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>
29 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
30 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
31 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
32 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
33 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
34 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
37 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
39 DEFINE_PER_CPU(int, bpf_prog_active);
40 static DEFINE_IDR(prog_idr);
41 static DEFINE_SPINLOCK(prog_idr_lock);
42 static DEFINE_IDR(map_idr);
43 static DEFINE_SPINLOCK(map_idr_lock);
45 int sysctl_unprivileged_bpf_disabled __read_mostly;
47 static const struct bpf_map_ops * const bpf_map_types[] = {
48 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
49 #define BPF_MAP_TYPE(_id, _ops) \
51 #include <linux/bpf_types.h>
57 * If we're handed a bigger struct than we know of, ensure all the unknown bits
58 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
59 * we don't know about yet.
61 * There is a ToCToU between this function call and the following
62 * copy_from_user() call. However, this is not a concern since this function is
63 * meant to be a future-proofing of bits.
65 int bpf_check_uarg_tail_zero(void __user *uaddr,
69 unsigned char __user *addr;
70 unsigned char __user *end;
74 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
77 if (unlikely(!access_ok(uaddr, actual_size)))
80 if (actual_size <= expected_size)
83 addr = uaddr + expected_size;
84 end = uaddr + actual_size;
86 for (; addr < end; addr++) {
87 err = get_user(val, addr);
97 const struct bpf_map_ops bpf_map_offload_ops = {
98 .map_alloc = bpf_map_offload_map_alloc,
99 .map_free = bpf_map_offload_map_free,
100 .map_check_btf = map_check_no_btf,
103 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
105 const struct bpf_map_ops *ops;
106 u32 type = attr->map_type;
110 if (type >= ARRAY_SIZE(bpf_map_types))
111 return ERR_PTR(-EINVAL);
112 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
113 ops = bpf_map_types[type];
115 return ERR_PTR(-EINVAL);
117 if (ops->map_alloc_check) {
118 err = ops->map_alloc_check(attr);
122 if (attr->map_ifindex)
123 ops = &bpf_map_offload_ops;
124 map = ops->map_alloc(attr);
128 map->map_type = type;
132 static u32 bpf_map_value_size(struct bpf_map *map)
134 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
135 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
136 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
137 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
138 return round_up(map->value_size, 8) * num_possible_cpus();
139 else if (IS_FD_MAP(map))
142 return map->value_size;
145 static void maybe_wait_bpf_programs(struct bpf_map *map)
147 /* Wait for any running BPF programs to complete so that
148 * userspace, when we return to it, knows that all programs
149 * that could be running use the new map value.
151 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
152 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
156 static int bpf_map_update_value(struct bpf_map *map, struct fd f, void *key,
157 void *value, __u64 flags)
161 /* Need to create a kthread, thus must support schedule */
162 if (bpf_map_is_dev_bound(map)) {
163 return bpf_map_offload_update_elem(map, key, value, flags);
164 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
165 map->map_type == BPF_MAP_TYPE_SOCKHASH ||
166 map->map_type == BPF_MAP_TYPE_SOCKMAP ||
167 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
168 return map->ops->map_update_elem(map, key, value, flags);
169 } else if (IS_FD_PROG_ARRAY(map)) {
170 return bpf_fd_array_map_update_elem(map, f.file, key, value,
174 bpf_disable_instrumentation();
175 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
176 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
177 err = bpf_percpu_hash_update(map, key, value, flags);
178 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
179 err = bpf_percpu_array_update(map, key, value, flags);
180 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
181 err = bpf_percpu_cgroup_storage_update(map, key, value,
183 } else if (IS_FD_ARRAY(map)) {
185 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
188 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
190 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
193 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
194 /* rcu_read_lock() is not needed */
195 err = bpf_fd_reuseport_array_update_elem(map, key, value,
197 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
198 map->map_type == BPF_MAP_TYPE_STACK) {
199 err = map->ops->map_push_elem(map, value, flags);
202 err = map->ops->map_update_elem(map, key, value, flags);
205 bpf_enable_instrumentation();
206 maybe_wait_bpf_programs(map);
211 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
217 if (bpf_map_is_dev_bound(map))
218 return bpf_map_offload_lookup_elem(map, key, value);
220 bpf_disable_instrumentation();
221 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
222 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
223 err = bpf_percpu_hash_copy(map, key, value);
224 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
225 err = bpf_percpu_array_copy(map, key, value);
226 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
227 err = bpf_percpu_cgroup_storage_copy(map, key, value);
228 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
229 err = bpf_stackmap_copy(map, key, value);
230 } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
231 err = bpf_fd_array_map_lookup_elem(map, key, value);
232 } else if (IS_FD_HASH(map)) {
233 err = bpf_fd_htab_map_lookup_elem(map, key, value);
234 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
235 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
236 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
237 map->map_type == BPF_MAP_TYPE_STACK) {
238 err = map->ops->map_peek_elem(map, value);
239 } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
240 /* struct_ops map requires directly updating "value" */
241 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
244 if (map->ops->map_lookup_elem_sys_only)
245 ptr = map->ops->map_lookup_elem_sys_only(map, key);
247 ptr = map->ops->map_lookup_elem(map, key);
254 if (flags & BPF_F_LOCK)
255 /* lock 'ptr' and copy everything but lock */
256 copy_map_value_locked(map, value, ptr, true);
258 copy_map_value(map, value, ptr);
259 /* mask lock, since value wasn't zero inited */
260 check_and_init_map_lock(map, value);
265 bpf_enable_instrumentation();
266 maybe_wait_bpf_programs(map);
271 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
273 /* We really just want to fail instead of triggering OOM killer
274 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
275 * which is used for lower order allocation requests.
277 * It has been observed that higher order allocation requests done by
278 * vmalloc with __GFP_NORETRY being set might fail due to not trying
279 * to reclaim memory from the page cache, thus we set
280 * __GFP_RETRY_MAYFAIL to avoid such situations.
283 const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
286 if (size >= SIZE_MAX)
289 /* kmalloc()'ed memory can't be mmap()'ed */
290 if (!mmapable && size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
291 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
297 BUG_ON(!PAGE_ALIGNED(size));
298 return vmalloc_user_node_flags(size, numa_node, GFP_KERNEL |
299 __GFP_RETRY_MAYFAIL | flags);
301 return __vmalloc_node_flags_caller(size, numa_node,
302 GFP_KERNEL | __GFP_RETRY_MAYFAIL |
303 flags, __builtin_return_address(0));
306 void *bpf_map_area_alloc(u64 size, int numa_node)
308 return __bpf_map_area_alloc(size, numa_node, false);
311 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
313 return __bpf_map_area_alloc(size, numa_node, true);
316 void bpf_map_area_free(void *area)
321 static u32 bpf_map_flags_retain_permanent(u32 flags)
323 /* Some map creation flags are not tied to the map object but
324 * rather to the map fd instead, so they have no meaning upon
325 * map object inspection since multiple file descriptors with
326 * different (access) properties can exist here. Thus, given
327 * this has zero meaning for the map itself, lets clear these
330 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
333 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
335 map->map_type = attr->map_type;
336 map->key_size = attr->key_size;
337 map->value_size = attr->value_size;
338 map->max_entries = attr->max_entries;
339 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
340 map->numa_node = bpf_map_attr_numa_node(attr);
343 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
345 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
347 if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
348 atomic_long_sub(pages, &user->locked_vm);
354 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
357 atomic_long_sub(pages, &user->locked_vm);
360 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size)
362 u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
363 struct user_struct *user;
366 if (size >= U32_MAX - PAGE_SIZE)
369 user = get_current_user();
370 ret = bpf_charge_memlock(user, pages);
382 void bpf_map_charge_finish(struct bpf_map_memory *mem)
384 bpf_uncharge_memlock(mem->user, mem->pages);
388 void bpf_map_charge_move(struct bpf_map_memory *dst,
389 struct bpf_map_memory *src)
393 /* Make sure src will not be used for the redundant uncharging. */
394 memset(src, 0, sizeof(struct bpf_map_memory));
397 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
401 ret = bpf_charge_memlock(map->memory.user, pages);
404 map->memory.pages += pages;
408 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
410 bpf_uncharge_memlock(map->memory.user, pages);
411 map->memory.pages -= pages;
414 static int bpf_map_alloc_id(struct bpf_map *map)
418 idr_preload(GFP_KERNEL);
419 spin_lock_bh(&map_idr_lock);
420 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
423 spin_unlock_bh(&map_idr_lock);
426 if (WARN_ON_ONCE(!id))
429 return id > 0 ? 0 : id;
432 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
436 /* Offloaded maps are removed from the IDR store when their device
437 * disappears - even if someone holds an fd to them they are unusable,
438 * the memory is gone, all ops will fail; they are simply waiting for
439 * refcnt to drop to be freed.
