projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
MIPS: OCTEON: Replace SIZEOF_FIELD() macro
[linux-2.6-block.git] / net / openvswitch / datapath.c
CommitLineData
c9422999 1// SPDX-License-Identifier: GPL-2.0-only
ccb1352e 2/*
ad552007 3 * Copyright (c) 2007-2014 Nicira, Inc.
ccb1352e
JG		 4 */
5
6#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7
8#include <linux/init.h>
9#include <linux/module.h>
10#include <linux/if_arp.h>
11#include <linux/if_vlan.h>
12#include <linux/in.h>
13#include <linux/ip.h>
14#include <linux/jhash.h>
15#include <linux/delay.h>
16#include <linux/time.h>
17#include <linux/etherdevice.h>
18#include <linux/genetlink.h>
19#include <linux/kernel.h>
20#include <linux/kthread.h>
21#include <linux/mutex.h>
22#include <linux/percpu.h>
23#include <linux/rcupdate.h>
24#include <linux/tcp.h>
25#include <linux/udp.h>
ccb1352e
JG		 26#include <linux/ethtool.h>
27#include <linux/wait.h>
ccb1352e
JG		 28#include <asm/div64.h>
29#include <linux/highmem.h>
30#include <linux/netfilter_bridge.h>
31#include <linux/netfilter_ipv4.h>
32#include <linux/inetdevice.h>
33#include <linux/list.h>
34#include <linux/openvswitch.h>
35#include <linux/rculist.h>
36#include <linux/dmi.h>
ccb1352e 37#include <net/genetlink.h>
46df7b81
PS		 38#include <net/net_namespace.h>
39#include <net/netns/generic.h>
ccb1352e
JG		 40
41#include "datapath.h"
42#include "flow.h"
e80857cc 43#include "flow_table.h"
e6445719 44#include "flow_netlink.h"
96fbc13d 45#include "meter.h"
ccb1352e 46#include "vport-internal_dev.h"
cff63a52 47#include "vport-netdev.h"
ccb1352e 48
c7d03a00 49unsigned int ovs_net_id __read_mostly;
8e4e1713 50
0c200ef9
PS		 51static struct genl_family dp_packet_genl_family;
52static struct genl_family dp_flow_genl_family;
53static struct genl_family dp_datapath_genl_family;
54
74ed7ab9
JS		 55static const struct nla_policy flow_policy[];
56
48e48a70 57static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
58 .name = OVS_FLOW_MCGROUP,
0c200ef9
PS		 59};
60
48e48a70 61static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
62 .name = OVS_DATAPATH_MCGROUP,
0c200ef9
PS		 63};
64
48e48a70 65static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
66 .name = OVS_VPORT_MCGROUP,
0c200ef9
PS		 67};
68
fb5d1e9e
JR		 69/* Check if need to build a reply message.
70 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
9b67aa4a
SG		 71static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
72 unsigned int group)
fb5d1e9e
JR		 73{
74 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
f8403a2e 75 genl_has_listeners(family, genl_info_net(info), group);
fb5d1e9e
JR		 76}
77
68eb5503 78static void ovs_notify(struct genl_family *family,
2a94fe48 79 struct sk_buff *skb, struct genl_info *info)
ed661185 80{
92c14d9b 81 genl_notify(family, skb, info, 0, GFP_KERNEL);
ed661185
TG		 82}
83
ccb1352e
JG		 84/**
85 * DOC: Locking:
86 *
8e4e1713
PS		 87 * All writes e.g. Writes to device state (add/remove datapath, port, set
88 * operations on vports, etc.), Writes to other state (flow table
89 * modifications, set miscellaneous datapath parameters, etc.) are protected
90 * by ovs_lock.
ccb1352e
JG		 91 *
92 * Reads are protected by RCU.
93 *
94 * There are a few special cases (mostly stats) that have their own
95 * synchronization but they nest under all of above and don't interact with
96 * each other.
8e4e1713
PS		 97 *
98 * The RTNL lock nests inside ovs_mutex.
ccb1352e
JG		 99 */
100
8e4e1713
PS		 101static DEFINE_MUTEX(ovs_mutex);
102
103void ovs_lock(void)
104{
105 mutex_lock(&ovs_mutex);
106}
107
108void ovs_unlock(void)
109{
110 mutex_unlock(&ovs_mutex);
111}
112
113#ifdef CONFIG_LOCKDEP
114int lockdep_ovsl_is_held(void)
115{
116 if (debug_locks)
117 return lockdep_is_held(&ovs_mutex);
118 else
119 return 1;
120}
121#endif
122
ccb1352e 123static struct vport *new_vport(const struct vport_parms *);
8055a89c 124static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
e8eedb85 125 const struct sw_flow_key *,
f2a4d086
WT		 126 const struct dp_upcall_info *,
127 uint32_t cutlen);
8055a89c 128static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
e8eedb85 129 const struct sw_flow_key *,
f2a4d086
WT		 130 const struct dp_upcall_info *,
131 uint32_t cutlen);
ccb1352e 132
8e4e1713 133/* Must be called with rcu_read_lock or ovs_mutex. */
971427f3 134const char *ovs_dp_name(const struct datapath *dp)
ccb1352e 135{
8e4e1713 136 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
c9db965c 137 return ovs_vport_name(vport);
ccb1352e
JG		 138}
139
12eb18f7 140static int get_dpifindex(const struct datapath *dp)
ccb1352e
JG		 141{
142 struct vport *local;
143 int ifindex;
144
145 rcu_read_lock();
146
15eac2a7 147 local = ovs_vport_rcu(dp, OVSP_LOCAL);
ccb1352e 148 if (local)
be4ace6e 149 ifindex = local->dev->ifindex;
ccb1352e
JG		 150 else
151 ifindex = 0;
152
153 rcu_read_unlock();
154
155 return ifindex;
156}
157
158static void destroy_dp_rcu(struct rcu_head *rcu)
159{
160 struct datapath *dp = container_of(rcu, struct datapath, rcu);
161
9b996e54 162 ovs_flow_tbl_destroy(&dp->table);
ccb1352e 163 free_percpu(dp->stats_percpu);
15eac2a7 164 kfree(dp->ports);
96fbc13d 165 ovs_meters_exit(dp);
ccb1352e
JG		 166 kfree(dp);
167}
168
15eac2a7
PS		 169static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
170 u16 port_no)
171{
172 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
173}
174
bb6f9a70 175/* Called with ovs_mutex or RCU read lock. */
15eac2a7
PS		 176struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
177{
178 struct vport *vport;
15eac2a7
PS		 179 struct hlist_head *head;
180
181 head = vport_hash_bucket(dp, port_no);
b67bfe0d 182 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
15eac2a7
PS		 183 if (vport->port_no == port_no)
184 return vport;
185 }
186 return NULL;
187}
188
8e4e1713 189/* Called with ovs_mutex. */
ccb1352e
JG		 190static struct vport *new_vport(const struct vport_parms *parms)
191{
192 struct vport *vport;
193
194 vport = ovs_vport_add(parms);
195 if (!IS_ERR(vport)) {
196 struct datapath *dp = parms->dp;
15eac2a7 197 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
ccb1352e 198
15eac2a7 199 hlist_add_head_rcu(&vport->dp_hash_node, head);
ccb1352e 200 }
ccb1352e
JG		 201 return vport;
202}
203
ccb1352e
JG		 204void ovs_dp_detach_port(struct vport *p)
205{
8e4e1713 206 ASSERT_OVSL();
ccb1352e
JG		 207
208 /* First drop references to device. */
15eac2a7 209 hlist_del_rcu(&p->dp_hash_node);
ccb1352e
JG		 210
211 /* Then destroy it. */
212 ovs_vport_del(p);
213}
214
215/* Must be called with rcu_read_lock. */
8c8b1b83 216void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
ccb1352e 217{
83c8df26 218 const struct vport *p = OVS_CB(skb)->input_vport;
ccb1352e
JG		 219 struct datapath *dp = p->dp;
220 struct sw_flow *flow;
d98612b8 221 struct sw_flow_actions *sf_acts;
ccb1352e 222 struct dp_stats_percpu *stats;
ccb1352e 223 u64 *stats_counter;
1bd7116f 224 u32 n_mask_hit;
aa733660 225 int error;
ccb1352e 226
404f2f10 227 stats = this_cpu_ptr(dp->stats_percpu);
ccb1352e 228
ccb1352e 229 /* Look up flow. */
04b7d136
TZ		 230 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
231 &n_mask_hit);
ccb1352e
JG		 232 if (unlikely(!flow)) {
233 struct dp_upcall_info upcall;
234
ccea7445 235 memset(&upcall, 0, sizeof(upcall));
ccb1352e 236 upcall.cmd = OVS_PACKET_CMD_MISS;
5cd667b0 237 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
7f8a436e 238 upcall.mru = OVS_CB(skb)->mru;
f2a4d086 239 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
c5eba0b6
LR		 240 if (unlikely(error))
241 kfree_skb(skb);
242 else
243 consume_skb(skb);
ccb1352e
JG		 244 stats_counter = &stats->n_missed;
245 goto out;
246 }
247
d98612b8
LJ		 248 ovs_flow_stats_update(flow, key->tp.flags, skb);
249 sf_acts = rcu_dereference(flow->sf_acts);
aa733660
YS		 250 error = ovs_execute_actions(dp, skb, sf_acts, key);
251 if (unlikely(error))
252 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
253 ovs_dp_name(dp), error);
ccb1352e 254
e298e505 255 stats_counter = &stats->n_hit;
ccb1352e
JG		 256
257out:
258 /* Update datapath statistics. */
df9d9fdf 259 u64_stats_update_begin(&stats->syncp);
ccb1352e 260 (*stats_counter)++;
1bd7116f 261 stats->n_mask_hit += n_mask_hit;
df9d9fdf 262 u64_stats_update_end(&stats->syncp);
ccb1352e
JG		 263}
264
ccb1352e 265int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
e8eedb85 266 const struct sw_flow_key *key,
f2a4d086
WT		 267 const struct dp_upcall_info *upcall_info,
268 uint32_t cutlen)
ccb1352e
JG		 269{
270 struct dp_stats_percpu *stats;
ccb1352e
JG		 271 int err;
272
15e47304 273 if (upcall_info->portid == 0) {
ccb1352e
JG		 274 err = -ENOTCONN;
275 goto err;
276 }
277
ccb1352e 278 if (!skb_is_gso(skb))
f2a4d086 279 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
ccb1352e 280 else
f2a4d086 281 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
ccb1352e
JG		 282 if (err)
283 goto err;
284
285 return 0;
286
287err:
404f2f10 288 stats = this_cpu_ptr(dp->stats_percpu);
ccb1352e 289
df9d9fdf 290 u64_stats_update_begin(&stats->syncp);
ccb1352e 291 stats->n_lost++;
df9d9fdf 292 u64_stats_update_end(&stats->syncp);
ccb1352e
JG		 293
294 return err;
295}
296
8055a89c 297static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
e8eedb85 298 const struct sw_flow_key *key,
f2a4d086
WT		 299 const struct dp_upcall_info *upcall_info,
300 uint32_t cutlen)
ccb1352e 301{
2734166e 302 unsigned int gso_type = skb_shinfo(skb)->gso_type;
0c19f846 303 struct sw_flow_key later_key;
ccb1352e
JG		 304 struct sk_buff *segs, *nskb;
305 int err;
306
9207f9d4 307 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
09c5e605 308 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
92e5dfc3
PS		 309 if (IS_ERR(segs))
310 return PTR_ERR(segs);
330966e5
FW		 311 if (segs == NULL)
312 return -EINVAL;
ccb1352e 313
0c19f846
WB		 314 if (gso_type & SKB_GSO_UDP) {
315 /* The initial flow key extracted by ovs_flow_key_extract()
316 * in this case is for a first fragment, so we need to
317 * properly mark later fragments.
