xref: /linux-6.15/include/linux/bpf.h (revision 520b7aa0)
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 
9 #include <linux/workqueue.h>
10 #include <linux/file.h>
11 #include <linux/percpu.h>
12 #include <linux/err.h>
13 #include <linux/rbtree_latch.h>
14 #include <linux/numa.h>
15 #include <linux/mm_types.h>
16 #include <linux/wait.h>
17 #include <linux/u64_stats_sync.h>
18 #include <linux/refcount.h>
19 #include <linux/mutex.h>
20 #include <linux/module.h>
21 #include <linux/kallsyms.h>
22 
23 struct bpf_verifier_env;
24 struct bpf_verifier_log;
25 struct perf_event;
26 struct bpf_prog;
27 struct bpf_prog_aux;
28 struct bpf_map;
29 struct sock;
30 struct seq_file;
31 struct btf;
32 struct btf_type;
33 struct exception_table_entry;
34 
35 extern struct idr btf_idr;
36 extern spinlock_t btf_idr_lock;
37 
38 /* map is generic key/value storage optionally accesible by eBPF programs */
39 struct bpf_map_ops {
40 	/* funcs callable from userspace (via syscall) */
41 	int (*map_alloc_check)(union bpf_attr *attr);
42 	struct bpf_map *(*map_alloc)(union bpf_attr *attr);
43 	void (*map_release)(struct bpf_map *map, struct file *map_file);
44 	void (*map_free)(struct bpf_map *map);
45 	int (*map_get_next_key)(struct bpf_map *map, void *key, void *next_key);
46 	void (*map_release_uref)(struct bpf_map *map);
47 	void *(*map_lookup_elem_sys_only)(struct bpf_map *map, void *key);
48 	int (*map_lookup_batch)(struct bpf_map *map, const union bpf_attr *attr,
49 				union bpf_attr __user *uattr);
50 	int (*map_lookup_and_delete_batch)(struct bpf_map *map,
51 					   const union bpf_attr *attr,
52 					   union bpf_attr __user *uattr);
53 	int (*map_update_batch)(struct bpf_map *map, const union bpf_attr *attr,
54 				union bpf_attr __user *uattr);
55 	int (*map_delete_batch)(struct bpf_map *map, const union bpf_attr *attr,
56 				union bpf_attr __user *uattr);
57 
58 	/* funcs callable from userspace and from eBPF programs */
59 	void *(*map_lookup_elem)(struct bpf_map *map, void *key);
60 	int (*map_update_elem)(struct bpf_map *map, void *key, void *value, u64 flags);
61 	int (*map_delete_elem)(struct bpf_map *map, void *key);
62 	int (*map_push_elem)(struct bpf_map *map, void *value, u64 flags);
63 	int (*map_pop_elem)(struct bpf_map *map, void *value);
64 	int (*map_peek_elem)(struct bpf_map *map, void *value);
65 
66 	/* funcs called by prog_array and perf_event_array map */
67 	void *(*map_fd_get_ptr)(struct bpf_map *map, struct file *map_file,
68 				int fd);
69 	void (*map_fd_put_ptr)(void *ptr);
70 	u32 (*map_gen_lookup)(struct bpf_map *map, struct bpf_insn *insn_buf);
71 	u32 (*map_fd_sys_lookup_elem)(void *ptr);
72 	void (*map_seq_show_elem)(struct bpf_map *map, void *key,
73 				  struct seq_file *m);
74 	int (*map_check_btf)(const struct bpf_map *map,
75 			     const struct btf *btf,
76 			     const struct btf_type *key_type,
77 			     const struct btf_type *value_type);
78 
79 	/* Prog poke tracking helpers. */
80 	int (*map_poke_track)(struct bpf_map *map, struct bpf_prog_aux *aux);
81 	void (*map_poke_untrack)(struct bpf_map *map, struct bpf_prog_aux *aux);
82 	void (*map_poke_run)(struct bpf_map *map, u32 key, struct bpf_prog *old,
83 			     struct bpf_prog *new);
84 
85 	/* Direct value access helpers. */
86 	int (*map_direct_value_addr)(const struct bpf_map *map,
87 				     u64 *imm, u32 off);
88 	int (*map_direct_value_meta)(const struct bpf_map *map,
89 				     u64 imm, u32 *off);
90 	int (*map_mmap)(struct bpf_map *map, struct vm_area_struct *vma);
91 };
92 
93 struct bpf_map_memory {
94 	u32 pages;
95 	struct user_struct *user;
96 };
97 
98 struct bpf_map {
99 	/* The first two cachelines with read-mostly members of which some
100 	 * are also accessed in fast-path (e.g. ops, max_entries).
101 	 */
102 	const struct bpf_map_ops *ops ____cacheline_aligned;
103 	struct bpf_map *inner_map_meta;
104 #ifdef CONFIG_SECURITY
105 	void *security;
106 #endif
107 	enum bpf_map_type map_type;
108 	u32 key_size;
109 	u32 value_size;
110 	u32 max_entries;
111 	u32 map_flags;
112 	int spin_lock_off; /* >=0 valid offset, <0 error */
113 	u32 id;
114 	int numa_node;
115 	u32 btf_key_type_id;
116 	u32 btf_value_type_id;
117 	struct btf *btf;
118 	struct bpf_map_memory memory;
119 	char name[BPF_OBJ_NAME_LEN];
120 	u32 btf_vmlinux_value_type_id;
121 	bool unpriv_array;
122 	bool frozen; /* write-once; write-protected by freeze_mutex */
123 	/* 22 bytes hole */
124 
125 	/* The 3rd and 4th cacheline with misc members to avoid false sharing
126 	 * particularly with refcounting.
127 	 */
128 	atomic64_t refcnt ____cacheline_aligned;
129 	atomic64_t usercnt;
130 	struct work_struct work;
131 	struct mutex freeze_mutex;
132 	u64 writecnt; /* writable mmap cnt; protected by freeze_mutex */
133 };
134 
135 static inline bool map_value_has_spin_lock(const struct bpf_map *map)
136 {
137 	return map->spin_lock_off >= 0;
138 }
139 
140 static inline void check_and_init_map_lock(struct bpf_map *map, void *dst)
141 {
142 	if (likely(!map_value_has_spin_lock(map)))
143 		return;
144 	*(struct bpf_spin_lock *)(dst + map->spin_lock_off) =
145 		(struct bpf_spin_lock){};
146 }
147 
148 /* copy everything but bpf_spin_lock */
149 static inline void copy_map_value(struct bpf_map *map, void *dst, void *src)
150 {
151 	if (unlikely(map_value_has_spin_lock(map))) {
152 		u32 off = map->spin_lock_off;
153 
154 		memcpy(dst, src, off);
155 		memcpy(dst + off + sizeof(struct bpf_spin_lock),
156 		       src + off + sizeof(struct bpf_spin_lock),
157 		       map->value_size - off - sizeof(struct bpf_spin_lock));
158 	} else {
159 		memcpy(dst, src, map->value_size);
160 	}
161 }
162 void copy_map_value_locked(struct bpf_map *map, void *dst, void *src,
163 			   bool lock_src);
164 
165 struct bpf_offload_dev;
166 struct bpf_offloaded_map;
167 
168 struct bpf_map_dev_ops {
169 	int (*map_get_next_key)(struct bpf_offloaded_map *map,
170 				void *key, void *next_key);
171 	int (*map_lookup_elem)(struct bpf_offloaded_map *map,
172 			       void *key, void *value);
173 	int (*map_update_elem)(struct bpf_offloaded_map *map,
174 			       void *key, void *value, u64 flags);
175 	int (*map_delete_elem)(struct bpf_offloaded_map *map, void *key);
176 };
177 
178 struct bpf_offloaded_map {
179 	struct bpf_map map;
180 	struct net_device *netdev;
181 	const struct bpf_map_dev_ops *dev_ops;
182 	void *dev_priv;
183 	struct list_head offloads;
184 };
185 
186 static inline struct bpf_offloaded_map *map_to_offmap(struct bpf_map *map)
187 {
188 	return container_of(map, struct bpf_offloaded_map, map);
189 }
190 
191 static inline bool bpf_map_offload_neutral(const struct bpf_map *map)
192 {
193 	return map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
194 }
195 
196 static inline bool bpf_map_support_seq_show(const struct bpf_map *map)
197 {
198 	return (map->btf_value_type_id || map->btf_vmlinux_value_type_id) &&
199 		map->ops->map_seq_show_elem;
200 }
201 
202 int map_check_no_btf(const struct bpf_map *map,
203 		     const struct btf *btf,
204 		     const struct btf_type *key_type,
205 		     const struct btf_type *value_type);
206 
207 extern const struct bpf_map_ops bpf_map_offload_ops;
208 
209 /* function argument constraints */
210 enum bpf_arg_type {
211 	ARG_DONTCARE = 0,	/* unused argument in helper function */
212 
213 	/* the following constraints used to prototype
214 	 * bpf_map_lookup/update/delete_elem() functions
215 	 */
216 	ARG_CONST_MAP_PTR,	/* const argument used as pointer to bpf_map */
217 	ARG_PTR_TO_MAP_KEY,	/* pointer to stack used as map key */
218 	ARG_PTR_TO_MAP_VALUE,	/* pointer to stack used as map value */
219 	ARG_PTR_TO_UNINIT_MAP_VALUE,	/* pointer to valid memory used to store a map value */
220 	ARG_PTR_TO_MAP_VALUE_OR_NULL,	/* pointer to stack used as map value or NULL */
221 
222 	/* the following constraints used to prototype bpf_memcmp() and other
223 	 * functions that access data on eBPF program stack
224 	 */
225 	ARG_PTR_TO_MEM,		/* pointer to valid memory (stack, packet, map value) */
226 	ARG_PTR_TO_MEM_OR_NULL, /* pointer to valid memory or NULL */
227 	ARG_PTR_TO_UNINIT_MEM,	/* pointer to memory does not need to be initialized,
228 				 * helper function must fill all bytes or clear
229 				 * them in error case.
