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);
857 if (btf_is_kernel(btf)) {
864 if (attr->btf_value_type_id) {
865 err = map_check_btf(map, btf, attr->btf_key_type_id,
866 attr->btf_value_type_id);
871 map->btf_key_type_id = attr->btf_key_type_id;
872 map->btf_value_type_id = attr->btf_value_type_id;
873 map->btf_vmlinux_value_type_id =
874 attr->btf_vmlinux_value_type_id;
877 err = security_bpf_map_alloc(map);
881 err = bpf_map_alloc_id(map);
885 bpf_map_save_memcg(map);
887 err = bpf_map_new_fd(map, f_flags);
889 /* failed to allocate fd.
890 * bpf_map_put_with_uref() is needed because the above
891 * bpf_map_alloc_id() has published the map
892 * to the userspace and the userspace may
893 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
895 bpf_map_put_with_uref(map);
902 security_bpf_map_free(map);
905 map->ops->map_free(map);
909 /* if error is returned, fd is released.
910 * On success caller should complete fd access with matching fdput()
912 struct bpf_map *__bpf_map_get(struct fd f)
915 return ERR_PTR(-EBADF);
916 if (f.file->f_op != &bpf_map_fops) {
918 return ERR_PTR(-EINVAL);
921 return f.file->private_data;
924 void bpf_map_inc(struct bpf_map *map)
926 atomic64_inc(&map->refcnt);
928 EXPORT_SYMBOL_GPL(bpf_map_inc);
930 void bpf_map_inc_with_uref(struct bpf_map *map)
932 atomic64_inc(&map->refcnt);
933 atomic64_inc(&map->usercnt);
935 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
937 struct bpf_map *bpf_map_get(u32 ufd)
939 struct fd f = fdget(ufd);
942 map = __bpf_map_get(f);
952 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
954 struct fd f = fdget(ufd);
957 map = __bpf_map_get(f);
961 bpf_map_inc_with_uref(map);
967 /* map_idr_lock should have been held */
968 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
972 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
974 return ERR_PTR(-ENOENT);
976 atomic64_inc(&map->usercnt);
981 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
983 spin_lock_bh(&map_idr_lock);
984 map = __bpf_map_inc_not_zero(map, false);
985 spin_unlock_bh(&map_idr_lock);
989 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
991 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
996 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
999 return memdup_user(ukey, key_size);
1002 return ERR_PTR(-EINVAL);
1007 /* last field in 'union bpf_attr' used by this command */
1008 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1010 static int map_lookup_elem(union bpf_attr *attr)
1012 void __user *ukey = u64_to_user_ptr(attr->key);
1013 void __user *uvalue = u64_to_user_ptr(attr->value);
1014 int ufd = attr->map_fd;
1015 struct bpf_map *map;
1021 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1024 if (attr->flags & ~BPF_F_LOCK)
1028 map = __bpf_map_get(f);
1030 return PTR_ERR(map);
1031 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1036 if ((attr->flags & BPF_F_LOCK) &&
1037 !map_value_has_spin_lock(map)) {
1042 key = __bpf_copy_key(ukey, map->key_size);
1048 value_size = bpf_map_value_size(map);
1051 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1055 err = bpf_map_copy_value(map, key, value, attr->flags);
1060 if (copy_to_user(uvalue, value, value_size) != 0)
1075 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1077 static int map_update_elem(union bpf_attr *attr)
1079 void __user *ukey = u64_to_user_ptr(attr->key);
1080 void __user *uvalue = u64_to_user_ptr(attr->value);
1081 int ufd = attr->map_fd;
1082 struct bpf_map *map;
1088 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1092 map = __bpf_map_get(f);
1094 return PTR_ERR(map);
1095 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1100 if ((attr->flags & BPF_F_LOCK) &&
1101 !map_value_has_spin_lock(map)) {
1106 key = __bpf_copy_key(ukey, map->key_size);
1112 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1113 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
1114 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
1115 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
1116 value_size = round_up(map->value_size, 8) * num_possible_cpus();
1118 value_size = map->value_size;
1121 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1126 if (copy_from_user(value, uvalue, value_size) != 0)
1129 err = bpf_map_update_value(map, f, key, value, attr->flags);
1140 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1142 static int map_delete_elem(union bpf_attr *attr)
1144 void __user *ukey = u64_to_user_ptr(attr->key);
1145 int ufd = attr->map_fd;
1146 struct bpf_map *map;
1151 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1155 map = __bpf_map_get(f);
1157 return PTR_ERR(map);
1158 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1163 key = __bpf_copy_key(ukey, map->key_size);
1169 if (bpf_map_is_dev_bound(map)) {
1170 err = bpf_map_offload_delete_elem(map, key);
1172 } else if (IS_FD_PROG_ARRAY(map) ||
1173 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1174 /* These maps require sleepable context */
1175 err = map->ops->map_delete_elem(map, key);
1179 bpf_disable_instrumentation();
1181 err = map->ops->map_delete_elem(map, key);
1183 bpf_enable_instrumentation();
1184 maybe_wait_bpf_programs(map);
1192 /* last field in 'union bpf_attr' used by this command */
1193 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1195 static int map_get_next_key(union bpf_attr *attr)
1197 void __user *ukey = u64_to_user_ptr(attr->key);
1198 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1199 int ufd = attr->map_fd;
1200 struct bpf_map *map;
1201 void *key, *next_key;
1205 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1209 map = __bpf_map_get(f);
1211 return PTR_ERR(map);
1212 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1218 key = __bpf_copy_key(ukey, map->key_size);
1228 next_key = kmalloc(map->key_size, GFP_USER);
1232 if (bpf_map_is_dev_bound(map)) {
1233 err = bpf_map_offload_get_next_key(map, key, next_key);
1238 err = map->ops->map_get_next_key(map, key, next_key);
1245 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1259 int generic_map_delete_batch(struct bpf_map *map,
1260 const union bpf_attr *attr,
1261 union bpf_attr __user *uattr)
1263 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1268 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1271 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1272 !map_value_has_spin_lock(map)) {
1276 max_count = attr->batch.count;
1280 key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1284 for (cp = 0; cp < max_count; cp++) {
1286 if (copy_from_user(key, keys + cp * map->key_size,
1290 if (bpf_map_is_dev_bound(map)) {
1291 err = bpf_map_offload_delete_elem(map, key);
1295 bpf_disable_instrumentation();
1297 err = map->ops->map_delete_elem(map, key);
1299 bpf_enable_instrumentation();
1300 maybe_wait_bpf_programs(map);
1304 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1311 int generic_map_update_batch(struct bpf_map *map,
1312 const union bpf_attr *attr,
1313 union bpf_attr __user *uattr)
1315 void __user *values = u64_to_user_ptr(attr->batch.values);
1316 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1317 u32 value_size, cp, max_count;
1318 int ufd = attr->map_fd;
1324 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1327 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1328 !map_value_has_spin_lock(map)) {
1332 value_size = bpf_map_value_size(map);
1334 max_count = attr->batch.count;
1338 key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1342 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1348 for (cp = 0; cp < max_count; cp++) {
1350 if (copy_from_user(key, keys + cp * map->key_size,
1352 copy_from_user(value, values + cp * value_size, value_size))
1355 err = bpf_map_update_value(map, f, key, value,
1356 attr->batch.elem_flags);
1362 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1370 #define MAP_LOOKUP_RETRIES 3
1372 int generic_map_lookup_batch(struct bpf_map *map,
1373 const union bpf_attr *attr,
1374 union bpf_attr __user *uattr)
1376 void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1377 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1378 void __user *values = u64_to_user_ptr(attr->batch.values);
1379 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1380 void *buf, *buf_prevkey, *prev_key, *key, *value;
1381 int err, retry = MAP_LOOKUP_RETRIES;
1382 u32 value_size, cp, max_count;
1384 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1387 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1388 !map_value_has_spin_lock(map))
1391 value_size = bpf_map_value_size(map);
1393 max_count = attr->batch.count;
1397 if (put_user(0, &uattr->batch.count))
1400 buf_prevkey = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1404 buf = kmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1412 if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1415 value = key + map->key_size;
1417 prev_key = buf_prevkey;
1419 for (cp = 0; cp < max_count;) {
1421 err = map->ops->map_get_next_key(map, prev_key, key);
1425 err = bpf_map_copy_value(map, key, value,
1426 attr->batch.elem_flags);
1428 if (err == -ENOENT) {
1440 if (copy_to_user(keys + cp * map->key_size, key,
1445 if (copy_to_user(values + cp * value_size, value, value_size)) {
1451 prev_key = buf_prevkey;
1453 swap(prev_key, key);
1454 retry = MAP_LOOKUP_RETRIES;
1461 if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1462 (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1471 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1473 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1475 void __user *ukey = u64_to_user_ptr(attr->key);
1476 void __user *uvalue = u64_to_user_ptr(attr->value);
1477 int ufd = attr->map_fd;
1478 struct bpf_map *map;
1484 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1488 map = __bpf_map_get(f);
1490 return PTR_ERR(map);
1491 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1492 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1497 key = __bpf_copy_key(ukey, map->key_size);
1503 value_size = map->value_size;
1506 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1510 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1511 map->map_type == BPF_MAP_TYPE_STACK) {
1512 err = map->ops->map_pop_elem(map, value);
1520 if (copy_to_user(uvalue, value, value_size) != 0) {
1536 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1538 static int map_freeze(const union bpf_attr *attr)
1540 int err = 0, ufd = attr->map_fd;
1541 struct bpf_map *map;
1544 if (CHECK_ATTR(BPF_MAP_FREEZE))
1548 map = __bpf_map_get(f);
1550 return PTR_ERR(map);
1552 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1557 mutex_lock(&map->freeze_mutex);
1559 if (map->writecnt) {
1563 if (READ_ONCE(map->frozen)) {
1567 if (!bpf_capable()) {
1572 WRITE_ONCE(map->frozen, true);
1574 mutex_unlock(&map->freeze_mutex);
1579 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1580 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1581 [_id] = & _name ## _prog_ops,
1582 #define BPF_MAP_TYPE(_id, _ops)
1583 #define BPF_LINK_TYPE(_id, _name)
1584 #include <linux/bpf_types.h>
1585 #undef BPF_PROG_TYPE
1587 #undef BPF_LINK_TYPE
1590 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1592 const struct bpf_prog_ops *ops;
1594 if (type >= ARRAY_SIZE(bpf_prog_types))
1596 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1597 ops = bpf_prog_types[type];
1601 if (!bpf_prog_is_dev_bound(prog->aux))
1602 prog->aux->ops = ops;
1604 prog->aux->ops = &bpf_offload_prog_ops;
1615 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1616 [BPF_AUDIT_LOAD] = "LOAD",
1617 [BPF_AUDIT_UNLOAD] = "UNLOAD",
1620 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1622 struct audit_context *ctx = NULL;
1623 struct audit_buffer *ab;
1625 if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1627 if (audit_enabled == AUDIT_OFF)
1629 if (op == BPF_AUDIT_LOAD)
1630 ctx = audit_context();
1631 ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1634 audit_log_format(ab, "prog-id=%u op=%s",
1635 prog->aux->id, bpf_audit_str[op]);
1639 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1643 idr_preload(GFP_KERNEL);
1644 spin_lock_bh(&prog_idr_lock);
1645 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1648 spin_unlock_bh(&prog_idr_lock);
1651 /* id is in [1, INT_MAX) */
1652 if (WARN_ON_ONCE(!id))
1655 return id > 0 ? 0 : id;
1658 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1660 /* cBPF to eBPF migrations are currently not in the idr store.