318 */
319 later_key = *key;
320 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
321 }
322
ccb1352e
JG		 323 /* Queue all of the segments. */
324 skb = segs;
325 do {
0c19f846
WB		 326 if (gso_type & SKB_GSO_UDP && skb != segs)
327 key = &later_key;
328
f2a4d086 329 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
ccb1352e
JG		 330 if (err)
331 break;
332
ccb1352e
JG		 333 } while ((skb = skb->next));
334
335 /* Free all of the segments. */
336 skb = segs;
337 do {
338 nskb = skb->next;
339 if (err)
340 kfree_skb(skb);
341 else
342 consume_skb(skb);
343 } while ((skb = nskb));
344 return err;
345}
346
8f0aad6f 347static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
494bea39 348 unsigned int hdrlen, int actions_attrlen)
c3ff8cfe
TG		 349{
350 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
bda56f14 351 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
b95e5928 352 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
bd1903b7
TZ		 353 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
354 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
c3ff8cfe
TG		 355
356 /* OVS_PACKET_ATTR_USERDATA */
8f0aad6f
WZ		 357 if (upcall_info->userdata)
358 size += NLA_ALIGN(upcall_info->userdata->nla_len);
359
360 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
361 if (upcall_info->egress_tun_info)
362 size += nla_total_size(ovs_tun_key_attr_size());
c3ff8cfe 363
ccea7445
NM		 364 /* OVS_PACKET_ATTR_ACTIONS */
365 if (upcall_info->actions_len)
494bea39 366 size += nla_total_size(actions_attrlen);
ccea7445 367
7f8a436e
JS		 368 /* OVS_PACKET_ATTR_MRU */
369 if (upcall_info->mru)
370 size += nla_total_size(sizeof(upcall_info->mru));
371
c3ff8cfe
TG		 372 return size;
373}
374
7f8a436e
JS		 375static void pad_packet(struct datapath *dp, struct sk_buff *skb)
376{
377 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
378 size_t plen = NLA_ALIGN(skb->len) - skb->len;
379
380 if (plen > 0)
b080db58 381 skb_put_zero(skb, plen);
7f8a436e
JS		 382 }
383}
384
8055a89c 385static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
e8eedb85 386 const struct sw_flow_key *key,
f2a4d086
WT		 387 const struct dp_upcall_info *upcall_info,
388 uint32_t cutlen)
ccb1352e
JG		 389{
390 struct ovs_header *upcall;
391 struct sk_buff *nskb = NULL;
4ee45ea0 392 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
ccb1352e 393 struct nlattr *nla;
795449d8 394 size_t len;
bda56f14 395 unsigned int hlen;
8055a89c 396 int err, dp_ifindex;
bd1903b7 397 u64 hash;
8055a89c
TG		 398
399 dp_ifindex = get_dpifindex(dp);
400 if (!dp_ifindex)
401 return -ENODEV;
ccb1352e 402
df8a39de 403 if (skb_vlan_tag_present(skb)) {
ccb1352e
JG		 404 nskb = skb_clone(skb, GFP_ATOMIC);
405 if (!nskb)
406 return -ENOMEM;
407
5968250c 408 nskb = __vlan_hwaccel_push_inside(nskb);
8aa51d64 409 if (!nskb)
ccb1352e
JG		 410 return -ENOMEM;
411
ccb1352e
JG		 412 skb = nskb;
413 }
414
415 if (nla_attr_size(skb->len) > USHRT_MAX) {
416 err = -EFBIG;
417 goto out;
418 }
419
bda56f14
TG		 420 /* Complete checksum if needed */
421 if (skb->ip_summed == CHECKSUM_PARTIAL &&
7529390d 422 (err = skb_csum_hwoffload_help(skb, 0)))
bda56f14
TG		 423 goto out;
424
425 /* Older versions of OVS user space enforce alignment of the last
426 * Netlink attribute to NLA_ALIGNTO which would require extensive
427 * padding logic. Only perform zerocopy if padding is not required.
428 */
429 if (dp->user_features & OVS_DP_F_UNALIGNED)
430 hlen = skb_zerocopy_headlen(skb);
431 else
432 hlen = skb->len;
433
494bea39
LZ		 434 len = upcall_msg_size(upcall_info, hlen - cutlen,
435 OVS_CB(skb)->acts_origlen);
551ddc05 436 user_skb = genlmsg_new(len, GFP_ATOMIC);
ccb1352e
JG		 437 if (!user_skb) {
438 err = -ENOMEM;
439 goto out;
440 }
441
442 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
443 0, upcall_info->cmd);
6f19893b
KL		 444 if (!upcall) {
445 err = -EINVAL;
446 goto out;
447 }
ccb1352e
JG		 448 upcall->dp_ifindex = dp_ifindex;
449
5b4237bb 450 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
a734d1f4
EC		 451 if (err)
452 goto out;
ccb1352e
JG		 453
454 if (upcall_info->userdata)
4490108b
BP		 455 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
456 nla_len(upcall_info->userdata),
457 nla_data(upcall_info->userdata));
ccb1352e 458
8f0aad6f 459 if (upcall_info->egress_tun_info) {
ae0be8de
MK		 460 nla = nla_nest_start_noflag(user_skb,
461 OVS_PACKET_ATTR_EGRESS_TUN_KEY);
0fff9bd4
KL		 462 if (!nla) {
463 err = -EMSGSIZE;
464 goto out;
465 }
fc4099f1
PS		 466 err = ovs_nla_put_tunnel_info(user_skb,
467 upcall_info->egress_tun_info);
a734d1f4
EC		 468 if (err)
469 goto out;
470
8f0aad6f
WZ		 471 nla_nest_end(user_skb, nla);
472 }
473
ccea7445 474 if (upcall_info->actions_len) {
ae0be8de 475 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
0fff9bd4
KL		 476 if (!nla) {
477 err = -EMSGSIZE;
478 goto out;
479 }
ccea7445
NM		 480 err = ovs_nla_put_actions(upcall_info->actions,
481 upcall_info->actions_len,
482 user_skb);
483 if (!err)
484 nla_nest_end(user_skb, nla);
485 else
486 nla_nest_cancel(user_skb, nla);
487 }
488
7f8a436e 489 /* Add OVS_PACKET_ATTR_MRU */
61ca533c
TZ		 490 if (upcall_info->mru &&
491 nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
492 err = -ENOBUFS;
493 goto out;
7f8a436e
JS		 494 }
495
b95e5928 496 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
61ca533c
TZ		 497 if (cutlen > 0 &&
498 nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
499 err = -ENOBUFS;
500 goto out;
b95e5928
WT		 501 }
502
bd1903b7
TZ		 503 /* Add OVS_PACKET_ATTR_HASH */
504 hash = skb_get_hash_raw(skb);
505 if (skb->sw_hash)
506 hash |= OVS_PACKET_HASH_SW_BIT;
507
508 if (skb->l4_hash)
509 hash |= OVS_PACKET_HASH_L4_BIT;
510
511 if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
512 err = -ENOBUFS;
513 goto out;
514 }
515
bda56f14
TG		 516 /* Only reserve room for attribute header, packet data is added
517 * in skb_zerocopy() */
518 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
519 err = -ENOBUFS;
520 goto out;
521 }
f2a4d086 522 nla->nla_len = nla_attr_size(skb->len - cutlen);
ccb1352e 523
f2a4d086 524 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
36d5fe6a
ZK		 525 if (err)
526 goto out;
ccb1352e 527
aea0bb4f 528 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
7f8a436e 529 pad_packet(dp, user_skb);
aea0bb4f 530
bda56f14 531 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
ccb1352e 532
bda56f14 533 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
4ee45ea0 534 user_skb = NULL;
ccb1352e 535out:
36d5fe6a
ZK		 536 if (err)
537 skb_tx_error(skb);
4ee45ea0 538 kfree_skb(user_skb);
ccb1352e
JG		 539 kfree_skb(nskb);
540 return err;
541}
542
ccb1352e
JG		 543static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
544{
545 struct ovs_header *ovs_header = info->userhdr;
7f8a436e 546 struct net *net = sock_net(skb->sk);
ccb1352e
JG		 547 struct nlattr **a = info->attrs;
548 struct sw_flow_actions *acts;
549 struct sk_buff *packet;
550 struct sw_flow *flow;
d98612b8 551 struct sw_flow_actions *sf_acts;
ccb1352e 552 struct datapath *dp;
83c8df26 553 struct vport *input_vport;
7f8a436e 554 u16 mru = 0;
bd1903b7 555 u64 hash;
ccb1352e
JG		 556 int len;
557 int err;
1ba39804 558 bool log = !a[OVS_PACKET_ATTR_PROBE];
ccb1352e
JG		 559
560 err = -EINVAL;
561 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
dded45fc 562 !a[OVS_PACKET_ATTR_ACTIONS])
ccb1352e
JG		 563 goto err;
564
565 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
566 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
567 err = -ENOMEM;
568 if (!packet)
569 goto err;
570 skb_reserve(packet, NET_IP_ALIGN);
571
32686a9d 572 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
ccb1352e 573
7f8a436e
JS		 574 /* Set packet's mru */
575 if (a[OVS_PACKET_ATTR_MRU]) {
576 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
577 packet->ignore_df = 1;
578 }
579 OVS_CB(packet)->mru = mru;
580
bd1903b7
TZ		 581 if (a[OVS_PACKET_ATTR_HASH]) {
582 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
583
584 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
585 !!(hash & OVS_PACKET_HASH_SW_BIT),
586 !!(hash & OVS_PACKET_HASH_L4_BIT));
587 }
588
ccb1352e 589 /* Build an sw_flow for sending this packet. */
23dabf88 590 flow = ovs_flow_alloc();
ccb1352e
JG		 591 err = PTR_ERR(flow);
592 if (IS_ERR(flow))
593 goto err_kfree_skb;
594
c2ac6673
JS		 595 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
596 packet, &flow->key, log);
ccb1352e
JG		 597 if (err)
598 goto err_flow_free;
599
7f8a436e 600 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
05da5898 601 &flow->key, &acts, log);
74f84a57
PS		 602 if (err)
603 goto err_flow_free;
ccb1352e 604
f5796684 605 rcu_assign_pointer(flow->sf_acts, acts);
ccb1352e 606 packet->priority = flow->key.phy.priority;
39c7caeb 607 packet->mark = flow->key.phy.skb_mark;
ccb1352e
JG		 608
609 rcu_read_lock();
7f8a436e 610 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
ccb1352e
JG		 611 err = -ENODEV;
612 if (!dp)
613 goto err_unlock;
614
83c8df26
PS		 615 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
616 if (!input_vport)
617 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
618
619 if (!input_vport)
620 goto err_unlock;
621
7f8a436e 622 packet->dev = input_vport->dev;
83c8df26 623 OVS_CB(packet)->input_vport = input_vport;
d98612b8 624 sf_acts = rcu_dereference(flow->sf_acts);
83c8df26 625
ccb1352e 626 local_bh_disable();
d98612b8 627 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
ccb1352e
JG		 628 local_bh_enable();
629 rcu_read_unlock();
630
03f0d916 631 ovs_flow_free(flow, false);
ccb1352e
JG		 632 return err;
633
634err_unlock:
635 rcu_read_unlock();
636err_flow_free:
03f0d916 637 ovs_flow_free(flow, false);
ccb1352e
JG		 638err_kfree_skb:
639 kfree_skb(packet);
640err:
641 return err;
642}
643
644static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
dded45fc 645 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
ccb1352e
JG		 646 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
647 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
1ba39804 648 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
7f8a436e 649 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
ccb1352e
JG		 650};
651
4534de83 652static const struct genl_ops dp_packet_genl_ops[] = {
ccb1352e 653 { .cmd = OVS_PACKET_CMD_EXECUTE,
ef6243ac 654 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 655 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 656 .doit = ovs_packet_cmd_execute
657 }
658};
659
56989f6d 660static struct genl_family dp_packet_genl_family __ro_after_init = {
0c200ef9
PS		 661 .hdrsize = sizeof(struct ovs_header),
662 .name = OVS_PACKET_FAMILY,
663 .version = OVS_PACKET_VERSION,
664 .maxattr = OVS_PACKET_ATTR_MAX,
3b0f31f2 665 .policy = packet_policy,
0c200ef9
PS		 666 .netnsok = true,
667 .parallel_ops = true,
668 .ops = dp_packet_genl_ops,
669 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
489111e5 670 .module = THIS_MODULE,
0c200ef9
PS		 671};
672
12eb18f7 673static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
1bd7116f 674 struct ovs_dp_megaflow_stats *mega_stats)
ccb1352e
JG		 675{