230 				 */
231 
232 	ARG_CONST_SIZE,		/* number of bytes accessed from memory */
233 	ARG_CONST_SIZE_OR_ZERO,	/* number of bytes accessed from memory or 0 */
234 
235 	ARG_PTR_TO_CTX,		/* pointer to context */
236 	ARG_PTR_TO_CTX_OR_NULL,	/* pointer to context or NULL */
237 	ARG_ANYTHING,		/* any (initialized) argument is ok */
238 	ARG_PTR_TO_SPIN_LOCK,	/* pointer to bpf_spin_lock */
239 	ARG_PTR_TO_SOCK_COMMON,	/* pointer to sock_common */
240 	ARG_PTR_TO_INT,		/* pointer to int */
241 	ARG_PTR_TO_LONG,	/* pointer to long */
242 	ARG_PTR_TO_SOCKET,	/* pointer to bpf_sock (fullsock) */
243 	ARG_PTR_TO_BTF_ID,	/* pointer to in-kernel struct */
244 };
245 
246 /* type of values returned from helper functions */
247 enum bpf_return_type {
248 	RET_INTEGER,			/* function returns integer */
249 	RET_VOID,			/* function doesn't return anything */
250 	RET_PTR_TO_MAP_VALUE,		/* returns a pointer to map elem value */
251 	RET_PTR_TO_MAP_VALUE_OR_NULL,	/* returns a pointer to map elem value or NULL */
252 	RET_PTR_TO_SOCKET_OR_NULL,	/* returns a pointer to a socket or NULL */
253 	RET_PTR_TO_TCP_SOCK_OR_NULL,	/* returns a pointer to a tcp_sock or NULL */
254 	RET_PTR_TO_SOCK_COMMON_OR_NULL,	/* returns a pointer to a sock_common or NULL */
255 };
256 
257 /* eBPF function prototype used by verifier to allow BPF_CALLs from eBPF programs
258  * to in-kernel helper functions and for adjusting imm32 field in BPF_CALL
259  * instructions after verifying
260  */
261 struct bpf_func_proto {
262 	u64 (*func)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
263 	bool gpl_only;
264 	bool pkt_access;
265 	enum bpf_return_type ret_type;
266 	union {
267 		struct {
268 			enum bpf_arg_type arg1_type;
269 			enum bpf_arg_type arg2_type;
270 			enum bpf_arg_type arg3_type;
271 			enum bpf_arg_type arg4_type;
272 			enum bpf_arg_type arg5_type;
273 		};
274 		enum bpf_arg_type arg_type[5];
275 	};
276 	int *btf_id; /* BTF ids of arguments */
277 };
278 
279 /* bpf_context is intentionally undefined structure. Pointer to bpf_context is
280  * the first argument to eBPF programs.
281  * For socket filters: 'struct bpf_context *' == 'struct sk_buff *'
282  */
283 struct bpf_context;
284 
285 enum bpf_access_type {
286 	BPF_READ = 1,
287 	BPF_WRITE = 2
288 };
289 
290 /* types of values stored in eBPF registers */
291 /* Pointer types represent:
292  * pointer
293  * pointer + imm
294  * pointer + (u16) var
295  * pointer + (u16) var + imm
296  * if (range > 0) then [ptr, ptr + range - off) is safe to access
297  * if (id > 0) means that some 'var' was added
298  * if (off > 0) means that 'imm' was added
299  */
300 enum bpf_reg_type {
301 	NOT_INIT = 0,		 /* nothing was written into register */
302 	SCALAR_VALUE,		 /* reg doesn't contain a valid pointer */
303 	PTR_TO_CTX,		 /* reg points to bpf_context */
304 	CONST_PTR_TO_MAP,	 /* reg points to struct bpf_map */
305 	PTR_TO_MAP_VALUE,	 /* reg points to map element value */
306 	PTR_TO_MAP_VALUE_OR_NULL,/* points to map elem value or NULL */
307 	PTR_TO_STACK,		 /* reg == frame_pointer + offset */
308 	PTR_TO_PACKET_META,	 /* skb->data - meta_len */
309 	PTR_TO_PACKET,		 /* reg points to skb->data */
310 	PTR_TO_PACKET_END,	 /* skb->data + headlen */
311 	PTR_TO_FLOW_KEYS,	 /* reg points to bpf_flow_keys */
312 	PTR_TO_SOCKET,		 /* reg points to struct bpf_sock */
313 	PTR_TO_SOCKET_OR_NULL,	 /* reg points to struct bpf_sock or NULL */
314 	PTR_TO_SOCK_COMMON,	 /* reg points to sock_common */
315 	PTR_TO_SOCK_COMMON_OR_NULL, /* reg points to sock_common or NULL */
316 	PTR_TO_TCP_SOCK,	 /* reg points to struct tcp_sock */
317 	PTR_TO_TCP_SOCK_OR_NULL, /* reg points to struct tcp_sock or NULL */
318 	PTR_TO_TP_BUFFER,	 /* reg points to a writable raw tp's buffer */
319 	PTR_TO_XDP_SOCK,	 /* reg points to struct xdp_sock */
320 	PTR_TO_BTF_ID,		 /* reg points to kernel struct */
321 };
322 
323 /* The information passed from prog-specific *_is_valid_access
324  * back to the verifier.
325  */
326 struct bpf_insn_access_aux {
327 	enum bpf_reg_type reg_type;
328 	union {
329 		int ctx_field_size;
330 		u32 btf_id;
331 	};
332 	struct bpf_verifier_log *log; /* for verbose logs */
333 };
334 
335 static inline void
336 bpf_ctx_record_field_size(struct bpf_insn_access_aux *aux, u32 size)
337 {
338 	aux->ctx_field_size = size;
339 }
340 
341 struct bpf_prog_ops {
342 	int (*test_run)(struct bpf_prog *prog, const union bpf_attr *kattr,
343 			union bpf_attr __user *uattr);
344 };
345 
346 struct bpf_verifier_ops {
347 	/* return eBPF function prototype for verification */
348 	const struct bpf_func_proto *
349 	(*get_func_proto)(enum bpf_func_id func_id,
350 			  const struct bpf_prog *prog);
351 
352 	/* return true if 'size' wide access at offset 'off' within bpf_context
353 	 * with 'type' (read or write) is allowed
354 	 */
355 	bool (*is_valid_access)(int off, int size, enum bpf_access_type type,
356 				const struct bpf_prog *prog,
357 				struct bpf_insn_access_aux *info);
358 	int (*gen_prologue)(struct bpf_insn *insn, bool direct_write,
359 			    const struct bpf_prog *prog);
360 	int (*gen_ld_abs)(const struct bpf_insn *orig,
361 			  struct bpf_insn *insn_buf);
362 	u32 (*convert_ctx_access)(enum bpf_access_type type,
363 				  const struct bpf_insn *src,
364 				  struct bpf_insn *dst,
365 				  struct bpf_prog *prog, u32 *target_size);
366 	int (*btf_struct_access)(struct bpf_verifier_log *log,
367 				 const struct btf_type *t, int off, int size,
368 				 enum bpf_access_type atype,
369 				 u32 *next_btf_id);
370 };
371 
372 struct bpf_prog_offload_ops {
373 	/* verifier basic callbacks */
374 	int (*insn_hook)(struct bpf_verifier_env *env,
375 			 int insn_idx, int prev_insn_idx);
376 	int (*finalize)(struct bpf_verifier_env *env);
377 	/* verifier optimization callbacks (called after .finalize) */
378 	int (*replace_insn)(struct bpf_verifier_env *env, u32 off,
379 			    struct bpf_insn *insn);
380 	int (*remove_insns)(struct bpf_verifier_env *env, u32 off, u32 cnt);
381 	/* program management callbacks */
382 	int (*prepare)(struct bpf_prog *prog);
383 	int (*translate)(struct bpf_prog *prog);
384 	void (*destroy)(struct bpf_prog *prog);
385 };
386 
387 struct bpf_prog_offload {
388 	struct bpf_prog		*prog;
389 	struct net_device	*netdev;
390 	struct bpf_offload_dev	*offdev;
391 	void			*dev_priv;
392 	struct list_head	offloads;
393 	bool			dev_state;
394 	bool			opt_failed;
395 	void			*jited_image;
396 	u32			jited_len;
397 };
398 
399 enum bpf_cgroup_storage_type {
400 	BPF_CGROUP_STORAGE_SHARED,
401 	BPF_CGROUP_STORAGE_PERCPU,
402 	__BPF_CGROUP_STORAGE_MAX
403 };
404 
405 #define MAX_BPF_CGROUP_STORAGE_TYPE __BPF_CGROUP_STORAGE_MAX
406 
407 /* The longest tracepoint has 12 args.
