projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'bpf-perf'
[deliverable/linux.git] / net / openvswitch / flow_netlink.c
CommitLineData
e6445719 1/*
971427f3 2 * Copyright (c) 2007-2014 Nicira, Inc.
e6445719
PS		 3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
2235ad1c
JP		 19#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
e6445719
PS		 21#include "flow.h"
22#include "datapath.h"
23#include <linux/uaccess.h>
24#include <linux/netdevice.h>
25#include <linux/etherdevice.h>
26#include <linux/if_ether.h>
27#include <linux/if_vlan.h>
28#include <net/llc_pdu.h>
29#include <linux/kernel.h>
30#include <linux/jhash.h>
31#include <linux/jiffies.h>
32#include <linux/llc.h>
33#include <linux/module.h>
34#include <linux/in.h>
35#include <linux/rcupdate.h>
36#include <linux/if_arp.h>
37#include <linux/ip.h>
38#include <linux/ipv6.h>
39#include <linux/sctp.h>
40#include <linux/tcp.h>
41#include <linux/udp.h>
42#include <linux/icmp.h>
43#include <linux/icmpv6.h>
44#include <linux/rculist.h>
f5796684 45#include <net/geneve.h>
e6445719
PS		 46#include <net/ip.h>
47#include <net/ipv6.h>
48#include <net/ndisc.h>
25cd9ba0 49#include <net/mpls.h>
614732ea 50#include <net/vxlan.h>
e6445719
PS		 51
52#include "flow_netlink.h"
53
81bfe3c3
TG		 54struct ovs_len_tbl {
55 int len;
56 const struct ovs_len_tbl *next;
57};
58
59#define OVS_ATTR_NESTED -1
982b5270 60#define OVS_ATTR_VARIABLE -2
81bfe3c3 61
a85311bf
PS		 62static void update_range(struct sw_flow_match *match,
63 size_t offset, size_t size, bool is_mask)
e6445719 64{
a85311bf 65 struct sw_flow_key_range *range;
e6445719
PS		 66 size_t start = rounddown(offset, sizeof(long));
67 size_t end = roundup(offset + size, sizeof(long));
68
69 if (!is_mask)
70 range = &match->range;
a85311bf 71 else
e6445719
PS		 72 range = &match->mask->range;
73
e6445719
PS		 74 if (range->start == range->end) {
75 range->start = start;
76 range->end = end;
77 return;
78 }
79
80 if (range->start > start)
81 range->start = start;
82
83 if (range->end < end)
84 range->end = end;
85}
86
87#define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
88 do { \
a85311bf
PS		 89 update_range(match, offsetof(struct sw_flow_key, field), \
90 sizeof((match)->key->field), is_mask); \
91 if (is_mask) \
92 (match)->mask->key.field = value; \
93 else \
e6445719 94 (match)->key->field = value; \
e6445719
PS		 95 } while (0)
96
f5796684
JG		 97#define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask) \
98 do { \
a85311bf 99 update_range(match, offset, len, is_mask); \
f5796684
JG		 100 if (is_mask) \
101 memcpy((u8 *)&(match)->mask->key + offset, value_p, \
a85311bf 102 len); \
f5796684
JG		 103 else \
104 memcpy((u8 *)(match)->key + offset, value_p, len); \
e6445719
PS		 105 } while (0)
106
f5796684
JG		 107#define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
108 SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
109 value_p, len, is_mask)
110
a85311bf
PS		 111#define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask) \
112 do { \
113 update_range(match, offsetof(struct sw_flow_key, field), \
114 sizeof((match)->key->field), is_mask); \
115 if (is_mask) \
116 memset((u8 *)&(match)->mask->key.field, value, \
117 sizeof((match)->mask->key.field)); \
118 else \
f47de068
PS		 119 memset((u8 *)&(match)->key->field, value, \
120 sizeof((match)->key->field)); \
f47de068 121 } while (0)
e6445719
PS		 122
123static bool match_validate(const struct sw_flow_match *match,
05da5898 124 u64 key_attrs, u64 mask_attrs, bool log)
e6445719
PS		 125{
126 u64 key_expected = 1 << OVS_KEY_ATTR_ETHERNET;
127 u64 mask_allowed = key_attrs; /* At most allow all key attributes */
128
129 /* The following mask attributes allowed only if they
130 * pass the validation tests. */
131 mask_allowed &= ~((1 << OVS_KEY_ATTR_IPV4)
132 | (1 << OVS_KEY_ATTR_IPV6)
133 | (1 << OVS_KEY_ATTR_TCP)
5eb26b15 134 | (1 << OVS_KEY_ATTR_TCP_FLAGS)
e6445719
PS		 135 | (1 << OVS_KEY_ATTR_UDP)
136 | (1 << OVS_KEY_ATTR_SCTP)
137 | (1 << OVS_KEY_ATTR_ICMP)
138 | (1 << OVS_KEY_ATTR_ICMPV6)
139 | (1 << OVS_KEY_ATTR_ARP)
25cd9ba0
SH		 140 | (1 << OVS_KEY_ATTR_ND)
141 | (1 << OVS_KEY_ATTR_MPLS));
e6445719
PS		 142
143 /* Always allowed mask fields. */
144 mask_allowed |= ((1 << OVS_KEY_ATTR_TUNNEL)
145 | (1 << OVS_KEY_ATTR_IN_PORT)
146 | (1 << OVS_KEY_ATTR_ETHERTYPE));
147
148 /* Check key attributes. */
149 if (match->key->eth.type == htons(ETH_P_ARP)
150 || match->key->eth.type == htons(ETH_P_RARP)) {
151 key_expected |= 1 << OVS_KEY_ATTR_ARP;
f2a01517 152 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
e6445719
PS		 153 mask_allowed |= 1 << OVS_KEY_ATTR_ARP;
154 }
155
25cd9ba0
SH		 156 if (eth_p_mpls(match->key->eth.type)) {
157 key_expected |= 1 << OVS_KEY_ATTR_MPLS;
158 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
159 mask_allowed |= 1 << OVS_KEY_ATTR_MPLS;
160 }
161
e6445719
PS		 162 if (match->key->eth.type == htons(ETH_P_IP)) {
163 key_expected |= 1 << OVS_KEY_ATTR_IPV4;
164 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
165 mask_allowed |= 1 << OVS_KEY_ATTR_IPV4;
166
167 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
168 if (match->key->ip.proto == IPPROTO_UDP) {
169 key_expected |= 1 << OVS_KEY_ATTR_UDP;
170 if (match->mask && (match->mask->key.ip.proto == 0xff))
171 mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
172 }
173
174 if (match->key->ip.proto == IPPROTO_SCTP) {
175 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
176 if (match->mask && (match->mask->key.ip.proto == 0xff))
177 mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
178 }
179
180 if (match->key->ip.proto == IPPROTO_TCP) {
181 key_expected |= 1 << OVS_KEY_ATTR_TCP;
5eb26b15
JR		 182 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
183 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
e6445719 184 mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
5eb26b15
JR		 185 mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
186 }
e6445719
PS		 187 }
188
189 if (match->key->ip.proto == IPPROTO_ICMP) {
190 key_expected |= 1 << OVS_KEY_ATTR_ICMP;
191 if (match->mask && (match->mask->key.ip.proto == 0xff))
192 mask_allowed |= 1 << OVS_KEY_ATTR_ICMP;
193 }
194 }
195 }
196
197 if (match->key->eth.type == htons(ETH_P_IPV6)) {
198 key_expected |= 1 << OVS_KEY_ATTR_IPV6;
199 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
200 mask_allowed |= 1 << OVS_KEY_ATTR_IPV6;
201
202 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
203 if (match->key->ip.proto == IPPROTO_UDP) {
204 key_expected |= 1 << OVS_KEY_ATTR_UDP;
205 if (match->mask && (match->mask->key.ip.proto == 0xff))
206 mask_allowed |= 1 << OVS_KEY_ATTR_UDP;
207 }
208
209 if (match->key->ip.proto == IPPROTO_SCTP) {
210 key_expected |= 1 << OVS_KEY_ATTR_SCTP;
211 if (match->mask && (match->mask->key.ip.proto == 0xff))
212 mask_allowed |= 1 << OVS_KEY_ATTR_SCTP;
213 }
214
215 if (match->key->ip.proto == IPPROTO_TCP) {
216 key_expected |= 1 << OVS_KEY_ATTR_TCP;
5eb26b15
JR		 217 key_expected |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
218 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
e6445719 219 mask_allowed |= 1 << OVS_KEY_ATTR_TCP;
5eb26b15
JR		 220 mask_allowed |= 1 << OVS_KEY_ATTR_TCP_FLAGS;
221 }
e6445719
PS		 222 }
223
224 if (match->key->ip.proto == IPPROTO_ICMPV6) {
225 key_expected |= 1 << OVS_KEY_ATTR_ICMPV6;
226 if (match->mask && (match->mask->key.ip.proto == 0xff))
227 mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6;
228
1139e241 229 if (match->key->tp.src ==
e6445719 230 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
1139e241 231 match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
e6445719 232 key_expected |= 1 << OVS_KEY_ATTR_ND;
f2a01517 233 if (match->mask && (match->mask->key.tp.src == htons(0xff)))
e6445719
PS		 234 mask_allowed |= 1 << OVS_KEY_ATTR_ND;
235 }
236 }
237 }
238 }
239
240 if ((key_attrs & key_expected) != key_expected) {
241 /* Key attributes check failed. */
05da5898
JR		 242 OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
243 (unsigned long long)key_attrs,
244 (unsigned long long)key_expected);
e6445719
PS		 245 return false;
246 }
247
248 if ((mask_attrs & mask_allowed) != mask_attrs) {
249 /* Mask attributes check failed. */
05da5898
JR		 250 OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
251 (unsigned long long)mask_attrs,
252 (unsigned long long)mask_allowed);
e6445719
PS		 253 return false;
254 }
255
256 return true;
257}
258
8f0aad6f
WZ		 259size_t ovs_tun_key_attr_size(void)
260{
261 /* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
262 * updating this function.
263 */
264 return nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
6b26ba3a
JB		 265 + nla_total_size(16) /* OVS_TUNNEL_KEY_ATTR_IPV[46]_SRC */
266 + nla_total_size(16) /* OVS_TUNNEL_KEY_ATTR_IPV[46]_DST */
8f0aad6f
WZ		 267 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
268 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
269 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
270 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
271 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_OAM */
272 + nla_total_size(256) /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
1dd144cf
TG		 273 /* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS is mutually exclusive with
274 * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
275 */
8f0aad6f
WZ		 276 + nla_total_size(2) /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
277 + nla_total_size(2); /* OVS_TUNNEL_KEY_ATTR_TP_DST */
278}
279
41af73e9
JS		 280size_t ovs_key_attr_size(void)
281{
282 /* Whenever adding new OVS_KEY_ FIELDS, we should consider
283 * updating this function.
