1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
5 #include <linux/bpf-cgroup.h>
6 #include <linux/bpf_trace.h>
7 #include <linux/bpf_lirc.h>
8 #include <linux/bpf_verifier.h>
10 #include <linux/syscalls.h>
11 #include <linux/slab.h>
12 #include <linux/sched/signal.h>
13 #include <linux/vmalloc.h>
14 #include <linux/mmzone.h>
15 #include <linux/anon_inodes.h>
16 #include <linux/fdtable.h>
17 #include <linux/file.h>
19 #include <linux/license.h>
20 #include <linux/filter.h>
21 #include <linux/kernel.h>
22 #include <linux/idr.h>
23 #include <linux/cred.h>
24 #include <linux/timekeeping.h>
25 #include <linux/ctype.h>
26 #include <linux/nospec.h>
27 #include <linux/audit.h>
28 #include <uapi/linux/btf.h>
29 #include <linux/pgtable.h>
30 #include <linux/bpf_lsm.h>
31 #include <linux/poll.h>
32 #include <linux/bpf-netns.h>
33 #include <linux/rcupdate_trace.h>
34 #include <linux/memcontrol.h>
36 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
37 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
38 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
39 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
40 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
41 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
44 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
46 DEFINE_PER_CPU(int, bpf_prog_active);
47 static DEFINE_IDR(prog_idr);
48 static DEFINE_SPINLOCK(prog_idr_lock);
49 static DEFINE_IDR(map_idr);
50 static DEFINE_SPINLOCK(map_idr_lock);
51 static DEFINE_IDR(link_idr);
52 static DEFINE_SPINLOCK(link_idr_lock);
54 int sysctl_unprivileged_bpf_disabled __read_mostly =
55 IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
57 static const struct bpf_map_ops * const bpf_map_types[] = {
58 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
59 #define BPF_MAP_TYPE(_id, _ops) \
61 #define BPF_LINK_TYPE(_id, _name)
62 #include <linux/bpf_types.h>
69 * If we're handed a bigger struct than we know of, ensure all the unknown bits
70 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
71 * we don't know about yet.
73 * There is a ToCToU between this function call and the following
74 * copy_from_user() call. However, this is not a concern since this function is
75 * meant to be a future-proofing of bits.
77 int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
83 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
86 if (actual_size <= expected_size)
90 res = memchr_inv(uaddr.kernel + expected_size, 0,
91 actual_size - expected_size) == NULL;
93 res = check_zeroed_user(uaddr.user + expected_size,
94 actual_size - expected_size);
97 return res ? 0 : -E2BIG;
100 const struct bpf_map_ops bpf_map_offload_ops = {
101 .map_meta_equal = bpf_map_meta_equal,
102 .map_alloc = bpf_map_offload_map_alloc,
103 .map_free = bpf_map_offload_map_free,
104 .map_check_btf = map_check_no_btf,
107 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
109 const struct bpf_map_ops *ops;
110 u32 type = attr->map_type;
114 if (type >= ARRAY_SIZE(bpf_map_types))
115 return ERR_PTR(-EINVAL);
116 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
117 ops = bpf_map_types[type];
119 return ERR_PTR(-EINVAL);
121 if (ops->map_alloc_check) {
122 err = ops->map_alloc_check(attr);
126 if (attr->map_ifindex)
127 ops = &bpf_map_offload_ops;
128 map = ops->map_alloc(attr);
132 map->map_type = type;
136 static void bpf_map_write_active_inc(struct bpf_map *map)
138 atomic64_inc(&map->writecnt);
141 static void bpf_map_write_active_dec(struct bpf_map *map)
143 atomic64_dec(&map->writecnt);
146 bool bpf_map_write_active(const struct bpf_map *map)
148 return atomic64_read(&map->writecnt) != 0;
151 static u32 bpf_map_value_size(const struct bpf_map *map)
153 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
154 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
155 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
156 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
157 return round_up(map->value_size, 8) * num_possible_cpus();
158 else if (IS_FD_MAP(map))
161 return map->value_size;
164 static void maybe_wait_bpf_programs(struct bpf_map *map)
166 /* Wait for any running BPF programs to complete so that
167 * userspace, when we return to it, knows that all programs
168 * that could be running use the new map value.
170 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
171 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
175 static int bpf_map_update_value(struct bpf_map *map, struct fd f, void *key,
176 void *value, __u64 flags)
180 /* Need to create a kthread, thus must support schedule */
181 if (bpf_map_is_dev_bound(map)) {
182 return bpf_map_offload_update_elem(map, key, value, flags);
183 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
184 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
185 return map->ops->map_update_elem(map, key, value, flags);
186 } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
187 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
188 return sock_map_update_elem_sys(map, key, value, flags);
189 } else if (IS_FD_PROG_ARRAY(map)) {
190 return bpf_fd_array_map_update_elem(map, f.file, key, value,
194 bpf_disable_instrumentation();
195 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
196 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
197 err = bpf_percpu_hash_update(map, key, value, flags);
198 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
199 err = bpf_percpu_array_update(map, key, value, flags);
200 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
201 err = bpf_percpu_cgroup_storage_update(map, key, value,
203 } else if (IS_FD_ARRAY(map)) {
205 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
208 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
210 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
213 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
214 /* rcu_read_lock() is not needed */
215 err = bpf_fd_reuseport_array_update_elem(map, key, value,
217 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
218 map->map_type == BPF_MAP_TYPE_STACK ||
219 map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
220 err = map->ops->map_push_elem(map, value, flags);
223 err = map->ops->map_update_elem(map, key, value, flags);
226 bpf_enable_instrumentation();
227 maybe_wait_bpf_programs(map);
232 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
238 if (bpf_map_is_dev_bound(map))
239 return bpf_map_offload_lookup_elem(map, key, value);
241 bpf_disable_instrumentation();
242 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
243 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
244 err = bpf_percpu_hash_copy(map, key, value);
245 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
246 err = bpf_percpu_array_copy(map, key, value);
247 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
248 err = bpf_percpu_cgroup_storage_copy(map, key, value);
249 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
250 err = bpf_stackmap_copy(map, key, value);
251 } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
252 err = bpf_fd_array_map_lookup_elem(map, key, value);
253 } else if (IS_FD_HASH(map)) {
254 err = bpf_fd_htab_map_lookup_elem(map, key, value);
255 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
256 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
257 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
258 map->map_type == BPF_MAP_TYPE_STACK ||
259 map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
260 err = map->ops->map_peek_elem(map, value);
261 } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
262 /* struct_ops map requires directly updating "value" */
263 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
266 if (map->ops->map_lookup_elem_sys_only)
267 ptr = map->ops->map_lookup_elem_sys_only(map, key);
269 ptr = map->ops->map_lookup_elem(map, key);
276 if (flags & BPF_F_LOCK)
277 /* lock 'ptr' and copy everything but lock */
278 copy_map_value_locked(map, value, ptr, true);
280 copy_map_value(map, value, ptr);
281 /* mask lock and timer, since value wasn't zero inited */
282 check_and_init_map_value(map, value);
287 bpf_enable_instrumentation();
288 maybe_wait_bpf_programs(map);
293 /* Please, do not use this function outside from the map creation path
294 * (e.g. in map update path) without taking care of setting the active
295 * memory cgroup (see at bpf_map_kmalloc_node() for example).
297 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
299 /* We really just want to fail instead of triggering OOM killer
300 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
301 * which is used for lower order allocation requests.
303 * It has been observed that higher order allocation requests done by
304 * vmalloc with __GFP_NORETRY being set might fail due to not trying
305 * to reclaim memory from the page cache, thus we set
306 * __GFP_RETRY_MAYFAIL to avoid such situations.
309 const gfp_t gfp = __GFP_NOWARN | __GFP_ZERO | __GFP_ACCOUNT;
310 unsigned int flags = 0;
311 unsigned long align = 1;
314 if (size >= SIZE_MAX)
317 /* kmalloc()'ed memory can't be mmap()'ed */
319 BUG_ON(!PAGE_ALIGNED(size));
322 } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
323 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
329 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
330 gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
331 flags, numa_node, __builtin_return_address(0));
334 void *bpf_map_area_alloc(u64 size, int numa_node)
336 return __bpf_map_area_alloc(size, numa_node, false);
339 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
341 return __bpf_map_area_alloc(size, numa_node, true);
344 void bpf_map_area_free(void *area)
349 static u32 bpf_map_flags_retain_permanent(u32 flags)
351 /* Some map creation flags are not tied to the map object but
352 * rather to the map fd instead, so they have no meaning upon
353 * map object inspection since multiple file descriptors with
354 * different (access) properties can exist here. Thus, given
355 * this has zero meaning for the map itself, lets clear these
358 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
361 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
363 map->map_type = attr->map_type;
364 map->key_size = attr->key_size;
365 map->value_size = attr->value_size;
366 map->max_entries = attr->max_entries;
367 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
368 map->numa_node = bpf_map_attr_numa_node(attr);
369 map->map_extra = attr->map_extra;
372 static int bpf_map_alloc_id(struct bpf_map *map)
376 idr_preload(GFP_KERNEL);
377 spin_lock_bh(&map_idr_lock);
378 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
381 spin_unlock_bh(&map_idr_lock);
384 if (WARN_ON_ONCE(!id))
387 return id > 0 ? 0 : id;
390 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
394 /* Offloaded maps are removed from the IDR store when their device
395 * disappears - even if someone holds an fd to them they are unusable,
396 * the memory is gone, all ops will fail; they are simply waiting for
397 * refcnt to drop to be freed.
403 spin_lock_irqsave(&map_idr_lock, flags);
405 __acquire(&map_idr_lock);
407 idr_remove(&map_idr, map->id);
411 spin_unlock_irqrestore(&map_idr_lock, flags);
413 __release(&map_idr_lock);
416 #ifdef CONFIG_MEMCG_KMEM
417 static void bpf_map_save_memcg(struct bpf_map *map)
419 map->memcg = get_mem_cgroup_from_mm(current->mm);
422 static void bpf_map_release_memcg(struct bpf_map *map)
424 mem_cgroup_put(map->memcg);
427 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
430 struct mem_cgroup *old_memcg;
433 old_memcg = set_active_memcg(map->memcg);
434 ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
435 set_active_memcg(old_memcg);
440 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
442 struct mem_cgroup *old_memcg;
445 old_memcg = set_active_memcg(map->memcg);
446 ptr = kzalloc(size, flags | __GFP_ACCOUNT);
447 set_active_memcg(old_memcg);
452 void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
453 size_t align, gfp_t flags)
455 struct mem_cgroup *old_memcg;
458 old_memcg = set_active_memcg(map->memcg);
459 ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
460 set_active_memcg(old_memcg);
466 static void bpf_map_save_memcg(struct bpf_map *map)
470 static void bpf_map_release_memcg(struct bpf_map *map)
475 /* called from workqueue */
476 static void bpf_map_free_deferred(struct work_struct *work)
478 struct bpf_map *map = container_of(work, struct bpf_map, work);
480 security_bpf_map_free(map);
481 bpf_map_release_memcg(map);
482 /* implementation dependent freeing */
483 map->ops->map_free(map);
486 static void bpf_map_put_uref(struct bpf_map *map)
488 if (atomic64_dec_and_test(&map->usercnt)) {
489 if (map->ops->map_release_uref)
490 map->ops->map_release_uref(map);
494 /* decrement map refcnt and schedule it for freeing via workqueue
495 * (unrelying map implementation ops->map_free() might sleep)
497 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
499 if (atomic64_dec_and_test(&map->refcnt)) {
500 /* bpf_map_free_id() must be called first */
501 bpf_map_free_id(map, do_idr_lock);
503 INIT_WORK(&map->work, bpf_map_free_deferred);
504 schedule_work(&map->work);
508 void bpf_map_put(struct bpf_map *map)
510 __bpf_map_put(map, true);
512 EXPORT_SYMBOL_GPL(bpf_map_put);
514 void bpf_map_put_with_uref(struct bpf_map *map)
516 bpf_map_put_uref(map);
520 static int bpf_map_release(struct inode *inode, struct file *filp)
522 struct bpf_map *map = filp->private_data;
524 if (map->ops->map_release)
525 map->ops->map_release(map, filp);
527 bpf_map_put_with_uref(map);
531 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
533 fmode_t mode = f.file->f_mode;
535 /* Our file permissions may have been overridden by global
536 * map permissions facing syscall side.
538 if (READ_ONCE(map->frozen))
539 mode &= ~FMODE_CAN_WRITE;
543 #ifdef CONFIG_PROC_FS
544 /* Provides an approximation of the map's memory footprint.
545 * Used only to provide a backward compatibility and display
546 * a reasonable "memlock" info.
548 static unsigned long bpf_map_memory_footprint(const struct bpf_map *map)
552 size = round_up(map->key_size + bpf_map_value_size(map), 8);
554 return round_up(map->max_entries * size, PAGE_SIZE);
557 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
559 const struct bpf_map *map = filp->private_data;
560 const struct bpf_array *array;
561 u32 type = 0, jited = 0;
563 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
564 array = container_of(map, struct bpf_array, map);
565 spin_lock(&array->aux->owner.lock);
566 type = array->aux->owner.type;
567 jited = array->aux->owner.jited;
568 spin_unlock(&array->aux->owner.lock);
577 "map_extra:\t%#llx\n"
586 (unsigned long long)map->map_extra,
587 bpf_map_memory_footprint(map),
589 READ_ONCE(map->frozen));
591 seq_printf(m, "owner_prog_type:\t%u\n", type);
592 seq_printf(m, "owner_jited:\t%u\n", jited);
597 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
600 /* We need this handler such that alloc_file() enables
601 * f_mode with FMODE_CAN_READ.
