1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
5 #include <linux/bpf_trace.h>
6 #include <linux/bpf_lirc.h>
7 #include <linux/bpf_verifier.h>
9 #include <linux/syscalls.h>
10 #include <linux/slab.h>
11 #include <linux/sched/signal.h>
12 #include <linux/vmalloc.h>
13 #include <linux/mmzone.h>
14 #include <linux/anon_inodes.h>
15 #include <linux/fdtable.h>
16 #include <linux/file.h>
18 #include <linux/license.h>
19 #include <linux/filter.h>
20 #include <linux/kernel.h>
21 #include <linux/idr.h>
22 #include <linux/cred.h>
23 #include <linux/timekeeping.h>
24 #include <linux/ctype.h>
25 #include <linux/nospec.h>
26 #include <linux/audit.h>
27 #include <uapi/linux/btf.h>
28 #include <linux/pgtable.h>
29 #include <linux/bpf_lsm.h>
30 #include <linux/poll.h>
31 #include <linux/bpf-netns.h>
32 #include <linux/rcupdate_trace.h>
33 #include <linux/memcontrol.h>
35 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
36 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
37 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
38 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
39 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
40 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
43 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
45 DEFINE_PER_CPU(int, bpf_prog_active);
46 static DEFINE_IDR(prog_idr);
47 static DEFINE_SPINLOCK(prog_idr_lock);
48 static DEFINE_IDR(map_idr);
49 static DEFINE_SPINLOCK(map_idr_lock);
50 static DEFINE_IDR(link_idr);
51 static DEFINE_SPINLOCK(link_idr_lock);
53 int sysctl_unprivileged_bpf_disabled __read_mostly;
55 static const struct bpf_map_ops * const bpf_map_types[] = {
56 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
57 #define BPF_MAP_TYPE(_id, _ops) \
59 #define BPF_LINK_TYPE(_id, _name)
60 #include <linux/bpf_types.h>
67 * If we're handed a bigger struct than we know of, ensure all the unknown bits
68 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
69 * we don't know about yet.
71 * There is a ToCToU between this function call and the following
72 * copy_from_user() call. However, this is not a concern since this function is
73 * meant to be a future-proofing of bits.
75 int bpf_check_uarg_tail_zero(void __user *uaddr,
79 unsigned char __user *addr = uaddr + expected_size;
82 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
85 if (actual_size <= expected_size)
88 res = check_zeroed_user(addr, actual_size - expected_size);
91 return res ? 0 : -E2BIG;
94 const struct bpf_map_ops bpf_map_offload_ops = {
95 .map_meta_equal = bpf_map_meta_equal,
96 .map_alloc = bpf_map_offload_map_alloc,
97 .map_free = bpf_map_offload_map_free,
98 .map_check_btf = map_check_no_btf,
101 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
103 const struct bpf_map_ops *ops;
104 u32 type = attr->map_type;
108 if (type >= ARRAY_SIZE(bpf_map_types))
109 return ERR_PTR(-EINVAL);
110 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
111 ops = bpf_map_types[type];
113 return ERR_PTR(-EINVAL);
115 if (ops->map_alloc_check) {
116 err = ops->map_alloc_check(attr);
120 if (attr->map_ifindex)
121 ops = &bpf_map_offload_ops;
122 map = ops->map_alloc(attr);
126 map->map_type = type;
130 static u32 bpf_map_value_size(const struct bpf_map *map)
132 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
133 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
134 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
135 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
136 return round_up(map->value_size, 8) * num_possible_cpus();
137 else if (IS_FD_MAP(map))
140 return map->value_size;
143 static void maybe_wait_bpf_programs(struct bpf_map *map)
145 /* Wait for any running BPF programs to complete so that
146 * userspace, when we return to it, knows that all programs
147 * that could be running use the new map value.
149 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
150 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
154 static int bpf_map_update_value(struct bpf_map *map, struct fd f, void *key,
155 void *value, __u64 flags)
159 /* Need to create a kthread, thus must support schedule */
160 if (bpf_map_is_dev_bound(map)) {
161 return bpf_map_offload_update_elem(map, key, value, flags);
162 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
163 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
164 return map->ops->map_update_elem(map, key, value, flags);
165 } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
166 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
167 return sock_map_update_elem_sys(map, key, value, flags);
168 } else if (IS_FD_PROG_ARRAY(map)) {
169 return bpf_fd_array_map_update_elem(map, f.file, key, value,
173 bpf_disable_instrumentation();
174 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
175 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
176 err = bpf_percpu_hash_update(map, key, value, flags);
177 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
178 err = bpf_percpu_array_update(map, key, value, flags);
179 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
180 err = bpf_percpu_cgroup_storage_update(map, key, value,
182 } else if (IS_FD_ARRAY(map)) {
184 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
187 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
189 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
192 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
193 /* rcu_read_lock() is not needed */
194 err = bpf_fd_reuseport_array_update_elem(map, key, value,
196 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
197 map->map_type == BPF_MAP_TYPE_STACK) {
198 err = map->ops->map_push_elem(map, value, flags);
201 err = map->ops->map_update_elem(map, key, value, flags);
204 bpf_enable_instrumentation();
205 maybe_wait_bpf_programs(map);
210 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
216 if (bpf_map_is_dev_bound(map))
217 return bpf_map_offload_lookup_elem(map, key, value);
219 bpf_disable_instrumentation();
220 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
221 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
222 err = bpf_percpu_hash_copy(map, key, value);
223 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
224 err = bpf_percpu_array_copy(map, key, value);
225 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
226 err = bpf_percpu_cgroup_storage_copy(map, key, value);
227 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
228 err = bpf_stackmap_copy(map, key, value);
229 } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
230 err = bpf_fd_array_map_lookup_elem(map, key, value);
231 } else if (IS_FD_HASH(map)) {
232 err = bpf_fd_htab_map_lookup_elem(map, key, value);
233 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
234 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
235 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
236 map->map_type == BPF_MAP_TYPE_STACK) {
237 err = map->ops->map_peek_elem(map, value);
238 } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
239 /* struct_ops map requires directly updating "value" */
240 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
243 if (map->ops->map_lookup_elem_sys_only)
244 ptr = map->ops->map_lookup_elem_sys_only(map, key);
246 ptr = map->ops->map_lookup_elem(map, key);
253 if (flags & BPF_F_LOCK)
254 /* lock 'ptr' and copy everything but lock */
255 copy_map_value_locked(map, value, ptr, true);
257 copy_map_value(map, value, ptr);
258 /* mask lock, since value wasn't zero inited */
259 check_and_init_map_lock(map, value);
264 bpf_enable_instrumentation();
265 maybe_wait_bpf_programs(map);
270 /* Please, do not use this function outside from the map creation path
271 * (e.g. in map update path) without taking care of setting the active
272 * memory cgroup (see at bpf_map_kmalloc_node() for example).
274 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
276 /* We really just want to fail instead of triggering OOM killer
277 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
278 * which is used for lower order allocation requests.
280 * It has been observed that higher order allocation requests done by
281 * vmalloc with __GFP_NORETRY being set might fail due to not trying
282 * to reclaim memory from the page cache, thus we set
283 * __GFP_RETRY_MAYFAIL to avoid such situations.
286 const gfp_t gfp = __GFP_NOWARN | __GFP_ZERO | __GFP_ACCOUNT;
287 unsigned int flags = 0;
288 unsigned long align = 1;
291 if (size >= SIZE_MAX)
294 /* kmalloc()'ed memory can't be mmap()'ed */
296 BUG_ON(!PAGE_ALIGNED(size));
299 } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
300 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
306 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
307 gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
308 flags, numa_node, __builtin_return_address(0));
311 void *bpf_map_area_alloc(u64 size, int numa_node)
313 return __bpf_map_area_alloc(size, numa_node, false);
316 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
318 return __bpf_map_area_alloc(size, numa_node, true);
321 void bpf_map_area_free(void *area)
326 static u32 bpf_map_flags_retain_permanent(u32 flags)
328 /* Some map creation flags are not tied to the map object but
329 * rather to the map fd instead, so they have no meaning upon
330 * map object inspection since multiple file descriptors with
331 * different (access) properties can exist here. Thus, given
332 * this has zero meaning for the map itself, lets clear these
335 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
338 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
340 map->map_type = attr->map_type;
341 map->key_size = attr->key_size;
342 map->value_size = attr->value_size;
343 map->max_entries = attr->max_entries;
344 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
345 map->numa_node = bpf_map_attr_numa_node(attr);
348 static int bpf_map_alloc_id(struct bpf_map *map)
352 idr_preload(GFP_KERNEL);
353 spin_lock_bh(&map_idr_lock);
354 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
357 spin_unlock_bh(&map_idr_lock);
360 if (WARN_ON_ONCE(!id))
363 return id > 0 ? 0 : id;
366 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
370 /* Offloaded maps are removed from the IDR store when their device
371 * disappears - even if someone holds an fd to them they are unusable,
372 * the memory is gone, all ops will fail; they are simply waiting for
373 * refcnt to drop to be freed.
379 spin_lock_irqsave(&map_idr_lock, flags);
381 __acquire(&map_idr_lock);
383 idr_remove(&map_idr, map->id);
387 spin_unlock_irqrestore(&map_idr_lock, flags);
389 __release(&map_idr_lock);
392 #ifdef CONFIG_MEMCG_KMEM
393 static void bpf_map_save_memcg(struct bpf_map *map)
395 map->memcg = get_mem_cgroup_from_mm(current->mm);
398 static void bpf_map_release_memcg(struct bpf_map *map)
400 mem_cgroup_put(map->memcg);
403 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
406 struct mem_cgroup *old_memcg;
409 old_memcg = set_active_memcg(map->memcg);
410 ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
411 set_active_memcg(old_memcg);
416 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
418 struct mem_cgroup *old_memcg;
421 old_memcg = set_active_memcg(map->memcg);
422 ptr = kzalloc(size, flags | __GFP_ACCOUNT);
423 set_active_memcg(old_memcg);
428 void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
429 size_t align, gfp_t flags)
431 struct mem_cgroup *old_memcg;
434 old_memcg = set_active_memcg(map->memcg);
435 ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
436 set_active_memcg(old_memcg);
442 static void bpf_map_save_memcg(struct bpf_map *map)
446 static void bpf_map_release_memcg(struct bpf_map *map)
451 /* called from workqueue */
452 static void bpf_map_free_deferred(struct work_struct *work)
454 struct bpf_map *map = container_of(work, struct bpf_map, work);
456 security_bpf_map_free(map);
457 bpf_map_release_memcg(map);
458 /* implementation dependent freeing */
459 map->ops->map_free(map);
462 static void bpf_map_put_uref(struct bpf_map *map)
464 if (atomic64_dec_and_test(&map->usercnt)) {
465 if (map->ops->map_release_uref)
466 map->ops->map_release_uref(map);
470 /* decrement map refcnt and schedule it for freeing via workqueue
471 * (unrelying map implementation ops->map_free() might sleep)
473 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
475 if (atomic64_dec_and_test(&map->refcnt)) {
476 /* bpf_map_free_id() must be called first */
477 bpf_map_free_id(map, do_idr_lock);
479 INIT_WORK(&map->work, bpf_map_free_deferred);
480 schedule_work(&map->work);
484 void bpf_map_put(struct bpf_map *map)
486 __bpf_map_put(map, true);
488 EXPORT_SYMBOL_GPL(bpf_map_put);
490 void bpf_map_put_with_uref(struct bpf_map *map)
492 bpf_map_put_uref(map);
496 static int bpf_map_release(struct inode *inode, struct file *filp)
498 struct bpf_map *map = filp->private_data;
500 if (map->ops->map_release)
501 map->ops->map_release(map, filp);
503 bpf_map_put_with_uref(map);
507 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
509 fmode_t mode = f.file->f_mode;
511 /* Our file permissions may have been overridden by global
512 * map permissions facing syscall side.
514 if (READ_ONCE(map->frozen))
515 mode &= ~FMODE_CAN_WRITE;
519 #ifdef CONFIG_PROC_FS
520 /* Provides an approximation of the map's memory footprint.
521 * Used only to provide a backward compatibility and display
522 * a reasonable "memlock" info.
