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 <stdarg.h>
9
10 #include <linux/atomic.h>
11 #include <linux/refcount.h>
12 #include <linux/compat.h>
13 #include <linux/skbuff.h>
14 #include <linux/linkage.h>
15 #include <linux/printk.h>
16 #include <linux/workqueue.h>
17 #include <linux/sched.h>
18 #include <linux/capability.h>
19 #include <linux/cryptohash.h>
20 #include <linux/set_memory.h>
21 #include <linux/kallsyms.h>
22 #include <linux/if_vlan.h>
23
24 #include <net/sch_generic.h>
25
26 #include <uapi/linux/filter.h>
27 #include <uapi/linux/bpf.h>
28
29 struct sk_buff;
30 struct sock;
31 struct seccomp_data;
32 struct bpf_prog_aux;
33 struct xdp_rxq_info;
34 struct xdp_buff;
35 struct sock_reuseport;
36
37 /* ArgX, context and stack frame pointer register positions. Note,
38 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
39 * calls in BPF_CALL instruction.
40 */
41 #define BPF_REG_ARG1 BPF_REG_1
42 #define BPF_REG_ARG2 BPF_REG_2
43 #define BPF_REG_ARG3 BPF_REG_3
44 #define BPF_REG_ARG4 BPF_REG_4
45 #define BPF_REG_ARG5 BPF_REG_5
46 #define BPF_REG_CTX BPF_REG_6
47 #define BPF_REG_FP BPF_REG_10
48
49 /* Additional register mappings for converted user programs. */
50 #define BPF_REG_A BPF_REG_0
51 #define BPF_REG_X BPF_REG_7
52 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
53 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
54 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
55
56 /* Kernel hidden auxiliary/helper register. */
57 #define BPF_REG_AX MAX_BPF_REG
58 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1)
59 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
60
61 /* unused opcode to mark special call to bpf_tail_call() helper */
62 #define BPF_TAIL_CALL 0xf0
63
64 /* unused opcode to mark call to interpreter with arguments */
65 #define BPF_CALL_ARGS 0xe0
66
67 /* unused opcode to mark speculation barrier for mitigating
68 * Speculative Store Bypass
69 */
70 #define BPF_NOSPEC 0xc0
71
72 /* As per nm, we expose JITed images as text (code) section for
73 * kallsyms. That way, tools like perf can find it to match
74 * addresses.
75 */
76 #define BPF_SYM_ELF_TYPE 't'
77
78 /* BPF program can access up to 512 bytes of stack space. */
79 #define MAX_BPF_STACK 512
80
81 /* Helper macros for filter block array initializers. */
82
83 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
84
85 #define BPF_ALU_REG(CLASS, OP, DST, SRC) \
86 ((struct bpf_insn) { \
87 .code = CLASS | BPF_OP(OP) | BPF_X, \
88 .dst_reg = DST, \
89 .src_reg = SRC, \
90 .off = 0, \
91 .imm = 0 })
92
93 #define BPF_ALU64_REG(OP, DST, SRC) \
94 ((struct bpf_insn) { \
95 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
96 .dst_reg = DST, \
97 .src_reg = SRC, \
98 .off = 0, \
99 .imm = 0 })
100
101 #define BPF_ALU32_REG(OP, DST, SRC) \
102 ((struct bpf_insn) { \
103 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
104 .dst_reg = DST, \
105 .src_reg = SRC, \
106 .off = 0, \
107 .imm = 0 })
108
109 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
110
111 #define BPF_ALU64_IMM(OP, DST, IMM) \
112 ((struct bpf_insn) { \
113 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
114 .dst_reg = DST, \
115 .src_reg = 0, \
116 .off = 0, \
117 .imm = IMM })
118
119 #define BPF_ALU32_IMM(OP, DST, IMM) \
120 ((struct bpf_insn) { \
121 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
122 .dst_reg = DST, \
123 .src_reg = 0, \
124 .off = 0, \
125 .imm = IMM })
126
127 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
128
129 #define BPF_ENDIAN(TYPE, DST, LEN) \
130 ((struct bpf_insn) { \
131 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
132 .dst_reg = DST, \
133 .src_reg = 0, \
134 .off = 0, \
135 .imm = LEN })
136
137 /* Short form of mov, dst_reg = src_reg */
138
139 #define BPF_MOV_REG(CLASS, DST, SRC) \
140 ((struct bpf_insn) { \
141 .code = CLASS | BPF_MOV | BPF_X, \
142 .dst_reg = DST, \
143 .src_reg = SRC, \
144 .off = 0, \
145 .imm = 0 })
146
147 #define BPF_MOV64_REG(DST, SRC) \
148 ((struct bpf_insn) { \
149 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
150 .dst_reg = DST, \
151 .src_reg = SRC, \
152 .off = 0, \
153 .imm = 0 })
