]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
a6057323 | 2 | * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks. |
a14bc59f BP |
3 | * Distributed under the terms of the GNU GPL version 2. |
4 | * | |
5 | * Significant portions of this file may be copied from parts of the Linux | |
6 | * kernel, by Linus Torvalds and others. | |
064af421 BP |
7 | */ |
8 | ||
9 | /* Functions for managing the dp interface/device. */ | |
10 | ||
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/fs.h> | |
14 | #include <linux/if_arp.h> | |
15 | #include <linux/if_bridge.h> | |
16 | #include <linux/if_vlan.h> | |
17 | #include <linux/in.h> | |
18 | #include <linux/ip.h> | |
19 | #include <linux/delay.h> | |
20 | #include <linux/time.h> | |
21 | #include <linux/etherdevice.h> | |
22 | #include <linux/kernel.h> | |
23 | #include <linux/kthread.h> | |
24 | #include <linux/llc.h> | |
25 | #include <linux/mutex.h> | |
26 | #include <linux/percpu.h> | |
27 | #include <linux/rcupdate.h> | |
28 | #include <linux/tcp.h> | |
29 | #include <linux/udp.h> | |
30 | #include <linux/version.h> | |
31 | #include <linux/ethtool.h> | |
32 | #include <linux/random.h> | |
33 | #include <linux/wait.h> | |
34 | #include <asm/system.h> | |
35 | #include <asm/div64.h> | |
36 | #include <asm/bug.h> | |
37 | #include <linux/netfilter_bridge.h> | |
38 | #include <linux/netfilter_ipv4.h> | |
39 | #include <linux/inetdevice.h> | |
40 | #include <linux/list.h> | |
41 | #include <linux/rculist.h> | |
42 | #include <linux/workqueue.h> | |
43 | #include <linux/dmi.h> | |
3c5f6de3 | 44 | #include <net/inet_ecn.h> |
064af421 BP |
45 | #include <net/llc.h> |
46 | ||
47 | #include "openvswitch/datapath-protocol.h" | |
48 | #include "datapath.h" | |
49 | #include "actions.h" | |
50 | #include "dp_dev.h" | |
51 | #include "flow.h" | |
52 | ||
53 | #include "compat.h" | |
54 | ||
55 | ||
56 | int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd); | |
57 | EXPORT_SYMBOL(dp_ioctl_hook); | |
58 | ||
064af421 | 59 | /* Datapaths. Protected on the read side by rcu_read_lock, on the write side |
0d3b8a34 | 60 | * by dp_mutex. |
064af421 BP |
61 | * |
62 | * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL | |
63 | * lock first. | |
64 | * | |
65 | * It is safe to access the datapath and net_bridge_port structures with just | |
66 | * dp_mutex. | |
67 | */ | |
68 | static struct datapath *dps[ODP_MAX]; | |
69 | static DEFINE_MUTEX(dp_mutex); | |
70 | ||
71 | /* Number of milliseconds between runs of the maintenance thread. */ | |
72 | #define MAINT_SLEEP_MSECS 1000 | |
73 | ||
74 | static int new_nbp(struct datapath *, struct net_device *, int port_no); | |
75 | ||
76 | /* Must be called with rcu_read_lock or dp_mutex. */ | |
77 | struct datapath *get_dp(int dp_idx) | |
78 | { | |
79 | if (dp_idx < 0 || dp_idx >= ODP_MAX) | |
80 | return NULL; | |
81 | return rcu_dereference(dps[dp_idx]); | |
82 | } | |
83 | EXPORT_SYMBOL_GPL(get_dp); | |
84 | ||
35f7605b | 85 | static struct datapath *get_dp_locked(int dp_idx) |
064af421 BP |
86 | { |
87 | struct datapath *dp; | |
88 | ||
89 | mutex_lock(&dp_mutex); | |
90 | dp = get_dp(dp_idx); | |
91 | if (dp) | |
92 | mutex_lock(&dp->mutex); | |
93 | mutex_unlock(&dp_mutex); | |
94 | return dp; | |
95 | } | |
96 | ||
97 | static inline size_t br_nlmsg_size(void) | |
98 | { | |
99 | return NLMSG_ALIGN(sizeof(struct ifinfomsg)) | |
100 | + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */ | |
101 | + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */ | |
102 | + nla_total_size(4) /* IFLA_MASTER */ | |
103 | + nla_total_size(4) /* IFLA_MTU */ | |
104 | + nla_total_size(4) /* IFLA_LINK */ | |
105 | + nla_total_size(1); /* IFLA_OPERSTATE */ | |
106 | } | |
107 | ||
108 | static int dp_fill_ifinfo(struct sk_buff *skb, | |
109 | const struct net_bridge_port *port, | |
110 | int event, unsigned int flags) | |
111 | { | |
112 | const struct datapath *dp = port->dp; | |
113 | const struct net_device *dev = port->dev; | |
114 | struct ifinfomsg *hdr; | |
115 | struct nlmsghdr *nlh; | |
116 | ||
117 | nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags); | |
118 | if (nlh == NULL) | |
119 | return -EMSGSIZE; | |
120 | ||
121 | hdr = nlmsg_data(nlh); | |
122 | hdr->ifi_family = AF_BRIDGE; | |
123 | hdr->__ifi_pad = 0; | |
124 | hdr->ifi_type = dev->type; | |
125 | hdr->ifi_index = dev->ifindex; | |
126 | hdr->ifi_flags = dev_get_flags(dev); | |
127 | hdr->ifi_change = 0; | |
128 | ||
129 | NLA_PUT_STRING(skb, IFLA_IFNAME, dev->name); | |
130 | NLA_PUT_U32(skb, IFLA_MASTER, dp->ports[ODPP_LOCAL]->dev->ifindex); | |
131 | NLA_PUT_U32(skb, IFLA_MTU, dev->mtu); | |
132 | #ifdef IFLA_OPERSTATE | |
133 | NLA_PUT_U8(skb, IFLA_OPERSTATE, | |
134 | netif_running(dev) ? dev->operstate : IF_OPER_DOWN); | |
135 | #endif | |
136 | ||
137 | if (dev->addr_len) | |
138 | NLA_PUT(skb, IFLA_ADDRESS, dev->addr_len, dev->dev_addr); | |
139 | ||
140 | if (dev->ifindex != dev->iflink) | |
141 | NLA_PUT_U32(skb, IFLA_LINK, dev->iflink); | |
142 | ||
143 | return nlmsg_end(skb, nlh); | |
144 | ||
145 | nla_put_failure: | |
146 | nlmsg_cancel(skb, nlh); | |
147 | return -EMSGSIZE; | |
148 | } | |
149 | ||
150 | static void dp_ifinfo_notify(int event, struct net_bridge_port *port) | |
151 | { | |
152 | struct net *net = dev_net(port->dev); | |
153 | struct sk_buff *skb; | |
154 | int err = -ENOBUFS; | |
155 | ||
156 | skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL); | |
157 | if (skb == NULL) | |
158 | goto errout; | |
159 | ||
160 | err = dp_fill_ifinfo(skb, port, event, 0); | |
161 | if (err < 0) { | |
162 | /* -EMSGSIZE implies BUG in br_nlmsg_size() */ | |
163 | WARN_ON(err == -EMSGSIZE); | |
164 | kfree_skb(skb); | |
165 | goto errout; | |
166 | } | |
cfe7c1f5 BP |
167 | rtnl_notify(skb, net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL); |
168 | return; | |
064af421 BP |
169 | errout: |
170 | if (err < 0) | |
171 | rtnl_set_sk_err(net, RTNLGRP_LINK, err); | |
172 | } | |
173 | ||
58c342f6 BP |
174 | static void release_dp(struct kobject *kobj) |
175 | { | |
176 | struct datapath *dp = container_of(kobj, struct datapath, ifobj); | |
177 | kfree(dp); | |
178 | } | |
179 | ||
35f7605b | 180 | static struct kobj_type dp_ktype = { |
58c342f6 BP |
181 | .release = release_dp |
182 | }; | |
183 | ||
064af421 BP |
184 | static int create_dp(int dp_idx, const char __user *devnamep) |
185 | { | |
186 | struct net_device *dp_dev; | |
187 | char devname[IFNAMSIZ]; | |
188 | struct datapath *dp; | |
189 | int err; | |
190 | int i; | |
191 | ||
192 | if (devnamep) { | |
193 | err = -EFAULT; | |
194 | if (strncpy_from_user(devname, devnamep, IFNAMSIZ - 1) < 0) | |
195 | goto err; | |
196 | devname[IFNAMSIZ - 1] = '\0'; | |
197 | } else { | |
198 | snprintf(devname, sizeof devname, "of%d", dp_idx); | |
199 | } | |
200 | ||
201 | rtnl_lock(); | |
202 | mutex_lock(&dp_mutex); | |
203 | err = -ENODEV; | |
204 | if (!try_module_get(THIS_MODULE)) | |
205 | goto err_unlock; | |
206 | ||
207 | /* Exit early if a datapath with that number already exists. | |
208 | * (We don't use -EEXIST because that's ambiguous with 'devname' | |
209 | * conflicting with an existing network device name.) */ | |
210 | err = -EBUSY; | |
211 | if (get_dp(dp_idx)) | |
212 | goto err_put_module; | |
213 | ||
214 | err = -ENOMEM; | |
215 | dp = kzalloc(sizeof *dp, GFP_KERNEL); | |
216 | if (dp == NULL) | |
217 | goto err_put_module; | |
828bc1f0 | 218 | INIT_LIST_HEAD(&dp->port_list); |
064af421 BP |
219 | mutex_init(&dp->mutex); |
220 | dp->dp_idx = dp_idx; | |
221 | for (i = 0; i < DP_N_QUEUES; i++) | |
222 | skb_queue_head_init(&dp->queues[i]); | |
223 | init_waitqueue_head(&dp->waitqueue); | |
224 | ||
58c342f6 | 225 | /* Initialize kobject for bridge. This will be added as |
b0c32774 | 226 | * /sys/class/net/<devname>/brif later, if sysfs is enabled. */ |
58c342f6 | 227 | dp->ifobj.kset = NULL; |
58c342f6 BP |
228 | kobject_init(&dp->ifobj, &dp_ktype); |
229 | ||
828bc1f0 BP |
230 | /* Allocate table. */ |
231 | err = -ENOMEM; | |
232 | rcu_assign_pointer(dp->table, dp_table_create(DP_L1_SIZE)); | |
233 | if (!dp->table) | |
234 | goto err_free_dp; | |
235 | ||
d6fbec6d | 236 | /* Set up our datapath device. */ |
064af421 BP |
237 | dp_dev = dp_dev_create(dp, devname, ODPP_LOCAL); |
