2 * Linux Socket Filter - Kernel level socket filtering
4 * Based on the design of the Berkeley Packet Filter. The new
5 * internal format has been designed by PLUMgrid:
7 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
11 * Jay Schulist <jschlst@samba.org>
12 * Alexei Starovoitov <ast@plumgrid.com>
13 * Daniel Borkmann <dborkman@redhat.com>
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
20 * Andi Kleen - Fix a few bad bugs and races.
21 * Kris Katterjohn - Added many additional checks in sk_chk_filter()
24 #include <linux/module.h>
25 #include <linux/types.h>
27 #include <linux/fcntl.h>
28 #include <linux/socket.h>
30 #include <linux/inet.h>
31 #include <linux/netdevice.h>
32 #include <linux/if_packet.h>
33 #include <linux/gfp.h>
35 #include <net/protocol.h>
36 #include <net/netlink.h>
37 #include <linux/skbuff.h>
39 #include <linux/errno.h>
40 #include <linux/timer.h>
41 #include <asm/uaccess.h>
42 #include <asm/unaligned.h>
43 #include <linux/filter.h>
44 #include <linux/ratelimit.h>
45 #include <linux/seccomp.h>
46 #include <linux/if_vlan.h>
49 #define BPF_R0 regs[BPF_REG_0]
50 #define BPF_R1 regs[BPF_REG_1]
51 #define BPF_R2 regs[BPF_REG_2]
52 #define BPF_R3 regs[BPF_REG_3]
53 #define BPF_R4 regs[BPF_REG_4]
54 #define BPF_R5 regs[BPF_REG_5]
55 #define BPF_R6 regs[BPF_REG_6]
56 #define BPF_R7 regs[BPF_REG_7]
57 #define BPF_R8 regs[BPF_REG_8]
58 #define BPF_R9 regs[BPF_REG_9]
59 #define BPF_R10 regs[BPF_REG_10]
62 #define DST regs[insn->dst_reg]
63 #define SRC regs[insn->src_reg]
64 #define FP regs[BPF_REG_FP]
65 #define ARG1 regs[BPF_REG_ARG1]
66 #define CTX regs[BPF_REG_CTX]
69 /* No hurry in this branch
71 * Exported for the bpf jit load helper.
73 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff
*skb
, int k
, unsigned int size
)
78 ptr
= skb_network_header(skb
) + k
- SKF_NET_OFF
;
79 else if (k
>= SKF_LL_OFF
)
80 ptr
= skb_mac_header(skb
) + k
- SKF_LL_OFF
;
81 if (ptr
>= skb
->head
&& ptr
+ size
<= skb_tail_pointer(skb
))
87 static inline void *load_pointer(const struct sk_buff
*skb
, int k
,
88 unsigned int size
, void *buffer
)
91 return skb_header_pointer(skb
, k
, size
, buffer
);
93 return bpf_internal_load_pointer_neg_helper(skb
, k
, size
);
97 * sk_filter - run a packet through a socket filter
98 * @sk: sock associated with &sk_buff
99 * @skb: buffer to filter
101 * Run the filter code and then cut skb->data to correct size returned by
102 * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
103 * than pkt_len we keep whole skb->data. This is the socket level
104 * wrapper to sk_run_filter. It returns 0 if the packet should
105 * be accepted or -EPERM if the packet should be tossed.
108 int sk_filter(struct sock
*sk
, struct sk_buff
*skb
)
111 struct sk_filter
*filter
;
114 * If the skb was allocated from pfmemalloc reserves, only
115 * allow SOCK_MEMALLOC sockets to use it as this socket is
116 * helping free memory
118 if (skb_pfmemalloc(skb
) && !sock_flag(sk
, SOCK_MEMALLOC
))
121 err
= security_sock_rcv_skb(sk
, skb
);
126 filter
= rcu_dereference(sk
->sk_filter
);
128 unsigned int pkt_len
= SK_RUN_FILTER(filter
, skb
);
130 err
= pkt_len
? pskb_trim(skb
, pkt_len
) : -EPERM
;
136 EXPORT_SYMBOL(sk_filter
);
138 /* Base function for offset calculation. Needs to go into .text section,
139 * therefore keeping it non-static as well; will also be used by JITs
140 * anyway later on, so do not let the compiler omit it.
142 noinline u64
__bpf_call_base(u64 r1
, u64 r2
, u64 r3
, u64 r4
, u64 r5
)
148 * __sk_run_filter - run a filter on a given context
149 * @ctx: buffer to run the filter on
150 * @insn: filter to apply
152 * Decode and apply filter instructions to the skb->data. Return length to
153 * keep, 0 for none. @ctx is the data we are operating on, @insn is the
154 * array of filter instructions.
156 static unsigned int __sk_run_filter(void *ctx
, const struct sock_filter_int
*insn
)
158 u64 stack
[MAX_BPF_STACK
/ sizeof(u64
)];
159 u64 regs
[MAX_BPF_REG
], tmp
;
160 static const void *jumptable
[256] = {
161 [0 ... 255] = &&default_label
,
162 /* Now overwrite non-defaults ... */
163 /* 32 bit ALU operations */
164 [BPF_ALU
| BPF_ADD
| BPF_X
] = &&ALU_ADD_X
,
165 [BPF_ALU
| BPF_ADD
| BPF_K
] = &&ALU_ADD_K
,
166 [BPF_ALU
| BPF_SUB
| BPF_X
] = &&ALU_SUB_X
,
167 [BPF_ALU
| BPF_SUB
| BPF_K
] = &&ALU_SUB_K
,
168 [BPF_ALU
| BPF_AND
| BPF_X
] = &&ALU_AND_X
,
169 [BPF_ALU
| BPF_AND
| BPF_K
] = &&ALU_AND_K
,
170 [BPF_ALU
| BPF_OR
| BPF_X
] = &&ALU_OR_X
,
171 [BPF_ALU
| BPF_OR
| BPF_K
] = &&ALU_OR_K
,
172 [BPF_ALU
| BPF_LSH
| BPF_X
] = &&ALU_LSH_X
,
173 [BPF_ALU
| BPF_LSH
| BPF_K
] = &&ALU_LSH_K
,
174 [BPF_ALU
| BPF_RSH
| BPF_X
] = &&ALU_RSH_X
,
175 [BPF_ALU
| BPF_RSH
| BPF_K
] = &&ALU_RSH_K
,
176 [BPF_ALU
| BPF_XOR
| BPF_X
