]> exis.tech > repos - linux.git/blob - include/linux/filter.h
KVM: Move more code under CONFIG_HAVE_KVM_IRQFD
[linux.git] / include / linux / filter.h
1 /*
2 * Linux Socket Filter Data Structures
3 */
4 #ifndef __LINUX_FILTER_H__
5 #define __LINUX_FILTER_H__
6
7 #include <linux/atomic.h>
8 #include <linux/compat.h>
9 #include <linux/workqueue.h>
10 #include <uapi/linux/filter.h>
11
12 /* Internally used and optimized filter representation with extended
13 * instruction set based on top of classic BPF.
14 */
15
16 /* instruction classes */
17 #define BPF_ALU64 0x07 /* alu mode in double word width */
18
19 /* ld/ldx fields */
20 #define BPF_DW 0x18 /* double word */
21 #define BPF_XADD 0xc0 /* exclusive add */
22
23 /* alu/jmp fields */
24 #define BPF_MOV 0xb0 /* mov reg to reg */
25 #define BPF_ARSH 0xc0 /* sign extending arithmetic shift right */
26
27 /* change endianness of a register */
28 #define BPF_END 0xd0 /* flags for endianness conversion: */
29 #define BPF_TO_LE 0x00 /* convert to little-endian */
30 #define BPF_TO_BE 0x08 /* convert to big-endian */
31 #define BPF_FROM_LE BPF_TO_LE
32 #define BPF_FROM_BE BPF_TO_BE
33
34 #define BPF_JNE 0x50 /* jump != */
35 #define BPF_JSGT 0x60 /* SGT is signed '>', GT in x86 */
36 #define BPF_JSGE 0x70 /* SGE is signed '>=', GE in x86 */
37 #define BPF_CALL 0x80 /* function call */
38 #define BPF_EXIT 0x90 /* function return */
39
40 /* Register numbers */
41 enum {
42 BPF_REG_0 = 0,
43 BPF_REG_1,
44 BPF_REG_2,
45 BPF_REG_3,
46 BPF_REG_4,
47 BPF_REG_5,
48 BPF_REG_6,
49 BPF_REG_7,
50 BPF_REG_8,
51 BPF_REG_9,
52 BPF_REG_10,
53 __MAX_BPF_REG,
54 };
55
56 /* BPF has 10 general purpose 64-bit registers and stack frame. */
57 #define MAX_BPF_REG __MAX_BPF_REG
58
59 /* ArgX, context and stack frame pointer register positions. Note,
60 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
61 * calls in BPF_CALL instruction.
62 */
63 #define BPF_REG_ARG1 BPF_REG_1
64 #define BPF_REG_ARG2 BPF_REG_2
65 #define BPF_REG_ARG3 BPF_REG_3
66 #define BPF_REG_ARG4 BPF_REG_4
67 #define BPF_REG_ARG5 BPF_REG_5
68 #define BPF_REG_CTX BPF_REG_6
69 #define BPF_REG_FP BPF_REG_10
70
71 /* Additional register mappings for converted user programs. */
72 #define BPF_REG_A BPF_REG_0
73 #define BPF_REG_X BPF_REG_7
74 #define BPF_REG_TMP BPF_REG_8
75
76 /* BPF program can access up to 512 bytes of stack space. */
77 #define MAX_BPF_STACK 512
78
79 /* Helper macros for filter block array initializers. */
80
81 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
82
83 #define BPF_ALU64_REG(OP, DST, SRC) \
84 ((struct sock_filter_int) { \
85 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
86 .dst_reg = DST, \
87 .src_reg = SRC, \
88 .off = 0, \
89 .imm = 0 })
90
91 #define BPF_ALU32_REG(OP, DST, SRC) \
92 ((struct sock_filter_int) { \
93 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
94 .dst_reg = DST, \
95 .src_reg = SRC, \
96 .off = 0, \
97 .imm = 0 })
98
99 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
100
101 #define BPF_ALU64_IMM(OP, DST, IMM) \
102 ((struct sock_filter_int) { \
103 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
104 .dst_reg = DST, \
105 .src_reg = 0, \
106 .off = 0, \
107 .imm = IMM })
108
109 #define BPF_ALU32_IMM(OP, DST, IMM) \
110 ((struct sock_filter_int) { \
111 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
112 .dst_reg = DST, \
113 .src_reg = 0, \
114 .off = 0, \
115 .imm = IMM })
116
117 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
118
119 #define BPF_ENDIAN(TYPE, DST, LEN) \
120 ((struct sock_filter_int) { \
121 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
122 .dst_reg = DST, \
123 .src_reg = 0, \
124 .off = 0, \
125 .imm = LEN })
126
127 /* Short form of mov, dst_reg = src_reg */
128
129 #define BPF_MOV64_REG(DST, SRC) \
130 ((struct sock_filter_int) { \
131 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
132 .dst_reg = DST, \
133 .src_reg = SRC, \
134 .off = 0, \
135 .imm = 0 })
136
137 #define BPF_MOV32_REG(DST, SRC) \
138 ((struct sock_filter_int) { \
139 .code = BPF_ALU | BPF_MOV | BPF_X, \
140 .dst_reg = DST, \
