at v5.12 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; 43struct module; 44 45extern struct idr btf_idr; 46extern spinlock_t btf_idr_lock; 47extern struct kobject *btf_kobj; 48 49typedef int (*bpf_iter_init_seq_priv_t)(void *private_data, 50 struct bpf_iter_aux_info *aux); 51typedef void (*bpf_iter_fini_seq_priv_t)(void *private_data); 52struct bpf_iter_seq_info { 53 const struct seq_operations *seq_ops; 54 bpf_iter_init_seq_priv_t init_seq_private; 55 bpf_iter_fini_seq_priv_t fini_seq_private; 56 u32 seq_priv_size; 57}; 58 59/* map is generic key/value storage optionally accesible by eBPF programs */ 60struct bpf_map_ops { 61 /* funcs callable from userspace (via syscall) */ 62 int (*map_alloc_check)(union bpf_attr *attr); 63 struct bpf_map *(*map_alloc)(union bpf_attr *attr); 64 void (*map_release)(struct bpf_map *map, struct file *map_file); 65 void (*map_free)(struct bpf_map *map); 66 int (*map_get_next_key)(struct bpf_map *map, void *key, void *next_key); 67 void (*map_release_uref)(struct bpf_map *map); 68 void *(*map_lookup_elem_sys_only)(struct bpf_map *map, void *key); 69 int (*map_lookup_batch)(struct bpf_map *map, const union bpf_attr *attr, 70 union bpf_attr __user *uattr); 71 int (*map_lookup_and_delete_batch)(struct bpf_map *map, 72 const union bpf_attr *attr, 73 union bpf_attr __user *uattr); 74 int (*map_update_batch)(struct bpf_map *map, const union bpf_attr *attr, 75 union bpf_attr __user *uattr); 76 int (*map_delete_batch)(struct bpf_map *map, const union bpf_attr *attr, 77 union bpf_attr __user *uattr); 78 79 /* funcs callable from userspace and from eBPF programs */ 80 void *(*map_lookup_elem)(struct bpf_map *map, void *key); 81 int (*map_update_elem)(struct bpf_map *map, void *key, void *value, u64 flags); 82 int (*map_delete_elem)(struct bpf_map *map, void *key); 83 int (*map_push_elem)(struct bpf_map *map, void *value, u64 flags); 84 int (*map_pop_elem)(struct bpf_map *map, void *value); 85 int (*map_peek_elem)(struct bpf_map *map, void *value); 86 87 /* funcs called by prog_array and perf_event_array map */ 88 void *(*map_fd_get_ptr)(struct bpf_map *map, struct file *map_file, 89 int fd); 90 void (*map_fd_put_ptr)(void *ptr); 91 int (*map_gen_lookup)(struct bpf_map *map, struct bpf_insn *insn_buf); 92 u32 (*map_fd_sys_lookup_elem)(void *ptr); 93 void (*map_seq_show_elem)(struct bpf_map *map, void *key, 94 struct seq_file *m); 95 int (*map_check_btf)(const struct bpf_map *map, 96 const struct btf *btf, 97 const struct btf_type *key_type, 98 const struct btf_type *value_type); 99 100 /* Prog poke tracking helpers. */ 101 int (*map_poke_track)(struct bpf_map *map, struct bpf_prog_aux *aux); 102 void (*map_poke_untrack)(struct bpf_map *map, struct bpf_prog_aux *aux); 103 void (*map_poke_run)(struct bpf_map *map, u32 key, struct bpf_prog *old, 104 struct bpf_prog *new); 105 106 /* Direct value access helpers. */ 107 int (*map_direct_value_addr)(const struct bpf_map *map, 108 u64 *imm, u32 off); 109 int (*map_direct_value_meta)(const struct bpf_map *map, 110 u64 imm, u32 *off); 111 int (*map_mmap)(struct bpf_map *map, struct vm_area_struct *vma); 112 __poll_t (*map_poll)(struct bpf_map *map, struct file *filp, 113 struct poll_table_struct *pts); 114 115 /* Functions called by bpf_local_storage maps */ 116 int (*map_local_storage_charge)(struct bpf_local_storage_map *smap, 117 void *owner, u32 size); 118 void (*map_local_storage_uncharge)(struct bpf_local_storage_map *smap, 119 void *owner, u32 size); 120 struct bpf_local_storage __rcu ** (*map_owner_storage_ptr)(void *owner); 121 122 /* map_meta_equal must be implemented for maps that can be 123 * used as an inner map. It is a runtime check to ensure 124 * an inner map can be inserted to an outer map. 125 * 126 * Some properties of the inner map has been used during the 127 * verification time. When inserting an inner map at the runtime, 128 * map_meta_equal has to ensure the inserting map has the same 129 * properties that the verifier has used earlier. 130 */ 131 bool (*map_meta_equal)(const struct bpf_map *meta0, 132 const struct bpf_map *meta1); 133 134 /* BTF name and id of struct allocated by map_alloc */ 135 const char * const map_btf_name; 136 int *map_btf_id; 137 138 /* bpf_iter info used to open a seq_file */ 139 const struct bpf_iter_seq_info *iter_seq_info; 140}; 141 142struct bpf_map { 143 /* The first two cachelines with read-mostly members of which some 144 * are also accessed in fast-path (e.g. ops, max_entries). 145 */ 146 const struct bpf_map_ops *ops ____cacheline_aligned; 147 struct bpf_map *inner_map_meta; 148#ifdef CONFIG_SECURITY 149 void *security; 150#endif 151 enum bpf_map_type map_type; 152 u32 key_size; 153 u32 value_size; 154 u32 max_entries; 155 u32 map_flags; 156 int spin_lock_off; /* >=0 valid offset, <0 error */ 157 u32 id; 158 int numa_node; 159 u32 btf_key_type_id; 160 u32 btf_value_type_id; 161 struct btf *btf; 162#ifdef CONFIG_MEMCG_KMEM 163 struct mem_cgroup *memcg; 164#endif 165 char name[BPF_OBJ_NAME_LEN]; 166 u32 btf_vmlinux_value_type_id; 167 bool bypass_spec_v1; 168 bool frozen; /* write-once; write-protected by freeze_mutex */ 169 /* 22 bytes hole */ 170 171 /* The 3rd and 4th cacheline with misc members to avoid false sharing 172 * particularly with refcounting. 173 */ 174 atomic64_t refcnt ____cacheline_aligned; 175 atomic64_t usercnt; 176 struct work_struct work; 177 struct mutex freeze_mutex; 178 u64 writecnt; /* writable mmap cnt; protected by freeze_mutex */ 179}; 180 181static inline bool map_value_has_spin_lock(const struct bpf_map *map) 182{ 183 return map->spin_lock_off >= 0; 184} 185 186static inline void check_and_init_map_lock(struct bpf_map *map, void *dst) 187{ 188 if (likely(!map_value_has_spin_lock(map))) 189 return; 190 *(struct bpf_spin_lock *)(dst + map->spin_lock_off) = 191 (struct bpf_spin_lock){}; 192} 193 194/* copy everything but bpf_spin_lock */ 195static inline void copy_map_value(struct bpf_map *map, void *dst, void *src) 196{ 197 if (unlikely(map_value_has_spin_lock(map))) { 198 u32 off = map->spin_lock_off; 199 200 memcpy(dst, src, off); 201 memcpy(dst + off + sizeof(struct bpf_spin_lock), 202 src + off + sizeof(struct bpf_spin_lock), 203 map->value_size - off - sizeof(struct bpf_spin_lock)); 204 } else { 205 memcpy(dst, src, map->value_size); 206 } 207} 208void copy_map_value_locked(struct bpf_map *map, void *dst, void *src, 209 bool lock_src); 210int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size); 211 212struct bpf_offload_dev; 213struct bpf_offloaded_map; 214 215struct bpf_map_dev_ops { 216 int (*map_get_next_key)(struct bpf_offloaded_map *map, 217 void *key, void *next_key); 218 int (*map_lookup_elem)(struct bpf_offloaded_map *map, 219 void *key, void *value); 220 int (*map_update_elem)(struct bpf_offloaded_map *map, 221 void *key, void *value, u64 flags); 222 int (*map_delete_elem)(struct bpf_offloaded_map *map, void *key); 223}; 224 225struct bpf_offloaded_map { 226 struct bpf_map map; 227 struct net_device *netdev; 228 const struct bpf_map_dev_ops *dev_ops; 229 void *dev_priv; 230 struct list_head offloads; 231}; 232 233static inline struct bpf_offloaded_map *map_to_offmap(struct bpf_map *map) 234{ 235 return container_of(map, struct bpf_offloaded_map, map); 236} 237 238static inline bool bpf_map_offload_neutral(const struct bpf_map *map) 239{ 240 return map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY; 241} 242 243static inline bool bpf_map_support_seq_show(const struct bpf_map *map) 244{ 245 return (map->btf_value_type_id || map->btf_vmlinux_value_type_id) && 246 map->ops->map_seq_show_elem; 247} 248 249int map_check_no_btf(const struct bpf_map *map, 250 const struct btf *btf, 251 const struct btf_type *key_type, 252 const struct btf_type *value_type); 253 254bool bpf_map_meta_equal(const struct bpf_map *meta0, 255 const struct bpf_map *meta1); 256 257extern const struct bpf_map_ops bpf_map_offload_ops; 258 259/* function argument constraints */ 260enum bpf_arg_type { 261 ARG_DONTCARE = 0, /* unused argument in helper function */ 262 263 /* the following constraints used to prototype 264 * bpf_map_lookup/update/delete_elem() functions 265 */ 266 ARG_CONST_MAP_PTR, /* const argument used as pointer to bpf_map */ 267 ARG_PTR_TO_MAP_KEY, /* pointer to stack used as map key */ 268 ARG_PTR_TO_MAP_VALUE, /* pointer to stack used as map value */ 269 ARG_PTR_TO_UNINIT_MAP_VALUE, /* pointer to valid memory used to store a map value */ 270 ARG_PTR_TO_MAP_VALUE_OR_NULL, /* pointer to stack used as map value or NULL */ 271 272 /* the following constraints used to prototype bpf_memcmp() and other 273 * functions that access data on eBPF