xref: /linux-6.15/include/linux/filter.h (revision cd65cd95)
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_2	/* scratch reg */
51 #define BPF_REG_D	BPF_REG_8	/* data, callee-saved */
52 #define BPF_REG_H	BPF_REG_9	/* hlen, callee-saved */
53 
54 /* Kernel hidden auxiliary/helper register for hardening step.
55  * Only used by eBPF JITs. It's nothing more than a temporary
56  * register that JITs use internally, only that here it's part
57  * of eBPF instructions that have been rewritten for blinding
58  * constants. See JIT pre-step in bpf_jit_blind_constants().
59  */
60 #define BPF_REG_AX		MAX_BPF_REG
61 #define MAX_BPF_JIT_REG		(MAX_BPF_REG + 1)
62 
63 /* unused opcode to mark special call to bpf_tail_call() helper */
64 #define BPF_TAIL_CALL	0xf0
65 
66 /* unused opcode to mark call to interpreter with arguments */
67 #define BPF_CALL_ARGS	0xe0
68 
69 /* As per nm, we expose JITed images as text (code) section for
70  * kallsyms. That way, tools like perf can find it to match
71  * addresses.
72  */
73 #define BPF_SYM_ELF_TYPE	't'
74 
75 /* BPF program can access up to 512 bytes of stack space. */
76 #define MAX_BPF_STACK	512
77 
78 /* Helper macros for filter block array initializers. */
79 
80 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
81 
82 #define BPF_ALU64_REG(OP, DST, SRC)				\
83 	((struct bpf_insn) {					\
84 		.code  = BPF_ALU64 | BPF_OP(OP) | BPF_X,	\
85 		.dst_reg = DST,					\
86 		.src_reg = SRC,					\
87 		.off   = 0,					\
88 		.imm   = 0 })
89 
90 #define BPF_ALU32_REG(OP, DST, SRC)				\
91 	((struct bpf_insn) {					\
92 		.code  = BPF_ALU | BPF_OP(OP) | BPF_X,		\
93 		.dst_reg = DST,					\
94 		.src_reg = SRC,					\
95 		.off   = 0,					\
96 		.imm   = 0 })
97 
98 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
99 
100 #define BPF_ALU64_IMM(OP, DST, IMM)				\
101 	((struct bpf_insn) {					\
102 		.code  = BPF_ALU64 | BPF_OP(OP) | BPF_K,	\
103 		.dst_reg = DST,					\
104 		.src_reg = 0,					\
105 		.off   = 0,					\
106 		.imm   = IMM })
107 
108 #define BPF_ALU32_IMM(OP, DST, IMM)				\
109 	((struct bpf_insn) {					\
110 		.code  = BPF_ALU | BPF_OP(OP) | BPF_K,		\
111 		.dst_reg = DST,					\
112 		.src_reg = 0,					\
113 		.off   = 0,					\
114 		.imm   = IMM })
115 
116 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
117 
118 #define BPF_ENDIAN(TYPE, DST, LEN)				\
119 	((struct bpf_insn) {					\
120 		.code  = BPF_ALU | BPF_END | BPF_SRC(TYPE),	\
121 		.dst_reg = DST,					\
122 		.src_reg = 0,					\
123 		.off   = 0,					\
124 		.imm   = LEN })
125 
126 /* Short form of mov, dst_reg = src_reg */
127 
128 #define BPF_MOV64_REG(DST, SRC)					\
129 	((struct bpf_insn) {					\
130 		.code  = BPF_ALU64 | BPF_MOV | BPF_X,		\
131 		.dst_reg = DST,					\
132 		.src_reg = SRC,					\
133 		.off   = 0,					\
134 		.imm   = 0 })
135 
136 #define BPF_MOV32_REG(DST, SRC)					\
137 	((struct bpf_insn) {					\
138 		.code  = BPF_ALU | BPF_MOV | BPF_X,		\
139 		.dst_reg = DST,					\
140 		.src_reg = SRC,					\
141 		.off   = 0,					\
142 		.imm   = 0 })
143 
144 /* Short form of mov, dst_reg = imm32 */
145 
146 #define BPF_MOV64_IMM(DST, IMM)					\
147 	((struct bpf_insn) {					\
148 		.code  = BPF_ALU64 | BPF_MOV | BPF_K,		\
149 		.dst_reg = DST,					\
150 		.src_reg = 0,					\
151 		.off   = 0,					\
152 		.imm   = IMM })
153 
154 #define BPF_MOV32_IMM(DST, IMM)					\
155 	((struct bpf_insn) {					\
156 		.code  = BPF_ALU | BPF_MOV | BPF_K,		\
157 		.dst_reg = DST,					\
158 		.src_reg = 0,					\
159 		.off   = 0,					\
160 		.imm   = IMM })
161 
