2 * Copyright (c) 2009, 2011 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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 #include <linux/kernel.h>
12 #include <asm/uaccess.h>
13 #include <linux/completion.h>
14 #include <linux/etherdevice.h>
15 #include <linux/if_bridge.h>
16 #include <linux/netdevice.h>
17 #include <linux/rtnetlink.h>
18 #include <net/genetlink.h>
20 #include "openvswitch/brcompat-netlink.h"
23 static struct genl_family brc_genl_family;
24 static struct genl_multicast_group brc_mc_group;
26 /* Time to wait for ovs-vswitchd to respond to a datapath action, in
28 #define BRC_TIMEOUT (HZ * 5)
30 /* Mutex to serialize ovs-brcompatd callbacks. (Some callbacks naturally hold
31 * br_ioctl_mutex, others hold rtnl_lock, but we can't take the former
32 * ourselves and we don't want to hold the latter over a potentially long
34 static DEFINE_MUTEX(brc_serial);
36 /* Userspace communication. */
37 static DEFINE_SPINLOCK(brc_lock); /* Ensure atomic access to these vars. */
38 static DECLARE_COMPLETION(brc_done); /* Userspace signaled operation done? */
39 static struct sk_buff *brc_reply; /* Reply from userspace. */
40 static u32 brc_seq; /* Sequence number for current op. */
42 static struct sk_buff *brc_send_command(struct sk_buff *, struct nlattr **attrs);
43 static int brc_send_simple_command(struct sk_buff *);
45 static struct sk_buff *brc_make_request(int op, const char *bridge,
48 struct sk_buff *skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
52 genlmsg_put(skb, 0, 0, &brc_genl_family, 0, op);
54 NLA_PUT_STRING(skb, BRC_GENL_A_DP_NAME, bridge);
56 NLA_PUT_STRING(skb, BRC_GENL_A_PORT_NAME, port);
65 static int brc_send_simple_command(struct sk_buff *request)
67 struct nlattr *attrs[BRC_GENL_A_MAX + 1];
68 struct sk_buff *reply;
71 reply = brc_send_command(request, attrs);
73 return PTR_ERR(reply);
75 error = nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
80 static int brc_add_del_bridge(char __user *uname, int add)
82 struct sk_buff *request;
85 if (!capable(CAP_NET_ADMIN))
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. */
152 static int old_deviceless(void __user *uarg)
154 unsigned long args[3];
156 if (copy_from_user(args, uarg, sizeof(args)))
160 case BRCTL_GET_BRIDGES:
161 return brc_get_bridges((int __user *)args[1], args[2]);
163 case BRCTL_ADD_BRIDGE:
164 return brc_add_del_bridge((void __user *)args[1], 1);
165 case BRCTL_DEL_BRIDGE:
166 return brc_add_del_bridge((void __user *)args[1], 0);
172 /* Called with the br_ioctl_mutex. */
174 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
175 brc_ioctl_deviceless_stub(unsigned int cmd, void __user *uarg)
177 brc_ioctl_deviceless_stub(struct net *net, unsigned int cmd, void __user *uarg)
183 return old_deviceless(uarg);
186 return brc_add_del_bridge(uarg, 1);
188 return brc_add_del_bridge(uarg, 0);
194 static int brc_add_del_port(struct net_device *dev, int port_ifindex, int add)
196 struct sk_buff *request;
197 struct net_device *port;
200 if (!capable(CAP_NET_ADMIN))
203 port = __dev_get_by_index(&init_net, port_ifindex);
207 /* Save name of dev and port because there's a race between the
208 * rtnl_unlock() and the brc_send_simple_command(). */
209 request = brc_make_request(add ? BRC_GENL_C_PORT_ADD : BRC_GENL_C_PORT_DEL,
210 dev->name, port->name);
215 err = brc_send_simple_command(request);
221 static int brc_get_bridge_info(struct net_device *dev,
222 struct __bridge_info __user *ub)
224 struct __bridge_info b;
226 memset(&b, 0, sizeof(struct __bridge_info));
228 /* First two bytes are the priority, which we should skip. This comes
229 * from struct bridge_id in br_private.h, which is unavailable to us.
