.nr_line_info = prog->aux->nr_linfo;
4898 if (info.nr_line_info && ulen) {
4899 __u8 __user *user_linfo;
4900
4901 user_linfo = u64_to_user_ptr(info.line_info);
4902 ulen = min_t(u32, info.nr_line_info, ulen);
4903 if (copy_to_user(user_linfo, prog->aux->linfo,
4904 info.line_info_rec_size * ulen))
4905 return -EFAULT;
4906 }
4907
4908 ulen = info.nr_jited_line_info;
4909 if (prog->aux->jited_linfo)
4910 info.nr_jited_line_info = prog->aux->nr_linfo;
4911 else
4912 info.nr_jited_line_info = 0;
4913 if (info.nr_jited_line_info && ulen) {
4914 if (bpf_dump_raw_ok(file->f_cred)) {
4915 unsigned long line_addr;
4916 __u64 __user *user_linfo;
4917 u32 i;
4918
4919 user_linfo = u64_to_user_ptr(info.jited_line_info);
4920 ulen = min_t(u32, info.nr_jited_line_info, ulen);
4921 for (i = 0; i < ulen; i++) {
4922 line_addr = (unsigned long)prog->aux->jited_linfo[i];
4923 if (put_user((__u64)line_addr, &user_linfo[i]))
4924 return -EFAULT;
4925 }
4926 } else {
4927 info.jited_line_info = 0;
4928 }
4929 }
4930
4931 ulen = info.nr_prog_tags;
4932 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4933 if (ulen) {
4934 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4935 u32 i;
4936
4937 user_prog_tags = u64_to_user_ptr(info.prog_tags);
4938 ulen = min_t(u32, info.nr_prog_tags, ulen);
4939 if (prog->aux->func_cnt) {
4940 for (i = 0; i < ulen; i++) {
4941 if (copy_to_user(user_prog_tags[i],
4942 prog->aux->func[i]->tag,
4943 BPF_TAG_SIZE))
4944 return -EFAULT;
4945 }
4946 } else {
4947 if (copy_to_user(user_prog_tags[0],
4948 prog->tag, BPF_TAG_SIZE))
4949 return -EFAULT;
4950 }
4951 }
4952
4953done:
4954 if (copy_to_user(uinfo, &info, info_len) ||
4955 put_user(info_len, &uattr->info.info_len))
4956 return -EFAULT;
4957
4958 return 0;
4959}
4960
4961static int bpf_map_get_info_by_fd(struct file *file,
4962 struct bpf_map *map,
4963 const union bpf_attr *attr,
4964 union bpf_attr __user *uattr)
4965{
4966 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4967 struct bpf_map_info info;
4968 u32 info_len = attr->info.info_len;
4969 int err;
4970
4971 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4972 if (err)
4973 return err;
4974 info_len = min_t(u32, sizeof(info), info_len);
4975
4976 memset(&info, 0, sizeof(info));
4977 info.type = map->map_type;
4978 info.id = map->id;
4979 info.key_size = map->key_size;
4980 info.value_size = map->value_size;
4981 info.max_entries = map->max_entries;
4982 info.map_flags = map->map_flags;
4983 info.map_extra = map->map_extra;
4984 memcpy(info.name, map->name, sizeof(map->name));
4985
4986 if (map->btf) {
4987 info.btf_id = btf_obj_id(map->btf);
4988 info.btf_key_type_id = map->btf_key_type_id;
4989 info.btf_value_type_id = map->btf_value_type_id;
4990 }
4991 info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4992 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS)
4993 bpf_map_struct_ops_info_fill(&info, map);
4994
4995 if (bpf_map_is_offloaded(map)) {
4996 err = bpf_map_offload_info_fill(&info, map);
4997 if (err)
4998 return err;
4999 }
5000
5001 if (copy_to_user(uinfo, &info, info_len) ||
5002 put_user(info_len, &uattr->info.info_len))
5003 return -EFAULT;
5004
5005 return 0;
5006}
5007
5008static int bpf_btf_get_info_by_fd(struct file *file,
5009 struct btf *btf,
5010 const union bpf_attr *attr,
5011 union bpf_attr __user *uattr)
5012{
5013 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
5014 u32 info_len = attr->info.info_len;
5015 int err;
5016
5017 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
5018 if (err)
5019 return err;
5020
5021 return btf_get_info_by_fd(btf, attr, uattr);
5022}
5023
5024static int bpf_link_get_info_by_fd(struct file *file,
5025 struct bpf_link *link,
5026 const union bpf_attr *attr,
5027 union bpf_attr __user *uattr)
5028{
5029 struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
5030 struct bpf_link_info info;
5031 u32 info_len = attr->info.info_len;
5032 int err;
5033
5034 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
5035 if (err)
5036 return err;
5037 info_len = min_t(u32, sizeof(info), info_len);
5038
5039 memset(&info, 0, sizeof(info));
5040 if (copy_from_user(&info, uinfo, info_len))
5041 return -EFAULT;
5042
5043 info.type = link->type;
5044 info.id = link->id;
5045 if (link->prog)
5046 info.prog_id = link->prog->aux->id;
5047
5048 if (link->ops->fill_link_info) {
5049 err = link->ops->fill_link_info(link, &info);
5050 if (err)
5051 return err;
5052 }
5053
5054 if (copy_to_user(uinfo, &info, info_len) ||
5055 put_user(info_len, &uattr->info.info_len))
5056 return -EFAULT;
5057
5058 return 0;
5059}
5060
5061
5062#define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
5063
5064static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
5065 union bpf_attr __user *uattr)
5066{
5067 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
5068 return -EINVAL;
5069
5070 CLASS(fd, f)(attr->info.bpf_fd);
5071 if (fd_empty(f))
5072 return -EBADFD;
5073
5074 if (fd_file(f)->f_op == &bpf_prog_fops)
5075 return bpf_prog_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr,
5076 uattr);
5077 else if (fd_file(f)->f_op == &bpf_map_fops)
5078 return bpf_map_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr,
5079 uattr);
5080 else if (fd_file(f)->f_op == &btf_fops)
5081 return bpf_btf_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr, uattr);
5082 else if (fd_file(f)->f_op == &bpf_link_fops || fd_file(f)->f_op == &bpf_link_fops_poll)
5083 return bpf_link_get_info_by_fd(fd_file(f), fd_file(f)->private_data,
5084 attr, uattr);
5085 return -EINVAL;
5086}
5087
5088#define BPF_BTF_LOAD_LAST_FIELD btf_token_fd
5089
5090static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size)
5091{
5092 struct bpf_token *token = NULL;
5093
5094 if (CHECK_ATTR(BPF_BTF_LOAD))
5095 return -EINVAL;
5096
5097 if (attr->btf_flags & ~BPF_F_TOKEN_FD)
5098 return -EINVAL;
5099
5100 if (attr->btf_flags & BPF_F_TOKEN_FD) {
5101 token = bpf_token_get_from_fd(attr->btf_token_fd);
5102 if (IS_ERR(token))
5103 return PTR_ERR(token);
5104 if (!bpf_token_allow_cmd(token, BPF_BTF_LOAD)) {
5105 bpf_token_put(token);
5106 token = NULL;
5107 }
5108 }
5109
5110 if (!bpf_token_capable(token, CAP_BPF)) {
5111 bpf_token_put(token);
5112 return -EPERM;
5113 }
5114
5115 bpf_token_put(token);
5116
5117 return btf_new_fd(attr, uattr, uattr_size);
5118}
5119
5120#define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
5121
5122static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
5123{
5124 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
5125 return -EINVAL;
5126
5127 if (!capable(CAP_SYS_ADMIN))
5128 return -EPERM;
5129
5130 return btf_get_fd_by_id(attr->btf_id);
5131}
5132
5133static int bpf_task_fd_query_copy(const union bpf_attr *attr,
5134 union bpf_attr __user *uattr,
5135 u32 prog_id, u32 fd_type,
5136 const char *buf, u64 probe_offset,
5137 u64 probe_addr)
5138{
5139 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
5140 u32 len = buf ? strlen(buf) : 0, input_len;
5141 int err = 0;
5142
5143 if (put_user(len, &uattr->task_fd_query.buf_len))
5144 return -EFAULT;
5145 input_len = attr->task_fd_query.buf_len;
5146 if (input_len && ubuf) {
5147 if (!len) {
5148 /* nothing to copy, just make ubuf NULL terminated */
5149 char zero = '\0';
5150
5151 if (put_user(zero, ubuf))
5152 return -EFAULT;
5153 } else if (input_len >= len + 1) {
5154 /* ubuf can hold the string with NULL terminator */
5155 if (copy_to_user(ubuf, buf, len + 1))
5156 return -EFAULT;
5157 } else {
5158 /* ubuf cannot hold the string with NULL terminator,
5159 * do a partial copy with NULL terminator.
5160 */
5161 char zero = '\0';
5162
5163 err = -ENOSPC;
5164 if (copy_to_user(ubuf, buf, input_len - 1))
5165 return -EFAULT;
5166 if (put_user(zero, ubuf + input_len - 1))
5167 return -EFAULT;
5168 }
5169 }
5170
5171 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
5172 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
5173 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
5174 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
5175 return -EFAULT;
5176
5177 return err;
5178}
5179
5180#define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
5181
5182static int bpf_task_fd_query(const union bpf_attr *attr,
5183 union bpf_attr __user *uattr)
5184{
5185 pid_t pid = attr->task_fd_query.pid;
5186 u32 fd = attr->task_fd_query.fd;
5187 const struct perf_event *event;
5188 struct task_struct *task;
5189 struct file *file;
5190 int err;
5191
5192 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
5193 return -EINVAL;
5194
5195 if (!capable(CAP_SYS_ADMIN))
5196 return -EPERM;
5197
5198 if (attr->task_fd_query.flags != 0)
5199 return -EINVAL;
5200
5201 rcu_read_lock();
5202 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
5203 rcu_read_unlock();
5204 if (!task)
5205 return -ENOENT;
5206
5207 err = 0;
5208 file = fget_task(task, fd);
5209 put_task_struct(task);
5210 if (!file)
5211 return -EBADF;
5212
5213 if (file->f_op == &bpf_link_fops || file->f_op == &bpf_link_fops_poll) {
5214 struct bpf_link *link = file->private_data;
5215
5216 if (link->ops == &bpf_raw_tp_link_lops) {
5217 struct bpf_raw_tp_link *raw_tp =
5218 container_of(link, struct bpf_raw_tp_link, link);
5219 struct bpf_raw_event_map *btp = raw_tp->btp;
5220
5221 err = bpf_task_fd_query_copy(attr, uattr,
5222 raw_tp->link.prog->aux->id,
5223 BPF_FD_TYPE_RAW_TRACEPOINT,
5224 btp->tp->name, 0, 0);
5225 goto put_file;
5226 }
5227 goto out_not_supp;
5228 }
5229
5230 event = perf_get_event(file);
5231 if (!IS_ERR(event)) {
5232 u64 probe_offset, probe_addr;
5233 u32 prog_id, fd_type;
5234 const char *buf;
5235
5236 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
5237 &buf, &probe_offset,
5238 &probe_addr, NULL);
5239 if (!err)
5240 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
5241 fd_type, buf,
5242 probe_offset,
5243 probe_addr);
5244 goto put_file;
5245 }
5246
5247out_not_supp:
5248 err = -ENOTSUPP;
5249put_file:
5250 fput(file);
5251 return err;
5252}
5253
5254#define BPF_MAP_BATCH_LAST_FIELD batch.flags
5255
5256#define BPF_DO_BATCH(fn, ...) \
5257 do { \
5258 if (!fn) { \
5259 err = -ENOTSUPP; \
5260 goto err_put; \
5261 } \
5262 err = fn(__VA_ARGS__); \
5263 } while (0)
5264
5265static int bpf_map_do_batch(const union bpf_attr *attr,
5266 union bpf_attr __user *uattr,
5267 int cmd)
5268{
5269 bool has_read = cmd == BPF_MAP_LOOKUP_BATCH ||
5270 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
5271 bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
5272 struct bpf_map *map;
5273 int err;
5274
5275 if (CHECK_ATTR(BPF_MAP_BATCH))
5276 return -EINVAL;
5277
5278 CLASS(fd, f)(attr->batch.map_fd);
5279
5280 map = __bpf_map_get(f);
5281 if (IS_ERR(map))
5282 return PTR_ERR(map);
5283 if (has_write)
5284 bpf_map_write_active_inc(map);
5285 if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
5286 err = -EPERM;
5287 goto err_put;
5288 }
5289 if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
5290 err = -EPERM;
5291 goto err_put;
5292 }
5293
5294 if (cmd == BPF_MAP_LOOKUP_BATCH)
5295 BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
5296 else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
5297 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
5298 else if (cmd == BPF_MAP_UPDATE_BATCH)
5299 BPF_DO_BATCH(map->ops->map_update_batch, map, fd_file(f), attr, uattr);
5300 else
5301 BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
5302err_put:
5303 if (has_write) {
5304 maybe_wait_bpf_programs(map);
5305 bpf_map_write_active_dec(map);
5306 }
5307 return err;
5308}
5309
5310#define BPF_LINK_CREATE_LAST_FIELD link_create.uprobe_multi.pid
5311static int link_create(union bpf_attr *attr, bpfptr_t uattr)
5312{
5313 struct bpf_prog *prog;
5314 int ret;
5315
5316 if (CHECK_ATTR(BPF_LINK_CREATE))
5317 return -EINVAL;
5318
5319 if (attr->link_create.attach_type == BPF_STRUCT_OPS)
5320 return bpf_struct_ops_link_create(attr);
5321
5322 prog = bpf_prog_get(attr->link_create.prog_fd);
5323 if (IS_ERR(prog))
5324 return PTR_ERR(prog);
5325
5326 ret = bpf_prog_attach_check_attach_type(prog,
5327 attr->link_create.attach_type);
5328 if (ret)
5329 goto out;
5330
5331 switch (prog->type) {
5332 case BPF_PROG_TYPE_CGROUP_SKB:
5333 case BPF_PROG_TYPE_CGROUP_SOCK:
5334 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
5335 case BPF_PROG_TYPE_SOCK_OPS:
5336 case BPF_PROG_TYPE_CGROUP_DEVICE:
5337 case BPF_PROG_TYPE_CGROUP_SYSCTL:
5338 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
5339 ret = cgroup_bpf_link_attach(attr, prog);
5340 break;
5341 case BPF_PROG_TYPE_EXT:
5342 ret = bpf_tracing_prog_attach(prog,
5343 attr->link_create.target_fd,
5344 attr->link_create.target_btf_id,
5345 attr->link_create.tracing.cookie);
5346 break;
5347 case BPF_PROG_TYPE_LSM:
5348 case BPF_PROG_TYPE_TRACING:
5349 if (attr->link_create.attach_type != prog->expected_attach_type) {
5350 ret = -EINVAL;
5351 goto out;
5352 }
5353 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
5354 ret = bpf_raw_tp_link_attach(prog, NULL, attr->link_create.tracing.cookie);
5355 else if (prog->expected_attach_type == BPF_TRACE_ITER)
5356 ret = bpf_iter_link_attach(attr, uattr, prog);
5357 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
5358 ret = cgroup_bpf_link_attach(attr, prog);
5359 else
5360 ret = bpf_tracing_prog_attach(prog,
5361 attr->link_create.target_fd,
5362 attr->link_create.target_btf_id,
5363 attr->link_create.tracing.cookie);
5364 break;
5365 case BPF_PROG_TYPE_FLOW_DISSECTOR:
5366 case BPF_PROG_TYPE_SK_LOOKUP:
5367 ret = netns_bpf_link_create(attr, prog);
5368 break;
5369 case BPF_PROG_TYPE_SK_MSG:
5370 case BPF_PROG_TYPE_SK_SKB:
5371 ret = sock_map_link_create(attr, prog);
5372 break;
5373#ifdef CONFIG_NET
5374 case BPF_PROG_TYPE_XDP:
5375 ret = bpf_xdp_link_attach(attr, prog);
5376 break;
5377 case BPF_PROG_TYPE_SCHED_CLS:
5378 if (attr->link_create.attach_type == BPF_TCX_INGRESS ||
5379 attr->link_create.attach_type == BPF_TCX_EGRESS)
5380 ret = tcx_link_attach(attr, prog);
5381 else
5382 ret = netkit_link_attach(attr, prog);
5383 break;
5384 case BPF_PROG_TYPE_NETFILTER:
5385 ret = bpf_nf_link_attach(attr, prog);
5386 break;
5387#endif
5388 case BPF_PROG_TYPE_PERF_EVENT:
5389 case BPF_PROG_TYPE_TRACEPOINT:
5390 ret = bpf_perf_link_attach(attr, prog);
5391 break;
5392 case BPF_PROG_TYPE_KPROBE:
5393 if (attr->link_create.attach_type == BPF_PERF_EVENT)
5394 ret = bpf_perf_link_attach(attr, prog);
5395 else if (attr->link_create.attach_type == BPF_TRACE_KPROBE_MULTI ||
5396 attr->link_create.attach_type == BPF_TRACE_KPROBE_SESSION)
5397 ret = bpf_kprobe_multi_link_attach(attr, prog);
5398 else if (attr->link_create.attach_type == BPF_TRACE_UPROBE_MULTI ||
5399 attr->link_create.attach_type == BPF_TRACE_UPROBE_SESSION)
5400 ret = bpf_uprobe_multi_link_attach(attr, prog);
5401 break;
5402 default:
5403 ret = -EINVAL;
5404 }
5405
5406out:
5407 if (ret < 0)
5408 bpf_prog_put(prog);
5409 return ret;
5410}
5411
5412static int link_update_map(struct bpf_link *link, union bpf_attr *attr)
5413{
5414 struct bpf_map *new_map, *old_map = NULL;
5415 int ret;
5416
5417 new_map = bpf_map_get(attr->link_update.new_map_fd);
5418 if (IS_ERR(new_map))
5419 return PTR_ERR(new_map);
5420
5421 if (attr->link_update.flags & BPF_F_REPLACE) {
5422 old_map = bpf_map_get(attr->link_update.old_map_fd);
5423 if (IS_ERR(old_map)) {
5424 ret = PTR_ERR(old_map);
5425 goto out_put;
5426 }
5427 } else if (attr->link_update.old_map_fd) {
5428 ret = -EINVAL;
5429 goto out_put;
5430 }
5431
5432 ret = link->ops->update_map(link, new_map, old_map);
5433
5434 if (old_map)
5435 bpf_map_put(old_map);
5436out_put:
5437 bpf_map_put(new_map);
5438 return ret;
5439}
5440
5441#define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
5442
5443static int link_update(union bpf_attr *attr)
5444{
5445 struct bpf_prog *old_prog = NULL, *new_prog;
5446 struct bpf_link *link;
5447 u32 flags;
5448 int ret;
5449
5450 if (CHECK_ATTR(BPF_LINK_UPDATE))
5451 return -EINVAL;
5452
5453 flags = attr->link_update.flags;
5454 if (flags & ~BPF_F_REPLACE)
5455 return -EINVAL;
5456
5457 link = bpf_link_get_from_fd(attr->link_update.link_fd);
5458 if (IS_ERR(link))
5459 return PTR_ERR(link);
5460
5461 if (link->ops->update_map) {
5462 ret = link_update_map(link, attr);
5463 goto out_put_link;
5464 }
5465
5466 new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
5467 if (IS_ERR(new_prog)) {
5468 ret = PTR_ERR(new_prog);
5469 goto out_put_link;
5470 }
5471
5472 if (flags & BPF_F_REPLACE) {
5473 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
5474 if (IS_ERR(old_prog)) {
5475 ret = PTR_ERR(old_prog);
5476 old_prog = NULL;
5477 goto out_put_progs;
5478 }
5479 } else if (attr->link_update.old_prog_fd) {
5480 ret = -EINVAL;
5481 goto out_put_progs;
5482 }
5483
5484 if (link->ops->update_prog)
5485 ret = link->ops->update_prog(link, new_prog, old_prog);
5486 else
5487 ret = -EINVAL;
5488
5489out_put_progs:
5490 if (old_prog)
5491 bpf_prog_put(old_prog);
5492 if (ret)
5493 bpf_prog_put(new_prog);
5494out_put_link:
5495 bpf_link_put_direct(link);
5496 return ret;
5497}
5498
5499#define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
5500
5501static int link_detach(union bpf_attr *attr)
5502{
5503 struct bpf_link *link;
5504 int ret;
5505
5506 if (CHECK_ATTR(BPF_LINK_DETACH))
5507 return -EINVAL;
5508
5509 link = bpf_link_get_from_fd(attr->link_detach.link_fd);
5510 if (IS_ERR(link))
5511 return PTR_ERR(link);
5512
5513 if (link->ops->detach)
5514 ret = link->ops->detach(link);
5515 else
5516 ret = -EOPNOTSUPP;
5517
5518 bpf_link_put_direct(link);
5519 return ret;
5520}
5521
5522struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
5523{
5524 return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
5525}
5526EXPORT_SYMBOL(bpf_link_inc_not_zero);
5527
5528struct bpf_link *bpf_link_by_id(u32 id)
5529{
5530 struct bpf_link *link;
5531
5532 if (!id)
5533 return ERR_PTR(-ENOENT);
5534
5535 spin_lock_bh(&link_idr_lock);
5536 /* before link is "settled", ID is 0, pretend it doesn't exist yet */
5537 link = idr_find(&link_idr, id);
5538 if (link) {
5539 if (link->id)
5540 link = bpf_link_inc_not_zero(link);
5541 else
5542 link = ERR_PTR(-EAGAIN);
5543 } else {
5544 link = ERR_PTR(-ENOENT);
5545 }
5546 spin_unlock_bh(&link_idr_lock);
5547 return link;
5548}
5549
5550struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
5551{
5552 struct bpf_link *link;
5553
5554 spin_lock_bh(&link_idr_lock);
5555again:
5556 link = idr_get_next(&link_idr, id);
5557 if (link) {
5558 link = bpf_link_inc_not_zero(link);
5559 if (IS_ERR(link)) {
5560 (*id)++;
5561 goto again;
5562 }
5563 }
5564 spin_unlock_bh(&link_idr_lock);
5565
5566 return link;
5567}
5568
5569#define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
5570
5571static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
5572{
5573 struct bpf_link *link;
5574 u32 id = attr->link_id;
5575 int fd;
5576
5577 if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
5578 return -EINVAL;
5579
5580 if (!capable(CAP_SYS_ADMIN))
5581 return -EPERM;
5582
5583 link = bpf_link_by_id(id);
5584 if (IS_ERR(link))
5585 return PTR_ERR(link);
5586
5587 fd = bpf_link_new_fd(link);
5588 if (fd < 0)
5589 bpf_link_put_direct(link);
5590
5591 return fd;
5592}
5593
5594DEFINE_MUTEX(bpf_stats_enabled_mutex);
5595
5596static int bpf_stats_release(struct inode *inode, struct file *file)
5597{
5598 mutex_lock(&bpf_stats_enabled_mutex);
5599 static_key_slow_dec(&bpf_stats_enabled_key.key);
5600 mutex_unlock(&bpf_stats_enabled_mutex);
5601 return 0;
5602}
5603
5604static const struct file_operations bpf_stats_fops = {
5605 .release = bpf_stats_release,
5606};
5607
5608static int bpf_enable_runtime_stats(void)
5609{
5610 int fd;
5611
5612 mutex_lock(&bpf_stats_enabled_mutex);
5613
5614 /* Set a very high limit to avoid overflow */
5615 if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
5616 mutex_unlock(&bpf_stats_enabled_mutex);
5617 return -EBUSY;
5618 }
5619
5620 fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
5621 if (fd >= 0)
5622 static_key_slow_inc(&bpf_stats_enabled_key.key);
5623
5624 mutex_unlock(&bpf_stats_enabled_mutex);
5625 return fd;
5626}
5627
5628#define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
5629
5630static int bpf_enable_stats(union bpf_attr *attr)
5631{
5632
5633 if (CHECK_ATTR(BPF_ENABLE_STATS))
5634 return -EINVAL;
5635
5636 if (!capable(CAP_SYS_ADMIN))
5637 return -EPERM;
5638
5639 switch (attr->enable_stats.type) {
5640 case BPF_STATS_RUN_TIME:
5641 return bpf_enable_runtime_stats();
5642 default:
5643 break;
5644 }
5645 return -EINVAL;
5646}
5647
5648#define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
5649
5650static int bpf_iter_create(union bpf_attr *attr)
5651{
5652 struct bpf_link *link;
5653 int err;
5654
5655 if (CHECK_ATTR(BPF_ITER_CREATE))
5656 return -EINVAL;
5657
5658 if (attr->iter_create.flags)
5659 return -EINVAL;
5660
5661 link = bpf_link_get_from_fd(attr->iter_create.link_fd);
5662 if (IS_ERR(link))
5663 return PTR_ERR(link);
5664
5665 err = bpf_iter_new_fd(link);
5666 bpf_link_put_direct(link);
5667
5668 return err;
5669}
5670
5671#define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
5672
5673static int bpf_prog_bind_map(union bpf_attr *attr)
5674{
5675 struct bpf_prog *prog;
5676 struct bpf_map *map;
5677 struct bpf_map **used_maps_old, **used_maps_new;
5678 int i, ret = 0;
5679
5680 if (CHECK_ATTR(BPF_PROG_BIND_MAP))
5681 return -EINVAL;
5682
5683 if (attr->prog_bind_map.flags)
5684 return -EINVAL;
5685
5686 prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
5687 if (IS_ERR(prog))
5688 return PTR_ERR(prog);
5689
5690 map = bpf_map_get(attr->prog_bind_map.map_fd);
5691 if (IS_ERR(map)) {
5692 ret = PTR_ERR(map);
5693 goto out_prog_put;
5694 }
5695
5696 mutex_lock(&prog->aux->used_maps_mutex);
5697
5698 used_maps_old = prog->aux->used_maps;
5699
5700 for (i = 0; i < prog->aux->used_map_cnt; i++)
5701 if (used_maps_old[i] == map) {
5702 bpf_map_put(map);
5703 goto out_unlock;
5704 }
5705
5706 used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
5707 sizeof(used_maps_new[0]),
5708 GFP_KERNEL);
5709 if (!used_maps_new) {
5710 ret = -ENOMEM;
5711 goto out_unlock;
5712 }
5713
5714 /* The bpf program will not access the bpf map, but for the sake of
5715 * simplicity, increase sleepable_refcnt for sleepable program as well.
