Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/* SPDX-License-Identifier: GPL-2.0-only */
2/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 */
4#ifndef _LINUX_BPF_H
5#define _LINUX_BPF_H 1
6
7#include <uapi/linux/bpf.h>
8#include <uapi/linux/filter.h>
9
10#include <crypto/sha2.h>
11#include <linux/workqueue.h>
12#include <linux/file.h>
13#include <linux/percpu.h>
14#include <linux/err.h>
15#include <linux/rbtree_latch.h>
16#include <linux/numa.h>
17#include <linux/mm_types.h>
18#include <linux/wait.h>
19#include <linux/refcount.h>
20#include <linux/mutex.h>
21#include <linux/module.h>
22#include <linux/kallsyms.h>
23#include <linux/capability.h>
24#include <linux/sched/mm.h>
25#include <linux/slab.h>
26#include <linux/percpu-refcount.h>
27#include <linux/stddef.h>
28#include <linux/bpfptr.h>
29#include <linux/btf.h>
30#include <linux/rcupdate_trace.h>
31#include <linux/static_call.h>
32#include <linux/memcontrol.h>
33#include <linux/cfi.h>
34#include <asm/rqspinlock.h>
35
36struct bpf_verifier_env;
37struct bpf_verifier_log;
38struct perf_event;
39struct bpf_prog;
40struct bpf_prog_aux;
41struct bpf_map;
42struct bpf_arena;
43struct sock;
44struct seq_file;
45struct btf;
46struct btf_type;
47struct exception_table_entry;
48struct seq_operations;
49struct bpf_iter_aux_info;
50struct bpf_local_storage;
51struct bpf_local_storage_map;
52struct kobject;
53struct mem_cgroup;
54struct module;
55struct bpf_func_state;
56struct ftrace_ops;
57struct cgroup;
58struct bpf_token;
59struct user_namespace;
60struct super_block;
61struct inode;
62
63extern struct idr btf_idr;
64extern spinlock_t btf_idr_lock;
65extern struct kobject *btf_kobj;
66extern struct bpf_mem_alloc bpf_global_ma, bpf_global_percpu_ma;
67extern bool bpf_global_ma_set;
68
69typedef u64 (*bpf_callback_t)(u64, u64, u64, u64, u64);
70typedef int (*bpf_iter_init_seq_priv_t)(void *private_data,
71 struct bpf_iter_aux_info *aux);
72typedef void (*bpf_iter_fini_seq_priv_t)(void *private_data);
73typedef unsigned int (*bpf_func_t)(const void *,
74 const struct bpf_insn *);
75struct bpf_iter_seq_info {
76 const struct seq_operations *seq_ops;
77 bpf_iter_init_seq_priv_t init_seq_private;
78 bpf_iter_fini_seq_priv_t fini_seq_private;
79 u32 seq_priv_size;
80};
81
82/* map is generic key/value storage optionally accessible by eBPF programs */
83struct bpf_map_ops {
84 /* funcs callable from userspace (via syscall) */
85 int (*map_alloc_check)(union bpf_attr *attr);
86 struct bpf_map *(*map_alloc)(union bpf_attr *attr);
87 void (*map_release)(struct bpf_map *map, struct file *map_file);
88 void (*map_free)(struct bpf_map *map);
89 int (*map_get_next_key)(struct bpf_map *map, void *key, void *next_key);
90 void (*map_release_uref)(struct bpf_map *map);
91 void *(*map_lookup_elem_sys_only)(struct bpf_map *map, void *key);
92 int (*map_lookup_batch)(struct bpf_map *map, const union bpf_attr *attr,
93 union bpf_attr __user *uattr);
94 int (*map_lookup_and_delete_elem)(struct bpf_map *map, void *key,
95 void *value, u64 flags);
96 int (*map_lookup_and_delete_batch)(struct bpf_map *map,
97 const union bpf_attr *attr,
98 union bpf_attr __user *uattr);
99 int (*map_update_batch)(struct bpf_map *map, struct file *map_file,
100 const union bpf_attr *attr,
101 union bpf_attr __user *uattr);
102 int (*map_delete_batch)(struct bpf_map *map, const union bpf_attr *attr,
103 union bpf_attr __user *uattr);
104
105 /* funcs callable from userspace and from eBPF programs */
106 void *(*map_lookup_elem)(struct bpf_map *map, void *key);
107 long (*map_update_elem)(struct bpf_map *map, void *key, void *value, u64 flags);
108 long (*map_delete_elem)(struct bpf_map *map, void *key);
109 long (*map_push_elem)(struct bpf_map *map, void *value, u64 flags);
110 long (*map_pop_elem)(struct bpf_map *map, void *value);
111 long (*map_peek_elem)(struct bpf_map *map, void *value);
112 void *(*map_lookup_percpu_elem)(struct bpf_map *map, void *key, u32 cpu);
113 int (*map_get_hash)(struct bpf_map *map, u32 hash_buf_size, void *hash_buf);
114
115 /* funcs called by prog_array and perf_event_array map */
116 void *(*map_fd_get_ptr)(struct bpf_map *map, struct file *map_file,
117 int fd);
118 /* If need_defer is true, the implementation should guarantee that
119 * the to-be-put element is still alive before the bpf program, which
120 * may manipulate it, exists.
121 */
122 void (*map_fd_put_ptr)(struct bpf_map *map, void *ptr, bool need_defer);
123 int (*map_gen_lookup)(struct bpf_map *map, struct bpf_insn *insn_buf);
124 u32 (*map_fd_sys_lookup_elem)(void *ptr);
125 void (*map_seq_show_elem)(struct bpf_map *map, void *key,
126 struct seq_file *m);
127 int (*map_check_btf)(struct bpf_map *map,
128 const struct btf *btf,
129 const struct btf_type *key_type,
130 const struct btf_type *value_type);
131
132 /* Prog poke tracking helpers. */
133 int (*map_poke_track)(struct bpf_map *map, struct bpf_prog_aux *aux);
134 void (*map_poke_untrack)(struct bpf_map *map, struct bpf_prog_aux *aux);
135 void (*map_poke_run)(struct bpf_map *map, u32 key, struct bpf_prog *old,
136 struct bpf_prog *new);
137
138 /* Direct value access helpers. */
139 int (*map_direct_value_addr)(const struct bpf_map *map,
140 u64 *imm, u32 off);
141 int (*map_direct_value_meta)(const struct bpf_map *map,
142 u64 imm, u32 *off);
143 int (*map_mmap)(struct bpf_map *map, struct vm_area_struct *vma);
144 __poll_t (*map_poll)(struct bpf_map *map, struct file *filp,
145 struct poll_table_struct *pts);
146 unsigned long (*map_get_unmapped_area)(struct file *filep, unsigned long addr,
147 unsigned long len, unsigned long pgoff,
148 unsigned long flags);
149
150 /* Functions called by bpf_local_storage maps */
151 int (*map_local_storage_charge)(struct bpf_local_storage_map *smap,
152 void *owner, u32 size);
153 void (*map_local_storage_uncharge)(struct bpf_local_storage_map *smap,
154 void *owner, u32 size);
155 struct bpf_local_storage __rcu ** (*map_owner_storage_ptr)(void *owner);
156
157 /* Misc helpers.*/
158 long (*map_redirect)(struct bpf_map *map, u64 key, u64 flags);
159
160 /* map_meta_equal must be implemented for maps that can be
161 * used as an inner map. It is a runtime check to ensure
162 * an inner map can be inserted to an outer map.
163 *
164 * Some properties of the inner map has been used during the
165 * verification time. When inserting an inner map at the runtime,
166 * map_meta_equal has to ensure the inserting map has the same
167 * properties that the verifier has used earlier.
168 */
169 bool (*map_meta_equal)(const struct bpf_map *meta0,
170 const struct bpf_map *meta1);
171
172
173 int (*map_set_for_each_callback_args)(struct bpf_verifier_env *env,
174 struct bpf_func_state *caller,
175 struct bpf_func_state *callee);
176 long (*map_for_each_callback)(struct bpf_map *map,
177 bpf_callback_t callback_fn,
178 void *callback_ctx, u64 flags);
179
180 u64 (*map_mem_usage)(const struct bpf_map *map);
181
182 /* BTF id of struct allocated by map_alloc */
183 int *map_btf_id;
184
185 /* bpf_iter info used to open a seq_file */
186 const struct bpf_iter_seq_info *iter_seq_info;
187};
188
189enum {
190 /* Support at most 11 fields in a BTF type */
191 BTF_FIELDS_MAX = 11,
192};
193
194enum btf_field_type {
195 BPF_SPIN_LOCK = (1 << 0),
196 BPF_TIMER = (1 << 1),
197 BPF_KPTR_UNREF = (1 << 2),
198 BPF_KPTR_REF = (1 << 3),
199 BPF_KPTR_PERCPU = (1 << 4),
200 BPF_KPTR = BPF_KPTR_UNREF | BPF_KPTR_REF | BPF_KPTR_PERCPU,
201 BPF_LIST_HEAD = (1 << 5),
202 BPF_LIST_NODE = (1 << 6),
203 BPF_RB_ROOT = (1 << 7),
204 BPF_RB_NODE = (1 << 8),
205 BPF_GRAPH_NODE = BPF_RB_NODE | BPF_LIST_NODE,
206 BPF_GRAPH_ROOT = BPF_RB_ROOT | BPF_LIST_HEAD,
207 BPF_REFCOUNT = (1 << 9),
208 BPF_WORKQUEUE = (1 << 10),
209 BPF_UPTR = (1 << 11),
210 BPF_RES_SPIN_LOCK = (1 << 12),
211 BPF_TASK_WORK = (1 << 13),
212};
213
214enum bpf_cgroup_storage_type {
215 BPF_CGROUP_STORAGE_SHARED,
216 BPF_CGROUP_STORAGE_PERCPU,
217 __BPF_CGROUP_STORAGE_MAX
218#define MAX_BPF_CGROUP_STORAGE_TYPE __BPF_CGROUP_STORAGE_MAX
219};
220
221#ifdef CONFIG_CGROUP_BPF
222# define for_each_cgroup_storage_type(stype) \
223 for (stype = 0; stype < MAX_BPF_CGROUP_STORAGE_TYPE; stype++)
224#else
225# define for_each_cgroup_storage_type(stype) for (; false; )
226#endif /* CONFIG_CGROUP_BPF */
227
228typedef void (*btf_dtor_kfunc_t)(void *);
229
230struct btf_field_kptr {
231 struct btf *btf;
232 struct module *module;
233 /* dtor used if btf_is_kernel(btf), otherwise the type is
234 * program-allocated, dtor is NULL, and __bpf_obj_drop_impl is used
235 */
236 btf_dtor_kfunc_t dtor;
237 u32 btf_id;
238};
239
240struct btf_field_graph_root {
241 struct btf *btf;
242 u32 value_btf_id;
243 u32 node_offset;
244 struct btf_record *value_rec;
245};
246
247struct btf_field {
248 u32 offset;
249 u32 size;
250 enum btf_field_type type;
251 union {
252 struct btf_field_kptr kptr;
253 struct btf_field_graph_root graph_root;
254 };
255};
256
257struct btf_record {
258 u32 cnt;
259 u32 field_mask;
260 int spin_lock_off;
261 int res_spin_lock_off;
262 int timer_off;
263 int wq_off;
264 int refcount_off;
265 int task_work_off;
266 struct btf_field fields[];
267};
268
269/* Non-opaque version of bpf_rb_node in uapi/linux/bpf.h */
270struct bpf_rb_node_kern {
271 struct rb_node rb_node;
272 void *owner;
273} __attribute__((aligned(8)));
274
275/* Non-opaque version of bpf_list_node in uapi/linux/bpf.h */
276struct bpf_list_node_kern {
277 struct list_head list_head;
278 void *owner;
279} __attribute__((aligned(8)));
280
281/* 'Ownership' of program-containing map is claimed by the first program
282 * that is going to use this map or by the first program which FD is
283 * stored in the map to make sure that all callers and callees have the
284 * same prog type, JITed flag and xdp_has_frags flag.
285 */
286struct bpf_map_owner {
287 enum bpf_prog_type type;
288 bool jited;
289 bool xdp_has_frags;
290 bool sleepable;
291 u64 storage_cookie[MAX_BPF_CGROUP_STORAGE_TYPE];
292 const struct btf_type *attach_func_proto;
293 enum bpf_attach_type expected_attach_type;
294};
295
296struct bpf_map {
297 u8 sha[SHA256_DIGEST_SIZE];
298 const struct bpf_map_ops *ops;
299 struct bpf_map *inner_map_meta;
300#ifdef CONFIG_SECURITY
301 void *security;
302#endif
303 enum bpf_map_type map_type;
304 u32 key_size;
305 u32 value_size;
306 u32 max_entries;
307 u64 map_extra; /* any per-map-type extra fields */
308 u32 map_flags;
309 u32 id;
310 struct btf_record *record;
311 int numa_node;
312 u32 btf_key_type_id;
313 u32 btf_value_type_id;
314 u32 btf_vmlinux_value_type_id;
315 struct btf *btf;
316#ifdef CONFIG_MEMCG
317 struct obj_cgroup *objcg;
318#endif
319 char name[BPF_OBJ_NAME_LEN];
320 struct mutex freeze_mutex;
321 atomic64_t refcnt;
322 atomic64_t usercnt;
323 /* rcu is used before freeing and work is only used during freeing */
324 union {
325 struct work_struct work;
326 struct rcu_head rcu;
327 };
328 atomic64_t writecnt;
329 spinlock_t owner_lock;
330 struct bpf_map_owner *owner;
331 bool bypass_spec_v1;
332 bool frozen; /* write-once; write-protected by freeze_mutex */
333 bool free_after_mult_rcu_gp;
334 bool free_after_rcu_gp;
335 atomic64_t sleepable_refcnt;
336 s64 __percpu *elem_count;
337 u64 cookie; /* write-once */
338 char *excl_prog_sha;
339};
340
341static inline const char *btf_field_type_name(enum btf_field_type type)
342{
343 switch (type) {
344 case BPF_SPIN_LOCK:
345 return "bpf_spin_lock";
346 case BPF_RES_SPIN_LOCK:
347 return "bpf_res_spin_lock";
348 case BPF_TIMER:
349 return "bpf_timer";
350 case BPF_WORKQUEUE:
351 return "bpf_wq";
352 case BPF_KPTR_UNREF:
353 case BPF_KPTR_REF:
354 return "kptr";
355 case BPF_KPTR_PERCPU:
356 return "percpu_kptr";
357 case BPF_UPTR:
358 return "uptr";
359 case BPF_LIST_HEAD:
360 return "bpf_list_head";
361 case BPF_LIST_NODE:
362 return "bpf_list_node";
363 case BPF_RB_ROOT:
364 return "bpf_rb_root";
365 case BPF_RB_NODE:
366 return "bpf_rb_node";
367 case BPF_REFCOUNT:
368 return "bpf_refcount";
369 case BPF_TASK_WORK:
370 return "bpf_task_work";
371 default:
372 WARN_ON_ONCE(1);
373 return "unknown";
374 }
375}
376
377#if IS_ENABLED(CONFIG_DEBUG_KERNEL)
378#define BPF_WARN_ONCE(cond, format...) WARN_ONCE(cond, format)
379#else
380#define BPF_WARN_ONCE(cond, format...) BUILD_BUG_ON_INVALID(cond)
381#endif
382
383static inline u32 btf_field_type_size(enum btf_field_type type)
384{
385 switch (type) {
386 case BPF_SPIN_LOCK:
387 return sizeof(struct bpf_spin_lock);
388 case BPF_RES_SPIN_LOCK:
389 return sizeof(struct bpf_res_spin_lock);
390 case BPF_TIMER:
391 return sizeof(struct bpf_timer);
392 case BPF_WORKQUEUE:
393 return sizeof(struct bpf_wq);
394 case BPF_KPTR_UNREF:
395 case BPF_KPTR_REF:
396 case BPF_KPTR_PERCPU:
397 case BPF_UPTR:
398 return sizeof(u64);
399 case BPF_LIST_HEAD:
400 return sizeof(struct bpf_list_head);
401 case BPF_LIST_NODE:
402 return sizeof(struct bpf_list_node);
403 case BPF_RB_ROOT:
404 return sizeof(struct bpf_rb_root);
405 case BPF_RB_NODE:
406 return sizeof(struct bpf_rb_node);
407 case BPF_REFCOUNT:
408 return sizeof(struct bpf_refcount);
409 case BPF_TASK_WORK:
410 return sizeof(struct bpf_task_work);
411 default:
412 WARN_ON_ONCE(1);
413 return 0;
414 }
415}
416
417static inline u32 btf_field_type_align(enum btf_field_type type)
418{
419 switch (type) {
420 case BPF_SPIN_LOCK:
421 return __alignof__(struct bpf_spin_lock);
422 case BPF_RES_SPIN_LOCK:
423 return __alignof__(struct bpf_res_spin_lock);
424 case BPF_TIMER:
425 return __alignof__(struct bpf_timer);
426 case BPF_WORKQUEUE:
427 return __alignof__(struct bpf_wq);
428 case BPF_KPTR_UNREF:
429 case BPF_KPTR_REF:
430 case BPF_KPTR_PERCPU:
431 case BPF_UPTR:
432 return __alignof__(u64);
433 case BPF_LIST_HEAD:
434 return __alignof__(struct bpf_list_head);
435 case BPF_LIST_NODE:
436 return __alignof__(struct bpf_list_node);
437 case BPF_RB_ROOT:
438 return __alignof__(struct bpf_rb_root);
439 case BPF_RB_NODE:
440 return __alignof__(struct bpf_rb_node);
441 case BPF_REFCOUNT:
442 return __alignof__(struct bpf_refcount);
443 case BPF_TASK_WORK:
444 return __alignof__(struct bpf_task_work);
445 default:
446 WARN_ON_ONCE(1);
447 return 0;
448 }
449}
450
451static inline void bpf_obj_init_field(const struct btf_field *field, void *addr)
452{
453 memset(addr, 0, field->size);
454
455 switch (field->type) {
456 case BPF_REFCOUNT:
457 refcount_set((refcount_t *)addr, 1);
458 break;
459 case BPF_RB_NODE:
460 RB_CLEAR_NODE((struct rb_node *)addr);
461 break;
462 case BPF_LIST_HEAD:
463 case BPF_LIST_NODE:
464 INIT_LIST_HEAD((struct list_head *)addr);
465 break;
466 case BPF_RB_ROOT:
467 /* RB_ROOT_CACHED 0-inits, no need to do anything after memset */
468 case BPF_SPIN_LOCK:
469 case BPF_RES_SPIN_LOCK:
470 case BPF_TIMER:
471 case BPF_WORKQUEUE:
472 case BPF_KPTR_UNREF:
473 case BPF_KPTR_REF:
474 case BPF_KPTR_PERCPU:
475 case BPF_UPTR:
476 case BPF_TASK_WORK:
477 break;
478 default:
479 WARN_ON_ONCE(1);
480 return;
481 }
482}
483
484static inline bool btf_record_has_field(const struct btf_record *rec, enum btf_field_type type)
485{
486 if (IS_ERR_OR_NULL(rec))
487 return false;
488 return rec->field_mask & type;
489}
490
491static inline void bpf_obj_init(const struct btf_record *rec, void *obj)
492{
493 int i;
494
495 if (IS_ERR_OR_NULL(rec))
496 return;
497 for (i = 0; i < rec->cnt; i++)
498 bpf_obj_init_field(&rec->fields[i], obj + rec->fields[i].offset);
499}
500
501/* 'dst' must be a temporary buffer and should not point to memory that is being
502 * used in parallel by a bpf program or bpf syscall, otherwise the access from
503 * the bpf program or bpf syscall may be corrupted by the reinitialization,
504 * leading to weird problems. Even 'dst' is newly-allocated from bpf memory
505 * allocator, it is still possible for 'dst' to be used in parallel by a bpf
506 * program or bpf syscall.
507 */
508static inline void check_and_init_map_value(struct bpf_map *map, void *dst)
509{
510 bpf_obj_init(map->record, dst);
511}
512
513/* memcpy that is used with 8-byte aligned pointers, power-of-8 size and
514 * forced to use 'long' read/writes to try to atomically copy long counters.
515 * Best-effort only. No barriers here, since it _will_ race with concurrent
516 * updates from BPF programs. Called from bpf syscall and mostly used with
517 * size 8 or 16 bytes, so ask compiler to inline it.
