]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * NETLINK Kernel-user communication protocol. | |
3 | * | |
113aa838 | 4 | * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
5 | * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> |
6 | * | |
7 | * This program is free software; you can redistribute it and/or | |
8 | * modify it under the terms of the GNU General Public License | |
9 | * as published by the Free Software Foundation; either version | |
10 | * 2 of the License, or (at your option) any later version. | |
746fac4d | 11 | * |
1da177e4 LT |
12 | * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith |
13 | * added netlink_proto_exit | |
14 | * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br> | |
15 | * use nlk_sk, as sk->protinfo is on a diet 8) | |
4fdb3bb7 HW |
16 | * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org> |
17 | * - inc module use count of module that owns | |
18 | * the kernel socket in case userspace opens | |
19 | * socket of same protocol | |
20 | * - remove all module support, since netlink is | |
21 | * mandatory if CONFIG_NET=y these days | |
1da177e4 LT |
22 | */ |
23 | ||
1da177e4 LT |
24 | #include <linux/module.h> |
25 | ||
4fc268d2 | 26 | #include <linux/capability.h> |
1da177e4 LT |
27 | #include <linux/kernel.h> |
28 | #include <linux/init.h> | |
1da177e4 LT |
29 | #include <linux/signal.h> |
30 | #include <linux/sched.h> | |
31 | #include <linux/errno.h> | |
32 | #include <linux/string.h> | |
33 | #include <linux/stat.h> | |
34 | #include <linux/socket.h> | |
35 | #include <linux/un.h> | |
36 | #include <linux/fcntl.h> | |
37 | #include <linux/termios.h> | |
38 | #include <linux/sockios.h> | |
39 | #include <linux/net.h> | |
40 | #include <linux/fs.h> | |
41 | #include <linux/slab.h> | |
42 | #include <asm/uaccess.h> | |
43 | #include <linux/skbuff.h> | |
44 | #include <linux/netdevice.h> | |
45 | #include <linux/rtnetlink.h> | |
46 | #include <linux/proc_fs.h> | |
47 | #include <linux/seq_file.h> | |
1da177e4 LT |
48 | #include <linux/notifier.h> |
49 | #include <linux/security.h> | |
50 | #include <linux/jhash.h> | |
51 | #include <linux/jiffies.h> | |
52 | #include <linux/random.h> | |
53 | #include <linux/bitops.h> | |
54 | #include <linux/mm.h> | |
55 | #include <linux/types.h> | |
54e0f520 | 56 | #include <linux/audit.h> |
af65bdfc | 57 | #include <linux/mutex.h> |
54e0f520 | 58 | |
457c4cbc | 59 | #include <net/net_namespace.h> |
1da177e4 LT |
60 | #include <net/sock.h> |
61 | #include <net/scm.h> | |
82ace47a | 62 | #include <net/netlink.h> |
1da177e4 | 63 | |
f7fa9b10 | 64 | #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8) |
b4ff4f04 | 65 | #define NLGRPLONGS(x) (NLGRPSZ(x)/sizeof(unsigned long)) |
1da177e4 LT |
66 | |
67 | struct netlink_sock { | |
68 | /* struct sock has to be the first member of netlink_sock */ | |
69 | struct sock sk; | |
15e47304 EB |
70 | u32 portid; |
71 | u32 dst_portid; | |
d629b836 | 72 | u32 dst_group; |
f7fa9b10 PM |
73 | u32 flags; |
74 | u32 subscriptions; | |
75 | u32 ngroups; | |
76 | unsigned long *groups; | |
1da177e4 LT |
77 | unsigned long state; |
78 | wait_queue_head_t wait; | |
79 | struct netlink_callback *cb; | |
af65bdfc PM |
80 | struct mutex *cb_mutex; |
81 | struct mutex cb_def_mutex; | |
cd40b7d3 | 82 | void (*netlink_rcv)(struct sk_buff *skb); |
03292745 | 83 | void (*netlink_bind)(int group); |
77247bbb | 84 | struct module *module; |
1da177e4 LT |
85 | }; |
86 | ||
5c398dc8 ED |
87 | struct listeners { |
88 | struct rcu_head rcu; | |
89 | unsigned long masks[0]; | |
6c04bb18 JB |
90 | }; |
91 | ||
77247bbb | 92 | #define NETLINK_KERNEL_SOCKET 0x1 |
9a4595bc | 93 | #define NETLINK_RECV_PKTINFO 0x2 |
be0c22a4 | 94 | #define NETLINK_BROADCAST_SEND_ERROR 0x4 |
38938bfe | 95 | #define NETLINK_RECV_NO_ENOBUFS 0x8 |
77247bbb | 96 | |
1da177e4 LT |
97 | static inline struct netlink_sock *nlk_sk(struct sock *sk) |
98 | { | |
32b21e03 | 99 | return container_of(sk, struct netlink_sock, sk); |
1da177e4 LT |
100 | } |
101 | ||
035c4c16 | 102 | static inline int netlink_is_kernel(struct sock *sk) |
aed81560 DL |
103 | { |
104 | return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET; | |
105 | } | |
106 | ||
15e47304 | 107 | struct nl_portid_hash { |
658cb354 ED |
108 | struct hlist_head *table; |
109 | unsigned long rehash_time; | |
1da177e4 | 110 | |
658cb354 ED |
111 | unsigned int mask; |
112 | unsigned int shift; | |
1da177e4 | 113 | |
658cb354 ED |
114 | unsigned int entries; |
115 | unsigned int max_shift; | |
1da177e4 | 116 | |
658cb354 | 117 | u32 rnd; |
1da177e4 LT |
118 | }; |
119 | ||
120 | struct netlink_table { | |
15e47304 | 121 | struct nl_portid_hash hash; |
658cb354 ED |
122 | struct hlist_head mc_list; |
123 | struct listeners __rcu *listeners; | |
9785e10a | 124 | unsigned int flags; |
658cb354 ED |
125 | unsigned int groups; |
126 | struct mutex *cb_mutex; | |
127 | struct module *module; | |
03292745 | 128 | void (*bind)(int group); |
658cb354 | 129 | int registered; |
1da177e4 LT |
130 | }; |
131 | ||
132 | static struct netlink_table *nl_table; | |
133 | ||
134 | static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait); | |
135 | ||
136 | static int netlink_dump(struct sock *sk); | |
1da177e4 LT |
137 | |
138 | static DEFINE_RWLOCK(nl_table_lock); | |
139 | static atomic_t nl_table_users = ATOMIC_INIT(0); | |
140 | ||
6d772ac5 ED |
141 | #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock)); |
142 | ||
e041c683 | 143 | static ATOMIC_NOTIFIER_HEAD(netlink_chain); |
1da177e4 | 144 | |
b57ef81f | 145 | static inline u32 netlink_group_mask(u32 group) |
d629b836 PM |
146 | { |
147 | return group ? 1 << (group - 1) : 0; | |
148 | } | |
149 | ||
15e47304 | 150 | static inline struct hlist_head *nl_portid_hashfn(struct nl_portid_hash *hash, u32 portid) |
1da177e4 | 151 | { |
15e47304 | 152 | return &hash->table[jhash_1word(portid, hash->rnd) & hash->mask]; |
1da177e4 LT |
153 | } |
154 | ||
658cb354 ED |
155 | static void netlink_destroy_callback(struct netlink_callback *cb) |
156 | { | |
157 | kfree_skb(cb->skb); | |
158 | kfree(cb); | |
159 | } | |
160 | ||
bfb253c9 ED |
161 | static void netlink_consume_callback(struct netlink_callback *cb) |
162 | { | |
163 | consume_skb(cb->skb); | |
164 | kfree(cb); | |
165 | } | |
166 | ||
1da177e4 LT |
167 | static void netlink_sock_destruct(struct sock *sk) |
168 | { | |
3f660d66 HX |
169 | struct netlink_sock *nlk = nlk_sk(sk); |
170 | ||
3f660d66 HX |
171 | if (nlk->cb) { |
172 | if (nlk->cb->done) | |
173 | nlk->cb->done(nlk->cb); | |
6dc878a8 G |
174 | |
175 | module_put(nlk->cb->module); | |
3f660d66 HX |
176 | netlink_destroy_callback(nlk->cb); |
177 | } | |
178 | ||
1da177e4 LT |
179 | skb_queue_purge(&sk->sk_receive_queue); |
180 | ||
181 | if (!sock_flag(sk, SOCK_DEAD)) { | |
6ac552fd | 182 | printk(KERN_ERR "Freeing alive netlink socket %p\n", sk); |
1da177e4 LT |
183 | return; |
184 | } | |
547b792c IJ |
185 | |
186 | WARN_ON(atomic_read(&sk->sk_rmem_alloc)); | |
187 | WARN_ON(atomic_read(&sk->sk_wmem_alloc)); | |
188 | WARN_ON(nlk_sk(sk)->groups); | |
1da177e4 LT |
189 | } |
190 | ||
6ac552fd PM |
191 | /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on |
192 | * SMP. Look, when several writers sleep and reader wakes them up, all but one | |
1da177e4 LT |
193 | * immediately hit write lock and grab all the cpus. Exclusive sleep solves |
194 | * this, _but_ remember, it adds useless work on UP machines. | |
195 | */ | |
196 | ||
d136f1bd | 197 | void netlink_table_grab(void) |
9a429c49 | 198 | __acquires(nl_table_lock) |
1da177e4 | 199 | { |
d136f1bd JB |
200 | might_sleep(); |
201 | ||
6abd219c | 202 | write_lock_irq(&nl_table_lock); |
1da177e4 LT |
203 | |
204 | if (atomic_read(&nl_table_users)) { | |
205 | DECLARE_WAITQUEUE(wait, current); | |
206 | ||
207 | add_wait_queue_exclusive(&nl_table_wait, &wait); | |
6ac552fd | 208 | for (;;) { |
1da177e4 LT |
209 | set_current_state(TASK_UNINTERRUPTIBLE); |
210 | if (atomic_read(&nl_table_users) == 0) | |
211 | break; | |
6abd219c | 212 | write_unlock_irq(&nl_table_lock); |
1da177e4 | 213 | schedule(); |
6abd219c | 214 | write_lock_irq(&nl_table_lock); |
1da177e4 LT |
215 | } |
216 | ||
217 | __set_current_state(TASK_RUNNING); | |
218 | remove_wait_queue(&nl_table_wait, &wait); | |
219 | } | |
220 | } | |
221 | ||
d136f1bd | 222 | void netlink_table_ungrab(void) |
9a429c49 | 223 | __releases(nl_table_lock) |
1da177e4 | 224 | { |
6abd219c | 225 | write_unlock_irq(&nl_table_lock); |
1da177e4 LT |
226 | wake_up(&nl_table_wait); |
227 | } | |
228 | ||
6ac552fd | 229 | static inline void |
1da177e4 LT |
230 | netlink_lock_table(void) |
231 | { | |
232 | /* read_lock() synchronizes us to netlink_table_grab */ | |
233 | ||
234 | read_lock(&nl_table_lock); | |
235 | atomic_inc(&nl_table_users); | |
236 | read_unlock(&nl_table_lock); | |
237 | } | |
238 | ||
6ac552fd | 239 | static inline void |
1da177e4 LT |
240 | netlink_unlock_table(void) |
241 | { | |
242 | if (atomic_dec_and_test(&nl_table_users)) | |
243 | wake_up(&nl_table_wait); | |
244 | } | |
245 | ||
15e47304 | 246 | static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid) |
1da177e4 | 247 | { |
15e47304 | 248 | struct nl_portid_hash *hash = &nl_table[protocol].hash; |
1da177e4 LT |
249 | struct hlist_head *head; |
250 | struct sock *sk; | |
1da177e4 LT |
251 | |
252 | read_lock(&nl_table_lock); | |
15e47304 | 253 | head = nl_portid_hashfn(hash, portid); |
b67bfe0d | 254 | sk_for_each(sk, head) { |
15e47304 | 255 | if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->portid == portid)) { |
1da177e4 LT |
256 | sock_hold(sk); |
257 | goto found; | |
258 | } | |
259 | } | |
260 | sk = NULL; | |
261 | found: | |
262 | read_unlock(&nl_table_lock); | |
263 | return sk; | |
264 | } | |
265 | ||
15e47304 | 266 | static struct hlist_head *nl_portid_hash_zalloc(size_t size) |
1da177e4 LT |
267 | { |
268 | if (size <= PAGE_SIZE) | |
ea72912c | 269 | return kzalloc(size, GFP_ATOMIC); |
1da177e4 LT |
270 | else |
271 | return (struct hlist_head *) | |