5716 */
5717 if (prog->sleepable)
5718 atomic64_inc(&map->sleepable_refcnt);
5719 memcpy(used_maps_new, used_maps_old,
5720 sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
5721 used_maps_new[prog->aux->used_map_cnt] = map;
5722
5723 prog->aux->used_map_cnt++;
5724 prog->aux->used_maps = used_maps_new;
5725
5726 kfree(used_maps_old);
5727
5728out_unlock:
5729 mutex_unlock(&prog->aux->used_maps_mutex);
5730
5731 if (ret)
5732 bpf_map_put(map);
5733out_prog_put:
5734 bpf_prog_put(prog);
5735 return ret;
5736}
5737
5738#define BPF_TOKEN_CREATE_LAST_FIELD token_create.bpffs_fd
5739
5740static int token_create(union bpf_attr *attr)
5741{
5742 if (CHECK_ATTR(BPF_TOKEN_CREATE))
5743 return -EINVAL;
5744
5745 /* no flags are supported yet */
5746 if (attr->token_create.flags)
5747 return -EINVAL;
5748
5749 return bpf_token_create(attr);
5750}
5751
5752static int __sys_bpf(enum bpf_cmd cmd, bpfptr_t uattr, unsigned int size)
5753{
5754 union bpf_attr attr;
5755 int err;
5756
5757 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
5758 if (err)
5759 return err;
5760 size = min_t(u32, size, sizeof(attr));
5761
5762 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
5763 memset(&attr, 0, sizeof(attr));
5764 if (copy_from_bpfptr(&attr, uattr, size) != 0)
5765 return -EFAULT;
5766
5767 err = security_bpf(cmd, &attr, size);
5768 if (err < 0)
5769 return err;
5770
5771 switch (cmd) {
5772 case BPF_MAP_CREATE:
5773 err = map_create(&attr);
5774 break;
5775 case BPF_MAP_LOOKUP_ELEM:
5776 err = map_lookup_elem(&attr);
5777 break;
5778 case BPF_MAP_UPDATE_ELEM:
5779 err = map_update_elem(&attr, uattr);
5780 break;
5781 case BPF_MAP_DELETE_ELEM:
5782 err = map_delete_elem(&attr, uattr);
5783 break;
5784 case BPF_MAP_GET_NEXT_KEY:
5785 err = map_get_next_key(&attr);
5786 break;
5787 case BPF_MAP_FREEZE:
5788 err = map_freeze(&attr);
5789 break;
5790 case BPF_PROG_LOAD:
5791 err = bpf_prog_load(&attr, uattr, size);
5792 break;
5793 case BPF_OBJ_PIN:
5794 err = bpf_obj_pin(&attr);
5795 break;
5796 case BPF_OBJ_GET:
5797 err = bpf_obj_get(&attr);
5798 break;
5799 case BPF_PROG_ATTACH:
5800 err = bpf_prog_attach(&attr);
5801 break;
5802 case BPF_PROG_DETACH:
5803 err = bpf_prog_detach(&attr);
5804 break;
5805 case BPF_PROG_QUERY:
5806 err = bpf_prog_query(&attr, uattr.user);
5807 break;
5808 case BPF_PROG_TEST_RUN:
5809 err = bpf_prog_test_run(&attr, uattr.user);
5810 break;
5811 case BPF_PROG_GET_NEXT_ID:
5812 err = bpf_obj_get_next_id(&attr, uattr.user,
5813 &prog_idr, &prog_idr_lock);
5814 break;
5815 case BPF_MAP_GET_NEXT_ID:
5816 err = bpf_obj_get_next_id(&attr, uattr.user,
5817 &map_idr, &map_idr_lock);
5818 break;
5819 case BPF_BTF_GET_NEXT_ID:
5820 err = bpf_obj_get_next_id(&attr, uattr.user,
5821 &btf_idr, &btf_idr_lock);
5822 break;
5823 case BPF_PROG_GET_FD_BY_ID:
5824 err = bpf_prog_get_fd_by_id(&attr);
5825 break;
5826 case BPF_MAP_GET_FD_BY_ID:
5827 err = bpf_map_get_fd_by_id(&attr);
5828 break;
5829 case BPF_OBJ_GET_INFO_BY_FD:
5830 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5831 break;
5832 case BPF_RAW_TRACEPOINT_OPEN:
5833 err = bpf_raw_tracepoint_open(&attr);
5834 break;
5835 case BPF_BTF_LOAD:
5836 err = bpf_btf_load(&attr, uattr, size);
5837 break;
5838 case BPF_BTF_GET_FD_BY_ID:
5839 err = bpf_btf_get_fd_by_id(&attr);
5840 break;
5841 case BPF_TASK_FD_QUERY:
5842 err = bpf_task_fd_query(&attr, uattr.user);
5843 break;
5844 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5845 err = map_lookup_and_delete_elem(&attr);
5846 break;
5847 case BPF_MAP_LOOKUP_BATCH:
5848 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5849 break;
5850 case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5851 err = bpf_map_do_batch(&attr, uattr.user,
5852 BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5853 break;
5854 case BPF_MAP_UPDATE_BATCH:
5855 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5856 break;
5857 case BPF_MAP_DELETE_BATCH:
5858 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5859 break;
5860 case BPF_LINK_CREATE:
5861 err = link_create(&attr, uattr);
5862 break;
5863 case BPF_LINK_UPDATE:
5864 err = link_update(&attr);
5865 break;
5866 case BPF_LINK_GET_FD_BY_ID:
5867 err = bpf_link_get_fd_by_id(&attr);
5868 break;
5869 case BPF_LINK_GET_NEXT_ID:
5870 err = bpf_obj_get_next_id(&attr, uattr.user,
5871 &link_idr, &link_idr_lock);
5872 break;
5873 case BPF_ENABLE_STATS:
5874 err = bpf_enable_stats(&attr);
5875 break;
5876 case BPF_ITER_CREATE:
5877 err = bpf_iter_create(&attr);
5878 break;
5879 case BPF_LINK_DETACH:
5880 err = link_detach(&attr);
5881 break;
5882 case BPF_PROG_BIND_MAP:
5883 err = bpf_prog_bind_map(&attr);
5884 break;
5885 case BPF_TOKEN_CREATE:
5886 err = token_create(&attr);
5887 break;
5888 default:
5889 err = -EINVAL;
5890 break;
5891 }
5892
5893 return err;
5894}
5895
5896SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5897{
5898 return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5899}
5900
5901static bool syscall_prog_is_valid_access(int off, int size,
5902 enum bpf_access_type type,
5903 const struct bpf_prog *prog,
5904 struct bpf_insn_access_aux *info)
5905{
5906 if (off < 0 || off >= U16_MAX)
5907 return false;
5908 if (off % size != 0)
5909 return false;
5910 return true;
5911}
5912
5913BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5914{
5915 switch (cmd) {
5916 case BPF_MAP_CREATE:
5917 case BPF_MAP_DELETE_ELEM:
5918 case BPF_MAP_UPDATE_ELEM:
5919 case BPF_MAP_FREEZE:
5920 case BPF_MAP_GET_FD_BY_ID:
5921 case BPF_PROG_LOAD:
5922 case BPF_BTF_LOAD:
5923 case BPF_LINK_CREATE:
5924 case BPF_RAW_TRACEPOINT_OPEN:
5925 break;
5926 default:
5927 return -EINVAL;
5928 }
5929 return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5930}
5931
5932
5933/* To shut up -Wmissing-prototypes.
5934 * This function is used by the kernel light skeleton
5935 * to load bpf programs when modules are loaded or during kernel boot.
5936 * See tools/lib/bpf/skel_internal.h
5937 */
5938int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5939
5940int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5941{
5942 struct bpf_prog * __maybe_unused prog;
5943 struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5944
5945 switch (cmd) {
5946#ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5947 case BPF_PROG_TEST_RUN:
5948 if (attr->test.data_in || attr->test.data_out ||
5949 attr->test.ctx_out || attr->test.duration ||
5950 attr->test.repeat || attr->test.flags)
5951 return -EINVAL;
5952
5953 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5954 if (IS_ERR(prog))
5955 return PTR_ERR(prog);
5956
5957 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5958 attr->test.ctx_size_in > U16_MAX) {
5959 bpf_prog_put(prog);
5960 return -EINVAL;
5961 }
5962
5963 run_ctx.bpf_cookie = 0;
5964 if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5965 /* recursion detected */
5966 __bpf_prog_exit_sleepable_recur(prog, 0, &run_ctx);
5967 bpf_prog_put(prog);
5968 return -EBUSY;
5969 }
5970 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5971 __bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5972 &run_ctx);
5973 bpf_prog_put(prog);
5974 return 0;
5975#endif
5976 default:
5977 return ____bpf_sys_bpf(cmd, attr, size);
5978 }
5979}
5980EXPORT_SYMBOL(kern_sys_bpf);
5981
5982static const struct bpf_func_proto bpf_sys_bpf_proto = {
5983 .func = bpf_sys_bpf,
5984 .gpl_only = false,
5985 .ret_type = RET_INTEGER,
5986 .arg1_type = ARG_ANYTHING,
5987 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5988 .arg3_type = ARG_CONST_SIZE,
5989};
5990
5991const struct bpf_func_proto * __weak
5992tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5993{
5994 return bpf_base_func_proto(func_id, prog);
5995}
5996
5997BPF_CALL_1(bpf_sys_close, u32, fd)
5998{
5999 /* When bpf program calls this helper there should not be
6000 * an fdget() without matching completed fdput().
6001 * This helper is allowed in the following callchain only:
6002 * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
6003 */
6004 return close_fd(fd);
6005}
6006
6007static const struct bpf_func_proto bpf_sys_close_proto = {
6008 .func = bpf_sys_close,
6009 .gpl_only = false,
6010 .ret_type = RET_INTEGER,
6011 .arg1_type = ARG_ANYTHING,
6012};
6013
6014BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
6015{
6016 *res = 0;
6017 if (flags)
6018 return -EINVAL;
6019
6020 if (name_sz <= 1 || name[name_sz - 1])
6021 return -EINVAL;
6022
6023 if (!bpf_dump_raw_ok(current_cred()))
6024 return -EPERM;
6025
6026 *res = kallsyms_lookup_name(name);
6027 return *res ? 0 : -ENOENT;
6028}
6029
6030static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
6031 .func = bpf_kallsyms_lookup_name,
6032 .gpl_only = false,
6033 .ret_type = RET_INTEGER,
6034 .arg1_type = ARG_PTR_TO_MEM,
6035 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
6036 .arg3_type = ARG_ANYTHING,
6037 .arg4_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_WRITE | MEM_ALIGNED,
6038 .arg4_size = sizeof(u64),
6039};
6040
6041static const struct bpf_func_proto *
6042syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6043{
6044 switch (func_id) {
6045 case BPF_FUNC_sys_bpf:
6046 return !bpf_token_capable(prog->aux->token, CAP_PERFMON)
6047 ? NULL : &bpf_sys_bpf_proto;
6048 case BPF_FUNC_btf_find_by_name_kind:
6049 return &bpf_btf_find_by_name_kind_proto;
6050 case BPF_FUNC_sys_close:
6051 return &bpf_sys_close_proto;
6052 case BPF_FUNC_kallsyms_lookup_name:
6053 return &bpf_kallsyms_lookup_name_proto;
6054 default:
6055 return tracing_prog_func_proto(func_id, prog);
6056 }
6057}
6058
6059const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
6060 .get_func_proto = syscall_prog_func_proto,
6061 .is_valid_access = syscall_prog_is_valid_access,
6062};
6063
6064const struct bpf_prog_ops bpf_syscall_prog_ops = {
6065 .test_run = bpf_prog_test_run_syscall,
6066};
6067
6068#ifdef CONFIG_SYSCTL
6069static int bpf_stats_handler(const struct ctl_table *table, int write,
6070 void *buffer, size_t *lenp, loff_t *ppos)
6071{
6072 struct static_key *key = (struct static_key *)table->data;
6073 static int saved_val;
6074 int val, ret;
6075 struct ctl_table tmp = {
6076 .data = &val,
6077 .maxlen = sizeof(val),
6078 .mode = table->mode,
6079 .extra1 = SYSCTL_ZERO,
6080 .extra2 = SYSCTL_ONE,
6081 };
6082
6083 if (write && !capable(CAP_SYS_ADMIN))
6084 return -EPERM;
6085
6086 mutex_lock(&bpf_stats_enabled_mutex);
6087 val = saved_val;
6088 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
6089 if (write && !ret && val != saved_val) {
6090 if (val)
6091 static_key_slow_inc(key);
6092 else
6093 static_key_slow_dec(key);
6094 saved_val = val;
6095 }
6096 mutex_unlock(&bpf_stats_enabled_mutex);
6097 return ret;
6098}
6099
6100void __weak unpriv_ebpf_notify(int new_state)
6101{
6102}
6103
6104static int bpf_unpriv_handler(const struct ctl_table *table, int write,
6105 void *buffer, size_t *lenp, loff_t *ppos)
6106{
6107 int ret, unpriv_enable = *(int *)table->data;
6108 bool locked_state = unpriv_enable == 1;
6109 struct ctl_table tmp = *table;
6110
6111 if (write && !capable(CAP_SYS_ADMIN))
6112 return -EPERM;
6113
6114 tmp.data = &unpriv_enable;
6115 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
6116 if (write && !ret) {
6117 if (locked_state && unpriv_enable != 1)
6118 return -EPERM;
6119 *(int *)table->data = unpriv_enable;
6120 }
6121
6122 if (write)
6123 unpriv_ebpf_notify(unpriv_enable);
6124
6125 return ret;
6126}
6127
6128static struct ctl_table bpf_syscall_table[] = {
6129 {
6130 .procname = "unprivileged_bpf_disabled",
6131 .data = &sysctl_unprivileged_bpf_disabled,
6132 .maxlen = sizeof(sysctl_unprivileged_bpf_disabled),
6133 .mode = 0644,
6134 .proc_handler = bpf_unpriv_handler,
6135 .extra1 = SYSCTL_ZERO,
6136 .extra2 = SYSCTL_TWO,
6137 },
6138 {
6139 .procname = "bpf_stats_enabled",
6140 .data = &bpf_stats_enabled_key.key,
6141 .mode = 0644,
6142 .proc_handler = bpf_stats_handler,
6143 },
6144};
6145
6146static int __init bpf_syscall_sysctl_init(void)
6147{
6148 register_sysctl_init("kernel", bpf_syscall_table);
6149 return 0;
6150}
6151late_initcall(bpf_syscall_sysctl_init);
6152#endif /* CONFIG_SYSCTL */
1// SPDX-License-Identifier: GPL-2.0-only
2/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 */
4#include <linux/bpf.h>
5#include <linux/bpf_trace.h>
6#include <linux/bpf_lirc.h>
7#include <linux/bpf_verifier.h>
8#include <linux/btf.h>
9#include <linux/syscalls.h>
10#include <linux/slab.h>
11#include <linux/sched/signal.h>
12#include <linux/vmalloc.h>
13#include <linux/mmzone.h>
14#include <linux/anon_inodes.h>
15#include <linux/fdtable.h>
16#include <linux/file.h>
17#include <linux/fs.h>
18#include <linux/license.h>
19#include <linux/filter.h>
20#include <linux/kernel.h>
21#include <linux/idr.h>
22#include <linux/cred.h>
23#include <linux/timekeeping.h>
24#include <linux/ctype.h>
25#include <linux/nospec.h>
26#include <linux/audit.h>
27#include <uapi/linux/btf.h>
28#include <linux/pgtable.h>
29#include <linux/bpf_lsm.h>
30#include <linux/poll.h>
31#include <linux/bpf-netns.h>
32#include <linux/rcupdate_trace.h>
33#include <linux/memcontrol.h>
34
35#define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
36 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
37 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
38#define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
39#define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
40#define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
41 IS_FD_HASH(map))
42
43#define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
44
45DEFINE_PER_CPU(int, bpf_prog_active);
46static DEFINE_IDR(prog_idr);
47static DEFINE_SPINLOCK(prog_idr_lock);
48static DEFINE_IDR(map_idr);
49static DEFINE_SPINLOCK(map_idr_lock);
50static DEFINE_IDR(link_idr);
51static DEFINE_SPINLOCK(link_idr_lock);
52
53int sysctl_unprivileged_bpf_disabled __read_mostly =
54 IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
55
56static const struct bpf_map_ops * const bpf_map_types[] = {
57#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
58#define BPF_MAP_TYPE(_id, _ops) \
59 [_id] = &_ops,
60#define BPF_LINK_TYPE(_id, _name)
61#include <linux/bpf_types.h>
62#undef BPF_PROG_TYPE
63#undef BPF_MAP_TYPE
64#undef BPF_LINK_TYPE
65};
66
67/*
68 * If we're handed a bigger struct than we know of, ensure all the unknown bits
69 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
70 * we don't know about yet.
71 *
72 * There is a ToCToU between this function call and the following
73 * copy_from_user() call. However, this is not a concern since this function is
74 * meant to be a future-proofing of bits.
75 */
76int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
77 size_t expected_size,
78 size_t actual_size)
79{
80 int res;
81
82 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
83 return -E2BIG;
84
85 if (actual_size <= expected_size)
86 return 0;
87
88 if (uaddr.is_kernel)
89 res = memchr_inv(uaddr.kernel + expected_size, 0,
90 actual_size - expected_size) == NULL;
91 else
92 res = check_zeroed_user(uaddr.user + expected_size,
93 actual_size - expected_size);
94 if (res < 0)
95 return res;
96 return res ? 0 : -E2BIG;
97}
98
99const struct bpf_map_ops bpf_map_offload_ops = {
100 .map_meta_equal = bpf_map_meta_equal,
101 .map_alloc = bpf_map_offload_map_alloc,
102 .map_free = bpf_map_offload_map_free,
103 .map_check_btf = map_check_no_btf,
104};
105
106static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
107{
108 const struct bpf_map_ops *ops;
109 u32 type = attr->map_type;
110 struct bpf_map *map;
111 int err;
112
113 if (type >= ARRAY_SIZE(bpf_map_types))
114 return ERR_PTR(-EINVAL);
115 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
116 ops = bpf_map_types[type];
117 if (!ops)
118 return ERR_PTR(-EINVAL);
119
120 if (ops->map_alloc_check) {
121 err = ops->map_alloc_check(attr);
122 if (err)
123 return ERR_PTR(err);
124 }
125 if (attr->map_ifindex)
126 ops = &bpf_map_offload_ops;
127 map = ops->map_alloc(attr);
128 if (IS_ERR(map))
129 return map;
130 map->ops = ops;
131 map->map_type = type;
132 return map;
133}
134
135static u32 bpf_map_value_size(const struct bpf_map *map)
136{
137 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
138 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
139 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
140 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
141 return round_up(map->value_size, 8) * num_possible_cpus();
142 else if (IS_FD_MAP(map))
143 return sizeof(u32);
144 else
145 return map->value_size;
146}
147
148static void maybe_wait_bpf_programs(struct bpf_map *map)
149{
150 /* Wait for any running BPF programs to complete so that
151 * userspace, when we return to it, knows that all programs
152 * that could be running use the new map value.
