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/etherdevice.h>
13 #include <linux/if_bridge.h>
14 #include <linux/netdevice.h>
15 #include <linux/rtnetlink.h>
16 #include <net/genetlink.h>
19 #include "openvswitch/brcompat-netlink.h"
20 #include "brc_procfs.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 (copy_from_user(name, uname, IFNAMSIZ))
88 name[IFNAMSIZ - 1] = 0;
89 request = brc_make_request(add ? BRC_GENL_C_DP_ADD : BRC_GENL_C_DP_DEL,
94 return brc_send_simple_command(request);
97 static int brc_get_indices(int op, const char *br_name,
98 int __user *uindices, int n)
100 struct nlattr *attrs[BRC_GENL_A_MAX + 1];
101 struct sk_buff *request, *reply;
111 request = brc_make_request(op, br_name, NULL);
115 reply = brc_send_command(request, attrs);
116 ret = PTR_ERR(reply);
120 ret = -nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
125 if (!attrs[BRC_GENL_A_IFINDEXES])
128 len = nla_len(attrs[BRC_GENL_A_IFINDEXES]);
129 indices = nla_data(attrs[BRC_GENL_A_IFINDEXES]);
130 if (len % sizeof(int))
133 n = min_t(int, n, len / sizeof(int));
134 ret = copy_to_user(uindices, indices, n * sizeof(int)) ? -EFAULT : n;
142 /* Called with br_ioctl_mutex. */
143 static int brc_get_bridges(int __user *uindices, int n)
145 return brc_get_indices(BRC_GENL_C_GET_BRIDGES, NULL, uindices, n);
148 /* Legacy deviceless bridge ioctl's. Called with br_ioctl_mutex. */
149 static int old_deviceless(void __user *uarg)
151 unsigned long args[3];
153 if (copy_from_user(args, uarg, sizeof(args)))
157 case BRCTL_GET_BRIDGES:
158 return brc_get_bridges((int __user *)args[1], args[2]);
160 case BRCTL_ADD_BRIDGE:
161 return brc_add_del_bridge((void __user *)args[1], 1);
162 case BRCTL_DEL_BRIDGE:
163 return brc_add_del_bridge((void __user *)args[1], 0);
169 /* Called with the br_ioctl_mutex. */
171 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
172 brc_ioctl_deviceless_stub(unsigned int cmd, void __user *uarg)
174 brc_ioctl_deviceless_stub(struct net *net, unsigned int cmd, void __user *uarg)
180 return old_deviceless(uarg);
183 return brc_add_del_bridge(uarg, 1);
185 return brc_add_del_bridge(uarg, 0);
191 static int brc_add_del_port(struct net_device *dev, int port_ifindex, int add)
193 struct sk_buff *request;
194 struct net_device *port;
197 port = __dev_get_by_index(&init_net, port_ifindex);
201 /* Save name of dev and port because there's a race between the
202 * rtnl_unlock() and the brc_send_simple_command(). */
203 request = brc_make_request(add ? BRC_GENL_C_PORT_ADD : BRC_GENL_C_PORT_DEL,
204 dev->name, port->name);
209 err = brc_send_simple_command(request);
215 static int brc_get_bridge_info(struct net_device *dev,
216 struct __bridge_info __user *ub)
218 struct __bridge_info b;
222 memset(&b, 0, sizeof(struct __bridge_info));
224 for (i=0; i<ETH_ALEN; i++)
225 id |= (u64)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
226 b.bridge_id = cpu_to_be64(id);
229 if (copy_to_user(ub, &b, sizeof(struct __bridge_info)))
235 static int brc_get_port_list(struct net_device *dev, int __user *uindices,
241 retval = brc_get_indices(BRC_GENL_C_GET_PORTS, dev->name,
249 * Format up to a page worth of forwarding table entries
