]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - include/linux/filter.h
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:


4e555d0ba6b632b9adb0b908e61970ce86b4e091
[mirror_ubuntu-bionic-kernel.git] / include / linux / filter.h
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
22 #include <net/sch_generic.h>
23
24 #include <uapi/linux/filter.h>
25 #include <uapi/linux/bpf.h>
26
27 struct sk_buff;
28 struct sock;
29 struct seccomp_data;
30 struct bpf_prog_aux;
31
32 /* ArgX, context and stack frame pointer register positions. Note,
33 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
34 * calls in BPF_CALL instruction.
35 */
36 #define BPF_REG_ARG1 BPF_REG_1
37 #define BPF_REG_ARG2 BPF_REG_2
38 #define BPF_REG_ARG3 BPF_REG_3
39 #define BPF_REG_ARG4 BPF_REG_4
40 #define BPF_REG_ARG5 BPF_REG_5
41 #define BPF_REG_CTX BPF_REG_6
42 #define BPF_REG_FP BPF_REG_10
43
44 /* Additional register mappings for converted user programs. */
45 #define BPF_REG_A BPF_REG_0
46 #define BPF_REG_X BPF_REG_7
47 #define BPF_REG_TMP BPF_REG_8
48
49 /* Kernel hidden auxiliary/helper register. */
50 #define BPF_REG_AX MAX_BPF_REG
51 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1)
52 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
53
54 /* unused opcode to mark special call to bpf_tail_call() helper */
55 #define BPF_TAIL_CALL 0xf0
56
57 /* As per nm, we expose JITed images as text (code) section for
58 * kallsyms. That way, tools like perf can find it to match
59 * addresses.
60 */
61 #define BPF_SYM_ELF_TYPE 't'
62
63 /* BPF program can access up to 512 bytes of stack space. */
64 #define MAX_BPF_STACK 512
65
66 /* Helper macros for filter block array initializers. */
67
68 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
69
70 #define BPF_ALU64_REG(OP, DST, SRC) \
71 ((struct bpf_insn) { \
72 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
73 .dst_reg = DST, \
74 .src_reg = SRC, \
75 .off = 0, \
76 .imm = 0 })
77
78 #define BPF_ALU32_REG(OP, DST, SRC) \
79 ((struct bpf_insn) { \
80 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
81 .dst_reg = DST, \
82 .src_reg = SRC, \
83 .off = 0, \
84 .imm = 0 })
85
86 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
87
88 #define BPF_ALU64_IMM(OP, DST, IMM) \
89 ((struct bpf_insn) { \
90 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
91 .dst_reg = DST, \
92 .src_reg = 0, \
93 .off = 0, \
94 .imm = IMM })
95
96 #define BPF_ALU32_IMM(OP, DST, IMM) \
97 ((struct bpf_insn) { \
98 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
99 .dst_reg = DST, \
100 .src_reg = 0, \
101 .off = 0, \
102 .imm = IMM })
103
104 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
105
106 #define BPF_ENDIAN(TYPE, DST, LEN) \
107 ((struct bpf_insn) { \
108 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
109 .dst_reg = DST, \
110 .src_reg = 0, \
111 .off = 0, \
112 .imm = LEN })
113
114 /* Short form of mov, dst_reg = src_reg */
115
116 #define BPF_MOV64_REG(DST, SRC) \
117 ((struct bpf_insn) { \
118 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
119 .dst_reg = DST, \
120 .src_reg = SRC, \
121 .off = 0, \
122 .imm = 0 })
123
124 #define BPF_MOV32_REG(DST, SRC) \
125 ((struct bpf_insn) { \
126 .code = BPF_ALU | BPF_MOV | BPF_X, \
127 .dst_reg = DST, \
128 .src_reg = SRC, \
129 .off = 0, \
130 .imm = 0 })
131
132 /* Short form of mov, dst_reg = imm32 */
133
134 #define BPF_MOV64_IMM(DST, IMM) \
135 ((struct bpf_insn) { \
136 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
137 .dst_reg = DST, \
138 .src_reg = 0, \
139 .off = 0, \
140 .imm = IMM })
141
142 #define BPF_MOV32_IMM(DST, IMM) \
143 ((struct bpf_insn) { \
144 .code = BPF_ALU | BPF_MOV | BPF_K, \
145 .dst_reg = DST, \
146 .src_reg = 0, \
147 .off = 0, \
148 .imm = IMM })
149
150 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
151 #define BPF_LD_IMM64(DST, IMM) \
152 BPF_LD_IMM64_RAW(DST, 0, IMM)
153
154 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
155 ((struct bpf_insn) { \
156 .code = BPF_LD | BPF_DW | BPF_IMM, \
157 .dst_reg = DST, \