524 static unsigned long bpf_map_memory_footprint(const struct bpf_map *map)
528 size = round_up(map->key_size + bpf_map_value_size(map), 8);
530 return round_up(map->max_entries * size, PAGE_SIZE);
533 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
535 const struct bpf_map *map = filp->private_data;
536 const struct bpf_array *array;
537 u32 type = 0, jited = 0;
539 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
540 array = container_of(map, struct bpf_array, map);
541 type = array->aux->type;
542 jited = array->aux->jited;
559 bpf_map_memory_footprint(map),
561 READ_ONCE(map->frozen));
563 seq_printf(m, "owner_prog_type:\t%u\n", type);
564 seq_printf(m, "owner_jited:\t%u\n", jited);
569 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
572 /* We need this handler such that alloc_file() enables
573 * f_mode with FMODE_CAN_READ.
578 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
579 size_t siz, loff_t *ppos)
581 /* We need this handler such that alloc_file() enables
582 * f_mode with FMODE_CAN_WRITE.
587 /* called for any extra memory-mapped regions (except initial) */
588 static void bpf_map_mmap_open(struct vm_area_struct *vma)
590 struct bpf_map *map = vma->vm_file->private_data;
592 if (vma->vm_flags & VM_MAYWRITE) {
593 mutex_lock(&map->freeze_mutex);
595 mutex_unlock(&map->freeze_mutex);
599 /* called for all unmapped memory region (including initial) */
600 static void bpf_map_mmap_close(struct vm_area_struct *vma)
602 struct bpf_map *map = vma->vm_file->private_data;
604 if (vma->vm_flags & VM_MAYWRITE) {
605 mutex_lock(&map->freeze_mutex);
607 mutex_unlock(&map->freeze_mutex);
611 static const struct vm_operations_struct bpf_map_default_vmops = {
612 .open = bpf_map_mmap_open,
613 .close = bpf_map_mmap_close,
616 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
618 struct bpf_map *map = filp->private_data;
621 if (!map->ops->map_mmap || map_value_has_spin_lock(map))
624 if (!(vma->vm_flags & VM_SHARED))
627 mutex_lock(&map->freeze_mutex);
629 if (vma->vm_flags & VM_WRITE) {
634 /* map is meant to be read-only, so do not allow mapping as
635 * writable, because it's possible to leak a writable page
636 * reference and allows user-space to still modify it after
637 * freezing, while verifier will assume contents do not change
639 if (map->map_flags & BPF_F_RDONLY_PROG) {
645 /* set default open/close callbacks */
646 vma->vm_ops = &bpf_map_default_vmops;
647 vma->vm_private_data = map;
648 vma->vm_flags &= ~VM_MAYEXEC;
649 if (!(vma->vm_flags & VM_WRITE))
650 /* disallow re-mapping with PROT_WRITE */
651 vma->vm_flags &= ~VM_MAYWRITE;
653 err = map->ops->map_mmap(map, vma);
657 if (vma->vm_flags & VM_MAYWRITE)
660 mutex_unlock(&map->freeze_mutex);
664 static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
666 struct bpf_map *map = filp->private_data;
668 if (map->ops->map_poll)
669 return map->ops->map_poll(map, filp, pts);
674 const struct file_operations bpf_map_fops = {
675 #ifdef CONFIG_PROC_FS
676 .show_fdinfo = bpf_map_show_fdinfo,
678 .release = bpf_map_release,
679 .read = bpf_dummy_read,
680 .write = bpf_dummy_write,
681 .mmap = bpf_map_mmap,
682 .poll = bpf_map_poll,
685 int bpf_map_new_fd(struct bpf_map *map, int flags)
689 ret = security_bpf_map(map, OPEN_FMODE(flags));
693 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
697 int bpf_get_file_flag(int flags)
699 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
701 if (flags & BPF_F_RDONLY)
703 if (flags & BPF_F_WRONLY)
708 /* helper macro to check that unused fields 'union bpf_attr' are zero */
709 #define CHECK_ATTR(CMD) \
710 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
711 sizeof(attr->CMD##_LAST_FIELD), 0, \
713 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
714 sizeof(attr->CMD##_LAST_FIELD)) != NULL
716 /* dst and src must have at least "size" number of bytes.
717 * Return strlen on success and < 0 on error.
719 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
721 const char *end = src + size;
722 const char *orig_src = src;
724 memset(dst, 0, size);
725 /* Copy all isalnum(), '_' and '.' chars. */
726 while (src < end && *src) {
727 if (!isalnum(*src) &&
728 *src != '_' && *src != '.')
733 /* No '\0' found in "size" number of bytes */
737 return src - orig_src;
740 int map_check_no_btf(const struct bpf_map *map,
741 const struct btf *btf,
742 const struct btf_type *key_type,
743 const struct btf_type *value_type)
748 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
749 u32 btf_key_id, u32 btf_value_id)
751 const struct btf_type *key_type, *value_type;
752 u32 key_size, value_size;
755 /* Some maps allow key to be unspecified. */
757 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
758 if (!key_type || key_size != map->key_size)
761 key_type = btf_type_by_id(btf, 0);
762 if (!map->ops->map_check_btf)
766 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
767 if (!value_type || value_size != map->value_size)
770 map->spin_lock_off = btf_find_spin_lock(btf, value_type);
772 if (map_value_has_spin_lock(map)) {
773 if (map->map_flags & BPF_F_RDONLY_PROG)
775 if (map->map_type != BPF_MAP_TYPE_HASH &&
776 map->map_type != BPF_MAP_TYPE_ARRAY &&
777 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
778 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
779 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
780 map->map_type != BPF_MAP_TYPE_TASK_STORAGE)
782 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
785 "verifier bug spin_lock_off %d value_size %d\n",
786 map->spin_lock_off, map->value_size);
791 if (map->ops->map_check_btf)
792 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
797 #define BPF_MAP_CREATE_LAST_FIELD btf_vmlinux_value_type_id
798 /* called via syscall */
799 static int map_create(union bpf_attr *attr)
801 int numa_node = bpf_map_attr_numa_node(attr);
806 err = CHECK_ATTR(BPF_MAP_CREATE);
810 if (attr->btf_vmlinux_value_type_id) {
811 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
812 attr->btf_key_type_id || attr->btf_value_type_id)
814 } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
818 f_flags = bpf_get_file_flag(attr->map_flags);
822 if (numa_node != NUMA_NO_NODE &&
823 ((unsigned int)numa_node >= nr_node_ids ||
824 !node_online(numa_node)))
827 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
828 map = find_and_alloc_map(attr);
832 err = bpf_obj_name_cpy(map->name, attr->map_name,
833 sizeof(attr->map_name));
837 atomic64_set(&map->refcnt, 1);
838 atomic64_set(&map->usercnt, 1);
839 mutex_init(&map->freeze_mutex);
841 map->spin_lock_off = -EINVAL;
842 if (attr->btf_key_type_id || attr->btf_value_type_id ||
843 /* Even the map's value is a kernel's struct,
844 * the bpf_prog.o must have BTF to begin with
845 * to figure out the corresponding kernel's
846 * counter part. Thus, attr->btf_fd has
849 attr->btf_vmlinux_value_type_id) {
852 btf = btf_get_by_fd(attr->btf_fd);
859 if (attr->btf_value_type_id) {
860 err = map_check_btf(map, btf, attr->btf_key_type_id,
861 attr->btf_value_type_id);
866 map->btf_key_type_id = attr->btf_key_type_id;
867 map->btf_value_type_id = attr->btf_value_type_id;
868 map->btf_vmlinux_value_type_id =
869 attr->btf_vmlinux_value_type_id;
872 err = security_bpf_map_alloc(map);
876 err = bpf_map_alloc_id(map);
880 bpf_map_save_memcg(map);
882 err = bpf_map_new_fd(map, f_flags);
884 /* failed to allocate fd.
885 * bpf_map_put_with_uref() is needed because the above
886 * bpf_map_alloc_id() has published the map
887 * to the userspace and the userspace may
888 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
890 bpf_map_put_with_uref(map);
897 security_bpf_map_free(map);
900 map->ops->map_free(map);
904 /* if error is returned, fd is released.
905 * On success caller should complete fd access with matching fdput()
907 struct bpf_map *__bpf_map_get(struct fd f)
910 return ERR_PTR(-EBADF);
911 if (f.file->f_op != &bpf_map_fops) {
913 return ERR_PTR(-EINVAL);
916 return f.file->private_data;
919 void bpf_map_inc(struct bpf_map *map)
921 atomic64_inc(&map->refcnt);
923 EXPORT_SYMBOL_GPL(bpf_map_inc);
925 void bpf_map_inc_with_uref(struct bpf_map *map)
927 atomic64_inc(&map->refcnt);
928 atomic64_inc(&map->usercnt);
930 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
932 struct bpf_map *bpf_map_get(u32 ufd)
934 struct fd f = fdget(ufd);
937 map = __bpf_map_get(f);
947 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
949 struct fd f = fdget(ufd);
952 map = __bpf_map_get(f);
956 bpf_map_inc_with_uref(map);
962 /* map_idr_lock should have been held */
963 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
967 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
969 return ERR_PTR(-ENOENT);
971 atomic64_inc(&map->usercnt);
976 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
978 spin_lock_bh(&map_idr_lock);
979 map = __bpf_map_inc_not_zero(map, false);
980 spin_unlock_bh(&map_idr_lock);
984 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
986 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
991 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
994 return memdup_user(ukey, key_size);
997 return ERR_PTR(-EINVAL);
1002 /* last field in 'union bpf_attr' used by this command */
1003 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1005 static int map_lookup_elem(union bpf_attr *attr)
1007 void __user *ukey = u64_to_user_ptr(attr->key);
1008 void __user *uvalue = u64_to_user_ptr(attr->value);
1009 int ufd = attr->map_fd;
1010 struct bpf_map *map;
1016 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1019 if (attr->flags & ~BPF_F_LOCK)
1023 map = __bpf_map_get(f);
1025 return PTR_ERR(map);
1026 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1031 if ((attr->flags & BPF_F_LOCK) &&
1032 !map_value_has_spin_lock(map)) {
1037 key = __bpf_copy_key(ukey, map->key_size);
1043 value_size = bpf_map_value_size(map);
1046 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1050 err = bpf_map_copy_value(map, key, value, attr->flags);
1055 if (copy_to_user(uvalue, value, value_size) != 0)
1070 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1072 static int map_update_elem(union bpf_attr *attr)
1074 void __user *ukey = u64_to_user_ptr(attr->key);
1075 void __user *uvalue = u64_to_user_ptr(attr->value);
1076 int ufd = attr->map_fd;
1077 struct bpf_map *map;
1083 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1087 map = __bpf_map_get(f);
1089 return PTR_ERR(map);
1090 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1095 if ((attr->flags & BPF_F_LOCK) &&
1096 !map_value_has_spin_lock(map)) {
1101 key = __bpf_copy_key(ukey, map->key_size);
1107 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1108 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
1109 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
1110 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
1111 value_size = round_up(map->value_size, 8) * num_possible_cpus();
1113 value_size = map->value_size;
1116 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1121 if (copy_from_user(value, uvalue, value_size) != 0)
1124 err = bpf_map_update_value(map, f, key, value, attr->flags);
1135 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1137 static int map_delete_elem(union bpf_attr *attr)
1139 void __user *ukey = u64_to_user_ptr(attr->key);
1140 int ufd = attr->map_fd;
1141 struct bpf_map *map;
1146 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1150 map = __bpf_map_get(f);
1152 return PTR_ERR(map);
1153 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1158 key = __bpf_copy_key(ukey, map->key_size);
1164 if (bpf_map_is_dev_bound(map)) {
1165 err = bpf_map_offload_delete_elem(map, key);
1167 } else if (IS_FD_PROG_ARRAY(map) ||
1168 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1169 /* These maps require sleepable context */
1170 err = map->ops->map_delete_elem(map, key);
1174 bpf_disable_instrumentation();
1176 err = map->ops->map_delete_elem(map, key);
1178 bpf_enable_instrumentation();
1179 maybe_wait_bpf_programs(map);
1187 /* last field in 'union bpf_attr' used by this command */
1188 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1190 static int map_get_next_key(union bpf_attr *attr)
1192 void __user *ukey = u64_to_user_ptr(attr->key);
1193 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1194 int ufd = attr->map_fd;
1195 struct bpf_map *map;
1196 void *key, *next_key;
1200 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1204 map = __bpf_map_get(f);
1206 return PTR_ERR(map);