250 * userbuf -- where to copy result
251 * maxnum -- maximum number of entries desired
252 * (limited to a page for sanity)
253 * offset -- number of records to skip
255 static int brc_get_fdb_entries(struct net_device *dev, void __user *userbuf,
256 unsigned long maxnum, unsigned long offset)
258 struct nlattr *attrs[BRC_GENL_A_MAX + 1];
259 struct sk_buff *request, *reply;
263 /* Clamp size to PAGE_SIZE, test maxnum to avoid overflow */
264 if (maxnum > PAGE_SIZE/sizeof(struct __fdb_entry))
265 maxnum = PAGE_SIZE/sizeof(struct __fdb_entry);
267 request = brc_make_request(BRC_GENL_C_FDB_QUERY, dev->name, NULL);
270 NLA_PUT_U64(request, BRC_GENL_A_FDB_COUNT, maxnum);
271 NLA_PUT_U64(request, BRC_GENL_A_FDB_SKIP, offset);
274 reply = brc_send_command(request, attrs);
275 retval = PTR_ERR(reply);
279 retval = -nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
284 if (!attrs[BRC_GENL_A_FDB_DATA])
286 len = nla_len(attrs[BRC_GENL_A_FDB_DATA]);
287 if (len % sizeof(struct __fdb_entry) ||
288 len / sizeof(struct __fdb_entry) > maxnum)
291 retval = len / sizeof(struct __fdb_entry);
292 if (copy_to_user(userbuf, nla_data(attrs[BRC_GENL_A_FDB_DATA]), len))
306 /* Legacy ioctl's through SIOCDEVPRIVATE. Called with rtnl_lock. */
307 static int old_dev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
309 unsigned long args[4];
311 if (copy_from_user(args, rq->ifr_data, sizeof(args)))
316 return brc_add_del_port(dev, args[1], 1);
318 return brc_add_del_port(dev, args[1], 0);
320 case BRCTL_GET_BRIDGE_INFO:
321 return brc_get_bridge_info(dev, (struct __bridge_info __user *)args[1]);
323 case BRCTL_GET_PORT_LIST:
324 return brc_get_port_list(dev, (int __user *)args[1], args[2]);
326 case BRCTL_GET_FDB_ENTRIES:
327 return brc_get_fdb_entries(dev, (void __user *)args[1],
334 /* Called with the rtnl_lock. */
335 static int brc_dev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
341 err = old_dev_ioctl(dev, rq, cmd);
345 return brc_add_del_port(dev, rq->ifr_ifindex, 1);
347 return brc_add_del_port(dev, rq->ifr_ifindex, 0);
358 static struct genl_family brc_genl_family = {
359 .id = GENL_ID_GENERATE,
361 .name = BRC_GENL_FAMILY_NAME,
363 .maxattr = BRC_GENL_A_MAX,
366 static int brc_genl_query(struct sk_buff *skb, struct genl_info *info)
369 struct sk_buff *ans_skb;
372 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
376 data = genlmsg_put_reply(ans_skb, info, &brc_genl_family,
377 0, BRC_GENL_C_QUERY_MC);
382 NLA_PUT_U32(ans_skb, BRC_GENL_A_MC_GROUP, brc_mc_group.id);
384 genlmsg_end(ans_skb, data);
385 return genlmsg_reply(ans_skb, info);
393 static struct genl_ops brc_genl_ops_query_dp = {
394 .cmd = BRC_GENL_C_QUERY_MC,
395 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
397 .doit = brc_genl_query,
401 /* Attribute policy: what each attribute may contain. */
402 static struct nla_policy brc_genl_policy[BRC_GENL_A_MAX + 1] = {
403 [BRC_GENL_A_ERR_CODE] = { .type = NLA_U32 },
405 [BRC_GENL_A_PROC_DIR] = { .type = NLA_NUL_STRING },
406 [BRC_GENL_A_PROC_NAME] = { .type = NLA_NUL_STRING },
407 [BRC_GENL_A_PROC_DATA] = { .type = NLA_NUL_STRING },
409 [BRC_GENL_A_FDB_DATA] = { .type = NLA_UNSPEC },
