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/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/sha.h>
25
26 #include <net/sch_generic.h>
27
28 #include <asm/byteorder.h>
29 #include <uapi/linux/filter.h>
30 #include <uapi/linux/bpf.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 call to interpreter with arguments */
73 #define BPF_CALL_ARGS 0xe0
74
75 /* As per nm, we expose JITed images as text (code) section for
76 * kallsyms. That way, tools like perf can find it to match
77 * addresses.
78 */
79 #define BPF_SYM_ELF_TYPE 't'
80
81 /* BPF program can access up to 512 bytes of stack space. */
82 #define MAX_BPF_STACK 512
83
84 /* Helper macros for filter block array initializers. */
85
86 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
87
88 #define BPF_ALU64_REG(OP, DST, SRC) \
89 ((struct bpf_insn) { \
90 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
91 .dst_reg = DST, \
92 .src_reg = SRC, \
93 .off = 0, \
94 .imm = 0 })
95
96 #define BPF_ALU32_REG(OP, DST, SRC) \
97 ((struct bpf_insn) { \
98 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
99 .dst_reg = DST, \
100 .src_reg = SRC, \
101 .off = 0, \
102 .imm = 0 })
103
104 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
105
106 #define BPF_ALU64_IMM(OP, DST, IMM) \
107 ((struct bpf_insn) { \
108 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
109 .dst_reg = DST, \
110 .src_reg = 0, \
111 .off = 0, \
112 .imm = IMM })
113
114 #define BPF_ALU32_IMM(OP, DST, IMM) \
115 ((struct bpf_insn) { \
116 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
117 .dst_reg = DST, \
118 .src_reg = 0, \
119 .off = 0, \
120 .imm = IMM })
121
122 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
123
124 #define BPF_ENDIAN(TYPE, DST, LEN) \
125 ((struct bpf_insn) { \
126 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
127 .dst_reg = DST, \
128 .src_reg = 0, \
129 .off = 0, \
130 .imm = LEN })
131
132 /* Short form of mov, dst_reg = src_reg */
133
134 #define BPF_MOV64_REG(DST, SRC) \
135 ((struct bpf_insn) { \
136 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
137 .dst_reg = DST, \
138 .src_reg = SRC, \
139 .off = 0, \
140 .imm = 0 })
141
142 #define BPF_MOV32_REG(DST, SRC) \
143 ((struct bpf_insn) { \
144 .code = BPF_ALU | BPF_MOV | BPF_X, \
145 .dst_reg = DST, \
146 .src_reg = SRC, \
147 .off = 0, \
148 .imm = 0 })
149
150 /* Short form of mov, dst_reg = imm32 */
151
152 #define BPF_MOV64_IMM(DST, IMM) \
153 ((struct bpf_insn) { \
154 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
155 .dst_reg = DST, \
156 .src_reg = 0, \
157 .off = 0, \
158 .imm = IMM })
159
160 #define BPF_MOV32_IMM(DST, IMM) \
161 ((struct bpf_insn) { \
162 .code = BPF_ALU | BPF_MOV | BPF_K, \
163 .dst_reg = DST, \
164 .src_reg = 0, \
165 .off = 0, \
166 .imm = IMM })
167
168 /* Special form of mov32, used for doing explicit zero extension on dst. */
169 #define BPF_ZEXT_REG(DST) \
170 ((struct bpf_insn) { \
171 .code = BPF_ALU | BPF_MOV | BPF_X, \
172 .dst_reg = DST, \
173 .src_reg = DST, \
174 .off = 0, \
175 .imm = 1 })
176
insn_is_zext(const struct bpf_insn * insn)177 static inline bool insn_is_zext(const struct bpf_insn *insn)
178 {
179 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1;
180 }
181
182 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
183 #define BPF_LD_IMM64(DST, IMM) \
184 BPF_LD_IMM64_RAW(DST, 0, IMM)
185
186 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
187 ((struct bpf_insn) { \
188 .code = BPF_LD | BPF_DW | BPF_IMM, \
189 .dst_reg = DST, \
190 .src_reg = SRC, \
191 .off = 0, \
192 .imm = (__u32) (IMM) }), \
193 ((struct bpf_insn) { \
194 .code = 0, /* zero is reserved opcode */ \
195 .dst_reg = 0, \
196 .src_reg = 0, \
197 .off = 0, \
198 .imm = ((__u64) (IMM)) >> 32 })
199
200 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
201 #define BPF_LD_MAP_FD(DST, MAP_FD) \
202 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
203
204 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
205
206 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
207 ((struct bpf_insn) { \
208 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
209 .dst_reg = DST, \
210 .src_reg = SRC, \
211 .off = 0, \
212 .imm = IMM })
213
214 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
215 ((struct bpf_insn) { \
216 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
217 .dst_reg = DST, \
218 .src_reg = SRC, \
219 .off = 0, \
220 .imm = IMM })
221
222 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
223
224 #define BPF_LD_ABS(SIZE, IMM) \
225 ((struct bpf_insn) { \
226 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
227 .dst_reg = 0, \
228 .src_reg = 0, \
229 .off = 0, \
230 .imm = IMM })
231
232 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
233
234 #define BPF_LD_IND(SIZE, SRC, IMM) \
235 ((struct bpf_insn) { \
236 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
237 .dst_reg = 0, \
238 .src_reg = SRC, \
239 .off = 0, \
240 .imm = IMM })
241
242 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
243
244 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
245 ((struct bpf_insn) { \
246 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
247 .dst_reg = DST, \
248 .src_reg = SRC, \
249 .off = OFF, \
250 .imm = 0 })
251
252 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
253
254 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
255 ((struct bpf_insn) { \
256 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
257 .dst_reg = DST, \
258 .src_reg = SRC, \
259 .off = OFF, \
260 .imm = 0 })
261
262 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
263