ea72912c ED |
272 | __get_free_pages(GFP_ATOMIC | __GFP_ZERO, |
273 | get_order(size)); | |
1da177e4 LT |
274 | } |
275 | ||
15e47304 | 276 | static void nl_portid_hash_free(struct hlist_head *table, size_t size) |
1da177e4 LT |
277 | { |
278 | if (size <= PAGE_SIZE) | |
279 | kfree(table); | |
280 | else | |
281 | free_pages((unsigned long)table, get_order(size)); | |
282 | } | |
283 | ||
15e47304 | 284 | static int nl_portid_hash_rehash(struct nl_portid_hash *hash, int grow) |
1da177e4 LT |
285 | { |
286 | unsigned int omask, mask, shift; | |
287 | size_t osize, size; | |
288 | struct hlist_head *otable, *table; | |
289 | int i; | |
290 | ||
291 | omask = mask = hash->mask; | |
292 | osize = size = (mask + 1) * sizeof(*table); | |
293 | shift = hash->shift; | |
294 | ||
295 | if (grow) { | |
296 | if (++shift > hash->max_shift) | |
297 | return 0; | |
298 | mask = mask * 2 + 1; | |
299 | size *= 2; | |
300 | } | |
301 | ||
15e47304 | 302 | table = nl_portid_hash_zalloc(size); |
1da177e4 LT |
303 | if (!table) |
304 | return 0; | |
305 | ||
1da177e4 LT |
306 | otable = hash->table; |
307 | hash->table = table; | |
308 | hash->mask = mask; | |
309 | hash->shift = shift; | |
310 | get_random_bytes(&hash->rnd, sizeof(hash->rnd)); | |
311 | ||
312 | for (i = 0; i <= omask; i++) { | |
313 | struct sock *sk; | |
b67bfe0d | 314 | struct hlist_node *tmp; |
1da177e4 | 315 | |
b67bfe0d | 316 | sk_for_each_safe(sk, tmp, &otable[i]) |
15e47304 | 317 | __sk_add_node(sk, nl_portid_hashfn(hash, nlk_sk(sk)->portid)); |
1da177e4 LT |
318 | } |
319 | ||
15e47304 | 320 | nl_portid_hash_free(otable, osize); |
1da177e4 LT |
321 | hash->rehash_time = jiffies + 10 * 60 * HZ; |
322 | return 1; | |
323 | } | |
324 | ||
15e47304 | 325 | static inline int nl_portid_hash_dilute(struct nl_portid_hash *hash, int len) |
1da177e4 LT |
326 | { |
327 | int avg = hash->entries >> hash->shift; | |
328 | ||
15e47304 | 329 | if (unlikely(avg > 1) && nl_portid_hash_rehash(hash, 1)) |
1da177e4 LT |
330 | return 1; |
331 | ||
332 | if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) { | |
15e47304 | 333 | nl_portid_hash_rehash(hash, 0); |
1da177e4 LT |
334 | return 1; |
335 | } | |
336 | ||
337 | return 0; | |
338 | } | |
339 | ||
90ddc4f0 | 340 | static const struct proto_ops netlink_ops; |
1da177e4 | 341 | |
4277a083 PM |
342 | static void |
343 | netlink_update_listeners(struct sock *sk) | |
344 | { | |
345 | struct netlink_table *tbl = &nl_table[sk->sk_protocol]; | |
4277a083 PM |
346 | unsigned long mask; |
347 | unsigned int i; | |
6d772ac5 ED |
348 | struct listeners *listeners; |
349 | ||
350 | listeners = nl_deref_protected(tbl->listeners); | |
351 | if (!listeners) | |
352 | return; | |
4277a083 | 353 | |
b4ff4f04 | 354 | for (i = 0; i < NLGRPLONGS(tbl->groups); i++) { |
4277a083 | 355 | mask = 0; |
b67bfe0d | 356 | sk_for_each_bound(sk, &tbl->mc_list) { |
b4ff4f04 JB |
357 | if (i < NLGRPLONGS(nlk_sk(sk)->ngroups)) |
358 | mask |= nlk_sk(sk)->groups[i]; | |
359 | } | |
6d772ac5 | 360 | listeners->masks[i] = mask; |
4277a083 PM |
361 | } |
362 | /* this function is only called with the netlink table "grabbed", which | |
363 | * makes sure updates are visible before bind or setsockopt return. */ | |
364 | } | |
365 | ||
15e47304 | 366 | static int netlink_insert(struct sock *sk, struct net *net, u32 portid) |
1da177e4 | 367 | { |
15e47304 | 368 | struct nl_portid_hash *hash = &nl_table[sk->sk_protocol].hash; |
1da177e4 LT |
369 | struct hlist_head *head; |
370 | int err = -EADDRINUSE; | |
371 | struct sock *osk; | |
1da177e4 LT |
372 | int len; |
373 | ||
374 | netlink_table_grab(); | |
15e47304 | 375 | head = nl_portid_hashfn(hash, portid); |
1da177e4 | 376 | len = 0; |
b67bfe0d | 377 | sk_for_each(osk, head) { |
15e47304 | 378 | if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->portid == portid)) |
1da177e4 LT |
379 | break; |
380 | len++; | |
381 | } | |
b67bfe0d | 382 | if (osk) |
1da177e4 LT |
383 | goto err; |
384 | ||
385 | err = -EBUSY; | |
15e47304 | 386 | if (nlk_sk(sk)->portid) |
1da177e4 LT |
387 | goto err; |
388 | ||
389 | err = -ENOMEM; | |
390 | if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX)) | |
391 | goto err; | |
392 | ||
15e47304 EB |
393 | if (len && nl_portid_hash_dilute(hash, len)) |
394 | head = nl_portid_hashfn(hash, portid); | |
1da177e4 | 395 | hash->entries++; |
15e47304 | 396 | nlk_sk(sk)->portid = portid; |
1da177e4 LT |
397 | sk_add_node(sk, head); |
398 | err = 0; | |
399 | ||
400 | err: | |
401 | netlink_table_ungrab(); | |
402 | return err; | |
403 | } | |
404 | ||
405 | static void netlink_remove(struct sock *sk) | |
406 | { | |
407 | netlink_table_grab(); | |
d470e3b4 DM |
408 | if (sk_del_node_init(sk)) |
409 | nl_table[sk->sk_protocol].hash.entries--; | |
f7fa9b10 | 410 | if (nlk_sk(sk)->subscriptions) |
1da177e4 LT |
411 | __sk_del_bind_node(sk); |
412 | netlink_table_ungrab(); | |
413 | } | |
414 | ||
415 | static struct proto netlink_proto = { | |
416 | .name = "NETLINK", | |
417 | .owner = THIS_MODULE, | |
418 | .obj_size = sizeof(struct netlink_sock), | |
419 | }; | |
420 | ||
1b8d7ae4 EB |
421 | static int __netlink_create(struct net *net, struct socket *sock, |
422 | struct mutex *cb_mutex, int protocol) | |
1da177e4 LT |
423 | { |
424 | struct sock *sk; | |
425 | struct netlink_sock *nlk; | |
ab33a171 PM |
426 | |
427 | sock->ops = &netlink_ops; | |
428 | ||
6257ff21 | 429 | sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto); |
ab33a171 PM |
430 | if (!sk) |
431 | return -ENOMEM; | |
432 | ||
433 | sock_init_data(sock, sk); | |
434 | ||
435 | nlk = nlk_sk(sk); | |
658cb354 | 436 | if (cb_mutex) { |
ffa4d721 | 437 | nlk->cb_mutex = cb_mutex; |
658cb354 | 438 | } else { |
ffa4d721 PM |
439 | nlk->cb_mutex = &nlk->cb_def_mutex; |
440 | mutex_init(nlk->cb_mutex); | |
441 | } | |
ab33a171 PM |
442 | init_waitqueue_head(&nlk->wait); |
443 | ||
444 | sk->sk_destruct = netlink_sock_destruct; | |
445 | sk->sk_protocol = protocol; | |
446 | return 0; | |
447 | } | |
448 | ||
3f378b68 EP |
449 | static int netlink_create(struct net *net, struct socket *sock, int protocol, |
450 | int kern) | |
ab33a171 PM |
451 | { |
452 | struct module *module = NULL; | |
af65bdfc | 453 | struct mutex *cb_mutex; |
f7fa9b10 | 454 | struct netlink_sock *nlk; |
03292745 | 455 | void (*bind)(int group); |
ab33a171 | 456 | int err = 0; |
1da177e4 LT |
457 | |
458 | sock->state = SS_UNCONNECTED; | |
459 | ||
460 | if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM) | |
461 | return -ESOCKTNOSUPPORT; | |
462 | ||
6ac552fd | 463 | if (protocol < 0 || protocol >= MAX_LINKS) |
1da177e4 LT |
464 | return -EPROTONOSUPPORT; |
465 | ||
77247bbb | 466 | netlink_lock_table(); |
95a5afca | 467 | #ifdef CONFIG_MODULES |
ab33a171 | 468 | if (!nl_table[protocol].registered) { |
77247bbb | 469 | netlink_unlock_table(); |
4fdb3bb7 | 470 | request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol); |
77247bbb | 471 | netlink_lock_table(); |
4fdb3bb7 | 472 | } |
ab33a171 PM |
473 | #endif |
474 | if (nl_table[protocol].registered && | |
475 | try_module_get(nl_table[protocol].module)) | |
476 | module = nl_table[protocol].module; | |
974c37e9 AD |
477 | else |
478 | err = -EPROTONOSUPPORT; | |
af65bdfc | 479 | cb_mutex = nl_table[protocol].cb_mutex; |
03292745 | 480 | bind = nl_table[protocol].bind; |
77247bbb | 481 | netlink_unlock_table(); |
4fdb3bb7 | 482 | |
974c37e9 AD |
483 | if (err < 0) |
484 | goto out; | |
485 | ||
6ac552fd PM |
486 | err = __netlink_create(net, sock, cb_mutex, protocol); |
487 | if (err < 0) | |
f7fa9b10 PM |
488 | goto out_module; |
489 | ||
6f756a8c | 490 | local_bh_disable(); |
c1fd3b94 | 491 | sock_prot_inuse_add(net, &netlink_proto, 1); |
6f756a8c DM |
492 | local_bh_enable(); |
493 | ||
f7fa9b10 | 494 | nlk = nlk_sk(sock->sk); |
f7fa9b10 | 495 | nlk->module = module; |
03292745 | 496 | nlk->netlink_bind = bind; |
ab33a171 PM |
497 | out: |
498 | return err; | |
1da177e4 | 499 | |
ab33a171 PM |
500 | out_module: |
501 | module_put(module); | |
502 | goto out; | |
1da177e4 LT |
503 | } |
504 | ||
505 | static int netlink_release(struct socket *sock) | |
506 | { | |
507 | struct sock *sk = sock->sk; | |
508 | struct netlink_sock *nlk; | |
509 | ||
510 | if (!sk) | |
511 | return 0; | |
512 | ||
513 | netlink_remove(sk); | |
ac57b3a9 | 514 | sock_orphan(sk); |
1da177e4 LT |
515 | nlk = nlk_sk(sk); |
516 | ||
3f660d66 HX |
517 | /* |
518 | * OK. Socket is unlinked, any packets that arrive now | |
519 | * will be purged. | |
520 | */ | |
1da177e4 | 521 | |
1da177e4 LT |
522 | sock->sk = NULL; |
523 | wake_up_interruptible_all(&nlk->wait); | |
524 | ||
525 | skb_queue_purge(&sk->sk_write_queue); | |
526 | ||
15e47304 | 527 | if (nlk->portid) { |
1da177e4 | 528 | struct netlink_notify n = { |
3b1e0a65 | 529 | .net = sock_net(sk), |
1da177e4 | 530 | .protocol = sk->sk_protocol, |
15e47304 | 531 | .portid = nlk->portid, |
1da177e4 | 532 | }; |
e041c683 AS |
533 | atomic_notifier_call_chain(&netlink_chain, |
534 | NETLINK_URELEASE, &n); | |
746fac4d | 535 | } |
4fdb3bb7 | 536 | |
5e7c001c | 537 | module_put(nlk->module); |
4fdb3bb7 | 538 | |
4277a083 | 539 | netlink_table_grab(); |
aed81560 | 540 | if (netlink_is_kernel(sk)) { |
869e58f8 DL |
541 | BUG_ON(nl_table[sk->sk_protocol].registered == 0); |
542 | if (--nl_table[sk->sk_protocol].registered == 0) { | |
6d772ac5 ED |
543 | struct listeners *old; |
544 | ||
545 | old = nl_deref_protected(nl_table[sk->sk_protocol].listeners); | |
546 | RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL); | |
547 | kfree_rcu(old, rcu); | |
869e58f8 | 548 | nl_table[sk->sk_protocol].module = NULL; |
9785e10a PNA |
549 | nl_table[sk->sk_protocol].bind = NULL; |
550 | nl_table[sk->sk_protocol].flags = 0; | |
869e58f8 DL |
551 | nl_table[sk->sk_protocol].registered = 0; |
552 | } | |
658cb354 | 553 | } else if (nlk->subscriptions) { |
4277a083 | 554 | netlink_update_listeners(sk); |
658cb354 | 555 | } |
4277a083 | 556 | netlink_table_ungrab(); |
77247bbb | 557 | |
f7fa9b10 PM |
558 | kfree(nlk->groups); |