153 */
154 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
155 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
156 synchronize_rcu();
157}
158
159static int bpf_map_update_value(struct bpf_map *map, struct fd f, void *key,
160 void *value, __u64 flags)
161{
162 int err;
163
164 /* Need to create a kthread, thus must support schedule */
165 if (bpf_map_is_dev_bound(map)) {
166 return bpf_map_offload_update_elem(map, key, value, flags);
167 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
168 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
169 return map->ops->map_update_elem(map, key, value, flags);
170 } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
171 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
172 return sock_map_update_elem_sys(map, key, value, flags);
173 } else if (IS_FD_PROG_ARRAY(map)) {
174 return bpf_fd_array_map_update_elem(map, f.file, key, value,
175 flags);
176 }
177
178 bpf_disable_instrumentation();
179 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
180 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
181 err = bpf_percpu_hash_update(map, key, value, flags);
182 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
183 err = bpf_percpu_array_update(map, key, value, flags);
184 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
185 err = bpf_percpu_cgroup_storage_update(map, key, value,
186 flags);
187 } else if (IS_FD_ARRAY(map)) {
188 rcu_read_lock();
189 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
190 flags);
191 rcu_read_unlock();
192 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
193 rcu_read_lock();
194 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
195 flags);
196 rcu_read_unlock();
197 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
198 /* rcu_read_lock() is not needed */
199 err = bpf_fd_reuseport_array_update_elem(map, key, value,
200 flags);
201 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
202 map->map_type == BPF_MAP_TYPE_STACK) {
203 err = map->ops->map_push_elem(map, value, flags);
204 } else {
205 rcu_read_lock();
206 err = map->ops->map_update_elem(map, key, value, flags);
207 rcu_read_unlock();
208 }
209 bpf_enable_instrumentation();
210 maybe_wait_bpf_programs(map);
211
212 return err;
213}
214
215static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
216 __u64 flags)
217{
218 void *ptr;
219 int err;
220
221 if (bpf_map_is_dev_bound(map))
222 return bpf_map_offload_lookup_elem(map, key, value);
223
224 bpf_disable_instrumentation();
225 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
226 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
227 err = bpf_percpu_hash_copy(map, key, value);
228 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
229 err = bpf_percpu_array_copy(map, key, value);
230 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
231 err = bpf_percpu_cgroup_storage_copy(map, key, value);
232 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
233 err = bpf_stackmap_copy(map, key, value);
234 } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
235 err = bpf_fd_array_map_lookup_elem(map, key, value);
236 } else if (IS_FD_HASH(map)) {
237 err = bpf_fd_htab_map_lookup_elem(map, key, value);
238 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
239 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
240 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
241 map->map_type == BPF_MAP_TYPE_STACK) {
242 err = map->ops->map_peek_elem(map, value);
243 } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
244 /* struct_ops map requires directly updating "value" */
245 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
246 } else {
247 rcu_read_lock();
248 if (map->ops->map_lookup_elem_sys_only)
249 ptr = map->ops->map_lookup_elem_sys_only(map, key);
250 else
251 ptr = map->ops->map_lookup_elem(map, key);
252 if (IS_ERR(ptr)) {
253 err = PTR_ERR(ptr);
254 } else if (!ptr) {
255 err = -ENOENT;
256 } else {
257 err = 0;
258 if (flags & BPF_F_LOCK)
259 /* lock 'ptr' and copy everything but lock */
260 copy_map_value_locked(map, value, ptr, true);
261 else
262 copy_map_value(map, value, ptr);
263 /* mask lock, since value wasn't zero inited */
264 check_and_init_map_lock(map, value);
265 }
266 rcu_read_unlock();
267 }
268
269 bpf_enable_instrumentation();
270 maybe_wait_bpf_programs(map);
271
272 return err;
273}
274
275/* Please, do not use this function outside from the map creation path
276 * (e.g. in map update path) without taking care of setting the active
277 * memory cgroup (see at bpf_map_kmalloc_node() for example).
278 */
279static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
280{
281 /* We really just want to fail instead of triggering OOM killer
282 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
283 * which is used for lower order allocation requests.
284 *
285 * It has been observed that higher order allocation requests done by
286 * vmalloc with __GFP_NORETRY being set might fail due to not trying
287 * to reclaim memory from the page cache, thus we set
288 * __GFP_RETRY_MAYFAIL to avoid such situations.
289 */
290
291 const gfp_t gfp = __GFP_NOWARN | __GFP_ZERO | __GFP_ACCOUNT;
292 unsigned int flags = 0;
293 unsigned long align = 1;
294 void *area;
295
296 if (size >= SIZE_MAX)
297 return NULL;
298
299 /* kmalloc()'ed memory can't be mmap()'ed */
300 if (mmapable) {
301 BUG_ON(!PAGE_ALIGNED(size));
302 align = SHMLBA;
303 flags = VM_USERMAP;
304 } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
305 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
306 numa_node);
307 if (area != NULL)
308 return area;
309 }
310
311 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
312 gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
313 flags, numa_node, __builtin_return_address(0));
314}
315
316void *bpf_map_area_alloc(u64 size, int numa_node)
317{
318 return __bpf_map_area_alloc(size, numa_node, false);
319}
320
321void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
322{
323 return __bpf_map_area_alloc(size, numa_node, true);
324}
325
326void bpf_map_area_free(void *area)
327{
328 kvfree(area);
329}
330
331static u32 bpf_map_flags_retain_permanent(u32 flags)
332{
333 /* Some map creation flags are not tied to the map object but
334 * rather to the map fd instead, so they have no meaning upon
335 * map object inspection since multiple file descriptors with
336 * different (access) properties can exist here. Thus, given
337 * this has zero meaning for the map itself, lets clear these
338 * from here.
339 */
340 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
341}
342
343void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
344{
345 map->map_type = attr->map_type;
346 map->key_size = attr->key_size;
347 map->value_size = attr->value_size;
348 map->max_entries = attr->max_entries;
349 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
350 map->numa_node = bpf_map_attr_numa_node(attr);
351}
352
353static int bpf_map_alloc_id(struct bpf_map *map)
354{
355 int id;
356
357 idr_preload(GFP_KERNEL);
358 spin_lock_bh(&map_idr_lock);
359 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
360 if (id > 0)
361 map->id = id;
362 spin_unlock_bh(&map_idr_lock);
363 idr_preload_end();
364
365 if (WARN_ON_ONCE(!id))
366 return -ENOSPC;
367
368 return id > 0 ? 0 : id;
369}
370
371void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
372{
373 unsigned long flags;
374
375 /* Offloaded maps are removed from the IDR store when their device
376 * disappears - even if someone holds an fd to them they are unusable,
377 * the memory is gone, all ops will fail; they are simply waiting for
378 * refcnt to drop to be freed.
379 */
380 if (!map->id)
381 return;
382
383 if (do_idr_lock)
384 spin_lock_irqsave(&map_idr_lock, flags);
385 else
386 __acquire(&map_idr_lock);
387
388 idr_remove(&map_idr, map->id);
389 map->id = 0;
390
391 if (do_idr_lock)
392 spin_unlock_irqrestore(&map_idr_lock, flags);
393 else
394 __release(&map_idr_lock);
395}
396
397#ifdef CONFIG_MEMCG_KMEM
398static void bpf_map_save_memcg(struct bpf_map *map)
399{
400 map->memcg = get_mem_cgroup_from_mm(current->mm);
401}
402
403static void bpf_map_release_memcg(struct bpf_map *map)
404{
405 mem_cgroup_put(map->memcg);
406}
407
408void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
409 int node)
410{
411 struct mem_cgroup *old_memcg;
412 void *ptr;
413
414 old_memcg = set_active_memcg(map->memcg);
415 ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
416 set_active_memcg(old_memcg);
417
418 return ptr;
419}
420
421void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
422{
423 struct mem_cgroup *old_memcg;
424 void *ptr;
425
426 old_memcg = set_active_memcg(map->memcg);
427 ptr = kzalloc(size, flags | __GFP_ACCOUNT);
428 set_active_memcg(old_memcg);
429
430 return ptr;
431}
432
433void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
434 size_t align, gfp_t flags)
435{
436 struct mem_cgroup *old_memcg;
437 void __percpu *ptr;
438
439 old_memcg = set_active_memcg(map->memcg);
440 ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
441 set_active_memcg(old_memcg);
442
443 return ptr;
444}
445
446#else
447static void bpf_map_save_memcg(struct bpf_map *map)
448{
449}
450
451static void bpf_map_release_memcg(struct bpf_map *map)
452{
453}
454#endif
455
456/* called from workqueue */
457static void bpf_map_free_deferred(struct work_struct *work)
458{
459 struct bpf_map *map = container_of(work, struct bpf_map, work);
460
461 security_bpf_map_free(map);
462 bpf_map_release_memcg(map);
463 /* implementation dependent freeing */
464 map->ops->map_free(map);
465}
466
467static void bpf_map_put_uref(struct bpf_map *map)
468{
469 if (atomic64_dec_and_test(&map->usercnt)) {
470 if (map->ops->map_release_uref)
471 map->ops->map_release_uref(map);
472 }
473}
474
475/* decrement map refcnt and schedule it for freeing via workqueue
476 * (unrelying map implementation ops->map_free() might sleep)
477 */
478static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
479{
480 if (atomic64_dec_and_test(&map->refcnt)) {
481 /* bpf_map_free_id() must be called first */
482 bpf_map_free_id(map, do_idr_lock);
483 btf_put(map->btf);
484 INIT_WORK(&map->work, bpf_map_free_deferred);
485 schedule_work(&map->work);
486 }
487}
488
489void bpf_map_put(struct bpf_map *map)
490{
491 __bpf_map_put(map, true);
492}
493EXPORT_SYMBOL_GPL(bpf_map_put);
494
495void bpf_map_put_with_uref(struct bpf_map *map)
496{
497 bpf_map_put_uref(map);
498 bpf_map_put(map);
499}
500
501static int bpf_map_release(struct inode *inode, struct file *filp)
502{
503 struct bpf_map *map = filp->private_data;
504
505 if (map->ops->map_release)
506 map->ops->map_release(map, filp);
507
508 bpf_map_put_with_uref(map);
509 return 0;
510}
511
512static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
513{
514 fmode_t mode = f.file->f_mode;
515
516 /* Our file permissions may have been overridden by global
517 * map permissions facing syscall side.
518 */
519 if (READ_ONCE(map->frozen))
520 mode &= ~FMODE_CAN_WRITE;
521 return mode;
522}
523
524#ifdef CONFIG_PROC_FS
525/* Provides an approximation of the map's memory footprint.
526 * Used only to provide a backward compatibility and display
527 * a reasonable "memlock" info.
528 */
529static unsigned long bpf_map_memory_footprint(const struct bpf_map *map)
530{
531 unsigned long size;
532
533 size = round_up(map->key_size + bpf_map_value_size(map), 8);
534
535 return round_up(map->max_entries * size, PAGE_SIZE);
536}
537
538static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
539{
540 const struct bpf_map *map = filp->private_data;
541 const struct bpf_array *array;
542 u32 type = 0, jited = 0;
543
544 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
545 array = container_of(map, struct bpf_array, map);
546 type = array->aux->type;
547 jited = array->aux->jited;
548 }
549
550 seq_printf(m,
551 "map_type:\t%u\n"
552 "key_size:\t%u\n"
553 "value_size:\t%u\n"
554 "max_entries:\t%u\n"
555 "map_flags:\t%#x\n"
556 "memlock:\t%lu\n"
557 "map_id:\t%u\n"
558 "frozen:\t%u\n",
559 map->map_type,
560 map->key_size,
561 map->value_size,
562 map->max_entries,
563 map->map_flags,
564 bpf_map_memory_footprint(map),
565 map->id,
566 READ_ONCE(map->frozen));
567 if (type) {
568 seq_printf(m, "owner_prog_type:\t%u\n", type);
569 seq_printf(m, "owner_jited:\t%u\n", jited);
570 }
571}
572#endif
573
574static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
575 loff_t *ppos)
576{
577 /* We need this handler such that alloc_file() enables
578 * f_mode with FMODE_CAN_READ.
579 */
580 return -EINVAL;
581}
582
583static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
584 size_t siz, loff_t *ppos)
585{
586 /* We need this handler such that alloc_file() enables
587 * f_mode with FMODE_CAN_WRITE.
588 */
589 return -EINVAL;
590}
591
592/* called for any extra memory-mapped regions (except initial) */
593static void bpf_map_mmap_open(struct vm_area_struct *vma)
594{
595 struct bpf_map *map = vma->vm_file->private_data;
596
597 if (vma->vm_flags & VM_MAYWRITE) {
598 mutex_lock(&map->freeze_mutex);
599 map->writecnt++;
600 mutex_unlock(&map->freeze_mutex);
601 }
602}
603
604/* called for all unmapped memory region (including initial) */
605static void bpf_map_mmap_close(struct vm_area_struct *vma)
606{
607 struct bpf_map *map = vma->vm_file->private_data;
608
609 if (vma->vm_flags & VM_MAYWRITE) {
610 mutex_lock(&map->freeze_mutex);
611 map->writecnt--;
612 mutex_unlock(&map->freeze_mutex);
613 }
614}
615
616static const struct vm_operations_struct bpf_map_default_vmops = {
617 .open = bpf_map_mmap_open,
618 .close = bpf_map_mmap_close,
619};
620
621static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
622{
623 struct bpf_map *map = filp->private_data;
624 int err;
625
626 if (!map->ops->map_mmap || map_value_has_spin_lock(map))
627 return -ENOTSUPP;
628
629 if (!(vma->vm_flags & VM_SHARED))
630 return -EINVAL;
631
632 mutex_lock(&map->freeze_mutex);
633
634 if (vma->vm_flags & VM_WRITE) {
635 if (map->frozen) {
636 err = -EPERM;
637 goto out;
638 }
639 /* map is meant to be read-only, so do not allow mapping as
640 * writable, because it's possible to leak a writable page
641 * reference and allows user-space to still modify it after
642 * freezing, while verifier will assume contents do not change
643 */
644 if (map->map_flags & BPF_F_RDONLY_PROG) {
645 err = -EACCES;
646 goto out;
647 }
648 }
649
650 /* set default open/close callbacks */
651 vma->vm_ops = &bpf_map_default_vmops;
652 vma->vm_private_data = map;
653 vma->vm_flags &= ~VM_MAYEXEC;
654 if (!(vma->vm_flags & VM_WRITE))
655 /* disallow re-mapping with PROT_WRITE */
656 vma->vm_flags &= ~VM_MAYWRITE;
657
658 err = map->ops->map_mmap(map, vma);
659 if (err)
660 goto out;
661
662 if (vma->vm_flags & VM_MAYWRITE)
663 map->writecnt++;
664out:
665 mutex_unlock(&map->freeze_mutex);
666 return err;
667}
668
669static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
670{
671 struct bpf_map *map = filp->private_data;
672
673 if (map->ops->map_poll)
674 return map->ops->map_poll(map, filp, pts);
675
676 return EPOLLERR;
677}
678
679const struct file_operations bpf_map_fops = {
680#ifdef CONFIG_PROC_FS
681 .show_fdinfo = bpf_map_show_fdinfo,
682#endif
683 .release = bpf_map_release,
684 .read = bpf_dummy_read,
685 .write = bpf_dummy_write,
686 .mmap = bpf_map_mmap,
687 .poll = bpf_map_poll,
688};
689
690int bpf_map_new_fd(struct bpf_map *map, int flags)
691{
692 int ret;
693
694 ret = security_bpf_map(map, OPEN_FMODE(flags));
695 if (ret < 0)
696 return ret;
697
698 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
699 flags | O_CLOEXEC);
700}
701
702int bpf_get_file_flag(int flags)
703{
704 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
705 return -EINVAL;
706 if (flags & BPF_F_RDONLY)
707 return O_RDONLY;
708 if (flags & BPF_F_WRONLY)
709 return O_WRONLY;
710 return O_RDWR;
711}
712
713/* helper macro to check that unused fields 'union bpf_attr' are zero */
714#define CHECK_ATTR(CMD) \
715 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
716 sizeof(attr->CMD##_LAST_FIELD), 0, \
717 sizeof(*attr) - \
718 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
719 sizeof(attr->CMD##_LAST_FIELD)) != NULL
720
721/* dst and src must have at least "size" number of bytes.
722 * Return strlen on success and < 0 on error.
723 */
724int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
725{
726 const char *end = src + size;
727 const char *orig_src = src;
728
729 memset(dst, 0, size);
730 /* Copy all isalnum(), '_' and '.' chars. */
731 while (src < end && *src) {
732 if (!isalnum(*src) &&
733 *src != '_' && *src != '.')
734 return -EINVAL;
735 *dst++ = *src++;
736 }
737
738 /* No '\0' found in "size" number of bytes */
739 if (src == end)
740 return -EINVAL;
741
742 return src - orig_src;
743}
744
745int map_check_no_btf(const struct bpf_map *map,
746 const struct btf *btf,
747 const struct btf_type *key_type,
748 const struct btf_type *value_type)
749{
750 return -ENOTSUPP;
751}
752
753static int map_check_btf(struct bpf_map *map, const struct btf *btf,
754 u32 btf_key_id, u32 btf_value_id)
755{
756 const struct btf_type *key_type, *value_type;
757 u32 key_size, value_size;
758 int ret = 0;
759
760 /* Some maps allow key to be unspecified. */
761 if (btf_key_id) {
762 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
763 if (!key_type || key_size != map->key_size)
764 return -EINVAL;
765 } else {
766 key_type = btf_type_by_id(btf, 0);
767 if (!map->ops->map_check_btf)
768 return -EINVAL;
769 }
770
771 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
772 if (!value_type || value_size != map->value_size)
773 return -EINVAL;
774
775 map->spin_lock_off = btf_find_spin_lock(btf, value_type);
776
777 if (map_value_has_spin_lock(map)) {
778 if (map->map_flags & BPF_F_RDONLY_PROG)
779 return -EACCES;
780 if (map->map_type != BPF_MAP_TYPE_HASH &&
781 map->map_type != BPF_MAP_TYPE_ARRAY &&
782 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
783 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
784 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
785 map->map_type != BPF_MAP_TYPE_TASK_STORAGE)
786 return -ENOTSUPP;
787 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
788 map->value_size) {
789 WARN_ONCE(1,
790 "verifier bug spin_lock_off %d value_size %d\n",
791 map->spin_lock_off, map->value_size);
792 return -EFAULT;
793 }
794 }
795
796 if (map->ops->map_check_btf)
797 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
798
799 return ret;
800}
801
802#define BPF_MAP_CREATE_LAST_FIELD btf_vmlinux_value_type_id
803/* called via syscall */
804static int map_create(union bpf_attr *attr)
805{
806 int numa_node = bpf_map_attr_numa_node(attr);
807 struct bpf_map *map;
808 int f_flags;
809 int err;
810
811 err = CHECK_ATTR(BPF_MAP_CREATE);
812 if (err)
813 return -EINVAL;
814
815 if (attr->btf_vmlinux_value_type_id) {
816 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
817 attr->btf_key_type_id || attr->btf_value_type_id)
818 return -EINVAL;
819 } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
820 return -EINVAL;
821 }
822
823 f_flags = bpf_get_file_flag(attr->map_flags);
824 if (f_flags < 0)
825 return f_flags;
826
827 if (numa_node != NUMA_NO_NODE &&
828 ((unsigned int)numa_node >= nr_node_ids ||
829 !node_online(numa_node)))
830 return -EINVAL;
831
832 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
833 map = find_and_alloc_map(attr);
834 if (IS_ERR(map))
835 return PTR_ERR(map);
836
837 err = bpf_obj_name_cpy(map->name, attr->map_name,
838 sizeof(attr->map_name));
839 if (err < 0)
840 goto free_map;
841
842 atomic64_set(&map->refcnt, 1);
843 atomic64_set(&map->usercnt, 1);
844 mutex_init(&map->freeze_mutex);
845
846 map->spin_lock_off = -EINVAL;
847 if (attr->btf_key_type_id || attr->btf_value_type_id ||
848 /* Even the map's value is a kernel's struct,
849 * the bpf_prog.o must have BTF to begin with
850 * to figure out the corresponding kernel's
851 * counter part. Thus, attr->btf_fd has
852 * to be valid also.
853 */
854 attr->btf_vmlinux_value_type_id) {
855 struct btf *btf;
856
857 btf = btf_get_by_fd(attr->btf_fd);
858 if (IS_ERR(btf)) {
859 err = PTR_ERR(btf);
860 goto free_map;
861 }
862 if (btf_is_kernel(btf)) {
863 btf_put(btf);
864 err = -EACCES;
865 goto free_map;
866 }
867 map->btf = btf;
868
869 if (attr->btf_value_type_id) {
870 err = map_check_btf(map, btf, attr->btf_key_type_id,
871 attr->btf_value_type_id);
872 if (err)
873 goto free_map;
874 }
875
876 map->btf_key_type_id = attr->btf_key_type_id;
877 map->btf_value_type_id = attr->btf_value_type_id;
878 map->btf_vmlinux_value_type_id =
879 attr->btf_vmlinux_value_type_id;
880 }
881
882 err = security_bpf_map_alloc(map);
883 if (err)
884 goto free_map;
885
886 err = bpf_map_alloc_id(map);
887 if (err)
888 goto free_map_sec;
889
890 bpf_map_save_memcg(map);
891
892 err = bpf_map_new_fd(map, f_flags);
893 if (err < 0) {
894 /* failed to allocate fd.
895 * bpf_map_put_with_uref() is needed because the above
896 * bpf_map_alloc_id() has published the map
897 * to the userspace and the userspace may
898 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
899 */
900 bpf_map_put_with_uref(map);
901 return err;
902 }
903
904 return err;
905
906free_map_sec:
907 security_bpf_map_free(map);
908free_map:
909 btf_put(map->btf);
910 map->ops->map_free(map);
911 return err;
912}
913
914/* if error is returned, fd is released.