1661 * Offloaded programs are removed from the store when their device
1662 * disappears - even if someone grabs an fd to them they are unusable,
1663 * simply waiting for refcnt to drop to be freed.
1669 spin_lock_bh(&prog_idr_lock);
1671 __acquire(&prog_idr_lock);
1673 idr_remove(&prog_idr, prog->aux->id);
1677 spin_unlock_bh(&prog_idr_lock);
1679 __release(&prog_idr_lock);
1682 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1684 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1686 kvfree(aux->func_info);
1687 kfree(aux->func_info_aux);
1688 free_uid(aux->user);
1689 security_bpf_prog_free(aux);
1690 bpf_prog_free(aux->prog);
1693 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1695 bpf_prog_kallsyms_del_all(prog);
1696 btf_put(prog->aux->btf);
1697 bpf_prog_free_linfo(prog);
1698 if (prog->aux->attach_btf)
1699 btf_put(prog->aux->attach_btf);
1702 if (prog->aux->sleepable)
1703 call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
1705 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1707 __bpf_prog_put_rcu(&prog->aux->rcu);
1711 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1713 if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1714 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1715 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
1716 /* bpf_prog_free_id() must be called first */
1717 bpf_prog_free_id(prog, do_idr_lock);
1718 __bpf_prog_put_noref(prog, true);
1722 void bpf_prog_put(struct bpf_prog *prog)
1724 __bpf_prog_put(prog, true);
1726 EXPORT_SYMBOL_GPL(bpf_prog_put);
1728 static int bpf_prog_release(struct inode *inode, struct file *filp)
1730 struct bpf_prog *prog = filp->private_data;
1736 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1737 struct bpf_prog_stats *stats)
1739 u64 nsecs = 0, cnt = 0, misses = 0;
1742 for_each_possible_cpu(cpu) {
1743 const struct bpf_prog_stats *st;
1745 u64 tnsecs, tcnt, tmisses;
1747 st = per_cpu_ptr(prog->stats, cpu);
1749 start = u64_stats_fetch_begin_irq(&st->syncp);
1752 tmisses = st->misses;
1753 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1758 stats->nsecs = nsecs;
1760 stats->misses = misses;
1763 #ifdef CONFIG_PROC_FS
1764 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1766 const struct bpf_prog *prog = filp->private_data;
1767 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1768 struct bpf_prog_stats stats;
1770 bpf_prog_get_stats(prog, &stats);
1771 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1778 "run_time_ns:\t%llu\n"
1780 "recursion_misses:\t%llu\n",
1784 prog->pages * 1ULL << PAGE_SHIFT,
1792 const struct file_operations bpf_prog_fops = {
1793 #ifdef CONFIG_PROC_FS
1794 .show_fdinfo = bpf_prog_show_fdinfo,
1796 .release = bpf_prog_release,
1797 .read = bpf_dummy_read,
1798 .write = bpf_dummy_write,
1801 int bpf_prog_new_fd(struct bpf_prog *prog)
1805 ret = security_bpf_prog(prog);
1809 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1810 O_RDWR | O_CLOEXEC);
1813 static struct bpf_prog *____bpf_prog_get(struct fd f)
1816 return ERR_PTR(-EBADF);
1817 if (f.file->f_op != &bpf_prog_fops) {
1819 return ERR_PTR(-EINVAL);
1822 return f.file->private_data;
1825 void bpf_prog_add(struct bpf_prog *prog, int i)
1827 atomic64_add(i, &prog->aux->refcnt);
1829 EXPORT_SYMBOL_GPL(bpf_prog_add);
1831 void bpf_prog_sub(struct bpf_prog *prog, int i)
1833 /* Only to be used for undoing previous bpf_prog_add() in some
1834 * error path. We still know that another entity in our call
1835 * path holds a reference to the program, thus atomic_sub() can
1836 * be safely used in such cases!
1838 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1840 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1842 void bpf_prog_inc(struct bpf_prog *prog)
1844 atomic64_inc(&prog->aux->refcnt);
1846 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1848 /* prog_idr_lock should have been held */
1849 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1853 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1856 return ERR_PTR(-ENOENT);
1860 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1862 bool bpf_prog_get_ok(struct bpf_prog *prog,
1863 enum bpf_prog_type *attach_type, bool attach_drv)
1865 /* not an attachment, just a refcount inc, always allow */
1869 if (prog->type != *attach_type)
1871 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1877 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1880 struct fd f = fdget(ufd);
1881 struct bpf_prog *prog;
1883 prog = ____bpf_prog_get(f);
1886 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1887 prog = ERR_PTR(-EINVAL);
1897 struct bpf_prog *bpf_prog_get(u32 ufd)
1899 return __bpf_prog_get(ufd, NULL, false);
1902 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1905 return __bpf_prog_get(ufd, &type, attach_drv);
1907 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1909 /* Initially all BPF programs could be loaded w/o specifying
1910 * expected_attach_type. Later for some of them specifying expected_attach_type
1911 * at load time became required so that program could be validated properly.
1912 * Programs of types that are allowed to be loaded both w/ and w/o (for
1913 * backward compatibility) expected_attach_type, should have the default attach
1914 * type assigned to expected_attach_type for the latter case, so that it can be
1915 * validated later at attach time.
1917 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1918 * prog type requires it but has some attach types that have to be backward
1921 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1923 switch (attr->prog_type) {
1924 case BPF_PROG_TYPE_CGROUP_SOCK:
1925 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1926 * exist so checking for non-zero is the way to go here.
