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>
8 #include <linux/syscalls.h>
9 #include <linux/slab.h>
10 #include <linux/sched/signal.h>
11 #include <linux/vmalloc.h>
12 #include <linux/mmzone.h>
13 #include <linux/anon_inodes.h>
14 #include <linux/fdtable.h>
15 #include <linux/file.h>
17 #include <linux/license.h>
18 #include <linux/filter.h>
19 #include <linux/version.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/bpf_lsm.h>
30 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
31 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
32 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
33 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
34 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
35 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
38 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
40 DEFINE_PER_CPU(int, bpf_prog_active);
41 static DEFINE_IDR(prog_idr);
42 static DEFINE_SPINLOCK(prog_idr_lock);
43 static DEFINE_IDR(map_idr);
44 static DEFINE_SPINLOCK(map_idr_lock);
46 int sysctl_unprivileged_bpf_disabled __read_mostly;
48 static const struct bpf_map_ops * const bpf_map_types[] = {
49 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
50 #define BPF_MAP_TYPE(_id, _ops) \
52 #include <linux/bpf_types.h>
58 * If we're handed a bigger struct than we know of, ensure all the unknown bits
59 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
60 * we don't know about yet.
62 * There is a ToCToU between this function call and the following
63 * copy_from_user() call. However, this is not a concern since this function is
64 * meant to be a future-proofing of bits.
66 int bpf_check_uarg_tail_zero(void __user *uaddr,
70 unsigned char __user *addr;
71 unsigned char __user *end;
75 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
78 if (unlikely(!access_ok(uaddr, actual_size)))
81 if (actual_size <= expected_size)
84 addr = uaddr + expected_size;
85 end = uaddr + actual_size;
87 for (; addr < end; addr++) {
88 err = get_user(val, addr);
98 const struct bpf_map_ops bpf_map_offload_ops = {
99 .map_alloc = bpf_map_offload_map_alloc,
100 .map_free = bpf_map_offload_map_free,
101 .map_check_btf = map_check_no_btf,
104 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
106 const struct bpf_map_ops *ops;
107 u32 type = attr->map_type;
111 if (type >= ARRAY_SIZE(bpf_map_types))
112 return ERR_PTR(-EINVAL);
113 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
114 ops = bpf_map_types[type];
116 return ERR_PTR(-EINVAL);
118 if (ops->map_alloc_check) {
119 err = ops->map_alloc_check(attr);
123 if (attr->map_ifindex)
124 ops = &bpf_map_offload_ops;
125 map = ops->map_alloc(attr);
129 map->map_type = type;
133 static u32 bpf_map_value_size(struct bpf_map *map)
135 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
136 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
137 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
138 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
139 return round_up(map->value_size, 8) * num_possible_cpus();
140 else if (IS_FD_MAP(map))
143 return map->value_size;
146 static void maybe_wait_bpf_programs(struct bpf_map *map)
148 /* Wait for any running BPF programs to complete so that
149 * userspace, when we return to it, knows that all programs
150 * that could be running use the new map value.
152 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
153 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
157 static int bpf_map_update_value(struct bpf_map *map, struct fd f, void *key,
158 void *value, __u64 flags)
162 /* Need to create a kthread, thus must support schedule */
163 if (bpf_map_is_dev_bound(map)) {
164 return bpf_map_offload_update_elem(map, key, value, flags);
165 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
166 map->map_type == BPF_MAP_TYPE_SOCKHASH ||
167 map->map_type == BPF_MAP_TYPE_SOCKMAP ||
168 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
169 return map->ops->map_update_elem(map, key, value, flags);
170 } else if (IS_FD_PROG_ARRAY(map)) {
171 return bpf_fd_array_map_update_elem(map, f.file, key, value,
175 bpf_disable_instrumentation();
176 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
177 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
178 err = bpf_percpu_hash_update(map, key, value, flags);
179 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
180 err = bpf_percpu_array_update(map, key, value, flags);
181 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
182 err = bpf_percpu_cgroup_storage_update(map, key, value,
184 } else if (IS_FD_ARRAY(map)) {
186 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
189 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
191 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
194 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
195 /* rcu_read_lock() is not needed */
196 err = bpf_fd_reuseport_array_update_elem(map, key, value,
198 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
199 map->map_type == BPF_MAP_TYPE_STACK) {
200 err = map->ops->map_push_elem(map, value, flags);
203 err = map->ops->map_update_elem(map, key, value, flags);
206 bpf_enable_instrumentation();
207 maybe_wait_bpf_programs(map);
212 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
218 if (bpf_map_is_dev_bound(map))
219 return bpf_map_offload_lookup_elem(map, key, value);
221 bpf_disable_instrumentation();
222 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
223 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
224 err = bpf_percpu_hash_copy(map, key, value);
225 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
226 err = bpf_percpu_array_copy(map, key, value);
227 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
228 err = bpf_percpu_cgroup_storage_copy(map, key, value);
229 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
230 err = bpf_stackmap_copy(map, key, value);
231 } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
232 err = bpf_fd_array_map_lookup_elem(map, key, value);
233 } else if (IS_FD_HASH(map)) {
234 err = bpf_fd_htab_map_lookup_elem(map, key, value);
235 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
236 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
237 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
238 map->map_type == BPF_MAP_TYPE_STACK) {
239 err = map->ops->map_peek_elem(map, value);
240 } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
241 /* struct_ops map requires directly updating "value" */
242 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
245 if (map->ops->map_lookup_elem_sys_only)
246 ptr = map->ops->map_lookup_elem_sys_only(map, key);
248 ptr = map->ops->map_lookup_elem(map, key);
255 if (flags & BPF_F_LOCK)
256 /* lock 'ptr' and copy everything but lock */
257 copy_map_value_locked(map, value, ptr, true);
259 copy_map_value(map, value, ptr);
260 /* mask lock, since value wasn't zero inited */
261 check_and_init_map_lock(map, value);
266 bpf_enable_instrumentation();
267 maybe_wait_bpf_programs(map);
272 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
274 /* We really just want to fail instead of triggering OOM killer
275 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
276 * which is used for lower order allocation requests.
278 * It has been observed that higher order allocation requests done by
279 * vmalloc with __GFP_NORETRY being set might fail due to not trying
280 * to reclaim memory from the page cache, thus we set
281 * __GFP_RETRY_MAYFAIL to avoid such situations.
284 const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
287 if (size >= SIZE_MAX)
290 /* kmalloc()'ed memory can't be mmap()'ed */
291 if (!mmapable && size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
292 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
298 BUG_ON(!PAGE_ALIGNED(size));
299 return vmalloc_user_node_flags(size, numa_node, GFP_KERNEL |
300 __GFP_RETRY_MAYFAIL | flags);
302 return __vmalloc_node_flags_caller(size, numa_node,
303 GFP_KERNEL | __GFP_RETRY_MAYFAIL |
304 flags, __builtin_return_address(0));
307 void *bpf_map_area_alloc(u64 size, int numa_node)
309 return __bpf_map_area_alloc(size, numa_node, false);
312 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
314 return __bpf_map_area_alloc(size, numa_node, true);
317 void bpf_map_area_free(void *area)
322 static u32 bpf_map_flags_retain_permanent(u32 flags)
324 /* Some map creation flags are not tied to the map object but
325 * rather to the map fd instead, so they have no meaning upon
326 * map object inspection since multiple file descriptors with
327 * different (access) properties can exist here. Thus, given
328 * this has zero meaning for the map itself, lets clear these
331 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
334 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
336 map->map_type = attr->map_type;
337 map->key_size = attr->key_size;
338 map->value_size = attr->value_size;
339 map->max_entries = attr->max_entries;
340 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
341 map->numa_node = bpf_map_attr_numa_node(attr);
344 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
346 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
348 if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
349 atomic_long_sub(pages, &user->locked_vm);
355 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
358 atomic_long_sub(pages, &user->locked_vm);
361 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size)
363 u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
364 struct user_struct *user;
367 if (size >= U32_MAX - PAGE_SIZE)
370 user = get_current_user();
371 ret = bpf_charge_memlock(user, pages);
383 void bpf_map_charge_finish(struct bpf_map_memory *mem)
385 bpf_uncharge_memlock(mem->user, mem->pages);
389 void bpf_map_charge_move(struct bpf_map_memory *dst,
390 struct bpf_map_memory *src)
394 /* Make sure src will not be used for the redundant uncharging. */
395 memset(src, 0, sizeof(struct bpf_map_memory));
398 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
402 ret = bpf_charge_memlock(map->memory.user, pages);
405 map->memory.pages += pages;
409 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
411 bpf_uncharge_memlock(map->memory.user, pages);
412 map->memory.pages -= pages;
415 static int bpf_map_alloc_id(struct bpf_map *map)
419 idr_preload(GFP_KERNEL);
420 spin_lock_bh(&map_idr_lock);
421 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
424 spin_unlock_bh(&map_idr_lock);
427 if (WARN_ON_ONCE(!id))
430 return id > 0 ? 0 : id;
433 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
437 /* Offloaded maps are removed from the IDR store when their device
438 * disappears - even if someone holds an fd to them they are unusable,
439 * the memory is gone, all ops will fail; they are simply waiting for
440 * refcnt to drop to be freed.
