projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
vfs: avoid duplicating creds in faccessat if possible
[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>
35d39fec 40#include <net/pkt_cls.h>
ccb1352e
JG		 41
42#include "datapath.h"
43#include "flow.h"
e80857cc 44#include "flow_table.h"
e6445719 45#include "flow_netlink.h"
96fbc13d 46#include "meter.h"
c4ab7b56 47#include "openvswitch_trace.h"
ccb1352e 48#include "vport-internal_dev.h"
cff63a52 49#include "vport-netdev.h"
ccb1352e 50
c7d03a00 51unsigned int ovs_net_id __read_mostly;
8e4e1713 52
0c200ef9
PS		 53static struct genl_family dp_packet_genl_family;
54static struct genl_family dp_flow_genl_family;
55static struct genl_family dp_datapath_genl_family;
56
74ed7ab9
JS		 57static const struct nla_policy flow_policy[];
58
48e48a70 59static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
60 .name = OVS_FLOW_MCGROUP,
0c200ef9
PS		 61};
62
48e48a70 63static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
64 .name = OVS_DATAPATH_MCGROUP,
0c200ef9
PS		 65};
66
48e48a70 67static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
68 .name = OVS_VPORT_MCGROUP,
0c200ef9
PS		 69};
70
fb5d1e9e
JR		 71/* Check if need to build a reply message.
72 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
9b67aa4a
SG		 73static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
74 unsigned int group)
fb5d1e9e
JR		 75{
76 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
f8403a2e 77 genl_has_listeners(family, genl_info_net(info), group);
fb5d1e9e
JR		 78}
79
68eb5503 80static void ovs_notify(struct genl_family *family,
2a94fe48 81 struct sk_buff *skb, struct genl_info *info)
ed661185 82{
92c14d9b 83 genl_notify(family, skb, info, 0, GFP_KERNEL);
ed661185
TG		 84}
85
ccb1352e
JG		 86/**
87 * DOC: Locking:
88 *
8e4e1713
PS		 89 * All writes e.g. Writes to device state (add/remove datapath, port, set
90 * operations on vports, etc.), Writes to other state (flow table
91 * modifications, set miscellaneous datapath parameters, etc.) are protected
92 * by ovs_lock.
ccb1352e
JG		 93 *
94 * Reads are protected by RCU.
95 *
96 * There are a few special cases (mostly stats) that have their own
97 * synchronization but they nest under all of above and don't interact with
98 * each other.
8e4e1713
PS		 99 *
100 * The RTNL lock nests inside ovs_mutex.
ccb1352e
JG		 101 */
102
8e4e1713
PS		 103static DEFINE_MUTEX(ovs_mutex);
104
105void ovs_lock(void)
106{
107 mutex_lock(&ovs_mutex);
108}
109
110void ovs_unlock(void)
111{
112 mutex_unlock(&ovs_mutex);
113}
114
115#ifdef CONFIG_LOCKDEP
116int lockdep_ovsl_is_held(void)
117{
118 if (debug_locks)
119 return lockdep_is_held(&ovs_mutex);
120 else
121 return 1;
122}
123#endif
124
ccb1352e 125static struct vport *new_vport(const struct vport_parms *);
8055a89c 126static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
e8eedb85 127 const struct sw_flow_key *,
f2a4d086
WT		 128 const struct dp_upcall_info *,
129 uint32_t cutlen);
8055a89c 130static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
e8eedb85 131 const struct sw_flow_key *,
f2a4d086
WT		 132 const struct dp_upcall_info *,
133 uint32_t cutlen);
ccb1352e 134
eac87c41
EC		 135static void ovs_dp_masks_rebalance(struct work_struct *work);
136
b83d23a2
MG		 137static int ovs_dp_set_upcall_portids(struct datapath *, const struct nlattr *);
138
8e4e1713 139/* Must be called with rcu_read_lock or ovs_mutex. */
971427f3 140const char *ovs_dp_name(const struct datapath *dp)
ccb1352e 141{
8e4e1713 142 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
c9db965c 143 return ovs_vport_name(vport);
ccb1352e
JG		 144}
145
12eb18f7 146static int get_dpifindex(const struct datapath *dp)
ccb1352e
JG		 147{
148 struct vport *local;
149 int ifindex;
150
151 rcu_read_lock();
152
15eac2a7 153 local = ovs_vport_rcu(dp, OVSP_LOCAL);
ccb1352e 154 if (local)
be4ace6e 155 ifindex = local->dev->ifindex;
ccb1352e
JG		 156 else
157 ifindex = 0;
158
159 rcu_read_unlock();
160
161 return ifindex;
162}
163
164static void destroy_dp_rcu(struct rcu_head *rcu)
165{
166 struct datapath *dp = container_of(rcu, struct datapath, rcu);
167
9b996e54 168 ovs_flow_tbl_destroy(&dp->table);
ccb1352e 169 free_percpu(dp->stats_percpu);
15eac2a7 170 kfree(dp->ports);
96fbc13d 171 ovs_meters_exit(dp);
076999e4 172 kfree(rcu_dereference_raw(dp->upcall_portids));
ccb1352e
JG		 173 kfree(dp);
174}
175
15eac2a7
PS		 176static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
177 u16 port_no)
178{
179 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
180}
181
bb6f9a70 182/* Called with ovs_mutex or RCU read lock. */
15eac2a7
PS		 183struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
184{
185 struct vport *vport;
15eac2a7
PS		 186 struct hlist_head *head;
187
188 head = vport_hash_bucket(dp, port_no);
53742e69 189 hlist_for_each_entry_rcu(vport, head, dp_hash_node,
cf3266ad 190 lockdep_ovsl_is_held()) {
15eac2a7
PS		 191 if (vport->port_no == port_no)
192 return vport;
193 }
194 return NULL;
195}
196
8e4e1713 197/* Called with ovs_mutex. */
ccb1352e
JG		 198static struct vport *new_vport(const struct vport_parms *parms)
199{
200 struct vport *vport;
201
202 vport = ovs_vport_add(parms);
203 if (!IS_ERR(vport)) {
204 struct datapath *dp = parms->dp;
15eac2a7 205 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
ccb1352e 206
15eac2a7 207 hlist_add_head_rcu(&vport->dp_hash_node, head);
ccb1352e 208 }
ccb1352e
JG		 209 return vport;
210}
211
1933ea36 212static void ovs_vport_update_upcall_stats(struct sk_buff *skb,
213 const struct dp_upcall_info *upcall_info,
214 bool upcall_result)
215{
216 struct vport *p = OVS_CB(skb)->input_vport;
217 struct vport_upcall_stats_percpu *stats;
218
219 if (upcall_info->cmd != OVS_PACKET_CMD_MISS &&
220 upcall_info->cmd != OVS_PACKET_CMD_ACTION)
221 return;
222
223 stats = this_cpu_ptr(p->upcall_stats);
224 u64_stats_update_begin(&stats->syncp);
225 if (upcall_result)
226 u64_stats_inc(&stats->n_success);
227 else
228 u64_stats_inc(&stats->n_fail);
229 u64_stats_update_end(&stats->syncp);
230}
231
ccb1352e
JG		 232void ovs_dp_detach_port(struct vport *p)
233{
8e4e1713 234 ASSERT_OVSL();
ccb1352e
JG		 235
236 /* First drop references to device. */
15eac2a7 237 hlist_del_rcu(&p->dp_hash_node);
ccb1352e 238
1933ea36 239 /* Free percpu memory */
240 free_percpu(p->upcall_stats);
241
ccb1352e
JG		 242 /* Then destroy it. */
243 ovs_vport_del(p);
244}
245
246/* Must be called with rcu_read_lock. */
8c8b1b83 247void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
ccb1352e 248{
83c8df26 249 const struct vport *p = OVS_CB(skb)->input_vport;
ccb1352e
JG		 250 struct datapath *dp = p->dp;
251 struct sw_flow *flow;
d98612b8 252 struct sw_flow_actions *sf_acts;
ccb1352e 253 struct dp_stats_percpu *stats;
ccb1352e 254 u64 *stats_counter;
1bd7116f 255 u32 n_mask_hit;
9d2f627b 256 u32 n_cache_hit;
aa733660 257 int error;
ccb1352e 258
404f2f10 259 stats = this_cpu_ptr(dp->stats_percpu);
ccb1352e 260
ccb1352e 261 /* Look up flow. */
04b7d136 262 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
9d2f627b 263 &n_mask_hit, &n_cache_hit);
ccb1352e
JG		 264 if (unlikely(!flow)) {
265 struct dp_upcall_info upcall;
266
ccea7445 267 memset(&upcall, 0, sizeof(upcall));
ccb1352e 268 upcall.cmd = OVS_PACKET_CMD_MISS;
b83d23a2
MG		 269
270 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
784dcfa5
MG		 271 upcall.portid =
272 ovs_dp_get_upcall_portid(dp, smp_processor_id());
b83d23a2
MG		 273 else
274 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
275
7f8a436e 276 upcall.mru = OVS_CB(skb)->mru;
f2a4d086 277 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
1100248a
MP		 278 switch (error) {
279 case 0:
280 case -EAGAIN:
281 case -ERESTARTSYS:
282 case -EINTR:
c5eba0b6 283 consume_skb(skb);
1100248a
MP		 284 break;
285 default:
286 kfree_skb(skb);
287 break;
288 }
ccb1352e
JG		 289 stats_counter = &stats->n_missed;
290 goto out;
291 }
292
d98612b8
LJ		 293 ovs_flow_stats_update(flow, key->tp.flags, skb);
294 sf_acts = rcu_dereference(flow->sf_acts);
aa733660
YS		 295 error = ovs_execute_actions(dp, skb, sf_acts, key);
296 if (unlikely(error))
297 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
cf3266ad 298 ovs_dp_name(dp), error);
ccb1352e 299
e298e505 300 stats_counter = &stats->n_hit;
ccb1352e
JG		 301
302out:
303 /* Update datapath statistics. */
df9d9fdf 304 u64_stats_update_begin(&stats->syncp);
ccb1352e 305 (*stats_counter)++;
1bd7116f 306 stats->n_mask_hit += n_mask_hit;
9d2f627b 307 stats->n_cache_hit += n_cache_hit;
df9d9fdf 308 u64_stats_update_end(&stats->syncp);
ccb1352e
JG		 309}
310
ccb1352e 311int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
e8eedb85 312 const struct sw_flow_key *key,
f2a4d086
WT		 313 const struct dp_upcall_info *upcall_info,
314 uint32_t cutlen)
ccb1352e
JG		 315{
316 struct dp_stats_percpu *stats;
ccb1352e
JG		 317 int err;
318
c4ab7b56
AC		 319 if (trace_ovs_dp_upcall_enabled())
320 trace_ovs_dp_upcall(dp, skb, key, upcall_info);
321
15e47304 322 if (upcall_info->portid == 0) {
ccb1352e
JG		 323 err = -ENOTCONN;
324 goto err;
325 }
326
ccb1352e 327 if (!skb_is_gso(skb))
f2a4d086 328 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
ccb1352e 329 else
f2a4d086 330 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
1933ea36 331
332 ovs_vport_update_upcall_stats(skb, upcall_info, !err);
ccb1352e
JG		 333 if (err)
334 goto err;
335
336 return 0;
337
338err:
404f2f10 339 stats = this_cpu_ptr(dp->stats_percpu);
ccb1352e 340
df9d9fdf 341 u64_stats_update_begin(&stats->syncp);
ccb1352e 342 stats->n_lost++;
df9d9fdf 343 u64_stats_update_end(&stats->syncp);
ccb1352e
JG		 344
345 return err;
346}
347
8055a89c 348static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
e8eedb85 349 const struct sw_flow_key *key,
f2a4d086 350 const struct dp_upcall_info *upcall_info,
cf3266ad 351 uint32_t cutlen)
ccb1352e 352{
2734166e 353 unsigned int gso_type = skb_shinfo(skb)->gso_type;
0c19f846 354 struct sw_flow_key later_key;
ccb1352e
JG		 355 struct sk_buff *segs, *nskb;
356 int err;
357
a08e7fd9 358 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
09c5e605 359 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
92e5dfc3
PS		 360 if (IS_ERR(segs))
361 return PTR_ERR(segs);
330966e5
FW		 362 if (segs == NULL)
363 return -EINVAL;
ccb1352e 364
0c19f846
WB		 365 if (gso_type & SKB_GSO_UDP) {
366 /* The initial flow key extracted by ovs_flow_key_extract()
367 * in this case is for a first fragment, so we need to
368 * properly mark later fragments.
