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 | ||
360d8738 | 78 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT |
1da177e4 LT |
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); |
755cceab TT |
88 | static void svc_sock_detach(struct svc_xprt *); |
89 | static void svc_sock_free(struct svc_xprt *); | |
1da177e4 LT |
90 | |
91 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk); | |
92 | static int svc_deferred_recv(struct svc_rqst *rqstp); | |
93 | static struct cache_deferred_req *svc_defer(struct cache_req *req); | |
94 | ||
36bdfc8b GB |
95 | /* apparently the "standard" is that clients close |
96 | * idle connections after 5 minutes, servers after | |
97 | * 6 minutes | |
98 | * http://www.connectathon.org/talks96/nfstcp.pdf | |
99 | */ | |
100 | static int svc_conn_age_period = 6*60; | |
101 | ||
ed07536e PZ |
102 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
103 | static struct lock_class_key svc_key[2]; | |
104 | static struct lock_class_key svc_slock_key[2]; | |
105 | ||
106 | static inline void svc_reclassify_socket(struct socket *sock) | |
107 | { | |
108 | struct sock *sk = sock->sk; | |
02b3d346 | 109 | BUG_ON(sock_owned_by_user(sk)); |
ed07536e PZ |
110 | switch (sk->sk_family) { |
111 | case AF_INET: | |
112 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
113 | &svc_slock_key[0], "sk_lock-AF_INET-NFSD", &svc_key[0]); | |
114 | break; | |
115 | ||
116 | case AF_INET6: | |
117 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
118 | &svc_slock_key[1], "sk_lock-AF_INET6-NFSD", &svc_key[1]); | |
119 | break; | |
120 | ||
121 | default: | |
122 | BUG(); | |
123 | } | |
124 | } | |
125 | #else | |
126 | static inline void svc_reclassify_socket(struct socket *sock) | |
127 | { | |
128 | } | |
129 | #endif | |
130 | ||
ad06e4bd CL |
131 | static char *__svc_print_addr(struct sockaddr *addr, char *buf, size_t len) |
132 | { | |
133 | switch (addr->sa_family) { | |
134 | case AF_INET: | |
135 | snprintf(buf, len, "%u.%u.%u.%u, port=%u", | |
136 | NIPQUAD(((struct sockaddr_in *) addr)->sin_addr), | |
582ee43d | 137 | ntohs(((struct sockaddr_in *) addr)->sin_port)); |
ad06e4bd | 138 | break; |
5a05ed73 | 139 | |
ad06e4bd CL |
140 | case AF_INET6: |
141 | snprintf(buf, len, "%x:%x:%x:%x:%x:%x:%x:%x, port=%u", | |
142 | NIP6(((struct sockaddr_in6 *) addr)->sin6_addr), | |
582ee43d | 143 | ntohs(((struct sockaddr_in6 *) addr)->sin6_port)); |
ad06e4bd | 144 | break; |
5a05ed73 | 145 | |
ad06e4bd CL |
146 | default: |
147 | snprintf(buf, len, "unknown address type: %d", addr->sa_family); | |
148 | break; | |
149 | } | |
150 | return buf; | |
151 | } | |
152 | ||
153 | /** | |
154 | * svc_print_addr - Format rq_addr field for printing | |
155 | * @rqstp: svc_rqst struct containing address to print | |
156 | * @buf: target buffer for formatted address | |
157 | * @len: length of target buffer | |
158 | * | |
159 | */ | |
160 | char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len) | |
161 | { | |
27459f09 | 162 | return __svc_print_addr(svc_addr(rqstp), buf, len); |
ad06e4bd CL |
163 | } |
164 | EXPORT_SYMBOL_GPL(svc_print_addr); | |
165 | ||
1da177e4 | 166 | /* |
3262c816 | 167 | * Queue up an idle server thread. Must have pool->sp_lock held. |
1da177e4 | 168 | * Note: this is really a stack rather than a queue, so that we only |
3262c816 | 169 | * use as many different threads as we need, and the rest don't pollute |
1da177e4 LT |
170 | * the cache. |
171 | */ | |
172 | static inline void | |
3262c816 | 173 | svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp) |
1da177e4 | 174 | { |
3262c816 | 175 | list_add(&rqstp->rq_list, &pool->sp_threads); |
1da177e4 LT |
176 | } |
177 | ||
178 | /* | |
3262c816 | 179 | * Dequeue an nfsd thread. Must have pool->sp_lock held. |
1da177e4 LT |
180 | */ |
181 | static inline void | |
3262c816 | 182 | svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp) |
1da177e4 LT |
183 | { |
184 | list_del(&rqstp->rq_list); | |
185 | } | |
186 | ||
187 | /* | |
188 | * Release an skbuff after use | |
189 | */ | |
5148bf4e | 190 | static void svc_release_skb(struct svc_rqst *rqstp) |
1da177e4 | 191 | { |
5148bf4e | 192 | struct sk_buff *skb = rqstp->rq_xprt_ctxt; |
1da177e4 LT |
193 | struct svc_deferred_req *dr = rqstp->rq_deferred; |
194 | ||
195 | if (skb) { | |
5148bf4e | 196 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
197 | |
198 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
199 | skb_free_datagram(rqstp->rq_sock->sk_sk, skb); | |
200 | } | |
201 | if (dr) { | |
202 | rqstp->rq_deferred = NULL; | |
203 | kfree(dr); | |
204 | } | |
205 | } | |
206 | ||
1da177e4 LT |
207 | /* |
208 | * Queue up a socket with data pending. If there are idle nfsd | |
209 | * processes, wake 'em up. | |
210 | * | |
211 | */ | |
212 | static void | |
213 | svc_sock_enqueue(struct svc_sock *svsk) | |
214 | { | |
215 | struct svc_serv *serv = svsk->sk_server; | |
bfd24160 | 216 | struct svc_pool *pool; |
1da177e4 | 217 | struct svc_rqst *rqstp; |
bfd24160 | 218 | int cpu; |
1da177e4 LT |
219 | |
220 | if (!(svsk->sk_flags & | |
221 | ( (1<<SK_CONN)|(1<<SK_DATA)|(1<<SK_CLOSE)|(1<<SK_DEFERRED)) )) | |
222 | return; | |
223 | if (test_bit(SK_DEAD, &svsk->sk_flags)) | |
224 | return; | |
225 | ||
bfd24160 GB |
226 | cpu = get_cpu(); |
227 | pool = svc_pool_for_cpu(svsk->sk_server, cpu); | |
228 | put_cpu(); | |
229 | ||
3262c816 | 230 | spin_lock_bh(&pool->sp_lock); |
1da177e4 | 231 | |
3262c816 GB |
232 | if (!list_empty(&pool->sp_threads) && |
233 | !list_empty(&pool->sp_sockets)) | |
1da177e4 LT |
234 | printk(KERN_ERR |
235 | "svc_sock_enqueue: threads and sockets both waiting??\n"); | |
236 | ||
237 | if (test_bit(SK_DEAD, &svsk->sk_flags)) { | |
238 | /* Don't enqueue dead sockets */ | |
239 | dprintk("svc: socket %p is dead, not enqueued\n", svsk->sk_sk); | |
240 | goto out_unlock; | |
241 | } | |
242 | ||
c081a0c7 GB |
243 | /* Mark socket as busy. It will remain in this state until the |
244 | * server has processed all pending data and put the socket back | |
245 | * on the idle list. We update SK_BUSY atomically because | |
246 | * it also guards against trying to enqueue the svc_sock twice. | |
247 | */ | |
248 | if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) { | |
249 | /* Don't enqueue socket while already enqueued */ | |
1da177e4 LT |
250 | dprintk("svc: socket %p busy, not enqueued\n", svsk->sk_sk); |
251 | goto out_unlock; | |
252 | } | |
3262c816 GB |
253 | BUG_ON(svsk->sk_pool != NULL); |
254 | svsk->sk_pool = pool; | |
1da177e4 | 255 | |
323bee32 TT |
256 | /* Handle pending connection */ |
257 | if (test_bit(SK_CONN, &svsk->sk_flags)) | |
258 | goto process; | |
259 | ||
260 | /* Handle close in-progress */ | |
261 | if (test_bit(SK_CLOSE, &svsk->sk_flags)) | |
262 | goto process; | |
263 | ||
264 | /* Check if we have space to reply to a request */ | |
265 | if (!svsk->sk_xprt.xpt_ops->xpo_has_wspace(&svsk->sk_xprt)) { | |
1da177e4 | 266 | /* Don't enqueue while not enough space for reply */ |
323bee32 | 267 | dprintk("svc: no write space, socket %p not enqueued\n", svsk); |
3262c816 | 268 | svsk->sk_pool = NULL; |
c081a0c7 | 269 | clear_bit(SK_BUSY, &svsk->sk_flags); |
1da177e4 LT |
270 | goto out_unlock; |
271 | } | |
1da177e4 | 272 | |
323bee32 | 273 | process: |
3262c816 GB |
274 | if (!list_empty(&pool->sp_threads)) { |
275 | rqstp = list_entry(pool->sp_threads.next, | |
1da177e4 LT |
276 | struct svc_rqst, |
277 | rq_list); | |
278 | dprintk("svc: socket %p served by daemon %p\n", | |
279 | svsk->sk_sk, rqstp); | |
3262c816 | 280 | svc_thread_dequeue(pool, rqstp); |
1da177e4 | 281 | if (rqstp->rq_sock) |
cca5172a | 282 | printk(KERN_ERR |
1da177e4 LT |
283 | "svc_sock_enqueue: server %p, rq_sock=%p!\n", |
284 | rqstp, rqstp->rq_sock); | |
285 | rqstp->rq_sock = svsk; | |
c45c357d | 286 | atomic_inc(&svsk->sk_inuse); |
c6b0a9f8 | 287 | rqstp->rq_reserved = serv->sv_max_mesg; |
5685f0fa | 288 | atomic_add(rqstp->rq_reserved, &svsk->sk_reserved); |
3262c816 | 289 | BUG_ON(svsk->sk_pool != pool); |
1da177e4 LT |
290 | wake_up(&rqstp->rq_wait); |
291 | } else { | |
292 | dprintk("svc: socket %p put into queue\n", svsk->sk_sk); | |
3262c816 GB |
293 | list_add_tail(&svsk->sk_ready, &pool->sp_sockets); |
294 | BUG_ON(svsk->sk_pool != pool); | |
1da177e4 LT |
295 | } |
296 | ||
297 | out_unlock: | |
3262c816 | 298 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 LT |
299 | } |
300 | ||
301 | /* | |
3262c816 | 302 | * Dequeue the first socket. Must be called with the pool->sp_lock held. |
1da177e4 LT |
303 | */ |
304 | static inline struct svc_sock * | |
3262c816 | 305 | svc_sock_dequeue(struct svc_pool *pool) |
1da177e4 LT |
306 | { |
307 | struct svc_sock *svsk; | |
308 | ||
3262c816 | 309 | if (list_empty(&pool->sp_sockets)) |
1da177e4 LT |
310 | return NULL; |
311 | ||
3262c816 | 312 | svsk = list_entry(pool->sp_sockets.next, |
1da177e4 LT |
313 | struct svc_sock, sk_ready); |
314 | list_del_init(&svsk->sk_ready); | |
315 | ||
316 | dprintk("svc: socket %p dequeued, inuse=%d\n", | |
c45c357d | 317 | svsk->sk_sk, atomic_read(&svsk->sk_inuse)); |
1da177e4 LT |
318 | |
319 | return svsk; | |
320 | } | |
321 | ||
322 | /* | |
323 | * Having read something from a socket, check whether it | |
324 | * needs to be re-enqueued. | |