1928 if (!attr->expected_attach_type)
1929 attr->expected_attach_type =
1930 BPF_CGROUP_INET_SOCK_CREATE;
1936 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1937 enum bpf_attach_type expected_attach_type,
1938 struct btf *attach_btf, u32 btf_id,
1939 struct bpf_prog *dst_prog)
1942 if (btf_id > BTF_MAX_TYPE)
1945 if (!attach_btf && !dst_prog)
1948 switch (prog_type) {
1949 case BPF_PROG_TYPE_TRACING:
1950 case BPF_PROG_TYPE_LSM:
1951 case BPF_PROG_TYPE_STRUCT_OPS:
1952 case BPF_PROG_TYPE_EXT:
1959 if (attach_btf && (!btf_id || dst_prog))
1962 if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
1963 prog_type != BPF_PROG_TYPE_EXT)
1966 switch (prog_type) {
1967 case BPF_PROG_TYPE_CGROUP_SOCK:
1968 switch (expected_attach_type) {
1969 case BPF_CGROUP_INET_SOCK_CREATE:
1970 case BPF_CGROUP_INET_SOCK_RELEASE:
1971 case BPF_CGROUP_INET4_POST_BIND:
1972 case BPF_CGROUP_INET6_POST_BIND:
1977 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1978 switch (expected_attach_type) {
1979 case BPF_CGROUP_INET4_BIND:
1980 case BPF_CGROUP_INET6_BIND:
1981 case BPF_CGROUP_INET4_CONNECT:
1982 case BPF_CGROUP_INET6_CONNECT:
1983 case BPF_CGROUP_INET4_GETPEERNAME:
1984 case BPF_CGROUP_INET6_GETPEERNAME:
1985 case BPF_CGROUP_INET4_GETSOCKNAME:
1986 case BPF_CGROUP_INET6_GETSOCKNAME:
1987 case BPF_CGROUP_UDP4_SENDMSG:
1988 case BPF_CGROUP_UDP6_SENDMSG:
1989 case BPF_CGROUP_UDP4_RECVMSG:
1990 case BPF_CGROUP_UDP6_RECVMSG:
1995 case BPF_PROG_TYPE_CGROUP_SKB:
1996 switch (expected_attach_type) {
1997 case BPF_CGROUP_INET_INGRESS:
1998 case BPF_CGROUP_INET_EGRESS:
2003 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2004 switch (expected_attach_type) {
2005 case BPF_CGROUP_SETSOCKOPT:
2006 case BPF_CGROUP_GETSOCKOPT:
2011 case BPF_PROG_TYPE_SK_LOOKUP:
2012 if (expected_attach_type == BPF_SK_LOOKUP)
2015 case BPF_PROG_TYPE_EXT:
2016 if (expected_attach_type)
2024 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2026 switch (prog_type) {
2027 case BPF_PROG_TYPE_SCHED_CLS:
2028 case BPF_PROG_TYPE_SCHED_ACT:
2029 case BPF_PROG_TYPE_XDP:
2030 case BPF_PROG_TYPE_LWT_IN:
2031 case BPF_PROG_TYPE_LWT_OUT:
2032 case BPF_PROG_TYPE_LWT_XMIT:
2033 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2034 case BPF_PROG_TYPE_SK_SKB:
2035 case BPF_PROG_TYPE_SK_MSG:
2036 case BPF_PROG_TYPE_LIRC_MODE2:
2037 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2038 case BPF_PROG_TYPE_CGROUP_DEVICE:
2039 case BPF_PROG_TYPE_CGROUP_SOCK:
2040 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2041 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2042 case BPF_PROG_TYPE_CGROUP_SYSCTL:
2043 case BPF_PROG_TYPE_SOCK_OPS:
2044 case BPF_PROG_TYPE_EXT: /* extends any prog */
2046 case BPF_PROG_TYPE_CGROUP_SKB:
2048 case BPF_PROG_TYPE_SK_REUSEPORT:
2049 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2055 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2057 switch (prog_type) {
2058 case BPF_PROG_TYPE_KPROBE:
2059 case BPF_PROG_TYPE_TRACEPOINT:
2060 case BPF_PROG_TYPE_PERF_EVENT:
2061 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2062 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2063 case BPF_PROG_TYPE_TRACING:
2064 case BPF_PROG_TYPE_LSM:
2065 case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2066 case BPF_PROG_TYPE_EXT: /* extends any prog */
2073 /* last field in 'union bpf_attr' used by this command */
2074 #define BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
2076 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
2078 enum bpf_prog_type type = attr->prog_type;
2079 struct bpf_prog *prog, *dst_prog = NULL;
2080 struct btf *attach_btf = NULL;
2085 if (CHECK_ATTR(BPF_PROG_LOAD))
2088 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2089 BPF_F_ANY_ALIGNMENT |
2090 BPF_F_TEST_STATE_FREQ |
2092 BPF_F_TEST_RND_HI32))
2095 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2096 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2100 /* copy eBPF program license from user space */
2101 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
2102 sizeof(license) - 1) < 0)
2104 license[sizeof(license) - 1] = 0;
2106 /* eBPF programs must be GPL compatible to use GPL-ed functions */
2107 is_gpl = license_is_gpl_compatible(license);
2109 if (attr->insn_cnt == 0 ||
2110 attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2112 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2113 type != BPF_PROG_TYPE_CGROUP_SKB &&
2117 if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2119 if (is_perfmon_prog_type(type) && !perfmon_capable())
2122 /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2123 * or btf, we need to check which one it is
2125 if (attr->attach_prog_fd) {
2126 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2127 if (IS_ERR(dst_prog)) {
2129 attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2130 if (IS_ERR(attach_btf))
2132 if (!btf_is_kernel(attach_btf)) {
2133 /* attaching through specifying bpf_prog's BTF
2134 * objects directly might be supported eventually
2136 btf_put(attach_btf);
2140 } else if (attr->attach_btf_id) {
2141 /* fall back to vmlinux BTF, if BTF type ID is specified */
2142 attach_btf = bpf_get_btf_vmlinux();
2143 if (IS_ERR(attach_btf))
2144 return PTR_ERR(attach_btf);
2147 btf_get(attach_btf);
2150 bpf_prog_load_fixup_attach_type(attr);
2151 if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2152 attach_btf, attr->attach_btf_id,
2155 bpf_prog_put(dst_prog);
2157 btf_put(attach_btf);
2161 /* plain bpf_prog allocation */
2162 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2165 bpf_prog_put(dst_prog);
2167 btf_put(attach_btf);
2171 prog->expected_attach_type = attr->expected_attach_type;
2172 prog->aux->attach_btf = attach_btf;
2173 prog->aux->attach_btf_id = attr->attach_btf_id;
2174 prog->aux->dst_prog = dst_prog;
2175 prog->aux->offload_requested = !!attr->prog_ifindex;
2176 prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2178 err = security_bpf_prog_alloc(prog->aux);
2182 prog->aux->user = get_current_user();
2183 prog->len = attr->insn_cnt;
2186 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
2187 bpf_prog_insn_size(prog)) != 0)
2190 prog->orig_prog = NULL;
2193 atomic64_set(&prog->aux->refcnt, 1);
2194 prog->gpl_compatible = is_gpl ? 1 : 0;
2196 if (bpf_prog_is_dev_bound(prog->aux)) {
2197 err = bpf_prog_offload_init(prog, attr);
2202 /* find program type: socket_filter vs tracing_filter */
2203 err = find_prog_type(type, prog);
2207 prog->aux->load_time = ktime_get_boottime_ns();
2208 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2209 sizeof(attr->prog_name));
2213 /* run eBPF verifier */
2214 err = bpf_check(&prog, attr, uattr);
2216 goto free_used_maps;
2218 prog = bpf_prog_select_runtime(prog, &err);
2220 goto free_used_maps;
2222 err = bpf_prog_alloc_id(prog);
2224 goto free_used_maps;
2226 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2227 * effectively publicly exposed. However, retrieving via
2228 * bpf_prog_get_fd_by_id() will take another reference,
2229 * therefore it cannot be gone underneath us.
2231 * Only for the time /after/ successful bpf_prog_new_fd()
2232 * and before returning to userspace, we might just hold
2233 * one reference and any parallel close on that fd could
2234 * rip everything out. Hence, below notifications must
2235 * happen before bpf_prog_new_fd().
2237 * Also, any failure handling from this point onwards must
2238 * be using bpf_prog_put() given the program is exposed.
2240 bpf_prog_kallsyms_add(prog);
2241 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2242 bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2244 err = bpf_prog_new_fd(prog);
2250 /* In case we have subprogs, we need to wait for a grace
2251 * period before we can tear down JIT memory since symbols
2252 * are already exposed under kallsyms.
2254 __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2257 free_uid(prog->aux->user);
2258 security_bpf_prog_free(prog->aux);
2260 if (prog->aux->attach_btf)
2261 btf_put(prog->aux->attach_btf);
2262 bpf_prog_free(prog);
2266 #define BPF_OBJ_LAST_FIELD file_flags
2268 static int bpf_obj_pin(const union bpf_attr *attr)
2270 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2273 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2276 static int bpf_obj_get(const union bpf_attr *attr)
2278 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2279 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2282 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2286 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2287 const struct bpf_link_ops *ops, struct bpf_prog *prog)
2289 atomic64_set(&link->refcnt, 1);
2296 static void bpf_link_free_id(int id)
2301 spin_lock_bh(&link_idr_lock);
2302 idr_remove(&link_idr, id);
2303 spin_unlock_bh(&link_idr_lock);
2306 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2307 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2308 * anon_inode's release() call. This helper marksbpf_link as
2309 * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2310 * is not decremented, it's the responsibility of a calling code that failed
2311 * to complete bpf_link initialization.
2313 void bpf_link_cleanup(struct bpf_link_primer *primer)
2315 primer->link->prog = NULL;
2316 bpf_link_free_id(primer->id);
2318 put_unused_fd(primer->fd);
2321 void bpf_link_inc(struct bpf_link *link)
2323 atomic64_inc(&link->refcnt);
2326 /* bpf_link_free is guaranteed to be called from process context */
2327 static void bpf_link_free(struct bpf_link *link)
2329 bpf_link_free_id(link->id);
2331 /* detach BPF program, clean up used resources */
2332 link->ops->release(link);
2333 bpf_prog_put(link->prog);
2335 /* free bpf_link and its containing memory */
2336 link->ops->dealloc(link);
2339 static void bpf_link_put_deferred(struct work_struct *work)
2341 struct bpf_link *link = container_of(work, struct bpf_link, work);
2343 bpf_link_free(link);
2346 /* bpf_link_put can be called from atomic context, but ensures that resources
2347 * are freed from process context
2349 void bpf_link_put(struct bpf_link *link)
2351 if (!atomic64_dec_and_test(&link->refcnt))
2355 INIT_WORK(&link->work, bpf_link_put_deferred);
2356 schedule_work(&link->work);
2358 bpf_link_free(link);
2362 static int bpf_link_release(struct inode *inode, struct file *filp)
2364 struct bpf_link *link = filp->private_data;
2370 #ifdef CONFIG_PROC_FS
2371 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2372 #define BPF_MAP_TYPE(_id, _ops)
2373 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2374 static const char *bpf_link_type_strs[] = {
2375 [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2376 #include <linux/bpf_types.h>
2378 #undef BPF_PROG_TYPE
2380 #undef BPF_LINK_TYPE
2382 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2384 const struct bpf_link *link = filp->private_data;
2385 const struct bpf_prog *prog = link->prog;
2386 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2388 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2394 bpf_link_type_strs[link->type],
2398 if (link->ops->show_fdinfo)
2399 link->ops->show_fdinfo(link, m);
2403 static const struct file_operations bpf_link_fops = {
2404 #ifdef CONFIG_PROC_FS
2405 .show_fdinfo = bpf_link_show_fdinfo,
2407 .release = bpf_link_release,
2408 .read = bpf_dummy_read,
2409 .write = bpf_dummy_write,
2412 static int bpf_link_alloc_id(struct bpf_link *link)
2416 idr_preload(GFP_KERNEL);
2417 spin_lock_bh(&link_idr_lock);
2418 id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2419 spin_unlock_bh(&link_idr_lock);
2425 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2426 * reserving unused FD and allocating ID from link_idr. This is to be paired
2427 * with bpf_link_settle() to install FD and ID and expose bpf_link to
2428 * user-space, if bpf_link is successfully attached. If not, bpf_link and
2429 * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2430 * transient state is passed around in struct bpf_link_primer.