238 | err = PTR_ERR(dp_dev); | |
239 | if (IS_ERR(dp_dev)) | |
828bc1f0 | 240 | goto err_destroy_table; |
064af421 BP |
241 | |
242 | err = new_nbp(dp, dp_dev, ODPP_LOCAL); | |
828bc1f0 BP |
243 | if (err) { |
244 | dp_dev_destroy(dp_dev); | |
064af421 | 245 | goto err_destroy_table; |
828bc1f0 | 246 | } |
064af421 BP |
247 | |
248 | dp->drop_frags = 0; | |
249 | dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); | |
250 | if (!dp->stats_percpu) | |
251 | goto err_destroy_local_port; | |
252 | ||
253 | rcu_assign_pointer(dps[dp_idx], dp); | |
254 | mutex_unlock(&dp_mutex); | |
255 | rtnl_unlock(); | |
256 | ||
2ba9026e | 257 | dp_sysfs_add_dp(dp); |
064af421 BP |
258 | |
259 | return 0; | |
260 | ||
261 | err_destroy_local_port: | |
72ca14c1 | 262 | dp_del_port(dp->ports[ODPP_LOCAL]); |
064af421 BP |
263 | err_destroy_table: |
264 | dp_table_destroy(dp->table, 0); | |
064af421 BP |
265 | err_free_dp: |
266 | kfree(dp); | |
267 | err_put_module: | |
268 | module_put(THIS_MODULE); | |
269 | err_unlock: | |
270 | mutex_unlock(&dp_mutex); | |
271 | rtnl_unlock(); | |
272 | err: | |
273 | return err; | |
274 | } | |
275 | ||
72ca14c1 | 276 | static void do_destroy_dp(struct datapath *dp) |
064af421 BP |
277 | { |
278 | struct net_bridge_port *p, *n; | |
279 | int i; | |
280 | ||
6fba0d0b BP |
281 | list_for_each_entry_safe (p, n, &dp->port_list, node) |
282 | if (p->port_no != ODPP_LOCAL) | |
72ca14c1 | 283 | dp_del_port(p); |
6fba0d0b | 284 | |
2ba9026e | 285 | dp_sysfs_del_dp(dp); |
064af421 | 286 | |
064af421 | 287 | rcu_assign_pointer(dps[dp->dp_idx], NULL); |
064af421 | 288 | |
72ca14c1 | 289 | dp_del_port(dp->ports[ODPP_LOCAL]); |
6fba0d0b | 290 | |
064af421 | 291 | dp_table_destroy(dp->table, 1); |
6fba0d0b | 292 | |
064af421 BP |
293 | for (i = 0; i < DP_N_QUEUES; i++) |
294 | skb_queue_purge(&dp->queues[i]); | |
295 | for (i = 0; i < DP_MAX_GROUPS; i++) | |
296 | kfree(dp->groups[i]); | |
297 | free_percpu(dp->stats_percpu); | |
58c342f6 | 298 | kobject_put(&dp->ifobj); |
064af421 BP |
299 | module_put(THIS_MODULE); |
300 | } | |
301 | ||
302 | static int destroy_dp(int dp_idx) | |
303 | { | |
064af421 | 304 | struct datapath *dp; |
064af421 BP |
305 | int err; |
306 | ||
307 | rtnl_lock(); | |
308 | mutex_lock(&dp_mutex); | |
309 | dp = get_dp(dp_idx); | |
310 | err = -ENODEV; | |
311 | if (!dp) | |
312 | goto err_unlock; | |
313 | ||
72ca14c1 | 314 | do_destroy_dp(dp); |
064af421 BP |
315 | err = 0; |
316 | ||
317 | err_unlock: | |
318 | mutex_unlock(&dp_mutex); | |
319 | rtnl_unlock(); | |
064af421 BP |
320 | return err; |
321 | } | |
322 | ||
58c342f6 BP |
323 | static void release_nbp(struct kobject *kobj) |
324 | { | |
325 | struct net_bridge_port *p = container_of(kobj, struct net_bridge_port, kobj); | |
326 | kfree(p); | |
327 | } | |
328 | ||
35f7605b | 329 | static struct kobj_type brport_ktype = { |
8fef8c71 | 330 | #ifdef CONFIG_SYSFS |
58c342f6 BP |
331 | .sysfs_ops = &brport_sysfs_ops, |
332 | #endif | |
333 | .release = release_nbp | |
334 | }; | |
335 | ||
064af421 BP |
336 | /* Called with RTNL lock and dp_mutex. */ |
337 | static int new_nbp(struct datapath *dp, struct net_device *dev, int port_no) | |
338 | { | |
339 | struct net_bridge_port *p; | |
340 | ||
341 | if (dev->br_port != NULL) | |
342 | return -EBUSY; | |
343 | ||
344 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
345 | if (!p) | |
346 | return -ENOMEM; | |
347 | ||
348 | dev_set_promiscuity(dev, 1); | |
349 | dev_hold(dev); | |
350 | p->port_no = port_no; | |
351 | p->dp = dp; | |
352 | p->dev = dev; | |
56fd8edf | 353 | atomic_set(&p->sflow_pool, 0); |
064af421 BP |
354 | if (!is_dp_dev(dev)) |
355 | rcu_assign_pointer(dev->br_port, p); | |
356 | else { | |
357 | /* It would make sense to assign dev->br_port here too, but | |
358 | * that causes packets received on internal ports to get caught | |
359 | * in dp_frame_hook(). In turn dp_frame_hook() can reject them | |
360 | * back to network stack, but that's a waste of time. */ | |
361 | } | |
2de32079 | 362 | dev_disable_lro(dev); |
064af421 BP |
363 | rcu_assign_pointer(dp->ports[port_no], p); |
364 | list_add_rcu(&p->node, &dp->port_list); | |
365 | dp->n_ports++; | |
366 | ||
58c342f6 BP |
367 | /* Initialize kobject for bridge. This will be added as |
368 | * /sys/class/net/<devname>/brport later, if sysfs is enabled. */ | |
58c342f6 | 369 | p->kobj.kset = NULL; |
58c342f6 BP |
370 | kobject_init(&p->kobj, &brport_ktype); |
371 | ||
064af421 BP |
372 | dp_ifinfo_notify(RTM_NEWLINK, p); |
373 | ||
374 | return 0; | |
375 | } | |
376 | ||
377 | static int add_port(int dp_idx, struct odp_port __user *portp) | |
378 | { | |
379 | struct net_device *dev; | |
380 | struct datapath *dp; | |
381 | struct odp_port port; | |
382 | int port_no; | |
383 | int err; | |
384 | ||
385 | err = -EFAULT; | |
386 | if (copy_from_user(&port, portp, sizeof port)) | |
387 | goto out; | |
388 | port.devname[IFNAMSIZ - 1] = '\0'; | |
064af421 BP |
389 | |
390 | rtnl_lock(); | |
391 | dp = get_dp_locked(dp_idx); | |
392 | err = -ENODEV; | |
393 | if (!dp) | |
394 | goto out_unlock_rtnl; | |
395 | ||
9ee3ae3e BP |
396 | for (port_no = 1; port_no < DP_MAX_PORTS; port_no++) |
397 | if (!dp->ports[port_no]) | |
398 | goto got_port_no; | |
3c71830a | 399 | err = -EFBIG; |
9ee3ae3e | 400 | goto out_unlock_dp; |
064af421 | 401 | |
9ee3ae3e | 402 | got_port_no: |
064af421 BP |
403 | if (!(port.flags & ODP_PORT_INTERNAL)) { |
404 | err = -ENODEV; | |
405 | dev = dev_get_by_name(&init_net, port.devname); | |
406 | if (!dev) | |
407 | goto out_unlock_dp; | |
408 | ||
409 | err = -EINVAL; | |
410 | if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER || | |
411 | is_dp_dev(dev)) | |
412 | goto out_put; | |
413 | } else { | |
414 | dev = dp_dev_create(dp, port.devname, port_no); | |
415 | err = PTR_ERR(dev); | |
416 | if (IS_ERR(dev)) | |
417 | goto out_unlock_dp; | |
418 | dev_hold(dev); | |
419 | } | |
420 | ||
421 | err = new_nbp(dp, dev, port_no); | |
422 | if (err) | |
423 | goto out_put; | |
424 | ||
a7786963 | 425 | set_dp_devs_mtu(dp, dev); |
2ba9026e | 426 | dp_sysfs_add_if(dp->ports[port_no]); |
064af421 | 427 | |
21755402 | 428 | err = __put_user(port_no, &portp->port); |
064af421 BP |
429 | |
430 | out_put: | |
431 | dev_put(dev); | |
432 | out_unlock_dp: | |
433 | mutex_unlock(&dp->mutex); | |
434 | out_unlock_rtnl: | |
435 | rtnl_unlock(); | |
436 | out: | |
437 | return err; | |
438 | } | |
439 | ||
72ca14c1 | 440 | int dp_del_port(struct net_bridge_port *p) |
064af421 BP |
441 | { |
442 | ASSERT_RTNL(); | |
443 | ||
2e7dd8ec | 444 | if (p->port_no != ODPP_LOCAL) |
0515ceb3 | 445 | dp_sysfs_del_if(p); |
064af421 BP |
446 | dp_ifinfo_notify(RTM_DELLINK, p); |
447 | ||
448 | p->dp->n_ports--; | |
449 | ||
450 | if (is_dp_dev(p->dev)) { | |
451 | /* Make sure that no packets arrive from now on, since | |
452 | * dp_dev_xmit() will try to find itself through | |
453 | * p->dp->ports[], and we're about to set that to null. */ | |
454 | netif_tx_disable(p->dev); | |
455 | } | |
456 | ||
457 | /* First drop references to device. */ | |
458 | dev_set_promiscuity(p->dev, -1); | |
459 | list_del_rcu(&p->node); | |
460 | rcu_assign_pointer(p->dp->ports[p->port_no], NULL); | |
461 | rcu_assign_pointer(p->dev->br_port, NULL); | |
462 | ||
463 | /* Then wait until no one is still using it, and destroy it. */ | |
464 | synchronize_rcu(); | |
465 | ||
58c342f6 | 466 | if (is_dp_dev(p->dev)) |
064af421 | 467 | dp_dev_destroy(p->dev); |
58c342f6 BP |
468 | dev_put(p->dev); |
469 | kobject_put(&p->kobj); | |
064af421 BP |
470 | |
471 | return 0; | |
472 | } | |
473 | ||
474 | static int del_port(int dp_idx, int port_no) | |
475 | { | |
064af421 BP |
476 | struct net_bridge_port *p; |
477 | struct datapath *dp; | |
478 | LIST_HEAD(dp_devs); | |
479 | int err; | |
480 | ||
481 | err = -EINVAL; | |
482 | if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL) | |
483 | goto out; | |
484 | ||
485 | rtnl_lock(); | |
486 | dp = get_dp_locked(dp_idx); | |
487 | err = -ENODEV; | |
488 | if (!dp) | |
489 | goto out_unlock_rtnl; | |
490 | ||
491 | p = dp->ports[port_no]; | |
492 | err = -ENOENT; | |
493 | if (!p) | |
494 | goto out_unlock_dp; | |
495 | ||
72ca14c1 | 496 | err = dp_del_port(p); |
064af421 BP |
497 | |
498 | out_unlock_dp: | |
