projects/linux·blob v5.13 · refs

4511 lines · 106.6 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 */
4#include <linux/bpf.h>
5#include <linux/bpf_trace.h>
6#include <linux/bpf_lirc.h>
7#include <linux/bpf_verifier.h>
8#include <linux/btf.h>
9#include <linux/syscalls.h>
10#include <linux/slab.h>
11#include <linux/sched/signal.h>
12#include <linux/vmalloc.h>
13#include <linux/mmzone.h>
14#include <linux/anon_inodes.h>
15#include <linux/fdtable.h>
16#include <linux/file.h>
17#include <linux/fs.h>
18#include <linux/license.h>
19#include <linux/filter.h>
20#include <linux/kernel.h>
21#include <linux/idr.h>
22#include <linux/cred.h>
23#include <linux/timekeeping.h>
24#include <linux/ctype.h>
25#include <linux/nospec.h>
26#include <linux/audit.h>
27#include <uapi/linux/btf.h>
28#include <linux/pgtable.h>
29#include <linux/bpf_lsm.h>
30#include <linux/poll.h>
31#include <linux/bpf-netns.h>
32#include <linux/rcupdate_trace.h>
33#include <linux/memcontrol.h>
34
35#define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
36 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
37 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
38#define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
39#define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
40#define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
41 IS_FD_HASH(map))
42
43#define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
44
45DEFINE_PER_CPU(int, bpf_prog_active);
46static DEFINE_IDR(prog_idr);
47static DEFINE_SPINLOCK(prog_idr_lock);
48static DEFINE_IDR(map_idr);
49static DEFINE_SPINLOCK(map_idr_lock);
50static DEFINE_IDR(link_idr);
51static DEFINE_SPINLOCK(link_idr_lock);
52
53int sysctl_unprivileged_bpf_disabled __read_mostly =
54 IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
55
56static const struct bpf_map_ops * const bpf_map_types[] = {
57#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
58#define BPF_MAP_TYPE(_id, _ops) \
59 [_id] = &_ops,
60#define BPF_LINK_TYPE(_id, _name)
61#include <linux/bpf_types.h>
62#undef BPF_PROG_TYPE
63#undef BPF_MAP_TYPE
64#undef BPF_LINK_TYPE
65};
66
67/*
68 * If we're handed a bigger struct than we know of, ensure all the unknown bits
69 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
70 * we don't know about yet.
71 *
72 * There is a ToCToU between this function call and the following
73 * copy_from_user() call. However, this is not a concern since this function is
74 * meant to be a future-proofing of bits.
75 */
76int bpf_check_uarg_tail_zero(void __user *uaddr,
77 size_t expected_size,
78 size_t actual_size)
79{
80 unsigned char __user *addr = uaddr + expected_size;
81 int res;
82
83 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
84 return -E2BIG;
85
86 if (actual_size <= expected_size)
87 return 0;
88
89 res = check_zeroed_user(addr, actual_size - expected_size);
90 if (res < 0)
91 return res;
92 return res ? 0 : -E2BIG;
93}
94
95const struct bpf_map_ops bpf_map_offload_ops = {
96 .map_meta_equal = bpf_map_meta_equal,
97 .map_alloc = bpf_map_offload_map_alloc,
98 .map_free = bpf_map_offload_map_free,
99 .map_check_btf = map_check_no_btf,
100};
101
102static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
103{
104 const struct bpf_map_ops *ops;
105 u32 type = attr->map_type;
106 struct bpf_map *map;
107 int err;
108
109 if (type >= ARRAY_SIZE(bpf_map_types))
110 return ERR_PTR(-EINVAL);
111 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
112 ops = bpf_map_types[type];
113 if (!ops)
114 return ERR_PTR(-EINVAL);
115
116 if (ops->map_alloc_check) {
117 err = ops->map_alloc_check(attr);
118 if (err)
119 return ERR_PTR(err);
120 }
121 if (attr->map_ifindex)
122 ops = &bpf_map_offload_ops;
123 map = ops->map_alloc(attr);
124 if (IS_ERR(map))
125 return map;
126 map->ops = ops;
127 map->map_type = type;
128 return map;
129}
130
131static u32 bpf_map_value_size(const struct bpf_map *map)
132{
133 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
134 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
135 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
136 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
137 return round_up(map->value_size, 8) * num_possible_cpus();
138 else if (IS_FD_MAP(map))
139 return sizeof(u32);
140 else
141 return map->value_size;
142}
143
144static void maybe_wait_bpf_programs(struct bpf_map *map)
145{
146 /* Wait for any running BPF programs to complete so that
147 * userspace, when we return to it, knows that all programs
148 * that could be running use the new map value.
149 */
150 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
151 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
152 synchronize_rcu();
153}
154
155static int bpf_map_update_value(struct bpf_map *map, struct fd f, void *key,
156 void *value, __u64 flags)
157{
158 int err;
159
160 /* Need to create a kthread, thus must support schedule */
161 if (bpf_map_is_dev_bound(map)) {
162 return bpf_map_offload_update_elem(map, key, value, flags);
163 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
164 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
165 return map->ops->map_update_elem(map, key, value, flags);
166 } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
167 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
168 return sock_map_update_elem_sys(map, key, value, flags);
169 } else if (IS_FD_PROG_ARRAY(map)) {
170 return bpf_fd_array_map_update_elem(map, f.file, key, value,
171 flags);
172 }
173
174 bpf_disable_instrumentation();
175 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
176 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
177 err = bpf_percpu_hash_update(map, key, value, flags);
178 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
179 err = bpf_percpu_array_update(map, key, value, flags);
180 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
181 err = bpf_percpu_cgroup_storage_update(map, key, value,
182 flags);
183 } else if (IS_FD_ARRAY(map)) {
184 rcu_read_lock();
185 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
186 flags);
187 rcu_read_unlock();
188 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
189 rcu_read_lock();
190 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
191 flags);
192 rcu_read_unlock();
193 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
194 /* rcu_read_lock() is not needed */
195 err = bpf_fd_reuseport_array_update_elem(map, key, value,
196 flags);
197 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
198 map->map_type == BPF_MAP_TYPE_STACK) {
199 err = map->ops->map_push_elem(map, value, flags);
200 } else {
201 rcu_read_lock();
202 err = map->ops->map_update_elem(map, key, value, flags);
203 rcu_read_unlock();
204 }
205 bpf_enable_instrumentation();
206 maybe_wait_bpf_programs(map);
207
208 return err;
209}
210
211static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
212 __u64 flags)
213{
214 void *ptr;
215 int err;
216
217 if (bpf_map_is_dev_bound(map))
218 return bpf_map_offload_lookup_elem(map, key, value);
219
220 bpf_disable_instrumentation();
221 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
222 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
223 err = bpf_percpu_hash_copy(map, key, value);
224 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
225 err = bpf_percpu_array_copy(map, key, value);
226 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
227 err = bpf_percpu_cgroup_storage_copy(map, key, value);
228 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
229 err = bpf_stackmap_copy(map, key, value);
230 } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
231 err = bpf_fd_array_map_lookup_elem(map, key, value);
232 } else if (IS_FD_HASH(map)) {
233 err = bpf_fd_htab_map_lookup_elem(map, key, value);
234 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
235 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
236 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
237 map->map_type == BPF_MAP_TYPE_STACK) {
238 err = map->ops->map_peek_elem(map, value);
239 } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
240 /* struct_ops map requires directly updating "value" */
241 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
242 } else {
243 rcu_read_lock();
244 if (map->ops->map_lookup_elem_sys_only)
245 ptr = map->ops->map_lookup_elem_sys_only(map, key);
246 else
247 ptr = map->ops->map_lookup_elem(map, key);
248 if (IS_ERR(ptr)) {
249 err = PTR_ERR(ptr);
250 } else if (!ptr) {
251 err = -ENOENT;
252 } else {
253 err = 0;
254 if (flags & BPF_F_LOCK)
255 /* lock 'ptr' and copy everything but lock */
256 copy_map_value_locked(map, value, ptr, true);
257 else
258 copy_map_value(map, value, ptr);
259 /* mask lock, since value wasn't zero inited */
260 check_and_init_map_lock(map, value);
261 }
262 rcu_read_unlock();
263 }
264
265 bpf_enable_instrumentation();
266 maybe_wait_bpf_programs(map);
267
268 return err;
269}
270
271/* Please, do not use this function outside from the map creation path
272 * (e.g. in map update path) without taking care of setting the active
273 * memory cgroup (see at bpf_map_kmalloc_node() for example).
274 */
275static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
276{
277 /* We really just want to fail instead of triggering OOM killer
278 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
279 * which is used for lower order allocation requests.
280 *
281 * It has been observed that higher order allocation requests done by
282 * vmalloc with __GFP_NORETRY being set might fail due to not trying
283 * to reclaim memory from the page cache, thus we set
284 * __GFP_RETRY_MAYFAIL to avoid such situations.
285 */
286
287 const gfp_t gfp = __GFP_NOWARN | __GFP_ZERO | __GFP_ACCOUNT;
288 unsigned int flags = 0;
289 unsigned long align = 1;
290 void *area;
291
292 if (size >= SIZE_MAX)
293 return NULL;
294
295 /* kmalloc()'ed memory can't be mmap()'ed */
296 if (mmapable) {
297 BUG_ON(!PAGE_ALIGNED(size));
298 align = SHMLBA;
299 flags = VM_USERMAP;
300 } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
301 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
302 numa_node);
303 if (area != NULL)
304 return area;
305 }
306
307 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
308 gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
309 flags, numa_node, __builtin_return_address(0));
310}
311
312void *bpf_map_area_alloc(u64 size, int numa_node)
313{
314 return __bpf_map_area_alloc(size, numa_node, false);
315}
316
317void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
318{
319 return __bpf_map_area_alloc(size, numa_node, true);
320}
321
322void bpf_map_area_free(void *area)
323{
324 kvfree(area);
325}
326
327static u32 bpf_map_flags_retain_permanent(u32 flags)
328{
329 /* Some map creation flags are not tied to the map object but
330 * rather to the map fd instead, so they have no meaning upon
331 * map object inspection since multiple file descriptors with
332 * different (access) properties can exist here. Thus, given
333 * this has zero meaning for the map itself, lets clear these
334 * from here.
335 */
336 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
337}
338
339void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
340{
341 map->map_type = attr->map_type;
342 map->key_size = attr->key_size;
343 map->value_size = attr->value_size;
344 map->max_entries = attr->max_entries;
345 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
346 map->numa_node = bpf_map_attr_numa_node(attr);
347}
348
349static int bpf_map_alloc_id(struct bpf_map *map)
350{
351 int id;
352
353 idr_preload(GFP_KERNEL);
354 spin_lock_bh(&map_idr_lock);
355 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
356 if (id > 0)
357 map->id = id;
358 spin_unlock_bh(&map_idr_lock);
359 idr_preload_end();
360
361 if (WARN_ON_ONCE(!id))
362 return -ENOSPC;
363
364 return id > 0 ? 0 : id;
365}
366
367void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
368{
369 unsigned long flags;
370
371 /* Offloaded maps are removed from the IDR store when their device
372 * disappears - even if someone holds an fd to them they are unusable,
373 * the memory is gone, all ops will fail; they are simply waiting for
374 * refcnt to drop to be freed.
375 */
376 if (!map->id)
377 return;
378
379 if (do_idr_lock)
380 spin_lock_irqsave(&map_idr_lock, flags);
381 else
382 __acquire(&map_idr_lock);
383
384 idr_remove(&map_idr, map->id);
385 map->id = 0;
386
387 if (do_idr_lock)
388 spin_unlock_irqrestore(&map_idr_lock, flags);
389 else
390 __release(&map_idr_lock);
391}
392
393#ifdef CONFIG_MEMCG_KMEM
394static void bpf_map_save_memcg(struct bpf_map *map)
395{
396 map->memcg = get_mem_cgroup_from_mm(current->mm);
397}
398
399static void bpf_map_release_memcg(struct bpf_map *map)
400{
401 mem_cgroup_put(map->memcg);
402}
403
404void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
405 int node)
406{
407 struct mem_cgroup *old_memcg;
408 void *ptr;
409
410 old_memcg = set_active_memcg(map->memcg);
411 ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
412 set_active_memcg(old_memcg);
413
414 return ptr;
415}
416
417void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
418{
419 struct mem_cgroup *old_memcg;
420 void *ptr;
421
422 old_memcg = set_active_memcg(map->memcg);
423 ptr = kzalloc(size, flags | __GFP_ACCOUNT);
424 set_active_memcg(old_memcg);
425
426 return ptr;
427}
428
429void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
430 size_t align, gfp_t flags)
431{
432 struct mem_cgroup *old_memcg;
433 void __percpu *ptr;
434
435 old_memcg = set_active_memcg(map->memcg);
436 ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
437 set_active_memcg(old_memcg);
438
439 return ptr;
440}
441
442#else
443static void bpf_map_save_memcg(struct bpf_map *map)
444{
445}
446
447static void bpf_map_release_memcg(struct bpf_map *map)
448{
449}
450#endif
451
452/* called from workqueue */
453static void bpf_map_free_deferred(struct work_struct *work)
454{
455 struct bpf_map *map = container_of(work, struct bpf_map, work);
456
457 security_bpf_map_free(map);
458 bpf_map_release_memcg(map);
459 /* implementation dependent freeing */
460 map->ops->map_free(map);
461}
462
463static void bpf_map_put_uref(struct bpf_map *map)
464{
465 if (atomic64_dec_and_test(&map->usercnt)) {
466 if (map->ops->map_release_uref)
467 map->ops->map_release_uref(map);
468 }
469}
470
471/* decrement map refcnt and schedule it for freeing via workqueue
472 * (unrelying map implementation ops->map_free() might sleep)
473 */
474static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
475{
476 if (atomic64_dec_and_test(&map->refcnt)) {
477 /* bpf_map_free_id() must be called first */
478 bpf_map_free_id(map, do_idr_lock);
479 btf_put(map->btf);
480 INIT_WORK(&map->work, bpf_map_free_deferred);
481 schedule_work(&map->work);
482 }
483}
484
485void bpf_map_put(struct bpf_map *map)
486{
487 __bpf_map_put(map, true);
488}
489EXPORT_SYMBOL_GPL(bpf_map_put);
490
491void bpf_map_put_with_uref(struct bpf_map *map)
492{
493 bpf_map_put_uref(map);
494 bpf_map_put(map);
495}
496
497static int bpf_map_release(struct inode *inode, struct file *filp)
498{
499 struct bpf_map *map = filp->private_data;
500
501 if (map->ops->map_release)
502 map->ops->map_release(map, filp);
503
504 bpf_map_put_with_uref(map);
505 return 0;
506}
507
508static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
509{
510 fmode_t mode = f.file->f_mode;
511
512 /* Our file permissions may have been overridden by global
513 * map permissions facing syscall side.
514 */
515 if (READ_ONCE(map->frozen))
516 mode &= ~FMODE_CAN_WRITE;
517 return mode;
518}
519
520#ifdef CONFIG_PROC_FS
521/* Provides an approximation of the map's memory footprint.
522 * Used only to provide a backward compatibility and display
523 * a reasonable "memlock" info.
524 */
525static unsigned long bpf_map_memory_footprint(const struct bpf_map *map)
526{
527 unsigned long size;
528
529 size = round_up(map->key_size + bpf_map_value_size(map), 8);
530
531 return round_up(map->max_entries * size, PAGE_SIZE);
532}
533
534static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
535{
536 const struct bpf_map *map = filp->private_data;
537 const struct bpf_array *array;
538 u32 type = 0, jited = 0;
539
540 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
541 array = container_of(map, struct bpf_array, map);
542 type = array->aux->type;
543 jited = array->aux->jited;
544 }
545
546 seq_printf(m,
547 "map_type:\t%u\n"
548 "key_size:\t%u\n"
549 "value_size:\t%u\n"
550 "max_entries:\t%u\n"
551 "map_flags:\t%#x\n"
552 "memlock:\t%lu\n"
553 "map_id:\t%u\n"
554 "frozen:\t%u\n",
555 map->map_type,
556 map->key_size,
557 map->value_size,
558 map->max_entries,
559 map->map_flags,
560 bpf_map_memory_footprint(map),
561 map->id,
562 READ_ONCE(map->frozen));
563 if (type) {
564 seq_printf(m, "owner_prog_type:\t%u\n", type);
565 seq_printf(m, "owner_jited:\t%u\n", jited);
566 }
567}
568#endif
569
570static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
571 loff_t *ppos)
572{
573 /* We need this handler such that alloc_file() enables
574 * f_mode with FMODE_CAN_READ.
575 */
576 return -EINVAL;
577}
578
579static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
580 size_t siz, loff_t *ppos)
581{
582 /* We need this handler such that alloc_file() enables
583 * f_mode with FMODE_CAN_WRITE.
