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> | |
c752f073 | 39 | #include <net/tcp_states.h> |
1da177e4 LT |
40 | #include <asm/uaccess.h> |
41 | #include <asm/ioctls.h> | |
42 | ||
43 | #include <linux/sunrpc/types.h> | |
44 | #include <linux/sunrpc/xdr.h> | |
45 | #include <linux/sunrpc/svcsock.h> | |
46 | #include <linux/sunrpc/stats.h> | |
47 | ||
48 | /* SMP locking strategy: | |
49 | * | |
3262c816 GB |
50 | * svc_pool->sp_lock protects most of the fields of that pool. |
51 | * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt. | |
52 | * when both need to be taken (rare), svc_serv->sv_lock is first. | |
53 | * BKL protects svc_serv->sv_nrthread. | |
1a68d952 | 54 | * svc_sock->sk_defer_lock protects the svc_sock->sk_deferred list |
c081a0c7 | 55 | * svc_sock->sk_flags.SK_BUSY prevents a svc_sock being enqueued multiply. |
1da177e4 LT |
56 | * |
57 | * Some flags can be set to certain values at any time | |
58 | * providing that certain rules are followed: | |
59 | * | |
1da177e4 | 60 | * SK_CONN, SK_DATA, can be set or cleared at any time. |
cca5172a | 61 | * after a set, svc_sock_enqueue must be called. |
1da177e4 LT |
62 | * after a clear, the socket must be read/accepted |
63 | * if this succeeds, it must be set again. | |
64 | * SK_CLOSE can set at any time. It is never cleared. | |
aaf68cfb N |
65 | * sk_inuse contains a bias of '1' until SK_DEAD is set. |
66 | * so when sk_inuse hits zero, we know the socket is dead | |
67 | * and no-one is using it. | |
68 | * SK_DEAD can only be set while SK_BUSY is held which ensures | |
69 | * no other thread will be using the socket or will try to | |
70 | * set SK_DEAD. | |
1da177e4 LT |
71 | * |
72 | */ | |
73 | ||
74 | #define RPCDBG_FACILITY RPCDBG_SVCSOCK | |
75 | ||
76 | ||
77 | static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, | |
6b174337 | 78 | int *errp, int flags); |
aaf68cfb | 79 | static void svc_delete_socket(struct svc_sock *svsk); |
1da177e4 LT |
80 | static void svc_udp_data_ready(struct sock *, int); |
81 | static int svc_udp_recvfrom(struct svc_rqst *); | |
82 | static int svc_udp_sendto(struct svc_rqst *); | |
83 | ||
84 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk); | |
85 | static int svc_deferred_recv(struct svc_rqst *rqstp); | |
86 | static struct cache_deferred_req *svc_defer(struct cache_req *req); | |
87 | ||
36bdfc8b GB |
88 | /* apparently the "standard" is that clients close |
89 | * idle connections after 5 minutes, servers after | |
90 | * 6 minutes | |
91 | * http://www.connectathon.org/talks96/nfstcp.pdf | |
92 | */ | |
93 | static int svc_conn_age_period = 6*60; | |
94 | ||
ed07536e PZ |
95 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
96 | static struct lock_class_key svc_key[2]; | |
97 | static struct lock_class_key svc_slock_key[2]; | |
98 | ||
99 | static inline void svc_reclassify_socket(struct socket *sock) | |
100 | { | |
101 | struct sock *sk = sock->sk; | |
102 | BUG_ON(sk->sk_lock.owner != NULL); | |
103 | switch (sk->sk_family) { | |
104 | case AF_INET: | |
105 | sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", | |
106 | &svc_slock_key[0], "sk_lock-AF_INET-NFSD", &svc_key[0]); | |
107 | break; | |
108 | ||
109 | case AF_INET6: | |
110 | sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", | |
111 | &svc_slock_key[1], "sk_lock-AF_INET6-NFSD", &svc_key[1]); | |
112 | break; | |
113 | ||
114 | default: | |
115 | BUG(); | |
116 | } | |
117 | } | |
118 | #else | |
119 | static inline void svc_reclassify_socket(struct socket *sock) | |
120 | { | |
121 | } | |
122 | #endif | |
123 | ||
1da177e4 | 124 | /* |
3262c816 | 125 | * Queue up an idle server thread. Must have pool->sp_lock held. |
1da177e4 | 126 | * Note: this is really a stack rather than a queue, so that we only |
3262c816 | 127 | * use as many different threads as we need, and the rest don't pollute |
1da177e4 LT |
128 | * the cache. |
129 | */ | |
130 | static inline void | |
3262c816 | 131 | svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp) |
1da177e4 | 132 | { |
3262c816 | 133 | list_add(&rqstp->rq_list, &pool->sp_threads); |
1da177e4 LT |
134 | } |
135 | ||
136 | /* | |
3262c816 | 137 | * Dequeue an nfsd thread. Must have pool->sp_lock held. |
1da177e4 LT |
138 | */ |
139 | static inline void | |
3262c816 | 140 | svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp) |
1da177e4 LT |
141 | { |
142 | list_del(&rqstp->rq_list); | |
143 | } | |
144 | ||
145 | /* | |
146 | * Release an skbuff after use | |