231 memcpy((u8 *)&b.bridge_id + 2, dev->dev_addr, ETH_ALEN);
234 if (copy_to_user(ub, &b, sizeof(struct __bridge_info)))
240 static int brc_get_port_list(struct net_device *dev, int __user *uindices,
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. */
312 static int old_dev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
314 unsigned long args[4];
316 if (copy_from_user(args, rq->ifr_data, sizeof(args)))
321 return brc_add_del_port(dev, args[1], 1);
323 return brc_add_del_port(dev, args[1], 0);
325 case BRCTL_GET_BRIDGE_INFO:
326 return brc_get_bridge_info(dev, (struct __bridge_info __user *)args[1]);
328 case BRCTL_GET_PORT_LIST:
329 return brc_get_port_list(dev, (int __user *)args[1], args[2]);
331 case BRCTL_GET_FDB_ENTRIES:
332 return brc_get_fdb_entries(dev, (void __user *)args[1],
339 /* Called with the rtnl_lock. */
340 static int brc_dev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
346 err = old_dev_ioctl(dev, rq, cmd);
350 return brc_add_del_port(dev, rq->ifr_ifindex, 1);
352 return brc_add_del_port(dev, rq->ifr_ifindex, 0);
363 static struct genl_family brc_genl_family = {
364 .id = GENL_ID_GENERATE,
366 .name = BRC_GENL_FAMILY_NAME,
368 .maxattr = BRC_GENL_A_MAX,
371 static int brc_genl_query(struct sk_buff *skb, struct genl_info *info)
374 struct sk_buff *ans_skb;
377 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
381 data = genlmsg_put_reply(ans_skb, info, &brc_genl_family,
382 0, BRC_GENL_C_QUERY_MC);
387 NLA_PUT_U32(ans_skb, BRC_GENL_A_MC_GROUP, brc_mc_group.id);
389 genlmsg_end(ans_skb, data);
390 return genlmsg_reply(ans_skb, info);
398 /* Attribute policy: what each attribute may contain. */
399 static struct nla_policy brc_genl_policy[BRC_GENL_A_MAX + 1] = {
400 [BRC_GENL_A_ERR_CODE] = { .type = NLA_U32 },
401 [BRC_GENL_A_FDB_DATA] = { .type = NLA_UNSPEC },
404 static int brc_genl_dp_result(struct sk_buff *skb, struct genl_info *info)
406 unsigned long int flags;
409 if (!info->attrs[BRC_GENL_A_ERR_CODE])
412 skb = skb_clone(skb, GFP_KERNEL);
416 spin_lock_irqsave(&brc_lock, flags);
417 if (brc_seq == info->snd_seq) {
420 kfree_skb(brc_reply);
429 spin_unlock_irqrestore(&brc_lock, flags);
434 static struct genl_ops brc_genl_ops[] = {
435 { .cmd = BRC_GENL_C_QUERY_MC,
436 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
438 .doit = brc_genl_query,
440 { .cmd = BRC_GENL_C_DP_RESULT,
441 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
442 .policy = brc_genl_policy,
443 .doit = brc_genl_dp_result,
447 static struct sk_buff *brc_send_command(struct sk_buff *request,
448 struct nlattr **attrs)
450 unsigned long int flags;
451 struct sk_buff *reply;
454 mutex_lock(&brc_serial);
456 /* Increment sequence number first, so that we ignore any replies
457 * to stale requests. */
458 spin_lock_irqsave(&brc_lock, flags);
459 nlmsg_hdr(request)->nlmsg_seq = ++brc_seq;
460 INIT_COMPLETION(brc_done);
461 spin_unlock_irqrestore(&brc_lock, flags);
463 nlmsg_end(request, nlmsg_hdr(request));
466 error = genlmsg_multicast(request, 0, brc_mc_group.id, GFP_KERNEL);
470 /* Wait for reply. */
472 if (!wait_for_completion_timeout(&brc_done, BRC_TIMEOUT)) {
473 pr_warn("timed out waiting for userspace\n");
478 spin_lock_irqsave(&brc_lock, flags);
481 spin_unlock_irqrestore(&brc_lock, flags);
483 mutex_unlock(&brc_serial);
485 /* Re-parse message. Can't fail, since it parsed correctly once
487 error = nlmsg_parse(nlmsg_hdr(reply), GENL_HDRLEN,
488 attrs, BRC_GENL_A_MAX, brc_genl_policy);
494 mutex_unlock(&brc_serial);
495 return ERR_PTR(error);
498 static int __init brc_init(void)
502 printk("Open vSwitch Bridge Compatibility, built "__DATE__" "__TIME__"\n");
504 /* Set the bridge ioctl handler */
505 brioctl_set(brc_ioctl_deviceless_stub);
507 /* Set the openvswitch_mod device ioctl handler */
508 dp_ioctl_hook = brc_dev_ioctl;
510 /* Randomize the initial sequence number. This is not a security
511 * feature; it only helps avoid crossed wires between userspace and
512 * the kernel when the module is unloaded and reloaded. */
513 brc_seq = net_random();
515 /* Register generic netlink family to communicate changes to
517 err = genl_register_family_with_ops(&brc_genl_family,
518 brc_genl_ops, ARRAY_SIZE(brc_genl_ops));
522 strcpy(brc_mc_group.name, "brcompat");
523 err = genl_register_mc_group(&brc_genl_family, &brc_mc_group);
530 genl_unregister_family(&brc_genl_family);
532 pr_emerg("failed to install!\n");
536 static void brc_cleanup(void)
538 /* Unregister ioctl hooks */
539 dp_ioctl_hook = NULL;
542 genl_unregister_family(&brc_genl_family);
545 module_init(brc_init);
546 module_exit(brc_cleanup);
548 MODULE_DESCRIPTION("Open vSwitch bridge compatibility");
549 MODULE_AUTHOR("Nicira Networks");
550 MODULE_LICENSE("GPL");