Commit | Line | Data |
---|---|---|
ccb1352e | 1 | /* |
03f0d916 | 2 | * Copyright (c) 2007-2013 Nicira, Inc. |
ccb1352e JG |
3 | * |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of version 2 of the GNU General Public | |
6 | * License as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, but | |
9 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public License | |
14 | * along with this program; if not, write to the Free Software | |
15 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA | |
16 | * 02110-1301, USA | |
17 | */ | |
18 | ||
19 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
20 | ||
21 | #include <linux/init.h> | |
22 | #include <linux/module.h> | |
23 | #include <linux/if_arp.h> | |
24 | #include <linux/if_vlan.h> | |
25 | #include <linux/in.h> | |
26 | #include <linux/ip.h> | |
27 | #include <linux/jhash.h> | |
28 | #include <linux/delay.h> | |
29 | #include <linux/time.h> | |
30 | #include <linux/etherdevice.h> | |
31 | #include <linux/genetlink.h> | |
32 | #include <linux/kernel.h> | |
33 | #include <linux/kthread.h> | |
34 | #include <linux/mutex.h> | |
35 | #include <linux/percpu.h> | |
36 | #include <linux/rcupdate.h> | |
37 | #include <linux/tcp.h> | |
38 | #include <linux/udp.h> | |
ccb1352e JG |
39 | #include <linux/ethtool.h> |
40 | #include <linux/wait.h> | |
ccb1352e JG |
41 | #include <asm/div64.h> |
42 | #include <linux/highmem.h> | |
43 | #include <linux/netfilter_bridge.h> | |
44 | #include <linux/netfilter_ipv4.h> | |
45 | #include <linux/inetdevice.h> | |
46 | #include <linux/list.h> | |
8e4e1713 | 47 | #include <linux/lockdep.h> |
ccb1352e JG |
48 | #include <linux/openvswitch.h> |
49 | #include <linux/rculist.h> | |
50 | #include <linux/dmi.h> | |
51 | #include <linux/workqueue.h> | |
52 | #include <net/genetlink.h> | |
46df7b81 PS |
53 | #include <net/net_namespace.h> |
54 | #include <net/netns/generic.h> | |
ccb1352e JG |
55 | |
56 | #include "datapath.h" | |
57 | #include "flow.h" | |
e6445719 | 58 | #include "flow_netlink.h" |
ccb1352e | 59 | #include "vport-internal_dev.h" |
cff63a52 | 60 | #include "vport-netdev.h" |
ccb1352e | 61 | |
8e4e1713 PS |
62 | int ovs_net_id __read_mostly; |
63 | ||
68eb5503 | 64 | static void ovs_notify(struct genl_family *family, |
2a94fe48 | 65 | struct sk_buff *skb, struct genl_info *info) |
ed661185 | 66 | { |
68eb5503 | 67 | genl_notify(family, skb, genl_info_net(info), info->snd_portid, |
2a94fe48 | 68 | 0, info->nlhdr, GFP_KERNEL); |
ed661185 TG |
69 | } |
70 | ||
ccb1352e JG |
71 | /** |
72 | * DOC: Locking: | |
73 | * | |
8e4e1713 PS |
74 | * All writes e.g. Writes to device state (add/remove datapath, port, set |
75 | * operations on vports, etc.), Writes to other state (flow table | |
76 | * modifications, set miscellaneous datapath parameters, etc.) are protected | |
77 | * by ovs_lock. | |
ccb1352e JG |
78 | * |
79 | * Reads are protected by RCU. | |
80 | * | |
81 | * There are a few special cases (mostly stats) that have their own | |
82 | * synchronization but they nest under all of above and don't interact with | |
83 | * each other. | |
8e4e1713 PS |
84 | * |
85 | * The RTNL lock nests inside ovs_mutex. | |
ccb1352e JG |
86 | */ |
87 | ||
8e4e1713 PS |
88 | static DEFINE_MUTEX(ovs_mutex); |
89 | ||
90 | void ovs_lock(void) | |
91 | { | |
92 | mutex_lock(&ovs_mutex); | |
93 | } | |
94 | ||
95 | void ovs_unlock(void) | |
96 | { | |
97 | mutex_unlock(&ovs_mutex); | |
98 | } | |
99 | ||
100 | #ifdef CONFIG_LOCKDEP | |
101 | int lockdep_ovsl_is_held(void) | |
102 | { | |
103 | if (debug_locks) | |
104 | return lockdep_is_held(&ovs_mutex); | |
105 | else | |
106 | return 1; | |
107 | } | |
108 | #endif | |
109 | ||
ccb1352e | 110 | static struct vport *new_vport(const struct vport_parms *); |
8055a89c | 111 | static int queue_gso_packets(struct datapath *dp, struct sk_buff *, |
ccb1352e | 112 | const struct dp_upcall_info *); |
8055a89c | 113 | static int queue_userspace_packet(struct datapath *dp, struct sk_buff *, |
ccb1352e JG |
114 | const struct dp_upcall_info *); |
115 | ||
8e4e1713 | 116 | /* Must be called with rcu_read_lock or ovs_mutex. */ |
46df7b81 | 117 | static struct datapath *get_dp(struct net *net, int dp_ifindex) |
ccb1352e JG |
118 | { |
119 | struct datapath *dp = NULL; | |
120 | struct net_device *dev; | |
121 | ||
122 | rcu_read_lock(); | |
46df7b81 | 123 | dev = dev_get_by_index_rcu(net, dp_ifindex); |
ccb1352e JG |
124 | if (dev) { |
125 | struct vport *vport = ovs_internal_dev_get_vport(dev); | |
126 | if (vport) | |
127 | dp = vport->dp; | |
128 | } | |
129 | rcu_read_unlock(); | |
130 | ||
131 | return dp; | |
132 | } | |
133 | ||
8e4e1713 | 134 | /* Must be called with rcu_read_lock or ovs_mutex. */ |
443cd88c | 135 | static const char *ovs_dp_name(const struct datapath *dp) |
ccb1352e | 136 | { |
8e4e1713 | 137 | struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL); |
ccb1352e JG |
138 | return vport->ops->get_name(vport); |
139 | } | |
140 | ||
141 | static int get_dpifindex(struct datapath *dp) | |
142 | { | |
143 | struct vport *local; | |
144 | int ifindex; | |
145 | ||
146 | rcu_read_lock(); | |
147 | ||
15eac2a7 | 148 | local = ovs_vport_rcu(dp, OVSP_LOCAL); |
ccb1352e | 149 | if (local) |
cff63a52 | 150 | ifindex = netdev_vport_priv(local)->dev->ifindex; |
ccb1352e JG |
151 | else |
152 | ifindex = 0; | |
153 | ||
154 | rcu_read_unlock(); | |
155 | ||
156 | return ifindex; | |
157 | } | |
158 | ||
159 | static void destroy_dp_rcu(struct rcu_head *rcu) | |
160 | { | |
161 | struct datapath *dp = container_of(rcu, struct datapath, rcu); | |
162 | ||
618ed0c8 | 163 | ovs_flow_tbl_destroy(&dp->table); |
ccb1352e | 164 | free_percpu(dp->stats_percpu); |
46df7b81 | 165 | release_net(ovs_dp_get_net(dp)); |
15eac2a7 | 166 | kfree(dp->ports); |
ccb1352e JG |
167 | kfree(dp); |
168 | } | |
169 | ||
15eac2a7 PS |
170 | static struct hlist_head *vport_hash_bucket(const struct datapath *dp, |
171 | u16 port_no) | |
172 | { | |
173 | return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)]; | |
174 | } | |
175 | ||
176 | struct 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 |
190 | static 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 |
204 | void 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. */ | |
216 | void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb) | |
217 | { | |
218 | struct datapath *dp = p->dp; | |
219 | struct sw_flow *flow; | |
220 | struct dp_stats_percpu *stats; | |
221 | struct sw_flow_key key; | |
222 | u64 *stats_counter; | |
1bd7116f | 223 | u32 n_mask_hit; |
ccb1352e | 224 | int error; |
ccb1352e | 225 | |
404f2f10 | 226 | stats = this_cpu_ptr(dp->stats_percpu); |
ccb1352e JG |
227 | |
228 | /* Extract flow from 'skb' into 'key'. */ | |
03f0d916 | 229 | error = ovs_flow_extract(skb, p->port_no, &key); |
ccb1352e JG |
230 | if (unlikely(error)) { |
231 | kfree_skb(skb); | |
232 | return; | |
233 | } | |
234 | ||
235 | /* Look up flow. */ | |
5bb50632 | 236 | flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit); |
ccb1352e JG |
237 | if (unlikely(!flow)) { |
238 | struct dp_upcall_info upcall; | |
239 | ||
240 | upcall.cmd = OVS_PACKET_CMD_MISS; | |
241 | upcall.key = &key; | |
242 | upcall.userdata = NULL; | |
15e47304 | 243 | upcall.portid = p->upcall_portid; |
ccb1352e JG |
244 | ovs_dp_upcall(dp, skb, &upcall); |
245 | consume_skb(skb); | |
246 | stats_counter = &stats->n_missed; | |
247 | goto out; | |
248 | } | |
249 | ||
250 | OVS_CB(skb)->flow = flow; | |
03f0d916 | 251 | OVS_CB(skb)->pkt_key = &key; |
ccb1352e | 252 | |
e298e505 | 253 | ovs_flow_stats_update(OVS_CB(skb)->flow, skb); |
ccb1352e | 254 | ovs_execute_actions(dp, skb); |
e298e505 | 255 | stats_counter = &stats->n_hit; |
ccb1352e JG |
256 | |
257 | out: | |
258 | /* Update datapath statistics. */ | |
259 | u64_stats_update_begin(&stats->sync); | |
260 | (*stats_counter)++; | |
1bd7116f | 261 | stats->n_mask_hit += n_mask_hit; |
ccb1352e JG |
262 | u64_stats_update_end(&stats->sync); |
263 | } | |
264 | ||
265 | static struct genl_family dp_packet_genl_family = { | |
266 | .id = GENL_ID_GENERATE, | |
267 | .hdrsize = sizeof(struct ovs_header), | |
268 | .name = OVS_PACKET_FAMILY, | |
269 | .version = OVS_PACKET_VERSION, | |
46df7b81 | 270 | .maxattr = OVS_PACKET_ATTR_MAX, |
3a4e0d6a PS |
271 | .netnsok = true, |
272 | .parallel_ops = true, | |
ccb1352e JG |
273 | }; |
274 | ||
275 | int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb, | |
46df7b81 | 276 | const struct dp_upcall_info *upcall_info) |
ccb1352e JG |
277 | { |
278 | struct dp_stats_percpu *stats; | |
ccb1352e JG |
279 | int err; |
280 | ||
15e47304 | 281 | if (upcall_info->portid == 0) { |
ccb1352e JG |
282 | err = -ENOTCONN; |
283 | goto err; | |
284 | } | |
285 | ||
ccb1352e | 286 | if (!skb_is_gso(skb)) |
