2 * Copyright (c) 2009 Nicira Networks.
3 * Distributed under the terms of the GNU GPL version 2.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
9 #include <linux/kernel.h>
10 #include <asm/uaccess.h>
11 #include <linux/completion.h>
12 #include <linux/delay.h>
13 #include <linux/etherdevice.h>
14 #include <linux/if_bridge.h>
15 #include <linux/rculist.h>
16 #include <linux/netdevice.h>
17 #include <linux/rtnetlink.h>
18 #include <net/genetlink.h>
21 #include "openvswitch/brcompat-netlink.h"
22 #include "brc_procfs.h"
26 static struct genl_family brc_genl_family;
27 static struct genl_multicast_group brc_mc_group;
29 /* Time to wait for ovs-vswitchd to respond to a datapath action, in
31 #define BRC_TIMEOUT (HZ * 5)
33 /* Mutex to serialize ovs-brcompatd callbacks. (Some callbacks naturally hold
34 * br_ioctl_mutex, others hold rtnl_lock, but we can't take the former
35 * ourselves and we don't want to hold the latter over a potentially long
37 static DEFINE_MUTEX(brc_serial);
39 /* Userspace communication. */
40 static DEFINE_SPINLOCK(brc_lock); /* Ensure atomic access to these vars. */
41 static DECLARE_COMPLETION(brc_done); /* Userspace signaled operation done? */
42 static struct sk_buff *brc_reply; /* Reply from userspace. */
43 static u32 brc_seq; /* Sequence number for current op. */
45 static struct sk_buff *brc_send_command(struct sk_buff *, struct nlattr **attrs);
46 static int brc_send_simple_command(struct sk_buff *);
48 static struct sk_buff *
49 brc_make_request(int op, const char *bridge, const char *port)
51 struct sk_buff *skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
55 genlmsg_put(skb, 0, 0, &brc_genl_family, 0, op);
57 NLA_PUT_STRING(skb, BRC_GENL_A_DP_NAME, bridge);
59 NLA_PUT_STRING(skb, BRC_GENL_A_PORT_NAME, port);
68 static int brc_send_simple_command(struct sk_buff *request)
70 struct nlattr *attrs[BRC_GENL_A_MAX + 1];
71 struct sk_buff *reply;
74 reply = brc_send_command(request, attrs);
76 return PTR_ERR(reply);
78 error = nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
83 static int brc_add_del_bridge(char __user *uname, int add)
85 struct sk_buff *request;
88 if (copy_from_user(name, uname, IFNAMSIZ))
91 name[IFNAMSIZ - 1] = 0;
92 request = brc_make_request(add ? BRC_GENL_C_DP_ADD : BRC_GENL_C_DP_DEL,
97 return brc_send_simple_command(request);
100 static int brc_get_indices(int op, const char *br_name,
101 int __user *uindices, int n)
103 struct nlattr *attrs[BRC_GENL_A_MAX + 1];
104 struct sk_buff *request, *reply;
114 request = brc_make_request(op, br_name, NULL);
118 reply = brc_send_command(request, attrs);
119 ret = PTR_ERR(reply);
123 ret = -nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
128 if (!attrs[BRC_GENL_A_IFINDEXES])
131 len = nla_len(attrs[BRC_GENL_A_IFINDEXES]);
132 indices = nla_data(attrs[BRC_GENL_A_IFINDEXES]);
133 if (len % sizeof(int))
136 n = min_t(int, n, len / sizeof(int));
137 ret = copy_to_user(uindices, indices, n * sizeof(int)) ? -EFAULT : n;
145 /* Called with br_ioctl_mutex. */
146 static int brc_get_bridges(int __user *uindices, int n)
148 return brc_get_indices(BRC_GENL_C_GET_BRIDGES, NULL, uindices, n);
151 /* Legacy deviceless bridge ioctl's. Called with br_ioctl_mutex. */
153 old_deviceless(void __user *uarg)
155 unsigned long args[3];
157 if (copy_from_user(args, uarg, sizeof(args)))
161 case BRCTL_GET_BRIDGES:
162 return brc_get_bridges((int __user *)args[1], args[2]);
164 case BRCTL_ADD_BRIDGE:
165 return brc_add_del_bridge((void __user *)args[1], 1);
166 case BRCTL_DEL_BRIDGE:
167 return brc_add_del_bridge((void __user *)args[1], 0);
173 /* Called with the br_ioctl_mutex. */
175 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
176 brc_ioctl_deviceless_stub(unsigned int cmd, void __user *uarg)