1207 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1213 key = __bpf_copy_key(ukey, map->key_size);
1223 next_key = kmalloc(map->key_size, GFP_USER);
1227 if (bpf_map_is_dev_bound(map)) {
1228 err = bpf_map_offload_get_next_key(map, key, next_key);
1233 err = map->ops->map_get_next_key(map, key, next_key);
1240 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1254 int generic_map_delete_batch(struct bpf_map *map,
1255 const union bpf_attr *attr,
1256 union bpf_attr __user *uattr)
1258 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1263 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1266 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1267 !map_value_has_spin_lock(map)) {
1271 max_count = attr->batch.count;
1275 key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1279 for (cp = 0; cp < max_count; cp++) {
1281 if (copy_from_user(key, keys + cp * map->key_size,
1285 if (bpf_map_is_dev_bound(map)) {
1286 err = bpf_map_offload_delete_elem(map, key);
1290 bpf_disable_instrumentation();
1292 err = map->ops->map_delete_elem(map, key);
1294 bpf_enable_instrumentation();
1295 maybe_wait_bpf_programs(map);
1299 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1306 int generic_map_update_batch(struct bpf_map *map,
1307 const union bpf_attr *attr,
1308 union bpf_attr __user *uattr)
1310 void __user *values = u64_to_user_ptr(attr->batch.values);
1311 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1312 u32 value_size, cp, max_count;
1313 int ufd = attr->map_fd;
1319 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1322 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1323 !map_value_has_spin_lock(map)) {
1327 value_size = bpf_map_value_size(map);
1329 max_count = attr->batch.count;
1333 key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1337 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1343 for (cp = 0; cp < max_count; cp++) {
1345 if (copy_from_user(key, keys + cp * map->key_size,
1347 copy_from_user(value, values + cp * value_size, value_size))
1350 err = bpf_map_update_value(map, f, key, value,
1351 attr->batch.elem_flags);
1357 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1365 #define MAP_LOOKUP_RETRIES 3
1367 int generic_map_lookup_batch(struct bpf_map *map,
1368 const union bpf_attr *attr,
1369 union bpf_attr __user *uattr)
1371 void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1372 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1373 void __user *values = u64_to_user_ptr(attr->batch.values);
1374 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1375 void *buf, *buf_prevkey, *prev_key, *key, *value;
1376 int err, retry = MAP_LOOKUP_RETRIES;
1377 u32 value_size, cp, max_count;
1379 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1382 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1383 !map_value_has_spin_lock(map))
1386 value_size = bpf_map_value_size(map);
1388 max_count = attr->batch.count;
1392 if (put_user(0, &uattr->batch.count))
1395 buf_prevkey = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1399 buf = kmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1407 if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1410 value = key + map->key_size;
1412 prev_key = buf_prevkey;
1414 for (cp = 0; cp < max_count;) {
1416 err = map->ops->map_get_next_key(map, prev_key, key);
1420 err = bpf_map_copy_value(map, key, value,
1421 attr->batch.elem_flags);
1423 if (err == -ENOENT) {
1435 if (copy_to_user(keys + cp * map->key_size, key,
1440 if (copy_to_user(values + cp * value_size, value, value_size)) {
1446 prev_key = buf_prevkey;
1448 swap(prev_key, key);
1449 retry = MAP_LOOKUP_RETRIES;
1456 if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1457 (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1466 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1468 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1470 void __user *ukey = u64_to_user_ptr(attr->key);
1471 void __user *uvalue = u64_to_user_ptr(attr->value);
1472 int ufd = attr->map_fd;
1473 struct bpf_map *map;
1479 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1483 map = __bpf_map_get(f);
1485 return PTR_ERR(map);
1486 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1487 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1492 key = __bpf_copy_key(ukey, map->key_size);
1498 value_size = map->value_size;
1501 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1505 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1506 map->map_type == BPF_MAP_TYPE_STACK) {
1507 err = map->ops->map_pop_elem(map, value);
1515 if (copy_to_user(uvalue, value, value_size) != 0) {
1531 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1533 static int map_freeze(const union bpf_attr *attr)
1535 int err = 0, ufd = attr->map_fd;
1536 struct bpf_map *map;
1539 if (CHECK_ATTR(BPF_MAP_FREEZE))
1543 map = __bpf_map_get(f);
1545 return PTR_ERR(map);
1547 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1552 mutex_lock(&map->freeze_mutex);
1554 if (map->writecnt) {
1558 if (READ_ONCE(map->frozen)) {
1562 if (!bpf_capable()) {
1567 WRITE_ONCE(map->frozen, true);
1569 mutex_unlock(&map->freeze_mutex);
1574 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1575 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1576 [_id] = & _name ## _prog_ops,
1577 #define BPF_MAP_TYPE(_id, _ops)
1578 #define BPF_LINK_TYPE(_id, _name)
1579 #include <linux/bpf_types.h>
1580 #undef BPF_PROG_TYPE
1582 #undef BPF_LINK_TYPE
1585 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1587 const struct bpf_prog_ops *ops;
1589 if (type >= ARRAY_SIZE(bpf_prog_types))
1591 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1592 ops = bpf_prog_types[type];
1596 if (!bpf_prog_is_dev_bound(prog->aux))
1597 prog->aux->ops = ops;
1599 prog->aux->ops = &bpf_offload_prog_ops;
1610 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1611 [BPF_AUDIT_LOAD] = "LOAD",
1612 [BPF_AUDIT_UNLOAD] = "UNLOAD",
1615 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1617 struct audit_context *ctx = NULL;
1618 struct audit_buffer *ab;
1620 if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1622 if (audit_enabled == AUDIT_OFF)
1624 if (op == BPF_AUDIT_LOAD)
1625 ctx = audit_context();
1626 ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1629 audit_log_format(ab, "prog-id=%u op=%s",
1630 prog->aux->id, bpf_audit_str[op]);
1634 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1638 idr_preload(GFP_KERNEL);
1639 spin_lock_bh(&prog_idr_lock);
1640 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1643 spin_unlock_bh(&prog_idr_lock);
1646 /* id is in [1, INT_MAX) */
1647 if (WARN_ON_ONCE(!id))
1650 return id > 0 ? 0 : id;
1653 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1655 /* cBPF to eBPF migrations are currently not in the idr store.
1656 * Offloaded programs are removed from the store when their device
1657 * disappears - even if someone grabs an fd to them they are unusable,
1658 * simply waiting for refcnt to drop to be freed.
1664 spin_lock_bh(&prog_idr_lock);
1666 __acquire(&prog_idr_lock);
1668 idr_remove(&prog_idr, prog->aux->id);
1672 spin_unlock_bh(&prog_idr_lock);
1674 __release(&prog_idr_lock);
1677 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1679 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1681 kvfree(aux->func_info);
1682 kfree(aux->func_info_aux);
1683 free_uid(aux->user);
1684 security_bpf_prog_free(aux);
1685 bpf_prog_free(aux->prog);
1688 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1690 bpf_prog_kallsyms_del_all(prog);
1691 btf_put(prog->aux->btf);
1692 bpf_prog_free_linfo(prog);
1693 if (prog->aux->attach_btf)
1694 btf_put(prog->aux->attach_btf);
1697 if (prog->aux->sleepable)
1698 call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
1700 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1702 __bpf_prog_put_rcu(&prog->aux->rcu);
1706 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1708 if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1709 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1710 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
1711 /* bpf_prog_free_id() must be called first */
1712 bpf_prog_free_id(prog, do_idr_lock);
1713 __bpf_prog_put_noref(prog, true);
1717 void bpf_prog_put(struct bpf_prog *prog)
1719 __bpf_prog_put(prog, true);
1721 EXPORT_SYMBOL_GPL(bpf_prog_put);
1723 static int bpf_prog_release(struct inode *inode, struct file *filp)
1725 struct bpf_prog *prog = filp->private_data;
1731 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1732 struct bpf_prog_stats *stats)
1734 u64 nsecs = 0, cnt = 0, misses = 0;
1737 for_each_possible_cpu(cpu) {
1738 const struct bpf_prog_stats *st;
1740 u64 tnsecs, tcnt, tmisses;
1742 st = per_cpu_ptr(prog->stats, cpu);
1744 start = u64_stats_fetch_begin_irq(&st->syncp);
1747 tmisses = st->misses;
1748 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1753 stats->nsecs = nsecs;
1755 stats->misses = misses;
1758 #ifdef CONFIG_PROC_FS
1759 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1761 const struct bpf_prog *prog = filp->private_data;
1762 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1763 struct bpf_prog_stats stats;
1765 bpf_prog_get_stats(prog, &stats);
1766 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1773 "run_time_ns:\t%llu\n"
1775 "recursion_misses:\t%llu\n",
1779 prog->pages * 1ULL << PAGE_SHIFT,
1787 const struct file_operations bpf_prog_fops = {
1788 #ifdef CONFIG_PROC_FS
1789 .show_fdinfo = bpf_prog_show_fdinfo,
1791 .release = bpf_prog_release,
1792 .read = bpf_dummy_read,
1793 .write = bpf_dummy_write,
1796 int bpf_prog_new_fd(struct bpf_prog *prog)
1800 ret = security_bpf_prog(prog);
1804 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1805 O_RDWR | O_CLOEXEC);
1808 static struct bpf_prog *____bpf_prog_get(struct fd f)
1811 return ERR_PTR(-EBADF);
1812 if (f.file->f_op != &bpf_prog_fops) {
1814 return ERR_PTR(-EINVAL);
1817 return f.file->private_data;
1820 void bpf_prog_add(struct bpf_prog *prog, int i)
1822 atomic64_add(i, &prog->aux->refcnt);
1824 EXPORT_SYMBOL_GPL(bpf_prog_add);
1826 void bpf_prog_sub(struct bpf_prog *prog, int i)
1828 /* Only to be used for undoing previous bpf_prog_add() in some
1829 * error path. We still know that another entity in our call
1830 * path holds a reference to the program, thus atomic_sub() can
1831 * be safely used in such cases!
1833 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1835 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1837 void bpf_prog_inc(struct bpf_prog *prog)
1839 atomic64_inc(&prog->aux->refcnt);
1841 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1843 /* prog_idr_lock should have been held */
1844 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1848 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1851 return ERR_PTR(-ENOENT);
1855 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1857 bool bpf_prog_get_ok(struct bpf_prog *prog,
1858 enum bpf_prog_type *attach_type, bool attach_drv)
1860 /* not an attachment, just a refcount inc, always allow */
1864 if (prog->type != *attach_type)
1866 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1872 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1875 struct fd f = fdget(ufd);
1876 struct bpf_prog *prog;
1878 prog = ____bpf_prog_get(f);
1881 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1882 prog = ERR_PTR(-EINVAL);
1892 struct bpf_prog *bpf_prog_get(u32 ufd)
1894 return __bpf_prog_get(ufd, NULL, false);
1897 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1900 return __bpf_prog_get(ufd, &type, attach_drv);
1902 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1904 /* Initially all BPF programs could be loaded w/o specifying
1905 * expected_attach_type. Later for some of them specifying expected_attach_type
1906 * at load time became required so that program could be validated properly.
1907 * Programs of types that are allowed to be loaded both w/ and w/o (for
1908 * backward compatibility) expected_attach_type, should have the default attach
1909 * type assigned to expected_attach_type for the latter case, so that it can be
1910 * validated later at attach time.
1912 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1913 * prog type requires it but has some attach types that have to be backward
1916 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1918 switch (attr->prog_type) {
1919 case BPF_PROG_TYPE_CGROUP_SOCK:
1920 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1921 * exist so checking for non-zero is the way to go here.
