Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Linux Socket Filter - Kernel level socket filtering | |
3 | * | |
bd4cf0ed AS |
4 | * Based on the design of the Berkeley Packet Filter. The new |
5 | * internal format has been designed by PLUMgrid: | |
1da177e4 | 6 | * |
bd4cf0ed AS |
7 | * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com |
8 | * | |
9 | * Authors: | |
10 | * | |
11 | * Jay Schulist <jschlst@samba.org> | |
12 | * Alexei Starovoitov <ast@plumgrid.com> | |
13 | * Daniel Borkmann <dborkman@redhat.com> | |
1da177e4 LT |
14 | * |
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. | |
19 | * | |
20 | * Andi Kleen - Fix a few bad bugs and races. | |
93699863 | 21 | * Kris Katterjohn - Added many additional checks in sk_chk_filter() |
1da177e4 LT |
22 | */ |
23 | ||
24 | #include <linux/module.h> | |
25 | #include <linux/types.h> | |
1da177e4 LT |
26 | #include <linux/mm.h> |
27 | #include <linux/fcntl.h> | |
28 | #include <linux/socket.h> | |
29 | #include <linux/in.h> | |
30 | #include <linux/inet.h> | |
31 | #include <linux/netdevice.h> | |
32 | #include <linux/if_packet.h> | |
5a0e3ad6 | 33 | #include <linux/gfp.h> |
1da177e4 LT |
34 | #include <net/ip.h> |
35 | #include <net/protocol.h> | |
4738c1db | 36 | #include <net/netlink.h> |
1da177e4 LT |
37 | #include <linux/skbuff.h> |
38 | #include <net/sock.h> | |
39 | #include <linux/errno.h> | |
40 | #include <linux/timer.h> | |
1da177e4 | 41 | #include <asm/uaccess.h> |
40daafc8 | 42 | #include <asm/unaligned.h> |
1da177e4 | 43 | #include <linux/filter.h> |
86e4ca66 | 44 | #include <linux/ratelimit.h> |
46b325c7 | 45 | #include <linux/seccomp.h> |
f3335031 | 46 | #include <linux/if_vlan.h> |
1da177e4 | 47 | |
43db6d65 SH |
48 | /** |
49 | * sk_filter - run a packet through a socket filter | |
50 | * @sk: sock associated with &sk_buff | |
51 | * @skb: buffer to filter | |
43db6d65 SH |
52 | * |
53 | * Run the filter code and then cut skb->data to correct size returned by | |
54 | * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller | |
55 | * than pkt_len we keep whole skb->data. This is the socket level | |
56 | * wrapper to sk_run_filter. It returns 0 if the packet should | |
57 | * be accepted or -EPERM if the packet should be tossed. | |
58 | * | |
59 | */ | |
60 | int sk_filter(struct sock *sk, struct sk_buff *skb) | |
61 | { | |
62 | int err; | |
63 | struct sk_filter *filter; | |
64 | ||
c93bdd0e MG |
65 | /* |
66 | * If the skb was allocated from pfmemalloc reserves, only | |
67 | * allow SOCK_MEMALLOC sockets to use it as this socket is | |
68 | * helping free memory | |
69 | */ | |
70 | if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) | |
71 | return -ENOMEM; | |
72 | ||
43db6d65 SH |
73 | err = security_sock_rcv_skb(sk, skb); |
74 | if (err) | |
75 | return err; | |
76 | ||
80f8f102 ED |
77 | rcu_read_lock(); |
78 | filter = rcu_dereference(sk->sk_filter); | |
43db6d65 | 79 | if (filter) { |
0a14842f | 80 | unsigned int pkt_len = SK_RUN_FILTER(filter, skb); |
0d7da9dd | 81 | |
43db6d65 SH |
82 | err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM; |
83 | } | |
80f8f102 | 84 | rcu_read_unlock(); |
43db6d65 SH |
85 | |
86 | return err; | |
87 | } | |
88 | EXPORT_SYMBOL(sk_filter); | |
89 | ||
bd4cf0ed AS |
90 | /* Helper to find the offset of pkt_type in sk_buff structure. We want |
91 | * to make sure its still a 3bit field starting at a byte boundary; | |
92 | * taken from arch/x86/net/bpf_jit_comp.c. | |
93 | */ | |
0dcceabb AS |
94 | #ifdef __BIG_ENDIAN_BITFIELD |
95 | #define PKT_TYPE_MAX (7 << 5) | |
96 | #else | |
bd4cf0ed | 97 | #define PKT_TYPE_MAX 7 |
0dcceabb | 98 | #endif |
bd4cf0ed AS |
99 | static unsigned int pkt_type_offset(void) |
100 | { | |
101 | struct sk_buff skb_probe = { .pkt_type = ~0, }; | |
102 | u8 *ct = (u8 *) &skb_probe; | |
103 | unsigned int off; | |
104 | ||
105 | for (off = 0; off < sizeof(struct sk_buff); off++) { | |
106 | if (ct[off] == PKT_TYPE_MAX) | |
107 | return off; | |
108 | } | |
109 | ||
110 | pr_err_once("Please fix %s, as pkt_type couldn't be found!\n", __func__); | |
111 | return -1; | |
112 | } | |
113 | ||
30743837 | 114 | static u64 __skb_get_pay_offset(u64 ctx, u64 a, u64 x, u64 r4, u64 r5) |
bd4cf0ed | 115 | { |
eb9672f4 | 116 | return __skb_get_poff((struct sk_buff *)(unsigned long) ctx); |
bd4cf0ed AS |
117 | } |
118 | ||
30743837 | 119 | static u64 __skb_get_nlattr(u64 ctx, u64 a, u64 x, u64 r4, u64 r5) |
bd4cf0ed | 120 | { |
eb9672f4 | 121 | struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx; |
bd4cf0ed AS |
122 | struct nlattr *nla; |
123 | ||
124 | if (skb_is_nonlinear(skb)) | |
125 | return 0; | |
126 | ||
05ab8f26 MK |
127 | if (skb->len < sizeof(struct nlattr)) |
128 | return 0; | |
129 | ||
30743837 | 130 | if (a > skb->len - sizeof(struct nlattr)) |
bd4cf0ed AS |
131 | return 0; |
132 | ||
30743837 | 133 | nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x); |
bd4cf0ed AS |
134 | if (nla) |
135 | return (void *) nla - (void *) skb->data; | |
136 | ||
137 | return 0; | |
138 | } | |
139 | ||
30743837 | 140 | static u64 __skb_get_nlattr_nest(u64 ctx, u64 a, u64 x, u64 r4, u64 r5) |
bd4cf0ed | 141 | { |
eb9672f4 | 142 | struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx; |
bd4cf0ed AS |
143 | struct nlattr *nla; |
144 | ||
145 | if (skb_is_nonlinear(skb)) | |
146 | return 0; | |
147 | ||
05ab8f26 MK |
148 | if (skb->len < sizeof(struct nlattr)) |
149 | return 0; | |
150 | ||
30743837 | 151 | if (a > skb->len - sizeof(struct nlattr)) |
bd4cf0ed AS |
152 | return 0; |
153 | ||
30743837 DB |
154 | nla = (struct nlattr *) &skb->data[a]; |
