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 | ||
22 | #include <linux/sched.h> | |
23 | #include <linux/errno.h> | |
24 | #include <linux/fcntl.h> | |
25 | #include <linux/net.h> | |
26 | #include <linux/in.h> | |
27 | #include <linux/inet.h> | |
28 | #include <linux/udp.h> | |
91483c4b | 29 | #include <linux/tcp.h> |
1da177e4 LT |
30 | #include <linux/unistd.h> |
31 | #include <linux/slab.h> | |
32 | #include <linux/netdevice.h> | |
33 | #include <linux/skbuff.h> | |
b41b66d6 | 34 | #include <linux/file.h> |
7dfb7103 | 35 | #include <linux/freezer.h> |
1da177e4 LT |
36 | #include <net/sock.h> |
37 | #include <net/checksum.h> | |
38 | #include <net/ip.h> | |
b92503b2 | 39 | #include <net/ipv6.h> |
c752f073 | 40 | #include <net/tcp_states.h> |
1da177e4 LT |
41 | #include <asm/uaccess.h> |
42 | #include <asm/ioctls.h> | |
43 | ||
44 | #include <linux/sunrpc/types.h> | |
ad06e4bd | 45 | #include <linux/sunrpc/clnt.h> |
1da177e4 LT |
46 | #include <linux/sunrpc/xdr.h> |
47 | #include <linux/sunrpc/svcsock.h> | |
48 | #include <linux/sunrpc/stats.h> | |
49 | ||
50 | /* SMP locking strategy: | |
51 | * | |
3262c816 GB |
52 | * svc_pool->sp_lock protects most of the fields of that pool. |
53 | * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt. | |
54 | * when both need to be taken (rare), svc_serv->sv_lock is first. | |
55 | * BKL protects svc_serv->sv_nrthread. | |
1a68d952 | 56 | * svc_sock->sk_defer_lock protects the svc_sock->sk_deferred list |
c081a0c7 | 57 | * svc_sock->sk_flags.SK_BUSY prevents a svc_sock being enqueued multiply. |
1da177e4 LT |
58 | * |
59 | * Some flags can be set to certain values at any time | |
60 | * providing that certain rules are followed: | |
61 | * | |
1da177e4 | 62 | * SK_CONN, SK_DATA, can be set or cleared at any time. |
cca5172a | 63 | * after a set, svc_sock_enqueue must be called. |
1da177e4 LT |
64 | * after a clear, the socket must be read/accepted |
65 | * if this succeeds, it must be set again. | |
66 | * SK_CLOSE can set at any time. It is never cleared. | |
aaf68cfb N |
67 | * sk_inuse contains a bias of '1' until SK_DEAD is set. |
68 | * so when sk_inuse hits zero, we know the socket is dead | |
69 | * and no-one is using it. | |
70 | * SK_DEAD can only be set while SK_BUSY is held which ensures | |
71 | * no other thread will be using the socket or will try to | |
72 | * set SK_DEAD. | |
1da177e4 LT |
73 | * |
74 | */ | |
75 | ||
76 | #define RPCDBG_FACILITY RPCDBG_SVCSOCK | |
77 | ||
78 | ||
79 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
6b174337 | 80 | int *errp, int flags); |
aaf68cfb | 81 | static void svc_delete_socket(struct svc_sock *svsk); |
1da177e4 LT |
82 | static void svc_udp_data_ready(struct sock *, int); |
83 | static int svc_udp_recvfrom(struct svc_rqst *); | |
84 | static int svc_udp_sendto(struct svc_rqst *); | |
cda1fd4a | 85 | static void svc_close_socket(struct svc_sock *svsk); |
1da177e4 LT |
86 | |
87 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk); | |
88 | static int svc_deferred_recv(struct svc_rqst *rqstp); | |
89 | static struct cache_deferred_req *svc_defer(struct cache_req *req); | |
90 | ||
36bdfc8b GB |
91 | /* apparently the "standard" is that clients close |
92 | * idle connections after 5 minutes, servers after | |
93 | * 6 minutes | |
94 | * http://www.connectathon.org/talks96/nfstcp.pdf | |
95 | */ | |
96 | static int svc_conn_age_period = 6*60; | |
97 | ||
ed07536e PZ |
98 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
99 | static struct lock_class_key svc_key[2]; | |
100 | static struct lock_class_key svc_slock_key[2]; | |
101 | ||
102 | static inline void svc_reclassify_socket(struct socket *sock) | |
103 | { | |
104 | struct sock *sk = sock->sk; | |
105 | BUG_ON(sk->sk_lock.owner != NULL); | |
106 | switch (sk->sk_family) { | |
107 | case AF_INET: | |
108 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
109 | &svc_slock_key[0], "sk_lock-AF_INET-NFSD", &svc_key[0]); | |
110 | break; | |
111 | ||
112 | case AF_INET6: | |
113 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
114 | &svc_slock_key[1], "sk_lock-AF_INET6-NFSD", &svc_key[1]); | |
115 | break; | |
116 | ||
117 | default: | |
118 | BUG(); | |
119 | } | |
120 | } | |
121 | #else | |
122 | static inline void svc_reclassify_socket(struct socket *sock) | |
123 | { | |
124 | } | |
125 | #endif | |
126 | ||
ad06e4bd CL |
127 | static char *__svc_print_addr(struct sockaddr *addr, char *buf, size_t len) |
128 | { | |
129 | switch (addr->sa_family) { | |
130 | case AF_INET: | |
131 | snprintf(buf, len, "%u.%u.%u.%u, port=%u", | |
132 | NIPQUAD(((struct sockaddr_in *) addr)->sin_addr), | |
133 | htons(((struct sockaddr_in *) addr)->sin_port)); | |
134 | break; | |
5a05ed73 | 135 | |
ad06e4bd CL |
136 | case AF_INET6: |
137 | snprintf(buf, len, "%x:%x:%x:%x:%x:%x:%x:%x, port=%u", | |
138 | NIP6(((struct sockaddr_in6 *) addr)->sin6_addr), | |
139 | htons(((struct sockaddr_in6 *) addr)->sin6_port)); | |
140 | break; | |
5a05ed73 | 141 | |
ad06e4bd CL |
142 | default: |
143 | snprintf(buf, len, "unknown address type: %d", addr->sa_family); | |
144 | break; | |
145 | } | |
146 | return buf; | |
147 | } | |
148 | ||
149 | /** | |
150 | * svc_print_addr - Format rq_addr field for printing | |
151 | * @rqstp: svc_rqst struct containing address to print | |
152 | * @buf: target buffer for formatted address | |
153 | * @len: length of target buffer | |
154 | * | |
155 | */ | |
156 | char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len) | |
157 | { | |
27459f09 | 158 | return __svc_print_addr(svc_addr(rqstp), buf, len); |
ad06e4bd CL |
159 | } |
160 | EXPORT_SYMBOL_GPL(svc_print_addr); | |
161 | ||
1da177e4 | 162 | /* |
3262c816 | 163 | * Queue up an idle server thread. Must have pool->sp_lock held. |
1da177e4 | 164 | * Note: this is really a stack rather than a queue, so that we only |
3262c816 | 165 | * use as many different threads as we need, and the rest don't pollute |
1da177e4 LT |
166 | * the cache. |
167 | */ | |
168 | static inline void | |
3262c816 | 169 | svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp) |
1da177e4 | 170 | { |
3262c816 | 171 | list_add(&rqstp->rq_list, &pool->sp_threads); |
1da177e4 LT |
172 | } |
173 | ||
174 | /* | |
3262c816 | 175 | * Dequeue an nfsd thread. Must have pool->sp_lock held. |
1da177e4 LT |
176 | */ |
177 | static inline void | |
3262c816 | 178 | svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp) |
1da177e4 LT |
179 | { |
180 | list_del(&rqstp->rq_list); | |
181 | } | |
182 | ||
183 | /* | |
184 | * Release an skbuff after use | |
185 | */ | |
186 | static inline void | |
187 | svc_release_skb(struct svc_rqst *rqstp) | |
188 | { | |
189 | struct sk_buff *skb = rqstp->rq_skbuff; | |
190 | struct svc_deferred_req *dr = rqstp->rq_deferred; | |
191 | ||
192 | if (skb) { | |
193 | rqstp->rq_skbuff = NULL; | |
194 | ||
195 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
196 | skb_free_datagram(rqstp->rq_sock->sk_sk, skb); | |
197 | } | |
198 | if (dr) { | |
199 | rqstp->rq_deferred = NULL; | |
200 | kfree(dr); | |
201 | } | |
202 | } | |
203 | ||
204 | /* | |
205 | * Any space to write? | |
206 | */ | |
207 | static inline unsigned long | |
208 | svc_sock_wspace(struct svc_sock *svsk) | |
209 | { | |
210 | int wspace; | |
211 | ||
212 | if (svsk->sk_sock->type == SOCK_STREAM) | |
213 | wspace = sk_stream_wspace(svsk->sk_sk); | |
214 | else | |
215 | wspace = sock_wspace(svsk->sk_sk); | |
216 | ||
217 | return wspace; | |
218 | } | |
219 | ||
220 | /* | |
221 | * Queue up a socket with data pending. If there are idle nfsd | |
222 | * processes, wake 'em up. | |
223 | * | |
224 | */ | |
225 | static void | |
226 | svc_sock_enqueue(struct svc_sock *svsk) | |
227 | { | |
228 | struct svc_serv *serv = svsk->sk_server; | |
bfd24160 | 229 | struct svc_pool *pool; |
1da177e4 | 230 | struct svc_rqst *rqstp; |
bfd24160 | 231 | int cpu; |
1da177e4 LT |
232 | |
233 | if (!(svsk->sk_flags & | |
234 | ( (1<<SK_CONN)|(1<<SK_DATA)|(1<<SK_CLOSE)|(1<<SK_DEFERRED)) )) | |
235 | return; | |
236 | if (test_bit(SK_DEAD, &svsk->sk_flags)) | |
237 | return; | |
238 | ||
bfd24160 GB |
239 | cpu = get_cpu(); |
240 | pool = svc_pool_for_cpu(svsk->sk_server, cpu); | |
241 | put_cpu(); | |
242 | ||
3262c816 | 243 | spin_lock_bh(&pool->sp_lock); |
1da177e4 | 244 | |
3262c816 GB |
245 | if (!list_empty(&pool->sp_threads) && |
246 | !list_empty(&pool->sp_sockets)) | |
1da177e4 LT |
247 | printk(KERN_ERR |
248 | "svc_sock_enqueue: threads and sockets both waiting??\n"); | |
249 | ||
250 | if (test_bit(SK_DEAD, &svsk->sk_flags)) { | |
251 | /* Don't enqueue dead sockets */ | |
252 | dprintk("svc: socket %p is dead, not enqueued\n", svsk->sk_sk); | |
253 | goto out_unlock; | |
254 | } | |
255 | ||
c081a0c7 GB |
256 | /* Mark socket as busy. It will remain in this state until the |
257 | * server has processed all pending data and put the socket back | |