147 | */ | |
148 | static inline void | |
149 | svc_release_skb(struct svc_rqst *rqstp) | |
150 | { | |
151 | struct sk_buff *skb = rqstp->rq_skbuff; | |
152 | struct svc_deferred_req *dr = rqstp->rq_deferred; | |
153 | ||
154 | if (skb) { | |
155 | rqstp->rq_skbuff = NULL; | |
156 | ||
157 | dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); | |
158 | skb_free_datagram(rqstp->rq_sock->sk_sk, skb); | |
159 | } | |
160 | if (dr) { | |
161 | rqstp->rq_deferred = NULL; | |
162 | kfree(dr); | |
163 | } | |
164 | } | |
165 | ||
166 | /* | |
167 | * Any space to write? | |
168 | */ | |
169 | static inline unsigned long | |
170 | svc_sock_wspace(struct svc_sock *svsk) | |
171 | { | |
172 | int wspace; | |
173 | ||
174 | if (svsk->sk_sock->type == SOCK_STREAM) | |
175 | wspace = sk_stream_wspace(svsk->sk_sk); | |
176 | else | |
177 | wspace = sock_wspace(svsk->sk_sk); | |
178 | ||
179 | return wspace; | |
180 | } | |
181 | ||
182 | /* | |
183 | * Queue up a socket with data pending. If there are idle nfsd | |
184 | * processes, wake 'em up. | |
185 | * | |
186 | */ | |
187 | static void | |
188 | svc_sock_enqueue(struct svc_sock *svsk) | |
189 | { | |
190 | struct svc_serv *serv = svsk->sk_server; | |
bfd24160 | 191 | struct svc_pool *pool; |
1da177e4 | 192 | struct svc_rqst *rqstp; |
bfd24160 | 193 | int cpu; |
1da177e4 LT |
194 | |
195 | if (!(svsk->sk_flags & | |
196 | ( (1<<SK_CONN)|(1<<SK_DATA)|(1<<SK_CLOSE)|(1<<SK_DEFERRED)) )) | |
197 | return; | |
198 | if (test_bit(SK_DEAD, &svsk->sk_flags)) | |
199 | return; | |
200 | ||
bfd24160 GB |
201 | cpu = get_cpu(); |
202 | pool = svc_pool_for_cpu(svsk->sk_server, cpu); | |
203 | put_cpu(); | |
204 | ||
3262c816 | 205 | spin_lock_bh(&pool->sp_lock); |
1da177e4 | 206 | |
3262c816 GB |
207 | if (!list_empty(&pool->sp_threads) && |
208 | !list_empty(&pool->sp_sockets)) | |
1da177e4 LT |
209 | printk(KERN_ERR |
210 | "svc_sock_enqueue: threads and sockets both waiting??\n"); | |
211 | ||
212 | if (test_bit(SK_DEAD, &svsk->sk_flags)) { | |
213 | /* Don't enqueue dead sockets */ | |
214 | dprintk("svc: socket %p is dead, not enqueued\n", svsk->sk_sk); | |
215 | goto out_unlock; | |
216 | } | |
217 | ||
c081a0c7 GB |
218 | /* Mark socket as busy. It will remain in this state until the |
219 | * server has processed all pending data and put the socket back | |
220 | * on the idle list. We update SK_BUSY atomically because | |
221 | * it also guards against trying to enqueue the svc_sock twice. | |
222 | */ | |
223 | if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) { | |
224 | /* Don't enqueue socket while already enqueued */ | |
1da177e4 LT |
225 | dprintk("svc: socket %p busy, not enqueued\n", svsk->sk_sk); |
226 | goto out_unlock; | |
227 | } | |
3262c816 GB |
228 | BUG_ON(svsk->sk_pool != NULL); |
229 | svsk->sk_pool = pool; | |
1da177e4 LT |
230 | |
231 | set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
c6b0a9f8 | 232 | if (((atomic_read(&svsk->sk_reserved) + serv->sv_max_mesg)*2 |
1da177e4 LT |
233 | > svc_sock_wspace(svsk)) |
234 | && !test_bit(SK_CLOSE, &svsk->sk_flags) | |
235 | && !test_bit(SK_CONN, &svsk->sk_flags)) { | |
236 | /* Don't enqueue while not enough space for reply */ | |
237 | dprintk("svc: socket %p no space, %d*2 > %ld, not enqueued\n", | |
c6b0a9f8 | 238 | svsk->sk_sk, atomic_read(&svsk->sk_reserved)+serv->sv_max_mesg, |
1da177e4 | 239 | svc_sock_wspace(svsk)); |
3262c816 | 240 | svsk->sk_pool = NULL; |
c081a0c7 | 241 | clear_bit(SK_BUSY, &svsk->sk_flags); |
1da177e4 LT |
242 | goto out_unlock; |
243 | } | |
244 | clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); | |
245 | ||
1da177e4 | 246 | |
3262c816 GB |
247 | if (!list_empty(&pool->sp_threads)) { |
248 | rqstp = list_entry(pool->sp_threads.next, | |
1da177e4 LT |
249 | struct svc_rqst, |
250 | rq_list); | |
251 | dprintk("svc: socket %p served by daemon %p\n", | |
252 | svsk->sk_sk, rqstp); | |
3262c816 | 253 | svc_thread_dequeue(pool, rqstp); |
1da177e4 | 254 | if (rqstp->rq_sock) |
cca5172a | 255 | printk(KERN_ERR |
1da177e4 LT |
256 | "svc_sock_enqueue: server %p, rq_sock=%p!\n", |
257 | rqstp, rqstp->rq_sock); | |
258 | rqstp->rq_sock = svsk; | |
c45c357d | 259 | atomic_inc(&svsk->sk_inuse); |
c6b0a9f8 | 260 | rqstp->rq_reserved = serv->sv_max_mesg; |
5685f0fa | 261 | atomic_add(rqstp->rq_reserved, &svsk->sk_reserved); |
3262c816 | 262 | BUG_ON(svsk->sk_pool != pool); |
1da177e4 LT |
263 | wake_up(&rqstp->rq_wait); |
264 | } else { | |