284 */
c2ac6673 285 BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 26);
41af73e9
JS		 286
287 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
288 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
8f0aad6f 289 + ovs_tun_key_attr_size()
41af73e9
JS		 290 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
291 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
292 + nla_total_size(4) /* OVS_KEY_ATTR_DP_HASH */
293 + nla_total_size(4) /* OVS_KEY_ATTR_RECIRC_ID */
fbccce59 294 + nla_total_size(4) /* OVS_KEY_ATTR_CT_STATE */
7f8a436e 295 + nla_total_size(2) /* OVS_KEY_ATTR_CT_ZONE */
182e3042 296 + nla_total_size(4) /* OVS_KEY_ATTR_CT_MARK */
33db4125 297 + nla_total_size(16) /* OVS_KEY_ATTR_CT_LABELS */
41af73e9
JS		 298 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
299 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
300 + nla_total_size(4) /* OVS_KEY_ATTR_VLAN */
301 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
302 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
303 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
304 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
305 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
306}
307
982b5270
JG		 308static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
309 [OVS_VXLAN_EXT_GBP] = { .len = sizeof(u32) },
310};
311
81bfe3c3
TG		 312static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
313 [OVS_TUNNEL_KEY_ATTR_ID] = { .len = sizeof(u64) },
314 [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = { .len = sizeof(u32) },
315 [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = { .len = sizeof(u32) },
316 [OVS_TUNNEL_KEY_ATTR_TOS] = { .len = 1 },
317 [OVS_TUNNEL_KEY_ATTR_TTL] = { .len = 1 },
318 [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
319 [OVS_TUNNEL_KEY_ATTR_CSUM] = { .len = 0 },
320 [OVS_TUNNEL_KEY_ATTR_TP_SRC] = { .len = sizeof(u16) },
321 [OVS_TUNNEL_KEY_ATTR_TP_DST] = { .len = sizeof(u16) },
322 [OVS_TUNNEL_KEY_ATTR_OAM] = { .len = 0 },
982b5270
JG		 323 [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS] = { .len = OVS_ATTR_VARIABLE },
324 [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS] = { .len = OVS_ATTR_NESTED,
325 .next = ovs_vxlan_ext_key_lens },
6b26ba3a
JB		 326 [OVS_TUNNEL_KEY_ATTR_IPV6_SRC] = { .len = sizeof(struct in6_addr) },
327 [OVS_TUNNEL_KEY_ATTR_IPV6_DST] = { .len = sizeof(struct in6_addr) },
81bfe3c3
TG		 328};
329
e6445719 330/* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute. */
81bfe3c3
TG		 331static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
332 [OVS_KEY_ATTR_ENCAP] = { .len = OVS_ATTR_NESTED },
333 [OVS_KEY_ATTR_PRIORITY] = { .len = sizeof(u32) },
334 [OVS_KEY_ATTR_IN_PORT] = { .len = sizeof(u32) },
335 [OVS_KEY_ATTR_SKB_MARK] = { .len = sizeof(u32) },
336 [OVS_KEY_ATTR_ETHERNET] = { .len = sizeof(struct ovs_key_ethernet) },
337 [OVS_KEY_ATTR_VLAN] = { .len = sizeof(__be16) },
338 [OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
339 [OVS_KEY_ATTR_IPV4] = { .len = sizeof(struct ovs_key_ipv4) },
340 [OVS_KEY_ATTR_IPV6] = { .len = sizeof(struct ovs_key_ipv6) },
341 [OVS_KEY_ATTR_TCP] = { .len = sizeof(struct ovs_key_tcp) },
342 [OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
343 [OVS_KEY_ATTR_UDP] = { .len = sizeof(struct ovs_key_udp) },
344 [OVS_KEY_ATTR_SCTP] = { .len = sizeof(struct ovs_key_sctp) },
345 [OVS_KEY_ATTR_ICMP] = { .len = sizeof(struct ovs_key_icmp) },
346 [OVS_KEY_ATTR_ICMPV6] = { .len = sizeof(struct ovs_key_icmpv6) },
347 [OVS_KEY_ATTR_ARP] = { .len = sizeof(struct ovs_key_arp) },
348 [OVS_KEY_ATTR_ND] = { .len = sizeof(struct ovs_key_nd) },
349 [OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
350 [OVS_KEY_ATTR_DP_HASH] = { .len = sizeof(u32) },
351 [OVS_KEY_ATTR_TUNNEL] = { .len = OVS_ATTR_NESTED,
352 .next = ovs_tunnel_key_lens, },
353 [OVS_KEY_ATTR_MPLS] = { .len = sizeof(struct ovs_key_mpls) },
fbccce59 354 [OVS_KEY_ATTR_CT_STATE] = { .len = sizeof(u32) },
7f8a436e 355 [OVS_KEY_ATTR_CT_ZONE] = { .len = sizeof(u16) },
182e3042 356 [OVS_KEY_ATTR_CT_MARK] = { .len = sizeof(u32) },
33db4125 357 [OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
e6445719
PS		 358};
359
982b5270
JG		 360static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
361{
362 return expected_len == attr_len ||
363 expected_len == OVS_ATTR_NESTED ||
364 expected_len == OVS_ATTR_VARIABLE;
365}
366
e6445719
PS		 367static bool is_all_zero(const u8 *fp, size_t size)
368{
369 int i;
370
371 if (!fp)
372 return false;
373
374 for (i = 0; i < size; i++)
375 if (fp[i])
376 return false;
377
378 return true;
379}
380
381static int __parse_flow_nlattrs(const struct nlattr *attr,
382 const struct nlattr *a[],
05da5898 383 u64 *attrsp, bool log, bool nz)
e6445719
PS		 384{
385 const struct nlattr *nla;
386 u64 attrs;
387 int rem;
388
389 attrs = *attrsp;
390 nla_for_each_nested(nla, attr, rem) {
391 u16 type = nla_type(nla);
392 int expected_len;
393
394 if (type > OVS_KEY_ATTR_MAX) {
05da5898 395 OVS_NLERR(log, "Key type %d is out of range max %d",
e6445719
PS		 396 type, OVS_KEY_ATTR_MAX);
397 return -EINVAL;
398 }
399
400 if (attrs & (1 << type)) {
05da5898 401 OVS_NLERR(log, "Duplicate key (type %d).", type);
e6445719
PS		 402 return -EINVAL;
403 }
404
81bfe3c3 405 expected_len = ovs_key_lens[type].len;
982b5270 406 if (!check_attr_len(nla_len(nla), expected_len)) {
05da5898
JR		 407 OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
408 type, nla_len(nla), expected_len);
e6445719
PS		 409 return -EINVAL;
410 }
411
412 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
413 attrs |= 1 << type;
414 a[type] = nla;
415 }
416 }
417 if (rem) {
05da5898 418 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
e6445719
PS		 419 return -EINVAL;
420 }
421
422 *attrsp = attrs;
423 return 0;
424}
425
426static int parse_flow_mask_nlattrs(const struct nlattr *attr,
05da5898
JR		 427 const struct nlattr *a[], u64 *attrsp,
428 bool log)
e6445719 429{
05da5898 430 return __parse_flow_nlattrs(attr, a, attrsp, log, true);
e6445719
PS		 431}
432
433static int parse_flow_nlattrs(const struct nlattr *attr,
05da5898
JR		 434 const struct nlattr *a[], u64 *attrsp,
435 bool log)
e6445719 436{
05da5898
JR		 437 return __parse_flow_nlattrs(attr, a, attrsp, log, false);
438}
439
440static int genev_tun_opt_from_nlattr(const struct nlattr *a,
441 struct sw_flow_match *match, bool is_mask,
442 bool log)
443{
444 unsigned long opt_key_offset;
445
446 if (nla_len(a) > sizeof(match->key->tun_opts)) {
447 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
448 nla_len(a), sizeof(match->key->tun_opts));
449 return -EINVAL;
450 }
451
452 if (nla_len(a) % 4 != 0) {
453 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
454 nla_len(a));
455 return -EINVAL;
456 }
457
458 /* We need to record the length of the options passed
459 * down, otherwise packets with the same format but
460 * additional options will be silently matched.
461 */
462 if (!is_mask) {
463 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
464 false);
465 } else {
466 /* This is somewhat unusual because it looks at
467 * both the key and mask while parsing the
468 * attributes (and by extension assumes the key
469 * is parsed first). Normally, we would verify
470 * that each is the correct length and that the
471 * attributes line up in the validate function.
472 * However, that is difficult because this is
473 * variable length and we won't have the
474 * information later.