559 | nlk->groups = NULL; | |
560 | ||
3755810c | 561 | local_bh_disable(); |
c1fd3b94 | 562 | sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1); |
3755810c | 563 | local_bh_enable(); |
1da177e4 LT |
564 | sock_put(sk); |
565 | return 0; | |
566 | } | |
567 | ||
568 | static int netlink_autobind(struct socket *sock) | |
569 | { | |
570 | struct sock *sk = sock->sk; | |
3b1e0a65 | 571 | struct net *net = sock_net(sk); |
15e47304 | 572 | struct nl_portid_hash *hash = &nl_table[sk->sk_protocol].hash; |
1da177e4 LT |
573 | struct hlist_head *head; |
574 | struct sock *osk; | |
15e47304 | 575 | s32 portid = task_tgid_vnr(current); |
1da177e4 LT |
576 | int err; |
577 | static s32 rover = -4097; | |
578 | ||
579 | retry: | |
580 | cond_resched(); | |
581 | netlink_table_grab(); | |
15e47304 | 582 | head = nl_portid_hashfn(hash, portid); |
b67bfe0d | 583 | sk_for_each(osk, head) { |
878628fb | 584 | if (!net_eq(sock_net(osk), net)) |
b4b51029 | 585 | continue; |
15e47304 EB |
586 | if (nlk_sk(osk)->portid == portid) { |
587 | /* Bind collision, search negative portid values. */ | |
588 | portid = rover--; | |
1da177e4 LT |
589 | if (rover > -4097) |
590 | rover = -4097; | |
591 | netlink_table_ungrab(); | |
592 | goto retry; | |
593 | } | |
594 | } | |
595 | netlink_table_ungrab(); | |
596 | ||
15e47304 | 597 | err = netlink_insert(sk, net, portid); |
1da177e4 LT |
598 | if (err == -EADDRINUSE) |
599 | goto retry; | |
d470e3b4 DM |
600 | |
601 | /* If 2 threads race to autobind, that is fine. */ | |
602 | if (err == -EBUSY) | |
603 | err = 0; | |
604 | ||
605 | return err; | |
1da177e4 LT |
606 | } |
607 | ||
b57ef81f | 608 | static inline int netlink_capable(const struct socket *sock, unsigned int flag) |
746fac4d | 609 | { |
9785e10a | 610 | return (nl_table[sock->sk->sk_protocol].flags & flag) || |
df008c91 | 611 | ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN); |
746fac4d | 612 | } |
1da177e4 | 613 | |
f7fa9b10 PM |
614 | static void |
615 | netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions) | |
616 | { | |
617 | struct netlink_sock *nlk = nlk_sk(sk); | |
618 | ||
619 | if (nlk->subscriptions && !subscriptions) | |
620 | __sk_del_bind_node(sk); | |
621 | else if (!nlk->subscriptions && subscriptions) | |
622 | sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list); | |
623 | nlk->subscriptions = subscriptions; | |
624 | } | |
625 | ||
b4ff4f04 | 626 | static int netlink_realloc_groups(struct sock *sk) |
513c2500 PM |
627 | { |
628 | struct netlink_sock *nlk = nlk_sk(sk); | |
629 | unsigned int groups; | |
b4ff4f04 | 630 | unsigned long *new_groups; |
513c2500 PM |
631 | int err = 0; |
632 | ||
b4ff4f04 JB |
633 | netlink_table_grab(); |
634 | ||
513c2500 | 635 | groups = nl_table[sk->sk_protocol].groups; |
b4ff4f04 | 636 | if (!nl_table[sk->sk_protocol].registered) { |
513c2500 | 637 | err = -ENOENT; |
b4ff4f04 JB |
638 | goto out_unlock; |
639 | } | |
513c2500 | 640 | |
b4ff4f04 JB |
641 | if (nlk->ngroups >= groups) |
642 | goto out_unlock; | |
513c2500 | 643 | |
b4ff4f04 JB |
644 | new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC); |
645 | if (new_groups == NULL) { | |
646 | err = -ENOMEM; | |
647 | goto out_unlock; | |
648 | } | |
6ac552fd | 649 | memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0, |
b4ff4f04 JB |
650 | NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups)); |
651 | ||
652 | nlk->groups = new_groups; | |
513c2500 | 653 | nlk->ngroups = groups; |
b4ff4f04 JB |
654 | out_unlock: |
655 | netlink_table_ungrab(); | |
656 | return err; | |
513c2500 PM |
657 | } |
658 | ||
6ac552fd PM |
659 | static int netlink_bind(struct socket *sock, struct sockaddr *addr, |
660 | int addr_len) | |
1da177e4 LT |
661 | { |
662 | struct sock *sk = sock->sk; | |
3b1e0a65 | 663 | struct net *net = sock_net(sk); |
1da177e4 LT |
664 | struct netlink_sock *nlk = nlk_sk(sk); |
665 | struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; | |
666 | int err; | |
746fac4d | 667 | |
4e4b5376 HFS |
668 | if (addr_len < sizeof(struct sockaddr_nl)) |
669 | return -EINVAL; | |
670 | ||
1da177e4 LT |
671 | if (nladdr->nl_family != AF_NETLINK) |
672 | return -EINVAL; | |
673 | ||
674 | /* Only superuser is allowed to listen multicasts */ | |
513c2500 | 675 | if (nladdr->nl_groups) { |
9785e10a | 676 | if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV)) |
513c2500 | 677 | return -EPERM; |
b4ff4f04 JB |
678 | err = netlink_realloc_groups(sk); |
679 | if (err) | |
680 | return err; | |
513c2500 | 681 | } |
1da177e4 | 682 | |
15e47304 EB |
683 | if (nlk->portid) { |
684 | if (nladdr->nl_pid != nlk->portid) | |
1da177e4 LT |
685 | return -EINVAL; |
686 | } else { | |
687 | err = nladdr->nl_pid ? | |
b4b51029 | 688 | netlink_insert(sk, net, nladdr->nl_pid) : |
1da177e4 LT |
689 | netlink_autobind(sock); |
690 | if (err) | |
691 | return err; | |
692 | } | |
693 | ||
513c2500 | 694 | if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0])) |
1da177e4 LT |
695 | return 0; |
696 | ||
697 | netlink_table_grab(); | |
f7fa9b10 | 698 | netlink_update_subscriptions(sk, nlk->subscriptions + |
746fac4d YH |
699 | hweight32(nladdr->nl_groups) - |
700 | hweight32(nlk->groups[0])); | |
701 | nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups; | |
4277a083 | 702 | netlink_update_listeners(sk); |
1da177e4 LT |
703 | netlink_table_ungrab(); |
704 | ||
03292745 PNA |
705 | if (nlk->netlink_bind && nlk->groups[0]) { |
706 | int i; | |
707 | ||
708 | for (i=0; i<nlk->ngroups; i++) { | |
709 | if (test_bit(i, nlk->groups)) | |
710 | nlk->netlink_bind(i); | |
711 | } | |
712 | } | |
713 | ||
1da177e4 LT |
714 | return 0; |
715 | } | |
716 | ||
717 | static int netlink_connect(struct socket *sock, struct sockaddr *addr, | |
718 | int alen, int flags) | |
719 | { | |
720 | int err = 0; | |
721 | struct sock *sk = sock->sk; | |
722 | struct netlink_sock *nlk = nlk_sk(sk); | |
6ac552fd | 723 | struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr; |
1da177e4 | 724 | |
6503d961 CG |
725 | if (alen < sizeof(addr->sa_family)) |
726 | return -EINVAL; | |
727 | ||
1da177e4 LT |
728 | if (addr->sa_family == AF_UNSPEC) { |
729 | sk->sk_state = NETLINK_UNCONNECTED; | |
15e47304 | 730 | nlk->dst_portid = 0; |
d629b836 | 731 | nlk->dst_group = 0; |
1da177e4 LT |
732 | return 0; |
733 | } | |
734 | if (addr->sa_family != AF_NETLINK) | |
735 | return -EINVAL; | |
736 | ||
737 | /* Only superuser is allowed to send multicasts */ | |
9785e10a | 738 | if (nladdr->nl_groups && !netlink_capable(sock, NL_CFG_F_NONROOT_SEND)) |
1da177e4 LT |
739 | return -EPERM; |
740 | ||
15e47304 | 741 | if (!nlk->portid) |
1da177e4 LT |
742 | err = netlink_autobind(sock); |
743 | ||
744 | if (err == 0) { | |
745 | sk->sk_state = NETLINK_CONNECTED; | |
15e47304 | 746 | nlk->dst_portid = nladdr->nl_pid; |
d629b836 | 747 | nlk->dst_group = ffs(nladdr->nl_groups); |
1da177e4 LT |
748 | } |
749 | ||
750 | return err; | |
751 | } | |
752 | ||
6ac552fd PM |
753 | static int netlink_getname(struct socket *sock, struct sockaddr *addr, |
754 | int *addr_len, int peer) | |
1da177e4 LT |
755 | { |
756 | struct sock *sk = sock->sk; | |
757 | struct netlink_sock *nlk = nlk_sk(sk); | |
13cfa97b | 758 | DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr); |
746fac4d | 759 | |
1da177e4 LT |
760 | nladdr->nl_family = AF_NETLINK; |
761 | nladdr->nl_pad = 0; | |
762 | *addr_len = sizeof(*nladdr); | |
763 | ||
764 | if (peer) { | |
15e47304 | 765 | nladdr->nl_pid = nlk->dst_portid; |
d629b836 | 766 | nladdr->nl_groups = netlink_group_mask(nlk->dst_group); |
1da177e4 | 767 | } else { |
15e47304 | 768 | nladdr->nl_pid = nlk->portid; |
513c2500 | 769 | nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0; |
1da177e4 LT |
770 | } |
771 | return 0; | |
772 | } | |
773 | ||
774 | static void netlink_overrun(struct sock *sk) | |
775 | { | |
38938bfe PNA |
776 | struct netlink_sock *nlk = nlk_sk(sk); |
777 | ||
778 | if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) { | |
779 | if (!test_and_set_bit(0, &nlk_sk(sk)->state)) { | |
780 | sk->sk_err = ENOBUFS; | |
781 | sk->sk_error_report(sk); | |
782 | } | |
1da177e4 | 783 | } |
38938bfe | 784 | atomic_inc(&sk->sk_drops); |
1da177e4 LT |
785 | } |
786 | ||
15e47304 | 787 | static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid) |
1da177e4 | 788 | { |
1da177e4 LT |
789 | struct sock *sock; |
790 | struct netlink_sock *nlk; | |
791 | ||
15e47304 | 792 | sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid); |
1da177e4 LT |
793 | if (!sock) |
794 | return ERR_PTR(-ECONNREFUSED); | |
795 | ||
796 | /* Don't bother queuing skb if kernel socket has no input function */ | |
797 | nlk = nlk_sk(sock); | |
cd40b7d3 | 798 | if (sock->sk_state == NETLINK_CONNECTED && |
15e47304 | 799 | nlk->dst_portid != nlk_sk(ssk)->portid) { |
1da177e4 LT |
800 | sock_put(sock); |
801 | return ERR_PTR(-ECONNREFUSED); | |
802 | } | |
803 | return sock; | |
804 | } | |
805 | ||
806 | struct sock *netlink_getsockbyfilp(struct file *filp) | |
807 | { | |
496ad9aa | 808 | struct inode *inode = file_inode(filp); |
1da177e4 LT |
809 | struct sock *sock; |
810 | ||
811 | if (!S_ISSOCK(inode->i_mode)) | |
812 | return ERR_PTR(-ENOTSOCK); | |
813 | ||
814 | sock = SOCKET_I(inode)->sk; | |
815 | if (sock->sk_family != AF_NETLINK) | |
816 | return ERR_PTR(-EINVAL); | |
817 | ||
818 | sock_hold(sock); | |
819 | return sock; | |
820 | } | |
821 | ||
822 | /* | |
823 | * Attach a skb to a netlink socket. | |
824 | * The caller must hold a reference to the destination socket. On error, the | |
825 | * reference is dropped. The skb is not send to the destination, just all | |
826 | * all error checks are performed and memory in the queue is reserved. | |
827 | * Return values: | |
828 | * < 0: error. skb freed, reference to sock dropped. | |
829 | * 0: continue | |
830 | * 1: repeat lookup - reference dropped while waiting for socket memory. | |
831 | */ | |
9457afee | 832 | int netlink_attachskb(struct sock *sk, struct sk_buff *skb, |