155 | if (nla->nla_len > skb->len - a) | |
bd4cf0ed AS |
156 | return 0; |
157 | ||
30743837 | 158 | nla = nla_find_nested(nla, x); |
bd4cf0ed AS |
159 | if (nla) |
160 | return (void *) nla - (void *) skb->data; | |
161 | ||
162 | return 0; | |
163 | } | |
164 | ||
30743837 | 165 | static u64 __get_raw_cpu_id(u64 ctx, u64 a, u64 x, u64 r4, u64 r5) |
bd4cf0ed AS |
166 | { |
167 | return raw_smp_processor_id(); | |
168 | } | |
169 | ||
4cd3675e | 170 | /* note that this only generates 32-bit random numbers */ |
30743837 | 171 | static u64 __get_random_u32(u64 ctx, u64 a, u64 x, u64 r4, u64 r5) |
4cd3675e | 172 | { |
eb9672f4 | 173 | return prandom_u32(); |
4cd3675e CG |
174 | } |
175 | ||
bd4cf0ed | 176 | static bool convert_bpf_extensions(struct sock_filter *fp, |
2695fb55 | 177 | struct bpf_insn **insnp) |
bd4cf0ed | 178 | { |
2695fb55 | 179 | struct bpf_insn *insn = *insnp; |
bd4cf0ed AS |
180 | |
181 | switch (fp->k) { | |
182 | case SKF_AD_OFF + SKF_AD_PROTOCOL: | |
183 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2); | |
184 | ||
e430f34e | 185 | /* A = *(u16 *) (CTX + offsetof(protocol)) */ |
f8f6d679 DB |
186 | *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, |
187 | offsetof(struct sk_buff, protocol)); | |
bd4cf0ed | 188 | /* A = ntohs(A) [emitting a nop or swap16] */ |
f8f6d679 | 189 | *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); |
bd4cf0ed AS |
190 | break; |
191 | ||
192 | case SKF_AD_OFF + SKF_AD_PKTTYPE: | |
9739eef1 AS |
193 | *insn = BPF_LDX_MEM(BPF_B, BPF_REG_A, BPF_REG_CTX, |
194 | pkt_type_offset()); | |
bd4cf0ed AS |
195 | if (insn->off < 0) |
196 | return false; | |
197 | insn++; | |
9739eef1 | 198 | *insn = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, PKT_TYPE_MAX); |
0dcceabb AS |
199 | #ifdef __BIG_ENDIAN_BITFIELD |
200 | insn++; | |
f666f87b | 201 | *insn = BPF_ALU32_IMM(BPF_RSH, BPF_REG_A, 5); |
0dcceabb | 202 | #endif |
bd4cf0ed AS |
203 | break; |
204 | ||
205 | case SKF_AD_OFF + SKF_AD_IFINDEX: | |
206 | case SKF_AD_OFF + SKF_AD_HATYPE: | |
bd4cf0ed AS |
207 | BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4); |
208 | BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2); | |
f8f6d679 DB |
209 | BUILD_BUG_ON(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)) < 0); |
210 | ||
211 | *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)), | |
212 | BPF_REG_TMP, BPF_REG_CTX, | |
213 | offsetof(struct sk_buff, dev)); | |
214 | /* if (tmp != 0) goto pc + 1 */ | |
215 | *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1); | |
216 | *insn++ = BPF_EXIT_INSN(); | |
217 | if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) | |
218 | *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP, | |
219 | offsetof(struct net_device, ifindex)); | |
220 | else | |
221 | *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP, | |
222 | offsetof(struct net_device, type)); | |
bd4cf0ed AS |
223 | break; |
224 | ||
225 | case SKF_AD_OFF + SKF_AD_MARK: | |
226 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4); | |
227 | ||
9739eef1 AS |
228 | *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, |
229 | offsetof(struct sk_buff, mark)); | |
bd4cf0ed AS |
230 | break; |
231 | ||
232 | case SKF_AD_OFF + SKF_AD_RXHASH: | |
233 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4); | |
234 | ||
9739eef1 AS |
235 | *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, |
236 | offsetof(struct sk_buff, hash)); | |
bd4cf0ed AS |
237 | break; |
238 | ||
239 | case SKF_AD_OFF + SKF_AD_QUEUE: | |
240 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2); | |
241 | ||
9739eef1 AS |
242 | *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, |
243 | offsetof(struct sk_buff, queue_mapping)); | |
bd4cf0ed AS |
244 | break; |
245 | ||
246 | case SKF_AD_OFF + SKF_AD_VLAN_TAG: | |
247 | case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT: | |
248 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2); | |
bd4cf0ed AS |
249 | BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000); |
250 | ||
e430f34e | 251 | /* A = *(u16 *) (CTX + offsetof(vlan_tci)) */ |
f8f6d679 DB |
252 | *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, |
253 | offsetof(struct sk_buff, vlan_tci)); | |
bd4cf0ed | 254 | if (fp->k == SKF_AD_OFF + SKF_AD_VLAN_TAG) { |
9739eef1 AS |
255 | *insn = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, |
256 | ~VLAN_TAG_PRESENT); | |
bd4cf0ed | 257 | } else { |
9739eef1 | 258 | /* A >>= 12 */ |
f8f6d679 | 259 | *insn++ = BPF_ALU32_IMM(BPF_RSH, BPF_REG_A, 12); |
9739eef1 AS |
260 | /* A &= 1 */ |
261 | *insn = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 1); | |
bd4cf0ed AS |
262 | } |
263 | break; | |
264 | ||
265 | case SKF_AD_OFF + SKF_AD_PAY_OFFSET: | |
266 | case SKF_AD_OFF + SKF_AD_NLATTR: | |
267 | case SKF_AD_OFF + SKF_AD_NLATTR_NEST: | |
268 | case SKF_AD_OFF + SKF_AD_CPU: | |
4cd3675e | 269 | case SKF_AD_OFF + SKF_AD_RANDOM: |
e430f34e | 270 | /* arg1 = CTX */ |
f8f6d679 | 271 | *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); |
bd4cf0ed | 272 | /* arg2 = A */ |
f8f6d679 | 273 | *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A); |
bd4cf0ed | 274 | /* arg3 = X */ |
f8f6d679 | 275 | *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X); |
e430f34e | 276 | /* Emit call(arg1=CTX, arg2=A, arg3=X) */ |
bd4cf0ed AS |
277 | switch (fp->k) { |
278 | case SKF_AD_OFF + SKF_AD_PAY_OFFSET: | |
f8f6d679 | 279 | *insn = BPF_EMIT_CALL(__skb_get_pay_offset); |
bd4cf0ed AS |
280 | break; |
281 | case SKF_AD_OFF + SKF_AD_NLATTR: | |
f8f6d679 | 282 | *insn = BPF_EMIT_CALL(__skb_get_nlattr); |
bd4cf0ed AS |