475 */
476 if (match->key->tun_opts_len != nla_len(a)) {
477 OVS_NLERR(log, "Geneve option len %d != mask len %d",
478 match->key->tun_opts_len, nla_len(a));
479 return -EINVAL;
480 }
481
482 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
483 }
484
d91641d9 485 opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
05da5898
JR		 486 SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
487 nla_len(a), is_mask);
488 return 0;
e6445719
PS		 489}
490
982b5270 491static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
1dd144cf
TG		 492 struct sw_flow_match *match, bool is_mask,
493 bool log)
494{
982b5270
JG		 495 struct nlattr *a;
496 int rem;
1dd144cf 497 unsigned long opt_key_offset;
614732ea 498 struct vxlan_metadata opts;
1dd144cf
TG		 499
500 BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
501
1dd144cf 502 memset(&opts, 0, sizeof(opts));
982b5270
JG		 503 nla_for_each_nested(a, attr, rem) {
504 int type = nla_type(a);
1dd144cf 505
982b5270
JG		 506 if (type > OVS_VXLAN_EXT_MAX) {
507 OVS_NLERR(log, "VXLAN extension %d out of range max %d",
508 type, OVS_VXLAN_EXT_MAX);
509 return -EINVAL;
510 }
511
512 if (!check_attr_len(nla_len(a),
513 ovs_vxlan_ext_key_lens[type].len)) {
514 OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
515 type, nla_len(a),
516 ovs_vxlan_ext_key_lens[type].len);
517 return -EINVAL;
518 }
519
520 switch (type) {
521 case OVS_VXLAN_EXT_GBP:
522 opts.gbp = nla_get_u32(a);
523 break;
524 default:
525 OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
526 type);
527 return -EINVAL;
528 }
529 }
530 if (rem) {
531 OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
532 rem);
533 return -EINVAL;
534 }
1dd144cf
TG		 535
536 if (!is_mask)
537 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
538 else
539 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
540
541 opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
542 SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
543 is_mask);
544 return 0;
545}
546
6b26ba3a
JB		 547static int ip_tun_from_nlattr(const struct nlattr *attr,
548 struct sw_flow_match *match, bool is_mask,
549 bool log)
e6445719
PS		 550{
551 struct nlattr *a;
552 int rem;
553 bool ttl = false;
6b26ba3a 554 __be16 tun_flags = 0, ipv4 = false, ipv6 = false;
1dd144cf 555 int opts_type = 0;
e6445719
PS		 556
557 nla_for_each_nested(a, attr, rem) {
558 int type = nla_type(a);
05da5898
JR		 559 int err;
560
e6445719 561 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
05da5898
JR		 562 OVS_NLERR(log, "Tunnel attr %d out of range max %d",
563 type, OVS_TUNNEL_KEY_ATTR_MAX);
e6445719
PS		 564 return -EINVAL;
565 }
566
982b5270
JG		 567 if (!check_attr_len(nla_len(a),
568 ovs_tunnel_key_lens[type].len)) {
05da5898 569 OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
81bfe3c3 570 type, nla_len(a), ovs_tunnel_key_lens[type].len);
e6445719
PS		 571 return -EINVAL;
572 }
573
574 switch (type) {
575 case OVS_TUNNEL_KEY_ATTR_ID:
576 SW_FLOW_KEY_PUT(match, tun_key.tun_id,
577 nla_get_be64(a), is_mask);
578 tun_flags |= TUNNEL_KEY;
579 break;
580 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
c1ea5d67 581 SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
67b61f6c 582 nla_get_in_addr(a), is_mask);
6b26ba3a 583 ipv4 = true;
e6445719
PS		 584 break;
585 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
c1ea5d67 586 SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
67b61f6c 587 nla_get_in_addr(a), is_mask);
6b26ba3a
JB		 588 ipv4 = true;
589 break;
590 case OVS_TUNNEL_KEY_ATTR_IPV6_SRC:
591 SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
592 nla_get_in6_addr(a), is_mask);
593 ipv6 = true;
594 break;
595 case OVS_TUNNEL_KEY_ATTR_IPV6_DST:
596 SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
597 nla_get_in6_addr(a), is_mask);
598 ipv6 = true;
e6445719
PS		 599 break;
600 case OVS_TUNNEL_KEY_ATTR_TOS:
7c383fb2 601 SW_FLOW_KEY_PUT(match, tun_key.tos,
e6445719
PS		 602 nla_get_u8(a), is_mask);
603 break;
604 case OVS_TUNNEL_KEY_ATTR_TTL:
7c383fb2 605 SW_FLOW_KEY_PUT(match, tun_key.ttl,
e6445719
PS		 606 nla_get_u8(a), is_mask);
607 ttl = true;
608 break;
609 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
610 tun_flags |= TUNNEL_DONT_FRAGMENT;
611 break;
612 case OVS_TUNNEL_KEY_ATTR_CSUM:
613 tun_flags |= TUNNEL_CSUM;
614 break;
8f0aad6f
WZ		 615 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
616 SW_FLOW_KEY_PUT(match, tun_key.tp_src,
617 nla_get_be16(a), is_mask);
618 break;
619 case OVS_TUNNEL_KEY_ATTR_TP_DST:
620 SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
621 nla_get_be16(a), is_mask);
622 break;
67fa0341
JG		 623 case OVS_TUNNEL_KEY_ATTR_OAM:
624 tun_flags |= TUNNEL_OAM;
625 break;
f5796684 626 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1dd144cf
TG		 627 if (opts_type) {
628 OVS_NLERR(log, "Multiple metadata blocks provided");
629 return -EINVAL;
630 }
631
05da5898
JR		 632 err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
633 if (err)
634 return err;
f5796684 635
1dd144cf
TG		 636 tun_flags |= TUNNEL_GENEVE_OPT;
637 opts_type = type;
638 break;
639 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
640 if (opts_type) {
641 OVS_NLERR(log, "Multiple metadata blocks provided");
642 return -EINVAL;
643 }
644
645 err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
646 if (err)
647 return err;
648
649 tun_flags |= TUNNEL_VXLAN_OPT;
650 opts_type = type;
f5796684 651 break;
e6445719 652 default:
6b26ba3a 653 OVS_NLERR(log, "Unknown IP tunnel attribute %d",
f5796684 654 type);
e6445719
PS		 655 return -EINVAL;
656 }
657 }
658
659 SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
00a93bab
JB		 660 if (is_mask)
661 SW_FLOW_KEY_MEMSET_FIELD(match, tun_proto, 0xff, true);
662 else
6b26ba3a
JB		 663 SW_FLOW_KEY_PUT(match, tun_proto, ipv6 ? AF_INET6 : AF_INET,
664 false);
e6445719
PS		 665
666 if (rem > 0) {
6b26ba3a 667 OVS_NLERR(log, "IP tunnel attribute has %d unknown bytes.",
05da5898 668 rem);
e6445719
PS		 669 return -EINVAL;
670 }
671
6b26ba3a
JB		 672 if (ipv4 && ipv6) {
673 OVS_NLERR(log, "Mixed IPv4 and IPv6 tunnel attributes");
674 return -EINVAL;
675 }
676
e6445719 677 if (!is_mask) {
6b26ba3a
JB		 678 if (!ipv4 && !ipv6) {
679 OVS_NLERR(log, "IP tunnel dst address not specified");
680 return -EINVAL;
681 }
682 if (ipv4 && !match->key->tun_key.u.ipv4.dst) {
05da5898 683 OVS_NLERR(log, "IPv4 tunnel dst address is zero");
e6445719
PS		 684 return -EINVAL;
685 }
6b26ba3a
JB		 686 if (ipv6 && ipv6_addr_any(&match->key->tun_key.u.ipv6.dst)) {
687 OVS_NLERR(log, "IPv6 tunnel dst address is zero");
688 return -EINVAL;
689 }
e6445719
PS		 690
691 if (!ttl) {
6b26ba3a 692 OVS_NLERR(log, "IP tunnel TTL not specified.");
e6445719
PS		 693 return -EINVAL;
694 }
695 }
696
1dd144cf
TG		 697 return opts_type;
698}
699
700static int vxlan_opt_to_nlattr(struct sk_buff *skb,
701 const void *tun_opts, int swkey_tun_opts_len)
702{
614732ea 703 const struct vxlan_metadata *opts = tun_opts;
1dd144cf
TG		 704 struct nlattr *nla;
705
706 nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
707 if (!nla)
708 return -EMSGSIZE;
709
710 if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
711 return -EMSGSIZE;
712
713 nla_nest_end(skb, nla);
e6445719
PS		 714 return 0;
715}
716
6b26ba3a
JB		 717static int __ip_tun_to_nlattr(struct sk_buff *skb,
718 const struct ip_tunnel_key *output,
719 const void *tun_opts, int swkey_tun_opts_len,
720 unsigned short tun_proto)
e6445719 721{
e6445719
PS		 722 if (output->tun_flags & TUNNEL_KEY &&
723 nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
724 return -EMSGSIZE;
6b26ba3a
JB		 725 switch (tun_proto) {
726 case AF_INET:
727 if (output->u.ipv4.src &&
728 nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
729 output->u.ipv4.src))
730 return -EMSGSIZE;
731 if (output->u.ipv4.dst &&
732 nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
733 output->u.ipv4.dst))
734 return -EMSGSIZE;
735 break;
736 case AF_INET6:
737 if (!ipv6_addr_any(&output->u.ipv6.src) &&
738 nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
739 &output->u.ipv6.src))
740 return -EMSGSIZE;
741 if (!ipv6_addr_any(&output->u.ipv6.dst) &&
742 nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
743 &output->u.ipv6.dst))
744 return -EMSGSIZE;
745 break;
746 }
7c383fb2
JB		 747 if (output->tos &&
748 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
e6445719 749 return -EMSGSIZE;
7c383fb2 750 if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
e6445719
PS		 751 return -EMSGSIZE;
752 if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
67fa0341 753 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
e6445719
PS		 754 return -EMSGSIZE;
755 if ((output->tun_flags & TUNNEL_CSUM) &&
67fa0341
JG		 756 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
757 return -EMSGSIZE;
8f0aad6f
WZ		 758 if (output->tp_src &&
759 nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
760 return -EMSGSIZE;
761 if (output->tp_dst &&
762 nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
763 return -EMSGSIZE;
67fa0341
JG		 764 if ((output->tun_flags & TUNNEL_OAM) &&
765 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
e6445719 766 return -EMSGSIZE;
1dd144cf
TG		 767 if (tun_opts) {
768 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
769 nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
770 swkey_tun_opts_len, tun_opts))
771 return -EMSGSIZE;
772 else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
773 vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
774 return -EMSGSIZE;
775 }
e6445719 776
e6445719
PS		 777 return 0;
778}
779
6b26ba3a
JB		 780static int ip_tun_to_nlattr(struct sk_buff *skb,
781 const struct ip_tunnel_key *output,
782 const void *tun_opts, int swkey_tun_opts_len,
783 unsigned short tun_proto)
f5796684
JG		 784{
785 struct nlattr *nla;
786 int err;
787
788 nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
789 if (!nla)
790 return -EMSGSIZE;
791
6b26ba3a
JB		 792 err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
793 tun_proto);
f5796684
JG		 794 if (err)
795 return err;
796
797 nla_nest_end(skb, nla);
798 return 0;
799}
800
8f0aad6f 801int ovs_nla_put_egress_tunnel_key(struct sk_buff *skb,
4c222798
PS		 802 const struct ip_tunnel_info *egress_tun_info,
803 const void *egress_tun_opts)
8f0aad6f 804{
6b26ba3a
JB		 805 return __ip_tun_to_nlattr(skb, &egress_tun_info->key,
806 egress_tun_opts,
807 egress_tun_info->options_len,
808 ip_tunnel_info_af(egress_tun_info));
8f0aad6f
WZ		 809}
810
c2ac6673
JS		 811static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
812 u64 *attrs, const struct nlattr **a,
813 bool is_mask, bool log)
e6445719 814{
971427f3
AZ		 815 if (*attrs & (1 << OVS_KEY_ATTR_DP_HASH)) {
816 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
817
818 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
819 *attrs &= ~(1 << OVS_KEY_ATTR_DP_HASH);
820 }
821
822 if (*attrs & (1 << OVS_KEY_ATTR_RECIRC_ID)) {
823 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
824
825 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
826 *attrs &= ~(1 << OVS_KEY_ATTR_RECIRC_ID);
827 }
828
e6445719
PS		 829 if (*attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
830 SW_FLOW_KEY_PUT(match, phy.priority,
831 nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
832 *attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
833 }
834
835 if (*attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
836 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
837
426cda5c 838 if (is_mask) {
e6445719 839 in_port = 0xffffffff; /* Always exact match in_port. */
426cda5c 840 } else if (in_port >= DP_MAX_PORTS) {
05da5898 841 OVS_NLERR(log, "Port %d exceeds max allowable %d",
426cda5c 842 in_port, DP_MAX_PORTS);
e6445719 843 return -EINVAL;
426cda5c 844 }
e6445719
PS		 845
846 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
847 *attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
848 } else if (!is_mask) {
849 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
850 }
851
852 if (*attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
853 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
854
855 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
856 *attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
857 }
858 if (*attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
6b26ba3a
JB		 859 if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
860 is_mask, log) < 0)
e6445719
PS		 861 return -EINVAL;
862 *attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
863 }
7f8a436e
JS		 864
865 if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
c2ac6673 866 ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
fbccce59 867 u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
7f8a436e 868
6f225952 869 if (!is_mask && !ovs_ct_state_supported(ct_state)) {
fbccce59 870 OVS_NLERR(log, "ct_state flags %08x unsupported",
6f225952
JS		 871 ct_state);
872 return -EINVAL;
873 }
7f8a436e
JS		 874
875 SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
876 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
877 }
878 if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