584 */
585 return -EINVAL;
586}
587
588/* called for any extra memory-mapped regions (except initial) */
589static void bpf_map_mmap_open(struct vm_area_struct *vma)
590{
591 struct bpf_map *map = vma->vm_file->private_data;
592
593 if (vma->vm_flags & VM_MAYWRITE) {
594 mutex_lock(&map->freeze_mutex);
595 map->writecnt++;
596 mutex_unlock(&map->freeze_mutex);
597 }
598}
599
600/* called for all unmapped memory region (including initial) */
601static void bpf_map_mmap_close(struct vm_area_struct *vma)
602{
603 struct bpf_map *map = vma->vm_file->private_data;
604
605 if (vma->vm_flags & VM_MAYWRITE) {
606 mutex_lock(&map->freeze_mutex);
607 map->writecnt--;
608 mutex_unlock(&map->freeze_mutex);
609 }
610}
611
612static const struct vm_operations_struct bpf_map_default_vmops = {
613 .open = bpf_map_mmap_open,
614 .close = bpf_map_mmap_close,
615};
616
617static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
618{
619 struct bpf_map *map = filp->private_data;
620 int err;
621
622 if (!map->ops->map_mmap || map_value_has_spin_lock(map))
623 return -ENOTSUPP;
624
625 if (!(vma->vm_flags & VM_SHARED))
626 return -EINVAL;
627
628 mutex_lock(&map->freeze_mutex);
629
630 if (vma->vm_flags & VM_WRITE) {
631 if (map->frozen) {
632 err = -EPERM;
633 goto out;
634 }
635 /* map is meant to be read-only, so do not allow mapping as
636 * writable, because it's possible to leak a writable page
637 * reference and allows user-space to still modify it after
638 * freezing, while verifier will assume contents do not change
639 */
640 if (map->map_flags & BPF_F_RDONLY_PROG) {
641 err = -EACCES;
642 goto out;
643 }
644 }
645
646 /* set default open/close callbacks */
647 vma->vm_ops = &bpf_map_default_vmops;
648 vma->vm_private_data = map;
649 vma->vm_flags &= ~VM_MAYEXEC;
650 if (!(vma->vm_flags & VM_WRITE))
651 /* disallow re-mapping with PROT_WRITE */
652 vma->vm_flags &= ~VM_MAYWRITE;
653
654 err = map->ops->map_mmap(map, vma);
655 if (err)
656 goto out;
657
658 if (vma->vm_flags & VM_MAYWRITE)
659 map->writecnt++;
660out:
661 mutex_unlock(&map->freeze_mutex);
662 return err;
663}
664
665static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
666{
667 struct bpf_map *map = filp->private_data;
668
669 if (map->ops->map_poll)
670 return map->ops->map_poll(map, filp, pts);
671
672 return EPOLLERR;
673}
674
675const struct file_operations bpf_map_fops = {
676#ifdef CONFIG_PROC_FS
677 .show_fdinfo = bpf_map_show_fdinfo,
678#endif
679 .release = bpf_map_release,
680 .read = bpf_dummy_read,
681 .write = bpf_dummy_write,
682 .mmap = bpf_map_mmap,
683 .poll = bpf_map_poll,
684};
685
686int bpf_map_new_fd(struct bpf_map *map, int flags)
687{
688 int ret;
689
690 ret = security_bpf_map(map, OPEN_FMODE(flags));
691 if (ret < 0)
692 return ret;
693
694 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
695 flags | O_CLOEXEC);
696}
697
698int bpf_get_file_flag(int flags)
699{
700 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
701 return -EINVAL;
702 if (flags & BPF_F_RDONLY)
703 return O_RDONLY;
704 if (flags & BPF_F_WRONLY)
705 return O_WRONLY;
706 return O_RDWR;
707}
708
709/* helper macro to check that unused fields 'union bpf_attr' are zero */
710#define CHECK_ATTR(CMD) \
711 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
712 sizeof(attr->CMD##_LAST_FIELD), 0, \
713 sizeof(*attr) - \
714 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
715 sizeof(attr->CMD##_LAST_FIELD)) != NULL
716
717/* dst and src must have at least "size" number of bytes.
718 * Return strlen on success and < 0 on error.
719 */
720int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
721{
722 const char *end = src + size;
723 const char *orig_src = src;
724
725 memset(dst, 0, size);
726 /* Copy all isalnum(), '_' and '.' chars. */
727 while (src < end && *src) {
728 if (!isalnum(*src) &&
729 *src != '_' && *src != '.')
730 return -EINVAL;
731 *dst++ = *src++;
732 }
733
734 /* No '\0' found in "size" number of bytes */
735 if (src == end)
736 return -EINVAL;
737
738 return src - orig_src;
739}
740
741int map_check_no_btf(const struct bpf_map *map,
742 const struct btf *btf,
743 const struct btf_type *key_type,
744 const struct btf_type *value_type)
745{
746 return -ENOTSUPP;
747}
748
749static int map_check_btf(struct bpf_map *map, const struct btf *btf,
750 u32 btf_key_id, u32 btf_value_id)
751{
752 const struct btf_type *key_type, *value_type;
753 u32 key_size, value_size;
754 int ret = 0;
755
756 /* Some maps allow key to be unspecified. */
757 if (btf_key_id) {
758 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
759 if (!key_type || key_size != map->key_size)
760 return -EINVAL;
761 } else {
762 key_type = btf_type_by_id(btf, 0);
763 if (!map->ops->map_check_btf)
764 return -EINVAL;
765 }
766
767 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
768 if (!value_type || value_size != map->value_size)
769 return -EINVAL;
770
771 map->spin_lock_off = btf_find_spin_lock(btf, value_type);
772
773 if (map_value_has_spin_lock(map)) {
774 if (map->map_flags & BPF_F_RDONLY_PROG)
775 return -EACCES;
776 if (map->map_type != BPF_MAP_TYPE_HASH &&
777 map->map_type != BPF_MAP_TYPE_ARRAY &&
778 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
779 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
780 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
781 map->map_type != BPF_MAP_TYPE_TASK_STORAGE)
782 return -ENOTSUPP;
783 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
784 map->value_size) {
785 WARN_ONCE(1,
786 "verifier bug spin_lock_off %d value_size %d\n",
787 map->spin_lock_off, map->value_size);
788 return -EFAULT;
789 }
790 }
791
792 if (map->ops->map_check_btf)
793 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
794
795 return ret;
796}
797
798#define BPF_MAP_CREATE_LAST_FIELD btf_vmlinux_value_type_id
799/* called via syscall */
800static int map_create(union bpf_attr *attr)
801{
802 int numa_node = bpf_map_attr_numa_node(attr);
803 struct bpf_map *map;
804 int f_flags;
805 int err;
806
807 err = CHECK_ATTR(BPF_MAP_CREATE);
808 if (err)
809 return -EINVAL;
810
811 if (attr->btf_vmlinux_value_type_id) {
812 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
813 attr->btf_key_type_id || attr->btf_value_type_id)
814 return -EINVAL;
815 } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
816 return -EINVAL;
817 }
818
819 f_flags = bpf_get_file_flag(attr->map_flags);
820 if (f_flags < 0)
821 return f_flags;
822
823 if (numa_node != NUMA_NO_NODE &&
824 ((unsigned int)numa_node >= nr_node_ids ||
825 !node_online(numa_node)))
826 return -EINVAL;
827
828 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
829 map = find_and_alloc_map(attr);
830 if (IS_ERR(map))
831 return PTR_ERR(map);
832
833 err = bpf_obj_name_cpy(map->name, attr->map_name,
834 sizeof(attr->map_name));
835 if (err < 0)
836 goto free_map;
837
838 atomic64_set(&map->refcnt, 1);
839 atomic64_set(&map->usercnt, 1);
840 mutex_init(&map->freeze_mutex);
841
842 map->spin_lock_off = -EINVAL;
843 if (attr->btf_key_type_id || attr->btf_value_type_id ||
844 /* Even the map's value is a kernel's struct,
845 * the bpf_prog.o must have BTF to begin with
846 * to figure out the corresponding kernel's
847 * counter part. Thus, attr->btf_fd has
848 * to be valid also.
849 */
850 attr->btf_vmlinux_value_type_id) {
851 struct btf *btf;
852
853 btf = btf_get_by_fd(attr->btf_fd);
854 if (IS_ERR(btf)) {
855 err = PTR_ERR(btf);
856 goto free_map;
857 }
858 if (btf_is_kernel(btf)) {
859 btf_put(btf);
860 err = -EACCES;
861 goto free_map;
862 }
863 map->btf = btf;
864
865 if (attr->btf_value_type_id) {
866 err = map_check_btf(map, btf, attr->btf_key_type_id,
867 attr->btf_value_type_id);
868 if (err)
869 goto free_map;
870 }
871
872 map->btf_key_type_id = attr->btf_key_type_id;
873 map->btf_value_type_id = attr->btf_value_type_id;
874 map->btf_vmlinux_value_type_id =
875 attr->btf_vmlinux_value_type_id;
876 }
877
878 err = security_bpf_map_alloc(map);
879 if (err)
880 goto free_map;
881
882 err = bpf_map_alloc_id(map);
883 if (err)
884 goto free_map_sec;
885
886 bpf_map_save_memcg(map);
887
888 err = bpf_map_new_fd(map, f_flags);
889 if (err < 0) {
890 /* failed to allocate fd.
891 * bpf_map_put_with_uref() is needed because the above
892 * bpf_map_alloc_id() has published the map
893 * to the userspace and the userspace may
894 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
895 */
896 bpf_map_put_with_uref(map);
897 return err;
898 }
899
900 return err;
901
902free_map_sec:
903 security_bpf_map_free(map);
904free_map:
905 btf_put(map->btf);
906 map->ops->map_free(map);
907 return err;
908}
909
910/* if error is returned, fd is released.
911 * On success caller should complete fd access with matching fdput()
912 */
913struct bpf_map *__bpf_map_get(struct fd f)
914{
915 if (!f.file)
916 return ERR_PTR(-EBADF);
917 if (f.file->f_op != &bpf_map_fops) {
918 fdput(f);
919 return ERR_PTR(-EINVAL);
920 }
921
922 return f.file->private_data;
923}
924
925void bpf_map_inc(struct bpf_map *map)
926{
927 atomic64_inc(&map->refcnt);
928}
929EXPORT_SYMBOL_GPL(bpf_map_inc);
930
931void bpf_map_inc_with_uref(struct bpf_map *map)
932{
933 atomic64_inc(&map->refcnt);
934 atomic64_inc(&map->usercnt);
935}
936EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
937
938struct bpf_map *bpf_map_get(u32 ufd)
939{
940 struct fd f = fdget(ufd);
941 struct bpf_map *map;
942
943 map = __bpf_map_get(f);
944 if (IS_ERR(map))
945 return map;
946
947 bpf_map_inc(map);
948 fdput(f);
949
950 return map;
951}
952
953struct bpf_map *bpf_map_get_with_uref(u32 ufd)
954{
955 struct fd f = fdget(ufd);
956 struct bpf_map *map;
957
958 map = __bpf_map_get(f);
959 if (IS_ERR(map))
960 return map;
961
962 bpf_map_inc_with_uref(map);
963 fdput(f);
964
965 return map;
966}
967
968/* map_idr_lock should have been held */
969static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
970{
971 int refold;
972
973 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
974 if (!refold)
975 return ERR_PTR(-ENOENT);
976 if (uref)
977 atomic64_inc(&map->usercnt);
978
979 return map;
980}
981
982struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
983{
984 spin_lock_bh(&map_idr_lock);
985 map = __bpf_map_inc_not_zero(map, false);
986 spin_unlock_bh(&map_idr_lock);
987
988 return map;
989}
990EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
991
992int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
993{
994 return -ENOTSUPP;
995}
996
997static void *__bpf_copy_key(void __user *ukey, u64 key_size)
998{
999 if (key_size)
1000 return memdup_user(ukey, key_size);
1001
1002 if (ukey)
1003 return ERR_PTR(-EINVAL);
1004
1005 return NULL;
1006}
1007
1008/* last field in 'union bpf_attr' used by this command */
1009#define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1010
1011static int map_lookup_elem(union bpf_attr *attr)
1012{
1013 void __user *ukey = u64_to_user_ptr(attr->key);
1014 void __user *uvalue = u64_to_user_ptr(attr->value);
1015 int ufd = attr->map_fd;
1016 struct bpf_map *map;
1017 void *key, *value;
1018 u32 value_size;
1019 struct fd f;
1020 int err;
1021
1022 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1023 return -EINVAL;
1024
1025 if (attr->flags & ~BPF_F_LOCK)
1026 return -EINVAL;
1027
1028 f = fdget(ufd);
1029 map = __bpf_map_get(f);
1030 if (IS_ERR(map))
1031 return PTR_ERR(map);
1032 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1033 err = -EPERM;
1034 goto err_put;
1035 }
1036
1037 if ((attr->flags & BPF_F_LOCK) &&
1038 !map_value_has_spin_lock(map)) {
1039 err = -EINVAL;
1040 goto err_put;
1041 }
1042
1043 key = __bpf_copy_key(ukey, map->key_size);
1044 if (IS_ERR(key)) {
1045 err = PTR_ERR(key);
1046 goto err_put;
1047 }
1048
1049 value_size = bpf_map_value_size(map);
1050
1051 err = -ENOMEM;
1052 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1053 if (!value)
1054 goto free_key;
1055
1056 err = bpf_map_copy_value(map, key, value, attr->flags);
1057 if (err)
1058 goto free_value;
1059
1060 err = -EFAULT;
1061 if (copy_to_user(uvalue, value, value_size) != 0)
1062 goto free_value;
1063
1064 err = 0;
1065
1066free_value:
1067 kfree(value);
1068free_key:
1069 kfree(key);
1070err_put:
1071 fdput(f);
1072 return err;
1073}
1074
1075
1076#define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1077
1078static int map_update_elem(union bpf_attr *attr)
1079{
1080 void __user *ukey = u64_to_user_ptr(attr->key);
1081 void __user *uvalue = u64_to_user_ptr(attr->value);
1082 int ufd = attr->map_fd;
1083 struct bpf_map *map;
1084 void *key, *value;
1085 u32 value_size;
1086 struct fd f;
1087 int err;
1088
1089 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1090 return -EINVAL;
1091
1092 f = fdget(ufd);
1093 map = __bpf_map_get(f);
1094 if (IS_ERR(map))
1095 return PTR_ERR(map);
1096 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1097 err = -EPERM;
1098 goto err_put;
1099 }
1100
1101 if ((attr->flags & BPF_F_LOCK) &&
1102 !map_value_has_spin_lock(map)) {
1103 err = -EINVAL;
1104 goto err_put;
1105 }
1106
1107 key = __bpf_copy_key(ukey, map->key_size);
1108 if (IS_ERR(key)) {
1109 err = PTR_ERR(key);
1110 goto err_put;
1111 }
1112
1113 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1114 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
1115 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
1116 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
1117 value_size = round_up(map->value_size, 8) * num_possible_cpus();
1118 else
1119 value_size = map->value_size;
1120
1121 err = -ENOMEM;
1122 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1123 if (!value)
1124 goto free_key;
1125
1126 err = -EFAULT;
1127 if (copy_from_user(value, uvalue, value_size) != 0)
1128 goto free_value;
1129
1130 err = bpf_map_update_value(map, f, key, value, attr->flags);
1131
1132free_value:
1133 kfree(value);
1134free_key:
1135 kfree(key);
1136err_put:
1137 fdput(f);
1138 return err;
1139}
1140
1141#define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1142
1143static int map_delete_elem(union bpf_attr *attr)
1144{
1145 void __user *ukey = u64_to_user_ptr(attr->key);
1146 int ufd = attr->map_fd;
1147 struct bpf_map *map;
1148 struct fd f;
1149 void *key;
1150 int err;
1151
1152 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1153 return -EINVAL;
1154
1155 f = fdget(ufd);
1156 map = __bpf_map_get(f);
1157 if (IS_ERR(map))
1158 return PTR_ERR(map);
1159 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1160 err = -EPERM;
1161 goto err_put;
1162 }
1163
1164 key = __bpf_copy_key(ukey, map->key_size);
1165 if (IS_ERR(key)) {
1166 err = PTR_ERR(key);
1167 goto err_put;
1168 }
1169
1170 if (bpf_map_is_dev_bound(map)) {
1171 err = bpf_map_offload_delete_elem(map, key);
1172 goto out;
1173 } else if (IS_FD_PROG_ARRAY(map) ||
1174 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1175 /* These maps require sleepable context */
1176 err = map->ops->map_delete_elem(map, key);
1177 goto out;
1178 }
1179
1180 bpf_disable_instrumentation();
1181 rcu_read_lock();
1182 err = map->ops->map_delete_elem(map, key);
1183 rcu_read_unlock();
1184 bpf_enable_instrumentation();
1185 maybe_wait_bpf_programs(map);
1186out:
1187 kfree(key);
1188err_put:
1189 fdput(f);
1190 return err;
1191}
1192
1193/* last field in 'union bpf_attr' used by this command */
1194#define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1195
1196static int map_get_next_key(union bpf_attr *attr)
1197{
1198 void __user *ukey = u64_to_user_ptr(attr->key);
1199 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1200 int ufd = attr->map_fd;
1201 struct bpf_map *map;
1202 void *key, *next_key;
1203 struct fd f;
1204 int err;
1205