1923 if (!attr->expected_attach_type)
1924 attr->expected_attach_type =
1925 BPF_CGROUP_INET_SOCK_CREATE;
1931 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1932 enum bpf_attach_type expected_attach_type,
1933 struct btf *attach_btf, u32 btf_id,
1934 struct bpf_prog *dst_prog)
1937 if (btf_id > BTF_MAX_TYPE)
1940 if (!attach_btf && !dst_prog)
1943 switch (prog_type) {
1944 case BPF_PROG_TYPE_TRACING:
1945 case BPF_PROG_TYPE_LSM:
1946 case BPF_PROG_TYPE_STRUCT_OPS:
1947 case BPF_PROG_TYPE_EXT:
1954 if (attach_btf && (!btf_id || dst_prog))
1957 if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
1958 prog_type != BPF_PROG_TYPE_EXT)
1961 switch (prog_type) {
1962 case BPF_PROG_TYPE_CGROUP_SOCK:
1963 switch (expected_attach_type) {
1964 case BPF_CGROUP_INET_SOCK_CREATE:
1965 case BPF_CGROUP_INET_SOCK_RELEASE:
1966 case BPF_CGROUP_INET4_POST_BIND:
1967 case BPF_CGROUP_INET6_POST_BIND:
1972 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1973 switch (expected_attach_type) {
1974 case BPF_CGROUP_INET4_BIND:
1975 case BPF_CGROUP_INET6_BIND:
1976 case BPF_CGROUP_INET4_CONNECT:
1977 case BPF_CGROUP_INET6_CONNECT:
1978 case BPF_CGROUP_INET4_GETPEERNAME:
1979 case BPF_CGROUP_INET6_GETPEERNAME:
1980 case BPF_CGROUP_INET4_GETSOCKNAME:
1981 case BPF_CGROUP_INET6_GETSOCKNAME:
1982 case BPF_CGROUP_UDP4_SENDMSG:
1983 case BPF_CGROUP_UDP6_SENDMSG:
1984 case BPF_CGROUP_UDP4_RECVMSG:
1985 case BPF_CGROUP_UDP6_RECVMSG:
1990 case BPF_PROG_TYPE_CGROUP_SKB:
1991 switch (expected_attach_type) {
1992 case BPF_CGROUP_INET_INGRESS:
1993 case BPF_CGROUP_INET_EGRESS:
1998 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1999 switch (expected_attach_type) {
2000 case BPF_CGROUP_SETSOCKOPT:
2001 case BPF_CGROUP_GETSOCKOPT:
2006 case BPF_PROG_TYPE_SK_LOOKUP:
2007 if (expected_attach_type == BPF_SK_LOOKUP)
2010 case BPF_PROG_TYPE_EXT:
2011 if (expected_attach_type)
2019 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2021 switch (prog_type) {
2022 case BPF_PROG_TYPE_SCHED_CLS:
2023 case BPF_PROG_TYPE_SCHED_ACT:
2024 case BPF_PROG_TYPE_XDP:
2025 case BPF_PROG_TYPE_LWT_IN:
2026 case BPF_PROG_TYPE_LWT_OUT:
2027 case BPF_PROG_TYPE_LWT_XMIT:
2028 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2029 case BPF_PROG_TYPE_SK_SKB:
2030 case BPF_PROG_TYPE_SK_MSG:
2031 case BPF_PROG_TYPE_LIRC_MODE2:
2032 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2033 case BPF_PROG_TYPE_CGROUP_DEVICE:
2034 case BPF_PROG_TYPE_CGROUP_SOCK:
2035 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2036 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2037 case BPF_PROG_TYPE_CGROUP_SYSCTL:
2038 case BPF_PROG_TYPE_SOCK_OPS:
2039 case BPF_PROG_TYPE_EXT: /* extends any prog */
2041 case BPF_PROG_TYPE_CGROUP_SKB:
2043 case BPF_PROG_TYPE_SK_REUSEPORT:
2044 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2050 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2052 switch (prog_type) {
2053 case BPF_PROG_TYPE_KPROBE:
2054 case BPF_PROG_TYPE_TRACEPOINT:
2055 case BPF_PROG_TYPE_PERF_EVENT:
2056 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2057 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2058 case BPF_PROG_TYPE_TRACING:
2059 case BPF_PROG_TYPE_LSM:
2060 case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2061 case BPF_PROG_TYPE_EXT: /* extends any prog */
2068 /* last field in 'union bpf_attr' used by this command */
2069 #define BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
2071 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
2073 enum bpf_prog_type type = attr->prog_type;
2074 struct bpf_prog *prog, *dst_prog = NULL;
2075 struct btf *attach_btf = NULL;
2080 if (CHECK_ATTR(BPF_PROG_LOAD))
2083 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2084 BPF_F_ANY_ALIGNMENT |
2085 BPF_F_TEST_STATE_FREQ |
2087 BPF_F_TEST_RND_HI32))
2090 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2091 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2095 /* copy eBPF program license from user space */
2096 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
2097 sizeof(license) - 1) < 0)
2099 license[sizeof(license) - 1] = 0;
2101 /* eBPF programs must be GPL compatible to use GPL-ed functions */
2102 is_gpl = license_is_gpl_compatible(license);
2104 if (attr->insn_cnt == 0 ||
2105 attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2107 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2108 type != BPF_PROG_TYPE_CGROUP_SKB &&
2112 if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2114 if (is_perfmon_prog_type(type) && !perfmon_capable())
2117 /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2118 * or btf, we need to check which one it is
2120 if (attr->attach_prog_fd) {
2121 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2122 if (IS_ERR(dst_prog)) {
2124 attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2125 if (IS_ERR(attach_btf))
2127 if (!btf_is_kernel(attach_btf)) {
2128 /* attaching through specifying bpf_prog's BTF
2129 * objects directly might be supported eventually
2131 btf_put(attach_btf);
2135 } else if (attr->attach_btf_id) {
2136 /* fall back to vmlinux BTF, if BTF type ID is specified */
2137 attach_btf = bpf_get_btf_vmlinux();
2138 if (IS_ERR(attach_btf))
2139 return PTR_ERR(attach_btf);
2142 btf_get(attach_btf);
2145 bpf_prog_load_fixup_attach_type(attr);
2146 if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2147 attach_btf, attr->attach_btf_id,
2150 bpf_prog_put(dst_prog);
2152 btf_put(attach_btf);
2156 /* plain bpf_prog allocation */
2157 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2160 bpf_prog_put(dst_prog);
2162 btf_put(attach_btf);
2166 prog->expected_attach_type = attr->expected_attach_type;
2167 prog->aux->attach_btf = attach_btf;
2168 prog->aux->attach_btf_id = attr->attach_btf_id;
2169 prog->aux->dst_prog = dst_prog;
2170 prog->aux->offload_requested = !!attr->prog_ifindex;
2171 prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2173 err = security_bpf_prog_alloc(prog->aux);
2177 prog->aux->user = get_current_user();
2178 prog->len = attr->insn_cnt;
2181 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
2182 bpf_prog_insn_size(prog)) != 0)
2185 prog->orig_prog = NULL;
2188 atomic64_set(&prog->aux->refcnt, 1);
2189 prog->gpl_compatible = is_gpl ? 1 : 0;
2191 if (bpf_prog_is_dev_bound(prog->aux)) {
2192 err = bpf_prog_offload_init(prog, attr);
2197 /* find program type: socket_filter vs tracing_filter */
2198 err = find_prog_type(type, prog);
2202 prog->aux->load_time = ktime_get_boottime_ns();
2203 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2204 sizeof(attr->prog_name));
2208 /* run eBPF verifier */
2209 err = bpf_check(&prog, attr, uattr);
2211 goto free_used_maps;
2213 prog = bpf_prog_select_runtime(prog, &err);
2215 goto free_used_maps;
2217 err = bpf_prog_alloc_id(prog);
2219 goto free_used_maps;
2221 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2222 * effectively publicly exposed. However, retrieving via
2223 * bpf_prog_get_fd_by_id() will take another reference,
2224 * therefore it cannot be gone underneath us.
2226 * Only for the time /after/ successful bpf_prog_new_fd()
2227 * and before returning to userspace, we might just hold
2228 * one reference and any parallel close on that fd could
2229 * rip everything out. Hence, below notifications must
2230 * happen before bpf_prog_new_fd().
2232 * Also, any failure handling from this point onwards must
2233 * be using bpf_prog_put() given the program is exposed.
2235 bpf_prog_kallsyms_add(prog);
2236 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2237 bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2239 err = bpf_prog_new_fd(prog);
2245 /* In case we have subprogs, we need to wait for a grace
2246 * period before we can tear down JIT memory since symbols
2247 * are already exposed under kallsyms.
2249 __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2252 free_uid(prog->aux->user);
2253 security_bpf_prog_free(prog->aux);
2255 if (prog->aux->attach_btf)
2256 btf_put(prog->aux->attach_btf);
2257 bpf_prog_free(prog);
2261 #define BPF_OBJ_LAST_FIELD file_flags
2263 static int bpf_obj_pin(const union bpf_attr *attr)
2265 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2268 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2271 static int bpf_obj_get(const union bpf_attr *attr)
2273 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2274 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2277 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2281 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2282 const struct bpf_link_ops *ops, struct bpf_prog *prog)
2284 atomic64_set(&link->refcnt, 1);
2291 static void bpf_link_free_id(int id)
2296 spin_lock_bh(&link_idr_lock);
2297 idr_remove(&link_idr, id);
2298 spin_unlock_bh(&link_idr_lock);
2301 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2302 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2303 * anon_inode's release() call. This helper marksbpf_link as
2304 * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2305 * is not decremented, it's the responsibility of a calling code that failed
2306 * to complete bpf_link initialization.
2308 void bpf_link_cleanup(struct bpf_link_primer *primer)
2310 primer->link->prog = NULL;
2311 bpf_link_free_id(primer->id);
2313 put_unused_fd(primer->fd);
2316 void bpf_link_inc(struct bpf_link *link)
2318 atomic64_inc(&link->refcnt);
2321 /* bpf_link_free is guaranteed to be called from process context */
2322 static void bpf_link_free(struct bpf_link *link)
2324 bpf_link_free_id(link->id);
2326 /* detach BPF program, clean up used resources */
2327 link->ops->release(link);
2328 bpf_prog_put(link->prog);
2330 /* free bpf_link and its containing memory */
2331 link->ops->dealloc(link);
2334 static void bpf_link_put_deferred(struct work_struct *work)
2336 struct bpf_link *link = container_of(work, struct bpf_link, work);
2338 bpf_link_free(link);
2341 /* bpf_link_put can be called from atomic context, but ensures that resources
2342 * are freed from process context
2344 void bpf_link_put(struct bpf_link *link)
2346 if (!atomic64_dec_and_test(&link->refcnt))
2350 INIT_WORK(&link->work, bpf_link_put_deferred);
2351 schedule_work(&link->work);
2353 bpf_link_free(link);
2357 static int bpf_link_release(struct inode *inode, struct file *filp)
2359 struct bpf_link *link = filp->private_data;
2365 #ifdef CONFIG_PROC_FS
2366 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2367 #define BPF_MAP_TYPE(_id, _ops)
2368 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2369 static const char *bpf_link_type_strs[] = {
2370 [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2371 #include <linux/bpf_types.h>
2373 #undef BPF_PROG_TYPE
2375 #undef BPF_LINK_TYPE
2377 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2379 const struct bpf_link *link = filp->private_data;
2380 const struct bpf_prog *prog = link->prog;
2381 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2383 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2389 bpf_link_type_strs[link->type],
2393 if (link->ops->show_fdinfo)
2394 link->ops->show_fdinfo(link, m);
2398 static const struct file_operations bpf_link_fops = {
2399 #ifdef CONFIG_PROC_FS
2400 .show_fdinfo = bpf_link_show_fdinfo,
2402 .release = bpf_link_release,
2403 .read = bpf_dummy_read,
2404 .write = bpf_dummy_write,
2407 static int bpf_link_alloc_id(struct bpf_link *link)
2411 idr_preload(GFP_KERNEL);
2412 spin_lock_bh(&link_idr_lock);
2413 id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2414 spin_unlock_bh(&link_idr_lock);
2420 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2421 * reserving unused FD and allocating ID from link_idr. This is to be paired
2422 * with bpf_link_settle() to install FD and ID and expose bpf_link to
2423 * user-space, if bpf_link is successfully attached. If not, bpf_link and
2424 * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2425 * transient state is passed around in struct bpf_link_primer.