154
155 #define BPF_MOV32_REG(DST, SRC) \
156 ((struct bpf_insn) { \
157 .code = BPF_ALU | BPF_MOV | BPF_X, \
158 .dst_reg = DST, \
159 .src_reg = SRC, \
160 .off = 0, \
161 .imm = 0 })
162
163 /* Short form of mov, dst_reg = imm32 */
164
165 #define BPF_MOV64_IMM(DST, IMM) \
166 ((struct bpf_insn) { \
167 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
168 .dst_reg = DST, \
169 .src_reg = 0, \
170 .off = 0, \
171 .imm = IMM })
172
173 #define BPF_MOV32_IMM(DST, IMM) \
174 ((struct bpf_insn) { \
175 .code = BPF_ALU | BPF_MOV | BPF_K, \
176 .dst_reg = DST, \
177 .src_reg = 0, \
178 .off = 0, \
179 .imm = IMM })
180
181 #define BPF_RAW_REG(insn, DST, SRC) \
182 ((struct bpf_insn) { \
183 .code = (insn).code, \
184 .dst_reg = DST, \
185 .src_reg = SRC, \
186 .off = (insn).off, \
187 .imm = (insn).imm })
188
189 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
190 #define BPF_LD_IMM64(DST, IMM) \
191 BPF_LD_IMM64_RAW(DST, 0, IMM)
192
193 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
194 ((struct bpf_insn) { \
195 .code = BPF_LD | BPF_DW | BPF_IMM, \
196 .dst_reg = DST, \
197 .src_reg = SRC, \
198 .off = 0, \
199 .imm = (__u32) (IMM) }), \
200 ((struct bpf_insn) { \
201 .code = 0, /* zero is reserved opcode */ \
202 .dst_reg = 0, \
203 .src_reg = 0, \
204 .off = 0, \
205 .imm = ((__u64) (IMM)) >> 32 })
206
207 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
208 #define BPF_LD_MAP_FD(DST, MAP_FD) \
209 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
210
211 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
212
213 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
214 ((struct bpf_insn) { \
215 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
216 .dst_reg = DST, \
217 .src_reg = SRC, \
218 .off = 0, \
219 .imm = IMM })
220
221 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
222 ((struct bpf_insn) { \
223 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
224 .dst_reg = DST, \
225 .src_reg = SRC, \
226 .off = 0, \
227 .imm = IMM })
228
229 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
230
231 #define BPF_LD_ABS(SIZE, IMM) \
232 ((struct bpf_insn) { \
233 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
234 .dst_reg = 0, \
235 .src_reg = 0, \
236 .off = 0, \
237 .imm = IMM })
238
239 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
240
241 #define BPF_LD_IND(SIZE, SRC, IMM) \
242 ((struct bpf_insn) { \
243 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
244 .dst_reg = 0, \
245 .src_reg = SRC, \
246 .off = 0, \
247 .imm = IMM })
248
249 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
250
251 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
252 ((struct bpf_insn) { \
253 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
254 .dst_reg = DST, \
255 .src_reg = SRC, \
256 .off = OFF, \
257 .imm = 0 })
258
259 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
260
261 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
262 ((struct bpf_insn) { \
263 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
264 .dst_reg = DST, \
265 .src_reg = SRC, \
266 .off = OFF, \
267 .imm = 0 })
268
269 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
270
271 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
272 ((struct bpf_insn) { \
273 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
274 .dst_reg = DST, \
275 .src_reg = SRC, \
276 .off = OFF, \
277 .imm = 0 })
278
279 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
280
281 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
282 ((struct bpf_insn) { \
283 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
284 .dst_reg = DST, \
285 .src_reg = 0, \
286 .off = OFF, \
287 .imm = IMM })
288
289 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
290
291 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
292 ((struct bpf_insn) { \
293 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
294 .dst_reg = DST, \
295 .src_reg = SRC, \
296 .off = OFF, \
297 .imm = 0 })
298
299 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
300
301 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
302 ((struct bpf_insn) { \