158 .src_reg = SRC, \
159 .off = 0, \
160 .imm = (__u32) (IMM) }), \
161 ((struct bpf_insn) { \
162 .code = 0, /* zero is reserved opcode */ \
163 .dst_reg = 0, \
164 .src_reg = 0, \
165 .off = 0, \
166 .imm = ((__u64) (IMM)) >> 32 })
167
168 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
169 #define BPF_LD_MAP_FD(DST, MAP_FD) \
170 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
171
172 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
173
174 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
175 ((struct bpf_insn) { \
176 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
177 .dst_reg = DST, \
178 .src_reg = SRC, \
179 .off = 0, \
180 .imm = IMM })
181
182 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
183 ((struct bpf_insn) { \
184 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
185 .dst_reg = DST, \
186 .src_reg = SRC, \
187 .off = 0, \
188 .imm = IMM })
189
190 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
191
192 #define BPF_LD_ABS(SIZE, IMM) \
193 ((struct bpf_insn) { \
194 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
195 .dst_reg = 0, \
196 .src_reg = 0, \
197 .off = 0, \
198 .imm = IMM })
199
200 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
201
202 #define BPF_LD_IND(SIZE, SRC, IMM) \
203 ((struct bpf_insn) { \
204 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
205 .dst_reg = 0, \
206 .src_reg = SRC, \
207 .off = 0, \
208 .imm = IMM })
209
210 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
211
212 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
213 ((struct bpf_insn) { \
214 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
215 .dst_reg = DST, \
216 .src_reg = SRC, \
217 .off = OFF, \
218 .imm = 0 })
219
220 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
221
222 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
223 ((struct bpf_insn) { \
224 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
225 .dst_reg = DST, \
226 .src_reg = SRC, \
227 .off = OFF, \
228 .imm = 0 })
229
230 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
231
232 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
233 ((struct bpf_insn) { \
234 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
235 .dst_reg = DST, \
236 .src_reg = SRC, \
237 .off = OFF, \
238 .imm = 0 })
239
240 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
241
242 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
243 ((struct bpf_insn) { \
244 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
245 .dst_reg = DST, \
246 .src_reg = 0, \
247 .off = OFF, \
248 .imm = IMM })
249
250 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
251
252 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
253 ((struct bpf_insn) { \
254 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
255 .dst_reg = DST, \
256 .src_reg = SRC, \
257 .off = OFF, \
258 .imm = 0 })
259
260 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
261
262 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
263 ((struct bpf_insn) { \
264 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
265 .dst_reg = DST, \
266 .src_reg = 0, \
267 .off = OFF, \
268 .imm = IMM })
269
270 /* Unconditional jumps, goto pc + off16 */
271
272 #define BPF_JMP_A(OFF) \
273 ((struct bpf_insn) { \
274 .code = BPF_JMP | BPF_JA, \
275 .dst_reg = 0, \
276 .src_reg = 0, \
277 .off = OFF, \
278 .imm = 0 })
279
280 /* Function call */
281
282 #define BPF_EMIT_CALL(FUNC) \
283 ((struct bpf_insn) { \
284 .code = BPF_JMP | BPF_CALL, \
285 .dst_reg = 0, \
286 .src_reg = 0, \
287 .off = 0, \
288 .imm = ((FUNC) - __bpf_call_base) })
289
290 /* Raw code statement block */
291
292 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
293 ((struct bpf_insn) { \
294 .code = CODE, \
295 .dst_reg = DST, \
296 .src_reg = SRC, \
297 .off = OFF, \
298 .imm = IMM })
299
300 /* Program exit */
301
302 #define BPF_EXIT_INSN() \
303 ((struct bpf_insn) { \
304 .code = BPF_JMP | BPF_EXIT, \
305 .dst_reg = 0, \
306 .src_reg = 0, \
307 .off = 0, \
308 .imm = 0 })
309