518 */
519static inline void bpf_long_memcpy(void *dst, const void *src, u32 size)
520{
521 const long *lsrc = src;
522 long *ldst = dst;
523
524 size /= sizeof(long);
525 while (size--)
526 data_race(*ldst++ = *lsrc++);
527}
528
529/* copy everything but bpf_spin_lock, bpf_timer, and kptrs. There could be one of each. */
530static inline void bpf_obj_memcpy(struct btf_record *rec,
531 void *dst, void *src, u32 size,
532 bool long_memcpy)
533{
534 u32 curr_off = 0;
535 int i;
536
537 if (IS_ERR_OR_NULL(rec)) {
538 if (long_memcpy)
539 bpf_long_memcpy(dst, src, round_up(size, 8));
540 else
541 memcpy(dst, src, size);
542 return;
543 }
544
545 for (i = 0; i < rec->cnt; i++) {
546 u32 next_off = rec->fields[i].offset;
547 u32 sz = next_off - curr_off;
548
549 memcpy(dst + curr_off, src + curr_off, sz);
550 curr_off += rec->fields[i].size + sz;
551 }
552 memcpy(dst + curr_off, src + curr_off, size - curr_off);
553}
554
555static inline void copy_map_value(struct bpf_map *map, void *dst, void *src)
556{
557 bpf_obj_memcpy(map->record, dst, src, map->value_size, false);
558}
559
560static inline void copy_map_value_long(struct bpf_map *map, void *dst, void *src)
561{
562 bpf_obj_memcpy(map->record, dst, src, map->value_size, true);
563}
564
565static inline void bpf_obj_swap_uptrs(const struct btf_record *rec, void *dst, void *src)
566{
567 unsigned long *src_uptr, *dst_uptr;
568 const struct btf_field *field;
569 int i;
570
571 if (!btf_record_has_field(rec, BPF_UPTR))
572 return;
573
574 for (i = 0, field = rec->fields; i < rec->cnt; i++, field++) {
575 if (field->type != BPF_UPTR)
576 continue;
577
578 src_uptr = src + field->offset;
579 dst_uptr = dst + field->offset;
580 swap(*src_uptr, *dst_uptr);
581 }
582}
583
584static inline void bpf_obj_memzero(struct btf_record *rec, void *dst, u32 size)
585{
586 u32 curr_off = 0;
587 int i;
588
589 if (IS_ERR_OR_NULL(rec)) {
590 memset(dst, 0, size);
591 return;
592 }
593
594 for (i = 0; i < rec->cnt; i++) {
595 u32 next_off = rec->fields[i].offset;
596 u32 sz = next_off - curr_off;
597
598 memset(dst + curr_off, 0, sz);
599 curr_off += rec->fields[i].size + sz;
600 }
601 memset(dst + curr_off, 0, size - curr_off);
602}
603
604static inline void zero_map_value(struct bpf_map *map, void *dst)
605{
606 bpf_obj_memzero(map->record, dst, map->value_size);
607}
608
609void copy_map_value_locked(struct bpf_map *map, void *dst, void *src,
610 bool lock_src);
611void bpf_timer_cancel_and_free(void *timer);
612void bpf_wq_cancel_and_free(void *timer);
613void bpf_task_work_cancel_and_free(void *timer);
614void bpf_list_head_free(const struct btf_field *field, void *list_head,
615 struct bpf_spin_lock *spin_lock);
616void bpf_rb_root_free(const struct btf_field *field, void *rb_root,
617 struct bpf_spin_lock *spin_lock);
618u64 bpf_arena_get_kern_vm_start(struct bpf_arena *arena);
619u64 bpf_arena_get_user_vm_start(struct bpf_arena *arena);
620int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size);
621
622struct bpf_offload_dev;
623struct bpf_offloaded_map;
624
625struct bpf_map_dev_ops {
626 int (*map_get_next_key)(struct bpf_offloaded_map *map,
627 void *key, void *next_key);
628 int (*map_lookup_elem)(struct bpf_offloaded_map *map,
629 void *key, void *value);
630 int (*map_update_elem)(struct bpf_offloaded_map *map,
631 void *key, void *value, u64 flags);
632 int (*map_delete_elem)(struct bpf_offloaded_map *map, void *key);
633};
634
635struct bpf_offloaded_map {
636 struct bpf_map map;
637 struct net_device *netdev;
638 const struct bpf_map_dev_ops *dev_ops;
639 void *dev_priv;
640 struct list_head offloads;
641};
642
643static inline struct bpf_offloaded_map *map_to_offmap(struct bpf_map *map)
644{
645 return container_of(map, struct bpf_offloaded_map, map);
646}
647
648static inline bool bpf_map_offload_neutral(const struct bpf_map *map)
649{
650 return map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
651}
652
653static inline bool bpf_map_support_seq_show(const struct bpf_map *map)
654{
655 return (map->btf_value_type_id || map->btf_vmlinux_value_type_id) &&
656 map->ops->map_seq_show_elem;
657}
658
659int map_check_no_btf(struct bpf_map *map,
660 const struct btf *btf,
661 const struct btf_type *key_type,
662 const struct btf_type *value_type);
663
664bool bpf_map_meta_equal(const struct bpf_map *meta0,
665 const struct bpf_map *meta1);
666
667static inline bool bpf_map_has_internal_structs(struct bpf_map *map)
668{
669 return btf_record_has_field(map->record, BPF_TIMER | BPF_WORKQUEUE | BPF_TASK_WORK);
670}
671
672void bpf_map_free_internal_structs(struct bpf_map *map, void *obj);
673
674int bpf_dynptr_from_file_sleepable(struct file *file, u32 flags,
675 struct bpf_dynptr *ptr__uninit);
676
677#if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
678void *bpf_arena_alloc_pages_non_sleepable(void *p__map, void *addr__ign, u32 page_cnt, int node_id,
679 u64 flags);
680void bpf_arena_free_pages_non_sleepable(void *p__map, void *ptr__ign, u32 page_cnt);
681#else
682static inline void *bpf_arena_alloc_pages_non_sleepable(void *p__map, void *addr__ign, u32 page_cnt,
683 int node_id, u64 flags)
684{
685 return NULL;
686}
687
688static inline void bpf_arena_free_pages_non_sleepable(void *p__map, void *ptr__ign, u32 page_cnt)
689{
690}
691#endif
692
693extern const struct bpf_map_ops bpf_map_offload_ops;
694
695/* bpf_type_flag contains a set of flags that are applicable to the values of
696 * arg_type, ret_type and reg_type. For example, a pointer value may be null,
697 * or a memory is read-only. We classify types into two categories: base types
698 * and extended types. Extended types are base types combined with a type flag.
699 *
700 * Currently there are no more than 32 base types in arg_type, ret_type and
701 * reg_types.
702 */
703#define BPF_BASE_TYPE_BITS 8
704
705enum bpf_type_flag {
706 /* PTR may be NULL. */
707 PTR_MAYBE_NULL = BIT(0 + BPF_BASE_TYPE_BITS),
708
709 /* MEM is read-only. When applied on bpf_arg, it indicates the arg is
710 * compatible with both mutable and immutable memory.
711 */
712 MEM_RDONLY = BIT(1 + BPF_BASE_TYPE_BITS),
713
714 /* MEM points to BPF ring buffer reservation. */
715 MEM_RINGBUF = BIT(2 + BPF_BASE_TYPE_BITS),
716
717 /* MEM is in user address space. */
718 MEM_USER = BIT(3 + BPF_BASE_TYPE_BITS),
719
720 /* MEM is a percpu memory. MEM_PERCPU tags PTR_TO_BTF_ID. When tagged
721 * with MEM_PERCPU, PTR_TO_BTF_ID _cannot_ be directly accessed. In
722 * order to drop this tag, it must be passed into bpf_per_cpu_ptr()
723 * or bpf_this_cpu_ptr(), which will return the pointer corresponding
724 * to the specified cpu.
725 */
726 MEM_PERCPU = BIT(4 + BPF_BASE_TYPE_BITS),
727
728 /* Indicates that the argument will be released. */
729 OBJ_RELEASE = BIT(5 + BPF_BASE_TYPE_BITS),
730
731 /* PTR is not trusted. This is only used with PTR_TO_BTF_ID, to mark
732 * unreferenced and referenced kptr loaded from map value using a load
733 * instruction, so that they can only be dereferenced but not escape the
734 * BPF program into the kernel (i.e. cannot be passed as arguments to
735 * kfunc or bpf helpers).
736 */
737 PTR_UNTRUSTED = BIT(6 + BPF_BASE_TYPE_BITS),
738
739 /* MEM can be uninitialized. */
740 MEM_UNINIT = BIT(7 + BPF_BASE_TYPE_BITS),
741
742 /* DYNPTR points to memory local to the bpf program. */
743 DYNPTR_TYPE_LOCAL = BIT(8 + BPF_BASE_TYPE_BITS),
744
745 /* DYNPTR points to a kernel-produced ringbuf record. */
746 DYNPTR_TYPE_RINGBUF = BIT(9 + BPF_BASE_TYPE_BITS),
747
748 /* Size is known at compile time. */
749 MEM_FIXED_SIZE = BIT(10 + BPF_BASE_TYPE_BITS),
750
751 /* MEM is of an allocated object of type in program BTF. This is used to
752 * tag PTR_TO_BTF_ID allocated using bpf_obj_new.
753 */
754 MEM_ALLOC = BIT(11 + BPF_BASE_TYPE_BITS),
755
756 /* PTR was passed from the kernel in a trusted context, and may be
757 * passed to kfuncs or BPF helper functions.
758 * Confusingly, this is _not_ the opposite of PTR_UNTRUSTED above.
759 * PTR_UNTRUSTED refers to a kptr that was read directly from a map
760 * without invoking bpf_kptr_xchg(). What we really need to know is
761 * whether a pointer is safe to pass to a kfunc or BPF helper function.
762 * While PTR_UNTRUSTED pointers are unsafe to pass to kfuncs and BPF
763 * helpers, they do not cover all possible instances of unsafe
764 * pointers. For example, a pointer that was obtained from walking a
765 * struct will _not_ get the PTR_UNTRUSTED type modifier, despite the
766 * fact that it may be NULL, invalid, etc. This is due to backwards
767 * compatibility requirements, as this was the behavior that was first
768 * introduced when kptrs were added. The behavior is now considered
769 * deprecated, and PTR_UNTRUSTED will eventually be removed.
770 *
771 * PTR_TRUSTED, on the other hand, is a pointer that the kernel
772 * guarantees to be valid and safe to pass to kfuncs and BPF helpers.
773 * For example, pointers passed to tracepoint arguments are considered
774 * PTR_TRUSTED, as are pointers that are passed to struct_ops
775 * callbacks. As alluded to above, pointers that are obtained from
776 * walking PTR_TRUSTED pointers are _not_ trusted. For example, if a
777 * struct task_struct *task is PTR_TRUSTED, then accessing
778 * task->last_wakee will lose the PTR_TRUSTED modifier when it's stored
779 * in a BPF register. Similarly, pointers passed to certain programs
780 * types such as kretprobes are not guaranteed to be valid, as they may
781 * for example contain an object that was recently freed.
782 */
783 PTR_TRUSTED = BIT(12 + BPF_BASE_TYPE_BITS),
784
785 /* MEM is tagged with rcu and memory access needs rcu_read_lock protection. */
786 MEM_RCU = BIT(13 + BPF_BASE_TYPE_BITS),
787
788 /* Used to tag PTR_TO_BTF_ID | MEM_ALLOC references which are non-owning.
789 * Currently only valid for linked-list and rbtree nodes. If the nodes
790 * have a bpf_refcount_field, they must be tagged MEM_RCU as well.
791 */
792 NON_OWN_REF = BIT(14 + BPF_BASE_TYPE_BITS),
793
794 /* DYNPTR points to sk_buff */
795 DYNPTR_TYPE_SKB = BIT(15 + BPF_BASE_TYPE_BITS),
796
797 /* DYNPTR points to xdp_buff */
798 DYNPTR_TYPE_XDP = BIT(16 + BPF_BASE_TYPE_BITS),
799
800 /* Memory must be aligned on some architectures, used in combination with
801 * MEM_FIXED_SIZE.
802 */
803 MEM_ALIGNED = BIT(17 + BPF_BASE_TYPE_BITS),
804
805 /* MEM is being written to, often combined with MEM_UNINIT. Non-presence
806 * of MEM_WRITE means that MEM is only being read. MEM_WRITE without the
807 * MEM_UNINIT means that memory needs to be initialized since it is also
808 * read.
809 */
810 MEM_WRITE = BIT(18 + BPF_BASE_TYPE_BITS),
811
812 /* DYNPTR points to skb_metadata_end()-skb_metadata_len() */
813 DYNPTR_TYPE_SKB_META = BIT(19 + BPF_BASE_TYPE_BITS),
814
815 /* DYNPTR points to file */
816 DYNPTR_TYPE_FILE = BIT(20 + BPF_BASE_TYPE_BITS),
817
818 __BPF_TYPE_FLAG_MAX,
819 __BPF_TYPE_LAST_FLAG = __BPF_TYPE_FLAG_MAX - 1,
820};
821
822#define DYNPTR_TYPE_FLAG_MASK (DYNPTR_TYPE_LOCAL | DYNPTR_TYPE_RINGBUF | DYNPTR_TYPE_SKB \
823 | DYNPTR_TYPE_XDP | DYNPTR_TYPE_SKB_META | DYNPTR_TYPE_FILE)
824
825/* Max number of base types. */
826#define BPF_BASE_TYPE_LIMIT (1UL << BPF_BASE_TYPE_BITS)
827
828/* Max number of all types. */
829#define BPF_TYPE_LIMIT (__BPF_TYPE_LAST_FLAG | (__BPF_TYPE_LAST_FLAG - 1))
830
831/* function argument constraints */
832enum bpf_arg_type {
833 ARG_DONTCARE = 0, /* unused argument in helper function */
834
835 /* the following constraints used to prototype
836 * bpf_map_lookup/update/delete_elem() functions
837 */
838 ARG_CONST_MAP_PTR, /* const argument used as pointer to bpf_map */
839 ARG_PTR_TO_MAP_KEY, /* pointer to stack used as map key */
840 ARG_PTR_TO_MAP_VALUE, /* pointer to stack used as map value */
841
842 /* Used to prototype bpf_memcmp() and other functions that access data
843 * on eBPF program stack
844 */
845 ARG_PTR_TO_MEM, /* pointer to valid memory (stack, packet, map value) */
846 ARG_PTR_TO_ARENA,
847
848 ARG_CONST_SIZE, /* number of bytes accessed from memory */
849 ARG_CONST_SIZE_OR_ZERO, /* number of bytes accessed from memory or 0 */
850
851 ARG_PTR_TO_CTX, /* pointer to context */
852 ARG_ANYTHING, /* any (initialized) argument is ok */
853 ARG_PTR_TO_SPIN_LOCK, /* pointer to bpf_spin_lock */
854 ARG_PTR_TO_SOCK_COMMON, /* pointer to sock_common */
855 ARG_PTR_TO_SOCKET, /* pointer to bpf_sock (fullsock) */
856 ARG_PTR_TO_BTF_ID, /* pointer to in-kernel struct */
857 ARG_PTR_TO_RINGBUF_MEM, /* pointer to dynamically reserved ringbuf memory */
858 ARG_CONST_ALLOC_SIZE_OR_ZERO, /* number of allocated bytes requested */
859 ARG_PTR_TO_BTF_ID_SOCK_COMMON, /* pointer to in-kernel sock_common or bpf-mirrored bpf_sock */
860 ARG_PTR_TO_PERCPU_BTF_ID, /* pointer to in-kernel percpu type */
861 ARG_PTR_TO_FUNC, /* pointer to a bpf program function */
862 ARG_PTR_TO_STACK, /* pointer to stack */
863 ARG_PTR_TO_CONST_STR, /* pointer to a null terminated read-only string */
864 ARG_PTR_TO_TIMER, /* pointer to bpf_timer */
865 ARG_KPTR_XCHG_DEST, /* pointer to destination that kptrs are bpf_kptr_xchg'd into */
866 ARG_PTR_TO_DYNPTR, /* pointer to bpf_dynptr. See bpf_type_flag for dynptr type */
867 __BPF_ARG_TYPE_MAX,
868
869 /* Extended arg_types. */
870 ARG_PTR_TO_MAP_VALUE_OR_NULL = PTR_MAYBE_NULL | ARG_PTR_TO_MAP_VALUE,
871 ARG_PTR_TO_MEM_OR_NULL = PTR_MAYBE_NULL | ARG_PTR_TO_MEM,
872 ARG_PTR_TO_CTX_OR_NULL = PTR_MAYBE_NULL | ARG_PTR_TO_CTX,
873 ARG_PTR_TO_SOCKET_OR_NULL = PTR_MAYBE_NULL | ARG_PTR_TO_SOCKET,
874 ARG_PTR_TO_STACK_OR_NULL = PTR_MAYBE_NULL | ARG_PTR_TO_STACK,
875 ARG_PTR_TO_BTF_ID_OR_NULL = PTR_MAYBE_NULL | ARG_PTR_TO_BTF_ID,
876 /* Pointer to memory does not need to be initialized, since helper function
877 * fills all bytes or clears them in error case.
878 */
879 ARG_PTR_TO_UNINIT_MEM = MEM_UNINIT | MEM_WRITE | ARG_PTR_TO_MEM,
880 /* Pointer to valid memory of size known at compile time. */
881 ARG_PTR_TO_FIXED_SIZE_MEM = MEM_FIXED_SIZE | ARG_PTR_TO_MEM,
882
883 /* This must be the last entry. Its purpose is to ensure the enum is
884 * wide enough to hold the higher bits reserved for bpf_type_flag.
885 */
886 __BPF_ARG_TYPE_LIMIT = BPF_TYPE_LIMIT,
887};
888static_assert(__BPF_ARG_TYPE_MAX <= BPF_BASE_TYPE_LIMIT);
889
890/* type of values returned from helper functions */
891enum bpf_return_type {
892 RET_INTEGER, /* function returns integer */
893 RET_VOID, /* function doesn't return anything */
894 RET_PTR_TO_MAP_VALUE, /* returns a pointer to map elem value */
895 RET_PTR_TO_SOCKET, /* returns a pointer to a socket */
896 RET_PTR_TO_TCP_SOCK, /* returns a pointer to a tcp_sock */
897 RET_PTR_TO_SOCK_COMMON, /* returns a pointer to a sock_common */
898 RET_PTR_TO_MEM, /* returns a pointer to memory */
899 RET_PTR_TO_MEM_OR_BTF_ID, /* returns a pointer to a valid memory or a btf_id */
900 RET_PTR_TO_BTF_ID, /* returns a pointer to a btf_id */
901 __BPF_RET_TYPE_MAX,
902
903 /* Extended ret_types. */
904 RET_PTR_TO_MAP_VALUE_OR_NULL = PTR_MAYBE_NULL | RET_PTR_TO_MAP_VALUE,
905 RET_PTR_TO_SOCKET_OR_NULL = PTR_MAYBE_NULL | RET_PTR_TO_SOCKET,
906 RET_PTR_TO_TCP_SOCK_OR_NULL = PTR_MAYBE_NULL | RET_PTR_TO_TCP_SOCK,
907 RET_PTR_TO_SOCK_COMMON_OR_NULL = PTR_MAYBE_NULL | RET_PTR_TO_SOCK_COMMON,
908 RET_PTR_TO_RINGBUF_MEM_OR_NULL = PTR_MAYBE_NULL | MEM_RINGBUF | RET_PTR_TO_MEM,
909 RET_PTR_TO_DYNPTR_MEM_OR_NULL = PTR_MAYBE_NULL | RET_PTR_TO_MEM,
910 RET_PTR_TO_BTF_ID_OR_NULL = PTR_MAYBE_NULL | RET_PTR_TO_BTF_ID,
911 RET_PTR_TO_BTF_ID_TRUSTED = PTR_TRUSTED | RET_PTR_TO_BTF_ID,
912
913 /* This must be the last entry. Its purpose is to ensure the enum is
914 * wide enough to hold the higher bits reserved for bpf_type_flag.
915 */
916 __BPF_RET_TYPE_LIMIT = BPF_TYPE_LIMIT,
917};
918static_assert(__BPF_RET_TYPE_MAX <= BPF_BASE_TYPE_LIMIT);
919
920/* eBPF function prototype used by verifier to allow BPF_CALLs from eBPF programs
921 * to in-kernel helper functions and for adjusting imm32 field in BPF_CALL
922 * instructions after verifying
923 */
924struct bpf_func_proto {
925 u64 (*func)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
926 bool gpl_only;
927 bool pkt_access;
928 bool might_sleep;
929 /* set to true if helper follows contract for llvm
930 * attribute bpf_fastcall:
931 * - void functions do not scratch r0
932 * - functions taking N arguments scratch only registers r1-rN
933 */
934 bool allow_fastcall;
935 enum bpf_return_type ret_type;
936 union {
937 struct {
938 enum bpf_arg_type arg1_type;
939 enum bpf_arg_type arg2_type;
940 enum bpf_arg_type arg3_type;
941 enum bpf_arg_type arg4_type;
942 enum bpf_arg_type arg5_type;
943 };
944 enum bpf_arg_type arg_type[5];
945 };
946 union {
947 struct {
948 u32 *arg1_btf_id;
949 u32 *arg2_btf_id;
950 u32 *arg3_btf_id;
951 u32 *arg4_btf_id;
952 u32 *arg5_btf_id;
953 };
954 u32 *arg_btf_id[5];
955 struct {
956 size_t arg1_size;
957 size_t arg2_size;
958 size_t arg3_size;
959 size_t arg4_size;
960 size_t arg5_size;
961 };
962 size_t arg_size[5];
963 };
964 int *ret_btf_id; /* return value btf_id */
965 bool (*allowed)(const struct bpf_prog *prog);
966};
967
968/* bpf_context is intentionally undefined structure. Pointer to bpf_context is
969 * the first argument to eBPF programs.
