projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'vfio-v4.8-rc1' of git://github.com/awilliam/linux-vfio
[deliverable/linux.git] / net / core / filter.c
CommitLineData
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.
4df95ff4 21 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
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>
10b89ee4 39#include <net/flow_dissector.h>
1da177e4
LT		 40#include <linux/errno.h>
41#include <linux/timer.h>
1da177e4 42#include <asm/uaccess.h>
40daafc8 43#include <asm/unaligned.h>
1da177e4 44#include <linux/filter.h>
86e4ca66 45#include <linux/ratelimit.h>
46b325c7 46#include <linux/seccomp.h>
f3335031 47#include <linux/if_vlan.h>
89aa0758 48#include <linux/bpf.h>
d691f9e8 49#include <net/sch_generic.h>
8d20aabe 50#include <net/cls_cgroup.h>
d3aa45ce 51#include <net/dst_metadata.h>
c46646d0 52#include <net/dst.h>
538950a1 53#include <net/sock_reuseport.h>
1da177e4 54
43db6d65 55/**
f4979fce 56 * sk_filter_trim_cap - run a packet through a socket filter
43db6d65
SH		 57 * @sk: sock associated with &sk_buff
58 * @skb: buffer to filter
f4979fce 59 * @cap: limit on how short the eBPF program may trim the packet
43db6d65 60 *
ff936a04
AS		 61 * Run the eBPF program and then cut skb->data to correct size returned by
62 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
43db6d65 63 * than pkt_len we keep whole skb->data. This is the socket level
ff936a04 64 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
43db6d65
SH		 65 * be accepted or -EPERM if the packet should be tossed.
66 *
67 */
f4979fce 68int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
43db6d65
SH		 69{
70 int err;
71 struct sk_filter *filter;
72
c93bdd0e
MG		 73 /*
74 * If the skb was allocated from pfmemalloc reserves, only
75 * allow SOCK_MEMALLOC sockets to use it as this socket is
76 * helping free memory
77 */
78 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
79 return -ENOMEM;
80
43db6d65
SH		 81 err = security_sock_rcv_skb(sk, skb);
82 if (err)
83 return err;
84
80f8f102
ED		 85 rcu_read_lock();
86 filter = rcu_dereference(sk->sk_filter);
43db6d65 87 if (filter) {
ff936a04 88 unsigned int pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
f4979fce 89 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
43db6d65 90 }
80f8f102 91 rcu_read_unlock();
43db6d65
SH		 92
93 return err;
94}
f4979fce 95EXPORT_SYMBOL(sk_filter_trim_cap);
43db6d65 96
30743837 97static u64 __skb_get_pay_offset(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
bd4cf0ed 98{
56193d1b 99 return skb_get_poff((struct sk_buff *)(unsigned long) ctx);
bd4cf0ed
AS		 100}
101
30743837 102static u64 __skb_get_nlattr(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
bd4cf0ed 103{
eb9672f4 104 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
bd4cf0ed
AS		 105 struct nlattr *nla;
106
107 if (skb_is_nonlinear(skb))
108 return 0;
109
05ab8f26
MK		 110 if (skb->len < sizeof(struct nlattr))
111 return 0;
112
30743837 113 if (a > skb->len - sizeof(struct nlattr))
bd4cf0ed
AS		 114 return 0;
115
30743837 116 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
bd4cf0ed
AS		 117 if (nla)
118 return (void *) nla - (void *) skb->data;
119
120 return 0;
121}
122
30743837 123static u64 __skb_get_nlattr_nest(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
bd4cf0ed 124{
eb9672f4 125 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
bd4cf0ed
AS		 126 struct nlattr *nla;
127
128 if (skb_is_nonlinear(skb))
129 return 0;
130
05ab8f26
MK		 131 if (skb->len < sizeof(struct nlattr))
132 return 0;
133
30743837 134 if (a > skb->len - sizeof(struct nlattr))
bd4cf0ed
AS		 135 return 0;
136
30743837
DB		 137 nla = (struct nlattr *) &skb->data[a];
138 if (nla->nla_len > skb->len - a)
bd4cf0ed
AS		 139 return 0;
140
30743837 141 nla = nla_find_nested(nla, x);
bd4cf0ed
AS		 142 if (nla)
143 return (void *) nla - (void *) skb->data;
144
145 return 0;
146}
147
30743837 148static u64 __get_raw_cpu_id(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
bd4cf0ed
AS		 149{
150 return raw_smp_processor_id();
151}
152
80b48c44
DB		 153static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
154 .func = __get_raw_cpu_id,
155 .gpl_only = false,
156 .ret_type = RET_INTEGER,
157};
158
9bac3d6d
AS		 159static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
160 struct bpf_insn *insn_buf)
161{
162 struct bpf_insn *insn = insn_buf;
163
164 switch (skb_field) {
165 case SKF_AD_MARK:
166 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
167
168 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
169 offsetof(struct sk_buff, mark));
170 break;
171
172 case SKF_AD_PKTTYPE:
173 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
174 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
175#ifdef __BIG_ENDIAN_BITFIELD
176 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
177#endif
178 break;
179
180 case SKF_AD_QUEUE:
181 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
182
183 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
184 offsetof(struct sk_buff, queue_mapping));
185 break;
c2497395 186
c2497395
AS		 187 case SKF_AD_VLAN_TAG:
188 case SKF_AD_VLAN_TAG_PRESENT:
189 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
190 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
191
192 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
193 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
194 offsetof(struct sk_buff, vlan_tci));
195 if (skb_field == SKF_AD_VLAN_TAG) {
196 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
197 ~VLAN_TAG_PRESENT);
198 } else {
199 /* dst_reg >>= 12 */
200 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
201 /* dst_reg &= 1 */
202 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
203 }
204 break;
9bac3d6d
AS		 205 }
206
207 return insn - insn_buf;
208}
209
bd4cf0ed 210static bool convert_bpf_extensions(struct sock_filter *fp,
2695fb55 211 struct bpf_insn **insnp)
bd4cf0ed 212{
2695fb55 213 struct bpf_insn *insn = *insnp;
9bac3d6d 214 u32 cnt;
bd4cf0ed
AS		 215
216 switch (fp->k) {
217 case SKF_AD_OFF + SKF_AD_PROTOCOL:
0b8c707d
DB		 218 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
219
220 /* A = *(u16 *) (CTX + offsetof(protocol)) */
221 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
222 offsetof(struct sk_buff, protocol));
223 /* A = ntohs(A) [emitting a nop or swap16] */
224 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
bd4cf0ed
AS		 225 break;
226
227 case SKF_AD_OFF + SKF_AD_PKTTYPE:
9bac3d6d
AS		 228 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
229 insn += cnt - 1;
bd4cf0ed
AS		 230 break;
231
232 case SKF_AD_OFF + SKF_AD_IFINDEX:
233 case SKF_AD_OFF + SKF_AD_HATYPE:
bd4cf0ed
AS		 234 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
235 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
f8f6d679
DB		 236 BUILD_BUG_ON(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)) < 0);
237
238 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)),
239 BPF_REG_TMP, BPF_REG_CTX,
240 offsetof(struct sk_buff, dev));
241 /* if (tmp != 0) goto pc + 1 */
242 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
243 *insn++ = BPF_EXIT_INSN();
244 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
245 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
246 offsetof(struct net_device, ifindex));
247 else
248 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
249 offsetof(struct net_device, type));
bd4cf0ed
AS		 250 break;
251
252 case SKF_AD_OFF + SKF_AD_MARK:
9bac3d6d
AS		 253 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
254 insn += cnt - 1;
bd4cf0ed
AS		 255 break;
256
257 case SKF_AD_OFF + SKF_AD_RXHASH:
258 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
259
9739eef1
AS		 260 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
261 offsetof(struct sk_buff, hash));
bd4cf0ed
AS		 262 break;
263
264 case SKF_AD_OFF + SKF_AD_QUEUE:
9bac3d6d
AS		 265 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
266 insn += cnt - 1;
bd4cf0ed
AS		 267 break;
268
269 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
c2497395
AS		 270 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
271 BPF_REG_A, BPF_REG_CTX, insn);
272 insn += cnt - 1;
273 break;
bd4cf0ed 274
c2497395
AS		 275 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
276 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
277 BPF_REG_A, BPF_REG_CTX, insn);
278 insn += cnt - 1;
bd4cf0ed
AS		 279 break;
280
27cd5452
MS		 281 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
282 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
283
284 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
285 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
286 offsetof(struct sk_buff, vlan_proto));
287 /* A = ntohs(A) [emitting a nop or swap16] */
288 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
289 break;
290
bd4cf0ed
AS		 291 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
292 case SKF_AD_OFF + SKF_AD_NLATTR:
293 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
294 case SKF_AD_OFF + SKF_AD_CPU:
4cd3675e 295 case SKF_AD_OFF + SKF_AD_RANDOM:
e430f34e 296 /* arg1 = CTX */
f8f6d679 297 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
bd4cf0ed 298 /* arg2 = A */
f8f6d679 299 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
bd4cf0ed 300 /* arg3 = X */
f8f6d679 301 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
e430f34e 302 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
bd4cf0ed
AS		 303 switch (fp->k) {
304 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
f8f6d679 305 *insn = BPF_EMIT_CALL(__skb_get_pay_offset);
bd4cf0ed
AS		 306 break;
307 case SKF_AD_OFF + SKF_AD_NLATTR:
f8f6d679 308 *insn = BPF_EMIT_CALL(__skb_get_nlattr);
bd4cf0ed
AS		 309 break;
310 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
f8f6d679 311 *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest);
bd4cf0ed
AS		 312 break;
313 case SKF_AD_OFF + SKF_AD_CPU:
f8f6d679 314 *insn = BPF_EMIT_CALL(__get_raw_cpu_id);
bd4cf0ed 315 break;
4cd3675e 316 case SKF_AD_OFF + SKF_AD_RANDOM:
3ad00405
DB		 317 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
318 bpf_user_rnd_init_once();
4cd3675e 319 break;
bd4cf0ed
AS		 320 }
321 break;
322
323 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
9739eef1
AS		 324 /* A ^= X */
325 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
bd4cf0ed
AS		 326 break;
327
328 default:
329 /* This is just a dummy call to avoid letting the compiler
330 * evict __bpf_call_base() as an optimization. Placed here
331 * where no-one bothers.
332 */
333 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
334 return false;
335 }
336
337 *insnp = insn;
338 return true;
339}
340
341/**
8fb575ca 342 * bpf_convert_filter - convert filter program
bd4cf0ed
AS		 343 * @prog: the user passed filter program
344 * @len: the length of the user passed filter program
345 * @new_prog: buffer where converted program will be stored
346 * @new_len: pointer to store length of converted program
347 *
348 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
349 * Conversion workflow:
350 *
351 * 1) First pass for calculating the new program length:
8fb575ca 352 * bpf_convert_filter(old_prog, old_len, NULL, &new_len)
bd4cf0ed
AS		 353 *
354 * 2) 2nd pass to remap in two passes: 1st pass finds new
355 * jump offsets, 2nd pass remapping:
2695fb55 356 * new_prog = kmalloc(sizeof(struct bpf_insn) * new_len);
8fb575ca 357 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len);
bd4cf0ed 358 */
d9e12f42
NS		 359static int bpf_convert_filter(struct sock_filter *prog, int len,
360 struct bpf_insn *new_prog, int *new_len)
bd4cf0ed
AS		 361{
362 int new_flen = 0, pass = 0, target, i;
2695fb55 363 struct bpf_insn *new_insn;
bd4cf0ed
AS		 364 struct sock_filter *fp;
365 int *addrs = NULL;
366 u8 bpf_src;
367
368 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
30743837 369 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
bd4cf0ed 370
6f9a093b 371 if (len <= 0 || len > BPF_MAXINSNS)
bd4cf0ed
AS		 372 return -EINVAL;
373
374 if (new_prog) {
658da937
DB		 375 addrs = kcalloc(len, sizeof(*addrs),
376 GFP_KERNEL | __GFP_NOWARN);
bd4cf0ed
AS		 377 if (!addrs)
378 return -ENOMEM;
379 }
380
381do_pass:
382 new_insn = new_prog;
383 fp = prog;
384
8b614aeb
DB		 385 /* Classic BPF related prologue emission. */
386 if (new_insn) {
387 /* Classic BPF expects A and X to be reset first. These need
388 * to be guaranteed to be the first two instructions.
389 */
390 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
391 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
392
393 /* All programs must keep CTX in callee saved BPF_REG_CTX.
