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 <crypto/sha1.h>
25 #include <linux/u64_stats_sync.h>
26
27 #include <net/sch_generic.h>
28
29 #include <asm/byteorder.h>
30 #include <uapi/linux/filter.h>
31
32 struct sk_buff;
33 struct sock;
34 struct seccomp_data;
35 struct bpf_prog_aux;
36 struct xdp_rxq_info;
37 struct xdp_buff;
38 struct sock_reuseport;
39 struct ctl_table;
40 struct ctl_table_header;
41
42 /* ArgX, context and stack frame pointer register positions. Note,
43 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
44 * calls in BPF_CALL instruction.
45 */
46 #define BPF_REG_ARG1 BPF_REG_1
47 #define BPF_REG_ARG2 BPF_REG_2
48 #define BPF_REG_ARG3 BPF_REG_3
49 #define BPF_REG_ARG4 BPF_REG_4
50 #define BPF_REG_ARG5 BPF_REG_5
51 #define BPF_REG_CTX BPF_REG_6
52 #define BPF_REG_FP BPF_REG_10
53
54 /* Additional register mappings for converted user programs. */
55 #define BPF_REG_A BPF_REG_0
56 #define BPF_REG_X BPF_REG_7
57 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
58 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
59 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
60
61 /* Kernel hidden auxiliary/helper register. */
62 #define BPF_REG_AX MAX_BPF_REG
63 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1)
64 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
65
66 /* unused opcode to mark special call to bpf_tail_call() helper */
67 #define BPF_TAIL_CALL 0xf0
68
69 /* unused opcode to mark special load instruction. Same as BPF_ABS */
70 #define BPF_PROBE_MEM 0x20
71
72 /* unused opcode to mark special ldsx instruction. Same as BPF_IND */
73 #define BPF_PROBE_MEMSX 0x40
74
75 /* unused opcode to mark special load instruction. Same as BPF_MSH */
76 #define BPF_PROBE_MEM32 0xa0
77
78 /* unused opcode to mark special atomic instruction */
79 #define BPF_PROBE_ATOMIC 0xe0
80
81 /* unused opcode to mark special ldsx instruction. Same as BPF_NOSPEC */
82 #define BPF_PROBE_MEM32SX 0xc0
83
84 /* unused opcode to mark call to interpreter with arguments */
85 #define BPF_CALL_ARGS 0xe0
86
87 /* unused opcode to mark speculation barrier for mitigating
88 * Spectre v1 and v4
89 */
90 #define BPF_NOSPEC 0xc0
91
92 /* As per nm, we expose JITed images as text (code) section for
93 * kallsyms. That way, tools like perf can find it to match
94 * addresses.
95 */
96 #define BPF_SYM_ELF_TYPE 't'
97
98 /* BPF program can access up to 512 bytes of stack space. */
99 #define MAX_BPF_STACK 512
100
101 /* Helper macros for filter block array initializers. */
102
103 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
104
105 #define BPF_ALU64_REG_OFF(OP, DST, SRC, OFF) \
106 ((struct bpf_insn) { \
107 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
108 .dst_reg = DST, \
109 .src_reg = SRC, \
110 .off = OFF, \
111 .imm = 0 })
112
113 #define BPF_ALU64_REG(OP, DST, SRC) \
114 BPF_ALU64_REG_OFF(OP, DST, SRC, 0)
115
116 #define BPF_ALU32_REG_OFF(OP, DST, SRC, OFF) \
117 ((struct bpf_insn) { \
118 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
119 .dst_reg = DST, \
120 .src_reg = SRC, \
121 .off = OFF, \
122 .imm = 0 })
123
124 #define BPF_ALU32_REG(OP, DST, SRC) \
125 BPF_ALU32_REG_OFF(OP, DST, SRC, 0)
126
127 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
128
129 #define BPF_ALU64_IMM_OFF(OP, DST, IMM, OFF) \
130 ((struct bpf_insn) { \
131 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
132 .dst_reg = DST, \
133 .src_reg = 0, \
134 .off = OFF, \
135 .imm = IMM })
136 #define BPF_ALU64_IMM(OP, DST, IMM) \
137 BPF_ALU64_IMM_OFF(OP, DST, IMM, 0)
138
139 #define BPF_ALU32_IMM_OFF(OP, DST, IMM, OFF) \
140 ((struct bpf_insn) { \
141 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
142 .dst_reg = DST, \
143 .src_reg = 0, \
144 .off = OFF, \
145 .imm = IMM })
146 #define BPF_ALU32_IMM(OP, DST, IMM) \
147 BPF_ALU32_IMM_OFF(OP, DST, IMM, 0)
148
149 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
150
151 #define BPF_ENDIAN(TYPE, DST, LEN) \
152 ((struct bpf_insn) { \
153 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
154 .dst_reg = DST, \
155 .src_reg = 0, \
156 .off = 0, \
157 .imm = LEN })
158
159 /* Byte Swap, bswap16/32/64 */
160
161 #define BPF_BSWAP(DST, LEN) \
162 ((struct bpf_insn) { \
163 .code = BPF_ALU64 | BPF_END | BPF_SRC(BPF_TO_LE), \
164 .dst_reg = DST, \
165 .src_reg = 0, \
166 .off = 0, \
167 .imm = LEN })
168
169 /* Short form of mov, dst_reg = src_reg */
170
171 #define BPF_MOV64_REG(DST, SRC) \
172 ((struct bpf_insn) { \
173 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
174 .dst_reg = DST, \
175 .src_reg = SRC, \
176 .off = 0, \
177 .imm = 0 })
178
179 #define BPF_MOV32_REG(DST, SRC) \
180 ((struct bpf_insn) { \
181 .code = BPF_ALU | BPF_MOV | BPF_X, \
182 .dst_reg = DST, \
183 .src_reg = SRC, \
184 .off = 0, \
185 .imm = 0 })
186
187 /* Special (internal-only) form of mov, used to resolve per-CPU addrs:
188 * dst_reg = src_reg + <percpu_base_off>
189 * BPF_ADDR_PERCPU is used as a special insn->off value.
190 */
191 #define BPF_ADDR_PERCPU (-1)
192
193 #define BPF_MOV64_PERCPU_REG(DST, SRC) \
194 ((struct bpf_insn) { \
195 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
196 .dst_reg = DST, \
197 .src_reg = SRC, \
198 .off = BPF_ADDR_PERCPU, \
199 .imm = 0 })
200
insn_is_mov_percpu_addr(const struct bpf_insn * insn)201 static inline bool insn_is_mov_percpu_addr(const struct bpf_insn *insn)
202 {
203 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) && insn->off == BPF_ADDR_PERCPU;
204 }
205
206 /* Short form of mov, dst_reg = imm32 */
207
208 #define BPF_MOV64_IMM(DST, IMM) \
209 ((struct bpf_insn) { \
210 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
211 .dst_reg = DST, \
212 .src_reg = 0, \
213 .off = 0, \
214 .imm = IMM })
215
216 #define BPF_MOV32_IMM(DST, IMM) \
217 ((struct bpf_insn) { \
218 .code = BPF_ALU | BPF_MOV | BPF_K, \
219 .dst_reg = DST, \
220 .src_reg = 0, \
221 .off = 0, \
222 .imm = IMM })
223
224 /* Short form of movsx, dst_reg = (s8,s16,s32)src_reg */
225
226 #define BPF_MOVSX64_REG(DST, SRC, OFF) \
227 ((struct bpf_insn) { \
228 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
229 .dst_reg = DST, \
230 .src_reg = SRC, \
231 .off = OFF, \
232 .imm = 0 })
233
234 #define BPF_MOVSX32_REG(DST, SRC, OFF) \
235 ((struct bpf_insn) { \
236 .code = BPF_ALU | BPF_MOV | BPF_X, \
237 .dst_reg = DST, \
238 .src_reg = SRC, \
239 .off = OFF, \
240 .imm = 0 })
241
242 /* Special form of mov32, used for doing explicit zero extension on dst. */
243 #define BPF_ZEXT_REG(DST) \
244 ((struct bpf_insn) { \
245 .code = BPF_ALU | BPF_MOV | BPF_X, \
246 .dst_reg = DST, \
247 .src_reg = DST, \
248 .off = 0, \
249 .imm = 1 })
250
insn_is_zext(const struct bpf_insn * insn)251 static inline bool insn_is_zext(const struct bpf_insn *insn)
252 {
253 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1;
254 }
255
256 /* addr_space_cast from as(0) to as(1) is for converting bpf arena pointers
257 * to pointers in user vma.