408  * See include/trace/bpf_probe.h
409  */
410 #define MAX_BPF_FUNC_ARGS 12
411 
412 struct bpf_prog_stats {
413 	u64 cnt;
414 	u64 nsecs;
415 	struct u64_stats_sync syncp;
416 } __aligned(2 * sizeof(u64));
417 
418 struct btf_func_model {
419 	u8 ret_size;
420 	u8 nr_args;
421 	u8 arg_size[MAX_BPF_FUNC_ARGS];
422 };
423 
424 /* Restore arguments before returning from trampoline to let original function
425  * continue executing. This flag is used for fentry progs when there are no
426  * fexit progs.
427  */
428 #define BPF_TRAMP_F_RESTORE_REGS	BIT(0)
429 /* Call original function after fentry progs, but before fexit progs.
430  * Makes sense for fentry/fexit, normal calls and indirect calls.
431  */
432 #define BPF_TRAMP_F_CALL_ORIG		BIT(1)
433 /* Skip current frame and return to parent.  Makes sense for fentry/fexit
434  * programs only. Should not be used with normal calls and indirect calls.
435  */
436 #define BPF_TRAMP_F_SKIP_FRAME		BIT(2)
437 
438 /* Each call __bpf_prog_enter + call bpf_func + call __bpf_prog_exit is ~50
439  * bytes on x86.  Pick a number to fit into BPF_IMAGE_SIZE / 2
440  */
441 #define BPF_MAX_TRAMP_PROGS 40
442 
443 struct bpf_tramp_progs {
444 	struct bpf_prog *progs[BPF_MAX_TRAMP_PROGS];
445 	int nr_progs;
446 };
447 
448 /* Different use cases for BPF trampoline:
449  * 1. replace nop at the function entry (kprobe equivalent)
450  *    flags = BPF_TRAMP_F_RESTORE_REGS
451  *    fentry = a set of programs to run before returning from trampoline
452  *
453  * 2. replace nop at the function entry (kprobe + kretprobe equivalent)
454  *    flags = BPF_TRAMP_F_CALL_ORIG | BPF_TRAMP_F_SKIP_FRAME
455  *    orig_call = fentry_ip + MCOUNT_INSN_SIZE
456  *    fentry = a set of program to run before calling original function
457  *    fexit = a set of program to run after original function
458  *
459  * 3. replace direct call instruction anywhere in the function body
460  *    or assign a function pointer for indirect call (like tcp_congestion_ops->cong_avoid)
461  *    With flags = 0
462  *      fentry = a set of programs to run before returning from trampoline
463  *    With flags = BPF_TRAMP_F_CALL_ORIG
464  *      orig_call = original callback addr or direct function addr
465  *      fentry = a set of program to run before calling original function
466  *      fexit = a set of program to run after original function
467  */
468 int arch_prepare_bpf_trampoline(void *image, void *image_end,
469 				const struct btf_func_model *m, u32 flags,
470 				struct bpf_tramp_progs *tprogs,
471 				void *orig_call);
472 /* these two functions are called from generated trampoline */
473 u64 notrace __bpf_prog_enter(void);
474 void notrace __bpf_prog_exit(struct bpf_prog *prog, u64 start);
475 
476 struct bpf_ksym {
477 	unsigned long		 start;
478 	unsigned long		 end;
479 	char			 name[KSYM_NAME_LEN];
480 	struct list_head	 lnode;
481 	struct latch_tree_node	 tnode;
482 	bool			 prog;
483 };
484 
485 enum bpf_tramp_prog_type {
486 	BPF_TRAMP_FENTRY,
487 	BPF_TRAMP_FEXIT,
488 	BPF_TRAMP_MODIFY_RETURN,
489 	BPF_TRAMP_MAX,
490 	BPF_TRAMP_REPLACE, /* more than MAX */
491 };
492 
493 struct bpf_trampoline {
494 	/* hlist for trampoline_table */
495 	struct hlist_node hlist;
496 	/* serializes access to fields of this trampoline */
497 	struct mutex mutex;
498 	refcount_t refcnt;
499 	u64 key;
500 	struct {
501 		struct btf_func_model model;
502 		void *addr;
503 		bool ftrace_managed;
504 	} func;
505 	/* if !NULL this is BPF_PROG_TYPE_EXT program that extends another BPF
506 	 * program by replacing one of its functions. func.addr is the address
507 	 * of the function it replaced.
508 	 */
509 	struct bpf_prog *extension_prog;
510 	/* list of BPF programs using this trampoline */
511 	struct hlist_head progs_hlist[BPF_TRAMP_MAX];
512 	/* Number of attached programs. A counter per kind. */
513 	int progs_cnt[BPF_TRAMP_MAX];
514 	/* Executable image of trampoline */
515 	void *image;
516 	u64 selector;
517 	struct bpf_ksym ksym;
518 };
519 
520 #define BPF_DISPATCHER_MAX 48 /* Fits in 2048B */
521 
522 struct bpf_dispatcher_prog {
523 	struct bpf_prog *prog;
524 	refcount_t users;
525 };
526 
527 struct bpf_dispatcher {
528 	/* dispatcher mutex */
529 	struct mutex mutex;
530 	void *func;
531 	struct bpf_dispatcher_prog progs[BPF_DISPATCHER_MAX];
532 	int num_progs;
533 	void *image;
534 	u32 image_off;
535 	struct bpf_ksym ksym;
536 };
537 
538 static __always_inline unsigned int bpf_dispatcher_nop_func(
539 	const void *ctx,
540 	const struct bpf_insn *insnsi,
541 	unsigned int (*bpf_func)(const void *,
542 				 const struct bpf_insn *))
543 {
544 	return bpf_func(ctx, insnsi);
545 }
546 #ifdef CONFIG_BPF_JIT
547 struct bpf_trampoline *bpf_trampoline_lookup(u64 key);
548 int bpf_trampoline_link_prog(struct bpf_prog *prog);
549 int bpf_trampoline_unlink_prog(struct bpf_prog *prog);
550 void bpf_trampoline_put(struct bpf_trampoline *tr);
551 #define BPF_DISPATCHER_INIT(_name) {				\
552 	.mutex = __MUTEX_INITIALIZER(_name.mutex),		\
553 	.func = &_name##_func,					\
554 	.progs = {},						\
555 	.num_progs = 0,						\
556 	.image = NULL,						\
557 	.image_off = 0,						\
558 	.ksym = {						\
559 		.name  = #_name,				\
560 		.lnode = LIST_HEAD_INIT(_name.ksym.lnode),	\
561 	},							\
562 }
563 
564 #define DEFINE_BPF_DISPATCHER(name)					\
565 	noinline unsigned int bpf_dispatcher_##name##_func(		\
566 		const void *ctx,					\
567 		const struct bpf_insn *insnsi,				\
568 		unsigned int (*bpf_func)(const void *,			\
569 					 const struct bpf_insn *))	\
570 	{								\
571 		return bpf_func(ctx, insnsi);				\
572 	}								\
573 	EXPORT_SYMBOL(bpf_dispatcher_##name##_func);			\
574 	struct bpf_dispatcher bpf_dispatcher_##name =			\
575 		BPF_DISPATCHER_INIT(bpf_dispatcher_##name);
576 #define DECLARE_BPF_DISPATCHER(name)					\
577 	unsigned int bpf_dispatcher_##name##_func(			\
578 		const void *ctx,					\
579 		const struct bpf_insn *insnsi,				\
580 		unsigned int (*bpf_func)(const void *,			\
581 					 const struct bpf_insn *));	\
582 	extern struct bpf_dispatcher bpf_dispatcher_##name;
583 #define BPF_DISPATCHER_FUNC(name) bpf_dispatcher_##name##_func
584 #define BPF_DISPATCHER_PTR(name) (&bpf_dispatcher_##name)
585 void bpf_dispatcher_change_prog(struct bpf_dispatcher *d, struct bpf_prog *from,
586 				struct bpf_prog *to);
587 /* Called only from JIT-enabled code, so there's no need for stubs. */
588 void *bpf_jit_alloc_exec_page(void);
589 void bpf_image_ksym_add(void *data, struct bpf_ksym *ksym);
590 void bpf_image_ksym_del(struct bpf_ksym *ksym);
591 void bpf_ksym_add(struct bpf_ksym *ksym);
592 void bpf_ksym_del(struct bpf_ksym *ksym);
593 #else
594 static inline struct bpf_trampoline *bpf_trampoline_lookup(u64 key)
595 {
596 	return NULL;
597 }
598 static inline int bpf_trampoline_link_prog(struct bpf_prog *prog)
599 {
600 	return -ENOTSUPP;
601 }
602 static inline int bpf_trampoline_unlink_prog(struct bpf_prog *prog)
603 {
604 	return -ENOTSUPP;
605 }
606 static inline void bpf_trampoline_put(struct bpf_trampoline *tr) {}
607 #define DEFINE_BPF_DISPATCHER(name)
608 #define DECLARE_BPF_DISPATCHER(name)
