]> git.proxmox.com Git - ovs.git/blame - datapath/flow_netlink.c
netdev-dpdk: add DPDK pdump capability
[ovs.git] / datapath / flow_netlink.c
CommitLineData
a097c0b2 1/*
e23775f2 2 * Copyright (c) 2007-2015 Nicira, Inc.
a097c0b2
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
0a0857df
JP
19#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
a097c0b2
PS
21#include <linux/uaccess.h>
22#include <linux/netdevice.h>
23#include <linux/etherdevice.h>
24#include <linux/if_ether.h>
25#include <linux/if_vlan.h>
26#include <net/llc_pdu.h>
27#include <linux/kernel.h>
28#include <linux/jhash.h>
29#include <linux/jiffies.h>
30#include <linux/llc.h>
31#include <linux/module.h>
32#include <linux/in.h>
33#include <linux/rcupdate.h>
34#include <linux/if_arp.h>
35#include <linux/ip.h>
36#include <linux/ipv6.h>
37#include <linux/sctp.h>
38#include <linux/tcp.h>
39#include <linux/udp.h>
40#include <linux/icmp.h>
41#include <linux/icmpv6.h>
42#include <linux/rculist.h>
c1fc1411 43#include <net/geneve.h>
a097c0b2
PS
44#include <net/ip.h>
45#include <net/ipv6.h>
46#include <net/ndisc.h>
2baf0e0c 47#include <net/mpls.h>
e23775f2 48#include <net/vxlan.h>
a097c0b2 49
2baf0e0c 50#include "datapath.h"
a94ebc39 51#include "conntrack.h"
2baf0e0c 52#include "flow.h"
a097c0b2 53#include "flow_netlink.h"
e23775f2 54#include "gso.h"
a097c0b2 55
ec959cdc
TG
56struct ovs_len_tbl {
57 int len;
58 const struct ovs_len_tbl *next;
59};
60
61#define OVS_ATTR_NESTED -1
6da061f4 62#define OVS_ATTR_VARIABLE -2
ec959cdc 63
f3ccd17d
PS
64static void update_range(struct sw_flow_match *match,
65 size_t offset, size_t size, bool is_mask)
a097c0b2 66{
f3ccd17d 67 struct sw_flow_key_range *range;
a097c0b2
PS
68 size_t start = rounddown(offset, sizeof(long));
69 size_t end = roundup(offset + size, sizeof(long));
70
71 if (!is_mask)
72 range = &match->range;
f3ccd17d 73 else
a097c0b2
PS
74 range = &match->mask->range;
75
a097c0b2
PS
76 if (range->start == range->end) {
77 range->start = start;
78 range->end = end;
79 return;
80 }
81
82 if (range->start > start)
83 range->start = start;
84
85 if (range->end < end)
86 range->end = end;
87}
88
89#define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
90 do { \
f3ccd17d
PS
91 update_range(match, offsetof(struct sw_flow_key, field), \
92 sizeof((match)->key->field), is_mask); \
93 if (is_mask) \
94 (match)->mask->key.field = value; \
95 else \
a097c0b2 96 (match)->key->field = value; \
a097c0b2
PS
97 } while (0)
98
f3ccd17d
PS
99#define SW_FLOW_KEY_MEMCPY_OFFSET(match, offset, value_p, len, is_mask) \
100 do { \
101 update_range(match, offset, len, is_mask); \
102 if (is_mask) \
103 memcpy((u8 *)&(match)->mask->key + offset, value_p, len);\
104 else \
105 memcpy((u8 *)(match)->key + offset, value_p, len); \
a097c0b2
PS
106 } while (0)
107
f3ccd17d 108#define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
c1fc1411
JG
109 SW_FLOW_KEY_MEMCPY_OFFSET(match, offsetof(struct sw_flow_key, field), \
110 value_p, len, is_mask)
111
f3ccd17d
PS
112#define SW_FLOW_KEY_MEMSET_FIELD(match, field, value, is_mask) \
113 do { \
114 update_range(match, offsetof(struct sw_flow_key, field), \
115 sizeof((match)->key->field), is_mask); \
116 if (is_mask) \
117 memset((u8 *)&(match)->mask->key.field, value, \
118 sizeof((match)->mask->key.field)); \
119 else \
0b496cda
DDP
120 memset((u8 *)&(match)->key->field, value, \
121 sizeof((match)->key->field)); \
0b496cda
DDP
122 } while (0)
123
a097c0b2 124static bool match_validate(const struct sw_flow_match *match,
9233cef7 125 u64 key_attrs, u64 mask_attrs, bool log)
a097c0b2
PS
126{
127 u64 key_expected = 1ULL << OVS_KEY_ATTR_ETHERNET;
128 u64 mask_allowed = key_attrs; /* At most allow all key attributes */
129
130 /* The following mask attributes allowed only if they
af465b67
PS
131 * pass the validation tests.
132 */
a097c0b2
PS
133 mask_allowed &= ~((1ULL << OVS_KEY_ATTR_IPV4)
134 | (1ULL << OVS_KEY_ATTR_IPV6)
135 | (1ULL << OVS_KEY_ATTR_TCP)
dc235f7f 136 | (1ULL << OVS_KEY_ATTR_TCP_FLAGS)
a097c0b2
PS
137 | (1ULL << OVS_KEY_ATTR_UDP)
138 | (1ULL << OVS_KEY_ATTR_SCTP)
139 | (1ULL << OVS_KEY_ATTR_ICMP)
140 | (1ULL << OVS_KEY_ATTR_ICMPV6)
141 | (1ULL << OVS_KEY_ATTR_ARP)
ccf43786
SH
142 | (1ULL << OVS_KEY_ATTR_ND)
143 | (1ULL << OVS_KEY_ATTR_MPLS));
a097c0b2
PS
144
145 /* Always allowed mask fields. */
146 mask_allowed |= ((1ULL << OVS_KEY_ATTR_TUNNEL)
147 | (1ULL << OVS_KEY_ATTR_IN_PORT)
148 | (1ULL << OVS_KEY_ATTR_ETHERTYPE));
149
150 /* Check key attributes. */
151 if (match->key->eth.type == htons(ETH_P_ARP)
152 || match->key->eth.type == htons(ETH_P_RARP)) {
153 key_expected |= 1ULL << OVS_KEY_ATTR_ARP;
154 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
155 mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
156 }
157
ccf43786
SH
158 if (eth_p_mpls(match->key->eth.type)) {
159 key_expected |= 1ULL << OVS_KEY_ATTR_MPLS;
160 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
161 mask_allowed |= 1ULL << OVS_KEY_ATTR_MPLS;
162 }
163
a097c0b2
PS
164 if (match->key->eth.type == htons(ETH_P_IP)) {
165 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
166 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
167 mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV4;
168
169 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
170 if (match->key->ip.proto == IPPROTO_UDP) {
171 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
172 if (match->mask && (match->mask->key.ip.proto == 0xff))
173 mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
174 }
175
176 if (match->key->ip.proto == IPPROTO_SCTP) {
177 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
178 if (match->mask && (match->mask->key.ip.proto == 0xff))
179 mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
180 }
181
182 if (match->key->ip.proto == IPPROTO_TCP) {
183 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
dc235f7f
JR
184 key_expected |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
185 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
a097c0b2 186 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
dc235f7f
JR
187 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
188 }
a097c0b2
PS
189 }
190
191 if (match->key->ip.proto == IPPROTO_ICMP) {
192 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
193 if (match->mask && (match->mask->key.ip.proto == 0xff))
194 mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
195 }
196 }
197 }
198
199 if (match->key->eth.type == htons(ETH_P_IPV6)) {
200 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
201 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
202 mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV6;
203
204 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
205 if (match->key->ip.proto == IPPROTO_UDP) {
206 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
207 if (match->mask && (match->mask->key.ip.proto == 0xff))
208 mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
209 }
210
211 if (match->key->ip.proto == IPPROTO_SCTP) {
212 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
213 if (match->mask && (match->mask->key.ip.proto == 0xff))
214 mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
215 }
216
217 if (match->key->ip.proto == IPPROTO_TCP) {
218 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
dc235f7f
JR
219 key_expected |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
220 if (match->mask && (match->mask->key.ip.proto == 0xff)) {
a097c0b2 221 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
dc235f7f
JR
222 mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP_FLAGS;
223 }
a097c0b2
PS
224 }
225
226 if (match->key->ip.proto == IPPROTO_ICMPV6) {
227 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
228 if (match->mask && (match->mask->key.ip.proto == 0xff))
229 mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMPV6;
230
708fb4c5 231 if (match->key->tp.src ==
a097c0b2 232 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
708fb4c5 233 match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
a097c0b2 234 key_expected |= 1ULL << OVS_KEY_ATTR_ND;
560f3099 235 if (match->mask && (match->mask->key.tp.src == htons(0xff)))
a097c0b2
PS
236 mask_allowed |= 1ULL << OVS_KEY_ATTR_ND;
237 }
238 }
239 }
240 }
241
242 if ((key_attrs & key_expected) != key_expected) {
243 /* Key attributes check failed. */
7d16c847 244 OVS_NLERR(log, "Missing key (keys=%llx, expected=%llx)",
9233cef7
JR
245 (unsigned long long)key_attrs,
246 (unsigned long long)key_expected);
a097c0b2
PS
247 return false;
248 }
249
250 if ((mask_attrs & mask_allowed) != mask_attrs) {
251 /* Mask attributes check failed. */
7d16c847 252 OVS_NLERR(log, "Unexpected mask (mask=%llx, allowed=%llx)",
9233cef7
JR
253 (unsigned long long)mask_attrs,
254 (unsigned long long)mask_allowed);
a097c0b2
PS
255 return false;
256 }
257
258 return true;
259}
260
8b7ea2d4
WZ
261size_t ovs_tun_key_attr_size(void)
262{
263 /* Whenever adding new OVS_TUNNEL_KEY_ FIELDS, we should consider
af465b67
PS
264 * updating this function.
265 */
f3464818 266 return nla_total_size_64bit(8) /* OVS_TUNNEL_KEY_ATTR_ID */
8a2d4905
PS
267 + nla_total_size(16) /* OVS_TUNNEL_KEY_ATTR_IPV[46]_SRC */
268 + nla_total_size(16) /* OVS_TUNNEL_KEY_ATTR_IPV[46]_DST */
8b7ea2d4
WZ
269 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
270 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
271 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
272 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
273 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_OAM */
274 + nla_total_size(256) /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
0c7930a3
TG
275 /* OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS is mutually exclusive with
276 * OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS and covered by it.
277 */
8b7ea2d4
WZ
278 + nla_total_size(2) /* OVS_TUNNEL_KEY_ATTR_TP_SRC */
279 + nla_total_size(2); /* OVS_TUNNEL_KEY_ATTR_TP_DST */
280}
281
4e25b8c1
JS
282size_t ovs_key_attr_size(void)
283{
284 /* Whenever adding new OVS_KEY_ FIELDS, we should consider
af465b67
PS
285 * updating this function.
