]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/net/sunrpc/svcsock.c | |
3 | * | |
4 | * These are the RPC server socket internals. | |
5 | * | |
6 | * The server scheduling algorithm does not always distribute the load | |
7 | * evenly when servicing a single client. May need to modify the | |
8 | * svc_sock_enqueue procedure... | |
9 | * | |
10 | * TCP support is largely untested and may be a little slow. The problem | |
11 | * is that we currently do two separate recvfrom's, one for the 4-byte | |
12 | * record length, and the second for the actual record. This could possibly | |
13 | * be improved by always reading a minimum size of around 100 bytes and | |
14 | * tucking any superfluous bytes away in a temporary store. Still, that | |
15 | * leaves write requests out in the rain. An alternative may be to peek at | |
16 | * the first skb in the queue, and if it matches the next TCP sequence | |
17 | * number, to extract the record marker. Yuck. | |
18 | * | |
19 | * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> | |
20 | */ | |
21 | ||
172589cc | 22 | #include <linux/kernel.h> |
1da177e4 LT |
23 | #include <linux/sched.h> |
24 | #include <linux/errno.h> | |
25 | #include <linux/fcntl.h> | |
26 | #include <linux/net.h> | |
27 | #include <linux/in.h> | |
28 | #include <linux/inet.h> | |
29 | #include <linux/udp.h> | |
91483c4b | 30 | #include <linux/tcp.h> |
1da177e4 LT |
31 | #include <linux/unistd.h> |
32 | #include <linux/slab.h> | |
33 | #include <linux/netdevice.h> | |
34 | #include <linux/skbuff.h> | |
b41b66d6 | 35 | #include <linux/file.h> |
7dfb7103 | 36 | #include <linux/freezer.h> |
1da177e4 LT |
37 | #include <net/sock.h> |
38 | #include <net/checksum.h> | |
39 | #include <net/ip.h> | |
b92503b2 | 40 | #include <net/ipv6.h> |
c752f073 | 41 | #include <net/tcp_states.h> |
1da177e4 LT |
42 | #include <asm/uaccess.h> |
43 | #include <asm/ioctls.h> | |
44 | ||
45 | #include <linux/sunrpc/types.h> | |
ad06e4bd | 46 | #include <linux/sunrpc/clnt.h> |
1da177e4 LT |
47 | #include <linux/sunrpc/xdr.h> |
48 | #include <linux/sunrpc/svcsock.h> | |
49 | #include <linux/sunrpc/stats.h> | |
50 | ||
51 | /* SMP locking strategy: | |
52 | * | |
3262c816 GB |
53 | * svc_pool->sp_lock protects most of the fields of that pool. |
54 | * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt. | |
55 | * when both need to be taken (rare), svc_serv->sv_lock is first. | |
56 | * BKL protects svc_serv->sv_nrthread. | |
7ac1bea5 N |
57 | * svc_sock->sk_lock protects the svc_sock->sk_deferred list |
58 | * and the ->sk_info_authunix cache. | |
c081a0c7 | 59 | * svc_sock->sk_flags.SK_BUSY prevents a svc_sock being enqueued multiply. |
1da177e4 LT |
60 | * |
61 | * Some flags can be set to certain values at any time | |
62 | * providing that certain rules are followed: | |
63 | * | |
1da177e4 | 64 | * SK_CONN, SK_DATA, can be set or cleared at any time. |
cca5172a | 65 | * after a set, svc_sock_enqueue must be called. |
1da177e4 LT |
66 | * after a clear, the socket must be read/accepted |
67 | * if this succeeds, it must be set again. | |
68 | * SK_CLOSE can set at any time. It is never cleared. | |
aaf68cfb N |
69 | * sk_inuse contains a bias of '1' until SK_DEAD is set. |
70 | * so when sk_inuse hits zero, we know the socket is dead | |
71 | * and no-one is using it. | |
72 | * SK_DEAD can only be set while SK_BUSY is held which ensures | |
73 | * no other thread will be using the socket or will try to | |
74 | * set SK_DEAD. | |
1da177e4 LT |
75 | * |
76 | */ | |
77 | ||
78 | #define RPCDBG_FACILITY RPCDBG_SVCSOCK | |
79 | ||
80 | ||
81 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
6b174337 | 82 | int *errp, int flags); |
aaf68cfb | 83 | static void svc_delete_socket(struct svc_sock *svsk); |
1da177e4 LT |
84 | static void svc_udp_data_ready(struct sock *, int); |
85 | static int svc_udp_recvfrom(struct svc_rqst *); | |
86 | static int svc_udp_sendto(struct svc_rqst *); | |
cda1fd4a | 87 | static void svc_close_socket(struct svc_sock *svsk); |
1da177e4 LT |
88 | |
89 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk); | |
90 | static int svc_deferred_recv(struct svc_rqst *rqstp); | |
91 | static struct cache_deferred_req *svc_defer(struct cache_req *req); | |
92 | ||
36bdfc8b GB |
93 | /* apparently the "standard" is that clients close |
94 | * idle connections after 5 minutes, servers after | |
95 | * 6 minutes | |
96 | * http://www.connectathon.org/talks96/nfstcp.pdf | |
97 | */ | |
98 | static int svc_conn_age_period = 6*60; | |
99 | ||
ed07536e PZ |
100 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
101 | static struct lock_class_key svc_key[2]; | |
102 | static struct lock_class_key svc_slock_key[2]; | |
103 | ||
104 | static inline void svc_reclassify_socket(struct socket *sock) | |
105 | { | |
106 | struct sock *sk = sock->sk; | |
02b3d346 | 107 | BUG_ON(sock_owned_by_user(sk)); |
ed07536e PZ |
108 | switch (sk->sk_family) { |
109 | case AF_INET: | |
110 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
111 | &svc_slock_key[0], "sk_lock-AF_INET-NFSD", &svc_key[0]); | |
112 | break; | |
113 | ||
114 | case AF_INET6: | |
115 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
116 | &svc_slock_key[1], "sk_lock-AF_INET6-NFSD", &svc_key[1]); | |
117 | break; | |
118 | ||
119 | default: | |
120 | BUG(); | |
121 | } | |
122 | } | |
123 | #else | |
124 | static inline void svc_reclassify_socket(struct socket *sock) | |
125 | { | |
126 | } | |
127 | #endif | |
128 | ||
ad06e4bd CL |
129 | static char *__svc_print_addr(struct sockaddr *addr, char *buf, size_t len) |
130 | { | |
131 | switch (addr->sa_family) { | |
132 | case AF_INET: | |
133 | snprintf(buf, len, "%u.%u.%u.%u, port=%u", | |
134 | NIPQUAD(((struct sockaddr_in *) addr)->sin_addr), | |
582ee43d | 135 | ntohs(((struct sockaddr_in *) addr)->sin_port)); |
ad06e4bd | 136 | break; |
5a05ed73 | 137 | |
ad06e4bd CL |
138 | case AF_INET6: |
139 | snprintf(buf, len, "%x:%x:%x:%x:%x:%x:%x:%x, port=%u", | |
140 | NIP6(((struct sockaddr_in6 *) addr)->sin6_addr), | |
582ee43d | 141 | ntohs(((struct sockaddr_in6 *) addr)->sin6_port)); |
ad06e4bd | 142 | break; |
5a05ed73 | 143 | |
ad06e4bd CL |
144 | default: |
145 | snprintf(buf, len, "unknown address type: %d", addr->sa_family); | |
146 | break; | |
147 | } | |
148 | return buf; | |
149 | } | |
150 | ||
151 | /** | |
152 | * svc_print_addr - Format rq_addr field for printing | |
153 | * @rqstp: svc_rqst struct containing address to print | |
154 | * @buf: target buffer for formatted address | |
155 | * @len: length of target buffer | |
156 | * | |
157 | */ | |
158 | char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len) | |
159 | { | |
27459f09 | 160 | return __svc_print_addr(svc_addr(rqstp), buf, len); |
ad06e4bd CL |
161 | } |
162 | EXPORT_SYMBOL_GPL(svc_print_addr); | |
163 | ||
1da177e4 | 164 | /* |
3262c816 | 165 | * Queue up an idle server thread. Must have pool->sp_lock held. |
1da177e4 | 166 | * Note: this is really a stack rather than a queue, so that we only |
3262c816 | 167 | * use as many different threads as we need, and the rest don't pollute |
1da177e4 LT |
168 | * the cache. |
169 | */ | |
170 | static inline void | |
3262c816 | 171 | svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp) |
1da177e4 | 172 | { |
3262c816 | 173 | list_add(&rqstp->rq_list, &pool->sp_threads); |
1da177e4 LT |
174 | } |
175 | ||
176 | /* | |
3262c816 | 177 | * Dequeue an nfsd thread. Must have pool->sp_lock held. |
1da177e4 LT |
178 | */ |
179 | static inline void | |
3262c816 | 180 | svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp) |
1da177e4 LT |
181 | { |
182 | list_del(&rqstp->rq_list); | |
183 | } | |
184 | ||
185 | /* | |
186 | * Release an skbuff after use | |
187 | */ | |
188 | static inline void | |
189 | svc_release_skb(struct svc_rqst *rqstp) | |
190 | { | |
191 | struct sk_buff *skb = rqstp->rq_skbuff; | |
192 | struct svc_deferred_req *dr = rqstp->rq_deferred; | |
193 | ||
194 | if (skb) { | |
195 | rqstp->rq_skbuff = NULL; | |
196 | ||
197 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
198 | skb_free_datagram(rqstp->rq_sock->sk_sk, skb); | |
199 | } | |
200 | if (dr) { | |
201 | rqstp->rq_deferred = NULL; | |
202 | kfree(dr); | |
203 | } | |
204 | } | |
205 | ||
206 | /* | |
207 | * Any space to write? | |
208 | */ | |
209 | static inline unsigned long | |
210 | svc_sock_wspace(struct svc_sock *svsk) | |
211 | { | |
212 | int wspace; | |
213 | ||
214 | if (svsk->sk_sock->type == SOCK_STREAM) | |
215 | wspace = sk_stream_wspace(svsk->sk_sk); | |
216 | else | |
217 | wspace = sock_wspace(svsk->sk_sk); | |
218 | ||
219 | return wspace; | |
220 | } | |
221 | ||
222 | /* | |
223 | * Queue up a socket with data pending. If there are idle nfsd | |
224 | * processes, wake 'em up. | |
225 | * | |
226 | */ | |
227 | static void | |
228 | svc_sock_enqueue(struct svc_sock *svsk) | |
229 | { | |
230 | struct svc_serv *serv = svsk->sk_server; | |
bfd24160 | 231 | struct svc_pool *pool; |
1da177e4 | 232 | struct svc_rqst *rqstp; |
bfd24160 | 233 | int cpu; |
1da177e4 LT |
234 | |
235 | if (!(svsk->sk_flags & | |