264 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
265 ((struct bpf_insn) { \
266 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
267 .dst_reg = DST, \
268 .src_reg = SRC, \
269 .off = OFF, \
270 .imm = 0 })
271
272 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
273
274 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
275 ((struct bpf_insn) { \
276 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
277 .dst_reg = DST, \
278 .src_reg = 0, \
279 .off = OFF, \
280 .imm = IMM })
281
282 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
283
284 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
285 ((struct bpf_insn) { \
286 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
287 .dst_reg = DST, \
288 .src_reg = SRC, \
289 .off = OFF, \
290 .imm = 0 })
291
292 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
293
294 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
295 ((struct bpf_insn) { \
296 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
297 .dst_reg = DST, \
298 .src_reg = 0, \
299 .off = OFF, \
300 .imm = IMM })
301
302 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
303
304 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
305 ((struct bpf_insn) { \
306 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
307 .dst_reg = DST, \
308 .src_reg = SRC, \
309 .off = OFF, \
310 .imm = 0 })
311
312 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
313
314 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
315 ((struct bpf_insn) { \
316 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
317 .dst_reg = DST, \
318 .src_reg = 0, \
319 .off = OFF, \
320 .imm = IMM })
321
322 /* Unconditional jumps, goto pc + off16 */
323
324 #define BPF_JMP_A(OFF) \
325 ((struct bpf_insn) { \
326 .code = BPF_JMP | BPF_JA, \
327 .dst_reg = 0, \
328 .src_reg = 0, \
329 .off = OFF, \
330 .imm = 0 })
331
332 /* Relative call */
333
334 #define BPF_CALL_REL(TGT) \
335 ((struct bpf_insn) { \
336 .code = BPF_JMP | BPF_CALL, \
337 .dst_reg = 0, \
338 .src_reg = BPF_PSEUDO_CALL, \
339 .off = 0, \
340 .imm = TGT })
341
342 /* Function call */
343
344 #define BPF_CAST_CALL(x) \
345 ((u64 (*)(u64, u64, u64, u64, u64))(x))
346
347 #define BPF_EMIT_CALL(FUNC) \
348 ((struct bpf_insn) { \
349 .code = BPF_JMP | BPF_CALL, \
350 .dst_reg = 0, \
351 .src_reg = 0, \
352 .off = 0, \
353 .imm = ((FUNC) - __bpf_call_base) })
354
355 /* Raw code statement block */
356
357 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
358 ((struct bpf_insn) { \
359 .code = CODE, \
360 .dst_reg = DST, \
361 .src_reg = SRC, \
362 .off = OFF, \
363 .imm = IMM })
364
365 /* Program exit */
366
367 #define BPF_EXIT_INSN() \
368 ((struct bpf_insn) { \
369 .code = BPF_JMP | BPF_EXIT, \
370 .dst_reg = 0, \
371 .src_reg = 0, \
372 .off = 0, \
373 .imm = 0 })
374
375 /* Internal classic blocks for direct assignment */
376
377 #define __BPF_STMT(CODE, K) \
378 ((struct sock_filter) BPF_STMT(CODE, K))
379
380 #define __BPF_JUMP(CODE, K, JT, JF) \
381 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
382
383 #define bytes_to_bpf_size(bytes) \
384 ({ \
385 int bpf_size = -EINVAL; \
386 \
387 if (bytes == sizeof(u8)) \
388 bpf_size = BPF_B; \
389 else if (bytes == sizeof(u16)) \
390 bpf_size = BPF_H; \
391 else if (bytes == sizeof(u32)) \
392 bpf_size = BPF_W; \
393 else if (bytes == sizeof(u64)) \
394 bpf_size = BPF_DW; \
395 \
396 bpf_size; \
397 })
398
399 #define bpf_size_to_bytes(bpf_size) \
400 ({ \
401 int bytes = -EINVAL; \
402 \
403 if (bpf_size == BPF_B) \
404 bytes = sizeof(u8); \
405 else if (bpf_size == BPF_H) \
406 bytes = sizeof(u16); \
407 else if (bpf_size == BPF_W) \
408 bytes = sizeof(u32); \
409 else if (bpf_size == BPF_DW) \
410 bytes = sizeof(u64); \
411 \
412 bytes; \
413 })
414
415 #define BPF_SIZEOF(type) \
416 ({ \
417 const int __size = bytes_to_bpf_size(sizeof(type)); \
418 BUILD_BUG_ON(__size < 0); \
419 __size; \
420 })
421
422 #define BPF_FIELD_SIZEOF(type, field) \
423 ({ \
424 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
425 BUILD_BUG_ON(__size < 0); \
426 __size; \
427 })
428
429 #define BPF_LDST_BYTES(insn) \
430 ({ \
431 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
432 WARN_ON(__size < 0); \
433 __size; \
434 })
435
436 #define __BPF_MAP_0(m, v, ...) v
437 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
438 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
439 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
440 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
441 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
442
443 #define __BPF_REG_0(...) __BPF_PAD(5)
444 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
445 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
446 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
447 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
448 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
449
450 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
451 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
452
453 #define __BPF_CAST(t, a) \
454 (__force t) \
455 (__force \
456 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
457 (unsigned long)0, (t)0))) a
458 #define __BPF_V void
459 #define __BPF_N
460
461 #define __BPF_DECL_ARGS(t, a) t a
462 #define __BPF_DECL_REGS(t, a) u64 a
463
464 #define __BPF_PAD(n) \
465 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
466 u64, __ur_3, u64, __ur_4, u64, __ur_5)
467
468 #define BPF_CALL_x(x, name, ...) \
469 static __always_inline \
470 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
471 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