286 */
9daf2348 287 BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 26);
4e25b8c1
JS
288
289 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
290 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
8b7ea2d4 291 + ovs_tun_key_attr_size()
4e25b8c1
JS
292 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
293 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
294 + nla_total_size(4) /* OVS_KEY_ATTR_DP_HASH */
295 + nla_total_size(4) /* OVS_KEY_ATTR_RECIRC_ID */
c05e2094 296 + nla_total_size(4) /* OVS_KEY_ATTR_CT_STATE */
a94ebc39 297 + nla_total_size(2) /* OVS_KEY_ATTR_CT_ZONE */
372ce973 298 + nla_total_size(4) /* OVS_KEY_ATTR_CT_MARK */
c05e2094 299 + nla_total_size(16) /* OVS_KEY_ATTR_CT_LABELS */
4e25b8c1
JS
300 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
301 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
302 + nla_total_size(4) /* OVS_KEY_ATTR_VLAN */
303 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
304 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
305 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
306 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
307 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
308}
309
6da061f4
JG
310static const struct ovs_len_tbl ovs_vxlan_ext_key_lens[OVS_VXLAN_EXT_MAX + 1] = {
311 [OVS_VXLAN_EXT_GBP] = { .len = sizeof(u32) },
312};
313
ec959cdc
TG
314static const struct ovs_len_tbl ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
315 [OVS_TUNNEL_KEY_ATTR_ID] = { .len = sizeof(u64) },
316 [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = { .len = sizeof(u32) },
317 [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = { .len = sizeof(u32) },
318 [OVS_TUNNEL_KEY_ATTR_TOS] = { .len = 1 },
319 [OVS_TUNNEL_KEY_ATTR_TTL] = { .len = 1 },
320 [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = { .len = 0 },
321 [OVS_TUNNEL_KEY_ATTR_CSUM] = { .len = 0 },
322 [OVS_TUNNEL_KEY_ATTR_TP_SRC] = { .len = sizeof(u16) },
323 [OVS_TUNNEL_KEY_ATTR_TP_DST] = { .len = sizeof(u16) },
324 [OVS_TUNNEL_KEY_ATTR_OAM] = { .len = 0 },
6da061f4
JG
325 [OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS] = { .len = OVS_ATTR_VARIABLE },
326 [OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS] = { .len = OVS_ATTR_NESTED,
327 .next = ovs_vxlan_ext_key_lens },
8a2d4905
PS
328 [OVS_TUNNEL_KEY_ATTR_IPV6_SRC] = { .len = sizeof(struct in6_addr) },
329 [OVS_TUNNEL_KEY_ATTR_IPV6_DST] = { .len = sizeof(struct in6_addr) },
a097c0b2
PS
330};
331
ec959cdc
TG
332/* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute. */
333static const struct ovs_len_tbl ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
334 [OVS_KEY_ATTR_ENCAP] = { .len = OVS_ATTR_NESTED },
335 [OVS_KEY_ATTR_PRIORITY] = { .len = sizeof(u32) },
336 [OVS_KEY_ATTR_IN_PORT] = { .len = sizeof(u32) },
337 [OVS_KEY_ATTR_SKB_MARK] = { .len = sizeof(u32) },
338 [OVS_KEY_ATTR_ETHERNET] = { .len = sizeof(struct ovs_key_ethernet) },
339 [OVS_KEY_ATTR_VLAN] = { .len = sizeof(__be16) },
340 [OVS_KEY_ATTR_ETHERTYPE] = { .len = sizeof(__be16) },
341 [OVS_KEY_ATTR_IPV4] = { .len = sizeof(struct ovs_key_ipv4) },
342 [OVS_KEY_ATTR_IPV6] = { .len = sizeof(struct ovs_key_ipv6) },
343 [OVS_KEY_ATTR_TCP] = { .len = sizeof(struct ovs_key_tcp) },
344 [OVS_KEY_ATTR_TCP_FLAGS] = { .len = sizeof(__be16) },
345 [OVS_KEY_ATTR_UDP] = { .len = sizeof(struct ovs_key_udp) },
346 [OVS_KEY_ATTR_SCTP] = { .len = sizeof(struct ovs_key_sctp) },
347 [OVS_KEY_ATTR_ICMP] = { .len = sizeof(struct ovs_key_icmp) },
348 [OVS_KEY_ATTR_ICMPV6] = { .len = sizeof(struct ovs_key_icmpv6) },
349 [OVS_KEY_ATTR_ARP] = { .len = sizeof(struct ovs_key_arp) },
350 [OVS_KEY_ATTR_ND] = { .len = sizeof(struct ovs_key_nd) },
351 [OVS_KEY_ATTR_RECIRC_ID] = { .len = sizeof(u32) },
352 [OVS_KEY_ATTR_DP_HASH] = { .len = sizeof(u32) },
353 [OVS_KEY_ATTR_TUNNEL] = { .len = OVS_ATTR_NESTED,
354 .next = ovs_tunnel_key_lens, },
355 [OVS_KEY_ATTR_MPLS] = { .len = sizeof(struct ovs_key_mpls) },
c05e2094 356 [OVS_KEY_ATTR_CT_STATE] = { .len = sizeof(u32) },
a94ebc39 357 [OVS_KEY_ATTR_CT_ZONE] = { .len = sizeof(u16) },
372ce973 358 [OVS_KEY_ATTR_CT_MARK] = { .len = sizeof(u32) },
c05e2094 359 [OVS_KEY_ATTR_CT_LABELS] = { .len = sizeof(struct ovs_key_ct_labels) },
ec959cdc 360};
a0fb56c1 361
6da061f4
JG
362static bool check_attr_len(unsigned int attr_len, unsigned int expected_len)
363{
364 return expected_len == attr_len ||
365 expected_len == OVS_ATTR_NESTED ||
366 expected_len == OVS_ATTR_VARIABLE;
367}
368
a097c0b2
PS
369static bool is_all_zero(const u8 *fp, size_t size)
370{
371 int i;
372
373 if (!fp)
374 return false;
375
376 for (i = 0; i < size; i++)
377 if (fp[i])
378 return false;
379
380 return true;
381}
382
383static int __parse_flow_nlattrs(const struct nlattr *attr,
384 const struct nlattr *a[],
9233cef7 385 u64 *attrsp, bool log, bool nz)
a097c0b2
PS
386{
387 const struct nlattr *nla;
388 u64 attrs;
389 int rem;
390
391 attrs = *attrsp;
392 nla_for_each_nested(nla, attr, rem) {
393 u16 type = nla_type(nla);
394 int expected_len;
395
396 if (type > OVS_KEY_ATTR_MAX) {
7d16c847 397 OVS_NLERR(log, "Key type %d is out of range max %d",
a097c0b2
PS
398 type, OVS_KEY_ATTR_MAX);
399 return -EINVAL;
400 }
401
402 if (attrs & (1ULL << type)) {
7d16c847 403 OVS_NLERR(log, "Duplicate key (type %d).", type);
a097c0b2
PS
404 return -EINVAL;
405 }
406
ec959cdc 407 expected_len = ovs_key_lens[type].len;
6da061f4 408 if (!check_attr_len(nla_len(nla), expected_len)) {
7d16c847
PS
409 OVS_NLERR(log, "Key %d has unexpected len %d expected %d",
410 type, nla_len(nla), expected_len);
a097c0b2
PS
411 return -EINVAL;
412 }
413
414 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
415 attrs |= 1ULL << type;
416 a[type] = nla;
417 }
418 }
419 if (rem) {
9233cef7 420 OVS_NLERR(log, "Message has %d unknown bytes.", rem);
a097c0b2
PS
421 return -EINVAL;
422 }
423
424 *attrsp = attrs;
425 return 0;
426}
427
428static int parse_flow_mask_nlattrs(const struct nlattr *attr,
9233cef7
JR
429 const struct nlattr *a[], u64 *attrsp,
430 bool log)
a097c0b2 431{
9233cef7 432 return __parse_flow_nlattrs(attr, a, attrsp, log, true);
a097c0b2
PS
433}
434
435static int parse_flow_nlattrs(const struct nlattr *attr,
9233cef7
JR
436 const struct nlattr *a[], u64 *attrsp,
437 bool log)
a097c0b2 438{
9233cef7 439 return __parse_flow_nlattrs(attr, a, attrsp, log, false);
a097c0b2
PS
440}
441
7d16c847
PS
442static int genev_tun_opt_from_nlattr(const struct nlattr *a,
443 struct sw_flow_match *match, bool is_mask,
444 bool log)
445{
446 unsigned long opt_key_offset;
447
448 if (nla_len(a) > sizeof(match->key->tun_opts)) {
449 OVS_NLERR(log, "Geneve option length err (len %d, max %zu).",
450 nla_len(a), sizeof(match->key->tun_opts));
451 return -EINVAL;
452 }
453
454 if (nla_len(a) % 4 != 0) {
455 OVS_NLERR(log, "Geneve opt len %d is not a multiple of 4.",
456 nla_len(a));
457 return -EINVAL;
458 }
459
460 /* We need to record the length of the options passed
461 * down, otherwise packets with the same format but
462 * additional options will be silently matched.
463 */
464 if (!is_mask) {
465 SW_FLOW_KEY_PUT(match, tun_opts_len, nla_len(a),
466 false);
467 } else {
468 /* This is somewhat unusual because it looks at
469 * both the key and mask while parsing the
470 * attributes (and by extension assumes the key
471 * is parsed first). Normally, we would verify
472 * that each is the correct length and that the
473 * attributes line up in the validate function.
474 * However, that is difficult because this is
475 * variable length and we won't have the
476 * information later.
477 */
478 if (match->key->tun_opts_len != nla_len(a)) {
479 OVS_NLERR(log, "Geneve option len %d != mask len %d",
480 match->key->tun_opts_len, nla_len(a));
481 return -EINVAL;
482 }
483
484 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
485 }
486
4b163224 487 opt_key_offset = TUN_METADATA_OFFSET(nla_len(a));
7d16c847
PS
488 SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, nla_data(a),
489 nla_len(a), is_mask);
490 return 0;
491}
492
6da061f4 493static int vxlan_tun_opt_from_nlattr(const struct nlattr *attr,
0c7930a3
TG
494 struct sw_flow_match *match, bool is_mask,
495 bool log)
496{
6da061f4
JG
497 struct nlattr *a;
498 int rem;
0c7930a3 499 unsigned long opt_key_offset;
e23775f2 500 struct vxlan_metadata opts;
0c7930a3
TG
501
502 BUILD_BUG_ON(sizeof(opts) > sizeof(match->key->tun_opts));
503
0c7930a3 504 memset(&opts, 0, sizeof(opts));
6da061f4
JG
505 nla_for_each_nested(a, attr, rem) {
506 int type = nla_type(a);
0c7930a3 507
6da061f4
JG
508 if (type > OVS_VXLAN_EXT_MAX) {
509 OVS_NLERR(log, "VXLAN extension %d out of range max %d",
510 type, OVS_VXLAN_EXT_MAX);
511 return -EINVAL;
512 }
513
514 if (!check_attr_len(nla_len(a),
515 ovs_vxlan_ext_key_lens[type].len)) {
516 OVS_NLERR(log, "VXLAN extension %d has unexpected len %d expected %d",
517 type, nla_len(a),
518 ovs_vxlan_ext_key_lens[type].len);
519 return -EINVAL;
520 }
521
522 switch (type) {
523 case OVS_VXLAN_EXT_GBP:
524 opts.gbp = nla_get_u32(a);
525 break;
526 default:
527 OVS_NLERR(log, "Unknown VXLAN extension attribute %d",
528 type);
529 return -EINVAL;
530 }
531 }
532 if (rem) {
533 OVS_NLERR(log, "VXLAN extension message has %d unknown bytes.",
534 rem);
535 return -EINVAL;
536 }
0c7930a3
TG
537
538 if (!is_mask)
539 SW_FLOW_KEY_PUT(match, tun_opts_len, sizeof(opts), false);
540 else
541 SW_FLOW_KEY_PUT(match, tun_opts_len, 0xff, true);
542
543 opt_key_offset = TUN_METADATA_OFFSET(sizeof(opts));
544 SW_FLOW_KEY_MEMCPY_OFFSET(match, opt_key_offset, &opts, sizeof(opts),
545 is_mask);
546 return 0;
547}
548
8a2d4905
PS
549static int ip_tun_from_nlattr(const struct nlattr *attr,
550 struct sw_flow_match *match, bool is_mask,
551 bool log)
a097c0b2 552{
8a2d4905 553 bool ttl = false, ipv4 = false, ipv6 = false;
a097c0b2 554 __be16 tun_flags = 0;
0c7930a3 555 int opts_type = 0;
8a2d4905
PS
556 struct nlattr *a;
557 int rem;
a097c0b2
PS
558
559 nla_for_each_nested(a, attr, rem) {
560 int type = nla_type(a);
7d16c847
PS
561 int err;
562
a097c0b2 563 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
7d16c847 564 OVS_NLERR(log, "Tunnel attr %d out of range max %d",
9233cef7 565 type, OVS_TUNNEL_KEY_ATTR_MAX);
a097c0b2
PS
566 return -EINVAL;
567 }
568
6da061f4
JG
569 if (!check_attr_len(nla_len(a),
570 ovs_tunnel_key_lens[type].len)) {
7d16c847 571 OVS_NLERR(log, "Tunnel attr %d has unexpected len %d expected %d",
ec959cdc 572 type, nla_len(a), ovs_tunnel_key_lens[type].len);
a097c0b2
PS
573 return -EINVAL;
574 }
575
576 switch (type) {
577 case OVS_TUNNEL_KEY_ATTR_ID:
578 SW_FLOW_KEY_PUT(match, tun_key.tun_id,
579 nla_get_be64(a), is_mask);
580 tun_flags |= TUNNEL_KEY;
581 break;
582 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
e23775f2 583 SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.src,
dd693f9b 584 nla_get_in_addr(a), is_mask);
8a2d4905 585 ipv4 = true;
a097c0b2
PS
586 break;
587 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
e23775f2 588 SW_FLOW_KEY_PUT(match, tun_key.u.ipv4.dst,
dd693f9b 589 nla_get_in_addr(a), is_mask);
8a2d4905
PS
590 ipv4 = true;
591 break;
592 case OVS_TUNNEL_KEY_ATTR_IPV6_SRC:
593 SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
594 nla_get_in6_addr(a), is_mask);
595 ipv6 = true;
596 break;
597 case OVS_TUNNEL_KEY_ATTR_IPV6_DST:
598 SW_FLOW_KEY_PUT(match, tun_key.u.ipv6.dst,
599 nla_get_in6_addr(a), is_mask);
600 ipv6 = true;
a097c0b2
PS
601 break;