] = &&ALU_XOR_X
,
177 [BPF_ALU
| BPF_XOR
| BPF_K
] = &&ALU_XOR_K
,
178 [BPF_ALU
| BPF_MUL
| BPF_X
] = &&ALU_MUL_X
,
179 [BPF_ALU
| BPF_MUL
| BPF_K
] = &&ALU_MUL_K
,
180 [BPF_ALU
| BPF_MOV
| BPF_X
] = &&ALU_MOV_X
,
181 [BPF_ALU
| BPF_MOV
| BPF_K
] = &&ALU_MOV_K
,
182 [BPF_ALU
| BPF_DIV
| BPF_X
] = &&ALU_DIV_X
,
183 [BPF_ALU
| BPF_DIV
| BPF_K
] = &&ALU_DIV_K
,
184 [BPF_ALU
| BPF_MOD
| BPF_X
] = &&ALU_MOD_X
,
185 [BPF_ALU
| BPF_MOD
| BPF_K
] = &&ALU_MOD_K
,
186 [BPF_ALU
| BPF_NEG
] = &&ALU_NEG
,
187 [BPF_ALU
| BPF_END
| BPF_TO_BE
] = &&ALU_END_TO_BE
,
188 [BPF_ALU
| BPF_END
| BPF_TO_LE
] = &&ALU_END_TO_LE
,
189 /* 64 bit ALU operations */
190 [BPF_ALU64
| BPF_ADD
| BPF_X
] = &&ALU64_ADD_X
,
191 [BPF_ALU64
| BPF_ADD
| BPF_K
] = &&ALU64_ADD_K
,
192 [BPF_ALU64
| BPF_SUB
| BPF_X
] = &&ALU64_SUB_X
,
193 [BPF_ALU64
| BPF_SUB
| BPF_K
] = &&ALU64_SUB_K
,
194 [BPF_ALU64
| BPF_AND
| BPF_X
] = &&ALU64_AND_X
,
195 [BPF_ALU64
| BPF_AND
| BPF_K
] = &&ALU64_AND_K
,
196 [BPF_ALU64
| BPF_OR
| BPF_X
] = &&ALU64_OR_X
,
197 [BPF_ALU64
| BPF_OR
| BPF_K
] = &&ALU64_OR_K
,
198 [BPF_ALU64
| BPF_LSH
| BPF_X
] = &&ALU64_LSH_X
,
199 [BPF_ALU64
| BPF_LSH
| BPF_K
] = &&ALU64_LSH_K
,
200 [BPF_ALU64
| BPF_RSH
| BPF_X
] = &&ALU64_RSH_X
,
201 [BPF_ALU64
| BPF_RSH
| BPF_K
] = &&ALU64_RSH_K
,
202 [BPF_ALU64
| BPF_XOR
| BPF_X
] = &&ALU64_XOR_X
,
203 [BPF_ALU64
| BPF_XOR
| BPF_K
] = &&ALU64_XOR_K
,
204 [BPF_ALU64
| BPF_MUL
| BPF_X
] = &&ALU64_MUL_X
,
205 [BPF_ALU64
| BPF_MUL
| BPF_K
] = &&ALU64_MUL_K
,
206 [BPF_ALU64
| BPF_MOV
| BPF_X
] = &&ALU64_MOV_X
,
207 [BPF_ALU64
| BPF_MOV
| BPF_K
] = &&ALU64_MOV_K
,
208 [BPF_ALU64
| BPF_ARSH
| BPF_X
] = &&ALU64_ARSH_X
,
209 [BPF_ALU64
| BPF_ARSH
| BPF_K
] = &&ALU64_ARSH_K
,
210 [BPF_ALU64
| BPF_DIV
| BPF_X
] = &&ALU64_DIV_X
,
211 [BPF_ALU64
| BPF_DIV
| BPF_K
] = &&ALU64_DIV_K
,
212 [BPF_ALU64
| BPF_MOD
| BPF_X
] = &&ALU64_MOD_X
,
213 [BPF_ALU64
| BPF_MOD
| BPF_K
] = &&ALU64_MOD_K
,
214 [BPF_ALU64
| BPF_NEG
] = &&ALU64_NEG
,
215 /* Call instruction */
216 [BPF_JMP
| BPF_CALL
] = &&JMP_CALL
,
218 [BPF_JMP
| BPF_JA
] = &&JMP_JA
,
219 [BPF_JMP
| BPF_JEQ
| BPF_X
] = &&JMP_JEQ_X
,
220 [BPF_JMP
| BPF_JEQ
| BPF_K
] = &&JMP_JEQ_K
,
221 [BPF_JMP
| BPF_JNE
| BPF_X
] = &&JMP_JNE_X
,
222 [BPF_JMP
| BPF_JNE
| BPF_K
] = &&JMP_JNE_K
,
223 [BPF_JMP
| BPF_JGT
| BPF_X
] = &&JMP_JGT_X
,
224 [BPF_JMP
| BPF_JGT
| BPF_K
] = &&JMP_JGT_K
,
225 [BPF_JMP
| BPF_JGE
| BPF_X
] = &&JMP_JGE_X
,
226 [BPF_JMP
| BPF_JGE
| BPF_K
] = &&JMP_JGE_K
,
227 [BPF_JMP
| BPF_JSGT
| BPF_X
] = &&JMP_JSGT_X
,
228 [BPF_JMP
| BPF_JSGT
| BPF_K
] = &&JMP_JSGT_K
,
229 [BPF_JMP
| BPF_JSGE
| BPF_X
] = &&JMP_JSGE_X
,
230 [BPF_JMP
| BPF_JSGE
| BPF_K
] = &&JMP_JSGE_K
,
231 [BPF_JMP
| BPF_JSET
| BPF_X
] = &&JMP_JSET_X
,
232 [BPF_JMP
| BPF_JSET
| BPF_K
] = &&JMP_JSET_K
,
234 [BPF_JMP
| BPF_EXIT
] = &&JMP_EXIT
,
235 /* Store instructions */
236 [BPF_STX
| BPF_MEM
| BPF_B
] = &&STX_MEM_B
,
237 [BPF_STX
| BPF_MEM
| BPF_H
] = &&STX_MEM_H
,
238 [BPF_STX
| BPF_MEM
| BPF_W
] = &&STX_MEM_W
,
239 [BPF_STX
| BPF_MEM
| BPF_DW
] = &&STX_MEM_DW
,
240 [BPF_STX
| BPF_XADD
| BPF_W
] = &&STX_XADD_W
,
241 [BPF_STX
| BPF_XADD
| BPF_DW
] = &&STX_XADD_DW
,
242 [BPF_ST
| BPF_MEM
| BPF_B
] = &&ST_MEM_B
,
243 [BPF_ST
| BPF_MEM
| BPF_H
] = &&ST_MEM_H
,
244 [BPF_ST
| BPF_MEM
| BPF_W
] = &&ST_MEM_W
,
245 [BPF_ST
| BPF_MEM
| BPF_DW
] = &&ST_MEM_DW
,
246 /* Load instructions */
247 [BPF_LDX
| BPF_MEM
| BPF_B
] = &&LDX_MEM_B
,
248 [BPF_LDX
| BPF_MEM
| BPF_H
] = &&LDX_MEM_H
,
249 [BPF_LDX
| BPF_MEM
| BPF_W
] = &&LDX_MEM_W
,
250 [BPF_LDX
| BPF_MEM
| BPF_DW
] = &&LDX_MEM_DW
,
251 [BPF_LD
| BPF_ABS
| BPF_W
] = &&LD_ABS_W
,
252 [BPF_LD
| BPF_ABS
| BPF_H
] = &&LD_ABS_H
,
253 [BPF_LD
| BPF_ABS
| BPF_B
] = &&LD_ABS_B
,
254 [BPF_LD
| BPF_IND
| BPF_W
] = &&LD_IND_W
,
255 [BPF_LD
| BPF_IND
| BPF_H
] = &&LD_IND_H
,
256 [BPF_LD
| BPF_IND
| BPF_B
] = &&LD_IND_B
,
261 #define CONT ({ insn++; goto select_insn; })
262 #define CONT_JMP ({ insn++; goto select_insn; })
264 FP
= (u64
) (unsigned long) &stack
[ARRAY_SIZE(stack
)];
265 ARG1
= (u64
) (unsigned long) ctx
;
267 /* Registers used in classic BPF programs need to be reset first. */
272 goto *jumptable
[insn
->code
];
275 #define ALU(OPCODE, OP) \
276 ALU64_##OPCODE##_X: \
280 DST = (u32) DST OP (u32) SRC; \
282 ALU64_##OPCODE##_K: \
286 DST = (u32) DST OP (u32) IMM; \
317 (*(s64
*) &DST
) >>= SRC
;
320 (*(s64
*) &DST
) >>= IMM
;
323 if (unlikely(SRC
== 0))
326 DST
= do_div(tmp
, SRC
);
329 if (unlikely(SRC
== 0))
332 DST
= do_div(tmp
, (u32
) SRC
);
336 DST
= do_div(tmp
, IMM
);
340 DST
= do_div(tmp
, (u32
) IMM
);
343 if (unlikely(SRC
== 0))
348 if (unlikely(SRC
== 0))
351 do_div(tmp
, (u32
) SRC
);
359 do_div(tmp
, (u32
) IMM
);
365 DST
= (__force u16
) cpu_to_be16(DST
);
368 DST
= (__force u32
) cpu_to_be32(DST