2431 * This is preferred way to create and initialize bpf_link, especially when
2432 * there are complicated and expensive operations inbetween creating bpf_link
2433 * itself and attaching it to BPF hook. By using bpf_link_prime() and
2434 * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2435 * expensive (and potentially failing) roll back operations in a rare case
2436 * that file, FD, or ID can't be allocated.
2438 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2443 fd = get_unused_fd_flags(O_CLOEXEC);
2448 id = bpf_link_alloc_id(link);
2454 file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2456 bpf_link_free_id(id);
2458 return PTR_ERR(file);
2461 primer->link = link;
2462 primer->file = file;
2468 int bpf_link_settle(struct bpf_link_primer *primer)
2470 /* make bpf_link fetchable by ID */
2471 spin_lock_bh(&link_idr_lock);
2472 primer->link->id = primer->id;
2473 spin_unlock_bh(&link_idr_lock);
2474 /* make bpf_link fetchable by FD */
2475 fd_install(primer->fd, primer->file);
2476 /* pass through installed FD */
2480 int bpf_link_new_fd(struct bpf_link *link)
2482 return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2485 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2487 struct fd f = fdget(ufd);
2488 struct bpf_link *link;
2491 return ERR_PTR(-EBADF);
2492 if (f.file->f_op != &bpf_link_fops) {
2494 return ERR_PTR(-EINVAL);
2497 link = f.file->private_data;
2504 struct bpf_tracing_link {
2505 struct bpf_link link;
2506 enum bpf_attach_type attach_type;
2507 struct bpf_trampoline *trampoline;
2508 struct bpf_prog *tgt_prog;
2511 static void bpf_tracing_link_release(struct bpf_link *link)
2513 struct bpf_tracing_link *tr_link =
2514 container_of(link, struct bpf_tracing_link, link);
2516 WARN_ON_ONCE(bpf_trampoline_unlink_prog(link->prog,
2517 tr_link->trampoline));
2519 bpf_trampoline_put(tr_link->trampoline);
2521 /* tgt_prog is NULL if target is a kernel function */
2522 if (tr_link->tgt_prog)
2523 bpf_prog_put(tr_link->tgt_prog);
2526 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2528 struct bpf_tracing_link *tr_link =
2529 container_of(link, struct bpf_tracing_link, link);
2534 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2535 struct seq_file *seq)
2537 struct bpf_tracing_link *tr_link =
2538 container_of(link, struct bpf_tracing_link, link);
2541 "attach_type:\t%d\n",
2542 tr_link->attach_type);
2545 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2546 struct bpf_link_info *info)
2548 struct bpf_tracing_link *tr_link =
2549 container_of(link, struct bpf_tracing_link, link);
2551 info->tracing.attach_type = tr_link->attach_type;
2556 static const struct bpf_link_ops bpf_tracing_link_lops = {
2557 .release = bpf_tracing_link_release,
2558 .dealloc = bpf_tracing_link_dealloc,
2559 .show_fdinfo = bpf_tracing_link_show_fdinfo,
2560 .fill_link_info = bpf_tracing_link_fill_link_info,
2563 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2567 struct bpf_link_primer link_primer;
2568 struct bpf_prog *tgt_prog = NULL;
2569 struct bpf_trampoline *tr = NULL;
2570 struct bpf_tracing_link *link;
2574 switch (prog->type) {
2575 case BPF_PROG_TYPE_TRACING:
2576 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2577 prog->expected_attach_type != BPF_TRACE_FEXIT &&
2578 prog->expected_attach_type != BPF_MODIFY_RETURN) {
2583 case BPF_PROG_TYPE_EXT:
2584 if (prog->expected_attach_type != 0) {
2589 case BPF_PROG_TYPE_LSM:
2590 if (prog->expected_attach_type != BPF_LSM_MAC) {
2600 if (!!tgt_prog_fd != !!btf_id) {
2606 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
2607 if (prog->type != BPF_PROG_TYPE_EXT) {
2612 tgt_prog = bpf_prog_get(tgt_prog_fd);
2613 if (IS_ERR(tgt_prog)) {
2614 err = PTR_ERR(tgt_prog);
2619 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
2622 link = kzalloc(sizeof(*link), GFP_USER);
2627 bpf_link_init(&link->link, BPF_LINK_TYPE_TRACING,
2628 &bpf_tracing_link_lops, prog);
2629 link->attach_type = prog->expected_attach_type;
2631 mutex_lock(&prog->aux->dst_mutex);
2633 /* There are a few possible cases here:
2635 * - if prog->aux->dst_trampoline is set, the program was just loaded
2636 * and not yet attached to anything, so we can use the values stored
2639 * - if prog->aux->dst_trampoline is NULL, the program has already been
2640 * attached to a target and its initial target was cleared (below)
2642 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
2643 * target_btf_id using the link_create API.
2645 * - if tgt_prog == NULL when this function was called using the old
2646 * raw_tracepoint_open API, and we need a target from prog->aux
2648 * The combination of no saved target in prog->aux, and no target
2649 * specified on load is illegal, and we reject that here.
2651 if (!prog->aux->dst_trampoline && !tgt_prog) {
2656 if (!prog->aux->dst_trampoline ||
2657 (key && key != prog->aux->dst_trampoline->key)) {
2658 /* If there is no saved target, or the specified target is
2659 * different from the destination specified at load time, we
2660 * need a new trampoline and a check for compatibility
2662 struct bpf_attach_target_info tgt_info = {};
2664 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
2669 tr = bpf_trampoline_get(key, &tgt_info);
2675 /* The caller didn't specify a target, or the target was the
2676 * same as the destination supplied during program load. This
2677 * means we can reuse the trampoline and reference from program
2678 * load time, and there is no need to allocate a new one. This
2679 * can only happen once for any program, as the saved values in
2680 * prog->aux are cleared below.
2682 tr = prog->aux->dst_trampoline;
2683 tgt_prog = prog->aux->dst_prog;
2686 err = bpf_link_prime(&link->link, &link_primer);
2690 err = bpf_trampoline_link_prog(prog, tr);
2692 bpf_link_cleanup(&link_primer);
2697 link->tgt_prog = tgt_prog;
2698 link->trampoline = tr;
2700 /* Always clear the trampoline and target prog from prog->aux to make
2701 * sure the original attach destination is not kept alive after a
2702 * program is (re-)attached to another target.
2704 if (prog->aux->dst_prog &&
2705 (tgt_prog_fd || tr != prog->aux->dst_trampoline))
2706 /* got extra prog ref from syscall, or attaching to different prog */
2707 bpf_prog_put(prog->aux->dst_prog);
2708 if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
2709 /* we allocated a new trampoline, so free the old one */
2710 bpf_trampoline_put(prog->aux->dst_trampoline);
2712 prog->aux->dst_prog = NULL;
2713 prog->aux->dst_trampoline = NULL;
2714 mutex_unlock(&prog->aux->dst_mutex);
2716 return bpf_link_settle(&link_primer);
2718 if (tr && tr != prog->aux->dst_trampoline)
2719 bpf_trampoline_put(tr);
2720 mutex_unlock(&prog->aux->dst_mutex);
2723 if (tgt_prog_fd && tgt_prog)
2724 bpf_prog_put(tgt_prog);
2728 struct bpf_raw_tp_link {
2729 struct bpf_link link;
2730 struct bpf_raw_event_map *btp;
2733 static void bpf_raw_tp_link_release(struct bpf_link *link)
2735 struct bpf_raw_tp_link *raw_tp =
2736 container_of(link, struct bpf_raw_tp_link, link);
2738 bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
2739 bpf_put_raw_tracepoint(raw_tp->btp);
2742 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
2744 struct bpf_raw_tp_link *raw_tp =
2745 container_of(link, struct bpf_raw_tp_link, link);
2750 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
2751 struct seq_file *seq)
2753 struct bpf_raw_tp_link *raw_tp_link =
2754 container_of(link, struct bpf_raw_tp_link, link);
2758 raw_tp_link->btp->tp->name);
2761 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
2762 struct bpf_link_info *info)
2764 struct bpf_raw_tp_link *raw_tp_link =
2765 container_of(link, struct bpf_raw_tp_link, link);
2766 char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
2767 const char *tp_name = raw_tp_link->btp->tp->name;
2768 u32 ulen = info->raw_tracepoint.tp_name_len;
2769 size_t tp_len = strlen(tp_name);
2774 info->raw_tracepoint.tp_name_len = tp_len + 1;
2779 if (ulen >= tp_len + 1) {
2780 if (copy_to_user(ubuf, tp_name, tp_len + 1))
2785 if (copy_to_user(ubuf, tp_name, ulen - 1))
2787 if (put_user(zero, ubuf + ulen - 1))
2795 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
2796 .release = bpf_raw_tp_link_release,
2797 .dealloc = bpf_raw_tp_link_dealloc,
2798 .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
2799 .fill_link_info = bpf_raw_tp_link_fill_link_info,
2802 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
2804 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
2806 struct bpf_link_primer link_primer;
2807 struct bpf_raw_tp_link *link;
2808 struct bpf_raw_event_map *btp;
2809 struct bpf_prog *prog;
2810 const char *tp_name;
2814 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
2817 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
2819 return PTR_ERR(prog);
2821 switch (prog->type) {
2822 case BPF_PROG_TYPE_TRACING:
2823 case BPF_PROG_TYPE_EXT:
2824 case BPF_PROG_TYPE_LSM:
2825 if (attr->raw_tracepoint.name) {
2826 /* The attach point for this category of programs
2827 * should be specified via btf_id during program load.