258 | * on the idle list. We update SK_BUSY atomically because | |
259 | * it also guards against trying to enqueue the svc_sock twice. | |
260 | */ | |
261 | if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) { | |
262 | /* Don't enqueue socket while already enqueued */ | |
1da177e4 LT |
263 | dprintk("svc: socket %p busy, not enqueued\n", svsk->sk_sk); |
264 | goto out_unlock; | |
265 | } | |
3262c816 GB |
266 | BUG_ON(svsk->sk_pool != NULL); |
267 | svsk->sk_pool = pool; | |
1da177e4 LT |
268 | |
269 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
c6b0a9f8 | 270 | if (((atomic_read(&svsk->sk_reserved) + serv->sv_max_mesg)*2 |
1da177e4 LT |
271 | > svc_sock_wspace(svsk)) |
272 | && !test_bit(SK_CLOSE, &svsk->sk_flags) | |
273 | && !test_bit(SK_CONN, &svsk->sk_flags)) { | |
274 | /* Don't enqueue while not enough space for reply */ | |
275 | dprintk("svc: socket %p no space, %d*2 > %ld, not enqueued\n", | |
c6b0a9f8 | 276 | svsk->sk_sk, atomic_read(&svsk->sk_reserved)+serv->sv_max_mesg, |
1da177e4 | 277 | svc_sock_wspace(svsk)); |
3262c816 | 278 | svsk->sk_pool = NULL; |
c081a0c7 | 279 | clear_bit(SK_BUSY, &svsk->sk_flags); |
1da177e4 LT |
280 | goto out_unlock; |
281 | } | |
282 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
283 | ||
1da177e4 | 284 | |
3262c816 GB |
285 | if (!list_empty(&pool->sp_threads)) { |
286 | rqstp = list_entry(pool->sp_threads.next, | |
1da177e4 LT |
287 | struct svc_rqst, |
288 | rq_list); | |
289 | dprintk("svc: socket %p served by daemon %p\n", | |
290 | svsk->sk_sk, rqstp); | |
3262c816 | 291 | svc_thread_dequeue(pool, rqstp); |
1da177e4 | 292 | if (rqstp->rq_sock) |
cca5172a | 293 | printk(KERN_ERR |
1da177e4 LT |
294 | "svc_sock_enqueue: server %p, rq_sock=%p!\n", |
295 | rqstp, rqstp->rq_sock); | |
296 | rqstp->rq_sock = svsk; | |
c45c357d | 297 | atomic_inc(&svsk->sk_inuse); |
c6b0a9f8 | 298 | rqstp->rq_reserved = serv->sv_max_mesg; |
5685f0fa | 299 | atomic_add(rqstp->rq_reserved, &svsk->sk_reserved); |
3262c816 | 300 | BUG_ON(svsk->sk_pool != pool); |
1da177e4 LT |
301 | wake_up(&rqstp->rq_wait); |
302 | } else { | |
303 | dprintk("svc: socket %p put into queue\n", svsk->sk_sk); | |
3262c816 GB |
304 | list_add_tail(&svsk->sk_ready, &pool->sp_sockets); |
305 | BUG_ON(svsk->sk_pool != pool); | |
1da177e4 LT |
306 | } |
307 | ||
308 | out_unlock: | |
3262c816 | 309 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 LT |
310 | } |
311 | ||
312 | /* | |
3262c816 | 313 | * Dequeue the first socket. Must be called with the pool->sp_lock held. |
1da177e4 LT |
314 | */ |
315 | static inline struct svc_sock * | |
3262c816 | 316 | svc_sock_dequeue(struct svc_pool *pool) |
1da177e4 LT |
317 | { |
318 | struct svc_sock *svsk; | |
319 | ||
3262c816 | 320 | if (list_empty(&pool->sp_sockets)) |
1da177e4 LT |
321 | return NULL; |
322 | ||
3262c816 | 323 | svsk = list_entry(pool->sp_sockets.next, |
1da177e4 LT |
324 | struct svc_sock, sk_ready); |
325 | list_del_init(&svsk->sk_ready); | |
326 | ||
327 | dprintk("svc: socket %p dequeued, inuse=%d\n", | |
c45c357d | 328 | svsk->sk_sk, atomic_read(&svsk->sk_inuse)); |
1da177e4 LT |
329 | |
330 | return svsk; | |
331 | } | |
332 | ||
333 | /* | |
334 | * Having read something from a socket, check whether it | |
335 | * needs to be re-enqueued. | |
336 | * Note: SK_DATA only gets cleared when a read-attempt finds | |
337 | * no (or insufficient) data. | |
338 | */ | |
339 | static inline void | |
340 | svc_sock_received(struct svc_sock *svsk) | |
341 | { | |
3262c816 | 342 | svsk->sk_pool = NULL; |
1da177e4 LT |
343 | clear_bit(SK_BUSY, &svsk->sk_flags); |
344 | svc_sock_enqueue(svsk); | |
345 | } | |
346 | ||
347 | ||
348 | /** | |
349 | * svc_reserve - change the space reserved for the reply to a request. | |
350 | * @rqstp: The request in question | |
351 | * @space: new max space to reserve | |
352 | * | |
353 | * Each request reserves some space on the output queue of the socket | |
354 | * to make sure the reply fits. This function reduces that reserved | |
355 | * space to be the amount of space used already, plus @space. | |
356 | * | |
357 | */ | |
358 | void svc_reserve(struct svc_rqst *rqstp, int space) | |
359 | { | |
360 | space += rqstp->rq_res.head[0].iov_len; | |
361 | ||
362 | if (space < rqstp->rq_reserved) { | |
363 | struct svc_sock *svsk = rqstp->rq_sock; | |
5685f0fa | 364 | atomic_sub((rqstp->rq_reserved - space), &svsk->sk_reserved); |
1da177e4 | 365 | rqstp->rq_reserved = space; |
1da177e4 LT |
366 | |
367 | svc_sock_enqueue(svsk); | |
368 | } | |
369 | } | |
370 | ||
371 | /* | |
372 | * Release a socket after use. | |
373 | */ | |
374 | static inline void | |
375 | svc_sock_put(struct svc_sock *svsk) | |
376 | { | |
aaf68cfb N |
377 | if (atomic_dec_and_test(&svsk->sk_inuse)) { |
378 | BUG_ON(! test_bit(SK_DEAD, &svsk->sk_flags)); | |
379 | ||
202dd450 | 380 | dprintk("svc: releasing dead socket\n"); |
d6740df9 NB |
381 | if (svsk->sk_sock->file) |
382 | sockfd_put(svsk->sk_sock); | |
383 | else | |
384 | sock_release(svsk->sk_sock); | |
385 | if (svsk->sk_info_authunix != NULL) | |
386 | svcauth_unix_info_release(svsk->sk_info_authunix); | |
1da177e4 LT |
387 | kfree(svsk); |
388 | } | |
1da177e4 LT |
389 | } |
390 | ||
391 | static void | |
392 | svc_sock_release(struct svc_rqst *rqstp) | |
393 | { | |
394 | struct svc_sock *svsk = rqstp->rq_sock; | |
395 | ||
396 | svc_release_skb(rqstp); | |
397 | ||
44524359 | 398 | svc_free_res_pages(rqstp); |
1da177e4 LT |
399 | rqstp->rq_res.page_len = 0; |
400 | rqstp->rq_res.page_base = 0; | |
401 | ||
402 | ||
403 | /* Reset response buffer and release | |
404 | * the reservation. | |
405 | * But first, check that enough space was reserved | |
406 | * for the reply, otherwise we have a bug! | |
407 | */ | |
408 | if ((rqstp->rq_res.len) > rqstp->rq_reserved) | |
409 | printk(KERN_ERR "RPC request reserved %d but used %d\n", | |
410 | rqstp->rq_reserved, | |
411 | rqstp->rq_res.len); | |
412 | ||
413 | rqstp->rq_res.head[0].iov_len = 0; | |
414 | svc_reserve(rqstp, 0); | |
415 | rqstp->rq_sock = NULL; | |
416 | ||
417 | svc_sock_put(svsk); | |
418 | } | |
419 | ||
420 | /* | |
421 | * External function to wake up a server waiting for data | |
3262c816 GB |
422 | * This really only makes sense for services like lockd |
423 | * which have exactly one thread anyway. | |
1da177e4 LT |
424 | */ |
425 | void | |
426 | svc_wake_up(struct svc_serv *serv) | |
427 | { | |
428 | struct svc_rqst *rqstp; | |
3262c816 GB |
429 | unsigned int i; |
430 | struct svc_pool *pool; | |
431 | ||
432 | for (i = 0; i < serv->sv_nrpools; i++) { | |
433 | pool = &serv->sv_pools[i]; | |
434 | ||
435 | spin_lock_bh(&pool->sp_lock); | |
436 | if (!list_empty(&pool->sp_threads)) { | |
437 | rqstp = list_entry(pool->sp_threads.next, | |
438 | struct svc_rqst, | |
439 | rq_list); | |
440 | dprintk("svc: daemon %p woken up.\n", rqstp); | |
441 | /* | |
442 | svc_thread_dequeue(pool, rqstp); | |
443 | rqstp->rq_sock = NULL; | |
444 | */ | |
445 | wake_up(&rqstp->rq_wait); | |
446 | } | |
447 | spin_unlock_bh(&pool->sp_lock); | |
1da177e4 | 448 | } |
1da177e4 LT |
449 | } |
450 | ||
b92503b2 CL |
451 | union svc_pktinfo_u { |
452 | struct in_pktinfo pkti; | |
b92503b2 | 453 | struct in6_pktinfo pkti6; |
b92503b2 | 454 | }; |
bc375ea7 DM |
455 | #define SVC_PKTINFO_SPACE \ |
456 | CMSG_SPACE(sizeof(union svc_pktinfo_u)) | |
b92503b2 CL |
457 | |
458 | static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) | |
459 | { | |
460 | switch (rqstp->rq_sock->sk_sk->sk_family) { | |
461 | case AF_INET: { | |
462 | struct in_pktinfo *pki = CMSG_DATA(cmh); | |
463 | ||
464 | cmh->cmsg_level = SOL_IP; | |
465 | cmh->cmsg_type = IP_PKTINFO; | |
466 | pki->ipi_ifindex = 0; | |
467 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; | |
468 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
469 | } | |
470 | break; | |
5a05ed73 | 471 | |
b92503b2 CL |
472 | case AF_INET6: { |
473 | struct in6_pktinfo *pki = CMSG_DATA(cmh); | |
474 | ||
475 | cmh->cmsg_level = SOL_IPV6; | |
476 | cmh->cmsg_type = IPV6_PKTINFO; | |
477 | pki->ipi6_ifindex = 0; | |
478 | ipv6_addr_copy(&pki->ipi6_addr, | |
479 | &rqstp->rq_daddr.addr6); | |
480 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
481 | } | |
482 | break; | |
b92503b2 CL |
483 | } |
484 | return; | |
485 | } | |
486 | ||
1da177e4 LT |
487 | /* |
488 | * Generic sendto routine | |
489 | */ | |
490 | static int | |
491 | svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) | |
492 | { | |
493 | struct svc_sock *svsk = rqstp->rq_sock; | |
494 | struct socket *sock = svsk->sk_sock; | |
495 | int slen; | |
bc375ea7 DM |
496 | union { |
497 | struct cmsghdr hdr; | |
498 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
499 | } buffer; | |
500 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 LT |
501 | int len = 0; |
502 | int result; | |
503 | int size; | |
