xref: /linux-6.15/include/linux/filter.h (revision 064223c1)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Linux Socket Filter Data Structures
4  */
5 #ifndef __LINUX_FILTER_H__
6 #define __LINUX_FILTER_H__
7 
8 #include <stdarg.h>
9 
10 #include <linux/atomic.h>
11 #include <linux/refcount.h>
12 #include <linux/compat.h>
13 #include <linux/skbuff.h>
14 #include <linux/linkage.h>
15 #include <linux/printk.h>
16 #include <linux/workqueue.h>
17 #include <linux/sched.h>
18 #include <linux/capability.h>
19 #include <linux/cryptohash.h>
20 #include <linux/set_memory.h>
21 #include <linux/kallsyms.h>
22 
23 #include <net/sch_generic.h>
24 
25 #include <uapi/linux/filter.h>
26 #include <uapi/linux/bpf.h>
27 
28 struct sk_buff;
29 struct sock;
30 struct seccomp_data;
31 struct bpf_prog_aux;
32 struct xdp_rxq_info;
33 struct xdp_buff;
34 
35 /* ArgX, context and stack frame pointer register positions. Note,
36  * Arg1, Arg2, Arg3, etc are used as argument mappings of function
37  * calls in BPF_CALL instruction.
38  */
39 #define BPF_REG_ARG1	BPF_REG_1
40 #define BPF_REG_ARG2	BPF_REG_2
41 #define BPF_REG_ARG3	BPF_REG_3
42 #define BPF_REG_ARG4	BPF_REG_4
43 #define BPF_REG_ARG5	BPF_REG_5
44 #define BPF_REG_CTX	BPF_REG_6
45 #define BPF_REG_FP	BPF_REG_10
46 
47 /* Additional register mappings for converted user programs. */
48 #define BPF_REG_A	BPF_REG_0
49 #define BPF_REG_X	BPF_REG_7
50 #define BPF_REG_TMP	BPF_REG_8
51 
52 /* Kernel hidden auxiliary/helper register for hardening step.
53  * Only used by eBPF JITs. It's nothing more than a temporary
54  * register that JITs use internally, only that here it's part
55  * of eBPF instructions that have been rewritten for blinding
56  * constants. See JIT pre-step in bpf_jit_blind_constants().
57  */
58 #define BPF_REG_AX		MAX_BPF_REG
59 #define MAX_BPF_JIT_REG		(MAX_BPF_REG + 1)
60 
61 /* unused opcode to mark special call to bpf_tail_call() helper */
62 #define BPF_TAIL_CALL	0xf0
63 
64 /* unused opcode to mark call to interpreter with arguments */
65 #define BPF_CALL_ARGS	0xe0
66 
67 /* As per nm, we expose JITed images as text (code) section for
68  * kallsyms. That way, tools like perf can find it to match
69  * addresses.
70  */
71 #define BPF_SYM_ELF_TYPE	't'
72 
73 /* BPF program can access up to 512 bytes of stack space. */
74 #define MAX_BPF_STACK	512
75 
76 /* Helper macros for filter block array initializers. */
77 
78 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
79 
80 #define BPF_ALU64_REG(OP, DST, SRC)				\
81 	((struct bpf_insn) {					\
82 		.code  = BPF_ALU64 | BPF_OP(OP) | BPF_X,	\
83 		.dst_reg = DST,					\
84 		.src_reg = SRC,					\
85 		.off   = 0,					\
86 		.imm   = 0 })
87 
88 #define BPF_ALU32_REG(OP, DST, SRC)				\
89 	((struct bpf_insn) {					\
90 		.code  = BPF_ALU | BPF_OP(OP) | BPF_X,		\
91 		.dst_reg = DST,					\
92 		.src_reg = SRC,					\
93 		.off   = 0,					\
94 		.imm   = 0 })
95 
96 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
97 
98 #define BPF_ALU64_IMM(OP, DST, IMM)				\
99 	((struct bpf_insn) {					\
100 		.code  = BPF_ALU64 | BPF_OP(OP) | BPF_K,	\
101 		.dst_reg = DST,					\
102 		.src_reg = 0,					\
103 		.off   = 0,					\
104 		.imm   = IMM })
105 
106 #define BPF_ALU32_IMM(OP, DST, IMM)				\
107 	((struct bpf_insn) {					\
108 		.code  = BPF_ALU | BPF_OP(OP) | BPF_K,		\
109 		.dst_reg = DST,					\
110 		.src_reg = 0,					\
111 		.off   = 0,					\
112 		.imm   = IMM })
113 
114 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
115 
116 #define BPF_ENDIAN(TYPE, DST, LEN)				\
117 	((struct bpf_insn) {					\
118 		.code  = BPF_ALU | BPF_END | BPF_SRC(TYPE),	\
119 		.dst_reg = DST,					\
120 		.src_reg = 0,					\
121 		.off   = 0,					\
122 		.imm   = LEN })
123 
124 /* Short form of mov, dst_reg = src_reg */
125 
