Linux kernel mirror (for testing) git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel os linux
at v5.12-rc6 2023 lines 66 kB view raw
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/refcount.h> 18#include <linux/mutex.h> 19#include <linux/module.h> 20#include <linux/kallsyms.h> 21#include <linux/capability.h> 22#include <linux/sched/mm.h> 23#include <linux/slab.h> 24#include <linux/percpu-refcount.h> 25 26struct bpf_verifier_env; 27struct bpf_verifier_log; 28struct perf_event; 29struct bpf_prog; 30struct bpf_prog_aux; 31struct bpf_map; 32struct sock; 33struct seq_file; 34struct btf; 35struct btf_type; 36struct exception_table_entry; 37struct seq_operations; 38struct bpf_iter_aux_info; 39struct bpf_local_storage; 40struct bpf_local_storage_map; 41struct kobject; 42struct mem_cgroup; 43 44extern struct idr btf_idr; 45extern spinlock_t btf_idr_lock; 46extern struct kobject *btf_kobj; 47 48typedef int (*bpf_iter_init_seq_priv_t)(void *private_data, 49 struct bpf_iter_aux_info *aux); 50typedef void (*bpf_iter_fini_seq_priv_t)(void *private_data); 51struct bpf_iter_seq_info { 52 const struct seq_operations *seq_ops; 53 bpf_iter_init_seq_priv_t init_seq_private; 54 bpf_iter_fini_seq_priv_t fini_seq_private; 55 u32 seq_priv_size; 56}; 57 58/* map is generic key/value storage optionally accesible by eBPF programs */ 59struct bpf_map_ops { 60 /* funcs callable from userspace (via syscall) */ 61 int (*map_alloc_check)(union bpf_attr *attr); 62 struct bpf_map *(*map_alloc)(union bpf_attr *attr); 63 void (*map_release)(struct bpf_map *map, struct file *map_file); 64 void (*map_free)(struct bpf_map *map); 65 int (*map_get_next_key)(struct bpf_map *map, void *key, void *next_key); 66 void (*map_release_uref)(struct bpf_map *map); 67 void *(*map_lookup_elem_sys_only)(struct bpf_map *map, void *key); 68 int (*map_lookup_batch)(struct bpf_map *map, const union bpf_attr *attr, 69 union bpf_attr __user *uattr); 70 int (*map_lookup_and_delete_batch)(struct bpf_map *map, 71 const union bpf_attr *attr, 72 union bpf_attr __user *uattr); 73 int (*map_update_batch)(struct bpf_map *map, const union bpf_attr *attr, 74 union bpf_attr __user *uattr); 75 int (*map_delete_batch)(struct bpf_map *map, const union bpf_attr *attr, 76 union bpf_attr __user *uattr); 77 78 /* funcs callable from userspace and from eBPF programs */ 79 void *(*map_lookup_elem)(struct bpf_map *map, void *key); 80 int (*map_update_elem)(struct bpf_map *map, void *key, void *value, u64 flags); 81 int (*map_delete_elem)(struct bpf_map *map, void *key); 82 int (*map_push_elem)(struct bpf_map *map, void *value, u64 flags); 83 int (*map_pop_elem)(struct bpf_map *map, void *value); 84 int (*map_peek_elem)(struct bpf_map *map, void *value); 85 86 /* funcs called by prog_array and perf_event_array map */ 87 void *(*map_fd_get_ptr)(struct bpf_map *map, struct file *map_file, 88 int fd); 89 void (*map_fd_put_ptr)(void *ptr); 90 int (*map_gen_lookup)(struct bpf_map *map, struct bpf_insn *insn_buf); 91 u32 (*map_fd_sys_lookup_elem)(void *ptr); 92 void (*map_seq_show_elem)(struct bpf_map *map, void *key, 93 struct seq_file *m); 94 int (*map_check_btf)(const struct bpf_map *map, 95 const struct btf *btf, 96 const struct btf_type *key_type, 97 const struct btf_type *value_type); 98 99 /* Prog poke tracking helpers. */ 100 int (*map_poke_track)(struct bpf_map *map, struct bpf_prog_aux *aux); 101 void (*map_poke_untrack)(struct bpf_map *map, struct bpf_prog_aux *aux); 102 void (*map_poke_run)(struct bpf_map *map, u32 key, struct bpf_prog *old, 103 struct bpf_prog *new); 104 105 /* Direct value access helpers. */ 106 int (*map_direct_value_addr)(const struct bpf_map *map, 107 u64 *imm, u32 off); 108 int (*map_direct_value_meta)(const struct bpf_map *map, 109 u64 imm, u32 *off); 110 int (*map_mmap)(struct bpf_map *map, struct vm_area_struct *vma); 111 __poll_t (*map_poll)(struct bpf_map *map, struct file *filp, 112 struct poll_table_struct *pts); 113 114 /* Functions called by bpf_local_storage maps */ 115 int (*map_local_storage_charge)(struct bpf_local_storage_map *smap, 116 void *owner, u32 size); 117 void (*map_local_storage_uncharge)(struct bpf_local_storage_map *smap, 118 void *owner, u32 size); 119 struct bpf_local_storage __rcu ** (*map_owner_storage_ptr)(void *owner); 120 121 /* map_meta_equal must be implemented for maps that can be 122 * used as an inner map. It is a runtime check to ensure 123 * an inner map can be inserted to an outer map. 124 * 125 * Some properties of the inner map has been used during the 126 * verification time. When inserting an inner map at the runtime, 127 * map_meta_equal has to ensure the inserting map has the same 128 * properties that the verifier has used earlier. 129 */ 130 bool (*map_meta_equal)(const struct bpf_map *meta0, 131 const struct bpf_map *meta1); 132 133 /* BTF name and id of struct allocated by map_alloc */ 134 const char * const map_btf_name; 135 int *map_btf_id; 136 137 /* bpf_iter info used to open a seq_file */ 138 const struct bpf_iter_seq_info *iter_seq_info; 139}; 140 141struct bpf_map { 142 /* The first two cachelines with read-mostly members of which some 143 * are also accessed in fast-path (e.g. ops, max_entries). 144 */ 145 const struct bpf_map_ops *ops ____cacheline_aligned; 146 struct bpf_map *inner_map_meta; 147#ifdef CONFIG_SECURITY 148 void *security; 149#endif 150 enum bpf_map_type map_type; 151 u32 key_size; 152 u32 value_size; 153 u32 max_entries; 154 u32 map_flags; 155 int spin_lock_off; /* >=0 valid offset, <0 error */ 156 u32 id; 157 int numa_node; 158 u32 btf_key_type_id; 159 u32 btf_value_type_id; 160 struct btf *btf; 161#ifdef CONFIG_MEMCG_KMEM 162 struct mem_cgroup *memcg; 163#endif 164 char name[BPF_OBJ_NAME_LEN]; 165 u32 btf_vmlinux_value_type_id; 166 bool bypass_spec_v1; 167 bool frozen; /* write-once; write-protected by freeze_mutex */ 168 /* 22 bytes hole */ 169 170 /* The 3rd and 4th cacheline with misc members to avoid false sharing 171 * particularly with refcounting. 172 */ 173 atomic64_t refcnt ____cacheline_aligned; 174 atomic64_t usercnt; 175 struct work_struct work; 176 struct mutex freeze_mutex; 177 u64 writecnt; /* writable mmap cnt; protected by freeze_mutex */ 178}; 179 180static inline bool map_value_has_spin_lock(const struct bpf_map *map) 181{ 182 return map->spin_lock_off >= 0; 183} 184 185static inline void check_and_init_map_lock(struct bpf_map *map, void *dst) 186{ 187 if (likely(!map_value_has_spin_lock(map))) 188 return; 189 *(struct bpf_spin_lock *)(dst + map->spin_lock_off) = 190 (struct bpf_spin_lock){}; 191} 192 193/* copy everything but bpf_spin_lock */ 194static inline void copy_map_value(struct bpf_map *map, void *dst, void *src) 195{ 196 if (unlikely(map_value_has_spin_lock(map))) { 197 u32 off = map->spin_lock_off; 198 199 memcpy(dst, src, off); 200 memcpy(dst + off + sizeof(struct bpf_spin_lock), 201 src + off + sizeof(struct bpf_spin_lock), 202 map->value_size - off - sizeof(struct bpf_spin_lock)); 203 } else { 204 memcpy(dst, src, map->value_size); 205 } 206} 207void copy_map_value_locked(struct bpf_map *map, void *dst, void *src, 208 bool lock_src); 209int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size); 210 211struct bpf_offload_dev; 212struct bpf_offloaded_map; 213 214struct bpf_map_dev_ops { 215 int (*map_get_next_key)(struct bpf_offloaded_map *map, 216 void *key, void *next_key); 217 int (*map_lookup_elem)(struct bpf_offloaded_map *map, 218 void *key, void *value); 219 int (*map_update_elem)(struct bpf_offloaded_map *map, 220 void *key, void *value, u64 flags); 221 int (*map_delete_elem)(struct bpf_offloaded_map *map, void *key); 222}; 223 224struct bpf_offloaded_map { 225 struct bpf_map map; 226 struct net_device *netdev; 227 const struct bpf_map_dev_ops *dev_ops; 228 void *dev_priv; 229 struct list_head offloads; 230}; 231 232static inline struct bpf_offloaded_map *map_to_offmap(struct bpf_map *map) 233{ 234 return container_of(map, struct bpf_offloaded_map, map); 235} 236 237static inline bool bpf_map_offload_neutral(const struct bpf_map *map) 238{ 239 return map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY; 240} 241 242static inline bool bpf_map_support_seq_show(const struct bpf_map *map) 243{ 244 return (map->btf_value_type_id || map->btf_vmlinux_value_type_id) && 245 map->ops->map_seq_show_elem; 246} 247 248int map_check_no_btf(const struct bpf_map *map, 249 const struct btf *btf, 250 const struct btf_type *key_type, 251 const struct btf_type *value_type); 252 253bool bpf_map_meta_equal(const struct bpf_map *meta0, 254 const struct bpf_map *meta1); 255 256extern const struct bpf_map_ops bpf_map_offload_ops; 257 258/* function argument constraints */ 259enum bpf_arg_type { 260 ARG_DONTCARE = 0, /* unused