446 spin_lock_irqsave(&map_idr_lock, flags);
448 __acquire(&map_idr_lock);
450 idr_remove(&map_idr, map->id);
454 spin_unlock_irqrestore(&map_idr_lock, flags);
456 __release(&map_idr_lock);
459 /* called from workqueue */
460 static void bpf_map_free_deferred(struct work_struct *work)
462 struct bpf_map *map = container_of(work, struct bpf_map, work);
463 struct bpf_map_memory mem;
465 bpf_map_charge_move(&mem, &map->memory);
466 security_bpf_map_free(map);
467 /* implementation dependent freeing */
468 map->ops->map_free(map);
469 bpf_map_charge_finish(&mem);
472 static void bpf_map_put_uref(struct bpf_map *map)
474 if (atomic64_dec_and_test(&map->usercnt)) {
475 if (map->ops->map_release_uref)
476 map->ops->map_release_uref(map);
480 /* decrement map refcnt and schedule it for freeing via workqueue
481 * (unrelying map implementation ops->map_free() might sleep)
483 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
485 if (atomic64_dec_and_test(&map->refcnt)) {
486 /* bpf_map_free_id() must be called first */
487 bpf_map_free_id(map, do_idr_lock);
489 INIT_WORK(&map->work, bpf_map_free_deferred);
490 schedule_work(&map->work);
494 void bpf_map_put(struct bpf_map *map)
496 __bpf_map_put(map, true);
498 EXPORT_SYMBOL_GPL(bpf_map_put);
500 void bpf_map_put_with_uref(struct bpf_map *map)
502 bpf_map_put_uref(map);
506 static int bpf_map_release(struct inode *inode, struct file *filp)
508 struct bpf_map *map = filp->private_data;
510 if (map->ops->map_release)
511 map->ops->map_release(map, filp);
513 bpf_map_put_with_uref(map);
517 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
519 fmode_t mode = f.file->f_mode;
521 /* Our file permissions may have been overridden by global
522 * map permissions facing syscall side.
524 if (READ_ONCE(map->frozen))
525 mode &= ~FMODE_CAN_WRITE;
529 #ifdef CONFIG_PROC_FS
530 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
532 const struct bpf_map *map = filp->private_data;
533 const struct bpf_array *array;
534 u32 type = 0, jited = 0;
536 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
537 array = container_of(map, struct bpf_array, map);
538 type = array->aux->type;
539 jited = array->aux->jited;
556 map->memory.pages * 1ULL << PAGE_SHIFT,
558 READ_ONCE(map->frozen));
560 seq_printf(m, "owner_prog_type:\t%u\n", type);
561 seq_printf(m, "owner_jited:\t%u\n", jited);
566 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
569 /* We need this handler such that alloc_file() enables
570 * f_mode with FMODE_CAN_READ.
575 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
576 size_t siz, loff_t *ppos)
578 /* We need this handler such that alloc_file() enables
579 * f_mode with FMODE_CAN_WRITE.
584 /* called for any extra memory-mapped regions (except initial) */
585 static void bpf_map_mmap_open(struct vm_area_struct *vma)
587 struct bpf_map *map = vma->vm_file->private_data;
589 if (vma->vm_flags & VM_MAYWRITE) {
590 mutex_lock(&map->freeze_mutex);
592 mutex_unlock(&map->freeze_mutex);
596 /* called for all unmapped memory region (including initial) */
597 static void bpf_map_mmap_close(struct vm_area_struct *vma)
599 struct bpf_map *map = vma->vm_file->private_data;
601 if (vma->vm_flags & VM_MAYWRITE) {
602 mutex_lock(&map->freeze_mutex);
604 mutex_unlock(&map->freeze_mutex);
608 static const struct vm_operations_struct bpf_map_default_vmops = {
609 .open = bpf_map_mmap_open,
610 .close = bpf_map_mmap_close,
613 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
615 struct bpf_map *map = filp->private_data;
618 if (!map->ops->map_mmap || map_value_has_spin_lock(map))
621 if (!(vma->vm_flags & VM_SHARED))
624 mutex_lock(&map->freeze_mutex);
626 if ((vma->vm_flags & VM_WRITE) && map->frozen) {
631 /* set default open/close callbacks */
632 vma->vm_ops = &bpf_map_default_vmops;
633 vma->vm_private_data = map;
634 vma->vm_flags &= ~VM_MAYEXEC;
635 if (!(vma->vm_flags & VM_WRITE))
636 /* disallow re-mapping with PROT_WRITE */
637 vma->vm_flags &= ~VM_MAYWRITE;
639 err = map->ops->map_mmap(map, vma);
643 if (vma->vm_flags & VM_MAYWRITE)
646 mutex_unlock(&map->freeze_mutex);
650 const struct file_operations bpf_map_fops = {
651 #ifdef CONFIG_PROC_FS
652 .show_fdinfo = bpf_map_show_fdinfo,
654 .release = bpf_map_release,
655 .read = bpf_dummy_read,
656 .write = bpf_dummy_write,
657 .mmap = bpf_map_mmap,
660 int bpf_map_new_fd(struct bpf_map *map, int flags)
664 ret = security_bpf_map(map, OPEN_FMODE(flags));
668 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
672 int bpf_get_file_flag(int flags)
674 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
676 if (flags & BPF_F_RDONLY)
678 if (flags & BPF_F_WRONLY)
683 /* helper macro to check that unused fields 'union bpf_attr' are zero */
684 #define CHECK_ATTR(CMD) \
685 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
686 sizeof(attr->CMD##_LAST_FIELD), 0, \
688 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
689 sizeof(attr->CMD##_LAST_FIELD)) != NULL
691 /* dst and src must have at least "size" number of bytes.
692 * Return strlen on success and < 0 on error.
694 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
696 const char *end = src + size;
697 const char *orig_src = src;
699 memset(dst, 0, size);
700 /* Copy all isalnum(), '_' and '.' chars. */
701 while (src < end && *src) {
702 if (!isalnum(*src) &&
703 *src != '_' && *src != '.')
708 /* No '\0' found in "size" number of bytes */
712 return src - orig_src;
715 int map_check_no_btf(const struct bpf_map *map,
716 const struct btf *btf,
717 const struct btf_type *key_type,
718 const struct btf_type *value_type)
723 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
724 u32 btf_key_id, u32 btf_value_id)
726 const struct btf_type *key_type, *value_type;
727 u32 key_size, value_size;
730 /* Some maps allow key to be unspecified. */
732 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
733 if (!key_type || key_size != map->key_size)
736 key_type = btf_type_by_id(btf, 0);
737 if (!map->ops->map_check_btf)
741 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
742 if (!value_type || value_size != map->value_size)
745 map->spin_lock_off = btf_find_spin_lock(btf, value_type);
747 if (map_value_has_spin_lock(map)) {
748 if (map->map_flags & BPF_F_RDONLY_PROG)
750 if (map->map_type != BPF_MAP_TYPE_HASH &&
751 map->map_type != BPF_MAP_TYPE_ARRAY &&
752 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
753 map->map_type != BPF_MAP_TYPE_SK_STORAGE)
755 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
758 "verifier bug spin_lock_off %d value_size %d\n",
759 map->spin_lock_off, map->value_size);
764 if (map->ops->map_check_btf)
765 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
770 #define BPF_MAP_CREATE_LAST_FIELD btf_vmlinux_value_type_id
771 /* called via syscall */
772 static int map_create(union bpf_attr *attr)
774 int numa_node = bpf_map_attr_numa_node(attr);
775 struct bpf_map_memory mem;
780 err = CHECK_ATTR(BPF_MAP_CREATE);
784 if (attr->btf_vmlinux_value_type_id) {
785 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
786 attr->btf_key_type_id || attr->btf_value_type_id)
788 } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
792 f_flags = bpf_get_file_flag(attr->map_flags);
796 if (numa_node != NUMA_NO_NODE &&
797 ((unsigned int)numa_node >= nr_node_ids ||
798 !node_online(numa_node)))
801 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
802 map = find_and_alloc_map(attr);
806 err = bpf_obj_name_cpy(map->name, attr->map_name,
807 sizeof(attr->map_name));
811 atomic64_set(&map->refcnt, 1);
812 atomic64_set(&map->usercnt, 1);
813 mutex_init(&map->freeze_mutex);
815 map->spin_lock_off = -EINVAL;
816 if (attr->btf_key_type_id || attr->btf_value_type_id ||
817 /* Even the map's value is a kernel's struct,
818 * the bpf_prog.o must have BTF to begin with
819 * to figure out the corresponding kernel's
820 * counter part. Thus, attr->btf_fd has
823 attr->btf_vmlinux_value_type_id) {
826 btf = btf_get_by_fd(attr->btf_fd);
833 if (attr->btf_value_type_id) {
834 err = map_check_btf(map, btf, attr->btf_key_type_id,
835 attr->btf_value_type_id);
840 map->btf_key_type_id = attr->btf_key_type_id;
841 map->btf_value_type_id = attr->btf_value_type_id;
842 map->btf_vmlinux_value_type_id =
843 attr->btf_vmlinux_value_type_id;
846 err = security_bpf_map_alloc(map);
850 err = bpf_map_alloc_id(map);
854 err = bpf_map_new_fd(map, f_flags);
856 /* failed to allocate fd.
857 * bpf_map_put_with_uref() is needed because the above
858 * bpf_map_alloc_id() has published the map
859 * to the userspace and the userspace may
860 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
862 bpf_map_put_with_uref(map);
869 security_bpf_map_free(map);
872 bpf_map_charge_move(&mem, &map->memory);
873 map->ops->map_free(map);
874 bpf_map_charge_finish(&mem);
878 /* if error is returned, fd is released.
