projects/linux·blob v7.2 · refs

6790 lines · 170.8 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 */
4#include <crypto/sha2.h>
5#include <linux/bpf.h>
6#include <linux/bpf-cgroup.h>
7#include <linux/bpf_trace.h>
8#include <linux/bpf_lirc.h>
9#include <linux/bpf_verifier.h>
10#include <linux/bsearch.h>
11#include <linux/btf.h>
12#include <linux/hex.h>
13#include <linux/syscalls.h>
14#include <linux/slab.h>
15#include <linux/sched/signal.h>
16#include <linux/vmalloc.h>
17#include <linux/mmzone.h>
18#include <linux/anon_inodes.h>
19#include <linux/fdtable.h>
20#include <linux/file.h>
21#include <linux/fs.h>
22#include <linux/license.h>
23#include <linux/filter.h>
24#include <linux/kernel.h>
25#include <linux/idr.h>
26#include <linux/cred.h>
27#include <linux/timekeeping.h>
28#include <linux/ctype.h>
29#include <linux/nospec.h>
30#include <linux/audit.h>
31#include <uapi/linux/btf.h>
32#include <linux/pgtable.h>
33#include <linux/bpf_lsm.h>
34#include <linux/poll.h>
35#include <linux/sort.h>
36#include <linux/bpf-netns.h>
37#include <linux/rcupdate_trace.h>
38#include <linux/memcontrol.h>
39#include <linux/trace_events.h>
40#include <linux/tracepoint.h>
41#include <linux/overflow.h>
42#include <linux/cookie.h>
43#include <linux/verification.h>
44#include <linux/btf_ids.h>
45
46#include <net/netfilter/nf_bpf_link.h>
47#include <net/netkit.h>
48#include <net/tcx.h>
49
50#define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
51 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
52 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
53#define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
54#define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
55#define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
56 IS_FD_HASH(map))
57
58#define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
59
60DEFINE_PER_CPU(int, bpf_prog_active);
61DEFINE_COOKIE(bpf_map_cookie);
62static DEFINE_IDR(prog_idr);
63static DEFINE_SPINLOCK(prog_idr_lock);
64static DEFINE_IDR(map_idr);
65static DEFINE_SPINLOCK(map_idr_lock);
66static DEFINE_IDR(link_idr);
67static DEFINE_SPINLOCK(link_idr_lock);
68
69int sysctl_unprivileged_bpf_disabled __read_mostly =
70 IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
71
72static const struct bpf_map_ops * const bpf_map_types[] = {
73#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
74#define BPF_MAP_TYPE(_id, _ops) \
75 [_id] = &_ops,
76#define BPF_LINK_TYPE(_id, _name)
77#include <linux/bpf_types.h>
78#undef BPF_PROG_TYPE
79#undef BPF_MAP_TYPE
80#undef BPF_LINK_TYPE
81};
82
83/*
84 * If we're handed a bigger struct than we know of, ensure all the unknown bits
85 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
86 * we don't know about yet.
87 *
88 * There is a ToCToU between this function call and the following
89 * copy_from_user() call. However, this is not a concern since this function is
90 * meant to be a future-proofing of bits.
91 */
92int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
93 size_t expected_size,
94 size_t actual_size)
95{
96 int res;
97
98 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
99 return -E2BIG;
100
101 if (actual_size <= expected_size)
102 return 0;
103
104 if (uaddr.is_kernel)
105 res = memchr_inv(uaddr.kernel + expected_size, 0,
106 actual_size - expected_size) == NULL;
107 else
108 res = check_zeroed_user(uaddr.user + expected_size,
109 actual_size - expected_size);
110 if (res < 0)
111 return res;
112 return res ? 0 : -E2BIG;
113}
114
115const struct bpf_map_ops bpf_map_offload_ops = {
116 .map_meta_equal = bpf_map_meta_equal,
117 .map_alloc = bpf_map_offload_map_alloc,
118 .map_free = bpf_map_offload_map_free,
119 .map_check_btf = map_check_no_btf,
120 .map_mem_usage = bpf_map_offload_map_mem_usage,
121};
122
123static void bpf_map_write_active_inc(struct bpf_map *map)
124{
125 atomic64_inc(&map->writecnt);
126}
127
128static void bpf_map_write_active_dec(struct bpf_map *map)
129{
130 atomic64_dec(&map->writecnt);
131}
132
133bool bpf_map_write_active(const struct bpf_map *map)
134{
135 return atomic64_read(&map->writecnt) != 0;
136}
137
138static u32 bpf_map_value_size(const struct bpf_map *map, u64 flags)
139{
140 if (flags & (BPF_F_CPU | BPF_F_ALL_CPUS))
141 return map->value_size;
142 else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
143 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
144 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
145 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
146 return round_up(map->value_size, 8) * num_possible_cpus();
147 else if (IS_FD_MAP(map))
148 return sizeof(u32);
149 else
150 return map->value_size;
151}
152
153static void maybe_wait_bpf_programs(struct bpf_map *map)
154{
155 /* Wait for any running non-sleepable BPF programs to complete so that
156 * userspace, when we return to it, knows that all non-sleepable
157 * programs that could be running use the new map value. For sleepable
158 * BPF programs, synchronize_rcu_tasks_trace() should be used to wait
159 * for the completions of these programs, but considering the waiting
160 * time can be very long and userspace may think it will hang forever,
161 * so don't handle sleepable BPF programs now.
162 */
163 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
164 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
165 synchronize_rcu_expedited();
166}
167
168static void unpin_uptr_kaddr(void *kaddr)
169{
170 if (kaddr)
171 unpin_user_page(virt_to_page(kaddr));
172}
173
174static void __bpf_obj_unpin_uptrs(struct btf_record *rec, u32 cnt, void *obj)
175{
176 const struct btf_field *field;
177 void **uptr_addr;
178 int i;
179
180 for (i = 0, field = rec->fields; i < cnt; i++, field++) {
181 if (field->type != BPF_UPTR)
182 continue;
183
184 uptr_addr = obj + field->offset;
185 unpin_uptr_kaddr(*uptr_addr);
186 }
187}
188
189static void bpf_obj_unpin_uptrs(struct btf_record *rec, void *obj)
190{
191 if (!btf_record_has_field(rec, BPF_UPTR))
192 return;
193
194 __bpf_obj_unpin_uptrs(rec, rec->cnt, obj);
195}
196
197static int bpf_obj_pin_uptrs(struct btf_record *rec, void *obj)
198{
199 const struct btf_field *field;
200 const struct btf_type *t;
201 unsigned long start, end;
202 struct page *page;
203 void **uptr_addr;
204 int i, err;
205
206 if (!btf_record_has_field(rec, BPF_UPTR))
207 return 0;
208
209 for (i = 0, field = rec->fields; i < rec->cnt; i++, field++) {
210 if (field->type != BPF_UPTR)
211 continue;
212
213 uptr_addr = obj + field->offset;
214 start = *(unsigned long *)uptr_addr;
215 if (!start)
216 continue;
217
218 t = btf_type_by_id(field->kptr.btf, field->kptr.btf_id);
219 /* t->size was checked for zero before */
220 if (check_add_overflow(start, t->size - 1, &end)) {
221 err = -EFAULT;
222 goto unpin_all;
223 }
224
225 /* The uptr's struct cannot span across two pages */
226 if ((start & PAGE_MASK) != (end & PAGE_MASK)) {
227 err = -EOPNOTSUPP;
228 goto unpin_all;
229 }
230
231 err = pin_user_pages_fast(start, 1, FOLL_LONGTERM | FOLL_WRITE, &page);
232 if (err != 1)
233 goto unpin_all;
234
235 if (PageHighMem(page)) {
236 err = -EOPNOTSUPP;
237 unpin_user_page(page);
238 goto unpin_all;
239 }
240
241 *uptr_addr = page_address(page) + offset_in_page(start);
242 }
243
244 return 0;
245
246unpin_all:
247 __bpf_obj_unpin_uptrs(rec, i, obj);
248 return err;
249}
250
251static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
252 void *key, void *value, __u64 flags)
253{
254 int err;
255
256 /* Need to create a kthread, thus must support schedule */
257 if (bpf_map_is_offloaded(map)) {
258 return bpf_map_offload_update_elem(map, key, value, flags);
259 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
260 map->map_type == BPF_MAP_TYPE_ARENA ||
261 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
262 return map->ops->map_update_elem(map, key, value, flags);
263 } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
264 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
265 return sock_map_update_elem_sys(map, key, value, flags);
266 } else if (IS_FD_PROG_ARRAY(map)) {
267 return bpf_fd_array_map_update_elem(map, map_file, key, value,
268 flags);
269 }
270
271 bpf_disable_instrumentation();
272 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
273 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
274 err = bpf_percpu_hash_update(map, key, value, flags);
275 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
276 err = bpf_percpu_array_update(map, key, value, flags);
277 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
278 err = bpf_percpu_cgroup_storage_update(map, key, value,
279 flags);
280 } else if (IS_FD_ARRAY(map)) {
281 err = bpf_fd_array_map_update_elem(map, map_file, key, value,
282 flags);
283 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
284 err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
285 flags);
286 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
287 /* rcu_read_lock() is not needed */
288 err = bpf_fd_reuseport_array_update_elem(map, key, value,
289 flags);
290 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
291 map->map_type == BPF_MAP_TYPE_STACK ||
292 map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
293 err = map->ops->map_push_elem(map, value, flags);
294 } else {
295 err = bpf_obj_pin_uptrs(map->record, value);
296 if (!err) {
297 rcu_read_lock();
298 err = map->ops->map_update_elem(map, key, value, flags);
299 rcu_read_unlock();
300 if (err)
301 bpf_obj_unpin_uptrs(map->record, value);
302 }
303 }
304 bpf_enable_instrumentation();
305
306 return err;
307}
308
309static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
310 __u64 flags)
311{
312 void *ptr;
313 int err;
314
315 if (bpf_map_is_offloaded(map))
316 return bpf_map_offload_lookup_elem(map, key, value);
317
318 bpf_disable_instrumentation();
319 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
320 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
321 err = bpf_percpu_hash_copy(map, key, value, flags);
322 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
323 err = bpf_percpu_array_copy(map, key, value, flags);
324 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
325 err = bpf_percpu_cgroup_storage_copy(map, key, value, flags);
326 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
327 err = bpf_stackmap_extract(map, key, value, false);
328 } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
329 err = bpf_fd_array_map_lookup_elem(map, key, value);
330 } else if (IS_FD_HASH(map)) {
331 err = bpf_fd_htab_map_lookup_elem(map, key, value);
332 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
333 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
334 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
335 map->map_type == BPF_MAP_TYPE_STACK ||
336 map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
337 err = map->ops->map_peek_elem(map, value);
338 } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
339 /* struct_ops map requires directly updating "value" */
340 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
341 } else {
342 rcu_read_lock();
343 if (map->ops->map_lookup_elem_sys_only)
344 ptr = map->ops->map_lookup_elem_sys_only(map, key);
345 else
346 ptr = map->ops->map_lookup_elem(map, key);
347 if (IS_ERR(ptr)) {
348 err = PTR_ERR(ptr);
349 } else if (!ptr) {
350 err = -ENOENT;
351 } else {
352 err = 0;
353 if (flags & BPF_F_LOCK)
354 /* lock 'ptr' and copy everything but lock */
355 copy_map_value_locked(map, value, ptr, true);
356 else
357 copy_map_value(map, value, ptr);
358 /* mask lock and timer, since value wasn't zero inited */
359 check_and_init_map_value(map, value);
360 }
361 rcu_read_unlock();
362 }
363
364 bpf_enable_instrumentation();
365
366 return err;
367}
368
369/* Please, do not use this function outside from the map creation path
370 * (e.g. in map update path) without taking care of setting the active
371 * memory cgroup (see at bpf_map_kmalloc_node() for example).
372 */
373static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
374{
375 /* We really just want to fail instead of triggering OOM killer
376 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
377 * which is used for lower order allocation requests.
378 *
379 * It has been observed that higher order allocation requests done by
380 * vmalloc with __GFP_NORETRY being set might fail due to not trying
381 * to reclaim memory from the page cache, thus we set
382 * __GFP_RETRY_MAYFAIL to avoid such situations.
383 */
384
385 gfp_t gfp = bpf_memcg_flags(__GFP_NOWARN | __GFP_ZERO);
386 unsigned int flags = 0;
387 unsigned long align = 1;
388 void *area;
389
390 if (size >= SIZE_MAX)
391 return NULL;
392
393 /* kmalloc()'ed memory can't be mmap()'ed */
394 if (mmapable) {
395 BUG_ON(!PAGE_ALIGNED(size));
396 align = SHMLBA;
397 flags = VM_USERMAP;
398 } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
399 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
400 numa_node);
401 if (area != NULL)
402 return area;
403 }
404
405 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
406 gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
407 flags, numa_node, __builtin_return_address(0));
408}
409
410void *bpf_map_area_alloc(u64 size, int numa_node)
411{
412 return __bpf_map_area_alloc(size, numa_node, false);
413}
414
415void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
416{
417 return __bpf_map_area_alloc(size, numa_node, true);
418}
419
420void bpf_map_area_free(void *area)
421{
422 kvfree(area);
423}
424
425static u32 bpf_map_flags_retain_permanent(u32 flags)
426{
427 /* Some map creation flags are not tied to the map object but
428 * rather to the map fd instead, so they have no meaning upon
429 * map object inspection since multiple file descriptors with
430 * different (access) properties can exist here. Thus, given
431 * this has zero meaning for the map itself, lets clear these
432 * from here.
433 */
434 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
435}
436
437void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
438{
439 map->map_type = attr->map_type;
440 map->key_size = attr->key_size;
441 map->value_size = attr->value_size;
442 map->max_entries = attr->max_entries;
443 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
444 map->numa_node = bpf_map_attr_numa_node(attr);
445 map->map_extra = attr->map_extra;
446}
447
448static int bpf_map_alloc_id(struct bpf_map *map)
449{
450 int id;
451
452 idr_preload(GFP_KERNEL);
453 spin_lock_bh(&map_idr_lock);
454 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
455 if (id > 0)
456 map->id = id;
457 spin_unlock_bh(&map_idr_lock);
458 idr_preload_end();
459
460 if (WARN_ON_ONCE(!id))
461 return -ENOSPC;
462
463 return id > 0 ? 0 : id;
464}
465
466void bpf_map_free_id(struct bpf_map *map)
467{
468 unsigned long flags;
469
470 /* Offloaded maps are removed from the IDR store when their device
471 * disappears - even if someone holds an fd to them they are unusable,
472 * the memory is gone, all ops will fail; they are simply waiting for
473 * refcnt to drop to be freed.
474 */
475 if (!map->id)
476 return;
477
478 spin_lock_irqsave(&map_idr_lock, flags);
479
480 idr_remove(&map_idr, map->id);
481 map->id = 0;
482
483 spin_unlock_irqrestore(&map_idr_lock, flags);
484}
485
486#ifdef CONFIG_MEMCG
487static void bpf_map_save_memcg(struct bpf_map *map)
488{
489 /* Currently if a map is created by a process belonging to the root
490 * memory cgroup, get_obj_cgroup_from_current() will return NULL.
491 * So we have to check map->objcg for being NULL each time it's
492 * being used.
493 */
494 if (memcg_bpf_enabled())
495 map->objcg = get_obj_cgroup_from_current();
496}
497
498static void bpf_map_release_memcg(struct bpf_map *map)
499{
500 if (map->objcg)
501 obj_cgroup_put(map->objcg);
502}
503
504static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
505{
506 if (map->objcg)
507 return get_mem_cgroup_from_objcg(map->objcg);
508
509 return root_mem_cgroup;
510}
511
512void bpf_map_memcg_enter(const struct bpf_map *map, struct mem_cgroup **old_memcg,
513 struct mem_cgroup **new_memcg)
514{
515 *new_memcg = bpf_map_get_memcg(map);
516 *old_memcg = set_active_memcg(*new_memcg);
517}
518
519void bpf_map_memcg_exit(struct mem_cgroup *old_memcg,
520 struct mem_cgroup *new_memcg)
521{
522 set_active_memcg(old_memcg);
523 mem_cgroup_put(new_memcg);
524}
525
526void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
527 int node)
528{
529 struct mem_cgroup *memcg, *old_memcg;
530 void *ptr;
531
532 bpf_map_memcg_enter(map, &old_memcg, &memcg);
533 ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
534 bpf_map_memcg_exit(old_memcg, memcg);
535
536 return ptr;
537}
538
539void *bpf_map_kmalloc_nolock(const struct bpf_map *map, size_t size, gfp_t flags,
540 int node)
541{
542 struct mem_cgroup *memcg, *old_memcg;
543 void *ptr;
544
545 bpf_map_memcg_enter(map, &old_memcg, &memcg);
546 ptr = kmalloc_nolock(size, flags | __GFP_ACCOUNT, node);
547 bpf_map_memcg_exit(old_memcg, memcg);
548
549 return ptr;
550}
551
552void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
553{
554 struct mem_cgroup *memcg, *old_memcg;
555 void *ptr;
556
557 bpf_map_memcg_enter(map, &old_memcg, &memcg);
558 ptr = kzalloc(size, flags | __GFP_ACCOUNT);
559 bpf_map_memcg_exit(old_memcg, memcg);
560
561 return ptr;
562}
563
564void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
565 gfp_t flags)
566{
567 struct mem_cgroup *memcg, *old_memcg;
568 void *ptr;
569
570 bpf_map_memcg_enter(map, &old_memcg, &memcg);
571 ptr = kvcalloc(n, size, flags | __GFP_ACCOUNT);
572 bpf_map_memcg_exit(old_memcg, memcg);
573
574 return ptr;
575}
576
577void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
578 size_t align, gfp_t flags)
579{
580 struct mem_cgroup *memcg, *old_memcg;
581 void __percpu *ptr;
582
583 bpf_map_memcg_enter(map, &old_memcg, &memcg);
584 ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
585 bpf_map_memcg_exit(old_memcg, memcg);
586
587 return ptr;
588}
589
590#else
591static void bpf_map_save_memcg(struct bpf_map *map)
592{
593}
594
595static void bpf_map_release_memcg(struct bpf_map *map)
596{
597}
598#endif
599
600static bool can_alloc_pages(void)
601{
602 return preempt_count() == 0 && !irqs_disabled() &&
603 !IS_ENABLED(CONFIG_PREEMPT_RT);
604}
605
606static struct page *__bpf_alloc_page(int nid)
607{
608 if (!can_alloc_pages())
609 return alloc_pages_nolock(__GFP_ACCOUNT, nid, 0);
610
611 return alloc_pages_node(nid,
612 GFP_KERNEL | __GFP_ZERO | __GFP_ACCOUNT
613 | __GFP_NOWARN,
614 0);
615}
616
617int bpf_map_alloc_pages(const struct bpf_map *map, int nid,
618 unsigned long nr_pages, struct page **pages)
619{
620 unsigned long i, j;
621 struct page *pg;
622 int ret = 0;
623
624 for (i = 0; i < nr_pages; i++) {
625 pg = __bpf_alloc_page(nid);
626
627 if (pg) {
628 pages[i] = pg;
629 continue;
630 }
631 for (j = 0; j < i; j++)
632 free_pages_nolock(pages[j], 0);
633 ret = -ENOMEM;
634 break;
635 }
636
637 return ret;
638}
639
640
641static int btf_field_cmp(const void *a, const void *b)
642{
643 const struct btf_field *f1 = a, *f2 = b;
644
645 if (f1->offset < f2->offset)
646 return -1;
647 else if (f1->offset > f2->offset)
648 return 1;
649 return 0;
650}
651
652struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
653 u32 field_mask)
654{
655 struct btf_field *field;
656
657 if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & field_mask))
658 return NULL;
659 field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
660 if (!field || !(field->type & field_mask))
661 return NULL;
662 return field;
663}
664
665void btf_record_free(struct btf_record *rec)
666{
667 int i;
668
669 if (IS_ERR_OR_NULL(rec))
670 return;
671 for (i = 0; i < rec->cnt; i++) {
672 switch (rec->fields[i].type) {
673 case BPF_KPTR_UNREF:
674 case BPF_KPTR_REF:
675 case BPF_KPTR_PERCPU:
676 case BPF_UPTR:
677 if (rec->fields[i].kptr.module)
678 module_put(rec->fields[i].kptr.module);
679 if (btf_is_kernel(rec->fields[i].kptr.btf))
680 btf_put(rec->fields[i].kptr.btf);
681 break;
682 case BPF_LIST_HEAD:
683 case BPF_LIST_NODE:
684 case BPF_RB_ROOT:
685 case BPF_RB_NODE:
686 case BPF_SPIN_LOCK:
687 case BPF_RES_SPIN_LOCK:
688 case BPF_TIMER:
689 case BPF_REFCOUNT:
690 case BPF_WORKQUEUE:
691 case BPF_TASK_WORK:
692 /* Nothing to release */
693 break;
694 default:
695 WARN_ON_ONCE(1);
696 continue;
697 }
698 }
699 kfree(rec);
700}
701
702void bpf_map_free_record(struct bpf_map *map)
703{
704 btf_record_free(map->record);
705 map->record = NULL;
706}
707
708struct btf_record *btf_record_dup(const struct btf_record *rec)
709{
710 const struct btf_field *fields;
711 struct btf_record *new_rec;
712 int ret, size, i;
713
714 if (IS_ERR_OR_NULL(rec))
715 return NULL;
716 size = struct_size(rec, fields, rec->cnt);
717 new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
718 if (!new_rec)
719 return ERR_PTR(-ENOMEM);
720 /* Do a deep copy of the btf_record */
721 fields = rec->fields;
722 new_rec->cnt = 0;
723 for (i = 0; i < rec->cnt; i++) {
724 switch (fields[i].type) {
725 case BPF_KPTR_UNREF:
726 case BPF_KPTR_REF:
727 case BPF_KPTR_PERCPU:
728 case BPF_UPTR:
729 if (btf_is_kernel(fields[i].kptr.btf))
730 btf_get(fields[i].kptr.btf);
731 if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
732 ret = -ENXIO;
733 goto free;
734 }
735 break;
736 case BPF_LIST_HEAD:
737 case BPF_LIST_NODE:
738 case BPF_RB_ROOT:
739 case BPF_RB_NODE:
740 case BPF_SPIN_LOCK:
741 case BPF_RES_SPIN_LOCK:
742 case BPF_TIMER:
743 case BPF_REFCOUNT:
744 case BPF_WORKQUEUE:
745 case BPF_TASK_WORK:
746 /* Nothing to acquire */
747 break;
748 default:
749 ret = -EFAULT;
750 WARN_ON_ONCE(1);
751 goto free;
752 }
753 new_rec->cnt++;
754 }
755 return new_rec;
756free:
757 btf_record_free(new_rec);
758 return ERR_PTR(ret);
759}
760
761bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
762{
763 bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
764 int size;
765
766 if (!a_has_fields && !b_has_fields)
767 return true;
768 if (a_has_fields != b_has_fields)
769 return false;
770 if (rec_a->cnt != rec_b->cnt)
771 return false;
772 size = struct_size(rec_a, fields, rec_a->cnt);
773 /* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
774 * members are zeroed out. So memcmp is safe to do without worrying
775 * about padding/unused fields.
776 *
777 * While spin_lock, timer, and kptr have no relation to map BTF,
778 * list_head metadata is specific to map BTF, the btf and value_rec
779 * members in particular. btf is the map BTF, while value_rec points to
780 * btf_record in that map BTF.
781 *
782 * So while by default, we don't rely on the map BTF (which the records
783 * were parsed from) matching for both records, which is not backwards
784 * compatible, in case list_head is part of it, we implicitly rely on
785 * that by way of depending on memcmp succeeding for it.
786 */
787 return !memcmp(rec_a, rec_b, size);
788}
789
790void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
791{
792 if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
793 return;
794 bpf_timer_cancel_and_free(obj + rec->timer_off);
795}
796
797void bpf_obj_free_workqueue(const struct btf_record *rec, void *obj)
798{
799 if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_WORKQUEUE)))
800 return;
801 bpf_wq_cancel_and_free(obj + rec->wq_off);
802}
803
804void bpf_obj_free_task_work(const struct btf_record *rec, void *obj)
805{
806 if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TASK_WORK)))
807 return;
808 bpf_task_work_cancel_and_free(obj + rec->task_work_off);
809}
810
811void bpf_obj_cancel_fields(struct bpf_map *map, void *obj)
812{
813 bpf_map_free_internal_structs(map, obj);
814}
815
816void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
817{
818 const struct btf_field *fields;
819 int i;
820
821 if (IS_ERR_OR_NULL(rec))
822 return;
823 fields = rec->fields;
824 for (i = 0; i < rec->cnt; i++) {
825 struct btf_struct_meta *pointee_struct_meta;
826 const struct btf_field *field = &fields[i];
827 void *field_ptr = obj + field->offset;
828 void *xchgd_field;
829
830 switch (fields[i].type) {
831 case BPF_SPIN_LOCK:
832 case BPF_RES_SPIN_LOCK:
833 break;
834 case BPF_TIMER:
835 bpf_timer_cancel_and_free(field_ptr);
836 break;
837 case BPF_WORKQUEUE:
838 bpf_wq_cancel_and_free(field_ptr);
839 break;
840 case BPF_TASK_WORK:
841 bpf_task_work_cancel_and_free(field_ptr);
842 break;
843 case BPF_KPTR_UNREF:
844 WRITE_ONCE(*(u64 *)field_ptr, 0);
845 break;
846 case BPF_KPTR_REF:
847 case BPF_KPTR_PERCPU:
848 xchgd_field = (void *)xchg((unsigned long *)field_ptr, 0);
849 if (!xchgd_field)
850 break;
851
852 if (!btf_is_kernel(field->kptr.btf)) {
853 pointee_struct_meta = btf_find_struct_meta(field->kptr.btf,
854 field->kptr.btf_id);
855 __bpf_obj_drop_impl(xchgd_field, pointee_struct_meta ?
856 pointee_struct_meta->record : NULL,
857 fields[i].type == BPF_KPTR_PERCPU);
858 } else {
859 field->kptr.dtor(xchgd_field);
860 }
861 break;
862 case BPF_UPTR:
863 /* The caller ensured that no one is using the uptr */
864 unpin_uptr_kaddr(*(void **)field_ptr);
865 break;
866 case BPF_LIST_HEAD:
867 if (WARN_ON_ONCE(rec->spin_lock_off < 0))
868 continue;
869 bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
870 break;
871 case BPF_RB_ROOT:
872 if (WARN_ON_ONCE(rec->spin_lock_off < 0))
873 continue;
874 bpf_rb_root_free(field, field_ptr, obj + rec->spin_lock_off);
875 break;
876 case BPF_LIST_NODE:
877 case BPF_RB_NODE:
878 case BPF_REFCOUNT:
879 break;
880 default:
881 WARN_ON_ONCE(1);
882 continue;
883 }
884 }
885}
886
887static void bpf_map_free(struct bpf_map *map)
888{
889 struct btf_record *rec = map->record;
890 struct btf *btf = map->btf;
891
892 /* implementation dependent freeing. Disabling migration to simplify
893 * the free of values or special fields allocated from bpf memory
894 * allocator.
895 */
896 kfree(map->excl_prog_sha);
897 migrate_disable();
898 map->ops->map_free(map);
899 migrate_enable();
900
901 /* Delay freeing of btf_record for maps, as map_free
902 * callback usually needs access to them. It is better to do it here
903 * than require each callback to do the free itself manually.
904 *
905 * Note that the btf_record stashed in map->inner_map_meta->record was
906 * already freed using the map_free callback for map in map case which
907 * eventually calls bpf_map_free_meta, since inner_map_meta is only a
908 * template bpf_map struct used during verification.
909 */
910 btf_record_free(rec);
911 /* Delay freeing of btf for maps, as map_free callback may need
912 * struct_meta info which will be freed with btf_put().
913 */
914 btf_put(btf);
915}
916
917/* called from workqueue */
918static void bpf_map_free_deferred(struct work_struct *work)
919{
920 struct bpf_map *map = container_of(work, struct bpf_map, work);
921
922 security_bpf_map_free(map);
923 bpf_map_release_memcg(map);
924 bpf_map_owner_free(map);
925 bpf_map_free(map);
926}
927
928static void bpf_map_put_uref(struct bpf_map *map)
929{
930 if (atomic64_dec_and_test(&map->usercnt)) {
931 if (map->ops->map_release_uref)
932 map->ops->map_release_uref(map);
933 }
934}
935
936static void bpf_map_free_in_work(struct bpf_map *map)
937{
938 INIT_WORK(&map->work, bpf_map_free_deferred);
939 /* Avoid spawning kworkers, since they all might contend
940 * for the same mutex like slab_mutex.
