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/capability.h>
18#include <linux/set_memory.h>
19#include <linux/kallsyms.h>
20#include <linux/if_vlan.h>
21#include <linux/vmalloc.h>
22#include <linux/sockptr.h>
23#include <crypto/sha1.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
31struct sk_buff;
32struct sock;
33struct seccomp_data;
34struct bpf_prog_aux;
35struct xdp_rxq_info;
36struct xdp_buff;
37struct sock_reuseport;
38struct ctl_table;
39struct 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 call to interpreter with arguments */
72#define BPF_CALL_ARGS 0xe0
73
74/* unused opcode to mark speculation barrier for mitigating
75 * Speculative Store Bypass
76 */
77#define BPF_NOSPEC 0xc0
78
79/* As per nm, we expose JITed images as text (code) section for
80 * kallsyms. That way, tools like perf can find it to match
81 * addresses.
82 */
83#define BPF_SYM_ELF_TYPE 't'
84
85/* BPF program can access up to 512 bytes of stack space. */
86#define MAX_BPF_STACK 512
87
88/* Helper macros for filter block array initializers. */
89
90/* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
91
92#define BPF_ALU64_REG(OP, DST, SRC) \
93 ((struct bpf_insn) { \
94 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
95 .dst_reg = DST, \
96 .src_reg = SRC, \
97 .off = 0, \
98 .imm = 0 })
99
100#define BPF_ALU32_REG(OP, DST, SRC) \
101 ((struct bpf_insn) { \
102 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
103 .dst_reg = DST, \
104 .src_reg = SRC, \
105 .off = 0, \
106 .imm = 0 })
107
108/* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
109
110#define BPF_ALU64_IMM(OP, DST, IMM) \
111 ((struct bpf_insn) { \
112 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
113 .dst_reg = DST, \
114 .src_reg = 0, \
115 .off = 0, \
116 .imm = IMM })
117
118#define BPF_ALU32_IMM(OP, DST, IMM) \
119 ((struct bpf_insn) { \
120 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
121 .dst_reg = DST, \
122 .src_reg = 0, \
123 .off = 0, \
124 .imm = IMM })
125
126/* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
127
128#define BPF_ENDIAN(TYPE, DST, LEN) \
129 ((struct bpf_insn) { \
130 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
131 .dst_reg = DST, \
132 .src_reg = 0, \
133 .off = 0, \
134 .imm = LEN })
135
136/* Short form of mov, dst_reg = src_reg */
137
138#define BPF_MOV64_REG(DST, SRC) \
139 ((struct bpf_insn) { \
140 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
141 .dst_reg = DST, \
142 .src_reg = SRC, \
143 .off = 0, \
144 .imm = 0 })
145
146#define BPF_MOV32_REG(DST, SRC) \
147 ((struct bpf_insn) { \
148 .code = BPF_ALU | BPF_MOV | BPF_X, \
149 .dst_reg = DST, \
150 .src_reg = SRC, \
151 .off = 0, \
152 .imm = 0 })
153
154/* Short form of mov, dst_reg = imm32 */
155
156#define BPF_MOV64_IMM(DST, IMM) \
157 ((struct bpf_insn) { \
158 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
159 .dst_reg = DST, \
160 .src_reg = 0, \
161 .off = 0, \
162 .imm = IMM })
163
164#define BPF_MOV32_IMM(DST, IMM) \
165 ((struct bpf_insn) { \
166 .code = BPF_ALU | BPF_MOV | BPF_K, \
167 .dst_reg = DST, \
168 .src_reg = 0, \
169 .off = 0, \
170 .imm = IMM })
171
172/* Special form of mov32, used for doing explicit zero extension on dst. */
173#define BPF_ZEXT_REG(DST) \
174 ((struct bpf_insn) { \
175 .code = BPF_ALU | BPF_MOV | BPF_X, \
176 .dst_reg = DST, \
177 .src_reg = DST, \
178 .off = 0, \
179 .imm = 1 })
180
181static inline bool insn_is_zext(const struct bpf_insn *insn)
182{
183 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1;
184}
185
186/* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
187#define BPF_LD_IMM64(DST, IMM) \
188 BPF_LD_IMM64_RAW(DST, 0, IMM)
189
190#define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
191 ((struct bpf_insn) { \
192 .code = BPF_LD | BPF_DW | BPF_IMM, \
193 .dst_reg = DST, \
194 .src_reg = SRC, \
195 .off = 0, \
196 .imm = (__u32) (IMM) }), \
197 ((struct bpf_insn) { \
198 .code = 0, /* zero is reserved opcode */ \
199 .dst_reg = 0, \
200 .src_reg = 0, \
201 .off = 0, \
202 .imm = ((__u64) (IMM)) >> 32 })
203
204/* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
205#define BPF_LD_MAP_FD(DST, MAP_FD) \
206 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
207
208/* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
209
210#define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
211 ((struct bpf_insn) { \
212 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
213 .dst_reg = DST, \
214 .src_reg = SRC, \
215 .off = 0, \
216 .imm = IMM })
217
218#define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
219 ((struct bpf_insn) { \
220 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
221 .dst_reg = DST, \
222 .src_reg = SRC, \
223 .off = 0, \
224 .imm = IMM })
225
226/* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
227
228#define BPF_LD_ABS(SIZE, IMM) \
229 ((struct bpf_insn) { \
230 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
231 .dst_reg = 0, \
232 .src_reg = 0, \
233 .off = 0, \
234 .imm = IMM })
235
236/* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
237
238#define BPF_LD_IND(SIZE, SRC, IMM) \
239 ((struct bpf_insn) { \
240 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
241 .dst_reg = 0, \
242 .src_reg = SRC, \
243 .off = 0, \
244 .imm = IMM })
245
246/* Memory load, dst_reg = *(uint *) (src_reg + off16) */
247