472 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
473 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
474 { \
475 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
476 } \
477 static __always_inline \
478 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
479
480 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
481 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
482 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
483 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
484 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
485 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
486
487 #define bpf_ctx_range(TYPE, MEMBER) \
488 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
489 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
490 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
491 #if BITS_PER_LONG == 64
492 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
493 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
494 #else
495 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
496 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
497 #endif /* BITS_PER_LONG == 64 */
498
499 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
500 ({ \
501 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \
502 *(PTR_SIZE) = (SIZE); \
503 offsetof(TYPE, MEMBER); \
504 })
505
506 /* A struct sock_filter is architecture independent. */
507 struct compat_sock_fprog {
508 u16 len;
509 compat_uptr_t filter; /* struct sock_filter * */
510 };
511
512 struct sock_fprog_kern {
513 u16 len;
514 struct sock_filter *filter;
515 };
516
517 /* Some arches need doubleword alignment for their instructions and/or data */
518 #define BPF_IMAGE_ALIGNMENT 8
519
520 struct bpf_binary_header {
521 u32 pages;
522 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT);
523 };
524
525 struct bpf_prog {
526 u16 pages; /* Number of allocated pages */
527 u16 jited:1, /* Is our filter JIT'ed? */
528 jit_requested:1,/* archs need to JIT the prog */
529 gpl_compatible:1, /* Is filter GPL compatible? */
530 cb_access:1, /* Is control block accessed? */
531 dst_needed:1, /* Do we need dst entry? */
532 blinded:1, /* Was blinded */
533 is_func:1, /* program is a bpf function */
534 kprobe_override:1, /* Do we override a kprobe? */
535 has_callchain_buf:1, /* callchain buffer allocated? */
536 enforce_expected_attach_type:1, /* Enforce expected_attach_type checking at attach time */
537 call_get_stack:1; /* Do we call bpf_get_stack() or bpf_get_stackid() */
538 enum bpf_prog_type type; /* Type of BPF program */
539 enum bpf_attach_type expected_attach_type; /* For some prog types */
540 u32 len; /* Number of filter blocks */
541 u32 jited_len; /* Size of jited insns in bytes */
542 u8 tag[BPF_TAG_SIZE];
543 struct bpf_prog_aux *aux; /* Auxiliary fields */
544 struct sock_fprog_kern *orig_prog; /* Original BPF program */
545 unsigned int (*bpf_func)(const void *ctx,
546 const struct bpf_insn *insn);
547 /* Instructions for interpreter */
548 struct sock_filter insns[0];
549 struct bpf_insn insnsi[];
550 };
551
552 struct sk_filter {
553 refcount_t refcnt;
554 struct rcu_head rcu;
555 struct bpf_prog *prog;
556 };
557
558 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
559
560 #define __BPF_PROG_RUN(prog, ctx, dfunc) ({ \
561 u32 __ret; \
562 cant_migrate(); \
563 if (static_branch_unlikely(&bpf_stats_enabled_key)) { \
564 struct bpf_prog_stats *__stats; \
565 u64 __start = sched_clock(); \
566 __ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func); \
567 __stats = this_cpu_ptr(prog->aux->stats); \
568 u64_stats_update_begin(&__stats->syncp); \
569 __stats->cnt++; \
570 __stats->nsecs += sched_clock() - __start; \
571 u64_stats_update_end(&__stats->syncp); \
572 } else { \
573 __ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func); \
574 } \
575 __ret; })
576
577 #define BPF_PROG_RUN(prog, ctx) \
578 __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nop_func)
579
580 /*
581 * Use in preemptible and therefore migratable context to make sure that
582 * the execution of the BPF program runs on one CPU.
583 *
584 * This uses migrate_disable/enable() explicitly to document that the
585 * invocation of a BPF program does not require reentrancy protection
586 * against a BPF program which is invoked from a preempting task.
587 *
588 * For non RT enabled kernels migrate_disable/enable() maps to
589 * preempt_disable/enable(), i.e. it disables also preemption.
590 */
bpf_prog_run_pin_on_cpu(const struct bpf_prog * prog,const void * ctx)591 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
592 const void *ctx)
593 {
594 u32 ret;
595
596 migrate_disable();
597 ret = __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nop_func);
598 migrate_enable();
599 return ret;
600 }
601
602 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
603
604 struct bpf_skb_data_end {
605 struct qdisc_skb_cb qdisc_cb;
606 void *data_meta;
607 void *data_end;
608 };
609
610 struct bpf_nh_params {
611 u32 nh_family;
612 union {
613 u32 ipv4_nh;
614 struct in6_addr ipv6_nh;
615 };
616 };
617
618 struct bpf_redirect_info {
619 u32 flags;
620 u32 tgt_index;
621 void *tgt_value;
622 struct bpf_map *map;
623 u32 kern_flags;
624 struct bpf_nh_params nh;
625 };
626
627 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
628
629 /* flags for bpf_redirect_info kern_flags */
630 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
631
632 /* Compute the linear packet data range [data, data_end) which
633 * will be accessed by various program types (cls_bpf, act_bpf,
634 * lwt, ...). Subsystems allowing direct data access must (!)
