Commit | Line | Data |
---|---|---|
ccb1352e | 1 | /* |
ad552007 | 2 | * Copyright (c) 2007-2014 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> | |
47 | #include <linux/openvswitch.h> | |
48 | #include <linux/rculist.h> | |
49 | #include <linux/dmi.h> | |
ccb1352e | 50 | #include <net/genetlink.h> |
46df7b81 PS |
51 | #include <net/net_namespace.h> |
52 | #include <net/netns/generic.h> | |
ccb1352e JG |
53 | |
54 | #include "datapath.h" | |
55 | #include "flow.h" | |
e80857cc | 56 | #include "flow_table.h" |
e6445719 | 57 | #include "flow_netlink.h" |
ccb1352e | 58 | #include "vport-internal_dev.h" |
cff63a52 | 59 | #include "vport-netdev.h" |
ccb1352e | 60 | |
8e4e1713 | 61 | int ovs_net_id __read_mostly; |
9ba559d9 | 62 | EXPORT_SYMBOL_GPL(ovs_net_id); |
8e4e1713 | 63 | |
0c200ef9 PS |
64 | static struct genl_family dp_packet_genl_family; |
65 | static struct genl_family dp_flow_genl_family; | |
66 | static struct genl_family dp_datapath_genl_family; | |
67 | ||
48e48a70 | 68 | static const struct genl_multicast_group ovs_dp_flow_multicast_group = { |
69 | .name = OVS_FLOW_MCGROUP, | |
0c200ef9 PS |
70 | }; |
71 | ||
48e48a70 | 72 | static const struct genl_multicast_group ovs_dp_datapath_multicast_group = { |
73 | .name = OVS_DATAPATH_MCGROUP, | |
0c200ef9 PS |
74 | }; |
75 | ||
48e48a70 | 76 | static const struct genl_multicast_group ovs_dp_vport_multicast_group = { |
77 | .name = OVS_VPORT_MCGROUP, | |
0c200ef9 PS |
78 | }; |
79 | ||
fb5d1e9e JR |
80 | /* Check if need to build a reply message. |
81 | * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */ | |
9b67aa4a SG |
82 | static bool ovs_must_notify(struct genl_family *family, struct genl_info *info, |
83 | unsigned int group) | |
fb5d1e9e JR |
84 | { |
85 | return info->nlhdr->nlmsg_flags & NLM_F_ECHO || | |
f8403a2e | 86 | genl_has_listeners(family, genl_info_net(info), group); |
fb5d1e9e JR |
87 | } |
88 | ||
68eb5503 | 89 | static void ovs_notify(struct genl_family *family, |
2a94fe48 | 90 | struct sk_buff *skb, struct genl_info *info) |
ed661185 | 91 | { |
68eb5503 | 92 | genl_notify(family, skb, genl_info_net(info), info->snd_portid, |
2a94fe48 | 93 | 0, info->nlhdr, GFP_KERNEL); |
ed661185 TG |
94 | } |
95 | ||
ccb1352e JG |
96 | /** |
97 | * DOC: Locking: | |
98 | * | |
8e4e1713 PS |
99 | * All writes e.g. Writes to device state (add/remove datapath, port, set |
100 | * operations on vports, etc.), Writes to other state (flow table | |
101 | * modifications, set miscellaneous datapath parameters, etc.) are protected | |
102 | * by ovs_lock. | |
ccb1352e JG |
103 | * |
104 | * Reads are protected by RCU. | |
105 | * | |
106 | * There are a few special cases (mostly stats) that have their own | |
107 | * synchronization but they nest under all of above and don't interact with | |
108 | * each other. | |
8e4e1713 PS |
109 | * |
110 | * The RTNL lock nests inside ovs_mutex. | |
ccb1352e JG |
111 | */ |
112 | ||
8e4e1713 PS |
113 | static DEFINE_MUTEX(ovs_mutex); |
114 | ||
115 | void ovs_lock(void) | |
116 | { | |
117 | mutex_lock(&ovs_mutex); | |
118 | } | |
119 | ||
120 | void ovs_unlock(void) | |
121 | { | |
122 | mutex_unlock(&ovs_mutex); | |
123 | } | |
124 | ||
125 | #ifdef CONFIG_LOCKDEP | |
126 | int lockdep_ovsl_is_held(void) | |
127 | { | |
128 | if (debug_locks) | |
129 | return lockdep_is_held(&ovs_mutex); | |
130 | else | |
131 | return 1; | |
132 | } | |
9ba559d9 | 133 | EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held); |
8e4e1713 PS |
134 | #endif |
135 | ||
ccb1352e | 136 | static struct vport *new_vport(const struct vport_parms *); |
8055a89c | 137 | static int queue_gso_packets(struct datapath *dp, struct sk_buff *, |
e8eedb85 | 138 | const struct sw_flow_key *, |
ccb1352e | 139 | const struct dp_upcall_info *); |
8055a89c | 140 | static int queue_userspace_packet(struct datapath *dp, struct sk_buff *, |
e8eedb85 | 141 | const struct sw_flow_key *, |
ccb1352e JG |
142 | const struct dp_upcall_info *); |
143 | ||
cc3a5ae6 AZ |
144 | /* Must be called with rcu_read_lock. */ |
145 | static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex) | |
ccb1352e | 146 | { |
cc3a5ae6 | 147 | struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex); |
ccb1352e | 148 | |
ccb1352e JG |
149 | if (dev) { |
150 | struct vport *vport = ovs_internal_dev_get_vport(dev); | |
151 | if (vport) | |
cc3a5ae6 | 152 | return vport->dp; |
ccb1352e | 153 | } |
cc3a5ae6 AZ |
154 | |
155 | return NULL; | |
156 | } | |
157 | ||
158 | /* The caller must hold either ovs_mutex or rcu_read_lock to keep the | |
159 | * returned dp pointer valid. | |
160 | */ | |
161 | static inline struct datapath *get_dp(struct net *net, int dp_ifindex) | |
162 | { | |
163 | struct datapath *dp; | |
164 | ||
165 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held()); | |
166 | rcu_read_lock(); | |
167 | dp = get_dp_rcu(net, dp_ifindex); | |
ccb1352e JG |
168 | rcu_read_unlock(); |
169 | ||
170 | return dp; | |
171 | } | |
172 | ||
8e4e1713 | 173 | /* Must be called with rcu_read_lock or ovs_mutex. */ |
971427f3 | 174 | const char *ovs_dp_name(const struct datapath *dp) |
ccb1352e | 175 | { |
8e4e1713 | 176 | struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL); |
ccb1352e JG |
177 | return vport->ops->get_name(vport); |
178 | } | |
179 | ||
12eb18f7 | 180 | static int get_dpifindex(const struct datapath *dp) |
ccb1352e JG |
181 | { |
182 | struct vport *local; | |
183 | int ifindex; | |
184 | ||
185 | rcu_read_lock(); | |
186 | ||
15eac2a7 | 187 | local = ovs_vport_rcu(dp, OVSP_LOCAL); |
ccb1352e | 188 | if (local) |
cff63a52 | 189 | ifindex = netdev_vport_priv(local)->dev->ifindex; |
ccb1352e JG |
190 | else |
191 | ifindex = 0; | |
192 | ||
193 | rcu_read_unlock(); | |
194 | ||
195 | return ifindex; | |
196 | } | |
197 | ||
198 | static void destroy_dp_rcu(struct rcu_head *rcu) | |
199 | { | |
200 | struct datapath *dp = container_of(rcu, struct datapath, rcu); | |
201 | ||
9b996e54 | 202 | ovs_flow_tbl_destroy(&dp->table); |
ccb1352e | 203 | free_percpu(dp->stats_percpu); |
46df7b81 | 204 | release_net(ovs_dp_get_net(dp)); |
15eac2a7 | 205 | kfree(dp->ports); |
ccb1352e JG |
206 | kfree(dp); |
207 | } | |
208 | ||
15eac2a7 PS |
209 | static struct hlist_head *vport_hash_bucket(const struct datapath *dp, |
210 | u16 port_no) | |
211 | { | |
212 | return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)]; | |
213 | } | |
214 | ||
bb6f9a70 | 215 | /* Called with ovs_mutex or RCU read lock. */ |
15eac2a7 PS |
216 | struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no) |
217 | { | |
218 | struct vport *vport; | |
15eac2a7 PS |
219 | struct hlist_head *head; |
220 | ||
221 | head = vport_hash_bucket(dp, port_no); | |
b67bfe0d | 222 | hlist_for_each_entry_rcu(vport, head, dp_hash_node) { |
15eac2a7 PS |
223 | if (vport->port_no == port_no) |
224 | return vport; | |
225 | } | |
226 | return NULL; | |
227 | } | |
228 | ||
8e4e1713 | 229 | /* Called with ovs_mutex. */ |
ccb1352e JG |
230 | static struct vport *new_vport(const struct vport_parms *parms) |
231 | { | |
232 | struct vport *vport; | |
233 | ||
234 | vport = ovs_vport_add(parms); | |
235 | if (!IS_ERR(vport)) { | |
236 | struct datapath *dp = parms->dp; | |
15eac2a7 | 237 | struct hlist_head *head = vport_hash_bucket(dp, vport->port_no); |
ccb1352e | 238 | |
15eac2a7 | 239 | hlist_add_head_rcu(&vport->dp_hash_node, head); |
ccb1352e | 240 | } |
ccb1352e JG |
241 | return vport; |
242 | } | |
243 | ||
ccb1352e JG |
244 | void ovs_dp_detach_port(struct vport *p) |
245 | { | |
8e4e1713 | 246 | ASSERT_OVSL(); |
ccb1352e JG |
247 | |
248 | /* First drop references to device. */ | |
15eac2a7 | 249 | hlist_del_rcu(&p->dp_hash_node); |
ccb1352e JG |
250 | |
251 | /* Then destroy it. */ | |
252 | ovs_vport_del(p); | |
253 | } | |
254 | ||
255 | /* Must be called with rcu_read_lock. */ | |
8c8b1b83 | 256 | void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key) |
ccb1352e | 257 | { |
83c8df26 | 258 | const struct vport *p = OVS_CB(skb)->input_vport; |
ccb1352e JG |
259 | struct datapath *dp = p->dp; |
260 | struct sw_flow *flow; | |
d98612b8 | 261 | struct sw_flow_actions *sf_acts; |
ccb1352e | 262 | struct dp_stats_percpu *stats; |
ccb1352e | 263 | u64 *stats_counter; |
1bd7116f | 264 | u32 n_mask_hit; |
ccb1352e | 265 | |
404f2f10 | 266 | stats = this_cpu_ptr(dp->stats_percpu); |
ccb1352e | 267 | |
ccb1352e | 268 | /* Look up flow. */ |
8c8b1b83 | 269 | flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit); |
ccb1352e JG |
270 | if (unlikely(!flow)) { |
271 | struct dp_upcall_info upcall; | |
8c8b1b83 | 272 | int error; |
ccb1352e JG |
273 | |
274 | upcall.cmd = OVS_PACKET_CMD_MISS; | |
ccb1352e | 275 | upcall.userdata = NULL; |
5cd667b0 | 276 | upcall.portid = ovs_vport_find_upcall_portid(p, skb); |
8f0aad6f | 277 | upcall.egress_tun_info = NULL; |
e8eedb85 | 278 | error = ovs_dp_upcall(dp, skb, key, &upcall); |
c5eba0b6 LR |
279 | if (unlikely(error)) |
280 | kfree_skb(skb); | |
281 | else | |
282 | consume_skb(skb); | |
ccb1352e JG |
283 | stats_counter = &stats->n_missed; |
284 | goto out; | |
285 | } | |
286 | ||
d98612b8 LJ |
287 | ovs_flow_stats_update(flow, key->tp.flags, skb); |
288 | sf_acts = rcu_dereference(flow->sf_acts); | |
289 | ovs_execute_actions(dp, skb, sf_acts, key); | |
ccb1352e | 290 | |
e298e505 | 291 | stats_counter = &stats->n_hit; |
ccb1352e JG |
292 | |
293 | out: | |
294 | /* Update datapath statistics. */ | |
df9d9fdf | 295 | u64_stats_update_begin(&stats->syncp); |
ccb1352e | 296 | (*stats_counter)++; |
1bd7116f | 297 | stats->n_mask_hit += n_mask_hit; |
df9d9fdf | 298 | u64_stats_update_end(&stats->syncp); |
ccb1352e JG |
299 | } |
300 | ||
ccb1352e | 301 | int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb, |
e8eedb85 | 302 | const struct sw_flow_key *key, |
46df7b81 | 303 | const struct dp_upcall_info *upcall_info) |
ccb1352e JG |
304 | { |
305 | struct dp_stats_percpu *stats; | |
ccb1352e JG |
306 | int err; |
307 | ||
15e47304 | 308 | if (upcall_info->portid == 0) { |
ccb1352e JG |
309 | err = -ENOTCONN; |
310 | goto err; | |
311 | } | |