178 brc_ioctl_deviceless_stub(struct net *net, unsigned int cmd, void __user *uarg)
184 return old_deviceless(uarg);
187 return brc_add_del_bridge(uarg, 1);
189 return brc_add_del_bridge(uarg, 0);
196 brc_add_del_port(struct net_device *dev, int port_ifindex, int add)
198 struct sk_buff *request;
199 struct net_device *port;
202 port = __dev_get_by_index(&init_net, port_ifindex);
206 /* Save name of dev and port because there's a race between the
207 * rtnl_unlock() and the brc_send_simple_command(). */
208 request = brc_make_request(add ? BRC_GENL_C_PORT_ADD : BRC_GENL_C_PORT_DEL,
209 dev->name, port->name);
214 err = brc_send_simple_command(request);
221 brc_get_bridge_info(struct net_device *dev, struct __bridge_info __user *ub)
223 struct __bridge_info b;
227 memset(&b, 0, sizeof(struct __bridge_info));
229 for (i=0; i<ETH_ALEN; i++)
230 id |= (u64)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
231 b.bridge_id = cpu_to_be64(id);
234 if (copy_to_user(ub, &b, sizeof(struct __bridge_info)))
241 brc_get_port_list(struct net_device *dev, int __user *uindices, int num)
246 retval = brc_get_indices(BRC_GENL_C_GET_PORTS, dev->name,
254 * Format up to a page worth of forwarding table entries
255 * userbuf -- where to copy result
256 * maxnum -- maximum number of entries desired
257 * (limited to a page for sanity)
258 * offset -- number of records to skip
260 static int brc_get_fdb_entries(struct net_device *dev, void __user *userbuf,
261 unsigned long maxnum, unsigned long offset)
263 struct nlattr *attrs[BRC_GENL_A_MAX + 1];
264 struct sk_buff *request, *reply;
268 /* Clamp size to PAGE_SIZE, test maxnum to avoid overflow */
269 if (maxnum > PAGE_SIZE/sizeof(struct __fdb_entry))
270 maxnum = PAGE_SIZE/sizeof(struct __fdb_entry);
272 request = brc_make_request(BRC_GENL_C_FDB_QUERY, dev->name, NULL);
275 NLA_PUT_U64(request, BRC_GENL_A_FDB_COUNT, maxnum);
276 NLA_PUT_U64(request, BRC_GENL_A_FDB_SKIP, offset);
279 reply = brc_send_command(request, attrs);
280 retval = PTR_ERR(reply);
284 retval = -nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
289 if (!attrs[BRC_GENL_A_FDB_DATA])
291 len = nla_len(attrs[BRC_GENL_A_FDB_DATA]);
292 if (len % sizeof(struct __fdb_entry) ||
293 len / sizeof(struct __fdb_entry) > maxnum)
296 retval = len / sizeof(struct __fdb_entry);
297 if (copy_to_user(userbuf, nla_data(attrs[BRC_GENL_A_FDB_DATA]), len))
311 /* Legacy ioctl's through SIOCDEVPRIVATE. Called with rtnl_lock. */
313 old_dev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
315 unsigned long args[4];
317 if (copy_from_user(args, rq->ifr_data, sizeof(args)))
322 return brc_add_del_port(dev, args[1], 1);
324 return brc_add_del_port(dev, args[1], 0);
326 case BRCTL_GET_BRIDGE_INFO:
327 return brc_get_bridge_info(dev, (struct __bridge_info __user *)args[1]);
329 case BRCTL_GET_PORT_LIST:
330 return brc_get_port_list(dev, (int __user *)args[1], args[2]);
332 case BRCTL_GET_FDB_ENTRIES:
333 return brc_get_fdb_entries(dev, (void __user *)args[1],
340 /* Called with the rtnl_lock. */
342 brc_dev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
348 err = old_dev_ioctl(dev, rq, cmd);
352 return brc_add_del_port(dev, rq->ifr_ifindex, 1);
354 return brc_add_del_port(dev, rq->ifr_ifindex, 0);
365 static struct genl_family brc_genl_family = {
366 .id = GENL_ID_GENERATE,
368 .name = BRC_GENL_FAMILY_NAME,
370 .maxattr = BRC_GENL_A_MAX,
373 static int brc_genl_query(struct sk_buff *skb, struct genl_info *info)
376 struct sk_buff *ans_skb;
379 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
383 data = genlmsg_put_reply(ans_skb, info, &brc_genl_family,
384 0, BRC_GENL_C_QUERY_MC);
389 NLA_PUT_U32(ans_skb, BRC_GENL_A_MC_GROUP, brc_mc_group.id);
391 genlmsg_end(ans_skb, data);