602 case OVS_TUNNEL_KEY_ATTR_TOS:
e23775f2 603 SW_FLOW_KEY_PUT(match, tun_key.tos,
a097c0b2
PS
604 nla_get_u8(a), is_mask);
605 break;
606 case OVS_TUNNEL_KEY_ATTR_TTL:
e23775f2 607 SW_FLOW_KEY_PUT(match, tun_key.ttl,
a097c0b2
PS
608 nla_get_u8(a), is_mask);
609 ttl = true;
610 break;
611 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
612 tun_flags |= TUNNEL_DONT_FRAGMENT;
613 break;
614 case OVS_TUNNEL_KEY_ATTR_CSUM:
615 tun_flags |= TUNNEL_CSUM;
616 break;
8b7ea2d4
WZ
617 case OVS_TUNNEL_KEY_ATTR_TP_SRC:
618 SW_FLOW_KEY_PUT(match, tun_key.tp_src,
619 nla_get_be16(a), is_mask);
620 break;
621 case OVS_TUNNEL_KEY_ATTR_TP_DST:
622 SW_FLOW_KEY_PUT(match, tun_key.tp_dst,
623 nla_get_be16(a), is_mask);
624 break;
94872594
JG
625 case OVS_TUNNEL_KEY_ATTR_OAM:
626 tun_flags |= TUNNEL_OAM;
627 break;
c1fc1411 628 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
0c7930a3
TG
629 if (opts_type) {
630 OVS_NLERR(log, "Multiple metadata blocks provided");
631 return -EINVAL;
632 }
633
7d16c847
PS
634 err = genev_tun_opt_from_nlattr(a, match, is_mask, log);
635 if (err)
636 return err;
c1fc1411 637
0c7930a3
TG
638 tun_flags |= TUNNEL_GENEVE_OPT;
639 opts_type = type;
640 break;
641 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
642 if (opts_type) {
643 OVS_NLERR(log, "Multiple metadata blocks provided");
644 return -EINVAL;
645 }
646
647 err = vxlan_tun_opt_from_nlattr(a, match, is_mask, log);
648 if (err)
649 return err;
650
651 tun_flags |= TUNNEL_VXLAN_OPT;
652 opts_type = type;
c1fc1411 653 break;
a097c0b2 654 default:
8a2d4905 655 OVS_NLERR(log, "Unknown IP tunnel attribute %d",
9233cef7 656 type);
a097c0b2
PS
657 return -EINVAL;
658 }
659 }
660
661 SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
8a2d4905
PS
662 if (is_mask)
663 SW_FLOW_KEY_MEMSET_FIELD(match, tun_proto, 0xff, true);
664 else
665 SW_FLOW_KEY_PUT(match, tun_proto, ipv6 ? AF_INET6 : AF_INET,
666 false);
a097c0b2
PS
667
668 if (rem > 0) {
8a2d4905 669 OVS_NLERR(log, "IP tunnel attribute has %d unknown bytes.",
9233cef7 670 rem);
a097c0b2
PS
671 return -EINVAL;
672 }
673
8a2d4905
PS
674 if (ipv4 && ipv6) {
675 OVS_NLERR(log, "Mixed IPv4 and IPv6 tunnel attributes");
676 return -EINVAL;
677 }
678
a097c0b2 679 if (!is_mask) {
8a2d4905
PS
680 if (!ipv4 && !ipv6) {
681 OVS_NLERR(log, "IP tunnel dst address not specified");
682 return -EINVAL;
683 }
684 if (ipv4 && !match->key->tun_key.u.ipv4.dst) {
7d16c847 685 OVS_NLERR(log, "IPv4 tunnel dst address is zero");
a097c0b2
PS
686 return -EINVAL;
687 }
8a2d4905
PS
688 if (ipv6 && ipv6_addr_any(&match->key->tun_key.u.ipv6.dst)) {
689 OVS_NLERR(log, "IPv6 tunnel dst address is zero");
690 return -EINVAL;
691 }
a097c0b2
PS
692
693 if (!ttl) {
8a2d4905 694 OVS_NLERR(log, "IP tunnel TTL not specified.");
a097c0b2
PS
695 return -EINVAL;
696 }
697 }
698
0c7930a3
TG
699 return opts_type;
700}
701
702static int vxlan_opt_to_nlattr(struct sk_buff *skb,
703 const void *tun_opts, int swkey_tun_opts_len)
704{
e23775f2 705 const struct vxlan_metadata *opts = tun_opts;
0c7930a3
TG
706 struct nlattr *nla;
707
708 nla = nla_nest_start(skb, OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS);
709 if (!nla)
710 return -EMSGSIZE;
711
712 if (nla_put_u32(skb, OVS_VXLAN_EXT_GBP, opts->gbp) < 0)
713 return -EMSGSIZE;
714
715 nla_nest_end(skb, nla);
a097c0b2
PS
716 return 0;
717}
718
8a2d4905
PS
719static int __ip_tun_to_nlattr(struct sk_buff *skb,
720 const struct ip_tunnel_key *output,
721 const void *tun_opts, int swkey_tun_opts_len,
722 unsigned short tun_proto)
a097c0b2 723{
a097c0b2 724 if (output->tun_flags & TUNNEL_KEY &&
f3464818
PS
725 nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id,
726 OVS_TUNNEL_KEY_ATTR_PAD))
a097c0b2 727 return -EMSGSIZE;
8a2d4905
PS
728 switch (tun_proto) {
729 case AF_INET:
730 if (output->u.ipv4.src &&
731 nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC,
732 output->u.ipv4.src))
733 return -EMSGSIZE;
734 if (output->u.ipv4.dst &&
735 nla_put_in_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST,
736 output->u.ipv4.dst))
737 return -EMSGSIZE;
738 break;
739 case AF_INET6:
740 if (!ipv6_addr_any(&output->u.ipv6.src) &&
741 nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_SRC,
742 &output->u.ipv6.src))
743 return -EMSGSIZE;
744 if (!ipv6_addr_any(&output->u.ipv6.dst) &&
745 nla_put_in6_addr(skb, OVS_TUNNEL_KEY_ATTR_IPV6_DST,
746 &output->u.ipv6.dst))
747 return -EMSGSIZE;
748 break;
749 }
e23775f2
PS
750 if (output->tos &&
751 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->tos))
a097c0b2 752 return -EMSGSIZE;
e23775f2 753 if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ttl))
a097c0b2
PS
754 return -EMSGSIZE;
755 if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
7d16c847 756 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
a097c0b2
PS
757 return -EMSGSIZE;
758 if ((output->tun_flags & TUNNEL_CSUM) &&
7d16c847 759 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
a097c0b2 760 return -EMSGSIZE;
8b7ea2d4 761 if (output->tp_src &&
7d16c847 762 nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_SRC, output->tp_src))
8b7ea2d4
WZ
763 return -EMSGSIZE;
764 if (output->tp_dst &&
7d16c847 765 nla_put_be16(skb, OVS_TUNNEL_KEY_ATTR_TP_DST, output->tp_dst))
8b7ea2d4 766 return -EMSGSIZE;
94872594 767 if ((output->tun_flags & TUNNEL_OAM) &&
7d16c847 768 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_OAM))
94872594 769 return -EMSGSIZE;
aad7cb91 770 if (swkey_tun_opts_len) {
0c7930a3
TG
771 if (output->tun_flags & TUNNEL_GENEVE_OPT &&
772 nla_put(skb, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS,
773 swkey_tun_opts_len, tun_opts))
774 return -EMSGSIZE;
e23775f2
PS
775 else if (output->tun_flags & TUNNEL_VXLAN_OPT &&
776 vxlan_opt_to_nlattr(skb, tun_opts, swkey_tun_opts_len))
0c7930a3
TG
777 return -EMSGSIZE;
778 }
a097c0b2 779
8b7ea2d4
WZ
780 return 0;
781}
782
8a2d4905
PS
783static int ip_tun_to_nlattr(struct sk_buff *skb,
784 const struct ip_tunnel_key *output,
785 const void *tun_opts, int swkey_tun_opts_len,
786 unsigned short tun_proto)
8b7ea2d4
WZ
787{
788 struct nlattr *nla;
789 int err;
790
791 nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
792 if (!nla)
793 return -EMSGSIZE;
794
8a2d4905
PS
795 err = __ip_tun_to_nlattr(skb, output, tun_opts, swkey_tun_opts_len,
796 tun_proto);
8b7ea2d4
WZ
797 if (err)
798 return err;
799
a097c0b2
PS
800 nla_nest_end(skb, nla);
801 return 0;
802}
803
aad7cb91
PS
804int ovs_nla_put_tunnel_info(struct sk_buff *skb,
805 struct ip_tunnel_info *tun_info)
8b7ea2d4 806{
8a2d4905
PS
807 return __ip_tun_to_nlattr(skb, &tun_info->key,
808 ip_tunnel_info_opts(tun_info),
809 tun_info->options_len,
810 ip_tunnel_info_af(tun_info));
8b7ea2d4 811}
a097c0b2 812
038e34ab
JS
813static int metadata_from_nlattrs(struct net *net, struct sw_flow_match *match,
814 u64 *attrs, const struct nlattr **a,
815 bool is_mask, bool log)
a097c0b2 816{
7804df20
AZ
817 if (*attrs & (1ULL << OVS_KEY_ATTR_DP_HASH)) {
818 u32 hash_val = nla_get_u32(a[OVS_KEY_ATTR_DP_HASH]);
819
820 SW_FLOW_KEY_PUT(match, ovs_flow_hash, hash_val, is_mask);
821 *attrs &= ~(1ULL << OVS_KEY_ATTR_DP_HASH);
822 }
823
a6059080
AZ
824 if (*attrs & (1ULL << OVS_KEY_ATTR_RECIRC_ID)) {
825 u32 recirc_id = nla_get_u32(a[OVS_KEY_ATTR_RECIRC_ID]);
826
827 SW_FLOW_KEY_PUT(match, recirc_id, recirc_id, is_mask);
828 *attrs &= ~(1ULL << OVS_KEY_ATTR_RECIRC_ID);
829 }
830
a097c0b2
PS
831 if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
832 SW_FLOW_KEY_PUT(match, phy.priority,
833 nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
834 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
835 }
836
837 if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
838 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
839
a473df5b 840 if (is_mask) {
a097c0b2 841 in_port = 0xffffffff; /* Always exact match in_port. */
a473df5b 842 } else if (in_port >= DP_MAX_PORTS) {
7d16c847 843 OVS_NLERR(log, "Port %d exceeds max allowable %d",
a473df5b 844 in_port, DP_MAX_PORTS);
a097c0b2 845 return -EINVAL;
a473df5b 846 }
a097c0b2
PS
847
848 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
849 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
850 } else if (!is_mask) {
851 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
852 }
853
854 if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
855 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
856
857 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
858 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
859 }
860 if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
8a2d4905 861 if (ip_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
0c7930a3 862 is_mask, log) < 0)
a097c0b2
PS
863 return -EINVAL;
864 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
865 }
a94ebc39
JS
866
867 if (*attrs & (1 << OVS_KEY_ATTR_CT_STATE) &&
038e34ab 868 ovs_ct_verify(net, OVS_KEY_ATTR_CT_STATE)) {
c05e2094
JS
869 u32 ct_state = nla_get_u32(a[OVS_KEY_ATTR_CT_STATE]);
870
871 if (ct_state & ~CT_SUPPORTED_MASK) {
872 OVS_NLERR(log, "ct_state flags %08x unsupported",
873 ct_state);
874 return -EINVAL;
875 }
a94ebc39
JS
876
877 SW_FLOW_KEY_PUT(match, ct.state, ct_state, is_mask);
878 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_STATE);
879 }
880 if (*attrs & (1 << OVS_KEY_ATTR_CT_ZONE) &&
038e34ab 881 ovs_ct_verify(net, OVS_KEY_ATTR_CT_ZONE)) {
a94ebc39
JS
882 u16 ct_zone = nla_get_u16(a[OVS_KEY_ATTR_CT_ZONE]);
883
884 SW_FLOW_KEY_PUT(match, ct.zone, ct_zone, is_mask);
885 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_ZONE);
886 }
372ce973 887 if (*attrs & (1 << OVS_KEY_ATTR_CT_MARK) &&
038e34ab 888 ovs_ct_verify(net, OVS_KEY_ATTR_CT_MARK)) {
372ce973
JS
889 u32 mark = nla_get_u32(a[OVS_KEY_ATTR_CT_MARK]);
890
891 SW_FLOW_KEY_PUT(match, ct.mark, mark, is_mask);
892 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_MARK);
893 }
c05e2094
JS
894 if (*attrs & (1 << OVS_KEY_ATTR_CT_LABELS) &&
895 ovs_ct_verify(net, OVS_KEY_ATTR_CT_LABELS)) {
896 const struct ovs_key_ct_labels *cl;
038e34ab 897
c05e2094
JS
898 cl = nla_data(a[OVS_KEY_ATTR_CT_LABELS]);
899 SW_FLOW_KEY_MEMCPY(match, ct.labels, cl->ct_labels,
038e34ab 900 sizeof(*cl), is_mask);
c05e2094 901 *attrs &= ~(1ULL << OVS_KEY_ATTR_CT_LABELS);
038e34ab 902 }
a097c0b2
PS
903 return 0;
904}
905
038e34ab
JS
906static int ovs_key_from_nlattrs(struct net *net, struct sw_flow_match *match,
907 u64 attrs, const struct nlattr **a,
908 bool is_mask, bool log)
a097c0b2
PS
909{
910 int err;
a097c0b2 911
038e34ab 912 err = metadata_from_nlattrs(net, match, &attrs, a, is_mask, log);
a097c0b2
PS
913 if (err)
914 return err;
915
916 if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
917 const struct ovs_key_ethernet *eth_key;
918
919 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
920 SW_FLOW_KEY_MEMCPY(match, eth.src,
921 eth_key->eth_src, ETH_ALEN, is_mask);
922 SW_FLOW_KEY_MEMCPY(match, eth.dst,
923 eth_key->eth_dst, ETH_ALEN, is_mask);
924 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
925 }
926
927 if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
928 __be16 tci;
929
930 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
931 if (!(tci & htons(VLAN_TAG_PRESENT))) {
932 if (is_mask)
7d16c847 933 OVS_NLERR(log, "VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.");
a097c0b2 934 else
7d16c847 935 OVS_NLERR(log, "VLAN TCI does not have VLAN_TAG_PRESENT bit set.");
a097c0b2
PS
936
937 return -EINVAL;
938 }
939
