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 A regs[insn->a_reg]
63 #define X regs[insn->x_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 /* Register for user BPF programs need to be reset first. */
272 goto *jumptable
[insn
->code
];
275 #define ALU(OPCODE, OP) \
276 ALU64_##OPCODE##_X: \
280 A = (u32) A OP (u32) X; \
282 ALU64_##OPCODE##_K: \
286 A = (u32) A OP (u32) K; \
323 if (unlikely(X
== 0))
329 if (unlikely(X
== 0))
332 A
= do_div(tmp
, (u32
) X
);
340 A
= do_div(tmp
, (u32
) K
);
343 if (unlikely(X
== 0))
348 if (unlikely(X
== 0))
351 do_div(tmp
, (u32
) X
);
359 do_div(tmp
, (u32
) K
);
365 A
= (__force u16
) cpu_to_be16(A
);
368 A
= (__force u32
) cpu_to_be32(A
);
371 A
= (__force u64
) cpu_to_be64(A
);
378 A
= (__force u16
) cpu_to_le16(A
);
381 A
= (__force u32
) cpu_to_le32(A
);
384 A
= (__force u64
) cpu_to_le64(A
);
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
) A
) > ((s64
) X
)) {
458 if (((s64
) A
) > ((s64
) K
)) {
464 if (((s64
) A
) >= ((s64
) X
)) {
470 if (((s64
) A
) >= ((s64
) K
)) {
490 /* STX and ST and LDX*/
491 #define LDST(SIZEOP, SIZE) \
493 *(SIZE *)(unsigned long) (A + insn->off) = X; \
496 *(SIZE *)(unsigned long) (A + insn->off) = K; \
499 A = *(SIZE *)(unsigned long) (X + insn->off); \
507 STX_XADD_W
: /* lock xadd *(u32 *)(A + insn->off) += X */
508 atomic_add((u32
) X
, (atomic_t
*)(unsigned long)
511 STX_XADD_DW
: /* lock xadd *(u64 *)(A + insn->off) += X */
512 atomic64_add((u64
) X
, (atomic64_t
*)(unsigned long)
515 LD_ABS_W
: /* BPF_R0 = ntohl(*(u32 *) (skb->data + K)) */
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.
534 * K == 32-bit immediate
537 * BPF_R0 - 8/16/32-bit skb data converted to cpu endianness
539 ptr
= load_pointer((struct sk_buff
*) ctx
, off
, 4, &tmp
);
540 if (likely(ptr
!= NULL
)) {
541 BPF_R0
= get_unaligned_be32(ptr
);
545 LD_ABS_H
: /* BPF_R0 = ntohs(*(u16 *) (skb->data + K)) */
548 ptr
= load_pointer((struct sk_buff
*) ctx
, off
, 2, &tmp
);
549 if (likely(ptr
!= NULL
)) {
550 BPF_R0
= get_unaligned_be16(ptr
);
554 LD_ABS_B
: /* BPF_R0 = *(u8 *) (ctx + K) */
557 ptr
= load_pointer((struct sk_buff
*) ctx
, off
, 1, &tmp
);
558 if (likely(ptr
!= NULL
)) {
563 LD_IND_W
: /* BPF_R0 = ntohl(*(u32 *) (skb->data + X + K)) */
566 LD_IND_H
: /* BPF_R0 = ntohs(*(u16 *) (skb->data + X + K)) */
569 LD_IND_B
: /* BPF_R0 = *(u8 *) (skb->data + X + K) */
574 /* If we ever reach this, we have a bug somewhere. */
575 WARN_RATELIMIT(1, "unknown opcode %02x\n", insn
->code
);
579 /* Helper to find the offset of pkt_type in sk_buff structure. We want
580 * to make sure its still a 3bit field starting at a byte boundary;
581 * taken from arch/x86/net/bpf_jit_comp.c.
583 #define PKT_TYPE_MAX 7
584 static unsigned int pkt_type_offset(void)
586 struct sk_buff skb_probe
= { .pkt_type
= ~0, };
587 u8
*ct
= (u8
*) &skb_probe
;
590 for (off
= 0; off
< sizeof(struct sk_buff
); off
++) {
591 if (ct
[off
] == PKT_TYPE_MAX
)
595 pr_err_once("Please fix %s, as pkt_type couldn't be found!\n", __func__
);
599 static u64
__skb_get_pay_offset(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
601 return __skb_get_poff((struct sk_buff
*)(unsigned long) ctx
);
604 static u64
__skb_get_nlattr(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
606 struct sk_buff
*skb
= (struct sk_buff
*)(unsigned long) ctx
;
609 if (skb_is_nonlinear(skb
))
612 if (skb
->len
< sizeof(struct nlattr
))
615 if (a
> skb
->len
- sizeof(struct nlattr
))
618 nla
= nla_find((struct nlattr
*) &skb
->data
[a
], skb
->len
- a
, x
);
620 return (void *) nla
- (void *) skb
->data
;
625 static u64
__skb_get_nlattr_nest(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
627 struct sk_buff
*skb
= (struct sk_buff
*)(unsigned long) ctx
;
630 if (skb_is_nonlinear(skb
))
633 if (skb
->len
< sizeof(struct nlattr
))
636 if (a
> skb
->len
- sizeof(struct nlattr
))
639 nla
= (struct nlattr
*) &skb
->data
[a
];
640 if (nla
->nla_len
> skb
->len
- a
)
643 nla
= nla_find_nested(nla
, x
);
645 return (void *) nla
- (void *) skb
->data
;
650 static u64
__get_raw_cpu_id(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
652 return raw_smp_processor_id();
655 /* note that this only generates 32-bit random numbers */
656 static u64
__get_random_u32(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
658 return prandom_u32();
661 static bool convert_bpf_extensions(struct sock_filter
*fp
,
662 struct sock_filter_int
**insnp
)
664 struct sock_filter_int
*insn
= *insnp
;
667 case SKF_AD_OFF
+ SKF_AD_PROTOCOL
:
668 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, protocol
) != 2);
670 /* A = *(u16 *) (ctx + offsetof(protocol)) */
671 *insn
= BPF_LDX_MEM(BPF_H