635 * ensure that cb[] area can be written to when BPF program is
636 * invoked (otherwise cb[] save/restore is necessary).
637 */
bpf_compute_data_pointers(struct sk_buff * skb)638 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
639 {
640 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
641
642 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
643 cb->data_meta = skb->data - skb_metadata_len(skb);
644 cb->data_end = skb->data + skb_headlen(skb);
645 }
646
647 /* Similar to bpf_compute_data_pointers(), except that save orginal
648 * data in cb->data and cb->meta_data for restore.
649 */
bpf_compute_and_save_data_end(struct sk_buff * skb,void ** saved_data_end)650 static inline void bpf_compute_and_save_data_end(
651 struct sk_buff *skb, void **saved_data_end)
652 {
653 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
654
655 *saved_data_end = cb->data_end;
656 cb->data_end = skb->data + skb_headlen(skb);
657 }
658
659 /* Restore data saved by bpf_compute_data_pointers(). */
bpf_restore_data_end(struct sk_buff * skb,void * saved_data_end)660 static inline void bpf_restore_data_end(
661 struct sk_buff *skb, void *saved_data_end)
662 {
663 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
664
665 cb->data_end = saved_data_end;
666 }
667
bpf_skb_cb(struct sk_buff * skb)668 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
669 {
670 /* eBPF programs may read/write skb->cb[] area to transfer meta
671 * data between tail calls. Since this also needs to work with
672 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
673 *
674 * In some socket filter cases, the cb unfortunately needs to be
675 * saved/restored so that protocol specific skb->cb[] data won't
676 * be lost. In any case, due to unpriviledged eBPF programs
677 * attached to sockets, we need to clear the bpf_skb_cb() area
678 * to not leak previous contents to user space.
679 */
680 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
681 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) !=
682 sizeof_field(struct qdisc_skb_cb, data));
683
684 return qdisc_skb_cb(skb)->data;
685 }
686
687 /* Must be invoked with migration disabled */
__bpf_prog_run_save_cb(const struct bpf_prog * prog,struct sk_buff * skb)688 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
689 struct sk_buff *skb)
690 {
691 u8 *cb_data = bpf_skb_cb(skb);
692 u8 cb_saved[BPF_SKB_CB_LEN];
693 u32 res;
694
695 if (unlikely(prog->cb_access)) {
696 memcpy(cb_saved, cb_data, sizeof(cb_saved));
697 memset(cb_data, 0, sizeof(cb_saved));
698 }
699
700 res = BPF_PROG_RUN(prog, skb);
701
702 if (unlikely(prog->cb_access))
703 memcpy(cb_data, cb_saved, sizeof(cb_saved));
704
705 return res;
706 }
707
bpf_prog_run_save_cb(const struct bpf_prog * prog,struct sk_buff * skb)708 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
709 struct sk_buff *skb)
710 {
711 u32 res;
712
713 migrate_disable();
714 res = __bpf_prog_run_save_cb(prog, skb);
715 migrate_enable();
716 return res;
717 }
718
bpf_prog_run_clear_cb(const struct bpf_prog * prog,struct sk_buff * skb)719 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
720 struct sk_buff *skb)
721 {
722 u8 *cb_data = bpf_skb_cb(skb);
723 u32 res;
724
725 if (unlikely(prog->cb_access))
726 memset(cb_data, 0, BPF_SKB_CB_LEN);
727
728 res = bpf_prog_run_pin_on_cpu(prog, skb);
729 return res;
730 }
731
DECLARE_BPF_DISPATCHER(xdp)732 DECLARE_BPF_DISPATCHER(xdp)
733
734 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
735 struct xdp_buff *xdp)
736 {
737 /* Caller needs to hold rcu_read_lock() (!), otherwise program
738 * can be released while still running, or map elements could be
739 * freed early while still having concurrent users. XDP fastpath
740 * already takes rcu_read_lock() when fetching the program, so
741 * it's not necessary here anymore.
742 */
743 return __BPF_PROG_RUN(prog, xdp, BPF_DISPATCHER_FUNC(xdp));
744 }
745
746 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog);
747
bpf_prog_insn_size(const struct bpf_prog * prog)748 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
749 {
750 return prog->len * sizeof(struct bpf_insn);
751 }
752
bpf_prog_tag_scratch_size(const struct bpf_prog * prog)753 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
754 {
755 return round_up(bpf_prog_insn_size(prog) +
756 sizeof(__be64) + 1, SHA1_BLOCK_SIZE);
757 }
758
bpf_prog_size(unsigned int proglen)759 static inline unsigned int bpf_prog_size(unsigned int proglen)
760 {
761 return max(sizeof(struct bpf_prog),
762 offsetof(struct bpf_prog, insns[proglen]));
763 }
764
bpf_prog_was_classic(const struct bpf_prog * prog)765 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
766 {
767 /* When classic BPF programs have been loaded and the arch
768 * does not have a classic BPF JIT (anymore), they have been
769 * converted via bpf_migrate_filter() to eBPF and thus always
770 * have an unspec program type.