609 #define BPF_DISPATCHER_FUNC(name) bpf_dispatcher_nop_func
610 #define BPF_DISPATCHER_PTR(name) NULL
611 static inline void bpf_dispatcher_change_prog(struct bpf_dispatcher *d,
612 					      struct bpf_prog *from,
613 					      struct bpf_prog *to) {}
614 static inline bool is_bpf_image_address(unsigned long address)
615 {
616 	return false;
617 }
618 #endif
619 
620 struct bpf_func_info_aux {
621 	u16 linkage;
622 	bool unreliable;
623 };
624 
625 enum bpf_jit_poke_reason {
626 	BPF_POKE_REASON_TAIL_CALL,
627 };
628 
629 /* Descriptor of pokes pointing /into/ the JITed image. */
630 struct bpf_jit_poke_descriptor {
631 	void *ip;
632 	union {
633 		struct {
634 			struct bpf_map *map;
635 			u32 key;
636 		} tail_call;
637 	};
638 	bool ip_stable;
639 	u8 adj_off;
640 	u16 reason;
641 };
642 
643 struct bpf_prog_aux {
644 	atomic64_t refcnt;
645 	u32 used_map_cnt;
646 	u32 max_ctx_offset;
647 	u32 max_pkt_offset;
648 	u32 max_tp_access;
649 	u32 stack_depth;
650 	u32 id;
651 	u32 func_cnt; /* used by non-func prog as the number of func progs */
652 	u32 func_idx; /* 0 for non-func prog, the index in func array for func prog */
653 	u32 attach_btf_id; /* in-kernel BTF type id to attach to */
654 	struct bpf_prog *linked_prog;
655 	bool verifier_zext; /* Zero extensions has been inserted by verifier. */
656 	bool offload_requested;
657 	bool attach_btf_trace; /* true if attaching to BTF-enabled raw tp */
658 	bool func_proto_unreliable;
659 	enum bpf_tramp_prog_type trampoline_prog_type;
660 	struct bpf_trampoline *trampoline;
661 	struct hlist_node tramp_hlist;
662 	/* BTF_KIND_FUNC_PROTO for valid attach_btf_id */
663 	const struct btf_type *attach_func_proto;
664 	/* function name for valid attach_btf_id */
665 	const char *attach_func_name;
666 	struct bpf_prog **func;
667 	void *jit_data; /* JIT specific data. arch dependent */
668 	struct bpf_jit_poke_descriptor *poke_tab;
669 	u32 size_poke_tab;
670 	struct bpf_ksym ksym;
671 	const struct bpf_prog_ops *ops;
672 	struct bpf_map **used_maps;
673 	struct bpf_prog *prog;
674 	struct user_struct *user;
675 	u64 load_time; /* ns since boottime */
676 	struct bpf_map *cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE];
677 	char name[BPF_OBJ_NAME_LEN];
678 #ifdef CONFIG_SECURITY
679 	void *security;
680 #endif
681 	struct bpf_prog_offload *offload;
682 	struct btf *btf;
683 	struct bpf_func_info *func_info;
684 	struct bpf_func_info_aux *func_info_aux;
685 	/* bpf_line_info loaded from userspace.  linfo->insn_off
686 	 * has the xlated insn offset.
687 	 * Both the main and sub prog share the same linfo.
688 	 * The subprog can access its first linfo by
689 	 * using the linfo_idx.
690 	 */
691 	struct bpf_line_info *linfo;
692 	/* jited_linfo is the jited addr of the linfo.  It has a
693 	 * one to one mapping to linfo:
694 	 * jited_linfo[i] is the jited addr for the linfo[i]->insn_off.
695 	 * Both the main and sub prog share the same jited_linfo.
696 	 * The subprog can access its first jited_linfo by
697 	 * using the linfo_idx.
698 	 */
699 	void **jited_linfo;
700 	u32 func_info_cnt;
701 	u32 nr_linfo;
702 	/* subprog can use linfo_idx to access its first linfo and
703 	 * jited_linfo.
704 	 * main prog always has linfo_idx == 0
705 	 */
706 	u32 linfo_idx;
707 	u32 num_exentries;
708 	struct exception_table_entry *extable;
709 	struct bpf_prog_stats __percpu *stats;
710 	union {
711 		struct work_struct work;
712 		struct rcu_head	rcu;
713 	};
714 };
715 
716 struct bpf_array_aux {
717 	/* 'Ownership' of prog array is claimed by the first program that
718 	 * is going to use this map or by the first program which FD is
719 	 * stored in the map to make sure that all callers and callees have
720 	 * the same prog type and JITed flag.
721 	 */
722 	enum bpf_prog_type type;
723 	bool jited;
724 	/* Programs with direct jumps into programs part of this array. */
725 	struct list_head poke_progs;
726 	struct bpf_map *map;
727 	struct mutex poke_mutex;
728 	struct work_struct work;
729 };
730 
731 struct bpf_struct_ops_value;
732 struct btf_type;
733 struct btf_member;
734 
735 #define BPF_STRUCT_OPS_MAX_NR_MEMBERS 64
736 struct bpf_struct_ops {
737 	const struct bpf_verifier_ops *verifier_ops;
738 	int (*init)(struct btf *btf);
739 	int (*check_member)(const struct btf_type *t,
740 			    const struct btf_member *member);
741 	int (*init_member)(const struct btf_type *t,
742 			   const struct btf_member *member,
743 			   void *kdata, const void *udata);
744 	int (*reg)(void *kdata);
745 	void (*unreg)(void *kdata);
746 	const struct btf_type *type;
747 	const struct btf_type *value_type;
748 	const char *name;
749 	struct btf_func_model func_models[BPF_STRUCT_OPS_MAX_NR_MEMBERS];
750 	u32 type_id;
751 	u32 value_id;
752 };
753 
754 #if defined(CONFIG_BPF_JIT) && defined(CONFIG_BPF_SYSCALL)
755 #define BPF_MODULE_OWNER ((void *)((0xeB9FUL << 2) + POISON_POINTER_DELTA))
756 const struct bpf_struct_ops *bpf_struct_ops_find(u32 type_id);
757 void bpf_struct_ops_init(struct btf *btf, struct bpf_verifier_log *log);
758 bool bpf_struct_ops_get(const void *kdata);
759 void bpf_struct_ops_put(const void *kdata);
760 int bpf_struct_ops_map_sys_lookup_elem(struct bpf_map *map, void *key,
761 				       void *value);
762 static inline bool bpf_try_module_get(const void *data, struct module *owner)
763 {
764 	if (owner == BPF_MODULE_OWNER)
765 		return bpf_struct_ops_get(data);
766 	else
767 		return try_module_get(owner);
768 }
769 static inline void bpf_module_put(const void *data, struct module *owner)
770 {
771 	if (owner == BPF_MODULE_OWNER)
772 		bpf_struct_ops_put(data);
773 	else
774 		module_put(owner);
775 }
776 #else
777 static inline const struct bpf_struct_ops *bpf_struct_ops_find(u32 type_id)
778 {
779 	return NULL;
780 }
781 static inline void bpf_struct_ops_init(struct btf *btf,
782 				       struct bpf_verifier_log *log)
783 {
784 }
785 static inline bool bpf_try_module_get(const void *data, struct module *owner)
786 {
787 	return try_module_get(owner);
788 }
789 static inline void bpf_module_put(const void *data, struct module *owner)
790 {
791 	module_put(owner);
792 }
793 static inline int bpf_struct_ops_map_sys_lookup_elem(struct bpf_map *map,
794 						     void *key,
795 						     void *value)
796 {
797 	return -EINVAL;
798 }
799 #endif
800 
801 struct bpf_array {
802 	struct bpf_map map;
803 	u32 elem_size;
804 	u32 index_mask;
805 	struct bpf_array_aux *aux;
806 	union {
807 		char value[0] __aligned(8);
808 		void *ptrs[0] __aligned(8);
809 		void __percpu *pptrs[0] __aligned(8);
810 	};
811 };
812 
813 #define BPF_COMPLEXITY_LIMIT_INSNS      1000000 /* yes. 1M insns */
814 #define MAX_TAIL_CALL_CNT 32
815 
816 #define BPF_F_ACCESS_MASK	(BPF_F_RDONLY |		\
817 				 BPF_F_RDONLY_PROG |	\
818 				 BPF_F_WRONLY |		\
819 				 BPF_F_WRONLY_PROG)
820 
821 #define BPF_MAP_CAN_READ	BIT(0)
822 #define BPF_MAP_CAN_WRITE	BIT(1)
823 
824 static inline u32 bpf_map_flags_to_cap(struct bpf_map *map)
825 {
826 	u32 access_flags = map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG);
827 
828 	/* Combination of BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG is
829 	 * not possible.