970 * For socket filters: 'struct bpf_context *' == 'struct sk_buff *'
971 */
972struct bpf_context;
973
974enum bpf_access_type {
975 BPF_READ = 1,
976 BPF_WRITE = 2
977};
978
979/* types of values stored in eBPF registers */
980/* Pointer types represent:
981 * pointer
982 * pointer + imm
983 * pointer + (u16) var
984 * pointer + (u16) var + imm
985 * if (range > 0) then [ptr, ptr + range - off) is safe to access
986 * if (id > 0) means that some 'var' was added
987 * if (off > 0) means that 'imm' was added
988 */
989enum bpf_reg_type {
990 NOT_INIT = 0, /* nothing was written into register */
991 SCALAR_VALUE, /* reg doesn't contain a valid pointer */
992 PTR_TO_CTX, /* reg points to bpf_context */
993 CONST_PTR_TO_MAP, /* reg points to struct bpf_map */
994 PTR_TO_MAP_VALUE, /* reg points to map element value */
995 PTR_TO_MAP_KEY, /* reg points to a map element key */
996 PTR_TO_STACK, /* reg == frame_pointer + offset */
997 PTR_TO_PACKET_META, /* skb->data - meta_len */
998 PTR_TO_PACKET, /* reg points to skb->data */
999 PTR_TO_PACKET_END, /* skb->data + headlen */
1000 PTR_TO_FLOW_KEYS, /* reg points to bpf_flow_keys */
1001 PTR_TO_SOCKET, /* reg points to struct bpf_sock */
1002 PTR_TO_SOCK_COMMON, /* reg points to sock_common */
1003 PTR_TO_TCP_SOCK, /* reg points to struct tcp_sock */
1004 PTR_TO_TP_BUFFER, /* reg points to a writable raw tp's buffer */
1005 PTR_TO_XDP_SOCK, /* reg points to struct xdp_sock */
1006 /* PTR_TO_BTF_ID points to a kernel struct that does not need
1007 * to be null checked by the BPF program. This does not imply the
1008 * pointer is _not_ null and in practice this can easily be a null
1009 * pointer when reading pointer chains. The assumption is program
1010 * context will handle null pointer dereference typically via fault
1011 * handling. The verifier must keep this in mind and can make no
1012 * assumptions about null or non-null when doing branch analysis.
1013 * Further, when passed into helpers the helpers can not, without
1014 * additional context, assume the value is non-null.
1015 */
1016 PTR_TO_BTF_ID,
1017 PTR_TO_MEM, /* reg points to valid memory region */
1018 PTR_TO_ARENA,
1019 PTR_TO_BUF, /* reg points to a read/write buffer */
1020 PTR_TO_FUNC, /* reg points to a bpf program function */
1021 PTR_TO_INSN, /* reg points to a bpf program instruction */
1022 CONST_PTR_TO_DYNPTR, /* reg points to a const struct bpf_dynptr */
1023 __BPF_REG_TYPE_MAX,
1024
1025 /* Extended reg_types. */
1026 PTR_TO_MAP_VALUE_OR_NULL = PTR_MAYBE_NULL | PTR_TO_MAP_VALUE,
1027 PTR_TO_SOCKET_OR_NULL = PTR_MAYBE_NULL | PTR_TO_SOCKET,
1028 PTR_TO_SOCK_COMMON_OR_NULL = PTR_MAYBE_NULL | PTR_TO_SOCK_COMMON,
1029 PTR_TO_TCP_SOCK_OR_NULL = PTR_MAYBE_NULL | PTR_TO_TCP_SOCK,
1030 /* PTR_TO_BTF_ID_OR_NULL points to a kernel struct that has not
1031 * been checked for null. Used primarily to inform the verifier
1032 * an explicit null check is required for this struct.
1033 */
1034 PTR_TO_BTF_ID_OR_NULL = PTR_MAYBE_NULL | PTR_TO_BTF_ID,
1035
1036 /* This must be the last entry. Its purpose is to ensure the enum is
1037 * wide enough to hold the higher bits reserved for bpf_type_flag.
1038 */
1039 __BPF_REG_TYPE_LIMIT = BPF_TYPE_LIMIT,
1040};
1041static_assert(__BPF_REG_TYPE_MAX <= BPF_BASE_TYPE_LIMIT);
1042
1043/* The information passed from prog-specific *_is_valid_access
1044 * back to the verifier.
1045 */
1046struct bpf_insn_access_aux {
1047 enum bpf_reg_type reg_type;
1048 bool is_ldsx;
1049 union {
1050 int ctx_field_size;
1051 struct {
1052 struct btf *btf;
1053 u32 btf_id;
1054 u32 ref_obj_id;
1055 };
1056 };
1057 struct bpf_verifier_log *log; /* for verbose logs */
1058 bool is_retval; /* is accessing function return value ? */
1059};
1060
1061static inline void
1062bpf_ctx_record_field_size(struct bpf_insn_access_aux *aux, u32 size)
1063{
1064 aux->ctx_field_size = size;
1065}
1066
1067static bool bpf_is_ldimm64(const struct bpf_insn *insn)
1068{
1069 return insn->code == (BPF_LD | BPF_IMM | BPF_DW);
1070}
1071
1072static inline bool bpf_pseudo_func(const struct bpf_insn *insn)
1073{
1074 return bpf_is_ldimm64(insn) && insn->src_reg == BPF_PSEUDO_FUNC;
1075}
1076
1077/* Given a BPF_ATOMIC instruction @atomic_insn, return true if it is an
1078 * atomic load or store, and false if it is a read-modify-write instruction.
1079 */
1080static inline bool
1081bpf_atomic_is_load_store(const struct bpf_insn *atomic_insn)
1082{
1083 switch (atomic_insn->imm) {
1084 case BPF_LOAD_ACQ:
1085 case BPF_STORE_REL:
1086 return true;
1087 default:
1088 return false;
1089 }
1090}
1091
1092struct bpf_prog_ops {
1093 int (*test_run)(struct bpf_prog *prog, const union bpf_attr *kattr,
1094 union bpf_attr __user *uattr);
1095};
1096
1097struct bpf_reg_state;
1098struct bpf_verifier_ops {
1099 /* return eBPF function prototype for verification */
1100 const struct bpf_func_proto *
1101 (*get_func_proto)(enum bpf_func_id func_id,
1102 const struct bpf_prog *prog);
1103
1104 /* return true if 'size' wide access at offset 'off' within bpf_context
1105 * with 'type' (read or write) is allowed
1106 */
1107 bool (*is_valid_access)(int off, int size, enum bpf_access_type type,
1108 const struct bpf_prog *prog,
1109 struct bpf_insn_access_aux *info);
1110 int (*gen_prologue)(struct bpf_insn *insn, bool direct_write,
1111 const struct bpf_prog *prog);
1112 int (*gen_epilogue)(struct bpf_insn *insn, const struct bpf_prog *prog,
1113 s16 ctx_stack_off);
1114 int (*gen_ld_abs)(const struct bpf_insn *orig,
1115 struct bpf_insn *insn_buf);
1116 u32 (*convert_ctx_access)(enum bpf_access_type type,
1117 const struct bpf_insn *src,
1118 struct bpf_insn *dst,
1119 struct bpf_prog *prog, u32 *target_size);
1120 int (*btf_struct_access)(struct bpf_verifier_log *log,
1121 const struct bpf_reg_state *reg,
1122 int off, int size);
1123};
1124
1125struct bpf_prog_offload_ops {
1126 /* verifier basic callbacks */
1127 int (*insn_hook)(struct bpf_verifier_env *env,
1128 int insn_idx, int prev_insn_idx);
1129 int (*finalize)(struct bpf_verifier_env *env);
1130 /* verifier optimization callbacks (called after .finalize) */
1131 int (*replace_insn)(struct bpf_verifier_env *env, u32 off,
1132 struct bpf_insn *insn);
1133 int (*remove_insns)(struct bpf_verifier_env *env, u32 off, u32 cnt);
1134 /* program management callbacks */
1135 int (*prepare)(struct bpf_prog *prog);
1136 int (*translate)(struct bpf_prog *prog);
1137 void (*destroy)(struct bpf_prog *prog);
1138};
1139
1140struct bpf_prog_offload {
1141 struct bpf_prog *prog;
1142 struct net_device *netdev;
1143 struct bpf_offload_dev *offdev;
1144 void *dev_priv;
1145 struct list_head offloads;
1146 bool dev_state;
1147 bool opt_failed;
1148 void *jited_image;
1149 u32 jited_len;
1150};
1151
1152/* The longest tracepoint has 12 args.
1153 * See include/trace/bpf_probe.h
1154 */
1155#define MAX_BPF_FUNC_ARGS 12
1156
1157/* The maximum number of arguments passed through registers
1158 * a single function may have.
1159 */
1160#define MAX_BPF_FUNC_REG_ARGS 5
1161
1162/* The argument is a structure or a union. */
1163#define BTF_FMODEL_STRUCT_ARG BIT(0)
1164
1165/* The argument is signed. */
1166#define BTF_FMODEL_SIGNED_ARG BIT(1)
1167
1168struct btf_func_model {
1169 u8 ret_size;
1170 u8 ret_flags;
1171 u8 nr_args;
1172 u8 arg_size[MAX_BPF_FUNC_ARGS];
1173 u8 arg_flags[MAX_BPF_FUNC_ARGS];
1174};
1175
1176/* Restore arguments before returning from trampoline to let original function
1177 * continue executing. This flag is used for fentry progs when there are no
1178 * fexit progs.
1179 */
1180#define BPF_TRAMP_F_RESTORE_REGS BIT(0)
1181/* Call original function after fentry progs, but before fexit progs.
1182 * Makes sense for fentry/fexit, normal calls and indirect calls.
1183 */
1184#define BPF_TRAMP_F_CALL_ORIG BIT(1)
1185/* Skip current frame and return to parent. Makes sense for fentry/fexit
1186 * programs only. Should not be used with normal calls and indirect calls.
1187 */
1188#define BPF_TRAMP_F_SKIP_FRAME BIT(2)
1189/* Store IP address of the caller on the trampoline stack,
1190 * so it's available for trampoline's programs.
1191 */
1192#define BPF_TRAMP_F_IP_ARG BIT(3)
1193/* Return the return value of fentry prog. Only used by bpf_struct_ops. */
1194#define BPF_TRAMP_F_RET_FENTRY_RET BIT(4)
1195
1196/* Get original function from stack instead of from provided direct address.
1197 * Makes sense for trampolines with fexit or fmod_ret programs.
1198 */
1199#define BPF_TRAMP_F_ORIG_STACK BIT(5)
1200
1201/* This trampoline is on a function with another ftrace_ops with IPMODIFY,
1202 * e.g., a live patch. This flag is set and cleared by ftrace call backs,
1203 */
1204#define BPF_TRAMP_F_SHARE_IPMODIFY BIT(6)
1205
1206/* Indicate that current trampoline is in a tail call context. Then, it has to
1207 * cache and restore tail_call_cnt to avoid infinite tail call loop.
1208 */
1209#define BPF_TRAMP_F_TAIL_CALL_CTX BIT(7)
1210
1211/*
1212 * Indicate the trampoline should be suitable to receive indirect calls;
1213 * without this indirectly calling the generated code can result in #UD/#CP,
1214 * depending on the CFI options.
1215 *
1216 * Used by bpf_struct_ops.
1217 *
1218 * Incompatible with FENTRY usage, overloads @func_addr argument.
1219 */
1220#define BPF_TRAMP_F_INDIRECT BIT(8)
1221
1222/* Each call __bpf_prog_enter + call bpf_func + call __bpf_prog_exit is ~50
1223 * bytes on x86.
1224 */
1225enum {
1226#if defined(__s390x__)
1227 BPF_MAX_TRAMP_LINKS = 27,
1228#else
1229 BPF_MAX_TRAMP_LINKS = 38,
1230#endif
1231};
1232
1233#define BPF_TRAMP_COOKIE_INDEX_SHIFT 8
1234#define BPF_TRAMP_IS_RETURN_SHIFT 63
1235
1236struct bpf_tramp_links {
1237 struct bpf_tramp_link *links[BPF_MAX_TRAMP_LINKS];
1238 int nr_links;
1239};
1240
1241struct bpf_tramp_run_ctx;
1242
1243/* Different use cases for BPF trampoline:
1244 * 1. replace nop at the function entry (kprobe equivalent)
1245 * flags = BPF_TRAMP_F_RESTORE_REGS
1246 * fentry = a set of programs to run before returning from trampoline
1247 *
1248 * 2. replace nop at the function entry (kprobe + kretprobe equivalent)
1249 * flags = BPF_TRAMP_F_CALL_ORIG | BPF_TRAMP_F_SKIP_FRAME
1250 * orig_call = fentry_ip + MCOUNT_INSN_SIZE
1251 * fentry = a set of program to run before calling original function
1252 * fexit = a set of program to run after original function
1253 *
1254 * 3. replace direct call instruction anywhere in the function body
1255 * or assign a function pointer for indirect call (like tcp_congestion_ops->cong_avoid)
1256 * With flags = 0
1257 * fentry = a set of programs to run before returning from trampoline
1258 * With flags = BPF_TRAMP_F_CALL_ORIG
1259 * orig_call = original callback addr or direct function addr
1260 * fentry = a set of program to run before calling original function
1261 * fexit = a set of program to run after original function
1262 */
1263struct bpf_tramp_image;
1264int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *image_end,
1265 const struct btf_func_model *m, u32 flags,
1266 struct bpf_tramp_links *tlinks,
1267 void *func_addr);
1268void *arch_alloc_bpf_trampoline(unsigned int size);
1269void arch_free_bpf_trampoline(void *image, unsigned int size);
1270int __must_check arch_protect_bpf_trampoline(void *image, unsigned int size);
1271int arch_bpf_trampoline_size(const struct btf_func_model *m, u32 flags,
1272 struct bpf_tramp_links *tlinks, void *func_addr);
1273
1274u64 notrace __bpf_prog_enter_sleepable_recur(struct bpf_prog *prog,
1275 struct bpf_tramp_run_ctx *run_ctx);
1276void notrace __bpf_prog_exit_sleepable_recur(struct bpf_prog *prog, u64 start,
1277 struct bpf_tramp_run_ctx *run_ctx);
1278void notrace __bpf_tramp_enter(struct bpf_tramp_image *tr);
1279void notrace __bpf_tramp_exit(struct bpf_tramp_image *tr);
1280typedef u64 (*bpf_trampoline_enter_t)(struct bpf_prog *prog,
1281 struct bpf_tramp_run_ctx *run_ctx);
1282typedef void (*bpf_trampoline_exit_t)(struct bpf_prog *prog, u64 start,
1283 struct bpf_tramp_run_ctx *run_ctx);
1284bpf_trampoline_enter_t bpf_trampoline_enter(const struct bpf_prog *prog);
1285bpf_trampoline_exit_t bpf_trampoline_exit(const struct bpf_prog *prog);
1286
1287#ifdef CONFIG_DYNAMIC_FTRACE_WITH_JMP
1288static inline bool bpf_trampoline_use_jmp(u64 flags)
1289{
1290 return flags & BPF_TRAMP_F_CALL_ORIG && !(flags & BPF_TRAMP_F_SKIP_FRAME);
1291}
1292#else
1293static inline bool bpf_trampoline_use_jmp(u64 flags)
1294{
1295 return false;
1296}
1297#endif
1298
1299struct bpf_ksym {
1300 unsigned long start;
1301 unsigned long end;
1302 char name[KSYM_NAME_LEN];
1303 struct list_head lnode;
1304 struct latch_tree_node tnode;
1305 bool prog;
1306 u32 fp_start;
1307 u32 fp_end;
1308};
1309
1310enum bpf_tramp_prog_type {
1311 BPF_TRAMP_FENTRY,
1312 BPF_TRAMP_FEXIT,
1313 BPF_TRAMP_MODIFY_RETURN,
1314 BPF_TRAMP_MAX,
1315 BPF_TRAMP_REPLACE, /* more than MAX */
1316 BPF_TRAMP_FSESSION,
1317};
1318
1319struct bpf_tramp_image {
1320 void *image;
1321 int size;
1322 struct bpf_ksym ksym;
1323 struct percpu_ref pcref;
1324 void *ip_after_call;
1325 void *ip_epilogue;
1326 union {
1327 struct rcu_head rcu;
1328 struct work_struct work;
1329 };
1330};
1331
1332struct bpf_trampoline {
1333 /* hlist for trampoline_key_table */
1334 struct hlist_node hlist_key;
1335 /* hlist for trampoline_ip_table */
1336 struct hlist_node hlist_ip;
1337 struct ftrace_ops *fops;
1338 /* serializes access to fields of this trampoline */
1339 struct mutex mutex;
1340 refcount_t refcnt;
1341 u32 flags;
1342 u64 key;
1343 unsigned long ip;
1344 struct {
1345 struct btf_func_model model;
1346 void *addr;
1347 bool ftrace_managed;
1348 } func;
1349 /* if !NULL this is BPF_PROG_TYPE_EXT program that extends another BPF
1350 * program by replacing one of its functions. func.addr is the address
1351 * of the function it replaced.
1352 */
1353 struct bpf_prog *extension_prog;
1354 /* list of BPF programs using this trampoline */
1355 struct hlist_head progs_hlist[BPF_TRAMP_MAX];
1356 /* Number of attached programs. A counter per kind. */
1357 int progs_cnt[BPF_TRAMP_MAX];
1358 /* Executable image of trampoline */
1359 struct bpf_tramp_image *cur_image;
1360};
1361
1362struct bpf_attach_target_info {
1363 struct btf_func_model fmodel;
1364 long tgt_addr;
1365 struct module *tgt_mod;
1366 const char *tgt_name;
1367 const struct btf_type *tgt_type;
1368};
1369
1370#define BPF_DISPATCHER_MAX 48 /* Fits in 2048B */
1371
1372struct bpf_dispatcher_prog {
1373 struct bpf_prog *prog;
1374 refcount_t users;
1375};
1376
1377struct bpf_dispatcher {
1378 /* dispatcher mutex */
1379 struct mutex mutex;
1380 void *func;
1381 struct bpf_dispatcher_prog progs[BPF_DISPATCHER_MAX];
1382 int num_progs;
1383 void *image;
1384 void *rw_image;
1385 u32 image_off;
1386 struct bpf_ksym ksym;
1387#ifdef CONFIG_HAVE_STATIC_CALL
1388 struct static_call_key *sc_key;
1389 void *sc_tramp;
1390#endif
1391};
1392
1393#ifndef __bpfcall
1394#define __bpfcall __nocfi
1395#endif
1396
1397static __always_inline __bpfcall unsigned int bpf_dispatcher_nop_func(
1398 const void *ctx,
1399 const struct bpf_insn *insnsi,
1400 bpf_func_t bpf_func)
1401{
1402 return bpf_func(ctx, insnsi);
1403}
1404
1405/* the implementation of the opaque uapi struct bpf_dynptr */
1406struct bpf_dynptr_kern {
1407 void *data;
1408 /* Size represents the number of usable bytes of dynptr data.
1409 * If for example the offset is at 4 for a local dynptr whose data is
1410 * of type u64, the number of usable bytes is 4.
