1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com 2 * 3 * This program is free software; you can redistribute it and/or 4 * modify it under the terms of version 2 of the GNU General Public 5 * License as published by the Free Software Foundation. 6 */ 7 #ifndef _UAPI__LINUX_BPF_H__ 8 #define _UAPI__LINUX_BPF_H__ 9 10 #include <linux/types.h> 11 #include <linux/bpf_common.h> 12 13 /* Extended instruction set based on top of classic BPF */ 14 15 /* instruction classes */ 16 #define BPF_ALU64 0x07 /* alu mode in double word width */ 17 18 /* ld/ldx fields */ 19 #define BPF_DW 0x18 /* double word */ 20 #define BPF_XADD 0xc0 /* exclusive add */ 21 22 /* alu/jmp fields */ 23 #define BPF_MOV 0xb0 /* mov reg to reg */ 24 #define BPF_ARSH 0xc0 /* sign extending arithmetic shift right */ 25 26 /* change endianness of a register */ 27 #define BPF_END 0xd0 /* flags for endianness conversion: */ 28 #define BPF_TO_LE 0x00 /* convert to little-endian */ 29 #define BPF_TO_BE 0x08 /* convert to big-endian */ 30 #define BPF_FROM_LE BPF_TO_LE 31 #define BPF_FROM_BE BPF_TO_BE 32 33 #define BPF_JNE 0x50 /* jump != */ 34 #define BPF_JSGT 0x60 /* SGT is signed '>', GT in x86 */ 35 #define BPF_JSGE 0x70 /* SGE is signed '>=', GE in x86 */ 36 #define BPF_CALL 0x80 /* function call */ 37 #define BPF_EXIT 0x90 /* function return */ 38 39 /* Register numbers */ 40 enum { 41 BPF_REG_0 = 0, 42 BPF_REG_1, 43 BPF_REG_2, 44 BPF_REG_3, 45 BPF_REG_4, 46 BPF_REG_5, 47 BPF_REG_6, 48 BPF_REG_7, 49 BPF_REG_8, 50 BPF_REG_9, 51 BPF_REG_10, 52 __MAX_BPF_REG, 53 }; 54 55 /* BPF has 10 general purpose 64-bit registers and stack frame. */ 56 #define MAX_BPF_REG __MAX_BPF_REG 57 58 struct bpf_insn { 59 __u8 code; /* opcode */ 60 __u8 dst_reg:4; /* dest register */ 61 __u8 src_reg:4; /* source register */ 62 __s16 off; /* signed offset */ 63 __s32 imm; /* signed immediate constant */ 64 }; 65 66 /* BPF syscall commands, see bpf(2) man-page for details. */ 67 enum bpf_cmd { 68 BPF_MAP_CREATE, 69 BPF_MAP_LOOKUP_ELEM, 70 BPF_MAP_UPDATE_ELEM, 71 BPF_MAP_DELETE_ELEM, 72 BPF_MAP_GET_NEXT_KEY, 73 BPF_PROG_LOAD, 74 BPF_OBJ_PIN, 75 BPF_OBJ_GET, 76 BPF_PROG_ATTACH, 77 BPF_PROG_DETACH, 78 }; 79 80 enum bpf_map_type { 81 BPF_MAP_TYPE_UNSPEC, 82 BPF_MAP_TYPE_HASH, 83 BPF_MAP_TYPE_ARRAY, 84 BPF_MAP_TYPE_PROG_ARRAY, 85 BPF_MAP_TYPE_PERF_EVENT_ARRAY, 86 BPF_MAP_TYPE_PERCPU_HASH, 87 BPF_MAP_TYPE_PERCPU_ARRAY, 88 BPF_MAP_TYPE_STACK_TRACE, 89 BPF_MAP_TYPE_CGROUP_ARRAY, 90 BPF_MAP_TYPE_LRU_HASH, 91 BPF_MAP_TYPE_LRU_PERCPU_HASH, 92 }; 93 94 enum bpf_prog_type { 95 BPF_PROG_TYPE_UNSPEC, 96 BPF_PROG_TYPE_SOCKET_FILTER, 97 BPF_PROG_TYPE_KPROBE, 98 BPF_PROG_TYPE_SCHED_CLS, 99 BPF_PROG_TYPE_SCHED_ACT, 100 BPF_PROG_TYPE_TRACEPOINT, 101 BPF_PROG_TYPE_XDP, 102 BPF_PROG_TYPE_PERF_EVENT, 103 BPF_PROG_TYPE_CGROUP_SKB, 104 BPF_PROG_TYPE_CGROUP_SOCK, 105 BPF_PROG_TYPE_LWT_IN, 106 BPF_PROG_TYPE_LWT_OUT, 107 BPF_PROG_TYPE_LWT_XMIT, 108 }; 109 110 enum bpf_attach_type { 111 BPF_CGROUP_INET_INGRESS, 112 BPF_CGROUP_INET_EGRESS, 113 BPF_CGROUP_INET_SOCK_CREATE, 114 __MAX_BPF_ATTACH_TYPE 115 }; 116 117 #define MAX_BPF_ATTACH_TYPE __MAX_BPF_ATTACH_TYPE 118 119 /* If BPF_F_ALLOW_OVERRIDE flag is used in BPF_PROG_ATTACH command 120 * to the given target_fd cgroup