argument in helper function */ 261 262 /* the following constraints used to prototype 263 * bpf_map_lookup/update/delete_elem() functions 264 */ 265 ARG_CONST_MAP_PTR, /* const argument used as pointer to bpf_map */ 266 ARG_PTR_TO_MAP_KEY, /* pointer to stack used as map key */ 267 ARG_PTR_TO_MAP_VALUE, /* pointer to stack used as map value */ 268 ARG_PTR_TO_UNINIT_MAP_VALUE, /* pointer to valid memory used to store a map value */ 269 ARG_PTR_TO_MAP_VALUE_OR_NULL, /* pointer to stack used as map value or NULL */ 270 271 /* the following constraints used to prototype bpf_memcmp() and other 272 * functions that access data on eBPF program stack 273 */ 274 ARG_PTR_TO_MEM, /* pointer to valid memory (stack, packet, map value) */ 275 ARG_PTR_TO_MEM_OR_NULL, /* pointer to valid memory or NULL */ 276 ARG_PTR_TO_UNINIT_MEM, /* pointer to memory does not need to be initialized, 277 * helper function must fill all bytes or clear 278 * them in error case. 279 */ 280 281 ARG_CONST_SIZE, /* number of bytes accessed from memory */ 282 ARG_CONST_SIZE_OR_ZERO, /* number of bytes accessed from memory or 0 */ 283 284 ARG_PTR_TO_CTX, /* pointer to context */ 285 ARG_PTR_TO_CTX_OR_NULL, /* pointer to context or NULL */ 286 ARG_ANYTHING, /* any (initialized) argument is ok */ 287 ARG_PTR_TO_SPIN_LOCK, /* pointer to bpf_spin_lock */ 288 ARG_PTR_TO_SOCK_COMMON, /* pointer to sock_common */ 289 ARG_PTR_TO_INT, /* pointer to int */ 290 ARG_PTR_TO_LONG, /* pointer to long */ 291 ARG_PTR_TO_SOCKET, /* pointer to bpf_sock (fullsock) */ 292 ARG_PTR_TO_SOCKET_OR_NULL, /* pointer to bpf_sock (fullsock) or NULL */ 293 ARG_PTR_TO_BTF_ID, /* pointer to in-kernel struct */ 294 ARG_PTR_TO_ALLOC_MEM, /* pointer to dynamically allocated memory */ 295 ARG_PTR_TO_ALLOC_MEM_OR_NULL, /* pointer to dynamically allocated memory or NULL */ 296 ARG_CONST_ALLOC_SIZE_OR_ZERO, /* number of allocated bytes requested */ 297 ARG_PTR_TO_BTF_ID_SOCK_COMMON, /* pointer to in-kernel sock_common or bpf-mirrored bpf_sock */ 298 ARG_PTR_TO_PERCPU_BTF_ID, /* pointer to in-kernel percpu type */ 299 __BPF_ARG_TYPE_MAX, 300}; 301 302/* type of values returned from helper functions */ 303enum bpf_return_type { 304 RET_INTEGER, /* function returns integer */ 305 RET_VOID, /* function doesn't return anything */ 306 RET_PTR_TO_MAP_VALUE, /* returns a pointer to map elem value */ 307 RET_PTR_TO_MAP_VALUE_OR_NULL, /* returns a pointer to map elem value or NULL */ 308 RET_PTR_TO_SOCKET_OR_NULL, /* returns a pointer to a socket or NULL */ 309 RET_PTR_TO_TCP_SOCK_OR_NULL, /* returns a pointer to a tcp_sock or NULL */ 310 RET_PTR_TO_SOCK_COMMON_OR_NULL, /* returns a pointer to a sock_common or NULL */ 311 RET_PTR_TO_ALLOC_MEM_OR_NULL, /* returns a pointer to dynamically allocated memory or NULL */ 312 RET_PTR_TO_BTF_ID_OR_NULL, /* returns a pointer to a btf_id or NULL */ 313 RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL, /* returns a pointer to a valid memory or a btf_id or NULL */ 314 RET_PTR_TO_MEM_OR_BTF_ID, /* returns a pointer to a valid memory or a btf_id */ 315 RET_PTR_TO_BTF_ID, /* returns a pointer to a btf_id */ 316}; 317 318/* eBPF function prototype used by verifier to allow BPF_CALLs from eBPF programs 319 * to in-kernel helper functions and for adjusting imm32 field in BPF_CALL 320 * instructions after verifying 321 */ 322struct bpf_func_proto { 323 u64 (*func)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 324 bool gpl_only; 325 bool pkt_access; 326 enum bpf_return_type ret_type; 327 union { 328 struct { 329 enum bpf_arg_type arg1_type; 330 enum bpf_arg_type arg2_type; 331 enum bpf_arg_type arg3_type; 332 enum bpf_arg_type arg4_type; 333 enum bpf_arg_type arg5_type; 334 }; 335 enum bpf_arg_type arg_type[5]; 336 }; 337 union { 338 struct { 339 u32 *arg1_btf_id; 340 u32 *arg2_btf_id; 341 u32 *arg3_btf_id; 342 u32 *arg4_btf_id; 343 u32 *arg5_btf_id; 344 }; 345 u32 *arg_btf_id[5]; 346 }; 347 int *ret_btf_id; /* return value btf_id */ 348 bool (*allowed)(const struct bpf_prog *prog); 349}; 350 351/* bpf_context is intentionally undefined structure. Pointer to bpf_context is 352 * the first argument to eBPF programs. 353 * For socket filters: 'struct bpf_context *' == 'struct sk_buff *' 354 */ 355struct bpf_context; 356 357enum bpf_access_type { 358 BPF_READ = 1, 359 BPF_WRITE = 2 360}; 361 362/* types of values stored in eBPF registers */ 363/* Pointer types represent: 364 * pointer 365 * pointer + imm 366 * pointer + (u16) var 367 * pointer + (u16) var + imm 368 * if (range > 0) then [ptr, ptr + range - off) is safe to access 369 * if (id > 0) means that some 'var' was added 370 * if (off > 0) means that 'imm' was added 371 */ 372enum bpf_reg_type { 373 NOT_INIT = 0, /* nothing was written into register */ 374 SCALAR_VALUE, /* reg doesn't contain a valid pointer */ 375 PTR_TO_CTX, /* reg points to bpf_context */ 376 CONST_PTR_TO_MAP, /* reg points to struct bpf_map */ 377 PTR_TO_MAP_VALUE, /* reg points to map element value */ 378 PTR_TO_MAP_VALUE_OR_NULL,/* points to map elem value or NULL */ 379 PTR_TO_STACK, /* reg == frame_pointer + offset */ 380 PTR_TO_PACKET_META, /* skb->data - meta_len */ 381 PTR_TO_PACKET, /* reg points to skb->data */ 382 PTR_TO_PACKET_END, /* skb->data + headlen */ 383 PTR_TO_FLOW_KEYS, /* reg points to bpf_flow_keys */ 384 PTR_TO_SOCKET, /* reg points to struct bpf_sock */ 385 PTR_TO_SOCKET_OR_NULL, /* reg points to struct bpf_sock or NULL */ 386 PTR_TO_SOCK_COMMON, /* reg points to sock_common */ 387 PTR_TO_SOCK_COMMON_OR_NULL, /* reg points to sock_common or NULL */ 388 PTR_TO_TCP_SOCK, /* reg points to struct tcp_sock */ 389 PTR_TO_TCP_SOCK_OR_NULL, /* reg points to struct tcp_sock or NULL */ 390 PTR_TO_TP_BUFFER, /* reg points to a writable raw tp's buffer */ 391 PTR_TO_XDP_SOCK, /* reg points to struct xdp_sock */ 392 /* PTR_TO_BTF_ID points to a kernel struct that does not need 393 * to be null checked by the BPF program. This does not imply the 394 * pointer is _not_ null and in practice this can easily be a null 395 * pointer when reading pointer chains. The assumption is program 396 * context will handle null pointer dereference typically via fault 397 * handling. The verifier must keep this in mind and can make no 398 * assumptions about null or non-null when doing branch analysis. 399 * Further, when passed into helpers the helpers can not, without 400 * additional context, assume the value is non-null. 401 */ 402 PTR_TO_BTF_ID, 403 /* PTR_TO_BTF_ID_OR_NULL points to a kernel struct that has not 404 * been checked for null. Used primarily to inform the verifier 405 * an explicit null check is required for this struct. 406 */ 407 PTR_TO_BTF_ID_OR_NULL, 408 PTR_TO_MEM, /* reg points to valid memory region */ 409 PTR_TO_MEM_OR_NULL, /* reg points to valid memory region or NULL */ 410 PTR_TO_RDONLY_BUF, /* reg points to a readonly buffer */ 411 PTR_TO_RDONLY_BUF_OR_NULL, /* reg points to a readonly buffer or NULL */ 412 PTR_TO_RDWR_BUF, /* reg points to a read/write buffer */ 413 PTR_TO_RDWR_BUF_OR_NULL, /* reg points to a read/write buffer or NULL */ 414 PTR_TO_PERCPU_BTF_ID, /* reg points to a percpu kernel variable */ 415}; 416 417/* The information passed from prog-specific *_is_valid_access 418 * back to the verifier. 419 */ 420struct bpf_insn_access_aux { 421 enum bpf_reg_type reg_type; 422 union { 423 int ctx_field_size; 424 struct { 425 struct btf *btf; 426 u32 btf_id; 427 }; 428 }; 429 struct bpf_verifier_log *log; /* for verbose logs */ 430}; 431 432static inline void 433bpf_ctx_record_field_size(struct bpf_insn_access_aux *aux, u32 size) 434{ 435 aux->ctx_field_size = size; 436} 437 438struct bpf_prog_ops { 439 int (*test_run)(struct bpf_prog *prog, const union bpf_attr *kattr, 440 union bpf_attr __user *uattr); 441}; 442 443struct bpf_verifier_ops { 444 /* return eBPF function prototype for verification */ 445 const struct bpf_func_proto * 446 (*get_func_proto)(enum bpf_func_id func_id, 447 const struct bpf_prog *prog); 448 449 /* return true if 'size' wide access at offset 'off' within bpf_context 450 * with 'type' (read or write) is allowed 451 */ 452 bool (*is_valid_access)(int off, int size, enum bpf_access_type type, 453 const struct bpf_prog *prog, 454 struct bpf_insn_access_aux *info); 455 int (*gen_prologue)(struct bpf_insn *insn, bool direct_write, 456 const struct bpf_prog *prog); 457 int (*gen_ld_abs)(const struct bpf_insn *orig, 458 struct bpf_insn *insn_buf); 459 u32 (*convert_ctx_access)(enum bpf_access_type type, 460 const struct bpf_insn *src, 461 struct bpf_insn *dst, 462 struct bpf_prog *prog, u32 *target_size); 463 int (*btf_struct_access)(struct bpf_verifier_log *log, 464 const struct btf *btf, 465 const struct btf_type *t, int off, int size, 466 enum