projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
net: phy: dp83869: Fix return paths to return proper values
[linux-2.6-block.git] / net / openvswitch / datapath.c
CommitLineData
c9422999 1// SPDX-License-Identifier: GPL-2.0-only
ccb1352e 2/*
ad552007 3 * Copyright (c) 2007-2014 Nicira, Inc.
ccb1352e
JG		 4 */
5
6#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7
8#include <linux/init.h>
9#include <linux/module.h>
10#include <linux/if_arp.h>
11#include <linux/if_vlan.h>
12#include <linux/in.h>
13#include <linux/ip.h>
14#include <linux/jhash.h>
15#include <linux/delay.h>
16#include <linux/time.h>
17#include <linux/etherdevice.h>
18#include <linux/genetlink.h>
19#include <linux/kernel.h>
20#include <linux/kthread.h>
21#include <linux/mutex.h>
22#include <linux/percpu.h>
23#include <linux/rcupdate.h>
24#include <linux/tcp.h>
25#include <linux/udp.h>
ccb1352e
JG		 26#include <linux/ethtool.h>
27#include <linux/wait.h>
ccb1352e
JG		 28#include <asm/div64.h>
29#include <linux/highmem.h>
30#include <linux/netfilter_bridge.h>
31#include <linux/netfilter_ipv4.h>
32#include <linux/inetdevice.h>
33#include <linux/list.h>
34#include <linux/openvswitch.h>
35#include <linux/rculist.h>
36#include <linux/dmi.h>
ccb1352e 37#include <net/genetlink.h>
46df7b81
PS		 38#include <net/net_namespace.h>
39#include <net/netns/generic.h>
ccb1352e
JG		 40
41#include "datapath.h"
42#include "flow.h"
e80857cc 43#include "flow_table.h"
e6445719 44#include "flow_netlink.h"
96fbc13d 45#include "meter.h"
ccb1352e 46#include "vport-internal_dev.h"
cff63a52 47#include "vport-netdev.h"
ccb1352e 48
c7d03a00 49unsigned int ovs_net_id __read_mostly;
8e4e1713 50
0c200ef9
PS		 51static struct genl_family dp_packet_genl_family;
52static struct genl_family dp_flow_genl_family;
53static struct genl_family dp_datapath_genl_family;
54
74ed7ab9
JS		 55static const struct nla_policy flow_policy[];
56
48e48a70 57static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
58 .name = OVS_FLOW_MCGROUP,
0c200ef9
PS		 59};
60
48e48a70 61static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
62 .name = OVS_DATAPATH_MCGROUP,
0c200ef9
PS		 63};
64
48e48a70 65static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
66 .name = OVS_VPORT_MCGROUP,
0c200ef9
PS		 67};
68
fb5d1e9e
JR		 69/* Check if need to build a reply message.
70 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
9b67aa4a
SG		 71static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
72 unsigned int group)
fb5d1e9e
JR		 73{
74 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
f8403a2e 75 genl_has_listeners(family, genl_info_net(info), group);
fb5d1e9e
JR		 76}
77
68eb5503 78static void ovs_notify(struct genl_family *family,
2a94fe48 79 struct sk_buff *skb, struct genl_info *info)
ed661185 80{
92c14d9b 81 genl_notify(family, skb, info, 0, GFP_KERNEL);
ed661185
TG		 82}
83
ccb1352e
JG		 84/**
85 * DOC: Locking:
86 *
8e4e1713
PS		 87 * All writes e.g. Writes to device state (add/remove datapath, port, set
88 * operations on vports, etc.), Writes to other state (flow table
89 * modifications, set miscellaneous datapath parameters, etc.) are protected
90 * by ovs_lock.
ccb1352e
JG		 91 *
92 * Reads are protected by RCU.
93 *
94 * There are a few special cases (mostly stats) that have their own
95 * synchronization but they nest under all of above and don't interact with
96 * each other.
8e4e1713
PS		 97 *
98 * The RTNL lock nests inside ovs_mutex.
ccb1352e
JG		 99 */
100
8e4e1713
PS		 101static DEFINE_MUTEX(ovs_mutex);
102
103void ovs_lock(void)
104{
105 mutex_lock(&ovs_mutex);
106}
107
108void ovs_unlock(void)
109{
110 mutex_unlock(&ovs_mutex);
111}
112
113#ifdef CONFIG_LOCKDEP
114int lockdep_ovsl_is_held(void)
115{
116 if (debug_locks)
117 return lockdep_is_held(&ovs_mutex);
118 else
119 return 1;
120}
121#endif
122
ccb1352e 123static struct vport *new_vport(const struct vport_parms *);
8055a89c 124static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
e8eedb85 125 const struct sw_flow_key *,
f2a4d086
WT		 126 const struct dp_upcall_info *,
127 uint32_t cutlen);
8055a89c 128static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
e8eedb85 129 const struct sw_flow_key *,
f2a4d086
WT		 130 const struct dp_upcall_info *,
131 uint32_t cutlen);
ccb1352e 132
8e4e1713 133/* Must be called with rcu_read_lock or ovs_mutex. */
971427f3 134const char *ovs_dp_name(const struct datapath *dp)
ccb1352e 135{
8e4e1713 136 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
c9db965c 137 return ovs_vport_name(vport);
ccb1352e
JG		 138}
139
12eb18f7 140static int get_dpifindex(const struct datapath *dp)
ccb1352e
JG		 141{
142 struct vport *local;
143 int ifindex;
144
145 rcu_read_lock();
146
15eac2a7 147 local = ovs_vport_rcu(dp, OVSP_LOCAL);
ccb1352e 148 if (local)
be4ace6e 149 ifindex = local->dev->ifindex;
ccb1352e
JG		 150 else
151 ifindex = 0;
152
153 rcu_read_unlock();
154
155 return ifindex;
156}
157
158static void destroy_dp_rcu(struct rcu_head *rcu)
159{
160 struct datapath *dp = container_of(rcu, struct datapath, rcu);
161
9b996e54 162 ovs_flow_tbl_destroy(&dp->table);
ccb1352e 163 free_percpu(dp->stats_percpu);
15eac2a7 164 kfree(dp->ports);
96fbc13d 165 ovs_meters_exit(dp);
ccb1352e
JG		 166 kfree(dp);
167}
168
15eac2a7
PS		 169static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
170 u16 port_no)
171{
172 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
173}
174
bb6f9a70 175/* Called with ovs_mutex or RCU read lock. */
15eac2a7
PS		 176struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
177{
178 struct vport *vport;
15eac2a7
PS		 179 struct hlist_head *head;
180
181 head = vport_hash_bucket(dp, port_no);
b67bfe0d 182 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
15eac2a7
PS		 183 if (vport->port_no == port_no)
184 return vport;
185 }
186 return NULL;
187}
188
8e4e1713 189/* Called with ovs_mutex. */
ccb1352e
JG		 190static struct vport *new_vport(const struct vport_parms *parms)
191{
192 struct vport *vport;
193
194 vport = ovs_vport_add(parms);
195 if (!IS_ERR(vport)) {
196 struct datapath *dp = parms->dp;
15eac2a7 197 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
ccb1352e 198
15eac2a7 199 hlist_add_head_rcu(&vport->dp_hash_node, head);
ccb1352e 200 }
ccb1352e
JG		 201 return vport;
202}
203
ccb1352e
JG		 204void ovs_dp_detach_port(struct vport *p)
205{
8e4e1713 206 ASSERT_OVSL();
ccb1352e
JG		 207
208 /* First drop references to device. */
15eac2a7 209 hlist_del_rcu(&p->dp_hash_node);
ccb1352e
JG		 210
211 /* Then destroy it. */
212 ovs_vport_del(p);
213}
214
215/* Must be called with rcu_read_lock. */
8c8b1b83 216void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
ccb1352e 217{
83c8df26 218 const struct vport *p = OVS_CB(skb)->input_vport;
ccb1352e
JG		 219 struct datapath *dp = p->dp;
220 struct sw_flow *flow;
d98612b8 221 struct sw_flow_actions *sf_acts;
ccb1352e 222 struct dp_stats_percpu *stats;
ccb1352e 223 u64 *stats_counter;
1bd7116f 224 u32 n_mask_hit;
aa733660 225 int error;
ccb1352e 226
404f2f10 227 stats = this_cpu_ptr(dp->stats_percpu);
ccb1352e 228
ccb1352e 229 /* Look up flow. */
04b7d136
TZ		 230 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
231 &n_mask_hit);
ccb1352e
JG		 232 if (unlikely(!flow)) {
233 struct dp_upcall_info upcall;
234
ccea7445 235 memset(&upcall, 0, sizeof(upcall));
ccb1352e 236 upcall.cmd = OVS_PACKET_CMD_MISS;
5cd667b0 237 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
7f8a436e 238 upcall.mru = OVS_CB(skb)->mru;
f2a4d086 239 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
c5eba0b6
LR		 240 if (unlikely(error))
241 kfree_skb(skb);
242 else
243 consume_skb(skb);
ccb1352e
JG		 244 stats_counter = &stats->n_missed;
245 goto out;
246 }
247
d98612b8
LJ		 248 ovs_flow_stats_update(flow, key->tp.flags, skb);
249 sf_acts = rcu_dereference(flow->sf_acts);
aa733660
YS		 250 error = ovs_execute_actions(dp, skb, sf_acts, key);
251 if (unlikely(error))
252 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
253 ovs_dp_name(dp), error);
ccb1352e 254
e298e505 255 stats_counter = &stats->n_hit;
ccb1352e
JG		 256
257out:
258 /* Update datapath statistics. */
df9d9fdf 259 u64_stats_update_begin(&stats->syncp);
ccb1352e 260 (*stats_counter)++;
1bd7116f 261 stats->n_mask_hit += n_mask_hit;
df9d9fdf 262 u64_stats_update_end(&stats->syncp);
ccb1352e
JG		 263}
264
ccb1352e 265int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
e8eedb85 266 const struct sw_flow_key *key,
f2a4d086
WT		 267 const struct dp_upcall_info *upcall_info,
268 uint32_t cutlen)
ccb1352e
JG		 269{
270 struct dp_stats_percpu *stats;
ccb1352e
JG		 271 int err;
272
15e47304 273 if (upcall_info->portid == 0) {
ccb1352e
JG		 274 err = -ENOTCONN;
275 goto err;
276 }
277
ccb1352e 278 if (!skb_is_gso(skb))
f2a4d086 279 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
ccb1352e 280 else
f2a4d086 281 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
ccb1352e
JG		 282 if (err)
283 goto err;
284
285 return 0;
286
287err:
404f2f10 288 stats = this_cpu_ptr(dp->stats_percpu);
ccb1352e 289
df9d9fdf 290 u64_stats_update_begin(&stats->syncp);
ccb1352e 291 stats->n_lost++;
df9d9fdf 292 u64_stats_update_end(&stats->syncp);
ccb1352e
JG		 293
294 return err;
295}
296
8055a89c 297static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
e8eedb85 298 const struct sw_flow_key *key,
f2a4d086
WT		 299 const struct dp_upcall_info *upcall_info,
300 uint32_t cutlen)
ccb1352e 301{
2734166e 302 unsigned int gso_type = skb_shinfo(skb)->gso_type;
0c19f846 303 struct sw_flow_key later_key;
ccb1352e
JG		 304 struct sk_buff *segs, *nskb;
305 int err;
306
9207f9d4 307 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
09c5e605 308 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
92e5dfc3
PS		 309 if (IS_ERR(segs))
310 return PTR_ERR(segs);
330966e5
FW		 311 if (segs == NULL)
312 return -EINVAL;
ccb1352e 313
0c19f846
WB		 314 if (gso_type & SKB_GSO_UDP) {
315 /* The initial flow key extracted by ovs_flow_key_extract()
316 * in this case is for a first fragment, so we need to
317 * properly mark later fragments.