283 | break; |
284 | case SKF_AD_OFF + SKF_AD_NLATTR_NEST: | |
f8f6d679 | 285 | *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest); |
bd4cf0ed AS |
286 | break; |
287 | case SKF_AD_OFF + SKF_AD_CPU: | |
f8f6d679 | 288 | *insn = BPF_EMIT_CALL(__get_raw_cpu_id); |
bd4cf0ed | 289 | break; |
4cd3675e | 290 | case SKF_AD_OFF + SKF_AD_RANDOM: |
f8f6d679 | 291 | *insn = BPF_EMIT_CALL(__get_random_u32); |
4cd3675e | 292 | break; |
bd4cf0ed AS |
293 | } |
294 | break; | |
295 | ||
296 | case SKF_AD_OFF + SKF_AD_ALU_XOR_X: | |
9739eef1 AS |
297 | /* A ^= X */ |
298 | *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X); | |
bd4cf0ed AS |
299 | break; |
300 | ||
301 | default: | |
302 | /* This is just a dummy call to avoid letting the compiler | |
303 | * evict __bpf_call_base() as an optimization. Placed here | |
304 | * where no-one bothers. | |
305 | */ | |
306 | BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0); | |
307 | return false; | |
308 | } | |
309 | ||
310 | *insnp = insn; | |
311 | return true; | |
312 | } | |
313 | ||
314 | /** | |
315 | * sk_convert_filter - convert filter program | |
316 | * @prog: the user passed filter program | |
317 | * @len: the length of the user passed filter program | |
318 | * @new_prog: buffer where converted program will be stored | |
319 | * @new_len: pointer to store length of converted program | |
320 | * | |
321 | * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style. | |
322 | * Conversion workflow: | |
323 | * | |
324 | * 1) First pass for calculating the new program length: | |
325 | * sk_convert_filter(old_prog, old_len, NULL, &new_len) | |
326 | * | |
327 | * 2) 2nd pass to remap in two passes: 1st pass finds new | |
328 | * jump offsets, 2nd pass remapping: | |
2695fb55 | 329 | * new_prog = kmalloc(sizeof(struct bpf_insn) * new_len); |
bd4cf0ed AS |
330 | * sk_convert_filter(old_prog, old_len, new_prog, &new_len); |
331 | * | |
332 | * User BPF's register A is mapped to our BPF register 6, user BPF | |
333 | * register X is mapped to BPF register 7; frame pointer is always | |
334 | * register 10; Context 'void *ctx' is stored in register 1, that is, | |
335 | * for socket filters: ctx == 'struct sk_buff *', for seccomp: | |
336 | * ctx == 'struct seccomp_data *'. | |
337 | */ | |
338 | int sk_convert_filter(struct sock_filter *prog, int len, | |
2695fb55 | 339 | struct bpf_insn *new_prog, int *new_len) |
bd4cf0ed AS |
340 | { |
341 | int new_flen = 0, pass = 0, target, i; | |
2695fb55 | 342 | struct bpf_insn *new_insn; |
bd4cf0ed AS |
343 | struct sock_filter *fp; |
344 | int *addrs = NULL; | |
345 | u8 bpf_src; | |
346 | ||
347 | BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK); | |
30743837 | 348 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); |
bd4cf0ed | 349 | |
6f9a093b | 350 | if (len <= 0 || len > BPF_MAXINSNS) |
bd4cf0ed AS |
351 | return -EINVAL; |
352 | ||
353 | if (new_prog) { | |
99e72a0f | 354 | addrs = kcalloc(len, sizeof(*addrs), GFP_KERNEL); |
bd4cf0ed AS |
355 | if (!addrs) |
356 | return -ENOMEM; | |
357 | } | |
358 | ||
359 | do_pass: | |
360 | new_insn = new_prog; | |
361 | fp = prog; | |
362 | ||
f8f6d679 DB |
363 | if (new_insn) |
364 | *new_insn = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1); | |
bd4cf0ed AS |
365 | new_insn++; |
366 | ||
367 | for (i = 0; i < len; fp++, i++) { | |
2695fb55 AS |
368 | struct bpf_insn tmp_insns[6] = { }; |
369 | struct bpf_insn *insn = tmp_insns; | |
bd4cf0ed AS |
370 | |
371 | if (addrs) | |
372 | addrs[i] = new_insn - new_prog; | |
373 | ||
374 | switch (fp->code) { | |
375 | /* All arithmetic insns and skb loads map as-is. */ | |
376 | case BPF_ALU | BPF_ADD | BPF_X: | |
377 | case BPF_ALU | BPF_ADD | BPF_K: | |
378 | case BPF_ALU | BPF_SUB | BPF_X: | |
379 | case BPF_ALU | BPF_SUB | BPF_K: | |
380 | case BPF_ALU | BPF_AND | BPF_X: | |
381 | case BPF_ALU | BPF_AND | BPF_K: | |
382 | case BPF_ALU | BPF_OR | BPF_X: | |
383 | case BPF_ALU | BPF_OR | BPF_K: | |
384 | case BPF_ALU | BPF_LSH | BPF_X: | |
385 | case BPF_ALU | BPF_LSH | BPF_K: | |
386 | case BPF_ALU | BPF_RSH | BPF_X: | |
387 | case BPF_ALU | BPF_RSH | BPF_K: | |
388 | case BPF_ALU | BPF_XOR | BPF_X: | |
389 | case BPF_ALU | BPF_XOR | BPF_K: | |
390 | case BPF_ALU | BPF_MUL | BPF_X: | |
391 | case BPF_ALU | BPF_MUL | BPF_K: | |
392 | case BPF_ALU | BPF_DIV | BPF_X: | |
393 | case BPF_ALU | BPF_DIV | BPF_K: | |
394 | case BPF_ALU | BPF_MOD | BPF_X: | |
395 | case BPF_ALU | BPF_MOD | BPF_K: | |
396 | case BPF_ALU | BPF_NEG: | |
397 | case BPF_LD | BPF_ABS | BPF_W: | |
398 | case BPF_LD | BPF_ABS | BPF_H: | |
399 | case BPF_LD | BPF_ABS | BPF_B: | |
400 | case BPF_LD | BPF_IND | BPF_W: | |
401 | case BPF_LD | BPF_IND | BPF_H: | |
402 | case BPF_LD | BPF_IND | BPF_B: | |
403 | /* Check for overloaded BPF extension and | |
404 | * directly convert it if found, otherwise | |
405 | * just move on with mapping. | |
406 | */ | |
407 | if (BPF_CLASS(fp->code) == BPF_LD && | |
408 | BPF_MODE(fp->code) == BPF_ABS && | |
409 | convert_bpf_extensions(fp, &insn)) | |
410 | break; | |
411 | ||
f8f6d679 | 412 | *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k); |
bd4cf0ed AS |
413 | break; |
414 | ||
f8f6d679 DB |
415 | /* Jump transformation cannot use BPF block macros |
416 | * everywhere as offset calculation and target updates | |
417 | * require a bit more work than the rest, i.e. jump | |
418 | * opcodes map as-is, but offsets need adjustment. | |
419 | */ | |
420 | ||
421 | #define BPF_EMIT_JMP \ | |
bd4cf0ed AS |
422 | do { \ |
423 | if (target >= len || target < 0) \ | |
424 | goto err; \ | |