126 #define BPF_MOV64_REG(DST, SRC)					\
127 	((struct bpf_insn) {					\
128 		.code  = BPF_ALU64 | BPF_MOV | BPF_X,		\
129 		.dst_reg = DST,					\
130 		.src_reg = SRC,					\
131 		.off   = 0,					\
132 		.imm   = 0 })
133 
134 #define BPF_MOV32_REG(DST, SRC)					\
135 	((struct bpf_insn) {					\
136 		.code  = BPF_ALU | BPF_MOV | BPF_X,		\
137 		.dst_reg = DST,					\
138 		.src_reg = SRC,					\
139 		.off   = 0,					\
140 		.imm   = 0 })
141 
142 /* Short form of mov, dst_reg = imm32 */
143 
144 #define BPF_MOV64_IMM(DST, IMM)					\
145 	((struct bpf_insn) {					\
146 		.code  = BPF_ALU64 | BPF_MOV | BPF_K,		\
147 		.dst_reg = DST,					\
148 		.src_reg = 0,					\
149 		.off   = 0,					\
150 		.imm   = IMM })
151 
152 #define BPF_MOV32_IMM(DST, IMM)					\
153 	((struct bpf_insn) {					\
154 		.code  = BPF_ALU | BPF_MOV | BPF_K,		\
155 		.dst_reg = DST,					\
156 		.src_reg = 0,					\
157 		.off   = 0,					\
158 		.imm   = IMM })
159 
160 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
161 #define BPF_LD_IMM64(DST, IMM)					\
162 	BPF_LD_IMM64_RAW(DST, 0, IMM)
163 
164 #define BPF_LD_IMM64_RAW(DST, SRC, IMM)				\
165 	((struct bpf_insn) {					\
166 		.code  = BPF_LD | BPF_DW | BPF_IMM,		\
167 		.dst_reg = DST,					\
168 		.src_reg = SRC,					\
169 		.off   = 0,					\
170 		.imm   = (__u32) (IMM) }),			\
171 	((struct bpf_insn) {					\
172 		.code  = 0, /* zero is reserved opcode */	\
173 		.dst_reg = 0,					\
174 		.src_reg = 0,					\
175 		.off   = 0,					\
176 		.imm   = ((__u64) (IMM)) >> 32 })
177 
178 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
179 #define BPF_LD_MAP_FD(DST, MAP_FD)				\
180 	BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
181 
182 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
183 
184 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM)			\
185 	((struct bpf_insn) {					\
186 		.code  = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE),	\
187 		.dst_reg = DST,					\
188 		.src_reg = SRC,					\
189 		.off   = 0,					\
190 		.imm   = IMM })
191 
192 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM)			\
193 	((struct bpf_insn) {					\
194 		.code  = BPF_ALU | BPF_MOV | BPF_SRC(TYPE),	\
195 		.dst_reg = DST,					\
196 		.src_reg = SRC,					\
197 		.off   = 0,					\
198 		.imm   = IMM })
199 
200 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
201 
202 #define BPF_LD_ABS(SIZE, IMM)					\
203 	((struct bpf_insn) {					\
204 		.code  = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS,	\
205 		.dst_reg = 0,					\
206 		.src_reg = 0,					\
207 		.off   = 0,					\
208 		.imm   = IMM })
209 
210 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
211 
212 #define BPF_LD_IND(SIZE, SRC, IMM)				\
213 	((struct bpf_insn) {					\
214 		.code  = BPF_LD | BPF_SIZE(SIZE) | BPF_IND,	\
215 		.dst_reg = 0,					\
216 		.src_reg = SRC,					\
217 		.off   = 0,					\
218 		.imm   = IMM })
219 
220 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
221 
222 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF)			\
223 	((struct bpf_insn) {					\
224 		.code  = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM,	\
225 		.dst_reg = DST,					\
226 		.src_reg = SRC,					\
227 		.off   = OFF,					\
228 		.imm   = 0 })
229 
230 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
231 
232 #define BPF_STX_MEM(SIZE, DST, SRC, OFF)			\
233 	((struct bpf_insn) {					\
234 		.code  = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM,	\
235 		.dst_reg = DST,					\
236 		.src_reg = SRC,					\
237 		.off   = OFF,					\
238 		.imm   = 0 })
239 
240 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
241 
242 #define BPF_STX_XADD(SIZE, DST, SRC, OFF)			\
243 	((struct bpf_insn) {					\
244 		.code  = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD,	\
245 		.dst_reg = DST,					\