445 spin_lock_irqsave(&map_idr_lock, flags);
447 __acquire(&map_idr_lock);
449 idr_remove(&map_idr, map->id);
453 spin_unlock_irqrestore(&map_idr_lock, flags);
455 __release(&map_idr_lock);
458 /* called from workqueue */
459 static void bpf_map_free_deferred(struct work_struct *work)
461 struct bpf_map *map = container_of(work, struct bpf_map, work);
462 struct bpf_map_memory mem;
464 bpf_map_charge_move(&mem, &map->memory);
465 security_bpf_map_free(map);
466 /* implementation dependent freeing */
467 map->ops->map_free(map);
468 bpf_map_charge_finish(&mem);
471 static void bpf_map_put_uref(struct bpf_map *map)
473 if (atomic64_dec_and_test(&map->usercnt)) {
474 if (map->ops->map_release_uref)
475 map->ops->map_release_uref(map);
479 /* decrement map refcnt and schedule it for freeing via workqueue
480 * (unrelying map implementation ops->map_free() might sleep)
482 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
484 if (atomic64_dec_and_test(&map->refcnt)) {
485 /* bpf_map_free_id() must be called first */
486 bpf_map_free_id(map, do_idr_lock);
488 INIT_WORK(&map->work, bpf_map_free_deferred);
489 schedule_work(&map->work);
493 void bpf_map_put(struct bpf_map *map)
495 __bpf_map_put(map, true);
497 EXPORT_SYMBOL_GPL(bpf_map_put);
499 void bpf_map_put_with_uref(struct bpf_map *map)
501 bpf_map_put_uref(map);
505 static int bpf_map_release(struct inode *inode, struct file *filp)
507 struct bpf_map *map = filp->private_data;
509 if (map->ops->map_release)
510 map->ops->map_release(map, filp);
512 bpf_map_put_with_uref(map);
516 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
518 fmode_t mode = f.file->f_mode;
520 /* Our file permissions may have been overridden by global
521 * map permissions facing syscall side.
523 if (READ_ONCE(map->frozen))
524 mode &= ~FMODE_CAN_WRITE;
528 #ifdef CONFIG_PROC_FS
529 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
531 const struct bpf_map *map = filp->private_data;
532 const struct bpf_array *array;
533 u32 type = 0, jited = 0;
535 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
536 array = container_of(map, struct bpf_array, map);
537 type = array->aux->type;
538 jited = array->aux->jited;
555 map->memory.pages * 1ULL << PAGE_SHIFT,
557 READ_ONCE(map->frozen));
559 seq_printf(m, "owner_prog_type:\t%u\n", type);
560 seq_printf(m, "owner_jited:\t%u\n", jited);
565 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
568 /* We need this handler such that alloc_file() enables
569 * f_mode with FMODE_CAN_READ.
574 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
575 size_t siz, loff_t *ppos)
577 /* We need this handler such that alloc_file() enables
578 * f_mode with FMODE_CAN_WRITE.
583 /* called for any extra memory-mapped regions (except initial) */
584 static void bpf_map_mmap_open(struct vm_area_struct *vma)
586 struct bpf_map *map = vma->vm_file->private_data;
588 bpf_map_inc_with_uref(map);
590 if (vma->vm_flags & VM_WRITE) {
591 mutex_lock(&map->freeze_mutex);
593 mutex_unlock(&map->freeze_mutex);
597 /* called for all unmapped memory region (including initial) */
598 static void bpf_map_mmap_close(struct vm_area_struct *vma)
600 struct bpf_map *map = vma->vm_file->private_data;
602 if (vma->vm_flags & VM_WRITE) {
603 mutex_lock(&map->freeze_mutex);
605 mutex_unlock(&map->freeze_mutex);
608 bpf_map_put_with_uref(map);
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) && map->frozen) {
634 /* set default open/close callbacks */
635 vma->vm_ops = &bpf_map_default_vmops;
636 vma->vm_private_data = map;
638 err = map->ops->map_mmap(map, vma);
642 bpf_map_inc_with_uref(map);
644 if (vma->vm_flags & VM_WRITE)
647 mutex_unlock(&map->freeze_mutex);
651 const struct file_operations bpf_map_fops = {
652 #ifdef CONFIG_PROC_FS
653 .show_fdinfo = bpf_map_show_fdinfo,
655 .release = bpf_map_release,
656 .read = bpf_dummy_read,
657 .write = bpf_dummy_write,
658 .mmap = bpf_map_mmap,
661 int bpf_map_new_fd(struct bpf_map *map, int flags)
665 ret = security_bpf_map(map, OPEN_FMODE(flags));
669 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
673 int bpf_get_file_flag(int flags)
675 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
677 if (flags & BPF_F_RDONLY)
679 if (flags & BPF_F_WRONLY)
684 /* helper macro to check that unused fields 'union bpf_attr' are zero */
685 #define CHECK_ATTR(CMD) \
686 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
687 sizeof(attr->CMD##_LAST_FIELD), 0, \
689 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
690 sizeof(attr->CMD##_LAST_FIELD)) != NULL
692 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
693 * Return 0 on success and < 0 on error.
695 static int bpf_obj_name_cpy(char *dst, const char *src)
697 const char *end = src + BPF_OBJ_NAME_LEN;
699 memset(dst, 0, BPF_OBJ_NAME_LEN);
700 /* Copy all isalnum(), '_' and '.' chars. */
701 while (src < end && *src) {
702 if (!isalnum(*src) &&
703 *src != '_' && *src != '.')
708 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
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);
810 atomic64_set(&map->refcnt, 1);
811 atomic64_set(&map->usercnt, 1);
812 mutex_init(&map->freeze_mutex);
814 map->spin_lock_off = -EINVAL;
815 if (attr->btf_key_type_id || attr->btf_value_type_id ||
816 /* Even the map's value is a kernel's struct,
817 * the bpf_prog.o must have BTF to begin with
818 * to figure out the corresponding kernel's
819 * counter part. Thus, attr->btf_fd has
822 attr->btf_vmlinux_value_type_id) {
825 btf = btf_get_by_fd(attr->btf_fd);
832 if (attr->btf_value_type_id) {
833 err = map_check_btf(map, btf, attr->btf_key_type_id,
834 attr->btf_value_type_id);
839 map->btf_key_type_id = attr->btf_key_type_id;
840 map->btf_value_type_id = attr->btf_value_type_id;
841 map->btf_vmlinux_value_type_id =
842 attr->btf_vmlinux_value_type_id;
845 err = security_bpf_map_alloc(map);
849 err = bpf_map_alloc_id(map);
853 err = bpf_map_new_fd(map, f_flags);
855 /* failed to allocate fd.
856 * bpf_map_put_with_uref() is needed because the above
857 * bpf_map_alloc_id() has published the map
858 * to the userspace and the userspace may
859 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
861 bpf_map_put_with_uref(map);
868 security_bpf_map_free(map);
871 bpf_map_charge_move(&mem, &map->memory);
872 map->ops->map_free(map);
873 bpf_map_charge_finish(&mem);
877 /* if error is returned, fd is released.