369 */
370 later_key = *key;
371 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
372 }
373
ccb1352e 374 /* Queue all of the segments. */
2cec4448 375 skb_list_walk_safe(segs, skb, nskb) {
0c19f846
WB		 376 if (gso_type & SKB_GSO_UDP && skb != segs)
377 key = &later_key;
378
f2a4d086 379 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
ccb1352e
JG		 380 if (err)
381 break;
382
2cec4448 383 }
ccb1352e
JG		 384
385 /* Free all of the segments. */
2cec4448 386 skb_list_walk_safe(segs, skb, nskb) {
ccb1352e
JG		 387 if (err)
388 kfree_skb(skb);
389 else
390 consume_skb(skb);
2cec4448 391 }
ccb1352e
JG		 392 return err;
393}
394
8f0aad6f 395static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
494bea39 396 unsigned int hdrlen, int actions_attrlen)
c3ff8cfe
TG		 397{
398 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
bda56f14 399 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
b95e5928 400 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
bd1903b7
TZ		 401 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
402 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
c3ff8cfe
TG		 403
404 /* OVS_PACKET_ATTR_USERDATA */
8f0aad6f
WZ		 405 if (upcall_info->userdata)
406 size += NLA_ALIGN(upcall_info->userdata->nla_len);
407
408 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
409 if (upcall_info->egress_tun_info)
410 size += nla_total_size(ovs_tun_key_attr_size());
c3ff8cfe 411
ccea7445
NM		 412 /* OVS_PACKET_ATTR_ACTIONS */
413 if (upcall_info->actions_len)
494bea39 414 size += nla_total_size(actions_attrlen);
ccea7445 415
7f8a436e
JS		 416 /* OVS_PACKET_ATTR_MRU */
417 if (upcall_info->mru)
418 size += nla_total_size(sizeof(upcall_info->mru));
419
c3ff8cfe
TG		 420 return size;
421}
422
7f8a436e
JS		 423static void pad_packet(struct datapath *dp, struct sk_buff *skb)
424{
425 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
426 size_t plen = NLA_ALIGN(skb->len) - skb->len;
427
428 if (plen > 0)
b080db58 429 skb_put_zero(skb, plen);
7f8a436e
JS		 430 }
431}
432
8055a89c 433static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
e8eedb85 434 const struct sw_flow_key *key,
f2a4d086
WT		 435 const struct dp_upcall_info *upcall_info,
436 uint32_t cutlen)
ccb1352e
JG		 437{
438 struct ovs_header *upcall;
439 struct sk_buff *nskb = NULL;
4ee45ea0 440 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
ccb1352e 441 struct nlattr *nla;
795449d8 442 size_t len;
bda56f14 443 unsigned int hlen;
8055a89c 444 int err, dp_ifindex;
bd1903b7 445 u64 hash;
8055a89c
TG		 446
447 dp_ifindex = get_dpifindex(dp);
448 if (!dp_ifindex)
449 return -ENODEV;
ccb1352e 450
df8a39de 451 if (skb_vlan_tag_present(skb)) {
ccb1352e
JG		 452 nskb = skb_clone(skb, GFP_ATOMIC);
453 if (!nskb)
454 return -ENOMEM;
455
5968250c 456 nskb = __vlan_hwaccel_push_inside(nskb);
8aa51d64 457 if (!nskb)
ccb1352e
JG		 458 return -ENOMEM;
459
ccb1352e
JG		 460 skb = nskb;
461 }
462
463 if (nla_attr_size(skb->len) > USHRT_MAX) {
464 err = -EFBIG;
465 goto out;
466 }
467
bda56f14
TG		 468 /* Complete checksum if needed */
469 if (skb->ip_summed == CHECKSUM_PARTIAL &&
7529390d 470 (err = skb_csum_hwoffload_help(skb, 0)))
bda56f14
TG		 471 goto out;
472
473 /* Older versions of OVS user space enforce alignment of the last
474 * Netlink attribute to NLA_ALIGNTO which would require extensive
475 * padding logic. Only perform zerocopy if padding is not required.
476 */
477 if (dp->user_features & OVS_DP_F_UNALIGNED)
478 hlen = skb_zerocopy_headlen(skb);
479 else
480 hlen = skb->len;
481
494bea39
LZ		 482 len = upcall_msg_size(upcall_info, hlen - cutlen,
483 OVS_CB(skb)->acts_origlen);
551ddc05 484 user_skb = genlmsg_new(len, GFP_ATOMIC);
ccb1352e
JG		 485 if (!user_skb) {
486 err = -ENOMEM;
487 goto out;
488 }
489
490 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
491 0, upcall_info->cmd);
6f19893b
KL		 492 if (!upcall) {
493 err = -EINVAL;
494 goto out;
495 }
ccb1352e
JG		 496 upcall->dp_ifindex = dp_ifindex;
497
5b4237bb 498 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
a734d1f4
EC		 499 if (err)
500 goto out;
ccb1352e
JG		 501
502 if (upcall_info->userdata)
4490108b
BP		 503 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
504 nla_len(upcall_info->userdata),
505 nla_data(upcall_info->userdata));
ccb1352e 506
8f0aad6f 507 if (upcall_info->egress_tun_info) {
ae0be8de
MK		 508 nla = nla_nest_start_noflag(user_skb,
509 OVS_PACKET_ATTR_EGRESS_TUN_KEY);
0fff9bd4
KL		 510 if (!nla) {
511 err = -EMSGSIZE;
512 goto out;
513 }
fc4099f1
PS		 514 err = ovs_nla_put_tunnel_info(user_skb,
515 upcall_info->egress_tun_info);
a734d1f4
EC		 516 if (err)
517 goto out;
518
8f0aad6f
WZ		 519 nla_nest_end(user_skb, nla);
520 }
521
ccea7445 522 if (upcall_info->actions_len) {
ae0be8de 523 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
0fff9bd4
KL		 524 if (!nla) {
525 err = -EMSGSIZE;
526 goto out;
527 }
ccea7445
NM		 528 err = ovs_nla_put_actions(upcall_info->actions,
529 upcall_info->actions_len,
530 user_skb);
531 if (!err)
532 nla_nest_end(user_skb, nla);
533 else
534 nla_nest_cancel(user_skb, nla);
535 }
536
7f8a436e 537 /* Add OVS_PACKET_ATTR_MRU */
61ca533c
TZ		 538 if (upcall_info->mru &&
539 nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
540 err = -ENOBUFS;
541 goto out;
7f8a436e
JS		 542 }
543
b95e5928 544 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
61ca533c
TZ		 545 if (cutlen > 0 &&
546 nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
547 err = -ENOBUFS;
548 goto out;
b95e5928
WT		 549 }
550
bd1903b7
TZ		 551 /* Add OVS_PACKET_ATTR_HASH */
552 hash = skb_get_hash_raw(skb);
553 if (skb->sw_hash)
554 hash |= OVS_PACKET_HASH_SW_BIT;
555
556 if (skb->l4_hash)
557 hash |= OVS_PACKET_HASH_L4_BIT;
558
559 if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
560 err = -ENOBUFS;
561 goto out;
562 }
563
bda56f14
TG		 564 /* Only reserve room for attribute header, packet data is added
565 * in skb_zerocopy() */
566 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
567 err = -ENOBUFS;
568 goto out;
569 }
f2a4d086 570 nla->nla_len = nla_attr_size(skb->len - cutlen);
ccb1352e 571
f2a4d086 572 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
36d5fe6a
ZK		 573 if (err)
574 goto out;
ccb1352e 575
aea0bb4f 576 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
7f8a436e 577 pad_packet(dp, user_skb);
aea0bb4f 578
bda56f14 579 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
ccb1352e 580
bda56f14 581 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
4ee45ea0 582 user_skb = NULL;
ccb1352e 583out:
36d5fe6a
ZK		 584 if (err)
585 skb_tx_error(skb);
c21ab2af
MP		 586 consume_skb(user_skb);
587 consume_skb(nskb);
588
ccb1352e
JG		 589 return err;
590}
591
ccb1352e
JG		 592static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
593{
594 struct ovs_header *ovs_header = info->userhdr;
7f8a436e 595 struct net *net = sock_net(skb->sk);
ccb1352e
JG		 596 struct nlattr **a = info->attrs;
597 struct sw_flow_actions *acts;
598 struct sk_buff *packet;
599 struct sw_flow *flow;
d98612b8 600 struct sw_flow_actions *sf_acts;
ccb1352e 601 struct datapath *dp;
83c8df26 602 struct vport *input_vport;
7f8a436e 603 u16 mru = 0;
bd1903b7 604 u64 hash;
ccb1352e
JG		 605 int len;
606 int err;
1ba39804 607 bool log = !a[OVS_PACKET_ATTR_PROBE];
ccb1352e
JG		 608
609 err = -EINVAL;
610 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
dded45fc 611 !a[OVS_PACKET_ATTR_ACTIONS])
ccb1352e
JG		 612 goto err;
613
614 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
615 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
616 err = -ENOMEM;
617 if (!packet)
618 goto err;
619 skb_reserve(packet, NET_IP_ALIGN);
620
32686a9d 621 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
ccb1352e 622
7f8a436e
JS		 623 /* Set packet's mru */
624 if (a[OVS_PACKET_ATTR_MRU]) {
625 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
626 packet->ignore_df = 1;
627 }
628 OVS_CB(packet)->mru = mru;
629
bd1903b7
TZ		 630 if (a[OVS_PACKET_ATTR_HASH]) {
631 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
632
633 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
634 !!(hash & OVS_PACKET_HASH_SW_BIT),
635 !!(hash & OVS_PACKET_HASH_L4_BIT));
636 }
637
ccb1352e 638 /* Build an sw_flow for sending this packet. */
23dabf88 639 flow = ovs_flow_alloc();
ccb1352e
JG		 640 err = PTR_ERR(flow);
641 if (IS_ERR(flow))
642 goto err_kfree_skb;
643
c2ac6673
JS		 644 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
645 packet, &flow->key, log);
ccb1352e
JG		 646 if (err)
647 goto err_flow_free;
648
7f8a436e 649 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
05da5898 650 &flow->key, &acts, log);
74f84a57
PS		 651 if (err)
652 goto err_flow_free;
ccb1352e 653
f5796684 654 rcu_assign_pointer(flow->sf_acts, acts);
ccb1352e 655 packet->priority = flow->key.phy.priority;
39c7caeb 656 packet->mark = flow->key.phy.skb_mark;
ccb1352e
JG		 657
658 rcu_read_lock();
7f8a436e 659 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
ccb1352e
JG		 660 err = -ENODEV;
661 if (!dp)
662 goto err_unlock;
663
83c8df26
PS		 664 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
665 if (!input_vport)
666 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
667
668 if (!input_vport)
669 goto err_unlock;
670
7f8a436e 671 packet->dev = input_vport->dev;
83c8df26 672 OVS_CB(packet)->input_vport = input_vport;
d98612b8 673 sf_acts = rcu_dereference(flow->sf_acts);
83c8df26 674
ccb1352e 675 local_bh_disable();
d98612b8 676 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
ccb1352e
JG		 677 local_bh_enable();
678 rcu_read_unlock();
679
03f0d916 680 ovs_flow_free(flow, false);
ccb1352e
JG		 681 return err;
682
683err_unlock:
684 rcu_read_unlock();
685err_flow_free:
03f0d916 686 ovs_flow_free(flow, false);
ccb1352e
JG		 687err_kfree_skb:
688 kfree_skb(packet);
689err:
690 return err;
691}
692
693static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
dded45fc 694 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
ccb1352e
JG		 695 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
696 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
1ba39804 697 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
7f8a436e 698 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
b5ab1f1b 699 [OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
ccb1352e
JG		 700};
701
66a9b928 702static const struct genl_small_ops dp_packet_genl_ops[] = {
ccb1352e 703 { .cmd = OVS_PACKET_CMD_EXECUTE,
ef6243ac 704 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 705 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 706 .doit = ovs_packet_cmd_execute
707 }
708};
709
56989f6d 710static struct genl_family dp_packet_genl_family __ro_after_init = {
0c200ef9
PS		 711 .hdrsize = sizeof(struct ovs_header),
712 .name = OVS_PACKET_FAMILY,
713 .version = OVS_PACKET_VERSION,
714 .maxattr = OVS_PACKET_ATTR_MAX,
3b0f31f2 715 .policy = packet_policy,
0c200ef9
PS		 716 .netnsok = true,
717 .parallel_ops = true,
66a9b928
JK		 718 .small_ops = dp_packet_genl_ops,
719 .n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
9c5d03d3 720 .resv_start_op = OVS_PACKET_CMD_EXECUTE + 1,
489111e5 721 .module = THIS_MODULE,
0c200ef9
PS		 722};
723
12eb18f7 724static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
1bd7116f 725 struct ovs_dp_megaflow_stats *mega_stats)
ccb1352e
JG		 726{
727 int i;
ccb1352e 728
1bd7116f
AZ		 729 memset(mega_stats, 0, sizeof(*mega_stats));
730
b637e498 731 stats->n_flows = ovs_flow_tbl_count(&dp->table);
1bd7116f 732 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
ccb1352e
JG		 733
734 stats->n_hit = stats->n_missed = stats->n_lost = 0;
1bd7116f 735
ccb1352e
JG		 736 for_each_possible_cpu(i) {
737 const struct dp_stats_percpu *percpu_stats;
738 struct dp_stats_percpu local_stats;
739 unsigned int start;
740
741 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
742
743 do {
d120d1a6 744 start = u64_stats_fetch_begin(&percpu_stats->syncp);
ccb1352e 745 local_stats = *percpu_stats;
d120d1a6 746 } while (u64_stats_fetch_retry(&percpu_stats->syncp, start));
ccb1352e
JG		 747
748 stats->n_hit += local_stats.n_hit;
749 stats->n_missed += local_stats.n_missed;
750 stats->n_lost += local_stats.n_lost;
1bd7116f 751 mega_stats->n_mask_hit += local_stats.n_mask_hit;
9d2f627b 752 mega_stats->n_cache_hit += local_stats.n_cache_hit;
ccb1352e
JG		 753 }
754}
755
74ed7ab9
JS		 756static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
757{
758 return ovs_identifier_is_ufid(sfid) &&
759 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
760}
761
762static bool should_fill_mask(uint32_t ufid_flags)
763{
764 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
765}
766
767static bool should_fill_actions(uint32_t ufid_flags)
c3ff8cfe 768{
74ed7ab9
JS		 769 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
770}
771
772static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
773 const struct sw_flow_id *sfid,
774 uint32_t ufid_flags)
775{
776 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
777
4e81c0b3
PA		 778 /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
779 * see ovs_nla_put_identifier()
780 */
74ed7ab9
JS		 781 if (sfid && ovs_identifier_is_ufid(sfid))
782 len += nla_total_size(sfid->ufid_len);
4e81c0b3
PA		 783 else
784 len += nla_total_size(ovs_key_attr_size());
74ed7ab9
JS		 785
786 /* OVS_FLOW_ATTR_KEY */
787 if (!sfid || should_fill_key(sfid, ufid_flags))
788 len += nla_total_size(ovs_key_attr_size());
789
790 /* OVS_FLOW_ATTR_MASK */
791 if (should_fill_mask(ufid_flags))
792 len += nla_total_size(ovs_key_attr_size());
793
794 /* OVS_FLOW_ATTR_ACTIONS */
795 if (should_fill_actions(ufid_flags))
8e2fed1c 796 len += nla_total_size(acts->orig_len);
74ed7ab9
JS		 797
798 return len
66c7a5ee 799 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
c3ff8cfe 800 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
66c7a5ee 801 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
c3ff8cfe
TG		 802}
803
ca7105f2
JS		 804/* Called with ovs_mutex or RCU read lock. */
805static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
806 struct sk_buff *skb)
807{
808 struct ovs_flow_stats stats;
809 __be16 tcp_flags;
810 unsigned long used;
ccb1352e 811
e298e505 812 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
0e9796b4 813
028d6a67 814 if (used &&
0238b720
ND		 815 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
816 OVS_FLOW_ATTR_PAD))
ca7105f2 817 return -EMSGSIZE;
ccb1352e 818
028d6a67 819 if (stats.n_packets &&
66c7a5ee
ND		 820 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
821 sizeof(struct ovs_flow_stats), &stats,
822 OVS_FLOW_ATTR_PAD))
ca7105f2 823 return -EMSGSIZE;
ccb1352e 824
e298e505
PS		 825 if ((u8)ntohs(tcp_flags) &&
826 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
ca7105f2
JS		 827 return -EMSGSIZE;
828
829 return 0;
830}
831
832/* Called with ovs_mutex or RCU read lock. */
833static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
834 struct sk_buff *skb, int skb_orig_len)
835{
836 struct nlattr *start;
837 int err;
ccb1352e
JG		 838
839 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
840 * this is the first flow to be dumped into 'skb'. This is unusual for
841 * Netlink but individual action lists can be longer than
842 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
843 * The userspace caller can always fetch the actions separately if it
844 * really wants them. (Most userspace callers in fact don't care.)