504 | struct page **ppage = xdr->pages; | |
505 | size_t base = xdr->page_base; | |
506 | unsigned int pglen = xdr->page_len; | |
507 | unsigned int flags = MSG_MORE; | |
ad06e4bd | 508 | char buf[RPC_MAX_ADDRBUFLEN]; |
1da177e4 LT |
509 | |
510 | slen = xdr->len; | |
511 | ||
512 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
b92503b2 CL |
513 | struct msghdr msg = { |
514 | .msg_name = &rqstp->rq_addr, | |
515 | .msg_namelen = rqstp->rq_addrlen, | |
516 | .msg_control = cmh, | |
517 | .msg_controllen = sizeof(buffer), | |
518 | .msg_flags = MSG_MORE, | |
519 | }; | |
520 | ||
521 | svc_set_cmsg_data(rqstp, cmh); | |
1da177e4 LT |
522 | |
523 | if (sock_sendmsg(sock, &msg, 0) < 0) | |
524 | goto out; | |
525 | } | |
526 | ||
527 | /* send head */ | |
528 | if (slen == xdr->head[0].iov_len) | |
529 | flags = 0; | |
44524359 N |
530 | len = kernel_sendpage(sock, rqstp->rq_respages[0], 0, |
531 | xdr->head[0].iov_len, flags); | |
1da177e4 LT |
532 | if (len != xdr->head[0].iov_len) |
533 | goto out; | |
534 | slen -= xdr->head[0].iov_len; | |
535 | if (slen == 0) | |
536 | goto out; | |
537 | ||
538 | /* send page data */ | |
539 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
540 | while (pglen > 0) { | |
541 | if (slen == size) | |
542 | flags = 0; | |
e6242e92 | 543 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
544 | if (result > 0) |
545 | len += result; | |
546 | if (result != size) | |
547 | goto out; | |
548 | slen -= size; | |
549 | pglen -= size; | |
550 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
551 | base = 0; | |
552 | ppage++; | |
553 | } | |
554 | /* send tail */ | |
555 | if (xdr->tail[0].iov_len) { | |
44524359 N |
556 | result = kernel_sendpage(sock, rqstp->rq_respages[0], |
557 | ((unsigned long)xdr->tail[0].iov_base) | |
cca5172a | 558 | & (PAGE_SIZE-1), |
1da177e4 LT |
559 | xdr->tail[0].iov_len, 0); |
560 | ||
561 | if (result > 0) | |
562 | len += result; | |
563 | } | |
564 | out: | |
ad06e4bd CL |
565 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", |
566 | rqstp->rq_sock, xdr->head[0].iov_base, xdr->head[0].iov_len, | |
567 | xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); | |
1da177e4 LT |
568 | |
569 | return len; | |
570 | } | |
571 | ||
80212d59 N |
572 | /* |
573 | * Report socket names for nfsdfs | |
574 | */ | |
575 | static int one_sock_name(char *buf, struct svc_sock *svsk) | |
576 | { | |
577 | int len; | |
578 | ||
579 | switch(svsk->sk_sk->sk_family) { | |
580 | case AF_INET: | |
581 | len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n", | |
582 | svsk->sk_sk->sk_protocol==IPPROTO_UDP? | |
583 | "udp" : "tcp", | |
584 | NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr), | |
585 | inet_sk(svsk->sk_sk)->num); | |
586 | break; | |
587 | default: | |
588 | len = sprintf(buf, "*unknown-%d*\n", | |
589 | svsk->sk_sk->sk_family); | |
590 | } | |
591 | return len; | |
592 | } | |
593 | ||
594 | int | |
b41b66d6 | 595 | svc_sock_names(char *buf, struct svc_serv *serv, char *toclose) |
80212d59 | 596 | { |
b41b66d6 | 597 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
598 | int len = 0; |
599 | ||
600 | if (!serv) | |
601 | return 0; | |
aaf68cfb | 602 | spin_lock_bh(&serv->sv_lock); |
80212d59 N |
603 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_list) { |
604 | int onelen = one_sock_name(buf+len, svsk); | |
b41b66d6 N |
605 | if (toclose && strcmp(toclose, buf+len) == 0) |
606 | closesk = svsk; | |
607 | else | |
608 | len += onelen; | |
80212d59 | 609 | } |
aaf68cfb | 610 | spin_unlock_bh(&serv->sv_lock); |
b41b66d6 | 611 | if (closesk) |
5680c446 N |
612 | /* Should unregister with portmap, but you cannot |
613 | * unregister just one protocol... | |
614 | */ | |
aaf68cfb | 615 | svc_close_socket(closesk); |
37a03472 N |
616 | else if (toclose) |
617 | return -ENOENT; | |
80212d59 N |
618 | return len; |
619 | } | |
620 | EXPORT_SYMBOL(svc_sock_names); | |
621 | ||
1da177e4 LT |
622 | /* |
623 | * Check input queue length | |
624 | */ | |
625 | static int | |
626 | svc_recv_available(struct svc_sock *svsk) | |
627 | { | |
1da177e4 LT |
628 | struct socket *sock = svsk->sk_sock; |
629 | int avail, err; | |
630 | ||
e6242e92 | 631 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
632 | |
633 | return (err >= 0)? avail : err; | |
634 | } | |
635 | ||
636 | /* | |
637 | * Generic recvfrom routine. | |
638 | */ | |
639 | static int | |
640 | svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, int buflen) | |
641 | { | |
067d7817 | 642 | struct svc_sock *svsk = rqstp->rq_sock; |
1ba95105 CL |
643 | struct msghdr msg = { |
644 | .msg_flags = MSG_DONTWAIT, | |
645 | }; | |
646 | int len; | |
1da177e4 | 647 | |
1ba95105 CL |
648 | len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, |
649 | msg.msg_flags); | |
1da177e4 LT |
650 | |
651 | /* sock_recvmsg doesn't fill in the name/namelen, so we must.. | |
1da177e4 | 652 | */ |
067d7817 CL |
653 | memcpy(&rqstp->rq_addr, &svsk->sk_remote, svsk->sk_remotelen); |
654 | rqstp->rq_addrlen = svsk->sk_remotelen; | |
1da177e4 LT |
655 | |
656 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", | |
1ba95105 | 657 | svsk, iov[0].iov_base, iov[0].iov_len, len); |
1da177e4 LT |
658 | |
659 | return len; | |
660 | } | |
661 | ||
662 | /* | |
663 | * Set socket snd and rcv buffer lengths | |
664 | */ | |
665 | static inline void | |
666 | svc_sock_setbufsize(struct socket *sock, unsigned int snd, unsigned int rcv) | |
667 | { | |
668 | #if 0 | |
669 | mm_segment_t oldfs; | |
670 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
671 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
672 | (char*)&snd, sizeof(snd)); | |
673 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
674 | (char*)&rcv, sizeof(rcv)); | |
675 | #else | |
676 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
677 | * which isn't acceptable. Until that is made conditional | |
678 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
679 | * DaveM said I could! | |
680 | */ | |
681 | lock_sock(sock->sk); | |
682 | sock->sk->sk_sndbuf = snd * 2; | |
683 | sock->sk->sk_rcvbuf = rcv * 2; | |
684 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; | |
685 | release_sock(sock->sk); | |
686 | #endif | |
687 | } | |
688 | /* | |
689 | * INET callback when data has been received on the socket. | |
690 | */ | |
691 | static void | |
692 | svc_udp_data_ready(struct sock *sk, int count) | |
693 | { | |
939bb7ef | 694 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 695 | |
939bb7ef NB |
696 | if (svsk) { |
697 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
698 | svsk, sk, count, test_bit(SK_BUSY, &svsk->sk_flags)); | |
699 | set_bit(SK_DATA, &svsk->sk_flags); | |
700 | svc_sock_enqueue(svsk); | |
701 | } | |
1da177e4 LT |
702 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
703 | wake_up_interruptible(sk->sk_sleep); | |
704 | } | |
705 | ||
706 | /* | |
707 | * INET callback when space is newly available on the socket. | |
708 | */ | |
709 | static void | |
710 | svc_write_space(struct sock *sk) | |
711 | { | |
712 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
713 | ||
714 | if (svsk) { | |
715 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
716 | svsk, sk, test_bit(SK_BUSY, &svsk->sk_flags)); | |
717 | svc_sock_enqueue(svsk); | |
718 | } | |
719 | ||
720 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { | |
939bb7ef | 721 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 LT |
722 | svsk); |
723 | wake_up_interruptible(sk->sk_sleep); | |
724 | } | |
725 | } | |
726 | ||
7a37f578 N |
727 | static inline void svc_udp_get_dest_address(struct svc_rqst *rqstp, |
728 | struct cmsghdr *cmh) | |
95756482 CL |
729 | { |
730 | switch (rqstp->rq_sock->sk_sk->sk_family) { | |
731 | case AF_INET: { | |
7a37f578 N |
732 | struct in_pktinfo *pki = CMSG_DATA(cmh); |
733 | rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; | |
95756482 | 734 | break; |
7a37f578 | 735 | } |
95756482 | 736 | case AF_INET6: { |
7a37f578 N |
737 | struct in6_pktinfo *pki = CMSG_DATA(cmh); |
738 | ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); | |
95756482 | 739 | break; |
7a37f578 | 740 | } |
95756482 | 741 | } |
95756482 CL |
742 | } |
743 | ||
1da177e4 LT |
744 | /* |
745 | * Receive a datagram from a UDP socket. | |
746 | */ | |
1da177e4 LT |
747 | static int |
748 | svc_udp_recvfrom(struct svc_rqst *rqstp) | |
749 | { | |
750 | struct svc_sock *svsk = rqstp->rq_sock; | |
751 | struct svc_serv *serv = svsk->sk_server; | |
752 | struct sk_buff *skb; | |
bc375ea7 DM |
753 | union { |
754 | struct cmsghdr hdr; | |
755 | long all[SVC_PKTINFO_SPACE / sizeof(long)]; | |
756 | } buffer; | |
757 | struct cmsghdr *cmh = &buffer.hdr; | |
1da177e4 | 758 | int err, len; |
7a37f578 N |
759 | struct msghdr msg = { |
760 | .msg_name = svc_addr(rqstp), | |
761 | .msg_control = cmh, | |
762 | .msg_controllen = sizeof(buffer), | |