318 */
319 later_key = *key;
320 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
321 }
322
ccb1352e
JG		 323 /* Queue all of the segments. */
324 skb = segs;
325 do {
0c19f846
WB		 326 if (gso_type & SKB_GSO_UDP && skb != segs)
327 key = &later_key;
328
f2a4d086 329 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
ccb1352e
JG		 330 if (err)
331 break;
332
ccb1352e
JG		 333 } while ((skb = skb->next));
334
335 /* Free all of the segments. */
336 skb = segs;
337 do {
338 nskb = skb->next;
339 if (err)
340 kfree_skb(skb);
341 else
342 consume_skb(skb);
343 } while ((skb = nskb));
344 return err;
345}
346
8f0aad6f 347static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
494bea39 348 unsigned int hdrlen, int actions_attrlen)
c3ff8cfe
TG		 349{
350 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
bda56f14 351 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
b95e5928 352 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
bd1903b7
TZ		 353 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
354 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
c3ff8cfe
TG		 355
356 /* OVS_PACKET_ATTR_USERDATA */
8f0aad6f
WZ		 357 if (upcall_info->userdata)
358 size += NLA_ALIGN(upcall_info->userdata->nla_len);
359
360 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
361 if (upcall_info->egress_tun_info)
362 size += nla_total_size(ovs_tun_key_attr_size());
c3ff8cfe 363
ccea7445
NM		 364 /* OVS_PACKET_ATTR_ACTIONS */
365 if (upcall_info->actions_len)
494bea39 366 size += nla_total_size(actions_attrlen);
ccea7445 367
7f8a436e
JS		 368 /* OVS_PACKET_ATTR_MRU */
369 if (upcall_info->mru)
370 size += nla_total_size(sizeof(upcall_info->mru));
371
c3ff8cfe
TG		 372 return size;
373}
374
7f8a436e
JS		 375static void pad_packet(struct datapath *dp, struct sk_buff *skb)
376{
377 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
378 size_t plen = NLA_ALIGN(skb->len) - skb->len;
379
380 if (plen > 0)
b080db58 381 skb_put_zero(skb, plen);
7f8a436e
JS		 382 }
383}
384
8055a89c 385static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
e8eedb85 386 const struct sw_flow_key *key,
f2a4d086
WT		 387 const struct dp_upcall_info *upcall_info,
388 uint32_t cutlen)
ccb1352e
JG		 389{
390 struct ovs_header *upcall;
391 struct sk_buff *nskb = NULL;
4ee45ea0 392 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
ccb1352e 393 struct nlattr *nla;
795449d8 394 size_t len;
bda56f14 395 unsigned int hlen;
8055a89c 396 int err, dp_ifindex;
bd1903b7 397 u64 hash;
8055a89c
TG		 398
399 dp_ifindex = get_dpifindex(dp);
400 if (!dp_ifindex)
401 return -ENODEV;
ccb1352e 402
df8a39de 403 if (skb_vlan_tag_present(skb)) {
ccb1352e
JG		 404 nskb = skb_clone(skb, GFP_ATOMIC);
405 if (!nskb)
406 return -ENOMEM;
407
5968250c 408 nskb = __vlan_hwaccel_push_inside(nskb);
8aa51d64 409 if (!nskb)
ccb1352e
JG		 410 return -ENOMEM;
411
ccb1352e
JG		 412 skb = nskb;
413 }
414
415 if (nla_attr_size(skb->len) > USHRT_MAX) {
416 err = -EFBIG;
417 goto out;
418 }
419
bda56f14
TG		 420 /* Complete checksum if needed */
421 if (skb->ip_summed == CHECKSUM_PARTIAL &&
7529390d 422 (err = skb_csum_hwoffload_help(skb, 0)))
bda56f14
TG		 423 goto out;
424
425 /* Older versions of OVS user space enforce alignment of the last
426 * Netlink attribute to NLA_ALIGNTO which would require extensive
427 * padding logic. Only perform zerocopy if padding is not required.
428 */
429 if (dp->user_features & OVS_DP_F_UNALIGNED)
430 hlen = skb_zerocopy_headlen(skb);
431 else
432 hlen = skb->len;
433
494bea39
LZ		 434 len = upcall_msg_size(upcall_info, hlen - cutlen,
435 OVS_CB(skb)->acts_origlen);
551ddc05 436 user_skb = genlmsg_new(len, GFP_ATOMIC);
ccb1352e
JG		 437 if (!user_skb) {
438 err = -ENOMEM;
439 goto out;
440 }
441
442 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
443 0, upcall_info->cmd);
6f19893b
KL		 444 if (!upcall) {
445 err = -EINVAL;
446 goto out;
447 }
ccb1352e
JG		 448 upcall->dp_ifindex = dp_ifindex;
449
5b4237bb 450 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
a734d1f4
EC		 451 if (err)
452 goto out;
ccb1352e
JG		 453
454 if (upcall_info->userdata)
4490108b
BP		 455 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
456 nla_len(upcall_info->userdata),
457 nla_data(upcall_info->userdata));
ccb1352e 458
8f0aad6f 459 if (upcall_info->egress_tun_info) {
ae0be8de
MK		 460 nla = nla_nest_start_noflag(user_skb,
461 OVS_PACKET_ATTR_EGRESS_TUN_KEY);
0fff9bd4
KL		 462 if (!nla) {
463 err = -EMSGSIZE;
464 goto out;
465 }
fc4099f1
PS		 466 err = ovs_nla_put_tunnel_info(user_skb,
467 upcall_info->egress_tun_info);
a734d1f4
EC		 468 if (err)
469 goto out;
470
8f0aad6f
WZ		 471 nla_nest_end(user_skb, nla);
472 }
473
ccea7445 474 if (upcall_info->actions_len) {
ae0be8de 475 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
0fff9bd4
KL		 476 if (!nla) {
477 err = -EMSGSIZE;
478 goto out;
479 }
ccea7445
NM		 480 err = ovs_nla_put_actions(upcall_info->actions,
481 upcall_info->actions_len,
482 user_skb);
483 if (!err)
484 nla_nest_end(user_skb, nla);
485 else
486 nla_nest_cancel(user_skb, nla);
487 }
488
7f8a436e 489 /* Add OVS_PACKET_ATTR_MRU */
61ca533c
TZ		 490 if (upcall_info->mru &&
491 nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
492 err = -ENOBUFS;
493 goto out;
7f8a436e
JS		 494 }
495
b95e5928 496 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
61ca533c
TZ		 497 if (cutlen > 0 &&
498 nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
499 err = -ENOBUFS;
500 goto out;
b95e5928
WT		 501 }
502
bd1903b7
TZ		 503 /* Add OVS_PACKET_ATTR_HASH */
504 hash = skb_get_hash_raw(skb);
505 if (skb->sw_hash)
506 hash |= OVS_PACKET_HASH_SW_BIT;
507
508 if (skb->l4_hash)
509 hash |= OVS_PACKET_HASH_L4_BIT;
510
511 if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
512 err = -ENOBUFS;
513 goto out;
514 }
515
bda56f14
TG		 516 /* Only reserve room for attribute header, packet data is added
517 * in skb_zerocopy() */
518 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
519 err = -ENOBUFS;
520 goto out;
521 }
f2a4d086 522 nla->nla_len = nla_attr_size(skb->len - cutlen);
ccb1352e 523
f2a4d086 524 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
36d5fe6a
ZK		 525 if (err)
526 goto out;
ccb1352e 527
aea0bb4f 528 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
7f8a436e 529 pad_packet(dp, user_skb);
aea0bb4f 530
bda56f14 531 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
ccb1352e 532
bda56f14 533 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
4ee45ea0 534 user_skb = NULL;
ccb1352e 535out:
36d5fe6a
ZK		 536 if (err)
537 skb_tx_error(skb);
4ee45ea0 538 kfree_skb(user_skb);
ccb1352e
JG		 539 kfree_skb(nskb);
540 return err;
541}
542
ccb1352e
JG		 543static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
544{
545 struct ovs_header *ovs_header = info->userhdr;
7f8a436e 546 struct net *net = sock_net(skb->sk);
ccb1352e
JG		 547 struct nlattr **a = info->attrs;
548 struct sw_flow_actions *acts;
549 struct sk_buff *packet;
550 struct sw_flow *flow;
d98612b8 551 struct sw_flow_actions *sf_acts;
ccb1352e 552 struct datapath *dp;
83c8df26 553 struct vport *input_vport;
7f8a436e 554 u16 mru = 0;
bd1903b7 555 u64 hash;
ccb1352e
JG		 556 int len;
557 int err;
1ba39804 558 bool log = !a[OVS_PACKET_ATTR_PROBE];
ccb1352e
JG		 559
560 err = -EINVAL;
561 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
dded45fc 562 !a[OVS_PACKET_ATTR_ACTIONS])
ccb1352e
JG		 563 goto err;
564
565 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
566 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
567 err = -ENOMEM;
568 if (!packet)
569 goto err;
570 skb_reserve(packet, NET_IP_ALIGN);
571
32686a9d 572 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
ccb1352e 573
7f8a436e
JS		 574 /* Set packet's mru */
575 if (a[OVS_PACKET_ATTR_MRU]) {
576 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
577 packet->ignore_df = 1;
578 }
579 OVS_CB(packet)->mru = mru;
580
bd1903b7
TZ		 581 if (a[OVS_PACKET_ATTR_HASH]) {
582 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
583
584 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
585 !!(hash & OVS_PACKET_HASH_SW_BIT),
586 !!(hash & OVS_PACKET_HASH_L4_BIT));
587 }
588
ccb1352e 589 /* Build an sw_flow for sending this packet. */
23dabf88 590 flow = ovs_flow_alloc();
ccb1352e
JG		 591 err = PTR_ERR(flow);
592 if (IS_ERR(flow))
593 goto err_kfree_skb;
594
c2ac6673
JS		 595 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
596 packet, &flow->key, log);
ccb1352e
JG		 597 if (err)
598 goto err_flow_free;
599
7f8a436e 600 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
05da5898 601 &flow->key, &acts, log);
74f84a57
PS		 602 if (err)
603 goto err_flow_free;
ccb1352e 604
f5796684 605 rcu_assign_pointer(flow->sf_acts, acts);
ccb1352e 606 packet->priority = flow->key.phy.priority;
39c7caeb 607 packet->mark = flow->key.phy.skb_mark;
ccb1352e
JG		 608
609 rcu_read_lock();
7f8a436e 610 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
ccb1352e
JG		 611 err = -ENODEV;
612 if (!dp)
613 goto err_unlock;
614
83c8df26
PS		 615 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
616 if (!input_vport)
617 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
618
619 if (!input_vport)
620 goto err_unlock;
621
7f8a436e 622 packet->dev = input_vport->dev;
83c8df26 623 OVS_CB(packet)->input_vport = input_vport;
d98612b8 624 sf_acts = rcu_dereference(flow->sf_acts);
83c8df26 625
ccb1352e 626 local_bh_disable();
d98612b8 627 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
ccb1352e
JG		 628 local_bh_enable();
629 rcu_read_unlock();
630
03f0d916 631 ovs_flow_free(flow, false);
ccb1352e
JG		 632 return err;
633
634err_unlock:
635 rcu_read_unlock();
636err_flow_free:
03f0d916 637 ovs_flow_free(flow, false);
ccb1352e
JG		 638err_kfree_skb:
639 kfree_skb(packet);
640err:
641 return err;
642}
643
644static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
dded45fc 645 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
ccb1352e
JG		 646 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
647 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
1ba39804 648 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
7f8a436e 649 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
ccb1352e
JG		 650};
651
4534de83 652static const struct genl_ops dp_packet_genl_ops[] = {
ccb1352e 653 { .cmd = OVS_PACKET_CMD_EXECUTE,
ef6243ac 654 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 655 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 656 .doit = ovs_packet_cmd_execute
657 }
658};
659
56989f6d 660static struct genl_family dp_packet_genl_family __ro_after_init = {
0c200ef9
PS		 661 .hdrsize = sizeof(struct ovs_header),
662 .name = OVS_PACKET_FAMILY,
663 .version = OVS_PACKET_VERSION,
664 .maxattr = OVS_PACKET_ATTR_MAX,
3b0f31f2 665 .policy = packet_policy,
0c200ef9
PS		 666 .netnsok = true,
667 .parallel_ops = true,
668 .ops = dp_packet_genl_ops,
669 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
489111e5 670 .module = THIS_MODULE,
0c200ef9
PS		 671};
672