246 		.src_reg = SRC,					\
247 		.off   = OFF,					\
248 		.imm   = 0 })
249 
250 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
251 
252 #define BPF_ST_MEM(SIZE, DST, OFF, IMM)				\
253 	((struct bpf_insn) {					\
254 		.code  = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM,	\
255 		.dst_reg = DST,					\
256 		.src_reg = 0,					\
257 		.off   = OFF,					\
258 		.imm   = IMM })
259 
260 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
261 
262 #define BPF_JMP_REG(OP, DST, SRC, OFF)				\
263 	((struct bpf_insn) {					\
264 		.code  = BPF_JMP | BPF_OP(OP) | BPF_X,		\
265 		.dst_reg = DST,					\
266 		.src_reg = SRC,					\
267 		.off   = OFF,					\
268 		.imm   = 0 })
269 
270 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
271 
272 #define BPF_JMP_IMM(OP, DST, IMM, OFF)				\
273 	((struct bpf_insn) {					\
274 		.code  = BPF_JMP | BPF_OP(OP) | BPF_K,		\
275 		.dst_reg = DST,					\
276 		.src_reg = 0,					\
277 		.off   = OFF,					\
278 		.imm   = IMM })
279 
280 /* Unconditional jumps, goto pc + off16 */
281 
282 #define BPF_JMP_A(OFF)						\
283 	((struct bpf_insn) {					\
284 		.code  = BPF_JMP | BPF_JA,			\
285 		.dst_reg = 0,					\
286 		.src_reg = 0,					\
287 		.off   = OFF,					\
288 		.imm   = 0 })
289 
290 /* Function call */
291 
292 #define BPF_EMIT_CALL(FUNC)					\
293 	((struct bpf_insn) {					\
294 		.code  = BPF_JMP | BPF_CALL,			\
295 		.dst_reg = 0,					\
296 		.src_reg = 0,					\
297 		.off   = 0,					\
298 		.imm   = ((FUNC) - __bpf_call_base) })
299 
300 /* Raw code statement block */
301 
302 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM)			\
303 	((struct bpf_insn) {					\
304 		.code  = CODE,					\
305 		.dst_reg = DST,					\
306 		.src_reg = SRC,					\
307 		.off   = OFF,					\
308 		.imm   = IMM })
309 
310 /* Program exit */
311 
312 #define BPF_EXIT_INSN()						\
313 	((struct bpf_insn) {					\
314 		.code  = BPF_JMP | BPF_EXIT,			\
315 		.dst_reg = 0,					\
316 		.src_reg = 0,					\
317 		.off   = 0,					\
318 		.imm   = 0 })
319 
320 /* Internal classic blocks for direct assignment */
321 
322 #define __BPF_STMT(CODE, K)					\
323 	((struct sock_filter) BPF_STMT(CODE, K))
324 
325 #define __BPF_JUMP(CODE, K, JT, JF)				\
326 	((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
327 
328 #define bytes_to_bpf_size(bytes)				\
329 ({								\
330 	int bpf_size = -EINVAL;					\
331 								\
332 	if (bytes == sizeof(u8))				\
333 		bpf_size = BPF_B;				\
334 	else if (bytes == sizeof(u16))				\
335 		bpf_size = BPF_H;				\
336 	else if (bytes == sizeof(u32))				\
337 		bpf_size = BPF_W;				\
338 	else if (bytes == sizeof(u64))				\
339 		bpf_size = BPF_DW;				\
340 								\
341 	bpf_size;						\
342 })
343 
344 #define bpf_size_to_bytes(bpf_size)				\
345 ({								\
346 	int bytes = -EINVAL;					\
347 								\
348 	if (bpf_size == BPF_B)					\
349 		bytes = sizeof(u8);				\
350 	else if (bpf_size == BPF_H)				\
351 		bytes = sizeof(u16);				\
352 	else if (bpf_size == BPF_W)				\
353 		bytes = sizeof(u32);				\
354 	else if (bpf_size == BPF_DW)				\
355 		bytes = sizeof(u64);				\
356 								\
357 	bytes;							\
358 })
359 
360 #define BPF_SIZEOF(type)					\
361 	({							\
362 		const int __size = bytes_to_bpf_size(sizeof(type)); \
363 		BUILD_BUG_ON(__size < 0);			\
364 		__size;						\
365 	})
366 
367 #define BPF_FIELD_SIZEOF(type, field)				\
368 	({							\
369 		const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
370 		BUILD_BUG_ON(__size < 0);			\
371 		__size;						\
372 	})
373 
374 #define BPF_LDST_BYTES(insn)					\
375 	({							\
376 		const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
377 		WARN_ON(__size < 0);				\
378 		__size;						\
379 	})
380 
381 #define __BPF_MAP_0(m, v, ...) v
382 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