394 * In eBPF case it's done by the compiler, here we need to
395 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
396 */
397 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
398 } else {
399 new_insn += 3;
400 }
bd4cf0ed
AS		 401
402 for (i = 0; i < len; fp++, i++) {
2695fb55
AS		 403 struct bpf_insn tmp_insns[6] = { };
404 struct bpf_insn *insn = tmp_insns;
bd4cf0ed
AS		 405
406 if (addrs)
407 addrs[i] = new_insn - new_prog;
408
409 switch (fp->code) {
410 /* All arithmetic insns and skb loads map as-is. */
411 case BPF_ALU | BPF_ADD | BPF_X:
412 case BPF_ALU | BPF_ADD | BPF_K:
413 case BPF_ALU | BPF_SUB | BPF_X:
414 case BPF_ALU | BPF_SUB | BPF_K:
415 case BPF_ALU | BPF_AND | BPF_X:
416 case BPF_ALU | BPF_AND | BPF_K:
417 case BPF_ALU | BPF_OR | BPF_X:
418 case BPF_ALU | BPF_OR | BPF_K:
419 case BPF_ALU | BPF_LSH | BPF_X:
420 case BPF_ALU | BPF_LSH | BPF_K:
421 case BPF_ALU | BPF_RSH | BPF_X:
422 case BPF_ALU | BPF_RSH | BPF_K:
423 case BPF_ALU | BPF_XOR | BPF_X:
424 case BPF_ALU | BPF_XOR | BPF_K:
425 case BPF_ALU | BPF_MUL | BPF_X:
426 case BPF_ALU | BPF_MUL | BPF_K:
427 case BPF_ALU | BPF_DIV | BPF_X:
428 case BPF_ALU | BPF_DIV | BPF_K:
429 case BPF_ALU | BPF_MOD | BPF_X:
430 case BPF_ALU | BPF_MOD | BPF_K:
431 case BPF_ALU | BPF_NEG:
432 case BPF_LD | BPF_ABS | BPF_W:
433 case BPF_LD | BPF_ABS | BPF_H:
434 case BPF_LD | BPF_ABS | BPF_B:
435 case BPF_LD | BPF_IND | BPF_W:
436 case BPF_LD | BPF_IND | BPF_H:
437 case BPF_LD | BPF_IND | BPF_B:
438 /* Check for overloaded BPF extension and
439 * directly convert it if found, otherwise
440 * just move on with mapping.
441 */
442 if (BPF_CLASS(fp->code) == BPF_LD &&
443 BPF_MODE(fp->code) == BPF_ABS &&
444 convert_bpf_extensions(fp, &insn))
445 break;
446
f8f6d679 447 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
bd4cf0ed
AS		 448 break;
449
f8f6d679
DB		 450 /* Jump transformation cannot use BPF block macros
451 * everywhere as offset calculation and target updates
452 * require a bit more work than the rest, i.e. jump
453 * opcodes map as-is, but offsets need adjustment.
454 */
455
456#define BPF_EMIT_JMP \
bd4cf0ed
AS		 457 do { \
458 if (target >= len || target < 0) \
459 goto err; \
460 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
461 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
462 insn->off -= insn - tmp_insns; \
463 } while (0)
464
f8f6d679
DB		 465 case BPF_JMP | BPF_JA:
466 target = i + fp->k + 1;
467 insn->code = fp->code;
468 BPF_EMIT_JMP;
bd4cf0ed
AS		 469 break;
470
471 case BPF_JMP | BPF_JEQ | BPF_K:
472 case BPF_JMP | BPF_JEQ | BPF_X:
473 case BPF_JMP | BPF_JSET | BPF_K:
474 case BPF_JMP | BPF_JSET | BPF_X:
475 case BPF_JMP | BPF_JGT | BPF_K:
476 case BPF_JMP | BPF_JGT | BPF_X:
477 case BPF_JMP | BPF_JGE | BPF_K:
478 case BPF_JMP | BPF_JGE | BPF_X:
479 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
480 /* BPF immediates are signed, zero extend
481 * immediate into tmp register and use it
482 * in compare insn.
483 */
f8f6d679 484 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
bd4cf0ed 485
e430f34e
AS		 486 insn->dst_reg = BPF_REG_A;
487 insn->src_reg = BPF_REG_TMP;
bd4cf0ed
AS		 488 bpf_src = BPF_X;
489 } else {
e430f34e 490 insn->dst_reg = BPF_REG_A;
bd4cf0ed
AS		 491 insn->imm = fp->k;
492 bpf_src = BPF_SRC(fp->code);
19539ce7 493 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
1da177e4 494 }
bd4cf0ed
AS		 495
496 /* Common case where 'jump_false' is next insn. */
497 if (fp->jf == 0) {
498 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
499 target = i + fp->jt + 1;
f8f6d679 500 BPF_EMIT_JMP;
bd4cf0ed 501 break;
1da177e4 502 }
bd4cf0ed
AS		 503
504 /* Convert JEQ into JNE when 'jump_true' is next insn. */
505 if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) {
506 insn->code = BPF_JMP | BPF_JNE | bpf_src;
507 target = i + fp->jf + 1;
f8f6d679 508 BPF_EMIT_JMP;
bd4cf0ed 509 break;
0b05b2a4 510 }
bd4cf0ed
AS		 511
512 /* Other jumps are mapped into two insns: Jxx and JA. */
513 target = i + fp->jt + 1;
514 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
f8f6d679 515 BPF_EMIT_JMP;
bd4cf0ed
AS		 516 insn++;
517
518 insn->code = BPF_JMP | BPF_JA;
519 target = i + fp->jf + 1;
f8f6d679 520 BPF_EMIT_JMP;
bd4cf0ed
AS		 521 break;
522
523 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
524 case BPF_LDX | BPF_MSH | BPF_B:
9739eef1 525 /* tmp = A */
f8f6d679 526 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A);
1268e253 527 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
f8f6d679 528 *insn++ = BPF_LD_ABS(BPF_B, fp->k);
9739eef1 529 /* A &= 0xf */
f8f6d679 530 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
9739eef1 531 /* A <<= 2 */
f8f6d679 532 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
9739eef1 533 /* X = A */
f8f6d679 534 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
9739eef1 535 /* A = tmp */
f8f6d679 536 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
bd4cf0ed
AS		 537 break;
538
6205b9cf
DB		 539 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
540 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
541 */
bd4cf0ed
AS		 542 case BPF_RET | BPF_A:
543 case BPF_RET | BPF_K:
6205b9cf
DB		 544 if (BPF_RVAL(fp->code) == BPF_K)
545 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
546 0, fp->k);
9739eef1 547 *insn = BPF_EXIT_INSN();
bd4cf0ed
AS		 548 break;
549
550 /* Store to stack. */
551 case BPF_ST:
552 case BPF_STX:
f8f6d679
DB		 553 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
554 BPF_ST ? BPF_REG_A : BPF_REG_X,
555 -(BPF_MEMWORDS - fp->k) * 4);
bd4cf0ed
AS		 556 break;
557
558 /* Load from stack. */
559 case BPF_LD | BPF_MEM:
560 case BPF_LDX | BPF_MEM:
f8f6d679
DB		 561 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
562 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
563 -(BPF_MEMWORDS - fp->k) * 4);
bd4cf0ed
AS		 564 break;
565
566 /* A = K or X = K */
567 case BPF_LD | BPF_IMM:
568 case BPF_LDX | BPF_IMM:
f8f6d679
DB		 569 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
570 BPF_REG_A : BPF_REG_X, fp->k);
bd4cf0ed
AS		 571 break;
572
573 /* X = A */
574 case BPF_MISC | BPF_TAX:
f8f6d679 575 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
bd4cf0ed
AS		 576 break;
577
578 /* A = X */
579 case BPF_MISC | BPF_TXA:
f8f6d679 580 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
bd4cf0ed
AS		 581 break;
582
583 /* A = skb->len or X = skb->len */
584 case BPF_LD | BPF_W | BPF_LEN:
585 case BPF_LDX | BPF_W | BPF_LEN:
f8f6d679
DB		 586 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
587 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
588 offsetof(struct sk_buff, len));
bd4cf0ed
AS		 589 break;
590
f8f6d679 591 /* Access seccomp_data fields. */
bd4cf0ed 592 case BPF_LDX | BPF_ABS | BPF_W:
9739eef1
AS		 593 /* A = *(u32 *) (ctx + K) */
594 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
bd4cf0ed
AS		 595 break;
596
ca9f1fd2 597 /* Unknown instruction. */
1da177e4 598 default:
bd4cf0ed 599 goto err;
1da177e4 600 }
bd4cf0ed
AS		 601
602 insn++;
603 if (new_prog)
604 memcpy(new_insn, tmp_insns,
605 sizeof(*insn) * (insn - tmp_insns));
bd4cf0ed 606 new_insn += insn - tmp_insns;
1da177e4
LT		 607 }
608
bd4cf0ed
AS		 609 if (!new_prog) {
610 /* Only calculating new length. */
611 *new_len = new_insn - new_prog;
612 return 0;
613 }
614
615 pass++;
616 if (new_flen != new_insn - new_prog) {
617 new_flen = new_insn - new_prog;
618 if (pass > 2)
619 goto err;
bd4cf0ed
AS		 620 goto do_pass;
621 }
622
623 kfree(addrs);
624 BUG_ON(*new_len != new_flen);
1da177e4 625 return 0;
bd4cf0ed
AS		 626err:
627 kfree(addrs);
628 return -EINVAL;
1da177e4
LT		 629}
630
bd4cf0ed 631/* Security:
bd4cf0ed 632 *
2d5311e4 633 * As we dont want to clear mem[] array for each packet going through
8ea6e345 634 * __bpf_prog_run(), we check that filter loaded by user never try to read
2d5311e4 635 * a cell if not previously written, and we check all branches to be sure
25985edc 636 * a malicious user doesn't try to abuse us.
2d5311e4 637 */
ec31a05c 638static int check_load_and_stores(const struct sock_filter *filter, int flen)
2d5311e4 639{
34805931 640 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
2d5311e4
ED		 641 int pc, ret = 0;
642
643 BUILD_BUG_ON(BPF_MEMWORDS > 16);
34805931 644
99e72a0f 645 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
2d5311e4
ED		 646 if (!masks)
647 return -ENOMEM;
34805931 648
2d5311e4
ED		 649 memset(masks, 0xff, flen * sizeof(*masks));
650
651 for (pc = 0; pc < flen; pc++) {
652 memvalid &= masks[pc];
653
654 switch (filter[pc].code) {
34805931
DB		 655 case BPF_ST:
656 case BPF_STX:
2d5311e4
ED		 657 memvalid |= (1 << filter[pc].k);
658 break;
34805931
DB		 659 case BPF_LD | BPF_MEM:
660 case BPF_LDX | BPF_MEM:
2d5311e4
ED		 661 if (!(memvalid & (1 << filter[pc].k))) {
662 ret = -EINVAL;
663 goto error;
664 }
665 break;
34805931
DB		 666 case BPF_JMP | BPF_JA:
667 /* A jump must set masks on target */
2d5311e4
ED		 668 masks[pc + 1 + filter[pc].k] &= memvalid;
669 memvalid = ~0;
670 break;
34805931
DB		 671 case BPF_JMP | BPF_JEQ | BPF_K:
672 case BPF_JMP | BPF_JEQ | BPF_X:
673 case BPF_JMP | BPF_JGE | BPF_K:
674 case BPF_JMP | BPF_JGE | BPF_X:
675 case BPF_JMP | BPF_JGT | BPF_K:
676 case BPF_JMP | BPF_JGT | BPF_X:
677 case BPF_JMP | BPF_JSET | BPF_K:
678 case BPF_JMP | BPF_JSET | BPF_X:
679 /* A jump must set masks on targets */
2d5311e4
ED		 680 masks[pc + 1 + filter[pc].jt] &= memvalid;
681 masks[pc + 1 + filter[pc].jf] &= memvalid;
682 memvalid = ~0;
683 break;
684 }
685 }
686error:
687 kfree(masks);
688 return ret;
689}
690
34805931
DB		 691static bool chk_code_allowed(u16 code_to_probe)
692{
693 static const bool codes[] = {
694 /* 32 bit ALU operations */
695 [BPF_ALU | BPF_ADD | BPF_K] = true,
696 [BPF_ALU | BPF_ADD | BPF_X] = true,
697 [BPF_ALU | BPF_SUB | BPF_K] = true,
698 [BPF_ALU | BPF_SUB | BPF_X] = true,
699 [BPF_ALU | BPF_MUL | BPF_K] = true,
700 [BPF_ALU | BPF_MUL | BPF_X] = true,
701 [BPF_ALU | BPF_DIV | BPF_K] = true,
702 [BPF_ALU | BPF_DIV | BPF_X] = true,
703 [BPF_ALU | BPF_MOD | BPF_K] = true,
704 [BPF_ALU | BPF_MOD | BPF_X] = true,
705 [BPF_ALU | BPF_AND | BPF_K] = true,
706 [BPF_ALU | BPF_AND | BPF_X] = true,
707 [BPF_ALU | BPF_OR | BPF_K] = true,
708 [BPF_ALU | BPF_OR | BPF_X] = true,
709 [BPF_ALU | BPF_XOR | BPF_K] = true,
710 [BPF_ALU | BPF_XOR | BPF_X] = true,
711 [BPF_ALU | BPF_LSH | BPF_K] = true,
712 [BPF_ALU | BPF_LSH | BPF_X] = true,
713 [BPF_ALU | BPF_RSH | BPF_K] = true,
714 [BPF_ALU | BPF_RSH | BPF_X] = true,
715 [BPF_ALU | BPF_NEG] = true,
716 /* Load instructions */
717 [BPF_LD | BPF_W | BPF_ABS] = true,
718 [BPF_LD | BPF_H | BPF_ABS] = true,
719 [BPF_LD | BPF_B | BPF_ABS] = true,
720 [BPF_LD | BPF_W | BPF_LEN] = true,
721 [BPF_LD | BPF_W | BPF_IND] = true,
722 [BPF_LD | BPF_H | BPF_IND] = true,
723 [BPF_LD | BPF_B | BPF_IND] = true,
724 [BPF_LD | BPF_IMM] = true,
725 [BPF_LD | BPF_MEM] = true,
726 [BPF_LDX | BPF_W | BPF_LEN] = true,
727 [BPF_LDX | BPF_B | BPF_MSH] = true,
728 [BPF_LDX | BPF_IMM] = true,
729 [BPF_LDX | BPF_MEM] = true,
730 /* Store instructions */
731 [BPF_ST] = true,
732 [BPF_STX] = true,
733 /* Misc instructions */
734 [BPF_MISC | BPF_TAX] = true,
735 [BPF_MISC | BPF_TXA] = true,
736 /* Return instructions */
737 [BPF_RET | BPF_K] = true,
738 [BPF_RET | BPF_A] = true,
739 /* Jump instructions */
740 [BPF_JMP | BPF_JA] = true,
741 [BPF_JMP | BPF_JEQ | BPF_K] = true,
742 [BPF_JMP | BPF_JEQ | BPF_X] = true,
743 [BPF_JMP | BPF_JGE | BPF_K] = true,
744 [BPF_JMP | BPF_JGE | BPF_X] = true,
745 [BPF_JMP | BPF_JGT | BPF_K] = true,
746 [BPF_JMP | BPF_JGT | BPF_X] = true,
747 [BPF_JMP | BPF_JSET | BPF_K] = true,
748 [BPF_JMP | BPF_JSET | BPF_X] = true,
749 };
750
751 if (code_to_probe >= ARRAY_SIZE(codes))
752 return false;
753
754 return codes[code_to_probe];
755}
756
f7bd9e36
DB		 757static bool bpf_check_basics_ok(const struct sock_filter *filter,
758 unsigned int flen)
759{
760 if (filter == NULL)
761 return false;
762 if (flen == 0 || flen > BPF_MAXINSNS)
763 return false;
764
765 return true;
766}
767
1da177e4 768/**
4df95ff4 769 * bpf_check_classic - verify socket filter code
1da177e4
LT		 770 * @filter: filter to verify
771 * @flen: length of filter
772 *
773 * Check the user's filter code. If we let some ugly
774 * filter code slip through kaboom! The filter must contain
93699863
KK		 775 * no references or jumps that are out of range, no illegal
776 * instructions, and must end with a RET instruction.