program stack 274 */ 275 ARG_PTR_TO_MEM, /* pointer to valid memory (stack, packet, map value) */ 276 ARG_PTR_TO_MEM_OR_NULL, /* pointer to valid memory or NULL */ 277 ARG_PTR_TO_UNINIT_MEM, /* pointer to memory does not need to be initialized, 278 * helper function must fill all bytes or clear 279 * them in error case. 280 */ 281 282 ARG_CONST_SIZE, /* number of bytes accessed from memory */ 283 ARG_CONST_SIZE_OR_ZERO, /* number of bytes accessed from memory or 0 */ 284 285 ARG_PTR_TO_CTX, /* pointer to context */ 286 ARG_PTR_TO_CTX_OR_NULL, /* pointer to context or NULL */ 287 ARG_ANYTHING, /* any (initialized) argument is ok */ 288 ARG_PTR_TO_SPIN_LOCK, /* pointer to bpf_spin_lock */ 289 ARG_PTR_TO_SOCK_COMMON, /* pointer to sock_common */ 290 ARG_PTR_TO_INT, /* pointer to int */ 291 ARG_PTR_TO_LONG, /* pointer to long */ 292 ARG_PTR_TO_SOCKET, /* pointer to bpf_sock (fullsock) */ 293 ARG_PTR_TO_SOCKET_OR_NULL, /* pointer to bpf_sock (fullsock) or NULL */ 294 ARG_PTR_TO_BTF_ID, /* pointer to in-kernel struct */ 295 ARG_PTR_TO_ALLOC_MEM, /* pointer to dynamically allocated memory */ 296 ARG_PTR_TO_ALLOC_MEM_OR_NULL, /* pointer to dynamically allocated memory or NULL */ 297 ARG_CONST_ALLOC_SIZE_OR_ZERO, /* number of allocated bytes requested */ 298 ARG_PTR_TO_BTF_ID_SOCK_COMMON, /* pointer to in-kernel sock_common or bpf-mirrored bpf_sock */ 299 ARG_PTR_TO_PERCPU_BTF_ID, /* pointer to in-kernel percpu type */ 300 __BPF_ARG_TYPE_MAX, 301}; 302 303/* type of values returned from helper functions */ 304enum bpf_return_type { 305 RET_INTEGER, /* function returns integer */ 306 RET_VOID, /* function doesn't return anything */ 307 RET_PTR_TO_MAP_VALUE, /* returns a pointer to map elem value */ 308 RET_PTR_TO_MAP_VALUE_OR_NULL, /* returns a pointer to map elem value or NULL */ 309 RET_PTR_TO_SOCKET_OR_NULL, /* returns a pointer to a socket or NULL */ 310 RET_PTR_TO_TCP_SOCK_OR_NULL, /* returns a pointer to a tcp_sock or NULL */ 311 RET_PTR_TO_SOCK_COMMON_OR_NULL, /* returns a pointer to a sock_common or NULL */ 312 RET_PTR_TO_ALLOC_MEM_OR_NULL, /* returns a pointer to dynamically allocated memory or NULL */ 313 RET_PTR_TO_BTF_ID_OR_NULL, /* returns a pointer to a btf_id or NULL */ 314 RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL, /* returns a pointer to a valid memory or a btf_id or NULL */ 315 RET_PTR_TO_MEM_OR_BTF_ID, /* returns a pointer to a valid memory or a btf_id */ 316 RET_PTR_TO_BTF_ID, /* returns a pointer to a btf_id */ 317}; 318 319/* eBPF function prototype used by verifier to allow BPF_CALLs from eBPF programs 320 * to in-kernel helper functions and for adjusting imm32 field in BPF_CALL 321 * instructions after verifying 322 */ 323struct bpf_func_proto { 324 u64 (*func)(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 325 bool gpl_only; 326 bool pkt_access; 327 enum bpf_return_type ret_type; 328 union { 329 struct { 330 enum bpf_arg_type arg1_type; 331 enum bpf_arg_type arg2_type; 332 enum bpf_arg_type arg3_type; 333 enum bpf_arg_type arg4_type; 334 enum bpf_arg_type arg5_type; 335 }; 336 enum bpf_arg_type arg_type[5]; 337 }; 338 union { 339 struct { 340 u32 *arg1_btf_id; 341 u32 *arg2_btf_id; 342 u32 *arg3_btf_id; 343 u32 *arg4_btf_id; 344 u32 *arg5_btf_id; 345 }; 346 u32 *arg_btf_id[5]; 347 }; 348 int *ret_btf_id; /* return value btf_id */ 349 bool (*allowed)(const struct bpf_prog *prog); 350}; 351 352/* bpf_context is intentionally undefined structure. Pointer to bpf_context is 353 * the first argument to eBPF programs. 354 * For socket filters: 'struct bpf_context *' == 'struct sk_buff *' 355 */ 356struct bpf_context; 357 358enum bpf_access_type { 359 BPF_READ = 1, 360 BPF_WRITE = 2 361}; 362 363/* types of values stored in eBPF registers */ 364/* Pointer types represent: 365 * pointer 366 * pointer + imm 367 * pointer + (u16) var 368 * pointer + (u16) var + imm 369 * if (range > 0) then [ptr, ptr + range - off) is safe to access 370 * if (id > 0) means that some 'var' was added 371 * if (off > 0) means that 'imm' was added 372 */ 373enum bpf_reg_type { 374 NOT_INIT = 0, /* nothing was written into register */ 375 SCALAR_VALUE, /* reg doesn't contain a valid pointer */ 376 PTR_TO_CTX, /* reg points to bpf_context */ 377 CONST_PTR_TO_MAP, /* reg points to struct bpf_map */ 378 PTR_TO_MAP_VALUE, /* reg points to map element value */ 379 PTR_TO_MAP_VALUE_OR_NULL,/* points to map elem value or NULL */ 380 PTR_TO_STACK, /* reg == frame_pointer + offset */ 381 PTR_TO_PACKET_META, /* skb->data - meta_len */ 382 PTR_TO_PACKET, /* reg points to skb->data */ 383 PTR_TO_PACKET_END, /* skb->data + headlen */ 384 PTR_TO_FLOW_KEYS, /* reg points to bpf_flow_keys */ 385 PTR_TO_SOCKET, /* reg points to struct bpf_sock */ 386 PTR_TO_SOCKET_OR_NULL, /* reg points to struct bpf_sock or NULL */ 387 PTR_TO_SOCK_COMMON, /* reg points to sock_common */ 388 PTR_TO_SOCK_COMMON_OR_NULL, /* reg points to sock_common or NULL */ 389 PTR_TO_TCP_SOCK, /* reg points to struct tcp_sock */ 390 PTR_TO_TCP_SOCK_OR_NULL, /* reg points to struct tcp_sock or NULL */ 391 PTR_TO_TP_BUFFER, /* reg points to a writable raw tp's buffer */ 392 PTR_TO_XDP_SOCK, /* reg points to struct xdp_sock */ 393 /* PTR_TO_BTF_ID points to a kernel struct that does not need 394 * to be null checked by the BPF program. This does not imply the 395 * pointer is _not_ null and in practice this can easily be a null 396 * pointer when reading pointer chains. The assumption is program 397 * context will handle null pointer dereference typically via fault 398 * handling. The verifier must keep this in mind and can make no 399 * assumptions about null or non-null when doing branch analysis. 400 * Further, when passed into helpers the helpers can not, without 401 * additional context, assume the value is non-null. 402 */ 403 PTR_TO_BTF_ID, 404 /* PTR_TO_BTF_ID_OR_NULL points to a kernel struct that has not 405 * been checked for null. Used primarily to inform the verifier 406 * an explicit null check is required for this struct. 407 */ 408 PTR_TO_BTF_ID_OR_NULL, 409 PTR_TO_MEM, /* reg points to valid memory region */ 410 PTR_TO_MEM_OR_NULL, /* reg points to valid memory region or NULL */ 411 PTR_TO_RDONLY_BUF, /* reg points to a readonly buffer */ 412 PTR_TO_RDONLY_BUF_OR_NULL, /* reg points to a readonly buffer or NULL */ 413 PTR_TO_RDWR_BUF, /* reg points to a read/write buffer */ 414 PTR_TO_RDWR_BUF_OR_NULL, /* reg points to a read/write buffer or NULL */ 415 PTR_TO_PERCPU_BTF_ID, /* reg points to a percpu kernel variable */ 416}; 417 418/* The information passed from prog-specific *_is_valid_access 419 * back to the verifier. 420 */ 421struct bpf_insn_access_aux { 422 enum bpf_reg_type reg_type; 423 union { 424 int ctx_field_size; 425 struct { 426 struct btf *btf; 427 u32 btf_id; 428 }; 429 }; 430 struct bpf_verifier_log *log; /* for verbose logs */ 431}; 432 433static inline void 434bpf_ctx_record_field_size(struct bpf_insn_access_aux *aux, u32 size) 435{ 436 aux->ctx_field_size = size; 437} 438 439struct bpf_prog_ops { 440 int (*test_run)(struct bpf_prog *prog, const union bpf_attr *kattr, 441 union bpf_attr __user *uattr); 442}; 443 444struct bpf_verifier_ops { 445 /* return eBPF function prototype for verification */ 446 const struct bpf_func_proto * 447 (*get_func_proto)(enum bpf_func_id func_id, 448 const struct bpf_prog *prog); 449 450 /* return true if 'size' wide access at offset 'off' within bpf_context 451 * with 'type' (read or write) is allowed 452 */ 453 bool (*is_valid_access)(int off, int size, enum bpf_access_type type, 454 const struct bpf_prog *prog, 455 struct bpf_insn_access_aux *info); 456 int (*gen_prologue)(struct bpf_insn *insn, bool direct_write, 457 const struct bpf_prog *prog); 458 int (*gen_ld_abs)(const struct bpf_insn *orig, 459 struct bpf_insn *insn_buf); 460 u32 (*convert_ctx_access)(enum bpf_access_type type, 461 const struct bpf_insn *src, 462 struct bpf_insn *dst, 463 struct bpf_prog *prog, u32 *target_size); 464 int (*btf_struct_access)(struct bpf_verifier_log *log, 465 const struct btf *btf, 466 const struct btf_type *t, int off, int size, 467 enum bpf_access_type atype, 468 u32 *next_btf_id); 469}; 470 471struct bpf_prog_offload_ops { 472 /* verifier basic callbacks */ 473 int (*insn_hook)(struct bpf_verifier_env *env, 474 int insn_idx, int prev_insn_idx); 475 int (*finalize)(struct bpf_verifier_env *env); 476 /* verifier optimization callbacks (called after .finalize) */ 477 int (*replace_insn)(struct bpf_verifier_env *env, u32 off, 478 struct bpf_insn *insn); 479 int (*remove_insns)(struct bpf_verifier_env *env, u32 off, u32 cnt); 480 /* program management callbacks */ 481 int (*prepare)(struct bpf_prog *prog); 482 int (*translate)(struct bpf_prog *prog); 483 void (*destroy)(struct bpf_prog *prog); 484}; 485 486struct bpf_prog_offload { 487 struct bpf_prog *prog; 488 struct net_device *netdev; 489 struct bpf_offload_dev *offdev; 490 void *dev_priv; 491 struct list_head offloads; 492 bool dev_state; 493 bool opt_failed; 494 void *jited_image; 495 u32 jited_len; 496}; 497 498enum bpf_cgroup_storage_type { 499 BPF_CGROUP_STORAGE_SHARED, 500 BPF_CGROUP_STORAGE_PERCPU, 501 __BPF_CGROUP_STORAGE_MAX 502}; 503 504#define MAX_BPF_CGROUP_STORAGE_TYPE __BPF_CGROUP_STORAGE_MAX 505 506/* The longest tracepoint has 12 args. 507 * See include/trace/bpf_probe.h 508 */ 509#define MAX_BPF_FUNC_ARGS 12 510 511struct btf_func_model { 512 u8 ret_size; 513 u8 nr_args; 514 u8 arg_size[MAX_BPF_FUNC_ARGS]; 515}; 516 517/* Restore arguments before returning from trampoline to let original function 518 * continue executing. This flag is used for fentry progs when there are no 519 * fexit progs. 