bpf_access_type atype, 467 u32 *next_btf_id); 468}; 469 470struct bpf_prog_offload_ops { 471 /* verifier basic callbacks */ 472 int (*insn_hook)(struct bpf_verifier_env *env, 473 int insn_idx, int prev_insn_idx); 474 int (*finalize)(struct bpf_verifier_env *env); 475 /* verifier optimization callbacks (called after .finalize) */ 476 int (*replace_insn)(struct bpf_verifier_env *env, u32 off, 477 struct bpf_insn *insn); 478 int (*remove_insns)(struct bpf_verifier_env *env, u32 off, u32 cnt); 479 /* program management callbacks */ 480 int (*prepare)(struct bpf_prog *prog); 481 int (*translate)(struct bpf_prog *prog); 482 void (*destroy)(struct bpf_prog *prog); 483}; 484 485struct bpf_prog_offload { 486 struct bpf_prog *prog; 487 struct net_device *netdev; 488 struct bpf_offload_dev *offdev; 489 void *dev_priv; 490 struct list_head offloads; 491 bool dev_state; 492 bool opt_failed; 493 void *jited_image; 494 u32 jited_len; 495}; 496 497enum bpf_cgroup_storage_type { 498 BPF_CGROUP_STORAGE_SHARED, 499 BPF_CGROUP_STORAGE_PERCPU, 500 __BPF_CGROUP_STORAGE_MAX 501}; 502 503#define MAX_BPF_CGROUP_STORAGE_TYPE __BPF_CGROUP_STORAGE_MAX 504 505/* The longest tracepoint has 12 args. 506 * See include/trace/bpf_probe.h 507 */ 508#define MAX_BPF_FUNC_ARGS 12 509 510struct btf_func_model { 511 u8 ret_size; 512 u8 nr_args; 513 u8 arg_size[MAX_BPF_FUNC_ARGS]; 514}; 515 516/* Restore arguments before returning from trampoline to let original function 517 * continue executing. This flag is used for fentry progs when there are no 518 * fexit progs. 519 */ 520#define BPF_TRAMP_F_RESTORE_REGS BIT(0) 521/* Call original function after fentry progs, but before fexit progs. 522 * Makes sense for fentry/fexit, normal calls and indirect calls. 523 */ 524#define BPF_TRAMP_F_CALL_ORIG BIT(1) 525/* Skip current frame and return to parent. Makes sense for fentry/fexit 526 * programs only. Should not be used with normal calls and indirect calls. 527 */ 528#define BPF_TRAMP_F_SKIP_FRAME BIT(2) 529 530/* Each call __bpf_prog_enter + call bpf_func + call __bpf_prog_exit is ~50 531 * bytes on x86. Pick a number to fit into BPF_IMAGE_SIZE / 2 532 */ 533#define BPF_MAX_TRAMP_PROGS 38 534 535struct bpf_tramp_progs { 536 struct bpf_prog *progs[BPF_MAX_TRAMP_PROGS]; 537 int nr_progs; 538}; 539 540/* Different use cases for BPF trampoline: 541 * 1. replace nop at the function entry (kprobe equivalent) 542 * flags = BPF_TRAMP_F_RESTORE_REGS 543 * fentry = a set of programs to run before returning from trampoline 544 * 545 * 2. replace nop at the function entry (kprobe + kretprobe equivalent) 546 * flags = BPF_TRAMP_F_CALL_ORIG | BPF_TRAMP_F_SKIP_FRAME 547 * orig_call = fentry_ip + MCOUNT_INSN_SIZE 548 * fentry = a set of program to run before calling original function 549 * fexit = a set of program to run after original function 550 * 551 * 3. replace direct call instruction anywhere in the function body 552 * or assign a function pointer for indirect call (like tcp_congestion_ops->cong_avoid) 553 * With flags = 0 554 * fentry = a set of programs to run before returning from trampoline 555 * With flags = BPF_TRAMP_F_CALL_ORIG 556 * orig_call = original callback addr or direct function addr 557 * fentry = a set of program to run before calling original function 558 * fexit = a set of program to run after original function 559 */ 560struct bpf_tramp_image; 561int arch_prepare_bpf_trampoline(struct bpf_tramp_image *tr, void *image, void *image_end, 562 const struct btf_func_model *m, u32 flags, 563 struct bpf_tramp_progs *tprogs, 564 void *orig_call); 565/* these two functions are called from generated trampoline */ 566u64 notrace __bpf_prog_enter(struct bpf_prog *prog); 567void notrace __bpf_prog_exit(struct bpf_prog *prog, u64 start); 568u64 notrace __bpf_prog_enter_sleepable(struct bpf_prog *prog); 569void notrace __bpf_prog_exit_sleepable(struct bpf_prog *prog, u64 start); 570void notrace __bpf_tramp_enter(struct bpf_tramp_image *tr); 571void notrace __bpf_tramp_exit(struct bpf_tramp_image *tr); 572 573struct bpf_ksym { 574 unsigned long start; 575 unsigned long end; 576 char name[KSYM_NAME_LEN]; 577 struct list_head lnode; 578 struct latch_tree_node tnode; 579 bool prog; 580}; 581 582enum bpf_tramp_prog_type { 583 BPF_TRAMP_FENTRY, 584 BPF_TRAMP_FEXIT, 585 BPF_TRAMP_MODIFY_RETURN, 586 BPF_TRAMP_MAX, 587 BPF_TRAMP_REPLACE, /* more than MAX */ 588}; 589 590struct bpf_tramp_image { 591 void *image; 592 struct bpf_ksym ksym; 593 struct percpu_ref pcref; 594 void *ip_after_call; 595 void *ip_epilogue; 596 union { 597 struct rcu_head rcu; 598 struct work_struct work; 599 }; 600}; 601 602struct bpf_trampoline { 603 /* hlist for trampoline_table */ 604 struct hlist_node hlist; 605 /* serializes access to fields of this trampoline */ 606 struct mutex mutex; 607 refcount_t refcnt; 608 u64 key; 609 struct { 610 struct btf_func_model model; 611 void *addr; 612 bool ftrace_managed; 613 } func; 614 /* if !NULL this is BPF_PROG_TYPE_EXT program that extends another BPF 615 * program by replacing one of its functions. func.addr is the address 616 * of the function it replaced. 617 */ 618 struct bpf_prog *extension_prog; 619 /* list of BPF programs using this trampoline */ 620 struct hlist_head progs_hlist[BPF_TRAMP_MAX]; 621 /* Number of attached programs. A counter per kind. */ 622 int progs_cnt[BPF_TRAMP_MAX]; 623 /* Executable image of trampoline */ 624 struct bpf_tramp_image *cur_image; 625 u64 selector; 626}; 627 628struct bpf_attach_target_info { 629 struct btf_func_model fmodel; 630 long tgt_addr; 631 const char *tgt_name; 632 const struct btf_type *tgt_type; 633}; 634 635#define BPF_DISPATCHER_MAX 48 /* Fits in 2048B */ 636 637struct bpf_dispatcher_prog { 638 struct bpf_prog *prog; 639 refcount_t users; 640}; 641 642struct bpf_dispatcher { 643 /* dispatcher mutex */ 644 struct mutex mutex; 645 void *func; 646 struct bpf_dispatcher_prog progs[BPF_DISPATCHER_MAX]; 647 int num_progs; 648 void *image; 649 u32 image_off; 650 struct bpf_ksym ksym; 651}; 652 653static __always_inline unsigned int bpf_dispatcher_nop_func( 654 const void *ctx, 655 const struct bpf_insn *insnsi, 656 unsigned int (*bpf_func)(const void *, 657 const struct bpf_insn *)) 658{ 659 return bpf_func(ctx, insnsi); 660} 661#ifdef CONFIG_BPF_JIT 662int bpf_trampoline_link_prog(struct bpf_prog *prog, struct bpf_trampoline *tr); 663int bpf_trampoline_unlink_prog(struct bpf_prog *prog, struct bpf_trampoline *tr); 664struct bpf_trampoline *bpf_trampoline_get(u64 key, 665 struct bpf_attach_target_info *tgt_info); 666void bpf_trampoline_put(struct bpf_trampoline *tr); 667#define BPF_DISPATCHER_INIT(_name) { \ 668 .mutex = __MUTEX_INITIALIZER(_name.mutex), \ 669 .func = &_name##_func, \ 670 .progs = {}, \ 671 .num_progs = 0, \ 672 .image = NULL, \ 673 .image_off = 0, \ 674 .ksym = { \ 675 .name = #_name, \ 676 .lnode = LIST_HEAD_INIT(_name.ksym.lnode), \ 677 }, \ 678} 679 680#define DEFINE_BPF_DISPATCHER(name) \ 681 noinline unsigned int bpf_dispatcher_##name##_func( \ 682 const void *ctx, \ 683 const struct bpf_insn *insnsi, \ 684 unsigned int (*bpf_func)(const void *, \ 685 const struct bpf_insn *)) \ 686 { \ 687 return bpf_func(ctx, insnsi); \ 688 } \ 689 EXPORT_SYMBOL(bpf_dispatcher_##name##_func); \ 690 struct bpf_dispatcher bpf_dispatcher_##name = \ 691 BPF_DISPATCHER_INIT(bpf_dispatcher_##name); 692#define DECLARE_BPF_DISPATCHER(name) \ 693 unsigned int bpf_dispatcher_##name##_func( \ 694 const void *ctx, \ 695 const struct bpf_insn *insnsi, \ 696 unsigned int (*bpf_func)(const void *, \ 697 const struct bpf_insn *)); \ 698 extern struct bpf_dispatcher bpf_dispatcher_##name; 699#define BPF_DISPATCHER_FUNC(name) bpf_dispatcher_##name##_func 700#define BPF_DISPATCHER_PTR(name) (&bpf_dispatcher_##name) 701void bpf_dispatcher_change_prog(struct bpf_dispatcher *d, struct bpf_prog *from, 702 struct bpf_prog *to); 703/* Called only from JIT-enabled code, so there's no need for stubs. */ 704void *bpf_jit_alloc_exec_page(void); 705void bpf_image_ksym_add(void *data, struct bpf_ksym *ksym); 706void bpf_image_ksym_del(struct bpf_ksym *ksym); 707void bpf_ksym_add(struct bpf_ksym *ksym); 708void bpf_ksym_del(struct bpf_ksym *ksym); 709int bpf_jit_charge_modmem(u32 pages); 710void bpf_jit_uncharge_modmem(u32 pages); 711#else 712static inline int bpf_trampoline_link_prog(struct bpf_prog *prog, 713 struct bpf_trampoline *tr) 714{ 715 return -ENOTSUPP; 716} 717static inline int bpf_trampoline_unlink_prog(struct bpf_prog *prog, 718 struct bpf_trampoline *tr) 719{ 720 return -ENOTSUPP; 721} 722static inline struct bpf_trampoline *bpf_trampoline_get(u64 key, 723 struct bpf_attach_target_info *tgt_info) 724{ 725 return ERR_PTR(-EOPNOTSUPP); 726} 727static inline void bpf_trampoline_put(struct