258 */
insn_is_cast_user(const struct bpf_insn * insn)259 static inline bool insn_is_cast_user(const struct bpf_insn *insn)
260 {
261 return insn->code == (BPF_ALU64 | BPF_MOV | BPF_X) &&
262 insn->off == BPF_ADDR_SPACE_CAST &&
263 insn->imm == 1U << 16;
264 }
265
266 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
267 #define BPF_LD_IMM64(DST, IMM) \
268 BPF_LD_IMM64_RAW(DST, 0, IMM)
269
270 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
271 ((struct bpf_insn) { \
272 .code = BPF_LD | BPF_DW | BPF_IMM, \
273 .dst_reg = DST, \
274 .src_reg = SRC, \
275 .off = 0, \
276 .imm = (__u32) (IMM) }), \
277 ((struct bpf_insn) { \
278 .code = 0, /* zero is reserved opcode */ \
279 .dst_reg = 0, \
280 .src_reg = 0, \
281 .off = 0, \
282 .imm = ((__u64) (IMM)) >> 32 })
283
284 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
285 #define BPF_LD_MAP_FD(DST, MAP_FD) \
286 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
287
288 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
289
290 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
291 ((struct bpf_insn) { \
292 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
293 .dst_reg = DST, \
294 .src_reg = SRC, \
295 .off = 0, \
296 .imm = IMM })
297
298 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
299 ((struct bpf_insn) { \
300 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
301 .dst_reg = DST, \
302 .src_reg = SRC, \
303 .off = 0, \
304 .imm = IMM })
305
306 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
307
308 #define BPF_LD_ABS(SIZE, IMM) \
309 ((struct bpf_insn) { \
310 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
311 .dst_reg = 0, \
312 .src_reg = 0, \
313 .off = 0, \
314 .imm = IMM })
315
316 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
317
318 #define BPF_LD_IND(SIZE, SRC, IMM) \
319 ((struct bpf_insn) { \
320 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
321 .dst_reg = 0, \
322 .src_reg = SRC, \
323 .off = 0, \
324 .imm = IMM })
325
326 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
327
328 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
329 ((struct bpf_insn) { \
330 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
331 .dst_reg = DST, \
332 .src_reg = SRC, \
333 .off = OFF, \
334 .imm = 0 })
335
336 /* Memory load, dst_reg = *(signed size *) (src_reg + off16) */
337
338 #define BPF_LDX_MEMSX(SIZE, DST, SRC, OFF) \
339 ((struct bpf_insn) { \
340 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEMSX, \
341 .dst_reg = DST, \
342 .src_reg = SRC, \
343 .off = OFF, \
344 .imm = 0 })
345
346 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
347
348 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
349 ((struct bpf_insn) { \
350 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
351 .dst_reg = DST, \
352 .src_reg = SRC, \
353 .off = OFF, \
354 .imm = 0 })
355
356
357 /*
358 * Atomic operations:
359 *
360 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg
361 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg
362 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg
363 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg
364 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg);
365 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg);
366 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg);
367 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg);
368 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg)
369 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg)
370 * BPF_LOAD_ACQ dst_reg = smp_load_acquire(src_reg + off16)
371 * BPF_STORE_REL smp_store_release(dst_reg + off16, src_reg)
372 */
373
374 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \
375 ((struct bpf_insn) { \
376 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \
377 .dst_reg = DST, \
378 .src_reg = SRC, \
379 .off = OFF, \
380 .imm = OP })
381
382 /* Legacy alias */
383 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF)
384
385 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
386
387 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
388 ((struct bpf_insn) { \
389 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
390 .dst_reg = DST, \
391 .src_reg = 0, \
392 .off = OFF, \
393 .imm = IMM })
394
395 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
396
397 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
398 ((struct bpf_insn) { \
399 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
400 .dst_reg = DST, \
401 .src_reg = SRC, \
402 .off = OFF, \
403 .imm = 0 })
404
405 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
406
407 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
408 ((struct bpf_insn) { \
409 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
410 .dst_reg = DST, \
411 .src_reg = 0, \
412 .off = OFF, \
413 .imm = IMM })
414
415 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
416
417 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
418 ((struct bpf_insn) { \
419 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
420 .dst_reg = DST, \
421 .src_reg = SRC, \
422 .off = OFF, \
423 .imm = 0 })
424
425 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
426
427 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
428 ((struct bpf_insn) { \
429 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
430 .dst_reg = DST, \
431 .src_reg = 0, \
432 .off = OFF, \
433 .imm = IMM })
434
435 /* Unconditional jumps, goto pc + off16 */
436
437 #define BPF_JMP_A(OFF) \
438 ((struct bpf_insn) { \
439 .code = BPF_JMP | BPF_JA, \
440 .dst_reg = 0, \
441 .src_reg = 0, \
442 .off = OFF, \
443 .imm = 0 })
444
445 /* Unconditional jumps, gotol pc + imm32 */
446
447 #define BPF_JMP32_A(IMM) \
448 ((struct bpf_insn) { \
449 .code = BPF_JMP32 | BPF_JA, \
450 .dst_reg = 0, \
451 .src_reg = 0, \
452 .off = 0, \
453 .imm = IMM })
454
455 /* Relative call */
456
457 #define BPF_CALL_REL(TGT) \
458 ((struct bpf_insn) { \
459 .code = BPF_JMP | BPF_CALL, \
460 .dst_reg = 0, \
461 .src_reg = BPF_PSEUDO_CALL, \
462 .off = 0, \
463 .imm = TGT })
464
465 /* Convert function address to BPF immediate */
466
467 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base)
468
469 #define BPF_EMIT_CALL(FUNC) \
470 ((struct bpf_insn) { \
471 .code = BPF_JMP | BPF_CALL, \
472 .dst_reg = 0, \
473 .src_reg = 0, \
474 .off = 0, \
475 .imm = BPF_CALL_IMM(FUNC) })
476
477 /* Kfunc call */
478
479 #define BPF_CALL_KFUNC(OFF, IMM) \
480 ((struct bpf_insn) { \
481 .code = BPF_JMP | BPF_CALL, \
482 .dst_reg = 0, \
483 .src_reg = BPF_PSEUDO_KFUNC_CALL, \
484 .off = OFF, \
485 .imm = IMM })
486
487 /* Raw code statement block */
488
489 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
490 ((struct bpf_insn) { \
491 .code = CODE, \
492 .dst_reg = DST, \
493 .src_reg = SRC, \
494 .off = OFF, \
495 .imm = IMM })
496
497 /* Program exit */
498
499 #define BPF_EXIT_INSN() \
500 ((struct bpf_insn) { \
501 .code = BPF_JMP | BPF_EXIT, \
502 .dst_reg = 0, \
503 .src_reg = 0, \
504 .off = 0, \
505 .imm = 0 })
506
507 /* Speculation barrier */
508
509 #define BPF_ST_NOSPEC() \