879 * On success caller should complete fd access with matching fdput()
881 struct bpf_map *__bpf_map_get(struct fd f)
884 return ERR_PTR(-EBADF);
885 if (f.file->f_op != &bpf_map_fops) {
887 return ERR_PTR(-EINVAL);
890 return f.file->private_data;
893 void bpf_map_inc(struct bpf_map *map)
895 atomic64_inc(&map->refcnt);
897 EXPORT_SYMBOL_GPL(bpf_map_inc);
899 void bpf_map_inc_with_uref(struct bpf_map *map)
901 atomic64_inc(&map->refcnt);
902 atomic64_inc(&map->usercnt);
904 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
906 struct bpf_map *bpf_map_get(u32 ufd)
908 struct fd f = fdget(ufd);
911 map = __bpf_map_get(f);
921 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
923 struct fd f = fdget(ufd);
926 map = __bpf_map_get(f);
930 bpf_map_inc_with_uref(map);
936 /* map_idr_lock should have been held */
937 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
941 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
943 return ERR_PTR(-ENOENT);
945 atomic64_inc(&map->usercnt);
950 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
952 spin_lock_bh(&map_idr_lock);
953 map = __bpf_map_inc_not_zero(map, false);
954 spin_unlock_bh(&map_idr_lock);
958 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
960 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
965 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
968 return memdup_user(ukey, key_size);
971 return ERR_PTR(-EINVAL);
976 /* last field in 'union bpf_attr' used by this command */
977 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
979 static int map_lookup_elem(union bpf_attr *attr)
981 void __user *ukey = u64_to_user_ptr(attr->key);
982 void __user *uvalue = u64_to_user_ptr(attr->value);
983 int ufd = attr->map_fd;
990 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
993 if (attr->flags & ~BPF_F_LOCK)
997 map = __bpf_map_get(f);
1000 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1005 if ((attr->flags & BPF_F_LOCK) &&
1006 !map_value_has_spin_lock(map)) {
1011 key = __bpf_copy_key(ukey, map->key_size);
1017 value_size = bpf_map_value_size(map);
1020 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1024 err = bpf_map_copy_value(map, key, value, attr->flags);
1029 if (copy_to_user(uvalue, value, value_size) != 0)
1044 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1046 static int map_update_elem(union bpf_attr *attr)
1048 void __user *ukey = u64_to_user_ptr(attr->key);
1049 void __user *uvalue = u64_to_user_ptr(attr->value);
1050 int ufd = attr->map_fd;
1051 struct bpf_map *map;
1057 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1061 map = __bpf_map_get(f);
1063 return PTR_ERR(map);
1064 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1069 if ((attr->flags & BPF_F_LOCK) &&
1070 !map_value_has_spin_lock(map)) {
1075 key = __bpf_copy_key(ukey, map->key_size);
1081 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1082 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
1083 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
1084 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
1085 value_size = round_up(map->value_size, 8) * num_possible_cpus();
1087 value_size = map->value_size;
1090 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1095 if (copy_from_user(value, uvalue, value_size) != 0)
1098 err = bpf_map_update_value(map, f, key, value, attr->flags);
1109 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1111 static int map_delete_elem(union bpf_attr *attr)
1113 void __user *ukey = u64_to_user_ptr(attr->key);
1114 int ufd = attr->map_fd;
1115 struct bpf_map *map;
1120 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1124 map = __bpf_map_get(f);
1126 return PTR_ERR(map);
1127 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1132 key = __bpf_copy_key(ukey, map->key_size);
1138 if (bpf_map_is_dev_bound(map)) {
1139 err = bpf_map_offload_delete_elem(map, key);
1141 } else if (IS_FD_PROG_ARRAY(map) ||
1142 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1143 /* These maps require sleepable context */
1144 err = map->ops->map_delete_elem(map, key);
1148 bpf_disable_instrumentation();
1150 err = map->ops->map_delete_elem(map, key);
1152 bpf_enable_instrumentation();
1153 maybe_wait_bpf_programs(map);
1161 /* last field in 'union bpf_attr' used by this command */
1162 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1164 static int map_get_next_key(union bpf_attr *attr)
1166 void __user *ukey = u64_to_user_ptr(attr->key);
1167 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1168 int ufd = attr->map_fd;
1169 struct bpf_map *map;
1170 void *key, *next_key;
1174 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1178 map = __bpf_map_get(f);
1180 return PTR_ERR(map);
1181 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1187 key = __bpf_copy_key(ukey, map->key_size);
1197 next_key = kmalloc(map->key_size, GFP_USER);
1201 if (bpf_map_is_dev_bound(map)) {
1202 err = bpf_map_offload_get_next_key(map, key, next_key);
1207 err = map->ops->map_get_next_key(map, key, next_key);
1214 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1228 int generic_map_delete_batch(struct bpf_map *map,
1229 const union bpf_attr *attr,
1230 union bpf_attr __user *uattr)
1232 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1237 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1240 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1241 !map_value_has_spin_lock(map)) {
1245 max_count = attr->batch.count;
1249 key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1253 for (cp = 0; cp < max_count; cp++) {
1255 if (copy_from_user(key, keys + cp * map->key_size,
1259 if (bpf_map_is_dev_bound(map)) {
1260 err = bpf_map_offload_delete_elem(map, key);
1264 bpf_disable_instrumentation();
1266 err = map->ops->map_delete_elem(map, key);
1268 bpf_enable_instrumentation();
1269 maybe_wait_bpf_programs(map);
1273 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1280 int generic_map_update_batch(struct bpf_map *map,
1281 const union bpf_attr *attr,
1282 union bpf_attr __user *uattr)
1284 void __user *values = u64_to_user_ptr(attr->batch.values);
1285 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1286 u32 value_size, cp, max_count;
1287 int ufd = attr->map_fd;
1293 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1296 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1297 !map_value_has_spin_lock(map)) {
1301 value_size = bpf_map_value_size(map);
1303 max_count = attr->batch.count;
1307 key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1311 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1317 for (cp = 0; cp < max_count; cp++) {
1319 if (copy_from_user(key, keys + cp * map->key_size,
1321 copy_from_user(value, values + cp * value_size, value_size))
1324 err = bpf_map_update_value(map, f, key, value,
1325 attr->batch.elem_flags);
1331 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1339 #define MAP_LOOKUP_RETRIES 3
1341 int generic_map_lookup_batch(struct bpf_map *map,
1342 const union bpf_attr *attr,
1343 union bpf_attr __user *uattr)
1345 void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1346 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1347 void __user *values = u64_to_user_ptr(attr->batch.values);
1348 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1349 void *buf, *buf_prevkey, *prev_key, *key, *value;
1350 int err, retry = MAP_LOOKUP_RETRIES;
1351 u32 value_size, cp, max_count;
1353 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1356 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1357 !map_value_has_spin_lock(map))
1360 value_size = bpf_map_value_size(map);
1362 max_count = attr->batch.count;
1366 if (put_user(0, &uattr->batch.count))
1369 buf_prevkey = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1373 buf = kmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1375 kvfree(buf_prevkey);
1381 if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1384 value = key + map->key_size;
1386 prev_key = buf_prevkey;
1388 for (cp = 0; cp < max_count;) {
1390 err = map->ops->map_get_next_key(map, prev_key, key);
1394 err = bpf_map_copy_value(map, key, value,
1395 attr->batch.elem_flags);
1397 if (err == -ENOENT) {
1409 if (copy_to_user(keys + cp * map->key_size, key,
1414 if (copy_to_user(values + cp * value_size, value, value_size)) {
1420 prev_key = buf_prevkey;
1422 swap(prev_key, key);
1423 retry = MAP_LOOKUP_RETRIES;
1430 if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1431 (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1440 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1442 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1444 void __user *ukey = u64_to_user_ptr(attr->key);
1445 void __user *uvalue = u64_to_user_ptr(attr->value);
1446 int ufd = attr->map_fd;
1447 struct bpf_map *map;
1453 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1457 map = __bpf_map_get(f);
1459 return PTR_ERR(map);
1460 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1465 key = __bpf_copy_key(ukey, map->key_size);
1471 value_size = map->value_size;
1474 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1478 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1479 map->map_type == BPF_MAP_TYPE_STACK) {
1480 err = map->ops->map_pop_elem(map, value);
1488 if (copy_to_user(uvalue, value, value_size) != 0)
1502 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1504 static int map_freeze(const union bpf_attr *attr)
1506 int err = 0, ufd = attr->map_fd;
1507 struct bpf_map *map;
1510 if (CHECK_ATTR(BPF_MAP_FREEZE))
1514 map = __bpf_map_get(f);
1516 return PTR_ERR(map);
1518 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1523 mutex_lock(&map->freeze_mutex);
1525 if (map->writecnt) {
1529 if (READ_ONCE(map->frozen)) {
1533 if (!capable(CAP_SYS_ADMIN)) {
1538 WRITE_ONCE(map->frozen, true);
1540 mutex_unlock(&map->freeze_mutex);
1545 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1546 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1547 [_id] = & _name ## _prog_ops,
1548 #define BPF_MAP_TYPE(_id, _ops)
1549 #include <linux/bpf_types.h>
1550 #undef BPF_PROG_TYPE
1554 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1556 const struct bpf_prog_ops *ops;
1558 if (type >= ARRAY_SIZE(bpf_prog_types))
1560 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1561 ops = bpf_prog_types[type];
1565 if (!bpf_prog_is_dev_bound(prog->aux))
1566 prog->aux->ops = ops;
1568 prog->aux->ops = &bpf_offload_prog_ops;
1579 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1580 [BPF_AUDIT_LOAD] = "LOAD",
1581 [BPF_AUDIT_UNLOAD] = "UNLOAD",
1584 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1586 struct audit_context *ctx = NULL;
1587 struct audit_buffer *ab;
1589 if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1591 if (audit_enabled == AUDIT_OFF)
1593 if (op == BPF_AUDIT_LOAD)
1594 ctx = audit_context();
1595 ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1598 audit_log_format(ab, "prog-id=%u op=%s",
1599 prog->aux->id, bpf_audit_str[op]);
1603 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1605 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1606 unsigned long user_bufs;
1609 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1610 if (user_bufs > memlock_limit) {
1611 atomic_long_sub(pages, &user->locked_vm);
1619 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1622 atomic_long_sub(pages, &user->locked_vm);
1625 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1627 struct user_struct *user = get_current_user();
1630 ret = __bpf_prog_charge(user, prog->pages);
1636 prog->aux->user = user;
1640 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1642 struct user_struct *user = prog->aux->user;
1644 __bpf_prog_uncharge(user, prog->pages);
1648 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1652 idr_preload(GFP_KERNEL);
1653 spin_lock_bh(&prog_idr_lock);
1654 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1657 spin_unlock_bh(&prog_idr_lock);
1660 /* id is in [1, INT_MAX) */
1661 if (WARN_ON_ONCE(!id))
1664 return id > 0 ? 0 : id;
1667 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1669 /* cBPF to eBPF migrations are currently not in the idr store.
