1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * Linux Socket Filter Data Structures
4 */
5 #ifndef __LINUX_FILTER_H__
6 #define __LINUX_FILTER_H__
7
8 #include <linux/atomic.h>
9 #include <linux/bpf.h>
10 #include <linux/refcount.h>
11 #include <linux/compat.h>
12 #include <linux/skbuff.h>
13 #include <linux/linkage.h>
14 #include <linux/printk.h>
15 #include <linux/workqueue.h>
16 #include <linux/sched.h>
17 #include <linux/sched/clock.h>
18 #include <linux/capability.h>
19 #include <linux/set_memory.h>
20 #include <linux/kallsyms.h>
21 #include <linux/if_vlan.h>
22 #include <linux/vmalloc.h>
23 #include <linux/sockptr.h>
24 #include <linux/u64_stats_sync.h>
25
26 #include <net/sch_generic.h>
27
28 #include <asm/byteorder.h>
29 #include <uapi/linux/filter.h>
30
31 struct sk_buff;
32 struct sock;
33 struct seccomp_data;
34 struct bpf_prog_aux;
35 struct xdp_rxq_info;
36 struct xdp_buff;
37 struct sock_reuseport;
38 struct ctl_table;
39 struct ctl_table_header;
40
41 /* ArgX, context and stack frame pointer register positions. Note,
42 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
43 * calls in BPF_CALL instruction.
44 */
45 #define BPF_REG_ARG1 BPF_REG_1
46 #define BPF_REG_ARG2 BPF_REG_2
47 #define BPF_REG_ARG3 BPF_REG_3
48 #define BPF_REG_ARG4 BPF_REG_4
49 #define BPF_REG_ARG5 BPF_REG_5
50 #define BPF_REG_CTX BPF_REG_6
51 #define BPF_REG_FP BPF_REG_10
52
53 /* Additional register mappings for converted user programs. */
54 #define BPF_REG_A BPF_REG_0
55 #define BPF_REG_X BPF_REG_7
56 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
57 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
58 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
59
60 /* Kernel hidden auxiliary/helper register. */
61 #define BPF_REG_AX MAX_BPF_REG
62 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1)
63 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
64
65 /* unused opcode to mark special call to bpf_tail_call() helper */
66 #define BPF_TAIL_CALL 0xf0
67
68 /* unused opcode to mark special load instruction. Same as BPF_ABS */
69 #define BPF_PROBE_MEM 0x20
70
71 /* unused opcode to mark special ldsx instruction. Same as BPF_IND */
72 #define BPF_PROBE_MEMSX 0x40
73
74 /* unused opcode to mark special load instruction. Same as BPF_MSH */
75 #define BPF_PROBE_MEM32 0xa0
76
77 /* unused opcode to mark special atomic instruction */
78 #define BPF_PROBE_ATOMIC 0xe0
79
80 /* unused opcode to mark special ldsx instruction. Same as BPF_NOSPEC */
81 #define BPF_PROBE_MEM32SX 0xc0
82
83 /* unused opcode to mark call to interpreter with arguments */
84 #define BPF_CALL_ARGS 0xe0
85
86 /* unused opcode to mark speculation barrier for mitigating
87 * Spectre v1 and v4
88 */
89 #define BPF_NOSPEC 0xc0
90
91 /* As per nm, we expose JITed images as text (code) section for
92 * kallsyms. That way, tools like perf can find it to match
93 * addresses.
94 */
95 #define BPF_SYM_ELF_TYPE 't'
96
97 /* BPF program can access up to 512 bytes of stack space. */
98 #define MAX_BPF_STACK 512
99
100 /* Helper macros for filter block array initializers. */
101
102 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
103
104 #define BPF_ALU64_REG_OFF(OP, DST, SRC, OFF) \
105 ((struct bpf_insn) { \
106 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
107 .dst_reg = DST, \
108 .src_reg = SRC, \
109 .off = OFF, \
110 .imm = 0 })
111
112 #define BPF_ALU64_REG(OP, DST, SRC) \
113 BPF_ALU64_REG_OFF(OP, DST, SRC, 0)
114
115 #define BPF_ALU32_REG_OFF(OP, DST, SRC, OFF) \
116 ((struct bpf_insn) { \
117 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
118 .dst_reg = DST, \
119 .src_reg = SRC, \
120 .off = OFF, \
121 .imm = 0 })
122
123 #define BPF_ALU32_REG(OP, DST, SRC) \
124 BPF_ALU32_REG_OFF(OP, DST, SRC, 0)
125
126 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
127
128 #define BPF_ALU64_IMM_OFF(OP, DST, IMM, OFF) \
129 ((struct bpf_insn) { \
130 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
131 .dst_reg = DST, \
132 .src_reg = 0, \
133 .off = OFF, \
134 .imm = IMM })
135 #define BPF_ALU64_IMM(OP, DST, IMM) \
136 BPF_ALU64_IMM_OFF(OP, DST, IMM, 0)
137
138 #define BPF_ALU32_IMM_OFF(OP, DST, IMM, OFF) \
139 ((struct bpf_insn) { \
140 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
141 .dst_reg = DST, \
142 .src_reg = 0, \
143 .off = OFF, \
144 .imm = IMM })
145 #define BPF_ALU32_IMM(OP, DST, IMM) \
146 BPF_ALU32_IMM_OFF(OP, DST, IMM, 0)
147
148 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
149
150 #define BPF_ENDIAN(TYPE, DST, LEN) \
151 ((struct bpf_insn) { \
152 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
153 .dst_reg = DST, \
154 .src_reg = 0, \
155 .off = 0, \
156 .imm = LEN })
157
158 /* Byte Swap, bswap16/32/64 */
159
160 #define BPF_BSWAP(DST, LEN) \
161 ((struct bpf_insn) { \
162 .code = BPF_ALU64 | BPF_END | BPF_SRC(BPF_TO_LE), \
163 .dst_reg = DST, \
164 .src_reg = 0, \
165 .off = 0, \
166 .imm = LEN })
167
168 /* Short form of mov, dst_reg = src_reg */
169
170 #define BPF_MOV64_REG(DST, SRC) \
171 ((struct bpf_insn) { \
172 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
173 .dst_reg = DST, \
174 .src_reg = SRC, \
175 .off = 0, \
176 .imm = 0 })
177
178 #define BPF_MOV32_REG(DST, SRC) \
179 ((struct bpf_insn) { \
180 .code = BPF_ALU | BPF_MOV | BPF_X, \
181 .dst_reg = DST, \
182 .src_reg = SRC, \
183 .off = 0, \
184 .imm = 0 })
185
186 /* Special (internal-only) form of mov, used to resolve per-CPU addrs:
187 * dst_reg = src_reg + <percpu_base_off>
188 * BPF_ADDR_PERCPU is used as a special insn->off value.