2426 * This is preferred way to create and initialize bpf_link, especially when
2427 * there are complicated and expensive operations inbetween creating bpf_link
2428 * itself and attaching it to BPF hook. By using bpf_link_prime() and
2429 * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2430 * expensive (and potentially failing) roll back operations in a rare case
2431 * that file, FD, or ID can't be allocated.
2433 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2438 fd = get_unused_fd_flags(O_CLOEXEC);
2443 id = bpf_link_alloc_id(link);
2449 file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2451 bpf_link_free_id(id);
2453 return PTR_ERR(file);
2456 primer->link = link;
2457 primer->file = file;
2463 int bpf_link_settle(struct bpf_link_primer *primer)
2465 /* make bpf_link fetchable by ID */
2466 spin_lock_bh(&link_idr_lock);
2467 primer->link->id = primer->id;
2468 spin_unlock_bh(&link_idr_lock);
2469 /* make bpf_link fetchable by FD */
2470 fd_install(primer->fd, primer->file);
2471 /* pass through installed FD */
2475 int bpf_link_new_fd(struct bpf_link *link)
2477 return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2480 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2482 struct fd f = fdget(ufd);
2483 struct bpf_link *link;
2486 return ERR_PTR(-EBADF);
2487 if (f.file->f_op != &bpf_link_fops) {
2489 return ERR_PTR(-EINVAL);
2492 link = f.file->private_data;
2499 struct bpf_tracing_link {
2500 struct bpf_link link;
2501 enum bpf_attach_type attach_type;
2502 struct bpf_trampoline *trampoline;
2503 struct bpf_prog *tgt_prog;
2506 static void bpf_tracing_link_release(struct bpf_link *link)
2508 struct bpf_tracing_link *tr_link =
2509 container_of(link, struct bpf_tracing_link, link);
2511 WARN_ON_ONCE(bpf_trampoline_unlink_prog(link->prog,
2512 tr_link->trampoline));
2514 bpf_trampoline_put(tr_link->trampoline);
2516 /* tgt_prog is NULL if target is a kernel function */
2517 if (tr_link->tgt_prog)
2518 bpf_prog_put(tr_link->tgt_prog);
2521 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2523 struct bpf_tracing_link *tr_link =
2524 container_of(link, struct bpf_tracing_link, link);
2529 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2530 struct seq_file *seq)
2532 struct bpf_tracing_link *tr_link =
2533 container_of(link, struct bpf_tracing_link, link);
2536 "attach_type:\t%d\n",
2537 tr_link->attach_type);
2540 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2541 struct bpf_link_info *info)
2543 struct bpf_tracing_link *tr_link =
2544 container_of(link, struct bpf_tracing_link, link);
2546 info->tracing.attach_type = tr_link->attach_type;
2551 static const struct bpf_link_ops bpf_tracing_link_lops = {
2552 .release = bpf_tracing_link_release,
2553 .dealloc = bpf_tracing_link_dealloc,
2554 .show_fdinfo = bpf_tracing_link_show_fdinfo,
2555 .fill_link_info = bpf_tracing_link_fill_link_info,
2558 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2562 struct bpf_link_primer link_primer;
2563 struct bpf_prog *tgt_prog = NULL;
2564 struct bpf_trampoline *tr = NULL;
2565 struct bpf_tracing_link *link;
2569 switch (prog->type) {
2570 case BPF_PROG_TYPE_TRACING:
2571 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2572 prog->expected_attach_type != BPF_TRACE_FEXIT &&
2573 prog->expected_attach_type != BPF_MODIFY_RETURN) {
2578 case BPF_PROG_TYPE_EXT:
2579 if (prog->expected_attach_type != 0) {
2584 case BPF_PROG_TYPE_LSM:
2585 if (prog->expected_attach_type != BPF_LSM_MAC) {
2595 if (!!tgt_prog_fd != !!btf_id) {
2601 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
2602 if (prog->type != BPF_PROG_TYPE_EXT) {
2607 tgt_prog = bpf_prog_get(tgt_prog_fd);
2608 if (IS_ERR(tgt_prog)) {
2609 err = PTR_ERR(tgt_prog);
2614 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
2617 link = kzalloc(sizeof(*link), GFP_USER);
2622 bpf_link_init(&link->link, BPF_LINK_TYPE_TRACING,
2623 &bpf_tracing_link_lops, prog);
2624 link->attach_type = prog->expected_attach_type;
2626 mutex_lock(&prog->aux->dst_mutex);
2628 /* There are a few possible cases here:
2630 * - if prog->aux->dst_trampoline is set, the program was just loaded
2631 * and not yet attached to anything, so we can use the values stored
2634 * - if prog->aux->dst_trampoline is NULL, the program has already been
2635 * attached to a target and its initial target was cleared (below)
2637 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
2638 * target_btf_id using the link_create API.
2640 * - if tgt_prog == NULL when this function was called using the old
2641 * raw_tracepoint_open API, and we need a target from prog->aux
2643 * The combination of no saved target in prog->aux, and no target
2644 * specified on load is illegal, and we reject that here.
2646 if (!prog->aux->dst_trampoline && !tgt_prog) {
2651 if (!prog->aux->dst_trampoline ||
2652 (key && key != prog->aux->dst_trampoline->key)) {
2653 /* If there is no saved target, or the specified target is
2654 * different from the destination specified at load time, we
2655 * need a new trampoline and a check for compatibility
2657 struct bpf_attach_target_info tgt_info = {};
2659 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
2664 tr = bpf_trampoline_get(key, &tgt_info);
2670 /* The caller didn't specify a target, or the target was the
2671 * same as the destination supplied during program load. This
2672 * means we can reuse the trampoline and reference from program
2673 * load time, and there is no need to allocate a new one. This
2674 * can only happen once for any program, as the saved values in
2675 * prog->aux are cleared below.
2677 tr = prog->aux->dst_trampoline;
2678 tgt_prog = prog->aux->dst_prog;
2681 err = bpf_link_prime(&link->link, &link_primer);
2685 err = bpf_trampoline_link_prog(prog, tr);
2687 bpf_link_cleanup(&link_primer);
2692 link->tgt_prog = tgt_prog;
2693 link->trampoline = tr;
2695 /* Always clear the trampoline and target prog from prog->aux to make
2696 * sure the original attach destination is not kept alive after a
2697 * program is (re-)attached to another target.
2699 if (prog->aux->dst_prog &&
2700 (tgt_prog_fd || tr != prog->aux->dst_trampoline))
2701 /* got extra prog ref from syscall, or attaching to different prog */
2702 bpf_prog_put(prog->aux->dst_prog);
2703 if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
2704 /* we allocated a new trampoline, so free the old one */
2705 bpf_trampoline_put(prog->aux->dst_trampoline);
2707 prog->aux->dst_prog = NULL;
2708 prog->aux->dst_trampoline = NULL;
2709 mutex_unlock(&prog->aux->dst_mutex);
2711 return bpf_link_settle(&link_primer);
2713 if (tr && tr != prog->aux->dst_trampoline)
2714 bpf_trampoline_put(tr);
2715 mutex_unlock(&prog->aux->dst_mutex);
2718 if (tgt_prog_fd && tgt_prog)
2719 bpf_prog_put(tgt_prog);
2723 struct bpf_raw_tp_link {
2724 struct bpf_link link;
2725 struct bpf_raw_event_map *btp;
2728 static void bpf_raw_tp_link_release(struct bpf_link *link)
2730 struct bpf_raw_tp_link *raw_tp =
2731 container_of(link, struct bpf_raw_tp_link, link);
2733 bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
2734 bpf_put_raw_tracepoint(raw_tp->btp);
2737 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
2739 struct bpf_raw_tp_link *raw_tp =
2740 container_of(link, struct bpf_raw_tp_link, link);
2745 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
2746 struct seq_file *seq)
2748 struct bpf_raw_tp_link *raw_tp_link =
2749 container_of(link, struct bpf_raw_tp_link, link);
2753 raw_tp_link->btp->tp->name);
2756 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
2757 struct bpf_link_info *info)
2759 struct bpf_raw_tp_link *raw_tp_link =
2760 container_of(link, struct bpf_raw_tp_link, link);
2761 char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
2762 const char *tp_name = raw_tp_link->btp->tp->name;
2763 u32 ulen = info->raw_tracepoint.tp_name_len;
2764 size_t tp_len = strlen(tp_name);
2769 info->raw_tracepoint.tp_name_len = tp_len + 1;
2774 if (ulen >= tp_len + 1) {
2775 if (copy_to_user(ubuf, tp_name, tp_len + 1))
2780 if (copy_to_user(ubuf, tp_name, ulen - 1))
2782 if (put_user(zero, ubuf + ulen - 1))
2790 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
2791 .release = bpf_raw_tp_link_release,
2792 .dealloc = bpf_raw_tp_link_dealloc,
2793 .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
2794 .fill_link_info = bpf_raw_tp_link_fill_link_info,
2797 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
2799 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
2801 struct bpf_link_primer link_primer;
2802 struct bpf_raw_tp_link *link;
2803 struct bpf_raw_event_map *btp;
2804 struct bpf_prog *prog;
2805 const char *tp_name;
2809 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
2812 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
2814 return PTR_ERR(prog);
2816 switch (prog->type) {
2817 case BPF_PROG_TYPE_TRACING:
2818 case BPF_PROG_TYPE_EXT:
2819 case BPF_PROG_TYPE_LSM:
2820 if (attr->raw_tracepoint.name) {
2821 /* The attach point for this category of programs
2822 * should be specified via btf_id during program load.
2827 if (prog->type == BPF_PROG_TYPE_TRACING &&
2828 prog->expected_attach_type == BPF_TRACE_RAW_TP) {
2829 tp_name = prog->aux->attach_func_name;
2832 err = bpf_tracing_prog_attach(prog, 0, 0);
2836 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2837 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2838 if (strncpy_from_user(buf,
2839 u64_to_user_ptr(attr->raw_tracepoint.name),
2840 sizeof(buf) - 1) < 0) {
2844 buf[sizeof(buf) - 1] = 0;
2852 btp = bpf_get_raw_tracepoint(tp_name);
2858 link = kzalloc(sizeof(*link), GFP_USER);
2863 bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
2864 &bpf_raw_tp_link_lops, prog);
2867 err = bpf_link_prime(&link->link, &link_primer);
2873 err = bpf_probe_register(link->btp, prog);
2875 bpf_link_cleanup(&link_primer);
2879 return bpf_link_settle(&link_primer);
2882 bpf_put_raw_tracepoint(btp);
2888 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2889 enum bpf_attach_type attach_type)
2891 switch (prog->type) {
2892 case BPF_PROG_TYPE_CGROUP_SOCK:
2893 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2894 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2895 case BPF_PROG_TYPE_SK_LOOKUP:
2896 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2897 case BPF_PROG_TYPE_CGROUP_SKB:
2898 if (!capable(CAP_NET_ADMIN))
2899 /* cg-skb progs can be loaded by unpriv user.
2900 * check permissions at attach time.
2903 return prog->enforce_expected_attach_type &&
2904 prog->expected_attach_type != attach_type ?