, BPF_REG_A
, BPF_REG_CTX
,
672 offsetof(struct sk_buff
, protocol
));
675 /* A = ntohs(A) [emitting a nop or swap16] */
676 insn
->code
= BPF_ALU
| BPF_END
| BPF_FROM_BE
;
677 insn
->a_reg
= BPF_REG_A
;
681 case SKF_AD_OFF
+ SKF_AD_PKTTYPE
:
682 *insn
= BPF_LDX_MEM(BPF_B
, BPF_REG_A
, BPF_REG_CTX
,
688 *insn
= BPF_ALU32_IMM(BPF_AND
, BPF_REG_A
, PKT_TYPE_MAX
);
691 case SKF_AD_OFF
+ SKF_AD_IFINDEX
:
692 case SKF_AD_OFF
+ SKF_AD_HATYPE
:
693 *insn
= BPF_LDX_MEM(size_to_bpf(FIELD_SIZEOF(struct sk_buff
, dev
)),
694 BPF_REG_TMP
, BPF_REG_CTX
,
695 offsetof(struct sk_buff
, dev
));
698 /* if (tmp != 0) goto pc+1 */
699 *insn
= BPF_JMP_IMM(BPF_JNE
, BPF_REG_TMP
, 0, 1);
702 *insn
= BPF_EXIT_INSN();
705 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device
, ifindex
) != 4);
706 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device
, type
) != 2);
708 insn
->a_reg
= BPF_REG_A
;
709 insn
->x_reg
= BPF_REG_TMP
;
711 if (fp
->k
== SKF_AD_OFF
+ SKF_AD_IFINDEX
) {
712 insn
->code
= BPF_LDX
| BPF_MEM
| BPF_W
;
713 insn
->off
= offsetof(struct net_device
, ifindex
);
715 insn
->code
= BPF_LDX
| BPF_MEM
| BPF_H
;
716 insn
->off
= offsetof(struct net_device
, type
);
720 case SKF_AD_OFF
+ SKF_AD_MARK
:
721 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, mark
) != 4);
723 *insn
= BPF_LDX_MEM(BPF_W
, BPF_REG_A
, BPF_REG_CTX
,
724 offsetof(struct sk_buff
, mark
));
727 case SKF_AD_OFF
+ SKF_AD_RXHASH
:
728 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, hash
) != 4);
730 *insn
= BPF_LDX_MEM(BPF_W
, BPF_REG_A
, BPF_REG_CTX
,
731 offsetof(struct sk_buff
, hash
));
734 case SKF_AD_OFF
+ SKF_AD_QUEUE
:
735 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, queue_mapping
) != 2);
737 *insn
= BPF_LDX_MEM(BPF_H
, BPF_REG_A
, BPF_REG_CTX
,
738 offsetof(struct sk_buff
, queue_mapping
));
741 case SKF_AD_OFF
+ SKF_AD_VLAN_TAG
:
742 case SKF_AD_OFF
+ SKF_AD_VLAN_TAG_PRESENT
:
743 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, vlan_tci
) != 2);
745 /* A = *(u16 *) (ctx + offsetof(vlan_tci)) */
746 *insn
= BPF_LDX_MEM(BPF_H
, BPF_REG_A
, BPF_REG_CTX
,
747 offsetof(struct sk_buff
, vlan_tci
));
750 BUILD_BUG_ON(VLAN_TAG_PRESENT
!= 0x1000);
752 if (fp
->k
== SKF_AD_OFF
+ SKF_AD_VLAN_TAG
) {
753 *insn
= BPF_ALU32_IMM(BPF_AND
, BPF_REG_A
,
757 *insn
= BPF_ALU32_IMM(BPF_RSH
, BPF_REG_A
, 12);
761 *insn
= BPF_ALU32_IMM(BPF_AND
, BPF_REG_A
, 1);
765 case SKF_AD_OFF
+ SKF_AD_PAY_OFFSET
:
766 case SKF_AD_OFF
+ SKF_AD_NLATTR
:
767 case SKF_AD_OFF
+ SKF_AD_NLATTR_NEST
:
768 case SKF_AD_OFF
+ SKF_AD_CPU
:
769 case SKF_AD_OFF
+ SKF_AD_RANDOM
:
771 *insn
= BPF_ALU64_REG(BPF_MOV
, BPF_REG_ARG1
, BPF_REG_CTX
);
775 *insn
= BPF_ALU64_REG(BPF_MOV
, BPF_REG_ARG2
, BPF_REG_A
);
779 *insn
= BPF_ALU64_REG(BPF_MOV
, BPF_REG_ARG3
, BPF_REG_X
);
782 /* Emit call(ctx, arg2=A, arg3=X) */
783 insn
->code
= BPF_JMP
| BPF_CALL
;
785 case SKF_AD_OFF
+ SKF_AD_PAY_OFFSET
:
786 insn
->imm
= __skb_get_pay_offset
- __bpf_call_base
;
788 case SKF_AD_OFF
+ SKF_AD_NLATTR
:
789 insn
->imm
= __skb_get_nlattr
- __bpf_call_base
;
791 case SKF_AD_OFF
+ SKF_AD_NLATTR_NEST
:
792 insn
->imm
= __skb_get_nlattr_nest
- __bpf_call_base
;
794 case SKF_AD_OFF
+ SKF_AD_CPU
:
795 insn
->imm
= __get_raw_cpu_id
- __bpf_call_base
;
797 case SKF_AD_OFF
+ SKF_AD_RANDOM
:
798 insn
->imm
= __get_random_u32
- __bpf_call_base
;
803 case SKF_AD_OFF
+ SKF_AD_ALU_XOR_X
:
805 *insn
= BPF_ALU32_REG(BPF_XOR
, BPF_REG_A
, BPF_REG_X
);
809 /* This is just a dummy call to avoid letting the compiler
810 * evict __bpf_call_base() as an optimization. Placed here
811 * where no-one bothers.
813 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
822 * sk_convert_filter - convert filter program
823 * @prog: the user passed filter program
824 * @len: the length of the user passed filter program
825 * @new_prog: buffer where converted program will be stored
826 * @new_len: pointer to store length of converted program
828 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
829 * Conversion workflow:
831 * 1) First pass for calculating the new program length:
832 * sk_convert_filter(old_prog, old_len, NULL, &new_len)
834 * 2) 2nd pass to remap in two passes: 1st pass finds new
835 * jump offsets, 2nd pass remapping:
836 * new_prog = kmalloc(sizeof(struct sock_filter_int) * new_len);
837 * sk_convert_filter(old_prog, old_len, new_prog, &new_len);
839 * User BPF's register A is mapped to our BPF register 6, user BPF
840 * register X is mapped to BPF register 7; frame pointer is always
841 * register 10; Context 'void *ctx' is stored in register 1, that is,
842 * for socket filters: ctx == 'struct sk_buff *', for seccomp:
843 * ctx == 'struct seccomp_data *'.