771 */
772 return prog->type == BPF_PROG_TYPE_UNSPEC;
773 }
774
bpf_ctx_off_adjust_machine(u32 size)775 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
776 {
777 const u32 size_machine = sizeof(unsigned long);
778
779 if (size > size_machine && size % size_machine == 0)
780 size = size_machine;
781
782 return size;
783 }
784
785 static inline bool
bpf_ctx_narrow_access_ok(u32 off,u32 size,u32 size_default)786 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
787 {
788 return size <= size_default && (size & (size - 1)) == 0;
789 }
790
791 static inline u8
bpf_ctx_narrow_access_offset(u32 off,u32 size,u32 size_default)792 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
793 {
794 u8 access_off = off & (size_default - 1);
795
796 #ifdef __LITTLE_ENDIAN
797 return access_off;
798 #else
799 return size_default - (access_off + size);
800 #endif
801 }
802
803 #define bpf_ctx_wide_access_ok(off, size, type, field) \
804 (size == sizeof(__u64) && \
805 off >= offsetof(type, field) && \
806 off + sizeof(__u64) <= offsetofend(type, field) && \
807 off % sizeof(__u64) == 0)
808
809 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
810
bpf_prog_lock_ro(struct bpf_prog * fp)811 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
812 {
813 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
814 if (!fp->jited) {
815 set_vm_flush_reset_perms(fp);
816 set_memory_ro((unsigned long)fp, fp->pages);
817 }
818 #endif
819 }
820
bpf_jit_binary_lock_ro(struct bpf_binary_header * hdr)821 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
822 {
823 set_vm_flush_reset_perms(hdr);
824 set_memory_ro((unsigned long)hdr, hdr->pages);
825 set_memory_x((unsigned long)hdr, hdr->pages);
826 }
827
828 static inline struct bpf_binary_header *
bpf_jit_binary_hdr(const struct bpf_prog * fp)829 bpf_jit_binary_hdr(const struct bpf_prog *fp)
830 {
831 unsigned long real_start = (unsigned long)fp->bpf_func;
832 unsigned long addr = real_start & PAGE_MASK;
833
834 return (void *)addr;
835 }
836
837 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
sk_filter(struct sock * sk,struct sk_buff * skb)838 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
839 {
840 return sk_filter_trim_cap(sk, skb, 1);
841 }
842
843 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
844 void bpf_prog_free(struct bpf_prog *fp);
845
846 bool bpf_opcode_in_insntable(u8 code);
847
848 void bpf_prog_free_linfo(struct bpf_prog *prog);
849 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
850 const u32 *insn_to_jit_off);
851 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
852 void bpf_prog_free_jited_linfo(struct bpf_prog *prog);
853 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog);
854
855 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
856 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
857 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
858 gfp_t gfp_extra_flags);
859 void __bpf_prog_free(struct bpf_prog *fp);
860
bpf_prog_unlock_free(struct bpf_prog * fp)861 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
862 {
863 __bpf_prog_free(fp);
864 }
865
866 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
867 unsigned int flen);
868
869 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
870 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
871 bpf_aux_classic_check_t trans, bool save_orig);
872 void bpf_prog_destroy(struct bpf_prog *fp);
873
874 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
875 int sk_attach_bpf(u32 ufd, struct sock *sk);
876 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
877 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
878 void sk_reuseport_prog_free(struct bpf_prog *prog);
879 int sk_detach_filter(struct sock *sk);
880 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
881 unsigned int len);
882
883 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
884 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
885
886 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
887 #define __bpf_call_base_args \
888 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
889 __bpf_call_base)
890
891 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
892 void bpf_jit_compile(struct bpf_prog *prog);
893 bool bpf_jit_needs_zext(void);
894 bool bpf_helper_changes_pkt_data(void *func);
895
bpf_dump_raw_ok(const struct cred * cred)896 static inline bool bpf_dump_raw_ok(const struct cred *cred)
897 {
898 /* Reconstruction of call-sites is dependent on kallsyms,
899 * thus make dump the same restriction.
900 */
901 return kallsyms_show_value(cred);
902 }
903
904 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
905 const struct bpf_insn *patch, u32 len);
906 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
907
908 void bpf_clear_redirect_map(struct bpf_map *map);
909
xdp_return_frame_no_direct(void)910 static inline bool xdp_return_frame_no_direct(void)
911 {
912 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
913
914 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
915 }
916
xdp_set_return_frame_no_direct(void)917 static inline void xdp_set_return_frame_no_direct(void)
918 {
919 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
920
921 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
922 }
923
xdp_clear_return_frame_no_direct(void)924 static inline void xdp_clear_return_frame_no_direct(void)
925 {
926 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
927
928 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
929 }
930
xdp_ok_fwd_dev(const struct net_device * fwd,unsigned int pktlen)931 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
932 unsigned int pktlen)
933 {
934 unsigned int len;
935
936 if (unlikely(!(fwd->flags & IFF_UP)))
937 return -ENETDOWN;
938
939 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
940 if (pktlen > len)
941 return -EMSGSIZE;
942
943 return 0;
944 }
945
946 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the
947 * same cpu context. Further for best results no more than a single map
948 * for the do_redirect/do_flush pair should be used. This limitation is
949 * because we only track one map and force a flush when the map changes.
950 * This does not appear to be a real limitation for existing software.
