Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
1/*
2 * Linux Socket Filter Data Structures
3 */
4#ifndef __LINUX_FILTER_H__
5#define __LINUX_FILTER_H__
6
7#include <stdarg.h>
8
9#include <linux/atomic.h>
10#include <linux/compat.h>
11#include <linux/skbuff.h>
12#include <linux/linkage.h>
13#include <linux/printk.h>
14#include <linux/workqueue.h>
15
16#include <asm/cacheflush.h>
17
18#include <uapi/linux/filter.h>
19#include <uapi/linux/bpf.h>
20
21struct sk_buff;
22struct sock;
23struct seccomp_data;
24struct bpf_prog_aux;
25
26/* ArgX, context and stack frame pointer register positions. Note,
27 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
28 * calls in BPF_CALL instruction.
29 */
30#define BPF_REG_ARG1 BPF_REG_1
31#define BPF_REG_ARG2 BPF_REG_2
32#define BPF_REG_ARG3 BPF_REG_3
33#define BPF_REG_ARG4 BPF_REG_4
34#define BPF_REG_ARG5 BPF_REG_5
35#define BPF_REG_CTX BPF_REG_6
36#define BPF_REG_FP BPF_REG_10
37
38/* Additional register mappings for converted user programs. */
39#define BPF_REG_A BPF_REG_0
40#define BPF_REG_X BPF_REG_7
41#define BPF_REG_TMP BPF_REG_8
42
43/* BPF program can access up to 512 bytes of stack space. */
44#define MAX_BPF_STACK 512
45
46/* Helper macros for filter block array initializers. */
47
48/* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
49
50#define BPF_ALU64_REG(OP, DST, SRC) \
51 ((struct bpf_insn) { \
52 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
53 .dst_reg = DST, \
54 .src_reg = SRC, \
55 .off = 0, \
56 .imm = 0 })
57
58#define BPF_ALU32_REG(OP, DST, SRC) \
59 ((struct bpf_insn) { \
60 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
61 .dst_reg = DST, \
62 .src_reg = SRC, \
63 .off = 0, \
64 .imm = 0 })
65
66/* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
67
68#define BPF_ALU64_IMM(OP, DST, IMM) \
69 ((struct bpf_insn) { \
70 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
71 .dst_reg = DST, \
72 .src_reg = 0, \
73 .off = 0, \
74 .imm = IMM })
75
76#define BPF_ALU32_IMM(OP, DST, IMM) \
77 ((struct bpf_insn) { \
78 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
79 .dst_reg = DST, \
80 .src_reg = 0, \
81 .off = 0, \
82 .imm = IMM })
83
84/* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
85
86#define BPF_ENDIAN(TYPE, DST, LEN) \
87 ((struct bpf_insn) { \
88 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
89 .dst_reg = DST, \
90 .src_reg = 0, \
91 .off = 0, \
92 .imm = LEN })
93
94/* Short form of mov, dst_reg = src_reg */
95
96#define BPF_MOV64_REG(DST, SRC) \
97 ((struct bpf_insn) { \
98 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
99 .dst_reg = DST, \
100 .src_reg = SRC, \
101 .off = 0, \
102 .imm = 0 })
103
104#define BPF_MOV32_REG(DST, SRC) \
105 ((struct bpf_insn) { \
106 .code = BPF_ALU | BPF_MOV | BPF_X, \
107 .dst_reg = DST, \
108 .src_reg = SRC, \
109 .off = 0, \
110 .imm = 0 })
111
112/* Short form of mov, dst_reg = imm32 */
113
114#define BPF_MOV64_IMM(DST, IMM) \
115 ((struct bpf_insn) { \
116 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
117 .dst_reg = DST, \
118 .src_reg = 0, \
119 .off = 0, \
120 .imm = IMM })
121
122#define BPF_MOV32_IMM(DST, IMM) \
123 ((struct bpf_insn) { \
124 .code = BPF_ALU | BPF_MOV | BPF_K, \
125 .dst_reg = DST, \
126 .src_reg = 0, \
127 .off = 0, \
128 .imm = IMM })
129
130/* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
131#define BPF_LD_IMM64(DST, IMM) \
132 BPF_LD_IMM64_RAW(DST, 0, IMM)
133
134#define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
135 ((struct bpf_insn) { \
136 .code = BPF_LD | BPF_DW | BPF_IMM, \
137 .dst_reg = DST, \
138 .src_reg = SRC, \
139 .off = 0, \
140 .imm = (__u32) (IMM) }), \
141 ((struct bpf_insn) { \
142 .code = 0, /* zero is reserved opcode */ \
143 .dst_reg = 0, \
144 .src_reg = 0, \
145 .off = 0, \
146 .imm = ((__u64) (IMM)) >> 32 })
147
148/* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
149#define BPF_LD_MAP_FD(DST, MAP_FD) \
150 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
151
152/* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
153
154#define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
155 ((struct bpf_insn) { \
156 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
157 .dst_reg = DST, \
158 .src_reg = SRC, \
159 .off = 0, \
160 .imm = IMM })
161
162#define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
163 ((struct bpf_insn) { \
164 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
165 .dst_reg = DST, \
166 .src_reg = SRC, \
167 .off = 0, \
168 .imm = IMM })
169
170/* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