12eb18f7 673static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
1bd7116f 674 struct ovs_dp_megaflow_stats *mega_stats)
ccb1352e
JG		 675{
676 int i;
ccb1352e 677
1bd7116f
AZ		 678 memset(mega_stats, 0, sizeof(*mega_stats));
679
b637e498 680 stats->n_flows = ovs_flow_tbl_count(&dp->table);
1bd7116f 681 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
ccb1352e
JG		 682
683 stats->n_hit = stats->n_missed = stats->n_lost = 0;
1bd7116f 684
ccb1352e
JG		 685 for_each_possible_cpu(i) {
686 const struct dp_stats_percpu *percpu_stats;
687 struct dp_stats_percpu local_stats;
688 unsigned int start;
689
690 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
691
692 do {
57a7744e 693 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
ccb1352e 694 local_stats = *percpu_stats;
57a7744e 695 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
ccb1352e
JG		 696
697 stats->n_hit += local_stats.n_hit;
698 stats->n_missed += local_stats.n_missed;
699 stats->n_lost += local_stats.n_lost;
1bd7116f 700 mega_stats->n_mask_hit += local_stats.n_mask_hit;
ccb1352e
JG		 701 }
702}
703
74ed7ab9
JS		 704static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
705{
706 return ovs_identifier_is_ufid(sfid) &&
707 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
708}
709
710static bool should_fill_mask(uint32_t ufid_flags)
711{
712 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
713}
714
715static bool should_fill_actions(uint32_t ufid_flags)
c3ff8cfe 716{
74ed7ab9
JS		 717 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
718}
719
720static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
721 const struct sw_flow_id *sfid,
722 uint32_t ufid_flags)
723{
724 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
725
726 /* OVS_FLOW_ATTR_UFID */
727 if (sfid && ovs_identifier_is_ufid(sfid))
728 len += nla_total_size(sfid->ufid_len);
729
730 /* OVS_FLOW_ATTR_KEY */
731 if (!sfid || should_fill_key(sfid, ufid_flags))
732 len += nla_total_size(ovs_key_attr_size());
733
734 /* OVS_FLOW_ATTR_MASK */
735 if (should_fill_mask(ufid_flags))
736 len += nla_total_size(ovs_key_attr_size());
737
738 /* OVS_FLOW_ATTR_ACTIONS */
739 if (should_fill_actions(ufid_flags))
8e2fed1c 740 len += nla_total_size(acts->orig_len);
74ed7ab9
JS		 741
742 return len
66c7a5ee 743 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
c3ff8cfe 744 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
66c7a5ee 745 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
c3ff8cfe
TG		 746}
747
ca7105f2
JS		 748/* Called with ovs_mutex or RCU read lock. */
749static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
750 struct sk_buff *skb)
751{
752 struct ovs_flow_stats stats;
753 __be16 tcp_flags;
754 unsigned long used;
ccb1352e 755
e298e505 756 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
0e9796b4 757
028d6a67 758 if (used &&
0238b720
ND		 759 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
760 OVS_FLOW_ATTR_PAD))
ca7105f2 761 return -EMSGSIZE;
ccb1352e 762
028d6a67 763 if (stats.n_packets &&
66c7a5ee
ND		 764 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
765 sizeof(struct ovs_flow_stats), &stats,
766 OVS_FLOW_ATTR_PAD))
ca7105f2 767 return -EMSGSIZE;
ccb1352e 768
e298e505
PS		 769 if ((u8)ntohs(tcp_flags) &&
770 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
ca7105f2
JS		 771 return -EMSGSIZE;
772
773 return 0;
774}
775
776/* Called with ovs_mutex or RCU read lock. */
777static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
778 struct sk_buff *skb, int skb_orig_len)
779{
780 struct nlattr *start;
781 int err;
ccb1352e
JG		 782
783 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
784 * this is the first flow to be dumped into 'skb'. This is unusual for
785 * Netlink but individual action lists can be longer than
786 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
787 * The userspace caller can always fetch the actions separately if it
788 * really wants them. (Most userspace callers in fact don't care.)
789 *
790 * This can only fail for dump operations because the skb is always
791 * properly sized for single flows.
792 */
ae0be8de 793 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
74f84a57 794 if (start) {
d57170b1
PS		 795 const struct sw_flow_actions *sf_acts;
796
663efa36 797 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
e6445719
PS		 798 err = ovs_nla_put_actions(sf_acts->actions,
799 sf_acts->actions_len, skb);
0e9796b4 800
74f84a57
PS		 801 if (!err)
802 nla_nest_end(skb, start);
803 else {
804 if (skb_orig_len)
ca7105f2 805 return err;
74f84a57
PS		 806
807 nla_nest_cancel(skb, start);
808 }
ca7105f2
JS		 809 } else if (skb_orig_len) {
810 return -EMSGSIZE;
811 }
812
813 return 0;
814}
815
816/* Called with ovs_mutex or RCU read lock. */
817static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
818 struct sk_buff *skb, u32 portid,
74ed7ab9 819 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
ca7105f2
JS		 820{
821 const int skb_orig_len = skb->len;
822 struct ovs_header *ovs_header;
823 int err;
824
825 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
826 flags, cmd);
827 if (!ovs_header)
828 return -EMSGSIZE;
829
830 ovs_header->dp_ifindex = dp_ifindex;
831
74ed7ab9 832 err = ovs_nla_put_identifier(flow, skb);
5b4237bb
JS		 833 if (err)
834 goto error;
835
74ed7ab9
JS		 836 if (should_fill_key(&flow->id, ufid_flags)) {
837 err = ovs_nla_put_masked_key(flow, skb);
838 if (err)
839 goto error;
840 }
841
842 if (should_fill_mask(ufid_flags)) {
843 err = ovs_nla_put_mask(flow, skb);
844 if (err)
845 goto error;
846 }
ca7105f2
JS		 847
848 err = ovs_flow_cmd_fill_stats(flow, skb);
849 if (err)
850 goto error;
851
74ed7ab9
JS		 852 if (should_fill_actions(ufid_flags)) {
853 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
854 if (err)
855 goto error;
856 }
ccb1352e 857
053c095a
JB		 858 genlmsg_end(skb, ovs_header);
859 return 0;
ccb1352e 860
ccb1352e
JG		 861error:
862 genlmsg_cancel(skb, ovs_header);
863 return err;
864}
865
0e9796b4
JR		 866/* May not be called with RCU read lock. */
867static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
74ed7ab9 868 const struct sw_flow_id *sfid,
fb5d1e9e 869 struct genl_info *info,
74ed7ab9
JS		 870 bool always,
871 uint32_t ufid_flags)
ccb1352e 872{
fb5d1e9e 873 struct sk_buff *skb;
74ed7ab9 874 size_t len;
ccb1352e 875
9b67aa4a 876 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
fb5d1e9e
JR		 877 return NULL;
878
74ed7ab9 879 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
551ddc05 880 skb = genlmsg_new(len, GFP_KERNEL);
fb5d1e9e
JR		 881 if (!skb)
882 return ERR_PTR(-ENOMEM);
883
884 return skb;
ccb1352e
JG		 885}
886
0e9796b4
JR		 887/* Called with ovs_mutex. */
888static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
889 int dp_ifindex,
890 struct genl_info *info, u8 cmd,
74ed7ab9 891 bool always, u32 ufid_flags)
ccb1352e
JG		 892{
893 struct sk_buff *skb;
894 int retval;
895
74ed7ab9
JS		 896 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
897 &flow->id, info, always, ufid_flags);
d0e992aa 898 if (IS_ERR_OR_NULL(skb))
fb5d1e9e 899 return skb;
ccb1352e 900
0e9796b4
JR		 901 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
902 info->snd_portid, info->snd_seq, 0,
74ed7ab9 903 cmd, ufid_flags);
ccb1352e
JG		 904 BUG_ON(retval < 0);
905 return skb;
906}
907
37bdc87b 908static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
ccb1352e 909{
7f8a436e 910 struct net *net = sock_net(skb->sk);
ccb1352e
JG		 911 struct nlattr **a = info->attrs;
912 struct ovs_header *ovs_header = info->userhdr;
74ed7ab9 913 struct sw_flow *flow = NULL, *new_flow;
03f0d916 914 struct sw_flow_mask mask;
ccb1352e
JG		 915 struct sk_buff *reply;
916 struct datapath *dp;
37bdc87b 917 struct sw_flow_actions *acts;
03f0d916 918 struct sw_flow_match match;
74ed7ab9 919 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
ccb1352e 920 int error;
05da5898 921 bool log = !a[OVS_FLOW_ATTR_PROBE];
ccb1352e 922
893f139b 923 /* Must have key and actions. */
ccb1352e 924 error = -EINVAL;
426cda5c 925 if (!a[OVS_FLOW_ATTR_KEY]) {
05da5898 926 OVS_NLERR(log, "Flow key attr not present in new flow.");
ccb1352e 927 goto error;
426cda5c
JG		 928 }
929 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
05da5898 930 OVS_NLERR(log, "Flow actions attr not present in new flow.");
893f139b 931 goto error;
426cda5c 932 }
03f0d916 933
893f139b
JR		 934 /* Most of the time we need to allocate a new flow, do it before
935 * locking.
936 */
937 new_flow = ovs_flow_alloc();
938 if (IS_ERR(new_flow)) {
939 error = PTR_ERR(new_flow);
940 goto error;
941 }
942
943 /* Extract key. */
2279994d 944 ovs_match_init(&match, &new_flow->key, false, &mask);
c2ac6673 945 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
05da5898 946 a[OVS_FLOW_ATTR_MASK], log);
ccb1352e 947 if (error)
893f139b 948 goto err_kfree_flow;
ccb1352e 949
74ed7ab9
JS		 950 /* Extract flow identifier. */
951 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
190aa3e7 952 &new_flow->key, log);
74ed7ab9
JS		 953 if (error)
954 goto err_kfree_flow;
74f84a57 955
190aa3e7 956 /* unmasked key is needed to match when ufid is not used. */
957 if (ovs_identifier_is_key(&new_flow->id))
958 match.key = new_flow->id.unmasked_key;
959
960 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
961
893f139b 962 /* Validate actions. */
7f8a436e
JS		 963 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
964 &new_flow->key, &acts, log);
37bdc87b 965 if (error) {
05da5898 966 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
2fdb957d 967 goto err_kfree_flow;
893f139b
JR		 968 }
969
74ed7ab9
JS		 970 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
971 ufid_flags);
893f139b
JR		 972 if (IS_ERR(reply)) {
973 error = PTR_ERR(reply);
974 goto err_kfree_acts;
ccb1352e
JG		 975 }
976
8e4e1713 977 ovs_lock();
7f8a436e 978 dp = get_dp(net, ovs_header->dp_ifindex);
893f139b
JR		 979 if (unlikely(!dp)) {
980 error = -ENODEV;
8e4e1713 981 goto err_unlock_ovs;
893f139b 982 }
74ed7ab9 983
03f0d916 984 /* Check if this is a duplicate flow */
74ed7ab9
JS		 985 if (ovs_identifier_is_ufid(&new_flow->id))
986 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
987 if (!flow)
190aa3e7 988 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
893f139b
JR		 989 if (likely(!flow)) {
990 rcu_assign_pointer(new_flow->sf_acts, acts);
ccb1352e
JG		 991
992 /* Put flow in bucket. */
893f139b
JR		 993 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
994 if (unlikely(error)) {
618ed0c8 995 acts = NULL;
893f139b
JR		 996 goto err_unlock_ovs;
997 }
998
999 if (unlikely(reply)) {
1000 error = ovs_flow_cmd_fill_info(new_flow,
1001 ovs_header->dp_ifindex,
1002 reply, info->snd_portid,
1003 info->snd_seq, 0,
74ed7ab9
JS		 1004 OVS_FLOW_CMD_NEW,
1005 ufid_flags);
893f139b 1006 BUG_ON(error < 0);
618ed0c8 1007 }
893f139b 1008 ovs_unlock();
ccb1352e 1009 } else {
37bdc87b
JR		 1010 struct sw_flow_actions *old_acts;
1011
ccb1352e
JG		 1012 /* Bail out if we're not allowed to modify an existing flow.