606 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
607 size_t siz, loff_t *ppos)
609 /* We need this handler such that alloc_file() enables
610 * f_mode with FMODE_CAN_WRITE.
615 /* called for any extra memory-mapped regions (except initial) */
616 static void bpf_map_mmap_open(struct vm_area_struct *vma)
618 struct bpf_map *map = vma->vm_file->private_data;
620 if (vma->vm_flags & VM_MAYWRITE)
621 bpf_map_write_active_inc(map);
624 /* called for all unmapped memory region (including initial) */
625 static void bpf_map_mmap_close(struct vm_area_struct *vma)
627 struct bpf_map *map = vma->vm_file->private_data;
629 if (vma->vm_flags & VM_MAYWRITE)
630 bpf_map_write_active_dec(map);
633 static const struct vm_operations_struct bpf_map_default_vmops = {
634 .open = bpf_map_mmap_open,
635 .close = bpf_map_mmap_close,
638 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
640 struct bpf_map *map = filp->private_data;
643 if (!map->ops->map_mmap || map_value_has_spin_lock(map) ||
644 map_value_has_timer(map))
647 if (!(vma->vm_flags & VM_SHARED))
650 mutex_lock(&map->freeze_mutex);
652 if (vma->vm_flags & VM_WRITE) {
657 /* map is meant to be read-only, so do not allow mapping as
658 * writable, because it's possible to leak a writable page
659 * reference and allows user-space to still modify it after
660 * freezing, while verifier will assume contents do not change
662 if (map->map_flags & BPF_F_RDONLY_PROG) {
668 /* set default open/close callbacks */
669 vma->vm_ops = &bpf_map_default_vmops;
670 vma->vm_private_data = map;
671 vma->vm_flags &= ~VM_MAYEXEC;
672 if (!(vma->vm_flags & VM_WRITE))
673 /* disallow re-mapping with PROT_WRITE */
674 vma->vm_flags &= ~VM_MAYWRITE;
676 err = map->ops->map_mmap(map, vma);
680 if (vma->vm_flags & VM_MAYWRITE)
681 bpf_map_write_active_inc(map);
683 mutex_unlock(&map->freeze_mutex);
687 static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
689 struct bpf_map *map = filp->private_data;
691 if (map->ops->map_poll)
692 return map->ops->map_poll(map, filp, pts);
697 const struct file_operations bpf_map_fops = {
698 #ifdef CONFIG_PROC_FS
699 .show_fdinfo = bpf_map_show_fdinfo,
701 .release = bpf_map_release,
702 .read = bpf_dummy_read,
703 .write = bpf_dummy_write,
704 .mmap = bpf_map_mmap,
705 .poll = bpf_map_poll,
708 int bpf_map_new_fd(struct bpf_map *map, int flags)
712 ret = security_bpf_map(map, OPEN_FMODE(flags));
716 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
720 int bpf_get_file_flag(int flags)
722 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
724 if (flags & BPF_F_RDONLY)
726 if (flags & BPF_F_WRONLY)
731 /* helper macro to check that unused fields 'union bpf_attr' are zero */
732 #define CHECK_ATTR(CMD) \
733 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
734 sizeof(attr->CMD##_LAST_FIELD), 0, \
736 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
737 sizeof(attr->CMD##_LAST_FIELD)) != NULL
739 /* dst and src must have at least "size" number of bytes.
740 * Return strlen on success and < 0 on error.
742 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
744 const char *end = src + size;
745 const char *orig_src = src;
747 memset(dst, 0, size);
748 /* Copy all isalnum(), '_' and '.' chars. */
749 while (src < end && *src) {
750 if (!isalnum(*src) &&
751 *src != '_' && *src != '.')
756 /* No '\0' found in "size" number of bytes */
760 return src - orig_src;
763 int map_check_no_btf(const struct bpf_map *map,
764 const struct btf *btf,
765 const struct btf_type *key_type,
766 const struct btf_type *value_type)
771 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
772 u32 btf_key_id, u32 btf_value_id)
774 const struct btf_type *key_type, *value_type;
775 u32 key_size, value_size;
778 /* Some maps allow key to be unspecified. */
780 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
781 if (!key_type || key_size != map->key_size)
784 key_type = btf_type_by_id(btf, 0);
785 if (!map->ops->map_check_btf)
789 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
790 if (!value_type || value_size != map->value_size)
793 map->spin_lock_off = btf_find_spin_lock(btf, value_type);
795 if (map_value_has_spin_lock(map)) {
796 if (map->map_flags & BPF_F_RDONLY_PROG)
798 if (map->map_type != BPF_MAP_TYPE_HASH &&
799 map->map_type != BPF_MAP_TYPE_ARRAY &&
800 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
801 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
802 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
803 map->map_type != BPF_MAP_TYPE_TASK_STORAGE)
805 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
808 "verifier bug spin_lock_off %d value_size %d\n",
809 map->spin_lock_off, map->value_size);
814 map->timer_off = btf_find_timer(btf, value_type);
815 if (map_value_has_timer(map)) {
816 if (map->map_flags & BPF_F_RDONLY_PROG)
818 if (map->map_type != BPF_MAP_TYPE_HASH &&
819 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
820 map->map_type != BPF_MAP_TYPE_ARRAY)
824 if (map->ops->map_check_btf)
825 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
830 #define BPF_MAP_CREATE_LAST_FIELD map_extra
831 /* called via syscall */
832 static int map_create(union bpf_attr *attr)
834 int numa_node = bpf_map_attr_numa_node(attr);
839 err = CHECK_ATTR(BPF_MAP_CREATE);
843 if (attr->btf_vmlinux_value_type_id) {
844 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
845 attr->btf_key_type_id || attr->btf_value_type_id)
847 } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
851 if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
852 attr->map_extra != 0)
855 f_flags = bpf_get_file_flag(attr->map_flags);
859 if (numa_node != NUMA_NO_NODE &&
860 ((unsigned int)numa_node >= nr_node_ids ||
861 !node_online(numa_node)))
864 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
865 map = find_and_alloc_map(attr);
869 err = bpf_obj_name_cpy(map->name, attr->map_name,
870 sizeof(attr->map_name));
874 atomic64_set(&map->refcnt, 1);
875 atomic64_set(&map->usercnt, 1);
876 mutex_init(&map->freeze_mutex);
878 map->spin_lock_off = -EINVAL;
879 map->timer_off = -EINVAL;
880 if (attr->btf_key_type_id || attr->btf_value_type_id ||
881 /* Even the map's value is a kernel's struct,
882 * the bpf_prog.o must have BTF to begin with
883 * to figure out the corresponding kernel's
884 * counter part. Thus, attr->btf_fd has
887 attr->btf_vmlinux_value_type_id) {
890 btf = btf_get_by_fd(attr->btf_fd);
895 if (btf_is_kernel(btf)) {
902 if (attr->btf_value_type_id) {
903 err = map_check_btf(map, btf, attr->btf_key_type_id,
904 attr->btf_value_type_id);
909 map->btf_key_type_id = attr->btf_key_type_id;
910 map->btf_value_type_id = attr->btf_value_type_id;
911 map->btf_vmlinux_value_type_id =
912 attr->btf_vmlinux_value_type_id;
915 err = security_bpf_map_alloc(map);
919 err = bpf_map_alloc_id(map);
923 bpf_map_save_memcg(map);
925 err = bpf_map_new_fd(map, f_flags);
927 /* failed to allocate fd.
928 * bpf_map_put_with_uref() is needed because the above
929 * bpf_map_alloc_id() has published the map
930 * to the userspace and the userspace may
931 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
933 bpf_map_put_with_uref(map);
940 security_bpf_map_free(map);
943 map->ops->map_free(map);
947 /* if error is returned, fd is released.
948 * On success caller should complete fd access with matching fdput()
950 struct bpf_map *__bpf_map_get(struct fd f)
953 return ERR_PTR(-EBADF);
954 if (f.file->f_op != &bpf_map_fops) {
956 return ERR_PTR(-EINVAL);
959 return f.file->private_data;
962 void bpf_map_inc(struct bpf_map *map)
964 atomic64_inc(&map->refcnt);
966 EXPORT_SYMBOL_GPL(bpf_map_inc);
968 void bpf_map_inc_with_uref(struct bpf_map *map)
970 atomic64_inc(&map->refcnt);
971 atomic64_inc(&map->usercnt);
973 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
975 struct bpf_map *bpf_map_get(u32 ufd)
977 struct fd f = fdget(ufd);
980 map = __bpf_map_get(f);
990 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
992 struct fd f = fdget(ufd);
995 map = __bpf_map_get(f);
999 bpf_map_inc_with_uref(map);
1005 /* map_idr_lock should have been held */
1006 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1010 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1012 return ERR_PTR(-ENOENT);
1014 atomic64_inc(&map->usercnt);
1019 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1021 spin_lock_bh(&map_idr_lock);
1022 map = __bpf_map_inc_not_zero(map, false);
1023 spin_unlock_bh(&map_idr_lock);
1027 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1029 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1034 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1037 return vmemdup_user(ukey, key_size);
1040 return ERR_PTR(-EINVAL);
1045 static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1048 return kvmemdup_bpfptr(ukey, key_size);
1050 if (!bpfptr_is_null(ukey))
1051 return ERR_PTR(-EINVAL);
1056 /* last field in 'union bpf_attr' used by this command */
1057 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1059 static int map_lookup_elem(union bpf_attr *attr)
1061 void __user *ukey = u64_to_user_ptr(attr->key);
1062 void __user *uvalue = u64_to_user_ptr(attr->value);
1063 int ufd = attr->map_fd;
1064 struct bpf_map *map;
1070 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1073 if (attr->flags & ~BPF_F_LOCK)
1077 map = __bpf_map_get(f);
1079 return PTR_ERR(map);
1080 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1085 if ((attr->flags & BPF_F_LOCK) &&
1086 !map_value_has_spin_lock(map)) {
1091 key = __bpf_copy_key(ukey, map->key_size);
1097 value_size = bpf_map_value_size(map);
1100 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1104 if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1105 if (copy_from_user(value, uvalue, value_size))
1108 err = bpf_map_copy_value(map, key, value, attr->flags);
1112 err = bpf_map_copy_value(map, key, value, attr->flags);
1117 if (copy_to_user(uvalue, value, value_size) != 0)
1132 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1134 static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1136 bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1137 bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1138 int ufd = attr->map_fd;
1139 struct bpf_map *map;
1145 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1149 map = __bpf_map_get(f);
1151 return PTR_ERR(map);
1152 bpf_map_write_active_inc(map);
1153 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1158 if ((attr->flags & BPF_F_LOCK) &&
1159 !map_value_has_spin_lock(map)) {
1164 key = ___bpf_copy_key(ukey, map->key_size);
1170 value_size = bpf_map_value_size(map);
1173 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1178 if (copy_from_bpfptr(value, uvalue, value_size) != 0)
1181 err = bpf_map_update_value(map, f, key, value, attr->flags);
1188 bpf_map_write_active_dec(map);
1193 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1195 static int map_delete_elem(union bpf_attr *attr)
1197 void __user *ukey = u64_to_user_ptr(attr->key);
1198 int ufd = attr->map_fd;
1199 struct bpf_map *map;
1204 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1208 map = __bpf_map_get(f);
1210 return PTR_ERR(map);
1211 bpf_map_write_active_inc(map);
1212 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1217 key = __bpf_copy_key(ukey, map->key_size);
1223 if (bpf_map_is_dev_bound(map)) {
1224 err = bpf_map_offload_delete_elem(map, key);
1226 } else if (IS_FD_PROG_ARRAY(map) ||
1227 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1228 /* These maps require sleepable context */
1229 err = map->ops->map_delete_elem(map, key);
1233 bpf_disable_instrumentation();
1235 err = map->ops->map_delete_elem(map, key);
1237 bpf_enable_instrumentation();
1238 maybe_wait_bpf_programs(map);
1242 bpf_map_write_active_dec(map);
1247 /* last field in 'union bpf_attr' used by this command */
1248 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1250 static int map_get_next_key(union bpf_attr *attr)
1252 void __user *ukey = u64_to_user_ptr(attr->key);
1253 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1254 int ufd = attr->map_fd;
1255 struct bpf_map *map;
1256 void *key, *next_key;
1260 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1264 map = __bpf_map_get(f);
1266 return PTR_ERR(map);
1267 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1273 key = __bpf_copy_key(ukey, map->key_size);
1283 next_key = kvmalloc(map->key_size, GFP_USER);
1287 if (bpf_map_is_dev_bound(map)) {
1288 err = bpf_map_offload_get_next_key(map, key, next_key);
1293 err = map->ops->map_get_next_key(map, key, next_key);
1300 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1314 int generic_map_delete_batch(struct bpf_map *map,
1315 const union bpf_attr *attr,
1316 union bpf_attr __user *uattr)
1318 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1323 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1326 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1327 !map_value_has_spin_lock(map)) {
1331 max_count = attr->batch.count;
1335 key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1339 for (cp = 0; cp < max_count; cp++) {
1341 if (copy_from_user(key, keys + cp * map->key_size,