c3d8d1e3 | 833 | long *timeo, struct sock *ssk) |
1da177e4 LT |
834 | { |
835 | struct netlink_sock *nlk; | |
836 | ||
837 | nlk = nlk_sk(sk); | |
838 | ||
839 | if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || | |
840 | test_bit(0, &nlk->state)) { | |
841 | DECLARE_WAITQUEUE(wait, current); | |
c3d8d1e3 | 842 | if (!*timeo) { |
aed81560 | 843 | if (!ssk || netlink_is_kernel(ssk)) |
1da177e4 LT |
844 | netlink_overrun(sk); |
845 | sock_put(sk); | |
846 | kfree_skb(skb); | |
847 | return -EAGAIN; | |
848 | } | |
849 | ||
850 | __set_current_state(TASK_INTERRUPTIBLE); | |
851 | add_wait_queue(&nlk->wait, &wait); | |
852 | ||
853 | if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || | |
854 | test_bit(0, &nlk->state)) && | |
855 | !sock_flag(sk, SOCK_DEAD)) | |
c3d8d1e3 | 856 | *timeo = schedule_timeout(*timeo); |
1da177e4 LT |
857 | |
858 | __set_current_state(TASK_RUNNING); | |
859 | remove_wait_queue(&nlk->wait, &wait); | |
860 | sock_put(sk); | |
861 | ||
862 | if (signal_pending(current)) { | |
863 | kfree_skb(skb); | |
c3d8d1e3 | 864 | return sock_intr_errno(*timeo); |
1da177e4 LT |
865 | } |
866 | return 1; | |
867 | } | |
868 | skb_set_owner_r(skb, sk); | |
869 | return 0; | |
870 | } | |
871 | ||
4a7e7c2a | 872 | static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 873 | { |
1da177e4 LT |
874 | int len = skb->len; |
875 | ||
1da177e4 LT |
876 | skb_queue_tail(&sk->sk_receive_queue, skb); |
877 | sk->sk_data_ready(sk, len); | |
4a7e7c2a ED |
878 | return len; |
879 | } | |
880 | ||
881 | int netlink_sendskb(struct sock *sk, struct sk_buff *skb) | |
882 | { | |
883 | int len = __netlink_sendskb(sk, skb); | |
884 | ||
1da177e4 LT |
885 | sock_put(sk); |
886 | return len; | |
887 | } | |
888 | ||
889 | void netlink_detachskb(struct sock *sk, struct sk_buff *skb) | |
890 | { | |
891 | kfree_skb(skb); | |
892 | sock_put(sk); | |
893 | } | |
894 | ||
b57ef81f | 895 | static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation) |
1da177e4 LT |
896 | { |
897 | int delta; | |
898 | ||
899 | skb_orphan(skb); | |
900 | ||
4305b541 | 901 | delta = skb->end - skb->tail; |
1da177e4 LT |
902 | if (delta * 2 < skb->truesize) |
903 | return skb; | |
904 | ||
905 | if (skb_shared(skb)) { | |
906 | struct sk_buff *nskb = skb_clone(skb, allocation); | |
907 | if (!nskb) | |
908 | return skb; | |
8460c00f | 909 | consume_skb(skb); |
1da177e4 LT |
910 | skb = nskb; |
911 | } | |
912 | ||
913 | if (!pskb_expand_head(skb, 0, -delta, allocation)) | |
914 | skb->truesize -= delta; | |
915 | ||
916 | return skb; | |
917 | } | |
918 | ||
b57ef81f | 919 | static void netlink_rcv_wake(struct sock *sk) |
cd40b7d3 DL |
920 | { |
921 | struct netlink_sock *nlk = nlk_sk(sk); | |
922 | ||
923 | if (skb_queue_empty(&sk->sk_receive_queue)) | |
924 | clear_bit(0, &nlk->state); | |
925 | if (!test_bit(0, &nlk->state)) | |
926 | wake_up_interruptible(&nlk->wait); | |
927 | } | |
928 | ||
3fbc2905 EB |
929 | static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb, |
930 | struct sock *ssk) | |
cd40b7d3 DL |
931 | { |
932 | int ret; | |
933 | struct netlink_sock *nlk = nlk_sk(sk); | |
934 | ||
935 | ret = -ECONNREFUSED; | |
936 | if (nlk->netlink_rcv != NULL) { | |
937 | ret = skb->len; | |
938 | skb_set_owner_r(skb, sk); | |
3fbc2905 | 939 | NETLINK_CB(skb).ssk = ssk; |
cd40b7d3 | 940 | nlk->netlink_rcv(skb); |
bfb253c9 ED |
941 | consume_skb(skb); |
942 | } else { | |
943 | kfree_skb(skb); | |
cd40b7d3 | 944 | } |
cd40b7d3 DL |
945 | sock_put(sk); |
946 | return ret; | |
947 | } | |
948 | ||
949 | int netlink_unicast(struct sock *ssk, struct sk_buff *skb, | |
15e47304 | 950 | u32 portid, int nonblock) |
1da177e4 LT |
951 | { |
952 | struct sock *sk; | |
953 | int err; | |
954 | long timeo; | |
955 | ||
956 | skb = netlink_trim(skb, gfp_any()); | |
957 | ||
958 | timeo = sock_sndtimeo(ssk, nonblock); | |
959 | retry: | |
15e47304 | 960 | sk = netlink_getsockbyportid(ssk, portid); |
1da177e4 LT |
961 | if (IS_ERR(sk)) { |
962 | kfree_skb(skb); | |
963 | return PTR_ERR(sk); | |
964 | } | |
cd40b7d3 | 965 | if (netlink_is_kernel(sk)) |
3fbc2905 | 966 | return netlink_unicast_kernel(sk, skb, ssk); |
cd40b7d3 | 967 | |
b1153f29 | 968 | if (sk_filter(sk, skb)) { |
84874607 | 969 | err = skb->len; |
b1153f29 SH |
970 | kfree_skb(skb); |
971 | sock_put(sk); | |
972 | return err; | |
973 | } | |
974 | ||
9457afee | 975 | err = netlink_attachskb(sk, skb, &timeo, ssk); |
1da177e4 LT |
976 | if (err == 1) |
977 | goto retry; | |
978 | if (err) | |
979 | return err; | |
980 | ||
7ee015e0 | 981 | return netlink_sendskb(sk, skb); |
1da177e4 | 982 | } |
6ac552fd | 983 | EXPORT_SYMBOL(netlink_unicast); |
1da177e4 | 984 | |
4277a083 PM |
985 | int netlink_has_listeners(struct sock *sk, unsigned int group) |
986 | { | |
987 | int res = 0; | |
5c398dc8 | 988 | struct listeners *listeners; |
4277a083 | 989 | |
aed81560 | 990 | BUG_ON(!netlink_is_kernel(sk)); |
b4ff4f04 JB |
991 | |
992 | rcu_read_lock(); | |
993 | listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners); | |
994 | ||
6d772ac5 | 995 | if (listeners && group - 1 < nl_table[sk->sk_protocol].groups) |
5c398dc8 | 996 | res = test_bit(group - 1, listeners->masks); |
b4ff4f04 JB |
997 | |
998 | rcu_read_unlock(); | |
999 | ||
4277a083 PM |
1000 | return res; |
1001 | } | |
1002 | EXPORT_SYMBOL_GPL(netlink_has_listeners); | |
1003 | ||
b57ef81f | 1004 | static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb) |
1da177e4 LT |
1005 | { |
1006 | struct netlink_sock *nlk = nlk_sk(sk); | |
1007 | ||
1008 | if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && | |
1009 | !test_bit(0, &nlk->state)) { | |
1010 | skb_set_owner_r(skb, sk); | |
4a7e7c2a | 1011 | __netlink_sendskb(sk, skb); |
2c645800 | 1012 | return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1); |
1da177e4 LT |
1013 | } |
1014 | return -1; | |
1015 | } | |
1016 | ||
1017 | struct netlink_broadcast_data { | |
1018 | struct sock *exclude_sk; | |
b4b51029 | 1019 | struct net *net; |
15e47304 | 1020 | u32 portid; |
1da177e4 LT |
1021 | u32 group; |
1022 | int failure; | |
ff491a73 | 1023 | int delivery_failure; |
1da177e4 LT |
1024 | int congested; |
1025 | int delivered; | |
7d877f3b | 1026 | gfp_t allocation; |
1da177e4 | 1027 | struct sk_buff *skb, *skb2; |
910a7e90 EB |
1028 | int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data); |
1029 | void *tx_data; | |
1da177e4 LT |
1030 | }; |
1031 | ||
b57ef81f | 1032 | static int do_one_broadcast(struct sock *sk, |
1da177e4 LT |
1033 | struct netlink_broadcast_data *p) |
1034 | { | |
1035 | struct netlink_sock *nlk = nlk_sk(sk); | |
1036 | int val; | |
1037 | ||
1038 | if (p->exclude_sk == sk) | |
1039 | goto out; | |
1040 | ||
15e47304 | 1041 | if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups || |
f7fa9b10 | 1042 | !test_bit(p->group - 1, nlk->groups)) |
1da177e4 LT |
1043 | goto out; |
1044 | ||
878628fb | 1045 | if (!net_eq(sock_net(sk), p->net)) |
b4b51029 EB |
1046 | goto out; |
1047 | ||
1da177e4 LT |
1048 | if (p->failure) { |
1049 | netlink_overrun(sk); | |
1050 | goto out; | |
1051 | } | |
1052 | ||
1053 | sock_hold(sk); | |
1054 | if (p->skb2 == NULL) { | |
68acc024 | 1055 | if (skb_shared(p->skb)) { |
1da177e4 LT |
1056 | p->skb2 = skb_clone(p->skb, p->allocation); |
1057 | } else { | |
68acc024 TC |
1058 | p->skb2 = skb_get(p->skb); |
1059 | /* | |
1060 | * skb ownership may have been set when | |
1061 | * delivered to a previous socket. | |
1062 | */ | |
1063 | skb_orphan(p->skb2); | |
1da177e4 LT |
1064 | } |
1065 | } | |
1066 | if (p->skb2 == NULL) { | |
1067 | netlink_overrun(sk); | |
1068 | /* Clone failed. Notify ALL listeners. */ | |
1069 | p->failure = 1; | |
be0c22a4 PNA |
1070 | if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR) |
1071 | p->delivery_failure = 1; | |
910a7e90 EB |
1072 | } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) { |
1073 | kfree_skb(p->skb2); | |
1074 | p->skb2 = NULL; | |
b1153f29 SH |
1075 | } else if (sk_filter(sk, p->skb2)) { |
1076 | kfree_skb(p->skb2); | |
1077 | p->skb2 = NULL; | |
1da177e4 LT |
1078 | } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) { |
1079 | netlink_overrun(sk); | |
be0c22a4 PNA |
1080 | if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR) |
1081 | p->delivery_failure = 1; | |
1da177e4 LT |
1082 | } else { |
1083 | p->congested |= val; | |
1084 | p->delivered = 1; | |
1085 | p->skb2 = NULL; | |
1086 | } | |
1087 | sock_put(sk); | |
1088 | ||
1089 | out: | |
1090 | return 0; | |
1091 | } | |
1092 | ||
15e47304 | 1093 | int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid, |
910a7e90 EB |
1094 | u32 group, gfp_t allocation, |
1095 | int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data), | |
1096 | void *filter_data) | |
1da177e4 | 1097 | { |
3b1e0a65 | 1098 | struct net *net = sock_net(ssk); |
1da177e4 | 1099 | struct netlink_broadcast_data info; |
1da177e4 LT |
1100 | struct sock *sk; |
1101 | ||
1102 | skb = netlink_trim(skb, allocation); | |
1103 | ||
1104 | info.exclude_sk = ssk; | |
b4b51029 | 1105 | info.net = net; |
15e47304 | 1106 | info.portid = portid; |
1da177e4 LT |
1107 | info.group = group; |
1108 | info.failure = 0; | |
ff491a73 | 1109 | info.delivery_failure = 0; |
1da177e4 LT |
1110 | info.congested = 0; |
1111 | info.delivered = 0; | |
1112 | info.allocation = allocation; | |
1113 | info.skb = skb; | |
1114 | info.skb2 = NULL; | |
910a7e90 EB |
1115 | info.tx_filter = filter; |
1116 | info.tx_data = filter_data; | |
1da177e4 LT |
1117 | |
1118 | /* While we sleep in clone, do not allow to change socket list */ | |
1119 | ||
1120 | netlink_lock_table(); | |
1121 | ||
b67bfe0d | 1122 | sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list) |
1da177e4 LT |
1123 | do_one_broadcast(sk, &info); |
1124 | ||
70d4bf6d | 1125 | consume_skb(skb); |
aa1c6a6f | 1126 | |
1da177e4 LT |
1127 | netlink_unlock_table(); |
1128 | ||
70d4bf6d NH |
1129 | if (info.delivery_failure) { |
1130 | kfree_skb(info.skb2); | |
ff491a73 | 1131 | return -ENOBUFS; |
658cb354 ED |
1132 | } |
1133 | consume_skb(info.skb2); | |