940 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
941 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
f3ccd17d 942 }
a097c0b2
PS
943
944 if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
945 __be16 eth_type;
946
947 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
948 if (is_mask) {
949 /* Always exact match EtherType. */
950 eth_type = htons(0xffff);
935fc582 951 } else if (!eth_proto_is_802_3(eth_type)) {
7d16c847 952 OVS_NLERR(log, "EtherType %x is less than min %x",
9233cef7 953 ntohs(eth_type), ETH_P_802_3_MIN);
a097c0b2
PS
954 return -EINVAL;
955 }
956
957 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
958 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
959 } else if (!is_mask) {
960 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
961 }
962
e23775f2 963 if (attrs & (1 << OVS_KEY_ATTR_IPV4)) {
a097c0b2
PS
964 const struct ovs_key_ipv4 *ipv4_key;
965
966 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
967 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
7d16c847 968 OVS_NLERR(log, "IPv4 frag type %d is out of range max %d",
9233cef7 969 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
a097c0b2
PS
970 return -EINVAL;
971 }
972 SW_FLOW_KEY_PUT(match, ip.proto,
973 ipv4_key->ipv4_proto, is_mask);
974 SW_FLOW_KEY_PUT(match, ip.tos,
975 ipv4_key->ipv4_tos, is_mask);
976 SW_FLOW_KEY_PUT(match, ip.ttl,
977 ipv4_key->ipv4_ttl, is_mask);
978 SW_FLOW_KEY_PUT(match, ip.frag,
979 ipv4_key->ipv4_frag, is_mask);
980 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
981 ipv4_key->ipv4_src, is_mask);
982 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
983 ipv4_key->ipv4_dst, is_mask);
e23775f2 984 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
a097c0b2
PS
985 }
986
987 if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
988 const struct ovs_key_ipv6 *ipv6_key;
989
990 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
991 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
7d16c847 992 OVS_NLERR(log, "IPv6 frag type %d is out of range max %d",
9233cef7 993 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
a097c0b2
PS
994 return -EINVAL;
995 }
a0fb56c1 996
65721038 997 if (!is_mask && ipv6_key->ipv6_label & htonl(0xFFF00000)) {
8a2d4905 998 OVS_NLERR(log, "IPv6 flow label %x is out of range (max=%x).\n",
67782539
JR
999 ntohl(ipv6_key->ipv6_label), (1 << 20) - 1);
1000 return -EINVAL;
1001 }
a0fb56c1 1002
a097c0b2
PS
1003 SW_FLOW_KEY_PUT(match, ipv6.label,
1004 ipv6_key->ipv6_label, is_mask);
1005 SW_FLOW_KEY_PUT(match, ip.proto,
1006 ipv6_key->ipv6_proto, is_mask);
1007 SW_FLOW_KEY_PUT(match, ip.tos,
1008 ipv6_key->ipv6_tclass, is_mask);
1009 SW_FLOW_KEY_PUT(match, ip.ttl,
1010 ipv6_key->ipv6_hlimit, is_mask);
1011 SW_FLOW_KEY_PUT(match, ip.frag,
1012 ipv6_key->ipv6_frag, is_mask);
1013 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1014 ipv6_key->ipv6_src,
1015 sizeof(match->key->ipv6.addr.src),
1016 is_mask);
1017 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1018 ipv6_key->ipv6_dst,
1019 sizeof(match->key->ipv6.addr.dst),
1020 is_mask);
1021
1022 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
1023 }
1024
1025 if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
1026 const struct ovs_key_arp *arp_key;
1027
1028 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1029 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
9233cef7 1030 OVS_NLERR(log, "Unknown ARP opcode (opcode=%d).",
a097c0b2
PS
1031 arp_key->arp_op);
1032 return -EINVAL;
1033 }
1034
1035 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1036 arp_key->arp_sip, is_mask);
1037 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1038 arp_key->arp_tip, is_mask);
1039 SW_FLOW_KEY_PUT(match, ip.proto,
1040 ntohs(arp_key->arp_op), is_mask);
1041 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1042 arp_key->arp_sha, ETH_ALEN, is_mask);
1043 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1044 arp_key->arp_tha, ETH_ALEN, is_mask);
1045
1046 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
1047 }
1048
ccf43786
SH
1049 if (attrs & (1ULL << OVS_KEY_ATTR_MPLS)) {
1050 const struct ovs_key_mpls *mpls_key;
1051
1052 mpls_key = nla_data(a[OVS_KEY_ATTR_MPLS]);
1053 SW_FLOW_KEY_PUT(match, mpls.top_lse,
1054 mpls_key->mpls_lse, is_mask);
1055
1056 attrs &= ~(1ULL << OVS_KEY_ATTR_MPLS);
62974663 1057 }
ccf43786 1058
a097c0b2
PS
1059 if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
1060 const struct ovs_key_tcp *tcp_key;
1061
1062 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
708fb4c5
JR
1063 SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask);
1064 SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask);
a097c0b2
PS
1065 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
1066 }
1067
dc235f7f 1068 if (attrs & (1ULL << OVS_KEY_ATTR_TCP_FLAGS)) {
cab29271
JS
1069 SW_FLOW_KEY_PUT(match, tp.flags,
1070 nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]),
1071 is_mask);
dc235f7f
JR
1072 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP_FLAGS);
1073 }
1074
a097c0b2
PS
1075 if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
1076 const struct ovs_key_udp *udp_key;
1077
1078 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
708fb4c5
JR
1079 SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask);
1080 SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask);
a097c0b2
PS
1081 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
1082 }
1083
1084 if (attrs & (1ULL << OVS_KEY_ATTR_SCTP)) {
1085 const struct ovs_key_sctp *sctp_key;
1086
1087 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
708fb4c5
JR
1088 SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask);
1089 SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask);
a097c0b2
PS
1090 attrs &= ~(1ULL << OVS_KEY_ATTR_SCTP);
1091 }
1092
1093 if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
1094 const struct ovs_key_icmp *icmp_key;
1095
1096 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
708fb4c5 1097 SW_FLOW_KEY_PUT(match, tp.src,
a097c0b2 1098 htons(icmp_key->icmp_type), is_mask);
708fb4c5 1099 SW_FLOW_KEY_PUT(match, tp.dst,
a097c0b2
PS
1100 htons(icmp_key->icmp_code), is_mask);
1101 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
1102 }
1103
1104 if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
1105 const struct ovs_key_icmpv6 *icmpv6_key;
1106
1107 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
708fb4c5 1108 SW_FLOW_KEY_PUT(match, tp.src,
a097c0b2 1109 htons(icmpv6_key->icmpv6_type), is_mask);
708fb4c5 1110 SW_FLOW_KEY_PUT(match, tp.dst,
a097c0b2
PS
1111 htons(icmpv6_key->icmpv6_code), is_mask);
1112 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
1113 }
1114
1115 if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
1116 const struct ovs_key_nd *nd_key;
1117
1118 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1119 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1120 nd_key->nd_target,
1121 sizeof(match->key->ipv6.nd.target),
1122 is_mask);
1123 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1124 nd_key->nd_sll, ETH_ALEN, is_mask);
1125 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1126 nd_key->nd_tll, ETH_ALEN, is_mask);
1127 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
1128 }
1129
a473df5b 1130 if (attrs != 0) {
7d16c847 1131 OVS_NLERR(log, "Unknown key attributes %llx",
a473df5b 1132 (unsigned long long)attrs);
a097c0b2 1133 return -EINVAL;
a473df5b 1134 }
a097c0b2
PS
1135
1136 return 0;
1137}
1138
ec959cdc
TG
1139static void nlattr_set(struct nlattr *attr, u8 val,
1140 const struct ovs_len_tbl *tbl)
a097c0b2 1141{
62974663
DDP
1142 struct nlattr *nla;
1143 int rem;
a097c0b2 1144
62974663
DDP
1145 /* The nlattr stream should already have been validated */
1146 nla_for_each_nested(nla, attr, rem) {
6da061f4
JG
1147 if (tbl[nla_type(nla)].len == OVS_ATTR_NESTED) {
1148 if (tbl[nla_type(nla)].next)
1149 tbl = tbl[nla_type(nla)].next;
1150 nlattr_set(nla, val, tbl);
1151 } else {
62974663 1152 memset(nla_data(nla), val, nla_len(nla));
6da061f4 1153 }
c05e2094
JS
1154
1155 if (nla_type(nla) == OVS_KEY_ATTR_CT_STATE)
1156 *(u32 *)nla_data(nla) &= CT_SUPPORTED_MASK;
62974663
DDP
1157 }
1158}
1159
1160static void mask_set_nlattr(struct nlattr *attr, u8 val)
1161{
ec959cdc 1162 nlattr_set(attr, val, ovs_key_lens);
a097c0b2
PS
1163}
1164
1165/**
1166 * ovs_nla_get_match - parses Netlink attributes into a flow key and
1167 * mask. In case the 'mask' is NULL, the flow is treated as exact match
1168 * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1169 * does not include any don't care bit.
038e34ab 1170 * @net: Used to determine per-namespace field support.
a097c0b2
PS
1171 * @match: receives the extracted flow match information.
1172 * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1173 * sequence. The fields should of the packet that triggered the creation
1174 * of this flow.
1175 * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1176 * attribute specifies the mask field of the wildcarded flow.
9233cef7
JR
1177 * @log: Boolean to allow kernel error logging. Normally true, but when
1178 * probing for feature compatibility this should be passed in as false to
1179 * suppress unnecessary error logging.
a097c0b2 1180 */
038e34ab 1181int ovs_nla_get_match(struct net *net, struct sw_flow_match *match,
f3ccd17d 1182 const struct nlattr *nla_key,
9233cef7
JR
1183 const struct nlattr *nla_mask,
1184 bool log)
a097c0b2
PS
1185{
1186 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1187 const struct nlattr *encap;
62974663 1188 struct nlattr *newmask = NULL;
a097c0b2
PS
1189 u64 key_attrs = 0;
1190 u64 mask_attrs = 0;
1191 bool encap_valid = false;
1192 int err;
1193
9233cef7 1194 err = parse_flow_nlattrs(nla_key, a, &key_attrs, log);
a097c0b2
PS
1195 if (err)
1196 return err;
1197
1198 if ((key_attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1199 (key_attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) &&
1200 (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1201 __be16 tci;
1202
1203 if (!((key_attrs & (1ULL << OVS_KEY_ATTR_VLAN)) &&
1204 (key_attrs & (1ULL << OVS_KEY_ATTR_ENCAP)))) {
9233cef7 1205 OVS_NLERR(log, "Invalid Vlan frame.");
a097c0b2
PS
1206 return -EINVAL;
1207 }
1208
1209 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1210 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1211 encap = a[OVS_KEY_ATTR_ENCAP];
1212 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1213 encap_valid = true;
1214
1215 if (tci & htons(VLAN_TAG_PRESENT)) {
9233cef7 1216 err = parse_flow_nlattrs(encap, a, &key_attrs, log);
a097c0b2
PS
1217 if (err)
1218 return err;
1219 } else if (!tci) {
1220 /* Corner case for truncated 802.1Q header. */
1221 if (nla_len(encap)) {
7d16c847 1222 OVS_NLERR(log, "Truncated 802.1Q header has non-zero encap attribute.");
a097c0b2
PS
1223 return -EINVAL;
1224 }
1225 } else {
7d16c847 1226 OVS_NLERR(log, "Encap attr is set for non-VLAN frame");
a097c0b2
PS
1227 return -EINVAL;
1228 }
1229 }
1230
038e34ab 1231 err = ovs_key_from_nlattrs(net, match, key_attrs, a, false, log);
a097c0b2
PS
1232 if (err)
1233 return err;
1234
f3ccd17d 1235 if (match->mask) {
f3ccd17d
PS
1236 if (!nla_mask) {
1237 /* Create an exact match mask. We need to set to 0xff
1238 * all the 'match->mask' fields that have been touched
1239 * in 'match->key'. We cannot simply memset
1240 * 'match->mask', because padding bytes and fields not
1241 * specified in 'match->key' should be left to 0.