941 */
942 queue_work(system_dfl_wq, &map->work);
943}
944
945static void bpf_map_free_rcu_gp(struct rcu_head *rcu)
946{
947 bpf_map_free_in_work(container_of(rcu, struct bpf_map, rcu));
948}
949
950/* decrement map refcnt and schedule it for freeing via workqueue
951 * (underlying map implementation ops->map_free() might sleep)
952 */
953void bpf_map_put(struct bpf_map *map)
954{
955 if (atomic64_dec_and_test(&map->refcnt)) {
956 /* bpf_map_free_id() must be called first */
957 bpf_map_free_id(map);
958
959 WARN_ON_ONCE(atomic64_read(&map->sleepable_refcnt));
960 /* RCU tasks trace grace period implies RCU grace period. */
961 if (READ_ONCE(map->free_after_mult_rcu_gp))
962 call_rcu_tasks_trace(&map->rcu, bpf_map_free_rcu_gp);
963 else if (READ_ONCE(map->free_after_rcu_gp))
964 call_rcu(&map->rcu, bpf_map_free_rcu_gp);
965 else
966 bpf_map_free_in_work(map);
967 }
968}
969EXPORT_SYMBOL_GPL(bpf_map_put);
970
971void bpf_map_put_with_uref(struct bpf_map *map)
972{
973 bpf_map_put_uref(map);
974 bpf_map_put(map);
975}
976
977static int bpf_map_release(struct inode *inode, struct file *filp)
978{
979 struct bpf_map *map = filp->private_data;
980
981 if (map->ops->map_release)
982 map->ops->map_release(map, filp);
983
984 bpf_map_put_with_uref(map);
985 return 0;
986}
987
988static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
989{
990 fmode_t mode = fd_file(f)->f_mode;
991
992 /* Our file permissions may have been overridden by global
993 * map permissions facing syscall side.
994 */
995 if (READ_ONCE(map->frozen))
996 mode &= ~FMODE_CAN_WRITE;
997 return mode;
998}
999
1000#ifdef CONFIG_PROC_FS
1001/* Show the memory usage of a bpf map */
1002static u64 bpf_map_memory_usage(const struct bpf_map *map)
1003{
1004 return map->ops->map_mem_usage(map);
1005}
1006
1007static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
1008{
1009 struct bpf_map *map = filp->private_data;
1010 u32 type = 0, jited = 0;
1011
1012 spin_lock(&map->owner_lock);
1013 if (map->owner) {
1014 type = map->owner->type;
1015 jited = map->owner->jited;
1016 }
1017 spin_unlock(&map->owner_lock);
1018
1019 seq_printf(m,
1020 "map_type:\t%u\n"
1021 "key_size:\t%u\n"
1022 "value_size:\t%u\n"
1023 "max_entries:\t%u\n"
1024 "map_flags:\t%#x\n"
1025 "map_extra:\t%#llx\n"
1026 "memlock:\t%llu\n"
1027 "map_id:\t%u\n"
1028 "frozen:\t%u\n",
1029 map->map_type,
1030 map->key_size,
1031 map->value_size,
1032 map->max_entries,
1033 map->map_flags,
1034 (unsigned long long)map->map_extra,
1035 bpf_map_memory_usage(map),
1036 map->id,
1037 READ_ONCE(map->frozen));
1038 if (type) {
1039 seq_printf(m, "owner_prog_type:\t%u\n", type);
1040 seq_printf(m, "owner_jited:\t%u\n", jited);
1041 }
1042}
1043#endif
1044
1045static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
1046 loff_t *ppos)
1047{
1048 /* We need this handler such that alloc_file() enables
1049 * f_mode with FMODE_CAN_READ.
1050 */
1051 return -EINVAL;
1052}
1053
1054static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
1055 size_t siz, loff_t *ppos)
1056{
1057 /* We need this handler such that alloc_file() enables
1058 * f_mode with FMODE_CAN_WRITE.
1059 */
1060 return -EINVAL;
1061}
1062
1063/* called for any extra memory-mapped regions (except initial) */
1064static void bpf_map_mmap_open(struct vm_area_struct *vma)
1065{
1066 struct bpf_map *map = vma->vm_file->private_data;
1067
1068 if (vma->vm_flags & VM_MAYWRITE)
1069 bpf_map_write_active_inc(map);
1070}
1071
1072/* called for all unmapped memory region (including initial) */
1073static void bpf_map_mmap_close(struct vm_area_struct *vma)
1074{
1075 struct bpf_map *map = vma->vm_file->private_data;
1076
1077 if (vma->vm_flags & VM_MAYWRITE)
1078 bpf_map_write_active_dec(map);
1079}
1080
1081static const struct vm_operations_struct bpf_map_default_vmops = {
1082 .open = bpf_map_mmap_open,
1083 .close = bpf_map_mmap_close,
1084};
1085
1086static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
1087{
1088 struct bpf_map *map = filp->private_data;
1089 int err = 0;
1090
1091 if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
1092 return -ENOTSUPP;
1093
1094 if (!(vma->vm_flags & VM_SHARED))
1095 return -EINVAL;
1096
1097 mutex_lock(&map->freeze_mutex);
1098
1099 if (vma->vm_flags & VM_WRITE) {
1100 if (map->frozen) {
1101 err = -EPERM;
1102 goto out;
1103 }
1104 /* map is meant to be read-only, so do not allow mapping as
1105 * writable, because it's possible to leak a writable page
1106 * reference and allows user-space to still modify it after
1107 * freezing, while verifier will assume contents do not change
1108 */
1109 if (map->map_flags & BPF_F_RDONLY_PROG) {
1110 err = -EACCES;
1111 goto out;
1112 }
1113 bpf_map_write_active_inc(map);
1114 }
1115out:
1116 mutex_unlock(&map->freeze_mutex);
1117 if (err)
1118 return err;
1119
1120 /* set default open/close callbacks */
1121 vma->vm_ops = &bpf_map_default_vmops;
1122 vma->vm_private_data = map;
1123 vm_flags_clear(vma, VM_MAYEXEC);
1124 /* If mapping is read-only, then disallow potentially re-mapping with
1125 * PROT_WRITE by dropping VM_MAYWRITE flag. This VM_MAYWRITE clearing
1126 * means that as far as BPF map's memory-mapped VMAs are concerned,
1127 * VM_WRITE and VM_MAYWRITE and equivalent, if one of them is set,
1128 * both should be set, so we can forget about VM_MAYWRITE and always
1129 * check just VM_WRITE
1130 */
1131 if (!(vma->vm_flags & VM_WRITE))
1132 vm_flags_clear(vma, VM_MAYWRITE);
1133
1134 err = map->ops->map_mmap(map, vma);
1135 if (err) {
1136 if (vma->vm_flags & VM_WRITE)
1137 bpf_map_write_active_dec(map);
1138 }
1139
1140 return err;
1141}
1142
1143static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
1144{
1145 struct bpf_map *map = filp->private_data;
1146
1147 if (map->ops->map_poll)
1148 return map->ops->map_poll(map, filp, pts);
1149
1150 return EPOLLERR;
1151}
1152
1153static unsigned long bpf_get_unmapped_area(struct file *filp, unsigned long addr,
1154 unsigned long len, unsigned long pgoff,
1155 unsigned long flags)
1156{
1157 struct bpf_map *map = filp->private_data;
1158
1159 if (map->ops->map_get_unmapped_area)
1160 return map->ops->map_get_unmapped_area(filp, addr, len, pgoff, flags);
1161#ifdef CONFIG_MMU
1162 return mm_get_unmapped_area(filp, addr, len, pgoff, flags);
1163#else
1164 return addr;
1165#endif
1166}
1167
1168const struct file_operations bpf_map_fops = {
1169#ifdef CONFIG_PROC_FS
1170 .show_fdinfo = bpf_map_show_fdinfo,
1171#endif
1172 .release = bpf_map_release,
1173 .read = bpf_dummy_read,
1174 .write = bpf_dummy_write,
1175 .mmap = bpf_map_mmap,
1176 .poll = bpf_map_poll,
1177 .get_unmapped_area = bpf_get_unmapped_area,
1178};
1179
1180int bpf_map_new_fd(struct bpf_map *map, int flags)
1181{
1182 int ret;
1183
1184 ret = security_bpf_map(map, OPEN_FMODE(flags));
1185 if (ret < 0)
1186 return ret;
1187
1188 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
1189 flags | O_CLOEXEC);
1190}
1191
1192int bpf_get_file_flag(int flags)
1193{
1194 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
1195 return -EINVAL;
1196 if (flags & BPF_F_RDONLY)
1197 return O_RDONLY;
1198 if (flags & BPF_F_WRONLY)
1199 return O_WRONLY;
1200 return O_RDWR;
1201}
1202
1203/* helper macro to check that unused fields 'union bpf_attr' are zero */
1204#define CHECK_ATTR(CMD) \
1205 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
1206 sizeof(attr->CMD##_LAST_FIELD), 0, \
1207 sizeof(*attr) - \
1208 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
1209 sizeof(attr->CMD##_LAST_FIELD)) != NULL
1210
1211/* dst and src must have at least "size" number of bytes.
1212 * Return strlen on success and < 0 on error.
1213 */
1214int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
1215{
1216 const char *end = src + size;
1217 const char *orig_src = src;
1218
1219 memset(dst, 0, size);
1220 /* Copy all isalnum(), '_' and '.' chars. */
1221 while (src < end && *src) {
1222 if (!isalnum(*src) &&
1223 *src != '_' && *src != '.')
1224 return -EINVAL;
1225 *dst++ = *src++;
1226 }
1227
1228 /* No '\0' found in "size" number of bytes */
1229 if (src == end)
1230 return -EINVAL;
1231
1232 return src - orig_src;
1233}
1234EXPORT_SYMBOL_GPL(bpf_obj_name_cpy);
1235
1236int map_check_no_btf(struct bpf_map *map,
1237 const struct btf *btf,
1238 const struct btf_type *key_type,
1239 const struct btf_type *value_type)
1240{
1241 return -ENOTSUPP;
1242}
1243
1244static int map_check_btf(struct bpf_map *map, struct bpf_token *token,
1245 const struct btf *btf, u32 btf_key_id, u32 btf_value_id)
1246{
1247 const struct btf_type *key_type, *value_type;
1248 u32 key_size, value_size;
1249 int ret = 0;
1250
1251 /* Some maps allow key to be unspecified. */
1252 if (btf_key_id) {
1253 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
1254 if (!key_type || key_size != map->key_size)
1255 return -EINVAL;
1256 } else {
1257 key_type = btf_type_by_id(btf, 0);
1258 if (!map->ops->map_check_btf)
1259 return -EINVAL;
1260 }
1261
1262 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1263 if (!value_type || value_size != map->value_size)
1264 return -EINVAL;
1265
1266 map->record = btf_parse_fields(btf, value_type,
1267 BPF_SPIN_LOCK | BPF_RES_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
1268 BPF_RB_ROOT | BPF_REFCOUNT | BPF_WORKQUEUE | BPF_UPTR |
1269 BPF_TASK_WORK,
1270 map->value_size);
1271 if (!IS_ERR_OR_NULL(map->record)) {
1272 int i;
1273
1274 if (!bpf_token_capable(token, CAP_BPF)) {
1275 ret = -EPERM;
1276 goto free_map_tab;
1277 }
1278 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1279 ret = -EACCES;
1280 goto free_map_tab;
1281 }
1282 for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1283 switch (map->record->field_mask & (1 << i)) {
1284 case 0:
1285 continue;
1286 case BPF_SPIN_LOCK:
1287 case BPF_RES_SPIN_LOCK:
1288 if (map->map_type != BPF_MAP_TYPE_HASH &&
1289 map->map_type != BPF_MAP_TYPE_RHASH &&
1290 map->map_type != BPF_MAP_TYPE_ARRAY &&
1291 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1292 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1293 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1294 map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1295 map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1296 ret = -EOPNOTSUPP;
1297 goto free_map_tab;
1298 }
1299 break;
1300 case BPF_TIMER:
1301 case BPF_WORKQUEUE:
1302 case BPF_TASK_WORK:
1303 if (map->map_type != BPF_MAP_TYPE_HASH &&
1304 map->map_type != BPF_MAP_TYPE_RHASH &&
1305 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1306 map->map_type != BPF_MAP_TYPE_ARRAY) {
1307 ret = -EOPNOTSUPP;
1308 goto free_map_tab;
1309 }
1310 break;
1311 case BPF_KPTR_UNREF:
1312 case BPF_KPTR_REF:
1313 case BPF_KPTR_PERCPU:
1314 case BPF_REFCOUNT:
1315 if (map->map_type != BPF_MAP_TYPE_HASH &&
1316 map->map_type != BPF_MAP_TYPE_RHASH &&
1317 map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
1318 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1319 map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH &&
1320 map->map_type != BPF_MAP_TYPE_ARRAY &&
1321 map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY &&
1322 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1323 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1324 map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1325 map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1326 ret = -EOPNOTSUPP;
1327 goto free_map_tab;
1328 }
1329 break;
1330 case BPF_UPTR:
1331 if (map->map_type != BPF_MAP_TYPE_TASK_STORAGE) {
1332 ret = -EOPNOTSUPP;
1333 goto free_map_tab;
1334 }
1335 break;
1336 case BPF_LIST_HEAD:
1337 case BPF_RB_ROOT:
1338 if (map->map_type != BPF_MAP_TYPE_HASH &&
1339 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1340 map->map_type != BPF_MAP_TYPE_ARRAY) {
1341 ret = -EOPNOTSUPP;
1342 goto free_map_tab;
1343 }
1344 break;
1345 default:
1346 /* Fail if map_type checks are missing for a field type */
1347 ret = -EOPNOTSUPP;
1348 goto free_map_tab;
1349 }
1350 }
1351 }
1352
1353 ret = btf_check_and_fixup_fields(btf, map->record);
1354 if (ret < 0)
1355 goto free_map_tab;
1356
1357 if (map->ops->map_check_btf) {
1358 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1359 if (ret < 0)
1360 goto free_map_tab;
1361 }
1362
1363 return ret;
1364free_map_tab:
1365 bpf_map_free_record(map);
1366 return ret;
1367}
1368
1369#define BPF_MAP_CREATE_LAST_FIELD excl_prog_hash_size
1370/* called via syscall */
1371static int map_create_alloc(union bpf_attr *attr, bpfptr_t uattr, struct bpf_verifier_log *log,
1372 struct bpf_map **mapp, struct bpf_token **tokenp)
1373{
1374 const struct bpf_map_ops *ops;
1375 struct bpf_token *token = NULL;
1376 int numa_node = bpf_map_attr_numa_node(attr);
1377 u32 map_type = attr->map_type;
1378 struct bpf_map *map;
1379 bool token_flag;
1380 int err;
1381
1382 err = CHECK_ATTR(BPF_MAP_CREATE);
1383 if (err) {
1384 bpf_log(log, "Invalid attr.\n");
1385 return -EINVAL;
1386 }
1387
1388 /* check BPF_F_TOKEN_FD flag, remember if it's set, and then clear it
1389 * to avoid per-map type checks tripping on unknown flag
1390 */
1391 token_flag = attr->map_flags & BPF_F_TOKEN_FD;
1392 attr->map_flags &= ~BPF_F_TOKEN_FD;
1393
1394 if (attr->btf_vmlinux_value_type_id) {
1395 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS) {
1396 bpf_log(log, "btf_vmlinux_value_type_id can only be used with struct_ops maps.\n");
1397 return -EINVAL;
1398 }
1399 if (attr->btf_key_type_id || attr->btf_value_type_id) {
1400 bpf_log(log, "btf_vmlinux_value_type_id is mutually exclusive with btf_key_type_id and btf_value_type_id.\n");
1401 return -EINVAL;
1402 }
1403 } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1404 bpf_log(log, "Invalid btf_value_type_id.\n");
1405 return -EINVAL;
1406 }
1407
1408 if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1409 attr->map_type != BPF_MAP_TYPE_ARENA &&
1410 attr->map_type != BPF_MAP_TYPE_RHASH &&
1411 attr->map_extra != 0) {
1412 bpf_log(log, "Invalid map_extra.\n");
1413 return -EINVAL;
1414 }
1415
1416 if (numa_node != NUMA_NO_NODE &&
1417 ((unsigned int)numa_node >= nr_node_ids ||
1418 !node_online(numa_node))) {
1419 bpf_log(log, "Invalid numa_node.\n");
1420 return -EINVAL;
1421 }
1422
1423 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1424 map_type = attr->map_type;
1425 if (map_type >= ARRAY_SIZE(bpf_map_types)) {
1426 bpf_log(log, "Invalid map_type.\n");
1427 return -EINVAL;
1428 }
1429 map_type = array_index_nospec(map_type, ARRAY_SIZE(bpf_map_types));
1430 ops = bpf_map_types[map_type];
1431 if (!ops)
1432 return -EINVAL;
1433
1434 if (ops->map_alloc_check) {
1435 err = ops->map_alloc_check(attr);
1436 if (err)
1437 return err;
1438 }
1439 if (attr->map_ifindex)
1440 ops = &bpf_map_offload_ops;
1441 if (!ops->map_mem_usage)
1442 return -EINVAL;
1443
1444 if (token_flag) {
1445 token = bpf_token_get_from_fd(attr->map_token_fd);
1446 if (IS_ERR(token)) {
1447 bpf_log(log, "Invalid map_token_fd.\n");
1448 return PTR_ERR(token);
1449 }
1450
1451 /* if current token doesn't grant map creation permissions,
1452 * then we can't use this token, so ignore it and rely on
1453 * system-wide capabilities checks
1454 */
1455 if (!bpf_token_allow_cmd(token, BPF_MAP_CREATE) ||
1456 !bpf_token_allow_map_type(token, attr->map_type)) {
1457 bpf_token_put(token);
1458 token = NULL;
1459 }
1460 }
1461
1462 err = -EPERM;
1463
1464 /* Intent here is for unprivileged_bpf_disabled to block BPF map
1465 * creation for unprivileged users; other actions depend
1466 * on fd availability and access to bpffs, so are dependent on
1467 * object creation success. Even with unprivileged BPF disabled,
1468 * capability checks are still carried out.
1469 */
1470 if (sysctl_unprivileged_bpf_disabled && !bpf_token_capable(token, CAP_BPF))
1471 goto put_token;
1472
1473 /* check privileged map type permissions */
1474 switch (map_type) {
1475 case BPF_MAP_TYPE_ARRAY:
1476 case BPF_MAP_TYPE_PERCPU_ARRAY:
1477 case BPF_MAP_TYPE_PROG_ARRAY:
1478 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1479 case BPF_MAP_TYPE_CGROUP_ARRAY:
1480 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1481 case BPF_MAP_TYPE_HASH:
1482 case BPF_MAP_TYPE_RHASH:
1483 case BPF_MAP_TYPE_PERCPU_HASH:
1484 case BPF_MAP_TYPE_HASH_OF_MAPS:
1485 case BPF_MAP_TYPE_RINGBUF:
1486 case BPF_MAP_TYPE_USER_RINGBUF:
1487 case BPF_MAP_TYPE_CGROUP_STORAGE:
1488 case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
1489 /* unprivileged */
1490 break;
1491 case BPF_MAP_TYPE_SK_STORAGE:
1492 case BPF_MAP_TYPE_INODE_STORAGE:
1493 case BPF_MAP_TYPE_TASK_STORAGE:
1494 case BPF_MAP_TYPE_CGRP_STORAGE:
1495 case BPF_MAP_TYPE_BLOOM_FILTER:
1496 case BPF_MAP_TYPE_LPM_TRIE:
1497 case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
1498 case BPF_MAP_TYPE_STACK_TRACE:
1499 case BPF_MAP_TYPE_QUEUE:
1500 case BPF_MAP_TYPE_STACK:
1501 case BPF_MAP_TYPE_LRU_HASH:
1502 case BPF_MAP_TYPE_LRU_PERCPU_HASH:
1503 case BPF_MAP_TYPE_STRUCT_OPS:
1504 case BPF_MAP_TYPE_CPUMAP:
1505 case BPF_MAP_TYPE_ARENA:
1506 case BPF_MAP_TYPE_INSN_ARRAY:
1507 if (!bpf_token_capable(token, CAP_BPF))
1508 goto put_token;
1509 break;
1510 case BPF_MAP_TYPE_SOCKMAP:
1511 case BPF_MAP_TYPE_SOCKHASH:
1512 case BPF_MAP_TYPE_DEVMAP:
1513 case BPF_MAP_TYPE_DEVMAP_HASH:
1514 case BPF_MAP_TYPE_XSKMAP:
1515 if (!bpf_token_capable(token, CAP_NET_ADMIN))
1516 goto put_token;
1517 break;
1518 default:
1519 WARN(1, "unsupported map type %d", map_type);
1520 goto put_token;
1521 }
1522
1523 map = ops->map_alloc(attr);
1524 if (IS_ERR(map)) {
1525 err = PTR_ERR(map);
1526 goto put_token;
1527 }
1528 map->ops = ops;
1529 map->map_type = map_type;
1530
1531 err = bpf_obj_name_cpy(map->name, attr->map_name,
1532 sizeof(attr->map_name));
1533 if (err < 0) {
1534 bpf_log(log, "Invalid map_name.\n");
1535 goto free_map;
1536 }
1537
1538 preempt_disable();
1539 map->cookie = gen_cookie_next(&bpf_map_cookie);
1540 preempt_enable();
1541
1542 atomic64_set(&map->refcnt, 1);
1543 atomic64_set(&map->usercnt, 1);
1544 mutex_init(&map->freeze_mutex);
1545 spin_lock_init(&map->owner_lock);
1546
1547 if (attr->btf_key_type_id || attr->btf_value_type_id ||
1548 /* Even the map's value is a kernel's struct,
1549 * the bpf_prog.o must have BTF to begin with
1550 * to figure out the corresponding kernel's
1551 * counter part. Thus, attr->btf_fd has
1552 * to be valid also.
1553 */
1554 attr->btf_vmlinux_value_type_id) {
1555 struct btf *btf;
1556
1557 btf = btf_get_by_fd(attr->btf_fd);
1558 if (IS_ERR(btf)) {
1559 bpf_log(log, "Invalid btf_fd.\n");
1560 err = PTR_ERR(btf);
1561 goto free_map;
1562 }
1563 if (btf_is_kernel(btf)) {
1564 btf_put(btf);
1565 err = -EACCES;
1566 goto free_map;
1567 }
1568 map->btf = btf;
1569
1570 if (attr->btf_value_type_id) {
1571 err = map_check_btf(map, token, btf, attr->btf_key_type_id,
1572 attr->btf_value_type_id);
1573 if (err)
1574 goto free_map;
1575 }
1576
1577 map->btf_key_type_id = attr->btf_key_type_id;
1578 map->btf_value_type_id = attr->btf_value_type_id;
1579 map->btf_vmlinux_value_type_id =
1580 attr->btf_vmlinux_value_type_id;
1581 }
1582
1583 if (attr->excl_prog_hash) {
1584 bpfptr_t uprog_hash = make_bpfptr(attr->excl_prog_hash, uattr.is_kernel);
1585
1586 if (attr->excl_prog_hash_size != SHA256_DIGEST_SIZE) {
1587 bpf_log(log, "Invalid excl_prog_hash_size.\n");
1588 err = -EINVAL;
1589 goto free_map;
1590 }
1591
1592 map->excl_prog_sha = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
1593 if (!map->excl_prog_sha) {
1594 err = -ENOMEM;
1595 goto free_map;
1596 }
1597
1598 if (copy_from_bpfptr(map->excl_prog_sha, uprog_hash, SHA256_DIGEST_SIZE)) {
1599 err = -EFAULT;
1600 goto free_map;
1601 }
1602
1603 /* See libbpf: emit_signature_match() */
1604 BUILD_BUG_ON(offsetof(struct bpf_map, excl) != SHA256_DIGEST_SIZE);
1605 BUILD_BUG_ON(!__same_type(map->excl, u32));
1606 BUILD_BUG_ON(offsetof(struct bpf_map, sha) != 0);
1607 BUILD_BUG_ON(!__same_type(map->sha, u8[SHA256_DIGEST_SIZE]));
1608 map->excl = 1;
1609 } else if (attr->excl_prog_hash_size) {
1610 bpf_log(log, "Invalid excl_prog_hash_size.\n");
1611 err = -EINVAL;
1612 goto free_map;
1613 }
1614
1615 *mapp = map;
1616 *tokenp = token;
1617 return 0;
1618
1619free_map:
1620 bpf_map_free(map);
1621put_token:
1622 bpf_token_put(token);
1623 return err;
1624}
1625
1626static int map_create(union bpf_attr *attr, bpfptr_t uattr, struct bpf_common_attr *attr_common,
1627 bpfptr_t uattr_common, u32 size_common)
1628{
1629 struct bpf_token *token = NULL;
1630 struct bpf_verifier_log *log;
1631 struct bpf_log_attr attr_log;
1632 struct bpf_map *map = NULL;
1633 int err, ret;
1634 int f_flags;
1635
1636 log = bpf_log_attr_create_vlog(&attr_log, attr_common, uattr_common, size_common);
1637 if (IS_ERR(log))
1638 return PTR_ERR(log);
1639
1640 err = map_create_alloc(attr, uattr, log, &map, &token);
1641
1642 /* preserve original error even if log finalization is successful */
1643 ret = bpf_log_attr_finalize(&attr_log, log);
1644 if (ret)
1645 err = ret;
1646
1647 kfree(log);
1648
1649 if (err)
1650 goto free_map;
1651
1652 f_flags = bpf_get_file_flag(attr->map_flags);
1653 if (f_flags < 0) {
1654 err = f_flags;
1655 goto free_map;
1656 }
1657
1658 err = security_bpf_map_create(map, attr, token, uattr.is_kernel);
1659 if (err)
1660 goto free_map_sec;
1661
1662 err = bpf_map_alloc_id(map);
1663 if (err)
1664 goto free_map_sec;
1665
1666 bpf_map_save_memcg(map);
1667 bpf_token_put(token);
1668
1669 err = bpf_map_new_fd(map, f_flags);
1670 if (err < 0) {
1671 /* failed to allocate fd.
1672 * bpf_map_put_with_uref() is needed because the above
1673 * bpf_map_alloc_id() has published the map
1674 * to the userspace and the userspace may
1675 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1676 */
1677 bpf_map_put_with_uref(map);
1678 return err;
1679 }
1680
1681 return err;
1682
1683free_map_sec:
1684 security_bpf_map_free(map);
1685free_map:
1686 if (map)
1687 bpf_map_free(map);
1688 bpf_token_put(token);
1689 return err;
1690}
1691
1692void bpf_map_inc(struct bpf_map *map)
1693{
1694 atomic64_inc(&map->refcnt);
1695}
1696EXPORT_SYMBOL_GPL(bpf_map_inc);
1697
1698void bpf_map_inc_with_uref(struct bpf_map *map)
1699{
1700 atomic64_inc(&map->refcnt);
1701 atomic64_inc(&map->usercnt);
1702}
1703EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1704
1705struct bpf_map *bpf_map_get(u32 ufd)
1706{
1707 CLASS(fd, f)(ufd);
1708 struct bpf_map *map = __bpf_map_get(f);
1709
1710 if (!IS_ERR(map))
1711 bpf_map_inc(map);
1712
1713 return map;
1714}
1715EXPORT_SYMBOL_NS(bpf_map_get, "BPF_INTERNAL");
1716
1717struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1718{
1719 CLASS(fd, f)(ufd);
1720 struct bpf_map *map = __bpf_map_get(f);
1721
1722 if (!IS_ERR(map))
1723 bpf_map_inc_with_uref(map);
1724
1725 return map;
1726}
1727
1728/* map_idr_lock should have been held or the map should have been
1729 * protected by rcu read lock.