325 | * Note: SK_DATA only gets cleared when a read-attempt finds | |
326 | * no (or insufficient) data. | |
327 | */ | |
328 | static inline void | |
329 | svc_sock_received(struct svc_sock *svsk) | |
330 | { | |
3262c816 | 331 | svsk->sk_pool = NULL; |
1da177e4 LT |
332 | clear_bit(SK_BUSY, &svsk->sk_flags); |
333 | svc_sock_enqueue(svsk); | |
334 | } | |
335 | ||
336 | ||
337 | /** | |
338 | * svc_reserve - change the space reserved for the reply to a request. | |
339 | * @rqstp: The request in question | |
340 | * @space: new max space to reserve | |
341 | * | |
342 | * Each request reserves some space on the output queue of the socket | |
343 | * to make sure the reply fits. This function reduces that reserved | |
344 | * space to be the amount of space used already, plus @space. | |
345 | * | |
346 | */ | |
347 | void svc_reserve(struct svc_rqst *rqstp, int space) | |
348 | { | |
349 | space += rqstp->rq_res.head[0].iov_len; | |
350 | ||
351 | if (space < rqstp->rq_reserved) { | |
352 | struct svc_sock *svsk = rqstp->rq_sock; | |
5685f0fa | 353 | atomic_sub((rqstp->rq_reserved - space), &svsk->sk_reserved); |
1da177e4 | 354 | rqstp->rq_reserved = space; |
1da177e4 LT |
355 | |
356 | svc_sock_enqueue(svsk); | |
357 | } | |
358 | } | |
359 | ||
360 | /* | |
361 | * Release a socket after use. | |
362 | */ | |
363 | static inline void | |
364 | svc_sock_put(struct svc_sock *svsk) | |
365 | { | |
aaf68cfb | 366 | if (atomic_dec_and_test(&svsk->sk_inuse)) { |
755cceab TT |
367 | BUG_ON(!test_bit(SK_DEAD, &svsk->sk_flags)); |
368 | svsk->sk_xprt.xpt_ops->xpo_free(&svsk->sk_xprt); | |
1da177e4 | 369 | } |
1da177e4 LT |
370 | } |
371 | ||
372 | static void | |
373 | svc_sock_release(struct svc_rqst *rqstp) | |
374 | { | |
375 | struct svc_sock *svsk = rqstp->rq_sock; | |
376 | ||
5148bf4e | 377 | rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp); |
1da177e4 | 378 | |
44524359 | 379 | svc_free_res_pages(rqstp); |
1da177e4 LT |
380 | rqstp->rq_res.page_len = 0; |
381 | rqstp->rq_res.page_base = 0; | |
382 | ||
383 | ||
384 | /* Reset response buffer and release | |
385 | * the reservation. | |
386 | * But first, check that enough space was reserved | |
387 | * for the reply, otherwise we have a bug! | |
388 | */ | |
389 | if ((rqstp->rq_res.len) > rqstp->rq_reserved) | |
390 | printk(KERN_ERR "RPC request reserved %d but used %d\n", | |
391 | rqstp->rq_reserved, | |
392 | rqstp->rq_res.len); | |
393 | ||
394 | rqstp->rq_res.head[0].iov_len = 0; | |
395 | svc_reserve(rqstp, 0); | |
396 | rqstp->rq_sock = NULL; | |
397 | ||
398 | svc_sock_put(svsk); | |
399 | } | |
400 | ||
401 | /* | |
402 | * External function to wake up a server waiting for data | |
3262c816 GB |
403 | * This really only makes sense for services like lockd |
404 | * which have exactly one thread anyway. | |
1da177e4 LT |
405 | */ |
406 | void | |
407 | svc_wake_up(struct svc_serv *serv) | |
408 | { | |
409 | struct svc_rqst *rqstp; | |
3262c816 GB |
410 | unsigned int i; |
411 | struct svc_pool *pool; | |
412 | ||
413 | for (i = 0; i < serv->sv_nrpools; i++) { | |
414 | pool = &serv->sv_pools[i]; | |
415 | ||
416 | spin_lock_bh(&pool->sp_lock); | |
417 | if (!list_empty(&pool->sp_threads)) { | |
418 | rqstp = list_entry(pool->sp_threads.next, | |
419 | struct svc_rqst, | |
420 | rq_list); | |
421 | dprintk("svc: daemon %p woken up.\n", rqstp); | |
422 | /* | |
423 | svc_thread_dequeue(pool, rqstp); | |
424 | rqstp->rq_sock = NULL; | |
425 | */ | |
426 | wake_up(&rqstp->rq_wait); | |
427 | } | |
428 | spin_unlock_bh(&pool->sp_lock); | |
1da177e4 | 429 | } |
1da177e4 LT |
430 | } |
431 | ||
b92503b2 CL |
432 | union svc_pktinfo_u { |
433 | struct in_pktinfo pkti; | |
b92503b2 | 434 | struct in6_pktinfo pkti6; |
b92503b2 | 435 | }; |
bc375ea7 DM |
436 | #define SVC_PKTINFO_SPACE \ |
437 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
438 | |
439 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
440 | { | |
441 | switch (rqstp->rq_sock->sk_sk->sk_family) { | |
442 | case AF_INET: { | |
443 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
444 | ||
445 | cmh->cmsg_level = SOL_IP; | |
446 | cmh->cmsg_type = IP_PKTINFO; | |
447 | pki->ipi_ifindex = 0; | |
448 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; | |
449 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
450 | } | |
451 | break; | |
5a05ed73 | 452 | |
b92503b2 CL |
453 | case AF_INET6: { |
454 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
455 | ||
456 | cmh->cmsg_level = SOL_IPV6; | |
457 | cmh->cmsg_type = IPV6_PKTINFO; | |
458 | pki->ipi6_ifindex = 0; | |
459 | ipv6_addr_copy(&pki->ipi6_addr, | |
460 | &rqstp->rq_daddr.addr6); | |
461 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
462 | } | |
463 | break; | |
b92503b2 CL |
464 | } |
465 | return; | |
466 | } | |
467 | ||
1da177e4 LT |
468 | /* |
469 | * Generic sendto routine | |
470 | */ | |
471 | static int | |
472 | svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) | |
473 | { | |
474 | struct svc_sock *svsk = rqstp->rq_sock; | |
475 | struct socket *sock = svsk->sk_sock; | |
476 | int slen; | |
bc375ea7 DM |
477 | union { |
478 | struct cmsghdr hdr; | |
479 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
480 | } buffer; | |
481 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 LT |
482 | int len = 0; |
483 | int result; | |
484 | int size; | |
485 | struct page **ppage = xdr->pages; | |
486 | size_t base = xdr->page_base; | |
487 | unsigned int pglen = xdr->page_len; | |
488 | unsigned int flags = MSG_MORE; | |
ad06e4bd | 489 | char buf[RPC_MAX_ADDRBUFLEN]; |
1da177e4 LT |
490 | |
491 | slen = xdr->len; | |
492 | ||
493 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
b92503b2 CL |
494 | struct msghdr msg = { |
495 | .msg_name = &rqstp->rq_addr, | |
496 | .msg_namelen = rqstp->rq_addrlen, | |
497 | .msg_control = cmh, | |
498 | .msg_controllen = sizeof(buffer), | |
499 | .msg_flags = MSG_MORE, | |
500 | }; | |
501 | ||
502 | svc_set_cmsg_data(rqstp, cmh); | |
1da177e4 LT |
503 | |
504 | if (sock_sendmsg(sock, &msg, 0) < 0) | |
505 | goto out; | |
506 | } | |
507 | ||
508 | /* send head */ | |
509 | if (slen == xdr->head[0].iov_len) | |
510 | flags = 0; | |
44524359 N |
511 | len = kernel_sendpage(sock, rqstp->rq_respages[0], 0, |
512 | xdr->head[0].iov_len, flags); | |
1da177e4 LT |
513 | if (len != xdr->head[0].iov_len) |
514 | goto out; | |
515 | slen -= xdr->head[0].iov_len; | |
516 | if (slen == 0) | |
517 | goto out; | |
518 | ||
519 | /* send page data */ | |
520 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
521 | while (pglen > 0) { | |
522 | if (slen == size) | |
523 | flags = 0; | |
e6242e92 | 524 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
525 | if (result > 0) |
526 | len += result; | |
527 | if (result != size) | |
528 | goto out; | |
529 | slen -= size; | |
530 | pglen -= size; | |
531 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
532 | base = 0; | |
533 | ppage++; | |
534 | } | |
535 | /* send tail */ | |
536 | if (xdr->tail[0].iov_len) { | |
44524359 N |
537 | result = kernel_sendpage(sock, rqstp->rq_respages[0], |
538 | ((unsigned long)xdr->tail[0].iov_base) | |
cca5172a | 539 | & (PAGE_SIZE-1), |
1da177e4 LT |
540 | xdr->tail[0].iov_len, 0); |
541 | ||
542 | if (result > 0) | |
543 | len += result; | |
544 | } | |
545 | out: | |
ad06e4bd CL |
546 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", |
547 | rqstp->rq_sock, xdr->head[0].iov_base, xdr->head[0].iov_len, | |
548 | xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); | |
1da177e4 LT |
549 | |
550 | return len; | |
551 | } | |
552 | ||
80212d59 N |
553 | /* |
554 | * Report socket names for nfsdfs | |
555 | */ | |
556 | static int one_sock_name(char *buf, struct svc_sock *svsk) | |
557 | { | |
558 | int len; | |
559 | ||
560 | switch(svsk->sk_sk->sk_family) { | |
561 | case AF_INET: | |
562 | len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n", | |
563 | svsk->sk_sk->sk_protocol==IPPROTO_UDP? | |
564 | "udp" : "tcp", | |
565 | NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr), | |
566 | inet_sk(svsk->sk_sk)->num); | |
567 | break; | |
568 | default: | |
569 | len = sprintf(buf, "*unknown-%d*\n", | |
570 | svsk->sk_sk->sk_family); | |
571 | } | |
572 | return len; | |
573 | } | |
574 | ||
575 | int | |
b41b66d6 | 576 | svc_sock_names(char *buf, struct svc_serv *serv, char *toclose) |
80212d59 | 577 | { |
b41b66d6 | 578 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
579 | int len = 0; |
580 | ||
581 | if (!serv) | |
582 | return 0; | |
aaf68cfb | 583 | spin_lock_bh(&serv->sv_lock); |
80212d59 N |
584 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_list) { |
585 | int onelen = one_sock_name(buf+len, svsk); | |
b41b66d6 N |
586 | if (toclose && strcmp(toclose, buf+len) == 0) |
587 | closesk = svsk; | |
588 | else | |
589 | len += onelen; | |
80212d59 | 590 | } |
aaf68cfb | 591 | spin_unlock_bh(&serv->sv_lock); |
b41b66d6 | 592 | if (closesk) |
5680c446 N |
593 | /* Should unregister with portmap, but you cannot |
594 | * unregister just one protocol... | |
595 | */ | |
aaf68cfb | 596 | svc_close_socket(closesk); |
37a03472 N |
597 | else if (toclose) |
598 | return -ENOENT; | |
80212d59 N |
599 | return len; |
600 | } | |
601 | EXPORT_SYMBOL(svc_sock_names); | |
602 | ||