312 | ||
ccb1352e | 313 | if (!skb_is_gso(skb)) |
e8eedb85 | 314 | err = queue_userspace_packet(dp, skb, key, upcall_info); |
ccb1352e | 315 | else |
e8eedb85 | 316 | err = queue_gso_packets(dp, skb, key, upcall_info); |
ccb1352e JG |
317 | if (err) |
318 | goto err; | |
319 | ||
320 | return 0; | |
321 | ||
322 | err: | |
404f2f10 | 323 | stats = this_cpu_ptr(dp->stats_percpu); |
ccb1352e | 324 | |
df9d9fdf | 325 | u64_stats_update_begin(&stats->syncp); |
ccb1352e | 326 | stats->n_lost++; |
df9d9fdf | 327 | u64_stats_update_end(&stats->syncp); |
ccb1352e JG |
328 | |
329 | return err; | |
330 | } | |
331 | ||
8055a89c | 332 | static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb, |
e8eedb85 | 333 | const struct sw_flow_key *key, |
ccb1352e JG |
334 | const struct dp_upcall_info *upcall_info) |
335 | { | |
a1b5d0dd | 336 | unsigned short gso_type = skb_shinfo(skb)->gso_type; |
ccb1352e JG |
337 | struct sw_flow_key later_key; |
338 | struct sk_buff *segs, *nskb; | |
e8eedb85 | 339 | struct ovs_skb_cb ovs_cb; |
ccb1352e JG |
340 | int err; |
341 | ||
e8eedb85 | 342 | ovs_cb = *OVS_CB(skb); |
09c5e605 | 343 | segs = __skb_gso_segment(skb, NETIF_F_SG, false); |
e8eedb85 | 344 | *OVS_CB(skb) = ovs_cb; |
92e5dfc3 PS |
345 | if (IS_ERR(segs)) |
346 | return PTR_ERR(segs); | |
330966e5 FW |
347 | if (segs == NULL) |
348 | return -EINVAL; | |
ccb1352e | 349 | |
e8eedb85 PS |
350 | if (gso_type & SKB_GSO_UDP) { |
351 | /* The initial flow key extracted by ovs_flow_key_extract() | |
352 | * in this case is for a first fragment, so we need to | |
353 | * properly mark later fragments. | |
354 | */ | |
355 | later_key = *key; | |
356 | later_key.ip.frag = OVS_FRAG_TYPE_LATER; | |
357 | } | |
358 | ||
ccb1352e JG |
359 | /* Queue all of the segments. */ |
360 | skb = segs; | |
361 | do { | |
e8eedb85 PS |
362 | *OVS_CB(skb) = ovs_cb; |
363 | if (gso_type & SKB_GSO_UDP && skb != segs) | |
364 | key = &later_key; | |
365 | ||
366 | err = queue_userspace_packet(dp, skb, key, upcall_info); | |
ccb1352e JG |
367 | if (err) |
368 | break; | |
369 | ||
ccb1352e JG |
370 | } while ((skb = skb->next)); |
371 | ||
372 | /* Free all of the segments. */ | |
373 | skb = segs; | |
374 | do { | |
375 | nskb = skb->next; | |
376 | if (err) | |
377 | kfree_skb(skb); | |
378 | else | |
379 | consume_skb(skb); | |
380 | } while ((skb = nskb)); | |
381 | return err; | |
382 | } | |
383 | ||
8f0aad6f | 384 | static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info, |
bda56f14 | 385 | unsigned int hdrlen) |
c3ff8cfe TG |
386 | { |
387 | size_t size = NLMSG_ALIGN(sizeof(struct ovs_header)) | |
bda56f14 | 388 | + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */ |
41af73e9 | 389 | + nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */ |
c3ff8cfe TG |
390 | |
391 | /* OVS_PACKET_ATTR_USERDATA */ | |
8f0aad6f WZ |
392 | if (upcall_info->userdata) |
393 | size += NLA_ALIGN(upcall_info->userdata->nla_len); | |
394 | ||
395 | /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */ | |
396 | if (upcall_info->egress_tun_info) | |
397 | size += nla_total_size(ovs_tun_key_attr_size()); | |
c3ff8cfe TG |
398 | |
399 | return size; | |
400 | } | |
401 | ||
8055a89c | 402 | static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb, |
e8eedb85 | 403 | const struct sw_flow_key *key, |
ccb1352e JG |
404 | const struct dp_upcall_info *upcall_info) |
405 | { | |
406 | struct ovs_header *upcall; | |
407 | struct sk_buff *nskb = NULL; | |
4ee45ea0 | 408 | struct sk_buff *user_skb = NULL; /* to be queued to userspace */ |
ccb1352e | 409 | struct nlattr *nla; |
795449d8 | 410 | struct genl_info info = { |
8055a89c | 411 | .dst_sk = ovs_dp_get_net(dp)->genl_sock, |
795449d8 TG |
412 | .snd_portid = upcall_info->portid, |
413 | }; | |
414 | size_t len; | |
bda56f14 | 415 | unsigned int hlen; |
8055a89c TG |
416 | int err, dp_ifindex; |
417 | ||
418 | dp_ifindex = get_dpifindex(dp); | |
419 | if (!dp_ifindex) | |
420 | return -ENODEV; | |
ccb1352e JG |
421 | |
422 | if (vlan_tx_tag_present(skb)) { | |
423 | nskb = skb_clone(skb, GFP_ATOMIC); | |
424 | if (!nskb) | |
425 | return -ENOMEM; | |
426 | ||
5968250c | 427 | nskb = __vlan_hwaccel_push_inside(nskb); |
8aa51d64 | 428 | if (!nskb) |
ccb1352e JG |
429 | return -ENOMEM; |
430 | ||
ccb1352e JG |
431 | skb = nskb; |
432 | } | |
433 | ||
434 | if (nla_attr_size(skb->len) > USHRT_MAX) { | |
435 | err = -EFBIG; | |
436 | goto out; | |
437 | } | |
438 | ||
bda56f14 TG |
439 | /* Complete checksum if needed */ |
440 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
441 | (err = skb_checksum_help(skb))) | |
442 | goto out; | |
443 | ||
444 | /* Older versions of OVS user space enforce alignment of the last | |
445 | * Netlink attribute to NLA_ALIGNTO which would require extensive | |
446 | * padding logic. Only perform zerocopy if padding is not required. | |
447 | */ | |
448 | if (dp->user_features & OVS_DP_F_UNALIGNED) | |
449 | hlen = skb_zerocopy_headlen(skb); | |
450 | else | |
451 | hlen = skb->len; | |
452 | ||
8f0aad6f | 453 | len = upcall_msg_size(upcall_info, hlen); |
795449d8 | 454 | user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC); |
ccb1352e JG |
455 | if (!user_skb) { |
456 | err = -ENOMEM; | |
457 | goto out; | |
458 | } | |
459 | ||
460 | upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, | |
461 | 0, upcall_info->cmd); | |
462 | upcall->dp_ifindex = dp_ifindex; | |
463 | ||
464 | nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY); | |
e8eedb85 | 465 | err = ovs_nla_put_flow(key, key, user_skb); |
f53e3831 | 466 | BUG_ON(err); |
ccb1352e JG |
467 | nla_nest_end(user_skb, nla); |
468 | ||
469 | if (upcall_info->userdata) | |
4490108b BP |
470 | __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA, |
471 | nla_len(upcall_info->userdata), | |
472 | nla_data(upcall_info->userdata)); | |
ccb1352e | 473 | |
8f0aad6f WZ |
474 | if (upcall_info->egress_tun_info) { |
475 | nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY); | |
476 | err = ovs_nla_put_egress_tunnel_key(user_skb, | |
477 | upcall_info->egress_tun_info); | |
478 | BUG_ON(err); | |
479 | nla_nest_end(user_skb, nla); | |
480 | } | |
481 | ||
bda56f14 TG |
482 | /* Only reserve room for attribute header, packet data is added |
483 | * in skb_zerocopy() */ | |
484 | if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) { | |
485 | err = -ENOBUFS; | |
486 | goto out; | |
487 | } | |
488 | nla->nla_len = nla_attr_size(skb->len); | |
ccb1352e | 489 | |
36d5fe6a ZK |
490 | err = skb_zerocopy(user_skb, skb, skb->len, hlen); |
491 | if (err) | |
492 | goto out; | |
ccb1352e | 493 | |
aea0bb4f TG |
494 | /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */ |
495 | if (!(dp->user_features & OVS_DP_F_UNALIGNED)) { | |
496 | size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len; | |
497 | ||
498 | if (plen > 0) | |
499 | memset(skb_put(user_skb, plen), 0, plen); | |
500 | } | |
501 | ||
bda56f14 | 502 | ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len; |
ccb1352e | 503 | |
bda56f14 | 504 | err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid); |
4ee45ea0 | 505 | user_skb = NULL; |
ccb1352e | 506 | out: |
36d5fe6a ZK |
507 | if (err) |
508 | skb_tx_error(skb); | |
4ee45ea0 | 509 | kfree_skb(user_skb); |
ccb1352e JG |
510 | kfree_skb(nskb); |
511 | return err; | |
512 | } | |
513 | ||
ccb1352e JG |
514 | static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info) |
515 | { | |
516 | struct ovs_header *ovs_header = info->userhdr; | |
517 | struct nlattr **a = info->attrs; | |
518 | struct sw_flow_actions *acts; | |
519 | struct sk_buff *packet; | |
520 | struct sw_flow *flow; | |
d98612b8 | 521 | struct sw_flow_actions *sf_acts; |
ccb1352e JG |
522 | struct datapath *dp; |
523 | struct ethhdr *eth; | |
83c8df26 | 524 | struct vport *input_vport; |
ccb1352e JG |
525 | int len; |
526 | int err; | |
05da5898 | 527 | bool log = !a[OVS_FLOW_ATTR_PROBE]; |
ccb1352e JG |
528 | |
529 | err = -EINVAL; | |
530 | if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] || | |
dded45fc | 531 | !a[OVS_PACKET_ATTR_ACTIONS]) |
ccb1352e JG |
532 | goto err; |
533 | ||
534 | len = nla_len(a[OVS_PACKET_ATTR_PACKET]); | |
535 | packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL); | |
536 | err = -ENOMEM; | |
537 | if (!packet) | |
538 | goto err; | |
539 | skb_reserve(packet, NET_IP_ALIGN); | |
540 | ||
32686a9d | 541 | nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len); |
ccb1352e JG |
542 | |
543 | skb_reset_mac_header(packet); | |
544 | eth = eth_hdr(packet); | |
545 | ||
546 | /* Normally, setting the skb 'protocol' field would be handled by a | |
547 | * call to eth_type_trans(), but it assumes there's a sending | |
548 | * device, which we may not have. */ | |
e5c5d22e | 549 | if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN) |
ccb1352e JG |
550 | packet->protocol = eth->h_proto; |
551 | else | |
552 | packet->protocol = htons(ETH_P_802_2); | |
553 | ||
554 | /* Build an sw_flow for sending this packet. */ | |
23dabf88 | 555 | flow = ovs_flow_alloc(); |
ccb1352e JG |
556 | err = PTR_ERR(flow); |
557 | if (IS_ERR(flow)) | |
558 | goto err_kfree_skb; | |
559 | ||
83c8df26 | 560 | err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet, |
05da5898 | 561 | &flow->key, log); |
ccb1352e JG |
562 | if (err) |
563 | goto err_flow_free; | |
564 | ||
e6445719 | 565 | err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], |
05da5898 | 566 | &flow->key, &acts, log); |
74f84a57 PS |
567 | if (err) |
568 | goto err_flow_free; | |
ccb1352e | 569 | |
f5796684 | 570 | rcu_assign_pointer(flow->sf_acts, acts); |
f5796684 | 571 | OVS_CB(packet)->egress_tun_info = NULL; |
ccb1352e | 572 | packet->priority = flow->key.phy.priority; |
39c7caeb | 573 | packet->mark = flow->key.phy.skb_mark; |
ccb1352e JG |
574 | |
575 | rcu_read_lock(); | |
cc3a5ae6 | 576 | dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex); |
ccb1352e JG |
577 | err = -ENODEV; |
578 | if (!dp) | |
579 | goto err_unlock; | |
580 | ||
83c8df26 PS |
581 | input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port); |