171
172#define BPF_LD_ABS(SIZE, IMM) \
173 ((struct bpf_insn) { \
174 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
175 .dst_reg = 0, \
176 .src_reg = 0, \
177 .off = 0, \
178 .imm = IMM })
179
180/* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
181
182#define BPF_LD_IND(SIZE, SRC, IMM) \
183 ((struct bpf_insn) { \
184 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
185 .dst_reg = 0, \
186 .src_reg = SRC, \
187 .off = 0, \
188 .imm = IMM })
189
190/* Memory load, dst_reg = *(uint *) (src_reg + off16) */
191
192#define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
193 ((struct bpf_insn) { \
194 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
195 .dst_reg = DST, \
196 .src_reg = SRC, \
197 .off = OFF, \
198 .imm = 0 })
199
200/* Memory store, *(uint *) (dst_reg + off16) = src_reg */
201
202#define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
203 ((struct bpf_insn) { \
204 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
205 .dst_reg = DST, \
206 .src_reg = SRC, \
207 .off = OFF, \
208 .imm = 0 })
209
210/* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
211
212#define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
213 ((struct bpf_insn) { \
214 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
215 .dst_reg = DST, \
216 .src_reg = SRC, \
217 .off = OFF, \
218 .imm = 0 })
219
220/* Memory store, *(uint *) (dst_reg + off16) = imm32 */
221
222#define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
223 ((struct bpf_insn) { \
224 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
225 .dst_reg = DST, \
226 .src_reg = 0, \
227 .off = OFF, \
228 .imm = IMM })
229
230/* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
231
232#define BPF_JMP_REG(OP, DST, SRC, OFF) \
233 ((struct bpf_insn) { \
234 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
235 .dst_reg = DST, \
236 .src_reg = SRC, \
237 .off = OFF, \
238 .imm = 0 })
239
240/* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
241
242#define BPF_JMP_IMM(OP, DST, IMM, OFF) \
243 ((struct bpf_insn) { \
244 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
245 .dst_reg = DST, \
246 .src_reg = 0, \
247 .off = OFF, \
248 .imm = IMM })
249
250/* Function call */
251
252#define BPF_EMIT_CALL(FUNC) \
253 ((struct bpf_insn) { \
254 .code = BPF_JMP | BPF_CALL, \
255 .dst_reg = 0, \
256 .src_reg = 0, \
257 .off = 0, \
258 .imm = ((FUNC) - __bpf_call_base) })
259
260/* Raw code statement block */
261
262#define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
263 ((struct bpf_insn) { \
264 .code = CODE, \
265 .dst_reg = DST, \
266 .src_reg = SRC, \
267 .off = OFF, \
268 .imm = IMM })
269
270/* Program exit */
271
272#define BPF_EXIT_INSN() \
273 ((struct bpf_insn) { \
274 .code = BPF_JMP | BPF_EXIT, \
275 .dst_reg = 0, \
276 .src_reg = 0, \
277 .off = 0, \
278 .imm = 0 })
279
280/* Internal classic blocks for direct assignment */
281
282#define __BPF_STMT(CODE, K) \
283 ((struct sock_filter) BPF_STMT(CODE, K))
284
285#define __BPF_JUMP(CODE, K, JT, JF) \
286 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
287
288#define bytes_to_bpf_size(bytes) \
289({ \
290 int bpf_size = -EINVAL; \
291 \
292 if (bytes == sizeof(u8)) \
293 bpf_size = BPF_B; \
294 else if (bytes == sizeof(u16)) \
295 bpf_size = BPF_H; \
296 else if (bytes == sizeof(u32)) \
297 bpf_size = BPF_W; \
298 else if (bytes == sizeof(u64)) \
299 bpf_size = BPF_DW; \
300 \
301 bpf_size; \
302})
303
304/* Macro to invoke filter function. */
305#define SK_RUN_FILTER(filter, ctx) \
306 (*filter->prog->bpf_func)(ctx, filter->prog->insnsi)
307
308#ifdef CONFIG_COMPAT
309/* A struct sock_filter is architecture independent. */
310struct compat_sock_fprog {
311 u16 len;
312 compat_uptr_t filter; /* struct sock_filter * */
313};
314#endif
315
316struct sock_fprog_kern {
317 u16 len;
318 struct sock_filter *filter;
319};
320
321struct bpf_binary_header {
322 unsigned int pages;
323 u8 image[];
324};
325
326struct bpf_prog {
327 u16 pages; /* Number of allocated pages */
328 bool jited; /* Is our filter JIT'ed? */
329 bool gpl_compatible; /* Is our filter GPL compatible? */
330 u32 len; /* Number of filter blocks */
331 enum bpf_prog_type type; /* Type of BPF program */
332 struct bpf_prog_aux *aux; /* Auxiliary fields */
333 struct sock_fprog_kern *orig_prog; /* Original BPF program */
334 unsigned int (*bpf_func)(const struct sk_buff *skb,
335 const struct bpf_insn *filter);
336 /* Instructions for interpreter */
337 union {
338 struct sock_filter insns[0];