265 | dprintk("svc: socket %p put into queue\n", svsk->sk_sk); | |
3262c816 GB |
266 | list_add_tail(&svsk->sk_ready, &pool->sp_sockets); |
267 | BUG_ON(svsk->sk_pool != pool); | |
1da177e4 LT |
268 | } |
269 | ||
270 | out_unlock: | |
3262c816 | 271 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 LT |
272 | } |
273 | ||
274 | /* | |
3262c816 | 275 | * Dequeue the first socket. Must be called with the pool->sp_lock held. |
1da177e4 LT |
276 | */ |
277 | static inline struct svc_sock * | |
3262c816 | 278 | svc_sock_dequeue(struct svc_pool *pool) |
1da177e4 LT |
279 | { |
280 | struct svc_sock *svsk; | |
281 | ||
3262c816 | 282 | if (list_empty(&pool->sp_sockets)) |
1da177e4 LT |
283 | return NULL; |
284 | ||
3262c816 | 285 | svsk = list_entry(pool->sp_sockets.next, |
1da177e4 LT |
286 | struct svc_sock, sk_ready); |
287 | list_del_init(&svsk->sk_ready); | |
288 | ||
289 | dprintk("svc: socket %p dequeued, inuse=%d\n", | |
c45c357d | 290 | svsk->sk_sk, atomic_read(&svsk->sk_inuse)); |
1da177e4 LT |
291 | |
292 | return svsk; | |
293 | } | |
294 | ||
295 | /* | |
296 | * Having read something from a socket, check whether it | |
297 | * needs to be re-enqueued. | |
298 | * Note: SK_DATA only gets cleared when a read-attempt finds | |
299 | * no (or insufficient) data. | |
300 | */ | |
301 | static inline void | |
302 | svc_sock_received(struct svc_sock *svsk) | |
303 | { | |
3262c816 | 304 | svsk->sk_pool = NULL; |
1da177e4 LT |
305 | clear_bit(SK_BUSY, &svsk->sk_flags); |
306 | svc_sock_enqueue(svsk); | |
307 | } | |
308 | ||
309 | ||
310 | /** | |
311 | * svc_reserve - change the space reserved for the reply to a request. | |
312 | * @rqstp: The request in question | |
313 | * @space: new max space to reserve | |
314 | * | |
315 | * Each request reserves some space on the output queue of the socket | |
316 | * to make sure the reply fits. This function reduces that reserved | |
317 | * space to be the amount of space used already, plus @space. | |
318 | * | |
319 | */ | |
320 | void svc_reserve(struct svc_rqst *rqstp, int space) | |
321 | { | |
322 | space += rqstp->rq_res.head[0].iov_len; | |
323 | ||
324 | if (space < rqstp->rq_reserved) { | |
325 | struct svc_sock *svsk = rqstp->rq_sock; | |
5685f0fa | 326 | atomic_sub((rqstp->rq_reserved - space), &svsk->sk_reserved); |
1da177e4 | 327 | rqstp->rq_reserved = space; |
1da177e4 LT |
328 | |
329 | svc_sock_enqueue(svsk); | |
330 | } | |
331 | } | |
332 | ||
333 | /* | |
334 | * Release a socket after use. | |
335 | */ | |
336 | static inline void | |
337 | svc_sock_put(struct svc_sock *svsk) | |
338 | { | |
aaf68cfb N |
339 | if (atomic_dec_and_test(&svsk->sk_inuse)) { |
340 | BUG_ON(! test_bit(SK_DEAD, &svsk->sk_flags)); | |
341 | ||
202dd450 | 342 | dprintk("svc: releasing dead socket\n"); |
d6740df9 NB |
343 | if (svsk->sk_sock->file) |
344 | sockfd_put(svsk->sk_sock); | |
345 | else | |
346 | sock_release(svsk->sk_sock); | |
347 | if (svsk->sk_info_authunix != NULL) | |
348 | svcauth_unix_info_release(svsk->sk_info_authunix); | |
1da177e4 LT |
349 | kfree(svsk); |
350 | } | |
1da177e4 LT |
351 | } |
352 | ||
353 | static void | |
354 | svc_sock_release(struct svc_rqst *rqstp) | |
355 | { | |
356 | struct svc_sock *svsk = rqstp->rq_sock; | |
357 | ||
358 | svc_release_skb(rqstp); | |
359 | ||
44524359 | 360 | svc_free_res_pages(rqstp); |
1da177e4 LT |
361 | rqstp->rq_res.page_len = 0; |
362 | rqstp->rq_res.page_base = 0; | |
363 | ||
364 | ||
365 | /* Reset response buffer and release | |
366 | * the reservation. | |
367 | * But first, check that enough space was reserved | |
368 | * for the reply, otherwise we have a bug! | |
369 | */ | |
370 | if ((rqstp->rq_res.len) > rqstp->rq_reserved) | |
371 | printk(KERN_ERR "RPC request reserved %d but used %d\n", | |
372 | rqstp->rq_reserved, | |
373 | rqstp->rq_res.len); | |
374 | ||
375 | rqstp->rq_res.head[0].iov_len = 0; | |
376 | svc_reserve(rqstp, 0); | |
377 | rqstp->rq_sock = NULL; | |
378 | ||
379 | svc_sock_put(svsk); | |
380 | } | |
381 | ||
382 | /* | |
383 | * External function to wake up a server waiting for data | |
3262c816 GB |
384 | * This really only makes sense for services like lockd |
385 | * which have exactly one thread anyway. | |
1da177e4 LT |
386 | */ |
387 | void | |
388 | svc_wake_up(struct svc_serv *serv) | |
389 | { | |
390 | struct svc_rqst *rqstp; | |
3262c816 GB |
391 | unsigned int i; |
392 | struct svc_pool *pool; | |
393 | ||
394 | for (i = 0; i < serv->sv_nrpools; i++) { | |
395 | pool = &serv->sv_pools[i]; | |