676 int i;
ccb1352e 677
1bd7116f
AZ		 678 memset(mega_stats, 0, sizeof(*mega_stats));
679
b637e498 680 stats->n_flows = ovs_flow_tbl_count(&dp->table);
1bd7116f 681 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
ccb1352e
JG		 682
683 stats->n_hit = stats->n_missed = stats->n_lost = 0;
1bd7116f 684
ccb1352e
JG		 685 for_each_possible_cpu(i) {
686 const struct dp_stats_percpu *percpu_stats;
687 struct dp_stats_percpu local_stats;
688 unsigned int start;
689
690 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
691
692 do {
57a7744e 693 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
ccb1352e 694 local_stats = *percpu_stats;
57a7744e 695 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
ccb1352e
JG		 696
697 stats->n_hit += local_stats.n_hit;
698 stats->n_missed += local_stats.n_missed;
699 stats->n_lost += local_stats.n_lost;
1bd7116f 700 mega_stats->n_mask_hit += local_stats.n_mask_hit;
ccb1352e
JG		 701 }
702}
703
74ed7ab9
JS		 704static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
705{
706 return ovs_identifier_is_ufid(sfid) &&
707 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
708}
709
710static bool should_fill_mask(uint32_t ufid_flags)
711{
712 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
713}
714
715static bool should_fill_actions(uint32_t ufid_flags)
c3ff8cfe 716{
74ed7ab9
JS		 717 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
718}
719
720static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
721 const struct sw_flow_id *sfid,
722 uint32_t ufid_flags)
723{
724 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
725
4e81c0b3
PA		 726 /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
727 * see ovs_nla_put_identifier()
728 */
74ed7ab9
JS		 729 if (sfid && ovs_identifier_is_ufid(sfid))
730 len += nla_total_size(sfid->ufid_len);
4e81c0b3
PA		 731 else
732 len += nla_total_size(ovs_key_attr_size());
74ed7ab9
JS		 733
734 /* OVS_FLOW_ATTR_KEY */
735 if (!sfid || should_fill_key(sfid, ufid_flags))
736 len += nla_total_size(ovs_key_attr_size());
737
738 /* OVS_FLOW_ATTR_MASK */
739 if (should_fill_mask(ufid_flags))
740 len += nla_total_size(ovs_key_attr_size());
741
742 /* OVS_FLOW_ATTR_ACTIONS */
743 if (should_fill_actions(ufid_flags))
8e2fed1c 744 len += nla_total_size(acts->orig_len);
74ed7ab9
JS		 745
746 return len
66c7a5ee 747 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
c3ff8cfe 748 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
66c7a5ee 749 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
c3ff8cfe
TG		 750}
751
ca7105f2
JS		 752/* Called with ovs_mutex or RCU read lock. */
753static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
754 struct sk_buff *skb)
755{
756 struct ovs_flow_stats stats;
757 __be16 tcp_flags;
758 unsigned long used;
ccb1352e 759
e298e505 760 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
0e9796b4 761
028d6a67 762 if (used &&
0238b720
ND		 763 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
764 OVS_FLOW_ATTR_PAD))
ca7105f2 765 return -EMSGSIZE;
ccb1352e 766
028d6a67 767 if (stats.n_packets &&
66c7a5ee
ND		 768 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
769 sizeof(struct ovs_flow_stats), &stats,
770 OVS_FLOW_ATTR_PAD))
ca7105f2 771 return -EMSGSIZE;
ccb1352e 772
e298e505
PS		 773 if ((u8)ntohs(tcp_flags) &&
774 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
ca7105f2
JS		 775 return -EMSGSIZE;
776
777 return 0;
778}
779
780/* Called with ovs_mutex or RCU read lock. */
781static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
782 struct sk_buff *skb, int skb_orig_len)
783{
784 struct nlattr *start;
785 int err;
ccb1352e
JG		 786
787 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
788 * this is the first flow to be dumped into 'skb'. This is unusual for
789 * Netlink but individual action lists can be longer than
790 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
791 * The userspace caller can always fetch the actions separately if it
792 * really wants them. (Most userspace callers in fact don't care.)
793 *
794 * This can only fail for dump operations because the skb is always
795 * properly sized for single flows.
796 */
ae0be8de 797 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
74f84a57 798 if (start) {
d57170b1
PS		 799 const struct sw_flow_actions *sf_acts;
800
663efa36 801 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
e6445719
PS		 802 err = ovs_nla_put_actions(sf_acts->actions,
803 sf_acts->actions_len, skb);
0e9796b4 804
74f84a57
PS		 805 if (!err)
806 nla_nest_end(skb, start);
807 else {
808 if (skb_orig_len)
ca7105f2 809 return err;
74f84a57
PS		 810
811 nla_nest_cancel(skb, start);
812 }
ca7105f2
JS		 813 } else if (skb_orig_len) {
814 return -EMSGSIZE;
815 }
816
817 return 0;
818}
819
820/* Called with ovs_mutex or RCU read lock. */
821static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
822 struct sk_buff *skb, u32 portid,
74ed7ab9 823 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
ca7105f2
JS		 824{
825 const int skb_orig_len = skb->len;
826 struct ovs_header *ovs_header;
827 int err;
828
829 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
830 flags, cmd);
831 if (!ovs_header)
832 return -EMSGSIZE;
833
834 ovs_header->dp_ifindex = dp_ifindex;
835
74ed7ab9 836 err = ovs_nla_put_identifier(flow, skb);
5b4237bb
JS		 837 if (err)
838 goto error;
839
74ed7ab9
JS		 840 if (should_fill_key(&flow->id, ufid_flags)) {
841 err = ovs_nla_put_masked_key(flow, skb);
842 if (err)
843 goto error;
844 }
845
846 if (should_fill_mask(ufid_flags)) {
847 err = ovs_nla_put_mask(flow, skb);
848 if (err)
849 goto error;
850 }
ca7105f2
JS		 851
852 err = ovs_flow_cmd_fill_stats(flow, skb);
853 if (err)
854 goto error;
855
74ed7ab9
JS		 856 if (should_fill_actions(ufid_flags)) {
857 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
858 if (err)
859 goto error;
860 }
ccb1352e 861
053c095a
JB		 862 genlmsg_end(skb, ovs_header);
863 return 0;
ccb1352e 864
ccb1352e
JG		 865error:
866 genlmsg_cancel(skb, ovs_header);
867 return err;
868}
869
0e9796b4
JR		 870/* May not be called with RCU read lock. */
871static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
74ed7ab9 872 const struct sw_flow_id *sfid,
fb5d1e9e 873 struct genl_info *info,
74ed7ab9
JS		 874 bool always,
875 uint32_t ufid_flags)
ccb1352e 876{
fb5d1e9e 877 struct sk_buff *skb;
74ed7ab9 878 size_t len;
ccb1352e 879
9b67aa4a 880 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
fb5d1e9e
JR		 881 return NULL;
882
74ed7ab9 883 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
551ddc05 884 skb = genlmsg_new(len, GFP_KERNEL);
fb5d1e9e
JR		 885 if (!skb)
886 return ERR_PTR(-ENOMEM);
887
888 return skb;
ccb1352e
JG		 889}
890
0e9796b4
JR		 891/* Called with ovs_mutex. */
892static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
893 int dp_ifindex,
894 struct genl_info *info, u8 cmd,
74ed7ab9 895 bool always, u32 ufid_flags)
ccb1352e
JG		 896{
897 struct sk_buff *skb;
898 int retval;
899
74ed7ab9
JS		 900 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
901 &flow->id, info, always, ufid_flags);
d0e992aa 902 if (IS_ERR_OR_NULL(skb))
fb5d1e9e 903 return skb;
ccb1352e 904
0e9796b4
JR		 905 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
906 info->snd_portid, info->snd_seq, 0,
74ed7ab9 907 cmd, ufid_flags);
8ffeb03f
PA		 908 if (WARN_ON_ONCE(retval < 0)) {
909 kfree_skb(skb);
910 skb = ERR_PTR(retval);
911 }
ccb1352e
JG		 912 return skb;
913}
914
37bdc87b 915static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
ccb1352e 916{
7f8a436e 917 struct net *net = sock_net(skb->sk);
ccb1352e
JG		 918 struct nlattr **a = info->attrs;
919 struct ovs_header *ovs_header = info->userhdr;
74ed7ab9 920 struct sw_flow *flow = NULL, *new_flow;
03f0d916 921 struct sw_flow_mask mask;
ccb1352e
JG		 922 struct sk_buff *reply;
923 struct datapath *dp;
37bdc87b 924 struct sw_flow_actions *acts;
03f0d916 925 struct sw_flow_match match;
74ed7ab9 926 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
ccb1352e 927 int error;
05da5898 928 bool log = !a[OVS_FLOW_ATTR_PROBE];
ccb1352e 929
893f139b 930 /* Must have key and actions. */
ccb1352e 931 error = -EINVAL;
426cda5c 932 if (!a[OVS_FLOW_ATTR_KEY]) {
05da5898 933 OVS_NLERR(log, "Flow key attr not present in new flow.");
ccb1352e 934 goto error;
426cda5c
JG		 935 }
936 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
05da5898 937 OVS_NLERR(log, "Flow actions attr not present in new flow.");
893f139b 938 goto error;
426cda5c 939 }
03f0d916 940
893f139b
JR		 941 /* Most of the time we need to allocate a new flow, do it before
942 * locking.
943 */
944 new_flow = ovs_flow_alloc();
945 if (IS_ERR(new_flow)) {
946 error = PTR_ERR(new_flow);
947 goto error;
948 }
949
950 /* Extract key. */
2279994d 951 ovs_match_init(&match, &new_flow->key, false, &mask);
c2ac6673 952 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
05da5898 953 a[OVS_FLOW_ATTR_MASK], log);
ccb1352e 954 if (error)
893f139b 955 goto err_kfree_flow;
ccb1352e 956
74ed7ab9
JS		 957 /* Extract flow identifier. */
958 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
190aa3e7 959 &new_flow->key, log);
74ed7ab9
JS		 960 if (error)
961 goto err_kfree_flow;
74f84a57 962
190aa3e7 963 /* unmasked key is needed to match when ufid is not used. */
964 if (ovs_identifier_is_key(&new_flow->id))
965 match.key = new_flow->id.unmasked_key;
966
967 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
968
893f139b 969 /* Validate actions. */
7f8a436e
JS		 970 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
971 &new_flow->key, &acts, log);
37bdc87b 972 if (error) {
05da5898 973 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
2fdb957d 974 goto err_kfree_flow;
893f139b
JR		 975 }
976
74ed7ab9
JS		 977 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
978 ufid_flags);
893f139b
JR		 979 if (IS_ERR(reply)) {
980 error = PTR_ERR(reply);
981 goto err_kfree_acts;
ccb1352e
JG		 982 }
983
8e4e1713 984 ovs_lock();
7f8a436e 985 dp = get_dp(net, ovs_header->dp_ifindex);
893f139b
JR		 986 if (unlikely(!dp)) {
987 error = -ENODEV;
8e4e1713 988 goto err_unlock_ovs;
893f139b 989 }
74ed7ab9 990
03f0d916 991 /* Check if this is a duplicate flow */
74ed7ab9
JS		 992 if (ovs_identifier_is_ufid(&new_flow->id))
993 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
994 if (!flow)
190aa3e7 995 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
893f139b
JR		 996 if (likely(!flow)) {
997 rcu_assign_pointer(new_flow->sf_acts, acts);
ccb1352e
JG		 998
999 /* Put flow in bucket. */
893f139b
JR		 1000 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1001 if (unlikely(error)) {
618ed0c8 1002 acts = NULL;
893f139b
JR		 1003 goto err_unlock_ovs;
1004 }
1005
1006 if (unlikely(reply)) {
1007 error = ovs_flow_cmd_fill_info(new_flow,
1008 ovs_header->dp_ifindex,
1009 reply, info->snd_portid,
1010 info->snd_seq, 0,
74ed7ab9
JS		 1011 OVS_FLOW_CMD_NEW,
1012 ufid_flags);
893f139b 1013 BUG_ON(error < 0);
618ed0c8 1014 }
893f139b 1015 ovs_unlock();
ccb1352e 1016 } else {
37bdc87b
JR		 1017 struct sw_flow_actions *old_acts;
1018
ccb1352e
JG		 1019 /* Bail out if we're not allowed to modify an existing flow.