);
371 DST
= (__force u64
) cpu_to_be64(DST
);
378 DST
= (__force u16
) cpu_to_le16(DST
);
381 DST
= (__force u32
) cpu_to_le32(DST
);
384 DST
= (__force u64
) cpu_to_le64(DST
);
391 /* Function call scratches BPF_R1-BPF_R5 registers,
392 * preserves BPF_R6-BPF_R9, and stores return value
395 BPF_R0
= (__bpf_call_base
+ insn
->imm
)(BPF_R1
, BPF_R2
, BPF_R3
,
452 if (((s64
) DST
) > ((s64
) SRC
)) {
458 if (((s64
) DST
) > ((s64
) IMM
)) {
464 if (((s64
) DST
) >= ((s64
) SRC
)) {
470 if (((s64
) DST
) >= ((s64
) IMM
)) {
490 /* STX and ST and LDX*/
491 #define LDST(SIZEOP, SIZE) \
493 *(SIZE *)(unsigned long) (DST + insn->off) = SRC; \
496 *(SIZE *)(unsigned long) (DST + insn->off) = IMM; \
499 DST = *(SIZE *)(unsigned long) (SRC + insn->off); \
507 STX_XADD_W
: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */
508 atomic_add((u32
) SRC
, (atomic_t
*)(unsigned long)
511 STX_XADD_DW
: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */
512 atomic64_add((u64
) SRC
, (atomic64_t
*)(unsigned long)
515 LD_ABS_W
: /* BPF_R0 = ntohl(*(u32 *) (skb->data + imm32)) */
518 /* BPF_LD + BPD_ABS and BPF_LD + BPF_IND insns are
519 * only appearing in the programs where ctx ==
520 * skb. All programs keep 'ctx' in regs[BPF_REG_CTX]
521 * == BPF_R6, sk_convert_filter() saves it in BPF_R6,
522 * internal BPF verifier will check that BPF_R6 ==
525 * BPF_ABS and BPF_IND are wrappers of function calls,
526 * so they scratch BPF_R1-BPF_R5 registers, preserve
527 * BPF_R6-BPF_R9, and store return value into BPF_R0.
530 * ctx == skb == BPF_R6 == CTX
533 * SRC == any register
534 * IMM == 32-bit immediate
537 * BPF_R0 - 8/16/32-bit skb data converted to cpu endianness
540 ptr
= load_pointer((struct sk_buff
*) (unsigned long) CTX
, off
, 4, &tmp
);
541 if (likely(ptr
!= NULL
)) {
542 BPF_R0
= get_unaligned_be32(ptr
);
547 LD_ABS_H
: /* BPF_R0 = ntohs(*(u16 *) (skb->data + imm32)) */
550 ptr
= load_pointer((struct sk_buff
*) (unsigned long) CTX
, off
, 2, &tmp
);
551 if (likely(ptr
!= NULL
)) {
552 BPF_R0
= get_unaligned_be16(ptr
);
557 LD_ABS_B
: /* BPF_R0 = *(u8 *) (skb->data + imm32) */
560 ptr
= load_pointer((struct sk_buff
*) (unsigned long) CTX
, off
, 1, &tmp
);
561 if (likely(ptr
!= NULL
)) {
567 LD_IND_W
: /* BPF_R0 = ntohl(*(u32 *) (skb->data + src_reg + imm32)) */
570 LD_IND_H
: /* BPF_R0 = ntohs(*(u16 *) (skb->data + src_reg + imm32)) */
573 LD_IND_B
: /* BPF_R0 = *(u8 *) (skb->data + src_reg + imm32) */
578 /* If we ever reach this, we have a bug somewhere. */
579 WARN_RATELIMIT(1, "unknown opcode %02x\n", insn
->code
);
583 /* Helper to find the offset of pkt_type in sk_buff structure. We want
584 * to make sure its still a 3bit field starting at a byte boundary;
585 * taken from arch/x86/net/bpf_jit_comp.c.
587 #ifdef __BIG_ENDIAN_BITFIELD
588 #define PKT_TYPE_MAX (7 << 5)
590 #define PKT_TYPE_MAX 7
592 static unsigned int pkt_type_offset(void)
594 struct sk_buff skb_probe
= { .pkt_type
= ~0, };
595 u8
*ct
= (u8
*) &skb_probe
;
598 for (off
= 0; off
< sizeof(struct sk_buff
); off
++) {
599 if (ct
[off
] == PKT_TYPE_MAX
)
603 pr_err_once("Please fix %s, as pkt_type couldn't be found!\n", __func__
);
607 static u64
__skb_get_pay_offset(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
609 return __skb_get_poff((struct sk_buff
*)(unsigned long) ctx
);
612 static u64
__skb_get_nlattr(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
614 struct sk_buff
*skb
= (struct sk_buff
*)(unsigned long) ctx
;
617 if (skb_is_nonlinear(skb
))
620 if (skb
->len
< sizeof(struct nlattr
))
623 if (a
> skb
->len
- sizeof(struct nlattr
))
626 nla
= nla_find((struct nlattr
*) &skb
->data
[a
], skb
->len
- a
, x
);
628 return (void *) nla
- (void *) skb
->data
;
633 static u64
__skb_get_nlattr_nest(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
635 struct sk_buff
*skb
= (struct sk_buff
*)(unsigned long) ctx
;
638 if (skb_is_nonlinear(skb
))
641 if (skb
->len
< sizeof(struct nlattr
))
644 if (a
> skb
->len
- sizeof(struct nlattr
))
647 nla
= (struct nlattr
*) &skb
->data
[a
];
648 if (nla
->nla_len
> skb
->len
- a
)
651 nla
= nla_find_nested(nla
, x
);
653 return (void *) nla
- (void *) skb
->data
;
658 static u64
__get_raw_cpu_id(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
660 return raw_smp_processor_id();
663 /* note that this only generates 32-bit random numbers */
664 static u64
__get_random_u32(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
666 return prandom_u32();
669 static bool convert_bpf_extensions(struct sock_filter
*fp
,
670 struct sock_filter_int
**insnp
)
672 struct sock_filter_int
*insn
= *insnp
;
675 case SKF_AD_OFF
+ SKF_AD_PROTOCOL
:
676 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, protocol
) != 2);