520 */ 521#define BPF_TRAMP_F_RESTORE_REGS BIT(0) 522/* Call original function after fentry progs, but before fexit progs. 523 * Makes sense for fentry/fexit, normal calls and indirect calls. 524 */ 525#define BPF_TRAMP_F_CALL_ORIG BIT(1) 526/* Skip current frame and return to parent. Makes sense for fentry/fexit 527 * programs only. Should not be used with normal calls and indirect calls. 528 */ 529#define BPF_TRAMP_F_SKIP_FRAME BIT(2) 530 531/* Each call __bpf_prog_enter + call bpf_func + call __bpf_prog_exit is ~50 532 * bytes on x86. Pick a number to fit into BPF_IMAGE_SIZE / 2 533 */ 534#define BPF_MAX_TRAMP_PROGS 38 535 536struct bpf_tramp_progs { 537 struct bpf_prog *progs[BPF_MAX_TRAMP_PROGS]; 538 int nr_progs; 539}; 540 541/* Different use cases for BPF trampoline: 542 * 1. replace nop at the function entry (kprobe equivalent) 543 * flags = BPF_TRAMP_F_RESTORE_REGS 544 * fentry = a set of programs to run before returning from trampoline 545 * 546 * 2. replace nop at the function entry (kprobe + kretprobe equivalent) 547 * flags = BPF_TRAMP_F_CALL_ORIG | BPF_TRAMP_F_SKIP_FRAME 548 * orig_call = fentry_ip + MCOUNT_INSN_SIZE 549 * fentry = a set of program to run before calling original function 550 * fexit = a set of program to run after original function 551 * 552 * 3. replace direct call instruction anywhere in the function body 553 * or assign a function pointer for indirect call (like tcp_congestion_ops->cong_avoid) 554 * With flags = 0 555 * fentry = a set of programs to run before returning from trampoline 556 * With flags = BPF_TRAMP_F_CALL_ORIG 557 * orig_call = original callback addr or direct function addr 558 * fentry = a set of program to run before calling original function 559 * fexit = a set of program to run after original function 560 */ 561struct bpf_tramp_image; 562int arch_prepare_bpf_trampoline(struct bpf_tramp_image *tr, void *image, void *image_end, 563 const struct btf_func_model *m, u32 flags, 564 struct bpf_tramp_progs *tprogs, 565 void *orig_call); 566/* these two functions are called from generated trampoline */ 567u64 notrace __bpf_prog_enter(struct bpf_prog *prog); 568void notrace __bpf_prog_exit(struct bpf_prog *prog, u64 start); 569u64 notrace __bpf_prog_enter_sleepable(struct bpf_prog *prog); 570void notrace __bpf_prog_exit_sleepable(struct bpf_prog *prog, u64 start); 571void notrace __bpf_tramp_enter(struct bpf_tramp_image *tr); 572void notrace __bpf_tramp_exit(struct bpf_tramp_image *tr); 573 574struct bpf_ksym { 575 unsigned long start; 576 unsigned long end; 577 char name[KSYM_NAME_LEN]; 578 struct list_head lnode; 579 struct latch_tree_node tnode; 580 bool prog; 581}; 582 583enum bpf_tramp_prog_type { 584 BPF_TRAMP_FENTRY, 585 BPF_TRAMP_FEXIT, 586 BPF_TRAMP_MODIFY_RETURN, 587 BPF_TRAMP_MAX, 588 BPF_TRAMP_REPLACE, /* more than MAX */ 589}; 590 591struct bpf_tramp_image { 592 void *image; 593 struct bpf_ksym ksym; 594 struct percpu_ref pcref; 595 void *ip_after_call; 596 void *ip_epilogue; 597 union { 598 struct rcu_head rcu; 599 struct work_struct work; 600 }; 601}; 602 603struct bpf_trampoline { 604 /* hlist for trampoline_table */ 605 struct hlist_node hlist; 606 /* serializes access to fields of this trampoline */ 607 struct mutex mutex; 608 refcount_t refcnt; 609 u64 key; 610 struct { 611 struct btf_func_model model; 612 void *addr; 613 bool ftrace_managed; 614 } func; 615 /* if !NULL this is BPF_PROG_TYPE_EXT program that extends another BPF 616 * program by replacing one of its functions. func.addr is the address 617 * of the function it replaced. 618 */ 619 struct bpf_prog *extension_prog; 620 /* list of BPF programs using this trampoline */ 621 struct hlist_head progs_hlist[BPF_TRAMP_MAX]; 622 /* Number of attached programs. A counter per kind. */ 623 int progs_cnt[BPF_TRAMP_MAX]; 624 /* Executable image of trampoline */ 625 struct bpf_tramp_image *cur_image; 626 u64 selector; 627 struct module *mod; 628}; 629 630struct bpf_attach_target_info { 631 struct btf_func_model fmodel; 632 long tgt_addr; 633 const char *tgt_name; 634 const struct btf_type *tgt_type; 635}; 636 637#define BPF_DISPATCHER_MAX 48 /* Fits in 2048B */ 638 639struct bpf_dispatcher_prog { 640 struct bpf_prog *prog; 641 refcount_t users; 642}; 643 644struct bpf_dispatcher { 645 /* dispatcher mutex */ 646 struct mutex mutex; 647 void *func; 648 struct bpf_dispatcher_prog progs[BPF_DISPATCHER_MAX]; 649 int num_progs; 650 void *image; 651 u32 image_off; 652 struct bpf_ksym ksym; 653}; 654 655static __always_inline unsigned int bpf_dispatcher_nop_func( 656 const void *ctx, 657 const struct bpf_insn *insnsi, 658 unsigned int (*bpf_func)(const void *, 659 const struct bpf_insn *)) 660{ 661 return bpf_func(ctx, insnsi); 662} 663#ifdef CONFIG_BPF_JIT 664int bpf_trampoline_link_prog(struct bpf_prog *prog, struct bpf_trampoline *tr); 665int bpf_trampoline_unlink_prog(struct bpf_prog *prog, struct bpf_trampoline *tr); 666struct bpf_trampoline *bpf_trampoline_get(u64 key, 667 struct bpf_attach_target_info *tgt_info); 668void bpf_trampoline_put(struct bpf_trampoline *tr); 669#define BPF_DISPATCHER_INIT(_name) { \ 670 .mutex = __MUTEX_INITIALIZER(_name.mutex), \ 671 .func = &_name##_func, \ 672 .progs = {}, \ 673 .num_progs = 0, \ 674 .image = NULL, \ 675 .image_off = 0, \ 676 .ksym = { \ 677 .name = #_name, \ 678 .lnode = LIST_HEAD_INIT(_name.ksym.lnode), \ 679 }, \ 680} 681 682#define DEFINE_BPF_DISPATCHER(name) \ 683 noinline unsigned int bpf_dispatcher_##name##_func( \ 684 const void *ctx, \ 685 const struct bpf_insn *insnsi, \ 686 unsigned int (*bpf_func)(const void *, \ 687 const struct bpf_insn *)) \ 688 { \ 689 return bpf_func(ctx, insnsi); \ 690 } \ 691 EXPORT_SYMBOL(bpf_dispatcher_##name##_func); \ 692 struct bpf_dispatcher bpf_dispatcher_##name = \ 693 BPF_DISPATCHER_INIT(bpf_dispatcher_##name); 694#define DECLARE_BPF_DISPATCHER(name) \ 695 unsigned int bpf_dispatcher_##name##_func( \ 696 const void *ctx, \ 697 const struct bpf_insn *insnsi, \ 698 unsigned int (*bpf_func)(const void *, \ 699 const struct bpf_insn *)); \ 700 extern struct bpf_dispatcher bpf_dispatcher_##name; 701#define BPF_DISPATCHER_FUNC(name) bpf_dispatcher_##name##_func 702#define BPF_DISPATCHER_PTR(name) (&bpf_dispatcher_##name) 703void bpf_dispatcher_change_prog(struct bpf_dispatcher *d, struct bpf_prog *from, 704 struct bpf_prog *to); 705/* Called only from JIT-enabled code, so there's no need for stubs. */ 706void *bpf_jit_alloc_exec_page(void); 707void bpf_image_ksym_add(void *data, struct bpf_ksym *ksym); 708void bpf_image_ksym_del(struct bpf_ksym *ksym); 709void bpf_ksym_add(struct bpf_ksym *ksym); 710void bpf_ksym_del(struct bpf_ksym *ksym); 711int bpf_jit_charge_modmem(u32 pages); 712void bpf_jit_uncharge_modmem(u32 pages); 713#else 714static inline int bpf_trampoline_link_prog(struct bpf_prog *prog, 715 struct bpf_trampoline *tr) 716{ 717 return -ENOTSUPP; 718} 719static inline int bpf_trampoline_unlink_prog(struct bpf_prog *prog, 720 struct bpf_trampoline *tr) 721{ 722 return -ENOTSUPP; 723} 724static inline struct bpf_trampoline *bpf_trampoline_get(u64 key, 725 struct bpf_attach_target_info *tgt_info) 726{ 727 return ERR_PTR(-EOPNOTSUPP); 728} 729static inline void bpf_trampoline_put(struct bpf_trampoline *tr) {} 730#define DEFINE_BPF_DISPATCHER(name) 731#define DECLARE_BPF_DISPATCHER(name) 732#define BPF_DISPATCHER_FUNC(name) bpf_dispatcher_nop_func 733#define BPF_DISPATCHER_PTR(name) NULL 734static inline void bpf_dispatcher_change_prog(struct bpf_dispatcher *d, 735 struct bpf_prog *from, 736 struct bpf_prog *to) {} 737static inline bool is_bpf_image_address(unsigned long address) 738{ 739 return false; 740} 741#endif 742 743struct