878 * On success caller should complete fd access with matching fdput()
880 struct bpf_map *__bpf_map_get(struct fd f)
883 return ERR_PTR(-EBADF);
884 if (f.file->f_op != &bpf_map_fops) {
886 return ERR_PTR(-EINVAL);
889 return f.file->private_data;
892 void bpf_map_inc(struct bpf_map *map)
894 atomic64_inc(&map->refcnt);
896 EXPORT_SYMBOL_GPL(bpf_map_inc);
898 void bpf_map_inc_with_uref(struct bpf_map *map)
900 atomic64_inc(&map->refcnt);
901 atomic64_inc(&map->usercnt);
903 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
905 struct bpf_map *bpf_map_get(u32 ufd)
907 struct fd f = fdget(ufd);
910 map = __bpf_map_get(f);
920 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
922 struct fd f = fdget(ufd);
925 map = __bpf_map_get(f);
929 bpf_map_inc_with_uref(map);
935 /* map_idr_lock should have been held */
936 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
940 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
942 return ERR_PTR(-ENOENT);
944 atomic64_inc(&map->usercnt);
949 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
951 spin_lock_bh(&map_idr_lock);
952 map = __bpf_map_inc_not_zero(map, false);
953 spin_unlock_bh(&map_idr_lock);
957 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
959 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
964 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
967 return memdup_user(ukey, key_size);
970 return ERR_PTR(-EINVAL);
975 /* last field in 'union bpf_attr' used by this command */
976 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
978 static int map_lookup_elem(union bpf_attr *attr)
980 void __user *ukey = u64_to_user_ptr(attr->key);
981 void __user *uvalue = u64_to_user_ptr(attr->value);
982 int ufd = attr->map_fd;
989 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
992 if (attr->flags & ~BPF_F_LOCK)
996 map = __bpf_map_get(f);
999 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1004 if ((attr->flags & BPF_F_LOCK) &&
1005 !map_value_has_spin_lock(map)) {
1010 key = __bpf_copy_key(ukey, map->key_size);
1016 value_size = bpf_map_value_size(map);
1019 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1023 err = bpf_map_copy_value(map, key, value, attr->flags);
1028 if (copy_to_user(uvalue, value, value_size) != 0)
1043 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1045 static int map_update_elem(union bpf_attr *attr)
1047 void __user *ukey = u64_to_user_ptr(attr->key);
1048 void __user *uvalue = u64_to_user_ptr(attr->value);
1049 int ufd = attr->map_fd;
1050 struct bpf_map *map;
1056 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1060 map = __bpf_map_get(f);
1062 return PTR_ERR(map);
1063 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1068 if ((attr->flags & BPF_F_LOCK) &&
1069 !map_value_has_spin_lock(map)) {
1074 key = __bpf_copy_key(ukey, map->key_size);
1080 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1081 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
1082 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
1083 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
1084 value_size = round_up(map->value_size, 8) * num_possible_cpus();
1086 value_size = map->value_size;
1089 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1094 if (copy_from_user(value, uvalue, value_size) != 0)
1097 err = bpf_map_update_value(map, f, key, value, attr->flags);
1108 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1110 static int map_delete_elem(union bpf_attr *attr)
1112 void __user *ukey = u64_to_user_ptr(attr->key);
1113 int ufd = attr->map_fd;
1114 struct bpf_map *map;
1119 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1123 map = __bpf_map_get(f);
1125 return PTR_ERR(map);
1126 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1131 key = __bpf_copy_key(ukey, map->key_size);
1137 if (bpf_map_is_dev_bound(map)) {
1138 err = bpf_map_offload_delete_elem(map, key);
1140 } else if (IS_FD_PROG_ARRAY(map) ||
1141 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1142 /* These maps require sleepable context */
1143 err = map->ops->map_delete_elem(map, key);
1147 bpf_disable_instrumentation();
1149 err = map->ops->map_delete_elem(map, key);
1151 bpf_enable_instrumentation();
1152 maybe_wait_bpf_programs(map);
1160 /* last field in 'union bpf_attr' used by this command */
1161 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1163 static int map_get_next_key(union bpf_attr *attr)
1165 void __user *ukey = u64_to_user_ptr(attr->key);
1166 void __user *unext_key = u64_to_user_ptr(attr->next_key);
1167 int ufd = attr->map_fd;
1168 struct bpf_map *map;
1169 void *key, *next_key;
1173 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1177 map = __bpf_map_get(f);
1179 return PTR_ERR(map);
1180 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1186 key = __bpf_copy_key(ukey, map->key_size);
1196 next_key = kmalloc(map->key_size, GFP_USER);
1200 if (bpf_map_is_dev_bound(map)) {
1201 err = bpf_map_offload_get_next_key(map, key, next_key);
1206 err = map->ops->map_get_next_key(map, key, next_key);
1213 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1227 int generic_map_delete_batch(struct bpf_map *map,
1228 const union bpf_attr *attr,
1229 union bpf_attr __user *uattr)
1231 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1236 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1239 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1240 !map_value_has_spin_lock(map)) {
1244 max_count = attr->batch.count;
1248 key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1252 for (cp = 0; cp < max_count; cp++) {
1254 if (copy_from_user(key, keys + cp * map->key_size,
1258 if (bpf_map_is_dev_bound(map)) {
1259 err = bpf_map_offload_delete_elem(map, key);
1263 bpf_disable_instrumentation();
1265 err = map->ops->map_delete_elem(map, key);
1267 bpf_enable_instrumentation();
1268 maybe_wait_bpf_programs(map);
1272 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1279 int generic_map_update_batch(struct bpf_map *map,
1280 const union bpf_attr *attr,
1281 union bpf_attr __user *uattr)
1283 void __user *values = u64_to_user_ptr(attr->batch.values);
1284 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1285 u32 value_size, cp, max_count;
1286 int ufd = attr->map_fd;
1292 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1295 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1296 !map_value_has_spin_lock(map)) {
1300 value_size = bpf_map_value_size(map);
1302 max_count = attr->batch.count;
1306 key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1310 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1316 for (cp = 0; cp < max_count; cp++) {
1318 if (copy_from_user(key, keys + cp * map->key_size,
1320 copy_from_user(value, values + cp * value_size, value_size))
1323 err = bpf_map_update_value(map, f, key, value,
1324 attr->batch.elem_flags);
1330 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1338 #define MAP_LOOKUP_RETRIES 3
1340 int generic_map_lookup_batch(struct bpf_map *map,
1341 const union bpf_attr *attr,
1342 union bpf_attr __user *uattr)
1344 void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1345 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1346 void __user *values = u64_to_user_ptr(attr->batch.values);
1347 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1348 void *buf, *buf_prevkey, *prev_key, *key, *value;
1349 int err, retry = MAP_LOOKUP_RETRIES;
1350 u32 value_size, cp, max_count;
1352 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1355 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1356 !map_value_has_spin_lock(map))
1359 value_size = bpf_map_value_size(map);
1361 max_count = attr->batch.count;
1365 if (put_user(0, &uattr->batch.count))
1368 buf_prevkey = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1372 buf = kmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1374 kvfree(buf_prevkey);
1380 if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1383 value = key + map->key_size;
1385 prev_key = buf_prevkey;
1387 for (cp = 0; cp < max_count;) {
1389 err = map->ops->map_get_next_key(map, prev_key, key);
1393 err = bpf_map_copy_value(map, key, value,
1394 attr->batch.elem_flags);
1396 if (err == -ENOENT) {
1408 if (copy_to_user(keys + cp * map->key_size, key,
1413 if (copy_to_user(values + cp * value_size, value, value_size)) {
1419 prev_key = buf_prevkey;
1421 swap(prev_key, key);
1422 retry = MAP_LOOKUP_RETRIES;
1429 if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1430 (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1439 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1441 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1443 void __user *ukey = u64_to_user_ptr(attr->key);
1444 void __user *uvalue = u64_to_user_ptr(attr->value);
1445 int ufd = attr->map_fd;
1446 struct bpf_map *map;
1452 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1456 map = __bpf_map_get(f);
1458 return PTR_ERR(map);
1459 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1464 key = __bpf_copy_key(ukey, map->key_size);
1470 value_size = map->value_size;
1473 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1477 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1478 map->map_type == BPF_MAP_TYPE_STACK) {
1479 err = map->ops->map_pop_elem(map, value);
1487 if (copy_to_user(uvalue, value, value_size) != 0)
1501 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1503 static int map_freeze(const union bpf_attr *attr)
1505 int err = 0, ufd = attr->map_fd;
1506 struct bpf_map *map;
1509 if (CHECK_ATTR(BPF_MAP_FREEZE))
1513 map = __bpf_map_get(f);
1515 return PTR_ERR(map);
1517 mutex_lock(&map->freeze_mutex);
1519 if (map->writecnt) {
1523 if (READ_ONCE(map->frozen)) {
1527 if (!capable(CAP_SYS_ADMIN)) {
1532 WRITE_ONCE(map->frozen, true);
1534 mutex_unlock(&map->freeze_mutex);
1539 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1540 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1541 [_id] = & _name ## _prog_ops,
1542 #define BPF_MAP_TYPE(_id, _ops)
1543 #include <linux/bpf_types.h>
1544 #undef BPF_PROG_TYPE
1548 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1550 const struct bpf_prog_ops *ops;
1552 if (type >= ARRAY_SIZE(bpf_prog_types))
1554 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1555 ops = bpf_prog_types[type];
1559 if (!bpf_prog_is_dev_bound(prog->aux))
1560 prog->aux->ops = ops;
1562 prog->aux->ops = &bpf_offload_prog_ops;
1573 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1574 [BPF_AUDIT_LOAD] = "LOAD",
1575 [BPF_AUDIT_UNLOAD] = "UNLOAD",
1578 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1580 struct audit_context *ctx = NULL;
1581 struct audit_buffer *ab;
1583 if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1585 if (audit_enabled == AUDIT_OFF)
1587 if (op == BPF_AUDIT_LOAD)
1588 ctx = audit_context();
1589 ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1592 audit_log_format(ab, "prog-id=%u op=%s",
1593 prog->aux->id, bpf_audit_str[op]);
1597 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1599 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1600 unsigned long user_bufs;
1603 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1604 if (user_bufs > memlock_limit) {
1605 atomic_long_sub(pages, &user->locked_vm);
1613 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1616 atomic_long_sub(pages, &user->locked_vm);
1619 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1621 struct user_struct *user = get_current_user();
1624 ret = __bpf_prog_charge(user, prog->pages);
1630 prog->aux->user = user;
1634 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1636 struct user_struct *user = prog->aux->user;
1638 __bpf_prog_uncharge(user, prog->pages);
1642 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1646 idr_preload(GFP_KERNEL);
1647 spin_lock_bh(&prog_idr_lock);
1648 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1651 spin_unlock_bh(&prog_idr_lock);
1654 /* id is in [1, INT_MAX) */
1655 if (WARN_ON_ONCE(!id))
1658 return id > 0 ? 0 : id;
1661 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1663 /* cBPF to eBPF migrations are currently not in the idr store.