141 .src_reg = SRC, \
142 .off = 0, \
143 .imm = 0 })
144
145 /* Short form of mov, dst_reg = imm32 */
146
147 #define BPF_MOV64_IMM(DST, IMM) \
148 ((struct sock_filter_int) { \
149 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
150 .dst_reg = DST, \
151 .src_reg = 0, \
152 .off = 0, \
153 .imm = IMM })
154
155 #define BPF_MOV32_IMM(DST, IMM) \
156 ((struct sock_filter_int) { \
157 .code = BPF_ALU | BPF_MOV | BPF_K, \
158 .dst_reg = DST, \
159 .src_reg = 0, \
160 .off = 0, \
161 .imm = IMM })
162
163 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
164
165 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
166 ((struct sock_filter_int) { \
167 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
168 .dst_reg = DST, \
169 .src_reg = SRC, \
170 .off = 0, \
171 .imm = IMM })
172
173 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
174 ((struct sock_filter_int) { \
175 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
176 .dst_reg = DST, \
177 .src_reg = SRC, \
178 .off = 0, \
179 .imm = IMM })
180
181 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
182
183 #define BPF_LD_ABS(SIZE, IMM) \
184 ((struct sock_filter_int) { \
185 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
186 .dst_reg = 0, \
187 .src_reg = 0, \
188 .off = 0, \
189 .imm = IMM })
190
191 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
192
193 #define BPF_LD_IND(SIZE, SRC, IMM) \
194 ((struct sock_filter_int) { \
195 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
196 .dst_reg = 0, \
197 .src_reg = SRC, \
198 .off = 0, \
199 .imm = IMM })
200
201 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
202
203 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
204 ((struct sock_filter_int) { \
205 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
206 .dst_reg = DST, \
207 .src_reg = SRC, \
208 .off = OFF, \
209 .imm = 0 })
210
211 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
212
213 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
214 ((struct sock_filter_int) { \
215 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
216 .dst_reg = DST, \
217 .src_reg = SRC, \
218 .off = OFF, \
219 .imm = 0 })
220
221 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
222
223 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
224 ((struct sock_filter_int) { \
225 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
226 .dst_reg = DST, \
227 .src_reg = 0, \
228 .off = OFF, \
229 .imm = IMM })
230
231 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
232
233 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
234 ((struct sock_filter_int) { \
235 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
236 .dst_reg = DST, \
237 .src_reg = SRC, \
238 .off = OFF, \
239 .imm = 0 })
240
241 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
242
243 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
244 ((struct sock_filter_int) { \
245 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
246 .dst_reg = DST, \
247 .src_reg = 0, \
248 .off = OFF, \
249 .imm = IMM })
250
251 /* Function call */
252
253 #define BPF_EMIT_CALL(FUNC) \
254 ((struct sock_filter_int) { \
255 .code = BPF_JMP | BPF_CALL, \
256 .dst_reg = 0, \
257 .src_reg = 0, \
258 .off = 0, \
259 .imm = ((FUNC) - __bpf_call_base) })
260
261 /* Raw code statement block */
262
263 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
264 ((struct sock_filter_int) { \
265 .code = CODE, \
266 .dst_reg = DST, \
267 .src_reg = SRC, \
268 .off = OFF, \
269 .imm = IMM })
270
271 /* Program exit */
272
273 #define BPF_EXIT_INSN() \
274 ((struct sock_filter_int) { \
275 .code = BPF_JMP | BPF_EXIT, \
276 .dst_reg = 0, \
277 .src_reg = 0, \
278 .off = 0, \
279 .imm = 0 })
280
281 #define bytes_to_bpf_size(bytes) \
282 ({ \
283 int bpf_size = -EINVAL; \
284 \
285 if (bytes == sizeof(u8)) \
286 bpf_size = BPF_B; \
287 else if (bytes == sizeof(u16)) \
288 bpf_size = BPF_H; \
289 else if (bytes == sizeof(u32)) \
290 bpf_size = BPF_W; \
291 else if (bytes == sizeof(u64)) \
292 bpf_size = BPF_DW; \
293 \
294 bpf_size; \
295 })
296
297 /* Macro to invoke filter function. */