951 */
952 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
953 struct xdp_buff *xdp, struct bpf_prog *prog);
954 int xdp_do_redirect(struct net_device *dev,
955 struct xdp_buff *xdp,
956 struct bpf_prog *prog);
957 void xdp_do_flush(void);
958
959 /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as
960 * it is no longer only flushing maps. Keep this define for compatibility
961 * until all drivers are updated - do not use xdp_do_flush_map() in new code!
962 */
963 #define xdp_do_flush_map xdp_do_flush
964
965 void bpf_warn_invalid_xdp_action(u32 act);
966
967 #ifdef CONFIG_INET
968 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
969 struct bpf_prog *prog, struct sk_buff *skb,
970 u32 hash);
971 #else
972 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)973 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
974 struct bpf_prog *prog, struct sk_buff *skb,
975 u32 hash)
976 {
977 return NULL;
978 }
979 #endif
980
981 #ifdef CONFIG_BPF_JIT
982 extern int bpf_jit_enable;
983 extern int bpf_jit_harden;
984 extern int bpf_jit_kallsyms;
985 extern long bpf_jit_limit;
986
987 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
988
989 struct bpf_binary_header *
990 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
991 unsigned int alignment,
992 bpf_jit_fill_hole_t bpf_fill_ill_insns);
993 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
994 u64 bpf_jit_alloc_exec_limit(void);
995 void *bpf_jit_alloc_exec(unsigned long size);
996 void bpf_jit_free_exec(void *addr);
997 void bpf_jit_free(struct bpf_prog *fp);
998
999 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1000 struct bpf_jit_poke_descriptor *poke);
1001
1002 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1003 const struct bpf_insn *insn, bool extra_pass,
1004 u64 *func_addr, bool *func_addr_fixed);
1005
1006 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
1007 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
1008
bpf_jit_dump(unsigned int flen,unsigned int proglen,u32 pass,void * image)1009 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
1010 u32 pass, void *image)
1011 {
1012 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
1013 proglen, pass, image, current->comm, task_pid_nr(current));
1014
1015 if (image)
1016 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
1017 16, 1, image, proglen, false);
1018 }
1019
bpf_jit_is_ebpf(void)1020 static inline bool bpf_jit_is_ebpf(void)
1021 {
1022 # ifdef CONFIG_HAVE_EBPF_JIT
1023 return true;
1024 # else
1025 return false;
1026 # endif
1027 }
1028
ebpf_jit_enabled(void)1029 static inline bool ebpf_jit_enabled(void)
1030 {
1031 return bpf_jit_enable && bpf_jit_is_ebpf();
1032 }
1033
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1034 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1035 {
1036 return fp->jited && bpf_jit_is_ebpf();
1037 }
1038
bpf_jit_blinding_enabled(struct bpf_prog * prog)1039 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1040 {
1041 /* These are the prerequisites, should someone ever have the
1042 * idea to call blinding outside of them, we make sure to
1043 * bail out.
1044 */
1045 if (!bpf_jit_is_ebpf())
1046 return false;
1047 if (!prog->jit_requested)
1048 return false;
1049 if (!bpf_jit_harden)
1050 return false;
1051 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
1052 return false;
1053
1054 return true;
1055 }
1056
bpf_jit_kallsyms_enabled(void)1057 static inline bool bpf_jit_kallsyms_enabled(void)
1058 {
1059 /* There are a couple of corner cases where kallsyms should
1060 * not be enabled f.e. on hardening.
1061 */
1062 if (bpf_jit_harden)
1063 return false;
1064 if (!bpf_jit_kallsyms)
1065 return false;
1066 if (bpf_jit_kallsyms == 1)
1067 return true;
1068
1069 return false;
1070 }
1071
1072 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
1073 unsigned long *off, char *sym);
1074 bool is_bpf_text_address(unsigned long addr);
1075 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1076 char *sym);
1077
1078 static inline const char *
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1079 bpf_address_lookup(unsigned long addr, unsigned long *size,
1080 unsigned long *off, char **modname, char *sym)
1081 {
1082 const char *ret = __bpf_address_lookup(addr, size, off, sym);
1083
1084 if (ret && modname)
1085 *modname = NULL;
1086 return ret;
1087 }
1088
1089 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1090 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1091
1092 #else /* CONFIG_BPF_JIT */
1093
ebpf_jit_enabled(void)1094 static inline bool ebpf_jit_enabled(void)
1095 {
1096 return false;
1097 }
1098
bpf_jit_blinding_enabled(struct bpf_prog * prog)1099 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1100 {
1101 return false;
1102 }
1103
bpf_prog_ebpf_jited(const struct bpf_prog * fp)1104 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1105 {
1106 return false;
1107 }
1108
1109 static inline int
bpf_jit_add_poke_descriptor(struct bpf_prog * prog,struct bpf_jit_poke_descriptor * poke)1110 bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1111 struct bpf_jit_poke_descriptor *poke)
1112 {
1113 return -ENOTSUPP;
1114 }
1115
bpf_jit_free(struct bpf_prog * fp)1116 static inline void bpf_jit_free(struct bpf_prog *fp)
1117 {
1118 bpf_prog_unlock_free(fp);
1119 }
1120
bpf_jit_kallsyms_enabled(void)1121 static inline bool bpf_jit_kallsyms_enabled(void)