303 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
304 .dst_reg = DST, \
305 .src_reg = 0, \
306 .off = OFF, \
307 .imm = IMM })
308
309 /* Unconditional jumps, goto pc + off16 */
310
311 #define BPF_JMP_A(OFF) \
312 ((struct bpf_insn) { \
313 .code = BPF_JMP | BPF_JA, \
314 .dst_reg = 0, \
315 .src_reg = 0, \
316 .off = OFF, \
317 .imm = 0 })
318
319 /* Relative call */
320
321 #define BPF_CALL_REL(TGT) \
322 ((struct bpf_insn) { \
323 .code = BPF_JMP | BPF_CALL, \
324 .dst_reg = 0, \
325 .src_reg = BPF_PSEUDO_CALL, \
326 .off = 0, \
327 .imm = TGT })
328
329 /* Function call */
330
331 #define BPF_CAST_CALL(x) \
332 ((u64 (*)(u64, u64, u64, u64, u64))(x))
333
334 #define BPF_EMIT_CALL(FUNC) \
335 ((struct bpf_insn) { \
336 .code = BPF_JMP | BPF_CALL, \
337 .dst_reg = 0, \
338 .src_reg = 0, \
339 .off = 0, \
340 .imm = ((FUNC) - __bpf_call_base) })
341
342 /* Raw code statement block */
343
344 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
345 ((struct bpf_insn) { \
346 .code = CODE, \
347 .dst_reg = DST, \
348 .src_reg = SRC, \
349 .off = OFF, \
350 .imm = IMM })
351
352 /* Program exit */
353
354 #define BPF_EXIT_INSN() \
355 ((struct bpf_insn) { \
356 .code = BPF_JMP | BPF_EXIT, \
357 .dst_reg = 0, \
358 .src_reg = 0, \
359 .off = 0, \
360 .imm = 0 })
361
362 /* Speculation barrier */
363
364 #define BPF_ST_NOSPEC() \
365 ((struct bpf_insn) { \
366 .code = BPF_ST | BPF_NOSPEC, \
367 .dst_reg = 0, \
368 .src_reg = 0, \
369 .off = 0, \
370 .imm = 0 })
371
372 /* Internal classic blocks for direct assignment */
373
374 #define __BPF_STMT(CODE, K) \
375 ((struct sock_filter) BPF_STMT(CODE, K))
376
377 #define __BPF_JUMP(CODE, K, JT, JF) \
378 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
379
380 #define bytes_to_bpf_size(bytes) \
381 ({ \
382 int bpf_size = -EINVAL; \
383 \
384 if (bytes == sizeof(u8)) \
385 bpf_size = BPF_B; \
386 else if (bytes == sizeof(u16)) \
387 bpf_size = BPF_H; \
388 else if (bytes == sizeof(u32)) \
389 bpf_size = BPF_W; \
390 else if (bytes == sizeof(u64)) \
391 bpf_size = BPF_DW; \
392 \
393 bpf_size; \
394 })
395
396 #define bpf_size_to_bytes(bpf_size) \
397 ({ \
398 int bytes = -EINVAL; \
399 \
400 if (bpf_size == BPF_B) \
401 bytes = sizeof(u8); \
402 else if (bpf_size == BPF_H) \
403 bytes = sizeof(u16); \
404 else if (bpf_size == BPF_W) \
405 bytes = sizeof(u32); \
406 else if (bpf_size == BPF_DW) \
407 bytes = sizeof(u64); \
408 \
409 bytes; \
410 })
411
412 #define BPF_SIZEOF(type) \
413 ({ \
414 const int __size = bytes_to_bpf_size(sizeof(type)); \
415 BUILD_BUG_ON(__size < 0); \
416 __size; \
417 })
418
419 #define BPF_FIELD_SIZEOF(type, field) \
420 ({ \
421 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
422 BUILD_BUG_ON(__size < 0); \
423 __size; \
424 })
425
426 #define BPF_LDST_BYTES(insn) \
427 ({ \
428 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
429 WARN_ON(__size < 0); \
430 __size; \
431 })
432
433 #define __BPF_MAP_0(m, v, ...) v
434 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
435 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
436 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
437 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
438 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
439
440 #define __BPF_REG_0(...) __BPF_PAD(5)
441 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
442 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
443 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
444 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
445 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
446
447 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
448 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
449
450 #define __BPF_CAST(t, a) \
451 (__force t) \
452 (__force \
453 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
454 (unsigned long)0, (t)0))) a
455 #define __BPF_V void
456 #define __BPF_N
457
458 #define __BPF_DECL_ARGS(t, a) t a
459 #define __BPF_DECL_REGS(t, a) u64 a
460
461 #define __BPF_PAD(n) \
462 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
463 u64, __ur_3, u64, __ur_4, u64, __ur_5)
464
465 #define BPF_CALL_x(x, name, ...) \