1206 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1207 return -EINVAL;
1208
1209 f = fdget(ufd);
1210 map = __bpf_map_get(f);
1211 if (IS_ERR(map))
1212 return PTR_ERR(map);
1213 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1214 err = -EPERM;
1215 goto err_put;
1216 }
1217
1218 if (ukey) {
1219 key = __bpf_copy_key(ukey, map->key_size);
1220 if (IS_ERR(key)) {
1221 err = PTR_ERR(key);
1222 goto err_put;
1223 }
1224 } else {
1225 key = NULL;
1226 }
1227
1228 err = -ENOMEM;
1229 next_key = kmalloc(map->key_size, GFP_USER);
1230 if (!next_key)
1231 goto free_key;
1232
1233 if (bpf_map_is_dev_bound(map)) {
1234 err = bpf_map_offload_get_next_key(map, key, next_key);
1235 goto out;
1236 }
1237
1238 rcu_read_lock();
1239 err = map->ops->map_get_next_key(map, key, next_key);
1240 rcu_read_unlock();
1241out:
1242 if (err)
1243 goto free_next_key;
1244
1245 err = -EFAULT;
1246 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1247 goto free_next_key;
1248
1249 err = 0;
1250
1251free_next_key:
1252 kfree(next_key);
1253free_key:
1254 kfree(key);
1255err_put:
1256 fdput(f);
1257 return err;
1258}
1259
1260int generic_map_delete_batch(struct bpf_map *map,
1261 const union bpf_attr *attr,
1262 union bpf_attr __user *uattr)
1263{
1264 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1265 u32 cp, max_count;
1266 int err = 0;
1267 void *key;
1268
1269 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1270 return -EINVAL;
1271
1272 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1273 !map_value_has_spin_lock(map)) {
1274 return -EINVAL;
1275 }
1276
1277 max_count = attr->batch.count;
1278 if (!max_count)
1279 return 0;
1280
1281 key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1282 if (!key)
1283 return -ENOMEM;
1284
1285 for (cp = 0; cp < max_count; cp++) {
1286 err = -EFAULT;
1287 if (copy_from_user(key, keys + cp * map->key_size,
1288 map->key_size))
1289 break;
1290
1291 if (bpf_map_is_dev_bound(map)) {
1292 err = bpf_map_offload_delete_elem(map, key);
1293 break;
1294 }
1295
1296 bpf_disable_instrumentation();
1297 rcu_read_lock();
1298 err = map->ops->map_delete_elem(map, key);
1299 rcu_read_unlock();
1300 bpf_enable_instrumentation();
1301 maybe_wait_bpf_programs(map);
1302 if (err)
1303 break;
1304 }
1305 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1306 err = -EFAULT;
1307
1308 kfree(key);
1309 return err;
1310}
1311
1312int generic_map_update_batch(struct bpf_map *map,
1313 const union bpf_attr *attr,
1314 union bpf_attr __user *uattr)
1315{
1316 void __user *values = u64_to_user_ptr(attr->batch.values);
1317 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1318 u32 value_size, cp, max_count;
1319 int ufd = attr->map_fd;
1320 void *key, *value;
1321 struct fd f;
1322 int err = 0;
1323
1324 f = fdget(ufd);
1325 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1326 return -EINVAL;
1327
1328 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1329 !map_value_has_spin_lock(map)) {
1330 return -EINVAL;
1331 }
1332
1333 value_size = bpf_map_value_size(map);
1334
1335 max_count = attr->batch.count;
1336 if (!max_count)
1337 return 0;
1338
1339 key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1340 if (!key)
1341 return -ENOMEM;
1342
1343 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1344 if (!value) {
1345 kfree(key);
1346 return -ENOMEM;
1347 }
1348
1349 for (cp = 0; cp < max_count; cp++) {
1350 err = -EFAULT;
1351 if (copy_from_user(key, keys + cp * map->key_size,
1352 map->key_size) ||
1353 copy_from_user(value, values + cp * value_size, value_size))
1354 break;
1355
1356 err = bpf_map_update_value(map, f, key, value,
1357 attr->batch.elem_flags);
1358
1359 if (err)
1360 break;
1361 }
1362
1363 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1364 err = -EFAULT;
1365
1366 kfree(value);
1367 kfree(key);
1368 return err;
1369}
1370
1371#define MAP_LOOKUP_RETRIES 3
1372
1373int generic_map_lookup_batch(struct bpf_map *map,
1374 const union bpf_attr *attr,
1375 union bpf_attr __user *uattr)
1376{
1377 void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1378 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1379 void __user *values = u64_to_user_ptr(attr->batch.values);
1380 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1381 void *buf, *buf_prevkey, *prev_key, *key, *value;
1382 int err, retry = MAP_LOOKUP_RETRIES;
1383 u32 value_size, cp, max_count;
1384
1385 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1386 return -EINVAL;
1387
1388 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1389 !map_value_has_spin_lock(map))
1390 return -EINVAL;
1391
1392 value_size = bpf_map_value_size(map);
1393
1394 max_count = attr->batch.count;
1395 if (!max_count)
1396 return 0;
1397
1398 if (put_user(0, &uattr->batch.count))
1399 return -EFAULT;
1400
1401 buf_prevkey = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1402 if (!buf_prevkey)
1403 return -ENOMEM;
1404
1405 buf = kmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1406 if (!buf) {
1407 kfree(buf_prevkey);
1408 return -ENOMEM;
1409 }
1410
1411 err = -EFAULT;
1412 prev_key = NULL;
1413 if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1414 goto free_buf;
1415 key = buf;
1416 value = key + map->key_size;
1417 if (ubatch)
1418 prev_key = buf_prevkey;
1419
1420 for (cp = 0; cp < max_count;) {
1421 rcu_read_lock();
1422 err = map->ops->map_get_next_key(map, prev_key, key);
1423 rcu_read_unlock();
1424 if (err)
1425 break;
1426 err = bpf_map_copy_value(map, key, value,
1427 attr->batch.elem_flags);
1428
1429 if (err == -ENOENT) {
1430 if (retry) {
1431 retry--;
1432 continue;
1433 }
1434 err = -EINTR;
1435 break;
1436 }
1437
1438 if (err)
1439 goto free_buf;
1440
1441 if (copy_to_user(keys + cp * map->key_size, key,
1442 map->key_size)) {
1443 err = -EFAULT;
1444 goto free_buf;
1445 }
1446 if (copy_to_user(values + cp * value_size, value, value_size)) {
1447 err = -EFAULT;
1448 goto free_buf;
1449 }
1450
1451 if (!prev_key)
1452 prev_key = buf_prevkey;
1453
1454 swap(prev_key, key);
1455 retry = MAP_LOOKUP_RETRIES;
1456 cp++;
1457 }
1458
1459 if (err == -EFAULT)
1460 goto free_buf;
1461
1462 if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1463 (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1464 err = -EFAULT;
1465
1466free_buf:
1467 kfree(buf_prevkey);
1468 kfree(buf);
1469 return err;
1470}
1471
1472#define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1473
1474static int map_lookup_and_delete_elem(union bpf_attr *attr)
1475{
1476 void __user *ukey = u64_to_user_ptr(attr->key);
1477 void __user *uvalue = u64_to_user_ptr(attr->value);
1478 int ufd = attr->map_fd;
1479 struct bpf_map *map;
1480 void *key, *value;
1481 u32 value_size;
1482 struct fd f;
1483 int err;
1484
1485 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1486 return -EINVAL;
1487
1488 f = fdget(ufd);
1489 map = __bpf_map_get(f);
1490 if (IS_ERR(map))
1491 return PTR_ERR(map);
1492 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1493 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1494 err = -EPERM;
1495 goto err_put;
1496 }
1497
1498 key = __bpf_copy_key(ukey, map->key_size);
1499 if (IS_ERR(key)) {
1500 err = PTR_ERR(key);
1501 goto err_put;
1502 }
1503
1504 value_size = map->value_size;
1505
1506 err = -ENOMEM;
1507 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1508 if (!value)
1509 goto free_key;
1510
1511 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1512 map->map_type == BPF_MAP_TYPE_STACK) {
1513 err = map->ops->map_pop_elem(map, value);
1514 } else {
1515 err = -ENOTSUPP;
1516 }
1517
1518 if (err)
1519 goto free_value;
1520
1521 if (copy_to_user(uvalue, value, value_size) != 0) {
1522 err = -EFAULT;
1523 goto free_value;
1524 }
1525
1526 err = 0;
1527
1528free_value:
1529 kfree(value);
1530free_key:
1531 kfree(key);
1532err_put:
1533 fdput(f);
1534 return err;
1535}
1536
1537#define BPF_MAP_FREEZE_LAST_FIELD map_fd
1538
1539static int map_freeze(const union bpf_attr *attr)
1540{
1541 int err = 0, ufd = attr->map_fd;
1542 struct bpf_map *map;
1543 struct fd f;
1544
1545 if (CHECK_ATTR(BPF_MAP_FREEZE))
1546 return -EINVAL;
1547
1548 f = fdget(ufd);
1549 map = __bpf_map_get(f);
1550 if (IS_ERR(map))
1551 return PTR_ERR(map);
1552
1553 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1554 fdput(f);
1555 return -ENOTSUPP;
1556 }
1557
1558 mutex_lock(&map->freeze_mutex);
1559
1560 if (map->writecnt) {
1561 err = -EBUSY;
1562 goto err_put;
1563 }
1564 if (READ_ONCE(map->frozen)) {
1565 err = -EBUSY;
1566 goto err_put;
1567 }
1568 if (!bpf_capable()) {
1569 err = -EPERM;
1570 goto err_put;
1571 }
1572
1573 WRITE_ONCE(map->frozen, true);
1574err_put:
1575 mutex_unlock(&map->freeze_mutex);
1576 fdput(f);
1577 return err;
1578}
1579
1580static const struct bpf_prog_ops * const bpf_prog_types[] = {
1581#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1582 [_id] = & _name ## _prog_ops,
1583#define BPF_MAP_TYPE(_id, _ops)
1584#define BPF_LINK_TYPE(_id, _name)
1585#include <linux/bpf_types.h>
1586#undef BPF_PROG_TYPE
1587#undef BPF_MAP_TYPE
1588#undef BPF_LINK_TYPE
1589};
1590
1591static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1592{
1593 const struct bpf_prog_ops *ops;
1594
1595 if (type >= ARRAY_SIZE(bpf_prog_types))
1596 return -EINVAL;
1597 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1598 ops = bpf_prog_types[type];
1599 if (!ops)
1600 return -EINVAL;
1601
1602 if (!bpf_prog_is_dev_bound(prog->aux))
1603 prog->aux->ops = ops;
1604 else
1605 prog->aux->ops = &bpf_offload_prog_ops;
1606 prog->type = type;
1607 return 0;
1608}
1609
1610enum bpf_audit {
1611 BPF_AUDIT_LOAD,
1612 BPF_AUDIT_UNLOAD,
1613 BPF_AUDIT_MAX,
1614};
1615
1616static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1617 [BPF_AUDIT_LOAD] = "LOAD",
1618 [BPF_AUDIT_UNLOAD] = "UNLOAD",
1619};
1620
1621static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1622{
1623 struct audit_context *ctx = NULL;
1624 struct audit_buffer *ab;
1625
1626 if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1627 return;
1628 if (audit_enabled == AUDIT_OFF)
1629 return;
1630 if (op == BPF_AUDIT_LOAD)
1631 ctx = audit_context();
1632 ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1633 if (unlikely(!ab))
1634 return;
1635 audit_log_format(ab, "prog-id=%u op=%s",
1636 prog->aux->id, bpf_audit_str[op]);
1637 audit_log_end(ab);
1638}
1639
1640static int bpf_prog_alloc_id(struct bpf_prog *prog)
1641{
1642 int id;
1643
1644 idr_preload(GFP_KERNEL);
1645 spin_lock_bh(&prog_idr_lock);
1646 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1647 if (id > 0)
1648 prog->aux->id = id;
1649 spin_unlock_bh(&prog_idr_lock);
1650 idr_preload_end();
1651
1652 /* id is in [1, INT_MAX) */
1653 if (WARN_ON_ONCE(!id))
1654 return -ENOSPC;
1655
1656 return id > 0 ? 0 : id;
1657}
1658
1659void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1660{
1661 /* cBPF to eBPF migrations are currently not in the idr store.
1662 * Offloaded programs are removed from the store when their device
1663 * disappears - even if someone grabs an fd to them they are unusable,
1664 * simply waiting for refcnt to drop to be freed.
1665 */
1666 if (!prog->aux->id)
1667 return;
1668
1669 if (do_idr_lock)
1670 spin_lock_bh(&prog_idr_lock);
1671 else
1672 __acquire(&prog_idr_lock);
1673
1674 idr_remove(&prog_idr, prog->aux->id);
1675 prog->aux->id = 0;
1676
1677 if (do_idr_lock)
1678 spin_unlock_bh(&prog_idr_lock);
1679 else
1680 __release(&prog_idr_lock);
1681}
1682
1683static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1684{
1685 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1686
1687 kvfree(aux->func_info);
1688 kfree(aux->func_info_aux);
1689 free_uid(aux->user);
1690 security_bpf_prog_free(aux);
1691 bpf_prog_free(aux->prog);
1692}
1693
1694static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1695{
1696 bpf_prog_kallsyms_del_all(prog);
1697 btf_put(prog->aux->btf);
1698 kvfree(prog->aux->jited_linfo);
1699 kvfree(prog->aux->linfo);
1700 kfree(prog->aux->kfunc_tab);
1701 if (prog->aux->attach_btf)
1702 btf_put(prog->aux->attach_btf);
1703
1704 if (deferred) {
1705 if (prog->aux->sleepable)
1706 call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
1707 else
1708 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1709 } else {
1710 __bpf_prog_put_rcu(&prog->aux->rcu);
1711 }
1712}
1713
1714static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1715{
1716 if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1717 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1718 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
1719 /* bpf_prog_free_id() must be called first */
1720 bpf_prog_free_id(prog, do_idr_lock);
1721 __bpf_prog_put_noref(prog, true);
1722 }
1723}
1724
1725void bpf_prog_put(struct bpf_prog *prog)
1726{
1727 __bpf_prog_put(prog, true);
1728}
1729EXPORT_SYMBOL_GPL(bpf_prog_put);
1730
1731static int bpf_prog_release(struct inode *inode, struct file *filp)
1732{
1733 struct bpf_prog *prog = filp->private_data;
1734
1735 bpf_prog_put(prog);
1736 return 0;
1737}
1738
1739static void bpf_prog_get_stats(const struct bpf_prog *prog,
1740 struct bpf_prog_stats *stats)
1741{
1742 u64 nsecs = 0, cnt = 0, misses = 0;
1743 int cpu;
1744
1745 for_each_possible_cpu(cpu) {
1746 const struct bpf_prog_stats *st;
1747 unsigned int start;
1748 u64 tnsecs, tcnt, tmisses;
1749
1750 st = per_cpu_ptr(prog->stats, cpu);
1751 do {
1752 start = u64_stats_fetch_begin_irq(&st->syncp);
1753 tnsecs = st->nsecs;
1754 tcnt = st->cnt;
1755 tmisses = st->misses;
1756 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1757 nsecs += tnsecs;
1758 cnt += tcnt;
1759 misses += tmisses;
1760 }
1761 stats->nsecs = nsecs;
1762 stats->cnt = cnt;
1763 stats->misses = misses;
1764}
1765
1766#ifdef CONFIG_PROC_FS
1767static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1768{
1769 const struct bpf_prog *prog = filp->private_data;
1770 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1771 struct bpf_prog_stats stats;
1772
1773 bpf_prog_get_stats(prog, &stats);
1774 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1775 seq_printf(m,
1776 "prog_type:\t%u\n"
1777 "prog_jited:\t%u\n"
1778 "prog_tag:\t%s\n"
1779 "memlock:\t%llu\n"
1780 "prog_id:\t%u\n"
1781 "run_time_ns:\t%llu\n"
1782 "run_cnt:\t%llu\n"
1783 "recursion_misses:\t%llu\n",
1784 prog->type,
1785 prog->jited,
1786 prog_tag,
1787 prog->pages * 1ULL << PAGE_SHIFT,
1788 prog->aux->id,
1789 stats.nsecs,
1790 stats.cnt,
1791 stats.misses);
1792}
1793#endif
1794
1795const struct file_operations bpf_prog_fops = {
1796#ifdef CONFIG_PROC_FS
1797 .show_fdinfo = bpf_prog_show_fdinfo,
1798#endif
1799 .release = bpf_prog_release,
1800 .read = bpf_dummy_read,
1801 .write = bpf_dummy_write,
1802};
1803
1804int bpf_prog_new_fd(struct bpf_prog *prog)
1805{
1806 int ret;
1807
1808 ret = security_bpf_prog(prog);
1809 if (ret < 0)
1810 return ret;
1811
1812 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1813 O_RDWR | O_CLOEXEC);
1814}
1815
1816static struct bpf_prog *____bpf_prog_get(struct fd f)
1817{
1818 if (!f.file)
1819 return ERR_PTR(-EBADF);
1820 if (f.file->f_op != &bpf_prog_fops) {
1821 fdput(f);
1822 return ERR_PTR(-EINVAL);
1823 }
1824
1825 return f.file->private_data;
1826}
1827
1828void bpf_prog_add(struct bpf_prog *prog, int i)
1829{
1830 atomic64_add(i, &prog->aux->refcnt);
1831}
1832EXPORT_SYMBOL_GPL(bpf_prog_add);
1833
1834void bpf_prog_sub(struct bpf_prog *prog, int i)
1835{
1836 /* Only to be used for undoing previous bpf_prog_add() in some
1837 * error path. We still know that another entity in our call
1838 * path holds a reference to the program, thus atomic_sub() can
1839 * be safely used in such cases!