ff491a73 | 1134 | |
1da177e4 LT |
1135 | if (info.delivered) { |
1136 | if (info.congested && (allocation & __GFP_WAIT)) | |
1137 | yield(); | |
1138 | return 0; | |
1139 | } | |
1da177e4 LT |
1140 | return -ESRCH; |
1141 | } | |
910a7e90 EB |
1142 | EXPORT_SYMBOL(netlink_broadcast_filtered); |
1143 | ||
15e47304 | 1144 | int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid, |
910a7e90 EB |
1145 | u32 group, gfp_t allocation) |
1146 | { | |
15e47304 | 1147 | return netlink_broadcast_filtered(ssk, skb, portid, group, allocation, |
910a7e90 EB |
1148 | NULL, NULL); |
1149 | } | |
6ac552fd | 1150 | EXPORT_SYMBOL(netlink_broadcast); |
1da177e4 LT |
1151 | |
1152 | struct netlink_set_err_data { | |
1153 | struct sock *exclude_sk; | |
15e47304 | 1154 | u32 portid; |
1da177e4 LT |
1155 | u32 group; |
1156 | int code; | |
1157 | }; | |
1158 | ||
b57ef81f | 1159 | static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p) |
1da177e4 LT |
1160 | { |
1161 | struct netlink_sock *nlk = nlk_sk(sk); | |
1a50307b | 1162 | int ret = 0; |
1da177e4 LT |
1163 | |
1164 | if (sk == p->exclude_sk) | |
1165 | goto out; | |
1166 | ||
09ad9bc7 | 1167 | if (!net_eq(sock_net(sk), sock_net(p->exclude_sk))) |
b4b51029 EB |
1168 | goto out; |
1169 | ||
15e47304 | 1170 | if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups || |
f7fa9b10 | 1171 | !test_bit(p->group - 1, nlk->groups)) |
1da177e4 LT |
1172 | goto out; |
1173 | ||
1a50307b PNA |
1174 | if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) { |
1175 | ret = 1; | |
1176 | goto out; | |
1177 | } | |
1178 | ||
1da177e4 LT |
1179 | sk->sk_err = p->code; |
1180 | sk->sk_error_report(sk); | |
1181 | out: | |
1a50307b | 1182 | return ret; |
1da177e4 LT |
1183 | } |
1184 | ||
4843b93c PNA |
1185 | /** |
1186 | * netlink_set_err - report error to broadcast listeners | |
1187 | * @ssk: the kernel netlink socket, as returned by netlink_kernel_create() | |
15e47304 | 1188 | * @portid: the PORTID of a process that we want to skip (if any) |
4843b93c PNA |
1189 | * @groups: the broadcast group that will notice the error |
1190 | * @code: error code, must be negative (as usual in kernelspace) | |
1a50307b PNA |
1191 | * |
1192 | * This function returns the number of broadcast listeners that have set the | |
1193 | * NETLINK_RECV_NO_ENOBUFS socket option. | |
4843b93c | 1194 | */ |
15e47304 | 1195 | int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code) |
1da177e4 LT |
1196 | { |
1197 | struct netlink_set_err_data info; | |
1da177e4 | 1198 | struct sock *sk; |
1a50307b | 1199 | int ret = 0; |
1da177e4 LT |
1200 | |
1201 | info.exclude_sk = ssk; | |
15e47304 | 1202 | info.portid = portid; |
1da177e4 | 1203 | info.group = group; |
4843b93c PNA |
1204 | /* sk->sk_err wants a positive error value */ |
1205 | info.code = -code; | |
1da177e4 LT |
1206 | |
1207 | read_lock(&nl_table_lock); | |
1208 | ||
b67bfe0d | 1209 | sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list) |
1a50307b | 1210 | ret += do_one_set_err(sk, &info); |
1da177e4 LT |
1211 | |
1212 | read_unlock(&nl_table_lock); | |
1a50307b | 1213 | return ret; |
1da177e4 | 1214 | } |
dd5b6ce6 | 1215 | EXPORT_SYMBOL(netlink_set_err); |
1da177e4 | 1216 | |
84659eb5 JB |
1217 | /* must be called with netlink table grabbed */ |
1218 | static void netlink_update_socket_mc(struct netlink_sock *nlk, | |
1219 | unsigned int group, | |
1220 | int is_new) | |
1221 | { | |
1222 | int old, new = !!is_new, subscriptions; | |
1223 | ||
1224 | old = test_bit(group - 1, nlk->groups); | |
1225 | subscriptions = nlk->subscriptions - old + new; | |
1226 | if (new) | |
1227 | __set_bit(group - 1, nlk->groups); | |
1228 | else | |
1229 | __clear_bit(group - 1, nlk->groups); | |
1230 | netlink_update_subscriptions(&nlk->sk, subscriptions); | |
1231 | netlink_update_listeners(&nlk->sk); | |
1232 | } | |
1233 | ||
9a4595bc | 1234 | static int netlink_setsockopt(struct socket *sock, int level, int optname, |
b7058842 | 1235 | char __user *optval, unsigned int optlen) |
9a4595bc PM |
1236 | { |
1237 | struct sock *sk = sock->sk; | |
1238 | struct netlink_sock *nlk = nlk_sk(sk); | |
eb496534 JB |
1239 | unsigned int val = 0; |
1240 | int err; | |
9a4595bc PM |
1241 | |
1242 | if (level != SOL_NETLINK) | |
1243 | return -ENOPROTOOPT; | |
1244 | ||
1245 | if (optlen >= sizeof(int) && | |
eb496534 | 1246 | get_user(val, (unsigned int __user *)optval)) |
9a4595bc PM |
1247 | return -EFAULT; |
1248 | ||
1249 | switch (optname) { | |
1250 | case NETLINK_PKTINFO: | |
1251 | if (val) | |
1252 | nlk->flags |= NETLINK_RECV_PKTINFO; | |
1253 | else | |
1254 | nlk->flags &= ~NETLINK_RECV_PKTINFO; | |
1255 | err = 0; | |
1256 | break; | |
1257 | case NETLINK_ADD_MEMBERSHIP: | |
1258 | case NETLINK_DROP_MEMBERSHIP: { | |
9785e10a | 1259 | if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV)) |
9a4595bc | 1260 | return -EPERM; |
b4ff4f04 JB |
1261 | err = netlink_realloc_groups(sk); |
1262 | if (err) | |
1263 | return err; | |
9a4595bc PM |
1264 | if (!val || val - 1 >= nlk->ngroups) |
1265 | return -EINVAL; | |
1266 | netlink_table_grab(); | |
84659eb5 JB |
1267 | netlink_update_socket_mc(nlk, val, |
1268 | optname == NETLINK_ADD_MEMBERSHIP); | |
9a4595bc | 1269 | netlink_table_ungrab(); |
03292745 PNA |
1270 | |
1271 | if (nlk->netlink_bind) | |
1272 | nlk->netlink_bind(val); | |
1273 | ||
9a4595bc PM |
1274 | err = 0; |
1275 | break; | |
1276 | } | |
be0c22a4 PNA |
1277 | case NETLINK_BROADCAST_ERROR: |
1278 | if (val) | |
1279 | nlk->flags |= NETLINK_BROADCAST_SEND_ERROR; | |
1280 | else | |
1281 | nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR; | |
1282 | err = 0; | |
1283 | break; | |
38938bfe PNA |
1284 | case NETLINK_NO_ENOBUFS: |
1285 | if (val) { | |
1286 | nlk->flags |= NETLINK_RECV_NO_ENOBUFS; | |
1287 | clear_bit(0, &nlk->state); | |
1288 | wake_up_interruptible(&nlk->wait); | |
658cb354 | 1289 | } else { |
38938bfe | 1290 | nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS; |
658cb354 | 1291 | } |
38938bfe PNA |
1292 | err = 0; |
1293 | break; | |
9a4595bc PM |
1294 | default: |
1295 | err = -ENOPROTOOPT; | |
1296 | } | |
1297 | return err; | |
1298 | } | |
1299 | ||
1300 | static int netlink_getsockopt(struct socket *sock, int level, int optname, | |
746fac4d | 1301 | char __user *optval, int __user *optlen) |
9a4595bc PM |
1302 | { |
1303 | struct sock *sk = sock->sk; | |
1304 | struct netlink_sock *nlk = nlk_sk(sk); | |
1305 | int len, val, err; | |
1306 | ||
1307 | if (level != SOL_NETLINK) | |
1308 | return -ENOPROTOOPT; | |
1309 | ||
1310 | if (get_user(len, optlen)) | |
1311 | return -EFAULT; | |
1312 | if (len < 0) | |
1313 | return -EINVAL; | |
1314 | ||
1315 | switch (optname) { | |
1316 | case NETLINK_PKTINFO: | |
1317 | if (len < sizeof(int)) | |
1318 | return -EINVAL; | |
1319 | len = sizeof(int); | |
1320 | val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0; | |
a27b58fe HC |
1321 | if (put_user(len, optlen) || |
1322 | put_user(val, optval)) | |
1323 | return -EFAULT; | |
9a4595bc PM |
1324 | err = 0; |
1325 | break; | |
be0c22a4 PNA |
1326 | case NETLINK_BROADCAST_ERROR: |
1327 | if (len < sizeof(int)) | |
1328 | return -EINVAL; | |
1329 | len = sizeof(int); | |
1330 | val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0; | |
1331 | if (put_user(len, optlen) || | |
1332 | put_user(val, optval)) | |
1333 | return -EFAULT; | |
1334 | err = 0; | |
1335 | break; | |
38938bfe PNA |
1336 | case NETLINK_NO_ENOBUFS: |
1337 | if (len < sizeof(int)) | |
1338 | return -EINVAL; | |
1339 | len = sizeof(int); | |
1340 | val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0; | |
1341 | if (put_user(len, optlen) || | |
1342 | put_user(val, optval)) | |
1343 | return -EFAULT; | |
1344 | err = 0; | |
1345 | break; | |
9a4595bc PM |
1346 | default: |
1347 | err = -ENOPROTOOPT; | |
1348 | } | |
1349 | return err; | |
1350 | } | |
1351 | ||
1352 | static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb) | |
1353 | { | |
1354 | struct nl_pktinfo info; | |
1355 | ||
1356 | info.group = NETLINK_CB(skb).dst_group; | |
1357 | put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info); | |
1358 | } | |
1359 | ||
1da177e4 LT |
1360 | static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock, |
1361 | struct msghdr *msg, size_t len) | |
1362 | { | |
1363 | struct sock_iocb *siocb = kiocb_to_siocb(kiocb); | |
1364 | struct sock *sk = sock->sk; | |
1365 | struct netlink_sock *nlk = nlk_sk(sk); | |
6ac552fd | 1366 | struct sockaddr_nl *addr = msg->msg_name; |
15e47304 | 1367 | u32 dst_portid; |
d629b836 | 1368 | u32 dst_group; |
1da177e4 LT |
1369 | struct sk_buff *skb; |
1370 | int err; | |
1371 | struct scm_cookie scm; | |
1372 | ||
1373 | if (msg->msg_flags&MSG_OOB) | |
1374 | return -EOPNOTSUPP; | |
1375 | ||
16e57262 | 1376 | if (NULL == siocb->scm) |
1da177e4 | 1377 | siocb->scm = &scm; |
16e57262 | 1378 | |
e0e3cea4 | 1379 | err = scm_send(sock, msg, siocb->scm, true); |
1da177e4 LT |
1380 | if (err < 0) |
1381 | return err; | |
1382 | ||
1383 | if (msg->msg_namelen) { | |
b47030c7 | 1384 | err = -EINVAL; |
1da177e4 | 1385 | if (addr->nl_family != AF_NETLINK) |
b47030c7 | 1386 | goto out; |
15e47304 | 1387 | dst_portid = addr->nl_pid; |
d629b836 | 1388 | dst_group = ffs(addr->nl_groups); |
b47030c7 | 1389 | err = -EPERM; |
15e47304 | 1390 | if ((dst_group || dst_portid) && |
9785e10a | 1391 | !netlink_capable(sock, NL_CFG_F_NONROOT_SEND)) |
b47030c7 | 1392 | goto out; |
1da177e4 | 1393 | } else { |
15e47304 | 1394 | dst_portid = nlk->dst_portid; |
d629b836 | 1395 | dst_group = nlk->dst_group; |
1da177e4 LT |
1396 | } |
1397 | ||
15e47304 | 1398 | if (!nlk->portid) { |
1da177e4 LT |
1399 | err = netlink_autobind(sock); |
1400 | if (err) | |
1401 | goto out; | |
1402 | } | |
1403 | ||
1404 | err = -EMSGSIZE; | |
1405 | if (len > sk->sk_sndbuf - 32) | |
1406 | goto out; | |
1407 | err = -ENOBUFS; | |
339bf98f | 1408 | skb = alloc_skb(len, GFP_KERNEL); |
6ac552fd | 1409 | if (skb == NULL) |
1da177e4 LT |
1410 | goto out; |
1411 | ||