1730 */
1731struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1732{
1733 int refold;
1734
1735 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1736 if (!refold)
1737 return ERR_PTR(-ENOENT);
1738 if (uref)
1739 atomic64_inc(&map->usercnt);
1740
1741 return map;
1742}
1743
1744struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1745{
1746 lockdep_assert(rcu_read_lock_held());
1747 return __bpf_map_inc_not_zero(map, false);
1748}
1749EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1750
1751int __weak bpf_stackmap_extract(struct bpf_map *map, void *key, void *value,
1752 bool delete)
1753{
1754 return -ENOTSUPP;
1755}
1756
1757static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1758{
1759 if (key_size)
1760 return vmemdup_user(ukey, key_size);
1761
1762 if (ukey)
1763 return ERR_PTR(-EINVAL);
1764
1765 return NULL;
1766}
1767
1768static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1769{
1770 if (key_size)
1771 return kvmemdup_bpfptr(ukey, key_size);
1772
1773 if (!bpfptr_is_null(ukey))
1774 return ERR_PTR(-EINVAL);
1775
1776 return NULL;
1777}
1778
1779/* last field in 'union bpf_attr' used by this command */
1780#define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1781
1782static int map_lookup_elem(union bpf_attr *attr)
1783{
1784 void __user *ukey = u64_to_user_ptr(attr->key);
1785 void __user *uvalue = u64_to_user_ptr(attr->value);
1786 struct bpf_map *map;
1787 void *key, *value;
1788 u32 value_size;
1789 int err;
1790
1791 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1792 return -EINVAL;
1793
1794 CLASS(fd, f)(attr->map_fd);
1795 map = __bpf_map_get(f);
1796 if (IS_ERR(map))
1797 return PTR_ERR(map);
1798 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ))
1799 return -EPERM;
1800
1801 err = bpf_map_check_op_flags(map, attr->flags, BPF_F_LOCK | BPF_F_CPU);
1802 if (err)
1803 return err;
1804
1805 key = __bpf_copy_key(ukey, map->key_size);
1806 if (IS_ERR(key))
1807 return PTR_ERR(key);
1808
1809 value_size = bpf_map_value_size(map, attr->flags);
1810
1811 err = -ENOMEM;
1812 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1813 if (!value)
1814 goto free_key;
1815
1816 if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1817 if (copy_from_user(value, uvalue, value_size))
1818 err = -EFAULT;
1819 else
1820 err = bpf_map_copy_value(map, key, value, attr->flags);
1821 goto free_value;
1822 }
1823
1824 err = bpf_map_copy_value(map, key, value, attr->flags);
1825 if (err)
1826 goto free_value;
1827
1828 err = -EFAULT;
1829 if (copy_to_user(uvalue, value, value_size) != 0)
1830 goto free_value;
1831
1832 err = 0;
1833
1834free_value:
1835 kvfree(value);
1836free_key:
1837 kvfree(key);
1838 return err;
1839}
1840
1841
1842#define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1843
1844static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1845{
1846 bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1847 bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1848 struct bpf_map *map;
1849 void *key, *value;
1850 u32 value_size;
1851 int err;
1852
1853 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1854 return -EINVAL;
1855
1856 CLASS(fd, f)(attr->map_fd);
1857 map = __bpf_map_get(f);
1858 if (IS_ERR(map))
1859 return PTR_ERR(map);
1860 bpf_map_write_active_inc(map);
1861 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1862 err = -EPERM;
1863 goto err_put;
1864 }
1865
1866 err = bpf_map_check_op_flags(map, attr->flags, ~0);
1867 if (err)
1868 goto err_put;
1869
1870 key = ___bpf_copy_key(ukey, map->key_size);
1871 if (IS_ERR(key)) {
1872 err = PTR_ERR(key);
1873 goto err_put;
1874 }
1875
1876 value_size = bpf_map_value_size(map, attr->flags);
1877 value = kvmemdup_bpfptr(uvalue, value_size);
1878 if (IS_ERR(value)) {
1879 err = PTR_ERR(value);
1880 goto free_key;
1881 }
1882
1883 err = bpf_map_update_value(map, fd_file(f), key, value, attr->flags);
1884 if (!err)
1885 maybe_wait_bpf_programs(map);
1886
1887 kvfree(value);
1888free_key:
1889 kvfree(key);
1890err_put:
1891 bpf_map_write_active_dec(map);
1892 return err;
1893}
1894
1895#define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1896
1897static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1898{
1899 bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1900 struct bpf_map *map;
1901 void *key;
1902 int err;
1903
1904 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1905 return -EINVAL;
1906
1907 CLASS(fd, f)(attr->map_fd);
1908 map = __bpf_map_get(f);
1909 if (IS_ERR(map))
1910 return PTR_ERR(map);
1911 bpf_map_write_active_inc(map);
1912 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1913 err = -EPERM;
1914 goto err_put;
1915 }
1916
1917 key = ___bpf_copy_key(ukey, map->key_size);
1918 if (IS_ERR(key)) {
1919 err = PTR_ERR(key);
1920 goto err_put;
1921 }
1922
1923 if (bpf_map_is_offloaded(map)) {
1924 err = bpf_map_offload_delete_elem(map, key);
1925 goto out;
1926 } else if (IS_FD_PROG_ARRAY(map) ||
1927 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1928 /* These maps require sleepable context */
1929 err = map->ops->map_delete_elem(map, key);
1930 goto out;
1931 }
1932
1933 bpf_disable_instrumentation();
1934 rcu_read_lock();
1935 err = map->ops->map_delete_elem(map, key);
1936 rcu_read_unlock();
1937 bpf_enable_instrumentation();
1938 if (!err)
1939 maybe_wait_bpf_programs(map);
1940out:
1941 kvfree(key);
1942err_put:
1943 bpf_map_write_active_dec(map);
1944 return err;
1945}
1946
1947/* last field in 'union bpf_attr' used by this command */
1948#define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1949
1950static int map_get_next_key(union bpf_attr *attr)
1951{
1952 void __user *ukey = u64_to_user_ptr(attr->key);
1953 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1954 struct bpf_map *map;
1955 void *key, *next_key;
1956 int err;
1957
1958 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1959 return -EINVAL;
1960
1961 CLASS(fd, f)(attr->map_fd);
1962 map = __bpf_map_get(f);
1963 if (IS_ERR(map))
1964 return PTR_ERR(map);
1965 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ))
1966 return -EPERM;
1967
1968 if (ukey) {
1969 key = __bpf_copy_key(ukey, map->key_size);
1970 if (IS_ERR(key))
1971 return PTR_ERR(key);
1972 } else {
1973 key = NULL;
1974 }
1975
1976 err = -ENOMEM;
1977 next_key = kvmalloc(map->key_size, GFP_USER);
1978 if (!next_key)
1979 goto free_key;
1980
1981 if (bpf_map_is_offloaded(map)) {
1982 err = bpf_map_offload_get_next_key(map, key, next_key);
1983 goto out;
1984 }
1985
1986 rcu_read_lock();
1987 err = map->ops->map_get_next_key(map, key, next_key);
1988 rcu_read_unlock();
1989out:
1990 if (err)
1991 goto free_next_key;
1992
1993 err = -EFAULT;
1994 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1995 goto free_next_key;
1996
1997 err = 0;
1998
1999free_next_key:
2000 kvfree(next_key);
2001free_key:
2002 kvfree(key);
2003 return err;
2004}
2005
2006int generic_map_delete_batch(struct bpf_map *map,
2007 const union bpf_attr *attr,
2008 union bpf_attr __user *uattr)
2009{
2010 void __user *keys = u64_to_user_ptr(attr->batch.keys);
2011 u32 cp, max_count;
2012 int err = 0;
2013 void *key;
2014
2015 if (attr->batch.elem_flags & ~BPF_F_LOCK)
2016 return -EINVAL;
2017
2018 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
2019 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
2020 return -EINVAL;
2021 }
2022
2023 max_count = attr->batch.count;
2024 if (!max_count)
2025 return 0;
2026
2027 if (put_user(0, &uattr->batch.count))
2028 return -EFAULT;
2029
2030 key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
2031 if (!key)
2032 return -ENOMEM;
2033
2034 for (cp = 0; cp < max_count; cp++) {
2035 err = -EFAULT;
2036 if (copy_from_user(key, keys + cp * map->key_size,
2037 map->key_size))
2038 break;
2039
2040 if (bpf_map_is_offloaded(map)) {
2041 err = bpf_map_offload_delete_elem(map, key);
2042 break;
2043 }
2044
2045 bpf_disable_instrumentation();
2046 rcu_read_lock();
2047 err = map->ops->map_delete_elem(map, key);
2048 rcu_read_unlock();
2049 bpf_enable_instrumentation();
2050 if (err)
2051 break;
2052 cond_resched();
2053 }
2054 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
2055 err = -EFAULT;
2056
2057 kvfree(key);
2058
2059 return err;
2060}
2061
2062int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
2063 const union bpf_attr *attr,
2064 union bpf_attr __user *uattr)
2065{
2066 void __user *values = u64_to_user_ptr(attr->batch.values);
2067 void __user *keys = u64_to_user_ptr(attr->batch.keys);
2068 u32 value_size, cp, max_count;
2069 void *key, *value;
2070 int err = 0;
2071
2072 err = bpf_map_check_op_flags(map, attr->batch.elem_flags,
2073 BPF_F_LOCK | BPF_F_CPU | BPF_F_ALL_CPUS);
2074 if (err)
2075 return err;
2076
2077 value_size = bpf_map_value_size(map, attr->batch.elem_flags);
2078
2079 max_count = attr->batch.count;
2080 if (!max_count)
2081 return 0;
2082
2083 if (put_user(0, &uattr->batch.count))
2084 return -EFAULT;
2085
2086 key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
2087 if (!key)
2088 return -ENOMEM;
2089
2090 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
2091 if (!value) {
2092 kvfree(key);
2093 return -ENOMEM;
2094 }
2095
2096 for (cp = 0; cp < max_count; cp++) {
2097 err = -EFAULT;
2098 if (copy_from_user(key, keys + cp * map->key_size,
2099 map->key_size) ||
2100 copy_from_user(value, values + cp * value_size, value_size))
2101 break;
2102
2103 err = bpf_map_update_value(map, map_file, key, value,
2104 attr->batch.elem_flags);
2105
2106 if (err)
2107 break;
2108 cond_resched();
2109 }
2110
2111 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
2112 err = -EFAULT;
2113
2114 kvfree(value);
2115 kvfree(key);
2116
2117 return err;
2118}
2119
2120int generic_map_lookup_batch(struct bpf_map *map,
2121 const union bpf_attr *attr,
2122 union bpf_attr __user *uattr)
2123{
2124 void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
2125 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
2126 void __user *values = u64_to_user_ptr(attr->batch.values);
2127 void __user *keys = u64_to_user_ptr(attr->batch.keys);
2128 void *buf, *buf_prevkey, *prev_key, *key, *value;
2129 u32 value_size, cp, max_count;
2130 int err;
2131
2132 err = bpf_map_check_op_flags(map, attr->batch.elem_flags, BPF_F_LOCK | BPF_F_CPU);
2133 if (err)
2134 return err;
2135
2136 value_size = bpf_map_value_size(map, attr->batch.elem_flags);
2137
2138 max_count = attr->batch.count;
2139 if (!max_count)
2140 return 0;
2141
2142 if (put_user(0, &uattr->batch.count))
2143 return -EFAULT;
2144
2145 buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
2146 if (!buf_prevkey)
2147 return -ENOMEM;
2148
2149 buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
2150 if (!buf) {
2151 kvfree(buf_prevkey);
2152 return -ENOMEM;
2153 }
2154
2155 err = -EFAULT;
2156 prev_key = NULL;
2157 if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
2158 goto free_buf;
2159 key = buf;
2160 value = key + map->key_size;
2161 if (ubatch)
2162 prev_key = buf_prevkey;
2163
2164 for (cp = 0; cp < max_count;) {
2165 rcu_read_lock();
2166 err = map->ops->map_get_next_key(map, prev_key, key);
2167 rcu_read_unlock();
2168 if (err)
2169 break;
2170 err = bpf_map_copy_value(map, key, value,
2171 attr->batch.elem_flags);
2172
2173 if (err == -ENOENT)
2174 goto next_key;
2175
2176 if (err)
2177 goto free_buf;
2178
2179 if (copy_to_user(keys + cp * map->key_size, key,
2180 map->key_size)) {
2181 err = -EFAULT;
2182 goto free_buf;
2183 }
2184 if (copy_to_user(values + cp * value_size, value, value_size)) {
2185 err = -EFAULT;
2186 goto free_buf;
2187 }
2188
2189 cp++;
2190next_key:
2191 if (!prev_key)
2192 prev_key = buf_prevkey;
2193
2194 swap(prev_key, key);
2195 cond_resched();
2196 }
2197
2198 if (err == -EFAULT)
2199 goto free_buf;
2200
2201 if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
2202 (cp && copy_to_user(uobatch, prev_key, map->key_size))))
2203 err = -EFAULT;
2204
2205free_buf:
2206 kvfree(buf_prevkey);
2207 kvfree(buf);
2208 return err;
2209}
2210
2211#define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
2212
2213static int map_lookup_and_delete_elem(union bpf_attr *attr)
2214{
2215 void __user *ukey = u64_to_user_ptr(attr->key);
2216 void __user *uvalue = u64_to_user_ptr(attr->value);
2217 struct bpf_map *map;
2218 void *key, *value;
2219 u32 value_size;
2220 int err;
2221
2222 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
2223 return -EINVAL;
2224
2225 if (attr->flags & ~BPF_F_LOCK)
2226 return -EINVAL;
2227
2228 CLASS(fd, f)(attr->map_fd);
2229 map = __bpf_map_get(f);
2230 if (IS_ERR(map))
2231 return PTR_ERR(map);
2232 bpf_map_write_active_inc(map);
2233 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
2234 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
2235 err = -EPERM;
2236 goto err_put;
2237 }
2238
2239 if (attr->flags &&
2240 (map->map_type == BPF_MAP_TYPE_QUEUE ||
2241 map->map_type == BPF_MAP_TYPE_STACK)) {
2242 err = -EINVAL;
2243 goto err_put;
2244 }
2245
2246 if ((attr->flags & BPF_F_LOCK) &&
2247 !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
2248 err = -EINVAL;
2249 goto err_put;
2250 }
2251
2252 key = __bpf_copy_key(ukey, map->key_size);
2253 if (IS_ERR(key)) {
2254 err = PTR_ERR(key);
2255 goto err_put;
2256 }
2257
2258 value_size = bpf_map_value_size(map, 0);
2259
2260 err = -ENOMEM;
2261 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
2262 if (!value)
2263 goto free_key;
2264
2265 err = -ENOTSUPP;
2266 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
2267 map->map_type == BPF_MAP_TYPE_STACK) {
2268 err = map->ops->map_pop_elem(map, value);
2269 } else if (map->map_type == BPF_MAP_TYPE_HASH ||
2270 map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
2271 map->map_type == BPF_MAP_TYPE_LRU_HASH ||
2272 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
2273 map->map_type == BPF_MAP_TYPE_RHASH ||
2274 map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
2275 if (!bpf_map_is_offloaded(map)) {
2276 bpf_disable_instrumentation();
2277 rcu_read_lock();
2278 err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
2279 rcu_read_unlock();
2280 bpf_enable_instrumentation();
2281 }
2282 }
2283
2284 if (err)
2285 goto free_value;
2286
2287 if (copy_to_user(uvalue, value, value_size) != 0) {
2288 err = -EFAULT;
2289 goto free_value;
2290 }
2291
2292 err = 0;
2293
2294free_value:
2295 kvfree(value);
2296free_key:
2297 kvfree(key);
2298err_put:
2299 bpf_map_write_active_dec(map);
2300 return err;
2301}
2302
2303#define BPF_MAP_FREEZE_LAST_FIELD map_fd
2304
2305static int map_freeze(const union bpf_attr *attr)
2306{
2307 int err = 0;
2308 struct bpf_map *map;
2309
2310 if (CHECK_ATTR(BPF_MAP_FREEZE))
2311 return -EINVAL;
2312
2313 CLASS(fd, f)(attr->map_fd);
2314 map = __bpf_map_get(f);
2315 if (IS_ERR(map))
2316 return PTR_ERR(map);
2317
2318 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record))
2319 return -ENOTSUPP;
2320
2321 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE))
2322 return -EPERM;
2323
2324 mutex_lock(&map->freeze_mutex);
2325 if (bpf_map_write_active(map)) {
2326 err = -EBUSY;
2327 goto err_put;
2328 }
2329 if (READ_ONCE(map->frozen)) {
2330 err = -EBUSY;
2331 goto err_put;
2332 }
2333
2334 WRITE_ONCE(map->frozen, true);
2335err_put:
2336 mutex_unlock(&map->freeze_mutex);
2337 return err;
2338}
2339
2340static const struct bpf_prog_ops * const bpf_prog_types[] = {
2341#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
2342 [_id] = & _name ## _prog_ops,
2343#define BPF_MAP_TYPE(_id, _ops)
2344#define BPF_LINK_TYPE(_id, _name)
2345#include <linux/bpf_types.h>
2346#undef BPF_PROG_TYPE
2347#undef BPF_MAP_TYPE
2348#undef BPF_LINK_TYPE
2349};
2350
2351static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
2352{
2353 const struct bpf_prog_ops *ops;
2354
2355 if (type >= ARRAY_SIZE(bpf_prog_types))
2356 return -EINVAL;
2357 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
2358 ops = bpf_prog_types[type];
2359 if (!ops)
2360 return -EINVAL;
2361
2362 if (!bpf_prog_is_offloaded(prog->aux))
2363 prog->aux->ops = ops;
2364 else
2365 prog->aux->ops = &bpf_offload_prog_ops;
2366 prog->type = type;
2367 return 0;
2368}
2369
2370enum bpf_audit {
2371 BPF_AUDIT_LOAD,
2372 BPF_AUDIT_UNLOAD,
2373 BPF_AUDIT_MAX,
2374};
2375
2376static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
2377 [BPF_AUDIT_LOAD] = "LOAD",
2378 [BPF_AUDIT_UNLOAD] = "UNLOAD",
2379};
2380
2381static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
2382{
2383 struct audit_context *ctx = NULL;
2384 struct audit_buffer *ab;
2385
2386 if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
2387 return;
2388 if (audit_enabled == AUDIT_OFF)
2389 return;
2390 if (!in_hardirq() && !irqs_disabled())
2391 ctx = audit_context();
2392 ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
2393 if (unlikely(!ab))
2394 return;
2395 audit_log_format(ab, "prog-id=%u op=%s",
2396 prog->aux->id, bpf_audit_str[op]);
2397 audit_log_end(ab);
2398}
2399
2400static int bpf_prog_alloc_id(struct bpf_prog *prog)
2401{
2402 int id;
2403
2404 idr_preload(GFP_KERNEL);
2405 spin_lock_bh(&prog_idr_lock);
2406 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2407 if (id > 0)
2408 prog->aux->id = id;
2409 spin_unlock_bh(&prog_idr_lock);
2410 idr_preload_end();
2411
2412 /* id is in [1, INT_MAX) */
2413 if (WARN_ON_ONCE(!id))
2414 return -ENOSPC;
2415
2416 return id > 0 ? 0 : id;
2417}
2418
2419void bpf_prog_free_id(struct bpf_prog *prog)
2420{
2421 unsigned long flags;
2422
2423 /* cBPF to eBPF migrations are currently not in the idr store.
2424 * Offloaded programs are removed from the store when their device
2425 * disappears - even if someone grabs an fd to them they are unusable,
2426 * simply waiting for refcnt to drop to be freed.
2427 */
2428 if (!prog->aux->id)
2429 return;
2430
2431 spin_lock_irqsave(&prog_idr_lock, flags);
2432 idr_remove(&prog_idr, prog->aux->id);
2433 prog->aux->id = 0;
2434 spin_unlock_irqrestore(&prog_idr_lock, flags);
2435}
2436
2437static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2438{
2439 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2440
2441 kvfree(aux->func_info);
2442 kfree(aux->func_info_aux);
2443 free_uid(aux->user);
2444 security_bpf_prog_free(aux->prog);
2445 bpf_prog_free(aux->prog);
2446}
2447
2448static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2449{
2450 bpf_prog_kallsyms_del_all(prog);
2451 btf_put(prog->aux->btf);
2452 module_put(prog->aux->mod);
2453 kvfree(prog->aux->jited_linfo);
2454 kvfree(prog->aux->linfo);
2455 kfree(prog->aux->kfunc_tab);
2456 kfree(prog->aux->ctx_arg_info);
2457 if (prog->aux->attach_btf)
2458 btf_put(prog->aux->attach_btf);
2459
2460 if (deferred) {
2461 if (prog->sleepable)
2462 call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2463 else
2464 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2465 } else {
2466 __bpf_prog_put_rcu(&prog->aux->rcu);
2467 }
2468}
2469
2470static void bpf_prog_put_deferred(struct work_struct *work)
2471{
2472 struct bpf_prog_aux *aux;
2473 struct bpf_prog *prog;
2474
2475 aux = container_of(work, struct bpf_prog_aux, work);
2476 prog = aux->prog;
2477 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2478 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2479 bpf_prog_free_id(prog);
2480 __bpf_prog_put_noref(prog, true);
2481}
2482
2483static void __bpf_prog_put(struct bpf_prog *prog)
2484{
2485 struct bpf_prog_aux *aux = prog->aux;
2486
2487 if (atomic64_dec_and_test(&aux->refcnt)) {
2488 if (in_hardirq() || irqs_disabled()) {
2489 INIT_WORK(&aux->work, bpf_prog_put_deferred);
2490 schedule_work(&aux->work);
2491 } else {
2492 bpf_prog_put_deferred(&aux->work);
2493 }
2494 }
2495}
2496
2497void bpf_prog_put(struct bpf_prog *prog)
2498{
2499 __bpf_prog_put(prog);
2500}
2501EXPORT_SYMBOL_GPL(bpf_prog_put);
2502
2503static int bpf_prog_release(struct inode *inode, struct file *filp)
2504{
2505 struct bpf_prog *prog = filp->private_data;
2506
2507 bpf_prog_put(prog);
2508 return 0;
2509}
2510
2511struct bpf_prog_kstats {
2512 u64 nsecs;
2513 u64 cnt;
2514 u64 misses;
2515};
2516
2517void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2518{
2519 struct bpf_prog_stats *stats;
2520 unsigned int flags;
2521
2522 if (unlikely(!prog->stats))
2523 return;
2524
2525 stats = this_cpu_ptr(prog->stats);
2526 flags = u64_stats_update_begin_irqsave(&stats->syncp);
2527 u64_stats_inc(&stats->misses);
2528 u64_stats_update_end_irqrestore(&stats->syncp, flags);
2529}
2530
2531static void bpf_prog_get_stats(const struct bpf_prog *prog,
2532 struct bpf_prog_kstats *stats)
2533{
2534 u64 nsecs = 0, cnt = 0, misses = 0;
2535 int cpu;
2536
2537 for_each_possible_cpu(cpu) {
2538 const struct bpf_prog_stats *st;
2539 unsigned int start;
2540 u64 tnsecs, tcnt, tmisses;
2541
2542 st = per_cpu_ptr(prog->stats, cpu);
2543 do {
2544 start = u64_stats_fetch_begin(&st->syncp);
2545 tnsecs = u64_stats_read(&st->nsecs);
2546 tcnt = u64_stats_read(&st->cnt);
2547 tmisses = u64_stats_read(&st->misses);
2548 } while (u64_stats_fetch_retry(&st->syncp, start));
2549 nsecs += tnsecs;
2550 cnt += tcnt;
2551 misses += tmisses;
2552 }
2553 stats->nsecs = nsecs;
2554 stats->cnt = cnt;
2555 stats->misses = misses;
2556}
2557
2558#ifdef CONFIG_PROC_FS
2559static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2560{
2561 const struct bpf_prog *prog = filp->private_data;
2562 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2563 struct bpf_prog_kstats stats;
2564
2565 bpf_prog_get_stats(prog, &stats);
2566 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2567 seq_printf(m,
2568 "prog_type:\t%u\n"
2569 "prog_jited:\t%u\n"
2570 "prog_tag:\t%s\n"
2571 "memlock:\t%llu\n"
2572 "prog_id:\t%u\n"
2573 "run_time_ns:\t%llu\n"
2574 "run_cnt:\t%llu\n"
2575 "recursion_misses:\t%llu\n"
2576 "verified_insns:\t%u\n",
2577 prog->type,
2578 prog->jited,
2579 prog_tag,
2580 prog->pages * 1ULL << PAGE_SHIFT,
2581 prog->aux->id,
2582 stats.nsecs,
2583 stats.cnt,
2584 stats.misses,
2585 prog->aux->verified_insns);
2586}
2587#endif
2588
2589const struct file_operations bpf_prog_fops = {
2590#ifdef CONFIG_PROC_FS
2591 .show_fdinfo = bpf_prog_show_fdinfo,
2592#endif
2593 .release = bpf_prog_release,
2594 .read = bpf_dummy_read,
2595 .write = bpf_dummy_write,
2596};
2597
2598int bpf_prog_new_fd(struct bpf_prog *prog)
2599{
2600 int ret;
2601
2602 ret = security_bpf_prog(prog);
2603 if (ret < 0)
2604 return ret;
2605
2606 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2607 O_RDWR | O_CLOEXEC);
2608}
2609
2610void bpf_prog_add(struct bpf_prog *prog, int i)
2611{
2612 atomic64_add(i, &prog->aux->refcnt);
2613}
2614EXPORT_SYMBOL_GPL(bpf_prog_add);
2615
2616void bpf_prog_sub(struct bpf_prog *prog, int i)
2617{
2618 /* Only to be used for undoing previous bpf_prog_add() in some
2619 * error path. We still know that another entity in our call
2620 * path holds a reference to the program, thus atomic_sub() can
2621 * be safely used in such cases!
2622 */
2623 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2624}
2625EXPORT_SYMBOL_GPL(bpf_prog_sub);
2626
2627void bpf_prog_inc(struct bpf_prog *prog)
2628{
2629 atomic64_inc(&prog->aux->refcnt);
2630}
2631EXPORT_SYMBOL_GPL(bpf_prog_inc);
2632
2633/* prog_idr_lock should have been held */
2634struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2635{
2636 int refold;
2637
2638 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2639
2640 if (!refold)
2641 return ERR_PTR(-ENOENT);
2642
2643 return prog;
2644}
2645EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2646
2647bool bpf_prog_get_ok(struct bpf_prog *prog,
2648 enum bpf_prog_type *attach_type, bool attach_drv)
2649{
2650 /* not an attachment, just a refcount inc, always allow */
2651 if (!attach_type)
2652 return true;
2653
2654 if (prog->type != *attach_type)
2655 return false;
2656 if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2657 return false;
2658
2659 return true;
2660}
2661
2662static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2663 bool attach_drv)
2664{
2665 CLASS(fd, f)(ufd);
2666 struct bpf_prog *prog;
2667
2668 if (fd_empty(f))
2669 return ERR_PTR(-EBADF);
2670 if (fd_file(f)->f_op != &bpf_prog_fops)
2671 return ERR_PTR(-EINVAL);
2672
2673 prog = fd_file(f)->private_data;
2674 if (!bpf_prog_get_ok(prog, attach_type, attach_drv))
2675 return ERR_PTR(-EINVAL);
2676
2677 bpf_prog_inc(prog);
2678 return prog;
2679}
2680
2681struct bpf_prog *bpf_prog_get(u32 ufd)
2682{
2683 return __bpf_prog_get(ufd, NULL, false);
2684}
2685
2686struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2687 bool attach_drv)
2688{
2689 return __bpf_prog_get(ufd, &type, attach_drv);
2690}
2691EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2692
2693/* Initially all BPF programs could be loaded w/o specifying
2694 * expected_attach_type. Later for some of them specifying expected_attach_type
2695 * at load time became required so that program could be validated properly.
2696 * Programs of types that are allowed to be loaded both w/ and w/o (for
2697 * backward compatibility) expected_attach_type, should have the default attach
2698 * type assigned to expected_attach_type for the latter case, so that it can be
2699 * validated later at attach time.
2700 *
2701 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2702 * prog type requires it but has some attach types that have to be backward
2703 * compatible.
2704 */
2705static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2706{
2707 switch (attr->prog_type) {
2708 case BPF_PROG_TYPE_CGROUP_SOCK:
2709 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2710 * exist so checking for non-zero is the way to go here.
