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 #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>
21 #include "openvswitch/brcompat-netlink.h"
22 #include "brc_procfs.h"
25 static struct genl_family brc_genl_family;
26 static struct genl_multicast_group brc_mc_group;
28 /* Time to wait for ovs-vswitchd to respond to a datapath action, in
30 #define BRC_TIMEOUT (HZ * 5)
32 /* Mutex to serialize ovs-brcompatd callbacks. (Some callbacks naturally hold
33 * br_ioctl_mutex, others hold rtnl_lock, but we can't take the former
34 * ourselves and we don't want to hold the latter over a potentially long
36 static DEFINE_MUTEX(brc_serial);
38 /* Userspace communication. */
39 static DEFINE_SPINLOCK(brc_lock); /* Ensure atomic access to these vars. */
40 static DECLARE_COMPLETION(brc_done); /* Userspace signaled operation done? */
41 static struct sk_buff *brc_reply; /* Reply from userspace. */
42 static u32 brc_seq; /* Sequence number for current op. */
44 static struct sk_buff *brc_send_command(struct sk_buff *, struct nlattr **attrs);
45 static int brc_send_simple_command(struct sk_buff *);
47 static struct sk_buff *brc_make_request(int op, const char *bridge,
50 struct sk_buff *skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
54 genlmsg_put(skb, 0, 0, &brc_genl_family, 0, op);
56 NLA_PUT_STRING(skb, BRC_GENL_A_DP_NAME, bridge);
58 NLA_PUT_STRING(skb, BRC_GENL_A_PORT_NAME, port);
67 static int brc_send_simple_command(struct sk_buff *request)
69 struct nlattr *attrs[BRC_GENL_A_MAX + 1];
70 struct sk_buff *reply;
73 reply = brc_send_command(request, attrs);
75 return PTR_ERR(reply);
77 error = nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
82 static int brc_add_del_bridge(char __user *uname, int add)
84 struct sk_buff *request;
87 if (!capable(CAP_NET_ADMIN))
90 if (copy_from_user(name, uname, IFNAMSIZ))
93 name[IFNAMSIZ - 1] = 0;
94 request = brc_make_request(add ? BRC_GENL_C_DP_ADD : BRC_GENL_C_DP_DEL,
99 return brc_send_simple_command(request);
102 static int brc_get_indices(int op, const char *br_name,
103 int __user *uindices, int n)
105 struct nlattr *attrs[BRC_GENL_A_MAX + 1];
106 struct sk_buff *request, *reply;
116 request = brc_make_request(op, br_name, NULL);
120 reply = brc_send_command(request, attrs);
121 ret = PTR_ERR(reply);
125 ret = -nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
130 if (!attrs[BRC_GENL_A_IFINDEXES])
133 len = nla_len(attrs[BRC_GENL_A_IFINDEXES]);
134 indices = nla_data(attrs[BRC_GENL_A_IFINDEXES]);
135 if (len % sizeof(int))
138 n = min_t(int, n, len / sizeof(int));
139 ret = copy_to_user(uindices, indices, n * sizeof(int)) ? -EFAULT : n;
147 /* Called with br_ioctl_mutex. */
148 static int brc_get_bridges(int __user *uindices, int n)
150 return brc_get_indices(BRC_GENL_C_GET_BRIDGES, NULL, uindices, n);
153 /* Legacy deviceless bridge ioctl's. Called with br_ioctl_mutex. */
154 static int old_deviceless(void __user *uarg)
156 unsigned long args[3];
158 if (copy_from_user(args, uarg, sizeof(args)))
162 case BRCTL_GET_BRIDGES:
163 return brc_get_bridges((int __user *)args[1], args[2]);
165 case BRCTL_ADD_BRIDGE:
166 return brc_add_del_bridge((void __user *)args[1], 1);
167 case BRCTL_DEL_BRIDGE:
168 return brc_add_del_bridge((void __user *)args[1], 0);
174 /* Called with the br_ioctl_mutex. */
176 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
177 brc_ioctl_deviceless_stub(unsigned int cmd, void __user *uarg)