236 | ( (1<<SK_CONN)|(1<<SK_DATA)|(1<<SK_CLOSE)|(1<<SK_DEFERRED)) )) | |
237 | return; | |
238 | if (test_bit(SK_DEAD, &svsk->sk_flags)) | |
239 | return; | |
240 | ||
bfd24160 GB |
241 | cpu = get_cpu(); |
242 | pool = svc_pool_for_cpu(svsk->sk_server, cpu); | |
243 | put_cpu(); | |
244 | ||
3262c816 | 245 | spin_lock_bh(&pool->sp_lock); |
1da177e4 | 246 | |
3262c816 GB |
247 | if (!list_empty(&pool->sp_threads) && |
248 | !list_empty(&pool->sp_sockets)) | |
1da177e4 LT |
249 | printk(KERN_ERR |
250 | "svc_sock_enqueue: threads and sockets both waiting??\n"); | |
251 | ||
252 | if (test_bit(SK_DEAD, &svsk->sk_flags)) { | |
253 | /* Don't enqueue dead sockets */ | |
254 | dprintk("svc: socket %p is dead, not enqueued\n", svsk->sk_sk); | |
255 | goto out_unlock; | |
256 | } | |
257 | ||
c081a0c7 GB |
258 | /* Mark socket as busy. It will remain in this state until the |
259 | * server has processed all pending data and put the socket back | |
260 | * on the idle list. We update SK_BUSY atomically because | |
261 | * it also guards against trying to enqueue the svc_sock twice. | |
262 | */ | |
263 | if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) { | |
264 | /* Don't enqueue socket while already enqueued */ | |
1da177e4 LT |
265 | dprintk("svc: socket %p busy, not enqueued\n", svsk->sk_sk); |
266 | goto out_unlock; | |
267 | } | |
3262c816 GB |
268 | BUG_ON(svsk->sk_pool != NULL); |
269 | svsk->sk_pool = pool; | |
1da177e4 LT |
270 | |
271 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
c6b0a9f8 | 272 | if (((atomic_read(&svsk->sk_reserved) + serv->sv_max_mesg)*2 |
1da177e4 LT |
273 | > svc_sock_wspace(svsk)) |
274 | && !test_bit(SK_CLOSE, &svsk->sk_flags) | |
275 | && !test_bit(SK_CONN, &svsk->sk_flags)) { | |
276 | /* Don't enqueue while not enough space for reply */ | |
277 | dprintk("svc: socket %p no space, %d*2 > %ld, not enqueued\n", | |
c6b0a9f8 | 278 | svsk->sk_sk, atomic_read(&svsk->sk_reserved)+serv->sv_max_mesg, |
1da177e4 | 279 | svc_sock_wspace(svsk)); |
3262c816 | 280 | svsk->sk_pool = NULL; |
c081a0c7 | 281 | clear_bit(SK_BUSY, &svsk->sk_flags); |
1da177e4 LT |
282 | goto out_unlock; |
283 | } | |
284 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
285 | ||
1da177e4 | 286 | |
3262c816 GB |
287 | if (!list_empty(&pool->sp_threads)) { |
288 | rqstp = list_entry(pool->sp_threads.next, | |
1da177e4 LT |
289 | struct svc_rqst, |
290 | rq_list); | |
291 | dprintk("svc: socket %p served by daemon %p\n", | |
292 | svsk->sk_sk, rqstp); | |
3262c816 | 293 | svc_thread_dequeue(pool, rqstp); |
1da177e4 | 294 | if (rqstp->rq_sock) |
cca5172a | 295 | printk(KERN_ERR |
1da177e4 LT |
296 | "svc_sock_enqueue: server %p, rq_sock=%p!\n", |
297 | rqstp, rqstp->rq_sock); | |
298 | rqstp->rq_sock = svsk; | |
c45c357d | 299 | atomic_inc(&svsk->sk_inuse); |
c6b0a9f8 | 300 | rqstp->rq_reserved = serv->sv_max_mesg; |
5685f0fa | 301 | atomic_add(rqstp->rq_reserved, &svsk->sk_reserved); |
3262c816 | 302 | BUG_ON(svsk->sk_pool != pool); |
1da177e4 LT |
303 | wake_up(&rqstp->rq_wait); |
304 | } else { | |
305 | dprintk("svc: socket %p put into queue\n", svsk->sk_sk); | |
3262c816 GB |
306 | list_add_tail(&svsk->sk_ready, &pool->sp_sockets); |
307 | BUG_ON(svsk->sk_pool != pool); | |
1da177e4 LT |
308 | } |
309 | ||
310 | out_unlock: | |
3262c816 | 311 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 LT |
312 | } |
313 | ||
314 | /* | |
3262c816 | 315 | * Dequeue the first socket. Must be called with the pool->sp_lock held. |
1da177e4 LT |
316 | */ |
317 | static inline struct svc_sock * | |
3262c816 | 318 | svc_sock_dequeue(struct svc_pool *pool) |
1da177e4 LT |
319 | { |
320 | struct svc_sock *svsk; | |
321 | ||
3262c816 | 322 | if (list_empty(&pool->sp_sockets)) |
1da177e4 LT |
323 | return NULL; |
324 | ||
3262c816 | 325 | svsk = list_entry(pool->sp_sockets.next, |
1da177e4 LT |
326 | struct svc_sock, sk_ready); |
327 | list_del_init(&svsk->sk_ready); | |
328 | ||
329 | dprintk("svc: socket %p dequeued, inuse=%d\n", | |
c45c357d | 330 | svsk->sk_sk, atomic_read(&svsk->sk_inuse)); |
1da177e4 LT |
331 | |
332 | return svsk; | |
333 | } | |
334 | ||
335 | /* | |
336 | * Having read something from a socket, check whether it | |
337 | * needs to be re-enqueued. | |
338 | * Note: SK_DATA only gets cleared when a read-attempt finds | |
339 | * no (or insufficient) data. | |
340 | */ | |
341 | static inline void | |
342 | svc_sock_received(struct svc_sock *svsk) | |
343 | { | |
3262c816 | 344 | svsk->sk_pool = NULL; |
1da177e4 LT |
345 | clear_bit(SK_BUSY, &svsk->sk_flags); |
346 | svc_sock_enqueue(svsk); | |
347 | } | |
348 | ||
349 | ||
350 | /** | |
351 | * svc_reserve - change the space reserved for the reply to a request. | |
352 | * @rqstp: The request in question | |
353 | * @space: new max space to reserve | |
354 | * | |
355 | * Each request reserves some space on the output queue of the socket | |
356 | * to make sure the reply fits. This function reduces that reserved | |
357 | * space to be the amount of space used already, plus @space. | |
358 | * | |
359 | */ | |
360 | void svc_reserve(struct svc_rqst *rqstp, int space) | |
361 | { | |
362 | space += rqstp->rq_res.head[0].iov_len; | |
363 | ||
364 | if (space < rqstp->rq_reserved) { | |
365 | struct svc_sock *svsk = rqstp->rq_sock; | |
5685f0fa | 366 | atomic_sub((rqstp->rq_reserved - space), &svsk->sk_reserved); |
1da177e4 | 367 | rqstp->rq_reserved = space; |
1da177e4 LT |
368 | |
369 | svc_sock_enqueue(svsk); | |
370 | } | |
371 | } | |
372 | ||
373 | /* | |
374 | * Release a socket after use. | |
375 | */ | |
376 | static inline void | |
377 | svc_sock_put(struct svc_sock *svsk) | |
378 | { | |
aaf68cfb N |
379 | if (atomic_dec_and_test(&svsk->sk_inuse)) { |
380 | BUG_ON(! test_bit(SK_DEAD, &svsk->sk_flags)); | |
381 | ||
202dd450 | 382 | dprintk("svc: releasing dead socket\n"); |
d6740df9 NB |
383 | if (svsk->sk_sock->file) |
384 | sockfd_put(svsk->sk_sock); | |
385 | else | |
386 | sock_release(svsk->sk_sock); | |
387 | if (svsk->sk_info_authunix != NULL) | |
388 | svcauth_unix_info_release(svsk->sk_info_authunix); | |
1da177e4 LT |
389 | kfree(svsk); |
390 | } | |
1da177e4 LT |
391 | } |
392 | ||
393 | static void | |
394 | svc_sock_release(struct svc_rqst *rqstp) | |
395 | { | |
396 | struct svc_sock *svsk = rqstp->rq_sock; | |
397 | ||
398 | svc_release_skb(rqstp); | |
399 | ||
44524359 | 400 | svc_free_res_pages(rqstp); |
1da177e4 LT |
401 | rqstp->rq_res.page_len = 0; |
402 | rqstp->rq_res.page_base = 0; | |
403 | ||
404 | ||
405 | /* Reset response buffer and release | |
406 | * the reservation. | |
407 | * But first, check that enough space was reserved | |
408 | * for the reply, otherwise we have a bug! | |
409 | */ | |
410 | if ((rqstp->rq_res.len) > rqstp->rq_reserved) | |
411 | printk(KERN_ERR "RPC request reserved %d but used %d\n", | |
412 | rqstp->rq_reserved, | |
413 | rqstp->rq_res.len); | |
414 | ||
415 | rqstp->rq_res.head[0].iov_len = 0; | |
416 | svc_reserve(rqstp, 0); | |
417 | rqstp->rq_sock = NULL; | |
418 | ||
419 | svc_sock_put(svsk); | |
420 | } | |
421 | ||
422 | /* | |
423 | * External function to wake up a server waiting for data | |
3262c816 GB |
424 | * This really only makes sense for services like lockd |
425 | * which have exactly one thread anyway. | |
1da177e4 LT |
426 | */ |
427 | void | |
428 | svc_wake_up(struct svc_serv *serv) | |
429 | { | |
430 | struct svc_rqst *rqstp; | |
3262c816 GB |
431 | unsigned int i; |
432 | struct svc_pool *pool; | |
433 | ||
434 | for (i = 0; i < serv->sv_nrpools; i++) { | |
435 | pool = &serv->sv_pools[i]; | |
436 | ||
437 | spin_lock_bh(&pool->sp_lock); | |
438 | if (!list_empty(&pool->sp_threads)) { | |
439 | rqstp = list_entry(pool->sp_threads.next, | |
440 | struct svc_rqst, | |
441 | rq_list); | |
442 | dprintk("svc: daemon %p woken up.\n", rqstp); | |
443 | /* | |
444 | svc_thread_dequeue(pool, rqstp); | |
445 | rqstp->rq_sock = NULL; | |
446 | */ | |
447 | wake_up(&rqstp->rq_wait); | |
448 | } | |
449 | spin_unlock_bh(&pool->sp_lock); | |
1da177e4 | 450 | } |
1da177e4 LT |
451 | } |
452 | ||
b92503b2 CL |
453 | union svc_pktinfo_u { |
454 | struct in_pktinfo pkti; | |
b92503b2 | 455 | struct in6_pktinfo pkti6; |
b92503b2 | 456 | }; |
bc375ea7 DM |
457 | #define SVC_PKTINFO_SPACE \ |
458 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
459 | |
460 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
461 | { | |
462 | switch (rqstp->rq_sock->sk_sk->sk_family) { | |
463 | case AF_INET: { | |
464 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
465 | ||
466 | cmh->cmsg_level = SOL_IP; | |
467 | cmh->cmsg_type = IP_PKTINFO; | |
468 | pki->ipi_ifindex = 0; | |
469 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; | |
470 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
471 | } | |
472 | break; | |
5a05ed73 | 473 | |
b92503b2 CL |
474 | case AF_INET6: { |
475 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
476 | ||
477 | cmh->cmsg_level = SOL_IPV6; | |
478 | cmh->cmsg_type = IPV6_PKTINFO; | |
479 | pki->ipi6_ifindex = 0; | |
480 | ipv6_addr_copy(&pki->ipi6_addr, | |
481 | &rqstp->rq_daddr.addr6); | |