915 * On success caller should complete fd access with matching fdput()
916 */
917struct bpf_map *__bpf_map_get(struct fd f)
918{
919 if (!f.file)
920 return ERR_PTR(-EBADF);
921 if (f.file->f_op != &bpf_map_fops) {
922 fdput(f);
923 return ERR_PTR(-EINVAL);
924 }
925
926 return f.file->private_data;
927}
928
929void bpf_map_inc(struct bpf_map *map)
930{
931 atomic64_inc(&map->refcnt);
932}
933EXPORT_SYMBOL_GPL(bpf_map_inc);
934
935void bpf_map_inc_with_uref(struct bpf_map *map)
936{
937 atomic64_inc(&map->refcnt);
938 atomic64_inc(&map->usercnt);
939}
940EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
941
942struct bpf_map *bpf_map_get(u32 ufd)
943{
944 struct fd f = fdget(ufd);
945 struct bpf_map *map;
946
947 map = __bpf_map_get(f);
948 if (IS_ERR(map))
949 return map;
950
951 bpf_map_inc(map);
952 fdput(f);
953
954 return map;
955}
956
957struct bpf_map *bpf_map_get_with_uref(u32 ufd)
958{
959 struct fd f = fdget(ufd);
960 struct bpf_map *map;
961
962 map = __bpf_map_get(f);
963 if (IS_ERR(map))
964 return map;
965
966 bpf_map_inc_with_uref(map);
967 fdput(f);
968
969 return map;
970}
971
972/* map_idr_lock should have been held */
973static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
974{
975 int refold;
976
977 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
978 if (!refold)
979 return ERR_PTR(-ENOENT);
980 if (uref)
981 atomic64_inc(&map->usercnt);
982
983 return map;
984}
985
986struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
987{
988 spin_lock_bh(&map_idr_lock);
989 map = __bpf_map_inc_not_zero(map, false);
990 spin_unlock_bh(&map_idr_lock);
991
992 return map;
993}
994EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
995
996int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
997{
998 return -ENOTSUPP;
999}
1000
1001static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1002{
1003 if (key_size)
1004 return memdup_user(ukey, key_size);
1005
1006 if (ukey)
1007 return ERR_PTR(-EINVAL);
1008
1009 return NULL;
1010}
1011
1012static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1013{
1014 if (key_size)
1015 return memdup_bpfptr(ukey, key_size);
1016
1017 if (!bpfptr_is_null(ukey))
1018 return ERR_PTR(-EINVAL);
1019
1020 return NULL;
1021}
1022
1023/* last field in 'union bpf_attr' used by this command */
1024#define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1025
1026static int map_lookup_elem(union bpf_attr *attr)
1027{
1028 void __user *ukey = u64_to_user_ptr(attr->key);
1029 void __user *uvalue = u64_to_user_ptr(attr->value);
1030 int ufd = attr->map_fd;
1031 struct bpf_map *map;
1032 void *key, *value;
1033 u32 value_size;
1034 struct fd f;
1035 int err;
1036
1037 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1038 return -EINVAL;
1039
1040 if (attr->flags & ~BPF_F_LOCK)
1041 return -EINVAL;
1042
1043 f = fdget(ufd);
1044 map = __bpf_map_get(f);
1045 if (IS_ERR(map))
1046 return PTR_ERR(map);
1047 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1048 err = -EPERM;
1049 goto err_put;
1050 }
1051
1052 if ((attr->flags & BPF_F_LOCK) &&
1053 !map_value_has_spin_lock(map)) {
1054 err = -EINVAL;
1055 goto err_put;
1056 }
1057
1058 key = __bpf_copy_key(ukey, map->key_size);
1059 if (IS_ERR(key)) {
1060 err = PTR_ERR(key);
1061 goto err_put;
1062 }
1063
1064 value_size = bpf_map_value_size(map);
1065
1066 err = -ENOMEM;
1067 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1068 if (!value)
1069 goto free_key;
1070
1071 err = bpf_map_copy_value(map, key, value, attr->flags);
1072 if (err)
1073 goto free_value;
1074
1075 err = -EFAULT;
1076 if (copy_to_user(uvalue, value, value_size) != 0)
1077 goto free_value;
1078
1079 err = 0;
1080
1081free_value:
1082 kfree(value);
1083free_key:
1084 kfree(key);
1085err_put:
1086 fdput(f);
1087 return err;
1088}
1089
1090
1091#define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1092
1093static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1094{
1095 bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1096 bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1097 int ufd = attr->map_fd;
1098 struct bpf_map *map;
1099 void *key, *value;
1100 u32 value_size;
1101 struct fd f;
1102 int err;
1103
1104 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1105 return -EINVAL;
1106
1107 f = fdget(ufd);
1108 map = __bpf_map_get(f);
1109 if (IS_ERR(map))
1110 return PTR_ERR(map);
1111 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1112 err = -EPERM;
1113 goto err_put;
1114 }
1115
1116 if ((attr->flags & BPF_F_LOCK) &&
1117 !map_value_has_spin_lock(map)) {
1118 err = -EINVAL;
1119 goto err_put;
1120 }
1121
1122 key = ___bpf_copy_key(ukey, map->key_size);
1123 if (IS_ERR(key)) {
1124 err = PTR_ERR(key);
1125 goto err_put;
1126 }
1127
1128 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1129 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
1130 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
1131 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
1132 value_size = round_up(map->value_size, 8) * num_possible_cpus();
1133 else
1134 value_size = map->value_size;
1135
1136 err = -ENOMEM;
1137 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1138 if (!value)
1139 goto free_key;
1140
1141 err = -EFAULT;
1142 if (copy_from_bpfptr(value, uvalue, value_size) != 0)
1143 goto free_value;
1144
1145 err = bpf_map_update_value(map, f, key, value, attr->flags);
1146
1147free_value:
1148 kfree(value);
1149free_key:
1150 kfree(key);
1151err_put:
1152 fdput(f);
1153 return err;
1154}
1155
1156#define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1157
1158static int map_delete_elem(union bpf_attr *attr)
1159{
1160 void __user *ukey = u64_to_user_ptr(attr->key);
1161 int ufd = attr->map_fd;
1162 struct bpf_map *map;
1163 struct fd f;
1164 void *key;
1165 int err;
1166
1167 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1168 return -EINVAL;
1169
1170 f = fdget(ufd);
1171 map = __bpf_map_get(f);
1172 if (IS_ERR(map))
1173 return PTR_ERR(map);
1174 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1175 err = -EPERM;
1176 goto err_put;
1177 }
1178
1179 key = __bpf_copy_key(ukey, map->key_size);
1180 if (IS_ERR(key)) {
1181 err = PTR_ERR(key);
1182 goto err_put;
1183 }
1184
1185 if (bpf_map_is_dev_bound(map)) {
1186 err = bpf_map_offload_delete_elem(map, key);
1187 goto out;
1188 } else if (IS_FD_PROG_ARRAY(map) ||
1189 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1190 /* These maps require sleepable context */
1191 err = map->ops->map_delete_elem(map, key);
1192 goto out;
1193 }
1194
1195 bpf_disable_instrumentation();
1196 rcu_read_lock();
1197 err = map->ops->map_delete_elem(map, key);
1198 rcu_read_unlock();
1199 bpf_enable_instrumentation();
1200 maybe_wait_bpf_programs(map);
1201out:
1202 kfree(key);
1203err_put:
1204 fdput(f);
1205 return err;
1206}
1207
1208/* last field in 'union bpf_attr' used by this command */
1209#define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1210
1211static int map_get_next_key(union bpf_attr *attr)
1212{
1213 void __user *ukey = u64_to_user_ptr(attr->key);
1214 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1215 int ufd = attr->map_fd;
1216 struct bpf_map *map;
1217 void *key, *next_key;
1218 struct fd f;
1219 int err;
1220
1221 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1222 return -EINVAL;
1223
1224 f = fdget(ufd);
1225 map = __bpf_map_get(f);
1226 if (IS_ERR(map))
1227 return PTR_ERR(map);
1228 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1229 err = -EPERM;
1230 goto err_put;
1231 }
1232
1233 if (ukey) {
1234 key = __bpf_copy_key(ukey, map->key_size);
1235 if (IS_ERR(key)) {
1236 err = PTR_ERR(key);
1237 goto err_put;
1238 }
1239 } else {
1240 key = NULL;
1241 }
1242
1243 err = -ENOMEM;
1244 next_key = kmalloc(map->key_size, GFP_USER);
1245 if (!next_key)
1246 goto free_key;
1247
1248 if (bpf_map_is_dev_bound(map)) {
1249 err = bpf_map_offload_get_next_key(map, key, next_key);
1250 goto out;
1251 }
1252
1253 rcu_read_lock();
1254 err = map->ops->map_get_next_key(map, key, next_key);
1255 rcu_read_unlock();
1256out:
1257 if (err)
1258 goto free_next_key;
1259
1260 err = -EFAULT;
1261 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1262 goto free_next_key;
1263
1264 err = 0;
1265
1266free_next_key:
1267 kfree(next_key);
1268free_key:
1269 kfree(key);
1270err_put:
1271 fdput(f);
1272 return err;
1273}
1274
1275int generic_map_delete_batch(struct bpf_map *map,
1276 const union bpf_attr *attr,
1277 union bpf_attr __user *uattr)
1278{
1279 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1280 u32 cp, max_count;
1281 int err = 0;
1282 void *key;
1283
1284 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1285 return -EINVAL;
1286
1287 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1288 !map_value_has_spin_lock(map)) {
1289 return -EINVAL;
1290 }
1291
1292 max_count = attr->batch.count;
1293 if (!max_count)
1294 return 0;
1295
1296 key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1297 if (!key)
1298 return -ENOMEM;
1299
1300 for (cp = 0; cp < max_count; cp++) {
1301 err = -EFAULT;
1302 if (copy_from_user(key, keys + cp * map->key_size,
1303 map->key_size))
1304 break;
1305
1306 if (bpf_map_is_dev_bound(map)) {
1307 err = bpf_map_offload_delete_elem(map, key);
1308 break;
1309 }
1310
1311 bpf_disable_instrumentation();
1312 rcu_read_lock();
1313 err = map->ops->map_delete_elem(map, key);
1314 rcu_read_unlock();
1315 bpf_enable_instrumentation();
1316 maybe_wait_bpf_programs(map);
1317 if (err)
1318 break;
1319 }
1320 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1321 err = -EFAULT;
1322
1323 kfree(key);
1324 return err;
1325}
1326
1327int generic_map_update_batch(struct bpf_map *map,
1328 const union bpf_attr *attr,
1329 union bpf_attr __user *uattr)
1330{
1331 void __user *values = u64_to_user_ptr(attr->batch.values);
1332 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1333 u32 value_size, cp, max_count;
1334 int ufd = attr->map_fd;
1335 void *key, *value;
1336 struct fd f;
1337 int err = 0;
1338
1339 f = fdget(ufd);
1340 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1341 return -EINVAL;
1342
1343 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1344 !map_value_has_spin_lock(map)) {
1345 return -EINVAL;
1346 }
1347
1348 value_size = bpf_map_value_size(map);
1349
1350 max_count = attr->batch.count;
1351 if (!max_count)
1352 return 0;
1353
1354 key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1355 if (!key)
1356 return -ENOMEM;
1357
1358 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1359 if (!value) {
1360 kfree(key);
1361 return -ENOMEM;
1362 }
1363
1364 for (cp = 0; cp < max_count; cp++) {
1365 err = -EFAULT;
1366 if (copy_from_user(key, keys + cp * map->key_size,
1367 map->key_size) ||
1368 copy_from_user(value, values + cp * value_size, value_size))
1369 break;
1370
1371 err = bpf_map_update_value(map, f, key, value,
1372 attr->batch.elem_flags);
1373
1374 if (err)
1375 break;
1376 }
1377
1378 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1379 err = -EFAULT;
1380
1381 kfree(value);
1382 kfree(key);
1383 return err;
1384}
1385
1386#define MAP_LOOKUP_RETRIES 3
1387
1388int generic_map_lookup_batch(struct bpf_map *map,
1389 const union bpf_attr *attr,
1390 union bpf_attr __user *uattr)
1391{
1392 void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1393 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1394 void __user *values = u64_to_user_ptr(attr->batch.values);
1395 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1396 void *buf, *buf_prevkey, *prev_key, *key, *value;
1397 int err, retry = MAP_LOOKUP_RETRIES;
1398 u32 value_size, cp, max_count;
1399
1400 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1401 return -EINVAL;
1402
1403 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1404 !map_value_has_spin_lock(map))
1405 return -EINVAL;
1406
1407 value_size = bpf_map_value_size(map);
1408
1409 max_count = attr->batch.count;
1410 if (!max_count)
1411 return 0;
1412
1413 if (put_user(0, &uattr->batch.count))
1414 return -EFAULT;
1415
1416 buf_prevkey = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1417 if (!buf_prevkey)
1418 return -ENOMEM;
1419
1420 buf = kmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1421 if (!buf) {
1422 kfree(buf_prevkey);
1423 return -ENOMEM;
1424 }
1425
1426 err = -EFAULT;
1427 prev_key = NULL;
1428 if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1429 goto free_buf;
1430 key = buf;
1431 value = key + map->key_size;
1432 if (ubatch)
1433 prev_key = buf_prevkey;
1434
1435 for (cp = 0; cp < max_count;) {
1436 rcu_read_lock();
1437 err = map->ops->map_get_next_key(map, prev_key, key);
1438 rcu_read_unlock();
1439 if (err)
1440 break;
1441 err = bpf_map_copy_value(map, key, value,
1442 attr->batch.elem_flags);
1443
1444 if (err == -ENOENT) {
1445 if (retry) {
1446 retry--;
1447 continue;
1448 }
1449 err = -EINTR;
1450 break;
1451 }
1452
1453 if (err)
1454 goto free_buf;
1455
1456 if (copy_to_user(keys + cp * map->key_size, key,
1457 map->key_size)) {
1458 err = -EFAULT;
1459 goto free_buf;
1460 }
1461 if (copy_to_user(values + cp * value_size, value, value_size)) {
1462 err = -EFAULT;
1463 goto free_buf;
1464 }
1465
1466 if (!prev_key)
1467 prev_key = buf_prevkey;
1468
1469 swap(prev_key, key);
1470 retry = MAP_LOOKUP_RETRIES;
1471 cp++;
1472 }
1473
1474 if (err == -EFAULT)
1475 goto free_buf;
1476
1477 if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1478 (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1479 err = -EFAULT;
1480
1481free_buf:
1482 kfree(buf_prevkey);
1483 kfree(buf);
1484 return err;
1485}
1486
1487#define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1488
1489static int map_lookup_and_delete_elem(union bpf_attr *attr)
1490{
1491 void __user *ukey = u64_to_user_ptr(attr->key);
1492 void __user *uvalue = u64_to_user_ptr(attr->value);
1493 int ufd = attr->map_fd;
1494 struct bpf_map *map;
1495 void *key, *value;
1496 u32 value_size;
1497 struct fd f;
1498 int err;
1499
1500 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1501 return -EINVAL;
1502
1503 if (attr->flags & ~BPF_F_LOCK)
1504 return -EINVAL;
1505
1506 f = fdget(ufd);
1507 map = __bpf_map_get(f);
1508 if (IS_ERR(map))
1509 return PTR_ERR(map);
1510 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1511 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1512 err = -EPERM;
1513 goto err_put;
1514 }
1515
1516 if (attr->flags &&
1517 (map->map_type == BPF_MAP_TYPE_QUEUE ||
1518 map->map_type == BPF_MAP_TYPE_STACK)) {
1519 err = -EINVAL;
1520 goto err_put;
1521 }
1522
1523 if ((attr->flags & BPF_F_LOCK) &&
1524 !map_value_has_spin_lock(map)) {
1525 err = -EINVAL;
1526 goto err_put;
1527 }
1528
1529 key = __bpf_copy_key(ukey, map->key_size);
1530 if (IS_ERR(key)) {
1531 err = PTR_ERR(key);
1532 goto err_put;
1533 }
1534
1535 value_size = bpf_map_value_size(map);
1536
1537 err = -ENOMEM;
1538 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1539 if (!value)
1540 goto free_key;
1541
1542 err = -ENOTSUPP;
1543 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1544 map->map_type == BPF_MAP_TYPE_STACK) {
1545 err = map->ops->map_pop_elem(map, value);
1546 } else if (map->map_type == BPF_MAP_TYPE_HASH ||
1547 map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1548 map->map_type == BPF_MAP_TYPE_LRU_HASH ||
1549 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1550 if (!bpf_map_is_dev_bound(map)) {
1551 bpf_disable_instrumentation();
1552 rcu_read_lock();
1553 err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
1554 rcu_read_unlock();
1555 bpf_enable_instrumentation();
1556 }
1557 }
1558
1559 if (err)
1560 goto free_value;
1561
1562 if (copy_to_user(uvalue, value, value_size) != 0) {
1563 err = -EFAULT;
1564 goto free_value;
1565 }
1566
1567 err = 0;
1568
1569free_value:
1570 kfree(value);
1571free_key:
1572 kfree(key);
1573err_put:
1574 fdput(f);
1575 return err;
1576}
1577
1578#define BPF_MAP_FREEZE_LAST_FIELD map_fd
1579
1580static int map_freeze(const union bpf_attr *attr)
1581{
1582 int err = 0, ufd = attr->map_fd;
1583 struct bpf_map *map;
1584 struct fd f;
1585
1586 if (CHECK_ATTR(BPF_MAP_FREEZE))
1587 return -EINVAL;
1588
1589 f = fdget(ufd);
1590 map = __bpf_map_get(f);
1591 if (IS_ERR(map))
1592 return PTR_ERR(map);
1593
1594 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1595 fdput(f);
1596 return -ENOTSUPP;
1597 }
1598
1599 mutex_lock(&map->freeze_mutex);
1600
1601 if (map->writecnt) {
1602 err = -EBUSY;
1603 goto err_put;
1604 }
1605 if (READ_ONCE(map->frozen)) {
1606 err = -EBUSY;
1607 goto err_put;
1608 }
1609 if (!bpf_capable()) {
1610 err = -EPERM;
1611 goto err_put;
1612 }
1613
1614 WRITE_ONCE(map->frozen, true);
1615err_put:
1616 mutex_unlock(&map->freeze_mutex);
1617 fdput(f);
1618 return err;
1619}
1620
1621static const struct bpf_prog_ops * const bpf_prog_types[] = {
1622#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1623 [_id] = & _name ## _prog_ops,
1624#define BPF_MAP_TYPE(_id, _ops)
1625#define BPF_LINK_TYPE(_id, _name)
1626#include <linux/bpf_types.h>
1627#undef BPF_PROG_TYPE
1628#undef BPF_MAP_TYPE
1629#undef BPF_LINK_TYPE
1630};
1631
1632static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1633{
1634 const struct bpf_prog_ops *ops;
1635
1636 if (type >= ARRAY_SIZE(bpf_prog_types))
1637 return -EINVAL;
1638 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1639 ops = bpf_prog_types[type];
1640 if (!ops)
1641 return -EINVAL;
1642
1643 if (!bpf_prog_is_dev_bound(prog->aux))
1644 prog->aux->ops = ops;
1645 else
1646 prog->aux->ops = &bpf_offload_prog_ops;
1647 prog->type = type;
1648 return 0;
1649}
1650
1651enum bpf_audit {
1652 BPF_AUDIT_LOAD,
1653 BPF_AUDIT_UNLOAD,
1654 BPF_AUDIT_MAX,
1655};
1656
1657static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1658 [BPF_AUDIT_LOAD] = "LOAD",
1659 [BPF_AUDIT_UNLOAD] = "UNLOAD",
1660};
1661
1662static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1663{
1664 struct audit_context *ctx = NULL;
1665 struct audit_buffer *ab;
1666
1667 if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1668 return;
1669 if (audit_enabled == AUDIT_OFF)
1670 return;
1671 if (op == BPF_AUDIT_LOAD)
1672 ctx = audit_context();
1673 ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1674 if (unlikely(!ab))
1675 return;
1676 audit_log_format(ab, "prog-id=%u op=%s",
1677 prog->aux->id, bpf_audit_str[op]);
1678 audit_log_end(ab);
1679}
1680
1681static int bpf_prog_alloc_id(struct bpf_prog *prog)
1682{
1683 int id;
1684
1685 idr_preload(GFP_KERNEL);
1686 spin_lock_bh(&prog_idr_lock);
1687 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1688 if (id > 0)
1689 prog->aux->id = id;
1690 spin_unlock_bh(&prog_idr_lock);
1691 idr_preload_end();
1692
1693 /* id is in [1, INT_MAX) */
1694 if (WARN_ON_ONCE(!id))
1695 return -ENOSPC;
1696
1697 return id > 0 ? 0 : id;
1698}
1699
1700void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1701{
1702 /* cBPF to eBPF migrations are currently not in the idr store.
1703 * Offloaded programs are removed from the store when their device
1704 * disappears - even if someone grabs an fd to them they are unusable,
1705 * simply waiting for refcnt to drop to be freed.
1706 */
1707 if (!prog->aux->id)
1708 return;
1709
1710 if (do_idr_lock)
1711 spin_lock_bh(&prog_idr_lock);
1712 else
1713 __acquire(&prog_idr_lock);
1714
1715 idr_remove(&prog_idr, prog->aux->id);
1716 prog->aux->id = 0;
1717
1718 if (do_idr_lock)
1719 spin_unlock_bh(&prog_idr_lock);
1720 else
1721 __release(&prog_idr_lock);
1722}
1723
1724static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1725{
1726 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1727
1728 kvfree(aux->func_info);
1729 kfree(aux->func_info_aux);
1730 free_uid(aux->user);
1731 security_bpf_prog_free(aux);
1732 bpf_prog_free(aux->prog);
1733}
1734
1735static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1736{
1737 bpf_prog_kallsyms_del_all(prog);
1738 btf_put(prog->aux->btf);
1739 kvfree(prog->aux->jited_linfo);
1740 kvfree(prog->aux->linfo);
1741 kfree(prog->aux->kfunc_tab);
1742 if (prog->aux->attach_btf)
1743 btf_put(prog->aux->attach_btf);
1744
1745 if (deferred) {
1746 if (prog->aux->sleepable)
1747 call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
1748 else
1749 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1750 } else {
1751 __bpf_prog_put_rcu(&prog->aux->rcu);
1752 }
1753}
1754
1755static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1756{
1757 if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1758 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1759 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
1760 /* bpf_prog_free_id() must be called first */
1761 bpf_prog_free_id(prog, do_idr_lock);
1762 __bpf_prog_put_noref(prog, true);
1763 }
1764}
1765
1766void bpf_prog_put(struct bpf_prog *prog)
1767{
1768 __bpf_prog_put(prog, true);
1769}
1770EXPORT_SYMBOL_GPL(bpf_prog_put);
1771
1772static int bpf_prog_release(struct inode *inode, struct file *filp)
1773{
1774 struct bpf_prog *prog = filp->private_data;
1775
1776 bpf_prog_put(prog);
1777 return 0;
1778}
1779
1780static void bpf_prog_get_stats(const struct bpf_prog *prog,
1781 struct bpf_prog_stats *stats)
1782{
1783 u64 nsecs = 0, cnt = 0, misses = 0;
1784 int cpu;
1785
1786 for_each_possible_cpu(cpu) {
1787 const struct bpf_prog_stats *st;
1788 unsigned int start;
1789 u64 tnsecs, tcnt, tmisses;
1790
1791 st = per_cpu_ptr(prog->stats, cpu);
1792 do {
1793 start = u64_stats_fetch_begin_irq(&st->syncp);
1794 tnsecs = st->nsecs;
1795 tcnt = st->cnt;
1796 tmisses = st->misses;
1797 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1798 nsecs += tnsecs;
1799 cnt += tcnt;
1800 misses += tmisses;
1801 }
1802 stats->nsecs = nsecs;
1803 stats->cnt = cnt;
1804 stats->misses = misses;
1805}
1806
1807#ifdef CONFIG_PROC_FS
1808static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1809{
1810 const struct bpf_prog *prog = filp->private_data;
1811 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1812 struct bpf_prog_stats stats;
1813
1814 bpf_prog_get_stats(prog, &stats);
1815 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1816 seq_printf(m,
1817 "prog_type:\t%u\n"
1818 "prog_jited:\t%u\n"
1819 "prog_tag:\t%s\n"
1820 "memlock:\t%llu\n"
1821 "prog_id:\t%u\n"
1822 "run_time_ns:\t%llu\n"
1823 "run_cnt:\t%llu\n"
1824 "recursion_misses:\t%llu\n",
1825 prog->type,
1826 prog->jited,
1827 prog_tag,
1828 prog->pages * 1ULL << PAGE_SHIFT,
1829 prog->aux->id,
1830 stats.nsecs,
1831 stats.cnt,
1832 stats.misses);
1833}
1834#endif
1835
1836const struct file_operations bpf_prog_fops = {
1837#ifdef CONFIG_PROC_FS
1838 .show_fdinfo = bpf_prog_show_fdinfo,
1839#endif
1840 .release = bpf_prog_release,
1841 .read = bpf_dummy_read,
1842 .write = bpf_dummy_write,
1843};
1844
1845int bpf_prog_new_fd(struct bpf_prog *prog)
1846{
1847 int ret;
1848
1849 ret = security_bpf_prog(prog);
1850 if (ret < 0)
1851 return ret;
1852
1853 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1854 O_RDWR | O_CLOEXEC);
1855}
1856
1857static struct bpf_prog *____bpf_prog_get(struct fd f)
1858{
1859 if (!f.file)
1860 return ERR_PTR(-EBADF);
1861 if (f.file->f_op != &bpf_prog_fops) {
1862 fdput(f);
1863 return ERR_PTR(-EINVAL);
1864 }
1865
1866 return f.file->private_data;
1867}
1868
1869void bpf_prog_add(struct bpf_prog *prog, int i)
1870{
1871 atomic64_add(i, &prog->aux->refcnt);
1872}
1873EXPORT_SYMBOL_GPL(bpf_prog_add);
1874
1875void bpf_prog_sub(struct bpf_prog *prog, int i)
1876{
1877 /* Only to be used for undoing previous bpf_prog_add() in some
1878 * error path. We still know that another entity in our call
1879 * path holds a reference to the program, thus atomic_sub() can
1880 * be safely used in such cases!