8055a89c | 287 | err = queue_userspace_packet(dp, skb, upcall_info); |
ccb1352e | 288 | else |
8055a89c | 289 | err = queue_gso_packets(dp, skb, upcall_info); |
ccb1352e JG |
290 | if (err) |
291 | goto err; | |
292 | ||
293 | return 0; | |
294 | ||
295 | err: | |
404f2f10 | 296 | stats = this_cpu_ptr(dp->stats_percpu); |
ccb1352e JG |
297 | |
298 | u64_stats_update_begin(&stats->sync); | |
299 | stats->n_lost++; | |
300 | u64_stats_update_end(&stats->sync); | |
301 | ||
302 | return err; | |
303 | } | |
304 | ||
8055a89c | 305 | static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb, |
ccb1352e JG |
306 | const struct dp_upcall_info *upcall_info) |
307 | { | |
a1b5d0dd | 308 | unsigned short gso_type = skb_shinfo(skb)->gso_type; |
ccb1352e JG |
309 | struct dp_upcall_info later_info; |
310 | struct sw_flow_key later_key; | |
311 | struct sk_buff *segs, *nskb; | |
312 | int err; | |
313 | ||
09c5e605 | 314 | segs = __skb_gso_segment(skb, NETIF_F_SG, false); |
92e5dfc3 PS |
315 | if (IS_ERR(segs)) |
316 | return PTR_ERR(segs); | |
ccb1352e JG |
317 | |
318 | /* Queue all of the segments. */ | |
319 | skb = segs; | |
320 | do { | |
8055a89c | 321 | err = queue_userspace_packet(dp, skb, upcall_info); |
ccb1352e JG |
322 | if (err) |
323 | break; | |
324 | ||
a1b5d0dd | 325 | if (skb == segs && gso_type & SKB_GSO_UDP) { |
ccb1352e JG |
326 | /* The initial flow key extracted by ovs_flow_extract() |
327 | * in this case is for a first fragment, so we need to | |
328 | * properly mark later fragments. | |
329 | */ | |
330 | later_key = *upcall_info->key; | |
331 | later_key.ip.frag = OVS_FRAG_TYPE_LATER; | |
332 | ||
333 | later_info = *upcall_info; | |
334 | later_info.key = &later_key; | |
335 | upcall_info = &later_info; | |
336 | } | |
337 | } while ((skb = skb->next)); | |
338 | ||
339 | /* Free all of the segments. */ | |
340 | skb = segs; | |
341 | do { | |
342 | nskb = skb->next; | |
343 | if (err) | |
344 | kfree_skb(skb); | |
345 | else | |
346 | consume_skb(skb); | |
347 | } while ((skb = nskb)); | |
348 | return err; | |
349 | } | |
350 | ||
c3ff8cfe TG |
351 | static size_t key_attr_size(void) |
352 | { | |
353 | return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */ | |
7d5437c7 PS |
354 | + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */ |
355 | + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */ | |
356 | + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */ | |
357 | + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */ | |
358 | + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */ | |
359 | + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */ | |
360 | + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */ | |
361 | + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */ | |
c3ff8cfe TG |
362 | + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */ |
363 | + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */ | |
364 | + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */ | |
365 | + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */ | |
366 | + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */ | |
367 | + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */ | |
368 | + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */ | |
369 | + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */ | |
370 | + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */ | |
371 | + nla_total_size(28); /* OVS_KEY_ATTR_ND */ | |
372 | } | |
373 | ||
bda56f14 TG |
374 | static size_t upcall_msg_size(const struct nlattr *userdata, |
375 | unsigned int hdrlen) | |
c3ff8cfe TG |
376 | { |
377 | size_t size = NLMSG_ALIGN(sizeof(struct ovs_header)) | |
bda56f14 | 378 | + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */ |
c3ff8cfe TG |
379 | + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */ |
380 | ||
381 | /* OVS_PACKET_ATTR_USERDATA */ | |
382 | if (userdata) | |
383 | size += NLA_ALIGN(userdata->nla_len); | |
384 | ||
385 | return size; | |
386 | } | |
387 | ||
8055a89c | 388 | static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb, |
ccb1352e JG |
389 | const struct dp_upcall_info *upcall_info) |
390 | { | |
391 | struct ovs_header *upcall; | |
392 | struct sk_buff *nskb = NULL; | |
393 | struct sk_buff *user_skb; /* to be queued to userspace */ | |
394 | struct nlattr *nla; | |
795449d8 | 395 | struct genl_info info = { |
8055a89c | 396 | .dst_sk = ovs_dp_get_net(dp)->genl_sock, |
795449d8 TG |
397 | .snd_portid = upcall_info->portid, |
398 | }; | |
399 | size_t len; | |
bda56f14 | 400 | unsigned int hlen; |
8055a89c TG |
401 | int err, dp_ifindex; |
402 | ||
403 | dp_ifindex = get_dpifindex(dp); | |
404 | if (!dp_ifindex) | |
405 | return -ENODEV; | |
ccb1352e JG |
406 | |
407 | if (vlan_tx_tag_present(skb)) { | |
408 | nskb = skb_clone(skb, GFP_ATOMIC); | |
409 | if (!nskb) | |
410 | return -ENOMEM; | |
411 | ||
86a9bad3 | 412 | nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb)); |
8aa51d64 | 413 | if (!nskb) |
ccb1352e JG |
414 | return -ENOMEM; |
415 | ||
416 | nskb->vlan_tci = 0; | |
417 | skb = nskb; | |
418 | } | |
419 | ||
420 | if (nla_attr_size(skb->len) > USHRT_MAX) { | |
421 | err = -EFBIG; | |
422 | goto out; | |
423 | } | |
424 | ||
bda56f14 TG |
425 | /* Complete checksum if needed */ |
426 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
427 | (err = skb_checksum_help(skb))) | |
428 | goto out; | |
429 | ||
430 | /* Older versions of OVS user space enforce alignment of the last | |
431 | * Netlink attribute to NLA_ALIGNTO which would require extensive | |
432 | * padding logic. Only perform zerocopy if padding is not required. | |
433 | */ | |
434 | if (dp->user_features & OVS_DP_F_UNALIGNED) | |
435 | hlen = skb_zerocopy_headlen(skb); | |
436 | else | |
437 | hlen = skb->len; | |
438 | ||
439 | len = upcall_msg_size(upcall_info->userdata, hlen); | |
795449d8 | 440 | user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC); |
ccb1352e JG |
441 | if (!user_skb) { |
442 | err = -ENOMEM; | |
443 | goto out; | |
444 | } | |
445 | ||
446 | upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, | |
447 | 0, upcall_info->cmd); | |
448 | upcall->dp_ifindex = dp_ifindex; | |
449 | ||
450 | nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY); | |
e6445719 | 451 | ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb); |
ccb1352e JG |
452 | nla_nest_end(user_skb, nla); |
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 | |
bda56f14 TG |
459 | /* Only reserve room for attribute header, packet data is added |
460 | * in skb_zerocopy() */ | |
461 | if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) { | |
462 | err = -ENOBUFS; | |
463 | goto out; | |
464 | } | |
465 | nla->nla_len = nla_attr_size(skb->len); | |
ccb1352e | 466 | |
bda56f14 | 467 | skb_zerocopy(user_skb, skb, skb->len, hlen); |
ccb1352e | 468 | |
aea0bb4f TG |
469 | /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */ |
470 | if (!(dp->user_features & OVS_DP_F_UNALIGNED)) { | |
471 | size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len; | |
472 | ||
473 | if (plen > 0) | |
474 | memset(skb_put(user_skb, plen), 0, plen); | |
475 | } | |
476 | ||
bda56f14 | 477 | ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len; |
ccb1352e | 478 | |
bda56f14 | 479 | err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid); |
ccb1352e JG |
480 | out: |
481 | kfree_skb(nskb); | |
482 | return err; | |
483 | } | |
484 | ||
ccb1352e JG |
485 | static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info) |
486 | { | |
487 | struct ovs_header *ovs_header = info->userhdr; | |
488 | struct nlattr **a = info->attrs; | |
489 | struct sw_flow_actions *acts; | |
490 | struct sk_buff *packet; | |
491 | struct sw_flow *flow; | |
492 | struct datapath *dp; | |
493 | struct ethhdr *eth; | |
494 | int len; | |
495 | int err; | |
ccb1352e JG |
496 | |
497 | err = -EINVAL; | |
498 | if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] || | |
dded45fc | 499 | !a[OVS_PACKET_ATTR_ACTIONS]) |
ccb1352e JG |
500 | goto err; |
501 | ||
502 | len = nla_len(a[OVS_PACKET_ATTR_PACKET]); | |
503 | packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL); | |
504 | err = -ENOMEM; | |
505 | if (!packet) | |
506 | goto err; | |
507 | skb_reserve(packet, NET_IP_ALIGN); | |
508 | ||
32686a9d | 509 | nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len); |
ccb1352e JG |
510 | |
511 | skb_reset_mac_header(packet); | |
512 | eth = eth_hdr(packet); | |
513 | ||
514 | /* Normally, setting the skb 'protocol' field would be handled by a | |
515 | * call to eth_type_trans(), but it assumes there's a sending | |
516 | * device, which we may not have. */ | |
e5c5d22e | 517 | if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN) |
ccb1352e JG |
518 | packet->protocol = eth->h_proto; |
519 | else | |
520 | packet->protocol = htons(ETH_P_802_2); | |