1670 * Offloaded programs are removed from the store when their device
1671 * disappears - even if someone grabs an fd to them they are unusable,
1672 * simply waiting for refcnt to drop to be freed.
1678 spin_lock_bh(&prog_idr_lock);
1680 __acquire(&prog_idr_lock);
1682 idr_remove(&prog_idr, prog->aux->id);
1686 spin_unlock_bh(&prog_idr_lock);
1688 __release(&prog_idr_lock);
1691 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1693 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1695 kvfree(aux->func_info);
1696 kfree(aux->func_info_aux);
1697 bpf_prog_uncharge_memlock(aux->prog);
1698 security_bpf_prog_free(aux);
1699 bpf_prog_free(aux->prog);
1702 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1704 bpf_prog_kallsyms_del_all(prog);
1705 btf_put(prog->aux->btf);
1706 bpf_prog_free_linfo(prog);
1709 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1711 __bpf_prog_put_rcu(&prog->aux->rcu);
1714 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1716 if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1717 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1718 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
1719 /* bpf_prog_free_id() must be called first */
1720 bpf_prog_free_id(prog, do_idr_lock);
1721 __bpf_prog_put_noref(prog, true);
1725 void bpf_prog_put(struct bpf_prog *prog)
1727 __bpf_prog_put(prog, true);
1729 EXPORT_SYMBOL_GPL(bpf_prog_put);
1731 static int bpf_prog_release(struct inode *inode, struct file *filp)
1733 struct bpf_prog *prog = filp->private_data;
1739 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1740 struct bpf_prog_stats *stats)
1742 u64 nsecs = 0, cnt = 0;
1745 for_each_possible_cpu(cpu) {
1746 const struct bpf_prog_stats *st;
1750 st = per_cpu_ptr(prog->aux->stats, cpu);
1752 start = u64_stats_fetch_begin_irq(&st->syncp);
1755 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1759 stats->nsecs = nsecs;
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"
1783 prog->pages * 1ULL << PAGE_SHIFT,
1790 const struct file_operations bpf_prog_fops = {
1791 #ifdef CONFIG_PROC_FS
1792 .show_fdinfo = bpf_prog_show_fdinfo,
1794 .release = bpf_prog_release,
1795 .read = bpf_dummy_read,
1796 .write = bpf_dummy_write,
1799 int bpf_prog_new_fd(struct bpf_prog *prog)
1803 ret = security_bpf_prog(prog);
1807 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1808 O_RDWR | O_CLOEXEC);
1811 static struct bpf_prog *____bpf_prog_get(struct fd f)
1814 return ERR_PTR(-EBADF);
1815 if (f.file->f_op != &bpf_prog_fops) {
1817 return ERR_PTR(-EINVAL);
1820 return f.file->private_data;
1823 void bpf_prog_add(struct bpf_prog *prog, int i)
1825 atomic64_add(i, &prog->aux->refcnt);
1827 EXPORT_SYMBOL_GPL(bpf_prog_add);
1829 void bpf_prog_sub(struct bpf_prog *prog, int i)
1831 /* Only to be used for undoing previous bpf_prog_add() in some
1832 * error path. We still know that another entity in our call
1833 * path holds a reference to the program, thus atomic_sub() can
1834 * be safely used in such cases!
1836 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1838 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1840 void bpf_prog_inc(struct bpf_prog *prog)
1842 atomic64_inc(&prog->aux->refcnt);
1844 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1846 /* prog_idr_lock should have been held */
1847 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1851 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1854 return ERR_PTR(-ENOENT);
1858 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1860 bool bpf_prog_get_ok(struct bpf_prog *prog,
1861 enum bpf_prog_type *attach_type, bool attach_drv)
1863 /* not an attachment, just a refcount inc, always allow */
1867 if (prog->type != *attach_type)
1869 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1875 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1878 struct fd f = fdget(ufd);
1879 struct bpf_prog *prog;
1881 prog = ____bpf_prog_get(f);
1884 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1885 prog = ERR_PTR(-EINVAL);
1895 struct bpf_prog *bpf_prog_get(u32 ufd)
1897 return __bpf_prog_get(ufd, NULL, false);
1900 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1903 return __bpf_prog_get(ufd, &type, attach_drv);
1905 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1907 /* Initially all BPF programs could be loaded w/o specifying
1908 * expected_attach_type. Later for some of them specifying expected_attach_type
1909 * at load time became required so that program could be validated properly.
1910 * Programs of types that are allowed to be loaded both w/ and w/o (for
1911 * backward compatibility) expected_attach_type, should have the default attach
1912 * type assigned to expected_attach_type for the latter case, so that it can be
1913 * validated later at attach time.
1915 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1916 * prog type requires it but has some attach types that have to be backward
1919 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1921 switch (attr->prog_type) {
1922 case BPF_PROG_TYPE_CGROUP_SOCK:
1923 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1924 * exist so checking for non-zero is the way to go here.
1926 if (!attr->expected_attach_type)
1927 attr->expected_attach_type =
1928 BPF_CGROUP_INET_SOCK_CREATE;
1934 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1935 enum bpf_attach_type expected_attach_type,
1936 u32 btf_id, u32 prog_fd)
1939 if (btf_id > BTF_MAX_TYPE)
1942 switch (prog_type) {
1943 case BPF_PROG_TYPE_TRACING:
1944 case BPF_PROG_TYPE_LSM:
1945 case BPF_PROG_TYPE_STRUCT_OPS:
1946 case BPF_PROG_TYPE_EXT:
1953 if (prog_fd && prog_type != BPF_PROG_TYPE_TRACING &&
1954 prog_type != BPF_PROG_TYPE_EXT)
1957 switch (prog_type) {
1958 case BPF_PROG_TYPE_CGROUP_SOCK:
1959 switch (expected_attach_type) {
1960 case BPF_CGROUP_INET_SOCK_CREATE:
1961 case BPF_CGROUP_INET4_POST_BIND:
1962 case BPF_CGROUP_INET6_POST_BIND:
1967 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1968 switch (expected_attach_type) {
1969 case BPF_CGROUP_INET4_BIND:
1970 case BPF_CGROUP_INET6_BIND:
1971 case BPF_CGROUP_INET4_CONNECT:
1972 case BPF_CGROUP_INET6_CONNECT:
1973 case BPF_CGROUP_UDP4_SENDMSG:
1974 case BPF_CGROUP_UDP6_SENDMSG:
1975 case BPF_CGROUP_UDP4_RECVMSG:
1976 case BPF_CGROUP_UDP6_RECVMSG:
1981 case BPF_PROG_TYPE_CGROUP_SKB:
1982 switch (expected_attach_type) {
1983 case BPF_CGROUP_INET_INGRESS:
1984 case BPF_CGROUP_INET_EGRESS:
1989 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1990 switch (expected_attach_type) {
1991 case BPF_CGROUP_SETSOCKOPT:
1992 case BPF_CGROUP_GETSOCKOPT:
1997 case BPF_PROG_TYPE_EXT:
1998 if (expected_attach_type)
2006 /* last field in 'union bpf_attr' used by this command */
2007 #define BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
2009 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
2011 enum bpf_prog_type type = attr->prog_type;
2012 struct bpf_prog *prog;
2017 if (CHECK_ATTR(BPF_PROG_LOAD))
2020 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2021 BPF_F_ANY_ALIGNMENT |
2022 BPF_F_TEST_STATE_FREQ |
2023 BPF_F_TEST_RND_HI32))
2026 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2027 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2028 !capable(CAP_SYS_ADMIN))
2031 /* copy eBPF program license from user space */
2032 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
2033 sizeof(license) - 1) < 0)
2035 license[sizeof(license) - 1] = 0;
2037 /* eBPF programs must be GPL compatible to use GPL-ed functions */
2038 is_gpl = license_is_gpl_compatible(license);
2040 if (attr->insn_cnt == 0 ||
2041 attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2043 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2044 type != BPF_PROG_TYPE_CGROUP_SKB &&
2045 !capable(CAP_SYS_ADMIN))
2048 bpf_prog_load_fixup_attach_type(attr);
2049 if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2050 attr->attach_btf_id,
2051 attr->attach_prog_fd))
2054 /* plain bpf_prog allocation */
2055 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2059 prog->expected_attach_type = attr->expected_attach_type;
2060 prog->aux->attach_btf_id = attr->attach_btf_id;
2061 if (attr->attach_prog_fd) {
2062 struct bpf_prog *tgt_prog;
2064 tgt_prog = bpf_prog_get(attr->attach_prog_fd);
2065 if (IS_ERR(tgt_prog)) {
2066 err = PTR_ERR(tgt_prog);
2067 goto free_prog_nouncharge;
2069 prog->aux->linked_prog = tgt_prog;
2072 prog->aux->offload_requested = !!attr->prog_ifindex;
2074 err = security_bpf_prog_alloc(prog->aux);
2076 goto free_prog_nouncharge;
2078 err = bpf_prog_charge_memlock(prog);
2082 prog->len = attr->insn_cnt;
2085 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
2086 bpf_prog_insn_size(prog)) != 0)
2089 prog->orig_prog = NULL;
2092 atomic64_set(&prog->aux->refcnt, 1);
2093 prog->gpl_compatible = is_gpl ? 1 : 0;
2095 if (bpf_prog_is_dev_bound(prog->aux)) {
2096 err = bpf_prog_offload_init(prog, attr);
2101 /* find program type: socket_filter vs tracing_filter */
2102 err = find_prog_type(type, prog);
2106 prog->aux->load_time = ktime_get_boottime_ns();
2107 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2108 sizeof(attr->prog_name));
2112 /* run eBPF verifier */
2113 err = bpf_check(&prog, attr, uattr);
2115 goto free_used_maps;
2117 prog = bpf_prog_select_runtime(prog, &err);
2119 goto free_used_maps;
2121 err = bpf_prog_alloc_id(prog);
2123 goto free_used_maps;
2125 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2126 * effectively publicly exposed. However, retrieving via
2127 * bpf_prog_get_fd_by_id() will take another reference,
2128 * therefore it cannot be gone underneath us.