582 | if (!input_vport) | |
583 | input_vport = ovs_vport_rcu(dp, OVSP_LOCAL); | |
584 | ||
585 | if (!input_vport) | |
586 | goto err_unlock; | |
587 | ||
588 | OVS_CB(packet)->input_vport = input_vport; | |
d98612b8 | 589 | sf_acts = rcu_dereference(flow->sf_acts); |
83c8df26 | 590 | |
ccb1352e | 591 | local_bh_disable(); |
d98612b8 | 592 | err = ovs_execute_actions(dp, packet, sf_acts, &flow->key); |
ccb1352e JG |
593 | local_bh_enable(); |
594 | rcu_read_unlock(); | |
595 | ||
03f0d916 | 596 | ovs_flow_free(flow, false); |
ccb1352e JG |
597 | return err; |
598 | ||
599 | err_unlock: | |
600 | rcu_read_unlock(); | |
601 | err_flow_free: | |
03f0d916 | 602 | ovs_flow_free(flow, false); |
ccb1352e JG |
603 | err_kfree_skb: |
604 | kfree_skb(packet); | |
605 | err: | |
606 | return err; | |
607 | } | |
608 | ||
609 | static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = { | |
dded45fc | 610 | [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN }, |
ccb1352e JG |
611 | [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED }, |
612 | [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED }, | |
613 | }; | |
614 | ||
4534de83 | 615 | static const struct genl_ops dp_packet_genl_ops[] = { |
ccb1352e JG |
616 | { .cmd = OVS_PACKET_CMD_EXECUTE, |
617 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
618 | .policy = packet_policy, | |
619 | .doit = ovs_packet_cmd_execute | |
620 | } | |
621 | }; | |
622 | ||
0c200ef9 PS |
623 | static struct genl_family dp_packet_genl_family = { |
624 | .id = GENL_ID_GENERATE, | |
625 | .hdrsize = sizeof(struct ovs_header), | |
626 | .name = OVS_PACKET_FAMILY, | |
627 | .version = OVS_PACKET_VERSION, | |
628 | .maxattr = OVS_PACKET_ATTR_MAX, | |
629 | .netnsok = true, | |
630 | .parallel_ops = true, | |
631 | .ops = dp_packet_genl_ops, | |
632 | .n_ops = ARRAY_SIZE(dp_packet_genl_ops), | |
633 | }; | |
634 | ||
12eb18f7 | 635 | static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats, |
1bd7116f | 636 | struct ovs_dp_megaflow_stats *mega_stats) |
ccb1352e JG |
637 | { |
638 | int i; | |
ccb1352e | 639 | |
1bd7116f AZ |
640 | memset(mega_stats, 0, sizeof(*mega_stats)); |
641 | ||
b637e498 | 642 | stats->n_flows = ovs_flow_tbl_count(&dp->table); |
1bd7116f | 643 | mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table); |
ccb1352e JG |
644 | |
645 | stats->n_hit = stats->n_missed = stats->n_lost = 0; | |
1bd7116f | 646 | |
ccb1352e JG |
647 | for_each_possible_cpu(i) { |
648 | const struct dp_stats_percpu *percpu_stats; | |
649 | struct dp_stats_percpu local_stats; | |
650 | unsigned int start; | |
651 | ||
652 | percpu_stats = per_cpu_ptr(dp->stats_percpu, i); | |
653 | ||
654 | do { | |
57a7744e | 655 | start = u64_stats_fetch_begin_irq(&percpu_stats->syncp); |
ccb1352e | 656 | local_stats = *percpu_stats; |
57a7744e | 657 | } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start)); |
ccb1352e JG |
658 | |
659 | stats->n_hit += local_stats.n_hit; | |
660 | stats->n_missed += local_stats.n_missed; | |
661 | stats->n_lost += local_stats.n_lost; | |
1bd7116f | 662 | mega_stats->n_mask_hit += local_stats.n_mask_hit; |
ccb1352e JG |
663 | } |
664 | } | |
665 | ||
c3ff8cfe TG |
666 | static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts) |
667 | { | |
668 | return NLMSG_ALIGN(sizeof(struct ovs_header)) | |
41af73e9 JS |
669 | + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_KEY */ |
670 | + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_MASK */ | |
c3ff8cfe TG |
671 | + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */ |
672 | + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */ | |
673 | + nla_total_size(8) /* OVS_FLOW_ATTR_USED */ | |
674 | + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */ | |
675 | } | |
676 | ||
bb6f9a70 | 677 | /* Called with ovs_mutex or RCU read lock. */ |
ca7105f2 JS |
678 | static int ovs_flow_cmd_fill_match(const struct sw_flow *flow, |
679 | struct sk_buff *skb) | |
ccb1352e | 680 | { |
ccb1352e | 681 | struct nlattr *nla; |
ccb1352e JG |
682 | int err; |
683 | ||
03f0d916 | 684 | /* Fill flow key. */ |
ccb1352e JG |
685 | nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY); |
686 | if (!nla) | |
ca7105f2 | 687 | return -EMSGSIZE; |
03f0d916 | 688 | |
e6445719 | 689 | err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb); |
03f0d916 | 690 | if (err) |
ca7105f2 JS |
691 | return err; |
692 | ||
03f0d916 AZ |
693 | nla_nest_end(skb, nla); |
694 | ||
ca7105f2 | 695 | /* Fill flow mask. */ |
03f0d916 AZ |
696 | nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK); |
697 | if (!nla) | |
ca7105f2 | 698 | return -EMSGSIZE; |
03f0d916 | 699 | |
e6445719 | 700 | err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb); |
ccb1352e | 701 | if (err) |
ca7105f2 | 702 | return err; |
03f0d916 | 703 | |
ccb1352e | 704 | nla_nest_end(skb, nla); |
ca7105f2 JS |
705 | return 0; |
706 | } | |
707 | ||
708 | /* Called with ovs_mutex or RCU read lock. */ | |
709 | static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow, | |
710 | struct sk_buff *skb) | |
711 | { | |
712 | struct ovs_flow_stats stats; | |
713 | __be16 tcp_flags; | |
714 | unsigned long used; | |
ccb1352e | 715 | |
e298e505 | 716 | ovs_flow_stats_get(flow, &stats, &used, &tcp_flags); |
0e9796b4 | 717 | |
028d6a67 DM |
718 | if (used && |
719 | nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used))) | |
ca7105f2 | 720 | return -EMSGSIZE; |
ccb1352e | 721 | |
028d6a67 | 722 | if (stats.n_packets && |
e298e505 | 723 | nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats)) |
ca7105f2 | 724 | return -EMSGSIZE; |
ccb1352e | 725 | |
e298e505 PS |
726 | if ((u8)ntohs(tcp_flags) && |
727 | nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags))) | |
ca7105f2 JS |
728 | return -EMSGSIZE; |
729 | ||
730 | return 0; | |
731 | } | |
732 | ||
733 | /* Called with ovs_mutex or RCU read lock. */ | |
734 | static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow, | |
735 | struct sk_buff *skb, int skb_orig_len) | |
736 | { | |
737 | struct nlattr *start; | |
738 | int err; | |
ccb1352e JG |
739 | |
740 | /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if | |
741 | * this is the first flow to be dumped into 'skb'. This is unusual for | |
742 | * Netlink but individual action lists can be longer than | |
743 | * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this. | |
744 | * The userspace caller can always fetch the actions separately if it | |
745 | * really wants them. (Most userspace callers in fact don't care.) | |
746 | * | |
747 | * This can only fail for dump operations because the skb is always | |
748 | * properly sized for single flows. | |
749 | */ | |
74f84a57 PS |
750 | start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS); |
751 | if (start) { | |
d57170b1 PS |
752 | const struct sw_flow_actions *sf_acts; |
753 | ||
663efa36 | 754 | sf_acts = rcu_dereference_ovsl(flow->sf_acts); |
e6445719 PS |
755 | err = ovs_nla_put_actions(sf_acts->actions, |
756 | sf_acts->actions_len, skb); | |
0e9796b4 | 757 | |
74f84a57 PS |
758 | if (!err) |
759 | nla_nest_end(skb, start); | |
760 | else { | |
761 | if (skb_orig_len) | |
ca7105f2 | 762 | return err; |
74f84a57 PS |
763 | |
764 | nla_nest_cancel(skb, start); | |
765 | } | |
ca7105f2 JS |
766 | } else if (skb_orig_len) { |
767 | return -EMSGSIZE; | |
768 | } | |
769 | ||
770 | return 0; | |
771 | } | |
772 | ||
773 | /* Called with ovs_mutex or RCU read lock. */ | |
774 | static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex, | |
775 | struct sk_buff *skb, u32 portid, | |
776 | u32 seq, u32 flags, u8 cmd) | |
777 | { | |
778 | const int skb_orig_len = skb->len; | |
779 | struct ovs_header *ovs_header; | |
780 | int err; | |
781 | ||
782 | ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, | |
783 | flags, cmd); | |
784 | if (!ovs_header) | |
785 | return -EMSGSIZE; | |
786 | ||
787 | ovs_header->dp_ifindex = dp_ifindex; | |
788 | ||
789 | err = ovs_flow_cmd_fill_match(flow, skb); | |
790 | if (err) | |
791 | goto error; | |
792 | ||
793 | err = ovs_flow_cmd_fill_stats(flow, skb); | |
794 | if (err) | |
795 | goto error; | |
796 | ||
797 | err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len); | |
798 | if (err) | |
799 | goto error; | |
ccb1352e JG |
800 | |
801 | return genlmsg_end(skb, ovs_header); | |
802 | ||
ccb1352e JG |
803 | error: |
804 | genlmsg_cancel(skb, ovs_header); | |
805 | return err; | |
806 | } | |
807 | ||
0e9796b4 JR |
808 | /* May not be called with RCU read lock. */ |
809 | static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts, | |
fb5d1e9e JR |
810 | struct genl_info *info, |
811 | bool always) | |
ccb1352e | 812 | { |
fb5d1e9e | 813 | struct sk_buff *skb; |
ccb1352e | 814 | |
9b67aa4a | 815 | if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0)) |
fb5d1e9e JR |
816 | return NULL; |
817 | ||
0e9796b4 | 818 | skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL); |
fb5d1e9e JR |
819 | if (!skb) |
820 | return ERR_PTR(-ENOMEM); | |
821 | ||
822 | return skb; | |
ccb1352e JG |
823 | } |
824 | ||
0e9796b4 JR |
825 | /* Called with ovs_mutex. */ |
826 | static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow, | |
827 | int dp_ifindex, | |
828 | struct genl_info *info, u8 cmd, | |
829 | bool always) | |
ccb1352e JG |
830 | { |
831 | struct sk_buff *skb; | |
832 | int retval; | |
833 | ||
0e9796b4 JR |
834 | skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info, |
835 | always); | |
d0e992aa | 836 | if (IS_ERR_OR_NULL(skb)) |
fb5d1e9e | 837 | return skb; |
ccb1352e | 838 | |
0e9796b4 JR |
839 | retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb, |
840 | info->snd_portid, info->snd_seq, 0, | |
841 | cmd); | |
ccb1352e JG |
842 | BUG_ON(retval < 0); |
843 | return skb; | |
844 | } | |
845 | ||
37bdc87b | 846 | static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info) |
ccb1352e JG |
847 | { |
848 | struct nlattr **a = info->attrs; | |
849 | struct ovs_header *ovs_header = info->userhdr; | |
893f139b | 850 | struct sw_flow *flow, *new_flow; |
03f0d916 | 851 | struct sw_flow_mask mask; |