521 | ||
522 | /* Build an sw_flow for sending this packet. */ | |
e298e505 | 523 | flow = ovs_flow_alloc(false); |
ccb1352e JG |
524 | err = PTR_ERR(flow); |
525 | if (IS_ERR(flow)) | |
526 | goto err_kfree_skb; | |
527 | ||
03f0d916 | 528 | err = ovs_flow_extract(packet, -1, &flow->key); |
ccb1352e JG |
529 | if (err) |
530 | goto err_flow_free; | |
531 | ||
e6445719 | 532 | err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]); |
ccb1352e JG |
533 | if (err) |
534 | goto err_flow_free; | |
e6445719 | 535 | acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS])); |
ccb1352e JG |
536 | err = PTR_ERR(acts); |
537 | if (IS_ERR(acts)) | |
538 | goto err_flow_free; | |
74f84a57 | 539 | |
e6445719 PS |
540 | err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], |
541 | &flow->key, 0, &acts); | |
ccb1352e | 542 | rcu_assign_pointer(flow->sf_acts, acts); |
74f84a57 PS |
543 | if (err) |
544 | goto err_flow_free; | |
ccb1352e JG |
545 | |
546 | OVS_CB(packet)->flow = flow; | |
03f0d916 | 547 | OVS_CB(packet)->pkt_key = &flow->key; |
ccb1352e | 548 | packet->priority = flow->key.phy.priority; |
39c7caeb | 549 | packet->mark = flow->key.phy.skb_mark; |
ccb1352e JG |
550 | |
551 | rcu_read_lock(); | |
46df7b81 | 552 | dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); |
ccb1352e JG |
553 | err = -ENODEV; |
554 | if (!dp) | |
555 | goto err_unlock; | |
556 | ||
557 | local_bh_disable(); | |
558 | err = ovs_execute_actions(dp, packet); | |
559 | local_bh_enable(); | |
560 | rcu_read_unlock(); | |
561 | ||
03f0d916 | 562 | ovs_flow_free(flow, false); |
ccb1352e JG |
563 | return err; |
564 | ||
565 | err_unlock: | |
566 | rcu_read_unlock(); | |
567 | err_flow_free: | |
03f0d916 | 568 | ovs_flow_free(flow, false); |
ccb1352e JG |
569 | err_kfree_skb: |
570 | kfree_skb(packet); | |
571 | err: | |
572 | return err; | |
573 | } | |
574 | ||
575 | static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = { | |
dded45fc | 576 | [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN }, |
ccb1352e JG |
577 | [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED }, |
578 | [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED }, | |
579 | }; | |
580 | ||
4534de83 | 581 | static const struct genl_ops dp_packet_genl_ops[] = { |
ccb1352e JG |
582 | { .cmd = OVS_PACKET_CMD_EXECUTE, |
583 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
584 | .policy = packet_policy, | |
585 | .doit = ovs_packet_cmd_execute | |
586 | } | |
587 | }; | |
588 | ||
1bd7116f AZ |
589 | static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats, |
590 | struct ovs_dp_megaflow_stats *mega_stats) | |
ccb1352e JG |
591 | { |
592 | int i; | |
ccb1352e | 593 | |
1bd7116f AZ |
594 | memset(mega_stats, 0, sizeof(*mega_stats)); |
595 | ||
b637e498 | 596 | stats->n_flows = ovs_flow_tbl_count(&dp->table); |
1bd7116f | 597 | mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table); |
ccb1352e JG |
598 | |
599 | stats->n_hit = stats->n_missed = stats->n_lost = 0; | |
1bd7116f | 600 | |
ccb1352e JG |
601 | for_each_possible_cpu(i) { |
602 | const struct dp_stats_percpu *percpu_stats; | |
603 | struct dp_stats_percpu local_stats; | |
604 | unsigned int start; | |
605 | ||
606 | percpu_stats = per_cpu_ptr(dp->stats_percpu, i); | |
607 | ||
608 | do { | |
609 | start = u64_stats_fetch_begin_bh(&percpu_stats->sync); | |
610 | local_stats = *percpu_stats; | |
611 | } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start)); | |
612 | ||
613 | stats->n_hit += local_stats.n_hit; | |
614 | stats->n_missed += local_stats.n_missed; | |
615 | stats->n_lost += local_stats.n_lost; | |
1bd7116f | 616 | mega_stats->n_mask_hit += local_stats.n_mask_hit; |
ccb1352e JG |
617 | } |
618 | } | |
619 | ||
620 | static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = { | |
621 | [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED }, | |
622 | [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED }, | |
623 | [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG }, | |
624 | }; | |
625 | ||
626 | static struct genl_family dp_flow_genl_family = { | |
627 | .id = GENL_ID_GENERATE, | |
628 | .hdrsize = sizeof(struct ovs_header), | |
629 | .name = OVS_FLOW_FAMILY, | |
630 | .version = OVS_FLOW_VERSION, | |
46df7b81 | 631 | .maxattr = OVS_FLOW_ATTR_MAX, |
3a4e0d6a PS |
632 | .netnsok = true, |
633 | .parallel_ops = true, | |
ccb1352e JG |
634 | }; |
635 | ||
636 | static struct genl_multicast_group ovs_dp_flow_multicast_group = { | |
637 | .name = OVS_FLOW_MCGROUP | |
638 | }; | |
639 | ||
c3ff8cfe TG |
640 | static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts) |
641 | { | |
642 | return NLMSG_ALIGN(sizeof(struct ovs_header)) | |
643 | + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */ | |
03f0d916 | 644 | + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */ |
c3ff8cfe TG |
645 | + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */ |
646 | + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */ | |
647 | + nla_total_size(8) /* OVS_FLOW_ATTR_USED */ | |
648 | + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */ | |
649 | } | |
650 | ||
8e4e1713 | 651 | /* Called with ovs_mutex. */ |
ccb1352e | 652 | static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp, |
15e47304 | 653 | struct sk_buff *skb, u32 portid, |
ccb1352e JG |
654 | u32 seq, u32 flags, u8 cmd) |
655 | { | |
656 | const int skb_orig_len = skb->len; | |
74f84a57 | 657 | struct nlattr *start; |
ccb1352e | 658 | struct ovs_flow_stats stats; |
e298e505 PS |
659 | __be16 tcp_flags; |
660 | unsigned long used; | |
ccb1352e JG |
661 | struct ovs_header *ovs_header; |
662 | struct nlattr *nla; | |
ccb1352e JG |
663 | int err; |
664 | ||
15e47304 | 665 | ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd); |
ccb1352e JG |
666 | if (!ovs_header) |
667 | return -EMSGSIZE; | |
668 | ||
669 | ovs_header->dp_ifindex = get_dpifindex(dp); | |
670 | ||
03f0d916 | 671 | /* Fill flow key. */ |
ccb1352e JG |
672 | nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY); |
673 | if (!nla) | |
674 | goto nla_put_failure; | |
03f0d916 | 675 | |
e6445719 | 676 | err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb); |
03f0d916 AZ |
677 | if (err) |
678 | goto error; | |
679 | nla_nest_end(skb, nla); | |
680 | ||
681 | nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK); | |
682 | if (!nla) | |
683 | goto nla_put_failure; | |
684 | ||
e6445719 | 685 | err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb); |
ccb1352e JG |
686 | if (err) |
687 | goto error; | |
03f0d916 | 688 | |
ccb1352e JG |
689 | nla_nest_end(skb, nla); |
690 | ||
e298e505 | 691 | ovs_flow_stats_get(flow, &stats, &used, &tcp_flags); |
028d6a67 DM |
692 | if (used && |
693 | nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used))) | |
694 | goto nla_put_failure; | |
ccb1352e | 695 | |
028d6a67 | 696 | if (stats.n_packets && |
e298e505 | 697 | nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats)) |
028d6a67 | 698 | goto nla_put_failure; |
ccb1352e | 699 | |
e298e505 PS |
700 | if ((u8)ntohs(tcp_flags) && |
701 | nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags))) | |
028d6a67 | 702 | goto nla_put_failure; |
ccb1352e JG |
703 | |
704 | /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if | |
705 | * this is the first flow to be dumped into 'skb'. This is unusual for | |
706 | * Netlink but individual action lists can be longer than | |
707 | * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this. | |
708 | * The userspace caller can always fetch the actions separately if it | |
709 | * really wants them. (Most userspace callers in fact don't care.) | |
710 | * | |
711 | * This can only fail for dump operations because the skb is always | |
712 | * properly sized for single flows. | |
713 | */ | |
74f84a57 PS |
714 | start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS); |
715 | if (start) { | |
d57170b1 PS |
716 | const struct sw_flow_actions *sf_acts; |
717 | ||
663efa36 | 718 | sf_acts = rcu_dereference_ovsl(flow->sf_acts); |
d57170b1 | 719 | |
e6445719 PS |
720 | err = ovs_nla_put_actions(sf_acts->actions, |
721 | sf_acts->actions_len, skb); | |
74f84a57 PS |
722 | if (!err) |
723 | nla_nest_end(skb, start); | |
724 | else { | |
725 | if (skb_orig_len) | |
726 | goto error; | |
727 | ||
728 | nla_nest_cancel(skb, start); | |
729 | } | |
730 | } else if (skb_orig_len) | |
731 | goto nla_put_failure; | |
ccb1352e JG |
732 | |
733 | return genlmsg_end(skb, ovs_header); | |
734 | ||
735 | nla_put_failure: | |
736 | err = -EMSGSIZE; | |
737 | error: | |
738 | genlmsg_cancel(skb, ovs_header); | |
739 | return err; | |
740 | } | |
741 | ||
795449d8 TG |
742 | static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow, |
743 | struct genl_info *info) | |
ccb1352e | 744 | { |