678 /* A = *(u16 *) (CTX + offsetof(protocol)) */
679 *insn
++ = BPF_LDX_MEM(BPF_H
, BPF_REG_A
, BPF_REG_CTX
,
680 offsetof(struct sk_buff
, protocol
));
681 /* A = ntohs(A) [emitting a nop or swap16] */
682 *insn
= BPF_ENDIAN(BPF_FROM_BE
, BPF_REG_A
, 16);
685 case SKF_AD_OFF
+ SKF_AD_PKTTYPE
:
686 *insn
= BPF_LDX_MEM(BPF_B
, BPF_REG_A
, BPF_REG_CTX
,
691 *insn
= BPF_ALU32_IMM(BPF_AND
, BPF_REG_A
, PKT_TYPE_MAX
);
692 #ifdef __BIG_ENDIAN_BITFIELD
694 *insn
= BPF_ALU32_IMM(BPF_RSH
, BPF_REG_A
, 5);
698 case SKF_AD_OFF
+ SKF_AD_IFINDEX
:
699 case SKF_AD_OFF
+ SKF_AD_HATYPE
:
700 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device
, ifindex
) != 4);
701 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device
, type
) != 2);
702 BUILD_BUG_ON(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff
, dev
)) < 0);
704 *insn
++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff
, dev
)),
705 BPF_REG_TMP
, BPF_REG_CTX
,
706 offsetof(struct sk_buff
, dev
));
707 /* if (tmp != 0) goto pc + 1 */
708 *insn
++ = BPF_JMP_IMM(BPF_JNE
, BPF_REG_TMP
, 0, 1);
709 *insn
++ = BPF_EXIT_INSN();
710 if (fp
->k
== SKF_AD_OFF
+ SKF_AD_IFINDEX
)
711 *insn
= BPF_LDX_MEM(BPF_W
, BPF_REG_A
, BPF_REG_TMP
,
712 offsetof(struct net_device
, ifindex
));
714 *insn
= BPF_LDX_MEM(BPF_H
, BPF_REG_A
, BPF_REG_TMP
,
715 offsetof(struct net_device
, type
));
718 case SKF_AD_OFF
+ SKF_AD_MARK
:
719 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, mark
) != 4);
721 *insn
= BPF_LDX_MEM(BPF_W
, BPF_REG_A
, BPF_REG_CTX
,
722 offsetof(struct sk_buff
, mark
));
725 case SKF_AD_OFF
+ SKF_AD_RXHASH
:
726 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, hash
) != 4);
728 *insn
= BPF_LDX_MEM(BPF_W
, BPF_REG_A
, BPF_REG_CTX
,
729 offsetof(struct sk_buff
, hash
));
732 case SKF_AD_OFF
+ SKF_AD_QUEUE
:
733 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, queue_mapping
) != 2);
735 *insn
= BPF_LDX_MEM(BPF_H
, BPF_REG_A
, BPF_REG_CTX
,
736 offsetof(struct sk_buff
, queue_mapping
));
739 case SKF_AD_OFF
+ SKF_AD_VLAN_TAG
:
740 case SKF_AD_OFF
+ SKF_AD_VLAN_TAG_PRESENT
:
741 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, vlan_tci
) != 2);
742 BUILD_BUG_ON(VLAN_TAG_PRESENT
!= 0x1000);
744 /* A = *(u16 *) (CTX + offsetof(vlan_tci)) */
745 *insn
++ = BPF_LDX_MEM(BPF_H
, BPF_REG_A
, BPF_REG_CTX
,
746 offsetof(struct sk_buff
, vlan_tci
));
747 if (fp
->k
== SKF_AD_OFF
+ SKF_AD_VLAN_TAG
) {
748 *insn
= BPF_ALU32_IMM(BPF_AND
, BPF_REG_A
,
752 *insn
++ = BPF_ALU32_IMM(BPF_RSH
, BPF_REG_A
, 12);
754 *insn
= BPF_ALU32_IMM(BPF_AND
, BPF_REG_A
, 1);
758 case SKF_AD_OFF
+ SKF_AD_PAY_OFFSET
:
759 case SKF_AD_OFF
+ SKF_AD_NLATTR
:
760 case SKF_AD_OFF
+ SKF_AD_NLATTR_NEST
:
761 case SKF_AD_OFF
+ SKF_AD_CPU
:
762 case SKF_AD_OFF
+ SKF_AD_RANDOM
:
764 *insn
++ = BPF_MOV64_REG(BPF_REG_ARG1
, BPF_REG_CTX
);
766 *insn
++ = BPF_MOV64_REG(BPF_REG_ARG2
, BPF_REG_A
);
768 *insn
++ = BPF_MOV64_REG(BPF_REG_ARG3
, BPF_REG_X
);
769 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
771 case SKF_AD_OFF
+ SKF_AD_PAY_OFFSET
:
772 *insn
= BPF_EMIT_CALL(__skb_get_pay_offset
);
774 case SKF_AD_OFF
+ SKF_AD_NLATTR
:
775 *insn
= BPF_EMIT_CALL(__skb_get_nlattr
);
777 case SKF_AD_OFF
+ SKF_AD_NLATTR_NEST
:
778 *insn
= BPF_EMIT_CALL(__skb_get_nlattr_nest
);
780 case SKF_AD_OFF
+ SKF_AD_CPU
:
781 *insn
= BPF_EMIT_CALL(__get_raw_cpu_id
);
783 case SKF_AD_OFF
+ SKF_AD_RANDOM
:
784 *insn
= BPF_EMIT_CALL(__get_random_u32
);
789 case SKF_AD_OFF
+ SKF_AD_ALU_XOR_X
:
791 *insn
= BPF_ALU32_REG(BPF_XOR
, BPF_REG_A
, BPF_REG_X
);
795 /* This is just a dummy call to avoid letting the compiler
796 * evict __bpf_call_base() as an optimization. Placed here
797 * where no-one bothers.
799 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
808 * sk_convert_filter - convert filter program
809 * @prog: the user passed filter program
810 * @len: the length of the user passed filter program
811 * @new_prog: buffer where converted program will be stored
812 * @new_len: pointer to store length of converted program
814 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
815 * Conversion workflow:
817 * 1) First pass for calculating the new program length:
818 * sk_convert_filter(old_prog, old_len, NULL, &new_len)
820 * 2) 2nd pass to remap in two passes: 1st pass finds new
821 * jump offsets, 2nd pass remapping:
822 * new_prog = kmalloc(sizeof(struct sock_filter_int) * new_len);
823 * sk_convert_filter(old_prog, old_len, new_prog, &new_len);
825 * User BPF's register A is mapped to our BPF register 6, user BPF
826 * register X is mapped to BPF register 7; frame pointer is always
827 * register 10; Context 'void *ctx' is stored in register 1, that is,
828 * for socket filters: ctx == 'struct sk_buff *', for seccomp:
829 * ctx == 'struct seccomp_data *'.