1840 */
1841 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1842}
1843EXPORT_SYMBOL_GPL(bpf_prog_sub);
1844
1845void bpf_prog_inc(struct bpf_prog *prog)
1846{
1847 atomic64_inc(&prog->aux->refcnt);
1848}
1849EXPORT_SYMBOL_GPL(bpf_prog_inc);
1850
1851/* prog_idr_lock should have been held */
1852struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1853{
1854 int refold;
1855
1856 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1857
1858 if (!refold)
1859 return ERR_PTR(-ENOENT);
1860
1861 return prog;
1862}
1863EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1864
1865bool bpf_prog_get_ok(struct bpf_prog *prog,
1866 enum bpf_prog_type *attach_type, bool attach_drv)
1867{
1868 /* not an attachment, just a refcount inc, always allow */
1869 if (!attach_type)
1870 return true;
1871
1872 if (prog->type != *attach_type)
1873 return false;
1874 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1875 return false;
1876
1877 return true;
1878}
1879
1880static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1881 bool attach_drv)
1882{
1883 struct fd f = fdget(ufd);
1884 struct bpf_prog *prog;
1885
1886 prog = ____bpf_prog_get(f);
1887 if (IS_ERR(prog))
1888 return prog;
1889 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1890 prog = ERR_PTR(-EINVAL);
1891 goto out;
1892 }
1893
1894 bpf_prog_inc(prog);
1895out:
1896 fdput(f);
1897 return prog;
1898}
1899
1900struct bpf_prog *bpf_prog_get(u32 ufd)
1901{
1902 return __bpf_prog_get(ufd, NULL, false);
1903}
1904
1905struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1906 bool attach_drv)
1907{
1908 return __bpf_prog_get(ufd, &type, attach_drv);
1909}
1910EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1911
1912/* Initially all BPF programs could be loaded w/o specifying
1913 * expected_attach_type. Later for some of them specifying expected_attach_type
1914 * at load time became required so that program could be validated properly.
1915 * Programs of types that are allowed to be loaded both w/ and w/o (for
1916 * backward compatibility) expected_attach_type, should have the default attach
1917 * type assigned to expected_attach_type for the latter case, so that it can be
1918 * validated later at attach time.
1919 *
1920 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1921 * prog type requires it but has some attach types that have to be backward
1922 * compatible.
1923 */
1924static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1925{
1926 switch (attr->prog_type) {
1927 case BPF_PROG_TYPE_CGROUP_SOCK:
1928 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1929 * exist so checking for non-zero is the way to go here.
1930 */
1931 if (!attr->expected_attach_type)
1932 attr->expected_attach_type =
1933 BPF_CGROUP_INET_SOCK_CREATE;
1934 break;
1935 }
1936}
1937
1938static int
1939bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1940 enum bpf_attach_type expected_attach_type,
1941 struct btf *attach_btf, u32 btf_id,
1942 struct bpf_prog *dst_prog)
1943{
1944 if (btf_id) {
1945 if (btf_id > BTF_MAX_TYPE)
1946 return -EINVAL;
1947
1948 if (!attach_btf && !dst_prog)
1949 return -EINVAL;
1950
1951 switch (prog_type) {
1952 case BPF_PROG_TYPE_TRACING:
1953 case BPF_PROG_TYPE_LSM:
1954 case BPF_PROG_TYPE_STRUCT_OPS:
1955 case BPF_PROG_TYPE_EXT:
1956 break;
1957 default:
1958 return -EINVAL;
1959 }
1960 }
1961
1962 if (attach_btf && (!btf_id || dst_prog))
1963 return -EINVAL;
1964
1965 if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
1966 prog_type != BPF_PROG_TYPE_EXT)
1967 return -EINVAL;
1968
1969 switch (prog_type) {
1970 case BPF_PROG_TYPE_CGROUP_SOCK:
1971 switch (expected_attach_type) {
1972 case BPF_CGROUP_INET_SOCK_CREATE:
1973 case BPF_CGROUP_INET_SOCK_RELEASE:
1974 case BPF_CGROUP_INET4_POST_BIND:
1975 case BPF_CGROUP_INET6_POST_BIND:
1976 return 0;
1977 default:
1978 return -EINVAL;
1979 }
1980 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1981 switch (expected_attach_type) {
1982 case BPF_CGROUP_INET4_BIND:
1983 case BPF_CGROUP_INET6_BIND:
1984 case BPF_CGROUP_INET4_CONNECT:
1985 case BPF_CGROUP_INET6_CONNECT:
1986 case BPF_CGROUP_INET4_GETPEERNAME:
1987 case BPF_CGROUP_INET6_GETPEERNAME:
1988 case BPF_CGROUP_INET4_GETSOCKNAME:
1989 case BPF_CGROUP_INET6_GETSOCKNAME:
1990 case BPF_CGROUP_UDP4_SENDMSG:
1991 case BPF_CGROUP_UDP6_SENDMSG:
1992 case BPF_CGROUP_UDP4_RECVMSG:
1993 case BPF_CGROUP_UDP6_RECVMSG:
1994 return 0;
1995 default:
1996 return -EINVAL;
1997 }
1998 case BPF_PROG_TYPE_CGROUP_SKB:
1999 switch (expected_attach_type) {
2000 case BPF_CGROUP_INET_INGRESS:
2001 case BPF_CGROUP_INET_EGRESS:
2002 return 0;
2003 default:
2004 return -EINVAL;
2005 }
2006 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2007 switch (expected_attach_type) {
2008 case BPF_CGROUP_SETSOCKOPT:
2009 case BPF_CGROUP_GETSOCKOPT:
2010 return 0;
2011 default:
2012 return -EINVAL;
2013 }
2014 case BPF_PROG_TYPE_SK_LOOKUP:
2015 if (expected_attach_type == BPF_SK_LOOKUP)
2016 return 0;
2017 return -EINVAL;
2018 case BPF_PROG_TYPE_EXT:
2019 if (expected_attach_type)
2020 return -EINVAL;
2021 fallthrough;
2022 default:
2023 return 0;
2024 }
2025}
2026
2027static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2028{
2029 switch (prog_type) {
2030 case BPF_PROG_TYPE_SCHED_CLS:
2031 case BPF_PROG_TYPE_SCHED_ACT:
2032 case BPF_PROG_TYPE_XDP:
2033 case BPF_PROG_TYPE_LWT_IN:
2034 case BPF_PROG_TYPE_LWT_OUT:
2035 case BPF_PROG_TYPE_LWT_XMIT:
2036 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2037 case BPF_PROG_TYPE_SK_SKB:
2038 case BPF_PROG_TYPE_SK_MSG:
2039 case BPF_PROG_TYPE_LIRC_MODE2:
2040 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2041 case BPF_PROG_TYPE_CGROUP_DEVICE:
2042 case BPF_PROG_TYPE_CGROUP_SOCK:
2043 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2044 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2045 case BPF_PROG_TYPE_CGROUP_SYSCTL:
2046 case BPF_PROG_TYPE_SOCK_OPS:
2047 case BPF_PROG_TYPE_EXT: /* extends any prog */
2048 return true;
2049 case BPF_PROG_TYPE_CGROUP_SKB:
2050 /* always unpriv */
2051 case BPF_PROG_TYPE_SK_REUSEPORT:
2052 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2053 default:
2054 return false;
2055 }
2056}
2057
2058static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2059{
2060 switch (prog_type) {
2061 case BPF_PROG_TYPE_KPROBE:
2062 case BPF_PROG_TYPE_TRACEPOINT:
2063 case BPF_PROG_TYPE_PERF_EVENT:
2064 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2065 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2066 case BPF_PROG_TYPE_TRACING:
2067 case BPF_PROG_TYPE_LSM:
2068 case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2069 case BPF_PROG_TYPE_EXT: /* extends any prog */
2070 return true;
2071 default:
2072 return false;
2073 }
2074}
2075
2076/* last field in 'union bpf_attr' used by this command */
2077#define BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
2078
2079static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
2080{
2081 enum bpf_prog_type type = attr->prog_type;
2082 struct bpf_prog *prog, *dst_prog = NULL;
2083 struct btf *attach_btf = NULL;
2084 int err;
2085 char license[128];
2086 bool is_gpl;
2087
2088 if (CHECK_ATTR(BPF_PROG_LOAD))
2089 return -EINVAL;
2090
2091 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2092 BPF_F_ANY_ALIGNMENT |
2093 BPF_F_TEST_STATE_FREQ |
2094 BPF_F_SLEEPABLE |
2095 BPF_F_TEST_RND_HI32))
2096 return -EINVAL;
2097
2098 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2099 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2100 !bpf_capable())
2101 return -EPERM;
2102
2103 /* copy eBPF program license from user space */
2104 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
2105 sizeof(license) - 1) < 0)
2106 return -EFAULT;
2107 license[sizeof(license) - 1] = 0;
2108
2109 /* eBPF programs must be GPL compatible to use GPL-ed functions */
2110 is_gpl = license_is_gpl_compatible(license);
2111
2112 if (attr->insn_cnt == 0 ||
2113 attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2114 return -E2BIG;
2115 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2116 type != BPF_PROG_TYPE_CGROUP_SKB &&
2117 !bpf_capable())
2118 return -EPERM;
2119
2120 if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2121 return -EPERM;
2122 if (is_perfmon_prog_type(type) && !perfmon_capable())
2123 return -EPERM;
2124
2125 /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2126 * or btf, we need to check which one it is
2127 */
2128 if (attr->attach_prog_fd) {
2129 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2130 if (IS_ERR(dst_prog)) {
2131 dst_prog = NULL;
2132 attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2133 if (IS_ERR(attach_btf))
2134 return -EINVAL;
2135 if (!btf_is_kernel(attach_btf)) {
2136 /* attaching through specifying bpf_prog's BTF
2137 * objects directly might be supported eventually
2138 */
2139 btf_put(attach_btf);
2140 return -ENOTSUPP;
2141 }
2142 }
2143 } else if (attr->attach_btf_id) {
2144 /* fall back to vmlinux BTF, if BTF type ID is specified */
2145 attach_btf = bpf_get_btf_vmlinux();
2146 if (IS_ERR(attach_btf))
2147 return PTR_ERR(attach_btf);
2148 if (!attach_btf)
2149 return -EINVAL;
2150 btf_get(attach_btf);
2151 }
2152
2153 bpf_prog_load_fixup_attach_type(attr);
2154 if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2155 attach_btf, attr->attach_btf_id,
2156 dst_prog)) {
2157 if (dst_prog)
2158 bpf_prog_put(dst_prog);
2159 if (attach_btf)
2160 btf_put(attach_btf);
2161 return -EINVAL;
2162 }
2163
2164 /* plain bpf_prog allocation */
2165 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2166 if (!prog) {
2167 if (dst_prog)
2168 bpf_prog_put(dst_prog);
2169 if (attach_btf)
2170 btf_put(attach_btf);
2171 return -ENOMEM;
2172 }
2173
2174 prog->expected_attach_type = attr->expected_attach_type;
2175 prog->aux->attach_btf = attach_btf;
2176 prog->aux->attach_btf_id = attr->attach_btf_id;
2177 prog->aux->dst_prog = dst_prog;
2178 prog->aux->offload_requested = !!attr->prog_ifindex;
2179 prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2180
2181 err = security_bpf_prog_alloc(prog->aux);
2182 if (err)
2183 goto free_prog;
2184
2185 prog->aux->user = get_current_user();
2186 prog->len = attr->insn_cnt;
2187
2188 err = -EFAULT;
2189 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
2190 bpf_prog_insn_size(prog)) != 0)
2191 goto free_prog_sec;
2192
2193 prog->orig_prog = NULL;
2194 prog->jited = 0;
2195
2196 atomic64_set(&prog->aux->refcnt, 1);
2197 prog->gpl_compatible = is_gpl ? 1 : 0;
2198
2199 if (bpf_prog_is_dev_bound(prog->aux)) {
2200 err = bpf_prog_offload_init(prog, attr);
2201 if (err)
2202 goto free_prog_sec;
2203 }
2204
2205 /* find program type: socket_filter vs tracing_filter */
2206 err = find_prog_type(type, prog);
2207 if (err < 0)
2208 goto free_prog_sec;
2209
2210 prog->aux->load_time = ktime_get_boottime_ns();
2211 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2212 sizeof(attr->prog_name));
2213 if (err < 0)
2214 goto free_prog_sec;
2215
2216 /* run eBPF verifier */
2217 err = bpf_check(&prog, attr, uattr);
2218 if (err < 0)
2219 goto free_used_maps;
2220
2221 prog = bpf_prog_select_runtime(prog, &err);
2222 if (err < 0)
2223 goto free_used_maps;
2224
2225 err = bpf_prog_alloc_id(prog);
2226 if (err)
2227 goto free_used_maps;
2228
2229 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2230 * effectively publicly exposed. However, retrieving via
2231 * bpf_prog_get_fd_by_id() will take another reference,
2232 * therefore it cannot be gone underneath us.
2233 *
2234 * Only for the time /after/ successful bpf_prog_new_fd()
2235 * and before returning to userspace, we might just hold
2236 * one reference and any parallel close on that fd could
2237 * rip everything out. Hence, below notifications must
2238 * happen before bpf_prog_new_fd().
2239 *
2240 * Also, any failure handling from this point onwards must
2241 * be using bpf_prog_put() given the program is exposed.
2242 */
2243 bpf_prog_kallsyms_add(prog);
2244 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2245 bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2246
2247 err = bpf_prog_new_fd(prog);
2248 if (err < 0)
2249 bpf_prog_put(prog);
2250 return err;
2251
2252free_used_maps:
2253 /* In case we have subprogs, we need to wait for a grace
2254 * period before we can tear down JIT memory since symbols
2255 * are already exposed under kallsyms.
2256 */
2257 __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2258 return err;
2259free_prog_sec:
2260 free_uid(prog->aux->user);
2261 security_bpf_prog_free(prog->aux);
2262free_prog:
2263 if (prog->aux->attach_btf)
2264 btf_put(prog->aux->attach_btf);
2265 bpf_prog_free(prog);
2266 return err;
2267}
2268
2269#define BPF_OBJ_LAST_FIELD file_flags
2270
2271static int bpf_obj_pin(const union bpf_attr *attr)
2272{
2273 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2274 return -EINVAL;
2275
2276 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2277}
2278
2279static int bpf_obj_get(const union bpf_attr *attr)
2280{
2281 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2282 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2283 return -EINVAL;
2284
2285 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2286 attr->file_flags);
2287}
2288
2289void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2290 const struct bpf_link_ops *ops, struct bpf_prog *prog)
2291{
2292 atomic64_set(&link->refcnt, 1);
2293 link->type = type;
2294 link->id = 0;
2295 link->ops = ops;
2296 link->prog = prog;
2297}
2298
2299static void bpf_link_free_id(int id)
2300{
2301 if (!id)
2302 return;
2303
2304 spin_lock_bh(&link_idr_lock);
2305 idr_remove(&link_idr, id);
2306 spin_unlock_bh(&link_idr_lock);
2307}
2308
2309/* Clean up bpf_link and corresponding anon_inode file and FD. After
2310 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2311 * anon_inode's release() call. This helper marksbpf_link as
2312 * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2313 * is not decremented, it's the responsibility of a calling code that failed
2314 * to complete bpf_link initialization.
2315 */
2316void bpf_link_cleanup(struct bpf_link_primer *primer)
2317{
2318 primer->link->prog = NULL;
2319 bpf_link_free_id(primer->id);
2320 fput(primer->file);
2321 put_unused_fd(primer->fd);
2322}
2323
2324void bpf_link_inc(struct bpf_link *link)
2325{
2326 atomic64_inc(&link->refcnt);
2327}
2328
2329/* bpf_link_free is guaranteed to be called from process context */
2330static void bpf_link_free(struct bpf_link *link)
2331{
2332 bpf_link_free_id(link->id);
2333 if (link->prog) {
2334 /* detach BPF program, clean up used resources */
2335 link->ops->release(link);
2336 bpf_prog_put(link->prog);
2337 }
2338 /* free bpf_link and its containing memory */
2339 link->ops->dealloc(link);
2340}
2341
2342static void bpf_link_put_deferred(struct work_struct *work)
2343{
2344 struct bpf_link *link = container_of(work, struct bpf_link, work);
2345
2346 bpf_link_free(link);
2347}
2348
2349/* bpf_link_put can be called from atomic context, but ensures that resources
2350 * are freed from process context
2351 */
2352void bpf_link_put(struct bpf_link *link)
2353{
2354 if (!atomic64_dec_and_test(&link->refcnt))
2355 return;
2356
2357 if (in_atomic()) {
2358 INIT_WORK(&link->work, bpf_link_put_deferred);
2359 schedule_work(&link->work);
2360 } else {
2361 bpf_link_free(link);
2362 }
2363}
2364
2365static int bpf_link_release(struct inode *inode, struct file *filp)
2366{
2367 struct bpf_link *link = filp->private_data;
2368
2369 bpf_link_put(link);
2370 return 0;
2371}
2372
2373#ifdef CONFIG_PROC_FS
2374#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2375#define BPF_MAP_TYPE(_id, _ops)
2376#define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2377static const char *bpf_link_type_strs[] = {
2378 [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2379#include <linux/bpf_types.h>
2380};
2381#undef BPF_PROG_TYPE
2382#undef BPF_MAP_TYPE
2383#undef BPF_LINK_TYPE
2384
2385static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2386{
2387 const struct bpf_link *link = filp->private_data;
2388 const struct bpf_prog *prog = link->prog;
2389 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2390
2391 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2392 seq_printf(m,
2393 "link_type:\t%s\n"
2394 "link_id:\t%u\n"
2395 "prog_tag:\t%s\n"
2396 "prog_id:\t%u\n",
2397 bpf_link_type_strs[link->type],
2398 link->id,
2399 prog_tag,
2400 prog->aux->id);
2401 if (link->ops->show_fdinfo)
2402 link->ops->show_fdinfo(link, m);
2403}
2404#endif
2405
2406static const struct file_operations bpf_link_fops = {
2407#ifdef CONFIG_PROC_FS
2408 .show_fdinfo = bpf_link_show_fdinfo,
2409#endif
2410 .release = bpf_link_release,
2411 .read = bpf_dummy_read,
2412 .write = bpf_dummy_write,
2413};
2414
2415static int bpf_link_alloc_id(struct bpf_link *link)
2416{
2417 int id;
2418
2419 idr_preload(GFP_KERNEL);
2420 spin_lock_bh(&link_idr_lock);
2421 id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2422 spin_unlock_bh(&link_idr_lock);
2423 idr_preload_end();
2424
2425 return id;
2426}
2427
2428/* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2429 * reserving unused FD and allocating ID from link_idr. This is to be paired
2430 * with bpf_link_settle() to install FD and ID and expose bpf_link to
2431 * user-space, if bpf_link is successfully attached. If not, bpf_link and
2432 * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2433 * transient state is passed around in struct bpf_link_primer.