189 */
190 #define BPF_ADDR_PERCPU (-1)
191
192 #define BPF_MOV64_PERCPU_REG(DST, SRC) \
193 ((struct bpf_insn) { \
194 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
195 .dst_reg = DST, \
196 .src_reg = SRC, \
197 .off = BPF_ADDR_PERCPU, \
198 .imm = 0 })
199
insn_is_mov_percpu_addr(const struct bpf_insn * insn)200 static inline bool insn_is_mov_percpu_addr(const struct bpf_insn *insn)
201 {
202 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) && insn->off == BPF_ADDR_PERCPU;
203 }
204
205 /* Short form of mov, dst_reg = imm32 */
206
207 #define BPF_MOV64_IMM(DST, IMM) \
208 ((struct bpf_insn) { \
209 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
210 .dst_reg = DST, \
211 .src_reg = 0, \
212 .off = 0, \
213 .imm = IMM })
214
215 #define BPF_MOV32_IMM(DST, IMM) \
216 ((struct bpf_insn) { \
217 .code = BPF_ALU | BPF_MOV | BPF_K, \
218 .dst_reg = DST, \
219 .src_reg = 0, \
220 .off = 0, \
221 .imm = IMM })
222
223 /* Short form of movsx, dst_reg = (s8,s16,s32)src_reg */
224
225 #define BPF_MOVSX64_REG(DST, SRC, OFF) \
226 ((struct bpf_insn) { \
227 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
228 .dst_reg = DST, \
229 .src_reg = SRC, \
230 .off = OFF, \
231 .imm = 0 })
232
233 #define BPF_MOVSX32_REG(DST, SRC, OFF) \
234 ((struct bpf_insn) { \
235 .code = BPF_ALU | BPF_MOV | BPF_X, \
236 .dst_reg = DST, \
237 .src_reg = SRC, \
238 .off = OFF, \
239 .imm = 0 })
240
241 /* Special form of mov32, used for doing explicit zero extension on dst. */
242 #define BPF_ZEXT_REG(DST) \
243 ((struct bpf_insn) { \
244 .code = BPF_ALU | BPF_MOV | BPF_X, \
245 .dst_reg = DST, \
246 .src_reg = DST, \
247 .off = 0, \
248 .imm = 1 })
249
insn_is_zext(const struct bpf_insn * insn)250 static inline bool insn_is_zext(const struct bpf_insn *insn)
251 {
252 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1;
253 }
254
255 /* addr_space_cast from as(0) to as(1) is for converting bpf arena pointers
256 * to pointers in user vma.
257 */
insn_is_cast_user(const struct bpf_insn * insn)258 static inline bool insn_is_cast_user(const struct bpf_insn *insn)
259 {
260 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) &&
261 insn->off == BPF_ADDR_SPACE_CAST &&
262 insn->imm == 1U << 16;
263 }
264
265 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
266 #define BPF_LD_IMM64(DST, IMM) \
267 BPF_LD_IMM64_RAW(DST, 0, IMM)
268
269 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
270 ((struct bpf_insn) { \
271 .code = BPF_LD | BPF_DW | BPF_IMM, \
272 .dst_reg = DST, \
273 .src_reg = SRC, \
274 .off = 0, \
275 .imm = (__u32) (IMM) }), \
276 ((struct bpf_insn) { \
277 .code = 0, /* zero is reserved opcode */ \
278 .dst_reg = 0, \
279 .src_reg = 0, \
280 .off = 0, \
281 .imm = ((__u64) (IMM)) >> 32 })
282
283 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
284 #define BPF_LD_MAP_FD(DST, MAP_FD) \
285 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
286
287 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
288
289 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
290 ((struct bpf_insn) { \
291 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
292 .dst_reg = DST, \
293 .src_reg = SRC, \
294 .off = 0, \
295 .imm = IMM })
296
297 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
298 ((struct bpf_insn) { \
299 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
300 .dst_reg = DST, \
301 .src_reg = SRC, \
302 .off = 0, \
303 .imm = IMM })
304
305 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
306
307 #define BPF_LD_ABS(SIZE, IMM) \
308 ((struct bpf_insn) { \
309 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
310 .dst_reg = 0, \
311 .src_reg = 0, \
312 .off = 0, \
313 .imm = IMM })
314
315 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
316
317 #define BPF_LD_IND(SIZE, SRC, IMM) \
318 ((struct bpf_insn) { \
319 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
320 .dst_reg = 0, \
321 .src_reg = SRC, \
322 .off = 0, \
323 .imm = IMM })
324
325 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
326
327 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
328 ((struct bpf_insn) { \
329 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
330 .dst_reg = DST, \
331 .src_reg = SRC, \
332 .off = OFF, \
333 .imm = 0 })
334
335 /* Memory load, dst_reg = *(signed size *) (src_reg + off16) */
336
337 #define BPF_LDX_MEMSX(SIZE, DST, SRC, OFF) \
338 ((struct bpf_insn) { \
339 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEMSX, \
340 .dst_reg = DST, \
341 .src_reg = SRC, \
342 .off = OFF, \
343 .imm = 0 })
344
345 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
346
347 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
348 ((struct bpf_insn) { \
349 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
350 .dst_reg = DST, \
351 .src_reg = SRC, \
352 .off = OFF, \
353 .imm = 0 })
354
355
356 /*
357 * Atomic operations:
358 *
359 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg
360 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg
361 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg
362 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg
363 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg);
364 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg);
365 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg);
366 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg);
367 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg)
368 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg)
369 * BPF_LOAD_ACQ dst_reg = smp_load_acquire(src_reg + off16)
370 * BPF_STORE_REL smp_store_release(dst_reg + off16, src_reg)
371 */
372
373 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \
374 ((struct bpf_insn) { \
375 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \
376 .dst_reg = DST, \
377 .src_reg = SRC, \
378 .off = OFF, \
379 .imm = OP })
380
381 /* Legacy alias */
382 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF)
383
384 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
385
386 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
387 ((struct bpf_insn) { \
388 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
389 .dst_reg = DST, \
390 .src_reg = 0, \
391 .off = OFF, \
392 .imm = IMM })
393
394 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
395
396 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
397 ((struct bpf_insn) { \
398 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
399 .dst_reg = DST, \
400 .src_reg = SRC, \
401 .off = OFF, \
402 .imm = 0 })
403
404 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
405
406 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
407 ((struct bpf_insn) { \
408 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
409 .dst_reg = DST, \
410 .src_reg = 0, \
411 .off = OFF, \
412 .imm = IMM })
413
414 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
415
416 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
417 ((struct bpf_insn) { \
418 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
419 .dst_reg = DST, \
420 .src_reg = SRC, \
421 .off = OFF, \
422 .imm = 0 })
423
424 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
425
426 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
427 ((struct bpf_insn) { \
428 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
429 .dst_reg = DST, \
430 .src_reg = 0, \
431 .off = OFF, \
432 .imm = IMM })
433
434 /* Unconditional jumps, goto pc + off16 */
435
436 #define BPF_JMP_A(OFF) \
437 ((struct bpf_insn) { \
438 .code = BPF_JMP | BPF_JA, \
439 .dst_reg = 0, \
440 .src_reg = 0, \
441 .off = OFF, \
442 .imm = 0 })
443
444 /* Unconditional jumps, gotol pc + imm32 */
445
446 #define BPF_JMP32_A(IMM) \
447 ((struct bpf_insn) { \
448 .code = BPF_JMP32 | BPF_JA, \
449 .dst_reg = 0, \
450 .src_reg = 0, \
451 .off = 0, \
452 .imm = IMM })
453
454 /* Relative call */
455
456 #define BPF_CALL_REL(TGT) \
457 ((struct bpf_insn) { \
458 .code = BPF_JMP | BPF_CALL, \
459 .dst_reg = 0, \
460 .src_reg = BPF_PSEUDO_CALL, \
461 .off = 0, \
462 .imm = TGT })
463
464 /* Convert function address to BPF immediate */
465
466 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base)
467