845 *
846 * This can only fail for dump operations because the skb is always
847 * properly sized for single flows.
848 */
ae0be8de 849 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
74f84a57 850 if (start) {
d57170b1
PS		 851 const struct sw_flow_actions *sf_acts;
852
663efa36 853 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
e6445719
PS		 854 err = ovs_nla_put_actions(sf_acts->actions,
855 sf_acts->actions_len, skb);
0e9796b4 856
74f84a57
PS		 857 if (!err)
858 nla_nest_end(skb, start);
859 else {
860 if (skb_orig_len)
ca7105f2 861 return err;
74f84a57
PS		 862
863 nla_nest_cancel(skb, start);
864 }
ca7105f2
JS		 865 } else if (skb_orig_len) {
866 return -EMSGSIZE;
867 }
868
869 return 0;
870}
871
872/* Called with ovs_mutex or RCU read lock. */
873static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
874 struct sk_buff *skb, u32 portid,
74ed7ab9 875 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
ca7105f2
JS		 876{
877 const int skb_orig_len = skb->len;
878 struct ovs_header *ovs_header;
879 int err;
880
881 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
882 flags, cmd);
883 if (!ovs_header)
884 return -EMSGSIZE;
885
886 ovs_header->dp_ifindex = dp_ifindex;
887
74ed7ab9 888 err = ovs_nla_put_identifier(flow, skb);
5b4237bb
JS		 889 if (err)
890 goto error;
891
74ed7ab9
JS		 892 if (should_fill_key(&flow->id, ufid_flags)) {
893 err = ovs_nla_put_masked_key(flow, skb);
894 if (err)
895 goto error;
896 }
897
898 if (should_fill_mask(ufid_flags)) {
899 err = ovs_nla_put_mask(flow, skb);
900 if (err)
901 goto error;
902 }
ca7105f2
JS		 903
904 err = ovs_flow_cmd_fill_stats(flow, skb);
905 if (err)
906 goto error;
907
74ed7ab9
JS		 908 if (should_fill_actions(ufid_flags)) {
909 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
910 if (err)
911 goto error;
912 }
ccb1352e 913
053c095a
JB		 914 genlmsg_end(skb, ovs_header);
915 return 0;
ccb1352e 916
ccb1352e
JG		 917error:
918 genlmsg_cancel(skb, ovs_header);
919 return err;
920}
921
0e9796b4
JR		 922/* May not be called with RCU read lock. */
923static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
74ed7ab9 924 const struct sw_flow_id *sfid,
fb5d1e9e 925 struct genl_info *info,
74ed7ab9
JS		 926 bool always,
927 uint32_t ufid_flags)
ccb1352e 928{
fb5d1e9e 929 struct sk_buff *skb;
74ed7ab9 930 size_t len;
ccb1352e 931
9b67aa4a 932 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
fb5d1e9e
JR		 933 return NULL;
934
74ed7ab9 935 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
551ddc05 936 skb = genlmsg_new(len, GFP_KERNEL);
fb5d1e9e
JR		 937 if (!skb)
938 return ERR_PTR(-ENOMEM);
939
940 return skb;
ccb1352e
JG		 941}
942
0e9796b4
JR		 943/* Called with ovs_mutex. */
944static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
945 int dp_ifindex,
946 struct genl_info *info, u8 cmd,
74ed7ab9 947 bool always, u32 ufid_flags)
ccb1352e
JG		 948{
949 struct sk_buff *skb;
950 int retval;
951
74ed7ab9
JS		 952 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
953 &flow->id, info, always, ufid_flags);
d0e992aa 954 if (IS_ERR_OR_NULL(skb))
fb5d1e9e 955 return skb;
ccb1352e 956
0e9796b4
JR		 957 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
958 info->snd_portid, info->snd_seq, 0,
74ed7ab9 959 cmd, ufid_flags);
8ffeb03f
PA		 960 if (WARN_ON_ONCE(retval < 0)) {
961 kfree_skb(skb);
962 skb = ERR_PTR(retval);
963 }
ccb1352e
JG		 964 return skb;
965}
966
37bdc87b 967static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
ccb1352e 968{
7f8a436e 969 struct net *net = sock_net(skb->sk);
ccb1352e
JG		 970 struct nlattr **a = info->attrs;
971 struct ovs_header *ovs_header = info->userhdr;
74ed7ab9 972 struct sw_flow *flow = NULL, *new_flow;
03f0d916 973 struct sw_flow_mask mask;
ccb1352e
JG		 974 struct sk_buff *reply;
975 struct datapath *dp;
68bb1010 976 struct sw_flow_key *key;
37bdc87b 977 struct sw_flow_actions *acts;
03f0d916 978 struct sw_flow_match match;
74ed7ab9 979 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
ccb1352e 980 int error;
05da5898 981 bool log = !a[OVS_FLOW_ATTR_PROBE];
ccb1352e 982
893f139b 983 /* Must have key and actions. */
ccb1352e 984 error = -EINVAL;
426cda5c 985 if (!a[OVS_FLOW_ATTR_KEY]) {
05da5898 986 OVS_NLERR(log, "Flow key attr not present in new flow.");
ccb1352e 987 goto error;
426cda5c
JG		 988 }
989 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
05da5898 990 OVS_NLERR(log, "Flow actions attr not present in new flow.");
893f139b 991 goto error;
426cda5c 992 }
03f0d916 993
893f139b
JR		 994 /* Most of the time we need to allocate a new flow, do it before
995 * locking.
996 */
997 new_flow = ovs_flow_alloc();
998 if (IS_ERR(new_flow)) {
999 error = PTR_ERR(new_flow);
1000 goto error;
1001 }
1002
1003 /* Extract key. */
68bb1010
EC		 1004 key = kzalloc(sizeof(*key), GFP_KERNEL);
1005 if (!key) {
1006 error = -ENOMEM;
0c598aed 1007 goto err_kfree_flow;
68bb1010
EC		 1008 }
1009
1010 ovs_match_init(&match, key, false, &mask);
c2ac6673 1011 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
05da5898 1012 a[OVS_FLOW_ATTR_MASK], log);
ccb1352e 1013 if (error)
0c598aed 1014 goto err_kfree_key;
ccb1352e 1015
68bb1010
EC		 1016 ovs_flow_mask_key(&new_flow->key, key, true, &mask);
1017
74ed7ab9
JS		 1018 /* Extract flow identifier. */
1019 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
68bb1010 1020 key, log);
74ed7ab9 1021 if (error)
0c598aed 1022 goto err_kfree_key;
74f84a57 1023
893f139b 1024 /* Validate actions. */
7f8a436e
JS		 1025 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
1026 &new_flow->key, &acts, log);
37bdc87b 1027 if (error) {
05da5898 1028 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
0c598aed 1029 goto err_kfree_key;
893f139b
JR		 1030 }
1031
74ed7ab9
JS		 1032 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1033 ufid_flags);
893f139b
JR		 1034 if (IS_ERR(reply)) {
1035 error = PTR_ERR(reply);
1036 goto err_kfree_acts;
ccb1352e
JG		 1037 }
1038
8e4e1713 1039 ovs_lock();
7f8a436e 1040 dp = get_dp(net, ovs_header->dp_ifindex);
893f139b
JR		 1041 if (unlikely(!dp)) {
1042 error = -ENODEV;
8e4e1713 1043 goto err_unlock_ovs;
893f139b 1044 }
74ed7ab9 1045
03f0d916 1046 /* Check if this is a duplicate flow */
74ed7ab9
JS		 1047 if (ovs_identifier_is_ufid(&new_flow->id))
1048 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1049 if (!flow)
68bb1010 1050 flow = ovs_flow_tbl_lookup(&dp->table, key);
893f139b
JR		 1051 if (likely(!flow)) {
1052 rcu_assign_pointer(new_flow->sf_acts, acts);
ccb1352e
JG		 1053
1054 /* Put flow in bucket. */
893f139b
JR		 1055 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1056 if (unlikely(error)) {
618ed0c8 1057 acts = NULL;
893f139b
JR		 1058 goto err_unlock_ovs;
1059 }
1060
1061 if (unlikely(reply)) {
1062 error = ovs_flow_cmd_fill_info(new_flow,
1063 ovs_header->dp_ifindex,
1064 reply, info->snd_portid,
1065 info->snd_seq, 0,
74ed7ab9
JS		 1066 OVS_FLOW_CMD_NEW,
1067 ufid_flags);
893f139b 1068 BUG_ON(error < 0);
618ed0c8 1069 }
893f139b 1070 ovs_unlock();
ccb1352e 1071 } else {
37bdc87b
JR		 1072 struct sw_flow_actions *old_acts;
1073
ccb1352e
JG		 1074 /* Bail out if we're not allowed to modify an existing flow.
1075 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1076 * because Generic Netlink treats the latter as a dump
1077 * request. We also accept NLM_F_EXCL in case that bug ever
1078 * gets fixed.
1079 */
893f139b
JR		 1080 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1081 | NLM_F_EXCL))) {
1082 error = -EEXIST;
8e4e1713 1083 goto err_unlock_ovs;
893f139b 1084 }
74ed7ab9
JS		 1085 /* The flow identifier has to be the same for flow updates.
1086 * Look for any overlapping flow.