845 int sk_convert_filter(struct sock_filter
*prog
, int len
,
846 struct sock_filter_int
*new_prog
, int *new_len
)
848 int new_flen
= 0, pass
= 0, target
, i
;
849 struct sock_filter_int
*new_insn
;
850 struct sock_filter
*fp
;
854 BUILD_BUG_ON(BPF_MEMWORDS
* sizeof(u32
) > MAX_BPF_STACK
);
855 BUILD_BUG_ON(BPF_REG_FP
+ 1 != MAX_BPF_REG
);
857 if (len
<= 0 || len
>= BPF_MAXINSNS
)
861 addrs
= kzalloc(len
* sizeof(*addrs
), GFP_KERNEL
);
871 *new_insn
= BPF_ALU64_REG(BPF_MOV
, BPF_REG_CTX
, BPF_REG_ARG1
);
875 for (i
= 0; i
< len
; fp
++, i
++) {
876 struct sock_filter_int tmp_insns
[6] = { };
877 struct sock_filter_int
*insn
= tmp_insns
;
880 addrs
[i
] = new_insn
- new_prog
;
883 /* All arithmetic insns and skb loads map as-is. */
884 case BPF_ALU
| BPF_ADD
| BPF_X
:
885 case BPF_ALU
| BPF_ADD
| BPF_K
:
886 case BPF_ALU
| BPF_SUB
| BPF_X
:
887 case BPF_ALU
| BPF_SUB
| BPF_K
:
888 case BPF_ALU
| BPF_AND
| BPF_X
:
889 case BPF_ALU
| BPF_AND
| BPF_K
:
890 case BPF_ALU
| BPF_OR
| BPF_X
:
891 case BPF_ALU
| BPF_OR
| BPF_K
:
892 case BPF_ALU
| BPF_LSH
| BPF_X
:
893 case BPF_ALU
| BPF_LSH
| BPF_K
:
894 case BPF_ALU
| BPF_RSH
| BPF_X
:
895 case BPF_ALU
| BPF_RSH
| BPF_K
:
896 case BPF_ALU
| BPF_XOR
| BPF_X
:
897 case BPF_ALU
| BPF_XOR
| BPF_K
:
898 case BPF_ALU
| BPF_MUL
| BPF_X
:
899 case BPF_ALU
| BPF_MUL
| BPF_K
:
900 case BPF_ALU
| BPF_DIV
| BPF_X
:
901 case BPF_ALU
| BPF_DIV
| BPF_K
:
902 case BPF_ALU
| BPF_MOD
| BPF_X
:
903 case BPF_ALU
| BPF_MOD
| BPF_K
:
904 case BPF_ALU
| BPF_NEG
:
905 case BPF_LD
| BPF_ABS
| BPF_W
:
906 case BPF_LD
| BPF_ABS
| BPF_H
:
907 case BPF_LD
| BPF_ABS
| BPF_B
:
908 case BPF_LD
| BPF_IND
| BPF_W
:
909 case BPF_LD
| BPF_IND
| BPF_H
:
910 case BPF_LD
| BPF_IND
| BPF_B
:
911 /* Check for overloaded BPF extension and
912 * directly convert it if found, otherwise
913 * just move on with mapping.
915 if (BPF_CLASS(fp
->code
) == BPF_LD
&&
916 BPF_MODE(fp
->code
) == BPF_ABS
&&
917 convert_bpf_extensions(fp
, &insn
))
920 insn
->code
= fp
->code
;
921 insn
->a_reg
= BPF_REG_A
;
922 insn
->x_reg
= BPF_REG_X
;
926 /* Jump opcodes map as-is, but offsets need adjustment. */
927 case BPF_JMP
| BPF_JA
:
928 target
= i
+ fp
->k
+ 1;
929 insn
->code
= fp
->code
;
932 if (target >= len || target < 0) \
934 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
935 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
936 insn->off -= insn - tmp_insns; \
942 case BPF_JMP
| BPF_JEQ
| BPF_K
:
943 case BPF_JMP
| BPF_JEQ
| BPF_X
:
944 case BPF_JMP
| BPF_JSET
| BPF_K
:
945 case BPF_JMP
| BPF_JSET
| BPF_X
:
946 case BPF_JMP
| BPF_JGT
| BPF_K
:
947 case BPF_JMP
| BPF_JGT
| BPF_X
:
948 case BPF_JMP
| BPF_JGE
| BPF_K
:
949 case BPF_JMP
| BPF_JGE
| BPF_X
:
950 if (BPF_SRC(fp
->code
) == BPF_K
&& (int) fp
->k
< 0) {
951 /* BPF immediates are signed, zero extend
952 * immediate into tmp register and use it
955 insn
->code
= BPF_ALU
| BPF_MOV
| BPF_K
;
956 insn
->a_reg
= BPF_REG_TMP
;
960 insn
->a_reg
= BPF_REG_A
;
961 insn
->x_reg
= BPF_REG_TMP
;
964 insn
->a_reg
= BPF_REG_A
;
965 insn
->x_reg
= BPF_REG_X
;
967 bpf_src
= BPF_SRC(fp
->code
);
970 /* Common case where 'jump_false' is next insn. */
972 insn
->code
= BPF_JMP
| BPF_OP(fp
->code
) | bpf_src
;
973 target
= i
+ fp
->jt
+ 1;
978 /* Convert JEQ into JNE when 'jump_true' is next insn. */
979 if (fp
->jt
== 0 && BPF_OP(fp
->code
) == BPF_JEQ
) {
980 insn
->code
= BPF_JMP
| BPF_JNE
| bpf_src
;
981 target
= i
+ fp
->jf
+ 1;
986 /* Other jumps are mapped into two insns: Jxx and JA. */
987 target
= i
+ fp
->jt
+ 1;
988 insn
->code
= BPF_JMP
| BPF_OP(fp
->code
) | bpf_src
;
992 insn
->code
= BPF_JMP
| BPF_JA
;
993 target
= i
+ fp
->jf
+ 1;
997 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
998 case BPF_LDX
| BPF_MSH
| BPF_B
:
1000 *insn
= BPF_ALU64_REG(BPF_MOV
, BPF_REG_TMP
, BPF_REG_A
);
1003 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
1004 *insn
= BPF_LD_ABS(BPF_B
, fp
->k
);
1008 *insn
= BPF_ALU32_IMM(BPF_AND
, BPF_REG_A
, 0xf);
1012 *insn
= BPF_ALU32_IMM(BPF_LSH
, BPF_REG_A
, 2);
1016 *insn
= BPF_ALU64_REG(BPF_MOV
, BPF_REG_X
, BPF_REG_A
);
1020 *insn
= BPF_ALU64_REG(BPF_MOV
, BPF_REG_A
, BPF_REG_TMP
);
1023 /* RET_K, RET_A are remaped into 2 insns. */
1024 case BPF_RET
| BPF_A
:
1025 case BPF_RET
| BPF_K
:
1026 insn
->code
= BPF_ALU
| BPF_MOV
|
1027 (BPF_RVAL(fp
->code
) == BPF_K
?