482 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
483 | } | |
484 | break; | |
b92503b2 CL |
485 | } |
486 | return; | |
487 | } | |
488 | ||
1da177e4 LT |
489 | /* |
490 | * Generic sendto routine | |
491 | */ | |
492 | static int | |
493 | svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) | |
494 | { | |
495 | struct svc_sock *svsk = rqstp->rq_sock; | |
496 | struct socket *sock = svsk->sk_sock; | |
497 | int slen; | |
bc375ea7 DM |
498 | union { |
499 | struct cmsghdr hdr; | |
500 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
501 | } buffer; | |
502 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 LT |
503 | int len = 0; |
504 | int result; | |
505 | int size; | |
506 | struct page **ppage = xdr->pages; | |
507 | size_t base = xdr->page_base; | |
508 | unsigned int pglen = xdr->page_len; | |
509 | unsigned int flags = MSG_MORE; | |
ad06e4bd | 510 | char buf[RPC_MAX_ADDRBUFLEN]; |
1da177e4 LT |
511 | |
512 | slen = xdr->len; | |
513 | ||
514 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
b92503b2 CL |
515 | struct msghdr msg = { |
516 | .msg_name = &rqstp->rq_addr, | |
517 | .msg_namelen = rqstp->rq_addrlen, | |
518 | .msg_control = cmh, | |
519 | .msg_controllen = sizeof(buffer), | |
520 | .msg_flags = MSG_MORE, | |
521 | }; | |
522 | ||
523 | svc_set_cmsg_data(rqstp, cmh); | |
1da177e4 LT |
524 | |
525 | if (sock_sendmsg(sock, &msg, 0) < 0) | |
526 | goto out; | |
527 | } | |
528 | ||
529 | /* send head */ | |
530 | if (slen == xdr->head[0].iov_len) | |
531 | flags = 0; | |
44524359 N |
532 | len = kernel_sendpage(sock, rqstp->rq_respages[0], 0, |
533 | xdr->head[0].iov_len, flags); | |
1da177e4 LT |
534 | if (len != xdr->head[0].iov_len) |
535 | goto out; | |
536 | slen -= xdr->head[0].iov_len; | |
537 | if (slen == 0) | |
538 | goto out; | |
539 | ||
540 | /* send page data */ | |
541 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
542 | while (pglen > 0) { | |
543 | if (slen == size) | |
544 | flags = 0; | |
e6242e92 | 545 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
546 | if (result > 0) |
547 | len += result; | |
548 | if (result != size) | |
549 | goto out; | |
550 | slen -= size; | |
551 | pglen -= size; | |
552 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
553 | base = 0; | |
554 | ppage++; | |
555 | } | |
556 | /* send tail */ | |
557 | if (xdr->tail[0].iov_len) { | |
44524359 N |
558 | result = kernel_sendpage(sock, rqstp->rq_respages[0], |
559 | ((unsigned long)xdr->tail[0].iov_base) | |
cca5172a | 560 | & (PAGE_SIZE-1), |
1da177e4 LT |
561 | xdr->tail[0].iov_len, 0); |
562 | ||
563 | if (result > 0) | |
564 | len += result; | |
565 | } | |
566 | out: | |
ad06e4bd CL |
567 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", |
568 | rqstp->rq_sock, xdr->head[0].iov_base, xdr->head[0].iov_len, | |
569 | xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); | |
1da177e4 LT |
570 | |
571 | return len; | |
572 | } | |
573 | ||
80212d59 N |
574 | /* |
575 | * Report socket names for nfsdfs | |
576 | */ | |
577 | static int one_sock_name(char *buf, struct svc_sock *svsk) | |
578 | { | |
579 | int len; | |
580 | ||
581 | switch(svsk->sk_sk->sk_family) { | |
582 | case AF_INET: | |
583 | len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n", | |
584 | svsk->sk_sk->sk_protocol==IPPROTO_UDP? | |
585 | "udp" : "tcp", | |
586 | NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr), | |
587 | inet_sk(svsk->sk_sk)->num); | |
588 | break; | |
589 | default: | |
590 | len = sprintf(buf, "*unknown-%d*\n", | |
591 | svsk->sk_sk->sk_family); | |
592 | } | |
593 | return len; | |
594 | } | |
595 | ||
596 | int | |
b41b66d6 | 597 | svc_sock_names(char *buf, struct svc_serv *serv, char *toclose) |
80212d59 | 598 | { |
b41b66d6 | 599 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
600 | int len = 0; |
601 | ||
602 | if (!serv) | |
603 | return 0; | |
aaf68cfb | 604 | spin_lock_bh(&serv->sv_lock); |
80212d59 N |
605 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_list) { |
606 | int onelen = one_sock_name(buf+len, svsk); | |
b41b66d6 N |
607 | if (toclose && strcmp(toclose, buf+len) == 0) |
608 | closesk = svsk; | |
609 | else | |
610 | len += onelen; | |
80212d59 | 611 | } |
aaf68cfb | 612 | spin_unlock_bh(&serv->sv_lock); |
b41b66d6 | 613 | if (closesk) |
5680c446 N |
614 | /* Should unregister with portmap, but you cannot |
615 | * unregister just one protocol... | |
616 | */ | |
aaf68cfb | 617 | svc_close_socket(closesk); |
37a03472 N |
618 | else if (toclose) |
619 | return -ENOENT; | |
80212d59 N |
620 | return len; |
621 | } | |
622 | EXPORT_SYMBOL(svc_sock_names); | |
623 | ||
1da177e4 LT |
624 | /* |
625 | * Check input queue length | |
626 | */ | |
627 | static int | |
628 | svc_recv_available(struct svc_sock *svsk) | |
629 | { | |
1da177e4 LT |
630 | struct socket *sock = svsk->sk_sock; |
631 | int avail, err; | |
632 | ||
e6242e92 | 633 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
634 | |
635 | return (err >= 0)? avail : err; | |
636 | } | |
637 | ||
638 | /* | |
639 | * Generic recvfrom routine. | |
640 | */ | |
641 | static int | |
642 | svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, int buflen) | |
643 | { | |
067d7817 | 644 | struct svc_sock *svsk = rqstp->rq_sock; |
1ba95105 CL |
645 | struct msghdr msg = { |
646 | .msg_flags = MSG_DONTWAIT, | |
647 | }; | |
a9747692 | 648 | struct sockaddr *sin; |
1ba95105 | 649 | int len; |
1da177e4 | 650 | |
1ba95105 CL |
651 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
652 | msg.msg_flags); | |
1da177e4 LT |
653 | |
654 | /* sock_recvmsg doesn't fill in the name/namelen, so we must.. | |
1da177e4 | 655 | */ |
067d7817 CL |
656 | memcpy(&rqstp->rq_addr, &svsk->sk_remote, svsk->sk_remotelen); |
657 | rqstp->rq_addrlen = svsk->sk_remotelen; | |
1da177e4 | 658 | |
a9747692 FM |
659 | /* Destination address in request is needed for binding the |
660 | * source address in RPC callbacks later. | |
661 | */ | |
662 | sin = (struct sockaddr *)&svsk->sk_local; | |
663 | switch (sin->sa_family) { | |
664 | case AF_INET: | |
665 | rqstp->rq_daddr.addr = ((struct sockaddr_in *)sin)->sin_addr; | |
666 | break; | |
667 | case AF_INET6: | |
668 | rqstp->rq_daddr.addr6 = ((struct sockaddr_in6 *)sin)->sin6_addr; | |
669 | break; | |
670 | } | |
671 | ||
1da177e4 | 672 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", |
1ba95105 | 673 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
1da177e4 LT |
674 | |
675 | return len; | |
676 | } | |
677 | ||
678 | /* | |
679 | * Set socket snd and rcv buffer lengths | |
680 | */ | |
681 | static inline void | |
682 | svc_sock_setbufsize(struct socket *sock, unsigned int snd, unsigned int rcv) | |
683 | { | |
684 | #if 0 | |
685 | mm_segment_t oldfs; | |
686 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
687 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
688 | (char*)&snd, sizeof(snd)); | |
689 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
690 | (char*)&rcv, sizeof(rcv)); | |
691 | #else | |
692 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
693 | * which isn't acceptable. Until that is made conditional | |
694 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
695 | * DaveM said I could! | |
696 | */ | |
697 | lock_sock(sock->sk); | |
698 | sock->sk->sk_sndbuf = snd * 2; | |
699 | sock->sk->sk_rcvbuf = rcv * 2; | |
700 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; | |
701 | release_sock(sock->sk); | |
702 | #endif | |
703 | } | |
704 | /* | |
705 | * INET callback when data has been received on the socket. | |
706 | */ | |
707 | static void | |
708 | svc_udp_data_ready(struct sock *sk, int count) | |
709 | { | |
939bb7ef | 710 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 711 | |
939bb7ef NB |
712 | if (svsk) { |
713 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
714 | svsk, sk, count, test_bit(SK_BUSY, &svsk->sk_flags)); | |
715 | set_bit(SK_DATA, &svsk->sk_flags); | |
716 | svc_sock_enqueue(svsk); | |
717 | } | |
1da177e4 LT |
718 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
719 | wake_up_interruptible(sk->sk_sleep); | |
720 | } | |
721 | ||
722 | /* | |
723 | * INET callback when space is newly available on the socket. | |
724 | */ | |
725 | static void | |
726 | svc_write_space(struct sock *sk) | |
727 | { | |
728 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
729 | ||
730 | if (svsk) { | |
731 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
732 | svsk, sk, test_bit(SK_BUSY, &svsk->sk_flags)); | |
733 | svc_sock_enqueue(svsk); | |
734 | } | |
735 | ||
736 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { | |
939bb7ef | 737 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 LT |
738 | svsk); |
739 | wake_up_interruptible(sk->sk_sleep); | |
740 | } | |
741 | } | |
742 | ||
7a37f578 N |
743 | static inline void svc_udp_get_dest_address(struct svc_rqst *rqstp, |
744 | struct cmsghdr *cmh) | |
95756482 CL |
745 | { |
746 | switch (rqstp->rq_sock->sk_sk->sk_family) { | |
747 | case AF_INET: { | |
7a37f578 N |
748 | struct in_pktinfo *pki = CMSG_DATA(cmh); |
749 | rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; | |