795449d8 | 745 | size_t len; |
ccb1352e | 746 | |
795449d8 | 747 | len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts)); |
ccb1352e | 748 | |
795449d8 | 749 | return genlmsg_new_unicast(len, info, GFP_KERNEL); |
ccb1352e JG |
750 | } |
751 | ||
752 | static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, | |
753 | struct datapath *dp, | |
795449d8 TG |
754 | struct genl_info *info, |
755 | u8 cmd) | |
ccb1352e JG |
756 | { |
757 | struct sk_buff *skb; | |
758 | int retval; | |
759 | ||
795449d8 | 760 | skb = ovs_flow_cmd_alloc_info(flow, info); |
ccb1352e JG |
761 | if (!skb) |
762 | return ERR_PTR(-ENOMEM); | |
763 | ||
795449d8 TG |
764 | retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid, |
765 | info->snd_seq, 0, cmd); | |
ccb1352e JG |
766 | BUG_ON(retval < 0); |
767 | return skb; | |
768 | } | |
769 | ||
770 | static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info) | |
771 | { | |
772 | struct nlattr **a = info->attrs; | |
773 | struct ovs_header *ovs_header = info->userhdr; | |
03f0d916 AZ |
774 | struct sw_flow_key key, masked_key; |
775 | struct sw_flow *flow = NULL; | |
776 | struct sw_flow_mask mask; | |
ccb1352e JG |
777 | struct sk_buff *reply; |
778 | struct datapath *dp; | |
74f84a57 | 779 | struct sw_flow_actions *acts = NULL; |
03f0d916 | 780 | struct sw_flow_match match; |
e298e505 | 781 | bool exact_5tuple; |
ccb1352e | 782 | int error; |
ccb1352e JG |
783 | |
784 | /* Extract key. */ | |
785 | error = -EINVAL; | |
786 | if (!a[OVS_FLOW_ATTR_KEY]) | |
787 | goto error; | |
03f0d916 AZ |
788 | |
789 | ovs_match_init(&match, &key, &mask); | |
e298e505 | 790 | error = ovs_nla_get_match(&match, &exact_5tuple, |
e6445719 | 791 | a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]); |
ccb1352e JG |
792 | if (error) |
793 | goto error; | |
794 | ||
795 | /* Validate actions. */ | |
796 | if (a[OVS_FLOW_ATTR_ACTIONS]) { | |
e6445719 | 797 | acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS])); |
74f84a57 PS |
798 | error = PTR_ERR(acts); |
799 | if (IS_ERR(acts)) | |
ccb1352e | 800 | goto error; |
74f84a57 | 801 | |
e6445719 PS |
802 | ovs_flow_mask_key(&masked_key, &key, &mask); |
803 | error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], | |
804 | &masked_key, 0, &acts); | |
03f0d916 AZ |
805 | if (error) { |
806 | OVS_NLERR("Flow actions may not be safe on all matching packets.\n"); | |
74f84a57 | 807 | goto err_kfree; |
03f0d916 | 808 | } |
ccb1352e JG |
809 | } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) { |
810 | error = -EINVAL; | |
811 | goto error; | |
812 | } | |
813 | ||
8e4e1713 | 814 | ovs_lock(); |
46df7b81 | 815 | dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); |
ccb1352e JG |
816 | error = -ENODEV; |
817 | if (!dp) | |
8e4e1713 | 818 | goto err_unlock_ovs; |
ccb1352e | 819 | |
03f0d916 | 820 | /* Check if this is a duplicate flow */ |
5bb50632 | 821 | flow = ovs_flow_tbl_lookup(&dp->table, &key); |
ccb1352e | 822 | if (!flow) { |
ccb1352e JG |
823 | /* Bail out if we're not allowed to create a new flow. */ |
824 | error = -ENOENT; | |
825 | if (info->genlhdr->cmd == OVS_FLOW_CMD_SET) | |
8e4e1713 | 826 | goto err_unlock_ovs; |
ccb1352e | 827 | |
ccb1352e | 828 | /* Allocate flow. */ |
e298e505 | 829 | flow = ovs_flow_alloc(!exact_5tuple); |
ccb1352e JG |
830 | if (IS_ERR(flow)) { |
831 | error = PTR_ERR(flow); | |
8e4e1713 | 832 | goto err_unlock_ovs; |
ccb1352e | 833 | } |
ccb1352e | 834 | |
03f0d916 AZ |
835 | flow->key = masked_key; |
836 | flow->unmasked_key = key; | |
ccb1352e JG |
837 | rcu_assign_pointer(flow->sf_acts, acts); |
838 | ||
839 | /* Put flow in bucket. */ | |
618ed0c8 PS |
840 | error = ovs_flow_tbl_insert(&dp->table, flow, &mask); |
841 | if (error) { | |
842 | acts = NULL; | |
843 | goto err_flow_free; | |
844 | } | |
ccb1352e | 845 | |
795449d8 | 846 | reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW); |
ccb1352e JG |
847 | } else { |
848 | /* We found a matching flow. */ | |
849 | struct sw_flow_actions *old_acts; | |
ccb1352e JG |
850 | |
851 | /* Bail out if we're not allowed to modify an existing flow. | |
852 | * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL | |
853 | * because Generic Netlink treats the latter as a dump | |
854 | * request. We also accept NLM_F_EXCL in case that bug ever | |
855 | * gets fixed. | |
856 | */ | |
857 | error = -EEXIST; | |
858 | if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW && | |
859 | info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL)) | |
8e4e1713 | 860 | goto err_unlock_ovs; |
ccb1352e | 861 | |
03f0d916 AZ |
862 | /* The unmasked key has to be the same for flow updates. */ |
863 | error = -EINVAL; | |
e6445719 | 864 | if (!ovs_flow_cmp_unmasked_key(flow, &match)) { |
03f0d916 AZ |
865 | OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n"); |
866 | goto err_unlock_ovs; | |
867 | } | |
868 | ||
ccb1352e | 869 | /* Update actions. */ |
8e4e1713 | 870 | old_acts = ovsl_dereference(flow->sf_acts); |
74f84a57 | 871 | rcu_assign_pointer(flow->sf_acts, acts); |
e6445719 | 872 | ovs_nla_free_flow_actions(old_acts); |
ccb1352e | 873 | |
795449d8 | 874 | reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW); |
ccb1352e JG |
875 | |
876 | /* Clear stats. */ | |
e298e505 PS |
877 | if (a[OVS_FLOW_ATTR_CLEAR]) |
878 | ovs_flow_stats_clear(flow); | |
ccb1352e | 879 | } |
8e4e1713 | 880 | ovs_unlock(); |
ccb1352e JG |
881 | |
882 | if (!IS_ERR(reply)) | |
2a94fe48 | 883 | ovs_notify(&dp_flow_genl_family, reply, info); |
ccb1352e | 884 | else |
68eb5503 | 885 | genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0, |
2a94fe48 | 886 | 0, PTR_ERR(reply)); |
ccb1352e JG |
887 | return 0; |
888 | ||
03f0d916 AZ |
889 | err_flow_free: |
890 | ovs_flow_free(flow, false); | |
8e4e1713 PS |
891 | err_unlock_ovs: |
892 | ovs_unlock(); | |
74f84a57 PS |
893 | err_kfree: |
894 | kfree(acts); | |
ccb1352e JG |
895 | error: |
896 | return error; | |
897 | } | |
898 | ||
899 | static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info) | |
900 | { | |
901 | struct nlattr **a = info->attrs; | |
902 | struct ovs_header *ovs_header = info->userhdr; | |
903 | struct sw_flow_key key; | |
904 | struct sk_buff *reply; | |
905 | struct sw_flow *flow; | |
906 | struct datapath *dp; | |
03f0d916 | 907 | struct sw_flow_match match; |
ccb1352e | 908 | int err; |
ccb1352e | 909 | |
03f0d916 AZ |
910 | if (!a[OVS_FLOW_ATTR_KEY]) { |
911 | OVS_NLERR("Flow get message rejected, Key attribute missing.\n"); | |
ccb1352e | 912 | return -EINVAL; |
03f0d916 AZ |
913 | } |
914 | ||
915 | ovs_match_init(&match, &key, NULL); | |
e298e505 | 916 | err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL); |
ccb1352e JG |
917 | if (err) |
918 | return err; | |
919 | ||
8e4e1713 | 920 | ovs_lock(); |
46df7b81 | 921 | dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); |
8e4e1713 PS |
922 | if (!dp) { |
923 | err = -ENODEV; | |
924 | goto unlock; | |
925 | } | |
ccb1352e | 926 | |
5bb50632 | 927 | flow = ovs_flow_tbl_lookup(&dp->table, &key); |
e6445719 | 928 | if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) { |
8e4e1713 PS |
929 | err = -ENOENT; |
930 | goto unlock; | |
931 | } | |
ccb1352e | 932 | |
795449d8 | 933 | reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW); |
8e4e1713 PS |
934 | if (IS_ERR(reply)) { |
935 | err = PTR_ERR(reply); | |
936 | goto unlock; | |
937 | } | |
ccb1352e | 938 | |
8e4e1713 | 939 | ovs_unlock(); |
ccb1352e | 940 | return genlmsg_reply(reply, info); |
8e4e1713 PS |
941 | unlock: |
942 | ovs_unlock(); | |
943 | return err; | |
ccb1352e JG |
944 | } |
945 | ||
946 | static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info) | |
947 | { | |
948 | struct nlattr **a = info->attrs; | |
949 | struct ovs_header *ovs_header = info->userhdr; | |
950 | struct sw_flow_key key; | |
951 | struct sk_buff *reply; | |
952 | struct sw_flow *flow; | |
953 | struct datapath *dp; | |
03f0d916 | 954 | struct sw_flow_match match; |
ccb1352e | 955 | int err; |
ccb1352e | 956 | |
8e4e1713 | 957 | ovs_lock(); |
46df7b81 | 958 | dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); |
8e4e1713 PS |
959 | if (!dp) { |
960 | err = -ENODEV; | |
961 | goto unlock; | |
962 | } | |
46df7b81 | 963 | |
8e4e1713 | 964 | if (!a[OVS_FLOW_ATTR_KEY]) { |
b637e498 | 965 | err = ovs_flow_tbl_flush(&dp->table); |
8e4e1713 PS |
966 | goto unlock; |
967 | } | |
03f0d916 AZ |
968 | |
969 | ovs_match_init(&match, &key, NULL); | |
e298e505 | 970 | err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL); |
ccb1352e | 971 | if (err) |
8e4e1713 | 972 | goto unlock; |
ccb1352e | 973 | |
5bb50632 | 974 | flow = ovs_flow_tbl_lookup(&dp->table, &key); |
e6445719 | 975 | if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) { |
8e4e1713 PS |
976 | err = -ENOENT; |
977 | goto unlock; | |