1345 if (bpf_map_is_dev_bound(map)) {
1346 err = bpf_map_offload_delete_elem(map, key);
1350 bpf_disable_instrumentation();
1352 err = map->ops->map_delete_elem(map, key);
1354 bpf_enable_instrumentation();
1355 maybe_wait_bpf_programs(map);
1359 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1366 int generic_map_update_batch(struct bpf_map *map,
1367 const union bpf_attr *attr,
1368 union bpf_attr __user *uattr)
1370 void __user *values = u64_to_user_ptr(attr->batch.values);
1371 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1372 u32 value_size, cp, max_count;
1373 int ufd = attr->batch.map_fd;
1378 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1381 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1382 !map_value_has_spin_lock(map)) {
1386 value_size = bpf_map_value_size(map);
1388 max_count = attr->batch.count;
1392 key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1396 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1402 f = fdget(ufd); /* bpf_map_do_batch() guarantees ufd is valid */
1403 for (cp = 0; cp < max_count; cp++) {
1405 if (copy_from_user(key, keys + cp * map->key_size,
1407 copy_from_user(value, values + cp * value_size, value_size))
1410 err = bpf_map_update_value(map, f, key, value,
1411 attr->batch.elem_flags);
1417 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1426 #define MAP_LOOKUP_RETRIES 3
1428 int generic_map_lookup_batch(struct bpf_map *map,
1429 const union bpf_attr *attr,
1430 union bpf_attr __user *uattr)
1432 void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1433 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1434 void __user *values = u64_to_user_ptr(attr->batch.values);
1435 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1436 void *buf, *buf_prevkey, *prev_key, *key, *value;
1437 int err, retry = MAP_LOOKUP_RETRIES;
1438 u32 value_size, cp, max_count;
1440 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1443 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1444 !map_value_has_spin_lock(map))
1447 value_size = bpf_map_value_size(map);
1449 max_count = attr->batch.count;
1453 if (put_user(0, &uattr->batch.count))
1456 buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1460 buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1462 kvfree(buf_prevkey);
1468 if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1471 value = key + map->key_size;
1473 prev_key = buf_prevkey;
1475 for (cp = 0; cp < max_count;) {
1477 err = map->ops->map_get_next_key(map, prev_key, key);
1481 err = bpf_map_copy_value(map, key, value,
1482 attr->batch.elem_flags);
1484 if (err == -ENOENT) {
1496 if (copy_to_user(keys + cp * map->key_size, key,
1501 if (copy_to_user(values + cp * value_size, value, value_size)) {
1507 prev_key = buf_prevkey;
1509 swap(prev_key, key);
1510 retry = MAP_LOOKUP_RETRIES;
1517 if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1518 (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1522 kvfree(buf_prevkey);
1527 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1529 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1531 void __user *ukey = u64_to_user_ptr(attr->key);
1532 void __user *uvalue = u64_to_user_ptr(attr->value);
1533 int ufd = attr->map_fd;
1534 struct bpf_map *map;
1540 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1543 if (attr->flags & ~BPF_F_LOCK)
1547 map = __bpf_map_get(f);
1549 return PTR_ERR(map);
1550 bpf_map_write_active_inc(map);
1551 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1552 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1558 (map->map_type == BPF_MAP_TYPE_QUEUE ||
1559 map->map_type == BPF_MAP_TYPE_STACK)) {
1564 if ((attr->flags & BPF_F_LOCK) &&
1565 !map_value_has_spin_lock(map)) {
1570 key = __bpf_copy_key(ukey, map->key_size);
1576 value_size = bpf_map_value_size(map);
1579 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1584 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1585 map->map_type == BPF_MAP_TYPE_STACK) {
1586 err = map->ops->map_pop_elem(map, value);
1587 } else if (map->map_type == BPF_MAP_TYPE_HASH ||
1588 map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1589 map->map_type == BPF_MAP_TYPE_LRU_HASH ||
1590 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1591 if (!bpf_map_is_dev_bound(map)) {
1592 bpf_disable_instrumentation();
1594 err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
1596 bpf_enable_instrumentation();
1603 if (copy_to_user(uvalue, value, value_size) != 0) {
1615 bpf_map_write_active_dec(map);
1620 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1622 static int map_freeze(const union bpf_attr *attr)
1624 int err = 0, ufd = attr->map_fd;
1625 struct bpf_map *map;
1628 if (CHECK_ATTR(BPF_MAP_FREEZE))
1632 map = __bpf_map_get(f);
1634 return PTR_ERR(map);
1636 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS ||
1637 map_value_has_timer(map)) {
1642 mutex_lock(&map->freeze_mutex);
1643 if (bpf_map_write_active(map)) {
1647 if (READ_ONCE(map->frozen)) {
1651 if (!bpf_capable()) {
1656 WRITE_ONCE(map->frozen, true);
1658 mutex_unlock(&map->freeze_mutex);
1663 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1664 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1665 [_id] = & _name ## _prog_ops,
1666 #define BPF_MAP_TYPE(_id, _ops)
1667 #define BPF_LINK_TYPE(_id, _name)
1668 #include <linux/bpf_types.h>
1669 #undef BPF_PROG_TYPE
1671 #undef BPF_LINK_TYPE
1674 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1676 const struct bpf_prog_ops *ops;
1678 if (type >= ARRAY_SIZE(bpf_prog_types))
1680 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1681 ops = bpf_prog_types[type];
1685 if (!bpf_prog_is_dev_bound(prog->aux))
1686 prog->aux->ops = ops;
1688 prog->aux->ops = &bpf_offload_prog_ops;
1699 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1700 [BPF_AUDIT_LOAD] = "LOAD",
1701 [BPF_AUDIT_UNLOAD] = "UNLOAD",
1704 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1706 struct audit_context *ctx = NULL;
1707 struct audit_buffer *ab;
1709 if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1711 if (audit_enabled == AUDIT_OFF)
1713 if (op == BPF_AUDIT_LOAD)
1714 ctx = audit_context();
1715 ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1718 audit_log_format(ab, "prog-id=%u op=%s",
1719 prog->aux->id, bpf_audit_str[op]);
1723 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1727 idr_preload(GFP_KERNEL);
1728 spin_lock_bh(&prog_idr_lock);
1729 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1732 spin_unlock_bh(&prog_idr_lock);
1735 /* id is in [1, INT_MAX) */
1736 if (WARN_ON_ONCE(!id))
1739 return id > 0 ? 0 : id;
1742 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1744 unsigned long flags;
1746 /* cBPF to eBPF migrations are currently not in the idr store.
1747 * Offloaded programs are removed from the store when their device
1748 * disappears - even if someone grabs an fd to them they are unusable,
1749 * simply waiting for refcnt to drop to be freed.
1755 spin_lock_irqsave(&prog_idr_lock, flags);
1757 __acquire(&prog_idr_lock);
1759 idr_remove(&prog_idr, prog->aux->id);
1763 spin_unlock_irqrestore(&prog_idr_lock, flags);
1765 __release(&prog_idr_lock);
1768 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1770 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1772 kvfree(aux->func_info);
1773 kfree(aux->func_info_aux);
1774 free_uid(aux->user);
1775 security_bpf_prog_free(aux);
1776 bpf_prog_free(aux->prog);
1779 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1781 bpf_prog_kallsyms_del_all(prog);
1782 btf_put(prog->aux->btf);
1783 kvfree(prog->aux->jited_linfo);
1784 kvfree(prog->aux->linfo);
1785 kfree(prog->aux->kfunc_tab);
1786 if (prog->aux->attach_btf)
1787 btf_put(prog->aux->attach_btf);
1790 if (prog->aux->sleepable)
1791 call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
1793 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1795 __bpf_prog_put_rcu(&prog->aux->rcu);
1799 static void bpf_prog_put_deferred(struct work_struct *work)
1801 struct bpf_prog_aux *aux;
1802 struct bpf_prog *prog;
1804 aux = container_of(work, struct bpf_prog_aux, work);
1806 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1807 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
1808 __bpf_prog_put_noref(prog, true);
1811 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1813 struct bpf_prog_aux *aux = prog->aux;
1815 if (atomic64_dec_and_test(&aux->refcnt)) {
1816 /* bpf_prog_free_id() must be called first */
1817 bpf_prog_free_id(prog, do_idr_lock);
1819 if (in_irq() || irqs_disabled()) {
1820 INIT_WORK(&aux->work, bpf_prog_put_deferred);
1821 schedule_work(&aux->work);
1823 bpf_prog_put_deferred(&aux->work);
1828 void bpf_prog_put(struct bpf_prog *prog)
1830 __bpf_prog_put(prog, true);
1832 EXPORT_SYMBOL_GPL(bpf_prog_put);
1834 static int bpf_prog_release(struct inode *inode, struct file *filp)
1836 struct bpf_prog *prog = filp->private_data;
1842 struct bpf_prog_kstats {
1848 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1849 struct bpf_prog_kstats *stats)
1851 u64 nsecs = 0, cnt = 0, misses = 0;
1854 for_each_possible_cpu(cpu) {
1855 const struct bpf_prog_stats *st;
1857 u64 tnsecs, tcnt, tmisses;
1859 st = per_cpu_ptr(prog->stats, cpu);
1861 start = u64_stats_fetch_begin_irq(&st->syncp);
1862 tnsecs = u64_stats_read(&st->nsecs);
1863 tcnt = u64_stats_read(&st->cnt);
1864 tmisses = u64_stats_read(&st->misses);
1865 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1870 stats->nsecs = nsecs;
1872 stats->misses = misses;
1875 #ifdef CONFIG_PROC_FS
1876 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1878 const struct bpf_prog *prog = filp->private_data;
1879 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1880 struct bpf_prog_kstats stats;
1882 bpf_prog_get_stats(prog, &stats);
1883 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1890 "run_time_ns:\t%llu\n"
1892 "recursion_misses:\t%llu\n"
1893 "verified_insns:\t%u\n",
1897 prog->pages * 1ULL << PAGE_SHIFT,
1902 prog->aux->verified_insns);
1906 const struct file_operations bpf_prog_fops = {
1907 #ifdef CONFIG_PROC_FS
1908 .show_fdinfo = bpf_prog_show_fdinfo,
1910 .release = bpf_prog_release,
1911 .read = bpf_dummy_read,
1912 .write = bpf_dummy_write,
1915 int bpf_prog_new_fd(struct bpf_prog *prog)
1919 ret = security_bpf_prog(prog);
1923 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1924 O_RDWR | O_CLOEXEC);
1927 static struct bpf_prog *____bpf_prog_get(struct fd f)
1930 return ERR_PTR(-EBADF);
1931 if (f.file->f_op != &bpf_prog_fops) {
1933 return ERR_PTR(-EINVAL);
1936 return f.file->private_data;
1939 void bpf_prog_add(struct bpf_prog *prog, int i)
1941 atomic64_add(i, &prog->aux->refcnt);
1943 EXPORT_SYMBOL_GPL(bpf_prog_add);
1945 void bpf_prog_sub(struct bpf_prog *prog, int i)
1947 /* Only to be used for undoing previous bpf_prog_add() in some
1948 * error path. We still know that another entity in our call
1949 * path holds a reference to the program, thus atomic_sub() can
1950 * be safely used in such cases!
1952 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1954 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1956 void bpf_prog_inc(struct bpf_prog *prog)
1958 atomic64_inc(&prog->aux->refcnt);
1960 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1962 /* prog_idr_lock should have been held */
1963 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1967 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1970 return ERR_PTR(-ENOENT);
1974 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1976 bool bpf_prog_get_ok(struct bpf_prog *prog,
1977 enum bpf_prog_type *attach_type, bool attach_drv)
1979 /* not an attachment, just a refcount inc, always allow */
1983 if (prog->type != *attach_type)
1985 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1991 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1994 struct fd f = fdget(ufd);
1995 struct bpf_prog *prog;
1997 prog = ____bpf_prog_get(f);
2000 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
2001 prog = ERR_PTR(-EINVAL);
2011 struct bpf_prog *bpf_prog_get(u32 ufd)
2013 return __bpf_prog_get(ufd, NULL, false);
2016 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2019 return __bpf_prog_get(ufd, &type, attach_drv);
2021 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2023 /* Initially all BPF programs could be loaded w/o specifying
2024 * expected_attach_type. Later for some of them specifying expected_attach_type
2025 * at load time became required so that program could be validated properly.
2026 * Programs of types that are allowed to be loaded both w/ and w/o (for
2027 * backward compatibility) expected_attach_type, should have the default attach
2028 * type assigned to expected_attach_type for the latter case, so that it can be
2029 * validated later at attach time.
2031 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2032 * prog type requires it but has some attach types that have to be backward
2035 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2037 switch (attr->prog_type) {
2038 case BPF_PROG_TYPE_CGROUP_SOCK:
2039 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2040 * exist so checking for non-zero is the way to go here.