1242 * Instead, we use a stream of netlink attributes,
7d16c847
PS
1243 * copied from 'key' and set to 0xff.
1244 * ovs_key_from_nlattrs() will take care of filling
1245 * 'match->mask' appropriately.
f3ccd17d
PS
1246 */
1247 newmask = kmemdup(nla_key,
1248 nla_total_size(nla_len(nla_key)),
1249 GFP_KERNEL);
1250 if (!newmask)
1251 return -ENOMEM;
62974663 1252
f3ccd17d 1253 mask_set_nlattr(newmask, 0xff);
62974663 1254
f3ccd17d
PS
1255 /* The userspace does not send tunnel attributes that
1256 * are 0, but we should not wildcard them nonetheless.
1257 */
bdcbce89 1258 if (match->key->tun_proto)
f3ccd17d
PS
1259 SW_FLOW_KEY_MEMSET_FIELD(match, tun_key,
1260 0xff, true);
0b496cda 1261
f3ccd17d
PS
1262 nla_mask = newmask;
1263 }
62974663 1264
9233cef7 1265 err = parse_flow_mask_nlattrs(nla_mask, a, &mask_attrs, log);
a097c0b2 1266 if (err)
62974663 1267 goto free_newmask;
a097c0b2 1268
f3ccd17d
PS
1269 /* Always match on tci. */
1270 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1271
62974663 1272 if (mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
a097c0b2
PS
1273 __be16 eth_type = 0;
1274 __be16 tci = 0;
1275
1276 if (!encap_valid) {
7d16c847 1277 OVS_NLERR(log, "Encap mask attribute is set for non-VLAN frame.");
62974663
DDP
1278 err = -EINVAL;
1279 goto free_newmask;
a097c0b2
PS
1280 }
1281
1282 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1283 if (a[OVS_KEY_ATTR_ETHERTYPE])
1284 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1285
1286 if (eth_type == htons(0xffff)) {
1287 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1288 encap = a[OVS_KEY_ATTR_ENCAP];
9233cef7
JR
1289 err = parse_flow_mask_nlattrs(encap, a,
1290 &mask_attrs, log);
3854ab21
AW
1291 if (err)
1292 goto free_newmask;
a097c0b2 1293 } else {
7d16c847 1294 OVS_NLERR(log, "VLAN frames must have an exact match on the TPID (mask=%x).",
9233cef7 1295 ntohs(eth_type));
62974663
DDP
1296 err = -EINVAL;
1297 goto free_newmask;
a097c0b2
PS
1298 }
1299
1300 if (a[OVS_KEY_ATTR_VLAN])
1301 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1302
1303 if (!(tci & htons(VLAN_TAG_PRESENT))) {
7d16c847 1304 OVS_NLERR(log, "VLAN tag present bit must have an exact match (tci_mask=%x).",
9233cef7 1305 ntohs(tci));
62974663
DDP
1306 err = -EINVAL;
1307 goto free_newmask;
a097c0b2
PS
1308 }
1309 }
1310
038e34ab
JS
1311 err = ovs_key_from_nlattrs(net, match, mask_attrs, a, true,
1312 log);
a097c0b2 1313 if (err)
62974663 1314 goto free_newmask;
a097c0b2
PS
1315 }
1316
9233cef7 1317 if (!match_validate(match, key_attrs, mask_attrs, log))
62974663 1318 err = -EINVAL;
a097c0b2 1319
62974663
DDP
1320free_newmask:
1321 kfree(newmask);
1322 return err;
a097c0b2
PS
1323}
1324
bc619e29
JS
1325static size_t get_ufid_len(const struct nlattr *attr, bool log)
1326{
1327 size_t len;
1328
1329 if (!attr)
1330 return 0;
1331
1332 len = nla_len(attr);
1333 if (len < 1 || len > MAX_UFID_LENGTH) {
1334 OVS_NLERR(log, "ufid size %u bytes exceeds the range (1, %d)",
1335 nla_len(attr), MAX_UFID_LENGTH);
1336 return 0;
1337 }
1338
1339 return len;
1340}
1341
1342/* Initializes 'flow->ufid', returning true if 'attr' contains a valid UFID,
1343 * or false otherwise.
1344 */
1345bool ovs_nla_get_ufid(struct sw_flow_id *sfid, const struct nlattr *attr,
1346 bool log)
1347{
1348 sfid->ufid_len = get_ufid_len(attr, log);
1349 if (sfid->ufid_len)
1350 memcpy(sfid->ufid, nla_data(attr), sfid->ufid_len);
1351
1352 return sfid->ufid_len;
1353}
1354
1355int ovs_nla_get_identifier(struct sw_flow_id *sfid, const struct nlattr *ufid,
1356 const struct sw_flow_key *key, bool log)
1357{
1358 struct sw_flow_key *new_key;
1359
1360 if (ovs_nla_get_ufid(sfid, ufid, log))
1361 return 0;
1362
1363 /* If UFID was not provided, use unmasked key. */
1364 new_key = kmalloc(sizeof(*new_key), GFP_KERNEL);
1365 if (!new_key)
1366 return -ENOMEM;
1367 memcpy(new_key, key, sizeof(*key));
1368 sfid->unmasked_key = new_key;
1369
1370 return 0;
1371}
1372
1373u32 ovs_nla_get_ufid_flags(const struct nlattr *attr)
1374{
1375 return attr ? nla_get_u32(attr) : 0;
1376}
1377
a097c0b2
PS
1378/**
1379 * ovs_nla_get_flow_metadata - parses Netlink attributes into a flow key.
c135bba1 1380 * @key: Receives extracted in_port, priority, tun_key and skb_mark.
a097c0b2
PS
1381 * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1382 * sequence.
9233cef7
JR
1383 * @log: Boolean to allow kernel error logging. Normally true, but when
1384 * probing for feature compatibility this should be passed in as false to
1385 * suppress unnecessary error logging.
a097c0b2
PS
1386 *
1387 * This parses a series of Netlink attributes that form a flow key, which must
1388 * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1389 * get the metadata, that is, the parts of the flow key that cannot be
1390 * extracted from the packet itself.
1391 */
7d16c847 1392
038e34ab 1393int ovs_nla_get_flow_metadata(struct net *net, const struct nlattr *attr,
9233cef7
JR
1394 struct sw_flow_key *key,
1395 bool log)
a097c0b2 1396{
a097c0b2 1397 const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
c135bba1 1398 struct sw_flow_match match;
a097c0b2
PS
1399 u64 attrs = 0;
1400 int err;
a097c0b2 1401
9233cef7 1402 err = parse_flow_nlattrs(attr, a, &attrs, log);
a097c0b2
PS
1403 if (err)
1404 return -EINVAL;
1405
1406 memset(&match, 0, sizeof(match));
c135bba1 1407 match.key = key;
a097c0b2 1408
a94ebc39 1409 memset(&key->ct, 0, sizeof(key->ct));
c135bba1 1410 key->phy.in_port = DP_MAX_PORTS;
a097c0b2 1411
038e34ab 1412 return metadata_from_nlattrs(net, &match, &attrs, a, false, log);
a097c0b2
PS
1413}
1414
db7f2238
JS
1415static int __ovs_nla_put_key(const struct sw_flow_key *swkey,
1416 const struct sw_flow_key *output, bool is_mask,
1417 struct sk_buff *skb)
a097c0b2
PS
1418{
1419 struct ovs_key_ethernet *eth_key;
1420 struct nlattr *nla, *encap;
a097c0b2 1421
7d16c847 1422 if (nla_put_u32(skb, OVS_KEY_ATTR_RECIRC_ID, output->recirc_id))
7804df20
AZ
1423 goto nla_put_failure;
1424
7d16c847 1425 if (nla_put_u32(skb, OVS_KEY_ATTR_DP_HASH, output->ovs_flow_hash))
a6059080
AZ
1426 goto nla_put_failure;
1427
a097c0b2
PS
1428 if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1429 goto nla_put_failure;
1430
8a2d4905 1431 if ((swkey->tun_proto || is_mask)) {
4b163224 1432 const void *opts = NULL;
c1fc1411 1433
fa6395df 1434 if (output->tun_key.tun_flags & TUNNEL_OPTIONS_PRESENT)
4b163224 1435 opts = TUN_METADATA_OPTS(output, swkey->tun_opts_len);
c1fc1411 1436
8a2d4905
PS
1437 if (ip_tun_to_nlattr(skb, &output->tun_key, opts,
1438 swkey->tun_opts_len, swkey->tun_proto))
c1fc1411
JG
1439 goto nla_put_failure;
1440 }
a097c0b2
PS
1441
1442 if (swkey->phy.in_port == DP_MAX_PORTS) {
1443 if (is_mask && (output->phy.in_port == 0xffff))
1444 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1445 goto nla_put_failure;
1446 } else {
1447 u16 upper_u16;
1448 upper_u16 = !is_mask ? 0 : 0xffff;
1449
1450 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1451 (upper_u16 << 16) | output->phy.in_port))
1452 goto nla_put_failure;
1453 }
1454
1455 if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1456 goto nla_put_failure;
1457
a94ebc39
JS
1458 if (ovs_ct_put_key(output, skb))
1459 goto nla_put_failure;
1460
a097c0b2
PS
1461 nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1462 if (!nla)
1463 goto nla_put_failure;
1464
1465 eth_key = nla_data(nla);
982a47ec
JP
1466 ether_addr_copy(eth_key->eth_src, output->eth.src);
1467 ether_addr_copy(eth_key->eth_dst, output->eth.dst);
a097c0b2
PS
1468
1469 if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1470 __be16 eth_type;
1471 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1472 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1473 nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1474 goto nla_put_failure;
1475 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1476 if (!swkey->eth.tci)
1477 goto unencap;
1478 } else
1479 encap = NULL;
1480
1481 if (swkey->eth.type == htons(ETH_P_802_2)) {
1482 /*
1483 * Ethertype 802.2 is represented in the netlink with omitted
1484 * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1485 * 0xffff in the mask attribute. Ethertype can also
1486 * be wildcarded.