2711 */
2712 if (!attr->expected_attach_type)
2713 attr->expected_attach_type =
2714 BPF_CGROUP_INET_SOCK_CREATE;
2715 break;
2716 case BPF_PROG_TYPE_SK_REUSEPORT:
2717 if (!attr->expected_attach_type)
2718 attr->expected_attach_type =
2719 BPF_SK_REUSEPORT_SELECT;
2720 break;
2721 }
2722}
2723
2724static int
2725bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2726 enum bpf_attach_type expected_attach_type,
2727 struct btf *attach_btf, u32 btf_id,
2728 struct bpf_prog *dst_prog,
2729 bool multi_func)
2730{
2731 if (btf_id) {
2732 if (btf_id > BTF_MAX_TYPE)
2733 return -EINVAL;
2734
2735 if (!attach_btf && !dst_prog)
2736 return -EINVAL;
2737
2738 switch (prog_type) {
2739 case BPF_PROG_TYPE_TRACING:
2740 case BPF_PROG_TYPE_LSM:
2741 case BPF_PROG_TYPE_STRUCT_OPS:
2742 case BPF_PROG_TYPE_EXT:
2743 break;
2744 default:
2745 return -EINVAL;
2746 }
2747 }
2748
2749 if (multi_func) {
2750 if (prog_type != BPF_PROG_TYPE_TRACING)
2751 return -EINVAL;
2752 if (!attach_btf || btf_id)
2753 return -EINVAL;
2754 return 0;
2755 }
2756
2757 if (attach_btf && (!btf_id || dst_prog))
2758 return -EINVAL;
2759
2760 if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2761 prog_type != BPF_PROG_TYPE_EXT)
2762 return -EINVAL;
2763
2764 switch (prog_type) {
2765 case BPF_PROG_TYPE_CGROUP_SOCK:
2766 switch (expected_attach_type) {
2767 case BPF_CGROUP_INET_SOCK_CREATE:
2768 case BPF_CGROUP_INET_SOCK_RELEASE:
2769 case BPF_CGROUP_INET4_POST_BIND:
2770 case BPF_CGROUP_INET6_POST_BIND:
2771 return 0;
2772 default:
2773 return -EINVAL;
2774 }
2775 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2776 switch (expected_attach_type) {
2777 case BPF_CGROUP_INET4_BIND:
2778 case BPF_CGROUP_INET6_BIND:
2779 case BPF_CGROUP_INET4_CONNECT:
2780 case BPF_CGROUP_INET6_CONNECT:
2781 case BPF_CGROUP_UNIX_CONNECT:
2782 case BPF_CGROUP_INET4_GETPEERNAME:
2783 case BPF_CGROUP_INET6_GETPEERNAME:
2784 case BPF_CGROUP_UNIX_GETPEERNAME:
2785 case BPF_CGROUP_INET4_GETSOCKNAME:
2786 case BPF_CGROUP_INET6_GETSOCKNAME:
2787 case BPF_CGROUP_UNIX_GETSOCKNAME:
2788 case BPF_CGROUP_UDP4_SENDMSG:
2789 case BPF_CGROUP_UDP6_SENDMSG:
2790 case BPF_CGROUP_UNIX_SENDMSG:
2791 case BPF_CGROUP_UDP4_RECVMSG:
2792 case BPF_CGROUP_UDP6_RECVMSG:
2793 case BPF_CGROUP_UNIX_RECVMSG:
2794 return 0;
2795 default:
2796 return -EINVAL;
2797 }
2798 case BPF_PROG_TYPE_CGROUP_SKB:
2799 switch (expected_attach_type) {
2800 case BPF_CGROUP_INET_INGRESS:
2801 case BPF_CGROUP_INET_EGRESS:
2802 return 0;
2803 default:
2804 return -EINVAL;
2805 }
2806 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2807 switch (expected_attach_type) {
2808 case BPF_CGROUP_SETSOCKOPT:
2809 case BPF_CGROUP_GETSOCKOPT:
2810 return 0;
2811 default:
2812 return -EINVAL;
2813 }
2814 case BPF_PROG_TYPE_SK_LOOKUP:
2815 if (expected_attach_type == BPF_SK_LOOKUP)
2816 return 0;
2817 return -EINVAL;
2818 case BPF_PROG_TYPE_SK_REUSEPORT:
2819 switch (expected_attach_type) {
2820 case BPF_SK_REUSEPORT_SELECT:
2821 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2822 return 0;
2823 default:
2824 return -EINVAL;
2825 }
2826 case BPF_PROG_TYPE_NETFILTER:
2827 if (expected_attach_type == BPF_NETFILTER)
2828 return 0;
2829 return -EINVAL;
2830 case BPF_PROG_TYPE_SYSCALL:
2831 case BPF_PROG_TYPE_EXT:
2832 if (expected_attach_type)
2833 return -EINVAL;
2834 fallthrough;
2835 default:
2836 return 0;
2837 }
2838}
2839
2840static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2841{
2842 switch (prog_type) {
2843 case BPF_PROG_TYPE_SCHED_CLS:
2844 case BPF_PROG_TYPE_SCHED_ACT:
2845 case BPF_PROG_TYPE_XDP:
2846 case BPF_PROG_TYPE_LWT_IN:
2847 case BPF_PROG_TYPE_LWT_OUT:
2848 case BPF_PROG_TYPE_LWT_XMIT:
2849 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2850 case BPF_PROG_TYPE_SK_SKB:
2851 case BPF_PROG_TYPE_SK_MSG:
2852 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2853 case BPF_PROG_TYPE_CGROUP_DEVICE:
2854 case BPF_PROG_TYPE_CGROUP_SOCK:
2855 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2856 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2857 case BPF_PROG_TYPE_CGROUP_SYSCTL:
2858 case BPF_PROG_TYPE_SOCK_OPS:
2859 case BPF_PROG_TYPE_EXT: /* extends any prog */
2860 case BPF_PROG_TYPE_NETFILTER:
2861 return true;
2862 case BPF_PROG_TYPE_CGROUP_SKB:
2863 /* always unpriv */
2864 case BPF_PROG_TYPE_SK_REUSEPORT:
2865 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2866 default:
2867 return false;
2868 }
2869}
2870
2871static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2872{
2873 switch (prog_type) {
2874 case BPF_PROG_TYPE_KPROBE:
2875 case BPF_PROG_TYPE_TRACEPOINT:
2876 case BPF_PROG_TYPE_PERF_EVENT:
2877 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2878 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2879 case BPF_PROG_TYPE_TRACING:
2880 case BPF_PROG_TYPE_LSM:
2881 case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2882 case BPF_PROG_TYPE_EXT: /* extends any prog */
2883 return true;
2884 default:
2885 return false;
2886 }
2887}
2888
2889static enum bpf_sig_keyring bpf_classify_keyring(s32 keyring_id)
2890{
2891 switch (keyring_id) {
2892 case 0:
2893 return BPF_SIG_KEYRING_BUILTIN;
2894 case (s32)(unsigned long)VERIFY_USE_SECONDARY_KEYRING:
2895 return BPF_SIG_KEYRING_SECONDARY;
2896 case (s32)(unsigned long)VERIFY_USE_PLATFORM_KEYRING:
2897 return BPF_SIG_KEYRING_PLATFORM;
2898 default:
2899 return BPF_SIG_KEYRING_USER;
2900 }
2901}
2902
2903static int bpf_prog_verify_signature(struct bpf_prog *prog, union bpf_attr *attr,
2904 bool is_kernel, s32 *keyring_serial)
2905{
2906 bpfptr_t usig = make_bpfptr(attr->signature, is_kernel);
2907 struct bpf_dynptr_kern sig_ptr, insns_ptr;
2908 struct bpf_key *key = NULL;
2909 void *sig;
2910 int err = 0;
2911
2912 /*
2913 * Don't attempt to use kmalloc_large or vmalloc for signatures.
2914 * Practical signature for BPF program should be below this limit.
2915 */
2916 if (attr->signature_size > KMALLOC_MAX_CACHE_SIZE)
2917 return -EINVAL;
2918
2919 if (system_keyring_id_check(attr->keyring_id) == 0)
2920 key = bpf_lookup_system_key(attr->keyring_id);
2921 else
2922 key = bpf_lookup_user_key(attr->keyring_id, 0);
2923
2924 if (!key)
2925 return -EINVAL;
2926
2927 sig = kvmemdup_bpfptr(usig, attr->signature_size);
2928 if (IS_ERR(sig)) {
2929 bpf_key_put(key);
2930 return PTR_ERR(sig);
2931 }
2932
2933 bpf_dynptr_init(&sig_ptr, sig, BPF_DYNPTR_TYPE_LOCAL, 0,
2934 attr->signature_size);
2935 bpf_dynptr_init(&insns_ptr, prog->insnsi, BPF_DYNPTR_TYPE_LOCAL, 0,
2936 prog->len * sizeof(struct bpf_insn));
2937
2938 err = bpf_verify_pkcs7_signature((struct bpf_dynptr *)&insns_ptr,
2939 (struct bpf_dynptr *)&sig_ptr, key);
2940 if (!err)
2941 *keyring_serial = bpf_key_serial(key);
2942 bpf_key_put(key);
2943 kvfree(sig);
2944 return err;
2945}
2946
2947static int bpf_prog_mark_insn_arrays_ready(struct bpf_prog *prog)
2948{
2949 int err;
2950 int i;
2951
2952 for (i = 0; i < prog->aux->used_map_cnt; i++) {
2953 if (prog->aux->used_maps[i]->map_type != BPF_MAP_TYPE_INSN_ARRAY)
2954 continue;
2955
2956 err = bpf_insn_array_ready(prog->aux->used_maps[i]);
2957 if (err)
2958 return err;
2959 }
2960
2961 return 0;
2962}
2963
2964extern int bpf_multi_func(void);
2965int __init __used bpf_multi_func(void) { return 0; }
2966
2967BTF_ID_LIST_GLOBAL_SINGLE(bpf_multi_func_btf_id, func, bpf_multi_func)
2968
2969/* last field in 'union bpf_attr' used by this command */
2970#define BPF_PROG_LOAD_LAST_FIELD keyring_id
2971
2972static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, struct bpf_log_attr *attr_log)
2973{
2974 enum bpf_prog_type type = attr->prog_type;
2975 struct bpf_prog *prog, *dst_prog = NULL;
2976 struct btf *attach_btf = NULL;
2977 struct bpf_token *token = NULL;
2978 bool bpf_cap;
2979 int err;
2980 char license[128];
2981 bool multi_func;
2982
2983 if (CHECK_ATTR(BPF_PROG_LOAD))
2984 return -EINVAL;
2985
2986 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2987 BPF_F_ANY_ALIGNMENT |
2988 BPF_F_TEST_STATE_FREQ |
2989 BPF_F_SLEEPABLE |
2990 BPF_F_TEST_RND_HI32 |
2991 BPF_F_XDP_HAS_FRAGS |
2992 BPF_F_XDP_DEV_BOUND_ONLY |
2993 BPF_F_TEST_REG_INVARIANTS |
2994 BPF_F_TOKEN_FD))
2995 return -EINVAL;
2996
2997 bpf_prog_load_fixup_attach_type(attr);
2998
2999 if (attr->prog_flags & BPF_F_TOKEN_FD) {
3000 token = bpf_token_get_from_fd(attr->prog_token_fd);
3001 if (IS_ERR(token))
3002 return PTR_ERR(token);
3003 /* if current token doesn't grant prog loading permissions,
3004 * then we can't use this token, so ignore it and rely on
3005 * system-wide capabilities checks
3006 */
3007 if (!bpf_token_allow_cmd(token, BPF_PROG_LOAD) ||
3008 !bpf_token_allow_prog_type(token, attr->prog_type,
3009 attr->expected_attach_type)) {
3010 bpf_token_put(token);
3011 token = NULL;
3012 }
3013 }
3014
3015 bpf_cap = bpf_token_capable(token, CAP_BPF);
3016 err = -EPERM;
3017
3018 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
3019 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
3020 !bpf_cap)
3021 goto put_token;
3022
3023 /* Intent here is for unprivileged_bpf_disabled to block BPF program
3024 * creation for unprivileged users; other actions depend
3025 * on fd availability and access to bpffs, so are dependent on
3026 * object creation success. Even with unprivileged BPF disabled,
3027 * capability checks are still carried out for these
3028 * and other operations.
3029 */
3030 if (sysctl_unprivileged_bpf_disabled && !bpf_cap)
3031 goto put_token;
3032
3033 if (attr->insn_cnt == 0 ||
3034 attr->insn_cnt > (bpf_cap ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS)) {
3035 err = -E2BIG;
3036 goto put_token;
3037 }
3038 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
3039 type != BPF_PROG_TYPE_CGROUP_SKB &&
3040 !bpf_cap)
3041 goto put_token;
3042
3043 if (is_net_admin_prog_type(type) && !bpf_token_capable(token, CAP_NET_ADMIN))
3044 goto put_token;
3045 if (is_perfmon_prog_type(type) && !bpf_token_capable(token, CAP_PERFMON))
3046 goto put_token;
3047
3048 multi_func = is_tracing_multi(attr->expected_attach_type);
3049
3050 /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
3051 * or btf, we need to check which one it is
3052 */
3053 if (attr->attach_prog_fd) {
3054 dst_prog = bpf_prog_get(attr->attach_prog_fd);
3055 if (IS_ERR(dst_prog)) {
3056 dst_prog = NULL;
3057 attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
3058 if (IS_ERR(attach_btf)) {
3059 err = -EINVAL;
3060 goto put_token;
3061 }
3062 if (!btf_is_kernel(attach_btf)) {
3063 /* attaching through specifying bpf_prog's BTF
3064 * objects directly might be supported eventually
3065 */
3066 btf_put(attach_btf);
3067 err = -ENOTSUPP;
3068 goto put_token;
3069 }
3070 }
3071 } else if (attr->attach_btf_id || multi_func) {
3072 /* fall back to vmlinux BTF, if BTF type ID is specified */
3073 attach_btf = bpf_get_btf_vmlinux();
3074 if (IS_ERR(attach_btf)) {
3075 err = PTR_ERR(attach_btf);
3076 goto put_token;
3077 }
3078 if (!attach_btf) {
3079 err = -EINVAL;
3080 goto put_token;
3081 }
3082 btf_get(attach_btf);
3083 }
3084
3085 if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
3086 attach_btf, attr->attach_btf_id,
3087 dst_prog, multi_func)) {
3088 if (dst_prog)
3089 bpf_prog_put(dst_prog);
3090 if (attach_btf)
3091 btf_put(attach_btf);
3092 err = -EINVAL;
3093 goto put_token;
3094 }
3095
3096 /* plain bpf_prog allocation */
3097 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
3098 if (!prog) {
3099 if (dst_prog)
3100 bpf_prog_put(dst_prog);
3101 if (attach_btf)
3102 btf_put(attach_btf);
3103 err = -EINVAL;
3104 goto put_token;
3105 }
3106
3107 prog->expected_attach_type = attr->expected_attach_type;
3108 prog->sleepable = !!(attr->prog_flags & BPF_F_SLEEPABLE);
3109 prog->aux->attach_btf = attach_btf;
3110 prog->aux->attach_btf_id = multi_func ? bpf_multi_func_btf_id[0] : attr->attach_btf_id;
3111 prog->aux->dst_prog = dst_prog;
3112 prog->aux->dev_bound = !!attr->prog_ifindex;
3113 prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
3114
3115 /* move token into prog->aux, reuse taken refcnt */
3116 prog->aux->token = token;
3117 token = NULL;
3118
3119 prog->aux->user = get_current_user();
3120 prog->len = attr->insn_cnt;
3121
3122 err = -EFAULT;
3123 if (copy_from_bpfptr(prog->insns,
3124 make_bpfptr(attr->insns, uattr.is_kernel),
3125 bpf_prog_insn_size(prog)) != 0)
3126 goto free_prog;
3127 /* copy eBPF program license from user space */
3128 if (strncpy_from_bpfptr(license,
3129 make_bpfptr(attr->license, uattr.is_kernel),
3130 sizeof(license) - 1) < 0)
3131 goto free_prog;
3132 license[sizeof(license) - 1] = 0;
3133
3134 /* eBPF programs must be GPL compatible to use GPL-ed functions */
3135 prog->gpl_compatible = license_is_gpl_compatible(license) ? 1 : 0;
3136 if (attr->signature) {
3137 err = bpf_prog_verify_signature(prog, attr, uattr.is_kernel,
3138 &prog->aux->sig.keyring_serial);
3139 if (err)
3140 goto free_prog;
3141 prog->aux->sig.keyring_type = bpf_classify_keyring(attr->keyring_id);
3142 prog->aux->sig.verdict = BPF_SIG_VERIFIED;
3143 } else {
3144 prog->aux->sig.keyring_type = BPF_SIG_KEYRING_NONE;
3145 prog->aux->sig.verdict = BPF_SIG_UNSIGNED;
3146 }
3147 prog->orig_prog = NULL;
3148 prog->jited = 0;
3149
3150 atomic64_set(&prog->aux->refcnt, 1);
3151
3152 if (bpf_prog_is_dev_bound(prog->aux)) {
3153 err = bpf_prog_dev_bound_init(prog, attr);
3154 if (err)
3155 goto free_prog;
3156 }
3157
3158 if (type == BPF_PROG_TYPE_EXT && dst_prog &&
3159 bpf_prog_is_dev_bound(dst_prog->aux)) {
3160 err = bpf_prog_dev_bound_inherit(prog, dst_prog);
3161 if (err)
3162 goto free_prog;
3163 }
3164
3165 /*
3166 * Bookkeeping for managing the program attachment chain.
3167 *
3168 * It might be tempting to set attach_tracing_prog flag at the attachment
3169 * time, but this will not prevent from loading bunch of tracing prog
3170 * first, then attach them one to another.
3171 *
3172 * The flag attach_tracing_prog is set for the whole program lifecycle, and
3173 * doesn't have to be cleared in bpf_tracing_link_release, since tracing
3174 * programs cannot change attachment target.
3175 */
3176 if (type == BPF_PROG_TYPE_TRACING && dst_prog &&
3177 dst_prog->type == BPF_PROG_TYPE_TRACING) {
3178 prog->aux->attach_tracing_prog = true;
3179 }
3180
3181 /* find program type: socket_filter vs tracing_filter */
3182 err = find_prog_type(type, prog);
3183 if (err < 0)
3184 goto free_prog;
3185
3186 prog->aux->load_time = ktime_get_boottime_ns();
3187 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
3188 sizeof(attr->prog_name));
3189 if (err < 0)
3190 goto free_prog;
3191
3192 err = security_bpf_prog_load(prog, attr, token, uattr.is_kernel);
3193 if (err)
3194 goto free_prog;
3195
3196 /* run eBPF verifier */
3197 err = bpf_check(&prog, attr, uattr, attr_log);
3198 if (err < 0)
3199 goto free_used_maps;
3200
3201 err = bpf_prog_mark_insn_arrays_ready(prog);
3202 if (err < 0)
3203 goto free_used_maps;
3204
3205 err = bpf_prog_alloc_id(prog);
3206 if (err)
3207 goto free_used_maps;
3208
3209 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
3210 * effectively publicly exposed. However, retrieving via
3211 * bpf_prog_get_fd_by_id() will take another reference,
3212 * therefore it cannot be gone underneath us.
3213 *
3214 * Only for the time /after/ successful bpf_prog_new_fd()
3215 * and before returning to userspace, we might just hold
3216 * one reference and any parallel close on that fd could
3217 * rip everything out. Hence, below notifications must
3218 * happen before bpf_prog_new_fd().
3219 *
3220 * Also, any failure handling from this point onwards must
3221 * be using bpf_prog_put() given the program is exposed.
3222 */
3223 bpf_prog_kallsyms_add(prog);
3224 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
3225 bpf_audit_prog(prog, BPF_AUDIT_LOAD);
3226
3227 err = bpf_prog_new_fd(prog);
3228 if (err < 0)
3229 bpf_prog_put(prog);
3230 return err;
3231
3232free_used_maps:
3233 /* In case we have subprogs, we need to wait for a grace
3234 * period before we can tear down JIT memory since symbols
3235 * are already exposed under kallsyms.
3236 */
3237 __bpf_prog_put_noref(prog, prog->aux->real_func_cnt);
3238 return err;
3239
3240free_prog:
3241 free_uid(prog->aux->user);
3242 if (prog->aux->attach_btf)
3243 btf_put(prog->aux->attach_btf);
3244 bpf_prog_free(prog);
3245put_token:
3246 bpf_token_put(token);
3247 return err;
3248}
3249
3250#define BPF_OBJ_LAST_FIELD path_fd
3251
3252static int bpf_obj_pin(const union bpf_attr *attr)
3253{
3254 int path_fd;
3255
3256 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags & ~BPF_F_PATH_FD)
3257 return -EINVAL;
3258
3259 /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
3260 if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
3261 return -EINVAL;
3262
3263 path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
3264 return bpf_obj_pin_user(attr->bpf_fd, path_fd,
3265 u64_to_user_ptr(attr->pathname));
3266}
3267
3268static int bpf_obj_get(const union bpf_attr *attr)
3269{
3270 int path_fd;
3271
3272 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
3273 attr->file_flags & ~(BPF_OBJ_FLAG_MASK | BPF_F_PATH_FD))
3274 return -EINVAL;
3275
3276 /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
3277 if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
3278 return -EINVAL;
3279
3280 path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
3281 return bpf_obj_get_user(path_fd, u64_to_user_ptr(attr->pathname),
3282 attr->file_flags);
3283}
3284
3285/* bpf_link_init_sleepable() allows to specify whether BPF link itself has
3286 * "sleepable" semantics, which normally would mean that BPF link's attach
3287 * hook can dereference link or link's underlying program for some time after
3288 * detachment due to RCU Tasks Trace-based lifetime protection scheme.
3289 * BPF program itself can be non-sleepable, yet, because it's transitively
3290 * reachable through BPF link, its freeing has to be delayed until after RCU
3291 * Tasks Trace GP.
3292 */
3293void bpf_link_init_sleepable(struct bpf_link *link, enum bpf_link_type type,
3294 const struct bpf_link_ops *ops, struct bpf_prog *prog,
3295 enum bpf_attach_type attach_type, bool sleepable)
3296{
3297 WARN_ON(ops->dealloc && ops->dealloc_deferred);
3298 atomic64_set(&link->refcnt, 1);
3299 link->type = type;
3300 link->sleepable = sleepable;
3301 link->id = 0;
3302 link->ops = ops;
3303 link->prog = prog;
3304 link->attach_type = attach_type;
3305}
3306
3307void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
3308 const struct bpf_link_ops *ops, struct bpf_prog *prog,
3309 enum bpf_attach_type attach_type)
3310{
3311 bpf_link_init_sleepable(link, type, ops, prog, attach_type, false);
3312}
3313
3314void bpf_tramp_link_init(struct bpf_tramp_link *link, enum bpf_link_type type,
3315 const struct bpf_link_ops *ops, struct bpf_prog *prog,
3316 enum bpf_attach_type attach_type, u64 cookie)
3317{
3318 bpf_link_init(&link->link, type, ops, prog, attach_type);
3319 link->node.link = &link->link;
3320 link->node.cookie = cookie;
3321}
3322
3323static void bpf_link_free_id(int id)
3324{
3325 if (!id)
3326 return;
3327
3328 spin_lock_bh(&link_idr_lock);
3329 idr_remove(&link_idr, id);
3330 spin_unlock_bh(&link_idr_lock);
3331}
3332
3333/* Clean up bpf_link and corresponding anon_inode file and FD. After
3334 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
3335 * anon_inode's release() call. This helper marks bpf_link as
3336 * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
3337 * is not decremented, it's the responsibility of a calling code that failed
3338 * to complete bpf_link initialization.
3339 * This helper eventually calls link's dealloc callback, but does not call
3340 * link's release callback.
3341 */
3342void bpf_link_cleanup(struct bpf_link_primer *primer)
3343{
3344 primer->link->prog = NULL;
3345 bpf_link_free_id(primer->id);
3346 fput(primer->file);
3347 put_unused_fd(primer->fd);
3348}
3349
3350void bpf_link_inc(struct bpf_link *link)
3351{
3352 atomic64_inc(&link->refcnt);
3353}
3354
3355static void bpf_link_dealloc(struct bpf_link *link)
3356{
3357 /* now that we know that bpf_link itself can't be reached, put underlying BPF program */
3358 if (link->prog)
3359 bpf_prog_put(link->prog);
3360
3361 /* free bpf_link and its containing memory */
3362 if (link->ops->dealloc_deferred)
3363 link->ops->dealloc_deferred(link);
3364 else
3365 link->ops->dealloc(link);
3366}
3367
3368static void bpf_link_defer_dealloc_rcu_gp(struct rcu_head *rcu)
3369{
3370 struct bpf_link *link = container_of(rcu, struct bpf_link, rcu);
3371
3372 bpf_link_dealloc(link);
3373}
3374
3375static bool bpf_link_is_tracepoint(struct bpf_link *link)
3376{
3377 /*
3378 * Only these combinations support a tracepoint bpf_link.
3379 * BPF_LINK_TYPE_TRACING raw_tp progs are hardcoded to use
3380 * bpf_raw_tp_link_lops and thus dealloc_deferred(), see
3381 * bpf_raw_tp_link_attach().
3382 */
3383 return link->type == BPF_LINK_TYPE_RAW_TRACEPOINT ||
3384 (link->type == BPF_LINK_TYPE_TRACING && link->attach_type == BPF_TRACE_RAW_TP);
3385}
3386
3387/* bpf_link_free is guaranteed to be called from process context */
3388static void bpf_link_free(struct bpf_link *link)
3389{
3390 const struct bpf_link_ops *ops = link->ops;
3391
3392 bpf_link_free_id(link->id);
3393 /* detach BPF program, clean up used resources */
3394 if (link->prog)
3395 ops->release(link);
3396 if (ops->dealloc_deferred) {
3397 /*
3398 * Schedule BPF link deallocation, which will only then
3399 * trigger putting BPF program refcount.
3400 * If underlying BPF program is sleepable or BPF link's target
3401 * attach hookpoint is sleepable or otherwise requires RCU GPs
3402 * to ensure link and its underlying BPF program is not
3403 * reachable anymore, we need to first wait for RCU tasks
3404 * trace sync, and then go through "classic" RCU grace period.
3405 *
3406 * For tracepoint BPF links, we need to go through SRCU grace
3407 * period wait instead when non-faultable tracepoint is used. We
3408 * don't need to chain SRCU grace period waits, however, for the
3409 * faultable case, since it exclusively uses RCU Tasks Trace.
3410 */
3411 if (link->sleepable || (link->prog && link->prog->sleepable))
3412 /* RCU Tasks Trace grace period implies RCU grace period. */
3413 call_rcu_tasks_trace(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
3414 /* We need to do a SRCU grace period wait for non-faultable tracepoint BPF links. */
3415 else if (bpf_link_is_tracepoint(link))
3416 call_tracepoint_unregister_atomic(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
3417 else
3418 call_rcu(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
3419 } else if (ops->dealloc) {
3420 bpf_link_dealloc(link);
3421 }
3422}
3423
3424static void bpf_link_put_deferred(struct work_struct *work)
3425{
3426 struct bpf_link *link = container_of(work, struct bpf_link, work);
3427
3428 bpf_link_free(link);
3429}
3430
3431/* bpf_link_put might be called from atomic context. It needs to be called
3432 * from sleepable context in order to acquire sleeping locks during the process.
3433 */
3434void bpf_link_put(struct bpf_link *link)
3435{
3436 if (!atomic64_dec_and_test(&link->refcnt))
3437 return;
3438
3439 INIT_WORK(&link->work, bpf_link_put_deferred);
3440 schedule_work(&link->work);
3441}
3442EXPORT_SYMBOL(bpf_link_put);
3443
3444static void bpf_link_put_direct(struct bpf_link *link)
3445{
3446 if (!atomic64_dec_and_test(&link->refcnt))
3447 return;
3448 bpf_link_free(link);
3449}
3450
3451static int bpf_link_release(struct inode *inode, struct file *filp)
3452{
3453 struct bpf_link *link = filp->private_data;
3454
3455 bpf_link_put_direct(link);
3456 return 0;
3457}
3458
3459#ifdef CONFIG_PROC_FS
3460#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
3461#define BPF_MAP_TYPE(_id, _ops)
3462#define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
3463static const char *bpf_link_type_strs[] = {
3464 [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
3465#include <linux/bpf_types.h>
3466};
3467#undef BPF_PROG_TYPE
3468#undef BPF_MAP_TYPE
3469#undef BPF_LINK_TYPE
3470
3471static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
3472{
3473 const struct bpf_link *link = filp->private_data;
3474 const struct bpf_prog *prog = link->prog;
3475 enum bpf_link_type type = link->type;
3476 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
3477
3478 if (type < ARRAY_SIZE(bpf_link_type_strs) && bpf_link_type_strs[type]) {
3479 if (link->type == BPF_LINK_TYPE_KPROBE_MULTI)
3480 seq_printf(m, "link_type:\t%s\n", link->flags == BPF_F_KPROBE_MULTI_RETURN ?
3481 "kretprobe_multi" : "kprobe_multi");
3482 else if (link->type == BPF_LINK_TYPE_UPROBE_MULTI)
3483 seq_printf(m, "link_type:\t%s\n", link->flags & BPF_F_UPROBE_MULTI_RETURN ?