383 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
384 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
385 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
386 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
387 
388 #define __BPF_REG_0(...) __BPF_PAD(5)
389 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
390 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
391 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
392 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
393 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
394 
395 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
396 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
397 
398 #define __BPF_CAST(t, a)						       \
399 	(__force t)							       \
400 	(__force							       \
401 	 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long),      \
402 				      (unsigned long)0, (t)0))) a
403 #define __BPF_V void
404 #define __BPF_N
405 
406 #define __BPF_DECL_ARGS(t, a) t   a
407 #define __BPF_DECL_REGS(t, a) u64 a
408 
409 #define __BPF_PAD(n)							       \
410 	__BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2,       \
411 		  u64, __ur_3, u64, __ur_4, u64, __ur_5)
412 
413 #define BPF_CALL_x(x, name, ...)					       \
414 	static __always_inline						       \
415 	u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__));   \
416 	u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__));	       \
417 	u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__))	       \
418 	{								       \
419 		return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
420 	}								       \
421 	static __always_inline						       \
422 	u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
423 
424 #define BPF_CALL_0(name, ...)	BPF_CALL_x(0, name, __VA_ARGS__)
425 #define BPF_CALL_1(name, ...)	BPF_CALL_x(1, name, __VA_ARGS__)
426 #define BPF_CALL_2(name, ...)	BPF_CALL_x(2, name, __VA_ARGS__)
427 #define BPF_CALL_3(name, ...)	BPF_CALL_x(3, name, __VA_ARGS__)
428 #define BPF_CALL_4(name, ...)	BPF_CALL_x(4, name, __VA_ARGS__)
429 #define BPF_CALL_5(name, ...)	BPF_CALL_x(5, name, __VA_ARGS__)
430 
431 #define bpf_ctx_range(TYPE, MEMBER)						\
432 	offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
433 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2)				\
434 	offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
435 
436 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE)				\
437 	({									\
438 		BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE));		\
439 		*(PTR_SIZE) = (SIZE);						\
440 		offsetof(TYPE, MEMBER);						\
441 	})
442 
443 #ifdef CONFIG_COMPAT
444 /* A struct sock_filter is architecture independent. */
445 struct compat_sock_fprog {
446 	u16		len;
447 	compat_uptr_t	filter;	/* struct sock_filter * */
448 };
449 #endif
450 
451 struct sock_fprog_kern {
452 	u16			len;
453 	struct sock_filter	*filter;
454 };
455 
456 struct bpf_binary_header {
457 	unsigned int pages;
458 	u8 image[];
459 };
460 
461 struct bpf_prog {
462 	u16			pages;		/* Number of allocated pages */
463 	u16			jited:1,	/* Is our filter JIT'ed? */
464 				jit_requested:1,/* archs need to JIT the prog */
465 				locked:1,	/* Program image locked? */
466 				gpl_compatible:1, /* Is filter GPL compatible? */
467 				cb_access:1,	/* Is control block accessed? */
468 				dst_needed:1,	/* Do we need dst entry? */
469 				blinded:1,	/* Was blinded */
470 				is_func:1,	/* program is a bpf function */
471 				kprobe_override:1; /* Do we override a kprobe? */
472 	enum bpf_prog_type	type;		/* Type of BPF program */
473 	enum bpf_attach_type	expected_attach_type; /* For some prog types */
474 	u32			len;		/* Number of filter blocks */
475 	u32			jited_len;	/* Size of jited insns in bytes */
476 	u8			tag[BPF_TAG_SIZE];