1da177e4 LT |
603 | /* |
604 | * Check input queue length | |
605 | */ | |
606 | static int | |
607 | svc_recv_available(struct svc_sock *svsk) | |
608 | { | |
1da177e4 LT |
609 | struct socket *sock = svsk->sk_sock; |
610 | int avail, err; | |
611 | ||
e6242e92 | 612 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
613 | |
614 | return (err >= 0)? avail : err; | |
615 | } | |
616 | ||
617 | /* | |
618 | * Generic recvfrom routine. | |
619 | */ | |
620 | static int | |
621 | svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, int buflen) | |
622 | { | |
067d7817 | 623 | struct svc_sock *svsk = rqstp->rq_sock; |
1ba95105 CL |
624 | struct msghdr msg = { |
625 | .msg_flags = MSG_DONTWAIT, | |
626 | }; | |
a9747692 | 627 | struct sockaddr *sin; |
1ba95105 | 628 | int len; |
1da177e4 | 629 | |
1ba95105 CL |
630 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
631 | msg.msg_flags); | |
1da177e4 LT |
632 | |
633 | /* sock_recvmsg doesn't fill in the name/namelen, so we must.. | |
1da177e4 | 634 | */ |
067d7817 CL |
635 | memcpy(&rqstp->rq_addr, &svsk->sk_remote, svsk->sk_remotelen); |
636 | rqstp->rq_addrlen = svsk->sk_remotelen; | |
1da177e4 | 637 | |
a9747692 FM |
638 | /* Destination address in request is needed for binding the |
639 | * source address in RPC callbacks later. | |
640 | */ | |
641 | sin = (struct sockaddr *)&svsk->sk_local; | |
642 | switch (sin->sa_family) { | |
643 | case AF_INET: | |
644 | rqstp->rq_daddr.addr = ((struct sockaddr_in *)sin)->sin_addr; | |
645 | break; | |
646 | case AF_INET6: | |
647 | rqstp->rq_daddr.addr6 = ((struct sockaddr_in6 *)sin)->sin6_addr; | |
648 | break; | |
649 | } | |
650 | ||
1da177e4 | 651 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", |
1ba95105 | 652 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
1da177e4 LT |
653 | |
654 | return len; | |
655 | } | |
656 | ||
657 | /* | |
658 | * Set socket snd and rcv buffer lengths | |
659 | */ | |
660 | static inline void | |
661 | svc_sock_setbufsize(struct socket *sock, unsigned int snd, unsigned int rcv) | |
662 | { | |
663 | #if 0 | |
664 | mm_segment_t oldfs; | |
665 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
666 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
667 | (char*)&snd, sizeof(snd)); | |
668 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
669 | (char*)&rcv, sizeof(rcv)); | |
670 | #else | |
671 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
672 | * which isn't acceptable. Until that is made conditional | |
673 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
674 | * DaveM said I could! | |
675 | */ | |
676 | lock_sock(sock->sk); | |
677 | sock->sk->sk_sndbuf = snd * 2; | |
678 | sock->sk->sk_rcvbuf = rcv * 2; | |
679 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; | |
680 | release_sock(sock->sk); | |
681 | #endif | |
682 | } | |
683 | /* | |
684 | * INET callback when data has been received on the socket. | |
685 | */ | |
686 | static void | |
687 | svc_udp_data_ready(struct sock *sk, int count) | |
688 | { | |
939bb7ef | 689 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 690 | |
939bb7ef NB |
691 | if (svsk) { |
692 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
693 | svsk, sk, count, test_bit(SK_BUSY, &svsk->sk_flags)); | |
694 | set_bit(SK_DATA, &svsk->sk_flags); | |
695 | svc_sock_enqueue(svsk); | |
696 | } | |
1da177e4 LT |
697 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
698 | wake_up_interruptible(sk->sk_sleep); | |
699 | } | |
700 | ||
701 | /* | |
702 | * INET callback when space is newly available on the socket. | |
703 | */ | |
704 | static void | |
705 | svc_write_space(struct sock *sk) | |
706 | { | |
707 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
708 | ||
709 | if (svsk) { | |
710 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
711 | svsk, sk, test_bit(SK_BUSY, &svsk->sk_flags)); | |
712 | svc_sock_enqueue(svsk); | |
713 | } | |
714 | ||
715 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { | |
939bb7ef | 716 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 LT |
717 | svsk); |
718 | wake_up_interruptible(sk->sk_sleep); | |
719 | } | |
720 | } | |
721 | ||
7a37f578 N |
722 | static inline void svc_udp_get_dest_address(struct svc_rqst *rqstp, |
723 | struct cmsghdr *cmh) | |
95756482 CL |
724 | { |
725 | switch (rqstp->rq_sock->sk_sk->sk_family) { | |
726 | case AF_INET: { | |
7a37f578 N |
727 | struct in_pktinfo *pki = CMSG_DATA(cmh); |
728 | rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; | |
95756482 | 729 | break; |
7a37f578 | 730 | } |
95756482 | 731 | case AF_INET6: { |
7a37f578 N |
732 | struct in6_pktinfo *pki = CMSG_DATA(cmh); |
733 | ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); | |
95756482 | 734 | break; |
7a37f578 | 735 | } |
95756482 | 736 | } |
95756482 CL |
737 | } |
738 | ||
1da177e4 LT |
739 | /* |
740 | * Receive a datagram from a UDP socket. | |
741 | */ | |
1da177e4 LT |
742 | static int |
743 | svc_udp_recvfrom(struct svc_rqst *rqstp) | |
744 | { | |
745 | struct svc_sock *svsk = rqstp->rq_sock; | |
746 | struct svc_serv *serv = svsk->sk_server; | |
747 | struct sk_buff *skb; | |
bc375ea7 DM |
748 | union { |
749 | struct cmsghdr hdr; | |
750 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
751 | } buffer; | |
752 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 | 753 | int err, len; |
7a37f578 N |
754 | struct msghdr msg = { |
755 | .msg_name = svc_addr(rqstp), | |
756 | .msg_control = cmh, | |
757 | .msg_controllen = sizeof(buffer), | |
758 | .msg_flags = MSG_DONTWAIT, | |
759 | }; | |
1da177e4 LT |
760 | |
761 | if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags)) | |
762 | /* udp sockets need large rcvbuf as all pending | |
763 | * requests are still in that buffer. sndbuf must | |
764 | * also be large enough that there is enough space | |
3262c816 GB |
765 | * for one reply per thread. We count all threads |
766 | * rather than threads in a particular pool, which | |
767 | * provides an upper bound on the number of threads | |
768 | * which will access the socket. | |
1da177e4 LT |
769 | */ |
770 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
771 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
772 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 LT |
773 | |
774 | if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) { | |
775 | svc_sock_received(svsk); | |
776 | return svc_deferred_recv(rqstp); | |
777 | } | |
778 | ||
779 | clear_bit(SK_DATA, &svsk->sk_flags); | |
05ed690e N |
780 | skb = NULL; |
781 | err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, | |
782 | 0, 0, MSG_PEEK | MSG_DONTWAIT); | |
783 | if (err >= 0) | |
784 | skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); | |
785 | ||
786 | if (skb == NULL) { | |
787 | if (err != -EAGAIN) { | |
788 | /* possibly an icmp error */ | |
789 | dprintk("svc: recvfrom returned error %d\n", -err); | |
790 | set_bit(SK_DATA, &svsk->sk_flags); | |
1da177e4 | 791 | } |
05ed690e N |
792 | svc_sock_received(svsk); |
793 | return -EAGAIN; | |
1da177e4 | 794 | } |
7a37f578 | 795 | rqstp->rq_addrlen = sizeof(rqstp->rq_addr); |
b7aa0bf7 ED |
796 | if (skb->tstamp.tv64 == 0) { |
797 | skb->tstamp = ktime_get_real(); | |
cca5172a | 798 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
799 | need that much accuracy */ |
800 | } | |
b7aa0bf7 | 801 | svsk->sk_sk->sk_stamp = skb->tstamp; |
1da177e4 LT |
802 | set_bit(SK_DATA, &svsk->sk_flags); /* there may be more data... */ |
803 | ||
804 | /* | |
805 | * Maybe more packets - kick another thread ASAP. | |
806 | */ | |
807 | svc_sock_received(svsk); | |
808 | ||
809 | len = skb->len - sizeof(struct udphdr); | |
810 | rqstp->rq_arg.len = len; | |
811 | ||
95756482 | 812 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 813 | |
7a37f578 N |
814 | if (cmh->cmsg_level != IPPROTO_IP || |
815 | cmh->cmsg_type != IP_PKTINFO) { | |
816 | if (net_ratelimit()) | |
817 | printk("rpcsvc: received unknown control message:" | |
818 | "%d/%d\n", | |
819 | cmh->cmsg_level, cmh->cmsg_type); | |
820 | skb_free_datagram(svsk->sk_sk, skb); | |
821 | return 0; | |
822 | } | |
823 | svc_udp_get_dest_address(rqstp, cmh); | |
1da177e4 LT |
824 | |
825 | if (skb_is_nonlinear(skb)) { | |
826 | /* we have to copy */ | |
827 | local_bh_disable(); | |
828 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
829 | local_bh_enable(); | |
830 | /* checksum error */ | |
831 | skb_free_datagram(svsk->sk_sk, skb); | |
832 | return 0; | |
833 | } | |
834 | local_bh_enable(); | |
cca5172a | 835 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 LT |
836 | } else { |
837 | /* we can use it in-place */ | |
838 | rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr); | |
839 | rqstp->rq_arg.head[0].iov_len = len; | |
fb286bb2 HX |
840 | if (skb_checksum_complete(skb)) { |
841 | skb_free_datagram(svsk->sk_sk, skb); | |
842 | return 0; | |
1da177e4 | 843 | } |
5148bf4e | 844 | rqstp->rq_xprt_ctxt = skb; |
1da177e4 LT |
845 | } |
846 | ||
847 | rqstp->rq_arg.page_base = 0; | |
848 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
849 | rqstp->rq_arg.head[0].iov_len = len; | |
850 | rqstp->rq_arg.page_len = 0; | |
44524359 | 851 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
852 | } else { |