the descendent cgroup will be able to 121 * override effective bpf program that was inherited from this cgroup 122 */ 123 #define BPF_F_ALLOW_OVERRIDE (1U << 0) 124 125 #define BPF_PSEUDO_MAP_FD 1 126 127 /* flags for BPF_MAP_UPDATE_ELEM command */ 128 #define BPF_ANY 0 /* create new element or update existing */ 129 #define BPF_NOEXIST 1 /* create new element if it didn't exist */ 130 #define BPF_EXIST 2 /* update existing element */ 131 132 #define BPF_F_NO_PREALLOC (1U << 0) 133 /* Instead of having one common LRU list in the 134 * BPF_MAP_TYPE_LRU_[PERCPU_]HASH map, use a percpu LRU list 135 * which can scale and perform better. 136 * Note, the LRU nodes (including free nodes) cannot be moved 137 * across different LRU lists. 138 */ 139 #define BPF_F_NO_COMMON_LRU (1U << 1) 140 141 union bpf_attr { 142 struct { /* anonymous struct used by BPF_MAP_CREATE command */ 143 __u32 map_type; /* one of enum bpf_map_type */ 144 __u32 key_size; /* size of key in bytes */ 145 __u32 value_size; /* size of value in bytes */ 146 __u32 max_entries; /* max number of entries in a map */ 147 __u32 map_flags; /* prealloc or not */ 148 }; 149 150 struct { /* anonymous struct used by BPF_MAP_*_ELEM commands */ 151 __u32 map_fd; 152 __aligned_u64 key; 153 union { 154 __aligned_u64 value; 155 __aligned_u64 next_key; 156 }; 157 __u64 flags; 158 }; 159 160 struct { /* anonymous struct used by BPF_PROG_LOAD command */ 161 __u32 prog_type; /* one of enum bpf_prog_type */ 162 __u32 insn_cnt; 163 __aligned_u64 insns; 164 __aligned_u64 license; 165 __u32 log_level; /* verbosity level of verifier */ 166 __u32 log_size; /* size of user buffer */ 167 __aligned_u64 log_buf; /* user supplied buffer */ 168 __u32 kern_version; /* checked when prog_type=kprobe */ 169 }; 170 171 struct { /* anonymous struct used by BPF_OBJ_* commands */ 172 __aligned_u64 pathname; 173 __u32 bpf_fd; 174 }; 175 176 struct { /* anonymous struct used by BPF_PROG_ATTACH/DETACH commands */ 177 __u32 target_fd; /* container object to attach to */ 178 __u32 attach_bpf_fd; /* eBPF program to attach */ 179 __u32 attach_type; 180 __u32 attach_flags; 181 }; 182 } __attribute__((aligned(8))); 183 184 /* BPF helper function descriptions: 185 * 186 * void *bpf_map_lookup_elem(&map, &key) 187 * Return: Map value or NULL 188 * 189 * int bpf_map_update_elem(&map, &key, &value, flags) 190 * Return: 0 on success or negative error 191 * 192 * int bpf_map_delete_elem(&map, &key) 193 * Return: 0 on success or negative error 194 * 195 * int bpf_probe_read(void *dst, int size, void *src) 196 * Return: 0 on success or negative error 197 * 198 * u64 bpf_ktime_get_ns(void) 199 * Return: current ktime 200 * 201 * int bpf_trace_printk(const char *fmt, int fmt_size, ...) 202 * Return: length of buffer written or negative error 203 * 204 * u32 bpf_prandom_u32(void) 205 * Return: random value 206 * 207 * u32 bpf_raw_smp_processor_id(void) 208 * Return: SMP processor ID 209 * 210 * int bpf_skb_store_bytes(skb, offset, from, len, flags) 211 * store bytes into packet 212 * @skb: pointer to skb 213 * @offset: offset within packet from skb->mac_header 214 * @from: pointer where to copy bytes from 215 * @len: number of bytes to store into packet 216 * @flags: bit 0 - if