1013 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1014 * because Generic Netlink treats the latter as a dump
1015 * request. We also accept NLM_F_EXCL in case that bug ever
1016 * gets fixed.
1017 */
893f139b
JR		 1018 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1019 | NLM_F_EXCL))) {
1020 error = -EEXIST;
8e4e1713 1021 goto err_unlock_ovs;
893f139b 1022 }
74ed7ab9
JS		 1023 /* The flow identifier has to be the same for flow updates.
1024 * Look for any overlapping flow.
1025 */
1026 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1027 if (ovs_identifier_is_key(&flow->id))
1028 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1029 &match);
1030 else /* UFID matches but key is different */
1031 flow = NULL;
4a46b24e
AW		 1032 if (!flow) {
1033 error = -ENOENT;
1034 goto err_unlock_ovs;
1035 }
893f139b 1036 }
37bdc87b
JR		 1037 /* Update actions. */
1038 old_acts = ovsl_dereference(flow->sf_acts);
1039 rcu_assign_pointer(flow->sf_acts, acts);
37bdc87b 1040
893f139b
JR		 1041 if (unlikely(reply)) {
1042 error = ovs_flow_cmd_fill_info(flow,
1043 ovs_header->dp_ifindex,
1044 reply, info->snd_portid,
1045 info->snd_seq, 0,
74ed7ab9
JS		 1046 OVS_FLOW_CMD_NEW,
1047 ufid_flags);
893f139b
JR		 1048 BUG_ON(error < 0);
1049 }
1050 ovs_unlock();
37bdc87b 1051
34ae932a 1052 ovs_nla_free_flow_actions_rcu(old_acts);
893f139b 1053 ovs_flow_free(new_flow, false);
37bdc87b 1054 }
893f139b
JR		 1055
1056 if (reply)
1057 ovs_notify(&dp_flow_genl_family, reply, info);
37bdc87b
JR		 1058 return 0;
1059
37bdc87b
JR		 1060err_unlock_ovs:
1061 ovs_unlock();
893f139b
JR		 1062 kfree_skb(reply);
1063err_kfree_acts:
34ae932a 1064 ovs_nla_free_flow_actions(acts);
893f139b
JR		 1065err_kfree_flow:
1066 ovs_flow_free(new_flow, false);
37bdc87b
JR		 1067error:
1068 return error;
1069}
ccb1352e 1070
2fdb957d 1071/* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
26063790 1072static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
7f8a436e 1073 const struct nlattr *a,
6b205b2c 1074 const struct sw_flow_key *key,
05da5898
JR		 1075 const struct sw_flow_mask *mask,
1076 bool log)
6b205b2c
JG		 1077{
1078 struct sw_flow_actions *acts;
1079 struct sw_flow_key masked_key;
1080 int error;
1081
ae5f2fb1 1082 ovs_flow_mask_key(&masked_key, key, true, mask);
7f8a436e 1083 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
6b205b2c 1084 if (error) {
05da5898
JR		 1085 OVS_NLERR(log,
1086 "Actions may not be safe on all matching packets");
6b205b2c
JG		 1087 return ERR_PTR(error);
1088 }
1089
1090 return acts;
1091}
1092
9cc9a5cb
TZ		 1093/* Factor out match-init and action-copy to avoid
1094 * "Wframe-larger-than=1024" warning. Because mask is only
1095 * used to get actions, we new a function to save some
1096 * stack space.
1097 *
1098 * If there are not key and action attrs, we return 0
1099 * directly. In the case, the caller will also not use the
1100 * match as before. If there is action attr, we try to get
1101 * actions and save them to *acts. Before returning from
1102 * the function, we reset the match->mask pointer. Because
1103 * we should not to return match object with dangling reference
1104 * to mask.
1105 * */
26063790
AB		 1106static noinline_for_stack int
1107ovs_nla_init_match_and_action(struct net *net,
1108 struct sw_flow_match *match,
1109 struct sw_flow_key *key,
1110 struct nlattr **a,
1111 struct sw_flow_actions **acts,
1112 bool log)
9cc9a5cb
TZ		 1113{
1114 struct sw_flow_mask mask;
1115 int error = 0;
1116
1117 if (a[OVS_FLOW_ATTR_KEY]) {
1118 ovs_match_init(match, key, true, &mask);
1119 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1120 a[OVS_FLOW_ATTR_MASK], log);
1121 if (error)
1122 goto error;
1123 }
1124
1125 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1126 if (!a[OVS_FLOW_ATTR_KEY]) {
1127 OVS_NLERR(log,
1128 "Flow key attribute not present in set flow.");
5829e62a
CJ		 1129 error = -EINVAL;
1130 goto error;
9cc9a5cb
TZ		 1131 }
1132
1133 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1134 &mask, log);
1135 if (IS_ERR(*acts)) {
1136 error = PTR_ERR(*acts);
1137 goto error;
1138 }
1139 }
1140
1141 /* On success, error is 0. */
1142error:
1143 match->mask = NULL;
1144 return error;
1145}
1146
37bdc87b
JR		 1147static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1148{
7f8a436e 1149 struct net *net = sock_net(skb->sk);
37bdc87b
JR		 1150 struct nlattr **a = info->attrs;
1151 struct ovs_header *ovs_header = info->userhdr;
6b205b2c 1152 struct sw_flow_key key;
37bdc87b 1153 struct sw_flow *flow;
37bdc87b
JR		 1154 struct sk_buff *reply = NULL;
1155 struct datapath *dp;
893f139b 1156 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
37bdc87b 1157 struct sw_flow_match match;
74ed7ab9
JS		 1158 struct sw_flow_id sfid;
1159 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
6f15cdbf 1160 int error = 0;
05da5898 1161 bool log = !a[OVS_FLOW_ATTR_PROBE];
74ed7ab9 1162 bool ufid_present;
37bdc87b 1163
74ed7ab9 1164 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
9cc9a5cb 1165 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
6f15cdbf
SG		 1166 OVS_NLERR(log,
1167 "Flow set message rejected, Key attribute missing.");
9cc9a5cb 1168 return -EINVAL;
6f15cdbf 1169 }
9cc9a5cb
TZ		 1170
1171 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1172 &acts, log);
37bdc87b
JR		 1173 if (error)
1174 goto error;
1175
9cc9a5cb 1176 if (acts) {
2fdb957d 1177 /* Can allocate before locking if have acts. */
74ed7ab9
JS		 1178 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1179 ufid_flags);
893f139b
JR		 1180 if (IS_ERR(reply)) {
1181 error = PTR_ERR(reply);
1182 goto err_kfree_acts;
be52c9e9 1183 }
37bdc87b 1184 }
0e9796b4 1185
37bdc87b 1186 ovs_lock();
7f8a436e 1187 dp = get_dp(net, ovs_header->dp_ifindex);
893f139b
JR		 1188 if (unlikely(!dp)) {
1189 error = -ENODEV;
37bdc87b 1190 goto err_unlock_ovs;
893f139b 1191 }
37bdc87b 1192 /* Check that the flow exists. */
74ed7ab9
JS		 1193 if (ufid_present)
1194 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1195 else
1196 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
893f139b
JR		 1197 if (unlikely(!flow)) {
1198 error = -ENOENT;
37bdc87b 1199 goto err_unlock_ovs;
893f139b 1200 }
4a46b24e 1201
37bdc87b 1202 /* Update actions, if present. */
893f139b 1203 if (likely(acts)) {
37bdc87b
JR		 1204 old_acts = ovsl_dereference(flow->sf_acts);
1205 rcu_assign_pointer(flow->sf_acts, acts);
893f139b
JR		 1206
1207 if (unlikely(reply)) {
1208 error = ovs_flow_cmd_fill_info(flow,
1209 ovs_header->dp_ifindex,
1210 reply, info->snd_portid,
1211 info->snd_seq, 0,
804fe108 1212 OVS_FLOW_CMD_SET,
74ed7ab9 1213 ufid_flags);
893f139b
JR		 1214 BUG_ON(error < 0);
1215 }
1216 } else {
1217 /* Could not alloc without acts before locking. */
1218 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
804fe108 1219 info, OVS_FLOW_CMD_SET, false,
74ed7ab9
JS		 1220 ufid_flags);
1221
b5ffe634 1222 if (IS_ERR(reply)) {
893f139b
JR		 1223 error = PTR_ERR(reply);
1224 goto err_unlock_ovs;
1225 }
ccb1352e 1226 }
37bdc87b 1227
37bdc87b
JR		 1228 /* Clear stats. */
1229 if (a[OVS_FLOW_ATTR_CLEAR])
1230 ovs_flow_stats_clear(flow);
8e4e1713 1231 ovs_unlock();
ccb1352e 1232
893f139b
JR		 1233 if (reply)
1234 ovs_notify(&dp_flow_genl_family, reply, info);
1235 if (old_acts)
34ae932a 1236 ovs_nla_free_flow_actions_rcu(old_acts);
fb5d1e9e 1237
ccb1352e
JG		 1238 return 0;
1239
8e4e1713
PS		 1240err_unlock_ovs:
1241 ovs_unlock();
893f139b
JR		 1242 kfree_skb(reply);
1243err_kfree_acts:
34ae932a 1244 ovs_nla_free_flow_actions(acts);
ccb1352e
JG		 1245error:
1246 return error;
1247}
1248
1249static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1250{
1251 struct nlattr **a = info->attrs;
1252 struct ovs_header *ovs_header = info->userhdr;
c2ac6673 1253 struct net *net = sock_net(skb->sk);
ccb1352e
JG		 1254 struct sw_flow_key key;
1255 struct sk_buff *reply;
1256 struct sw_flow *flow;
1257 struct datapath *dp;
03f0d916 1258 struct sw_flow_match match;
74ed7ab9
JS		 1259 struct sw_flow_id ufid;
1260 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1261 int err = 0;
05da5898 1262 bool log = !a[OVS_FLOW_ATTR_PROBE];
74ed7ab9 1263 bool ufid_present;
ccb1352e 1264
74ed7ab9
JS		 1265 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1266 if (a[OVS_FLOW_ATTR_KEY]) {
2279994d 1267 ovs_match_init(&match, &key, true, NULL);
c2ac6673 1268 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
74ed7ab9
JS		 1269 log);
1270 } else if (!ufid_present) {
05da5898
JR		 1271 OVS_NLERR(log,
1272 "Flow get message rejected, Key attribute missing.");
74ed7ab9 1273 err = -EINVAL;
03f0d916 1274 }
ccb1352e
JG		 1275 if (err)
1276 return err;
1277
8e4e1713 1278 ovs_lock();
46df7b81 1279 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
8e4e1713
PS		 1280 if (!dp) {
1281 err = -ENODEV;
1282 goto unlock;
1283 }
ccb1352e 1284
74ed7ab9
JS		 1285 if (ufid_present)
1286 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1287 else
1288 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
4a46b24e 1289 if (!flow) {
8e4e1713
PS		 1290 err = -ENOENT;
1291 goto unlock;
1292 }
ccb1352e 1293
0e9796b4 1294 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
804fe108 1295 OVS_FLOW_CMD_GET, true, ufid_flags);
8e4e1713
PS		 1296 if (IS_ERR(reply)) {
1297 err = PTR_ERR(reply);
1298 goto unlock;
1299 }
ccb1352e 1300
8e4e1713 1301 ovs_unlock();
ccb1352e 1302 return genlmsg_reply(reply, info);
8e4e1713
PS		 1303unlock:
1304 ovs_unlock();
1305 return err;
ccb1352e
JG		 1306}
1307
1308static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1309{
1310 struct nlattr **a = info->attrs;
1311 struct ovs_header *ovs_header = info->userhdr;
c2ac6673 1312 struct net *net = sock_net(skb->sk);
ccb1352e
JG		 1313 struct sw_flow_key key;
1314 struct sk_buff *reply;
74ed7ab9 1315 struct sw_flow *flow = NULL;
ccb1352e 1316 struct datapath *dp;
03f0d916 1317 struct sw_flow_match match;
74ed7ab9
JS		 1318 struct sw_flow_id ufid;
1319 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
ccb1352e 1320 int err;
05da5898 1321 bool log = !a[OVS_FLOW_ATTR_PROBE];
74ed7ab9 1322 bool ufid_present;
ccb1352e 1323
74ed7ab9
JS		 1324 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1325 if (a[OVS_FLOW_ATTR_KEY]) {
2279994d 1326 ovs_match_init(&match, &key, true, NULL);
c2ac6673
JS		 1327 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1328 NULL, log);
aed06778
JR		 1329 if (unlikely(err))
1330 return err;
1331 }
1332
8e4e1713 1333 ovs_lock();
46df7b81 1334 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
aed06778 1335 if (unlikely(!dp)) {
8e4e1713
PS		 1336 err = -ENODEV;
1337 goto unlock;
1338 }
46df7b81 1339
74ed7ab9 1340 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
b637e498 1341 err = ovs_flow_tbl_flush(&dp->table);
8e4e1713
PS		 1342 goto unlock;
1343 }
03f0d916 1344
74ed7ab9
JS		 1345 if (ufid_present)
1346 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1347 else
1348 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
4a46b24e 1349 if (unlikely(!flow)) {
8e4e1713
PS		 1350 err = -ENOENT;
1351 goto unlock;
1352 }