1020 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1021 * because Generic Netlink treats the latter as a dump
1022 * request. We also accept NLM_F_EXCL in case that bug ever
1023 * gets fixed.
1024 */
893f139b
JR		 1025 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1026 | NLM_F_EXCL))) {
1027 error = -EEXIST;
8e4e1713 1028 goto err_unlock_ovs;
893f139b 1029 }
74ed7ab9
JS		 1030 /* The flow identifier has to be the same for flow updates.
1031 * Look for any overlapping flow.
1032 */
1033 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1034 if (ovs_identifier_is_key(&flow->id))
1035 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1036 &match);
1037 else /* UFID matches but key is different */
1038 flow = NULL;
4a46b24e
AW		 1039 if (!flow) {
1040 error = -ENOENT;
1041 goto err_unlock_ovs;
1042 }
893f139b 1043 }
37bdc87b
JR		 1044 /* Update actions. */
1045 old_acts = ovsl_dereference(flow->sf_acts);
1046 rcu_assign_pointer(flow->sf_acts, acts);
37bdc87b 1047
893f139b
JR		 1048 if (unlikely(reply)) {
1049 error = ovs_flow_cmd_fill_info(flow,
1050 ovs_header->dp_ifindex,
1051 reply, info->snd_portid,
1052 info->snd_seq, 0,
74ed7ab9
JS		 1053 OVS_FLOW_CMD_NEW,
1054 ufid_flags);
893f139b
JR		 1055 BUG_ON(error < 0);
1056 }
1057 ovs_unlock();
37bdc87b 1058
34ae932a 1059 ovs_nla_free_flow_actions_rcu(old_acts);
893f139b 1060 ovs_flow_free(new_flow, false);
37bdc87b 1061 }
893f139b
JR		 1062
1063 if (reply)
1064 ovs_notify(&dp_flow_genl_family, reply, info);
37bdc87b
JR		 1065 return 0;
1066
37bdc87b
JR		 1067err_unlock_ovs:
1068 ovs_unlock();
893f139b
JR		 1069 kfree_skb(reply);
1070err_kfree_acts:
34ae932a 1071 ovs_nla_free_flow_actions(acts);
893f139b
JR		 1072err_kfree_flow:
1073 ovs_flow_free(new_flow, false);
37bdc87b
JR		 1074error:
1075 return error;
1076}
ccb1352e 1077
2fdb957d 1078/* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
26063790 1079static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
7f8a436e 1080 const struct nlattr *a,
6b205b2c 1081 const struct sw_flow_key *key,
05da5898
JR		 1082 const struct sw_flow_mask *mask,
1083 bool log)
6b205b2c
JG		 1084{
1085 struct sw_flow_actions *acts;
1086 struct sw_flow_key masked_key;
1087 int error;
1088
ae5f2fb1 1089 ovs_flow_mask_key(&masked_key, key, true, mask);
7f8a436e 1090 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
6b205b2c 1091 if (error) {
05da5898
JR		 1092 OVS_NLERR(log,
1093 "Actions may not be safe on all matching packets");
6b205b2c
JG		 1094 return ERR_PTR(error);
1095 }
1096
1097 return acts;
1098}
1099
9cc9a5cb
TZ		 1100/* Factor out match-init and action-copy to avoid
1101 * "Wframe-larger-than=1024" warning. Because mask is only
1102 * used to get actions, we new a function to save some
1103 * stack space.
1104 *
1105 * If there are not key and action attrs, we return 0
1106 * directly. In the case, the caller will also not use the
1107 * match as before. If there is action attr, we try to get
1108 * actions and save them to *acts. Before returning from
1109 * the function, we reset the match->mask pointer. Because
1110 * we should not to return match object with dangling reference
1111 * to mask.
1112 * */
26063790
AB		 1113static noinline_for_stack int
1114ovs_nla_init_match_and_action(struct net *net,
1115 struct sw_flow_match *match,
1116 struct sw_flow_key *key,
1117 struct nlattr **a,
1118 struct sw_flow_actions **acts,
1119 bool log)
9cc9a5cb
TZ		 1120{
1121 struct sw_flow_mask mask;
1122 int error = 0;
1123
1124 if (a[OVS_FLOW_ATTR_KEY]) {
1125 ovs_match_init(match, key, true, &mask);
1126 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1127 a[OVS_FLOW_ATTR_MASK], log);
1128 if (error)
1129 goto error;
1130 }
1131
1132 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1133 if (!a[OVS_FLOW_ATTR_KEY]) {
1134 OVS_NLERR(log,
1135 "Flow key attribute not present in set flow.");
5829e62a
CJ		 1136 error = -EINVAL;
1137 goto error;
9cc9a5cb
TZ		 1138 }
1139
1140 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1141 &mask, log);
1142 if (IS_ERR(*acts)) {
1143 error = PTR_ERR(*acts);
1144 goto error;
1145 }
1146 }
1147
1148 /* On success, error is 0. */
1149error:
1150 match->mask = NULL;
1151 return error;
1152}
1153
37bdc87b
JR		 1154static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1155{
7f8a436e 1156 struct net *net = sock_net(skb->sk);
37bdc87b
JR		 1157 struct nlattr **a = info->attrs;
1158 struct ovs_header *ovs_header = info->userhdr;
6b205b2c 1159 struct sw_flow_key key;
37bdc87b 1160 struct sw_flow *flow;
37bdc87b
JR		 1161 struct sk_buff *reply = NULL;
1162 struct datapath *dp;
893f139b 1163 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
37bdc87b 1164 struct sw_flow_match match;
74ed7ab9
JS		 1165 struct sw_flow_id sfid;
1166 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
6f15cdbf 1167 int error = 0;
05da5898 1168 bool log = !a[OVS_FLOW_ATTR_PROBE];
74ed7ab9 1169 bool ufid_present;
37bdc87b 1170
74ed7ab9 1171 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
9cc9a5cb 1172 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
6f15cdbf
SG		 1173 OVS_NLERR(log,
1174 "Flow set message rejected, Key attribute missing.");
9cc9a5cb 1175 return -EINVAL;
6f15cdbf 1176 }
9cc9a5cb
TZ		 1177
1178 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1179 &acts, log);
37bdc87b
JR		 1180 if (error)
1181 goto error;
1182
9cc9a5cb 1183 if (acts) {
2fdb957d 1184 /* Can allocate before locking if have acts. */
74ed7ab9
JS		 1185 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1186 ufid_flags);
893f139b
JR		 1187 if (IS_ERR(reply)) {
1188 error = PTR_ERR(reply);
1189 goto err_kfree_acts;
be52c9e9 1190 }
37bdc87b 1191 }
0e9796b4 1192
37bdc87b 1193 ovs_lock();
7f8a436e 1194 dp = get_dp(net, ovs_header->dp_ifindex);
893f139b
JR		 1195 if (unlikely(!dp)) {
1196 error = -ENODEV;
37bdc87b 1197 goto err_unlock_ovs;
893f139b 1198 }
37bdc87b 1199 /* Check that the flow exists. */
74ed7ab9
JS		 1200 if (ufid_present)
1201 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1202 else
1203 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
893f139b
JR		 1204 if (unlikely(!flow)) {
1205 error = -ENOENT;
37bdc87b 1206 goto err_unlock_ovs;
893f139b 1207 }
4a46b24e 1208
37bdc87b 1209 /* Update actions, if present. */
893f139b 1210 if (likely(acts)) {
37bdc87b
JR		 1211 old_acts = ovsl_dereference(flow->sf_acts);
1212 rcu_assign_pointer(flow->sf_acts, acts);
893f139b
JR		 1213
1214 if (unlikely(reply)) {
1215 error = ovs_flow_cmd_fill_info(flow,
1216 ovs_header->dp_ifindex,
1217 reply, info->snd_portid,
1218 info->snd_seq, 0,
804fe108 1219 OVS_FLOW_CMD_SET,
74ed7ab9 1220 ufid_flags);
893f139b
JR		 1221 BUG_ON(error < 0);
1222 }
1223 } else {
1224 /* Could not alloc without acts before locking. */
1225 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
804fe108 1226 info, OVS_FLOW_CMD_SET, false,
74ed7ab9
JS		 1227 ufid_flags);
1228
b5ffe634 1229 if (IS_ERR(reply)) {
893f139b
JR		 1230 error = PTR_ERR(reply);
1231 goto err_unlock_ovs;
1232 }
ccb1352e 1233 }
37bdc87b 1234
37bdc87b
JR		 1235 /* Clear stats. */
1236 if (a[OVS_FLOW_ATTR_CLEAR])
1237 ovs_flow_stats_clear(flow);
8e4e1713 1238 ovs_unlock();
ccb1352e 1239
893f139b
JR		 1240 if (reply)
1241 ovs_notify(&dp_flow_genl_family, reply, info);
1242 if (old_acts)
34ae932a 1243 ovs_nla_free_flow_actions_rcu(old_acts);
fb5d1e9e 1244
ccb1352e
JG		 1245 return 0;
1246
8e4e1713
PS		 1247err_unlock_ovs:
1248 ovs_unlock();
893f139b
JR		 1249 kfree_skb(reply);
1250err_kfree_acts:
34ae932a 1251 ovs_nla_free_flow_actions(acts);
ccb1352e
JG		 1252error:
1253 return error;
1254}
1255
1256static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1257{
1258 struct nlattr **a = info->attrs;
1259 struct ovs_header *ovs_header = info->userhdr;
c2ac6673 1260 struct net *net = sock_net(skb->sk);
ccb1352e
JG		 1261 struct sw_flow_key key;
1262 struct sk_buff *reply;
1263 struct sw_flow *flow;
1264 struct datapath *dp;
03f0d916 1265 struct sw_flow_match match;
74ed7ab9
JS		 1266 struct sw_flow_id ufid;
1267 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1268 int err = 0;
05da5898 1269 bool log = !a[OVS_FLOW_ATTR_PROBE];
74ed7ab9 1270 bool ufid_present;
ccb1352e 1271
74ed7ab9
JS		 1272 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1273 if (a[OVS_FLOW_ATTR_KEY]) {
2279994d 1274 ovs_match_init(&match, &key, true, NULL);
c2ac6673 1275 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
74ed7ab9
JS		 1276 log);
1277 } else if (!ufid_present) {
05da5898
JR		 1278 OVS_NLERR(log,
1279 "Flow get message rejected, Key attribute missing.");
74ed7ab9 1280 err = -EINVAL;
03f0d916 1281 }
ccb1352e
JG		 1282 if (err)
1283 return err;
1284
8e4e1713 1285 ovs_lock();
46df7b81 1286 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
8e4e1713
PS		 1287 if (!dp) {
1288 err = -ENODEV;
1289 goto unlock;
1290 }
ccb1352e 1291
74ed7ab9
JS		 1292 if (ufid_present)
1293 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1294 else
1295 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
4a46b24e 1296 if (!flow) {
8e4e1713
PS		 1297 err = -ENOENT;
1298 goto unlock;
1299 }
ccb1352e 1300
0e9796b4 1301 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
804fe108 1302 OVS_FLOW_CMD_GET, true, ufid_flags);
8e4e1713
PS		 1303 if (IS_ERR(reply)) {
1304 err = PTR_ERR(reply);
1305 goto unlock;
1306 }
ccb1352e 1307
8e4e1713 1308 ovs_unlock();
ccb1352e 1309 return genlmsg_reply(reply, info);
8e4e1713
PS		 1310unlock:
1311 ovs_unlock();
1312 return err;
ccb1352e
JG		 1313}
1314
1315static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1316{
1317 struct nlattr **a = info->attrs;
1318 struct ovs_header *ovs_header = info->userhdr;
c2ac6673 1319 struct net *net = sock_net(skb->sk);
ccb1352e
JG		 1320 struct sw_flow_key key;
1321 struct sk_buff *reply;
74ed7ab9 1322 struct sw_flow *flow = NULL;
ccb1352e 1323 struct datapath *dp;
03f0d916 1324 struct sw_flow_match match;
74ed7ab9
JS		 1325 struct sw_flow_id ufid;
1326 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
ccb1352e 1327 int err;
05da5898 1328 bool log = !a[OVS_FLOW_ATTR_PROBE];
74ed7ab9 1329 bool ufid_present;
ccb1352e 1330
74ed7ab9
JS		 1331 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1332 if (a[OVS_FLOW_ATTR_KEY]) {
2279994d 1333 ovs_match_init(&match, &key, true, NULL);
c2ac6673
JS		 1334 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1335 NULL, log);
aed06778
JR		 1336 if (unlikely(err))
1337 return err;
1338 }
1339
8e4e1713 1340 ovs_lock();
46df7b81 1341 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
aed06778 1342 if (unlikely(!dp)) {
8e4e1713
PS		 1343 err = -ENODEV;
1344 goto unlock;
1345 }
46df7b81 1346
74ed7ab9 1347 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
b637e498 1348 err = ovs_flow_tbl_flush(&dp->table);
8e4e1713
PS		 1349 goto unlock;
1350 }
03f0d916 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 (unlikely(!flow)) {
8e4e1713
PS		 1357 err = -ENOENT;
1358 goto unlock;
1359 }
ccb1352e 1360
b637e498 1361 ovs_flow_tbl_remove(&dp->table, flow);
aed06778 1362 ovs_unlock();
ccb1352e 1363