3484 "uretprobe_multi" : "uprobe_multi");
3485 else
3486 seq_printf(m, "link_type:\t%s\n", bpf_link_type_strs[type]);
3487 } else {
3488 WARN_ONCE(1, "missing BPF_LINK_TYPE(...) for link type %u\n", type);
3489 seq_printf(m, "link_type:\t<%u>\n", type);
3490 }
3491 seq_printf(m, "link_id:\t%u\n", link->id);
3492
3493 if (prog) {
3494 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
3495 seq_printf(m,
3496 "prog_tag:\t%s\n"
3497 "prog_id:\t%u\n",
3498 prog_tag,
3499 prog->aux->id);
3500 }
3501 if (link->ops->show_fdinfo)
3502 link->ops->show_fdinfo(link, m);
3503}
3504#endif
3505
3506static __poll_t bpf_link_poll(struct file *file, struct poll_table_struct *pts)
3507{
3508 struct bpf_link *link = file->private_data;
3509
3510 return link->ops->poll(file, pts);
3511}
3512
3513static const struct file_operations bpf_link_fops = {
3514#ifdef CONFIG_PROC_FS
3515 .show_fdinfo = bpf_link_show_fdinfo,
3516#endif
3517 .release = bpf_link_release,
3518 .read = bpf_dummy_read,
3519 .write = bpf_dummy_write,
3520};
3521
3522static const struct file_operations bpf_link_fops_poll = {
3523#ifdef CONFIG_PROC_FS
3524 .show_fdinfo = bpf_link_show_fdinfo,
3525#endif
3526 .release = bpf_link_release,
3527 .read = bpf_dummy_read,
3528 .write = bpf_dummy_write,
3529 .poll = bpf_link_poll,
3530};
3531
3532static int bpf_link_alloc_id(struct bpf_link *link)
3533{
3534 int id;
3535
3536 idr_preload(GFP_KERNEL);
3537 spin_lock_bh(&link_idr_lock);
3538 id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
3539 spin_unlock_bh(&link_idr_lock);
3540 idr_preload_end();
3541
3542 return id;
3543}
3544
3545/* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
3546 * reserving unused FD and allocating ID from link_idr. This is to be paired
3547 * with bpf_link_settle() to install FD and ID and expose bpf_link to
3548 * user-space, if bpf_link is successfully attached. If not, bpf_link and
3549 * pre-allocated resources are to be freed with bpf_cleanup() call. All the
3550 * transient state is passed around in struct bpf_link_primer.
3551 * This is preferred way to create and initialize bpf_link, especially when
3552 * there are complicated and expensive operations in between creating bpf_link
3553 * itself and attaching it to BPF hook. By using bpf_link_prime() and
3554 * bpf_link_settle() kernel code using bpf_link doesn't have to perform
3555 * expensive (and potentially failing) roll back operations in a rare case
3556 * that file, FD, or ID can't be allocated.
3557 */
3558int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
3559{
3560 struct file *file;
3561 int fd, id;
3562
3563 fd = get_unused_fd_flags(O_CLOEXEC);
3564 if (fd < 0)
3565 return fd;
3566
3567
3568 id = bpf_link_alloc_id(link);
3569 if (id < 0) {
3570 put_unused_fd(fd);
3571 return id;
3572 }
3573
3574 file = anon_inode_getfile("bpf_link",
3575 link->ops->poll ? &bpf_link_fops_poll : &bpf_link_fops,
3576 link, O_CLOEXEC);
3577 if (IS_ERR(file)) {
3578 bpf_link_free_id(id);
3579 put_unused_fd(fd);
3580 return PTR_ERR(file);
3581 }
3582
3583 primer->link = link;
3584 primer->file = file;
3585 primer->fd = fd;
3586 primer->id = id;
3587 return 0;
3588}
3589
3590int bpf_link_settle(struct bpf_link_primer *primer)
3591{
3592 /* make bpf_link fetchable by ID */
3593 spin_lock_bh(&link_idr_lock);
3594 primer->link->id = primer->id;
3595 spin_unlock_bh(&link_idr_lock);
3596 /* make bpf_link fetchable by FD */
3597 fd_install(primer->fd, primer->file);
3598 /* pass through installed FD */
3599 return primer->fd;
3600}
3601
3602int bpf_link_new_fd(struct bpf_link *link)
3603{
3604 return anon_inode_getfd("bpf-link",
3605 link->ops->poll ? &bpf_link_fops_poll : &bpf_link_fops,
3606 link, O_CLOEXEC);
3607}
3608
3609struct bpf_link *bpf_link_get_from_fd(u32 ufd)
3610{
3611 CLASS(fd, f)(ufd);
3612 struct bpf_link *link;
3613
3614 if (fd_empty(f))
3615 return ERR_PTR(-EBADF);
3616 if (fd_file(f)->f_op != &bpf_link_fops && fd_file(f)->f_op != &bpf_link_fops_poll)
3617 return ERR_PTR(-EINVAL);
3618
3619 link = fd_file(f)->private_data;
3620 bpf_link_inc(link);
3621 return link;
3622}
3623EXPORT_SYMBOL_NS(bpf_link_get_from_fd, "BPF_INTERNAL");
3624
3625static void bpf_tracing_link_release(struct bpf_link *link)
3626{
3627 struct bpf_tracing_link *tr_link =
3628 container_of(link, struct bpf_tracing_link, link.link);
3629
3630 WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link.node,
3631 tr_link->trampoline,
3632 tr_link->tgt_prog));
3633
3634 bpf_trampoline_put(tr_link->trampoline);
3635
3636 /* tgt_prog is NULL if target is a kernel function */
3637 if (tr_link->tgt_prog)
3638 bpf_prog_put(tr_link->tgt_prog);
3639}
3640
3641static void bpf_tracing_link_dealloc(struct bpf_link *link)
3642{
3643 struct bpf_tracing_link *tr_link = container_of(link, struct bpf_tracing_link, link.link);
3644
3645 kfree(tr_link);
3646}
3647
3648static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
3649 struct seq_file *seq)
3650{
3651 struct bpf_tracing_link *tr_link = container_of(link, struct bpf_tracing_link, link.link);
3652
3653 u32 target_btf_id, target_obj_id;
3654
3655 bpf_trampoline_unpack_key(tr_link->trampoline->key,
3656 &target_obj_id, &target_btf_id);
3657 seq_printf(seq,
3658 "attach_type:\t%d\n"
3659 "target_obj_id:\t%u\n"
3660 "target_btf_id:\t%u\n"
3661 "cookie:\t%llu\n",
3662 link->attach_type,
3663 target_obj_id,
3664 target_btf_id,
3665 tr_link->link.node.cookie);
3666}
3667
3668static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
3669 struct bpf_link_info *info)
3670{
3671 struct bpf_tracing_link *tr_link = container_of(link, struct bpf_tracing_link, link.link);
3672
3673 info->tracing.attach_type = link->attach_type;
3674 info->tracing.cookie = tr_link->link.node.cookie;
3675 bpf_trampoline_unpack_key(tr_link->trampoline->key,
3676 &info->tracing.target_obj_id,
3677 &info->tracing.target_btf_id);
3678
3679 return 0;
3680}
3681
3682static const struct bpf_link_ops bpf_tracing_link_lops = {
3683 .release = bpf_tracing_link_release,
3684 .dealloc = bpf_tracing_link_dealloc,
3685 .show_fdinfo = bpf_tracing_link_show_fdinfo,
3686 .fill_link_info = bpf_tracing_link_fill_link_info,
3687};
3688
3689static int bpf_tracing_prog_attach(struct bpf_prog *prog,
3690 int tgt_prog_fd,
3691 u32 btf_id,
3692 u64 bpf_cookie,
3693 enum bpf_attach_type attach_type)
3694{
3695 struct bpf_link_primer link_primer;
3696 struct bpf_prog *tgt_prog = NULL;
3697 struct bpf_trampoline *tr = NULL;
3698 struct bpf_tracing_link *link;
3699 u64 key = 0;
3700 int err;
3701
3702 switch (prog->type) {
3703 case BPF_PROG_TYPE_TRACING:
3704 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
3705 prog->expected_attach_type != BPF_TRACE_FEXIT &&
3706 prog->expected_attach_type != BPF_TRACE_FSESSION &&
3707 prog->expected_attach_type != BPF_MODIFY_RETURN) {
3708 err = -EINVAL;
3709 goto out_put_prog;
3710 }
3711 break;
3712 case BPF_PROG_TYPE_EXT:
3713 if (prog->expected_attach_type != 0) {
3714 err = -EINVAL;
3715 goto out_put_prog;
3716 }
3717 break;
3718 case BPF_PROG_TYPE_LSM:
3719 if (prog->expected_attach_type != BPF_LSM_MAC) {
3720 err = -EINVAL;
3721 goto out_put_prog;
3722 }
3723 break;
3724 default:
3725 err = -EINVAL;
3726 goto out_put_prog;
3727 }
3728
3729 if (!!tgt_prog_fd != !!btf_id) {
3730 err = -EINVAL;
3731 goto out_put_prog;
3732 }
3733
3734 if (tgt_prog_fd) {
3735 /*
3736 * For now we only allow new targets for BPF_PROG_TYPE_EXT. If this
3737 * part would be changed to implement the same for
3738 * BPF_PROG_TYPE_TRACING, do not forget to update the way how
3739 * attach_tracing_prog flag is set.
3740 */
3741 if (prog->type != BPF_PROG_TYPE_EXT) {
3742 err = -EINVAL;
3743 goto out_put_prog;
3744 }
3745
3746 tgt_prog = bpf_prog_get(tgt_prog_fd);
3747 if (IS_ERR(tgt_prog)) {
3748 err = PTR_ERR(tgt_prog);
3749 tgt_prog = NULL;
3750 goto out_put_prog;
3751 }
3752
3753 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3754 }
3755
3756 link = kzalloc_obj(*link, GFP_USER);
3757 if (!link) {
3758 err = -ENOMEM;
3759 goto out_put_prog;
3760 }
3761 bpf_tramp_link_init(&link->link, BPF_LINK_TYPE_TRACING,
3762 &bpf_tracing_link_lops, prog, attach_type, bpf_cookie);
3763
3764 if (prog->expected_attach_type == BPF_TRACE_FSESSION) {
3765 link->fexit.link = &link->link.link;
3766 link->fexit.cookie = bpf_cookie;
3767 }
3768
3769 mutex_lock(&prog->aux->dst_mutex);
3770
3771 /* There are a few possible cases here:
3772 *
3773 * - if prog->aux->dst_trampoline is set, the program was just loaded
3774 * and not yet attached to anything, so we can use the values stored
3775 * in prog->aux
3776 *
3777 * - if prog->aux->dst_trampoline is NULL, the program has already been
3778 * attached to a target and its initial target was cleared (below)
3779 *
3780 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3781 * target_btf_id using the link_create API.
3782 *
3783 * - if tgt_prog == NULL when this function was called using the old
3784 * raw_tracepoint_open API, and we need a target from prog->aux
3785 *
3786 * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3787 * was detached and is going for re-attachment.
3788 *
3789 * - if prog->aux->dst_trampoline is NULL and tgt_prog and prog->aux->attach_btf
3790 * are NULL, then program was already attached and user did not provide
3791 * tgt_prog_fd so we have no way to find out or create trampoline
3792 */
3793 if (!prog->aux->dst_trampoline && !tgt_prog) {
3794 /*
3795 * Allow re-attach for TRACING and LSM programs. If it's
3796 * currently linked, bpf_trampoline_link_prog will fail.
3797 * EXT programs need to specify tgt_prog_fd, so they
3798 * re-attach in separate code path.
3799 */
3800 if (prog->type != BPF_PROG_TYPE_TRACING &&
3801 prog->type != BPF_PROG_TYPE_LSM) {
3802 err = -EINVAL;
3803 goto out_unlock;
3804 }
3805 /* We can allow re-attach only if we have valid attach_btf. */
3806 if (!prog->aux->attach_btf) {
3807 err = -EINVAL;
3808 goto out_unlock;
3809 }
3810 btf_id = prog->aux->attach_btf_id;
3811 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3812 }
3813
3814 if (!prog->aux->dst_trampoline ||
3815 (key && key != prog->aux->dst_trampoline->key)) {
3816 /* If there is no saved target, or the specified target is
3817 * different from the destination specified at load time, we
3818 * need a new trampoline and a check for compatibility
3819 */
3820 struct bpf_attach_target_info tgt_info = {};
3821
3822 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3823 &tgt_info);
3824 if (err)
3825 goto out_unlock;
3826
3827 if (tgt_info.tgt_mod) {
3828 module_put(prog->aux->mod);
3829 prog->aux->mod = tgt_info.tgt_mod;
3830 }
3831
3832 tr = bpf_trampoline_get(key, &tgt_info);
3833 if (!tr) {
3834 err = -ENOMEM;
3835 goto out_unlock;
3836 }
3837 } else {
3838 /* The caller didn't specify a target, or the target was the
3839 * same as the destination supplied during program load. This
3840 * means we can reuse the trampoline and reference from program
3841 * load time, and there is no need to allocate a new one. This
3842 * can only happen once for any program, as the saved values in
3843 * prog->aux are cleared below.
3844 */
3845 tr = prog->aux->dst_trampoline;
3846 tgt_prog = prog->aux->dst_prog;
3847 }
3848 /*
3849 * It is to prevent modifying struct pt_regs via kprobe_write_ctx=true
3850 * freplace prog. Without this check, kprobe_write_ctx=true freplace
3851 * prog is allowed to attach to kprobe_write_ctx=false kprobe prog, and
3852 * then modify the registers of the kprobe prog's target kernel
3853 * function.
3854 *
3855 * This also blocks the combination of uprobe+freplace, because it is
3856 * unable to recognize the use of the tgt_prog as an uprobe or a kprobe
3857 * by tgt_prog itself. At attach time, uprobe/kprobe is recognized by
3858 * the target perf event flags in __perf_event_set_bpf_prog().
3859 */
3860 if (prog->type == BPF_PROG_TYPE_EXT &&
3861 prog->aux->kprobe_write_ctx != tgt_prog->aux->kprobe_write_ctx) {
3862 err = -EINVAL;
3863 goto out_unlock;
3864 }
3865
3866 err = bpf_link_prime(&link->link.link, &link_primer);
3867 if (err)
3868 goto out_unlock;
3869
3870 err = bpf_trampoline_link_prog(&link->link.node, tr, tgt_prog);
3871 if (err) {
3872 bpf_link_cleanup(&link_primer);
3873 link = NULL;
3874 goto out_unlock;
3875 }
3876
3877 link->tgt_prog = tgt_prog;
3878 link->trampoline = tr;
3879
3880 /* Always clear the trampoline and target prog from prog->aux to make
3881 * sure the original attach destination is not kept alive after a
3882 * program is (re-)attached to another target.
3883 */
3884 if (prog->aux->dst_prog &&
3885 (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3886 /* got extra prog ref from syscall, or attaching to different prog */
3887 bpf_prog_put(prog->aux->dst_prog);
3888 if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3889 /* we allocated a new trampoline, so free the old one */
3890 bpf_trampoline_put(prog->aux->dst_trampoline);
3891
3892 prog->aux->dst_prog = NULL;
3893 prog->aux->dst_trampoline = NULL;
3894 mutex_unlock(&prog->aux->dst_mutex);
3895
3896 return bpf_link_settle(&link_primer);
3897out_unlock:
3898 if (tr && tr != prog->aux->dst_trampoline)
3899 bpf_trampoline_put(tr);
3900 mutex_unlock(&prog->aux->dst_mutex);
3901 kfree(link);
3902out_put_prog:
3903 if (tgt_prog_fd && tgt_prog)
3904 bpf_prog_put(tgt_prog);
3905 return err;
3906}
3907
3908static void bpf_raw_tp_link_release(struct bpf_link *link)
3909{
3910 struct bpf_raw_tp_link *raw_tp =
3911 container_of(link, struct bpf_raw_tp_link, link);
3912
3913 bpf_probe_unregister(raw_tp->btp, raw_tp);
3914 bpf_put_raw_tracepoint(raw_tp->btp);
3915}
3916
3917static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3918{
3919 struct bpf_raw_tp_link *raw_tp =
3920 container_of(link, struct bpf_raw_tp_link, link);
3921
3922 kfree(raw_tp);
3923}
3924
3925static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3926 struct seq_file *seq)
3927{
3928 struct bpf_raw_tp_link *raw_tp_link =
3929 container_of(link, struct bpf_raw_tp_link, link);
3930
3931 seq_printf(seq,
3932 "tp_name:\t%s\n"
3933 "cookie:\t%llu\n",
3934 raw_tp_link->btp->tp->name,
3935 raw_tp_link->cookie);
3936}
3937
3938static int bpf_copy_to_user(char __user *ubuf, const char *buf, u32 ulen,
3939 u32 len)
3940{
3941 if (ulen >= len + 1) {
3942 if (copy_to_user(ubuf, buf, len + 1))
3943 return -EFAULT;
3944 } else {
3945 char zero = '\0';
3946
3947 if (copy_to_user(ubuf, buf, ulen - 1))
3948 return -EFAULT;
3949 if (put_user(zero, ubuf + ulen - 1))
3950 return -EFAULT;
3951 return -ENOSPC;
3952 }
3953
3954 return 0;
3955}
3956
3957static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3958 struct bpf_link_info *info)
3959{
3960 struct bpf_raw_tp_link *raw_tp_link =
3961 container_of(link, struct bpf_raw_tp_link, link);
3962 char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3963 const char *tp_name = raw_tp_link->btp->tp->name;
3964 u32 ulen = info->raw_tracepoint.tp_name_len;
3965 size_t tp_len = strlen(tp_name);
3966
3967 if (!ulen ^ !ubuf)
3968 return -EINVAL;
3969
3970 info->raw_tracepoint.tp_name_len = tp_len + 1;
3971 info->raw_tracepoint.cookie = raw_tp_link->cookie;
3972
3973 if (!ubuf)
3974 return 0;
3975
3976 return bpf_copy_to_user(ubuf, tp_name, ulen, tp_len);
3977}
3978
3979static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3980 .release = bpf_raw_tp_link_release,
3981 .dealloc_deferred = bpf_raw_tp_link_dealloc,
3982 .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3983 .fill_link_info = bpf_raw_tp_link_fill_link_info,
3984};
3985
3986#ifdef CONFIG_PERF_EVENTS
3987struct bpf_perf_link {
3988 struct bpf_link link;
3989 struct file *perf_file;
3990};
3991
3992static void bpf_perf_link_release(struct bpf_link *link)
3993{
3994 struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3995 struct perf_event *event = perf_link->perf_file->private_data;
3996
3997 perf_event_free_bpf_prog(event);
3998 fput(perf_link->perf_file);
3999}
4000
4001static void bpf_perf_link_dealloc(struct bpf_link *link)
4002{
4003 struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
4004
4005 kfree(perf_link);
4006}
4007
4008static int bpf_perf_link_fill_common(const struct perf_event *event,
4009 char __user *uname, u32 *ulenp,
4010 u64 *probe_offset, u64 *probe_addr,
4011 u32 *fd_type, unsigned long *missed)
4012{
4013 const char *buf;
4014 u32 prog_id, ulen;
4015 size_t len;
4016 int err;
4017
4018 ulen = *ulenp;
4019 if (!ulen ^ !uname)
4020 return -EINVAL;
4021
4022 err = bpf_get_perf_event_info(event, &prog_id, fd_type, &buf,
4023 probe_offset, probe_addr, missed);
4024 if (err)
4025 return err;
4026
4027 if (buf) {
4028 len = strlen(buf);
4029 *ulenp = len + 1;
4030 } else {
4031 *ulenp = 1;
4032 }
4033 if (!uname)
4034 return 0;
4035
4036 if (buf) {
4037 err = bpf_copy_to_user(uname, buf, ulen, len);
4038 if (err)
4039 return err;
4040 } else {
4041 char zero = '\0';
4042
4043 if (put_user(zero, uname))
4044 return -EFAULT;
4045 }
4046 return 0;
4047}
4048
4049#ifdef CONFIG_KPROBE_EVENTS
4050static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
4051 struct bpf_link_info *info)
4052{
4053 unsigned long missed;
4054 char __user *uname;
4055 u64 addr, offset;
4056 u32 ulen, type;
4057 int err;
4058
4059 uname = u64_to_user_ptr(info->perf_event.kprobe.func_name);
4060 ulen = info->perf_event.kprobe.name_len;
4061 err = bpf_perf_link_fill_common(event, uname, &ulen, &offset, &addr,
4062 &type, &missed);
4063 if (err)
4064 return err;
4065 if (type == BPF_FD_TYPE_KRETPROBE)
4066 info->perf_event.type = BPF_PERF_EVENT_KRETPROBE;
4067 else
4068 info->perf_event.type = BPF_PERF_EVENT_KPROBE;
4069 info->perf_event.kprobe.name_len = ulen;
4070 info->perf_event.kprobe.offset = offset;
4071 info->perf_event.kprobe.missed = missed;
4072 if (!kallsyms_show_value(current_cred()))
4073 addr = 0;
4074 info->perf_event.kprobe.addr = addr;
4075 info->perf_event.kprobe.cookie = event->bpf_cookie;
4076 return 0;
4077}
4078
4079static void bpf_perf_link_fdinfo_kprobe(const struct perf_event *event,
4080 struct seq_file *seq)
4081{
4082 const char *name;
4083 int err;
4084 u32 prog_id, type;
4085 u64 offset, addr;
4086 unsigned long missed;
4087
4088 err = bpf_get_perf_event_info(event, &prog_id, &type, &name,
4089 &offset, &addr, &missed);
4090 if (err)
4091 return;
4092
4093 seq_printf(seq,
4094 "name:\t%s\n"
4095 "offset:\t%#llx\n"
4096 "missed:\t%lu\n"
4097 "addr:\t%#llx\n"
4098 "event_type:\t%s\n"
4099 "cookie:\t%llu\n",
4100 name, offset, missed, addr,
4101 type == BPF_FD_TYPE_KRETPROBE ? "kretprobe" : "kprobe",
4102 event->bpf_cookie);
4103}
4104#endif
4105
4106#ifdef CONFIG_UPROBE_EVENTS
4107static int bpf_perf_link_fill_uprobe(const struct perf_event *event,
4108 struct bpf_link_info *info)
4109{
4110 u64 ref_ctr_offset, offset;
4111 char __user *uname;
4112 u32 ulen, type;
4113 int err;
4114
4115 uname = u64_to_user_ptr(info->perf_event.uprobe.file_name);
4116 ulen = info->perf_event.uprobe.name_len;
4117 err = bpf_perf_link_fill_common(event, uname, &ulen, &offset, &ref_ctr_offset,
4118 &type, NULL);
4119 if (err)
4120 return err;
4121
4122 if (type == BPF_FD_TYPE_URETPROBE)
4123 info->perf_event.type = BPF_PERF_EVENT_URETPROBE;
4124 else
4125 info->perf_event.type = BPF_PERF_EVENT_UPROBE;
4126 info->perf_event.uprobe.name_len = ulen;
4127 info->perf_event.uprobe.offset = offset;
4128 info->perf_event.uprobe.cookie = event->bpf_cookie;
4129 info->perf_event.uprobe.ref_ctr_offset = ref_ctr_offset;
4130 return 0;
4131}
4132
4133static void bpf_perf_link_fdinfo_uprobe(const struct perf_event *event,
4134 struct seq_file *seq)
4135{
4136 const char *name;
4137 int err;
4138 u32 prog_id, type;
4139 u64 offset, ref_ctr_offset;
4140 unsigned long missed;
4141
4142 err = bpf_get_perf_event_info(event, &prog_id, &type, &name,
4143 &offset, &ref_ctr_offset, &missed);
4144 if (err)
4145 return;
4146
4147 seq_printf(seq,
4148 "name:\t%s\n"
4149 "offset:\t%#llx\n"
4150 "ref_ctr_offset:\t%#llx\n"
4151 "event_type:\t%s\n"
4152 "cookie:\t%llu\n",
4153 name, offset, ref_ctr_offset,
4154 type == BPF_FD_TYPE_URETPROBE ? "uretprobe" : "uprobe",
4155 event->bpf_cookie);
4156}
4157#endif
4158
4159static int bpf_perf_link_fill_probe(const struct perf_event *event,
4160 struct bpf_link_info *info)
4161{
4162#ifdef CONFIG_KPROBE_EVENTS
4163 if (event->tp_event->flags & TRACE_EVENT_FL_KPROBE)
4164 return bpf_perf_link_fill_kprobe(event, info);
4165#endif
4166#ifdef CONFIG_UPROBE_EVENTS
4167 if (event->tp_event->flags & TRACE_EVENT_FL_UPROBE)
4168 return bpf_perf_link_fill_uprobe(event, info);
4169#endif
4170 return -EOPNOTSUPP;
4171}
4172
4173static int bpf_perf_link_fill_tracepoint(const struct perf_event *event,
4174 struct bpf_link_info *info)
4175{
4176 char __user *uname;
4177 u32 ulen;
4178 int err;
4179
4180 uname = u64_to_user_ptr(info->perf_event.tracepoint.tp_name);
4181 ulen = info->perf_event.tracepoint.name_len;
4182 err = bpf_perf_link_fill_common(event, uname, &ulen, NULL, NULL, NULL, NULL);
4183 if (err)
4184 return err;
4185
4186 info->perf_event.type = BPF_PERF_EVENT_TRACEPOINT;
4187 info->perf_event.tracepoint.name_len = ulen;
4188 info->perf_event.tracepoint.cookie = event->bpf_cookie;
4189 return 0;
4190}
4191
4192static int bpf_perf_link_fill_perf_event(const struct perf_event *event,
4193 struct bpf_link_info *info)
4194{
4195 info->perf_event.event.type = event->attr.type;
4196 info->perf_event.event.config = event->attr.config;
4197 info->perf_event.event.cookie = event->bpf_cookie;
4198 info->perf_event.type = BPF_PERF_EVENT_EVENT;
4199 return 0;
4200}
4201
4202static int bpf_perf_link_fill_link_info(const struct bpf_link *link,
4203 struct bpf_link_info *info)
4204{
4205 struct bpf_perf_link *perf_link;
4206 const struct perf_event *event;
4207
4208 perf_link = container_of(link, struct bpf_perf_link, link);
4209 event = perf_get_event(perf_link->perf_file);
4210 if (IS_ERR(event))
4211 return PTR_ERR(event);
4212
4213 switch (event->prog->type) {
4214 case BPF_PROG_TYPE_PERF_EVENT:
4215 return bpf_perf_link_fill_perf_event(event, info);
4216 case BPF_PROG_TYPE_TRACEPOINT:
4217 return bpf_perf_link_fill_tracepoint(event, info);
4218 case BPF_PROG_TYPE_KPROBE:
4219 return bpf_perf_link_fill_probe(event, info);
4220 default:
4221 return -EOPNOTSUPP;
4222 }
4223}
4224
4225static void bpf_perf_event_link_show_fdinfo(const struct perf_event *event,
4226 struct seq_file *seq)
4227{
4228 seq_printf(seq,
4229 "type:\t%u\n"
4230 "config:\t%llu\n"
4231 "event_type:\t%s\n"
4232 "cookie:\t%llu\n",
4233 event->attr.type, event->attr.config,
4234 "event", event->bpf_cookie);
4235}
4236
4237static void bpf_tracepoint_link_show_fdinfo(const struct perf_event *event,
4238 struct seq_file *seq)
4239{
4240 int err;
4241 const char *name;
4242 u32 prog_id;
4243
4244 err = bpf_get_perf_event_info(event, &prog_id, NULL, &name, NULL,
4245 NULL, NULL);
4246 if (err)
4247 return;
4248
4249 seq_printf(seq,
4250 "tp_name:\t%s\n"
4251 "event_type:\t%s\n"
4252 "cookie:\t%llu\n",
4253 name, "tracepoint", event->bpf_cookie);
4254}
4255
4256static void bpf_probe_link_show_fdinfo(const struct perf_event *event,
4257 struct seq_file *seq)
4258{
4259#ifdef CONFIG_KPROBE_EVENTS
4260 if (event->tp_event->flags & TRACE_EVENT_FL_KPROBE)
4261 return bpf_perf_link_fdinfo_kprobe(event, seq);
4262#endif
4263
4264#ifdef CONFIG_UPROBE_EVENTS
4265 if (event->tp_event->flags & TRACE_EVENT_FL_UPROBE)
4266 return bpf_perf_link_fdinfo_uprobe(event, seq);
4267#endif
4268}
4269
4270static void bpf_perf_link_show_fdinfo(const struct bpf_link *link,
4271 struct seq_file *seq)
4272{
4273 struct bpf_perf_link *perf_link;
4274 const struct perf_event *event;
4275
4276 perf_link = container_of(link, struct bpf_perf_link, link);
4277 event = perf_get_event(perf_link->perf_file);
4278 if (IS_ERR(event))
4279 return;
4280
4281 switch (event->prog->type) {
4282 case BPF_PROG_TYPE_PERF_EVENT:
4283 return bpf_perf_event_link_show_fdinfo(event, seq);
4284 case BPF_PROG_TYPE_TRACEPOINT:
4285 return bpf_tracepoint_link_show_fdinfo(event, seq);
4286 case BPF_PROG_TYPE_KPROBE:
4287 return bpf_probe_link_show_fdinfo(event, seq);
4288 default:
4289 return;
4290 }
4291}
4292
4293static const struct bpf_link_ops bpf_perf_link_lops = {
4294 .release = bpf_perf_link_release,
4295 .dealloc = bpf_perf_link_dealloc,
4296 .fill_link_info = bpf_perf_link_fill_link_info,
4297 .show_fdinfo = bpf_perf_link_show_fdinfo,
4298};
4299
4300static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
4301{
4302 struct bpf_link_primer link_primer;
4303 struct bpf_perf_link *link;
4304 struct perf_event *event;
4305 struct file *perf_file;
4306 int err;
4307
4308 if (attr->link_create.flags)
4309 return -EINVAL;
4310
4311 perf_file = perf_event_get(attr->link_create.target_fd);
4312 if (IS_ERR(perf_file))
4313 return PTR_ERR(perf_file);
4314
4315 link = kzalloc_obj(*link, GFP_USER);
4316 if (!link) {
4317 err = -ENOMEM;
4318 goto out_put_file;
4319 }
4320 bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog,
4321 attr->link_create.attach_type);
4322 link->perf_file = perf_file;
4323
4324 err = bpf_link_prime(&link->link, &link_primer);
4325 if (err) {
4326 kfree(link);
4327 goto out_put_file;
4328 }
4329
4330 event = perf_file->private_data;
4331 err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
4332 if (err) {
4333 bpf_link_cleanup(&link_primer);
4334 goto out_put_file;
4335 }
4336 /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
4337 bpf_prog_inc(prog);
4338
4339 return bpf_link_settle(&link_primer);
4340
4341out_put_file:
4342 fput(perf_file);
4343 return err;
4344}
4345#else
4346static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
4347{
4348 return -EOPNOTSUPP;
4349}
4350#endif /* CONFIG_PERF_EVENTS */
4351
4352static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
4353 const char __user *user_tp_name, u64 cookie,
4354 enum bpf_attach_type attach_type)
4355{
4356 struct bpf_link_primer link_primer;
4357 struct bpf_raw_tp_link *link;
4358 struct bpf_raw_event_map *btp;
4359 const char *tp_name;
4360 char buf[128];
4361 int err;
4362
4363 switch (prog->type) {
4364 case BPF_PROG_TYPE_TRACING:
4365 case BPF_PROG_TYPE_EXT:
4366 case BPF_PROG_TYPE_LSM:
4367 if (user_tp_name)
4368 /* The attach point for this category of programs
4369 * should be specified via btf_id during program load.