853 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 | 854 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
172589cc | 855 | DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); |
1da177e4 LT |
856 | } |
857 | ||
858 | if (serv->sv_stats) | |
859 | serv->sv_stats->netudpcnt++; | |
860 | ||
861 | return len; | |
862 | } | |
863 | ||
864 | static int | |
865 | svc_udp_sendto(struct svc_rqst *rqstp) | |
866 | { | |
867 | int error; | |
868 | ||
869 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
870 | if (error == -ECONNREFUSED) | |
871 | /* ICMP error on earlier request. */ | |
872 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
873 | ||
874 | return error; | |
875 | } | |
876 | ||
e831fe65 TT |
877 | static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) |
878 | { | |
879 | } | |
880 | ||
323bee32 TT |
881 | static int svc_udp_has_wspace(struct svc_xprt *xprt) |
882 | { | |
883 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
884 | struct svc_serv *serv = svsk->sk_server; | |
885 | unsigned long required; | |
886 | ||
887 | /* | |
888 | * Set the SOCK_NOSPACE flag before checking the available | |
889 | * sock space. | |
890 | */ | |
891 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
892 | required = atomic_read(&svsk->sk_reserved) + serv->sv_max_mesg; | |
893 | if (required*2 > sock_wspace(svsk->sk_sk)) | |
894 | return 0; | |
895 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
896 | return 1; | |
897 | } | |
898 | ||
38a417cc TT |
899 | static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) |
900 | { | |
901 | BUG(); | |
902 | return NULL; | |
903 | } | |
904 | ||
360d8738 | 905 | static struct svc_xprt_ops svc_udp_ops = { |
5d137990 TT |
906 | .xpo_recvfrom = svc_udp_recvfrom, |
907 | .xpo_sendto = svc_udp_sendto, | |
5148bf4e | 908 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
909 | .xpo_detach = svc_sock_detach, |
910 | .xpo_free = svc_sock_free, | |
e831fe65 | 911 | .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, |
323bee32 | 912 | .xpo_has_wspace = svc_udp_has_wspace, |
38a417cc | 913 | .xpo_accept = svc_udp_accept, |
360d8738 TT |
914 | }; |
915 | ||
916 | static struct svc_xprt_class svc_udp_class = { | |
917 | .xcl_name = "udp", | |
918 | .xcl_ops = &svc_udp_ops, | |
49023155 | 919 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, |
360d8738 TT |
920 | }; |
921 | ||
1da177e4 LT |
922 | static void |
923 | svc_udp_init(struct svc_sock *svsk) | |
924 | { | |
7a37f578 N |
925 | int one = 1; |
926 | mm_segment_t oldfs; | |
927 | ||
360d8738 | 928 | svc_xprt_init(&svc_udp_class, &svsk->sk_xprt); |
1da177e4 LT |
929 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
930 | svsk->sk_sk->sk_write_space = svc_write_space; | |
1da177e4 LT |
931 | |
932 | /* initialise setting must have enough space to | |
cca5172a | 933 | * receive and respond to one request. |
1da177e4 LT |
934 | * svc_udp_recvfrom will re-adjust if necessary |
935 | */ | |
936 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
937 | 3 * svsk->sk_server->sv_max_mesg, |
938 | 3 * svsk->sk_server->sv_max_mesg); | |
1da177e4 LT |
939 | |
940 | set_bit(SK_DATA, &svsk->sk_flags); /* might have come in before data_ready set up */ | |
941 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
7a37f578 N |
942 | |
943 | oldfs = get_fs(); | |
944 | set_fs(KERNEL_DS); | |
945 | /* make sure we get destination address info */ | |
946 | svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO, | |
947 | (char __user *)&one, sizeof(one)); | |
948 | set_fs(oldfs); | |
1da177e4 LT |
949 | } |
950 | ||
951 | /* | |
952 | * A data_ready event on a listening socket means there's a connection | |
953 | * pending. Do not use state_change as a substitute for it. | |
954 | */ | |
955 | static void | |
956 | svc_tcp_listen_data_ready(struct sock *sk, int count_unused) | |
957 | { | |
939bb7ef | 958 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
959 | |
960 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 961 | sk, sk->sk_state); |
1da177e4 | 962 | |
939bb7ef NB |
963 | /* |
964 | * This callback may called twice when a new connection | |
965 | * is established as a child socket inherits everything | |
966 | * from a parent LISTEN socket. | |
967 | * 1) data_ready method of the parent socket will be called | |
968 | * when one of child sockets become ESTABLISHED. | |
969 | * 2) data_ready method of the child socket may be called | |
970 | * when it receives data before the socket is accepted. | |
971 | * In case of 2, we should ignore it silently. | |
972 | */ | |
973 | if (sk->sk_state == TCP_LISTEN) { | |
974 | if (svsk) { | |
975 | set_bit(SK_CONN, &svsk->sk_flags); | |
976 | svc_sock_enqueue(svsk); | |
977 | } else | |
978 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 979 | } |
939bb7ef | 980 | |
1da177e4 LT |
981 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
982 | wake_up_interruptible_all(sk->sk_sleep); | |
983 | } | |
984 | ||
985 | /* | |
986 | * A state change on a connected socket means it's dying or dead. | |
987 | */ | |
988 | static void | |
989 | svc_tcp_state_change(struct sock *sk) | |
990 | { | |
939bb7ef | 991 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
992 | |
993 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 994 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 995 | |
939bb7ef | 996 | if (!svsk) |
1da177e4 | 997 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef NB |
998 | else { |
999 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1000 | svc_sock_enqueue(svsk); | |
1da177e4 | 1001 | } |
1da177e4 LT |
1002 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
1003 | wake_up_interruptible_all(sk->sk_sleep); | |
1004 | } | |
1005 | ||
1006 | static void | |
1007 | svc_tcp_data_ready(struct sock *sk, int count) | |
1008 | { | |
939bb7ef | 1009 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
1010 | |
1011 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
1012 | sk, sk->sk_user_data); |
1013 | if (svsk) { | |
1014 | set_bit(SK_DATA, &svsk->sk_flags); | |
1015 | svc_sock_enqueue(svsk); | |
1016 | } | |
1da177e4 LT |
1017 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
1018 | wake_up_interruptible(sk->sk_sleep); | |
1019 | } | |
1020 | ||
bcdb81ae CL |
1021 | static inline int svc_port_is_privileged(struct sockaddr *sin) |
1022 | { | |
1023 | switch (sin->sa_family) { | |
1024 | case AF_INET: | |
1025 | return ntohs(((struct sockaddr_in *)sin)->sin_port) | |
1026 | < PROT_SOCK; | |
bcdb81ae CL |
1027 | case AF_INET6: |
1028 | return ntohs(((struct sockaddr_in6 *)sin)->sin6_port) | |
1029 | < PROT_SOCK; | |
bcdb81ae CL |
1030 | default: |
1031 | return 0; | |
1032 | } | |
1033 | } | |
1034 | ||
1da177e4 LT |
1035 | /* |
1036 | * Accept a TCP connection | |
1037 | */ | |
38a417cc | 1038 | static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) |
1da177e4 | 1039 | { |
38a417cc | 1040 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); |
cdd88b9f AM |
1041 | struct sockaddr_storage addr; |
1042 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
1da177e4 LT |
1043 | struct svc_serv *serv = svsk->sk_server; |
1044 | struct socket *sock = svsk->sk_sock; | |
1045 | struct socket *newsock; | |
1da177e4 LT |
1046 | struct svc_sock *newsvsk; |
1047 | int err, slen; | |
ad06e4bd | 1048 | char buf[RPC_MAX_ADDRBUFLEN]; |
1da177e4 LT |
1049 | |
1050 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
1051 | if (!sock) | |
38a417cc | 1052 | return NULL; |
1da177e4 | 1053 | |
e6242e92 SS |
1054 | clear_bit(SK_CONN, &svsk->sk_flags); |
1055 | err = kernel_accept(sock, &newsock, O_NONBLOCK); | |
1056 | if (err < 0) { | |
1da177e4 LT |
1057 | if (err == -ENOMEM) |
1058 | printk(KERN_WARNING "%s: no more sockets!\n", | |
1059 | serv->sv_name); | |
e6242e92 | 1060 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
1061 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
1062 | serv->sv_name, -err); | |
38a417cc | 1063 | return NULL; |
1da177e4 | 1064 | } |
e6242e92 | 1065 | |
1da177e4 | 1066 | set_bit(SK_CONN, &svsk->sk_flags); |
1da177e4 | 1067 | |
cdd88b9f | 1068 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 LT |
1069 | if (err < 0) { |
1070 | if (net_ratelimit()) | |
1071 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
1072 | serv->sv_name, -err); | |
1073 | goto failed; /* aborted connection or whatever */ | |
1074 | } | |
1075 | ||
1076 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
1077 | * hosts here, but when we get encryption, the IP of the host won't |
1078 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 1079 | */ |
cdd88b9f | 1080 | if (!svc_port_is_privileged(sin)) { |
1da177e4 | 1081 | dprintk(KERN_WARNING |
ad06e4bd | 1082 | "%s: connect from unprivileged port: %s\n", |
cca5172a | 1083 | serv->sv_name, |
cdd88b9f | 1084 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 1085 | } |
ad06e4bd | 1086 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 1087 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1088 | |
1089 | /* make sure that a write doesn't block forever when | |
1090 | * low on memory | |
1091 | */ | |
1092 | newsock->sk->sk_sndtimeo = HZ*30; | |
1093 | ||
6b174337 CL |
1094 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
1095 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) | |
1da177e4 | 1096 | goto failed; |