1087 */
1088 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1089 if (ovs_identifier_is_key(&flow->id))
1090 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1091 &match);
1092 else /* UFID matches but key is different */
1093 flow = NULL;
4a46b24e
AW		 1094 if (!flow) {
1095 error = -ENOENT;
1096 goto err_unlock_ovs;
1097 }
893f139b 1098 }
37bdc87b
JR		 1099 /* Update actions. */
1100 old_acts = ovsl_dereference(flow->sf_acts);
1101 rcu_assign_pointer(flow->sf_acts, acts);
37bdc87b 1102
893f139b
JR		 1103 if (unlikely(reply)) {
1104 error = ovs_flow_cmd_fill_info(flow,
1105 ovs_header->dp_ifindex,
1106 reply, info->snd_portid,
1107 info->snd_seq, 0,
74ed7ab9
JS		 1108 OVS_FLOW_CMD_NEW,
1109 ufid_flags);
893f139b
JR		 1110 BUG_ON(error < 0);
1111 }
1112 ovs_unlock();
37bdc87b 1113
34ae932a 1114 ovs_nla_free_flow_actions_rcu(old_acts);
893f139b 1115 ovs_flow_free(new_flow, false);
37bdc87b 1116 }
893f139b
JR		 1117
1118 if (reply)
1119 ovs_notify(&dp_flow_genl_family, reply, info);
68bb1010
EC		 1120
1121 kfree(key);
37bdc87b
JR		 1122 return 0;
1123
37bdc87b
JR		 1124err_unlock_ovs:
1125 ovs_unlock();
893f139b
JR		 1126 kfree_skb(reply);
1127err_kfree_acts:
34ae932a 1128 ovs_nla_free_flow_actions(acts);
68bb1010
EC		 1129err_kfree_key:
1130 kfree(key);
0c598aed
FP		 1131err_kfree_flow:
1132 ovs_flow_free(new_flow, false);
37bdc87b
JR		 1133error:
1134 return error;
1135}
ccb1352e 1136
2fdb957d 1137/* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
cf3266ad
TZ		 1138static noinline_for_stack
1139struct sw_flow_actions *get_flow_actions(struct net *net,
1140 const struct nlattr *a,
1141 const struct sw_flow_key *key,
1142 const struct sw_flow_mask *mask,
1143 bool log)
6b205b2c
JG		 1144{
1145 struct sw_flow_actions *acts;
1146 struct sw_flow_key masked_key;
1147 int error;
1148
ae5f2fb1 1149 ovs_flow_mask_key(&masked_key, key, true, mask);
7f8a436e 1150 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
6b205b2c 1151 if (error) {
05da5898
JR		 1152 OVS_NLERR(log,
1153 "Actions may not be safe on all matching packets");
6b205b2c
JG		 1154 return ERR_PTR(error);
1155 }
1156
1157 return acts;
1158}
1159
9cc9a5cb
TZ		 1160/* Factor out match-init and action-copy to avoid
1161 * "Wframe-larger-than=1024" warning. Because mask is only
1162 * used to get actions, we new a function to save some
1163 * stack space.
1164 *
1165 * If there are not key and action attrs, we return 0
1166 * directly. In the case, the caller will also not use the
1167 * match as before. If there is action attr, we try to get
1168 * actions and save them to *acts. Before returning from
1169 * the function, we reset the match->mask pointer. Because
1170 * we should not to return match object with dangling reference
1171 * to mask.
1172 * */
26063790
AB		 1173static noinline_for_stack int
1174ovs_nla_init_match_and_action(struct net *net,
1175 struct sw_flow_match *match,
1176 struct sw_flow_key *key,
1177 struct nlattr **a,
1178 struct sw_flow_actions **acts,
1179 bool log)
9cc9a5cb
TZ		 1180{
1181 struct sw_flow_mask mask;
1182 int error = 0;
1183
1184 if (a[OVS_FLOW_ATTR_KEY]) {
1185 ovs_match_init(match, key, true, &mask);
1186 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1187 a[OVS_FLOW_ATTR_MASK], log);
1188 if (error)
1189 goto error;
1190 }
1191
1192 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1193 if (!a[OVS_FLOW_ATTR_KEY]) {
1194 OVS_NLERR(log,
1195 "Flow key attribute not present in set flow.");
5829e62a
CJ		 1196 error = -EINVAL;
1197 goto error;
9cc9a5cb
TZ		 1198 }
1199
1200 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1201 &mask, log);
1202 if (IS_ERR(*acts)) {
1203 error = PTR_ERR(*acts);
1204 goto error;
1205 }
1206 }
1207
1208 /* On success, error is 0. */
1209error:
1210 match->mask = NULL;
1211 return error;
1212}
1213
37bdc87b
JR		 1214static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1215{
7f8a436e 1216 struct net *net = sock_net(skb->sk);
37bdc87b
JR		 1217 struct nlattr **a = info->attrs;
1218 struct ovs_header *ovs_header = info->userhdr;
6b205b2c 1219 struct sw_flow_key key;
37bdc87b 1220 struct sw_flow *flow;
37bdc87b
JR		 1221 struct sk_buff *reply = NULL;
1222 struct datapath *dp;
893f139b 1223 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
37bdc87b 1224 struct sw_flow_match match;
74ed7ab9
JS		 1225 struct sw_flow_id sfid;
1226 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
6f15cdbf 1227 int error = 0;
05da5898 1228 bool log = !a[OVS_FLOW_ATTR_PROBE];
74ed7ab9 1229 bool ufid_present;
37bdc87b 1230
74ed7ab9 1231 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
9cc9a5cb 1232 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
6f15cdbf
SG		 1233 OVS_NLERR(log,
1234 "Flow set message rejected, Key attribute missing.");
9cc9a5cb 1235 return -EINVAL;
6f15cdbf 1236 }
9cc9a5cb
TZ		 1237
1238 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1239 &acts, log);
37bdc87b
JR		 1240 if (error)
1241 goto error;
1242
9cc9a5cb 1243 if (acts) {
2fdb957d 1244 /* Can allocate before locking if have acts. */
74ed7ab9
JS		 1245 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1246 ufid_flags);
893f139b
JR		 1247 if (IS_ERR(reply)) {
1248 error = PTR_ERR(reply);
1249 goto err_kfree_acts;
be52c9e9 1250 }
37bdc87b 1251 }
0e9796b4 1252
37bdc87b 1253 ovs_lock();
7f8a436e 1254 dp = get_dp(net, ovs_header->dp_ifindex);
893f139b
JR		 1255 if (unlikely(!dp)) {
1256 error = -ENODEV;
37bdc87b 1257 goto err_unlock_ovs;
893f139b 1258 }
37bdc87b 1259 /* Check that the flow exists. */
74ed7ab9
JS		 1260 if (ufid_present)
1261 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1262 else
1263 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
893f139b
JR		 1264 if (unlikely(!flow)) {
1265 error = -ENOENT;
37bdc87b 1266 goto err_unlock_ovs;
893f139b 1267 }
4a46b24e 1268
37bdc87b 1269 /* Update actions, if present. */
893f139b 1270 if (likely(acts)) {
37bdc87b
JR		 1271 old_acts = ovsl_dereference(flow->sf_acts);
1272 rcu_assign_pointer(flow->sf_acts, acts);
893f139b
JR		 1273
1274 if (unlikely(reply)) {
1275 error = ovs_flow_cmd_fill_info(flow,
1276 ovs_header->dp_ifindex,
1277 reply, info->snd_portid,
1278 info->snd_seq, 0,
804fe108 1279 OVS_FLOW_CMD_SET,
74ed7ab9 1280 ufid_flags);
893f139b
JR		 1281 BUG_ON(error < 0);
1282 }
1283 } else {
1284 /* Could not alloc without acts before locking. */
1285 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
804fe108 1286 info, OVS_FLOW_CMD_SET, false,
74ed7ab9
JS		 1287 ufid_flags);
1288
b5ffe634 1289 if (IS_ERR(reply)) {
893f139b
JR		 1290 error = PTR_ERR(reply);
1291 goto err_unlock_ovs;
1292 }
ccb1352e 1293 }
37bdc87b 1294
37bdc87b
JR		 1295 /* Clear stats. */
1296 if (a[OVS_FLOW_ATTR_CLEAR])
1297 ovs_flow_stats_clear(flow);
8e4e1713 1298 ovs_unlock();
ccb1352e 1299
893f139b
JR		 1300 if (reply)
1301 ovs_notify(&dp_flow_genl_family, reply, info);
1302 if (old_acts)
34ae932a 1303 ovs_nla_free_flow_actions_rcu(old_acts);
fb5d1e9e 1304
ccb1352e
JG		 1305 return 0;
1306
8e4e1713
PS		 1307err_unlock_ovs:
1308 ovs_unlock();
893f139b
JR		 1309 kfree_skb(reply);
1310err_kfree_acts:
34ae932a 1311 ovs_nla_free_flow_actions(acts);
ccb1352e
JG		 1312error:
1313 return error;
1314}
1315
1316static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1317{
1318 struct nlattr **a = info->attrs;
1319 struct ovs_header *ovs_header = info->userhdr;
c2ac6673 1320 struct net *net = sock_net(skb->sk);
ccb1352e
JG		 1321 struct sw_flow_key key;
1322 struct sk_buff *reply;
1323 struct sw_flow *flow;
1324 struct datapath *dp;
03f0d916 1325 struct sw_flow_match match;
74ed7ab9
JS		 1326 struct sw_flow_id ufid;
1327 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1328 int err = 0;
05da5898 1329 bool log = !a[OVS_FLOW_ATTR_PROBE];
74ed7ab9 1330 bool ufid_present;
ccb1352e 1331
74ed7ab9
JS		 1332 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1333 if (a[OVS_FLOW_ATTR_KEY]) {
2279994d 1334 ovs_match_init(&match, &key, true, NULL);
c2ac6673 1335 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
74ed7ab9
JS		 1336 log);
1337 } else if (!ufid_present) {
05da5898
JR		 1338 OVS_NLERR(log,
1339 "Flow get message rejected, Key attribute missing.");
74ed7ab9 1340 err = -EINVAL;
03f0d916 1341 }
ccb1352e
JG		 1342 if (err)
1343 return err;
1344
8e4e1713 1345 ovs_lock();
46df7b81 1346 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
8e4e1713
PS		 1347 if (!dp) {
1348 err = -ENODEV;
1349 goto unlock;
1350 }
ccb1352e 1351
74ed7ab9
JS		 1352 if (ufid_present)
1353 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1354 else
1355 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
4a46b24e 1356 if (!flow) {
8e4e1713
PS		 1357 err = -ENOENT;
1358 goto unlock;
1359 }
ccb1352e 1360
0e9796b4 1361 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
804fe108 1362 OVS_FLOW_CMD_GET, true, ufid_flags);
8e4e1713
PS		 1363 if (IS_ERR(reply)) {
1364 err = PTR_ERR(reply);
1365 goto unlock;
1366 }
ccb1352e 1367
8e4e1713 1368 ovs_unlock();
ccb1352e 1369 return genlmsg_reply(reply, info);
8e4e1713
PS		 1370unlock:
1371 ovs_unlock();
1372 return err;
ccb1352e
JG		 1373}
1374
1375static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1376{
1377 struct nlattr **a = info->attrs;
1378 struct ovs_header *ovs_header = info->userhdr;
c2ac6673 1379 struct net *net = sock_net(skb->sk);
ccb1352e
JG		 1380 struct sw_flow_key key;
1381 struct sk_buff *reply;
74ed7ab9 1382 struct sw_flow *flow = NULL;
ccb1352e 1383 struct datapath *dp;
03f0d916 1384 struct sw_flow_match match;
74ed7ab9
JS		 1385 struct sw_flow_id ufid;
1386 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
ccb1352e 1387 int err;
05da5898 1388 bool log = !a[OVS_FLOW_ATTR_PROBE];
74ed7ab9 1389 bool ufid_present;
ccb1352e 1390
74ed7ab9
JS		 1391 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1392 if (a[OVS_FLOW_ATTR_KEY]) {
2279994d 1393 ovs_match_init(&match, &key, true, NULL);
c2ac6673
JS		 1394 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1395 NULL, log);
aed06778
JR		 1396 if (unlikely(err))
1397 return err;
1398 }
1399
8e4e1713 1400 ovs_lock();
46df7b81 1401 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
aed06778 1402 if (unlikely(!dp)) {
8e4e1713
PS		 1403 err = -ENODEV;
1404 goto unlock;
1405 }
46df7b81 1406
74ed7ab9 1407 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
b637e498 1408 err = ovs_flow_tbl_flush(&dp->table);
8e4e1713
PS		 1409 goto unlock;
1410 }
03f0d916 1411
74ed7ab9
JS		 1412 if (ufid_present)
1413 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1414 else
1415 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
4a46b24e 1416 if (unlikely(!flow)) {
8e4e1713
PS		 1417 err = -ENOENT;
1418 goto unlock;
1419 }
ccb1352e 1420
b637e498 1421 ovs_flow_tbl_remove(&dp->table, flow);
aed06778 1422 ovs_unlock();
ccb1352e 1423
aed06778 1424 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
74ed7ab9 1425 &flow->id, info, false, ufid_flags);
aed06778 1426 if (likely(reply)) {
b90f5aa4 1427 if (!IS_ERR(reply)) {
aed06778
JR		 1428 rcu_read_lock(); /*To keep RCU checker happy. */
1429 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1430 reply, info->snd_portid,
1431 info->snd_seq, 0,
74ed7ab9
JS		 1432 OVS_FLOW_CMD_DEL,
1433 ufid_flags);
aed06778 1434 rcu_read_unlock();
8a574f86
PA		 1435 if (WARN_ON_ONCE(err < 0)) {
1436 kfree_skb(reply);
1437 goto out_free;
1438 }
aed06778
JR		 1439
1440 ovs_notify(&dp_flow_genl_family, reply, info);
1441 } else {
cf3266ad
TZ		 1442 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1443 PTR_ERR(reply));
aed06778 1444 }
fb5d1e9e 1445 }
ccb1352e 1446
8a574f86 1447out_free:
aed06778 1448 ovs_flow_free(flow, true);
ccb1352e 1449 return 0;
8e4e1713
PS		 1450unlock:
1451 ovs_unlock();
1452 return err;
ccb1352e
JG		 1453}
1454
1455static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1456{
74ed7ab9 1457 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
ccb1352e 1458 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
b637e498 1459 struct table_instance *ti;
ccb1352e 1460 struct datapath *dp;
74ed7ab9
JS		 1461 u32 ufid_flags;
1462 int err;
1463
8cb08174
JB		 1464 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1465 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
74ed7ab9
JS		 1466 if (err)
1467 return err;
1468 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
ccb1352e 1469
d57170b1 1470 rcu_read_lock();
cc3a5ae6 1471 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
8e4e1713 1472 if (!dp) {
d57170b1 1473 rcu_read_unlock();
ccb1352e 1474 return -ENODEV;
8e4e1713 1475 }
ccb1352e 1476
b637e498 1477 ti = rcu_dereference(dp->table.ti);
ccb1352e
JG		 1478 for (;;) {