bpf_func_info_aux { 744 u16 linkage; 745 bool unreliable; 746}; 747 748enum bpf_jit_poke_reason { 749 BPF_POKE_REASON_TAIL_CALL, 750}; 751 752/* Descriptor of pokes pointing /into/ the JITed image. */ 753struct bpf_jit_poke_descriptor { 754 void *tailcall_target; 755 void *tailcall_bypass; 756 void *bypass_addr; 757 union { 758 struct { 759 struct bpf_map *map; 760 u32 key; 761 } tail_call; 762 }; 763 bool tailcall_target_stable; 764 u8 adj_off; 765 u16 reason; 766 u32 insn_idx; 767}; 768 769/* reg_type info for ctx arguments */ 770struct bpf_ctx_arg_aux { 771 u32 offset; 772 enum bpf_reg_type reg_type; 773 u32 btf_id; 774}; 775 776struct btf_mod_pair { 777 struct btf *btf; 778 struct module *module; 779}; 780 781struct bpf_prog_aux { 782 atomic64_t refcnt; 783 u32 used_map_cnt; 784 u32 used_btf_cnt; 785 u32 max_ctx_offset; 786 u32 max_pkt_offset; 787 u32 max_tp_access; 788 u32 stack_depth; 789 u32 id; 790 u32 func_cnt; /* used by non-func prog as the number of func progs */ 791 u32 func_idx; /* 0 for non-func prog, the index in func array for func prog */ 792 u32 attach_btf_id; /* in-kernel BTF type id to attach to */ 793 u32 ctx_arg_info_size; 794 u32 max_rdonly_access; 795 u32 max_rdwr_access; 796 struct btf *attach_btf; 797 const struct bpf_ctx_arg_aux *ctx_arg_info; 798 struct mutex dst_mutex; /* protects dst_* pointers below, *after* prog becomes visible */ 799 struct bpf_prog *dst_prog; 800 struct bpf_trampoline *dst_trampoline; 801 enum bpf_prog_type saved_dst_prog_type; 802 enum bpf_attach_type saved_dst_attach_type; 803 bool verifier_zext; /* Zero extensions has been inserted by verifier. */ 804 bool offload_requested; 805 bool attach_btf_trace; /* true if attaching to BTF-enabled raw tp */ 806 bool func_proto_unreliable; 807 bool sleepable; 808 bool tail_call_reachable; 809 struct hlist_node tramp_hlist; 810 /* BTF_KIND_FUNC_PROTO for valid attach_btf_id */ 811 const struct btf_type *attach_func_proto; 812 /* function name for valid attach_btf_id */ 813 const char *attach_func_name; 814 struct bpf_prog **func; 815 void *jit_data; /* JIT specific data. arch dependent */ 816 struct bpf_jit_poke_descriptor *poke_tab; 817 u32 size_poke_tab; 818 struct bpf_ksym ksym; 819 const struct bpf_prog_ops *ops; 820 struct bpf_map **used_maps; 821 struct mutex used_maps_mutex; /* mutex for used_maps and used_map_cnt */ 822 struct btf_mod_pair *used_btfs; 823 struct bpf_prog *prog; 824 struct user_struct *user; 825 u64 load_time; /* ns since boottime */ 826 struct bpf_map *cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]; 827 char name[BPF_OBJ_NAME_LEN]; 828#ifdef CONFIG_SECURITY 829 void *security; 830#endif 831 struct bpf_prog_offload *offload; 832 struct btf *btf; 833 struct bpf_func_info *func_info; 834 struct bpf_func_info_aux *func_info_aux; 835 /* bpf_line_info loaded from userspace. linfo->insn_off 836 * has the xlated insn offset. 837 * Both the main and sub prog share the same linfo. 838 * The subprog can access its first linfo by 839 * using the linfo_idx. 840 */ 841 struct bpf_line_info *linfo; 842 /* jited_linfo is the jited addr of the linfo. It has a 843 * one to one mapping to linfo: 844 * jited_linfo[i] is the jited addr for the linfo[i]->insn_off. 845 * Both the main and sub prog share the same jited_linfo. 846 * The subprog can access its first jited_linfo by 847 * using the linfo_idx. 848 */ 849 void **jited_linfo; 850 u32 func_info_cnt; 851 u32 nr_linfo; 852 /* subprog can use linfo_idx to access its first linfo and 853 * jited_linfo. 854 * main prog always has linfo_idx == 0 855 */ 856 u32 linfo_idx; 857 u32 num_exentries; 858 struct exception_table_entry *extable; 859 union { 860 struct work_struct work; 861 struct rcu_head rcu; 862 }; 863}; 864 865struct bpf_array_aux { 866 /* 'Ownership' of prog array is claimed by the first program that 867 * is going to use this map or by the first program which FD is 868 * stored in the map to make sure that all callers and callees have 869 * the same prog type and JITed flag. 870 */ 871 enum bpf_prog_type type; 872 bool jited; 873 /* Programs with direct jumps into programs part of this array. */ 874 struct list_head poke_progs; 875 struct bpf_map *map; 876 struct mutex poke_mutex; 877 struct work_struct work; 878}; 879 880struct bpf_link { 881 atomic64_t refcnt; 882 u32 id; 883 enum bpf_link_type type; 884 const struct bpf_link_ops *ops; 885 struct bpf_prog *prog; 886 struct work_struct work; 887}; 888 889struct bpf_link_ops { 890 void (*release)(struct bpf_link *link); 891 void (*dealloc)(struct bpf_link *link); 892 int (*detach)(struct bpf_link *link); 893 int (*update_prog)(struct bpf_link *link, struct bpf_prog *new_prog, 894 struct bpf_prog *old_prog); 895 void (*show_fdinfo)(const struct bpf_link *link, struct seq_file *seq); 896 int (*fill_link_info)(const struct bpf_link *link, 897 struct bpf_link_info *info); 898}; 899 900struct bpf_link_primer { 901 struct bpf_link *link; 902 struct file *file; 903 int fd; 904 u32 id; 905}; 906 907struct bpf_struct_ops_value; 908struct btf_type; 909struct btf_member; 910 911#define BPF_STRUCT_OPS_MAX_NR_MEMBERS 64 912struct bpf_struct_ops { 913 const struct bpf_verifier_ops *verifier_ops; 914 int (*init)(struct btf *btf); 915 int (*check_member)(const struct btf_type *t, 916 const struct btf_member *member); 917 int (*init_member)(const struct btf_type *t, 918 const struct btf_member *member, 919 void *kdata, const void *udata); 920 int (*reg)(void *kdata); 921 void (*unreg)(void *kdata); 922 const struct btf_type *type; 923 const struct btf_type *value_type; 924 const char *name; 925 struct btf_func_model func_models[BPF_STRUCT_OPS_MAX_NR_MEMBERS]; 926 u32 type_id; 927 u32 value_id; 928}; 929 930#if defined(CONFIG_BPF_JIT) && defined(CONFIG_BPF_SYSCALL) 931#define BPF_MODULE_OWNER ((void *)((0xeB9FUL << 2) + POISON_POINTER_DELTA)) 932const struct bpf_struct_ops *bpf_struct_ops_find(u32 type_id); 933void bpf_struct_ops_init(struct btf *btf, struct bpf_verifier_log *log); 934bool bpf_struct_ops_get(const void *kdata); 935void bpf_struct_ops_put(const void *kdata); 936int bpf_struct_ops_map_sys_lookup_elem(struct bpf_map *map, void *key, 937 void *value); 938static inline bool bpf_try_module_get(const void *data, struct module *owner) 939{ 940 if (owner == BPF_MODULE_OWNER) 941 return bpf_struct_ops_get(data); 942 else 943 return try_module_get(owner); 944} 945static inline void bpf_module_put(const void *data, struct module *owner) 946{ 947 if (owner == BPF_MODULE_OWNER) 948 bpf_struct_ops_put(data); 949 else 950 module_put(owner); 951} 952#else 953static inline const struct bpf_struct_ops *bpf_struct_ops_find(u32 type_id) 954{ 955 return NULL; 956} 957static inline void bpf_struct_ops_init(struct btf *btf, 958 struct bpf_verifier_log *log) 959{ 960} 961static inline bool bpf_try_module_get(const void *data, struct module *owner) 962{ 963 return try_module_get(owner); 964} 965static inline void bpf_module_put(const void *data, struct module *owner) 966{ 967 module_put(owner); 968} 969static inline int bpf_struct_ops_map_sys_lookup_elem(struct bpf_map *map, 970 void *key, 971 void *value) 972{ 973 return -EINVAL; 974} 975#endif 976 977struct bpf_array { 978 struct bpf_map map; 979 u32 elem_size; 980 u32 index_mask; 981 struct bpf_array_aux *aux; 982 union { 983 char value[0] __aligned(8); 984 void *ptrs[0] __aligned(8); 985 void __percpu *pptrs[0] __aligned(8); 986 }; 987}; 988 989#define BPF_COMPLEXITY_LIMIT_INSNS 1000000 /* yes. 1M insns */ 990#define MAX_TAIL_CALL_CNT 32 991 992#define BPF_F_ACCESS_MASK (BPF_F_RDONLY | \ 993 BPF_F_RDONLY_PROG | \ 994 BPF_F_WRONLY | \ 995 BPF_F_WRONLY_PROG) 996 997#define BPF_MAP_CAN_READ BIT(0) 998#define BPF_MAP_CAN_WRITE BIT(1) 999 1000static inline u32 bpf_map_flags_to_cap(struct bpf_map *map) 1001{ 1002 u32 access_flags = map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG); 1003 1004 /* Combination of BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG is 1005 * not possible. 