468 #define BPF_EMIT_CALL(FUNC) \
469 ((struct bpf_insn) { \
470 .code = BPF_JMP | BPF_CALL, \
471 .dst_reg = 0, \
472 .src_reg = 0, \
473 .off = 0, \
474 .imm = BPF_CALL_IMM(FUNC) })
475
476 /* Kfunc call */
477
478 #define BPF_CALL_KFUNC(OFF, IMM) \
479 ((struct bpf_insn) { \
480 .code = BPF_JMP | BPF_CALL, \
481 .dst_reg = 0, \
482 .src_reg = BPF_PSEUDO_KFUNC_CALL, \
483 .off = OFF, \
484 .imm = IMM })
485
486 /* Raw code statement block */
487
488 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
489 ((struct bpf_insn) { \
490 .code = CODE, \
491 .dst_reg = DST, \
492 .src_reg = SRC, \
493 .off = OFF, \
494 .imm = IMM })
495
496 /* Program exit */
497
498 #define BPF_EXIT_INSN() \
499 ((struct bpf_insn) { \
500 .code = BPF_JMP | BPF_EXIT, \
501 .dst_reg = 0, \
502 .src_reg = 0, \
503 .off = 0, \
504 .imm = 0 })
505
506 /* Speculation barrier */
507
508 #define BPF_ST_NOSPEC() \
509 ((struct bpf_insn) { \
510 .code = BPF_ST | BPF_NOSPEC, \
511 .dst_reg = 0, \
512 .src_reg = 0, \
513 .off = 0, \
514 .imm = 0 })
515
516 /* Internal classic blocks for direct assignment */
517
518 #define __BPF_STMT(CODE, K) \
519 ((struct sock_filter) BPF_STMT(CODE, K))
520
521 #define __BPF_JUMP(CODE, K, JT, JF) \
522 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
523
524 #define bytes_to_bpf_size(bytes) \
525 ({ \
526 int bpf_size = -EINVAL; \
527 \
528 if (bytes == sizeof(u8)) \
529 bpf_size = BPF_B; \
530 else if (bytes == sizeof(u16)) \
531 bpf_size = BPF_H; \
532 else if (bytes == sizeof(u32)) \
533 bpf_size = BPF_W; \
534 else if (bytes == sizeof(u64)) \
535 bpf_size = BPF_DW; \
536 \
537 bpf_size; \
538 })
539
540 #define bpf_size_to_bytes(bpf_size) \
541 ({ \
542 int bytes = -EINVAL; \
543 \
544 if (bpf_size == BPF_B) \
545 bytes = sizeof(u8); \
546 else if (bpf_size == BPF_H) \
547 bytes = sizeof(u16); \
548 else if (bpf_size == BPF_W) \
549 bytes = sizeof(u32); \
550 else if (bpf_size == BPF_DW) \
551 bytes = sizeof(u64); \
552 \
553 bytes; \
554 })
555
556 #define BPF_SIZEOF(type) \
557 ({ \
558 const int __size = bytes_to_bpf_size(sizeof(type)); \
559 BUILD_BUG_ON(__size < 0); \
560 __size; \
561 })
562
563 #define BPF_FIELD_SIZEOF(type, field) \
564 ({ \
565 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
566 BUILD_BUG_ON(__size < 0); \
567 __size; \
568 })
569
570 #define BPF_LDST_BYTES(insn) \
571 ({ \
572 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
573 WARN_ON(__size < 0); \
574 __size; \
575 })
576
577 #define __BPF_MAP_0(m, v, ...) v
578 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
579 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
580 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
581 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
582 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
583
584 #define __BPF_REG_0(...) __BPF_PAD(5)
585 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
586 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
587 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
588 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
589 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
590
591 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
592 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
593
594 #define __BPF_CAST(t, a) \
595 (__force t) \
596 (__force \
597 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
598 (unsigned long)0, (t)0))) a
599 #define __BPF_V void
600 #define __BPF_N
601
602 #define __BPF_DECL_ARGS(t, a) t a
603 #define __BPF_DECL_REGS(t, a) u64 a
604
605 #define __BPF_PAD(n) \
606 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
607 u64, __ur_3, u64, __ur_4, u64, __ur_5)
608
609 #define BPF_CALL_x(x, attr, name, ...) \
610 static __always_inline \
611 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
612 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
613 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
614 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
615 { \
616 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
617 } \
618 static __always_inline \
619 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
620
621 #define __NOATTR
622 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, __NOATTR, name, __VA_ARGS__)
623 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, __NOATTR, name, __VA_ARGS__)
624 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, __NOATTR, name, __VA_ARGS__)
625 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__)
626 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, __NOATTR, name, __VA_ARGS__)
627 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, __NOATTR, name, __VA_ARGS__)
628
629 #define NOTRACE_BPF_CALL_1(name, ...) BPF_CALL_x(1, notrace, name, __VA_ARGS__)
630
631 #define bpf_ctx_range(TYPE, MEMBER) \
632 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
633 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
634 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
635 #if BITS_PER_LONG == 64
636 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
637 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
638 #else
639 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
640 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
641 #endif /* BITS_PER_LONG == 64 */
642
643 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
644 ({ \
645 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \
646 *(PTR_SIZE) = (SIZE); \
647 offsetof(TYPE, MEMBER); \
648 })
649
650 /* A struct sock_filter is architecture independent. */
651 struct compat_sock_fprog {
652 u16 len;
653 compat_uptr_t filter; /* struct sock_filter * */
654 };
655
656 struct sock_fprog_kern {
657 u16 len;
658 struct sock_filter *filter;
659 };
660
661 /* Some arches need doubleword alignment for their instructions and/or data */
662 #define BPF_IMAGE_ALIGNMENT 8
663
664 struct bpf_binary_header {
665 u32 size;
666 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT);
667 };
668
669 struct bpf_prog_stats {
670 u64_stats_t cnt;
671 u64_stats_t nsecs;
672 u64_stats_t misses;
673 struct u64_stats_sync syncp;
674 } __aligned(2 * sizeof(u64));
675
676 struct bpf_timed_may_goto {
677 u64 count;
678 u64 timestamp;
679 };
680
681 struct sk_filter {
682 refcount_t refcnt;
683 struct rcu_head rcu;
684 struct bpf_prog *prog;
685 };
686
687 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
688
689 extern struct mutex nf_conn_btf_access_lock;
690 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
691 const struct bpf_reg_state *reg,
692 int off, int size);
693
694 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx,
695 const struct bpf_insn *insnsi,
696 unsigned int (*bpf_func)(const void *,
697 const struct bpf_insn *));
698
__bpf_prog_run(const struct bpf_prog * prog,const void * ctx,bpf_dispatcher_fn dfunc)699 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog,
700 const void *ctx,
701 bpf_dispatcher_fn dfunc)
702 {
703 u32 ret;
704
705 cant_migrate();
706 if (static_branch_unlikely(&bpf_stats_enabled_key)) {
707 struct bpf_prog_stats *stats;
708 u64 duration, start = sched_clock();
709 unsigned long flags;
710
711 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
712
713 duration = sched_clock() - start;
714 if (likely(prog->stats)) {
715 stats = this_cpu_ptr(prog->stats);
716 flags = u64_stats_update_begin_irqsave(&stats->syncp);
717 u64_stats_inc(&stats->cnt);
718 u64_stats_add(&stats->nsecs, duration);
719 u64_stats_update_end_irqrestore(&stats->syncp, flags);
720 }
721 } else {
722 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
723 }
724 return ret;
725 }
726
bpf_prog_run(const struct bpf_prog * prog,const void * ctx)727 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx)
728 {
729 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func);
730 }
731
732 /*
733 * Use in preemptible and therefore migratable context to make sure that
734 * the execution of the BPF program runs on one CPU.
735 *
736 * This uses migrate_disable/enable() explicitly to document that the
737 * invocation of a BPF program does not require reentrancy protection
738 * against a BPF program which is invoked from a preempting task.