ccb1352e 1353
b637e498 1354 ovs_flow_tbl_remove(&dp->table, flow);
aed06778 1355 ovs_unlock();
ccb1352e 1356
aed06778 1357 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
74ed7ab9 1358 &flow->id, info, false, ufid_flags);
aed06778 1359 if (likely(reply)) {
b90f5aa4 1360 if (!IS_ERR(reply)) {
aed06778
JR		 1361 rcu_read_lock(); /*To keep RCU checker happy. */
1362 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1363 reply, info->snd_portid,
1364 info->snd_seq, 0,
74ed7ab9
JS		 1365 OVS_FLOW_CMD_DEL,
1366 ufid_flags);
aed06778
JR		 1367 rcu_read_unlock();
1368 BUG_ON(err < 0);
1369
1370 ovs_notify(&dp_flow_genl_family, reply, info);
1371 } else {
1372 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1373 }
fb5d1e9e 1374 }
ccb1352e 1375
aed06778 1376 ovs_flow_free(flow, true);
ccb1352e 1377 return 0;
8e4e1713
PS		 1378unlock:
1379 ovs_unlock();
1380 return err;
ccb1352e
JG		 1381}
1382
1383static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1384{
74ed7ab9 1385 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
ccb1352e 1386 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
b637e498 1387 struct table_instance *ti;
ccb1352e 1388 struct datapath *dp;
74ed7ab9
JS		 1389 u32 ufid_flags;
1390 int err;
1391
8cb08174
JB		 1392 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1393 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
74ed7ab9
JS		 1394 if (err)
1395 return err;
1396 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
ccb1352e 1397
d57170b1 1398 rcu_read_lock();
cc3a5ae6 1399 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
8e4e1713 1400 if (!dp) {
d57170b1 1401 rcu_read_unlock();
ccb1352e 1402 return -ENODEV;
8e4e1713 1403 }
ccb1352e 1404
b637e498 1405 ti = rcu_dereference(dp->table.ti);
ccb1352e
JG		 1406 for (;;) {
1407 struct sw_flow *flow;
1408 u32 bucket, obj;
1409
1410 bucket = cb->args[0];
1411 obj = cb->args[1];
b637e498 1412 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
ccb1352e
JG		 1413 if (!flow)
1414 break;
1415
0e9796b4 1416 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
15e47304 1417 NETLINK_CB(cb->skb).portid,
ccb1352e 1418 cb->nlh->nlmsg_seq, NLM_F_MULTI,
804fe108 1419 OVS_FLOW_CMD_GET, ufid_flags) < 0)
ccb1352e
JG		 1420 break;
1421
1422 cb->args[0] = bucket;
1423 cb->args[1] = obj;
1424 }
d57170b1 1425 rcu_read_unlock();
ccb1352e
JG		 1426 return skb->len;
1427}
1428
0c200ef9
PS		 1429static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1430 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
05da5898 1431 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
0c200ef9
PS		 1432 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1433 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
05da5898 1434 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
74ed7ab9
JS		 1435 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1436 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
0c200ef9
PS		 1437};
1438
48e48a70 1439static const struct genl_ops dp_flow_genl_ops[] = {
ccb1352e 1440 { .cmd = OVS_FLOW_CMD_NEW,
ef6243ac 1441 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1442 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
37bdc87b 1443 .doit = ovs_flow_cmd_new
ccb1352e
JG		 1444 },
1445 { .cmd = OVS_FLOW_CMD_DEL,
ef6243ac 1446 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1447 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 1448 .doit = ovs_flow_cmd_del
1449 },
1450 { .cmd = OVS_FLOW_CMD_GET,
ef6243ac 1451 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
ccb1352e 1452 .flags = 0, /* OK for unprivileged users. */
ccb1352e
JG		 1453 .doit = ovs_flow_cmd_get,
1454 .dumpit = ovs_flow_cmd_dump
1455 },
1456 { .cmd = OVS_FLOW_CMD_SET,
ef6243ac 1457 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1458 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
37bdc87b 1459 .doit = ovs_flow_cmd_set,
ccb1352e
JG		 1460 },
1461};
1462
56989f6d 1463static struct genl_family dp_flow_genl_family __ro_after_init = {
ccb1352e 1464 .hdrsize = sizeof(struct ovs_header),
0c200ef9
PS		 1465 .name = OVS_FLOW_FAMILY,
1466 .version = OVS_FLOW_VERSION,
1467 .maxattr = OVS_FLOW_ATTR_MAX,
3b0f31f2 1468 .policy = flow_policy,
3a4e0d6a
PS		 1469 .netnsok = true,
1470 .parallel_ops = true,
0c200ef9
PS		 1471 .ops = dp_flow_genl_ops,
1472 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1473 .mcgrps = &ovs_dp_flow_multicast_group,
1474 .n_mcgrps = 1,
489111e5 1475 .module = THIS_MODULE,
ccb1352e
JG		 1476};
1477
c3ff8cfe
TG		 1478static size_t ovs_dp_cmd_msg_size(void)
1479{
1480 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1481
1482 msgsize += nla_total_size(IFNAMSIZ);
66c7a5ee
ND		 1483 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1484 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
45fb9c35 1485 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
c3ff8cfe
TG		 1486
1487 return msgsize;
1488}
1489
8ec609d8 1490/* Called with ovs_mutex. */
ccb1352e 1491static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
15e47304 1492 u32 portid, u32 seq, u32 flags, u8 cmd)
ccb1352e
JG		 1493{
1494 struct ovs_header *ovs_header;
1495 struct ovs_dp_stats dp_stats;
1bd7116f 1496 struct ovs_dp_megaflow_stats dp_megaflow_stats;
ccb1352e
JG		 1497 int err;
1498
15e47304 1499 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
ccb1352e
JG		 1500 flags, cmd);
1501 if (!ovs_header)
1502 goto error;
1503
1504 ovs_header->dp_ifindex = get_dpifindex(dp);
1505
ccb1352e 1506 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
ccb1352e
JG		 1507 if (err)
1508 goto nla_put_failure;
1509
1bd7116f 1510 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
66c7a5ee
ND		 1511 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1512 &dp_stats, OVS_DP_ATTR_PAD))
1bd7116f
AZ		 1513 goto nla_put_failure;
1514
66c7a5ee
ND		 1515 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1516 sizeof(struct ovs_dp_megaflow_stats),
1517 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
028d6a67 1518 goto nla_put_failure;
ccb1352e 1519
43d4be9c
TG		 1520 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1521 goto nla_put_failure;
1522
053c095a
JB		 1523 genlmsg_end(skb, ovs_header);
1524 return 0;
ccb1352e
JG		 1525
1526nla_put_failure:
1527 genlmsg_cancel(skb, ovs_header);
1528error:
1529 return -EMSGSIZE;
1530}
1531
263ea090 1532static struct sk_buff *ovs_dp_cmd_alloc_info(void)
ccb1352e 1533{
551ddc05 1534 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
ccb1352e
JG		 1535}
1536
bb6f9a70 1537/* Called with rcu_read_lock or ovs_mutex. */
46df7b81 1538static struct datapath *lookup_datapath(struct net *net,
12eb18f7 1539 const struct ovs_header *ovs_header,
ccb1352e
JG		 1540 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1541{
1542 struct datapath *dp;
1543
1544 if (!a[OVS_DP_ATTR_NAME])
46df7b81 1545 dp = get_dp(net, ovs_header->dp_ifindex);
ccb1352e
JG		 1546 else {
1547 struct vport *vport;
1548
46df7b81 1549 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
ccb1352e 1550 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
ccb1352e
JG		 1551 }
1552 return dp ? dp : ERR_PTR(-ENODEV);
1553}
1554
44da5ae5
TG		 1555static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1556{
1557 struct datapath *dp;
1558
1559 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
3c7eacfc 1560 if (IS_ERR(dp))
44da5ae5
TG		 1561 return;
1562
1563 WARN(dp->user_features, "Dropping previously announced user features\n");
1564 dp->user_features = 0;
1565}
1566
95a7233c
PB		 1567DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1568
1569static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
43d4be9c 1570{
95a7233c
PB		 1571 u32 user_features = 0;
1572
1573 if (a[OVS_DP_ATTR_USER_FEATURES]) {
1574 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1575
1576 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1577 OVS_DP_F_UNALIGNED |
1578 OVS_DP_F_TC_RECIRC_SHARING))
1579 return -EOPNOTSUPP;
1580
1581#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1582 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1583 return -EOPNOTSUPP;
1584#endif
1585 }
1586
1587 dp->user_features = user_features;
1588
1589 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1590 static_branch_enable(&tc_recirc_sharing_support);
1591 else
1592 static_branch_disable(&tc_recirc_sharing_support);
1593
1594 return 0;
43d4be9c
TG		 1595}
1596
eec62ead
TZ		 1597static int ovs_dp_stats_init(struct datapath *dp)
1598{
1599 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1600 if (!dp->stats_percpu)
1601 return -ENOMEM;
1602
1603 return 0;
1604}
1605
1606static int ovs_dp_vport_init(struct datapath *dp)
1607{
1608 int i;
1609
1610 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1611 sizeof(struct hlist_head),
1612 GFP_KERNEL);
1613 if (!dp->ports)
1614 return -ENOMEM;
1615
1616 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1617 INIT_HLIST_HEAD(&dp->ports[i]);
1618
1619 return 0;
1620}
1621
ccb1352e
JG		 1622static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1623{
1624 struct nlattr **a = info->attrs;
1625 struct vport_parms parms;
1626 struct sk_buff *reply;
1627 struct datapath *dp;
1628 struct vport *vport;
46df7b81 1629 struct ovs_net *ovs_net;
eec62ead 1630 int err;
ccb1352e
JG		 1631
1632 err = -EINVAL;
1633 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1634 goto err;
1635
263ea090 1636 reply = ovs_dp_cmd_alloc_info();
6093ae9a
JR		 1637 if (!reply)
1638 return -ENOMEM;
ccb1352e
JG		 1639
1640 err = -ENOMEM;
1641 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1642 if (dp == NULL)
eec62ead 1643 goto err_destroy_reply;
46df7b81 1644
efd7ef1c 1645 ovs_dp_set_net(dp, sock_net(skb->sk));
ccb1352e
JG		 1646
1647 /* Allocate table. */
b637e498
PS		 1648 err = ovs_flow_tbl_init(&dp->table);
1649 if (err)
eec62ead 1650 goto err_destroy_dp;
ccb1352e 1651
eec62ead
TZ		 1652 err = ovs_dp_stats_init(dp);
1653 if (err)
ccb1352e 1654 goto err_destroy_table;
ccb1352e 1655
eec62ead
TZ		 1656 err = ovs_dp_vport_init(dp);
1657 if (err)
1658 goto err_destroy_stats;
15eac2a7 1659
96fbc13d
AZ		 1660 err = ovs_meters_init(dp);
1661 if (err)
eec62ead 1662 goto err_destroy_ports;
96fbc13d 1663
ccb1352e
JG		 1664 /* Set up our datapath device. */
1665 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1666 parms.type = OVS_VPORT_TYPE_INTERNAL;
1667 parms.options = NULL;
1668 parms.dp = dp;
1669 parms.port_no = OVSP_LOCAL;
5cd667b0 1670 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
ccb1352e 1671
95a7233c
PB		 1672 err = ovs_dp_change(dp, a);
1673 if (err)
1674 goto err_destroy_meters;
43d4be9c 1675
6093ae9a
JR		 1676 /* So far only local changes have been made, now need the lock. */
1677 ovs_lock();
1678
ccb1352e
JG		 1679 vport = new_vport(&parms);
1680 if (IS_ERR(vport)) {
1681 err = PTR_ERR(vport);
1682 if (err == -EBUSY)
1683 err = -EEXIST;
1684
44da5ae5
TG		 1685 if (err == -EEXIST) {
1686 /* An outdated user space instance that does not understand
1687 * the concept of user_features has attempted to create a new
1688 * datapath and is likely to reuse it. Drop all user features.