1122 {
1123 return false;
1124 }
1125
1126 static inline const char *
__bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char * sym)1127 __bpf_address_lookup(unsigned long addr, unsigned long *size,
1128 unsigned long *off, char *sym)
1129 {
1130 return NULL;
1131 }
1132
is_bpf_text_address(unsigned long addr)1133 static inline bool is_bpf_text_address(unsigned long addr)
1134 {
1135 return false;
1136 }
1137
bpf_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * sym)1138 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1139 char *type, char *sym)
1140 {
1141 return -ERANGE;
1142 }
1143
1144 static inline const char *
bpf_address_lookup(unsigned long addr,unsigned long * size,unsigned long * off,char ** modname,char * sym)1145 bpf_address_lookup(unsigned long addr, unsigned long *size,
1146 unsigned long *off, char **modname, char *sym)
1147 {
1148 return NULL;
1149 }
1150
bpf_prog_kallsyms_add(struct bpf_prog * fp)1151 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1152 {
1153 }
1154
bpf_prog_kallsyms_del(struct bpf_prog * fp)1155 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1156 {
1157 }
1158
1159 #endif /* CONFIG_BPF_JIT */
1160
1161 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1162
1163 #define BPF_ANC BIT(15)
1164
bpf_needs_clear_a(const struct sock_filter * first)1165 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1166 {
1167 switch (first->code) {
1168 case BPF_RET | BPF_K:
1169 case BPF_LD | BPF_W | BPF_LEN:
1170 return false;
1171
1172 case BPF_LD | BPF_W | BPF_ABS:
1173 case BPF_LD | BPF_H | BPF_ABS:
1174 case BPF_LD | BPF_B | BPF_ABS:
1175 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1176 return true;
1177 return false;
1178
1179 default:
1180 return true;
1181 }
1182 }
1183
bpf_anc_helper(const struct sock_filter * ftest)1184 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1185 {
1186 BUG_ON(ftest->code & BPF_ANC);
1187
1188 switch (ftest->code) {
1189 case BPF_LD | BPF_W | BPF_ABS:
1190 case BPF_LD | BPF_H | BPF_ABS:
1191 case BPF_LD | BPF_B | BPF_ABS:
1192 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1193 return BPF_ANC | SKF_AD_##CODE
1194 switch (ftest->k) {
1195 BPF_ANCILLARY(PROTOCOL);
1196 BPF_ANCILLARY(PKTTYPE);
1197 BPF_ANCILLARY(IFINDEX);
1198 BPF_ANCILLARY(NLATTR);
1199 BPF_ANCILLARY(NLATTR_NEST);
1200 BPF_ANCILLARY(MARK);
1201 BPF_ANCILLARY(QUEUE);
1202 BPF_ANCILLARY(HATYPE);
1203 BPF_ANCILLARY(RXHASH);
1204 BPF_ANCILLARY(CPU);
1205 BPF_ANCILLARY(ALU_XOR_X);
1206 BPF_ANCILLARY(VLAN_TAG);
1207 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1208 BPF_ANCILLARY(PAY_OFFSET);
1209 BPF_ANCILLARY(RANDOM);
1210 BPF_ANCILLARY(VLAN_TPID);
1211 }
1212 fallthrough;
1213 default:
1214 return ftest->code;
1215 }
1216 }
1217
1218 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1219 int k, unsigned int size);
1220
bpf_load_pointer(const struct sk_buff * skb,int k,unsigned int size,void * buffer)1221 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
1222 unsigned int size, void *buffer)
1223 {
1224 if (k >= 0)
1225 return skb_header_pointer(skb, k, size, buffer);
1226
1227 return bpf_internal_load_pointer_neg_helper(skb, k, size);
1228 }
1229
bpf_tell_extensions(void)1230 static inline int bpf_tell_extensions(void)
1231 {
1232 return SKF_AD_MAX;
1233 }
1234
1235 struct bpf_sock_addr_kern {
1236 struct sock *sk;
1237 struct sockaddr *uaddr;
1238 /* Temporary "register" to make indirect stores to nested structures
1239 * defined above. We need three registers to make such a store, but
1240 * only two (src and dst) are available at convert_ctx_access time
1241 */
1242 u64 tmp_reg;
1243 void *t_ctx; /* Attach type specific context. */
1244 };
1245
1246 struct bpf_sock_ops_kern {
1247 struct sock *sk;
1248 union {
1249 u32 args[4];
1250 u32 reply;
1251 u32 replylong[4];
1252 };
1253 struct sk_buff *syn_skb;
1254 struct sk_buff *skb;
1255 void *skb_data_end;
1256 u8 op;
1257 u8 is_fullsock;
1258 u8 remaining_opt_len;
1259 u64 temp; /* temp and everything after is not
1260 * initialized to 0 before calling
1261 * the BPF program. New fields that
1262 * should be initialized to 0 should
1263 * be inserted before temp.
1264 * temp is scratch storage used by
1265 * sock_ops_convert_ctx_access
1266 * as temporary storage of a register.
1267 */
1268 };
1269
1270 struct bpf_sysctl_kern {
1271 struct ctl_table_header *head;
1272 struct ctl_table *table;
1273 void *cur_val;
1274 size_t cur_len;
1275 void *new_val;
1276 size_t new_len;
1277 int new_updated;
1278 int write;
1279 loff_t *ppos;
1280 /* Temporary "register" for indirect stores to ppos. */
1281 u64 tmp_reg;
1282 };
1283
1284 struct bpf_sockopt_kern {
1285 struct sock *sk;
1286 u8 *optval;
1287 u8 *optval_end;
1288 s32 level;
1289 s32 optname;
1290 s32 optlen;
1291 s32 retval;
1292 };
1293
1294 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len);
1295
1296 struct bpf_sk_lookup_kern {
1297 u16 family;
1298 u16 protocol;
1299 __be16 sport;
1300 u16 dport;
1301 struct {
1302 __be32 saddr;
1303 __be32 daddr;
1304 } v4;
1305 struct {
1306 const struct in6_addr *saddr;
1307 const struct in6_addr *daddr;
1308 } v6;
1309 struct sock *selected_sk;
1310 bool no_reuseport;
1311 };
1312
1313 extern struct static_key_false bpf_sk_lookup_enabled;
1314
1315 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup.