4370 */
4371 return -EINVAL;
4372 if (prog->type == BPF_PROG_TYPE_TRACING &&
4373 prog->expected_attach_type == BPF_TRACE_RAW_TP) {
4374 tp_name = prog->aux->attach_func_name;
4375 break;
4376 }
4377 return bpf_tracing_prog_attach(prog, 0, 0, 0, attach_type);
4378 case BPF_PROG_TYPE_RAW_TRACEPOINT:
4379 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
4380 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
4381 return -EFAULT;
4382 buf[sizeof(buf) - 1] = 0;
4383 tp_name = buf;
4384 break;
4385 default:
4386 return -EINVAL;
4387 }
4388
4389 btp = bpf_get_raw_tracepoint(tp_name);
4390 if (!btp)
4391 return -ENOENT;
4392
4393 if (prog->sleepable && !tracepoint_is_faultable(btp->tp)) {
4394 bpf_put_raw_tracepoint(btp);
4395 return -EINVAL;
4396 }
4397
4398 link = kzalloc_obj(*link, GFP_USER);
4399 if (!link) {
4400 err = -ENOMEM;
4401 goto out_put_btp;
4402 }
4403 bpf_link_init_sleepable(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
4404 &bpf_raw_tp_link_lops, prog, attach_type,
4405 tracepoint_is_faultable(btp->tp));
4406 link->btp = btp;
4407 link->cookie = cookie;
4408
4409 err = bpf_link_prime(&link->link, &link_primer);
4410 if (err) {
4411 kfree(link);
4412 goto out_put_btp;
4413 }
4414
4415 err = bpf_probe_register(link->btp, link);
4416 if (err) {
4417 bpf_link_cleanup(&link_primer);
4418 goto out_put_btp;
4419 }
4420
4421 return bpf_link_settle(&link_primer);
4422
4423out_put_btp:
4424 bpf_put_raw_tracepoint(btp);
4425 return err;
4426}
4427
4428#define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.cookie
4429
4430static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
4431{
4432 struct bpf_prog *prog;
4433 void __user *tp_name;
4434 __u64 cookie;
4435 int fd;
4436
4437 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
4438 return -EINVAL;
4439
4440 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
4441 if (IS_ERR(prog))
4442 return PTR_ERR(prog);
4443
4444 tp_name = u64_to_user_ptr(attr->raw_tracepoint.name);
4445 cookie = attr->raw_tracepoint.cookie;
4446 fd = bpf_raw_tp_link_attach(prog, tp_name, cookie, prog->expected_attach_type);
4447 if (fd < 0)
4448 bpf_prog_put(prog);
4449 return fd;
4450}
4451
4452static enum bpf_prog_type
4453attach_type_to_prog_type(enum bpf_attach_type attach_type)
4454{
4455 switch (attach_type) {
4456 case BPF_CGROUP_INET_INGRESS:
4457 case BPF_CGROUP_INET_EGRESS:
4458 return BPF_PROG_TYPE_CGROUP_SKB;
4459 case BPF_CGROUP_INET_SOCK_CREATE:
4460 case BPF_CGROUP_INET_SOCK_RELEASE:
4461 case BPF_CGROUP_INET4_POST_BIND:
4462 case BPF_CGROUP_INET6_POST_BIND:
4463 return BPF_PROG_TYPE_CGROUP_SOCK;
4464 case BPF_CGROUP_INET4_BIND:
4465 case BPF_CGROUP_INET6_BIND:
4466 case BPF_CGROUP_INET4_CONNECT:
4467 case BPF_CGROUP_INET6_CONNECT:
4468 case BPF_CGROUP_UNIX_CONNECT:
4469 case BPF_CGROUP_INET4_GETPEERNAME:
4470 case BPF_CGROUP_INET6_GETPEERNAME:
4471 case BPF_CGROUP_UNIX_GETPEERNAME:
4472 case BPF_CGROUP_INET4_GETSOCKNAME:
4473 case BPF_CGROUP_INET6_GETSOCKNAME:
4474 case BPF_CGROUP_UNIX_GETSOCKNAME:
4475 case BPF_CGROUP_UDP4_SENDMSG:
4476 case BPF_CGROUP_UDP6_SENDMSG:
4477 case BPF_CGROUP_UNIX_SENDMSG:
4478 case BPF_CGROUP_UDP4_RECVMSG:
4479 case BPF_CGROUP_UDP6_RECVMSG:
4480 case BPF_CGROUP_UNIX_RECVMSG:
4481 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
4482 case BPF_CGROUP_SOCK_OPS:
4483 return BPF_PROG_TYPE_SOCK_OPS;
4484 case BPF_CGROUP_DEVICE:
4485 return BPF_PROG_TYPE_CGROUP_DEVICE;
4486 case BPF_SK_MSG_VERDICT:
4487 return BPF_PROG_TYPE_SK_MSG;
4488 case BPF_SK_SKB_STREAM_PARSER:
4489 case BPF_SK_SKB_STREAM_VERDICT:
4490 case BPF_SK_SKB_VERDICT:
4491 return BPF_PROG_TYPE_SK_SKB;
4492 case BPF_LIRC_MODE2:
4493 return BPF_PROG_TYPE_LIRC_MODE2;
4494 case BPF_FLOW_DISSECTOR:
4495 return BPF_PROG_TYPE_FLOW_DISSECTOR;
4496 case BPF_CGROUP_SYSCTL:
4497 return BPF_PROG_TYPE_CGROUP_SYSCTL;
4498 case BPF_CGROUP_GETSOCKOPT:
4499 case BPF_CGROUP_SETSOCKOPT:
4500 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
4501 case BPF_TRACE_ITER:
4502 case BPF_TRACE_RAW_TP:
4503 case BPF_TRACE_FENTRY:
4504 case BPF_TRACE_FEXIT:
4505 case BPF_TRACE_FSESSION:
4506 case BPF_TRACE_FSESSION_MULTI:
4507 case BPF_TRACE_FENTRY_MULTI:
4508 case BPF_TRACE_FEXIT_MULTI:
4509 case BPF_MODIFY_RETURN:
4510 return BPF_PROG_TYPE_TRACING;
4511 case BPF_LSM_MAC:
4512 return BPF_PROG_TYPE_LSM;
4513 case BPF_SK_LOOKUP:
4514 return BPF_PROG_TYPE_SK_LOOKUP;
4515 case BPF_XDP:
4516 return BPF_PROG_TYPE_XDP;
4517 case BPF_LSM_CGROUP:
4518 return BPF_PROG_TYPE_LSM;
4519 case BPF_TCX_INGRESS:
4520 case BPF_TCX_EGRESS:
4521 case BPF_NETKIT_PRIMARY:
4522 case BPF_NETKIT_PEER:
4523 return BPF_PROG_TYPE_SCHED_CLS;
4524 default:
4525 return BPF_PROG_TYPE_UNSPEC;
4526 }
4527}
4528
4529static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
4530 enum bpf_attach_type attach_type)
4531{
4532 enum bpf_prog_type ptype;
4533
4534 switch (prog->type) {
4535 case BPF_PROG_TYPE_CGROUP_SOCK:
4536 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4537 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4538 case BPF_PROG_TYPE_SK_LOOKUP:
4539 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
4540 case BPF_PROG_TYPE_CGROUP_SKB:
4541 if (!bpf_token_capable(prog->aux->token, CAP_NET_ADMIN))
4542 /* cg-skb progs can be loaded by unpriv user.
4543 * check permissions at attach time.
4544 */
4545 return -EPERM;
4546
4547 ptype = attach_type_to_prog_type(attach_type);
4548 if (prog->type != ptype)
4549 return -EINVAL;
4550
4551 return prog->enforce_expected_attach_type &&
4552 prog->expected_attach_type != attach_type ?
4553 -EINVAL : 0;
4554 case BPF_PROG_TYPE_EXT:
4555 return 0;
4556 case BPF_PROG_TYPE_NETFILTER:
4557 if (attach_type != BPF_NETFILTER)
4558 return -EINVAL;
4559 return 0;
4560 case BPF_PROG_TYPE_PERF_EVENT:
4561 case BPF_PROG_TYPE_TRACEPOINT:
4562 if (attach_type != BPF_PERF_EVENT)
4563 return -EINVAL;
4564 return 0;
4565 case BPF_PROG_TYPE_KPROBE:
4566 if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI &&
4567 attach_type != BPF_TRACE_KPROBE_MULTI)
4568 return -EINVAL;
4569 if (prog->expected_attach_type == BPF_TRACE_KPROBE_SESSION &&
4570 attach_type != BPF_TRACE_KPROBE_SESSION)
4571 return -EINVAL;
4572 if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI &&
4573 attach_type != BPF_TRACE_UPROBE_MULTI)
4574 return -EINVAL;
4575 if (prog->expected_attach_type == BPF_TRACE_UPROBE_SESSION &&
4576 attach_type != BPF_TRACE_UPROBE_SESSION)
4577 return -EINVAL;
4578 if (attach_type != BPF_PERF_EVENT &&
4579 attach_type != BPF_TRACE_KPROBE_MULTI &&
4580 attach_type != BPF_TRACE_KPROBE_SESSION &&
4581 attach_type != BPF_TRACE_UPROBE_MULTI &&
4582 attach_type != BPF_TRACE_UPROBE_SESSION)
4583 return -EINVAL;
4584 return 0;
4585 case BPF_PROG_TYPE_SCHED_CLS:
4586 if (attach_type != BPF_TCX_INGRESS &&
4587 attach_type != BPF_TCX_EGRESS &&
4588 attach_type != BPF_NETKIT_PRIMARY &&
4589 attach_type != BPF_NETKIT_PEER)
4590 return -EINVAL;
4591 return 0;
4592 default:
4593 ptype = attach_type_to_prog_type(attach_type);
4594 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type)
4595 return -EINVAL;
4596 return 0;
4597 }
4598}
4599
4600static bool is_cgroup_prog_type(enum bpf_prog_type ptype, enum bpf_attach_type atype,
4601 bool check_atype)
4602{
4603 switch (ptype) {
4604 case BPF_PROG_TYPE_CGROUP_DEVICE:
4605 case BPF_PROG_TYPE_CGROUP_SKB:
4606 case BPF_PROG_TYPE_CGROUP_SOCK:
4607 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4608 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4609 case BPF_PROG_TYPE_CGROUP_SYSCTL:
4610 case BPF_PROG_TYPE_SOCK_OPS:
4611 return true;
4612 case BPF_PROG_TYPE_LSM:
4613 return check_atype ? atype == BPF_LSM_CGROUP : true;
4614 default:
4615 return false;
4616 }
4617}
4618
4619#define BPF_PROG_ATTACH_LAST_FIELD expected_revision
4620
4621#define BPF_F_ATTACH_MASK_BASE \
4622 (BPF_F_ALLOW_OVERRIDE | \
4623 BPF_F_ALLOW_MULTI | \
4624 BPF_F_REPLACE | \
4625 BPF_F_PREORDER)
4626
4627#define BPF_F_ATTACH_MASK_MPROG \
4628 (BPF_F_REPLACE | \
4629 BPF_F_BEFORE | \
4630 BPF_F_AFTER | \
4631 BPF_F_ID | \
4632 BPF_F_LINK)
4633
4634static int bpf_prog_attach(const union bpf_attr *attr)
4635{
4636 enum bpf_prog_type ptype;
4637 struct bpf_prog *prog;
4638 int ret;
4639
4640 if (CHECK_ATTR(BPF_PROG_ATTACH))
4641 return -EINVAL;
4642
4643 ptype = attach_type_to_prog_type(attr->attach_type);
4644 if (ptype == BPF_PROG_TYPE_UNSPEC)
4645 return -EINVAL;
4646 if (bpf_mprog_supported(ptype)) {
4647 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
4648 return -EINVAL;
4649 } else if (is_cgroup_prog_type(ptype, 0, false)) {
4650 if (attr->attach_flags & ~(BPF_F_ATTACH_MASK_BASE | BPF_F_ATTACH_MASK_MPROG))
4651 return -EINVAL;
4652 } else {
4653 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_BASE)
4654 return -EINVAL;
4655 if (attr->relative_fd ||
4656 attr->expected_revision)
4657 return -EINVAL;
4658 }
4659
4660 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
4661 if (IS_ERR(prog))
4662 return PTR_ERR(prog);
4663
4664 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
4665 bpf_prog_put(prog);
4666 return -EINVAL;
4667 }
4668
4669 if (is_cgroup_prog_type(ptype, prog->expected_attach_type, true)) {
4670 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
4671 goto out;
4672 }
4673
4674 switch (ptype) {
4675 case BPF_PROG_TYPE_SK_SKB:
4676 case BPF_PROG_TYPE_SK_MSG:
4677 ret = sock_map_get_from_fd(attr, prog);
4678 break;
4679 case BPF_PROG_TYPE_LIRC_MODE2:
4680 ret = lirc_prog_attach(attr, prog);
4681 break;
4682 case BPF_PROG_TYPE_FLOW_DISSECTOR:
4683 ret = netns_bpf_prog_attach(attr, prog);
4684 break;
4685 case BPF_PROG_TYPE_SCHED_CLS:
4686 if (attr->attach_type == BPF_TCX_INGRESS ||
4687 attr->attach_type == BPF_TCX_EGRESS)
4688 ret = tcx_prog_attach(attr, prog);
4689 else
4690 ret = netkit_prog_attach(attr, prog);
4691 break;
4692 default:
4693 ret = -EINVAL;
4694 }
4695out:
4696 if (ret)
4697 bpf_prog_put(prog);
4698 return ret;
4699}
4700
4701#define BPF_PROG_DETACH_LAST_FIELD expected_revision
4702
4703static int bpf_prog_detach(const union bpf_attr *attr)
4704{
4705 struct bpf_prog *prog = NULL;
4706 enum bpf_prog_type ptype;
4707 int ret;
4708
4709 if (CHECK_ATTR(BPF_PROG_DETACH))
4710 return -EINVAL;
4711
4712 ptype = attach_type_to_prog_type(attr->attach_type);
4713 if (bpf_mprog_supported(ptype)) {
4714 if (ptype == BPF_PROG_TYPE_UNSPEC)
4715 return -EINVAL;
4716 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
4717 return -EINVAL;
4718 if (attr->attach_bpf_fd) {
4719 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
4720 if (IS_ERR(prog))
4721 return PTR_ERR(prog);
4722 } else if (!bpf_mprog_detach_empty(ptype)) {
4723 return -EPERM;
4724 }
4725 } else if (is_cgroup_prog_type(ptype, 0, false)) {
4726 if (attr->attach_flags || attr->relative_fd)
4727 return -EINVAL;
4728 } else if (attr->attach_flags ||
4729 attr->relative_fd ||
4730 attr->expected_revision) {
4731 return -EINVAL;
4732 }
4733
4734 switch (ptype) {
4735 case BPF_PROG_TYPE_SK_MSG:
4736 case BPF_PROG_TYPE_SK_SKB:
4737 ret = sock_map_prog_detach(attr, ptype);
4738 break;
4739 case BPF_PROG_TYPE_LIRC_MODE2:
4740 ret = lirc_prog_detach(attr);
4741 break;
4742 case BPF_PROG_TYPE_FLOW_DISSECTOR:
4743 ret = netns_bpf_prog_detach(attr, ptype);
4744 break;
4745 case BPF_PROG_TYPE_CGROUP_DEVICE:
4746 case BPF_PROG_TYPE_CGROUP_SKB:
4747 case BPF_PROG_TYPE_CGROUP_SOCK:
4748 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4749 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4750 case BPF_PROG_TYPE_CGROUP_SYSCTL:
4751 case BPF_PROG_TYPE_SOCK_OPS:
4752 case BPF_PROG_TYPE_LSM:
4753 ret = cgroup_bpf_prog_detach(attr, ptype);
4754 break;
4755 case BPF_PROG_TYPE_SCHED_CLS:
4756 if (attr->attach_type == BPF_TCX_INGRESS ||
4757 attr->attach_type == BPF_TCX_EGRESS)
4758 ret = tcx_prog_detach(attr, prog);
4759 else
4760 ret = netkit_prog_detach(attr, prog);
4761 break;
4762 default:
4763 ret = -EINVAL;
4764 }
4765
4766 if (prog)
4767 bpf_prog_put(prog);
4768 return ret;
4769}
4770
4771#define BPF_PROG_QUERY_LAST_FIELD query.revision
4772
4773static int bpf_prog_query(const union bpf_attr *attr,
4774 union bpf_attr __user *uattr, u32 uattr_size)
4775{
4776 if (!bpf_net_capable())
4777 return -EPERM;
4778 if (CHECK_ATTR(BPF_PROG_QUERY))
4779 return -EINVAL;
4780 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
4781 return -EINVAL;
4782
4783 switch (attr->query.attach_type) {
4784 case BPF_CGROUP_INET_INGRESS:
4785 case BPF_CGROUP_INET_EGRESS:
4786 case BPF_CGROUP_INET_SOCK_CREATE:
4787 case BPF_CGROUP_INET_SOCK_RELEASE:
4788 case BPF_CGROUP_INET4_BIND:
4789 case BPF_CGROUP_INET6_BIND:
4790 case BPF_CGROUP_INET4_POST_BIND:
4791 case BPF_CGROUP_INET6_POST_BIND:
4792 case BPF_CGROUP_INET4_CONNECT:
4793 case BPF_CGROUP_INET6_CONNECT:
4794 case BPF_CGROUP_UNIX_CONNECT:
4795 case BPF_CGROUP_INET4_GETPEERNAME:
4796 case BPF_CGROUP_INET6_GETPEERNAME:
4797 case BPF_CGROUP_UNIX_GETPEERNAME:
4798 case BPF_CGROUP_INET4_GETSOCKNAME:
4799 case BPF_CGROUP_INET6_GETSOCKNAME:
4800 case BPF_CGROUP_UNIX_GETSOCKNAME:
4801 case BPF_CGROUP_UDP4_SENDMSG:
4802 case BPF_CGROUP_UDP6_SENDMSG:
4803 case BPF_CGROUP_UNIX_SENDMSG:
4804 case BPF_CGROUP_UDP4_RECVMSG:
4805 case BPF_CGROUP_UDP6_RECVMSG:
4806 case BPF_CGROUP_UNIX_RECVMSG:
4807 case BPF_CGROUP_SOCK_OPS:
4808 case BPF_CGROUP_DEVICE:
4809 case BPF_CGROUP_SYSCTL:
4810 case BPF_CGROUP_GETSOCKOPT:
4811 case BPF_CGROUP_SETSOCKOPT:
4812 case BPF_LSM_CGROUP:
4813 return cgroup_bpf_prog_query(attr, uattr, uattr_size);
4814 case BPF_LIRC_MODE2:
4815 return lirc_prog_query(attr, uattr);
4816 case BPF_FLOW_DISSECTOR:
4817 case BPF_SK_LOOKUP:
4818 return netns_bpf_prog_query(attr, uattr);
4819 case BPF_SK_SKB_STREAM_PARSER:
4820 case BPF_SK_SKB_STREAM_VERDICT:
4821 case BPF_SK_MSG_VERDICT:
4822 case BPF_SK_SKB_VERDICT:
4823 return sock_map_bpf_prog_query(attr, uattr);
4824 case BPF_TCX_INGRESS:
4825 case BPF_TCX_EGRESS:
4826 return tcx_prog_query(attr, uattr);
4827 case BPF_NETKIT_PRIMARY:
4828 case BPF_NETKIT_PEER:
4829 return netkit_prog_query(attr, uattr);
4830 default:
4831 return -EINVAL;
4832 }
4833}
4834
4835#define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
4836
4837static int bpf_prog_test_run(const union bpf_attr *attr,
4838 union bpf_attr __user *uattr)
4839{
4840 struct bpf_prog *prog;
4841 int ret = -ENOTSUPP;
4842
4843 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
4844 return -EINVAL;
4845
4846 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
4847 (!attr->test.ctx_size_in && attr->test.ctx_in))
4848 return -EINVAL;
4849
4850 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
4851 (!attr->test.ctx_size_out && attr->test.ctx_out))
4852 return -EINVAL;
4853
4854 prog = bpf_prog_get(attr->test.prog_fd);
4855 if (IS_ERR(prog))
4856 return PTR_ERR(prog);
4857
4858 if (prog->aux->ops->test_run)
4859 ret = prog->aux->ops->test_run(prog, attr, uattr);
4860
4861 bpf_prog_put(prog);
4862 return ret;
4863}
4864
4865#define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
4866
4867static int bpf_obj_get_next_id(const union bpf_attr *attr,
4868 union bpf_attr __user *uattr,
4869 struct idr *idr,
4870 spinlock_t *lock)
4871{
4872 u32 next_id = attr->start_id;
4873 int err = 0;
4874
4875 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
4876 return -EINVAL;
4877
4878 if (!capable(CAP_SYS_ADMIN))
4879 return -EPERM;
4880
4881 next_id++;
4882 spin_lock_bh(lock);
4883 if (!idr_get_next(idr, &next_id))
4884 err = -ENOENT;
4885 spin_unlock_bh(lock);
4886
4887 if (!err)
4888 err = put_user(next_id, &uattr->next_id);
4889
4890 return err;
4891}
4892
4893struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
4894{
4895 struct bpf_map *map;
4896
4897 spin_lock_bh(&map_idr_lock);
4898again:
4899 map = idr_get_next(&map_idr, id);
4900 if (map) {
4901 map = __bpf_map_inc_not_zero(map, false);
4902 if (IS_ERR(map)) {
4903 (*id)++;
4904 goto again;
4905 }
4906 }
4907 spin_unlock_bh(&map_idr_lock);
4908
4909 return map;
4910}
4911
4912struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
4913{
4914 struct bpf_prog *prog;
4915
4916 spin_lock_bh(&prog_idr_lock);
4917again:
4918 prog = idr_get_next(&prog_idr, id);
4919 if (prog) {
4920 prog = bpf_prog_inc_not_zero(prog);
4921 if (IS_ERR(prog)) {
4922 (*id)++;
4923 goto again;
4924 }
4925 }
4926 spin_unlock_bh(&prog_idr_lock);
4927
4928 return prog;
4929}
4930
4931#define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
4932
4933struct bpf_prog *bpf_prog_by_id(u32 id)
4934{
4935 struct bpf_prog *prog;
4936
4937 if (!id)
4938 return ERR_PTR(-ENOENT);
4939
4940 spin_lock_bh(&prog_idr_lock);
4941 prog = idr_find(&prog_idr, id);
4942 if (prog)
4943 prog = bpf_prog_inc_not_zero(prog);
4944 else
4945 prog = ERR_PTR(-ENOENT);
4946 spin_unlock_bh(&prog_idr_lock);
4947 return prog;
4948}
4949
4950static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
4951{
4952 struct bpf_prog *prog;
4953 u32 id = attr->prog_id;
4954 int fd;
4955
4956 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
4957 return -EINVAL;
4958
4959 if (!capable(CAP_SYS_ADMIN))
4960 return -EPERM;
4961
4962 prog = bpf_prog_by_id(id);
4963 if (IS_ERR(prog))
4964 return PTR_ERR(prog);
4965
4966 fd = bpf_prog_new_fd(prog);
4967 if (fd < 0)
4968 bpf_prog_put(prog);
4969
4970 return fd;
4971}
4972
4973#define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
4974
4975static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
4976{
4977 struct bpf_map *map;
4978 u32 id = attr->map_id;
4979 int f_flags;
4980 int fd;
4981
4982 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
4983 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
4984 return -EINVAL;
4985
4986 if (!capable(CAP_SYS_ADMIN))
4987 return -EPERM;
4988
4989 f_flags = bpf_get_file_flag(attr->open_flags);
4990 if (f_flags < 0)
4991 return f_flags;
4992
4993 spin_lock_bh(&map_idr_lock);
4994 map = idr_find(&map_idr, id);
4995 if (map)
4996 map = __bpf_map_inc_not_zero(map, true);
4997 else
4998 map = ERR_PTR(-ENOENT);
4999 spin_unlock_bh(&map_idr_lock);
5000
5001 if (IS_ERR(map))
5002 return PTR_ERR(map);
5003
5004 fd = bpf_map_new_fd(map, f_flags);
5005 if (fd < 0)
5006 bpf_map_put_with_uref(map);
5007
5008 return fd;
5009}
5010
5011static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
5012 unsigned long addr, u32 *off,
5013 u32 *type)
5014{
5015 const struct bpf_map *map;
5016 int i;
5017
5018 mutex_lock(&prog->aux->used_maps_mutex);
5019 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
5020 map = prog->aux->used_maps[i];
5021 if (map == (void *)addr) {
5022 *type = BPF_PSEUDO_MAP_FD;
5023 goto out;
5024 }
5025 if (!map->ops->map_direct_value_meta)
5026 continue;
5027 if (!map->ops->map_direct_value_meta(map, addr, off)) {
5028 *type = BPF_PSEUDO_MAP_VALUE;
5029 goto out;
5030 }
5031 }
5032 map = NULL;
5033
5034out:
5035 mutex_unlock(&prog->aux->used_maps_mutex);
5036 return map;
5037}
5038
5039static void prepare_dump_pseudo_call(struct bpf_insn *insn)
5040{
5041 s32 call_off = insn->imm;
5042
5043 /*
5044 * BPF_CALL_ARGS only exists for interpreter fallback.
5045 * 1. For interpreter (BPF_CALL_ARGS): insn->off is the index of
5046 * interpreters_args array, so here using bpf_call_args_imm()
5047 * to get the real address offset.
5048 * 2. For JIT (BPF_CALL): insn->off is the subprog id.
5049 */
5050 if (insn->code == (BPF_JMP | BPF_CALL_ARGS))
5051 insn->imm = bpf_call_args_imm(insn->off);
5052 else
5053 insn->imm = insn->off;
5054
5055 /* Avoid dumping a truncated and misleading pc-relative offset. */
5056 if (call_off > S16_MAX || call_off < S16_MIN)
5057 insn->off = 0;
5058 else
5059 insn->off = call_off;
5060}
5061
5062static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
5063 const struct cred *f_cred)
5064{
5065 const struct bpf_map *map;
5066 struct bpf_insn *insns;
5067 u32 off, type;
5068 u64 imm;
5069 u8 code;
5070 int i;
5071
5072 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
5073 GFP_USER);
5074 if (!insns)
5075 return insns;
5076
5077 for (i = 0; i < prog->len; i++) {
5078 code = insns[i].code;
5079
5080 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
5081 insns[i].code = BPF_JMP | BPF_CALL;
5082 insns[i].imm = BPF_FUNC_tail_call;
5083 /* fall-through */
5084 }
5085 if (code == (BPF_JMP | BPF_CALL) ||
5086 code == (BPF_JMP | BPF_CALL_ARGS)) {
5087 /* Restore the legacy xlated dump layout. */
5088 if (insns[i].src_reg == BPF_PSEUDO_CALL)
5089 prepare_dump_pseudo_call(&insns[i]);
5090 if (code == (BPF_JMP | BPF_CALL_ARGS))
5091 insns[i].code = BPF_JMP | BPF_CALL;
5092 if (!bpf_dump_raw_ok(f_cred))
5093 insns[i].imm = 0;
5094 continue;
5095 }
5096 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
5097 insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
5098 continue;
5099 }
5100
5101 if ((BPF_CLASS(code) == BPF_LDX || BPF_CLASS(code) == BPF_STX ||
5102 BPF_CLASS(code) == BPF_ST) && BPF_MODE(code) == BPF_PROBE_MEM32) {
5103 insns[i].code = BPF_CLASS(code) | BPF_SIZE(code) | BPF_MEM;
5104 continue;
5105 }
5106
5107 if (code != (BPF_LD | BPF_IMM | BPF_DW))
5108 continue;
5109
5110 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
5111 map = bpf_map_from_imm(prog, imm, &off, &type);
5112 if (map) {
5113 insns[i].src_reg = type;
5114 insns[i].imm = map->id;
5115 insns[i + 1].imm = off;
5116 continue;
5117 }
5118 }
5119
5120 return insns;
5121}
5122
5123static int set_info_rec_size(struct bpf_prog_info *info)
5124{
5125 /*
5126 * Ensure info.*_rec_size is the same as kernel expected size
5127 *
5128 * or
5129 *
5130 * Only allow zero *_rec_size if both _rec_size and _cnt are
5131 * zero. In this case, the kernel will set the expected
5132 * _rec_size back to the info.