c2ac6673 879 ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
7f8a436e
JS		 880 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
881
882 SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
883 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
884 }
182e3042 885 if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
c2ac6673 886 ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
182e3042
JS		 887 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
888
889 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
890 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
891 }
33db4125
JS		 892 if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
893 ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
894 const struct ovs_key_ct_labels *cl;
c2ac6673 895
33db4125
JS		 896 cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
897 SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
c2ac6673 898 sizeof(*cl), is_mask);
33db4125 899 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
c2ac6673 900 }
e6445719
PS		 901 return 0;
902}
903
c2ac6673
JS		 904static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
905 u64 attrs, const struct nlattr **a,
906 bool is_mask, bool log)
e6445719
PS		 907{
908 int err;
e6445719 909
c2ac6673 910 err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
e6445719
PS		 911 if (err)
912 return err;
913
914 if (attrs & (1 << OVS_KEY_ATTR_ETHERNET)) {
915 const struct ovs_key_ethernet *eth_key;
916
917 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
918 SW_FLOW_KEY_MEMCPY(match, eth.src,
919 eth_key->eth_src, ETH_ALEN, is_mask);
920 SW_FLOW_KEY_MEMCPY(match, eth.dst,
921 eth_key->eth_dst, ETH_ALEN, is_mask);
922 attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
923 }
924
925 if (attrs & (1 << OVS_KEY_ATTR_VLAN)) {
926 __be16 tci;
927
928 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
929 if (!(tci & htons(VLAN_TAG_PRESENT))) {
930 if (is_mask)
05da5898 931 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
e6445719 932 else
05da5898 933 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
e6445719
PS		 934
935 return -EINVAL;
936 }
937
938 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
939 attrs &= ~(1 << OVS_KEY_ATTR_VLAN);
a85311bf 940 }
e6445719
PS		 941
942 if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
943 __be16 eth_type;
944
945 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
946 if (is_mask) {
947 /* Always exact match EtherType. */
948 eth_type = htons(0xffff);
6713fc9b 949 } else if (!eth_proto_is_802_3(eth_type)) {
05da5898
JR		 950 OVS_NLERR(log, "EtherType %x is less than min %x",
951 ntohs(eth_type), ETH_P_802_3_MIN);
e6445719
PS		 952 return -EINVAL;
953 }
954
955 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
956 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
957 } else if (!is_mask) {
958 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
959 }
960
961 if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
962 const struct ovs_key_ipv4 *ipv4_key;
963
964 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
965 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
05da5898
JR		 966 OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
967 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
e6445719
PS		 968 return -EINVAL;
969 }
970 SW_FLOW_KEY_PUT(match, ip.proto,
971 ipv4_key->ipv4_proto, is_mask);
972 SW_FLOW_KEY_PUT(match, ip.tos,
973 ipv4_key->ipv4_tos, is_mask);
974 SW_FLOW_KEY_PUT(match, ip.ttl,
975 ipv4_key->ipv4_ttl, is_mask);
976 SW_FLOW_KEY_PUT(match, ip.frag,
977 ipv4_key->ipv4_frag, is_mask);
978 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
979 ipv4_key->ipv4_src, is_mask);
980 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
981 ipv4_key->ipv4_dst, is_mask);
982 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
983 }
984
985 if (attrs & (1 << OVS_KEY_ATTR_IPV6)) {
986 const struct ovs_key_ipv6 *ipv6_key;
987
988 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
989 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
05da5898
JR		 990 OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
991 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
e6445719
PS		 992 return -EINVAL;
993 }
fecaef85 994
d3052bb5 995 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
14591433 996 OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
fecaef85
JR		 997 ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
998 return -EINVAL;
999 }
1000
e6445719
PS		 1001 SW_FLOW_KEY_PUT(match, ipv6.label,
1002 ipv6_key->ipv6_label, is_mask);
1003 SW_FLOW_KEY_PUT(match, ip.proto,
1004 ipv6_key->ipv6_proto, is_mask);
1005 SW_FLOW_KEY_PUT(match, ip.tos,
1006 ipv6_key->ipv6_tclass, is_mask);
1007 SW_FLOW_KEY_PUT(match, ip.ttl,
1008 ipv6_key->ipv6_hlimit, is_mask);
1009 SW_FLOW_KEY_PUT(match, ip.frag,
1010 ipv6_key->ipv6_frag, is_mask);
1011 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1012 ipv6_key->ipv6_src,
1013 sizeof(match->key->ipv6.addr.src),
1014 is_mask);
1015 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1016 ipv6_key->ipv6_dst,
1017 sizeof(match->key->ipv6.addr.dst),
1018 is_mask);
1019
1020 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1021 }
1022
1023 if (attrs & (1 << OVS_KEY_ATTR_ARP)) {
1024 const struct ovs_key_arp *arp_key;
1025
1026 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1027 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
05da5898 1028 OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
e6445719
PS		 1029 arp_key->arp_op);
1030 return -EINVAL;
1031 }
1032
1033 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1034 arp_key->arp_sip, is_mask);
1035 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1036 arp_key->arp_tip, is_mask);
1037 SW_FLOW_KEY_PUT(match, ip.proto,
1038 ntohs(arp_key->arp_op), is_mask);
1039 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1040 arp_key->arp_sha, ETH_ALEN, is_mask);
1041 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1042 arp_key->arp_tha, ETH_ALEN, is_mask);
1043
1044 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1045 }
1046
25cd9ba0
SH		 1047 if (attrs & (1 << OVS_KEY_ATTR_MPLS)) {
1048 const struct ovs_key_mpls *mpls_key;
1049
1050 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1051 SW_FLOW_KEY_PUT(match, mpls.top_lse,
1052 mpls_key->mpls_lse, is_mask);
1053
1054 attrs &= ~(1 << OVS_KEY_ATTR_MPLS);
1055 }
1056
e6445719
PS		 1057 if (attrs & (1 << OVS_KEY_ATTR_TCP)) {
1058 const struct ovs_key_tcp *tcp_key;
1059
1060 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1139e241
JR		 1061 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1062 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
e6445719
PS		 1063 attrs &= ~(1 << OVS_KEY_ATTR_TCP);
1064 }
1065
5eb26b15 1066 if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) {
1b760fb9
JS		 1067 SW_FLOW_KEY_PUT(match, tp.flags,
1068 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1069 is_mask);
5eb26b15
JR		 1070 attrs &= ~(1 << OVS_KEY_ATTR_TCP_FLAGS);
1071 }
1072
e6445719
PS		 1073 if (attrs & (1 << OVS_KEY_ATTR_UDP)) {
1074 const struct ovs_key_udp *udp_key;
1075
1076 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1139e241
JR		 1077 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1078 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
e6445719
PS		 1079 attrs &= ~(1 << OVS_KEY_ATTR_UDP);
1080 }
1081
1082 if (attrs & (1 << OVS_KEY_ATTR_SCTP)) {
1083 const struct ovs_key_sctp *sctp_key;
1084
1085 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1139e241
JR		 1086 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1087 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
e6445719
PS		 1088 attrs &= ~(1 << OVS_KEY_ATTR_SCTP);
1089 }
1090
1091 if (attrs & (1 << OVS_KEY_ATTR_ICMP)) {
1092 const struct ovs_key_icmp *icmp_key;
1093
1094 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1139e241 1095 SW_FLOW_KEY_PUT(match, tp.src,
e6445719 1096 htons(icmp_key->icmp_type), is_mask);
1139e241 1097 SW_FLOW_KEY_PUT(match, tp.dst,
e6445719
PS		 1098 htons(icmp_key->icmp_code), is_mask);
1099 attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
1100 }
1101
1102 if (attrs & (1 << OVS_KEY_ATTR_ICMPV6)) {
1103 const struct ovs_key_icmpv6 *icmpv6_key;
1104
1105 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1139e241 1106 SW_FLOW_KEY_PUT(match, tp.src,
e6445719 1107 htons(icmpv6_key->icmpv6_type), is_mask);
1139e241 1108 SW_FLOW_KEY_PUT(match, tp.dst,
e6445719
PS		 1109 htons(icmpv6_key->icmpv6_code), is_mask);
1110 attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
1111 }
1112
1113 if (attrs & (1 << OVS_KEY_ATTR_ND)) {
1114 const struct ovs_key_nd *nd_key;
1115
1116 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1117 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1118 nd_key->nd_target,
1119 sizeof(match->key->ipv6.nd.target),
1120 is_mask);
1121 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1122 nd_key->nd_sll, ETH_ALEN, is_mask);
1123 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1124 nd_key->nd_tll, ETH_ALEN, is_mask);
1125 attrs &= ~(1 << OVS_KEY_ATTR_ND);
1126 }
1127
426cda5c 1128 if (attrs != 0) {
05da5898 1129 OVS_NLERR(log, "Unknown key attributes %llx",
426cda5c 1130 (unsigned long long)attrs);
e6445719 1131 return -EINVAL;
426cda5c 1132 }
e6445719
PS		 1133
1134 return 0;
1135}
1136
81bfe3c3
TG		 1137static void nlattr_set(struct nlattr *attr, u8 val,
1138 const struct ovs_len_tbl *tbl)
e6445719 1139{
f47de068
PS		 1140 struct nlattr *nla;
1141 int rem;
e6445719 1142
f47de068
PS		 1143 /* The nlattr stream should already have been validated */
1144 nla_for_each_nested(nla, attr, rem) {
982b5270
JG		 1145 if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED) {
1146 if (tbl[nla_type(nla)].next)
1147 tbl = tbl[nla_type(nla)].next;
1148 nlattr_set(nla, val, tbl);
1149 } else {
f47de068 1150 memset(nla_data(nla), val, nla_len(nla));
982b5270 1151 }
f47de068
PS		 1152 }
1153}
1154
1155static void mask_set_nlattr(struct nlattr *attr, u8 val)
1156{
81bfe3c3 1157 nlattr_set(attr, val, ovs_key_lens);
e6445719
PS		 1158}
1159
1160/**
1161 * ovs_nla_get_match - parses Netlink attributes into a flow key and
1162 * mask. In case the 'mask' is NULL, the flow is treated as exact match
1163 * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1164 * does not include any don't care bit.
c2ac6673 1165 * @net: Used to determine per-namespace field support.
e6445719
PS		 1166 * @match: receives the extracted flow match information.
1167 * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1168 * sequence. The fields should of the packet that triggered the creation
1169 * of this flow.
1170 * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1171 * attribute specifies the mask field of the wildcarded flow.
05da5898
JR		 1172 * @log: Boolean to allow kernel error logging. Normally true, but when
1173 * probing for feature compatibility this should be passed in as false to
1174 * suppress unnecessary error logging.
e6445719 1175 */
c2ac6673 1176int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
a85311bf 1177 const struct nlattr *nla_key,
05da5898
JR		 1178 const struct nlattr *nla_mask,
1179 bool log)
e6445719
PS		 1180{
1181 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1182 const struct nlattr *encap;
f47de068 1183 struct nlattr *newmask = NULL;
e6445719
PS		 1184 u64 key_attrs = 0;
1185 u64 mask_attrs = 0;
1186 bool encap_valid = false;
1187 int err;
1188
05da5898 1189 err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
e6445719
PS		 1190 if (err)
1191 return err;
1192
1193 if ((key_attrs & (1 << OVS_KEY_ATTR_ETHERNET)) &&
1194 (key_attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) &&
1195 (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1196 __be16 tci;
1197
1198 if (!((key_attrs & (1 << OVS_KEY_ATTR_VLAN)) &&
1199 (key_attrs & (1 << OVS_KEY_ATTR_ENCAP)))) {
05da5898 1200 OVS_NLERR(log, "Invalid Vlan frame.");
e6445719
PS		 1201 return -EINVAL;
1202 }
1203
1204 key_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1205 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1206 encap = a[OVS_KEY_ATTR_ENCAP];
1207 key_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1208 encap_valid = true;
1209
1210 if (tci & htons(VLAN_TAG_PRESENT)) {
05da5898 1211 err = parse_flow_nlattrs(encap, a, &key_attrs, log);
e6445719
PS		 1212 if (err)
1213 return err;
1214 } else if (!tci) {
1215 /* Corner case for truncated 802.1Q header. */
1216 if (nla_len(encap)) {
05da5898 1217 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
e6445719
PS		 1218 return -EINVAL;
1219 }
1220 } else {
05da5898 1221 OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
e6445719
PS		 1222 return -EINVAL;
1223 }
1224 }
1225
c2ac6673 1226 err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
e6445719
PS		 1227 if (err)
1228 return err;
1229
a85311bf
PS		 1230 if (match->mask) {
1231 if (!nla_mask) {
1232 /* Create an exact match mask. We need to set to 0xff
1233 * all the 'match->mask' fields that have been touched
1234 * in 'match->key'. We cannot simply memset
1235 * 'match->mask', because padding bytes and fields not
1236 * specified in 'match->key' should be left to 0.