15e47304 | 1412 | NETLINK_CB(skb).portid = nlk->portid; |
d629b836 | 1413 | NETLINK_CB(skb).dst_group = dst_group; |
dbe9a417 | 1414 | NETLINK_CB(skb).creds = siocb->scm->creds; |
1da177e4 | 1415 | |
1da177e4 | 1416 | err = -EFAULT; |
6ac552fd | 1417 | if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) { |
1da177e4 LT |
1418 | kfree_skb(skb); |
1419 | goto out; | |
1420 | } | |
1421 | ||
1422 | err = security_netlink_send(sk, skb); | |
1423 | if (err) { | |
1424 | kfree_skb(skb); | |
1425 | goto out; | |
1426 | } | |
1427 | ||
d629b836 | 1428 | if (dst_group) { |
1da177e4 | 1429 | atomic_inc(&skb->users); |
15e47304 | 1430 | netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL); |
1da177e4 | 1431 | } |
15e47304 | 1432 | err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT); |
1da177e4 LT |
1433 | |
1434 | out: | |
b47030c7 | 1435 | scm_destroy(siocb->scm); |
1da177e4 LT |
1436 | return err; |
1437 | } | |
1438 | ||
1439 | static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock, | |
1440 | struct msghdr *msg, size_t len, | |
1441 | int flags) | |
1442 | { | |
1443 | struct sock_iocb *siocb = kiocb_to_siocb(kiocb); | |
1444 | struct scm_cookie scm; | |
1445 | struct sock *sk = sock->sk; | |
1446 | struct netlink_sock *nlk = nlk_sk(sk); | |
1447 | int noblock = flags&MSG_DONTWAIT; | |
1448 | size_t copied; | |
68d6ac6d | 1449 | struct sk_buff *skb, *data_skb; |
b44d211e | 1450 | int err, ret; |
1da177e4 LT |
1451 | |
1452 | if (flags&MSG_OOB) | |
1453 | return -EOPNOTSUPP; | |
1454 | ||
1455 | copied = 0; | |
1456 | ||
6ac552fd PM |
1457 | skb = skb_recv_datagram(sk, flags, noblock, &err); |
1458 | if (skb == NULL) | |
1da177e4 LT |
1459 | goto out; |
1460 | ||
68d6ac6d JB |
1461 | data_skb = skb; |
1462 | ||
1dacc76d JB |
1463 | #ifdef CONFIG_COMPAT_NETLINK_MESSAGES |
1464 | if (unlikely(skb_shinfo(skb)->frag_list)) { | |
1dacc76d | 1465 | /* |
68d6ac6d JB |
1466 | * If this skb has a frag_list, then here that means that we |
1467 | * will have to use the frag_list skb's data for compat tasks | |
1468 | * and the regular skb's data for normal (non-compat) tasks. | |
1dacc76d | 1469 | * |
68d6ac6d JB |
1470 | * If we need to send the compat skb, assign it to the |
1471 | * 'data_skb' variable so that it will be used below for data | |
1472 | * copying. We keep 'skb' for everything else, including | |
1473 | * freeing both later. | |
1dacc76d | 1474 | */ |
68d6ac6d JB |
1475 | if (flags & MSG_CMSG_COMPAT) |
1476 | data_skb = skb_shinfo(skb)->frag_list; | |
1dacc76d JB |
1477 | } |
1478 | #endif | |
1479 | ||
1da177e4 LT |
1480 | msg->msg_namelen = 0; |
1481 | ||
68d6ac6d | 1482 | copied = data_skb->len; |
1da177e4 LT |
1483 | if (len < copied) { |
1484 | msg->msg_flags |= MSG_TRUNC; | |
1485 | copied = len; | |
1486 | } | |
1487 | ||
68d6ac6d JB |
1488 | skb_reset_transport_header(data_skb); |
1489 | err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied); | |
1da177e4 LT |
1490 | |
1491 | if (msg->msg_name) { | |
6ac552fd | 1492 | struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name; |
1da177e4 LT |
1493 | addr->nl_family = AF_NETLINK; |
1494 | addr->nl_pad = 0; | |
15e47304 | 1495 | addr->nl_pid = NETLINK_CB(skb).portid; |
d629b836 | 1496 | addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group); |
1da177e4 LT |
1497 | msg->msg_namelen = sizeof(*addr); |
1498 | } | |
1499 | ||
cc9a06cd PM |
1500 | if (nlk->flags & NETLINK_RECV_PKTINFO) |
1501 | netlink_cmsg_recv_pktinfo(msg, skb); | |
1502 | ||
1da177e4 LT |
1503 | if (NULL == siocb->scm) { |
1504 | memset(&scm, 0, sizeof(scm)); | |
1505 | siocb->scm = &scm; | |
1506 | } | |
1507 | siocb->scm->creds = *NETLINK_CREDS(skb); | |
188ccb55 | 1508 | if (flags & MSG_TRUNC) |
68d6ac6d | 1509 | copied = data_skb->len; |
daa3766e | 1510 | |
1da177e4 LT |
1511 | skb_free_datagram(sk, skb); |
1512 | ||
b44d211e AV |
1513 | if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) { |
1514 | ret = netlink_dump(sk); | |
1515 | if (ret) { | |
1516 | sk->sk_err = ret; | |
1517 | sk->sk_error_report(sk); | |
1518 | } | |
1519 | } | |
1da177e4 LT |
1520 | |
1521 | scm_recv(sock, msg, siocb->scm, flags); | |
1da177e4 LT |
1522 | out: |
1523 | netlink_rcv_wake(sk); | |
1524 | return err ? : copied; | |
1525 | } | |
1526 | ||
1527 | static void netlink_data_ready(struct sock *sk, int len) | |
1528 | { | |
cd40b7d3 | 1529 | BUG(); |
1da177e4 LT |
1530 | } |
1531 | ||
1532 | /* | |
746fac4d | 1533 | * We export these functions to other modules. They provide a |
1da177e4 LT |
1534 | * complete set of kernel non-blocking support for message |
1535 | * queueing. | |
1536 | */ | |
1537 | ||
1538 | struct sock * | |
9f00d977 PNA |
1539 | __netlink_kernel_create(struct net *net, int unit, struct module *module, |
1540 | struct netlink_kernel_cfg *cfg) | |
1da177e4 LT |
1541 | { |
1542 | struct socket *sock; | |
1543 | struct sock *sk; | |
77247bbb | 1544 | struct netlink_sock *nlk; |
5c398dc8 | 1545 | struct listeners *listeners = NULL; |
a31f2d17 PNA |
1546 | struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL; |
1547 | unsigned int groups; | |
1da177e4 | 1548 | |
fab2caf6 | 1549 | BUG_ON(!nl_table); |
1da177e4 | 1550 | |
6ac552fd | 1551 | if (unit < 0 || unit >= MAX_LINKS) |
1da177e4 LT |
1552 | return NULL; |
1553 | ||
1554 | if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock)) | |
1555 | return NULL; | |
1556 | ||
23fe1866 PE |
1557 | /* |
1558 | * We have to just have a reference on the net from sk, but don't | |
1559 | * get_net it. Besides, we cannot get and then put the net here. | |
1560 | * So we create one inside init_net and the move it to net. | |
1561 | */ | |
1562 | ||
1563 | if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0) | |
1564 | goto out_sock_release_nosk; | |
1565 | ||
1566 | sk = sock->sk; | |
edf02087 | 1567 | sk_change_net(sk, net); |
4fdb3bb7 | 1568 | |
a31f2d17 | 1569 | if (!cfg || cfg->groups < 32) |
4277a083 | 1570 | groups = 32; |
a31f2d17 PNA |
1571 | else |
1572 | groups = cfg->groups; | |
4277a083 | 1573 | |
5c398dc8 | 1574 | listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL); |
4277a083 PM |
1575 | if (!listeners) |
1576 | goto out_sock_release; | |
1577 | ||
1da177e4 | 1578 | sk->sk_data_ready = netlink_data_ready; |
a31f2d17 PNA |
1579 | if (cfg && cfg->input) |
1580 | nlk_sk(sk)->netlink_rcv = cfg->input; | |
1da177e4 | 1581 | |
b4b51029 | 1582 | if (netlink_insert(sk, net, 0)) |
77247bbb | 1583 | goto out_sock_release; |
4fdb3bb7 | 1584 | |
77247bbb PM |
1585 | nlk = nlk_sk(sk); |
1586 | nlk->flags |= NETLINK_KERNEL_SOCKET; | |
4fdb3bb7 | 1587 | |
4fdb3bb7 | 1588 | netlink_table_grab(); |
b4b51029 EB |
1589 | if (!nl_table[unit].registered) { |
1590 | nl_table[unit].groups = groups; | |
5c398dc8 | 1591 | rcu_assign_pointer(nl_table[unit].listeners, listeners); |
b4b51029 EB |
1592 | nl_table[unit].cb_mutex = cb_mutex; |
1593 | nl_table[unit].module = module; | |
9785e10a PNA |
1594 | if (cfg) { |
1595 | nl_table[unit].bind = cfg->bind; | |
1596 | nl_table[unit].flags = cfg->flags; | |
1597 | } | |
b4b51029 | 1598 | nl_table[unit].registered = 1; |
f937f1f4 JJ |
1599 | } else { |
1600 | kfree(listeners); | |
869e58f8 | 1601 | nl_table[unit].registered++; |
b4b51029 | 1602 | } |
4fdb3bb7 | 1603 | netlink_table_ungrab(); |
77247bbb PM |
1604 | return sk; |
1605 | ||
4fdb3bb7 | 1606 | out_sock_release: |
4277a083 | 1607 | kfree(listeners); |
9dfbec1f | 1608 | netlink_kernel_release(sk); |
23fe1866 PE |
1609 | return NULL; |
1610 | ||
1611 | out_sock_release_nosk: | |
4fdb3bb7 | 1612 | sock_release(sock); |
77247bbb | 1613 | return NULL; |
1da177e4 | 1614 | } |
9f00d977 | 1615 | EXPORT_SYMBOL(__netlink_kernel_create); |
b7c6ba6e DL |
1616 | |
1617 | void | |
1618 | netlink_kernel_release(struct sock *sk) | |
1619 | { | |
edf02087 | 1620 | sk_release_kernel(sk); |
b7c6ba6e DL |
1621 | } |
1622 | EXPORT_SYMBOL(netlink_kernel_release); | |
1623 | ||
d136f1bd | 1624 | int __netlink_change_ngroups(struct sock *sk, unsigned int groups) |
b4ff4f04 | 1625 | { |
5c398dc8 | 1626 | struct listeners *new, *old; |
b4ff4f04 | 1627 | struct netlink_table *tbl = &nl_table[sk->sk_protocol]; |
b4ff4f04 JB |
1628 | |
1629 | if (groups < 32) | |
1630 | groups = 32; | |
1631 | ||
b4ff4f04 | 1632 | if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) { |
5c398dc8 ED |
1633 | new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC); |
1634 | if (!new) | |
d136f1bd | 1635 | return -ENOMEM; |
6d772ac5 | 1636 | old = nl_deref_protected(tbl->listeners); |
5c398dc8 ED |
1637 | memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups)); |
1638 | rcu_assign_pointer(tbl->listeners, new); | |
1639 | ||
37b6b935 | 1640 | kfree_rcu(old, rcu); |
b4ff4f04 JB |
1641 | } |
1642 | tbl->groups = groups; | |
1643 | ||
d136f1bd JB |
1644 | return 0; |
1645 | } | |
1646 | ||
1647 | /** | |
1648 | * netlink_change_ngroups - change number of multicast groups | |
1649 | * | |
1650 | * This changes the number of multicast groups that are available | |
1651 | * on a certain netlink family. Note that it is not possible to | |
1652 | * change the number of groups to below 32. Also note that it does | |
1653 | * not implicitly call netlink_clear_multicast_users() when the | |
1654 | * number of groups is reduced. | |
1655 | * | |
1656 | * @sk: The kernel netlink socket, as returned by netlink_kernel_create(). | |
1657 | * @groups: The new number of groups. | |
1658 | */ | |
1659 | int netlink_change_ngroups(struct sock *sk, unsigned int groups) | |
1660 | { | |
1661 | int err; | |
1662 | ||
1663 | netlink_table_grab(); | |
1664 | err = __netlink_change_ngroups(sk, groups); | |
b4ff4f04 | 1665 | netlink_table_ungrab(); |
d136f1bd | 1666 | |
b4ff4f04 JB |
1667 | return err; |
1668 | } | |
b4ff4f04 | 1669 | |
b8273570 JB |
1670 | void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group) |
1671 | { | |
1672 | struct sock *sk; | |
b8273570 JB |
1673 | struct netlink_table *tbl = &nl_table[ksk->sk_protocol]; |
1674 | ||
b67bfe0d | 1675 | sk_for_each_bound(sk, &tbl->mc_list) |
b8273570 JB |