162 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
163 #define BPF_LD_IMM64(DST, IMM)					\
164 	BPF_LD_IMM64_RAW(DST, 0, IMM)
165 
166 #define BPF_LD_IMM64_RAW(DST, SRC, IMM)				\
167 	((struct bpf_insn) {					\
168 		.code  = BPF_LD | BPF_DW | BPF_IMM,		\
169 		.dst_reg = DST,					\
170 		.src_reg = SRC,					\
171 		.off   = 0,					\
172 		.imm   = (__u32) (IMM) }),			\
173 	((struct bpf_insn) {					\
174 		.code  = 0, /* zero is reserved opcode */	\
175 		.dst_reg = 0,					\
176 		.src_reg = 0,					\
177 		.off   = 0,					\
178 		.imm   = ((__u64) (IMM)) >> 32 })
179 
180 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
181 #define BPF_LD_MAP_FD(DST, MAP_FD)				\
182 	BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
183 
184 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
185 
186 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM)			\
187 	((struct bpf_insn) {					\
188 		.code  = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE),	\
189 		.dst_reg = DST,					\
190 		.src_reg = SRC,					\
191 		.off   = 0,					\
192 		.imm   = IMM })
193 
194 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM)			\
195 	((struct bpf_insn) {					\
196 		.code  = BPF_ALU | BPF_MOV | BPF_SRC(TYPE),	\
197 		.dst_reg = DST,					\
198 		.src_reg = SRC,					\
199 		.off   = 0,					\
200 		.imm   = IMM })
201 
202 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
203 
204 #define BPF_LD_ABS(SIZE, IMM)					\
205 	((struct bpf_insn) {					\
206 		.code  = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS,	\
207 		.dst_reg = 0,					\
208 		.src_reg = 0,					\
209 		.off   = 0,					\
210 		.imm   = IMM })
211 
212 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
213 
214 #define BPF_LD_IND(SIZE, SRC, IMM)				\
215 	((struct bpf_insn) {					\
216 		.code  = BPF_LD | BPF_SIZE(SIZE) | BPF_IND,	\
217 		.dst_reg = 0,					\
218 		.src_reg = SRC,					\
219 		.off   = 0,					\
220 		.imm   = IMM })
221 
222 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
223 
224 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF)			\
225 	((struct bpf_insn) {					\
226 		.code  = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM,	\
227 		.dst_reg = DST,					\
228 		.src_reg = SRC,					\
229 		.off   = OFF,					\
230 		.imm   = 0 })
231 
232 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
233 
234 #define BPF_STX_MEM(SIZE, DST, SRC, OFF)			\
235 	((struct bpf_insn) {					\
236 		.code  = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM,	\
237 		.dst_reg = DST,					\
238 		.src_reg = SRC,					\
239 		.off   = OFF,					\
240 		.imm   = 0 })
241 
242 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
243 
244 #define BPF_STX_XADD(SIZE, DST, SRC, OFF)			\
245 	((struct bpf_insn) {					\
246 		.code  = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD,	\
247 		.dst_reg = DST,					\
248 		.src_reg = SRC,					\
249 		.off   = OFF,					\
250 		.imm   = 0 })
251 
252 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
253 
254 #define BPF_ST_MEM(SIZE, DST, OFF, IMM)				\
255 	((struct bpf_insn) {					\
256 		.code  = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM,	\
257 		.dst_reg = DST,					\
258 		.src_reg = 0,					\
259 		.off   = OFF,					\
260 		.imm   = IMM })
261 
262 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
263 
264 #define BPF_JMP_REG(OP, DST, SRC, OFF)				\
265 	((struct bpf_insn) {					\
266 		.code  = BPF_JMP | BPF_OP(OP) | BPF_X,		\
267 		.dst_reg = DST,					\
268 		.src_reg = SRC,					\
269 		.off   = OFF,					\
270 		.imm   = 0 })
271 
272 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