477 	struct bpf_prog_aux	*aux;		/* Auxiliary fields */
478 	struct sock_fprog_kern	*orig_prog;	/* Original BPF program */
479 	unsigned int		(*bpf_func)(const void *ctx,
480 					    const struct bpf_insn *insn);
481 	/* Instructions for interpreter */
482 	union {
483 		struct sock_filter	insns[0];
484 		struct bpf_insn		insnsi[0];
485 	};
486 };
487 
488 struct sk_filter {
489 	refcount_t	refcnt;
490 	struct rcu_head	rcu;
491 	struct bpf_prog	*prog;
492 };
493 
494 #define BPF_PROG_RUN(filter, ctx)  (*(filter)->bpf_func)(ctx, (filter)->insnsi)
495 
496 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
497 
498 struct bpf_skb_data_end {
499 	struct qdisc_skb_cb qdisc_cb;
500 	void *data_meta;
501 	void *data_end;
502 };
503 
504 struct sk_msg_buff {
505 	void *data;
506 	void *data_end;
507 	__u32 apply_bytes;
508 	__u32 cork_bytes;
509 	int sg_copybreak;
510 	int sg_start;
511 	int sg_curr;
512 	int sg_end;
513 	struct scatterlist sg_data[MAX_SKB_FRAGS];
514 	bool sg_copy[MAX_SKB_FRAGS];
515 	__u32 key;
516 	__u32 flags;
517 	struct bpf_map *map;
518 	struct sk_buff *skb;
519 	struct list_head list;
520 };
521 
522 /* Compute the linear packet data range [data, data_end) which
523  * will be accessed by various program types (cls_bpf, act_bpf,
524  * lwt, ...). Subsystems allowing direct data access must (!)
525  * ensure that cb[] area can be written to when BPF program is
526  * invoked (otherwise cb[] save/restore is necessary).
527  */
528 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
529 {
530 	struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
531 
532 	BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
533 	cb->data_meta = skb->data - skb_metadata_len(skb);
534 	cb->data_end  = skb->data + skb_headlen(skb);
535 }
536 
537 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
538 {
539 	/* eBPF programs may read/write skb->cb[] area to transfer meta
540 	 * data between tail calls. Since this also needs to work with
541 	 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
542 	 *
543 	 * In some socket filter cases, the cb unfortunately needs to be
544 	 * saved/restored so that protocol specific skb->cb[] data won't
545 	 * be lost. In any case, due to unpriviledged eBPF programs
546 	 * attached to sockets, we need to clear the bpf_skb_cb() area
547 	 * to not leak previous contents to user space.
548 	 */
549 	BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
550 	BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
551 		     FIELD_SIZEOF(struct qdisc_skb_cb, data));
552 
553 	return qdisc_skb_cb(skb)->data;
554 }
555 
556 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
557 				       struct sk_buff *skb)
558 {
559 	u8 *cb_data = bpf_skb_cb(skb);
560 	u8 cb_saved[BPF_SKB_CB_LEN];
561 	u32 res;
562 
563 	if (unlikely(prog->cb_access)) {
564 		memcpy(cb_saved, cb_data, sizeof(cb_saved));
565 		memset(cb_data, 0, sizeof(cb_saved));
566 	}
567 
568 	res = BPF_PROG_RUN(prog, skb);
569 
570 	if (unlikely(prog->cb_access))
571 		memcpy(cb_data, cb_saved, sizeof(cb_saved));
572 
573 	return res;
574 }
575 
576 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
577 					struct sk_buff *skb)
578 {
579 	u8 *cb_data = bpf_skb_cb(skb);
580 
581 	if (unlikely(prog->cb_access))
582 		memset(cb_data, 0, BPF_SKB_CB_LEN);
583 
584 	return BPF_PROG_RUN(prog, skb);
585 }
586 
587 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
588 					    struct xdp_buff *xdp)
589 {
590 	/* Caller needs to hold rcu_read_lock() (!), otherwise program
591 	 * can be released while still running, or map elements could be
592 	 * freed early while still having concurrent users. XDP fastpath
593 	 * already takes rcu_read_lock() when fetching the program, so
594 	 * it's not necessary here anymore.