1411 *
1412 * The upper 8 bits are reserved. It is as follows:
1413 * Bits 0 - 23 = size
1414 * Bits 24 - 30 = dynptr type
1415 * Bit 31 = whether dynptr is read-only
1416 */
1417 u32 size;
1418 u32 offset;
1419} __aligned(8);
1420
1421enum bpf_dynptr_type {
1422 BPF_DYNPTR_TYPE_INVALID,
1423 /* Points to memory that is local to the bpf program */
1424 BPF_DYNPTR_TYPE_LOCAL,
1425 /* Underlying data is a ringbuf record */
1426 BPF_DYNPTR_TYPE_RINGBUF,
1427 /* Underlying data is a sk_buff */
1428 BPF_DYNPTR_TYPE_SKB,
1429 /* Underlying data is a xdp_buff */
1430 BPF_DYNPTR_TYPE_XDP,
1431 /* Points to skb_metadata_end()-skb_metadata_len() */
1432 BPF_DYNPTR_TYPE_SKB_META,
1433 /* Underlying data is a file */
1434 BPF_DYNPTR_TYPE_FILE,
1435};
1436
1437int bpf_dynptr_check_size(u64 size);
1438u64 __bpf_dynptr_size(const struct bpf_dynptr_kern *ptr);
1439const void *__bpf_dynptr_data(const struct bpf_dynptr_kern *ptr, u64 len);
1440void *__bpf_dynptr_data_rw(const struct bpf_dynptr_kern *ptr, u64 len);
1441bool __bpf_dynptr_is_rdonly(const struct bpf_dynptr_kern *ptr);
1442int __bpf_dynptr_write(const struct bpf_dynptr_kern *dst, u64 offset,
1443 void *src, u64 len, u64 flags);
1444void *bpf_dynptr_slice_rdwr(const struct bpf_dynptr *p, u64 offset,
1445 void *buffer__nullable, u64 buffer__szk);
1446
1447static inline int bpf_dynptr_check_off_len(const struct bpf_dynptr_kern *ptr, u64 offset, u64 len)
1448{
1449 u64 size = __bpf_dynptr_size(ptr);
1450
1451 if (len > size || offset > size - len)
1452 return -E2BIG;
1453
1454 return 0;
1455}
1456
1457#ifdef CONFIG_BPF_JIT
1458int bpf_trampoline_link_prog(struct bpf_tramp_link *link,
1459 struct bpf_trampoline *tr,
1460 struct bpf_prog *tgt_prog);
1461int bpf_trampoline_unlink_prog(struct bpf_tramp_link *link,
1462 struct bpf_trampoline *tr,
1463 struct bpf_prog *tgt_prog);
1464struct bpf_trampoline *bpf_trampoline_get(u64 key,
1465 struct bpf_attach_target_info *tgt_info);
1466void bpf_trampoline_put(struct bpf_trampoline *tr);
1467int arch_prepare_bpf_dispatcher(void *image, void *buf, s64 *funcs, int num_funcs);
1468
1469/*
1470 * When the architecture supports STATIC_CALL replace the bpf_dispatcher_fn
1471 * indirection with a direct call to the bpf program. If the architecture does
1472 * not have STATIC_CALL, avoid a double-indirection.
1473 */
1474#ifdef CONFIG_HAVE_STATIC_CALL
1475
1476#define __BPF_DISPATCHER_SC_INIT(_name) \
1477 .sc_key = &STATIC_CALL_KEY(_name), \
1478 .sc_tramp = STATIC_CALL_TRAMP_ADDR(_name),
1479
1480#define __BPF_DISPATCHER_SC(name) \
1481 DEFINE_STATIC_CALL(bpf_dispatcher_##name##_call, bpf_dispatcher_nop_func)
1482
1483#define __BPF_DISPATCHER_CALL(name) \
1484 static_call(bpf_dispatcher_##name##_call)(ctx, insnsi, bpf_func)
1485
1486#define __BPF_DISPATCHER_UPDATE(_d, _new) \
1487 __static_call_update((_d)->sc_key, (_d)->sc_tramp, (_new))
1488
1489#else
1490#define __BPF_DISPATCHER_SC_INIT(name)
1491#define __BPF_DISPATCHER_SC(name)
1492#define __BPF_DISPATCHER_CALL(name) bpf_func(ctx, insnsi)
1493#define __BPF_DISPATCHER_UPDATE(_d, _new)
1494#endif
1495
1496#define BPF_DISPATCHER_INIT(_name) { \
1497 .mutex = __MUTEX_INITIALIZER(_name.mutex), \
1498 .func = &_name##_func, \
1499 .progs = {}, \
1500 .num_progs = 0, \
1501 .image = NULL, \
1502 .image_off = 0, \
1503 .ksym = { \
1504 .name = #_name, \
1505 .lnode = LIST_HEAD_INIT(_name.ksym.lnode), \
1506 }, \
1507 __BPF_DISPATCHER_SC_INIT(_name##_call) \
1508}
1509
1510#define DEFINE_BPF_DISPATCHER(name) \
1511 __BPF_DISPATCHER_SC(name); \
1512 noinline __bpfcall unsigned int bpf_dispatcher_##name##_func( \
1513 const void *ctx, \
1514 const struct bpf_insn *insnsi, \
1515 bpf_func_t bpf_func) \
1516 { \
1517 return __BPF_DISPATCHER_CALL(name); \
1518 } \
1519 EXPORT_SYMBOL(bpf_dispatcher_##name##_func); \
1520 struct bpf_dispatcher bpf_dispatcher_##name = \
1521 BPF_DISPATCHER_INIT(bpf_dispatcher_##name);
1522
1523#define DECLARE_BPF_DISPATCHER(name) \
1524 unsigned int bpf_dispatcher_##name##_func( \
1525 const void *ctx, \
1526 const struct bpf_insn *insnsi, \
1527 bpf_func_t bpf_func); \
1528 extern struct bpf_dispatcher bpf_dispatcher_##name;
1529
1530#define BPF_DISPATCHER_FUNC(name) bpf_dispatcher_##name##_func
1531#define BPF_DISPATCHER_PTR(name) (&bpf_dispatcher_##name)
1532void bpf_dispatcher_change_prog(struct bpf_dispatcher *d, struct bpf_prog *from,
1533 struct bpf_prog *to);
1534/* Called only from JIT-enabled code, so there's no need for stubs. */
1535void bpf_image_ksym_init(void *data, unsigned int size, struct bpf_ksym *ksym);
1536void bpf_image_ksym_add(struct bpf_ksym *ksym);
1537void bpf_image_ksym_del(struct bpf_ksym *ksym);
1538void bpf_ksym_add(struct bpf_ksym *ksym);
1539void bpf_ksym_del(struct bpf_ksym *ksym);
1540bool bpf_has_frame_pointer(unsigned long ip);
1541int bpf_jit_charge_modmem(u32 size);
1542void bpf_jit_uncharge_modmem(u32 size);
1543bool bpf_prog_has_trampoline(const struct bpf_prog *prog);
1544bool bpf_insn_is_indirect_target(const struct bpf_verifier_env *env, const struct bpf_prog *prog,
1545 int insn_idx);
1546#else
1547static inline int bpf_trampoline_link_prog(struct bpf_tramp_link *link,
1548 struct bpf_trampoline *tr,
1549 struct bpf_prog *tgt_prog)
1550{
1551 return -ENOTSUPP;
1552}
1553static inline int bpf_trampoline_unlink_prog(struct bpf_tramp_link *link,
1554 struct bpf_trampoline *tr,
1555 struct bpf_prog *tgt_prog)
1556{
1557 return -ENOTSUPP;
1558}
1559static inline struct bpf_trampoline *bpf_trampoline_get(u64 key,
1560 struct bpf_attach_target_info *tgt_info)
1561{
1562 return NULL;
1563}
1564static inline void bpf_trampoline_put(struct bpf_trampoline *tr) {}
1565#define DEFINE_BPF_DISPATCHER(name)
1566#define DECLARE_BPF_DISPATCHER(name)
1567#define BPF_DISPATCHER_FUNC(name) bpf_dispatcher_nop_func
1568#define BPF_DISPATCHER_PTR(name) NULL
1569static inline void bpf_dispatcher_change_prog(struct bpf_dispatcher *d,
1570 struct bpf_prog *from,
1571 struct bpf_prog *to) {}
1572static inline bool is_bpf_image_address(unsigned long address)
1573{
1574 return false;
1575}
1576static inline bool bpf_prog_has_trampoline(const struct bpf_prog *prog)
1577{
1578 return false;
1579}
1580#endif
1581
1582struct bpf_func_info_aux {
1583 u16 linkage;
1584 bool unreliable;
1585 bool called : 1;
1586 bool verified : 1;
1587};
1588
1589enum bpf_jit_poke_reason {
1590 BPF_POKE_REASON_TAIL_CALL,
1591};
1592
1593/* Descriptor of pokes pointing /into/ the JITed image. */
1594struct bpf_jit_poke_descriptor {
1595 void *tailcall_target;
1596 void *tailcall_bypass;
1597 void *bypass_addr;
1598 void *aux;
1599 union {
1600 struct {
1601 struct bpf_map *map;
1602 u32 key;
1603 } tail_call;
1604 };
1605 bool tailcall_target_stable;
1606 u8 adj_off;
1607 u16 reason;
1608 u32 insn_idx;
1609};
1610
1611/* reg_type info for ctx arguments */
1612struct bpf_ctx_arg_aux {
1613 u32 offset;
1614 enum bpf_reg_type reg_type;
1615 struct btf *btf;
1616 u32 btf_id;
1617 u32 ref_obj_id;
1618 bool refcounted;
1619};
1620
1621struct btf_mod_pair {
1622 struct btf *btf;
1623 struct module *module;
1624};
1625
1626struct bpf_kfunc_desc_tab;
1627
1628enum bpf_stream_id {
1629 BPF_STDOUT = 1,
1630 BPF_STDERR = 2,
1631};
1632
1633struct bpf_stream_elem {
1634 struct llist_node node;
1635 int total_len;
1636 int consumed_len;
1637 char str[];
1638};
1639
1640enum {
1641 /* 100k bytes */
1642 BPF_STREAM_MAX_CAPACITY = 100000ULL,
1643};
1644
1645struct bpf_stream {
1646 atomic_t capacity;
1647 struct llist_head log; /* list of in-flight stream elements in LIFO order */
1648
1649 struct mutex lock; /* lock protecting backlog_{head,tail} */
1650 struct llist_node *backlog_head; /* list of in-flight stream elements in FIFO order */
1651 struct llist_node *backlog_tail; /* tail of the list above */
1652};
1653
1654struct bpf_stream_stage {
1655 struct llist_head log;
1656 int len;
1657};
1658
1659struct bpf_prog_aux {
1660 atomic64_t refcnt;
1661 u32 used_map_cnt;
1662 u32 used_btf_cnt;
1663 u32 max_ctx_offset;
1664 u32 max_pkt_offset;
1665 u32 max_tp_access;
1666 u32 stack_depth;
1667 u32 id;
1668 u32 func_cnt; /* used by non-func prog as the number of func progs */
1669 u32 real_func_cnt; /* includes hidden progs, only used for JIT and freeing progs */
1670 u32 func_idx; /* 0 for non-func prog, the index in func array for func prog */
1671 u32 attach_btf_id; /* in-kernel BTF type id to attach to */
1672 u32 attach_st_ops_member_off;
1673 u32 ctx_arg_info_size;
1674 u32 max_rdonly_access;
1675 u32 max_rdwr_access;
1676 u32 subprog_start;
1677 struct btf *attach_btf;
1678 struct bpf_ctx_arg_aux *ctx_arg_info;
1679 void __percpu *priv_stack_ptr;
1680 struct mutex dst_mutex; /* protects dst_* pointers below, *after* prog becomes visible */
1681 struct bpf_prog *dst_prog;
1682 struct bpf_trampoline *dst_trampoline;
1683 enum bpf_prog_type saved_dst_prog_type;
1684 enum bpf_attach_type saved_dst_attach_type;
1685 bool verifier_zext; /* Zero extensions has been inserted by verifier. */
1686 bool dev_bound; /* Program is bound to the netdev. */
1687 bool offload_requested; /* Program is bound and offloaded to the netdev. */
1688 bool attach_btf_trace; /* true if attaching to BTF-enabled raw tp */
1689 bool attach_tracing_prog; /* true if tracing another tracing program */
1690 bool func_proto_unreliable;
1691 bool tail_call_reachable;
1692 bool xdp_has_frags;
1693 bool exception_cb;
1694 bool exception_boundary;
1695 bool is_extended; /* true if extended by freplace program */
1696 bool jits_use_priv_stack;
1697 bool priv_stack_requested;
1698 bool changes_pkt_data;
1699 bool might_sleep;
1700 bool kprobe_write_ctx;
1701 u64 prog_array_member_cnt; /* counts how many times as member of prog_array */
1702 struct mutex ext_mutex; /* mutex for is_extended and prog_array_member_cnt */
1703 struct bpf_arena *arena;
1704 void (*recursion_detected)(struct bpf_prog *prog); /* callback if recursion is detected */
1705 /* BTF_KIND_FUNC_PROTO for valid attach_btf_id */
1706 const struct btf_type *attach_func_proto;
1707 /* function name for valid attach_btf_id */
1708 const char *attach_func_name;
1709 struct bpf_prog **func;
1710 struct bpf_prog_aux *main_prog_aux;
1711 void *jit_data; /* JIT specific data. arch dependent */
1712 struct bpf_jit_poke_descriptor *poke_tab;
1713 struct bpf_kfunc_desc_tab *kfunc_tab;
1714 struct bpf_kfunc_btf_tab *kfunc_btf_tab;
1715 u32 size_poke_tab;
1716#ifdef CONFIG_FINEIBT
1717 struct bpf_ksym ksym_prefix;
1718#endif
1719 struct bpf_ksym ksym;
1720 const struct bpf_prog_ops *ops;
1721 const struct bpf_struct_ops *st_ops;
1722 struct bpf_map **used_maps;
1723 struct mutex used_maps_mutex; /* mutex for used_maps and used_map_cnt */
1724 struct btf_mod_pair *used_btfs;
1725 struct bpf_prog *prog;
1726 struct user_struct *user;
1727 u64 load_time; /* ns since boottime */
1728 u32 verified_insns;
1729 int cgroup_atype; /* enum cgroup_bpf_attach_type */
1730 struct bpf_map *cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE];
1731 char name[BPF_OBJ_NAME_LEN];
1732 u64 (*bpf_exception_cb)(u64 cookie, u64 sp, u64 bp, u64, u64);
1733#ifdef CONFIG_SECURITY
1734 void *security;
1735#endif
1736 struct bpf_token *token;
1737 struct bpf_prog_offload *offload;
1738 struct btf *btf;
1739 struct bpf_func_info *func_info;
1740 struct bpf_func_info_aux *func_info_aux;
1741 /* bpf_line_info loaded from userspace. linfo->insn_off
1742 * has the xlated insn offset.
1743 * Both the main and sub prog share the same linfo.
1744 * The subprog can access its first linfo by
1745 * using the linfo_idx.
1746 */
1747 struct bpf_line_info *linfo;
1748 /* jited_linfo is the jited addr of the linfo. It has a
1749 * one to one mapping to linfo:
1750 * jited_linfo[i] is the jited addr for the linfo[i]->insn_off.
1751 * Both the main and sub prog share the same jited_linfo.
1752 * The subprog can access its first jited_linfo by
1753 * using the linfo_idx.
1754 */
1755 void **jited_linfo;
1756 u32 func_info_cnt;
1757 u32 nr_linfo;
1758 /* subprog can use linfo_idx to access its first linfo and
1759 * jited_linfo.
1760 * main prog always has linfo_idx == 0
1761 */
1762 u32 linfo_idx;
1763 struct module *mod;
1764 u32 num_exentries;
1765 struct exception_table_entry *extable;
1766 union {
1767 struct work_struct work;
1768 struct rcu_head rcu;
1769 };
1770 struct bpf_stream stream[2];
1771 struct mutex st_ops_assoc_mutex;
1772 struct bpf_map __rcu *st_ops_assoc;
1773};
1774
1775#define BPF_NR_CONTEXTS 4 /* normal, softirq, hardirq, NMI */
1776
1777struct bpf_prog {
1778 u16 pages; /* Number of allocated pages */
1779 u16 jited:1, /* Is our filter JIT'ed? */
1780 jit_requested:1,/* archs need to JIT the prog */
1781 gpl_compatible:1, /* Is filter GPL compatible? */
1782 cb_access:1, /* Is control block accessed? */
1783 dst_needed:1, /* Do we need dst entry? */
1784 blinding_requested:1, /* needs constant blinding */
1785 blinded:1, /* Was blinded */
1786 is_func:1, /* program is a bpf function */
1787 kprobe_override:1, /* Do we override a kprobe? */
1788 has_callchain_buf:1, /* callchain buffer allocated? */
1789 enforce_expected_attach_type:1, /* Enforce expected_attach_type checking at attach time */
1790 call_get_stack:1, /* Do we call bpf_get_stack() or bpf_get_stackid() */
1791 call_get_func_ip:1, /* Do we call get_func_ip() */
1792 call_session_cookie:1, /* Do we call bpf_session_cookie() */
1793 tstamp_type_access:1, /* Accessed __sk_buff->tstamp_type */
1794 sleepable:1; /* BPF program is sleepable */
1795 enum bpf_prog_type type; /* Type of BPF program */
1796 enum bpf_attach_type expected_attach_type; /* For some prog types */
1797 u32 len; /* Number of filter blocks */
1798 u32 jited_len; /* Size of jited insns in bytes */
1799 union {
1800 u8 digest[SHA256_DIGEST_SIZE];
1801 u8 tag[BPF_TAG_SIZE];
1802 };
1803 struct bpf_prog_stats __percpu *stats;
1804 u8 __percpu *active; /* u8[BPF_NR_CONTEXTS] for recursion protection */
1805 unsigned int (*bpf_func)(const void *ctx,
1806 const struct bpf_insn *insn);
1807 struct bpf_prog_aux *aux; /* Auxiliary fields */
1808 struct sock_fprog_kern *orig_prog; /* Original BPF program */
1809 /* Instructions for interpreter */
1810 union {
1811 DECLARE_FLEX_ARRAY(struct sock_filter, insns);
1812 DECLARE_FLEX_ARRAY(struct bpf_insn, insnsi);
1813 };
1814};
1815
1816struct bpf_array_aux {
1817 /* Programs with direct jumps into programs part of this array. */
1818 struct list_head poke_progs;
1819 struct bpf_map *map;
1820 struct mutex poke_mutex;
1821 struct work_struct work;
1822};
1823
1824struct bpf_link {
1825 atomic64_t refcnt;
1826 u32 id;
1827 enum bpf_link_type type;
1828 const struct bpf_link_ops *ops;
1829 struct bpf_prog *prog;
1830
1831 u32 flags;
1832 enum bpf_attach_type attach_type;
1833
1834 /* rcu is used before freeing, work can be used to schedule that
1835 * RCU-based freeing before that, so they never overlap
1836 */
1837 union {
1838 struct rcu_head rcu;
1839 struct work_struct work;
1840 };
1841 /* whether BPF link itself has "sleepable" semantics, which can differ
1842 * from underlying BPF program having a "sleepable" semantics, as BPF
1843 * link's semantics is determined by target attach hook
1844 */
1845 bool sleepable;
1846};
1847
1848struct bpf_link_ops {
1849 void (*release)(struct bpf_link *link);
1850 /* deallocate link resources callback, called without RCU grace period
1851 * waiting
1852 */
1853 void (*dealloc)(struct bpf_link *link);
1854 /* deallocate link resources callback, called after RCU grace period;
1855 * if either the underlying BPF program is sleepable or BPF link's
1856 * target hook is sleepable, we'll go through tasks trace RCU GP and
1857 * then "classic" RCU GP; this need for chaining tasks trace and
1858 * classic RCU GPs is designated by setting bpf_link->sleepable flag
1859 *
1860 * For non-sleepable tracepoint links we go through SRCU gp instead,
1861 * since RCU is not used in that case. Sleepable tracepoints still
1862 * follow the scheme above.
1863 */
1864 void (*dealloc_deferred)(struct bpf_link *link);
1865 int (*detach)(struct bpf_link *link);
1866 int (*update_prog)(struct bpf_link *link, struct bpf_prog *new_prog,
1867 struct bpf_prog *old_prog);
1868 void (*show_fdinfo)(const struct bpf_link *link, struct seq_file *seq);
1869 int (*fill_link_info)(const struct bpf_link *link,
1870 struct bpf_link_info *info);
1871 int (*update_map)(struct bpf_link *link, struct bpf_map *new_map,
1872 struct bpf_map *old_map);
1873 __poll_t (*poll)(struct file *file, struct poll_table_struct *pts);
1874};
1875
1876struct bpf_tramp_link {
1877 struct bpf_link link;
1878 struct hlist_node tramp_hlist;
1879 u64 cookie;
1880};
1881
1882struct bpf_shim_tramp_link {
1883 struct bpf_tramp_link link;
1884 struct bpf_trampoline *trampoline;
1885};
1886
1887struct bpf_tracing_link {
1888 struct bpf_tramp_link link;
1889 struct bpf_trampoline *trampoline;
1890 struct bpf_prog *tgt_prog;
1891};
1892
1893struct bpf_fsession_link {
1894 struct bpf_tracing_link link;
1895 struct bpf_tramp_link fexit;
1896};
1897
1898struct bpf_raw_tp_link {
1899 struct bpf_link link;
1900 struct bpf_raw_event_map *btp;
1901 u64 cookie;
1902};
1903
1904struct bpf_link_primer {
1905 struct bpf_link *link;
1906 struct file *file;
1907 int fd;
1908 u32 id;
1909};
1910
1911struct bpf_mount_opts {
1912 kuid_t uid;
1913 kgid_t gid;
1914 umode_t mode;
1915
1916 /* BPF token-related delegation options */
1917 u64 delegate_cmds;
1918 u64 delegate_maps;
1919 u64 delegate_progs;
1920 u64 delegate_attachs;
1921};
1922
1923struct bpf_token {
1924 struct work_struct work;
1925 atomic64_t refcnt;
1926 struct user_namespace *userns;
1927 u64 allowed_cmds;
1928 u64 allowed_maps;
1929 u64 allowed_progs;
1930 u64 allowed_attachs;
1931#ifdef CONFIG_SECURITY
1932 void *security;
1933#endif
1934};
1935
1936struct bpf_struct_ops_value;
1937struct btf_member;
1938
1939#define BPF_STRUCT_OPS_MAX_NR_MEMBERS 64
1940/**
1941 * struct bpf_struct_ops - A structure of callbacks allowing a subsystem to
1942 * define a BPF_MAP_TYPE_STRUCT_OPS map type composed
1943 * of BPF_PROG_TYPE_STRUCT_OPS progs.