ccb1352e JG |
852 | struct sk_buff *reply; |
853 | struct datapath *dp; | |
37bdc87b | 854 | struct sw_flow_actions *acts; |
03f0d916 | 855 | struct sw_flow_match match; |
ccb1352e | 856 | int error; |
05da5898 | 857 | bool log = !a[OVS_FLOW_ATTR_PROBE]; |
ccb1352e | 858 | |
893f139b | 859 | /* Must have key and actions. */ |
ccb1352e | 860 | error = -EINVAL; |
426cda5c | 861 | if (!a[OVS_FLOW_ATTR_KEY]) { |
05da5898 | 862 | OVS_NLERR(log, "Flow key attr not present in new flow."); |
ccb1352e | 863 | goto error; |
426cda5c JG |
864 | } |
865 | if (!a[OVS_FLOW_ATTR_ACTIONS]) { | |
05da5898 | 866 | OVS_NLERR(log, "Flow actions attr not present in new flow."); |
893f139b | 867 | goto error; |
426cda5c | 868 | } |
03f0d916 | 869 | |
893f139b JR |
870 | /* Most of the time we need to allocate a new flow, do it before |
871 | * locking. | |
872 | */ | |
873 | new_flow = ovs_flow_alloc(); | |
874 | if (IS_ERR(new_flow)) { | |
875 | error = PTR_ERR(new_flow); | |
876 | goto error; | |
877 | } | |
878 | ||
879 | /* Extract key. */ | |
880 | ovs_match_init(&match, &new_flow->unmasked_key, &mask); | |
05da5898 JR |
881 | error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], |
882 | a[OVS_FLOW_ATTR_MASK], log); | |
ccb1352e | 883 | if (error) |
893f139b | 884 | goto err_kfree_flow; |
ccb1352e | 885 | |
893f139b | 886 | ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask); |
74f84a57 | 887 | |
893f139b | 888 | /* Validate actions. */ |
893f139b | 889 | error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key, |
05da5898 | 890 | &acts, log); |
37bdc87b | 891 | if (error) { |
05da5898 | 892 | OVS_NLERR(log, "Flow actions may not be safe on all matching packets."); |
2fdb957d | 893 | goto err_kfree_flow; |
893f139b JR |
894 | } |
895 | ||
896 | reply = ovs_flow_cmd_alloc_info(acts, info, false); | |
897 | if (IS_ERR(reply)) { | |
898 | error = PTR_ERR(reply); | |
899 | goto err_kfree_acts; | |
ccb1352e JG |
900 | } |
901 | ||
8e4e1713 | 902 | ovs_lock(); |
46df7b81 | 903 | dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); |
893f139b JR |
904 | if (unlikely(!dp)) { |
905 | error = -ENODEV; | |
8e4e1713 | 906 | goto err_unlock_ovs; |
893f139b | 907 | } |
03f0d916 | 908 | /* Check if this is a duplicate flow */ |
893f139b JR |
909 | flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key); |
910 | if (likely(!flow)) { | |
911 | rcu_assign_pointer(new_flow->sf_acts, acts); | |
ccb1352e JG |
912 | |
913 | /* Put flow in bucket. */ | |
893f139b JR |
914 | error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask); |
915 | if (unlikely(error)) { | |
618ed0c8 | 916 | acts = NULL; |
893f139b JR |
917 | goto err_unlock_ovs; |
918 | } | |
919 | ||
920 | if (unlikely(reply)) { | |
921 | error = ovs_flow_cmd_fill_info(new_flow, | |
922 | ovs_header->dp_ifindex, | |
923 | reply, info->snd_portid, | |
924 | info->snd_seq, 0, | |
925 | OVS_FLOW_CMD_NEW); | |
926 | BUG_ON(error < 0); | |
618ed0c8 | 927 | } |
893f139b | 928 | ovs_unlock(); |
ccb1352e | 929 | } else { |
37bdc87b JR |
930 | struct sw_flow_actions *old_acts; |
931 | ||
ccb1352e JG |
932 | /* Bail out if we're not allowed to modify an existing flow. |
933 | * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL | |
934 | * because Generic Netlink treats the latter as a dump | |
935 | * request. We also accept NLM_F_EXCL in case that bug ever | |
936 | * gets fixed. | |
937 | */ | |
893f139b JR |
938 | if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE |
939 | | NLM_F_EXCL))) { | |
940 | error = -EEXIST; | |
8e4e1713 | 941 | goto err_unlock_ovs; |
893f139b | 942 | } |
03f0d916 | 943 | /* The unmasked key has to be the same for flow updates. */ |
893f139b | 944 | if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) { |
05da5898 | 945 | /* Look for any overlapping flow. */ |
4a46b24e AW |
946 | flow = ovs_flow_tbl_lookup_exact(&dp->table, &match); |
947 | if (!flow) { | |
948 | error = -ENOENT; | |
949 | goto err_unlock_ovs; | |
950 | } | |
893f139b | 951 | } |
37bdc87b JR |
952 | /* Update actions. */ |
953 | old_acts = ovsl_dereference(flow->sf_acts); | |
954 | rcu_assign_pointer(flow->sf_acts, acts); | |
37bdc87b | 955 | |
893f139b JR |
956 | if (unlikely(reply)) { |
957 | error = ovs_flow_cmd_fill_info(flow, | |
958 | ovs_header->dp_ifindex, | |
959 | reply, info->snd_portid, | |
960 | info->snd_seq, 0, | |
961 | OVS_FLOW_CMD_NEW); | |
962 | BUG_ON(error < 0); | |
963 | } | |
964 | ovs_unlock(); | |
37bdc87b | 965 | |
893f139b JR |
966 | ovs_nla_free_flow_actions(old_acts); |
967 | ovs_flow_free(new_flow, false); | |
37bdc87b | 968 | } |
893f139b JR |
969 | |
970 | if (reply) | |
971 | ovs_notify(&dp_flow_genl_family, reply, info); | |
37bdc87b JR |
972 | return 0; |
973 | ||
37bdc87b JR |
974 | err_unlock_ovs: |
975 | ovs_unlock(); | |
893f139b JR |
976 | kfree_skb(reply); |
977 | err_kfree_acts: | |
37bdc87b | 978 | kfree(acts); |
893f139b JR |
979 | err_kfree_flow: |
980 | ovs_flow_free(new_flow, false); | |
37bdc87b JR |
981 | error: |
982 | return error; | |
983 | } | |
ccb1352e | 984 | |
2fdb957d | 985 | /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */ |
6b205b2c JG |
986 | static struct sw_flow_actions *get_flow_actions(const struct nlattr *a, |
987 | const struct sw_flow_key *key, | |
05da5898 JR |
988 | const struct sw_flow_mask *mask, |
989 | bool log) | |
6b205b2c JG |
990 | { |
991 | struct sw_flow_actions *acts; | |
992 | struct sw_flow_key masked_key; | |
993 | int error; | |
994 | ||
6b205b2c | 995 | ovs_flow_mask_key(&masked_key, key, mask); |
05da5898 | 996 | error = ovs_nla_copy_actions(a, &masked_key, &acts, log); |
6b205b2c | 997 | if (error) { |
05da5898 JR |
998 | OVS_NLERR(log, |
999 | "Actions may not be safe on all matching packets"); | |
6b205b2c JG |
1000 | return ERR_PTR(error); |
1001 | } | |
1002 | ||
1003 | return acts; | |
1004 | } | |
1005 | ||
37bdc87b JR |
1006 | static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info) |
1007 | { | |
1008 | struct nlattr **a = info->attrs; | |
1009 | struct ovs_header *ovs_header = info->userhdr; | |
6b205b2c | 1010 | struct sw_flow_key key; |
37bdc87b JR |
1011 | struct sw_flow *flow; |
1012 | struct sw_flow_mask mask; | |
1013 | struct sk_buff *reply = NULL; | |
1014 | struct datapath *dp; | |
893f139b | 1015 | struct sw_flow_actions *old_acts = NULL, *acts = NULL; |
37bdc87b JR |
1016 | struct sw_flow_match match; |
1017 | int error; | |
05da5898 | 1018 | bool log = !a[OVS_FLOW_ATTR_PROBE]; |
37bdc87b JR |
1019 | |
1020 | /* Extract key. */ | |
1021 | error = -EINVAL; | |
426cda5c | 1022 | if (!a[OVS_FLOW_ATTR_KEY]) { |
05da5898 | 1023 | OVS_NLERR(log, "Flow key attribute not present in set flow."); |
37bdc87b | 1024 | goto error; |
426cda5c | 1025 | } |
37bdc87b JR |
1026 | |
1027 | ovs_match_init(&match, &key, &mask); | |
05da5898 JR |
1028 | error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], |
1029 | a[OVS_FLOW_ATTR_MASK], log); | |
37bdc87b JR |
1030 | if (error) |
1031 | goto error; | |
1032 | ||
1033 | /* Validate actions. */ | |
1034 | if (a[OVS_FLOW_ATTR_ACTIONS]) { | |
05da5898 JR |
1035 | acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask, |
1036 | log); | |
6b205b2c JG |
1037 | if (IS_ERR(acts)) { |
1038 | error = PTR_ERR(acts); | |
37bdc87b | 1039 | goto error; |
893f139b | 1040 | } |
893f139b | 1041 | |
2fdb957d | 1042 | /* Can allocate before locking if have acts. */ |
893f139b JR |
1043 | reply = ovs_flow_cmd_alloc_info(acts, info, false); |
1044 | if (IS_ERR(reply)) { | |
1045 | error = PTR_ERR(reply); | |
1046 | goto err_kfree_acts; | |
be52c9e9 | 1047 | } |
37bdc87b | 1048 | } |
0e9796b4 | 1049 | |
37bdc87b JR |
1050 | ovs_lock(); |
1051 | dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); | |
893f139b JR |
1052 | if (unlikely(!dp)) { |
1053 | error = -ENODEV; | |
37bdc87b | 1054 | goto err_unlock_ovs; |
893f139b | 1055 | } |
37bdc87b | 1056 | /* Check that the flow exists. */ |
4a46b24e | 1057 | flow = ovs_flow_tbl_lookup_exact(&dp->table, &match); |
893f139b JR |
1058 | if (unlikely(!flow)) { |
1059 | error = -ENOENT; | |
37bdc87b | 1060 | goto err_unlock_ovs; |
893f139b | 1061 | } |
4a46b24e | 1062 | |
37bdc87b | 1063 | /* Update actions, if present. */ |
893f139b | 1064 | if (likely(acts)) { |
37bdc87b JR |
1065 | old_acts = ovsl_dereference(flow->sf_acts); |
1066 | rcu_assign_pointer(flow->sf_acts, acts); | |
893f139b JR |
1067 | |
1068 | if (unlikely(reply)) { | |
1069 | error = ovs_flow_cmd_fill_info(flow, | |
1070 | ovs_header->dp_ifindex, | |
1071 | reply, info->snd_portid, | |
1072 | info->snd_seq, 0, | |
1073 | OVS_FLOW_CMD_NEW); | |
1074 | BUG_ON(error < 0); | |
1075 | } | |
1076 | } else { | |
1077 | /* Could not alloc without acts before locking. */ | |
1078 | reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, | |
1079 | info, OVS_FLOW_CMD_NEW, false); | |
1080 | if (unlikely(IS_ERR(reply))) { | |
1081 | error = PTR_ERR(reply); | |
1082 | goto err_unlock_ovs; | |
1083 | } | |
ccb1352e | 1084 | } |
37bdc87b | 1085 | |
37bdc87b JR |
1086 | /* Clear stats. */ |
1087 | if (a[OVS_FLOW_ATTR_CLEAR]) | |
1088 | ovs_flow_stats_clear(flow); | |
8e4e1713 | 1089 | ovs_unlock(); |
ccb1352e | 1090 | |
893f139b JR |
1091 | if (reply) |
1092 | ovs_notify(&dp_flow_genl_family, reply, info); | |
1093 | if (old_acts) | |
1094 | ovs_nla_free_flow_actions(old_acts); | |
fb5d1e9e | 1095 | |
ccb1352e JG |
1096 | return 0; |
1097 | ||
8e4e1713 PS |
1098 | err_unlock_ovs: |
1099 | ovs_unlock(); | |
893f139b JR |
1100 | kfree_skb(reply); |
1101 | err_kfree_acts: | |
74f84a57 | 1102 | kfree(acts); |
ccb1352e JG |
1103 | error: |
1104 | return error; | |
1105 | } | |
1106 | ||
1107 | static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info) | |
1108 | { | |
1109 | struct nlattr **a = info->attrs; | |
1110 | struct ovs_header *ovs_header = info->userhdr; | |
1111 | struct sw_flow_key key; | |
1112 | struct sk_buff *reply; | |
1113 | struct sw_flow *flow; | |
1114 | struct datapath *dp; | |
03f0d916 | 1115 | struct sw_flow_match match; |
ccb1352e | 1116 | int err; |
05da5898 | 1117 | bool log = !a[OVS_FLOW_ATTR_PROBE]; |