1689 */
1690 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1691 ovs_dp_reset_user_features(skb, info);
1692 }
1693
4c76bf69 1694 ovs_unlock();
96fbc13d 1695 goto err_destroy_meters;
ccb1352e
JG		 1696 }
1697
6093ae9a
JR		 1698 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1699 info->snd_seq, 0, OVS_DP_CMD_NEW);
1700 BUG_ON(err < 0);
ccb1352e 1701
46df7b81 1702 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
59a35d60 1703 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
8e4e1713
PS		 1704
1705 ovs_unlock();
ccb1352e 1706
2a94fe48 1707 ovs_notify(&dp_datapath_genl_family, reply, info);
ccb1352e
JG		 1708 return 0;
1709
96fbc13d 1710err_destroy_meters:
96fbc13d 1711 ovs_meters_exit(dp);
eec62ead 1712err_destroy_ports:
15eac2a7 1713 kfree(dp->ports);
eec62ead 1714err_destroy_stats:
ccb1352e
JG		 1715 free_percpu(dp->stats_percpu);
1716err_destroy_table:
9b996e54 1717 ovs_flow_tbl_destroy(&dp->table);
eec62ead 1718err_destroy_dp:
ccb1352e 1719 kfree(dp);
eec62ead 1720err_destroy_reply:
6093ae9a 1721 kfree_skb(reply);
ccb1352e
JG		 1722err:
1723 return err;
1724}
1725
8e4e1713 1726/* Called with ovs_mutex. */
46df7b81 1727static void __dp_destroy(struct datapath *dp)
ccb1352e 1728{
15eac2a7 1729 int i;
ccb1352e 1730
15eac2a7
PS		 1731 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1732 struct vport *vport;
b67bfe0d 1733 struct hlist_node *n;
15eac2a7 1734
b67bfe0d 1735 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
15eac2a7
PS		 1736 if (vport->port_no != OVSP_LOCAL)
1737 ovs_dp_detach_port(vport);
1738 }
ccb1352e 1739
59a35d60 1740 list_del_rcu(&dp->list_node);
ccb1352e 1741
8e4e1713 1742 /* OVSP_LOCAL is datapath internal port. We need to make sure that
e80857cc 1743 * all ports in datapath are destroyed first before freeing datapath.
ccb1352e 1744 */
8e4e1713 1745 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
ccb1352e 1746
e80857cc 1747 /* RCU destroy the flow table */
ccb1352e 1748 call_rcu(&dp->rcu, destroy_dp_rcu);
46df7b81
PS		 1749}
1750
1751static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1752{
1753 struct sk_buff *reply;
1754 struct datapath *dp;
1755 int err;
1756
263ea090 1757 reply = ovs_dp_cmd_alloc_info();
6093ae9a
JR		 1758 if (!reply)
1759 return -ENOMEM;
1760
8e4e1713 1761 ovs_lock();
46df7b81
PS		 1762 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1763 err = PTR_ERR(dp);
1764 if (IS_ERR(dp))
6093ae9a 1765 goto err_unlock_free;
46df7b81 1766
6093ae9a
JR		 1767 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1768 info->snd_seq, 0, OVS_DP_CMD_DEL);
1769 BUG_ON(err < 0);
46df7b81
PS		 1770
1771 __dp_destroy(dp);
8e4e1713 1772 ovs_unlock();
ccb1352e 1773
2a94fe48 1774 ovs_notify(&dp_datapath_genl_family, reply, info);
ccb1352e
JG		 1775
1776 return 0;
6093ae9a
JR		 1777
1778err_unlock_free:
8e4e1713 1779 ovs_unlock();
6093ae9a 1780 kfree_skb(reply);
8e4e1713 1781 return err;
ccb1352e
JG		 1782}
1783
1784static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1785{
1786 struct sk_buff *reply;
1787 struct datapath *dp;
1788 int err;
1789
263ea090 1790 reply = ovs_dp_cmd_alloc_info();
6093ae9a
JR		 1791 if (!reply)
1792 return -ENOMEM;
1793
8e4e1713 1794 ovs_lock();
46df7b81 1795 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
8e4e1713 1796 err = PTR_ERR(dp);
ccb1352e 1797 if (IS_ERR(dp))
6093ae9a 1798 goto err_unlock_free;
ccb1352e 1799
95a7233c
PB		 1800 err = ovs_dp_change(dp, info->attrs);
1801 if (err)
1802 goto err_unlock_free;
43d4be9c 1803
6093ae9a 1804 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
804fe108 1805 info->snd_seq, 0, OVS_DP_CMD_SET);
6093ae9a 1806 BUG_ON(err < 0);
ccb1352e 1807
8e4e1713 1808 ovs_unlock();
2a94fe48 1809 ovs_notify(&dp_datapath_genl_family, reply, info);
ccb1352e
JG		 1810
1811 return 0;
6093ae9a
JR		 1812
1813err_unlock_free:
8e4e1713 1814 ovs_unlock();
6093ae9a 1815 kfree_skb(reply);
8e4e1713 1816 return err;
ccb1352e
JG		 1817}
1818
1819static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1820{
1821 struct sk_buff *reply;
1822 struct datapath *dp;
8e4e1713 1823 int err;
ccb1352e 1824
263ea090 1825 reply = ovs_dp_cmd_alloc_info();
6093ae9a
JR		 1826 if (!reply)
1827 return -ENOMEM;
1828
8ec609d8 1829 ovs_lock();
46df7b81 1830 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
8e4e1713
PS		 1831 if (IS_ERR(dp)) {
1832 err = PTR_ERR(dp);
6093ae9a 1833 goto err_unlock_free;
8e4e1713 1834 }
6093ae9a 1835 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
804fe108 1836 info->snd_seq, 0, OVS_DP_CMD_GET);
6093ae9a 1837 BUG_ON(err < 0);
8ec609d8 1838 ovs_unlock();
ccb1352e
JG		 1839
1840 return genlmsg_reply(reply, info);
8e4e1713 1841
6093ae9a 1842err_unlock_free:
8ec609d8 1843 ovs_unlock();
6093ae9a 1844 kfree_skb(reply);
8e4e1713 1845 return err;
ccb1352e
JG		 1846}
1847
1848static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1849{
46df7b81 1850 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
ccb1352e
JG		 1851 struct datapath *dp;
1852 int skip = cb->args[0];
1853 int i = 0;
1854
8ec609d8
PS		 1855 ovs_lock();
1856 list_for_each_entry(dp, &ovs_net->dps, list_node) {
77676fdb 1857 if (i >= skip &&
15e47304 1858 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
ccb1352e 1859 cb->nlh->nlmsg_seq, NLM_F_MULTI,
804fe108 1860 OVS_DP_CMD_GET) < 0)
ccb1352e
JG		 1861 break;
1862 i++;
1863 }
8ec609d8 1864 ovs_unlock();
ccb1352e
JG		 1865
1866 cb->args[0] = i;
1867
1868 return skb->len;
1869}
1870
0c200ef9
PS		 1871static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1872 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1873 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1874 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1875};
1876
48e48a70 1877static const struct genl_ops dp_datapath_genl_ops[] = {
ccb1352e 1878 { .cmd = OVS_DP_CMD_NEW,
ef6243ac 1879 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1880 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 1881 .doit = ovs_dp_cmd_new
1882 },
1883 { .cmd = OVS_DP_CMD_DEL,
ef6243ac 1884 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1885 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 1886 .doit = ovs_dp_cmd_del
1887 },
1888 { .cmd = OVS_DP_CMD_GET,
ef6243ac 1889 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
ccb1352e 1890 .flags = 0, /* OK for unprivileged users. */
ccb1352e
JG		 1891 .doit = ovs_dp_cmd_get,
1892 .dumpit = ovs_dp_cmd_dump
1893 },
1894 { .cmd = OVS_DP_CMD_SET,
ef6243ac 1895 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1896 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 1897 .doit = ovs_dp_cmd_set,
1898 },
1899};
1900
56989f6d 1901static struct genl_family dp_datapath_genl_family __ro_after_init = {
ccb1352e 1902 .hdrsize = sizeof(struct ovs_header),
0c200ef9
PS		 1903 .name = OVS_DATAPATH_FAMILY,
1904 .version = OVS_DATAPATH_VERSION,
1905 .maxattr = OVS_DP_ATTR_MAX,
3b0f31f2 1906 .policy = datapath_policy,
3a4e0d6a
PS		 1907 .netnsok = true,
1908 .parallel_ops = true,
0c200ef9
PS		 1909 .ops = dp_datapath_genl_ops,
1910 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1911 .mcgrps = &ovs_dp_datapath_multicast_group,
1912 .n_mcgrps = 1,
489111e5 1913 .module = THIS_MODULE,
ccb1352e
JG		 1914};
1915
8e4e1713 1916/* Called with ovs_mutex or RCU read lock. */
ccb1352e 1917static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
9354d452 1918 struct net *net, u32 portid, u32 seq,
d4e4fdf9 1919 u32 flags, u8 cmd, gfp_t gfp)
ccb1352e
JG		 1920{
1921 struct ovs_header *ovs_header;
1922 struct ovs_vport_stats vport_stats;
1923 int err;
1924
15e47304 1925 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
ccb1352e
JG		 1926 flags, cmd);
1927 if (!ovs_header)
1928 return -EMSGSIZE;
1929
1930 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1931
028d6a67
DM		 1932 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1933 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
5cd667b0 1934 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
9354d452
JB		 1935 ovs_vport_name(vport)) ||
1936 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
028d6a67 1937 goto nla_put_failure;
ccb1352e 1938
9354d452 1939 if (!net_eq(net, dev_net(vport->dev))) {
d4e4fdf9 1940 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
9354d452
JB		 1941
1942 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1943 goto nla_put_failure;
1944 }
1945
ccb1352e 1946 ovs_vport_get_stats(vport, &vport_stats);
66c7a5ee
ND		 1947 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1948 sizeof(struct ovs_vport_stats), &vport_stats,
1949 OVS_VPORT_ATTR_PAD))
028d6a67 1950 goto nla_put_failure;
ccb1352e 1951
5cd667b0
AW		 1952 if (ovs_vport_get_upcall_portids(vport, skb))
1953 goto nla_put_failure;
1954
ccb1352e
JG		 1955 err = ovs_vport_get_options(vport, skb);
1956 if (err == -EMSGSIZE)
1957 goto error;
1958
053c095a
JB		 1959 genlmsg_end(skb, ovs_header);
1960 return 0;
ccb1352e
JG		 1961
1962nla_put_failure:
1963 err = -EMSGSIZE;
1964error:
1965 genlmsg_cancel(skb, ovs_header);
1966 return err;
1967}
1968
6093ae9a
JR		 1969static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1970{
1971 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1972}
1973
1974/* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
9354d452
JB		 1975struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1976 u32 portid, u32 seq, u8 cmd)
ccb1352e
JG		 1977{