1944 * @verifier_ops: A structure of callbacks that are invoked by the verifier
1945 * when determining whether the struct_ops progs in the
1946 * struct_ops map are valid.
1947 * @init: A callback that is invoked a single time, and before any other
1948 * callback, to initialize the structure. A nonzero return value means
1949 * the subsystem could not be initialized.
1950 * @check_member: When defined, a callback invoked by the verifier to allow
1951 * the subsystem to determine if an entry in the struct_ops map
1952 * is valid. A nonzero return value means that the map is
1953 * invalid and should be rejected by the verifier.
1954 * @init_member: A callback that is invoked for each member of the struct_ops
1955 * map to allow the subsystem to initialize the member. A nonzero
1956 * value means the member could not be initialized. This callback
1957 * is exclusive with the @type, @type_id, @value_type, and
1958 * @value_id fields.
1959 * @reg: A callback that is invoked when the struct_ops map has been
1960 * initialized and is being attached to. Zero means the struct_ops map
1961 * has been successfully registered and is live. A nonzero return value
1962 * means the struct_ops map could not be registered.
1963 * @unreg: A callback that is invoked when the struct_ops map should be
1964 * unregistered.
1965 * @update: A callback that is invoked when the live struct_ops map is being
1966 * updated to contain new values. This callback is only invoked when
1967 * the struct_ops map is loaded with BPF_F_LINK. If not defined, the
1968 * it is assumed that the struct_ops map cannot be updated.
1969 * @validate: A callback that is invoked after all of the members have been
1970 * initialized. This callback should perform static checks on the
1971 * map, meaning that it should either fail or succeed
1972 * deterministically. A struct_ops map that has been validated may
1973 * not necessarily succeed in being registered if the call to @reg
1974 * fails. For example, a valid struct_ops map may be loaded, but
1975 * then fail to be registered due to there being another active
1976 * struct_ops map on the system in the subsystem already. For this
1977 * reason, if this callback is not defined, the check is skipped as
1978 * the struct_ops map will have final verification performed in
1979 * @reg.
1980 * @cfi_stubs: Pointer to a structure of stub functions for CFI. These stubs
1981 * provide the correct Control Flow Integrity hashes for the
1982 * trampolines generated by BPF struct_ops.
1983 * @owner: The module that owns this struct_ops. Used for module reference
1984 * counting to ensure the module providing the struct_ops cannot be
1985 * unloaded while in use.
1986 * @name: The name of the struct bpf_struct_ops object.
1987 * @func_models: Func models
1988 */
1989struct bpf_struct_ops {
1990 const struct bpf_verifier_ops *verifier_ops;
1991 int (*init)(struct btf *btf);
1992 int (*check_member)(const struct btf_type *t,
1993 const struct btf_member *member,
1994 const struct bpf_prog *prog);
1995 int (*init_member)(const struct btf_type *t,
1996 const struct btf_member *member,
1997 void *kdata, const void *udata);
1998 int (*reg)(void *kdata, struct bpf_link *link);
1999 void (*unreg)(void *kdata, struct bpf_link *link);
2000 int (*update)(void *kdata, void *old_kdata, struct bpf_link *link);
2001 int (*validate)(void *kdata);
2002 void *cfi_stubs;
2003 struct module *owner;
2004 const char *name;
2005 struct btf_func_model func_models[BPF_STRUCT_OPS_MAX_NR_MEMBERS];
2006};
2007
2008/* Every member of a struct_ops type has an instance even a member is not
2009 * an operator (function pointer). The "info" field will be assigned to
2010 * prog->aux->ctx_arg_info of BPF struct_ops programs to provide the
2011 * argument information required by the verifier to verify the program.
2012 *
2013 * btf_ctx_access() will lookup prog->aux->ctx_arg_info to find the
2014 * corresponding entry for an given argument.
2015 */
2016struct bpf_struct_ops_arg_info {
2017 struct bpf_ctx_arg_aux *info;
2018 u32 cnt;
2019};
2020
2021struct bpf_struct_ops_desc {
2022 struct bpf_struct_ops *st_ops;
2023
2024 const struct btf_type *type;
2025 const struct btf_type *value_type;
2026 u32 type_id;
2027 u32 value_id;
2028
2029 /* Collection of argument information for each member */
2030 struct bpf_struct_ops_arg_info *arg_info;
2031};
2032
2033enum bpf_struct_ops_state {
2034 BPF_STRUCT_OPS_STATE_INIT,
2035 BPF_STRUCT_OPS_STATE_INUSE,
2036 BPF_STRUCT_OPS_STATE_TOBEFREE,
2037 BPF_STRUCT_OPS_STATE_READY,
2038};
2039
2040struct bpf_struct_ops_common_value {
2041 refcount_t refcnt;
2042 enum bpf_struct_ops_state state;
2043};
2044
2045static inline bool bpf_prog_get_recursion_context(struct bpf_prog *prog)
2046{
2047#ifdef CONFIG_ARM64
2048 u8 rctx = interrupt_context_level();
2049 u8 *active = this_cpu_ptr(prog->active);
2050 u32 val;
2051
2052 preempt_disable();
2053 active[rctx]++;
2054 val = le32_to_cpu(*(__le32 *)active);
2055 preempt_enable();
2056 if (val != BIT(rctx * 8))
2057 return false;
2058
2059 return true;
2060#else
2061 return this_cpu_inc_return(*(int __percpu *)(prog->active)) == 1;
2062#endif
2063}
2064
2065static inline void bpf_prog_put_recursion_context(struct bpf_prog *prog)
2066{
2067#ifdef CONFIG_ARM64
2068 u8 rctx = interrupt_context_level();
2069 u8 *active = this_cpu_ptr(prog->active);
2070
2071 preempt_disable();
2072 active[rctx]--;
2073 preempt_enable();
2074#else
2075 this_cpu_dec(*(int __percpu *)(prog->active));
2076#endif
2077}
2078
2079#if defined(CONFIG_BPF_JIT) && defined(CONFIG_BPF_SYSCALL)
2080/* This macro helps developer to register a struct_ops type and generate
2081 * type information correctly. Developers should use this macro to register
2082 * a struct_ops type instead of calling __register_bpf_struct_ops() directly.
2083 */
2084#define register_bpf_struct_ops(st_ops, type) \
2085 ({ \
2086 struct bpf_struct_ops_##type { \
2087 struct bpf_struct_ops_common_value common; \
2088 struct type data ____cacheline_aligned_in_smp; \
2089 }; \
2090 BTF_TYPE_EMIT(struct bpf_struct_ops_##type); \
2091 __register_bpf_struct_ops(st_ops); \
2092 })
2093#define BPF_MODULE_OWNER ((void *)((0xeB9FUL << 2) + POISON_POINTER_DELTA))
2094bool bpf_struct_ops_get(const void *kdata);
2095void bpf_struct_ops_put(const void *kdata);
2096int bpf_struct_ops_supported(const struct bpf_struct_ops *st_ops, u32 moff);
2097int bpf_struct_ops_map_sys_lookup_elem(struct bpf_map *map, void *key,
2098 void *value);
2099int bpf_struct_ops_prepare_trampoline(struct bpf_tramp_links *tlinks,
2100 struct bpf_tramp_link *link,
2101 const struct btf_func_model *model,
2102 void *stub_func,
2103 void **image, u32 *image_off,
2104 bool allow_alloc);
2105void bpf_struct_ops_image_free(void *image);
2106static inline bool bpf_try_module_get(const void *data, struct module *owner)
2107{
2108 if (owner == BPF_MODULE_OWNER)
2109 return bpf_struct_ops_get(data);
2110 else
2111 return try_module_get(owner);
2112}
2113static inline void bpf_module_put(const void *data, struct module *owner)
2114{
2115 if (owner == BPF_MODULE_OWNER)
2116 bpf_struct_ops_put(data);
2117 else
2118 module_put(owner);
2119}
2120int bpf_struct_ops_link_create(union bpf_attr *attr);
2121int bpf_prog_assoc_struct_ops(struct bpf_prog *prog, struct bpf_map *map);
2122void bpf_prog_disassoc_struct_ops(struct bpf_prog *prog);
2123void *bpf_prog_get_assoc_struct_ops(const struct bpf_prog_aux *aux);
2124u32 bpf_struct_ops_id(const void *kdata);
2125
2126#ifdef CONFIG_NET
2127/* Define it here to avoid the use of forward declaration */
2128struct bpf_dummy_ops_state {
2129 int val;
2130};
2131
2132struct bpf_dummy_ops {
2133 int (*test_1)(struct bpf_dummy_ops_state *cb);
2134 int (*test_2)(struct bpf_dummy_ops_state *cb, int a1, unsigned short a2,
2135 char a3, unsigned long a4);
2136 int (*test_sleepable)(struct bpf_dummy_ops_state *cb);
2137};
2138
2139int bpf_struct_ops_test_run(struct bpf_prog *prog, const union bpf_attr *kattr,
2140 union bpf_attr __user *uattr);
2141#endif
2142int bpf_struct_ops_desc_init(struct bpf_struct_ops_desc *st_ops_desc,
2143 struct btf *btf,
2144 struct bpf_verifier_log *log);
2145void bpf_map_struct_ops_info_fill(struct bpf_map_info *info, struct bpf_map *map);
2146void bpf_struct_ops_desc_release(struct bpf_struct_ops_desc *st_ops_desc);
2147#else
2148#define register_bpf_struct_ops(st_ops, type) ({ (void *)(st_ops); 0; })
2149static inline bool bpf_try_module_get(const void *data, struct module *owner)
2150{
2151 return try_module_get(owner);
2152}
2153static inline void bpf_module_put(const void *data, struct module *owner)
2154{
2155 module_put(owner);
2156}
2157static inline int bpf_struct_ops_supported(const struct bpf_struct_ops *st_ops, u32 moff)
2158{
2159 return -ENOTSUPP;
2160}
2161static inline int bpf_struct_ops_map_sys_lookup_elem(struct bpf_map *map,
2162 void *key,
2163 void *value)
2164{
2165 return -EINVAL;
2166}
2167static inline int bpf_struct_ops_link_create(union bpf_attr *attr)
2168{
2169 return -EOPNOTSUPP;
2170}
2171static inline int bpf_prog_assoc_struct_ops(struct bpf_prog *prog, struct bpf_map *map)
2172{
2173 return -EOPNOTSUPP;
2174}
2175static inline void bpf_prog_disassoc_struct_ops(struct bpf_prog *prog)
2176{
2177}
2178static inline void *bpf_prog_get_assoc_struct_ops(const struct bpf_prog_aux *aux)
2179{
2180 return NULL;
2181}
2182static inline void bpf_map_struct_ops_info_fill(struct bpf_map_info *info, struct bpf_map *map)
2183{
2184}
2185
2186static inline void bpf_struct_ops_desc_release(struct bpf_struct_ops_desc *st_ops_desc)
2187{
2188}
2189
2190#endif
2191
2192static inline int bpf_fsession_cnt(struct bpf_tramp_links *links)
2193{
2194 struct bpf_tramp_links fentries = links[BPF_TRAMP_FENTRY];
2195 int cnt = 0;
2196
2197 for (int i = 0; i < links[BPF_TRAMP_FENTRY].nr_links; i++) {
2198 if (fentries.links[i]->link.prog->expected_attach_type == BPF_TRACE_FSESSION)
2199 cnt++;
2200 }
2201
2202 return cnt;
2203}
2204
2205static inline bool bpf_prog_calls_session_cookie(struct bpf_tramp_link *link)
2206{
2207 return link->link.prog->call_session_cookie;
2208}
2209
2210static inline int bpf_fsession_cookie_cnt(struct bpf_tramp_links *links)
2211{
2212 struct bpf_tramp_links fentries = links[BPF_TRAMP_FENTRY];
2213 int cnt = 0;
2214
2215 for (int i = 0; i < links[BPF_TRAMP_FENTRY].nr_links; i++) {
2216 if (bpf_prog_calls_session_cookie(fentries.links[i]))
2217 cnt++;
2218 }
2219
2220 return cnt;
2221}
2222
2223int bpf_prog_ctx_arg_info_init(struct bpf_prog *prog,
2224 const struct bpf_ctx_arg_aux *info, u32 cnt);
2225
2226#if defined(CONFIG_CGROUP_BPF) && defined(CONFIG_BPF_LSM)
2227int bpf_trampoline_link_cgroup_shim(struct bpf_prog *prog,
2228 int cgroup_atype,
2229 enum bpf_attach_type attach_type);
2230void bpf_trampoline_unlink_cgroup_shim(struct bpf_prog *prog);
2231#else
2232static inline int bpf_trampoline_link_cgroup_shim(struct bpf_prog *prog,
2233 int cgroup_atype,
2234 enum bpf_attach_type attach_type)
2235{
2236 return -EOPNOTSUPP;
2237}
2238static inline void bpf_trampoline_unlink_cgroup_shim(struct bpf_prog *prog)
2239{
2240}
2241#endif
2242
2243struct bpf_array {
2244 struct bpf_map map;
2245 u32 elem_size;
2246 u32 index_mask;
2247 struct bpf_array_aux *aux;
2248 union {
2249 DECLARE_FLEX_ARRAY(char, value) __aligned(8);
2250 DECLARE_FLEX_ARRAY(void *, ptrs) __aligned(8);
2251 DECLARE_FLEX_ARRAY(void __percpu *, pptrs) __aligned(8);
2252 };
2253};
2254
2255/*
2256 * The bpf_array_get_next_key() function may be used for all array-like
2257 * maps, i.e., maps with u32 keys with range [0 ,..., max_entries)
2258 */
2259int bpf_array_get_next_key(struct bpf_map *map, void *key, void *next_key);
2260
2261#define BPF_COMPLEXITY_LIMIT_INSNS 1000000 /* yes. 1M insns */
2262#define MAX_TAIL_CALL_CNT 33
2263
2264/* Maximum number of loops for bpf_loop and bpf_iter_num.
2265 * It's enum to expose it (and thus make it discoverable) through BTF.
2266 */
2267enum {
2268 BPF_MAX_LOOPS = 8 * 1024 * 1024,
2269 BPF_MAX_TIMED_LOOPS = 0xffff,
2270};
2271
2272#define BPF_F_ACCESS_MASK (BPF_F_RDONLY | \
2273 BPF_F_RDONLY_PROG | \
2274 BPF_F_WRONLY | \
2275 BPF_F_WRONLY_PROG)
2276
2277#define BPF_MAP_CAN_READ BIT(0)
2278#define BPF_MAP_CAN_WRITE BIT(1)
2279
2280/* Maximum number of user-producer ring buffer samples that can be drained in
2281 * a call to bpf_user_ringbuf_drain().
2282 */
2283#define BPF_MAX_USER_RINGBUF_SAMPLES (128 * 1024)
2284
2285static inline u32 bpf_map_flags_to_cap(struct bpf_map *map)
2286{
2287 u32 access_flags = map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG);
2288
2289 /* Combination of BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG is
2290 * not possible.
2291 */
2292 if (access_flags & BPF_F_RDONLY_PROG)
2293 return BPF_MAP_CAN_READ;
2294 else if (access_flags & BPF_F_WRONLY_PROG)
2295 return BPF_MAP_CAN_WRITE;
2296 else
2297 return BPF_MAP_CAN_READ | BPF_MAP_CAN_WRITE;
2298}
2299
2300static inline bool bpf_map_flags_access_ok(u32 access_flags)
2301{
2302 return (access_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) !=
2303 (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG);
2304}
2305
2306static inline struct bpf_map_owner *bpf_map_owner_alloc(struct bpf_map *map)
2307{
2308 return kzalloc_obj(*map->owner, GFP_ATOMIC);
2309}
2310
2311static inline void bpf_map_owner_free(struct bpf_map *map)
2312{
2313 kfree(map->owner);
2314}
2315
2316struct bpf_event_entry {
2317 struct perf_event *event;
2318 struct file *perf_file;
2319 struct file *map_file;
2320 struct rcu_head rcu;
2321};
2322
2323static inline bool map_type_contains_progs(struct bpf_map *map)
2324{
2325 return map->map_type == BPF_MAP_TYPE_PROG_ARRAY ||
2326 map->map_type == BPF_MAP_TYPE_DEVMAP ||
2327 map->map_type == BPF_MAP_TYPE_CPUMAP;
2328}
2329
2330bool bpf_prog_map_compatible(struct bpf_map *map, const struct bpf_prog *fp);
2331int bpf_prog_calc_tag(struct bpf_prog *fp);
2332
2333const struct bpf_func_proto *bpf_get_trace_printk_proto(void);
2334const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void);
2335
2336const struct bpf_func_proto *bpf_get_perf_event_read_value_proto(void);
2337
2338typedef unsigned long (*bpf_ctx_copy_t)(void *dst, const void *src,
2339 unsigned long off, unsigned long len);
2340typedef u32 (*bpf_convert_ctx_access_t)(enum bpf_access_type type,
2341 const struct bpf_insn *src,
2342 struct bpf_insn *dst,
2343 struct bpf_prog *prog,
2344 u32 *target_size);
2345
2346u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
2347 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy);
2348
2349/* an array of programs to be executed under rcu_lock.
2350 *
2351 * Typical usage:
2352 * ret = bpf_prog_run_array(rcu_dereference(&bpf_prog_array), ctx, bpf_prog_run);
2353 *
2354 * the structure returned by bpf_prog_array_alloc() should be populated
2355 * with program pointers and the last pointer must be NULL.
2356 * The user has to keep refcnt on the program and make sure the program
2357 * is removed from the array before bpf_prog_put().
2358 * The 'struct bpf_prog_array *' should only be replaced with xchg()
2359 * since other cpus are walking the array of pointers in parallel.
2360 */
2361struct bpf_prog_array_item {
2362 struct bpf_prog *prog;
2363 union {
2364 struct bpf_cgroup_storage *cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE];
2365 u64 bpf_cookie;
2366 };
2367};
2368
2369struct bpf_prog_array {
2370 struct rcu_head rcu;
2371 struct bpf_prog_array_item items[];
2372};
2373
2374/* to avoid allocating empty bpf_prog_array for cgroups that
2375 * don't have bpf program attached use one global 'bpf_empty_prog_array'
2376 * It will not be modified the caller of bpf_prog_array_alloc()
2377 * (since caller requested prog_cnt == 0)
2378 * that pointer should be 'freed' by bpf_prog_array_free()
2379 */
2380extern struct bpf_prog_array bpf_empty_prog_array;
2381
2382struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags);
2383void bpf_prog_array_free(struct bpf_prog_array *progs);
2384/* Use when traversal over the bpf_prog_array uses tasks_trace rcu */
2385void bpf_prog_array_free_sleepable(struct bpf_prog_array *progs);
2386int bpf_prog_array_length(struct bpf_prog_array *progs);
2387bool bpf_prog_array_is_empty(struct bpf_prog_array *array);
2388int bpf_prog_array_copy_to_user(struct bpf_prog_array *progs,
2389 __u32 __user *prog_ids, u32 cnt);
2390
2391void bpf_prog_array_delete_safe(struct bpf_prog_array *progs,
2392 struct bpf_prog *old_prog);
2393int bpf_prog_array_delete_safe_at(struct bpf_prog_array *array, int index);
2394int bpf_prog_array_update_at(struct bpf_prog_array *array, int index,
2395 struct bpf_prog *prog);
2396int bpf_prog_array_copy_info(struct bpf_prog_array *array,
2397 u32 *prog_ids, u32 request_cnt,
2398 u32 *prog_cnt);
2399int bpf_prog_array_copy(struct bpf_prog_array *old_array,
2400 struct bpf_prog *exclude_prog,
2401 struct bpf_prog *include_prog,
2402 u64 bpf_cookie,
2403 struct bpf_prog_array **new_array);
2404
2405struct bpf_run_ctx {};
2406
2407struct bpf_cg_run_ctx {
2408 struct bpf_run_ctx run_ctx;
2409 const struct bpf_prog_array_item *prog_item;
2410 int retval;
2411};
2412
2413struct bpf_trace_run_ctx {
2414 struct bpf_run_ctx run_ctx;
2415 u64 bpf_cookie;
2416 bool is_uprobe;
2417};
2418
2419struct bpf_tramp_run_ctx {
2420 struct bpf_run_ctx run_ctx;
2421 u64 bpf_cookie;
2422 struct bpf_run_ctx *saved_run_ctx;
2423};
2424
2425static inline struct bpf_run_ctx *bpf_set_run_ctx(struct bpf_run_ctx *new_ctx)
2426{
2427 struct bpf_run_ctx *old_ctx = NULL;
2428
2429#ifdef CONFIG_BPF_SYSCALL
2430 old_ctx = current->bpf_ctx;
2431 current->bpf_ctx = new_ctx;
2432#endif
2433 return old_ctx;
2434}
2435
2436static inline void bpf_reset_run_ctx(struct bpf_run_ctx *old_ctx)
2437{
2438#ifdef CONFIG_BPF_SYSCALL
2439 current->bpf_ctx = old_ctx;
2440#endif
2441}
2442
2443/* BPF program asks to bypass CAP_NET_BIND_SERVICE in bind. */
2444#define BPF_RET_BIND_NO_CAP_NET_BIND_SERVICE (1 << 0)
2445/* BPF program asks to set CN on the packet. */
2446#define BPF_RET_SET_CN (1 << 0)
2447
2448typedef u32 (*bpf_prog_run_fn)(const struct bpf_prog *prog, const void *ctx);
2449
2450static __always_inline u32
2451bpf_prog_run_array(const struct bpf_prog_array *array,
2452 const void *ctx, bpf_prog_run_fn run_prog)
2453{
2454 const struct bpf_prog_array_item *item;
2455 const struct bpf_prog *prog;
2456 struct bpf_run_ctx *old_run_ctx;
2457 struct bpf_trace_run_ctx run_ctx;
2458 u32 ret = 1;
2459
2460 RCU_LOCKDEP_WARN(!rcu_read_lock_held(), "no rcu lock held");
2461
2462 if (unlikely(!array))
2463 return ret;
2464
2465 run_ctx.is_uprobe = false;
2466
2467 migrate_disable();
2468 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
2469 item = &array->items[0];
2470 while ((prog = READ_ONCE(item->prog))) {
2471 run_ctx.bpf_cookie = item->bpf_cookie;
2472 ret &= run_prog(prog, ctx);
2473 item++;
2474 }
2475 bpf_reset_run_ctx(old_run_ctx);
2476 migrate_enable();
2477 return ret;
2478}
2479
2480/* Notes on RCU design for bpf_prog_arrays containing sleepable programs:
2481 *
2482 * We use the tasks_trace rcu flavor read section to protect the bpf_prog_array
2483 * overall. As a result, we must use the bpf_prog_array_free_sleepable
2484 * in order to use the tasks_trace rcu grace period.