cdd88b9f | 1097 | memcpy(&newsvsk->sk_remote, sin, slen); |
067d7817 | 1098 | newsvsk->sk_remotelen = slen; |
a9747692 FM |
1099 | err = kernel_getsockname(newsock, sin, &slen); |
1100 | if (unlikely(err < 0)) { | |
1101 | dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); | |
1102 | slen = offsetof(struct sockaddr, sa_data); | |
1103 | } | |
1104 | memcpy(&newsvsk->sk_local, sin, slen); | |
067d7817 | 1105 | |
e79eff1f | 1106 | svc_sock_received(newsvsk); |
1da177e4 | 1107 | |
f9f3cc4f TT |
1108 | if (serv->sv_stats) |
1109 | serv->sv_stats->nettcpconn++; | |
1110 | ||
1111 | return &newsvsk->sk_xprt; | |
1112 | ||
1113 | failed: | |
1114 | sock_release(newsock); | |
1115 | return NULL; | |
1116 | } | |
1117 | ||
1118 | /* | |
1119 | * Make sure that we don't have too many active connections. If we | |
1120 | * have, something must be dropped. | |
1121 | * | |
1122 | * There's no point in trying to do random drop here for DoS | |
1123 | * prevention. The NFS clients does 1 reconnect in 15 seconds. An | |
1124 | * attacker can easily beat that. | |
1125 | * | |
1126 | * The only somewhat efficient mechanism would be if drop old | |
1127 | * connections from the same IP first. But right now we don't even | |
1128 | * record the client IP in svc_sock. | |
1129 | */ | |
1130 | static void svc_check_conn_limits(struct svc_serv *serv) | |
1131 | { | |
1da177e4 LT |
1132 | if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) { |
1133 | struct svc_sock *svsk = NULL; | |
1134 | spin_lock_bh(&serv->sv_lock); | |
1135 | if (!list_empty(&serv->sv_tempsocks)) { | |
1136 | if (net_ratelimit()) { | |
1137 | /* Try to help the admin */ | |
1138 | printk(KERN_NOTICE "%s: too many open TCP " | |
f9f3cc4f TT |
1139 | "sockets, consider increasing the " |
1140 | "number of nfsd threads\n", | |
1141 | serv->sv_name); | |
1da177e4 LT |
1142 | } |
1143 | /* | |
1144 | * Always select the oldest socket. It's not fair, | |
1145 | * but so is life | |
1146 | */ | |
1147 | svsk = list_entry(serv->sv_tempsocks.prev, | |
1148 | struct svc_sock, | |
1149 | sk_list); | |
1150 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
c45c357d | 1151 | atomic_inc(&svsk->sk_inuse); |
1da177e4 LT |
1152 | } |
1153 | spin_unlock_bh(&serv->sv_lock); | |
1154 | ||
1155 | if (svsk) { | |
1156 | svc_sock_enqueue(svsk); | |
1157 | svc_sock_put(svsk); | |
1158 | } | |
1da177e4 | 1159 | } |
1da177e4 LT |
1160 | } |
1161 | ||
1162 | /* | |
1163 | * Receive data from a TCP socket. | |
1164 | */ | |
1165 | static int | |
1166 | svc_tcp_recvfrom(struct svc_rqst *rqstp) | |
1167 | { | |
1168 | struct svc_sock *svsk = rqstp->rq_sock; | |
1169 | struct svc_serv *serv = svsk->sk_server; | |
1170 | int len; | |
3cc03b16 | 1171 | struct kvec *vec; |
1da177e4 LT |
1172 | int pnum, vlen; |
1173 | ||
1174 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
1175 | svsk, test_bit(SK_DATA, &svsk->sk_flags), | |
1176 | test_bit(SK_CONN, &svsk->sk_flags), | |
1177 | test_bit(SK_CLOSE, &svsk->sk_flags)); | |
1178 | ||
1179 | if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) { | |
1180 | svc_sock_received(svsk); | |
1181 | return svc_deferred_recv(rqstp); | |
1182 | } | |
1183 | ||
1da177e4 LT |
1184 | if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags)) |
1185 | /* sndbuf needs to have room for one request | |
1186 | * per thread, otherwise we can stall even when the | |
1187 | * network isn't a bottleneck. | |
3262c816 GB |
1188 | * |
1189 | * We count all threads rather than threads in a | |
1190 | * particular pool, which provides an upper bound | |
1191 | * on the number of threads which will access the socket. | |
1192 | * | |
1da177e4 | 1193 | * rcvbuf just needs to be able to hold a few requests. |
cca5172a | 1194 | * Normally they will be removed from the queue |
1da177e4 LT |
1195 | * as soon a a complete request arrives. |
1196 | */ | |
1197 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
1198 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
1199 | 3 * serv->sv_max_mesg); | |
1da177e4 LT |
1200 | |
1201 | clear_bit(SK_DATA, &svsk->sk_flags); | |
1202 | ||
1203 | /* Receive data. If we haven't got the record length yet, get | |
1204 | * the next four bytes. Otherwise try to gobble up as much as | |
1205 | * possible up to the complete record length. | |
1206 | */ | |
1207 | if (svsk->sk_tcplen < 4) { | |
1208 | unsigned long want = 4 - svsk->sk_tcplen; | |
1209 | struct kvec iov; | |
1210 | ||
1211 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; | |
1212 | iov.iov_len = want; | |
1213 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
1214 | goto error; | |
1215 | svsk->sk_tcplen += len; | |
1216 | ||
1217 | if (len < want) { | |
1218 | dprintk("svc: short recvfrom while reading record length (%d of %lu)\n", | |
cca5172a | 1219 | len, want); |
1da177e4 LT |
1220 | svc_sock_received(svsk); |
1221 | return -EAGAIN; /* record header not complete */ | |
1222 | } | |
1223 | ||
1224 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
1225 | if (!(svsk->sk_reclen & 0x80000000)) { | |
1226 | /* FIXME: technically, a record can be fragmented, | |
1227 | * and non-terminal fragments will not have the top | |
1228 | * bit set in the fragment length header. | |
1229 | * But apparently no known nfs clients send fragmented | |
1230 | * records. */ | |
34e9a63b N |
1231 | if (net_ratelimit()) |
1232 | printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx" | |
1233 | " (non-terminal)\n", | |
1234 | (unsigned long) svsk->sk_reclen); | |
1da177e4 LT |
1235 | goto err_delete; |
1236 | } | |
1237 | svsk->sk_reclen &= 0x7fffffff; | |
1238 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); | |
c6b0a9f8 | 1239 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
34e9a63b N |
1240 | if (net_ratelimit()) |
1241 | printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx" | |
1242 | " (large)\n", | |
1243 | (unsigned long) svsk->sk_reclen); | |
1da177e4 LT |
1244 | goto err_delete; |
1245 | } | |
1246 | } | |
1247 | ||
1248 | /* Check whether enough data is available */ | |
1249 | len = svc_recv_available(svsk); | |
1250 | if (len < 0) | |
1251 | goto error; | |
1252 | ||
1253 | if (len < svsk->sk_reclen) { | |
1254 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
1255 | len, svsk->sk_reclen); | |
1256 | svc_sock_received(svsk); | |
1257 | return -EAGAIN; /* record not complete */ | |
1258 | } | |
1259 | len = svsk->sk_reclen; | |
1260 | set_bit(SK_DATA, &svsk->sk_flags); | |
1261 | ||
3cc03b16 | 1262 | vec = rqstp->rq_vec; |
1da177e4 LT |
1263 | vec[0] = rqstp->rq_arg.head[0]; |
1264 | vlen = PAGE_SIZE; | |
1265 | pnum = 1; | |
1266 | while (vlen < len) { | |
44524359 | 1267 | vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]); |
1da177e4 LT |
1268 | vec[pnum].iov_len = PAGE_SIZE; |
1269 | pnum++; | |
1270 | vlen += PAGE_SIZE; | |
1271 | } | |
44524359 | 1272 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
1273 | |
1274 | /* Now receive data */ | |
1275 | len = svc_recvfrom(rqstp, vec, pnum, len); | |
1276 | if (len < 0) | |
1277 | goto error; | |
1278 | ||
1279 | dprintk("svc: TCP complete record (%d bytes)\n", len); | |
1280 | rqstp->rq_arg.len = len; | |
1281 | rqstp->rq_arg.page_base = 0; | |
1282 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
1283 | rqstp->rq_arg.head[0].iov_len = len; | |
1284 | rqstp->rq_arg.page_len = 0; | |
1285 | } else { | |
1286 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
1287 | } | |
1288 | ||
5148bf4e | 1289 | rqstp->rq_xprt_ctxt = NULL; |
1da177e4 LT |
1290 | rqstp->rq_prot = IPPROTO_TCP; |
1291 | ||
1292 | /* Reset TCP read info */ | |
1293 | svsk->sk_reclen = 0; | |
1294 | svsk->sk_tcplen = 0; | |
1295 | ||
1296 | svc_sock_received(svsk); | |
1297 | if (serv->sv_stats) | |
1298 | serv->sv_stats->nettcpcnt++; | |
1299 | ||
1300 | return len; | |
1301 | ||
1302 | err_delete: | |
d7979ae4 | 1303 | set_bit(SK_CLOSE, &svsk->sk_flags); |
1da177e4 LT |
1304 | return -EAGAIN; |
1305 | ||
1306 | error: | |
1307 | if (len == -EAGAIN) { | |
1308 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
1309 | svc_sock_received(svsk); | |
1310 | } else { | |
1311 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", | |
1312 | svsk->sk_server->sv_name, -len); | |
93fbf1a5 | 1313 | goto err_delete; |
1da177e4 LT |
1314 | } |
1315 | ||
1316 | return len; | |
1317 | } | |
1318 | ||
1319 | /* | |
1320 | * Send out data on TCP socket. | |
1321 | */ | |
1322 | static int | |
1323 | svc_tcp_sendto(struct svc_rqst *rqstp) | |
1324 | { | |
1325 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
1326 | int sent; | |
d8ed029d | 1327 | __be32 reclen; |
1da177e4 LT |
1328 | |
1329 | /* Set up the first element of the reply kvec. | |
1330 | * Any other kvecs that may be in use have been taken | |
1331 | * care of by the server implementation itself. | |
1332 | */ | |
1333 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
1334 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
1335 | ||
1336 | if (test_bit(SK_DEAD, &rqstp->rq_sock->sk_flags)) | |
1337 | return -ENOTCONN; | |
1338 | ||
1339 | sent = svc_sendto(rqstp, &rqstp->rq_res); | |
1340 | if (sent != xbufp->len) { | |
1341 | printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n", | |
1342 | rqstp->rq_sock->sk_server->sv_name, | |