bpf_trampoline *tr) {} 728#define DEFINE_BPF_DISPATCHER(name) 729#define DECLARE_BPF_DISPATCHER(name) 730#define BPF_DISPATCHER_FUNC(name) bpf_dispatcher_nop_func 731#define BPF_DISPATCHER_PTR(name) NULL 732static inline void bpf_dispatcher_change_prog(struct bpf_dispatcher *d, 733 struct bpf_prog *from, 734 struct bpf_prog *to) {} 735static inline bool is_bpf_image_address(unsigned long address) 736{ 737 return false; 738} 739#endif 740 741struct bpf_func_info_aux { 742 u16 linkage; 743 bool unreliable; 744}; 745 746enum bpf_jit_poke_reason { 747 BPF_POKE_REASON_TAIL_CALL, 748}; 749 750/* Descriptor of pokes pointing /into/ the JITed image. */ 751struct bpf_jit_poke_descriptor { 752 void *tailcall_target; 753 void *tailcall_bypass; 754 void *bypass_addr; 755 union { 756 struct { 757 struct bpf_map *map; 758 u32 key; 759 } tail_call; 760 }; 761 bool tailcall_target_stable; 762 u8 adj_off; 763 u16 reason; 764 u32 insn_idx; 765}; 766 767/* reg_type info for ctx arguments */ 768struct bpf_ctx_arg_aux { 769 u32 offset; 770 enum bpf_reg_type reg_type; 771 u32 btf_id; 772}; 773 774struct btf_mod_pair { 775 struct btf *btf; 776 struct module *module; 777}; 778 779struct bpf_prog_aux { 780 atomic64_t refcnt; 781 u32 used_map_cnt; 782 u32 used_btf_cnt; 783 u32 max_ctx_offset; 784 u32 max_pkt_offset; 785 u32 max_tp_access; 786 u32 stack_depth; 787 u32 id; 788 u32 func_cnt; /* used by non-func prog as the number of func progs */ 789 u32 func_idx; /* 0 for non-func prog, the index in func array for func prog */ 790 u32 attach_btf_id; /* in-kernel BTF type id to attach to */ 791 u32 ctx_arg_info_size; 792 u32 max_rdonly_access; 793 u32 max_rdwr_access; 794 struct btf *attach_btf; 795 const struct bpf_ctx_arg_aux *ctx_arg_info; 796 struct mutex dst_mutex; /* protects dst_* pointers below, *after* prog becomes visible */ 797 struct bpf_prog *dst_prog; 798 struct bpf_trampoline *dst_trampoline; 799 enum bpf_prog_type saved_dst_prog_type; 800 enum bpf_attach_type saved_dst_attach_type; 801 bool verifier_zext; /* Zero extensions has been inserted by verifier. */ 802 bool offload_requested; 803 bool attach_btf_trace; /* true if attaching to BTF-enabled raw tp */ 804 bool func_proto_unreliable; 805 bool sleepable; 806 bool tail_call_reachable; 807 struct hlist_node tramp_hlist; 808 /* BTF_KIND_FUNC_PROTO for valid attach_btf_id */ 809 const struct btf_type *attach_func_proto; 810 /* function name for valid attach_btf_id */ 811 const char *attach_func_name; 812 struct bpf_prog **func; 813 void *jit_data; /* JIT specific data. arch dependent */ 814 struct bpf_jit_poke_descriptor *poke_tab; 815 u32 size_poke_tab; 816 struct bpf_ksym ksym; 817 const struct bpf_prog_ops *ops; 818 struct bpf_map **used_maps; 819 struct mutex used_maps_mutex; /* mutex for used_maps and used_map_cnt */ 820 struct btf_mod_pair *used_btfs; 821 struct bpf_prog *prog; 822 struct user_struct *user; 823 u64 load_time; /* ns since boottime */ 824 struct bpf_map *cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]; 825 char name[BPF_OBJ_NAME_LEN]; 826#ifdef CONFIG_SECURITY 827 void *security; 828#endif 829 struct bpf_prog_offload *offload; 830 struct btf *btf; 831 struct bpf_func_info *func_info; 832 struct bpf_func_info_aux *func_info_aux; 833 /* bpf_line_info loaded from userspace. linfo->insn_off 834 * has the xlated insn offset. 835 * Both the main and sub prog share the same linfo. 836 * The subprog can access its first linfo by 837 * using the linfo_idx. 838 */ 839 struct bpf_line_info *linfo; 840 /* jited_linfo is the jited addr of the linfo. It has a 841 * one to one mapping to linfo: 842 * jited_linfo[i] is the jited addr for the linfo[i]->insn_off. 843 * Both the main and sub prog share the same jited_linfo. 844 * The subprog can access its first jited_linfo by 845 * using the linfo_idx. 846 */ 847 void **jited_linfo; 848 u32 func_info_cnt; 849 u32 nr_linfo; 850 /* subprog can use linfo_idx to access its first linfo and 851 * jited_linfo. 852 * main prog always has linfo_idx == 0 853 */ 854 u32 linfo_idx; 855 u32 num_exentries; 856 struct exception_table_entry *extable; 857 union { 858 struct work_struct work; 859 struct rcu_head rcu; 860 }; 861}; 862 863struct bpf_array_aux { 864 /* 'Ownership' of prog array is claimed by the first program that 865 * is going to use this map or by the first program which FD is 866 * stored in the map to make sure that all callers and callees have 867 * the same prog type and JITed flag. 868 */ 869 enum bpf_prog_type type; 870 bool jited; 871 /* Programs with direct jumps into programs part of this array. */ 872 struct list_head poke_progs; 873 struct bpf_map *map; 874 struct mutex poke_mutex; 875 struct work_struct work; 876}; 877 878struct bpf_link { 879 atomic64_t refcnt; 880 u32 id; 881 enum bpf_link_type type; 882 const struct bpf_link_ops *ops; 883 struct bpf_prog *prog; 884 struct work_struct work; 885}; 886 887struct bpf_link_ops { 888 void (*release)(struct bpf_link *link); 889 void (*dealloc)(struct bpf_link *link); 890 int (*detach)(struct bpf_link *link); 891 int (*update_prog)(struct bpf_link *link, struct bpf_prog *new_prog, 892 struct bpf_prog *old_prog); 893 void (*show_fdinfo)(const struct bpf_link *link, struct seq_file *seq); 894 int (*fill_link_info)(const struct bpf_link *link, 895 struct bpf_link_info *info); 896}; 897 898struct bpf_link_primer { 899 struct bpf_link *link; 900 struct file *file; 901 int fd; 902 u32 id; 903}; 904 905struct bpf_struct_ops_value; 906struct btf_type; 907struct btf_member; 908 909#define BPF_STRUCT_OPS_MAX_NR_MEMBERS 64 910struct bpf_struct_ops { 911 const struct bpf_verifier_ops *verifier_ops; 912 int (*init)(struct btf *btf); 913 int (*check_member)(const struct btf_type *t, 914 const struct btf_member *member); 915 int (*init_member)(const struct btf_type *t, 916 const struct btf_member *member, 917 void *kdata, const void *udata); 918 int (*reg)(void *kdata); 919 void (*unreg)(void *kdata); 920 const struct btf_type *type; 921 const struct btf_type *value_type; 922 const char *name; 923 struct btf_func_model func_models[BPF_STRUCT_OPS_MAX_NR_MEMBERS]; 924 u32 type_id; 925 u32 value_id; 926}; 927 928#if defined(CONFIG_BPF_JIT) && defined(CONFIG_BPF_SYSCALL) 929#define BPF_MODULE_OWNER ((void *)((0xeB9FUL << 2) + POISON_POINTER_DELTA)) 930const struct bpf_struct_ops *bpf_struct_ops_find(u32 type_id); 931void bpf_struct_ops_init(struct btf *btf, struct bpf_verifier_log *log); 932bool bpf_struct_ops_get(const void *kdata); 933void bpf_struct_ops_put(const void *kdata); 934int bpf_struct_ops_map_sys_lookup_elem(struct bpf_map *map, void *key, 935 void *value); 936static inline bool bpf_try_module_get(const void *data, struct module *owner) 937{ 938 if (owner == BPF_MODULE_OWNER) 939 return bpf_struct_ops_get(data); 940 else 941 return try_module_get(owner); 942} 943static inline void bpf_module_put(const void *data, struct module *owner) 944{ 945 if (owner == BPF_MODULE_OWNER) 946 bpf_struct_ops_put(data); 947 else 948 module_put(owner); 949} 950#else 951static inline const struct bpf_struct_ops *bpf_struct_ops_find(u32 type_id) 952{ 953 return NULL; 954} 955static inline void bpf_struct_ops_init(struct btf *btf, 956 struct bpf_verifier_log *log) 957{ 958} 959static inline bool bpf_try_module_get(const void *data, struct module *owner) 960{ 961 return try_module_get(owner); 962} 963static inline void bpf_module_put(const void *data, struct module *owner) 964{ 965 module_put(owner); 966} 967static inline int bpf_struct_ops_map_sys_lookup_elem(struct bpf_map *map, 968 void *key, 969 void *value) 970{ 971 return -EINVAL; 972} 973#endif 974 975struct bpf_array { 976 struct bpf_map map; 977 u32 elem_size; 978 u32 index_mask; 979 struct bpf_array_aux *aux; 980 union { 981 char value[0] __aligned(8); 982 void *ptrs[0] __aligned(8); 983 void __percpu *pptrs[0] __aligned(8); 984 }; 985}; 986 987#define BPF_COMPLEXITY_LIMIT_INSNS 1000000 /* yes. 1M insns */ 988#define MAX_TAIL_CALL_CNT 32 989 990#define BPF_F_ACCESS_MASK (BPF_F_RDONLY | \ 991 BPF_F_RDONLY_PROG | \ 992 BPF_F_WRONLY | \ 993 BPF_F_WRONLY_PROG) 994 995#define BPF_MAP_CAN_READ BIT(0) 996#define BPF_MAP_CAN_WRITE BIT(1) 997 998static inline u32 bpf_map_flags_to_cap(struct bpf_map *map) 999{ 1000 u32 access_flags = map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG); 1001 1002 /* Combination of BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG is 1003 * not possible. 