95756482 | 750 | break; |
7a37f578 | 751 | } |
95756482 | 752 | case AF_INET6: { |
7a37f578 N |
753 | struct in6_pktinfo *pki = CMSG_DATA(cmh); |
754 | ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); | |
95756482 | 755 | break; |
7a37f578 | 756 | } |
95756482 | 757 | } |
95756482 CL |
758 | } |
759 | ||
1da177e4 LT |
760 | /* |
761 | * Receive a datagram from a UDP socket. | |
762 | */ | |
1da177e4 LT |
763 | static int |
764 | svc_udp_recvfrom(struct svc_rqst *rqstp) | |
765 | { | |
766 | struct svc_sock *svsk = rqstp->rq_sock; | |
767 | struct svc_serv *serv = svsk->sk_server; | |
768 | struct sk_buff *skb; | |
bc375ea7 DM |
769 | union { |
770 | struct cmsghdr hdr; | |
771 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
772 | } buffer; | |
773 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 | 774 | int err, len; |
7a37f578 N |
775 | struct msghdr msg = { |
776 | .msg_name = svc_addr(rqstp), | |
777 | .msg_control = cmh, | |
778 | .msg_controllen = sizeof(buffer), | |
779 | .msg_flags = MSG_DONTWAIT, | |
780 | }; | |
1da177e4 LT |
781 | |
782 | if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags)) | |
783 | /* udp sockets need large rcvbuf as all pending | |
784 | * requests are still in that buffer. sndbuf must | |
785 | * also be large enough that there is enough space | |
3262c816 GB |
786 | * for one reply per thread. We count all threads |
787 | * rather than threads in a particular pool, which | |
788 | * provides an upper bound on the number of threads | |
789 | * which will access the socket. | |
1da177e4 LT |
790 | */ |
791 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
792 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
793 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 LT |
794 | |
795 | if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) { | |
796 | svc_sock_received(svsk); | |
797 | return svc_deferred_recv(rqstp); | |
798 | } | |
799 | ||
aaf68cfb N |
800 | if (test_bit(SK_CLOSE, &svsk->sk_flags)) { |
801 | svc_delete_socket(svsk); | |
802 | return 0; | |
803 | } | |
804 | ||
1da177e4 | 805 | clear_bit(SK_DATA, &svsk->sk_flags); |
05ed690e N |
806 | skb = NULL; |
807 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, | |
808 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
809 | if (err >= 0) | |
810 | skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); | |
811 | ||
812 | if (skb == NULL) { | |
813 | if (err != -EAGAIN) { | |
814 | /* possibly an icmp error */ | |
815 | dprintk("svc: recvfrom returned error %d\n", -err); | |
816 | set_bit(SK_DATA, &svsk->sk_flags); | |
1da177e4 | 817 | } |
05ed690e N |
818 | svc_sock_received(svsk); |
819 | return -EAGAIN; | |
1da177e4 | 820 | } |
7a37f578 | 821 | rqstp->rq_addrlen = sizeof(rqstp->rq_addr); |
b7aa0bf7 ED |
822 | if (skb->tstamp.tv64 == 0) { |
823 | skb->tstamp = ktime_get_real(); | |
cca5172a | 824 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
825 | need that much accuracy */ |
826 | } | |
b7aa0bf7 | 827 | svsk->sk_sk->sk_stamp = skb->tstamp; |
1da177e4 LT |
828 | set_bit(SK_DATA, &svsk->sk_flags); /* there may be more data... */ |
829 | ||
830 | /* | |
831 | * Maybe more packets - kick another thread ASAP. | |
832 | */ | |
833 | svc_sock_received(svsk); | |
834 | ||
835 | len = skb->len - sizeof(struct udphdr); | |
836 | rqstp->rq_arg.len = len; | |
837 | ||
95756482 | 838 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 839 | |
7a37f578 N |
840 | if (cmh->cmsg_level != IPPROTO_IP || |
841 | cmh->cmsg_type != IP_PKTINFO) { | |
842 | if (net_ratelimit()) | |
843 | printk("rpcsvc: received unknown control message:" | |
844 | "%d/%d\n", | |
845 | cmh->cmsg_level, cmh->cmsg_type); | |
846 | skb_free_datagram(svsk->sk_sk, skb); | |
847 | return 0; | |
848 | } | |
849 | svc_udp_get_dest_address(rqstp, cmh); | |
1da177e4 LT |
850 | |
851 | if (skb_is_nonlinear(skb)) { | |
852 | /* we have to copy */ | |
853 | local_bh_disable(); | |
854 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
855 | local_bh_enable(); | |
856 | /* checksum error */ | |
857 | skb_free_datagram(svsk->sk_sk, skb); | |
858 | return 0; | |
859 | } | |
860 | local_bh_enable(); | |
cca5172a | 861 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 LT |
862 | } else { |
863 | /* we can use it in-place */ | |
864 | rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr); | |
865 | rqstp->rq_arg.head[0].iov_len = len; | |
fb286bb2 HX |
866 | if (skb_checksum_complete(skb)) { |
867 | skb_free_datagram(svsk->sk_sk, skb); | |
868 | return 0; | |
1da177e4 LT |
869 | } |
870 | rqstp->rq_skbuff = skb; | |
871 | } | |
872 | ||
873 | rqstp->rq_arg.page_base = 0; | |
874 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
875 | rqstp->rq_arg.head[0].iov_len = len; | |
876 | rqstp->rq_arg.page_len = 0; | |
44524359 | 877 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
878 | } else { |
879 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 880 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 881 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 LT |
882 | } |
883 | ||
884 | if (serv->sv_stats) | |
885 | serv->sv_stats->netudpcnt++; | |
886 | ||
887 | return len; | |
888 | } | |
889 | ||
890 | static int | |
891 | svc_udp_sendto(struct svc_rqst *rqstp) | |
892 | { | |
893 | int error; | |
894 | ||
895 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
896 | if (error == -ECONNREFUSED) | |
897 | /* ICMP error on earlier request. */ | |
898 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
899 | ||
900 | return error; | |
901 | } | |
902 | ||
903 | static void | |
904 | svc_udp_init(struct svc_sock *svsk) | |
905 | { | |
7a37f578 N |
906 | int one = 1; |
907 | mm_segment_t oldfs; | |
908 | ||
1da177e4 LT |
909 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
910 | svsk->sk_sk->sk_write_space = svc_write_space; | |
911 | svsk->sk_recvfrom = svc_udp_recvfrom; | |
912 | svsk->sk_sendto = svc_udp_sendto; | |
913 | ||
914 | /* initialise setting must have enough space to | |
cca5172a | 915 | * receive and respond to one request. |
1da177e4 LT |
916 | * svc_udp_recvfrom will re-adjust if necessary |
917 | */ | |
918 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
919 | 3 * svsk->sk_server->sv_max_mesg, |
920 | 3 * svsk->sk_server->sv_max_mesg); | |
1da177e4 LT |
921 | |
922 | set_bit(SK_DATA, &svsk->sk_flags); /* might have come in before data_ready set up */ | |
923 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
7a37f578 N |
924 | |
925 | oldfs = get_fs(); | |
926 | set_fs(KERNEL_DS); | |
927 | /* make sure we get destination address info */ | |
928 | svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO, | |
929 | (char __user *)&one, sizeof(one)); | |
930 | set_fs(oldfs); | |
1da177e4 LT |
931 | } |
932 | ||
933 | /* | |
934 | * A data_ready event on a listening socket means there's a connection | |
935 | * pending. Do not use state_change as a substitute for it. | |
936 | */ | |
937 | static void | |
938 | svc_tcp_listen_data_ready(struct sock *sk, int count_unused) | |
939 | { | |
939bb7ef | 940 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
941 | |
942 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 943 | sk, sk->sk_state); |
1da177e4 | 944 | |
939bb7ef NB |
945 | /* |
946 | * This callback may called twice when a new connection | |
947 | * is established as a child socket inherits everything | |
948 | * from a parent LISTEN socket. | |
949 | * 1) data_ready method of the parent socket will be called | |
950 | * when one of child sockets become ESTABLISHED. | |
951 | * 2) data_ready method of the child socket may be called | |
952 | * when it receives data before the socket is accepted. | |
953 | * In case of 2, we should ignore it silently. | |
954 | */ | |
955 | if (sk->sk_state == TCP_LISTEN) { | |
956 | if (svsk) { | |
957 | set_bit(SK_CONN, &svsk->sk_flags); | |
958 | svc_sock_enqueue(svsk); | |
959 | } else | |
960 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 961 | } |
939bb7ef | 962 | |
1da177e4 LT |
963 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
964 | wake_up_interruptible_all(sk->sk_sleep); | |
965 | } | |
966 | ||
967 | /* | |
968 | * A state change on a connected socket means it's dying or dead. | |
969 | */ | |
970 | static void | |
971 | svc_tcp_state_change(struct sock *sk) | |
972 | { | |
939bb7ef | 973 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
974 | |
975 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 976 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 977 | |
939bb7ef | 978 | if (!svsk) |
1da177e4 | 979 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef NB |
980 | else { |
981 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
982 | svc_sock_enqueue(svsk); | |
1da177e4 | 983 | } |
1da177e4 LT |
984 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
985 | wake_up_interruptible_all(sk->sk_sleep); | |
986 | } | |
987 | ||
988 | static void | |
989 | svc_tcp_data_ready(struct sock *sk, int count) | |
990 | { | |
939bb7ef | 991 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
992 | |
993 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