466 static __always_inline \
467 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
468 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
469 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
470 { \
471 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
472 } \
473 static __always_inline \
474 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
475
476 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
477 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
478 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
479 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
480 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
481 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
482
483 #define bpf_ctx_range(TYPE, MEMBER) \
484 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
485 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
486 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
487
488 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
489 ({ \
490 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE)); \
491 *(PTR_SIZE) = (SIZE); \
492 offsetof(TYPE, MEMBER); \
493 })
494
495 #ifdef CONFIG_COMPAT
496 /* A struct sock_filter is architecture independent. */
497 struct compat_sock_fprog {
498 u16 len;
499 compat_uptr_t filter; /* struct sock_filter * */
500 };
501 #endif
502
503 struct sock_fprog_kern {
504 u16 len;
505 struct sock_filter *filter;
506 };
507
508 struct bpf_binary_header {
509 u32 pages;
510 /* Some arches need word alignment for their instructions */
511 u8 image[] __aligned(4);
512 };
513
514 struct bpf_prog {
515 u16 pages; /* Number of allocated pages */
516 u16 jited:1, /* Is our filter JIT'ed? */
517 jit_requested:1,/* archs need to JIT the prog */
518 undo_set_mem:1, /* Passed set_memory_ro() checkpoint */
519 gpl_compatible:1, /* Is filter GPL compatible? */
520 cb_access:1, /* Is control block accessed? */
521 dst_needed:1, /* Do we need dst entry? */
522 blinded:1, /* Was blinded */
523 is_func:1, /* program is a bpf function */
524 kprobe_override:1, /* Do we override a kprobe? */
525 has_callchain_buf:1; /* callchain buffer allocated? */
526 enum bpf_prog_type type; /* Type of BPF program */
527 enum bpf_attach_type expected_attach_type; /* For some prog types */
528 u32 len; /* Number of filter blocks */
529 u32 jited_len; /* Size of jited insns in bytes */
530 u8 tag[BPF_TAG_SIZE];
531 struct bpf_prog_aux *aux; /* Auxiliary fields */
532 struct sock_fprog_kern *orig_prog; /* Original BPF program */
533 unsigned int (*bpf_func)(const void *ctx,
534 const struct bpf_insn *insn);
535 /* Instructions for interpreter */
536 union {
537 struct sock_filter insns[0];
538 struct bpf_insn insnsi[0];
539 };
540 };
541
542 struct sk_filter {
543 refcount_t refcnt;
544 struct rcu_head rcu;
545 struct bpf_prog *prog;
546 };
547
548 #define BPF_PROG_RUN(filter, ctx) (*(filter)->bpf_func)(ctx, (filter)->insnsi)
549
550 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
551
552 struct bpf_skb_data_end {
553 struct qdisc_skb_cb qdisc_cb;
554 void *data_meta;
555 void *data_end;
556 };
557
558 struct sk_msg_buff {
559 void *data;
560 void *data_end;
561 __u32 apply_bytes;
562 __u32 cork_bytes;
563 int sg_copybreak;
564 int sg_start;
565 int sg_curr;
566 int sg_end;
567 struct scatterlist sg_data[MAX_SKB_FRAGS];
568 bool sg_copy[MAX_SKB_FRAGS];
569 __u32 flags;
570 struct sock *sk_redir;
571 struct sock *sk;
572 struct sk_buff *skb;
573 struct list_head list;
574 };
575
576 struct bpf_redirect_info {
577 u32 ifindex;
578 u32 flags;
579 struct bpf_map *map;
580 struct bpf_map *map_to_flush;
581 u32 kern_flags;
582 };
583
584 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
585
586 /* flags for bpf_redirect_info kern_flags */
587 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
588
589 /* Compute the linear packet data range [data, data_end) which
590 * will be accessed by various program types (cls_bpf, act_bpf,
591 * lwt, ...). Subsystems allowing direct data access must (!)
592 * ensure that cb[] area can be written to when BPF program is
593 * invoked (otherwise cb[] save/restore is necessary).