2130 * Only for the time /after/ successful bpf_prog_new_fd()
2131 * and before returning to userspace, we might just hold
2132 * one reference and any parallel close on that fd could
2133 * rip everything out. Hence, below notifications must
2134 * happen before bpf_prog_new_fd().
2136 * Also, any failure handling from this point onwards must
2137 * be using bpf_prog_put() given the program is exposed.
2139 bpf_prog_kallsyms_add(prog);
2140 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2141 bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2143 err = bpf_prog_new_fd(prog);
2149 /* In case we have subprogs, we need to wait for a grace
2150 * period before we can tear down JIT memory since symbols
2151 * are already exposed under kallsyms.
2153 __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2156 bpf_prog_uncharge_memlock(prog);
2158 security_bpf_prog_free(prog->aux);
2159 free_prog_nouncharge:
2160 bpf_prog_free(prog);
2164 #define BPF_OBJ_LAST_FIELD file_flags
2166 static int bpf_obj_pin(const union bpf_attr *attr)
2168 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2171 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2174 static int bpf_obj_get(const union bpf_attr *attr)
2176 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2177 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2180 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2184 void bpf_link_init(struct bpf_link *link, const struct bpf_link_ops *ops,
2185 struct bpf_prog *prog)
2187 atomic64_set(&link->refcnt, 1);
2192 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2193 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2194 * anon_inode's release() call. This helper manages marking bpf_link as
2195 * defunct, releases anon_inode file and puts reserved FD.
2197 void bpf_link_cleanup(struct bpf_link *link, struct file *link_file,
2202 put_unused_fd(link_fd);
2205 void bpf_link_inc(struct bpf_link *link)
2207 atomic64_inc(&link->refcnt);
2210 /* bpf_link_free is guaranteed to be called from process context */
2211 static void bpf_link_free(struct bpf_link *link)
2214 /* detach BPF program, clean up used resources */
2215 link->ops->release(link);
2216 bpf_prog_put(link->prog);
2218 /* free bpf_link and its containing memory */
2219 link->ops->dealloc(link);
2222 static void bpf_link_put_deferred(struct work_struct *work)
2224 struct bpf_link *link = container_of(work, struct bpf_link, work);
2226 bpf_link_free(link);
2229 /* bpf_link_put can be called from atomic context, but ensures that resources
2230 * are freed from process context
2232 void bpf_link_put(struct bpf_link *link)
2234 if (!atomic64_dec_and_test(&link->refcnt))
2238 INIT_WORK(&link->work, bpf_link_put_deferred);
2239 schedule_work(&link->work);
2241 bpf_link_free(link);
2245 static int bpf_link_release(struct inode *inode, struct file *filp)
2247 struct bpf_link *link = filp->private_data;
2253 #ifdef CONFIG_PROC_FS
2254 static const struct bpf_link_ops bpf_raw_tp_lops;
2255 static const struct bpf_link_ops bpf_tracing_link_lops;
2257 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2259 const struct bpf_link *link = filp->private_data;
2260 const struct bpf_prog *prog = link->prog;
2261 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2262 const char *link_type;
2264 if (link->ops == &bpf_raw_tp_lops)
2265 link_type = "raw_tracepoint";
2266 else if (link->ops == &bpf_tracing_link_lops)
2267 link_type = "tracing";
2268 #ifdef CONFIG_CGROUP_BPF
2269 else if (link->ops == &bpf_cgroup_link_lops)
2270 link_type = "cgroup";
2273 link_type = "unknown";
2275 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2286 static const struct file_operations bpf_link_fops = {
2287 #ifdef CONFIG_PROC_FS
2288 .show_fdinfo = bpf_link_show_fdinfo,
2290 .release = bpf_link_release,
2291 .read = bpf_dummy_read,
2292 .write = bpf_dummy_write,
2295 int bpf_link_new_fd(struct bpf_link *link)
2297 return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2300 /* Similar to bpf_link_new_fd, create anon_inode for given bpf_link, but
2301 * instead of immediately installing fd in fdtable, just reserve it and
2302 * return. Caller then need to either install it with fd_install(fd, file) or
2303 * release with put_unused_fd(fd).
2304 * This is useful for cases when bpf_link attachment/detachment are
2305 * complicated and expensive operations and should be delayed until all the fd
2306 * reservation and anon_inode creation succeeds.
2308 struct file *bpf_link_new_file(struct bpf_link *link, int *reserved_fd)
2313 fd = get_unused_fd_flags(O_CLOEXEC);
2317 file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2327 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2329 struct fd f = fdget(ufd);
2330 struct bpf_link *link;
2333 return ERR_PTR(-EBADF);
2334 if (f.file->f_op != &bpf_link_fops) {
2336 return ERR_PTR(-EINVAL);
2339 link = f.file->private_data;
2346 struct bpf_tracing_link {
2347 struct bpf_link link;
2350 static void bpf_tracing_link_release(struct bpf_link *link)
2352 WARN_ON_ONCE(bpf_trampoline_unlink_prog(link->prog));
2355 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2357 struct bpf_tracing_link *tr_link =
2358 container_of(link, struct bpf_tracing_link, link);
2363 static const struct bpf_link_ops bpf_tracing_link_lops = {
2364 .release = bpf_tracing_link_release,
2365 .dealloc = bpf_tracing_link_dealloc,
2368 static int bpf_tracing_prog_attach(struct bpf_prog *prog)
2370 struct bpf_tracing_link *link;
2371 struct file *link_file;
2374 switch (prog->type) {
2375 case BPF_PROG_TYPE_TRACING:
2376 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2377 prog->expected_attach_type != BPF_TRACE_FEXIT &&
2378 prog->expected_attach_type != BPF_MODIFY_RETURN) {
2383 case BPF_PROG_TYPE_EXT:
2384 if (prog->expected_attach_type != 0) {
2389 case BPF_PROG_TYPE_LSM:
2390 if (prog->expected_attach_type != BPF_LSM_MAC) {
2400 link = kzalloc(sizeof(*link), GFP_USER);
2405 bpf_link_init(&link->link, &bpf_tracing_link_lops, prog);
2407 link_file = bpf_link_new_file(&link->link, &link_fd);
2408 if (IS_ERR(link_file)) {
2410 err = PTR_ERR(link_file);
2414 err = bpf_trampoline_link_prog(prog);
2416 bpf_link_cleanup(&link->link, link_file, link_fd);
2420 fd_install(link_fd, link_file);
2428 struct bpf_raw_tp_link {
2429 struct bpf_link link;
2430 struct bpf_raw_event_map *btp;
2433 static void bpf_raw_tp_link_release(struct bpf_link *link)
2435 struct bpf_raw_tp_link *raw_tp =
2436 container_of(link, struct bpf_raw_tp_link, link);
2438 bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
2439 bpf_put_raw_tracepoint(raw_tp->btp);
2442 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
2444 struct bpf_raw_tp_link *raw_tp =
2445 container_of(link, struct bpf_raw_tp_link, link);
2450 static const struct bpf_link_ops bpf_raw_tp_lops = {
2451 .release = bpf_raw_tp_link_release,
2452 .dealloc = bpf_raw_tp_link_dealloc,
2455 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
2457 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
2459 struct bpf_raw_tp_link *link;
2460 struct bpf_raw_event_map *btp;
2461 struct file *link_file;
2462 struct bpf_prog *prog;
2463 const char *tp_name;
2467 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
2470 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
2472 return PTR_ERR(prog);
2474 switch (prog->type) {
2475 case BPF_PROG_TYPE_TRACING:
2476 case BPF_PROG_TYPE_EXT:
2477 case BPF_PROG_TYPE_LSM:
2478 if (attr->raw_tracepoint.name) {
2479 /* The attach point for this category of programs
2480 * should be specified via btf_id during program load.
2485 if (prog->type == BPF_PROG_TYPE_TRACING &&
2486 prog->expected_attach_type == BPF_TRACE_RAW_TP) {
2487 tp_name = prog->aux->attach_func_name;
2490 return bpf_tracing_prog_attach(prog);
2491 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2492 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2493 if (strncpy_from_user(buf,
2494 u64_to_user_ptr(attr->raw_tracepoint.name),
2495 sizeof(buf) - 1) < 0) {
2499 buf[sizeof(buf) - 1] = 0;
2507 btp = bpf_get_raw_tracepoint(tp_name);
2513 link = kzalloc(sizeof(*link), GFP_USER);
2518 bpf_link_init(&link->link, &bpf_raw_tp_lops, prog);
2521 link_file = bpf_link_new_file(&link->link, &link_fd);
2522 if (IS_ERR(link_file)) {
2524 err = PTR_ERR(link_file);
2528 err = bpf_probe_register(link->btp, prog);
2530 bpf_link_cleanup(&link->link, link_file, link_fd);
2534 fd_install(link_fd, link_file);
2538 bpf_put_raw_tracepoint(btp);
2544 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2545 enum bpf_attach_type attach_type)
2547 switch (prog->type) {
2548 case BPF_PROG_TYPE_CGROUP_SOCK:
2549 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2550 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2551 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2552 case BPF_PROG_TYPE_CGROUP_SKB:
2553 return prog->enforce_expected_attach_type &&
2554 prog->expected_attach_type != attach_type ?