2485 *
2486 * When a non-sleepable program is inside the array, we take the rcu read
2487 * section and disable preemption for that program alone, so it can access
2488 * rcu-protected dynamically sized maps.
2489 */
2490static __always_inline u32
2491bpf_prog_run_array_uprobe(const struct bpf_prog_array *array,
2492 const void *ctx, bpf_prog_run_fn run_prog)
2493{
2494 const struct bpf_prog_array_item *item;
2495 const struct bpf_prog *prog;
2496 struct bpf_run_ctx *old_run_ctx;
2497 struct bpf_trace_run_ctx run_ctx;
2498 u32 ret = 1;
2499
2500 might_fault();
2501 RCU_LOCKDEP_WARN(!rcu_read_lock_trace_held(), "no rcu lock held");
2502
2503 if (unlikely(!array))
2504 return ret;
2505
2506 migrate_disable();
2507
2508 run_ctx.is_uprobe = true;
2509
2510 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
2511 item = &array->items[0];
2512 while ((prog = READ_ONCE(item->prog))) {
2513 if (!prog->sleepable)
2514 rcu_read_lock();
2515
2516 run_ctx.bpf_cookie = item->bpf_cookie;
2517 ret &= run_prog(prog, ctx);
2518 item++;
2519
2520 if (!prog->sleepable)
2521 rcu_read_unlock();
2522 }
2523 bpf_reset_run_ctx(old_run_ctx);
2524 migrate_enable();
2525 return ret;
2526}
2527
2528bool bpf_jit_bypass_spec_v1(void);
2529bool bpf_jit_bypass_spec_v4(void);
2530
2531#define bpf_rcu_lock_held() \
2532 (rcu_read_lock_held() || rcu_read_lock_trace_held() || rcu_read_lock_bh_held())
2533
2534#ifdef CONFIG_BPF_SYSCALL
2535DECLARE_PER_CPU(int, bpf_prog_active);
2536extern struct mutex bpf_stats_enabled_mutex;
2537
2538/*
2539 * Block execution of BPF programs attached to instrumentation (perf,
2540 * kprobes, tracepoints) to prevent deadlocks on map operations as any of
2541 * these events can happen inside a region which holds a map bucket lock
2542 * and can deadlock on it.
2543 */
2544static inline void bpf_disable_instrumentation(void)
2545{
2546 migrate_disable();
2547 this_cpu_inc(bpf_prog_active);
2548}
2549
2550static inline void bpf_enable_instrumentation(void)
2551{
2552 this_cpu_dec(bpf_prog_active);
2553 migrate_enable();
2554}
2555
2556extern const struct super_operations bpf_super_ops;
2557extern const struct file_operations bpf_map_fops;
2558extern const struct file_operations bpf_prog_fops;
2559extern const struct file_operations bpf_iter_fops;
2560extern const struct file_operations bpf_token_fops;
2561
2562#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
2563 extern const struct bpf_prog_ops _name ## _prog_ops; \
2564 extern const struct bpf_verifier_ops _name ## _verifier_ops;
2565#define BPF_MAP_TYPE(_id, _ops) \
2566 extern const struct bpf_map_ops _ops;
2567#define BPF_LINK_TYPE(_id, _name)
2568#include <linux/bpf_types.h>
2569#undef BPF_PROG_TYPE
2570#undef BPF_MAP_TYPE
2571#undef BPF_LINK_TYPE
2572
2573extern const struct bpf_prog_ops bpf_offload_prog_ops;
2574extern const struct bpf_verifier_ops tc_cls_act_analyzer_ops;
2575extern const struct bpf_verifier_ops xdp_analyzer_ops;
2576
2577struct bpf_prog *bpf_prog_get(u32 ufd);
2578struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2579 bool attach_drv);
2580void bpf_prog_add(struct bpf_prog *prog, int i);
2581void bpf_prog_sub(struct bpf_prog *prog, int i);
2582void bpf_prog_inc(struct bpf_prog *prog);
2583struct bpf_prog * __must_check bpf_prog_inc_not_zero(struct bpf_prog *prog);
2584void bpf_prog_put(struct bpf_prog *prog);
2585
2586void bpf_prog_free_id(struct bpf_prog *prog);
2587void bpf_map_free_id(struct bpf_map *map);
2588
2589struct btf_field *btf_record_find(const struct btf_record *rec,
2590 u32 offset, u32 field_mask);
2591void btf_record_free(struct btf_record *rec);
2592void bpf_map_free_record(struct bpf_map *map);
2593struct btf_record *btf_record_dup(const struct btf_record *rec);
2594bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b);
2595void bpf_obj_free_timer(const struct btf_record *rec, void *obj);
2596void bpf_obj_free_workqueue(const struct btf_record *rec, void *obj);
2597void bpf_obj_free_task_work(const struct btf_record *rec, void *obj);
2598void bpf_obj_free_fields(const struct btf_record *rec, void *obj);
2599void __bpf_obj_drop_impl(void *p, const struct btf_record *rec, bool percpu);
2600
2601struct bpf_map *bpf_map_get(u32 ufd);
2602struct bpf_map *bpf_map_get_with_uref(u32 ufd);
2603
2604/*
2605 * The __bpf_map_get() and __btf_get_by_fd() functions parse a file
2606 * descriptor and return a corresponding map or btf object.
2607 * Their names are double underscored to emphasize the fact that they
2608 * do not increase refcnt. To also increase refcnt use corresponding
2609 * bpf_map_get() and btf_get_by_fd() functions.
2610 */
2611
2612static inline struct bpf_map *__bpf_map_get(struct fd f)
2613{
2614 if (fd_empty(f))
2615 return ERR_PTR(-EBADF);
2616 if (unlikely(fd_file(f)->f_op != &bpf_map_fops))
2617 return ERR_PTR(-EINVAL);
2618 return fd_file(f)->private_data;
2619}
2620
2621static inline struct btf *__btf_get_by_fd(struct fd f)
2622{
2623 if (fd_empty(f))
2624 return ERR_PTR(-EBADF);
2625 if (unlikely(fd_file(f)->f_op != &btf_fops))
2626 return ERR_PTR(-EINVAL);
2627 return fd_file(f)->private_data;
2628}
2629
2630void bpf_map_inc(struct bpf_map *map);
2631void bpf_map_inc_with_uref(struct bpf_map *map);
2632struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref);
2633struct bpf_map * __must_check bpf_map_inc_not_zero(struct bpf_map *map);
2634void bpf_map_put_with_uref(struct bpf_map *map);
2635void bpf_map_put(struct bpf_map *map);
2636void *bpf_map_area_alloc(u64 size, int numa_node);
2637void *bpf_map_area_mmapable_alloc(u64 size, int numa_node);
2638void bpf_map_area_free(void *base);
2639bool bpf_map_write_active(const struct bpf_map *map);
2640void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr);
2641int generic_map_lookup_batch(struct bpf_map *map,
2642 const union bpf_attr *attr,
2643 union bpf_attr __user *uattr);
2644int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
2645 const union bpf_attr *attr,
2646 union bpf_attr __user *uattr);
2647int generic_map_delete_batch(struct bpf_map *map,
2648 const union bpf_attr *attr,
2649 union bpf_attr __user *uattr);
2650struct bpf_map *bpf_map_get_curr_or_next(u32 *id);
2651struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id);
2652
2653
2654int bpf_map_alloc_pages(const struct bpf_map *map, int nid,
2655 unsigned long nr_pages, struct page **page_array);
2656#ifdef CONFIG_MEMCG
2657void bpf_map_memcg_enter(const struct bpf_map *map, struct mem_cgroup **old_memcg,
2658 struct mem_cgroup **new_memcg);
2659void bpf_map_memcg_exit(struct mem_cgroup *old_memcg,
2660 struct mem_cgroup *memcg);
2661void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
2662 int node);
2663void *bpf_map_kmalloc_nolock(const struct bpf_map *map, size_t size, gfp_t flags,
2664 int node);
2665void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags);
2666void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
2667 gfp_t flags);
2668void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
2669 size_t align, gfp_t flags);
2670#else
2671/*
2672 * These specialized allocators have to be macros for their allocations to be
2673 * accounted separately (to have separate alloc_tag).
2674 */
2675#define bpf_map_kmalloc_node(_map, _size, _flags, _node) \
2676 kmalloc_node(_size, _flags, _node)
2677#define bpf_map_kmalloc_nolock(_map, _size, _flags, _node) \
2678 kmalloc_nolock(_size, _flags, _node)
2679#define bpf_map_kzalloc(_map, _size, _flags) \
2680 kzalloc(_size, _flags)
2681#define bpf_map_kvcalloc(_map, _n, _size, _flags) \
2682 kvcalloc(_n, _size, _flags)
2683#define bpf_map_alloc_percpu(_map, _size, _align, _flags) \
2684 __alloc_percpu_gfp(_size, _align, _flags)
2685static inline void bpf_map_memcg_enter(const struct bpf_map *map, struct mem_cgroup **old_memcg,
2686 struct mem_cgroup **new_memcg)
2687{
2688 *new_memcg = NULL;
2689 *old_memcg = NULL;
2690}
2691
2692static inline void bpf_map_memcg_exit(struct mem_cgroup *old_memcg,
2693 struct mem_cgroup *memcg)
2694{
2695}
2696#endif
2697
2698static inline int
2699bpf_map_init_elem_count(struct bpf_map *map)
2700{
2701 size_t size = sizeof(*map->elem_count), align = size;
2702 gfp_t flags = GFP_USER | __GFP_NOWARN;
2703
2704 map->elem_count = bpf_map_alloc_percpu(map, size, align, flags);
2705 if (!map->elem_count)
2706 return -ENOMEM;
2707
2708 return 0;
2709}
2710
2711static inline void
2712bpf_map_free_elem_count(struct bpf_map *map)
2713{
2714 free_percpu(map->elem_count);
2715}
2716
2717static inline void bpf_map_inc_elem_count(struct bpf_map *map)
2718{
2719 this_cpu_inc(*map->elem_count);
2720}
2721
2722static inline void bpf_map_dec_elem_count(struct bpf_map *map)
2723{
2724 this_cpu_dec(*map->elem_count);
2725}
2726
2727extern int sysctl_unprivileged_bpf_disabled;
2728
2729bool bpf_token_capable(const struct bpf_token *token, int cap);
2730
2731static inline bool bpf_allow_ptr_leaks(const struct bpf_token *token)
2732{
2733 return bpf_token_capable(token, CAP_PERFMON);
2734}
2735
2736static inline bool bpf_allow_uninit_stack(const struct bpf_token *token)
2737{
2738 return bpf_token_capable(token, CAP_PERFMON);
2739}
2740
2741static inline bool bpf_bypass_spec_v1(const struct bpf_token *token)
2742{
2743 return bpf_jit_bypass_spec_v1() ||
2744 cpu_mitigations_off() ||
2745 bpf_token_capable(token, CAP_PERFMON);
2746}
2747
2748static inline bool bpf_bypass_spec_v4(const struct bpf_token *token)
2749{
2750 return bpf_jit_bypass_spec_v4() ||
2751 cpu_mitigations_off() ||
2752 bpf_token_capable(token, CAP_PERFMON);
2753}
2754
2755int bpf_map_new_fd(struct bpf_map *map, int flags);
2756int bpf_prog_new_fd(struct bpf_prog *prog);
2757
2758void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2759 const struct bpf_link_ops *ops, struct bpf_prog *prog,
2760 enum bpf_attach_type attach_type);
2761void bpf_link_init_sleepable(struct bpf_link *link, enum bpf_link_type type,
2762 const struct bpf_link_ops *ops, struct bpf_prog *prog,
2763 enum bpf_attach_type attach_type, bool sleepable);
2764int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer);
2765int bpf_link_settle(struct bpf_link_primer *primer);
2766void bpf_link_cleanup(struct bpf_link_primer *primer);
2767void bpf_link_inc(struct bpf_link *link);
2768struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link);
2769void bpf_link_put(struct bpf_link *link);
2770int bpf_link_new_fd(struct bpf_link *link);
2771struct bpf_link *bpf_link_get_from_fd(u32 ufd);
2772struct bpf_link *bpf_link_get_curr_or_next(u32 *id);
2773
2774void bpf_token_inc(struct bpf_token *token);
2775void bpf_token_put(struct bpf_token *token);
2776int bpf_token_create(union bpf_attr *attr);
2777struct bpf_token *bpf_token_get_from_fd(u32 ufd);
2778int bpf_token_get_info_by_fd(struct bpf_token *token,
2779 const union bpf_attr *attr,
2780 union bpf_attr __user *uattr);
2781
2782bool bpf_token_allow_cmd(const struct bpf_token *token, enum bpf_cmd cmd);
2783bool bpf_token_allow_map_type(const struct bpf_token *token, enum bpf_map_type type);
2784bool bpf_token_allow_prog_type(const struct bpf_token *token,
2785 enum bpf_prog_type prog_type,
2786 enum bpf_attach_type attach_type);
2787
2788int bpf_obj_pin_user(u32 ufd, int path_fd, const char __user *pathname);
2789int bpf_obj_get_user(int path_fd, const char __user *pathname, int flags);
2790struct inode *bpf_get_inode(struct super_block *sb, const struct inode *dir,
2791 umode_t mode);
2792
2793#define BPF_ITER_FUNC_PREFIX "bpf_iter_"
2794#define DEFINE_BPF_ITER_FUNC(target, args...) \
2795 extern int bpf_iter_ ## target(args); \
2796 int __init bpf_iter_ ## target(args) { return 0; }
2797
2798/*
2799 * The task type of iterators.
2800 *
2801 * For BPF task iterators, they can be parameterized with various
2802 * parameters to visit only some of tasks.
2803 *
2804 * BPF_TASK_ITER_ALL (default)
2805 * Iterate over resources of every task.
2806 *
2807 * BPF_TASK_ITER_TID
2808 * Iterate over resources of a task/tid.
2809 *
2810 * BPF_TASK_ITER_TGID
2811 * Iterate over resources of every task of a process / task group.
2812 */
2813enum bpf_iter_task_type {
2814 BPF_TASK_ITER_ALL = 0,
2815 BPF_TASK_ITER_TID,
2816 BPF_TASK_ITER_TGID,
2817};
2818
2819struct bpf_iter_aux_info {
2820 /* for map_elem iter */
2821 struct bpf_map *map;
2822
2823 /* for cgroup iter */
2824 struct {
2825 struct cgroup *start; /* starting cgroup */
2826 enum bpf_cgroup_iter_order order;
2827 } cgroup;
2828 struct {
2829 enum bpf_iter_task_type type;
2830 u32 pid;
2831 } task;
2832};
2833
2834typedef int (*bpf_iter_attach_target_t)(struct bpf_prog *prog,
2835 union bpf_iter_link_info *linfo,
2836 struct bpf_iter_aux_info *aux);
2837typedef void (*bpf_iter_detach_target_t)(struct bpf_iter_aux_info *aux);
2838typedef void (*bpf_iter_show_fdinfo_t) (const struct bpf_iter_aux_info *aux,
2839 struct seq_file *seq);
2840typedef int (*bpf_iter_fill_link_info_t)(const struct bpf_iter_aux_info *aux,
2841 struct bpf_link_info *info);
2842typedef const struct bpf_func_proto *
2843(*bpf_iter_get_func_proto_t)(enum bpf_func_id func_id,
2844 const struct bpf_prog *prog);
2845
2846enum bpf_iter_feature {
2847 BPF_ITER_RESCHED = BIT(0),
2848};
2849
2850#define BPF_ITER_CTX_ARG_MAX 2
2851struct bpf_iter_reg {
2852 const char *target;
2853 bpf_iter_attach_target_t attach_target;
2854 bpf_iter_detach_target_t detach_target;
2855 bpf_iter_show_fdinfo_t show_fdinfo;
2856 bpf_iter_fill_link_info_t fill_link_info;
2857 bpf_iter_get_func_proto_t get_func_proto;
2858 u32 ctx_arg_info_size;
2859 u32 feature;
2860 struct bpf_ctx_arg_aux ctx_arg_info[BPF_ITER_CTX_ARG_MAX];
2861 const struct bpf_iter_seq_info *seq_info;
2862};
2863
2864struct bpf_iter_meta {
2865 __bpf_md_ptr(struct seq_file *, seq);
2866 u64 session_id;
2867 u64 seq_num;
2868};
2869
2870struct bpf_iter__bpf_map_elem {
2871 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2872 __bpf_md_ptr(struct bpf_map *, map);
2873 __bpf_md_ptr(void *, key);
2874 __bpf_md_ptr(void *, value);
2875};
2876
2877int bpf_iter_reg_target(const struct bpf_iter_reg *reg_info);
2878void bpf_iter_unreg_target(const struct bpf_iter_reg *reg_info);
2879int bpf_iter_prog_supported(struct bpf_prog *prog);
2880const struct bpf_func_proto *
2881bpf_iter_get_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog);
2882int bpf_iter_link_attach(const union bpf_attr *attr, bpfptr_t uattr, struct bpf_prog *prog);
2883int bpf_iter_new_fd(struct bpf_link *link);
2884bool bpf_link_is_iter(struct bpf_link *link);
2885struct bpf_prog *bpf_iter_get_info(struct bpf_iter_meta *meta, bool in_stop);
2886int bpf_iter_run_prog(struct bpf_prog *prog, void *ctx);
2887void bpf_iter_map_show_fdinfo(const struct bpf_iter_aux_info *aux,
2888 struct seq_file *seq);
2889int bpf_iter_map_fill_link_info(const struct bpf_iter_aux_info *aux,
2890 struct bpf_link_info *info);
2891
2892int map_set_for_each_callback_args(struct bpf_verifier_env *env,
2893 struct bpf_func_state *caller,
2894 struct bpf_func_state *callee);
2895
2896int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value, u64 flags);
2897int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value, u64 flags);
2898int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
2899 u64 flags);
2900int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
2901 u64 flags);
2902
2903int bpf_stackmap_extract(struct bpf_map *map, void *key, void *value, bool delete);
2904
2905int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
2906 void *key, void *value, u64 map_flags);
2907int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value);
2908int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file,
2909 void *key, void *value, u64 map_flags);
2910int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value);
2911
2912int bpf_get_file_flag(int flags);
2913int bpf_check_uarg_tail_zero(bpfptr_t uaddr, size_t expected_size,
2914 size_t actual_size);
2915
2916/* verify correctness of eBPF program */
2917int bpf_check(struct bpf_prog **fp, union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size);
2918
2919#ifndef CONFIG_BPF_JIT_ALWAYS_ON
2920void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth);
2921#endif
2922
2923struct btf *bpf_get_btf_vmlinux(void);
2924
2925/* Map specifics */
2926struct xdp_frame;
2927struct sk_buff;
2928struct bpf_dtab_netdev;
2929struct bpf_cpu_map_entry;
2930
2931void __dev_flush(struct list_head *flush_list);
2932int dev_xdp_enqueue(struct net_device *dev, struct xdp_frame *xdpf,
2933 struct net_device *dev_rx);
2934int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_frame *xdpf,
2935 struct net_device *dev_rx);
2936int dev_map_enqueue_multi(struct xdp_frame *xdpf, struct net_device *dev_rx,
2937 struct bpf_map *map, bool exclude_ingress);
2938int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, struct sk_buff *skb,
2939 const struct bpf_prog *xdp_prog);
2940int dev_map_redirect_multi(struct net_device *dev, struct sk_buff *skb,
2941 const struct bpf_prog *xdp_prog,
2942 struct bpf_map *map, bool exclude_ingress);
2943
2944void __cpu_map_flush(struct list_head *flush_list);
2945int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_frame *xdpf,
2946 struct net_device *dev_rx);
2947int cpu_map_generic_redirect(struct bpf_cpu_map_entry *rcpu,
2948 struct sk_buff *skb);
2949
2950/* Return map's numa specified by userspace */
2951static inline int bpf_map_attr_numa_node(const union bpf_attr *attr)
2952{
2953 return (attr->map_flags & BPF_F_NUMA_NODE) ?