739 */
bpf_prog_run_pin_on_cpu(const struct bpf_prog * prog,const void * ctx)740 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
741 const void *ctx)
742 {
743 u32 ret;
744
745 migrate_disable();
746 ret = bpf_prog_run(prog, ctx);
747 migrate_enable();
748 return ret;
749 }
750
751 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
752
753 struct bpf_skb_data_end {
754 struct qdisc_skb_cb qdisc_cb;
755 void *data_meta;
756 void *data_end;
757 };
758
759 struct bpf_nh_params {
760 u32 nh_family;
761 union {
762 u32 ipv4_nh;
763 struct in6_addr ipv6_nh;
764 };
765 };
766
767 /* flags for bpf_redirect_info kern_flags */
768 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
769 #define BPF_RI_F_RI_INIT BIT(1)
770 #define BPF_RI_F_CPU_MAP_INIT BIT(2)
771 #define BPF_RI_F_DEV_MAP_INIT BIT(3)
772 #define BPF_RI_F_XSK_MAP_INIT BIT(4)
773
774 struct bpf_redirect_info {
775 u64 tgt_index;
776 void *tgt_value;
777 struct bpf_map *map;
778 u32 flags;
779 u32 map_id;
780 enum bpf_map_type map_type;
781 struct bpf_nh_params nh;
782 u32 kern_flags;
783 };
784
785 struct bpf_net_context {
786 struct bpf_redirect_info ri;
787 struct list_head cpu_map_flush_list;
788 struct list_head dev_map_flush_list;
789 struct list_head xskmap_map_flush_list;
790 };
791
bpf_net_ctx_set(struct bpf_net_context * bpf_net_ctx)792 static inline struct bpf_net_context *bpf_net_ctx_set(struct bpf_net_context *bpf_net_ctx)
793 {
794 struct task_struct *tsk = current;
795
796 if (tsk->bpf_net_context != NULL)
797 return NULL;
798 bpf_net_ctx->ri.kern_flags = 0;
799
800 tsk->bpf_net_context = bpf_net_ctx;
801 return bpf_net_ctx;
802 }
803
bpf_net_ctx_clear(struct bpf_net_context * bpf_net_ctx)804 static inline void bpf_net_ctx_clear(struct bpf_net_context *bpf_net_ctx)
805 {
806 if (bpf_net_ctx)
807 current->bpf_net_context = NULL;
808 }
809
bpf_net_ctx_get(void)810 static inline struct bpf_net_context *bpf_net_ctx_get(void)
811 {
812 return current->bpf_net_context;
813 }
814
bpf_net_ctx_get_ri(void)815 static inline struct bpf_redirect_info *bpf_net_ctx_get_ri(void)
816 {
817 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
818
819 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_RI_INIT)) {
820 memset(&bpf_net_ctx->ri, 0, offsetof(struct bpf_net_context, ri.nh));
821 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_RI_INIT;
822 }
823
824 return &bpf_net_ctx->ri;
825 }
826
bpf_net_ctx_get_cpu_map_flush_list(void)827 static inline struct list_head *bpf_net_ctx_get_cpu_map_flush_list(void)
828 {
829 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
830
831 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_CPU_MAP_INIT)) {
832 INIT_LIST_HEAD(&bpf_net_ctx->cpu_map_flush_list);
833 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_CPU_MAP_INIT;
834 }
835
836 return &bpf_net_ctx->cpu_map_flush_list;
837 }
838
bpf_net_ctx_get_dev_flush_list(void)839 static inline struct list_head *bpf_net_ctx_get_dev_flush_list(void)
840 {
841 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
842
843 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_DEV_MAP_INIT)) {
844 INIT_LIST_HEAD(&bpf_net_ctx->dev_map_flush_list);
845 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_DEV_MAP_INIT;
846 }
847
848 return &bpf_net_ctx->dev_map_flush_list;
849 }
850
bpf_net_ctx_get_xskmap_flush_list(void)851 static inline struct list_head *bpf_net_ctx_get_xskmap_flush_list(void)
852 {
853 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
854
855 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_XSK_MAP_INIT)) {
856 INIT_LIST_HEAD(&bpf_net_ctx->xskmap_map_flush_list);
857 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_XSK_MAP_INIT;
858 }
859
860 return &bpf_net_ctx->xskmap_map_flush_list;
861 }
862
bpf_net_ctx_get_all_used_flush_lists(struct list_head ** lh_map,struct list_head ** lh_dev,struct list_head ** lh_xsk)863 static inline void bpf_net_ctx_get_all_used_flush_lists(struct list_head **lh_map,
864 struct list_head **lh_dev,
865 struct list_head **lh_xsk)
866 {
867 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
868 u32 kern_flags = bpf_net_ctx->ri.kern_flags;
869 struct list_head *lh;
870
871 *lh_map = *lh_dev = *lh_xsk = NULL;
872
873 if (!IS_ENABLED(CONFIG_BPF_SYSCALL))
874 return;
875
876 lh = &bpf_net_ctx->dev_map_flush_list;
877 if (kern_flags & BPF_RI_F_DEV_MAP_INIT && !list_empty(lh))
878 *lh_dev = lh;
879
880 lh = &bpf_net_ctx->cpu_map_flush_list;
881 if (kern_flags & BPF_RI_F_CPU_MAP_INIT && !list_empty(lh))
882 *lh_map = lh;
883
884 lh = &bpf_net_ctx->xskmap_map_flush_list;
885 if (IS_ENABLED(CONFIG_XDP_SOCKETS) &&
886 kern_flags & BPF_RI_F_XSK_MAP_INIT && !list_empty(lh))
887 *lh_xsk = lh;
888 }
889
890 /* Compute the linear packet data range [data, data_end) which
891 * will be accessed by various program types (cls_bpf, act_bpf,
892 * lwt, ...). Subsystems allowing direct data access must (!)
893 * ensure that cb[] area can be written to when BPF program is
894 * invoked (otherwise cb[] save/restore is necessary).
895 */
bpf_compute_data_pointers(struct sk_buff * skb)896 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
897 {
898 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
899
900 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
901 cb->data_meta = skb->data - skb_metadata_len(skb);
902 cb->data_end = skb->data + skb_headlen(skb);
903 }
904
bpf_prog_run_data_pointers(const struct bpf_prog * prog,struct sk_buff * skb)905 static inline int bpf_prog_run_data_pointers(
906 const struct bpf_prog *prog,
907 struct sk_buff *skb)
908 {
909 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
910 void *save_data_meta, *save_data_end;
911 int res;
912
913 save_data_meta = cb->data_meta;
914 save_data_end = cb->data_end;
915
916 bpf_compute_data_pointers(skb);
917 res = bpf_prog_run(prog, skb);
918
919 cb->data_meta = save_data_meta;
920 cb->data_end = save_data_end;
921
922 return res;
923 }
924
925 /* Similar to bpf_compute_data_pointers(), except that save orginal
926 * data in cb->data and cb->meta_data for restore.
927 */
bpf_compute_and_save_data_end(struct sk_buff * skb,void ** saved_data_end)928 static inline void bpf_compute_and_save_data_end(
929 struct sk_buff *skb, void **saved_data_end)
930 {
931 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
932
933 *saved_data_end = cb->data_end;
934 cb->data_end = skb->data + skb_headlen(skb);
935 }
936
937 /* Restore data saved by bpf_compute_and_save_data_end(). */
bpf_restore_data_end(struct sk_buff * skb,void * saved_data_end)938 static inline void bpf_restore_data_end(
939 struct sk_buff *skb, void *saved_data_end)
940 {
941 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
942
943 cb->data_end = saved_data_end;
944 }
945
bpf_skb_cb(const struct sk_buff * skb)946 static inline u8 *bpf_skb_cb(const struct sk_buff *skb)
947 {
948 /* eBPF programs may read/write skb->cb[] area to transfer meta
949 * data between tail calls. Since this also needs to work with
950 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
951 *
952 * In some socket filter cases, the cb unfortunately needs to be
953 * saved/restored so that protocol specific skb->cb[] data won't
954 * be lost. In any case, due to unpriviledged eBPF programs
955 * attached to sockets, we need to clear the bpf_skb_cb() area
956 * to not leak previous contents to user space.