1316 *
1317 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and
1318 * SK_DROP. Their meaning is as follows:
1319 *
1320 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result
1321 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup
1322 * SK_DROP : terminate lookup with -ECONNREFUSED
1323 *
1324 * This macro aggregates return values and selected sockets from
1325 * multiple BPF programs according to following rules in order:
1326 *
1327 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk,
1328 * macro result is SK_PASS and last ctx.selected_sk is used.
1329 * 2. If any program returned SK_DROP return value,
1330 * macro result is SK_DROP.
1331 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL.
1332 *
1333 * Caller must ensure that the prog array is non-NULL, and that the
1334 * array as well as the programs it contains remain valid.
1335 */
1336 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \
1337 ({ \
1338 struct bpf_sk_lookup_kern *_ctx = &(ctx); \
1339 struct bpf_prog_array_item *_item; \
1340 struct sock *_selected_sk = NULL; \
1341 bool _no_reuseport = false; \
1342 struct bpf_prog *_prog; \
1343 bool _all_pass = true; \
1344 u32 _ret; \
1345 \
1346 migrate_disable(); \
1347 _item = &(array)->items[0]; \
1348 while ((_prog = READ_ONCE(_item->prog))) { \
1349 /* restore most recent selection */ \
1350 _ctx->selected_sk = _selected_sk; \
1351 _ctx->no_reuseport = _no_reuseport; \
1352 \
1353 _ret = func(_prog, _ctx); \
1354 if (_ret == SK_PASS && _ctx->selected_sk) { \
1355 /* remember last non-NULL socket */ \
1356 _selected_sk = _ctx->selected_sk; \
1357 _no_reuseport = _ctx->no_reuseport; \
1358 } else if (_ret == SK_DROP && _all_pass) { \
1359 _all_pass = false; \
1360 } \
1361 _item++; \
1362 } \
1363 _ctx->selected_sk = _selected_sk; \
1364 _ctx->no_reuseport = _no_reuseport; \
1365 migrate_enable(); \
1366 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \
1367 })
1368
bpf_sk_lookup_run_v4(struct net * net,int protocol,const __be32 saddr,const __be16 sport,const __be32 daddr,const u16 dport,struct sock ** psk)1369 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol,
1370 const __be32 saddr, const __be16 sport,
1371 const __be32 daddr, const u16 dport,
1372 struct sock **psk)
1373 {
1374 struct bpf_prog_array *run_array;
1375 struct sock *selected_sk = NULL;
1376 bool no_reuseport = false;
1377
1378 rcu_read_lock();
1379 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1380 if (run_array) {
1381 struct bpf_sk_lookup_kern ctx = {
1382 .family = AF_INET,
1383 .protocol = protocol,
1384 .v4.saddr = saddr,
1385 .v4.daddr = daddr,
1386 .sport = sport,
1387 .dport = dport,
1388 };
1389 u32 act;
1390
1391 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, BPF_PROG_RUN);
1392 if (act == SK_PASS) {
1393 selected_sk = ctx.selected_sk;
1394 no_reuseport = ctx.no_reuseport;
1395 } else {
1396 selected_sk = ERR_PTR(-ECONNREFUSED);
1397 }
1398 }
1399 rcu_read_unlock();
1400 *psk = selected_sk;
1401 return no_reuseport;
1402 }
1403
1404 #if IS_ENABLED(CONFIG_IPV6)
bpf_sk_lookup_run_v6(struct net * net,int protocol,const struct in6_addr * saddr,const __be16 sport,const struct in6_addr * daddr,const u16 dport,struct sock ** psk)1405 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol,
1406 const struct in6_addr *saddr,
1407 const __be16 sport,
1408 const struct in6_addr *daddr,
1409 const u16 dport,
1410 struct sock **psk)
1411 {
1412 struct bpf_prog_array *run_array;
1413 struct sock *selected_sk = NULL;
1414 bool no_reuseport = false;
1415
1416 rcu_read_lock();
1417 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1418 if (run_array) {
1419 struct bpf_sk_lookup_kern ctx = {
1420 .family = AF_INET6,
1421 .protocol = protocol,
1422 .v6.saddr = saddr,
1423 .v6.daddr = daddr,
1424 .sport = sport,
1425 .dport = dport,
1426 };
1427 u32 act;
1428
1429 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, BPF_PROG_RUN);
1430 if (act == SK_PASS) {
1431 selected_sk = ctx.selected_sk;
1432 no_reuseport = ctx.no_reuseport;
1433 } else {
1434 selected_sk = ERR_PTR(-ECONNREFUSED);
1435 }
1436 }
1437 rcu_read_unlock();
1438 *psk = selected_sk;
1439 return no_reuseport;
1440 }
1441 #endif /* IS_ENABLED(CONFIG_IPV6) */
1442
1443 #endif /* __LINUX_FILTER_H__ */
1444