1343 | (sent<0)?"got error":"sent only", | |
1344 | sent, xbufp->len); | |
aaf68cfb N |
1345 | set_bit(SK_CLOSE, &rqstp->rq_sock->sk_flags); |
1346 | svc_sock_enqueue(rqstp->rq_sock); | |
1da177e4 LT |
1347 | sent = -EAGAIN; |
1348 | } | |
1349 | return sent; | |
1350 | } | |
1351 | ||
e831fe65 TT |
1352 | /* |
1353 | * Setup response header. TCP has a 4B record length field. | |
1354 | */ | |
1355 | static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) | |
1356 | { | |
1357 | struct kvec *resv = &rqstp->rq_res.head[0]; | |
1358 | ||
1359 | /* tcp needs a space for the record length... */ | |
1360 | svc_putnl(resv, 0); | |
1361 | } | |
1362 | ||
323bee32 TT |
1363 | static int svc_tcp_has_wspace(struct svc_xprt *xprt) |
1364 | { | |
1365 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1366 | struct svc_serv *serv = svsk->sk_server; | |
1367 | int required; | |
1368 | int wspace; | |
1369 | ||
1370 | /* | |
1371 | * Set the SOCK_NOSPACE flag before checking the available | |
1372 | * sock space. | |
1373 | */ | |
1374 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
1375 | required = atomic_read(&svsk->sk_reserved) + serv->sv_max_mesg; | |
1376 | wspace = sk_stream_wspace(svsk->sk_sk); | |
1377 | ||
1378 | if (wspace < sk_stream_min_wspace(svsk->sk_sk)) | |
1379 | return 0; | |
1380 | if (required * 2 > wspace) | |
1381 | return 0; | |
1382 | ||
1383 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
1384 | return 1; | |
1385 | } | |
1386 | ||
360d8738 | 1387 | static struct svc_xprt_ops svc_tcp_ops = { |
5d137990 TT |
1388 | .xpo_recvfrom = svc_tcp_recvfrom, |
1389 | .xpo_sendto = svc_tcp_sendto, | |
5148bf4e | 1390 | .xpo_release_rqst = svc_release_skb, |
755cceab TT |
1391 | .xpo_detach = svc_sock_detach, |
1392 | .xpo_free = svc_sock_free, | |
e831fe65 | 1393 | .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, |
323bee32 | 1394 | .xpo_has_wspace = svc_tcp_has_wspace, |
38a417cc | 1395 | .xpo_accept = svc_tcp_accept, |
360d8738 TT |
1396 | }; |
1397 | ||
1398 | static struct svc_xprt_class svc_tcp_class = { | |
1399 | .xcl_name = "tcp", | |
1400 | .xcl_ops = &svc_tcp_ops, | |
49023155 | 1401 | .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, |
360d8738 TT |
1402 | }; |
1403 | ||
1404 | void svc_init_xprt_sock(void) | |
1405 | { | |
1406 | svc_reg_xprt_class(&svc_tcp_class); | |
1407 | svc_reg_xprt_class(&svc_udp_class); | |
1408 | } | |
1409 | ||
1410 | void svc_cleanup_xprt_sock(void) | |
1411 | { | |
1412 | svc_unreg_xprt_class(&svc_tcp_class); | |
1413 | svc_unreg_xprt_class(&svc_udp_class); | |
1414 | } | |
1415 | ||
1da177e4 LT |
1416 | static void |
1417 | svc_tcp_init(struct svc_sock *svsk) | |
1418 | { | |
1419 | struct sock *sk = svsk->sk_sk; | |
1420 | struct tcp_sock *tp = tcp_sk(sk); | |
1421 | ||
360d8738 | 1422 | svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt); |
1da177e4 LT |
1423 | |
1424 | if (sk->sk_state == TCP_LISTEN) { | |
1425 | dprintk("setting up TCP socket for listening\n"); | |
38a417cc | 1426 | set_bit(SK_LISTENER, &svsk->sk_flags); |
1da177e4 LT |
1427 | sk->sk_data_ready = svc_tcp_listen_data_ready; |
1428 | set_bit(SK_CONN, &svsk->sk_flags); | |
1429 | } else { | |
1430 | dprintk("setting up TCP socket for reading\n"); | |
1431 | sk->sk_state_change = svc_tcp_state_change; | |
1432 | sk->sk_data_ready = svc_tcp_data_ready; | |
1433 | sk->sk_write_space = svc_write_space; | |
1434 | ||
1435 | svsk->sk_reclen = 0; | |
1436 | svsk->sk_tcplen = 0; | |
1437 | ||
1438 | tp->nonagle = 1; /* disable Nagle's algorithm */ | |
1439 | ||
1440 | /* initialise setting must have enough space to | |
cca5172a | 1441 | * receive and respond to one request. |
1da177e4 LT |
1442 | * svc_tcp_recvfrom will re-adjust if necessary |
1443 | */ | |
1444 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
1445 | 3 * svsk->sk_server->sv_max_mesg, |
1446 | 3 * svsk->sk_server->sv_max_mesg); | |
1da177e4 LT |
1447 | |
1448 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1449 | set_bit(SK_DATA, &svsk->sk_flags); | |
cca5172a | 1450 | if (sk->sk_state != TCP_ESTABLISHED) |
1da177e4 LT |
1451 | set_bit(SK_CLOSE, &svsk->sk_flags); |
1452 | } | |
1453 | } | |
1454 | ||
1455 | void | |
1456 | svc_sock_update_bufs(struct svc_serv *serv) | |
1457 | { | |
1458 | /* | |
1459 | * The number of server threads has changed. Update | |
1460 | * rcvbuf and sndbuf accordingly on all sockets | |
1461 | */ | |
1462 | struct list_head *le; | |
1463 | ||
1464 | spin_lock_bh(&serv->sv_lock); | |
1465 | list_for_each(le, &serv->sv_permsocks) { | |
cca5172a | 1466 | struct svc_sock *svsk = |
1da177e4 LT |
1467 | list_entry(le, struct svc_sock, sk_list); |
1468 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1469 | } | |
1470 | list_for_each(le, &serv->sv_tempsocks) { | |
1471 | struct svc_sock *svsk = | |
1472 | list_entry(le, struct svc_sock, sk_list); | |
1473 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1474 | } | |
1475 | spin_unlock_bh(&serv->sv_lock); | |
1476 | } | |
1477 | ||
1478 | /* | |
3262c816 GB |
1479 | * Receive the next request on any socket. This code is carefully |
1480 | * organised not to touch any cachelines in the shared svc_serv | |
1481 | * structure, only cachelines in the local svc_pool. | |
1da177e4 LT |
1482 | */ |
1483 | int | |
6fb2b47f | 1484 | svc_recv(struct svc_rqst *rqstp, long timeout) |
1da177e4 | 1485 | { |
27459f09 | 1486 | struct svc_sock *svsk = NULL; |
6fb2b47f | 1487 | struct svc_serv *serv = rqstp->rq_server; |
3262c816 | 1488 | struct svc_pool *pool = rqstp->rq_pool; |
44524359 | 1489 | int len, i; |
1da177e4 LT |
1490 | int pages; |
1491 | struct xdr_buf *arg; | |
1492 | DECLARE_WAITQUEUE(wait, current); | |
1493 | ||
1494 | dprintk("svc: server %p waiting for data (to = %ld)\n", | |
1495 | rqstp, timeout); | |
1496 | ||
1497 | if (rqstp->rq_sock) | |
cca5172a | 1498 | printk(KERN_ERR |
1da177e4 LT |
1499 | "svc_recv: service %p, socket not NULL!\n", |
1500 | rqstp); | |
1501 | if (waitqueue_active(&rqstp->rq_wait)) | |
cca5172a | 1502 | printk(KERN_ERR |
1da177e4 LT |
1503 | "svc_recv: service %p, wait queue active!\n", |
1504 | rqstp); | |
1505 | ||
1da177e4 LT |
1506 | |
1507 | /* now allocate needed pages. If we get a failure, sleep briefly */ | |
c6b0a9f8 | 1508 | pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE; |
44524359 N |
1509 | for (i=0; i < pages ; i++) |
1510 | while (rqstp->rq_pages[i] == NULL) { | |
1511 | struct page *p = alloc_page(GFP_KERNEL); | |
1512 | if (!p) | |
1513 | schedule_timeout_uninterruptible(msecs_to_jiffies(500)); | |
1514 | rqstp->rq_pages[i] = p; | |
1da177e4 | 1515 | } |
250f3915 N |
1516 | rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */ |
1517 | BUG_ON(pages >= RPCSVC_MAXPAGES); | |
1da177e4 LT |
1518 | |
1519 | /* Make arg->head point to first page and arg->pages point to rest */ | |
1520 | arg = &rqstp->rq_arg; | |
44524359 | 1521 | arg->head[0].iov_base = page_address(rqstp->rq_pages[0]); |
1da177e4 | 1522 | arg->head[0].iov_len = PAGE_SIZE; |
44524359 | 1523 | arg->pages = rqstp->rq_pages + 1; |
1da177e4 LT |
1524 | arg->page_base = 0; |
1525 | /* save at least one page for response */ | |
1526 | arg->page_len = (pages-2)*PAGE_SIZE; | |
1527 | arg->len = (pages-1)*PAGE_SIZE; | |
1528 | arg->tail[0].iov_len = 0; | |
3e1d1d28 CL |
1529 | |
1530 | try_to_freeze(); | |
1887b935 | 1531 | cond_resched(); |
1da177e4 LT |
1532 | if (signalled()) |
1533 | return -EINTR; | |
1534 | ||
3262c816 GB |
1535 | spin_lock_bh(&pool->sp_lock); |
1536 | if ((svsk = svc_sock_dequeue(pool)) != NULL) { | |
1da177e4 | 1537 | rqstp->rq_sock = svsk; |
c45c357d | 1538 | atomic_inc(&svsk->sk_inuse); |
c6b0a9f8 | 1539 | rqstp->rq_reserved = serv->sv_max_mesg; |
5685f0fa | 1540 | atomic_add(rqstp->rq_reserved, &svsk->sk_reserved); |
1da177e4 LT |
1541 | } else { |
1542 | /* No data pending. Go to sleep */ | |
3262c816 | 1543 | svc_thread_enqueue(pool, rqstp); |
1da177e4 LT |
1544 | |
1545 | /* | |
1546 | * We have to be able to interrupt this wait | |
1547 | * to bring down the daemons ... | |
1548 | */ | |
1549 | set_current_state(TASK_INTERRUPTIBLE); | |
1550 | add_wait_queue(&rqstp->rq_wait, &wait); | |
3262c816 | 1551 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 LT |
1552 | |
1553 | schedule_timeout(timeout); | |
1554 | ||
3e1d1d28 | 1555 | try_to_freeze(); |
1da177e4 | 1556 | |
3262c816 | 1557 | spin_lock_bh(&pool->sp_lock); |
1da177e4 LT |
1558 | remove_wait_queue(&rqstp->rq_wait, &wait); |
1559 | ||
1560 | if (!(svsk = rqstp->rq_sock)) { | |
3262c816 GB |
1561 | svc_thread_dequeue(pool, rqstp); |
1562 | spin_unlock_bh(&pool->sp_lock); | |
1da177e4 LT |
1563 | dprintk("svc: server %p, no data yet\n", rqstp); |
1564 | return signalled()? -EINTR : -EAGAIN; | |
1565 | } | |
1566 | } | |
3262c816 | 1567 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 | 1568 | |
d7979ae4 TT |
1569 | len = 0; |
1570 | if (test_bit(SK_CLOSE, &svsk->sk_flags)) { | |
1571 | dprintk("svc_recv: found SK_CLOSE\n"); | |
1572 | svc_delete_socket(svsk); | |
38a417cc TT |
1573 | } else if (test_bit(SK_LISTENER, &svsk->sk_flags)) { |
1574 | struct svc_xprt *newxpt; | |
1575 | newxpt = svsk->sk_xprt.xpt_ops->xpo_accept(&svsk->sk_xprt); | |
f9f3cc4f TT |
1576 | if (newxpt) |
1577 | svc_check_conn_limits(svsk->sk_server); | |