2911 static enum bpf_prog_type
2912 attach_type_to_prog_type(enum bpf_attach_type attach_type)
2914 switch (attach_type) {
2915 case BPF_CGROUP_INET_INGRESS:
2916 case BPF_CGROUP_INET_EGRESS:
2917 return BPF_PROG_TYPE_CGROUP_SKB;
2918 case BPF_CGROUP_INET_SOCK_CREATE:
2919 case BPF_CGROUP_INET_SOCK_RELEASE:
2920 case BPF_CGROUP_INET4_POST_BIND:
2921 case BPF_CGROUP_INET6_POST_BIND:
2922 return BPF_PROG_TYPE_CGROUP_SOCK;
2923 case BPF_CGROUP_INET4_BIND:
2924 case BPF_CGROUP_INET6_BIND:
2925 case BPF_CGROUP_INET4_CONNECT:
2926 case BPF_CGROUP_INET6_CONNECT:
2927 case BPF_CGROUP_INET4_GETPEERNAME:
2928 case BPF_CGROUP_INET6_GETPEERNAME:
2929 case BPF_CGROUP_INET4_GETSOCKNAME:
2930 case BPF_CGROUP_INET6_GETSOCKNAME:
2931 case BPF_CGROUP_UDP4_SENDMSG:
2932 case BPF_CGROUP_UDP6_SENDMSG:
2933 case BPF_CGROUP_UDP4_RECVMSG:
2934 case BPF_CGROUP_UDP6_RECVMSG:
2935 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2936 case BPF_CGROUP_SOCK_OPS:
2937 return BPF_PROG_TYPE_SOCK_OPS;
2938 case BPF_CGROUP_DEVICE:
2939 return BPF_PROG_TYPE_CGROUP_DEVICE;
2940 case BPF_SK_MSG_VERDICT:
2941 return BPF_PROG_TYPE_SK_MSG;
2942 case BPF_SK_SKB_STREAM_PARSER:
2943 case BPF_SK_SKB_STREAM_VERDICT:
2944 return BPF_PROG_TYPE_SK_SKB;
2945 case BPF_LIRC_MODE2:
2946 return BPF_PROG_TYPE_LIRC_MODE2;
2947 case BPF_FLOW_DISSECTOR:
2948 return BPF_PROG_TYPE_FLOW_DISSECTOR;
2949 case BPF_CGROUP_SYSCTL:
2950 return BPF_PROG_TYPE_CGROUP_SYSCTL;
2951 case BPF_CGROUP_GETSOCKOPT:
2952 case BPF_CGROUP_SETSOCKOPT:
2953 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
2954 case BPF_TRACE_ITER:
2955 return BPF_PROG_TYPE_TRACING;
2957 return BPF_PROG_TYPE_SK_LOOKUP;
2959 return BPF_PROG_TYPE_XDP;
2961 return BPF_PROG_TYPE_UNSPEC;
2965 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
2967 #define BPF_F_ATTACH_MASK \
2968 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
2970 static int bpf_prog_attach(const union bpf_attr *attr)
2972 enum bpf_prog_type ptype;
2973 struct bpf_prog *prog;
2976 if (CHECK_ATTR(BPF_PROG_ATTACH))
2979 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
2982 ptype = attach_type_to_prog_type(attr->attach_type);
2983 if (ptype == BPF_PROG_TYPE_UNSPEC)
2986 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
2988 return PTR_ERR(prog);
2990 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
2996 case BPF_PROG_TYPE_SK_SKB:
2997 case BPF_PROG_TYPE_SK_MSG:
2998 ret = sock_map_get_from_fd(attr, prog);
3000 case BPF_PROG_TYPE_LIRC_MODE2:
3001 ret = lirc_prog_attach(attr, prog);
3003 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3004 ret = netns_bpf_prog_attach(attr, prog);
3006 case BPF_PROG_TYPE_CGROUP_DEVICE:
3007 case BPF_PROG_TYPE_CGROUP_SKB:
3008 case BPF_PROG_TYPE_CGROUP_SOCK:
3009 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3010 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3011 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3012 case BPF_PROG_TYPE_SOCK_OPS:
3013 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3024 #define BPF_PROG_DETACH_LAST_FIELD attach_type
3026 static int bpf_prog_detach(const union bpf_attr *attr)
3028 enum bpf_prog_type ptype;
3030 if (CHECK_ATTR(BPF_PROG_DETACH))
3033 ptype = attach_type_to_prog_type(attr->attach_type);
3036 case BPF_PROG_TYPE_SK_MSG:
3037 case BPF_PROG_TYPE_SK_SKB:
3038 return sock_map_prog_detach(attr, ptype);
3039 case BPF_PROG_TYPE_LIRC_MODE2:
3040 return lirc_prog_detach(attr);
3041 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3042 return netns_bpf_prog_detach(attr, ptype);
3043 case BPF_PROG_TYPE_CGROUP_DEVICE:
3044 case BPF_PROG_TYPE_CGROUP_SKB:
3045 case BPF_PROG_TYPE_CGROUP_SOCK:
3046 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3047 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3048 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3049 case BPF_PROG_TYPE_SOCK_OPS:
3050 return cgroup_bpf_prog_detach(attr, ptype);
3056 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
3058 static int bpf_prog_query(const union bpf_attr *attr,
3059 union bpf_attr __user *uattr)
3061 if (!capable(CAP_NET_ADMIN))
3063 if (CHECK_ATTR(BPF_PROG_QUERY))
3065 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3068 switch (attr->query.attach_type) {
3069 case BPF_CGROUP_INET_INGRESS:
3070 case BPF_CGROUP_INET_EGRESS:
3071 case BPF_CGROUP_INET_SOCK_CREATE:
3072 case BPF_CGROUP_INET_SOCK_RELEASE:
3073 case BPF_CGROUP_INET4_BIND:
3074 case BPF_CGROUP_INET6_BIND:
3075 case BPF_CGROUP_INET4_POST_BIND:
3076 case BPF_CGROUP_INET6_POST_BIND:
3077 case BPF_CGROUP_INET4_CONNECT:
3078 case BPF_CGROUP_INET6_CONNECT:
3079 case BPF_CGROUP_INET4_GETPEERNAME:
3080 case BPF_CGROUP_INET6_GETPEERNAME:
3081 case BPF_CGROUP_INET4_GETSOCKNAME:
3082 case BPF_CGROUP_INET6_GETSOCKNAME:
3083 case BPF_CGROUP_UDP4_SENDMSG:
3084 case BPF_CGROUP_UDP6_SENDMSG:
3085 case BPF_CGROUP_UDP4_RECVMSG:
3086 case BPF_CGROUP_UDP6_RECVMSG:
3087 case BPF_CGROUP_SOCK_OPS:
3088 case BPF_CGROUP_DEVICE:
3089 case BPF_CGROUP_SYSCTL:
3090 case BPF_CGROUP_GETSOCKOPT:
3091 case BPF_CGROUP_SETSOCKOPT:
3092 return cgroup_bpf_prog_query(attr, uattr);
3093 case BPF_LIRC_MODE2:
3094 return lirc_prog_query(attr, uattr);
3095 case BPF_FLOW_DISSECTOR:
3097 return netns_bpf_prog_query(attr, uattr);
3103 #define BPF_PROG_TEST_RUN_LAST_FIELD test.cpu
3105 static int bpf_prog_test_run(const union bpf_attr *attr,
3106 union bpf_attr __user *uattr)
3108 struct bpf_prog *prog;
3109 int ret = -ENOTSUPP;
3111 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3114 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3115 (!attr->test.ctx_size_in && attr->test.ctx_in))
3118 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3119 (!attr->test.ctx_size_out && attr->test.ctx_out))
3122 prog = bpf_prog_get(attr->test.prog_fd);
3124 return PTR_ERR(prog);
3126 if (prog->aux->ops->test_run)
3127 ret = prog->aux->ops->test_run(prog, attr, uattr);
3133 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3135 static int bpf_obj_get_next_id(const union bpf_attr *attr,
3136 union bpf_attr __user *uattr,
3140 u32 next_id = attr->start_id;
3143 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3146 if (!capable(CAP_SYS_ADMIN))
3151 if (!idr_get_next(idr, &next_id))
3153 spin_unlock_bh(lock);
3156 err = put_user(next_id, &uattr->next_id);
3161 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3163 struct bpf_map *map;
3165 spin_lock_bh(&map_idr_lock);
3167 map = idr_get_next(&map_idr, id);
3169 map = __bpf_map_inc_not_zero(map, false);
3175 spin_unlock_bh(&map_idr_lock);
3180 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3182 struct bpf_prog *prog;
3184 spin_lock_bh(&prog_idr_lock);
3186 prog = idr_get_next(&prog_idr, id);
3188 prog = bpf_prog_inc_not_zero(prog);
3194 spin_unlock_bh(&prog_idr_lock);
3199 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3201 struct bpf_prog *bpf_prog_by_id(u32 id)
3203 struct bpf_prog *prog;
3206 return ERR_PTR(-ENOENT);
3208 spin_lock_bh(&prog_idr_lock);
3209 prog = idr_find(&prog_idr, id);
3211 prog = bpf_prog_inc_not_zero(prog);
3213 prog = ERR_PTR(-ENOENT);
3214 spin_unlock_bh(&prog_idr_lock);
3218 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3220 struct bpf_prog *prog;
3221 u32 id = attr->prog_id;
3224 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3227 if (!capable(CAP_SYS_ADMIN))
3230 prog = bpf_prog_by_id(id);
3232 return PTR_ERR(prog);
3234 fd = bpf_prog_new_fd(prog);
3241 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3243 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3245 struct bpf_map *map;
3246 u32 id = attr->map_id;
3250 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3251 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3254 if (!capable(CAP_SYS_ADMIN))
3257 f_flags = bpf_get_file_flag(attr->open_flags);
3261 spin_lock_bh(&map_idr_lock);
3262 map = idr_find(&map_idr, id);
3264 map = __bpf_map_inc_not_zero(map, true);
3266 map = ERR_PTR(-ENOENT);
3267 spin_unlock_bh(&map_idr_lock);
3270 return PTR_ERR(map);
3272 fd = bpf_map_new_fd(map, f_flags);
3274 bpf_map_put_with_uref(map);
3279 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3280 unsigned long addr, u32 *off,
3283 const struct bpf_map *map;
3286 mutex_lock(&prog->aux->used_maps_mutex);
3287 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3288 map = prog->aux->used_maps[i];
3289 if (map == (void *)addr) {
3290 *type = BPF_PSEUDO_MAP_FD;
3293 if (!map->ops->map_direct_value_meta)
3295 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3296 *type = BPF_PSEUDO_MAP_VALUE;
3303 mutex_unlock(&prog->aux->used_maps_mutex);
3307 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3308 const struct cred *f_cred)
3310 const struct bpf_map *map;
3311 struct bpf_insn *insns;
3317 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3322 for (i = 0; i < prog->len; i++) {
3323 code = insns[i].code;
3325 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3326 insns[i].code = BPF_JMP | BPF_CALL;
3327 insns[i].imm = BPF_FUNC_tail_call;
3330 if (code == (BPF_JMP | BPF_CALL) ||
3331 code == (BPF_JMP | BPF_CALL_ARGS)) {
3332 if (code == (BPF_JMP | BPF_CALL_ARGS))
3333 insns[i].code = BPF_JMP | BPF_CALL;
3334 if (!bpf_dump_raw_ok(f_cred))
3338 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3339 insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3343 if (code != (BPF_LD | BPF_IMM | BPF_DW))
3346 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3347 map = bpf_map_from_imm(prog, imm, &off, &type);
3349 insns[i].src_reg = type;
3350 insns[i].imm = map->id;
3351 insns[i + 1].imm = off;
3359 static int set_info_rec_size(struct bpf_prog_info *info)
3362 * Ensure info.*_rec_size is the same as kernel expected size
3366 * Only allow zero *_rec_size if both _rec_size and _cnt are
3367 * zero. In this case, the kernel will set the expected
3368 * _rec_size back to the info.