763 | .msg_flags = MSG_DONTWAIT, | |
764 | }; | |
1da177e4 LT |
765 | |
766 | if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags)) | |
767 | /* udp sockets need large rcvbuf as all pending | |
768 | * requests are still in that buffer. sndbuf must | |
769 | * also be large enough that there is enough space | |
3262c816 GB |
770 | * for one reply per thread. We count all threads |
771 | * rather than threads in a particular pool, which | |
772 | * provides an upper bound on the number of threads | |
773 | * which will access the socket. | |
1da177e4 LT |
774 | */ |
775 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
776 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
777 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 LT |
778 | |
779 | if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) { | |
780 | svc_sock_received(svsk); | |
781 | return svc_deferred_recv(rqstp); | |
782 | } | |
783 | ||
aaf68cfb N |
784 | if (test_bit(SK_CLOSE, &svsk->sk_flags)) { |
785 | svc_delete_socket(svsk); | |
786 | return 0; | |
787 | } | |
788 | ||
1da177e4 | 789 | clear_bit(SK_DATA, &svsk->sk_flags); |
418106d6 AB |
790 | while ((err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, |
791 | 0, 0, MSG_PEEK | MSG_DONTWAIT)) < 0 || | |
7a37f578 | 792 | (skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err)) == NULL) { |
1da177e4 LT |
793 | if (err == -EAGAIN) { |
794 | svc_sock_received(svsk); | |
795 | return err; | |
796 | } | |
797 | /* possibly an icmp error */ | |
798 | dprintk("svc: recvfrom returned error %d\n", -err); | |
799 | } | |
7a37f578 | 800 | rqstp->rq_addrlen = sizeof(rqstp->rq_addr); |
a61bbcf2 PM |
801 | if (skb->tstamp.off_sec == 0) { |
802 | struct timeval tv; | |
803 | ||
804 | tv.tv_sec = xtime.tv_sec; | |
4bcde03d | 805 | tv.tv_usec = xtime.tv_nsec / NSEC_PER_USEC; |
a61bbcf2 | 806 | skb_set_timestamp(skb, &tv); |
cca5172a | 807 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
808 | need that much accuracy */ |
809 | } | |
a61bbcf2 | 810 | skb_get_timestamp(skb, &svsk->sk_sk->sk_stamp); |
1da177e4 LT |
811 | set_bit(SK_DATA, &svsk->sk_flags); /* there may be more data... */ |
812 | ||
813 | /* | |
814 | * Maybe more packets - kick another thread ASAP. | |
815 | */ | |
816 | svc_sock_received(svsk); | |
817 | ||
818 | len = skb->len - sizeof(struct udphdr); | |
819 | rqstp->rq_arg.len = len; | |
820 | ||
95756482 | 821 | rqstp->rq_prot = IPPROTO_UDP; |
27459f09 | 822 | |
7a37f578 N |
823 | if (cmh->cmsg_level != IPPROTO_IP || |
824 | cmh->cmsg_type != IP_PKTINFO) { | |
825 | if (net_ratelimit()) | |
826 | printk("rpcsvc: received unknown control message:" | |
827 | "%d/%d\n", | |
828 | cmh->cmsg_level, cmh->cmsg_type); | |
829 | skb_free_datagram(svsk->sk_sk, skb); | |
830 | return 0; | |
831 | } | |
832 | svc_udp_get_dest_address(rqstp, cmh); | |
1da177e4 LT |
833 | |
834 | if (skb_is_nonlinear(skb)) { | |
835 | /* we have to copy */ | |
836 | local_bh_disable(); | |
837 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
838 | local_bh_enable(); | |
839 | /* checksum error */ | |
840 | skb_free_datagram(svsk->sk_sk, skb); | |
841 | return 0; | |
842 | } | |
843 | local_bh_enable(); | |
cca5172a | 844 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 LT |
845 | } else { |
846 | /* we can use it in-place */ | |
847 | rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr); | |
848 | rqstp->rq_arg.head[0].iov_len = len; | |
fb286bb2 HX |
849 | if (skb_checksum_complete(skb)) { |
850 | skb_free_datagram(svsk->sk_sk, skb); | |
851 | return 0; | |
1da177e4 LT |
852 | } |
853 | rqstp->rq_skbuff = skb; | |
854 | } | |
855 | ||
856 | rqstp->rq_arg.page_base = 0; | |
857 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
858 | rqstp->rq_arg.head[0].iov_len = len; | |
859 | rqstp->rq_arg.page_len = 0; | |
44524359 | 860 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
861 | } else { |
862 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 N |
863 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
864 | (rqstp->rq_arg.page_len + PAGE_SIZE - 1)/ PAGE_SIZE; | |
1da177e4 LT |
865 | } |
866 | ||
867 | if (serv->sv_stats) | |
868 | serv->sv_stats->netudpcnt++; | |
869 | ||
870 | return len; | |
871 | } | |
872 | ||
873 | static int | |
874 | svc_udp_sendto(struct svc_rqst *rqstp) | |
875 | { | |
876 | int error; | |
877 | ||
878 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
879 | if (error == -ECONNREFUSED) | |
880 | /* ICMP error on earlier request. */ | |
881 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
882 | ||
883 | return error; | |
884 | } | |
885 | ||
886 | static void | |
887 | svc_udp_init(struct svc_sock *svsk) | |
888 | { | |
7a37f578 N |
889 | int one = 1; |
890 | mm_segment_t oldfs; | |
891 | ||
1da177e4 LT |
892 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; |
893 | svsk->sk_sk->sk_write_space = svc_write_space; | |
894 | svsk->sk_recvfrom = svc_udp_recvfrom; | |
895 | svsk->sk_sendto = svc_udp_sendto; | |
896 | ||
897 | /* initialise setting must have enough space to | |
cca5172a | 898 | * receive and respond to one request. |
1da177e4 LT |
899 | * svc_udp_recvfrom will re-adjust if necessary |
900 | */ | |
901 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
902 | 3 * svsk->sk_server->sv_max_mesg, |
903 | 3 * svsk->sk_server->sv_max_mesg); | |
1da177e4 LT |
904 | |
905 | set_bit(SK_DATA, &svsk->sk_flags); /* might have come in before data_ready set up */ | |
906 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
7a37f578 N |
907 | |
908 | oldfs = get_fs(); | |
909 | set_fs(KERNEL_DS); | |
910 | /* make sure we get destination address info */ | |
911 | svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO, | |
912 | (char __user *)&one, sizeof(one)); | |
913 | set_fs(oldfs); | |
1da177e4 LT |
914 | } |
915 | ||
916 | /* | |
917 | * A data_ready event on a listening socket means there's a connection | |
918 | * pending. Do not use state_change as a substitute for it. | |
919 | */ | |
920 | static void | |
921 | svc_tcp_listen_data_ready(struct sock *sk, int count_unused) | |
922 | { | |
939bb7ef | 923 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
924 | |
925 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 926 | sk, sk->sk_state); |
1da177e4 | 927 | |
939bb7ef NB |
928 | /* |
929 | * This callback may called twice when a new connection | |
930 | * is established as a child socket inherits everything | |
931 | * from a parent LISTEN socket. | |
932 | * 1) data_ready method of the parent socket will be called | |
933 | * when one of child sockets become ESTABLISHED. | |
934 | * 2) data_ready method of the child socket may be called | |
935 | * when it receives data before the socket is accepted. | |
936 | * In case of 2, we should ignore it silently. | |
937 | */ | |
938 | if (sk->sk_state == TCP_LISTEN) { | |
939 | if (svsk) { | |
940 | set_bit(SK_CONN, &svsk->sk_flags); | |
941 | svc_sock_enqueue(svsk); | |
942 | } else | |
943 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 944 | } |
939bb7ef | 945 | |
1da177e4 LT |
946 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
947 | wake_up_interruptible_all(sk->sk_sleep); | |
948 | } | |
949 | ||
950 | /* | |
951 | * A state change on a connected socket means it's dying or dead. | |
952 | */ | |
953 | static void | |
954 | svc_tcp_state_change(struct sock *sk) | |
955 | { | |
939bb7ef | 956 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
957 | |
958 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 959 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 960 | |
939bb7ef | 961 | if (!svsk) |
1da177e4 | 962 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef NB |
963 | else { |
964 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
965 | svc_sock_enqueue(svsk); | |
1da177e4 | 966 | } |
1da177e4 LT |
967 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
968 | wake_up_interruptible_all(sk->sk_sleep); | |
969 | } | |
970 | ||
971 | static void | |
972 | svc_tcp_data_ready(struct sock *sk, int count) | |
973 | { | |
939bb7ef | 974 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
975 | |
976 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
977 | sk, sk->sk_user_data); |
978 | if (svsk) { | |
979 | set_bit(SK_DATA, &svsk->sk_flags); | |
980 | svc_sock_enqueue(svsk); | |
981 | } | |
1da177e4 LT |
982 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
983 | wake_up_interruptible(sk->sk_sleep); | |
984 | } | |
985 | ||
bcdb81ae CL |
986 | static inline int svc_port_is_privileged(struct sockaddr *sin) |
987 | { | |
988 | switch (sin->sa_family) { | |
989 | case AF_INET: | |
990 | return ntohs(((struct sockaddr_in *)sin)->sin_port) | |
991 | < PROT_SOCK; | |
bcdb81ae CL |
992 | case AF_INET6: |
993 | return ntohs(((struct sockaddr_in6 *)sin)->sin6_port) | |
994 | < PROT_SOCK; | |
bcdb81ae CL |
995 | default: |
996 | return 0; | |
997 | } | |