1006 */ 1007 if (access_flags & BPF_F_RDONLY_PROG) 1008 return BPF_MAP_CAN_READ; 1009 else if (access_flags & BPF_F_WRONLY_PROG) 1010 return BPF_MAP_CAN_WRITE; 1011 else 1012 return BPF_MAP_CAN_READ | BPF_MAP_CAN_WRITE; 1013} 1014 1015static inline bool bpf_map_flags_access_ok(u32 access_flags) 1016{ 1017 return (access_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) != 1018 (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG); 1019} 1020 1021struct bpf_event_entry { 1022 struct perf_event *event; 1023 struct file *perf_file; 1024 struct file *map_file; 1025 struct rcu_head rcu; 1026}; 1027 1028bool bpf_prog_array_compatible(struct bpf_array *array, const struct bpf_prog *fp); 1029int bpf_prog_calc_tag(struct bpf_prog *fp); 1030 1031const struct bpf_func_proto *bpf_get_trace_printk_proto(void); 1032 1033typedef unsigned long (*bpf_ctx_copy_t)(void *dst, const void *src, 1034 unsigned long off, unsigned long len); 1035typedef u32 (*bpf_convert_ctx_access_t)(enum bpf_access_type type, 1036 const struct bpf_insn *src, 1037 struct bpf_insn *dst, 1038 struct bpf_prog *prog, 1039 u32 *target_size); 1040 1041u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, 1042 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy); 1043 1044/* an array of programs to be executed under rcu_lock. 1045 * 1046 * Typical usage: 1047 * ret = BPF_PROG_RUN_ARRAY(&bpf_prog_array, ctx, BPF_PROG_RUN); 1048 * 1049 * the structure returned by bpf_prog_array_alloc() should be populated 1050 * with program pointers and the last pointer must be NULL. 1051 * The user has to keep refcnt on the program and make sure the program 1052 * is removed from the array before bpf_prog_put(). 1053 * The 'struct bpf_prog_array *' should only be replaced with xchg() 1054 * since other cpus are walking the array of pointers in parallel. 1055 */ 1056struct bpf_prog_array_item { 1057 struct bpf_prog *prog; 1058 struct bpf_cgroup_storage *cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]; 1059}; 1060 1061struct bpf_prog_array { 1062 struct rcu_head rcu; 1063 struct bpf_prog_array_item items[]; 1064}; 1065 1066struct bpf_prog_array *bpf_prog_array_alloc(u32 prog_cnt, gfp_t flags); 1067void bpf_prog_array_free(struct bpf_prog_array *progs); 1068int bpf_prog_array_length(struct bpf_prog_array *progs); 1069bool bpf_prog_array_is_empty(struct bpf_prog_array *array); 1070int bpf_prog_array_copy_to_user(struct bpf_prog_array *progs, 1071 __u32 __user *prog_ids, u32 cnt); 1072 1073void bpf_prog_array_delete_safe(struct bpf_prog_array *progs, 1074 struct bpf_prog *old_prog); 1075int bpf_prog_array_delete_safe_at(struct bpf_prog_array *array, int index); 1076int bpf_prog_array_update_at(struct bpf_prog_array *array, int index, 1077 struct bpf_prog *prog); 1078int bpf_prog_array_copy_info(struct bpf_prog_array *array, 1079 u32 *prog_ids, u32 request_cnt, 1080 u32 *prog_cnt); 1081int bpf_prog_array_copy(struct bpf_prog_array *old_array, 1082 struct bpf_prog *exclude_prog, 1083 struct bpf_prog *include_prog, 1084 struct bpf_prog_array **new_array); 1085 1086/* BPF program asks to bypass CAP_NET_BIND_SERVICE in bind. */ 1087#define BPF_RET_BIND_NO_CAP_NET_BIND_SERVICE (1 << 0) 1088/* BPF program asks to set CN on the packet. */ 1089#define BPF_RET_SET_CN (1 << 0) 1090 1091#define BPF_PROG_RUN_ARRAY_FLAGS(array, ctx, func, ret_flags) \ 1092 ({ \ 1093 struct bpf_prog_array_item *_item; \ 1094 struct bpf_prog *_prog; \ 1095 struct bpf_prog_array *_array; \ 1096 u32 _ret = 1; \ 1097 u32 func_ret; \ 1098 migrate_disable(); \ 1099 rcu_read_lock(); \ 1100 _array = rcu_dereference(array); \ 1101 _item = &_array->items[0]; \ 1102 while ((_prog = READ_ONCE(_item->prog))) { \ 1103 bpf_cgroup_storage_set(_item->cgroup_storage); \ 1104 func_ret = func(_prog, ctx); \ 1105 _ret &= (func_ret & 1); \ 1106 *(ret_flags) |= (func_ret >> 1); \ 1107 _item++; \ 1108 } \ 1109 rcu_read_unlock(); \ 1110 migrate_enable(); \ 1111 _ret; \ 1112 }) 1113 1114#define __BPF_PROG_RUN_ARRAY(array, ctx, func, check_non_null, set_cg_storage) \ 1115 ({ \ 1116 struct bpf_prog_array_item *_item; \ 1117 struct bpf_prog *_prog; \ 1118 struct bpf_prog_array *_array; \ 1119 u32 _ret = 1; \ 1120 migrate_disable(); \ 1121 rcu_read_lock(); \ 1122 _array = rcu_dereference(array); \ 1123 if (unlikely(check_non_null && !_array))\ 1124 goto _out; \ 1125 _item = &_array->items[0]; \ 1126 while ((_prog = READ_ONCE(_item->prog))) { \ 1127 if (set_cg_storage) \ 1128 bpf_cgroup_storage_set(_item->cgroup_storage); \ 1129 _ret &= func(_prog, ctx); \ 1130 _item++; \ 1131 } \ 1132_out: \ 1133 rcu_read_unlock(); \ 1134 migrate_enable(); \ 1135 _ret; \ 1136 }) 1137 1138/* To be used by __cgroup_bpf_run_filter_skb for EGRESS BPF progs 1139 * so BPF programs can request cwr for TCP packets. 1140 * 1141 * Current cgroup skb programs can only return 0 or 1 (0 to drop the 1142 * packet. This macro changes the behavior so the low order bit 1143 * indicates whether the packet should be dropped (0) or not (1) 1144 * and the next bit is a congestion notification bit. This could be 1145 * used by TCP to call tcp_enter_cwr() 1146 * 1147 * Hence, new allowed return values of CGROUP EGRESS BPF programs are: 1148 * 0: drop packet 1149 * 1: keep packet 1150 * 2: drop packet and cn 1151 * 3: keep packet and cn 1152 * 1153 * This macro then converts it to one of the NET_XMIT or an error 1154 * code that is then interpreted as drop packet (and no cn): 1155 * 0: NET_XMIT_SUCCESS skb should be transmitted 1156 * 1: NET_XMIT_DROP skb should be dropped and cn 1157 * 2: NET_XMIT_CN skb should be transmitted and cn 1158 * 3: -EPERM skb should be dropped 1159 */ 1160#define BPF_PROG_CGROUP_INET_EGRESS_RUN_ARRAY(array, ctx, func) \ 1161 ({ \ 1162 u32 _flags = 0; \ 1163 bool _cn; \ 1164 u32 _ret; \ 1165 _ret = BPF_PROG_RUN_ARRAY_FLAGS(array, ctx, func, &_flags); \ 1166 _cn = _flags & BPF_RET_SET_CN; \ 1167 if (_ret) \ 1168 _ret = (_cn ? NET_XMIT_CN : NET_XMIT_SUCCESS); \ 1169 else \ 1170 _ret = (_cn ? NET_XMIT_DROP : -EPERM); \ 1171 _ret; \ 1172 }) 1173 1174#define BPF_PROG_RUN_ARRAY(array, ctx, func) \ 1175 __BPF_PROG_RUN_ARRAY(array, ctx, func, false, true) 1176 1177#define BPF_PROG_RUN_ARRAY_CHECK(array, ctx, func) \ 1178 __BPF_PROG_RUN_ARRAY(array, ctx, func, true, false) 1179 1180#ifdef CONFIG_BPF_SYSCALL 1181DECLARE_PER_CPU(int, bpf_prog_active); 1182extern struct mutex bpf_stats_enabled_mutex; 1183 1184/* 1185 * Block execution of BPF programs attached to instrumentation (perf, 1186 * kprobes, tracepoints) to prevent deadlocks on map operations as any of 1187 * these events can happen inside a region which holds a map bucket lock 1188 * and can deadlock on it. 1189 * 1190 * Use the preemption safe inc/dec variants on RT because migrate disable 1191 * is preemptible on RT and preemption in the middle of the RMW operation 1192 * might lead to inconsistent state. Use the raw variants for non RT 1193 * kernels as migrate_disable() maps to preempt_disable() so the slightly 1194 * more expensive save operation can be avoided. 1195 */ 1196static inline void bpf_disable_instrumentation(void) 1197{ 1198 migrate_disable(); 1199 if (IS_ENABLED(CONFIG_PREEMPT_RT)) 1200 this_cpu_inc(bpf_prog_active); 1201 else 1202 __this_cpu_inc(bpf_prog_active); 1203} 1204 1205static inline void bpf_enable_instrumentation(void) 1206{ 1207 if (IS_ENABLED(CONFIG_PREEMPT_RT)) 1208 this_cpu_dec(bpf_prog_active); 1209 else 1210 __this_cpu_dec(bpf_prog_active); 1211 migrate_enable(); 1212} 1213 1214extern const struct file_operations bpf_map_fops; 1215extern const struct file_operations bpf_prog_fops; 1216extern const struct file_operations bpf_iter_fops; 1217 1218#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \ 1219 extern const struct bpf_prog_ops _name ## _prog_ops; \ 1220 extern const struct bpf_verifier_ops _name ## _verifier_ops; 1221#define BPF_MAP_TYPE(_id, _ops) \ 1222 extern const struct bpf_map_ops _ops; 1223#define BPF_LINK_TYPE(_id, _name) 1224#include <linux/bpf_types.h> 1225#undef BPF_PROG_TYPE 1226#undef BPF_MAP_TYPE 1227#undef BPF_LINK_TYPE 1228 1229extern const struct bpf_prog_ops bpf_offload_prog_ops; 1230extern const struct bpf_verifier_ops tc_cls_act_analyzer_ops; 1231extern const struct bpf_verifier_ops xdp_analyzer_ops; 1232 1233struct bpf_prog *bpf_prog_get(u32 ufd); 1234struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type, 1235 bool attach_drv); 1236void bpf_prog_add(struct bpf_prog *prog, int i); 1237void bpf_prog_sub(struct bpf_prog *prog, int i); 1238void bpf_prog_inc(struct bpf_prog *prog); 1239struct bpf_prog * __must_check bpf_prog_inc_not_zero(struct bpf_prog *prog); 1240void bpf_prog_put(struct bpf_prog *prog); 1241 1242void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock); 1243void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock); 1244 1245struct bpf_map *bpf_map_get(u32 ufd); 1246struct bpf_map *bpf_map_get_with_uref(u32 ufd); 1247struct bpf_map *__bpf_map_get(struct fd f); 1248void bpf_map_inc(struct bpf_map *map); 1249void bpf_map_inc_with_uref(struct bpf_map *map); 1250struct bpf_map * __must_check bpf_map_inc_not_zero(struct bpf_map *map); 1251void bpf_map_put_with_uref(struct bpf_map *map); 1252void bpf_map_put(struct bpf_map *map); 1253void *bpf_map_area_alloc(u64 size, int numa_node); 1254void *bpf_map_area_mmapable_alloc(u64 size, int numa_node); 1255void bpf_map_area_free(void *base); 1256void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr); 1257int generic_map_lookup_batch(struct bpf_map *map, 1258 const union bpf_attr *attr, 1259 union bpf_attr __user *uattr); 1260int generic_map_update_batch(struct bpf_map *map, 1261 const union bpf_attr *attr, 1262 union bpf_attr __user *uattr); 1263int generic_map_delete_batch(struct bpf_map *map, 1264 const union bpf_attr *attr, 1265 union bpf_attr __user *uattr); 1266struct bpf_map *bpf_map_get_curr_or_next(u32 *id); 1267struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id); 1268 1269#ifdef CONFIG_MEMCG_KMEM 1270void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags, 1271 int node); 1272void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags); 1273void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size, 1274 size_t align, gfp_t flags); 1275#else 1276static inline void * 1277bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags, 1278 int node) 1279{ 1280 return kmalloc_node(size, flags, node); 1281} 1282 1283static inline void * 1284bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags) 1285{ 1286 return kzalloc(size, flags); 1287} 1288 1289static inline void __percpu * 1290bpf_map_alloc_percpu(const struct bpf_map *map, size_t size, size_t align, 1291 gfp_t flags) 1292{ 1293 return __alloc_percpu_gfp(size, align, flags); 1294} 1295#endif 1296 1297extern int sysctl_unprivileged_bpf_disabled; 1298 1299static inline bool bpf_allow_ptr_leaks(void) 1300{ 1301 return perfmon_capable(); 1302} 1303 1304static inline bool bpf_allow_uninit_stack(void) 1305{ 1306 return perfmon_capable(); 1307} 1308 1309static inline bool bpf_allow_ptr_to_map_access(void) 1310{ 1311 return perfmon_capable(); 1312} 1313 1314static inline bool bpf_bypass_spec_v1(void) 1315{ 1316 return perfmon_capable(); 1317} 1318 1319static inline bool bpf_bypass_spec_v4(void) 1320{ 1321 return perfmon_capable(); 1322} 1323 1324int bpf_map_new_fd(struct bpf_map *map, int flags); 1325int bpf_prog_new_fd(struct bpf_prog *prog); 1326 1327void bpf_link_init(struct bpf_link *link, enum bpf_link_type type, 1328 const struct bpf_link_ops *ops, struct bpf_prog *prog); 1329int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer); 1330int bpf_link_settle(struct bpf_link_primer *primer); 1331void bpf_link_cleanup(struct bpf_link_primer *primer); 1332void bpf_link_inc(struct bpf_link *link); 1333void bpf_link_put(struct bpf_link *link); 1334int bpf_link_new_fd(struct bpf_link *link); 1335struct file *bpf_link_new_file(struct bpf_link *link, int *reserved_fd); 1336struct bpf_link *bpf_link_get_from_fd(u32 ufd); 1337 1338int bpf_obj_pin_user(u32 ufd, const char __user *pathname); 1339int bpf_obj_get_user(const char __user *pathname, int flags); 1340 1341#define BPF_ITER_FUNC_PREFIX "bpf_iter_" 1342#define DEFINE_BPF_ITER_FUNC(target, args...) \ 1343 extern int bpf_iter_ ## target(args); \ 1344 int __init bpf_iter_ ## target(args) { return 0; } 1345 1346struct bpf_iter_aux_info { 1347 struct bpf_map *map; 1348}; 1349 1350typedef int (*bpf_iter_attach_target_t)(struct bpf_prog *prog, 1351 union bpf_iter_link_info *linfo, 1352 struct bpf_iter_aux_info *aux); 1353typedef void (*bpf_iter_detach_target_t)(struct bpf_iter_aux_info *aux); 1354typedef void (*bpf_iter_show_fdinfo_t) (const struct bpf_iter_aux_info *aux, 1355 struct seq_file *seq); 1356typedef int (*bpf_iter_fill_link_info_t)(const struct bpf_iter_aux_info *aux, 1357 struct bpf_link_info *info); 1358 1359enum bpf_iter_feature { 1360 BPF_ITER_RESCHED = BIT(0), 1361}; 1362 1363#define BPF_ITER_CTX_ARG_MAX 2 1364struct bpf_iter_reg { 1365 const char *target; 1366 bpf_iter_attach_target_t attach_target; 1367 bpf_iter_detach_target_t detach_target; 1368 bpf_iter_show_fdinfo_t show_fdinfo; 1369 bpf_iter_fill_link_info_t fill_link_info; 1370 u32 ctx_arg_info_size; 1371 u32 feature; 1372 struct bpf_ctx_arg_aux ctx_arg_info[BPF_ITER_CTX_ARG_MAX]; 1373 const struct bpf_iter_seq_info *seq_info; 1374}; 1375 1376struct bpf_iter_meta { 1377 __bpf_md_ptr(struct seq_file *, seq); 1378 u64 session_id; 1379 u64 seq_num; 1380}; 1381 1382struct bpf_iter__bpf_map_elem { 1383 __bpf_md_ptr(struct bpf_iter_meta *, meta); 1384 __bpf_md_ptr(struct bpf_map *, map); 1385 __bpf_md_ptr(void *, key); 1386 __bpf_md_ptr(void *, value); 1387}; 1388 1389int bpf_iter_reg_target(const struct bpf_iter_reg *reg_info); 1390void bpf_iter_unreg_target(const struct bpf_iter_reg *reg_info); 1391bool bpf_iter_prog_supported(struct bpf_prog *prog); 1392int bpf_iter_link_attach(const union bpf_attr *attr, struct bpf_prog *prog); 1393int bpf_iter_new_fd(struct bpf_link *link); 1394bool bpf_link_is_iter(struct bpf_link *link); 1395struct bpf_prog *bpf_iter_get_info(struct bpf_iter_meta *meta, bool in_stop); 1396int bpf_iter_run_prog(struct bpf_prog *prog, void *ctx); 1397void bpf_iter_map_show_fdinfo(const struct bpf_iter_aux_info *aux, 1398 struct seq_file *seq); 1399int bpf_iter_map_fill_link_info(const struct bpf_iter_aux_info *aux, 1400 struct bpf_link_info *info); 1401 1402int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value); 1403int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value); 1404int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value, 1405 u64 flags); 1406int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value, 1407 u64 flags); 1408 1409int bpf_stackmap_copy(struct bpf_map *map, void *key, void *value); 1410 1411int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file, 1412 void *key, void *value, u64 map_flags); 1413int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value); 1414int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file, 1415 void *key, void *value, u64 map_flags); 1416int bpf_fd_htab_map_lookup_elem(struct bpf_map *map, void *key, u32 *value); 1417 1418int bpf_get_file_flag(int flags); 1419int bpf_check_uarg_tail_zero(void __user *uaddr, size_t expected_size, 1420 size_t actual_size); 1421 1422/* memcpy that is used with 8-byte aligned pointers, power-of-8 size and 1423 * forced to use 'long' read/writes to try to atomically copy long counters. 1424 * Best-effort only. No barriers here, since it _will_ race with concurrent 1425 * updates from BPF programs. Called from bpf syscall and mostly used with 1426 * size 8 or 16 bytes, so ask compiler to inline it. 1427 */ 1428static inline void bpf_long_memcpy(void *dst, const void *src, u32 size) 1429{ 1430 const long *lsrc = src; 1431 long *ldst = dst; 1432 1433 size /= sizeof(long); 1434 while (size--) 1435 *ldst++ = *lsrc++; 1436} 1437 1438/* verify correctness of eBPF program */ 1439int bpf_check(struct bpf_prog **fp, union bpf_attr *attr, 1440 union bpf_attr __user *uattr); 1441 1442#ifndef CONFIG_BPF_JIT_ALWAYS_ON 1443void bpf_patch_call_args(struct bpf_insn *insn, u32 stack_depth); 1444#endif 1445 1446struct btf *bpf_get_btf_vmlinux(void); 1447 1448/* Map specifics */ 1449struct xdp_buff; 1450struct sk_buff; 1451 1452struct bpf_dtab_netdev *__dev_map_lookup_elem(struct bpf_map *map, u32 key); 1453struct bpf_dtab_netdev *__dev_map_hash_lookup_elem(struct bpf_map *map, u32 key); 1454void __dev_flush(void); 1455int dev_xdp_enqueue(struct net_device *dev, struct xdp_buff *xdp, 1456 struct net_device *dev_rx); 1457int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp, 1458 struct net_device *dev_rx); 1459int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, struct sk_buff *skb, 1460 struct bpf_prog *xdp_prog); 1461bool dev_map_can_have_prog(struct bpf_map *map); 1462 1463struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key); 1464void __cpu_map_flush(void); 1465int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_buff *xdp, 1466 struct net_device *dev_rx); 1467bool cpu_map_prog_allowed(struct bpf_map *map); 1468 1469/* Return map's numa specified by userspace */ 1470static inline int bpf_map_attr_numa_node(const union bpf_attr *attr) 1471{ 1472 return (attr->map_flags & BPF_F_NUMA_NODE) ? 