ccb1352e | 1118 | |
03f0d916 | 1119 | if (!a[OVS_FLOW_ATTR_KEY]) { |
05da5898 JR |
1120 | OVS_NLERR(log, |
1121 | "Flow get message rejected, Key attribute missing."); | |
ccb1352e | 1122 | return -EINVAL; |
03f0d916 AZ |
1123 | } |
1124 | ||
1125 | ovs_match_init(&match, &key, NULL); | |
05da5898 | 1126 | err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL, log); |
ccb1352e JG |
1127 | if (err) |
1128 | return err; | |
1129 | ||
8e4e1713 | 1130 | ovs_lock(); |
46df7b81 | 1131 | dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); |
8e4e1713 PS |
1132 | if (!dp) { |
1133 | err = -ENODEV; | |
1134 | goto unlock; | |
1135 | } | |
ccb1352e | 1136 | |
4a46b24e AW |
1137 | flow = ovs_flow_tbl_lookup_exact(&dp->table, &match); |
1138 | if (!flow) { | |
8e4e1713 PS |
1139 | err = -ENOENT; |
1140 | goto unlock; | |
1141 | } | |
ccb1352e | 1142 | |
0e9796b4 JR |
1143 | reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info, |
1144 | OVS_FLOW_CMD_NEW, true); | |
8e4e1713 PS |
1145 | if (IS_ERR(reply)) { |
1146 | err = PTR_ERR(reply); | |
1147 | goto unlock; | |
1148 | } | |
ccb1352e | 1149 | |
8e4e1713 | 1150 | ovs_unlock(); |
ccb1352e | 1151 | return genlmsg_reply(reply, info); |
8e4e1713 PS |
1152 | unlock: |
1153 | ovs_unlock(); | |
1154 | return err; | |
ccb1352e JG |
1155 | } |
1156 | ||
1157 | static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info) | |
1158 | { | |
1159 | struct nlattr **a = info->attrs; | |
1160 | struct ovs_header *ovs_header = info->userhdr; | |
1161 | struct sw_flow_key key; | |
1162 | struct sk_buff *reply; | |
1163 | struct sw_flow *flow; | |
1164 | struct datapath *dp; | |
03f0d916 | 1165 | struct sw_flow_match match; |
ccb1352e | 1166 | int err; |
05da5898 | 1167 | bool log = !a[OVS_FLOW_ATTR_PROBE]; |
ccb1352e | 1168 | |
aed06778 JR |
1169 | if (likely(a[OVS_FLOW_ATTR_KEY])) { |
1170 | ovs_match_init(&match, &key, NULL); | |
05da5898 JR |
1171 | err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL, |
1172 | log); | |
aed06778 JR |
1173 | if (unlikely(err)) |
1174 | return err; | |
1175 | } | |
1176 | ||
8e4e1713 | 1177 | ovs_lock(); |
46df7b81 | 1178 | dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); |
aed06778 | 1179 | if (unlikely(!dp)) { |
8e4e1713 PS |
1180 | err = -ENODEV; |
1181 | goto unlock; | |
1182 | } | |
46df7b81 | 1183 | |
aed06778 | 1184 | if (unlikely(!a[OVS_FLOW_ATTR_KEY])) { |
b637e498 | 1185 | err = ovs_flow_tbl_flush(&dp->table); |
8e4e1713 PS |
1186 | goto unlock; |
1187 | } | |
03f0d916 | 1188 | |
4a46b24e AW |
1189 | flow = ovs_flow_tbl_lookup_exact(&dp->table, &match); |
1190 | if (unlikely(!flow)) { | |
8e4e1713 PS |
1191 | err = -ENOENT; |
1192 | goto unlock; | |
1193 | } | |
ccb1352e | 1194 | |
b637e498 | 1195 | ovs_flow_tbl_remove(&dp->table, flow); |
aed06778 | 1196 | ovs_unlock(); |
ccb1352e | 1197 | |
aed06778 JR |
1198 | reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts, |
1199 | info, false); | |
1200 | if (likely(reply)) { | |
1201 | if (likely(!IS_ERR(reply))) { | |
1202 | rcu_read_lock(); /*To keep RCU checker happy. */ | |
1203 | err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, | |
1204 | reply, info->snd_portid, | |
1205 | info->snd_seq, 0, | |
1206 | OVS_FLOW_CMD_DEL); | |
1207 | rcu_read_unlock(); | |
1208 | BUG_ON(err < 0); | |
1209 | ||
1210 | ovs_notify(&dp_flow_genl_family, reply, info); | |
1211 | } else { | |
1212 | netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply)); | |
1213 | } | |
fb5d1e9e | 1214 | } |
ccb1352e | 1215 | |
aed06778 | 1216 | ovs_flow_free(flow, true); |
ccb1352e | 1217 | return 0; |
8e4e1713 PS |
1218 | unlock: |
1219 | ovs_unlock(); | |
1220 | return err; | |
ccb1352e JG |
1221 | } |
1222 | ||
1223 | static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) | |
1224 | { | |
1225 | struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); | |
b637e498 | 1226 | struct table_instance *ti; |
ccb1352e | 1227 | struct datapath *dp; |
ccb1352e | 1228 | |
d57170b1 | 1229 | rcu_read_lock(); |
cc3a5ae6 | 1230 | dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex); |
8e4e1713 | 1231 | if (!dp) { |
d57170b1 | 1232 | rcu_read_unlock(); |
ccb1352e | 1233 | return -ENODEV; |
8e4e1713 | 1234 | } |
ccb1352e | 1235 | |
b637e498 | 1236 | ti = rcu_dereference(dp->table.ti); |
ccb1352e JG |
1237 | for (;;) { |
1238 | struct sw_flow *flow; | |
1239 | u32 bucket, obj; | |
1240 | ||
1241 | bucket = cb->args[0]; | |
1242 | obj = cb->args[1]; | |
b637e498 | 1243 | flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj); |
ccb1352e JG |
1244 | if (!flow) |
1245 | break; | |
1246 | ||
0e9796b4 | 1247 | if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb, |
15e47304 | 1248 | NETLINK_CB(cb->skb).portid, |
ccb1352e JG |
1249 | cb->nlh->nlmsg_seq, NLM_F_MULTI, |
1250 | OVS_FLOW_CMD_NEW) < 0) | |
1251 | break; | |
1252 | ||
1253 | cb->args[0] = bucket; | |
1254 | cb->args[1] = obj; | |
1255 | } | |
d57170b1 | 1256 | rcu_read_unlock(); |
ccb1352e JG |
1257 | return skb->len; |
1258 | } | |
1259 | ||
0c200ef9 PS |
1260 | static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = { |
1261 | [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED }, | |
05da5898 | 1262 | [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED }, |
0c200ef9 PS |
1263 | [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED }, |
1264 | [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG }, | |
05da5898 | 1265 | [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG }, |
0c200ef9 PS |
1266 | }; |
1267 | ||
48e48a70 | 1268 | static const struct genl_ops dp_flow_genl_ops[] = { |
ccb1352e JG |
1269 | { .cmd = OVS_FLOW_CMD_NEW, |
1270 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1271 | .policy = flow_policy, | |
37bdc87b | 1272 | .doit = ovs_flow_cmd_new |
ccb1352e JG |
1273 | }, |
1274 | { .cmd = OVS_FLOW_CMD_DEL, | |
1275 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1276 | .policy = flow_policy, | |
1277 | .doit = ovs_flow_cmd_del | |
1278 | }, | |
1279 | { .cmd = OVS_FLOW_CMD_GET, | |
1280 | .flags = 0, /* OK for unprivileged users. */ | |
1281 | .policy = flow_policy, | |
1282 | .doit = ovs_flow_cmd_get, | |
1283 | .dumpit = ovs_flow_cmd_dump | |
1284 | }, | |
1285 | { .cmd = OVS_FLOW_CMD_SET, | |
1286 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1287 | .policy = flow_policy, | |
37bdc87b | 1288 | .doit = ovs_flow_cmd_set, |
ccb1352e JG |
1289 | }, |
1290 | }; | |
1291 | ||
0c200ef9 | 1292 | static struct genl_family dp_flow_genl_family = { |
ccb1352e JG |
1293 | .id = GENL_ID_GENERATE, |
1294 | .hdrsize = sizeof(struct ovs_header), | |
0c200ef9 PS |
1295 | .name = OVS_FLOW_FAMILY, |
1296 | .version = OVS_FLOW_VERSION, | |
1297 | .maxattr = OVS_FLOW_ATTR_MAX, | |
3a4e0d6a PS |
1298 | .netnsok = true, |
1299 | .parallel_ops = true, | |
0c200ef9 PS |
1300 | .ops = dp_flow_genl_ops, |
1301 | .n_ops = ARRAY_SIZE(dp_flow_genl_ops), | |
1302 | .mcgrps = &ovs_dp_flow_multicast_group, | |
1303 | .n_mcgrps = 1, | |
ccb1352e JG |
1304 | }; |
1305 | ||
c3ff8cfe TG |
1306 | static size_t ovs_dp_cmd_msg_size(void) |
1307 | { | |
1308 | size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header)); | |
1309 | ||
1310 | msgsize += nla_total_size(IFNAMSIZ); | |
1311 | msgsize += nla_total_size(sizeof(struct ovs_dp_stats)); | |
1bd7116f | 1312 | msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats)); |
45fb9c35 | 1313 | msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */ |
c3ff8cfe TG |
1314 | |
1315 | return msgsize; | |
1316 | } | |
1317 | ||
8ec609d8 | 1318 | /* Called with ovs_mutex. */ |
ccb1352e | 1319 | static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb, |
15e47304 | 1320 | u32 portid, u32 seq, u32 flags, u8 cmd) |
ccb1352e JG |
1321 | { |
1322 | struct ovs_header *ovs_header; | |
1323 | struct ovs_dp_stats dp_stats; | |
1bd7116f | 1324 | struct ovs_dp_megaflow_stats dp_megaflow_stats; |
ccb1352e JG |
1325 | int err; |
1326 | ||
15e47304 | 1327 | ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family, |
ccb1352e JG |
1328 | flags, cmd); |
1329 | if (!ovs_header) | |
1330 | goto error; | |
1331 | ||
1332 | ovs_header->dp_ifindex = get_dpifindex(dp); | |
1333 | ||
ccb1352e | 1334 | err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp)); |
ccb1352e JG |
1335 | if (err) |
1336 | goto nla_put_failure; | |
1337 | ||
1bd7116f AZ |
1338 | get_dp_stats(dp, &dp_stats, &dp_megaflow_stats); |
1339 | if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), | |
1340 | &dp_stats)) | |
1341 | goto nla_put_failure; | |
1342 | ||
1343 | if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS, | |
1344 | sizeof(struct ovs_dp_megaflow_stats), | |
1345 | &dp_megaflow_stats)) | |
028d6a67 | 1346 | goto nla_put_failure; |
ccb1352e | 1347 | |
43d4be9c TG |
1348 | if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features)) |
1349 | goto nla_put_failure; | |
1350 | ||
ccb1352e JG |
1351 | return genlmsg_end(skb, ovs_header); |
1352 | ||
1353 | nla_put_failure: | |
1354 | genlmsg_cancel(skb, ovs_header); | |
1355 | error: | |
1356 | return -EMSGSIZE; | |
1357 | } | |
1358 | ||
6093ae9a | 1359 | static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info) |
ccb1352e | 1360 | { |
6093ae9a | 1361 | return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL); |
ccb1352e JG |
1362 | } |
1363 | ||
bb6f9a70 | 1364 | /* Called with rcu_read_lock or ovs_mutex. */ |
46df7b81 | 1365 | static struct datapath *lookup_datapath(struct net *net, |
12eb18f7 | 1366 | const struct ovs_header *ovs_header, |
ccb1352e JG |
1367 | struct nlattr *a[OVS_DP_ATTR_MAX + 1]) |
1368 | { | |
1369 | struct datapath *dp; | |
1370 | ||
1371 | if (!a[OVS_DP_ATTR_NAME]) | |
46df7b81 | 1372 | dp = get_dp(net, ovs_header->dp_ifindex); |
ccb1352e JG |
1373 | else { |
1374 | struct vport *vport; | |
1375 | ||
46df7b81 | 1376 | vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME])); |
ccb1352e | 1377 | dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL; |
ccb1352e JG |
1378 | } |
1379 | return dp ? dp : ERR_PTR(-ENODEV); | |
1380 | } | |
1381 | ||
44da5ae5 TG |