273 
274 #define BPF_JMP_IMM(OP, DST, IMM, OFF)				\
275 	((struct bpf_insn) {					\
276 		.code  = BPF_JMP | BPF_OP(OP) | BPF_K,		\
277 		.dst_reg = DST,					\
278 		.src_reg = 0,					\
279 		.off   = OFF,					\
280 		.imm   = IMM })
281 
282 /* Unconditional jumps, goto pc + off16 */
283 
284 #define BPF_JMP_A(OFF)						\
285 	((struct bpf_insn) {					\
286 		.code  = BPF_JMP | BPF_JA,			\
287 		.dst_reg = 0,					\
288 		.src_reg = 0,					\
289 		.off   = OFF,					\
290 		.imm   = 0 })
291 
292 /* Function call */
293 
294 #define BPF_EMIT_CALL(FUNC)					\
295 	((struct bpf_insn) {					\
296 		.code  = BPF_JMP | BPF_CALL,			\
297 		.dst_reg = 0,					\
298 		.src_reg = 0,					\
299 		.off   = 0,					\
300 		.imm   = ((FUNC) - __bpf_call_base) })
301 
302 /* Raw code statement block */
303 
304 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM)			\
305 	((struct bpf_insn) {					\
306 		.code  = CODE,					\
307 		.dst_reg = DST,					\
308 		.src_reg = SRC,					\
309 		.off   = OFF,					\
310 		.imm   = IMM })
311 
312 /* Program exit */
313 
314 #define BPF_EXIT_INSN()						\
315 	((struct bpf_insn) {					\
316 		.code  = BPF_JMP | BPF_EXIT,			\
317 		.dst_reg = 0,					\
318 		.src_reg = 0,					\
319 		.off   = 0,					\
320 		.imm   = 0 })
321 
322 /* Internal classic blocks for direct assignment */
323 
324 #define __BPF_STMT(CODE, K)					\
325 	((struct sock_filter) BPF_STMT(CODE, K))
326 
327 #define __BPF_JUMP(CODE, K, JT, JF)				\
328 	((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
329 
330 #define bytes_to_bpf_size(bytes)				\
331 ({								\
332 	int bpf_size = -EINVAL;					\
333 								\
334 	if (bytes == sizeof(u8))				\
335 		bpf_size = BPF_B;				\
336 	else if (bytes == sizeof(u16))				\
337 		bpf_size = BPF_H;				\
338 	else if (bytes == sizeof(u32))				\
339 		bpf_size = BPF_W;				\
340 	else if (bytes == sizeof(u64))				\
341 		bpf_size = BPF_DW;				\
342 								\
343 	bpf_size;						\
344 })
345 
346 #define bpf_size_to_bytes(bpf_size)				\
347 ({								\
348 	int bytes = -EINVAL;					\
349 								\
350 	if (bpf_size == BPF_B)					\
351 		bytes = sizeof(u8);				\
352 	else if (bpf_size == BPF_H)				\
353 		bytes = sizeof(u16);				\
354 	else if (bpf_size == BPF_W)				\
355 		bytes = sizeof(u32);				\
356 	else if (bpf_size == BPF_DW)				\
357 		bytes = sizeof(u64);				\
358 								\
359 	bytes;							\
360 })
361 
362 #define BPF_SIZEOF(type)					\
363 	({							\
364 		const int __size = bytes_to_bpf_size(sizeof(type)); \
365 		BUILD_BUG_ON(__size < 0);			\
366 		__size;						\
367 	})
368 
369 #define BPF_FIELD_SIZEOF(type, field)				\
370 	({							\
371 		const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
372 		BUILD_BUG_ON(__size < 0);			\
373 		__size;						\
374 	})
375 
376 #define BPF_LDST_BYTES(insn)					\
377 	({							\
378 		const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
379 		WARN_ON(__size < 0);				\
380 		__size;						\
381 	})
382 
383 #define __BPF_MAP_0(m, v, ...) v
384 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
385 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
386 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
387 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
388 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
389 
390 #define __BPF_REG_0(...) __BPF_PAD(5)
391 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
392 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
393 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
394 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