2434 * This is preferred way to create and initialize bpf_link, especially when
2435 * there are complicated and expensive operations inbetween creating bpf_link
2436 * itself and attaching it to BPF hook. By using bpf_link_prime() and
2437 * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2438 * expensive (and potentially failing) roll back operations in a rare case
2439 * that file, FD, or ID can't be allocated.
2440 */
2441int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2442{
2443 struct file *file;
2444 int fd, id;
2445
2446 fd = get_unused_fd_flags(O_CLOEXEC);
2447 if (fd < 0)
2448 return fd;
2449
2450
2451 id = bpf_link_alloc_id(link);
2452 if (id < 0) {
2453 put_unused_fd(fd);
2454 return id;
2455 }
2456
2457 file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2458 if (IS_ERR(file)) {
2459 bpf_link_free_id(id);
2460 put_unused_fd(fd);
2461 return PTR_ERR(file);
2462 }
2463
2464 primer->link = link;
2465 primer->file = file;
2466 primer->fd = fd;
2467 primer->id = id;
2468 return 0;
2469}
2470
2471int bpf_link_settle(struct bpf_link_primer *primer)
2472{
2473 /* make bpf_link fetchable by ID */
2474 spin_lock_bh(&link_idr_lock);
2475 primer->link->id = primer->id;
2476 spin_unlock_bh(&link_idr_lock);
2477 /* make bpf_link fetchable by FD */
2478 fd_install(primer->fd, primer->file);
2479 /* pass through installed FD */
2480 return primer->fd;
2481}
2482
2483int bpf_link_new_fd(struct bpf_link *link)
2484{
2485 return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2486}
2487
2488struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2489{
2490 struct fd f = fdget(ufd);
2491 struct bpf_link *link;
2492
2493 if (!f.file)
2494 return ERR_PTR(-EBADF);
2495 if (f.file->f_op != &bpf_link_fops) {
2496 fdput(f);
2497 return ERR_PTR(-EINVAL);
2498 }
2499
2500 link = f.file->private_data;
2501 bpf_link_inc(link);
2502 fdput(f);
2503
2504 return link;
2505}
2506
2507struct bpf_tracing_link {
2508 struct bpf_link link;
2509 enum bpf_attach_type attach_type;
2510 struct bpf_trampoline *trampoline;
2511 struct bpf_prog *tgt_prog;
2512};
2513
2514static void bpf_tracing_link_release(struct bpf_link *link)
2515{
2516 struct bpf_tracing_link *tr_link =
2517 container_of(link, struct bpf_tracing_link, link);
2518
2519 WARN_ON_ONCE(bpf_trampoline_unlink_prog(link->prog,
2520 tr_link->trampoline));
2521
2522 bpf_trampoline_put(tr_link->trampoline);
2523
2524 /* tgt_prog is NULL if target is a kernel function */
2525 if (tr_link->tgt_prog)
2526 bpf_prog_put(tr_link->tgt_prog);
2527}
2528
2529static void bpf_tracing_link_dealloc(struct bpf_link *link)
2530{
2531 struct bpf_tracing_link *tr_link =
2532 container_of(link, struct bpf_tracing_link, link);
2533
2534 kfree(tr_link);
2535}
2536
2537static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2538 struct seq_file *seq)
2539{
2540 struct bpf_tracing_link *tr_link =
2541 container_of(link, struct bpf_tracing_link, link);
2542
2543 seq_printf(seq,
2544 "attach_type:\t%d\n",
2545 tr_link->attach_type);
2546}
2547
2548static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2549 struct bpf_link_info *info)
2550{
2551 struct bpf_tracing_link *tr_link =
2552 container_of(link, struct bpf_tracing_link, link);
2553
2554 info->tracing.attach_type = tr_link->attach_type;
2555 bpf_trampoline_unpack_key(tr_link->trampoline->key,
2556 &info->tracing.target_obj_id,
2557 &info->tracing.target_btf_id);
2558
2559 return 0;
2560}
2561
2562static const struct bpf_link_ops bpf_tracing_link_lops = {
2563 .release = bpf_tracing_link_release,
2564 .dealloc = bpf_tracing_link_dealloc,
2565 .show_fdinfo = bpf_tracing_link_show_fdinfo,
2566 .fill_link_info = bpf_tracing_link_fill_link_info,
2567};
2568
2569static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2570 int tgt_prog_fd,
2571 u32 btf_id)
2572{
2573 struct bpf_link_primer link_primer;
2574 struct bpf_prog *tgt_prog = NULL;
2575 struct bpf_trampoline *tr = NULL;
2576 struct bpf_tracing_link *link;
2577 u64 key = 0;
2578 int err;
2579
2580 switch (prog->type) {
2581 case BPF_PROG_TYPE_TRACING:
2582 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2583 prog->expected_attach_type != BPF_TRACE_FEXIT &&
2584 prog->expected_attach_type != BPF_MODIFY_RETURN) {
2585 err = -EINVAL;
2586 goto out_put_prog;
2587 }
2588 break;
2589 case BPF_PROG_TYPE_EXT:
2590 if (prog->expected_attach_type != 0) {
2591 err = -EINVAL;
2592 goto out_put_prog;
2593 }
2594 break;
2595 case BPF_PROG_TYPE_LSM:
2596 if (prog->expected_attach_type != BPF_LSM_MAC) {
2597 err = -EINVAL;
2598 goto out_put_prog;
2599 }
2600 break;
2601 default:
2602 err = -EINVAL;
2603 goto out_put_prog;
2604 }
2605
2606 if (!!tgt_prog_fd != !!btf_id) {
2607 err = -EINVAL;
2608 goto out_put_prog;
2609 }
2610
2611 if (tgt_prog_fd) {
2612 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
2613 if (prog->type != BPF_PROG_TYPE_EXT) {
2614 err = -EINVAL;
2615 goto out_put_prog;
2616 }
2617
2618 tgt_prog = bpf_prog_get(tgt_prog_fd);
2619 if (IS_ERR(tgt_prog)) {
2620 err = PTR_ERR(tgt_prog);
2621 tgt_prog = NULL;
2622 goto out_put_prog;
2623 }
2624
2625 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
2626 }
2627
2628 link = kzalloc(sizeof(*link), GFP_USER);
2629 if (!link) {
2630 err = -ENOMEM;
2631 goto out_put_prog;
2632 }
2633 bpf_link_init(&link->link, BPF_LINK_TYPE_TRACING,
2634 &bpf_tracing_link_lops, prog);
2635 link->attach_type = prog->expected_attach_type;
2636
2637 mutex_lock(&prog->aux->dst_mutex);
2638
2639 /* There are a few possible cases here:
2640 *
2641 * - if prog->aux->dst_trampoline is set, the program was just loaded
2642 * and not yet attached to anything, so we can use the values stored
2643 * in prog->aux
2644 *
2645 * - if prog->aux->dst_trampoline is NULL, the program has already been
2646 * attached to a target and its initial target was cleared (below)
2647 *
2648 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
2649 * target_btf_id using the link_create API.
2650 *
2651 * - if tgt_prog == NULL when this function was called using the old
2652 * raw_tracepoint_open API, and we need a target from prog->aux
2653 *
2654 * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
2655 * was detached and is going for re-attachment.
2656 */
2657 if (!prog->aux->dst_trampoline && !tgt_prog) {
2658 /*
2659 * Allow re-attach for TRACING and LSM programs. If it's
2660 * currently linked, bpf_trampoline_link_prog will fail.
2661 * EXT programs need to specify tgt_prog_fd, so they
2662 * re-attach in separate code path.
2663 */
2664 if (prog->type != BPF_PROG_TYPE_TRACING &&
2665 prog->type != BPF_PROG_TYPE_LSM) {
2666 err = -EINVAL;
2667 goto out_unlock;
2668 }
2669 btf_id = prog->aux->attach_btf_id;
2670 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
2671 }
2672
2673 if (!prog->aux->dst_trampoline ||
2674 (key && key != prog->aux->dst_trampoline->key)) {
2675 /* If there is no saved target, or the specified target is
2676 * different from the destination specified at load time, we
2677 * need a new trampoline and a check for compatibility
2678 */
2679 struct bpf_attach_target_info tgt_info = {};
2680
2681 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
2682 &tgt_info);
2683 if (err)
2684 goto out_unlock;
2685
2686 tr = bpf_trampoline_get(key, &tgt_info);
2687 if (!tr) {
2688 err = -ENOMEM;
2689 goto out_unlock;
2690 }
2691 } else {
2692 /* The caller didn't specify a target, or the target was the
2693 * same as the destination supplied during program load. This
2694 * means we can reuse the trampoline and reference from program
2695 * load time, and there is no need to allocate a new one. This
2696 * can only happen once for any program, as the saved values in
2697 * prog->aux are cleared below.
2698 */
2699 tr = prog->aux->dst_trampoline;
2700 tgt_prog = prog->aux->dst_prog;
2701 }
2702
2703 err = bpf_link_prime(&link->link, &link_primer);
2704 if (err)
2705 goto out_unlock;
2706
2707 err = bpf_trampoline_link_prog(prog, tr);
2708 if (err) {
2709 bpf_link_cleanup(&link_primer);
2710 link = NULL;
2711 goto out_unlock;
2712 }
2713
2714 link->tgt_prog = tgt_prog;
2715 link->trampoline = tr;
2716
2717 /* Always clear the trampoline and target prog from prog->aux to make
2718 * sure the original attach destination is not kept alive after a
2719 * program is (re-)attached to another target.
2720 */
2721 if (prog->aux->dst_prog &&
2722 (tgt_prog_fd || tr != prog->aux->dst_trampoline))
2723 /* got extra prog ref from syscall, or attaching to different prog */
2724 bpf_prog_put(prog->aux->dst_prog);
2725 if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
2726 /* we allocated a new trampoline, so free the old one */
2727 bpf_trampoline_put(prog->aux->dst_trampoline);
2728
2729 prog->aux->dst_prog = NULL;
2730 prog->aux->dst_trampoline = NULL;
2731 mutex_unlock(&prog->aux->dst_mutex);
2732
2733 return bpf_link_settle(&link_primer);
2734out_unlock:
2735 if (tr && tr != prog->aux->dst_trampoline)
2736 bpf_trampoline_put(tr);
2737 mutex_unlock(&prog->aux->dst_mutex);
2738 kfree(link);
2739out_put_prog:
2740 if (tgt_prog_fd && tgt_prog)
2741 bpf_prog_put(tgt_prog);
2742 return err;
2743}
2744
2745struct bpf_raw_tp_link {
2746 struct bpf_link link;
2747 struct bpf_raw_event_map *btp;
2748};
2749
2750static void bpf_raw_tp_link_release(struct bpf_link *link)
2751{
2752 struct bpf_raw_tp_link *raw_tp =
2753 container_of(link, struct bpf_raw_tp_link, link);
2754
2755 bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
2756 bpf_put_raw_tracepoint(raw_tp->btp);
2757}
2758
2759static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
2760{
2761 struct bpf_raw_tp_link *raw_tp =
2762 container_of(link, struct bpf_raw_tp_link, link);
2763
2764 kfree(raw_tp);
2765}
2766
2767static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
2768 struct seq_file *seq)
2769{
2770 struct bpf_raw_tp_link *raw_tp_link =
2771 container_of(link, struct bpf_raw_tp_link, link);
2772
2773 seq_printf(seq,
2774 "tp_name:\t%s\n",
2775 raw_tp_link->btp->tp->name);
2776}
2777
2778static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
2779 struct bpf_link_info *info)
2780{
2781 struct bpf_raw_tp_link *raw_tp_link =
2782 container_of(link, struct bpf_raw_tp_link, link);
2783 char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
2784 const char *tp_name = raw_tp_link->btp->tp->name;
2785 u32 ulen = info->raw_tracepoint.tp_name_len;
2786 size_t tp_len = strlen(tp_name);
2787
2788 if (!ulen ^ !ubuf)
2789 return -EINVAL;
2790
2791 info->raw_tracepoint.tp_name_len = tp_len + 1;
2792
2793 if (!ubuf)
2794 return 0;
2795
2796 if (ulen >= tp_len + 1) {
2797 if (copy_to_user(ubuf, tp_name, tp_len + 1))
2798 return -EFAULT;
2799 } else {
2800 char zero = '\0';
2801
2802 if (copy_to_user(ubuf, tp_name, ulen - 1))
2803 return -EFAULT;
2804 if (put_user(zero, ubuf + ulen - 1))
2805 return -EFAULT;
2806 return -ENOSPC;
2807 }
2808
2809 return 0;
2810}
2811
2812static const struct bpf_link_ops bpf_raw_tp_link_lops = {
2813 .release = bpf_raw_tp_link_release,
2814 .dealloc = bpf_raw_tp_link_dealloc,
2815 .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
2816 .fill_link_info = bpf_raw_tp_link_fill_link_info,
2817};
2818
2819#define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
2820
2821static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
2822{
2823 struct bpf_link_primer link_primer;
2824 struct bpf_raw_tp_link *link;
2825 struct bpf_raw_event_map *btp;
2826 struct bpf_prog *prog;
2827 const char *tp_name;
2828 char buf[128];
2829 int err;
2830
2831 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
2832 return -EINVAL;
2833
2834 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
2835 if (IS_ERR(prog))
2836 return PTR_ERR(prog);
2837
2838 switch (prog->type) {
2839 case BPF_PROG_TYPE_TRACING:
2840 case BPF_PROG_TYPE_EXT:
2841 case BPF_PROG_TYPE_LSM:
2842 if (attr->raw_tracepoint.name) {
2843 /* The attach point for this category of programs
2844 * should be specified via btf_id during program load.
2845 */
2846 err = -EINVAL;
2847 goto out_put_prog;
2848 }
2849 if (prog->type == BPF_PROG_TYPE_TRACING &&
2850 prog->expected_attach_type == BPF_TRACE_RAW_TP) {
2851 tp_name = prog->aux->attach_func_name;
2852 break;
2853 }
2854 err = bpf_tracing_prog_attach(prog, 0, 0);
2855 if (err >= 0)
2856 return err;
2857 goto out_put_prog;
2858 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2859 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2860 if (strncpy_from_user(buf,
2861 u64_to_user_ptr(attr->raw_tracepoint.name),
2862 sizeof(buf) - 1) < 0) {
2863 err = -EFAULT;
2864 goto out_put_prog;
2865 }
2866 buf[sizeof(buf) - 1] = 0;
2867 tp_name = buf;
2868 break;
2869 default:
2870 err = -EINVAL;
2871 goto out_put_prog;
2872 }
2873
2874 btp = bpf_get_raw_tracepoint(tp_name);
2875 if (!btp) {
2876 err = -ENOENT;
2877 goto out_put_prog;
2878 }
2879
2880 link = kzalloc(sizeof(*link), GFP_USER);
2881 if (!link) {
2882 err = -ENOMEM;
2883 goto out_put_btp;
2884 }
2885 bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
2886 &bpf_raw_tp_link_lops, prog);
2887 link->btp = btp;
2888
2889 err = bpf_link_prime(&link->link, &link_primer);
2890 if (err) {
2891 kfree(link);
2892 goto out_put_btp;
2893 }
2894
2895 err = bpf_probe_register(link->btp, prog);
2896 if (err) {
2897 bpf_link_cleanup(&link_primer);
2898 goto out_put_btp;
2899 }
2900
2901 return bpf_link_settle(&link_primer);
2902
2903out_put_btp:
2904 bpf_put_raw_tracepoint(btp);
2905out_put_prog:
2906 bpf_prog_put(prog);
2907 return err;
2908}
2909
2910static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2911 enum bpf_attach_type attach_type)
2912{
2913 switch (prog->type) {
2914 case BPF_PROG_TYPE_CGROUP_SOCK:
2915 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2916 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2917 case BPF_PROG_TYPE_SK_LOOKUP:
2918 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2919 case BPF_PROG_TYPE_CGROUP_SKB:
2920 if (!capable(CAP_NET_ADMIN))
2921 /* cg-skb progs can be loaded by unpriv user.