1664 * Offloaded programs are removed from the store when their device
1665 * disappears - even if someone grabs an fd to them they are unusable,
1666 * simply waiting for refcnt to drop to be freed.
1672 spin_lock_bh(&prog_idr_lock);
1674 __acquire(&prog_idr_lock);
1676 idr_remove(&prog_idr, prog->aux->id);
1680 spin_unlock_bh(&prog_idr_lock);
1682 __release(&prog_idr_lock);
1685 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1687 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1689 kvfree(aux->func_info);
1690 kfree(aux->func_info_aux);
1691 bpf_prog_uncharge_memlock(aux->prog);
1692 security_bpf_prog_free(aux);
1693 bpf_prog_free(aux->prog);
1696 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1698 bpf_prog_kallsyms_del_all(prog);
1699 btf_put(prog->aux->btf);
1700 bpf_prog_free_linfo(prog);
1703 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1705 __bpf_prog_put_rcu(&prog->aux->rcu);
1708 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1710 if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1711 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1712 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
1713 /* bpf_prog_free_id() must be called first */
1714 bpf_prog_free_id(prog, do_idr_lock);
1715 __bpf_prog_put_noref(prog, true);
1719 void bpf_prog_put(struct bpf_prog *prog)
1721 __bpf_prog_put(prog, true);
1723 EXPORT_SYMBOL_GPL(bpf_prog_put);
1725 static int bpf_prog_release(struct inode *inode, struct file *filp)
1727 struct bpf_prog *prog = filp->private_data;
1733 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1734 struct bpf_prog_stats *stats)
1736 u64 nsecs = 0, cnt = 0;
1739 for_each_possible_cpu(cpu) {
1740 const struct bpf_prog_stats *st;
1744 st = per_cpu_ptr(prog->aux->stats, cpu);
1746 start = u64_stats_fetch_begin_irq(&st->syncp);
1749 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1753 stats->nsecs = nsecs;
1757 #ifdef CONFIG_PROC_FS
1758 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1760 const struct bpf_prog *prog = filp->private_data;
1761 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1762 struct bpf_prog_stats stats;
1764 bpf_prog_get_stats(prog, &stats);
1765 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1772 "run_time_ns:\t%llu\n"
1777 prog->pages * 1ULL << PAGE_SHIFT,
1784 const struct file_operations bpf_prog_fops = {
1785 #ifdef CONFIG_PROC_FS
1786 .show_fdinfo = bpf_prog_show_fdinfo,
1788 .release = bpf_prog_release,
1789 .read = bpf_dummy_read,
1790 .write = bpf_dummy_write,
1793 int bpf_prog_new_fd(struct bpf_prog *prog)
1797 ret = security_bpf_prog(prog);
1801 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1802 O_RDWR | O_CLOEXEC);
1805 static struct bpf_prog *____bpf_prog_get(struct fd f)
1808 return ERR_PTR(-EBADF);
1809 if (f.file->f_op != &bpf_prog_fops) {
1811 return ERR_PTR(-EINVAL);
1814 return f.file->private_data;
1817 void bpf_prog_add(struct bpf_prog *prog, int i)
1819 atomic64_add(i, &prog->aux->refcnt);
1821 EXPORT_SYMBOL_GPL(bpf_prog_add);
1823 void bpf_prog_sub(struct bpf_prog *prog, int i)
1825 /* Only to be used for undoing previous bpf_prog_add() in some
1826 * error path. We still know that another entity in our call
1827 * path holds a reference to the program, thus atomic_sub() can
1828 * be safely used in such cases!
1830 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1832 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1834 void bpf_prog_inc(struct bpf_prog *prog)
1836 atomic64_inc(&prog->aux->refcnt);
1838 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1840 /* prog_idr_lock should have been held */
1841 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1845 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1848 return ERR_PTR(-ENOENT);
1852 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1854 bool bpf_prog_get_ok(struct bpf_prog *prog,
1855 enum bpf_prog_type *attach_type, bool attach_drv)
1857 /* not an attachment, just a refcount inc, always allow */
1861 if (prog->type != *attach_type)
1863 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1869 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1872 struct fd f = fdget(ufd);
1873 struct bpf_prog *prog;
1875 prog = ____bpf_prog_get(f);
1878 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1879 prog = ERR_PTR(-EINVAL);
1889 struct bpf_prog *bpf_prog_get(u32 ufd)
1891 return __bpf_prog_get(ufd, NULL, false);
1894 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1897 return __bpf_prog_get(ufd, &type, attach_drv);
1899 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1901 /* Initially all BPF programs could be loaded w/o specifying
1902 * expected_attach_type. Later for some of them specifying expected_attach_type
1903 * at load time became required so that program could be validated properly.
1904 * Programs of types that are allowed to be loaded both w/ and w/o (for
1905 * backward compatibility) expected_attach_type, should have the default attach
1906 * type assigned to expected_attach_type for the latter case, so that it can be
1907 * validated later at attach time.
1909 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1910 * prog type requires it but has some attach types that have to be backward
1913 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1915 switch (attr->prog_type) {
1916 case BPF_PROG_TYPE_CGROUP_SOCK:
1917 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1918 * exist so checking for non-zero is the way to go here.
1920 if (!attr->expected_attach_type)
1921 attr->expected_attach_type =
1922 BPF_CGROUP_INET_SOCK_CREATE;
1928 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1929 enum bpf_attach_type expected_attach_type,
1930 u32 btf_id, u32 prog_fd)
1933 if (btf_id > BTF_MAX_TYPE)
1936 switch (prog_type) {
1937 case BPF_PROG_TYPE_TRACING:
1938 case BPF_PROG_TYPE_STRUCT_OPS:
1939 case BPF_PROG_TYPE_EXT:
1946 if (prog_fd && prog_type != BPF_PROG_TYPE_TRACING &&
1947 prog_type != BPF_PROG_TYPE_EXT)
1950 switch (prog_type) {
1951 case BPF_PROG_TYPE_CGROUP_SOCK:
1952 switch (expected_attach_type) {
1953 case BPF_CGROUP_INET_SOCK_CREATE:
1954 case BPF_CGROUP_INET4_POST_BIND:
1955 case BPF_CGROUP_INET6_POST_BIND:
1960 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1961 switch (expected_attach_type) {
1962 case BPF_CGROUP_INET4_BIND:
1963 case BPF_CGROUP_INET6_BIND:
1964 case BPF_CGROUP_INET4_CONNECT:
1965 case BPF_CGROUP_INET6_CONNECT:
1966 case BPF_CGROUP_UDP4_SENDMSG:
1967 case BPF_CGROUP_UDP6_SENDMSG:
1968 case BPF_CGROUP_UDP4_RECVMSG:
1969 case BPF_CGROUP_UDP6_RECVMSG:
1974 case BPF_PROG_TYPE_CGROUP_SKB:
1975 switch (expected_attach_type) {
1976 case BPF_CGROUP_INET_INGRESS:
1977 case BPF_CGROUP_INET_EGRESS:
1982 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1983 switch (expected_attach_type) {
1984 case BPF_CGROUP_SETSOCKOPT:
1985 case BPF_CGROUP_GETSOCKOPT:
1990 case BPF_PROG_TYPE_EXT:
1991 if (expected_attach_type)
1999 /* last field in 'union bpf_attr' used by this command */
2000 #define BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
2002 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
2004 enum bpf_prog_type type = attr->prog_type;
2005 struct bpf_prog *prog;
2010 if (CHECK_ATTR(BPF_PROG_LOAD))
2013 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2014 BPF_F_ANY_ALIGNMENT |
2015 BPF_F_TEST_STATE_FREQ |
2016 BPF_F_TEST_RND_HI32))
2019 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2020 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2021 !capable(CAP_SYS_ADMIN))
2024 /* copy eBPF program license from user space */
2025 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
2026 sizeof(license) - 1) < 0)
2028 license[sizeof(license) - 1] = 0;
2030 /* eBPF programs must be GPL compatible to use GPL-ed functions */
2031 is_gpl = license_is_gpl_compatible(license);
2033 if (attr->insn_cnt == 0 ||
2034 attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2036 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2037 type != BPF_PROG_TYPE_CGROUP_SKB &&
2038 !capable(CAP_SYS_ADMIN))
2041 bpf_prog_load_fixup_attach_type(attr);
2042 if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2043 attr->attach_btf_id,
2044 attr->attach_prog_fd))
2047 /* plain bpf_prog allocation */
2048 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2052 prog->expected_attach_type = attr->expected_attach_type;
2053 prog->aux->attach_btf_id = attr->attach_btf_id;
2054 if (attr->attach_prog_fd) {
2055 struct bpf_prog *tgt_prog;
2057 tgt_prog = bpf_prog_get(attr->attach_prog_fd);
2058 if (IS_ERR(tgt_prog)) {
2059 err = PTR_ERR(tgt_prog);
2060 goto free_prog_nouncharge;
2062 prog->aux->linked_prog = tgt_prog;
2065 prog->aux->offload_requested = !!attr->prog_ifindex;
2067 err = security_bpf_prog_alloc(prog->aux);
2069 goto free_prog_nouncharge;
2071 err = bpf_prog_charge_memlock(prog);
2075 prog->len = attr->insn_cnt;
2078 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
2079 bpf_prog_insn_size(prog)) != 0)
2082 prog->orig_prog = NULL;
2085 atomic64_set(&prog->aux->refcnt, 1);
2086 prog->gpl_compatible = is_gpl ? 1 : 0;
2088 if (bpf_prog_is_dev_bound(prog->aux)) {
2089 err = bpf_prog_offload_init(prog, attr);
2094 /* find program type: socket_filter vs tracing_filter */
2095 err = find_prog_type(type, prog);
2099 prog->aux->load_time = ktime_get_boottime_ns();
2100 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
2104 /* run eBPF verifier */
2105 err = bpf_check(&prog, attr, uattr);
2107 goto free_used_maps;
2109 prog = bpf_prog_select_runtime(prog, &err);
2111 goto free_used_maps;
2113 err = bpf_prog_alloc_id(prog);
2115 goto free_used_maps;
2117 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2118 * effectively publicly exposed. However, retrieving via
2119 * bpf_prog_get_fd_by_id() will take another reference,
2120 * therefore it cannot be gone underneath us.