395 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
396 
397 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
398 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
399 
400 #define __BPF_CAST(t, a)						       \
401 	(__force t)							       \
402 	(__force							       \
403 	 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long),      \
404 				      (unsigned long)0, (t)0))) a
405 #define __BPF_V void
406 #define __BPF_N
407 
408 #define __BPF_DECL_ARGS(t, a) t   a
409 #define __BPF_DECL_REGS(t, a) u64 a
410 
411 #define __BPF_PAD(n)							       \
412 	__BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2,       \
413 		  u64, __ur_3, u64, __ur_4, u64, __ur_5)
414 
415 #define BPF_CALL_x(x, name, ...)					       \
416 	static __always_inline						       \
417 	u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__));   \
418 	u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__));	       \
419 	u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__))	       \
420 	{								       \
421 		return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
422 	}								       \
423 	static __always_inline						       \
424 	u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
425 
426 #define BPF_CALL_0(name, ...)	BPF_CALL_x(0, name, __VA_ARGS__)
427 #define BPF_CALL_1(name, ...)	BPF_CALL_x(1, name, __VA_ARGS__)
428 #define BPF_CALL_2(name, ...)	BPF_CALL_x(2, name, __VA_ARGS__)
429 #define BPF_CALL_3(name, ...)	BPF_CALL_x(3, name, __VA_ARGS__)
430 #define BPF_CALL_4(name, ...)	BPF_CALL_x(4, name, __VA_ARGS__)
431 #define BPF_CALL_5(name, ...)	BPF_CALL_x(5, name, __VA_ARGS__)
432 
433 #define bpf_ctx_range(TYPE, MEMBER)						\
434 	offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
435 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2)				\
436 	offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
437 
438 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE)				\
439 	({									\
440 		BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE));		\
441 		*(PTR_SIZE) = (SIZE);						\
442 		offsetof(TYPE, MEMBER);						\
443 	})
444 
445 #ifdef CONFIG_COMPAT
446 /* A struct sock_filter is architecture independent. */
447 struct compat_sock_fprog {
448 	u16		len;
449 	compat_uptr_t	filter;	/* struct sock_filter * */
450 };
451 #endif
452 
453 struct sock_fprog_kern {
454 	u16			len;
455 	struct sock_filter	*filter;
456 };
457 
458 struct bpf_binary_header {
459 	unsigned int pages;
460 	u8 image[];
461 };
462 
463 struct bpf_prog {
464 	u16			pages;		/* Number of allocated pages */
465 	u16			jited:1,	/* Is our filter JIT'ed? */
466 				jit_requested:1,/* archs need to JIT the prog */
467 				locked:1,	/* Program image locked? */
468 				gpl_compatible:1, /* Is filter GPL compatible? */
469 				cb_access:1,	/* Is control block accessed? */
470 				dst_needed:1,	/* Do we need dst entry? */
471 				blinded:1,	/* Was blinded */
472 				is_func:1,	/* program is a bpf function */
473 				kprobe_override:1, /* Do we override a kprobe? */
474 				has_callchain_buf:1; /* callchain buffer allocated? */
475 	enum bpf_prog_type	type;		/* Type of BPF program */
476 	enum bpf_attach_type	expected_attach_type; /* For some prog types */
477 	u32			len;		/* Number of filter blocks */
478 	u32			jited_len;	/* Size of jited insns in bytes */
479 	u8			tag[BPF_TAG_SIZE];
480 	struct bpf_prog_aux	*aux;		/* Auxiliary fields */
481 	struct sock_fprog_kern	*orig_prog;	/* Original BPF program */
482 	unsigned int		(*bpf_func)(const void *ctx,
483 					    const struct bpf_insn *insn);