830 	 */
831 	if (access_flags & BPF_F_RDONLY_PROG)
832 		return BPF_MAP_CAN_READ;
833 	else if (access_flags & BPF_F_WRONLY_PROG)
834 		return BPF_MAP_CAN_WRITE;
835 	else
836 		return BPF_MAP_CAN_READ | BPF_MAP_CAN_WRITE;
837 }
838 
839 static inline bool bpf_map_flags_access_ok(u32 access_flags)
840 {
841 	return (access_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) !=
842 	       (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG);
843 }
844 
845 struct bpf_event_entry {
846 	struct perf_event *event;
847 	struct file *perf_file;
848 	struct file *map_file;
849 	struct rcu_head rcu;
850 };
851 
852 bool bpf_prog_array_compatible(struct bpf_array *array, const struct bpf_prog *fp);
853 int bpf_prog_calc_tag(struct bpf_prog *fp);
854 const char *kernel_type_name(u32 btf_type_id);
855 
856 const struct bpf_func_proto *bpf_get_trace_printk_proto(void);
857 
858 typedef unsigned long (*bpf_ctx_copy_t)(void *dst, const void *src,
859 					unsigned long off, unsigned long len);
860 typedef u32 (*bpf_convert_ctx_access_t)(enum bpf_access_type type,
861 					const struct bpf_insn *src,
862 					struct bpf_insn *dst,
863 					struct bpf_prog *prog,
864 					u32 *target_size);
865 
866 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
867 		     void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy);
868 
869 /* an array of programs to be executed under rcu_lock.
870  *
871  * Typical usage:
872  * ret = BPF_PROG_RUN_ARRAY(&bpf_prog_array, ctx, BPF_PROG_RUN);
873  *
874  * the structure returned by bpf_prog_array_alloc() should be populated
875  * with program pointers and the last pointer must be NULL.
876  * The user has to keep refcnt on the program and make sure the program
877  * is removed from the array before bpf_prog_put().
878  * The 'struct bpf_prog_array *' should only be replaced with xchg()
879  * since other cpus are walking the array of pointers in parallel.
880  */
881 struct bpf_prog_array_item {
882 	struct bpf_prog *prog;
883 	struct bpf_cgroup_storage *cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE];
884 };
885 
886 struct bpf_prog_array {
887 	struct rcu_head rcu;
888 	struct bpf_prog_array_item items[];
889 };
890 
891 struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags);
892 void bpf_prog_array_free(struct bpf_prog_array *progs);
893 int bpf_prog_array_length(struct bpf_prog_array *progs);
894 bool bpf_prog_array_is_empty(struct bpf_prog_array *array);
895 int bpf_prog_array_copy_to_user(struct bpf_prog_array *progs,
896 				__u32 __user *prog_ids, u32 cnt);
897 
898 void bpf_prog_array_delete_safe(struct bpf_prog_array *progs,
899 				struct bpf_prog *old_prog);
900 int bpf_prog_array_copy_info(struct bpf_prog_array *array,
901 			     u32 *prog_ids, u32 request_cnt,
902 			     u32 *prog_cnt);
903 int bpf_prog_array_copy(struct bpf_prog_array *old_array,
904 			struct bpf_prog *exclude_prog,
905 			struct bpf_prog *include_prog,
906 			struct bpf_prog_array **new_array);
907 
908 #define __BPF_PROG_RUN_ARRAY(array, ctx, func, check_non_null)	\
909 	({						\
910 		struct bpf_prog_array_item *_item;	\
911 		struct bpf_prog *_prog;			\
912 		struct bpf_prog_array *_array;		\
913 		u32 _ret = 1;				\
914 		migrate_disable();			\
915 		rcu_read_lock();			\
916 		_array = rcu_dereference(array);	\
917 		if (unlikely(check_non_null && !_array))\
918 			goto _out;			\
919 		_item = &_array->items[0];		\
920 		while ((_prog = READ_ONCE(_item->prog))) {		\
921 			bpf_cgroup_storage_set(_item->cgroup_storage);	\
922 			_ret &= func(_prog, ctx);	\
923 			_item++;			\
924 		}					\
925 _out:							\
926 		rcu_read_unlock();			\
927 		migrate_enable();			\
928 		_ret;					\
929 	 })
930 
931 /* To be used by __cgroup_bpf_run_filter_skb for EGRESS BPF progs
932  * so BPF programs can request cwr for TCP packets.
933  *
934  * Current cgroup skb programs can only return 0 or 1 (0 to drop the
935  * packet. This macro changes the behavior so the low order bit
936  * indicates whether the packet should be dropped (0) or not (1)
937  * and the next bit is a congestion notification bit. This could be
938  * used by TCP to call tcp_enter_cwr()
939  *
940  * Hence, new allowed return values of CGROUP EGRESS BPF programs are:
941  *   0: drop packet
942  *   1: keep packet
943  *   2: drop packet and cn
944  *   3: keep packet and cn
945  *
946  * This macro then converts it to one of the NET_XMIT or an error
947  * code that is then interpreted as drop packet (and no cn):
948  *   0: NET_XMIT_SUCCESS  skb should be transmitted
949  *   1: NET_XMIT_DROP     skb should be dropped and cn
950  *   2: NET_XMIT_CN       skb should be transmitted and cn
951  *   3: -EPERM            skb should be dropped
952  */
953 #define BPF_PROG_CGROUP_INET_EGRESS_RUN_ARRAY(array, ctx, func)		\
954 	({						\
955 		struct bpf_prog_array_item *_item;	\
956 		struct bpf_prog *_prog;			\
957 		struct bpf_prog_array *_array;		\
958 		u32 ret;				\
959 		u32 _ret = 1;				\
960 		u32 _cn = 0;				\
961 		migrate_disable();			\
962 		rcu_read_lock();			\
963 		_array = rcu_dereference(array);	\
964 		_item = &_array->items[0];		\
965 		while ((_prog = READ_ONCE(_item->prog))) {		\
966 			bpf_cgroup_storage_set(_item->cgroup_storage);	\
967 			ret = func(_prog, ctx);		\
968 			_ret &= (ret & 1);		\
969 			_cn |= (ret & 2);		\
970 			_item++;			\
971 		}					\
972 		rcu_read_unlock();			\
973 		migrate_enable();			\
974 		if (_ret)				\
975 			_ret = (_cn ? NET_XMIT_CN : NET_XMIT_SUCCESS);	\
976 		else					\
977 			_ret = (_cn ? NET_XMIT_DROP : -EPERM);		\
978 		_ret;					\
979 	})
980 
981 #define BPF_PROG_RUN_ARRAY(array, ctx, func)		\
982 	__BPF_PROG_RUN_ARRAY(array, ctx, func, false)
983 
984 #define BPF_PROG_RUN_ARRAY_CHECK(array, ctx, func)	\
985 	__BPF_PROG_RUN_ARRAY(array, ctx, func, true)
986 
987 #ifdef CONFIG_BPF_SYSCALL
988 DECLARE_PER_CPU(int, bpf_prog_active);
989 
990 /*
991  * Block execution of BPF programs attached to instrumentation (perf,
992  * kprobes, tracepoints) to prevent deadlocks on map operations as any of
993  * these events can happen inside a region which holds a map bucket lock
994  * and can deadlock on it.