1487 */
1488 if (is_mask && output->eth.type)
1489 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1490 output->eth.type))
1491 goto nla_put_failure;
1492 goto unencap;
1493 }
1494
1495 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1496 goto nla_put_failure;
1497
1498 if (swkey->eth.type == htons(ETH_P_IP)) {
1499 struct ovs_key_ipv4 *ipv4_key;
1500
1501 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1502 if (!nla)
1503 goto nla_put_failure;
1504 ipv4_key = nla_data(nla);
1505 ipv4_key->ipv4_src = output->ipv4.addr.src;
1506 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1507 ipv4_key->ipv4_proto = output->ip.proto;
1508 ipv4_key->ipv4_tos = output->ip.tos;
1509 ipv4_key->ipv4_ttl = output->ip.ttl;
1510 ipv4_key->ipv4_frag = output->ip.frag;
1511 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1512 struct ovs_key_ipv6 *ipv6_key;
1513
1514 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1515 if (!nla)
1516 goto nla_put_failure;
1517 ipv6_key = nla_data(nla);
1518 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1519 sizeof(ipv6_key->ipv6_src));
1520 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1521 sizeof(ipv6_key->ipv6_dst));
1522 ipv6_key->ipv6_label = output->ipv6.label;
1523 ipv6_key->ipv6_proto = output->ip.proto;
1524 ipv6_key->ipv6_tclass = output->ip.tos;
1525 ipv6_key->ipv6_hlimit = output->ip.ttl;
1526 ipv6_key->ipv6_frag = output->ip.frag;
1527 } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1528 swkey->eth.type == htons(ETH_P_RARP)) {
1529 struct ovs_key_arp *arp_key;
1530
1531 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1532 if (!nla)
1533 goto nla_put_failure;
1534 arp_key = nla_data(nla);
1535 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1536 arp_key->arp_sip = output->ipv4.addr.src;
1537 arp_key->arp_tip = output->ipv4.addr.dst;
1538 arp_key->arp_op = htons(output->ip.proto);
982a47ec
JP
1539 ether_addr_copy(arp_key->arp_sha, output->ipv4.arp.sha);
1540 ether_addr_copy(arp_key->arp_tha, output->ipv4.arp.tha);
ccf43786
SH
1541 } else if (eth_p_mpls(swkey->eth.type)) {
1542 struct ovs_key_mpls *mpls_key;
1543
1544 nla = nla_reserve(skb, OVS_KEY_ATTR_MPLS, sizeof(*mpls_key));
1545 if (!nla)
1546 goto nla_put_failure;
1547 mpls_key = nla_data(nla);
1548 mpls_key->mpls_lse = output->mpls.top_lse;
a097c0b2
PS
1549 }
1550
1551 if ((swkey->eth.type == htons(ETH_P_IP) ||
1552 swkey->eth.type == htons(ETH_P_IPV6)) &&
1553 swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1554
1555 if (swkey->ip.proto == IPPROTO_TCP) {
1556 struct ovs_key_tcp *tcp_key;
1557
1558 nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1559 if (!nla)
1560 goto nla_put_failure;
1561 tcp_key = nla_data(nla);
708fb4c5
JR
1562 tcp_key->tcp_src = output->tp.src;
1563 tcp_key->tcp_dst = output->tp.dst;
1564 if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS,
1565 output->tp.flags))
1566 goto nla_put_failure;
a097c0b2
PS
1567 } else if (swkey->ip.proto == IPPROTO_UDP) {
1568 struct ovs_key_udp *udp_key;
1569
1570 nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1571 if (!nla)
1572 goto nla_put_failure;
1573 udp_key = nla_data(nla);
708fb4c5
JR
1574 udp_key->udp_src = output->tp.src;
1575 udp_key->udp_dst = output->tp.dst;
a097c0b2
PS
1576 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1577 struct ovs_key_sctp *sctp_key;
1578
1579 nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1580 if (!nla)
1581 goto nla_put_failure;
1582 sctp_key = nla_data(nla);
708fb4c5
JR
1583 sctp_key->sctp_src = output->tp.src;
1584 sctp_key->sctp_dst = output->tp.dst;
a097c0b2
PS
1585 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1586 swkey->ip.proto == IPPROTO_ICMP) {
1587 struct ovs_key_icmp *icmp_key;
1588
1589 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1590 if (!nla)
1591 goto nla_put_failure;
1592 icmp_key = nla_data(nla);
708fb4c5
JR
1593 icmp_key->icmp_type = ntohs(output->tp.src);
1594 icmp_key->icmp_code = ntohs(output->tp.dst);
a097c0b2
PS
1595 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1596 swkey->ip.proto == IPPROTO_ICMPV6) {
1597 struct ovs_key_icmpv6 *icmpv6_key;
1598
1599 nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1600 sizeof(*icmpv6_key));
1601 if (!nla)
1602 goto nla_put_failure;
1603 icmpv6_key = nla_data(nla);
708fb4c5
JR
1604 icmpv6_key->icmpv6_type = ntohs(output->tp.src);
1605 icmpv6_key->icmpv6_code = ntohs(output->tp.dst);
a097c0b2
PS
1606
1607 if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1608 icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1609 struct ovs_key_nd *nd_key;
1610
1611 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1612 if (!nla)
1613 goto nla_put_failure;
1614 nd_key = nla_data(nla);
1615 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1616 sizeof(nd_key->nd_target));
982a47ec
JP
1617 ether_addr_copy(nd_key->nd_sll, output->ipv6.nd.sll);
1618 ether_addr_copy(nd_key->nd_tll, output->ipv6.nd.tll);
a097c0b2
PS
1619 }
1620 }
1621 }
1622
1623unencap:
1624 if (encap)
1625 nla_nest_end(skb, encap);
1626
1627 return 0;
1628
1629nla_put_failure:
1630 return -EMSGSIZE;
1631}
1632
db7f2238
JS
1633int ovs_nla_put_key(const struct sw_flow_key *swkey,
1634 const struct sw_flow_key *output, int attr, bool is_mask,
1635 struct sk_buff *skb)
1636{
1637 int err;
1638 struct nlattr *nla;
1639
1640 nla = nla_nest_start(skb, attr);
1641 if (!nla)
1642 return -EMSGSIZE;
1643 err = __ovs_nla_put_key(swkey, output, is_mask, skb);
1644 if (err)
1645 return err;
1646 nla_nest_end(skb, nla);
1647
1648 return 0;
1649}
1650
1651/* Called with ovs_mutex or RCU read lock. */
bc619e29
JS
1652int ovs_nla_put_identifier(const struct sw_flow *flow, struct sk_buff *skb)
1653{
1654 if (ovs_identifier_is_ufid(&flow->id))
1655 return nla_put(skb, OVS_FLOW_ATTR_UFID, flow->id.ufid_len,
1656 flow->id.ufid);
1657
1658 return ovs_nla_put_key(flow->id.unmasked_key, flow->id.unmasked_key,
1659 OVS_FLOW_ATTR_KEY, false, skb);
1660}
1661
1662/* Called with ovs_mutex or RCU read lock. */
1663int ovs_nla_put_masked_key(const struct sw_flow *flow, struct sk_buff *skb)
db7f2238 1664{
b037bc5a 1665 return ovs_nla_put_key(&flow->key, &flow->key,
db7f2238
JS
1666 OVS_FLOW_ATTR_KEY, false, skb);
1667}
1668
1669/* Called with ovs_mutex or RCU read lock. */
1670int ovs_nla_put_mask(const struct sw_flow *flow, struct sk_buff *skb)
1671{
1672 return ovs_nla_put_key(&flow->key, &flow->mask->key,
1673 OVS_FLOW_ATTR_MASK, true, skb);
1674}
1675
a097c0b2
PS
1676#define MAX_ACTIONS_BUFSIZE (32 * 1024)
1677
9233cef7 1678static struct sw_flow_actions *nla_alloc_flow_actions(int size, bool log)
a097c0b2
PS
1679{
1680 struct sw_flow_actions *sfa;
1681
a473df5b 1682 if (size > MAX_ACTIONS_BUFSIZE) {
7d16c847 1683 OVS_NLERR(log, "Flow action size %u bytes exceeds max", size);
a097c0b2 1684 return ERR_PTR(-EINVAL);
a473df5b 1685 }
a097c0b2
PS
1686
1687 sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
1688 if (!sfa)
1689 return ERR_PTR(-ENOMEM);
1690
1691 sfa->actions_len = 0;
1692 return sfa;
1693}
1694
e23775f2 1695static void ovs_nla_free_set_action(const struct nlattr *a)
a097c0b2 1696{
e23775f2
PS
1697 const struct nlattr *ovs_key = nla_data(a);
1698 struct ovs_tunnel_info *ovs_tun;
1699
1700 switch (nla_type(ovs_key)) {
1701 case OVS_KEY_ATTR_TUNNEL_INFO:
1702 ovs_tun = nla_data(ovs_key);
1703 ovs_dst_release((struct dst_entry *)ovs_tun->tun_dst);
1704 break;
1705 }
1706}
1707
1708void ovs_nla_free_flow_actions(struct sw_flow_actions *sf_acts)
1709{
1710 const struct nlattr *a;
1711 int rem;
1712
1713 if (!sf_acts)
1714 return;
1715
1716 nla_for_each_attr(a, sf_acts->actions, sf_acts->actions_len, rem) {
1717 switch (nla_type(a)) {
1718 case OVS_ACTION_ATTR_SET:
1719 ovs_nla_free_set_action(a);
1720 break;
a94ebc39
JS
1721 case OVS_ACTION_ATTR_CT:
1722 ovs_ct_free_action(a);
1723 break;
e23775f2
PS
1724 }
1725 }
1726
a097c0b2
PS
1727 kfree(sf_acts);
1728}
1729
e23775f2
PS
1730static void __ovs_nla_free_flow_actions(struct rcu_head *head)
1731{
1732 ovs_nla_free_flow_actions(container_of(head, struct sw_flow_actions, rcu));
1733}
1734
a097c0b2 1735/* Schedules 'sf_acts' to be freed after the next RCU grace period.
8a2d4905 1736 * The caller must hold rcu_read_lock for this to be sensible. */
e23775f2 1737void ovs_nla_free_flow_actions_rcu(struct sw_flow_actions *sf_acts)
a097c0b2 1738{
e23775f2 1739 call_rcu(&sf_acts->rcu, __ovs_nla_free_flow_actions);
a097c0b2
PS
1740}
1741
1742static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa,
9233cef7 1743 int attr_len, bool log)
a097c0b2
PS
1744{
1745
1746 struct sw_flow_actions *acts;
1747 int new_acts_size;
1748 int req_size = NLA_ALIGN(attr_len);
1749 int next_offset = offsetof(struct sw_flow_actions, actions) +
1750 (*sfa)->actions_len;
1751
1752 if (req_size <= (ksize(*sfa) - next_offset))
1753 goto out;
1754
1755 new_acts_size = ksize(*sfa) * 2;
1756
1757 if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
1758 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
1759 return ERR_PTR(-EMSGSIZE);
1760 new_acts_size = MAX_ACTIONS_BUFSIZE;
1761 }
1762
9233cef7 1763 acts = nla_alloc_flow_actions(new_acts_size, log);
a097c0b2
PS
1764 if (IS_ERR(acts))
1765 return (void *)acts;
1766
1767 memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
1768 acts->actions_len = (*sfa)->actions_len;
c3bb15b3 1769 acts->orig_len = (*sfa)->orig_len;
a097c0b2
PS
1770 kfree(*sfa);
1771 *sfa = acts;
1772
1773out:
1774 (*sfa)->actions_len += req_size;
1775 return (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
1776}
1777
7d16c847
PS
1778static struct nlattr *__add_action(struct sw_flow_actions **sfa,
1779 int attrtype, void *data, int len, bool log)
a097c0b2
PS
1780{
1781 struct nlattr *a;
1782
9233cef7 1783 a = reserve_sfa_size(sfa, nla_attr_size(len), log);
a097c0b2 1784 if (IS_ERR(a))
f0cd669f 1785 return a;
a097c0b2
PS
1786
1787 a->nla_type = attrtype;
1788 a->nla_len = nla_attr_size(len);
1789
1790 if (data)
1791 memcpy(nla_data(a), data, len);
1792 memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
1793
f0cd669f
JG
1794 return a;
1795}
1796
a94ebc39
JS
1797int ovs_nla_add_action(struct sw_flow_actions **sfa, int attrtype, void *data,
1798 int len, bool log)
f0cd669f
JG
1799{
1800 struct nlattr *a;
1801
9233cef7 1802 a = __add_action(sfa, attrtype, data, len, log);
f0cd669f 1803
54125dd0 1804 return PTR_ERR_OR_ZERO(a);
a097c0b2
PS
1805}
1806
1807static inline int add_nested_action_start(struct sw_flow_actions **sfa,
9233cef7 1808 int attrtype, bool log)
a097c0b2
PS
1809{
1810 int used = (*sfa)->actions_len;
1811 int err;
1812
a94ebc39 1813 err = ovs_nla_add_action(sfa, attrtype, NULL, 0, log);