2922 * check permissions at attach time.
2923 */
2924 return -EPERM;
2925 return prog->enforce_expected_attach_type &&
2926 prog->expected_attach_type != attach_type ?
2927 -EINVAL : 0;
2928 default:
2929 return 0;
2930 }
2931}
2932
2933static enum bpf_prog_type
2934attach_type_to_prog_type(enum bpf_attach_type attach_type)
2935{
2936 switch (attach_type) {
2937 case BPF_CGROUP_INET_INGRESS:
2938 case BPF_CGROUP_INET_EGRESS:
2939 return BPF_PROG_TYPE_CGROUP_SKB;
2940 case BPF_CGROUP_INET_SOCK_CREATE:
2941 case BPF_CGROUP_INET_SOCK_RELEASE:
2942 case BPF_CGROUP_INET4_POST_BIND:
2943 case BPF_CGROUP_INET6_POST_BIND:
2944 return BPF_PROG_TYPE_CGROUP_SOCK;
2945 case BPF_CGROUP_INET4_BIND:
2946 case BPF_CGROUP_INET6_BIND:
2947 case BPF_CGROUP_INET4_CONNECT:
2948 case BPF_CGROUP_INET6_CONNECT:
2949 case BPF_CGROUP_INET4_GETPEERNAME:
2950 case BPF_CGROUP_INET6_GETPEERNAME:
2951 case BPF_CGROUP_INET4_GETSOCKNAME:
2952 case BPF_CGROUP_INET6_GETSOCKNAME:
2953 case BPF_CGROUP_UDP4_SENDMSG:
2954 case BPF_CGROUP_UDP6_SENDMSG:
2955 case BPF_CGROUP_UDP4_RECVMSG:
2956 case BPF_CGROUP_UDP6_RECVMSG:
2957 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2958 case BPF_CGROUP_SOCK_OPS:
2959 return BPF_PROG_TYPE_SOCK_OPS;
2960 case BPF_CGROUP_DEVICE:
2961 return BPF_PROG_TYPE_CGROUP_DEVICE;
2962 case BPF_SK_MSG_VERDICT:
2963 return BPF_PROG_TYPE_SK_MSG;
2964 case BPF_SK_SKB_STREAM_PARSER:
2965 case BPF_SK_SKB_STREAM_VERDICT:
2966 case BPF_SK_SKB_VERDICT:
2967 return BPF_PROG_TYPE_SK_SKB;
2968 case BPF_LIRC_MODE2:
2969 return BPF_PROG_TYPE_LIRC_MODE2;
2970 case BPF_FLOW_DISSECTOR:
2971 return BPF_PROG_TYPE_FLOW_DISSECTOR;
2972 case BPF_CGROUP_SYSCTL:
2973 return BPF_PROG_TYPE_CGROUP_SYSCTL;
2974 case BPF_CGROUP_GETSOCKOPT:
2975 case BPF_CGROUP_SETSOCKOPT:
2976 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
2977 case BPF_TRACE_ITER:
2978 return BPF_PROG_TYPE_TRACING;
2979 case BPF_SK_LOOKUP:
2980 return BPF_PROG_TYPE_SK_LOOKUP;
2981 case BPF_XDP:
2982 return BPF_PROG_TYPE_XDP;
2983 default:
2984 return BPF_PROG_TYPE_UNSPEC;
2985 }
2986}
2987
2988#define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
2989
2990#define BPF_F_ATTACH_MASK \
2991 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
2992
2993static int bpf_prog_attach(const union bpf_attr *attr)
2994{
2995 enum bpf_prog_type ptype;
2996 struct bpf_prog *prog;
2997 int ret;
2998
2999 if (CHECK_ATTR(BPF_PROG_ATTACH))
3000 return -EINVAL;
3001
3002 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
3003 return -EINVAL;
3004
3005 ptype = attach_type_to_prog_type(attr->attach_type);
3006 if (ptype == BPF_PROG_TYPE_UNSPEC)
3007 return -EINVAL;
3008
3009 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3010 if (IS_ERR(prog))
3011 return PTR_ERR(prog);
3012
3013 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3014 bpf_prog_put(prog);
3015 return -EINVAL;
3016 }
3017
3018 switch (ptype) {
3019 case BPF_PROG_TYPE_SK_SKB:
3020 case BPF_PROG_TYPE_SK_MSG:
3021 ret = sock_map_get_from_fd(attr, prog);
3022 break;
3023 case BPF_PROG_TYPE_LIRC_MODE2:
3024 ret = lirc_prog_attach(attr, prog);
3025 break;
3026 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3027 ret = netns_bpf_prog_attach(attr, prog);
3028 break;
3029 case BPF_PROG_TYPE_CGROUP_DEVICE:
3030 case BPF_PROG_TYPE_CGROUP_SKB:
3031 case BPF_PROG_TYPE_CGROUP_SOCK:
3032 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3033 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3034 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3035 case BPF_PROG_TYPE_SOCK_OPS:
3036 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3037 break;
3038 default:
3039 ret = -EINVAL;
3040 }
3041
3042 if (ret)
3043 bpf_prog_put(prog);
3044 return ret;
3045}
3046
3047#define BPF_PROG_DETACH_LAST_FIELD attach_type
3048
3049static int bpf_prog_detach(const union bpf_attr *attr)
3050{
3051 enum bpf_prog_type ptype;
3052
3053 if (CHECK_ATTR(BPF_PROG_DETACH))
3054 return -EINVAL;
3055
3056 ptype = attach_type_to_prog_type(attr->attach_type);
3057
3058 switch (ptype) {
3059 case BPF_PROG_TYPE_SK_MSG:
3060 case BPF_PROG_TYPE_SK_SKB:
3061 return sock_map_prog_detach(attr, ptype);
3062 case BPF_PROG_TYPE_LIRC_MODE2:
3063 return lirc_prog_detach(attr);
3064 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3065 return netns_bpf_prog_detach(attr, ptype);
3066 case BPF_PROG_TYPE_CGROUP_DEVICE:
3067 case BPF_PROG_TYPE_CGROUP_SKB:
3068 case BPF_PROG_TYPE_CGROUP_SOCK:
3069 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3070 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3071 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3072 case BPF_PROG_TYPE_SOCK_OPS:
3073 return cgroup_bpf_prog_detach(attr, ptype);
3074 default:
3075 return -EINVAL;
3076 }
3077}
3078
3079#define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
3080
3081static int bpf_prog_query(const union bpf_attr *attr,
3082 union bpf_attr __user *uattr)
3083{
3084 if (!capable(CAP_NET_ADMIN))
3085 return -EPERM;
3086 if (CHECK_ATTR(BPF_PROG_QUERY))
3087 return -EINVAL;
3088 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3089 return -EINVAL;
3090
3091 switch (attr->query.attach_type) {
3092 case BPF_CGROUP_INET_INGRESS:
3093 case BPF_CGROUP_INET_EGRESS:
3094 case BPF_CGROUP_INET_SOCK_CREATE:
3095 case BPF_CGROUP_INET_SOCK_RELEASE:
3096 case BPF_CGROUP_INET4_BIND:
3097 case BPF_CGROUP_INET6_BIND:
3098 case BPF_CGROUP_INET4_POST_BIND:
3099 case BPF_CGROUP_INET6_POST_BIND:
3100 case BPF_CGROUP_INET4_CONNECT:
3101 case BPF_CGROUP_INET6_CONNECT:
3102 case BPF_CGROUP_INET4_GETPEERNAME:
3103 case BPF_CGROUP_INET6_GETPEERNAME:
3104 case BPF_CGROUP_INET4_GETSOCKNAME:
3105 case BPF_CGROUP_INET6_GETSOCKNAME:
3106 case BPF_CGROUP_UDP4_SENDMSG:
3107 case BPF_CGROUP_UDP6_SENDMSG:
3108 case BPF_CGROUP_UDP4_RECVMSG:
3109 case BPF_CGROUP_UDP6_RECVMSG:
3110 case BPF_CGROUP_SOCK_OPS:
3111 case BPF_CGROUP_DEVICE:
3112 case BPF_CGROUP_SYSCTL:
3113 case BPF_CGROUP_GETSOCKOPT:
3114 case BPF_CGROUP_SETSOCKOPT:
3115 return cgroup_bpf_prog_query(attr, uattr);
3116 case BPF_LIRC_MODE2:
3117 return lirc_prog_query(attr, uattr);
3118 case BPF_FLOW_DISSECTOR:
3119 case BPF_SK_LOOKUP:
3120 return netns_bpf_prog_query(attr, uattr);
3121 default:
3122 return -EINVAL;
3123 }
3124}
3125
3126#define BPF_PROG_TEST_RUN_LAST_FIELD test.cpu
3127
3128static int bpf_prog_test_run(const union bpf_attr *attr,
3129 union bpf_attr __user *uattr)
3130{
3131 struct bpf_prog *prog;
3132 int ret = -ENOTSUPP;
3133
3134 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3135 return -EINVAL;
3136
3137 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3138 (!attr->test.ctx_size_in && attr->test.ctx_in))
3139 return -EINVAL;
3140
3141 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3142 (!attr->test.ctx_size_out && attr->test.ctx_out))
3143 return -EINVAL;
3144
3145 prog = bpf_prog_get(attr->test.prog_fd);
3146 if (IS_ERR(prog))
3147 return PTR_ERR(prog);
3148
3149 if (prog->aux->ops->test_run)
3150 ret = prog->aux->ops->test_run(prog, attr, uattr);
3151
3152 bpf_prog_put(prog);
3153 return ret;
3154}
3155
3156#define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3157
3158static int bpf_obj_get_next_id(const union bpf_attr *attr,
3159 union bpf_attr __user *uattr,
3160 struct idr *idr,
3161 spinlock_t *lock)
3162{
3163 u32 next_id = attr->start_id;
3164 int err = 0;
3165
3166 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3167 return -EINVAL;
3168
3169 if (!capable(CAP_SYS_ADMIN))
3170 return -EPERM;
3171
3172 next_id++;
3173 spin_lock_bh(lock);
3174 if (!idr_get_next(idr, &next_id))
3175 err = -ENOENT;
3176 spin_unlock_bh(lock);
3177
3178 if (!err)
3179 err = put_user(next_id, &uattr->next_id);
3180
3181 return err;
3182}
3183
3184struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3185{
3186 struct bpf_map *map;
3187
3188 spin_lock_bh(&map_idr_lock);
3189again:
3190 map = idr_get_next(&map_idr, id);
3191 if (map) {
3192 map = __bpf_map_inc_not_zero(map, false);
3193 if (IS_ERR(map)) {
3194 (*id)++;
3195 goto again;
3196 }
3197 }
3198 spin_unlock_bh(&map_idr_lock);
3199
3200 return map;
3201}
3202
3203struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3204{
3205 struct bpf_prog *prog;
3206
3207 spin_lock_bh(&prog_idr_lock);
3208again:
3209 prog = idr_get_next(&prog_idr, id);
3210 if (prog) {
3211 prog = bpf_prog_inc_not_zero(prog);
3212 if (IS_ERR(prog)) {
3213 (*id)++;
3214 goto again;
3215 }
3216 }
3217 spin_unlock_bh(&prog_idr_lock);
3218
3219 return prog;
3220}
3221
3222#define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3223
3224struct bpf_prog *bpf_prog_by_id(u32 id)
3225{
3226 struct bpf_prog *prog;
3227
3228 if (!id)
3229 return ERR_PTR(-ENOENT);
3230
3231 spin_lock_bh(&prog_idr_lock);
3232 prog = idr_find(&prog_idr, id);
3233 if (prog)
3234 prog = bpf_prog_inc_not_zero(prog);
3235 else
3236 prog = ERR_PTR(-ENOENT);
3237 spin_unlock_bh(&prog_idr_lock);
3238 return prog;
3239}
3240
3241static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3242{
3243 struct bpf_prog *prog;
3244 u32 id = attr->prog_id;
3245 int fd;
3246
3247 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3248 return -EINVAL;
3249
3250 if (!capable(CAP_SYS_ADMIN))
3251 return -EPERM;
3252
3253 prog = bpf_prog_by_id(id);
3254 if (IS_ERR(prog))
3255 return PTR_ERR(prog);
3256
3257 fd = bpf_prog_new_fd(prog);
3258 if (fd < 0)
3259 bpf_prog_put(prog);
3260
3261 return fd;
3262}
3263
3264#define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3265
3266static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3267{
3268 struct bpf_map *map;
3269 u32 id = attr->map_id;
3270 int f_flags;
3271 int fd;
3272
3273 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3274 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3275 return -EINVAL;
3276
3277 if (!capable(CAP_SYS_ADMIN))
3278 return -EPERM;
3279
3280 f_flags = bpf_get_file_flag(attr->open_flags);
3281 if (f_flags < 0)
3282 return f_flags;
3283
3284 spin_lock_bh(&map_idr_lock);
3285 map = idr_find(&map_idr, id);
3286 if (map)
3287 map = __bpf_map_inc_not_zero(map, true);
3288 else
3289 map = ERR_PTR(-ENOENT);
3290 spin_unlock_bh(&map_idr_lock);
3291
3292 if (IS_ERR(map))
3293 return PTR_ERR(map);
3294
3295 fd = bpf_map_new_fd(map, f_flags);
3296 if (fd < 0)
3297 bpf_map_put_with_uref(map);
3298
3299 return fd;
3300}
3301
3302static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3303 unsigned long addr, u32 *off,
3304 u32 *type)
3305{
3306 const struct bpf_map *map;
3307 int i;
3308
3309 mutex_lock(&prog->aux->used_maps_mutex);
3310 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3311 map = prog->aux->used_maps[i];
3312 if (map == (void *)addr) {
3313 *type = BPF_PSEUDO_MAP_FD;
3314 goto out;
3315 }
3316 if (!map->ops->map_direct_value_meta)
3317 continue;
3318 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3319 *type = BPF_PSEUDO_MAP_VALUE;
3320 goto out;
3321 }
3322 }
3323 map = NULL;
3324
3325out:
3326 mutex_unlock(&prog->aux->used_maps_mutex);
3327 return map;
3328}
3329
3330static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3331 const struct cred *f_cred)
3332{
3333 const struct bpf_map *map;
3334 struct bpf_insn *insns;
3335 u32 off, type;
3336 u64 imm;
3337 u8 code;
3338 int i;
3339
3340 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3341 GFP_USER);
3342 if (!insns)
3343 return insns;
3344
3345 for (i = 0; i < prog->len; i++) {
3346 code = insns[i].code;
3347
3348 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3349 insns[i].code = BPF_JMP | BPF_CALL;
3350 insns[i].imm = BPF_FUNC_tail_call;
3351 /* fall-through */
3352 }
3353 if (code == (BPF_JMP | BPF_CALL) ||
3354 code == (BPF_JMP | BPF_CALL_ARGS)) {
3355 if (code == (BPF_JMP | BPF_CALL_ARGS))
3356 insns[i].code = BPF_JMP | BPF_CALL;
3357 if (!bpf_dump_raw_ok(f_cred))
3358 insns[i].imm = 0;
3359 continue;
3360 }
3361 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3362 insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3363 continue;
3364 }
3365
3366 if (code != (BPF_LD | BPF_IMM | BPF_DW))
3367 continue;
3368
3369 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3370 map = bpf_map_from_imm(prog, imm, &off, &type);
3371 if (map) {
3372 insns[i].src_reg = type;
3373 insns[i].imm = map->id;
3374 insns[i + 1].imm = off;
3375 continue;
3376 }
3377 }
3378
3379 return insns;
3380}
3381
3382static int set_info_rec_size(struct bpf_prog_info *info)
3383{
3384 /*
3385 * Ensure info.*_rec_size is the same as kernel expected size
3386 *
3387 * or
3388 *
3389 * Only allow zero *_rec_size if both _rec_size and _cnt are
3390 * zero. In this case, the kernel will set the expected
3391 * _rec_size back to the info.