957 */
958 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
959 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) !=
960 sizeof_field(struct qdisc_skb_cb, data));
961
962 return qdisc_skb_cb(skb)->data;
963 }
964
965 /* Must be invoked with migration disabled */
__bpf_prog_run_save_cb(const struct bpf_prog * prog,const void * ctx)966 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
967 const void *ctx)
968 {
969 const struct sk_buff *skb = ctx;
970 u8 *cb_data = bpf_skb_cb(skb);
971 u8 cb_saved[BPF_SKB_CB_LEN];
972 u32 res;
973
974 if (unlikely(prog->cb_access)) {
975 memcpy(cb_saved, cb_data, sizeof(cb_saved));
976 memset(cb_data, 0, sizeof(cb_saved));
977 }
978
979 res = bpf_prog_run(prog, skb);
980
981 if (unlikely(prog->cb_access))
982 memcpy(cb_data, cb_saved, sizeof(cb_saved));
983
984 return res;
985 }
986
bpf_prog_run_save_cb(const struct bpf_prog * prog,struct sk_buff * skb)987 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
988 struct sk_buff *skb)
989 {
990 u32 res;
991
992 migrate_disable();
993 res = __bpf_prog_run_save_cb(prog, skb);
994 migrate_enable();
995 return res;
996 }
997
bpf_prog_run_clear_cb(const struct bpf_prog * prog,struct sk_buff * skb)998 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
999 struct sk_buff *skb)
1000 {
1001 u8 *cb_data = bpf_skb_cb(skb);
1002 u32 res;
1003
1004 if (unlikely(prog->cb_access))
1005 memset(cb_data, 0, BPF_SKB_CB_LEN);
1006
1007 res = bpf_prog_run_pin_on_cpu(prog, skb);
1008 return res;
1009 }
1010
1011 DECLARE_BPF_DISPATCHER(xdp)
1012
1013 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
1014
1015 u32 xdp_master_redirect(struct xdp_buff *xdp);
1016
1017 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog);
1018
bpf_prog_insn_size(const struct bpf_prog * prog)1019 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
1020 {
1021 return prog->len * sizeof(struct bpf_insn);
1022 }
1023
bpf_prog_size(unsigned int proglen)1024 static inline unsigned int bpf_prog_size(unsigned int proglen)
1025 {
1026 return max(sizeof(struct bpf_prog),
1027 offsetof(struct bpf_prog, insns[proglen]));
1028 }
1029
bpf_prog_was_classic(const struct bpf_prog * prog)1030 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
1031 {
1032 /* When classic BPF programs have been loaded and the arch
1033 * does not have a classic BPF JIT (anymore), they have been
1034 * converted via bpf_migrate_filter() to eBPF and thus always
1035 * have an unspec program type.
1036 */
1037 return prog->type == BPF_PROG_TYPE_UNSPEC;
1038 }
1039
bpf_ctx_off_adjust_machine(u32 size)1040 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
1041 {
1042 const u32 size_machine = sizeof(unsigned long);
1043
1044 if (size > size_machine && size % size_machine == 0)
1045 size = size_machine;
1046
1047 return size;
1048 }
1049
1050 static inline bool
bpf_ctx_narrow_access_ok(u32 off,u32 size,u32 size_default)1051 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
1052 {
1053 return size <= size_default && (size & (size - 1)) == 0;
1054 }
1055
1056 static inline u8
bpf_ctx_narrow_access_offset(u32 off,u32 size,u32 size_default)1057 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
1058 {
1059 u8 access_off = off & (size_default - 1);
1060
1061 #ifdef __LITTLE_ENDIAN
1062 return access_off;
1063 #else
1064 return size_default - (access_off + size);
1065 #endif
1066 }
1067
1068 #define bpf_ctx_wide_access_ok(off, size, type, field) \
1069 (size == sizeof(__u64) && \
1070 off >= offsetof(type, field) && \
1071 off + sizeof(__u64) <= offsetofend(type, field) && \
1072 off % sizeof(__u64) == 0)
1073
1074 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
1075
bpf_prog_lock_ro(struct bpf_prog * fp)1076 static inline int __must_check bpf_prog_lock_ro(struct bpf_prog *fp)
1077 {
1078 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1079 if (!fp->jited) {
1080 set_vm_flush_reset_perms(fp);
1081 return set_memory_ro((unsigned long)fp, fp->pages);
1082 }
1083 #endif
1084 return 0;
1085 }
1086
1087 static inline int __must_check
bpf_jit_binary_lock_ro(struct bpf_binary_header * hdr)1088 bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
1089 {
1090 set_vm_flush_reset_perms(hdr);
1091 return set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT);
1092 }
1093
1094 enum skb_drop_reason
1095 sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
1096
sk_filter(struct sock * sk,struct sk_buff * skb)1097 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
1098 {
1099 enum skb_drop_reason drop_reason;
1100
1101 drop_reason = sk_filter_trim_cap(sk, skb, 1);
1102 return drop_reason ? -EPERM : 0;
1103 }
1104
1105 static inline enum skb_drop_reason
sk_filter_reason(struct sock * sk,struct sk_buff * skb)1106 sk_filter_reason(struct sock *sk, struct sk_buff *skb)
1107 {
1108 return sk_filter_trim_cap(sk, skb, 1);
1109 }
1110
1111 struct bpf_prog *__bpf_prog_select_runtime(struct bpf_verifier_env *env, struct bpf_prog *fp,
1112 int *err);
1113 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
1114 void bpf_prog_free(struct bpf_prog *fp);
1115
1116 bool bpf_opcode_in_insntable(u8 code);
1117
1118 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
1119 const u32 *insn_to_jit_off);
1120 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
1121 void bpf_prog_jit_attempt_done(struct bpf_prog *prog);
1122
1123 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
1124 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
1125 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
1126 gfp_t gfp_extra_flags);
1127 void __bpf_prog_free(struct bpf_prog *fp);
1128
bpf_prog_unlock_free(struct bpf_prog * fp)1129 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
1130 {
1131 __bpf_prog_free(fp);
1132 }
1133
1134 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
1135 unsigned int flen);
1136
1137 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
1138 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1139 bpf_aux_classic_check_t trans, bool save_orig);
1140 void bpf_prog_destroy(struct bpf_prog *fp);
1141
1142 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
1143 int sk_attach_bpf(u32 ufd, struct sock *sk);
1144 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
1145 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
1146 void sk_reuseport_prog_free(struct bpf_prog *prog);
1147 int sk_detach_filter(struct sock *sk);
1148 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len);
1149
1150 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
1151 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
1152
1153 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
1154
1155 struct bpf_prog *bpf_int_jit_compile(struct bpf_verifier_env *env, struct bpf_prog *prog);
1156 void bpf_jit_compile(struct bpf_prog *prog);
1157 bool bpf_jit_needs_zext(void);
1158 bool bpf_jit_inlines_helper_call(s32 imm);
1159 bool bpf_jit_supports_subprog_tailcalls(void);
1160 bool bpf_jit_supports_percpu_insn(void);
1161 bool bpf_jit_supports_kfunc_call(void);
1162 bool bpf_jit_supports_far_kfunc_call(void);
1163 bool bpf_jit_supports_exceptions(void);
1164 bool bpf_jit_supports_ptr_xchg(void);
1165 bool bpf_jit_supports_arena(void);
1166 bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena);
1167 bool bpf_jit_supports_private_stack(void);
1168 bool bpf_jit_supports_timed_may_goto(void);
1169 bool bpf_jit_supports_fsession(void);
1170 u64 bpf_arch_uaddress_limit(void);
1171 void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie);
1172 u64 arch_bpf_timed_may_goto(void);
1173 u64 bpf_check_timed_may_goto(struct bpf_timed_may_goto *);
1174 bool bpf_helper_changes_pkt_data(enum bpf_func_id func_id);
1175
bpf_dump_raw_ok(const struct cred * cred)1176 static inline bool bpf_dump_raw_ok(const struct cred *cred)
1177 {
1178 /* Reconstruction of call-sites is dependent on kallsyms,
1179 * thus make dump the same restriction.