2122 * Only for the time /after/ successful bpf_prog_new_fd()
2123 * and before returning to userspace, we might just hold
2124 * one reference and any parallel close on that fd could
2125 * rip everything out. Hence, below notifications must
2126 * happen before bpf_prog_new_fd().
2128 * Also, any failure handling from this point onwards must
2129 * be using bpf_prog_put() given the program is exposed.
2131 bpf_prog_kallsyms_add(prog);
2132 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2133 bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2135 err = bpf_prog_new_fd(prog);
2141 /* In case we have subprogs, we need to wait for a grace
2142 * period before we can tear down JIT memory since symbols
2143 * are already exposed under kallsyms.
2145 __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2148 bpf_prog_uncharge_memlock(prog);
2150 security_bpf_prog_free(prog->aux);
2151 free_prog_nouncharge:
2152 bpf_prog_free(prog);
2156 #define BPF_OBJ_LAST_FIELD file_flags
2158 static int bpf_obj_pin(const union bpf_attr *attr)
2160 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2163 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2166 static int bpf_obj_get(const union bpf_attr *attr)
2168 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2169 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2172 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2178 const struct bpf_link_ops *ops;
2179 struct bpf_prog *prog;
2180 struct work_struct work;
2183 void bpf_link_init(struct bpf_link *link, const struct bpf_link_ops *ops,
2184 struct bpf_prog *prog)
2186 atomic64_set(&link->refcnt, 1);
2191 void bpf_link_inc(struct bpf_link *link)
2193 atomic64_inc(&link->refcnt);
2196 /* bpf_link_free is guaranteed to be called from process context */
2197 static void bpf_link_free(struct bpf_link *link)
2199 struct bpf_prog *prog;
2201 /* remember prog locally, because release below will free link memory */
2203 /* extra clean up and kfree of container link struct */
2204 link->ops->release(link);
2205 /* no more accesing of link members after this point */
2209 static void bpf_link_put_deferred(struct work_struct *work)
2211 struct bpf_link *link = container_of(work, struct bpf_link, work);
2213 bpf_link_free(link);
2216 /* bpf_link_put can be called from atomic context, but ensures that resources
2217 * are freed from process context
2219 void bpf_link_put(struct bpf_link *link)
2221 if (!atomic64_dec_and_test(&link->refcnt))
2225 INIT_WORK(&link->work, bpf_link_put_deferred);
2226 schedule_work(&link->work);
2228 bpf_link_free(link);
2232 static int bpf_link_release(struct inode *inode, struct file *filp)
2234 struct bpf_link *link = filp->private_data;
2240 #ifdef CONFIG_PROC_FS
2241 static const struct bpf_link_ops bpf_raw_tp_lops;
2242 static const struct bpf_link_ops bpf_tracing_link_lops;
2243 static const struct bpf_link_ops bpf_xdp_link_lops;
2245 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2247 const struct bpf_link *link = filp->private_data;
2248 const struct bpf_prog *prog = link->prog;
2249 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2250 const char *link_type;
2252 if (link->ops == &bpf_raw_tp_lops)
2253 link_type = "raw_tracepoint";
2254 else if (link->ops == &bpf_tracing_link_lops)
2255 link_type = "tracing";
2257 link_type = "unknown";
2259 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2270 const struct file_operations bpf_link_fops = {
2271 #ifdef CONFIG_PROC_FS
2272 .show_fdinfo = bpf_link_show_fdinfo,
2274 .release = bpf_link_release,
2275 .read = bpf_dummy_read,
2276 .write = bpf_dummy_write,
2279 int bpf_link_new_fd(struct bpf_link *link)
2281 return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2284 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2286 struct fd f = fdget(ufd);
2287 struct bpf_link *link;
2290 return ERR_PTR(-EBADF);
2291 if (f.file->f_op != &bpf_link_fops) {
2293 return ERR_PTR(-EINVAL);
2296 link = f.file->private_data;
2303 struct bpf_tracing_link {
2304 struct bpf_link link;
2307 static void bpf_tracing_link_release(struct bpf_link *link)
2309 struct bpf_tracing_link *tr_link =
2310 container_of(link, struct bpf_tracing_link, link);
2312 WARN_ON_ONCE(bpf_trampoline_unlink_prog(link->prog));
2316 static const struct bpf_link_ops bpf_tracing_link_lops = {
2317 .release = bpf_tracing_link_release,
2320 static int bpf_tracing_prog_attach(struct bpf_prog *prog)
2322 struct bpf_tracing_link *link;
2325 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2326 prog->expected_attach_type != BPF_TRACE_FEXIT &&
2327 prog->type != BPF_PROG_TYPE_EXT) {
2332 link = kzalloc(sizeof(*link), GFP_USER);
2337 bpf_link_init(&link->link, &bpf_tracing_link_lops, prog);
2339 err = bpf_trampoline_link_prog(prog);
2343 link_fd = bpf_link_new_fd(&link->link);
2345 WARN_ON_ONCE(bpf_trampoline_unlink_prog(prog));
2358 struct bpf_raw_tp_link {
2359 struct bpf_link link;
2360 struct bpf_raw_event_map *btp;
2363 static void bpf_raw_tp_link_release(struct bpf_link *link)
2365 struct bpf_raw_tp_link *raw_tp =
2366 container_of(link, struct bpf_raw_tp_link, link);
2368 bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
2369 bpf_put_raw_tracepoint(raw_tp->btp);
2373 static const struct bpf_link_ops bpf_raw_tp_lops = {
2374 .release = bpf_raw_tp_link_release,
2377 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
2379 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
2381 struct bpf_raw_tp_link *raw_tp;
2382 struct bpf_raw_event_map *btp;
2383 struct bpf_prog *prog;
2384 const char *tp_name;
2388 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
2391 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
2393 return PTR_ERR(prog);
2395 if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
2396 prog->type != BPF_PROG_TYPE_TRACING &&
2397 prog->type != BPF_PROG_TYPE_EXT &&
2398 prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
2403 if (prog->type == BPF_PROG_TYPE_TRACING ||
2404 prog->type == BPF_PROG_TYPE_EXT) {
2405 if (attr->raw_tracepoint.name) {
2406 /* The attach point for this category of programs
2407 * should be specified via btf_id during program load.
2412 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
2413 tp_name = prog->aux->attach_func_name;
2415 return bpf_tracing_prog_attach(prog);
2417 if (strncpy_from_user(buf,
2418 u64_to_user_ptr(attr->raw_tracepoint.name),
2419 sizeof(buf) - 1) < 0) {
2423 buf[sizeof(buf) - 1] = 0;
2427 btp = bpf_get_raw_tracepoint(tp_name);
2433 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
2438 bpf_link_init(&raw_tp->link, &bpf_raw_tp_lops, prog);
2441 err = bpf_probe_register(raw_tp->btp, prog);
2445 link_fd = bpf_link_new_fd(&raw_tp->link);
2447 bpf_probe_unregister(raw_tp->btp, prog);
2456 bpf_put_raw_tracepoint(btp);
2462 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2463 enum bpf_attach_type attach_type)
2465 switch (prog->type) {
2466 case BPF_PROG_TYPE_CGROUP_SOCK:
2467 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2468 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2469 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2470 case BPF_PROG_TYPE_CGROUP_SKB:
2471 return prog->enforce_expected_attach_type &&
2472 prog->expected_attach_type != attach_type ?