2832 if (prog->type == BPF_PROG_TYPE_TRACING &&
2833 prog->expected_attach_type == BPF_TRACE_RAW_TP) {
2834 tp_name = prog->aux->attach_func_name;
2837 err = bpf_tracing_prog_attach(prog, 0, 0);
2841 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2842 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2843 if (strncpy_from_user(buf,
2844 u64_to_user_ptr(attr->raw_tracepoint.name),
2845 sizeof(buf) - 1) < 0) {
2849 buf[sizeof(buf) - 1] = 0;
2857 btp = bpf_get_raw_tracepoint(tp_name);
2863 link = kzalloc(sizeof(*link), GFP_USER);
2868 bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
2869 &bpf_raw_tp_link_lops, prog);
2872 err = bpf_link_prime(&link->link, &link_primer);
2878 err = bpf_probe_register(link->btp, prog);
2880 bpf_link_cleanup(&link_primer);
2884 return bpf_link_settle(&link_primer);
2887 bpf_put_raw_tracepoint(btp);
2893 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2894 enum bpf_attach_type attach_type)
2896 switch (prog->type) {
2897 case BPF_PROG_TYPE_CGROUP_SOCK:
2898 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2899 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2900 case BPF_PROG_TYPE_SK_LOOKUP:
2901 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2902 case BPF_PROG_TYPE_CGROUP_SKB:
2903 if (!capable(CAP_NET_ADMIN))
2904 /* cg-skb progs can be loaded by unpriv user.
2905 * check permissions at attach time.
2908 return prog->enforce_expected_attach_type &&
2909 prog->expected_attach_type != attach_type ?
2916 static enum bpf_prog_type
2917 attach_type_to_prog_type(enum bpf_attach_type attach_type)
2919 switch (attach_type) {
2920 case BPF_CGROUP_INET_INGRESS:
2921 case BPF_CGROUP_INET_EGRESS:
2922 return BPF_PROG_TYPE_CGROUP_SKB;
2923 case BPF_CGROUP_INET_SOCK_CREATE:
2924 case BPF_CGROUP_INET_SOCK_RELEASE:
2925 case BPF_CGROUP_INET4_POST_BIND:
2926 case BPF_CGROUP_INET6_POST_BIND:
2927 return BPF_PROG_TYPE_CGROUP_SOCK;
2928 case BPF_CGROUP_INET4_BIND:
2929 case BPF_CGROUP_INET6_BIND:
2930 case BPF_CGROUP_INET4_CONNECT:
2931 case BPF_CGROUP_INET6_CONNECT:
2932 case BPF_CGROUP_INET4_GETPEERNAME:
2933 case BPF_CGROUP_INET6_GETPEERNAME:
2934 case BPF_CGROUP_INET4_GETSOCKNAME:
2935 case BPF_CGROUP_INET6_GETSOCKNAME:
2936 case BPF_CGROUP_UDP4_SENDMSG:
2937 case BPF_CGROUP_UDP6_SENDMSG:
2938 case BPF_CGROUP_UDP4_RECVMSG:
2939 case BPF_CGROUP_UDP6_RECVMSG:
2940 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2941 case BPF_CGROUP_SOCK_OPS:
2942 return BPF_PROG_TYPE_SOCK_OPS;
2943 case BPF_CGROUP_DEVICE:
2944 return BPF_PROG_TYPE_CGROUP_DEVICE;
2945 case BPF_SK_MSG_VERDICT:
2946 return BPF_PROG_TYPE_SK_MSG;
2947 case BPF_SK_SKB_STREAM_PARSER:
2948 case BPF_SK_SKB_STREAM_VERDICT:
2949 return BPF_PROG_TYPE_SK_SKB;
2950 case BPF_LIRC_MODE2:
2951 return BPF_PROG_TYPE_LIRC_MODE2;
2952 case BPF_FLOW_DISSECTOR:
2953 return BPF_PROG_TYPE_FLOW_DISSECTOR;
2954 case BPF_CGROUP_SYSCTL:
2955 return BPF_PROG_TYPE_CGROUP_SYSCTL;
2956 case BPF_CGROUP_GETSOCKOPT:
2957 case BPF_CGROUP_SETSOCKOPT:
2958 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
2959 case BPF_TRACE_ITER:
2960 return BPF_PROG_TYPE_TRACING;
2962 return BPF_PROG_TYPE_SK_LOOKUP;
2964 return BPF_PROG_TYPE_XDP;
2966 return BPF_PROG_TYPE_UNSPEC;
2970 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
2972 #define BPF_F_ATTACH_MASK \
2973 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
2975 static int bpf_prog_attach(const union bpf_attr *attr)
2977 enum bpf_prog_type ptype;
2978 struct bpf_prog *prog;
2981 if (CHECK_ATTR(BPF_PROG_ATTACH))
2984 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
2987 ptype = attach_type_to_prog_type(attr->attach_type);
2988 if (ptype == BPF_PROG_TYPE_UNSPEC)
2991 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
2993 return PTR_ERR(prog);
2995 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3001 case BPF_PROG_TYPE_SK_SKB:
3002 case BPF_PROG_TYPE_SK_MSG:
3003 ret = sock_map_get_from_fd(attr, prog);
3005 case BPF_PROG_TYPE_LIRC_MODE2:
3006 ret = lirc_prog_attach(attr, prog);
3008 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3009 ret = netns_bpf_prog_attach(attr, prog);
3011 case BPF_PROG_TYPE_CGROUP_DEVICE:
3012 case BPF_PROG_TYPE_CGROUP_SKB:
3013 case BPF_PROG_TYPE_CGROUP_SOCK:
3014 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3015 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3016 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3017 case BPF_PROG_TYPE_SOCK_OPS:
3018 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3029 #define BPF_PROG_DETACH_LAST_FIELD attach_type
3031 static int bpf_prog_detach(const union bpf_attr *attr)
3033 enum bpf_prog_type ptype;
3035 if (CHECK_ATTR(BPF_PROG_DETACH))
3038 ptype = attach_type_to_prog_type(attr->attach_type);
3041 case BPF_PROG_TYPE_SK_MSG:
3042 case BPF_PROG_TYPE_SK_SKB:
3043 return sock_map_prog_detach(attr, ptype);
3044 case BPF_PROG_TYPE_LIRC_MODE2:
3045 return lirc_prog_detach(attr);
3046 case BPF_PROG_TYPE_FLOW_DISSECTOR:
3047 return netns_bpf_prog_detach(attr, ptype);
3048 case BPF_PROG_TYPE_CGROUP_DEVICE:
3049 case BPF_PROG_TYPE_CGROUP_SKB:
3050 case BPF_PROG_TYPE_CGROUP_SOCK:
3051 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3052 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3053 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3054 case BPF_PROG_TYPE_SOCK_OPS:
3055 return cgroup_bpf_prog_detach(attr, ptype);
3061 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
3063 static int bpf_prog_query(const union bpf_attr *attr,
3064 union bpf_attr __user *uattr)
3066 if (!capable(CAP_NET_ADMIN))
3068 if (CHECK_ATTR(BPF_PROG_QUERY))
3070 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3073 switch (attr->query.attach_type) {
3074 case BPF_CGROUP_INET_INGRESS:
3075 case BPF_CGROUP_INET_EGRESS:
3076 case BPF_CGROUP_INET_SOCK_CREATE:
3077 case BPF_CGROUP_INET_SOCK_RELEASE:
3078 case BPF_CGROUP_INET4_BIND:
3079 case BPF_CGROUP_INET6_BIND:
3080 case BPF_CGROUP_INET4_POST_BIND:
3081 case BPF_CGROUP_INET6_POST_BIND:
3082 case BPF_CGROUP_INET4_CONNECT:
3083 case BPF_CGROUP_INET6_CONNECT:
3084 case BPF_CGROUP_INET4_GETPEERNAME:
3085 case BPF_CGROUP_INET6_GETPEERNAME:
3086 case BPF_CGROUP_INET4_GETSOCKNAME:
3087 case BPF_CGROUP_INET6_GETSOCKNAME:
3088 case BPF_CGROUP_UDP4_SENDMSG:
3089 case BPF_CGROUP_UDP6_SENDMSG:
3090 case BPF_CGROUP_UDP4_RECVMSG:
3091 case BPF_CGROUP_UDP6_RECVMSG:
3092 case BPF_CGROUP_SOCK_OPS:
3093 case BPF_CGROUP_DEVICE:
3094 case BPF_CGROUP_SYSCTL:
3095 case BPF_CGROUP_GETSOCKOPT:
3096 case BPF_CGROUP_SETSOCKOPT:
3097 return cgroup_bpf_prog_query(attr, uattr);
3098 case BPF_LIRC_MODE2:
3099 return lirc_prog_query(attr, uattr);
3100 case BPF_FLOW_DISSECTOR:
3102 return netns_bpf_prog_query(attr, uattr);
3108 #define BPF_PROG_TEST_RUN_LAST_FIELD test.cpu
3110 static int bpf_prog_test_run(const union bpf_attr *attr,
3111 union bpf_attr __user *uattr)
3113 struct bpf_prog *prog;
3114 int ret = -ENOTSUPP;
3116 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3119 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3120 (!attr->test.ctx_size_in && attr->test.ctx_in))
3123 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3124 (!attr->test.ctx_size_out && attr->test.ctx_out))