5133 */
5134
5135 if ((info->nr_func_info || info->func_info_rec_size) &&
5136 info->func_info_rec_size != sizeof(struct bpf_func_info))
5137 return -EINVAL;
5138
5139 if ((info->nr_line_info || info->line_info_rec_size) &&
5140 info->line_info_rec_size != sizeof(struct bpf_line_info))
5141 return -EINVAL;
5142
5143 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
5144 info->jited_line_info_rec_size != sizeof(__u64))
5145 return -EINVAL;
5146
5147 info->func_info_rec_size = sizeof(struct bpf_func_info);
5148 info->line_info_rec_size = sizeof(struct bpf_line_info);
5149 info->jited_line_info_rec_size = sizeof(__u64);
5150
5151 return 0;
5152}
5153
5154static int bpf_prog_get_info_by_fd(struct file *file,
5155 struct bpf_prog *prog,
5156 const union bpf_attr *attr,
5157 union bpf_attr __user *uattr)
5158{
5159 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
5160 struct btf *attach_btf = bpf_prog_get_target_btf(prog);
5161 struct bpf_prog_info info;
5162 u32 info_len = attr->info.info_len;
5163 struct bpf_prog_kstats stats;
5164 char __user *uinsns;
5165 u32 ulen, len;
5166 int err;
5167
5168 len = offsetofend(struct bpf_prog_info, attach_btf_id);
5169 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), len, info_len);
5170 if (err)
5171 return err;
5172 info_len = min_t(u32, sizeof(info), info_len);
5173
5174 memset(&info, 0, sizeof(info));
5175 if (copy_from_user(&info, uinfo, info_len))
5176 return -EFAULT;
5177
5178 info.type = prog->type;
5179 info.id = prog->aux->id;
5180 info.load_time = prog->aux->load_time;
5181 info.created_by_uid = from_kuid_munged(current_user_ns(),
5182 prog->aux->user->uid);
5183 info.gpl_compatible = prog->gpl_compatible;
5184
5185 memcpy(info.tag, prog->tag, sizeof(prog->tag));
5186 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
5187
5188 mutex_lock(&prog->aux->used_maps_mutex);
5189 ulen = info.nr_map_ids;
5190 info.nr_map_ids = prog->aux->used_map_cnt;
5191 ulen = min_t(u32, info.nr_map_ids, ulen);
5192 if (ulen) {
5193 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
5194 u32 i;
5195
5196 for (i = 0; i < ulen; i++)
5197 if (put_user(prog->aux->used_maps[i]->id,
5198 &user_map_ids[i])) {
5199 mutex_unlock(&prog->aux->used_maps_mutex);
5200 return -EFAULT;
5201 }
5202 }
5203 mutex_unlock(&prog->aux->used_maps_mutex);
5204
5205 err = set_info_rec_size(&info);
5206 if (err)
5207 return err;
5208
5209 bpf_prog_get_stats(prog, &stats);
5210 info.run_time_ns = stats.nsecs;
5211 info.run_cnt = stats.cnt;
5212 info.recursion_misses = stats.misses;
5213
5214 info.verified_insns = prog->aux->verified_insns;
5215 if (prog->aux->btf)
5216 info.btf_id = btf_obj_id(prog->aux->btf);
5217
5218 if (!bpf_capable()) {
5219 info.jited_prog_len = 0;
5220 info.xlated_prog_len = 0;
5221 info.nr_jited_ksyms = 0;
5222 info.nr_jited_func_lens = 0;
5223 info.nr_func_info = 0;
5224 info.nr_line_info = 0;
5225 info.nr_jited_line_info = 0;
5226 goto done;
5227 }
5228
5229 ulen = info.xlated_prog_len;
5230 info.xlated_prog_len = bpf_prog_insn_size(prog);
5231 if (info.xlated_prog_len && ulen) {
5232 struct bpf_insn *insns_sanitized;
5233 bool fault;
5234
5235 if (!prog->blinded || bpf_dump_raw_ok(file->f_cred)) {
5236 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
5237 if (!insns_sanitized)
5238 return -ENOMEM;
5239 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
5240 ulen = min_t(u32, info.xlated_prog_len, ulen);
5241 fault = copy_to_user(uinsns, insns_sanitized, ulen);
5242 kfree(insns_sanitized);
5243 if (fault)
5244 return -EFAULT;
5245 } else {
5246 info.xlated_prog_insns = 0;
5247 }
5248 }
5249
5250 if (bpf_prog_is_offloaded(prog->aux)) {
5251 err = bpf_prog_offload_info_fill(&info, prog);
5252 if (err)
5253 return err;
5254 goto done;
5255 }
5256
5257 /* NOTE: the following code is supposed to be skipped for offload.
5258 * bpf_prog_offload_info_fill() is the place to fill similar fields
5259 * for offload.
5260 */
5261 ulen = info.jited_prog_len;
5262 if (prog->aux->func_cnt) {
5263 u32 i;
5264
5265 info.jited_prog_len = 0;
5266 for (i = 0; i < prog->aux->func_cnt; i++)
5267 info.jited_prog_len += prog->aux->func[i]->jited_len;
5268 } else {
5269 info.jited_prog_len = prog->jited_len;
5270 }
5271
5272 if (info.jited_prog_len && ulen) {
5273 if (bpf_dump_raw_ok(file->f_cred)) {
5274 uinsns = u64_to_user_ptr(info.jited_prog_insns);
5275 ulen = min_t(u32, info.jited_prog_len, ulen);
5276
5277 /* for multi-function programs, copy the JITed
5278 * instructions for all the functions
5279 */
5280 if (prog->aux->func_cnt) {
5281 u32 len, free, i;
5282 u8 *img;
5283
5284 free = ulen;
5285 for (i = 0; i < prog->aux->func_cnt; i++) {
5286 len = prog->aux->func[i]->jited_len;
5287 len = min_t(u32, len, free);
5288 img = (u8 *) prog->aux->func[i]->bpf_func;
5289 if (copy_to_user(uinsns, img, len))
5290 return -EFAULT;
5291 uinsns += len;
5292 free -= len;
5293 if (!free)
5294 break;
5295 }
5296 } else {
5297 if (copy_to_user(uinsns, prog->bpf_func, ulen))
5298 return -EFAULT;
5299 }
5300 } else {
5301 info.jited_prog_insns = 0;
5302 }
5303 }
5304
5305 ulen = info.nr_jited_ksyms;
5306 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
5307 if (ulen) {
5308 if (bpf_dump_raw_ok(file->f_cred)) {
5309 unsigned long ksym_addr;
5310 u64 __user *user_ksyms;
5311 u32 i;
5312
5313 /* copy the address of the kernel symbol
5314 * corresponding to each function
5315 */
5316 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
5317 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
5318 if (prog->aux->func_cnt) {
5319 for (i = 0; i < ulen; i++) {
5320 ksym_addr = (unsigned long)
5321 prog->aux->func[i]->bpf_func;
5322 if (put_user((u64) ksym_addr,
5323 &user_ksyms[i]))
5324 return -EFAULT;
5325 }
5326 } else {
5327 ksym_addr = (unsigned long) prog->bpf_func;
5328 if (put_user((u64) ksym_addr, &user_ksyms[0]))
5329 return -EFAULT;
5330 }
5331 } else {
5332 info.jited_ksyms = 0;
5333 }
5334 }
5335
5336 ulen = info.nr_jited_func_lens;
5337 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
5338 if (ulen) {
5339 if (bpf_dump_raw_ok(file->f_cred)) {
5340 u32 __user *user_lens;
5341 u32 func_len, i;
5342
5343 /* copy the JITed image lengths for each function */
5344 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
5345 user_lens = u64_to_user_ptr(info.jited_func_lens);
5346 if (prog->aux->func_cnt) {
5347 for (i = 0; i < ulen; i++) {
5348 func_len =
5349 prog->aux->func[i]->jited_len;
5350 if (put_user(func_len, &user_lens[i]))
5351 return -EFAULT;
5352 }
5353 } else {
5354 func_len = prog->jited_len;
5355 if (put_user(func_len, &user_lens[0]))
5356 return -EFAULT;
5357 }
5358 } else {
5359 info.jited_func_lens = 0;
5360 }
5361 }
5362
5363 info.attach_btf_id = prog->aux->attach_btf_id;
5364 if (attach_btf)
5365 info.attach_btf_obj_id = btf_obj_id(attach_btf);
5366
5367 ulen = info.nr_func_info;
5368 info.nr_func_info = prog->aux->func_info_cnt;
5369 if (info.nr_func_info && ulen) {
5370 char __user *user_finfo;
5371
5372 user_finfo = u64_to_user_ptr(info.func_info);
5373 ulen = min_t(u32, info.nr_func_info, ulen);
5374 if (copy_to_user(user_finfo, prog->aux->func_info,
5375 info.func_info_rec_size * ulen))
5376 return -EFAULT;
5377 }
5378
5379 ulen = info.nr_line_info;
5380 info.nr_line_info = prog->aux->nr_linfo;
5381 if (info.nr_line_info && ulen) {
5382 __u8 __user *user_linfo;
5383
5384 user_linfo = u64_to_user_ptr(info.line_info);
5385 ulen = min_t(u32, info.nr_line_info, ulen);
5386 if (copy_to_user(user_linfo, prog->aux->linfo,
5387 info.line_info_rec_size * ulen))
5388 return -EFAULT;
5389 }
5390
5391 ulen = info.nr_jited_line_info;
5392 if (prog->aux->jited_linfo)
5393 info.nr_jited_line_info = prog->aux->nr_linfo;
5394 else
5395 info.nr_jited_line_info = 0;
5396 if (info.nr_jited_line_info && ulen) {
5397 if (bpf_dump_raw_ok(file->f_cred)) {
5398 unsigned long line_addr;
5399 __u64 __user *user_linfo;
5400 u32 i;
5401
5402 user_linfo = u64_to_user_ptr(info.jited_line_info);
5403 ulen = min_t(u32, info.nr_jited_line_info, ulen);
5404 for (i = 0; i < ulen; i++) {
5405 line_addr = (unsigned long)prog->aux->jited_linfo[i];
5406 if (put_user((__u64)line_addr, &user_linfo[i]))
5407 return -EFAULT;
5408 }
5409 } else {
5410 info.jited_line_info = 0;
5411 }
5412 }
5413
5414 ulen = info.nr_prog_tags;
5415 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
5416 if (ulen) {
5417 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
5418 u32 i;
5419
5420 user_prog_tags = u64_to_user_ptr(info.prog_tags);
5421 ulen = min_t(u32, info.nr_prog_tags, ulen);
5422 if (prog->aux->func_cnt) {
5423 for (i = 0; i < ulen; i++) {
5424 if (copy_to_user(user_prog_tags[i],
5425 prog->aux->func[i]->tag,
5426 BPF_TAG_SIZE))
5427 return -EFAULT;
5428 }
5429 } else {
5430 if (copy_to_user(user_prog_tags[0],
5431 prog->tag, BPF_TAG_SIZE))
5432 return -EFAULT;
5433 }
5434 }
5435
5436done:
5437 if (copy_to_user(uinfo, &info, info_len) ||
5438 put_user(info_len, &uattr->info.info_len))
5439 return -EFAULT;
5440
5441 return 0;
5442}
5443
5444static int bpf_map_get_info_by_fd(struct file *file,
5445 struct bpf_map *map,
5446 const union bpf_attr *attr,
5447 union bpf_attr __user *uattr)
5448{
5449 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
5450 struct bpf_map_info info;
5451 u32 info_len = attr->info.info_len, len;
5452 int err;
5453
5454 len = offsetofend(struct bpf_map_info, hash_size);
5455 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), len, info_len);
5456 if (err)
5457 return err;
5458 info_len = min_t(u32, sizeof(info), info_len);
5459
5460 memset(&info, 0, sizeof(info));
5461 if (copy_from_user(&info, uinfo, info_len))
5462 return -EFAULT;
5463
5464 info.type = map->map_type;
5465 info.id = map->id;
5466 info.key_size = map->key_size;
5467 info.value_size = map->value_size;
5468 info.max_entries = map->max_entries;
5469 info.map_flags = map->map_flags;
5470 info.map_extra = map->map_extra;
5471 memcpy(info.name, map->name, sizeof(map->name));
5472
5473 if (map->btf) {
5474 info.btf_id = btf_obj_id(map->btf);
5475 info.btf_key_type_id = map->btf_key_type_id;
5476 info.btf_value_type_id = map->btf_value_type_id;
5477 }
5478 info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
5479 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS)
5480 bpf_map_struct_ops_info_fill(&info, map);
5481
5482 if (bpf_map_is_offloaded(map)) {
5483 err = bpf_map_offload_info_fill(&info, map);
5484 if (err)
5485 return err;
5486 }
5487
5488 if (info.hash) {
5489 char __user *uhash = u64_to_user_ptr(info.hash);
5490
5491 if (!map->ops->map_get_hash)
5492 return -EINVAL;
5493 if (info.hash_size != sizeof(map->sha))
5494 return -EINVAL;
5495 if (!READ_ONCE(map->frozen))
5496 return -EPERM;
5497
5498 err = map->ops->map_get_hash(map);
5499 if (err != 0)
5500 return err;
5501
5502 if (copy_to_user(uhash, map->sha, sizeof(map->sha)) != 0)
5503 return -EFAULT;
5504 } else if (info.hash_size) {
5505 return -EINVAL;
5506 }
5507
5508 if (copy_to_user(uinfo, &info, info_len) ||
5509 put_user(info_len, &uattr->info.info_len))
5510 return -EFAULT;
5511
5512 return 0;
5513}
5514
5515static int bpf_btf_get_info_by_fd(struct file *file,
5516 struct btf *btf,
5517 const union bpf_attr *attr,
5518 union bpf_attr __user *uattr)
5519{
5520 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
5521 u32 info_len = attr->info.info_len;
5522 int err;
5523
5524 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
5525 if (err)
5526 return err;
5527
5528 return btf_get_info_by_fd(btf, attr, uattr);
5529}
5530
5531static int bpf_link_get_info_by_fd(struct file *file,
5532 struct bpf_link *link,
5533 const union bpf_attr *attr,
5534 union bpf_attr __user *uattr)
5535{
5536 struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
5537 struct bpf_link_info info;
5538 u32 info_len = attr->info.info_len;
5539 int err;
5540
5541 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
5542 if (err)
5543 return err;
5544 info_len = min_t(u32, sizeof(info), info_len);
5545
5546 memset(&info, 0, sizeof(info));
5547 if (copy_from_user(&info, uinfo, info_len))
5548 return -EFAULT;
5549
5550 info.type = link->type;
5551 info.id = link->id;
5552 if (link->prog)
5553 info.prog_id = link->prog->aux->id;
5554
5555 if (link->ops->fill_link_info) {
5556 err = link->ops->fill_link_info(link, &info);
5557 if (err)
5558 return err;
5559 }
5560
5561 if (copy_to_user(uinfo, &info, info_len) ||
5562 put_user(info_len, &uattr->info.info_len))
5563 return -EFAULT;
5564
5565 return 0;
5566}
5567
5568
5569static int token_get_info_by_fd(struct file *file,
5570 struct bpf_token *token,
5571 const union bpf_attr *attr,
5572 union bpf_attr __user *uattr)
5573{
5574 struct bpf_token_info __user *uinfo = u64_to_user_ptr(attr->info.info);
5575 u32 info_len = attr->info.info_len;
5576 int err;
5577
5578 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
5579 if (err)
5580 return err;
5581 return bpf_token_get_info_by_fd(token, attr, uattr);
5582}
5583
5584#define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
5585
5586static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
5587 union bpf_attr __user *uattr)
5588{
5589 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
5590 return -EINVAL;
5591
5592 CLASS(fd, f)(attr->info.bpf_fd);
5593 if (fd_empty(f))
5594 return -EBADFD;
5595
5596 if (fd_file(f)->f_op == &bpf_prog_fops)
5597 return bpf_prog_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr,
5598 uattr);
5599 else if (fd_file(f)->f_op == &bpf_map_fops)
5600 return bpf_map_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr,
5601 uattr);
5602 else if (fd_file(f)->f_op == &btf_fops)
5603 return bpf_btf_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr, uattr);
5604 else if (fd_file(f)->f_op == &bpf_link_fops || fd_file(f)->f_op == &bpf_link_fops_poll)
5605 return bpf_link_get_info_by_fd(fd_file(f), fd_file(f)->private_data,
5606 attr, uattr);
5607 else if (fd_file(f)->f_op == &bpf_token_fops)
5608 return token_get_info_by_fd(fd_file(f), fd_file(f)->private_data,
5609 attr, uattr);
5610 return -EINVAL;
5611}
5612
5613#define BPF_BTF_LOAD_LAST_FIELD btf_token_fd
5614
5615static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr, struct bpf_log_attr *attr_log)
5616{
5617 struct bpf_token *token = NULL;
5618
5619 if (CHECK_ATTR(BPF_BTF_LOAD))
5620 return -EINVAL;
5621
5622 if (attr->btf_flags & ~BPF_F_TOKEN_FD)
5623 return -EINVAL;
5624
5625 if (attr->btf_flags & BPF_F_TOKEN_FD) {
5626 token = bpf_token_get_from_fd(attr->btf_token_fd);
5627 if (IS_ERR(token))
5628 return PTR_ERR(token);
5629 if (!bpf_token_allow_cmd(token, BPF_BTF_LOAD)) {
5630 bpf_token_put(token);
5631 token = NULL;
5632 }
5633 }
5634
5635 if (!bpf_token_capable(token, CAP_BPF)) {
5636 bpf_token_put(token);
5637 return -EPERM;
5638 }
5639
5640 bpf_token_put(token);
5641
5642 return btf_new_fd(attr, uattr, attr_log);
5643}
5644
5645#define BPF_BTF_GET_FD_BY_ID_LAST_FIELD fd_by_id_token_fd
5646
5647static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
5648{
5649 struct bpf_token *token = NULL;
5650
5651 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
5652 return -EINVAL;
5653
5654 if (attr->open_flags & ~BPF_F_TOKEN_FD)
5655 return -EINVAL;
5656
5657 if (attr->open_flags & BPF_F_TOKEN_FD) {
5658 token = bpf_token_get_from_fd(attr->fd_by_id_token_fd);
5659 if (IS_ERR(token))
5660 return PTR_ERR(token);
5661 if (!bpf_token_allow_cmd(token, BPF_BTF_GET_FD_BY_ID)) {
5662 bpf_token_put(token);
5663 token = NULL;
5664 }
5665 }
5666
5667 if (!bpf_token_capable(token, CAP_SYS_ADMIN)) {
5668 bpf_token_put(token);
5669 return -EPERM;
5670 }
5671
5672 bpf_token_put(token);
5673
5674 return btf_get_fd_by_id(attr->btf_id);
5675}
5676
5677static int bpf_task_fd_query_copy(const union bpf_attr *attr,
5678 union bpf_attr __user *uattr,
5679 u32 prog_id, u32 fd_type,
5680 const char *buf, u64 probe_offset,
5681 u64 probe_addr)
5682{
5683 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
5684 u32 len = buf ? strlen(buf) : 0, input_len;
5685 int err = 0;
5686
5687 if (put_user(len, &uattr->task_fd_query.buf_len))
5688 return -EFAULT;
5689 input_len = attr->task_fd_query.buf_len;
5690 if (input_len && ubuf) {
5691 if (!len) {
5692 /* nothing to copy, just make ubuf NULL terminated */
5693 char zero = '\0';
5694
5695 if (put_user(zero, ubuf))
5696 return -EFAULT;
5697 } else {
5698 err = bpf_copy_to_user(ubuf, buf, input_len, len);
5699 if (err == -EFAULT)
5700 return err;
5701 }
5702 }
5703
5704 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
5705 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
5706 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
5707 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
5708 return -EFAULT;
5709
5710 return err;
5711}
5712
5713#define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
5714
5715static int bpf_task_fd_query(const union bpf_attr *attr,
5716 union bpf_attr __user *uattr)
5717{
5718 pid_t pid = attr->task_fd_query.pid;
5719 u32 fd = attr->task_fd_query.fd;
5720 const struct perf_event *event;
5721 struct task_struct *task;
5722 struct file *file;
5723 int err;
5724
5725 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
5726 return -EINVAL;
5727
5728 if (!capable(CAP_SYS_ADMIN))
5729 return -EPERM;
5730
5731 if (attr->task_fd_query.flags != 0)
5732 return -EINVAL;
5733
5734 rcu_read_lock();
5735 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
5736 rcu_read_unlock();
5737 if (!task)
5738 return -ENOENT;
5739
5740 err = 0;
5741 file = fget_task(task, fd);
5742 put_task_struct(task);
5743 if (!file)
5744 return -EBADF;
5745
5746 if (file->f_op == &bpf_link_fops || file->f_op == &bpf_link_fops_poll) {
5747 struct bpf_link *link = file->private_data;
5748
5749 if (link->ops == &bpf_raw_tp_link_lops) {
5750 struct bpf_raw_tp_link *raw_tp =
5751 container_of(link, struct bpf_raw_tp_link, link);
5752 struct bpf_raw_event_map *btp = raw_tp->btp;
5753
5754 err = bpf_task_fd_query_copy(attr, uattr,
5755 raw_tp->link.prog->aux->id,
5756 BPF_FD_TYPE_RAW_TRACEPOINT,
5757 btp->tp->name, 0, 0);
5758 goto put_file;
5759 }
5760 goto out_not_supp;
5761 }
5762
5763 event = perf_get_event(file);
5764 if (!IS_ERR(event)) {
5765 u64 probe_offset, probe_addr;
5766 u32 prog_id, fd_type;
5767 const char *buf;
5768
5769 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
5770 &buf, &probe_offset,
5771 &probe_addr, NULL);
5772 if (!err)
5773 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
5774 fd_type, buf,
5775 probe_offset,
5776 probe_addr);
5777 goto put_file;
5778 }
5779
5780out_not_supp:
5781 err = -ENOTSUPP;
5782put_file:
5783 fput(file);
5784 return err;
5785}
5786
5787#define BPF_MAP_BATCH_LAST_FIELD batch.flags
5788
5789#define BPF_DO_BATCH(fn, ...) \
5790 do { \
5791 if (!fn) { \
5792 err = -ENOTSUPP; \
5793 goto err_put; \
5794 } \
5795 err = fn(__VA_ARGS__); \
5796 } while (0)
5797
5798static int bpf_map_do_batch(const union bpf_attr *attr,
5799 union bpf_attr __user *uattr,
5800 int cmd)
5801{
5802 bool has_read = cmd == BPF_MAP_LOOKUP_BATCH ||
5803 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
5804 bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
5805 struct bpf_map *map;
5806 int err;
5807
5808 if (CHECK_ATTR(BPF_MAP_BATCH))
5809 return -EINVAL;
5810
5811 CLASS(fd, f)(attr->batch.map_fd);
5812
5813 map = __bpf_map_get(f);
5814 if (IS_ERR(map))
5815 return PTR_ERR(map);
5816 if (has_write)
5817 bpf_map_write_active_inc(map);
5818 if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
5819 err = -EPERM;
5820 goto err_put;
5821 }
5822 if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
5823 err = -EPERM;
5824 goto err_put;
5825 }
5826
5827 if (cmd == BPF_MAP_LOOKUP_BATCH)
5828 BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
5829 else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
5830 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
5831 else if (cmd == BPF_MAP_UPDATE_BATCH)
5832 BPF_DO_BATCH(map->ops->map_update_batch, map, fd_file(f), attr, uattr);
5833 else
5834 BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
5835err_put:
5836 if (has_write) {
5837 maybe_wait_bpf_programs(map);
5838 bpf_map_write_active_dec(map);
5839 }
5840 return err;
5841}
5842
5843#define BPF_LINK_CREATE_LAST_FIELD link_create.uprobe_multi.path_fd
5844static int link_create(union bpf_attr *attr, bpfptr_t uattr)
5845{
5846 struct bpf_prog *prog;
5847 int ret;
5848
5849 if (CHECK_ATTR(BPF_LINK_CREATE))
5850 return -EINVAL;
5851
5852 if (attr->link_create.attach_type == BPF_STRUCT_OPS)
5853 return bpf_struct_ops_link_create(attr);
5854
5855 prog = bpf_prog_get(attr->link_create.prog_fd);
5856 if (IS_ERR(prog))
5857 return PTR_ERR(prog);
5858
5859 ret = bpf_prog_attach_check_attach_type(prog,
5860 attr->link_create.attach_type);
5861 if (ret)
5862 goto out;
5863
5864 switch (prog->type) {
5865 case BPF_PROG_TYPE_CGROUP_SKB:
5866 case BPF_PROG_TYPE_CGROUP_SOCK:
5867 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
5868 case BPF_PROG_TYPE_SOCK_OPS:
5869 case BPF_PROG_TYPE_CGROUP_DEVICE:
5870 case BPF_PROG_TYPE_CGROUP_SYSCTL:
5871 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
5872 ret = cgroup_bpf_link_attach(attr, prog);
5873 break;
5874 case BPF_PROG_TYPE_EXT:
5875 ret = bpf_tracing_prog_attach(prog,
5876 attr->link_create.target_fd,
5877 attr->link_create.target_btf_id,
5878 attr->link_create.tracing.cookie,
5879 attr->link_create.attach_type);
5880 break;
5881 case BPF_PROG_TYPE_LSM:
5882 case BPF_PROG_TYPE_TRACING:
5883 if (attr->link_create.attach_type != prog->expected_attach_type) {
5884 ret = -EINVAL;
5885 goto out;
5886 }
5887 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
5888 ret = bpf_raw_tp_link_attach(prog, NULL, attr->link_create.tracing.cookie,
5889 attr->link_create.attach_type);
5890 else if (prog->expected_attach_type == BPF_TRACE_ITER)
5891 ret = bpf_iter_link_attach(attr, uattr, prog);
5892 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
5893 ret = cgroup_bpf_link_attach(attr, prog);
5894 else if (is_tracing_multi(prog->expected_attach_type))
5895 ret = bpf_tracing_multi_attach(prog, attr);
5896 else
5897 ret = bpf_tracing_prog_attach(prog,
5898 attr->link_create.target_fd,
5899 attr->link_create.target_btf_id,
5900 attr->link_create.tracing.cookie,
5901 attr->link_create.attach_type);
5902 break;
5903 case BPF_PROG_TYPE_FLOW_DISSECTOR:
5904 case BPF_PROG_TYPE_SK_LOOKUP:
5905 ret = netns_bpf_link_create(attr, prog);
5906 break;
5907 case BPF_PROG_TYPE_SK_MSG:
5908 case BPF_PROG_TYPE_SK_SKB:
5909 ret = sock_map_link_create(attr, prog);
5910 break;
5911#ifdef CONFIG_NET
5912 case BPF_PROG_TYPE_XDP:
5913 ret = bpf_xdp_link_attach(attr, prog);
5914 break;
5915 case BPF_PROG_TYPE_SCHED_CLS:
5916 if (attr->link_create.attach_type == BPF_TCX_INGRESS ||
5917 attr->link_create.attach_type == BPF_TCX_EGRESS)
5918 ret = tcx_link_attach(attr, prog);
5919 else
5920 ret = netkit_link_attach(attr, prog);
5921 break;
5922 case BPF_PROG_TYPE_NETFILTER:
5923 ret = bpf_nf_link_attach(attr, prog);
5924 break;
5925#endif
5926 case BPF_PROG_TYPE_PERF_EVENT:
5927 case BPF_PROG_TYPE_TRACEPOINT:
5928 ret = bpf_perf_link_attach(attr, prog);
5929 break;
5930 case BPF_PROG_TYPE_KPROBE:
5931 if (attr->link_create.attach_type == BPF_PERF_EVENT)
5932 ret = bpf_perf_link_attach(attr, prog);
5933 else if (attr->link_create.attach_type == BPF_TRACE_KPROBE_MULTI ||