499 | mutex_unlock(&dp->mutex); | |
500 | out_unlock_rtnl: | |
501 | rtnl_unlock(); | |
502 | out: | |
064af421 BP |
503 | return err; |
504 | } | |
505 | ||
506 | /* Must be called with rcu_read_lock. */ | |
507 | static void | |
508 | do_port_input(struct net_bridge_port *p, struct sk_buff *skb) | |
509 | { | |
2de32079 JG |
510 | /* LRO isn't suitable for bridging. We turn it off but make sure |
511 | * that it wasn't reactivated. */ | |
512 | if (skb_warn_if_lro(skb)) | |
513 | return; | |
514 | ||
064af421 BP |
515 | /* Make our own copy of the packet. Otherwise we will mangle the |
516 | * packet for anyone who came before us (e.g. tcpdump via AF_PACKET). | |
517 | * (No one comes after us, since we tell handle_bridge() that we took | |
518 | * the packet.) */ | |
519 | skb = skb_share_check(skb, GFP_ATOMIC); | |
520 | if (!skb) | |
521 | return; | |
522 | ||
523 | /* Push the Ethernet header back on. */ | |
524 | skb_push(skb, ETH_HLEN); | |
525 | skb_reset_mac_header(skb); | |
526 | dp_process_received_packet(skb, p); | |
527 | } | |
528 | ||
529 | /* Must be called with rcu_read_lock and with bottom-halves disabled. */ | |
530 | void dp_process_received_packet(struct sk_buff *skb, struct net_bridge_port *p) | |
531 | { | |
532 | struct datapath *dp = p->dp; | |
533 | struct dp_stats_percpu *stats; | |
534 | struct odp_flow_key key; | |
535 | struct sw_flow *flow; | |
536 | ||
537 | WARN_ON_ONCE(skb_shared(skb)); | |
064af421 | 538 | |
635c9298 | 539 | compute_ip_summed(skb, false); |
659586ef | 540 | OVS_CB(skb)->tun_id = 0; |
a063b0df | 541 | |
064af421 BP |
542 | /* BHs are off so we don't have to use get_cpu()/put_cpu() here. */ |
543 | stats = percpu_ptr(dp->stats_percpu, smp_processor_id()); | |
544 | ||
545 | if (flow_extract(skb, p ? p->port_no : ODPP_NONE, &key)) { | |
546 | if (dp->drop_frags) { | |
547 | kfree_skb(skb); | |
548 | stats->n_frags++; | |
549 | return; | |
550 | } | |
551 | } | |
552 | ||
553 | flow = dp_table_lookup(rcu_dereference(dp->table), &key); | |
554 | if (flow) { | |
555 | struct sw_flow_actions *acts = rcu_dereference(flow->sf_acts); | |
556 | flow_used(flow, skb); | |
557 | execute_actions(dp, skb, &key, acts->actions, acts->n_actions, | |
558 | GFP_ATOMIC); | |
559 | stats->n_hit++; | |
560 | } else { | |
561 | stats->n_missed++; | |
659586ef | 562 | dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id); |
064af421 BP |
563 | } |
564 | } | |
565 | ||
566 | /* | |
567 | * Used as br_handle_frame_hook. (Cannot run bridge at the same time, even on | |
568 | * different set of devices!) | |
569 | */ | |
570 | #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22) | |
571 | /* Called with rcu_read_lock and bottom-halves disabled. */ | |
572 | static struct sk_buff *dp_frame_hook(struct net_bridge_port *p, | |
573 | struct sk_buff *skb) | |
574 | { | |
575 | do_port_input(p, skb); | |
576 | return NULL; | |
577 | } | |
578 | #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0) | |
579 | /* Called with rcu_read_lock and bottom-halves disabled. */ | |
580 | static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb) | |
581 | { | |
582 | do_port_input(p, *pskb); | |
583 | return 1; | |
584 | } | |
585 | #else | |
586 | #error | |
587 | #endif | |
588 | ||
f7fed000 | 589 | #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID) |
8cdaca99 JG |
590 | /* This code is based on a skb_checksum_setup from net/dev/core.c from a |
591 | * combination of Lenny's 2.6.26 Xen kernel and Xen's | |
592 | * linux-2.6.18-92.1.10.el5.xs5.0.0.394.644. We can't call this function | |
593 | * directly because it isn't exported in all versions. */ | |
064af421 BP |
594 | static int skb_pull_up_to(struct sk_buff *skb, void *ptr) |
595 | { | |
596 | if (ptr < (void *)skb->tail) | |
597 | return 1; | |
598 | if (__pskb_pull_tail(skb, | |
599 | ptr - (void *)skb->data - skb_headlen(skb))) { | |
600 | return 1; | |
601 | } else { | |
602 | return 0; | |
603 | } | |
604 | } | |
605 | ||
b2f460c7 | 606 | int vswitch_skb_checksum_setup(struct sk_buff *skb) |
064af421 | 607 | { |
8cdaca99 JG |
608 | struct iphdr *iph; |
609 | unsigned char *th; | |
610 | int err = -EPROTO; | |
611 | __u16 csum_start, csum_offset; | |
612 | ||
613 | if (!skb->proto_csum_blank) | |
614 | return 0; | |
615 | ||
616 | if (skb->protocol != htons(ETH_P_IP)) | |
617 | goto out; | |
618 | ||
a063b0df | 619 | if (!skb_pull_up_to(skb, skb_network_header(skb) + sizeof(struct iphdr))) |
8cdaca99 JG |
620 | goto out; |
621 | ||
622 | iph = ip_hdr(skb); | |
623 | th = skb_network_header(skb) + 4 * iph->ihl; | |
624 | ||
625 | csum_start = th - skb->head; | |
626 | switch (iph->protocol) { | |
627 | case IPPROTO_TCP: | |
628 | csum_offset = offsetof(struct tcphdr, check); | |
629 | break; | |
630 | case IPPROTO_UDP: | |
631 | csum_offset = offsetof(struct udphdr, check); | |
632 | break; | |
633 | default: | |
634 | if (net_ratelimit()) | |
635 | printk(KERN_ERR "Attempting to checksum a non-" | |
636 | "TCP/UDP packet, dropping a protocol" | |
637 | " %d packet", iph->protocol); | |
638 | goto out; | |
064af421 | 639 | } |
8cdaca99 JG |
640 | |
641 | if (!skb_pull_up_to(skb, th + csum_offset + 2)) | |
642 | goto out; | |
643 | ||
644 | skb->ip_summed = CHECKSUM_PARTIAL; | |
645 | skb->proto_csum_blank = 0; | |
646 | ||
647 | #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22) | |
648 | skb->csum_start = csum_start; | |
649 | skb->csum_offset = csum_offset; | |
650 | #else | |
651 | skb_set_transport_header(skb, csum_start - skb_headroom(skb)); | |
652 | skb->csum = csum_offset; | |
653 | #endif | |
654 | ||
655 | err = 0; | |
656 | ||
064af421 | 657 | out: |
8cdaca99 | 658 | return err; |
064af421 | 659 | } |
53d3bbbc | 660 | #endif /* CONFIG_XEN && HAVE_PROTO_DATA_VALID */ |
064af421 | 661 | |
a6057323 JG |
662 | /* Types of checksums that we can receive (these all refer to L4 checksums): |
663 | * 1. CHECKSUM_NONE: Device that did not compute checksum, contains full | |
664 | * (though not verified) checksum in packet but not in skb->csum. Packets | |
665 | * from the bridge local port will also have this type. | |
666 | * 2. CHECKSUM_COMPLETE (CHECKSUM_HW): Good device that computes checksums, | |
667 | * also the GRE module. This is the same as CHECKSUM_NONE, except it has | |
668 | * a valid skb->csum. Importantly, both contain a full checksum (not | |
669 | * verified) in the packet itself. The only difference is that if the | |
670 | * packet gets to L4 processing on this machine (not in DomU) we won't | |
671 | * have to recompute the checksum to verify. Most hardware devices do not | |
672 | * produce packets with this type, even if they support receive checksum | |
673 | * offloading (they produce type #5). | |
674 | * 3. CHECKSUM_PARTIAL (CHECKSUM_HW): Packet without full checksum and needs to | |
675 | * be computed if it is sent off box. Unfortunately on earlier kernels, | |
676 | * this case is impossible to distinguish from #2, despite having opposite | |
677 | * meanings. Xen adds an extra field on earlier kernels (see #4) in order | |
678 | * to distinguish the different states. The only real user of this type | |
679 | * with bridging is Xen (on later kernels). | |
680 | * 4. CHECKSUM_UNNECESSARY (with proto_csum_blank true): This packet was | |
681 | * generated locally by a Xen DomU and has a partial checksum. If it is | |
682 | * handled on this machine (Dom0 or DomU), then the checksum will not be | |
7dab847a | 683 | * computed. If it goes off box, the checksum in the packet needs to be |
a6057323 JG |
684 | * completed. Calling skb_checksum_setup converts this to CHECKSUM_HW |
685 | * (CHECKSUM_PARTIAL) so that the checksum can be completed. In later | |
686 | * kernels, this combination is replaced with CHECKSUM_PARTIAL. | |
687 | * 5. CHECKSUM_UNNECESSARY (with proto_csum_blank false): Packet with a correct | |
688 | * full checksum or using a protocol without a checksum. skb->csum is | |
689 | * undefined. This is common from devices with receive checksum | |
690 | * offloading. This is somewhat similar to CHECKSUM_NONE, except that | |
691 | * nobody will try to verify the checksum with CHECKSUM_UNNECESSARY. | |
692 | * | |
693 | * Note that on earlier kernels, CHECKSUM_COMPLETE and CHECKSUM_PARTIAL are | |