978 | } | |
ccb1352e | 979 | |
795449d8 | 980 | reply = ovs_flow_cmd_alloc_info(flow, info); |
8e4e1713 PS |
981 | if (!reply) { |
982 | err = -ENOMEM; | |
983 | goto unlock; | |
984 | } | |
ccb1352e | 985 | |
b637e498 | 986 | ovs_flow_tbl_remove(&dp->table, flow); |
ccb1352e | 987 | |
15e47304 | 988 | err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid, |
ccb1352e JG |
989 | info->snd_seq, 0, OVS_FLOW_CMD_DEL); |
990 | BUG_ON(err < 0); | |
991 | ||
03f0d916 | 992 | ovs_flow_free(flow, true); |
8e4e1713 | 993 | ovs_unlock(); |
ccb1352e | 994 | |
2a94fe48 | 995 | ovs_notify(&dp_flow_genl_family, reply, info); |
ccb1352e | 996 | return 0; |
8e4e1713 PS |
997 | unlock: |
998 | ovs_unlock(); | |
999 | return err; | |
ccb1352e JG |
1000 | } |
1001 | ||
1002 | static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) | |
1003 | { | |
1004 | struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); | |
b637e498 | 1005 | struct table_instance *ti; |
ccb1352e | 1006 | struct datapath *dp; |
ccb1352e | 1007 | |
d57170b1 | 1008 | rcu_read_lock(); |
46df7b81 | 1009 | dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); |
8e4e1713 | 1010 | if (!dp) { |
d57170b1 | 1011 | rcu_read_unlock(); |
ccb1352e | 1012 | return -ENODEV; |
8e4e1713 | 1013 | } |
ccb1352e | 1014 | |
b637e498 | 1015 | ti = rcu_dereference(dp->table.ti); |
ccb1352e JG |
1016 | for (;;) { |
1017 | struct sw_flow *flow; | |
1018 | u32 bucket, obj; | |
1019 | ||
1020 | bucket = cb->args[0]; | |
1021 | obj = cb->args[1]; | |
b637e498 | 1022 | flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj); |
ccb1352e JG |
1023 | if (!flow) |
1024 | break; | |
1025 | ||
1026 | if (ovs_flow_cmd_fill_info(flow, dp, skb, | |
15e47304 | 1027 | NETLINK_CB(cb->skb).portid, |
ccb1352e JG |
1028 | cb->nlh->nlmsg_seq, NLM_F_MULTI, |
1029 | OVS_FLOW_CMD_NEW) < 0) | |
1030 | break; | |
1031 | ||
1032 | cb->args[0] = bucket; | |
1033 | cb->args[1] = obj; | |
1034 | } | |
d57170b1 | 1035 | rcu_read_unlock(); |
ccb1352e JG |
1036 | return skb->len; |
1037 | } | |
1038 | ||
4534de83 | 1039 | static const struct genl_ops dp_flow_genl_ops[] = { |
ccb1352e JG |
1040 | { .cmd = OVS_FLOW_CMD_NEW, |
1041 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1042 | .policy = flow_policy, | |
1043 | .doit = ovs_flow_cmd_new_or_set | |
1044 | }, | |
1045 | { .cmd = OVS_FLOW_CMD_DEL, | |
1046 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1047 | .policy = flow_policy, | |
1048 | .doit = ovs_flow_cmd_del | |
1049 | }, | |
1050 | { .cmd = OVS_FLOW_CMD_GET, | |
1051 | .flags = 0, /* OK for unprivileged users. */ | |
1052 | .policy = flow_policy, | |
1053 | .doit = ovs_flow_cmd_get, | |
1054 | .dumpit = ovs_flow_cmd_dump | |
1055 | }, | |
1056 | { .cmd = OVS_FLOW_CMD_SET, | |
1057 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1058 | .policy = flow_policy, | |
1059 | .doit = ovs_flow_cmd_new_or_set, | |
1060 | }, | |
1061 | }; | |
1062 | ||
1063 | static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = { | |
1064 | [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, | |
1065 | [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 }, | |
43d4be9c | 1066 | [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 }, |
ccb1352e JG |
1067 | }; |
1068 | ||
1069 | static struct genl_family dp_datapath_genl_family = { | |
1070 | .id = GENL_ID_GENERATE, | |
1071 | .hdrsize = sizeof(struct ovs_header), | |
1072 | .name = OVS_DATAPATH_FAMILY, | |
1073 | .version = OVS_DATAPATH_VERSION, | |
46df7b81 | 1074 | .maxattr = OVS_DP_ATTR_MAX, |
3a4e0d6a PS |
1075 | .netnsok = true, |
1076 | .parallel_ops = true, | |
ccb1352e JG |
1077 | }; |
1078 | ||
1079 | static struct genl_multicast_group ovs_dp_datapath_multicast_group = { | |
1080 | .name = OVS_DATAPATH_MCGROUP | |
1081 | }; | |
1082 | ||
c3ff8cfe TG |
1083 | static size_t ovs_dp_cmd_msg_size(void) |
1084 | { | |
1085 | size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header)); | |
1086 | ||
1087 | msgsize += nla_total_size(IFNAMSIZ); | |
1088 | msgsize += nla_total_size(sizeof(struct ovs_dp_stats)); | |
1bd7116f | 1089 | msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats)); |
c3ff8cfe TG |
1090 | |
1091 | return msgsize; | |
1092 | } | |
1093 | ||
ccb1352e | 1094 | static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb, |
15e47304 | 1095 | u32 portid, u32 seq, u32 flags, u8 cmd) |
ccb1352e JG |
1096 | { |
1097 | struct ovs_header *ovs_header; | |
1098 | struct ovs_dp_stats dp_stats; | |
1bd7116f | 1099 | struct ovs_dp_megaflow_stats dp_megaflow_stats; |
ccb1352e JG |
1100 | int err; |
1101 | ||
15e47304 | 1102 | ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family, |
ccb1352e JG |
1103 | flags, cmd); |
1104 | if (!ovs_header) | |
1105 | goto error; | |
1106 | ||
1107 | ovs_header->dp_ifindex = get_dpifindex(dp); | |
1108 | ||
1109 | rcu_read_lock(); | |
1110 | err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp)); | |
1111 | rcu_read_unlock(); | |
1112 | if (err) | |
1113 | goto nla_put_failure; | |
1114 | ||
1bd7116f AZ |
1115 | get_dp_stats(dp, &dp_stats, &dp_megaflow_stats); |
1116 | if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), | |
1117 | &dp_stats)) | |
1118 | goto nla_put_failure; | |
1119 | ||
1120 | if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS, | |
1121 | sizeof(struct ovs_dp_megaflow_stats), | |
1122 | &dp_megaflow_stats)) | |
028d6a67 | 1123 | goto nla_put_failure; |
ccb1352e | 1124 | |
43d4be9c TG |
1125 | if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features)) |
1126 | goto nla_put_failure; | |
1127 | ||
ccb1352e JG |
1128 | return genlmsg_end(skb, ovs_header); |
1129 | ||
1130 | nla_put_failure: | |
1131 | genlmsg_cancel(skb, ovs_header); | |
1132 | error: | |
1133 | return -EMSGSIZE; | |
1134 | } | |
1135 | ||
795449d8 TG |
1136 | static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, |
1137 | struct genl_info *info, u8 cmd) | |
ccb1352e JG |
1138 | { |
1139 | struct sk_buff *skb; | |
1140 | int retval; | |
1141 | ||
795449d8 | 1142 | skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL); |
ccb1352e JG |
1143 | if (!skb) |
1144 | return ERR_PTR(-ENOMEM); | |
1145 | ||
795449d8 | 1146 | retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd); |
ccb1352e JG |
1147 | if (retval < 0) { |
1148 | kfree_skb(skb); | |
1149 | return ERR_PTR(retval); | |
1150 | } | |
1151 | return skb; | |
1152 | } | |
1153 | ||
8e4e1713 | 1154 | /* Called with ovs_mutex. */ |
46df7b81 PS |
1155 | static struct datapath *lookup_datapath(struct net *net, |
1156 | struct ovs_header *ovs_header, | |
ccb1352e JG |
1157 | struct nlattr *a[OVS_DP_ATTR_MAX + 1]) |
1158 | { | |
1159 | struct datapath *dp; | |
1160 | ||
1161 | if (!a[OVS_DP_ATTR_NAME]) | |
46df7b81 | 1162 | dp = get_dp(net, ovs_header->dp_ifindex); |
ccb1352e JG |
1163 | else { |
1164 | struct vport *vport; | |
1165 | ||
1166 | rcu_read_lock(); | |
46df7b81 | 1167 | vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME])); |
ccb1352e JG |
1168 | dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL; |
1169 | rcu_read_unlock(); | |
1170 | } | |
1171 | return dp ? dp : ERR_PTR(-ENODEV); | |
1172 | } | |
1173 | ||
44da5ae5 TG |
1174 | static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info) |
1175 | { | |
1176 | struct datapath *dp; | |
1177 | ||
1178 | dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); | |
1179 | if (!dp) | |
1180 | return; | |
1181 | ||
1182 | WARN(dp->user_features, "Dropping previously announced user features\n"); | |
1183 | dp->user_features = 0; | |
1184 | } | |
1185 | ||
43d4be9c TG |
1186 | static void ovs_dp_change(struct datapath *dp, struct nlattr **a) |
1187 | { | |
1188 | if (a[OVS_DP_ATTR_USER_FEATURES]) | |
1189 | dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]); | |
1190 | } | |
1191 | ||
ccb1352e JG |
1192 | static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) |
1193 | { | |
1194 | struct nlattr **a = info->attrs; | |
1195 | struct vport_parms parms; | |
1196 | struct sk_buff *reply; | |
1197 | struct datapath *dp; | |
1198 | struct vport *vport; | |
46df7b81 | 1199 | struct ovs_net *ovs_net; |
15eac2a7 | 1200 | int err, i; |
ccb1352e JG |
1201 | |
1202 | err = -EINVAL; | |
1203 | if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID]) | |
1204 | goto err; | |
1205 | ||
8e4e1713 | 1206 | ovs_lock(); |
ccb1352e JG |
1207 | |
1208 | err = -ENOMEM; | |
1209 | dp = kzalloc(sizeof(*dp), GFP_KERNEL); | |
1210 | if (dp == NULL) | |
8e4e1713 | 1211 | goto err_unlock_ovs; |
46df7b81 | 1212 | |
46df7b81 | 1213 | ovs_dp_set_net(dp, hold_net(sock_net(skb->sk))); |
ccb1352e JG |
1214 | |
1215 | /* Allocate table. */ | |
b637e498 PS |
1216 | err = ovs_flow_tbl_init(&dp->table); |
1217 | if (err) | |
ccb1352e JG |
1218 | goto err_free_dp; |
1219 | ||
1220 | dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); | |