248#define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
249 ((struct bpf_insn) { \
250 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
251 .dst_reg = DST, \
252 .src_reg = SRC, \
253 .off = OFF, \
254 .imm = 0 })
255
256/* Memory store, *(uint *) (dst_reg + off16) = src_reg */
257
258#define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
259 ((struct bpf_insn) { \
260 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
261 .dst_reg = DST, \
262 .src_reg = SRC, \
263 .off = OFF, \
264 .imm = 0 })
265
266
267/*
268 * Atomic operations:
269 *
270 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg
271 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg
272 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg
273 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg
274 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg);
275 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg);
276 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg);
277 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg);
278 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg)
279 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg)
280 */
281
282#define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \
283 ((struct bpf_insn) { \
284 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \
285 .dst_reg = DST, \
286 .src_reg = SRC, \
287 .off = OFF, \
288 .imm = OP })
289
290/* Legacy alias */
291#define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF)
292
293/* Memory store, *(uint *) (dst_reg + off16) = imm32 */
294
295#define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
296 ((struct bpf_insn) { \
297 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
298 .dst_reg = DST, \
299 .src_reg = 0, \
300 .off = OFF, \
301 .imm = IMM })
302
303/* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
304
305#define BPF_JMP_REG(OP, DST, SRC, OFF) \
306 ((struct bpf_insn) { \
307 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
308 .dst_reg = DST, \
309 .src_reg = SRC, \
310 .off = OFF, \
311 .imm = 0 })
312
313/* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
314
315#define BPF_JMP_IMM(OP, DST, IMM, OFF) \
316 ((struct bpf_insn) { \
317 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
318 .dst_reg = DST, \
319 .src_reg = 0, \
320 .off = OFF, \
321 .imm = IMM })
322
323/* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
324
325#define BPF_JMP32_REG(OP, DST, SRC, OFF) \
326 ((struct bpf_insn) { \
327 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
328 .dst_reg = DST, \
329 .src_reg = SRC, \
330 .off = OFF, \
331 .imm = 0 })
332
333/* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
334
335#define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
336 ((struct bpf_insn) { \
337 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
338 .dst_reg = DST, \
339 .src_reg = 0, \
340 .off = OFF, \
341 .imm = IMM })
342
343/* Unconditional jumps, goto pc + off16 */
344
345#define BPF_JMP_A(OFF) \
346 ((struct bpf_insn) { \
347 .code = BPF_JMP | BPF_JA, \
348 .dst_reg = 0, \
349 .src_reg = 0, \
350 .off = OFF, \
351 .imm = 0 })
352
353/* Relative call */
354
355#define BPF_CALL_REL(TGT) \
356 ((struct bpf_insn) { \
357 .code = BPF_JMP | BPF_CALL, \
358 .dst_reg = 0, \
359 .src_reg = BPF_PSEUDO_CALL, \
360 .off = 0, \
361 .imm = TGT })
362
363/* Convert function address to BPF immediate */
364
365#define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base)
366
367#define BPF_EMIT_CALL(FUNC) \
368 ((struct bpf_insn) { \
369 .code = BPF_JMP | BPF_CALL, \
370 .dst_reg = 0, \
371 .src_reg = 0, \
372 .off = 0, \
373 .imm = BPF_CALL_IMM(FUNC) })
374
375/* Raw code statement block */
376
377#define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
378 ((struct bpf_insn) { \
379 .code = CODE, \
380 .dst_reg = DST, \
381 .src_reg = SRC, \
382 .off = OFF, \
383 .imm = IMM })
384
385/* Program exit */
386
387#define BPF_EXIT_INSN() \
388 ((struct bpf_insn) { \
389 .code = BPF_JMP | BPF_EXIT, \
390 .dst_reg = 0, \
391 .src_reg = 0, \
392 .off = 0, \
393 .imm = 0 })
394
395/* Speculation barrier */
396
397#define BPF_ST_NOSPEC() \
398 ((struct bpf_insn) { \
399 .code = BPF_ST | BPF_NOSPEC, \
400 .dst_reg = 0, \
401 .src_reg = 0, \
402 .off = 0, \
403 .imm = 0 })
404
405/* Internal classic blocks for direct assignment */
406
407#define __BPF_STMT(CODE, K) \
408 ((struct sock_filter) BPF_STMT(CODE, K))
409
410#define __BPF_JUMP(CODE, K, JT, JF) \
411 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
412
413#define bytes_to_bpf_size(bytes) \
414({ \
415 int bpf_size = -EINVAL; \
416 \
417 if (bytes == sizeof(u8)) \
418 bpf_size = BPF_B; \
419 else if (bytes == sizeof(u16)) \
420 bpf_size = BPF_H; \
421 else if (bytes == sizeof(u32)) \
422 bpf_size = BPF_W; \
423 else if (bytes == sizeof(u64)) \
424 bpf_size = BPF_DW; \
425 \
426 bpf_size; \
427})
428
429#define bpf_size_to_bytes(bpf_size) \
430({ \
431 int bytes = -EINVAL; \
432 \
433 if (bpf_size == BPF_B) \
434 bytes = sizeof(u8); \
435 else if (bpf_size == BPF_H) \
436 bytes = sizeof(u16); \
437 else if (bpf_size == BPF_W) \
438 bytes = sizeof(u32); \
439 else if (bpf_size == BPF_DW) \
440 bytes = sizeof(u64); \
441 \
442 bytes; \
443})
444
445#define BPF_SIZEOF(type) \
446 ({ \
447 const int __size = bytes_to_bpf_size(sizeof(type)); \
448 BUILD_BUG_ON(__size < 0); \
449 __size; \
450 })
451