998 | } | |
999 | ||
1da177e4 LT |
1000 | /* |
1001 | * Accept a TCP connection | |
1002 | */ | |
1003 | static void | |
1004 | svc_tcp_accept(struct svc_sock *svsk) | |
1005 | { | |
cdd88b9f AM |
1006 | struct sockaddr_storage addr; |
1007 | struct sockaddr *sin = (struct sockaddr *) &addr; | |
1da177e4 LT |
1008 | struct svc_serv *serv = svsk->sk_server; |
1009 | struct socket *sock = svsk->sk_sock; | |
1010 | struct socket *newsock; | |
1da177e4 LT |
1011 | struct svc_sock *newsvsk; |
1012 | int err, slen; | |
ad06e4bd | 1013 | char buf[RPC_MAX_ADDRBUFLEN]; |
1da177e4 LT |
1014 | |
1015 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
1016 | if (!sock) | |
1017 | return; | |
1018 | ||
e6242e92 SS |
1019 | clear_bit(SK_CONN, &svsk->sk_flags); |
1020 | err = kernel_accept(sock, &newsock, O_NONBLOCK); | |
1021 | if (err < 0) { | |
1da177e4 LT |
1022 | if (err == -ENOMEM) |
1023 | printk(KERN_WARNING "%s: no more sockets!\n", | |
1024 | serv->sv_name); | |
e6242e92 | 1025 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
1026 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
1027 | serv->sv_name, -err); | |
e6242e92 | 1028 | return; |
1da177e4 | 1029 | } |
e6242e92 | 1030 | |
1da177e4 LT |
1031 | set_bit(SK_CONN, &svsk->sk_flags); |
1032 | svc_sock_enqueue(svsk); | |
1033 | ||
cdd88b9f | 1034 | err = kernel_getpeername(newsock, sin, &slen); |
1da177e4 LT |
1035 | if (err < 0) { |
1036 | if (net_ratelimit()) | |
1037 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
1038 | serv->sv_name, -err); | |
1039 | goto failed; /* aborted connection or whatever */ | |
1040 | } | |
1041 | ||
1042 | /* Ideally, we would want to reject connections from unauthorized | |
ad06e4bd CL |
1043 | * hosts here, but when we get encryption, the IP of the host won't |
1044 | * tell us anything. For now just warn about unpriv connections. | |
1da177e4 | 1045 | */ |
cdd88b9f | 1046 | if (!svc_port_is_privileged(sin)) { |
1da177e4 | 1047 | dprintk(KERN_WARNING |
ad06e4bd | 1048 | "%s: connect from unprivileged port: %s\n", |
cca5172a | 1049 | serv->sv_name, |
cdd88b9f | 1050 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 | 1051 | } |
ad06e4bd | 1052 | dprintk("%s: connect from %s\n", serv->sv_name, |
cdd88b9f | 1053 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1054 | |
1055 | /* make sure that a write doesn't block forever when | |
1056 | * low on memory | |
1057 | */ | |
1058 | newsock->sk->sk_sndtimeo = HZ*30; | |
1059 | ||
6b174337 CL |
1060 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
1061 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) | |
1da177e4 | 1062 | goto failed; |
cdd88b9f | 1063 | memcpy(&newsvsk->sk_remote, sin, slen); |
067d7817 CL |
1064 | newsvsk->sk_remotelen = slen; |
1065 | ||
e79eff1f | 1066 | svc_sock_received(newsvsk); |
1da177e4 LT |
1067 | |
1068 | /* make sure that we don't have too many active connections. | |
1069 | * If we have, something must be dropped. | |
1070 | * | |
1071 | * There's no point in trying to do random drop here for | |
1072 | * DoS prevention. The NFS clients does 1 reconnect in 15 | |
1073 | * seconds. An attacker can easily beat that. | |
1074 | * | |
1075 | * The only somewhat efficient mechanism would be if drop | |
1076 | * old connections from the same IP first. But right now | |
1077 | * we don't even record the client IP in svc_sock. | |
1078 | */ | |
1079 | if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) { | |
1080 | struct svc_sock *svsk = NULL; | |
1081 | spin_lock_bh(&serv->sv_lock); | |
1082 | if (!list_empty(&serv->sv_tempsocks)) { | |
1083 | if (net_ratelimit()) { | |
1084 | /* Try to help the admin */ | |
1085 | printk(KERN_NOTICE "%s: too many open TCP " | |
1086 | "sockets, consider increasing the " | |
1087 | "number of nfsd threads\n", | |
1088 | serv->sv_name); | |
ad06e4bd CL |
1089 | printk(KERN_NOTICE |
1090 | "%s: last TCP connect from %s\n", | |
1091 | serv->sv_name, buf); | |
1da177e4 LT |
1092 | } |
1093 | /* | |
1094 | * Always select the oldest socket. It's not fair, | |
1095 | * but so is life | |
1096 | */ | |
1097 | svsk = list_entry(serv->sv_tempsocks.prev, | |
1098 | struct svc_sock, | |
1099 | sk_list); | |
1100 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
c45c357d | 1101 | atomic_inc(&svsk->sk_inuse); |
1da177e4 LT |
1102 | } |
1103 | spin_unlock_bh(&serv->sv_lock); | |
1104 | ||
1105 | if (svsk) { | |
1106 | svc_sock_enqueue(svsk); | |
1107 | svc_sock_put(svsk); | |
1108 | } | |
1109 | ||
1110 | } | |
1111 | ||
1112 | if (serv->sv_stats) | |
1113 | serv->sv_stats->nettcpconn++; | |
1114 | ||
1115 | return; | |
1116 | ||
1117 | failed: | |
1118 | sock_release(newsock); | |
1119 | return; | |
1120 | } | |
1121 | ||
1122 | /* | |
1123 | * Receive data from a TCP socket. | |
1124 | */ | |
1125 | static int | |
1126 | svc_tcp_recvfrom(struct svc_rqst *rqstp) | |
1127 | { | |
1128 | struct svc_sock *svsk = rqstp->rq_sock; | |
1129 | struct svc_serv *serv = svsk->sk_server; | |
1130 | int len; | |
3cc03b16 | 1131 | struct kvec *vec; |
1da177e4 LT |
1132 | int pnum, vlen; |
1133 | ||
1134 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
1135 | svsk, test_bit(SK_DATA, &svsk->sk_flags), | |
1136 | test_bit(SK_CONN, &svsk->sk_flags), | |
1137 | test_bit(SK_CLOSE, &svsk->sk_flags)); | |
1138 | ||
1139 | if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) { | |
1140 | svc_sock_received(svsk); | |
1141 | return svc_deferred_recv(rqstp); | |
1142 | } | |
1143 | ||
1144 | if (test_bit(SK_CLOSE, &svsk->sk_flags)) { | |
1145 | svc_delete_socket(svsk); | |
1146 | return 0; | |
1147 | } | |
1148 | ||
1a047060 | 1149 | if (svsk->sk_sk->sk_state == TCP_LISTEN) { |
1da177e4 LT |
1150 | svc_tcp_accept(svsk); |
1151 | svc_sock_received(svsk); | |
1152 | return 0; | |
1153 | } | |
1154 | ||
1155 | if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags)) | |
1156 | /* sndbuf needs to have room for one request | |
1157 | * per thread, otherwise we can stall even when the | |
1158 | * network isn't a bottleneck. | |
3262c816 GB |
1159 | * |
1160 | * We count all threads rather than threads in a | |
1161 | * particular pool, which provides an upper bound | |
1162 | * on the number of threads which will access the socket. | |
1163 | * | |
1da177e4 | 1164 | * rcvbuf just needs to be able to hold a few requests. |
cca5172a | 1165 | * Normally they will be removed from the queue |
1da177e4 LT |
1166 | * as soon a a complete request arrives. |
1167 | */ | |
1168 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
1169 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
1170 | 3 * serv->sv_max_mesg); | |
1da177e4 LT |
1171 | |
1172 | clear_bit(SK_DATA, &svsk->sk_flags); | |
1173 | ||
1174 | /* Receive data. If we haven't got the record length yet, get | |
1175 | * the next four bytes. Otherwise try to gobble up as much as | |
1176 | * possible up to the complete record length. | |
1177 | */ | |
1178 | if (svsk->sk_tcplen < 4) { | |
1179 | unsigned long want = 4 - svsk->sk_tcplen; | |
1180 | struct kvec iov; | |
1181 | ||
1182 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; | |
1183 | iov.iov_len = want; | |
1184 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
1185 | goto error; | |
1186 | svsk->sk_tcplen += len; | |
1187 | ||
1188 | if (len < want) { | |
1189 | dprintk("svc: short recvfrom while reading record length (%d of %lu)\n", | |
cca5172a | 1190 | len, want); |
1da177e4 LT |
1191 | svc_sock_received(svsk); |
1192 | return -EAGAIN; /* record header not complete */ | |
1193 | } | |
1194 | ||
1195 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
1196 | if (!(svsk->sk_reclen & 0x80000000)) { | |
1197 | /* FIXME: technically, a record can be fragmented, | |
1198 | * and non-terminal fragments will not have the top | |
1199 | * bit set in the fragment length header. | |
1200 | * But apparently no known nfs clients send fragmented | |
1201 | * records. */ | |
34e9a63b N |
1202 | if (net_ratelimit()) |
1203 | printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx" | |
1204 | " (non-terminal)\n", | |
1205 | (unsigned long) svsk->sk_reclen); | |
1da177e4 LT |
1206 | goto err_delete; |
1207 | } | |
1208 | svsk->sk_reclen &= 0x7fffffff; | |
1209 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); | |
c6b0a9f8 | 1210 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
34e9a63b N |
1211 | if (net_ratelimit()) |
1212 | printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx" | |
1213 | " (large)\n", | |
1214 | (unsigned long) svsk->sk_reclen); | |
1da177e4 LT |
1215 | goto err_delete; |
1216 | } | |
1217 | } | |
1218 | ||
1219 | /* Check whether enough data is available */ | |
1220 | len = svc_recv_available(svsk); | |
1221 | if (len < 0) | |
1222 | goto error; | |
1223 | ||
1224 | if (len < svsk->sk_reclen) { | |
1225 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
1226 | len, svsk->sk_reclen); | |