3392 */
3393
3394 if ((info->nr_func_info || info->func_info_rec_size) &&
3395 info->func_info_rec_size != sizeof(struct bpf_func_info))
3396 return -EINVAL;
3397
3398 if ((info->nr_line_info || info->line_info_rec_size) &&
3399 info->line_info_rec_size != sizeof(struct bpf_line_info))
3400 return -EINVAL;
3401
3402 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3403 info->jited_line_info_rec_size != sizeof(__u64))
3404 return -EINVAL;
3405
3406 info->func_info_rec_size = sizeof(struct bpf_func_info);
3407 info->line_info_rec_size = sizeof(struct bpf_line_info);
3408 info->jited_line_info_rec_size = sizeof(__u64);
3409
3410 return 0;
3411}
3412
3413static int bpf_prog_get_info_by_fd(struct file *file,
3414 struct bpf_prog *prog,
3415 const union bpf_attr *attr,
3416 union bpf_attr __user *uattr)
3417{
3418 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3419 struct bpf_prog_info info;
3420 u32 info_len = attr->info.info_len;
3421 struct bpf_prog_stats stats;
3422 char __user *uinsns;
3423 u32 ulen;
3424 int err;
3425
3426 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3427 if (err)
3428 return err;
3429 info_len = min_t(u32, sizeof(info), info_len);
3430
3431 memset(&info, 0, sizeof(info));
3432 if (copy_from_user(&info, uinfo, info_len))
3433 return -EFAULT;
3434
3435 info.type = prog->type;
3436 info.id = prog->aux->id;
3437 info.load_time = prog->aux->load_time;
3438 info.created_by_uid = from_kuid_munged(current_user_ns(),
3439 prog->aux->user->uid);
3440 info.gpl_compatible = prog->gpl_compatible;
3441
3442 memcpy(info.tag, prog->tag, sizeof(prog->tag));
3443 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3444
3445 mutex_lock(&prog->aux->used_maps_mutex);
3446 ulen = info.nr_map_ids;
3447 info.nr_map_ids = prog->aux->used_map_cnt;
3448 ulen = min_t(u32, info.nr_map_ids, ulen);
3449 if (ulen) {
3450 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3451 u32 i;
3452
3453 for (i = 0; i < ulen; i++)
3454 if (put_user(prog->aux->used_maps[i]->id,
3455 &user_map_ids[i])) {
3456 mutex_unlock(&prog->aux->used_maps_mutex);
3457 return -EFAULT;
3458 }
3459 }
3460 mutex_unlock(&prog->aux->used_maps_mutex);
3461
3462 err = set_info_rec_size(&info);
3463 if (err)
3464 return err;
3465
3466 bpf_prog_get_stats(prog, &stats);
3467 info.run_time_ns = stats.nsecs;
3468 info.run_cnt = stats.cnt;
3469 info.recursion_misses = stats.misses;
3470
3471 if (!bpf_capable()) {
3472 info.jited_prog_len = 0;
3473 info.xlated_prog_len = 0;
3474 info.nr_jited_ksyms = 0;
3475 info.nr_jited_func_lens = 0;
3476 info.nr_func_info = 0;
3477 info.nr_line_info = 0;
3478 info.nr_jited_line_info = 0;
3479 goto done;
3480 }
3481
3482 ulen = info.xlated_prog_len;
3483 info.xlated_prog_len = bpf_prog_insn_size(prog);
3484 if (info.xlated_prog_len && ulen) {
3485 struct bpf_insn *insns_sanitized;
3486 bool fault;
3487
3488 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
3489 info.xlated_prog_insns = 0;
3490 goto done;
3491 }
3492 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
3493 if (!insns_sanitized)
3494 return -ENOMEM;
3495 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3496 ulen = min_t(u32, info.xlated_prog_len, ulen);
3497 fault = copy_to_user(uinsns, insns_sanitized, ulen);
3498 kfree(insns_sanitized);
3499 if (fault)
3500 return -EFAULT;
3501 }
3502
3503 if (bpf_prog_is_dev_bound(prog->aux)) {
3504 err = bpf_prog_offload_info_fill(&info, prog);
3505 if (err)
3506 return err;
3507 goto done;
3508 }
3509
3510 /* NOTE: the following code is supposed to be skipped for offload.
3511 * bpf_prog_offload_info_fill() is the place to fill similar fields
3512 * for offload.
3513 */
3514 ulen = info.jited_prog_len;
3515 if (prog->aux->func_cnt) {
3516 u32 i;
3517
3518 info.jited_prog_len = 0;
3519 for (i = 0; i < prog->aux->func_cnt; i++)
3520 info.jited_prog_len += prog->aux->func[i]->jited_len;
3521 } else {
3522 info.jited_prog_len = prog->jited_len;
3523 }
3524
3525 if (info.jited_prog_len && ulen) {
3526 if (bpf_dump_raw_ok(file->f_cred)) {
3527 uinsns = u64_to_user_ptr(info.jited_prog_insns);
3528 ulen = min_t(u32, info.jited_prog_len, ulen);
3529
3530 /* for multi-function programs, copy the JITed
3531 * instructions for all the functions
3532 */
3533 if (prog->aux->func_cnt) {
3534 u32 len, free, i;
3535 u8 *img;
3536
3537 free = ulen;
3538 for (i = 0; i < prog->aux->func_cnt; i++) {
3539 len = prog->aux->func[i]->jited_len;
3540 len = min_t(u32, len, free);
3541 img = (u8 *) prog->aux->func[i]->bpf_func;
3542 if (copy_to_user(uinsns, img, len))
3543 return -EFAULT;
3544 uinsns += len;
3545 free -= len;
3546 if (!free)
3547 break;
3548 }
3549 } else {
3550 if (copy_to_user(uinsns, prog->bpf_func, ulen))
3551 return -EFAULT;
3552 }
3553 } else {
3554 info.jited_prog_insns = 0;
3555 }
3556 }
3557
3558 ulen = info.nr_jited_ksyms;
3559 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
3560 if (ulen) {
3561 if (bpf_dump_raw_ok(file->f_cred)) {
3562 unsigned long ksym_addr;
3563 u64 __user *user_ksyms;
3564 u32 i;
3565
3566 /* copy the address of the kernel symbol
3567 * corresponding to each function
3568 */
3569 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
3570 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
3571 if (prog->aux->func_cnt) {
3572 for (i = 0; i < ulen; i++) {
3573 ksym_addr = (unsigned long)
3574 prog->aux->func[i]->bpf_func;
3575 if (put_user((u64) ksym_addr,
3576 &user_ksyms[i]))
3577 return -EFAULT;
3578 }
3579 } else {
3580 ksym_addr = (unsigned long) prog->bpf_func;
3581 if (put_user((u64) ksym_addr, &user_ksyms[0]))
3582 return -EFAULT;
3583 }
3584 } else {
3585 info.jited_ksyms = 0;
3586 }
3587 }
3588
3589 ulen = info.nr_jited_func_lens;
3590 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
3591 if (ulen) {
3592 if (bpf_dump_raw_ok(file->f_cred)) {
3593 u32 __user *user_lens;
3594 u32 func_len, i;
3595
3596 /* copy the JITed image lengths for each function */
3597 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
3598 user_lens = u64_to_user_ptr(info.jited_func_lens);
3599 if (prog->aux->func_cnt) {
3600 for (i = 0; i < ulen; i++) {
3601 func_len =
3602 prog->aux->func[i]->jited_len;
3603 if (put_user(func_len, &user_lens[i]))
3604 return -EFAULT;
3605 }
3606 } else {
3607 func_len = prog->jited_len;
3608 if (put_user(func_len, &user_lens[0]))
3609 return -EFAULT;
3610 }
3611 } else {
3612 info.jited_func_lens = 0;
3613 }
3614 }
3615
3616 if (prog->aux->btf)
3617 info.btf_id = btf_obj_id(prog->aux->btf);
3618
3619 ulen = info.nr_func_info;
3620 info.nr_func_info = prog->aux->func_info_cnt;
3621 if (info.nr_func_info && ulen) {
3622 char __user *user_finfo;
3623
3624 user_finfo = u64_to_user_ptr(info.func_info);
3625 ulen = min_t(u32, info.nr_func_info, ulen);
3626 if (copy_to_user(user_finfo, prog->aux->func_info,
3627 info.func_info_rec_size * ulen))
3628 return -EFAULT;
3629 }
3630
3631 ulen = info.nr_line_info;
3632 info.nr_line_info = prog->aux->nr_linfo;
3633 if (info.nr_line_info && ulen) {
3634 __u8 __user *user_linfo;
3635
3636 user_linfo = u64_to_user_ptr(info.line_info);
3637 ulen = min_t(u32, info.nr_line_info, ulen);
3638 if (copy_to_user(user_linfo, prog->aux->linfo,
3639 info.line_info_rec_size * ulen))
3640 return -EFAULT;
3641 }
3642
3643 ulen = info.nr_jited_line_info;
3644 if (prog->aux->jited_linfo)
3645 info.nr_jited_line_info = prog->aux->nr_linfo;
3646 else
3647 info.nr_jited_line_info = 0;
3648 if (info.nr_jited_line_info && ulen) {
3649 if (bpf_dump_raw_ok(file->f_cred)) {
3650 __u64 __user *user_linfo;
3651 u32 i;
3652
3653 user_linfo = u64_to_user_ptr(info.jited_line_info);
3654 ulen = min_t(u32, info.nr_jited_line_info, ulen);
3655 for (i = 0; i < ulen; i++) {
3656 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
3657 &user_linfo[i]))
3658 return -EFAULT;
3659 }
3660 } else {
3661 info.jited_line_info = 0;
3662 }
3663 }
3664
3665 ulen = info.nr_prog_tags;
3666 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
3667 if (ulen) {
3668 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
3669 u32 i;
3670
3671 user_prog_tags = u64_to_user_ptr(info.prog_tags);
3672 ulen = min_t(u32, info.nr_prog_tags, ulen);
3673 if (prog->aux->func_cnt) {
3674 for (i = 0; i < ulen; i++) {
3675 if (copy_to_user(user_prog_tags[i],
3676 prog->aux->func[i]->tag,
3677 BPF_TAG_SIZE))
3678 return -EFAULT;
3679 }
3680 } else {
3681 if (copy_to_user(user_prog_tags[0],
3682 prog->tag, BPF_TAG_SIZE))
3683 return -EFAULT;
3684 }
3685 }
3686
3687done:
3688 if (copy_to_user(uinfo, &info, info_len) ||
3689 put_user(info_len, &uattr->info.info_len))
3690 return -EFAULT;
3691
3692 return 0;
3693}
3694
3695static int bpf_map_get_info_by_fd(struct file *file,
3696 struct bpf_map *map,
3697 const union bpf_attr *attr,
3698 union bpf_attr __user *uattr)
3699{
3700 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3701 struct bpf_map_info info;
3702 u32 info_len = attr->info.info_len;
3703 int err;
3704
3705 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3706 if (err)
3707 return err;
3708 info_len = min_t(u32, sizeof(info), info_len);
3709
3710 memset(&info, 0, sizeof(info));
3711 info.type = map->map_type;
3712 info.id = map->id;
3713 info.key_size = map->key_size;
3714 info.value_size = map->value_size;
3715 info.max_entries = map->max_entries;
3716 info.map_flags = map->map_flags;
3717 memcpy(info.name, map->name, sizeof(map->name));
3718
3719 if (map->btf) {
3720 info.btf_id = btf_obj_id(map->btf);
3721 info.btf_key_type_id = map->btf_key_type_id;
3722 info.btf_value_type_id = map->btf_value_type_id;
3723 }
3724 info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
3725
3726 if (bpf_map_is_dev_bound(map)) {
3727 err = bpf_map_offload_info_fill(&info, map);
3728 if (err)
3729 return err;
3730 }
3731
3732 if (copy_to_user(uinfo, &info, info_len) ||
3733 put_user(info_len, &uattr->info.info_len))
3734 return -EFAULT;
3735
3736 return 0;
3737}
3738
3739static int bpf_btf_get_info_by_fd(struct file *file,
3740 struct btf *btf,
3741 const union bpf_attr *attr,
3742 union bpf_attr __user *uattr)
3743{
3744 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3745 u32 info_len = attr->info.info_len;
3746 int err;
3747
3748 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
3749 if (err)
3750 return err;
3751
3752 return btf_get_info_by_fd(btf, attr, uattr);
3753}
3754
3755static int bpf_link_get_info_by_fd(struct file *file,
3756 struct bpf_link *link,
3757 const union bpf_attr *attr,
3758 union bpf_attr __user *uattr)
3759{
3760 struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3761 struct bpf_link_info info;
3762 u32 info_len = attr->info.info_len;
3763 int err;
3764
3765 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3766 if (err)
3767 return err;
3768 info_len = min_t(u32, sizeof(info), info_len);
3769
3770 memset(&info, 0, sizeof(info));
3771 if (copy_from_user(&info, uinfo, info_len))
3772 return -EFAULT;
3773
3774 info.type = link->type;
3775 info.id = link->id;
3776 info.prog_id = link->prog->aux->id;
3777
3778 if (link->ops->fill_link_info) {
3779 err = link->ops->fill_link_info(link, &info);
3780 if (err)
3781 return err;
3782 }
3783
3784 if (copy_to_user(uinfo, &info, info_len) ||
3785 put_user(info_len, &uattr->info.info_len))
3786 return -EFAULT;
3787
3788 return 0;
3789}
3790
3791
3792#define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
3793
3794static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
3795 union bpf_attr __user *uattr)
3796{
3797 int ufd = attr->info.bpf_fd;
3798 struct fd f;
3799 int err;
3800
3801 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
3802 return -EINVAL;
3803
3804 f = fdget(ufd);
3805 if (!f.file)
3806 return -EBADFD;
3807
3808 if (f.file->f_op == &bpf_prog_fops)
3809 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
3810 uattr);
3811 else if (f.file->f_op == &bpf_map_fops)
3812 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
3813 uattr);
3814 else if (f.file->f_op == &btf_fops)
3815 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
3816 else if (f.file->f_op == &bpf_link_fops)
3817 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
3818 attr, uattr);
3819 else
3820 err = -EINVAL;
3821
3822 fdput(f);
3823 return err;
3824}
3825
3826#define BPF_BTF_LOAD_LAST_FIELD btf_log_level
3827
3828static int bpf_btf_load(const union bpf_attr *attr)
3829{
3830 if (CHECK_ATTR(BPF_BTF_LOAD))
3831 return -EINVAL;
3832
3833 if (!bpf_capable())
3834 return -EPERM;
3835
3836 return btf_new_fd(attr);
3837}
3838
3839#define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
3840
3841static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
3842{
3843 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
3844 return -EINVAL;
3845
3846 if (!capable(CAP_SYS_ADMIN))
3847 return -EPERM;
3848
3849 return btf_get_fd_by_id(attr->btf_id);
3850}
3851
3852static int bpf_task_fd_query_copy(const union bpf_attr *attr,
3853 union bpf_attr __user *uattr,
3854 u32 prog_id, u32 fd_type,
3855 const char *buf, u64 probe_offset,
3856 u64 probe_addr)
3857{
3858 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
3859 u32 len = buf ? strlen(buf) : 0, input_len;
3860 int err = 0;
3861
3862 if (put_user(len, &uattr->task_fd_query.buf_len))
3863 return -EFAULT;
3864 input_len = attr->task_fd_query.buf_len;
3865 if (input_len && ubuf) {
3866 if (!len) {
3867 /* nothing to copy, just make ubuf NULL terminated */
3868 char zero = '\0';
3869
3870 if (put_user(zero, ubuf))
3871 return -EFAULT;
3872 } else if (input_len >= len + 1) {
3873 /* ubuf can hold the string with NULL terminator */
3874 if (copy_to_user(ubuf, buf, len + 1))
3875 return -EFAULT;
3876 } else {
3877 /* ubuf cannot hold the string with NULL terminator,
3878 * do a partial copy with NULL terminator.