510 ((struct bpf_insn) { \
511 .code = BPF_ST | BPF_NOSPEC, \
512 .dst_reg = 0, \
513 .src_reg = 0, \
514 .off = 0, \
515 .imm = 0 })
516
517 /* Internal classic blocks for direct assignment */
518
519 #define __BPF_STMT(CODE, K) \
520 ((struct sock_filter) BPF_STMT(CODE, K))
521
522 #define __BPF_JUMP(CODE, K, JT, JF) \
523 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
524
525 #define bytes_to_bpf_size(bytes) \
526 ({ \
527 int bpf_size = -EINVAL; \
528 \
529 if (bytes == sizeof(u8)) \
530 bpf_size = BPF_B; \
531 else if (bytes == sizeof(u16)) \
532 bpf_size = BPF_H; \
533 else if (bytes == sizeof(u32)) \
534 bpf_size = BPF_W; \
535 else if (bytes == sizeof(u64)) \
536 bpf_size = BPF_DW; \
537 \
538 bpf_size; \
539 })
540
541 #define bpf_size_to_bytes(bpf_size) \
542 ({ \
543 int bytes = -EINVAL; \
544 \
545 if (bpf_size == BPF_B) \
546 bytes = sizeof(u8); \
547 else if (bpf_size == BPF_H) \
548 bytes = sizeof(u16); \
549 else if (bpf_size == BPF_W) \
550 bytes = sizeof(u32); \
551 else if (bpf_size == BPF_DW) \
552 bytes = sizeof(u64); \
553 \
554 bytes; \
555 })
556
557 #define BPF_SIZEOF(type) \
558 ({ \
559 const int __size = bytes_to_bpf_size(sizeof(type)); \
560 BUILD_BUG_ON(__size < 0); \
561 __size; \
562 })
563
564 #define BPF_FIELD_SIZEOF(type, field) \
565 ({ \
566 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
567 BUILD_BUG_ON(__size < 0); \
568 __size; \
569 })
570
571 #define BPF_LDST_BYTES(insn) \
572 ({ \
573 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
574 WARN_ON(__size < 0); \
575 __size; \
576 })
577
578 #define __BPF_MAP_0(m, v, ...) v
579 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
580 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
581 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
582 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
583 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
584
585 #define __BPF_REG_0(...) __BPF_PAD(5)
586 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
587 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
588 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
589 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
590 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
591
592 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
593 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
594
595 #define __BPF_CAST(t, a) \
596 (__force t) \
597 (__force \
598 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
599 (unsigned long)0, (t)0))) a
600 #define __BPF_V void
601 #define __BPF_N
602
603 #define __BPF_DECL_ARGS(t, a) t a
604 #define __BPF_DECL_REGS(t, a) u64 a
605
606 #define __BPF_PAD(n) \
607 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
608 u64, __ur_3, u64, __ur_4, u64, __ur_5)
609
610 #define BPF_CALL_x(x, attr, name, ...) \
611 static __always_inline \
612 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
613 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
614 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
615 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
616 { \
617 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
618 } \
619 static __always_inline \
620 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
621
622 #define __NOATTR
623 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, __NOATTR, name, __VA_ARGS__)
624 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, __NOATTR, name, __VA_ARGS__)
625 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, __NOATTR, name, __VA_ARGS__)
626 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__)
627 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, __NOATTR, name, __VA_ARGS__)
628 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, __NOATTR, name, __VA_ARGS__)
629
630 #define NOTRACE_BPF_CALL_1(name, ...) BPF_CALL_x(1, notrace, name, __VA_ARGS__)
631
632 #define bpf_ctx_range(TYPE, MEMBER) \
633 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
634 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
635 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
636 #if BITS_PER_LONG == 64
637 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
638 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
639 #else
640 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
641 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
642 #endif /* BITS_PER_LONG == 64 */
643
644 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
645 ({ \
646 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \
647 *(PTR_SIZE) = (SIZE); \
648 offsetof(TYPE, MEMBER); \
649 })
650
651 /* A struct sock_filter is architecture independent. */
652 struct compat_sock_fprog {
653 u16 len;
654 compat_uptr_t filter; /* struct sock_filter * */
655 };
656
657 struct sock_fprog_kern {
658 u16 len;
659 struct sock_filter *filter;
660 };
661
662 /* Some arches need doubleword alignment for their instructions and/or data */
663 #define BPF_IMAGE_ALIGNMENT 8
664
665 struct bpf_binary_header {
666 u32 size;
667 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT);
668 };
669
670 struct bpf_prog_stats {
671 u64_stats_t cnt;
672 u64_stats_t nsecs;
673 u64_stats_t misses;
674 struct u64_stats_sync syncp;
675 } __aligned(2 * sizeof(u64));
676
677 struct bpf_timed_may_goto {
678 u64 count;
679 u64 timestamp;
680 };
681
682 struct sk_filter {
683 refcount_t refcnt;
684 struct rcu_head rcu;
685 struct bpf_prog *prog;
686 };
687
688 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
689
690 extern struct mutex nf_conn_btf_access_lock;
691 extern int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
692 const struct bpf_reg_state *reg,
693 int off, int size);
694
695 typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx,
696 const struct bpf_insn *insnsi,
697 unsigned int (*bpf_func)(const void *,
698 const struct bpf_insn *));
699
__bpf_prog_run(const struct bpf_prog * prog,const void * ctx,bpf_dispatcher_fn dfunc)700 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog,
701 const void *ctx,
702 bpf_dispatcher_fn dfunc)
703 {
704 u32 ret;
705
706 cant_migrate();
707 if (static_branch_unlikely(&bpf_stats_enabled_key)) {
708 struct bpf_prog_stats *stats;
709 u64 duration, start = sched_clock();
710 unsigned long flags;
711
712 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
713
714 duration = sched_clock() - start;
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 } else {
721 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
722 }
723 return ret;
724 }
725
bpf_prog_run(const struct bpf_prog * prog,const void * ctx)726 static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx)
727 {
728 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func);
729 }
730
731 /*
732 * Use in preemptible and therefore migratable context to make sure that
733 * the execution of the BPF program runs on one CPU.
734 *
735 * This uses migrate_disable/enable() explicitly to document that the
736 * invocation of a BPF program does not require reentrancy protection
737 * against a BPF program which is invoked from a preempting task.