1004 */ 1005 if (access_flags & BPF_F_RDONLY_PROG) 1006 return BPF_MAP_CAN_READ; 1007 else if (access_flags & BPF_F_WRONLY_PROG) 1008 return BPF_MAP_CAN_WRITE; 1009 else 1010 return BPF_MAP_CAN_READ | BPF_MAP_CAN_WRITE; 1011} 1012 1013static inline bool bpf_map_flags_access_ok(u32 access_flags) 1014{ 1015 return (access_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) != 1016 (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG); 1017} 1018 1019struct bpf_event_entry { 1020 struct perf_event *event; 1021 struct file *perf_file; 1022 struct file *map_file; 1023 struct rcu_head rcu; 1024}; 1025 1026bool bpf_prog_array_compatible(struct bpf_array *array, const struct bpf_prog *fp); 1027int bpf_prog_calc_tag(struct bpf_prog *fp); 1028 1029const struct bpf_func_proto *bpf_get_trace_printk_proto(void); 1030 1031typedef unsigned long (*bpf_ctx_copy_t)(void *dst, const void *src, 1032 unsigned long off, unsigned long len); 1033typedef u32 (*bpf_convert_ctx_access_t)(enum bpf_access_type type, 1034 const struct bpf_insn *src, 1035 struct bpf_insn *dst, 1036 struct bpf_prog *prog, 1037 u32 *target_size); 1038 1039u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, 1040 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy); 1041 1042/* an array of programs to be executed under rcu_lock. 1043 * 1044 * Typical usage: 1045 * ret = BPF_PROG_RUN_ARRAY(&bpf_prog_array, ctx, BPF_PROG_RUN); 1046 * 1047 * the structure returned by bpf_prog_array_alloc() should be populated 1048 * with program pointers and the last pointer must be NULL. 1049 * The user has to keep refcnt on the program and make sure the program 1050 * is removed from the array before bpf_prog_put(). 1051 * The 'struct bpf_prog_array *' should only be replaced with xchg() 1052 * since other cpus are walking the array of pointers in parallel. 1053 */ 1054struct bpf_prog_array_item { 1055 struct bpf_prog *prog; 1056 struct bpf_cgroup_storage *cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]; 1057}; 1058 1059struct bpf_prog_array { 1060 struct rcu_head rcu; 1061 struct bpf_prog_array_item items[]; 1062}; 1063 1064struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags); 1065void bpf_prog_array_free(struct bpf_prog_array *progs); 1066int bpf_prog_array_length(struct bpf_prog_array *progs); 1067bool bpf_prog_array_is_empty(struct bpf_prog_array *array); 1068int bpf_prog_array_copy_to_user(struct bpf_prog_array *progs, 1069 __u32 __user *prog_ids, u32 cnt); 1070 1071void bpf_prog_array_delete_safe(struct bpf_prog_array *progs, 1072 struct bpf_prog *old_prog); 1073int bpf_prog_array_delete_safe_at(struct bpf_prog_array *array, int index); 1074int bpf_prog_array_update_at(struct bpf_prog_array *array, int index, 1075 struct bpf_prog *prog); 1076int bpf_prog_array_copy_info(struct bpf_prog_array *array, 1077 u32 *prog_ids, u32 request_cnt, 1078 u32 *prog_cnt); 1079int bpf_prog_array_copy(struct bpf_prog_array *old_array, 1080 struct bpf_prog *exclude_prog, 1081 struct bpf_prog *include_prog, 1082 struct bpf_prog_array **new_array); 1083 1084/* BPF program asks to bypass CAP_NET_BIND_SERVICE in bind. */ 1085#define BPF_RET_BIND_NO_CAP_NET_BIND_SERVICE (1 << 0) 1086/* BPF program asks to set CN on the packet. */ 1087#define BPF_RET_SET_CN (1 << 0) 1088 1089#define BPF_PROG_RUN_ARRAY_FLAGS(array, ctx, func, ret_flags) \ 1090 ({ \ 1091 struct bpf_prog_array_item *_item; \ 1092 struct bpf_prog *_prog; \ 1093 struct bpf_prog_array *_array; \ 1094 u32 _ret = 1; \ 1095 u32 func_ret; \ 1096 migrate_disable(); \ 1097 rcu_read_lock(); \ 1098 _array = rcu_dereference(array); \ 1099 _item = &_array->items[0]; \ 1100 while ((_prog = READ_ONCE(_item->prog))) { \ 1101 bpf_cgroup_storage_set(_item->cgroup_storage); \ 1102 func_ret = func(_prog, ctx); \ 1103 _ret &= (func_ret & 1); \ 1104 *(ret_flags) |= (func_ret >> 1); \ 1105 _item++; \ 1106 } \ 1107 rcu_read_unlock(); \ 1108 migrate_enable(); \ 1109 _ret; \ 1110 }) 1111 1112#define __BPF_PROG_RUN_ARRAY(array, ctx, func, check_non_null, set_cg_storage) \ 1113 ({ \ 1114 struct bpf_prog_array_item *_item; \ 1115 struct bpf_prog *_prog; \ 1116 struct bpf_prog_array *_array; \ 1117 u32 _ret = 1; \ 1118 migrate_disable(); \ 1119 rcu_read_lock(); \ 1120 _array = rcu_dereference(array); \ 1121 if (unlikely(check_non_null && !_array))\ 1122 goto _out; \ 1123 _item = &_array->items[0]; \ 1124 while ((_prog = READ_ONCE(_item->prog))) { \ 1125 if (set_cg_storage) \ 1126 bpf_cgroup_storage_set(_item->cgroup_storage); \ 1127 _ret &= func(_prog, ctx); \ 1128 _item++; \ 1129 } \ 1130_out: \ 1131 rcu_read_unlock(); \ 1132 migrate_enable(); \ 1133 _ret; \ 1134 }) 1135 1136/* To be used by __cgroup_bpf_run_filter_skb for EGRESS BPF progs 1137 * so BPF programs can request cwr for TCP packets. 1138 * 1139 * Current cgroup skb programs can only return 0 or 1 (0 to drop the 1140 * packet. This macro changes the behavior so the low order bit 1141 * indicates whether the packet should be dropped (0) or not (1) 1142 * and the next bit is a congestion notification bit. This could be 1143 * used by TCP to call tcp_enter_cwr() 1144 * 1145 * Hence, new allowed return values of CGROUP EGRESS BPF programs are: 1146 * 0: drop packet 1147 * 1: keep packet 1148 * 2: drop packet and cn 1149 * 3: keep packet and cn 1150 * 1151 * This macro then converts it to one of the NET_XMIT or an error 1152 * code that is then interpreted as drop packet (and no cn): 1153 * 0: NET_XMIT_SUCCESS skb should be transmitted 1154 * 1: NET_XMIT_DROP skb should be dropped and cn 1155 * 2: NET_XMIT_CN skb should be transmitted and cn 1156 * 3: -EPERM skb should be dropped 1157 */ 1158#define BPF_PROG_CGROUP_INET_EGRESS_RUN_ARRAY(array, ctx, func) \ 1159 ({ \ 1160 u32 _flags = 0; \ 1161 bool _cn; \ 1162 u32 _ret; \ 1163 _ret = BPF_PROG_RUN_ARRAY_FLAGS(array, ctx, func, &_flags); \ 1164 _cn = _flags & BPF_RET_SET_CN; \ 1165 if (_ret) \ 1166 _ret = (_cn ? NET_XMIT_CN : NET_XMIT_SUCCESS); \ 1167 else \ 1168 _ret = (_cn ? NET_XMIT_DROP : -EPERM); \ 1169 _ret; \ 1170 }) 1171 1172#define BPF_PROG_RUN_ARRAY(array, ctx, func) \ 1173 __BPF_PROG_RUN_ARRAY(array, ctx, func, false, true) 1174 1175#define BPF_PROG_RUN_ARRAY_CHECK(array, ctx, func) \ 1176 __BPF_PROG_RUN_ARRAY(array, ctx, func, true, false) 1177 1178#ifdef CONFIG_BPF_SYSCALL 1179DECLARE_PER_CPU(int, bpf_prog_active); 1180extern struct mutex bpf_stats_enabled_mutex; 1181 1182/* 1183 * Block execution of BPF programs attached to instrumentation (perf, 1184 * kprobes, tracepoints) to prevent deadlocks on map operations as any of 1185 * these events can happen inside a region which holds a map bucket lock 1186 * and can deadlock on it. 1187 * 1188 * Use the preemption safe inc/dec variants on RT because migrate disable 1189 * is preemptible on RT and preemption in the middle of the RMW operation 1190 * might lead to inconsistent state. Use the raw variants for non RT 1191 * kernels as migrate_disable() maps to preempt_disable() so the slightly 1192 * more expensive save operation can be avoided. 1193 */ 1194static inline void bpf_disable_instrumentation(void) 1195{ 1196 migrate_disable(); 1197 if (IS_ENABLED(CONFIG_PREEMPT_RT)) 1198 this_cpu_inc(bpf_prog_active); 1199 else 1200 __this_cpu_inc(bpf_prog_active); 1201} 1202 1203static inline void bpf_enable_instrumentation(void) 1204{ 1205 if (IS_ENABLED(CONFIG_PREEMPT_RT)) 1206 this_cpu_dec(bpf_prog_active); 1207 else 1208 __this_cpu_dec(bpf_prog_active); 1209 migrate_enable(); 1210} 1211 1212extern const struct file_operations bpf_map_fops; 1213extern const struct file_operations bpf_prog_fops; 1214extern const struct file_operations bpf_iter_fops; 1215 1216#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \ 1217 extern const struct bpf_prog_ops _name ## _prog_ops; \ 1218 extern const struct bpf_verifier_ops _name ## _verifier_ops; 1219#define BPF_MAP_TYPE(_id, _ops) \ 1220 extern const struct bpf_map_ops _ops; 1221#define BPF_LINK_TYPE(_id, _name) 1222#include <linux/bpf_types.h> 1223#undef BPF_PROG_TYPE 1224#undef BPF_MAP_TYPE 1225#undef BPF_LINK_TYPE 1226 1227extern const struct bpf_prog_ops bpf_offload_prog_ops; 1228extern const struct bpf_verifier_ops tc_cls_act_analyzer_ops; 1229extern const struct bpf_verifier_ops xdp_analyzer_ops; 1230 1231struct bpf_prog *bpf_prog_get(u32 ufd); 1232struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type, 1233 bool attach_drv); 1234void bpf_prog_add(struct bpf_prog *prog, int i); 1235void bpf_prog_sub(struct bpf_prog *prog, int i); 1236void bpf_prog_inc(struct bpf_prog *prog); 1237struct bpf_prog * __must_check bpf_prog_inc_not_zero(struct bpf_prog *prog); 1238void bpf_prog_put(struct bpf_prog *prog); 1239 1240void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock); 1241void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock); 1242 1243struct bpf_map *bpf_map_get(u32 ufd); 1244struct bpf_map *bpf_map_get_with_uref(u32 ufd); 1245struct bpf_map *__bpf_map_get(struct fd f); 1246void bpf_map_inc(struct bpf_map *map); 1247void bpf_map_inc_with_uref(struct bpf_map *map); 1248struct bpf_map * __must_check bpf_map_inc_not_zero(struct bpf_map *map); 1249void bpf_map_put_with_uref(struct bpf_map *map); 1250void bpf_map_put(struct bpf_map *map); 1251void *bpf_map_area_alloc(u64 size, int numa_node); 1252void *bpf_map_area_mmapable_alloc(u64 size, int numa_node); 1253void bpf_map_area_free(void *base); 1254void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr); 1255int generic_map_lookup_batch(struct bpf_map *map, 1256 const union bpf_attr *attr, 1257 union bpf_attr __user *uattr); 1258int generic_map_update_batch(struct bpf_map *map, 1259 const union bpf_attr *attr, 1260 union bpf_attr __user *uattr); 1261int generic_map_delete_batch(struct bpf_map *map, 1262 const union bpf_attr *attr, 1263 union bpf_attr __user *uattr); 1264struct bpf_map *bpf_map_get_curr_or_next(u32 *id); 1265struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id); 1266 1267#ifdef CONFIG_MEMCG_KMEM 1268void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags, 1269 int node); 1270void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags); 1271void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size, 1272 size_t align, gfp_t flags); 1273#else 1274static inline void * 1275bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags, 1276 int node) 1277{ 1278 return kmalloc_node(size, flags, node); 1279} 1280 1281static inline void * 1282bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags) 1283{ 1284 return kzalloc(size, flags); 1285} 1286 1287static inline void __percpu * 1288bpf_map_alloc_percpu(const struct bpf_map *map, size_t size, size_t align, 1289 gfp_t flags) 1290{ 1291 return __alloc_percpu_gfp(size, align, flags); 1292} 1293#endif 1294 1295extern int sysctl_unprivileged_bpf_disabled; 1296 1297static inline bool bpf_allow_ptr_leaks(void) 1298{ 1299 return perfmon_capable(); 1300} 1301 1302static inline bool bpf_allow_uninit_stack(void) 1303{ 1304 return perfmon_capable(); 1305} 1306 1307static inline bool bpf_allow_ptr_to_map_access(void) 1308{ 1309 return perfmon_capable(); 1310} 1311 1312static inline bool bpf_bypass_spec_v1(void) 1313{ 1314 return perfmon_capable(); 1315} 1316 1317static inline bool bpf_bypass_spec_v4(void) 1318{ 1319 return perfmon_capable(); 1320} 1321 1322int bpf_map_new_fd(struct bpf_map *map, int flags); 1323int bpf_prog_new_fd(struct bpf_prog *prog); 1324 1325void bpf_link_init(struct bpf_link *link, enum bpf_link_type type, 1326 const struct bpf_link_ops *ops, struct bpf_prog *prog); 1327int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer); 1328int bpf_link_settle(struct bpf_link_primer *primer); 1329void bpf_link_cleanup(struct bpf_link_primer *primer); 1330void bpf_link_inc(struct bpf_link *link); 1331void bpf_link_put(struct bpf_link *link); 1332int bpf_link_new_fd(struct bpf_link *link); 1333struct file *bpf_link_new_file(struct bpf_link *link, int *reserved_fd); 1334struct bpf_link *bpf_link_get_from_fd(u32 ufd); 1335 1336int bpf_obj_pin_user(u32 ufd, const char __user *pathname); 1337int bpf_obj_get_user(const char __user *pathname, int flags); 1338 1339#define BPF_ITER_FUNC_PREFIX "bpf_iter_" 1340#define DEFINE_BPF_ITER_FUNC(target, args...) \ 1341 extern int bpf_iter_ ## target(args); \ 1342 int __init bpf_iter_ ## target(args) { return 0; } 1343 1344struct bpf_iter_aux_info { 1345 struct bpf_map *map; 1346}; 1347 1348typedef int (*bpf_iter_attach_target_t)(struct bpf_prog *prog, 1349 union bpf_iter_link_info *linfo, 1350 struct bpf_iter_aux_info *aux); 1351typedef void (*bpf_iter_detach_target_t)(struct bpf_iter_aux_info *aux); 1352typedef void (*bpf_iter_show_fdinfo_t) (const struct bpf_iter_aux_info *aux, 1353 struct seq_file *seq); 1354typedef int (*bpf_iter_fill_link_info_t)(const struct bpf_iter_aux_info *aux, 1355 struct bpf_link_info *info); 1356 1357enum bpf_iter_feature { 1358 BPF_ITER_RESCHED = BIT(0), 1359}; 1360 1361#define BPF_ITER_CTX_ARG_MAX 2 1362struct bpf_iter_reg { 1363 const char *target; 1364 bpf_iter_attach_target_t attach_target; 1365 bpf_iter_detach_target_t detach_target; 1366 bpf_iter_show_fdinfo_t show_fdinfo; 1367 bpf_iter_fill_link_info_t fill_link_info; 1368 u32 ctx_arg_info_size; 1369 u32 feature; 1370 struct bpf_ctx_arg_aux ctx_arg_info[BPF_ITER_CTX_ARG_MAX]; 1371 const struct bpf_iter_seq_info *seq_info; 1372}; 1373 1374struct bpf_iter_meta { 1375 __bpf_md_ptr(struct seq_file *, seq); 1376 u64 session_id; 1377 u64 seq_num; 1378}; 1379 1380struct bpf_iter__bpf_map_elem { 1381 __bpf_md_ptr(struct bpf_iter_meta *, meta); 1382 __bpf_md_ptr(struct bpf_map *, map); 1383 __bpf_md_ptr(void *, key); 1384 __bpf_md_ptr(void *, value); 1385}; 1386 1387int bpf_iter_reg_target(const struct bpf_iter_reg *reg_info); 1388void bpf_iter_unreg_target(const struct bpf_iter_reg *reg_info); 1389bool bpf_iter_prog_supported(struct bpf_prog *prog); 1390int bpf_iter_link_attach(const union bpf_attr *attr, struct bpf_prog *prog); 1391int bpf_iter_new_fd(struct bpf_link *link); 1392bool bpf_link_is_iter(struct bpf_link *link); 1393struct bpf_prog *bpf_iter_get_info(struct bpf_iter_meta *meta, bool in_stop); 1394int bpf_iter_run_prog(struct bpf_prog *prog, void *ctx); 1395void bpf_iter_map_show_fdinfo(const struct bpf_iter_aux_info *aux, 1396 struct seq_file *seq); 1397int bpf_iter_map_fill_link_info(const struct bpf_iter_aux_info *aux, 1398 struct bpf_link_info *info); 1399 1400int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value); 1401int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value); 1402int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value, 1403 u64 flags); 1404int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value, 1405 u64 flags); 1406 1407int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value); 1408 1409int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file, 1410 void *key, void *value, u64 map_flags); 1411int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value); 1412int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file, 1413 void *key, void *value, u64 map_flags); 1414int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value); 1415 1416int bpf_get_file_flag(int flags); 1417int bpf_check_uarg_tail_zero(void __user *uaddr, size_t expected_size, 1418 size_t actual_size); 1419 1420/* memcpy that is used with 8-byte aligned pointers, power-of-8 size and 1421 * forced to use 'long' read/writes to try to atomically copy long counters. 1422 * Best-effort only. No barriers here, since it _will_ race with concurrent 1423 * updates from BPF programs. Called from bpf syscall and mostly used with 1424 * size 8 or 16 bytes, so ask compiler to inline it. 1425 */ 1426static inline void bpf_long_memcpy(void *dst, const void *src, u32 size) 1427{ 1428 const long *lsrc = src; 1429 long *ldst = dst; 1430 1431 size /= sizeof(long); 1432 while (size--) 1433 *ldst++ = *lsrc++; 1434} 1435 1436/* verify correctness of eBPF program */ 1437int bpf_check(struct bpf_prog **fp, union bpf_attr *attr, 1438 union bpf_attr __user *uattr); 1439 1440#ifndef CONFIG_BPF_JIT_ALWAYS_ON 1441void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth); 1442#endif 1443 1444struct btf *bpf_get_btf_vmlinux(void); 1445 1446/* Map specifics */ 1447struct xdp_buff; 1448struct sk_buff; 1449 1450struct bpf_dtab_netdev *__dev_map_lookup_elem(struct bpf_map *map, u32 key); 1451struct bpf_dtab_netdev *__dev_map_hash_lookup_elem(struct bpf_map *map, u32 key); 1452void __dev_flush(void); 1453int dev_xdp_enqueue(struct net_device *dev, struct xdp_buff *xdp, 1454 struct net_device *dev_rx); 1455int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp, 1456 struct net_device *dev_rx); 1457int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, struct sk_buff *skb, 1458 struct bpf_prog *xdp_prog); 1459bool dev_map_can_have_prog(struct bpf_map *map); 1460 1461struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key); 1462void __cpu_map_flush(void); 1463int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_buff *xdp, 1464 struct net_device *dev_rx); 1465bool cpu_map_prog_allowed(struct bpf_map *map); 1466 1467/* Return map's numa specified by userspace */ 1468static inline int bpf_map_attr_numa_node(const union bpf_attr *attr) 1469{ 1470 return (attr->map_flags & BPF_F_NUMA_NODE) ? 