1030 insn
->x_reg
= BPF_REG_A
;
1034 *insn
= BPF_EXIT_INSN();
1037 /* Store to stack. */
1040 insn
->code
= BPF_STX
| BPF_MEM
| BPF_W
;
1041 insn
->a_reg
= BPF_REG_FP
;
1042 insn
->x_reg
= fp
->code
== BPF_ST
?
1043 BPF_REG_A
: BPF_REG_X
;
1044 insn
->off
= -(BPF_MEMWORDS
- fp
->k
) * 4;
1047 /* Load from stack. */
1048 case BPF_LD
| BPF_MEM
:
1049 case BPF_LDX
| BPF_MEM
:
1050 insn
->code
= BPF_LDX
| BPF_MEM
| BPF_W
;
1051 insn
->a_reg
= BPF_CLASS(fp
->code
) == BPF_LD
?
1052 BPF_REG_A
: BPF_REG_X
;
1053 insn
->x_reg
= BPF_REG_FP
;
1054 insn
->off
= -(BPF_MEMWORDS
- fp
->k
) * 4;
1057 /* A = K or X = K */
1058 case BPF_LD
| BPF_IMM
:
1059 case BPF_LDX
| BPF_IMM
:
1060 insn
->code
= BPF_ALU
| BPF_MOV
| BPF_K
;
1061 insn
->a_reg
= BPF_CLASS(fp
->code
) == BPF_LD
?
1062 BPF_REG_A
: BPF_REG_X
;
1067 case BPF_MISC
| BPF_TAX
:
1068 *insn
= BPF_ALU64_REG(BPF_MOV
, BPF_REG_X
, BPF_REG_A
);
1072 case BPF_MISC
| BPF_TXA
:
1073 *insn
= BPF_ALU64_REG(BPF_MOV
, BPF_REG_A
, BPF_REG_X
);
1076 /* A = skb->len or X = skb->len */
1077 case BPF_LD
| BPF_W
| BPF_LEN
:
1078 case BPF_LDX
| BPF_W
| BPF_LEN
:
1079 insn
->code
= BPF_LDX
| BPF_MEM
| BPF_W
;
1080 insn
->a_reg
= BPF_CLASS(fp
->code
) == BPF_LD
?
1081 BPF_REG_A
: BPF_REG_X
;
1082 insn
->x_reg
= BPF_REG_CTX
;
1083 insn
->off
= offsetof(struct sk_buff
, len
);
1086 /* access seccomp_data fields */
1087 case BPF_LDX
| BPF_ABS
| BPF_W
:
1088 /* A = *(u32 *) (ctx + K) */
1089 *insn
= BPF_LDX_MEM(BPF_W
, BPF_REG_A
, BPF_REG_CTX
, fp
->k
);
1098 memcpy(new_insn
, tmp_insns
,
1099 sizeof(*insn
) * (insn
- tmp_insns
));
1101 new_insn
+= insn
- tmp_insns
;
1105 /* Only calculating new length. */
1106 *new_len
= new_insn
- new_prog
;
1111 if (new_flen
!= new_insn
- new_prog
) {
1112 new_flen
= new_insn
- new_prog
;
1120 BUG_ON(*new_len
!= new_flen
);
1129 * A BPF program is able to use 16 cells of memory to store intermediate
1130 * values (check u32 mem[BPF_MEMWORDS] in sk_run_filter()).
1132 * As we dont want to clear mem[] array for each packet going through
1133 * sk_run_filter(), we check that filter loaded by user never try to read
1134 * a cell if not previously written, and we check all branches to be sure
1135 * a malicious user doesn't try to abuse us.
1137 static int check_load_and_stores(struct sock_filter
*filter
, int flen
)
1139 u16
*masks
, memvalid
= 0; /* one bit per cell, 16 cells */
1142 BUILD_BUG_ON(BPF_MEMWORDS
> 16);
1143 masks
= kmalloc(flen
* sizeof(*masks
), GFP_KERNEL
);
1146 memset(masks
, 0xff, flen
* sizeof(*masks
));
1148 for (pc
= 0; pc
< flen
; pc
++) {
1149 memvalid
&= masks
[pc
];
1151 switch (filter
[pc
].code
) {
1154 memvalid
|= (1 << filter
[pc
].k
);
1158 if (!(memvalid
& (1 << filter
[pc
].k
))) {
1164 /* a jump must set masks on target */
1165 masks
[pc
+ 1 + filter
[pc
].k
] &= memvalid
;
1168 case BPF_S_JMP_JEQ_K
:
1169 case BPF_S_JMP_JEQ_X
:
1170 case BPF_S_JMP_JGE_K
:
1171 case BPF_S_JMP_JGE_X
:
1172 case BPF_S_JMP_JGT_K
:
1173 case BPF_S_JMP_JGT_X
:
1174 case BPF_S_JMP_JSET_X
:
1175 case BPF_S_JMP_JSET_K
:
1176 /* a jump must set masks on targets */
1177 masks
[pc
+ 1 + filter
[pc
].jt
] &= memvalid
;
1178 masks
[pc
+ 1 + filter
[pc
].jf
] &= memvalid
;
1189 * sk_chk_filter - verify socket filter code
1190 * @filter: filter to verify
1191 * @flen: length of filter
1193 * Check the user's filter code. If we let some ugly
1194 * filter code slip through kaboom! The filter must contain
1195 * no references or jumps that are out of range, no illegal
1196 * instructions, and must end with a RET instruction.
1198 * All jumps are forward as they are not signed.
1200 * Returns 0 if the rule set is legal or -EINVAL if not.
1202 int sk_chk_filter(struct sock_filter
*filter
, unsigned int flen
)
1205 * Valid instructions are initialized to non-0.
1206 * Invalid instructions are initialized to 0.