425 | insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \ | |
426 | /* Adjust pc relative offset for 2nd or 3rd insn. */ \ | |
427 | insn->off -= insn - tmp_insns; \ | |
428 | } while (0) | |
429 | ||
f8f6d679 DB |
430 | case BPF_JMP | BPF_JA: |
431 | target = i + fp->k + 1; | |
432 | insn->code = fp->code; | |
433 | BPF_EMIT_JMP; | |
bd4cf0ed AS |
434 | break; |
435 | ||
436 | case BPF_JMP | BPF_JEQ | BPF_K: | |
437 | case BPF_JMP | BPF_JEQ | BPF_X: | |
438 | case BPF_JMP | BPF_JSET | BPF_K: | |
439 | case BPF_JMP | BPF_JSET | BPF_X: | |
440 | case BPF_JMP | BPF_JGT | BPF_K: | |
441 | case BPF_JMP | BPF_JGT | BPF_X: | |
442 | case BPF_JMP | BPF_JGE | BPF_K: | |
443 | case BPF_JMP | BPF_JGE | BPF_X: | |
444 | if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) { | |
445 | /* BPF immediates are signed, zero extend | |
446 | * immediate into tmp register and use it | |
447 | * in compare insn. | |
448 | */ | |
f8f6d679 | 449 | *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k); |
bd4cf0ed | 450 | |
e430f34e AS |
451 | insn->dst_reg = BPF_REG_A; |
452 | insn->src_reg = BPF_REG_TMP; | |
bd4cf0ed AS |
453 | bpf_src = BPF_X; |
454 | } else { | |
e430f34e AS |
455 | insn->dst_reg = BPF_REG_A; |
456 | insn->src_reg = BPF_REG_X; | |
bd4cf0ed AS |
457 | insn->imm = fp->k; |
458 | bpf_src = BPF_SRC(fp->code); | |
1da177e4 | 459 | } |
bd4cf0ed AS |
460 | |
461 | /* Common case where 'jump_false' is next insn. */ | |
462 | if (fp->jf == 0) { | |
463 | insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; | |
464 | target = i + fp->jt + 1; | |
f8f6d679 | 465 | BPF_EMIT_JMP; |
bd4cf0ed | 466 | break; |
1da177e4 | 467 | } |
bd4cf0ed AS |
468 | |
469 | /* Convert JEQ into JNE when 'jump_true' is next insn. */ | |
470 | if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) { | |
471 | insn->code = BPF_JMP | BPF_JNE | bpf_src; | |
472 | target = i + fp->jf + 1; | |
f8f6d679 | 473 | BPF_EMIT_JMP; |
bd4cf0ed | 474 | break; |
0b05b2a4 | 475 | } |
bd4cf0ed AS |
476 | |
477 | /* Other jumps are mapped into two insns: Jxx and JA. */ | |
478 | target = i + fp->jt + 1; | |
479 | insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; | |
f8f6d679 | 480 | BPF_EMIT_JMP; |
bd4cf0ed AS |
481 | insn++; |
482 | ||
483 | insn->code = BPF_JMP | BPF_JA; | |
484 | target = i + fp->jf + 1; | |
f8f6d679 | 485 | BPF_EMIT_JMP; |
bd4cf0ed AS |
486 | break; |
487 | ||
488 | /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */ | |
489 | case BPF_LDX | BPF_MSH | BPF_B: | |
9739eef1 | 490 | /* tmp = A */ |
f8f6d679 | 491 | *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A); |
1268e253 | 492 | /* A = BPF_R0 = *(u8 *) (skb->data + K) */ |
f8f6d679 | 493 | *insn++ = BPF_LD_ABS(BPF_B, fp->k); |
9739eef1 | 494 | /* A &= 0xf */ |
f8f6d679 | 495 | *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf); |
9739eef1 | 496 | /* A <<= 2 */ |
f8f6d679 | 497 | *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2); |
9739eef1 | 498 | /* X = A */ |
f8f6d679 | 499 | *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); |
9739eef1 | 500 | /* A = tmp */ |
f8f6d679 | 501 | *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP); |
bd4cf0ed AS |
502 | break; |
503 | ||
504 | /* RET_K, RET_A are remaped into 2 insns. */ | |
505 | case BPF_RET | BPF_A: | |
506 | case BPF_RET | BPF_K: | |
f8f6d679 DB |
507 | *insn++ = BPF_MOV32_RAW(BPF_RVAL(fp->code) == BPF_K ? |
508 | BPF_K : BPF_X, BPF_REG_0, | |
509 | BPF_REG_A, fp->k); | |
9739eef1 | 510 | *insn = BPF_EXIT_INSN(); |
bd4cf0ed AS |
511 | break; |
512 | ||
513 | /* Store to stack. */ | |
514 | case BPF_ST: | |
515 | case BPF_STX: | |
f8f6d679 DB |
516 | *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) == |
517 | BPF_ST ? BPF_REG_A : BPF_REG_X, | |
518 | -(BPF_MEMWORDS - fp->k) * 4); | |
bd4cf0ed AS |
519 | break; |
520 | ||
521 | /* Load from stack. */ | |
522 | case BPF_LD | BPF_MEM: | |
523 | case BPF_LDX | BPF_MEM: | |
f8f6d679 DB |
524 | *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? |
525 | BPF_REG_A : BPF_REG_X, BPF_REG_FP, | |
526 | -(BPF_MEMWORDS - fp->k) * 4); | |
bd4cf0ed AS |
527 | break; |
528 | ||
529 | /* A = K or X = K */ | |
530 | case BPF_LD | BPF_IMM: | |
531 | case BPF_LDX | BPF_IMM: | |
f8f6d679 DB |
532 | *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ? |
533 | BPF_REG_A : BPF_REG_X, fp->k); | |
bd4cf0ed AS |
534 | break; |
535 | ||
536 | /* X = A */ | |
537 | case BPF_MISC | BPF_TAX: | |
f8f6d679 | 538 | *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); |
bd4cf0ed AS |
539 | break; |
540 | ||
541 | /* A = X */ | |
542 | case BPF_MISC | BPF_TXA: | |
f8f6d679 | 543 | *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X); |
bd4cf0ed AS |
544 | break; |
545 | ||
546 | /* A = skb->len or X = skb->len */ | |
547 | case BPF_LD | BPF_W | BPF_LEN: | |
548 | case BPF_LDX | BPF_W | BPF_LEN: | |
f8f6d679 DB |
549 | *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? |
550 | BPF_REG_A : BPF_REG_X, BPF_REG_CTX, | |
551 | offsetof(struct sk_buff, len)); | |
bd4cf0ed AS |
552 | break; |
553 | ||
f8f6d679 | 554 | /* Access seccomp_data fields. */ |
bd4cf0ed | 555 | case BPF_LDX | BPF_ABS | BPF_W: |
9739eef1 AS |
556 | /* A = *(u32 *) (ctx + K) */ |
557 | *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k); | |
bd4cf0ed AS |
558 | break; |
559 | ||
f8f6d679 | 560 | /* Unkown instruction. */ |
1da177e4 | 561 | default: |
bd4cf0ed | 562 | goto err; |
1da177e4 | 563 | } |
bd4cf0ed AS |
564 | |
565 | insn++; | |
566 | if (new_prog) | |
567 | memcpy(new_insn, tmp_insns, | |
568 | sizeof(*insn) * (insn - tmp_insns)); | |
bd4cf0ed | 569 | new_insn += insn - tmp_insns; |