396 | ||
397 | spin_lock_bh(&pool->sp_lock); | |
398 | if (!list_empty(&pool->sp_threads)) { | |
399 | rqstp = list_entry(pool->sp_threads.next, | |
400 | struct svc_rqst, | |
401 | rq_list); | |
402 | dprintk("svc: daemon %p woken up.\n", rqstp); | |
403 | /* | |
404 | svc_thread_dequeue(pool, rqstp); | |
405 | rqstp->rq_sock = NULL; | |
406 | */ | |
407 | wake_up(&rqstp->rq_wait); | |
408 | } | |
409 | spin_unlock_bh(&pool->sp_lock); | |
1da177e4 | 410 | } |
1da177e4 LT |
411 | } |
412 | ||
413 | /* | |
414 | * Generic sendto routine | |
415 | */ | |
416 | static int | |
417 | svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) | |
418 | { | |
419 | struct svc_sock *svsk = rqstp->rq_sock; | |
420 | struct socket *sock = svsk->sk_sock; | |
421 | int slen; | |
422 | char buffer[CMSG_SPACE(sizeof(struct in_pktinfo))]; | |
423 | struct cmsghdr *cmh = (struct cmsghdr *)buffer; | |
424 | struct in_pktinfo *pki = (struct in_pktinfo *)CMSG_DATA(cmh); | |
425 | int len = 0; | |
426 | int result; | |
427 | int size; | |
428 | struct page **ppage = xdr->pages; | |
429 | size_t base = xdr->page_base; | |
430 | unsigned int pglen = xdr->page_len; | |
431 | unsigned int flags = MSG_MORE; | |
432 | ||
433 | slen = xdr->len; | |
434 | ||
435 | if (rqstp->rq_prot == IPPROTO_UDP) { | |
436 | /* set the source and destination */ | |
437 | struct msghdr msg; | |
438 | msg.msg_name = &rqstp->rq_addr; | |
439 | msg.msg_namelen = sizeof(rqstp->rq_addr); | |
440 | msg.msg_iov = NULL; | |
441 | msg.msg_iovlen = 0; | |
442 | msg.msg_flags = MSG_MORE; | |
443 | ||
444 | msg.msg_control = cmh; | |
445 | msg.msg_controllen = sizeof(buffer); | |
446 | cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); | |
447 | cmh->cmsg_level = SOL_IP; | |
448 | cmh->cmsg_type = IP_PKTINFO; | |
449 | pki->ipi_ifindex = 0; | |
450 | pki->ipi_spec_dst.s_addr = rqstp->rq_daddr; | |
451 | ||
452 | if (sock_sendmsg(sock, &msg, 0) < 0) | |
453 | goto out; | |
454 | } | |
455 | ||
456 | /* send head */ | |
457 | if (slen == xdr->head[0].iov_len) | |
458 | flags = 0; | |
44524359 N |
459 | len = kernel_sendpage(sock, rqstp->rq_respages[0], 0, |
460 | xdr->head[0].iov_len, flags); | |
1da177e4 LT |
461 | if (len != xdr->head[0].iov_len) |
462 | goto out; | |
463 | slen -= xdr->head[0].iov_len; | |
464 | if (slen == 0) | |
465 | goto out; | |
466 | ||
467 | /* send page data */ | |
468 | size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; | |
469 | while (pglen > 0) { | |
470 | if (slen == size) | |
471 | flags = 0; | |
e6242e92 | 472 | result = kernel_sendpage(sock, *ppage, base, size, flags); |
1da177e4 LT |
473 | if (result > 0) |
474 | len += result; | |
475 | if (result != size) | |
476 | goto out; | |
477 | slen -= size; | |
478 | pglen -= size; | |
479 | size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; | |
480 | base = 0; | |
481 | ppage++; | |
482 | } | |
483 | /* send tail */ | |
484 | if (xdr->tail[0].iov_len) { | |
44524359 N |
485 | result = kernel_sendpage(sock, rqstp->rq_respages[0], |
486 | ((unsigned long)xdr->tail[0].iov_base) | |
cca5172a | 487 | & (PAGE_SIZE-1), |
1da177e4 LT |
488 | xdr->tail[0].iov_len, 0); |
489 | ||
490 | if (result > 0) | |
491 | len += result; | |
492 | } | |
493 | out: | |
494 | dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %x)\n", | |
495 | rqstp->rq_sock, xdr->head[0].iov_base, xdr->head[0].iov_len, xdr->len, len, | |
496 | rqstp->rq_addr.sin_addr.s_addr); | |
497 | ||
498 | return len; | |
499 | } | |
500 | ||
80212d59 N |
501 | /* |
502 | * Report socket names for nfsdfs | |
503 | */ | |
504 | static int one_sock_name(char *buf, struct svc_sock *svsk) | |
505 | { | |
506 | int len; | |
507 | ||
508 | switch(svsk->sk_sk->sk_family) { | |
509 | case AF_INET: | |
510 | len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n", | |
511 | svsk->sk_sk->sk_protocol==IPPROTO_UDP? | |
512 | "udp" : "tcp", | |
513 | NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr), | |
514 | inet_sk(svsk->sk_sk)->num); | |
515 | break; | |
516 | default: | |
517 | len = sprintf(buf, "*unknown-%d*\n", | |
518 | svsk->sk_sk->sk_family); | |
519 | } | |
520 | return len; | |
521 | } | |
522 | ||
523 | int | |
b41b66d6 | 524 | svc_sock_names(char *buf, struct svc_serv *serv, char *toclose) |
80212d59 | 525 | { |
b41b66d6 | 526 | struct svc_sock *svsk, *closesk = NULL; |
80212d59 N |
527 | int len = 0; |
528 | ||
529 | if (!serv) | |
530 | return 0; | |
aaf68cfb | 531 | spin_lock_bh(&serv->sv_lock); |
80212d59 N |