594 */
bpf_compute_data_pointers(struct sk_buff * skb)595 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
596 {
597 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
598
599 BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
600 cb->data_meta = skb->data - skb_metadata_len(skb);
601 cb->data_end = skb->data + skb_headlen(skb);
602 }
603
bpf_skb_cb(struct sk_buff * skb)604 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
605 {
606 /* eBPF programs may read/write skb->cb[] area to transfer meta
607 * data between tail calls. Since this also needs to work with
608 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
609 *
610 * In some socket filter cases, the cb unfortunately needs to be
611 * saved/restored so that protocol specific skb->cb[] data won't
612 * be lost. In any case, due to unpriviledged eBPF programs
613 * attached to sockets, we need to clear the bpf_skb_cb() area
614 * to not leak previous contents to user space.
615 */
616 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
617 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
618 FIELD_SIZEOF(struct qdisc_skb_cb, data));
619
620 return qdisc_skb_cb(skb)->data;
621 }
622
bpf_prog_run_save_cb(const struct bpf_prog * prog,struct sk_buff * skb)623 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
624 struct sk_buff *skb)
625 {
626 u8 *cb_data = bpf_skb_cb(skb);
627 u8 cb_saved[BPF_SKB_CB_LEN];
628 u32 res;
629
630 if (unlikely(prog->cb_access)) {
631 memcpy(cb_saved, cb_data, sizeof(cb_saved));
632 memset(cb_data, 0, sizeof(cb_saved));
633 }
634
635 res = BPF_PROG_RUN(prog, skb);
636
637 if (unlikely(prog->cb_access))
638 memcpy(cb_data, cb_saved, sizeof(cb_saved));
639
640 return res;
641 }
642
bpf_prog_run_clear_cb(const struct bpf_prog * prog,struct sk_buff * skb)643 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
644 struct sk_buff *skb)
645 {
646 u8 *cb_data = bpf_skb_cb(skb);
647
648 if (unlikely(prog->cb_access))
649 memset(cb_data, 0, BPF_SKB_CB_LEN);
650
651 return BPF_PROG_RUN(prog, skb);
652 }
653
bpf_prog_run_xdp(const struct bpf_prog * prog,struct xdp_buff * xdp)654 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
655 struct xdp_buff *xdp)
656 {
657 /* Caller needs to hold rcu_read_lock() (!), otherwise program
658 * can be released while still running, or map elements could be
659 * freed early while still having concurrent users. XDP fastpath
660 * already takes rcu_read_lock() when fetching the program, so
661 * it's not necessary here anymore.
662 */
663 return BPF_PROG_RUN(prog, xdp);
664 }
665
bpf_prog_insn_size(const struct bpf_prog * prog)666 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
667 {
668 return prog->len * sizeof(struct bpf_insn);
669 }
670
bpf_prog_tag_scratch_size(const struct bpf_prog * prog)671 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
672 {
673 return round_up(bpf_prog_insn_size(prog) +
674 sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
675 }
676
bpf_prog_size(unsigned int proglen)677 static inline unsigned int bpf_prog_size(unsigned int proglen)
678 {
679 return max(sizeof(struct bpf_prog),
680 offsetof(struct bpf_prog, insns[proglen]));
681 }
682
bpf_prog_was_classic(const struct bpf_prog * prog)683 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
684 {
685 /* When classic BPF programs have been loaded and the arch
686 * does not have a classic BPF JIT (anymore), they have been
687 * converted via bpf_migrate_filter() to eBPF and thus always
688 * have an unspec program type.