1da177e4 LT |
570 | } |
571 | ||
bd4cf0ed AS |
572 | if (!new_prog) { |
573 | /* Only calculating new length. */ | |
574 | *new_len = new_insn - new_prog; | |
575 | return 0; | |
576 | } | |
577 | ||
578 | pass++; | |
579 | if (new_flen != new_insn - new_prog) { | |
580 | new_flen = new_insn - new_prog; | |
581 | if (pass > 2) | |
582 | goto err; | |
bd4cf0ed AS |
583 | goto do_pass; |
584 | } | |
585 | ||
586 | kfree(addrs); | |
587 | BUG_ON(*new_len != new_flen); | |
1da177e4 | 588 | return 0; |
bd4cf0ed AS |
589 | err: |
590 | kfree(addrs); | |
591 | return -EINVAL; | |
1da177e4 LT |
592 | } |
593 | ||
bd4cf0ed AS |
594 | /* Security: |
595 | * | |
2d5311e4 | 596 | * A BPF program is able to use 16 cells of memory to store intermediate |
bd4cf0ed AS |
597 | * values (check u32 mem[BPF_MEMWORDS] in sk_run_filter()). |
598 | * | |
2d5311e4 ED |
599 | * As we dont want to clear mem[] array for each packet going through |
600 | * sk_run_filter(), we check that filter loaded by user never try to read | |
601 | * a cell if not previously written, and we check all branches to be sure | |
25985edc | 602 | * a malicious user doesn't try to abuse us. |
2d5311e4 | 603 | */ |
ec31a05c | 604 | static int check_load_and_stores(const struct sock_filter *filter, int flen) |
2d5311e4 | 605 | { |
34805931 | 606 | u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */ |
2d5311e4 ED |
607 | int pc, ret = 0; |
608 | ||
609 | BUILD_BUG_ON(BPF_MEMWORDS > 16); | |
34805931 | 610 | |
99e72a0f | 611 | masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL); |
2d5311e4 ED |
612 | if (!masks) |
613 | return -ENOMEM; | |
34805931 | 614 | |
2d5311e4 ED |
615 | memset(masks, 0xff, flen * sizeof(*masks)); |
616 | ||
617 | for (pc = 0; pc < flen; pc++) { | |
618 | memvalid &= masks[pc]; | |
619 | ||
620 | switch (filter[pc].code) { | |
34805931 DB |
621 | case BPF_ST: |
622 | case BPF_STX: | |
2d5311e4 ED |
623 | memvalid |= (1 << filter[pc].k); |
624 | break; | |
34805931 DB |
625 | case BPF_LD | BPF_MEM: |
626 | case BPF_LDX | BPF_MEM: | |
2d5311e4 ED |
627 | if (!(memvalid & (1 << filter[pc].k))) { |
628 | ret = -EINVAL; | |
629 | goto error; | |
630 | } | |
631 | break; | |
34805931 DB |
632 | case BPF_JMP | BPF_JA: |
633 | /* A jump must set masks on target */ | |
2d5311e4 ED |
634 | masks[pc + 1 + filter[pc].k] &= memvalid; |
635 | memvalid = ~0; | |
636 | break; | |
34805931 DB |
637 | case BPF_JMP | BPF_JEQ | BPF_K: |
638 | case BPF_JMP | BPF_JEQ | BPF_X: | |
639 | case BPF_JMP | BPF_JGE | BPF_K: | |
640 | case BPF_JMP | BPF_JGE | BPF_X: | |
641 | case BPF_JMP | BPF_JGT | BPF_K: | |
642 | case BPF_JMP | BPF_JGT | BPF_X: | |
643 | case BPF_JMP | BPF_JSET | BPF_K: | |
644 | case BPF_JMP | BPF_JSET | BPF_X: | |
645 | /* A jump must set masks on targets */ | |
2d5311e4 ED |
646 | masks[pc + 1 + filter[pc].jt] &= memvalid; |
647 | masks[pc + 1 + filter[pc].jf] &= memvalid; | |
648 | memvalid = ~0; | |
649 | break; | |
650 | } | |
651 | } | |
652 | error: | |
653 | kfree(masks); | |
654 | return ret; | |
655 | } | |
656 | ||
34805931 DB |
657 | static bool chk_code_allowed(u16 code_to_probe) |
658 | { | |
659 | static const bool codes[] = { | |
660 | /* 32 bit ALU operations */ | |
661 | [BPF_ALU | BPF_ADD | BPF_K] = true, | |
662 | [BPF_ALU | BPF_ADD | BPF_X] = true, | |
663 | [BPF_ALU | BPF_SUB | BPF_K] = true, | |
664 | [BPF_ALU | BPF_SUB | BPF_X] = true, | |
665 | [BPF_ALU | BPF_MUL | BPF_K] = true, | |
666 | [BPF_ALU | BPF_MUL | BPF_X] = true, | |
667 | [BPF_ALU | BPF_DIV | BPF_K] = true, | |
668 | [BPF_ALU | BPF_DIV | BPF_X] = true, | |
669 | [BPF_ALU | BPF_MOD | BPF_K] = true, | |
670 | [BPF_ALU | BPF_MOD | BPF_X] = true, | |
671 | [BPF_ALU | BPF_AND | BPF_K] = true, | |
672 | [BPF_ALU | BPF_AND | BPF_X] = true, | |
673 | [BPF_ALU | BPF_OR | BPF_K] = true, | |
674 | [BPF_ALU | BPF_OR | BPF_X] = true, | |
675 | [BPF_ALU | BPF_XOR | BPF_K] = true, | |
676 | [BPF_ALU | BPF_XOR | BPF_X] = true, | |
677 | [BPF_ALU | BPF_LSH | BPF_K] = true, | |
678 | [BPF_ALU | BPF_LSH | BPF_X] = true, | |
679 | [BPF_ALU | BPF_RSH | BPF_K] = true, | |
680 | [BPF_ALU | BPF_RSH | BPF_X] = true, | |
681 | [BPF_ALU | BPF_NEG] = true, | |
682 | /* Load instructions */ | |
683 | [BPF_LD | BPF_W | BPF_ABS] = true, | |
684 | [BPF_LD | BPF_H | BPF_ABS] = true, | |
685 | [BPF_LD | BPF_B | BPF_ABS] = true, | |
686 | [BPF_LD | BPF_W | BPF_LEN] = true, | |
687 | [BPF_LD | BPF_W | BPF_IND] = true, | |
688 | [BPF_LD | BPF_H | BPF_IND] = true, | |
689 | [BPF_LD | BPF_B | BPF_IND] = true, | |
690 | [BPF_LD | BPF_IMM] = true, | |
691 | [BPF_LD | BPF_MEM] = true, | |
692 | [BPF_LDX | BPF_W | BPF_LEN] = true, | |
693 | [BPF_LDX | BPF_B | BPF_MSH] = true, | |
694 | [BPF_LDX | BPF_IMM] = true, | |
695 | [BPF_LDX | BPF_MEM] = true, | |
696 | /* Store instructions */ | |
697 | [BPF_ST] = true, | |
698 | [BPF_STX] = true, | |
699 | /* Misc instructions */ | |
700 | [BPF_MISC | BPF_TAX] = true, | |
701 | [BPF_MISC | BPF_TXA] = true, | |
702 | /* Return instructions */ | |
703 | [BPF_RET | BPF_K] = true, | |
704 | [BPF_RET | BPF_A] = true, | |
705 | /* Jump instructions */ | |
706 | [BPF_JMP | BPF_JA] = true, | |
707 | [BPF_JMP | BPF_JEQ | BPF_K] = true, | |
708 | [BPF_JMP | BPF_JEQ | BPF_X] = true, | |
709 | [BPF_JMP | BPF_JGE | BPF_K] = true, | |
710 | [BPF_JMP | BPF_JGE | BPF_X] = true, | |
711 | [BPF_JMP | BPF_JGT | BPF_K] = true, | |
712 | [BPF_JMP | BPF_JGT | BPF_X] = true, | |
713 | [BPF_JMP | BPF_JSET | BPF_K] = true, | |
714 | [BPF_JMP | BPF_JSET | BPF_X] = true, | |
715 | }; | |
716 | ||
717 | if (code_to_probe >= ARRAY_SIZE(codes)) | |
718 | return false; | |