595 	 */
596 	return BPF_PROG_RUN(prog, xdp);
597 }
598 
599 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
600 {
601 	return prog->len * sizeof(struct bpf_insn);
602 }
603 
604 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
605 {
606 	return round_up(bpf_prog_insn_size(prog) +
607 			sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
608 }
609 
610 static inline unsigned int bpf_prog_size(unsigned int proglen)
611 {
612 	return max(sizeof(struct bpf_prog),
613 		   offsetof(struct bpf_prog, insns[proglen]));
614 }
615 
616 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
617 {
618 	/* When classic BPF programs have been loaded and the arch
619 	 * does not have a classic BPF JIT (anymore), they have been
620 	 * converted via bpf_migrate_filter() to eBPF and thus always
621 	 * have an unspec program type.
622 	 */
623 	return prog->type == BPF_PROG_TYPE_UNSPEC;
624 }
625 
626 static inline bool
627 bpf_ctx_narrow_access_ok(u32 off, u32 size, const u32 size_default)
628 {
629 	bool off_ok;
630 #ifdef __LITTLE_ENDIAN
631 	off_ok = (off & (size_default - 1)) == 0;
632 #else
633 	off_ok = (off & (size_default - 1)) + size == size_default;
634 #endif
635 	return off_ok && size <= size_default && (size & (size - 1)) == 0;
636 }
637 
638 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
639 
640 #ifdef CONFIG_ARCH_HAS_SET_MEMORY
641 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
642 {
643 	fp->locked = 1;
644 	WARN_ON_ONCE(set_memory_ro((unsigned long)fp, fp->pages));
645 }
646 
647 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
648 {
649 	if (fp->locked) {
650 		WARN_ON_ONCE(set_memory_rw((unsigned long)fp, fp->pages));
651 		/* In case set_memory_rw() fails, we want to be the first
652 		 * to crash here instead of some random place later on.
653 		 */
654 		fp->locked = 0;
655 	}
656 }
657 
658 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
659 {
660 	WARN_ON_ONCE(set_memory_ro((unsigned long)hdr, hdr->pages));
661 }
662 
663 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
664 {
665 	WARN_ON_ONCE(set_memory_rw((unsigned long)hdr, hdr->pages));
666 }
667 #else
668 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
669 {
670 }
671 
672 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
673 {
674 }
675 
676 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
677 {
678 }
679 
680 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
681 {
682 }
683 #endif /* CONFIG_ARCH_HAS_SET_MEMORY */
684 
685 static inline struct bpf_binary_header *
686 bpf_jit_binary_hdr(const struct bpf_prog *fp)
687 {
688 	unsigned long real_start = (unsigned long)fp->bpf_func;
689 	unsigned long addr = real_start & PAGE_MASK;
690 
691 	return (void *)addr;
692 }
693 
694 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
695 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
696 {
697 	return sk_filter_trim_cap(sk, skb, 1);
698 }
699 
700 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
701 void bpf_prog_free(struct bpf_prog *fp);
702 
703 bool bpf_opcode_in_insntable(u8 code);
704 
705 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
706 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
707 				  gfp_t gfp_extra_flags);
708 void __bpf_prog_free(struct bpf_prog *fp);
709 
710 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
711 {
712 	bpf_prog_unlock_ro(fp);
713 	__bpf_prog_free(fp);
714 }
715 
716 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
717 				       unsigned int flen);
718 
719 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
720 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
721 			      bpf_aux_classic_check_t trans, bool save_orig);
722 void bpf_prog_destroy(struct bpf_prog *fp);
723 
724 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
725 int sk_attach_bpf(u32 ufd, struct sock *sk);
726 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
727 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
728 int sk_detach_filter(struct sock *sk);
729 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
730 		  unsigned int len);
731 
732 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
733 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
734 
735 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
736 #define __bpf_call_base_args \
737 	((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
738 	 __bpf_call_base)
739 
740 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
741 void bpf_jit_compile(struct bpf_prog *prog);
742 bool bpf_helper_changes_pkt_data(void *func);
743 
744 static inline bool bpf_dump_raw_ok(void)
745 {
746 	/* Reconstruction of call-sites is dependent on kallsyms,
747 	 * thus make dump the same restriction.