392 return genlmsg_reply(ans_skb, info);
400 static struct genl_ops brc_genl_ops_query_dp = {
401 .cmd = BRC_GENL_C_QUERY_MC,
402 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
404 .doit = brc_genl_query,
408 /* Attribute policy: what each attribute may contain. */
409 static struct nla_policy brc_genl_policy[BRC_GENL_A_MAX + 1] = {
410 [BRC_GENL_A_ERR_CODE] = { .type = NLA_U32 },
412 [BRC_GENL_A_PROC_DIR] = { .type = NLA_NUL_STRING },
413 [BRC_GENL_A_PROC_NAME] = { .type = NLA_NUL_STRING },
414 [BRC_GENL_A_PROC_DATA] = { .type = NLA_NUL_STRING },
416 [BRC_GENL_A_FDB_DATA] = { .type = NLA_UNSPEC },
420 brc_genl_dp_result(struct sk_buff *skb, struct genl_info *info)
422 unsigned long int flags;
425 if (!info->attrs[BRC_GENL_A_ERR_CODE])
428 skb = skb_clone(skb, GFP_KERNEL);
432 spin_lock_irqsave(&brc_lock, flags);
433 if (brc_seq == info->snd_seq) {
437 kfree_skb(brc_reply);
446 spin_unlock_irqrestore(&brc_lock, flags);
451 static struct genl_ops brc_genl_ops_dp_result = {
452 .cmd = BRC_GENL_C_DP_RESULT,
453 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
454 .policy = brc_genl_policy,
455 .doit = brc_genl_dp_result,
459 static struct genl_ops brc_genl_ops_set_proc = {
460 .cmd = BRC_GENL_C_SET_PROC,
461 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
462 .policy = brc_genl_policy,
463 .doit = brc_genl_set_proc,
467 static struct sk_buff *brc_send_command(struct sk_buff *request, struct nlattr **attrs)
469 unsigned long int flags;
470 struct sk_buff *reply;
473 mutex_lock(&brc_serial);
475 /* Increment sequence number first, so that we ignore any replies
476 * to stale requests. */
477 spin_lock_irqsave(&brc_lock, flags);
478 nlmsg_hdr(request)->nlmsg_seq = ++brc_seq;
479 INIT_COMPLETION(brc_done);
480 spin_unlock_irqrestore(&brc_lock, flags);
482 nlmsg_end(request, nlmsg_hdr(request));
485 error = genlmsg_multicast(request, 0, brc_mc_group.id, GFP_KERNEL);
489 /* Wait for reply. */
491 if (!wait_for_completion_timeout(&brc_done, BRC_TIMEOUT))
495 spin_lock_irqsave(&brc_lock, flags);
498 spin_unlock_irqrestore(&brc_lock, flags);
500 mutex_unlock(&brc_serial);
502 /* Re-parse message. Can't fail, since it parsed correctly once
504 error = nlmsg_parse(nlmsg_hdr(reply), GENL_HDRLEN,
505 attrs, BRC_GENL_A_MAX, brc_genl_policy);
511 mutex_unlock(&brc_serial);
512 return ERR_PTR(error);
516 __init brc_init(void)
521 printk("Open vSwitch Bridge Compatibility, built "__DATE__" "__TIME__"\n");
524 for (i=0; i<ODP_MAX; i++) {
527 printk(KERN_EMERG "brcompat: no datapaths may exist!\n");
533 /* Set the bridge ioctl handler */
534 brioctl_set(brc_ioctl_deviceless_stub);
536 /* Set the openvswitch_mod device ioctl handler */
537 dp_ioctl_hook = brc_dev_ioctl;
539 /* Randomize the initial sequence number. This is not a security
540 * feature; it only helps avoid crossed wires between userspace and
541 * the kernel when the module is unloaded and reloaded. */
542 brc_seq = net_random();
544 /* Register generic netlink family to communicate changes to
546 err = genl_register_family(&brc_genl_family);
550 err = genl_register_ops(&brc_genl_family, &brc_genl_ops_query_dp);
554 err = genl_register_ops(&brc_genl_family, &brc_genl_ops_dp_result);
558 err = genl_register_ops(&brc_genl_family, &brc_genl_ops_set_proc);
562 strcpy(brc_mc_group.name, "brcompat");
563 err = genl_register_mc_group(&brc_genl_family, &brc_mc_group);
570 genl_unregister_family(&brc_genl_family);
572 printk(KERN_EMERG "brcompat: failed to install!");
579 /* Unregister ioctl hooks */
580 dp_ioctl_hook = NULL;
583 genl_unregister_family(&brc_genl_family);
587 module_init(brc_init);
588 module_exit(brc_cleanup);
590 MODULE_DESCRIPTION("Open vSwitch bridge compatibility");
591 MODULE_AUTHOR("Nicira Networks");
592 MODULE_LICENSE("GPL");