3371 if ((info->nr_func_info || info->func_info_rec_size) &&
3372 info->func_info_rec_size != sizeof(struct bpf_func_info))
3375 if ((info->nr_line_info || info->line_info_rec_size) &&
3376 info->line_info_rec_size != sizeof(struct bpf_line_info))
3379 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3380 info->jited_line_info_rec_size != sizeof(__u64))
3383 info->func_info_rec_size = sizeof(struct bpf_func_info);
3384 info->line_info_rec_size = sizeof(struct bpf_line_info);
3385 info->jited_line_info_rec_size = sizeof(__u64);
3390 static int bpf_prog_get_info_by_fd(struct file *file,
3391 struct bpf_prog *prog,
3392 const union bpf_attr *attr,
3393 union bpf_attr __user *uattr)
3395 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3396 struct bpf_prog_info info;
3397 u32 info_len = attr->info.info_len;
3398 struct bpf_prog_stats stats;
3399 char __user *uinsns;
3403 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3406 info_len = min_t(u32, sizeof(info), info_len);
3408 memset(&info, 0, sizeof(info));
3409 if (copy_from_user(&info, uinfo, info_len))
3412 info.type = prog->type;
3413 info.id = prog->aux->id;
3414 info.load_time = prog->aux->load_time;
3415 info.created_by_uid = from_kuid_munged(current_user_ns(),
3416 prog->aux->user->uid);
3417 info.gpl_compatible = prog->gpl_compatible;
3419 memcpy(info.tag, prog->tag, sizeof(prog->tag));
3420 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3422 mutex_lock(&prog->aux->used_maps_mutex);
3423 ulen = info.nr_map_ids;
3424 info.nr_map_ids = prog->aux->used_map_cnt;
3425 ulen = min_t(u32, info.nr_map_ids, ulen);
3427 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3430 for (i = 0; i < ulen; i++)
3431 if (put_user(prog->aux->used_maps[i]->id,
3432 &user_map_ids[i])) {
3433 mutex_unlock(&prog->aux->used_maps_mutex);
3437 mutex_unlock(&prog->aux->used_maps_mutex);
3439 err = set_info_rec_size(&info);
3443 bpf_prog_get_stats(prog, &stats);
3444 info.run_time_ns = stats.nsecs;
3445 info.run_cnt = stats.cnt;
3446 info.recursion_misses = stats.misses;
3448 if (!bpf_capable()) {
3449 info.jited_prog_len = 0;
3450 info.xlated_prog_len = 0;
3451 info.nr_jited_ksyms = 0;
3452 info.nr_jited_func_lens = 0;
3453 info.nr_func_info = 0;
3454 info.nr_line_info = 0;
3455 info.nr_jited_line_info = 0;
3459 ulen = info.xlated_prog_len;
3460 info.xlated_prog_len = bpf_prog_insn_size(prog);
3461 if (info.xlated_prog_len && ulen) {
3462 struct bpf_insn *insns_sanitized;
3465 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
3466 info.xlated_prog_insns = 0;
3469 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
3470 if (!insns_sanitized)
3472 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3473 ulen = min_t(u32, info.xlated_prog_len, ulen);
3474 fault = copy_to_user(uinsns, insns_sanitized, ulen);
3475 kfree(insns_sanitized);
3480 if (bpf_prog_is_dev_bound(prog->aux)) {
3481 err = bpf_prog_offload_info_fill(&info, prog);
3487 /* NOTE: the following code is supposed to be skipped for offload.
3488 * bpf_prog_offload_info_fill() is the place to fill similar fields
3491 ulen = info.jited_prog_len;
3492 if (prog->aux->func_cnt) {
3495 info.jited_prog_len = 0;
3496 for (i = 0; i < prog->aux->func_cnt; i++)
3497 info.jited_prog_len += prog->aux->func[i]->jited_len;
3499 info.jited_prog_len = prog->jited_len;
3502 if (info.jited_prog_len && ulen) {
3503 if (bpf_dump_raw_ok(file->f_cred)) {
3504 uinsns = u64_to_user_ptr(info.jited_prog_insns);
3505 ulen = min_t(u32, info.jited_prog_len, ulen);
3507 /* for multi-function programs, copy the JITed
3508 * instructions for all the functions
3510 if (prog->aux->func_cnt) {
3515 for (i = 0; i < prog->aux->func_cnt; i++) {
3516 len = prog->aux->func[i]->jited_len;
3517 len = min_t(u32, len, free);
3518 img = (u8 *) prog->aux->func[i]->bpf_func;
3519 if (copy_to_user(uinsns, img, len))
3527 if (copy_to_user(uinsns, prog->bpf_func, ulen))
3531 info.jited_prog_insns = 0;
3535 ulen = info.nr_jited_ksyms;
3536 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
3538 if (bpf_dump_raw_ok(file->f_cred)) {
3539 unsigned long ksym_addr;
3540 u64 __user *user_ksyms;
3543 /* copy the address of the kernel symbol
3544 * corresponding to each function
3546 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
3547 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
3548 if (prog->aux->func_cnt) {
3549 for (i = 0; i < ulen; i++) {
3550 ksym_addr = (unsigned long)
3551 prog->aux->func[i]->bpf_func;
3552 if (put_user((u64) ksym_addr,
3557 ksym_addr = (unsigned long) prog->bpf_func;
3558 if (put_user((u64) ksym_addr, &user_ksyms[0]))
3562 info.jited_ksyms = 0;
3566 ulen = info.nr_jited_func_lens;
3567 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
3569 if (bpf_dump_raw_ok(file->f_cred)) {
3570 u32 __user *user_lens;
3573 /* copy the JITed image lengths for each function */
3574 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
3575 user_lens = u64_to_user_ptr(info.jited_func_lens);
3576 if (prog->aux->func_cnt) {
3577 for (i = 0; i < ulen; i++) {
3579 prog->aux->func[i]->jited_len;
3580 if (put_user(func_len, &user_lens[i]))
3584 func_len = prog->jited_len;
3585 if (put_user(func_len, &user_lens[0]))
3589 info.jited_func_lens = 0;
3594 info.btf_id = btf_obj_id(prog->aux->btf);
3596 ulen = info.nr_func_info;
3597 info.nr_func_info = prog->aux->func_info_cnt;
3598 if (info.nr_func_info && ulen) {
3599 char __user *user_finfo;
3601 user_finfo = u64_to_user_ptr(info.func_info);
3602 ulen = min_t(u32, info.nr_func_info, ulen);
3603 if (copy_to_user(user_finfo, prog->aux->func_info,
3604 info.func_info_rec_size * ulen))
3608 ulen = info.nr_line_info;
3609 info.nr_line_info = prog->aux->nr_linfo;
3610 if (info.nr_line_info && ulen) {
3611 __u8 __user *user_linfo;
3613 user_linfo = u64_to_user_ptr(info.line_info);
3614 ulen = min_t(u32, info.nr_line_info, ulen);
3615 if (copy_to_user(user_linfo, prog->aux->linfo,
3616 info.line_info_rec_size * ulen))
3620 ulen = info.nr_jited_line_info;
3621 if (prog->aux->jited_linfo)
3622 info.nr_jited_line_info = prog->aux->nr_linfo;
3624 info.nr_jited_line_info = 0;
3625 if (info.nr_jited_line_info && ulen) {
3626 if (bpf_dump_raw_ok(file->f_cred)) {
3627 __u64 __user *user_linfo;
3630 user_linfo = u64_to_user_ptr(info.jited_line_info);
3631 ulen = min_t(u32, info.nr_jited_line_info, ulen);
3632 for (i = 0; i < ulen; i++) {
3633 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
3638 info.jited_line_info = 0;
3642 ulen = info.nr_prog_tags;
3643 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
3645 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
3648 user_prog_tags = u64_to_user_ptr(info.prog_tags);
3649 ulen = min_t(u32, info.nr_prog_tags, ulen);
3650 if (prog->aux->func_cnt) {
3651 for (i = 0; i < ulen; i++) {
3652 if (copy_to_user(user_prog_tags[i],
3653 prog->aux->func[i]->tag,
3658 if (copy_to_user(user_prog_tags[0],
3659 prog->tag, BPF_TAG_SIZE))
3665 if (copy_to_user(uinfo, &info, info_len) ||
3666 put_user(info_len, &uattr->info.info_len))
3672 static int bpf_map_get_info_by_fd(struct file *file,
3673 struct bpf_map *map,
3674 const union bpf_attr *attr,
3675 union bpf_attr __user *uattr)
3677 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3678 struct bpf_map_info info;
3679 u32 info_len = attr->info.info_len;
3682 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3685 info_len = min_t(u32, sizeof(info), info_len);
3687 memset(&info, 0, sizeof(info));
3688 info.type = map->map_type;
3690 info.key_size = map->key_size;
3691 info.value_size = map->value_size;
3692 info.max_entries = map->max_entries;
3693 info.map_flags = map->map_flags;
3694 memcpy(info.name, map->name, sizeof(map->name));
3697 info.btf_id = btf_obj_id(map->btf);
3698 info.btf_key_type_id = map->btf_key_type_id;
3699 info.btf_value_type_id = map->btf_value_type_id;
3701 info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
3703 if (bpf_map_is_dev_bound(map)) {
3704 err = bpf_map_offload_info_fill(&info, map);
3709 if (copy_to_user(uinfo, &info, info_len) ||
3710 put_user(info_len, &uattr->info.info_len))
3716 static int bpf_btf_get_info_by_fd(struct file *file,
3718 const union bpf_attr *attr,
3719 union bpf_attr __user *uattr)
3721 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3722 u32 info_len = attr->info.info_len;
3725 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
3729 return btf_get_info_by_fd(btf, attr, uattr);
3732 static int bpf_link_get_info_by_fd(struct file *file,
3733 struct bpf_link *link,
3734 const union bpf_attr *attr,
3735 union bpf_attr __user *uattr)
3737 struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3738 struct bpf_link_info info;
3739 u32 info_len = attr->info.info_len;
3742 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3745 info_len = min_t(u32, sizeof(info), info_len);
3747 memset(&info, 0, sizeof(info));
3748 if (copy_from_user(&info, uinfo, info_len))
3751 info.type = link->type;
3753 info.prog_id = link->prog->aux->id;
3755 if (link->ops->fill_link_info) {
3756 err = link->ops->fill_link_info(link, &info);
3761 if (copy_to_user(uinfo, &info, info_len) ||
3762 put_user(info_len, &uattr->info.info_len))
3769 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
3771 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
3772 union bpf_attr __user *uattr)
3774 int ufd = attr->info.bpf_fd;
3778 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
3785 if (f.file->f_op == &bpf_prog_fops)
3786 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
3788 else if (f.file->f_op == &bpf_map_fops)
3789 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
3791 else if (f.file->f_op == &btf_fops)
3792 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
3793 else if (f.file->f_op == &bpf_link_fops)
3794 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
3803 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
3805 static int bpf_btf_load(const union bpf_attr *attr)
3807 if (CHECK_ATTR(BPF_BTF_LOAD))
3813 return btf_new_fd(attr);
3816 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
3818 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
3820 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
3823 if (!capable(CAP_SYS_ADMIN))
3826 return btf_get_fd_by_id(attr->btf_id);
3829 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
3830 union bpf_attr __user *uattr,
3831 u32 prog_id, u32 fd_type,
3832 const char *buf, u64 probe_offset,
3835 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
3836 u32 len = buf ? strlen(buf) : 0, input_len;
3839 if (put_user(len, &uattr->task_fd_query.buf_len))
3841 input_len = attr->task_fd_query.buf_len;
3842 if (input_len && ubuf) {
3844 /* nothing to copy, just make ubuf NULL terminated */
3847 if (put_user(zero, ubuf))
3849 } else if (input_len >= len + 1) {
3850 /* ubuf can hold the string with NULL terminator */
3851 if (copy_to_user(ubuf, buf, len + 1))
3854 /* ubuf cannot hold the string with NULL terminator,
3855 * do a partial copy with NULL terminator.