aed06778 1364 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
74ed7ab9 1365 &flow->id, info, false, ufid_flags);
aed06778 1366 if (likely(reply)) {
b90f5aa4 1367 if (!IS_ERR(reply)) {
aed06778
JR		 1368 rcu_read_lock(); /*To keep RCU checker happy. */
1369 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1370 reply, info->snd_portid,
1371 info->snd_seq, 0,
74ed7ab9
JS		 1372 OVS_FLOW_CMD_DEL,
1373 ufid_flags);
aed06778 1374 rcu_read_unlock();
8a574f86
PA		 1375 if (WARN_ON_ONCE(err < 0)) {
1376 kfree_skb(reply);
1377 goto out_free;
1378 }
aed06778
JR		 1379
1380 ovs_notify(&dp_flow_genl_family, reply, info);
1381 } else {
1382 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1383 }
fb5d1e9e 1384 }
ccb1352e 1385
8a574f86 1386out_free:
aed06778 1387 ovs_flow_free(flow, true);
ccb1352e 1388 return 0;
8e4e1713
PS		 1389unlock:
1390 ovs_unlock();
1391 return err;
ccb1352e
JG		 1392}
1393
1394static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1395{
74ed7ab9 1396 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
ccb1352e 1397 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
b637e498 1398 struct table_instance *ti;
ccb1352e 1399 struct datapath *dp;
74ed7ab9
JS		 1400 u32 ufid_flags;
1401 int err;
1402
8cb08174
JB		 1403 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1404 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
74ed7ab9
JS		 1405 if (err)
1406 return err;
1407 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
ccb1352e 1408
d57170b1 1409 rcu_read_lock();
cc3a5ae6 1410 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
8e4e1713 1411 if (!dp) {
d57170b1 1412 rcu_read_unlock();
ccb1352e 1413 return -ENODEV;
8e4e1713 1414 }
ccb1352e 1415
b637e498 1416 ti = rcu_dereference(dp->table.ti);
ccb1352e
JG		 1417 for (;;) {
1418 struct sw_flow *flow;
1419 u32 bucket, obj;
1420
1421 bucket = cb->args[0];
1422 obj = cb->args[1];
b637e498 1423 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
ccb1352e
JG		 1424 if (!flow)
1425 break;
1426
0e9796b4 1427 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
15e47304 1428 NETLINK_CB(cb->skb).portid,
ccb1352e 1429 cb->nlh->nlmsg_seq, NLM_F_MULTI,
804fe108 1430 OVS_FLOW_CMD_GET, ufid_flags) < 0)
ccb1352e
JG		 1431 break;
1432
1433 cb->args[0] = bucket;
1434 cb->args[1] = obj;
1435 }
d57170b1 1436 rcu_read_unlock();
ccb1352e
JG		 1437 return skb->len;
1438}
1439
0c200ef9
PS		 1440static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1441 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
05da5898 1442 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
0c200ef9
PS		 1443 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1444 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
05da5898 1445 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
74ed7ab9
JS		 1446 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1447 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
0c200ef9
PS		 1448};
1449
48e48a70 1450static const struct genl_ops dp_flow_genl_ops[] = {
ccb1352e 1451 { .cmd = OVS_FLOW_CMD_NEW,
ef6243ac 1452 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1453 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
37bdc87b 1454 .doit = ovs_flow_cmd_new
ccb1352e
JG		 1455 },
1456 { .cmd = OVS_FLOW_CMD_DEL,
ef6243ac 1457 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1458 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 1459 .doit = ovs_flow_cmd_del
1460 },
1461 { .cmd = OVS_FLOW_CMD_GET,
ef6243ac 1462 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
ccb1352e 1463 .flags = 0, /* OK for unprivileged users. */
ccb1352e
JG		 1464 .doit = ovs_flow_cmd_get,
1465 .dumpit = ovs_flow_cmd_dump
1466 },
1467 { .cmd = OVS_FLOW_CMD_SET,
ef6243ac 1468 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1469 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
37bdc87b 1470 .doit = ovs_flow_cmd_set,
ccb1352e
JG		 1471 },
1472};
1473
56989f6d 1474static struct genl_family dp_flow_genl_family __ro_after_init = {
ccb1352e 1475 .hdrsize = sizeof(struct ovs_header),
0c200ef9
PS		 1476 .name = OVS_FLOW_FAMILY,
1477 .version = OVS_FLOW_VERSION,
1478 .maxattr = OVS_FLOW_ATTR_MAX,
3b0f31f2 1479 .policy = flow_policy,
3a4e0d6a
PS		 1480 .netnsok = true,
1481 .parallel_ops = true,
0c200ef9
PS		 1482 .ops = dp_flow_genl_ops,
1483 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1484 .mcgrps = &ovs_dp_flow_multicast_group,
1485 .n_mcgrps = 1,
489111e5 1486 .module = THIS_MODULE,
ccb1352e
JG		 1487};
1488
c3ff8cfe
TG		 1489static size_t ovs_dp_cmd_msg_size(void)
1490{
1491 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1492
1493 msgsize += nla_total_size(IFNAMSIZ);
66c7a5ee
ND		 1494 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1495 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
45fb9c35 1496 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
c3ff8cfe
TG		 1497
1498 return msgsize;
1499}
1500
8ec609d8 1501/* Called with ovs_mutex. */
ccb1352e 1502static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
15e47304 1503 u32 portid, u32 seq, u32 flags, u8 cmd)
ccb1352e
JG		 1504{
1505 struct ovs_header *ovs_header;
1506 struct ovs_dp_stats dp_stats;
1bd7116f 1507 struct ovs_dp_megaflow_stats dp_megaflow_stats;
ccb1352e
JG		 1508 int err;
1509
15e47304 1510 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
ccb1352e
JG		 1511 flags, cmd);
1512 if (!ovs_header)
1513 goto error;
1514
1515 ovs_header->dp_ifindex = get_dpifindex(dp);
1516
ccb1352e 1517 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
ccb1352e
JG		 1518 if (err)
1519 goto nla_put_failure;
1520
1bd7116f 1521 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
66c7a5ee
ND		 1522 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1523 &dp_stats, OVS_DP_ATTR_PAD))
1bd7116f
AZ		 1524 goto nla_put_failure;
1525
66c7a5ee
ND		 1526 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1527 sizeof(struct ovs_dp_megaflow_stats),
1528 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
028d6a67 1529 goto nla_put_failure;
ccb1352e 1530
43d4be9c
TG		 1531 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1532 goto nla_put_failure;
1533
053c095a
JB		 1534 genlmsg_end(skb, ovs_header);
1535 return 0;
ccb1352e
JG		 1536
1537nla_put_failure:
1538 genlmsg_cancel(skb, ovs_header);
1539error:
1540 return -EMSGSIZE;
1541}
1542
263ea090 1543static struct sk_buff *ovs_dp_cmd_alloc_info(void)
ccb1352e 1544{
551ddc05 1545 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
ccb1352e
JG		 1546}
1547
bb6f9a70 1548/* Called with rcu_read_lock or ovs_mutex. */
46df7b81 1549static struct datapath *lookup_datapath(struct net *net,
12eb18f7 1550 const struct ovs_header *ovs_header,
ccb1352e
JG		 1551 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1552{
1553 struct datapath *dp;
1554
1555 if (!a[OVS_DP_ATTR_NAME])
46df7b81 1556 dp = get_dp(net, ovs_header->dp_ifindex);
ccb1352e
JG		 1557 else {
1558 struct vport *vport;
1559
46df7b81 1560 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
ccb1352e 1561 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
ccb1352e
JG		 1562 }
1563 return dp ? dp : ERR_PTR(-ENODEV);
1564}
1565
44da5ae5
TG		 1566static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1567{
1568 struct datapath *dp;
1569
1570 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
3c7eacfc 1571 if (IS_ERR(dp))
44da5ae5
TG		 1572 return;
1573
1574 WARN(dp->user_features, "Dropping previously announced user features\n");
1575 dp->user_features = 0;
1576}
1577
95a7233c
PB		 1578DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1579
1580static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
43d4be9c 1581{
95a7233c
PB		 1582 u32 user_features = 0;
1583
1584 if (a[OVS_DP_ATTR_USER_FEATURES]) {
1585 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1586
1587 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1588 OVS_DP_F_UNALIGNED |
1589 OVS_DP_F_TC_RECIRC_SHARING))
1590 return -EOPNOTSUPP;
1591
1592#if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1593 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1594 return -EOPNOTSUPP;
1595#endif
1596 }
1597
1598 dp->user_features = user_features;
1599
1600 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1601 static_branch_enable(&tc_recirc_sharing_support);
1602 else
1603 static_branch_disable(&tc_recirc_sharing_support);
1604
1605 return 0;
43d4be9c
TG		 1606}
1607
eec62ead
TZ		 1608static int ovs_dp_stats_init(struct datapath *dp)
1609{
1610 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1611 if (!dp->stats_percpu)
1612 return -ENOMEM;
1613
1614 return 0;
1615}
1616
1617static int ovs_dp_vport_init(struct datapath *dp)
1618{
1619 int i;
1620
1621 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1622 sizeof(struct hlist_head),
1623 GFP_KERNEL);
1624 if (!dp->ports)
1625 return -ENOMEM;
1626
1627 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1628 INIT_HLIST_HEAD(&dp->ports[i]);
1629
1630 return 0;
1631}
1632
ccb1352e
JG		 1633static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1634{
1635 struct nlattr **a = info->attrs;
1636 struct vport_parms parms;
1637 struct sk_buff *reply;
1638 struct datapath *dp;
1639 struct vport *vport;
46df7b81 1640 struct ovs_net *ovs_net;
eec62ead 1641 int err;
ccb1352e
JG		 1642
1643 err = -EINVAL;
1644 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1645 goto err;
1646
263ea090 1647 reply = ovs_dp_cmd_alloc_info();
6093ae9a
JR		 1648 if (!reply)
1649 return -ENOMEM;
ccb1352e
JG		 1650
1651 err = -ENOMEM;
1652 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1653 if (dp == NULL)
eec62ead 1654 goto err_destroy_reply;
46df7b81 1655
efd7ef1c 1656 ovs_dp_set_net(dp, sock_net(skb->sk));
ccb1352e
JG		 1657
1658 /* Allocate table. */
b637e498
PS		 1659 err = ovs_flow_tbl_init(&dp->table);
1660 if (err)
eec62ead 1661 goto err_destroy_dp;
ccb1352e 1662
eec62ead
TZ		 1663 err = ovs_dp_stats_init(dp);
1664 if (err)
ccb1352e 1665 goto err_destroy_table;
ccb1352e 1666
eec62ead
TZ		 1667 err = ovs_dp_vport_init(dp);
1668 if (err)
1669 goto err_destroy_stats;
15eac2a7 1670
96fbc13d
AZ		 1671 err = ovs_meters_init(dp);
1672 if (err)
eec62ead 1673 goto err_destroy_ports;
96fbc13d 1674
ccb1352e
JG		 1675 /* Set up our datapath device. */
1676 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1677 parms.type = OVS_VPORT_TYPE_INTERNAL;
1678 parms.options = NULL;
1679 parms.dp = dp;
1680 parms.port_no = OVSP_LOCAL;
5cd667b0 1681 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
ccb1352e 1682
95a7233c
PB		 1683 err = ovs_dp_change(dp, a);
1684 if (err)
1685 goto err_destroy_meters;
43d4be9c 1686
6093ae9a
JR		 1687 /* So far only local changes have been made, now need the lock. */
1688 ovs_lock();
1689
ccb1352e
JG		 1690 vport = new_vport(&parms);
1691 if (IS_ERR(vport)) {
1692 err = PTR_ERR(vport);
1693 if (err == -EBUSY)
1694 err = -EEXIST;
1695
44da5ae5
TG		 1696 if (err == -EEXIST) {
1697 /* An outdated user space instance that does not understand
1698 * the concept of user_features has attempted to create a new
1699 * datapath and is likely to reuse it. Drop all user features.