2561 static enum bpf_prog_type
2562 attach_type_to_prog_type(enum bpf_attach_type attach_type)
2564 switch (attach_type) {
2565 case BPF_CGROUP_INET_INGRESS:
2566 case BPF_CGROUP_INET_EGRESS:
2567 return BPF_PROG_TYPE_CGROUP_SKB;
2569 case BPF_CGROUP_INET_SOCK_CREATE:
2570 case BPF_CGROUP_INET4_POST_BIND:
2571 case BPF_CGROUP_INET6_POST_BIND:
2572 return BPF_PROG_TYPE_CGROUP_SOCK;
2573 case BPF_CGROUP_INET4_BIND:
2574 case BPF_CGROUP_INET6_BIND:
2575 case BPF_CGROUP_INET4_CONNECT:
2576 case BPF_CGROUP_INET6_CONNECT:
2577 case BPF_CGROUP_UDP4_SENDMSG:
2578 case BPF_CGROUP_UDP6_SENDMSG:
2579 case BPF_CGROUP_UDP4_RECVMSG:
2580 case BPF_CGROUP_UDP6_RECVMSG:
2581 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2582 case BPF_CGROUP_SOCK_OPS:
2583 return BPF_PROG_TYPE_SOCK_OPS;
2584 case BPF_CGROUP_DEVICE:
2585 return BPF_PROG_TYPE_CGROUP_DEVICE;
2586 case BPF_SK_MSG_VERDICT:
2587 return BPF_PROG_TYPE_SK_MSG;
2588 case BPF_SK_SKB_STREAM_PARSER:
2589 case BPF_SK_SKB_STREAM_VERDICT:
2590 return BPF_PROG_TYPE_SK_SKB;
2591 case BPF_LIRC_MODE2:
2592 return BPF_PROG_TYPE_LIRC_MODE2;
2593 case BPF_FLOW_DISSECTOR:
2594 return BPF_PROG_TYPE_FLOW_DISSECTOR;
2595 case BPF_CGROUP_SYSCTL:
2596 return BPF_PROG_TYPE_CGROUP_SYSCTL;
2597 case BPF_CGROUP_GETSOCKOPT:
2598 case BPF_CGROUP_SETSOCKOPT:
2599 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
2601 return BPF_PROG_TYPE_UNSPEC;
2605 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
2607 #define BPF_F_ATTACH_MASK \
2608 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
2610 static int bpf_prog_attach(const union bpf_attr *attr)
2612 enum bpf_prog_type ptype;
2613 struct bpf_prog *prog;
2616 if (!capable(CAP_NET_ADMIN))
2619 if (CHECK_ATTR(BPF_PROG_ATTACH))
2622 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
2625 ptype = attach_type_to_prog_type(attr->attach_type);
2626 if (ptype == BPF_PROG_TYPE_UNSPEC)
2629 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
2631 return PTR_ERR(prog);
2633 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
2639 case BPF_PROG_TYPE_SK_SKB:
2640 case BPF_PROG_TYPE_SK_MSG:
2641 ret = sock_map_get_from_fd(attr, prog);
2643 case BPF_PROG_TYPE_LIRC_MODE2:
2644 ret = lirc_prog_attach(attr, prog);
2646 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2647 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
2649 case BPF_PROG_TYPE_CGROUP_DEVICE:
2650 case BPF_PROG_TYPE_CGROUP_SKB:
2651 case BPF_PROG_TYPE_CGROUP_SOCK:
2652 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2653 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2654 case BPF_PROG_TYPE_CGROUP_SYSCTL:
2655 case BPF_PROG_TYPE_SOCK_OPS:
2656 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
2667 #define BPF_PROG_DETACH_LAST_FIELD attach_type
2669 static int bpf_prog_detach(const union bpf_attr *attr)
2671 enum bpf_prog_type ptype;
2673 if (!capable(CAP_NET_ADMIN))
2676 if (CHECK_ATTR(BPF_PROG_DETACH))
2679 ptype = attach_type_to_prog_type(attr->attach_type);
2682 case BPF_PROG_TYPE_SK_MSG:
2683 case BPF_PROG_TYPE_SK_SKB:
2684 return sock_map_get_from_fd(attr, NULL);
2685 case BPF_PROG_TYPE_LIRC_MODE2:
2686 return lirc_prog_detach(attr);
2687 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2688 return skb_flow_dissector_bpf_prog_detach(attr);
2689 case BPF_PROG_TYPE_CGROUP_DEVICE:
2690 case BPF_PROG_TYPE_CGROUP_SKB:
2691 case BPF_PROG_TYPE_CGROUP_SOCK:
2692 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2693 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2694 case BPF_PROG_TYPE_CGROUP_SYSCTL:
2695 case BPF_PROG_TYPE_SOCK_OPS:
2696 return cgroup_bpf_prog_detach(attr, ptype);
2702 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2704 static int bpf_prog_query(const union bpf_attr *attr,
2705 union bpf_attr __user *uattr)
2707 if (!capable(CAP_NET_ADMIN))
2709 if (CHECK_ATTR(BPF_PROG_QUERY))
2711 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2714 switch (attr->query.attach_type) {
2715 case BPF_CGROUP_INET_INGRESS:
2716 case BPF_CGROUP_INET_EGRESS:
2717 case BPF_CGROUP_INET_SOCK_CREATE:
2718 case BPF_CGROUP_INET4_BIND:
2719 case BPF_CGROUP_INET6_BIND:
2720 case BPF_CGROUP_INET4_POST_BIND:
2721 case BPF_CGROUP_INET6_POST_BIND:
2722 case BPF_CGROUP_INET4_CONNECT:
2723 case BPF_CGROUP_INET6_CONNECT:
2724 case BPF_CGROUP_UDP4_SENDMSG:
2725 case BPF_CGROUP_UDP6_SENDMSG:
2726 case BPF_CGROUP_UDP4_RECVMSG:
2727 case BPF_CGROUP_UDP6_RECVMSG:
2728 case BPF_CGROUP_SOCK_OPS:
2729 case BPF_CGROUP_DEVICE:
2730 case BPF_CGROUP_SYSCTL:
2731 case BPF_CGROUP_GETSOCKOPT:
2732 case BPF_CGROUP_SETSOCKOPT:
2733 return cgroup_bpf_prog_query(attr, uattr);
2734 case BPF_LIRC_MODE2:
2735 return lirc_prog_query(attr, uattr);
2736 case BPF_FLOW_DISSECTOR:
2737 return skb_flow_dissector_prog_query(attr, uattr);
2743 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2745 static int bpf_prog_test_run(const union bpf_attr *attr,
2746 union bpf_attr __user *uattr)
2748 struct bpf_prog *prog;
2749 int ret = -ENOTSUPP;
2751 if (!capable(CAP_SYS_ADMIN))
2753 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2756 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2757 (!attr->test.ctx_size_in && attr->test.ctx_in))
2760 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2761 (!attr->test.ctx_size_out && attr->test.ctx_out))
2764 prog = bpf_prog_get(attr->test.prog_fd);
2766 return PTR_ERR(prog);
2768 if (prog->aux->ops->test_run)
2769 ret = prog->aux->ops->test_run(prog, attr, uattr);
2775 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2777 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2778 union bpf_attr __user *uattr,
2782 u32 next_id = attr->start_id;
2785 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2788 if (!capable(CAP_SYS_ADMIN))
2793 if (!idr_get_next(idr, &next_id))
2795 spin_unlock_bh(lock);
2798 err = put_user(next_id, &uattr->next_id);
2803 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2805 struct bpf_prog *bpf_prog_by_id(u32 id)
2807 struct bpf_prog *prog;
2810 return ERR_PTR(-ENOENT);
2812 spin_lock_bh(&prog_idr_lock);
2813 prog = idr_find(&prog_idr, id);
2815 prog = bpf_prog_inc_not_zero(prog);
2817 prog = ERR_PTR(-ENOENT);
2818 spin_unlock_bh(&prog_idr_lock);
2822 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2824 struct bpf_prog *prog;
2825 u32 id = attr->prog_id;
2828 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2831 if (!capable(CAP_SYS_ADMIN))
2834 prog = bpf_prog_by_id(id);
2836 return PTR_ERR(prog);
2838 fd = bpf_prog_new_fd(prog);
2845 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2847 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2849 struct bpf_map *map;
2850 u32 id = attr->map_id;
2854 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2855 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2858 if (!capable(CAP_SYS_ADMIN))
2861 f_flags = bpf_get_file_flag(attr->open_flags);
2865 spin_lock_bh(&map_idr_lock);
2866 map = idr_find(&map_idr, id);
2868 map = __bpf_map_inc_not_zero(map, true);
2870 map = ERR_PTR(-ENOENT);
2871 spin_unlock_bh(&map_idr_lock);
2874 return PTR_ERR(map);
2876 fd = bpf_map_new_fd(map, f_flags);
2878 bpf_map_put_with_uref(map);
2883 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2884 unsigned long addr, u32 *off,
2887 const struct bpf_map *map;
2890 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2891 map = prog->aux->used_maps[i];
2892 if (map == (void *)addr) {
2893 *type = BPF_PSEUDO_MAP_FD;
2896 if (!map->ops->map_direct_value_meta)
2898 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2899 *type = BPF_PSEUDO_MAP_VALUE;
2907 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2909 const struct bpf_map *map;
2910 struct bpf_insn *insns;
2915 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2920 for (i = 0; i < prog->len; i++) {
2921 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2922 insns[i].code = BPF_JMP | BPF_CALL;
2923 insns[i].imm = BPF_FUNC_tail_call;
2926 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2927 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2928 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2929 insns[i].code = BPF_JMP | BPF_CALL;
2930 if (!bpf_dump_raw_ok())
2935 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2938 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2939 map = bpf_map_from_imm(prog, imm, &off, &type);
2941 insns[i].src_reg = type;
2942 insns[i].imm = map->id;
2943 insns[i + 1].imm = off;
2951 static int set_info_rec_size(struct bpf_prog_info *info)
2954 * Ensure info.*_rec_size is the same as kernel expected size
2958 * Only allow zero *_rec_size if both _rec_size and _cnt are
2959 * zero. In this case, the kernel will set the expected
2960 * _rec_size back to the info.