452#define BPF_FIELD_SIZEOF(type, field) \
453 ({ \
454 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
455 BUILD_BUG_ON(__size < 0); \
456 __size; \
457 })
458
459#define BPF_LDST_BYTES(insn) \
460 ({ \
461 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
462 WARN_ON(__size < 0); \
463 __size; \
464 })
465
466#define __BPF_MAP_0(m, v, ...) v
467#define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
468#define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
469#define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
470#define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
471#define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
472
473#define __BPF_REG_0(...) __BPF_PAD(5)
474#define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
475#define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
476#define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
477#define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
478#define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
479
480#define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
481#define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
482
483#define __BPF_CAST(t, a) \
484 (__force t) \
485 (__force \
486 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
487 (unsigned long)0, (t)0))) a
488#define __BPF_V void
489#define __BPF_N
490
491#define __BPF_DECL_ARGS(t, a) t a
492#define __BPF_DECL_REGS(t, a) u64 a
493
494#define __BPF_PAD(n) \
495 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
496 u64, __ur_3, u64, __ur_4, u64, __ur_5)
497
498#define BPF_CALL_x(x, name, ...) \
499 static __always_inline \
500 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
501 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
502 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
503 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
504 { \
505 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
506 } \
507 static __always_inline \
508 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
509
510#define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
511#define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
512#define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
513#define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
514#define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
515#define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
516
517#define bpf_ctx_range(TYPE, MEMBER) \
518 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
519#define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
520 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
521#if BITS_PER_LONG == 64
522# define bpf_ctx_range_ptr(TYPE, MEMBER) \
523 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
524#else
525# define bpf_ctx_range_ptr(TYPE, MEMBER) \
526 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
527#endif /* BITS_PER_LONG == 64 */
528
529#define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
530 ({ \
531 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \
532 *(PTR_SIZE) = (SIZE); \
533 offsetof(TYPE, MEMBER); \
534 })
535
536/* A struct sock_filter is architecture independent. */
537struct compat_sock_fprog {
538 u16 len;
539 compat_uptr_t filter; /* struct sock_filter * */
540};
541
542struct sock_fprog_kern {
543 u16 len;
544 struct sock_filter *filter;
545};
546
547/* Some arches need doubleword alignment for their instructions and/or data */
548#define BPF_IMAGE_ALIGNMENT 8
549
550struct bpf_binary_header {
551 u32 size;
552 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT);
553};
554
555struct bpf_prog_stats {
556 u64_stats_t cnt;
557 u64_stats_t nsecs;
558 u64_stats_t misses;
559 struct u64_stats_sync syncp;
560} __aligned(2 * sizeof(u64));
561
562struct sk_filter {
563 refcount_t refcnt;
564 struct rcu_head rcu;
565 struct bpf_prog *prog;
566};
567
568DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
569
570typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx,
571 const struct bpf_insn *insnsi,
572 unsigned int (*bpf_func)(const void *,
573 const struct bpf_insn *));
574
575static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog,
576 const void *ctx,
577 bpf_dispatcher_fn dfunc)
578{
579 u32 ret;
580
581 cant_migrate();
582 if (static_branch_unlikely(&bpf_stats_enabled_key)) {
583 struct bpf_prog_stats *stats;
584 u64 start = sched_clock();
585 unsigned long flags;
586
587 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
588 stats = this_cpu_ptr(prog->stats);
589 flags = u64_stats_update_begin_irqsave(&stats->syncp);
590 u64_stats_inc(&stats->cnt);
591 u64_stats_add(&stats->nsecs, sched_clock() - start);
592 u64_stats_update_end_irqrestore(&stats->syncp, flags);
593 } else {
594 ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
595 }
596 return ret;
597}
598
599static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx)
600{
601 return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func);
602}
603
604/*
605 * Use in preemptible and therefore migratable context to make sure that
606 * the execution of the BPF program runs on one CPU.
607 *
608 * This uses migrate_disable/enable() explicitly to document that the
609 * invocation of a BPF program does not require reentrancy protection
610 * against a BPF program which is invoked from a preempting task.