831 int sk_convert_filter(struct sock_filter
*prog
, int len
,
832 struct sock_filter_int
*new_prog
, int *new_len
)
834 int new_flen
= 0, pass
= 0, target
, i
;
835 struct sock_filter_int
*new_insn
;
836 struct sock_filter
*fp
;
840 BUILD_BUG_ON(BPF_MEMWORDS
* sizeof(u32
) > MAX_BPF_STACK
);
841 BUILD_BUG_ON(BPF_REG_FP
+ 1 != MAX_BPF_REG
);
843 if (len
<= 0 || len
> BPF_MAXINSNS
)
847 addrs
= kcalloc(len
, sizeof(*addrs
), GFP_KERNEL
);
857 *new_insn
= BPF_MOV64_REG(BPF_REG_CTX
, BPF_REG_ARG1
);
860 for (i
= 0; i
< len
; fp
++, i
++) {
861 struct sock_filter_int tmp_insns
[6] = { };
862 struct sock_filter_int
*insn
= tmp_insns
;
865 addrs
[i
] = new_insn
- new_prog
;
868 /* All arithmetic insns and skb loads map as-is. */
869 case BPF_ALU
| BPF_ADD
| BPF_X
:
870 case BPF_ALU
| BPF_ADD
| BPF_K
:
871 case BPF_ALU
| BPF_SUB
| BPF_X
:
872 case BPF_ALU
| BPF_SUB
| BPF_K
:
873 case BPF_ALU
| BPF_AND
| BPF_X
:
874 case BPF_ALU
| BPF_AND
| BPF_K
:
875 case BPF_ALU
| BPF_OR
| BPF_X
:
876 case BPF_ALU
| BPF_OR
| BPF_K
:
877 case BPF_ALU
| BPF_LSH
| BPF_X
:
878 case BPF_ALU
| BPF_LSH
| BPF_K
:
879 case BPF_ALU
| BPF_RSH
| BPF_X
:
880 case BPF_ALU
| BPF_RSH
| BPF_K
:
881 case BPF_ALU
| BPF_XOR
| BPF_X
:
882 case BPF_ALU
| BPF_XOR
| BPF_K
:
883 case BPF_ALU
| BPF_MUL
| BPF_X
:
884 case BPF_ALU
| BPF_MUL
| BPF_K
:
885 case BPF_ALU
| BPF_DIV
| BPF_X
:
886 case BPF_ALU
| BPF_DIV
| BPF_K
:
887 case BPF_ALU
| BPF_MOD
| BPF_X
:
888 case BPF_ALU
| BPF_MOD
| BPF_K
:
889 case BPF_ALU
| BPF_NEG
:
890 case BPF_LD
| BPF_ABS
| BPF_W
:
891 case BPF_LD
| BPF_ABS
| BPF_H
:
892 case BPF_LD
| BPF_ABS
| BPF_B
:
893 case BPF_LD
| BPF_IND
| BPF_W
:
894 case BPF_LD
| BPF_IND
| BPF_H
:
895 case BPF_LD
| BPF_IND
| BPF_B
:
896 /* Check for overloaded BPF extension and
897 * directly convert it if found, otherwise
898 * just move on with mapping.
900 if (BPF_CLASS(fp
->code
) == BPF_LD
&&
901 BPF_MODE(fp
->code
) == BPF_ABS
&&
902 convert_bpf_extensions(fp
, &insn
))
905 *insn
= BPF_RAW_INSN(fp
->code
, BPF_REG_A
, BPF_REG_X
, 0, fp
->k
);
908 /* Jump transformation cannot use BPF block macros
909 * everywhere as offset calculation and target updates
910 * require a bit more work than the rest, i.e. jump
911 * opcodes map as-is, but offsets need adjustment.
914 #define BPF_EMIT_JMP \
916 if (target >= len || target < 0) \
918 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
919 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
920 insn->off -= insn - tmp_insns; \
923 case BPF_JMP
| BPF_JA
:
924 target
= i
+ fp
->k
+ 1;
925 insn
->code
= fp
->code
;
929 case BPF_JMP
| BPF_JEQ
| BPF_K
:
930 case BPF_JMP
| BPF_JEQ
| BPF_X
:
931 case BPF_JMP
| BPF_JSET
| BPF_K
:
932 case BPF_JMP
| BPF_JSET
| BPF_X
:
933 case BPF_JMP
| BPF_JGT
| BPF_K
:
934 case BPF_JMP
| BPF_JGT
| BPF_X
:
935 case BPF_JMP
| BPF_JGE
| BPF_K
:
936 case BPF_JMP
| BPF_JGE
| BPF_X
:
937 if (BPF_SRC(fp
->code
) == BPF_K
&& (int) fp
->k
< 0) {
938 /* BPF immediates are signed, zero extend
939 * immediate into tmp register and use it
942 *insn
++ = BPF_MOV32_IMM(BPF_REG_TMP
, fp
->k
);
944 insn
->dst_reg
= BPF_REG_A
;
945 insn
->src_reg
= BPF_REG_TMP
;
948 insn
->dst_reg
= BPF_REG_A
;
949 insn
->src_reg
= BPF_REG_X
;
951 bpf_src
= BPF_SRC(fp
->code
);
954 /* Common case where 'jump_false' is next insn. */
956 insn
->code
= BPF_JMP
| BPF_OP(fp
->code
) | bpf_src
;
957 target
= i
+ fp
->jt
+ 1;
962 /* Convert JEQ into JNE when 'jump_true' is next insn. */
963 if (fp
->jt
== 0 && BPF_OP(fp
->code
) == BPF_JEQ
) {
964 insn
->code
= BPF_JMP
| BPF_JNE
| bpf_src
;
965 target
= i
+ fp
->jf
+ 1;
970 /* Other jumps are mapped into two insns: Jxx and JA. */
971 target
= i
+ fp
->jt
+ 1;
972 insn
->code
= BPF_JMP
| BPF_OP(fp
->code
) | bpf_src
;
976 insn
->code
= BPF_JMP
| BPF_JA
;
977 target
= i
+ fp
->jf
+ 1;
981 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
982 case BPF_LDX
| BPF_MSH
| BPF_B
:
984 *insn
++ = BPF_MOV64_REG(BPF_REG_TMP
, BPF_REG_A
);
985 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
986 *insn
++ = BPF_LD_ABS(BPF_B
, fp
->k
);
988 *insn
++ = BPF_ALU32_IMM(BPF_AND
, BPF_REG_A
, 0xf);
990 *insn
++ = BPF_ALU32_IMM(BPF_LSH
, BPF_REG_A
, 2);
992 *insn
++ = BPF_MOV64_REG(BPF_REG_X
, BPF_REG_A
);
994 *insn
= BPF_MOV64_REG(BPF_REG_A
, BPF_REG_TMP
);
997 /* RET_K, RET_A are remaped into 2 insns. */
998 case BPF_RET
| BPF_A
:
999 case BPF_RET
| BPF_K
:
1000 *insn
++ = BPF_MOV32_RAW(BPF_RVAL(fp
->code
) == BPF_K
?
1001 BPF_K
: BPF_X
, BPF_REG_0
,
1003 *insn
= BPF_EXIT_INSN();
1006 /* Store to stack. */
1009 *insn
= BPF_STX_MEM(BPF_W
, BPF_REG_FP
, BPF_CLASS(fp
->code
) ==
1010 BPF_ST
? BPF_REG_A
: BPF_REG_X
,
1011 -(BPF_MEMWORDS
- fp
->k
) * 4);
1014 /* Load from stack. */
1015 case BPF_LD
| BPF_MEM
:
1016 case BPF_LDX
| BPF_MEM
:
1017 *insn
= BPF_LDX_MEM(BPF_W
, BPF_CLASS(fp
->code
) == BPF_LD
?