694 | * both defined as CHECKSUM_HW. Normally the meaning of CHECKSUM_HW is clear | |
695 | * based on whether it is on the transmit or receive path. After the datapath | |
696 | * it will be intepreted as CHECKSUM_PARTIAL. If the packet already has a | |
697 | * checksum, we will panic. Since we can receive packets with checksums, we | |
698 | * assume that all CHECKSUM_HW packets have checksums and map them to | |
699 | * CHECKSUM_NONE, which has a similar meaning (the it is only different if the | |
700 | * packet is processed by the local IP stack, in which case it will need to | |
701 | * be reverified). If we receive a packet with CHECKSUM_HW that really means | |
702 | * CHECKSUM_PARTIAL, it will be sent with the wrong checksum. However, there | |
703 | * shouldn't be any devices that do this with bridging. | |
704 | * | |
705 | * The bridge has similar behavior and this function closely resembles | |
706 | * skb_forward_csum(). It is slightly different because we are only concerned | |
707 | * with bridging and not other types of forwarding and can get away with | |
708 | * slightly more optimal behavior.*/ | |
635c9298 JG |
709 | void |
710 | compute_ip_summed(struct sk_buff *skb, bool xmit) | |
a063b0df | 711 | { |
635c9298 JG |
712 | /* For our convenience these defines change repeatedly between kernel |
713 | * versions, so we can't just copy them over... */ | |
714 | switch (skb->ip_summed) { | |
715 | case CHECKSUM_NONE: | |
716 | OVS_CB(skb)->ip_summed = OVS_CSUM_NONE; | |
717 | break; | |
718 | case CHECKSUM_UNNECESSARY: | |
719 | OVS_CB(skb)->ip_summed = OVS_CSUM_UNNECESSARY; | |
720 | break; | |
a063b0df JG |
721 | #ifdef CHECKSUM_HW |
722 | /* In theory this could be either CHECKSUM_PARTIAL or CHECKSUM_COMPLETE. | |
723 | * However, we should only get CHECKSUM_PARTIAL packets from Xen, which | |
724 | * uses some special fields to represent this (see below). Since we | |
725 | * can only make one type work, pick the one that actually happens in | |
635c9298 JG |
726 | * practice. |
727 | * | |
728 | * The one exception to this is if we are on the transmit path | |
729 | * (basically after skb_checksum_setup() has been run) the type has | |
730 | * already been converted, so we should stay with that. */ | |
731 | case CHECKSUM_HW: | |
732 | if (!xmit) | |
733 | OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE; | |
734 | else | |
735 | OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL; | |
736 | ||
737 | break; | |
738 | #else | |
739 | case CHECKSUM_COMPLETE: | |
740 | OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE; | |
741 | break; | |
742 | case CHECKSUM_PARTIAL: | |
743 | OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL; | |
744 | break; | |
a063b0df | 745 | #endif |
635c9298 JG |
746 | default: |
747 | printk(KERN_ERR "openvswitch: unknown checksum type %d\n", | |
748 | skb->ip_summed); | |
749 | /* None seems the safest... */ | |
750 | OVS_CB(skb)->ip_summed = OVS_CSUM_NONE; | |
751 | } | |
752 | ||
a063b0df JG |
753 | #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID) |
754 | /* Xen has a special way of representing CHECKSUM_PARTIAL on older | |
635c9298 | 755 | * kernels. It should not be set on the transmit path though. */ |
a063b0df | 756 | if (skb->proto_csum_blank) |
635c9298 JG |
757 | OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL; |
758 | ||
759 | WARN_ON_ONCE(skb->proto_csum_blank && xmit); | |
a063b0df JG |
760 | #endif |
761 | } | |
762 | ||
a6057323 JG |
763 | void |
764 | forward_ip_summed(struct sk_buff *skb) | |
765 | { | |
766 | #ifdef CHECKSUM_HW | |
635c9298 | 767 | if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE) |
a6057323 JG |
768 | skb->ip_summed = CHECKSUM_NONE; |
769 | #endif | |
770 | } | |
771 | ||
cb5087ca BP |
772 | /* Append each packet in 'skb' list to 'queue'. There will be only one packet |
773 | * unless we broke up a GSO packet. */ | |
774 | static int | |
775 | queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue, | |
776 | int queue_no, u32 arg) | |
777 | { | |
778 | struct sk_buff *nskb; | |
779 | int port_no; | |
780 | int err; | |
781 | ||
782 | port_no = ODPP_LOCAL; | |
783 | if (skb->dev) { | |
784 | if (skb->dev->br_port) | |
785 | port_no = skb->dev->br_port->port_no; | |
786 | else if (is_dp_dev(skb->dev)) | |
787 | port_no = dp_dev_priv(skb->dev)->port_no; | |
788 | } | |
789 | ||
790 | do { | |
791 | struct odp_msg *header; | |
792 | ||
793 | nskb = skb->next; | |
794 | skb->next = NULL; | |
795 | ||
796 | /* If a checksum-deferred packet is forwarded to the | |
797 | * controller, correct the pointers and checksum. This happens | |
798 | * on a regular basis only on Xen, on which VMs can pass up | |
799 | * packets that do not have their checksum computed. | |
800 | */ | |
801 | err = vswitch_skb_checksum_setup(skb); | |
802 | if (err) | |
803 | goto err_kfree_skbs; | |
804 | #ifndef CHECKSUM_HW | |
805 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
806 | #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22) | |
807 | /* Until 2.6.22, the start of the transport header was | |
808 | * also the start of data to be checksummed. Linux | |
809 | * 2.6.22 introduced the csum_start field for this | |
810 | * purpose, but we should point the transport header to | |
811 | * it anyway for backward compatibility, as | |
812 | * dev_queue_xmit() does even in 2.6.28. */ | |
813 | skb_set_transport_header(skb, skb->csum_start - | |
814 | skb_headroom(skb)); | |
815 | #endif | |
816 | err = skb_checksum_help(skb); | |
817 | if (err) | |
818 | goto err_kfree_skbs; | |
819 | } | |
820 | #else | |
821 | if (skb->ip_summed == CHECKSUM_HW) { | |
822 | err = skb_checksum_help(skb, 0); | |
823 | if (err) | |
824 | goto err_kfree_skbs; | |
825 | } | |
826 | #endif | |
827 | ||
828 | err = skb_cow(skb, sizeof *header); | |
829 | if (err) | |
830 | goto err_kfree_skbs; | |
831 | ||
832 | header = (struct odp_msg*)__skb_push(skb, sizeof *header); | |
833 | header->type = queue_no; | |
834 | header->length = skb->len; | |
835 | header->port = port_no; | |
836 | header->reserved = 0; | |
837 | header->arg = arg; | |
838 | skb_queue_tail(queue, skb); | |
839 | ||
840 | skb = nskb; | |
841 | } while (skb); | |
842 | return 0; | |
843 | ||
844 | err_kfree_skbs: | |
845 | kfree_skb(skb); | |
846 | while ((skb = nskb) != NULL) { | |
847 | nskb = skb->next; | |
848 | kfree_skb(skb); | |
849 | } | |
850 | return err; | |
851 | } | |
852 | ||
064af421 BP |
853 | int |
854 | dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no, | |
855 | u32 arg) | |
856 | { | |
857 | struct dp_stats_percpu *stats; | |
858 | struct sk_buff_head *queue; | |
064af421 BP |
859 | int err; |
860 | ||
861 | WARN_ON_ONCE(skb_shared(skb)); | |
72b06300 | 862 | BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR); |
064af421 BP |
863 | queue = &dp->queues[queue_no]; |
864 | err = -ENOBUFS; | |
865 | if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN) | |
866 | goto err_kfree_skb; | |
867 | ||
a6057323 JG |
868 | forward_ip_summed(skb); |
869 | ||
064af421 BP |
870 | /* Break apart GSO packets into their component pieces. Otherwise |
871 | * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */ | |
872 | if (skb_is_gso(skb)) { | |
873 | struct sk_buff *nskb = skb_gso_segment(skb, 0); | |
874 | if (nskb) { | |
875 | kfree_skb(skb); | |
876 | skb = nskb; | |
877 | if (unlikely(IS_ERR(skb))) { | |
878 | err = PTR_ERR(skb); | |
879 | goto err; | |
880 | } | |
881 | } else { | |
882 | /* XXX This case might not be possible. It's hard to | |
883 | * tell from the skb_gso_segment() code and comment. */ | |
884 | } | |
885 | } | |
886 | ||
cb5087ca | 887 | err = queue_control_packets(skb, queue, queue_no, arg); |
064af421 | 888 | wake_up_interruptible(&dp->waitqueue); |
cb5087ca | 889 | return err; |
064af421 BP |
890 | |
891 | err_kfree_skb: | |
892 | kfree_skb(skb); | |
893 | err: | |
894 | stats = percpu_ptr(dp->stats_percpu, get_cpu()); | |
895 | stats->n_lost++; | |
896 | put_cpu(); | |
897 | ||
898 | return err; | |
899 | } | |
900 | ||
901 | static int flush_flows(struct datapath *dp) | |
902 | { | |
903 | dp->n_flows = 0; | |
904 | return dp_table_flush(dp); | |
905 | } | |
906 | ||
907 | static int validate_actions(const struct sw_flow_actions *actions) | |