1881 */
1882 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1883}
1884EXPORT_SYMBOL_GPL(bpf_prog_sub);
1885
1886void bpf_prog_inc(struct bpf_prog *prog)
1887{
1888 atomic64_inc(&prog->aux->refcnt);
1889}
1890EXPORT_SYMBOL_GPL(bpf_prog_inc);
1891
1892/* prog_idr_lock should have been held */
1893struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1894{
1895 int refold;
1896
1897 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1898
1899 if (!refold)
1900 return ERR_PTR(-ENOENT);
1901
1902 return prog;
1903}
1904EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1905
1906bool bpf_prog_get_ok(struct bpf_prog *prog,
1907 enum bpf_prog_type *attach_type, bool attach_drv)
1908{
1909 /* not an attachment, just a refcount inc, always allow */
1910 if (!attach_type)
1911 return true;
1912
1913 if (prog->type != *attach_type)
1914 return false;
1915 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1916 return false;
1917
1918 return true;
1919}
1920
1921static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1922 bool attach_drv)
1923{
1924 struct fd f = fdget(ufd);
1925 struct bpf_prog *prog;
1926
1927 prog = ____bpf_prog_get(f);
1928 if (IS_ERR(prog))
1929 return prog;
1930 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1931 prog = ERR_PTR(-EINVAL);
1932 goto out;
1933 }
1934
1935 bpf_prog_inc(prog);
1936out:
1937 fdput(f);
1938 return prog;
1939}
1940
1941struct bpf_prog *bpf_prog_get(u32 ufd)
1942{
1943 return __bpf_prog_get(ufd, NULL, false);
1944}
1945
1946struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1947 bool attach_drv)
1948{
1949 return __bpf_prog_get(ufd, &type, attach_drv);
1950}
1951EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1952
1953/* Initially all BPF programs could be loaded w/o specifying
1954 * expected_attach_type. Later for some of them specifying expected_attach_type
1955 * at load time became required so that program could be validated properly.
1956 * Programs of types that are allowed to be loaded both w/ and w/o (for
1957 * backward compatibility) expected_attach_type, should have the default attach
1958 * type assigned to expected_attach_type for the latter case, so that it can be
1959 * validated later at attach time.
1960 *
1961 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1962 * prog type requires it but has some attach types that have to be backward
1963 * compatible.
1964 */
1965static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1966{
1967 switch (attr->prog_type) {
1968 case BPF_PROG_TYPE_CGROUP_SOCK:
1969 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1970 * exist so checking for non-zero is the way to go here.
1971 */
1972 if (!attr->expected_attach_type)
1973 attr->expected_attach_type =
1974 BPF_CGROUP_INET_SOCK_CREATE;
1975 break;
1976 case BPF_PROG_TYPE_SK_REUSEPORT:
1977 if (!attr->expected_attach_type)
1978 attr->expected_attach_type =
1979 BPF_SK_REUSEPORT_SELECT;
1980 break;
1981 }
1982}
1983
1984static int
1985bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1986 enum bpf_attach_type expected_attach_type,
1987 struct btf *attach_btf, u32 btf_id,
1988 struct bpf_prog *dst_prog)
1989{
1990 if (btf_id) {
1991 if (btf_id > BTF_MAX_TYPE)
1992 return -EINVAL;
1993
1994 if (!attach_btf && !dst_prog)
1995 return -EINVAL;
1996
1997 switch (prog_type) {
1998 case BPF_PROG_TYPE_TRACING:
1999 case BPF_PROG_TYPE_LSM:
2000 case BPF_PROG_TYPE_STRUCT_OPS:
2001 case BPF_PROG_TYPE_EXT:
2002 break;
2003 default:
2004 return -EINVAL;
2005 }
2006 }
2007
2008 if (attach_btf && (!btf_id || dst_prog))
2009 return -EINVAL;
2010
2011 if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2012 prog_type != BPF_PROG_TYPE_EXT)
2013 return -EINVAL;
2014
2015 switch (prog_type) {
2016 case BPF_PROG_TYPE_CGROUP_SOCK:
2017 switch (expected_attach_type) {
2018 case BPF_CGROUP_INET_SOCK_CREATE:
2019 case BPF_CGROUP_INET_SOCK_RELEASE:
2020 case BPF_CGROUP_INET4_POST_BIND:
2021 case BPF_CGROUP_INET6_POST_BIND:
2022 return 0;
2023 default:
2024 return -EINVAL;
2025 }
2026 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2027 switch (expected_attach_type) {
2028 case BPF_CGROUP_INET4_BIND:
2029 case BPF_CGROUP_INET6_BIND:
2030 case BPF_CGROUP_INET4_CONNECT:
2031 case BPF_CGROUP_INET6_CONNECT:
2032 case BPF_CGROUP_INET4_GETPEERNAME:
2033 case BPF_CGROUP_INET6_GETPEERNAME:
2034 case BPF_CGROUP_INET4_GETSOCKNAME:
2035 case BPF_CGROUP_INET6_GETSOCKNAME:
2036 case BPF_CGROUP_UDP4_SENDMSG:
2037 case BPF_CGROUP_UDP6_SENDMSG:
2038 case BPF_CGROUP_UDP4_RECVMSG:
2039 case BPF_CGROUP_UDP6_RECVMSG:
2040 return 0;
2041 default:
2042 return -EINVAL;
2043 }
2044 case BPF_PROG_TYPE_CGROUP_SKB:
2045 switch (expected_attach_type) {
2046 case BPF_CGROUP_INET_INGRESS:
2047 case BPF_CGROUP_INET_EGRESS:
2048 return 0;
2049 default:
2050 return -EINVAL;
2051 }
2052 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2053 switch (expected_attach_type) {
2054 case BPF_CGROUP_SETSOCKOPT:
2055 case BPF_CGROUP_GETSOCKOPT:
2056 return 0;
2057 default:
2058 return -EINVAL;
2059 }
2060 case BPF_PROG_TYPE_SK_LOOKUP:
2061 if (expected_attach_type == BPF_SK_LOOKUP)
2062 return 0;
2063 return -EINVAL;
2064 case BPF_PROG_TYPE_SK_REUSEPORT:
2065 switch (expected_attach_type) {
2066 case BPF_SK_REUSEPORT_SELECT:
2067 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2068 return 0;
2069 default:
2070 return -EINVAL;
2071 }
2072 case BPF_PROG_TYPE_SYSCALL:
2073 case BPF_PROG_TYPE_EXT:
2074 if (expected_attach_type)
2075 return -EINVAL;
2076 fallthrough;
2077 default:
2078 return 0;
2079 }
2080}
2081
2082static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2083{
2084 switch (prog_type) {
2085 case BPF_PROG_TYPE_SCHED_CLS:
2086 case BPF_PROG_TYPE_SCHED_ACT:
2087 case BPF_PROG_TYPE_XDP:
2088 case BPF_PROG_TYPE_LWT_IN:
2089 case BPF_PROG_TYPE_LWT_OUT:
2090 case BPF_PROG_TYPE_LWT_XMIT:
2091 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2092 case BPF_PROG_TYPE_SK_SKB:
2093 case BPF_PROG_TYPE_SK_MSG:
2094 case BPF_PROG_TYPE_LIRC_MODE2:
2095 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2096 case BPF_PROG_TYPE_CGROUP_DEVICE:
2097 case BPF_PROG_TYPE_CGROUP_SOCK:
2098 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2099 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2100 case BPF_PROG_TYPE_CGROUP_SYSCTL:
2101 case BPF_PROG_TYPE_SOCK_OPS:
2102 case BPF_PROG_TYPE_EXT: /* extends any prog */
2103 return true;
2104 case BPF_PROG_TYPE_CGROUP_SKB:
2105 /* always unpriv */
2106 case BPF_PROG_TYPE_SK_REUSEPORT:
2107 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2108 default:
2109 return false;
2110 }
2111}
2112
2113static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2114{
2115 switch (prog_type) {
2116 case BPF_PROG_TYPE_KPROBE:
2117 case BPF_PROG_TYPE_TRACEPOINT:
2118 case BPF_PROG_TYPE_PERF_EVENT:
2119 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2120 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2121 case BPF_PROG_TYPE_TRACING:
2122 case BPF_PROG_TYPE_LSM:
2123 case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2124 case BPF_PROG_TYPE_EXT: /* extends any prog */
2125 return true;
2126 default:
2127 return false;
2128 }
2129}
2130
2131/* last field in 'union bpf_attr' used by this command */
2132#define BPF_PROG_LOAD_LAST_FIELD fd_array
2133
2134static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr)
2135{
2136 enum bpf_prog_type type = attr->prog_type;
2137 struct bpf_prog *prog, *dst_prog = NULL;
2138 struct btf *attach_btf = NULL;
2139 int err;
2140 char license[128];
2141 bool is_gpl;
2142
2143 if (CHECK_ATTR(BPF_PROG_LOAD))
2144 return -EINVAL;
2145
2146 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2147 BPF_F_ANY_ALIGNMENT |
2148 BPF_F_TEST_STATE_FREQ |
2149 BPF_F_SLEEPABLE |
2150 BPF_F_TEST_RND_HI32))
2151 return -EINVAL;
2152
2153 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2154 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2155 !bpf_capable())
2156 return -EPERM;
2157
2158 /* copy eBPF program license from user space */
2159 if (strncpy_from_bpfptr(license,
2160 make_bpfptr(attr->license, uattr.is_kernel),
2161 sizeof(license) - 1) < 0)
2162 return -EFAULT;
2163 license[sizeof(license) - 1] = 0;
2164
2165 /* eBPF programs must be GPL compatible to use GPL-ed functions */
2166 is_gpl = license_is_gpl_compatible(license);
2167
2168 if (attr->insn_cnt == 0 ||
2169 attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2170 return -E2BIG;
2171 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2172 type != BPF_PROG_TYPE_CGROUP_SKB &&
2173 !bpf_capable())
2174 return -EPERM;
2175
2176 if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2177 return -EPERM;
2178 if (is_perfmon_prog_type(type) && !perfmon_capable())
2179 return -EPERM;
2180
2181 /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2182 * or btf, we need to check which one it is
2183 */
2184 if (attr->attach_prog_fd) {
2185 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2186 if (IS_ERR(dst_prog)) {
2187 dst_prog = NULL;
2188 attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2189 if (IS_ERR(attach_btf))
2190 return -EINVAL;
2191 if (!btf_is_kernel(attach_btf)) {
2192 /* attaching through specifying bpf_prog's BTF
2193 * objects directly might be supported eventually
2194 */
2195 btf_put(attach_btf);
2196 return -ENOTSUPP;
2197 }
2198 }
2199 } else if (attr->attach_btf_id) {
2200 /* fall back to vmlinux BTF, if BTF type ID is specified */
2201 attach_btf = bpf_get_btf_vmlinux();
2202 if (IS_ERR(attach_btf))
2203 return PTR_ERR(attach_btf);
2204 if (!attach_btf)
2205 return -EINVAL;
2206 btf_get(attach_btf);
2207 }
2208
2209 bpf_prog_load_fixup_attach_type(attr);
2210 if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2211 attach_btf, attr->attach_btf_id,
2212 dst_prog)) {
2213 if (dst_prog)
2214 bpf_prog_put(dst_prog);
2215 if (attach_btf)
2216 btf_put(attach_btf);
2217 return -EINVAL;
2218 }
2219
2220 /* plain bpf_prog allocation */
2221 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2222 if (!prog) {
2223 if (dst_prog)
2224 bpf_prog_put(dst_prog);
2225 if (attach_btf)
2226 btf_put(attach_btf);
2227 return -ENOMEM;
2228 }
2229
2230 prog->expected_attach_type = attr->expected_attach_type;
2231 prog->aux->attach_btf = attach_btf;
2232 prog->aux->attach_btf_id = attr->attach_btf_id;
2233 prog->aux->dst_prog = dst_prog;
2234 prog->aux->offload_requested = !!attr->prog_ifindex;
2235 prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2236
2237 err = security_bpf_prog_alloc(prog->aux);
2238 if (err)
2239 goto free_prog;
2240
2241 prog->aux->user = get_current_user();
2242 prog->len = attr->insn_cnt;
2243
2244 err = -EFAULT;
2245 if (copy_from_bpfptr(prog->insns,
2246 make_bpfptr(attr->insns, uattr.is_kernel),
2247 bpf_prog_insn_size(prog)) != 0)
2248 goto free_prog_sec;
2249
2250 prog->orig_prog = NULL;
2251 prog->jited = 0;
2252
2253 atomic64_set(&prog->aux->refcnt, 1);
2254 prog->gpl_compatible = is_gpl ? 1 : 0;
2255
2256 if (bpf_prog_is_dev_bound(prog->aux)) {
2257 err = bpf_prog_offload_init(prog, attr);
2258 if (err)
2259 goto free_prog_sec;
2260 }
2261
2262 /* find program type: socket_filter vs tracing_filter */
2263 err = find_prog_type(type, prog);
2264 if (err < 0)
2265 goto free_prog_sec;
2266
2267 prog->aux->load_time = ktime_get_boottime_ns();
2268 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2269 sizeof(attr->prog_name));
2270 if (err < 0)
2271 goto free_prog_sec;
2272
2273 /* run eBPF verifier */
2274 err = bpf_check(&prog, attr, uattr);
2275 if (err < 0)
2276 goto free_used_maps;
2277
2278 prog = bpf_prog_select_runtime(prog, &err);
2279 if (err < 0)
2280 goto free_used_maps;
2281
2282 err = bpf_prog_alloc_id(prog);
2283 if (err)
2284 goto free_used_maps;
2285
2286 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2287 * effectively publicly exposed. However, retrieving via
2288 * bpf_prog_get_fd_by_id() will take another reference,
2289 * therefore it cannot be gone underneath us.
2290 *
2291 * Only for the time /after/ successful bpf_prog_new_fd()
2292 * and before returning to userspace, we might just hold
2293 * one reference and any parallel close on that fd could
2294 * rip everything out. Hence, below notifications must
2295 * happen before bpf_prog_new_fd().
2296 *
2297 * Also, any failure handling from this point onwards must
2298 * be using bpf_prog_put() given the program is exposed.
2299 */
2300 bpf_prog_kallsyms_add(prog);
2301 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2302 bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2303
2304 err = bpf_prog_new_fd(prog);
2305 if (err < 0)
2306 bpf_prog_put(prog);
2307 return err;
2308
2309free_used_maps:
2310 /* In case we have subprogs, we need to wait for a grace
2311 * period before we can tear down JIT memory since symbols
2312 * are already exposed under kallsyms.
2313 */
2314 __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2315 return err;
2316free_prog_sec:
2317 free_uid(prog->aux->user);
2318 security_bpf_prog_free(prog->aux);
2319free_prog:
2320 if (prog->aux->attach_btf)
2321 btf_put(prog->aux->attach_btf);
2322 bpf_prog_free(prog);
2323 return err;
2324}
2325
2326#define BPF_OBJ_LAST_FIELD file_flags
2327
2328static int bpf_obj_pin(const union bpf_attr *attr)
2329{
2330 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2331 return -EINVAL;
2332
2333 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2334}
2335
2336static int bpf_obj_get(const union bpf_attr *attr)
2337{
2338 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2339 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2340 return -EINVAL;
2341
2342 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2343 attr->file_flags);
2344}
2345
2346void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2347 const struct bpf_link_ops *ops, struct bpf_prog *prog)
2348{
2349 atomic64_set(&link->refcnt, 1);
2350 link->type = type;
2351 link->id = 0;
2352 link->ops = ops;
2353 link->prog = prog;
2354}
2355
2356static void bpf_link_free_id(int id)
2357{
2358 if (!id)
2359 return;
2360
2361 spin_lock_bh(&link_idr_lock);
2362 idr_remove(&link_idr, id);
2363 spin_unlock_bh(&link_idr_lock);
2364}
2365
2366/* Clean up bpf_link and corresponding anon_inode file and FD. After
2367 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2368 * anon_inode's release() call. This helper marksbpf_link as
2369 * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2370 * is not decremented, it's the responsibility of a calling code that failed
2371 * to complete bpf_link initialization.
2372 */
2373void bpf_link_cleanup(struct bpf_link_primer *primer)
2374{
2375 primer->link->prog = NULL;
2376 bpf_link_free_id(primer->id);
2377 fput(primer->file);
2378 put_unused_fd(primer->fd);
2379}
2380
2381void bpf_link_inc(struct bpf_link *link)
2382{
2383 atomic64_inc(&link->refcnt);
2384}
2385
2386/* bpf_link_free is guaranteed to be called from process context */
2387static void bpf_link_free(struct bpf_link *link)
2388{
2389 bpf_link_free_id(link->id);
2390 if (link->prog) {
2391 /* detach BPF program, clean up used resources */
2392 link->ops->release(link);
2393 bpf_prog_put(link->prog);
2394 }
2395 /* free bpf_link and its containing memory */
2396 link->ops->dealloc(link);
2397}
2398
2399static void bpf_link_put_deferred(struct work_struct *work)
2400{
2401 struct bpf_link *link = container_of(work, struct bpf_link, work);
2402
2403 bpf_link_free(link);
2404}
2405
2406/* bpf_link_put can be called from atomic context, but ensures that resources
2407 * are freed from process context
2408 */
2409void bpf_link_put(struct bpf_link *link)
2410{
2411 if (!atomic64_dec_and_test(&link->refcnt))
2412 return;
2413
2414 if (in_atomic()) {
2415 INIT_WORK(&link->work, bpf_link_put_deferred);
2416 schedule_work(&link->work);
2417 } else {
2418 bpf_link_free(link);
2419 }
2420}
2421
2422static int bpf_link_release(struct inode *inode, struct file *filp)
2423{
2424 struct bpf_link *link = filp->private_data;
2425
2426 bpf_link_put(link);
2427 return 0;
2428}
2429
2430#ifdef CONFIG_PROC_FS
2431#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2432#define BPF_MAP_TYPE(_id, _ops)
2433#define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2434static const char *bpf_link_type_strs[] = {
2435 [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2436#include <linux/bpf_types.h>
2437};
2438#undef BPF_PROG_TYPE
2439#undef BPF_MAP_TYPE
2440#undef BPF_LINK_TYPE
2441
2442static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2443{
2444 const struct bpf_link *link = filp->private_data;
2445 const struct bpf_prog *prog = link->prog;
2446 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2447
2448 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2449 seq_printf(m,
2450 "link_type:\t%s\n"
2451 "link_id:\t%u\n"
2452 "prog_tag:\t%s\n"
2453 "prog_id:\t%u\n",
2454 bpf_link_type_strs[link->type],
2455 link->id,
2456 prog_tag,
2457 prog->aux->id);
2458 if (link->ops->show_fdinfo)
2459 link->ops->show_fdinfo(link, m);
2460}
2461#endif
2462
2463static const struct file_operations bpf_link_fops = {
2464#ifdef CONFIG_PROC_FS
2465 .show_fdinfo = bpf_link_show_fdinfo,
2466#endif
2467 .release = bpf_link_release,
2468 .read = bpf_dummy_read,
2469 .write = bpf_dummy_write,
2470};
2471
2472static int bpf_link_alloc_id(struct bpf_link *link)
2473{
2474 int id;
2475
2476 idr_preload(GFP_KERNEL);
2477 spin_lock_bh(&link_idr_lock);
2478 id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2479 spin_unlock_bh(&link_idr_lock);
2480 idr_preload_end();
2481
2482 return id;
2483}
2484
2485/* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2486 * reserving unused FD and allocating ID from link_idr. This is to be paired
2487 * with bpf_link_settle() to install FD and ID and expose bpf_link to
2488 * user-space, if bpf_link is successfully attached. If not, bpf_link and
2489 * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2490 * transient state is passed around in struct bpf_link_primer.
2491 * This is preferred way to create and initialize bpf_link, especially when
2492 * there are complicated and expensive operations inbetween creating bpf_link
2493 * itself and attaching it to BPF hook. By using bpf_link_prime() and
2494 * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2495 * expensive (and potentially failing) roll back operations in a rare case
2496 * that file, FD, or ID can't be allocated.