310 /* Internal classic blocks for direct assignment */
311
312 #define __BPF_STMT(CODE, K) \
313 ((struct sock_filter) BPF_STMT(CODE, K))
314
315 #define __BPF_JUMP(CODE, K, JT, JF) \
316 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
317
318 #define bytes_to_bpf_size(bytes) \
319 ({ \
320 int bpf_size = -EINVAL; \
321 \
322 if (bytes == sizeof(u8)) \
323 bpf_size = BPF_B; \
324 else if (bytes == sizeof(u16)) \
325 bpf_size = BPF_H; \
326 else if (bytes == sizeof(u32)) \
327 bpf_size = BPF_W; \
328 else if (bytes == sizeof(u64)) \
329 bpf_size = BPF_DW; \
330 \
331 bpf_size; \
332 })
333
334 #define bpf_size_to_bytes(bpf_size) \
335 ({ \
336 int bytes = -EINVAL; \
337 \
338 if (bpf_size == BPF_B) \
339 bytes = sizeof(u8); \
340 else if (bpf_size == BPF_H) \
341 bytes = sizeof(u16); \
342 else if (bpf_size == BPF_W) \
343 bytes = sizeof(u32); \
344 else if (bpf_size == BPF_DW) \
345 bytes = sizeof(u64); \
346 \
347 bytes; \
348 })
349
350 #define BPF_SIZEOF(type) \
351 ({ \
352 const int __size = bytes_to_bpf_size(sizeof(type)); \
353 BUILD_BUG_ON(__size < 0); \
354 __size; \
355 })
356
357 #define BPF_FIELD_SIZEOF(type, field) \
358 ({ \
359 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
360 BUILD_BUG_ON(__size < 0); \
361 __size; \
362 })
363
364 #define BPF_LDST_BYTES(insn) \
365 ({ \
366 const int __size = bpf_size_to_bytes(BPF_SIZE(insn->code)); \
367 WARN_ON(__size < 0); \
368 __size; \
369 })
370
371 #define __BPF_MAP_0(m, v, ...) v
372 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
373 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
374 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
375 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
376 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
377
378 #define __BPF_REG_0(...) __BPF_PAD(5)
379 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
380 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
381 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
382 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
383 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
384
385 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
386 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
387
388 #define __BPF_CAST(t, a) \
389 (__force t) \
390 (__force \
391 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
392 (unsigned long)0, (t)0))) a
393 #define __BPF_V void
394 #define __BPF_N
395
396 #define __BPF_DECL_ARGS(t, a) t a
397 #define __BPF_DECL_REGS(t, a) u64 a
398
399 #define __BPF_PAD(n) \
400 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
401 u64, __ur_3, u64, __ur_4, u64, __ur_5)
402
403 #define BPF_CALL_x(x, name, ...) \
404 static __always_inline \
405 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
406 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
407 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
408 { \
409 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
410 } \
411 static __always_inline \
412 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
413
414 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
415 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
416 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
417 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
418 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
419 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
420
421 #define bpf_ctx_range(TYPE, MEMBER) \
422 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
423 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
424 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
425
426 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
427 ({ \
428 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE)); \
429 *(PTR_SIZE) = (SIZE); \
430 offsetof(TYPE, MEMBER); \
431 })
432
433 #ifdef CONFIG_COMPAT
434 /* A struct sock_filter is architecture independent. */
435 struct compat_sock_fprog {
436 u16 len;
437 compat_uptr_t filter; /* struct sock_filter * */
438 };
439 #endif
440
441 struct sock_fprog_kern {
442 u16 len;