908 | { | |
909 | unsigned int i; | |
910 | ||
911 | for (i = 0; i < actions->n_actions; i++) { | |
912 | const union odp_action *a = &actions->actions[i]; | |
913 | switch (a->type) { | |
914 | case ODPAT_OUTPUT: | |
915 | if (a->output.port >= DP_MAX_PORTS) | |
916 | return -EINVAL; | |
917 | break; | |
918 | ||
919 | case ODPAT_OUTPUT_GROUP: | |
920 | if (a->output_group.group >= DP_MAX_GROUPS) | |
921 | return -EINVAL; | |
922 | break; | |
923 | ||
924 | case ODPAT_SET_VLAN_VID: | |
925 | if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK)) | |
926 | return -EINVAL; | |
927 | break; | |
928 | ||
929 | case ODPAT_SET_VLAN_PCP: | |
a4fbb689 JT |
930 | if (a->vlan_pcp.vlan_pcp |
931 | & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT)) | |
064af421 BP |
932 | return -EINVAL; |
933 | break; | |
934 | ||
3c5f6de3 JG |
935 | case ODPAT_SET_NW_TOS: |
936 | if (a->nw_tos.nw_tos & INET_ECN_MASK) | |
937 | return -EINVAL; | |
938 | break; | |
939 | ||
064af421 BP |
940 | default: |
941 | if (a->type >= ODPAT_N_ACTIONS) | |
942 | return -EOPNOTSUPP; | |
943 | break; | |
944 | } | |
945 | } | |
946 | ||
947 | return 0; | |
948 | } | |
949 | ||
950 | static struct sw_flow_actions *get_actions(const struct odp_flow *flow) | |
951 | { | |
952 | struct sw_flow_actions *actions; | |
953 | int error; | |
954 | ||
955 | actions = flow_actions_alloc(flow->n_actions); | |
956 | error = PTR_ERR(actions); | |
957 | if (IS_ERR(actions)) | |
958 | goto error; | |
959 | ||
960 | error = -EFAULT; | |
961 | if (copy_from_user(actions->actions, flow->actions, | |
962 | flow->n_actions * sizeof(union odp_action))) | |
963 | goto error_free_actions; | |
964 | error = validate_actions(actions); | |
965 | if (error) | |
966 | goto error_free_actions; | |
967 | ||
968 | return actions; | |
969 | ||
970 | error_free_actions: | |
971 | kfree(actions); | |
972 | error: | |
973 | return ERR_PTR(error); | |
974 | } | |
975 | ||
976 | static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats) | |
977 | { | |
978 | if (flow->used.tv_sec) { | |
979 | stats->used_sec = flow->used.tv_sec; | |
980 | stats->used_nsec = flow->used.tv_nsec; | |
981 | } else { | |
982 | stats->used_sec = 0; | |
983 | stats->used_nsec = 0; | |
984 | } | |
985 | stats->n_packets = flow->packet_count; | |
986 | stats->n_bytes = flow->byte_count; | |
987 | stats->ip_tos = flow->ip_tos; | |
988 | stats->tcp_flags = flow->tcp_flags; | |
f1aa2072 | 989 | stats->error = 0; |
064af421 BP |
990 | } |
991 | ||
992 | static void clear_stats(struct sw_flow *flow) | |
993 | { | |
994 | flow->used.tv_sec = flow->used.tv_nsec = 0; | |
995 | flow->tcp_flags = 0; | |
996 | flow->ip_tos = 0; | |
997 | flow->packet_count = 0; | |
998 | flow->byte_count = 0; | |
999 | } | |
1000 | ||
1001 | static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp) | |
1002 | { | |
1003 | struct odp_flow_put uf; | |
6fa58f7a | 1004 | struct sw_flow *flow; |
064af421 BP |
1005 | struct dp_table *table; |
1006 | struct odp_flow_stats stats; | |
1007 | int error; | |
1008 | ||
1009 | error = -EFAULT; | |
1010 | if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put))) | |
1011 | goto error; | |
834377ea | 1012 | memset(uf.flow.key.reserved, 0, sizeof uf.flow.key.reserved); |
064af421 | 1013 | |
064af421 | 1014 | table = rcu_dereference(dp->table); |
6fa58f7a BP |
1015 | flow = dp_table_lookup(table, &uf.flow.key); |
1016 | if (!flow) { | |
1017 | /* No such flow. */ | |
064af421 BP |
1018 | struct sw_flow_actions *acts; |
1019 | ||
1020 | error = -ENOENT; | |
1021 | if (!(uf.flags & ODPPF_CREATE)) | |
1022 | goto error; | |
1023 | ||
1024 | /* Expand table, if necessary, to make room. */ | |
6fa58f7a BP |
1025 | if (dp->n_flows >= table->n_buckets) { |
1026 | error = -ENOSPC; | |
1027 | if (table->n_buckets >= DP_MAX_BUCKETS) | |
1028 | goto error; | |
1029 | ||
064af421 BP |
1030 | error = dp_table_expand(dp); |
1031 | if (error) | |
1032 | goto error; | |
6fa58f7a | 1033 | table = rcu_dereference(dp->table); |
064af421 BP |
1034 | } |
1035 | ||
1036 | /* Allocate flow. */ | |
1037 | error = -ENOMEM; | |
1038 | flow = kmem_cache_alloc(flow_cache, GFP_KERNEL); | |
1039 | if (flow == NULL) | |
1040 | goto error; | |
1041 | flow->key = uf.flow.key; | |
1042 | spin_lock_init(&flow->lock); | |
1043 | clear_stats(flow); | |
1044 | ||
1045 | /* Obtain actions. */ | |
1046 | acts = get_actions(&uf.flow); | |
1047 | error = PTR_ERR(acts); | |
1048 | if (IS_ERR(acts)) | |
1049 | goto error_free_flow; | |
1050 | rcu_assign_pointer(flow->sf_acts, acts); | |
1051 | ||
1052 | /* Put flow in bucket. */ | |
6fa58f7a BP |
1053 | error = dp_table_insert(table, flow); |
1054 | if (error) | |
1055 | goto error_free_flow_acts; | |
064af421 BP |
1056 | dp->n_flows++; |
1057 | memset(&stats, 0, sizeof(struct odp_flow_stats)); | |
1058 | } else { | |
1059 | /* We found a matching flow. */ | |
064af421 BP |
1060 | struct sw_flow_actions *old_acts, *new_acts; |
1061 | unsigned long int flags; | |
1062 | ||
1063 | /* Bail out if we're not allowed to modify an existing flow. */ | |
1064 | error = -EEXIST; | |
1065 | if (!(uf.flags & ODPPF_MODIFY)) | |
1066 | goto error; | |
1067 | ||
1068 | /* Swap actions. */ | |
1069 | new_acts = get_actions(&uf.flow); | |
1070 | error = PTR_ERR(new_acts); | |
1071 | if (IS_ERR(new_acts)) | |
1072 | goto error; | |
1073 | old_acts = rcu_dereference(flow->sf_acts); | |
1074 | if (old_acts->n_actions != new_acts->n_actions || | |
1075 | memcmp(old_acts->actions, new_acts->actions, | |
1076 | sizeof(union odp_action) * old_acts->n_actions)) { | |
1077 | rcu_assign_pointer(flow->sf_acts, new_acts); | |
1078 | flow_deferred_free_acts(old_acts); | |
1079 | } else { | |
1080 | kfree(new_acts); | |
1081 | } | |
1082 | ||
1083 | /* Fetch stats, then clear them if necessary. */ | |
1084 | spin_lock_irqsave(&flow->lock, flags); | |
1085 | get_stats(flow, &stats); | |
1086 | if (uf.flags & ODPPF_ZERO_STATS) | |
1087 | clear_stats(flow); | |
1088 | spin_unlock_irqrestore(&flow->lock, flags); | |
1089 | } | |
1090 | ||
1091 | /* Copy stats to userspace. */ | |
1092 | if (__copy_to_user(&ufp->flow.stats, &stats, | |
1093 | sizeof(struct odp_flow_stats))) | |
1094 | return -EFAULT; | |
1095 | return 0; | |
1096 | ||
6fa58f7a BP |
1097 | error_free_flow_acts: |
1098 | kfree(flow->sf_acts); | |
064af421 BP |
1099 | error_free_flow: |
1100 | kmem_cache_free(flow_cache, flow); | |
1101 | error: | |
1102 | return error; | |
1103 | } | |
1104 | ||
1105 | static int put_actions(const struct sw_flow *flow, struct odp_flow __user *ufp) | |
1106 | { | |
1107 | union odp_action __user *actions; | |
1108 | struct sw_flow_actions *sf_acts; | |
1109 | u32 n_actions; | |
1110 | ||
1111 | if (__get_user(actions, &ufp->actions) || | |
1112 | __get_user(n_actions, &ufp->n_actions)) | |
1113 | return -EFAULT; | |
1114 | ||
1115 | if (!n_actions) | |
1116 | return 0; | |
064af421 BP |
1117 | |
1118 | sf_acts = rcu_dereference(flow->sf_acts); | |
1119 | if (__put_user(sf_acts->n_actions, &ufp->n_actions) || | |
1120 | (actions && copy_to_user(actions, sf_acts->actions, | |
1121 | sizeof(union odp_action) * | |
1122 | min(sf_acts->n_actions, n_actions)))) | |
1123 | return -EFAULT; | |
1124 | ||
1125 | return 0; | |
1126 | } | |
1127 | ||
18fdbe16 JG |
1128 | static int answer_query(struct sw_flow *flow, u32 query_flags, |
1129 | struct odp_flow __user *ufp) | |
064af421 BP |
1130 | { |
1131 | struct odp_flow_stats stats; | |
1132 | unsigned long int flags; | |
1133 | ||
1134 | spin_lock_irqsave(&flow->lock, flags); | |
1135 | get_stats(flow, &stats); | |
18fdbe16 JG |
1136 | |
1137 | if (query_flags & ODPFF_ZERO_TCP_FLAGS) { | |
1138 | flow->tcp_flags = 0; | |
1139 | } | |
064af421 BP |
1140 | spin_unlock_irqrestore(&flow->lock, flags); |
1141 | ||
1142 | if (__copy_to_user(&ufp->stats, &stats, sizeof(struct odp_flow_stats))) | |
1143 | return -EFAULT; | |
1144 | return put_actions(flow, ufp); | |
1145 | } | |
1146 | ||
f1aa2072 | 1147 | static int del_flow(struct datapath *dp, struct odp_flow __user *ufp) |
064af421 BP |
1148 | { |
1149 | struct dp_table *table = rcu_dereference(dp->table); | |
1150 | struct odp_flow uf; | |
1151 | struct sw_flow *flow; | |
1152 | int error; | |
1153 | ||
1154 | error = -EFAULT; | |
1155 | if (copy_from_user(&uf, ufp, sizeof uf)) | |