1473 attr->numa_node : NUMA_NO_NODE; 1474} 1475 1476struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type); 1477int array_map_alloc_check(union bpf_attr *attr); 1478 1479int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr, 1480 union bpf_attr __user *uattr); 1481int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr, 1482 union bpf_attr __user *uattr); 1483int bpf_prog_test_run_tracing(struct bpf_prog *prog, 1484 const union bpf_attr *kattr, 1485 union bpf_attr __user *uattr); 1486int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog, 1487 const union bpf_attr *kattr, 1488 union bpf_attr __user *uattr); 1489int bpf_prog_test_run_raw_tp(struct bpf_prog *prog, 1490 const union bpf_attr *kattr, 1491 union bpf_attr __user *uattr); 1492bool btf_ctx_access(int off, int size, enum bpf_access_type type, 1493 const struct bpf_prog *prog, 1494 struct bpf_insn_access_aux *info); 1495int btf_struct_access(struct bpf_verifier_log *log, const struct btf *btf, 1496 const struct btf_type *t, int off, int size, 1497 enum bpf_access_type atype, 1498 u32 *next_btf_id); 1499bool btf_struct_ids_match(struct bpf_verifier_log *log, 1500 const struct btf *btf, u32 id, int off, 1501 const struct btf *need_btf, u32 need_type_id); 1502 1503int btf_distill_func_proto(struct bpf_verifier_log *log, 1504 struct btf *btf, 1505 const struct btf_type *func_proto, 1506 const char *func_name, 1507 struct btf_func_model *m); 1508 1509struct bpf_reg_state; 1510int btf_check_func_arg_match(struct bpf_verifier_env *env, int subprog, 1511 struct bpf_reg_state *regs); 1512int btf_prepare_func_args(struct bpf_verifier_env *env, int subprog, 1513 struct bpf_reg_state *reg); 1514int btf_check_type_match(struct bpf_verifier_log *log, const struct bpf_prog *prog, 1515 struct btf *btf, const struct btf_type *t); 1516 1517struct bpf_prog *bpf_prog_by_id(u32 id); 1518struct bpf_link *bpf_link_by_id(u32 id); 1519 1520const struct bpf_func_proto *bpf_base_func_proto(enum bpf_func_id func_id); 1521#else /* !CONFIG_BPF_SYSCALL */ 1522static inline struct bpf_prog *bpf_prog_get(u32 ufd) 1523{ 1524 return ERR_PTR(-EOPNOTSUPP); 1525} 1526 1527static inline struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, 1528 enum bpf_prog_type type, 1529 bool attach_drv) 1530{ 1531 return ERR_PTR(-EOPNOTSUPP); 1532} 1533 1534static inline void bpf_prog_add(struct bpf_prog *prog, int i) 1535{ 1536} 1537 1538static inline void bpf_prog_sub(struct bpf_prog *prog, int i) 1539{ 1540} 1541 1542static inline void bpf_prog_put(struct bpf_prog *prog) 1543{ 1544} 1545 1546static inline void bpf_prog_inc(struct bpf_prog *prog) 1547{ 1548} 1549 1550static inline struct bpf_prog *__must_check 1551bpf_prog_inc_not_zero(struct bpf_prog *prog) 1552{ 1553 return ERR_PTR(-EOPNOTSUPP); 1554} 1555 1556static inline void bpf_link_init(struct bpf_link *link, enum bpf_link_type type, 1557 const struct bpf_link_ops *ops, 1558 struct bpf_prog *prog) 1559{ 1560} 1561 1562static inline int bpf_link_prime(struct bpf_link *link, 1563 struct bpf_link_primer *primer) 1564{ 1565 return -EOPNOTSUPP; 1566} 1567 1568static inline int bpf_link_settle(struct bpf_link_primer *primer) 1569{ 1570 return -EOPNOTSUPP; 1571} 1572 1573static inline void bpf_link_cleanup(struct bpf_link_primer *primer) 1574{ 1575} 1576 1577static inline void bpf_link_inc(struct bpf_link *link) 1578{ 1579} 1580 1581static inline void bpf_link_put(struct bpf_link *link) 1582{ 1583} 1584 1585static inline int bpf_obj_get_user(const char __user *pathname, int flags) 1586{ 1587 return -EOPNOTSUPP; 1588} 1589 1590static inline struct net_device *__dev_map_lookup_elem(struct bpf_map *map, 1591 u32 key) 1592{ 1593 return NULL; 1594} 1595 1596static inline struct net_device *__dev_map_hash_lookup_elem(struct bpf_map *map, 1597 u32 key) 1598{ 1599 return NULL; 1600} 1601static inline bool dev_map_can_have_prog(struct bpf_map *map) 1602{ 1603 return false; 1604} 1605 1606static inline void __dev_flush(void) 1607{ 1608} 1609 1610struct xdp_buff; 1611struct bpf_dtab_netdev; 1612 1613static inline 1614int dev_xdp_enqueue(struct net_device *dev, struct xdp_buff *xdp, 1615 struct net_device *dev_rx) 1616{ 1617 return 0; 1618} 1619 1620static inline 1621int dev_map_enqueue(struct bpf_dtab_netdev *dst, struct xdp_buff *xdp, 1622 struct net_device *dev_rx) 1623{ 1624 return 0; 1625} 1626 1627struct sk_buff; 1628 1629static inline int dev_map_generic_redirect(struct bpf_dtab_netdev *dst, 1630 struct sk_buff *skb, 1631 struct bpf_prog *xdp_prog) 1632{ 1633 return 0; 1634} 1635 1636static inline 1637struct bpf_cpu_map_entry *__cpu_map_lookup_elem(struct bpf_map *map, u32 key) 1638{ 1639 return NULL; 1640} 1641 1642static inline void __cpu_map_flush(void) 1643{ 1644} 1645 1646static inline int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, 1647 struct xdp_buff *xdp, 1648 struct net_device *dev_rx) 1649{ 1650 return 0; 1651} 1652 1653static inline bool cpu_map_prog_allowed(struct bpf_map *map) 1654{ 1655 return false; 1656} 1657 1658static inline struct bpf_prog *bpf_prog_get_type_path(const char *name, 1659 enum bpf_prog_type type) 1660{ 1661 return ERR_PTR(-EOPNOTSUPP); 1662} 1663 1664static inline int bpf_prog_test_run_xdp(struct bpf_prog *prog, 1665 const union bpf_attr *kattr, 1666 union bpf_attr __user *uattr) 1667{ 1668 return -ENOTSUPP; 1669} 1670 1671static inline int bpf_prog_test_run_skb(struct bpf_prog *prog, 1672 const union bpf_attr *kattr, 1673 union bpf_attr __user *uattr) 1674{ 1675 return -ENOTSUPP; 1676} 1677 1678static inline int bpf_prog_test_run_tracing(struct bpf_prog *prog, 1679 const union bpf_attr *kattr, 1680 union bpf_attr __user *uattr) 1681{ 1682 return -ENOTSUPP; 1683} 1684 1685static inline int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog, 1686 const union bpf_attr *kattr, 1687 union bpf_attr __user *uattr) 1688{ 1689 return -ENOTSUPP; 1690} 1691 1692static inline void bpf_map_put(struct bpf_map *map) 1693{ 1694} 1695 1696static inline struct bpf_prog *bpf_prog_by_id(u32 id) 1697{ 1698 return ERR_PTR(-ENOTSUPP); 1699} 1700 1701static inline const struct bpf_func_proto * 1702bpf_base_func_proto(enum bpf_func_id func_id) 1703{ 1704 return NULL; 1705} 1706#endif /* CONFIG_BPF_SYSCALL */ 1707 1708void __bpf_free_used_btfs(struct bpf_prog_aux *aux, 1709 struct btf_mod_pair *used_btfs, u32 len); 1710 1711static inline struct bpf_prog *bpf_prog_get_type(u32 ufd, 1712 enum bpf_prog_type type) 1713{ 1714 return bpf_prog_get_type_dev(ufd, type, false); 1715} 1716 1717void __bpf_free_used_maps(struct bpf_prog_aux *aux, 1718 struct bpf_map **used_maps, u32 len); 1719 1720bool bpf_prog_get_ok(struct bpf_prog *, enum bpf_prog_type *, bool); 1721 1722int bpf_prog_offload_compile(struct bpf_prog *prog); 1723void bpf_prog_offload_destroy(struct bpf_prog *prog); 1724int bpf_prog_offload_info_fill(struct bpf_prog_info *info, 1725 struct bpf_prog *prog); 1726 1727int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map); 1728 1729int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value); 1730int bpf_map_offload_update_elem(struct bpf_map *map, 1731 void *key, void *value, u64 flags); 1732int bpf_map_offload_delete_elem(struct bpf_map *map, void *key); 1733int bpf_map_offload_get_next_key(struct bpf_map *map, 1734 void *key, void *next_key); 1735 1736bool bpf_offload_prog_map_match(struct bpf_prog *prog, struct bpf_map *map); 1737 1738struct bpf_offload_dev * 1739bpf_offload_dev_create(const struct bpf_prog_offload_ops *ops, void *priv); 1740void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev); 1741void *bpf_offload_dev_priv(struct bpf_offload_dev *offdev); 1742int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev, 1743 struct net_device *netdev); 1744void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev, 1745 struct net_device *netdev); 1746bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev); 1747 1748#if defined(CONFIG_NET) && defined(CONFIG_BPF_SYSCALL) 1749int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr); 1750 1751static inline bool bpf_prog_is_dev_bound(const struct bpf_prog_aux *aux) 1752{ 1753 return aux->offload_requested; 1754} 1755 1756static inline bool bpf_map_is_dev_bound(struct bpf_map *map) 1757{ 1758 return unlikely(map->ops == &bpf_map_offload_ops); 1759} 1760 1761struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr); 1762void bpf_map_offload_map_free(struct bpf_map *map); 1763#else 1764static inline int bpf_prog_offload_init(struct bpf_prog *prog, 1765 union bpf_attr *attr) 1766{ 1767 return -EOPNOTSUPP; 1768} 1769 1770static inline bool bpf_prog_is_dev_bound(struct bpf_prog_aux *aux) 1771{ 1772 return false; 1773} 1774 1775static inline bool bpf_map_is_dev_bound(struct bpf_map *map) 1776{ 1777 return false; 1778} 1779 1780static inline struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr) 1781{ 1782 return ERR_PTR(-EOPNOTSUPP); 1783} 1784 1785static inline void bpf_map_offload_map_free(struct bpf_map *map) 1786{ 1787} 1788#endif /* CONFIG_NET && CONFIG_BPF_SYSCALL */ 1789 1790#if defined(CONFIG_BPF_STREAM_PARSER) 1791int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog, 1792 struct bpf_prog *old, u32 which); 1793int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog); 1794int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype); 1795int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value, u64 flags); 1796void