443 struct sock_filter *filter;
444 };
445
446 struct bpf_binary_header {
447 unsigned int pages;
448
449 /* Some arches need word alignment for their instructions */
450 u8 image[] __aligned(4);
451 };
452
453 struct bpf_prog {
454 u16 pages; /* Number of allocated pages */
455 u16 jited:1, /* Is our filter JIT'ed? */
456 locked:1, /* Program image locked? */
457 gpl_compatible:1, /* Is filter GPL compatible? */
458 cb_access:1, /* Is control block accessed? */
459 dst_needed:1; /* Do we need dst entry? */
460 enum bpf_prog_type type; /* Type of BPF program */
461 u32 len; /* Number of filter blocks */
462 u32 jited_len; /* Size of jited insns in bytes */
463 u8 tag[BPF_TAG_SIZE];
464 struct bpf_prog_aux *aux; /* Auxiliary fields */
465 struct sock_fprog_kern *orig_prog; /* Original BPF program */
466 unsigned int (*bpf_func)(const void *ctx,
467 const struct bpf_insn *insn);
468 /* Instructions for interpreter */
469 union {
470 struct sock_filter insns[0];
471 struct bpf_insn insnsi[0];
472 };
473 };
474
475 struct sk_filter {
476 refcount_t refcnt;
477 struct rcu_head rcu;
478 struct bpf_prog *prog;
479 };
480
481 #define BPF_PROG_RUN(filter, ctx) (*(filter)->bpf_func)(ctx, (filter)->insnsi)
482
483 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
484
485 struct bpf_skb_data_end {
486 struct qdisc_skb_cb qdisc_cb;
487 void *data_meta;
488 void *data_end;
489 };
490
491 struct xdp_buff {
492 void *data;
493 void *data_end;
494 void *data_meta;
495 void *data_hard_start;
496 };
497
498 /* Compute the linear packet data range [data, data_end) which
499 * will be accessed by various program types (cls_bpf, act_bpf,
500 * lwt, ...). Subsystems allowing direct data access must (!)
501 * ensure that cb[] area can be written to when BPF program is
502 * invoked (otherwise cb[] save/restore is necessary).
503 */
504 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
505 {
506 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
507
508 BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
509 cb->data_meta = skb->data - skb_metadata_len(skb);
510 cb->data_end = skb->data + skb_headlen(skb);
511 }
512
513 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
514 {
515 /* eBPF programs may read/write skb->cb[] area to transfer meta
516 * data between tail calls. Since this also needs to work with
517 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
518 *
519 * In some socket filter cases, the cb unfortunately needs to be
520 * saved/restored so that protocol specific skb->cb[] data won't
521 * be lost. In any case, due to unpriviledged eBPF programs
522 * attached to sockets, we need to clear the bpf_skb_cb() area
523 * to not leak previous contents to user space.
524 */
525 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
526 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
527 FIELD_SIZEOF(struct qdisc_skb_cb, data));
528
529 return qdisc_skb_cb(skb)->data;
530 }
531
532 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
533 struct sk_buff *skb)
534 {
535 u8 *cb_data = bpf_skb_cb(skb);
536 u8 cb_saved[BPF_SKB_CB_LEN];
537 u32 res;
538
539 if (unlikely(prog->cb_access)) {
540 memcpy(cb_saved, cb_data, sizeof(cb_saved));
541 memset(cb_data, 0, sizeof(cb_saved));
542 }
543
544 res = BPF_PROG_RUN(prog, skb);
545
546 if (unlikely(prog->cb_access))
547 memcpy(cb_data, cb_saved, sizeof(cb_saved));
548
549 return res;
550 }
551
552 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
553 struct sk_buff *skb)
554 {
555 u8 *cb_data = bpf_skb_cb(skb);
556
557 if (unlikely(prog->cb_access))
558 memset(cb_data, 0, BPF_SKB_CB_LEN);
559
560 return BPF_PROG_RUN(prog, skb);
561 }
562
563 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
564 struct xdp_buff *xdp)
565 {
566 /* Caller needs to hold rcu_read_lock() (!), otherwise program
567 * can be released while still running, or map elements could be
568 * freed early while still having concurrent users. XDP fastpath
569 * already takes rcu_read_lock() when fetching the program, so
570 * it's not necessary here anymore.