1700 */
1701 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1702 ovs_dp_reset_user_features(skb, info);
1703 }
1704
4c76bf69 1705 ovs_unlock();
96fbc13d 1706 goto err_destroy_meters;
ccb1352e
JG		 1707 }
1708
6093ae9a
JR		 1709 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1710 info->snd_seq, 0, OVS_DP_CMD_NEW);
1711 BUG_ON(err < 0);
ccb1352e 1712
46df7b81 1713 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
59a35d60 1714 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
8e4e1713
PS		 1715
1716 ovs_unlock();
ccb1352e 1717
2a94fe48 1718 ovs_notify(&dp_datapath_genl_family, reply, info);
ccb1352e
JG		 1719 return 0;
1720
96fbc13d 1721err_destroy_meters:
96fbc13d 1722 ovs_meters_exit(dp);
eec62ead 1723err_destroy_ports:
15eac2a7 1724 kfree(dp->ports);
eec62ead 1725err_destroy_stats:
ccb1352e
JG		 1726 free_percpu(dp->stats_percpu);
1727err_destroy_table:
9b996e54 1728 ovs_flow_tbl_destroy(&dp->table);
eec62ead 1729err_destroy_dp:
ccb1352e 1730 kfree(dp);
eec62ead 1731err_destroy_reply:
6093ae9a 1732 kfree_skb(reply);
ccb1352e
JG		 1733err:
1734 return err;
1735}
1736
8e4e1713 1737/* Called with ovs_mutex. */
46df7b81 1738static void __dp_destroy(struct datapath *dp)
ccb1352e 1739{
15eac2a7 1740 int i;
ccb1352e 1741
15eac2a7
PS		 1742 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1743 struct vport *vport;
b67bfe0d 1744 struct hlist_node *n;
15eac2a7 1745
b67bfe0d 1746 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
15eac2a7
PS		 1747 if (vport->port_no != OVSP_LOCAL)
1748 ovs_dp_detach_port(vport);
1749 }
ccb1352e 1750
59a35d60 1751 list_del_rcu(&dp->list_node);
ccb1352e 1752
8e4e1713 1753 /* OVSP_LOCAL is datapath internal port. We need to make sure that
e80857cc 1754 * all ports in datapath are destroyed first before freeing datapath.
ccb1352e 1755 */
8e4e1713 1756 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
ccb1352e 1757
e80857cc 1758 /* RCU destroy the flow table */
ccb1352e 1759 call_rcu(&dp->rcu, destroy_dp_rcu);
46df7b81
PS		 1760}
1761
1762static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1763{
1764 struct sk_buff *reply;
1765 struct datapath *dp;
1766 int err;
1767
263ea090 1768 reply = ovs_dp_cmd_alloc_info();
6093ae9a
JR		 1769 if (!reply)
1770 return -ENOMEM;
1771
8e4e1713 1772 ovs_lock();
46df7b81
PS		 1773 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1774 err = PTR_ERR(dp);
1775 if (IS_ERR(dp))
6093ae9a 1776 goto err_unlock_free;
46df7b81 1777
6093ae9a
JR		 1778 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1779 info->snd_seq, 0, OVS_DP_CMD_DEL);
1780 BUG_ON(err < 0);
46df7b81
PS		 1781
1782 __dp_destroy(dp);
8e4e1713 1783 ovs_unlock();
ccb1352e 1784
2a94fe48 1785 ovs_notify(&dp_datapath_genl_family, reply, info);
ccb1352e
JG		 1786
1787 return 0;
6093ae9a
JR		 1788
1789err_unlock_free:
8e4e1713 1790 ovs_unlock();
6093ae9a 1791 kfree_skb(reply);
8e4e1713 1792 return err;
ccb1352e
JG		 1793}
1794
1795static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1796{
1797 struct sk_buff *reply;
1798 struct datapath *dp;
1799 int err;
1800
263ea090 1801 reply = ovs_dp_cmd_alloc_info();
6093ae9a
JR		 1802 if (!reply)
1803 return -ENOMEM;
1804
8e4e1713 1805 ovs_lock();
46df7b81 1806 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
8e4e1713 1807 err = PTR_ERR(dp);
ccb1352e 1808 if (IS_ERR(dp))
6093ae9a 1809 goto err_unlock_free;
ccb1352e 1810
95a7233c
PB		 1811 err = ovs_dp_change(dp, info->attrs);
1812 if (err)
1813 goto err_unlock_free;
43d4be9c 1814
6093ae9a 1815 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
804fe108 1816 info->snd_seq, 0, OVS_DP_CMD_SET);
6093ae9a 1817 BUG_ON(err < 0);
ccb1352e 1818
8e4e1713 1819 ovs_unlock();
2a94fe48 1820 ovs_notify(&dp_datapath_genl_family, reply, info);
ccb1352e
JG		 1821
1822 return 0;
6093ae9a
JR		 1823
1824err_unlock_free:
8e4e1713 1825 ovs_unlock();
6093ae9a 1826 kfree_skb(reply);
8e4e1713 1827 return err;
ccb1352e
JG		 1828}
1829
1830static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1831{
1832 struct sk_buff *reply;
1833 struct datapath *dp;
8e4e1713 1834 int err;
ccb1352e 1835
263ea090 1836 reply = ovs_dp_cmd_alloc_info();
6093ae9a
JR		 1837 if (!reply)
1838 return -ENOMEM;
1839
8ec609d8 1840 ovs_lock();
46df7b81 1841 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
8e4e1713
PS		 1842 if (IS_ERR(dp)) {
1843 err = PTR_ERR(dp);
6093ae9a 1844 goto err_unlock_free;
8e4e1713 1845 }
6093ae9a 1846 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
804fe108 1847 info->snd_seq, 0, OVS_DP_CMD_GET);
6093ae9a 1848 BUG_ON(err < 0);
8ec609d8 1849 ovs_unlock();
ccb1352e
JG		 1850
1851 return genlmsg_reply(reply, info);
8e4e1713 1852
6093ae9a 1853err_unlock_free:
8ec609d8 1854 ovs_unlock();
6093ae9a 1855 kfree_skb(reply);
8e4e1713 1856 return err;
ccb1352e
JG		 1857}
1858
1859static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1860{
46df7b81 1861 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
ccb1352e
JG		 1862 struct datapath *dp;
1863 int skip = cb->args[0];
1864 int i = 0;
1865
8ec609d8
PS		 1866 ovs_lock();
1867 list_for_each_entry(dp, &ovs_net->dps, list_node) {
77676fdb 1868 if (i >= skip &&
15e47304 1869 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
ccb1352e 1870 cb->nlh->nlmsg_seq, NLM_F_MULTI,
804fe108 1871 OVS_DP_CMD_GET) < 0)
ccb1352e
JG		 1872 break;
1873 i++;
1874 }
8ec609d8 1875 ovs_unlock();
ccb1352e
JG		 1876
1877 cb->args[0] = i;
1878
1879 return skb->len;
1880}
1881
0c200ef9
PS		 1882static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1883 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1884 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1885 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1886};
1887
48e48a70 1888static const struct genl_ops dp_datapath_genl_ops[] = {
ccb1352e 1889 { .cmd = OVS_DP_CMD_NEW,
ef6243ac 1890 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1891 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 1892 .doit = ovs_dp_cmd_new
1893 },
1894 { .cmd = OVS_DP_CMD_DEL,
ef6243ac 1895 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1896 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 1897 .doit = ovs_dp_cmd_del
1898 },
1899 { .cmd = OVS_DP_CMD_GET,
ef6243ac 1900 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
ccb1352e 1901 .flags = 0, /* OK for unprivileged users. */
ccb1352e
JG		 1902 .doit = ovs_dp_cmd_get,
1903 .dumpit = ovs_dp_cmd_dump
1904 },
1905 { .cmd = OVS_DP_CMD_SET,
ef6243ac 1906 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 1907 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 1908 .doit = ovs_dp_cmd_set,
1909 },
1910};
1911
56989f6d 1912static struct genl_family dp_datapath_genl_family __ro_after_init = {
ccb1352e 1913 .hdrsize = sizeof(struct ovs_header),
0c200ef9
PS		 1914 .name = OVS_DATAPATH_FAMILY,
1915 .version = OVS_DATAPATH_VERSION,
1916 .maxattr = OVS_DP_ATTR_MAX,
3b0f31f2 1917 .policy = datapath_policy,
3a4e0d6a
PS		 1918 .netnsok = true,
1919 .parallel_ops = true,
0c200ef9
PS		 1920 .ops = dp_datapath_genl_ops,
1921 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1922 .mcgrps = &ovs_dp_datapath_multicast_group,
1923 .n_mcgrps = 1,
489111e5 1924 .module = THIS_MODULE,
ccb1352e
JG		 1925};
1926
8e4e1713 1927/* Called with ovs_mutex or RCU read lock. */
ccb1352e 1928static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
9354d452 1929 struct net *net, u32 portid, u32 seq,
d4e4fdf9 1930 u32 flags, u8 cmd, gfp_t gfp)
ccb1352e
JG		 1931{
1932 struct ovs_header *ovs_header;
1933 struct ovs_vport_stats vport_stats;
1934 int err;
1935
15e47304 1936 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
ccb1352e
JG		 1937 flags, cmd);
1938 if (!ovs_header)
1939 return -EMSGSIZE;
1940
1941 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1942
028d6a67
DM		 1943 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1944 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
5cd667b0 1945 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
9354d452
JB		 1946 ovs_vport_name(vport)) ||
1947 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
028d6a67 1948 goto nla_put_failure;
ccb1352e 1949
9354d452 1950 if (!net_eq(net, dev_net(vport->dev))) {
d4e4fdf9 1951 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
9354d452
JB		 1952
1953 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1954 goto nla_put_failure;
1955 }
1956
ccb1352e 1957 ovs_vport_get_stats(vport, &vport_stats);
66c7a5ee
ND		 1958 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1959 sizeof(struct ovs_vport_stats), &vport_stats,
1960 OVS_VPORT_ATTR_PAD))
028d6a67 1961 goto nla_put_failure;
ccb1352e 1962
5cd667b0
AW		 1963 if (ovs_vport_get_upcall_portids(vport, skb))
1964 goto nla_put_failure;
1965
ccb1352e
JG		 1966 err = ovs_vport_get_options(vport, skb);
1967 if (err == -EMSGSIZE)
1968 goto error;
1969
053c095a
JB		 1970 genlmsg_end(skb, ovs_header);
1971 return 0;
ccb1352e
JG		 1972
1973nla_put_failure:
1974 err = -EMSGSIZE;
1975error:
1976 genlmsg_cancel(skb, ovs_header);
1977 return err;
1978}
1979
6093ae9a
JR		 1980static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1981{
1982 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1983}
1984
1985/* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
9354d452
JB		 1986struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1987 u32 portid, u32 seq, u8 cmd)
ccb1352e
JG		 1988{
1989 struct sk_buff *skb;