412 static int brc_genl_dp_result(struct sk_buff *skb, struct genl_info *info)
414 unsigned long int flags;
417 if (!info->attrs[BRC_GENL_A_ERR_CODE])
420 skb = skb_clone(skb, GFP_KERNEL);
424 spin_lock_irqsave(&brc_lock, flags);
425 if (brc_seq == info->snd_seq) {
429 kfree_skb(brc_reply);
438 spin_unlock_irqrestore(&brc_lock, flags);
443 static struct genl_ops brc_genl_ops_dp_result = {
444 .cmd = BRC_GENL_C_DP_RESULT,
445 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
446 .policy = brc_genl_policy,
447 .doit = brc_genl_dp_result,
451 static struct genl_ops brc_genl_ops_set_proc = {
452 .cmd = BRC_GENL_C_SET_PROC,
453 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
454 .policy = brc_genl_policy,
455 .doit = brc_genl_set_proc,
459 static struct sk_buff *brc_send_command(struct sk_buff *request,
460 struct nlattr **attrs)
462 unsigned long int flags;
463 struct sk_buff *reply;
466 mutex_lock(&brc_serial);
468 /* Increment sequence number first, so that we ignore any replies
469 * to stale requests. */
470 spin_lock_irqsave(&brc_lock, flags);
471 nlmsg_hdr(request)->nlmsg_seq = ++brc_seq;
472 INIT_COMPLETION(brc_done);
473 spin_unlock_irqrestore(&brc_lock, flags);
475 nlmsg_end(request, nlmsg_hdr(request));
478 error = genlmsg_multicast(request, 0, brc_mc_group.id, GFP_KERNEL);
482 /* Wait for reply. */
484 if (!wait_for_completion_timeout(&brc_done, BRC_TIMEOUT)) {
485 printk(KERN_WARNING "brcompat: timed out waiting for userspace\n");
490 spin_lock_irqsave(&brc_lock, flags);
493 spin_unlock_irqrestore(&brc_lock, flags);
495 mutex_unlock(&brc_serial);
497 /* Re-parse message. Can't fail, since it parsed correctly once
499 error = nlmsg_parse(nlmsg_hdr(reply), GENL_HDRLEN,
500 attrs, BRC_GENL_A_MAX, brc_genl_policy);
506 mutex_unlock(&brc_serial);
507 return ERR_PTR(error);
510 static int __init brc_init(void)
514 printk("Open vSwitch Bridge Compatibility, built "__DATE__" "__TIME__"\n");
516 /* Set the bridge ioctl handler */
517 brioctl_set(brc_ioctl_deviceless_stub);
519 /* Set the openvswitch_mod device ioctl handler */
520 dp_ioctl_hook = brc_dev_ioctl;
522 /* Randomize the initial sequence number. This is not a security
523 * feature; it only helps avoid crossed wires between userspace and
524 * the kernel when the module is unloaded and reloaded. */
525 brc_seq = net_random();
527 /* Register generic netlink family to communicate changes to
529 err = genl_register_family(&brc_genl_family);
533 err = genl_register_ops(&brc_genl_family, &brc_genl_ops_query_dp);
537 err = genl_register_ops(&brc_genl_family, &brc_genl_ops_dp_result);
541 err = genl_register_ops(&brc_genl_family, &brc_genl_ops_set_proc);
545 strcpy(brc_mc_group.name, "brcompat");
546 err = genl_register_mc_group(&brc_genl_family, &brc_mc_group);
553 genl_unregister_family(&brc_genl_family);
555 printk(KERN_EMERG "brcompat: failed to install!");
559 static void brc_cleanup(void)
561 /* Unregister ioctl hooks */
562 dp_ioctl_hook = NULL;
565 genl_unregister_family(&brc_genl_family);
569 module_init(brc_init);
570 module_exit(brc_cleanup);
572 MODULE_DESCRIPTION("Open vSwitch bridge compatibility");
573 MODULE_AUTHOR("Nicira Networks");
574 MODULE_LICENSE("GPL");