1208 static const u8 codes
[] = {
1209 [BPF_ALU
|BPF_ADD
|BPF_K
] = BPF_S_ALU_ADD_K
,
1210 [BPF_ALU
|BPF_ADD
|BPF_X
] = BPF_S_ALU_ADD_X
,
1211 [BPF_ALU
|BPF_SUB
|BPF_K
] = BPF_S_ALU_SUB_K
,
1212 [BPF_ALU
|BPF_SUB
|BPF_X
] = BPF_S_ALU_SUB_X
,
1213 [BPF_ALU
|BPF_MUL
|BPF_K
] = BPF_S_ALU_MUL_K
,
1214 [BPF_ALU
|BPF_MUL
|BPF_X
] = BPF_S_ALU_MUL_X
,
1215 [BPF_ALU
|BPF_DIV
|BPF_X
] = BPF_S_ALU_DIV_X
,
1216 [BPF_ALU
|BPF_MOD
|BPF_K
] = BPF_S_ALU_MOD_K
,
1217 [BPF_ALU
|BPF_MOD
|BPF_X
] = BPF_S_ALU_MOD_X
,
1218 [BPF_ALU
|BPF_AND
|BPF_K
] = BPF_S_ALU_AND_K
,
1219 [BPF_ALU
|BPF_AND
|BPF_X
] = BPF_S_ALU_AND_X
,
1220 [BPF_ALU
|BPF_OR
|BPF_K
] = BPF_S_ALU_OR_K
,
1221 [BPF_ALU
|BPF_OR
|BPF_X
] = BPF_S_ALU_OR_X
,
1222 [BPF_ALU
|BPF_XOR
|BPF_K
] = BPF_S_ALU_XOR_K
,
1223 [BPF_ALU
|BPF_XOR
|BPF_X
] = BPF_S_ALU_XOR_X
,
1224 [BPF_ALU
|BPF_LSH
|BPF_K
] = BPF_S_ALU_LSH_K
,
1225 [BPF_ALU
|BPF_LSH
|BPF_X
] = BPF_S_ALU_LSH_X
,
1226 [BPF_ALU
|BPF_RSH
|BPF_K
] = BPF_S_ALU_RSH_K
,
1227 [BPF_ALU
|BPF_RSH
|BPF_X
] = BPF_S_ALU_RSH_X
,
1228 [BPF_ALU
|BPF_NEG
] = BPF_S_ALU_NEG
,
1229 [BPF_LD
|BPF_W
|BPF_ABS
] = BPF_S_LD_W_ABS
,
1230 [BPF_LD
|BPF_H
|BPF_ABS
] = BPF_S_LD_H_ABS
,
1231 [BPF_LD
|BPF_B
|BPF_ABS
] = BPF_S_LD_B_ABS
,
1232 [BPF_LD
|BPF_W
|BPF_LEN
] = BPF_S_LD_W_LEN
,
1233 [BPF_LD
|BPF_W
|BPF_IND
] = BPF_S_LD_W_IND
,
1234 [BPF_LD
|BPF_H
|BPF_IND
] = BPF_S_LD_H_IND
,
1235 [BPF_LD
|BPF_B
|BPF_IND
] = BPF_S_LD_B_IND
,
1236 [BPF_LD
|BPF_IMM
] = BPF_S_LD_IMM
,
1237 [BPF_LDX
|BPF_W
|BPF_LEN
] = BPF_S_LDX_W_LEN
,
1238 [BPF_LDX
|BPF_B
|BPF_MSH
] = BPF_S_LDX_B_MSH
,
1239 [BPF_LDX
|BPF_IMM
] = BPF_S_LDX_IMM
,
1240 [BPF_MISC
|BPF_TAX
] = BPF_S_MISC_TAX
,
1241 [BPF_MISC
|BPF_TXA
] = BPF_S_MISC_TXA
,
1242 [BPF_RET
|BPF_K
] = BPF_S_RET_K
,
1243 [BPF_RET
|BPF_A
] = BPF_S_RET_A
,
1244 [BPF_ALU
|BPF_DIV
|BPF_K
] = BPF_S_ALU_DIV_K
,
1245 [BPF_LD
|BPF_MEM
] = BPF_S_LD_MEM
,
1246 [BPF_LDX
|BPF_MEM
] = BPF_S_LDX_MEM
,
1247 [BPF_ST
] = BPF_S_ST
,
1248 [BPF_STX
] = BPF_S_STX
,
1249 [BPF_JMP
|BPF_JA
] = BPF_S_JMP_JA
,
1250 [BPF_JMP
|BPF_JEQ
|BPF_K
] = BPF_S_JMP_JEQ_K
,
1251 [BPF_JMP
|BPF_JEQ
|BPF_X
] = BPF_S_JMP_JEQ_X
,
1252 [BPF_JMP
|BPF_JGE
|BPF_K
] = BPF_S_JMP_JGE_K
,
1253 [BPF_JMP
|BPF_JGE
|BPF_X
] = BPF_S_JMP_JGE_X
,
1254 [BPF_JMP
|BPF_JGT
|BPF_K
] = BPF_S_JMP_JGT_K
,
1255 [BPF_JMP
|BPF_JGT
|BPF_X
] = BPF_S_JMP_JGT_X
,
1256 [BPF_JMP
|BPF_JSET
|BPF_K
] = BPF_S_JMP_JSET_K
,
1257 [BPF_JMP
|BPF_JSET
|BPF_X
] = BPF_S_JMP_JSET_X
,
1262 if (flen
== 0 || flen
> BPF_MAXINSNS
)
1265 /* check the filter code now */
1266 for (pc
= 0; pc
< flen
; pc
++) {
1267 struct sock_filter
*ftest
= &filter
[pc
];
1268 u16 code
= ftest
->code
;
1270 if (code
>= ARRAY_SIZE(codes
))
1275 /* Some instructions need special checks */
1277 case BPF_S_ALU_DIV_K
:
1278 case BPF_S_ALU_MOD_K
:
1279 /* check for division by zero */
1287 /* check for invalid memory addresses */
1288 if (ftest
->k
>= BPF_MEMWORDS
)
1293 * Note, the large ftest->k might cause loops.