2479 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
2481 #define BPF_F_ATTACH_MASK \
2482 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
2484 static int bpf_prog_attach(const union bpf_attr *attr)
2486 enum bpf_prog_type ptype;
2487 struct bpf_prog *prog;
2490 if (!capable(CAP_NET_ADMIN))
2493 if (CHECK_ATTR(BPF_PROG_ATTACH))
2496 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
2499 switch (attr->attach_type) {
2500 case BPF_CGROUP_INET_INGRESS:
2501 case BPF_CGROUP_INET_EGRESS:
2502 ptype = BPF_PROG_TYPE_CGROUP_SKB;
2504 case BPF_CGROUP_INET_SOCK_CREATE:
2505 case BPF_CGROUP_INET4_POST_BIND:
2506 case BPF_CGROUP_INET6_POST_BIND:
2507 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2509 case BPF_CGROUP_INET4_BIND:
2510 case BPF_CGROUP_INET6_BIND:
2511 case BPF_CGROUP_INET4_CONNECT:
2512 case BPF_CGROUP_INET6_CONNECT:
2513 case BPF_CGROUP_UDP4_SENDMSG:
2514 case BPF_CGROUP_UDP6_SENDMSG:
2515 case BPF_CGROUP_UDP4_RECVMSG:
2516 case BPF_CGROUP_UDP6_RECVMSG:
2517 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2519 case BPF_CGROUP_SOCK_OPS:
2520 ptype = BPF_PROG_TYPE_SOCK_OPS;
2522 case BPF_CGROUP_DEVICE:
2523 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2525 case BPF_SK_MSG_VERDICT:
2526 ptype = BPF_PROG_TYPE_SK_MSG;
2528 case BPF_SK_SKB_STREAM_PARSER:
2529 case BPF_SK_SKB_STREAM_VERDICT:
2530 ptype = BPF_PROG_TYPE_SK_SKB;
2532 case BPF_LIRC_MODE2:
2533 ptype = BPF_PROG_TYPE_LIRC_MODE2;
2535 case BPF_FLOW_DISSECTOR:
2536 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
2538 case BPF_CGROUP_SYSCTL:
2539 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2541 case BPF_CGROUP_GETSOCKOPT:
2542 case BPF_CGROUP_SETSOCKOPT:
2543 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2549 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
2551 return PTR_ERR(prog);
2553 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
2559 case BPF_PROG_TYPE_SK_SKB:
2560 case BPF_PROG_TYPE_SK_MSG:
2561 ret = sock_map_get_from_fd(attr, prog);
2563 case BPF_PROG_TYPE_LIRC_MODE2:
2564 ret = lirc_prog_attach(attr, prog);
2566 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2567 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
2570 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
2578 #define BPF_PROG_DETACH_LAST_FIELD attach_type
2580 static int bpf_prog_detach(const union bpf_attr *attr)
2582 enum bpf_prog_type ptype;
2584 if (!capable(CAP_NET_ADMIN))
2587 if (CHECK_ATTR(BPF_PROG_DETACH))
2590 switch (attr->attach_type) {
2591 case BPF_CGROUP_INET_INGRESS:
2592 case BPF_CGROUP_INET_EGRESS:
2593 ptype = BPF_PROG_TYPE_CGROUP_SKB;
2595 case BPF_CGROUP_INET_SOCK_CREATE:
2596 case BPF_CGROUP_INET4_POST_BIND:
2597 case BPF_CGROUP_INET6_POST_BIND:
2598 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2600 case BPF_CGROUP_INET4_BIND:
2601 case BPF_CGROUP_INET6_BIND:
2602 case BPF_CGROUP_INET4_CONNECT:
2603 case BPF_CGROUP_INET6_CONNECT:
2604 case BPF_CGROUP_UDP4_SENDMSG:
2605 case BPF_CGROUP_UDP6_SENDMSG:
2606 case BPF_CGROUP_UDP4_RECVMSG:
2607 case BPF_CGROUP_UDP6_RECVMSG:
2608 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2610 case BPF_CGROUP_SOCK_OPS:
2611 ptype = BPF_PROG_TYPE_SOCK_OPS;
2613 case BPF_CGROUP_DEVICE:
2614 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2616 case BPF_SK_MSG_VERDICT:
2617 return sock_map_get_from_fd(attr, NULL);
2618 case BPF_SK_SKB_STREAM_PARSER:
2619 case BPF_SK_SKB_STREAM_VERDICT:
2620 return sock_map_get_from_fd(attr, NULL);
2621 case BPF_LIRC_MODE2:
2622 return lirc_prog_detach(attr);
2623 case BPF_FLOW_DISSECTOR:
2624 return skb_flow_dissector_bpf_prog_detach(attr);
2625 case BPF_CGROUP_SYSCTL:
2626 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2628 case BPF_CGROUP_GETSOCKOPT:
2629 case BPF_CGROUP_SETSOCKOPT:
2630 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2636 return cgroup_bpf_prog_detach(attr, ptype);
2639 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2641 static int bpf_prog_query(const union bpf_attr *attr,
2642 union bpf_attr __user *uattr)
2644 if (!capable(CAP_NET_ADMIN))
2646 if (CHECK_ATTR(BPF_PROG_QUERY))
2648 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2651 switch (attr->query.attach_type) {
2652 case BPF_CGROUP_INET_INGRESS:
2653 case BPF_CGROUP_INET_EGRESS:
2654 case BPF_CGROUP_INET_SOCK_CREATE:
2655 case BPF_CGROUP_INET4_BIND:
2656 case BPF_CGROUP_INET6_BIND:
2657 case BPF_CGROUP_INET4_POST_BIND:
2658 case BPF_CGROUP_INET6_POST_BIND:
2659 case BPF_CGROUP_INET4_CONNECT:
2660 case BPF_CGROUP_INET6_CONNECT:
2661 case BPF_CGROUP_UDP4_SENDMSG:
2662 case BPF_CGROUP_UDP6_SENDMSG:
2663 case BPF_CGROUP_UDP4_RECVMSG:
2664 case BPF_CGROUP_UDP6_RECVMSG:
2665 case BPF_CGROUP_SOCK_OPS:
2666 case BPF_CGROUP_DEVICE:
2667 case BPF_CGROUP_SYSCTL:
2668 case BPF_CGROUP_GETSOCKOPT:
2669 case BPF_CGROUP_SETSOCKOPT:
2671 case BPF_LIRC_MODE2:
2672 return lirc_prog_query(attr, uattr);
2673 case BPF_FLOW_DISSECTOR:
2674 return skb_flow_dissector_prog_query(attr, uattr);
2679 return cgroup_bpf_prog_query(attr, uattr);
2682 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2684 static int bpf_prog_test_run(const union bpf_attr *attr,
2685 union bpf_attr __user *uattr)
2687 struct bpf_prog *prog;
2688 int ret = -ENOTSUPP;
2690 if (!capable(CAP_SYS_ADMIN))
2692 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2695 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2696 (!attr->test.ctx_size_in && attr->test.ctx_in))
2699 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2700 (!attr->test.ctx_size_out && attr->test.ctx_out))
2703 prog = bpf_prog_get(attr->test.prog_fd);
2705 return PTR_ERR(prog);
2707 if (prog->aux->ops->test_run)
2708 ret = prog->aux->ops->test_run(prog, attr, uattr);
2714 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2716 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2717 union bpf_attr __user *uattr,
2721 u32 next_id = attr->start_id;
2724 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2727 if (!capable(CAP_SYS_ADMIN))
2732 if (!idr_get_next(idr, &next_id))
2734 spin_unlock_bh(lock);
2737 err = put_user(next_id, &uattr->next_id);
2742 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2744 struct bpf_prog *bpf_prog_by_id(u32 id)
2746 struct bpf_prog *prog;
2749 return ERR_PTR(-ENOENT);
2751 spin_lock_bh(&prog_idr_lock);
2752 prog = idr_find(&prog_idr, id);
2754 prog = bpf_prog_inc_not_zero(prog);
2756 prog = ERR_PTR(-ENOENT);
2757 spin_unlock_bh(&prog_idr_lock);
2761 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2763 struct bpf_prog *prog;
2764 u32 id = attr->prog_id;
2767 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2770 if (!capable(CAP_SYS_ADMIN))
2773 prog = bpf_prog_by_id(id);
2775 return PTR_ERR(prog);
2777 fd = bpf_prog_new_fd(prog);
2784 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2786 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2788 struct bpf_map *map;
2789 u32 id = attr->map_id;
2793 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2794 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2797 if (!capable(CAP_SYS_ADMIN))
2800 f_flags = bpf_get_file_flag(attr->open_flags);
2804 spin_lock_bh(&map_idr_lock);
2805 map = idr_find(&map_idr, id);
2807 map = __bpf_map_inc_not_zero(map, true);
2809 map = ERR_PTR(-ENOENT);
2810 spin_unlock_bh(&map_idr_lock);
2813 return PTR_ERR(map);
2815 fd = bpf_map_new_fd(map, f_flags);
2817 bpf_map_put_with_uref(map);
2822 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2823 unsigned long addr, u32 *off,
2826 const struct bpf_map *map;
2829 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2830 map = prog->aux->used_maps[i];
2831 if (map == (void *)addr) {
2832 *type = BPF_PSEUDO_MAP_FD;
2835 if (!map->ops->map_direct_value_meta)
2837 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2838 *type = BPF_PSEUDO_MAP_VALUE;
2846 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2848 const struct bpf_map *map;
2849 struct bpf_insn *insns;
2854 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2859 for (i = 0; i < prog->len; i++) {
2860 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2861 insns[i].code = BPF_JMP | BPF_CALL;
2862 insns[i].imm = BPF_FUNC_tail_call;
2865 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2866 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2867 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2868 insns[i].code = BPF_JMP | BPF_CALL;
2869 if (!bpf_dump_raw_ok())
2874 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2877 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2878 map = bpf_map_from_imm(prog, imm, &off, &type);
2880 insns[i].src_reg = type;
2881 insns[i].imm = map->id;
2882 insns[i + 1].imm = off;
2890 static int set_info_rec_size(struct bpf_prog_info *info)
2893 * Ensure info.*_rec_size is the same as kernel expected size
2897 * Only allow zero *_rec_size if both _rec_size and _cnt are
2898 * zero. In this case, the kernel will set the expected
2899 * _rec_size back to the info.