1479 struct sw_flow *flow;
1480 u32 bucket, obj;
1481
1482 bucket = cb->args[0];
1483 obj = cb->args[1];
b637e498 1484 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
ccb1352e
JG		 1485 if (!flow)
1486 break;
1487
0e9796b4 1488 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
15e47304 1489 NETLINK_CB(cb->skb).portid,
ccb1352e 1490 cb->nlh->nlmsg_seq, NLM_F_MULTI,
804fe108 1491 OVS_FLOW_CMD_GET, ufid_flags) < 0)
ccb1352e
JG		 1492 break;
1493
1494 cb->args[0] = bucket;
1495 cb->args[1] = obj;
1496 }
d57170b1 1497 rcu_read_unlock();
ccb1352e
JG		 1498 return skb->len;
1499}
1500
0c200ef9
PS		 1501static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1502 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
05da5898 1503 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
0c200ef9
PS		 1504 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1505 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
05da5898 1506 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
74ed7ab9
JS		 1507 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1508 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
0c200ef9
PS		 1509};
1510
66a9b928 1511static const struct genl_small_ops dp_flow_genl_ops[] = {
ccb1352e 1512 { .cmd = OVS_FLOW_CMD_NEW,
ef6243ac 1513 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1514 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
37bdc87b 1515 .doit = ovs_flow_cmd_new
ccb1352e
JG		 1516 },
1517 { .cmd = OVS_FLOW_CMD_DEL,
ef6243ac 1518 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1519 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 1520 .doit = ovs_flow_cmd_del
1521 },
1522 { .cmd = OVS_FLOW_CMD_GET,
ef6243ac 1523 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
ccb1352e 1524 .flags = 0, /* OK for unprivileged users. */
ccb1352e
JG		 1525 .doit = ovs_flow_cmd_get,
1526 .dumpit = ovs_flow_cmd_dump
1527 },
1528 { .cmd = OVS_FLOW_CMD_SET,
ef6243ac 1529 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1530 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
37bdc87b 1531 .doit = ovs_flow_cmd_set,
ccb1352e
JG		 1532 },
1533};
1534
56989f6d 1535static struct genl_family dp_flow_genl_family __ro_after_init = {
ccb1352e 1536 .hdrsize = sizeof(struct ovs_header),
0c200ef9
PS		 1537 .name = OVS_FLOW_FAMILY,
1538 .version = OVS_FLOW_VERSION,
1539 .maxattr = OVS_FLOW_ATTR_MAX,
3b0f31f2 1540 .policy = flow_policy,
3a4e0d6a
PS		 1541 .netnsok = true,
1542 .parallel_ops = true,
66a9b928
JK		 1543 .small_ops = dp_flow_genl_ops,
1544 .n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
9c5d03d3 1545 .resv_start_op = OVS_FLOW_CMD_SET + 1,
0c200ef9
PS		 1546 .mcgrps = &ovs_dp_flow_multicast_group,
1547 .n_mcgrps = 1,
489111e5 1548 .module = THIS_MODULE,
ccb1352e
JG		 1549};
1550
c3ff8cfe
TG		 1551static size_t ovs_dp_cmd_msg_size(void)
1552{
1553 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1554
1555 msgsize += nla_total_size(IFNAMSIZ);
66c7a5ee
ND		 1556 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1557 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
45fb9c35 1558 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
9bf24f59 1559 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
347541e2 1560 msgsize += nla_total_size(sizeof(u32) * nr_cpu_ids); /* OVS_DP_ATTR_PER_CPU_PIDS */
c3ff8cfe
TG		 1561
1562 return msgsize;
1563}
1564
8ec609d8 1565/* Called with ovs_mutex. */
ccb1352e 1566static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
15e47304 1567 u32 portid, u32 seq, u32 flags, u8 cmd)
ccb1352e
JG		 1568{
1569 struct ovs_header *ovs_header;
1570 struct ovs_dp_stats dp_stats;
1bd7116f 1571 struct ovs_dp_megaflow_stats dp_megaflow_stats;
347541e2
AZ		 1572 struct dp_nlsk_pids *pids = ovsl_dereference(dp->upcall_portids);
1573 int err, pids_len;
ccb1352e 1574
15e47304 1575 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
cf3266ad 1576 flags, cmd);
ccb1352e
JG		 1577 if (!ovs_header)
1578 goto error;
1579
1580 ovs_header->dp_ifindex = get_dpifindex(dp);
1581
ccb1352e 1582 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
ccb1352e
JG		 1583 if (err)
1584 goto nla_put_failure;
1585
1bd7116f 1586 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
66c7a5ee
ND		 1587 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1588 &dp_stats, OVS_DP_ATTR_PAD))
1bd7116f
AZ		 1589 goto nla_put_failure;
1590
66c7a5ee
ND		 1591 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1592 sizeof(struct ovs_dp_megaflow_stats),
1593 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
028d6a67 1594 goto nla_put_failure;
ccb1352e 1595
43d4be9c
TG		 1596 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1597 goto nla_put_failure;
1598
9bf24f59
EC		 1599 if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1600 ovs_flow_tbl_masks_cache_size(&dp->table)))
1601 goto nla_put_failure;
1602
347541e2
AZ		 1603 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU && pids) {
1604 pids_len = min(pids->n_pids, nr_cpu_ids) * sizeof(u32);
1605 if (nla_put(skb, OVS_DP_ATTR_PER_CPU_PIDS, pids_len, &pids->pids))
1606 goto nla_put_failure;
1607 }
1608
053c095a
JB		 1609 genlmsg_end(skb, ovs_header);
1610 return 0;
ccb1352e
JG		 1611
1612nla_put_failure:
1613 genlmsg_cancel(skb, ovs_header);
1614error:
1615 return -EMSGSIZE;
1616}
1617
263ea090 1618static struct sk_buff *ovs_dp_cmd_alloc_info(void)
ccb1352e 1619{
551ddc05 1620 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
ccb1352e
JG		 1621}
1622
bb6f9a70 1623/* Called with rcu_read_lock or ovs_mutex. */
46df7b81 1624static struct datapath *lookup_datapath(struct net *net,
12eb18f7 1625 const struct ovs_header *ovs_header,
ccb1352e
JG		 1626 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1627{
1628 struct datapath *dp;
1629
1630 if (!a[OVS_DP_ATTR_NAME])
46df7b81 1631 dp = get_dp(net, ovs_header->dp_ifindex);
ccb1352e
JG		 1632 else {
1633 struct vport *vport;
1634
46df7b81 1635 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
ccb1352e 1636 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
ccb1352e
JG		 1637 }
1638 return dp ? dp : ERR_PTR(-ENODEV);
1639}
1640
cf3266ad
TZ		 1641static void ovs_dp_reset_user_features(struct sk_buff *skb,
1642 struct genl_info *info)
44da5ae5
TG		 1643{
1644 struct datapath *dp;
1645
cf3266ad
TZ		 1646 dp = lookup_datapath(sock_net(skb->sk), info->userhdr,
1647 info->attrs);
3c7eacfc 1648 if (IS_ERR(dp))
44da5ae5
TG		 1649 return;
1650
fd954cc1
AC		 1651 pr_warn("%s: Dropping previously announced user features\n",
1652 ovs_dp_name(dp));
44da5ae5
TG		 1653 dp->user_features = 0;
1654}
1655
b83d23a2
MG		 1656static int ovs_dp_set_upcall_portids(struct datapath *dp,
1657 const struct nlattr *ids)
1658{
1659 struct dp_nlsk_pids *old, *dp_nlsk_pids;
1660
1661 if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
1662 return -EINVAL;
1663
1664 old = ovsl_dereference(dp->upcall_portids);
1665
1666 dp_nlsk_pids = kmalloc(sizeof(*dp_nlsk_pids) + nla_len(ids),
1667 GFP_KERNEL);
1668 if (!dp_nlsk_pids)
1669 return -ENOMEM;
1670
1671 dp_nlsk_pids->n_pids = nla_len(ids) / sizeof(u32);
1672 nla_memcpy(dp_nlsk_pids->pids, ids, nla_len(ids));
1673
1674 rcu_assign_pointer(dp->upcall_portids, dp_nlsk_pids);
1675
1676 kfree_rcu(old, rcu);
1677
1678 return 0;
1679}
1680
1681u32 ovs_dp_get_upcall_portid(const struct datapath *dp, uint32_t cpu_id)
1682{
1683 struct dp_nlsk_pids *dp_nlsk_pids;
1684
1685 dp_nlsk_pids = rcu_dereference(dp->upcall_portids);
1686
1687 if (dp_nlsk_pids) {
1688 if (cpu_id < dp_nlsk_pids->n_pids) {
1689 return dp_nlsk_pids->pids[cpu_id];
784dcfa5
MG		 1690 } else if (dp_nlsk_pids->n_pids > 0 &&
1691 cpu_id >= dp_nlsk_pids->n_pids) {
1692 /* If the number of netlink PIDs is mismatched with
1693 * the number of CPUs as seen by the kernel, log this
1694 * and send the upcall to an arbitrary socket (0) in
1695 * order to not drop packets
b83d23a2
MG		 1696 */
1697 pr_info_ratelimited("cpu_id mismatch with handler threads");
784dcfa5
MG		 1698 return dp_nlsk_pids->pids[cpu_id %
1699 dp_nlsk_pids->n_pids];
b83d23a2
MG		 1700 } else {
1701 return 0;
1702 }
1703 } else {
1704 return 0;
1705 }
1706}
1707
95a7233c 1708static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
43d4be9c 1709{
35d39fec 1710 u32 user_features = 0, old_features = dp->user_features;
b83d23a2 1711 int err;
95a7233c
PB		 1712
1713 if (a[OVS_DP_ATTR_USER_FEATURES]) {
1714 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1715
1716 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1717 OVS_DP_F_UNALIGNED |
b83d23a2
MG		 1718 OVS_DP_F_TC_RECIRC_SHARING |
1719 OVS_DP_F_DISPATCH_UPCALL_PER_CPU))
95a7233c
PB		 1720 return -EOPNOTSUPP;
1721
1722#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1723 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1724 return -EOPNOTSUPP;
1725#endif
1726 }
1727
9bf24f59
EC		 1728 if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1729 int err;
1730 u32 cache_size;
1731
1732 cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1733 err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1734 if (err)
1735 return err;
1736 }
1737
95a7233c
PB		 1738 dp->user_features = user_features;
1739
b83d23a2
MG		 1740 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU &&
1741 a[OVS_DP_ATTR_PER_CPU_PIDS]) {
1742 /* Upcall Netlink Port IDs have been updated */
1743 err = ovs_dp_set_upcall_portids(dp,
1744 a[OVS_DP_ATTR_PER_CPU_PIDS]);
1745 if (err)
1746 return err;
1747 }
1748
35d39fec
PB		 1749 if ((dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1750 !(old_features & OVS_DP_F_TC_RECIRC_SHARING))
1751 tc_skb_ext_tc_enable();
1752 else if (!(dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1753 (old_features & OVS_DP_F_TC_RECIRC_SHARING))
1754 tc_skb_ext_tc_disable();
95a7233c
PB		 1755
1756 return 0;
43d4be9c
TG		 1757}
1758
eec62ead
TZ		 1759static int ovs_dp_stats_init(struct datapath *dp)
1760{
1761 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1762 if (!dp->stats_percpu)
1763 return -ENOMEM;
1764
1765 return 0;
1766}
1767
1768static int ovs_dp_vport_init(struct datapath *dp)
1769{
1770 int i;
1771
1772 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1773 sizeof(struct hlist_head),
1774 GFP_KERNEL);
1775 if (!dp->ports)
1776 return -ENOMEM;
1777
1778 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1779 INIT_HLIST_HEAD(&dp->ports[i]);
1780
1781 return 0;
1782}
1783
ccb1352e
JG		 1784static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1785{
1786 struct nlattr **a = info->attrs;
1787 struct vport_parms parms;
1788 struct sk_buff *reply;
1789 struct datapath *dp;
1790 struct vport *vport;
46df7b81 1791 struct ovs_net *ovs_net;
eec62ead 1792 int err;
ccb1352e
JG		 1793
1794 err = -EINVAL;
1795 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1796 goto err;
1797
263ea090 1798 reply = ovs_dp_cmd_alloc_info();
6093ae9a
JR		 1799 if (!reply)
1800 return -ENOMEM;
ccb1352e
JG		 1801
1802 err = -ENOMEM;
1803 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1804 if (dp == NULL)
eec62ead 1805 goto err_destroy_reply;
46df7b81 1806
efd7ef1c 1807 ovs_dp_set_net(dp, sock_net(skb->sk));
ccb1352e
JG		 1808
1809 /* Allocate table. */
b637e498
PS		 1810 err = ovs_flow_tbl_init(&dp->table);
1811 if (err)
eec62ead 1812 goto err_destroy_dp;
ccb1352e 1813
eec62ead
TZ		 1814 err = ovs_dp_stats_init(dp);
1815 if (err)
ccb1352e 1816 goto err_destroy_table;
ccb1352e 1817
eec62ead
TZ		 1818 err = ovs_dp_vport_init(dp);
1819 if (err)
1820 goto err_destroy_stats;
15eac2a7 1821
96fbc13d
AZ		 1822 err = ovs_meters_init(dp);
1823 if (err)
eec62ead 1824 goto err_destroy_ports;
96fbc13d 1825
ccb1352e
JG		 1826 /* Set up our datapath device. */
1827 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1828 parms.type = OVS_VPORT_TYPE_INTERNAL;
1829 parms.options = NULL;
1830 parms.dp = dp;
1831 parms.port_no = OVSP_LOCAL;
5cd667b0 1832 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
54c4ef34
AZ		 1833 parms.desired_ifindex = a[OVS_DP_ATTR_IFINDEX]
1834 ? nla_get_u32(a[OVS_DP_ATTR_IFINDEX]) : 0;
ccb1352e 1835
6093ae9a
JR		 1836 /* So far only local changes have been made, now need the lock. */
1837 ovs_lock();
1838
fea07a48
EC		 1839 err = ovs_dp_change(dp, a);
1840 if (err)
1841 goto err_unlock_and_destroy_meters;
1842
ccb1352e
JG		 1843 vport = new_vport(&parms);
1844 if (IS_ERR(vport)) {
1845 err = PTR_ERR(vport);
1846 if (err == -EBUSY)
1847 err = -EEXIST;
1848
44da5ae5
TG		 1849 if (err == -EEXIST) {
1850 /* An outdated user space instance that does not understand
1851 * the concept of user_features has attempted to create a new
1852 * datapath and is likely to reuse it. Drop all user features.