1294 * Compare this with conditional jumps below,
1295 * where offsets are limited. --ANK (981016)
1297 if (ftest
->k
>= (unsigned int)(flen
-pc
-1))
1300 case BPF_S_JMP_JEQ_K
:
1301 case BPF_S_JMP_JEQ_X
:
1302 case BPF_S_JMP_JGE_K
:
1303 case BPF_S_JMP_JGE_X
:
1304 case BPF_S_JMP_JGT_K
:
1305 case BPF_S_JMP_JGT_X
:
1306 case BPF_S_JMP_JSET_X
:
1307 case BPF_S_JMP_JSET_K
:
1308 /* for conditionals both must be safe */
1309 if (pc
+ ftest
->jt
+ 1 >= flen
||
1310 pc
+ ftest
->jf
+ 1 >= flen
)
1313 case BPF_S_LD_W_ABS
:
1314 case BPF_S_LD_H_ABS
:
1315 case BPF_S_LD_B_ABS
:
1317 #define ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1318 code = BPF_S_ANC_##CODE; \
1322 ANCILLARY(PROTOCOL
);
1326 ANCILLARY(NLATTR_NEST
);
1332 ANCILLARY(ALU_XOR_X
);
1333 ANCILLARY(VLAN_TAG
);
1334 ANCILLARY(VLAN_TAG_PRESENT
);
1335 ANCILLARY(PAY_OFFSET
);
1339 /* ancillary operation unknown or unsupported */
1340 if (anc_found
== false && ftest
->k
>= SKF_AD_OFF
)
1346 /* last instruction must be a RET code */
1347 switch (filter
[flen
- 1].code
) {
1350 return check_load_and_stores(filter
, flen
);
1354 EXPORT_SYMBOL(sk_chk_filter
);
1356 static int sk_store_orig_filter(struct sk_filter
*fp
,
1357 const struct sock_fprog
*fprog
)
1359 unsigned int fsize
= sk_filter_proglen(fprog
);
1360 struct sock_fprog_kern
*fkprog
;
1362 fp
->orig_prog
= kmalloc(sizeof(*fkprog
), GFP_KERNEL
);
1366 fkprog
= fp
->orig_prog
;
1367 fkprog
->len
= fprog
->len
;
1368 fkprog
->filter
= kmemdup(fp
->insns
, fsize
, GFP_KERNEL
);
1369 if (!fkprog
->filter
) {
1370 kfree(fp
->orig_prog
);
1377 static void sk_release_orig_filter(struct sk_filter
*fp
)
1379 struct sock_fprog_kern
*fprog
= fp
->orig_prog
;
1382 kfree(fprog
->filter
);
1388 * sk_filter_release_rcu - Release a socket filter by rcu_head
1389 * @rcu: rcu_head that contains the sk_filter to free
1391 static void sk_filter_release_rcu(struct rcu_head
*rcu
)
1393 struct sk_filter
*fp
= container_of(rcu
, struct sk_filter
, rcu
);
1395 sk_release_orig_filter(fp
);
1400 * sk_filter_release - release a socket filter
1401 * @fp: filter to remove
1403 * Remove a filter from a socket and release its resources.
1405 static void sk_filter_release(struct sk_filter
*fp
)
1407 if (atomic_dec_and_test(&fp
->refcnt
))
1408 call_rcu(&fp
->rcu
, sk_filter_release_rcu
);
1411 void sk_filter_uncharge(struct sock
*sk
, struct sk_filter
*fp
)
1413 atomic_sub(sk_filter_size(fp
->len
), &sk
->sk_omem_alloc
);
1414 sk_filter_release(fp
);
1417 void sk_filter_charge(struct sock
*sk
, struct sk_filter
*fp
)
1419 atomic_inc(&fp
->refcnt
);
1420 atomic_add(sk_filter_size(fp
->len
), &sk
->sk_omem_alloc
);
1423 static struct sk_filter
*__sk_migrate_realloc(struct sk_filter
*fp
,
1427 struct sk_filter
*fp_new
;
1430 return krealloc(fp
, len
, GFP_KERNEL
);
1432 fp_new
= sock_kmalloc(sk
, len
, GFP_KERNEL
);
1435 /* As we're kepping orig_prog in fp_new along,
1436 * we need to make sure we're not evicting it
1439 fp
->orig_prog
= NULL
;
1440 sk_filter_uncharge(sk
, fp
);
1446 static struct sk_filter
*__sk_migrate_filter(struct sk_filter
*fp
,
1449 struct sock_filter
*old_prog
;
1450 struct sk_filter
*old_fp
;
1451 int i
, err
, new_len
, old_len
= fp
->len
;
1453 /* We are free to overwrite insns et al right here as it
1454 * won't be used at this point in time anymore internally
1455 * after the migration to the internal BPF instruction
1458 BUILD_BUG_ON(sizeof(struct sock_filter
) !=
1459 sizeof(struct sock_filter_int
));
1461 /* For now, we need to unfiddle BPF_S_* identifiers in place.
1462 * This can sooner or later on be subject to removal, e.g. when
1463 * JITs have been converted.
1465 for (i
= 0; i
< fp
->len
; i
++)
1466 sk_decode_filter(&fp
->insns
[i
], &fp
->insns
[i
]);
1468 /* Conversion cannot happen on overlapping memory areas,
1469 * so we need to keep the user BPF around until the 2nd
1470 * pass. At this time, the user BPF is stored in fp->insns.
1472 old_prog
= kmemdup(fp
->insns
, old_len
* sizeof(struct sock_filter
),
1479 /* 1st pass: calculate the new program length. */
1480 err
= sk_convert_filter(old_prog
, old_len
, NULL
, &new_len
);
1484 /* Expand fp for appending the new filter representation. */
1486 fp
= __sk_migrate_realloc(old_fp
, sk
, sk_filter_size(new_len
));
1488 /* The old_fp is still around in case we couldn't
1489 * allocate new memory, so uncharge on that one.
1498 /* 2nd pass: remap sock_filter insns into sock_filter_int insns. */
1499 err
= sk_convert_filter(old_prog
, old_len
, fp
->insnsi
, &new_len
);
1501 /* 2nd sk_convert_filter() can fail only if it fails
1502 * to allocate memory, remapping must succeed. Note,
1503 * that at this time old_fp has already been released
1504 * by __sk_migrate_realloc().
1508 sk_filter_select_runtime(fp
);
1516 /* Rollback filter setup. */
1518 sk_filter_uncharge(sk
, fp
);
1521 return ERR_PTR(err
);
1524 void __weak
bpf_int_jit_compile(struct sk_filter
*prog
)
1529 * sk_filter_select_runtime - select execution runtime for BPF program
1530 * @fp: sk_filter populated with internal BPF program
1532 * try to JIT internal BPF program, if JIT is not available select interpreter
1533 * BPF program will be executed via SK_RUN_FILTER() macro
1535 void sk_filter_select_runtime(struct sk_filter
*fp
)
1537 fp
->bpf_func
= (void *) __sk_run_filter
;
1539 /* Probe if internal BPF can be JITed */
1540 bpf_int_jit_compile(fp
);
1542 EXPORT_SYMBOL_GPL(sk_filter_select_runtime
);
1544 /* free internal BPF program */
1545 void sk_filter_free(struct sk_filter
*fp
)
1549 EXPORT_SYMBOL_GPL(sk_filter_free
);
1551 static struct sk_filter
*__sk_prepare_filter(struct sk_filter
*fp
,
1556 fp
->bpf_func
= NULL
;
1559 err
= sk_chk_filter(fp
->insns
, fp
->len
);
1561 return ERR_PTR(err
);
1563 /* Probe if we can JIT compile the filter and if so, do
1564 * the compilation of the filter.