2902 if ((info->nr_func_info || info->func_info_rec_size) &&
2903 info->func_info_rec_size != sizeof(struct bpf_func_info))
2906 if ((info->nr_line_info || info->line_info_rec_size) &&
2907 info->line_info_rec_size != sizeof(struct bpf_line_info))
2910 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2911 info->jited_line_info_rec_size != sizeof(__u64))
2914 info->func_info_rec_size = sizeof(struct bpf_func_info);
2915 info->line_info_rec_size = sizeof(struct bpf_line_info);
2916 info->jited_line_info_rec_size = sizeof(__u64);
2921 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2922 const union bpf_attr *attr,
2923 union bpf_attr __user *uattr)
2925 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2926 struct bpf_prog_info info = {};
2927 u32 info_len = attr->info.info_len;
2928 struct bpf_prog_stats stats;
2929 char __user *uinsns;
2933 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2936 info_len = min_t(u32, sizeof(info), info_len);
2938 if (copy_from_user(&info, uinfo, info_len))
2941 info.type = prog->type;
2942 info.id = prog->aux->id;
2943 info.load_time = prog->aux->load_time;
2944 info.created_by_uid = from_kuid_munged(current_user_ns(),
2945 prog->aux->user->uid);
2946 info.gpl_compatible = prog->gpl_compatible;
2948 memcpy(info.tag, prog->tag, sizeof(prog->tag));
2949 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2951 ulen = info.nr_map_ids;
2952 info.nr_map_ids = prog->aux->used_map_cnt;
2953 ulen = min_t(u32, info.nr_map_ids, ulen);
2955 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2958 for (i = 0; i < ulen; i++)
2959 if (put_user(prog->aux->used_maps[i]->id,
2964 err = set_info_rec_size(&info);
2968 bpf_prog_get_stats(prog, &stats);
2969 info.run_time_ns = stats.nsecs;
2970 info.run_cnt = stats.cnt;
2972 if (!capable(CAP_SYS_ADMIN)) {
2973 info.jited_prog_len = 0;
2974 info.xlated_prog_len = 0;
2975 info.nr_jited_ksyms = 0;
2976 info.nr_jited_func_lens = 0;
2977 info.nr_func_info = 0;
2978 info.nr_line_info = 0;
2979 info.nr_jited_line_info = 0;
2983 ulen = info.xlated_prog_len;
2984 info.xlated_prog_len = bpf_prog_insn_size(prog);
2985 if (info.xlated_prog_len && ulen) {
2986 struct bpf_insn *insns_sanitized;
2989 if (prog->blinded && !bpf_dump_raw_ok()) {
2990 info.xlated_prog_insns = 0;
2993 insns_sanitized = bpf_insn_prepare_dump(prog);
2994 if (!insns_sanitized)
2996 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2997 ulen = min_t(u32, info.xlated_prog_len, ulen);
2998 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2999 kfree(insns_sanitized);
3004 if (bpf_prog_is_dev_bound(prog->aux)) {
3005 err = bpf_prog_offload_info_fill(&info, prog);
3011 /* NOTE: the following code is supposed to be skipped for offload.
3012 * bpf_prog_offload_info_fill() is the place to fill similar fields
3015 ulen = info.jited_prog_len;
3016 if (prog->aux->func_cnt) {
3019 info.jited_prog_len = 0;
3020 for (i = 0; i < prog->aux->func_cnt; i++)
3021 info.jited_prog_len += prog->aux->func[i]->jited_len;
3023 info.jited_prog_len = prog->jited_len;
3026 if (info.jited_prog_len && ulen) {
3027 if (bpf_dump_raw_ok()) {
3028 uinsns = u64_to_user_ptr(info.jited_prog_insns);
3029 ulen = min_t(u32, info.jited_prog_len, ulen);
3031 /* for multi-function programs, copy the JITed
3032 * instructions for all the functions
3034 if (prog->aux->func_cnt) {
3039 for (i = 0; i < prog->aux->func_cnt; i++) {
3040 len = prog->aux->func[i]->jited_len;
3041 len = min_t(u32, len, free);
3042 img = (u8 *) prog->aux->func[i]->bpf_func;
3043 if (copy_to_user(uinsns, img, len))
3051 if (copy_to_user(uinsns, prog->bpf_func, ulen))
3055 info.jited_prog_insns = 0;
3059 ulen = info.nr_jited_ksyms;
3060 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
3062 if (bpf_dump_raw_ok()) {
3063 unsigned long ksym_addr;
3064 u64 __user *user_ksyms;
3067 /* copy the address of the kernel symbol
3068 * corresponding to each function
3070 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
3071 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
3072 if (prog->aux->func_cnt) {
3073 for (i = 0; i < ulen; i++) {
3074 ksym_addr = (unsigned long)
3075 prog->aux->func[i]->bpf_func;
3076 if (put_user((u64) ksym_addr,
3081 ksym_addr = (unsigned long) prog->bpf_func;
3082 if (put_user((u64) ksym_addr, &user_ksyms[0]))
3086 info.jited_ksyms = 0;
3090 ulen = info.nr_jited_func_lens;
3091 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
3093 if (bpf_dump_raw_ok()) {
3094 u32 __user *user_lens;
3097 /* copy the JITed image lengths for each function */
3098 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
3099 user_lens = u64_to_user_ptr(info.jited_func_lens);
3100 if (prog->aux->func_cnt) {
3101 for (i = 0; i < ulen; i++) {
3103 prog->aux->func[i]->jited_len;
3104 if (put_user(func_len, &user_lens[i]))
3108 func_len = prog->jited_len;
3109 if (put_user(func_len, &user_lens[0]))
3113 info.jited_func_lens = 0;
3118 info.btf_id = btf_id(prog->aux->btf);
3120 ulen = info.nr_func_info;
3121 info.nr_func_info = prog->aux->func_info_cnt;
3122 if (info.nr_func_info && ulen) {
3123 char __user *user_finfo;
3125 user_finfo = u64_to_user_ptr(info.func_info);
3126 ulen = min_t(u32, info.nr_func_info, ulen);
3127 if (copy_to_user(user_finfo, prog->aux->func_info,
3128 info.func_info_rec_size * ulen))
3132 ulen = info.nr_line_info;
3133 info.nr_line_info = prog->aux->nr_linfo;
3134 if (info.nr_line_info && ulen) {
3135 __u8 __user *user_linfo;
3137 user_linfo = u64_to_user_ptr(info.line_info);
3138 ulen = min_t(u32, info.nr_line_info, ulen);
3139 if (copy_to_user(user_linfo, prog->aux->linfo,
3140 info.line_info_rec_size * ulen))
3144 ulen = info.nr_jited_line_info;
3145 if (prog->aux->jited_linfo)
3146 info.nr_jited_line_info = prog->aux->nr_linfo;
3148 info.nr_jited_line_info = 0;
3149 if (info.nr_jited_line_info && ulen) {
3150 if (bpf_dump_raw_ok()) {
3151 __u64 __user *user_linfo;
3154 user_linfo = u64_to_user_ptr(info.jited_line_info);
3155 ulen = min_t(u32, info.nr_jited_line_info, ulen);
3156 for (i = 0; i < ulen; i++) {
3157 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
3162 info.jited_line_info = 0;
3166 ulen = info.nr_prog_tags;
3167 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
3169 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
3172 user_prog_tags = u64_to_user_ptr(info.prog_tags);
3173 ulen = min_t(u32, info.nr_prog_tags, ulen);
3174 if (prog->aux->func_cnt) {
3175 for (i = 0; i < ulen; i++) {
3176 if (copy_to_user(user_prog_tags[i],
3177 prog->aux->func[i]->tag,
3182 if (copy_to_user(user_prog_tags[0],
3183 prog->tag, BPF_TAG_SIZE))
3189 if (copy_to_user(uinfo, &info, info_len) ||
3190 put_user(info_len, &uattr->info.info_len))
3196 static int bpf_map_get_info_by_fd(struct bpf_map *map,
3197 const union bpf_attr *attr,
3198 union bpf_attr __user *uattr)
3200 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3201 struct bpf_map_info info = {};
3202 u32 info_len = attr->info.info_len;
3205 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3208 info_len = min_t(u32, sizeof(info), info_len);
3210 info.type = map->map_type;
3212 info.key_size = map->key_size;
3213 info.value_size = map->value_size;
3214 info.max_entries = map->max_entries;
3215 info.map_flags = map->map_flags;
3216 memcpy(info.name, map->name, sizeof(map->name));
3219 info.btf_id = btf_id(map->btf);
3220 info.btf_key_type_id = map->btf_key_type_id;
3221 info.btf_value_type_id = map->btf_value_type_id;
3223 info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