689 */
690 return prog->type == BPF_PROG_TYPE_UNSPEC;
691 }
692
bpf_ctx_off_adjust_machine(u32 size)693 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
694 {
695 const u32 size_machine = sizeof(unsigned long);
696
697 if (size > size_machine && size % size_machine == 0)
698 size = size_machine;
699
700 return size;
701 }
702
703 static inline bool
bpf_ctx_narrow_access_ok(u32 off,u32 size,u32 size_default)704 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
705 {
706 return size <= size_default && (size & (size - 1)) == 0;
707 }
708
709 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
710
bpf_prog_lock_ro(struct bpf_prog * fp)711 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
712 {
713 fp->undo_set_mem = 1;
714 set_memory_ro((unsigned long)fp, fp->pages);
715 }
716
bpf_prog_unlock_ro(struct bpf_prog * fp)717 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
718 {
719 if (fp->undo_set_mem)
720 set_memory_rw((unsigned long)fp, fp->pages);
721 }
722
bpf_jit_binary_lock_ro(struct bpf_binary_header * hdr)723 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
724 {
725 set_memory_ro((unsigned long)hdr, hdr->pages);
726 set_memory_x((unsigned long)hdr, hdr->pages);
727 }
728
bpf_jit_binary_unlock_ro(struct bpf_binary_header * hdr)729 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
730 {
731 set_memory_rw((unsigned long)hdr, hdr->pages);
732 }
733
734 static inline struct bpf_binary_header *
bpf_jit_binary_hdr(const struct bpf_prog * fp)735 bpf_jit_binary_hdr(const struct bpf_prog *fp)
736 {
737 unsigned long real_start = (unsigned long)fp->bpf_func;
738 unsigned long addr = real_start & PAGE_MASK;
739
740 return (void *)addr;
741 }
742
743 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
sk_filter(struct sock * sk,struct sk_buff * skb)744 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
745 {
746 return sk_filter_trim_cap(sk, skb, 1);
747 }
748
749 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
750 void bpf_prog_free(struct bpf_prog *fp);
751
752 bool bpf_opcode_in_insntable(u8 code);
753
754 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
755 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
756 gfp_t gfp_extra_flags);
757 void __bpf_prog_free(struct bpf_prog *fp);
758
bpf_prog_unlock_free(struct bpf_prog * fp)759 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
760 {
761 bpf_prog_unlock_ro(fp);
762 __bpf_prog_free(fp);
763 }
764
765 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
766 unsigned int flen);
767
768 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
769 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
770 bpf_aux_classic_check_t trans, bool save_orig);
771 void bpf_prog_destroy(struct bpf_prog *fp);
772
773 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
774 int sk_attach_bpf(u32 ufd, struct sock *sk);
775 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
776 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
777 void sk_reuseport_prog_free(struct bpf_prog *prog);
778 int sk_detach_filter(struct sock *sk);
779 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
780 unsigned int len);
781
782 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
783 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
784
785 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
786 #define __bpf_call_base_args \
787 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
788 (void *)__bpf_call_base)
789
790 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
791 void bpf_jit_compile(struct bpf_prog *prog);
792 bool bpf_helper_changes_pkt_data(void *func);
793
bpf_dump_raw_ok(const struct cred * cred)794 static inline bool bpf_dump_raw_ok(const struct cred *cred)
795 {
796 /* Reconstruction of call-sites is dependent on kallsyms,
797 * thus make dump the same restriction.
798 */
799 return kallsyms_show_value(cred);
800 }
801
802 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
803 const struct bpf_insn *patch, u32 len);
804
805 void bpf_clear_redirect_map(struct bpf_map *map);
806
xdp_return_frame_no_direct(void)807 static inline bool xdp_return_frame_no_direct(void)
808 {
809 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
810
811 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
812 }
813
xdp_set_return_frame_no_direct(void)814 static inline void xdp_set_return_frame_no_direct(void)
815 {
816 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
817
818 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
819 }
820
xdp_clear_return_frame_no_direct(void)821 static inline void xdp_clear_return_frame_no_direct(void)
822 {
823 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
824
825 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
826 }
827
xdp_ok_fwd_dev(const struct net_device * fwd,unsigned int pktlen)828 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
829 unsigned int pktlen)
830 {
831 unsigned int len;
832
833 if (unlikely(!(fwd->flags & IFF_UP)))
834 return -ENETDOWN;
835
836 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
837 if (pktlen > len)
838 return -EMSGSIZE;
839
840 return 0;
841 }
842
843 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
844 * same cpu context. Further for best results no more than a single map
845 * for the do_redirect/do_flush pair should be used. This limitation is
846 * because we only track one map and force a flush when the map changes.
847 * This does not appear to be a real limitation for existing software.