1da177e4 777 *
7b11f69f
KK		 778 * All jumps are forward as they are not signed.
779 *
780 * Returns 0 if the rule set is legal or -EINVAL if not.
1da177e4 781 */
d9e12f42
NS		 782static int bpf_check_classic(const struct sock_filter *filter,
783 unsigned int flen)
1da177e4 784{
aa1113d9 785 bool anc_found;
34805931 786 int pc;
1da177e4 787
34805931 788 /* Check the filter code now */
1da177e4 789 for (pc = 0; pc < flen; pc++) {
ec31a05c 790 const struct sock_filter *ftest = &filter[pc];
93699863 791
34805931
DB		 792 /* May we actually operate on this code? */
793 if (!chk_code_allowed(ftest->code))
cba328fc 794 return -EINVAL;
34805931 795
93699863 796 /* Some instructions need special checks */
34805931
DB		 797 switch (ftest->code) {
798 case BPF_ALU | BPF_DIV | BPF_K:
799 case BPF_ALU | BPF_MOD | BPF_K:
800 /* Check for division by zero */
b6069a95
ED		 801 if (ftest->k == 0)
802 return -EINVAL;
803 break;
229394e8
RV		 804 case BPF_ALU | BPF_LSH | BPF_K:
805 case BPF_ALU | BPF_RSH | BPF_K:
806 if (ftest->k >= 32)
807 return -EINVAL;
808 break;
34805931
DB		 809 case BPF_LD | BPF_MEM:
810 case BPF_LDX | BPF_MEM:
811 case BPF_ST:
812 case BPF_STX:
813 /* Check for invalid memory addresses */
93699863
KK		 814 if (ftest->k >= BPF_MEMWORDS)
815 return -EINVAL;
816 break;
34805931
DB		 817 case BPF_JMP | BPF_JA:
818 /* Note, the large ftest->k might cause loops.
93699863
KK		 819 * Compare this with conditional jumps below,
820 * where offsets are limited. --ANK (981016)
821 */
34805931 822 if (ftest->k >= (unsigned int)(flen - pc - 1))
93699863 823 return -EINVAL;
01f2f3f6 824 break;
34805931
DB		 825 case BPF_JMP | BPF_JEQ | BPF_K:
826 case BPF_JMP | BPF_JEQ | BPF_X:
827 case BPF_JMP | BPF_JGE | BPF_K:
828 case BPF_JMP | BPF_JGE | BPF_X:
829 case BPF_JMP | BPF_JGT | BPF_K:
830 case BPF_JMP | BPF_JGT | BPF_X:
831 case BPF_JMP | BPF_JSET | BPF_K:
832 case BPF_JMP | BPF_JSET | BPF_X:
833 /* Both conditionals must be safe */
e35bedf3 834 if (pc + ftest->jt + 1 >= flen ||
93699863
KK		 835 pc + ftest->jf + 1 >= flen)
836 return -EINVAL;
cba328fc 837 break;
34805931
DB		 838 case BPF_LD | BPF_W | BPF_ABS:
839 case BPF_LD | BPF_H | BPF_ABS:
840 case BPF_LD | BPF_B | BPF_ABS:
aa1113d9 841 anc_found = false;
34805931
DB		 842 if (bpf_anc_helper(ftest) & BPF_ANC)
843 anc_found = true;
844 /* Ancillary operation unknown or unsupported */
aa1113d9
DB		 845 if (anc_found == false && ftest->k >= SKF_AD_OFF)
846 return -EINVAL;
01f2f3f6
HPP		 847 }
848 }
93699863 849
34805931 850 /* Last instruction must be a RET code */
01f2f3f6 851 switch (filter[flen - 1].code) {
34805931
DB		 852 case BPF_RET | BPF_K:
853 case BPF_RET | BPF_A:
2d5311e4 854 return check_load_and_stores(filter, flen);
cba328fc 855 }
34805931 856
cba328fc 857 return -EINVAL;
1da177e4
LT		 858}
859
7ae457c1
AS		 860static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
861 const struct sock_fprog *fprog)
a3ea269b 862{
009937e7 863 unsigned int fsize = bpf_classic_proglen(fprog);
a3ea269b
DB		 864 struct sock_fprog_kern *fkprog;
865
866 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
867 if (!fp->orig_prog)
868 return -ENOMEM;
869
870 fkprog = fp->orig_prog;
871 fkprog->len = fprog->len;
658da937
DB		 872
873 fkprog->filter = kmemdup(fp->insns, fsize,
874 GFP_KERNEL | __GFP_NOWARN);
a3ea269b
DB		 875 if (!fkprog->filter) {
876 kfree(fp->orig_prog);
877 return -ENOMEM;
878 }
879
880 return 0;
881}
882
7ae457c1 883static void bpf_release_orig_filter(struct bpf_prog *fp)
a3ea269b
DB		 884{
885 struct sock_fprog_kern *fprog = fp->orig_prog;
886
887 if (fprog) {
888 kfree(fprog->filter);
889 kfree(fprog);
890 }
891}
892
7ae457c1
AS		 893static void __bpf_prog_release(struct bpf_prog *prog)
894{
24701ece 895 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
89aa0758
AS		 896 bpf_prog_put(prog);
897 } else {
898 bpf_release_orig_filter(prog);
899 bpf_prog_free(prog);
900 }
7ae457c1
AS		 901}
902
34c5bd66
PN		 903static void __sk_filter_release(struct sk_filter *fp)
904{
7ae457c1
AS		 905 __bpf_prog_release(fp->prog);
906 kfree(fp);
34c5bd66
PN		 907}
908
47e958ea 909/**
46bcf14f 910 * sk_filter_release_rcu - Release a socket filter by rcu_head
47e958ea
PE		 911 * @rcu: rcu_head that contains the sk_filter to free
912 */
fbc907f0 913static void sk_filter_release_rcu(struct rcu_head *rcu)
47e958ea
PE		 914{
915 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
916
34c5bd66 917 __sk_filter_release(fp);
47e958ea 918}
fbc907f0
DB		 919
920/**
921 * sk_filter_release - release a socket filter
922 * @fp: filter to remove
923 *
924 * Remove a filter from a socket and release its resources.
925 */
926static void sk_filter_release(struct sk_filter *fp)
927{
928 if (atomic_dec_and_test(&fp->refcnt))
929 call_rcu(&fp->rcu, sk_filter_release_rcu);
930}
931
932void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
933{
7ae457c1 934 u32 filter_size = bpf_prog_size(fp->prog->len);
fbc907f0 935
278571ba
AS		 936 atomic_sub(filter_size, &sk->sk_omem_alloc);
937 sk_filter_release(fp);
fbc907f0 938}
47e958ea 939
278571ba
AS		 940/* try to charge the socket memory if there is space available
941 * return true on success
942 */
943bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
bd4cf0ed 944{
7ae457c1 945 u32 filter_size = bpf_prog_size(fp->prog->len);
278571ba
AS		 946
947 /* same check as in sock_kmalloc() */
948 if (filter_size <= sysctl_optmem_max &&
949 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
950 atomic_inc(&fp->refcnt);
951 atomic_add(filter_size, &sk->sk_omem_alloc);
952 return true;
bd4cf0ed 953 }
278571ba 954 return false;
bd4cf0ed
AS		 955}
956
7ae457c1 957static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
bd4cf0ed
AS		 958{
959 struct sock_filter *old_prog;
7ae457c1 960 struct bpf_prog *old_fp;
34805931 961 int err, new_len, old_len = fp->len;
bd4cf0ed
AS		 962
963 /* We are free to overwrite insns et al right here as it
964 * won't be used at this point in time anymore internally
965 * after the migration to the internal BPF instruction
966 * representation.
967 */
968 BUILD_BUG_ON(sizeof(struct sock_filter) !=
2695fb55 969 sizeof(struct bpf_insn));
bd4cf0ed 970
bd4cf0ed
AS		 971 /* Conversion cannot happen on overlapping memory areas,
972 * so we need to keep the user BPF around until the 2nd
973 * pass. At this time, the user BPF is stored in fp->insns.
974 */
975 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
658da937 976 GFP_KERNEL | __GFP_NOWARN);
bd4cf0ed
AS		 977 if (!old_prog) {
978 err = -ENOMEM;
979 goto out_err;
980 }
981
982 /* 1st pass: calculate the new program length. */
8fb575ca 983 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len);
bd4cf0ed
AS		 984 if (err)
985 goto out_err_free;
986
987 /* Expand fp for appending the new filter representation. */
988 old_fp = fp;
60a3b225 989 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
bd4cf0ed
AS		 990 if (!fp) {
991 /* The old_fp is still around in case we couldn't
992 * allocate new memory, so uncharge on that one.
993 */
994 fp = old_fp;
995 err = -ENOMEM;
996 goto out_err_free;
997 }
998
bd4cf0ed
AS		 999 fp->len = new_len;
1000
2695fb55 1001 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
8fb575ca 1002 err = bpf_convert_filter(old_prog, old_len, fp->insnsi, &new_len);
bd4cf0ed 1003 if (err)
8fb575ca 1004 /* 2nd bpf_convert_filter() can fail only if it fails
bd4cf0ed
AS		 1005 * to allocate memory, remapping must succeed. Note,
1006 * that at this time old_fp has already been released
278571ba 1007 * by krealloc().
bd4cf0ed
AS		 1008 */
1009 goto out_err_free;
1010
d1c55ab5
DB		 1011 /* We are guaranteed to never error here with cBPF to eBPF
1012 * transitions, since there's no issue with type compatibility
1013 * checks on program arrays.
1014 */
1015 fp = bpf_prog_select_runtime(fp, &err);
5fe821a9 1016
bd4cf0ed
AS		 1017 kfree(old_prog);
1018 return fp;
1019
1020out_err_free:
1021 kfree(old_prog);
1022out_err:
7ae457c1 1023 __bpf_prog_release(fp);
bd4cf0ed
AS		 1024 return ERR_PTR(err);
1025}
1026
ac67eb2c
DB		 1027static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1028 bpf_aux_classic_check_t trans)
302d6637
JP		 1029{
1030 int err;
1031
bd4cf0ed 1032 fp->bpf_func = NULL;
a91263d5 1033 fp->jited = 0;
302d6637 1034
4df95ff4 1035 err = bpf_check_classic(fp->insns, fp->len);
418c96ac 1036 if (err) {
7ae457c1 1037 __bpf_prog_release(fp);
bd4cf0ed 1038 return ERR_PTR(err);
418c96ac 1039 }
302d6637 1040
4ae92bc7
NS		 1041 /* There might be additional checks and transformations
1042 * needed on classic filters, f.e. in case of seccomp.
1043 */
1044 if (trans) {
1045 err = trans(fp->insns, fp->len);
1046 if (err) {
1047 __bpf_prog_release(fp);
1048 return ERR_PTR(err);
1049 }
1050 }
1051
bd4cf0ed
AS		 1052 /* Probe if we can JIT compile the filter and if so, do
1053 * the compilation of the filter.
1054 */
302d6637 1055 bpf_jit_compile(fp);
bd4cf0ed
AS		 1056
1057 /* JIT compiler couldn't process this filter, so do the
1058 * internal BPF translation for the optimized interpreter.
1059 */
5fe821a9 1060 if (!fp->jited)
7ae457c1 1061 fp = bpf_migrate_filter(fp);
bd4cf0ed
AS		 1062
1063 return fp;
302d6637
JP		 1064}
1065
1066/**
7ae457c1 1067 * bpf_prog_create - create an unattached filter
c6c4b97c 1068 * @pfp: the unattached filter that is created
677a9fd3 1069 * @fprog: the filter program
302d6637 1070 *
c6c4b97c 1071 * Create a filter independent of any socket. We first run some
302d6637
JP		 1072 * sanity checks on it to make sure it does not explode on us later.