484 	/* Instructions for interpreter */
485 	union {
486 		struct sock_filter	insns[0];
487 		struct bpf_insn		insnsi[0];
488 	};
489 };
490 
491 struct sk_filter {
492 	refcount_t	refcnt;
493 	struct rcu_head	rcu;
494 	struct bpf_prog	*prog;
495 };
496 
497 #define BPF_PROG_RUN(filter, ctx)  (*(filter)->bpf_func)(ctx, (filter)->insnsi)
498 
499 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
500 
501 struct bpf_skb_data_end {
502 	struct qdisc_skb_cb qdisc_cb;
503 	void *data_meta;
504 	void *data_end;
505 };
506 
507 struct sk_msg_buff {
508 	void *data;
509 	void *data_end;
510 	__u32 apply_bytes;
511 	__u32 cork_bytes;
512 	int sg_copybreak;
513 	int sg_start;
514 	int sg_curr;
515 	int sg_end;
516 	struct scatterlist sg_data[MAX_SKB_FRAGS];
517 	bool sg_copy[MAX_SKB_FRAGS];
518 	__u32 key;
519 	__u32 flags;
520 	struct bpf_map *map;
521 	struct sk_buff *skb;
522 	struct list_head list;
523 };
524 
525 /* Compute the linear packet data range [data, data_end) which
526  * will be accessed by various program types (cls_bpf, act_bpf,
527  * lwt, ...). Subsystems allowing direct data access must (!)
528  * ensure that cb[] area can be written to when BPF program is
529  * invoked (otherwise cb[] save/restore is necessary).
530  */
531 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
532 {
533 	struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
534 
535 	BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
536 	cb->data_meta = skb->data - skb_metadata_len(skb);
537 	cb->data_end  = skb->data + skb_headlen(skb);
538 }
539 
540 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
541 {
542 	/* eBPF programs may read/write skb->cb[] area to transfer meta
543 	 * data between tail calls. Since this also needs to work with
544 	 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
545 	 *
546 	 * In some socket filter cases, the cb unfortunately needs to be
547 	 * saved/restored so that protocol specific skb->cb[] data won't
548 	 * be lost. In any case, due to unpriviledged eBPF programs
549 	 * attached to sockets, we need to clear the bpf_skb_cb() area
550 	 * to not leak previous contents to user space.
551 	 */
552 	BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
553 	BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
554 		     FIELD_SIZEOF(struct qdisc_skb_cb, data));
555 
556 	return qdisc_skb_cb(skb)->data;
557 }
558 
559 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
560 				       struct sk_buff *skb)
561 {
562 	u8 *cb_data = bpf_skb_cb(skb);
563 	u8 cb_saved[BPF_SKB_CB_LEN];
564 	u32 res;
565 
566 	if (unlikely(prog->cb_access)) {
567 		memcpy(cb_saved, cb_data, sizeof(cb_saved));
568 		memset(cb_data, 0, sizeof(cb_saved));
569 	}
570 
571 	res = BPF_PROG_RUN(prog, skb);
572 
573 	if (unlikely(prog->cb_access))
574 		memcpy(cb_data, cb_saved, sizeof(cb_saved));
575 
576 	return res;
577 }
578 
579 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
580 					struct sk_buff *skb)
581 {
582 	u8 *cb_data = bpf_skb_cb(skb);
583 
584 	if (unlikely(prog->cb_access))
585 		memset(cb_data, 0, BPF_SKB_CB_LEN);
586 
587 	return BPF_PROG_RUN(prog, skb);
588 }
589 
590 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
591 					    struct xdp_buff *xdp)
592 {
593 	/* Caller needs to hold rcu_read_lock() (!), otherwise program
594 	 * can be released while still running, or map elements could be
595 	 * freed early while still having concurrent users. XDP fastpath
596 	 * already takes rcu_read_lock() when fetching the program, so
597 	 * it's not necessary here anymore.