1978 struct sk_buff *skb;
1979 int retval;
1980
d4e4fdf9 1981 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
ccb1352e
JG		 1982 if (!skb)
1983 return ERR_PTR(-ENOMEM);
1984
d4e4fdf9
GN		 1985 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
1986 GFP_KERNEL);
a9341512
JG		 1987 BUG_ON(retval < 0);
1988
ccb1352e
JG		 1989 return skb;
1990}
1991
8e4e1713 1992/* Called with ovs_mutex or RCU read lock. */
46df7b81 1993static struct vport *lookup_vport(struct net *net,
12eb18f7 1994 const struct ovs_header *ovs_header,
ccb1352e
JG		 1995 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1996{
1997 struct datapath *dp;
1998 struct vport *vport;
1999
9354d452
JB		 2000 if (a[OVS_VPORT_ATTR_IFINDEX])
2001 return ERR_PTR(-EOPNOTSUPP);
ccb1352e 2002 if (a[OVS_VPORT_ATTR_NAME]) {
46df7b81 2003 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
ccb1352e
JG		 2004 if (!vport)
2005 return ERR_PTR(-ENODEV);
651a68ea
BP		 2006 if (ovs_header->dp_ifindex &&
2007 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2008 return ERR_PTR(-ENODEV);
ccb1352e
JG		 2009 return vport;
2010 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2011 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2012
2013 if (port_no >= DP_MAX_PORTS)
2014 return ERR_PTR(-EFBIG);
2015
46df7b81 2016 dp = get_dp(net, ovs_header->dp_ifindex);
ccb1352e
JG		 2017 if (!dp)
2018 return ERR_PTR(-ENODEV);
2019
8e4e1713 2020 vport = ovs_vport_ovsl_rcu(dp, port_no);
ccb1352e 2021 if (!vport)
14408dba 2022 return ERR_PTR(-ENODEV);
ccb1352e
JG		 2023 return vport;
2024 } else
2025 return ERR_PTR(-EINVAL);
9354d452 2026
ccb1352e
JG		 2027}
2028
6b660c41 2029static unsigned int ovs_get_max_headroom(struct datapath *dp)
3a927bc7 2030{
6b660c41 2031 unsigned int dev_headroom, max_headroom = 0;
3a927bc7
PA		 2032 struct net_device *dev;
2033 struct vport *vport;
2034 int i;
2035
2036 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2037 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2038 dev = vport->dev;
2039 dev_headroom = netdev_get_fwd_headroom(dev);
2040 if (dev_headroom > max_headroom)
2041 max_headroom = dev_headroom;
2042 }
2043 }
2044
6b660c41
TY		 2045 return max_headroom;
2046}
2047
2048/* Called with ovs_mutex */
2049static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2050{
2051 struct vport *vport;
2052 int i;
2053
2054 dp->max_headroom = new_headroom;
3a927bc7
PA		 2055 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2056 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
6b660c41 2057 netdev_set_rx_headroom(vport->dev, new_headroom);
3a927bc7
PA		 2058}
2059
ccb1352e
JG		 2060static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2061{
2062 struct nlattr **a = info->attrs;
2063 struct ovs_header *ovs_header = info->userhdr;
2064 struct vport_parms parms;
2065 struct sk_buff *reply;
2066 struct vport *vport;
2067 struct datapath *dp;
6b660c41 2068 unsigned int new_headroom;
ccb1352e
JG		 2069 u32 port_no;
2070 int err;
2071
ccb1352e
JG		 2072 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2073 !a[OVS_VPORT_ATTR_UPCALL_PID])
6093ae9a 2074 return -EINVAL;
9354d452
JB		 2075 if (a[OVS_VPORT_ATTR_IFINDEX])
2076 return -EOPNOTSUPP;
6093ae9a
JR		 2077
2078 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2079 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2080 if (port_no >= DP_MAX_PORTS)
2081 return -EFBIG;
2082
2083 reply = ovs_vport_cmd_alloc_info();
2084 if (!reply)
2085 return -ENOMEM;
ccb1352e 2086
8e4e1713 2087 ovs_lock();
62b9c8d0 2088restart:
46df7b81 2089 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
ccb1352e
JG		 2090 err = -ENODEV;
2091 if (!dp)
6093ae9a 2092 goto exit_unlock_free;
ccb1352e 2093
6093ae9a 2094 if (port_no) {
8e4e1713 2095 vport = ovs_vport_ovsl(dp, port_no);
ccb1352e
JG		 2096 err = -EBUSY;
2097 if (vport)
6093ae9a 2098 goto exit_unlock_free;
ccb1352e
JG		 2099 } else {
2100 for (port_no = 1; ; port_no++) {
2101 if (port_no >= DP_MAX_PORTS) {
2102 err = -EFBIG;
6093ae9a 2103 goto exit_unlock_free;
ccb1352e 2104 }
8e4e1713 2105 vport = ovs_vport_ovsl(dp, port_no);
ccb1352e
JG		 2106 if (!vport)
2107 break;
2108 }
2109 }
2110
2111 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2112 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2113 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2114 parms.dp = dp;
2115 parms.port_no = port_no;
5cd667b0 2116 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
ccb1352e
JG		 2117
2118 vport = new_vport(&parms);
2119 err = PTR_ERR(vport);
62b9c8d0
TG		 2120 if (IS_ERR(vport)) {
2121 if (err == -EAGAIN)
2122 goto restart;
6093ae9a 2123 goto exit_unlock_free;
62b9c8d0 2124 }
ccb1352e 2125
9354d452
JB		 2126 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2127 info->snd_portid, info->snd_seq, 0,
d4e4fdf9 2128 OVS_VPORT_CMD_NEW, GFP_KERNEL);
3a927bc7 2129
6b660c41
TY		 2130 new_headroom = netdev_get_fwd_headroom(vport->dev);
2131
2132 if (new_headroom > dp->max_headroom)
2133 ovs_update_headroom(dp, new_headroom);
3a927bc7
PA		 2134 else
2135 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2136
6093ae9a
JR		 2137 BUG_ON(err < 0);
2138 ovs_unlock();
ed661185 2139
2a94fe48 2140 ovs_notify(&dp_vport_genl_family, reply, info);
6093ae9a 2141 return 0;
ccb1352e 2142
6093ae9a 2143exit_unlock_free:
8e4e1713 2144 ovs_unlock();
6093ae9a 2145 kfree_skb(reply);
ccb1352e
JG		 2146 return err;
2147}
2148
2149static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2150{
2151 struct nlattr **a = info->attrs;
2152 struct sk_buff *reply;
2153 struct vport *vport;
2154 int err;
2155
6093ae9a
JR		 2156 reply = ovs_vport_cmd_alloc_info();
2157 if (!reply)
2158 return -ENOMEM;
2159
8e4e1713 2160 ovs_lock();
46df7b81 2161 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
ccb1352e
JG		 2162 err = PTR_ERR(vport);
2163 if (IS_ERR(vport))
6093ae9a 2164 goto exit_unlock_free;
ccb1352e 2165
ccb1352e 2166 if (a[OVS_VPORT_ATTR_TYPE] &&
f44f3408 2167 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
ccb1352e 2168 err = -EINVAL;
6093ae9a 2169 goto exit_unlock_free;
a9341512
JG		 2170 }
2171
f44f3408 2172 if (a[OVS_VPORT_ATTR_OPTIONS]) {
ccb1352e 2173 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
f44f3408 2174 if (err)
6093ae9a 2175 goto exit_unlock_free;
f44f3408 2176 }
a9341512 2177
5cd667b0
AW		 2178
2179 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2180 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2181
2182 err = ovs_vport_set_upcall_portids(vport, ids);
2183 if (err)
2184 goto exit_unlock_free;
2185 }
ccb1352e 2186
9354d452
JB		 2187 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2188 info->snd_portid, info->snd_seq, 0,
d4e4fdf9 2189 OVS_VPORT_CMD_SET, GFP_KERNEL);
a9341512 2190 BUG_ON(err < 0);
ccb1352e 2191
8e4e1713 2192 ovs_unlock();
2a94fe48 2193 ovs_notify(&dp_vport_genl_family, reply, info);
8e4e1713 2194 return 0;
ccb1352e 2195
6093ae9a 2196exit_unlock_free:
8e4e1713 2197 ovs_unlock();
6093ae9a 2198 kfree_skb(reply);
ccb1352e
JG		 2199 return err;
2200}
2201
2202static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2203{
6b660c41 2204 bool update_headroom = false;
ccb1352e
JG		 2205 struct nlattr **a = info->attrs;
2206 struct sk_buff *reply;
3a927bc7 2207 struct datapath *dp;
ccb1352e 2208 struct vport *vport;
6b660c41 2209 unsigned int new_headroom;
ccb1352e
JG		 2210 int err;
2211
6093ae9a
JR		 2212 reply = ovs_vport_cmd_alloc_info();
2213 if (!reply)
2214 return -ENOMEM;
2215
8e4e1713 2216 ovs_lock();
46df7b81 2217 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
ccb1352e
JG		 2218 err = PTR_ERR(vport);
2219 if (IS_ERR(vport))
6093ae9a 2220 goto exit_unlock_free;
ccb1352e
JG		 2221
2222 if (vport->port_no == OVSP_LOCAL) {
2223 err = -EINVAL;
6093ae9a 2224 goto exit_unlock_free;
ccb1352e
JG		 2225 }
2226
9354d452
JB		 2227 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2228 info->snd_portid, info->snd_seq, 0,
d4e4fdf9 2229 OVS_VPORT_CMD_DEL, GFP_KERNEL);
6093ae9a 2230 BUG_ON(err < 0);
3a927bc7
PA		 2231
2232 /* the vport deletion may trigger dp headroom update */
2233 dp = vport->dp;
2234 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
6b660c41
TY		 2235 update_headroom = true;
2236
3a927bc7 2237 netdev_reset_rx_headroom(vport->dev);
ccb1352e 2238 ovs_dp_detach_port(vport);
3a927bc7 2239
6b660c41
TY		 2240 if (update_headroom) {
2241 new_headroom = ovs_get_max_headroom(dp);
2242
2243 if (new_headroom < dp->max_headroom)
2244 ovs_update_headroom(dp, new_headroom);
2245 }
6093ae9a 2246 ovs_unlock();
ccb1352e 2247
2a94fe48 2248 ovs_notify(&dp_vport_genl_family, reply, info);
6093ae9a 2249 return 0;
ccb1352e 2250
6093ae9a 2251exit_unlock_free:
8e4e1713 2252 ovs_unlock();
6093ae9a 2253 kfree_skb(reply);
ccb1352e
JG		 2254 return err;
2255}
2256
2257static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2258{
2259 struct nlattr **a = info->attrs;
2260 struct ovs_header *ovs_header = info->userhdr;
2261 struct sk_buff *reply;
2262 struct vport *vport;
2263 int err;
2264
6093ae9a
JR		 2265 reply = ovs_vport_cmd_alloc_info();
2266 if (!reply)
2267 return -ENOMEM;
2268
ccb1352e 2269 rcu_read_lock();
46df7b81 2270 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
ccb1352e
JG		 2271 err = PTR_ERR(vport);
2272 if (IS_ERR(vport))
6093ae9a 2273 goto exit_unlock_free;
9354d452
JB		 2274 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2275 info->snd_portid, info->snd_seq, 0,