719 | ||
720 | return codes[code_to_probe]; | |
721 | } | |
722 | ||
1da177e4 LT |
723 | /** |
724 | * sk_chk_filter - verify socket filter code | |
725 | * @filter: filter to verify | |
726 | * @flen: length of filter | |
727 | * | |
728 | * Check the user's filter code. If we let some ugly | |
729 | * filter code slip through kaboom! The filter must contain | |
93699863 KK |
730 | * no references or jumps that are out of range, no illegal |
731 | * instructions, and must end with a RET instruction. | |
1da177e4 | 732 | * |
7b11f69f KK |
733 | * All jumps are forward as they are not signed. |
734 | * | |
735 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
1da177e4 | 736 | */ |
ec31a05c | 737 | int sk_chk_filter(const struct sock_filter *filter, unsigned int flen) |
1da177e4 | 738 | { |
aa1113d9 | 739 | bool anc_found; |
34805931 | 740 | int pc; |
1da177e4 | 741 | |
1b93ae64 | 742 | if (flen == 0 || flen > BPF_MAXINSNS) |
1da177e4 LT |
743 | return -EINVAL; |
744 | ||
34805931 | 745 | /* Check the filter code now */ |
1da177e4 | 746 | for (pc = 0; pc < flen; pc++) { |
ec31a05c | 747 | const struct sock_filter *ftest = &filter[pc]; |
93699863 | 748 | |
34805931 DB |
749 | /* May we actually operate on this code? */ |
750 | if (!chk_code_allowed(ftest->code)) | |
cba328fc | 751 | return -EINVAL; |
34805931 | 752 | |
93699863 | 753 | /* Some instructions need special checks */ |
34805931 DB |
754 | switch (ftest->code) { |
755 | case BPF_ALU | BPF_DIV | BPF_K: | |
756 | case BPF_ALU | BPF_MOD | BPF_K: | |
757 | /* Check for division by zero */ | |
b6069a95 ED |
758 | if (ftest->k == 0) |
759 | return -EINVAL; | |
760 | break; | |
34805931 DB |
761 | case BPF_LD | BPF_MEM: |
762 | case BPF_LDX | BPF_MEM: | |
763 | case BPF_ST: | |
764 | case BPF_STX: | |
765 | /* Check for invalid memory addresses */ | |
93699863 KK |
766 | if (ftest->k >= BPF_MEMWORDS) |
767 | return -EINVAL; | |
768 | break; | |
34805931 DB |
769 | case BPF_JMP | BPF_JA: |
770 | /* Note, the large ftest->k might cause loops. | |
93699863 KK |
771 | * Compare this with conditional jumps below, |
772 | * where offsets are limited. --ANK (981016) | |
773 | */ | |
34805931 | 774 | if (ftest->k >= (unsigned int)(flen - pc - 1)) |
93699863 | 775 | return -EINVAL; |
01f2f3f6 | 776 | break; |
34805931 DB |
777 | case BPF_JMP | BPF_JEQ | BPF_K: |
778 | case BPF_JMP | BPF_JEQ | BPF_X: | |
779 | case BPF_JMP | BPF_JGE | BPF_K: | |
780 | case BPF_JMP | BPF_JGE | BPF_X: | |
781 | case BPF_JMP | BPF_JGT | BPF_K: | |
782 | case BPF_JMP | BPF_JGT | BPF_X: | |
783 | case BPF_JMP | BPF_JSET | BPF_K: | |
784 | case BPF_JMP | BPF_JSET | BPF_X: | |
785 | /* Both conditionals must be safe */ | |
e35bedf3 | 786 | if (pc + ftest->jt + 1 >= flen || |
93699863 KK |
787 | pc + ftest->jf + 1 >= flen) |
788 | return -EINVAL; | |
cba328fc | 789 | break; |
34805931 DB |
790 | case BPF_LD | BPF_W | BPF_ABS: |
791 | case BPF_LD | BPF_H | BPF_ABS: | |
792 | case BPF_LD | BPF_B | BPF_ABS: | |
aa1113d9 | 793 | anc_found = false; |
34805931 DB |
794 | if (bpf_anc_helper(ftest) & BPF_ANC) |
795 | anc_found = true; | |
796 | /* Ancillary operation unknown or unsupported */ | |
aa1113d9 DB |
797 | if (anc_found == false && ftest->k >= SKF_AD_OFF) |
798 | return -EINVAL; | |
01f2f3f6 HPP |
799 | } |
800 | } | |
93699863 | 801 | |
34805931 | 802 | /* Last instruction must be a RET code */ |
01f2f3f6 | 803 | switch (filter[flen - 1].code) { |
34805931 DB |
804 | case BPF_RET | BPF_K: |
805 | case BPF_RET | BPF_A: | |
2d5311e4 | 806 | return check_load_and_stores(filter, flen); |
cba328fc | 807 | } |
34805931 | 808 | |
cba328fc | 809 | return -EINVAL; |
1da177e4 | 810 | } |
b715631f | 811 | EXPORT_SYMBOL(sk_chk_filter); |
1da177e4 | 812 | |
a3ea269b DB |
813 | static int sk_store_orig_filter(struct sk_filter *fp, |
814 | const struct sock_fprog *fprog) | |
815 | { | |
816 | unsigned int fsize = sk_filter_proglen(fprog); | |
817 | struct sock_fprog_kern *fkprog; | |
818 | ||
819 | fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL); | |
820 | if (!fp->orig_prog) | |
821 | return -ENOMEM; | |
822 | ||
823 | fkprog = fp->orig_prog; | |
824 | fkprog->len = fprog->len; | |
825 | fkprog->filter = kmemdup(fp->insns, fsize, GFP_KERNEL); | |
826 | if (!fkprog->filter) { | |
827 | kfree(fp->orig_prog); | |
828 | return -ENOMEM; | |
829 | } | |
830 | ||
831 | return 0; | |
832 | } | |
833 | ||
834 | static void sk_release_orig_filter(struct sk_filter *fp) | |
835 | { | |
836 | struct sock_fprog_kern *fprog = fp->orig_prog; | |
837 | ||
838 | if (fprog) { | |
839 | kfree(fprog->filter); | |
840 | kfree(fprog); | |
841 | } | |
842 | } | |
843 | ||
47e958ea | 844 | /** |
46bcf14f | 845 | * sk_filter_release_rcu - Release a socket filter by rcu_head |
47e958ea PE |
846 | * @rcu: rcu_head that contains the sk_filter to free |
847 | */ | |
fbc907f0 | 848 | static void sk_filter_release_rcu(struct rcu_head *rcu) |
47e958ea PE |
849 | { |
850 | struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu); | |
851 | ||
a3ea269b | 852 | sk_release_orig_filter(fp); |
5fe821a9 | 853 | sk_filter_free(fp); |
47e958ea | 854 | } |
fbc907f0 DB |
855 | |
856 | /** | |
857 | * sk_filter_release - release a socket filter | |
858 | * @fp: filter to remove | |
859 | * | |
860 | * Remove a filter from a socket and release its resources. | |
861 | */ | |
862 | static void sk_filter_release(struct sk_filter *fp) | |
863 | { | |
864 | if (atomic_dec_and_test(&fp->refcnt)) | |
865 | call_rcu(&fp->rcu, sk_filter_release_rcu); | |
866 | } | |
867 | ||
868 | void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) | |