3860 if (copy_to_user(ubuf, buf, input_len - 1))
3862 if (put_user(zero, ubuf + input_len - 1))
3867 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
3868 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
3869 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
3870 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
3876 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
3878 static int bpf_task_fd_query(const union bpf_attr *attr,
3879 union bpf_attr __user *uattr)
3881 pid_t pid = attr->task_fd_query.pid;
3882 u32 fd = attr->task_fd_query.fd;
3883 const struct perf_event *event;
3884 struct task_struct *task;
3888 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
3891 if (!capable(CAP_SYS_ADMIN))
3894 if (attr->task_fd_query.flags != 0)
3897 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3902 file = fget_task(task, fd);
3903 put_task_struct(task);
3907 if (file->f_op == &bpf_link_fops) {
3908 struct bpf_link *link = file->private_data;
3910 if (link->ops == &bpf_raw_tp_link_lops) {
3911 struct bpf_raw_tp_link *raw_tp =
3912 container_of(link, struct bpf_raw_tp_link, link);
3913 struct bpf_raw_event_map *btp = raw_tp->btp;
3915 err = bpf_task_fd_query_copy(attr, uattr,
3916 raw_tp->link.prog->aux->id,
3917 BPF_FD_TYPE_RAW_TRACEPOINT,
3918 btp->tp->name, 0, 0);
3924 event = perf_get_event(file);
3925 if (!IS_ERR(event)) {
3926 u64 probe_offset, probe_addr;
3927 u32 prog_id, fd_type;
3930 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
3931 &buf, &probe_offset,
3934 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
3948 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
3950 #define BPF_DO_BATCH(fn) \
3956 err = fn(map, attr, uattr); \
3959 static int bpf_map_do_batch(const union bpf_attr *attr,
3960 union bpf_attr __user *uattr,
3963 struct bpf_map *map;
3967 if (CHECK_ATTR(BPF_MAP_BATCH))
3970 ufd = attr->batch.map_fd;
3972 map = __bpf_map_get(f);
3974 return PTR_ERR(map);
3976 if ((cmd == BPF_MAP_LOOKUP_BATCH ||
3977 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH) &&
3978 !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
3983 if (cmd != BPF_MAP_LOOKUP_BATCH &&
3984 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
3989 if (cmd == BPF_MAP_LOOKUP_BATCH)
3990 BPF_DO_BATCH(map->ops->map_lookup_batch);
3991 else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
3992 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
3993 else if (cmd == BPF_MAP_UPDATE_BATCH)
3994 BPF_DO_BATCH(map->ops->map_update_batch);
3996 BPF_DO_BATCH(map->ops->map_delete_batch);
4003 static int tracing_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
4005 if (attr->link_create.attach_type != prog->expected_attach_type)
4008 if (prog->expected_attach_type == BPF_TRACE_ITER)
4009 return bpf_iter_link_attach(attr, prog);
4010 else if (prog->type == BPF_PROG_TYPE_EXT)
4011 return bpf_tracing_prog_attach(prog,
4012 attr->link_create.target_fd,
4013 attr->link_create.target_btf_id);
4017 #define BPF_LINK_CREATE_LAST_FIELD link_create.iter_info_len
4018 static int link_create(union bpf_attr *attr)
4020 enum bpf_prog_type ptype;
4021 struct bpf_prog *prog;
4024 if (CHECK_ATTR(BPF_LINK_CREATE))
4027 prog = bpf_prog_get(attr->link_create.prog_fd);
4029 return PTR_ERR(prog);
4031 ret = bpf_prog_attach_check_attach_type(prog,
4032 attr->link_create.attach_type);
4036 if (prog->type == BPF_PROG_TYPE_EXT) {
4037 ret = tracing_bpf_link_attach(attr, prog);
4041 ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4042 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4048 case BPF_PROG_TYPE_CGROUP_SKB:
4049 case BPF_PROG_TYPE_CGROUP_SOCK:
4050 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4051 case BPF_PROG_TYPE_SOCK_OPS:
4052 case BPF_PROG_TYPE_CGROUP_DEVICE:
4053 case BPF_PROG_TYPE_CGROUP_SYSCTL:
4054 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4055 ret = cgroup_bpf_link_attach(attr, prog);
4057 case BPF_PROG_TYPE_TRACING:
4058 ret = tracing_bpf_link_attach(attr, prog);
4060 case BPF_PROG_TYPE_FLOW_DISSECTOR:
4061 case BPF_PROG_TYPE_SK_LOOKUP:
4062 ret = netns_bpf_link_create(attr, prog);
4065 case BPF_PROG_TYPE_XDP:
4066 ret = bpf_xdp_link_attach(attr, prog);
4079 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4081 static int link_update(union bpf_attr *attr)
4083 struct bpf_prog *old_prog = NULL, *new_prog;
4084 struct bpf_link *link;
4088 if (CHECK_ATTR(BPF_LINK_UPDATE))
4091 flags = attr->link_update.flags;
4092 if (flags & ~BPF_F_REPLACE)
4095 link = bpf_link_get_from_fd(attr->link_update.link_fd);
4097 return PTR_ERR(link);
4099 new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4100 if (IS_ERR(new_prog)) {
4101 ret = PTR_ERR(new_prog);
4105 if (flags & BPF_F_REPLACE) {
4106 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4107 if (IS_ERR(old_prog)) {
4108 ret = PTR_ERR(old_prog);
4112 } else if (attr->link_update.old_prog_fd) {
4117 if (link->ops->update_prog)
4118 ret = link->ops->update_prog(link, new_prog, old_prog);
4124 bpf_prog_put(old_prog);
4126 bpf_prog_put(new_prog);
4132 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4134 static int link_detach(union bpf_attr *attr)
4136 struct bpf_link *link;
4139 if (CHECK_ATTR(BPF_LINK_DETACH))
4142 link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4144 return PTR_ERR(link);
4146 if (link->ops->detach)
4147 ret = link->ops->detach(link);
4155 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4157 return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4160 struct bpf_link *bpf_link_by_id(u32 id)
4162 struct bpf_link *link;
4165 return ERR_PTR(-ENOENT);
4167 spin_lock_bh(&link_idr_lock);
4168 /* before link is "settled", ID is 0, pretend it doesn't exist yet */
4169 link = idr_find(&link_idr, id);
4172 link = bpf_link_inc_not_zero(link);
4174 link = ERR_PTR(-EAGAIN);
4176 link = ERR_PTR(-ENOENT);
4178 spin_unlock_bh(&link_idr_lock);
4182 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4184 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4186 struct bpf_link *link;
4187 u32 id = attr->link_id;
4190 if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4193 if (!capable(CAP_SYS_ADMIN))
4196 link = bpf_link_by_id(id);
4198 return PTR_ERR(link);
4200 fd = bpf_link_new_fd(link);
4207 DEFINE_MUTEX(bpf_stats_enabled_mutex);
4209 static int bpf_stats_release(struct inode *inode, struct file *file)
4211 mutex_lock(&bpf_stats_enabled_mutex);
4212 static_key_slow_dec(&bpf_stats_enabled_key.key);
4213 mutex_unlock(&bpf_stats_enabled_mutex);
4217 static const struct file_operations bpf_stats_fops = {
4218 .release = bpf_stats_release,
4221 static int bpf_enable_runtime_stats(void)
4225 mutex_lock(&bpf_stats_enabled_mutex);
4227 /* Set a very high limit to avoid overflow */
4228 if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4229 mutex_unlock(&bpf_stats_enabled_mutex);
4233 fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4235 static_key_slow_inc(&bpf_stats_enabled_key.key);
4237 mutex_unlock(&bpf_stats_enabled_mutex);
4241 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4243 static int bpf_enable_stats(union bpf_attr *attr)
4246 if (CHECK_ATTR(BPF_ENABLE_STATS))
4249 if (!capable(CAP_SYS_ADMIN))
4252 switch (attr->enable_stats.type) {
4253 case BPF_STATS_RUN_TIME:
4254 return bpf_enable_runtime_stats();
4261 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4263 static int bpf_iter_create(union bpf_attr *attr)
4265 struct bpf_link *link;
4268 if (CHECK_ATTR(BPF_ITER_CREATE))
4271 if (attr->iter_create.flags)
4274 link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4276 return PTR_ERR(link);
4278 err = bpf_iter_new_fd(link);
4284 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4286 static int bpf_prog_bind_map(union bpf_attr *attr)
4288 struct bpf_prog *prog;
4289 struct bpf_map *map;
4290 struct bpf_map **used_maps_old, **used_maps_new;
4293 if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4296 if (attr->prog_bind_map.flags)
4299 prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4301 return PTR_ERR(prog);
4303 map = bpf_map_get(attr->prog_bind_map.map_fd);
4309 mutex_lock(&prog->aux->used_maps_mutex);
4311 used_maps_old = prog->aux->used_maps;
4313 for (i = 0; i < prog->aux->used_map_cnt; i++)
4314 if (used_maps_old[i] == map) {
4319 used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4320 sizeof(used_maps_new[0]),
4322 if (!used_maps_new) {
4327 memcpy(used_maps_new, used_maps_old,
4328 sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4329 used_maps_new[prog->aux->used_map_cnt] = map;
4331 prog->aux->used_map_cnt++;
4332 prog->aux->used_maps = used_maps_new;
4334 kfree(used_maps_old);
4337 mutex_unlock(&prog->aux->used_maps_mutex);
4346 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
4348 union bpf_attr attr;
4351 if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
4354 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
4357 size = min_t(u32, size, sizeof(attr));
4359 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
4360 memset(&attr, 0, sizeof(attr));
4361 if (copy_from_user(&attr, uattr, size) != 0)
4364 err = security_bpf(cmd, &attr, size);
4369 case BPF_MAP_CREATE:
4370 err = map_create(&attr);
4372 case BPF_MAP_LOOKUP_ELEM:
4373 err = map_lookup_elem(&attr);
4375 case BPF_MAP_UPDATE_ELEM:
4376 err = map_update_elem(&attr);
4378 case BPF_MAP_DELETE_ELEM:
4379 err = map_delete_elem(&attr);
4381 case BPF_MAP_GET_NEXT_KEY:
4382 err = map_get_next_key(&attr);
4384 case BPF_MAP_FREEZE:
4385 err = map_freeze(&attr);
4388 err = bpf_prog_load(&attr, uattr);
4391 err = bpf_obj_pin(&attr);
4394 err = bpf_obj_get(&attr);
4396 case BPF_PROG_ATTACH:
4397 err = bpf_prog_attach(&attr);
4399 case BPF_PROG_DETACH:
4400 err = bpf_prog_detach(&attr);
4402 case BPF_PROG_QUERY:
4403 err = bpf_prog_query(&attr, uattr);
4405 case BPF_PROG_TEST_RUN:
4406 err = bpf_prog_test_run(&attr, uattr);
4408 case BPF_PROG_GET_NEXT_ID:
4409 err = bpf_obj_get_next_id(&attr, uattr,
4410 &prog_idr, &prog_idr_lock);
4412 case BPF_MAP_GET_NEXT_ID:
4413 err = bpf_obj_get_next_id(&attr, uattr,
4414 &map_idr, &map_idr_lock);
4416 case BPF_BTF_GET_NEXT_ID:
4417 err = bpf_obj_get_next_id(&attr, uattr,
4418 &btf_idr, &btf_idr_lock);
4420 case BPF_PROG_GET_FD_BY_ID:
4421 err = bpf_prog_get_fd_by_id(&attr);
4423 case BPF_MAP_GET_FD_BY_ID:
4424 err = bpf_map_get_fd_by_id(&attr);
4426 case BPF_OBJ_GET_INFO_BY_FD:
4427 err = bpf_obj_get_info_by_fd(&attr, uattr);
4429 case BPF_RAW_TRACEPOINT_OPEN:
4430 err = bpf_raw_tracepoint_open(&attr);
4433 err = bpf_btf_load(&attr);
4435 case BPF_BTF_GET_FD_BY_ID:
4436 err = bpf_btf_get_fd_by_id(&attr);
4438 case BPF_TASK_FD_QUERY:
4439 err = bpf_task_fd_query(&attr, uattr);
4441 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
4442 err = map_lookup_and_delete_elem(&attr);
4444 case BPF_MAP_LOOKUP_BATCH:
4445 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_LOOKUP_BATCH);
4447 case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
4448 err = bpf_map_do_batch(&attr, uattr,
4449 BPF_MAP_LOOKUP_AND_DELETE_BATCH);
4451 case BPF_MAP_UPDATE_BATCH:
4452 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_UPDATE_BATCH);
4454 case BPF_MAP_DELETE_BATCH:
4455 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_DELETE_BATCH);
4457 case BPF_LINK_CREATE:
4458 err = link_create(&attr);
4460 case BPF_LINK_UPDATE:
4461 err = link_update(&attr);
4463 case BPF_LINK_GET_FD_BY_ID:
4464 err = bpf_link_get_fd_by_id(&attr);
4466 case BPF_LINK_GET_NEXT_ID:
4467 err = bpf_obj_get_next_id(&attr, uattr,
4468 &link_idr, &link_idr_lock);
4470 case BPF_ENABLE_STATS:
4471 err = bpf_enable_stats(&attr);
4473 case BPF_ITER_CREATE:
4474 err = bpf_iter_create(&attr);
4476 case BPF_LINK_DETACH:
4477 err = link_detach(&attr);
4479 case BPF_PROG_BIND_MAP:
4480 err = bpf_prog_bind_map(&attr);