1471 attr->numa_node : NUMA_NO_NODE; 1472} 1473 1474struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type); 1475int array_map_alloc_check(union bpf_attr *attr); 1476 1477int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr, 1478 union bpf_attr __user *uattr); 1479int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr, 1480 union bpf_attr __user *uattr); 1481int bpf_prog_test_run_tracing(struct bpf_prog *prog, 1482 const union bpf_attr *kattr, 1483 union bpf_attr __user *uattr); 1484int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog, 1485 const union bpf_attr *kattr, 1486 union bpf_attr __user *uattr); 1487int bpf_prog_test_run_raw_tp(struct bpf_prog *prog, 1488 const union bpf_attr *kattr, 1489 union bpf_attr __user *uattr); 1490bool btf_ctx_access(int off, int size, enum bpf_access_type type, 1491 const struct bpf_prog *prog, 1492 struct bpf_insn_access_aux *info); 1493int btf_struct_access(struct bpf_verifier_log *log, const struct btf *btf, 1494 const struct btf_type *t, int off, int size, 1495 enum bpf_access_type atype, 1496 u32 *next_btf_id); 1497bool btf_struct_ids_match(struct bpf_verifier_log *log, 1498 const struct btf *btf, u32 id, int off, 1499 const struct btf *need_btf, u32 need_type_id); 1500 1501int btf_distill_func_proto(struct bpf_verifier_log *log, 1502 struct btf *btf, 1503 const struct btf_type *func_proto, 1504 const char *func_name, 1505 struct btf_func_model *m); 1506 1507struct bpf_reg_state; 1508int btf_check_func_arg_match(struct bpf_verifier_env *env, int subprog, 1509 struct bpf_reg_state *regs); 1510int btf_prepare_func_args(struct bpf_verifier_env *env, int subprog, 1511 struct bpf_reg_state *reg); 1512int btf_check_type_match(struct bpf_verifier_log *log, const struct bpf_prog *prog, 1513 struct btf *btf, const struct btf_type *t); 1514 1515struct bpf_prog *bpf_prog_by_id(u32 id); 1516struct bpf_link *bpf_link_by_id(u32 id); 1517 1518const struct bpf_func_proto *bpf_base_func_proto(enum bpf_func_id func_id); 1519#else /* !CONFIG_BPF_SYSCALL */ 1520static inline struct bpf_prog *bpf_prog_get(u32 ufd) 1521{ 1522 return ERR_PTR(-EOPNOTSUPP); 1523} 1524 1525static inline struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, 1526 enum bpf_prog_type type, 1527 bool attach_drv) 1528{ 1529 return ERR_PTR(-EOPNOTSUPP); 1530} 1531 1532static inline void bpf_prog_add(struct bpf_prog *prog, int i) 1533{ 1534} 1535 1536static inline void bpf_prog_sub(struct bpf_prog *prog, int i) 1537{ 1538} 1539 1540static inline void bpf_prog_put(struct bpf_prog *prog) 1541{ 1542} 1543 1544static inline void bpf_prog_inc(struct bpf_prog *prog) 1545{ 1546} 1547 1548static inline struct bpf_prog *__must_check 1549bpf_prog_inc_not_zero(struct bpf_prog *prog) 1550{ 1551 return ERR_PTR(-EOPNOTSUPP); 1552} 1553 1554static inline void bpf_link_init(struct bpf_link *link, enum bpf_link_type type, 1555 const struct bpf_link_ops *ops, 1556 struct bpf_prog *prog) 1557{ 1558} 1559 1560static inline int bpf_link_prime(struct bpf_link *link, 1561 struct bpf_link_primer *primer) 1562{ 1563 return -EOPNOTSUPP; 1564} 1565 1566static inline int bpf_link_settle(struct bpf_link_primer *primer) 1567{ 1568 return -EOPNOTSUPP; 1569} 1570 1571static inline void bpf_link_cleanup(struct bpf_link_primer *primer) 1572{ 1573} 1574 1575static inline void bpf_link_inc(struct bpf_link *link) 1576{ 1577} 1578 1579static inline void bpf_link_put(struct bpf_link *link) 1580{ 1581} 1582 1583static inline int bpf_obj_get_user(const char __user *pathname, int flags) 1584{ 1585 return -EOPNOTSUPP; 1586} 1587 1588static inline struct net_device *__dev_map_lookup_elem(struct bpf_map *map, 1589 u32 key) 1590{ 1591 return NULL; 1592} 1593 1594static inline struct net_device *__dev_map_hash_lookup_elem(struct bpf_map *map, 1595 u32 key) 1596{ 1597 return NULL; 1598} 1599static inline bool dev_map_can_have_prog(struct bpf_map *map) 1600{ 1601 return false; 1602} 1603 1604static inline void __dev_flush(void) 1605{ 1606} 1607 1608struct xdp_buff; 1609struct bpf_dtab_netdev; 1610 1611static inline 1612int dev_xdp_enqueue(struct net_device *dev, struct xdp_buff *xdp, 1613 struct net_device *dev_rx) 1614{ 1615 return 0; 1616} 1617 1618static inline 1619int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp, 1620 struct net_device *dev_rx) 1621{ 1622 return 0; 1623} 1624 1625struct sk_buff; 1626 1627static inline int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, 1628 struct sk_buff *skb, 1629 struct bpf_prog *xdp_prog) 1630{ 1631 return 0; 1632} 1633 1634static inline 1635struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key) 1636{ 1637 return NULL; 1638} 1639 1640static inline void __cpu_map_flush(void) 1641{ 1642} 1643 1644static inline int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, 1645 struct xdp_buff *xdp, 1646 struct net_device *dev_rx) 1647{ 1648 return 0; 1649} 1650 1651static inline bool cpu_map_prog_allowed(struct bpf_map *map) 1652{ 1653 return false; 1654} 1655 1656static inline struct bpf_prog *bpf_prog_get_type_path(const char *name, 1657 enum bpf_prog_type type) 1658{ 1659 return ERR_PTR(-EOPNOTSUPP); 1660} 1661 1662static inline int bpf_prog_test_run_xdp(struct bpf_prog *prog, 1663 const union bpf_attr *kattr, 1664 union bpf_attr __user *uattr) 1665{ 1666 return -ENOTSUPP; 1667} 1668 1669static inline int bpf_prog_test_run_skb(struct bpf_prog *prog, 1670 const union bpf_attr *kattr, 1671 union bpf_attr __user *uattr) 1672{ 1673 return -ENOTSUPP; 1674} 1675 1676static inline int bpf_prog_test_run_tracing(struct bpf_prog *prog, 1677 const union bpf_attr *kattr, 1678 union bpf_attr __user *uattr) 1679{ 1680 return -ENOTSUPP; 1681} 1682 1683static inline int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog, 1684 const union bpf_attr *kattr, 1685 union bpf_attr __user *uattr) 1686{ 1687 return -ENOTSUPP; 1688} 1689 1690static inline void bpf_map_put(struct bpf_map *map) 1691{ 1692} 1693 1694static inline struct bpf_prog *bpf_prog_by_id(u32 id) 1695{ 1696 return ERR_PTR(-ENOTSUPP); 1697} 1698 1699static inline const struct bpf_func_proto * 1700bpf_base_func_proto(enum bpf_func_id func_id) 1701{ 1702 return NULL; 1703} 1704#endif /* CONFIG_BPF_SYSCALL */ 1705 1706void __bpf_free_used_btfs(struct bpf_prog_aux *aux, 1707 struct btf_mod_pair *used_btfs, u32 len); 1708 1709static inline struct bpf_prog *bpf_prog_get_type(u32 ufd, 1710 enum bpf_prog_type type) 1711{ 1712 return bpf_prog_get_type_dev(ufd, type, false); 1713} 1714 1715void __bpf_free_used_maps(struct bpf_prog_aux *aux, 1716 struct bpf_map **used_maps, u32 len); 1717 1718bool bpf_prog_get_ok(struct bpf_prog *, enum bpf_prog_type *, bool); 1719 1720int bpf_prog_offload_compile(struct bpf_prog *prog); 1721void bpf_prog_offload_destroy(struct bpf_prog *prog); 1722int bpf_prog_offload_info_fill(struct bpf_prog_info *info, 1723 struct bpf_prog *prog); 1724 1725int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map); 1726 1727int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value); 1728int bpf_map_offload_update_elem(struct bpf_map *map, 1729 void *key, void *value, u64 flags); 1730int bpf_map_offload_delete_elem(struct bpf_map *map, void *key); 1731int bpf_map_offload_get_next_key(struct bpf_map *map, 1732 void *key, void *next_key); 1733 1734bool bpf_offload_prog_map_match(struct bpf_prog *prog, struct bpf_map *map); 1735 1736struct bpf_offload_dev * 1737bpf_offload_dev_create(const struct bpf_prog_offload_ops *ops, void *priv); 1738void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev); 1739void *bpf_offload_dev_priv(struct bpf_offload_dev *offdev); 1740int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev, 1741 struct net_device *netdev); 1742void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev, 1743 struct net_device *netdev); 1744bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev); 1745 1746#if defined(CONFIG_NET) && defined(CONFIG_BPF_SYSCALL) 1747int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr); 1748 1749static inline bool bpf_prog_is_dev_bound(const struct bpf_prog_aux *aux) 1750{ 1751 return aux->offload_requested; 1752} 1753 1754static inline bool bpf_map_is_dev_bound(struct bpf_map *map) 1755{ 1756 return unlikely(map->ops == &bpf_map_offload_ops); 1757} 1758 1759struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr); 1760void bpf_map_offload_map_free(struct bpf_map *map); 1761#else 1762static inline int bpf_prog_offload_init(struct bpf_prog *prog, 1763 union bpf_attr *attr) 1764{ 1765 return -EOPNOTSUPP; 1766} 1767 1768static inline bool bpf_prog_is_dev_bound(struct bpf_prog_aux *aux) 1769{ 1770 return false; 1771} 1772 1773static inline bool bpf_map_is_dev_bound(struct bpf_map *map) 1774{ 1775 return false; 1776} 1777 1778static inline struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr) 1779{ 1780 return ERR_PTR(-EOPNOTSUPP); 1781} 1782 1783static inline void bpf_map_offload_map_free(struct bpf_map *map) 1784{ 1785} 1786#endif /* CONFIG_NET && CONFIG_BPF_SYSCALL */ 1787 1788#if defined(CONFIG_BPF_STREAM_PARSER) 1789int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog, 1790 struct bpf_prog *old, u32 which); 1791int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog); 1792int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype); 