2497 */
2498int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2499{
2500 struct file *file;
2501 int fd, id;
2502
2503 fd = get_unused_fd_flags(O_CLOEXEC);
2504 if (fd < 0)
2505 return fd;
2506
2507
2508 id = bpf_link_alloc_id(link);
2509 if (id < 0) {
2510 put_unused_fd(fd);
2511 return id;
2512 }
2513
2514 file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2515 if (IS_ERR(file)) {
2516 bpf_link_free_id(id);
2517 put_unused_fd(fd);
2518 return PTR_ERR(file);
2519 }
2520
2521 primer->link = link;
2522 primer->file = file;
2523 primer->fd = fd;
2524 primer->id = id;
2525 return 0;
2526}
2527
2528int bpf_link_settle(struct bpf_link_primer *primer)
2529{
2530 /* make bpf_link fetchable by ID */
2531 spin_lock_bh(&link_idr_lock);
2532 primer->link->id = primer->id;
2533 spin_unlock_bh(&link_idr_lock);
2534 /* make bpf_link fetchable by FD */
2535 fd_install(primer->fd, primer->file);
2536 /* pass through installed FD */
2537 return primer->fd;
2538}
2539
2540int bpf_link_new_fd(struct bpf_link *link)
2541{
2542 return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2543}
2544
2545struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2546{
2547 struct fd f = fdget(ufd);
2548 struct bpf_link *link;
2549
2550 if (!f.file)
2551 return ERR_PTR(-EBADF);
2552 if (f.file->f_op != &bpf_link_fops) {
2553 fdput(f);
2554 return ERR_PTR(-EINVAL);
2555 }
2556
2557 link = f.file->private_data;
2558 bpf_link_inc(link);
2559 fdput(f);
2560
2561 return link;
2562}
2563
2564struct bpf_tracing_link {
2565 struct bpf_link link;
2566 enum bpf_attach_type attach_type;
2567 struct bpf_trampoline *trampoline;
2568 struct bpf_prog *tgt_prog;
2569};
2570
2571static void bpf_tracing_link_release(struct bpf_link *link)
2572{
2573 struct bpf_tracing_link *tr_link =
2574 container_of(link, struct bpf_tracing_link, link);
2575
2576 WARN_ON_ONCE(bpf_trampoline_unlink_prog(link->prog,
2577 tr_link->trampoline));
2578
2579 bpf_trampoline_put(tr_link->trampoline);
2580
2581 /* tgt_prog is NULL if target is a kernel function */
2582 if (tr_link->tgt_prog)
2583 bpf_prog_put(tr_link->tgt_prog);
2584}
2585
2586static void bpf_tracing_link_dealloc(struct bpf_link *link)
2587{
2588 struct bpf_tracing_link *tr_link =
2589 container_of(link, struct bpf_tracing_link, link);
2590
2591 kfree(tr_link);
2592}
2593
2594static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2595 struct seq_file *seq)
2596{
2597 struct bpf_tracing_link *tr_link =
2598 container_of(link, struct bpf_tracing_link, link);
2599
2600 seq_printf(seq,
2601 "attach_type:\t%d\n",
2602 tr_link->attach_type);
2603}
2604
2605static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2606 struct bpf_link_info *info)
2607{
2608 struct bpf_tracing_link *tr_link =
2609 container_of(link, struct bpf_tracing_link, link);
2610
2611 info->tracing.attach_type = tr_link->attach_type;
2612 bpf_trampoline_unpack_key(tr_link->trampoline->key,
2613 &info->tracing.target_obj_id,
2614 &info->tracing.target_btf_id);
2615
2616 return 0;
2617}
2618
2619static const struct bpf_link_ops bpf_tracing_link_lops = {
2620 .release = bpf_tracing_link_release,
2621 .dealloc = bpf_tracing_link_dealloc,
2622 .show_fdinfo = bpf_tracing_link_show_fdinfo,
2623 .fill_link_info = bpf_tracing_link_fill_link_info,
2624};
2625
2626static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2627 int tgt_prog_fd,
2628 u32 btf_id)
2629{
2630 struct bpf_link_primer link_primer;
2631 struct bpf_prog *tgt_prog = NULL;
2632 struct bpf_trampoline *tr = NULL;
2633 struct bpf_tracing_link *link;
2634 u64 key = 0;
2635 int err;
2636
2637 switch (prog->type) {
2638 case BPF_PROG_TYPE_TRACING:
2639 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2640 prog->expected_attach_type != BPF_TRACE_FEXIT &&
2641 prog->expected_attach_type != BPF_MODIFY_RETURN) {
2642 err = -EINVAL;
2643 goto out_put_prog;
2644 }
2645 break;
2646 case BPF_PROG_TYPE_EXT:
2647 if (prog->expected_attach_type != 0) {
2648 err = -EINVAL;
2649 goto out_put_prog;
2650 }
2651 break;
2652 case BPF_PROG_TYPE_LSM:
2653 if (prog->expected_attach_type != BPF_LSM_MAC) {
2654 err = -EINVAL;
2655 goto out_put_prog;
2656 }
2657 break;
2658 default:
2659 err = -EINVAL;
2660 goto out_put_prog;
2661 }
2662
2663 if (!!tgt_prog_fd != !!btf_id) {
2664 err = -EINVAL;
2665 goto out_put_prog;
2666 }
2667
2668 if (tgt_prog_fd) {
2669 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
2670 if (prog->type != BPF_PROG_TYPE_EXT) {
2671 err = -EINVAL;
2672 goto out_put_prog;
2673 }
2674
2675 tgt_prog = bpf_prog_get(tgt_prog_fd);
2676 if (IS_ERR(tgt_prog)) {
2677 err = PTR_ERR(tgt_prog);
2678 tgt_prog = NULL;
2679 goto out_put_prog;
2680 }
2681
2682 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
2683 }
2684
2685 link = kzalloc(sizeof(*link), GFP_USER);
2686 if (!link) {
2687 err = -ENOMEM;
2688 goto out_put_prog;
2689 }
2690 bpf_link_init(&link->link, BPF_LINK_TYPE_TRACING,
2691 &bpf_tracing_link_lops, prog);
2692 link->attach_type = prog->expected_attach_type;
2693
2694 mutex_lock(&prog->aux->dst_mutex);
2695
2696 /* There are a few possible cases here:
2697 *
2698 * - if prog->aux->dst_trampoline is set, the program was just loaded
2699 * and not yet attached to anything, so we can use the values stored
2700 * in prog->aux
2701 *
2702 * - if prog->aux->dst_trampoline is NULL, the program has already been
2703 * attached to a target and its initial target was cleared (below)
2704 *
2705 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
2706 * target_btf_id using the link_create API.
2707 *
2708 * - if tgt_prog == NULL when this function was called using the old
2709 * raw_tracepoint_open API, and we need a target from prog->aux
2710 *
2711 * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
2712 * was detached and is going for re-attachment.
2713 */
2714 if (!prog->aux->dst_trampoline && !tgt_prog) {
2715 /*
2716 * Allow re-attach for TRACING and LSM programs. If it's
2717 * currently linked, bpf_trampoline_link_prog will fail.
2718 * EXT programs need to specify tgt_prog_fd, so they
2719 * re-attach in separate code path.
2720 */
2721 if (prog->type != BPF_PROG_TYPE_TRACING &&
2722 prog->type != BPF_PROG_TYPE_LSM) {
2723 err = -EINVAL;
2724 goto out_unlock;
2725 }
2726 btf_id = prog->aux->attach_btf_id;
2727 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
2728 }
2729
2730 if (!prog->aux->dst_trampoline ||
2731 (key && key != prog->aux->dst_trampoline->key)) {
2732 /* If there is no saved target, or the specified target is
2733 * different from the destination specified at load time, we
2734 * need a new trampoline and a check for compatibility
2735 */
2736 struct bpf_attach_target_info tgt_info = {};
2737
2738 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
2739 &tgt_info);
2740 if (err)
2741 goto out_unlock;
2742
2743 tr = bpf_trampoline_get(key, &tgt_info);
2744 if (!tr) {
2745 err = -ENOMEM;
2746 goto out_unlock;
2747 }
2748 } else {
2749 /* The caller didn't specify a target, or the target was the
2750 * same as the destination supplied during program load. This
2751 * means we can reuse the trampoline and reference from program
2752 * load time, and there is no need to allocate a new one. This
2753 * can only happen once for any program, as the saved values in
2754 * prog->aux are cleared below.
2755 */
2756 tr = prog->aux->dst_trampoline;
2757 tgt_prog = prog->aux->dst_prog;
2758 }
2759
2760 err = bpf_link_prime(&link->link, &link_primer);
2761 if (err)
2762 goto out_unlock;
2763
2764 err = bpf_trampoline_link_prog(prog, tr);
2765 if (err) {
2766 bpf_link_cleanup(&link_primer);
2767 link = NULL;
2768 goto out_unlock;
2769 }
2770
2771 link->tgt_prog = tgt_prog;
2772 link->trampoline = tr;
2773
2774 /* Always clear the trampoline and target prog from prog->aux to make
2775 * sure the original attach destination is not kept alive after a
2776 * program is (re-)attached to another target.
2777 */
2778 if (prog->aux->dst_prog &&
2779 (tgt_prog_fd || tr != prog->aux->dst_trampoline))
2780 /* got extra prog ref from syscall, or attaching to different prog */
2781 bpf_prog_put(prog->aux->dst_prog);
2782 if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
2783 /* we allocated a new trampoline, so free the old one */
2784 bpf_trampoline_put(prog->aux->dst_trampoline);
2785
2786 prog->aux->dst_prog = NULL;
2787 prog->aux->dst_trampoline = NULL;
2788 mutex_unlock(&prog->aux->dst_mutex);
2789
2790 return bpf_link_settle(&link_primer);
2791out_unlock:
2792 if (tr && tr != prog->aux->dst_trampoline)
2793 bpf_trampoline_put(tr);
2794 mutex_unlock(&prog->aux->dst_mutex);
2795 kfree(link);
2796out_put_prog:
2797 if (tgt_prog_fd && tgt_prog)
2798 bpf_prog_put(tgt_prog);
2799 return err;
2800}
2801
2802struct bpf_raw_tp_link {
2803 struct bpf_link link;
2804 struct bpf_raw_event_map *btp;
2805};
2806
2807static void bpf_raw_tp_link_release(struct bpf_link *link)
2808{
2809 struct bpf_raw_tp_link *raw_tp =
2810 container_of(link, struct bpf_raw_tp_link, link);
2811
2812 bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
2813 bpf_put_raw_tracepoint(raw_tp->btp);
2814}
2815
2816static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
2817{
2818 struct bpf_raw_tp_link *raw_tp =
2819 container_of(link, struct bpf_raw_tp_link, link);
2820
2821 kfree(raw_tp);
2822}
2823
2824static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
2825 struct seq_file *seq)
2826{
2827 struct bpf_raw_tp_link *raw_tp_link =
2828 container_of(link, struct bpf_raw_tp_link, link);
2829
2830 seq_printf(seq,
2831 "tp_name:\t%s\n",
2832 raw_tp_link->btp->tp->name);
2833}
2834
2835static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
2836 struct bpf_link_info *info)
2837{
2838 struct bpf_raw_tp_link *raw_tp_link =
2839 container_of(link, struct bpf_raw_tp_link, link);
2840 char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
2841 const char *tp_name = raw_tp_link->btp->tp->name;
2842 u32 ulen = info->raw_tracepoint.tp_name_len;
2843 size_t tp_len = strlen(tp_name);
2844
2845 if (!ulen ^ !ubuf)
2846 return -EINVAL;
2847
2848 info->raw_tracepoint.tp_name_len = tp_len + 1;
2849
2850 if (!ubuf)
2851 return 0;
2852
2853 if (ulen >= tp_len + 1) {
2854 if (copy_to_user(ubuf, tp_name, tp_len + 1))
2855 return -EFAULT;
2856 } else {
2857 char zero = '\0';
2858
2859 if (copy_to_user(ubuf, tp_name, ulen - 1))
2860 return -EFAULT;
2861 if (put_user(zero, ubuf + ulen - 1))
2862 return -EFAULT;
2863 return -ENOSPC;
2864 }
2865
2866 return 0;
2867}
2868
2869static const struct bpf_link_ops bpf_raw_tp_link_lops = {
2870 .release = bpf_raw_tp_link_release,
2871 .dealloc = bpf_raw_tp_link_dealloc,
2872 .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
2873 .fill_link_info = bpf_raw_tp_link_fill_link_info,
2874};
2875
2876#define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
2877
2878static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
2879{
2880 struct bpf_link_primer link_primer;
2881 struct bpf_raw_tp_link *link;
2882 struct bpf_raw_event_map *btp;
2883 struct bpf_prog *prog;
2884 const char *tp_name;
2885 char buf[128];
2886 int err;
2887
2888 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
2889 return -EINVAL;
2890
2891 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
2892 if (IS_ERR(prog))
2893 return PTR_ERR(prog);
2894
2895 switch (prog->type) {
2896 case BPF_PROG_TYPE_TRACING:
2897 case BPF_PROG_TYPE_EXT:
2898 case BPF_PROG_TYPE_LSM:
2899 if (attr->raw_tracepoint.name) {
2900 /* The attach point for this category of programs
2901 * should be specified via btf_id during program load.
2902 */
2903 err = -EINVAL;
2904 goto out_put_prog;
2905 }
2906 if (prog->type == BPF_PROG_TYPE_TRACING &&
2907 prog->expected_attach_type == BPF_TRACE_RAW_TP) {
2908 tp_name = prog->aux->attach_func_name;
2909 break;
2910 }
2911 err = bpf_tracing_prog_attach(prog, 0, 0);
2912 if (err >= 0)
2913 return err;
2914 goto out_put_prog;
2915 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2916 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2917 if (strncpy_from_user(buf,
2918 u64_to_user_ptr(attr->raw_tracepoint.name),
2919 sizeof(buf) - 1) < 0) {
2920 err = -EFAULT;
2921 goto out_put_prog;
2922 }
2923 buf[sizeof(buf) - 1] = 0;
2924 tp_name = buf;
2925 break;
2926 default:
2927 err = -EINVAL;
2928 goto out_put_prog;
2929 }
2930
2931 btp = bpf_get_raw_tracepoint(tp_name);
2932 if (!btp) {
2933 err = -ENOENT;
2934 goto out_put_prog;
2935 }
2936
2937 link = kzalloc(sizeof(*link), GFP_USER);
2938 if (!link) {
2939 err = -ENOMEM;
2940 goto out_put_btp;
2941 }
2942 bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
2943 &bpf_raw_tp_link_lops, prog);
2944 link->btp = btp;
2945
2946 err = bpf_link_prime(&link->link, &link_primer);
2947 if (err) {
2948 kfree(link);
2949 goto out_put_btp;
2950 }
2951
2952 err = bpf_probe_register(link->btp, prog);
2953 if (err) {
2954 bpf_link_cleanup(&link_primer);
2955 goto out_put_btp;
2956 }
2957
2958 return bpf_link_settle(&link_primer);
2959
2960out_put_btp:
2961 bpf_put_raw_tracepoint(btp);
2962out_put_prog:
2963 bpf_prog_put(prog);
2964 return err;
2965}
2966
2967static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2968 enum bpf_attach_type attach_type)
2969{
2970 switch (prog->type) {
2971 case BPF_PROG_TYPE_CGROUP_SOCK:
2972 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2973 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2974 case BPF_PROG_TYPE_SK_LOOKUP:
2975 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2976 case BPF_PROG_TYPE_CGROUP_SKB:
2977 if (!capable(CAP_NET_ADMIN))
2978 /* cg-skb progs can be loaded by unpriv user.
2979 * check permissions at attach time.
2980 */
2981 return -EPERM;
2982 return prog->enforce_expected_attach_type &&
2983 prog->expected_attach_type != attach_type ?
2984 -EINVAL : 0;
2985 default:
2986 return 0;
2987 }
2988}
2989
2990static enum bpf_prog_type
2991attach_type_to_prog_type(enum bpf_attach_type attach_type)
2992{
2993 switch (attach_type) {
2994 case BPF_CGROUP_INET_INGRESS:
2995 case BPF_CGROUP_INET_EGRESS:
2996 return BPF_PROG_TYPE_CGROUP_SKB;
2997 case BPF_CGROUP_INET_SOCK_CREATE:
2998 case BPF_CGROUP_INET_SOCK_RELEASE:
2999 case BPF_CGROUP_INET4_POST_BIND:
3000 case BPF_CGROUP_INET6_POST_BIND:
3001 return BPF_PROG_TYPE_CGROUP_SOCK;
3002 case BPF_CGROUP_INET4_BIND:
3003 case BPF_CGROUP_INET6_BIND:
3004 case BPF_CGROUP_INET4_CONNECT:
3005 case BPF_CGROUP_INET6_CONNECT:
3006 case BPF_CGROUP_INET4_GETPEERNAME:
3007 case BPF_CGROUP_INET6_GETPEERNAME:
3008 case BPF_CGROUP_INET4_GETSOCKNAME:
3009 case BPF_CGROUP_INET6_GETSOCKNAME:
3010 case BPF_CGROUP_UDP4_SENDMSG:
3011 case BPF_CGROUP_UDP6_SENDMSG:
3012 case BPF_CGROUP_UDP4_RECVMSG:
3013 case BPF_CGROUP_UDP6_RECVMSG:
3014 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3015 case BPF_CGROUP_SOCK_OPS:
3016 return BPF_PROG_TYPE_SOCK_OPS;
3017 case BPF_CGROUP_DEVICE:
3018 return BPF_PROG_TYPE_CGROUP_DEVICE;
3019 case BPF_SK_MSG_VERDICT:
3020 return BPF_PROG_TYPE_SK_MSG;
3021 case BPF_SK_SKB_STREAM_PARSER:
3022 case BPF_SK_SKB_STREAM_VERDICT:
3023 case BPF_SK_SKB_VERDICT:
3024 return BPF_PROG_TYPE_SK_SKB;
3025 case BPF_LIRC_MODE2:
3026 return BPF_PROG_TYPE_LIRC_MODE2;
3027 case BPF_FLOW_DISSECTOR:
3028 return BPF_PROG_TYPE_FLOW_DISSECTOR;
3029 case BPF_CGROUP_SYSCTL:
3030 return BPF_PROG_TYPE_CGROUP_SYSCTL;
3031 case BPF_CGROUP_GETSOCKOPT:
3032 case BPF_CGROUP_SETSOCKOPT:
3033 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3034 case BPF_TRACE_ITER:
3035 return BPF_PROG_TYPE_TRACING;
3036 case BPF_SK_LOOKUP:
3037 return BPF_PROG_TYPE_SK_LOOKUP;
3038 case BPF_XDP:
3039 return BPF_PROG_TYPE_XDP;
3040 default:
3041 return BPF_PROG_TYPE_UNSPEC;
3042 }
3043}
3044
3045#define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
3046
3047#define BPF_F_ATTACH_MASK \
3048 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
3049
3050static int bpf_prog_attach(const union bpf_attr *attr)
3051{
3052 enum bpf_prog_type ptype;
3053 struct bpf_prog *prog;
3054 int ret;
3055
3056 if (CHECK_ATTR(BPF_PROG_ATTACH))
3057 return -EINVAL;
3058
3059 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
3060 return -EINVAL;
3061
3062 ptype = attach_type_to_prog_type(attr->attach_type);
3063 if (ptype == BPF_PROG_TYPE_UNSPEC)
3064 return -EINVAL;
3065
3066 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3067 if (IS_ERR(prog))
3068 return PTR_ERR(prog);
3069
3070 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3071 bpf_prog_put(prog);
3072 return -EINVAL;
3073 }
3074
3075 switch (ptype) {
3076 case BPF_PROG_TYPE_SK_SKB:
3077 case BPF_PROG_TYPE_SK_MSG:
3078 ret = sock_map_get_from_fd(attr, prog);
3079 break;
3080 case BPF_PROG_TYPE_LIRC_MODE2:
3081 ret = lirc_prog_attach(attr, prog);
3082 break;
3083 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3084 ret = netns_bpf_prog_attach(attr, prog);
3085 break;
3086 case BPF_PROG_TYPE_CGROUP_DEVICE:
3087 case BPF_PROG_TYPE_CGROUP_SKB:
3088 case BPF_PROG_TYPE_CGROUP_SOCK:
3089 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3090 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3091 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3092 case BPF_PROG_TYPE_SOCK_OPS:
3093 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3094 break;
3095 default:
3096 ret = -EINVAL;
3097 }
3098
3099 if (ret)
3100 bpf_prog_put(prog);
3101 return ret;
3102}
3103
3104#define BPF_PROG_DETACH_LAST_FIELD attach_type
3105
3106static int bpf_prog_detach(const union bpf_attr *attr)
3107{
3108 enum bpf_prog_type ptype;
3109
3110 if (CHECK_ATTR(BPF_PROG_DETACH))
3111 return -EINVAL;
3112
3113 ptype = attach_type_to_prog_type(attr->attach_type);
3114
3115 switch (ptype) {
3116 case BPF_PROG_TYPE_SK_MSG:
3117 case BPF_PROG_TYPE_SK_SKB:
3118 return sock_map_prog_detach(attr, ptype);
3119 case BPF_PROG_TYPE_LIRC_MODE2:
3120 return lirc_prog_detach(attr);
3121 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3122 return netns_bpf_prog_detach(attr, ptype);
3123 case BPF_PROG_TYPE_CGROUP_DEVICE:
3124 case BPF_PROG_TYPE_CGROUP_SKB:
3125 case BPF_PROG_TYPE_CGROUP_SOCK:
3126 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3127 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3128 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3129 case BPF_PROG_TYPE_SOCK_OPS:
3130 return cgroup_bpf_prog_detach(attr, ptype);
3131 default:
3132 return -EINVAL;
3133 }
3134}
3135
3136#define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
3137
3138static int bpf_prog_query(const union bpf_attr *attr,
3139 union bpf_attr __user *uattr)
3140{
3141 if (!capable(CAP_NET_ADMIN))
3142 return -EPERM;
3143 if (CHECK_ATTR(BPF_PROG_QUERY))
3144 return -EINVAL;
3145 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3146 return -EINVAL;
3147
3148 switch (attr->query.attach_type) {
3149 case BPF_CGROUP_INET_INGRESS:
3150 case BPF_CGROUP_INET_EGRESS:
3151 case BPF_CGROUP_INET_SOCK_CREATE:
3152 case BPF_CGROUP_INET_SOCK_RELEASE:
3153 case BPF_CGROUP_INET4_BIND:
3154 case BPF_CGROUP_INET6_BIND:
3155 case BPF_CGROUP_INET4_POST_BIND:
3156 case BPF_CGROUP_INET6_POST_BIND:
3157 case BPF_CGROUP_INET4_CONNECT:
3158 case BPF_CGROUP_INET6_CONNECT:
3159 case BPF_CGROUP_INET4_GETPEERNAME:
3160 case BPF_CGROUP_INET6_GETPEERNAME:
3161 case BPF_CGROUP_INET4_GETSOCKNAME:
3162 case BPF_CGROUP_INET6_GETSOCKNAME:
3163 case BPF_CGROUP_UDP4_SENDMSG:
3164 case BPF_CGROUP_UDP6_SENDMSG:
3165 case BPF_CGROUP_UDP4_RECVMSG:
3166 case BPF_CGROUP_UDP6_RECVMSG:
3167 case BPF_CGROUP_SOCK_OPS:
3168 case BPF_CGROUP_DEVICE:
3169 case BPF_CGROUP_SYSCTL:
3170 case BPF_CGROUP_GETSOCKOPT:
3171 case BPF_CGROUP_SETSOCKOPT:
3172 return cgroup_bpf_prog_query(attr, uattr);
3173 case BPF_LIRC_MODE2:
3174 return lirc_prog_query(attr, uattr);
3175 case BPF_FLOW_DISSECTOR:
3176 case BPF_SK_LOOKUP:
3177 return netns_bpf_prog_query(attr, uattr);
3178 default:
3179 return -EINVAL;
3180 }
3181}
3182
3183#define BPF_PROG_TEST_RUN_LAST_FIELD test.cpu
3184
3185static int bpf_prog_test_run(const union bpf_attr *attr,
3186 union bpf_attr __user *uattr)
3187{