1180 */
1181 return kallsyms_show_value(cred);
1182 }
1183
1184 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
1185 const struct bpf_insn *patch, u32 len);
1186
1187 #ifdef CONFIG_BPF_SYSCALL
1188 struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off,
1189 const struct bpf_insn *patch, u32 len);
1190 struct bpf_insn_aux_data *bpf_dup_insn_aux_data(struct bpf_verifier_env *env);
1191 void bpf_restore_insn_aux_data(struct bpf_verifier_env *env,
1192 struct bpf_insn_aux_data *orig_insn_aux);
1193 #else
bpf_patch_insn_data(struct bpf_verifier_env * env,u32 off,const struct bpf_insn * patch,u32 len)1194 static inline struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off,
1195 const struct bpf_insn *patch, u32 len)
1196 {
1197 return ERR_PTR(-ENOTSUPP);
1198 }
1199
bpf_dup_insn_aux_data(struct bpf_verifier_env * env)1200 static inline struct bpf_insn_aux_data *bpf_dup_insn_aux_data(struct bpf_verifier_env *env)
1201 {
1202 return NULL;
1203 }
1204
bpf_restore_insn_aux_data(struct bpf_verifier_env * env,struct bpf_insn_aux_data * orig_insn_aux)1205 static inline void bpf_restore_insn_aux_data(struct bpf_verifier_env *env,
1206 struct bpf_insn_aux_data *orig_insn_aux)
1207 {
1208 }
1209 #endif /* CONFIG_BPF_SYSCALL */
1210
1211 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
1212
xdp_return_frame_no_direct(void)1213 static inline bool xdp_return_frame_no_direct(void)
1214 {
1215 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1216
1217 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
1218 }
1219
xdp_set_return_frame_no_direct(void)1220 static inline void xdp_set_return_frame_no_direct(void)
1221 {
1222 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1223
1224 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
1225 }
1226
xdp_clear_return_frame_no_direct(void)1227 static inline void xdp_clear_return_frame_no_direct(void)
1228 {
1229 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1230
1231 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
1232 }
1233
xdp_ok_fwd_dev(const struct net_device * fwd,unsigned int pktlen)1234 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
1235 unsigned int pktlen)
1236 {
1237 unsigned int len;
1238
1239 if (unlikely(!(fwd->flags & IFF_UP)))
1240 return -ENETDOWN;
1241
1242 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
1243 if (pktlen > len)
1244 return -EMSGSIZE;
1245
1246 return 0;
1247 }
1248
1249 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the
1250 * same cpu context. Further for best results no more than a single map
1251 * for the do_redirect/do_flush pair should be used. This limitation is
1252 * because we only track one map and force a flush when the map changes.
1253 * This does not appear to be a real limitation for existing software.
1254 */
1255 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
1256 struct xdp_buff *xdp, const struct bpf_prog *prog);
1257 int xdp_do_redirect(struct net_device *dev,
1258 struct xdp_buff *xdp,
1259 const struct bpf_prog *prog);
1260 int xdp_do_redirect_frame(struct net_device *dev,
1261 struct xdp_buff *xdp,
1262 struct xdp_frame *xdpf,
1263 const struct bpf_prog *prog);
1264 void xdp_do_flush(void);
1265
1266 void bpf_warn_invalid_xdp_action(const struct net_device *dev,
1267 const struct bpf_prog *prog, u32 act);
1268
1269 #ifdef CONFIG_INET
1270 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1271 struct bpf_prog *prog, struct sk_buff *skb,
1272 struct sock *migrating_sk,
1273 u32 hash);
1274 #else
1275 static inline struct sock *
bpf_run_sk_reuseport(struct sock_reuseport * reuse,struct sock * sk,struct bpf_prog * prog,struct sk_buff * skb,struct sock * migrating_sk,u32 hash)1276 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1277 struct bpf_prog *prog, struct sk_buff *skb,
1278 struct sock *migrating_sk,
1279 u32 hash)
1280 {
1281 return NULL;
1282 }
1283 #endif
1284
1285 #ifdef CONFIG_BPF_JIT
1286 extern int bpf_jit_enable;
1287 extern int bpf_jit_harden;
1288 extern int bpf_jit_kallsyms;
1289 extern long bpf_jit_limit;
1290 extern long bpf_jit_limit_max;
1291
1292 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
1293
1294 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size);
1295
1296 struct bpf_binary_header *
1297 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1298 unsigned int alignment,
1299 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1300 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
1301 u64 bpf_jit_alloc_exec_limit(void);
1302 void *bpf_jit_alloc_exec(unsigned long size);
1303 void bpf_jit_free_exec(void *addr);
1304 void bpf_jit_free(struct bpf_prog *fp);
1305 struct bpf_binary_header *
1306 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp);
1307
1308 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns);
1309 void bpf_prog_pack_free(void *ptr, u32 size);
1310
bpf_prog_kallsyms_verify_off(const struct bpf_prog * fp)1311 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
1312 {
1313 return list_empty(&fp->aux->ksym.lnode) ||
1314 fp->aux->ksym.lnode.prev == LIST_POISON2;
1315 }
1316
1317 struct bpf_binary_header *
1318 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image,
1319 unsigned int alignment,
1320 struct bpf_binary_header **rw_hdr,
1321 u8 **rw_image,
1322 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1323 int bpf_jit_binary_pack_finalize(struct bpf_binary_header *ro_header,
1324 struct bpf_binary_header *rw_header);
1325 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header,
1326 struct bpf_binary_header *rw_header);
1327
1328 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1329 struct bpf_jit_poke_descriptor *poke);
1330
1331 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1332 const struct bpf_insn *insn, bool extra_pass,
1333 u64 *func_addr, bool *func_addr_fixed);
1334
1335 const char *bpf_jit_get_prog_name(struct bpf_prog *prog);
1336
1337 struct bpf_prog *bpf_jit_blind_constants(struct bpf_verifier_env *env, struct bpf_prog *prog);
1338 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
1339
bpf_prog_need_blind(const struct bpf_prog * prog)1340 static inline bool bpf_prog_need_blind(const struct bpf_prog *prog)
1341 {
1342 return prog->blinding_requested && !prog->blinded;
1343 }
1344
bpf_jit_dump(unsigned int flen,unsigned int proglen,u32 pass,void * image)1345 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
1346 u32 pass, void *image)
1347 {
1348 pr_err("flen=%u proglen=%u pass=%u image=%p from=%s pid=%d\n", flen,
1349 proglen, pass, image, current->comm, task_pid_nr(current));
1350
1351 if (image)
1352 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
1353 16, 1, image, proglen, false);
1354 }
1355
bpf_jit_is_ebpf(void)1356 static inline bool bpf_jit_is_ebpf(void)
1357 {
1358 # ifdef CONFIG_HAVE_EBPF_JIT
1359 return true;
1360 # else
1361 return false;
1362 # endif
1363 }
1364
ebpf_jit_enabled(void)1365 static inline bool ebpf_jit_enabled(void)
1366 {
1367 return bpf_jit_enable && bpf_jit_is_ebpf();
1368 }
1369
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1370 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1371 {
1372 return fp->jited && bpf_jit_is_ebpf();
1373 }
1374
bpf_jit_blinding_enabled(struct bpf_prog * prog)1375 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1376 {
1377 /* These are the prerequisites, should someone ever have the
1378 * idea to call blinding outside of them, we make sure to
1379 * bail out.
1380 */
1381 if (!bpf_jit_is_ebpf())
1382 return false;
1383 if (!prog->jit_requested)
1384 return false;
1385 if (!bpf_jit_harden)
1386 return false;
1387 if (bpf_jit_harden == 1 && bpf_token_capable(prog->aux->token, CAP_BPF))
1388 return false;
1389
1390 return true;
1391 }
1392
bpf_jit_kallsyms_enabled(void)1393 static inline bool bpf_jit_kallsyms_enabled(void)
1394 {
1395 /* There are a couple of corner cases where kallsyms should
1396 * not be enabled f.e. on hardening.