1227 | svc_sock_received(svsk); | |
1228 | return -EAGAIN; /* record not complete */ | |
1229 | } | |
1230 | len = svsk->sk_reclen; | |
1231 | set_bit(SK_DATA, &svsk->sk_flags); | |
1232 | ||
3cc03b16 | 1233 | vec = rqstp->rq_vec; |
1da177e4 LT |
1234 | vec[0] = rqstp->rq_arg.head[0]; |
1235 | vlen = PAGE_SIZE; | |
1236 | pnum = 1; | |
1237 | while (vlen < len) { | |
44524359 | 1238 | vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]); |
1da177e4 LT |
1239 | vec[pnum].iov_len = PAGE_SIZE; |
1240 | pnum++; | |
1241 | vlen += PAGE_SIZE; | |
1242 | } | |
44524359 | 1243 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
1244 | |
1245 | /* Now receive data */ | |
1246 | len = svc_recvfrom(rqstp, vec, pnum, len); | |
1247 | if (len < 0) | |
1248 | goto error; | |
1249 | ||
1250 | dprintk("svc: TCP complete record (%d bytes)\n", len); | |
1251 | rqstp->rq_arg.len = len; | |
1252 | rqstp->rq_arg.page_base = 0; | |
1253 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
1254 | rqstp->rq_arg.head[0].iov_len = len; | |
1255 | rqstp->rq_arg.page_len = 0; | |
1256 | } else { | |
1257 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
1258 | } | |
1259 | ||
1260 | rqstp->rq_skbuff = NULL; | |
1261 | rqstp->rq_prot = IPPROTO_TCP; | |
1262 | ||
1263 | /* Reset TCP read info */ | |
1264 | svsk->sk_reclen = 0; | |
1265 | svsk->sk_tcplen = 0; | |
1266 | ||
1267 | svc_sock_received(svsk); | |
1268 | if (serv->sv_stats) | |
1269 | serv->sv_stats->nettcpcnt++; | |
1270 | ||
1271 | return len; | |
1272 | ||
1273 | err_delete: | |
1274 | svc_delete_socket(svsk); | |
1275 | return -EAGAIN; | |
1276 | ||
1277 | error: | |
1278 | if (len == -EAGAIN) { | |
1279 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
1280 | svc_sock_received(svsk); | |
1281 | } else { | |
1282 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", | |
1283 | svsk->sk_server->sv_name, -len); | |
93fbf1a5 | 1284 | goto err_delete; |
1da177e4 LT |
1285 | } |
1286 | ||
1287 | return len; | |
1288 | } | |
1289 | ||
1290 | /* | |
1291 | * Send out data on TCP socket. | |
1292 | */ | |
1293 | static int | |
1294 | svc_tcp_sendto(struct svc_rqst *rqstp) | |
1295 | { | |
1296 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
1297 | int sent; | |
d8ed029d | 1298 | __be32 reclen; |
1da177e4 LT |
1299 | |
1300 | /* Set up the first element of the reply kvec. | |
1301 | * Any other kvecs that may be in use have been taken | |
1302 | * care of by the server implementation itself. | |
1303 | */ | |
1304 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
1305 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
1306 | ||
1307 | if (test_bit(SK_DEAD, &rqstp->rq_sock->sk_flags)) | |
1308 | return -ENOTCONN; | |
1309 | ||
1310 | sent = svc_sendto(rqstp, &rqstp->rq_res); | |
1311 | if (sent != xbufp->len) { | |
1312 | printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n", | |
1313 | rqstp->rq_sock->sk_server->sv_name, | |
1314 | (sent<0)?"got error":"sent only", | |
1315 | sent, xbufp->len); | |
aaf68cfb N |
1316 | set_bit(SK_CLOSE, &rqstp->rq_sock->sk_flags); |
1317 | svc_sock_enqueue(rqstp->rq_sock); | |
1da177e4 LT |
1318 | sent = -EAGAIN; |
1319 | } | |
1320 | return sent; | |
1321 | } | |
1322 | ||
1323 | static void | |
1324 | svc_tcp_init(struct svc_sock *svsk) | |
1325 | { | |
1326 | struct sock *sk = svsk->sk_sk; | |
1327 | struct tcp_sock *tp = tcp_sk(sk); | |
1328 | ||
1329 | svsk->sk_recvfrom = svc_tcp_recvfrom; | |
1330 | svsk->sk_sendto = svc_tcp_sendto; | |
1331 | ||
1332 | if (sk->sk_state == TCP_LISTEN) { | |
1333 | dprintk("setting up TCP socket for listening\n"); | |
1334 | sk->sk_data_ready = svc_tcp_listen_data_ready; | |
1335 | set_bit(SK_CONN, &svsk->sk_flags); | |
1336 | } else { | |
1337 | dprintk("setting up TCP socket for reading\n"); | |
1338 | sk->sk_state_change = svc_tcp_state_change; | |
1339 | sk->sk_data_ready = svc_tcp_data_ready; | |
1340 | sk->sk_write_space = svc_write_space; | |
1341 | ||
1342 | svsk->sk_reclen = 0; | |
1343 | svsk->sk_tcplen = 0; | |
1344 | ||
1345 | tp->nonagle = 1; /* disable Nagle's algorithm */ | |
1346 | ||
1347 | /* initialise setting must have enough space to | |
cca5172a | 1348 | * receive and respond to one request. |
1da177e4 LT |
1349 | * svc_tcp_recvfrom will re-adjust if necessary |
1350 | */ | |
1351 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
1352 | 3 * svsk->sk_server->sv_max_mesg, |
1353 | 3 * svsk->sk_server->sv_max_mesg); | |
1da177e4 LT |
1354 | |
1355 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1356 | set_bit(SK_DATA, &svsk->sk_flags); | |
cca5172a | 1357 | if (sk->sk_state != TCP_ESTABLISHED) |
1da177e4 LT |
1358 | set_bit(SK_CLOSE, &svsk->sk_flags); |
1359 | } | |
1360 | } | |
1361 | ||
1362 | void | |
1363 | svc_sock_update_bufs(struct svc_serv *serv) | |
1364 | { | |
1365 | /* | |
1366 | * The number of server threads has changed. Update | |
1367 | * rcvbuf and sndbuf accordingly on all sockets | |
1368 | */ | |
1369 | struct list_head *le; | |
1370 | ||
1371 | spin_lock_bh(&serv->sv_lock); | |
1372 | list_for_each(le, &serv->sv_permsocks) { | |
cca5172a | 1373 | struct svc_sock *svsk = |
1da177e4 LT |
1374 | list_entry(le, struct svc_sock, sk_list); |
1375 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1376 | } | |
1377 | list_for_each(le, &serv->sv_tempsocks) { | |
1378 | struct svc_sock *svsk = | |
1379 | list_entry(le, struct svc_sock, sk_list); | |
1380 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1381 | } | |
1382 | spin_unlock_bh(&serv->sv_lock); | |
1383 | } | |
1384 | ||
1385 | /* | |
3262c816 GB |
1386 | * Receive the next request on any socket. This code is carefully |
1387 | * organised not to touch any cachelines in the shared svc_serv | |
1388 | * structure, only cachelines in the local svc_pool. | |
1da177e4 LT |
1389 | */ |
1390 | int | |
6fb2b47f | 1391 | svc_recv(struct svc_rqst *rqstp, long timeout) |
1da177e4 | 1392 | { |
27459f09 | 1393 | struct svc_sock *svsk = NULL; |
6fb2b47f | 1394 | struct svc_serv *serv = rqstp->rq_server; |
3262c816 | 1395 | struct svc_pool *pool = rqstp->rq_pool; |
44524359 | 1396 | int len, i; |
1da177e4 LT |
1397 | int pages; |
1398 | struct xdr_buf *arg; | |
1399 | DECLARE_WAITQUEUE(wait, current); | |
1400 | ||
1401 | dprintk("svc: server %p waiting for data (to = %ld)\n", | |
1402 | rqstp, timeout); | |
1403 | ||
1404 | if (rqstp->rq_sock) | |
cca5172a | 1405 | printk(KERN_ERR |
1da177e4 LT |
1406 | "svc_recv: service %p, socket not NULL!\n", |
1407 | rqstp); | |
1408 | if (waitqueue_active(&rqstp->rq_wait)) | |
cca5172a | 1409 | printk(KERN_ERR |
1da177e4 LT |
1410 | "svc_recv: service %p, wait queue active!\n", |
1411 | rqstp); | |
1412 | ||
1da177e4 LT |
1413 | |
1414 | /* now allocate needed pages. If we get a failure, sleep briefly */ | |
c6b0a9f8 | 1415 | pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE; |
44524359 N |
1416 | for (i=0; i < pages ; i++) |
1417 | while (rqstp->rq_pages[i] == NULL) { | |
1418 | struct page *p = alloc_page(GFP_KERNEL); | |
1419 | if (!p) | |
1420 | schedule_timeout_uninterruptible(msecs_to_jiffies(500)); | |
1421 | rqstp->rq_pages[i] = p; | |
1da177e4 | 1422 | } |
250f3915 N |
1423 | rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */ |
1424 | BUG_ON(pages >= RPCSVC_MAXPAGES); | |
1da177e4 LT |
1425 | |
1426 | /* Make arg->head point to first page and arg->pages point to rest */ | |
1427 | arg = &rqstp->rq_arg; | |
44524359 | 1428 | arg->head[0].iov_base = page_address(rqstp->rq_pages[0]); |
1da177e4 | 1429 | arg->head[0].iov_len = PAGE_SIZE; |
44524359 | 1430 | arg->pages = rqstp->rq_pages + 1; |
1da177e4 LT |
1431 | arg->page_base = 0; |
1432 | /* save at least one page for response */ | |
1433 | arg->page_len = (pages-2)*PAGE_SIZE; | |
1434 | arg->len = (pages-1)*PAGE_SIZE; | |
1435 | arg->tail[0].iov_len = 0; | |
3e1d1d28 CL |
1436 | |
1437 | try_to_freeze(); | |
1887b935 | 1438 | cond_resched(); |
1da177e4 LT |
1439 | if (signalled()) |
1440 | return -EINTR; | |
1441 | ||
3262c816 GB |
1442 | spin_lock_bh(&pool->sp_lock); |
1443 | if ((svsk = svc_sock_dequeue(pool)) != NULL) { | |
1da177e4 | 1444 | rqstp->rq_sock = svsk; |
c45c357d | 1445 | atomic_inc(&svsk->sk_inuse); |
c6b0a9f8 | 1446 | rqstp->rq_reserved = serv->sv_max_mesg; |
5685f0fa | 1447 | atomic_add(rqstp->rq_reserved, &svsk->sk_reserved); |
1da177e4 LT |
1448 | } else { |
1449 | /* No data pending. Go to sleep */ | |
3262c816 | 1450 | svc_thread_enqueue(pool, rqstp); |
1da177e4 LT |
1451 | |
1452 | /* | |
1453 | * We have to be able to interrupt this wait | |
1454 | * to bring down the daemons ... | |
1455 | */ | |
1456 | set_current_state(TASK_INTERRUPTIBLE); | |
1457 | add_wait_queue(&rqstp->rq_wait, &wait); | |
3262c816 | 1458 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 LT |
1459 | |
1460 | schedule_timeout(timeout); | |
1461 | ||
3e1d1d28 | 1462 | try_to_freeze(); |
1da177e4 | 1463 | |