738 */
bpf_prog_run_pin_on_cpu(const struct bpf_prog * prog,const void * ctx)739 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
740 const void *ctx)
741 {
742 u32 ret;
743
744 migrate_disable();
745 ret = bpf_prog_run(prog, ctx);
746 migrate_enable();
747 return ret;
748 }
749
750 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
751
752 struct bpf_skb_data_end {
753 struct qdisc_skb_cb qdisc_cb;
754 void *data_meta;
755 void *data_end;
756 };
757
758 struct bpf_nh_params {
759 u32 nh_family;
760 union {
761 u32 ipv4_nh;
762 struct in6_addr ipv6_nh;
763 };
764 };
765
766 /* flags for bpf_redirect_info kern_flags */
767 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
768 #define BPF_RI_F_RI_INIT BIT(1)
769 #define BPF_RI_F_CPU_MAP_INIT BIT(2)
770 #define BPF_RI_F_DEV_MAP_INIT BIT(3)
771 #define BPF_RI_F_XSK_MAP_INIT BIT(4)
772
773 struct bpf_redirect_info {
774 u64 tgt_index;
775 void *tgt_value;
776 struct bpf_map *map;
777 u32 flags;
778 u32 map_id;
779 enum bpf_map_type map_type;
780 struct bpf_nh_params nh;
781 u32 kern_flags;
782 };
783
784 struct bpf_net_context {
785 struct bpf_redirect_info ri;
786 struct list_head cpu_map_flush_list;
787 struct list_head dev_map_flush_list;
788 struct list_head xskmap_map_flush_list;
789 };
790
bpf_net_ctx_set(struct bpf_net_context * bpf_net_ctx)791 static inline struct bpf_net_context *bpf_net_ctx_set(struct bpf_net_context *bpf_net_ctx)
792 {
793 struct task_struct *tsk = current;
794
795 if (tsk->bpf_net_context != NULL)
796 return NULL;
797 bpf_net_ctx->ri.kern_flags = 0;
798
799 tsk->bpf_net_context = bpf_net_ctx;
800 return bpf_net_ctx;
801 }
802
bpf_net_ctx_clear(struct bpf_net_context * bpf_net_ctx)803 static inline void bpf_net_ctx_clear(struct bpf_net_context *bpf_net_ctx)
804 {
805 if (bpf_net_ctx)
806 current->bpf_net_context = NULL;
807 }
808
bpf_net_ctx_get(void)809 static inline struct bpf_net_context *bpf_net_ctx_get(void)
810 {
811 return current->bpf_net_context;
812 }
813
bpf_net_ctx_get_ri(void)814 static inline struct bpf_redirect_info *bpf_net_ctx_get_ri(void)
815 {
816 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
817
818 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_RI_INIT)) {
819 memset(&bpf_net_ctx->ri, 0, offsetof(struct bpf_net_context, ri.nh));
820 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_RI_INIT;
821 }
822
823 return &bpf_net_ctx->ri;
824 }
825
bpf_net_ctx_get_cpu_map_flush_list(void)826 static inline struct list_head *bpf_net_ctx_get_cpu_map_flush_list(void)
827 {
828 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
829
830 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_CPU_MAP_INIT)) {
831 INIT_LIST_HEAD(&bpf_net_ctx->cpu_map_flush_list);
832 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_CPU_MAP_INIT;
833 }
834
835 return &bpf_net_ctx->cpu_map_flush_list;
836 }
837
bpf_net_ctx_get_dev_flush_list(void)838 static inline struct list_head *bpf_net_ctx_get_dev_flush_list(void)
839 {
840 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
841
842 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_DEV_MAP_INIT)) {
843 INIT_LIST_HEAD(&bpf_net_ctx->dev_map_flush_list);
844 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_DEV_MAP_INIT;
845 }
846
847 return &bpf_net_ctx->dev_map_flush_list;
848 }
849
bpf_net_ctx_get_xskmap_flush_list(void)850 static inline struct list_head *bpf_net_ctx_get_xskmap_flush_list(void)
851 {
852 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
853
854 if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_XSK_MAP_INIT)) {
855 INIT_LIST_HEAD(&bpf_net_ctx->xskmap_map_flush_list);
856 bpf_net_ctx->ri.kern_flags |= BPF_RI_F_XSK_MAP_INIT;
857 }
858
859 return &bpf_net_ctx->xskmap_map_flush_list;
860 }
861
bpf_net_ctx_get_all_used_flush_lists(struct list_head ** lh_map,struct list_head ** lh_dev,struct list_head ** lh_xsk)862 static inline void bpf_net_ctx_get_all_used_flush_lists(struct list_head **lh_map,
863 struct list_head **lh_dev,
864 struct list_head **lh_xsk)
865 {
866 struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
867 u32 kern_flags = bpf_net_ctx->ri.kern_flags;
868 struct list_head *lh;
869
870 *lh_map = *lh_dev = *lh_xsk = NULL;
871
872 if (!IS_ENABLED(CONFIG_BPF_SYSCALL))
873 return;
874
875 lh = &bpf_net_ctx->dev_map_flush_list;
876 if (kern_flags & BPF_RI_F_DEV_MAP_INIT && !list_empty(lh))
877 *lh_dev = lh;
878
879 lh = &bpf_net_ctx->cpu_map_flush_list;
880 if (kern_flags & BPF_RI_F_CPU_MAP_INIT && !list_empty(lh))
881 *lh_map = lh;
882
883 lh = &bpf_net_ctx->xskmap_map_flush_list;
884 if (IS_ENABLED(CONFIG_XDP_SOCKETS) &&
885 kern_flags & BPF_RI_F_XSK_MAP_INIT && !list_empty(lh))
886 *lh_xsk = lh;
887 }
888
889 /* Compute the linear packet data range [data, data_end) which
890 * will be accessed by various program types (cls_bpf, act_bpf,
891 * lwt, ...). Subsystems allowing direct data access must (!)
892 * ensure that cb[] area can be written to when BPF program is
893 * invoked (otherwise cb[] save/restore is necessary).
894 */
bpf_compute_data_pointers(struct sk_buff * skb)895 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
896 {
897 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
898
899 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
900 cb->data_meta = skb->data - skb_metadata_len(skb);
901 cb->data_end = skb->data + skb_headlen(skb);
902 }
903
bpf_prog_run_data_pointers(const struct bpf_prog * prog,struct sk_buff * skb)904 static inline int bpf_prog_run_data_pointers(
905 const struct bpf_prog *prog,
906 struct sk_buff *skb)
907 {
908 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
909 void *save_data_meta, *save_data_end;
910 int res;
911
912 save_data_meta = cb->data_meta;
913 save_data_end = cb->data_end;
914
915 bpf_compute_data_pointers(skb);
916 res = bpf_prog_run(prog, skb);
917
918 cb->data_meta = save_data_meta;
919 cb->data_end = save_data_end;
920
921 return res;
922 }
923
924 /* Similar to bpf_compute_data_pointers(), except that save orginal
925 * data in cb->data and cb->meta_data for restore.
926 */
bpf_compute_and_save_data_end(struct sk_buff * skb,void ** saved_data_end)927 static inline void bpf_compute_and_save_data_end(
928 struct sk_buff *skb, void **saved_data_end)
929 {
930 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
931
932 *saved_data_end = cb->data_end;
933 cb->data_end = skb->data + skb_headlen(skb);
934 }
935
936 /* Restore data saved by bpf_compute_and_save_data_end(). */
bpf_restore_data_end(struct sk_buff * skb,void * saved_data_end)937 static inline void bpf_restore_data_end(
938 struct sk_buff *skb, void *saved_data_end)
939 {
940 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
941
942 cb->data_end = saved_data_end;
943 }
944
bpf_skb_cb(const struct sk_buff * skb)945 static inline u8 *bpf_skb_cb(const struct sk_buff *skb)
946 {
947 /* eBPF programs may read/write skb->cb[] area to transfer meta
948 * data between tail calls. Since this also needs to work with
949 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
950 *
951 * In some socket filter cases, the cb unfortunately needs to be
952 * saved/restored so that protocol specific skb->cb[] data won't
953 * be lost. In any case, due to unpriviledged eBPF programs
954 * attached to sockets, we need to clear the bpf_skb_cb() area
955 * to not leak previous contents to user space.