2042 if (!attr->expected_attach_type)
2043 attr->expected_attach_type =
2044 BPF_CGROUP_INET_SOCK_CREATE;
2046 case BPF_PROG_TYPE_SK_REUSEPORT:
2047 if (!attr->expected_attach_type)
2048 attr->expected_attach_type =
2049 BPF_SK_REUSEPORT_SELECT;
2055 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2056 enum bpf_attach_type expected_attach_type,
2057 struct btf *attach_btf, u32 btf_id,
2058 struct bpf_prog *dst_prog)
2061 if (btf_id > BTF_MAX_TYPE)
2064 if (!attach_btf && !dst_prog)
2067 switch (prog_type) {
2068 case BPF_PROG_TYPE_TRACING:
2069 case BPF_PROG_TYPE_LSM:
2070 case BPF_PROG_TYPE_STRUCT_OPS:
2071 case BPF_PROG_TYPE_EXT:
2078 if (attach_btf && (!btf_id || dst_prog))
2081 if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2082 prog_type != BPF_PROG_TYPE_EXT)
2085 switch (prog_type) {
2086 case BPF_PROG_TYPE_CGROUP_SOCK:
2087 switch (expected_attach_type) {
2088 case BPF_CGROUP_INET_SOCK_CREATE:
2089 case BPF_CGROUP_INET_SOCK_RELEASE:
2090 case BPF_CGROUP_INET4_POST_BIND:
2091 case BPF_CGROUP_INET6_POST_BIND:
2096 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2097 switch (expected_attach_type) {
2098 case BPF_CGROUP_INET4_BIND:
2099 case BPF_CGROUP_INET6_BIND:
2100 case BPF_CGROUP_INET4_CONNECT:
2101 case BPF_CGROUP_INET6_CONNECT:
2102 case BPF_CGROUP_INET4_GETPEERNAME:
2103 case BPF_CGROUP_INET6_GETPEERNAME:
2104 case BPF_CGROUP_INET4_GETSOCKNAME:
2105 case BPF_CGROUP_INET6_GETSOCKNAME:
2106 case BPF_CGROUP_UDP4_SENDMSG:
2107 case BPF_CGROUP_UDP6_SENDMSG:
2108 case BPF_CGROUP_UDP4_RECVMSG:
2109 case BPF_CGROUP_UDP6_RECVMSG:
2114 case BPF_PROG_TYPE_CGROUP_SKB:
2115 switch (expected_attach_type) {
2116 case BPF_CGROUP_INET_INGRESS:
2117 case BPF_CGROUP_INET_EGRESS:
2122 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2123 switch (expected_attach_type) {
2124 case BPF_CGROUP_SETSOCKOPT:
2125 case BPF_CGROUP_GETSOCKOPT:
2130 case BPF_PROG_TYPE_SK_LOOKUP:
2131 if (expected_attach_type == BPF_SK_LOOKUP)
2134 case BPF_PROG_TYPE_SK_REUSEPORT:
2135 switch (expected_attach_type) {
2136 case BPF_SK_REUSEPORT_SELECT:
2137 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2142 case BPF_PROG_TYPE_SYSCALL:
2143 case BPF_PROG_TYPE_EXT:
2144 if (expected_attach_type)
2152 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2154 switch (prog_type) {
2155 case BPF_PROG_TYPE_SCHED_CLS:
2156 case BPF_PROG_TYPE_SCHED_ACT:
2157 case BPF_PROG_TYPE_XDP:
2158 case BPF_PROG_TYPE_LWT_IN:
2159 case BPF_PROG_TYPE_LWT_OUT:
2160 case BPF_PROG_TYPE_LWT_XMIT:
2161 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2162 case BPF_PROG_TYPE_SK_SKB:
2163 case BPF_PROG_TYPE_SK_MSG:
2164 case BPF_PROG_TYPE_LIRC_MODE2:
2165 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2166 case BPF_PROG_TYPE_CGROUP_DEVICE:
2167 case BPF_PROG_TYPE_CGROUP_SOCK:
2168 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2169 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2170 case BPF_PROG_TYPE_CGROUP_SYSCTL:
2171 case BPF_PROG_TYPE_SOCK_OPS:
2172 case BPF_PROG_TYPE_EXT: /* extends any prog */
2174 case BPF_PROG_TYPE_CGROUP_SKB:
2176 case BPF_PROG_TYPE_SK_REUSEPORT:
2177 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2183 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2185 switch (prog_type) {
2186 case BPF_PROG_TYPE_KPROBE:
2187 case BPF_PROG_TYPE_TRACEPOINT:
2188 case BPF_PROG_TYPE_PERF_EVENT:
2189 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2190 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2191 case BPF_PROG_TYPE_TRACING:
2192 case BPF_PROG_TYPE_LSM:
2193 case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2194 case BPF_PROG_TYPE_EXT: /* extends any prog */
2201 /* last field in 'union bpf_attr' used by this command */
2202 #define BPF_PROG_LOAD_LAST_FIELD core_relo_rec_size
2204 static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr)
2206 enum bpf_prog_type type = attr->prog_type;
2207 struct bpf_prog *prog, *dst_prog = NULL;
2208 struct btf *attach_btf = NULL;
2213 if (CHECK_ATTR(BPF_PROG_LOAD))
2216 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2217 BPF_F_ANY_ALIGNMENT |
2218 BPF_F_TEST_STATE_FREQ |
2220 BPF_F_TEST_RND_HI32))
2223 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2224 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2228 /* copy eBPF program license from user space */
2229 if (strncpy_from_bpfptr(license,
2230 make_bpfptr(attr->license, uattr.is_kernel),
2231 sizeof(license) - 1) < 0)
2233 license[sizeof(license) - 1] = 0;
2235 /* eBPF programs must be GPL compatible to use GPL-ed functions */
2236 is_gpl = license_is_gpl_compatible(license);
2238 if (attr->insn_cnt == 0 ||
2239 attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2241 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2242 type != BPF_PROG_TYPE_CGROUP_SKB &&
2246 if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2248 if (is_perfmon_prog_type(type) && !perfmon_capable())
2251 /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2252 * or btf, we need to check which one it is
2254 if (attr->attach_prog_fd) {
2255 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2256 if (IS_ERR(dst_prog)) {
2258 attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2259 if (IS_ERR(attach_btf))
2261 if (!btf_is_kernel(attach_btf)) {
2262 /* attaching through specifying bpf_prog's BTF
2263 * objects directly might be supported eventually
2265 btf_put(attach_btf);
2269 } else if (attr->attach_btf_id) {
2270 /* fall back to vmlinux BTF, if BTF type ID is specified */
2271 attach_btf = bpf_get_btf_vmlinux();
2272 if (IS_ERR(attach_btf))
2273 return PTR_ERR(attach_btf);
2276 btf_get(attach_btf);
2279 bpf_prog_load_fixup_attach_type(attr);
2280 if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2281 attach_btf, attr->attach_btf_id,
2284 bpf_prog_put(dst_prog);
2286 btf_put(attach_btf);
2290 /* plain bpf_prog allocation */
2291 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2294 bpf_prog_put(dst_prog);
2296 btf_put(attach_btf);
2300 prog->expected_attach_type = attr->expected_attach_type;
2301 prog->aux->attach_btf = attach_btf;
2302 prog->aux->attach_btf_id = attr->attach_btf_id;
2303 prog->aux->dst_prog = dst_prog;
2304 prog->aux->offload_requested = !!attr->prog_ifindex;
2305 prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2307 err = security_bpf_prog_alloc(prog->aux);
2311 prog->aux->user = get_current_user();
2312 prog->len = attr->insn_cnt;
2315 if (copy_from_bpfptr(prog->insns,
2316 make_bpfptr(attr->insns, uattr.is_kernel),
2317 bpf_prog_insn_size(prog)) != 0)
2320 prog->orig_prog = NULL;
2323 atomic64_set(&prog->aux->refcnt, 1);
2324 prog->gpl_compatible = is_gpl ? 1 : 0;
2326 if (bpf_prog_is_dev_bound(prog->aux)) {
2327 err = bpf_prog_offload_init(prog, attr);
2332 /* find program type: socket_filter vs tracing_filter */
2333 err = find_prog_type(type, prog);
2337 prog->aux->load_time = ktime_get_boottime_ns();
2338 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2339 sizeof(attr->prog_name));
2343 /* run eBPF verifier */
2344 err = bpf_check(&prog, attr, uattr);
2346 goto free_used_maps;
2348 prog = bpf_prog_select_runtime(prog, &err);
2350 goto free_used_maps;
2352 err = bpf_prog_alloc_id(prog);
2354 goto free_used_maps;
2356 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2357 * effectively publicly exposed. However, retrieving via
2358 * bpf_prog_get_fd_by_id() will take another reference,
2359 * therefore it cannot be gone underneath us.
2361 * Only for the time /after/ successful bpf_prog_new_fd()
2362 * and before returning to userspace, we might just hold
2363 * one reference and any parallel close on that fd could
2364 * rip everything out. Hence, below notifications must
2365 * happen before bpf_prog_new_fd().
2367 * Also, any failure handling from this point onwards must
2368 * be using bpf_prog_put() given the program is exposed.
2370 bpf_prog_kallsyms_add(prog);
2371 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2372 bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2374 err = bpf_prog_new_fd(prog);
2380 /* In case we have subprogs, we need to wait for a grace
2381 * period before we can tear down JIT memory since symbols
2382 * are already exposed under kallsyms.
2384 __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2387 free_uid(prog->aux->user);
2388 security_bpf_prog_free(prog->aux);
2390 if (prog->aux->attach_btf)
2391 btf_put(prog->aux->attach_btf);
2392 bpf_prog_free(prog);
2396 #define BPF_OBJ_LAST_FIELD file_flags
2398 static int bpf_obj_pin(const union bpf_attr *attr)
2400 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2403 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2406 static int bpf_obj_get(const union bpf_attr *attr)
2408 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2409 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2412 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2416 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2417 const struct bpf_link_ops *ops, struct bpf_prog *prog)
2419 atomic64_set(&link->refcnt, 1);
2426 static void bpf_link_free_id(int id)
2431 spin_lock_bh(&link_idr_lock);
2432 idr_remove(&link_idr, id);
2433 spin_unlock_bh(&link_idr_lock);
2436 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2437 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2438 * anon_inode's release() call. This helper marksbpf_link as
2439 * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2440 * is not decremented, it's the responsibility of a calling code that failed
2441 * to complete bpf_link initialization.
2443 void bpf_link_cleanup(struct bpf_link_primer *primer)
2445 primer->link->prog = NULL;
2446 bpf_link_free_id(primer->id);
2448 put_unused_fd(primer->fd);
2451 void bpf_link_inc(struct bpf_link *link)
2453 atomic64_inc(&link->refcnt);
2456 /* bpf_link_free is guaranteed to be called from process context */
2457 static void bpf_link_free(struct bpf_link *link)
2459 bpf_link_free_id(link->id);
2461 /* detach BPF program, clean up used resources */
2462 link->ops->release(link);
2463 bpf_prog_put(link->prog);
2465 /* free bpf_link and its containing memory */
2466 link->ops->dealloc(link);
2469 static void bpf_link_put_deferred(struct work_struct *work)
2471 struct bpf_link *link = container_of(work, struct bpf_link, work);
2473 bpf_link_free(link);
2476 /* bpf_link_put can be called from atomic context, but ensures that resources
2477 * are freed from process context
2479 void bpf_link_put(struct bpf_link *link)
2481 if (!atomic64_dec_and_test(&link->refcnt))
2485 INIT_WORK(&link->work, bpf_link_put_deferred);
2486 schedule_work(&link->work);
2488 bpf_link_free(link);
2492 static int bpf_link_release(struct inode *inode, struct file *filp)
2494 struct bpf_link *link = filp->private_data;
2500 #ifdef CONFIG_PROC_FS
2501 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2502 #define BPF_MAP_TYPE(_id, _ops)
2503 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2504 static const char *bpf_link_type_strs[] = {
2505 [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2506 #include <linux/bpf_types.h>
2508 #undef BPF_PROG_TYPE
2510 #undef BPF_LINK_TYPE
2512 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2514 const struct bpf_link *link = filp->private_data;
2515 const struct bpf_prog *prog = link->prog;
2516 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2518 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2524 bpf_link_type_strs[link->type],
2528 if (link->ops->show_fdinfo)
2529 link->ops->show_fdinfo(link, m);
2533 static const struct file_operations bpf_link_fops = {
2534 #ifdef CONFIG_PROC_FS
2535 .show_fdinfo = bpf_link_show_fdinfo,
2537 .release = bpf_link_release,
2538 .read = bpf_dummy_read,
2539 .write = bpf_dummy_write,
2542 static int bpf_link_alloc_id(struct bpf_link *link)
2546 idr_preload(GFP_KERNEL);
2547 spin_lock_bh(&link_idr_lock);
2548 id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2549 spin_unlock_bh(&link_idr_lock);
2555 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2556 * reserving unused FD and allocating ID from link_idr. This is to be paired
2557 * with bpf_link_settle() to install FD and ID and expose bpf_link to
2558 * user-space, if bpf_link is successfully attached. If not, bpf_link and
2559 * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2560 * transient state is passed around in struct bpf_link_primer.
2561 * This is preferred way to create and initialize bpf_link, especially when
2562 * there are complicated and expensive operations inbetween creating bpf_link
2563 * itself and attaching it to BPF hook. By using bpf_link_prime() and
2564 * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2565 * expensive (and potentially failing) roll back operations in a rare case
2566 * that file, FD, or ID can't be allocated.