1156 | goto error; | |
834377ea | 1157 | memset(uf.key.reserved, 0, sizeof uf.key.reserved); |
064af421 BP |
1158 | |
1159 | flow = dp_table_lookup(table, &uf.key); | |
1160 | error = -ENOENT; | |
1161 | if (!flow) | |
1162 | goto error; | |
1163 | ||
f1aa2072 BP |
1164 | /* XXX redundant lookup */ |
1165 | error = dp_table_delete(table, flow); | |
1166 | if (error) | |
1167 | goto error; | |
064af421 | 1168 | |
f1aa2072 BP |
1169 | /* XXX These statistics might lose a few packets, since other CPUs can |
1170 | * be using this flow. We used to synchronize_rcu() to make sure that | |
1171 | * we get completely accurate stats, but that blows our performance, | |
1172 | * badly. */ | |
1173 | dp->n_flows--; | |
d65349ea | 1174 | error = answer_query(flow, 0, ufp); |
f1aa2072 | 1175 | flow_deferred_free(flow); |
064af421 BP |
1176 | |
1177 | error: | |
1178 | return error; | |
1179 | } | |
1180 | ||
f1aa2072 | 1181 | static int query_flows(struct datapath *dp, const struct odp_flowvec *flowvec) |
064af421 BP |
1182 | { |
1183 | struct dp_table *table = rcu_dereference(dp->table); | |
1184 | int i; | |
1185 | for (i = 0; i < flowvec->n_flows; i++) { | |
1186 | struct __user odp_flow *ufp = &flowvec->flows[i]; | |
1187 | struct odp_flow uf; | |
1188 | struct sw_flow *flow; | |
1189 | int error; | |
1190 | ||
1191 | if (__copy_from_user(&uf, ufp, sizeof uf)) | |
1192 | return -EFAULT; | |
834377ea | 1193 | memset(uf.key.reserved, 0, sizeof uf.key.reserved); |
064af421 BP |
1194 | |
1195 | flow = dp_table_lookup(table, &uf.key); | |
1196 | if (!flow) | |
f1aa2072 | 1197 | error = __put_user(ENOENT, &ufp->stats.error); |
064af421 | 1198 | else |
d65349ea | 1199 | error = answer_query(flow, uf.flags, ufp); |
064af421 BP |
1200 | if (error) |
1201 | return -EFAULT; | |
1202 | } | |
1203 | return flowvec->n_flows; | |
1204 | } | |
1205 | ||
1206 | struct list_flows_cbdata { | |
1207 | struct odp_flow __user *uflows; | |
1208 | int n_flows; | |
1209 | int listed_flows; | |
1210 | }; | |
1211 | ||
1212 | static int list_flow(struct sw_flow *flow, void *cbdata_) | |
1213 | { | |
1214 | struct list_flows_cbdata *cbdata = cbdata_; | |
1215 | struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++]; | |
1216 | int error; | |
1217 | ||
1218 | if (__copy_to_user(&ufp->key, &flow->key, sizeof flow->key)) | |
1219 | return -EFAULT; | |
18fdbe16 | 1220 | error = answer_query(flow, 0, ufp); |
064af421 BP |
1221 | if (error) |
1222 | return error; | |
1223 | ||
1224 | if (cbdata->listed_flows >= cbdata->n_flows) | |
1225 | return cbdata->listed_flows; | |
1226 | return 0; | |
1227 | } | |
1228 | ||
1229 | static int list_flows(struct datapath *dp, const struct odp_flowvec *flowvec) | |
1230 | { | |
1231 | struct list_flows_cbdata cbdata; | |
1232 | int error; | |
1233 | ||
1234 | if (!flowvec->n_flows) | |
1235 | return 0; | |
1236 | ||
1237 | cbdata.uflows = flowvec->flows; | |
1238 | cbdata.n_flows = flowvec->n_flows; | |
1239 | cbdata.listed_flows = 0; | |
1240 | error = dp_table_foreach(rcu_dereference(dp->table), | |
1241 | list_flow, &cbdata); | |
1242 | return error ? error : cbdata.listed_flows; | |
1243 | } | |
1244 | ||
1245 | static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp, | |
1246 | int (*function)(struct datapath *, | |
1247 | const struct odp_flowvec *)) | |
1248 | { | |
1249 | struct odp_flowvec __user *uflowvec; | |
1250 | struct odp_flowvec flowvec; | |
1251 | int retval; | |
1252 | ||
1253 | uflowvec = (struct odp_flowvec __user *)argp; | |
1254 | if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) || | |
1255 | copy_from_user(&flowvec, uflowvec, sizeof flowvec)) | |
1256 | return -EFAULT; | |
1257 | ||
1258 | if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow)) | |
1259 | return -EINVAL; | |
1260 | ||
1261 | if (!access_ok(VERIFY_WRITE, flowvec.flows, | |
1262 | flowvec.n_flows * sizeof(struct odp_flow))) | |
1263 | return -EFAULT; | |
1264 | ||
1265 | retval = function(dp, &flowvec); | |
1266 | return (retval < 0 ? retval | |
1267 | : retval == flowvec.n_flows ? 0 | |
1268 | : __put_user(retval, &uflowvec->n_flows)); | |
1269 | } | |
1270 | ||
1271 | static int do_execute(struct datapath *dp, const struct odp_execute *executep) | |
1272 | { | |
1273 | struct odp_execute execute; | |
1274 | struct odp_flow_key key; | |
1275 | struct sk_buff *skb; | |
1276 | struct sw_flow_actions *actions; | |
a393b897 | 1277 | struct ethhdr *eth; |
064af421 BP |
1278 | int err; |
1279 | ||
1280 | err = -EFAULT; | |
1281 | if (copy_from_user(&execute, executep, sizeof execute)) | |
1282 | goto error; | |
1283 | ||
1284 | err = -EINVAL; | |
1285 | if (execute.length < ETH_HLEN || execute.length > 65535) | |
1286 | goto error; | |
1287 | ||
1288 | err = -ENOMEM; | |
1289 | actions = flow_actions_alloc(execute.n_actions); | |
1290 | if (!actions) | |
1291 | goto error; | |
1292 | ||
1293 | err = -EFAULT; | |
1294 | if (copy_from_user(actions->actions, execute.actions, | |
1295 | execute.n_actions * sizeof *execute.actions)) | |
1296 | goto error_free_actions; | |
1297 | ||
1298 | err = validate_actions(actions); | |
1299 | if (err) | |
1300 | goto error_free_actions; | |
1301 | ||
1302 | err = -ENOMEM; | |
1303 | skb = alloc_skb(execute.length, GFP_KERNEL); | |
1304 | if (!skb) | |
1305 | goto error_free_actions; | |
659586ef | 1306 | |
064af421 BP |
1307 | if (execute.in_port < DP_MAX_PORTS) { |
1308 | struct net_bridge_port *p = dp->ports[execute.in_port]; | |
1309 | if (p) | |
1310 | skb->dev = p->dev; | |
1311 | } | |
1312 | ||
1313 | err = -EFAULT; | |
1314 | if (copy_from_user(skb_put(skb, execute.length), execute.data, | |
1315 | execute.length)) | |
1316 | goto error_free_skb; | |
1317 | ||
a393b897 JP |
1318 | skb_reset_mac_header(skb); |
1319 | eth = eth_hdr(skb); | |
1320 | ||
de3f65ea JP |
1321 | /* Normally, setting the skb 'protocol' field would be handled by a |
1322 | * call to eth_type_trans(), but it assumes there's a sending | |
1323 | * device, which we may not have. */ | |
a393b897 JP |
1324 | if (ntohs(eth->h_proto) >= 1536) |
1325 | skb->protocol = eth->h_proto; | |
1326 | else | |
1327 | skb->protocol = htons(ETH_P_802_2); | |
1328 | ||
064af421 BP |
1329 | flow_extract(skb, execute.in_port, &key); |
1330 | err = execute_actions(dp, skb, &key, actions->actions, | |
1331 | actions->n_actions, GFP_KERNEL); | |
1332 | kfree(actions); | |
1333 | return err; | |
1334 | ||
1335 | error_free_skb: | |
1336 | kfree_skb(skb); | |
1337 | error_free_actions: | |
1338 | kfree(actions); | |
1339 | error: | |
1340 | return err; | |
1341 | } | |
1342 | ||
16190191 | 1343 | static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp) |
064af421 BP |
1344 | { |
1345 | struct odp_stats stats; | |
1346 | int i; | |
1347 | ||
1348 | stats.n_flows = dp->n_flows; | |
6fa58f7a BP |
1349 | stats.cur_capacity = rcu_dereference(dp->table)->n_buckets; |
1350 | stats.max_capacity = DP_MAX_BUCKETS; | |
064af421 BP |
1351 | stats.n_ports = dp->n_ports; |
1352 | stats.max_ports = DP_MAX_PORTS; | |
1353 | stats.max_groups = DP_MAX_GROUPS; | |
1354 | stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0; | |
1355 | for_each_possible_cpu(i) { | |
1356 | const struct dp_stats_percpu *s; | |
1357 | s = percpu_ptr(dp->stats_percpu, i); | |
1358 | stats.n_frags += s->n_frags; | |
1359 | stats.n_hit += s->n_hit; | |
1360 | stats.n_missed += s->n_missed; | |
1361 | stats.n_lost += s->n_lost; | |
1362 | } | |
1363 | stats.max_miss_queue = DP_MAX_QUEUE_LEN; | |
1364 | stats.max_action_queue = DP_MAX_QUEUE_LEN; | |
1365 | return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0; | |
1366 | } | |
1367 | ||
1dcf111b JP |
1368 | /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */ |
1369 | int dp_min_mtu(const struct datapath *dp) | |
1370 | { | |
1371 | struct net_bridge_port *p; | |
1372 | int mtu = 0; | |
1373 | ||
1374 | ASSERT_RTNL(); | |
1375 | ||
1376 | list_for_each_entry_rcu (p, &dp->port_list, node) { | |
1377 | struct net_device *dev = p->dev; | |
1378 | ||
1379 | /* Skip any internal ports, since that's what we're trying to | |
1380 | * set. */ | |
1381 | if (is_dp_dev(dev)) | |
1382 | continue; | |
1383 | ||
1384 | if (!mtu || dev->mtu < mtu) | |
1385 | mtu = dev->mtu; | |
1386 | } | |
1387 | ||
1388 | return mtu ? mtu : ETH_DATA_LEN; | |