532 | list_for_each_entry(svsk, &serv->sv_permsocks, sk_list) { |
533 | int onelen = one_sock_name(buf+len, svsk); | |
b41b66d6 N |
534 | if (toclose && strcmp(toclose, buf+len) == 0) |
535 | closesk = svsk; | |
536 | else | |
537 | len += onelen; | |
80212d59 | 538 | } |
aaf68cfb | 539 | spin_unlock_bh(&serv->sv_lock); |
b41b66d6 | 540 | if (closesk) |
5680c446 N |
541 | /* Should unregister with portmap, but you cannot |
542 | * unregister just one protocol... | |
543 | */ | |
aaf68cfb | 544 | svc_close_socket(closesk); |
37a03472 N |
545 | else if (toclose) |
546 | return -ENOENT; | |
80212d59 N |
547 | return len; |
548 | } | |
549 | EXPORT_SYMBOL(svc_sock_names); | |
550 | ||
1da177e4 LT |
551 | /* |
552 | * Check input queue length | |
553 | */ | |
554 | static int | |
555 | svc_recv_available(struct svc_sock *svsk) | |
556 | { | |
1da177e4 LT |
557 | struct socket *sock = svsk->sk_sock; |
558 | int avail, err; | |
559 | ||
e6242e92 | 560 | err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); |
1da177e4 LT |
561 | |
562 | return (err >= 0)? avail : err; | |
563 | } | |
564 | ||
565 | /* | |
566 | * Generic recvfrom routine. | |
567 | */ | |
568 | static int | |
569 | svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, int buflen) | |
570 | { | |
571 | struct msghdr msg; | |
572 | struct socket *sock; | |
573 | int len, alen; | |
574 | ||
575 | rqstp->rq_addrlen = sizeof(rqstp->rq_addr); | |
576 | sock = rqstp->rq_sock->sk_sock; | |
577 | ||
578 | msg.msg_name = &rqstp->rq_addr; | |
579 | msg.msg_namelen = sizeof(rqstp->rq_addr); | |
580 | msg.msg_control = NULL; | |
581 | msg.msg_controllen = 0; | |
582 | ||
583 | msg.msg_flags = MSG_DONTWAIT; | |
584 | ||
585 | len = kernel_recvmsg(sock, &msg, iov, nr, buflen, MSG_DONTWAIT); | |
586 | ||
587 | /* sock_recvmsg doesn't fill in the name/namelen, so we must.. | |
588 | * possibly we should cache this in the svc_sock structure | |
589 | * at accept time. FIXME | |
590 | */ | |
591 | alen = sizeof(rqstp->rq_addr); | |
e6242e92 | 592 | kernel_getpeername(sock, (struct sockaddr *)&rqstp->rq_addr, &alen); |
1da177e4 LT |
593 | |
594 | dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", | |
595 | rqstp->rq_sock, iov[0].iov_base, iov[0].iov_len, len); | |
596 | ||
597 | return len; | |
598 | } | |
599 | ||
600 | /* | |
601 | * Set socket snd and rcv buffer lengths | |
602 | */ | |
603 | static inline void | |
604 | svc_sock_setbufsize(struct socket *sock, unsigned int snd, unsigned int rcv) | |
605 | { | |
606 | #if 0 | |
607 | mm_segment_t oldfs; | |
608 | oldfs = get_fs(); set_fs(KERNEL_DS); | |
609 | sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, | |
610 | (char*)&snd, sizeof(snd)); | |
611 | sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
612 | (char*)&rcv, sizeof(rcv)); | |
613 | #else | |
614 | /* sock_setsockopt limits use to sysctl_?mem_max, | |
615 | * which isn't acceptable. Until that is made conditional | |
616 | * on not having CAP_SYS_RESOURCE or similar, we go direct... | |
617 | * DaveM said I could! | |
618 | */ | |
619 | lock_sock(sock->sk); | |
620 | sock->sk->sk_sndbuf = snd * 2; | |
621 | sock->sk->sk_rcvbuf = rcv * 2; | |
622 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; | |
623 | release_sock(sock->sk); | |
624 | #endif | |
625 | } | |
626 | /* | |
627 | * INET callback when data has been received on the socket. | |
628 | */ | |
629 | static void | |
630 | svc_udp_data_ready(struct sock *sk, int count) | |
631 | { | |
939bb7ef | 632 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 | 633 | |
939bb7ef NB |
634 | if (svsk) { |
635 | dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", | |
636 | svsk, sk, count, test_bit(SK_BUSY, &svsk->sk_flags)); | |
637 | set_bit(SK_DATA, &svsk->sk_flags); | |
638 | svc_sock_enqueue(svsk); | |
639 | } | |
1da177e4 LT |
640 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
641 | wake_up_interruptible(sk->sk_sleep); | |
642 | } | |
643 | ||
644 | /* | |
645 | * INET callback when space is newly available on the socket. | |
646 | */ | |
647 | static void | |
648 | svc_write_space(struct sock *sk) | |
649 | { | |
650 | struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); | |
651 | ||
652 | if (svsk) { | |
653 | dprintk("svc: socket %p(inet %p), write_space busy=%d\n", | |
654 | svsk, sk, test_bit(SK_BUSY, &svsk->sk_flags)); | |
655 | svc_sock_enqueue(svsk); | |
656 | } | |
657 | ||
658 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { | |
939bb7ef | 659 | dprintk("RPC svc_write_space: someone sleeping on %p\n", |
1da177e4 LT |
660 | svsk); |
661 | wake_up_interruptible(sk->sk_sleep); | |
662 | } | |
663 | } | |
664 | ||
665 | /* | |
666 | * Receive a datagram from a UDP socket. | |
667 | */ | |
1da177e4 LT |
668 | static int |
669 | svc_udp_recvfrom(struct svc_rqst *rqstp) | |
670 | { | |
671 | struct svc_sock *svsk = rqstp->rq_sock; | |
672 | struct svc_serv *serv = svsk->sk_server; | |
673 | struct sk_buff *skb; | |
674 | int err, len; | |
675 | ||
676 | if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags)) | |
677 | /* udp sockets need large rcvbuf as all pending | |
678 | * requests are still in that buffer. sndbuf must | |
679 | * also be large enough that there is enough space | |
3262c816 GB |
680 | * for one reply per thread. We count all threads |
681 | * rather than threads in a particular pool, which | |
682 | * provides an upper bound on the number of threads | |
683 | * which will access the socket. | |
1da177e4 LT |
684 | */ |
685 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
686 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
687 | (serv->sv_nrthreads+3) * serv->sv_max_mesg); | |
1da177e4 LT |
688 | |
689 | if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) { | |
690 | svc_sock_received(svsk); | |
691 | return svc_deferred_recv(rqstp); | |
692 | } | |
693 | ||
aaf68cfb N |
694 | if (test_bit(SK_CLOSE, &svsk->sk_flags)) { |
695 | svc_delete_socket(svsk); | |
696 | return 0; | |
697 | } | |
698 | ||
1da177e4 LT |
699 | clear_bit(SK_DATA, &svsk->sk_flags); |
700 | while ((skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err)) == NULL) { | |
701 | if (err == -EAGAIN) { | |
702 | svc_sock_received(svsk); | |
703 | return err; | |
704 | } | |
705 | /* possibly an icmp error */ | |
706 | dprintk("svc: recvfrom returned error %d\n", -err); | |
707 | } | |
a61bbcf2 PM |
708 | if (skb->tstamp.off_sec == 0) { |
709 | struct timeval tv; | |
710 | ||
711 | tv.tv_sec = xtime.tv_sec; | |
4bcde03d | 712 | tv.tv_usec = xtime.tv_nsec / NSEC_PER_USEC; |
a61bbcf2 | 713 | skb_set_timestamp(skb, &tv); |
cca5172a | 714 | /* Don't enable netstamp, sunrpc doesn't |
1da177e4 LT |
715 | need that much accuracy */ |
716 | } | |
a61bbcf2 | 717 | skb_get_timestamp(skb, &svsk->sk_sk->sk_stamp); |
1da177e4 LT |
718 | set_bit(SK_DATA, &svsk->sk_flags); /* there may be more data... */ |
719 | ||
720 | /* | |
721 | * Maybe more packets - kick another thread ASAP. | |
722 | */ | |
723 | svc_sock_received(svsk); | |
724 | ||
725 | len = skb->len - sizeof(struct udphdr); | |
726 | rqstp->rq_arg.len = len; | |
727 | ||
728 | rqstp->rq_prot = IPPROTO_UDP; | |
729 | ||
730 | /* Get sender address */ | |
731 | rqstp->rq_addr.sin_family = AF_INET; | |
732 | rqstp->rq_addr.sin_port = skb->h.uh->source; | |
733 | rqstp->rq_addr.sin_addr.s_addr = skb->nh.iph->saddr; | |
734 | rqstp->rq_daddr = skb->nh.iph->daddr; | |
735 | ||
736 | if (skb_is_nonlinear(skb)) { | |
737 | /* we have to copy */ | |
738 | local_bh_disable(); | |
739 | if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { | |
740 | local_bh_enable(); | |
741 | /* checksum error */ | |
742 | skb_free_datagram(svsk->sk_sk, skb); | |
743 | return 0; | |
744 | } | |
745 | local_bh_enable(); | |
cca5172a | 746 | skb_free_datagram(svsk->sk_sk, skb); |
1da177e4 LT |
747 | } else { |
748 | /* we can use it in-place */ | |
749 | rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr); | |
750 | rqstp->rq_arg.head[0].iov_len = len; | |
fb286bb2 HX |
751 | if (skb_checksum_complete(skb)) { |
752 | skb_free_datagram(svsk->sk_sk, skb); | |
753 | return 0; | |
1da177e4 LT |
754 | } |
755 | rqstp->rq_skbuff = skb; | |
756 | } | |
757 | ||
758 | rqstp->rq_arg.page_base = 0; | |
759 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
760 | rqstp->rq_arg.head[0].iov_len = len; | |
761 | rqstp->rq_arg.page_len = 0; | |
44524359 | 762 | rqstp->rq_respages = rqstp->rq_pages+1; |
1da177e4 LT |
763 | } else { |
764 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
44524359 N |
765 | rqstp->rq_respages = rqstp->rq_pages + 1 + |
766 | (rqstp->rq_arg.page_len + PAGE_SIZE - 1)/ PAGE_SIZE; | |
1da177e4 LT |
767 | } |
768 | ||
769 | if (serv->sv_stats) | |
770 | serv->sv_stats->netudpcnt++; | |
771 | ||
772 | return len; | |
773 | } | |
774 | ||
775 | static int | |
776 | svc_udp_sendto(struct svc_rqst *rqstp) | |
777 | { | |
778 | int error; | |
779 | ||
780 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