1073 * If an error occurs or there is insufficient memory for the filter
1074 * a negative errno code is returned. On success the return is zero.
1075 */
7ae457c1 1076int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
302d6637 1077{
009937e7 1078 unsigned int fsize = bpf_classic_proglen(fprog);
7ae457c1 1079 struct bpf_prog *fp;
302d6637
JP		 1080
1081 /* Make sure new filter is there and in the right amounts. */
f7bd9e36 1082 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
302d6637
JP		 1083 return -EINVAL;
1084
60a3b225 1085 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
302d6637
JP		 1086 if (!fp)
1087 return -ENOMEM;
a3ea269b 1088
302d6637
JP		 1089 memcpy(fp->insns, fprog->filter, fsize);
1090
302d6637 1091 fp->len = fprog->len;
a3ea269b
DB		 1092 /* Since unattached filters are not copied back to user
1093 * space through sk_get_filter(), we do not need to hold
1094 * a copy here, and can spare us the work.
1095 */
1096 fp->orig_prog = NULL;
302d6637 1097
7ae457c1 1098 /* bpf_prepare_filter() already takes care of freeing
bd4cf0ed
AS		 1099 * memory in case something goes wrong.
1100 */
4ae92bc7 1101 fp = bpf_prepare_filter(fp, NULL);
bd4cf0ed
AS		 1102 if (IS_ERR(fp))
1103 return PTR_ERR(fp);
302d6637
JP		 1104
1105 *pfp = fp;
1106 return 0;
302d6637 1107}
7ae457c1 1108EXPORT_SYMBOL_GPL(bpf_prog_create);
302d6637 1109
ac67eb2c
DB		 1110/**
1111 * bpf_prog_create_from_user - create an unattached filter from user buffer
1112 * @pfp: the unattached filter that is created
1113 * @fprog: the filter program
1114 * @trans: post-classic verifier transformation handler
bab18991 1115 * @save_orig: save classic BPF program
ac67eb2c
DB		 1116 *
1117 * This function effectively does the same as bpf_prog_create(), only
1118 * that it builds up its insns buffer from user space provided buffer.
1119 * It also allows for passing a bpf_aux_classic_check_t handler.
1120 */
1121int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
bab18991 1122 bpf_aux_classic_check_t trans, bool save_orig)
ac67eb2c
DB		 1123{
1124 unsigned int fsize = bpf_classic_proglen(fprog);
1125 struct bpf_prog *fp;
bab18991 1126 int err;
ac67eb2c
DB		 1127
1128 /* Make sure new filter is there and in the right amounts. */
f7bd9e36 1129 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
ac67eb2c
DB		 1130 return -EINVAL;
1131
1132 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1133 if (!fp)
1134 return -ENOMEM;
1135
1136 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1137 __bpf_prog_free(fp);
1138 return -EFAULT;
1139 }
1140
1141 fp->len = fprog->len;
ac67eb2c
DB		 1142 fp->orig_prog = NULL;
1143
bab18991
DB		 1144 if (save_orig) {
1145 err = bpf_prog_store_orig_filter(fp, fprog);
1146 if (err) {
1147 __bpf_prog_free(fp);
1148 return -ENOMEM;
1149 }
1150 }
1151
ac67eb2c
DB		 1152 /* bpf_prepare_filter() already takes care of freeing
1153 * memory in case something goes wrong.
1154 */
1155 fp = bpf_prepare_filter(fp, trans);
1156 if (IS_ERR(fp))
1157 return PTR_ERR(fp);
1158
1159 *pfp = fp;
1160 return 0;
1161}
2ea273d7 1162EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
ac67eb2c 1163
7ae457c1 1164void bpf_prog_destroy(struct bpf_prog *fp)
302d6637 1165{
7ae457c1 1166 __bpf_prog_release(fp);
302d6637 1167}
7ae457c1 1168EXPORT_SYMBOL_GPL(bpf_prog_destroy);
302d6637 1169
8ced425e 1170static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
49b31e57
DB		 1171{
1172 struct sk_filter *fp, *old_fp;
1173
1174 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1175 if (!fp)
1176 return -ENOMEM;
1177
1178 fp->prog = prog;
1179 atomic_set(&fp->refcnt, 0);
1180
1181 if (!sk_filter_charge(sk, fp)) {
1182 kfree(fp);
1183 return -ENOMEM;
1184 }
1185
8ced425e
HFS		 1186 old_fp = rcu_dereference_protected(sk->sk_filter,
1187 lockdep_sock_is_held(sk));
49b31e57 1188 rcu_assign_pointer(sk->sk_filter, fp);
8ced425e 1189
49b31e57
DB		 1190 if (old_fp)
1191 sk_filter_uncharge(sk, old_fp);
1192
1193 return 0;
1194}
1195
538950a1
CG		 1196static int __reuseport_attach_prog(struct bpf_prog *prog, struct sock *sk)
1197{
1198 struct bpf_prog *old_prog;
1199 int err;
1200
1201 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1202 return -ENOMEM;
1203
fa463497 1204 if (sk_unhashed(sk) && sk->sk_reuseport) {
538950a1
CG		 1205 err = reuseport_alloc(sk);
1206 if (err)
1207 return err;
1208 } else if (!rcu_access_pointer(sk->sk_reuseport_cb)) {
1209 /* The socket wasn't bound with SO_REUSEPORT */
1210 return -EINVAL;
1211 }
1212
1213 old_prog = reuseport_attach_prog(sk, prog);
1214 if (old_prog)
1215 bpf_prog_destroy(old_prog);
1216
1217 return 0;
1218}
1219
1220static
1221struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1da177e4 1222{
009937e7 1223 unsigned int fsize = bpf_classic_proglen(fprog);
7ae457c1 1224 struct bpf_prog *prog;
1da177e4
LT		 1225 int err;
1226
d59577b6 1227 if (sock_flag(sk, SOCK_FILTER_LOCKED))
538950a1 1228 return ERR_PTR(-EPERM);
d59577b6 1229
1da177e4 1230 /* Make sure new filter is there and in the right amounts. */
f7bd9e36 1231 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
538950a1 1232 return ERR_PTR(-EINVAL);
1da177e4 1233
f7bd9e36 1234 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
7ae457c1 1235 if (!prog)
538950a1 1236 return ERR_PTR(-ENOMEM);
a3ea269b 1237
7ae457c1 1238 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
c0d1379a 1239 __bpf_prog_free(prog);
538950a1 1240 return ERR_PTR(-EFAULT);
1da177e4
LT		 1241 }
1242
7ae457c1 1243 prog->len = fprog->len;
1da177e4 1244
7ae457c1 1245 err = bpf_prog_store_orig_filter(prog, fprog);
a3ea269b 1246 if (err) {
c0d1379a 1247 __bpf_prog_free(prog);
538950a1 1248 return ERR_PTR(-ENOMEM);
a3ea269b
DB		 1249 }
1250
7ae457c1 1251 /* bpf_prepare_filter() already takes care of freeing
bd4cf0ed
AS		 1252 * memory in case something goes wrong.
1253 */
538950a1
CG		 1254 return bpf_prepare_filter(prog, NULL);
1255}
1256
1257/**
1258 * sk_attach_filter - attach a socket filter
1259 * @fprog: the filter program
1260 * @sk: the socket to use
1261 *
1262 * Attach the user's filter code. We first run some sanity checks on
1263 * it to make sure it does not explode on us later. If an error
1264 * occurs or there is insufficient memory for the filter a negative
1265 * errno code is returned. On success the return is zero.
1266 */
8ced425e 1267int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
538950a1
CG		 1268{
1269 struct bpf_prog *prog = __get_filter(fprog, sk);
1270 int err;
1271
7ae457c1
AS		 1272 if (IS_ERR(prog))
1273 return PTR_ERR(prog);
1274
8ced425e 1275 err = __sk_attach_prog(prog, sk);
49b31e57 1276 if (err < 0) {
7ae457c1 1277 __bpf_prog_release(prog);
49b31e57 1278 return err;
278571ba
AS		 1279 }
1280
d3904b73 1281 return 0;
1da177e4 1282}
8ced425e 1283EXPORT_SYMBOL_GPL(sk_attach_filter);
1da177e4 1284
538950a1 1285int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
89aa0758 1286{
538950a1 1287 struct bpf_prog *prog = __get_filter(fprog, sk);
49b31e57 1288 int err;
89aa0758 1289
538950a1
CG		 1290 if (IS_ERR(prog))
1291 return PTR_ERR(prog);
1292
1293 err = __reuseport_attach_prog(prog, sk);
1294 if (err < 0) {
1295 __bpf_prog_release(prog);
1296 return err;
1297 }
1298
1299 return 0;
1300}
1301
1302static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1303{
89aa0758 1304 if (sock_flag(sk, SOCK_FILTER_LOCKED))
538950a1 1305 return ERR_PTR(-EPERM);
89aa0758 1306
113214be 1307 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
538950a1
CG		 1308}
1309
1310int sk_attach_bpf(u32 ufd, struct sock *sk)
1311{
1312 struct bpf_prog *prog = __get_bpf(ufd, sk);
1313 int err;
1314
1315 if (IS_ERR(prog))
1316 return PTR_ERR(prog);
1317
8ced425e 1318 err = __sk_attach_prog(prog, sk);
49b31e57 1319 if (err < 0) {
89aa0758 1320 bpf_prog_put(prog);
49b31e57 1321 return err;
89aa0758
AS		 1322 }
1323
89aa0758
AS		 1324 return 0;
1325}
1326
538950a1
CG		 1327int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1328{
1329 struct bpf_prog *prog = __get_bpf(ufd, sk);
1330 int err;
1331
1332 if (IS_ERR(prog))
1333 return PTR_ERR(prog);
1334
1335 err = __reuseport_attach_prog(prog, sk);
1336 if (err < 0) {
1337 bpf_prog_put(prog);
1338 return err;
1339 }
1340
1341 return 0;
1342}
1343
21cafc1d
DB		 1344struct bpf_scratchpad {
1345 union {
1346 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1347 u8 buff[MAX_BPF_STACK];
1348 };
1349};
1350
1351static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
91bc4822 1352
db58ba45
AS		 1353static inline int bpf_try_make_writable(struct sk_buff *skb,
1354 unsigned int write_len)
1355{
1356 int err;
1357
1358 if (!skb_cloned(skb))
1359 return 0;
1360 if (skb_clone_writable(skb, write_len))
1361 return 0;
1362 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1363 if (!err)
1364 bpf_compute_data_end(skb);
1365 return err;
1366}
1367
91bc4822 1368static u64 bpf_skb_store_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 flags)
608cd71a 1369{
21cafc1d 1370 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
608cd71a 1371 struct sk_buff *skb = (struct sk_buff *) (long) r1;
a166151c 1372 int offset = (int) r2;
608cd71a
AS		 1373 void *from = (void *) (long) r3;
1374 unsigned int len = (unsigned int) r4;
608cd71a
AS		 1375 void *ptr;
1376
8afd54c8 1377 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
781c53bc
DB		 1378 return -EINVAL;
1379
608cd71a
AS		 1380 /* bpf verifier guarantees that:
1381 * 'from' pointer points to bpf program stack
1382 * 'len' bytes of it were initialized
1383 * 'len' > 0
1384 * 'skb' is a valid pointer to 'struct sk_buff'
1385 *
1386 * so check for invalid 'offset' and too large 'len'
1387 */
21cafc1d 1388 if (unlikely((u32) offset > 0xffff || len > sizeof(sp->buff)))
608cd71a 1389 return -EFAULT;
db58ba45 1390 if (unlikely(bpf_try_make_writable(skb, offset + len)))
608cd71a
AS		 1391 return -EFAULT;
1392
21cafc1d 1393 ptr = skb_header_pointer(skb, offset, len, sp->buff);
608cd71a
AS		 1394 if (unlikely(!ptr))
1395 return -EFAULT;
1396
781c53bc 1397 if (flags & BPF_F_RECOMPUTE_CSUM)
91bc4822 1398 skb_postpull_rcsum(skb, ptr, len);
608cd71a
AS		 1399
1400 memcpy(ptr, from, len);
1401
21cafc1d 1402 if (ptr == sp->buff)
608cd71a
AS		 1403 /* skb_store_bits cannot return -EFAULT here */
1404 skb_store_bits(skb, offset, ptr, len);
1405
781c53bc 1406 if (flags & BPF_F_RECOMPUTE_CSUM)
f8ffad69 1407 skb_postpush_rcsum(skb, ptr, len);
8afd54c8
DB		 1408 if (flags & BPF_F_INVALIDATE_HASH)
1409 skb_clear_hash(skb);
f8ffad69 1410
608cd71a
AS		 1411 return 0;
1412}
1413
577c50aa 1414static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
608cd71a
AS		 1415 .func = bpf_skb_store_bytes,
1416 .gpl_only = false,
1417 .ret_type = RET_INTEGER,
1418 .arg1_type = ARG_PTR_TO_CTX,
1419 .arg2_type = ARG_ANYTHING,
1420 .arg3_type = ARG_PTR_TO_STACK,
1421 .arg4_type = ARG_CONST_STACK_SIZE,
91bc4822
AS		 1422 .arg5_type = ARG_ANYTHING,
1423};
1424
05c74e5e
DB		 1425static u64 bpf_skb_load_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1426{
1427 const struct sk_buff *skb = (const struct sk_buff *)(unsigned long) r1;
1428 int offset = (int) r2;
1429 void *to = (void *)(unsigned long) r3;
1430 unsigned int len = (unsigned int) r4;
1431 void *ptr;
1432
074f528e
DB		 1433 if (unlikely((u32) offset > 0xffff))
1434 goto err_clear;
05c74e5e
DB		 1435
1436 ptr = skb_header_pointer(skb, offset, len, to);