2963 if ((info->nr_func_info || info->func_info_rec_size) &&
2964 info->func_info_rec_size != sizeof(struct bpf_func_info))
2967 if ((info->nr_line_info || info->line_info_rec_size) &&
2968 info->line_info_rec_size != sizeof(struct bpf_line_info))
2971 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2972 info->jited_line_info_rec_size != sizeof(__u64))
2975 info->func_info_rec_size = sizeof(struct bpf_func_info);
2976 info->line_info_rec_size = sizeof(struct bpf_line_info);
2977 info->jited_line_info_rec_size = sizeof(__u64);
2982 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2983 const union bpf_attr *attr,
2984 union bpf_attr __user *uattr)
2986 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2987 struct bpf_prog_info info;
2988 u32 info_len = attr->info.info_len;
2989 struct bpf_prog_stats stats;
2990 char __user *uinsns;
2994 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2997 info_len = min_t(u32, sizeof(info), info_len);
2999 memset(&info, 0, sizeof(info));
3000 if (copy_from_user(&info, uinfo, info_len))
3003 info.type = prog->type;
3004 info.id = prog->aux->id;
3005 info.load_time = prog->aux->load_time;
3006 info.created_by_uid = from_kuid_munged(current_user_ns(),
3007 prog->aux->user->uid);
3008 info.gpl_compatible = prog->gpl_compatible;
3010 memcpy(info.tag, prog->tag, sizeof(prog->tag));
3011 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3013 ulen = info.nr_map_ids;
3014 info.nr_map_ids = prog->aux->used_map_cnt;
3015 ulen = min_t(u32, info.nr_map_ids, ulen);
3017 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3020 for (i = 0; i < ulen; i++)
3021 if (put_user(prog->aux->used_maps[i]->id,
3026 err = set_info_rec_size(&info);
3030 bpf_prog_get_stats(prog, &stats);
3031 info.run_time_ns = stats.nsecs;
3032 info.run_cnt = stats.cnt;
3034 if (!capable(CAP_SYS_ADMIN)) {
3035 info.jited_prog_len = 0;
3036 info.xlated_prog_len = 0;
3037 info.nr_jited_ksyms = 0;
3038 info.nr_jited_func_lens = 0;
3039 info.nr_func_info = 0;
3040 info.nr_line_info = 0;
3041 info.nr_jited_line_info = 0;
3045 ulen = info.xlated_prog_len;
3046 info.xlated_prog_len = bpf_prog_insn_size(prog);
3047 if (info.xlated_prog_len && ulen) {
3048 struct bpf_insn *insns_sanitized;
3051 if (prog->blinded && !bpf_dump_raw_ok()) {
3052 info.xlated_prog_insns = 0;
3055 insns_sanitized = bpf_insn_prepare_dump(prog);
3056 if (!insns_sanitized)
3058 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3059 ulen = min_t(u32, info.xlated_prog_len, ulen);
3060 fault = copy_to_user(uinsns, insns_sanitized, ulen);
3061 kfree(insns_sanitized);
3066 if (bpf_prog_is_dev_bound(prog->aux)) {
3067 err = bpf_prog_offload_info_fill(&info, prog);
3073 /* NOTE: the following code is supposed to be skipped for offload.
3074 * bpf_prog_offload_info_fill() is the place to fill similar fields
3077 ulen = info.jited_prog_len;
3078 if (prog->aux->func_cnt) {
3081 info.jited_prog_len = 0;
3082 for (i = 0; i < prog->aux->func_cnt; i++)
3083 info.jited_prog_len += prog->aux->func[i]->jited_len;
3085 info.jited_prog_len = prog->jited_len;
3088 if (info.jited_prog_len && ulen) {
3089 if (bpf_dump_raw_ok()) {
3090 uinsns = u64_to_user_ptr(info.jited_prog_insns);
3091 ulen = min_t(u32, info.jited_prog_len, ulen);
3093 /* for multi-function programs, copy the JITed
3094 * instructions for all the functions
3096 if (prog->aux->func_cnt) {
3101 for (i = 0; i < prog->aux->func_cnt; i++) {
3102 len = prog->aux->func[i]->jited_len;
3103 len = min_t(u32, len, free);
3104 img = (u8 *) prog->aux->func[i]->bpf_func;
3105 if (copy_to_user(uinsns, img, len))
3113 if (copy_to_user(uinsns, prog->bpf_func, ulen))
3117 info.jited_prog_insns = 0;
3121 ulen = info.nr_jited_ksyms;
3122 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
3124 if (bpf_dump_raw_ok()) {
3125 unsigned long ksym_addr;
3126 u64 __user *user_ksyms;
3129 /* copy the address of the kernel symbol
3130 * corresponding to each function
3132 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
3133 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
3134 if (prog->aux->func_cnt) {
3135 for (i = 0; i < ulen; i++) {
3136 ksym_addr = (unsigned long)
3137 prog->aux->func[i]->bpf_func;
3138 if (put_user((u64) ksym_addr,
3143 ksym_addr = (unsigned long) prog->bpf_func;
3144 if (put_user((u64) ksym_addr, &user_ksyms[0]))
3148 info.jited_ksyms = 0;
3152 ulen = info.nr_jited_func_lens;
3153 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
3155 if (bpf_dump_raw_ok()) {
3156 u32 __user *user_lens;
3159 /* copy the JITed image lengths for each function */
3160 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
3161 user_lens = u64_to_user_ptr(info.jited_func_lens);
3162 if (prog->aux->func_cnt) {
3163 for (i = 0; i < ulen; i++) {
3165 prog->aux->func[i]->jited_len;
3166 if (put_user(func_len, &user_lens[i]))
3170 func_len = prog->jited_len;
3171 if (put_user(func_len, &user_lens[0]))
3175 info.jited_func_lens = 0;
3180 info.btf_id = btf_id(prog->aux->btf);
3182 ulen = info.nr_func_info;
3183 info.nr_func_info = prog->aux->func_info_cnt;
3184 if (info.nr_func_info && ulen) {
3185 char __user *user_finfo;
3187 user_finfo = u64_to_user_ptr(info.func_info);
3188 ulen = min_t(u32, info.nr_func_info, ulen);
3189 if (copy_to_user(user_finfo, prog->aux->func_info,
3190 info.func_info_rec_size * ulen))
3194 ulen = info.nr_line_info;
3195 info.nr_line_info = prog->aux->nr_linfo;
3196 if (info.nr_line_info && ulen) {
3197 __u8 __user *user_linfo;
3199 user_linfo = u64_to_user_ptr(info.line_info);
3200 ulen = min_t(u32, info.nr_line_info, ulen);
3201 if (copy_to_user(user_linfo, prog->aux->linfo,
3202 info.line_info_rec_size * ulen))
3206 ulen = info.nr_jited_line_info;
3207 if (prog->aux->jited_linfo)
3208 info.nr_jited_line_info = prog->aux->nr_linfo;
3210 info.nr_jited_line_info = 0;
3211 if (info.nr_jited_line_info && ulen) {
3212 if (bpf_dump_raw_ok()) {
3213 __u64 __user *user_linfo;
3216 user_linfo = u64_to_user_ptr(info.jited_line_info);
3217 ulen = min_t(u32, info.nr_jited_line_info, ulen);
3218 for (i = 0; i < ulen; i++) {
3219 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
3224 info.jited_line_info = 0;
3228 ulen = info.nr_prog_tags;
3229 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
3231 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
3234 user_prog_tags = u64_to_user_ptr(info.prog_tags);
3235 ulen = min_t(u32, info.nr_prog_tags, ulen);
3236 if (prog->aux->func_cnt) {
3237 for (i = 0; i < ulen; i++) {
3238 if (copy_to_user(user_prog_tags[i],
3239 prog->aux->func[i]->tag,
3244 if (copy_to_user(user_prog_tags[0],
3245 prog->tag, BPF_TAG_SIZE))
3251 if (copy_to_user(uinfo, &info, info_len) ||
3252 put_user(info_len, &uattr->info.info_len))
3258 static int bpf_map_get_info_by_fd(struct bpf_map *map,
3259 const union bpf_attr *attr,
3260 union bpf_attr __user *uattr)
3262 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3263 struct bpf_map_info info;
3264 u32 info_len = attr->info.info_len;
3267 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3270 info_len = min_t(u32, sizeof(info), info_len);
3272 memset(&info, 0, sizeof(info));
3273 info.type = map->map_type;
3275 info.key_size = map->key_size;
3276 info.value_size = map->value_size;
3277 info.max_entries = map->max_entries;
3278 info.map_flags = map->map_flags;
3279 memcpy(info.name, map->name, sizeof(map->name));
3282 info.btf_id = btf_id(map->btf);
3283 info.btf_key_type_id = map->btf_key_type_id;
3284 info.btf_value_type_id = map->btf_value_type_id;
3286 info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
3288 if (bpf_map_is_dev_bound(map)) {
3289 err = bpf_map_offload_info_fill(&info, map);
3294 if (copy_to_user(uinfo, &info, info_len) ||
3295 put_user(info_len, &uattr->info.info_len))
3301 static int bpf_btf_get_info_by_fd(struct btf *btf,
3302 const union bpf_attr *attr,
3303 union bpf_attr __user *uattr)
3305 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3306 u32 info_len = attr->info.info_len;
3309 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
3313 return btf_get_info_by_fd(btf, attr, uattr);
3316 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
3318 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
3319 union bpf_attr __user *uattr)
3321 int ufd = attr->info.bpf_fd;
3325 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
3332 if (f.file->f_op == &bpf_prog_fops)
3333 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
3335 else if (f.file->f_op == &bpf_map_fops)
3336 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
3338 else if (f.file->f_op == &btf_fops)
3339 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
3347 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
3349 static int bpf_btf_load(const union bpf_attr *attr)
3351 if (CHECK_ATTR(BPF_BTF_LOAD))
3354 if (!capable(CAP_SYS_ADMIN))
3357 return btf_new_fd(attr);
3360 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
3362 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
3364 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
3367 if (!capable(CAP_SYS_ADMIN))
3370 return btf_get_fd_by_id(attr->btf_id);
3373 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
3374 union bpf_attr __user *uattr,
3375 u32 prog_id, u32 fd_type,
3376 const char *buf, u64 probe_offset,
3379 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
3380 u32 len = buf ? strlen(buf) : 0, input_len;
3383 if (put_user(len, &uattr->task_fd_query.buf_len))
3385 input_len = attr->task_fd_query.buf_len;
3386 if (input_len && ubuf) {
3388 /* nothing to copy, just make ubuf NULL terminated */
3391 if (put_user(zero, ubuf))
3393 } else if (input_len >= len + 1) {
3394 /* ubuf can hold the string with NULL terminator */
3395 if (copy_to_user(ubuf, buf, len + 1))
3398 /* ubuf cannot hold the string with NULL terminator,
3399 * do a partial copy with NULL terminator.