1566 bpf_jit_compile(fp
);
1568 /* JIT compiler couldn't process this filter, so do the
1569 * internal BPF translation for the optimized interpreter.
1572 fp
= __sk_migrate_filter(fp
, sk
);
1578 * sk_unattached_filter_create - create an unattached filter
1579 * @fprog: the filter program
1580 * @pfp: the unattached filter that is created
1582 * Create a filter independent of any socket. We first run some
1583 * sanity checks on it to make sure it does not explode on us later.
1584 * If an error occurs or there is insufficient memory for the filter
1585 * a negative errno code is returned. On success the return is zero.
1587 int sk_unattached_filter_create(struct sk_filter
**pfp
,
1588 struct sock_fprog_kern
*fprog
)
1590 unsigned int fsize
= sk_filter_proglen(fprog
);
1591 struct sk_filter
*fp
;
1593 /* Make sure new filter is there and in the right amounts. */
1594 if (fprog
->filter
== NULL
)
1597 fp
= kmalloc(sk_filter_size(fprog
->len
), GFP_KERNEL
);
1601 memcpy(fp
->insns
, fprog
->filter
, fsize
);
1603 atomic_set(&fp
->refcnt
, 1);
1604 fp
->len
= fprog
->len
;
1605 /* Since unattached filters are not copied back to user
1606 * space through sk_get_filter(), we do not need to hold
1607 * a copy here, and can spare us the work.
1609 fp
->orig_prog
= NULL
;
1611 /* __sk_prepare_filter() already takes care of uncharging
1612 * memory in case something goes wrong.
1614 fp
= __sk_prepare_filter(fp
, NULL
);
1621 EXPORT_SYMBOL_GPL(sk_unattached_filter_create
);
1623 void sk_unattached_filter_destroy(struct sk_filter
*fp
)
1625 sk_filter_release(fp
);
1627 EXPORT_SYMBOL_GPL(sk_unattached_filter_destroy
);
1630 * sk_attach_filter - attach a socket filter
1631 * @fprog: the filter program
1632 * @sk: the socket to use
1634 * Attach the user's filter code. We first run some sanity checks on
1635 * it to make sure it does not explode on us later. If an error
1636 * occurs or there is insufficient memory for the filter a negative
1637 * errno code is returned. On success the return is zero.
1639 int sk_attach_filter(struct sock_fprog
*fprog
, struct sock
*sk
)
1641 struct sk_filter
*fp
, *old_fp
;
1642 unsigned int fsize
= sk_filter_proglen(fprog
);
1643 unsigned int sk_fsize
= sk_filter_size(fprog
->len
);
1646 if (sock_flag(sk
, SOCK_FILTER_LOCKED
))
1649 /* Make sure new filter is there and in the right amounts. */
1650 if (fprog
->filter
== NULL
)
1653 fp
= sock_kmalloc(sk
, sk_fsize
, GFP_KERNEL
);
1657 if (copy_from_user(fp
->insns
, fprog
->filter
, fsize
)) {
1658 sock_kfree_s(sk
, fp
, sk_fsize
);
1662 atomic_set(&fp
->refcnt
, 1);
1663 fp
->len
= fprog
->len
;
1665 err
= sk_store_orig_filter(fp
, fprog
);
1667 sk_filter_uncharge(sk
, fp
);
1671 /* __sk_prepare_filter() already takes care of uncharging
1672 * memory in case something goes wrong.
1674 fp
= __sk_prepare_filter(fp
, sk
);
1678 old_fp
= rcu_dereference_protected(sk
->sk_filter
,
1679 sock_owned_by_user(sk
));
1680 rcu_assign_pointer(sk
->sk_filter
, fp
);
1683 sk_filter_uncharge(sk
, old_fp
);
1687 EXPORT_SYMBOL_GPL(sk_attach_filter
);
1689 int sk_detach_filter(struct sock
*sk
)
1692 struct sk_filter
*filter
;
1694 if (sock_flag(sk
, SOCK_FILTER_LOCKED
))
1697 filter
= rcu_dereference_protected(sk
->sk_filter
,
1698 sock_owned_by_user(sk
));
1700 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1701 sk_filter_uncharge(sk
, filter
);
1707 EXPORT_SYMBOL_GPL(sk_detach_filter
);
1709 void sk_decode_filter(struct sock_filter
*filt
, struct sock_filter
*to
)
1711 static const u16 decodes
[] = {
1712 [BPF_S_ALU_ADD_K
] = BPF_ALU
|BPF_ADD
|BPF_K
,
1713 [BPF_S_ALU_ADD_X
] = BPF_ALU
|BPF_ADD
|BPF_X
,
1714 [BPF_S_ALU_SUB_K
] = BPF_ALU
|BPF_SUB
|BPF_K