1990 int retval;
1991
d4e4fdf9 1992 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
ccb1352e
JG		 1993 if (!skb)
1994 return ERR_PTR(-ENOMEM);
1995
d4e4fdf9
GN		 1996 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
1997 GFP_KERNEL);
a9341512
JG		 1998 BUG_ON(retval < 0);
1999
ccb1352e
JG		 2000 return skb;
2001}
2002
8e4e1713 2003/* Called with ovs_mutex or RCU read lock. */
46df7b81 2004static struct vport *lookup_vport(struct net *net,
12eb18f7 2005 const struct ovs_header *ovs_header,
ccb1352e
JG		 2006 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2007{
2008 struct datapath *dp;
2009 struct vport *vport;
2010
9354d452
JB		 2011 if (a[OVS_VPORT_ATTR_IFINDEX])
2012 return ERR_PTR(-EOPNOTSUPP);
ccb1352e 2013 if (a[OVS_VPORT_ATTR_NAME]) {
46df7b81 2014 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
ccb1352e
JG		 2015 if (!vport)
2016 return ERR_PTR(-ENODEV);
651a68ea
BP		 2017 if (ovs_header->dp_ifindex &&
2018 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2019 return ERR_PTR(-ENODEV);
ccb1352e
JG		 2020 return vport;
2021 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2022 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2023
2024 if (port_no >= DP_MAX_PORTS)
2025 return ERR_PTR(-EFBIG);
2026
46df7b81 2027 dp = get_dp(net, ovs_header->dp_ifindex);
ccb1352e
JG		 2028 if (!dp)
2029 return ERR_PTR(-ENODEV);
2030
8e4e1713 2031 vport = ovs_vport_ovsl_rcu(dp, port_no);
ccb1352e 2032 if (!vport)
14408dba 2033 return ERR_PTR(-ENODEV);
ccb1352e
JG		 2034 return vport;
2035 } else
2036 return ERR_PTR(-EINVAL);
9354d452 2037
ccb1352e
JG		 2038}
2039
6b660c41 2040static unsigned int ovs_get_max_headroom(struct datapath *dp)
3a927bc7 2041{
6b660c41 2042 unsigned int dev_headroom, max_headroom = 0;
3a927bc7
PA		 2043 struct net_device *dev;
2044 struct vport *vport;
2045 int i;
2046
2047 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2048 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2049 dev = vport->dev;
2050 dev_headroom = netdev_get_fwd_headroom(dev);
2051 if (dev_headroom > max_headroom)
2052 max_headroom = dev_headroom;
2053 }
2054 }
2055
6b660c41
TY		 2056 return max_headroom;
2057}
2058
2059/* Called with ovs_mutex */
2060static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2061{
2062 struct vport *vport;
2063 int i;
2064
2065 dp->max_headroom = new_headroom;
3a927bc7
PA		 2066 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2067 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
6b660c41 2068 netdev_set_rx_headroom(vport->dev, new_headroom);
3a927bc7
PA		 2069}
2070
ccb1352e
JG		 2071static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2072{
2073 struct nlattr **a = info->attrs;
2074 struct ovs_header *ovs_header = info->userhdr;
2075 struct vport_parms parms;
2076 struct sk_buff *reply;
2077 struct vport *vport;
2078 struct datapath *dp;
6b660c41 2079 unsigned int new_headroom;
ccb1352e
JG		 2080 u32 port_no;
2081 int err;
2082
ccb1352e
JG		 2083 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2084 !a[OVS_VPORT_ATTR_UPCALL_PID])
6093ae9a 2085 return -EINVAL;
9354d452
JB		 2086 if (a[OVS_VPORT_ATTR_IFINDEX])
2087 return -EOPNOTSUPP;
6093ae9a
JR		 2088
2089 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2090 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2091 if (port_no >= DP_MAX_PORTS)
2092 return -EFBIG;
2093
2094 reply = ovs_vport_cmd_alloc_info();
2095 if (!reply)
2096 return -ENOMEM;
ccb1352e 2097
8e4e1713 2098 ovs_lock();
62b9c8d0 2099restart:
46df7b81 2100 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
ccb1352e
JG		 2101 err = -ENODEV;
2102 if (!dp)
6093ae9a 2103 goto exit_unlock_free;
ccb1352e 2104
6093ae9a 2105 if (port_no) {
8e4e1713 2106 vport = ovs_vport_ovsl(dp, port_no);
ccb1352e
JG		 2107 err = -EBUSY;
2108 if (vport)
6093ae9a 2109 goto exit_unlock_free;
ccb1352e
JG		 2110 } else {
2111 for (port_no = 1; ; port_no++) {
2112 if (port_no >= DP_MAX_PORTS) {
2113 err = -EFBIG;
6093ae9a 2114 goto exit_unlock_free;
ccb1352e 2115 }
8e4e1713 2116 vport = ovs_vport_ovsl(dp, port_no);
ccb1352e
JG		 2117 if (!vport)
2118 break;
2119 }
2120 }
2121
2122 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2123 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2124 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2125 parms.dp = dp;
2126 parms.port_no = port_no;
5cd667b0 2127 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
ccb1352e
JG		 2128
2129 vport = new_vport(&parms);
2130 err = PTR_ERR(vport);
62b9c8d0
TG		 2131 if (IS_ERR(vport)) {
2132 if (err == -EAGAIN)
2133 goto restart;
6093ae9a 2134 goto exit_unlock_free;
62b9c8d0 2135 }
ccb1352e 2136
9354d452
JB		 2137 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2138 info->snd_portid, info->snd_seq, 0,
d4e4fdf9 2139 OVS_VPORT_CMD_NEW, GFP_KERNEL);
3a927bc7 2140
6b660c41
TY		 2141 new_headroom = netdev_get_fwd_headroom(vport->dev);
2142
2143 if (new_headroom > dp->max_headroom)
2144 ovs_update_headroom(dp, new_headroom);
3a927bc7
PA		 2145 else
2146 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2147
6093ae9a
JR		 2148 BUG_ON(err < 0);
2149 ovs_unlock();
ed661185 2150
2a94fe48 2151 ovs_notify(&dp_vport_genl_family, reply, info);
6093ae9a 2152 return 0;
ccb1352e 2153
6093ae9a 2154exit_unlock_free:
8e4e1713 2155 ovs_unlock();
6093ae9a 2156 kfree_skb(reply);
ccb1352e
JG		 2157 return err;
2158}
2159
2160static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2161{
2162 struct nlattr **a = info->attrs;
2163 struct sk_buff *reply;
2164 struct vport *vport;
2165 int err;
2166
6093ae9a
JR		 2167 reply = ovs_vport_cmd_alloc_info();
2168 if (!reply)
2169 return -ENOMEM;
2170
8e4e1713 2171 ovs_lock();
46df7b81 2172 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
ccb1352e
JG		 2173 err = PTR_ERR(vport);
2174 if (IS_ERR(vport))
6093ae9a 2175 goto exit_unlock_free;
ccb1352e 2176
ccb1352e 2177 if (a[OVS_VPORT_ATTR_TYPE] &&
f44f3408 2178 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
ccb1352e 2179 err = -EINVAL;
6093ae9a 2180 goto exit_unlock_free;
a9341512
JG		 2181 }
2182
f44f3408 2183 if (a[OVS_VPORT_ATTR_OPTIONS]) {
ccb1352e 2184 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
f44f3408 2185 if (err)
6093ae9a 2186 goto exit_unlock_free;
f44f3408 2187 }
a9341512 2188
5cd667b0
AW		 2189
2190 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2191 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2192
2193 err = ovs_vport_set_upcall_portids(vport, ids);
2194 if (err)
2195 goto exit_unlock_free;
2196 }
ccb1352e 2197
9354d452
JB		 2198 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2199 info->snd_portid, info->snd_seq, 0,
d4e4fdf9 2200 OVS_VPORT_CMD_SET, GFP_KERNEL);
a9341512 2201 BUG_ON(err < 0);
ccb1352e 2202
8e4e1713 2203 ovs_unlock();
2a94fe48 2204 ovs_notify(&dp_vport_genl_family, reply, info);
8e4e1713 2205 return 0;
ccb1352e 2206
6093ae9a 2207exit_unlock_free:
8e4e1713 2208 ovs_unlock();
6093ae9a 2209 kfree_skb(reply);
ccb1352e
JG		 2210 return err;
2211}
2212
2213static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2214{
6b660c41 2215 bool update_headroom = false;
ccb1352e
JG		 2216 struct nlattr **a = info->attrs;
2217 struct sk_buff *reply;
3a927bc7 2218 struct datapath *dp;
ccb1352e 2219 struct vport *vport;
6b660c41 2220 unsigned int new_headroom;
ccb1352e
JG		 2221 int err;
2222
6093ae9a
JR		 2223 reply = ovs_vport_cmd_alloc_info();
2224 if (!reply)
2225 return -ENOMEM;
2226
8e4e1713 2227 ovs_lock();
46df7b81 2228 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
ccb1352e
JG		 2229 err = PTR_ERR(vport);
2230 if (IS_ERR(vport))
6093ae9a 2231 goto exit_unlock_free;
ccb1352e
JG		 2232
2233 if (vport->port_no == OVSP_LOCAL) {
2234 err = -EINVAL;
6093ae9a 2235 goto exit_unlock_free;
ccb1352e
JG		 2236 }
2237
9354d452
JB		 2238 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2239 info->snd_portid, info->snd_seq, 0,
d4e4fdf9 2240 OVS_VPORT_CMD_DEL, GFP_KERNEL);
6093ae9a 2241 BUG_ON(err < 0);
3a927bc7
PA		 2242
2243 /* the vport deletion may trigger dp headroom update */
2244 dp = vport->dp;
2245 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
6b660c41
TY		 2246 update_headroom = true;
2247
3a927bc7 2248 netdev_reset_rx_headroom(vport->dev);
ccb1352e 2249 ovs_dp_detach_port(vport);
3a927bc7 2250
6b660c41
TY		 2251 if (update_headroom) {
2252 new_headroom = ovs_get_max_headroom(dp);
2253
2254 if (new_headroom < dp->max_headroom)
2255 ovs_update_headroom(dp, new_headroom);
2256 }
6093ae9a 2257 ovs_unlock();
ccb1352e 2258
2a94fe48 2259 ovs_notify(&dp_vport_genl_family, reply, info);
6093ae9a 2260 return 0;
ccb1352e 2261
6093ae9a 2262exit_unlock_free:
8e4e1713 2263 ovs_unlock();
6093ae9a 2264 kfree_skb(reply);
ccb1352e
JG		 2265 return err;
2266}
2267
2268static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2269{
2270 struct nlattr **a = info->attrs;
2271 struct ovs_header *ovs_header = info->userhdr;
2272 struct sk_buff *reply;
2273 struct vport *vport;
2274 int err;
2275
6093ae9a
JR		 2276 reply = ovs_vport_cmd_alloc_info();
2277 if (!reply)
2278 return -ENOMEM;
2279
ccb1352e 2280 rcu_read_lock();
46df7b81 2281 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
ccb1352e
JG		 2282 err = PTR_ERR(vport);
2283 if (IS_ERR(vport))
6093ae9a 2284 goto exit_unlock_free;
9354d452
JB		 2285 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2286 info->snd_portid, info->snd_seq, 0,