d4e4fdf9 2276 OVS_VPORT_CMD_GET, GFP_ATOMIC);
6093ae9a 2277 BUG_ON(err < 0);
ccb1352e
JG		 2278 rcu_read_unlock();
2279
2280 return genlmsg_reply(reply, info);
2281
6093ae9a 2282exit_unlock_free:
ccb1352e 2283 rcu_read_unlock();
6093ae9a 2284 kfree_skb(reply);
ccb1352e
JG		 2285 return err;
2286}
2287
2288static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2289{
2290 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2291 struct datapath *dp;
15eac2a7
PS		 2292 int bucket = cb->args[0], skip = cb->args[1];
2293 int i, j = 0;
ccb1352e 2294
42ee19e2 2295 rcu_read_lock();
cc3a5ae6 2296 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
42ee19e2
JR		 2297 if (!dp) {
2298 rcu_read_unlock();
ccb1352e 2299 return -ENODEV;
42ee19e2 2300 }
15eac2a7 2301 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
ccb1352e 2302 struct vport *vport;
15eac2a7
PS		 2303
2304 j = 0;
b67bfe0d 2305 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
15eac2a7
PS		 2306 if (j >= skip &&
2307 ovs_vport_cmd_fill_info(vport, skb,
9354d452 2308 sock_net(skb->sk),
15e47304 2309 NETLINK_CB(cb->skb).portid,
15eac2a7
PS		 2310 cb->nlh->nlmsg_seq,
2311 NLM_F_MULTI,
d4e4fdf9
GN		 2312 OVS_VPORT_CMD_GET,
2313 GFP_ATOMIC) < 0)
15eac2a7
PS		 2314 goto out;
2315
2316 j++;
2317 }
2318 skip = 0;
ccb1352e 2319 }
15eac2a7 2320out:
ccb1352e
JG		 2321 rcu_read_unlock();
2322
15eac2a7
PS		 2323 cb->args[0] = i;
2324 cb->args[1] = j;
ccb1352e 2325
15eac2a7 2326 return skb->len;
ccb1352e
JG		 2327}
2328
0c200ef9
PS		 2329static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2330 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2331 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2332 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2333 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
ea8564c8 2334 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
0c200ef9 2335 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
9354d452
JB		 2336 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2337 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
0c200ef9
PS		 2338};
2339
48e48a70 2340static const struct genl_ops dp_vport_genl_ops[] = {
ccb1352e 2341 { .cmd = OVS_VPORT_CMD_NEW,
ef6243ac 2342 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 2343 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 2344 .doit = ovs_vport_cmd_new
2345 },
2346 { .cmd = OVS_VPORT_CMD_DEL,
ef6243ac 2347 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 2348 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 2349 .doit = ovs_vport_cmd_del
2350 },
2351 { .cmd = OVS_VPORT_CMD_GET,
ef6243ac 2352 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
ccb1352e 2353 .flags = 0, /* OK for unprivileged users. */
ccb1352e
JG		 2354 .doit = ovs_vport_cmd_get,
2355 .dumpit = ovs_vport_cmd_dump
2356 },
2357 { .cmd = OVS_VPORT_CMD_SET,
ef6243ac 2358 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 2359 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 2360 .doit = ovs_vport_cmd_set,
2361 },
2362};
2363
56989f6d 2364struct genl_family dp_vport_genl_family __ro_after_init = {
0c200ef9
PS		 2365 .hdrsize = sizeof(struct ovs_header),
2366 .name = OVS_VPORT_FAMILY,
2367 .version = OVS_VPORT_VERSION,
2368 .maxattr = OVS_VPORT_ATTR_MAX,
3b0f31f2 2369 .policy = vport_policy,
0c200ef9
PS		 2370 .netnsok = true,
2371 .parallel_ops = true,
2372 .ops = dp_vport_genl_ops,
2373 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2374 .mcgrps = &ovs_dp_vport_multicast_group,
2375 .n_mcgrps = 1,
489111e5 2376 .module = THIS_MODULE,
ccb1352e
JG		 2377};
2378
0c200ef9
PS		 2379static struct genl_family * const dp_genl_families[] = {
2380 &dp_datapath_genl_family,
2381 &dp_vport_genl_family,
2382 &dp_flow_genl_family,
2383 &dp_packet_genl_family,
96fbc13d 2384 &dp_meter_genl_family,
11efd5cb
YHW		 2385#if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2386 &dp_ct_limit_genl_family,
2387#endif
ccb1352e
JG		 2388};
2389
2390static void dp_unregister_genl(int n_families)
2391{
2392 int i;
2393
2394 for (i = 0; i < n_families; i++)
0c200ef9 2395 genl_unregister_family(dp_genl_families[i]);
ccb1352e
JG		 2396}
2397
56989f6d 2398static int __init dp_register_genl(void)
ccb1352e 2399{
ccb1352e
JG		 2400 int err;
2401 int i;
2402
ccb1352e 2403 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
ccb1352e 2404
0c200ef9 2405 err = genl_register_family(dp_genl_families[i]);
ccb1352e
JG		 2406 if (err)
2407 goto error;
ccb1352e
JG		 2408 }
2409
2410 return 0;
2411
2412error:
0c200ef9 2413 dp_unregister_genl(i);
ccb1352e
JG		 2414 return err;
2415}
2416
46df7b81
PS		 2417static int __net_init ovs_init_net(struct net *net)
2418{
2419 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2420
2421 INIT_LIST_HEAD(&ovs_net->dps);
8e4e1713 2422 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
11efd5cb 2423 return ovs_ct_init(net);
46df7b81
PS		 2424}
2425
7b4577a9
PS		 2426static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2427 struct list_head *head)
46df7b81 2428{
8e4e1713 2429 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
7b4577a9
PS		 2430 struct datapath *dp;
2431
2432 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2433 int i;
2434
2435 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2436 struct vport *vport;
2437
2438 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
7b4577a9
PS		 2439 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2440 continue;
2441
be4ace6e 2442 if (dev_net(vport->dev) == dnet)
7b4577a9
PS		 2443 list_add(&vport->detach_list, head);
2444 }
2445 }
2446 }
2447}
2448
2449static void __net_exit ovs_exit_net(struct net *dnet)
2450{
2451 struct datapath *dp, *dp_next;
2452 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2453 struct vport *vport, *vport_next;
2454 struct net *net;
2455 LIST_HEAD(head);
46df7b81 2456
c2ac6673 2457 ovs_ct_exit(dnet);
8e4e1713 2458 ovs_lock();
46df7b81
PS		 2459 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2460 __dp_destroy(dp);
7b4577a9 2461
f0b07bb1 2462 down_read(&net_rwsem);
7b4577a9
PS		 2463 for_each_net(net)
2464 list_vports_from_net(net, dnet, &head);
f0b07bb1 2465 up_read(&net_rwsem);
7b4577a9
PS		 2466
2467 /* Detach all vports from given namespace. */
2468 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2469 list_del(&vport->detach_list);
2470 ovs_dp_detach_port(vport);
2471 }
2472
8e4e1713
PS		 2473 ovs_unlock();
2474
2475 cancel_work_sync(&ovs_net->dp_notify_work);
46df7b81
PS		 2476}
2477
2478static struct pernet_operations ovs_net_ops = {
2479 .init = ovs_init_net,
2480 .exit = ovs_exit_net,
2481 .id = &ovs_net_id,
2482 .size = sizeof(struct ovs_net),
2483};
2484
ccb1352e
JG		 2485static int __init dp_init(void)
2486{
ccb1352e
JG		 2487 int err;
2488
3523b29b 2489 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
ccb1352e
JG		 2490
2491 pr_info("Open vSwitch switching datapath\n");
2492
971427f3 2493 err = action_fifos_init();
ccb1352e
JG		 2494 if (err)
2495 goto error;
2496
971427f3
AZ		 2497 err = ovs_internal_dev_rtnl_link_register();
2498 if (err)
2499 goto error_action_fifos_exit;
2500
5b9e7e16
JP		 2501 err = ovs_flow_init();
2502 if (err)
2503 goto error_unreg_rtnl_link;
2504
ccb1352e
JG		 2505 err = ovs_vport_init();
2506 if (err)
2507 goto error_flow_exit;
2508
46df7b81 2509 err = register_pernet_device(&ovs_net_ops);
ccb1352e
JG		 2510 if (err)
2511 goto error_vport_exit;
2512
46df7b81
PS		 2513 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2514 if (err)
2515 goto error_netns_exit;
2516
62b9c8d0
TG		 2517 err = ovs_netdev_init();
2518 if (err)
2519 goto error_unreg_notifier;
2520
ccb1352e
JG		 2521 err = dp_register_genl();
2522 if (err < 0)
62b9c8d0 2523 goto error_unreg_netdev;
ccb1352e 2524
ccb1352e
JG		 2525 return 0;
2526
62b9c8d0
TG		 2527error_unreg_netdev:
2528 ovs_netdev_exit();
ccb1352e
JG		 2529error_unreg_notifier:
2530 unregister_netdevice_notifier(&ovs_dp_device_notifier);
46df7b81
PS		 2531error_netns_exit:
2532 unregister_pernet_device(&ovs_net_ops);
ccb1352e
JG		 2533error_vport_exit:
2534 ovs_vport_exit();
2535error_flow_exit:
2536 ovs_flow_exit();
5b9e7e16
JP		 2537error_unreg_rtnl_link:
2538 ovs_internal_dev_rtnl_link_unregister();
971427f3
AZ		 2539error_action_fifos_exit:
2540 action_fifos_exit();
ccb1352e
JG		 2541error:
2542 return err;
2543}
2544
2545static void dp_cleanup(void)
2546{
ccb1352e 2547 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
62b9c8d0 2548 ovs_netdev_exit();
ccb1352e 2549 unregister_netdevice_notifier(&ovs_dp_device_notifier);
46df7b81
PS		 2550 unregister_pernet_device(&ovs_net_ops);
2551 rcu_barrier();
ccb1352e
JG		 2552 ovs_vport_exit();
2553 ovs_flow_exit();
5b9e7e16 2554 ovs_internal_dev_rtnl_link_unregister();
971427f3 2555 action_fifos_exit();
ccb1352e
JG		 2556}
2557
2558module_init(dp_init);
2559module_exit(dp_cleanup);
2560
2561MODULE_DESCRIPTION("Open vSwitch switching datapath");
2562MODULE_LICENSE("GPL");
ed227099
TLSC		 2563MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2564MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2565MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2566MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
96fbc13d 2567MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
11efd5cb 2568MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
Linux 6.x block layer and io_uring tree(s)