995  *
996  * Use the preemption safe inc/dec variants on RT because migrate disable
997  * is preemptible on RT and preemption in the middle of the RMW operation
998  * might lead to inconsistent state. Use the raw variants for non RT
999  * kernels as migrate_disable() maps to preempt_disable() so the slightly
1000  * more expensive save operation can be avoided.
1001  */
1002 static inline void bpf_disable_instrumentation(void)
1003 {
1004 	migrate_disable();
1005 	if (IS_ENABLED(CONFIG_PREEMPT_RT))
1006 		this_cpu_inc(bpf_prog_active);
1007 	else
1008 		__this_cpu_inc(bpf_prog_active);
1009 }
1010 
1011 static inline void bpf_enable_instrumentation(void)
1012 {
1013 	if (IS_ENABLED(CONFIG_PREEMPT_RT))
1014 		this_cpu_dec(bpf_prog_active);
1015 	else
1016 		__this_cpu_dec(bpf_prog_active);
1017 	migrate_enable();
1018 }
1019 
1020 extern const struct file_operations bpf_map_fops;
1021 extern const struct file_operations bpf_prog_fops;
1022 
1023 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1024 	extern const struct bpf_prog_ops _name ## _prog_ops; \
1025 	extern const struct bpf_verifier_ops _name ## _verifier_ops;
1026 #define BPF_MAP_TYPE(_id, _ops) \
1027 	extern const struct bpf_map_ops _ops;
1028 #include <linux/bpf_types.h>
1029 #undef BPF_PROG_TYPE
1030 #undef BPF_MAP_TYPE
1031 
1032 extern const struct bpf_prog_ops bpf_offload_prog_ops;
1033 extern const struct bpf_verifier_ops tc_cls_act_analyzer_ops;
1034 extern const struct bpf_verifier_ops xdp_analyzer_ops;
1035 
1036 struct bpf_prog *bpf_prog_get(u32 ufd);
1037 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1038 				       bool attach_drv);
1039 void bpf_prog_add(struct bpf_prog *prog, int i);
1040 void bpf_prog_sub(struct bpf_prog *prog, int i);
1041 void bpf_prog_inc(struct bpf_prog *prog);
1042 struct bpf_prog * __must_check bpf_prog_inc_not_zero(struct bpf_prog *prog);
1043 void bpf_prog_put(struct bpf_prog *prog);
1044 int __bpf_prog_charge(struct user_struct *user, u32 pages);
1045 void __bpf_prog_uncharge(struct user_struct *user, u32 pages);
1046 void __bpf_free_used_maps(struct bpf_prog_aux *aux,
1047 			  struct bpf_map **used_maps, u32 len);
1048 
1049 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock);
1050 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock);
1051 
1052 struct bpf_map *bpf_map_get(u32 ufd);
1053 struct bpf_map *bpf_map_get_with_uref(u32 ufd);
1054 struct bpf_map *__bpf_map_get(struct fd f);
1055 void bpf_map_inc(struct bpf_map *map);
1056 void bpf_map_inc_with_uref(struct bpf_map *map);
1057 struct bpf_map * __must_check bpf_map_inc_not_zero(struct bpf_map *map);
1058 void bpf_map_put_with_uref(struct bpf_map *map);
1059 void bpf_map_put(struct bpf_map *map);
1060 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages);
1061 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages);
1062 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size);
1063 void bpf_map_charge_finish(struct bpf_map_memory *mem);
1064 void bpf_map_charge_move(struct bpf_map_memory *dst,
1065 			 struct bpf_map_memory *src);
1066 void *bpf_map_area_alloc(u64 size, int numa_node);
1067 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node);
1068 void bpf_map_area_free(void *base);
1069 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr);
1070 int  generic_map_lookup_batch(struct bpf_map *map,
1071 			      const union bpf_attr *attr,
1072 			      union bpf_attr __user *uattr);
1073 int  generic_map_update_batch(struct bpf_map *map,
1074 			      const union bpf_attr *attr,
1075 			      union bpf_attr __user *uattr);
1076 int  generic_map_delete_batch(struct bpf_map *map,
1077 			      const union bpf_attr *attr,
1078 			      union bpf_attr __user *uattr);
1079 
1080 extern int sysctl_unprivileged_bpf_disabled;
1081 
1082 int bpf_map_new_fd(struct bpf_map *map, int flags);
1083 int bpf_prog_new_fd(struct bpf_prog *prog);
1084 
1085 struct bpf_link;
1086 
1087 struct bpf_link_ops {
1088 	void (*release)(struct bpf_link *link);
1089 	void (*dealloc)(struct bpf_link *link);
1090 };
1091 
1092 void bpf_link_init(struct bpf_link *link, const struct bpf_link_ops *ops,
1093 		   struct bpf_prog *prog);
1094 void bpf_link_inc(struct bpf_link *link);
1095 void bpf_link_put(struct bpf_link *link);
1096 int bpf_link_new_fd(struct bpf_link *link);
1097 struct file *bpf_link_new_file(struct bpf_link *link, int *reserved_fd);
1098 struct bpf_link *bpf_link_get_from_fd(u32 ufd);
1099 
1100 int bpf_obj_pin_user(u32 ufd, const char __user *pathname);
1101 int bpf_obj_get_user(const char __user *pathname, int flags);
1102 
1103 int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value);
1104 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value);
1105 int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
1106 			   u64 flags);
1107 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
1108 			    u64 flags);
1109 
1110 int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value);
1111 
1112 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
1113 				 void *key, void *value, u64 map_flags);
1114 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value);
1115 int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file,
1116 				void *key, void *value, u64 map_flags);
1117 int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value);
1118 
1119 int bpf_get_file_flag(int flags);
1120 int bpf_check_uarg_tail_zero(void __user *uaddr, size_t expected_size,
1121 			     size_t actual_size);
1122 
1123 /* memcpy that is used with 8-byte aligned pointers, power-of-8 size and
1124  * forced to use 'long' read/writes to try to atomically copy long counters.
1125  * Best-effort only.  No barriers here, since it _will_ race with concurrent
1126  * updates from BPF programs. Called from bpf syscall and mostly used with
1127  * size 8 or 16 bytes, so ask compiler to inline it.
1128  */
1129 static inline void bpf_long_memcpy(void *dst, const void *src, u32 size)
1130 {
1131 	const long *lsrc = src;
1132 	long *ldst = dst;
1133 
1134 	size /= sizeof(long);
1135 	while (size--)
1136 		*ldst++ = *lsrc++;
1137 }
1138 
1139 /* verify correctness of eBPF program */
1140 int bpf_check(struct bpf_prog **fp, union bpf_attr *attr,
1141 	      union bpf_attr __user *uattr);
1142 void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth);
1143 
1144 /* Map specifics */
1145 struct xdp_buff;
1146 struct sk_buff;
1147 
1148 struct bpf_dtab_netdev *__dev_map_lookup_elem(struct bpf_map *map, u32 key);
1149 struct bpf_dtab_netdev *__dev_map_hash_lookup_elem(struct bpf_map *map, u32 key);
1150 void __dev_flush(void);
1151 int dev_xdp_enqueue(struct net_device *dev, struct xdp_buff *xdp,
1152 		    struct net_device *dev_rx);
1153 int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp,
1154 		    struct net_device *dev_rx);
1155 int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, struct sk_buff *skb,
1156 			     struct bpf_prog *xdp_prog);
1157 
1158 struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key);
1159 void __cpu_map_flush(void);
1160 int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_buff *xdp,
1161 		    struct net_device *dev_rx);
1162 
1163 /* Return map's numa specified by userspace */
1164 static inline int bpf_map_attr_numa_node(const union bpf_attr *attr)
1165 {
1166 	return (attr->map_flags & BPF_F_NUMA_NODE) ?