339 struct bpf_insn insnsi[0];
340 };
341};
342
343struct sk_filter {
344 atomic_t refcnt;
345 struct rcu_head rcu;
346 struct bpf_prog *prog;
347};
348
349#define BPF_PROG_RUN(filter, ctx) (*filter->bpf_func)(ctx, filter->insnsi)
350
351static inline unsigned int bpf_prog_size(unsigned int proglen)
352{
353 return max(sizeof(struct bpf_prog),
354 offsetof(struct bpf_prog, insns[proglen]));
355}
356
357#define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
358
359#ifdef CONFIG_DEBUG_SET_MODULE_RONX
360static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
361{
362 set_memory_ro((unsigned long)fp, fp->pages);
363}
364
365static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
366{
367 set_memory_rw((unsigned long)fp, fp->pages);
368}
369#else
370static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
371{
372}
373
374static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
375{
376}
377#endif /* CONFIG_DEBUG_SET_MODULE_RONX */
378
379int sk_filter(struct sock *sk, struct sk_buff *skb);
380
381int bpf_prog_select_runtime(struct bpf_prog *fp);
382void bpf_prog_free(struct bpf_prog *fp);
383
384struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
385struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
386 gfp_t gfp_extra_flags);
387void __bpf_prog_free(struct bpf_prog *fp);
388
389static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
390{
391 bpf_prog_unlock_ro(fp);
392 __bpf_prog_free(fp);
393}
394
395typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
396 unsigned int flen);
397
398int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
399int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
400 bpf_aux_classic_check_t trans);
401void bpf_prog_destroy(struct bpf_prog *fp);
402
403int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
404int sk_attach_bpf(u32 ufd, struct sock *sk);
405int sk_detach_filter(struct sock *sk);
406int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
407 unsigned int len);
408
409bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
410void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
411
412u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
413void bpf_int_jit_compile(struct bpf_prog *fp);
414
415#ifdef CONFIG_BPF_JIT
416typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
417
418struct bpf_binary_header *
419bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
420 unsigned int alignment,
421 bpf_jit_fill_hole_t bpf_fill_ill_insns);
422void bpf_jit_binary_free(struct bpf_binary_header *hdr);
423
424void bpf_jit_compile(struct bpf_prog *fp);
425void bpf_jit_free(struct bpf_prog *fp);
426
427static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
428 u32 pass, void *image)
429{
430 pr_err("flen=%u proglen=%u pass=%u image=%pK\n",
431 flen, proglen, pass, image);
432 if (image)
433 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
434 16, 1, image, proglen, false);
435}
436#else
437static inline void bpf_jit_compile(struct bpf_prog *fp)
438{
439}
440
441static inline void bpf_jit_free(struct bpf_prog *fp)
442{
443 bpf_prog_unlock_free(fp);
444}
445#endif /* CONFIG_BPF_JIT */
446
447#define BPF_ANC BIT(15)
448
449static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
450{
451 BUG_ON(ftest->code & BPF_ANC);
452
453 switch (ftest->code) {
454 case BPF_LD | BPF_W | BPF_ABS:
455 case BPF_LD | BPF_H | BPF_ABS:
456 case BPF_LD | BPF_B | BPF_ABS:
457#define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
458 return BPF_ANC | SKF_AD_##CODE
459 switch (ftest->k) {
460 BPF_ANCILLARY(PROTOCOL);
461 BPF_ANCILLARY(PKTTYPE);
462 BPF_ANCILLARY(IFINDEX);
463 BPF_ANCILLARY(NLATTR);
464 BPF_ANCILLARY(NLATTR_NEST);
465 BPF_ANCILLARY(MARK);
466 BPF_ANCILLARY(QUEUE);
467 BPF_ANCILLARY(HATYPE);
468 BPF_ANCILLARY(RXHASH);
469 BPF_ANCILLARY(CPU);
470 BPF_ANCILLARY(ALU_XOR_X);
471 BPF_ANCILLARY(VLAN_TAG);
472 BPF_ANCILLARY(VLAN_TAG_PRESENT);
473 BPF_ANCILLARY(PAY_OFFSET);
474 BPF_ANCILLARY(RANDOM);
475 BPF_ANCILLARY(VLAN_TPID);
476 }
477 /* Fallthrough. */
478 default:
479 return ftest->code;
480 }
481}
482
483void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
484 int k, unsigned int size);
485
486static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
487 unsigned int size, void *buffer)
488{
489 if (k >= 0)
490 return skb_header_pointer(skb, k, size, buffer);
491
492 return bpf_internal_load_pointer_neg_helper(skb, k, size);
493}
494
495static inline int bpf_tell_extensions(void)
496{
497 return SKF_AD_MAX;
498}
499
500#endif /* __LINUX_FILTER_H__ */