994 | sk, sk->sk_user_data); |
995 | if (svsk) { | |
996 | set_bit(SK_DATA, &svsk->sk_flags); | |
997 | svc_sock_enqueue(svsk); | |
998 | } | |
1da177e4 LT |
999 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
1000 | wake_up_interruptible(sk->sk_sleep); | |
1001 | } | |
1002 | ||
bcdb81ae CL |
1003 | static inline int svc_port_is_privileged(struct sockaddr *sin) |
1004 | { | |
1005 | switch (sin->sa_family) { | |
1006 | case AF_INET: | |
1007 | return ntohs(((struct sockaddr_in *)sin)->sin_port) | |
1008 | < PROT_SOCK; | |
bcdb81ae CL |
1009 | case AF_INET6: |
1010 | return ntohs(((struct sockaddr_in6 *)sin)->sin6_port) | |
1011 | < PROT_SOCK; | |
bcdb81ae CL |
1012 | default: |
1013 | return 0; | |
1014 | } | |
1015 | } | |
1016 | ||
1da177e4 LT |
1017 | /* |
1018 | * Accept a TCP connection | |
1019 | */ | |
1020 | static void | |
1021 | svc_tcp_accept(struct svc_sock *svsk) | |
1022 | { | |
cdd88b9f AM |
1023 | struct sockaddr_storage addr; |
1024 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
1da177e4 LT |
1025 | struct svc_serv *serv = svsk->sk_server; |
1026 | struct socket *sock = svsk->sk_sock; | |
1027 | struct socket *newsock; | |
1da177e4 LT |
1028 | struct svc_sock *newsvsk; |
1029 | int err, slen; | |
ad06e4bd | 1030 | char buf[RPC_MAX_ADDRBUFLEN]; |
1da177e4 LT |
1031 | |
1032 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
1033 | if (!sock) | |
1034 | return; | |
1035 | ||
e6242e92 SS |
1036 | clear_bit(SK_CONN, &svsk->sk_flags); |
1037 | err = kernel_accept(sock, &newsock, O_NONBLOCK); | |
1038 | if (err < 0) { | |
1da177e4 LT |
1039 | if (err == -ENOMEM) |
1040 | printk(KERN_WARNING "%s: no more sockets!\n", | |
1041 | serv->sv_name); | |
e6242e92 | 1042 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
1043 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
1044 | serv->sv_name, -err); | |
e6242e92 | 1045 | return; |
1da177e4 | 1046 | } |
e6242e92 | 1047 | |
1da177e4 LT |
1048 | set_bit(SK_CONN, &svsk->sk_flags); |
1049 | svc_sock_enqueue(svsk); | |
1050 | ||
cdd88b9f | 1051 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 LT |
1052 | if (err < 0) { |
1053 | if (net_ratelimit()) | |
1054 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
1055 | serv->sv_name, -err); | |
1056 | goto failed; /* aborted connection or whatever */ | |
1057 | } | |
1058 | ||
1059 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
1060 | * hosts here, but when we get encryption, the IP of the host won't |
1061 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 1062 | */ |
cdd88b9f | 1063 | if (!svc_port_is_privileged(sin)) { |
1da177e4 | 1064 | dprintk(KERN_WARNING |
ad06e4bd | 1065 | "%s: connect from unprivileged port: %s\n", |
cca5172a | 1066 | serv->sv_name, |
cdd88b9f | 1067 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 1068 | } |
ad06e4bd | 1069 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 1070 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1071 | |
1072 | /* make sure that a write doesn't block forever when | |
1073 | * low on memory | |
1074 | */ | |
1075 | newsock->sk->sk_sndtimeo = HZ*30; | |
1076 | ||
6b174337 CL |
1077 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
1078 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) | |
1da177e4 | 1079 | goto failed; |
cdd88b9f | 1080 | memcpy(&newsvsk->sk_remote, sin, slen); |
067d7817 | 1081 | newsvsk->sk_remotelen = slen; |
a9747692 FM |
1082 | err = kernel_getsockname(newsock, sin, &slen); |
1083 | if (unlikely(err < 0)) { | |
1084 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); | |
1085 | slen = offsetof(struct sockaddr, sa_data); | |
1086 | } | |
1087 | memcpy(&newsvsk->sk_local, sin, slen); | |
067d7817 | 1088 | |
e79eff1f | 1089 | svc_sock_received(newsvsk); |
1da177e4 LT |
1090 | |
1091 | /* make sure that we don't have too many active connections. | |
1092 | * If we have, something must be dropped. | |
1093 | * | |
1094 | * There's no point in trying to do random drop here for | |
1095 | * DoS prevention. The NFS clients does 1 reconnect in 15 | |
1096 | * seconds. An attacker can easily beat that. | |
1097 | * | |
1098 | * The only somewhat efficient mechanism would be if drop | |
1099 | * old connections from the same IP first. But right now | |
1100 | * we don't even record the client IP in svc_sock. | |
1101 | */ | |
1102 | if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) { | |
1103 | struct svc_sock *svsk = NULL; | |
1104 | spin_lock_bh(&serv->sv_lock); | |
1105 | if (!list_empty(&serv->sv_tempsocks)) { | |
1106 | if (net_ratelimit()) { | |
1107 | /* Try to help the admin */ | |
1108 | printk(KERN_NOTICE "%s: too many open TCP " | |
1109 | "sockets, consider increasing the " | |
1110 | "number of nfsd threads\n", | |
1111 | serv->sv_name); | |
ad06e4bd CL |
1112 | printk(KERN_NOTICE |
1113 | "%s: last TCP connect from %s\n", | |
9db619e6 WW |
1114 | serv->sv_name, __svc_print_addr(sin, |
1115 | buf, sizeof(buf))); | |
1da177e4 LT |
1116 | } |
1117 | /* | |
1118 | * Always select the oldest socket. It's not fair, | |
1119 | * but so is life | |
1120 | */ | |
1121 | svsk = list_entry(serv->sv_tempsocks.prev, | |
1122 | struct svc_sock, | |
1123 | sk_list); | |
1124 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
c45c357d | 1125 | atomic_inc(&svsk->sk_inuse); |
1da177e4 LT |
1126 | } |
1127 | spin_unlock_bh(&serv->sv_lock); | |
1128 | ||
1129 | if (svsk) { | |
1130 | svc_sock_enqueue(svsk); | |
1131 | svc_sock_put(svsk); | |
1132 | } | |
1133 | ||
1134 | } | |
1135 | ||
1136 | if (serv->sv_stats) | |
1137 | serv->sv_stats->nettcpconn++; | |
1138 | ||
1139 | return; | |
1140 | ||
1141 | failed: | |
1142 | sock_release(newsock); | |
1143 | return; | |
1144 | } | |
1145 | ||
1146 | /* | |
1147 | * Receive data from a TCP socket. | |
1148 | */ | |
1149 | static int | |
1150 | svc_tcp_recvfrom(struct svc_rqst *rqstp) | |
1151 | { | |
1152 | struct svc_sock *svsk = rqstp->rq_sock; | |
1153 | struct svc_serv *serv = svsk->sk_server; | |
1154 | int len; | |
3cc03b16 | 1155 | struct kvec *vec; |
1da177e4 LT |
1156 | int pnum, vlen; |
1157 | ||
1158 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
1159 | svsk, test_bit(SK_DATA, &svsk->sk_flags), | |
1160 | test_bit(SK_CONN, &svsk->sk_flags), | |
1161 | test_bit(SK_CLOSE, &svsk->sk_flags)); | |
1162 | ||
1163 | if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) { | |
1164 | svc_sock_received(svsk); | |
1165 | return svc_deferred_recv(rqstp); | |
1166 | } | |
1167 | ||
1168 | if (test_bit(SK_CLOSE, &svsk->sk_flags)) { | |
1169 | svc_delete_socket(svsk); | |
1170 | return 0; | |
1171 | } | |
1172 | ||
1a047060 | 1173 | if (svsk->sk_sk->sk_state == TCP_LISTEN) { |
1da177e4 LT |
1174 | svc_tcp_accept(svsk); |
1175 | svc_sock_received(svsk); | |
1176 | return 0; | |
1177 | } | |
1178 | ||
1179 | if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags)) | |
1180 | /* sndbuf needs to have room for one request | |
1181 | * per thread, otherwise we can stall even when the | |
1182 | * network isn't a bottleneck. | |
3262c816 GB |
1183 | * |
1184 | * We count all threads rather than threads in a | |
1185 | * particular pool, which provides an upper bound | |
1186 | * on the number of threads which will access the socket. | |
1187 | * | |
1da177e4 | 1188 | * rcvbuf just needs to be able to hold a few requests. |
cca5172a | 1189 | * Normally they will be removed from the queue |
1da177e4 LT |
1190 | * as soon a a complete request arrives. |
1191 | */ | |
1192 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
1193 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
1194 | 3 * serv->sv_max_mesg); | |
1da177e4 LT |
1195 | |
1196 | clear_bit(SK_DATA, &svsk->sk_flags); | |
1197 | ||
1198 | /* Receive data. If we haven't got the record length yet, get | |
1199 | * the next four bytes. Otherwise try to gobble up as much as | |
1200 | * possible up to the complete record length. | |
1201 | */ | |
1202 | if (svsk->sk_tcplen < 4) { | |
1203 | unsigned long want = 4 - svsk->sk_tcplen; | |
1204 | struct kvec iov; | |
1205 | ||
1206 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; | |
1207 | iov.iov_len = want; | |
1208 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
1209 | goto error; | |
1210 | svsk->sk_tcplen += len; | |
1211 | ||
1212 | if (len < want) { | |
1213 | dprintk("svc: short recvfrom while reading record length (%d of %lu)\n", | |
cca5172a | 1214 | len, want); |
1da177e4 LT |
1215 | svc_sock_received(svsk); |
1216 | return -EAGAIN; /* record header not complete */ | |
1217 | } | |
1218 | ||
1219 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
1220 | if (!(svsk->sk_reclen & 0x80000000)) { | |
1221 | /* FIXME: technically, a record can be fragmented, | |
1222 | * and non-terminal fragments will not have the top | |
1223 | * bit set in the fragment length header. | |
1224 | * But apparently no known nfs clients send fragmented | |
1225 | * records. */ | |
34e9a63b N |
1226 | if (net_ratelimit()) |
1227 | printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx" | |
1228 | " (non-terminal)\n", | |
1229 | (unsigned long) svsk->sk_reclen); | |
1da177e4 LT |
1230 | goto err_delete; |