1167 		attr->numa_node : NUMA_NO_NODE;
1168 }
1169 
1170 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type);
1171 int array_map_alloc_check(union bpf_attr *attr);
1172 
1173 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1174 			  union bpf_attr __user *uattr);
1175 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
1176 			  union bpf_attr __user *uattr);
1177 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
1178 			      const union bpf_attr *kattr,
1179 			      union bpf_attr __user *uattr);
1180 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1181 				     const union bpf_attr *kattr,
1182 				     union bpf_attr __user *uattr);
1183 bool btf_ctx_access(int off, int size, enum bpf_access_type type,
1184 		    const struct bpf_prog *prog,
1185 		    struct bpf_insn_access_aux *info);
1186 int btf_struct_access(struct bpf_verifier_log *log,
1187 		      const struct btf_type *t, int off, int size,
1188 		      enum bpf_access_type atype,
1189 		      u32 *next_btf_id);
1190 int btf_resolve_helper_id(struct bpf_verifier_log *log,
1191 			  const struct bpf_func_proto *fn, int);
1192 
1193 int btf_distill_func_proto(struct bpf_verifier_log *log,
1194 			   struct btf *btf,
1195 			   const struct btf_type *func_proto,
1196 			   const char *func_name,
1197 			   struct btf_func_model *m);
1198 
1199 struct bpf_reg_state;
1200 int btf_check_func_arg_match(struct bpf_verifier_env *env, int subprog,
1201 			     struct bpf_reg_state *regs);
1202 int btf_prepare_func_args(struct bpf_verifier_env *env, int subprog,
1203 			  struct bpf_reg_state *reg);
1204 int btf_check_type_match(struct bpf_verifier_env *env, struct bpf_prog *prog,
1205 			 struct btf *btf, const struct btf_type *t);
1206 
1207 struct bpf_prog *bpf_prog_by_id(u32 id);
1208 
1209 #else /* !CONFIG_BPF_SYSCALL */
1210 static inline struct bpf_prog *bpf_prog_get(u32 ufd)
1211 {
1212 	return ERR_PTR(-EOPNOTSUPP);
1213 }
1214 
1215 static inline struct bpf_prog *bpf_prog_get_type_dev(u32 ufd,
1216 						     enum bpf_prog_type type,
1217 						     bool attach_drv)
1218 {
1219 	return ERR_PTR(-EOPNOTSUPP);
1220 }
1221 
1222 static inline void bpf_prog_add(struct bpf_prog *prog, int i)
1223 {
1224 }
1225 
1226 static inline void bpf_prog_sub(struct bpf_prog *prog, int i)
1227 {
1228 }
1229 
1230 static inline void bpf_prog_put(struct bpf_prog *prog)
1231 {
1232 }
1233 
1234 static inline void bpf_prog_inc(struct bpf_prog *prog)
1235 {
1236 }
1237 
1238 static inline struct bpf_prog *__must_check
1239 bpf_prog_inc_not_zero(struct bpf_prog *prog)
1240 {
1241 	return ERR_PTR(-EOPNOTSUPP);
1242 }
1243 
1244 static inline int __bpf_prog_charge(struct user_struct *user, u32 pages)
1245 {
1246 	return 0;
1247 }
1248 
1249 static inline void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1250 {
1251 }
1252 
1253 static inline int bpf_obj_get_user(const char __user *pathname, int flags)
1254 {
1255 	return -EOPNOTSUPP;
1256 }
1257 
1258 static inline struct net_device  *__dev_map_lookup_elem(struct bpf_map *map,
1259 						       u32 key)
1260 {
1261 	return NULL;
1262 }
1263 
1264 static inline struct net_device  *__dev_map_hash_lookup_elem(struct bpf_map *map,
1265 							     u32 key)
1266 {
1267 	return NULL;
1268 }
1269 
1270 static inline void __dev_flush(void)
1271 {
1272 }
1273 
1274 struct xdp_buff;
1275 struct bpf_dtab_netdev;
1276 
1277 static inline
1278 int dev_xdp_enqueue(struct net_device *dev, struct xdp_buff *xdp,
1279 		    struct net_device *dev_rx)
1280 {
1281 	return 0;
1282 }
1283 
1284 static inline
1285 int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp,
1286 		    struct net_device *dev_rx)
1287 {
1288 	return 0;
1289 }
1290 
1291 struct sk_buff;
1292 
1293 static inline int dev_map_generic_redirect(struct bpf_dtab_netdev *dst,
1294 					   struct sk_buff *skb,
1295 					   struct bpf_prog *xdp_prog)
1296 {
1297 	return 0;
1298 }
1299 
1300 static inline
1301 struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key)
1302 {
1303 	return NULL;
1304 }
1305 
1306 static inline void __cpu_map_flush(void)
1307 {
1308 }
1309 
1310 static inline int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu,
1311 				  struct xdp_buff *xdp,
1312 				  struct net_device *dev_rx)
1313 {
1314 	return 0;
1315 }
1316 
1317 static inline struct bpf_prog *bpf_prog_get_type_path(const char *name,
1318 				enum bpf_prog_type type)
1319 {
1320 	return ERR_PTR(-EOPNOTSUPP);
1321 }
1322 
1323 static inline int bpf_prog_test_run_xdp(struct bpf_prog *prog,
1324 					const union bpf_attr *kattr,
1325 					union bpf_attr __user *uattr)
1326 {
1327 	return -ENOTSUPP;
1328 }
1329 
1330 static inline int bpf_prog_test_run_skb(struct bpf_prog *prog,
1331 					const union bpf_attr *kattr,
1332 					union bpf_attr __user *uattr)
1333 {
1334 	return -ENOTSUPP;
1335 }
1336 
1337 static inline int bpf_prog_test_run_tracing(struct bpf_prog *prog,
1338 					    const union bpf_attr *kattr,
1339 					    union bpf_attr __user *uattr)
1340 {
1341 	return -ENOTSUPP;
1342 }
1343 
1344 static inline int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1345 						   const union bpf_attr *kattr,
1346 						   union bpf_attr __user *uattr)
1347 {
1348 	return -ENOTSUPP;
1349 }
1350 
1351 static inline void bpf_map_put(struct bpf_map *map)
1352 {
1353 }
1354 
1355 static inline struct bpf_prog *bpf_prog_by_id(u32 id)
1356 {
1357 	return ERR_PTR(-ENOTSUPP);
1358 }
1359 #endif /* CONFIG_BPF_SYSCALL */
1360 
1361 static inline struct bpf_prog *bpf_prog_get_type(u32 ufd,
1362 						 enum bpf_prog_type type)
1363 {
1364 	return bpf_prog_get_type_dev(ufd, type, false);
1365 }
1366 
1367 bool bpf_prog_get_ok(struct bpf_prog *, enum bpf_prog_type *, bool);
1368 
1369 int bpf_prog_offload_compile(struct bpf_prog *prog);
1370 void bpf_prog_offload_destroy(struct bpf_prog *prog);
1371 int bpf_prog_offload_info_fill(struct bpf_prog_info *info,
1372 			       struct bpf_prog *prog);
1373 
1374 int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map);
1375 
1376 int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value);
1377 int bpf_map_offload_update_elem(struct bpf_map *map,
1378 				void *key, void *value, u64 flags);
1379 int bpf_map_offload_delete_elem(struct bpf_map *map, void *key);
1380 int bpf_map_offload_get_next_key(struct bpf_map *map,
1381 				 void *key, void *next_key);
1382 
1383 bool bpf_offload_prog_map_match(struct bpf_prog *prog, struct bpf_map *map);
1384 
1385 struct bpf_offload_dev *
1386 bpf_offload_dev_create(const struct bpf_prog_offload_ops *ops, void *priv);
1387 void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev);
1388 void *bpf_offload_dev_priv(struct bpf_offload_dev *offdev);
1389 int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev,
1390 				    struct net_device *netdev);
1391 void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev,
1392 				       struct net_device *netdev);
1393 bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev);
1394 
1395 #if defined(CONFIG_NET) && defined(CONFIG_BPF_SYSCALL)
1396 int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr);
1397 
1398 static inline bool bpf_prog_is_dev_bound(const struct bpf_prog_aux *aux)
1399 {
1400 	return aux->offload_requested;
1401 }
1402 
1403 static inline bool bpf_map_is_dev_bound(struct bpf_map *map)
1404 {
1405 	return unlikely(map->ops == &bpf_map_offload_ops);
1406 }
1407 
1408 struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr);
1409 void bpf_map_offload_map_free(struct bpf_map *map);
1410 #else
1411 static inline int bpf_prog_offload_init(struct bpf_prog *prog,
1412 					union bpf_attr *attr)
1413 {
1414 	return -EOPNOTSUPP;
1415 }
1416 
1417 static inline bool bpf_prog_is_dev_bound(struct bpf_prog_aux *aux)