571 */
572 return BPF_PROG_RUN(prog, xdp);
573 }
574
575 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
576 {
577 return prog->len * sizeof(struct bpf_insn);
578 }
579
580 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
581 {
582 return round_up(bpf_prog_insn_size(prog) +
583 sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
584 }
585
586 static inline unsigned int bpf_prog_size(unsigned int proglen)
587 {
588 return max(sizeof(struct bpf_prog),
589 offsetof(struct bpf_prog, insns[proglen]));
590 }
591
592 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
593 {
594 /* When classic BPF programs have been loaded and the arch
595 * does not have a classic BPF JIT (anymore), they have been
596 * converted via bpf_migrate_filter() to eBPF and thus always
597 * have an unspec program type.
598 */
599 return prog->type == BPF_PROG_TYPE_UNSPEC;
600 }
601
602 static inline bool
603 bpf_ctx_narrow_access_ok(u32 off, u32 size, const u32 size_default)
604 {
605 bool off_ok;
606 #ifdef __LITTLE_ENDIAN
607 off_ok = (off & (size_default - 1)) == 0;
608 #else
609 off_ok = (off & (size_default - 1)) + size == size_default;
610 #endif
611 return off_ok && size <= size_default && (size & (size - 1)) == 0;
612 }
613
614 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
615
616 #ifdef CONFIG_ARCH_HAS_SET_MEMORY
617 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
618 {
619 fp->locked = 1;
620 WARN_ON_ONCE(set_memory_ro((unsigned long)fp, fp->pages));
621 }
622
623 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
624 {
625 if (fp->locked) {
626 WARN_ON_ONCE(set_memory_rw((unsigned long)fp, fp->pages));
627 /* In case set_memory_rw() fails, we want to be the first
628 * to crash here instead of some random place later on.
629 */
630 fp->locked = 0;
631 }
632 }
633
634 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
635 {
636 WARN_ON_ONCE(set_memory_ro((unsigned long)hdr, hdr->pages));
637 }
638
639 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
640 {
641 WARN_ON_ONCE(set_memory_rw((unsigned long)hdr, hdr->pages));
642 }
643 #else
644 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
645 {
646 }
647
648 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
649 {
650 }
651
652 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
653 {
654 }
655
656 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
657 {
658 }
659 #endif /* CONFIG_ARCH_HAS_SET_MEMORY */
660
661 static inline struct bpf_binary_header *
662 bpf_jit_binary_hdr(const struct bpf_prog *fp)
663 {
664 unsigned long real_start = (unsigned long)fp->bpf_func;
665 unsigned long addr = real_start & PAGE_MASK;
666
667 return (void *)addr;
668 }
669
670 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
671 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
672 {
673 return sk_filter_trim_cap(sk, skb, 1);
674 }
675
676 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
677 void bpf_prog_free(struct bpf_prog *fp);
678
679 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
680 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
681 gfp_t gfp_extra_flags);
682 void __bpf_prog_free(struct bpf_prog *fp);
683
684 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
685 {
686 bpf_prog_unlock_ro(fp);
687 __bpf_prog_free(fp);
688 }
689
690 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
691 unsigned int flen);
692
693 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
694 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
695 bpf_aux_classic_check_t trans, bool save_orig);
696 void bpf_prog_destroy(struct bpf_prog *fp);
697
698 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
699 int sk_attach_bpf(u32 ufd, struct sock *sk);
700 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
701 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
702 int sk_detach_filter(struct sock *sk);
703 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
704 unsigned int len);
705
706 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
707 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
708
709 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
710
711 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
712 void bpf_jit_compile(struct bpf_prog *prog);
713 bool bpf_helper_changes_pkt_data(void *func);
714
715 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
716 const struct bpf_insn *patch, u32 len);
717
718 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
719 * same cpu context. Further for best results no more than a single map
720 * for the do_redirect/do_flush pair should be used. This limitation is
721 * because we only track one map and force a flush when the map changes.