1231 | } | |
1232 | svsk->sk_reclen &= 0x7fffffff; | |
1233 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); | |
c6b0a9f8 | 1234 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
34e9a63b N |
1235 | if (net_ratelimit()) |
1236 | printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx" | |
1237 | " (large)\n", | |
1238 | (unsigned long) svsk->sk_reclen); | |
1da177e4 LT |
1239 | goto err_delete; |
1240 | } | |
1241 | } | |
1242 | ||
1243 | /* Check whether enough data is available */ | |
1244 | len = svc_recv_available(svsk); | |
1245 | if (len < 0) | |
1246 | goto error; | |
1247 | ||
1248 | if (len < svsk->sk_reclen) { | |
1249 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
1250 | len, svsk->sk_reclen); | |
1251 | svc_sock_received(svsk); | |
1252 | return -EAGAIN; /* record not complete */ | |
1253 | } | |
1254 | len = svsk->sk_reclen; | |
1255 | set_bit(SK_DATA, &svsk->sk_flags); | |
1256 | ||
3cc03b16 | 1257 | vec = rqstp->rq_vec; |
1da177e4 LT |
1258 | vec[0] = rqstp->rq_arg.head[0]; |
1259 | vlen = PAGE_SIZE; | |
1260 | pnum = 1; | |
1261 | while (vlen < len) { | |
44524359 | 1262 | vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]); |
1da177e4 LT |
1263 | vec[pnum].iov_len = PAGE_SIZE; |
1264 | pnum++; | |
1265 | vlen += PAGE_SIZE; | |
1266 | } | |
44524359 | 1267 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
1268 | |
1269 | /* Now receive data */ | |
1270 | len = svc_recvfrom(rqstp, vec, pnum, len); | |
1271 | if (len < 0) | |
1272 | goto error; | |
1273 | ||
1274 | dprintk("svc: TCP complete record (%d bytes)\n", len); | |
1275 | rqstp->rq_arg.len = len; | |
1276 | rqstp->rq_arg.page_base = 0; | |
1277 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
1278 | rqstp->rq_arg.head[0].iov_len = len; | |
1279 | rqstp->rq_arg.page_len = 0; | |
1280 | } else { | |
1281 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
1282 | } | |
1283 | ||
1284 | rqstp->rq_skbuff = NULL; | |
1285 | rqstp->rq_prot = IPPROTO_TCP; | |
1286 | ||
1287 | /* Reset TCP read info */ | |
1288 | svsk->sk_reclen = 0; | |
1289 | svsk->sk_tcplen = 0; | |
1290 | ||
1291 | svc_sock_received(svsk); | |
1292 | if (serv->sv_stats) | |
1293 | serv->sv_stats->nettcpcnt++; | |
1294 | ||
1295 | return len; | |
1296 | ||
1297 | err_delete: | |
1298 | svc_delete_socket(svsk); | |
1299 | return -EAGAIN; | |
1300 | ||
1301 | error: | |
1302 | if (len == -EAGAIN) { | |
1303 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
1304 | svc_sock_received(svsk); | |
1305 | } else { | |
1306 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", | |
1307 | svsk->sk_server->sv_name, -len); | |
93fbf1a5 | 1308 | goto err_delete; |
1da177e4 LT |
1309 | } |
1310 | ||
1311 | return len; | |
1312 | } | |
1313 | ||
1314 | /* | |
1315 | * Send out data on TCP socket. | |
1316 | */ | |
1317 | static int | |
1318 | svc_tcp_sendto(struct svc_rqst *rqstp) | |
1319 | { | |
1320 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
1321 | int sent; | |
d8ed029d | 1322 | __be32 reclen; |
1da177e4 LT |
1323 | |
1324 | /* Set up the first element of the reply kvec. | |
1325 | * Any other kvecs that may be in use have been taken | |
1326 | * care of by the server implementation itself. | |
1327 | */ | |
1328 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
1329 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
1330 | ||
1331 | if (test_bit(SK_DEAD, &rqstp->rq_sock->sk_flags)) | |
1332 | return -ENOTCONN; | |
1333 | ||
1334 | sent = svc_sendto(rqstp, &rqstp->rq_res); | |
1335 | if (sent != xbufp->len) { | |
1336 | printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n", | |
1337 | rqstp->rq_sock->sk_server->sv_name, | |
1338 | (sent<0)?"got error":"sent only", | |
1339 | sent, xbufp->len); | |
aaf68cfb N |
1340 | set_bit(SK_CLOSE, &rqstp->rq_sock->sk_flags); |
1341 | svc_sock_enqueue(rqstp->rq_sock); | |
1da177e4 LT |
1342 | sent = -EAGAIN; |
1343 | } | |
1344 | return sent; | |
1345 | } | |
1346 | ||
1347 | static void | |
1348 | svc_tcp_init(struct svc_sock *svsk) | |
1349 | { | |
1350 | struct sock *sk = svsk->sk_sk; | |
1351 | struct tcp_sock *tp = tcp_sk(sk); | |
1352 | ||
1353 | svsk->sk_recvfrom = svc_tcp_recvfrom; | |
1354 | svsk->sk_sendto = svc_tcp_sendto; | |
1355 | ||
1356 | if (sk->sk_state == TCP_LISTEN) { | |
1357 | dprintk("setting up TCP socket for listening\n"); | |
1358 | sk->sk_data_ready = svc_tcp_listen_data_ready; | |
1359 | set_bit(SK_CONN, &svsk->sk_flags); | |
1360 | } else { | |
1361 | dprintk("setting up TCP socket for reading\n"); | |
1362 | sk->sk_state_change = svc_tcp_state_change; | |
1363 | sk->sk_data_ready = svc_tcp_data_ready; | |
1364 | sk->sk_write_space = svc_write_space; | |
1365 | ||
1366 | svsk->sk_reclen = 0; | |
1367 | svsk->sk_tcplen = 0; | |
1368 | ||
1369 | tp->nonagle = 1; /* disable Nagle's algorithm */ | |
1370 | ||
1371 | /* initialise setting must have enough space to | |
cca5172a | 1372 | * receive and respond to one request. |
1da177e4 LT |
1373 | * svc_tcp_recvfrom will re-adjust if necessary |
1374 | */ | |
1375 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
1376 | 3 * svsk->sk_server->sv_max_mesg, |
1377 | 3 * svsk->sk_server->sv_max_mesg); | |
1da177e4 LT |
1378 | |
1379 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1380 | set_bit(SK_DATA, &svsk->sk_flags); | |
cca5172a | 1381 | if (sk->sk_state != TCP_ESTABLISHED) |
1da177e4 LT |
1382 | set_bit(SK_CLOSE, &svsk->sk_flags); |
1383 | } | |
1384 | } | |
1385 | ||
1386 | void | |
1387 | svc_sock_update_bufs(struct svc_serv *serv) | |
1388 | { | |
1389 | /* | |
1390 | * The number of server threads has changed. Update | |
1391 | * rcvbuf and sndbuf accordingly on all sockets | |
1392 | */ | |
1393 | struct list_head *le; | |
1394 | ||
1395 | spin_lock_bh(&serv->sv_lock); | |
1396 | list_for_each(le, &serv->sv_permsocks) { | |
cca5172a | 1397 | struct svc_sock *svsk = |
1da177e4 LT |
1398 | list_entry(le, struct svc_sock, sk_list); |
1399 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1400 | } | |
1401 | list_for_each(le, &serv->sv_tempsocks) { | |
1402 | struct svc_sock *svsk = | |
1403 | list_entry(le, struct svc_sock, sk_list); | |
1404 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1405 | } | |
1406 | spin_unlock_bh(&serv->sv_lock); | |
1407 | } | |
1408 | ||
1409 | /* | |
3262c816 GB |
1410 | * Receive the next request on any socket. This code is carefully |
1411 | * organised not to touch any cachelines in the shared svc_serv | |
1412 | * structure, only cachelines in the local svc_pool. | |
1da177e4 LT |
1413 | */ |
1414 | int | |
6fb2b47f | 1415 | svc_recv(struct svc_rqst *rqstp, long timeout) |
1da177e4 | 1416 | { |
27459f09 | 1417 | struct svc_sock *svsk = NULL; |
6fb2b47f | 1418 | struct svc_serv *serv = rqstp->rq_server; |
3262c816 | 1419 | struct svc_pool *pool = rqstp->rq_pool; |
44524359 | 1420 | int len, i; |
1da177e4 LT |
1421 | int pages; |
1422 | struct xdr_buf *arg; | |
1423 | DECLARE_WAITQUEUE(wait, current); | |
1424 | ||
1425 | dprintk("svc: server %p waiting for data (to = %ld)\n", | |
1426 | rqstp, timeout); | |
1427 | ||
1428 | if (rqstp->rq_sock) | |
cca5172a | 1429 | printk(KERN_ERR |
1da177e4 LT |
1430 | "svc_recv: service %p, socket not NULL!\n", |
1431 | rqstp); | |
1432 | if (waitqueue_active(&rqstp->rq_wait)) | |
cca5172a | 1433 | printk(KERN_ERR |
1da177e4 LT |
1434 | "svc_recv: service %p, wait queue active!\n", |
1435 | rqstp); | |
1436 | ||
1da177e4 LT |
1437 | |
1438 | /* now allocate needed pages. If we get a failure, sleep briefly */ | |
c6b0a9f8 | 1439 | pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE; |
44524359 N |
1440 | for (i=0; i < pages ; i++) |
1441 | while (rqstp->rq_pages[i] == NULL) { | |
1442 | struct page *p = alloc_page(GFP_KERNEL); | |
1443 | if (!p) | |
1444 | schedule_timeout_uninterruptible(msecs_to_jiffies(500)); | |
1445 | rqstp->rq_pages[i] = p; | |
1da177e4 | 1446 | } |
250f3915 N |
1447 | rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */ |
1448 | BUG_ON(pages >= RPCSVC_MAXPAGES); | |
1da177e4 LT |
1449 | |
1450 | /* Make arg->head point to first page and arg->pages point to rest */ | |
1451 | arg = &rqstp->rq_arg; | |
44524359 | 1452 | arg->head[0].iov_base = page_address(rqstp->rq_pages[0]); |
1da177e4 | 1453 | arg->head[0].iov_len = PAGE_SIZE; |
44524359 | 1454 | arg->pages = rqstp->rq_pages + 1; |
1da177e4 LT |
1455 | arg->page_base = 0; |
1456 | /* save at least one page for response */ | |
1457 | arg->page_len = (pages-2)*PAGE_SIZE; | |
1458 | arg->len = (pages-1)*PAGE_SIZE; | |
1459 | arg->tail[0].iov_len = 0; | |
3e1d1d28 CL |
1460 | |
1461 | try_to_freeze(); | |
1887b935 | 1462 | cond_resched(); |
1da177e4 LT |
1463 | if (signalled()) |
1464 | return -EINTR; | |
1465 | ||
3262c816 GB |
1466 | spin_lock_bh(&pool->sp_lock); |
1467 | if ((svsk = svc_sock_dequeue(pool)) != NULL) { | |
1da177e4 | 1468 | rqstp->rq_sock = svsk; |
c45c357d | 1469 | atomic_inc(&svsk->sk_inuse); |
c6b0a9f8 | 1470 | rqstp->rq_reserved = serv->sv_max_mesg; |
5685f0fa | 1471 | atomic_add(rqstp->rq_reserved, &svsk->sk_reserved); |
1da177e4 LT |
1472 | } else { |
1473 | /* No data pending. Go to sleep */ | |