179 brc_ioctl_deviceless_stub(struct net *net, unsigned int cmd, void __user *uarg)
185 return old_deviceless(uarg);
188 return brc_add_del_bridge(uarg, 1);
190 return brc_add_del_bridge(uarg, 0);
196 static int brc_add_del_port(struct net_device *dev, int port_ifindex, int add)
198 struct sk_buff *request;
199 struct net_device *port;
202 if (!capable(CAP_NET_ADMIN))
205 port = __dev_get_by_index(&init_net, port_ifindex);
209 /* Save name of dev and port because there's a race between the
210 * rtnl_unlock() and the brc_send_simple_command(). */
211 request = brc_make_request(add ? BRC_GENL_C_PORT_ADD : BRC_GENL_C_PORT_DEL,
212 dev->name, port->name);
217 err = brc_send_simple_command(request);
223 static int brc_get_bridge_info(struct net_device *dev,
224 struct __bridge_info __user *ub)
226 struct __bridge_info b;
230 memset(&b, 0, sizeof(struct __bridge_info));
232 for (i=0; i<ETH_ALEN; i++)
233 id |= (u64)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
234 b.bridge_id = cpu_to_be64(id);
237 if (copy_to_user(ub, &b, sizeof(struct __bridge_info)))
243 static int brc_get_port_list(struct net_device *dev, int __user *uindices,
249 retval = brc_get_indices(BRC_GENL_C_GET_PORTS, dev->name,
257 * Format up to a page worth of forwarding table entries
258 * userbuf -- where to copy result
259 * maxnum -- maximum number of entries desired
260 * (limited to a page for sanity)
261 * offset -- number of records to skip
263 static int brc_get_fdb_entries(struct net_device *dev, void __user *userbuf,
264 unsigned long maxnum, unsigned long offset)
266 struct nlattr *attrs[BRC_GENL_A_MAX + 1];
267 struct sk_buff *request, *reply;
271 /* Clamp size to PAGE_SIZE, test maxnum to avoid overflow */
272 if (maxnum > PAGE_SIZE/sizeof(struct __fdb_entry))
273 maxnum = PAGE_SIZE/sizeof(struct __fdb_entry);
275 request = brc_make_request(BRC_GENL_C_FDB_QUERY, dev->name, NULL);
278 NLA_PUT_U64(request, BRC_GENL_A_FDB_COUNT, maxnum);
279 NLA_PUT_U64(request, BRC_GENL_A_FDB_SKIP, offset);
282 reply = brc_send_command(request, attrs);
283 retval = PTR_ERR(reply);
287 retval = -nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
292 if (!attrs[BRC_GENL_A_FDB_DATA])
294 len = nla_len(attrs[BRC_GENL_A_FDB_DATA]);
295 if (len % sizeof(struct __fdb_entry) ||
296 len / sizeof(struct __fdb_entry) > maxnum)
299 retval = len / sizeof(struct __fdb_entry);
300 if (copy_to_user(userbuf, nla_data(attrs[BRC_GENL_A_FDB_DATA]), len))
314 /* Legacy ioctl's through SIOCDEVPRIVATE. Called with rtnl_lock. */
315 static int old_dev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
317 unsigned long args[4];
319 if (copy_from_user(args, rq->ifr_data, sizeof(args)))
324 return brc_add_del_port(dev, args[1], 1);
326 return brc_add_del_port(dev, args[1], 0);
328 case BRCTL_GET_BRIDGE_INFO:
329 return brc_get_bridge_info(dev, (struct __bridge_info __user *)args[1]);
331 case BRCTL_GET_PORT_LIST:
332 return brc_get_port_list(dev, (int __user *)args[1], args[2]);
334 case BRCTL_GET_FDB_ENTRIES:
335 return brc_get_fdb_entries(dev, (void __user *)args[1],
342 /* Called with the rtnl_lock. */
343 static int brc_dev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
349 err = old_dev_ioctl(dev, rq, cmd);
353 return brc_add_del_port(dev, rq->ifr_ifindex, 1);
355 return brc_add_del_port(dev, rq->ifr_ifindex, 0);
366 static struct genl_family brc_genl_family = {
367 .id = GENL_ID_GENERATE,
369 .name = BRC_GENL_FAMILY_NAME,