611 */
612static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
613 const void *ctx)
614{
615 u32 ret;
616
617 migrate_disable();
618 ret = bpf_prog_run(prog, ctx);
619 migrate_enable();
620 return ret;
621}
622
623#define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
624
625struct bpf_skb_data_end {
626 struct qdisc_skb_cb qdisc_cb;
627 void *data_meta;
628 void *data_end;
629};
630
631struct bpf_nh_params {
632 u32 nh_family;
633 union {
634 u32 ipv4_nh;
635 struct in6_addr ipv6_nh;
636 };
637};
638
639struct bpf_redirect_info {
640 u32 flags;
641 u32 tgt_index;
642 void *tgt_value;
643 struct bpf_map *map;
644 u32 map_id;
645 enum bpf_map_type map_type;
646 u32 kern_flags;
647 struct bpf_nh_params nh;
648};
649
650DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
651
652/* flags for bpf_redirect_info kern_flags */
653#define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
654
655/* Compute the linear packet data range [data, data_end) which
656 * will be accessed by various program types (cls_bpf, act_bpf,
657 * lwt, ...). Subsystems allowing direct data access must (!)
658 * ensure that cb[] area can be written to when BPF program is
659 * invoked (otherwise cb[] save/restore is necessary).
660 */
661static inline void bpf_compute_data_pointers(struct sk_buff *skb)
662{
663 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
664
665 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
666 cb->data_meta = skb->data - skb_metadata_len(skb);
667 cb->data_end = skb->data + skb_headlen(skb);
668}
669
670/* Similar to bpf_compute_data_pointers(), except that save orginal
671 * data in cb->data and cb->meta_data for restore.
672 */
673static inline void bpf_compute_and_save_data_end(
674 struct sk_buff *skb, void **saved_data_end)
675{
676 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
677
678 *saved_data_end = cb->data_end;
679 cb->data_end = skb->data + skb_headlen(skb);
680}
681
682/* Restore data saved by bpf_compute_data_pointers(). */
683static inline void bpf_restore_data_end(
684 struct sk_buff *skb, void *saved_data_end)
685{
686 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
687
688 cb->data_end = saved_data_end;
689}
690
691static inline u8 *bpf_skb_cb(const struct sk_buff *skb)
692{
693 /* eBPF programs may read/write skb->cb[] area to transfer meta
694 * data between tail calls. Since this also needs to work with
695 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
696 *
697 * In some socket filter cases, the cb unfortunately needs to be
698 * saved/restored so that protocol specific skb->cb[] data won't
699 * be lost. In any case, due to unpriviledged eBPF programs
700 * attached to sockets, we need to clear the bpf_skb_cb() area
701 * to not leak previous contents to user space.
702 */
703 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
704 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) !=
705 sizeof_field(struct qdisc_skb_cb, data));
706
707 return qdisc_skb_cb(skb)->data;
708}
709
710/* Must be invoked with migration disabled */
711static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
712 const void *ctx)
713{
714 const struct sk_buff *skb = ctx;
715 u8 *cb_data = bpf_skb_cb(skb);
716 u8 cb_saved[BPF_SKB_CB_LEN];
717 u32 res;
718
719 if (unlikely(prog->cb_access)) {
720 memcpy(cb_saved, cb_data, sizeof(cb_saved));
721 memset(cb_data, 0, sizeof(cb_saved));
722 }
723
724 res = bpf_prog_run(prog, skb);
725
726 if (unlikely(prog->cb_access))
727 memcpy(cb_data, cb_saved, sizeof(cb_saved));
728
729 return res;
730}
731
732static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
733 struct sk_buff *skb)
734{
735 u32 res;
736
737 migrate_disable();
738 res = __bpf_prog_run_save_cb(prog, skb);
739 migrate_enable();
740 return res;
741}
742
743static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
744 struct sk_buff *skb)
745{
746 u8 *cb_data = bpf_skb_cb(skb);
747 u32 res;
748
749 if (unlikely(prog->cb_access))
750 memset(cb_data, 0, BPF_SKB_CB_LEN);
751
752 res = bpf_prog_run_pin_on_cpu(prog, skb);
753 return res;
754}
755
756DECLARE_BPF_DISPATCHER(xdp)
757
758DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
759
760u32 xdp_master_redirect(struct xdp_buff *xdp);
761
762static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
763 struct xdp_buff *xdp)
764{
765 /* Driver XDP hooks are invoked within a single NAPI poll cycle and thus
766 * under local_bh_disable(), which provides the needed RCU protection
767 * for accessing map entries.
768 */
769 u32 act = __bpf_prog_run(prog, xdp, BPF_DISPATCHER_FUNC(xdp));
770
771 if (static_branch_unlikely(&bpf_master_redirect_enabled_key)) {
772 if (act == XDP_TX && netif_is_bond_slave(xdp->rxq->dev))
773 act = xdp_master_redirect(xdp);
774 }
775
776 return act;
777}
778
779void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog);
780
781static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
782{
783 return prog->len * sizeof(struct bpf_insn);
784}
785
786static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
787{
788 return round_up(bpf_prog_insn_size(prog) +
789 sizeof(__be64) + 1, SHA1_BLOCK_SIZE);
790}
791
792static inline unsigned int bpf_prog_size(unsigned int proglen)
793{
794 return max(sizeof(struct bpf_prog),
795 offsetof(struct bpf_prog, insns[proglen]));
796}
797
798static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
799{
800 /* When classic BPF programs have been loaded and the arch
801 * does not have a classic BPF JIT (anymore), they have been
802 * converted via bpf_migrate_filter() to eBPF and thus always
803 * have an unspec program type.