3127 prog = bpf_prog_get(attr->test.prog_fd);
3129 return PTR_ERR(prog);
3131 if (prog->aux->ops->test_run)
3132 ret = prog->aux->ops->test_run(prog, attr, uattr);
3138 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3140 static int bpf_obj_get_next_id(const union bpf_attr *attr,
3141 union bpf_attr __user *uattr,
3145 u32 next_id = attr->start_id;
3148 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3151 if (!capable(CAP_SYS_ADMIN))
3156 if (!idr_get_next(idr, &next_id))
3158 spin_unlock_bh(lock);
3161 err = put_user(next_id, &uattr->next_id);
3166 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3168 struct bpf_map *map;
3170 spin_lock_bh(&map_idr_lock);
3172 map = idr_get_next(&map_idr, id);
3174 map = __bpf_map_inc_not_zero(map, false);
3180 spin_unlock_bh(&map_idr_lock);
3185 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3187 struct bpf_prog *prog;
3189 spin_lock_bh(&prog_idr_lock);
3191 prog = idr_get_next(&prog_idr, id);
3193 prog = bpf_prog_inc_not_zero(prog);
3199 spin_unlock_bh(&prog_idr_lock);
3204 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3206 struct bpf_prog *bpf_prog_by_id(u32 id)
3208 struct bpf_prog *prog;
3211 return ERR_PTR(-ENOENT);
3213 spin_lock_bh(&prog_idr_lock);
3214 prog = idr_find(&prog_idr, id);
3216 prog = bpf_prog_inc_not_zero(prog);
3218 prog = ERR_PTR(-ENOENT);
3219 spin_unlock_bh(&prog_idr_lock);
3223 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3225 struct bpf_prog *prog;
3226 u32 id = attr->prog_id;
3229 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3232 if (!capable(CAP_SYS_ADMIN))
3235 prog = bpf_prog_by_id(id);
3237 return PTR_ERR(prog);
3239 fd = bpf_prog_new_fd(prog);
3246 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3248 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3250 struct bpf_map *map;
3251 u32 id = attr->map_id;
3255 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3256 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3259 if (!capable(CAP_SYS_ADMIN))
3262 f_flags = bpf_get_file_flag(attr->open_flags);
3266 spin_lock_bh(&map_idr_lock);
3267 map = idr_find(&map_idr, id);
3269 map = __bpf_map_inc_not_zero(map, true);
3271 map = ERR_PTR(-ENOENT);
3272 spin_unlock_bh(&map_idr_lock);
3275 return PTR_ERR(map);
3277 fd = bpf_map_new_fd(map, f_flags);
3279 bpf_map_put_with_uref(map);
3284 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3285 unsigned long addr, u32 *off,
3288 const struct bpf_map *map;
3291 mutex_lock(&prog->aux->used_maps_mutex);
3292 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3293 map = prog->aux->used_maps[i];
3294 if (map == (void *)addr) {
3295 *type = BPF_PSEUDO_MAP_FD;
3298 if (!map->ops->map_direct_value_meta)
3300 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3301 *type = BPF_PSEUDO_MAP_VALUE;
3308 mutex_unlock(&prog->aux->used_maps_mutex);
3312 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3313 const struct cred *f_cred)
3315 const struct bpf_map *map;
3316 struct bpf_insn *insns;
3322 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3327 for (i = 0; i < prog->len; i++) {
3328 code = insns[i].code;
3330 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3331 insns[i].code = BPF_JMP | BPF_CALL;
3332 insns[i].imm = BPF_FUNC_tail_call;
3335 if (code == (BPF_JMP | BPF_CALL) ||
3336 code == (BPF_JMP | BPF_CALL_ARGS)) {
3337 if (code == (BPF_JMP | BPF_CALL_ARGS))
3338 insns[i].code = BPF_JMP | BPF_CALL;
3339 if (!bpf_dump_raw_ok(f_cred))
3343 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3344 insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3348 if (code != (BPF_LD | BPF_IMM | BPF_DW))
3351 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3352 map = bpf_map_from_imm(prog, imm, &off, &type);
3354 insns[i].src_reg = type;
3355 insns[i].imm = map->id;
3356 insns[i + 1].imm = off;
3364 static int set_info_rec_size(struct bpf_prog_info *info)
3367 * Ensure info.*_rec_size is the same as kernel expected size
3371 * Only allow zero *_rec_size if both _rec_size and _cnt are
3372 * zero. In this case, the kernel will set the expected
3373 * _rec_size back to the info.
3376 if ((info->nr_func_info || info->func_info_rec_size) &&
3377 info->func_info_rec_size != sizeof(struct bpf_func_info))
3380 if ((info->nr_line_info || info->line_info_rec_size) &&
3381 info->line_info_rec_size != sizeof(struct bpf_line_info))
3384 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3385 info->jited_line_info_rec_size != sizeof(__u64))
3388 info->func_info_rec_size = sizeof(struct bpf_func_info);
3389 info->line_info_rec_size = sizeof(struct bpf_line_info);
3390 info->jited_line_info_rec_size = sizeof(__u64);
3395 static int bpf_prog_get_info_by_fd(struct file *file,
3396 struct bpf_prog *prog,
3397 const union bpf_attr *attr,
3398 union bpf_attr __user *uattr)
3400 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3401 struct bpf_prog_info info;
3402 u32 info_len = attr->info.info_len;
3403 struct bpf_prog_stats stats;
3404 char __user *uinsns;
3408 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3411 info_len = min_t(u32, sizeof(info), info_len);
3413 memset(&info, 0, sizeof(info));
3414 if (copy_from_user(&info, uinfo, info_len))
3417 info.type = prog->type;
3418 info.id = prog->aux->id;
3419 info.load_time = prog->aux->load_time;
3420 info.created_by_uid = from_kuid_munged(current_user_ns(),
3421 prog->aux->user->uid);
3422 info.gpl_compatible = prog->gpl_compatible;
3424 memcpy(info.tag, prog->tag, sizeof(prog->tag));
3425 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3427 mutex_lock(&prog->aux->used_maps_mutex);
3428 ulen = info.nr_map_ids;
3429 info.nr_map_ids = prog->aux->used_map_cnt;
3430 ulen = min_t(u32, info.nr_map_ids, ulen);
3432 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3435 for (i = 0; i < ulen; i++)
3436 if (put_user(prog->aux->used_maps[i]->id,
3437 &user_map_ids[i])) {
3438 mutex_unlock(&prog->aux->used_maps_mutex);
3442 mutex_unlock(&prog->aux->used_maps_mutex);
3444 err = set_info_rec_size(&info);
3448 bpf_prog_get_stats(prog, &stats);
3449 info.run_time_ns = stats.nsecs;
3450 info.run_cnt = stats.cnt;
3451 info.recursion_misses = stats.misses;
3453 if (!bpf_capable()) {
3454 info.jited_prog_len = 0;
3455 info.xlated_prog_len = 0;
3456 info.nr_jited_ksyms = 0;
3457 info.nr_jited_func_lens = 0;
3458 info.nr_func_info = 0;
3459 info.nr_line_info = 0;
3460 info.nr_jited_line_info = 0;
3464 ulen = info.xlated_prog_len;
3465 info.xlated_prog_len = bpf_prog_insn_size(prog);
3466 if (info.xlated_prog_len && ulen) {
3467 struct bpf_insn *insns_sanitized;
3470 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
3471 info.xlated_prog_insns = 0;
3474 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
3475 if (!insns_sanitized)
3477 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3478 ulen = min_t(u32, info.xlated_prog_len, ulen);
3479 fault = copy_to_user(uinsns, insns_sanitized, ulen);
3480 kfree(insns_sanitized);
3485 if (bpf_prog_is_dev_bound(prog->aux)) {
3486 err = bpf_prog_offload_info_fill(&info, prog);
3492 /* NOTE: the following code is supposed to be skipped for offload.