2568 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2573 fd = get_unused_fd_flags(O_CLOEXEC);
2578 id = bpf_link_alloc_id(link);
2584 file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2586 bpf_link_free_id(id);
2588 return PTR_ERR(file);
2591 primer->link = link;
2592 primer->file = file;
2598 int bpf_link_settle(struct bpf_link_primer *primer)
2600 /* make bpf_link fetchable by ID */
2601 spin_lock_bh(&link_idr_lock);
2602 primer->link->id = primer->id;
2603 spin_unlock_bh(&link_idr_lock);
2604 /* make bpf_link fetchable by FD */
2605 fd_install(primer->fd, primer->file);
2606 /* pass through installed FD */
2610 int bpf_link_new_fd(struct bpf_link *link)
2612 return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2615 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2617 struct fd f = fdget(ufd);
2618 struct bpf_link *link;
2621 return ERR_PTR(-EBADF);
2622 if (f.file->f_op != &bpf_link_fops) {
2624 return ERR_PTR(-EINVAL);
2627 link = f.file->private_data;
2634 struct bpf_tracing_link {
2635 struct bpf_link link;
2636 enum bpf_attach_type attach_type;
2637 struct bpf_trampoline *trampoline;
2638 struct bpf_prog *tgt_prog;
2641 static void bpf_tracing_link_release(struct bpf_link *link)
2643 struct bpf_tracing_link *tr_link =
2644 container_of(link, struct bpf_tracing_link, link);
2646 WARN_ON_ONCE(bpf_trampoline_unlink_prog(link->prog,
2647 tr_link->trampoline));
2649 bpf_trampoline_put(tr_link->trampoline);
2651 /* tgt_prog is NULL if target is a kernel function */
2652 if (tr_link->tgt_prog)
2653 bpf_prog_put(tr_link->tgt_prog);
2656 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2658 struct bpf_tracing_link *tr_link =
2659 container_of(link, struct bpf_tracing_link, link);
2664 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2665 struct seq_file *seq)
2667 struct bpf_tracing_link *tr_link =
2668 container_of(link, struct bpf_tracing_link, link);
2671 "attach_type:\t%d\n",
2672 tr_link->attach_type);
2675 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2676 struct bpf_link_info *info)
2678 struct bpf_tracing_link *tr_link =
2679 container_of(link, struct bpf_tracing_link, link);
2681 info->tracing.attach_type = tr_link->attach_type;
2682 bpf_trampoline_unpack_key(tr_link->trampoline->key,
2683 &info->tracing.target_obj_id,
2684 &info->tracing.target_btf_id);
2689 static const struct bpf_link_ops bpf_tracing_link_lops = {
2690 .release = bpf_tracing_link_release,
2691 .dealloc = bpf_tracing_link_dealloc,
2692 .show_fdinfo = bpf_tracing_link_show_fdinfo,
2693 .fill_link_info = bpf_tracing_link_fill_link_info,
2696 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2700 struct bpf_link_primer link_primer;
2701 struct bpf_prog *tgt_prog = NULL;
2702 struct bpf_trampoline *tr = NULL;
2703 struct bpf_tracing_link *link;
2707 switch (prog->type) {
2708 case BPF_PROG_TYPE_TRACING:
2709 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2710 prog->expected_attach_type != BPF_TRACE_FEXIT &&
2711 prog->expected_attach_type != BPF_MODIFY_RETURN) {
2716 case BPF_PROG_TYPE_EXT:
2717 if (prog->expected_attach_type != 0) {
2722 case BPF_PROG_TYPE_LSM:
2723 if (prog->expected_attach_type != BPF_LSM_MAC) {
2733 if (!!tgt_prog_fd != !!btf_id) {
2739 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
2740 if (prog->type != BPF_PROG_TYPE_EXT) {
2745 tgt_prog = bpf_prog_get(tgt_prog_fd);
2746 if (IS_ERR(tgt_prog)) {
2747 err = PTR_ERR(tgt_prog);
2752 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
2755 link = kzalloc(sizeof(*link), GFP_USER);
2760 bpf_link_init(&link->link, BPF_LINK_TYPE_TRACING,
2761 &bpf_tracing_link_lops, prog);
2762 link->attach_type = prog->expected_attach_type;
2764 mutex_lock(&prog->aux->dst_mutex);
2766 /* There are a few possible cases here:
2768 * - if prog->aux->dst_trampoline is set, the program was just loaded
2769 * and not yet attached to anything, so we can use the values stored
2772 * - if prog->aux->dst_trampoline is NULL, the program has already been
2773 * attached to a target and its initial target was cleared (below)
2775 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
2776 * target_btf_id using the link_create API.
2778 * - if tgt_prog == NULL when this function was called using the old
2779 * raw_tracepoint_open API, and we need a target from prog->aux
2781 * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
2782 * was detached and is going for re-attachment.
2784 if (!prog->aux->dst_trampoline && !tgt_prog) {
2786 * Allow re-attach for TRACING and LSM programs. If it's
2787 * currently linked, bpf_trampoline_link_prog will fail.
2788 * EXT programs need to specify tgt_prog_fd, so they
2789 * re-attach in separate code path.
2791 if (prog->type != BPF_PROG_TYPE_TRACING &&
2792 prog->type != BPF_PROG_TYPE_LSM) {
2796 btf_id = prog->aux->attach_btf_id;
2797 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
2800 if (!prog->aux->dst_trampoline ||
2801 (key && key != prog->aux->dst_trampoline->key)) {
2802 /* If there is no saved target, or the specified target is
2803 * different from the destination specified at load time, we
2804 * need a new trampoline and a check for compatibility
2806 struct bpf_attach_target_info tgt_info = {};
2808 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
2813 tr = bpf_trampoline_get(key, &tgt_info);
2819 /* The caller didn't specify a target, or the target was the
2820 * same as the destination supplied during program load. This
2821 * means we can reuse the trampoline and reference from program
2822 * load time, and there is no need to allocate a new one. This
2823 * can only happen once for any program, as the saved values in
2824 * prog->aux are cleared below.
2826 tr = prog->aux->dst_trampoline;
2827 tgt_prog = prog->aux->dst_prog;
2830 err = bpf_link_prime(&link->link, &link_primer);
2834 err = bpf_trampoline_link_prog(prog, tr);
2836 bpf_link_cleanup(&link_primer);
2841 link->tgt_prog = tgt_prog;
2842 link->trampoline = tr;
2844 /* Always clear the trampoline and target prog from prog->aux to make
2845 * sure the original attach destination is not kept alive after a
2846 * program is (re-)attached to another target.
2848 if (prog->aux->dst_prog &&
2849 (tgt_prog_fd || tr != prog->aux->dst_trampoline))
2850 /* got extra prog ref from syscall, or attaching to different prog */
2851 bpf_prog_put(prog->aux->dst_prog);
2852 if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
2853 /* we allocated a new trampoline, so free the old one */
2854 bpf_trampoline_put(prog->aux->dst_trampoline);
2856 prog->aux->dst_prog = NULL;
2857 prog->aux->dst_trampoline = NULL;
2858 mutex_unlock(&prog->aux->dst_mutex);
2860 return bpf_link_settle(&link_primer);
2862 if (tr && tr != prog->aux->dst_trampoline)
2863 bpf_trampoline_put(tr);
2864 mutex_unlock(&prog->aux->dst_mutex);
2867 if (tgt_prog_fd && tgt_prog)
2868 bpf_prog_put(tgt_prog);
2872 struct bpf_raw_tp_link {
2873 struct bpf_link link;
2874 struct bpf_raw_event_map *btp;
2877 static void bpf_raw_tp_link_release(struct bpf_link *link)
2879 struct bpf_raw_tp_link *raw_tp =
2880 container_of(link, struct bpf_raw_tp_link, link);
2882 bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
2883 bpf_put_raw_tracepoint(raw_tp->btp);
2886 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
2888 struct bpf_raw_tp_link *raw_tp =
2889 container_of(link, struct bpf_raw_tp_link, link);
2894 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
2895 struct seq_file *seq)
2897 struct bpf_raw_tp_link *raw_tp_link =
2898 container_of(link, struct bpf_raw_tp_link, link);
2902 raw_tp_link->btp->tp->name);
2905 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
2906 struct bpf_link_info *info)
2908 struct bpf_raw_tp_link *raw_tp_link =
2909 container_of(link, struct bpf_raw_tp_link, link);
2910 char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
2911 const char *tp_name = raw_tp_link->btp->tp->name;
2912 u32 ulen = info->raw_tracepoint.tp_name_len;
2913 size_t tp_len = strlen(tp_name);
2918 info->raw_tracepoint.tp_name_len = tp_len + 1;
2923 if (ulen >= tp_len + 1) {
2924 if (copy_to_user(ubuf, tp_name, tp_len + 1))
2929 if (copy_to_user(ubuf, tp_name, ulen - 1))
2931 if (put_user(zero, ubuf + ulen - 1))
2939 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
2940 .release = bpf_raw_tp_link_release,
2941 .dealloc = bpf_raw_tp_link_dealloc,
2942 .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
2943 .fill_link_info = bpf_raw_tp_link_fill_link_info,
2946 #ifdef CONFIG_PERF_EVENTS
2947 struct bpf_perf_link {
2948 struct bpf_link link;
2949 struct file *perf_file;
2952 static void bpf_perf_link_release(struct bpf_link *link)
2954 struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
2955 struct perf_event *event = perf_link->perf_file->private_data;
2957 perf_event_free_bpf_prog(event);
2958 fput(perf_link->perf_file);
2961 static void bpf_perf_link_dealloc(struct bpf_link *link)
2963 struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
2968 static const struct bpf_link_ops bpf_perf_link_lops = {
2969 .release = bpf_perf_link_release,
2970 .dealloc = bpf_perf_link_dealloc,
2973 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2975 struct bpf_link_primer link_primer;
2976 struct bpf_perf_link *link;
2977 struct perf_event *event;
2978 struct file *perf_file;
2981 if (attr->link_create.flags)
2984 perf_file = perf_event_get(attr->link_create.target_fd);
2985 if (IS_ERR(perf_file))
2986 return PTR_ERR(perf_file);
2988 link = kzalloc(sizeof(*link), GFP_USER);
2993 bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
2994 link->perf_file = perf_file;
2996 err = bpf_link_prime(&link->link, &link_primer);
3002 event = perf_file->private_data;
3003 err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3005 bpf_link_cleanup(&link_primer);
3008 /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3011 return bpf_link_settle(&link_primer);
3017 #endif /* CONFIG_PERF_EVENTS */
3019 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3021 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3023 struct bpf_link_primer link_primer;
3024 struct bpf_raw_tp_link *link;
3025 struct bpf_raw_event_map *btp;
3026 struct bpf_prog *prog;
3027 const char *tp_name;
3031 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3034 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3036 return PTR_ERR(prog);
3038 switch (prog->type) {
3039 case BPF_PROG_TYPE_TRACING:
3040 case BPF_PROG_TYPE_EXT:
3041 case BPF_PROG_TYPE_LSM:
3042 if (attr->raw_tracepoint.name) {
3043 /* The attach point for this category of programs
3044 * should be specified via btf_id during program load.
3049 if (prog->type == BPF_PROG_TYPE_TRACING &&
3050 prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3051 tp_name = prog->aux->attach_func_name;
3054 err = bpf_tracing_prog_attach(prog, 0, 0);
3058 case BPF_PROG_TYPE_RAW_TRACEPOINT:
3059 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3060 if (strncpy_from_user(buf,
3061 u64_to_user_ptr(attr->raw_tracepoint.name),
3062 sizeof(buf) - 1) < 0) {
3066 buf[sizeof(buf) - 1] = 0;
3074 btp = bpf_get_raw_tracepoint(tp_name);
3080 link = kzalloc(sizeof(*link), GFP_USER);
3085 bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3086 &bpf_raw_tp_link_lops, prog);
3089 err = bpf_link_prime(&link->link, &link_primer);
3095 err = bpf_probe_register(link->btp, prog);
3097 bpf_link_cleanup(&link_primer);
3101 return bpf_link_settle(&link_primer);
3104 bpf_put_raw_tracepoint(btp);
3110 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3111 enum bpf_attach_type attach_type)
3113 switch (prog->type) {
3114 case BPF_PROG_TYPE_CGROUP_SOCK:
3115 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3116 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3117 case BPF_PROG_TYPE_SK_LOOKUP:
3118 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3119 case BPF_PROG_TYPE_CGROUP_SKB:
3120 if (!capable(CAP_NET_ADMIN))
3121 /* cg-skb progs can be loaded by unpriv user.
3122 * check permissions at attach time.
3125 return prog->enforce_expected_attach_type &&
3126 prog->expected_attach_type != attach_type ?