804 */
805 return prog->type == BPF_PROG_TYPE_UNSPEC;
806}
807
808static inline u32 bpf_ctx_off_adjust_machine(u32 size)
809{
810 const u32 size_machine = sizeof(unsigned long);
811
812 if (size > size_machine && size % size_machine == 0)
813 size = size_machine;
814
815 return size;
816}
817
818static inline bool
819bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
820{
821 return size <= size_default && (size & (size - 1)) == 0;
822}
823
824static inline u8
825bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
826{
827 u8 access_off = off & (size_default - 1);
828
829#ifdef __LITTLE_ENDIAN
830 return access_off;
831#else
832 return size_default - (access_off + size);
833#endif
834}
835
836#define bpf_ctx_wide_access_ok(off, size, type, field) \
837 (size == sizeof(__u64) && \
838 off >= offsetof(type, field) && \
839 off + sizeof(__u64) <= offsetofend(type, field) && \
840 off % sizeof(__u64) == 0)
841
842#define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
843
844static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
845{
846#ifndef CONFIG_BPF_JIT_ALWAYS_ON
847 if (!fp->jited) {
848 set_vm_flush_reset_perms(fp);
849 set_memory_ro((unsigned long)fp, fp->pages);
850 }
851#endif
852}
853
854static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
855{
856 set_vm_flush_reset_perms(hdr);
857 set_memory_ro((unsigned long)hdr, hdr->size >> PAGE_SHIFT);
858 set_memory_x((unsigned long)hdr, hdr->size >> PAGE_SHIFT);
859}
860
861int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
862static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
863{
864 return sk_filter_trim_cap(sk, skb, 1);
865}
866
867struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
868void bpf_prog_free(struct bpf_prog *fp);
869
870bool bpf_opcode_in_insntable(u8 code);
871
872void bpf_prog_free_linfo(struct bpf_prog *prog);
873void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
874 const u32 *insn_to_jit_off);
875int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
876void bpf_prog_jit_attempt_done(struct bpf_prog *prog);
877
878struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
879struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
880struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
881 gfp_t gfp_extra_flags);
882void __bpf_prog_free(struct bpf_prog *fp);
883
884static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
885{
886 __bpf_prog_free(fp);
887}
888
889typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
890 unsigned int flen);
891
892int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
893int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
894 bpf_aux_classic_check_t trans, bool save_orig);
895void bpf_prog_destroy(struct bpf_prog *fp);
896
897int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
898int sk_attach_bpf(u32 ufd, struct sock *sk);
899int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
900int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
901void sk_reuseport_prog_free(struct bpf_prog *prog);
902int sk_detach_filter(struct sock *sk);
903int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
904 unsigned int len);
905
906bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
907void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
908
909u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
910#define __bpf_call_base_args \
911 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
912 (void *)__bpf_call_base)
913
914struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
915void bpf_jit_compile(struct bpf_prog *prog);
916bool bpf_jit_needs_zext(void);
917bool bpf_jit_supports_subprog_tailcalls(void);
918bool bpf_jit_supports_kfunc_call(void);
919bool bpf_helper_changes_pkt_data(void *func);
920
921static inline bool bpf_dump_raw_ok(const struct cred *cred)
922{
923 /* Reconstruction of call-sites is dependent on kallsyms,
924 * thus make dump the same restriction.
925 */
926 return kallsyms_show_value(cred);
927}
928
929struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
930 const struct bpf_insn *patch, u32 len);
931int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
932
933void bpf_clear_redirect_map(struct bpf_map *map);
934
935static inline bool xdp_return_frame_no_direct(void)
936{
937 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
938
939 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
940}
941
942static inline void xdp_set_return_frame_no_direct(void)
943{
944 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
945
946 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
947}
948
949static inline void xdp_clear_return_frame_no_direct(void)
950{
951 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
952
953 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
954}
955
956static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
957 unsigned int pktlen)
958{
959 unsigned int len;
960
961 if (unlikely(!(fwd->flags & IFF_UP)))
962 return -ENETDOWN;
963
964 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
965 if (pktlen > len)
966 return -EMSGSIZE;
967
968 return 0;
969}
970
971/* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the
972 * same cpu context. Further for best results no more than a single map
973 * for the do_redirect/do_flush pair should be used. This limitation is
974 * because we only track one map and force a flush when the map changes.
975 * This does not appear to be a real limitation for existing software.
976 */
977int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
978 struct xdp_buff *xdp, struct bpf_prog *prog);
979int xdp_do_redirect(struct net_device *dev,
980 struct xdp_buff *xdp,
981 struct bpf_prog *prog);
982int xdp_do_redirect_frame(struct net_device *dev,
983 struct xdp_buff *xdp,
984 struct xdp_frame *xdpf,
985 struct bpf_prog *prog);
986void xdp_do_flush(void);
987
988/* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as
989 * it is no longer only flushing maps. Keep this define for compatibility
990 * until all drivers are updated - do not use xdp_do_flush_map() in new code!