1237 * Instead, we use a stream of netlink attributes,
1238 * copied from 'key' and set to 0xff.
1239 * ovs_key_from_nlattrs() will take care of filling
1240 * 'match->mask' appropriately.
1241 */
1242 newmask = kmemdup(nla_key,
1243 nla_total_size(nla_len(nla_key)),
1244 GFP_KERNEL);
1245 if (!newmask)
1246 return -ENOMEM;
f47de068 1247
a85311bf 1248 mask_set_nlattr(newmask, 0xff);
f47de068 1249
a85311bf
PS		 1250 /* The userspace does not send tunnel attributes that
1251 * are 0, but we should not wildcard them nonetheless.
1252 */
00a93bab 1253 if (match->key->tun_proto)
a85311bf
PS		 1254 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1255 0xff, true);
f47de068 1256
a85311bf
PS		 1257 nla_mask = newmask;
1258 }
f47de068 1259
05da5898 1260 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
e6445719 1261 if (err)
f47de068 1262 goto free_newmask;
e6445719 1263
a85311bf
PS		 1264 /* Always match on tci. */
1265 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1266
f47de068 1267 if (mask_attrs & 1 << OVS_KEY_ATTR_ENCAP) {
e6445719
PS		 1268 __be16 eth_type = 0;
1269 __be16 tci = 0;
1270
1271 if (!encap_valid) {
05da5898 1272 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
f47de068
PS		 1273 err = -EINVAL;
1274 goto free_newmask;
e6445719
PS		 1275 }
1276
1277 mask_attrs &= ~(1 << OVS_KEY_ATTR_ENCAP);
1278 if (a[OVS_KEY_ATTR_ETHERTYPE])
1279 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1280
1281 if (eth_type == htons(0xffff)) {
1282 mask_attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1283 encap = a[OVS_KEY_ATTR_ENCAP];
05da5898
JR		 1284 err = parse_flow_mask_nlattrs(encap, a,
1285 &mask_attrs, log);
f47de068
PS		 1286 if (err)
1287 goto free_newmask;
e6445719 1288 } else {
05da5898
JR		 1289 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
1290 ntohs(eth_type));
f47de068
PS		 1291 err = -EINVAL;
1292 goto free_newmask;
e6445719
PS		 1293 }
1294
1295 if (a[OVS_KEY_ATTR_VLAN])
1296 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1297
1298 if (!(tci & htons(VLAN_TAG_PRESENT))) {
05da5898
JR		 1299 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
1300 ntohs(tci));
f47de068
PS		 1301 err = -EINVAL;
1302 goto free_newmask;
e6445719
PS		 1303 }
1304 }
1305
c2ac6673
JS		 1306 err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1307 log);
e6445719 1308 if (err)
f47de068 1309 goto free_newmask;
e6445719
PS		 1310 }
1311
05da5898 1312 if (!match_validate(match, key_attrs, mask_attrs, log))
f47de068 1313 err = -EINVAL;
e6445719 1314
f47de068
PS		 1315free_newmask:
1316 kfree(newmask);
1317 return err;
e6445719
PS		 1318}
1319
74ed7ab9
JS		 1320static size_t get_ufid_len(const struct nlattr *attr, bool log)
1321{
1322 size_t len;
1323
1324 if (!attr)
1325 return 0;
1326
1327 len = nla_len(attr);
1328 if (len < 1 || len > MAX_UFID_LENGTH) {
1329 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1330 nla_len(attr), MAX_UFID_LENGTH);
1331 return 0;
1332 }
1333
1334 return len;
1335}
1336
1337/* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1338 * or false otherwise.
1339 */
1340bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1341 bool log)
1342{
1343 sfid->ufid_len = get_ufid_len(attr, log);
1344 if (sfid->ufid_len)
1345 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1346
1347 return sfid->ufid_len;
1348}
1349
1350int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1351 const struct sw_flow_key *key, bool log)
1352{
1353 struct sw_flow_key *new_key;
1354
1355 if (ovs_nla_get_ufid(sfid, ufid, log))
1356 return 0;
1357
1358 /* If UFID was not provided, use unmasked key. */
1359 new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1360 if (!new_key)
1361 return -ENOMEM;
1362 memcpy(new_key, key, sizeof(*key));
1363 sfid->unmasked_key = new_key;
1364
1365 return 0;
1366}
1367
1368u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1369{
1370 return attr ? nla_get_u32(attr) : 0;
1371}
1372
e6445719
PS		 1373/**
1374 * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
83c8df26 1375 * @key: Receives extracted in_port, priority, tun_key and skb_mark.
e6445719
PS		 1376 * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1377 * sequence.
05da5898
JR		 1378 * @log: Boolean to allow kernel error logging. Normally true, but when
1379 * probing for feature compatibility this should be passed in as false to
1380 * suppress unnecessary error logging.
e6445719
PS		 1381 *
1382 * This parses a series of Netlink attributes that form a flow key, which must
1383 * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1384 * get the metadata, that is, the parts of the flow key that cannot be
1385 * extracted from the packet itself.
1386 */
1387
c2ac6673 1388int ovs_nla_get_flow_metadata(struct net *net, const struct nlattr *attr,
05da5898
JR		 1389 struct sw_flow_key *key,
1390 bool log)
e6445719 1391{
e6445719 1392 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
83c8df26 1393 struct sw_flow_match match;
e6445719
PS		 1394 u64 attrs = 0;
1395 int err;
e6445719 1396
05da5898 1397 err = parse_flow_nlattrs(attr, a, &attrs, log);
e6445719
PS		 1398 if (err)
1399 return -EINVAL;
1400
1401 memset(&match, 0, sizeof(match));
83c8df26 1402 match.key = key;
e6445719 1403
7f8a436e 1404 memset(&key->ct, 0, sizeof(key->ct));
83c8df26 1405 key->phy.in_port = DP_MAX_PORTS;
e6445719 1406
c2ac6673 1407 return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
e6445719
PS		 1408}
1409
5b4237bb
JS		 1410static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1411 const struct sw_flow_key *output, bool is_mask,
1412 struct sk_buff *skb)
e6445719
PS		 1413{
1414 struct ovs_key_ethernet *eth_key;
1415 struct nlattr *nla, *encap;
e6445719 1416
971427f3
AZ		 1417 if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
1418 goto nla_put_failure;
1419
1420 if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
1421 goto nla_put_failure;
1422
e6445719
PS		 1423 if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1424 goto nla_put_failure;
1425
00a93bab 1426 if ((swkey->tun_proto || is_mask)) {
d91641d9 1427 const void *opts = NULL;
f5796684
JG		 1428
1429 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
d91641d9 1430 opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
f5796684 1431
6b26ba3a
JB		 1432 if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
1433 swkey->tun_opts_len, swkey->tun_proto))
f5796684
JG		 1434 goto nla_put_failure;
1435 }
e6445719
PS		 1436
1437 if (swkey->phy.in_port == DP_MAX_PORTS) {
1438 if (is_mask && (output->phy.in_port == 0xffff))
1439 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1440 goto nla_put_failure;
1441 } else {
1442 u16 upper_u16;
1443 upper_u16 = !is_mask ? 0 : 0xffff;
1444
1445 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1446 (upper_u16 << 16) | output->phy.in_port))
1447 goto nla_put_failure;
1448 }
1449
1450 if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1451 goto nla_put_failure;
1452
7f8a436e
JS		 1453 if (ovs_ct_put_key(output, skb))
1454 goto nla_put_failure;
1455
e6445719
PS		 1456 nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1457 if (!nla)
1458 goto nla_put_failure;
1459
1460 eth_key = nla_data(nla);
8c63ff09
JP		 1461 ether_addr_copy(eth_key->eth_src, output->eth.src);
1462 ether_addr_copy(eth_key->eth_dst, output->eth.dst);
e6445719
PS		 1463
1464 if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1465 __be16 eth_type;
1466 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1467 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1468 nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1469 goto nla_put_failure;
1470 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1471 if (!swkey->eth.tci)
1472 goto unencap;
1473 } else
1474 encap = NULL;
1475
1476 if (swkey->eth.type == htons(ETH_P_802_2)) {
1477 /*
1478 * Ethertype 802.2 is represented in the netlink with omitted
1479 * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1480 * 0xffff in the mask attribute. Ethertype can also
1481 * be wildcarded.