3133 static enum bpf_prog_type
3134 attach_type_to_prog_type(enum bpf_attach_type attach_type)
3136 switch (attach_type) {
3137 case BPF_CGROUP_INET_INGRESS:
3138 case BPF_CGROUP_INET_EGRESS:
3139 return BPF_PROG_TYPE_CGROUP_SKB;
3140 case BPF_CGROUP_INET_SOCK_CREATE:
3141 case BPF_CGROUP_INET_SOCK_RELEASE:
3142 case BPF_CGROUP_INET4_POST_BIND:
3143 case BPF_CGROUP_INET6_POST_BIND:
3144 return BPF_PROG_TYPE_CGROUP_SOCK;
3145 case BPF_CGROUP_INET4_BIND:
3146 case BPF_CGROUP_INET6_BIND:
3147 case BPF_CGROUP_INET4_CONNECT:
3148 case BPF_CGROUP_INET6_CONNECT:
3149 case BPF_CGROUP_INET4_GETPEERNAME:
3150 case BPF_CGROUP_INET6_GETPEERNAME:
3151 case BPF_CGROUP_INET4_GETSOCKNAME:
3152 case BPF_CGROUP_INET6_GETSOCKNAME:
3153 case BPF_CGROUP_UDP4_SENDMSG:
3154 case BPF_CGROUP_UDP6_SENDMSG:
3155 case BPF_CGROUP_UDP4_RECVMSG:
3156 case BPF_CGROUP_UDP6_RECVMSG:
3157 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3158 case BPF_CGROUP_SOCK_OPS:
3159 return BPF_PROG_TYPE_SOCK_OPS;
3160 case BPF_CGROUP_DEVICE:
3161 return BPF_PROG_TYPE_CGROUP_DEVICE;
3162 case BPF_SK_MSG_VERDICT:
3163 return BPF_PROG_TYPE_SK_MSG;
3164 case BPF_SK_SKB_STREAM_PARSER:
3165 case BPF_SK_SKB_STREAM_VERDICT:
3166 case BPF_SK_SKB_VERDICT:
3167 return BPF_PROG_TYPE_SK_SKB;
3168 case BPF_LIRC_MODE2:
3169 return BPF_PROG_TYPE_LIRC_MODE2;
3170 case BPF_FLOW_DISSECTOR:
3171 return BPF_PROG_TYPE_FLOW_DISSECTOR;
3172 case BPF_CGROUP_SYSCTL:
3173 return BPF_PROG_TYPE_CGROUP_SYSCTL;
3174 case BPF_CGROUP_GETSOCKOPT:
3175 case BPF_CGROUP_SETSOCKOPT:
3176 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3177 case BPF_TRACE_ITER:
3178 return BPF_PROG_TYPE_TRACING;
3180 return BPF_PROG_TYPE_SK_LOOKUP;
3182 return BPF_PROG_TYPE_XDP;
3184 return BPF_PROG_TYPE_UNSPEC;
3188 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
3190 #define BPF_F_ATTACH_MASK \
3191 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
3193 static int bpf_prog_attach(const union bpf_attr *attr)
3195 enum bpf_prog_type ptype;
3196 struct bpf_prog *prog;
3199 if (CHECK_ATTR(BPF_PROG_ATTACH))
3202 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
3205 ptype = attach_type_to_prog_type(attr->attach_type);
3206 if (ptype == BPF_PROG_TYPE_UNSPEC)
3209 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3211 return PTR_ERR(prog);
3213 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3219 case BPF_PROG_TYPE_SK_SKB:
3220 case BPF_PROG_TYPE_SK_MSG:
3221 ret = sock_map_get_from_fd(attr, prog);
3223 case BPF_PROG_TYPE_LIRC_MODE2:
3224 ret = lirc_prog_attach(attr, prog);
3226 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3227 ret = netns_bpf_prog_attach(attr, prog);
3229 case BPF_PROG_TYPE_CGROUP_DEVICE:
3230 case BPF_PROG_TYPE_CGROUP_SKB:
3231 case BPF_PROG_TYPE_CGROUP_SOCK:
3232 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3233 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3234 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3235 case BPF_PROG_TYPE_SOCK_OPS:
3236 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3247 #define BPF_PROG_DETACH_LAST_FIELD attach_type
3249 static int bpf_prog_detach(const union bpf_attr *attr)
3251 enum bpf_prog_type ptype;
3253 if (CHECK_ATTR(BPF_PROG_DETACH))
3256 ptype = attach_type_to_prog_type(attr->attach_type);
3259 case BPF_PROG_TYPE_SK_MSG:
3260 case BPF_PROG_TYPE_SK_SKB:
3261 return sock_map_prog_detach(attr, ptype);
3262 case BPF_PROG_TYPE_LIRC_MODE2:
3263 return lirc_prog_detach(attr);
3264 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3265 return netns_bpf_prog_detach(attr, ptype);
3266 case BPF_PROG_TYPE_CGROUP_DEVICE:
3267 case BPF_PROG_TYPE_CGROUP_SKB:
3268 case BPF_PROG_TYPE_CGROUP_SOCK:
3269 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3270 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3271 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3272 case BPF_PROG_TYPE_SOCK_OPS:
3273 return cgroup_bpf_prog_detach(attr, ptype);
3279 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
3281 static int bpf_prog_query(const union bpf_attr *attr,
3282 union bpf_attr __user *uattr)
3284 if (!capable(CAP_NET_ADMIN))
3286 if (CHECK_ATTR(BPF_PROG_QUERY))
3288 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3291 switch (attr->query.attach_type) {
3292 case BPF_CGROUP_INET_INGRESS:
3293 case BPF_CGROUP_INET_EGRESS:
3294 case BPF_CGROUP_INET_SOCK_CREATE:
3295 case BPF_CGROUP_INET_SOCK_RELEASE:
3296 case BPF_CGROUP_INET4_BIND:
3297 case BPF_CGROUP_INET6_BIND:
3298 case BPF_CGROUP_INET4_POST_BIND:
3299 case BPF_CGROUP_INET6_POST_BIND:
3300 case BPF_CGROUP_INET4_CONNECT:
3301 case BPF_CGROUP_INET6_CONNECT:
3302 case BPF_CGROUP_INET4_GETPEERNAME:
3303 case BPF_CGROUP_INET6_GETPEERNAME:
3304 case BPF_CGROUP_INET4_GETSOCKNAME:
3305 case BPF_CGROUP_INET6_GETSOCKNAME:
3306 case BPF_CGROUP_UDP4_SENDMSG:
3307 case BPF_CGROUP_UDP6_SENDMSG:
3308 case BPF_CGROUP_UDP4_RECVMSG:
3309 case BPF_CGROUP_UDP6_RECVMSG:
3310 case BPF_CGROUP_SOCK_OPS:
3311 case BPF_CGROUP_DEVICE:
3312 case BPF_CGROUP_SYSCTL:
3313 case BPF_CGROUP_GETSOCKOPT:
3314 case BPF_CGROUP_SETSOCKOPT:
3315 return cgroup_bpf_prog_query(attr, uattr);
3316 case BPF_LIRC_MODE2:
3317 return lirc_prog_query(attr, uattr);
3318 case BPF_FLOW_DISSECTOR:
3320 return netns_bpf_prog_query(attr, uattr);
3326 #define BPF_PROG_TEST_RUN_LAST_FIELD test.cpu
3328 static int bpf_prog_test_run(const union bpf_attr *attr,
3329 union bpf_attr __user *uattr)
3331 struct bpf_prog *prog;
3332 int ret = -ENOTSUPP;
3334 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3337 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3338 (!attr->test.ctx_size_in && attr->test.ctx_in))
3341 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3342 (!attr->test.ctx_size_out && attr->test.ctx_out))
3345 prog = bpf_prog_get(attr->test.prog_fd);
3347 return PTR_ERR(prog);
3349 if (prog->aux->ops->test_run)
3350 ret = prog->aux->ops->test_run(prog, attr, uattr);
3356 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3358 static int bpf_obj_get_next_id(const union bpf_attr *attr,
3359 union bpf_attr __user *uattr,
3363 u32 next_id = attr->start_id;
3366 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3369 if (!capable(CAP_SYS_ADMIN))
3374 if (!idr_get_next(idr, &next_id))
3376 spin_unlock_bh(lock);
3379 err = put_user(next_id, &uattr->next_id);
3384 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3386 struct bpf_map *map;
3388 spin_lock_bh(&map_idr_lock);
3390 map = idr_get_next(&map_idr, id);
3392 map = __bpf_map_inc_not_zero(map, false);
3398 spin_unlock_bh(&map_idr_lock);
3403 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3405 struct bpf_prog *prog;
3407 spin_lock_bh(&prog_idr_lock);
3409 prog = idr_get_next(&prog_idr, id);
3411 prog = bpf_prog_inc_not_zero(prog);
3417 spin_unlock_bh(&prog_idr_lock);
3422 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3424 struct bpf_prog *bpf_prog_by_id(u32 id)
3426 struct bpf_prog *prog;
3429 return ERR_PTR(-ENOENT);
3431 spin_lock_bh(&prog_idr_lock);
3432 prog = idr_find(&prog_idr, id);
3434 prog = bpf_prog_inc_not_zero(prog);
3436 prog = ERR_PTR(-ENOENT);
3437 spin_unlock_bh(&prog_idr_lock);
3441 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3443 struct bpf_prog *prog;
3444 u32 id = attr->prog_id;
3447 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3450 if (!capable(CAP_SYS_ADMIN))
3453 prog = bpf_prog_by_id(id);
3455 return PTR_ERR(prog);
3457 fd = bpf_prog_new_fd(prog);
3464 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3466 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3468 struct bpf_map *map;
3469 u32 id = attr->map_id;
3473 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3474 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3477 if (!capable(CAP_SYS_ADMIN))
3480 f_flags = bpf_get_file_flag(attr->open_flags);
3484 spin_lock_bh(&map_idr_lock);
3485 map = idr_find(&map_idr, id);
3487 map = __bpf_map_inc_not_zero(map, true);
3489 map = ERR_PTR(-ENOENT);
3490 spin_unlock_bh(&map_idr_lock);
3493 return PTR_ERR(map);
3495 fd = bpf_map_new_fd(map, f_flags);
3497 bpf_map_put_with_uref(map);
3502 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3503 unsigned long addr, u32 *off,
3506 const struct bpf_map *map;
3509 mutex_lock(&prog->aux->used_maps_mutex);
3510 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3511 map = prog->aux->used_maps[i];
3512 if (map == (void *)addr) {
3513 *type = BPF_PSEUDO_MAP_FD;
3516 if (!map->ops->map_direct_value_meta)
3518 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3519 *type = BPF_PSEUDO_MAP_VALUE;
3526 mutex_unlock(&prog->aux->used_maps_mutex);
3530 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3531 const struct cred *f_cred)
3533 const struct bpf_map *map;
3534 struct bpf_insn *insns;
3540 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3545 for (i = 0; i < prog->len; i++) {
3546 code = insns[i].code;
3548 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3549 insns[i].code = BPF_JMP | BPF_CALL;
3550 insns[i].imm = BPF_FUNC_tail_call;
3553 if (code == (BPF_JMP | BPF_CALL) ||
3554 code == (BPF_JMP | BPF_CALL_ARGS)) {
3555 if (code == (BPF_JMP | BPF_CALL_ARGS))
3556 insns[i].code = BPF_JMP | BPF_CALL;
3557 if (!bpf_dump_raw_ok(f_cred))
3561 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3562 insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3566 if (code != (BPF_LD | BPF_IMM | BPF_DW))
3569 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3570 map = bpf_map_from_imm(prog, imm, &off, &type);
3572 insns[i].src_reg = type;
3573 insns[i].imm = map->id;
3574 insns[i + 1].imm = off;
3582 static int set_info_rec_size(struct bpf_prog_info *info)
3585 * Ensure info.*_rec_size is the same as kernel expected size
3589 * Only allow zero *_rec_size if both _rec_size and _cnt are
3590 * zero. In this case, the kernel will set the expected
3591 * _rec_size back to the info.
3594 if ((info->nr_func_info || info->func_info_rec_size) &&
3595 info->func_info_rec_size != sizeof(struct bpf_func_info))
3598 if ((info->nr_line_info || info->line_info_rec_size) &&
3599 info->line_info_rec_size != sizeof(struct bpf_line_info))
3602 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3603 info->jited_line_info_rec_size != sizeof(__u64))
3606 info->func_info_rec_size = sizeof(struct bpf_func_info);
3607 info->line_info_rec_size = sizeof(struct bpf_line_info);
3608 info->jited_line_info_rec_size = sizeof(__u64);
3613 static int bpf_prog_get_info_by_fd(struct file *file,
3614 struct bpf_prog *prog,
3615 const union bpf_attr *attr,
3616 union bpf_attr __user *uattr)
3618 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3619 struct bpf_prog_info info;
3620 u32 info_len = attr->info.info_len;
3621 struct bpf_prog_kstats stats;
3622 char __user *uinsns;
3626 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
3629 info_len = min_t(u32, sizeof(info), info_len);
3631 memset(&info, 0, sizeof(info));
3632 if (copy_from_user(&info, uinfo, info_len))
3635 info.type = prog->type;
3636 info.id = prog->aux->id;
3637 info.load_time = prog->aux->load_time;
3638 info.created_by_uid = from_kuid_munged(current_user_ns(),
3639 prog->aux->user->uid);
3640 info.gpl_compatible = prog->gpl_compatible;
3642 memcpy(info.tag, prog->tag, sizeof(prog->tag));
3643 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3645 mutex_lock(&prog->aux->used_maps_mutex);
3646 ulen = info.nr_map_ids;
3647 info.nr_map_ids = prog->aux->used_map_cnt;
3648 ulen = min_t(u32, info.nr_map_ids, ulen);
3650 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3653 for (i = 0; i < ulen; i++)
3654 if (put_user(prog->aux->used_maps[i]->id,
3655 &user_map_ids[i])) {
3656 mutex_unlock(&prog->aux->used_maps_mutex);
3660 mutex_unlock(&prog->aux->used_maps_mutex);
3662 err = set_info_rec_size(&info);
3666 bpf_prog_get_stats(prog, &stats);
3667 info.run_time_ns = stats.nsecs;
3668 info.run_cnt = stats.cnt;
3669 info.recursion_misses = stats.misses;
3671 info.verified_insns = prog->aux->verified_insns;
3673 if (!bpf_capable()) {
3674 info.jited_prog_len = 0;
3675 info.xlated_prog_len = 0;
3676 info.nr_jited_ksyms = 0;
3677 info.nr_jited_func_lens = 0;
3678 info.nr_func_info = 0;
3679 info.nr_line_info = 0;
3680 info.nr_jited_line_info = 0;
3684 ulen = info.xlated_prog_len;
3685 info.xlated_prog_len = bpf_prog_insn_size(prog);
3686 if (info.xlated_prog_len && ulen) {
3687 struct bpf_insn *insns_sanitized;
3690 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
3691 info.xlated_prog_insns = 0;
3694 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
3695 if (!insns_sanitized)
3697 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3698 ulen = min_t(u32, info.xlated_prog_len, ulen);
3699 fault = copy_to_user(uinsns, insns_sanitized, ulen);
3700 kfree(insns_sanitized);
3705 if (bpf_prog_is_dev_bound(prog->aux)) {
3706 err = bpf_prog_offload_info_fill(&info, prog);
3712 /* NOTE: the following code is supposed to be skipped for offload.