1437 if (unlikely(!ptr))
074f528e 1438 goto err_clear;
05c74e5e
DB		 1439 if (ptr != to)
1440 memcpy(to, ptr, len);
1441
1442 return 0;
074f528e
DB		 1443err_clear:
1444 memset(to, 0, len);
1445 return -EFAULT;
05c74e5e
DB		 1446}
1447
577c50aa 1448static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
05c74e5e
DB		 1449 .func = bpf_skb_load_bytes,
1450 .gpl_only = false,
1451 .ret_type = RET_INTEGER,
1452 .arg1_type = ARG_PTR_TO_CTX,
1453 .arg2_type = ARG_ANYTHING,
074f528e 1454 .arg3_type = ARG_PTR_TO_RAW_STACK,
05c74e5e
DB		 1455 .arg4_type = ARG_CONST_STACK_SIZE,
1456};
1457
a166151c 1458static u64 bpf_l3_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
91bc4822
AS		 1459{
1460 struct sk_buff *skb = (struct sk_buff *) (long) r1;
a166151c 1461 int offset = (int) r2;
91bc4822
AS		 1462 __sum16 sum, *ptr;
1463
781c53bc
DB		 1464 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1465 return -EINVAL;
a166151c 1466 if (unlikely((u32) offset > 0xffff))
91bc4822 1467 return -EFAULT;
db58ba45 1468 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(sum))))
91bc4822
AS		 1469 return -EFAULT;
1470
1471 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1472 if (unlikely(!ptr))
1473 return -EFAULT;
1474
781c53bc 1475 switch (flags & BPF_F_HDR_FIELD_MASK) {
8050c0f0
DB		 1476 case 0:
1477 if (unlikely(from != 0))
1478 return -EINVAL;
1479
1480 csum_replace_by_diff(ptr, to);
1481 break;
91bc4822
AS		 1482 case 2:
1483 csum_replace2(ptr, from, to);
1484 break;
1485 case 4:
1486 csum_replace4(ptr, from, to);
1487 break;
1488 default:
1489 return -EINVAL;
1490 }
1491
1492 if (ptr == &sum)
1493 /* skb_store_bits guaranteed to not return -EFAULT here */
1494 skb_store_bits(skb, offset, ptr, sizeof(sum));
1495
1496 return 0;
1497}
1498
577c50aa 1499static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
91bc4822
AS		 1500 .func = bpf_l3_csum_replace,
1501 .gpl_only = false,
1502 .ret_type = RET_INTEGER,
1503 .arg1_type = ARG_PTR_TO_CTX,
1504 .arg2_type = ARG_ANYTHING,
1505 .arg3_type = ARG_ANYTHING,
1506 .arg4_type = ARG_ANYTHING,
1507 .arg5_type = ARG_ANYTHING,
1508};
1509
a166151c 1510static u64 bpf_l4_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
91bc4822
AS		 1511{
1512 struct sk_buff *skb = (struct sk_buff *) (long) r1;
781c53bc 1513 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
2f72959a 1514 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
a166151c 1515 int offset = (int) r2;
91bc4822
AS		 1516 __sum16 sum, *ptr;
1517
2f72959a
DB		 1518 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_PSEUDO_HDR |
1519 BPF_F_HDR_FIELD_MASK)))
781c53bc 1520 return -EINVAL;
a166151c 1521 if (unlikely((u32) offset > 0xffff))
91bc4822 1522 return -EFAULT;
db58ba45 1523 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(sum))))
91bc4822
AS		 1524 return -EFAULT;
1525
1526 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1527 if (unlikely(!ptr))
1528 return -EFAULT;
2f72959a
DB		 1529 if (is_mmzero && !*ptr)
1530 return 0;
91bc4822 1531
781c53bc 1532 switch (flags & BPF_F_HDR_FIELD_MASK) {
7d672345
DB		 1533 case 0:
1534 if (unlikely(from != 0))
1535 return -EINVAL;
1536
1537 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1538 break;
91bc4822
AS		 1539 case 2:
1540 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1541 break;
1542 case 4:
1543 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1544 break;
1545 default:
1546 return -EINVAL;
1547 }
1548
2f72959a
DB		 1549 if (is_mmzero && !*ptr)
1550 *ptr = CSUM_MANGLED_0;
91bc4822
AS		 1551 if (ptr == &sum)
1552 /* skb_store_bits guaranteed to not return -EFAULT here */
1553 skb_store_bits(skb, offset, ptr, sizeof(sum));
1554
1555 return 0;
1556}
1557
577c50aa 1558static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
91bc4822
AS		 1559 .func = bpf_l4_csum_replace,
1560 .gpl_only = false,
1561 .ret_type = RET_INTEGER,
1562 .arg1_type = ARG_PTR_TO_CTX,
1563 .arg2_type = ARG_ANYTHING,
1564 .arg3_type = ARG_ANYTHING,
1565 .arg4_type = ARG_ANYTHING,
1566 .arg5_type = ARG_ANYTHING,
608cd71a
AS		 1567};
1568
7d672345
DB		 1569static u64 bpf_csum_diff(u64 r1, u64 from_size, u64 r3, u64 to_size, u64 seed)
1570{
21cafc1d 1571 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
7d672345
DB		 1572 u64 diff_size = from_size + to_size;
1573 __be32 *from = (__be32 *) (long) r1;
1574 __be32 *to = (__be32 *) (long) r3;
1575 int i, j = 0;
1576
1577 /* This is quite flexible, some examples:
1578 *
1579 * from_size == 0, to_size > 0, seed := csum --> pushing data
1580 * from_size > 0, to_size == 0, seed := csum --> pulling data
1581 * from_size > 0, to_size > 0, seed := 0 --> diffing data
1582 *
1583 * Even for diffing, from_size and to_size don't need to be equal.
1584 */
1585 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1586 diff_size > sizeof(sp->diff)))
1587 return -EINVAL;
1588
1589 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1590 sp->diff[j] = ~from[i];
1591 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1592 sp->diff[j] = to[i];
1593
1594 return csum_partial(sp->diff, diff_size, seed);
1595}
1596
577c50aa 1597static const struct bpf_func_proto bpf_csum_diff_proto = {
7d672345
DB		 1598 .func = bpf_csum_diff,
1599 .gpl_only = false,
1600 .ret_type = RET_INTEGER,
1601 .arg1_type = ARG_PTR_TO_STACK,
1602 .arg2_type = ARG_CONST_STACK_SIZE_OR_ZERO,
1603 .arg3_type = ARG_PTR_TO_STACK,
1604 .arg4_type = ARG_CONST_STACK_SIZE_OR_ZERO,
1605 .arg5_type = ARG_ANYTHING,
1606};
1607
a70b506e
DB		 1608static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1609{
1610 if (skb_at_tc_ingress(skb))
1611 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1612
1613 return dev_forward_skb(dev, skb);
1614}
1615
1616static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
1617{
1618 int ret;
1619
1620 if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
1621 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
1622 kfree_skb(skb);
1623 return -ENETDOWN;
1624 }
1625
1626 skb->dev = dev;
1627
1628 __this_cpu_inc(xmit_recursion);
1629 ret = dev_queue_xmit(skb);
1630 __this_cpu_dec(xmit_recursion);
1631
1632 return ret;
1633}
1634
3896d655
AS		 1635static u64 bpf_clone_redirect(u64 r1, u64 ifindex, u64 flags, u64 r4, u64 r5)
1636{
a70b506e 1637 struct sk_buff *skb = (struct sk_buff *) (long) r1;
3896d655
AS		 1638 struct net_device *dev;
1639
781c53bc
DB		 1640 if (unlikely(flags & ~(BPF_F_INGRESS)))
1641 return -EINVAL;
1642
3896d655
AS		 1643 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
1644 if (unlikely(!dev))
1645 return -EINVAL;
1646
a70b506e
DB		 1647 skb = skb_clone(skb, GFP_ATOMIC);
1648 if (unlikely(!skb))
3896d655
AS		 1649 return -ENOMEM;
1650
a70b506e
DB		 1651 return flags & BPF_F_INGRESS ?
1652 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
3896d655
AS		 1653}
1654
577c50aa 1655static const struct bpf_func_proto bpf_clone_redirect_proto = {
3896d655
AS		 1656 .func = bpf_clone_redirect,
1657 .gpl_only = false,
1658 .ret_type = RET_INTEGER,
1659 .arg1_type = ARG_PTR_TO_CTX,
1660 .arg2_type = ARG_ANYTHING,
1661 .arg3_type = ARG_ANYTHING,
1662};
1663
27b29f63
AS		 1664struct redirect_info {
1665 u32 ifindex;
1666 u32 flags;
1667};
1668
1669static DEFINE_PER_CPU(struct redirect_info, redirect_info);
781c53bc 1670
27b29f63
AS		 1671static u64 bpf_redirect(u64 ifindex, u64 flags, u64 r3, u64 r4, u64 r5)
1672{
1673 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1674
781c53bc
DB		 1675 if (unlikely(flags & ~(BPF_F_INGRESS)))
1676 return TC_ACT_SHOT;
1677
27b29f63
AS		 1678 ri->ifindex = ifindex;
1679 ri->flags = flags;
781c53bc 1680
27b29f63
AS		 1681 return TC_ACT_REDIRECT;
1682}
1683
1684int skb_do_redirect(struct sk_buff *skb)
1685{
1686 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1687 struct net_device *dev;
1688
1689 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
1690 ri->ifindex = 0;
1691 if (unlikely(!dev)) {
1692 kfree_skb(skb);
1693 return -EINVAL;
1694 }
1695
a70b506e
DB		 1696 return ri->flags & BPF_F_INGRESS ?
1697 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
27b29f63
AS		 1698}
1699
577c50aa 1700static const struct bpf_func_proto bpf_redirect_proto = {
27b29f63
AS		 1701 .func = bpf_redirect,
1702 .gpl_only = false,
1703 .ret_type = RET_INTEGER,
1704 .arg1_type = ARG_ANYTHING,
1705 .arg2_type = ARG_ANYTHING,
1706};
1707
8d20aabe
DB		 1708static u64 bpf_get_cgroup_classid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1709{
1710 return task_get_classid((struct sk_buff *) (unsigned long) r1);
1711}
1712
1713static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
1714 .func = bpf_get_cgroup_classid,
1715 .gpl_only = false,
1716 .ret_type = RET_INTEGER,
1717 .arg1_type = ARG_PTR_TO_CTX,
1718};
1719
c46646d0
DB		 1720static u64 bpf_get_route_realm(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1721{
808c1b69 1722 return dst_tclassid((struct sk_buff *) (unsigned long) r1);
c46646d0
DB		 1723}
1724
1725static const struct bpf_func_proto bpf_get_route_realm_proto = {
1726 .func = bpf_get_route_realm,
1727 .gpl_only = false,
1728 .ret_type = RET_INTEGER,
1729 .arg1_type = ARG_PTR_TO_CTX,
1730};
1731
13c5c240
DB		 1732static u64 bpf_get_hash_recalc(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1733{
1734 /* If skb_clear_hash() was called due to mangling, we can
1735 * trigger SW recalculation here. Later access to hash
1736 * can then use the inline skb->hash via context directly
1737 * instead of calling this helper again.
1738 */
1739 return skb_get_hash((struct sk_buff *) (unsigned long) r1);
1740}
1741
1742static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
1743 .func = bpf_get_hash_recalc,
1744 .gpl_only = false,
1745 .ret_type = RET_INTEGER,
1746 .arg1_type = ARG_PTR_TO_CTX,
1747};
1748
4e10df9a
AS		 1749static u64 bpf_skb_vlan_push(u64 r1, u64 r2, u64 vlan_tci, u64 r4, u64 r5)
1750{
1751 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1752 __be16 vlan_proto = (__force __be16) r2;
db58ba45 1753 int ret;
4e10df9a
AS		 1754
1755 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
1756 vlan_proto != htons(ETH_P_8021AD)))
1757 vlan_proto = htons(ETH_P_8021Q);
1758
db58ba45
AS		 1759 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
1760 bpf_compute_data_end(skb);
1761 return ret;
4e10df9a
AS		 1762}
1763
1764const struct bpf_func_proto bpf_skb_vlan_push_proto = {
1765 .func = bpf_skb_vlan_push,
1766 .gpl_only = false,
1767 .ret_type = RET_INTEGER,
1768 .arg1_type = ARG_PTR_TO_CTX,
1769 .arg2_type = ARG_ANYTHING,
1770 .arg3_type = ARG_ANYTHING,
1771};
4d9c5c53 1772EXPORT_SYMBOL_GPL(bpf_skb_vlan_push_proto);
4e10df9a
AS		 1773
1774static u64 bpf_skb_vlan_pop(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1775{
1776 struct sk_buff *skb = (struct sk_buff *) (long) r1;
db58ba45 1777 int ret;
4e10df9a 1778
db58ba45
AS		 1779 ret = skb_vlan_pop(skb);
1780 bpf_compute_data_end(skb);
1781 return ret;
4e10df9a
AS		 1782}
1783
1784const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
1785 .func = bpf_skb_vlan_pop,
1786 .gpl_only = false,
1787 .ret_type = RET_INTEGER,
1788 .arg1_type = ARG_PTR_TO_CTX,
1789};
4d9c5c53 1790EXPORT_SYMBOL_GPL(bpf_skb_vlan_pop_proto);
4e10df9a 1791
6578171a
DB		 1792static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
1793{
1794 /* Caller already did skb_cow() with len as headroom,
1795 * so no need to do it here.