956 */
957 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
958 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) !=
959 sizeof_field(struct qdisc_skb_cb, data));
960
961 return qdisc_skb_cb(skb)->data;
962 }
963
964 /* Must be invoked with migration disabled */
__bpf_prog_run_save_cb(const struct bpf_prog * prog,const void * ctx)965 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
966 const void *ctx)
967 {
968 const struct sk_buff *skb = ctx;
969 u8 *cb_data = bpf_skb_cb(skb);
970 u8 cb_saved[BPF_SKB_CB_LEN];
971 u32 res;
972
973 if (unlikely(prog->cb_access)) {
974 memcpy(cb_saved, cb_data, sizeof(cb_saved));
975 memset(cb_data, 0, sizeof(cb_saved));
976 }
977
978 res = bpf_prog_run(prog, skb);
979
980 if (unlikely(prog->cb_access))
981 memcpy(cb_data, cb_saved, sizeof(cb_saved));
982
983 return res;
984 }
985
bpf_prog_run_save_cb(const struct bpf_prog * prog,struct sk_buff * skb)986 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
987 struct sk_buff *skb)
988 {
989 u32 res;
990
991 migrate_disable();
992 res = __bpf_prog_run_save_cb(prog, skb);
993 migrate_enable();
994 return res;
995 }
996
bpf_prog_run_clear_cb(const struct bpf_prog * prog,struct sk_buff * skb)997 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
998 struct sk_buff *skb)
999 {
1000 u8 *cb_data = bpf_skb_cb(skb);
1001 u32 res;
1002
1003 if (unlikely(prog->cb_access))
1004 memset(cb_data, 0, BPF_SKB_CB_LEN);
1005
1006 res = bpf_prog_run_pin_on_cpu(prog, skb);
1007 return res;
1008 }
1009
1010 DECLARE_BPF_DISPATCHER(xdp)
1011
1012 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
1013
1014 u32 xdp_master_redirect(struct xdp_buff *xdp);
1015
1016 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog);
1017
bpf_prog_insn_size(const struct bpf_prog * prog)1018 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
1019 {
1020 return prog->len * sizeof(struct bpf_insn);
1021 }
1022
bpf_prog_size(unsigned int proglen)1023 static inline unsigned int bpf_prog_size(unsigned int proglen)
1024 {
1025 return max(sizeof(struct bpf_prog),
1026 offsetof(struct bpf_prog, insns[proglen]));
1027 }
1028
bpf_prog_was_classic(const struct bpf_prog * prog)1029 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
1030 {
1031 /* When classic BPF programs have been loaded and the arch
1032 * does not have a classic BPF JIT (anymore), they have been
1033 * converted via bpf_migrate_filter() to eBPF and thus always
1034 * have an unspec program type.
1035 */
1036 return prog->type == BPF_PROG_TYPE_UNSPEC;
1037 }
1038
bpf_ctx_off_adjust_machine(u32 size)1039 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
1040 {
1041 const u32 size_machine = sizeof(unsigned long);
1042
1043 if (size > size_machine && size % size_machine == 0)
1044 size = size_machine;
1045
1046 return size;
1047 }
1048
1049 static inline bool
bpf_ctx_narrow_access_ok(u32 off,u32 size,u32 size_default)1050 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
1051 {
1052 return size <= size_default && (size & (size - 1)) == 0;
1053 }
1054
1055 static inline u8
bpf_ctx_narrow_access_offset(u32 off,u32 size,u32 size_default)1056 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
1057 {
1058 u8 access_off = off & (size_default - 1);
1059
1060 #ifdef __LITTLE_ENDIAN
1061 return access_off;
1062 #else
1063 return size_default - (access_off + size);
1064 #endif
1065 }
1066
1067 #define bpf_ctx_wide_access_ok(off, size, type, field) \
1068 (size == sizeof(__u64) && \
1069 off >= offsetof(type, field) && \
1070 off + sizeof(__u64) <= offsetofend(type, field) && \
1071 off % sizeof(__u64) == 0)
1072
1073 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
1074
bpf_prog_lock_ro(struct bpf_prog * fp)1075 static inline int __must_check bpf_prog_lock_ro(struct bpf_prog *fp)
1076 {
1077 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1078 if (!fp->jited) {
1079 set_vm_flush_reset_perms(fp);
1080 return set_memory_ro((unsigned long)fp, fp->pages);
1081 }
1082 #endif
1083 return 0;
1084 }
1085
1086 static inline int __must_check
bpf_jit_binary_lock_ro(struct bpf_binary_header * hdr)1087 bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
1088 {
1089 set_vm_flush_reset_perms(hdr);
1090 return set_memory_rox((unsigned long)hdr, hdr->size >> PAGE_SHIFT);
1091 }
1092
1093 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap,
1094 enum skb_drop_reason *reason);
1095
sk_filter(struct sock * sk,struct sk_buff * skb)1096 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
1097 {
1098 enum skb_drop_reason ignore_reason;
1099
1100 return sk_filter_trim_cap(sk, skb, 1, &ignore_reason);
1101 }
1102
sk_filter_reason(struct sock * sk,struct sk_buff * skb,enum skb_drop_reason * reason)1103 static inline int sk_filter_reason(struct sock *sk, struct sk_buff *skb,
1104 enum skb_drop_reason *reason)
1105 {
1106 return sk_filter_trim_cap(sk, skb, 1, reason);
1107 }
1108
1109 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
1110 void bpf_prog_free(struct bpf_prog *fp);
1111
1112 bool bpf_opcode_in_insntable(u8 code);
1113
1114 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
1115 const u32 *insn_to_jit_off);
1116 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
1117 void bpf_prog_jit_attempt_done(struct bpf_prog *prog);
1118
1119 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
1120 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
1121 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
1122 gfp_t gfp_extra_flags);
1123 void __bpf_prog_free(struct bpf_prog *fp);
1124
bpf_prog_unlock_free(struct bpf_prog * fp)1125 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
1126 {
1127 __bpf_prog_free(fp);
1128 }
1129
1130 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
1131 unsigned int flen);
1132
1133 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
1134 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1135 bpf_aux_classic_check_t trans, bool save_orig);
1136 void bpf_prog_destroy(struct bpf_prog *fp);
1137
1138 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
1139 int sk_attach_bpf(u32 ufd, struct sock *sk);
1140 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
1141 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
1142 void sk_reuseport_prog_free(struct bpf_prog *prog);
1143 int sk_detach_filter(struct sock *sk);
1144 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len);
1145
1146 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
1147 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
1148
1149 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
1150 #define __bpf_call_base_args \
1151 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
1152 (void *)__bpf_call_base)
1153
1154 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
1155 void bpf_jit_compile(struct bpf_prog *prog);
1156 bool bpf_jit_needs_zext(void);
1157 bool bpf_jit_inlines_helper_call(s32 imm);
1158 bool bpf_jit_supports_subprog_tailcalls(void);
1159 bool bpf_jit_supports_percpu_insn(void);
1160 bool bpf_jit_supports_kfunc_call(void);
1161 bool bpf_jit_supports_far_kfunc_call(void);
1162 bool bpf_jit_supports_exceptions(void);
1163 bool bpf_jit_supports_ptr_xchg(void);
1164 bool bpf_jit_supports_arena(void);
1165 bool bpf_jit_supports_insn(struct bpf_insn *insn, bool in_arena);
1166 bool bpf_jit_supports_private_stack(void);
1167 bool bpf_jit_supports_timed_may_goto(void);
1168 u64 bpf_arch_uaddress_limit(void);
1169 void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp), void *cookie);
1170 u64 arch_bpf_timed_may_goto(void);
1171 u64 bpf_check_timed_may_goto(struct bpf_timed_may_goto *);
1172 bool bpf_helper_changes_pkt_data(enum bpf_func_id func_id);
1173
bpf_dump_raw_ok(const struct cred * cred)1174 static inline bool bpf_dump_raw_ok(const struct cred *cred)
1175 {
1176 /* Reconstruction of call-sites is dependent on kallsyms,
1177 * thus make dump the same restriction.