3262c816 | 1474 | svc_thread_enqueue(pool, rqstp); |
1da177e4 LT |
1475 | |
1476 | /* | |
1477 | * We have to be able to interrupt this wait | |
1478 | * to bring down the daemons ... | |
1479 | */ | |
1480 | set_current_state(TASK_INTERRUPTIBLE); | |
1481 | add_wait_queue(&rqstp->rq_wait, &wait); | |
3262c816 | 1482 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 LT |
1483 | |
1484 | schedule_timeout(timeout); | |
1485 | ||
3e1d1d28 | 1486 | try_to_freeze(); |
1da177e4 | 1487 | |
3262c816 | 1488 | spin_lock_bh(&pool->sp_lock); |
1da177e4 LT |
1489 | remove_wait_queue(&rqstp->rq_wait, &wait); |
1490 | ||
1491 | if (!(svsk = rqstp->rq_sock)) { | |
3262c816 GB |
1492 | svc_thread_dequeue(pool, rqstp); |
1493 | spin_unlock_bh(&pool->sp_lock); | |
1da177e4 LT |
1494 | dprintk("svc: server %p, no data yet\n", rqstp); |
1495 | return signalled()? -EINTR : -EAGAIN; | |
1496 | } | |
1497 | } | |
3262c816 | 1498 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 | 1499 | |
3262c816 GB |
1500 | dprintk("svc: server %p, pool %u, socket %p, inuse=%d\n", |
1501 | rqstp, pool->sp_id, svsk, atomic_read(&svsk->sk_inuse)); | |
1da177e4 LT |
1502 | len = svsk->sk_recvfrom(rqstp); |
1503 | dprintk("svc: got len=%d\n", len); | |
1504 | ||
1505 | /* No data, incomplete (TCP) read, or accept() */ | |
1506 | if (len == 0 || len == -EAGAIN) { | |
1507 | rqstp->rq_res.len = 0; | |
1508 | svc_sock_release(rqstp); | |
1509 | return -EAGAIN; | |
1510 | } | |
1511 | svsk->sk_lastrecv = get_seconds(); | |
36bdfc8b | 1512 | clear_bit(SK_OLD, &svsk->sk_flags); |
1da177e4 | 1513 | |
bcdb81ae | 1514 | rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp)); |
1da177e4 LT |
1515 | rqstp->rq_chandle.defer = svc_defer; |
1516 | ||
1517 | if (serv->sv_stats) | |
1518 | serv->sv_stats->netcnt++; | |
1519 | return len; | |
1520 | } | |
1521 | ||
cca5172a | 1522 | /* |
1da177e4 LT |
1523 | * Drop request |
1524 | */ | |
1525 | void | |
1526 | svc_drop(struct svc_rqst *rqstp) | |
1527 | { | |
1528 | dprintk("svc: socket %p dropped request\n", rqstp->rq_sock); | |
1529 | svc_sock_release(rqstp); | |
1530 | } | |
1531 | ||
1532 | /* | |
1533 | * Return reply to client. | |
1534 | */ | |
1535 | int | |
1536 | svc_send(struct svc_rqst *rqstp) | |
1537 | { | |
1538 | struct svc_sock *svsk; | |
1539 | int len; | |
1540 | struct xdr_buf *xb; | |
1541 | ||
1542 | if ((svsk = rqstp->rq_sock) == NULL) { | |
1543 | printk(KERN_WARNING "NULL socket pointer in %s:%d\n", | |
1544 | __FILE__, __LINE__); | |
1545 | return -EFAULT; | |
1546 | } | |
1547 | ||
1548 | /* release the receive skb before sending the reply */ | |
1549 | svc_release_skb(rqstp); | |
1550 | ||
1551 | /* calculate over-all length */ | |
1552 | xb = & rqstp->rq_res; | |
1553 | xb->len = xb->head[0].iov_len + | |
1554 | xb->page_len + | |
1555 | xb->tail[0].iov_len; | |
1556 | ||
57b47a53 IM |
1557 | /* Grab svsk->sk_mutex to serialize outgoing data. */ |
1558 | mutex_lock(&svsk->sk_mutex); | |
1da177e4 LT |
1559 | if (test_bit(SK_DEAD, &svsk->sk_flags)) |
1560 | len = -ENOTCONN; | |
1561 | else | |
1562 | len = svsk->sk_sendto(rqstp); | |
57b47a53 | 1563 | mutex_unlock(&svsk->sk_mutex); |
1da177e4 LT |
1564 | svc_sock_release(rqstp); |
1565 | ||
1566 | if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN) | |
1567 | return 0; | |
1568 | return len; | |
1569 | } | |
1570 | ||
36bdfc8b GB |
1571 | /* |
1572 | * Timer function to close old temporary sockets, using | |
1573 | * a mark-and-sweep algorithm. | |
1574 | */ | |
1575 | static void | |
1576 | svc_age_temp_sockets(unsigned long closure) | |
1577 | { | |
1578 | struct svc_serv *serv = (struct svc_serv *)closure; | |
1579 | struct svc_sock *svsk; | |
1580 | struct list_head *le, *next; | |
1581 | LIST_HEAD(to_be_aged); | |
1582 | ||
1583 | dprintk("svc_age_temp_sockets\n"); | |
1584 | ||
1585 | if (!spin_trylock_bh(&serv->sv_lock)) { | |
1586 | /* busy, try again 1 sec later */ | |
1587 | dprintk("svc_age_temp_sockets: busy\n"); | |
1588 | mod_timer(&serv->sv_temptimer, jiffies + HZ); | |
1589 | return; | |
1590 | } | |
1591 | ||
1592 | list_for_each_safe(le, next, &serv->sv_tempsocks) { | |
1593 | svsk = list_entry(le, struct svc_sock, sk_list); | |
1594 | ||
1595 | if (!test_and_set_bit(SK_OLD, &svsk->sk_flags)) | |
1596 | continue; | |
7a1fa065 | 1597 | if (atomic_read(&svsk->sk_inuse) > 1 || test_bit(SK_BUSY, &svsk->sk_flags)) |
36bdfc8b | 1598 | continue; |
c45c357d | 1599 | atomic_inc(&svsk->sk_inuse); |
36bdfc8b GB |
1600 | list_move(le, &to_be_aged); |
1601 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1602 | set_bit(SK_DETACHED, &svsk->sk_flags); | |
1603 | } | |
1604 | spin_unlock_bh(&serv->sv_lock); | |
1605 | ||
1606 | while (!list_empty(&to_be_aged)) { | |
1607 | le = to_be_aged.next; | |
1608 | /* fiddling the sk_list node is safe 'cos we're SK_DETACHED */ | |
1609 | list_del_init(le); | |
1610 | svsk = list_entry(le, struct svc_sock, sk_list); | |
1611 | ||
1612 | dprintk("queuing svsk %p for closing, %lu seconds old\n", | |
1613 | svsk, get_seconds() - svsk->sk_lastrecv); | |
1614 | ||
1615 | /* a thread will dequeue and close it soon */ | |
1616 | svc_sock_enqueue(svsk); | |
1617 | svc_sock_put(svsk); | |
1618 | } | |
1619 | ||
1620 | mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ); | |
1621 | } | |
1622 | ||
1da177e4 LT |
1623 | /* |
1624 | * Initialize socket for RPC use and create svc_sock struct | |
1625 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1626 | */ | |
6b174337 CL |
1627 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1628 | struct socket *sock, | |
1629 | int *errp, int flags) | |
1da177e4 LT |
1630 | { |
1631 | struct svc_sock *svsk; | |
1632 | struct sock *inet; | |
6b174337 CL |
1633 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
1634 | int is_temporary = flags & SVC_SOCK_TEMPORARY; | |
1da177e4 LT |
1635 | |
1636 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1637 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1638 | *errp = -ENOMEM; |
1639 | return NULL; | |
1640 | } | |
1da177e4 LT |
1641 | |
1642 | inet = sock->sk; | |
1643 | ||
1644 | /* Register socket with portmapper */ | |
1645 | if (*errp >= 0 && pmap_register) | |
1646 | *errp = svc_register(serv, inet->sk_protocol, | |
1647 | ntohs(inet_sk(inet)->sport)); | |
1648 | ||
1649 | if (*errp < 0) { | |
1650 | kfree(svsk); | |
1651 | return NULL; | |
1652 | } | |
1653 | ||
1654 | set_bit(SK_BUSY, &svsk->sk_flags); | |
1655 | inet->sk_user_data = svsk; | |
1656 | svsk->sk_sock = sock; | |
1657 | svsk->sk_sk = inet; | |
1658 | svsk->sk_ostate = inet->sk_state_change; | |
1659 | svsk->sk_odata = inet->sk_data_ready; | |
1660 | svsk->sk_owspace = inet->sk_write_space; | |
1661 | svsk->sk_server = serv; | |
aaf68cfb | 1662 | atomic_set(&svsk->sk_inuse, 1); |
1da177e4 | 1663 | svsk->sk_lastrecv = get_seconds(); |
7ac1bea5 | 1664 | spin_lock_init(&svsk->sk_lock); |
1da177e4 LT |
1665 | INIT_LIST_HEAD(&svsk->sk_deferred); |
1666 | INIT_LIST_HEAD(&svsk->sk_ready); | |
57b47a53 | 1667 | mutex_init(&svsk->sk_mutex); |
1da177e4 LT |
1668 | |
1669 | /* Initialize the socket */ | |
1670 | if (sock->type == SOCK_DGRAM) | |
1671 | svc_udp_init(svsk); | |
1672 | else | |
1673 | svc_tcp_init(svsk); | |
1674 | ||
1675 | spin_lock_bh(&serv->sv_lock); | |
6b174337 | 1676 | if (is_temporary) { |
1da177e4 LT |
1677 | set_bit(SK_TEMP, &svsk->sk_flags); |
1678 | list_add(&svsk->sk_list, &serv->sv_tempsocks); | |
1679 | serv->sv_tmpcnt++; | |
36bdfc8b GB |
1680 | if (serv->sv_temptimer.function == NULL) { |
1681 | /* setup timer to age temp sockets */ | |
1682 | setup_timer(&serv->sv_temptimer, svc_age_temp_sockets, | |
1683 | (unsigned long)serv); | |
1684 | mod_timer(&serv->sv_temptimer, | |
1685 | jiffies + svc_conn_age_period * HZ); | |
1686 | } | |
1da177e4 LT |
1687 | } else { |
1688 | clear_bit(SK_TEMP, &svsk->sk_flags); | |
1689 | list_add(&svsk->sk_list, &serv->sv_permsocks); | |
1690 | } | |
1691 | spin_unlock_bh(&serv->sv_lock); | |
1692 | ||
1693 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", | |
1694 | svsk, svsk->sk_sk); | |
1695 | ||
1da177e4 LT |
1696 | return svsk; |
1697 | } | |
1698 | ||
b41b66d6 N |
1699 | int svc_addsock(struct svc_serv *serv, |
1700 | int fd, | |
1701 | char *name_return, | |
1702 | int *proto) | |
1703 | { | |
1704 | int err = 0; | |
1705 | struct socket *so = sockfd_lookup(fd, &err); | |
1706 | struct svc_sock *svsk = NULL; | |
1707 | ||
1708 | if (!so) | |
1709 | return err; | |
1710 | if (so->sk->sk_family != AF_INET) | |
1711 | err = -EAFNOSUPPORT; | |
1712 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1713 | so->sk->sk_protocol != IPPROTO_UDP) | |
1714 | err = -EPROTONOSUPPORT; | |
1715 | else if (so->state > SS_UNCONNECTED) | |
1716 | err = -EISCONN; | |
1717 | else { | |
6b174337 | 1718 | svsk = svc_setup_socket(serv, so, &err, SVC_SOCK_DEFAULTS); |
e79eff1f N |
1719 | if (svsk) { |
1720 | svc_sock_received(svsk); | |
b41b66d6 | 1721 | err = 0; |
e79eff1f | 1722 | } |
b41b66d6 N |
1723 | } |
1724 | if (err) { | |
1725 | sockfd_put(so); | |
1726 | return err; | |
1727 | } | |
1728 | if (proto) *proto = so->sk->sk_protocol; | |