1676 | netlink_update_socket_mc(nlk_sk(sk), group, 0); |
1677 | } | |
1678 | ||
84659eb5 JB |
1679 | /** |
1680 | * netlink_clear_multicast_users - kick off multicast listeners | |
1681 | * | |
1682 | * This function removes all listeners from the given group. | |
1683 | * @ksk: The kernel netlink socket, as returned by | |
1684 | * netlink_kernel_create(). | |
1685 | * @group: The multicast group to clear. | |
1686 | */ | |
1687 | void netlink_clear_multicast_users(struct sock *ksk, unsigned int group) | |
1688 | { | |
84659eb5 | 1689 | netlink_table_grab(); |
b8273570 | 1690 | __netlink_clear_multicast_users(ksk, group); |
84659eb5 JB |
1691 | netlink_table_ungrab(); |
1692 | } | |
84659eb5 | 1693 | |
a46621a3 | 1694 | struct nlmsghdr * |
15e47304 | 1695 | __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags) |
a46621a3 DV |
1696 | { |
1697 | struct nlmsghdr *nlh; | |
1698 | int size = NLMSG_LENGTH(len); | |
1699 | ||
1700 | nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size)); | |
1701 | nlh->nlmsg_type = type; | |
1702 | nlh->nlmsg_len = size; | |
1703 | nlh->nlmsg_flags = flags; | |
15e47304 | 1704 | nlh->nlmsg_pid = portid; |
a46621a3 DV |
1705 | nlh->nlmsg_seq = seq; |
1706 | if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0) | |
1707 | memset(NLMSG_DATA(nlh) + len, 0, NLMSG_ALIGN(size) - size); | |
1708 | return nlh; | |
1709 | } | |
1710 | EXPORT_SYMBOL(__nlmsg_put); | |
1711 | ||
1da177e4 LT |
1712 | /* |
1713 | * It looks a bit ugly. | |
1714 | * It would be better to create kernel thread. | |
1715 | */ | |
1716 | ||
1717 | static int netlink_dump(struct sock *sk) | |
1718 | { | |
1719 | struct netlink_sock *nlk = nlk_sk(sk); | |
1720 | struct netlink_callback *cb; | |
c7ac8679 | 1721 | struct sk_buff *skb = NULL; |
1da177e4 | 1722 | struct nlmsghdr *nlh; |
bf8b79e4 | 1723 | int len, err = -ENOBUFS; |
c7ac8679 | 1724 | int alloc_size; |
1da177e4 | 1725 | |
af65bdfc | 1726 | mutex_lock(nlk->cb_mutex); |
1da177e4 LT |
1727 | |
1728 | cb = nlk->cb; | |
1729 | if (cb == NULL) { | |
bf8b79e4 TG |
1730 | err = -EINVAL; |
1731 | goto errout_skb; | |
1da177e4 LT |
1732 | } |
1733 | ||
c7ac8679 GR |
1734 | alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE); |
1735 | ||
1736 | skb = sock_rmalloc(sk, alloc_size, 0, GFP_KERNEL); | |
1737 | if (!skb) | |
c63d6ea3 | 1738 | goto errout_skb; |
c7ac8679 | 1739 | |
1da177e4 LT |
1740 | len = cb->dump(skb, cb); |
1741 | ||
1742 | if (len > 0) { | |
af65bdfc | 1743 | mutex_unlock(nlk->cb_mutex); |
b1153f29 SH |
1744 | |
1745 | if (sk_filter(sk, skb)) | |
1746 | kfree_skb(skb); | |
4a7e7c2a ED |
1747 | else |
1748 | __netlink_sendskb(sk, skb); | |
1da177e4 LT |
1749 | return 0; |
1750 | } | |
1751 | ||
bf8b79e4 TG |
1752 | nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI); |
1753 | if (!nlh) | |
1754 | goto errout_skb; | |
1755 | ||
670dc283 JB |
1756 | nl_dump_check_consistent(cb, nlh); |
1757 | ||
bf8b79e4 TG |
1758 | memcpy(nlmsg_data(nlh), &len, sizeof(len)); |
1759 | ||
b1153f29 SH |
1760 | if (sk_filter(sk, skb)) |
1761 | kfree_skb(skb); | |
4a7e7c2a ED |
1762 | else |
1763 | __netlink_sendskb(sk, skb); | |
1da177e4 | 1764 | |
a8f74b22 TG |
1765 | if (cb->done) |
1766 | cb->done(cb); | |
1da177e4 | 1767 | nlk->cb = NULL; |
af65bdfc | 1768 | mutex_unlock(nlk->cb_mutex); |
1da177e4 | 1769 | |
6dc878a8 | 1770 | module_put(cb->module); |
bfb253c9 | 1771 | netlink_consume_callback(cb); |
1da177e4 | 1772 | return 0; |
1797754e | 1773 | |
bf8b79e4 | 1774 | errout_skb: |
af65bdfc | 1775 | mutex_unlock(nlk->cb_mutex); |
bf8b79e4 | 1776 | kfree_skb(skb); |
bf8b79e4 | 1777 | return err; |
1da177e4 LT |
1778 | } |
1779 | ||
6dc878a8 G |
1780 | int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb, |
1781 | const struct nlmsghdr *nlh, | |
1782 | struct netlink_dump_control *control) | |
1da177e4 LT |
1783 | { |
1784 | struct netlink_callback *cb; | |
1785 | struct sock *sk; | |
1786 | struct netlink_sock *nlk; | |
b44d211e | 1787 | int ret; |
1da177e4 | 1788 | |
0da974f4 | 1789 | cb = kzalloc(sizeof(*cb), GFP_KERNEL); |
1da177e4 LT |
1790 | if (cb == NULL) |
1791 | return -ENOBUFS; | |
1792 | ||
80d326fa PNA |
1793 | cb->dump = control->dump; |
1794 | cb->done = control->done; | |
1da177e4 | 1795 | cb->nlh = nlh; |
7175c883 | 1796 | cb->data = control->data; |
6dc878a8 | 1797 | cb->module = control->module; |
80d326fa | 1798 | cb->min_dump_alloc = control->min_dump_alloc; |
1da177e4 LT |
1799 | atomic_inc(&skb->users); |
1800 | cb->skb = skb; | |
1801 | ||
15e47304 | 1802 | sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid); |
1da177e4 LT |
1803 | if (sk == NULL) { |
1804 | netlink_destroy_callback(cb); | |
1805 | return -ECONNREFUSED; | |
1806 | } | |
1807 | nlk = nlk_sk(sk); | |
6dc878a8 | 1808 | |
af65bdfc | 1809 | mutex_lock(nlk->cb_mutex); |
6dc878a8 | 1810 | /* A dump is in progress... */ |
3f660d66 | 1811 | if (nlk->cb) { |
af65bdfc | 1812 | mutex_unlock(nlk->cb_mutex); |
1da177e4 | 1813 | netlink_destroy_callback(cb); |
6dc878a8 G |
1814 | ret = -EBUSY; |
1815 | goto out; | |
1da177e4 | 1816 | } |
6dc878a8 G |
1817 | /* add reference of module which cb->dump belongs to */ |
1818 | if (!try_module_get(cb->module)) { | |
1819 | mutex_unlock(nlk->cb_mutex); | |
1820 | netlink_destroy_callback(cb); | |
1821 | ret = -EPROTONOSUPPORT; | |
1822 | goto out; | |
1823 | } | |
1824 | ||
1da177e4 | 1825 | nlk->cb = cb; |
af65bdfc | 1826 | mutex_unlock(nlk->cb_mutex); |
1da177e4 | 1827 | |
b44d211e | 1828 | ret = netlink_dump(sk); |
6dc878a8 | 1829 | out: |
1da177e4 | 1830 | sock_put(sk); |
5c58298c | 1831 | |
b44d211e AV |
1832 | if (ret) |
1833 | return ret; | |
1834 | ||
5c58298c DL |
1835 | /* We successfully started a dump, by returning -EINTR we |
1836 | * signal not to send ACK even if it was requested. | |
1837 | */ | |
1838 | return -EINTR; | |
1da177e4 | 1839 | } |
6dc878a8 | 1840 | EXPORT_SYMBOL(__netlink_dump_start); |
1da177e4 LT |
1841 | |
1842 | void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err) | |
1843 | { | |
1844 | struct sk_buff *skb; | |
1845 | struct nlmsghdr *rep; | |
1846 | struct nlmsgerr *errmsg; | |
339bf98f | 1847 | size_t payload = sizeof(*errmsg); |
1da177e4 | 1848 | |
339bf98f TG |
1849 | /* error messages get the original request appened */ |
1850 | if (err) | |
1851 | payload += nlmsg_len(nlh); | |
1da177e4 | 1852 | |
339bf98f | 1853 | skb = nlmsg_new(payload, GFP_KERNEL); |
1da177e4 LT |
1854 | if (!skb) { |
1855 | struct sock *sk; | |
1856 | ||
3b1e0a65 | 1857 | sk = netlink_lookup(sock_net(in_skb->sk), |
b4b51029 | 1858 | in_skb->sk->sk_protocol, |
15e47304 | 1859 | NETLINK_CB(in_skb).portid); |
1da177e4 LT |
1860 | if (sk) { |
1861 | sk->sk_err = ENOBUFS; | |
1862 | sk->sk_error_report(sk); | |
1863 | sock_put(sk); | |
1864 | } | |
1865 | return; | |
1866 | } | |
1867 | ||
15e47304 | 1868 | rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, |
5dba93ae | 1869 | NLMSG_ERROR, payload, 0); |
bf8b79e4 | 1870 | errmsg = nlmsg_data(rep); |
1da177e4 | 1871 | errmsg->error = err; |
bf8b79e4 | 1872 | memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh)); |
15e47304 | 1873 | netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT); |
1da177e4 | 1874 | } |
6ac552fd | 1875 | EXPORT_SYMBOL(netlink_ack); |
1da177e4 | 1876 | |
cd40b7d3 | 1877 | int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *, |
1d00a4eb | 1878 | struct nlmsghdr *)) |
82ace47a | 1879 | { |
82ace47a TG |
1880 | struct nlmsghdr *nlh; |
1881 | int err; | |
1882 | ||
1883 | while (skb->len >= nlmsg_total_size(0)) { | |
cd40b7d3 DL |
1884 | int msglen; |
1885 | ||
b529ccf2 | 1886 | nlh = nlmsg_hdr(skb); |
d35b6856 | 1887 | err = 0; |
82ace47a | 1888 | |
ad8e4b75 | 1889 | if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len) |
82ace47a TG |
1890 | return 0; |
1891 | ||
d35b6856 TG |
1892 | /* Only requests are handled by the kernel */ |
1893 | if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) | |
5c58298c | 1894 | goto ack; |
45e7ae7f TG |
1895 | |
1896 | /* Skip control messages */ | |
1897 | if (nlh->nlmsg_type < NLMSG_MIN_TYPE) | |
5c58298c | 1898 | goto ack; |
d35b6856 | 1899 | |
1d00a4eb | 1900 | err = cb(skb, nlh); |
5c58298c DL |
1901 | if (err == -EINTR) |
1902 | goto skip; | |
1903 | ||
1904 | ack: | |
d35b6856 | 1905 | if (nlh->nlmsg_flags & NLM_F_ACK || err) |
82ace47a | 1906 | netlink_ack(skb, nlh, err); |
82ace47a | 1907 | |
5c58298c | 1908 | skip: |
6ac552fd | 1909 | msglen = NLMSG_ALIGN(nlh->nlmsg_len); |
cd40b7d3 DL |
1910 | if (msglen > skb->len) |
1911 | msglen = skb->len; | |
1912 | skb_pull(skb, msglen); | |
82ace47a TG |
1913 | } |
1914 | ||
1915 | return 0; | |
1916 | } | |
6ac552fd | 1917 | EXPORT_SYMBOL(netlink_rcv_skb); |
82ace47a | 1918 | |
d387f6ad TG |
1919 | /** |
1920 | * nlmsg_notify - send a notification netlink message | |
1921 | * @sk: netlink socket to use | |
1922 | * @skb: notification message | |
15e47304 | 1923 | * @portid: destination netlink portid for reports or 0 |
d387f6ad TG |
1924 | * @group: destination multicast group or 0 |
1925 | * @report: 1 to report back, 0 to disable | |
1926 | * @flags: allocation flags | |
1927 | */ | |
15e47304 | 1928 | int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid, |
d387f6ad TG |
1929 | unsigned int group, int report, gfp_t flags) |
1930 | { | |
1931 | int err = 0; | |
1932 | ||
1933 | if (group) { | |
15e47304 | 1934 | int exclude_portid = 0; |
d387f6ad TG |
1935 | |
1936 | if (report) { | |
1937 | atomic_inc(&skb->users); | |
15e47304 | 1938 | exclude_portid = portid; |
d387f6ad TG |
1939 | } |
1940 | ||
1ce85fe4 PNA |
1941 | /* errors reported via destination sk->sk_err, but propagate |
1942 | * delivery errors if NETLINK_BROADCAST_ERROR flag is set */ | |
15e47304 | 1943 | err = nlmsg_multicast(sk, skb, exclude_portid, group, flags); |
d387f6ad TG |
1944 | } |
1945 | ||
1ce85fe4 PNA |
1946 | if (report) { |
1947 | int err2; | |
1948 | ||
15e47304 | 1949 | err2 = nlmsg_unicast(sk, skb, portid); |
1ce85fe4 PNA |
1950 | if (!err || err == -ESRCH) |
1951 | err = err2; | |
1952 | } | |
