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 (copy_from_user(name, uname, IFNAMSIZ))
90 name[IFNAMSIZ - 1] = 0;
91 request = brc_make_request(add ? BRC_GENL_C_DP_ADD : BRC_GENL_C_DP_DEL,
96 return brc_send_simple_command(request);
99 static int brc_get_indices(int op, const char *br_name,
100 int __user *uindices, int n)
102 struct nlattr *attrs[BRC_GENL_A_MAX + 1];
103 struct sk_buff *request, *reply;
113 request = brc_make_request(op, br_name, NULL);
117 reply = brc_send_command(request, attrs);
118 ret = PTR_ERR(reply);
122 ret = -nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
127 if (!attrs[BRC_GENL_A_IFINDEXES])
130 len = nla_len(attrs[BRC_GENL_A_IFINDEXES]);
131 indices = nla_data(attrs[BRC_GENL_A_IFINDEXES]);
132 if (len % sizeof(int))
135 n = min_t(int, n, len / sizeof(int));
136 ret = copy_to_user(uindices, indices, n * sizeof(int)) ? -EFAULT : n;
144 /* Called with br_ioctl_mutex. */
145 static int brc_get_bridges(int __user *uindices, int n)
147 return brc_get_indices(BRC_GENL_C_GET_BRIDGES, NULL, uindices, n);
150 /* Legacy deviceless bridge ioctl's. Called with br_ioctl_mutex. */
151 static int old_deviceless(void __user *uarg)
153 unsigned long args[3];
155 if (copy_from_user(args, uarg, sizeof(args)))
159 case BRCTL_GET_BRIDGES:
160 return brc_get_bridges((int __user *)args[1], args[2]);
162 case BRCTL_ADD_BRIDGE:
163 return brc_add_del_bridge((void __user *)args[1], 1);
164 case BRCTL_DEL_BRIDGE:
165 return brc_add_del_bridge((void __user *)args[1], 0);
171 /* Called with the br_ioctl_mutex. */
173 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
174 brc_ioctl_deviceless_stub(unsigned int cmd, void __user *uarg)
176 brc_ioctl_deviceless_stub(struct net *net, unsigned int cmd, void __user *uarg)
182 return old_deviceless(uarg);
185 return brc_add_del_bridge(uarg, 1);
187 return brc_add_del_bridge(uarg, 0);
193 static int brc_add_del_port(struct net_device *dev, int port_ifindex, int add)
195 struct sk_buff *request;
196 struct net_device *port;
199 port = __dev_get_by_index(&init_net, port_ifindex);
203 /* Save name of dev and port because there's a race between the
204 * rtnl_unlock() and the brc_send_simple_command(). */
205 request = brc_make_request(add ? BRC_GENL_C_PORT_ADD : BRC_GENL_C_PORT_DEL,
206 dev->name, port->name);
211 err = brc_send_simple_command(request);
217 static int brc_get_bridge_info(struct net_device *dev,
218 struct __bridge_info __user *ub)
220 struct __bridge_info b;
224 memset(&b, 0, sizeof(struct __bridge_info));
226 for (i=0; i<ETH_ALEN; i++)
227 id |= (u64)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
228 b.bridge_id = cpu_to_be64(id);
231 if (copy_to_user(ub, &b, sizeof(struct __bridge_info)))
237 static int brc_get_port_list(struct net_device *dev, int __user *uindices,
243 retval = brc_get_indices(BRC_GENL_C_GET_PORTS, dev->name,
251 * Format up to a page worth of forwarding table entries
252 * userbuf -- where to copy result
253 * maxnum -- maximum number of entries desired
254 * (limited to a page for sanity)
255 * offset -- number of records to skip