1853 */
1854 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1855 ovs_dp_reset_user_features(skb, info);
1856 }
1857
a87406f4 1858 goto err_destroy_portids;
ccb1352e
JG		 1859 }
1860
1933ea36 1861 vport->upcall_stats = netdev_alloc_pcpu_stats(struct vport_upcall_stats_percpu);
1862 if (!vport->upcall_stats) {
1863 err = -ENOMEM;
95637d91 1864 goto err_destroy_vport;
1933ea36 1865 }
1866
6093ae9a
JR		 1867 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1868 info->snd_seq, 0, OVS_DP_CMD_NEW);
1869 BUG_ON(err < 0);
ccb1352e 1870
46df7b81 1871 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
59a35d60 1872 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
8e4e1713
PS		 1873
1874 ovs_unlock();
ccb1352e 1875
2a94fe48 1876 ovs_notify(&dp_datapath_genl_family, reply, info);
ccb1352e
JG		 1877 return 0;
1878
95637d91
AC		 1879err_destroy_vport:
1880 ovs_dp_detach_port(vport);
a87406f4
AZ		 1881err_destroy_portids:
1882 kfree(rcu_dereference_raw(dp->upcall_portids));
fea07a48
EC		 1883err_unlock_and_destroy_meters:
1884 ovs_unlock();
96fbc13d 1885 ovs_meters_exit(dp);
eec62ead 1886err_destroy_ports:
15eac2a7 1887 kfree(dp->ports);
eec62ead 1888err_destroy_stats:
ccb1352e
JG		 1889 free_percpu(dp->stats_percpu);
1890err_destroy_table:
9b996e54 1891 ovs_flow_tbl_destroy(&dp->table);
eec62ead 1892err_destroy_dp:
ccb1352e 1893 kfree(dp);
eec62ead 1894err_destroy_reply:
6093ae9a 1895 kfree_skb(reply);
ccb1352e
JG		 1896err:
1897 return err;
1898}
1899
8e4e1713 1900/* Called with ovs_mutex. */
46df7b81 1901static void __dp_destroy(struct datapath *dp)
ccb1352e 1902{
1f3a090b 1903 struct flow_table *table = &dp->table;
15eac2a7 1904 int i;
ccb1352e 1905
35d39fec
PB		 1906 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1907 tc_skb_ext_tc_disable();
1908
15eac2a7
PS		 1909 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1910 struct vport *vport;
b67bfe0d 1911 struct hlist_node *n;
15eac2a7 1912
b67bfe0d 1913 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
15eac2a7
PS		 1914 if (vport->port_no != OVSP_LOCAL)
1915 ovs_dp_detach_port(vport);
1916 }
ccb1352e 1917
59a35d60 1918 list_del_rcu(&dp->list_node);
ccb1352e 1919
8e4e1713 1920 /* OVSP_LOCAL is datapath internal port. We need to make sure that
e80857cc 1921 * all ports in datapath are destroyed first before freeing datapath.
ccb1352e 1922 */
8e4e1713 1923 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
ccb1352e 1924
1f3a090b
TZ		 1925 /* Flush sw_flow in the tables. RCU cb only releases resource
1926 * such as dp, ports and tables. That may avoid some issues
1927 * such as RCU usage warning.
1928 */
1929 table_instance_flow_flush(table, ovsl_dereference(table->ti),
1930 ovsl_dereference(table->ufid_ti));
1931
1932 /* RCU destroy the ports, meters and flow tables. */
ccb1352e 1933 call_rcu(&dp->rcu, destroy_dp_rcu);
46df7b81
PS		 1934}
1935
1936static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1937{
1938 struct sk_buff *reply;
1939 struct datapath *dp;
1940 int err;
1941
263ea090 1942 reply = ovs_dp_cmd_alloc_info();
6093ae9a
JR		 1943 if (!reply)
1944 return -ENOMEM;
1945
8e4e1713 1946 ovs_lock();
46df7b81
PS		 1947 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1948 err = PTR_ERR(dp);
1949 if (IS_ERR(dp))
6093ae9a 1950 goto err_unlock_free;
46df7b81 1951
6093ae9a
JR		 1952 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1953 info->snd_seq, 0, OVS_DP_CMD_DEL);
1954 BUG_ON(err < 0);
46df7b81
PS		 1955
1956 __dp_destroy(dp);
8e4e1713 1957 ovs_unlock();
ccb1352e 1958
2a94fe48 1959 ovs_notify(&dp_datapath_genl_family, reply, info);
ccb1352e
JG		 1960
1961 return 0;
6093ae9a
JR		 1962
1963err_unlock_free:
8e4e1713 1964 ovs_unlock();
6093ae9a 1965 kfree_skb(reply);
8e4e1713 1966 return err;
ccb1352e
JG		 1967}
1968
1969static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1970{
1971 struct sk_buff *reply;
1972 struct datapath *dp;
1973 int err;
1974
263ea090 1975 reply = ovs_dp_cmd_alloc_info();
6093ae9a
JR		 1976 if (!reply)
1977 return -ENOMEM;
1978
8e4e1713 1979 ovs_lock();
46df7b81 1980 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
8e4e1713 1981 err = PTR_ERR(dp);
ccb1352e 1982 if (IS_ERR(dp))
6093ae9a 1983 goto err_unlock_free;
ccb1352e 1984
95a7233c
PB		 1985 err = ovs_dp_change(dp, info->attrs);
1986 if (err)
1987 goto err_unlock_free;
43d4be9c 1988
6093ae9a 1989 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
804fe108 1990 info->snd_seq, 0, OVS_DP_CMD_SET);
6093ae9a 1991 BUG_ON(err < 0);
ccb1352e 1992
8e4e1713 1993 ovs_unlock();
2a94fe48 1994 ovs_notify(&dp_datapath_genl_family, reply, info);
ccb1352e
JG		 1995
1996 return 0;
6093ae9a
JR		 1997
1998err_unlock_free:
8e4e1713 1999 ovs_unlock();
6093ae9a 2000 kfree_skb(reply);
8e4e1713 2001 return err;
ccb1352e
JG		 2002}
2003
2004static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
2005{
2006 struct sk_buff *reply;
2007 struct datapath *dp;
8e4e1713 2008 int err;
ccb1352e 2009
263ea090 2010 reply = ovs_dp_cmd_alloc_info();
6093ae9a
JR		 2011 if (!reply)
2012 return -ENOMEM;
2013
8ec609d8 2014 ovs_lock();
46df7b81 2015 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
8e4e1713
PS		 2016 if (IS_ERR(dp)) {
2017 err = PTR_ERR(dp);
6093ae9a 2018 goto err_unlock_free;
8e4e1713 2019 }
6093ae9a 2020 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
804fe108 2021 info->snd_seq, 0, OVS_DP_CMD_GET);
6093ae9a 2022 BUG_ON(err < 0);
8ec609d8 2023 ovs_unlock();
ccb1352e
JG		 2024
2025 return genlmsg_reply(reply, info);
8e4e1713 2026
6093ae9a 2027err_unlock_free:
8ec609d8 2028 ovs_unlock();
6093ae9a 2029 kfree_skb(reply);
8e4e1713 2030 return err;
ccb1352e
JG		 2031}
2032
2033static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2034{
46df7b81 2035 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
ccb1352e
JG		 2036 struct datapath *dp;
2037 int skip = cb->args[0];
2038 int i = 0;
2039
8ec609d8
PS		 2040 ovs_lock();
2041 list_for_each_entry(dp, &ovs_net->dps, list_node) {
77676fdb 2042 if (i >= skip &&
15e47304 2043 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
ccb1352e 2044 cb->nlh->nlmsg_seq, NLM_F_MULTI,
804fe108 2045 OVS_DP_CMD_GET) < 0)
ccb1352e
JG		 2046 break;
2047 i++;
2048 }
8ec609d8 2049 ovs_unlock();
ccb1352e
JG		 2050
2051 cb->args[0] = i;
2052
2053 return skb->len;
2054}
2055
0c200ef9
PS		 2056static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
2057 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2058 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2059 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
9bf24f59
EC		 2060 [OVS_DP_ATTR_MASKS_CACHE_SIZE] = NLA_POLICY_RANGE(NLA_U32, 0,
2061 PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
54c4ef34 2062 [OVS_DP_ATTR_IFINDEX] = {.type = NLA_U32 },
0c200ef9
PS		 2063};
2064
66a9b928 2065static const struct genl_small_ops dp_datapath_genl_ops[] = {
ccb1352e 2066 { .cmd = OVS_DP_CMD_NEW,
ef6243ac 2067 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 2068 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 2069 .doit = ovs_dp_cmd_new
2070 },
2071 { .cmd = OVS_DP_CMD_DEL,
ef6243ac 2072 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 2073 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 2074 .doit = ovs_dp_cmd_del
2075 },
2076 { .cmd = OVS_DP_CMD_GET,
ef6243ac 2077 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
ccb1352e 2078 .flags = 0, /* OK for unprivileged users. */
ccb1352e
JG		 2079 .doit = ovs_dp_cmd_get,
2080 .dumpit = ovs_dp_cmd_dump
2081 },
2082 { .cmd = OVS_DP_CMD_SET,
ef6243ac 2083 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 2084 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 2085 .doit = ovs_dp_cmd_set,
2086 },
2087};
2088
56989f6d 2089static struct genl_family dp_datapath_genl_family __ro_after_init = {
ccb1352e 2090 .hdrsize = sizeof(struct ovs_header),
0c200ef9
PS		 2091 .name = OVS_DATAPATH_FAMILY,
2092 .version = OVS_DATAPATH_VERSION,
2093 .maxattr = OVS_DP_ATTR_MAX,
3b0f31f2 2094 .policy = datapath_policy,
3a4e0d6a
PS		 2095 .netnsok = true,
2096 .parallel_ops = true,
66a9b928
JK		 2097 .small_ops = dp_datapath_genl_ops,
2098 .n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
9c5d03d3 2099 .resv_start_op = OVS_DP_CMD_SET + 1,
0c200ef9
PS		 2100 .mcgrps = &ovs_dp_datapath_multicast_group,
2101 .n_mcgrps = 1,
489111e5 2102 .module = THIS_MODULE,
ccb1352e
JG		 2103};
2104
8e4e1713 2105/* Called with ovs_mutex or RCU read lock. */
ccb1352e 2106static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
9354d452 2107 struct net *net, u32 portid, u32 seq,
d4e4fdf9 2108 u32 flags, u8 cmd, gfp_t gfp)
ccb1352e
JG		 2109{
2110 struct ovs_header *ovs_header;
2111 struct ovs_vport_stats vport_stats;
2112 int err;
2113
15e47304 2114 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
ccb1352e
JG		 2115 flags, cmd);
2116 if (!ovs_header)
2117 return -EMSGSIZE;
2118
2119 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
2120
028d6a67
DM		 2121 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
2122 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
5cd667b0 2123 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
9354d452
JB		 2124 ovs_vport_name(vport)) ||
2125 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
028d6a67 2126 goto nla_put_failure;
ccb1352e 2127
9354d452 2128 if (!net_eq(net, dev_net(vport->dev))) {
d4e4fdf9 2129 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
9354d452
JB		 2130
2131 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2132 goto nla_put_failure;
2133 }
2134
ccb1352e 2135 ovs_vport_get_stats(vport, &vport_stats);
66c7a5ee
ND		 2136 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2137 sizeof(struct ovs_vport_stats), &vport_stats,
2138 OVS_VPORT_ATTR_PAD))
028d6a67 2139 goto nla_put_failure;
ccb1352e 2140
1933ea36 2141 if (ovs_vport_get_upcall_stats(vport, skb))
2142 goto nla_put_failure;
2143
5cd667b0
AW		 2144 if (ovs_vport_get_upcall_portids(vport, skb))
2145 goto nla_put_failure;
2146
ccb1352e
JG		 2147 err = ovs_vport_get_options(vport, skb);
2148 if (err == -EMSGSIZE)
2149 goto error;
2150
053c095a
JB		 2151 genlmsg_end(skb, ovs_header);
2152 return 0;
ccb1352e
JG		 2153
2154nla_put_failure:
2155 err = -EMSGSIZE;
2156error:
2157 genlmsg_cancel(skb, ovs_header);
2158 return err;
2159}
2160
6093ae9a
JR		 2161static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2162{
2163 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2164}