1018 BPF_REG_A
: BPF_REG_X
, BPF_REG_FP
,
1019 -(BPF_MEMWORDS
- fp
->k
) * 4);
1022 /* A = K or X = K */
1023 case BPF_LD
| BPF_IMM
:
1024 case BPF_LDX
| BPF_IMM
:
1025 *insn
= BPF_MOV32_IMM(BPF_CLASS(fp
->code
) == BPF_LD
?
1026 BPF_REG_A
: BPF_REG_X
, fp
->k
);
1030 case BPF_MISC
| BPF_TAX
:
1031 *insn
= BPF_MOV64_REG(BPF_REG_X
, BPF_REG_A
);
1035 case BPF_MISC
| BPF_TXA
:
1036 *insn
= BPF_MOV64_REG(BPF_REG_A
, BPF_REG_X
);
1039 /* A = skb->len or X = skb->len */
1040 case BPF_LD
| BPF_W
| BPF_LEN
:
1041 case BPF_LDX
| BPF_W
| BPF_LEN
:
1042 *insn
= BPF_LDX_MEM(BPF_W
, BPF_CLASS(fp
->code
) == BPF_LD
?
1043 BPF_REG_A
: BPF_REG_X
, BPF_REG_CTX
,
1044 offsetof(struct sk_buff
, len
));
1047 /* Access seccomp_data fields. */
1048 case BPF_LDX
| BPF_ABS
| BPF_W
:
1049 /* A = *(u32 *) (ctx + K) */
1050 *insn
= BPF_LDX_MEM(BPF_W
, BPF_REG_A
, BPF_REG_CTX
, fp
->k
);
1053 /* Unkown instruction. */
1060 memcpy(new_insn
, tmp_insns
,
1061 sizeof(*insn
) * (insn
- tmp_insns
));
1062 new_insn
+= insn
- tmp_insns
;
1066 /* Only calculating new length. */
1067 *new_len
= new_insn
- new_prog
;
1072 if (new_flen
!= new_insn
- new_prog
) {
1073 new_flen
= new_insn
- new_prog
;
1080 BUG_ON(*new_len
!= new_flen
);
1089 * A BPF program is able to use 16 cells of memory to store intermediate
1090 * values (check u32 mem[BPF_MEMWORDS] in sk_run_filter()).
1092 * As we dont want to clear mem[] array for each packet going through
1093 * sk_run_filter(), we check that filter loaded by user never try to read
1094 * a cell if not previously written, and we check all branches to be sure
1095 * a malicious user doesn't try to abuse us.
1097 static int check_load_and_stores(struct sock_filter
*filter
, int flen
)
1099 u16
*masks
, memvalid
= 0; /* One bit per cell, 16 cells */
1102 BUILD_BUG_ON(BPF_MEMWORDS
> 16);
1104 masks
= kmalloc_array(flen
, sizeof(*masks
), GFP_KERNEL
);
1108 memset(masks
, 0xff, flen
* sizeof(*masks
));
1110 for (pc
= 0; pc
< flen
; pc
++) {
1111 memvalid
&= masks
[pc
];
1113 switch (filter
[pc
].code
) {
1116 memvalid
|= (1 << filter
[pc
].k
);
1118 case BPF_LD
| BPF_MEM
:
1119 case BPF_LDX
| BPF_MEM
:
1120 if (!(memvalid
& (1 << filter
[pc
].k
))) {
1125 case BPF_JMP
| BPF_JA
:
1126 /* A jump must set masks on target */
1127 masks
[pc
+ 1 + filter
[pc
].k
] &= memvalid
;
1130 case BPF_JMP
| BPF_JEQ
| BPF_K
:
1131 case BPF_JMP
| BPF_JEQ
| BPF_X
:
1132 case BPF_JMP
| BPF_JGE
| BPF_K
:
1133 case BPF_JMP
| BPF_JGE
| BPF_X
:
1134 case BPF_JMP
| BPF_JGT
| BPF_K
:
1135 case BPF_JMP
| BPF_JGT
| BPF_X
:
1136 case BPF_JMP
| BPF_JSET
| BPF_K
:
1137 case BPF_JMP
| BPF_JSET
| BPF_X
:
1138 /* A jump must set masks on targets */
1139 masks
[pc
+ 1 + filter
[pc
].jt
] &= memvalid
;
1140 masks
[pc
+ 1 + filter
[pc
].jf
] &= memvalid
;
1150 static bool chk_code_allowed(u16 code_to_probe
)
1152 static const bool codes
[] = {
1153 /* 32 bit ALU operations */
1154 [BPF_ALU
| BPF_ADD
| BPF_K
] = true,
1155 [BPF_ALU
| BPF_ADD
| BPF_X
] = true,
1156 [BPF_ALU
| BPF_SUB
| BPF_K
] = true,
1157 [BPF_ALU
| BPF_SUB
| BPF_X
] = true,
1158 [BPF_ALU
| BPF_MUL
| BPF_K
] = true,
1159 [BPF_ALU
| BPF_MUL
| BPF_X
] = true,
1160 [BPF_ALU
| BPF_DIV
| BPF_K
] = true,
1161 [BPF_ALU
| BPF_DIV
| BPF_X
] = true,
1162 [BPF_ALU
| BPF_MOD
| BPF_K
] = true,
1163 [BPF_ALU
| BPF_MOD
| BPF_X
] = true,
1164 [BPF_ALU
| BPF_AND
| BPF_K
] = true,
1165 [BPF_ALU
| BPF_AND
| BPF_X
] = true,
1166 [BPF_ALU
| BPF_OR
| BPF_K
] = true,
1167 [BPF_ALU
| BPF_OR
| BPF_X
] = true,
1168 [BPF_ALU
| BPF_XOR
| BPF_K
] = true,
1169 [BPF_ALU
| BPF_XOR
| BPF_X
] = true,
1170 [BPF_ALU
| BPF_LSH
| BPF_K
] = true,
1171 [BPF_ALU
| BPF_LSH
| BPF_X
] = true,
1172 [BPF_ALU
| BPF_RSH
| BPF_K
] = true,
1173 [BPF_ALU
| BPF_RSH
| BPF_X
] = true,
1174 [BPF_ALU
| BPF_NEG
] = true,
1175 /* Load instructions */
1176 [BPF_LD
| BPF_W
| BPF_ABS
] = true,
1177 [BPF_LD
| BPF_H
| BPF_ABS
] = true,
1178 [BPF_LD
| BPF_B
| BPF_ABS
] = true,
1179 [BPF_LD
| BPF_W
| BPF_LEN
] = true,
1180 [BPF_LD
| BPF_W
| BPF_IND
] = true,
1181 [BPF_LD
| BPF_H
| BPF_IND
] = true,
1182 [BPF_LD
| BPF_B
| BPF_IND
] = true,
1183 [BPF_LD
| BPF_IMM
] = true,
1184 [BPF_LD
| BPF_MEM
] = true,
1185 [BPF_LDX
| BPF_W
| BPF_LEN
] = true,
1186 [BPF_LDX
| BPF_B
| BPF_MSH
] = true,
1187 [BPF_LDX
| BPF_IMM
] = true,
1188 [BPF_LDX
| BPF_MEM
] = true,
1189 /* Store instructions */
1192 /* Misc instructions */
1193 [BPF_MISC
| BPF_TAX
] = true,
1194 [BPF_MISC
| BPF_TXA
] = true,
1195 /* Return instructions */
1196 [BPF_RET
| BPF_K
] = true,
1197 [BPF_RET
| BPF_A
] = true,
1198 /* Jump instructions */
1199 [BPF_JMP
| BPF_JA
] = true,
1200 [BPF_JMP
| BPF_JEQ
| BPF_K
] = true,
1201 [BPF_JMP
| BPF_JEQ
| BPF_X
] = true,
1202 [BPF_JMP
| BPF_JGE
| BPF_K
] = true,
1203 [BPF_JMP
| BPF_JGE
| BPF_X
] = true,
1204 [BPF_JMP
| BPF_JGT
| BPF_K
] = true,
1205 [BPF_JMP
| BPF_JGT
| BPF_X
] = true,
1206 [BPF_JMP
| BPF_JSET
| BPF_K
] = true,
1207 [BPF_JMP
| BPF_JSET
| BPF_X
] = true,
1210 if (code_to_probe
>= ARRAY_SIZE(codes
))
1213 return codes
[code_to_probe
];
1217 * sk_chk_filter - verify socket filter code
1218 * @filter: filter to verify
1219 * @flen: length of filter
1221 * Check the user's filter code. If we let some ugly
1222 * filter code slip through kaboom! The filter must contain
1223 * no references or jumps that are out of range, no illegal
1224 * instructions, and must end with a RET instruction.