1178 */
1179 return kallsyms_show_value(cred);
1180 }
1181
1182 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
1183 const struct bpf_insn *patch, u32 len);
1184 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
1185
xdp_return_frame_no_direct(void)1186 static inline bool xdp_return_frame_no_direct(void)
1187 {
1188 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1189
1190 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
1191 }
1192
xdp_set_return_frame_no_direct(void)1193 static inline void xdp_set_return_frame_no_direct(void)
1194 {
1195 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1196
1197 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
1198 }
1199
xdp_clear_return_frame_no_direct(void)1200 static inline void xdp_clear_return_frame_no_direct(void)
1201 {
1202 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1203
1204 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
1205 }
1206
xdp_ok_fwd_dev(const struct net_device * fwd,unsigned int pktlen)1207 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
1208 unsigned int pktlen)
1209 {
1210 unsigned int len;
1211
1212 if (unlikely(!(fwd->flags & IFF_UP)))
1213 return -ENETDOWN;
1214
1215 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
1216 if (pktlen > len)
1217 return -EMSGSIZE;
1218
1219 return 0;
1220 }
1221
1222 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the
1223 * same cpu context. Further for best results no more than a single map
1224 * for the do_redirect/do_flush pair should be used. This limitation is
1225 * because we only track one map and force a flush when the map changes.
1226 * This does not appear to be a real limitation for existing software.
1227 */
1228 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
1229 struct xdp_buff *xdp, const struct bpf_prog *prog);
1230 int xdp_do_redirect(struct net_device *dev,
1231 struct xdp_buff *xdp,
1232 const struct bpf_prog *prog);
1233 int xdp_do_redirect_frame(struct net_device *dev,
1234 struct xdp_buff *xdp,
1235 struct xdp_frame *xdpf,
1236 const struct bpf_prog *prog);
1237 void xdp_do_flush(void);
1238
1239 void bpf_warn_invalid_xdp_action(const struct net_device *dev,
1240 const struct bpf_prog *prog, u32 act);
1241
1242 #ifdef CONFIG_INET
1243 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1244 struct bpf_prog *prog, struct sk_buff *skb,
1245 struct sock *migrating_sk,
1246 u32 hash);
1247 #else
1248 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)1249 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1250 struct bpf_prog *prog, struct sk_buff *skb,
1251 struct sock *migrating_sk,
1252 u32 hash)
1253 {
1254 return NULL;
1255 }
1256 #endif
1257
1258 #ifdef CONFIG_BPF_JIT
1259 extern int bpf_jit_enable;
1260 extern int bpf_jit_harden;
1261 extern int bpf_jit_kallsyms;
1262 extern long bpf_jit_limit;
1263 extern long bpf_jit_limit_max;
1264
1265 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
1266
1267 void bpf_jit_fill_hole_with_zero(void *area, unsigned int size);
1268
1269 struct bpf_binary_header *
1270 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1271 unsigned int alignment,
1272 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1273 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
1274 u64 bpf_jit_alloc_exec_limit(void);
1275 void *bpf_jit_alloc_exec(unsigned long size);
1276 void bpf_jit_free_exec(void *addr);
1277 void bpf_jit_free(struct bpf_prog *fp);
1278 struct bpf_binary_header *
1279 bpf_jit_binary_pack_hdr(const struct bpf_prog *fp);
1280
1281 void *bpf_prog_pack_alloc(u32 size, bpf_jit_fill_hole_t bpf_fill_ill_insns);
1282 void bpf_prog_pack_free(void *ptr, u32 size);
1283
bpf_prog_kallsyms_verify_off(const struct bpf_prog * fp)1284 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
1285 {
1286 return list_empty(&fp->aux->ksym.lnode) ||
1287 fp->aux->ksym.lnode.prev == LIST_POISON2;
1288 }
1289
1290 struct bpf_binary_header *
1291 bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image,
1292 unsigned int alignment,
1293 struct bpf_binary_header **rw_hdr,
1294 u8 **rw_image,
1295 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1296 int bpf_jit_binary_pack_finalize(struct bpf_binary_header *ro_header,
1297 struct bpf_binary_header *rw_header);
1298 void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header,
1299 struct bpf_binary_header *rw_header);
1300
1301 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1302 struct bpf_jit_poke_descriptor *poke);
1303
1304 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1305 const struct bpf_insn *insn, bool extra_pass,
1306 u64 *func_addr, bool *func_addr_fixed);
1307
1308 const char *bpf_jit_get_prog_name(struct bpf_prog *prog);
1309
1310 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
1311 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
1312
bpf_jit_dump(unsigned int flen,unsigned int proglen,u32 pass,void * image)1313 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
1314 u32 pass, void *image)
1315 {
1316 pr_err("flen=%u proglen=%u pass=%u image=%p from=%s pid=%d\n", flen,
1317 proglen, pass, image, current->comm, task_pid_nr(current));
1318
1319 if (image)
1320 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
1321 16, 1, image, proglen, false);
1322 }
1323
bpf_jit_is_ebpf(void)1324 static inline bool bpf_jit_is_ebpf(void)
1325 {
1326 # ifdef CONFIG_HAVE_EBPF_JIT
1327 return true;
1328 # else
1329 return false;
1330 # endif
1331 }
1332
ebpf_jit_enabled(void)1333 static inline bool ebpf_jit_enabled(void)
1334 {
1335 return bpf_jit_enable && bpf_jit_is_ebpf();
1336 }
1337
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1338 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1339 {
1340 return fp->jited && bpf_jit_is_ebpf();
1341 }
1342
bpf_jit_blinding_enabled(struct bpf_prog * prog)1343 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1344 {
1345 /* These are the prerequisites, should someone ever have the
1346 * idea to call blinding outside of them, we make sure to
1347 * bail out.
1348 */
1349 if (!bpf_jit_is_ebpf())
1350 return false;
1351 if (!prog->jit_requested)
1352 return false;
1353 if (!bpf_jit_harden)
1354 return false;
1355 if (bpf_jit_harden == 1 && bpf_token_capable(prog->aux->token, CAP_BPF))
1356 return false;
1357
1358 return true;
1359 }
1360
bpf_jit_kallsyms_enabled(void)1361 static inline bool bpf_jit_kallsyms_enabled(void)
1362 {
1363 /* There are a couple of corner cases where kallsyms should
1364 * not be enabled f.e. on hardening.