sock_map_unhash(struct sock *sk); 1797void sock_map_close(struct sock *sk, long timeout); 1798#else 1799static inline int sock_map_prog_update(struct bpf_map *map, 1800 struct bpf_prog *prog, 1801 struct bpf_prog *old, u32 which) 1802{ 1803 return -EOPNOTSUPP; 1804} 1805 1806static inline int sock_map_get_from_fd(const union bpf_attr *attr, 1807 struct bpf_prog *prog) 1808{ 1809 return -EINVAL; 1810} 1811 1812static inline int sock_map_prog_detach(const union bpf_attr *attr, 1813 enum bpf_prog_type ptype) 1814{ 1815 return -EOPNOTSUPP; 1816} 1817 1818static inline int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value, 1819 u64 flags) 1820{ 1821 return -EOPNOTSUPP; 1822} 1823#endif /* CONFIG_BPF_STREAM_PARSER */ 1824 1825#if defined(CONFIG_INET) && defined(CONFIG_BPF_SYSCALL) 1826void bpf_sk_reuseport_detach(struct sock *sk); 1827int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map, void *key, 1828 void *value); 1829int bpf_fd_reuseport_array_update_elem(struct bpf_map *map, void *key, 1830 void *value, u64 map_flags); 1831#else 1832static inline void bpf_sk_reuseport_detach(struct sock *sk) 1833{ 1834} 1835 1836#ifdef CONFIG_BPF_SYSCALL 1837static inline int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map, 1838 void *key, void *value) 1839{ 1840 return -EOPNOTSUPP; 1841} 1842 1843static inline int bpf_fd_reuseport_array_update_elem(struct bpf_map *map, 1844 void *key, void *value, 1845 u64 map_flags) 1846{ 1847 return -EOPNOTSUPP; 1848} 1849#endif /* CONFIG_BPF_SYSCALL */ 1850#endif /* defined(CONFIG_INET) && defined(CONFIG_BPF_SYSCALL) */ 1851 1852/* verifier prototypes for helper functions called from eBPF programs */ 1853extern const struct bpf_func_proto bpf_map_lookup_elem_proto; 1854extern const struct bpf_func_proto bpf_map_update_elem_proto; 1855extern const struct bpf_func_proto bpf_map_delete_elem_proto; 1856extern const struct bpf_func_proto bpf_map_push_elem_proto; 1857extern const struct bpf_func_proto bpf_map_pop_elem_proto; 1858extern const struct bpf_func_proto bpf_map_peek_elem_proto; 1859 1860extern const struct bpf_func_proto bpf_get_prandom_u32_proto; 1861extern const struct bpf_func_proto bpf_get_smp_processor_id_proto; 1862extern const struct bpf_func_proto bpf_get_numa_node_id_proto; 1863extern const struct bpf_func_proto bpf_tail_call_proto; 1864extern const struct bpf_func_proto bpf_ktime_get_ns_proto; 1865extern const struct bpf_func_proto bpf_ktime_get_boot_ns_proto; 1866extern const struct bpf_func_proto bpf_get_current_pid_tgid_proto; 1867extern const struct bpf_func_proto bpf_get_current_uid_gid_proto; 1868extern const struct bpf_func_proto bpf_get_current_comm_proto; 1869extern const struct bpf_func_proto bpf_get_stackid_proto; 1870extern const struct bpf_func_proto bpf_get_stack_proto; 1871extern const struct bpf_func_proto bpf_get_task_stack_proto; 1872extern const struct bpf_func_proto bpf_get_stackid_proto_pe; 1873extern const struct bpf_func_proto bpf_get_stack_proto_pe; 1874extern const struct bpf_func_proto bpf_sock_map_update_proto; 1875extern const struct bpf_func_proto bpf_sock_hash_update_proto; 1876extern const struct bpf_func_proto bpf_get_current_cgroup_id_proto; 1877extern const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto; 1878extern const struct bpf_func_proto bpf_msg_redirect_hash_proto; 1879extern const struct bpf_func_proto bpf_msg_redirect_map_proto; 1880extern const struct bpf_func_proto bpf_sk_redirect_hash_proto; 1881extern const struct bpf_func_proto bpf_sk_redirect_map_proto; 1882extern const struct bpf_func_proto bpf_spin_lock_proto; 1883extern const struct bpf_func_proto bpf_spin_unlock_proto; 1884extern const struct bpf_func_proto bpf_get_local_storage_proto; 1885extern const struct bpf_func_proto bpf_strtol_proto; 1886extern const struct bpf_func_proto bpf_strtoul_proto; 1887extern const struct bpf_func_proto bpf_tcp_sock_proto; 1888extern const struct bpf_func_proto bpf_jiffies64_proto; 1889extern const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto; 1890extern const struct bpf_func_proto bpf_event_output_data_proto; 1891extern const struct bpf_func_proto bpf_ringbuf_output_proto; 1892extern const struct bpf_func_proto bpf_ringbuf_reserve_proto; 1893extern const struct bpf_func_proto bpf_ringbuf_submit_proto; 1894extern const struct bpf_func_proto bpf_ringbuf_discard_proto; 1895extern const struct bpf_func_proto bpf_ringbuf_query_proto; 1896extern const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto; 1897extern const struct bpf_func_proto bpf_skc_to_tcp_sock_proto; 1898extern const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto; 1899extern const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto; 1900extern const struct bpf_func_proto bpf_skc_to_udp6_sock_proto; 1901extern const struct bpf_func_proto bpf_copy_from_user_proto; 1902extern const struct bpf_func_proto bpf_snprintf_btf_proto; 1903extern const struct bpf_func_proto bpf_per_cpu_ptr_proto; 1904extern const struct bpf_func_proto bpf_this_cpu_ptr_proto; 1905extern const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto; 1906extern const struct bpf_func_proto bpf_sock_from_file_proto; 1907extern const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto; 1908 1909const struct bpf_func_proto *bpf_tracing_func_proto( 1910 enum bpf_func_id func_id, const struct bpf_prog *prog); 1911 1912const struct bpf_func_proto *tracing_prog_func_proto( 1913 enum bpf_func_id func_id, const struct bpf_prog *prog); 1914 1915/* Shared helpers among cBPF and eBPF. */ 1916void bpf_user_rnd_init_once(void); 1917u64 bpf_user_rnd_u32(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 1918u64 bpf_get_raw_cpu_id(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 1919 1920#if defined(CONFIG_NET) 1921bool bpf_sock_common_is_valid_access(int off, int size, 1922 enum bpf_access_type type, 1923 struct bpf_insn_access_aux *info); 1924bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 1925 struct bpf_insn_access_aux *info); 1926u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 1927 const struct bpf_insn *si, 1928 struct bpf_insn *insn_buf, 1929 struct bpf_prog *prog, 1930 u32 *target_size); 1931#else 1932static inline bool bpf_sock_common_is_valid_access(int off, int size, 1933 enum bpf_access_type type, 1934 struct bpf_insn_access_aux *info) 1935{ 1936 return false; 1937} 1938static inline bool bpf_sock_is_valid_access(int off, int size, 1939 enum bpf_access_type type, 1940 struct bpf_insn_access_aux *info) 1941{ 1942 return false; 1943} 1944static inline u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 1945 const struct bpf_insn *si, 1946 struct bpf_insn *insn_buf, 1947 struct bpf_prog *prog, 1948 u32 *target_size) 1949{ 1950 return 0; 1951} 1952#endif 1953 1954#ifdef CONFIG_INET 1955struct sk_reuseport_kern { 1956 struct sk_buff *skb; 1957 struct sock *sk; 1958 struct sock *selected_sk; 1959 void *data_end; 1960 u32 hash; 1961 u32 reuseport_id; 1962 bool bind_inany; 1963}; 1964bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 1965 struct bpf_insn_access_aux *info); 1966 1967u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 1968 const struct bpf_insn *si, 1969 struct bpf_insn *insn_buf, 1970 struct bpf_prog *prog, 1971 u32 *target_size); 1972 1973bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 1974 struct bpf_insn_access_aux *info); 1975 1976u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 1977 const struct bpf_insn *si, 1978 struct bpf_insn *insn_buf, 1979 struct bpf_prog *prog, 1980 u32 *target_size); 1981#else 1982static inline bool bpf_tcp_sock_is_valid_access(int off, int size, 1983 enum bpf_access_type type, 1984 struct bpf_insn_access_aux *info) 1985{ 1986 return false; 1987} 1988 1989static inline u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 1990 const struct bpf_insn *si, 1991 struct bpf_insn *insn_buf, 1992 struct bpf_prog *prog, 1993 u32 *target_size) 1994{ 1995 return 0; 1996} 1997static inline bool bpf_xdp_sock_is_valid_access(int off, int size, 1998 enum bpf_access_type type, 1999 struct bpf_insn_access_aux *info) 2000{ 2001 return false; 2002} 2003 2004static inline u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 2005 const struct bpf_insn *si, 2006 struct bpf_insn *insn_buf, 2007 struct bpf_prog *prog, 2008 u32 *target_size) 2009{ 2010 return 0; 2011} 2012#endif /* CONFIG_INET */ 2013 2014enum bpf_text_poke_type { 2015 BPF_MOD_CALL, 2016 BPF_MOD_JUMP, 2017}; 2018 2019int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, 2020 void *addr1, void *addr2); 2021 2022struct btf_id_set; 2023bool btf_id_set_contains(const struct btf_id_set *set, u32 id); 2024 2025#endif /* _LINUX_BPF_H */