1397 */
1398 if (bpf_jit_harden)
1399 return false;
1400 if (!bpf_jit_kallsyms)
1401 return false;
1402 if (bpf_jit_kallsyms == 1)
1403 return true;
1404
1405 return false;
1406 }
1407
1408 int bpf_address_lookup(unsigned long addr, unsigned long *size,
1409 unsigned long *off, char *sym);
1410 bool is_bpf_text_address(unsigned long addr);
1411 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1412 char *sym);
1413 struct bpf_prog *bpf_prog_ksym_find(unsigned long addr);
1414
1415 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1416 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1417
1418 #else /* CONFIG_BPF_JIT */
1419
ebpf_jit_enabled(void)1420 static inline bool ebpf_jit_enabled(void)
1421 {
1422 return false;
1423 }
1424
bpf_jit_blinding_enabled(struct bpf_prog * prog)1425 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1426 {
1427 return false;
1428 }
1429
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1430 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1431 {
1432 return false;
1433 }
1434
1435 static inline int
bpf_jit_add_poke_descriptor(struct bpf_prog * prog,struct bpf_jit_poke_descriptor * poke)1436 bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1437 struct bpf_jit_poke_descriptor *poke)
1438 {
1439 return -ENOTSUPP;
1440 }
1441
bpf_jit_free(struct bpf_prog * fp)1442 static inline void bpf_jit_free(struct bpf_prog *fp)
1443 {
1444 bpf_prog_unlock_free(fp);
1445 }
1446
bpf_jit_kallsyms_enabled(void)1447 static inline bool bpf_jit_kallsyms_enabled(void)
1448 {
1449 return false;
1450 }
1451
1452 static inline int
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char * sym)1453 bpf_address_lookup(unsigned long addr, unsigned long *size,
1454 unsigned long *off, char *sym)
1455 {
1456 return 0;
1457 }
1458
is_bpf_text_address(unsigned long addr)1459 static inline bool is_bpf_text_address(unsigned long addr)
1460 {
1461 return false;
1462 }
1463
bpf_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * sym)1464 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1465 char *type, char *sym)
1466 {
1467 return -ERANGE;
1468 }
1469
bpf_prog_ksym_find(unsigned long addr)1470 static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
1471 {
1472 return NULL;
1473 }
1474
bpf_prog_kallsyms_add(struct bpf_prog * fp)1475 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1476 {
1477 }
1478
bpf_prog_kallsyms_del(struct bpf_prog * fp)1479 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1480 {
1481 }
1482
bpf_prog_need_blind(const struct bpf_prog * prog)1483 static inline bool bpf_prog_need_blind(const struct bpf_prog *prog)
1484 {
1485 return false;
1486 }
1487
1488 static inline
bpf_jit_blind_constants(struct bpf_verifier_env * env,struct bpf_prog * prog)1489 struct bpf_prog *bpf_jit_blind_constants(struct bpf_verifier_env *env, struct bpf_prog *prog)
1490 {
1491 return prog;
1492 }
1493
bpf_jit_prog_release_other(struct bpf_prog * fp,struct bpf_prog * fp_other)1494 static inline void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other)
1495 {
1496 }
1497 #endif /* CONFIG_BPF_JIT */
1498
1499 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1500
1501 #define BPF_ANC BIT(15)
1502
bpf_needs_clear_a(const struct sock_filter * first)1503 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1504 {
1505 switch (first->code) {
1506 case BPF_RET | BPF_K:
1507 case BPF_LD | BPF_W | BPF_LEN:
1508 return false;
1509
1510 case BPF_LD | BPF_W | BPF_ABS:
1511 case BPF_LD | BPF_H | BPF_ABS:
1512 case BPF_LD | BPF_B | BPF_ABS:
1513 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1514 return true;
1515 return false;
1516
1517 default:
1518 return true;
1519 }
1520 }
1521
bpf_anc_helper(const struct sock_filter * ftest)1522 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1523 {
1524 BUG_ON(ftest->code & BPF_ANC);
1525
1526 switch (ftest->code) {
1527 case BPF_LD | BPF_W | BPF_ABS:
1528 case BPF_LD | BPF_H | BPF_ABS:
1529 case BPF_LD | BPF_B | BPF_ABS:
1530 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1531 return BPF_ANC | SKF_AD_##CODE
1532 switch (ftest->k) {
1533 BPF_ANCILLARY(PROTOCOL);
1534 BPF_ANCILLARY(PKTTYPE);
1535 BPF_ANCILLARY(IFINDEX);
1536 BPF_ANCILLARY(NLATTR);
1537 BPF_ANCILLARY(NLATTR_NEST);
1538 BPF_ANCILLARY(MARK);
1539 BPF_ANCILLARY(QUEUE);
1540 BPF_ANCILLARY(HATYPE);
1541 BPF_ANCILLARY(RXHASH);
1542 BPF_ANCILLARY(CPU);
1543 BPF_ANCILLARY(ALU_XOR_X);
1544 BPF_ANCILLARY(VLAN_TAG);
1545 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1546 BPF_ANCILLARY(PAY_OFFSET);
1547 BPF_ANCILLARY(RANDOM);
1548 BPF_ANCILLARY(VLAN_TPID);
1549 }
1550 fallthrough;
1551 default:
1552 return ftest->code;
1553 }
1554 }
1555
1556 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1557 int k, unsigned int size);
1558
bpf_tell_extensions(void)1559 static inline int bpf_tell_extensions(void)
1560 {
1561 return SKF_AD_MAX;
1562 }
1563
1564 struct bpf_sock_addr_kern {
1565 struct sock *sk;
1566 struct sockaddr_unsized *uaddr;
1567 /* Temporary "register" to make indirect stores to nested structures
1568 * defined above. We need three registers to make such a store, but
1569 * only two (src and dst) are available at convert_ctx_access time
1570 */
1571 u64 tmp_reg;
1572 void *t_ctx; /* Attach type specific context. */
1573 u32 uaddrlen;
1574 };
1575
1576 struct bpf_sock_ops_kern {
1577 struct sock *sk;
1578 union {
1579 u32 args[4];
1580 u32 reply;
1581 u32 replylong[4];
1582 };
1583 struct sk_buff *syn_skb;
1584 struct sk_buff *skb;
1585 void *skb_data_end;
1586 u8 op;
1587 u8 is_fullsock;
1588 u8 is_locked_tcp_sock;
1589 u8 remaining_opt_len;
1590 u64 temp; /* temp and everything after is not
1591 * initialized to 0 before calling
1592 * the BPF program. New fields that
1593 * should be initialized to 0 should
1594 * be inserted before temp.
1595 * temp is scratch storage used by
1596 * sock_ops_convert_ctx_access
1597 * as temporary storage of a register.
1598 */
1599 };
1600
1601 struct bpf_sysctl_kern {
1602 struct ctl_table_header *head;
1603 const struct ctl_table *table;
1604 void *cur_val;
1605 size_t cur_len;
1606 void *new_val;
1607 size_t new_len;
1608 int new_updated;
1609 int write;
1610 loff_t *ppos;
1611 /* Temporary "register" for indirect stores to ppos. */
1612 u64 tmp_reg;
1613 };
1614
1615 #define BPF_SOCKOPT_KERN_BUF_SIZE 32
1616 struct bpf_sockopt_buf {
1617 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE];
1618 };
1619
1620 struct bpf_sockopt_kern {
1621 struct sock *sk;
1622 u8 *optval;
1623 u8 *optval_end;
1624 s32 level;
1625 s32 optname;
1626 s32 optlen;
1627 /* for retval in struct bpf_cg_run_ctx */
1628 struct task_struct *current_task;
1629 /* Temporary "register" for indirect stores to ppos. */
1630 u64 tmp_reg;
1631 };
1632
1633 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len);
1634
1635 struct bpf_sk_lookup_kern {
1636 u16 family;
1637 u16 protocol;
1638 __be16 sport;
1639 u16 dport;
1640 struct {
1641 __be32 saddr;
1642 __be32 daddr;
1643 } v4;
1644 struct {
1645 const struct in6_addr *saddr;
1646 const struct in6_addr *daddr;
1647 } v6;
1648 struct sock *selected_sk;
1649 u32 ingress_ifindex;
1650 bool no_reuseport;
1651 };
1652
1653 extern struct static_key_false bpf_sk_lookup_enabled;
1654
1655 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup.
1656 *
1657 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and
1658 * SK_DROP. Their meaning is as follows:
1659 *
1660 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result
1661 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup
1662 * SK_DROP : terminate lookup with -ECONNREFUSED
1663 *
1664 * This macro aggregates return values and selected sockets from
1665 * multiple BPF programs according to following rules in order:
1666 *
1667 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk,
1668 * macro result is SK_PASS and last ctx.selected_sk is used.
1669 * 2. If any program returned SK_DROP return value,
1670 * macro result is SK_DROP.
1671 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL.
1672 *
1673 * Caller must ensure that the prog array is non-NULL, and that the
1674 * array as well as the programs it contains remain valid.