a097c0b2
PS
1814 if (err)
1815 return err;
1816
1817 return used;
1818}
1819
1820static inline void add_nested_action_end(struct sw_flow_actions *sfa,
1821 int st_offset)
1822{
1823 struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions +
1824 st_offset);
1825
1826 a->nla_len = sfa->actions_len - st_offset;
1827}
1828
a94ebc39 1829static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
ccf43786
SH
1830 const struct sw_flow_key *key,
1831 int depth, struct sw_flow_actions **sfa,
9233cef7 1832 __be16 eth_type, __be16 vlan_tci, bool log);
ccf43786 1833
a94ebc39 1834static int validate_and_copy_sample(struct net *net, const struct nlattr *attr,
a097c0b2 1835 const struct sw_flow_key *key, int depth,
ccf43786 1836 struct sw_flow_actions **sfa,
9233cef7 1837 __be16 eth_type, __be16 vlan_tci, bool log)
a097c0b2
PS
1838{
1839 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
1840 const struct nlattr *probability, *actions;
1841 const struct nlattr *a;
1842 int rem, start, err, st_acts;
1843
1844 memset(attrs, 0, sizeof(attrs));
1845 nla_for_each_nested(a, attr, rem) {
1846 int type = nla_type(a);
1847 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
1848 return -EINVAL;
1849 attrs[type] = a;
1850 }
1851 if (rem)
1852 return -EINVAL;
1853
1854 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
1855 if (!probability || nla_len(probability) != sizeof(u32))
1856 return -EINVAL;
1857
1858 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
1859 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
1860 return -EINVAL;
1861
1862 /* validation done, copy sample action. */
9233cef7 1863 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE, log);
a097c0b2
PS
1864 if (start < 0)
1865 return start;
a94ebc39
JS
1866 err = ovs_nla_add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY,
1867 nla_data(probability), sizeof(u32), log);
a097c0b2
PS
1868 if (err)
1869 return err;
9233cef7 1870 st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS, log);
a097c0b2
PS
1871 if (st_acts < 0)
1872 return st_acts;
1873
a94ebc39 1874 err = __ovs_nla_copy_actions(net, actions, key, depth + 1, sfa,
9233cef7 1875 eth_type, vlan_tci, log);
a097c0b2
PS
1876 if (err)
1877 return err;
1878
1879 add_nested_action_end(*sfa, st_acts);
1880 add_nested_action_end(*sfa, start);
1881
1882 return 0;
1883}
1884
a097c0b2
PS
1885void ovs_match_init(struct sw_flow_match *match,
1886 struct sw_flow_key *key,
1887 struct sw_flow_mask *mask)
1888{
1889 memset(match, 0, sizeof(*match));
1890 match->key = key;
1891 match->mask = mask;
1892
1893 memset(key, 0, sizeof(*key));
1894
1895 if (mask) {
1896 memset(&mask->key, 0, sizeof(mask->key));
1897 mask->range.start = mask->range.end = 0;
1898 }
1899}
1900
4b163224
TG
1901static int validate_geneve_opts(struct sw_flow_key *key)
1902{
1903 struct geneve_opt *option;
1904 int opts_len = key->tun_opts_len;
1905 bool crit_opt = false;
1906
1907 option = (struct geneve_opt *)TUN_METADATA_OPTS(key, key->tun_opts_len);
1908 while (opts_len > 0) {
1909 int len;
1910
1911 if (opts_len < sizeof(*option))
1912 return -EINVAL;
1913
1914 len = sizeof(*option) + option->length * 4;
1915 if (len > opts_len)
1916 return -EINVAL;
1917
1918 crit_opt |= !!(option->type & GENEVE_CRIT_OPT_TYPE);
1919
1920 option = (struct geneve_opt *)((u8 *)option + len);
1921 opts_len -= len;
1922 };
1923
1924 key->tun_key.tun_flags |= crit_opt ? TUNNEL_CRIT_OPT : 0;
1925
1926 return 0;
1927}
1928
a097c0b2 1929static int validate_and_copy_set_tun(const struct nlattr *attr,
9233cef7 1930 struct sw_flow_actions **sfa, bool log)
a097c0b2
PS
1931{
1932 struct sw_flow_match match;
1933 struct sw_flow_key key;
e23775f2
PS
1934 struct metadata_dst *tun_dst;
1935 struct ip_tunnel_info *tun_info;
1936 struct ovs_tunnel_info *ovs_tun;
f0cd669f 1937 struct nlattr *a;
e23775f2 1938 int err = 0, start, opts_type;
a097c0b2
PS
1939
1940 ovs_match_init(&match, &key, NULL);
8a2d4905 1941 opts_type = ip_tun_from_nlattr(nla_data(attr), &match, false, log);
0c7930a3
TG
1942 if (opts_type < 0)
1943 return opts_type;
a097c0b2 1944
c1fc1411 1945 if (key.tun_opts_len) {
0c7930a3
TG
1946 switch (opts_type) {
1947 case OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS:
1948 err = validate_geneve_opts(&key);
1949 if (err < 0)
1950 return err;
1951 break;
1952 case OVS_TUNNEL_KEY_ATTR_VXLAN_OPTS:
1953 break;
1954 }
c1fc1411
JG
1955 };
1956
9233cef7 1957 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET, log);
a097c0b2
PS
1958 if (start < 0)
1959 return start;
1960
e23775f2
PS
1961 tun_dst = metadata_dst_alloc(key.tun_opts_len, GFP_KERNEL);
1962 if (!tun_dst)
1963 return -ENOMEM;
1964
e02c0ed7
PS
1965 err = dst_cache_init(&tun_dst->u.tun_info.dst_cache, GFP_KERNEL);
1966 if (err) {
1967 dst_release((struct dst_entry *)tun_dst);
1968 return err;
1969 }
f0cd669f 1970 a = __add_action(sfa, OVS_KEY_ATTR_TUNNEL_INFO, NULL,
e23775f2
PS
1971 sizeof(*ovs_tun), log);
1972 if (IS_ERR(a)) {
1973 ovs_dst_release((struct dst_entry *)tun_dst);
f0cd669f 1974 return PTR_ERR(a);
e23775f2 1975 }
f0cd669f 1976
e23775f2
PS
1977 ovs_tun = nla_data(a);
1978 ovs_tun->tun_dst = tun_dst;
c1fc1411 1979
e23775f2
PS
1980 tun_info = &tun_dst->u.tun_info;
1981 tun_info->mode = IP_TUNNEL_INFO_TX;
8a2d4905
PS
1982 if (key.tun_proto == AF_INET6)
1983 tun_info->mode |= IP_TUNNEL_INFO_IPV6;
e23775f2 1984 tun_info->key = key.tun_key;
f0cd669f 1985
e23775f2
PS
1986 /* We need to store the options in the action itself since
1987 * everything else will go away after flow setup. We can append
1988 * it to tun_info and then point there.
1989 */
1990 ip_tunnel_info_opts_set(tun_info,
1991 TUN_METADATA_OPTS(&key, key.tun_opts_len),
1992 key.tun_opts_len);
a097c0b2
PS
1993 add_nested_action_end(*sfa, start);
1994
1995 return err;
1996}
1997
b940b3d7
JR
1998/* Return false if there are any non-masked bits set.
1999 * Mask follows data immediately, before any netlink padding.
2000 */
2001static bool validate_masked(u8 *data, int len)
2002{
2003 u8 *mask = data + len;
2004
2005 while (len--)
2006 if (*data++ & ~*mask++)
2007 return false;
2008
2009 return true;
2010}
2011
a097c0b2
PS
2012static int validate_set(const struct nlattr *a,
2013 const struct sw_flow_key *flow_key,
2014 struct sw_flow_actions **sfa,
b940b3d7 2015 bool *skip_copy, __be16 eth_type, bool masked, bool log)
a097c0b2
PS
2016{
2017 const struct nlattr *ovs_key = nla_data(a);
2018 int key_type = nla_type(ovs_key);
b940b3d7 2019 size_t key_len;
a097c0b2
PS
2020
2021 /* There can be only one key in a action */
2022 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
2023 return -EINVAL;
2024
b940b3d7
JR
2025 key_len = nla_len(ovs_key);
2026 if (masked)
2027 key_len /= 2;
2028
a097c0b2 2029 if (key_type > OVS_KEY_ATTR_MAX ||
6da061f4 2030 !check_attr_len(key_len, ovs_key_lens[key_type].len))
a097c0b2
PS
2031 return -EINVAL;
2032
b940b3d7
JR
2033 if (masked && !validate_masked(nla_data(ovs_key), key_len))
2034 return -EINVAL;
2035
a097c0b2
PS
2036 switch (key_type) {
2037 const struct ovs_key_ipv4 *ipv4_key;
2038 const struct ovs_key_ipv6 *ipv6_key;
2039 int err;
2040
2041 case OVS_KEY_ATTR_PRIORITY:
2042 case OVS_KEY_ATTR_SKB_MARK:
372ce973 2043 case OVS_KEY_ATTR_CT_MARK:
c05e2094 2044 case OVS_KEY_ATTR_CT_LABELS:
a097c0b2
PS
2045 case OVS_KEY_ATTR_ETHERNET:
2046 break;
2047
2048 case OVS_KEY_ATTR_TUNNEL:
4c3c5643 2049#ifndef USE_UPSTREAM_TUNNEL
2baf0e0c
PS
2050 if (eth_p_mpls(eth_type))
2051 return -EINVAL;
4c3c5643 2052#endif
b940b3d7
JR
2053 if (masked)
2054 return -EINVAL; /* Masked tunnel set not supported. */
2055
2056 *skip_copy = true;
9233cef7 2057 err = validate_and_copy_set_tun(a, sfa, log);
a097c0b2
PS
2058 if (err)
2059 return err;
2060 break;
2061
2062 case OVS_KEY_ATTR_IPV4:
ccf43786 2063 if (eth_type != htons(ETH_P_IP))
a097c0b2
PS
2064 return -EINVAL;
2065
a097c0b2 2066 ipv4_key = nla_data(ovs_key);
a097c0b2 2067
b940b3d7
JR
2068 if (masked) {
2069 const struct ovs_key_ipv4 *mask = ipv4_key + 1;
a097c0b2 2070
b940b3d7
JR
2071 /* Non-writeable fields. */
2072 if (mask->ipv4_proto || mask->ipv4_frag)
2073 return -EINVAL;
2074 } else {
2075 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
2076 return -EINVAL;
2077
2078 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
2079 return -EINVAL;
2080 }
a097c0b2
PS
2081 break;
2082
2083 case OVS_KEY_ATTR_IPV6:
ccf43786 2084 if (eth_type != htons(ETH_P_IPV6))
a097c0b2
PS
2085 return -EINVAL;
2086
a097c0b2 2087 ipv6_key = nla_data(ovs_key);
a0fb56c1 2088
b940b3d7
JR
2089 if (masked) {
2090 const struct ovs_key_ipv6 *mask = ipv6_key + 1;
a097c0b2 2091
b940b3d7
JR
2092 /* Non-writeable fields. */
2093 if (mask->ipv6_proto || mask->ipv6_frag)
2094 return -EINVAL;
a097c0b2 2095
b940b3d7
JR
2096 /* Invalid bits in the flow label mask? */
2097 if (ntohl(mask->ipv6_label) & 0xFFF00000)
2098 return -EINVAL;
2099 } else {
2100 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
2101 return -EINVAL;
2102
2103 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
2104 return -EINVAL;
2105 }
a097c0b2
PS
2106 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
2107 return -EINVAL;
2108
2109 break;
2110
2111 case OVS_KEY_ATTR_TCP:
b940b3d7
JR
2112 if ((eth_type != htons(ETH_P_IP) &&
2113 eth_type != htons(ETH_P_IPV6)) ||
2114 flow_key->ip.proto != IPPROTO_TCP)
a097c0b2
PS
2115 return -EINVAL;
2116
b940b3d7 2117 break;
a097c0b2
PS
2118
2119 case OVS_KEY_ATTR_UDP:
b940b3d7
JR
2120 if ((eth_type != htons(ETH_P_IP) &&
2121 eth_type != htons(ETH_P_IPV6)) ||
2122 flow_key->ip.proto != IPPROTO_UDP)
a097c0b2
PS
2123 return -EINVAL;
2124
b940b3d7 2125 break;
ccf43786
SH
2126
2127 case OVS_KEY_ATTR_MPLS:
2128 if (!eth_p_mpls(eth_type))
2129 return -EINVAL;
2130 break;
a097c0b2
PS
2131
2132 case OVS_KEY_ATTR_SCTP:
b940b3d7
JR
2133 if ((eth_type != htons(ETH_P_IP) &&
2134 eth_type != htons(ETH_P_IPV6)) ||
2135 flow_key->ip.proto != IPPROTO_SCTP)
a097c0b2
PS
2136 return -EINVAL;
2137
b940b3d7 2138 break;
a097c0b2
PS
2139
2140 default:
2141 return -EINVAL;
2142 }
2143
b940b3d7
JR
2144 /* Convert non-masked non-tunnel set actions to masked set actions. */
2145 if (!masked && key_type != OVS_KEY_ATTR_TUNNEL) {
2146 int start, len = key_len * 2;
2147 struct nlattr *at;
2148
2149 *skip_copy = true;
2150
2151 start = add_nested_action_start(sfa,
2152 OVS_ACTION_ATTR_SET_TO_MASKED,
2153 log);
2154 if (start < 0)
2155 return start;
2156
2157 at = __add_action(sfa, key_type, NULL, len, log);
2158 if (IS_ERR(at))
2159 return PTR_ERR(at);
2160
2161 memcpy(nla_data(at), nla_data(ovs_key), key_len); /* Key. */
2162 memset(nla_data(at) + key_len, 0xff, key_len); /* Mask. */