2954 attr->numa_node : NUMA_NO_NODE;
2955}
2956
2957struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type);
2958int array_map_alloc_check(union bpf_attr *attr);
2959
2960int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
2961 union bpf_attr __user *uattr);
2962int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
2963 union bpf_attr __user *uattr);
2964int bpf_prog_test_run_tracing(struct bpf_prog *prog,
2965 const union bpf_attr *kattr,
2966 union bpf_attr __user *uattr);
2967int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
2968 const union bpf_attr *kattr,
2969 union bpf_attr __user *uattr);
2970int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
2971 const union bpf_attr *kattr,
2972 union bpf_attr __user *uattr);
2973int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog,
2974 const union bpf_attr *kattr,
2975 union bpf_attr __user *uattr);
2976int bpf_prog_test_run_nf(struct bpf_prog *prog,
2977 const union bpf_attr *kattr,
2978 union bpf_attr __user *uattr);
2979bool btf_ctx_access(int off, int size, enum bpf_access_type type,
2980 const struct bpf_prog *prog,
2981 struct bpf_insn_access_aux *info);
2982
2983static inline bool bpf_tracing_ctx_access(int off, int size,
2984 enum bpf_access_type type)
2985{
2986 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
2987 return false;
2988 if (type != BPF_READ)
2989 return false;
2990 if (off % size != 0)
2991 return false;
2992 return true;
2993}
2994
2995static inline bool bpf_tracing_btf_ctx_access(int off, int size,
2996 enum bpf_access_type type,
2997 const struct bpf_prog *prog,
2998 struct bpf_insn_access_aux *info)
2999{
3000 if (!bpf_tracing_ctx_access(off, size, type))
3001 return false;
3002 return btf_ctx_access(off, size, type, prog, info);
3003}
3004
3005int btf_struct_access(struct bpf_verifier_log *log,
3006 const struct bpf_reg_state *reg,
3007 int off, int size, enum bpf_access_type atype,
3008 u32 *next_btf_id, enum bpf_type_flag *flag, const char **field_name);
3009bool btf_struct_ids_match(struct bpf_verifier_log *log,
3010 const struct btf *btf, u32 id, int off,
3011 const struct btf *need_btf, u32 need_type_id,
3012 bool strict);
3013
3014int btf_distill_func_proto(struct bpf_verifier_log *log,
3015 struct btf *btf,
3016 const struct btf_type *func_proto,
3017 const char *func_name,
3018 struct btf_func_model *m);
3019
3020struct bpf_reg_state;
3021int btf_prepare_func_args(struct bpf_verifier_env *env, int subprog);
3022int btf_check_type_match(struct bpf_verifier_log *log, const struct bpf_prog *prog,
3023 struct btf *btf, const struct btf_type *t);
3024const char *btf_find_decl_tag_value(const struct btf *btf, const struct btf_type *pt,
3025 int comp_idx, const char *tag_key);
3026int btf_find_next_decl_tag(const struct btf *btf, const struct btf_type *pt,
3027 int comp_idx, const char *tag_key, int last_id);
3028
3029struct bpf_prog *bpf_prog_by_id(u32 id);
3030struct bpf_link *bpf_link_by_id(u32 id);
3031
3032const struct bpf_func_proto *bpf_base_func_proto(enum bpf_func_id func_id,
3033 const struct bpf_prog *prog);
3034void bpf_task_storage_free(struct task_struct *task);
3035void bpf_cgrp_storage_free(struct cgroup *cgroup);
3036bool bpf_prog_has_kfunc_call(const struct bpf_prog *prog);
3037const struct btf_func_model *
3038bpf_jit_find_kfunc_model(const struct bpf_prog *prog,
3039 const struct bpf_insn *insn);
3040int bpf_get_kfunc_addr(const struct bpf_prog *prog, u32 func_id,
3041 u16 btf_fd_idx, u8 **func_addr);
3042
3043struct bpf_core_ctx {
3044 struct bpf_verifier_log *log;
3045 const struct btf *btf;
3046};
3047
3048bool btf_nested_type_is_trusted(struct bpf_verifier_log *log,
3049 const struct bpf_reg_state *reg,
3050 const char *field_name, u32 btf_id, const char *suffix);
3051
3052bool btf_type_ids_nocast_alias(struct bpf_verifier_log *log,
3053 const struct btf *reg_btf, u32 reg_id,
3054 const struct btf *arg_btf, u32 arg_id);
3055
3056int bpf_core_apply(struct bpf_core_ctx *ctx, const struct bpf_core_relo *relo,
3057 int relo_idx, void *insn);
3058
3059static inline bool unprivileged_ebpf_enabled(void)
3060{
3061 return !sysctl_unprivileged_bpf_disabled;
3062}
3063
3064/* Not all bpf prog type has the bpf_ctx.
3065 * For the bpf prog type that has initialized the bpf_ctx,
3066 * this function can be used to decide if a kernel function
3067 * is called by a bpf program.
3068 */
3069static inline bool has_current_bpf_ctx(void)
3070{
3071 return !!current->bpf_ctx;
3072}
3073
3074void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog);
3075
3076void bpf_dynptr_init(struct bpf_dynptr_kern *ptr, void *data,
3077 enum bpf_dynptr_type type, u32 offset, u32 size);
3078void bpf_dynptr_set_null(struct bpf_dynptr_kern *ptr);
3079void bpf_dynptr_set_rdonly(struct bpf_dynptr_kern *ptr);
3080void bpf_prog_report_arena_violation(bool write, unsigned long addr, unsigned long fault_ip);
3081
3082#else /* !CONFIG_BPF_SYSCALL */
3083static inline struct bpf_prog *bpf_prog_get(u32 ufd)
3084{
3085 return ERR_PTR(-EOPNOTSUPP);
3086}
3087
3088static inline struct bpf_prog *bpf_prog_get_type_dev(u32 ufd,
3089 enum bpf_prog_type type,
3090 bool attach_drv)
3091{
3092 return ERR_PTR(-EOPNOTSUPP);
3093}
3094
3095static inline void bpf_prog_add(struct bpf_prog *prog, int i)
3096{
3097}
3098
3099static inline void bpf_prog_sub(struct bpf_prog *prog, int i)
3100{
3101}
3102
3103static inline void bpf_prog_put(struct bpf_prog *prog)
3104{
3105}
3106
3107static inline void bpf_prog_inc(struct bpf_prog *prog)
3108{
3109}
3110
3111static inline struct bpf_prog *__must_check
3112bpf_prog_inc_not_zero(struct bpf_prog *prog)
3113{
3114 return ERR_PTR(-EOPNOTSUPP);
3115}
3116
3117static inline void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
3118 const struct bpf_link_ops *ops,
3119 struct bpf_prog *prog, enum bpf_attach_type attach_type)
3120{
3121}
3122
3123static inline void bpf_link_init_sleepable(struct bpf_link *link, enum bpf_link_type type,
3124 const struct bpf_link_ops *ops, struct bpf_prog *prog,
3125 enum bpf_attach_type attach_type, bool sleepable)
3126{
3127}
3128
3129static inline int bpf_link_prime(struct bpf_link *link,
3130 struct bpf_link_primer *primer)
3131{
3132 return -EOPNOTSUPP;
3133}
3134
3135static inline int bpf_link_settle(struct bpf_link_primer *primer)
3136{
3137 return -EOPNOTSUPP;
3138}
3139
3140static inline void bpf_link_cleanup(struct bpf_link_primer *primer)
3141{
3142}
3143
3144static inline void bpf_link_inc(struct bpf_link *link)
3145{
3146}
3147
3148static inline struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
3149{
3150 return NULL;
3151}
3152
3153static inline void bpf_link_put(struct bpf_link *link)
3154{
3155}
3156
3157static inline int bpf_obj_get_user(const char __user *pathname, int flags)
3158{
3159 return -EOPNOTSUPP;
3160}
3161
3162static inline bool bpf_token_capable(const struct bpf_token *token, int cap)
3163{
3164 return capable(cap) || (cap != CAP_SYS_ADMIN && capable(CAP_SYS_ADMIN));
3165}
3166
3167static inline void bpf_token_inc(struct bpf_token *token)
3168{
3169}
3170
3171static inline void bpf_token_put(struct bpf_token *token)
3172{
3173}
3174
3175static inline struct bpf_token *bpf_token_get_from_fd(u32 ufd)
3176{
3177 return ERR_PTR(-EOPNOTSUPP);
3178}
3179
3180static inline int bpf_token_get_info_by_fd(struct bpf_token *token,
3181 const union bpf_attr *attr,
3182 union bpf_attr __user *uattr)
3183{
3184 return -EOPNOTSUPP;
3185}
3186
3187static inline void __dev_flush(struct list_head *flush_list)
3188{
3189}
3190
3191struct xdp_frame;
3192struct bpf_dtab_netdev;
3193struct bpf_cpu_map_entry;
3194
3195static inline
3196int dev_xdp_enqueue(struct net_device *dev, struct xdp_frame *xdpf,
3197 struct net_device *dev_rx)
3198{
3199 return 0;
3200}
3201
3202static inline
3203int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_frame *xdpf,
3204 struct net_device *dev_rx)
3205{
3206 return 0;
3207}
3208
3209static inline
3210int dev_map_enqueue_multi(struct xdp_frame *xdpf, struct net_device *dev_rx,
3211 struct bpf_map *map, bool exclude_ingress)
3212{
3213 return 0;
3214}
3215
3216struct sk_buff;
3217
3218static inline int dev_map_generic_redirect(struct bpf_dtab_netdev *dst,
3219 struct sk_buff *skb,
3220 const struct bpf_prog *xdp_prog)
3221{
3222 return 0;
3223}
3224
3225static inline
3226int dev_map_redirect_multi(struct net_device *dev, struct sk_buff *skb,
3227 const struct bpf_prog *xdp_prog,
3228 struct bpf_map *map, bool exclude_ingress)
3229{
3230 return 0;
3231}
3232
3233static inline void __cpu_map_flush(struct list_head *flush_list)
3234{
3235}
3236
3237static inline int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu,
3238 struct xdp_frame *xdpf,
3239 struct net_device *dev_rx)
3240{
3241 return 0;
3242}
3243
3244static inline int cpu_map_generic_redirect(struct bpf_cpu_map_entry *rcpu,
3245 struct sk_buff *skb)
3246{
3247 return -EOPNOTSUPP;
3248}
3249
3250static inline struct bpf_prog *bpf_prog_get_type_path(const char *name,
3251 enum bpf_prog_type type)
3252{
3253 return ERR_PTR(-EOPNOTSUPP);
3254}
3255
3256static inline int bpf_prog_test_run_xdp(struct bpf_prog *prog,
3257 const union bpf_attr *kattr,
3258 union bpf_attr __user *uattr)
3259{
3260 return -ENOTSUPP;
3261}
3262
3263static inline int bpf_prog_test_run_skb(struct bpf_prog *prog,
3264 const union bpf_attr *kattr,
3265 union bpf_attr __user *uattr)
3266{
3267 return -ENOTSUPP;
3268}
3269
3270static inline int bpf_prog_test_run_tracing(struct bpf_prog *prog,
3271 const union bpf_attr *kattr,
3272 union bpf_attr __user *uattr)
3273{
3274 return -ENOTSUPP;
3275}
3276
3277static inline int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
3278 const union bpf_attr *kattr,
3279 union bpf_attr __user *uattr)
3280{
3281 return -ENOTSUPP;
3282}
3283
3284static inline int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog,
3285 const union bpf_attr *kattr,
3286 union bpf_attr __user *uattr)
3287{
3288 return -ENOTSUPP;
3289}
3290
3291static inline void bpf_map_put(struct bpf_map *map)
3292{
3293}
3294
3295static inline struct bpf_prog *bpf_prog_by_id(u32 id)
3296{
3297 return ERR_PTR(-ENOTSUPP);
3298}
3299
3300static inline int btf_struct_access(struct bpf_verifier_log *log,
3301 const struct bpf_reg_state *reg,
3302 int off, int size, enum bpf_access_type atype,
3303 u32 *next_btf_id, enum bpf_type_flag *flag,
3304 const char **field_name)
3305{
3306 return -EACCES;
3307}
3308
3309static inline const struct bpf_func_proto *
3310bpf_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
3311{
3312 return NULL;
3313}
3314
3315static inline void bpf_task_storage_free(struct task_struct *task)
3316{
3317}
3318
3319static inline bool bpf_prog_has_kfunc_call(const struct bpf_prog *prog)
3320{
3321 return false;
3322}
3323
3324static inline const struct btf_func_model *
3325bpf_jit_find_kfunc_model(const struct bpf_prog *prog,
3326 const struct bpf_insn *insn)
3327{
3328 return NULL;
3329}
3330
3331static inline int
3332bpf_get_kfunc_addr(const struct bpf_prog *prog, u32 func_id,
3333 u16 btf_fd_idx, u8 **func_addr)
3334{
3335 return -ENOTSUPP;
3336}
3337
3338static inline bool unprivileged_ebpf_enabled(void)
3339{
3340 return false;
3341}
3342
3343static inline bool has_current_bpf_ctx(void)
3344{
3345 return false;
3346}
3347
3348static inline void bpf_prog_inc_misses_counter(struct bpf_prog *prog)
3349{
3350}
3351
3352static inline void bpf_cgrp_storage_free(struct cgroup *cgroup)
3353{
3354}
3355
3356static inline void bpf_dynptr_init(struct bpf_dynptr_kern *ptr, void *data,
3357 enum bpf_dynptr_type type, u32 offset, u32 size)
3358{
3359}
3360
3361static inline void bpf_dynptr_set_null(struct bpf_dynptr_kern *ptr)
3362{
3363}
3364
3365static inline void bpf_dynptr_set_rdonly(struct bpf_dynptr_kern *ptr)
3366{
3367}
3368
3369static inline void bpf_prog_report_arena_violation(bool write, unsigned long addr,
3370 unsigned long fault_ip)
3371{
3372}
3373#endif /* CONFIG_BPF_SYSCALL */
3374
3375static inline bool bpf_net_capable(void)
3376{
3377 return capable(CAP_NET_ADMIN) || capable(CAP_SYS_ADMIN);
3378}
3379
3380static __always_inline int
3381bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
3382{
3383 int ret = -EFAULT;
3384
3385 if (IS_ENABLED(CONFIG_BPF_EVENTS))
3386 ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
3387 if (unlikely(ret < 0))
3388 memset(dst, 0, size);
3389 return ret;
3390}
3391
3392void __bpf_free_used_btfs(struct btf_mod_pair *used_btfs, u32 len);
3393
3394static inline struct bpf_prog *bpf_prog_get_type(u32 ufd,
3395 enum bpf_prog_type type)
3396{
3397 return bpf_prog_get_type_dev(ufd, type, false);
3398}
3399
3400void __bpf_free_used_maps(struct bpf_prog_aux *aux,
3401 struct bpf_map **used_maps, u32 len);
3402
3403bool bpf_prog_get_ok(struct bpf_prog *, enum bpf_prog_type *, bool);
3404
3405int bpf_prog_offload_compile(struct bpf_prog *prog);
3406void bpf_prog_dev_bound_destroy(struct bpf_prog *prog);
3407int bpf_prog_offload_info_fill(struct bpf_prog_info *info,
3408 struct bpf_prog *prog);
3409
3410int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map);
3411
3412int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value);
3413int bpf_map_offload_update_elem(struct bpf_map *map,
3414 void *key, void *value, u64 flags);
3415int bpf_map_offload_delete_elem(struct bpf_map *map, void *key);
3416int bpf_map_offload_get_next_key(struct bpf_map *map,
3417 void *key, void *next_key);
3418
3419bool bpf_offload_prog_map_match(struct bpf_prog *prog, struct bpf_map *map);
3420
3421struct bpf_offload_dev *
3422bpf_offload_dev_create(const struct bpf_prog_offload_ops *ops, void *priv);
3423void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev);
3424void *bpf_offload_dev_priv(struct bpf_offload_dev *offdev);
3425int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev,
3426 struct net_device *netdev);
3427void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev,
3428 struct net_device *netdev);
3429bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev);
3430
3431void unpriv_ebpf_notify(int new_state);
3432
3433#if defined(CONFIG_NET) && defined(CONFIG_BPF_SYSCALL)
3434int bpf_dev_bound_kfunc_check(struct bpf_verifier_log *log,
3435 struct bpf_prog_aux *prog_aux);
3436void *bpf_dev_bound_resolve_kfunc(struct bpf_prog *prog, u32 func_id);
3437int bpf_prog_dev_bound_init(struct bpf_prog *prog, union bpf_attr *attr);
3438int bpf_prog_dev_bound_inherit(struct bpf_prog *new_prog, struct bpf_prog *old_prog);
3439void bpf_dev_bound_netdev_unregister(struct net_device *dev);
3440
3441static inline bool bpf_prog_is_dev_bound(const struct bpf_prog_aux *aux)
3442{
3443 return aux->dev_bound;
3444}
3445
3446static inline bool bpf_prog_is_offloaded(const struct bpf_prog_aux *aux)
3447{
3448 return aux->offload_requested;
3449}
3450
3451bool bpf_prog_dev_bound_match(const struct bpf_prog *lhs, const struct bpf_prog *rhs);
3452
3453static inline bool bpf_map_is_offloaded(struct bpf_map *map)
3454{
3455 return unlikely(map->ops == &bpf_map_offload_ops);
3456}
3457
3458struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr);
3459void bpf_map_offload_map_free(struct bpf_map *map);
3460u64 bpf_map_offload_map_mem_usage(const struct bpf_map *map);
3461int bpf_prog_test_run_syscall(struct bpf_prog *prog,
3462 const union bpf_attr *kattr,
3463 union bpf_attr __user *uattr);
3464
3465int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog);
3466int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype);
3467int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value, u64 flags);
3468int sock_map_bpf_prog_query(const union bpf_attr *attr,
3469 union bpf_attr __user *uattr);
3470int sock_map_link_create(const union bpf_attr *attr, struct bpf_prog *prog);
3471
3472void sock_map_unhash(struct sock *sk);
3473void sock_map_destroy(struct sock *sk);
3474void sock_map_close(struct sock *sk, long timeout);
3475#else
3476static inline int bpf_dev_bound_kfunc_check(struct bpf_verifier_log *log,
3477 struct bpf_prog_aux *prog_aux)
3478{
3479 return -EOPNOTSUPP;
3480}
3481
3482static inline void *bpf_dev_bound_resolve_kfunc(struct bpf_prog *prog,
3483 u32 func_id)
3484{
3485 return NULL;
3486}
3487
3488static inline int bpf_prog_dev_bound_init(struct bpf_prog *prog,
3489 union bpf_attr *attr)
3490{
3491 return -EOPNOTSUPP;
3492}
3493
3494static inline int bpf_prog_dev_bound_inherit(struct bpf_prog *new_prog,
3495 struct bpf_prog *old_prog)
3496{
3497 return -EOPNOTSUPP;
3498}
3499
3500static inline void bpf_dev_bound_netdev_unregister(struct net_device *dev)
3501{
3502}
3503
3504static inline bool bpf_prog_is_dev_bound(const struct bpf_prog_aux *aux)
3505{
3506 return false;
3507}
3508
3509static inline bool bpf_prog_is_offloaded(struct bpf_prog_aux *aux)
3510{
3511 return false;
3512}
3513
3514static inline bool bpf_prog_dev_bound_match(const struct bpf_prog *lhs, const struct bpf_prog *rhs)
3515{
3516 return false;
3517}
3518
3519static inline bool bpf_map_is_offloaded(struct bpf_map *map)
3520{
3521 return false;
3522}
3523
3524static inline struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr)
3525{
3526 return ERR_PTR(-EOPNOTSUPP);
3527}
3528
3529static inline void bpf_map_offload_map_free(struct bpf_map *map)
3530{
3531}
3532
3533static inline u64 bpf_map_offload_map_mem_usage(const struct bpf_map *map)
3534{
3535 return 0;
3536}
3537
3538static inline int bpf_prog_test_run_syscall(struct bpf_prog *prog,