,
1715 [BPF_S_ALU_SUB_X
] = BPF_ALU
|BPF_SUB
|BPF_X
,
1716 [BPF_S_ALU_MUL_K
] = BPF_ALU
|BPF_MUL
|BPF_K
,
1717 [BPF_S_ALU_MUL_X
] = BPF_ALU
|BPF_MUL
|BPF_X
,
1718 [BPF_S_ALU_DIV_X
] = BPF_ALU
|BPF_DIV
|BPF_X
,
1719 [BPF_S_ALU_MOD_K
] = BPF_ALU
|BPF_MOD
|BPF_K
,
1720 [BPF_S_ALU_MOD_X
] = BPF_ALU
|BPF_MOD
|BPF_X
,
1721 [BPF_S_ALU_AND_K
] = BPF_ALU
|BPF_AND
|BPF_K
,
1722 [BPF_S_ALU_AND_X
] = BPF_ALU
|BPF_AND
|BPF_X
,
1723 [BPF_S_ALU_OR_K
] = BPF_ALU
|BPF_OR
|BPF_K
,
1724 [BPF_S_ALU_OR_X
] = BPF_ALU
|BPF_OR
|BPF_X
,
1725 [BPF_S_ALU_XOR_K
] = BPF_ALU
|BPF_XOR
|BPF_K
,
1726 [BPF_S_ALU_XOR_X
] = BPF_ALU
|BPF_XOR
|BPF_X
,
1727 [BPF_S_ALU_LSH_K
] = BPF_ALU
|BPF_LSH
|BPF_K
,
1728 [BPF_S_ALU_LSH_X
] = BPF_ALU
|BPF_LSH
|BPF_X
,
1729 [BPF_S_ALU_RSH_K
] = BPF_ALU
|BPF_RSH
|BPF_K
,
1730 [BPF_S_ALU_RSH_X
] = BPF_ALU
|BPF_RSH
|BPF_X
,
1731 [BPF_S_ALU_NEG
] = BPF_ALU
|BPF_NEG
,
1732 [BPF_S_LD_W_ABS
] = BPF_LD
|BPF_W
|BPF_ABS
,
1733 [BPF_S_LD_H_ABS
] = BPF_LD
|BPF_H
|BPF_ABS
,
1734 [BPF_S_LD_B_ABS
] = BPF_LD
|BPF_B
|BPF_ABS
,
1735 [BPF_S_ANC_PROTOCOL
] = BPF_LD
|BPF_B
|BPF_ABS
,
1736 [BPF_S_ANC_PKTTYPE
] = BPF_LD
|BPF_B
|BPF_ABS
,
1737 [BPF_S_ANC_IFINDEX
] = BPF_LD
|BPF_B
|BPF_ABS
,
1738 [BPF_S_ANC_NLATTR
] = BPF_LD
|BPF_B
|BPF_ABS
,
1739 [BPF_S_ANC_NLATTR_NEST
] = BPF_LD
|BPF_B
|BPF_ABS
,
1740 [BPF_S_ANC_MARK
] = BPF_LD
|BPF_B
|BPF_ABS
,
1741 [BPF_S_ANC_QUEUE
] = BPF_LD
|BPF_B
|BPF_ABS
,
1742 [BPF_S_ANC_HATYPE
] = BPF_LD
|BPF_B
|BPF_ABS
,
1743 [BPF_S_ANC_RXHASH
] = BPF_LD
|BPF_B
|BPF_ABS
,
1744 [BPF_S_ANC_CPU
] = BPF_LD
|BPF_B
|BPF_ABS
,
1745 [BPF_S_ANC_ALU_XOR_X
] = BPF_LD
|BPF_B
|BPF_ABS
,
1746 [BPF_S_ANC_VLAN_TAG
] = BPF_LD
|BPF_B
|BPF_ABS
,
1747 [BPF_S_ANC_VLAN_TAG_PRESENT
] = BPF_LD
|BPF_B
|BPF_ABS
,
1748 [BPF_S_ANC_PAY_OFFSET
] = BPF_LD
|BPF_B
|BPF_ABS
,
1749 [BPF_S_ANC_RANDOM
] = BPF_LD
|BPF_B
|BPF_ABS
,
1750 [BPF_S_LD_W_LEN
] = BPF_LD
|BPF_W
|BPF_LEN
,
1751 [BPF_S_LD_W_IND
] = BPF_LD
|BPF_W
|BPF_IND
,
1752 [BPF_S_LD_H_IND
] = BPF_LD
|BPF_H
|BPF_IND
,
1753 [BPF_S_LD_B_IND
] = BPF_LD
|BPF_B
|BPF_IND
,
1754 [BPF_S_LD_IMM
] = BPF_LD
|BPF_IMM
,
1755 [BPF_S_LDX_W_LEN
] = BPF_LDX
|BPF_W
|BPF_LEN
,
1756 [BPF_S_LDX_B_MSH
] = BPF_LDX
|BPF_B
|BPF_MSH
,
1757 [BPF_S_LDX_IMM
] = BPF_LDX
|BPF_IMM
,
1758 [BPF_S_MISC_TAX
] = BPF_MISC
|BPF_TAX
,
1759 [BPF_S_MISC_TXA
] = BPF_MISC
|BPF_TXA
,
1760 [BPF_S_RET_K
] = BPF_RET
|BPF_K
,
1761 [BPF_S_RET_A
] = BPF_RET
|BPF_A
,
1762 [BPF_S_ALU_DIV_K
] = BPF_ALU
|BPF_DIV
|BPF_K
,
1763 [BPF_S_LD_MEM
] = BPF_LD
|BPF_MEM
,
1764 [BPF_S_LDX_MEM
] = BPF_LDX
|BPF_MEM
,
1765 [BPF_S_ST
] = BPF_ST
,
1766 [BPF_S_STX
] = BPF_STX
,
1767 [BPF_S_JMP_JA
] = BPF_JMP
|BPF_JA
,
1768 [BPF_S_JMP_JEQ_K
] = BPF_JMP
|BPF_JEQ
|BPF_K
,
1769 [BPF_S_JMP_JEQ_X
] = BPF_JMP
|BPF_JEQ
|BPF_X
,
1770 [BPF_S_JMP_JGE_K
] = BPF_JMP
|BPF_JGE
|BPF_K
,
1771 [BPF_S_JMP_JGE_X
] = BPF_JMP
|BPF_JGE
|BPF_X
,
1772 [BPF_S_JMP_JGT_K
] = BPF_JMP
|BPF_JGT
|BPF_K
,
1773 [BPF_S_JMP_JGT_X
] = BPF_JMP
|BPF_JGT
|BPF_X
,
1774 [BPF_S_JMP_JSET_K
] = BPF_JMP
|BPF_JSET
|BPF_K
,
1775 [BPF_S_JMP_JSET_X
] = BPF_JMP
|BPF_JSET
|BPF_X
,
1781 to
->code
= decodes
[code
];
1787 int sk_get_filter(struct sock
*sk
, struct sock_filter __user
*ubuf
,
1790 struct sock_fprog_kern
*fprog
;
1791 struct sk_filter
*filter
;
1795 filter
= rcu_dereference_protected(sk
->sk_filter
,
1796 sock_owned_by_user(sk
));
1800 /* We're copying the filter that has been originally attached,
1801 * so no conversion/decode needed anymore.
1803 fprog
= filter
->orig_prog
;
1807 /* User space only enquires number of filter blocks. */
1811 if (len
< fprog
->len
)
1815 if (copy_to_user(ubuf
, fprog
->filter
, sk_filter_proglen(fprog
)))
1818 /* Instead of bytes, the API requests to return the number