d387f6ad TG |
1953 | |
1954 | return err; | |
1955 | } | |
6ac552fd | 1956 | EXPORT_SYMBOL(nlmsg_notify); |
d387f6ad | 1957 | |
1da177e4 LT |
1958 | #ifdef CONFIG_PROC_FS |
1959 | struct nl_seq_iter { | |
e372c414 | 1960 | struct seq_net_private p; |
1da177e4 LT |
1961 | int link; |
1962 | int hash_idx; | |
1963 | }; | |
1964 | ||
1965 | static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos) | |
1966 | { | |
1967 | struct nl_seq_iter *iter = seq->private; | |
1968 | int i, j; | |
1969 | struct sock *s; | |
1da177e4 LT |
1970 | loff_t off = 0; |
1971 | ||
6ac552fd | 1972 | for (i = 0; i < MAX_LINKS; i++) { |
15e47304 | 1973 | struct nl_portid_hash *hash = &nl_table[i].hash; |
1da177e4 LT |
1974 | |
1975 | for (j = 0; j <= hash->mask; j++) { | |
b67bfe0d | 1976 | sk_for_each(s, &hash->table[j]) { |
1218854a | 1977 | if (sock_net(s) != seq_file_net(seq)) |
b4b51029 | 1978 | continue; |
1da177e4 LT |
1979 | if (off == pos) { |
1980 | iter->link = i; | |
1981 | iter->hash_idx = j; | |
1982 | return s; | |
1983 | } | |
1984 | ++off; | |
1985 | } | |
1986 | } | |
1987 | } | |
1988 | return NULL; | |
1989 | } | |
1990 | ||
1991 | static void *netlink_seq_start(struct seq_file *seq, loff_t *pos) | |
9a429c49 | 1992 | __acquires(nl_table_lock) |
1da177e4 LT |
1993 | { |
1994 | read_lock(&nl_table_lock); | |
1995 | return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN; | |
1996 | } | |
1997 | ||
1998 | static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
1999 | { | |
2000 | struct sock *s; | |
2001 | struct nl_seq_iter *iter; | |
2002 | int i, j; | |
2003 | ||
2004 | ++*pos; | |
2005 | ||
2006 | if (v == SEQ_START_TOKEN) | |
2007 | return netlink_seq_socket_idx(seq, 0); | |
746fac4d | 2008 | |
b4b51029 EB |
2009 | iter = seq->private; |
2010 | s = v; | |
2011 | do { | |
2012 | s = sk_next(s); | |
1218854a | 2013 | } while (s && sock_net(s) != seq_file_net(seq)); |
1da177e4 LT |
2014 | if (s) |
2015 | return s; | |
2016 | ||
1da177e4 LT |
2017 | i = iter->link; |
2018 | j = iter->hash_idx + 1; | |
2019 | ||
2020 | do { | |
15e47304 | 2021 | struct nl_portid_hash *hash = &nl_table[i].hash; |
1da177e4 LT |
2022 | |
2023 | for (; j <= hash->mask; j++) { | |
2024 | s = sk_head(&hash->table[j]); | |
1218854a | 2025 | while (s && sock_net(s) != seq_file_net(seq)) |
b4b51029 | 2026 | s = sk_next(s); |
1da177e4 LT |
2027 | if (s) { |
2028 | iter->link = i; | |
2029 | iter->hash_idx = j; | |
2030 | return s; | |
2031 | } | |
2032 | } | |
2033 | ||
2034 | j = 0; | |
2035 | } while (++i < MAX_LINKS); | |
2036 | ||
2037 | return NULL; | |
2038 | } | |
2039 | ||
2040 | static void netlink_seq_stop(struct seq_file *seq, void *v) | |
9a429c49 | 2041 | __releases(nl_table_lock) |
1da177e4 LT |
2042 | { |
2043 | read_unlock(&nl_table_lock); | |
2044 | } | |
2045 | ||
2046 | ||
2047 | static int netlink_seq_show(struct seq_file *seq, void *v) | |
2048 | { | |
658cb354 | 2049 | if (v == SEQ_START_TOKEN) { |
1da177e4 LT |
2050 | seq_puts(seq, |
2051 | "sk Eth Pid Groups " | |
cf0aa4e0 | 2052 | "Rmem Wmem Dump Locks Drops Inode\n"); |
658cb354 | 2053 | } else { |
1da177e4 LT |
2054 | struct sock *s = v; |
2055 | struct netlink_sock *nlk = nlk_sk(s); | |
2056 | ||
9f1e0ad0 | 2057 | seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %pK %-8d %-8d %-8lu\n", |
1da177e4 LT |
2058 | s, |
2059 | s->sk_protocol, | |
15e47304 | 2060 | nlk->portid, |
513c2500 | 2061 | nlk->groups ? (u32)nlk->groups[0] : 0, |
31e6d363 ED |
2062 | sk_rmem_alloc_get(s), |
2063 | sk_wmem_alloc_get(s), | |
1da177e4 | 2064 | nlk->cb, |
38938bfe | 2065 | atomic_read(&s->sk_refcnt), |
cf0aa4e0 MY |
2066 | atomic_read(&s->sk_drops), |
2067 | sock_i_ino(s) | |
1da177e4 LT |
2068 | ); |
2069 | ||
2070 | } | |
2071 | return 0; | |
2072 | } | |
2073 | ||
56b3d975 | 2074 | static const struct seq_operations netlink_seq_ops = { |
1da177e4 LT |
2075 | .start = netlink_seq_start, |
2076 | .next = netlink_seq_next, | |
2077 | .stop = netlink_seq_stop, | |
2078 | .show = netlink_seq_show, | |
2079 | }; | |
2080 | ||
2081 | ||
2082 | static int netlink_seq_open(struct inode *inode, struct file *file) | |
2083 | { | |
e372c414 DL |
2084 | return seq_open_net(inode, file, &netlink_seq_ops, |
2085 | sizeof(struct nl_seq_iter)); | |
b4b51029 EB |
2086 | } |
2087 | ||
da7071d7 | 2088 | static const struct file_operations netlink_seq_fops = { |
1da177e4 LT |
2089 | .owner = THIS_MODULE, |
2090 | .open = netlink_seq_open, | |
2091 | .read = seq_read, | |
2092 | .llseek = seq_lseek, | |
e372c414 | 2093 | .release = seq_release_net, |
1da177e4 LT |
2094 | }; |
2095 | ||
2096 | #endif | |
2097 | ||
2098 | int netlink_register_notifier(struct notifier_block *nb) | |
2099 | { | |
e041c683 | 2100 | return atomic_notifier_chain_register(&netlink_chain, nb); |
1da177e4 | 2101 | } |
6ac552fd | 2102 | EXPORT_SYMBOL(netlink_register_notifier); |
1da177e4 LT |
2103 | |
2104 | int netlink_unregister_notifier(struct notifier_block *nb) | |
2105 | { | |
e041c683 | 2106 | return atomic_notifier_chain_unregister(&netlink_chain, nb); |
1da177e4 | 2107 | } |
6ac552fd | 2108 | EXPORT_SYMBOL(netlink_unregister_notifier); |
746fac4d | 2109 | |
90ddc4f0 | 2110 | static const struct proto_ops netlink_ops = { |
1da177e4 LT |
2111 | .family = PF_NETLINK, |
2112 | .owner = THIS_MODULE, | |
2113 | .release = netlink_release, | |
2114 | .bind = netlink_bind, | |
2115 | .connect = netlink_connect, | |
2116 | .socketpair = sock_no_socketpair, | |
2117 | .accept = sock_no_accept, | |
2118 | .getname = netlink_getname, | |
2119 | .poll = datagram_poll, | |
2120 | .ioctl = sock_no_ioctl, | |
2121 | .listen = sock_no_listen, | |
2122 | .shutdown = sock_no_shutdown, | |
9a4595bc PM |
2123 | .setsockopt = netlink_setsockopt, |
2124 | .getsockopt = netlink_getsockopt, | |
1da177e4 LT |
2125 | .sendmsg = netlink_sendmsg, |
2126 | .recvmsg = netlink_recvmsg, | |
2127 | .mmap = sock_no_mmap, | |
2128 | .sendpage = sock_no_sendpage, | |
2129 | }; | |
2130 | ||
ec1b4cf7 | 2131 | static const struct net_proto_family netlink_family_ops = { |
1da177e4 LT |
2132 | .family = PF_NETLINK, |
2133 | .create = netlink_create, | |
2134 | .owner = THIS_MODULE, /* for consistency 8) */ | |
2135 | }; | |
2136 | ||
4665079c | 2137 | static int __net_init netlink_net_init(struct net *net) |
b4b51029 EB |
2138 | { |
2139 | #ifdef CONFIG_PROC_FS | |
d4beaa66 | 2140 | if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops)) |
b4b51029 EB |
2141 | return -ENOMEM; |
2142 | #endif | |
2143 | return 0; | |
2144 | } | |
2145 | ||
4665079c | 2146 | static void __net_exit netlink_net_exit(struct net *net) |
b4b51029 EB |
2147 | { |
2148 | #ifdef CONFIG_PROC_FS | |
ece31ffd | 2149 | remove_proc_entry("netlink", net->proc_net); |
b4b51029 EB |
2150 | #endif |
2151 | } | |
2152 | ||
b963ea89 DM |
2153 | static void __init netlink_add_usersock_entry(void) |
2154 | { | |
5c398dc8 | 2155 | struct listeners *listeners; |
b963ea89 DM |
2156 | int groups = 32; |
2157 | ||
5c398dc8 | 2158 | listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL); |
b963ea89 | 2159 | if (!listeners) |
5c398dc8 | 2160 | panic("netlink_add_usersock_entry: Cannot allocate listeners\n"); |
b963ea89 DM |
2161 | |
2162 | netlink_table_grab(); | |
2163 | ||
2164 | nl_table[NETLINK_USERSOCK].groups = groups; | |
5c398dc8 | 2165 | rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners); |
b963ea89 DM |
2166 | nl_table[NETLINK_USERSOCK].module = THIS_MODULE; |
2167 | nl_table[NETLINK_USERSOCK].registered = 1; | |
9785e10a | 2168 | nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND; |
b963ea89 DM |
2169 | |
2170 | netlink_table_ungrab(); | |
2171 | } | |
2172 | ||
022cbae6 | 2173 | static struct pernet_operations __net_initdata netlink_net_ops = { |
b4b51029 EB |
2174 | .init = netlink_net_init, |
2175 | .exit = netlink_net_exit, | |
2176 | }; | |
2177 | ||
1da177e4 LT |
2178 | static int __init netlink_proto_init(void) |
2179 | { | |
1da177e4 | 2180 | int i; |
26ff5ddc | 2181 | unsigned long limit; |
1da177e4 LT |
2182 | unsigned int order; |
2183 | int err = proto_register(&netlink_proto, 0); | |
2184 | ||
2185 | if (err != 0) | |
2186 | goto out; | |
2187 | ||
fab25745 | 2188 | BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb)); |
1da177e4 | 2189 | |
0da974f4 | 2190 | nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL); |
fab2caf6 AM |
2191 | if (!nl_table) |
2192 | goto panic; | |
1da177e4 | 2193 | |
4481374c JB |
2194 | if (totalram_pages >= (128 * 1024)) |
2195 | limit = totalram_pages >> (21 - PAGE_SHIFT); | |
1da177e4 | 2196 | else |
4481374c | 2197 | limit = totalram_pages >> (23 - PAGE_SHIFT); |
1da177e4 | 2198 | |
26ff5ddc DC |
2199 | order = get_bitmask_order(limit) - 1 + PAGE_SHIFT; |
2200 | limit = (1UL << order) / sizeof(struct hlist_head); | |
2201 | order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1; | |
1da177e4 LT |
2202 | |
2203 | for (i = 0; i < MAX_LINKS; i++) { | |
15e47304 | 2204 | struct nl_portid_hash *hash = &nl_table[i].hash; |
1da177e4 | 2205 | |
15e47304 | 2206 | hash->table = nl_portid_hash_zalloc(1 * sizeof(*hash->table)); |
1da177e4 LT |
2207 | if (!hash->table) { |
2208 | while (i-- > 0) | |
15e47304 | 2209 | nl_portid_hash_free(nl_table[i].hash.table, |
1da177e4 LT |
2210 | 1 * sizeof(*hash->table)); |
2211 | kfree(nl_table); | |
fab2caf6 | 2212 | goto panic; |
1da177e4 | 2213 | } |
1da177e4 LT |
2214 | hash->max_shift = order; |
2215 | hash->shift = 0; | |
2216 | hash->mask = 0; | |
2217 | hash->rehash_time = jiffies; | |
2218 | } | |
2219 | ||
b963ea89 DM |
2220 | netlink_add_usersock_entry(); |
2221 | ||
1da177e4 | 2222 | sock_register(&netlink_family_ops); |
b4b51029 | 2223 | register_pernet_subsys(&netlink_net_ops); |
746fac4d | 2224 | /* The netlink device handler may be needed early. */ |
1da177e4 LT |
2225 | rtnetlink_init(); |
2226 | out: | |
2227 | return err; | |
fab2caf6 AM |
2228 | panic: |
2229 | panic("netlink_init: Cannot allocate nl_table\n"); | |
1da177e4 LT |
2230 | } |
2231 | ||
1da177e4 | 2232 | core_initcall(netlink_proto_init); |