1482 */
1483 if (is_mask && output->eth.type)
1484 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1485 output->eth.type))
1486 goto nla_put_failure;
1487 goto unencap;
1488 }
1489
1490 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1491 goto nla_put_failure;
1492
1493 if (swkey->eth.type == htons(ETH_P_IP)) {
1494 struct ovs_key_ipv4 *ipv4_key;
1495
1496 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1497 if (!nla)
1498 goto nla_put_failure;
1499 ipv4_key = nla_data(nla);
1500 ipv4_key->ipv4_src = output->ipv4.addr.src;
1501 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1502 ipv4_key->ipv4_proto = output->ip.proto;
1503 ipv4_key->ipv4_tos = output->ip.tos;
1504 ipv4_key->ipv4_ttl = output->ip.ttl;
1505 ipv4_key->ipv4_frag = output->ip.frag;
1506 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1507 struct ovs_key_ipv6 *ipv6_key;
1508
1509 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1510 if (!nla)
1511 goto nla_put_failure;
1512 ipv6_key = nla_data(nla);
1513 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1514 sizeof(ipv6_key->ipv6_src));
1515 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1516 sizeof(ipv6_key->ipv6_dst));
1517 ipv6_key->ipv6_label = output->ipv6.label;
1518 ipv6_key->ipv6_proto = output->ip.proto;
1519 ipv6_key->ipv6_tclass = output->ip.tos;
1520 ipv6_key->ipv6_hlimit = output->ip.ttl;
1521 ipv6_key->ipv6_frag = output->ip.frag;
1522 } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1523 swkey->eth.type == htons(ETH_P_RARP)) {
1524 struct ovs_key_arp *arp_key;
1525
1526 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1527 if (!nla)
1528 goto nla_put_failure;
1529 arp_key = nla_data(nla);
1530 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1531 arp_key->arp_sip = output->ipv4.addr.src;
1532 arp_key->arp_tip = output->ipv4.addr.dst;
1533 arp_key->arp_op = htons(output->ip.proto);
8c63ff09
JP		 1534 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1535 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
25cd9ba0
SH		 1536 } else if (eth_p_mpls(swkey->eth.type)) {
1537 struct ovs_key_mpls *mpls_key;
1538
1539 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1540 if (!nla)
1541 goto nla_put_failure;
1542 mpls_key = nla_data(nla);
1543 mpls_key->mpls_lse = output->mpls.top_lse;
e6445719
PS		 1544 }
1545
1546 if ((swkey->eth.type == htons(ETH_P_IP) ||
1547 swkey->eth.type == htons(ETH_P_IPV6)) &&
1548 swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1549
1550 if (swkey->ip.proto == IPPROTO_TCP) {
1551 struct ovs_key_tcp *tcp_key;
1552
1553 nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1554 if (!nla)
1555 goto nla_put_failure;
1556 tcp_key = nla_data(nla);
1139e241
JR		 1557 tcp_key->tcp_src = output->tp.src;
1558 tcp_key->tcp_dst = output->tp.dst;
1559 if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1560 output->tp.flags))
1561 goto nla_put_failure;
e6445719
PS		 1562 } else if (swkey->ip.proto == IPPROTO_UDP) {
1563 struct ovs_key_udp *udp_key;
1564
1565 nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1566 if (!nla)
1567 goto nla_put_failure;
1568 udp_key = nla_data(nla);
1139e241
JR		 1569 udp_key->udp_src = output->tp.src;
1570 udp_key->udp_dst = output->tp.dst;
e6445719
PS		 1571 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1572 struct ovs_key_sctp *sctp_key;
1573
1574 nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1575 if (!nla)
1576 goto nla_put_failure;
1577 sctp_key = nla_data(nla);
1139e241
JR		 1578 sctp_key->sctp_src = output->tp.src;
1579 sctp_key->sctp_dst = output->tp.dst;
e6445719
PS		 1580 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1581 swkey->ip.proto == IPPROTO_ICMP) {
1582 struct ovs_key_icmp *icmp_key;
1583
1584 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1585 if (!nla)
1586 goto nla_put_failure;
1587 icmp_key = nla_data(nla);
1139e241
JR		 1588 icmp_key->icmp_type = ntohs(output->tp.src);
1589 icmp_key->icmp_code = ntohs(output->tp.dst);
e6445719
PS		 1590 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1591 swkey->ip.proto == IPPROTO_ICMPV6) {
1592 struct ovs_key_icmpv6 *icmpv6_key;
1593
1594 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1595 sizeof(*icmpv6_key));
1596 if (!nla)
1597 goto nla_put_failure;
1598 icmpv6_key = nla_data(nla);
1139e241
JR		 1599 icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1600 icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
e6445719
PS		 1601
1602 if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1603 icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1604 struct ovs_key_nd *nd_key;
1605
1606 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1607 if (!nla)
1608 goto nla_put_failure;
1609 nd_key = nla_data(nla);
1610 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1611 sizeof(nd_key->nd_target));
8c63ff09
JP		 1612 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1613 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
e6445719
PS		 1614 }
1615 }
1616 }
1617
1618unencap:
1619 if (encap)
1620 nla_nest_end(skb, encap);
1621
1622 return 0;
1623
1624nla_put_failure:
1625 return -EMSGSIZE;
1626}
1627
5b4237bb
JS		 1628int ovs_nla_put_key(const struct sw_flow_key *swkey,
1629 const struct sw_flow_key *output, int attr, bool is_mask,
1630 struct sk_buff *skb)
1631{
1632 int err;
1633 struct nlattr *nla;
1634
1635 nla = nla_nest_start(skb, attr);
1636 if (!nla)
1637 return -EMSGSIZE;
1638 err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1639 if (err)
1640 return err;
1641 nla_nest_end(skb, nla);
1642
1643 return 0;
1644}
1645
1646/* Called with ovs_mutex or RCU read lock. */
74ed7ab9
JS		 1647int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1648{
1649 if (ovs_identifier_is_ufid(&flow->id))
1650 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1651 flow->id.ufid);
1652
1653 return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1654 OVS_FLOW_ATTR_KEY, false, skb);
1655}
1656
1657/* Called with ovs_mutex or RCU read lock. */
1658int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
5b4237bb 1659{
26ad0b83 1660 return ovs_nla_put_key(&flow->key, &flow->key,
5b4237bb
JS		 1661 OVS_FLOW_ATTR_KEY, false, skb);
1662}
1663
1664/* Called with ovs_mutex or RCU read lock. */
1665int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1666{
1667 return ovs_nla_put_key(&flow->key, &flow->mask->key,
1668 OVS_FLOW_ATTR_MASK, true, skb);
1669}
1670
e6445719
PS		 1671#define MAX_ACTIONS_BUFSIZE (32 * 1024)
1672
05da5898 1673static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
e6445719
PS		 1674{
1675 struct sw_flow_actions *sfa;
1676
426cda5c 1677 if (size > MAX_ACTIONS_BUFSIZE) {
05da5898 1678 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
e6445719 1679 return ERR_PTR(-EINVAL);
426cda5c 1680 }
e6445719
PS		 1681
1682 sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1683 if (!sfa)
1684 return ERR_PTR(-ENOMEM);
1685
1686 sfa->actions_len = 0;
1687 return sfa;
1688}
1689
34ae932a
TG		 1690static void ovs_nla_free_set_action(const struct nlattr *a)
1691{
1692 const struct nlattr *ovs_key = nla_data(a);
1693 struct ovs_tunnel_info *ovs_tun;
1694
1695 switch (nla_type(ovs_key)) {
1696 case OVS_KEY_ATTR_TUNNEL_INFO:
1697 ovs_tun = nla_data(ovs_key);
1698 dst_release((struct dst_entry *)ovs_tun->tun_dst);
1699 break;
1700 }
1701}
1702
1703void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1704{
1705 const struct nlattr *a;
1706 int rem;
1707
1708 if (!sf_acts)
1709 return;
1710
1711 nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1712 switch (nla_type(a)) {
1713 case OVS_ACTION_ATTR_SET:
1714 ovs_nla_free_set_action(a);
1715 break;
7f8a436e
JS		 1716 case OVS_ACTION_ATTR_CT:
1717 ovs_ct_free_action(a);
1718 break;
34ae932a
TG		 1719 }
1720 }
1721
1722 kfree(sf_acts);
1723}
1724
1725static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1726{
1727 ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1728}
1729
e6445719
PS		 1730/* Schedules 'sf_acts' to be freed after the next RCU grace period.
1731 * The caller must hold rcu_read_lock for this to be sensible. */
34ae932a 1732void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
e6445719 1733{
34ae932a 1734 call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
e6445719
PS		 1735}
1736
1737static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
05da5898 1738 int attr_len, bool log)
e6445719
PS		 1739{
1740
1741 struct sw_flow_actions *acts;
1742 int new_acts_size;
1743 int req_size = NLA_ALIGN(attr_len);
1744 int next_offset = offsetof(struct sw_flow_actions, actions) +
1745 (*sfa)->actions_len;
1746
1747 if (req_size <= (ksize(*sfa) - next_offset))
1748 goto out;
1749
1750 new_acts_size = ksize(*sfa) * 2;
1751
1752 if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1753 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1754 return ERR_PTR(-EMSGSIZE);
1755 new_acts_size = MAX_ACTIONS_BUFSIZE;
1756 }
1757
05da5898 1758 acts = nla_alloc_flow_actions(new_acts_size, log);
e6445719
PS		 1759 if (IS_ERR(acts))
1760 return (void *)acts;
1761
1762 memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1763 acts->actions_len = (*sfa)->actions_len;
8e2fed1c 1764 acts->orig_len = (*sfa)->orig_len;
e6445719
PS		 1765 kfree(*sfa);
1766 *sfa = acts;
1767
1768out:
1769 (*sfa)->actions_len += req_size;
1770 return (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1771}
1772
f0b128c1 1773static struct nlattr *__add_action(struct sw_flow_actions **sfa,
05da5898 1774 int attrtype, void *data, int len, bool log)
e6445719
PS		 1775{
1776 struct nlattr *a;
1777
05da5898 1778 a = reserve_sfa_size(sfa, nla_attr_size(len), log);
e6445719 1779 if (IS_ERR(a))
f0b128c1 1780 return a;
e6445719
PS		 1781
1782 a->nla_type = attrtype;
1783 a->nla_len = nla_attr_size(len);
1784
1785 if (data)
1786 memcpy(nla_data(a), data, len);
1787 memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1788
f0b128c1
JG		 1789 return a;
1790}
1791
7f8a436e
JS		 1792int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
1793 int len, bool log)
f0b128c1
JG		 1794{
1795 struct nlattr *a;
1796
05da5898 1797 a = __add_action(sfa, attrtype, data, len, log);
f0b128c1 1798
f35423c1 1799 return PTR_ERR_OR_ZERO(a);
e6445719
PS		 1800}
1801
1802static inline int add_nested_action_start(struct sw_flow_actions **sfa,
05da5898 1803 int attrtype, bool log)
e6445719
PS		 1804{
1805 int used = (*sfa)->actions_len;
1806 int err;
1807
7f8a436e 1808 err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