1221 | if (!dp->stats_percpu) { | |
1222 | err = -ENOMEM; | |
1223 | goto err_destroy_table; | |
1224 | } | |
1225 | ||
827da44c JS |
1226 | for_each_possible_cpu(i) { |
1227 | struct dp_stats_percpu *dpath_stats; | |
1228 | dpath_stats = per_cpu_ptr(dp->stats_percpu, i); | |
1229 | u64_stats_init(&dpath_stats->sync); | |
1230 | } | |
1231 | ||
15eac2a7 | 1232 | dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head), |
e6445719 | 1233 | GFP_KERNEL); |
15eac2a7 PS |
1234 | if (!dp->ports) { |
1235 | err = -ENOMEM; | |
1236 | goto err_destroy_percpu; | |
1237 | } | |
1238 | ||
1239 | for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) | |
1240 | INIT_HLIST_HEAD(&dp->ports[i]); | |
1241 | ||
ccb1352e JG |
1242 | /* Set up our datapath device. */ |
1243 | parms.name = nla_data(a[OVS_DP_ATTR_NAME]); | |
1244 | parms.type = OVS_VPORT_TYPE_INTERNAL; | |
1245 | parms.options = NULL; | |
1246 | parms.dp = dp; | |
1247 | parms.port_no = OVSP_LOCAL; | |
15e47304 | 1248 | parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]); |
ccb1352e | 1249 | |
43d4be9c TG |
1250 | ovs_dp_change(dp, a); |
1251 | ||
ccb1352e JG |
1252 | vport = new_vport(&parms); |
1253 | if (IS_ERR(vport)) { | |
1254 | err = PTR_ERR(vport); | |
1255 | if (err == -EBUSY) | |
1256 | err = -EEXIST; | |
1257 | ||
44da5ae5 TG |
1258 | if (err == -EEXIST) { |
1259 | /* An outdated user space instance that does not understand | |
1260 | * the concept of user_features has attempted to create a new | |
1261 | * datapath and is likely to reuse it. Drop all user features. | |
1262 | */ | |
1263 | if (info->genlhdr->version < OVS_DP_VER_FEATURES) | |
1264 | ovs_dp_reset_user_features(skb, info); | |
1265 | } | |
1266 | ||
15eac2a7 | 1267 | goto err_destroy_ports_array; |
ccb1352e JG |
1268 | } |
1269 | ||
795449d8 | 1270 | reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW); |
ccb1352e JG |
1271 | err = PTR_ERR(reply); |
1272 | if (IS_ERR(reply)) | |
1273 | goto err_destroy_local_port; | |
1274 | ||
46df7b81 | 1275 | ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id); |
59a35d60 | 1276 | list_add_tail_rcu(&dp->list_node, &ovs_net->dps); |
8e4e1713 PS |
1277 | |
1278 | ovs_unlock(); | |
ccb1352e | 1279 | |
2a94fe48 | 1280 | ovs_notify(&dp_datapath_genl_family, reply, info); |
ccb1352e JG |
1281 | return 0; |
1282 | ||
1283 | err_destroy_local_port: | |
8e4e1713 | 1284 | ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL)); |
15eac2a7 PS |
1285 | err_destroy_ports_array: |
1286 | kfree(dp->ports); | |
ccb1352e JG |
1287 | err_destroy_percpu: |
1288 | free_percpu(dp->stats_percpu); | |
1289 | err_destroy_table: | |
618ed0c8 | 1290 | ovs_flow_tbl_destroy(&dp->table); |
ccb1352e | 1291 | err_free_dp: |
46df7b81 | 1292 | release_net(ovs_dp_get_net(dp)); |
ccb1352e | 1293 | kfree(dp); |
8e4e1713 PS |
1294 | err_unlock_ovs: |
1295 | ovs_unlock(); | |
ccb1352e JG |
1296 | err: |
1297 | return err; | |
1298 | } | |
1299 | ||
8e4e1713 | 1300 | /* Called with ovs_mutex. */ |
46df7b81 | 1301 | static void __dp_destroy(struct datapath *dp) |
ccb1352e | 1302 | { |
15eac2a7 | 1303 | int i; |
ccb1352e | 1304 | |
15eac2a7 PS |
1305 | for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) { |
1306 | struct vport *vport; | |
b67bfe0d | 1307 | struct hlist_node *n; |
15eac2a7 | 1308 | |
b67bfe0d | 1309 | hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node) |
15eac2a7 PS |
1310 | if (vport->port_no != OVSP_LOCAL) |
1311 | ovs_dp_detach_port(vport); | |
1312 | } | |
ccb1352e | 1313 | |
59a35d60 | 1314 | list_del_rcu(&dp->list_node); |
ccb1352e | 1315 | |
8e4e1713 PS |
1316 | /* OVSP_LOCAL is datapath internal port. We need to make sure that |
1317 | * all port in datapath are destroyed first before freeing datapath. | |
ccb1352e | 1318 | */ |
8e4e1713 | 1319 | ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL)); |
ccb1352e JG |
1320 | |
1321 | call_rcu(&dp->rcu, destroy_dp_rcu); | |
46df7b81 PS |
1322 | } |
1323 | ||
1324 | static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info) | |
1325 | { | |
1326 | struct sk_buff *reply; | |
1327 | struct datapath *dp; | |
1328 | int err; | |
1329 | ||
8e4e1713 | 1330 | ovs_lock(); |
46df7b81 PS |
1331 | dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); |
1332 | err = PTR_ERR(dp); | |
1333 | if (IS_ERR(dp)) | |
8e4e1713 | 1334 | goto unlock; |
46df7b81 | 1335 | |
795449d8 | 1336 | reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL); |
46df7b81 PS |
1337 | err = PTR_ERR(reply); |
1338 | if (IS_ERR(reply)) | |
8e4e1713 | 1339 | goto unlock; |
46df7b81 PS |
1340 | |
1341 | __dp_destroy(dp); | |
8e4e1713 | 1342 | ovs_unlock(); |
ccb1352e | 1343 | |
2a94fe48 | 1344 | ovs_notify(&dp_datapath_genl_family, reply, info); |
ccb1352e JG |
1345 | |
1346 | return 0; | |
8e4e1713 PS |
1347 | unlock: |
1348 | ovs_unlock(); | |
1349 | return err; | |
ccb1352e JG |
1350 | } |
1351 | ||
1352 | static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info) | |
1353 | { | |
1354 | struct sk_buff *reply; | |
1355 | struct datapath *dp; | |
1356 | int err; | |
1357 | ||
8e4e1713 | 1358 | ovs_lock(); |
46df7b81 | 1359 | dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); |
8e4e1713 | 1360 | err = PTR_ERR(dp); |
ccb1352e | 1361 | if (IS_ERR(dp)) |
8e4e1713 | 1362 | goto unlock; |
ccb1352e | 1363 | |
43d4be9c TG |
1364 | ovs_dp_change(dp, info->attrs); |
1365 | ||
795449d8 | 1366 | reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW); |
ccb1352e JG |
1367 | if (IS_ERR(reply)) { |
1368 | err = PTR_ERR(reply); | |
68eb5503 | 1369 | genl_set_err(&dp_datapath_genl_family, sock_net(skb->sk), 0, |
2a94fe48 | 1370 | 0, err); |
8e4e1713 PS |
1371 | err = 0; |
1372 | goto unlock; | |
ccb1352e JG |
1373 | } |
1374 | ||
8e4e1713 | 1375 | ovs_unlock(); |
2a94fe48 | 1376 | ovs_notify(&dp_datapath_genl_family, reply, info); |
ccb1352e JG |
1377 | |
1378 | return 0; | |
8e4e1713 PS |
1379 | unlock: |
1380 | ovs_unlock(); | |
1381 | return err; | |
ccb1352e JG |
1382 | } |
1383 | ||
1384 | static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info) | |
1385 | { | |
1386 | struct sk_buff *reply; | |
1387 | struct datapath *dp; | |
8e4e1713 | 1388 | int err; |
ccb1352e | 1389 | |
8e4e1713 | 1390 | ovs_lock(); |
46df7b81 | 1391 | dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); |
8e4e1713 PS |
1392 | if (IS_ERR(dp)) { |
1393 | err = PTR_ERR(dp); | |
1394 | goto unlock; | |
1395 | } | |
ccb1352e | 1396 | |
795449d8 | 1397 | reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW); |
8e4e1713 PS |
1398 | if (IS_ERR(reply)) { |
1399 | err = PTR_ERR(reply); | |
1400 | goto unlock; | |
1401 | } | |
ccb1352e | 1402 | |
8e4e1713 | 1403 | ovs_unlock(); |
ccb1352e | 1404 | return genlmsg_reply(reply, info); |
8e4e1713 PS |
1405 | |
1406 | unlock: | |
1407 | ovs_unlock(); | |
1408 | return err; | |
ccb1352e JG |
1409 | } |
1410 | ||
1411 | static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) | |
1412 | { | |
46df7b81 | 1413 | struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id); |
ccb1352e JG |
1414 | struct datapath *dp; |
1415 | int skip = cb->args[0]; | |
1416 | int i = 0; | |
1417 | ||
59a35d60 PS |
1418 | rcu_read_lock(); |
1419 | list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) { | |
77676fdb | 1420 | if (i >= skip && |
15e47304 | 1421 | ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid, |
ccb1352e JG |
1422 | cb->nlh->nlmsg_seq, NLM_F_MULTI, |
1423 | OVS_DP_CMD_NEW) < 0) | |
1424 | break; | |
1425 | i++; | |
1426 | } | |
59a35d60 | 1427 | rcu_read_unlock(); |
ccb1352e JG |
1428 | |
1429 | cb->args[0] = i; | |
1430 | ||
1431 | return skb->len; | |
1432 | } | |
1433 | ||
4534de83 | 1434 | static const struct genl_ops dp_datapath_genl_ops[] = { |
ccb1352e JG |
1435 | { .cmd = OVS_DP_CMD_NEW, |
1436 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1437 | .policy = datapath_policy, | |
1438 | .doit = ovs_dp_cmd_new | |
1439 | }, | |
1440 | { .cmd = OVS_DP_CMD_DEL, | |
1441 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1442 | .policy = datapath_policy, | |
1443 | .doit = ovs_dp_cmd_del | |
1444 | }, | |
1445 | { .cmd = OVS_DP_CMD_GET, | |
1446 | .flags = 0, /* OK for unprivileged users. */ | |
1447 | .policy = datapath_policy, | |
1448 | .doit = ovs_dp_cmd_get, | |
1449 | .dumpit = ovs_dp_cmd_dump | |
1450 | }, | |
1451 | { .cmd = OVS_DP_CMD_SET, | |
1452 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1453 | .policy = datapath_policy, | |
1454 | .doit = ovs_dp_cmd_set, | |
1455 | }, | |
1456 | }; | |
1457 | ||
1458 | static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = { | |
1459 | [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, | |
1460 | [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) }, | |
1461 | [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 }, | |