1382 | static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info) |
1383 | { | |
1384 | struct datapath *dp; | |
1385 | ||
1386 | dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); | |
3c7eacfc | 1387 | if (IS_ERR(dp)) |
44da5ae5 TG |
1388 | return; |
1389 | ||
1390 | WARN(dp->user_features, "Dropping previously announced user features\n"); | |
1391 | dp->user_features = 0; | |
1392 | } | |
1393 | ||
12eb18f7 | 1394 | static void ovs_dp_change(struct datapath *dp, struct nlattr *a[]) |
43d4be9c TG |
1395 | { |
1396 | if (a[OVS_DP_ATTR_USER_FEATURES]) | |
1397 | dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]); | |
1398 | } | |
1399 | ||
ccb1352e JG |
1400 | static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info) |
1401 | { | |
1402 | struct nlattr **a = info->attrs; | |
1403 | struct vport_parms parms; | |
1404 | struct sk_buff *reply; | |
1405 | struct datapath *dp; | |
1406 | struct vport *vport; | |
46df7b81 | 1407 | struct ovs_net *ovs_net; |
15eac2a7 | 1408 | int err, i; |
ccb1352e JG |
1409 | |
1410 | err = -EINVAL; | |
1411 | if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID]) | |
1412 | goto err; | |
1413 | ||
6093ae9a JR |
1414 | reply = ovs_dp_cmd_alloc_info(info); |
1415 | if (!reply) | |
1416 | return -ENOMEM; | |
ccb1352e JG |
1417 | |
1418 | err = -ENOMEM; | |
1419 | dp = kzalloc(sizeof(*dp), GFP_KERNEL); | |
1420 | if (dp == NULL) | |
6093ae9a | 1421 | goto err_free_reply; |
46df7b81 | 1422 | |
46df7b81 | 1423 | ovs_dp_set_net(dp, hold_net(sock_net(skb->sk))); |
ccb1352e JG |
1424 | |
1425 | /* Allocate table. */ | |
b637e498 PS |
1426 | err = ovs_flow_tbl_init(&dp->table); |
1427 | if (err) | |
ccb1352e JG |
1428 | goto err_free_dp; |
1429 | ||
1c213bd2 | 1430 | dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu); |
ccb1352e JG |
1431 | if (!dp->stats_percpu) { |
1432 | err = -ENOMEM; | |
1433 | goto err_destroy_table; | |
1434 | } | |
1435 | ||
15eac2a7 | 1436 | dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head), |
e6445719 | 1437 | GFP_KERNEL); |
15eac2a7 PS |
1438 | if (!dp->ports) { |
1439 | err = -ENOMEM; | |
1440 | goto err_destroy_percpu; | |
1441 | } | |
1442 | ||
1443 | for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) | |
1444 | INIT_HLIST_HEAD(&dp->ports[i]); | |
1445 | ||
ccb1352e JG |
1446 | /* Set up our datapath device. */ |
1447 | parms.name = nla_data(a[OVS_DP_ATTR_NAME]); | |
1448 | parms.type = OVS_VPORT_TYPE_INTERNAL; | |
1449 | parms.options = NULL; | |
1450 | parms.dp = dp; | |
1451 | parms.port_no = OVSP_LOCAL; | |
5cd667b0 | 1452 | parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID]; |
ccb1352e | 1453 | |
43d4be9c TG |
1454 | ovs_dp_change(dp, a); |
1455 | ||
6093ae9a JR |
1456 | /* So far only local changes have been made, now need the lock. */ |
1457 | ovs_lock(); | |
1458 | ||
ccb1352e JG |
1459 | vport = new_vport(&parms); |
1460 | if (IS_ERR(vport)) { | |
1461 | err = PTR_ERR(vport); | |
1462 | if (err == -EBUSY) | |
1463 | err = -EEXIST; | |
1464 | ||
44da5ae5 TG |
1465 | if (err == -EEXIST) { |
1466 | /* An outdated user space instance that does not understand | |
1467 | * the concept of user_features has attempted to create a new | |
1468 | * datapath and is likely to reuse it. Drop all user features. | |
1469 | */ | |
1470 | if (info->genlhdr->version < OVS_DP_VER_FEATURES) | |
1471 | ovs_dp_reset_user_features(skb, info); | |
1472 | } | |
1473 | ||
15eac2a7 | 1474 | goto err_destroy_ports_array; |
ccb1352e JG |
1475 | } |
1476 | ||
6093ae9a JR |
1477 | err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, |
1478 | info->snd_seq, 0, OVS_DP_CMD_NEW); | |
1479 | BUG_ON(err < 0); | |
ccb1352e | 1480 | |
46df7b81 | 1481 | ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id); |
59a35d60 | 1482 | list_add_tail_rcu(&dp->list_node, &ovs_net->dps); |
8e4e1713 PS |
1483 | |
1484 | ovs_unlock(); | |
ccb1352e | 1485 | |
2a94fe48 | 1486 | ovs_notify(&dp_datapath_genl_family, reply, info); |
ccb1352e JG |
1487 | return 0; |
1488 | ||
15eac2a7 | 1489 | err_destroy_ports_array: |
6093ae9a | 1490 | ovs_unlock(); |
15eac2a7 | 1491 | kfree(dp->ports); |
ccb1352e JG |
1492 | err_destroy_percpu: |
1493 | free_percpu(dp->stats_percpu); | |
1494 | err_destroy_table: | |
9b996e54 | 1495 | ovs_flow_tbl_destroy(&dp->table); |
ccb1352e | 1496 | err_free_dp: |
46df7b81 | 1497 | release_net(ovs_dp_get_net(dp)); |
ccb1352e | 1498 | kfree(dp); |
6093ae9a JR |
1499 | err_free_reply: |
1500 | kfree_skb(reply); | |
ccb1352e JG |
1501 | err: |
1502 | return err; | |
1503 | } | |
1504 | ||
8e4e1713 | 1505 | /* Called with ovs_mutex. */ |
46df7b81 | 1506 | static void __dp_destroy(struct datapath *dp) |
ccb1352e | 1507 | { |
15eac2a7 | 1508 | int i; |
ccb1352e | 1509 | |
15eac2a7 PS |
1510 | for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) { |
1511 | struct vport *vport; | |
b67bfe0d | 1512 | struct hlist_node *n; |
15eac2a7 | 1513 | |
b67bfe0d | 1514 | hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node) |
15eac2a7 PS |
1515 | if (vport->port_no != OVSP_LOCAL) |
1516 | ovs_dp_detach_port(vport); | |
1517 | } | |
ccb1352e | 1518 | |
59a35d60 | 1519 | list_del_rcu(&dp->list_node); |
ccb1352e | 1520 | |
8e4e1713 | 1521 | /* OVSP_LOCAL is datapath internal port. We need to make sure that |
e80857cc | 1522 | * all ports in datapath are destroyed first before freeing datapath. |
ccb1352e | 1523 | */ |
8e4e1713 | 1524 | ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL)); |
ccb1352e | 1525 | |
e80857cc | 1526 | /* RCU destroy the flow table */ |
ccb1352e | 1527 | call_rcu(&dp->rcu, destroy_dp_rcu); |
46df7b81 PS |
1528 | } |
1529 | ||
1530 | static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info) | |
1531 | { | |
1532 | struct sk_buff *reply; | |
1533 | struct datapath *dp; | |
1534 | int err; | |
1535 | ||
6093ae9a JR |
1536 | reply = ovs_dp_cmd_alloc_info(info); |
1537 | if (!reply) | |
1538 | return -ENOMEM; | |
1539 | ||
8e4e1713 | 1540 | ovs_lock(); |
46df7b81 PS |
1541 | dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); |
1542 | err = PTR_ERR(dp); | |
1543 | if (IS_ERR(dp)) | |
6093ae9a | 1544 | goto err_unlock_free; |
46df7b81 | 1545 | |
6093ae9a JR |
1546 | err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, |
1547 | info->snd_seq, 0, OVS_DP_CMD_DEL); | |
1548 | BUG_ON(err < 0); | |
46df7b81 PS |
1549 | |
1550 | __dp_destroy(dp); | |
8e4e1713 | 1551 | ovs_unlock(); |
ccb1352e | 1552 | |
2a94fe48 | 1553 | ovs_notify(&dp_datapath_genl_family, reply, info); |
ccb1352e JG |
1554 | |
1555 | return 0; | |
6093ae9a JR |
1556 | |
1557 | err_unlock_free: | |
8e4e1713 | 1558 | ovs_unlock(); |
6093ae9a | 1559 | kfree_skb(reply); |
8e4e1713 | 1560 | return err; |
ccb1352e JG |
1561 | } |
1562 | ||
1563 | static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info) | |
1564 | { | |
1565 | struct sk_buff *reply; | |
1566 | struct datapath *dp; | |
1567 | int err; | |
1568 | ||
6093ae9a JR |
1569 | reply = ovs_dp_cmd_alloc_info(info); |
1570 | if (!reply) | |
1571 | return -ENOMEM; | |
1572 | ||
8e4e1713 | 1573 | ovs_lock(); |
46df7b81 | 1574 | dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); |
8e4e1713 | 1575 | err = PTR_ERR(dp); |
ccb1352e | 1576 | if (IS_ERR(dp)) |
6093ae9a | 1577 | goto err_unlock_free; |
ccb1352e | 1578 | |
43d4be9c TG |
1579 | ovs_dp_change(dp, info->attrs); |
1580 | ||
6093ae9a JR |
1581 | err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, |
1582 | info->snd_seq, 0, OVS_DP_CMD_NEW); | |
1583 | BUG_ON(err < 0); | |
ccb1352e | 1584 | |
8e4e1713 | 1585 | ovs_unlock(); |
2a94fe48 | 1586 | ovs_notify(&dp_datapath_genl_family, reply, info); |
ccb1352e JG |
1587 | |
1588 | return 0; | |
6093ae9a JR |
1589 | |
1590 | err_unlock_free: | |
8e4e1713 | 1591 | ovs_unlock(); |
6093ae9a | 1592 | kfree_skb(reply); |
8e4e1713 | 1593 | return err; |
ccb1352e JG |
1594 | } |
1595 | ||
1596 | static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info) | |
1597 | { | |
1598 | struct sk_buff *reply; | |
1599 | struct datapath *dp; | |
8e4e1713 | 1600 | int err; |
ccb1352e | 1601 | |
6093ae9a JR |
1602 | reply = ovs_dp_cmd_alloc_info(info); |
1603 | if (!reply) | |
1604 | return -ENOMEM; | |
1605 | ||
8ec609d8 | 1606 | ovs_lock(); |
46df7b81 | 1607 | dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs); |
8e4e1713 PS |
1608 | if (IS_ERR(dp)) { |
1609 | err = PTR_ERR(dp); | |
6093ae9a | 1610 | goto err_unlock_free; |
8e4e1713 | 1611 | } |
6093ae9a JR |
1612 | err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid, |
1613 | info->snd_seq, 0, OVS_DP_CMD_NEW); | |
1614 | BUG_ON(err < 0); | |
8ec609d8 | 1615 | ovs_unlock(); |
ccb1352e JG |
1616 | |
1617 | return genlmsg_reply(reply, info); | |
8e4e1713 | 1618 | |
6093ae9a | 1619 | err_unlock_free: |
8ec609d8 | 1620 | ovs_unlock(); |
6093ae9a | 1621 | kfree_skb(reply); |
8e4e1713 | 1622 | return err; |
ccb1352e JG |
1623 | } |
1624 | ||
1625 | static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) | |
1626 | { | |
46df7b81 | 1627 | struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id); |
ccb1352e JG |
1628 | struct datapath *dp; |
1629 | int skip = cb->args[0]; | |
1630 | int i = 0; | |
1631 | ||
8ec609d8 PS |
1632 | ovs_lock(); |
1633 | list_for_each_entry(dp, &ovs_net->dps, list_node) { | |
77676fdb | 1634 | if (i >= skip && |
15e47304 | 1635 | ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid, |
ccb1352e JG |
1636 | cb->nlh->nlmsg_seq, NLM_F_MULTI, |
1637 | OVS_DP_CMD_NEW) < 0) | |
1638 | break; | |
1639 | i++; | |
1640 | } | |
8ec609d8 | 1641 | ovs_unlock(); |
ccb1352e JG |
1642 | |
1643 | cb->args[0] = i; | |
1644 | ||
1645 | return skb->len; | |
1646 | } | |
1647 | ||
0c200ef9 PS |
1648 | static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = { |
1649 | [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, | |
1650 | [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 }, | |
1651 | [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 }, | |
1652 | }; | |
1653 | ||
48e48a70 | 1654 | static const struct genl_ops dp_datapath_genl_ops[] = { |
ccb1352e JG |
1655 | { .cmd = OVS_DP_CMD_NEW, |