d4e4fdf9 2287 OVS_VPORT_CMD_GET, GFP_ATOMIC);
6093ae9a 2288 BUG_ON(err < 0);
ccb1352e
JG		 2289 rcu_read_unlock();
2290
2291 return genlmsg_reply(reply, info);
2292
6093ae9a 2293exit_unlock_free:
ccb1352e 2294 rcu_read_unlock();
6093ae9a 2295 kfree_skb(reply);
ccb1352e
JG		 2296 return err;
2297}
2298
2299static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2300{
2301 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2302 struct datapath *dp;
15eac2a7
PS		 2303 int bucket = cb->args[0], skip = cb->args[1];
2304 int i, j = 0;
ccb1352e 2305
42ee19e2 2306 rcu_read_lock();
cc3a5ae6 2307 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
42ee19e2
JR		 2308 if (!dp) {
2309 rcu_read_unlock();
ccb1352e 2310 return -ENODEV;
42ee19e2 2311 }
15eac2a7 2312 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
ccb1352e 2313 struct vport *vport;
15eac2a7
PS		 2314
2315 j = 0;
b67bfe0d 2316 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
15eac2a7
PS		 2317 if (j >= skip &&
2318 ovs_vport_cmd_fill_info(vport, skb,
9354d452 2319 sock_net(skb->sk),
15e47304 2320 NETLINK_CB(cb->skb).portid,
15eac2a7
PS		 2321 cb->nlh->nlmsg_seq,
2322 NLM_F_MULTI,
d4e4fdf9
GN		 2323 OVS_VPORT_CMD_GET,
2324 GFP_ATOMIC) < 0)
15eac2a7
PS		 2325 goto out;
2326
2327 j++;
2328 }
2329 skip = 0;
ccb1352e 2330 }
15eac2a7 2331out:
ccb1352e
JG		 2332 rcu_read_unlock();
2333
15eac2a7
PS		 2334 cb->args[0] = i;
2335 cb->args[1] = j;
ccb1352e 2336
15eac2a7 2337 return skb->len;
ccb1352e
JG		 2338}
2339
0c200ef9
PS		 2340static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2341 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2342 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2343 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2344 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
ea8564c8 2345 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
0c200ef9 2346 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
9354d452
JB		 2347 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2348 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
0c200ef9
PS		 2349};
2350
48e48a70 2351static const struct genl_ops dp_vport_genl_ops[] = {
ccb1352e 2352 { .cmd = OVS_VPORT_CMD_NEW,
ef6243ac 2353 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 2354 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 2355 .doit = ovs_vport_cmd_new
2356 },
2357 { .cmd = OVS_VPORT_CMD_DEL,
ef6243ac 2358 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 2359 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 2360 .doit = ovs_vport_cmd_del
2361 },
2362 { .cmd = OVS_VPORT_CMD_GET,
ef6243ac 2363 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
ccb1352e 2364 .flags = 0, /* OK for unprivileged users. */
ccb1352e
JG		 2365 .doit = ovs_vport_cmd_get,
2366 .dumpit = ovs_vport_cmd_dump
2367 },
2368 { .cmd = OVS_VPORT_CMD_SET,
ef6243ac 2369 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
4a92602a 2370 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
ccb1352e
JG		 2371 .doit = ovs_vport_cmd_set,
2372 },
2373};
2374
56989f6d 2375struct genl_family dp_vport_genl_family __ro_after_init = {
0c200ef9
PS		 2376 .hdrsize = sizeof(struct ovs_header),
2377 .name = OVS_VPORT_FAMILY,
2378 .version = OVS_VPORT_VERSION,
2379 .maxattr = OVS_VPORT_ATTR_MAX,
3b0f31f2 2380 .policy = vport_policy,
0c200ef9
PS		 2381 .netnsok = true,
2382 .parallel_ops = true,
2383 .ops = dp_vport_genl_ops,
2384 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2385 .mcgrps = &ovs_dp_vport_multicast_group,
2386 .n_mcgrps = 1,
489111e5 2387 .module = THIS_MODULE,
ccb1352e
JG		 2388};
2389
0c200ef9
PS		 2390static struct genl_family * const dp_genl_families[] = {
2391 &dp_datapath_genl_family,
2392 &dp_vport_genl_family,
2393 &dp_flow_genl_family,
2394 &dp_packet_genl_family,
96fbc13d 2395 &dp_meter_genl_family,
11efd5cb
YHW		 2396#if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2397 &dp_ct_limit_genl_family,
2398#endif
ccb1352e
JG		 2399};
2400
2401static void dp_unregister_genl(int n_families)
2402{
2403 int i;
2404
2405 for (i = 0; i < n_families; i++)
0c200ef9 2406 genl_unregister_family(dp_genl_families[i]);
ccb1352e
JG		 2407}
2408
56989f6d 2409static int __init dp_register_genl(void)
ccb1352e 2410{
ccb1352e
JG		 2411 int err;
2412 int i;
2413
ccb1352e 2414 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
ccb1352e 2415
0c200ef9 2416 err = genl_register_family(dp_genl_families[i]);
ccb1352e
JG		 2417 if (err)
2418 goto error;
ccb1352e
JG		 2419 }
2420
2421 return 0;
2422
2423error:
0c200ef9 2424 dp_unregister_genl(i);
ccb1352e
JG		 2425 return err;
2426}
2427
46df7b81
PS		 2428static int __net_init ovs_init_net(struct net *net)
2429{
2430 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2431
2432 INIT_LIST_HEAD(&ovs_net->dps);
8e4e1713 2433 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
11efd5cb 2434 return ovs_ct_init(net);
46df7b81
PS		 2435}
2436
7b4577a9
PS		 2437static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2438 struct list_head *head)
46df7b81 2439{
8e4e1713 2440 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
7b4577a9
PS		 2441 struct datapath *dp;
2442
2443 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2444 int i;
2445
2446 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2447 struct vport *vport;
2448
2449 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
7b4577a9
PS		 2450 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2451 continue;
2452
be4ace6e 2453 if (dev_net(vport->dev) == dnet)
7b4577a9
PS		 2454 list_add(&vport->detach_list, head);
2455 }
2456 }
2457 }
2458}
2459
2460static void __net_exit ovs_exit_net(struct net *dnet)
2461{
2462 struct datapath *dp, *dp_next;
2463 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2464 struct vport *vport, *vport_next;
2465 struct net *net;
2466 LIST_HEAD(head);
46df7b81 2467
c2ac6673 2468 ovs_ct_exit(dnet);
8e4e1713 2469 ovs_lock();
46df7b81
PS		 2470 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2471 __dp_destroy(dp);
7b4577a9 2472
f0b07bb1 2473 down_read(&net_rwsem);
7b4577a9
PS		 2474 for_each_net(net)
2475 list_vports_from_net(net, dnet, &head);
f0b07bb1 2476 up_read(&net_rwsem);
7b4577a9
PS		 2477
2478 /* Detach all vports from given namespace. */
2479 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2480 list_del(&vport->detach_list);
2481 ovs_dp_detach_port(vport);
2482 }
2483
8e4e1713
PS		 2484 ovs_unlock();
2485
2486 cancel_work_sync(&ovs_net->dp_notify_work);
46df7b81
PS		 2487}
2488
2489static struct pernet_operations ovs_net_ops = {
2490 .init = ovs_init_net,
2491 .exit = ovs_exit_net,
2492 .id = &ovs_net_id,
2493 .size = sizeof(struct ovs_net),
2494};
2495
ccb1352e
JG		 2496static int __init dp_init(void)
2497{
ccb1352e
JG		 2498 int err;
2499
3523b29b 2500 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
ccb1352e
JG		 2501
2502 pr_info("Open vSwitch switching datapath\n");
2503
971427f3 2504 err = action_fifos_init();
ccb1352e
JG		 2505 if (err)
2506 goto error;
2507
971427f3
AZ		 2508 err = ovs_internal_dev_rtnl_link_register();
2509 if (err)
2510 goto error_action_fifos_exit;
2511
5b9e7e16
JP		 2512 err = ovs_flow_init();
2513 if (err)
2514 goto error_unreg_rtnl_link;
2515
ccb1352e
JG		 2516 err = ovs_vport_init();
2517 if (err)
2518 goto error_flow_exit;
2519
46df7b81 2520 err = register_pernet_device(&ovs_net_ops);
ccb1352e
JG		 2521 if (err)
2522 goto error_vport_exit;
2523
46df7b81
PS		 2524 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2525 if (err)
2526 goto error_netns_exit;
2527
62b9c8d0
TG		 2528 err = ovs_netdev_init();
2529 if (err)
2530 goto error_unreg_notifier;
2531
ccb1352e
JG		 2532 err = dp_register_genl();
2533 if (err < 0)
62b9c8d0 2534 goto error_unreg_netdev;
ccb1352e 2535
ccb1352e
JG		 2536 return 0;
2537
62b9c8d0
TG		 2538error_unreg_netdev:
2539 ovs_netdev_exit();
ccb1352e
JG		 2540error_unreg_notifier:
2541 unregister_netdevice_notifier(&ovs_dp_device_notifier);
46df7b81
PS		 2542error_netns_exit:
2543 unregister_pernet_device(&ovs_net_ops);
ccb1352e
JG		 2544error_vport_exit:
2545 ovs_vport_exit();
2546error_flow_exit:
2547 ovs_flow_exit();
5b9e7e16
JP		 2548error_unreg_rtnl_link:
2549 ovs_internal_dev_rtnl_link_unregister();
971427f3
AZ		 2550error_action_fifos_exit:
2551 action_fifos_exit();
ccb1352e
JG		 2552error:
2553 return err;
2554}
2555
2556static void dp_cleanup(void)
2557{
ccb1352e 2558 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
62b9c8d0 2559 ovs_netdev_exit();
ccb1352e 2560 unregister_netdevice_notifier(&ovs_dp_device_notifier);
46df7b81
PS		 2561 unregister_pernet_device(&ovs_net_ops);
2562 rcu_barrier();
ccb1352e
JG		 2563 ovs_vport_exit();
2564 ovs_flow_exit();
5b9e7e16 2565 ovs_internal_dev_rtnl_link_unregister();
971427f3 2566 action_fifos_exit();
ccb1352e
JG		 2567}
2568
2569module_init(dp_init);
2570module_exit(dp_cleanup);
2571
2572MODULE_DESCRIPTION("Open vSwitch switching datapath");
2573MODULE_LICENSE("GPL");
ed227099
TLSC		 2574MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2575MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2576MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2577MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
96fbc13d 2578MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
11efd5cb 2579MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
Linux 6.x block layer and io_uring tree(s)