3713 * bpf_prog_offload_info_fill() is the place to fill similar fields
3716 ulen = info.jited_prog_len;
3717 if (prog->aux->func_cnt) {
3720 info.jited_prog_len = 0;
3721 for (i = 0; i < prog->aux->func_cnt; i++)
3722 info.jited_prog_len += prog->aux->func[i]->jited_len;
3724 info.jited_prog_len = prog->jited_len;
3727 if (info.jited_prog_len && ulen) {
3728 if (bpf_dump_raw_ok(file->f_cred)) {
3729 uinsns = u64_to_user_ptr(info.jited_prog_insns);
3730 ulen = min_t(u32, info.jited_prog_len, ulen);
3732 /* for multi-function programs, copy the JITed
3733 * instructions for all the functions
3735 if (prog->aux->func_cnt) {
3740 for (i = 0; i < prog->aux->func_cnt; i++) {
3741 len = prog->aux->func[i]->jited_len;
3742 len = min_t(u32, len, free);
3743 img = (u8 *) prog->aux->func[i]->bpf_func;
3744 if (copy_to_user(uinsns, img, len))
3752 if (copy_to_user(uinsns, prog->bpf_func, ulen))
3756 info.jited_prog_insns = 0;
3760 ulen = info.nr_jited_ksyms;
3761 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
3763 if (bpf_dump_raw_ok(file->f_cred)) {
3764 unsigned long ksym_addr;
3765 u64 __user *user_ksyms;
3768 /* copy the address of the kernel symbol
3769 * corresponding to each function
3771 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
3772 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
3773 if (prog->aux->func_cnt) {
3774 for (i = 0; i < ulen; i++) {
3775 ksym_addr = (unsigned long)
3776 prog->aux->func[i]->bpf_func;
3777 if (put_user((u64) ksym_addr,
3782 ksym_addr = (unsigned long) prog->bpf_func;
3783 if (put_user((u64) ksym_addr, &user_ksyms[0]))
3787 info.jited_ksyms = 0;
3791 ulen = info.nr_jited_func_lens;
3792 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
3794 if (bpf_dump_raw_ok(file->f_cred)) {
3795 u32 __user *user_lens;
3798 /* copy the JITed image lengths for each function */
3799 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
3800 user_lens = u64_to_user_ptr(info.jited_func_lens);
3801 if (prog->aux->func_cnt) {
3802 for (i = 0; i < ulen; i++) {
3804 prog->aux->func[i]->jited_len;
3805 if (put_user(func_len, &user_lens[i]))
3809 func_len = prog->jited_len;
3810 if (put_user(func_len, &user_lens[0]))
3814 info.jited_func_lens = 0;
3819 info.btf_id = btf_obj_id(prog->aux->btf);
3821 ulen = info.nr_func_info;
3822 info.nr_func_info = prog->aux->func_info_cnt;
3823 if (info.nr_func_info && ulen) {
3824 char __user *user_finfo;
3826 user_finfo = u64_to_user_ptr(info.func_info);
3827 ulen = min_t(u32, info.nr_func_info, ulen);
3828 if (copy_to_user(user_finfo, prog->aux->func_info,
3829 info.func_info_rec_size * ulen))
3833 ulen = info.nr_line_info;
3834 info.nr_line_info = prog->aux->nr_linfo;
3835 if (info.nr_line_info && ulen) {
3836 __u8 __user *user_linfo;
3838 user_linfo = u64_to_user_ptr(info.line_info);
3839 ulen = min_t(u32, info.nr_line_info, ulen);
3840 if (copy_to_user(user_linfo, prog->aux->linfo,
3841 info.line_info_rec_size * ulen))
3845 ulen = info.nr_jited_line_info;
3846 if (prog->aux->jited_linfo)
3847 info.nr_jited_line_info = prog->aux->nr_linfo;
3849 info.nr_jited_line_info = 0;
3850 if (info.nr_jited_line_info && ulen) {
3851 if (bpf_dump_raw_ok(file->f_cred)) {
3852 __u64 __user *user_linfo;
3855 user_linfo = u64_to_user_ptr(info.jited_line_info);
3856 ulen = min_t(u32, info.nr_jited_line_info, ulen);
3857 for (i = 0; i < ulen; i++) {
3858 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
3863 info.jited_line_info = 0;
3867 ulen = info.nr_prog_tags;
3868 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
3870 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
3873 user_prog_tags = u64_to_user_ptr(info.prog_tags);
3874 ulen = min_t(u32, info.nr_prog_tags, ulen);
3875 if (prog->aux->func_cnt) {
3876 for (i = 0; i < ulen; i++) {
3877 if (copy_to_user(user_prog_tags[i],
3878 prog->aux->func[i]->tag,
3883 if (copy_to_user(user_prog_tags[0],
3884 prog->tag, BPF_TAG_SIZE))
3890 if (copy_to_user(uinfo, &info, info_len) ||
3891 put_user(info_len, &uattr->info.info_len))
3897 static int bpf_map_get_info_by_fd(struct file *file,
3898 struct bpf_map *map,
3899 const union bpf_attr *attr,
3900 union bpf_attr __user *uattr)
3902 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3903 struct bpf_map_info info;
3904 u32 info_len = attr->info.info_len;
3907 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
3910 info_len = min_t(u32, sizeof(info), info_len);
3912 memset(&info, 0, sizeof(info));
3913 info.type = map->map_type;
3915 info.key_size = map->key_size;
3916 info.value_size = map->value_size;
3917 info.max_entries = map->max_entries;
3918 info.map_flags = map->map_flags;
3919 info.map_extra = map->map_extra;
3920 memcpy(info.name, map->name, sizeof(map->name));
3923 info.btf_id = btf_obj_id(map->btf);
3924 info.btf_key_type_id = map->btf_key_type_id;
3925 info.btf_value_type_id = map->btf_value_type_id;
3927 info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
3929 if (bpf_map_is_dev_bound(map)) {
3930 err = bpf_map_offload_info_fill(&info, map);
3935 if (copy_to_user(uinfo, &info, info_len) ||
3936 put_user(info_len, &uattr->info.info_len))
3942 static int bpf_btf_get_info_by_fd(struct file *file,
3944 const union bpf_attr *attr,
3945 union bpf_attr __user *uattr)
3947 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3948 u32 info_len = attr->info.info_len;
3951 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
3955 return btf_get_info_by_fd(btf, attr, uattr);
3958 static int bpf_link_get_info_by_fd(struct file *file,
3959 struct bpf_link *link,
3960 const union bpf_attr *attr,
3961 union bpf_attr __user *uattr)
3963 struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3964 struct bpf_link_info info;
3965 u32 info_len = attr->info.info_len;
3968 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
3971 info_len = min_t(u32, sizeof(info), info_len);
3973 memset(&info, 0, sizeof(info));
3974 if (copy_from_user(&info, uinfo, info_len))
3977 info.type = link->type;
3979 info.prog_id = link->prog->aux->id;
3981 if (link->ops->fill_link_info) {
3982 err = link->ops->fill_link_info(link, &info);
3987 if (copy_to_user(uinfo, &info, info_len) ||
3988 put_user(info_len, &uattr->info.info_len))
3995 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
3997 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
3998 union bpf_attr __user *uattr)
4000 int ufd = attr->info.bpf_fd;
4004 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4011 if (f.file->f_op == &bpf_prog_fops)
4012 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
4014 else if (f.file->f_op == &bpf_map_fops)
4015 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
4017 else if (f.file->f_op == &btf_fops)
4018 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
4019 else if (f.file->f_op == &bpf_link_fops)
4020 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
4029 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
4031 static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr)
4033 if (CHECK_ATTR(BPF_BTF_LOAD))
4039 return btf_new_fd(attr, uattr);
4042 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4044 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4046 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4049 if (!capable(CAP_SYS_ADMIN))
4052 return btf_get_fd_by_id(attr->btf_id);
4055 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
4056 union bpf_attr __user *uattr,
4057 u32 prog_id, u32 fd_type,
4058 const char *buf, u64 probe_offset,
4061 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
4062 u32 len = buf ? strlen(buf) : 0, input_len;
4065 if (put_user(len, &uattr->task_fd_query.buf_len))
4067 input_len = attr->task_fd_query.buf_len;
4068 if (input_len && ubuf) {
4070 /* nothing to copy, just make ubuf NULL terminated */
4073 if (put_user(zero, ubuf))
4075 } else if (input_len >= len + 1) {
4076 /* ubuf can hold the string with NULL terminator */
4077 if (copy_to_user(ubuf, buf, len + 1))
4080 /* ubuf cannot hold the string with NULL terminator,
4081 * do a partial copy with NULL terminator.