3262c816 | 1464 | spin_lock_bh(&pool->sp_lock); |
1da177e4 LT |
1465 | remove_wait_queue(&rqstp->rq_wait, &wait); |
1466 | ||
1467 | if (!(svsk = rqstp->rq_sock)) { | |
3262c816 GB |
1468 | svc_thread_dequeue(pool, rqstp); |
1469 | spin_unlock_bh(&pool->sp_lock); | |
1da177e4 LT |
1470 | dprintk("svc: server %p, no data yet\n", rqstp); |
1471 | return signalled()? -EINTR : -EAGAIN; | |
1472 | } | |
1473 | } | |
3262c816 | 1474 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 | 1475 | |
3262c816 GB |
1476 | dprintk("svc: server %p, pool %u, socket %p, inuse=%d\n", |
1477 | rqstp, pool->sp_id, svsk, atomic_read(&svsk->sk_inuse)); | |
1da177e4 LT |
1478 | len = svsk->sk_recvfrom(rqstp); |
1479 | dprintk("svc: got len=%d\n", len); | |
1480 | ||
1481 | /* No data, incomplete (TCP) read, or accept() */ | |
1482 | if (len == 0 || len == -EAGAIN) { | |
1483 | rqstp->rq_res.len = 0; | |
1484 | svc_sock_release(rqstp); | |
1485 | return -EAGAIN; | |
1486 | } | |
1487 | svsk->sk_lastrecv = get_seconds(); | |
36bdfc8b | 1488 | clear_bit(SK_OLD, &svsk->sk_flags); |
1da177e4 | 1489 | |
bcdb81ae | 1490 | rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp)); |
1da177e4 LT |
1491 | rqstp->rq_chandle.defer = svc_defer; |
1492 | ||
1493 | if (serv->sv_stats) | |
1494 | serv->sv_stats->netcnt++; | |
1495 | return len; | |
1496 | } | |
1497 | ||
cca5172a | 1498 | /* |
1da177e4 LT |
1499 | * Drop request |
1500 | */ | |
1501 | void | |
1502 | svc_drop(struct svc_rqst *rqstp) | |
1503 | { | |
1504 | dprintk("svc: socket %p dropped request\n", rqstp->rq_sock); | |
1505 | svc_sock_release(rqstp); | |
1506 | } | |
1507 | ||
1508 | /* | |
1509 | * Return reply to client. | |
1510 | */ | |
1511 | int | |
1512 | svc_send(struct svc_rqst *rqstp) | |
1513 | { | |
1514 | struct svc_sock *svsk; | |
1515 | int len; | |
1516 | struct xdr_buf *xb; | |
1517 | ||
1518 | if ((svsk = rqstp->rq_sock) == NULL) { | |
1519 | printk(KERN_WARNING "NULL socket pointer in %s:%d\n", | |
1520 | __FILE__, __LINE__); | |
1521 | return -EFAULT; | |
1522 | } | |
1523 | ||
1524 | /* release the receive skb before sending the reply */ | |
1525 | svc_release_skb(rqstp); | |
1526 | ||
1527 | /* calculate over-all length */ | |
1528 | xb = & rqstp->rq_res; | |
1529 | xb->len = xb->head[0].iov_len + | |
1530 | xb->page_len + | |
1531 | xb->tail[0].iov_len; | |
1532 | ||
57b47a53 IM |
1533 | /* Grab svsk->sk_mutex to serialize outgoing data. */ |
1534 | mutex_lock(&svsk->sk_mutex); | |
1da177e4 LT |
1535 | if (test_bit(SK_DEAD, &svsk->sk_flags)) |
1536 | len = -ENOTCONN; | |
1537 | else | |
1538 | len = svsk->sk_sendto(rqstp); | |
57b47a53 | 1539 | mutex_unlock(&svsk->sk_mutex); |
1da177e4 LT |
1540 | svc_sock_release(rqstp); |
1541 | ||
1542 | if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN) | |
1543 | return 0; | |
1544 | return len; | |
1545 | } | |
1546 | ||
36bdfc8b GB |
1547 | /* |
1548 | * Timer function to close old temporary sockets, using | |
1549 | * a mark-and-sweep algorithm. | |
1550 | */ | |
1551 | static void | |
1552 | svc_age_temp_sockets(unsigned long closure) | |
1553 | { | |
1554 | struct svc_serv *serv = (struct svc_serv *)closure; | |
1555 | struct svc_sock *svsk; | |
1556 | struct list_head *le, *next; | |
1557 | LIST_HEAD(to_be_aged); | |
1558 | ||
1559 | dprintk("svc_age_temp_sockets\n"); | |
1560 | ||
1561 | if (!spin_trylock_bh(&serv->sv_lock)) { | |
1562 | /* busy, try again 1 sec later */ | |
1563 | dprintk("svc_age_temp_sockets: busy\n"); | |
1564 | mod_timer(&serv->sv_temptimer, jiffies + HZ); | |
1565 | return; | |
1566 | } | |
1567 | ||
1568 | list_for_each_safe(le, next, &serv->sv_tempsocks) { | |
1569 | svsk = list_entry(le, struct svc_sock, sk_list); | |
1570 | ||
1571 | if (!test_and_set_bit(SK_OLD, &svsk->sk_flags)) | |
1572 | continue; | |
c45c357d | 1573 | if (atomic_read(&svsk->sk_inuse) || test_bit(SK_BUSY, &svsk->sk_flags)) |
36bdfc8b | 1574 | continue; |
c45c357d | 1575 | atomic_inc(&svsk->sk_inuse); |
36bdfc8b GB |
1576 | list_move(le, &to_be_aged); |
1577 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1578 | set_bit(SK_DETACHED, &svsk->sk_flags); | |
1579 | } | |
1580 | spin_unlock_bh(&serv->sv_lock); | |
1581 | ||
1582 | while (!list_empty(&to_be_aged)) { | |
1583 | le = to_be_aged.next; | |
1584 | /* fiddling the sk_list node is safe 'cos we're SK_DETACHED */ | |
1585 | list_del_init(le); | |
1586 | svsk = list_entry(le, struct svc_sock, sk_list); | |
1587 | ||
1588 | dprintk("queuing svsk %p for closing, %lu seconds old\n", | |
1589 | svsk, get_seconds() - svsk->sk_lastrecv); | |
1590 | ||
1591 | /* a thread will dequeue and close it soon */ | |
1592 | svc_sock_enqueue(svsk); | |
1593 | svc_sock_put(svsk); | |
1594 | } | |
1595 | ||
1596 | mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ); | |
1597 | } | |
1598 | ||
1da177e4 LT |
1599 | /* |
1600 | * Initialize socket for RPC use and create svc_sock struct | |
1601 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1602 | */ | |
6b174337 CL |
1603 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1604 | struct socket *sock, | |
1605 | int *errp, int flags) | |
1da177e4 LT |
1606 | { |
1607 | struct svc_sock *svsk; | |
1608 | struct sock *inet; | |
6b174337 CL |
1609 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
1610 | int is_temporary = flags & SVC_SOCK_TEMPORARY; | |
1da177e4 LT |
1611 | |
1612 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1613 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1614 | *errp = -ENOMEM; |
1615 | return NULL; | |
1616 | } | |
1da177e4 LT |
1617 | |
1618 | inet = sock->sk; | |
1619 | ||
1620 | /* Register socket with portmapper */ | |
1621 | if (*errp >= 0 && pmap_register) | |
1622 | *errp = svc_register(serv, inet->sk_protocol, | |
1623 | ntohs(inet_sk(inet)->sport)); | |
1624 | ||
1625 | if (*errp < 0) { | |
1626 | kfree(svsk); | |
1627 | return NULL; | |
1628 | } | |
1629 | ||
1630 | set_bit(SK_BUSY, &svsk->sk_flags); | |
1631 | inet->sk_user_data = svsk; | |
1632 | svsk->sk_sock = sock; | |
1633 | svsk->sk_sk = inet; | |
1634 | svsk->sk_ostate = inet->sk_state_change; | |
1635 | svsk->sk_odata = inet->sk_data_ready; | |
1636 | svsk->sk_owspace = inet->sk_write_space; | |
1637 | svsk->sk_server = serv; | |
aaf68cfb | 1638 | atomic_set(&svsk->sk_inuse, 1); |
1da177e4 | 1639 | svsk->sk_lastrecv = get_seconds(); |
1a68d952 | 1640 | spin_lock_init(&svsk->sk_defer_lock); |
1da177e4 LT |
1641 | INIT_LIST_HEAD(&svsk->sk_deferred); |
1642 | INIT_LIST_HEAD(&svsk->sk_ready); | |
57b47a53 | 1643 | mutex_init(&svsk->sk_mutex); |
1da177e4 LT |
1644 | |
1645 | /* Initialize the socket */ | |
1646 | if (sock->type == SOCK_DGRAM) | |
1647 | svc_udp_init(svsk); | |
1648 | else | |
1649 | svc_tcp_init(svsk); | |
1650 | ||
1651 | spin_lock_bh(&serv->sv_lock); | |
6b174337 | 1652 | if (is_temporary) { |
1da177e4 LT |
1653 | set_bit(SK_TEMP, &svsk->sk_flags); |
1654 | list_add(&svsk->sk_list, &serv->sv_tempsocks); | |
1655 | serv->sv_tmpcnt++; | |
36bdfc8b GB |
1656 | if (serv->sv_temptimer.function == NULL) { |
1657 | /* setup timer to age temp sockets */ | |
1658 | setup_timer(&serv->sv_temptimer, svc_age_temp_sockets, | |
1659 | (unsigned long)serv); | |
1660 | mod_timer(&serv->sv_temptimer, | |
1661 | jiffies + svc_conn_age_period * HZ); | |
1662 | } | |
1da177e4 LT |
1663 | } else { |
1664 | clear_bit(SK_TEMP, &svsk->sk_flags); | |
1665 | list_add(&svsk->sk_list, &serv->sv_permsocks); | |
1666 | } | |
1667 | spin_unlock_bh(&serv->sv_lock); | |
1668 | ||
1669 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", | |
1670 | svsk, svsk->sk_sk); | |
1671 | ||
1da177e4 LT |
1672 | return svsk; |
1673 | } | |
1674 | ||
b41b66d6 N |
1675 | int svc_addsock(struct svc_serv *serv, |
1676 | int fd, | |
1677 | char *name_return, | |
1678 | int *proto) | |
1679 | { | |
1680 | int err = 0; | |
1681 | struct socket *so = sockfd_lookup(fd, &err); | |
1682 | struct svc_sock *svsk = NULL; | |
1683 | ||
1684 | if (!so) | |
1685 | return err; | |
1686 | if (so->sk->sk_family != AF_INET) | |
1687 | err = -EAFNOSUPPORT; | |
1688 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1689 | so->sk->sk_protocol != IPPROTO_UDP) | |
1690 | err = -EPROTONOSUPPORT; | |
1691 | else if (so->state > SS_UNCONNECTED) | |
1692 | err = -EISCONN; | |
1693 | else { | |
6b174337 | 1694 | svsk = svc_setup_socket(serv, so, &err, SVC_SOCK_DEFAULTS); |
e79eff1f N |
1695 | if (svsk) { |
1696 | svc_sock_received(svsk); | |
b41b66d6 | 1697 | err = 0; |
e79eff1f | 1698 | } |
b41b66d6 N |
1699 | } |
1700 | if (err) { | |
1701 | sockfd_put(so); | |
1702 | return err; | |
1703 | } | |
1704 | if (proto) *proto = so->sk->sk_protocol; | |
1705 | return one_sock_name(name_return, svsk); | |
1706 | } | |
1707 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1708 | ||
1da177e4 LT |
1709 | /* |
1710 | * Create socket for RPC service. | |
1711 | */ | |
6b174337 | 1712 | static int svc_create_socket(struct svc_serv *serv, int protocol, |