3879 */
3880 char zero = '\0';
3881
3882 err = -ENOSPC;
3883 if (copy_to_user(ubuf, buf, input_len - 1))
3884 return -EFAULT;
3885 if (put_user(zero, ubuf + input_len - 1))
3886 return -EFAULT;
3887 }
3888 }
3889
3890 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
3891 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
3892 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
3893 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
3894 return -EFAULT;
3895
3896 return err;
3897}
3898
3899#define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
3900
3901static int bpf_task_fd_query(const union bpf_attr *attr,
3902 union bpf_attr __user *uattr)
3903{
3904 pid_t pid = attr->task_fd_query.pid;
3905 u32 fd = attr->task_fd_query.fd;
3906 const struct perf_event *event;
3907 struct task_struct *task;
3908 struct file *file;
3909 int err;
3910
3911 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
3912 return -EINVAL;
3913
3914 if (!capable(CAP_SYS_ADMIN))
3915 return -EPERM;
3916
3917 if (attr->task_fd_query.flags != 0)
3918 return -EINVAL;
3919
3920 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3921 if (!task)
3922 return -ENOENT;
3923
3924 err = 0;
3925 file = fget_task(task, fd);
3926 put_task_struct(task);
3927 if (!file)
3928 return -EBADF;
3929
3930 if (file->f_op == &bpf_link_fops) {
3931 struct bpf_link *link = file->private_data;
3932
3933 if (link->ops == &bpf_raw_tp_link_lops) {
3934 struct bpf_raw_tp_link *raw_tp =
3935 container_of(link, struct bpf_raw_tp_link, link);
3936 struct bpf_raw_event_map *btp = raw_tp->btp;
3937
3938 err = bpf_task_fd_query_copy(attr, uattr,
3939 raw_tp->link.prog->aux->id,
3940 BPF_FD_TYPE_RAW_TRACEPOINT,
3941 btp->tp->name, 0, 0);
3942 goto put_file;
3943 }
3944 goto out_not_supp;
3945 }
3946
3947 event = perf_get_event(file);
3948 if (!IS_ERR(event)) {
3949 u64 probe_offset, probe_addr;
3950 u32 prog_id, fd_type;
3951 const char *buf;
3952
3953 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
3954 &buf, &probe_offset,
3955 &probe_addr);
3956 if (!err)
3957 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
3958 fd_type, buf,
3959 probe_offset,
3960 probe_addr);
3961 goto put_file;
3962 }
3963
3964out_not_supp:
3965 err = -ENOTSUPP;
3966put_file:
3967 fput(file);
3968 return err;
3969}
3970
3971#define BPF_MAP_BATCH_LAST_FIELD batch.flags
3972
3973#define BPF_DO_BATCH(fn) \
3974 do { \
3975 if (!fn) { \
3976 err = -ENOTSUPP; \
3977 goto err_put; \
3978 } \
3979 err = fn(map, attr, uattr); \
3980 } while (0)
3981
3982static int bpf_map_do_batch(const union bpf_attr *attr,
3983 union bpf_attr __user *uattr,
3984 int cmd)
3985{
3986 struct bpf_map *map;
3987 int err, ufd;
3988 struct fd f;
3989
3990 if (CHECK_ATTR(BPF_MAP_BATCH))
3991 return -EINVAL;
3992
3993 ufd = attr->batch.map_fd;
3994 f = fdget(ufd);
3995 map = __bpf_map_get(f);
3996 if (IS_ERR(map))
3997 return PTR_ERR(map);
3998
3999 if ((cmd == BPF_MAP_LOOKUP_BATCH ||
4000 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH) &&
4001 !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4002 err = -EPERM;
4003 goto err_put;
4004 }
4005
4006 if (cmd != BPF_MAP_LOOKUP_BATCH &&
4007 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4008 err = -EPERM;
4009 goto err_put;
4010 }
4011
4012 if (cmd == BPF_MAP_LOOKUP_BATCH)
4013 BPF_DO_BATCH(map->ops->map_lookup_batch);
4014 else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4015 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
4016 else if (cmd == BPF_MAP_UPDATE_BATCH)
4017 BPF_DO_BATCH(map->ops->map_update_batch);
4018 else
4019 BPF_DO_BATCH(map->ops->map_delete_batch);
4020
4021err_put:
4022 fdput(f);
4023 return err;
4024}
4025
4026static int tracing_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
4027{
4028 if (attr->link_create.attach_type != prog->expected_attach_type)
4029 return -EINVAL;
4030
4031 if (prog->expected_attach_type == BPF_TRACE_ITER)
4032 return bpf_iter_link_attach(attr, prog);
4033 else if (prog->type == BPF_PROG_TYPE_EXT)
4034 return bpf_tracing_prog_attach(prog,
4035 attr->link_create.target_fd,
4036 attr->link_create.target_btf_id);
4037 return -EINVAL;
4038}
4039
4040#define BPF_LINK_CREATE_LAST_FIELD link_create.iter_info_len
4041static int link_create(union bpf_attr *attr)
4042{
4043 enum bpf_prog_type ptype;
4044 struct bpf_prog *prog;
4045 int ret;
4046
4047 if (CHECK_ATTR(BPF_LINK_CREATE))
4048 return -EINVAL;
4049
4050 prog = bpf_prog_get(attr->link_create.prog_fd);
4051 if (IS_ERR(prog))
4052 return PTR_ERR(prog);
4053
4054 ret = bpf_prog_attach_check_attach_type(prog,
4055 attr->link_create.attach_type);
4056 if (ret)
4057 goto out;
4058
4059 if (prog->type == BPF_PROG_TYPE_EXT) {
4060 ret = tracing_bpf_link_attach(attr, prog);
4061 goto out;
4062 }
4063
4064 ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4065 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4066 ret = -EINVAL;
4067 goto out;
4068 }
4069
4070 switch (ptype) {
4071 case BPF_PROG_TYPE_CGROUP_SKB:
4072 case BPF_PROG_TYPE_CGROUP_SOCK:
4073 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4074 case BPF_PROG_TYPE_SOCK_OPS:
4075 case BPF_PROG_TYPE_CGROUP_DEVICE:
4076 case BPF_PROG_TYPE_CGROUP_SYSCTL:
4077 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4078 ret = cgroup_bpf_link_attach(attr, prog);
4079 break;
4080 case BPF_PROG_TYPE_TRACING:
4081 ret = tracing_bpf_link_attach(attr, prog);
4082 break;
4083 case BPF_PROG_TYPE_FLOW_DISSECTOR:
4084 case BPF_PROG_TYPE_SK_LOOKUP:
4085 ret = netns_bpf_link_create(attr, prog);
4086 break;
4087#ifdef CONFIG_NET
4088 case BPF_PROG_TYPE_XDP:
4089 ret = bpf_xdp_link_attach(attr, prog);
4090 break;
4091#endif
4092 default:
4093 ret = -EINVAL;
4094 }
4095
4096out:
4097 if (ret < 0)
4098 bpf_prog_put(prog);
4099 return ret;
4100}
4101
4102#define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4103
4104static int link_update(union bpf_attr *attr)
4105{
4106 struct bpf_prog *old_prog = NULL, *new_prog;
4107 struct bpf_link *link;
4108 u32 flags;
4109 int ret;
4110
4111 if (CHECK_ATTR(BPF_LINK_UPDATE))
4112 return -EINVAL;
4113
4114 flags = attr->link_update.flags;
4115 if (flags & ~BPF_F_REPLACE)
4116 return -EINVAL;
4117
4118 link = bpf_link_get_from_fd(attr->link_update.link_fd);
4119 if (IS_ERR(link))
4120 return PTR_ERR(link);
4121
4122 new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4123 if (IS_ERR(new_prog)) {
4124 ret = PTR_ERR(new_prog);
4125 goto out_put_link;
4126 }
4127
4128 if (flags & BPF_F_REPLACE) {
4129 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4130 if (IS_ERR(old_prog)) {
4131 ret = PTR_ERR(old_prog);
4132 old_prog = NULL;
4133 goto out_put_progs;
4134 }
4135 } else if (attr->link_update.old_prog_fd) {
4136 ret = -EINVAL;
4137 goto out_put_progs;
4138 }
4139
4140 if (link->ops->update_prog)
4141 ret = link->ops->update_prog(link, new_prog, old_prog);
4142 else
4143 ret = -EINVAL;
4144
4145out_put_progs:
4146 if (old_prog)
4147 bpf_prog_put(old_prog);
4148 if (ret)
4149 bpf_prog_put(new_prog);
4150out_put_link:
4151 bpf_link_put(link);
4152 return ret;
4153}
4154
4155#define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4156
4157static int link_detach(union bpf_attr *attr)
4158{
4159 struct bpf_link *link;
4160 int ret;
4161
4162 if (CHECK_ATTR(BPF_LINK_DETACH))
4163 return -EINVAL;
4164
4165 link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4166 if (IS_ERR(link))
4167 return PTR_ERR(link);
4168
4169 if (link->ops->detach)
4170 ret = link->ops->detach(link);
4171 else
4172 ret = -EOPNOTSUPP;
4173
4174 bpf_link_put(link);
4175 return ret;
4176}
4177
4178static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4179{
4180 return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4181}
4182
4183struct bpf_link *bpf_link_by_id(u32 id)
4184{
4185 struct bpf_link *link;
4186
4187 if (!id)
4188 return ERR_PTR(-ENOENT);
4189
4190 spin_lock_bh(&link_idr_lock);
4191 /* before link is "settled", ID is 0, pretend it doesn't exist yet */
4192 link = idr_find(&link_idr, id);
4193 if (link) {
4194 if (link->id)
4195 link = bpf_link_inc_not_zero(link);
4196 else
4197 link = ERR_PTR(-EAGAIN);
4198 } else {
4199 link = ERR_PTR(-ENOENT);
4200 }
4201 spin_unlock_bh(&link_idr_lock);
4202 return link;
4203}
4204
4205#define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4206
4207static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4208{
4209 struct bpf_link *link;
4210 u32 id = attr->link_id;
4211 int fd;
4212
4213 if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4214 return -EINVAL;
4215
4216 if (!capable(CAP_SYS_ADMIN))
4217 return -EPERM;
4218
4219 link = bpf_link_by_id(id);
4220 if (IS_ERR(link))
4221 return PTR_ERR(link);
4222
4223 fd = bpf_link_new_fd(link);
4224 if (fd < 0)
4225 bpf_link_put(link);
4226
4227 return fd;
4228}
4229
4230DEFINE_MUTEX(bpf_stats_enabled_mutex);
4231
4232static int bpf_stats_release(struct inode *inode, struct file *file)
4233{
4234 mutex_lock(&bpf_stats_enabled_mutex);
4235 static_key_slow_dec(&bpf_stats_enabled_key.key);
4236 mutex_unlock(&bpf_stats_enabled_mutex);
4237 return 0;
4238}
4239
4240static const struct file_operations bpf_stats_fops = {
4241 .release = bpf_stats_release,
4242};
4243
4244static int bpf_enable_runtime_stats(void)
4245{
4246 int fd;
4247
4248 mutex_lock(&bpf_stats_enabled_mutex);
4249
4250 /* Set a very high limit to avoid overflow */
4251 if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4252 mutex_unlock(&bpf_stats_enabled_mutex);
4253 return -EBUSY;
4254 }
4255
4256 fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4257 if (fd >= 0)
4258 static_key_slow_inc(&bpf_stats_enabled_key.key);
4259
4260 mutex_unlock(&bpf_stats_enabled_mutex);
4261 return fd;
4262}
4263
4264#define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4265
4266static int bpf_enable_stats(union bpf_attr *attr)
4267{
4268
4269 if (CHECK_ATTR(BPF_ENABLE_STATS))
4270 return -EINVAL;
4271
4272 if (!capable(CAP_SYS_ADMIN))
4273 return -EPERM;
4274
4275 switch (attr->enable_stats.type) {
4276 case BPF_STATS_RUN_TIME:
4277 return bpf_enable_runtime_stats();
4278 default:
4279 break;
4280 }
4281 return -EINVAL;
4282}
4283
4284#define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4285
4286static int bpf_iter_create(union bpf_attr *attr)
4287{
4288 struct bpf_link *link;
4289 int err;
4290
4291 if (CHECK_ATTR(BPF_ITER_CREATE))
4292 return -EINVAL;
4293
4294 if (attr->iter_create.flags)
4295 return -EINVAL;
4296
4297 link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4298 if (IS_ERR(link))
4299 return PTR_ERR(link);
4300
4301 err = bpf_iter_new_fd(link);
4302 bpf_link_put(link);
4303
4304 return err;
4305}
4306
4307#define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4308
4309static int bpf_prog_bind_map(union bpf_attr *attr)
4310{
4311 struct bpf_prog *prog;
4312 struct bpf_map *map;
4313 struct bpf_map **used_maps_old, **used_maps_new;
4314 int i, ret = 0;
4315
4316 if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4317 return -EINVAL;
4318
4319 if (attr->prog_bind_map.flags)
4320 return -EINVAL;
4321
4322 prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4323 if (IS_ERR(prog))
4324 return PTR_ERR(prog);
4325
4326 map = bpf_map_get(attr->prog_bind_map.map_fd);
4327 if (IS_ERR(map)) {
4328 ret = PTR_ERR(map);
4329 goto out_prog_put;
4330 }
4331
4332 mutex_lock(&prog->aux->used_maps_mutex);
4333
4334 used_maps_old = prog->aux->used_maps;
4335
4336 for (i = 0; i < prog->aux->used_map_cnt; i++)
4337 if (used_maps_old[i] == map) {
4338 bpf_map_put(map);
4339 goto out_unlock;
4340 }
4341
4342 used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4343 sizeof(used_maps_new[0]),
4344 GFP_KERNEL);
4345 if (!used_maps_new) {
4346 ret = -ENOMEM;
4347 goto out_unlock;
4348 }
4349
4350 memcpy(used_maps_new, used_maps_old,
4351 sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4352 used_maps_new[prog->aux->used_map_cnt] = map;
4353
4354 prog->aux->used_map_cnt++;
4355 prog->aux->used_maps = used_maps_new;
4356
4357 kfree(used_maps_old);
4358
4359out_unlock:
4360 mutex_unlock(&prog->aux->used_maps_mutex);
4361
4362 if (ret)
4363 bpf_map_put(map);
4364out_prog_put:
4365 bpf_prog_put(prog);
4366 return ret;
4367}
4368
4369SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
4370{
4371 union bpf_attr attr;
4372 int err;
4373
4374 if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
4375 return -EPERM;
4376
4377 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
4378 if (err)
4379 return err;
4380 size = min_t(u32, size, sizeof(attr));
4381
4382 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
4383 memset(&attr, 0, sizeof(attr));
4384 if (copy_from_user(&attr, uattr, size) != 0)
4385 return -EFAULT;
4386
4387 err = security_bpf(cmd, &attr, size);
4388 if (err < 0)
4389 return err;
4390
4391 switch (cmd) {
4392 case BPF_MAP_CREATE:
4393 err = map_create(&attr);
4394 break;
4395 case BPF_MAP_LOOKUP_ELEM:
4396 err = map_lookup_elem(&attr);
4397 break;
4398 case BPF_MAP_UPDATE_ELEM:
4399 err = map_update_elem(&attr);
4400 break;
4401 case BPF_MAP_DELETE_ELEM:
4402 err = map_delete_elem(&attr);
4403 break;
4404 case BPF_MAP_GET_NEXT_KEY:
4405 err = map_get_next_key(&attr);
4406 break;
4407 case BPF_MAP_FREEZE:
4408 err = map_freeze(&attr);
4409 break;
4410 case BPF_PROG_LOAD:
4411 err = bpf_prog_load(&attr, uattr);
4412 break;
4413 case BPF_OBJ_PIN:
4414 err = bpf_obj_pin(&attr);
4415 break;
4416 case BPF_OBJ_GET:
4417 err = bpf_obj_get(&attr);
4418 break;
4419 case BPF_PROG_ATTACH:
4420 err = bpf_prog_attach(&attr);
4421 break;
4422 case BPF_PROG_DETACH:
4423 err = bpf_prog_detach(&attr);
4424 break;
4425 case BPF_PROG_QUERY:
4426 err = bpf_prog_query(&attr, uattr);
4427 break;
4428 case BPF_PROG_TEST_RUN:
4429 err = bpf_prog_test_run(&attr, uattr);
4430 break;
4431 case BPF_PROG_GET_NEXT_ID:
4432 err = bpf_obj_get_next_id(&attr, uattr,
4433 &prog_idr, &prog_idr_lock);
4434 break;
4435 case BPF_MAP_GET_NEXT_ID:
4436 err = bpf_obj_get_next_id(&attr, uattr,
4437 &map_idr, &map_idr_lock);
4438 break;
4439 case BPF_BTF_GET_NEXT_ID:
4440 err = bpf_obj_get_next_id(&attr, uattr,
4441 &btf_idr, &btf_idr_lock);
4442 break;
4443 case BPF_PROG_GET_FD_BY_ID:
4444 err = bpf_prog_get_fd_by_id(&attr);
4445 break;
4446 case BPF_MAP_GET_FD_BY_ID:
4447 err = bpf_map_get_fd_by_id(&attr);
4448 break;
4449 case BPF_OBJ_GET_INFO_BY_FD:
4450 err = bpf_obj_get_info_by_fd(&attr, uattr);
4451 break;
4452 case BPF_RAW_TRACEPOINT_OPEN:
4453 err = bpf_raw_tracepoint_open(&attr);
4454 break;
4455 case BPF_BTF_LOAD:
4456 err = bpf_btf_load(&attr);
4457 break;
4458 case BPF_BTF_GET_FD_BY_ID:
4459 err = bpf_btf_get_fd_by_id(&attr);
4460 break;
4461 case BPF_TASK_FD_QUERY:
4462 err = bpf_task_fd_query(&attr, uattr);
4463 break;
4464 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
4465 err = map_lookup_and_delete_elem(&attr);
4466 break;
4467 case BPF_MAP_LOOKUP_BATCH:
4468 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_LOOKUP_BATCH);
4469 break;
4470 case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
4471 err = bpf_map_do_batch(&attr, uattr,
4472 BPF_MAP_LOOKUP_AND_DELETE_BATCH);
4473 break;
4474 case BPF_MAP_UPDATE_BATCH:
4475 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_UPDATE_BATCH);
4476 break;
4477 case BPF_MAP_DELETE_BATCH:
4478 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_DELETE_BATCH);
4479 break;
4480 case BPF_LINK_CREATE:
4481 err = link_create(&attr);
4482 break;
4483 case BPF_LINK_UPDATE:
4484 err = link_update(&attr);
4485 break;
4486 case BPF_LINK_GET_FD_BY_ID:
4487 err = bpf_link_get_fd_by_id(&attr);
4488 break;
4489 case BPF_LINK_GET_NEXT_ID:
4490 err = bpf_obj_get_next_id(&attr, uattr,
4491 &link_idr, &link_idr_lock);
4492 break;
4493 case BPF_ENABLE_STATS:
4494 err = bpf_enable_stats(&attr);
4495 break;
4496 case BPF_ITER_CREATE:
4497 err = bpf_iter_create(&attr);
4498 break;
4499 case BPF_LINK_DETACH:
4500 err = link_detach(&attr);
4501 break;
4502 case BPF_PROG_BIND_MAP:
4503 err = bpf_prog_bind_map(&attr);
4504 break;
4505 default:
4506 err = -EINVAL;
4507 break;
4508 }
4509
4510 return err;
4511}