748 	 */
749 	return kallsyms_show_value() == 1;
750 }
751 
752 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
753 				       const struct bpf_insn *patch, u32 len);
754 
755 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
756  * same cpu context. Further for best results no more than a single map
757  * for the do_redirect/do_flush pair should be used. This limitation is
758  * because we only track one map and force a flush when the map changes.
759  * This does not appear to be a real limitation for existing software.
760  */
761 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
762 			    struct bpf_prog *prog);
763 int xdp_do_redirect(struct net_device *dev,
764 		    struct xdp_buff *xdp,
765 		    struct bpf_prog *prog);
766 void xdp_do_flush_map(void);
767 
768 void bpf_warn_invalid_xdp_action(u32 act);
769 
770 struct sock *do_sk_redirect_map(struct sk_buff *skb);
771 struct sock *do_msg_redirect_map(struct sk_msg_buff *md);
772 
773 #ifdef CONFIG_BPF_JIT
774 extern int bpf_jit_enable;
775 extern int bpf_jit_harden;
776 extern int bpf_jit_kallsyms;
777 
778 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
779 
780 struct bpf_binary_header *
781 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
782 		     unsigned int alignment,
783 		     bpf_jit_fill_hole_t bpf_fill_ill_insns);
784 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
785 
786 void bpf_jit_free(struct bpf_prog *fp);
787 
788 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
789 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
790 
791 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
792 				u32 pass, void *image)
793 {
794 	pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
795 	       proglen, pass, image, current->comm, task_pid_nr(current));
796 
797 	if (image)
798 		print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
799 			       16, 1, image, proglen, false);
800 }
801 
802 static inline bool bpf_jit_is_ebpf(void)
803 {
804 # ifdef CONFIG_HAVE_EBPF_JIT
805 	return true;
806 # else
807 	return false;
808 # endif
809 }
810 
811 static inline bool ebpf_jit_enabled(void)
812 {
813 	return bpf_jit_enable && bpf_jit_is_ebpf();
814 }
815 
816 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
817 {
818 	return fp->jited && bpf_jit_is_ebpf();
819 }
820 
821 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
822 {
823 	/* These are the prerequisites, should someone ever have the
824 	 * idea to call blinding outside of them, we make sure to
825 	 * bail out.
826 	 */
827 	if (!bpf_jit_is_ebpf())
828 		return false;
829 	if (!prog->jit_requested)
830 		return false;
831 	if (!bpf_jit_harden)
832 		return false;
833 	if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
834 		return false;
835 
836 	return true;
837 }
838 
839 static inline bool bpf_jit_kallsyms_enabled(void)
840 {
841 	/* There are a couple of corner cases where kallsyms should
842 	 * not be enabled f.e. on hardening.
843 	 */
844 	if (bpf_jit_harden)
845 		return false;
846 	if (!bpf_jit_kallsyms)
847 		return false;
848 	if (bpf_jit_kallsyms == 1)
849 		return true;
850 
851 	return false;
852 }
853 
854 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
855 				 unsigned long *off, char *sym);
856 bool is_bpf_text_address(unsigned long addr);
857 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
858 		    char *sym);
859 
860 static inline const char *
861 bpf_address_lookup(unsigned long addr, unsigned long *size,
862 		   unsigned long *off, char **modname, char *sym)
863 {
864 	const char *ret = __bpf_address_lookup(addr, size, off, sym);
865 
866 	if (ret && modname)
867 		*modname = NULL;
868 	return ret;
869 }
870 
871 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
872 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
873 
874 #else /* CONFIG_BPF_JIT */
875 
876 static inline bool ebpf_jit_enabled(void)
877 {
878 	return false;