848 */
849 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
850 struct xdp_buff *xdp, struct bpf_prog *prog);
851 int xdp_do_redirect(struct net_device *dev,
852 struct xdp_buff *xdp,
853 struct bpf_prog *prog);
854 void xdp_do_flush_map(void);
855
856 void bpf_warn_invalid_xdp_action(u32 act);
857
858 struct sock *do_sk_redirect_map(struct sk_buff *skb);
859 struct sock *do_msg_redirect_map(struct sk_msg_buff *md);
860
861 #ifdef CONFIG_INET
862 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
863 struct bpf_prog *prog, struct sk_buff *skb,
864 u32 hash);
865 #else
866 static inline struct sock *
bpf_run_sk_reuseport(struct sock_reuseport * reuse,struct sock * sk,struct bpf_prog * prog,struct sk_buff * skb,u32 hash)867 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
868 struct bpf_prog *prog, struct sk_buff *skb,
869 u32 hash)
870 {
871 return NULL;
872 }
873 #endif
874
875 #ifdef CONFIG_BPF_JIT
876 extern int bpf_jit_enable;
877 extern int bpf_jit_harden;
878 extern int bpf_jit_kallsyms;
879 extern long bpf_jit_limit;
880 extern long bpf_jit_limit_max;
881
882 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
883
884 struct bpf_binary_header *
885 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
886 unsigned int alignment,
887 bpf_jit_fill_hole_t bpf_fill_ill_insns);
888 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
889 u64 bpf_jit_alloc_exec_limit(void);
890 void *bpf_jit_alloc_exec(unsigned long size);
891 void bpf_jit_free_exec(void *addr);
892 void bpf_jit_free(struct bpf_prog *fp);
893
894 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
895 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
896
bpf_jit_dump(unsigned int flen,unsigned int proglen,u32 pass,void * image)897 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
898 u32 pass, void *image)
899 {
900 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
901 proglen, pass, image, current->comm, task_pid_nr(current));
902
903 if (image)
904 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
905 16, 1, image, proglen, false);
906 }
907
bpf_jit_is_ebpf(void)908 static inline bool bpf_jit_is_ebpf(void)
909 {
910 # ifdef CONFIG_HAVE_EBPF_JIT
911 return true;
912 # else
913 return false;
914 # endif
915 }
916
ebpf_jit_enabled(void)917 static inline bool ebpf_jit_enabled(void)
918 {
919 return bpf_jit_enable && bpf_jit_is_ebpf();
920 }
921
bpf_prog_ebpf_jited(const struct bpf_prog * fp)922 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
923 {
924 return fp->jited && bpf_jit_is_ebpf();
925 }
926
bpf_jit_blinding_enabled(struct bpf_prog * prog)927 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
928 {
929 /* These are the prerequisites, should someone ever have the
930 * idea to call blinding outside of them, we make sure to
931 * bail out.
932 */
933 if (!bpf_jit_is_ebpf())
934 return false;
935 if (!prog->jit_requested)
936 return false;
937 if (!bpf_jit_harden)
938 return false;
939 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
940 return false;
941
942 return true;
943 }
944
bpf_jit_kallsyms_enabled(void)945 static inline bool bpf_jit_kallsyms_enabled(void)
946 {
947 /* There are a couple of corner cases where kallsyms should
948 * not be enabled f.e. on hardening.
949 */
950 if (bpf_jit_harden)
951 return false;
952 if (!bpf_jit_kallsyms)
953 return false;
954 if (bpf_jit_kallsyms == 1)
955 return true;
956
957 return false;
958 }
959
960 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
961 unsigned long *off, char *sym);
962 bool is_bpf_text_address(unsigned long addr);
963 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
964 char *sym);
965
966 static inline const char *
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)967 bpf_address_lookup(unsigned long addr, unsigned long *size,
968 unsigned long *off, char **modname, char *sym)
969 {
970 const char *ret = __bpf_address_lookup(addr, size, off, sym);
971
972 if (ret && modname)
973 *modname = NULL;
974 return ret;
975 }
976
977 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
978 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
979
980 #else /* CONFIG_BPF_JIT */
981
ebpf_jit_enabled(void)982 static inline bool ebpf_jit_enabled(void)
983 {
984 return false;
985 }
986
bpf_prog_ebpf_jited(const struct bpf_prog * fp)987 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
988 {
989 return false;
990 }
991