38a417cc | 1578 | svc_sock_received(svsk); |
d7979ae4 TT |
1579 | } else { |
1580 | dprintk("svc: server %p, pool %u, socket %p, inuse=%d\n", | |
1581 | rqstp, pool->sp_id, svsk, atomic_read(&svsk->sk_inuse)); | |
1582 | len = svsk->sk_xprt.xpt_ops->xpo_recvfrom(rqstp); | |
1583 | dprintk("svc: got len=%d\n", len); | |
1584 | } | |
1da177e4 LT |
1585 | |
1586 | /* No data, incomplete (TCP) read, or accept() */ | |
1587 | if (len == 0 || len == -EAGAIN) { | |
1588 | rqstp->rq_res.len = 0; | |
1589 | svc_sock_release(rqstp); | |
1590 | return -EAGAIN; | |
1591 | } | |
1592 | svsk->sk_lastrecv = get_seconds(); | |
36bdfc8b | 1593 | clear_bit(SK_OLD, &svsk->sk_flags); |
1da177e4 | 1594 | |
bcdb81ae | 1595 | rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp)); |
1da177e4 LT |
1596 | rqstp->rq_chandle.defer = svc_defer; |
1597 | ||
1598 | if (serv->sv_stats) | |
1599 | serv->sv_stats->netcnt++; | |
1600 | return len; | |
1601 | } | |
1602 | ||
cca5172a | 1603 | /* |
1da177e4 LT |
1604 | * Drop request |
1605 | */ | |
1606 | void | |
1607 | svc_drop(struct svc_rqst *rqstp) | |
1608 | { | |
1609 | dprintk("svc: socket %p dropped request\n", rqstp->rq_sock); | |
1610 | svc_sock_release(rqstp); | |
1611 | } | |
1612 | ||
1613 | /* | |
1614 | * Return reply to client. | |
1615 | */ | |
1616 | int | |
1617 | svc_send(struct svc_rqst *rqstp) | |
1618 | { | |
1619 | struct svc_sock *svsk; | |
1620 | int len; | |
1621 | struct xdr_buf *xb; | |
1622 | ||
1623 | if ((svsk = rqstp->rq_sock) == NULL) { | |
1624 | printk(KERN_WARNING "NULL socket pointer in %s:%d\n", | |
1625 | __FILE__, __LINE__); | |
1626 | return -EFAULT; | |
1627 | } | |
1628 | ||
1629 | /* release the receive skb before sending the reply */ | |
5148bf4e | 1630 | rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp); |
1da177e4 LT |
1631 | |
1632 | /* calculate over-all length */ | |
1633 | xb = & rqstp->rq_res; | |
1634 | xb->len = xb->head[0].iov_len + | |
1635 | xb->page_len + | |
1636 | xb->tail[0].iov_len; | |
1637 | ||
57b47a53 IM |
1638 | /* Grab svsk->sk_mutex to serialize outgoing data. */ |
1639 | mutex_lock(&svsk->sk_mutex); | |
1da177e4 LT |
1640 | if (test_bit(SK_DEAD, &svsk->sk_flags)) |
1641 | len = -ENOTCONN; | |
1642 | else | |
5d137990 | 1643 | len = svsk->sk_xprt.xpt_ops->xpo_sendto(rqstp); |
57b47a53 | 1644 | mutex_unlock(&svsk->sk_mutex); |
1da177e4 LT |
1645 | svc_sock_release(rqstp); |
1646 | ||
1647 | if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN) | |
1648 | return 0; | |
1649 | return len; | |
1650 | } | |
1651 | ||
36bdfc8b GB |
1652 | /* |
1653 | * Timer function to close old temporary sockets, using | |
1654 | * a mark-and-sweep algorithm. | |
1655 | */ | |
1656 | static void | |
1657 | svc_age_temp_sockets(unsigned long closure) | |
1658 | { | |
1659 | struct svc_serv *serv = (struct svc_serv *)closure; | |
1660 | struct svc_sock *svsk; | |
1661 | struct list_head *le, *next; | |
1662 | LIST_HEAD(to_be_aged); | |
1663 | ||
1664 | dprintk("svc_age_temp_sockets\n"); | |
1665 | ||
1666 | if (!spin_trylock_bh(&serv->sv_lock)) { | |
1667 | /* busy, try again 1 sec later */ | |
1668 | dprintk("svc_age_temp_sockets: busy\n"); | |
1669 | mod_timer(&serv->sv_temptimer, jiffies + HZ); | |
1670 | return; | |
1671 | } | |
1672 | ||
1673 | list_for_each_safe(le, next, &serv->sv_tempsocks) { | |
1674 | svsk = list_entry(le, struct svc_sock, sk_list); | |
1675 | ||
1676 | if (!test_and_set_bit(SK_OLD, &svsk->sk_flags)) | |
1677 | continue; | |
7a1fa065 | 1678 | if (atomic_read(&svsk->sk_inuse) > 1 || test_bit(SK_BUSY, &svsk->sk_flags)) |
36bdfc8b | 1679 | continue; |
c45c357d | 1680 | atomic_inc(&svsk->sk_inuse); |
36bdfc8b GB |
1681 | list_move(le, &to_be_aged); |
1682 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1683 | set_bit(SK_DETACHED, &svsk->sk_flags); | |
1684 | } | |
1685 | spin_unlock_bh(&serv->sv_lock); | |
1686 | ||
1687 | while (!list_empty(&to_be_aged)) { | |
1688 | le = to_be_aged.next; | |
1689 | /* fiddling the sk_list node is safe 'cos we're SK_DETACHED */ | |
1690 | list_del_init(le); | |
1691 | svsk = list_entry(le, struct svc_sock, sk_list); | |
1692 | ||
1693 | dprintk("queuing svsk %p for closing, %lu seconds old\n", | |
1694 | svsk, get_seconds() - svsk->sk_lastrecv); | |
1695 | ||
1696 | /* a thread will dequeue and close it soon */ | |
1697 | svc_sock_enqueue(svsk); | |
1698 | svc_sock_put(svsk); | |
1699 | } | |
1700 | ||
1701 | mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ); | |
1702 | } | |
1703 | ||
1da177e4 LT |
1704 | /* |
1705 | * Initialize socket for RPC use and create svc_sock struct | |
1706 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1707 | */ | |
6b174337 CL |
1708 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1709 | struct socket *sock, | |
1710 | int *errp, int flags) | |
1da177e4 LT |
1711 | { |
1712 | struct svc_sock *svsk; | |
1713 | struct sock *inet; | |
6b174337 CL |
1714 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
1715 | int is_temporary = flags & SVC_SOCK_TEMPORARY; | |
1da177e4 LT |
1716 | |
1717 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1718 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1719 | *errp = -ENOMEM; |
1720 | return NULL; | |
1721 | } | |
1da177e4 LT |
1722 | |
1723 | inet = sock->sk; | |
1724 | ||
1725 | /* Register socket with portmapper */ | |
1726 | if (*errp >= 0 && pmap_register) | |
1727 | *errp = svc_register(serv, inet->sk_protocol, | |
1728 | ntohs(inet_sk(inet)->sport)); | |
1729 | ||
1730 | if (*errp < 0) { | |
1731 | kfree(svsk); | |
1732 | return NULL; | |
1733 | } | |
1734 | ||
1735 | set_bit(SK_BUSY, &svsk->sk_flags); | |
1736 | inet->sk_user_data = svsk; | |
1737 | svsk->sk_sock = sock; | |
1738 | svsk->sk_sk = inet; | |
1739 | svsk->sk_ostate = inet->sk_state_change; | |
1740 | svsk->sk_odata = inet->sk_data_ready; | |
1741 | svsk->sk_owspace = inet->sk_write_space; | |
1742 | svsk->sk_server = serv; | |
aaf68cfb | 1743 | atomic_set(&svsk->sk_inuse, 1); |
1da177e4 | 1744 | svsk->sk_lastrecv = get_seconds(); |
7ac1bea5 | 1745 | spin_lock_init(&svsk->sk_lock); |
1da177e4 LT |
1746 | INIT_LIST_HEAD(&svsk->sk_deferred); |
1747 | INIT_LIST_HEAD(&svsk->sk_ready); | |
57b47a53 | 1748 | mutex_init(&svsk->sk_mutex); |
1da177e4 LT |
1749 | |
1750 | /* Initialize the socket */ | |
1751 | if (sock->type == SOCK_DGRAM) | |
1752 | svc_udp_init(svsk); | |
1753 | else | |
1754 | svc_tcp_init(svsk); | |
1755 | ||
1756 | spin_lock_bh(&serv->sv_lock); | |
6b174337 | 1757 | if (is_temporary) { |
1da177e4 LT |
1758 | set_bit(SK_TEMP, &svsk->sk_flags); |
1759 | list_add(&svsk->sk_list, &serv->sv_tempsocks); | |
1760 | serv->sv_tmpcnt++; | |
36bdfc8b GB |
1761 | if (serv->sv_temptimer.function == NULL) { |
1762 | /* setup timer to age temp sockets */ | |
1763 | setup_timer(&serv->sv_temptimer, svc_age_temp_sockets, | |
1764 | (unsigned long)serv); | |
1765 | mod_timer(&serv->sv_temptimer, | |
1766 | jiffies + svc_conn_age_period * HZ); | |
1767 | } | |
1da177e4 LT |
1768 | } else { |
1769 | clear_bit(SK_TEMP, &svsk->sk_flags); | |
1770 | list_add(&svsk->sk_list, &serv->sv_permsocks); | |
1771 | } | |
1772 | spin_unlock_bh(&serv->sv_lock); | |
1773 | ||
1774 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", | |
1775 | svsk, svsk->sk_sk); | |
1776 | ||
1da177e4 LT |
1777 | return svsk; |
1778 | } | |
1779 | ||
b41b66d6 N |
1780 | int svc_addsock(struct svc_serv *serv, |
1781 | int fd, | |
1782 | char *name_return, | |
1783 | int *proto) | |
1784 | { | |
1785 | int err = 0; | |
1786 | struct socket *so = sockfd_lookup(fd, &err); | |
1787 | struct svc_sock *svsk = NULL; | |
1788 | ||
1789 | if (!so) | |
1790 | return err; | |
1791 | if (so->sk->sk_family != AF_INET) | |
1792 | err = -EAFNOSUPPORT; | |
1793 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1794 | so->sk->sk_protocol != IPPROTO_UDP) | |
1795 | err = -EPROTONOSUPPORT; | |
1796 | else if (so->state > SS_UNCONNECTED) | |
1797 | err = -EISCONN; | |
1798 | else { | |
6b174337 | 1799 | svsk = svc_setup_socket(serv, so, &err, SVC_SOCK_DEFAULTS); |
e79eff1f N |
1800 | if (svsk) { |
1801 | svc_sock_received(svsk); | |
b41b66d6 | 1802 | err = 0; |
e79eff1f | 1803 | } |
b41b66d6 N |
1804 | } |
1805 | if (err) { | |
1806 | sockfd_put(so); | |
1807 | return err; | |
1808 | } | |
1809 | if (proto) *proto = so->sk->sk_protocol; | |
1810 | return one_sock_name(name_return, svsk); | |
1811 | } | |
1812 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1813 | ||
1da177e4 LT |
1814 | /* |
1815 | * Create socket for RPC service. | |
1816 | */ | |
6b174337 | 1817 | static int svc_create_socket(struct svc_serv *serv, int protocol, |
77f1f67a | 1818 | struct sockaddr *sin, int len, int flags) |
1da177e4 LT |
1819 | { |
1820 | struct svc_sock *svsk; | |
1821 | struct socket *sock; | |
1822 | int error; | |
1823 | int type; | |
ad06e4bd | 1824 | char buf[RPC_MAX_ADDRBUFLEN]; |
1da177e4 | 1825 | |
ad06e4bd CL |
1826 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1827 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1828 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1829 | |
1830 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1831 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1832 | "sockets supported\n"); | |
1833 | return -EINVAL; | |
1834 | } | |