869 | { | |
870 | atomic_sub(sk_filter_size(fp->len), &sk->sk_omem_alloc); | |
871 | sk_filter_release(fp); | |
872 | } | |
873 | ||
874 | void sk_filter_charge(struct sock *sk, struct sk_filter *fp) | |
875 | { | |
876 | atomic_inc(&fp->refcnt); | |
877 | atomic_add(sk_filter_size(fp->len), &sk->sk_omem_alloc); | |
878 | } | |
47e958ea | 879 | |
bd4cf0ed AS |
880 | static struct sk_filter *__sk_migrate_realloc(struct sk_filter *fp, |
881 | struct sock *sk, | |
882 | unsigned int len) | |
883 | { | |
884 | struct sk_filter *fp_new; | |
885 | ||
886 | if (sk == NULL) | |
887 | return krealloc(fp, len, GFP_KERNEL); | |
888 | ||
889 | fp_new = sock_kmalloc(sk, len, GFP_KERNEL); | |
890 | if (fp_new) { | |
eb9672f4 | 891 | *fp_new = *fp; |
285276e7 | 892 | /* As we're keeping orig_prog in fp_new along, |
bd4cf0ed AS |
893 | * we need to make sure we're not evicting it |
894 | * from the old fp. | |
895 | */ | |
896 | fp->orig_prog = NULL; | |
897 | sk_filter_uncharge(sk, fp); | |
898 | } | |
899 | ||
900 | return fp_new; | |
901 | } | |
902 | ||
903 | static struct sk_filter *__sk_migrate_filter(struct sk_filter *fp, | |
904 | struct sock *sk) | |
905 | { | |
906 | struct sock_filter *old_prog; | |
907 | struct sk_filter *old_fp; | |
34805931 | 908 | int err, new_len, old_len = fp->len; |
bd4cf0ed AS |
909 | |
910 | /* We are free to overwrite insns et al right here as it | |
911 | * won't be used at this point in time anymore internally | |
912 | * after the migration to the internal BPF instruction | |
913 | * representation. | |
914 | */ | |
915 | BUILD_BUG_ON(sizeof(struct sock_filter) != | |
2695fb55 | 916 | sizeof(struct bpf_insn)); |
bd4cf0ed | 917 | |
bd4cf0ed AS |
918 | /* Conversion cannot happen on overlapping memory areas, |
919 | * so we need to keep the user BPF around until the 2nd | |
920 | * pass. At this time, the user BPF is stored in fp->insns. | |
921 | */ | |
922 | old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter), | |
923 | GFP_KERNEL); | |
924 | if (!old_prog) { | |
925 | err = -ENOMEM; | |
926 | goto out_err; | |
927 | } | |
928 | ||
929 | /* 1st pass: calculate the new program length. */ | |
930 | err = sk_convert_filter(old_prog, old_len, NULL, &new_len); | |
931 | if (err) | |
932 | goto out_err_free; | |
933 | ||
934 | /* Expand fp for appending the new filter representation. */ | |
935 | old_fp = fp; | |
936 | fp = __sk_migrate_realloc(old_fp, sk, sk_filter_size(new_len)); | |
937 | if (!fp) { | |
938 | /* The old_fp is still around in case we couldn't | |
939 | * allocate new memory, so uncharge on that one. | |
940 | */ | |
941 | fp = old_fp; | |
942 | err = -ENOMEM; | |
943 | goto out_err_free; | |
944 | } | |
945 | ||
bd4cf0ed AS |
946 | fp->len = new_len; |
947 | ||
2695fb55 | 948 | /* 2nd pass: remap sock_filter insns into bpf_insn insns. */ |
bd4cf0ed AS |
949 | err = sk_convert_filter(old_prog, old_len, fp->insnsi, &new_len); |
950 | if (err) | |
951 | /* 2nd sk_convert_filter() can fail only if it fails | |
952 | * to allocate memory, remapping must succeed. Note, | |
953 | * that at this time old_fp has already been released | |
954 | * by __sk_migrate_realloc(). | |
955 | */ | |
956 | goto out_err_free; | |
957 | ||
5fe821a9 AS |
958 | sk_filter_select_runtime(fp); |
959 | ||
bd4cf0ed AS |
960 | kfree(old_prog); |
961 | return fp; | |
962 | ||
963 | out_err_free: | |
964 | kfree(old_prog); | |
965 | out_err: | |
966 | /* Rollback filter setup. */ | |
967 | if (sk != NULL) | |
968 | sk_filter_uncharge(sk, fp); | |
969 | else | |
970 | kfree(fp); | |
971 | return ERR_PTR(err); | |
972 | } | |
973 | ||
974 | static struct sk_filter *__sk_prepare_filter(struct sk_filter *fp, | |
975 | struct sock *sk) | |
302d6637 JP |
976 | { |
977 | int err; | |
978 | ||
bd4cf0ed | 979 | fp->bpf_func = NULL; |
f8bbbfc3 | 980 | fp->jited = 0; |
302d6637 JP |
981 | |
982 | err = sk_chk_filter(fp->insns, fp->len); | |
418c96ac LY |
983 | if (err) { |
984 | if (sk != NULL) | |
985 | sk_filter_uncharge(sk, fp); | |
986 | else | |
987 | kfree(fp); | |
bd4cf0ed | 988 | return ERR_PTR(err); |
418c96ac | 989 | } |
302d6637 | 990 | |
bd4cf0ed AS |
991 | /* Probe if we can JIT compile the filter and if so, do |
992 | * the compilation of the filter. | |
993 | */ | |
302d6637 | 994 | bpf_jit_compile(fp); |
bd4cf0ed AS |
995 | |
996 | /* JIT compiler couldn't process this filter, so do the | |
997 | * internal BPF translation for the optimized interpreter. | |
998 | */ | |
5fe821a9 | 999 | if (!fp->jited) |
bd4cf0ed AS |
1000 | fp = __sk_migrate_filter(fp, sk); |
1001 | ||
1002 | return fp; | |
302d6637 JP |
1003 | } |
1004 | ||
1005 | /** | |
1006 | * sk_unattached_filter_create - create an unattached filter | |
c6c4b97c | 1007 | * @pfp: the unattached filter that is created |
677a9fd3 | 1008 | * @fprog: the filter program |
302d6637 | 1009 | * |
c6c4b97c | 1010 | * Create a filter independent of any socket. We first run some |
302d6637 JP |
1011 | * sanity checks on it to make sure it does not explode on us later. |
1012 | * If an error occurs or there is insufficient memory for the filter | |
1013 | * a negative errno code is returned. On success the return is zero. | |
1014 | */ | |
1015 | int sk_unattached_filter_create(struct sk_filter **pfp, | |
b1fcd35c | 1016 | struct sock_fprog_kern *fprog) |
302d6637 | 1017 | { |
a3ea269b | 1018 | unsigned int fsize = sk_filter_proglen(fprog); |
302d6637 | 1019 | struct sk_filter *fp; |
302d6637 JP |
1020 | |
1021 | /* Make sure new filter is there and in the right amounts. */ | |
1022 | if (fprog->filter == NULL) | |
1023 | return -EINVAL; | |