1656 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1657 | .policy = datapath_policy, | |
1658 | .doit = ovs_dp_cmd_new | |
1659 | }, | |
1660 | { .cmd = OVS_DP_CMD_DEL, | |
1661 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1662 | .policy = datapath_policy, | |
1663 | .doit = ovs_dp_cmd_del | |
1664 | }, | |
1665 | { .cmd = OVS_DP_CMD_GET, | |
1666 | .flags = 0, /* OK for unprivileged users. */ | |
1667 | .policy = datapath_policy, | |
1668 | .doit = ovs_dp_cmd_get, | |
1669 | .dumpit = ovs_dp_cmd_dump | |
1670 | }, | |
1671 | { .cmd = OVS_DP_CMD_SET, | |
1672 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
1673 | .policy = datapath_policy, | |
1674 | .doit = ovs_dp_cmd_set, | |
1675 | }, | |
1676 | }; | |
1677 | ||
0c200ef9 | 1678 | static struct genl_family dp_datapath_genl_family = { |
ccb1352e JG |
1679 | .id = GENL_ID_GENERATE, |
1680 | .hdrsize = sizeof(struct ovs_header), | |
0c200ef9 PS |
1681 | .name = OVS_DATAPATH_FAMILY, |
1682 | .version = OVS_DATAPATH_VERSION, | |
1683 | .maxattr = OVS_DP_ATTR_MAX, | |
3a4e0d6a PS |
1684 | .netnsok = true, |
1685 | .parallel_ops = true, | |
0c200ef9 PS |
1686 | .ops = dp_datapath_genl_ops, |
1687 | .n_ops = ARRAY_SIZE(dp_datapath_genl_ops), | |
1688 | .mcgrps = &ovs_dp_datapath_multicast_group, | |
1689 | .n_mcgrps = 1, | |
ccb1352e JG |
1690 | }; |
1691 | ||
8e4e1713 | 1692 | /* Called with ovs_mutex or RCU read lock. */ |
ccb1352e | 1693 | static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb, |
15e47304 | 1694 | u32 portid, u32 seq, u32 flags, u8 cmd) |
ccb1352e JG |
1695 | { |
1696 | struct ovs_header *ovs_header; | |
1697 | struct ovs_vport_stats vport_stats; | |
1698 | int err; | |
1699 | ||
15e47304 | 1700 | ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family, |
ccb1352e JG |
1701 | flags, cmd); |
1702 | if (!ovs_header) | |
1703 | return -EMSGSIZE; | |
1704 | ||
1705 | ovs_header->dp_ifindex = get_dpifindex(vport->dp); | |
1706 | ||
028d6a67 DM |
1707 | if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) || |
1708 | nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) || | |
5cd667b0 AW |
1709 | nla_put_string(skb, OVS_VPORT_ATTR_NAME, |
1710 | vport->ops->get_name(vport))) | |
028d6a67 | 1711 | goto nla_put_failure; |
ccb1352e JG |
1712 | |
1713 | ovs_vport_get_stats(vport, &vport_stats); | |
028d6a67 DM |
1714 | if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats), |
1715 | &vport_stats)) | |
1716 | goto nla_put_failure; | |
ccb1352e | 1717 | |
5cd667b0 AW |
1718 | if (ovs_vport_get_upcall_portids(vport, skb)) |
1719 | goto nla_put_failure; | |
1720 | ||
ccb1352e JG |
1721 | err = ovs_vport_get_options(vport, skb); |
1722 | if (err == -EMSGSIZE) | |
1723 | goto error; | |
1724 | ||
1725 | return genlmsg_end(skb, ovs_header); | |
1726 | ||
1727 | nla_put_failure: | |
1728 | err = -EMSGSIZE; | |
1729 | error: | |
1730 | genlmsg_cancel(skb, ovs_header); | |
1731 | return err; | |
1732 | } | |
1733 | ||
6093ae9a JR |
1734 | static struct sk_buff *ovs_vport_cmd_alloc_info(void) |
1735 | { | |
1736 | return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); | |
1737 | } | |
1738 | ||
1739 | /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */ | |
15e47304 | 1740 | struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid, |
ccb1352e JG |
1741 | u32 seq, u8 cmd) |
1742 | { | |
1743 | struct sk_buff *skb; | |
1744 | int retval; | |
1745 | ||
1746 | skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); | |
1747 | if (!skb) | |
1748 | return ERR_PTR(-ENOMEM); | |
1749 | ||
15e47304 | 1750 | retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd); |
a9341512 JG |
1751 | BUG_ON(retval < 0); |
1752 | ||
ccb1352e JG |
1753 | return skb; |
1754 | } | |
1755 | ||
8e4e1713 | 1756 | /* Called with ovs_mutex or RCU read lock. */ |
46df7b81 | 1757 | static struct vport *lookup_vport(struct net *net, |
12eb18f7 | 1758 | const struct ovs_header *ovs_header, |
ccb1352e JG |
1759 | struct nlattr *a[OVS_VPORT_ATTR_MAX + 1]) |
1760 | { | |
1761 | struct datapath *dp; | |
1762 | struct vport *vport; | |
1763 | ||
1764 | if (a[OVS_VPORT_ATTR_NAME]) { | |
46df7b81 | 1765 | vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME])); |
ccb1352e JG |
1766 | if (!vport) |
1767 | return ERR_PTR(-ENODEV); | |
651a68ea BP |
1768 | if (ovs_header->dp_ifindex && |
1769 | ovs_header->dp_ifindex != get_dpifindex(vport->dp)) | |
1770 | return ERR_PTR(-ENODEV); | |
ccb1352e JG |
1771 | return vport; |
1772 | } else if (a[OVS_VPORT_ATTR_PORT_NO]) { | |
1773 | u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]); | |
1774 | ||
1775 | if (port_no >= DP_MAX_PORTS) | |
1776 | return ERR_PTR(-EFBIG); | |
1777 | ||
46df7b81 | 1778 | dp = get_dp(net, ovs_header->dp_ifindex); |
ccb1352e JG |
1779 | if (!dp) |
1780 | return ERR_PTR(-ENODEV); | |
1781 | ||
8e4e1713 | 1782 | vport = ovs_vport_ovsl_rcu(dp, port_no); |
ccb1352e | 1783 | if (!vport) |
14408dba | 1784 | return ERR_PTR(-ENODEV); |
ccb1352e JG |
1785 | return vport; |
1786 | } else | |
1787 | return ERR_PTR(-EINVAL); | |
1788 | } | |
1789 | ||
1790 | static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info) | |
1791 | { | |
1792 | struct nlattr **a = info->attrs; | |
1793 | struct ovs_header *ovs_header = info->userhdr; | |
1794 | struct vport_parms parms; | |
1795 | struct sk_buff *reply; | |
1796 | struct vport *vport; | |
1797 | struct datapath *dp; | |
1798 | u32 port_no; | |
1799 | int err; | |
1800 | ||
ccb1352e JG |
1801 | if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] || |
1802 | !a[OVS_VPORT_ATTR_UPCALL_PID]) | |
6093ae9a JR |
1803 | return -EINVAL; |
1804 | ||
1805 | port_no = a[OVS_VPORT_ATTR_PORT_NO] | |
1806 | ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0; | |
1807 | if (port_no >= DP_MAX_PORTS) | |
1808 | return -EFBIG; | |
1809 | ||
1810 | reply = ovs_vport_cmd_alloc_info(); | |
1811 | if (!reply) | |
1812 | return -ENOMEM; | |
ccb1352e | 1813 | |
8e4e1713 | 1814 | ovs_lock(); |
62b9c8d0 | 1815 | restart: |
46df7b81 | 1816 | dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex); |
ccb1352e JG |
1817 | err = -ENODEV; |
1818 | if (!dp) | |
6093ae9a | 1819 | goto exit_unlock_free; |
ccb1352e | 1820 | |
6093ae9a | 1821 | if (port_no) { |
8e4e1713 | 1822 | vport = ovs_vport_ovsl(dp, port_no); |
ccb1352e JG |
1823 | err = -EBUSY; |
1824 | if (vport) | |
6093ae9a | 1825 | goto exit_unlock_free; |
ccb1352e JG |
1826 | } else { |
1827 | for (port_no = 1; ; port_no++) { | |
1828 | if (port_no >= DP_MAX_PORTS) { | |
1829 | err = -EFBIG; | |
6093ae9a | 1830 | goto exit_unlock_free; |
ccb1352e | 1831 | } |
8e4e1713 | 1832 | vport = ovs_vport_ovsl(dp, port_no); |
ccb1352e JG |
1833 | if (!vport) |
1834 | break; | |
1835 | } | |
1836 | } | |
1837 | ||
1838 | parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]); | |
1839 | parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]); | |
1840 | parms.options = a[OVS_VPORT_ATTR_OPTIONS]; | |
1841 | parms.dp = dp; | |
1842 | parms.port_no = port_no; | |
5cd667b0 | 1843 | parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID]; |
ccb1352e JG |
1844 | |
1845 | vport = new_vport(&parms); | |
1846 | err = PTR_ERR(vport); | |
62b9c8d0 TG |
1847 | if (IS_ERR(vport)) { |
1848 | if (err == -EAGAIN) | |
1849 | goto restart; | |
6093ae9a | 1850 | goto exit_unlock_free; |
62b9c8d0 | 1851 | } |
ccb1352e | 1852 | |
6093ae9a JR |
1853 | err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid, |
1854 | info->snd_seq, 0, OVS_VPORT_CMD_NEW); | |
1855 | BUG_ON(err < 0); | |
1856 | ovs_unlock(); | |
ed661185 | 1857 | |
2a94fe48 | 1858 | ovs_notify(&dp_vport_genl_family, reply, info); |
6093ae9a | 1859 | return 0; |
ccb1352e | 1860 | |
6093ae9a | 1861 | exit_unlock_free: |
8e4e1713 | 1862 | ovs_unlock(); |
6093ae9a | 1863 | kfree_skb(reply); |
ccb1352e JG |
1864 | return err; |
1865 | } | |
1866 | ||
1867 | static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info) | |
1868 | { | |
1869 | struct nlattr **a = info->attrs; | |
1870 | struct sk_buff *reply; | |
1871 | struct vport *vport; | |
1872 | int err; | |
1873 | ||
6093ae9a JR |
1874 | reply = ovs_vport_cmd_alloc_info(); |
1875 | if (!reply) | |
1876 | return -ENOMEM; | |
1877 | ||
8e4e1713 | 1878 | ovs_lock(); |
46df7b81 | 1879 | vport = lookup_vport(sock_net(skb->sk), info->userhdr, a); |
ccb1352e JG |
1880 | err = PTR_ERR(vport); |
1881 | if (IS_ERR(vport)) | |
6093ae9a | 1882 | goto exit_unlock_free; |
ccb1352e | 1883 | |
ccb1352e | 1884 | if (a[OVS_VPORT_ATTR_TYPE] && |
f44f3408 | 1885 | nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) { |
ccb1352e | 1886 | err = -EINVAL; |
6093ae9a | 1887 | goto exit_unlock_free; |
a9341512 JG |
1888 | } |
1889 | ||
f44f3408 | 1890 | if (a[OVS_VPORT_ATTR_OPTIONS]) { |
ccb1352e | 1891 | err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]); |
f44f3408 | 1892 | if (err) |
6093ae9a | 1893 | goto exit_unlock_free; |
f44f3408 | 1894 | } |
a9341512 | 1895 | |
5cd667b0 AW |
1896 | |
1897 | if (a[OVS_VPORT_ATTR_UPCALL_PID]) { | |
1898 | struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID]; | |
1899 | ||
1900 | err = ovs_vport_set_upcall_portids(vport, ids); | |
1901 | if (err) | |
1902 | goto exit_unlock_free; | |
1903 | } | |
ccb1352e | 1904 | |
a9341512 JG |
1905 | err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid, |
1906 | info->snd_seq, 0, OVS_VPORT_CMD_NEW); | |
1907 | BUG_ON(err < 0); | |
ccb1352e | 1908 | |
8e4e1713 | 1909 | ovs_unlock(); |
2a94fe48 | 1910 | ovs_notify(&dp_vport_genl_family, reply, info); |
8e4e1713 | 1911 | return 0; |
ccb1352e | 1912 | |
6093ae9a | 1913 | exit_unlock_free: |
8e4e1713 | 1914 | ovs_unlock(); |
6093ae9a | 1915 | kfree_skb(reply); |
ccb1352e JG |
1916 | return err; |
1917 | } | |
1918 | ||
1919 | static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info) | |
1920 | { | |
1921 | struct nlattr **a = info->attrs; | |
1922 | struct sk_buff *reply; | |
1923 | struct vport *vport; | |
1924 | int err; | |
1925 | ||
6093ae9a JR |
1926 | reply = ovs_vport_cmd_alloc_info(); |
1927 | if (!reply) | |
1928 | return -ENOMEM; | |
1929 | ||
8e4e1713 | 1930 | ovs_lock(); |