3493 * bpf_prog_offload_info_fill() is the place to fill similar fields
3496 ulen = info.jited_prog_len;
3497 if (prog->aux->func_cnt) {
3500 info.jited_prog_len = 0;
3501 for (i = 0; i < prog->aux->func_cnt; i++)
3502 info.jited_prog_len += prog->aux->func[i]->jited_len;
3504 info.jited_prog_len = prog->jited_len;
3507 if (info.jited_prog_len && ulen) {
3508 if (bpf_dump_raw_ok(file->f_cred)) {
3509 uinsns = u64_to_user_ptr(info.jited_prog_insns);
3510 ulen = min_t(u32, info.jited_prog_len, ulen);
3512 /* for multi-function programs, copy the JITed
3513 * instructions for all the functions
3515 if (prog->aux->func_cnt) {
3520 for (i = 0; i < prog->aux->func_cnt; i++) {
3521 len = prog->aux->func[i]->jited_len;
3522 len = min_t(u32, len, free);
3523 img = (u8 *) prog->aux->func[i]->bpf_func;
3524 if (copy_to_user(uinsns, img, len))
3532 if (copy_to_user(uinsns, prog->bpf_func, ulen))
3536 info.jited_prog_insns = 0;
3540 ulen = info.nr_jited_ksyms;
3541 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
3543 if (bpf_dump_raw_ok(file->f_cred)) {
3544 unsigned long ksym_addr;
3545 u64 __user *user_ksyms;
3548 /* copy the address of the kernel symbol
3549 * corresponding to each function
3551 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
3552 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
3553 if (prog->aux->func_cnt) {
3554 for (i = 0; i < ulen; i++) {
3555 ksym_addr = (unsigned long)
3556 prog->aux->func[i]->bpf_func;
3557 if (put_user((u64) ksym_addr,
3562 ksym_addr = (unsigned long) prog->bpf_func;
3563 if (put_user((u64) ksym_addr, &user_ksyms[0]))
3567 info.jited_ksyms = 0;
3571 ulen = info.nr_jited_func_lens;
3572 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
3574 if (bpf_dump_raw_ok(file->f_cred)) {
3575 u32 __user *user_lens;
3578 /* copy the JITed image lengths for each function */
3579 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
3580 user_lens = u64_to_user_ptr(info.jited_func_lens);
3581 if (prog->aux->func_cnt) {
3582 for (i = 0; i < ulen; i++) {
3584 prog->aux->func[i]->jited_len;
3585 if (put_user(func_len, &user_lens[i]))
3589 func_len = prog->jited_len;
3590 if (put_user(func_len, &user_lens[0]))
3594 info.jited_func_lens = 0;
3599 info.btf_id = btf_obj_id(prog->aux->btf);
3601 ulen = info.nr_func_info;
3602 info.nr_func_info = prog->aux->func_info_cnt;
3603 if (info.nr_func_info && ulen) {
3604 char __user *user_finfo;
3606 user_finfo = u64_to_user_ptr(info.func_info);
3607 ulen = min_t(u32, info.nr_func_info, ulen);
3608 if (copy_to_user(user_finfo, prog->aux->func_info,
3609 info.func_info_rec_size * ulen))
3613 ulen = info.nr_line_info;
3614 info.nr_line_info = prog->aux->nr_linfo;
3615 if (info.nr_line_info && ulen) {
3616 __u8 __user *user_linfo;
3618 user_linfo = u64_to_user_ptr(info.line_info);
3619 ulen = min_t(u32, info.nr_line_info, ulen);
3620 if (copy_to_user(user_linfo, prog->aux->linfo,
3621 info.line_info_rec_size * ulen))
3625 ulen = info.nr_jited_line_info;
3626 if (prog->aux->jited_linfo)
3627 info.nr_jited_line_info = prog->aux->nr_linfo;
3629 info.nr_jited_line_info = 0;
3630 if (info.nr_jited_line_info && ulen) {
3631 if (bpf_dump_raw_ok(file->f_cred)) {
3632 __u64 __user *user_linfo;
3635 user_linfo = u64_to_user_ptr(info.jited_line_info);
3636 ulen = min_t(u32, info.nr_jited_line_info, ulen);
3637 for (i = 0; i < ulen; i++) {
3638 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
3643 info.jited_line_info = 0;
3647 ulen = info.nr_prog_tags;
3648 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
3650 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
3653 user_prog_tags = u64_to_user_ptr(info.prog_tags);
3654 ulen = min_t(u32, info.nr_prog_tags, ulen);
3655 if (prog->aux->func_cnt) {
3656 for (i = 0; i < ulen; i++) {
3657 if (copy_to_user(user_prog_tags[i],
3658 prog->aux->func[i]->tag,
3663 if (copy_to_user(user_prog_tags[0],
3664 prog->tag, BPF_TAG_SIZE))
3670 if (copy_to_user(uinfo, &info, info_len) ||
3671 put_user(info_len, &uattr->info.info_len))
3677 static int bpf_map_get_info_by_fd(struct file *file,
3678 struct bpf_map *map,
3679 const union bpf_attr *attr,
3680 union bpf_attr __user *uattr)
3682 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3683 struct bpf_map_info info;
3684 u32 info_len = attr->info.info_len;
3687 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3690 info_len = min_t(u32, sizeof(info), info_len);
3692 memset(&info, 0, sizeof(info));
3693 info.type = map->map_type;
3695 info.key_size = map->key_size;
3696 info.value_size = map->value_size;
3697 info.max_entries = map->max_entries;
3698 info.map_flags = map->map_flags;
3699 memcpy(info.name, map->name, sizeof(map->name));
3702 info.btf_id = btf_obj_id(map->btf);
3703 info.btf_key_type_id = map->btf_key_type_id;
3704 info.btf_value_type_id = map->btf_value_type_id;
3706 info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
3708 if (bpf_map_is_dev_bound(map)) {
3709 err = bpf_map_offload_info_fill(&info, map);
3714 if (copy_to_user(uinfo, &info, info_len) ||
3715 put_user(info_len, &uattr->info.info_len))
3721 static int bpf_btf_get_info_by_fd(struct file *file,
3723 const union bpf_attr *attr,
3724 union bpf_attr __user *uattr)
3726 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3727 u32 info_len = attr->info.info_len;
3730 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
3734 return btf_get_info_by_fd(btf, attr, uattr);
3737 static int bpf_link_get_info_by_fd(struct file *file,
3738 struct bpf_link *link,
3739 const union bpf_attr *attr,
3740 union bpf_attr __user *uattr)
3742 struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3743 struct bpf_link_info info;
3744 u32 info_len = attr->info.info_len;
3747 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3750 info_len = min_t(u32, sizeof(info), info_len);
3752 memset(&info, 0, sizeof(info));
3753 if (copy_from_user(&info, uinfo, info_len))
3756 info.type = link->type;
3758 info.prog_id = link->prog->aux->id;
3760 if (link->ops->fill_link_info) {
3761 err = link->ops->fill_link_info(link, &info);
3766 if (copy_to_user(uinfo, &info, info_len) ||
3767 put_user(info_len, &uattr->info.info_len))
3774 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
3776 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
3777 union bpf_attr __user *uattr)
3779 int ufd = attr->info.bpf_fd;
3783 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
3790 if (f.file->f_op == &bpf_prog_fops)
3791 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
3793 else if (f.file->f_op == &bpf_map_fops)
3794 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
3796 else if (f.file->f_op == &btf_fops)
3797 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
3798 else if (f.file->f_op == &bpf_link_fops)
3799 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
3808 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
3810 static int bpf_btf_load(const union bpf_attr *attr)
3812 if (CHECK_ATTR(BPF_BTF_LOAD))
3818 return btf_new_fd(attr);
3821 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
3823 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
3825 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
3828 if (!capable(CAP_SYS_ADMIN))
3831 return btf_get_fd_by_id(attr->btf_id);
3834 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
3835 union bpf_attr __user *uattr,
3836 u32 prog_id, u32 fd_type,
3837 const char *buf, u64 probe_offset,
3840 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
3841 u32 len = buf ? strlen(buf) : 0, input_len;
3844 if (put_user(len, &uattr->task_fd_query.buf_len))
3846 input_len = attr->task_fd_query.buf_len;
3847 if (input_len && ubuf) {
3849 /* nothing to copy, just make ubuf NULL terminated */
3852 if (put_user(zero, ubuf))
3854 } else if (input_len >= len + 1) {
3855 /* ubuf can hold the string with NULL terminator */
3856 if (copy_to_user(ubuf, buf, len + 1))
3859 /* ubuf cannot hold the string with NULL terminator,
3860 * do a partial copy with NULL terminator.
3865 if (copy_to_user(ubuf, buf, input_len - 1))
3867 if (put_user(zero, ubuf + input_len - 1))
3872 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
3873 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
3874 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
3875 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
3881 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
3883 static int bpf_task_fd_query(const union bpf_attr *attr,
3884 union bpf_attr __user *uattr)
3886 pid_t pid = attr->task_fd_query.pid;
3887 u32 fd = attr->task_fd_query.fd;
3888 const struct perf_event *event;
3889 struct task_struct *task;
3893 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
3896 if (!capable(CAP_SYS_ADMIN))
3899 if (attr->task_fd_query.flags != 0)
3902 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3907 file = fget_task(task, fd);
3908 put_task_struct(task);
3912 if (file->f_op == &bpf_link_fops) {
3913 struct bpf_link *link = file->private_data;
3915 if (link->ops == &bpf_raw_tp_link_lops) {
3916 struct bpf_raw_tp_link *raw_tp =
3917 container_of(link, struct bpf_raw_tp_link, link);
3918 struct bpf_raw_event_map *btp = raw_tp->btp;
3920 err = bpf_task_fd_query_copy(attr, uattr,
3921 raw_tp->link.prog->aux->id,
3922 BPF_FD_TYPE_RAW_TRACEPOINT,
3923 btp->tp->name, 0, 0);
3929 event = perf_get_event(file);
3930 if (!IS_ERR(event)) {
3931 u64 probe_offset, probe_addr;
3932 u32 prog_id, fd_type;
3935 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
3936 &buf, &probe_offset,
3939 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
3953 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
3955 #define BPF_DO_BATCH(fn) \
3961 err = fn(map, attr, uattr); \
3964 static int bpf_map_do_batch(const union bpf_attr *attr,
3965 union bpf_attr __user *uattr,
3968 struct bpf_map *map;
3972 if (CHECK_ATTR(BPF_MAP_BATCH))
3975 ufd = attr->batch.map_fd;