1835 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; | |
1836 | ||
77f1f67a CL |
1837 | error = sock_create_kern(sin->sa_family, type, protocol, &sock); |
1838 | if (error < 0) | |
1da177e4 LT |
1839 | return error; |
1840 | ||
ed07536e PZ |
1841 | svc_reclassify_socket(sock); |
1842 | ||
18114746 | 1843 | if (type == SOCK_STREAM) |
77f1f67a CL |
1844 | sock->sk->sk_reuse = 1; /* allow address reuse */ |
1845 | error = kernel_bind(sock, sin, len); | |
18114746 ES |
1846 | if (error < 0) |
1847 | goto bummer; | |
1da177e4 LT |
1848 | |
1849 | if (protocol == IPPROTO_TCP) { | |
e6242e92 | 1850 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1851 | goto bummer; |
1852 | } | |
1853 | ||
e79eff1f N |
1854 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { |
1855 | svc_sock_received(svsk); | |
6b174337 | 1856 | return ntohs(inet_sk(svsk->sk_sk)->sport); |
e79eff1f | 1857 | } |
1da177e4 LT |
1858 | |
1859 | bummer: | |
1860 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1861 | sock_release(sock); | |
1862 | return error; | |
1863 | } | |
1864 | ||
755cceab TT |
1865 | /* |
1866 | * Detach the svc_sock from the socket so that no | |
1867 | * more callbacks occur. | |
1868 | */ | |
1869 | static void svc_sock_detach(struct svc_xprt *xprt) | |
1870 | { | |
1871 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1872 | struct sock *sk = svsk->sk_sk; | |
1873 | ||
1874 | dprintk("svc: svc_sock_detach(%p)\n", svsk); | |
1875 | ||
1876 | /* put back the old socket callbacks */ | |
1877 | sk->sk_state_change = svsk->sk_ostate; | |
1878 | sk->sk_data_ready = svsk->sk_odata; | |
1879 | sk->sk_write_space = svsk->sk_owspace; | |
1880 | } | |
1881 | ||
1882 | /* | |
1883 | * Free the svc_sock's socket resources and the svc_sock itself. | |
1884 | */ | |
1885 | static void svc_sock_free(struct svc_xprt *xprt) | |
1886 | { | |
1887 | struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); | |
1888 | dprintk("svc: svc_sock_free(%p)\n", svsk); | |
1889 | ||
1890 | if (svsk->sk_info_authunix != NULL) | |
1891 | svcauth_unix_info_release(svsk->sk_info_authunix); | |
1892 | if (svsk->sk_sock->file) | |
1893 | sockfd_put(svsk->sk_sock); | |
1894 | else | |
1895 | sock_release(svsk->sk_sock); | |
1896 | kfree(svsk); | |
1897 | } | |
1898 | ||
1da177e4 LT |
1899 | /* |
1900 | * Remove a dead socket | |
1901 | */ | |
aaf68cfb | 1902 | static void |
1da177e4 LT |
1903 | svc_delete_socket(struct svc_sock *svsk) |
1904 | { | |
1905 | struct svc_serv *serv; | |
1906 | struct sock *sk; | |
1907 | ||
1908 | dprintk("svc: svc_delete_socket(%p)\n", svsk); | |
1909 | ||
1910 | serv = svsk->sk_server; | |
1911 | sk = svsk->sk_sk; | |
1912 | ||
755cceab | 1913 | svsk->sk_xprt.xpt_ops->xpo_detach(&svsk->sk_xprt); |
1da177e4 LT |
1914 | |
1915 | spin_lock_bh(&serv->sv_lock); | |
1916 | ||
36bdfc8b GB |
1917 | if (!test_and_set_bit(SK_DETACHED, &svsk->sk_flags)) |
1918 | list_del_init(&svsk->sk_list); | |
cca5172a | 1919 | /* |
3262c816 GB |
1920 | * We used to delete the svc_sock from whichever list |
1921 | * it's sk_ready node was on, but we don't actually | |
1922 | * need to. This is because the only time we're called | |
1923 | * while still attached to a queue, the queue itself | |
1924 | * is about to be destroyed (in svc_destroy). | |
1925 | */ | |
aaf68cfb N |
1926 | if (!test_and_set_bit(SK_DEAD, &svsk->sk_flags)) { |
1927 | BUG_ON(atomic_read(&svsk->sk_inuse)<2); | |
1928 | atomic_dec(&svsk->sk_inuse); | |
1da177e4 LT |
1929 | if (test_bit(SK_TEMP, &svsk->sk_flags)) |
1930 | serv->sv_tmpcnt--; | |
aaf68cfb | 1931 | } |
1da177e4 | 1932 | |
d6740df9 | 1933 | spin_unlock_bh(&serv->sv_lock); |
aaf68cfb N |
1934 | } |
1935 | ||
cda1fd4a | 1936 | static void svc_close_socket(struct svc_sock *svsk) |
aaf68cfb N |
1937 | { |
1938 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1939 | if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) | |
1940 | /* someone else will have to effect the close */ | |
1941 | return; | |
1942 | ||
1943 | atomic_inc(&svsk->sk_inuse); | |
1944 | svc_delete_socket(svsk); | |
1945 | clear_bit(SK_BUSY, &svsk->sk_flags); | |
d6740df9 | 1946 | svc_sock_put(svsk); |
1da177e4 LT |
1947 | } |
1948 | ||
cda1fd4a N |
1949 | void svc_force_close_socket(struct svc_sock *svsk) |
1950 | { | |
1951 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1952 | if (test_bit(SK_BUSY, &svsk->sk_flags)) { | |
1953 | /* Waiting to be processed, but no threads left, | |
1954 | * So just remove it from the waiting list | |
1955 | */ | |
1956 | list_del_init(&svsk->sk_ready); | |
1957 | clear_bit(SK_BUSY, &svsk->sk_flags); | |
1958 | } | |
1959 | svc_close_socket(svsk); | |
1960 | } | |
1961 | ||
6b174337 CL |
1962 | /** |
1963 | * svc_makesock - Make a socket for nfsd and lockd | |
1964 | * @serv: RPC server structure | |
1965 | * @protocol: transport protocol to use | |
1966 | * @port: port to use | |
482fb94e | 1967 | * @flags: requested socket characteristics |
6b174337 | 1968 | * |
1da177e4 | 1969 | */ |
482fb94e CL |
1970 | int svc_makesock(struct svc_serv *serv, int protocol, unsigned short port, |
1971 | int flags) | |
1da177e4 | 1972 | { |
6b174337 CL |
1973 | struct sockaddr_in sin = { |
1974 | .sin_family = AF_INET, | |
1975 | .sin_addr.s_addr = INADDR_ANY, | |
1976 | .sin_port = htons(port), | |
1977 | }; | |
1da177e4 LT |
1978 | |
1979 | dprintk("svc: creating socket proto = %d\n", protocol); | |
77f1f67a CL |
1980 | return svc_create_socket(serv, protocol, (struct sockaddr *) &sin, |
1981 | sizeof(sin), flags); | |
1da177e4 LT |
1982 | } |
1983 | ||
1984 | /* | |
cca5172a | 1985 | * Handle defer and revisit of requests |
1da177e4 LT |
1986 | */ |
1987 | ||
1988 | static void svc_revisit(struct cache_deferred_req *dreq, int too_many) | |
1989 | { | |
1990 | struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle); | |
1da177e4 LT |
1991 | struct svc_sock *svsk; |
1992 | ||
1993 | if (too_many) { | |
1994 | svc_sock_put(dr->svsk); | |
1995 | kfree(dr); | |
1996 | return; | |
1997 | } | |
1998 | dprintk("revisit queued\n"); | |
1999 | svsk = dr->svsk; | |
2000 | dr->svsk = NULL; | |
7ac1bea5 | 2001 | spin_lock(&svsk->sk_lock); |
1da177e4 | 2002 | list_add(&dr->handle.recent, &svsk->sk_deferred); |
7ac1bea5 | 2003 | spin_unlock(&svsk->sk_lock); |
1da177e4 LT |
2004 | set_bit(SK_DEFERRED, &svsk->sk_flags); |
2005 | svc_sock_enqueue(svsk); | |
2006 | svc_sock_put(svsk); | |
2007 | } | |
2008 | ||
2009 | static struct cache_deferred_req * | |
2010 | svc_defer(struct cache_req *req) | |
2011 | { | |
2012 | struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle); | |
2013 | int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len); | |
2014 | struct svc_deferred_req *dr; | |
2015 | ||
2016 | if (rqstp->rq_arg.page_len) | |
2017 | return NULL; /* if more than a page, give up FIXME */ | |
2018 | if (rqstp->rq_deferred) { | |
2019 | dr = rqstp->rq_deferred; | |
2020 | rqstp->rq_deferred = NULL; | |
2021 | } else { | |
2022 | int skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
2023 | /* FIXME maybe discard if size too large */ | |
2024 | dr = kmalloc(size, GFP_KERNEL); | |
2025 | if (dr == NULL) | |
2026 | return NULL; | |
2027 | ||
2028 | dr->handle.owner = rqstp->rq_server; | |
2029 | dr->prot = rqstp->rq_prot; | |
24422222 CL |
2030 | memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen); |
2031 | dr->addrlen = rqstp->rq_addrlen; | |
1918e341 | 2032 | dr->daddr = rqstp->rq_daddr; |
1da177e4 LT |
2033 | dr->argslen = rqstp->rq_arg.len >> 2; |
2034 | memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2); | |
2035 | } | |
c45c357d | 2036 | atomic_inc(&rqstp->rq_sock->sk_inuse); |
1da177e4 | 2037 | dr->svsk = rqstp->rq_sock; |
1da177e4 LT |
2038 | |
2039 | dr->handle.revisit = svc_revisit; | |
2040 | return &dr->handle; | |
2041 | } | |
2042 | ||
2043 | /* | |
2044 | * recv data from a deferred request into an active one | |
2045 | */ | |
2046 | static int svc_deferred_recv(struct svc_rqst *rqstp) | |
2047 | { | |
2048 | struct svc_deferred_req *dr = rqstp->rq_deferred; | |
2049 | ||
2050 | rqstp->rq_arg.head[0].iov_base = dr->args; | |
2051 | rqstp->rq_arg.head[0].iov_len = dr->argslen<<2; | |
2052 | rqstp->rq_arg.page_len = 0; | |
2053 | rqstp->rq_arg.len = dr->argslen<<2; | |
2054 | rqstp->rq_prot = dr->prot; | |
24422222 CL |
2055 | memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen); |
2056 | rqstp->rq_addrlen = dr->addrlen; | |
1918e341 | 2057 | rqstp->rq_daddr = dr->daddr; |
44524359 | 2058 | rqstp->rq_respages = rqstp->rq_pages; |
1da177e4 LT |
2059 | return dr->argslen<<2; |
2060 | } | |
2061 | ||
2062 | ||
2063 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk) | |
2064 | { | |
2065 | struct svc_deferred_req *dr = NULL; | |
cca5172a | 2066 | |
1da177e4 LT |
2067 | if (!test_bit(SK_DEFERRED, &svsk->sk_flags)) |
2068 | return NULL; | |
7ac1bea5 | 2069 | spin_lock(&svsk->sk_lock); |
1da177e4 LT |
2070 | clear_bit(SK_DEFERRED, &svsk->sk_flags); |
2071 | if (!list_empty(&svsk->sk_deferred)) { | |
2072 | dr = list_entry(svsk->sk_deferred.next, | |
2073 | struct svc_deferred_req, | |
2074 | handle.recent); | |
2075 | list_del_init(&dr->handle.recent); | |
2076 | set_bit(SK_DEFERRED, &svsk->sk_flags); | |
2077 | } | |
7ac1bea5 | 2078 | spin_unlock(&svsk->sk_lock); |
1da177e4 LT |
2079 | return dr; |
2080 | } |