991 */
992#define xdp_do_flush_map xdp_do_flush
993
994void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act);
995
996#ifdef CONFIG_INET
997struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
998 struct bpf_prog *prog, struct sk_buff *skb,
999 struct sock *migrating_sk,
1000 u32 hash);
1001#else
1002static inline struct sock *
1003bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1004 struct bpf_prog *prog, struct sk_buff *skb,
1005 struct sock *migrating_sk,
1006 u32 hash)
1007{
1008 return NULL;
1009}
1010#endif
1011
1012#ifdef CONFIG_BPF_JIT
1013extern int bpf_jit_enable;
1014extern int bpf_jit_harden;
1015extern int bpf_jit_kallsyms;
1016extern long bpf_jit_limit;
1017extern long bpf_jit_limit_max;
1018
1019typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
1020
1021struct bpf_binary_header *
1022bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1023 unsigned int alignment,
1024 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1025void bpf_jit_binary_free(struct bpf_binary_header *hdr);
1026u64 bpf_jit_alloc_exec_limit(void);
1027void *bpf_jit_alloc_exec(unsigned long size);
1028void bpf_jit_free_exec(void *addr);
1029void bpf_jit_free(struct bpf_prog *fp);
1030struct bpf_binary_header *
1031bpf_jit_binary_pack_hdr(const struct bpf_prog *fp);
1032
1033static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog *fp)
1034{
1035 return list_empty(&fp->aux->ksym.lnode) ||
1036 fp->aux->ksym.lnode.prev == LIST_POISON2;
1037}
1038
1039struct bpf_binary_header *
1040bpf_jit_binary_pack_alloc(unsigned int proglen, u8 **ro_image,
1041 unsigned int alignment,
1042 struct bpf_binary_header **rw_hdr,
1043 u8 **rw_image,
1044 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1045int bpf_jit_binary_pack_finalize(struct bpf_prog *prog,
1046 struct bpf_binary_header *ro_header,
1047 struct bpf_binary_header *rw_header);
1048void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header,
1049 struct bpf_binary_header *rw_header);
1050
1051int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1052 struct bpf_jit_poke_descriptor *poke);
1053
1054int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1055 const struct bpf_insn *insn, bool extra_pass,
1056 u64 *func_addr, bool *func_addr_fixed);
1057
1058struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
1059void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
1060
1061static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
1062 u32 pass, void *image)
1063{
1064 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
1065 proglen, pass, image, current->comm, task_pid_nr(current));
1066
1067 if (image)
1068 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
1069 16, 1, image, proglen, false);
1070}
1071
1072static inline bool bpf_jit_is_ebpf(void)
1073{
1074# ifdef CONFIG_HAVE_EBPF_JIT
1075 return true;
1076# else
1077 return false;
1078# endif
1079}
1080
1081static inline bool ebpf_jit_enabled(void)
1082{
1083 return bpf_jit_enable && bpf_jit_is_ebpf();
1084}
1085
1086static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1087{
1088 return fp->jited && bpf_jit_is_ebpf();
1089}
1090
1091static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1092{
1093 /* These are the prerequisites, should someone ever have the
1094 * idea to call blinding outside of them, we make sure to
1095 * bail out.
1096 */
1097 if (!bpf_jit_is_ebpf())
1098 return false;
1099 if (!prog->jit_requested)
1100 return false;
1101 if (!bpf_jit_harden)
1102 return false;
1103 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
1104 return false;
1105
1106 return true;
1107}
1108
1109static inline bool bpf_jit_kallsyms_enabled(void)
1110{
1111 /* There are a couple of corner cases where kallsyms should
1112 * not be enabled f.e. on hardening.
1113 */
1114 if (bpf_jit_harden)
1115 return false;
1116 if (!bpf_jit_kallsyms)
1117 return false;
1118 if (bpf_jit_kallsyms == 1)
1119 return true;
1120
1121 return false;
1122}
1123
1124const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
1125 unsigned long *off, char *sym);
1126bool is_bpf_text_address(unsigned long addr);
1127int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1128 char *sym);
1129
1130static inline const char *
1131bpf_address_lookup(unsigned long addr, unsigned long *size,
1132 unsigned long *off, char **modname, char *sym)
1133{
1134 const char *ret = __bpf_address_lookup(addr, size, off, sym);
1135
1136 if (ret && modname)
1137 *modname = NULL;
1138 return ret;
1139}
1140
1141void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1142void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1143
1144#else /* CONFIG_BPF_JIT */
1145
1146static inline bool ebpf_jit_enabled(void)
1147{
1148 return false;
1149}
1150
1151static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1152{
1153 return false;
1154}
1155
1156static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1157{
1158 return false;
1159}
1160
1161static inline int
1162bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1163 struct bpf_jit_poke_descriptor *poke)
1164{
1165 return -ENOTSUPP;
1166}
1167
1168static inline void bpf_jit_free(struct bpf_prog *fp)
1169{
1170 bpf_prog_unlock_free(fp);
1171}
1172
1173static inline bool bpf_jit_kallsyms_enabled(void)
1174{
1175 return false;
1176}
1177
1178static inline const char *
1179__bpf_address_lookup(unsigned long addr, unsigned long *size,