3977 map = __bpf_map_get(f);
3979 return PTR_ERR(map);
3981 if ((cmd == BPF_MAP_LOOKUP_BATCH ||
3982 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH) &&
3983 !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
3988 if (cmd != BPF_MAP_LOOKUP_BATCH &&
3989 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
3994 if (cmd == BPF_MAP_LOOKUP_BATCH)
3995 BPF_DO_BATCH(map->ops->map_lookup_batch);
3996 else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
3997 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
3998 else if (cmd == BPF_MAP_UPDATE_BATCH)
3999 BPF_DO_BATCH(map->ops->map_update_batch);
4001 BPF_DO_BATCH(map->ops->map_delete_batch);
4008 static int tracing_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
4010 if (attr->link_create.attach_type != prog->expected_attach_type)
4013 if (prog->expected_attach_type == BPF_TRACE_ITER)
4014 return bpf_iter_link_attach(attr, prog);
4015 else if (prog->type == BPF_PROG_TYPE_EXT)
4016 return bpf_tracing_prog_attach(prog,
4017 attr->link_create.target_fd,
4018 attr->link_create.target_btf_id);
4022 #define BPF_LINK_CREATE_LAST_FIELD link_create.iter_info_len
4023 static int link_create(union bpf_attr *attr)
4025 enum bpf_prog_type ptype;
4026 struct bpf_prog *prog;
4029 if (CHECK_ATTR(BPF_LINK_CREATE))
4032 prog = bpf_prog_get(attr->link_create.prog_fd);
4034 return PTR_ERR(prog);
4036 ret = bpf_prog_attach_check_attach_type(prog,
4037 attr->link_create.attach_type);
4041 if (prog->type == BPF_PROG_TYPE_EXT) {
4042 ret = tracing_bpf_link_attach(attr, prog);
4046 ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4047 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4053 case BPF_PROG_TYPE_CGROUP_SKB:
4054 case BPF_PROG_TYPE_CGROUP_SOCK:
4055 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4056 case BPF_PROG_TYPE_SOCK_OPS:
4057 case BPF_PROG_TYPE_CGROUP_DEVICE:
4058 case BPF_PROG_TYPE_CGROUP_SYSCTL:
4059 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4060 ret = cgroup_bpf_link_attach(attr, prog);
4062 case BPF_PROG_TYPE_TRACING:
4063 ret = tracing_bpf_link_attach(attr, prog);
4065 case BPF_PROG_TYPE_FLOW_DISSECTOR:
4066 case BPF_PROG_TYPE_SK_LOOKUP:
4067 ret = netns_bpf_link_create(attr, prog);
4070 case BPF_PROG_TYPE_XDP:
4071 ret = bpf_xdp_link_attach(attr, prog);
4084 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4086 static int link_update(union bpf_attr *attr)
4088 struct bpf_prog *old_prog = NULL, *new_prog;
4089 struct bpf_link *link;
4093 if (CHECK_ATTR(BPF_LINK_UPDATE))
4096 flags = attr->link_update.flags;
4097 if (flags & ~BPF_F_REPLACE)
4100 link = bpf_link_get_from_fd(attr->link_update.link_fd);
4102 return PTR_ERR(link);
4104 new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4105 if (IS_ERR(new_prog)) {
4106 ret = PTR_ERR(new_prog);
4110 if (flags & BPF_F_REPLACE) {
4111 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4112 if (IS_ERR(old_prog)) {
4113 ret = PTR_ERR(old_prog);
4117 } else if (attr->link_update.old_prog_fd) {
4122 if (link->ops->update_prog)
4123 ret = link->ops->update_prog(link, new_prog, old_prog);
4129 bpf_prog_put(old_prog);
4131 bpf_prog_put(new_prog);
4137 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4139 static int link_detach(union bpf_attr *attr)
4141 struct bpf_link *link;
4144 if (CHECK_ATTR(BPF_LINK_DETACH))
4147 link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4149 return PTR_ERR(link);
4151 if (link->ops->detach)
4152 ret = link->ops->detach(link);
4160 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4162 return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4165 struct bpf_link *bpf_link_by_id(u32 id)
4167 struct bpf_link *link;
4170 return ERR_PTR(-ENOENT);
4172 spin_lock_bh(&link_idr_lock);
4173 /* before link is "settled", ID is 0, pretend it doesn't exist yet */
4174 link = idr_find(&link_idr, id);
4177 link = bpf_link_inc_not_zero(link);
4179 link = ERR_PTR(-EAGAIN);
4181 link = ERR_PTR(-ENOENT);
4183 spin_unlock_bh(&link_idr_lock);
4187 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4189 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4191 struct bpf_link *link;
4192 u32 id = attr->link_id;
4195 if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4198 if (!capable(CAP_SYS_ADMIN))
4201 link = bpf_link_by_id(id);
4203 return PTR_ERR(link);
4205 fd = bpf_link_new_fd(link);
4212 DEFINE_MUTEX(bpf_stats_enabled_mutex);
4214 static int bpf_stats_release(struct inode *inode, struct file *file)
4216 mutex_lock(&bpf_stats_enabled_mutex);
4217 static_key_slow_dec(&bpf_stats_enabled_key.key);
4218 mutex_unlock(&bpf_stats_enabled_mutex);
4222 static const struct file_operations bpf_stats_fops = {
4223 .release = bpf_stats_release,
4226 static int bpf_enable_runtime_stats(void)
4230 mutex_lock(&bpf_stats_enabled_mutex);
4232 /* Set a very high limit to avoid overflow */
4233 if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4234 mutex_unlock(&bpf_stats_enabled_mutex);
4238 fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4240 static_key_slow_inc(&bpf_stats_enabled_key.key);
4242 mutex_unlock(&bpf_stats_enabled_mutex);
4246 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4248 static int bpf_enable_stats(union bpf_attr *attr)
4251 if (CHECK_ATTR(BPF_ENABLE_STATS))
4254 if (!capable(CAP_SYS_ADMIN))
4257 switch (attr->enable_stats.type) {
4258 case BPF_STATS_RUN_TIME:
4259 return bpf_enable_runtime_stats();
4266 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4268 static int bpf_iter_create(union bpf_attr *attr)
4270 struct bpf_link *link;
4273 if (CHECK_ATTR(BPF_ITER_CREATE))
4276 if (attr->iter_create.flags)
4279 link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4281 return PTR_ERR(link);
4283 err = bpf_iter_new_fd(link);
4289 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4291 static int bpf_prog_bind_map(union bpf_attr *attr)
4293 struct bpf_prog *prog;
4294 struct bpf_map *map;
4295 struct bpf_map **used_maps_old, **used_maps_new;
4298 if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4301 if (attr->prog_bind_map.flags)
4304 prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4306 return PTR_ERR(prog);
4308 map = bpf_map_get(attr->prog_bind_map.map_fd);
4314 mutex_lock(&prog->aux->used_maps_mutex);
4316 used_maps_old = prog->aux->used_maps;
4318 for (i = 0; i < prog->aux->used_map_cnt; i++)
4319 if (used_maps_old[i] == map) {
4324 used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4325 sizeof(used_maps_new[0]),
4327 if (!used_maps_new) {
4332 memcpy(used_maps_new, used_maps_old,
4333 sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4334 used_maps_new[prog->aux->used_map_cnt] = map;
4336 prog->aux->used_map_cnt++;
4337 prog->aux->used_maps = used_maps_new;
4339 kfree(used_maps_old);
4342 mutex_unlock(&prog->aux->used_maps_mutex);
4351 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
4353 union bpf_attr attr;
4356 if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
4359 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
4362 size = min_t(u32, size, sizeof(attr));
4364 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
4365 memset(&attr, 0, sizeof(attr));
4366 if (copy_from_user(&attr, uattr, size) != 0)
4369 err = security_bpf(cmd, &attr, size);
4374 case BPF_MAP_CREATE:
4375 err = map_create(&attr);
4377 case BPF_MAP_LOOKUP_ELEM:
4378 err = map_lookup_elem(&attr);
4380 case BPF_MAP_UPDATE_ELEM:
4381 err = map_update_elem(&attr);
4383 case BPF_MAP_DELETE_ELEM:
4384 err = map_delete_elem(&attr);
4386 case BPF_MAP_GET_NEXT_KEY:
4387 err = map_get_next_key(&attr);
4389 case BPF_MAP_FREEZE:
4390 err = map_freeze(&attr);
4393 err = bpf_prog_load(&attr, uattr);
4396 err = bpf_obj_pin(&attr);
4399 err = bpf_obj_get(&attr);
4401 case BPF_PROG_ATTACH:
4402 err = bpf_prog_attach(&attr);
4404 case BPF_PROG_DETACH:
4405 err = bpf_prog_detach(&attr);
4407 case BPF_PROG_QUERY:
4408 err = bpf_prog_query(&attr, uattr);
4410 case BPF_PROG_TEST_RUN:
4411 err = bpf_prog_test_run(&attr, uattr);
4413 case BPF_PROG_GET_NEXT_ID:
4414 err = bpf_obj_get_next_id(&attr, uattr,
4415 &prog_idr, &prog_idr_lock);
4417 case BPF_MAP_GET_NEXT_ID:
4418 err = bpf_obj_get_next_id(&attr, uattr,
4419 &map_idr, &map_idr_lock);
4421 case BPF_BTF_GET_NEXT_ID:
4422 err = bpf_obj_get_next_id(&attr, uattr,
4423 &btf_idr, &btf_idr_lock);
4425 case BPF_PROG_GET_FD_BY_ID:
4426 err = bpf_prog_get_fd_by_id(&attr);
4428 case BPF_MAP_GET_FD_BY_ID:
4429 err = bpf_map_get_fd_by_id(&attr);
4431 case BPF_OBJ_GET_INFO_BY_FD:
4432 err = bpf_obj_get_info_by_fd(&attr, uattr);
4434 case BPF_RAW_TRACEPOINT_OPEN:
4435 err = bpf_raw_tracepoint_open(&attr);
4438 err = bpf_btf_load(&attr);
4440 case BPF_BTF_GET_FD_BY_ID:
4441 err = bpf_btf_get_fd_by_id(&attr);
4443 case BPF_TASK_FD_QUERY:
4444 err = bpf_task_fd_query(&attr, uattr);
4446 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
4447 err = map_lookup_and_delete_elem(&attr);
4449 case BPF_MAP_LOOKUP_BATCH:
4450 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_LOOKUP_BATCH);
4452 case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
4453 err = bpf_map_do_batch(&attr, uattr,
4454 BPF_MAP_LOOKUP_AND_DELETE_BATCH);
4456 case BPF_MAP_UPDATE_BATCH:
4457 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_UPDATE_BATCH);
4459 case BPF_MAP_DELETE_BATCH:
4460 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_DELETE_BATCH);
4462 case BPF_LINK_CREATE:
4463 err = link_create(&attr);
4465 case BPF_LINK_UPDATE:
4466 err = link_update(&attr);
4468 case BPF_LINK_GET_FD_BY_ID:
4469 err = bpf_link_get_fd_by_id(&attr);
4471 case BPF_LINK_GET_NEXT_ID:
4472 err = bpf_obj_get_next_id(&attr, uattr,
4473 &link_idr, &link_idr_lock);
4475 case BPF_ENABLE_STATS:
4476 err = bpf_enable_stats(&attr);
4478 case BPF_ITER_CREATE:
4479 err = bpf_iter_create(&attr);
4481 case BPF_LINK_DETACH:
4482 err = link_detach(&attr);
4484 case BPF_PROG_BIND_MAP:
4485 err = bpf_prog_bind_map(&attr);