598 	 */
599 	return BPF_PROG_RUN(prog, xdp);
600 }
601 
602 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
603 {
604 	return prog->len * sizeof(struct bpf_insn);
605 }
606 
607 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
608 {
609 	return round_up(bpf_prog_insn_size(prog) +
610 			sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
611 }
612 
613 static inline unsigned int bpf_prog_size(unsigned int proglen)
614 {
615 	return max(sizeof(struct bpf_prog),
616 		   offsetof(struct bpf_prog, insns[proglen]));
617 }
618 
619 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
620 {
621 	/* When classic BPF programs have been loaded and the arch
622 	 * does not have a classic BPF JIT (anymore), they have been
623 	 * converted via bpf_migrate_filter() to eBPF and thus always
624 	 * have an unspec program type.
625 	 */
626 	return prog->type == BPF_PROG_TYPE_UNSPEC;
627 }
628 
629 static inline bool
630 bpf_ctx_narrow_access_ok(u32 off, u32 size, const u32 size_default)
631 {
632 	bool off_ok;
633 #ifdef __LITTLE_ENDIAN
634 	off_ok = (off & (size_default - 1)) == 0;
635 #else
636 	off_ok = (off & (size_default - 1)) + size == size_default;
637 #endif
638 	return off_ok && size <= size_default && (size & (size - 1)) == 0;
639 }
640 
641 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
642 
643 #ifdef CONFIG_ARCH_HAS_SET_MEMORY
644 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
645 {
646 	fp->locked = 1;
647 	WARN_ON_ONCE(set_memory_ro((unsigned long)fp, fp->pages));
648 }
649 
650 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
651 {
652 	if (fp->locked) {
653 		WARN_ON_ONCE(set_memory_rw((unsigned long)fp, fp->pages));
654 		/* In case set_memory_rw() fails, we want to be the first
655 		 * to crash here instead of some random place later on.
656 		 */
657 		fp->locked = 0;
658 	}
659 }
660 
661 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
662 {
663 	WARN_ON_ONCE(set_memory_ro((unsigned long)hdr, hdr->pages));
664 }
665 
666 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
667 {
668 	WARN_ON_ONCE(set_memory_rw((unsigned long)hdr, hdr->pages));
669 }
670 #else
671 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
672 {
673 }
674 
675 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
676 {
677 }
678 
679 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
680 {
681 }
682 
683 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
684 {
685 }
686 #endif /* CONFIG_ARCH_HAS_SET_MEMORY */
687 
688 static inline struct bpf_binary_header *
689 bpf_jit_binary_hdr(const struct bpf_prog *fp)
690 {
691 	unsigned long real_start = (unsigned long)fp->bpf_func;
692 	unsigned long addr = real_start & PAGE_MASK;
693 
694 	return (void *)addr;
695 }
696 
697 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
698 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
699 {
700 	return sk_filter_trim_cap(sk, skb, 1);
701 }
702 
703 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
704 void bpf_prog_free(struct bpf_prog *fp);
705 
706 bool bpf_opcode_in_insntable(u8 code);
707 
708 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
709 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
710 				  gfp_t gfp_extra_flags);
711 void __bpf_prog_free(struct bpf_prog *fp);
712 
713 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
714 {
715 	bpf_prog_unlock_ro(fp);
716 	__bpf_prog_free(fp);
717 }
718 
719 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
720 				       unsigned int flen);
721 
722 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
723 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
724 			      bpf_aux_classic_check_t trans, bool save_orig);
725 void bpf_prog_destroy(struct bpf_prog *fp);
726 
727 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
728 int sk_attach_bpf(u32 ufd, struct sock *sk);
729 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
730 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
731 int sk_detach_filter(struct sock *sk);
732 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
733 		  unsigned int len);
734 
735 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
736 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
737 
738 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
739 #define __bpf_call_base_args \
740 	((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
741 	 __bpf_call_base)
742 
743 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
744 void bpf_jit_compile(struct bpf_prog *prog);
745 bool bpf_helper_changes_pkt_data(void *func);
746 
747 static inline bool bpf_dump_raw_ok(void)
748 {
749 	/* Reconstruction of call-sites is dependent on kallsyms,
750 	 * thus make dump the same restriction.
751 	 */
752 	return kallsyms_show_value() == 1;
753 }
754 
755 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
756 				       const struct bpf_insn *patch, u32 len);
757 
758 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
759  * same cpu context. Further for best results no more than a single map
760  * for the do_redirect/do_flush pair should be used. This limitation is
761  * because we only track one map and force a flush when the map changes.