1729 | return one_sock_name(name_return, svsk); | |
1730 | } | |
1731 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1732 | ||
1da177e4 LT |
1733 | /* |
1734 | * Create socket for RPC service. | |
1735 | */ | |
6b174337 | 1736 | static int svc_create_socket(struct svc_serv *serv, int protocol, |
77f1f67a | 1737 | struct sockaddr *sin, int len, int flags) |
1da177e4 LT |
1738 | { |
1739 | struct svc_sock *svsk; | |
1740 | struct socket *sock; | |
1741 | int error; | |
1742 | int type; | |
ad06e4bd | 1743 | char buf[RPC_MAX_ADDRBUFLEN]; |
1da177e4 | 1744 | |
ad06e4bd CL |
1745 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1746 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1747 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1748 | |
1749 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1750 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1751 | "sockets supported\n"); | |
1752 | return -EINVAL; | |
1753 | } | |
1754 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; | |
1755 | ||
77f1f67a CL |
1756 | error = sock_create_kern(sin->sa_family, type, protocol, &sock); |
1757 | if (error < 0) | |
1da177e4 LT |
1758 | return error; |
1759 | ||
ed07536e PZ |
1760 | svc_reclassify_socket(sock); |
1761 | ||
18114746 | 1762 | if (type == SOCK_STREAM) |
77f1f67a CL |
1763 | sock->sk->sk_reuse = 1; /* allow address reuse */ |
1764 | error = kernel_bind(sock, sin, len); | |
18114746 ES |
1765 | if (error < 0) |
1766 | goto bummer; | |
1da177e4 LT |
1767 | |
1768 | if (protocol == IPPROTO_TCP) { | |
e6242e92 | 1769 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1770 | goto bummer; |
1771 | } | |
1772 | ||
e79eff1f N |
1773 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { |
1774 | svc_sock_received(svsk); | |
6b174337 | 1775 | return ntohs(inet_sk(svsk->sk_sk)->sport); |
e79eff1f | 1776 | } |
1da177e4 LT |
1777 | |
1778 | bummer: | |
1779 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1780 | sock_release(sock); | |
1781 | return error; | |
1782 | } | |
1783 | ||
1784 | /* | |
1785 | * Remove a dead socket | |
1786 | */ | |
aaf68cfb | 1787 | static void |
1da177e4 LT |
1788 | svc_delete_socket(struct svc_sock *svsk) |
1789 | { | |
1790 | struct svc_serv *serv; | |
1791 | struct sock *sk; | |
1792 | ||
1793 | dprintk("svc: svc_delete_socket(%p)\n", svsk); | |
1794 | ||
1795 | serv = svsk->sk_server; | |
1796 | sk = svsk->sk_sk; | |
1797 | ||
1798 | sk->sk_state_change = svsk->sk_ostate; | |
1799 | sk->sk_data_ready = svsk->sk_odata; | |
1800 | sk->sk_write_space = svsk->sk_owspace; | |
1801 | ||
1802 | spin_lock_bh(&serv->sv_lock); | |
1803 | ||
36bdfc8b GB |
1804 | if (!test_and_set_bit(SK_DETACHED, &svsk->sk_flags)) |
1805 | list_del_init(&svsk->sk_list); | |
cca5172a | 1806 | /* |
3262c816 GB |
1807 | * We used to delete the svc_sock from whichever list |
1808 | * it's sk_ready node was on, but we don't actually | |
1809 | * need to. This is because the only time we're called | |
1810 | * while still attached to a queue, the queue itself | |
1811 | * is about to be destroyed (in svc_destroy). | |
1812 | */ | |
aaf68cfb N |
1813 | if (!test_and_set_bit(SK_DEAD, &svsk->sk_flags)) { |
1814 | BUG_ON(atomic_read(&svsk->sk_inuse)<2); | |
1815 | atomic_dec(&svsk->sk_inuse); | |
1da177e4 LT |
1816 | if (test_bit(SK_TEMP, &svsk->sk_flags)) |
1817 | serv->sv_tmpcnt--; | |
aaf68cfb | 1818 | } |
1da177e4 | 1819 | |
d6740df9 | 1820 | spin_unlock_bh(&serv->sv_lock); |
aaf68cfb N |
1821 | } |
1822 | ||
cda1fd4a | 1823 | static void svc_close_socket(struct svc_sock *svsk) |
aaf68cfb N |
1824 | { |
1825 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1826 | if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) | |
1827 | /* someone else will have to effect the close */ | |
1828 | return; | |
1829 | ||
1830 | atomic_inc(&svsk->sk_inuse); | |
1831 | svc_delete_socket(svsk); | |
1832 | clear_bit(SK_BUSY, &svsk->sk_flags); | |
d6740df9 | 1833 | svc_sock_put(svsk); |
1da177e4 LT |
1834 | } |
1835 | ||
cda1fd4a N |
1836 | void svc_force_close_socket(struct svc_sock *svsk) |
1837 | { | |
1838 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1839 | if (test_bit(SK_BUSY, &svsk->sk_flags)) { | |
1840 | /* Waiting to be processed, but no threads left, | |
1841 | * So just remove it from the waiting list | |
1842 | */ | |
1843 | list_del_init(&svsk->sk_ready); | |
1844 | clear_bit(SK_BUSY, &svsk->sk_flags); | |
1845 | } | |
1846 | svc_close_socket(svsk); | |
1847 | } | |
1848 | ||
6b174337 CL |
1849 | /** |
1850 | * svc_makesock - Make a socket for nfsd and lockd | |
1851 | * @serv: RPC server structure | |
1852 | * @protocol: transport protocol to use | |
1853 | * @port: port to use | |
482fb94e | 1854 | * @flags: requested socket characteristics |
6b174337 | 1855 | * |
1da177e4 | 1856 | */ |
482fb94e CL |
1857 | int svc_makesock(struct svc_serv *serv, int protocol, unsigned short port, |
1858 | int flags) | |
1da177e4 | 1859 | { |
6b174337 CL |
1860 | struct sockaddr_in sin = { |
1861 | .sin_family = AF_INET, | |
1862 | .sin_addr.s_addr = INADDR_ANY, | |
1863 | .sin_port = htons(port), | |
1864 | }; | |
1da177e4 LT |
1865 | |
1866 | dprintk("svc: creating socket proto = %d\n", protocol); | |
77f1f67a CL |
1867 | return svc_create_socket(serv, protocol, (struct sockaddr *) &sin, |
1868 | sizeof(sin), flags); | |
1da177e4 LT |
1869 | } |
1870 | ||
1871 | /* | |
cca5172a | 1872 | * Handle defer and revisit of requests |
1da177e4 LT |
1873 | */ |
1874 | ||
1875 | static void svc_revisit(struct cache_deferred_req *dreq, int too_many) | |
1876 | { | |
1877 | struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle); | |
1da177e4 LT |
1878 | struct svc_sock *svsk; |
1879 | ||
1880 | if (too_many) { | |
1881 | svc_sock_put(dr->svsk); | |
1882 | kfree(dr); | |
1883 | return; | |
1884 | } | |
1885 | dprintk("revisit queued\n"); | |
1886 | svsk = dr->svsk; | |
1887 | dr->svsk = NULL; | |
7ac1bea5 | 1888 | spin_lock(&svsk->sk_lock); |
1da177e4 | 1889 | list_add(&dr->handle.recent, &svsk->sk_deferred); |
7ac1bea5 | 1890 | spin_unlock(&svsk->sk_lock); |
1da177e4 LT |
1891 | set_bit(SK_DEFERRED, &svsk->sk_flags); |
1892 | svc_sock_enqueue(svsk); | |
1893 | svc_sock_put(svsk); | |
1894 | } | |
1895 | ||
1896 | static struct cache_deferred_req * | |
1897 | svc_defer(struct cache_req *req) | |
1898 | { | |
1899 | struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle); | |
1900 | int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len); | |
1901 | struct svc_deferred_req *dr; | |
1902 | ||
1903 | if (rqstp->rq_arg.page_len) | |
1904 | return NULL; /* if more than a page, give up FIXME */ | |
1905 | if (rqstp->rq_deferred) { | |
1906 | dr = rqstp->rq_deferred; | |
1907 | rqstp->rq_deferred = NULL; | |
1908 | } else { | |
1909 | int skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
1910 | /* FIXME maybe discard if size too large */ | |
1911 | dr = kmalloc(size, GFP_KERNEL); | |
1912 | if (dr == NULL) | |
1913 | return NULL; | |
1914 | ||
1915 | dr->handle.owner = rqstp->rq_server; | |
1916 | dr->prot = rqstp->rq_prot; | |
24422222 CL |
1917 | memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen); |
1918 | dr->addrlen = rqstp->rq_addrlen; | |
1918e341 | 1919 | dr->daddr = rqstp->rq_daddr; |
1da177e4 LT |
1920 | dr->argslen = rqstp->rq_arg.len >> 2; |
1921 | memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2); | |
1922 | } | |
c45c357d | 1923 | atomic_inc(&rqstp->rq_sock->sk_inuse); |
1da177e4 | 1924 | dr->svsk = rqstp->rq_sock; |
1da177e4 LT |
1925 | |
1926 | dr->handle.revisit = svc_revisit; | |
1927 | return &dr->handle; | |
1928 | } | |
1929 | ||
1930 | /* | |
1931 | * recv data from a deferred request into an active one | |
1932 | */ | |
1933 | static int svc_deferred_recv(struct svc_rqst *rqstp) | |
1934 | { | |
1935 | struct svc_deferred_req *dr = rqstp->rq_deferred; | |
1936 | ||
1937 | rqstp->rq_arg.head[0].iov_base = dr->args; | |
1938 | rqstp->rq_arg.head[0].iov_len = dr->argslen<<2; | |
1939 | rqstp->rq_arg.page_len = 0; | |
1940 | rqstp->rq_arg.len = dr->argslen<<2; | |
1941 | rqstp->rq_prot = dr->prot; | |
24422222 CL |
1942 | memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen); |
1943 | rqstp->rq_addrlen = dr->addrlen; | |
1918e341 | 1944 | rqstp->rq_daddr = dr->daddr; |
44524359 | 1945 | rqstp->rq_respages = rqstp->rq_pages; |
1da177e4 LT |
1946 | return dr->argslen<<2; |
1947 | } | |
1948 | ||
1949 | ||
1950 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk) | |
1951 | { | |
1952 | struct svc_deferred_req *dr = NULL; | |
cca5172a | 1953 | |
1da177e4 LT |
1954 | if (!test_bit(SK_DEFERRED, &svsk->sk_flags)) |
1955 | return NULL; | |
7ac1bea5 | 1956 | spin_lock(&svsk->sk_lock); |
1da177e4 LT |
1957 | clear_bit(SK_DEFERRED, &svsk->sk_flags); |
1958 | if (!list_empty(&svsk->sk_deferred)) { | |
1959 | dr = list_entry(svsk->sk_deferred.next, | |
1960 | struct svc_deferred_req, | |
1961 | handle.recent); | |
1962 | list_del_init(&dr->handle.recent); | |
1963 | set_bit(SK_DEFERRED, &svsk->sk_flags); | |
1964 | } | |
7ac1bea5 | 1965 | spin_unlock(&svsk->sk_lock); |
1da177e4 LT |
1966 | return dr; |
1967 | } |