3539 const union bpf_attr *kattr,
3540 union bpf_attr __user *uattr)
3541{
3542 return -ENOTSUPP;
3543}
3544
3545#ifdef CONFIG_BPF_SYSCALL
3546static inline int sock_map_get_from_fd(const union bpf_attr *attr,
3547 struct bpf_prog *prog)
3548{
3549 return -EINVAL;
3550}
3551
3552static inline int sock_map_prog_detach(const union bpf_attr *attr,
3553 enum bpf_prog_type ptype)
3554{
3555 return -EOPNOTSUPP;
3556}
3557
3558static inline int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
3559 u64 flags)
3560{
3561 return -EOPNOTSUPP;
3562}
3563
3564static inline int sock_map_bpf_prog_query(const union bpf_attr *attr,
3565 union bpf_attr __user *uattr)
3566{
3567 return -EINVAL;
3568}
3569
3570static inline int sock_map_link_create(const union bpf_attr *attr, struct bpf_prog *prog)
3571{
3572 return -EOPNOTSUPP;
3573}
3574#endif /* CONFIG_BPF_SYSCALL */
3575#endif /* CONFIG_NET && CONFIG_BPF_SYSCALL */
3576
3577static __always_inline void
3578bpf_prog_inc_misses_counters(const struct bpf_prog_array *array)
3579{
3580 const struct bpf_prog_array_item *item;
3581 struct bpf_prog *prog;
3582
3583 if (unlikely(!array))
3584 return;
3585
3586 item = &array->items[0];
3587 while ((prog = READ_ONCE(item->prog))) {
3588 bpf_prog_inc_misses_counter(prog);
3589 item++;
3590 }
3591}
3592
3593#if defined(CONFIG_INET) && defined(CONFIG_BPF_SYSCALL)
3594void bpf_sk_reuseport_detach(struct sock *sk);
3595int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map, void *key,
3596 void *value);
3597int bpf_fd_reuseport_array_update_elem(struct bpf_map *map, void *key,
3598 void *value, u64 map_flags);
3599#else
3600static inline void bpf_sk_reuseport_detach(struct sock *sk)
3601{
3602}
3603
3604#ifdef CONFIG_BPF_SYSCALL
3605static inline int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map,
3606 void *key, void *value)
3607{
3608 return -EOPNOTSUPP;
3609}
3610
3611static inline int bpf_fd_reuseport_array_update_elem(struct bpf_map *map,
3612 void *key, void *value,
3613 u64 map_flags)
3614{
3615 return -EOPNOTSUPP;
3616}
3617#endif /* CONFIG_BPF_SYSCALL */
3618#endif /* defined(CONFIG_INET) && defined(CONFIG_BPF_SYSCALL) */
3619
3620#if defined(CONFIG_KEYS) && defined(CONFIG_BPF_SYSCALL)
3621
3622struct bpf_key *bpf_lookup_user_key(s32 serial, u64 flags);
3623struct bpf_key *bpf_lookup_system_key(u64 id);
3624void bpf_key_put(struct bpf_key *bkey);
3625int bpf_verify_pkcs7_signature(struct bpf_dynptr *data_p,
3626 struct bpf_dynptr *sig_p,
3627 struct bpf_key *trusted_keyring);
3628
3629#else
3630static inline struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags)
3631{
3632 return NULL;
3633}
3634
3635static inline struct bpf_key *bpf_lookup_system_key(u64 id)
3636{
3637 return NULL;
3638}
3639
3640static inline void bpf_key_put(struct bpf_key *bkey)
3641{
3642}
3643
3644static inline int bpf_verify_pkcs7_signature(struct bpf_dynptr *data_p,
3645 struct bpf_dynptr *sig_p,
3646 struct bpf_key *trusted_keyring)
3647{
3648 return -EOPNOTSUPP;
3649}
3650#endif /* defined(CONFIG_KEYS) && defined(CONFIG_BPF_SYSCALL) */
3651
3652/* verifier prototypes for helper functions called from eBPF programs */
3653extern const struct bpf_func_proto bpf_map_lookup_elem_proto;
3654extern const struct bpf_func_proto bpf_map_update_elem_proto;
3655extern const struct bpf_func_proto bpf_map_delete_elem_proto;
3656extern const struct bpf_func_proto bpf_map_push_elem_proto;
3657extern const struct bpf_func_proto bpf_map_pop_elem_proto;
3658extern const struct bpf_func_proto bpf_map_peek_elem_proto;
3659extern const struct bpf_func_proto bpf_map_lookup_percpu_elem_proto;
3660
3661extern const struct bpf_func_proto bpf_get_prandom_u32_proto;
3662extern const struct bpf_func_proto bpf_get_smp_processor_id_proto;
3663extern const struct bpf_func_proto bpf_get_numa_node_id_proto;
3664extern const struct bpf_func_proto bpf_tail_call_proto;
3665extern const struct bpf_func_proto bpf_ktime_get_ns_proto;
3666extern const struct bpf_func_proto bpf_ktime_get_boot_ns_proto;
3667extern const struct bpf_func_proto bpf_ktime_get_tai_ns_proto;
3668extern const struct bpf_func_proto bpf_get_current_pid_tgid_proto;
3669extern const struct bpf_func_proto bpf_get_current_uid_gid_proto;
3670extern const struct bpf_func_proto bpf_get_current_comm_proto;
3671extern const struct bpf_func_proto bpf_get_stackid_proto;
3672extern const struct bpf_func_proto bpf_get_stack_proto;
3673extern const struct bpf_func_proto bpf_get_stack_sleepable_proto;
3674extern const struct bpf_func_proto bpf_get_task_stack_proto;
3675extern const struct bpf_func_proto bpf_get_task_stack_sleepable_proto;
3676extern const struct bpf_func_proto bpf_get_stackid_proto_pe;
3677extern const struct bpf_func_proto bpf_get_stack_proto_pe;
3678extern const struct bpf_func_proto bpf_sock_map_update_proto;
3679extern const struct bpf_func_proto bpf_sock_hash_update_proto;
3680extern const struct bpf_func_proto bpf_get_current_cgroup_id_proto;
3681extern const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto;
3682extern const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto;
3683extern const struct bpf_func_proto bpf_current_task_under_cgroup_proto;
3684extern const struct bpf_func_proto bpf_msg_redirect_hash_proto;
3685extern const struct bpf_func_proto bpf_msg_redirect_map_proto;
3686extern const struct bpf_func_proto bpf_sk_redirect_hash_proto;
3687extern const struct bpf_func_proto bpf_sk_redirect_map_proto;
3688extern const struct bpf_func_proto bpf_spin_lock_proto;
3689extern const struct bpf_func_proto bpf_spin_unlock_proto;
3690extern const struct bpf_func_proto bpf_get_local_storage_proto;
3691extern const struct bpf_func_proto bpf_strtol_proto;
3692extern const struct bpf_func_proto bpf_strtoul_proto;
3693extern const struct bpf_func_proto bpf_tcp_sock_proto;
3694extern const struct bpf_func_proto bpf_jiffies64_proto;
3695extern const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto;
3696extern const struct bpf_func_proto bpf_event_output_data_proto;
3697extern const struct bpf_func_proto bpf_ringbuf_output_proto;
3698extern const struct bpf_func_proto bpf_ringbuf_reserve_proto;
3699extern const struct bpf_func_proto bpf_ringbuf_submit_proto;
3700extern const struct bpf_func_proto bpf_ringbuf_discard_proto;
3701extern const struct bpf_func_proto bpf_ringbuf_query_proto;
3702extern const struct bpf_func_proto bpf_ringbuf_reserve_dynptr_proto;
3703extern const struct bpf_func_proto bpf_ringbuf_submit_dynptr_proto;
3704extern const struct bpf_func_proto bpf_ringbuf_discard_dynptr_proto;
3705extern const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto;
3706extern const struct bpf_func_proto bpf_skc_to_tcp_sock_proto;
3707extern const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto;
3708extern const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto;
3709extern const struct bpf_func_proto bpf_skc_to_udp6_sock_proto;
3710extern const struct bpf_func_proto bpf_skc_to_unix_sock_proto;
3711extern const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto;
3712extern const struct bpf_func_proto bpf_copy_from_user_proto;
3713extern const struct bpf_func_proto bpf_snprintf_btf_proto;
3714extern const struct bpf_func_proto bpf_snprintf_proto;
3715extern const struct bpf_func_proto bpf_per_cpu_ptr_proto;
3716extern const struct bpf_func_proto bpf_this_cpu_ptr_proto;
3717extern const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto;
3718extern const struct bpf_func_proto bpf_sock_from_file_proto;
3719extern const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto;
3720extern const struct bpf_func_proto bpf_task_storage_get_recur_proto;
3721extern const struct bpf_func_proto bpf_task_storage_get_proto;
3722extern const struct bpf_func_proto bpf_task_storage_delete_recur_proto;
3723extern const struct bpf_func_proto bpf_task_storage_delete_proto;
3724extern const struct bpf_func_proto bpf_for_each_map_elem_proto;
3725extern const struct bpf_func_proto bpf_btf_find_by_name_kind_proto;
3726extern const struct bpf_func_proto bpf_sk_setsockopt_proto;
3727extern const struct bpf_func_proto bpf_sk_getsockopt_proto;
3728extern const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto;
3729extern const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto;
3730extern const struct bpf_func_proto bpf_find_vma_proto;
3731extern const struct bpf_func_proto bpf_loop_proto;
3732extern const struct bpf_func_proto bpf_copy_from_user_task_proto;
3733extern const struct bpf_func_proto bpf_set_retval_proto;
3734extern const struct bpf_func_proto bpf_get_retval_proto;
3735extern const struct bpf_func_proto bpf_user_ringbuf_drain_proto;
3736extern const struct bpf_func_proto bpf_cgrp_storage_get_proto;
3737extern const struct bpf_func_proto bpf_cgrp_storage_delete_proto;
3738
3739const struct bpf_func_proto *tracing_prog_func_proto(
3740 enum bpf_func_id func_id, const struct bpf_prog *prog);
3741
3742/* Shared helpers among cBPF and eBPF. */
3743void bpf_user_rnd_init_once(void);
3744u64 bpf_user_rnd_u32(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
3745u64 bpf_get_raw_cpu_id(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
3746
3747#if defined(CONFIG_NET)
3748bool bpf_sock_common_is_valid_access(int off, int size,
3749 enum bpf_access_type type,
3750 struct bpf_insn_access_aux *info);
3751bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
3752 struct bpf_insn_access_aux *info);
3753u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
3754 const struct bpf_insn *si,
3755 struct bpf_insn *insn_buf,
3756 struct bpf_prog *prog,
3757 u32 *target_size);
3758int bpf_dynptr_from_skb_rdonly(struct __sk_buff *skb, u64 flags,
3759 struct bpf_dynptr *ptr);
3760#else
3761static inline bool bpf_sock_common_is_valid_access(int off, int size,
3762 enum bpf_access_type type,
3763 struct bpf_insn_access_aux *info)
3764{
3765 return false;
3766}
3767static inline bool bpf_sock_is_valid_access(int off, int size,
3768 enum bpf_access_type type,
3769 struct bpf_insn_access_aux *info)
3770{
3771 return false;
3772}
3773static inline u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
3774 const struct bpf_insn *si,
3775 struct bpf_insn *insn_buf,
3776 struct bpf_prog *prog,
3777 u32 *target_size)
3778{
3779 return 0;
3780}
3781static inline int bpf_dynptr_from_skb_rdonly(struct __sk_buff *skb, u64 flags,
3782 struct bpf_dynptr *ptr)
3783{
3784 return -EOPNOTSUPP;
3785}
3786#endif
3787
3788#ifdef CONFIG_INET
3789struct sk_reuseport_kern {
3790 struct sk_buff *skb;
3791 struct sock *sk;
3792 struct sock *selected_sk;
3793 struct sock *migrating_sk;
3794 void *data_end;
3795 u32 hash;
3796 u32 reuseport_id;
3797 bool bind_inany;
3798};
3799bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
3800 struct bpf_insn_access_aux *info);
3801
3802u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
3803 const struct bpf_insn *si,
3804 struct bpf_insn *insn_buf,
3805 struct bpf_prog *prog,
3806 u32 *target_size);
3807
3808bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
3809 struct bpf_insn_access_aux *info);
3810
3811u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
3812 const struct bpf_insn *si,
3813 struct bpf_insn *insn_buf,
3814 struct bpf_prog *prog,
3815 u32 *target_size);
3816#else
3817static inline bool bpf_tcp_sock_is_valid_access(int off, int size,
3818 enum bpf_access_type type,
3819 struct bpf_insn_access_aux *info)
3820{
3821 return false;
3822}
3823
3824static inline u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
3825 const struct bpf_insn *si,
3826 struct bpf_insn *insn_buf,
3827 struct bpf_prog *prog,
3828 u32 *target_size)
3829{
3830 return 0;
3831}
3832static inline bool bpf_xdp_sock_is_valid_access(int off, int size,
3833 enum bpf_access_type type,
3834 struct bpf_insn_access_aux *info)
3835{
3836 return false;
3837}
3838
3839static inline u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
3840 const struct bpf_insn *si,
3841 struct bpf_insn *insn_buf,
3842 struct bpf_prog *prog,
3843 u32 *target_size)
3844{
3845 return 0;
3846}
3847#endif /* CONFIG_INET */
3848
3849enum bpf_text_poke_type {
3850 BPF_MOD_NOP,
3851 BPF_MOD_CALL,
3852 BPF_MOD_JUMP,
3853};
3854
3855int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type old_t,
3856 enum bpf_text_poke_type new_t, void *old_addr,
3857 void *new_addr);
3858
3859void bpf_arch_poke_desc_update(struct bpf_jit_poke_descriptor *poke,
3860 struct bpf_prog *new, struct bpf_prog *old);
3861
3862void *bpf_arch_text_copy(void *dst, void *src, size_t len);
3863int bpf_arch_text_invalidate(void *dst, size_t len);
3864
3865struct btf_id_set;
3866bool btf_id_set_contains(const struct btf_id_set *set, u32 id);
3867
3868#define MAX_BPRINTF_VARARGS 12
3869#define MAX_BPRINTF_BUF 1024
3870
3871/* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary
3872 * arguments representation.
3873 */
3874#define MAX_BPRINTF_BIN_ARGS 512
3875
3876struct bpf_bprintf_buffers {
3877 char bin_args[MAX_BPRINTF_BIN_ARGS];
3878 char buf[MAX_BPRINTF_BUF];
3879};
3880
3881struct bpf_bprintf_data {
3882 u32 *bin_args;
3883 char *buf;
3884 bool get_bin_args;
3885 bool get_buf;
3886};
3887
3888int bpf_bprintf_prepare(const char *fmt, u32 fmt_size, const u64 *raw_args,
3889 u32 num_args, struct bpf_bprintf_data *data);
3890void bpf_bprintf_cleanup(struct bpf_bprintf_data *data);
3891int bpf_try_get_buffers(struct bpf_bprintf_buffers **bufs);
3892void bpf_put_buffers(void);
3893
3894void bpf_prog_stream_init(struct bpf_prog *prog);
3895void bpf_prog_stream_free(struct bpf_prog *prog);
3896int bpf_prog_stream_read(struct bpf_prog *prog, enum bpf_stream_id stream_id, void __user *buf, int len);
3897void bpf_stream_stage_init(struct bpf_stream_stage *ss);
3898void bpf_stream_stage_free(struct bpf_stream_stage *ss);
3899__printf(2, 3)
3900int bpf_stream_stage_printk(struct bpf_stream_stage *ss, const char *fmt, ...);
3901int bpf_stream_stage_commit(struct bpf_stream_stage *ss, struct bpf_prog *prog,
3902 enum bpf_stream_id stream_id);
3903int bpf_stream_stage_dump_stack(struct bpf_stream_stage *ss);
3904
3905#define bpf_stream_printk(ss, ...) bpf_stream_stage_printk(&ss, __VA_ARGS__)
3906#define bpf_stream_dump_stack(ss) bpf_stream_stage_dump_stack(&ss)
3907
3908#define bpf_stream_stage(ss, prog, stream_id, expr) \
3909 ({ \
3910 bpf_stream_stage_init(&ss); \
3911 (expr); \
3912 bpf_stream_stage_commit(&ss, prog, stream_id); \
3913 bpf_stream_stage_free(&ss); \
3914 })
3915
3916#ifdef CONFIG_BPF_LSM
3917void bpf_cgroup_atype_get(u32 attach_btf_id, int cgroup_atype);
3918void bpf_cgroup_atype_put(int cgroup_atype);
3919#else
3920static inline void bpf_cgroup_atype_get(u32 attach_btf_id, int cgroup_atype) {}
3921static inline void bpf_cgroup_atype_put(int cgroup_atype) {}
3922#endif /* CONFIG_BPF_LSM */
3923
3924struct key;
3925
3926#ifdef CONFIG_KEYS
3927struct bpf_key {
3928 struct key *key;
3929 bool has_ref;
3930};
3931#endif /* CONFIG_KEYS */
3932
3933static inline bool type_is_alloc(u32 type)
3934{
3935 return type & MEM_ALLOC;
3936}
3937
3938static inline gfp_t bpf_memcg_flags(gfp_t flags)
3939{
3940 if (memcg_bpf_enabled())
3941 return flags | __GFP_ACCOUNT;
3942 return flags;
3943}
3944
3945static inline bool bpf_is_subprog(const struct bpf_prog *prog)
3946{
3947 return prog->aux->func_idx != 0;
3948}
3949
3950const struct bpf_line_info *bpf_find_linfo(const struct bpf_prog *prog, u32 insn_off);
3951void bpf_get_linfo_file_line(struct btf *btf, const struct bpf_line_info *linfo,
3952 const char **filep, const char **linep, int *nump);
3953int bpf_prog_get_file_line(struct bpf_prog *prog, unsigned long ip, const char **filep,
3954 const char **linep, int *nump);
3955struct bpf_prog *bpf_prog_find_from_stack(void);
3956
3957int bpf_insn_array_init(struct bpf_map *map, const struct bpf_prog *prog);
3958int bpf_insn_array_ready(struct bpf_map *map);
3959void bpf_insn_array_release(struct bpf_map *map);
3960void bpf_insn_array_adjust(struct bpf_map *map, u32 off, u32 len);
3961void bpf_insn_array_adjust_after_remove(struct bpf_map *map, u32 off, u32 len);
3962
3963#ifdef CONFIG_BPF_SYSCALL
3964void bpf_prog_update_insn_ptrs(struct bpf_prog *prog, u32 *offsets, void *image);
3965#else
3966static inline void
3967bpf_prog_update_insn_ptrs(struct bpf_prog *prog, u32 *offsets, void *image)
3968{
3969}
3970#endif
3971
3972static inline bool bpf_map_supports_cpu_flags(enum bpf_map_type map_type)
3973{
3974 switch (map_type) {
3975 case BPF_MAP_TYPE_PERCPU_ARRAY:
3976 case BPF_MAP_TYPE_PERCPU_HASH:
3977 case BPF_MAP_TYPE_LRU_PERCPU_HASH:
3978 case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
3979 return true;
3980 default:
3981 return false;
3982 }
3983}
3984
3985static inline int bpf_map_check_op_flags(struct bpf_map *map, u64 flags, u64 allowed_flags)
3986{
3987 u32 cpu;
3988
3989 if ((u32)flags & ~allowed_flags)
3990 return -EINVAL;
3991
3992 if ((flags & BPF_F_LOCK) && !btf_record_has_field(map->record, BPF_SPIN_LOCK))
3993 return -EINVAL;
3994
3995 if (!(flags & BPF_F_CPU) && flags >> 32)
3996 return -EINVAL;
3997
3998 if (flags & (BPF_F_CPU | BPF_F_ALL_CPUS)) {
3999 if (!bpf_map_supports_cpu_flags(map->map_type))
4000 return -EINVAL;
4001 if ((flags & BPF_F_CPU) && (flags & BPF_F_ALL_CPUS))
4002 return -EINVAL;
4003
4004 cpu = flags >> 32;
4005 if ((flags & BPF_F_CPU) && cpu >= num_possible_cpus())
4006 return -ERANGE;
4007 }
4008
4009 return 0;
4010}
4011
4012#endif /* _LINUX_BPF_H */