762  * This does not appear to be a real limitation for existing software.
763  */
764 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
765 			    struct xdp_buff *xdp, struct bpf_prog *prog);
766 int xdp_do_redirect(struct net_device *dev,
767 		    struct xdp_buff *xdp,
768 		    struct bpf_prog *prog);
769 void xdp_do_flush_map(void);
770 
771 void bpf_warn_invalid_xdp_action(u32 act);
772 
773 struct sock *do_sk_redirect_map(struct sk_buff *skb);
774 struct sock *do_msg_redirect_map(struct sk_msg_buff *md);
775 
776 #ifdef CONFIG_BPF_JIT
777 extern int bpf_jit_enable;
778 extern int bpf_jit_harden;
779 extern int bpf_jit_kallsyms;
780 
781 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
782 
783 struct bpf_binary_header *
784 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
785 		     unsigned int alignment,
786 		     bpf_jit_fill_hole_t bpf_fill_ill_insns);
787 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
788 
789 void bpf_jit_free(struct bpf_prog *fp);
790 
791 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
792 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
793 
794 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
795 				u32 pass, void *image)
796 {
797 	pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
798 	       proglen, pass, image, current->comm, task_pid_nr(current));
799 
800 	if (image)
801 		print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
802 			       16, 1, image, proglen, false);
803 }
804 
805 static inline bool bpf_jit_is_ebpf(void)
806 {
807 # ifdef CONFIG_HAVE_EBPF_JIT
808 	return true;
809 # else
810 	return false;
811 # endif
812 }
813 
814 static inline bool ebpf_jit_enabled(void)
815 {
816 	return bpf_jit_enable && bpf_jit_is_ebpf();
817 }
818 
819 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
820 {
821 	return fp->jited && bpf_jit_is_ebpf();
822 }
823 
824 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
825 {
826 	/* These are the prerequisites, should someone ever have the
827 	 * idea to call blinding outside of them, we make sure to
828 	 * bail out.
829 	 */
830 	if (!bpf_jit_is_ebpf())
831 		return false;
832 	if (!prog->jit_requested)
833 		return false;
834 	if (!bpf_jit_harden)
835 		return false;
836 	if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
837 		return false;
838 
839 	return true;
840 }
841 
842 static inline bool bpf_jit_kallsyms_enabled(void)
843 {
844 	/* There are a couple of corner cases where kallsyms should
845 	 * not be enabled f.e. on hardening.
846 	 */
847 	if (bpf_jit_harden)
848 		return false;
849 	if (!bpf_jit_kallsyms)
850 		return false;
851 	if (bpf_jit_kallsyms == 1)
852 		return true;
853 
854 	return false;
855 }
856 
857 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
858 				 unsigned long *off, char *sym);
859 bool is_bpf_text_address(unsigned long addr);
860 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
861 		    char *sym);
862 
863 static inline const char *
864 bpf_address_lookup(unsigned long addr, unsigned long *size,
865 		   unsigned long *off, char **modname, char *sym)
866 {
867 	const char *ret = __bpf_address_lookup(addr, size, off, sym);
868 
869 	if (ret && modname)
870 		*modname = NULL;
871 	return ret;
872 }
873 
874 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
875 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
876 
877 #else /* CONFIG_BPF_JIT */
878 
879 static inline bool ebpf_jit_enabled(void)
880 {
881 	return false;
882 }
883 
884 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
885 {
886 	return false;
887 }
888 
889 static inline void bpf_jit_free(struct bpf_prog *fp)
890 {
891 	bpf_prog_unlock_free(fp);
892 }
893 