1389 | } | |
1390 | ||
a7786963 JG |
1391 | /* Sets the MTU of all datapath devices to the minimum of the ports. 'dev' |
1392 | * is the device whose MTU may have changed. Must be called with RTNL lock | |
1393 | * and dp_mutex. */ | |
1394 | void set_dp_devs_mtu(const struct datapath *dp, struct net_device *dev) | |
1395 | { | |
1396 | struct net_bridge_port *p; | |
1397 | int mtu; | |
1398 | ||
1399 | ASSERT_RTNL(); | |
1400 | ||
1401 | if (is_dp_dev(dev)) | |
1402 | return; | |
1403 | ||
1404 | mtu = dp_min_mtu(dp); | |
1405 | ||
1406 | list_for_each_entry_rcu (p, &dp->port_list, node) { | |
1407 | struct net_device *br_dev = p->dev; | |
1408 | ||
1409 | if (is_dp_dev(br_dev)) | |
1410 | dev_set_mtu(br_dev, mtu); | |
1411 | } | |
1412 | } | |
1413 | ||
064af421 BP |
1414 | static int |
1415 | put_port(const struct net_bridge_port *p, struct odp_port __user *uop) | |
1416 | { | |
1417 | struct odp_port op; | |
1418 | memset(&op, 0, sizeof op); | |
1419 | strncpy(op.devname, p->dev->name, sizeof op.devname); | |
1420 | op.port = p->port_no; | |
1421 | op.flags = is_dp_dev(p->dev) ? ODP_PORT_INTERNAL : 0; | |
1422 | return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0; | |
1423 | } | |
1424 | ||
1425 | static int | |
1426 | query_port(struct datapath *dp, struct odp_port __user *uport) | |
1427 | { | |
1428 | struct odp_port port; | |
1429 | ||
1430 | if (copy_from_user(&port, uport, sizeof port)) | |
1431 | return -EFAULT; | |
1432 | if (port.devname[0]) { | |
1433 | struct net_bridge_port *p; | |
1434 | struct net_device *dev; | |
1435 | int err; | |
1436 | ||
1437 | port.devname[IFNAMSIZ - 1] = '\0'; | |
1438 | ||
1439 | dev = dev_get_by_name(&init_net, port.devname); | |
1440 | if (!dev) | |
1441 | return -ENODEV; | |
1442 | ||
1443 | p = dev->br_port; | |
1444 | if (!p && is_dp_dev(dev)) { | |
1445 | struct dp_dev *dp_dev = dp_dev_priv(dev); | |
1446 | if (dp_dev->dp == dp) | |
1447 | p = dp->ports[dp_dev->port_no]; | |
1448 | } | |
1449 | err = p && p->dp == dp ? put_port(p, uport) : -ENOENT; | |
1450 | dev_put(dev); | |
1451 | ||
1452 | return err; | |
1453 | } else { | |
1454 | if (port.port >= DP_MAX_PORTS) | |
1455 | return -EINVAL; | |
1456 | if (!dp->ports[port.port]) | |
1457 | return -ENOENT; | |
1458 | return put_port(dp->ports[port.port], uport); | |
1459 | } | |
1460 | } | |
1461 | ||
1462 | static int | |
1463 | list_ports(struct datapath *dp, struct odp_portvec __user *pvp) | |
1464 | { | |
1465 | struct odp_portvec pv; | |
1466 | struct net_bridge_port *p; | |
1467 | int idx; | |
1468 | ||
1469 | if (copy_from_user(&pv, pvp, sizeof pv)) | |
1470 | return -EFAULT; | |
1471 | ||
1472 | idx = 0; | |
1473 | if (pv.n_ports) { | |
1474 | list_for_each_entry_rcu (p, &dp->port_list, node) { | |
1475 | if (put_port(p, &pv.ports[idx])) | |
1476 | return -EFAULT; | |
1477 | if (idx++ >= pv.n_ports) | |
1478 | break; | |
1479 | } | |
1480 | } | |
f4ba4c4f | 1481 | return put_user(dp->n_ports, &pvp->n_ports); |
064af421 BP |
1482 | } |
1483 | ||
1484 | /* RCU callback for freeing a dp_port_group */ | |
1485 | static void free_port_group(struct rcu_head *rcu) | |
1486 | { | |
1487 | struct dp_port_group *g = container_of(rcu, struct dp_port_group, rcu); | |
1488 | kfree(g); | |
1489 | } | |
1490 | ||
1491 | static int | |
1492 | set_port_group(struct datapath *dp, const struct odp_port_group __user *upg) | |
1493 | { | |
1494 | struct odp_port_group pg; | |
1495 | struct dp_port_group *new_group, *old_group; | |
1496 | int error; | |
1497 | ||
1498 | error = -EFAULT; | |
1499 | if (copy_from_user(&pg, upg, sizeof pg)) | |
1500 | goto error; | |
1501 | ||
1502 | error = -EINVAL; | |
1503 | if (pg.n_ports > DP_MAX_PORTS || pg.group >= DP_MAX_GROUPS) | |
1504 | goto error; | |
1505 | ||
1506 | error = -ENOMEM; | |
1507 | new_group = kmalloc(sizeof *new_group + sizeof(u16) * pg.n_ports, | |
1508 | GFP_KERNEL); | |
1509 | if (!new_group) | |
1510 | goto error; | |
1511 | ||
1512 | new_group->n_ports = pg.n_ports; | |
1513 | error = -EFAULT; | |
1514 | if (copy_from_user(new_group->ports, pg.ports, | |
1515 | sizeof(u16) * pg.n_ports)) | |
1516 | goto error_free; | |
1517 | ||
1518 | old_group = rcu_dereference(dp->groups[pg.group]); | |
1519 | rcu_assign_pointer(dp->groups[pg.group], new_group); | |
1520 | if (old_group) | |
1521 | call_rcu(&old_group->rcu, free_port_group); | |
1522 | return 0; | |
1523 | ||
1524 | error_free: | |
1525 | kfree(new_group); | |
1526 | error: | |
1527 | return error; | |
1528 | } | |
1529 | ||
1530 | static int | |
1531 | get_port_group(struct datapath *dp, struct odp_port_group *upg) | |
1532 | { | |
1533 | struct odp_port_group pg; | |
1534 | struct dp_port_group *g; | |
1535 | u16 n_copy; | |
1536 | ||
1537 | if (copy_from_user(&pg, upg, sizeof pg)) | |
1538 | return -EFAULT; | |
1539 | ||
1540 | if (pg.group >= DP_MAX_GROUPS) | |
1541 | return -EINVAL; | |
1542 | ||
1543 | g = dp->groups[pg.group]; | |
1544 | n_copy = g ? min_t(int, g->n_ports, pg.n_ports) : 0; | |
1545 | if (n_copy && copy_to_user(pg.ports, g->ports, n_copy * sizeof(u16))) | |
1546 | return -EFAULT; | |
1547 | ||
1548 | if (put_user(g ? g->n_ports : 0, &upg->n_ports)) | |
1549 | return -EFAULT; | |
1550 | ||
1551 | return 0; | |
1552 | } | |
1553 | ||
7c40efc9 BP |
1554 | static int get_listen_mask(const struct file *f) |
1555 | { | |
1556 | return (long)f->private_data; | |
1557 | } | |
1558 | ||
1559 | static void set_listen_mask(struct file *f, int listen_mask) | |
1560 | { | |
1561 | f->private_data = (void*)(long)listen_mask; | |
1562 | } | |
1563 | ||
064af421 BP |
1564 | static long openvswitch_ioctl(struct file *f, unsigned int cmd, |
1565 | unsigned long argp) | |
1566 | { | |
1567 | int dp_idx = iminor(f->f_dentry->d_inode); | |
1568 | struct datapath *dp; | |
1569 | int drop_frags, listeners, port_no; | |
72b06300 | 1570 | unsigned int sflow_probability; |
064af421 BP |
1571 | int err; |
1572 | ||
1573 | /* Handle commands with special locking requirements up front. */ | |
1574 | switch (cmd) { | |
1575 | case ODP_DP_CREATE: | |
e86c8696 BP |
1576 | err = create_dp(dp_idx, (char __user *)argp); |
1577 | goto exit; | |
064af421 BP |
1578 | |
1579 | case ODP_DP_DESTROY: | |
e86c8696 BP |
1580 | err = destroy_dp(dp_idx); |
1581 | goto exit; | |
064af421 BP |
1582 | |
1583 | case ODP_PORT_ADD: | |
e86c8696 BP |
1584 | err = add_port(dp_idx, (struct odp_port __user *)argp); |
1585 | goto exit; | |
064af421 BP |
1586 | |
1587 | case ODP_PORT_DEL: | |
1588 | err = get_user(port_no, (int __user *)argp); | |
e86c8696 BP |
1589 | if (!err) |
1590 | err = del_port(dp_idx, port_no); | |
1591 | goto exit; | |
064af421 BP |
1592 | } |
1593 | ||
1594 | dp = get_dp_locked(dp_idx); | |
e86c8696 | 1595 | err = -ENODEV; |
064af421 | 1596 | if (!dp) |
e86c8696 | 1597 | goto exit; |
064af421 BP |
1598 | |
1599 | switch (cmd) { | |
1600 | case ODP_DP_STATS: | |
1601 | err = get_dp_stats(dp, (struct odp_stats __user *)argp); | |
1602 | break; | |
1603 | ||
1604 | case ODP_GET_DROP_FRAGS: | |
1605 | err = put_user(dp->drop_frags, (int __user *)argp); | |
1606 | break; | |
1607 | ||
1608 | case ODP_SET_DROP_FRAGS: | |
1609 | err = get_user(drop_frags, (int __user *)argp); | |
1610 | if (err) | |
1611 | break; | |
1612 | err = -EINVAL; | |
1613 | if (drop_frags != 0 && drop_frags != 1) | |
1614 | break; | |
1615 | dp->drop_frags = drop_frags; | |
1616 | err = 0; | |
1617 | break; | |
1618 | ||
1619 | case ODP_GET_LISTEN_MASK: | |
7c40efc9 | 1620 | err = put_user(get_listen_mask(f), (int __user *)argp); |
064af421 BP |
1621 | break; |
1622 | ||
1623 | case ODP_SET_LISTEN_MASK: | |
1624 | err = get_user(listeners, (int __user *)argp); | |
1625 | if (err) | |
1626 | break; | |
1627 | err = -EINVAL; | |
1628 | if (listeners & ~ODPL_ALL) | |
1629 | break; | |
1630 | err = 0; | |
7c40efc9 | 1631 | set_listen_mask(f, listeners); |
064af421 BP |
1632 | break; |
1633 | ||
72b06300 BP |
1634 | case ODP_GET_SFLOW_PROBABILITY: |
1635 | err = put_user(dp->sflow_probability, (unsigned int __user *)argp); | |
1636 | break; | |
1637 | ||
1638 | case ODP_SET_SFLOW_PROBABILITY: | |
1639 | err = get_user(sflow_probability, (unsigned int __user *)argp); | |
1640 | if (!err) | |
1641 | dp->sflow_probability = sflow_probability; | |
1642 | break; | |