true, recompute skb->csum 217 * other bits - reserved 218 * Return: 0 on success or negative error 219 * 220 * int bpf_l3_csum_replace(skb, offset, from, to, flags) 221 * recompute IP checksum 222 * @skb: pointer to skb 223 * @offset: offset within packet where IP checksum is located 224 * @from: old value of header field 225 * @to: new value of header field 226 * @flags: bits 0-3 - size of header field 227 * other bits - reserved 228 * Return: 0 on success or negative error 229 * 230 * int bpf_l4_csum_replace(skb, offset, from, to, flags) 231 * recompute TCP/UDP checksum 232 * @skb: pointer to skb 233 * @offset: offset within packet where TCP/UDP checksum is located 234 * @from: old value of header field 235 * @to: new value of header field 236 * @flags: bits 0-3 - size of header field 237 * bit 4 - is pseudo header 238 * other bits - reserved 239 * Return: 0 on success or negative error 240 * 241 * int bpf_tail_call(ctx, prog_array_map, index) 242 * jump into another BPF program 243 * @ctx: context pointer passed to next program 244 * @prog_array_map: pointer to map which type is BPF_MAP_TYPE_PROG_ARRAY 245 * @index: index inside array that selects specific program to run 246 * Return: 0 on success or negative error 247 * 248 * int bpf_clone_redirect(skb, ifindex, flags) 249 * redirect to another netdev 250 * @skb: pointer to skb 251 * @ifindex: ifindex of the net device 252 * @flags: bit 0 - if set, redirect to ingress instead of egress 253 * other bits - reserved 254 * Return: 0 on success or negative error 255 * 256 * u64 bpf_get_current_pid_tgid(void) 257 * Return: current->tgid << 32 | current->pid 258 * 259 * u64 bpf_get_current_uid_gid(void) 260 * Return: current_gid << 32 | current_uid 261 * 262 * int bpf_get_current_comm(char *buf, int size_of_buf) 263 * stores current->comm into buf 264 * Return: 0 on success or negative error 265 * 266 * u32 bpf_get_cgroup_classid(skb) 267 * retrieve a proc's classid 268 * @skb: pointer to skb 269 * Return: classid if != 0 270 * 271 * int bpf_skb_vlan_push(skb, vlan_proto, vlan_tci) 272 * Return: 0 on success or negative error 273 * 274 * int bpf_skb_vlan_pop(skb) 275 * Return: 0 on success or negative error 276 * 277 * int bpf_skb_get_tunnel_key(skb, key, size, flags) 278 * int bpf_skb_set_tunnel_key(skb, key, size, flags) 279 * retrieve or populate tunnel metadata 280 * @skb: pointer to skb 281 * @key: pointer to 'struct bpf_tunnel_key' 282 * @size: size of 'struct bpf_tunnel_key' 283 * @flags: room for future extensions 284 * Return: 0 on success or negative error 285 * 286 * u64 bpf_perf_event_read(&map, index) 287 * Return: Number events read or error code 288 * 289 * int bpf_redirect(ifindex, flags) 290 * redirect to another netdev 291 * @ifindex: ifindex of the net device 292 * @flags: bit 0 - if set, redirect to ingress instead of egress 293 * other bits - reserved 294 * Return: TC_ACT_REDIRECT 295 * 296 * u32 bpf_get_route_realm(skb) 297 * retrieve a dst's tclassid 298 * @skb: pointer to skb 299 * Return: realm if != 0 300 * 301 * int bpf_perf_event_output(ctx, map, index, data, size) 302 * output perf raw sample 303 * @ctx: struct pt_regs* 304 * @map: pointer to perf_event_array map 305 * @index: index of event in the map 306 * @data: data on stack to be output as raw data 307 * @size: size of data 308 * Return: 0 on success or negative error 309 * 310 * int bpf_get_stackid(ctx, map, flags) 311 * walk user or kernel stack and return id 312 * @ctx: struct pt_regs* 313 * @map: pointer to stack_trace map 314 * @flags: bits 0-7 - numer of stack frames to skip 315 * bit 8 - collect user stack instead of kernel 316 * bit 9 - compare stacks by hash only 317 * bit 10 - if two different stacks hash into the same stackid 318 * discard old 319 * other bits - reserved 320 * Return: >= 0 stackid on success or negative error 321 * 322 * s64 bpf_csum_diff(from, from_size, to, to_size, seed) 323 * calculate csum diff 324 * @from: raw from buffer 325 * @from_size: length of from buffer 326 * @to: raw to buffer 327 * @to_size: length of to buffer 328 * @seed: optional seed 329 * Return: csum result or negative error code 330 * 331 * int bpf_skb_get_tunnel_opt(skb, opt, size) 332 * retrieve tunnel options metadata 333 * @skb: pointer to skb 334 * @opt: pointer to raw tunnel option data 335 * @size: size of @opt 336 * Return: option size 337 * 338 * int bpf_skb_set_tunnel_opt(skb, opt, size) 339 * populate tunnel options metadata 340 * @skb: pointer to skb 341 * @opt: pointer to raw tunnel option data 342 * @size: size of @opt 343 * Return: 0 on success or negative error 344 * 345 * int bpf_skb_change_proto(skb, proto, flags) 346 * Change protocol of the skb. Currently supported is v4 -> v6, 347 * v6 -> v4 transitions. The helper will also resize the skb. eBPF 348 * program is expected to fill the new headers via skb_store_bytes 349 * and lX_csum_replace. 350 * @skb: pointer to skb 351 * @proto: new skb->protocol type 352 * @flags: reserved 353 * Return: 0 on success or negative error 354 * 355 * int bpf_skb_change_type(skb, type) 356 * Change packet type of skb. 357 * @skb: pointer to skb 358 * @type: new skb->pkt_type type 359 * Return: 0 on success or negative error 360 * 361 * int bpf_skb_under_cgroup(skb, map, index) 362 * Check cgroup2 membership of skb 363 * @skb: pointer to skb 364 * @map: pointer to bpf_map in BPF_MAP_TYPE_CGROUP_ARRAY type 365 * @index: index of the cgroup in the bpf_map 366 * Return: 367 * == 0 skb failed the cgroup2 descendant test 368 * == 1 skb succeeded the cgroup2 descendant test 369 * < 0 error 370 * 371 * u32 bpf_get_hash_recalc(skb) 372 * Retrieve and possibly recalculate skb->hash. 373 * @skb: pointer to skb 374 * Return: hash 375 * 376 * u64 bpf_get_current_task(void) 377 * Returns current task_struct 378 * Return: current 379 * 380 * int bpf_probe_write_user(void *dst, void *src, int len) 381 * safely attempt to write to a location 382 * @dst: destination address in userspace 383 * @src: source address on stack 384 * @len: number of bytes to copy 385 * Return: 0 on success or negative error 386 * 387 * int bpf_current_task_under_cgroup(map, index) 388 * Check cgroup2 membership of current task 389 * @map: pointer to bpf_map in BPF_MAP_TYPE_CGROUP_ARRAY type 390 * @index: index of the cgroup in the bpf_map 391 * Return: 392 * == 0 current failed the cgroup2 descendant test 393 * == 1 current succeeded the cgroup2 descendant test 394 * < 0 error 395 * 396 * int bpf_skb_change_tail(skb, len, flags) 397 * The helper will resize the skb to the given new size, to be used f.e. 398 * with control messages. 