1365 */
1366 if (bpf_jit_harden)
1367 return false;
1368 if (!bpf_jit_kallsyms)
1369 return false;
1370 if (bpf_jit_kallsyms == 1)
1371 return true;
1372
1373 return false;
1374 }
1375
1376 int __bpf_address_lookup(unsigned long addr, unsigned long *size,
1377 unsigned long *off, char *sym);
1378 bool is_bpf_text_address(unsigned long addr);
1379 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1380 char *sym);
1381 struct bpf_prog *bpf_prog_ksym_find(unsigned long addr);
1382
1383 static inline int
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1384 bpf_address_lookup(unsigned long addr, unsigned long *size,
1385 unsigned long *off, char **modname, char *sym)
1386 {
1387 int ret = __bpf_address_lookup(addr, size, off, sym);
1388
1389 if (ret && modname)
1390 *modname = NULL;
1391 return ret;
1392 }
1393
1394 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1395 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1396
1397 #else /* CONFIG_BPF_JIT */
1398
ebpf_jit_enabled(void)1399 static inline bool ebpf_jit_enabled(void)
1400 {
1401 return false;
1402 }
1403
bpf_jit_blinding_enabled(struct bpf_prog * prog)1404 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1405 {
1406 return false;
1407 }
1408
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1409 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1410 {
1411 return false;
1412 }
1413
1414 static inline int
bpf_jit_add_poke_descriptor(struct bpf_prog * prog,struct bpf_jit_poke_descriptor * poke)1415 bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1416 struct bpf_jit_poke_descriptor *poke)
1417 {
1418 return -ENOTSUPP;
1419 }
1420
bpf_jit_free(struct bpf_prog * fp)1421 static inline void bpf_jit_free(struct bpf_prog *fp)
1422 {
1423 bpf_prog_unlock_free(fp);
1424 }
1425
bpf_jit_kallsyms_enabled(void)1426 static inline bool bpf_jit_kallsyms_enabled(void)
1427 {
1428 return false;
1429 }
1430
1431 static inline int
__bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char * sym)1432 __bpf_address_lookup(unsigned long addr, unsigned long *size,
1433 unsigned long *off, char *sym)
1434 {
1435 return 0;
1436 }
1437
is_bpf_text_address(unsigned long addr)1438 static inline bool is_bpf_text_address(unsigned long addr)
1439 {
1440 return false;
1441 }
1442
bpf_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * sym)1443 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1444 char *type, char *sym)
1445 {
1446 return -ERANGE;
1447 }
1448
bpf_prog_ksym_find(unsigned long addr)1449 static inline struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
1450 {
1451 return NULL;
1452 }
1453
1454 static inline int
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1455 bpf_address_lookup(unsigned long addr, unsigned long *size,
1456 unsigned long *off, char **modname, char *sym)
1457 {
1458 return 0;
1459 }
1460
bpf_prog_kallsyms_add(struct bpf_prog * fp)1461 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1462 {
1463 }
1464
bpf_prog_kallsyms_del(struct bpf_prog * fp)1465 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1466 {
1467 }
1468
1469 #endif /* CONFIG_BPF_JIT */
1470
1471 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1472
1473 #define BPF_ANC BIT(15)
1474
bpf_needs_clear_a(const struct sock_filter * first)1475 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1476 {
1477 switch (first->code) {
1478 case BPF_RET | BPF_K:
1479 case BPF_LD | BPF_W | BPF_LEN:
1480 return false;
1481
1482 case BPF_LD | BPF_W | BPF_ABS:
1483 case BPF_LD | BPF_H | BPF_ABS:
1484 case BPF_LD | BPF_B | BPF_ABS:
1485 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1486 return true;
1487 return false;
1488
1489 default:
1490 return true;
1491 }
1492 }
1493
bpf_anc_helper(const struct sock_filter * ftest)1494 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1495 {
1496 BUG_ON(ftest->code & BPF_ANC);
1497
1498 switch (ftest->code) {
1499 case BPF_LD | BPF_W | BPF_ABS:
1500 case BPF_LD | BPF_H | BPF_ABS:
1501 case BPF_LD | BPF_B | BPF_ABS:
1502 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1503 return BPF_ANC | SKF_AD_##CODE
1504 switch (ftest->k) {
1505 BPF_ANCILLARY(PROTOCOL);
1506 BPF_ANCILLARY(PKTTYPE);
1507 BPF_ANCILLARY(IFINDEX);
1508 BPF_ANCILLARY(NLATTR);
1509 BPF_ANCILLARY(NLATTR_NEST);
1510 BPF_ANCILLARY(MARK);
1511 BPF_ANCILLARY(QUEUE);
1512 BPF_ANCILLARY(HATYPE);
1513 BPF_ANCILLARY(RXHASH);
1514 BPF_ANCILLARY(CPU);
1515 BPF_ANCILLARY(ALU_XOR_X);
1516 BPF_ANCILLARY(VLAN_TAG);
1517 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1518 BPF_ANCILLARY(PAY_OFFSET);
1519 BPF_ANCILLARY(RANDOM);
1520 BPF_ANCILLARY(VLAN_TPID);
1521 }
1522 fallthrough;
1523 default:
1524 return ftest->code;
1525 }
1526 }
1527
1528 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1529 int k, unsigned int size);
1530
bpf_tell_extensions(void)1531 static inline int bpf_tell_extensions(void)
1532 {
1533 return SKF_AD_MAX;
1534 }
1535
1536 struct bpf_sock_addr_kern {
1537 struct sock *sk;
1538 struct sockaddr *uaddr;
1539 /* Temporary "register" to make indirect stores to nested structures
1540 * defined above. We need three registers to make such a store, but
1541 * only two (src and dst) are available at convert_ctx_access time
1542 */
1543 u64 tmp_reg;
1544 void *t_ctx; /* Attach type specific context. */
1545 u32 uaddrlen;
1546 };
1547
1548 struct bpf_sock_ops_kern {
1549 struct sock *sk;
1550 union {
1551 u32 args[4];
1552 u32 reply;
1553 u32 replylong[4];
1554 };
1555 struct sk_buff *syn_skb;
1556 struct sk_buff *skb;
1557 void *skb_data_end;
1558 u8 op;
1559 u8 is_fullsock;
1560 u8 is_locked_tcp_sock;
1561 u8 remaining_opt_len;
1562 u64 temp; /* temp and everything after is not
1563 * initialized to 0 before calling
1564 * the BPF program. New fields that
1565 * should be initialized to 0 should
1566 * be inserted before temp.
1567 * temp is scratch storage used by
1568 * sock_ops_convert_ctx_access
1569 * as temporary storage of a register.
1570 */
1571 };
1572
1573 struct bpf_sysctl_kern {
1574 struct ctl_table_header *head;
1575 const struct ctl_table *table;
1576 void *cur_val;
1577 size_t cur_len;
1578 void *new_val;
1579 size_t new_len;
1580 int new_updated;
1581 int write;
1582 loff_t *ppos;
1583 /* Temporary "register" for indirect stores to ppos. */
1584 u64 tmp_reg;
1585 };
1586
1587 #define BPF_SOCKOPT_KERN_BUF_SIZE 32
1588 struct bpf_sockopt_buf {
1589 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE];
1590 };
1591
1592 struct bpf_sockopt_kern {
1593 struct sock *sk;
1594 u8 *optval;
1595 u8 *optval_end;
1596 s32 level;
1597 s32 optname;
1598 s32 optlen;
1599 /* for retval in struct bpf_cg_run_ctx */
1600 struct task_struct *current_task;
1601 /* Temporary "register" for indirect stores to ppos. */
1602 u64 tmp_reg;
1603 };
1604
1605 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len);
1606
1607 struct bpf_sk_lookup_kern {
1608 u16 family;
1609 u16 protocol;
1610 __be16 sport;
1611 u16 dport;
1612 struct {
1613 __be32 saddr;
1614 __be32 daddr;
1615 } v4;
1616 struct {
1617 const struct in6_addr *saddr;
1618 const struct in6_addr *daddr;
1619 } v6;
1620 struct sock *selected_sk;
1621 u32 ingress_ifindex;
1622 bool no_reuseport;
1623 };
1624
1625 extern struct static_key_false bpf_sk_lookup_enabled;
1626
1627 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup.
1628 *
1629 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and
1630 * SK_DROP. Their meaning is as follows:
1631 *
1632 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result
1633 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup
1634 * SK_DROP : terminate lookup with -ECONNREFUSED
1635 *
1636 * This macro aggregates return values and selected sockets from
1637 * multiple BPF programs according to following rules in order:
1638 *
1639 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk,
1640 * macro result is SK_PASS and last ctx.selected_sk is used.