2163 /* Clear non-writeable bits from otherwise writeable fields. */
2164 if (key_type == OVS_KEY_ATTR_IPV6) {
2165 struct ovs_key_ipv6 *mask = nla_data(at) + key_len;
2166
2167 mask->ipv6_label &= htonl(0x000FFFFF);
2168 }
2169 add_nested_action_end(*sfa, start);
2170 }
2171
a097c0b2
PS
2172 return 0;
2173}
2174
2175static int validate_userspace(const struct nlattr *attr)
2176{
2177 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
2178 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
2179 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
8b7ea2d4 2180 [OVS_USERSPACE_ATTR_EGRESS_TUN_PORT] = {.type = NLA_U32 },
a097c0b2
PS
2181 };
2182 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
2183 int error;
2184
2185 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
2186 attr, userspace_policy);
2187 if (error)
2188 return error;
2189
2190 if (!a[OVS_USERSPACE_ATTR_PID] ||
2191 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
2192 return -EINVAL;
2193
2194 return 0;
2195}
2196
2197static int copy_action(const struct nlattr *from,
9233cef7 2198 struct sw_flow_actions **sfa, bool log)
a097c0b2
PS
2199{
2200 int totlen = NLA_ALIGN(from->nla_len);
2201 struct nlattr *to;
2202
9233cef7 2203 to = reserve_sfa_size(sfa, from->nla_len, log);
a097c0b2
PS
2204 if (IS_ERR(to))
2205 return PTR_ERR(to);
2206
2207 memcpy(to, from, totlen);
2208 return 0;
2209}
2210
a94ebc39 2211static int __ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
ccf43786
SH
2212 const struct sw_flow_key *key,
2213 int depth, struct sw_flow_actions **sfa,
9233cef7 2214 __be16 eth_type, __be16 vlan_tci, bool log)
a097c0b2
PS
2215{
2216 const struct nlattr *a;
2217 int rem, err;
2218
2219 if (depth >= SAMPLE_ACTION_DEPTH)
2220 return -EOVERFLOW;
2221
2222 nla_for_each_nested(a, attr, rem) {
2223 /* Expected argument lengths, (u32)-1 for variable length. */
2224 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
2225 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
a6059080 2226 [OVS_ACTION_ATTR_RECIRC] = sizeof(u32),
a097c0b2 2227 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
ccf43786
SH
2228 [OVS_ACTION_ATTR_PUSH_MPLS] = sizeof(struct ovs_action_push_mpls),
2229 [OVS_ACTION_ATTR_POP_MPLS] = sizeof(__be16),
a097c0b2
PS
2230 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
2231 [OVS_ACTION_ATTR_POP_VLAN] = 0,
2232 [OVS_ACTION_ATTR_SET] = (u32)-1,
b940b3d7 2233 [OVS_ACTION_ATTR_SET_MASKED] = (u32)-1,
7804df20 2234 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1,
a94ebc39
JS
2235 [OVS_ACTION_ATTR_HASH] = sizeof(struct ovs_action_hash),
2236 [OVS_ACTION_ATTR_CT] = (u32)-1,
4c7804f1 2237 [OVS_ACTION_ATTR_TRUNC] = sizeof(struct ovs_action_trunc),
a097c0b2
PS
2238 };
2239 const struct ovs_action_push_vlan *vlan;
2240 int type = nla_type(a);
2241 bool skip_copy;
2242
2243 if (type > OVS_ACTION_ATTR_MAX ||
2244 (action_lens[type] != nla_len(a) &&
2245 action_lens[type] != (u32)-1))
2246 return -EINVAL;
2247
2248 skip_copy = false;
2249 switch (type) {
2250 case OVS_ACTION_ATTR_UNSPEC:
2251 return -EINVAL;
2252
2253 case OVS_ACTION_ATTR_USERSPACE:
2254 err = validate_userspace(a);
2255 if (err)
2256 return err;
2257 break;
2258
2259 case OVS_ACTION_ATTR_OUTPUT:
2260 if (nla_get_u32(a) >= DP_MAX_PORTS)
2261 return -EINVAL;
2262 break;
2263
4c7804f1
WT
2264 case OVS_ACTION_ATTR_TRUNC: {
2265 const struct ovs_action_trunc *trunc = nla_data(a);
2266
2267 if (trunc->max_len < ETH_HLEN)
2268 return -EINVAL;
2269 break;
2270 }
2271
7804df20
AZ
2272 case OVS_ACTION_ATTR_HASH: {
2273 const struct ovs_action_hash *act_hash = nla_data(a);
2274
2275 switch (act_hash->hash_alg) {
2276 case OVS_HASH_ALG_L4:
2277 break;
2278 default:
2279 return -EINVAL;
2280 }
2281
2282 break;
2283 }
a097c0b2
PS
2284
2285 case OVS_ACTION_ATTR_POP_VLAN:
e0b8f73f 2286 vlan_tci = htons(0);
a097c0b2
PS
2287 break;
2288
2289 case OVS_ACTION_ATTR_PUSH_VLAN:
2290 vlan = nla_data(a);
2291 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
2292 return -EINVAL;
2293 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
2294 return -EINVAL;
ccf43786 2295 vlan_tci = vlan->vlan_tci;
a097c0b2
PS
2296 break;
2297
a6059080
AZ
2298 case OVS_ACTION_ATTR_RECIRC:
2299 break;
2300
ccf43786
SH
2301 case OVS_ACTION_ATTR_PUSH_MPLS: {
2302 const struct ovs_action_push_mpls *mpls = nla_data(a);
2303
2304 if (!eth_p_mpls(mpls->mpls_ethertype))
2305 return -EINVAL;
2306 /* Prohibit push MPLS other than to a white list
2307 * for packets that have a known tag order.
e0b8f73f 2308 */
ccf43786
SH
2309 if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2310 (eth_type != htons(ETH_P_IP) &&
2311 eth_type != htons(ETH_P_IPV6) &&
2312 eth_type != htons(ETH_P_ARP) &&
2313 eth_type != htons(ETH_P_RARP) &&
2314 !eth_p_mpls(eth_type)))
2315 return -EINVAL;
2316 eth_type = mpls->mpls_ethertype;
2317 break;
2318 }
2319
2320 case OVS_ACTION_ATTR_POP_MPLS:
2321 if (vlan_tci & htons(VLAN_TAG_PRESENT) ||
2322 !eth_p_mpls(eth_type))
2323 return -EINVAL;
2324
2325 /* Disallow subsequent L2.5+ set and mpls_pop actions
2326 * as there is no check here to ensure that the new
2327 * eth_type is valid and thus set actions could
2328 * write off the end of the packet or otherwise
2329 * corrupt it.
2330 *
2331 * Support for these actions is planned using packet
2332 * recirculation.
2333 */
2334 eth_type = htons(0);
2335 break;
2336
a097c0b2 2337 case OVS_ACTION_ATTR_SET:
2baf0e0c 2338 err = validate_set(a, key, sfa,
b940b3d7
JR
2339 &skip_copy, eth_type, false, log);
2340 if (err)
2341 return err;
2342 break;
2343
2344 case OVS_ACTION_ATTR_SET_MASKED:
2345 err = validate_set(a, key, sfa,
2346 &skip_copy, eth_type, true, log);
a097c0b2
PS
2347 if (err)
2348 return err;
2349 break;
2350
2351 case OVS_ACTION_ATTR_SAMPLE:
a94ebc39 2352 err = validate_and_copy_sample(net, a, key, depth, sfa,
9233cef7 2353 eth_type, vlan_tci, log);
a097c0b2
PS
2354 if (err)
2355 return err;
2356 skip_copy = true;
2357 break;
2358
a94ebc39
JS
2359 case OVS_ACTION_ATTR_CT:
2360 err = ovs_ct_copy_action(net, a, key, sfa, log);
2361 if (err)
2362 return err;
2363 skip_copy = true;
2364 break;
2365
a097c0b2 2366 default:
7d16c847 2367 OVS_NLERR(log, "Unknown Action type %d", type);
a097c0b2
PS
2368 return -EINVAL;
2369 }
2370 if (!skip_copy) {
9233cef7 2371 err = copy_action(a, sfa, log);
a097c0b2
PS
2372 if (err)
2373 return err;
2374 }
2375 }
2376
2377 if (rem > 0)
2378 return -EINVAL;
2379
2380 return 0;
2381}
2382
b940b3d7 2383/* 'key' must be the masked key. */
a94ebc39 2384int ovs_nla_copy_actions(struct net *net, const struct nlattr *attr,
ccf43786 2385 const struct sw_flow_key *key,
9233cef7 2386 struct sw_flow_actions **sfa, bool log)
ccf43786 2387{
ff27161e
PS
2388 int err;
2389
9233cef7 2390 *sfa = nla_alloc_flow_actions(nla_len(attr), log);
ff27161e
PS
2391 if (IS_ERR(*sfa))
2392 return PTR_ERR(*sfa);
2393
c3bb15b3 2394 (*sfa)->orig_len = nla_len(attr);
a94ebc39 2395 err = __ovs_nla_copy_actions(net, attr, key, 0, sfa, key->eth.type,
9233cef7 2396 key->eth.tci, log);
ff27161e 2397 if (err)
e23775f2 2398 ovs_nla_free_flow_actions(*sfa);
ff27161e
PS
2399
2400 return err;
ccf43786
SH
2401}
2402
a097c0b2
PS
2403static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
2404{
2405 const struct nlattr *a;
2406 struct nlattr *start;
2407 int err = 0, rem;
2408
2409 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
2410 if (!start)
2411 return -EMSGSIZE;
2412
2413 nla_for_each_nested(a, attr, rem) {
2414 int type = nla_type(a);
2415 struct nlattr *st_sample;
2416
2417 switch (type) {
2418 case OVS_SAMPLE_ATTR_PROBABILITY:
2419 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY,
2420 sizeof(u32), nla_data(a)))
2421 return -EMSGSIZE;
2422 break;
2423 case OVS_SAMPLE_ATTR_ACTIONS:
2424 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
2425 if (!st_sample)
2426 return -EMSGSIZE;
2427 err = ovs_nla_put_actions(nla_data(a), nla_len(a), skb);
2428 if (err)
2429 return err;
2430 nla_nest_end(skb, st_sample);
2431 break;
2432 }
2433 }
2434
2435 nla_nest_end(skb, start);
2436 return err;
2437}
2438
2439static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
2440{
2441 const struct nlattr *ovs_key = nla_data(a);
2442 int key_type = nla_type(ovs_key);
2443 struct nlattr *start;
2444 int err;
2445
2446 switch (key_type) {
f0cd669f 2447 case OVS_KEY_ATTR_TUNNEL_INFO: {
e23775f2
PS
2448 struct ovs_tunnel_info *ovs_tun = nla_data(ovs_key);
2449 struct ip_tunnel_info *tun_info = &ovs_tun->tun_dst->u.tun_info;
f0cd669f 2450
a097c0b2
PS
2451 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2452 if (!start)
2453 return -EMSGSIZE;
2454
29e7e7d0
PS
2455 err = ip_tun_to_nlattr(skb, &tun_info->key,
2456 ip_tunnel_info_opts(tun_info),
2457 tun_info->options_len,
2458 ip_tunnel_info_af(tun_info));
a097c0b2
PS
2459 if (err)
2460 return err;
2461 nla_nest_end(skb, start);
2462 break;
f0cd669f 2463 }
a097c0b2
PS
2464 default:
2465 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
2466 return -EMSGSIZE;
2467 break;
2468 }
2469
2470 return 0;
2471}
2472
b940b3d7
JR
2473static int masked_set_action_to_set_action_attr(const struct nlattr *a,
2474 struct sk_buff *skb)
2475{
2476 const struct nlattr *ovs_key = nla_data(a);
60642e4f 2477 struct nlattr *nla;
b940b3d7
JR
2478 size_t key_len = nla_len(ovs_key) / 2;
2479
2480 /* Revert the conversion we did from a non-masked set action to
2481 * masked set action.
2482 */
60642e4f
PS
2483 nla = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
2484 if (!nla)
b940b3d7
JR
2485 return -EMSGSIZE;
2486
60642e4f
PS
2487 if (nla_put(skb, nla_type(ovs_key), key_len, nla_data(ovs_key)))
2488 return -EMSGSIZE;
2489
2490 nla_nest_end(skb, nla);
b940b3d7
JR
2491 return 0;
2492}
2493
a097c0b2
PS
2494int ovs_nla_put_actions(const struct nlattr *attr, int len, struct sk_buff *skb)
2495{
2496 const struct nlattr *a;
2497 int rem, err;
2498
2499 nla_for_each_attr(a, attr, len, rem) {
2500 int type = nla_type(a);
2501
2502 switch (type) {
2503 case OVS_ACTION_ATTR_SET:
2504 err = set_action_to_attr(a, skb);
2505 if (err)
2506 return err;
2507 break;
2508
b940b3d7
JR
2509 case OVS_ACTION_ATTR_SET_TO_MASKED:
2510 err = masked_set_action_to_set_action_attr(a, skb);
2511 if (err)
2512 return err;
2513 break;
2514
a097c0b2
PS
2515 case OVS_ACTION_ATTR_SAMPLE:
2516 err = sample_action_to_attr(a, skb);
2517 if (err)
2518 return err;
2519 break;
a94ebc39
JS
2520
2521 case OVS_ACTION_ATTR_CT:
2522 err = ovs_ct_action_to_attr(nla_data(a), skb);
2523 if (err)
2524 return err;
2525 break;
2526
a097c0b2
PS
2527 default:
2528 if (nla_put(skb, type, nla_len(a), nla_data(a)))
2529 return -EMSGSIZE;
2530 break;
2531 }
2532 }
2533
2534 return 0;
2535}