879 }
880 
881 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
882 {
883 	return false;
884 }
885 
886 static inline void bpf_jit_free(struct bpf_prog *fp)
887 {
888 	bpf_prog_unlock_free(fp);
889 }
890 
891 static inline bool bpf_jit_kallsyms_enabled(void)
892 {
893 	return false;
894 }
895 
896 static inline const char *
897 __bpf_address_lookup(unsigned long addr, unsigned long *size,
898 		     unsigned long *off, char *sym)
899 {
900 	return NULL;
901 }
902 
903 static inline bool is_bpf_text_address(unsigned long addr)
904 {
905 	return false;
906 }
907 
908 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
909 				  char *type, char *sym)
910 {
911 	return -ERANGE;
912 }
913 
914 static inline const char *
915 bpf_address_lookup(unsigned long addr, unsigned long *size,
916 		   unsigned long *off, char **modname, char *sym)
917 {
918 	return NULL;
919 }
920 
921 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
922 {
923 }
924 
925 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
926 {
927 }
928 #endif /* CONFIG_BPF_JIT */
929 
930 #define BPF_ANC		BIT(15)
931 
932 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
933 {
934 	switch (first->code) {
935 	case BPF_RET | BPF_K:
936 	case BPF_LD | BPF_W | BPF_LEN:
937 		return false;
938 
939 	case BPF_LD | BPF_W | BPF_ABS:
940 	case BPF_LD | BPF_H | BPF_ABS:
941 	case BPF_LD | BPF_B | BPF_ABS:
942 		if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
943 			return true;
944 		return false;
945 
946 	default:
947 		return true;
948 	}
949 }
950 
951 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
952 {
953 	BUG_ON(ftest->code & BPF_ANC);
954 
955 	switch (ftest->code) {
956 	case BPF_LD | BPF_W | BPF_ABS:
957 	case BPF_LD | BPF_H | BPF_ABS:
958 	case BPF_LD | BPF_B | BPF_ABS:
959 #define BPF_ANCILLARY(CODE)	case SKF_AD_OFF + SKF_AD_##CODE:	\
960 				return BPF_ANC | SKF_AD_##CODE
961 		switch (ftest->k) {
962 		BPF_ANCILLARY(PROTOCOL);
963 		BPF_ANCILLARY(PKTTYPE);
964 		BPF_ANCILLARY(IFINDEX);
965 		BPF_ANCILLARY(NLATTR);
966 		BPF_ANCILLARY(NLATTR_NEST);
967 		BPF_ANCILLARY(MARK);
968 		BPF_ANCILLARY(QUEUE);
969 		BPF_ANCILLARY(HATYPE);
970 		BPF_ANCILLARY(RXHASH);
971 		BPF_ANCILLARY(CPU);
972 		BPF_ANCILLARY(ALU_XOR_X);
973 		BPF_ANCILLARY(VLAN_TAG);
974 		BPF_ANCILLARY(VLAN_TAG_PRESENT);
975 		BPF_ANCILLARY(PAY_OFFSET);
976 		BPF_ANCILLARY(RANDOM);
977 		BPF_ANCILLARY(VLAN_TPID);
978 		}
979 		/* Fallthrough. */
980 	default:
981 		return ftest->code;
982 	}
983 }
984 
985 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
986 					   int k, unsigned int size);
987 
988 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
989 				     unsigned int size, void *buffer)
990 {
991 	if (k >= 0)
992 		return skb_header_pointer(skb, k, size, buffer);
993 
994 	return bpf_internal_load_pointer_neg_helper(skb, k, size);
995 }
996 
997 static inline int bpf_tell_extensions(void)
998 {
999 	return SKF_AD_MAX;
1000 }
1001 
1002 struct bpf_sock_addr_kern {
1003 	struct sock *sk;
1004 	struct sockaddr *uaddr;
1005 	/* Temporary "register" to make indirect stores to nested structures
1006 	 * defined above. We need three registers to make such a store, but
1007 	 * only two (src and dst) are available at convert_ctx_access time
1008 	 */
1009 	u64 tmp_reg;
1010 };
1011 
1012 struct bpf_sock_ops_kern {
1013 	struct	sock *sk;
1014 	u32	op;
1015 	union {
1016 		u32 args[4];
1017 		u32 reply;
1018 		u32 replylong[4];
1019 	};
1020 	u32	is_fullsock;
1021 	u64	temp;			/* temp and everything after is not
1022 					 * initialized to 0 before calling
1023 					 * the BPF program. New fields that
1024 					 * should be initialized to 0 should
1025 					 * be inserted before temp.
1026 					 * temp is scratch storage used by
1027 					 * sock_ops_convert_ctx_access
1028 					 * as temporary storage of a register.
1029 					 */
1030 };
1031 
1032 #endif /* __LINUX_FILTER_H__ */
1033