371 .maxattr = BRC_GENL_A_MAX,
374 static int brc_genl_query(struct sk_buff *skb, struct genl_info *info)
377 struct sk_buff *ans_skb;
380 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
384 data = genlmsg_put_reply(ans_skb, info, &brc_genl_family,
385 0, BRC_GENL_C_QUERY_MC);
390 NLA_PUT_U32(ans_skb, BRC_GENL_A_MC_GROUP, brc_mc_group.id);
392 genlmsg_end(ans_skb, data);
393 return genlmsg_reply(ans_skb, info);
401 static struct genl_ops brc_genl_ops_query_dp = {
402 .cmd = BRC_GENL_C_QUERY_MC,
403 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
405 .doit = brc_genl_query,
409 /* Attribute policy: what each attribute may contain. */
410 static struct nla_policy brc_genl_policy[BRC_GENL_A_MAX + 1] = {
411 [BRC_GENL_A_ERR_CODE] = { .type = NLA_U32 },
413 [BRC_GENL_A_PROC_DIR] = { .type = NLA_NUL_STRING },
414 [BRC_GENL_A_PROC_NAME] = { .type = NLA_NUL_STRING },
415 [BRC_GENL_A_PROC_DATA] = { .type = NLA_NUL_STRING },
417 [BRC_GENL_A_FDB_DATA] = { .type = NLA_UNSPEC },
420 static int 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) {
436 kfree_skb(brc_reply);
445 spin_unlock_irqrestore(&brc_lock, flags);
450 static struct genl_ops brc_genl_ops_dp_result = {
451 .cmd = BRC_GENL_C_DP_RESULT,
452 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
453 .policy = brc_genl_policy,
454 .doit = brc_genl_dp_result,
458 static struct genl_ops brc_genl_ops_set_proc = {
459 .cmd = BRC_GENL_C_SET_PROC,
460 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
461 .policy = brc_genl_policy,
462 .doit = brc_genl_set_proc,
466 static struct sk_buff *brc_send_command(struct sk_buff *request,
467 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)) {
492 pr_warn("timed out waiting for userspace\n");
497 spin_lock_irqsave(&brc_lock, flags);
500 spin_unlock_irqrestore(&brc_lock, flags);
502 mutex_unlock(&brc_serial);
504 /* Re-parse message. Can't fail, since it parsed correctly once
506 error = nlmsg_parse(nlmsg_hdr(reply), GENL_HDRLEN,
507 attrs, BRC_GENL_A_MAX, brc_genl_policy);
513 mutex_unlock(&brc_serial);
514 return ERR_PTR(error);
517 static int __init brc_init(void)
521 printk("Open vSwitch Bridge Compatibility, built "__DATE__" "__TIME__"\n");
523 /* Set the bridge ioctl handler */
524 brioctl_set(brc_ioctl_deviceless_stub);
526 /* Set the openvswitch_mod device ioctl handler */
527 dp_ioctl_hook = brc_dev_ioctl;
529 /* Randomize the initial sequence number. This is not a security
530 * feature; it only helps avoid crossed wires between userspace and
531 * the kernel when the module is unloaded and reloaded. */
532 brc_seq = net_random();
534 /* Register generic netlink family to communicate changes to
536 err = genl_register_family(&brc_genl_family);
540 err = genl_register_ops(&brc_genl_family, &brc_genl_ops_query_dp);
544 err = genl_register_ops(&brc_genl_family, &brc_genl_ops_dp_result);
548 err = genl_register_ops(&brc_genl_family, &brc_genl_ops_set_proc);
552 strcpy(brc_mc_group.name, "brcompat");
553 err = genl_register_mc_group(&brc_genl_family, &brc_mc_group);
560 genl_unregister_family(&brc_genl_family);
562 pr_emerg("failed to install!\n");
566 static void brc_cleanup(void)
568 /* Unregister ioctl hooks */
569 dp_ioctl_hook = NULL;
572 genl_unregister_family(&brc_genl_family);
576 module_init(brc_init);
577 module_exit(brc_cleanup);
579 MODULE_DESCRIPTION("Open vSwitch bridge compatibility");
580 MODULE_AUTHOR("Nicira Networks");
581 MODULE_LICENSE("GPL");