257 static int brc_get_fdb_entries(struct net_device *dev, void __user *userbuf,
258 unsigned long maxnum, unsigned long offset)
260 struct nlattr *attrs[BRC_GENL_A_MAX + 1];
261 struct sk_buff *request, *reply;
265 /* Clamp size to PAGE_SIZE, test maxnum to avoid overflow */
266 if (maxnum > PAGE_SIZE/sizeof(struct __fdb_entry))
267 maxnum = PAGE_SIZE/sizeof(struct __fdb_entry);
269 request = brc_make_request(BRC_GENL_C_FDB_QUERY, dev->name, NULL);
272 NLA_PUT_U64(request, BRC_GENL_A_FDB_COUNT, maxnum);
273 NLA_PUT_U64(request, BRC_GENL_A_FDB_SKIP, offset);
276 reply = brc_send_command(request, attrs);
277 retval = PTR_ERR(reply);
281 retval = -nla_get_u32(attrs[BRC_GENL_A_ERR_CODE]);
286 if (!attrs[BRC_GENL_A_FDB_DATA])
288 len = nla_len(attrs[BRC_GENL_A_FDB_DATA]);
289 if (len % sizeof(struct __fdb_entry) ||
290 len / sizeof(struct __fdb_entry) > maxnum)
293 retval = len / sizeof(struct __fdb_entry);
294 if (copy_to_user(userbuf, nla_data(attrs[BRC_GENL_A_FDB_DATA]), len))
308 /* Legacy ioctl's through SIOCDEVPRIVATE. Called with rtnl_lock. */
309 static int old_dev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
311 unsigned long args[4];
313 if (copy_from_user(args, rq->ifr_data, sizeof(args)))
318 return brc_add_del_port(dev, args[1], 1);
320 return brc_add_del_port(dev, args[1], 0);
322 case BRCTL_GET_BRIDGE_INFO:
323 return brc_get_bridge_info(dev, (struct __bridge_info __user *)args[1]);
325 case BRCTL_GET_PORT_LIST:
326 return brc_get_port_list(dev, (int __user *)args[1], args[2]);
328 case BRCTL_GET_FDB_ENTRIES:
329 return brc_get_fdb_entries(dev, (void __user *)args[1],
336 /* Called with the rtnl_lock. */
337 static int brc_dev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
343 err = old_dev_ioctl(dev, rq, cmd);
347 return brc_add_del_port(dev, rq->ifr_ifindex, 1);
349 return brc_add_del_port(dev, rq->ifr_ifindex, 0);
360 static struct genl_family brc_genl_family = {
361 .id = GENL_ID_GENERATE,
363 .name = BRC_GENL_FAMILY_NAME,
365 .maxattr = BRC_GENL_A_MAX,
368 static int brc_genl_query(struct sk_buff *skb, struct genl_info *info)
371 struct sk_buff *ans_skb;
374 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
378 data = genlmsg_put_reply(ans_skb, info, &brc_genl_family,
379 0, BRC_GENL_C_QUERY_MC);
384 NLA_PUT_U32(ans_skb, BRC_GENL_A_MC_GROUP, brc_mc_group.id);
386 genlmsg_end(ans_skb, data);
387 return genlmsg_reply(ans_skb, info);
395 static struct genl_ops brc_genl_ops_query_dp = {
396 .cmd = BRC_GENL_C_QUERY_MC,
397 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
399 .doit = brc_genl_query,
403 /* Attribute policy: what each attribute may contain. */
404 static struct nla_policy brc_genl_policy[BRC_GENL_A_MAX + 1] = {
405 [BRC_GENL_A_ERR_CODE] = { .type = NLA_U32 },
407 [BRC_GENL_A_PROC_DIR] = { .type = NLA_NUL_STRING },
408 [BRC_GENL_A_PROC_NAME] = { .type = NLA_NUL_STRING },
409 [BRC_GENL_A_PROC_DATA] = { .type = NLA_NUL_STRING },
411 [BRC_GENL_A_FDB_DATA] = { .type = NLA_UNSPEC },
414 static int brc_genl_dp_result(struct sk_buff *skb, struct genl_info *info)