1796 */
1797 skb_push(skb, len);
1798 memmove(skb->data, skb->data + len, off);
1799 memset(skb->data + off, 0, len);
1800
1801 /* No skb_postpush_rcsum(skb, skb->data + off, len)
1802 * needed here as it does not change the skb->csum
1803 * result for checksum complete when summing over
1804 * zeroed blocks.
1805 */
1806 return 0;
1807}
1808
1809static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
1810{
1811 /* skb_ensure_writable() is not needed here, as we're
1812 * already working on an uncloned skb.
1813 */
1814 if (unlikely(!pskb_may_pull(skb, off + len)))
1815 return -ENOMEM;
1816
1817 skb_postpull_rcsum(skb, skb->data + off, len);
1818 memmove(skb->data + len, skb->data, off);
1819 __skb_pull(skb, len);
1820
1821 return 0;
1822}
1823
1824static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
1825{
1826 bool trans_same = skb->transport_header == skb->network_header;
1827 int ret;
1828
1829 /* There's no need for __skb_push()/__skb_pull() pair to
1830 * get to the start of the mac header as we're guaranteed
1831 * to always start from here under eBPF.
1832 */
1833 ret = bpf_skb_generic_push(skb, off, len);
1834 if (likely(!ret)) {
1835 skb->mac_header -= len;
1836 skb->network_header -= len;
1837 if (trans_same)
1838 skb->transport_header = skb->network_header;
1839 }
1840
1841 return ret;
1842}
1843
1844static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
1845{
1846 bool trans_same = skb->transport_header == skb->network_header;
1847 int ret;
1848
1849 /* Same here, __skb_push()/__skb_pull() pair not needed. */
1850 ret = bpf_skb_generic_pop(skb, off, len);
1851 if (likely(!ret)) {
1852 skb->mac_header += len;
1853 skb->network_header += len;
1854 if (trans_same)
1855 skb->transport_header = skb->network_header;
1856 }
1857
1858 return ret;
1859}
1860
1861static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
1862{
1863 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
1864 u32 off = skb->network_header - skb->mac_header;
1865 int ret;
1866
1867 ret = skb_cow(skb, len_diff);
1868 if (unlikely(ret < 0))
1869 return ret;
1870
1871 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
1872 if (unlikely(ret < 0))
1873 return ret;
1874
1875 if (skb_is_gso(skb)) {
1876 /* SKB_GSO_UDP stays as is. SKB_GSO_TCPV4 needs to
1877 * be changed into SKB_GSO_TCPV6.
1878 */
1879 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
1880 skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV4;
1881 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV6;
1882 }
1883
1884 /* Due to IPv6 header, MSS needs to be downgraded. */
1885 skb_shinfo(skb)->gso_size -= len_diff;
1886 /* Header must be checked, and gso_segs recomputed. */
1887 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1888 skb_shinfo(skb)->gso_segs = 0;
1889 }
1890
1891 skb->protocol = htons(ETH_P_IPV6);
1892 skb_clear_hash(skb);
1893
1894 return 0;
1895}
1896
1897static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
1898{
1899 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
1900 u32 off = skb->network_header - skb->mac_header;
1901 int ret;
1902
1903 ret = skb_unclone(skb, GFP_ATOMIC);
1904 if (unlikely(ret < 0))
1905 return ret;
1906
1907 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
1908 if (unlikely(ret < 0))
1909 return ret;
1910
1911 if (skb_is_gso(skb)) {
1912 /* SKB_GSO_UDP stays as is. SKB_GSO_TCPV6 needs to
1913 * be changed into SKB_GSO_TCPV4.
1914 */
1915 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) {
1916 skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV6;
1917 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV4;
1918 }
1919
1920 /* Due to IPv4 header, MSS can be upgraded. */
1921 skb_shinfo(skb)->gso_size += len_diff;
1922 /* Header must be checked, and gso_segs recomputed. */
1923 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1924 skb_shinfo(skb)->gso_segs = 0;
1925 }
1926
1927 skb->protocol = htons(ETH_P_IP);
1928 skb_clear_hash(skb);
1929
1930 return 0;
1931}
1932
1933static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
1934{
1935 __be16 from_proto = skb->protocol;
1936
1937 if (from_proto == htons(ETH_P_IP) &&
1938 to_proto == htons(ETH_P_IPV6))
1939 return bpf_skb_proto_4_to_6(skb);
1940
1941 if (from_proto == htons(ETH_P_IPV6) &&
1942 to_proto == htons(ETH_P_IP))
1943 return bpf_skb_proto_6_to_4(skb);
1944
1945 return -ENOTSUPP;
1946}
1947
1948static u64 bpf_skb_change_proto(u64 r1, u64 r2, u64 flags, u64 r4, u64 r5)
1949{
1950 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1951 __be16 proto = (__force __be16) r2;
1952 int ret;
1953
1954 if (unlikely(flags))
1955 return -EINVAL;
1956
1957 /* General idea is that this helper does the basic groundwork
1958 * needed for changing the protocol, and eBPF program fills the
1959 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
1960 * and other helpers, rather than passing a raw buffer here.
1961 *
1962 * The rationale is to keep this minimal and without a need to
1963 * deal with raw packet data. F.e. even if we would pass buffers
1964 * here, the program still needs to call the bpf_lX_csum_replace()
1965 * helpers anyway. Plus, this way we keep also separation of
1966 * concerns, since f.e. bpf_skb_store_bytes() should only take
1967 * care of stores.
1968 *
1969 * Currently, additional options and extension header space are
1970 * not supported, but flags register is reserved so we can adapt
1971 * that. For offloads, we mark packet as dodgy, so that headers
1972 * need to be verified first.
1973 */
1974 ret = bpf_skb_proto_xlat(skb, proto);
1975 bpf_compute_data_end(skb);
1976 return ret;
1977}
1978
1979static const struct bpf_func_proto bpf_skb_change_proto_proto = {
1980 .func = bpf_skb_change_proto,
1981 .gpl_only = false,
1982 .ret_type = RET_INTEGER,
1983 .arg1_type = ARG_PTR_TO_CTX,
1984 .arg2_type = ARG_ANYTHING,
1985 .arg3_type = ARG_ANYTHING,
1986};
1987
d2485c42
DB		 1988static u64 bpf_skb_change_type(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1989{
1990 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1991 u32 pkt_type = r2;
1992
1993 /* We only allow a restricted subset to be changed for now. */
1994 if (unlikely(skb->pkt_type > PACKET_OTHERHOST ||
1995 pkt_type > PACKET_OTHERHOST))
1996 return -EINVAL;
1997
1998 skb->pkt_type = pkt_type;
1999 return 0;
2000}
2001
2002static const struct bpf_func_proto bpf_skb_change_type_proto = {
2003 .func = bpf_skb_change_type,
2004 .gpl_only = false,
2005 .ret_type = RET_INTEGER,
2006 .arg1_type = ARG_PTR_TO_CTX,
2007 .arg2_type = ARG_ANYTHING,
2008};
2009
4e10df9a
AS		 2010bool bpf_helper_changes_skb_data(void *func)
2011{
2012 if (func == bpf_skb_vlan_push)
2013 return true;
2014 if (func == bpf_skb_vlan_pop)
2015 return true;
3697649f
DB		 2016 if (func == bpf_skb_store_bytes)
2017 return true;
6578171a
DB		 2018 if (func == bpf_skb_change_proto)
2019 return true;
3697649f
DB		 2020 if (func == bpf_l3_csum_replace)
2021 return true;
2022 if (func == bpf_l4_csum_replace)
2023 return true;
2024
4e10df9a
AS		 2025 return false;
2026}
2027
555c8a86 2028static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
aa7145c1 2029 unsigned long off, unsigned long len)
555c8a86 2030{
aa7145c1 2031 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
555c8a86
DB		 2032
2033 if (unlikely(!ptr))
2034 return len;
2035 if (ptr != dst_buff)
2036 memcpy(dst_buff, ptr, len);
2037
2038 return 0;
2039}
2040
2041static u64 bpf_skb_event_output(u64 r1, u64 r2, u64 flags, u64 r4,
2042 u64 meta_size)
2043{
2044 struct sk_buff *skb = (struct sk_buff *)(long) r1;
2045 struct bpf_map *map = (struct bpf_map *)(long) r2;
2046 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
2047 void *meta = (void *)(long) r4;
2048
2049 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
2050 return -EINVAL;
2051 if (unlikely(skb_size > skb->len))
2052 return -EFAULT;
2053
2054 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
2055 bpf_skb_copy);
2056}
2057
2058static const struct bpf_func_proto bpf_skb_event_output_proto = {
2059 .func = bpf_skb_event_output,
2060 .gpl_only = true,
2061 .ret_type = RET_INTEGER,
2062 .arg1_type = ARG_PTR_TO_CTX,
2063 .arg2_type = ARG_CONST_MAP_PTR,
2064 .arg3_type = ARG_ANYTHING,
2065 .arg4_type = ARG_PTR_TO_STACK,
2066 .arg5_type = ARG_CONST_STACK_SIZE,
2067};
2068
c6c33454
DB		 2069static unsigned short bpf_tunnel_key_af(u64 flags)
2070{
2071 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
2072}
2073
d3aa45ce
AS		 2074static u64 bpf_skb_get_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5)
2075{
2076 struct sk_buff *skb = (struct sk_buff *) (long) r1;
2077 struct bpf_tunnel_key *to = (struct bpf_tunnel_key *) (long) r2;
c6c33454
DB		 2078 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2079 u8 compat[sizeof(struct bpf_tunnel_key)];
074f528e
DB		 2080 void *to_orig = to;
2081 int err;
d3aa45ce 2082
074f528e
DB		 2083 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
2084 err = -EINVAL;
2085 goto err_clear;
2086 }
2087 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
2088 err = -EPROTO;
2089 goto err_clear;
2090 }
c6c33454 2091 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
074f528e 2092 err = -EINVAL;
c6c33454 2093 switch (size) {
4018ab18 2094 case offsetof(struct bpf_tunnel_key, tunnel_label):
c0e760c9 2095 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4018ab18 2096 goto set_compat;
c6c33454
DB		 2097 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
2098 /* Fixup deprecated structure layouts here, so we have
2099 * a common path later on.
2100 */
2101 if (ip_tunnel_info_af(info) != AF_INET)
074f528e 2102 goto err_clear;
4018ab18 2103set_compat:
c6c33454
DB		 2104 to = (struct bpf_tunnel_key *)compat;
2105 break;
2106 default:
074f528e 2107 goto err_clear;
c6c33454
DB		 2108 }
2109 }
d3aa45ce
AS		 2110
2111 to->tunnel_id = be64_to_cpu(info->key.tun_id);
c6c33454
DB		 2112 to->tunnel_tos = info->key.tos;
2113 to->tunnel_ttl = info->key.ttl;
2114
4018ab18 2115 if (flags & BPF_F_TUNINFO_IPV6) {
c6c33454
DB		 2116 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
2117 sizeof(to->remote_ipv6));
4018ab18
DB		 2118 to->tunnel_label = be32_to_cpu(info->key.label);
2119 } else {
c6c33454 2120 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4018ab18 2121 }
c6c33454
DB		 2122
2123 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
074f528e 2124 memcpy(to_orig, to, size);
d3aa45ce
AS		 2125
2126 return 0;
074f528e
DB		 2127err_clear:
2128 memset(to_orig, 0, size);
2129 return err;
d3aa45ce
AS		 2130}
2131
577c50aa 2132static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
d3aa45ce
AS		 2133 .func = bpf_skb_get_tunnel_key,
2134 .gpl_only = false,
2135 .ret_type = RET_INTEGER,
2136 .arg1_type = ARG_PTR_TO_CTX,
074f528e 2137 .arg2_type = ARG_PTR_TO_RAW_STACK,
d3aa45ce
AS		 2138 .arg3_type = ARG_CONST_STACK_SIZE,
2139 .arg4_type = ARG_ANYTHING,
2140};
2141
14ca0751
DB		 2142static u64 bpf_skb_get_tunnel_opt(u64 r1, u64 r2, u64 size, u64 r4, u64 r5)
2143{
2144 struct sk_buff *skb = (struct sk_buff *) (long) r1;
2145 u8 *to = (u8 *) (long) r2;
2146 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
074f528e 2147 int err;
14ca0751
DB		 2148
2149 if (unlikely(!info ||
074f528e
DB		 2150 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
2151 err = -ENOENT;
2152 goto err_clear;
2153 }
2154 if (unlikely(size < info->options_len)) {
2155 err = -ENOMEM;
2156 goto err_clear;
2157 }
14ca0751
DB		 2158
2159 ip_tunnel_info_opts_get(to, info);
074f528e
DB		 2160 if (size > info->options_len)
2161 memset(to + info->options_len, 0, size - info->options_len);
14ca0751
DB		 2162
2163 return info->options_len;
074f528e
DB		 2164err_clear:
2165 memset(to, 0, size);
2166 return err;
14ca0751
DB		 2167}
2168
2169static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
2170 .func = bpf_skb_get_tunnel_opt,
2171 .gpl_only = false,
2172 .ret_type = RET_INTEGER,
2173 .arg1_type = ARG_PTR_TO_CTX,
074f528e 2174 .arg2_type = ARG_PTR_TO_RAW_STACK,
14ca0751
DB		 2175 .arg3_type = ARG_CONST_STACK_SIZE,
2176};
2177
d3aa45ce
AS		 2178static struct metadata_dst __percpu *md_dst;
2179
2180static u64 bpf_skb_set_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5)
2181{
2182 struct sk_buff *skb = (struct sk_buff *) (long) r1;
2183 struct bpf_tunnel_key *from = (struct bpf_tunnel_key *) (long) r2;
2184 struct metadata_dst *md = this_cpu_ptr(md_dst);
c6c33454 2185 u8 compat[sizeof(struct bpf_tunnel_key)];
d3aa45ce
AS		 2186 struct ip_tunnel_info *info;
2187
22080870
DB		 2188 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
2189 BPF_F_DONT_FRAGMENT)))
d3aa45ce 2190 return -EINVAL;
c6c33454
DB		 2191 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
2192 switch (size) {
4018ab18 2193 case offsetof(struct bpf_tunnel_key, tunnel_label):
c0e760c9 2194 case offsetof(struct bpf_tunnel_key, tunnel_ext):
c6c33454
DB		 2195 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
2196 /* Fixup deprecated structure layouts here, so we have
2197 * a common path later on.