1226 * All jumps are forward as they are not signed.
1228 * Returns 0 if the rule set is legal or -EINVAL if not.
1230 int sk_chk_filter(struct sock_filter
*filter
, unsigned int flen
)
1235 if (flen
== 0 || flen
> BPF_MAXINSNS
)
1238 /* Check the filter code now */
1239 for (pc
= 0; pc
< flen
; pc
++) {
1240 struct sock_filter
*ftest
= &filter
[pc
];
1242 /* May we actually operate on this code? */
1243 if (!chk_code_allowed(ftest
->code
))
1246 /* Some instructions need special checks */
1247 switch (ftest
->code
) {
1248 case BPF_ALU
| BPF_DIV
| BPF_K
:
1249 case BPF_ALU
| BPF_MOD
| BPF_K
:
1250 /* Check for division by zero */
1254 case BPF_LD
| BPF_MEM
:
1255 case BPF_LDX
| BPF_MEM
:
1258 /* Check for invalid memory addresses */
1259 if (ftest
->k
>= BPF_MEMWORDS
)
1262 case BPF_JMP
| BPF_JA
:
1263 /* Note, the large ftest->k might cause loops.
1264 * Compare this with conditional jumps below,
1265 * where offsets are limited. --ANK (981016)
1267 if (ftest
->k
>= (unsigned int)(flen
- pc
- 1))
1270 case BPF_JMP
| BPF_JEQ
| BPF_K
:
1271 case BPF_JMP
| BPF_JEQ
| BPF_X
:
1272 case BPF_JMP
| BPF_JGE
| BPF_K
:
1273 case BPF_JMP
| BPF_JGE
| BPF_X
:
1274 case BPF_JMP
| BPF_JGT
| BPF_K
:
1275 case BPF_JMP
| BPF_JGT
| BPF_X
:
1276 case BPF_JMP
| BPF_JSET
| BPF_K
:
1277 case BPF_JMP
| BPF_JSET
| BPF_X
:
1278 /* Both conditionals must be safe */
1279 if (pc
+ ftest
->jt
+ 1 >= flen
||
1280 pc
+ ftest
->jf
+ 1 >= flen
)
1283 case BPF_LD
| BPF_W
| BPF_ABS
:
1284 case BPF_LD
| BPF_H
| BPF_ABS
:
1285 case BPF_LD
| BPF_B
| BPF_ABS
:
1287 if (bpf_anc_helper(ftest
) & BPF_ANC
)
1289 /* Ancillary operation unknown or unsupported */
1290 if (anc_found
== false && ftest
->k
>= SKF_AD_OFF
)
1295 /* Last instruction must be a RET code */
1296 switch (filter
[flen
- 1].code
) {
1297 case BPF_RET
| BPF_K
:
1298 case BPF_RET
| BPF_A
:
1299 return check_load_and_stores(filter
, flen
);
1304 EXPORT_SYMBOL(sk_chk_filter
);
1306 static int sk_store_orig_filter(struct sk_filter
*fp
,
1307 const struct sock_fprog
*fprog
)
1309 unsigned int fsize
= sk_filter_proglen(fprog
);
1310 struct sock_fprog_kern
*fkprog
;
1312 fp
->orig_prog
= kmalloc(sizeof(*fkprog
), GFP_KERNEL
);
1316 fkprog
= fp
->orig_prog
;
1317 fkprog
->len
= fprog
->len
;
1318 fkprog
->filter
= kmemdup(fp
->insns
, fsize
, GFP_KERNEL
);
1319 if (!fkprog
->filter
) {
1320 kfree(fp
->orig_prog
);
1327 static void sk_release_orig_filter(struct sk_filter
*fp
)
1329 struct sock_fprog_kern
*fprog
= fp
->orig_prog
;
1332 kfree(fprog
->filter
);
1338 * sk_filter_release_rcu - Release a socket filter by rcu_head
1339 * @rcu: rcu_head that contains the sk_filter to free
1341 static void sk_filter_release_rcu(struct rcu_head
*rcu
)
1343 struct sk_filter
*fp
= container_of(rcu
, struct sk_filter
, rcu
);
1345 sk_release_orig_filter(fp
);
1350 * sk_filter_release - release a socket filter
1351 * @fp: filter to remove
1353 * Remove a filter from a socket and release its resources.
1355 static void sk_filter_release(struct sk_filter
*fp
)
1357 if (atomic_dec_and_test(&fp
->refcnt
))
1358 call_rcu(&fp
->rcu
, sk_filter_release_rcu
);
1361 void sk_filter_uncharge(struct sock
*sk
, struct sk_filter
*fp
)
1363 atomic_sub(sk_filter_size(fp
->len
), &sk
->sk_omem_alloc
);
1364 sk_filter_release(fp
);
1367 void sk_filter_charge(struct sock
*sk
, struct sk_filter
*fp
)
1369 atomic_inc(&fp
->refcnt
);
1370 atomic_add(sk_filter_size(fp
->len
), &sk
->sk_omem_alloc
);
1373 static struct sk_filter
*__sk_migrate_realloc(struct sk_filter
*fp
,
1377 struct sk_filter
*fp_new
;
1380 return krealloc(fp
, len
, GFP_KERNEL
);
1382 fp_new
= sock_kmalloc(sk
, len
, GFP_KERNEL
);
1385 /* As we're keeping orig_prog in fp_new along,
1386 * we need to make sure we're not evicting it
1389 fp
->orig_prog
= NULL
;
1390 sk_filter_uncharge(sk
, fp
);
1396 static struct sk_filter
*__sk_migrate_filter(struct sk_filter
*fp
,
1399 struct sock_filter
*old_prog
;
1400 struct sk_filter
*old_fp
;
1401 int err
, new_len
, old_len
= fp
->len
;
1403 /* We are free to overwrite insns et al right here as it
1404 * won't be used at this point in time anymore internally
1405 * after the migration to the internal BPF instruction
1408 BUILD_BUG_ON(sizeof(struct sock_filter
) !=
1409 sizeof(struct sock_filter_int
));
1411 /* Conversion cannot happen on overlapping memory areas,
1412 * so we need to keep the user BPF around until the 2nd
1413 * pass. At this time, the user BPF is stored in fp->insns.