781 | if (error == -ECONNREFUSED) | |
782 | /* ICMP error on earlier request. */ | |
783 | error = svc_sendto(rqstp, &rqstp->rq_res); | |
784 | ||
785 | return error; | |
786 | } | |
787 | ||
788 | static void | |
789 | svc_udp_init(struct svc_sock *svsk) | |
790 | { | |
791 | svsk->sk_sk->sk_data_ready = svc_udp_data_ready; | |
792 | svsk->sk_sk->sk_write_space = svc_write_space; | |
793 | svsk->sk_recvfrom = svc_udp_recvfrom; | |
794 | svsk->sk_sendto = svc_udp_sendto; | |
795 | ||
796 | /* initialise setting must have enough space to | |
cca5172a | 797 | * receive and respond to one request. |
1da177e4 LT |
798 | * svc_udp_recvfrom will re-adjust if necessary |
799 | */ | |
800 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
801 | 3 * svsk->sk_server->sv_max_mesg, |
802 | 3 * svsk->sk_server->sv_max_mesg); | |
1da177e4 LT |
803 | |
804 | set_bit(SK_DATA, &svsk->sk_flags); /* might have come in before data_ready set up */ | |
805 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
806 | } | |
807 | ||
808 | /* | |
809 | * A data_ready event on a listening socket means there's a connection | |
810 | * pending. Do not use state_change as a substitute for it. | |
811 | */ | |
812 | static void | |
813 | svc_tcp_listen_data_ready(struct sock *sk, int count_unused) | |
814 | { | |
939bb7ef | 815 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
816 | |
817 | dprintk("svc: socket %p TCP (listen) state change %d\n", | |
939bb7ef | 818 | sk, sk->sk_state); |
1da177e4 | 819 | |
939bb7ef NB |
820 | /* |
821 | * This callback may called twice when a new connection | |
822 | * is established as a child socket inherits everything | |
823 | * from a parent LISTEN socket. | |
824 | * 1) data_ready method of the parent socket will be called | |
825 | * when one of child sockets become ESTABLISHED. | |
826 | * 2) data_ready method of the child socket may be called | |
827 | * when it receives data before the socket is accepted. | |
828 | * In case of 2, we should ignore it silently. | |
829 | */ | |
830 | if (sk->sk_state == TCP_LISTEN) { | |
831 | if (svsk) { | |
832 | set_bit(SK_CONN, &svsk->sk_flags); | |
833 | svc_sock_enqueue(svsk); | |
834 | } else | |
835 | printk("svc: socket %p: no user data\n", sk); | |
1da177e4 | 836 | } |
939bb7ef | 837 | |
1da177e4 LT |
838 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
839 | wake_up_interruptible_all(sk->sk_sleep); | |
840 | } | |
841 | ||
842 | /* | |
843 | * A state change on a connected socket means it's dying or dead. | |
844 | */ | |
845 | static void | |
846 | svc_tcp_state_change(struct sock *sk) | |
847 | { | |
939bb7ef | 848 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
849 | |
850 | dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", | |
939bb7ef | 851 | sk, sk->sk_state, sk->sk_user_data); |
1da177e4 | 852 | |
939bb7ef | 853 | if (!svsk) |
1da177e4 | 854 | printk("svc: socket %p: no user data\n", sk); |
939bb7ef NB |
855 | else { |
856 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
857 | svc_sock_enqueue(svsk); | |
1da177e4 | 858 | } |
1da177e4 LT |
859 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
860 | wake_up_interruptible_all(sk->sk_sleep); | |
861 | } | |
862 | ||
863 | static void | |
864 | svc_tcp_data_ready(struct sock *sk, int count) | |
865 | { | |
939bb7ef | 866 | struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; |
1da177e4 LT |
867 | |
868 | dprintk("svc: socket %p TCP data ready (svsk %p)\n", | |
939bb7ef NB |
869 | sk, sk->sk_user_data); |
870 | if (svsk) { | |
871 | set_bit(SK_DATA, &svsk->sk_flags); | |
872 | svc_sock_enqueue(svsk); | |
873 | } | |
1da177e4 LT |
874 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) |
875 | wake_up_interruptible(sk->sk_sleep); | |
876 | } | |
877 | ||
878 | /* | |
879 | * Accept a TCP connection | |
880 | */ | |
881 | static void | |
882 | svc_tcp_accept(struct svc_sock *svsk) | |
883 | { | |
884 | struct sockaddr_in sin; | |
885 | struct svc_serv *serv = svsk->sk_server; | |
886 | struct socket *sock = svsk->sk_sock; | |
887 | struct socket *newsock; | |
1da177e4 LT |
888 | struct svc_sock *newsvsk; |
889 | int err, slen; | |
890 | ||
891 | dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); | |
892 | if (!sock) | |
893 | return; | |
894 | ||
e6242e92 SS |
895 | clear_bit(SK_CONN, &svsk->sk_flags); |
896 | err = kernel_accept(sock, &newsock, O_NONBLOCK); | |
897 | if (err < 0) { | |
1da177e4 LT |
898 | if (err == -ENOMEM) |
899 | printk(KERN_WARNING "%s: no more sockets!\n", | |