1418 {
1419 	return false;
1420 }
1421 
1422 static inline bool bpf_map_is_dev_bound(struct bpf_map *map)
1423 {
1424 	return false;
1425 }
1426 
1427 static inline struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr)
1428 {
1429 	return ERR_PTR(-EOPNOTSUPP);
1430 }
1431 
1432 static inline void bpf_map_offload_map_free(struct bpf_map *map)
1433 {
1434 }
1435 #endif /* CONFIG_NET && CONFIG_BPF_SYSCALL */
1436 
1437 #if defined(CONFIG_BPF_STREAM_PARSER)
1438 int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog, u32 which);
1439 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog);
1440 void sock_map_unhash(struct sock *sk);
1441 void sock_map_close(struct sock *sk, long timeout);
1442 #else
1443 static inline int sock_map_prog_update(struct bpf_map *map,
1444 				       struct bpf_prog *prog, u32 which)
1445 {
1446 	return -EOPNOTSUPP;
1447 }
1448 
1449 static inline int sock_map_get_from_fd(const union bpf_attr *attr,
1450 				       struct bpf_prog *prog)
1451 {
1452 	return -EINVAL;
1453 }
1454 #endif /* CONFIG_BPF_STREAM_PARSER */
1455 
1456 #if defined(CONFIG_INET) && defined(CONFIG_BPF_SYSCALL)
1457 void bpf_sk_reuseport_detach(struct sock *sk);
1458 int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map, void *key,
1459 				       void *value);
1460 int bpf_fd_reuseport_array_update_elem(struct bpf_map *map, void *key,
1461 				       void *value, u64 map_flags);
1462 #else
1463 static inline void bpf_sk_reuseport_detach(struct sock *sk)
1464 {
1465 }
1466 
1467 #ifdef CONFIG_BPF_SYSCALL
1468 static inline int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map,
1469 						     void *key, void *value)
1470 {
1471 	return -EOPNOTSUPP;
1472 }
1473 
1474 static inline int bpf_fd_reuseport_array_update_elem(struct bpf_map *map,
1475 						     void *key, void *value,
1476 						     u64 map_flags)
1477 {
1478 	return -EOPNOTSUPP;
1479 }
1480 #endif /* CONFIG_BPF_SYSCALL */
1481 #endif /* defined(CONFIG_INET) && defined(CONFIG_BPF_SYSCALL) */
1482 
1483 /* verifier prototypes for helper functions called from eBPF programs */
1484 extern const struct bpf_func_proto bpf_map_lookup_elem_proto;
1485 extern const struct bpf_func_proto bpf_map_update_elem_proto;
1486 extern const struct bpf_func_proto bpf_map_delete_elem_proto;
1487 extern const struct bpf_func_proto bpf_map_push_elem_proto;
1488 extern const struct bpf_func_proto bpf_map_pop_elem_proto;
1489 extern const struct bpf_func_proto bpf_map_peek_elem_proto;
1490 
1491 extern const struct bpf_func_proto bpf_get_prandom_u32_proto;
1492 extern const struct bpf_func_proto bpf_get_smp_processor_id_proto;
1493 extern const struct bpf_func_proto bpf_get_numa_node_id_proto;
1494 extern const struct bpf_func_proto bpf_tail_call_proto;
1495 extern const struct bpf_func_proto bpf_ktime_get_ns_proto;
1496 extern const struct bpf_func_proto bpf_get_current_pid_tgid_proto;
1497 extern const struct bpf_func_proto bpf_get_current_uid_gid_proto;
1498 extern const struct bpf_func_proto bpf_get_current_comm_proto;
1499 extern const struct bpf_func_proto bpf_get_stackid_proto;
1500 extern const struct bpf_func_proto bpf_get_stack_proto;
1501 extern const struct bpf_func_proto bpf_sock_map_update_proto;
1502 extern const struct bpf_func_proto bpf_sock_hash_update_proto;
1503 extern const struct bpf_func_proto bpf_get_current_cgroup_id_proto;
1504 extern const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto;
1505 extern const struct bpf_func_proto bpf_msg_redirect_hash_proto;
1506 extern const struct bpf_func_proto bpf_msg_redirect_map_proto;
1507 extern const struct bpf_func_proto bpf_sk_redirect_hash_proto;
1508 extern const struct bpf_func_proto bpf_sk_redirect_map_proto;
1509 extern const struct bpf_func_proto bpf_spin_lock_proto;
1510 extern const struct bpf_func_proto bpf_spin_unlock_proto;
1511 extern const struct bpf_func_proto bpf_get_local_storage_proto;
1512 extern const struct bpf_func_proto bpf_strtol_proto;
1513 extern const struct bpf_func_proto bpf_strtoul_proto;
1514 extern const struct bpf_func_proto bpf_tcp_sock_proto;
1515 extern const struct bpf_func_proto bpf_jiffies64_proto;
1516 extern const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto;
1517 
1518 const struct bpf_func_proto *bpf_tracing_func_proto(
1519 	enum bpf_func_id func_id, const struct bpf_prog *prog);
1520 
1521 /* Shared helpers among cBPF and eBPF. */
1522 void bpf_user_rnd_init_once(void);
1523 u64 bpf_user_rnd_u32(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
1524 
1525 #if defined(CONFIG_NET)
1526 bool bpf_sock_common_is_valid_access(int off, int size,
1527 				     enum bpf_access_type type,
1528 				     struct bpf_insn_access_aux *info);
1529 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
1530 			      struct bpf_insn_access_aux *info);
1531 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
1532 				const struct bpf_insn *si,
1533 				struct bpf_insn *insn_buf,
1534 				struct bpf_prog *prog,
1535 				u32 *target_size);
1536 #else
1537 static inline bool bpf_sock_common_is_valid_access(int off, int size,
1538 						   enum bpf_access_type type,
1539 						   struct bpf_insn_access_aux *info)
1540 {
1541 	return false;
1542 }
1543 static inline bool bpf_sock_is_valid_access(int off, int size,
1544 					    enum bpf_access_type type,
1545 					    struct bpf_insn_access_aux *info)
1546 {
1547 	return false;
1548 }
1549 static inline u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
1550 					      const struct bpf_insn *si,
1551 					      struct bpf_insn *insn_buf,
1552 					      struct bpf_prog *prog,
1553 					      u32 *target_size)
1554 {
1555 	return 0;
1556 }
1557 #endif
1558 
1559 #ifdef CONFIG_INET
1560 struct sk_reuseport_kern {
1561 	struct sk_buff *skb;
1562 	struct sock *sk;
1563 	struct sock *selected_sk;
1564 	void *data_end;
1565 	u32 hash;
1566 	u32 reuseport_id;
1567 	bool bind_inany;
1568 };
1569 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
1570 				  struct bpf_insn_access_aux *info);
1571 
1572 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
1573 				    const struct bpf_insn *si,
1574 				    struct bpf_insn *insn_buf,
1575 				    struct bpf_prog *prog,
1576 				    u32 *target_size);
1577 
1578 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
1579 				  struct bpf_insn_access_aux *info);
1580 
1581 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
1582 				    const struct bpf_insn *si,
1583 				    struct bpf_insn *insn_buf,
1584 				    struct bpf_prog *prog,
1585 				    u32 *target_size);
1586 #else
1587 static inline bool bpf_tcp_sock_is_valid_access(int off, int size,
1588 						enum bpf_access_type type,
1589 						struct bpf_insn_access_aux *info)
1590 {
1591 	return false;
1592 }
1593 
1594 static inline u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
1595 						  const struct bpf_insn *si,
1596 						  struct bpf_insn *insn_buf,
1597 						  struct bpf_prog *prog,
1598 						  u32 *target_size)
1599 {
1600 	return 0;
1601 }
1602 static inline bool bpf_xdp_sock_is_valid_access(int off, int size,
1603 						enum bpf_access_type type,
1604 						struct bpf_insn_access_aux *info)
1605 {
1606 	return false;
1607 }
1608 
1609 static inline u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
1610 						  const struct bpf_insn *si,
1611 						  struct bpf_insn *insn_buf,
1612 						  struct bpf_prog *prog,
1613 						  u32 *target_size)
1614 {
1615 	return 0;
1616 }
1617 #endif /* CONFIG_INET */
1618 
1619 enum bpf_text_poke_type {
1620 	BPF_MOD_CALL,
1621 	BPF_MOD_JUMP,
1622 };
1623 
1624 int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t,
1625 		       void *addr1, void *addr2);
1626 
1627 #endif /* _LINUX_BPF_H */
1628