1415 old_prog
= kmemdup(fp
->insns
, old_len
* sizeof(struct sock_filter
),
1422 /* 1st pass: calculate the new program length. */
1423 err
= sk_convert_filter(old_prog
, old_len
, NULL
, &new_len
);
1427 /* Expand fp for appending the new filter representation. */
1429 fp
= __sk_migrate_realloc(old_fp
, sk
, sk_filter_size(new_len
));
1431 /* The old_fp is still around in case we couldn't
1432 * allocate new memory, so uncharge on that one.
1441 /* 2nd pass: remap sock_filter insns into sock_filter_int insns. */
1442 err
= sk_convert_filter(old_prog
, old_len
, fp
->insnsi
, &new_len
);
1444 /* 2nd sk_convert_filter() can fail only if it fails
1445 * to allocate memory, remapping must succeed. Note,
1446 * that at this time old_fp has already been released
1447 * by __sk_migrate_realloc().
1451 sk_filter_select_runtime(fp
);
1459 /* Rollback filter setup. */
1461 sk_filter_uncharge(sk
, fp
);
1464 return ERR_PTR(err
);
1467 void __weak
bpf_int_jit_compile(struct sk_filter
*prog
)
1472 * sk_filter_select_runtime - select execution runtime for BPF program
1473 * @fp: sk_filter populated with internal BPF program
1475 * try to JIT internal BPF program, if JIT is not available select interpreter
1476 * BPF program will be executed via SK_RUN_FILTER() macro
1478 void sk_filter_select_runtime(struct sk_filter
*fp
)
1480 fp
->bpf_func
= (void *) __sk_run_filter
;
1482 /* Probe if internal BPF can be JITed */
1483 bpf_int_jit_compile(fp
);
1485 EXPORT_SYMBOL_GPL(sk_filter_select_runtime
);
1487 /* free internal BPF program */
1488 void sk_filter_free(struct sk_filter
*fp
)
1492 EXPORT_SYMBOL_GPL(sk_filter_free
);
1494 static struct sk_filter
*__sk_prepare_filter(struct sk_filter
*fp
,
1499 fp
->bpf_func
= NULL
;
1502 err
= sk_chk_filter(fp
->insns
, fp
->len
);
1505 sk_filter_uncharge(sk
, fp
);
1508 return ERR_PTR(err
);
1511 /* Probe if we can JIT compile the filter and if so, do
1512 * the compilation of the filter.
1514 bpf_jit_compile(fp
);
1516 /* JIT compiler couldn't process this filter, so do the
1517 * internal BPF translation for the optimized interpreter.
1520 fp
= __sk_migrate_filter(fp
, sk
);
1526 * sk_unattached_filter_create - create an unattached filter
1527 * @pfp: the unattached filter that is created
1528 * @fprog: the filter program
1530 * Create a filter independent of any socket. We first run some
1531 * sanity checks on it to make sure it does not explode on us later.
1532 * If an error occurs or there is insufficient memory for the filter
1533 * a negative errno code is returned. On success the return is zero.
1535 int sk_unattached_filter_create(struct sk_filter
**pfp
,
1536 struct sock_fprog_kern
*fprog
)
1538 unsigned int fsize
= sk_filter_proglen(fprog
);
1539 struct sk_filter
*fp
;
1541 /* Make sure new filter is there and in the right amounts. */
1542 if (fprog
->filter
== NULL
)
1545 fp
= kmalloc(sk_filter_size(fprog
->len
), GFP_KERNEL
);
1549 memcpy(fp
->insns
, fprog
->filter
, fsize
);
1551 atomic_set(&fp
->refcnt
, 1);
1552 fp
->len
= fprog
->len
;
1553 /* Since unattached filters are not copied back to user
1554 * space through sk_get_filter(), we do not need to hold
1555 * a copy here, and can spare us the work.
1557 fp
->orig_prog
= NULL
;
1559 /* __sk_prepare_filter() already takes care of uncharging
1560 * memory in case something goes wrong.
1562 fp
= __sk_prepare_filter(fp
, NULL
);
1569 EXPORT_SYMBOL_GPL(sk_unattached_filter_create
);
1571 void sk_unattached_filter_destroy(struct sk_filter
*fp
)
1573 sk_filter_release(fp
);
1575 EXPORT_SYMBOL_GPL(sk_unattached_filter_destroy
);
1578 * sk_attach_filter - attach a socket filter
1579 * @fprog: the filter program
1580 * @sk: the socket to use
1582 * Attach the user's filter code. We first run some sanity checks on
1583 * it to make sure it does not explode on us later. If an error
1584 * occurs or there is insufficient memory for the filter a negative
1585 * errno code is returned. On success the return is zero.
1587 int sk_attach_filter(struct sock_fprog
*fprog
, struct sock
*sk
)
1589 struct sk_filter
*fp
, *old_fp
;
1590 unsigned int fsize
= sk_filter_proglen(fprog
);
1591 unsigned int sk_fsize
= sk_filter_size(fprog
->len
);
1594 if (sock_flag(sk
, SOCK_FILTER_LOCKED
))
1597 /* Make sure new filter is there and in the right amounts. */
1598 if (fprog
->filter
== NULL
)
1601 fp
= sock_kmalloc(sk
, sk_fsize
, GFP_KERNEL
);
1605 if (copy_from_user(fp
->insns
, fprog
->filter
, fsize
)) {
1606 sock_kfree_s(sk
, fp
, sk_fsize
);
1610 atomic_set(&fp
->refcnt
, 1);
1611 fp
->len
= fprog
->len
;
1613 err
= sk_store_orig_filter(fp
, fprog
);
1615 sk_filter_uncharge(sk
, fp
);
1619 /* __sk_prepare_filter() already takes care of uncharging
1620 * memory in case something goes wrong.
1622 fp
= __sk_prepare_filter(fp
, sk
);
1626 old_fp
= rcu_dereference_protected(sk
->sk_filter
,
1627 sock_owned_by_user(sk
));
1628 rcu_assign_pointer(sk
->sk_filter
, fp
);
1631 sk_filter_uncharge(sk
, old_fp
);
1635 EXPORT_SYMBOL_GPL(sk_attach_filter
);
1637 int sk_detach_filter(struct sock
*sk
)
1640 struct sk_filter
*filter
;
1642 if (sock_flag(sk
, SOCK_FILTER_LOCKED
))
1645 filter
= rcu_dereference_protected(sk
->sk_filter
,
1646 sock_owned_by_user(sk
));
1648 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1649 sk_filter_uncharge(sk
, filter
);
1655 EXPORT_SYMBOL_GPL(sk_detach_filter
);
1657 int sk_get_filter(struct sock
*sk
, struct sock_filter __user
*ubuf
,
1660 struct sock_fprog_kern
*fprog
;
1661 struct sk_filter
*filter
;
1665 filter
= rcu_dereference_protected(sk
->sk_filter
,
1666 sock_owned_by_user(sk
));
1670 /* We're copying the filter that has been originally attached,
1671 * so no conversion/decode needed anymore.
1673 fprog
= filter
->orig_prog
;
1677 /* User space only enquires number of filter blocks. */
1681 if (len
< fprog
->len
)
1685 if (copy_to_user(ubuf
, fprog
->filter
, sk_filter_proglen(fprog
)))
1688 /* Instead of bytes, the API requests to return the number