3404 if (copy_to_user(ubuf, buf, input_len - 1))
3406 if (put_user(zero, ubuf + input_len - 1))
3411 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
3412 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
3413 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
3414 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
3420 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
3422 static int bpf_task_fd_query(const union bpf_attr *attr,
3423 union bpf_attr __user *uattr)
3425 pid_t pid = attr->task_fd_query.pid;
3426 u32 fd = attr->task_fd_query.fd;
3427 const struct perf_event *event;
3428 struct files_struct *files;
3429 struct task_struct *task;
3433 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
3436 if (!capable(CAP_SYS_ADMIN))
3439 if (attr->task_fd_query.flags != 0)
3442 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3446 files = get_files_struct(task);
3447 put_task_struct(task);
3452 spin_lock(&files->file_lock);
3453 file = fcheck_files(files, fd);
3458 spin_unlock(&files->file_lock);
3459 put_files_struct(files);
3464 if (file->f_op == &bpf_link_fops) {
3465 struct bpf_link *link = file->private_data;
3467 if (link->ops == &bpf_raw_tp_lops) {
3468 struct bpf_raw_tp_link *raw_tp =
3469 container_of(link, struct bpf_raw_tp_link, link);
3470 struct bpf_raw_event_map *btp = raw_tp->btp;
3472 err = bpf_task_fd_query_copy(attr, uattr,
3473 raw_tp->link.prog->aux->id,
3474 BPF_FD_TYPE_RAW_TRACEPOINT,
3475 btp->tp->name, 0, 0);
3481 event = perf_get_event(file);
3482 if (!IS_ERR(event)) {
3483 u64 probe_offset, probe_addr;
3484 u32 prog_id, fd_type;
3487 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
3488 &buf, &probe_offset,
3491 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
3506 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
3508 #define BPF_DO_BATCH(fn) \
3514 err = fn(map, attr, uattr); \
3517 static int bpf_map_do_batch(const union bpf_attr *attr,
3518 union bpf_attr __user *uattr,
3521 struct bpf_map *map;
3525 if (CHECK_ATTR(BPF_MAP_BATCH))
3528 ufd = attr->batch.map_fd;
3530 map = __bpf_map_get(f);
3532 return PTR_ERR(map);
3534 if ((cmd == BPF_MAP_LOOKUP_BATCH ||
3535 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH) &&
3536 !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
3541 if (cmd != BPF_MAP_LOOKUP_BATCH &&
3542 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
3547 if (cmd == BPF_MAP_LOOKUP_BATCH)
3548 BPF_DO_BATCH(map->ops->map_lookup_batch);
3549 else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
3550 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
3551 else if (cmd == BPF_MAP_UPDATE_BATCH)
3552 BPF_DO_BATCH(map->ops->map_update_batch);
3554 BPF_DO_BATCH(map->ops->map_delete_batch);
3561 #define BPF_LINK_CREATE_LAST_FIELD link_create.flags
3562 static int link_create(union bpf_attr *attr)
3564 enum bpf_prog_type ptype;
3565 struct bpf_prog *prog;
3568 if (!capable(CAP_NET_ADMIN))
3571 if (CHECK_ATTR(BPF_LINK_CREATE))
3574 ptype = attach_type_to_prog_type(attr->link_create.attach_type);
3575 if (ptype == BPF_PROG_TYPE_UNSPEC)
3578 prog = bpf_prog_get_type(attr->link_create.prog_fd, ptype);
3580 return PTR_ERR(prog);
3582 ret = bpf_prog_attach_check_attach_type(prog,
3583 attr->link_create.attach_type);
3588 case BPF_PROG_TYPE_CGROUP_SKB:
3589 case BPF_PROG_TYPE_CGROUP_SOCK:
3590 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3591 case BPF_PROG_TYPE_SOCK_OPS:
3592 case BPF_PROG_TYPE_CGROUP_DEVICE:
3593 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3594 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3595 ret = cgroup_bpf_link_attach(attr, prog);
3607 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
3609 static int link_update(union bpf_attr *attr)
3611 struct bpf_prog *old_prog = NULL, *new_prog;
3612 struct bpf_link *link;
3616 if (!capable(CAP_NET_ADMIN))
3619 if (CHECK_ATTR(BPF_LINK_UPDATE))
3622 flags = attr->link_update.flags;
3623 if (flags & ~BPF_F_REPLACE)
3626 link = bpf_link_get_from_fd(attr->link_update.link_fd);
3628 return PTR_ERR(link);
3630 new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
3631 if (IS_ERR(new_prog)) {
3632 ret = PTR_ERR(new_prog);
3636 if (flags & BPF_F_REPLACE) {
3637 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
3638 if (IS_ERR(old_prog)) {
3639 ret = PTR_ERR(old_prog);
3643 } else if (attr->link_update.old_prog_fd) {
3648 #ifdef CONFIG_CGROUP_BPF
3649 if (link->ops == &bpf_cgroup_link_lops) {
3650 ret = cgroup_bpf_replace(link, old_prog, new_prog);
3658 bpf_prog_put(old_prog);
3660 bpf_prog_put(new_prog);
3666 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
3668 union bpf_attr attr;
3671 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
3674 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
3677 size = min_t(u32, size, sizeof(attr));
3679 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
3680 memset(&attr, 0, sizeof(attr));
3681 if (copy_from_user(&attr, uattr, size) != 0)
3684 err = security_bpf(cmd, &attr, size);
3689 case BPF_MAP_CREATE:
3690 err = map_create(&attr);
3692 case BPF_MAP_LOOKUP_ELEM:
3693 err = map_lookup_elem(&attr);
3695 case BPF_MAP_UPDATE_ELEM:
3696 err = map_update_elem(&attr);
3698 case BPF_MAP_DELETE_ELEM:
3699 err = map_delete_elem(&attr);
3701 case BPF_MAP_GET_NEXT_KEY:
3702 err = map_get_next_key(&attr);
3704 case BPF_MAP_FREEZE:
3705 err = map_freeze(&attr);
3708 err = bpf_prog_load(&attr, uattr);
3711 err = bpf_obj_pin(&attr);
3714 err = bpf_obj_get(&attr);
3716 case BPF_PROG_ATTACH:
3717 err = bpf_prog_attach(&attr);
3719 case BPF_PROG_DETACH:
3720 err = bpf_prog_detach(&attr);
3722 case BPF_PROG_QUERY:
3723 err = bpf_prog_query(&attr, uattr);
3725 case BPF_PROG_TEST_RUN:
3726 err = bpf_prog_test_run(&attr, uattr);
3728 case BPF_PROG_GET_NEXT_ID:
3729 err = bpf_obj_get_next_id(&attr, uattr,
3730 &prog_idr, &prog_idr_lock);
3732 case BPF_MAP_GET_NEXT_ID:
3733 err = bpf_obj_get_next_id(&attr, uattr,
3734 &map_idr, &map_idr_lock);
3736 case BPF_BTF_GET_NEXT_ID:
3737 err = bpf_obj_get_next_id(&attr, uattr,
3738 &btf_idr, &btf_idr_lock);
3740 case BPF_PROG_GET_FD_BY_ID:
3741 err = bpf_prog_get_fd_by_id(&attr);
3743 case BPF_MAP_GET_FD_BY_ID:
3744 err = bpf_map_get_fd_by_id(&attr);
3746 case BPF_OBJ_GET_INFO_BY_FD:
3747 err = bpf_obj_get_info_by_fd(&attr, uattr);
3749 case BPF_RAW_TRACEPOINT_OPEN:
3750 err = bpf_raw_tracepoint_open(&attr);
3753 err = bpf_btf_load(&attr);
3755 case BPF_BTF_GET_FD_BY_ID:
3756 err = bpf_btf_get_fd_by_id(&attr);
3758 case BPF_TASK_FD_QUERY:
3759 err = bpf_task_fd_query(&attr, uattr);
3761 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
3762 err = map_lookup_and_delete_elem(&attr);
3764 case BPF_MAP_LOOKUP_BATCH:
3765 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_LOOKUP_BATCH);
3767 case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
3768 err = bpf_map_do_batch(&attr, uattr,
3769 BPF_MAP_LOOKUP_AND_DELETE_BATCH);
3771 case BPF_MAP_UPDATE_BATCH:
3772 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_UPDATE_BATCH);
3774 case BPF_MAP_DELETE_BATCH:
3775 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_DELETE_BATCH);
3777 case BPF_LINK_CREATE:
3778 err = link_create(&attr);
3780 case BPF_LINK_UPDATE:
3781 err = link_update(&attr);