1180 unsigned long *off, char *sym)
1181{
1182 return NULL;
1183}
1184
1185static inline bool is_bpf_text_address(unsigned long addr)
1186{
1187 return false;
1188}
1189
1190static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1191 char *type, char *sym)
1192{
1193 return -ERANGE;
1194}
1195
1196static inline const char *
1197bpf_address_lookup(unsigned long addr, unsigned long *size,
1198 unsigned long *off, char **modname, char *sym)
1199{
1200 return NULL;
1201}
1202
1203static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1204{
1205}
1206
1207static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1208{
1209}
1210
1211#endif /* CONFIG_BPF_JIT */
1212
1213void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1214
1215#define BPF_ANC BIT(15)
1216
1217static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1218{
1219 switch (first->code) {
1220 case BPF_RET | BPF_K:
1221 case BPF_LD | BPF_W | BPF_LEN:
1222 return false;
1223
1224 case BPF_LD | BPF_W | BPF_ABS:
1225 case BPF_LD | BPF_H | BPF_ABS:
1226 case BPF_LD | BPF_B | BPF_ABS:
1227 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1228 return true;
1229 return false;
1230
1231 default:
1232 return true;
1233 }
1234}
1235
1236static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1237{
1238 BUG_ON(ftest->code & BPF_ANC);
1239
1240 switch (ftest->code) {
1241 case BPF_LD | BPF_W | BPF_ABS:
1242 case BPF_LD | BPF_H | BPF_ABS:
1243 case BPF_LD | BPF_B | BPF_ABS:
1244#define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1245 return BPF_ANC | SKF_AD_##CODE
1246 switch (ftest->k) {
1247 BPF_ANCILLARY(PROTOCOL);
1248 BPF_ANCILLARY(PKTTYPE);
1249 BPF_ANCILLARY(IFINDEX);
1250 BPF_ANCILLARY(NLATTR);
1251 BPF_ANCILLARY(NLATTR_NEST);
1252 BPF_ANCILLARY(MARK);
1253 BPF_ANCILLARY(QUEUE);
1254 BPF_ANCILLARY(HATYPE);
1255 BPF_ANCILLARY(RXHASH);
1256 BPF_ANCILLARY(CPU);
1257 BPF_ANCILLARY(ALU_XOR_X);
1258 BPF_ANCILLARY(VLAN_TAG);
1259 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1260 BPF_ANCILLARY(PAY_OFFSET);
1261 BPF_ANCILLARY(RANDOM);
1262 BPF_ANCILLARY(VLAN_TPID);
1263 }
1264 fallthrough;
1265 default:
1266 return ftest->code;
1267 }
1268}
1269
1270void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1271 int k, unsigned int size);
1272
1273static inline int bpf_tell_extensions(void)
1274{
1275 return SKF_AD_MAX;
1276}
1277
1278struct bpf_sock_addr_kern {
1279 struct sock *sk;
1280 struct sockaddr *uaddr;
1281 /* Temporary "register" to make indirect stores to nested structures
1282 * defined above. We need three registers to make such a store, but
1283 * only two (src and dst) are available at convert_ctx_access time
1284 */
1285 u64 tmp_reg;
1286 void *t_ctx; /* Attach type specific context. */
1287};
1288
1289struct bpf_sock_ops_kern {
1290 struct sock *sk;
1291 union {
1292 u32 args[4];
1293 u32 reply;
1294 u32 replylong[4];
1295 };
1296 struct sk_buff *syn_skb;
1297 struct sk_buff *skb;
1298 void *skb_data_end;
1299 u8 op;
1300 u8 is_fullsock;
1301 u8 remaining_opt_len;
1302 u64 temp; /* temp and everything after is not
1303 * initialized to 0 before calling
1304 * the BPF program. New fields that
1305 * should be initialized to 0 should
1306 * be inserted before temp.
1307 * temp is scratch storage used by
1308 * sock_ops_convert_ctx_access
1309 * as temporary storage of a register.
1310 */
1311};
1312
1313struct bpf_sysctl_kern {
1314 struct ctl_table_header *head;
1315 struct ctl_table *table;
1316 void *cur_val;
1317 size_t cur_len;
1318 void *new_val;
1319 size_t new_len;
1320 int new_updated;
1321 int write;
1322 loff_t *ppos;
1323 /* Temporary "register" for indirect stores to ppos. */
1324 u64 tmp_reg;
1325};
1326
1327#define BPF_SOCKOPT_KERN_BUF_SIZE 32
1328struct bpf_sockopt_buf {
1329 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE];
1330};
1331
1332struct bpf_sockopt_kern {
1333 struct sock *sk;
1334 u8 *optval;
1335 u8 *optval_end;
1336 s32 level;
1337 s32 optname;
1338 s32 optlen;
1339 /* for retval in struct bpf_cg_run_ctx */
1340 struct task_struct *current_task;
1341 /* Temporary "register" for indirect stores to ppos. */
1342 u64 tmp_reg;
1343};
1344
1345int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len);
1346
1347struct bpf_sk_lookup_kern {
1348 u16 family;
1349 u16 protocol;
1350 __be16 sport;
1351 u16 dport;
1352 struct {
1353 __be32 saddr;
1354 __be32 daddr;
1355 } v4;
1356 struct {
1357 const struct in6_addr *saddr;
1358 const struct in6_addr *daddr;
1359 } v6;
1360 struct sock *selected_sk;
1361 u32 ingress_ifindex;
1362 bool no_reuseport;
1363};
1364
1365extern struct static_key_false bpf_sk_lookup_enabled;
1366
1367/* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup.
1368 *
1369 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and
1370 * SK_DROP. Their meaning is as follows:
1371 *
1372 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result
1373 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup
1374 * SK_DROP : terminate lookup with -ECONNREFUSED
1375 *
1376 * This macro aggregates return values and selected sockets from
1377 * multiple BPF programs according to following rules in order:
1378 *
1379 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk,
1380 * macro result is SK_PASS and last ctx.selected_sk is used.
1381 * 2. If any program returned SK_DROP return value,
1382 * macro result is SK_DROP.
1383 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL.