e6445719
PS		 1809 if (err)
1810 return err;
1811
1812 return used;
1813}
1814
1815static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1816 int st_offset)
1817{
1818 struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1819 st_offset);
1820
1821 a->nla_len = sfa->actions_len - st_offset;
1822}
1823
7f8a436e 1824static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
25cd9ba0
SH		 1825 const struct sw_flow_key *key,
1826 int depth, struct sw_flow_actions **sfa,
05da5898 1827 __be16 eth_type, __be16 vlan_tci, bool log);
25cd9ba0 1828
7f8a436e 1829static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
e6445719 1830 const struct sw_flow_key *key, int depth,
25cd9ba0 1831 struct sw_flow_actions **sfa,
05da5898 1832 __be16 eth_type, __be16 vlan_tci, bool log)
e6445719
PS		 1833{
1834 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1835 const struct nlattr *probability, *actions;
1836 const struct nlattr *a;
1837 int rem, start, err, st_acts;
1838
1839 memset(attrs, 0, sizeof(attrs));
1840 nla_for_each_nested(a, attr, rem) {
1841 int type = nla_type(a);
1842 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1843 return -EINVAL;
1844 attrs[type] = a;
1845 }
1846 if (rem)
1847 return -EINVAL;
1848
1849 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1850 if (!probability || nla_len(probability) != sizeof(u32))
1851 return -EINVAL;
1852
1853 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1854 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1855 return -EINVAL;
1856
1857 /* validation done, copy sample action. */
05da5898 1858 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
e6445719
PS		 1859 if (start < 0)
1860 return start;
7f8a436e
JS		 1861 err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1862 nla_data(probability), sizeof(u32), log);
e6445719
PS		 1863 if (err)
1864 return err;
05da5898 1865 st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
e6445719
PS		 1866 if (st_acts < 0)
1867 return st_acts;
1868
7f8a436e 1869 err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
05da5898 1870 eth_type, vlan_tci, log);
e6445719
PS		 1871 if (err)
1872 return err;
1873
1874 add_nested_action_end(*sfa, st_acts);
1875 add_nested_action_end(*sfa, start);
1876
1877 return 0;
1878}
1879
e6445719
PS		 1880void ovs_match_init(struct sw_flow_match *match,
1881 struct sw_flow_key *key,
1882 struct sw_flow_mask *mask)
1883{
1884 memset(match, 0, sizeof(*match));
1885 match->key = key;
1886 match->mask = mask;
1887
1888 memset(key, 0, sizeof(*key));
1889
1890 if (mask) {
1891 memset(&mask->key, 0, sizeof(mask->key));
1892 mask->range.start = mask->range.end = 0;
1893 }
1894}
1895
d91641d9
TG		 1896static int validate_geneve_opts(struct sw_flow_key *key)
1897{
1898 struct geneve_opt *option;
1899 int opts_len = key->tun_opts_len;
1900 bool crit_opt = false;
1901
1902 option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1903 while (opts_len > 0) {
1904 int len;
1905
1906 if (opts_len < sizeof(*option))
1907 return -EINVAL;
1908
1909 len = sizeof(*option) + option->length * 4;
1910 if (len > opts_len)
1911 return -EINVAL;
1912
1913 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1914
1915 option = (struct geneve_opt *)((u8 *)option + len);
1916 opts_len -= len;
1917 };
1918
1919 key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1920
1921 return 0;
1922}
1923
e6445719 1924static int validate_and_copy_set_tun(const struct nlattr *attr,
05da5898 1925 struct sw_flow_actions **sfa, bool log)
e6445719
PS		 1926{
1927 struct sw_flow_match match;
1928 struct sw_flow_key key;
34ae932a 1929 struct metadata_dst *tun_dst;
1d8fff90 1930 struct ip_tunnel_info *tun_info;
34ae932a 1931 struct ovs_tunnel_info *ovs_tun;
f0b128c1 1932 struct nlattr *a;
13101602 1933 int err = 0, start, opts_type;
e6445719
PS		 1934
1935 ovs_match_init(&match, &key, NULL);
6b26ba3a 1936 opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
1dd144cf
TG		 1937 if (opts_type < 0)
1938 return opts_type;
e6445719 1939
f5796684 1940 if (key.tun_opts_len) {
1dd144cf
TG		 1941 switch (opts_type) {
1942 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1943 err = validate_geneve_opts(&key);
1944 if (err < 0)
1945 return err;
1946 break;
1947 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1948 break;
1949 }
f5796684
JG		 1950 };
1951
05da5898 1952 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
e6445719
PS		 1953 if (start < 0)
1954 return start;
1955
34ae932a
TG		 1956 tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
1957 if (!tun_dst)
1958 return -ENOMEM;
1959
f0b128c1 1960 a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
34ae932a
TG		 1961 sizeof(*ovs_tun), log);
1962 if (IS_ERR(a)) {
1963 dst_release((struct dst_entry *)tun_dst);
f0b128c1 1964 return PTR_ERR(a);
34ae932a
TG		 1965 }
1966
1967 ovs_tun = nla_data(a);
1968 ovs_tun->tun_dst = tun_dst;
f0b128c1 1969
34ae932a
TG		 1970 tun_info = &tun_dst->u.tun_info;
1971 tun_info->mode = IP_TUNNEL_INFO_TX;
00a93bab
JB		 1972 if (key.tun_proto == AF_INET6)
1973 tun_info->mode |= IP_TUNNEL_INFO_IPV6;
1d8fff90 1974 tun_info->key = key.tun_key;
f0b128c1 1975
4c222798
PS		 1976 /* We need to store the options in the action itself since
1977 * everything else will go away after flow setup. We can append
1978 * it to tun_info and then point there.
1979 */
1980 ip_tunnel_info_opts_set(tun_info,
1981 TUN_METADATA_OPTS(&key, key.tun_opts_len),
1982 key.tun_opts_len);
e6445719
PS		 1983 add_nested_action_end(*sfa, start);
1984
1985 return err;
1986}
1987
83d2b9ba
JR		 1988/* Return false if there are any non-masked bits set.
1989 * Mask follows data immediately, before any netlink padding.
1990 */
1991static bool validate_masked(u8 *data, int len)
1992{
1993 u8 *mask = data + len;
1994
1995 while (len--)
1996 if (*data++ & ~*mask++)
1997 return false;
1998
1999 return true;
2000}
2001
e6445719
PS		 2002static int validate_set(const struct nlattr *a,
2003 const struct sw_flow_key *flow_key,
2004 struct sw_flow_actions **sfa,
83d2b9ba 2005 bool *skip_copy, __be16 eth_type, bool masked, bool log)
e6445719
PS		 2006{
2007 const struct nlattr *ovs_key = nla_data(a);
2008 int key_type = nla_type(ovs_key);
83d2b9ba 2009 size_t key_len;
e6445719
PS		 2010
2011 /* There can be only one key in a action */
2012 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2013 return -EINVAL;
2014
83d2b9ba
JR		 2015 key_len = nla_len(ovs_key);
2016 if (masked)
2017 key_len /= 2;
2018
e6445719 2019 if (key_type > OVS_KEY_ATTR_MAX ||
982b5270 2020 !check_attr_len(key_len, ovs_key_lens[key_type].len))
e6445719
PS		 2021 return -EINVAL;
2022
83d2b9ba
JR		 2023 if (masked && !validate_masked(nla_data(ovs_key), key_len))
2024 return -EINVAL;
2025
e6445719
PS		 2026 switch (key_type) {
2027 const struct ovs_key_ipv4 *ipv4_key;
2028 const struct ovs_key_ipv6 *ipv6_key;
2029 int err;
2030
2031 case OVS_KEY_ATTR_PRIORITY:
2032 case OVS_KEY_ATTR_SKB_MARK:
182e3042 2033 case OVS_KEY_ATTR_CT_MARK:
33db4125 2034 case OVS_KEY_ATTR_CT_LABELS:
e6445719
PS		 2035 case OVS_KEY_ATTR_ETHERNET:
2036 break;
2037
2038 case OVS_KEY_ATTR_TUNNEL:
25cd9ba0
SH		 2039 if (eth_p_mpls(eth_type))
2040 return -EINVAL;
2041
83d2b9ba
JR		 2042 if (masked)
2043 return -EINVAL; /* Masked tunnel set not supported. */
2044
2045 *skip_copy = true;
05da5898 2046 err = validate_and_copy_set_tun(a, sfa, log);
e6445719
PS		 2047 if (err)
2048 return err;
2049 break;
2050
2051 case OVS_KEY_ATTR_IPV4:
25cd9ba0 2052 if (eth_type != htons(ETH_P_IP))
e6445719
PS		 2053 return -EINVAL;
2054
e6445719 2055 ipv4_key = nla_data(ovs_key);
e6445719 2056
83d2b9ba
JR		 2057 if (masked) {
2058 const struct ovs_key_ipv4 *mask = ipv4_key + 1;
e6445719 2059
83d2b9ba
JR		 2060 /* Non-writeable fields. */
2061 if (mask->ipv4_proto || mask->ipv4_frag)
2062 return -EINVAL;
2063 } else {
2064 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2065 return -EINVAL;
2066
2067 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2068 return -EINVAL;
2069 }
e6445719
PS		 2070 break;
2071
2072 case OVS_KEY_ATTR_IPV6:
25cd9ba0 2073 if (eth_type != htons(ETH_P_IPV6))
e6445719
PS		 2074 return -EINVAL;
2075
e6445719 2076 ipv6_key = nla_data(ovs_key);
e6445719 2077
83d2b9ba
JR		 2078 if (masked) {
2079 const struct ovs_key_ipv6 *mask = ipv6_key + 1;
2080
2081 /* Non-writeable fields. */
2082 if (mask->ipv6_proto || mask->ipv6_frag)
2083 return -EINVAL;
2084
2085 /* Invalid bits in the flow label mask? */
2086 if (ntohl(mask->ipv6_label) & 0xFFF00000)
2087 return -EINVAL;
2088 } else {
2089 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2090 return -EINVAL;
e6445719 2091
83d2b9ba
JR		 2092 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2093 return -EINVAL;
2094 }
e6445719
PS		 2095 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2096 return -EINVAL;
2097
2098 break;
2099
2100 case OVS_KEY_ATTR_TCP:
83d2b9ba
JR		 2101 if ((eth_type != htons(ETH_P_IP) &&
2102 eth_type != htons(ETH_P_IPV6)) ||
2103 flow_key->ip.proto != IPPROTO_TCP)
e6445719
PS		 2104 return -EINVAL;
2105
83d2b9ba 2106 break;
e6445719
PS		 2107
2108 case OVS_KEY_ATTR_UDP:
83d2b9ba
JR		 2109 if ((eth_type != htons(ETH_P_IP) &&
2110 eth_type != htons(ETH_P_IPV6)) ||
2111 flow_key->ip.proto != IPPROTO_UDP)
e6445719
PS		 2112 return -EINVAL;
2113
83d2b9ba 2114 break;
25cd9ba0
SH		 2115
2116 case OVS_KEY_ATTR_MPLS:
2117 if (!eth_p_mpls(eth_type))
2118 return -EINVAL;
2119 break;
e6445719
PS		 2120
2121 case OVS_KEY_ATTR_SCTP:
83d2b9ba
JR		 2122 if ((eth_type != htons(ETH_P_IP) &&
2123 eth_type != htons(ETH_P_IPV6)) ||
2124 flow_key->ip.proto != IPPROTO_SCTP)
e6445719
PS		 2125 return -EINVAL;
2126
83d2b9ba 2127 break;
e6445719
PS		 2128
2129 default:
2130 return -EINVAL;
2131 }
2132
83d2b9ba
JR		 2133 /* Convert non-masked non-tunnel set actions to masked set actions. */
2134 if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2135 int start, len = key_len * 2;
2136 struct nlattr *at;
2137
2138 *skip_copy = true;
2139
2140 start = add_nested_action_start(sfa,
2141 OVS_ACTION_ATTR_SET_TO_MASKED,
2142 log);
2143 if (start < 0)
2144 return start;
2145
2146 at = __add_action(sfa, key_type, NULL, len, log);
2147 if (IS_ERR(at))
2148 return PTR_ERR(at);
2149
2150 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2151 memset(nla_data(at) + key_len, 0xff, key_len); /* Mask. */
2152 /* Clear non-writeable bits from otherwise writeable fields. */
2153 if (key_type == OVS_KEY_ATTR_IPV6) {