3188 struct bpf_prog *prog;
3189 int ret = -ENOTSUPP;
3190
3191 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3192 return -EINVAL;
3193
3194 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3195 (!attr->test.ctx_size_in && attr->test.ctx_in))
3196 return -EINVAL;
3197
3198 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3199 (!attr->test.ctx_size_out && attr->test.ctx_out))
3200 return -EINVAL;
3201
3202 prog = bpf_prog_get(attr->test.prog_fd);
3203 if (IS_ERR(prog))
3204 return PTR_ERR(prog);
3205
3206 if (prog->aux->ops->test_run)
3207 ret = prog->aux->ops->test_run(prog, attr, uattr);
3208
3209 bpf_prog_put(prog);
3210 return ret;
3211}
3212
3213#define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3214
3215static int bpf_obj_get_next_id(const union bpf_attr *attr,
3216 union bpf_attr __user *uattr,
3217 struct idr *idr,
3218 spinlock_t *lock)
3219{
3220 u32 next_id = attr->start_id;
3221 int err = 0;
3222
3223 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3224 return -EINVAL;
3225
3226 if (!capable(CAP_SYS_ADMIN))
3227 return -EPERM;
3228
3229 next_id++;
3230 spin_lock_bh(lock);
3231 if (!idr_get_next(idr, &next_id))
3232 err = -ENOENT;
3233 spin_unlock_bh(lock);
3234
3235 if (!err)
3236 err = put_user(next_id, &uattr->next_id);
3237
3238 return err;
3239}
3240
3241struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3242{
3243 struct bpf_map *map;
3244
3245 spin_lock_bh(&map_idr_lock);
3246again:
3247 map = idr_get_next(&map_idr, id);
3248 if (map) {
3249 map = __bpf_map_inc_not_zero(map, false);
3250 if (IS_ERR(map)) {
3251 (*id)++;
3252 goto again;
3253 }
3254 }
3255 spin_unlock_bh(&map_idr_lock);
3256
3257 return map;
3258}
3259
3260struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3261{
3262 struct bpf_prog *prog;
3263
3264 spin_lock_bh(&prog_idr_lock);
3265again:
3266 prog = idr_get_next(&prog_idr, id);
3267 if (prog) {
3268 prog = bpf_prog_inc_not_zero(prog);
3269 if (IS_ERR(prog)) {
3270 (*id)++;
3271 goto again;
3272 }
3273 }
3274 spin_unlock_bh(&prog_idr_lock);
3275
3276 return prog;
3277}
3278
3279#define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3280
3281struct bpf_prog *bpf_prog_by_id(u32 id)
3282{
3283 struct bpf_prog *prog;
3284
3285 if (!id)
3286 return ERR_PTR(-ENOENT);
3287
3288 spin_lock_bh(&prog_idr_lock);
3289 prog = idr_find(&prog_idr, id);
3290 if (prog)
3291 prog = bpf_prog_inc_not_zero(prog);
3292 else
3293 prog = ERR_PTR(-ENOENT);
3294 spin_unlock_bh(&prog_idr_lock);
3295 return prog;
3296}
3297
3298static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3299{
3300 struct bpf_prog *prog;
3301 u32 id = attr->prog_id;
3302 int fd;
3303
3304 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3305 return -EINVAL;
3306
3307 if (!capable(CAP_SYS_ADMIN))
3308 return -EPERM;
3309
3310 prog = bpf_prog_by_id(id);
3311 if (IS_ERR(prog))
3312 return PTR_ERR(prog);
3313
3314 fd = bpf_prog_new_fd(prog);
3315 if (fd < 0)
3316 bpf_prog_put(prog);
3317
3318 return fd;
3319}
3320
3321#define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3322
3323static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3324{
3325 struct bpf_map *map;
3326 u32 id = attr->map_id;
3327 int f_flags;
3328 int fd;
3329
3330 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3331 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3332 return -EINVAL;
3333
3334 if (!capable(CAP_SYS_ADMIN))
3335 return -EPERM;
3336
3337 f_flags = bpf_get_file_flag(attr->open_flags);
3338 if (f_flags < 0)
3339 return f_flags;
3340
3341 spin_lock_bh(&map_idr_lock);
3342 map = idr_find(&map_idr, id);
3343 if (map)
3344 map = __bpf_map_inc_not_zero(map, true);
3345 else
3346 map = ERR_PTR(-ENOENT);
3347 spin_unlock_bh(&map_idr_lock);
3348
3349 if (IS_ERR(map))
3350 return PTR_ERR(map);
3351
3352 fd = bpf_map_new_fd(map, f_flags);
3353 if (fd < 0)
3354 bpf_map_put_with_uref(map);
3355
3356 return fd;
3357}
3358
3359static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3360 unsigned long addr, u32 *off,
3361 u32 *type)
3362{
3363 const struct bpf_map *map;
3364 int i;
3365
3366 mutex_lock(&prog->aux->used_maps_mutex);
3367 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3368 map = prog->aux->used_maps[i];
3369 if (map == (void *)addr) {
3370 *type = BPF_PSEUDO_MAP_FD;
3371 goto out;
3372 }
3373 if (!map->ops->map_direct_value_meta)
3374 continue;
3375 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3376 *type = BPF_PSEUDO_MAP_VALUE;
3377 goto out;
3378 }
3379 }
3380 map = NULL;
3381
3382out:
3383 mutex_unlock(&prog->aux->used_maps_mutex);
3384 return map;
3385}
3386
3387static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3388 const struct cred *f_cred)
3389{
3390 const struct bpf_map *map;
3391 struct bpf_insn *insns;
3392 u32 off, type;
3393 u64 imm;
3394 u8 code;
3395 int i;
3396
3397 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3398 GFP_USER);
3399 if (!insns)
3400 return insns;
3401
3402 for (i = 0; i < prog->len; i++) {
3403 code = insns[i].code;
3404
3405 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3406 insns[i].code = BPF_JMP | BPF_CALL;
3407 insns[i].imm = BPF_FUNC_tail_call;
3408 /* fall-through */
3409 }
3410 if (code == (BPF_JMP | BPF_CALL) ||
3411 code == (BPF_JMP | BPF_CALL_ARGS)) {
3412 if (code == (BPF_JMP | BPF_CALL_ARGS))
3413 insns[i].code = BPF_JMP | BPF_CALL;
3414 if (!bpf_dump_raw_ok(f_cred))
3415 insns[i].imm = 0;
3416 continue;
3417 }
3418 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3419 insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3420 continue;
3421 }
3422
3423 if (code != (BPF_LD | BPF_IMM | BPF_DW))
3424 continue;
3425
3426 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3427 map = bpf_map_from_imm(prog, imm, &off, &type);
3428 if (map) {
3429 insns[i].src_reg = type;
3430 insns[i].imm = map->id;
3431 insns[i + 1].imm = off;
3432 continue;
3433 }
3434 }
3435
3436 return insns;
3437}
3438
3439static int set_info_rec_size(struct bpf_prog_info *info)
3440{
3441 /*
3442 * Ensure info.*_rec_size is the same as kernel expected size
3443 *
3444 * or
3445 *
3446 * Only allow zero *_rec_size if both _rec_size and _cnt are
3447 * zero. In this case, the kernel will set the expected
3448 * _rec_size back to the info.
3449 */
3450
3451 if ((info->nr_func_info || info->func_info_rec_size) &&
3452 info->func_info_rec_size != sizeof(struct bpf_func_info))
3453 return -EINVAL;
3454
3455 if ((info->nr_line_info || info->line_info_rec_size) &&
3456 info->line_info_rec_size != sizeof(struct bpf_line_info))
3457 return -EINVAL;
3458
3459 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3460 info->jited_line_info_rec_size != sizeof(__u64))
3461 return -EINVAL;
3462
3463 info->func_info_rec_size = sizeof(struct bpf_func_info);
3464 info->line_info_rec_size = sizeof(struct bpf_line_info);
3465 info->jited_line_info_rec_size = sizeof(__u64);
3466
3467 return 0;
3468}
3469
3470static int bpf_prog_get_info_by_fd(struct file *file,
3471 struct bpf_prog *prog,
3472 const union bpf_attr *attr,
3473 union bpf_attr __user *uattr)
3474{
3475 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3476 struct bpf_prog_info info;
3477 u32 info_len = attr->info.info_len;
3478 struct bpf_prog_stats stats;
3479 char __user *uinsns;
3480 u32 ulen;
3481 int err;
3482
3483 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
3484 if (err)
3485 return err;
3486 info_len = min_t(u32, sizeof(info), info_len);
3487
3488 memset(&info, 0, sizeof(info));
3489 if (copy_from_user(&info, uinfo, info_len))
3490 return -EFAULT;
3491
3492 info.type = prog->type;
3493 info.id = prog->aux->id;
3494 info.load_time = prog->aux->load_time;
3495 info.created_by_uid = from_kuid_munged(current_user_ns(),
3496 prog->aux->user->uid);
3497 info.gpl_compatible = prog->gpl_compatible;
3498
3499 memcpy(info.tag, prog->tag, sizeof(prog->tag));
3500 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3501
3502 mutex_lock(&prog->aux->used_maps_mutex);
3503 ulen = info.nr_map_ids;
3504 info.nr_map_ids = prog->aux->used_map_cnt;
3505 ulen = min_t(u32, info.nr_map_ids, ulen);
3506 if (ulen) {
3507 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3508 u32 i;
3509
3510 for (i = 0; i < ulen; i++)
3511 if (put_user(prog->aux->used_maps[i]->id,
3512 &user_map_ids[i])) {
3513 mutex_unlock(&prog->aux->used_maps_mutex);
3514 return -EFAULT;
3515 }
3516 }
3517 mutex_unlock(&prog->aux->used_maps_mutex);
3518
3519 err = set_info_rec_size(&info);
3520 if (err)
3521 return err;
3522
3523 bpf_prog_get_stats(prog, &stats);
3524 info.run_time_ns = stats.nsecs;
3525 info.run_cnt = stats.cnt;
3526 info.recursion_misses = stats.misses;
3527
3528 if (!bpf_capable()) {
3529 info.jited_prog_len = 0;
3530 info.xlated_prog_len = 0;
3531 info.nr_jited_ksyms = 0;
3532 info.nr_jited_func_lens = 0;
3533 info.nr_func_info = 0;
3534 info.nr_line_info = 0;
3535 info.nr_jited_line_info = 0;
3536 goto done;
3537 }
3538
3539 ulen = info.xlated_prog_len;
3540 info.xlated_prog_len = bpf_prog_insn_size(prog);
3541 if (info.xlated_prog_len && ulen) {
3542 struct bpf_insn *insns_sanitized;
3543 bool fault;
3544
3545 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
3546 info.xlated_prog_insns = 0;
3547 goto done;
3548 }
3549 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
3550 if (!insns_sanitized)
3551 return -ENOMEM;
3552 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3553 ulen = min_t(u32, info.xlated_prog_len, ulen);
3554 fault = copy_to_user(uinsns, insns_sanitized, ulen);
3555 kfree(insns_sanitized);
3556 if (fault)
3557 return -EFAULT;
3558 }
3559
3560 if (bpf_prog_is_dev_bound(prog->aux)) {
3561 err = bpf_prog_offload_info_fill(&info, prog);
3562 if (err)
3563 return err;
3564 goto done;
3565 }
3566
3567 /* NOTE: the following code is supposed to be skipped for offload.
3568 * bpf_prog_offload_info_fill() is the place to fill similar fields
3569 * for offload.
3570 */
3571 ulen = info.jited_prog_len;
3572 if (prog->aux->func_cnt) {
3573 u32 i;
3574
3575 info.jited_prog_len = 0;
3576 for (i = 0; i < prog->aux->func_cnt; i++)
3577 info.jited_prog_len += prog->aux->func[i]->jited_len;
3578 } else {
3579 info.jited_prog_len = prog->jited_len;
3580 }
3581
3582 if (info.jited_prog_len && ulen) {
3583 if (bpf_dump_raw_ok(file->f_cred)) {
3584 uinsns = u64_to_user_ptr(info.jited_prog_insns);
3585 ulen = min_t(u32, info.jited_prog_len, ulen);
3586
3587 /* for multi-function programs, copy the JITed
3588 * instructions for all the functions
3589 */
3590 if (prog->aux->func_cnt) {
3591 u32 len, free, i;
3592 u8 *img;
3593
3594 free = ulen;
3595 for (i = 0; i < prog->aux->func_cnt; i++) {
3596 len = prog->aux->func[i]->jited_len;
3597 len = min_t(u32, len, free);
3598 img = (u8 *) prog->aux->func[i]->bpf_func;
3599 if (copy_to_user(uinsns, img, len))
3600 return -EFAULT;
3601 uinsns += len;
3602 free -= len;
3603 if (!free)
3604 break;
3605 }
3606 } else {
3607 if (copy_to_user(uinsns, prog->bpf_func, ulen))
3608 return -EFAULT;
3609 }
3610 } else {
3611 info.jited_prog_insns = 0;
3612 }
3613 }
3614
3615 ulen = info.nr_jited_ksyms;
3616 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
3617 if (ulen) {
3618 if (bpf_dump_raw_ok(file->f_cred)) {
3619 unsigned long ksym_addr;
3620 u64 __user *user_ksyms;
3621 u32 i;
3622
3623 /* copy the address of the kernel symbol
3624 * corresponding to each function
3625 */
3626 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
3627 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
3628 if (prog->aux->func_cnt) {
3629 for (i = 0; i < ulen; i++) {
3630 ksym_addr = (unsigned long)
3631 prog->aux->func[i]->bpf_func;
3632 if (put_user((u64) ksym_addr,
3633 &user_ksyms[i]))
3634 return -EFAULT;
3635 }
3636 } else {
3637 ksym_addr = (unsigned long) prog->bpf_func;
3638 if (put_user((u64) ksym_addr, &user_ksyms[0]))
3639 return -EFAULT;
3640 }
3641 } else {
3642 info.jited_ksyms = 0;
3643 }
3644 }
3645
3646 ulen = info.nr_jited_func_lens;
3647 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
3648 if (ulen) {
3649 if (bpf_dump_raw_ok(file->f_cred)) {
3650 u32 __user *user_lens;
3651 u32 func_len, i;
3652
3653 /* copy the JITed image lengths for each function */
3654 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
3655 user_lens = u64_to_user_ptr(info.jited_func_lens);
3656 if (prog->aux->func_cnt) {
3657 for (i = 0; i < ulen; i++) {
3658 func_len =
3659 prog->aux->func[i]->jited_len;
3660 if (put_user(func_len, &user_lens[i]))
3661 return -EFAULT;
3662 }
3663 } else {
3664 func_len = prog->jited_len;
3665 if (put_user(func_len, &user_lens[0]))
3666 return -EFAULT;
3667 }
3668 } else {
3669 info.jited_func_lens = 0;
3670 }
3671 }
3672
3673 if (prog->aux->btf)
3674 info.btf_id = btf_obj_id(prog->aux->btf);
3675
3676 ulen = info.nr_func_info;
3677 info.nr_func_info = prog->aux->func_info_cnt;
3678 if (info.nr_func_info && ulen) {
3679 char __user *user_finfo;
3680
3681 user_finfo = u64_to_user_ptr(info.func_info);
3682 ulen = min_t(u32, info.nr_func_info, ulen);
3683 if (copy_to_user(user_finfo, prog->aux->func_info,
3684 info.func_info_rec_size * ulen))
3685 return -EFAULT;
3686 }
3687
3688 ulen = info.nr_line_info;
3689 info.nr_line_info = prog->aux->nr_linfo;
3690 if (info.nr_line_info && ulen) {
3691 __u8 __user *user_linfo;
3692
3693 user_linfo = u64_to_user_ptr(info.line_info);
3694 ulen = min_t(u32, info.nr_line_info, ulen);
3695 if (copy_to_user(user_linfo, prog->aux->linfo,
3696 info.line_info_rec_size * ulen))
3697 return -EFAULT;
3698 }
3699
3700 ulen = info.nr_jited_line_info;
3701 if (prog->aux->jited_linfo)
3702 info.nr_jited_line_info = prog->aux->nr_linfo;
3703 else
3704 info.nr_jited_line_info = 0;
3705 if (info.nr_jited_line_info && ulen) {
3706 if (bpf_dump_raw_ok(file->f_cred)) {
3707 __u64 __user *user_linfo;
3708 u32 i;
3709
3710 user_linfo = u64_to_user_ptr(info.jited_line_info);
3711 ulen = min_t(u32, info.nr_jited_line_info, ulen);
3712 for (i = 0; i < ulen; i++) {
3713 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
3714 &user_linfo[i]))
3715 return -EFAULT;
3716 }
3717 } else {
3718 info.jited_line_info = 0;
3719 }
3720 }
3721
3722 ulen = info.nr_prog_tags;
3723 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
3724 if (ulen) {
3725 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
3726 u32 i;
3727
3728 user_prog_tags = u64_to_user_ptr(info.prog_tags);
3729 ulen = min_t(u32, info.nr_prog_tags, ulen);
3730 if (prog->aux->func_cnt) {
3731 for (i = 0; i < ulen; i++) {
3732 if (copy_to_user(user_prog_tags[i],
3733 prog->aux->func[i]->tag,
3734 BPF_TAG_SIZE))
3735 return -EFAULT;
3736 }
3737 } else {
3738 if (copy_to_user(user_prog_tags[0],
3739 prog->tag, BPF_TAG_SIZE))
3740 return -EFAULT;
3741 }
3742 }
3743
3744done:
3745 if (copy_to_user(uinfo, &info, info_len) ||
3746 put_user(info_len, &uattr->info.info_len))
3747 return -EFAULT;
3748
3749 return 0;
3750}
3751
3752static int bpf_map_get_info_by_fd(struct file *file,
3753 struct bpf_map *map,
3754 const union bpf_attr *attr,
3755 union bpf_attr __user *uattr)
3756{
3757 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3758 struct bpf_map_info info;
3759 u32 info_len = attr->info.info_len;
3760 int err;
3761
3762 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
3763 if (err)
3764 return err;
3765 info_len = min_t(u32, sizeof(info), info_len);
3766
3767 memset(&info, 0, sizeof(info));
3768 info.type = map->map_type;
3769 info.id = map->id;
3770 info.key_size = map->key_size;
3771 info.value_size = map->value_size;
3772 info.max_entries = map->max_entries;
3773 info.map_flags = map->map_flags;
3774 memcpy(info.name, map->name, sizeof(map->name));
3775
3776 if (map->btf) {
3777 info.btf_id = btf_obj_id(map->btf);
3778 info.btf_key_type_id = map->btf_key_type_id;
3779 info.btf_value_type_id = map->btf_value_type_id;
3780 }
3781 info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
3782
3783 if (bpf_map_is_dev_bound(map)) {
3784 err = bpf_map_offload_info_fill(&info, map);
3785 if (err)
3786 return err;
3787 }
3788
3789 if (copy_to_user(uinfo, &info, info_len) ||
3790 put_user(info_len, &uattr->info.info_len))
3791 return -EFAULT;
3792
3793 return 0;
3794}
3795
3796static int bpf_btf_get_info_by_fd(struct file *file,
3797 struct btf *btf,
3798 const union bpf_attr *attr,
3799 union bpf_attr __user *uattr)
3800{
3801 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3802 u32 info_len = attr->info.info_len;
3803 int err;
3804
3805 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
3806 if (err)
3807 return err;
3808
3809 return btf_get_info_by_fd(btf, attr, uattr);
3810}
3811
3812static int bpf_link_get_info_by_fd(struct file *file,
3813 struct bpf_link *link,
3814 const union bpf_attr *attr,
3815 union bpf_attr __user *uattr)
3816{
3817 struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3818 struct bpf_link_info info;
3819 u32 info_len = attr->info.info_len;
3820 int err;
3821
3822 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
3823 if (err)
3824 return err;
3825 info_len = min_t(u32, sizeof(info), info_len);
3826
3827 memset(&info, 0, sizeof(info));
3828 if (copy_from_user(&info, uinfo, info_len))
3829 return -EFAULT;
3830
3831 info.type = link->type;
3832 info.id = link->id;
3833 info.prog_id = link->prog->aux->id;
3834
3835 if (link->ops->fill_link_info) {
3836 err = link->ops->fill_link_info(link, &info);
3837 if (err)
3838 return err;
3839 }
3840
3841 if (copy_to_user(uinfo, &info, info_len) ||
3842 put_user(info_len, &uattr->info.info_len))
3843 return -EFAULT;
3844
3845 return 0;
3846}
3847
3848
3849#define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
3850
3851static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
3852 union bpf_attr __user *uattr)
3853{
3854 int ufd = attr->info.bpf_fd;
3855 struct fd f;
3856 int err;
3857
3858 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
3859 return -EINVAL;
3860
3861 f = fdget(ufd);
3862 if (!f.file)
3863 return -EBADFD;
3864
3865 if (f.file->f_op == &bpf_prog_fops)
3866 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
3867 uattr);
3868 else if (f.file->f_op == &bpf_map_fops)
3869 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
3870 uattr);
3871 else if (f.file->f_op == &btf_fops)
3872 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
3873 else if (f.file->f_op == &bpf_link_fops)
3874 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
3875 attr, uattr);
3876 else
3877 err = -EINVAL;
3878
3879 fdput(f);
3880 return err;
3881}
3882
3883#define BPF_BTF_LOAD_LAST_FIELD btf_log_level
3884
3885static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr)
3886{
3887 if (CHECK_ATTR(BPF_BTF_LOAD))
3888 return -EINVAL;
3889
3890 if (!bpf_capable())
3891 return -EPERM;
3892
3893 return btf_new_fd(attr, uattr);
3894}
3895
3896#define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
3897
3898static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
3899{
3900 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
3901 return -EINVAL;
3902
3903 if (!capable(CAP_SYS_ADMIN))
3904 return -EPERM;
3905
3906 return btf_get_fd_by_id(attr->btf_id);
3907}
3908
3909static int bpf_task_fd_query_copy(const union bpf_attr *attr,
3910 union bpf_attr __user *uattr,
3911 u32 prog_id, u32 fd_type,
3912 const char *buf, u64 probe_offset,
3913 u64 probe_addr)
3914{
3915 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
3916 u32 len = buf ? strlen(buf) : 0, input_len;
3917 int err = 0;
3918
3919 if (put_user(len, &uattr->task_fd_query.buf_len))
3920 return -EFAULT;
3921 input_len = attr->task_fd_query.buf_len;
3922 if (input_len && ubuf) {
3923 if (!len) {
3924 /* nothing to copy, just make ubuf NULL terminated */
3925 char zero = '\0';
3926
3927 if (put_user(zero, ubuf))
3928 return -EFAULT;
3929 } else if (input_len >= len + 1) {
3930 /* ubuf can hold the string with NULL terminator */
3931 if (copy_to_user(ubuf, buf, len + 1))
3932 return -EFAULT;
3933 } else {
3934 /* ubuf cannot hold the string with NULL terminator,
3935 * do a partial copy with NULL terminator.