399 * @skb: pointer to skb 400 * @len: new skb length 401 * @flags: reserved 402 * Return: 0 on success or negative error 403 * 404 * int bpf_skb_pull_data(skb, len) 405 * The helper will pull in non-linear data in case the skb is non-linear 406 * and not all of len are part of the linear section. Only needed for 407 * read/write with direct packet access. 408 * @skb: pointer to skb 409 * @len: len to make read/writeable 410 * Return: 0 on success or negative error 411 * 412 * s64 bpf_csum_update(skb, csum) 413 * Adds csum into skb->csum in case of CHECKSUM_COMPLETE. 414 * @skb: pointer to skb 415 * @csum: csum to add 416 * Return: csum on success or negative error 417 * 418 * void bpf_set_hash_invalid(skb) 419 * Invalidate current skb->hash. 420 * @skb: pointer to skb 421 * 422 * int bpf_get_numa_node_id() 423 * Return: Id of current NUMA node. 424 * 425 * int bpf_skb_change_head() 426 * Grows headroom of skb and adjusts MAC header offset accordingly. 427 * Will extends/reallocae as required automatically. 428 * May change skb data pointer and will thus invalidate any check 429 * performed for direct packet access. 430 * @skb: pointer to skb 431 * @len: length of header to be pushed in front 432 * @flags: Flags (unused for now) 433 * Return: 0 on success or negative error 434 * 435 * int bpf_xdp_adjust_head(xdp_md, delta) 436 * Adjust the xdp_md.data by delta 437 * @xdp_md: pointer to xdp_md 438 * @delta: An positive/negative integer to be added to xdp_md.data 439 * Return: 0 on success or negative on error 440 */ 441 #define __BPF_FUNC_MAPPER(FN) \ 442 FN(unspec), \ 443 FN(map_lookup_elem), \ 444 FN(map_update_elem), \ 445 FN(map_delete_elem), \ 446 FN(probe_read), \ 447 FN(ktime_get_ns), \ 448 FN(trace_printk), \ 449 FN(get_prandom_u32), \ 450 FN(get_smp_processor_id), \ 451 FN(skb_store_bytes), \ 452 FN(l3_csum_replace), \ 453 FN(l4_csum_replace), \ 454 FN(tail_call), \ 455 FN(clone_redirect), \ 456 FN(get_current_pid_tgid), \ 457 FN(get_current_uid_gid), \ 458 FN(get_current_comm), \ 459 FN(get_cgroup_classid), \ 460 FN(skb_vlan_push), \ 461 FN(skb_vlan_pop), \ 462 FN(skb_get_tunnel_key), \ 463 FN(skb_set_tunnel_key), \ 464 FN(perf_event_read), \ 465 FN(redirect), \ 466 FN(get_route_realm), \ 467 FN(perf_event_output), \ 468 FN(skb_load_bytes), \ 469 FN(get_stackid), \ 470 FN(csum_diff), \ 471 FN(skb_get_tunnel_opt), \ 472 FN(skb_set_tunnel_opt), \ 473 FN(skb_change_proto), \ 474 FN(skb_change_type), \ 475 FN(skb_under_cgroup), \ 476 FN(get_hash_recalc), \ 477 FN(get_current_task), \ 478 FN(probe_write_user), \ 479 FN(current_task_under_cgroup), \ 480 FN(skb_change_tail), \ 481 FN(skb_pull_data), \ 482 FN(csum_update), \ 483 FN(set_hash_invalid), \ 484 FN(get_numa_node_id), \ 485 FN(skb_change_head), \ 486 FN(xdp_adjust_head), 487 488 /* integer value in 'imm' field of BPF_CALL instruction selects which helper 489 * function eBPF program intends to call 490 */ 491 #define __BPF_ENUM_FN(x) BPF_FUNC_ ## x 492 enum bpf_func_id { 493 __BPF_FUNC_MAPPER(__BPF_ENUM_FN) 494 __BPF_FUNC_MAX_ID, 495 }; 496 #undef __BPF_ENUM_FN 497 498 /* All flags used by eBPF helper functions, placed here. */ 499 500 /* BPF_FUNC_skb_store_bytes flags. */ 501 #define BPF_F_RECOMPUTE_CSUM (1ULL << 0) 502 #define BPF_F_INVALIDATE_HASH (1ULL << 1) 503 504 /* BPF_FUNC_l3_csum_replace and BPF_FUNC_l4_csum_replace flags. 