4086 if (copy_to_user(ubuf, buf, input_len - 1))
4088 if (put_user(zero, ubuf + input_len - 1))
4093 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
4094 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
4095 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
4096 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
4102 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
4104 static int bpf_task_fd_query(const union bpf_attr *attr,
4105 union bpf_attr __user *uattr)
4107 pid_t pid = attr->task_fd_query.pid;
4108 u32 fd = attr->task_fd_query.fd;
4109 const struct perf_event *event;
4110 struct task_struct *task;
4114 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
4117 if (!capable(CAP_SYS_ADMIN))
4120 if (attr->task_fd_query.flags != 0)
4123 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
4128 file = fget_task(task, fd);
4129 put_task_struct(task);
4133 if (file->f_op == &bpf_link_fops) {
4134 struct bpf_link *link = file->private_data;
4136 if (link->ops == &bpf_raw_tp_link_lops) {
4137 struct bpf_raw_tp_link *raw_tp =
4138 container_of(link, struct bpf_raw_tp_link, link);
4139 struct bpf_raw_event_map *btp = raw_tp->btp;
4141 err = bpf_task_fd_query_copy(attr, uattr,
4142 raw_tp->link.prog->aux->id,
4143 BPF_FD_TYPE_RAW_TRACEPOINT,
4144 btp->tp->name, 0, 0);
4150 event = perf_get_event(file);
4151 if (!IS_ERR(event)) {
4152 u64 probe_offset, probe_addr;
4153 u32 prog_id, fd_type;
4156 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
4157 &buf, &probe_offset,
4160 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
4174 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
4176 #define BPF_DO_BATCH(fn) \
4182 err = fn(map, attr, uattr); \
4185 static int bpf_map_do_batch(const union bpf_attr *attr,
4186 union bpf_attr __user *uattr,
4189 bool has_read = cmd == BPF_MAP_LOOKUP_BATCH ||
4190 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
4191 bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
4192 struct bpf_map *map;
4196 if (CHECK_ATTR(BPF_MAP_BATCH))
4199 ufd = attr->batch.map_fd;
4201 map = __bpf_map_get(f);
4203 return PTR_ERR(map);
4205 bpf_map_write_active_inc(map);
4206 if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4210 if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4215 if (cmd == BPF_MAP_LOOKUP_BATCH)
4216 BPF_DO_BATCH(map->ops->map_lookup_batch);
4217 else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4218 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
4219 else if (cmd == BPF_MAP_UPDATE_BATCH)
4220 BPF_DO_BATCH(map->ops->map_update_batch);
4222 BPF_DO_BATCH(map->ops->map_delete_batch);
4225 bpf_map_write_active_dec(map);
4230 static int tracing_bpf_link_attach(const union bpf_attr *attr, bpfptr_t uattr,
4231 struct bpf_prog *prog)
4233 if (attr->link_create.attach_type != prog->expected_attach_type)
4236 if (prog->expected_attach_type == BPF_TRACE_ITER)
4237 return bpf_iter_link_attach(attr, uattr, prog);
4238 else if (prog->type == BPF_PROG_TYPE_EXT)
4239 return bpf_tracing_prog_attach(prog,
4240 attr->link_create.target_fd,
4241 attr->link_create.target_btf_id);
4245 #define BPF_LINK_CREATE_LAST_FIELD link_create.iter_info_len
4246 static int link_create(union bpf_attr *attr, bpfptr_t uattr)
4248 enum bpf_prog_type ptype;
4249 struct bpf_prog *prog;
4252 if (CHECK_ATTR(BPF_LINK_CREATE))
4255 prog = bpf_prog_get(attr->link_create.prog_fd);
4257 return PTR_ERR(prog);
4259 ret = bpf_prog_attach_check_attach_type(prog,
4260 attr->link_create.attach_type);
4264 switch (prog->type) {
4265 case BPF_PROG_TYPE_EXT:
4266 ret = tracing_bpf_link_attach(attr, uattr, prog);
4268 case BPF_PROG_TYPE_PERF_EVENT:
4269 case BPF_PROG_TYPE_KPROBE:
4270 case BPF_PROG_TYPE_TRACEPOINT:
4271 if (attr->link_create.attach_type != BPF_PERF_EVENT) {
4278 ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4279 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4287 case BPF_PROG_TYPE_CGROUP_SKB:
4288 case BPF_PROG_TYPE_CGROUP_SOCK:
4289 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4290 case BPF_PROG_TYPE_SOCK_OPS:
4291 case BPF_PROG_TYPE_CGROUP_DEVICE:
4292 case BPF_PROG_TYPE_CGROUP_SYSCTL:
4293 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4294 ret = cgroup_bpf_link_attach(attr, prog);
4296 case BPF_PROG_TYPE_TRACING:
4297 ret = tracing_bpf_link_attach(attr, uattr, prog);
4299 case BPF_PROG_TYPE_FLOW_DISSECTOR:
4300 case BPF_PROG_TYPE_SK_LOOKUP:
4301 ret = netns_bpf_link_create(attr, prog);
4304 case BPF_PROG_TYPE_XDP:
4305 ret = bpf_xdp_link_attach(attr, prog);
4308 #ifdef CONFIG_PERF_EVENTS
4309 case BPF_PROG_TYPE_PERF_EVENT:
4310 case BPF_PROG_TYPE_TRACEPOINT:
4311 case BPF_PROG_TYPE_KPROBE:
4312 ret = bpf_perf_link_attach(attr, prog);
4325 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4327 static int link_update(union bpf_attr *attr)
4329 struct bpf_prog *old_prog = NULL, *new_prog;
4330 struct bpf_link *link;
4334 if (CHECK_ATTR(BPF_LINK_UPDATE))
4337 flags = attr->link_update.flags;
4338 if (flags & ~BPF_F_REPLACE)
4341 link = bpf_link_get_from_fd(attr->link_update.link_fd);
4343 return PTR_ERR(link);
4345 new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4346 if (IS_ERR(new_prog)) {
4347 ret = PTR_ERR(new_prog);
4351 if (flags & BPF_F_REPLACE) {
4352 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4353 if (IS_ERR(old_prog)) {
4354 ret = PTR_ERR(old_prog);
4358 } else if (attr->link_update.old_prog_fd) {
4363 if (link->ops->update_prog)
4364 ret = link->ops->update_prog(link, new_prog, old_prog);
4370 bpf_prog_put(old_prog);
4372 bpf_prog_put(new_prog);
4378 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4380 static int link_detach(union bpf_attr *attr)
4382 struct bpf_link *link;
4385 if (CHECK_ATTR(BPF_LINK_DETACH))
4388 link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4390 return PTR_ERR(link);
4392 if (link->ops->detach)
4393 ret = link->ops->detach(link);
4401 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4403 return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4406 struct bpf_link *bpf_link_by_id(u32 id)
4408 struct bpf_link *link;
4411 return ERR_PTR(-ENOENT);
4413 spin_lock_bh(&link_idr_lock);
4414 /* before link is "settled", ID is 0, pretend it doesn't exist yet */
4415 link = idr_find(&link_idr, id);
4418 link = bpf_link_inc_not_zero(link);
4420 link = ERR_PTR(-EAGAIN);
4422 link = ERR_PTR(-ENOENT);
4424 spin_unlock_bh(&link_idr_lock);
4428 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4430 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4432 struct bpf_link *link;
4433 u32 id = attr->link_id;
4436 if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4439 if (!capable(CAP_SYS_ADMIN))
4442 link = bpf_link_by_id(id);
4444 return PTR_ERR(link);
4446 fd = bpf_link_new_fd(link);
4453 DEFINE_MUTEX(bpf_stats_enabled_mutex);
4455 static int bpf_stats_release(struct inode *inode, struct file *file)
4457 mutex_lock(&bpf_stats_enabled_mutex);
4458 static_key_slow_dec(&bpf_stats_enabled_key.key);
4459 mutex_unlock(&bpf_stats_enabled_mutex);
4463 static const struct file_operations bpf_stats_fops = {
4464 .release = bpf_stats_release,
4467 static int bpf_enable_runtime_stats(void)
4471 mutex_lock(&bpf_stats_enabled_mutex);
4473 /* Set a very high limit to avoid overflow */
4474 if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4475 mutex_unlock(&bpf_stats_enabled_mutex);
4479 fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4481 static_key_slow_inc(&bpf_stats_enabled_key.key);
4483 mutex_unlock(&bpf_stats_enabled_mutex);
4487 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4489 static int bpf_enable_stats(union bpf_attr *attr)
4492 if (CHECK_ATTR(BPF_ENABLE_STATS))
4495 if (!capable(CAP_SYS_ADMIN))
4498 switch (attr->enable_stats.type) {
4499 case BPF_STATS_RUN_TIME:
4500 return bpf_enable_runtime_stats();
4507 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4509 static int bpf_iter_create(union bpf_attr *attr)
4511 struct bpf_link *link;
4514 if (CHECK_ATTR(BPF_ITER_CREATE))
4517 if (attr->iter_create.flags)
4520 link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4522 return PTR_ERR(link);
4524 err = bpf_iter_new_fd(link);
4530 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4532 static int bpf_prog_bind_map(union bpf_attr *attr)
4534 struct bpf_prog *prog;
4535 struct bpf_map *map;
4536 struct bpf_map **used_maps_old, **used_maps_new;
4539 if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4542 if (attr->prog_bind_map.flags)
4545 prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4547 return PTR_ERR(prog);
4549 map = bpf_map_get(attr->prog_bind_map.map_fd);
4555 mutex_lock(&prog->aux->used_maps_mutex);
4557 used_maps_old = prog->aux->used_maps;
4559 for (i = 0; i < prog->aux->used_map_cnt; i++)
4560 if (used_maps_old[i] == map) {
4565 used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4566 sizeof(used_maps_new[0]),
4568 if (!used_maps_new) {
4573 memcpy(used_maps_new, used_maps_old,
4574 sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4575 used_maps_new[prog->aux->used_map_cnt] = map;
4577 prog->aux->used_map_cnt++;
4578 prog->aux->used_maps = used_maps_new;
4580 kfree(used_maps_old);
4583 mutex_unlock(&prog->aux->used_maps_mutex);
4592 static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
4594 union bpf_attr attr;
4597 if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
4600 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
4603 size = min_t(u32, size, sizeof(attr));
4605 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
4606 memset(&attr, 0, sizeof(attr));
4607 if (copy_from_bpfptr(&attr, uattr, size) != 0)
4610 err = security_bpf(cmd, &attr, size);
4615 case BPF_MAP_CREATE:
4616 err = map_create(&attr);
4618 case BPF_MAP_LOOKUP_ELEM:
4619 err = map_lookup_elem(&attr);
4621 case BPF_MAP_UPDATE_ELEM:
4622 err = map_update_elem(&attr, uattr);
4624 case BPF_MAP_DELETE_ELEM:
4625 err = map_delete_elem(&attr);
4627 case BPF_MAP_GET_NEXT_KEY:
4628 err = map_get_next_key(&attr);
4630 case BPF_MAP_FREEZE:
4631 err = map_freeze(&attr);
4634 err = bpf_prog_load(&attr, uattr);
4637 err = bpf_obj_pin(&attr);
4640 err = bpf_obj_get(&attr);
4642 case BPF_PROG_ATTACH:
4643 err = bpf_prog_attach(&attr);
4645 case BPF_PROG_DETACH:
4646 err = bpf_prog_detach(&attr);
4648 case BPF_PROG_QUERY:
4649 err = bpf_prog_query(&attr, uattr.user);
4651 case BPF_PROG_TEST_RUN:
4652 err = bpf_prog_test_run(&attr, uattr.user);
4654 case BPF_PROG_GET_NEXT_ID:
4655 err = bpf_obj_get_next_id(&attr, uattr.user,
4656 &prog_idr, &prog_idr_lock);
4658 case BPF_MAP_GET_NEXT_ID:
4659 err = bpf_obj_get_next_id(&attr, uattr.user,
4660 &map_idr, &map_idr_lock);
4662 case BPF_BTF_GET_NEXT_ID:
4663 err = bpf_obj_get_next_id(&attr, uattr.user,
4664 &btf_idr, &btf_idr_lock);
4666 case BPF_PROG_GET_FD_BY_ID:
4667 err = bpf_prog_get_fd_by_id(&attr);
4669 case BPF_MAP_GET_FD_BY_ID:
4670 err = bpf_map_get_fd_by_id(&attr);
4672 case BPF_OBJ_GET_INFO_BY_FD:
4673 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
4675 case BPF_RAW_TRACEPOINT_OPEN:
4676 err = bpf_raw_tracepoint_open(&attr);
4679 err = bpf_btf_load(&attr, uattr);
4681 case BPF_BTF_GET_FD_BY_ID:
4682 err = bpf_btf_get_fd_by_id(&attr);
4684 case BPF_TASK_FD_QUERY:
4685 err = bpf_task_fd_query(&attr, uattr.user);
4687 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
4688 err = map_lookup_and_delete_elem(&attr);
4690 case BPF_MAP_LOOKUP_BATCH:
4691 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
4693 case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
4694 err = bpf_map_do_batch(&attr, uattr.user,
4695 BPF_MAP_LOOKUP_AND_DELETE_BATCH);
4697 case BPF_MAP_UPDATE_BATCH:
4698 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
4700 case BPF_MAP_DELETE_BATCH:
4701 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
4703 case BPF_LINK_CREATE:
4704 err = link_create(&attr, uattr);
4706 case BPF_LINK_UPDATE:
4707 err = link_update(&attr);
4709 case BPF_LINK_GET_FD_BY_ID:
4710 err = bpf_link_get_fd_by_id(&attr);
4712 case BPF_LINK_GET_NEXT_ID:
4713 err = bpf_obj_get_next_id(&attr, uattr.user,
4714 &link_idr, &link_idr_lock);
4716 case BPF_ENABLE_STATS:
4717 err = bpf_enable_stats(&attr);
4719 case BPF_ITER_CREATE:
4720 err = bpf_iter_create(&attr);
4722 case BPF_LINK_DETACH:
4723 err = link_detach(&attr);
4725 case BPF_PROG_BIND_MAP:
4726 err = bpf_prog_bind_map(&attr);
4736 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
4738 return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
4741 static bool syscall_prog_is_valid_access(int off, int size,
4742 enum bpf_access_type type,
4743 const struct bpf_prog *prog,
4744 struct bpf_insn_access_aux *info)
4746 if (off < 0 || off >= U16_MAX)
4748 if (off % size != 0)
4753 BPF_CALL_3(bpf_sys_bpf, int, cmd, void *, attr, u32, attr_size)
4756 case BPF_MAP_CREATE:
4757 case BPF_MAP_UPDATE_ELEM:
4758 case BPF_MAP_FREEZE:
4762 /* case BPF_PROG_TEST_RUN:
4763 * is not part of this list to prevent recursive test_run
4768 return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
4771 static const struct bpf_func_proto bpf_sys_bpf_proto = {
4772 .func = bpf_sys_bpf,
4774 .ret_type = RET_INTEGER,
4775 .arg1_type = ARG_ANYTHING,
4776 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4777 .arg3_type = ARG_CONST_SIZE,
4780 const struct bpf_func_proto * __weak
4781 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4783 return bpf_base_func_proto(func_id);
4786 BPF_CALL_1(bpf_sys_close, u32, fd)
4788 /* When bpf program calls this helper there should not be
4789 * an fdget() without matching completed fdput().
4790 * This helper is allowed in the following callchain only:
4791 * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
4793 return close_fd(fd);
4796 static const struct bpf_func_proto bpf_sys_close_proto = {
4797 .func = bpf_sys_close,
4799 .ret_type = RET_INTEGER,
4800 .arg1_type = ARG_ANYTHING,
4803 BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
4808 if (name_sz <= 1 || name[name_sz - 1])
4811 if (!bpf_dump_raw_ok(current_cred()))
4814 *res = kallsyms_lookup_name(name);
4815 return *res ? 0 : -ENOENT;
4818 const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
4819 .func = bpf_kallsyms_lookup_name,
4821 .ret_type = RET_INTEGER,
4822 .arg1_type = ARG_PTR_TO_MEM,
4823 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
4824 .arg3_type = ARG_ANYTHING,
4825 .arg4_type = ARG_PTR_TO_LONG,
4828 static const struct bpf_func_proto *
4829 syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4832 case BPF_FUNC_sys_bpf:
4833 return &bpf_sys_bpf_proto;
4834 case BPF_FUNC_btf_find_by_name_kind:
4835 return &bpf_btf_find_by_name_kind_proto;
4836 case BPF_FUNC_sys_close:
4837 return &bpf_sys_close_proto;
4838 case BPF_FUNC_kallsyms_lookup_name:
4839 return &bpf_kallsyms_lookup_name_proto;
4841 return tracing_prog_func_proto(func_id, prog);
4845 const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
4846 .get_func_proto = syscall_prog_func_proto,
4847 .is_valid_access = syscall_prog_is_valid_access,
4850 const struct bpf_prog_ops bpf_syscall_prog_ops = {
4851 .test_run = bpf_prog_test_run_syscall,