1462 | [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 }, | |
1463 | [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 }, | |
1464 | [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, | |
1465 | }; | |
1466 | ||
68eb5503 | 1467 | struct genl_family dp_vport_genl_family = { |
ccb1352e JG |
1468 | .id = GENL_ID_GENERATE, |
1469 | .hdrsize = sizeof(struct ovs_header), | |
1470 | .name = OVS_VPORT_FAMILY, | |
1471 | .version = OVS_VPORT_VERSION, | |
46df7b81 | 1472 | .maxattr = OVS_VPORT_ATTR_MAX, |
3a4e0d6a PS |
1473 | .netnsok = true, |
1474 | .parallel_ops = true, | |
ccb1352e JG |
1475 | }; |
1476 | ||
443cd88c | 1477 | static struct genl_multicast_group ovs_dp_vport_multicast_group = { |
ccb1352e JG |
1478 | .name = OVS_VPORT_MCGROUP |
1479 | }; | |
1480 | ||
8e4e1713 | 1481 | /* Called with ovs_mutex or RCU read lock. */ |
ccb1352e | 1482 | static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb, |
15e47304 | 1483 | u32 portid, u32 seq, u32 flags, u8 cmd) |
ccb1352e JG |
1484 | { |
1485 | struct ovs_header *ovs_header; | |
1486 | struct ovs_vport_stats vport_stats; | |
1487 | int err; | |
1488 | ||
15e47304 | 1489 | ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family, |
ccb1352e JG |
1490 | flags, cmd); |
1491 | if (!ovs_header) | |
1492 | return -EMSGSIZE; | |
1493 | ||
1494 | ovs_header->dp_ifindex = get_dpifindex(vport->dp); | |
1495 | ||
028d6a67 DM |
1496 | if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) || |
1497 | nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) || | |
1498 | nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) || | |
15e47304 | 1499 | nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid)) |
028d6a67 | 1500 | goto nla_put_failure; |
ccb1352e JG |
1501 | |
1502 | ovs_vport_get_stats(vport, &vport_stats); | |
028d6a67 DM |
1503 | if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats), |
1504 | &vport_stats)) | |
1505 | goto nla_put_failure; | |
ccb1352e JG |
1506 | |
1507 | err = ovs_vport_get_options(vport, skb); | |
1508 | if (err == -EMSGSIZE) | |
1509 | goto error; | |
1510 | ||
1511 | return genlmsg_end(skb, ovs_header); | |
1512 | ||
1513 | nla_put_failure: | |
1514 | err = -EMSGSIZE; | |
1515 | error: | |
1516 | genlmsg_cancel(skb, ovs_header); | |
1517 | return err; | |
1518 | } | |
1519 | ||
8e4e1713 | 1520 | /* Called with ovs_mutex or RCU read lock. */ |
15e47304 | 1521 | struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid, |
ccb1352e JG |
1522 | u32 seq, u8 cmd) |
1523 | { | |
1524 | struct sk_buff *skb; | |
1525 | int retval; | |
1526 | ||
1527 | skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); | |
1528 | if (!skb) | |
1529 | return ERR_PTR(-ENOMEM); | |
1530 | ||
15e47304 | 1531 | retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd); |
a9341512 JG |
1532 | BUG_ON(retval < 0); |
1533 | ||
ccb1352e JG |
1534 | return skb; |
1535 | } | |
1536 | ||
8e4e1713 | 1537 | /* Called with ovs_mutex or RCU read lock. */ |
46df7b81 PS |
1538 | static struct vport *lookup_vport(struct net *net, |
1539 | struct ovs_header *ovs_header, | |
ccb1352e JG |
1540 | struct nlattr *a[OVS_VPORT_ATTR_MAX + 1]) |
1541 | { | |
1542 | struct datapath *dp; | |
1543 | struct vport *vport; | |
1544 | ||
1545 | if (a[OVS_VPORT_ATTR_NAME]) { | |
46df7b81 | 1546 | vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME])); |
ccb1352e JG |
1547 | if (!vport) |
1548 | return ERR_PTR(-ENODEV); | |
651a68ea BP |
1549 | if (ovs_header->dp_ifindex && |
1550 | ovs_header->dp_ifindex != get_dpifindex(vport->dp)) | |
1551 | return ERR_PTR(-ENODEV); | |
ccb1352e JG |
1552 | return vport; |
1553 | } else if (a[OVS_VPORT_ATTR_PORT_NO]) { | |
1554 | u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); | |
1555 | ||
1556 | if (port_no >= DP_MAX_PORTS) | |
1557 | return ERR_PTR(-EFBIG); | |
1558 | ||
46df7b81 | 1559 | dp = get_dp(net, ovs_header->dp_ifindex); |
ccb1352e JG |
1560 | if (!dp) |
1561 | return ERR_PTR(-ENODEV); | |
1562 | ||
8e4e1713 | 1563 | vport = ovs_vport_ovsl_rcu(dp, port_no); |
ccb1352e | 1564 | if (!vport) |
14408dba | 1565 | return ERR_PTR(-ENODEV); |
ccb1352e JG |
1566 | return vport; |
1567 | } else | |
1568 | return ERR_PTR(-EINVAL); | |
1569 | } | |
1570 | ||
1571 | static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info) | |
1572 | { | |
1573 | struct nlattr **a = info->attrs; | |
1574 | struct ovs_header *ovs_header = info->userhdr; | |
1575 | struct vport_parms parms; | |
1576 | struct sk_buff *reply; | |
1577 | struct vport *vport; | |
1578 | struct datapath *dp; | |
1579 | u32 port_no; | |
1580 | int err; | |
1581 | ||
1582 | err = -EINVAL; | |
1583 | if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] || | |
1584 | !a[OVS_VPORT_ATTR_UPCALL_PID]) | |
1585 | goto exit; | |
1586 | ||
8e4e1713 | 1587 | ovs_lock(); |
46df7b81 | 1588 | dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); |
ccb1352e JG |
1589 | err = -ENODEV; |
1590 | if (!dp) | |
1591 | goto exit_unlock; | |
1592 | ||
1593 | if (a[OVS_VPORT_ATTR_PORT_NO]) { | |
1594 | port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); | |
1595 | ||
1596 | err = -EFBIG; | |
1597 | if (port_no >= DP_MAX_PORTS) | |
1598 | goto exit_unlock; | |
1599 | ||
8e4e1713 | 1600 | vport = ovs_vport_ovsl(dp, port_no); |
ccb1352e JG |
1601 | err = -EBUSY; |
1602 | if (vport) | |
1603 | goto exit_unlock; | |
1604 | } else { | |
1605 | for (port_no = 1; ; port_no++) { | |
1606 | if (port_no >= DP_MAX_PORTS) { | |
1607 | err = -EFBIG; | |
1608 | goto exit_unlock; | |
1609 | } | |
8e4e1713 | 1610 | vport = ovs_vport_ovsl(dp, port_no); |
ccb1352e JG |
1611 | if (!vport) |
1612 | break; | |
1613 | } | |
1614 | } | |
1615 | ||
1616 | parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]); | |
1617 | parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]); | |
1618 | parms.options = a[OVS_VPORT_ATTR_OPTIONS]; | |
1619 | parms.dp = dp; | |
1620 | parms.port_no = port_no; | |
15e47304 | 1621 | parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]); |
ccb1352e JG |
1622 | |
1623 | vport = new_vport(&parms); | |
1624 | err = PTR_ERR(vport); | |
1625 | if (IS_ERR(vport)) | |
1626 | goto exit_unlock; | |
1627 | ||
cb7c5bdf | 1628 | err = 0; |
15e47304 | 1629 | reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq, |
ccb1352e JG |
1630 | OVS_VPORT_CMD_NEW); |
1631 | if (IS_ERR(reply)) { | |
1632 | err = PTR_ERR(reply); | |
1633 | ovs_dp_detach_port(vport); | |
1634 | goto exit_unlock; | |
1635 | } | |
ed661185 | 1636 | |
2a94fe48 | 1637 | ovs_notify(&dp_vport_genl_family, reply, info); |
ccb1352e JG |
1638 | |
1639 | exit_unlock: | |
8e4e1713 | 1640 | ovs_unlock(); |
ccb1352e JG |
1641 | exit: |
1642 | return err; | |
1643 | } | |
1644 | ||
1645 | static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info) | |
1646 | { | |
1647 | struct nlattr **a = info->attrs; | |
1648 | struct sk_buff *reply; | |
1649 | struct vport *vport; | |
1650 | int err; | |
1651 | ||
8e4e1713 | 1652 | ovs_lock(); |
46df7b81 | 1653 | vport = lookup_vport(sock_net(skb->sk), info->userhdr, a); |
ccb1352e JG |
1654 | err = PTR_ERR(vport); |
1655 | if (IS_ERR(vport)) | |
1656 | goto exit_unlock; | |
1657 | ||
ccb1352e | 1658 | if (a[OVS_VPORT_ATTR_TYPE] && |
f44f3408 | 1659 | nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) { |
ccb1352e | 1660 | err = -EINVAL; |
f44f3408 JG |
1661 | goto exit_unlock; |
1662 | } | |
ccb1352e | 1663 | |
a9341512 JG |
1664 | reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); |
1665 | if (!reply) { | |
1666 | err = -ENOMEM; | |
1667 | goto exit_unlock; | |
1668 | } | |
1669 | ||
f44f3408 | 1670 | if (a[OVS_VPORT_ATTR_OPTIONS]) { |
ccb1352e | 1671 | err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]); |
f44f3408 JG |
1672 | if (err) |
1673 | goto exit_free; | |
1674 | } | |
a9341512 | 1675 | |
03fbf8b3 | 1676 | if (a[OVS_VPORT_ATTR_UPCALL_PID]) |
15e47304 | 1677 | vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]); |
ccb1352e | 1678 | |
a9341512 JG |
1679 | err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid, |
1680 | info->snd_seq, 0, OVS_VPORT_CMD_NEW); | |
1681 | BUG_ON(err < 0); | |
ccb1352e | 1682 | |
8e4e1713 | 1683 | ovs_unlock(); |
2a94fe48 | 1684 | ovs_notify(&dp_vport_genl_family, reply, info); |
8e4e1713 | 1685 | return 0; |
ccb1352e | 1686 | |
a9341512 JG |
1687 | exit_free: |
1688 | kfree_skb(reply); | |
ccb1352e | 1689 | exit_unlock: |
8e4e1713 | 1690 | ovs_unlock(); |
ccb1352e JG |
1691 | return err; |
1692 | } | |
1693 | ||
1694 | static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info) | |
1695 | { | |
1696 | struct nlattr **a = info->attrs; | |
1697 | struct sk_buff *reply; | |
1698 | struct vport *vport; | |
1699 | int err; | |
1700 | ||
8e4e1713 | 1701 | ovs_lock(); |
46df7b81 | 1702 | vport = lookup_vport(sock_net(skb->sk), info->userhdr, a); |