2198 */
2199 memcpy(compat, from, size);
2200 memset(compat + size, 0, sizeof(compat) - size);
2201 from = (struct bpf_tunnel_key *)compat;
2202 break;
2203 default:
2204 return -EINVAL;
2205 }
2206 }
c0e760c9
DB		 2207 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
2208 from->tunnel_ext))
4018ab18 2209 return -EINVAL;
d3aa45ce
AS		 2210
2211 skb_dst_drop(skb);
2212 dst_hold((struct dst_entry *) md);
2213 skb_dst_set(skb, (struct dst_entry *) md);
2214
2215 info = &md->u.tun_info;
2216 info->mode = IP_TUNNEL_INFO_TX;
c6c33454 2217
db3c6139 2218 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
22080870
DB		 2219 if (flags & BPF_F_DONT_FRAGMENT)
2220 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
2221
d3aa45ce 2222 info->key.tun_id = cpu_to_be64(from->tunnel_id);
c6c33454
DB		 2223 info->key.tos = from->tunnel_tos;
2224 info->key.ttl = from->tunnel_ttl;
2225
2226 if (flags & BPF_F_TUNINFO_IPV6) {
2227 info->mode |= IP_TUNNEL_INFO_IPV6;
2228 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
2229 sizeof(from->remote_ipv6));
4018ab18
DB		 2230 info->key.label = cpu_to_be32(from->tunnel_label) &
2231 IPV6_FLOWLABEL_MASK;
c6c33454
DB		 2232 } else {
2233 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
2da897e5
DB		 2234 if (flags & BPF_F_ZERO_CSUM_TX)
2235 info->key.tun_flags &= ~TUNNEL_CSUM;
c6c33454 2236 }
d3aa45ce
AS		 2237
2238 return 0;
2239}
2240
577c50aa 2241static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
d3aa45ce
AS		 2242 .func = bpf_skb_set_tunnel_key,
2243 .gpl_only = false,
2244 .ret_type = RET_INTEGER,
2245 .arg1_type = ARG_PTR_TO_CTX,
2246 .arg2_type = ARG_PTR_TO_STACK,
2247 .arg3_type = ARG_CONST_STACK_SIZE,
2248 .arg4_type = ARG_ANYTHING,
2249};
2250
14ca0751
DB		 2251static u64 bpf_skb_set_tunnel_opt(u64 r1, u64 r2, u64 size, u64 r4, u64 r5)
2252{
2253 struct sk_buff *skb = (struct sk_buff *) (long) r1;
2254 u8 *from = (u8 *) (long) r2;
2255 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2256 const struct metadata_dst *md = this_cpu_ptr(md_dst);
2257
2258 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
2259 return -EINVAL;
fca5fdf6 2260 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
14ca0751
DB		 2261 return -ENOMEM;
2262
2263 ip_tunnel_info_opts_set(info, from, size);
2264
2265 return 0;
2266}
2267
2268static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
2269 .func = bpf_skb_set_tunnel_opt,
2270 .gpl_only = false,
2271 .ret_type = RET_INTEGER,
2272 .arg1_type = ARG_PTR_TO_CTX,
2273 .arg2_type = ARG_PTR_TO_STACK,
2274 .arg3_type = ARG_CONST_STACK_SIZE,
2275};
2276
2277static const struct bpf_func_proto *
2278bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
d3aa45ce
AS		 2279{
2280 if (!md_dst) {
14ca0751
DB		 2281 /* Race is not possible, since it's called from verifier
2282 * that is holding verifier mutex.
d3aa45ce 2283 */
fca5fdf6 2284 md_dst = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
14ca0751 2285 GFP_KERNEL);
d3aa45ce
AS		 2286 if (!md_dst)
2287 return NULL;
2288 }
14ca0751
DB		 2289
2290 switch (which) {
2291 case BPF_FUNC_skb_set_tunnel_key:
2292 return &bpf_skb_set_tunnel_key_proto;
2293 case BPF_FUNC_skb_set_tunnel_opt:
2294 return &bpf_skb_set_tunnel_opt_proto;
2295 default:
2296 return NULL;
2297 }
d3aa45ce
AS		 2298}
2299
4a482f34
MKL		 2300#ifdef CONFIG_SOCK_CGROUP_DATA
2301static u64 bpf_skb_in_cgroup(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
2302{
2303 struct sk_buff *skb = (struct sk_buff *)(long)r1;
2304 struct bpf_map *map = (struct bpf_map *)(long)r2;
2305 struct bpf_array *array = container_of(map, struct bpf_array, map);
2306 struct cgroup *cgrp;
2307 struct sock *sk;
2308 u32 i = (u32)r3;
2309
2310 sk = skb->sk;
2311 if (!sk || !sk_fullsock(sk))
2312 return -ENOENT;
2313
2314 if (unlikely(i >= array->map.max_entries))
2315 return -E2BIG;
2316
2317 cgrp = READ_ONCE(array->ptrs[i]);
2318 if (unlikely(!cgrp))
2319 return -EAGAIN;
2320
2321 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), cgrp);
2322}
2323
2324static const struct bpf_func_proto bpf_skb_in_cgroup_proto = {
2325 .func = bpf_skb_in_cgroup,
2326 .gpl_only = false,
2327 .ret_type = RET_INTEGER,
2328 .arg1_type = ARG_PTR_TO_CTX,
2329 .arg2_type = ARG_CONST_MAP_PTR,
2330 .arg3_type = ARG_ANYTHING,
2331};
2332#endif
2333
d4052c4a
DB		 2334static const struct bpf_func_proto *
2335sk_filter_func_proto(enum bpf_func_id func_id)
89aa0758
AS		 2336{
2337 switch (func_id) {
2338 case BPF_FUNC_map_lookup_elem:
2339 return &bpf_map_lookup_elem_proto;
2340 case BPF_FUNC_map_update_elem:
2341 return &bpf_map_update_elem_proto;
2342 case BPF_FUNC_map_delete_elem:
2343 return &bpf_map_delete_elem_proto;
03e69b50
DB		 2344 case BPF_FUNC_get_prandom_u32:
2345 return &bpf_get_prandom_u32_proto;
c04167ce 2346 case BPF_FUNC_get_smp_processor_id:
80b48c44 2347 return &bpf_get_raw_smp_processor_id_proto;
04fd61ab
AS		 2348 case BPF_FUNC_tail_call:
2349 return &bpf_tail_call_proto;
17ca8cbf
DB		 2350 case BPF_FUNC_ktime_get_ns:
2351 return &bpf_ktime_get_ns_proto;
0756ea3e 2352 case BPF_FUNC_trace_printk:
1be7f75d
AS		 2353 if (capable(CAP_SYS_ADMIN))
2354 return bpf_get_trace_printk_proto();
89aa0758
AS		 2355 default:
2356 return NULL;
2357 }
2358}
2359
608cd71a
AS		 2360static const struct bpf_func_proto *
2361tc_cls_act_func_proto(enum bpf_func_id func_id)
2362{
2363 switch (func_id) {
2364 case BPF_FUNC_skb_store_bytes:
2365 return &bpf_skb_store_bytes_proto;
05c74e5e
DB		 2366 case BPF_FUNC_skb_load_bytes:
2367 return &bpf_skb_load_bytes_proto;
7d672345
DB		 2368 case BPF_FUNC_csum_diff:
2369 return &bpf_csum_diff_proto;
91bc4822
AS		 2370 case BPF_FUNC_l3_csum_replace:
2371 return &bpf_l3_csum_replace_proto;
2372 case BPF_FUNC_l4_csum_replace:
2373 return &bpf_l4_csum_replace_proto;
3896d655
AS		 2374 case BPF_FUNC_clone_redirect:
2375 return &bpf_clone_redirect_proto;
8d20aabe
DB		 2376 case BPF_FUNC_get_cgroup_classid:
2377 return &bpf_get_cgroup_classid_proto;
4e10df9a
AS		 2378 case BPF_FUNC_skb_vlan_push:
2379 return &bpf_skb_vlan_push_proto;
2380 case BPF_FUNC_skb_vlan_pop:
2381 return &bpf_skb_vlan_pop_proto;
6578171a
DB		 2382 case BPF_FUNC_skb_change_proto:
2383 return &bpf_skb_change_proto_proto;
d2485c42
DB		 2384 case BPF_FUNC_skb_change_type:
2385 return &bpf_skb_change_type_proto;
d3aa45ce
AS		 2386 case BPF_FUNC_skb_get_tunnel_key:
2387 return &bpf_skb_get_tunnel_key_proto;
2388 case BPF_FUNC_skb_set_tunnel_key:
14ca0751
DB		 2389 return bpf_get_skb_set_tunnel_proto(func_id);
2390 case BPF_FUNC_skb_get_tunnel_opt:
2391 return &bpf_skb_get_tunnel_opt_proto;
2392 case BPF_FUNC_skb_set_tunnel_opt:
2393 return bpf_get_skb_set_tunnel_proto(func_id);
27b29f63
AS		 2394 case BPF_FUNC_redirect:
2395 return &bpf_redirect_proto;
c46646d0
DB		 2396 case BPF_FUNC_get_route_realm:
2397 return &bpf_get_route_realm_proto;
13c5c240
DB		 2398 case BPF_FUNC_get_hash_recalc:
2399 return &bpf_get_hash_recalc_proto;
bd570ff9 2400 case BPF_FUNC_perf_event_output:
555c8a86 2401 return &bpf_skb_event_output_proto;
80b48c44
DB		 2402 case BPF_FUNC_get_smp_processor_id:
2403 return &bpf_get_smp_processor_id_proto;
4a482f34
MKL		 2404#ifdef CONFIG_SOCK_CGROUP_DATA
2405 case BPF_FUNC_skb_in_cgroup:
2406 return &bpf_skb_in_cgroup_proto;
2407#endif
608cd71a
AS		 2408 default:
2409 return sk_filter_func_proto(func_id);
2410 }
2411}
2412
6a773a15
BB		 2413static const struct bpf_func_proto *
2414xdp_func_proto(enum bpf_func_id func_id)
2415{
2416 return sk_filter_func_proto(func_id);
2417}
2418
d691f9e8 2419static bool __is_valid_access(int off, int size, enum bpf_access_type type)
89aa0758 2420{
9bac3d6d
AS		 2421 if (off < 0 || off >= sizeof(struct __sk_buff))
2422 return false;
4936e352 2423 /* The verifier guarantees that size > 0. */
9bac3d6d
AS		 2424 if (off % size != 0)
2425 return false;
4936e352 2426 if (size != sizeof(__u32))
9bac3d6d
AS		 2427 return false;
2428
2429 return true;
2430}
2431
d691f9e8 2432static bool sk_filter_is_valid_access(int off, int size,
19de99f7
AS		 2433 enum bpf_access_type type,
2434 enum bpf_reg_type *reg_type)
d691f9e8 2435{
db58ba45
AS		 2436 switch (off) {
2437 case offsetof(struct __sk_buff, tc_classid):
2438 case offsetof(struct __sk_buff, data):
2439 case offsetof(struct __sk_buff, data_end):
045efa82 2440 return false;
db58ba45 2441 }
045efa82 2442
d691f9e8
AS		 2443 if (type == BPF_WRITE) {
2444 switch (off) {
2445 case offsetof(struct __sk_buff, cb[0]) ...
4936e352 2446 offsetof(struct __sk_buff, cb[4]):
d691f9e8
AS		 2447 break;
2448 default:
2449 return false;
2450 }
2451 }
2452
2453 return __is_valid_access(off, size, type);
2454}
2455
2456static bool tc_cls_act_is_valid_access(int off, int size,
19de99f7
AS		 2457 enum bpf_access_type type,
2458 enum bpf_reg_type *reg_type)
d691f9e8
AS		 2459{
2460 if (type == BPF_WRITE) {
2461 switch (off) {
2462 case offsetof(struct __sk_buff, mark):
2463 case offsetof(struct __sk_buff, tc_index):
754f1e6a 2464 case offsetof(struct __sk_buff, priority):
d691f9e8 2465 case offsetof(struct __sk_buff, cb[0]) ...