722 * This does not appear to be a real limitation for existing software.
723 */
724 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
725 struct bpf_prog *prog);
726 int xdp_do_redirect(struct net_device *dev,
727 struct xdp_buff *xdp,
728 struct bpf_prog *prog);
729 void xdp_do_flush_map(void);
730
731 /* Drivers not supporting XDP metadata can use this helper, which
732 * rejects any room expansion for metadata as a result.
733 */
734 static __always_inline void
735 xdp_set_data_meta_invalid(struct xdp_buff *xdp)
736 {
737 xdp->data_meta = xdp->data + 1;
738 }
739
740 static __always_inline bool
741 xdp_data_meta_unsupported(const struct xdp_buff *xdp)
742 {
743 return unlikely(xdp->data_meta > xdp->data);
744 }
745
746 void bpf_warn_invalid_xdp_action(u32 act);
747
748 struct sock *do_sk_redirect_map(struct sk_buff *skb);
749
750 #ifdef CONFIG_BPF_JIT
751 extern int bpf_jit_enable;
752 extern int bpf_jit_harden;
753 extern int bpf_jit_kallsyms;
754
755 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
756
757 struct bpf_binary_header *
758 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
759 unsigned int alignment,
760 bpf_jit_fill_hole_t bpf_fill_ill_insns);
761 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
762 u64 bpf_jit_alloc_exec_limit(void);
763 void *bpf_jit_alloc_exec(unsigned long size);
764 void bpf_jit_free_exec(void *addr);
765 void bpf_jit_free(struct bpf_prog *fp);
766
767 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
768 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
769
770 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
771 u32 pass, void *image)
772 {
773 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
774 proglen, pass, image, current->comm, task_pid_nr(current));
775
776 if (image)
777 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
778 16, 1, image, proglen, false);
779 }
780
781 static inline bool bpf_jit_is_ebpf(void)
782 {
783 # ifdef CONFIG_HAVE_EBPF_JIT
784 return true;
785 # else
786 return false;
787 # endif
788 }
789
790 static inline bool ebpf_jit_enabled(void)
791 {
792 return bpf_jit_enable && bpf_jit_is_ebpf();
793 }
794
795 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
796 {
797 return fp->jited && bpf_jit_is_ebpf();
798 }
799
800 static inline bool bpf_jit_blinding_enabled(void)
801 {
802 /* These are the prerequisites, should someone ever have the
803 * idea to call blinding outside of them, we make sure to
804 * bail out.
805 */
806 if (!bpf_jit_is_ebpf())
807 return false;
808 if (!bpf_jit_enable)
809 return false;
810 if (!bpf_jit_harden)
811 return false;
812 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
813 return false;
814
815 return true;
816 }
817
818 static inline bool bpf_jit_kallsyms_enabled(void)
819 {
820 /* There are a couple of corner cases where kallsyms should
821 * not be enabled f.e. on hardening.