3936 */
3937 char zero = '\0';
3938
3939 err = -ENOSPC;
3940 if (copy_to_user(ubuf, buf, input_len - 1))
3941 return -EFAULT;
3942 if (put_user(zero, ubuf + input_len - 1))
3943 return -EFAULT;
3944 }
3945 }
3946
3947 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
3948 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
3949 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
3950 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
3951 return -EFAULT;
3952
3953 return err;
3954}
3955
3956#define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
3957
3958static int bpf_task_fd_query(const union bpf_attr *attr,
3959 union bpf_attr __user *uattr)
3960{
3961 pid_t pid = attr->task_fd_query.pid;
3962 u32 fd = attr->task_fd_query.fd;
3963 const struct perf_event *event;
3964 struct task_struct *task;
3965 struct file *file;
3966 int err;
3967
3968 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
3969 return -EINVAL;
3970
3971 if (!capable(CAP_SYS_ADMIN))
3972 return -EPERM;
3973
3974 if (attr->task_fd_query.flags != 0)
3975 return -EINVAL;
3976
3977 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3978 if (!task)
3979 return -ENOENT;
3980
3981 err = 0;
3982 file = fget_task(task, fd);
3983 put_task_struct(task);
3984 if (!file)
3985 return -EBADF;
3986
3987 if (file->f_op == &bpf_link_fops) {
3988 struct bpf_link *link = file->private_data;
3989
3990 if (link->ops == &bpf_raw_tp_link_lops) {
3991 struct bpf_raw_tp_link *raw_tp =
3992 container_of(link, struct bpf_raw_tp_link, link);
3993 struct bpf_raw_event_map *btp = raw_tp->btp;
3994
3995 err = bpf_task_fd_query_copy(attr, uattr,
3996 raw_tp->link.prog->aux->id,
3997 BPF_FD_TYPE_RAW_TRACEPOINT,
3998 btp->tp->name, 0, 0);
3999 goto put_file;
4000 }
4001 goto out_not_supp;
4002 }
4003
4004 event = perf_get_event(file);
4005 if (!IS_ERR(event)) {
4006 u64 probe_offset, probe_addr;
4007 u32 prog_id, fd_type;
4008 const char *buf;
4009
4010 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
4011 &buf, &probe_offset,
4012 &probe_addr);
4013 if (!err)
4014 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
4015 fd_type, buf,
4016 probe_offset,
4017 probe_addr);
4018 goto put_file;
4019 }
4020
4021out_not_supp:
4022 err = -ENOTSUPP;
4023put_file:
4024 fput(file);
4025 return err;
4026}
4027
4028#define BPF_MAP_BATCH_LAST_FIELD batch.flags
4029
4030#define BPF_DO_BATCH(fn) \
4031 do { \
4032 if (!fn) { \
4033 err = -ENOTSUPP; \
4034 goto err_put; \
4035 } \
4036 err = fn(map, attr, uattr); \
4037 } while (0)
4038
4039static int bpf_map_do_batch(const union bpf_attr *attr,
4040 union bpf_attr __user *uattr,
4041 int cmd)
4042{
4043 struct bpf_map *map;
4044 int err, ufd;
4045 struct fd f;
4046
4047 if (CHECK_ATTR(BPF_MAP_BATCH))
4048 return -EINVAL;
4049
4050 ufd = attr->batch.map_fd;
4051 f = fdget(ufd);
4052 map = __bpf_map_get(f);
4053 if (IS_ERR(map))
4054 return PTR_ERR(map);
4055
4056 if ((cmd == BPF_MAP_LOOKUP_BATCH ||
4057 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH) &&
4058 !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4059 err = -EPERM;
4060 goto err_put;
4061 }
4062
4063 if (cmd != BPF_MAP_LOOKUP_BATCH &&
4064 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4065 err = -EPERM;
4066 goto err_put;
4067 }
4068
4069 if (cmd == BPF_MAP_LOOKUP_BATCH)
4070 BPF_DO_BATCH(map->ops->map_lookup_batch);
4071 else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4072 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
4073 else if (cmd == BPF_MAP_UPDATE_BATCH)
4074 BPF_DO_BATCH(map->ops->map_update_batch);
4075 else
4076 BPF_DO_BATCH(map->ops->map_delete_batch);
4077
4078err_put:
4079 fdput(f);
4080 return err;
4081}
4082
4083static int tracing_bpf_link_attach(const union bpf_attr *attr, bpfptr_t uattr,
4084 struct bpf_prog *prog)
4085{
4086 if (attr->link_create.attach_type != prog->expected_attach_type)
4087 return -EINVAL;
4088
4089 if (prog->expected_attach_type == BPF_TRACE_ITER)
4090 return bpf_iter_link_attach(attr, uattr, prog);
4091 else if (prog->type == BPF_PROG_TYPE_EXT)
4092 return bpf_tracing_prog_attach(prog,
4093 attr->link_create.target_fd,
4094 attr->link_create.target_btf_id);
4095 return -EINVAL;
4096}
4097
4098#define BPF_LINK_CREATE_LAST_FIELD link_create.iter_info_len
4099static int link_create(union bpf_attr *attr, bpfptr_t uattr)
4100{
4101 enum bpf_prog_type ptype;
4102 struct bpf_prog *prog;
4103 int ret;
4104
4105 if (CHECK_ATTR(BPF_LINK_CREATE))
4106 return -EINVAL;
4107
4108 prog = bpf_prog_get(attr->link_create.prog_fd);
4109 if (IS_ERR(prog))
4110 return PTR_ERR(prog);
4111
4112 ret = bpf_prog_attach_check_attach_type(prog,
4113 attr->link_create.attach_type);
4114 if (ret)
4115 goto out;
4116
4117 if (prog->type == BPF_PROG_TYPE_EXT) {
4118 ret = tracing_bpf_link_attach(attr, uattr, prog);
4119 goto out;
4120 }
4121
4122 ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4123 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4124 ret = -EINVAL;
4125 goto out;
4126 }
4127
4128 switch (ptype) {
4129 case BPF_PROG_TYPE_CGROUP_SKB:
4130 case BPF_PROG_TYPE_CGROUP_SOCK:
4131 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4132 case BPF_PROG_TYPE_SOCK_OPS:
4133 case BPF_PROG_TYPE_CGROUP_DEVICE:
4134 case BPF_PROG_TYPE_CGROUP_SYSCTL:
4135 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4136 ret = cgroup_bpf_link_attach(attr, prog);
4137 break;
4138 case BPF_PROG_TYPE_TRACING:
4139 ret = tracing_bpf_link_attach(attr, uattr, prog);
4140 break;
4141 case BPF_PROG_TYPE_FLOW_DISSECTOR:
4142 case BPF_PROG_TYPE_SK_LOOKUP:
4143 ret = netns_bpf_link_create(attr, prog);
4144 break;
4145#ifdef CONFIG_NET
4146 case BPF_PROG_TYPE_XDP:
4147 ret = bpf_xdp_link_attach(attr, prog);
4148 break;
4149#endif
4150 default:
4151 ret = -EINVAL;
4152 }
4153
4154out:
4155 if (ret < 0)
4156 bpf_prog_put(prog);
4157 return ret;
4158}
4159
4160#define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4161
4162static int link_update(union bpf_attr *attr)
4163{
4164 struct bpf_prog *old_prog = NULL, *new_prog;
4165 struct bpf_link *link;
4166 u32 flags;
4167 int ret;
4168
4169 if (CHECK_ATTR(BPF_LINK_UPDATE))
4170 return -EINVAL;
4171
4172 flags = attr->link_update.flags;
4173 if (flags & ~BPF_F_REPLACE)
4174 return -EINVAL;
4175
4176 link = bpf_link_get_from_fd(attr->link_update.link_fd);
4177 if (IS_ERR(link))
4178 return PTR_ERR(link);
4179
4180 new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4181 if (IS_ERR(new_prog)) {
4182 ret = PTR_ERR(new_prog);
4183 goto out_put_link;
4184 }
4185
4186 if (flags & BPF_F_REPLACE) {
4187 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4188 if (IS_ERR(old_prog)) {
4189 ret = PTR_ERR(old_prog);
4190 old_prog = NULL;
4191 goto out_put_progs;
4192 }
4193 } else if (attr->link_update.old_prog_fd) {
4194 ret = -EINVAL;
4195 goto out_put_progs;
4196 }
4197
4198 if (link->ops->update_prog)
4199 ret = link->ops->update_prog(link, new_prog, old_prog);
4200 else
4201 ret = -EINVAL;
4202
4203out_put_progs:
4204 if (old_prog)
4205 bpf_prog_put(old_prog);
4206 if (ret)
4207 bpf_prog_put(new_prog);
4208out_put_link:
4209 bpf_link_put(link);
4210 return ret;
4211}
4212
4213#define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4214
4215static int link_detach(union bpf_attr *attr)
4216{
4217 struct bpf_link *link;
4218 int ret;
4219
4220 if (CHECK_ATTR(BPF_LINK_DETACH))
4221 return -EINVAL;
4222
4223 link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4224 if (IS_ERR(link))
4225 return PTR_ERR(link);
4226
4227 if (link->ops->detach)
4228 ret = link->ops->detach(link);
4229 else
4230 ret = -EOPNOTSUPP;
4231
4232 bpf_link_put(link);
4233 return ret;
4234}
4235
4236static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4237{
4238 return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4239}
4240
4241struct bpf_link *bpf_link_by_id(u32 id)
4242{
4243 struct bpf_link *link;
4244
4245 if (!id)
4246 return ERR_PTR(-ENOENT);
4247
4248 spin_lock_bh(&link_idr_lock);
4249 /* before link is "settled", ID is 0, pretend it doesn't exist yet */
4250 link = idr_find(&link_idr, id);
4251 if (link) {
4252 if (link->id)
4253 link = bpf_link_inc_not_zero(link);
4254 else
4255 link = ERR_PTR(-EAGAIN);
4256 } else {
4257 link = ERR_PTR(-ENOENT);
4258 }
4259 spin_unlock_bh(&link_idr_lock);
4260 return link;
4261}
4262
4263#define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4264
4265static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4266{
4267 struct bpf_link *link;
4268 u32 id = attr->link_id;
4269 int fd;
4270
4271 if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4272 return -EINVAL;
4273
4274 if (!capable(CAP_SYS_ADMIN))
4275 return -EPERM;
4276
4277 link = bpf_link_by_id(id);
4278 if (IS_ERR(link))
4279 return PTR_ERR(link);
4280
4281 fd = bpf_link_new_fd(link);
4282 if (fd < 0)
4283 bpf_link_put(link);
4284
4285 return fd;
4286}
4287
4288DEFINE_MUTEX(bpf_stats_enabled_mutex);
4289
4290static int bpf_stats_release(struct inode *inode, struct file *file)
4291{
4292 mutex_lock(&bpf_stats_enabled_mutex);
4293 static_key_slow_dec(&bpf_stats_enabled_key.key);
4294 mutex_unlock(&bpf_stats_enabled_mutex);
4295 return 0;
4296}
4297
4298static const struct file_operations bpf_stats_fops = {
4299 .release = bpf_stats_release,
4300};
4301
4302static int bpf_enable_runtime_stats(void)
4303{
4304 int fd;
4305
4306 mutex_lock(&bpf_stats_enabled_mutex);
4307
4308 /* Set a very high limit to avoid overflow */
4309 if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4310 mutex_unlock(&bpf_stats_enabled_mutex);
4311 return -EBUSY;
4312 }
4313
4314 fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4315 if (fd >= 0)
4316 static_key_slow_inc(&bpf_stats_enabled_key.key);
4317
4318 mutex_unlock(&bpf_stats_enabled_mutex);
4319 return fd;
4320}
4321
4322#define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4323
4324static int bpf_enable_stats(union bpf_attr *attr)
4325{
4326
4327 if (CHECK_ATTR(BPF_ENABLE_STATS))
4328 return -EINVAL;
4329
4330 if (!capable(CAP_SYS_ADMIN))
4331 return -EPERM;
4332
4333 switch (attr->enable_stats.type) {
4334 case BPF_STATS_RUN_TIME:
4335 return bpf_enable_runtime_stats();
4336 default:
4337 break;
4338 }
4339 return -EINVAL;
4340}
4341
4342#define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4343
4344static int bpf_iter_create(union bpf_attr *attr)
4345{
4346 struct bpf_link *link;
4347 int err;
4348
4349 if (CHECK_ATTR(BPF_ITER_CREATE))
4350 return -EINVAL;
4351
4352 if (attr->iter_create.flags)
4353 return -EINVAL;
4354
4355 link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4356 if (IS_ERR(link))
4357 return PTR_ERR(link);
4358
4359 err = bpf_iter_new_fd(link);
4360 bpf_link_put(link);
4361
4362 return err;
4363}
4364
4365#define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4366
4367static int bpf_prog_bind_map(union bpf_attr *attr)
4368{
4369 struct bpf_prog *prog;
4370 struct bpf_map *map;
4371 struct bpf_map **used_maps_old, **used_maps_new;
4372 int i, ret = 0;
4373
4374 if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4375 return -EINVAL;
4376
4377 if (attr->prog_bind_map.flags)
4378 return -EINVAL;
4379
4380 prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4381 if (IS_ERR(prog))
4382 return PTR_ERR(prog);
4383
4384 map = bpf_map_get(attr->prog_bind_map.map_fd);
4385 if (IS_ERR(map)) {
4386 ret = PTR_ERR(map);
4387 goto out_prog_put;
4388 }
4389
4390 mutex_lock(&prog->aux->used_maps_mutex);
4391
4392 used_maps_old = prog->aux->used_maps;
4393
4394 for (i = 0; i < prog->aux->used_map_cnt; i++)
4395 if (used_maps_old[i] == map) {
4396 bpf_map_put(map);
4397 goto out_unlock;
4398 }
4399
4400 used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4401 sizeof(used_maps_new[0]),
4402 GFP_KERNEL);
4403 if (!used_maps_new) {
4404 ret = -ENOMEM;
4405 goto out_unlock;
4406 }
4407
4408 memcpy(used_maps_new, used_maps_old,
4409 sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4410 used_maps_new[prog->aux->used_map_cnt] = map;
4411
4412 prog->aux->used_map_cnt++;
4413 prog->aux->used_maps = used_maps_new;
4414
4415 kfree(used_maps_old);
4416
4417out_unlock:
4418 mutex_unlock(&prog->aux->used_maps_mutex);
4419
4420 if (ret)
4421 bpf_map_put(map);
4422out_prog_put:
4423 bpf_prog_put(prog);
4424 return ret;
4425}
4426
4427static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
4428{
4429 union bpf_attr attr;
4430 int err;
4431
4432 if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
4433 return -EPERM;
4434
4435 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
4436 if (err)
4437 return err;
4438 size = min_t(u32, size, sizeof(attr));
4439
4440 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
4441 memset(&attr, 0, sizeof(attr));
4442 if (copy_from_bpfptr(&attr, uattr, size) != 0)
4443 return -EFAULT;
4444
4445 err = security_bpf(cmd, &attr, size);
4446 if (err < 0)
4447 return err;
4448
4449 switch (cmd) {
4450 case BPF_MAP_CREATE:
4451 err = map_create(&attr);
4452 break;
4453 case BPF_MAP_LOOKUP_ELEM:
4454 err = map_lookup_elem(&attr);
4455 break;
4456 case BPF_MAP_UPDATE_ELEM:
4457 err = map_update_elem(&attr, uattr);
4458 break;
4459 case BPF_MAP_DELETE_ELEM:
4460 err = map_delete_elem(&attr);
4461 break;
4462 case BPF_MAP_GET_NEXT_KEY:
4463 err = map_get_next_key(&attr);
4464 break;
4465 case BPF_MAP_FREEZE:
4466 err = map_freeze(&attr);
4467 break;
4468 case BPF_PROG_LOAD:
4469 err = bpf_prog_load(&attr, uattr);
4470 break;
4471 case BPF_OBJ_PIN:
4472 err = bpf_obj_pin(&attr);
4473 break;
4474 case BPF_OBJ_GET:
4475 err = bpf_obj_get(&attr);
4476 break;
4477 case BPF_PROG_ATTACH:
4478 err = bpf_prog_attach(&attr);
4479 break;
4480 case BPF_PROG_DETACH:
4481 err = bpf_prog_detach(&attr);
4482 break;
4483 case BPF_PROG_QUERY:
4484 err = bpf_prog_query(&attr, uattr.user);
4485 break;
4486 case BPF_PROG_TEST_RUN:
4487 err = bpf_prog_test_run(&attr, uattr.user);
4488 break;
4489 case BPF_PROG_GET_NEXT_ID:
4490 err = bpf_obj_get_next_id(&attr, uattr.user,
4491 &prog_idr, &prog_idr_lock);
4492 break;
4493 case BPF_MAP_GET_NEXT_ID:
4494 err = bpf_obj_get_next_id(&attr, uattr.user,
4495 &map_idr, &map_idr_lock);
4496 break;
4497 case BPF_BTF_GET_NEXT_ID:
4498 err = bpf_obj_get_next_id(&attr, uattr.user,
4499 &btf_idr, &btf_idr_lock);
4500 break;
4501 case BPF_PROG_GET_FD_BY_ID:
4502 err = bpf_prog_get_fd_by_id(&attr);
4503 break;
4504 case BPF_MAP_GET_FD_BY_ID:
4505 err = bpf_map_get_fd_by_id(&attr);
4506 break;
4507 case BPF_OBJ_GET_INFO_BY_FD:
4508 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
4509 break;
4510 case BPF_RAW_TRACEPOINT_OPEN:
4511 err = bpf_raw_tracepoint_open(&attr);
4512 break;
4513 case BPF_BTF_LOAD:
4514 err = bpf_btf_load(&attr, uattr);
4515 break;
4516 case BPF_BTF_GET_FD_BY_ID:
4517 err = bpf_btf_get_fd_by_id(&attr);
4518 break;
4519 case BPF_TASK_FD_QUERY:
4520 err = bpf_task_fd_query(&attr, uattr.user);
4521 break;
4522 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
4523 err = map_lookup_and_delete_elem(&attr);
4524 break;
4525 case BPF_MAP_LOOKUP_BATCH:
4526 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
4527 break;
4528 case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
4529 err = bpf_map_do_batch(&attr, uattr.user,
4530 BPF_MAP_LOOKUP_AND_DELETE_BATCH);
4531 break;
4532 case BPF_MAP_UPDATE_BATCH:
4533 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
4534 break;
4535 case BPF_MAP_DELETE_BATCH:
4536 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
4537 break;
4538 case BPF_LINK_CREATE:
4539 err = link_create(&attr, uattr);
4540 break;
4541 case BPF_LINK_UPDATE:
4542 err = link_update(&attr);
4543 break;
4544 case BPF_LINK_GET_FD_BY_ID:
4545 err = bpf_link_get_fd_by_id(&attr);
4546 break;
4547 case BPF_LINK_GET_NEXT_ID:
4548 err = bpf_obj_get_next_id(&attr, uattr.user,
4549 &link_idr, &link_idr_lock);
4550 break;
4551 case BPF_ENABLE_STATS:
4552 err = bpf_enable_stats(&attr);
4553 break;
4554 case BPF_ITER_CREATE:
4555 err = bpf_iter_create(&attr);
4556 break;
4557 case BPF_LINK_DETACH:
4558 err = link_detach(&attr);
4559 break;
4560 case BPF_PROG_BIND_MAP:
4561 err = bpf_prog_bind_map(&attr);
4562 break;
4563 default:
4564 err = -EINVAL;
4565 break;
4566 }
4567
4568 return err;
4569}
4570
4571SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
4572{
4573 return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
4574}
4575
4576static bool syscall_prog_is_valid_access(int off, int size,
4577 enum bpf_access_type type,
4578 const struct bpf_prog *prog,
4579 struct bpf_insn_access_aux *info)
4580{
4581 if (off < 0 || off >= U16_MAX)
4582 return false;
4583 if (off % size != 0)
4584 return false;
4585 return true;
4586}
4587
4588BPF_CALL_3(bpf_sys_bpf, int, cmd, void *, attr, u32, attr_size)
4589{
4590 switch (cmd) {
4591 case BPF_MAP_CREATE:
4592 case BPF_MAP_UPDATE_ELEM:
4593 case BPF_MAP_FREEZE:
4594 case BPF_PROG_LOAD:
4595 case BPF_BTF_LOAD:
4596 break;
4597 /* case BPF_PROG_TEST_RUN:
4598 * is not part of this list to prevent recursive test_run
4599 */
4600 default:
4601 return -EINVAL;
4602 }
4603 return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
4604}
4605
4606static const struct bpf_func_proto bpf_sys_bpf_proto = {
4607 .func = bpf_sys_bpf,
4608 .gpl_only = false,
4609 .ret_type = RET_INTEGER,
4610 .arg1_type = ARG_ANYTHING,
4611 .arg2_type = ARG_PTR_TO_MEM,
4612 .arg3_type = ARG_CONST_SIZE,
4613};
4614
4615const struct bpf_func_proto * __weak
4616tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4617{
4618 return bpf_base_func_proto(func_id);
4619}
4620
4621BPF_CALL_1(bpf_sys_close, u32, fd)
4622{
4623 /* When bpf program calls this helper there should not be
4624 * an fdget() without matching completed fdput().
4625 * This helper is allowed in the following callchain only:
4626 * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
4627 */
4628 return close_fd(fd);
4629}
4630
4631static const struct bpf_func_proto bpf_sys_close_proto = {
4632 .func = bpf_sys_close,
4633 .gpl_only = false,
4634 .ret_type = RET_INTEGER,
4635 .arg1_type = ARG_ANYTHING,
4636};
4637
4638static const struct bpf_func_proto *
4639syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4640{
4641 switch (func_id) {
4642 case BPF_FUNC_sys_bpf:
4643 return &bpf_sys_bpf_proto;
4644 case BPF_FUNC_btf_find_by_name_kind:
4645 return &bpf_btf_find_by_name_kind_proto;
4646 case BPF_FUNC_sys_close:
4647 return &bpf_sys_close_proto;
4648 default:
4649 return tracing_prog_func_proto(func_id, prog);
4650 }
4651}
4652
4653const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
4654 .get_func_proto = syscall_prog_func_proto,
4655 .is_valid_access = syscall_prog_is_valid_access,
4656};
4657
4658const struct bpf_prog_ops bpf_syscall_prog_ops = {
4659 .test_run = bpf_prog_test_run_syscall,
4660};