77f1f67a | 1713 | struct sockaddr *sin, int len, int flags) |
1da177e4 LT |
1714 | { |
1715 | struct svc_sock *svsk; | |
1716 | struct socket *sock; | |
1717 | int error; | |
1718 | int type; | |
ad06e4bd | 1719 | char buf[RPC_MAX_ADDRBUFLEN]; |
1da177e4 | 1720 | |
ad06e4bd CL |
1721 | dprintk("svc: svc_create_socket(%s, %d, %s)\n", |
1722 | serv->sv_program->pg_name, protocol, | |
77f1f67a | 1723 | __svc_print_addr(sin, buf, sizeof(buf))); |
1da177e4 LT |
1724 | |
1725 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1726 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1727 | "sockets supported\n"); | |
1728 | return -EINVAL; | |
1729 | } | |
1730 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; | |
1731 | ||
77f1f67a CL |
1732 | error = sock_create_kern(sin->sa_family, type, protocol, &sock); |
1733 | if (error < 0) | |
1da177e4 LT |
1734 | return error; |
1735 | ||
ed07536e PZ |
1736 | svc_reclassify_socket(sock); |
1737 | ||
18114746 | 1738 | if (type == SOCK_STREAM) |
77f1f67a CL |
1739 | sock->sk->sk_reuse = 1; /* allow address reuse */ |
1740 | error = kernel_bind(sock, sin, len); | |
18114746 ES |
1741 | if (error < 0) |
1742 | goto bummer; | |
1da177e4 LT |
1743 | |
1744 | if (protocol == IPPROTO_TCP) { | |
e6242e92 | 1745 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1746 | goto bummer; |
1747 | } | |
1748 | ||
e79eff1f N |
1749 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { |
1750 | svc_sock_received(svsk); | |
6b174337 | 1751 | return ntohs(inet_sk(svsk->sk_sk)->sport); |
e79eff1f | 1752 | } |
1da177e4 LT |
1753 | |
1754 | bummer: | |
1755 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1756 | sock_release(sock); | |
1757 | return error; | |
1758 | } | |
1759 | ||
1760 | /* | |
1761 | * Remove a dead socket | |
1762 | */ | |
aaf68cfb | 1763 | static void |
1da177e4 LT |
1764 | svc_delete_socket(struct svc_sock *svsk) |
1765 | { | |
1766 | struct svc_serv *serv; | |
1767 | struct sock *sk; | |
1768 | ||
1769 | dprintk("svc: svc_delete_socket(%p)\n", svsk); | |
1770 | ||
1771 | serv = svsk->sk_server; | |
1772 | sk = svsk->sk_sk; | |
1773 | ||
1774 | sk->sk_state_change = svsk->sk_ostate; | |
1775 | sk->sk_data_ready = svsk->sk_odata; | |
1776 | sk->sk_write_space = svsk->sk_owspace; | |
1777 | ||
1778 | spin_lock_bh(&serv->sv_lock); | |
1779 | ||
36bdfc8b GB |
1780 | if (!test_and_set_bit(SK_DETACHED, &svsk->sk_flags)) |
1781 | list_del_init(&svsk->sk_list); | |
cca5172a | 1782 | /* |
3262c816 GB |
1783 | * We used to delete the svc_sock from whichever list |
1784 | * it's sk_ready node was on, but we don't actually | |
1785 | * need to. This is because the only time we're called | |
1786 | * while still attached to a queue, the queue itself | |
1787 | * is about to be destroyed (in svc_destroy). | |
1788 | */ | |
aaf68cfb N |
1789 | if (!test_and_set_bit(SK_DEAD, &svsk->sk_flags)) { |
1790 | BUG_ON(atomic_read(&svsk->sk_inuse)<2); | |
1791 | atomic_dec(&svsk->sk_inuse); | |
1da177e4 LT |
1792 | if (test_bit(SK_TEMP, &svsk->sk_flags)) |
1793 | serv->sv_tmpcnt--; | |
aaf68cfb | 1794 | } |
1da177e4 | 1795 | |
d6740df9 | 1796 | spin_unlock_bh(&serv->sv_lock); |
aaf68cfb N |
1797 | } |
1798 | ||
cda1fd4a | 1799 | static void svc_close_socket(struct svc_sock *svsk) |
aaf68cfb N |
1800 | { |
1801 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1802 | if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) | |
1803 | /* someone else will have to effect the close */ | |
1804 | return; | |
1805 | ||
1806 | atomic_inc(&svsk->sk_inuse); | |
1807 | svc_delete_socket(svsk); | |
1808 | clear_bit(SK_BUSY, &svsk->sk_flags); | |
d6740df9 | 1809 | svc_sock_put(svsk); |
1da177e4 LT |
1810 | } |
1811 | ||
cda1fd4a N |
1812 | void svc_force_close_socket(struct svc_sock *svsk) |
1813 | { | |
1814 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1815 | if (test_bit(SK_BUSY, &svsk->sk_flags)) { | |
1816 | /* Waiting to be processed, but no threads left, | |
1817 | * So just remove it from the waiting list | |
1818 | */ | |
1819 | list_del_init(&svsk->sk_ready); | |
1820 | clear_bit(SK_BUSY, &svsk->sk_flags); | |
1821 | } | |
1822 | svc_close_socket(svsk); | |
1823 | } | |
1824 | ||
6b174337 CL |
1825 | /** |
1826 | * svc_makesock - Make a socket for nfsd and lockd | |
1827 | * @serv: RPC server structure | |
1828 | * @protocol: transport protocol to use | |
1829 | * @port: port to use | |
482fb94e | 1830 | * @flags: requested socket characteristics |
6b174337 | 1831 | * |
1da177e4 | 1832 | */ |
482fb94e CL |
1833 | int svc_makesock(struct svc_serv *serv, int protocol, unsigned short port, |
1834 | int flags) | |
1da177e4 | 1835 | { |
6b174337 CL |
1836 | struct sockaddr_in sin = { |
1837 | .sin_family = AF_INET, | |
1838 | .sin_addr.s_addr = INADDR_ANY, | |
1839 | .sin_port = htons(port), | |
1840 | }; | |
1da177e4 LT |
1841 | |
1842 | dprintk("svc: creating socket proto = %d\n", protocol); | |
77f1f67a CL |
1843 | return svc_create_socket(serv, protocol, (struct sockaddr *) &sin, |
1844 | sizeof(sin), flags); | |
1da177e4 LT |
1845 | } |
1846 | ||
1847 | /* | |
cca5172a | 1848 | * Handle defer and revisit of requests |
1da177e4 LT |
1849 | */ |
1850 | ||
1851 | static void svc_revisit(struct cache_deferred_req *dreq, int too_many) | |
1852 | { | |
1853 | struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle); | |
1da177e4 LT |
1854 | struct svc_sock *svsk; |
1855 | ||
1856 | if (too_many) { | |
1857 | svc_sock_put(dr->svsk); | |
1858 | kfree(dr); | |
1859 | return; | |
1860 | } | |
1861 | dprintk("revisit queued\n"); | |
1862 | svsk = dr->svsk; | |
1863 | dr->svsk = NULL; | |
1a68d952 | 1864 | spin_lock_bh(&svsk->sk_defer_lock); |
1da177e4 | 1865 | list_add(&dr->handle.recent, &svsk->sk_deferred); |
1a68d952 | 1866 | spin_unlock_bh(&svsk->sk_defer_lock); |
1da177e4 LT |
1867 | set_bit(SK_DEFERRED, &svsk->sk_flags); |
1868 | svc_sock_enqueue(svsk); | |
1869 | svc_sock_put(svsk); | |
1870 | } | |
1871 | ||
1872 | static struct cache_deferred_req * | |
1873 | svc_defer(struct cache_req *req) | |
1874 | { | |
1875 | struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle); | |
1876 | int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len); | |
1877 | struct svc_deferred_req *dr; | |
1878 | ||
1879 | if (rqstp->rq_arg.page_len) | |
1880 | return NULL; /* if more than a page, give up FIXME */ | |
1881 | if (rqstp->rq_deferred) { | |
1882 | dr = rqstp->rq_deferred; | |
1883 | rqstp->rq_deferred = NULL; | |
1884 | } else { | |
1885 | int skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
1886 | /* FIXME maybe discard if size too large */ | |
1887 | dr = kmalloc(size, GFP_KERNEL); | |
1888 | if (dr == NULL) | |
1889 | return NULL; | |
1890 | ||
1891 | dr->handle.owner = rqstp->rq_server; | |
1892 | dr->prot = rqstp->rq_prot; | |
24422222 CL |
1893 | memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen); |
1894 | dr->addrlen = rqstp->rq_addrlen; | |
1918e341 | 1895 | dr->daddr = rqstp->rq_daddr; |
1da177e4 LT |
1896 | dr->argslen = rqstp->rq_arg.len >> 2; |
1897 | memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2); | |
1898 | } | |
c45c357d | 1899 | atomic_inc(&rqstp->rq_sock->sk_inuse); |
1da177e4 | 1900 | dr->svsk = rqstp->rq_sock; |
1da177e4 LT |
1901 | |
1902 | dr->handle.revisit = svc_revisit; | |
1903 | return &dr->handle; | |
1904 | } | |
1905 | ||
1906 | /* | |
1907 | * recv data from a deferred request into an active one | |
1908 | */ | |
1909 | static int svc_deferred_recv(struct svc_rqst *rqstp) | |
1910 | { | |
1911 | struct svc_deferred_req *dr = rqstp->rq_deferred; | |
1912 | ||
1913 | rqstp->rq_arg.head[0].iov_base = dr->args; | |
1914 | rqstp->rq_arg.head[0].iov_len = dr->argslen<<2; | |
1915 | rqstp->rq_arg.page_len = 0; | |
1916 | rqstp->rq_arg.len = dr->argslen<<2; | |
1917 | rqstp->rq_prot = dr->prot; | |
24422222 CL |
1918 | memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen); |
1919 | rqstp->rq_addrlen = dr->addrlen; | |
1918e341 | 1920 | rqstp->rq_daddr = dr->daddr; |
44524359 | 1921 | rqstp->rq_respages = rqstp->rq_pages; |
1da177e4 LT |
1922 | return dr->argslen<<2; |
1923 | } | |
1924 | ||
1925 | ||
1926 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk) | |
1927 | { | |
1928 | struct svc_deferred_req *dr = NULL; | |
cca5172a | 1929 | |
1da177e4 LT |
1930 | if (!test_bit(SK_DEFERRED, &svsk->sk_flags)) |
1931 | return NULL; | |
1a68d952 | 1932 | spin_lock_bh(&svsk->sk_defer_lock); |
1da177e4 LT |
1933 | clear_bit(SK_DEFERRED, &svsk->sk_flags); |
1934 | if (!list_empty(&svsk->sk_deferred)) { | |
1935 | dr = list_entry(svsk->sk_deferred.next, | |
1936 | struct svc_deferred_req, | |
1937 | handle.recent); | |
1938 | list_del_init(&dr->handle.recent); | |
1939 | set_bit(SK_DEFERRED, &svsk->sk_flags); | |
1940 | } | |
1a68d952 | 1941 | spin_unlock_bh(&svsk->sk_defer_lock); |
1da177e4 LT |
1942 | return dr; |
1943 | } |