298 #define SK_RUN_FILTER(filter, ctx) (*filter->bpf_func)(ctx, filter->insnsi)
299
300 struct sock_filter_int {
301 __u8 code; /* opcode */
302 __u8 dst_reg:4; /* dest register */
303 __u8 src_reg:4; /* source register */
304 __s16 off; /* signed offset */
305 __s32 imm; /* signed immediate constant */
306 };
307
308 #ifdef CONFIG_COMPAT
309 /* A struct sock_filter is architecture independent. */
310 struct compat_sock_fprog {
311 u16 len;
312 compat_uptr_t filter; /* struct sock_filter * */
313 };
314 #endif
315
316 struct sock_fprog_kern {
317 u16 len;
318 struct sock_filter *filter;
319 };
320
321 struct sk_buff;
322 struct sock;
323 struct seccomp_data;
324
325 struct sk_filter {
326 atomic_t refcnt;
327 u32 jited:1, /* Is our filter JIT'ed? */
328 len:31; /* Number of filter blocks */
329 struct sock_fprog_kern *orig_prog; /* Original BPF program */
330 struct rcu_head rcu;
331 unsigned int (*bpf_func)(const struct sk_buff *skb,
332 const struct sock_filter_int *filter);
333 union {
334 struct sock_filter insns[0];
335 struct sock_filter_int insnsi[0];
336 struct work_struct work;
337 };
338 };
339
340 static inline unsigned int sk_filter_size(unsigned int proglen)
341 {
342 return max(sizeof(struct sk_filter),
343 offsetof(struct sk_filter, insns[proglen]));
344 }
345
346 #define sk_filter_proglen(fprog) \
347 (fprog->len * sizeof(fprog->filter[0]))
348
349 int sk_filter(struct sock *sk, struct sk_buff *skb);
350
351 void sk_filter_select_runtime(struct sk_filter *fp);
352 void sk_filter_free(struct sk_filter *fp);
353
354 int sk_convert_filter(struct sock_filter *prog, int len,
355 struct sock_filter_int *new_prog, int *new_len);
356
357 int sk_unattached_filter_create(struct sk_filter **pfp,
358 struct sock_fprog_kern *fprog);
359 void sk_unattached_filter_destroy(struct sk_filter *fp);
360
361 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
362 int sk_detach_filter(struct sock *sk);
363
364 int sk_chk_filter(struct sock_filter *filter, unsigned int flen);
365 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
366 unsigned int len);
367
368 void sk_filter_charge(struct sock *sk, struct sk_filter *fp);
369 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
370
371 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
372 void bpf_int_jit_compile(struct sk_filter *fp);
373
374 #define BPF_ANC BIT(15)
375
376 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
377 {
378 BUG_ON(ftest->code & BPF_ANC);
379
380 switch (ftest->code) {
381 case BPF_LD | BPF_W | BPF_ABS:
382 case BPF_LD | BPF_H | BPF_ABS:
383 case BPF_LD | BPF_B | BPF_ABS:
384 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
385 return BPF_ANC | SKF_AD_##CODE
386 switch (ftest->k) {
387 BPF_ANCILLARY(PROTOCOL);
388 BPF_ANCILLARY(PKTTYPE);
389 BPF_ANCILLARY(IFINDEX);
390 BPF_ANCILLARY(NLATTR);
391 BPF_ANCILLARY(NLATTR_NEST);
392 BPF_ANCILLARY(MARK);
393 BPF_ANCILLARY(QUEUE);
394 BPF_ANCILLARY(HATYPE);
395 BPF_ANCILLARY(RXHASH);
396 BPF_ANCILLARY(CPU);
397 BPF_ANCILLARY(ALU_XOR_X);
398 BPF_ANCILLARY(VLAN_TAG);
399 BPF_ANCILLARY(VLAN_TAG_PRESENT);
400 BPF_ANCILLARY(PAY_OFFSET);
401 BPF_ANCILLARY(RANDOM);
402 }
403 /* Fallthrough. */
404 default:
405 return ftest->code;
406 }
407 }
408
409 #ifdef CONFIG_BPF_JIT
410 #include <stdarg.h>
411 #include <linux/linkage.h>
412 #include <linux/printk.h>
413
414 void bpf_jit_compile(struct sk_filter *fp);
415 void bpf_jit_free(struct sk_filter *fp);
416
417 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
418 u32 pass, void *image)
419 {
420 pr_err("flen=%u proglen=%u pass=%u image=%pK\n",
421 flen, proglen, pass, image);
422 if (image)
423 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
424 16, 1, image, proglen, false);
425 }
426 #else
427 #include <linux/slab.h>
428
429 static inline void bpf_jit_compile(struct sk_filter *fp)
430 {
431 }
432
433 static inline void bpf_jit_free(struct sk_filter *fp)
434 {
435 kfree(fp);
436 }
437 #endif /* CONFIG_BPF_JIT */
438
439 static inline int bpf_tell_extensions(void)
440 {
441 return SKF_AD_MAX;
442 }
443
444 #endif /* __LINUX_FILTER_H__ */