46df7b81 | 1931 | vport = lookup_vport(sock_net(skb->sk), info->userhdr, a); |
ccb1352e JG |
1932 | err = PTR_ERR(vport); |
1933 | if (IS_ERR(vport)) | |
6093ae9a | 1934 | goto exit_unlock_free; |
ccb1352e JG |
1935 | |
1936 | if (vport->port_no == OVSP_LOCAL) { | |
1937 | err = -EINVAL; | |
6093ae9a | 1938 | goto exit_unlock_free; |
ccb1352e JG |
1939 | } |
1940 | ||
6093ae9a JR |
1941 | err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid, |
1942 | info->snd_seq, 0, OVS_VPORT_CMD_DEL); | |
1943 | BUG_ON(err < 0); | |
ccb1352e | 1944 | ovs_dp_detach_port(vport); |
6093ae9a | 1945 | ovs_unlock(); |
ccb1352e | 1946 | |
2a94fe48 | 1947 | ovs_notify(&dp_vport_genl_family, reply, info); |
6093ae9a | 1948 | return 0; |
ccb1352e | 1949 | |
6093ae9a | 1950 | exit_unlock_free: |
8e4e1713 | 1951 | ovs_unlock(); |
6093ae9a | 1952 | kfree_skb(reply); |
ccb1352e JG |
1953 | return err; |
1954 | } | |
1955 | ||
1956 | static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info) | |
1957 | { | |
1958 | struct nlattr **a = info->attrs; | |
1959 | struct ovs_header *ovs_header = info->userhdr; | |
1960 | struct sk_buff *reply; | |
1961 | struct vport *vport; | |
1962 | int err; | |
1963 | ||
6093ae9a JR |
1964 | reply = ovs_vport_cmd_alloc_info(); |
1965 | if (!reply) | |
1966 | return -ENOMEM; | |
1967 | ||
ccb1352e | 1968 | rcu_read_lock(); |
46df7b81 | 1969 | vport = lookup_vport(sock_net(skb->sk), ovs_header, a); |
ccb1352e JG |
1970 | err = PTR_ERR(vport); |
1971 | if (IS_ERR(vport)) | |
6093ae9a JR |
1972 | goto exit_unlock_free; |
1973 | err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid, | |
1974 | info->snd_seq, 0, OVS_VPORT_CMD_NEW); | |
1975 | BUG_ON(err < 0); | |
ccb1352e JG |
1976 | rcu_read_unlock(); |
1977 | ||
1978 | return genlmsg_reply(reply, info); | |
1979 | ||
6093ae9a | 1980 | exit_unlock_free: |
ccb1352e | 1981 | rcu_read_unlock(); |
6093ae9a | 1982 | kfree_skb(reply); |
ccb1352e JG |
1983 | return err; |
1984 | } | |
1985 | ||
1986 | static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb) | |
1987 | { | |
1988 | struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh)); | |
1989 | struct datapath *dp; | |
15eac2a7 PS |
1990 | int bucket = cb->args[0], skip = cb->args[1]; |
1991 | int i, j = 0; | |
ccb1352e | 1992 | |
42ee19e2 | 1993 | rcu_read_lock(); |
cc3a5ae6 | 1994 | dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex); |
42ee19e2 JR |
1995 | if (!dp) { |
1996 | rcu_read_unlock(); | |
ccb1352e | 1997 | return -ENODEV; |
42ee19e2 | 1998 | } |
15eac2a7 | 1999 | for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) { |
ccb1352e | 2000 | struct vport *vport; |
15eac2a7 PS |
2001 | |
2002 | j = 0; | |
b67bfe0d | 2003 | hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) { |
15eac2a7 PS |
2004 | if (j >= skip && |
2005 | ovs_vport_cmd_fill_info(vport, skb, | |
15e47304 | 2006 | NETLINK_CB(cb->skb).portid, |
15eac2a7 PS |
2007 | cb->nlh->nlmsg_seq, |
2008 | NLM_F_MULTI, | |
2009 | OVS_VPORT_CMD_NEW) < 0) | |
2010 | goto out; | |
2011 | ||
2012 | j++; | |
2013 | } | |
2014 | skip = 0; | |
ccb1352e | 2015 | } |
15eac2a7 | 2016 | out: |
ccb1352e JG |
2017 | rcu_read_unlock(); |
2018 | ||
15eac2a7 PS |
2019 | cb->args[0] = i; |
2020 | cb->args[1] = j; | |
ccb1352e | 2021 | |
15eac2a7 | 2022 | return skb->len; |
ccb1352e JG |
2023 | } |
2024 | ||
0c200ef9 PS |
2025 | static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = { |
2026 | [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 }, | |
2027 | [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) }, | |
2028 | [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 }, | |
2029 | [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 }, | |
2030 | [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 }, | |
2031 | [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, | |
2032 | }; | |
2033 | ||
48e48a70 | 2034 | static const struct genl_ops dp_vport_genl_ops[] = { |
ccb1352e JG |
2035 | { .cmd = OVS_VPORT_CMD_NEW, |
2036 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
2037 | .policy = vport_policy, | |
2038 | .doit = ovs_vport_cmd_new | |
2039 | }, | |
2040 | { .cmd = OVS_VPORT_CMD_DEL, | |
2041 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
2042 | .policy = vport_policy, | |
2043 | .doit = ovs_vport_cmd_del | |
2044 | }, | |
2045 | { .cmd = OVS_VPORT_CMD_GET, | |
2046 | .flags = 0, /* OK for unprivileged users. */ | |
2047 | .policy = vport_policy, | |
2048 | .doit = ovs_vport_cmd_get, | |
2049 | .dumpit = ovs_vport_cmd_dump | |
2050 | }, | |
2051 | { .cmd = OVS_VPORT_CMD_SET, | |
2052 | .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */ | |
2053 | .policy = vport_policy, | |
2054 | .doit = ovs_vport_cmd_set, | |
2055 | }, | |
2056 | }; | |
2057 | ||
0c200ef9 PS |
2058 | struct genl_family dp_vport_genl_family = { |
2059 | .id = GENL_ID_GENERATE, | |
2060 | .hdrsize = sizeof(struct ovs_header), | |
2061 | .name = OVS_VPORT_FAMILY, | |
2062 | .version = OVS_VPORT_VERSION, | |
2063 | .maxattr = OVS_VPORT_ATTR_MAX, | |
2064 | .netnsok = true, | |
2065 | .parallel_ops = true, | |
2066 | .ops = dp_vport_genl_ops, | |
2067 | .n_ops = ARRAY_SIZE(dp_vport_genl_ops), | |
2068 | .mcgrps = &ovs_dp_vport_multicast_group, | |
2069 | .n_mcgrps = 1, | |
ccb1352e JG |
2070 | }; |
2071 | ||
0c200ef9 PS |
2072 | static struct genl_family * const dp_genl_families[] = { |
2073 | &dp_datapath_genl_family, | |
2074 | &dp_vport_genl_family, | |
2075 | &dp_flow_genl_family, | |
2076 | &dp_packet_genl_family, | |
ccb1352e JG |
2077 | }; |
2078 | ||
2079 | static void dp_unregister_genl(int n_families) | |
2080 | { | |
2081 | int i; | |
2082 | ||
2083 | for (i = 0; i < n_families; i++) | |
0c200ef9 | 2084 | genl_unregister_family(dp_genl_families[i]); |
ccb1352e JG |
2085 | } |
2086 | ||
2087 | static int dp_register_genl(void) | |
2088 | { | |
ccb1352e JG |
2089 | int err; |
2090 | int i; | |
2091 | ||
ccb1352e | 2092 | for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) { |
ccb1352e | 2093 | |
0c200ef9 | 2094 | err = genl_register_family(dp_genl_families[i]); |
ccb1352e JG |
2095 | if (err) |
2096 | goto error; | |
ccb1352e JG |
2097 | } |
2098 | ||
2099 | return 0; | |
2100 | ||
2101 | error: | |
0c200ef9 | 2102 | dp_unregister_genl(i); |
ccb1352e JG |
2103 | return err; |
2104 | } | |
2105 | ||
46df7b81 PS |
2106 | static int __net_init ovs_init_net(struct net *net) |
2107 | { | |
2108 | struct ovs_net *ovs_net = net_generic(net, ovs_net_id); | |
2109 | ||
2110 | INIT_LIST_HEAD(&ovs_net->dps); | |
8e4e1713 | 2111 | INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq); |
46df7b81 PS |
2112 | return 0; |
2113 | } | |
2114 | ||
2115 | static void __net_exit ovs_exit_net(struct net *net) | |
2116 | { | |
46df7b81 | 2117 | struct datapath *dp, *dp_next; |
8e4e1713 | 2118 | struct ovs_net *ovs_net = net_generic(net, ovs_net_id); |
46df7b81 | 2119 | |
8e4e1713 | 2120 | ovs_lock(); |
46df7b81 PS |
2121 | list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node) |
2122 | __dp_destroy(dp); | |
8e4e1713 PS |
2123 | ovs_unlock(); |
2124 | ||
2125 | cancel_work_sync(&ovs_net->dp_notify_work); | |
46df7b81 PS |
2126 | } |
2127 | ||
2128 | static struct pernet_operations ovs_net_ops = { | |
2129 | .init = ovs_init_net, | |
2130 | .exit = ovs_exit_net, | |
2131 | .id = &ovs_net_id, | |
2132 | .size = sizeof(struct ovs_net), | |
2133 | }; | |
2134 | ||
ccb1352e JG |
2135 | static int __init dp_init(void) |
2136 | { | |
ccb1352e JG |
2137 | int err; |
2138 | ||
3523b29b | 2139 | BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb)); |
ccb1352e JG |
2140 | |
2141 | pr_info("Open vSwitch switching datapath\n"); | |
2142 | ||
971427f3 | 2143 | err = action_fifos_init(); |
ccb1352e JG |
2144 | if (err) |
2145 | goto error; | |
2146 | ||
971427f3 AZ |
2147 | err = ovs_internal_dev_rtnl_link_register(); |
2148 | if (err) | |
2149 | goto error_action_fifos_exit; | |
2150 | ||
5b9e7e16 JP |
2151 | err = ovs_flow_init(); |
2152 | if (err) | |
2153 | goto error_unreg_rtnl_link; | |
2154 | ||
ccb1352e JG |
2155 | err = ovs_vport_init(); |
2156 | if (err) | |
2157 | goto error_flow_exit; | |
2158 | ||
46df7b81 | 2159 | err = register_pernet_device(&ovs_net_ops); |
ccb1352e JG |
2160 | if (err) |
2161 | goto error_vport_exit; | |
2162 | ||
46df7b81 PS |
2163 | err = register_netdevice_notifier(&ovs_dp_device_notifier); |
2164 | if (err) | |
2165 | goto error_netns_exit; | |
2166 | ||
62b9c8d0 TG |
2167 | err = ovs_netdev_init(); |
2168 | if (err) | |
2169 | goto error_unreg_notifier; | |
2170 | ||
ccb1352e JG |
2171 | err = dp_register_genl(); |
2172 | if (err < 0) | |
62b9c8d0 | 2173 | goto error_unreg_netdev; |
ccb1352e | 2174 | |
ccb1352e JG |
2175 | return 0; |
2176 | ||
62b9c8d0 TG |
2177 | error_unreg_netdev: |
2178 | ovs_netdev_exit(); | |
ccb1352e JG |
2179 | error_unreg_notifier: |
2180 | unregister_netdevice_notifier(&ovs_dp_device_notifier); | |
46df7b81 PS |
2181 | error_netns_exit: |
2182 | unregister_pernet_device(&ovs_net_ops); | |
ccb1352e JG |
2183 | error_vport_exit: |
2184 | ovs_vport_exit(); | |
2185 | error_flow_exit: | |
2186 | ovs_flow_exit(); | |
5b9e7e16 JP |
2187 | error_unreg_rtnl_link: |
2188 | ovs_internal_dev_rtnl_link_unregister(); | |
971427f3 AZ |
2189 | error_action_fifos_exit: |
2190 | action_fifos_exit(); | |
ccb1352e JG |
2191 | error: |
2192 | return err; | |
2193 | } | |
2194 | ||
2195 | static void dp_cleanup(void) | |
2196 | { | |
ccb1352e | 2197 | dp_unregister_genl(ARRAY_SIZE(dp_genl_families)); |
62b9c8d0 | 2198 | ovs_netdev_exit(); |
ccb1352e | 2199 | unregister_netdevice_notifier(&ovs_dp_device_notifier); |
46df7b81 PS |
2200 | unregister_pernet_device(&ovs_net_ops); |
2201 | rcu_barrier(); | |
ccb1352e JG |
2202 | ovs_vport_exit(); |
2203 | ovs_flow_exit(); | |
5b9e7e16 | 2204 | ovs_internal_dev_rtnl_link_unregister(); |
971427f3 | 2205 | action_fifos_exit(); |
ccb1352e JG |
2206 | } |
2207 | ||
2208 | module_init(dp_init); | |
2209 | module_exit(dp_cleanup); | |
2210 | ||
2211 | MODULE_DESCRIPTION("Open vSwitch switching datapath"); | |
2212 | MODULE_LICENSE("GPL"); |