09c37a2c
DB		 2466 offsetof(struct __sk_buff, cb[4]):
2467 case offsetof(struct __sk_buff, tc_classid):
d691f9e8
AS		 2468 break;
2469 default:
2470 return false;
2471 }
2472 }
19de99f7
AS		 2473
2474 switch (off) {
2475 case offsetof(struct __sk_buff, data):
2476 *reg_type = PTR_TO_PACKET;
2477 break;
2478 case offsetof(struct __sk_buff, data_end):
2479 *reg_type = PTR_TO_PACKET_END;
2480 break;
2481 }
2482
d691f9e8
AS		 2483 return __is_valid_access(off, size, type);
2484}
2485
6a773a15
BB		 2486static bool __is_valid_xdp_access(int off, int size,
2487 enum bpf_access_type type)
2488{
2489 if (off < 0 || off >= sizeof(struct xdp_md))
2490 return false;
2491 if (off % size != 0)
2492 return false;
2493 if (size != 4)
2494 return false;
2495
2496 return true;
2497}
2498
2499static bool xdp_is_valid_access(int off, int size,
2500 enum bpf_access_type type,
2501 enum bpf_reg_type *reg_type)
2502{
2503 if (type == BPF_WRITE)
2504 return false;
2505
2506 switch (off) {
2507 case offsetof(struct xdp_md, data):
2508 *reg_type = PTR_TO_PACKET;
2509 break;
2510 case offsetof(struct xdp_md, data_end):
2511 *reg_type = PTR_TO_PACKET_END;
2512 break;
2513 }
2514
2515 return __is_valid_xdp_access(off, size, type);
2516}
2517
2518void bpf_warn_invalid_xdp_action(u32 act)
2519{
2520 WARN_ONCE(1, "Illegal XDP return value %u, expect packet loss\n", act);
2521}
2522EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
2523
d691f9e8
AS		 2524static u32 bpf_net_convert_ctx_access(enum bpf_access_type type, int dst_reg,
2525 int src_reg, int ctx_off,
ff936a04
AS		 2526 struct bpf_insn *insn_buf,
2527 struct bpf_prog *prog)
9bac3d6d
AS		 2528{
2529 struct bpf_insn *insn = insn_buf;
2530
2531 switch (ctx_off) {
2532 case offsetof(struct __sk_buff, len):
2533 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4);
2534
2535 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2536 offsetof(struct sk_buff, len));
2537 break;
2538
0b8c707d
DB		 2539 case offsetof(struct __sk_buff, protocol):
2540 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
2541
2542 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
2543 offsetof(struct sk_buff, protocol));
2544 break;
2545
27cd5452
MS		 2546 case offsetof(struct __sk_buff, vlan_proto):
2547 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
2548
2549 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
2550 offsetof(struct sk_buff, vlan_proto));
2551 break;
2552
bcad5718
DB		 2553 case offsetof(struct __sk_buff, priority):
2554 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, priority) != 4);
2555
754f1e6a
DB		 2556 if (type == BPF_WRITE)
2557 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg,
2558 offsetof(struct sk_buff, priority));
2559 else
2560 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2561 offsetof(struct sk_buff, priority));
bcad5718
DB		 2562 break;
2563
37e82c2f
AS		 2564 case offsetof(struct __sk_buff, ingress_ifindex):
2565 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, skb_iif) != 4);
2566
2567 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2568 offsetof(struct sk_buff, skb_iif));
2569 break;
2570
2571 case offsetof(struct __sk_buff, ifindex):
2572 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
2573
2574 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)),
2575 dst_reg, src_reg,
2576 offsetof(struct sk_buff, dev));
2577 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
2578 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, dst_reg,
2579 offsetof(struct net_device, ifindex));
2580 break;
2581
ba7591d8
DB		 2582 case offsetof(struct __sk_buff, hash):
2583 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
2584
2585 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2586 offsetof(struct sk_buff, hash));
2587 break;
2588
9bac3d6d 2589 case offsetof(struct __sk_buff, mark):
d691f9e8
AS		 2590 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
2591
2592 if (type == BPF_WRITE)
2593 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg,
2594 offsetof(struct sk_buff, mark));
2595 else
2596 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2597 offsetof(struct sk_buff, mark));
2598 break;
9bac3d6d
AS		 2599
2600 case offsetof(struct __sk_buff, pkt_type):
2601 return convert_skb_access(SKF_AD_PKTTYPE, dst_reg, src_reg, insn);
2602
2603 case offsetof(struct __sk_buff, queue_mapping):
2604 return convert_skb_access(SKF_AD_QUEUE, dst_reg, src_reg, insn);
c2497395 2605
c2497395
AS		 2606 case offsetof(struct __sk_buff, vlan_present):
2607 return convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
2608 dst_reg, src_reg, insn);
2609
2610 case offsetof(struct __sk_buff, vlan_tci):
2611 return convert_skb_access(SKF_AD_VLAN_TAG,
2612 dst_reg, src_reg, insn);
d691f9e8
AS		 2613
2614 case offsetof(struct __sk_buff, cb[0]) ...
2615 offsetof(struct __sk_buff, cb[4]):
2616 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
2617
ff936a04 2618 prog->cb_access = 1;
d691f9e8
AS		 2619 ctx_off -= offsetof(struct __sk_buff, cb[0]);
2620 ctx_off += offsetof(struct sk_buff, cb);
2621 ctx_off += offsetof(struct qdisc_skb_cb, data);
2622 if (type == BPF_WRITE)
2623 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg, ctx_off);
2624 else
2625 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, ctx_off);
2626 break;
2627
045efa82
DB		 2628 case offsetof(struct __sk_buff, tc_classid):
2629 ctx_off -= offsetof(struct __sk_buff, tc_classid);
2630 ctx_off += offsetof(struct sk_buff, cb);
2631 ctx_off += offsetof(struct qdisc_skb_cb, tc_classid);
09c37a2c
DB		 2632 if (type == BPF_WRITE)
2633 *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg, ctx_off);
2634 else
2635 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, ctx_off);
045efa82
DB		 2636 break;
2637
db58ba45
AS		 2638 case offsetof(struct __sk_buff, data):
2639 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, data)),
2640 dst_reg, src_reg,
2641 offsetof(struct sk_buff, data));
2642 break;
2643
2644 case offsetof(struct __sk_buff, data_end):
2645 ctx_off -= offsetof(struct __sk_buff, data_end);
2646 ctx_off += offsetof(struct sk_buff, cb);
2647 ctx_off += offsetof(struct bpf_skb_data_end, data_end);
2648 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(sizeof(void *)),
2649 dst_reg, src_reg, ctx_off);
2650 break;
2651
d691f9e8
AS		 2652 case offsetof(struct __sk_buff, tc_index):
2653#ifdef CONFIG_NET_SCHED
2654 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tc_index) != 2);
2655
2656 if (type == BPF_WRITE)
2657 *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg,
2658 offsetof(struct sk_buff, tc_index));
2659 else
2660 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
2661 offsetof(struct sk_buff, tc_index));
2662 break;
2663#else
2664 if (type == BPF_WRITE)
2665 *insn++ = BPF_MOV64_REG(dst_reg, dst_reg);
2666 else
2667 *insn++ = BPF_MOV64_IMM(dst_reg, 0);
2668 break;
2669#endif
9bac3d6d
AS		 2670 }
2671
2672 return insn - insn_buf;
89aa0758
AS		 2673}
2674
6a773a15
BB		 2675static u32 xdp_convert_ctx_access(enum bpf_access_type type, int dst_reg,
2676 int src_reg, int ctx_off,
2677 struct bpf_insn *insn_buf,
2678 struct bpf_prog *prog)
2679{
2680 struct bpf_insn *insn = insn_buf;
2681
2682 switch (ctx_off) {
2683 case offsetof(struct xdp_md, data):
2684 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct xdp_buff, data)),
2685 dst_reg, src_reg,
2686 offsetof(struct xdp_buff, data));
2687 break;
2688 case offsetof(struct xdp_md, data_end):
2689 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct xdp_buff, data_end)),
2690 dst_reg, src_reg,
2691 offsetof(struct xdp_buff, data_end));
2692 break;
2693 }
2694
2695 return insn - insn_buf;
2696}
2697
d4052c4a 2698static const struct bpf_verifier_ops sk_filter_ops = {
4936e352
DB		 2699 .get_func_proto = sk_filter_func_proto,
2700 .is_valid_access = sk_filter_is_valid_access,
2701 .convert_ctx_access = bpf_net_convert_ctx_access,
89aa0758
AS		 2702};
2703
608cd71a 2704static const struct bpf_verifier_ops tc_cls_act_ops = {
4936e352
DB		 2705 .get_func_proto = tc_cls_act_func_proto,
2706 .is_valid_access = tc_cls_act_is_valid_access,
2707 .convert_ctx_access = bpf_net_convert_ctx_access,
608cd71a
AS		 2708};
2709
6a773a15
BB		 2710static const struct bpf_verifier_ops xdp_ops = {
2711 .get_func_proto = xdp_func_proto,
2712 .is_valid_access = xdp_is_valid_access,
2713 .convert_ctx_access = xdp_convert_ctx_access,
2714};
2715
d4052c4a 2716static struct bpf_prog_type_list sk_filter_type __read_mostly = {
4936e352
DB		 2717 .ops = &sk_filter_ops,
2718 .type = BPF_PROG_TYPE_SOCKET_FILTER,
89aa0758
AS		 2719};
2720
96be4325 2721static struct bpf_prog_type_list sched_cls_type __read_mostly = {
4936e352
DB		 2722 .ops = &tc_cls_act_ops,
2723 .type = BPF_PROG_TYPE_SCHED_CLS,
96be4325
DB		 2724};
2725
94caee8c 2726static struct bpf_prog_type_list sched_act_type __read_mostly = {
4936e352
DB		 2727 .ops = &tc_cls_act_ops,
2728 .type = BPF_PROG_TYPE_SCHED_ACT,
94caee8c
DB		 2729};
2730
6a773a15
BB		 2731static struct bpf_prog_type_list xdp_type __read_mostly = {
2732 .ops = &xdp_ops,
2733 .type = BPF_PROG_TYPE_XDP,
2734};
2735
d4052c4a 2736static int __init register_sk_filter_ops(void)
89aa0758 2737{
d4052c4a 2738 bpf_register_prog_type(&sk_filter_type);
96be4325 2739 bpf_register_prog_type(&sched_cls_type);
94caee8c 2740 bpf_register_prog_type(&sched_act_type);
6a773a15 2741 bpf_register_prog_type(&xdp_type);
96be4325 2742
89aa0758
AS		 2743 return 0;
2744}
d4052c4a
DB		 2745late_initcall(register_sk_filter_ops);
2746
8ced425e 2747int sk_detach_filter(struct sock *sk)
55b33325
PE		 2748{
2749 int ret = -ENOENT;
2750 struct sk_filter *filter;
2751
d59577b6
VB		 2752 if (sock_flag(sk, SOCK_FILTER_LOCKED))
2753 return -EPERM;
2754
8ced425e
HFS		 2755 filter = rcu_dereference_protected(sk->sk_filter,
2756 lockdep_sock_is_held(sk));
55b33325 2757 if (filter) {
a9b3cd7f 2758 RCU_INIT_POINTER(sk->sk_filter, NULL);
46bcf14f 2759 sk_filter_uncharge(sk, filter);
55b33325
PE		 2760 ret = 0;
2761 }
a3ea269b 2762
55b33325
PE		 2763 return ret;
2764}
8ced425e 2765EXPORT_SYMBOL_GPL(sk_detach_filter);
a8fc9277 2766
a3ea269b
DB		 2767int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
2768 unsigned int len)
a8fc9277 2769{
a3ea269b 2770 struct sock_fprog_kern *fprog;
a8fc9277 2771 struct sk_filter *filter;
a3ea269b 2772 int ret = 0;
a8fc9277
PE		 2773
2774 lock_sock(sk);
2775 filter = rcu_dereference_protected(sk->sk_filter,
8ced425e 2776 lockdep_sock_is_held(sk));
a8fc9277
PE		 2777 if (!filter)
2778 goto out;
a3ea269b
DB		 2779
2780 /* We're copying the filter that has been originally attached,
93d08b69
DB		 2781 * so no conversion/decode needed anymore. eBPF programs that
2782 * have no original program cannot be dumped through this.
a3ea269b 2783 */
93d08b69 2784 ret = -EACCES;
7ae457c1 2785 fprog = filter->prog->orig_prog;
93d08b69
DB		 2786 if (!fprog)
2787 goto out;
a3ea269b
DB		 2788
2789 ret = fprog->len;
a8fc9277 2790 if (!len)
a3ea269b 2791 /* User space only enquires number of filter blocks. */
a8fc9277 2792 goto out;
a3ea269b 2793
a8fc9277 2794 ret = -EINVAL;
a3ea269b 2795 if (len < fprog->len)
a8fc9277
PE		 2796 goto out;
2797
2798 ret = -EFAULT;
009937e7 2799 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
a3ea269b 2800 goto out;
a8fc9277 2801
a3ea269b
DB		 2802 /* Instead of bytes, the API requests to return the number
2803 * of filter blocks.
2804 */
2805 ret = fprog->len;
a8fc9277
PE		 2806out:
2807 release_sock(sk);
2808 return ret;
2809}
Linux kernel - Deliverables
This page took 1.0496 seconds and 5 git commands to generate.