1024 | ||
d45ed4a4 | 1025 | fp = kmalloc(sk_filter_size(fprog->len), GFP_KERNEL); |
302d6637 JP |
1026 | if (!fp) |
1027 | return -ENOMEM; | |
a3ea269b | 1028 | |
302d6637 JP |
1029 | memcpy(fp->insns, fprog->filter, fsize); |
1030 | ||
1031 | atomic_set(&fp->refcnt, 1); | |
1032 | fp->len = fprog->len; | |
a3ea269b DB |
1033 | /* Since unattached filters are not copied back to user |
1034 | * space through sk_get_filter(), we do not need to hold | |
1035 | * a copy here, and can spare us the work. | |
1036 | */ | |
1037 | fp->orig_prog = NULL; | |
302d6637 | 1038 | |
bd4cf0ed AS |
1039 | /* __sk_prepare_filter() already takes care of uncharging |
1040 | * memory in case something goes wrong. | |
1041 | */ | |
1042 | fp = __sk_prepare_filter(fp, NULL); | |
1043 | if (IS_ERR(fp)) | |
1044 | return PTR_ERR(fp); | |
302d6637 JP |
1045 | |
1046 | *pfp = fp; | |
1047 | return 0; | |
302d6637 JP |
1048 | } |
1049 | EXPORT_SYMBOL_GPL(sk_unattached_filter_create); | |
1050 | ||
1051 | void sk_unattached_filter_destroy(struct sk_filter *fp) | |
1052 | { | |
1053 | sk_filter_release(fp); | |
1054 | } | |
1055 | EXPORT_SYMBOL_GPL(sk_unattached_filter_destroy); | |
1056 | ||
1da177e4 LT |
1057 | /** |
1058 | * sk_attach_filter - attach a socket filter | |
1059 | * @fprog: the filter program | |
1060 | * @sk: the socket to use | |
1061 | * | |
1062 | * Attach the user's filter code. We first run some sanity checks on | |
1063 | * it to make sure it does not explode on us later. If an error | |
1064 | * occurs or there is insufficient memory for the filter a negative | |
1065 | * errno code is returned. On success the return is zero. | |
1066 | */ | |
1067 | int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk) | |
1068 | { | |
d3904b73 | 1069 | struct sk_filter *fp, *old_fp; |
a3ea269b | 1070 | unsigned int fsize = sk_filter_proglen(fprog); |
d45ed4a4 | 1071 | unsigned int sk_fsize = sk_filter_size(fprog->len); |
1da177e4 LT |
1072 | int err; |
1073 | ||
d59577b6 VB |
1074 | if (sock_flag(sk, SOCK_FILTER_LOCKED)) |
1075 | return -EPERM; | |
1076 | ||
1da177e4 | 1077 | /* Make sure new filter is there and in the right amounts. */ |
e35bedf3 KK |
1078 | if (fprog->filter == NULL) |
1079 | return -EINVAL; | |
1da177e4 | 1080 | |
d45ed4a4 | 1081 | fp = sock_kmalloc(sk, sk_fsize, GFP_KERNEL); |
1da177e4 LT |
1082 | if (!fp) |
1083 | return -ENOMEM; | |
a3ea269b | 1084 | |
1da177e4 | 1085 | if (copy_from_user(fp->insns, fprog->filter, fsize)) { |
d45ed4a4 | 1086 | sock_kfree_s(sk, fp, sk_fsize); |
1da177e4 LT |
1087 | return -EFAULT; |
1088 | } | |
1089 | ||
1090 | atomic_set(&fp->refcnt, 1); | |
1091 | fp->len = fprog->len; | |
1092 | ||
a3ea269b DB |
1093 | err = sk_store_orig_filter(fp, fprog); |
1094 | if (err) { | |
1095 | sk_filter_uncharge(sk, fp); | |
1096 | return -ENOMEM; | |
1097 | } | |
1098 | ||
bd4cf0ed AS |
1099 | /* __sk_prepare_filter() already takes care of uncharging |
1100 | * memory in case something goes wrong. | |
1101 | */ | |
1102 | fp = __sk_prepare_filter(fp, sk); | |
1103 | if (IS_ERR(fp)) | |
1104 | return PTR_ERR(fp); | |
1da177e4 | 1105 | |
f91ff5b9 ED |
1106 | old_fp = rcu_dereference_protected(sk->sk_filter, |
1107 | sock_owned_by_user(sk)); | |
d3904b73 | 1108 | rcu_assign_pointer(sk->sk_filter, fp); |
d3904b73 | 1109 | |
9b013e05 | 1110 | if (old_fp) |
46bcf14f | 1111 | sk_filter_uncharge(sk, old_fp); |
a3ea269b | 1112 | |
d3904b73 | 1113 | return 0; |
1da177e4 | 1114 | } |
5ff3f073 | 1115 | EXPORT_SYMBOL_GPL(sk_attach_filter); |
1da177e4 | 1116 | |
55b33325 PE |
1117 | int sk_detach_filter(struct sock *sk) |
1118 | { | |
1119 | int ret = -ENOENT; | |
1120 | struct sk_filter *filter; | |
1121 | ||
d59577b6 VB |
1122 | if (sock_flag(sk, SOCK_FILTER_LOCKED)) |
1123 | return -EPERM; | |
1124 | ||
f91ff5b9 ED |
1125 | filter = rcu_dereference_protected(sk->sk_filter, |
1126 | sock_owned_by_user(sk)); | |
55b33325 | 1127 | if (filter) { |
a9b3cd7f | 1128 | RCU_INIT_POINTER(sk->sk_filter, NULL); |
46bcf14f | 1129 | sk_filter_uncharge(sk, filter); |
55b33325 PE |
1130 | ret = 0; |
1131 | } | |
a3ea269b | 1132 | |
55b33325 PE |
1133 | return ret; |
1134 | } | |
5ff3f073 | 1135 | EXPORT_SYMBOL_GPL(sk_detach_filter); |
a8fc9277 | 1136 | |
a3ea269b DB |
1137 | int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf, |
1138 | unsigned int len) | |
a8fc9277 | 1139 | { |
a3ea269b | 1140 | struct sock_fprog_kern *fprog; |
a8fc9277 | 1141 | struct sk_filter *filter; |
a3ea269b | 1142 | int ret = 0; |
a8fc9277 PE |
1143 | |
1144 | lock_sock(sk); | |
1145 | filter = rcu_dereference_protected(sk->sk_filter, | |
a3ea269b | 1146 | sock_owned_by_user(sk)); |
a8fc9277 PE |
1147 | if (!filter) |
1148 | goto out; | |
a3ea269b DB |
1149 | |
1150 | /* We're copying the filter that has been originally attached, | |
1151 | * so no conversion/decode needed anymore. | |
1152 | */ | |
1153 | fprog = filter->orig_prog; | |
1154 | ||
1155 | ret = fprog->len; | |
a8fc9277 | 1156 | if (!len) |
a3ea269b | 1157 | /* User space only enquires number of filter blocks. */ |
a8fc9277 | 1158 | goto out; |
a3ea269b | 1159 | |
a8fc9277 | 1160 | ret = -EINVAL; |
a3ea269b | 1161 | if (len < fprog->len) |
a8fc9277 PE |
1162 | goto out; |
1163 | ||
1164 | ret = -EFAULT; | |
a3ea269b DB |
1165 | if (copy_to_user(ubuf, fprog->filter, sk_filter_proglen(fprog))) |
1166 | goto out; | |
a8fc9277 | 1167 | |
a3ea269b DB |
1168 | /* Instead of bytes, the API requests to return the number |
1169 | * of filter blocks. | |
1170 | */ | |
1171 | ret = fprog->len; | |
a8fc9277 PE |
1172 | out: |
1173 | release_sock(sk); | |
1174 | return ret; | |
1175 | } |