2165
2166/* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
9354d452
JB		 2167struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2168 u32 portid, u32 seq, u8 cmd)
ccb1352e
JG		 2169{
2170 struct sk_buff *skb;
2171 int retval;
2172
d4e4fdf9 2173 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
ccb1352e
JG		 2174 if (!skb)
2175 return ERR_PTR(-ENOMEM);
2176
d4e4fdf9
GN		 2177 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2178 GFP_KERNEL);
a9341512
JG		 2179 BUG_ON(retval < 0);
2180
ccb1352e
JG		 2181 return skb;
2182}
2183
8e4e1713 2184/* Called with ovs_mutex or RCU read lock. */
46df7b81 2185static struct vport *lookup_vport(struct net *net,
12eb18f7 2186 const struct ovs_header *ovs_header,
ccb1352e
JG		 2187 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2188{
2189 struct datapath *dp;
2190 struct vport *vport;
2191
9354d452
JB		 2192 if (a[OVS_VPORT_ATTR_IFINDEX])
2193 return ERR_PTR(-EOPNOTSUPP);
ccb1352e 2194 if (a[OVS_VPORT_ATTR_NAME]) {
46df7b81 2195 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
ccb1352e
JG		 2196 if (!vport)
2197 return ERR_PTR(-ENODEV);
651a68ea
BP		 2198 if (ovs_header->dp_ifindex &&
2199 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2200 return ERR_PTR(-ENODEV);
ccb1352e
JG		 2201 return vport;
2202 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2203 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2204
2205 if (port_no >= DP_MAX_PORTS)
2206 return ERR_PTR(-EFBIG);
2207
46df7b81 2208 dp = get_dp(net, ovs_header->dp_ifindex);
ccb1352e
JG		 2209 if (!dp)
2210 return ERR_PTR(-ENODEV);
2211
8e4e1713 2212 vport = ovs_vport_ovsl_rcu(dp, port_no);
ccb1352e 2213 if (!vport)
14408dba 2214 return ERR_PTR(-ENODEV);
ccb1352e
JG		 2215 return vport;
2216 } else
2217 return ERR_PTR(-EINVAL);
9354d452 2218
ccb1352e
JG		 2219}
2220
6b660c41 2221static unsigned int ovs_get_max_headroom(struct datapath *dp)
3a927bc7 2222{
6b660c41 2223 unsigned int dev_headroom, max_headroom = 0;
3a927bc7
PA		 2224 struct net_device *dev;
2225 struct vport *vport;
2226 int i;
2227
2228 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
53742e69 2229 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
cf3266ad 2230 lockdep_ovsl_is_held()) {
3a927bc7
PA		 2231 dev = vport->dev;
2232 dev_headroom = netdev_get_fwd_headroom(dev);
2233 if (dev_headroom > max_headroom)
2234 max_headroom = dev_headroom;
2235 }
2236 }
2237
6b660c41
TY		 2238 return max_headroom;
2239}
2240
2241/* Called with ovs_mutex */
2242static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2243{
2244 struct vport *vport;
2245 int i;
2246
2247 dp->max_headroom = new_headroom;
cf3266ad 2248 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
53742e69 2249 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
cf3266ad 2250 lockdep_ovsl_is_held())
6b660c41 2251 netdev_set_rx_headroom(vport->dev, new_headroom);
cf3266ad 2252 }
3a927bc7
PA		 2253}
2254
ccb1352e
JG		 2255static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2256{
2257 struct nlattr **a = info->attrs;
2258 struct ovs_header *ovs_header = info->userhdr;
2259 struct vport_parms parms;
2260 struct sk_buff *reply;
2261 struct vport *vport;
2262 struct datapath *dp;
6b660c41 2263 unsigned int new_headroom;
ccb1352e
JG		 2264 u32 port_no;
2265 int err;
2266
ccb1352e
JG		 2267 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2268 !a[OVS_VPORT_ATTR_UPCALL_PID])
6093ae9a 2269 return -EINVAL;
54c4ef34
AZ		 2270
2271 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2272
2273 if (a[OVS_VPORT_ATTR_IFINDEX] && parms.type != OVS_VPORT_TYPE_INTERNAL)
9354d452 2274 return -EOPNOTSUPP;
6093ae9a
JR		 2275
2276 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2277 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2278 if (port_no >= DP_MAX_PORTS)
2279 return -EFBIG;
2280
2281 reply = ovs_vport_cmd_alloc_info();
2282 if (!reply)
2283 return -ENOMEM;
ccb1352e 2284
8e4e1713 2285 ovs_lock();
62b9c8d0 2286restart:
46df7b81 2287 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
ccb1352e
JG		 2288 err = -ENODEV;
2289 if (!dp)
6093ae9a 2290 goto exit_unlock_free;
ccb1352e 2291
6093ae9a 2292 if (port_no) {
8e4e1713 2293 vport = ovs_vport_ovsl(dp, port_no);
ccb1352e
JG		 2294 err = -EBUSY;
2295 if (vport)
6093ae9a 2296 goto exit_unlock_free;
ccb1352e
JG		 2297 } else {
2298 for (port_no = 1; ; port_no++) {
2299 if (port_no >= DP_MAX_PORTS) {
2300 err = -EFBIG;
6093ae9a 2301 goto exit_unlock_free;
ccb1352e 2302 }
8e4e1713 2303 vport = ovs_vport_ovsl(dp, port_no);
ccb1352e
JG		 2304 if (!vport)
2305 break;
2306 }
2307 }
2308
2309 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
ccb1352e
JG		 2310 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2311 parms.dp = dp;
2312 parms.port_no = port_no;
5cd667b0 2313 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
54c4ef34
AZ		 2314 parms.desired_ifindex = a[OVS_VPORT_ATTR_IFINDEX]
2315 ? nla_get_u32(a[OVS_VPORT_ATTR_IFINDEX]) : 0;
ccb1352e
JG		 2316
2317 vport = new_vport(&parms);
2318 err = PTR_ERR(vport);
62b9c8d0
TG		 2319 if (IS_ERR(vport)) {
2320 if (err == -EAGAIN)
2321 goto restart;
6093ae9a 2322 goto exit_unlock_free;
62b9c8d0 2323 }
ccb1352e 2324
1933ea36 2325 vport->upcall_stats = netdev_alloc_pcpu_stats(struct vport_upcall_stats_percpu);
2326 if (!vport->upcall_stats) {
2327 err = -ENOMEM;
95637d91 2328 goto exit_unlock_free_vport;
1933ea36 2329 }
2330
9354d452
JB		 2331 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2332 info->snd_portid, info->snd_seq, 0,
d4e4fdf9 2333 OVS_VPORT_CMD_NEW, GFP_KERNEL);
3a927bc7 2334
6b660c41
TY		 2335 new_headroom = netdev_get_fwd_headroom(vport->dev);
2336
2337 if (new_headroom > dp->max_headroom)
2338 ovs_update_headroom(dp, new_headroom);
3a927bc7
PA		 2339 else
2340 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2341
6093ae9a
JR		 2342 BUG_ON(err < 0);
2343 ovs_unlock();
ed661185 2344
2a94fe48 2345 ovs_notify(&dp_vport_genl_family, reply, info);
6093ae9a 2346 return 0;
ccb1352e 2347
95637d91
AC		 2348exit_unlock_free_vport:
2349 ovs_dp_detach_port(vport);
6093ae9a 2350exit_unlock_free:
8e4e1713 2351 ovs_unlock();
6093ae9a 2352 kfree_skb(reply);
ccb1352e
JG		 2353 return err;
2354}
2355
2356static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2357{
2358 struct nlattr **a = info->attrs;
2359 struct sk_buff *reply;
2360 struct vport *vport;
2361 int err;
2362
6093ae9a
JR		 2363 reply = ovs_vport_cmd_alloc_info();
2364 if (!reply)
2365 return -ENOMEM;
2366
8e4e1713 2367 ovs_lock();
46df7b81 2368 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
ccb1352e
JG		 2369 err = PTR_ERR(vport);
2370 if (IS_ERR(vport))
6093ae9a 2371 goto exit_unlock_free;
ccb1352e 2372
ccb1352e 2373 if (a[OVS_VPORT_ATTR_TYPE] &&
f44f3408 2374 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
ccb1352e 2375 err = -EINVAL;
6093ae9a 2376 goto exit_unlock_free;
a9341512
JG		 2377 }
2378
f44f3408 2379 if (a[OVS_VPORT_ATTR_OPTIONS]) {
ccb1352e 2380 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
f44f3408 2381 if (err)
6093ae9a 2382 goto exit_unlock_free;
f44f3408 2383 }
a9341512 2384
5cd667b0
AW		 2385
2386 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2387 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2388
2389 err = ovs_vport_set_upcall_portids(vport, ids);
2390 if (err)
2391 goto exit_unlock_free;
2392 }
ccb1352e 2393
9354d452
JB		 2394 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2395 info->snd_portid, info->snd_seq, 0,
d4e4fdf9 2396 OVS_VPORT_CMD_SET, GFP_KERNEL);
a9341512 2397 BUG_ON(err < 0);
ccb1352e 2398
8e4e1713 2399 ovs_unlock();
2a94fe48 2400 ovs_notify(&dp_vport_genl_family, reply, info);
8e4e1713 2401 return 0;
ccb1352e 2402
6093ae9a 2403exit_unlock_free:
8e4e1713 2404 ovs_unlock();
6093ae9a 2405 kfree_skb(reply);
ccb1352e
JG		 2406 return err;
2407}
2408
2409static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2410{
6b660c41 2411 bool update_headroom = false;
ccb1352e
JG		 2412 struct nlattr **a = info->attrs;
2413 struct sk_buff *reply;
3a927bc7 2414 struct datapath *dp;
ccb1352e 2415 struct vport *vport;
6b660c41 2416 unsigned int new_headroom;
ccb1352e
JG		 2417 int err;
2418
6093ae9a
JR		 2419 reply = ovs_vport_cmd_alloc_info();
2420 if (!reply)
2421 return -ENOMEM;
2422
8e4e1713 2423 ovs_lock();
46df7b81 2424 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
ccb1352e
JG		 2425 err = PTR_ERR(vport);
2426 if (IS_ERR(vport))
6093ae9a 2427 goto exit_unlock_free;
ccb1352e
JG		 2428
2429 if (vport->port_no == OVSP_LOCAL) {
2430 err = -EINVAL;
6093ae9a 2431 goto exit_unlock_free;
ccb1352e
JG		 2432 }
2433
9354d452
JB		 2434 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2435 info->snd_portid, info->snd_seq, 0,
d4e4fdf9 2436 OVS_VPORT_CMD_DEL, GFP_KERNEL);
6093ae9a 2437 BUG_ON(err < 0);
3a927bc7
PA		 2438
2439 /* the vport deletion may trigger dp headroom update */
2440 dp = vport->dp;
2441 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
6b660c41
TY		 2442 update_headroom = true;
2443
3a927bc7 2444 netdev_reset_rx_headroom(vport->dev);
ccb1352e 2445 ovs_dp_detach_port(vport);
3a927bc7 2446
6b660c41
TY		 2447 if (update_headroom) {
2448 new_headroom = ovs_get_max_headroom(dp);
2449
2450 if (new_headroom < dp->max_headroom)
2451 ovs_update_headroom(dp, new_headroom);
2452 }
6093ae9a 2453 ovs_unlock();
ccb1352e 2454
2a94fe48 2455 ovs_notify(&dp_vport_genl_family, reply, info);
6093ae9a 2456 return 0;
ccb1352e 2457
6093ae9a 2458exit_unlock_free:
8e4e1713 2459 ovs_unlock();
6093ae9a 2460 kfree_skb(reply);
ccb1352e
JG		 2461 return err;
2462}
2463
2464static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2465{
2466 struct nlattr **a = info->attrs;
2467 struct ovs_header *ovs_header = info->userhdr;
2468 struct sk_buff *reply;
2469 struct vport *vport;
2470 int err;
2471
6093ae9a
JR		 2472 reply = ovs_vport_cmd_alloc_info();
2473 if (!reply)
2474 return -ENOMEM;
2475
ccb1352e 2476 rcu_read_lock();
46df7b81 2477 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
ccb1352e
JG		 2478 err = PTR_ERR(vport);
2479 if (IS_ERR(vport))
6093ae9a 2480 goto exit_unlock_free;
9354d452
JB		 2481 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2482 info->snd_portid, info->snd_seq, 0,
d4e4fdf9 2483 OVS_VPORT_CMD_GET, GFP_ATOMIC);
6093ae9a 2484 BUG_ON(err < 0);
ccb1352e
JG		 2485 rcu_read_unlock();
2486
2487 return genlmsg_reply(reply, info);