5934 attr->link_create.attach_type == BPF_TRACE_KPROBE_SESSION)
5935 ret = bpf_kprobe_multi_link_attach(attr, prog);
5936 else if (attr->link_create.attach_type == BPF_TRACE_UPROBE_MULTI ||
5937 attr->link_create.attach_type == BPF_TRACE_UPROBE_SESSION)
5938 ret = bpf_uprobe_multi_link_attach(attr, prog);
5939 break;
5940 default:
5941 ret = -EINVAL;
5942 }
5943
5944out:
5945 if (ret < 0)
5946 bpf_prog_put(prog);
5947 return ret;
5948}
5949
5950static int link_update_map(struct bpf_link *link, union bpf_attr *attr)
5951{
5952 struct bpf_map *new_map, *old_map = NULL;
5953 int ret;
5954
5955 new_map = bpf_map_get(attr->link_update.new_map_fd);
5956 if (IS_ERR(new_map))
5957 return PTR_ERR(new_map);
5958
5959 if (attr->link_update.flags & BPF_F_REPLACE) {
5960 old_map = bpf_map_get(attr->link_update.old_map_fd);
5961 if (IS_ERR(old_map)) {
5962 ret = PTR_ERR(old_map);
5963 goto out_put;
5964 }
5965 } else if (attr->link_update.old_map_fd) {
5966 ret = -EINVAL;
5967 goto out_put;
5968 }
5969
5970 ret = link->ops->update_map(link, new_map, old_map);
5971
5972 if (old_map)
5973 bpf_map_put(old_map);
5974out_put:
5975 bpf_map_put(new_map);
5976 return ret;
5977}
5978
5979#define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
5980
5981static int link_update(union bpf_attr *attr)
5982{
5983 struct bpf_prog *old_prog = NULL, *new_prog;
5984 struct bpf_link *link;
5985 u32 flags;
5986 int ret;
5987
5988 if (CHECK_ATTR(BPF_LINK_UPDATE))
5989 return -EINVAL;
5990
5991 flags = attr->link_update.flags;
5992 if (flags & ~BPF_F_REPLACE)
5993 return -EINVAL;
5994
5995 link = bpf_link_get_from_fd(attr->link_update.link_fd);
5996 if (IS_ERR(link))
5997 return PTR_ERR(link);
5998
5999 if (link->ops->update_map) {
6000 ret = link_update_map(link, attr);
6001 goto out_put_link;
6002 }
6003
6004 new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
6005 if (IS_ERR(new_prog)) {
6006 ret = PTR_ERR(new_prog);
6007 goto out_put_link;
6008 }
6009
6010 if (flags & BPF_F_REPLACE) {
6011 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
6012 if (IS_ERR(old_prog)) {
6013 ret = PTR_ERR(old_prog);
6014 old_prog = NULL;
6015 goto out_put_progs;
6016 }
6017 } else if (attr->link_update.old_prog_fd) {
6018 ret = -EINVAL;
6019 goto out_put_progs;
6020 }
6021
6022 if (link->ops->update_prog)
6023 ret = link->ops->update_prog(link, new_prog, old_prog);
6024 else
6025 ret = -EINVAL;
6026
6027out_put_progs:
6028 if (old_prog)
6029 bpf_prog_put(old_prog);
6030 if (ret)
6031 bpf_prog_put(new_prog);
6032out_put_link:
6033 bpf_link_put_direct(link);
6034 return ret;
6035}
6036
6037#define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
6038
6039static int link_detach(union bpf_attr *attr)
6040{
6041 struct bpf_link *link;
6042 int ret;
6043
6044 if (CHECK_ATTR(BPF_LINK_DETACH))
6045 return -EINVAL;
6046
6047 link = bpf_link_get_from_fd(attr->link_detach.link_fd);
6048 if (IS_ERR(link))
6049 return PTR_ERR(link);
6050
6051 if (link->ops->detach)
6052 ret = link->ops->detach(link);
6053 else
6054 ret = -EOPNOTSUPP;
6055
6056 bpf_link_put_direct(link);
6057 return ret;
6058}
6059
6060struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
6061{
6062 return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
6063}
6064EXPORT_SYMBOL(bpf_link_inc_not_zero);
6065
6066struct bpf_link *bpf_link_by_id(u32 id)
6067{
6068 struct bpf_link *link;
6069
6070 if (!id)
6071 return ERR_PTR(-ENOENT);
6072
6073 spin_lock_bh(&link_idr_lock);
6074 /* before link is "settled", ID is 0, pretend it doesn't exist yet */
6075 link = idr_find(&link_idr, id);
6076 if (link) {
6077 if (link->id)
6078 link = bpf_link_inc_not_zero(link);
6079 else
6080 link = ERR_PTR(-EAGAIN);
6081 } else {
6082 link = ERR_PTR(-ENOENT);
6083 }
6084 spin_unlock_bh(&link_idr_lock);
6085 return link;
6086}
6087
6088struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
6089{
6090 struct bpf_link *link;
6091
6092 spin_lock_bh(&link_idr_lock);
6093again:
6094 link = idr_get_next(&link_idr, id);
6095 if (link) {
6096 link = bpf_link_inc_not_zero(link);
6097 if (IS_ERR(link)) {
6098 (*id)++;
6099 goto again;
6100 }
6101 }
6102 spin_unlock_bh(&link_idr_lock);
6103
6104 return link;
6105}
6106
6107#define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
6108
6109static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
6110{
6111 struct bpf_link *link;
6112 u32 id = attr->link_id;
6113 int fd;
6114
6115 if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
6116 return -EINVAL;
6117
6118 if (!capable(CAP_SYS_ADMIN))
6119 return -EPERM;
6120
6121 link = bpf_link_by_id(id);
6122 if (IS_ERR(link))
6123 return PTR_ERR(link);
6124
6125 fd = bpf_link_new_fd(link);
6126 if (fd < 0)
6127 bpf_link_put_direct(link);
6128
6129 return fd;
6130}
6131
6132DEFINE_MUTEX(bpf_stats_enabled_mutex);
6133
6134static int bpf_stats_release(struct inode *inode, struct file *file)
6135{
6136 mutex_lock(&bpf_stats_enabled_mutex);
6137 static_key_slow_dec(&bpf_stats_enabled_key.key);
6138 mutex_unlock(&bpf_stats_enabled_mutex);
6139 return 0;
6140}
6141
6142static const struct file_operations bpf_stats_fops = {
6143 .release = bpf_stats_release,
6144};
6145
6146static int bpf_enable_runtime_stats(void)
6147{
6148 int fd;
6149
6150 mutex_lock(&bpf_stats_enabled_mutex);
6151
6152 /* Set a very high limit to avoid overflow */
6153 if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
6154 mutex_unlock(&bpf_stats_enabled_mutex);
6155 return -EBUSY;
6156 }
6157
6158 fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
6159 if (fd >= 0)
6160 static_key_slow_inc(&bpf_stats_enabled_key.key);
6161
6162 mutex_unlock(&bpf_stats_enabled_mutex);
6163 return fd;
6164}
6165
6166#define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
6167
6168static int bpf_enable_stats(union bpf_attr *attr)
6169{
6170
6171 if (CHECK_ATTR(BPF_ENABLE_STATS))
6172 return -EINVAL;
6173
6174 if (!capable(CAP_SYS_ADMIN))
6175 return -EPERM;
6176
6177 switch (attr->enable_stats.type) {
6178 case BPF_STATS_RUN_TIME:
6179 return bpf_enable_runtime_stats();
6180 default:
6181 break;
6182 }
6183 return -EINVAL;
6184}
6185
6186#define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
6187
6188static int bpf_iter_create(union bpf_attr *attr)
6189{
6190 struct bpf_link *link;
6191 int err;
6192
6193 if (CHECK_ATTR(BPF_ITER_CREATE))
6194 return -EINVAL;
6195
6196 if (attr->iter_create.flags)
6197 return -EINVAL;
6198
6199 link = bpf_link_get_from_fd(attr->iter_create.link_fd);
6200 if (IS_ERR(link))
6201 return PTR_ERR(link);
6202
6203 err = bpf_iter_new_fd(link);
6204 bpf_link_put_direct(link);
6205
6206 return err;
6207}
6208
6209#define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
6210
6211static int bpf_prog_bind_map(union bpf_attr *attr)
6212{
6213 struct bpf_prog *prog;
6214 struct bpf_map *map;
6215 struct bpf_map **used_maps_old, **used_maps_new;
6216 int i, ret = 0;
6217
6218 if (CHECK_ATTR(BPF_PROG_BIND_MAP))
6219 return -EINVAL;
6220
6221 if (attr->prog_bind_map.flags)
6222 return -EINVAL;
6223
6224 prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
6225 if (IS_ERR(prog))
6226 return PTR_ERR(prog);
6227
6228 map = bpf_map_get(attr->prog_bind_map.map_fd);
6229 if (IS_ERR(map)) {
6230 ret = PTR_ERR(map);
6231 goto out_prog_put;
6232 }
6233
6234 mutex_lock(&prog->aux->used_maps_mutex);
6235
6236 used_maps_old = prog->aux->used_maps;
6237
6238 for (i = 0; i < prog->aux->used_map_cnt; i++)
6239 if (used_maps_old[i] == map) {
6240 bpf_map_put(map);
6241 goto out_unlock;
6242 }
6243
6244 used_maps_new = kmalloc_objs(used_maps_new[0],
6245 prog->aux->used_map_cnt + 1);
6246 if (!used_maps_new) {
6247 ret = -ENOMEM;
6248 goto out_unlock;
6249 }
6250
6251 /* The bpf program will not access the bpf map, but for the sake of
6252 * simplicity, increase sleepable_refcnt for sleepable program as well.
6253 */
6254 if (prog->sleepable)
6255 atomic64_inc(&map->sleepable_refcnt);
6256 memcpy(used_maps_new, used_maps_old,
6257 sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
6258 used_maps_new[prog->aux->used_map_cnt] = map;
6259
6260 prog->aux->used_map_cnt++;
6261 prog->aux->used_maps = used_maps_new;
6262
6263 kfree(used_maps_old);
6264
6265out_unlock:
6266 mutex_unlock(&prog->aux->used_maps_mutex);
6267
6268 if (ret)
6269 bpf_map_put(map);
6270out_prog_put:
6271 bpf_prog_put(prog);
6272 return ret;
6273}
6274
6275#define BPF_TOKEN_CREATE_LAST_FIELD token_create.bpffs_fd
6276
6277static int token_create(union bpf_attr *attr)
6278{
6279 if (CHECK_ATTR(BPF_TOKEN_CREATE))
6280 return -EINVAL;
6281
6282 /* no flags are supported yet */
6283 if (attr->token_create.flags)
6284 return -EINVAL;
6285
6286 return bpf_token_create(attr);
6287}
6288
6289#define BPF_PROG_STREAM_READ_BY_FD_LAST_FIELD prog_stream_read.prog_fd
6290
6291static int prog_stream_read(union bpf_attr *attr)
6292{
6293 char __user *buf = u64_to_user_ptr(attr->prog_stream_read.stream_buf);
6294 u32 len = attr->prog_stream_read.stream_buf_len;
6295 struct bpf_prog *prog;
6296 int ret;
6297
6298 if (CHECK_ATTR(BPF_PROG_STREAM_READ_BY_FD))
6299 return -EINVAL;
6300
6301 prog = bpf_prog_get(attr->prog_stream_read.prog_fd);
6302 if (IS_ERR(prog))
6303 return PTR_ERR(prog);
6304
6305 ret = bpf_prog_stream_read(prog, attr->prog_stream_read.stream_id, buf, len);
6306 bpf_prog_put(prog);
6307
6308 return ret;
6309}
6310
6311#define BPF_PROG_ASSOC_STRUCT_OPS_LAST_FIELD prog_assoc_struct_ops.flags
6312
6313static int prog_assoc_struct_ops(union bpf_attr *attr)
6314{
6315 struct bpf_prog *prog;
6316 struct bpf_map *map;
6317 int ret;
6318
6319 if (CHECK_ATTR(BPF_PROG_ASSOC_STRUCT_OPS))
6320 return -EINVAL;
6321
6322 if (attr->prog_assoc_struct_ops.flags)
6323 return -EINVAL;
6324
6325 prog = bpf_prog_get(attr->prog_assoc_struct_ops.prog_fd);
6326 if (IS_ERR(prog))
6327 return PTR_ERR(prog);
6328
6329 if (prog->type == BPF_PROG_TYPE_STRUCT_OPS) {
6330 ret = -EINVAL;
6331 goto put_prog;
6332 }
6333
6334 map = bpf_map_get(attr->prog_assoc_struct_ops.map_fd);
6335 if (IS_ERR(map)) {
6336 ret = PTR_ERR(map);
6337 goto put_prog;
6338 }
6339
6340 if (map->map_type != BPF_MAP_TYPE_STRUCT_OPS) {
6341 ret = -EINVAL;
6342 goto put_map;
6343 }
6344
6345 ret = bpf_prog_assoc_struct_ops(prog, map);
6346
6347put_map:
6348 bpf_map_put(map);
6349put_prog:
6350 bpf_prog_put(prog);
6351 return ret;
6352}
6353
6354static int __sys_bpf(enum bpf_cmd cmd, bpfptr_t uattr, unsigned int size,
6355 bpfptr_t uattr_common, unsigned int size_common)
6356{
6357 struct bpf_common_attr attr_common;
6358 u32 offsetof_log_true_size = 0;
6359 struct bpf_log_attr attr_log;
6360 union bpf_attr attr;
6361 int err;
6362
6363 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
6364 if (err)
6365 return err;
6366 size = min_t(u32, size, sizeof(attr));
6367
6368 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
6369 memset(&attr, 0, sizeof(attr));
6370 if (copy_from_bpfptr(&attr, uattr, size) != 0)
6371 return -EFAULT;
6372
6373 memset(&attr_common, 0, sizeof(attr_common));
6374 if (cmd & BPF_COMMON_ATTRS) {
6375 err = bpf_check_uarg_tail_zero(uattr_common,
6376 offsetofend(struct bpf_common_attr, log_true_size),
6377 size_common);
6378 if (err)
6379 return err;
6380
6381 cmd &= ~BPF_COMMON_ATTRS;
6382 size_common = min_t(u32, size_common, sizeof(attr_common));
6383 if (copy_from_bpfptr(&attr_common, uattr_common, size_common) != 0)
6384 return -EFAULT;
6385 } else {
6386 size_common = 0;
6387 }
6388
6389 err = security_bpf(cmd, &attr, size, uattr.is_kernel);
6390 if (err < 0)
6391 return err;
6392
6393 switch (cmd) {
6394 case BPF_MAP_CREATE:
6395 err = map_create(&attr, uattr, &attr_common, uattr_common, size_common);
6396 break;
6397 case BPF_MAP_LOOKUP_ELEM:
6398 err = map_lookup_elem(&attr);
6399 break;
6400 case BPF_MAP_UPDATE_ELEM:
6401 err = map_update_elem(&attr, uattr);
6402 break;
6403 case BPF_MAP_DELETE_ELEM:
6404 err = map_delete_elem(&attr, uattr);
6405 break;
6406 case BPF_MAP_GET_NEXT_KEY:
6407 err = map_get_next_key(&attr);
6408 break;
6409 case BPF_MAP_FREEZE:
6410 err = map_freeze(&attr);
6411 break;
6412 case BPF_PROG_LOAD:
6413 if (size >= offsetofend(union bpf_attr, log_true_size))
6414 offsetof_log_true_size = offsetof(union bpf_attr, log_true_size);
6415 err = bpf_log_attr_init(&attr_log, attr.log_buf, attr.log_size, attr.log_level,
6416 offsetof_log_true_size, uattr, &attr_common, uattr_common,
6417 size_common);
6418 err = err ?: bpf_prog_load(&attr, uattr, &attr_log);
6419 break;
6420 case BPF_OBJ_PIN:
6421 err = bpf_obj_pin(&attr);
6422 break;
6423 case BPF_OBJ_GET:
6424 err = bpf_obj_get(&attr);
6425 break;
6426 case BPF_PROG_ATTACH:
6427 err = bpf_prog_attach(&attr);
6428 break;
6429 case BPF_PROG_DETACH:
6430 err = bpf_prog_detach(&attr);
6431 break;
6432 case BPF_PROG_QUERY:
6433 err = bpf_prog_query(&attr, uattr.user, size);
6434 break;
6435 case BPF_PROG_TEST_RUN:
6436 err = bpf_prog_test_run(&attr, uattr.user);
6437 break;
6438 case BPF_PROG_GET_NEXT_ID:
6439 err = bpf_obj_get_next_id(&attr, uattr.user,
6440 &prog_idr, &prog_idr_lock);
6441 break;
6442 case BPF_MAP_GET_NEXT_ID:
6443 err = bpf_obj_get_next_id(&attr, uattr.user,
6444 &map_idr, &map_idr_lock);
6445 break;
6446 case BPF_BTF_GET_NEXT_ID:
6447 err = bpf_obj_get_next_id(&attr, uattr.user,
6448 &btf_idr, &btf_idr_lock);
6449 break;
6450 case BPF_PROG_GET_FD_BY_ID:
6451 err = bpf_prog_get_fd_by_id(&attr);
6452 break;
6453 case BPF_MAP_GET_FD_BY_ID:
6454 err = bpf_map_get_fd_by_id(&attr);
6455 break;
6456 case BPF_OBJ_GET_INFO_BY_FD:
6457 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
6458 break;
6459 case BPF_RAW_TRACEPOINT_OPEN:
6460 err = bpf_raw_tracepoint_open(&attr);
6461 break;
6462 case BPF_BTF_LOAD:
6463 if (size >= offsetofend(union bpf_attr, btf_log_true_size))
6464 offsetof_log_true_size = offsetof(union bpf_attr, btf_log_true_size);
6465 err = bpf_log_attr_init(&attr_log, attr.btf_log_buf, attr.btf_log_size,
6466 attr.btf_log_level, offsetof_log_true_size, uattr,
6467 &attr_common, uattr_common, size_common);
6468 err = err ?: bpf_btf_load(&attr, uattr, &attr_log);
6469 break;
6470 case BPF_BTF_GET_FD_BY_ID:
6471 err = bpf_btf_get_fd_by_id(&attr);
6472 break;
6473 case BPF_TASK_FD_QUERY:
6474 err = bpf_task_fd_query(&attr, uattr.user);
6475 break;
6476 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
6477 err = map_lookup_and_delete_elem(&attr);
6478 break;
6479 case BPF_MAP_LOOKUP_BATCH:
6480 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
6481 break;
6482 case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
6483 err = bpf_map_do_batch(&attr, uattr.user,
6484 BPF_MAP_LOOKUP_AND_DELETE_BATCH);
6485 break;
6486 case BPF_MAP_UPDATE_BATCH:
6487 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
6488 break;
6489 case BPF_MAP_DELETE_BATCH:
6490 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
6491 break;
6492 case BPF_LINK_CREATE:
6493 err = link_create(&attr, uattr);
6494 break;
6495 case BPF_LINK_UPDATE:
6496 err = link_update(&attr);
6497 break;
6498 case BPF_LINK_GET_FD_BY_ID:
6499 err = bpf_link_get_fd_by_id(&attr);
6500 break;
6501 case BPF_LINK_GET_NEXT_ID:
6502 err = bpf_obj_get_next_id(&attr, uattr.user,
6503 &link_idr, &link_idr_lock);
6504 break;
6505 case BPF_ENABLE_STATS:
6506 err = bpf_enable_stats(&attr);
6507 break;
6508 case BPF_ITER_CREATE:
6509 err = bpf_iter_create(&attr);
6510 break;
6511 case BPF_LINK_DETACH:
6512 err = link_detach(&attr);
6513 break;
6514 case BPF_PROG_BIND_MAP:
6515 err = bpf_prog_bind_map(&attr);
6516 break;
6517 case BPF_TOKEN_CREATE:
6518 err = token_create(&attr);
6519 break;
6520 case BPF_PROG_STREAM_READ_BY_FD:
6521 err = prog_stream_read(&attr);
6522 break;
6523 case BPF_PROG_ASSOC_STRUCT_OPS:
6524 err = prog_assoc_struct_ops(&attr);
6525 break;
6526 default:
6527 err = -EINVAL;
6528 break;
6529 }
6530
6531 return err;
6532}
6533
6534SYSCALL_DEFINE5(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size,
6535 struct bpf_common_attr __user *, uattr_common, unsigned int, size_common)
6536{
6537 return __sys_bpf(cmd, USER_BPFPTR(uattr), size, USER_BPFPTR(uattr_common), size_common);
6538}
6539
6540static bool syscall_prog_is_valid_access(int off, int size,
6541 enum bpf_access_type type,
6542 const struct bpf_prog *prog,
6543 struct bpf_insn_access_aux *info)
6544{
6545 if (off < 0 || off >= U16_MAX)
6546 return false;
6547 /* No alignment requirements for syscall ctx accesses. */
6548 return true;
6549}
6550
6551BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
6552{
6553 switch (cmd) {
6554 case BPF_MAP_CREATE:
6555 case BPF_MAP_DELETE_ELEM:
6556 case BPF_MAP_UPDATE_ELEM:
6557 case BPF_MAP_FREEZE:
6558 case BPF_MAP_GET_FD_BY_ID:
6559 case BPF_PROG_LOAD:
6560 case BPF_BTF_LOAD:
6561 case BPF_LINK_CREATE:
6562 case BPF_RAW_TRACEPOINT_OPEN:
6563 break;
6564 default:
6565 return -EINVAL;
6566 }
6567 return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size, KERNEL_BPFPTR(NULL), 0);
6568}
6569
6570
6571/* To shut up -Wmissing-prototypes.
6572 * This function is used by the kernel light skeleton
6573 * to load bpf programs when modules are loaded or during kernel boot.
6574 * See tools/lib/bpf/skel_internal.h
6575 */
6576int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
6577
6578int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
6579{
6580 struct bpf_prog * __maybe_unused prog;
6581 struct bpf_tramp_run_ctx __maybe_unused run_ctx;
6582
6583 switch (cmd) {
6584#ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
6585 case BPF_PROG_TEST_RUN:
6586 if (attr->test.data_in || attr->test.data_out ||
6587 attr->test.ctx_out || attr->test.duration ||
6588 attr->test.repeat || attr->test.flags)
6589 return -EINVAL;
6590
6591 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
6592 if (IS_ERR(prog))
6593 return PTR_ERR(prog);
6594
6595 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
6596 attr->test.ctx_size_in > U16_MAX) {
6597 bpf_prog_put(prog);
6598 return -EINVAL;
6599 }
6600
6601 run_ctx.bpf_cookie = 0;
6602 if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
6603 /* recursion detected */
6604 __bpf_prog_exit_sleepable_recur(prog, 0, &run_ctx);
6605 bpf_prog_put(prog);
6606 return -EBUSY;
6607 }
6608 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
6609 __bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
6610 &run_ctx);
6611 bpf_prog_put(prog);
6612 return 0;
6613#endif
6614 default:
6615 return ____bpf_sys_bpf(cmd, attr, size);
6616 }
6617}
6618EXPORT_SYMBOL_NS(kern_sys_bpf, "BPF_INTERNAL");
6619
6620static const struct bpf_func_proto bpf_sys_bpf_proto = {
6621 .func = bpf_sys_bpf,
6622 .gpl_only = false,
6623 .ret_type = RET_INTEGER,
6624 .arg1_type = ARG_ANYTHING,
6625 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6626 .arg3_type = ARG_CONST_SIZE,
6627};
6628
6629const struct bpf_func_proto * __weak
6630tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6631{
6632 return bpf_base_func_proto(func_id, prog);
6633}
6634
6635BPF_CALL_1(bpf_sys_close, u32, fd)
6636{
6637 /* When bpf program calls this helper there should not be
6638 * an fdget() without matching completed fdput().
6639 * This helper is allowed in the following callchain only:
6640 * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
6641 */
6642 return close_fd(fd);
6643}
6644
6645static const struct bpf_func_proto bpf_sys_close_proto = {
6646 .func = bpf_sys_close,
6647 .gpl_only = false,
6648 .ret_type = RET_INTEGER,
6649 .arg1_type = ARG_ANYTHING,
6650};
6651
6652BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
6653{
6654 *res = 0;
6655 if (flags)
6656 return -EINVAL;
6657
6658 if (name_sz <= 1 || name[name_sz - 1])
6659 return -EINVAL;
6660
6661 if (!bpf_dump_raw_ok(current_cred()))
6662 return -EPERM;
6663
6664 *res = kallsyms_lookup_name(name);
6665 return *res ? 0 : -ENOENT;
6666}
6667
6668static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
6669 .func = bpf_kallsyms_lookup_name,
6670 .gpl_only = false,
6671 .ret_type = RET_INTEGER,
6672 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6673 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
6674 .arg3_type = ARG_ANYTHING,
6675 .arg4_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_WRITE | MEM_ALIGNED,
6676 .arg4_size = sizeof(u64),
6677};
6678
6679static const struct bpf_func_proto *
6680syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6681{
6682 switch (func_id) {
6683 case BPF_FUNC_sys_bpf:
6684 return !bpf_token_capable(prog->aux->token, CAP_PERFMON)
6685 ? NULL : &bpf_sys_bpf_proto;
6686 case BPF_FUNC_btf_find_by_name_kind:
6687 return &bpf_btf_find_by_name_kind_proto;
6688 case BPF_FUNC_sys_close:
6689 return &bpf_sys_close_proto;
6690 case BPF_FUNC_kallsyms_lookup_name:
6691 return &bpf_kallsyms_lookup_name_proto;
6692 default:
6693 return tracing_prog_func_proto(func_id, prog);
6694 }
6695}
6696
6697const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
6698 .get_func_proto = syscall_prog_func_proto,
6699 .is_valid_access = syscall_prog_is_valid_access,
6700};
6701
6702const struct bpf_prog_ops bpf_syscall_prog_ops = {
6703 .test_run = bpf_prog_test_run_syscall,
6704};
6705
6706#ifdef CONFIG_SYSCTL
6707static int bpf_stats_handler(const struct ctl_table *table, int write,
6708 void *buffer, size_t *lenp, loff_t *ppos)
6709{
6710 struct static_key *key = (struct static_key *)table->data;
6711 static int saved_val;
6712 int val, ret;
6713 struct ctl_table tmp = {
6714 .data = &val,
6715 .maxlen = sizeof(val),
6716 .mode = table->mode,
6717 .extra1 = SYSCTL_ZERO,
6718 .extra2 = SYSCTL_ONE,
6719 };
6720
6721 if (write && !capable(CAP_SYS_ADMIN))
6722 return -EPERM;
6723
6724 mutex_lock(&bpf_stats_enabled_mutex);
6725 val = saved_val;
6726 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
6727 if (write && !ret && val != saved_val) {
6728 if (val)
6729 static_key_slow_inc(key);
6730 else
6731 static_key_slow_dec(key);
6732 saved_val = val;
6733 }
6734 mutex_unlock(&bpf_stats_enabled_mutex);
6735 return ret;
6736}
6737
6738void __weak unpriv_ebpf_notify(int new_state)
6739{
6740}
6741
6742static int bpf_unpriv_handler(const struct ctl_table *table, int write,
6743 void *buffer, size_t *lenp, loff_t *ppos)
6744{
6745 int ret, unpriv_enable = *(int *)table->data;
6746 bool locked_state = unpriv_enable == 1;
6747 struct ctl_table tmp = *table;
6748
6749 if (write && !capable(CAP_SYS_ADMIN))
6750 return -EPERM;
6751
6752 tmp.data = &unpriv_enable;
6753 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
6754 if (write && !ret) {
6755 if (locked_state && unpriv_enable != 1)
6756 return -EPERM;
6757 *(int *)table->data = unpriv_enable;
6758 }
6759
6760 if (write)
6761 unpriv_ebpf_notify(unpriv_enable);
6762
6763 return ret;
6764}
6765
6766static const struct ctl_table bpf_syscall_table[] = {
6767 {
6768 .procname = "unprivileged_bpf_disabled",
6769 .data = &sysctl_unprivileged_bpf_disabled,
6770 .maxlen = sizeof(sysctl_unprivileged_bpf_disabled),
6771 .mode = 0644,
6772 .proc_handler = bpf_unpriv_handler,
6773 .extra1 = SYSCTL_ZERO,
6774 .extra2 = SYSCTL_TWO,
6775 },
6776 {
6777 .procname = "bpf_stats_enabled",
6778 .data = &bpf_stats_enabled_key.key,
6779 .mode = 0644,
6780 .proc_handler = bpf_stats_handler,
6781 },
6782};
6783
6784static int __init bpf_syscall_sysctl_init(void)
6785{
6786 register_sysctl_init("kernel", bpf_syscall_table);
6787 return 0;
6788}
6789late_initcall(bpf_syscall_sysctl_init);
6790#endif /* CONFIG_SYSCTL */