1384 *
1385 * Caller must ensure that the prog array is non-NULL, and that the
1386 * array as well as the programs it contains remain valid.
1387 */
1388#define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \
1389 ({ \
1390 struct bpf_sk_lookup_kern *_ctx = &(ctx); \
1391 struct bpf_prog_array_item *_item; \
1392 struct sock *_selected_sk = NULL; \
1393 bool _no_reuseport = false; \
1394 struct bpf_prog *_prog; \
1395 bool _all_pass = true; \
1396 u32 _ret; \
1397 \
1398 migrate_disable(); \
1399 _item = &(array)->items[0]; \
1400 while ((_prog = READ_ONCE(_item->prog))) { \
1401 /* restore most recent selection */ \
1402 _ctx->selected_sk = _selected_sk; \
1403 _ctx->no_reuseport = _no_reuseport; \
1404 \
1405 _ret = func(_prog, _ctx); \
1406 if (_ret == SK_PASS && _ctx->selected_sk) { \
1407 /* remember last non-NULL socket */ \
1408 _selected_sk = _ctx->selected_sk; \
1409 _no_reuseport = _ctx->no_reuseport; \
1410 } else if (_ret == SK_DROP && _all_pass) { \
1411 _all_pass = false; \
1412 } \
1413 _item++; \
1414 } \
1415 _ctx->selected_sk = _selected_sk; \
1416 _ctx->no_reuseport = _no_reuseport; \
1417 migrate_enable(); \
1418 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \
1419 })
1420
1421static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol,
1422 const __be32 saddr, const __be16 sport,
1423 const __be32 daddr, const u16 dport,
1424 const int ifindex, struct sock **psk)
1425{
1426 struct bpf_prog_array *run_array;
1427 struct sock *selected_sk = NULL;
1428 bool no_reuseport = false;
1429
1430 rcu_read_lock();
1431 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1432 if (run_array) {
1433 struct bpf_sk_lookup_kern ctx = {
1434 .family = AF_INET,
1435 .protocol = protocol,
1436 .v4.saddr = saddr,
1437 .v4.daddr = daddr,
1438 .sport = sport,
1439 .dport = dport,
1440 .ingress_ifindex = ifindex,
1441 };
1442 u32 act;
1443
1444 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1445 if (act == SK_PASS) {
1446 selected_sk = ctx.selected_sk;
1447 no_reuseport = ctx.no_reuseport;
1448 } else {
1449 selected_sk = ERR_PTR(-ECONNREFUSED);
1450 }
1451 }
1452 rcu_read_unlock();
1453 *psk = selected_sk;
1454 return no_reuseport;
1455}
1456
1457#if IS_ENABLED(CONFIG_IPV6)
1458static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol,
1459 const struct in6_addr *saddr,
1460 const __be16 sport,
1461 const struct in6_addr *daddr,
1462 const u16 dport,
1463 const int ifindex, struct sock **psk)
1464{
1465 struct bpf_prog_array *run_array;
1466 struct sock *selected_sk = NULL;
1467 bool no_reuseport = false;
1468
1469 rcu_read_lock();
1470 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1471 if (run_array) {
1472 struct bpf_sk_lookup_kern ctx = {
1473 .family = AF_INET6,
1474 .protocol = protocol,
1475 .v6.saddr = saddr,
1476 .v6.daddr = daddr,
1477 .sport = sport,
1478 .dport = dport,
1479 .ingress_ifindex = ifindex,
1480 };
1481 u32 act;
1482
1483 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, bpf_prog_run);
1484 if (act == SK_PASS) {
1485 selected_sk = ctx.selected_sk;
1486 no_reuseport = ctx.no_reuseport;
1487 } else {
1488 selected_sk = ERR_PTR(-ECONNREFUSED);
1489 }
1490 }
1491 rcu_read_unlock();
1492 *psk = selected_sk;
1493 return no_reuseport;
1494}
1495#endif /* IS_ENABLED(CONFIG_IPV6) */
1496
1497static __always_inline int __bpf_xdp_redirect_map(struct bpf_map *map, u32 ifindex,
1498 u64 flags, const u64 flag_mask,
1499 void *lookup_elem(struct bpf_map *map, u32 key))
1500{
1501 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
1502 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX;
1503
1504 /* Lower bits of the flags are used as return code on lookup failure */
1505 if (unlikely(flags & ~(action_mask | flag_mask)))
1506 return XDP_ABORTED;
1507
1508 ri->tgt_value = lookup_elem(map, ifindex);
1509 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) {
1510 /* If the lookup fails we want to clear out the state in the
1511 * redirect_info struct completely, so that if an eBPF program
1512 * performs multiple lookups, the last one always takes
1513 * precedence.
1514 */
1515 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */
1516 ri->map_type = BPF_MAP_TYPE_UNSPEC;
1517 return flags & action_mask;
1518 }
1519
1520 ri->tgt_index = ifindex;
1521 ri->map_id = map->id;
1522 ri->map_type = map->map_type;
1523
1524 if (flags & BPF_F_BROADCAST) {
1525 WRITE_ONCE(ri->map, map);
1526 ri->flags = flags;
1527 } else {
1528 WRITE_ONCE(ri->map, NULL);
1529 ri->flags = 0;
1530 }
1531
1532 return XDP_REDIRECT;
1533}
1534
1535#endif /* __LINUX_FILTER_H__ */
1536

source code of linux/include/linux/filter.h