bpf_jit_free(struct bpf_prog * fp)992 static inline void bpf_jit_free(struct bpf_prog *fp)
993 {
994 bpf_prog_unlock_free(fp);
995 }
996
bpf_jit_kallsyms_enabled(void)997 static inline bool bpf_jit_kallsyms_enabled(void)
998 {
999 return false;
1000 }
1001
1002 static inline const char *
__bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char * sym)1003 __bpf_address_lookup(unsigned long addr, unsigned long *size,
1004 unsigned long *off, char *sym)
1005 {
1006 return NULL;
1007 }
1008
is_bpf_text_address(unsigned long addr)1009 static inline bool is_bpf_text_address(unsigned long addr)
1010 {
1011 return false;
1012 }
1013
bpf_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * sym)1014 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1015 char *type, char *sym)
1016 {
1017 return -ERANGE;
1018 }
1019
1020 static inline const char *
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1021 bpf_address_lookup(unsigned long addr, unsigned long *size,
1022 unsigned long *off, char **modname, char *sym)
1023 {
1024 return NULL;
1025 }
1026
bpf_prog_kallsyms_add(struct bpf_prog * fp)1027 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1028 {
1029 }
1030
bpf_prog_kallsyms_del(struct bpf_prog * fp)1031 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1032 {
1033 }
1034 #endif /* CONFIG_BPF_JIT */
1035
1036 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp);
1037 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1038
1039 #define BPF_ANC BIT(15)
1040
bpf_needs_clear_a(const struct sock_filter * first)1041 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1042 {
1043 switch (first->code) {
1044 case BPF_RET | BPF_K:
1045 case BPF_LD | BPF_W | BPF_LEN:
1046 return false;
1047
1048 case BPF_LD | BPF_W | BPF_ABS:
1049 case BPF_LD | BPF_H | BPF_ABS:
1050 case BPF_LD | BPF_B | BPF_ABS:
1051 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1052 return true;
1053 return false;
1054
1055 default:
1056 return true;
1057 }
1058 }
1059
bpf_anc_helper(const struct sock_filter * ftest)1060 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1061 {
1062 BUG_ON(ftest->code & BPF_ANC);
1063
1064 switch (ftest->code) {
1065 case BPF_LD | BPF_W | BPF_ABS:
1066 case BPF_LD | BPF_H | BPF_ABS:
1067 case BPF_LD | BPF_B | BPF_ABS:
1068 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1069 return BPF_ANC | SKF_AD_##CODE
1070 switch (ftest->k) {
1071 BPF_ANCILLARY(PROTOCOL);
1072 BPF_ANCILLARY(PKTTYPE);
1073 BPF_ANCILLARY(IFINDEX);
1074 BPF_ANCILLARY(NLATTR);
1075 BPF_ANCILLARY(NLATTR_NEST);
1076 BPF_ANCILLARY(MARK);
1077 BPF_ANCILLARY(QUEUE);
1078 BPF_ANCILLARY(HATYPE);
1079 BPF_ANCILLARY(RXHASH);
1080 BPF_ANCILLARY(CPU);
1081 BPF_ANCILLARY(ALU_XOR_X);
1082 BPF_ANCILLARY(VLAN_TAG);
1083 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1084 BPF_ANCILLARY(PAY_OFFSET);
1085 BPF_ANCILLARY(RANDOM);
1086 BPF_ANCILLARY(VLAN_TPID);
1087 }
1088 /* Fallthrough. */
1089 default:
1090 return ftest->code;
1091 }
1092 }
1093
1094 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1095 int k, unsigned int size);
1096
bpf_load_pointer(const struct sk_buff * skb,int k,unsigned int size,void * buffer)1097 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
1098 unsigned int size, void *buffer)
1099 {
1100 if (k >= 0)
1101 return skb_header_pointer(skb, k, size, buffer);
1102
1103 return bpf_internal_load_pointer_neg_helper(skb, k, size);
1104 }
1105
bpf_tell_extensions(void)1106 static inline int bpf_tell_extensions(void)
1107 {
1108 return SKF_AD_MAX;
1109 }
1110
1111 struct bpf_sock_addr_kern {
1112 struct sock *sk;
1113 struct sockaddr *uaddr;
1114 /* Temporary "register" to make indirect stores to nested structures
1115 * defined above. We need three registers to make such a store, but
1116 * only two (src and dst) are available at convert_ctx_access time
1117 */
1118 u64 tmp_reg;
1119 void *t_ctx; /* Attach type specific context. */
1120 };
1121
1122 struct bpf_sock_ops_kern {
1123 struct sock *sk;
1124 u32 op;
1125 union {
1126 u32 args[4];
1127 u32 reply;
1128 u32 replylong[4];
1129 };
1130 u32 is_fullsock;
1131 u64 temp; /* temp and everything after is not
1132 * initialized to 0 before calling
1133 * the BPF program. New fields that
1134 * should be initialized to 0 should
1135 * be inserted before temp.
1136 * temp is scratch storage used by
1137 * sock_ops_convert_ctx_access
1138 * as temporary storage of a register.
1139 */
1140 };
1141
1142 #endif /* __LINUX_FILTER_H__ */
1143