3225 if (bpf_map_is_dev_bound(map)) {
3226 err = bpf_map_offload_info_fill(&info, map);
3231 if (copy_to_user(uinfo, &info, info_len) ||
3232 put_user(info_len, &uattr->info.info_len))
3238 static int bpf_btf_get_info_by_fd(struct btf *btf,
3239 const union bpf_attr *attr,
3240 union bpf_attr __user *uattr)
3242 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3243 u32 info_len = attr->info.info_len;
3246 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
3250 return btf_get_info_by_fd(btf, attr, uattr);
3253 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
3255 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
3256 union bpf_attr __user *uattr)
3258 int ufd = attr->info.bpf_fd;
3262 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
3269 if (f.file->f_op == &bpf_prog_fops)
3270 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
3272 else if (f.file->f_op == &bpf_map_fops)
3273 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
3275 else if (f.file->f_op == &btf_fops)
3276 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
3284 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
3286 static int bpf_btf_load(const union bpf_attr *attr)
3288 if (CHECK_ATTR(BPF_BTF_LOAD))
3291 if (!capable(CAP_SYS_ADMIN))
3294 return btf_new_fd(attr);
3297 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
3299 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
3301 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
3304 if (!capable(CAP_SYS_ADMIN))
3307 return btf_get_fd_by_id(attr->btf_id);
3310 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
3311 union bpf_attr __user *uattr,
3312 u32 prog_id, u32 fd_type,
3313 const char *buf, u64 probe_offset,
3316 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
3317 u32 len = buf ? strlen(buf) : 0, input_len;
3320 if (put_user(len, &uattr->task_fd_query.buf_len))
3322 input_len = attr->task_fd_query.buf_len;
3323 if (input_len && ubuf) {
3325 /* nothing to copy, just make ubuf NULL terminated */
3328 if (put_user(zero, ubuf))
3330 } else if (input_len >= len + 1) {
3331 /* ubuf can hold the string with NULL terminator */
3332 if (copy_to_user(ubuf, buf, len + 1))
3335 /* ubuf cannot hold the string with NULL terminator,
3336 * do a partial copy with NULL terminator.
3341 if (copy_to_user(ubuf, buf, input_len - 1))
3343 if (put_user(zero, ubuf + input_len - 1))
3348 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
3349 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
3350 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
3351 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
3357 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
3359 static int bpf_task_fd_query(const union bpf_attr *attr,
3360 union bpf_attr __user *uattr)
3362 pid_t pid = attr->task_fd_query.pid;
3363 u32 fd = attr->task_fd_query.fd;
3364 const struct perf_event *event;
3365 struct files_struct *files;
3366 struct task_struct *task;
3370 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
3373 if (!capable(CAP_SYS_ADMIN))
3376 if (attr->task_fd_query.flags != 0)
3379 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3383 files = get_files_struct(task);
3384 put_task_struct(task);
3389 spin_lock(&files->file_lock);
3390 file = fcheck_files(files, fd);
3395 spin_unlock(&files->file_lock);
3396 put_files_struct(files);
3401 if (file->f_op == &bpf_link_fops) {
3402 struct bpf_link *link = file->private_data;
3404 if (link->ops == &bpf_raw_tp_lops) {
3405 struct bpf_raw_tp_link *raw_tp =
3406 container_of(link, struct bpf_raw_tp_link, link);
3407 struct bpf_raw_event_map *btp = raw_tp->btp;
3409 err = bpf_task_fd_query_copy(attr, uattr,
3410 raw_tp->link.prog->aux->id,
3411 BPF_FD_TYPE_RAW_TRACEPOINT,
3412 btp->tp->name, 0, 0);
3418 event = perf_get_event(file);
3419 if (!IS_ERR(event)) {
3420 u64 probe_offset, probe_addr;
3421 u32 prog_id, fd_type;
3424 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
3425 &buf, &probe_offset,
3428 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
3443 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
3445 #define BPF_DO_BATCH(fn) \
3451 err = fn(map, attr, uattr); \
3454 static int bpf_map_do_batch(const union bpf_attr *attr,
3455 union bpf_attr __user *uattr,
3458 struct bpf_map *map;
3462 if (CHECK_ATTR(BPF_MAP_BATCH))
3465 ufd = attr->batch.map_fd;
3467 map = __bpf_map_get(f);
3469 return PTR_ERR(map);
3471 if ((cmd == BPF_MAP_LOOKUP_BATCH ||
3472 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH) &&
3473 !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
3478 if (cmd != BPF_MAP_LOOKUP_BATCH &&
3479 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
3484 if (cmd == BPF_MAP_LOOKUP_BATCH)
3485 BPF_DO_BATCH(map->ops->map_lookup_batch);
3486 else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
3487 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
3488 else if (cmd == BPF_MAP_UPDATE_BATCH)
3489 BPF_DO_BATCH(map->ops->map_update_batch);
3491 BPF_DO_BATCH(map->ops->map_delete_batch);
3498 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
3500 union bpf_attr attr = {};
3503 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
3506 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
3509 size = min_t(u32, size, sizeof(attr));
3511 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
3512 if (copy_from_user(&attr, uattr, size) != 0)
3515 err = security_bpf(cmd, &attr, size);
3520 case BPF_MAP_CREATE:
3521 err = map_create(&attr);
3523 case BPF_MAP_LOOKUP_ELEM:
3524 err = map_lookup_elem(&attr);
3526 case BPF_MAP_UPDATE_ELEM:
3527 err = map_update_elem(&attr);
3529 case BPF_MAP_DELETE_ELEM:
3530 err = map_delete_elem(&attr);
3532 case BPF_MAP_GET_NEXT_KEY:
3533 err = map_get_next_key(&attr);
3535 case BPF_MAP_FREEZE:
3536 err = map_freeze(&attr);
3539 err = bpf_prog_load(&attr, uattr);
3542 err = bpf_obj_pin(&attr);
3545 err = bpf_obj_get(&attr);
3547 case BPF_PROG_ATTACH:
3548 err = bpf_prog_attach(&attr);
3550 case BPF_PROG_DETACH:
3551 err = bpf_prog_detach(&attr);
3553 case BPF_PROG_QUERY:
3554 err = bpf_prog_query(&attr, uattr);
3556 case BPF_PROG_TEST_RUN:
3557 err = bpf_prog_test_run(&attr, uattr);
3559 case BPF_PROG_GET_NEXT_ID:
3560 err = bpf_obj_get_next_id(&attr, uattr,
3561 &prog_idr, &prog_idr_lock);
3563 case BPF_MAP_GET_NEXT_ID:
3564 err = bpf_obj_get_next_id(&attr, uattr,
3565 &map_idr, &map_idr_lock);
3567 case BPF_BTF_GET_NEXT_ID:
3568 err = bpf_obj_get_next_id(&attr, uattr,
3569 &btf_idr, &btf_idr_lock);
3571 case BPF_PROG_GET_FD_BY_ID:
3572 err = bpf_prog_get_fd_by_id(&attr);
3574 case BPF_MAP_GET_FD_BY_ID:
3575 err = bpf_map_get_fd_by_id(&attr);
3577 case BPF_OBJ_GET_INFO_BY_FD:
3578 err = bpf_obj_get_info_by_fd(&attr, uattr);
3580 case BPF_RAW_TRACEPOINT_OPEN:
3581 err = bpf_raw_tracepoint_open(&attr);
3584 err = bpf_btf_load(&attr);
3586 case BPF_BTF_GET_FD_BY_ID:
3587 err = bpf_btf_get_fd_by_id(&attr);
3589 case BPF_TASK_FD_QUERY:
3590 err = bpf_task_fd_query(&attr, uattr);
3592 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
3593 err = map_lookup_and_delete_elem(&attr);
3595 case BPF_MAP_LOOKUP_BATCH:
3596 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_LOOKUP_BATCH);
3598 case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
3599 err = bpf_map_do_batch(&attr, uattr,
3600 BPF_MAP_LOOKUP_AND_DELETE_BATCH);
3602 case BPF_MAP_UPDATE_BATCH:
3603 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_UPDATE_BATCH);
3605 case BPF_MAP_DELETE_BATCH:
3606 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_DELETE_BATCH);