1675 */
1676 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \
1677 ({ \
1678 struct bpf_sk_lookup_kern *_ctx = &(ctx); \
1679 struct bpf_prog_array_item *_item; \
1680 struct sock *_selected_sk = NULL; \
1681 bool _no_reuseport = false; \
1682 struct bpf_prog *_prog; \
1683 bool _all_pass = true; \
1684 u32 _ret; \
1685 \
1686 migrate_disable(); \
1687 _item = &(array)->items[0]; \
1688 while ((_prog = READ_ONCE(_item->prog))) { \
1689 /* restore most recent selection */ \
1690 _ctx->selected_sk = _selected_sk; \
1691 _ctx->no_reuseport = _no_reuseport; \
1692 \
1693 _ret = func(_prog, _ctx); \
1694 if (_ret == SK_PASS && _ctx->selected_sk) { \
1695 /* remember last non-NULL socket */ \
1696 _selected_sk = _ctx->selected_sk; \
1697 _no_reuseport = _ctx->no_reuseport; \
1698 } else if (_ret == SK_DROP && _all_pass) { \
1699 _all_pass = false; \
1700 } \
1701 _item++; \
1702 } \
1703 _ctx->selected_sk = _selected_sk; \
1704 _ctx->no_reuseport = _no_reuseport; \
1705 migrate_enable(); \
1706 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \
1707 })
1708
bpf_sk_lookup_run_v4(const struct net * net,int protocol,const __be32 saddr,const __be16 sport,const __be32 daddr,const u16 dport,const int ifindex,struct sock ** psk)1709 static inline bool bpf_sk_lookup_run_v4(const struct net *net, int protocol,
1710 const __be32 saddr, const __be16 sport,
1711 const __be32 daddr, const u16 dport,
1712 const int ifindex, struct sock **psk)
1713 {
1714 struct bpf_prog_array *run_array;
1715 struct sock *selected_sk = NULL;
1716 bool no_reuseport = false;
1717
1718 rcu_read_lock();
1719 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1720 if (run_array) {
1721 struct bpf_sk_lookup_kern ctx = {
1722 .family = AF_INET,
1723 .protocol = protocol,
1724 .v4.saddr = saddr,
1725 .v4.daddr = daddr,
1726 .sport = sport,
1727 .dport = dport,
1728 .ingress_ifindex = ifindex,
1729 };
1730 u32 act;
1731
1732 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1733 if (act == SK_PASS) {
1734 selected_sk = ctx.selected_sk;
1735 no_reuseport = ctx.no_reuseport;
1736 } else {
1737 selected_sk = ERR_PTR(-ECONNREFUSED);
1738 }
1739 }
1740 rcu_read_unlock();
1741 *psk = selected_sk;
1742 return no_reuseport;
1743 }
1744
1745 #if IS_ENABLED(CONFIG_IPV6)
bpf_sk_lookup_run_v6(const struct net * net,int protocol,const struct in6_addr * saddr,const __be16 sport,const struct in6_addr * daddr,const u16 dport,const int ifindex,struct sock ** psk)1746 static inline bool bpf_sk_lookup_run_v6(const struct net *net, int protocol,
1747 const struct in6_addr *saddr,
1748 const __be16 sport,
1749 const struct in6_addr *daddr,
1750 const u16 dport,
1751 const int ifindex, struct sock **psk)
1752 {
1753 struct bpf_prog_array *run_array;
1754 struct sock *selected_sk = NULL;
1755 bool no_reuseport = false;
1756
1757 rcu_read_lock();
1758 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1759 if (run_array) {
1760 struct bpf_sk_lookup_kern ctx = {
1761 .family = AF_INET6,
1762 .protocol = protocol,
1763 .v6.saddr = saddr,
1764 .v6.daddr = daddr,
1765 .sport = sport,
1766 .dport = dport,
1767 .ingress_ifindex = ifindex,
1768 };
1769 u32 act;
1770
1771 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1772 if (act == SK_PASS) {
1773 selected_sk = ctx.selected_sk;
1774 no_reuseport = ctx.no_reuseport;
1775 } else {
1776 selected_sk = ERR_PTR(-ECONNREFUSED);
1777 }
1778 }
1779 rcu_read_unlock();
1780 *psk = selected_sk;
1781 return no_reuseport;
1782 }
1783 #endif /* IS_ENABLED(CONFIG_IPV6) */
1784
__bpf_xdp_redirect_map(struct bpf_map * map,u64 index,u64 flags,const u64 flag_mask,void * lookup_elem (struct bpf_map * map,u32 key))1785 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index,
1786 u64 flags, const u64 flag_mask,
1787 void *lookup_elem(struct bpf_map *map, u32 key))
1788 {
1789 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1790 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX;
1791
1792 /* Lower bits of the flags are used as return code on lookup failure */
1793 if (unlikely(flags & ~(action_mask | flag_mask)))
1794 return XDP_ABORTED;
1795
1796 ri->tgt_value = lookup_elem(map, index);
1797 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) {
1798 /* If the lookup fails we want to clear out the state in the
1799 * redirect_info struct completely, so that if an eBPF program
1800 * performs multiple lookups, the last one always takes
1801 * precedence.
1802 */
1803 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */
1804 ri->map_type = BPF_MAP_TYPE_UNSPEC;
1805 return flags & action_mask;
1806 }
1807
1808 ri->tgt_index = index;
1809 ri->map_id = map->id;
1810 ri->map_type = map->map_type;
1811
1812 if (flags & BPF_F_BROADCAST) {
1813 WRITE_ONCE(ri->map, map);
1814 ri->flags = flags;
1815 } else {
1816 WRITE_ONCE(ri->map, NULL);
1817 ri->flags = 0;
1818 }
1819
1820 return XDP_REDIRECT;
1821 }
1822
1823 #ifdef CONFIG_NET
1824 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len);
1825 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from,
1826 u32 len, u64 flags);
1827 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len);
1828 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len);
1829 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len);
1830 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
1831 void *buf, unsigned long len, bool flush);
1832 int __bpf_skb_meta_store_bytes(struct sk_buff *skb, u32 offset,
1833 const void *from, u32 len, u64 flags);
1834 void *bpf_skb_meta_pointer(struct sk_buff *skb, u32 offset);
1835 #else /* CONFIG_NET */
__bpf_skb_load_bytes(const struct sk_buff * skb,u32 offset,void * to,u32 len)1836 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset,
1837 void *to, u32 len)
1838 {
1839 return -EOPNOTSUPP;
1840 }
1841
__bpf_skb_store_bytes(struct sk_buff * skb,u32 offset,const void * from,u32 len,u64 flags)1842 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset,
1843 const void *from, u32 len, u64 flags)
1844 {
1845 return -EOPNOTSUPP;
1846 }
1847
__bpf_xdp_load_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)1848 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset,
1849 void *buf, u32 len)
1850 {
1851 return -EOPNOTSUPP;
1852 }
1853
__bpf_xdp_store_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)1854 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset,
1855 void *buf, u32 len)
1856 {
1857 return -EOPNOTSUPP;
1858 }
1859
bpf_xdp_pointer(struct xdp_buff * xdp,u32 offset,u32 len)1860 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
1861 {
1862 return NULL;
1863 }
1864
bpf_xdp_copy_buf(struct xdp_buff * xdp,unsigned long off,void * buf,unsigned long len,bool flush)1865 static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf,
1866 unsigned long len, bool flush)
1867 {
1868 }
1869
__bpf_skb_meta_store_bytes(struct sk_buff * skb,u32 offset,const void * from,u32 len,u64 flags)1870 static inline int __bpf_skb_meta_store_bytes(struct sk_buff *skb, u32 offset,
1871 const void *from, u32 len,
1872 u64 flags)
1873 {
1874 return -EOPNOTSUPP;
1875 }
1876
bpf_skb_meta_pointer(struct sk_buff * skb,u32 offset)1877 static inline void *bpf_skb_meta_pointer(struct sk_buff *skb, u32 offset)
1878 {
1879 return ERR_PTR(-EOPNOTSUPP);
1880 }
1881 #endif /* CONFIG_NET */
1882
1883 #endif /* __LINUX_FILTER_H__ */
1884