1793int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value, u64 flags); 1794void sock_map_unhash(struct sock *sk); 1795void sock_map_close(struct sock *sk, long timeout); 1796#else 1797static inline int sock_map_prog_update(struct bpf_map *map, 1798 struct bpf_prog *prog, 1799 struct bpf_prog *old, u32 which) 1800{ 1801 return -EOPNOTSUPP; 1802} 1803 1804static inline int sock_map_get_from_fd(const union bpf_attr *attr, 1805 struct bpf_prog *prog) 1806{ 1807 return -EINVAL; 1808} 1809 1810static inline int sock_map_prog_detach(const union bpf_attr *attr, 1811 enum bpf_prog_type ptype) 1812{ 1813 return -EOPNOTSUPP; 1814} 1815 1816static inline int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value, 1817 u64 flags) 1818{ 1819 return -EOPNOTSUPP; 1820} 1821#endif /* CONFIG_BPF_STREAM_PARSER */ 1822 1823#if defined(CONFIG_INET) && defined(CONFIG_BPF_SYSCALL) 1824void bpf_sk_reuseport_detach(struct sock *sk); 1825int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map, void *key, 1826 void *value); 1827int bpf_fd_reuseport_array_update_elem(struct bpf_map *map, void *key, 1828 void *value, u64 map_flags); 1829#else 1830static inline void bpf_sk_reuseport_detach(struct sock *sk) 1831{ 1832} 1833 1834#ifdef CONFIG_BPF_SYSCALL 1835static inline int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map, 1836 void *key, void *value) 1837{ 1838 return -EOPNOTSUPP; 1839} 1840 1841static inline int bpf_fd_reuseport_array_update_elem(struct bpf_map *map, 1842 void *key, void *value, 1843 u64 map_flags) 1844{ 1845 return -EOPNOTSUPP; 1846} 1847#endif /* CONFIG_BPF_SYSCALL */ 1848#endif /* defined(CONFIG_INET) && defined(CONFIG_BPF_SYSCALL) */ 1849 1850/* verifier prototypes for helper functions called from eBPF programs */ 1851extern const struct bpf_func_proto bpf_map_lookup_elem_proto; 1852extern const struct bpf_func_proto bpf_map_update_elem_proto; 1853extern const struct bpf_func_proto bpf_map_delete_elem_proto; 1854extern const struct bpf_func_proto bpf_map_push_elem_proto; 1855extern const struct bpf_func_proto bpf_map_pop_elem_proto; 1856extern const struct bpf_func_proto bpf_map_peek_elem_proto; 1857 1858extern const struct bpf_func_proto bpf_get_prandom_u32_proto; 1859extern const struct bpf_func_proto bpf_get_smp_processor_id_proto; 1860extern const struct bpf_func_proto bpf_get_numa_node_id_proto; 1861extern const struct bpf_func_proto bpf_tail_call_proto; 1862extern const struct bpf_func_proto bpf_ktime_get_ns_proto; 1863extern const struct bpf_func_proto bpf_ktime_get_boot_ns_proto; 1864extern const struct bpf_func_proto bpf_get_current_pid_tgid_proto; 1865extern const struct bpf_func_proto bpf_get_current_uid_gid_proto; 1866extern const struct bpf_func_proto bpf_get_current_comm_proto; 1867extern const struct bpf_func_proto bpf_get_stackid_proto; 1868extern const struct bpf_func_proto bpf_get_stack_proto; 1869extern const struct bpf_func_proto bpf_get_task_stack_proto; 1870extern const struct bpf_func_proto bpf_get_stackid_proto_pe; 1871extern const struct bpf_func_proto bpf_get_stack_proto_pe; 1872extern const struct bpf_func_proto bpf_sock_map_update_proto; 1873extern const struct bpf_func_proto bpf_sock_hash_update_proto; 1874extern const struct bpf_func_proto bpf_get_current_cgroup_id_proto; 1875extern const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto; 1876extern const struct bpf_func_proto bpf_msg_redirect_hash_proto; 1877extern const struct bpf_func_proto bpf_msg_redirect_map_proto; 1878extern const struct bpf_func_proto bpf_sk_redirect_hash_proto; 1879extern const struct bpf_func_proto bpf_sk_redirect_map_proto; 1880extern const struct bpf_func_proto bpf_spin_lock_proto; 1881extern const struct bpf_func_proto bpf_spin_unlock_proto; 1882extern const struct bpf_func_proto bpf_get_local_storage_proto; 1883extern const struct bpf_func_proto bpf_strtol_proto; 1884extern const struct bpf_func_proto bpf_strtoul_proto; 1885extern const struct bpf_func_proto bpf_tcp_sock_proto; 1886extern const struct bpf_func_proto bpf_jiffies64_proto; 1887extern const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto; 1888extern const struct bpf_func_proto bpf_event_output_data_proto; 1889extern const struct bpf_func_proto bpf_ringbuf_output_proto; 1890extern const struct bpf_func_proto bpf_ringbuf_reserve_proto; 1891extern const struct bpf_func_proto bpf_ringbuf_submit_proto; 1892extern const struct bpf_func_proto bpf_ringbuf_discard_proto; 1893extern const struct bpf_func_proto bpf_ringbuf_query_proto; 1894extern const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto; 1895extern const struct bpf_func_proto bpf_skc_to_tcp_sock_proto; 1896extern const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto; 1897extern const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto; 1898extern const struct bpf_func_proto bpf_skc_to_udp6_sock_proto; 1899extern const struct bpf_func_proto bpf_copy_from_user_proto; 1900extern const struct bpf_func_proto bpf_snprintf_btf_proto; 1901extern const struct bpf_func_proto bpf_per_cpu_ptr_proto; 1902extern const struct bpf_func_proto bpf_this_cpu_ptr_proto; 1903extern const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto; 1904extern const struct bpf_func_proto bpf_sock_from_file_proto; 1905extern const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto; 1906 1907const struct bpf_func_proto *bpf_tracing_func_proto( 1908 enum bpf_func_id func_id, const struct bpf_prog *prog); 1909 1910const struct bpf_func_proto *tracing_prog_func_proto( 1911 enum bpf_func_id func_id, const struct bpf_prog *prog); 1912 1913/* Shared helpers among cBPF and eBPF. */ 1914void bpf_user_rnd_init_once(void); 1915u64 bpf_user_rnd_u32(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 1916u64 bpf_get_raw_cpu_id(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 1917 1918#if defined(CONFIG_NET) 1919bool bpf_sock_common_is_valid_access(int off, int size, 1920 enum bpf_access_type type, 1921 struct bpf_insn_access_aux *info); 1922bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 1923 struct bpf_insn_access_aux *info); 1924u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 1925 const struct bpf_insn *si, 1926 struct bpf_insn *insn_buf, 1927 struct bpf_prog *prog, 1928 u32 *target_size); 1929#else 1930static inline bool bpf_sock_common_is_valid_access(int off, int size, 1931 enum bpf_access_type type, 1932 struct bpf_insn_access_aux *info) 1933{ 1934 return false; 1935} 1936static inline bool bpf_sock_is_valid_access(int off, int size, 1937 enum bpf_access_type type, 1938 struct bpf_insn_access_aux *info) 1939{ 1940 return false; 1941} 1942static inline u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 1943 const struct bpf_insn *si, 1944 struct bpf_insn *insn_buf, 1945 struct bpf_prog *prog, 1946 u32 *target_size) 1947{ 1948 return 0; 1949} 1950#endif 1951 1952#ifdef CONFIG_INET 1953struct sk_reuseport_kern { 1954 struct sk_buff *skb; 1955 struct sock *sk; 1956 struct sock *selected_sk; 1957 void *data_end; 1958 u32 hash; 1959 u32 reuseport_id; 1960 bool bind_inany; 1961}; 1962bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 1963 struct bpf_insn_access_aux *info); 1964 1965u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 1966 const struct bpf_insn *si, 1967 struct bpf_insn *insn_buf, 1968 struct bpf_prog *prog, 1969 u32 *target_size); 1970 1971bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 1972 struct bpf_insn_access_aux *info); 1973 1974u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 1975 const struct bpf_insn *si, 1976 struct bpf_insn *insn_buf, 1977 struct bpf_prog *prog, 1978 u32 *target_size); 1979#else 1980static inline bool bpf_tcp_sock_is_valid_access(int off, int size, 1981 enum bpf_access_type type, 1982 struct bpf_insn_access_aux *info) 1983{ 1984 return false; 1985} 1986 1987static inline u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 1988 const struct bpf_insn *si, 1989 struct bpf_insn *insn_buf, 1990 struct bpf_prog *prog, 1991 u32 *target_size) 1992{ 1993 return 0; 1994} 1995static inline bool bpf_xdp_sock_is_valid_access(int off, int size, 1996 enum bpf_access_type type, 1997 struct bpf_insn_access_aux *info) 1998{ 1999 return false; 2000} 2001 2002static inline u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 2003 const struct bpf_insn *si, 2004 struct bpf_insn *insn_buf, 2005 struct bpf_prog *prog, 2006 u32 *target_size) 2007{ 2008 return 0; 2009} 2010#endif /* CONFIG_INET */ 2011 2012enum bpf_text_poke_type { 2013 BPF_MOD_CALL, 2014 BPF_MOD_JUMP, 2015}; 2016 2017int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, 2018 void *addr1, void *addr2); 2019 2020struct btf_id_set; 2021bool btf_id_set_contains(const struct btf_id_set *set, u32 id); 2022 2023#endif /* _LINUX_BPF_H */