505 * First 4 bits are for passing the header field size. 506 */ 507 #define BPF_F_HDR_FIELD_MASK 0xfULL 508 509 /* BPF_FUNC_l4_csum_replace flags. */ 510 #define BPF_F_PSEUDO_HDR (1ULL << 4) 511 #define BPF_F_MARK_MANGLED_0 (1ULL << 5) 512 513 /* BPF_FUNC_clone_redirect and BPF_FUNC_redirect flags. */ 514 #define BPF_F_INGRESS (1ULL << 0) 515 516 /* BPF_FUNC_skb_set_tunnel_key and BPF_FUNC_skb_get_tunnel_key flags. */ 517 #define BPF_F_TUNINFO_IPV6 (1ULL << 0) 518 519 /* BPF_FUNC_get_stackid flags. */ 520 #define BPF_F_SKIP_FIELD_MASK 0xffULL 521 #define BPF_F_USER_STACK (1ULL << 8) 522 #define BPF_F_FAST_STACK_CMP (1ULL << 9) 523 #define BPF_F_REUSE_STACKID (1ULL << 10) 524 525 /* BPF_FUNC_skb_set_tunnel_key flags. */ 526 #define BPF_F_ZERO_CSUM_TX (1ULL << 1) 527 #define BPF_F_DONT_FRAGMENT (1ULL << 2) 528 529 /* BPF_FUNC_perf_event_output and BPF_FUNC_perf_event_read flags. */ 530 #define BPF_F_INDEX_MASK 0xffffffffULL 531 #define BPF_F_CURRENT_CPU BPF_F_INDEX_MASK 532 /* BPF_FUNC_perf_event_output for sk_buff input context. */ 533 #define BPF_F_CTXLEN_MASK (0xfffffULL << 32) 534 535 /* user accessible mirror of in-kernel sk_buff. 536 * new fields can only be added to the end of this structure 537 */ 538 struct __sk_buff { 539 __u32 len; 540 __u32 pkt_type; 541 __u32 mark; 542 __u32 queue_mapping; 543 __u32 protocol; 544 __u32 vlan_present; 545 __u32 vlan_tci; 546 __u32 vlan_proto; 547 __u32 priority; 548 __u32 ingress_ifindex; 549 __u32 ifindex; 550 __u32 tc_index; 551 __u32 cb[5]; 552 __u32 hash; 553 __u32 tc_classid; 554 __u32 data; 555 __u32 data_end; 556 }; 557 558 struct bpf_tunnel_key { 559 __u32 tunnel_id; 560 union { 561 __u32 remote_ipv4; 562 __u32 remote_ipv6[4]; 563 }; 564 __u8 tunnel_tos; 565 __u8 tunnel_ttl; 566 __u16 tunnel_ext; 567 __u32 tunnel_label; 568 }; 569 570 /* Generic BPF return codes which all BPF program types may support. 571 * The values are binary compatible with their TC_ACT_* counter-part to 572 * provide backwards compatibility with existing SCHED_CLS and SCHED_ACT 573 * programs. 574 * 575 * XDP is handled seprately, see XDP_*. 576 */ 577 enum bpf_ret_code { 578 BPF_OK = 0, 579 /* 1 reserved */ 580 BPF_DROP = 2, 581 /* 3-6 reserved */ 582 BPF_REDIRECT = 7, 583 /* >127 are reserved for prog type specific return codes */ 584 }; 585 586 struct bpf_sock { 587 __u32 bound_dev_if; 588 __u32 family; 589 __u32 type; 590 __u32 protocol; 591 }; 592 593 #define XDP_PACKET_HEADROOM 256 594 595 /* User return codes for XDP prog type. 596 * A valid XDP program must return one of these defined values. All other 597 * return codes are reserved for future use. Unknown return codes will result 598 * in packet drop. 599 */ 600 enum xdp_action { 601 XDP_ABORTED = 0, 602 XDP_DROP, 603 XDP_PASS, 604 XDP_TX, 605 }; 606 607 /* user accessible metadata for XDP packet hook 608 * new fields must be added to the end of this structure 609 */ 610 struct xdp_md { 611 __u32 data; 612 __u32 data_end; 613 }; 614 615 #endif /* _UAPI__LINUX_BPF_H__ */ 616