416 unsigned long int flags;
419 if (!info->attrs[BRC_GENL_A_ERR_CODE])
422 skb = skb_clone(skb, GFP_KERNEL);
426 spin_lock_irqsave(&brc_lock, flags);
427 if (brc_seq == info->snd_seq) {
430 kfree_skb(brc_reply);
439 spin_unlock_irqrestore(&brc_lock, flags);
444 static struct genl_ops brc_genl_ops_dp_result = {
445 .cmd = BRC_GENL_C_DP_RESULT,
446 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
447 .policy = brc_genl_policy,
448 .doit = brc_genl_dp_result,
452 static struct genl_ops brc_genl_ops_set_proc = {
453 .cmd = BRC_GENL_C_SET_PROC,
454 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privelege. */
455 .policy = brc_genl_policy,
456 .doit = brc_genl_set_proc,
460 static struct sk_buff *brc_send_command(struct sk_buff *request,
461 struct nlattr **attrs)
463 unsigned long int flags;
464 struct sk_buff *reply;
467 mutex_lock(&brc_serial);
469 /* Increment sequence number first, so that we ignore any replies
470 * to stale requests. */
471 spin_lock_irqsave(&brc_lock, flags);
472 nlmsg_hdr(request)->nlmsg_seq = ++brc_seq;
473 INIT_COMPLETION(brc_done);
474 spin_unlock_irqrestore(&brc_lock, flags);
476 nlmsg_end(request, nlmsg_hdr(request));
479 error = genlmsg_multicast(request, 0, brc_mc_group.id, GFP_KERNEL);
483 /* Wait for reply. */
485 if (!wait_for_completion_timeout(&brc_done, BRC_TIMEOUT)) {
486 pr_warn("timed out waiting for userspace\n");
491 spin_lock_irqsave(&brc_lock, flags);
494 spin_unlock_irqrestore(&brc_lock, flags);
496 mutex_unlock(&brc_serial);
498 /* Re-parse message. Can't fail, since it parsed correctly once
500 error = nlmsg_parse(nlmsg_hdr(reply), GENL_HDRLEN,
501 attrs, BRC_GENL_A_MAX, brc_genl_policy);
507 mutex_unlock(&brc_serial);
508 return ERR_PTR(error);
511 static int __init brc_init(void)
515 printk("Open vSwitch Bridge Compatibility, built "__DATE__" "__TIME__"\n");
517 /* Set the bridge ioctl handler */
518 brioctl_set(brc_ioctl_deviceless_stub);
520 /* Set the openvswitch_mod device ioctl handler */
521 dp_ioctl_hook = brc_dev_ioctl;
523 /* Randomize the initial sequence number. This is not a security
524 * feature; it only helps avoid crossed wires between userspace and
525 * the kernel when the module is unloaded and reloaded. */
526 brc_seq = net_random();
528 /* Register generic netlink family to communicate changes to
530 err = genl_register_family(&brc_genl_family);
534 err = genl_register_ops(&brc_genl_family, &brc_genl_ops_query_dp);
538 err = genl_register_ops(&brc_genl_family, &brc_genl_ops_dp_result);
542 err = genl_register_ops(&brc_genl_family, &brc_genl_ops_set_proc);
546 strcpy(brc_mc_group.name, "brcompat");
547 err = genl_register_mc_group(&brc_genl_family, &brc_mc_group);
554 genl_unregister_family(&brc_genl_family);
556 pr_emerg("failed to install!\n");
560 static void brc_cleanup(void)
562 /* Unregister ioctl hooks */
563 dp_ioctl_hook = NULL;
566 genl_unregister_family(&brc_genl_family);
570 module_init(brc_init);
571 module_exit(brc_cleanup);
573 MODULE_DESCRIPTION("Open vSwitch bridge compatibility");
574 MODULE_AUTHOR("Nicira Networks");
575 MODULE_LICENSE("GPL");