1643 | ||
064af421 BP |
1644 | case ODP_PORT_QUERY: |
1645 | err = query_port(dp, (struct odp_port __user *)argp); | |
1646 | break; | |
1647 | ||
1648 | case ODP_PORT_LIST: | |
1649 | err = list_ports(dp, (struct odp_portvec __user *)argp); | |
1650 | break; | |
1651 | ||
1652 | case ODP_PORT_GROUP_SET: | |
1653 | err = set_port_group(dp, (struct odp_port_group __user *)argp); | |
1654 | break; | |
1655 | ||
1656 | case ODP_PORT_GROUP_GET: | |
1657 | err = get_port_group(dp, (struct odp_port_group __user *)argp); | |
1658 | break; | |
1659 | ||
1660 | case ODP_FLOW_FLUSH: | |
1661 | err = flush_flows(dp); | |
1662 | break; | |
1663 | ||
1664 | case ODP_FLOW_PUT: | |
1665 | err = put_flow(dp, (struct odp_flow_put __user *)argp); | |
1666 | break; | |
1667 | ||
1668 | case ODP_FLOW_DEL: | |
f1aa2072 | 1669 | err = del_flow(dp, (struct odp_flow __user *)argp); |
064af421 BP |
1670 | break; |
1671 | ||
f1aa2072 BP |
1672 | case ODP_FLOW_GET: |
1673 | err = do_flowvec_ioctl(dp, argp, query_flows); | |
064af421 BP |
1674 | break; |
1675 | ||
1676 | case ODP_FLOW_LIST: | |
1677 | err = do_flowvec_ioctl(dp, argp, list_flows); | |
1678 | break; | |
1679 | ||
1680 | case ODP_EXECUTE: | |
1681 | err = do_execute(dp, (struct odp_execute __user *)argp); | |
1682 | break; | |
1683 | ||
1684 | default: | |
1685 | err = -ENOIOCTLCMD; | |
1686 | break; | |
1687 | } | |
1688 | mutex_unlock(&dp->mutex); | |
e86c8696 | 1689 | exit: |
064af421 BP |
1690 | return err; |
1691 | } | |
1692 | ||
1693 | static int dp_has_packet_of_interest(struct datapath *dp, int listeners) | |
1694 | { | |
1695 | int i; | |
1696 | for (i = 0; i < DP_N_QUEUES; i++) { | |
1697 | if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i])) | |
1698 | return 1; | |
1699 | } | |
1700 | return 0; | |
1701 | } | |
1702 | ||
1703 | ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes, | |
1704 | loff_t *ppos) | |
1705 | { | |
6fba0d0b | 1706 | /* XXX is there sufficient synchronization here? */ |
7c40efc9 | 1707 | int listeners = get_listen_mask(f); |
064af421 BP |
1708 | int dp_idx = iminor(f->f_dentry->d_inode); |
1709 | struct datapath *dp = get_dp(dp_idx); | |
1710 | struct sk_buff *skb; | |
1711 | struct iovec __user iov; | |
1712 | size_t copy_bytes; | |
1713 | int retval; | |
1714 | ||
1715 | if (!dp) | |
1716 | return -ENODEV; | |
1717 | ||
1718 | if (nbytes == 0 || !listeners) | |
1719 | return 0; | |
1720 | ||
1721 | for (;;) { | |
1722 | int i; | |
1723 | ||
1724 | for (i = 0; i < DP_N_QUEUES; i++) { | |
1725 | if (listeners & (1 << i)) { | |
1726 | skb = skb_dequeue(&dp->queues[i]); | |
1727 | if (skb) | |
1728 | goto success; | |
1729 | } | |
1730 | } | |
1731 | ||
1732 | if (f->f_flags & O_NONBLOCK) { | |
1733 | retval = -EAGAIN; | |
1734 | goto error; | |
1735 | } | |
1736 | ||
1737 | wait_event_interruptible(dp->waitqueue, | |
1738 | dp_has_packet_of_interest(dp, | |
1739 | listeners)); | |
1740 | ||
1741 | if (signal_pending(current)) { | |
1742 | retval = -ERESTARTSYS; | |
1743 | goto error; | |
1744 | } | |
1745 | } | |
1746 | success: | |
1b378b99 | 1747 | copy_bytes = min_t(size_t, skb->len, nbytes); |
064af421 BP |
1748 | iov.iov_base = buf; |
1749 | iov.iov_len = copy_bytes; | |
1750 | retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len); | |
1751 | if (!retval) | |
1752 | retval = copy_bytes; | |
1753 | kfree_skb(skb); | |
1754 | ||
1755 | error: | |
1756 | return retval; | |
1757 | } | |
1758 | ||
1759 | static unsigned int openvswitch_poll(struct file *file, poll_table *wait) | |
1760 | { | |
6fba0d0b | 1761 | /* XXX is there sufficient synchronization here? */ |
064af421 BP |
1762 | int dp_idx = iminor(file->f_dentry->d_inode); |
1763 | struct datapath *dp = get_dp(dp_idx); | |
1764 | unsigned int mask; | |
1765 | ||
1766 | if (dp) { | |
1767 | mask = 0; | |
1768 | poll_wait(file, &dp->waitqueue, wait); | |
7c40efc9 | 1769 | if (dp_has_packet_of_interest(dp, get_listen_mask(file))) |
064af421 BP |
1770 | mask |= POLLIN | POLLRDNORM; |
1771 | } else { | |
1772 | mask = POLLIN | POLLRDNORM | POLLHUP; | |
1773 | } | |
1774 | return mask; | |
1775 | } | |
1776 | ||
1777 | struct file_operations openvswitch_fops = { | |
1778 | /* XXX .aio_read = openvswitch_aio_read, */ | |
1779 | .read = openvswitch_read, | |
1780 | .poll = openvswitch_poll, | |
1781 | .unlocked_ioctl = openvswitch_ioctl, | |
1782 | /* XXX .fasync = openvswitch_fasync, */ | |
1783 | }; | |
1784 | ||
1785 | static int major; | |
22d24ebf BP |
1786 | |
1787 | #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,27) | |
064af421 BP |
1788 | static struct llc_sap *dp_stp_sap; |
1789 | ||
1790 | static int dp_stp_rcv(struct sk_buff *skb, struct net_device *dev, | |
1791 | struct packet_type *pt, struct net_device *orig_dev) | |
1792 | { | |
1793 | /* We don't really care about STP packets, we just listen for them for | |
1794 | * mutual exclusion with the bridge module, so this just discards | |
1795 | * them. */ | |
1796 | kfree_skb(skb); | |
1797 | return 0; | |
1798 | } | |
1799 | ||
22d24ebf | 1800 | static int dp_avoid_bridge_init(void) |
064af421 | 1801 | { |
064af421 BP |
1802 | /* Register to receive STP packets because the bridge module also |
1803 | * attempts to do so. Since there can only be a single listener for a | |
1804 | * given protocol, this provides mutual exclusion against the bridge | |
1805 | * module, preventing both of them from being loaded at the same | |
1806 | * time. */ | |
1807 | dp_stp_sap = llc_sap_open(LLC_SAP_BSPAN, dp_stp_rcv); | |
1808 | if (!dp_stp_sap) { | |
1809 | printk(KERN_ERR "openvswitch: can't register sap for STP (probably the bridge module is loaded)\n"); | |
1810 | return -EADDRINUSE; | |
1811 | } | |
22d24ebf BP |
1812 | return 0; |
1813 | } | |
1814 | ||
1815 | static void dp_avoid_bridge_exit(void) | |
1816 | { | |
1817 | llc_sap_put(dp_stp_sap); | |
1818 | } | |
1819 | #else /* Linux 2.6.27 or later. */ | |
1820 | static int dp_avoid_bridge_init(void) | |
1821 | { | |
1822 | /* Linux 2.6.27 introduces a way for multiple clients to register for | |
1823 | * STP packets, which interferes with what we try to do above. | |
1824 | * Instead, just check whether there's a bridge hook defined. This is | |
1825 | * not as safe--the bridge module is willing to load over the top of | |
1826 | * us--but it provides a little bit of protection. */ | |
1827 | if (br_handle_frame_hook) { | |
1828 | printk(KERN_ERR "openvswitch: bridge module is loaded, cannot load over it\n"); | |
1829 | return -EADDRINUSE; | |
1830 | } | |
1831 | return 0; | |
1832 | } | |
1833 | ||
1834 | static void dp_avoid_bridge_exit(void) | |
1835 | { | |
1836 | /* Nothing to do. */ | |
1837 | } | |
1838 | #endif /* Linux 2.6.27 or later */ | |
1839 | ||
1840 | static int __init dp_init(void) | |
1841 | { | |
1842 | int err; | |
1843 | ||
1844 | printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR); | |
1845 | ||
1846 | err = dp_avoid_bridge_init(); | |
1847 | if (err) | |
1848 | return err; | |
064af421 BP |
1849 | |
1850 | err = flow_init(); | |
1851 | if (err) | |
1852 | goto error; | |
1853 | ||
1854 | err = register_netdevice_notifier(&dp_device_notifier); | |
1855 | if (err) | |
1856 | goto error_flow_exit; | |
1857 | ||
1858 | major = register_chrdev(0, "openvswitch", &openvswitch_fops); | |
1859 | if (err < 0) | |
1860 | goto error_unreg_notifier; | |
1861 | ||
1862 | /* Hook into callback used by the bridge to intercept packets. | |
1863 | * Parasites we are. */ | |
1864 | br_handle_frame_hook = dp_frame_hook; | |
1865 | ||
1866 | return 0; | |
1867 | ||
1868 | error_unreg_notifier: | |
1869 | unregister_netdevice_notifier(&dp_device_notifier); | |
1870 | error_flow_exit: | |
1871 | flow_exit(); | |
1872 | error: | |
1873 | return err; | |
1874 | } | |
1875 | ||
1876 | static void dp_cleanup(void) | |
1877 | { | |
1878 | rcu_barrier(); | |
1879 | unregister_chrdev(major, "openvswitch"); | |
1880 | unregister_netdevice_notifier(&dp_device_notifier); | |
1881 | flow_exit(); | |
1882 | br_handle_frame_hook = NULL; | |
22d24ebf | 1883 | dp_avoid_bridge_exit(); |
064af421 BP |
1884 | } |
1885 | ||
1886 | module_init(dp_init); | |
1887 | module_exit(dp_cleanup); | |
1888 | ||
1889 | MODULE_DESCRIPTION("Open vSwitch switching datapath"); | |
1890 | MODULE_LICENSE("GPL"); |