2154 struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2155
2156 mask->ipv6_label &= htonl(0x000FFFFF);
2157 }
2158 add_nested_action_end(*sfa, start);
2159 }
2160
e6445719
PS		 2161 return 0;
2162}
2163
2164static int validate_userspace(const struct nlattr *attr)
2165{
2166 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2167 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2168 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
8f0aad6f 2169 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
e6445719
PS		 2170 };
2171 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2172 int error;
2173
2174 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2175 attr, userspace_policy);
2176 if (error)
2177 return error;
2178
2179 if (!a[OVS_USERSPACE_ATTR_PID] ||
2180 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2181 return -EINVAL;
2182
2183 return 0;
2184}
2185
2186static int copy_action(const struct nlattr *from,
05da5898 2187 struct sw_flow_actions **sfa, bool log)
e6445719
PS		 2188{
2189 int totlen = NLA_ALIGN(from->nla_len);
2190 struct nlattr *to;
2191
05da5898 2192 to = reserve_sfa_size(sfa, from->nla_len, log);
e6445719
PS		 2193 if (IS_ERR(to))
2194 return PTR_ERR(to);
2195
2196 memcpy(to, from, totlen);
2197 return 0;
2198}
2199
7f8a436e 2200static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
25cd9ba0
SH		 2201 const struct sw_flow_key *key,
2202 int depth, struct sw_flow_actions **sfa,
05da5898 2203 __be16 eth_type, __be16 vlan_tci, bool log)
e6445719
PS		 2204{
2205 const struct nlattr *a;
2206 int rem, err;
2207
2208 if (depth >= SAMPLE_ACTION_DEPTH)
2209 return -EOVERFLOW;
2210
2211 nla_for_each_nested(a, attr, rem) {
2212 /* Expected argument lengths, (u32)-1 for variable length. */
2213 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2214 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
971427f3 2215 [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
e6445719 2216 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
25cd9ba0
SH		 2217 [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2218 [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
e6445719
PS		 2219 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2220 [OVS_ACTION_ATTR_POP_VLAN] = 0,
2221 [OVS_ACTION_ATTR_SET] = (u32)-1,
83d2b9ba 2222 [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
971427f3 2223 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
7f8a436e
JS		 2224 [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2225 [OVS_ACTION_ATTR_CT] = (u32)-1,
e6445719
PS		 2226 };
2227 const struct ovs_action_push_vlan *vlan;
2228 int type = nla_type(a);
2229 bool skip_copy;
2230
2231 if (type > OVS_ACTION_ATTR_MAX ||
2232 (action_lens[type] != nla_len(a) &&
2233 action_lens[type] != (u32)-1))
2234 return -EINVAL;
2235
2236 skip_copy = false;
2237 switch (type) {
2238 case OVS_ACTION_ATTR_UNSPEC:
2239 return -EINVAL;
2240
2241 case OVS_ACTION_ATTR_USERSPACE:
2242 err = validate_userspace(a);
2243 if (err)
2244 return err;
2245 break;
2246
2247 case OVS_ACTION_ATTR_OUTPUT:
2248 if (nla_get_u32(a) >= DP_MAX_PORTS)
2249 return -EINVAL;
2250 break;
2251
971427f3
AZ		 2252 case OVS_ACTION_ATTR_HASH: {
2253 const struct ovs_action_hash *act_hash = nla_data(a);
2254
2255 switch (act_hash->hash_alg) {
2256 case OVS_HASH_ALG_L4:
2257 break;
2258 default:
2259 return -EINVAL;
2260 }
2261
2262 break;
2263 }
e6445719
PS		 2264
2265 case OVS_ACTION_ATTR_POP_VLAN:
25cd9ba0 2266 vlan_tci = htons(0);
e6445719
PS		 2267 break;
2268
2269 case OVS_ACTION_ATTR_PUSH_VLAN:
2270 vlan = nla_data(a);
2271 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2272 return -EINVAL;
2273 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2274 return -EINVAL;
25cd9ba0 2275 vlan_tci = vlan->vlan_tci;
e6445719
PS		 2276 break;
2277
971427f3
AZ		 2278 case OVS_ACTION_ATTR_RECIRC:
2279 break;
2280
25cd9ba0
SH		 2281 case OVS_ACTION_ATTR_PUSH_MPLS: {
2282 const struct ovs_action_push_mpls *mpls = nla_data(a);
2283
25cd9ba0
SH		 2284 if (!eth_p_mpls(mpls->mpls_ethertype))
2285 return -EINVAL;
2286 /* Prohibit push MPLS other than to a white list
2287 * for packets that have a known tag order.
2288 */
2289 if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2290 (eth_type != htons(ETH_P_IP) &&
2291 eth_type != htons(ETH_P_IPV6) &&
2292 eth_type != htons(ETH_P_ARP) &&
2293 eth_type != htons(ETH_P_RARP) &&
2294 !eth_p_mpls(eth_type)))
2295 return -EINVAL;
2296 eth_type = mpls->mpls_ethertype;
2297 break;
2298 }
2299
2300 case OVS_ACTION_ATTR_POP_MPLS:
2301 if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2302 !eth_p_mpls(eth_type))
2303 return -EINVAL;
2304
2305 /* Disallow subsequent L2.5+ set and mpls_pop actions
2306 * as there is no check here to ensure that the new
2307 * eth_type is valid and thus set actions could
2308 * write off the end of the packet or otherwise
2309 * corrupt it.
2310 *
2311 * Support for these actions is planned using packet
2312 * recirculation.
2313 */
2314 eth_type = htons(0);
2315 break;
2316
e6445719 2317 case OVS_ACTION_ATTR_SET:
25cd9ba0 2318 err = validate_set(a, key, sfa,
83d2b9ba
JR		 2319 &skip_copy, eth_type, false, log);
2320 if (err)
2321 return err;
2322 break;
2323
2324 case OVS_ACTION_ATTR_SET_MASKED:
2325 err = validate_set(a, key, sfa,
2326 &skip_copy, eth_type, true, log);
e6445719
PS		 2327 if (err)
2328 return err;
2329 break;
2330
2331 case OVS_ACTION_ATTR_SAMPLE:
7f8a436e 2332 err = validate_and_copy_sample(net, a, key, depth, sfa,
05da5898 2333 eth_type, vlan_tci, log);
e6445719
PS		 2334 if (err)
2335 return err;
2336 skip_copy = true;
2337 break;
2338
7f8a436e
JS		 2339 case OVS_ACTION_ATTR_CT:
2340 err = ovs_ct_copy_action(net, a, key, sfa, log);
2341 if (err)
2342 return err;
2343 skip_copy = true;
2344 break;
2345
e6445719 2346 default:
05da5898 2347 OVS_NLERR(log, "Unknown Action type %d", type);
e6445719
PS		 2348 return -EINVAL;
2349 }
2350 if (!skip_copy) {
05da5898 2351 err = copy_action(a, sfa, log);
e6445719
PS		 2352 if (err)
2353 return err;
2354 }
2355 }
2356
2357 if (rem > 0)
2358 return -EINVAL;
2359
2360 return 0;
2361}
2362
83d2b9ba 2363/* 'key' must be the masked key. */
7f8a436e 2364int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
25cd9ba0 2365 const struct sw_flow_key *key,
05da5898 2366 struct sw_flow_actions **sfa, bool log)
25cd9ba0 2367{
2fdb957d
PS		 2368 int err;
2369
05da5898 2370 *sfa = nla_alloc_flow_actions(nla_len(attr), log);
2fdb957d
PS		 2371 if (IS_ERR(*sfa))
2372 return PTR_ERR(*sfa);
2373
8e2fed1c 2374 (*sfa)->orig_len = nla_len(attr);
7f8a436e 2375 err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
05da5898 2376 key->eth.tci, log);
2fdb957d 2377 if (err)
34ae932a 2378 ovs_nla_free_flow_actions(*sfa);
2fdb957d
PS		 2379
2380 return err;
25cd9ba0
SH		 2381}
2382
e6445719
PS		 2383static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2384{
2385 const struct nlattr *a;
2386 struct nlattr *start;
2387 int err = 0, rem;
2388
2389 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2390 if (!start)
2391 return -EMSGSIZE;
2392
2393 nla_for_each_nested(a, attr, rem) {
2394 int type = nla_type(a);
2395 struct nlattr *st_sample;
2396
2397 switch (type) {
2398 case OVS_SAMPLE_ATTR_PROBABILITY:
2399 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2400 sizeof(u32), nla_data(a)))
2401 return -EMSGSIZE;
2402 break;
2403 case OVS_SAMPLE_ATTR_ACTIONS:
2404 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2405 if (!st_sample)
2406 return -EMSGSIZE;
2407 err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2408 if (err)
2409 return err;
2410 nla_nest_end(skb, st_sample);
2411 break;
2412 }
2413 }
2414
2415 nla_nest_end(skb, start);
2416 return err;
2417}
2418
2419static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2420{
2421 const struct nlattr *ovs_key = nla_data(a);
2422 int key_type = nla_type(ovs_key);
2423 struct nlattr *start;
2424 int err;
2425
2426 switch (key_type) {
f0b128c1 2427 case OVS_KEY_ATTR_TUNNEL_INFO: {
34ae932a
TG		 2428 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2429 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
f0b128c1 2430
e6445719
PS		 2431 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2432 if (!start)
2433 return -EMSGSIZE;
2434
6b26ba3a
JB		 2435 err = ip_tun_to_nlattr(skb, &tun_info->key,
2436 tun_info->options_len ?
4c222798 2437 ip_tunnel_info_opts(tun_info) : NULL,
6b26ba3a
JB		 2438 tun_info->options_len,
2439 ip_tunnel_info_af(tun_info));
e6445719
PS		 2440 if (err)
2441 return err;
2442 nla_nest_end(skb, start);
2443 break;
f0b128c1 2444 }
e6445719
PS		 2445 default:
2446 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2447 return -EMSGSIZE;
2448 break;
2449 }
2450
2451 return 0;
2452}
2453
83d2b9ba
JR		 2454static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2455 struct sk_buff *skb)
2456{
2457 const struct nlattr *ovs_key = nla_data(a);
f4f8e738 2458 struct nlattr *nla;
83d2b9ba
JR		 2459 size_t key_len = nla_len(ovs_key) / 2;
2460
2461 /* Revert the conversion we did from a non-masked set action to
2462 * masked set action.
2463 */
f4f8e738
JS		 2464 nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2465 if (!nla)
83d2b9ba
JR		 2466 return -EMSGSIZE;
2467
f4f8e738
JS		 2468 if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2469 return -EMSGSIZE;
2470
2471 nla_nest_end(skb, nla);
83d2b9ba
JR		 2472 return 0;
2473}
2474
e6445719
PS		 2475int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2476{
2477 const struct nlattr *a;
2478 int rem, err;
2479
2480 nla_for_each_attr(a, attr, len, rem) {
2481 int type = nla_type(a);
2482
2483 switch (type) {
2484 case OVS_ACTION_ATTR_SET:
2485 err = set_action_to_attr(a, skb);
2486 if (err)
2487 return err;
2488 break;
2489
83d2b9ba
JR		 2490 case OVS_ACTION_ATTR_SET_TO_MASKED:
2491 err = masked_set_action_to_set_action_attr(a, skb);
2492 if (err)
2493 return err;
2494 break;
2495
e6445719
PS		 2496 case OVS_ACTION_ATTR_SAMPLE:
2497 err = sample_action_to_attr(a, skb);
2498 if (err)
2499 return err;
2500 break;
7f8a436e
JS		 2501
2502 case OVS_ACTION_ATTR_CT:
2503 err = ovs_ct_action_to_attr(nla_data(a), skb);
2504 if (err)
2505 return err;
2506 break;
2507
e6445719
PS		 2508 default:
2509 if (nla_put(skb, type, nla_len(a), nla_data(a)))
2510 return -EMSGSIZE;
2511 break;
2512 }
2513 }
2514
2515 return 0;
2516}
Linux kernel - Deliverables
This page took 0.775845 seconds and 5 git commands to generate.