894 static inline bool bpf_jit_kallsyms_enabled(void)
895 {
896 	return false;
897 }
898 
899 static inline const char *
900 __bpf_address_lookup(unsigned long addr, unsigned long *size,
901 		     unsigned long *off, char *sym)
902 {
903 	return NULL;
904 }
905 
906 static inline bool is_bpf_text_address(unsigned long addr)
907 {
908 	return false;
909 }
910 
911 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
912 				  char *type, char *sym)
913 {
914 	return -ERANGE;
915 }
916 
917 static inline const char *
918 bpf_address_lookup(unsigned long addr, unsigned long *size,
919 		   unsigned long *off, char **modname, char *sym)
920 {
921 	return NULL;
922 }
923 
924 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
925 {
926 }
927 
928 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
929 {
930 }
931 #endif /* CONFIG_BPF_JIT */
932 
933 #define BPF_ANC		BIT(15)
934 
935 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
936 {
937 	switch (first->code) {
938 	case BPF_RET | BPF_K:
939 	case BPF_LD | BPF_W | BPF_LEN:
940 		return false;
941 
942 	case BPF_LD | BPF_W | BPF_ABS:
943 	case BPF_LD | BPF_H | BPF_ABS:
944 	case BPF_LD | BPF_B | BPF_ABS:
945 		if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
946 			return true;
947 		return false;
948 
949 	default:
950 		return true;
951 	}
952 }
953 
954 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
955 {
956 	BUG_ON(ftest->code & BPF_ANC);
957 
958 	switch (ftest->code) {
959 	case BPF_LD | BPF_W | BPF_ABS:
960 	case BPF_LD | BPF_H | BPF_ABS:
961 	case BPF_LD | BPF_B | BPF_ABS:
962 #define BPF_ANCILLARY(CODE)	case SKF_AD_OFF + SKF_AD_##CODE:	\
963 				return BPF_ANC | SKF_AD_##CODE
964 		switch (ftest->k) {
965 		BPF_ANCILLARY(PROTOCOL);
966 		BPF_ANCILLARY(PKTTYPE);
967 		BPF_ANCILLARY(IFINDEX);
968 		BPF_ANCILLARY(NLATTR);
969 		BPF_ANCILLARY(NLATTR_NEST);
970 		BPF_ANCILLARY(MARK);
971 		BPF_ANCILLARY(QUEUE);
972 		BPF_ANCILLARY(HATYPE);
973 		BPF_ANCILLARY(RXHASH);
974 		BPF_ANCILLARY(CPU);
975 		BPF_ANCILLARY(ALU_XOR_X);
976 		BPF_ANCILLARY(VLAN_TAG);
977 		BPF_ANCILLARY(VLAN_TAG_PRESENT);
978 		BPF_ANCILLARY(PAY_OFFSET);
979 		BPF_ANCILLARY(RANDOM);
980 		BPF_ANCILLARY(VLAN_TPID);
981 		}
982 		/* Fallthrough. */
983 	default:
984 		return ftest->code;
985 	}
986 }
987 
988 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
989 					   int k, unsigned int size);
990 
991 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
992 				     unsigned int size, void *buffer)
993 {
994 	if (k >= 0)
995 		return skb_header_pointer(skb, k, size, buffer);
996 
997 	return bpf_internal_load_pointer_neg_helper(skb, k, size);
998 }
999 
1000 static inline int bpf_tell_extensions(void)
1001 {
1002 	return SKF_AD_MAX;
1003 }
1004 
1005 struct bpf_sock_addr_kern {
1006 	struct sock *sk;
1007 	struct sockaddr *uaddr;
1008 	/* Temporary "register" to make indirect stores to nested structures
1009 	 * defined above. We need three registers to make such a store, but
1010 	 * only two (src and dst) are available at convert_ctx_access time
1011 	 */
1012 	u64 tmp_reg;
1013 };
1014 
1015 struct bpf_sock_ops_kern {
1016 	struct	sock *sk;
1017 	u32	op;
1018 	union {
1019 		u32 args[4];
1020 		u32 reply;
1021 		u32 replylong[4];
1022 	};
1023 	u32	is_fullsock;
1024 	u64	temp;			/* temp and everything after is not
1025 					 * initialized to 0 before calling
1026 					 * the BPF program. New fields that
1027 					 * should be initialized to 0 should
1028 					 * be inserted before temp.
1029 					 * temp is scratch storage used by
1030 					 * sock_ops_convert_ctx_access
1031 					 * as temporary storage of a register.
1032 					 */
1033 };
1034 
1035 #endif /* __LINUX_FILTER_H__ */
1036