ccb1352e JG |
1703 | err = PTR_ERR(vport); |
1704 | if (IS_ERR(vport)) | |
1705 | goto exit_unlock; | |
1706 | ||
1707 | if (vport->port_no == OVSP_LOCAL) { | |
1708 | err = -EINVAL; | |
1709 | goto exit_unlock; | |
1710 | } | |
1711 | ||
74f84a57 PS |
1712 | reply = ovs_vport_cmd_build_info(vport, info->snd_portid, |
1713 | info->snd_seq, OVS_VPORT_CMD_DEL); | |
ccb1352e JG |
1714 | err = PTR_ERR(reply); |
1715 | if (IS_ERR(reply)) | |
1716 | goto exit_unlock; | |
1717 | ||
734907e8 | 1718 | err = 0; |
ccb1352e JG |
1719 | ovs_dp_detach_port(vport); |
1720 | ||
2a94fe48 | 1721 | ovs_notify(&dp_vport_genl_family, reply, info); |
ccb1352e JG |
1722 | |
1723 | exit_unlock: | |
8e4e1713 | 1724 | ovs_unlock(); |
ccb1352e JG |
1725 | return err; |
1726 | } | |
1727 | ||
1728 | static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info) | |
1729 | { | |
1730 | struct nlattr **a = info->attrs; | |
1731 | struct ovs_header *ovs_header = info->userhdr; | |
1732 | struct sk_buff *reply; | |
1733 | struct vport *vport; | |
1734 | int err; | |
1735 | ||
1736 | rcu_read_lock(); | |
46df7b81 | 1737 | vport = lookup_vport(sock_net(skb->sk), ovs_header, a); |
ccb1352e JG |
1738 | err = PTR_ERR(vport); |
1739 | if (IS_ERR(vport)) | |
1740 | goto exit_unlock; | |
1741 | ||
74f84a57 PS |
1742 | reply = ovs_vport_cmd_build_info(vport, info->snd_portid, |
1743 | info->snd_seq, OVS_VPORT_CMD_NEW); | |
ccb1352e JG |
1744 | err = PTR_ERR(reply); |
1745 | if (IS_ERR(reply)) | |
1746 | goto exit_unlock; | |
1747 | ||
1748 | rcu_read_unlock(); | |
1749 | ||
1750 | return genlmsg_reply(reply, info); | |
1751 | ||
1752 | exit_unlock: | |
1753 | rcu_read_unlock(); | |
1754 | return err; | |
1755 | } | |
1756 | ||
1757 | static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) | |
1758 | { | |
1759 | struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); | |
1760 | struct datapath *dp; | |
15eac2a7 PS |
1761 | int bucket = cb->args[0], skip = cb->args[1]; |
1762 | int i, j = 0; | |
ccb1352e | 1763 | |
46df7b81 | 1764 | dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); |
ccb1352e JG |
1765 | if (!dp) |
1766 | return -ENODEV; | |
1767 | ||
1768 | rcu_read_lock(); | |
15eac2a7 | 1769 | for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) { |
ccb1352e | 1770 | struct vport *vport; |
15eac2a7 PS |
1771 | |
1772 | j = 0; | |
b67bfe0d | 1773 | hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) { |
15eac2a7 PS |
1774 | if (j >= skip && |
1775 | ovs_vport_cmd_fill_info(vport, skb, | |
15e47304 | 1776 | NETLINK_CB(cb->skb).portid, |
15eac2a7 PS |
1777 | cb->nlh->nlmsg_seq, |
1778 | NLM_F_MULTI, | |
1779 | OVS_VPORT_CMD_NEW) < 0) | |
1780 | goto out; | |
1781 | ||
1782 | j++; | |
1783 | } | |
1784 | skip = 0; | |
ccb1352e | 1785 | } |
15eac2a7 | 1786 | out: |
ccb1352e JG |
1787 | rcu_read_unlock(); |
1788 | ||
15eac2a7 PS |
1789 | cb->args[0] = i; |
1790 | cb->args[1] = j; | |
ccb1352e | 1791 | |
15eac2a7 | 1792 | return skb->len; |
ccb1352e JG |
1793 | } |
1794 | ||
4534de83 | 1795 | static const struct genl_ops dp_vport_genl_ops[] = { |
ccb1352e JG |
1796 | { .cmd = OVS_VPORT_CMD_NEW, |
1797 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1798 | .policy = vport_policy, | |
1799 | .doit = ovs_vport_cmd_new | |
1800 | }, | |
1801 | { .cmd = OVS_VPORT_CMD_DEL, | |
1802 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1803 | .policy = vport_policy, | |
1804 | .doit = ovs_vport_cmd_del | |
1805 | }, | |
1806 | { .cmd = OVS_VPORT_CMD_GET, | |
1807 | .flags = 0, /* OK for unprivileged users. */ | |
1808 | .policy = vport_policy, | |
1809 | .doit = ovs_vport_cmd_get, | |
1810 | .dumpit = ovs_vport_cmd_dump | |
1811 | }, | |
1812 | { .cmd = OVS_VPORT_CMD_SET, | |
1813 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1814 | .policy = vport_policy, | |
1815 | .doit = ovs_vport_cmd_set, | |
1816 | }, | |
1817 | }; | |
1818 | ||
1819 | struct genl_family_and_ops { | |
1820 | struct genl_family *family; | |
4534de83 | 1821 | const struct genl_ops *ops; |
ccb1352e | 1822 | int n_ops; |
2a94fe48 | 1823 | const struct genl_multicast_group *group; |
ccb1352e JG |
1824 | }; |
1825 | ||
1826 | static const struct genl_family_and_ops dp_genl_families[] = { | |
1827 | { &dp_datapath_genl_family, | |
1828 | dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops), | |
1829 | &ovs_dp_datapath_multicast_group }, | |
1830 | { &dp_vport_genl_family, | |
1831 | dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops), | |
1832 | &ovs_dp_vport_multicast_group }, | |
1833 | { &dp_flow_genl_family, | |
1834 | dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops), | |
1835 | &ovs_dp_flow_multicast_group }, | |
1836 | { &dp_packet_genl_family, | |
1837 | dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops), | |
1838 | NULL }, | |
1839 | }; | |
1840 | ||
1841 | static void dp_unregister_genl(int n_families) | |
1842 | { | |
1843 | int i; | |
1844 | ||
1845 | for (i = 0; i < n_families; i++) | |
1846 | genl_unregister_family(dp_genl_families[i].family); | |
1847 | } | |
1848 | ||
1849 | static int dp_register_genl(void) | |
1850 | { | |
1851 | int n_registered; | |
1852 | int err; | |
1853 | int i; | |
1854 | ||
1855 | n_registered = 0; | |
1856 | for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) { | |
1857 | const struct genl_family_and_ops *f = &dp_genl_families[i]; | |
1858 | ||
c53ed742 JB |
1859 | f->family->ops = f->ops; |
1860 | f->family->n_ops = f->n_ops; | |
2a94fe48 JB |
1861 | f->family->mcgrps = f->group; |
1862 | f->family->n_mcgrps = f->group ? 1 : 0; | |
c53ed742 | 1863 | err = genl_register_family(f->family); |
ccb1352e JG |
1864 | if (err) |
1865 | goto error; | |
1866 | n_registered++; | |
ccb1352e JG |
1867 | } |
1868 | ||
1869 | return 0; | |
1870 | ||
1871 | error: | |
1872 | dp_unregister_genl(n_registered); | |
1873 | return err; | |
1874 | } | |
1875 | ||
46df7b81 PS |
1876 | static int __net_init ovs_init_net(struct net *net) |
1877 | { | |
1878 | struct ovs_net *ovs_net = net_generic(net, ovs_net_id); | |
1879 | ||
1880 | INIT_LIST_HEAD(&ovs_net->dps); | |
8e4e1713 | 1881 | INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq); |
46df7b81 PS |
1882 | return 0; |
1883 | } | |
1884 | ||
1885 | static void __net_exit ovs_exit_net(struct net *net) | |
1886 | { | |
46df7b81 | 1887 | struct datapath *dp, *dp_next; |
8e4e1713 | 1888 | struct ovs_net *ovs_net = net_generic(net, ovs_net_id); |
46df7b81 | 1889 | |
8e4e1713 | 1890 | ovs_lock(); |
46df7b81 PS |
1891 | list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node) |
1892 | __dp_destroy(dp); | |
8e4e1713 PS |
1893 | ovs_unlock(); |
1894 | ||
1895 | cancel_work_sync(&ovs_net->dp_notify_work); | |
46df7b81 PS |
1896 | } |
1897 | ||
1898 | static struct pernet_operations ovs_net_ops = { | |
1899 | .init = ovs_init_net, | |
1900 | .exit = ovs_exit_net, | |
1901 | .id = &ovs_net_id, | |
1902 | .size = sizeof(struct ovs_net), | |
1903 | }; | |
1904 | ||
ccb1352e JG |
1905 | static int __init dp_init(void) |
1906 | { | |
ccb1352e JG |
1907 | int err; |
1908 | ||
3523b29b | 1909 | BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb)); |
ccb1352e JG |
1910 | |
1911 | pr_info("Open vSwitch switching datapath\n"); | |
1912 | ||
1913 | err = ovs_flow_init(); | |
1914 | if (err) | |
1915 | goto error; | |
1916 | ||
1917 | err = ovs_vport_init(); | |
1918 | if (err) | |
1919 | goto error_flow_exit; | |
1920 | ||
46df7b81 | 1921 | err = register_pernet_device(&ovs_net_ops); |
ccb1352e JG |
1922 | if (err) |
1923 | goto error_vport_exit; | |
1924 | ||
46df7b81 PS |
1925 | err = register_netdevice_notifier(&ovs_dp_device_notifier); |
1926 | if (err) | |
1927 | goto error_netns_exit; | |
1928 | ||
ccb1352e JG |
1929 | err = dp_register_genl(); |
1930 | if (err < 0) | |
1931 | goto error_unreg_notifier; | |
1932 | ||
ccb1352e JG |
1933 | return 0; |
1934 | ||
1935 | error_unreg_notifier: | |
1936 | unregister_netdevice_notifier(&ovs_dp_device_notifier); | |
46df7b81 PS |
1937 | error_netns_exit: |
1938 | unregister_pernet_device(&ovs_net_ops); | |
ccb1352e JG |
1939 | error_vport_exit: |
1940 | ovs_vport_exit(); | |
1941 | error_flow_exit: | |
1942 | ovs_flow_exit(); | |
1943 | error: | |
1944 | return err; | |
1945 | } | |
1946 | ||
1947 | static void dp_cleanup(void) | |
1948 | { | |
ccb1352e JG |
1949 | dp_unregister_genl(ARRAY_SIZE(dp_genl_families)); |
1950 | unregister_netdevice_notifier(&ovs_dp_device_notifier); | |
46df7b81 PS |
1951 | unregister_pernet_device(&ovs_net_ops); |
1952 | rcu_barrier(); | |
ccb1352e JG |
1953 | ovs_vport_exit(); |
1954 | ovs_flow_exit(); | |
1955 | } | |
1956 | ||
1957 | module_init(dp_init); | |
1958 | module_exit(dp_cleanup); | |
1959 | ||
1960 | MODULE_DESCRIPTION("Open vSwitch switching datapath"); | |
1961 | MODULE_LICENSE("GPL"); |