2488
6093ae9a 2489exit_unlock_free:
ccb1352e 2490 rcu_read_unlock();
6093ae9a 2491 kfree_skb(reply);
ccb1352e
JG		 2492 return err;
2493}
2494
2495static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2496{
2497 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2498 struct datapath *dp;
15eac2a7
PS		 2499 int bucket = cb->args[0], skip = cb->args[1];
2500 int i, j = 0;
ccb1352e 2501
42ee19e2 2502 rcu_read_lock();
cc3a5ae6 2503 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
42ee19e2
JR		 2504 if (!dp) {
2505 rcu_read_unlock();
ccb1352e 2506 return -ENODEV;
42ee19e2 2507 }
15eac2a7 2508 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
ccb1352e 2509 struct vport *vport;
15eac2a7
PS		 2510
2511 j = 0;
b67bfe0d 2512 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
15eac2a7
PS		 2513 if (j >= skip &&
2514 ovs_vport_cmd_fill_info(vport, skb,
9354d452 2515 sock_net(skb->sk),
15e47304 2516 NETLINK_CB(cb->skb).portid,
15eac2a7
PS		 2517 cb->nlh->nlmsg_seq,
2518 NLM_F_MULTI,
d4e4fdf9
GN		 2519 OVS_VPORT_CMD_GET,
2520 GFP_ATOMIC) < 0)
15eac2a7
PS		 2521 goto out;
2522
2523 j++;
2524 }
2525 skip = 0;
ccb1352e 2526 }
15eac2a7 2527out:
ccb1352e
JG		 2528 rcu_read_unlock();
2529
15eac2a7
PS		 2530 cb->args[0] = i;
2531 cb->args[1] = j;
ccb1352e 2532
15eac2a7 2533 return skb->len;
ccb1352e
JG		 2534}
2535
eac87c41
EC		 2536static void ovs_dp_masks_rebalance(struct work_struct *work)
2537{
a65878d6
EC		 2538 struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2539 masks_rebalance.work);
2540 struct datapath *dp;
eac87c41
EC		 2541
2542 ovs_lock();
a65878d6
EC		 2543
2544 list_for_each_entry(dp, &ovs_net->dps, list_node)
2545 ovs_flow_masks_rebalance(&dp->table);
2546
eac87c41
EC		 2547 ovs_unlock();
2548
a65878d6 2549 schedule_delayed_work(&ovs_net->masks_rebalance,
eac87c41
EC		 2550 msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2551}
2552
0c200ef9
PS		 2553static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2554 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2555 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2556 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2557 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
ea8564c8 2558 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
0c200ef9 2559 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
9354d452
JB		 2560 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2561 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
1933ea36 2562 [OVS_VPORT_ATTR_UPCALL_STATS] = { .type = NLA_NESTED },
0c200ef9
PS		 2563};
2564
66a9b928 2565static const struct genl_small_ops dp_vport_genl_ops[] = {
ccb1352e 2566 { .cmd = OVS_VPORT_CMD_NEW,
ef6243ac 2567 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 2568 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 2569 .doit = ovs_vport_cmd_new
2570 },
2571 { .cmd = OVS_VPORT_CMD_DEL,
ef6243ac 2572 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 2573 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 2574 .doit = ovs_vport_cmd_del
2575 },
2576 { .cmd = OVS_VPORT_CMD_GET,
ef6243ac 2577 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
ccb1352e 2578 .flags = 0, /* OK for unprivileged users. */
ccb1352e
JG		 2579 .doit = ovs_vport_cmd_get,
2580 .dumpit = ovs_vport_cmd_dump
2581 },
2582 { .cmd = OVS_VPORT_CMD_SET,
ef6243ac 2583 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 2584 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 2585 .doit = ovs_vport_cmd_set,
2586 },
2587};
2588
56989f6d 2589struct genl_family dp_vport_genl_family __ro_after_init = {
0c200ef9
PS		 2590 .hdrsize = sizeof(struct ovs_header),
2591 .name = OVS_VPORT_FAMILY,
2592 .version = OVS_VPORT_VERSION,
2593 .maxattr = OVS_VPORT_ATTR_MAX,
3b0f31f2 2594 .policy = vport_policy,
0c200ef9
PS		 2595 .netnsok = true,
2596 .parallel_ops = true,
66a9b928
JK		 2597 .small_ops = dp_vport_genl_ops,
2598 .n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
e4ba4554 2599 .resv_start_op = OVS_VPORT_CMD_SET + 1,
0c200ef9
PS		 2600 .mcgrps = &ovs_dp_vport_multicast_group,
2601 .n_mcgrps = 1,
489111e5 2602 .module = THIS_MODULE,
ccb1352e
JG		 2603};
2604
0c200ef9
PS		 2605static struct genl_family * const dp_genl_families[] = {
2606 &dp_datapath_genl_family,
2607 &dp_vport_genl_family,
2608 &dp_flow_genl_family,
2609 &dp_packet_genl_family,
96fbc13d 2610 &dp_meter_genl_family,
11efd5cb
YHW		 2611#if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2612 &dp_ct_limit_genl_family,
2613#endif
ccb1352e
JG		 2614};
2615
2616static void dp_unregister_genl(int n_families)
2617{
2618 int i;
2619
2620 for (i = 0; i < n_families; i++)
0c200ef9 2621 genl_unregister_family(dp_genl_families[i]);
ccb1352e
JG		 2622}
2623
56989f6d 2624static int __init dp_register_genl(void)
ccb1352e 2625{
ccb1352e
JG		 2626 int err;
2627 int i;
2628
ccb1352e 2629 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
ccb1352e 2630
0c200ef9 2631 err = genl_register_family(dp_genl_families[i]);
ccb1352e
JG		 2632 if (err)
2633 goto error;
ccb1352e
JG		 2634 }
2635
2636 return 0;
2637
2638error:
0c200ef9 2639 dp_unregister_genl(i);
ccb1352e
JG		 2640 return err;
2641}
2642
46df7b81
PS		 2643static int __net_init ovs_init_net(struct net *net)
2644{
2645 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
e0afe914 2646 int err;
46df7b81
PS		 2647
2648 INIT_LIST_HEAD(&ovs_net->dps);
8e4e1713 2649 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
a65878d6 2650 INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
e0afe914
EC		 2651
2652 err = ovs_ct_init(net);
2653 if (err)
2654 return err;
2655
a65878d6
EC		 2656 schedule_delayed_work(&ovs_net->masks_rebalance,
2657 msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
e0afe914 2658 return 0;
46df7b81
PS		 2659}
2660
7b4577a9
PS		 2661static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2662 struct list_head *head)
46df7b81 2663{
8e4e1713 2664 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
7b4577a9
PS		 2665 struct datapath *dp;
2666
2667 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2668 int i;
2669
2670 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2671 struct vport *vport;
2672
2673 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
7b4577a9
PS		 2674 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2675 continue;
2676
be4ace6e 2677 if (dev_net(vport->dev) == dnet)
7b4577a9
PS		 2678 list_add(&vport->detach_list, head);
2679 }
2680 }
2681 }
2682}
2683
2684static void __net_exit ovs_exit_net(struct net *dnet)
2685{
2686 struct datapath *dp, *dp_next;
2687 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2688 struct vport *vport, *vport_next;
2689 struct net *net;
2690 LIST_HEAD(head);
46df7b81 2691
8e4e1713 2692 ovs_lock();
27de77ce
TZ		 2693
2694 ovs_ct_exit(dnet);
2695
46df7b81
PS		 2696 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2697 __dp_destroy(dp);
7b4577a9 2698
f0b07bb1 2699 down_read(&net_rwsem);
7b4577a9
PS		 2700 for_each_net(net)
2701 list_vports_from_net(net, dnet, &head);
f0b07bb1 2702 up_read(&net_rwsem);
7b4577a9
PS		 2703
2704 /* Detach all vports from given namespace. */
2705 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2706 list_del(&vport->detach_list);
2707 ovs_dp_detach_port(vport);
2708 }
2709
8e4e1713
PS		 2710 ovs_unlock();
2711
a65878d6 2712 cancel_delayed_work_sync(&ovs_net->masks_rebalance);
8e4e1713 2713 cancel_work_sync(&ovs_net->dp_notify_work);
46df7b81
PS		 2714}
2715
2716static struct pernet_operations ovs_net_ops = {
2717 .init = ovs_init_net,
2718 .exit = ovs_exit_net,
2719 .id = &ovs_net_id,
2720 .size = sizeof(struct ovs_net),
2721};
2722
ccb1352e
JG		 2723static int __init dp_init(void)
2724{
ccb1352e
JG		 2725 int err;
2726
cf3266ad
TZ		 2727 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2728 sizeof_field(struct sk_buff, cb));
ccb1352e
JG		 2729
2730 pr_info("Open vSwitch switching datapath\n");
2731
971427f3 2732 err = action_fifos_init();
ccb1352e
JG		 2733 if (err)
2734 goto error;
2735
971427f3
AZ		 2736 err = ovs_internal_dev_rtnl_link_register();
2737 if (err)
2738 goto error_action_fifos_exit;
2739
5b9e7e16
JP		 2740 err = ovs_flow_init();
2741 if (err)
2742 goto error_unreg_rtnl_link;
2743
ccb1352e
JG		 2744 err = ovs_vport_init();
2745 if (err)
2746 goto error_flow_exit;
2747
46df7b81 2748 err = register_pernet_device(&ovs_net_ops);
ccb1352e
JG		 2749 if (err)
2750 goto error_vport_exit;
2751
46df7b81
PS		 2752 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2753 if (err)
2754 goto error_netns_exit;
2755
62b9c8d0
TG		 2756 err = ovs_netdev_init();
2757 if (err)
2758 goto error_unreg_notifier;
2759
ccb1352e
JG		 2760 err = dp_register_genl();
2761 if (err < 0)
62b9c8d0 2762 goto error_unreg_netdev;
ccb1352e 2763
ccb1352e
JG		 2764 return 0;
2765
62b9c8d0
TG		 2766error_unreg_netdev:
2767 ovs_netdev_exit();
ccb1352e
JG		 2768error_unreg_notifier:
2769 unregister_netdevice_notifier(&ovs_dp_device_notifier);
46df7b81
PS		 2770error_netns_exit:
2771 unregister_pernet_device(&ovs_net_ops);
ccb1352e
JG		 2772error_vport_exit:
2773 ovs_vport_exit();
2774error_flow_exit:
2775 ovs_flow_exit();
5b9e7e16
JP		 2776error_unreg_rtnl_link:
2777 ovs_internal_dev_rtnl_link_unregister();
971427f3
AZ		 2778error_action_fifos_exit:
2779 action_fifos_exit();
ccb1352e
JG		 2780error:
2781 return err;
2782}
2783
2784static void dp_cleanup(void)
2785{
ccb1352e 2786 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
62b9c8d0 2787 ovs_netdev_exit();
ccb1352e 2788 unregister_netdevice_notifier(&ovs_dp_device_notifier);
46df7b81
PS		 2789 unregister_pernet_device(&ovs_net_ops);
2790 rcu_barrier();
ccb1352e
JG		 2791 ovs_vport_exit();
2792 ovs_flow_exit();
5b9e7e16 2793 ovs_internal_dev_rtnl_link_unregister();
971427f3 2794 action_fifos_exit();
ccb1352e
JG		 2795}
2796
2797module_init(dp_init);
2798module_exit(dp_cleanup);
2799
2800MODULE_DESCRIPTION("Open vSwitch switching datapath");
2801MODULE_LICENSE("GPL");
ed227099
TLSC		 2802MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2803MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2804MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2805MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
96fbc13d 2806MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
11efd5cb 2807MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
Linux 6.x block layer and io_uring tree(s)