1641 * 2. If any program returned SK_DROP return value,
1642 * macro result is SK_DROP.
1643 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL.
1644 *
1645 * Caller must ensure that the prog array is non-NULL, and that the
1646 * array as well as the programs it contains remain valid.
1647 */
1648 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \
1649 ({ \
1650 struct bpf_sk_lookup_kern *_ctx = &(ctx); \
1651 struct bpf_prog_array_item *_item; \
1652 struct sock *_selected_sk = NULL; \
1653 bool _no_reuseport = false; \
1654 struct bpf_prog *_prog; \
1655 bool _all_pass = true; \
1656 u32 _ret; \
1657 \
1658 migrate_disable(); \
1659 _item = &(array)->items[0]; \
1660 while ((_prog = READ_ONCE(_item->prog))) { \
1661 /* restore most recent selection */ \
1662 _ctx->selected_sk = _selected_sk; \
1663 _ctx->no_reuseport = _no_reuseport; \
1664 \
1665 _ret = func(_prog, _ctx); \
1666 if (_ret == SK_PASS && _ctx->selected_sk) { \
1667 /* remember last non-NULL socket */ \
1668 _selected_sk = _ctx->selected_sk; \
1669 _no_reuseport = _ctx->no_reuseport; \
1670 } else if (_ret == SK_DROP && _all_pass) { \
1671 _all_pass = false; \
1672 } \
1673 _item++; \
1674 } \
1675 _ctx->selected_sk = _selected_sk; \
1676 _ctx->no_reuseport = _no_reuseport; \
1677 migrate_enable(); \
1678 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \
1679 })
1680
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)1681 static inline bool bpf_sk_lookup_run_v4(const struct net *net, int protocol,
1682 const __be32 saddr, const __be16 sport,
1683 const __be32 daddr, const u16 dport,
1684 const int ifindex, struct sock **psk)
1685 {
1686 struct bpf_prog_array *run_array;
1687 struct sock *selected_sk = NULL;
1688 bool no_reuseport = false;
1689
1690 rcu_read_lock();
1691 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1692 if (run_array) {
1693 struct bpf_sk_lookup_kern ctx = {
1694 .family = AF_INET,
1695 .protocol = protocol,
1696 .v4.saddr = saddr,
1697 .v4.daddr = daddr,
1698 .sport = sport,
1699 .dport = dport,
1700 .ingress_ifindex = ifindex,
1701 };
1702 u32 act;
1703
1704 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1705 if (act == SK_PASS) {
1706 selected_sk = ctx.selected_sk;
1707 no_reuseport = ctx.no_reuseport;
1708 } else {
1709 selected_sk = ERR_PTR(-ECONNREFUSED);
1710 }
1711 }
1712 rcu_read_unlock();
1713 *psk = selected_sk;
1714 return no_reuseport;
1715 }
1716
1717 #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)1718 static inline bool bpf_sk_lookup_run_v6(const struct net *net, int protocol,
1719 const struct in6_addr *saddr,
1720 const __be16 sport,
1721 const struct in6_addr *daddr,
1722 const u16 dport,
1723 const int ifindex, struct sock **psk)
1724 {
1725 struct bpf_prog_array *run_array;
1726 struct sock *selected_sk = NULL;
1727 bool no_reuseport = false;
1728
1729 rcu_read_lock();
1730 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1731 if (run_array) {
1732 struct bpf_sk_lookup_kern ctx = {
1733 .family = AF_INET6,
1734 .protocol = protocol,
1735 .v6.saddr = saddr,
1736 .v6.daddr = daddr,
1737 .sport = sport,
1738 .dport = dport,
1739 .ingress_ifindex = ifindex,
1740 };
1741 u32 act;
1742
1743 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1744 if (act == SK_PASS) {
1745 selected_sk = ctx.selected_sk;
1746 no_reuseport = ctx.no_reuseport;
1747 } else {
1748 selected_sk = ERR_PTR(-ECONNREFUSED);
1749 }
1750 }
1751 rcu_read_unlock();
1752 *psk = selected_sk;
1753 return no_reuseport;
1754 }
1755 #endif /* IS_ENABLED(CONFIG_IPV6) */
1756
__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))1757 static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 index,
1758 u64 flags, const u64 flag_mask,
1759 void *lookup_elem(struct bpf_map *map, u32 key))
1760 {
1761 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
1762 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX;
1763
1764 /* Lower bits of the flags are used as return code on lookup failure */
1765 if (unlikely(flags & ~(action_mask | flag_mask)))
1766 return XDP_ABORTED;
1767
1768 ri->tgt_value = lookup_elem(map, index);
1769 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) {
1770 /* If the lookup fails we want to clear out the state in the
1771 * redirect_info struct completely, so that if an eBPF program
1772 * performs multiple lookups, the last one always takes
1773 * precedence.
1774 */
1775 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */
1776 ri->map_type = BPF_MAP_TYPE_UNSPEC;
1777 return flags & action_mask;
1778 }
1779
1780 ri->tgt_index = index;
1781 ri->map_id = map->id;
1782 ri->map_type = map->map_type;
1783
1784 if (flags & BPF_F_BROADCAST) {
1785 WRITE_ONCE(ri->map, map);
1786 ri->flags = flags;
1787 } else {
1788 WRITE_ONCE(ri->map, NULL);
1789 ri->flags = 0;
1790 }
1791
1792 return XDP_REDIRECT;
1793 }
1794
1795 #ifdef CONFIG_NET
1796 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len);
1797 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from,
1798 u32 len, u64 flags);
1799 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len);
1800 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len);
1801 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len);
1802 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
1803 void *buf, unsigned long len, bool flush);
1804 void *bpf_skb_meta_pointer(struct sk_buff *skb, u32 offset);
1805 #else /* CONFIG_NET */
__bpf_skb_load_bytes(const struct sk_buff * skb,u32 offset,void * to,u32 len)1806 static inline int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset,
1807 void *to, u32 len)
1808 {
1809 return -EOPNOTSUPP;
1810 }
1811
__bpf_skb_store_bytes(struct sk_buff * skb,u32 offset,const void * from,u32 len,u64 flags)1812 static inline int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset,
1813 const void *from, u32 len, u64 flags)
1814 {
1815 return -EOPNOTSUPP;
1816 }
1817
__bpf_xdp_load_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)1818 static inline int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset,
1819 void *buf, u32 len)
1820 {
1821 return -EOPNOTSUPP;
1822 }
1823
__bpf_xdp_store_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)1824 static inline int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset,
1825 void *buf, u32 len)
1826 {
1827 return -EOPNOTSUPP;
1828 }
1829
bpf_xdp_pointer(struct xdp_buff * xdp,u32 offset,u32 len)1830 static inline void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
1831 {
1832 return NULL;
1833 }
1834
bpf_xdp_copy_buf(struct xdp_buff * xdp,unsigned long off,void * buf,unsigned long len,bool flush)1835 static inline void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, void *buf,
1836 unsigned long len, bool flush)
1837 {
1838 }
1839
bpf_skb_meta_pointer(struct sk_buff * skb,u32 offset)1840 static inline void *bpf_skb_meta_pointer(struct sk_buff *skb, u32 offset)
1841 {
1842 return ERR_PTR(-EOPNOTSUPP);
1843 }
1844 #endif /* CONFIG_NET */
1845
1846 #endif /* __LINUX_FILTER_H__ */
1847