822 */
823 if (bpf_jit_harden)
824 return false;
825 if (!bpf_jit_kallsyms)
826 return false;
827 if (bpf_jit_kallsyms == 1)
828 return true;
829
830 return false;
831 }
832
833 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
834 unsigned long *off, char *sym);
835 bool is_bpf_text_address(unsigned long addr);
836 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
837 char *sym);
838
839 static inline const char *
840 bpf_address_lookup(unsigned long addr, unsigned long *size,
841 unsigned long *off, char **modname, char *sym)
842 {
843 const char *ret = __bpf_address_lookup(addr, size, off, sym);
844
845 if (ret && modname)
846 *modname = NULL;
847 return ret;
848 }
849
850 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
851 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
852
853 #else /* CONFIG_BPF_JIT */
854
855 static inline bool ebpf_jit_enabled(void)
856 {
857 return false;
858 }
859
860 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
861 {
862 return false;
863 }
864
865 static inline void bpf_jit_free(struct bpf_prog *fp)
866 {
867 bpf_prog_unlock_free(fp);
868 }
869
870 static inline bool bpf_jit_kallsyms_enabled(void)
871 {
872 return false;
873 }
874
875 static inline const char *
876 __bpf_address_lookup(unsigned long addr, unsigned long *size,
877 unsigned long *off, char *sym)
878 {
879 return NULL;
880 }
881
882 static inline bool is_bpf_text_address(unsigned long addr)
883 {
884 return false;
885 }
886
887 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
888 char *type, char *sym)
889 {
890 return -ERANGE;
891 }
892
893 static inline const char *
894 bpf_address_lookup(unsigned long addr, unsigned long *size,
895 unsigned long *off, char **modname, char *sym)
896 {
897 return NULL;
898 }
899
900 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
901 {
902 }
903
904 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
905 {
906 }
907 #endif /* CONFIG_BPF_JIT */
908
909 #define BPF_ANC BIT(15)
910
911 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
912 {
913 switch (first->code) {
914 case BPF_RET | BPF_K:
915 case BPF_LD | BPF_W | BPF_LEN:
916 return false;
917
918 case BPF_LD | BPF_W | BPF_ABS:
919 case BPF_LD | BPF_H | BPF_ABS:
920 case BPF_LD | BPF_B | BPF_ABS:
921 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
922 return true;
923 return false;
924
925 default:
926 return true;
927 }
928 }
929
930 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
931 {
932 BUG_ON(ftest->code & BPF_ANC);
933
934 switch (ftest->code) {
935 case BPF_LD | BPF_W | BPF_ABS:
936 case BPF_LD | BPF_H | BPF_ABS:
937 case BPF_LD | BPF_B | BPF_ABS:
938 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
939 return BPF_ANC | SKF_AD_##CODE
940 switch (ftest->k) {
941 BPF_ANCILLARY(PROTOCOL);
942 BPF_ANCILLARY(PKTTYPE);
943 BPF_ANCILLARY(IFINDEX);
944 BPF_ANCILLARY(NLATTR);
945 BPF_ANCILLARY(NLATTR_NEST);
946 BPF_ANCILLARY(MARK);
947 BPF_ANCILLARY(QUEUE);
948 BPF_ANCILLARY(HATYPE);
949 BPF_ANCILLARY(RXHASH);
950 BPF_ANCILLARY(CPU);
951 BPF_ANCILLARY(ALU_XOR_X);
952 BPF_ANCILLARY(VLAN_TAG);
953 BPF_ANCILLARY(VLAN_TAG_PRESENT);
954 BPF_ANCILLARY(PAY_OFFSET);
955 BPF_ANCILLARY(RANDOM);
956 BPF_ANCILLARY(VLAN_TPID);
957 }
958 /* Fallthrough. */
959 default:
960 return ftest->code;
961 }
962 }
963
964 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
965 int k, unsigned int size);
966
967 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
968 unsigned int size, void *buffer)
969 {
970 if (k >= 0)
971 return skb_header_pointer(skb, k, size, buffer);
972
973 return bpf_internal_load_pointer_neg_helper(skb, k, size);
974 }
975
976 static inline int bpf_tell_extensions(void)
977 {
978 return SKF_AD_MAX;
979 }
980
981 struct bpf_sock_ops_kern {
982 struct sock *sk;
983 u32 op;
984 union {
985 u32 reply;
986 u32 replylong[4];
987 };
988 };
989
990 #endif /* __LINUX_FILTER_H__ */
ubuntu-bionic-kernel mirror