900 | serv->sv_name); | |
e6242e92 | 901 | else if (err != -EAGAIN && net_ratelimit()) |
1da177e4 LT |
902 | printk(KERN_WARNING "%s: accept failed (err %d)!\n", |
903 | serv->sv_name, -err); | |
e6242e92 | 904 | return; |
1da177e4 | 905 | } |
e6242e92 | 906 | |
1da177e4 LT |
907 | set_bit(SK_CONN, &svsk->sk_flags); |
908 | svc_sock_enqueue(svsk); | |
909 | ||
910 | slen = sizeof(sin); | |
e6242e92 | 911 | err = kernel_getpeername(newsock, (struct sockaddr *) &sin, &slen); |
1da177e4 LT |
912 | if (err < 0) { |
913 | if (net_ratelimit()) | |
914 | printk(KERN_WARNING "%s: peername failed (err %d)!\n", | |
915 | serv->sv_name, -err); | |
916 | goto failed; /* aborted connection or whatever */ | |
917 | } | |
918 | ||
919 | /* Ideally, we would want to reject connections from unauthorized | |
920 | * hosts here, but when we get encription, the IP of the host won't | |
921 | * tell us anything. For now just warn about unpriv connections. | |
922 | */ | |
923 | if (ntohs(sin.sin_port) >= 1024) { | |
924 | dprintk(KERN_WARNING | |
925 | "%s: connect from unprivileged port: %u.%u.%u.%u:%d\n", | |
cca5172a | 926 | serv->sv_name, |
1da177e4 LT |
927 | NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port)); |
928 | } | |
929 | ||
930 | dprintk("%s: connect from %u.%u.%u.%u:%04x\n", serv->sv_name, | |
931 | NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port)); | |
932 | ||
933 | /* make sure that a write doesn't block forever when | |
934 | * low on memory | |
935 | */ | |
936 | newsock->sk->sk_sndtimeo = HZ*30; | |
937 | ||
6b174337 CL |
938 | if (!(newsvsk = svc_setup_socket(serv, newsock, &err, |
939 | (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) | |
1da177e4 LT |
940 | goto failed; |
941 | ||
942 | ||
943 | /* make sure that we don't have too many active connections. | |
944 | * If we have, something must be dropped. | |
945 | * | |
946 | * There's no point in trying to do random drop here for | |
947 | * DoS prevention. The NFS clients does 1 reconnect in 15 | |
948 | * seconds. An attacker can easily beat that. | |
949 | * | |
950 | * The only somewhat efficient mechanism would be if drop | |
951 | * old connections from the same IP first. But right now | |
952 | * we don't even record the client IP in svc_sock. | |
953 | */ | |
954 | if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) { | |
955 | struct svc_sock *svsk = NULL; | |
956 | spin_lock_bh(&serv->sv_lock); | |
957 | if (!list_empty(&serv->sv_tempsocks)) { | |
958 | if (net_ratelimit()) { | |
959 | /* Try to help the admin */ | |
960 | printk(KERN_NOTICE "%s: too many open TCP " | |
961 | "sockets, consider increasing the " | |
962 | "number of nfsd threads\n", | |
963 | serv->sv_name); | |
964 | printk(KERN_NOTICE "%s: last TCP connect from " | |
965 | "%u.%u.%u.%u:%d\n", | |
966 | serv->sv_name, | |
967 | NIPQUAD(sin.sin_addr.s_addr), | |
968 | ntohs(sin.sin_port)); | |
969 | } | |
970 | /* | |
971 | * Always select the oldest socket. It's not fair, | |
972 | * but so is life | |
973 | */ | |
974 | svsk = list_entry(serv->sv_tempsocks.prev, | |
975 | struct svc_sock, | |
976 | sk_list); | |
977 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
c45c357d | 978 | atomic_inc(&svsk->sk_inuse); |
1da177e4 LT |
979 | } |
980 | spin_unlock_bh(&serv->sv_lock); | |
981 | ||
982 | if (svsk) { | |
983 | svc_sock_enqueue(svsk); | |
984 | svc_sock_put(svsk); | |
985 | } | |
986 | ||
987 | } | |
988 | ||
989 | if (serv->sv_stats) | |
990 | serv->sv_stats->nettcpconn++; | |
991 | ||
992 | return; | |
993 | ||
994 | failed: | |
995 | sock_release(newsock); | |
996 | return; | |
997 | } | |
998 | ||
999 | /* | |
1000 | * Receive data from a TCP socket. | |
1001 | */ | |
1002 | static int | |
1003 | svc_tcp_recvfrom(struct svc_rqst *rqstp) | |
1004 | { | |
1005 | struct svc_sock *svsk = rqstp->rq_sock; | |
1006 | struct svc_serv *serv = svsk->sk_server; | |
1007 | int len; | |
3cc03b16 | 1008 | struct kvec *vec; |
1da177e4 LT |
1009 | int pnum, vlen; |
1010 | ||
1011 | dprintk("svc: tcp_recv %p data %d conn %d close %d\n", | |
1012 | svsk, test_bit(SK_DATA, &svsk->sk_flags), | |
1013 | test_bit(SK_CONN, &svsk->sk_flags), | |
1014 | test_bit(SK_CLOSE, &svsk->sk_flags)); | |
1015 | ||
1016 | if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) { | |
1017 | svc_sock_received(svsk); | |
1018 | return svc_deferred_recv(rqstp); | |
1019 | } | |
1020 | ||
1021 | if (test_bit(SK_CLOSE, &svsk->sk_flags)) { | |
1022 | svc_delete_socket(svsk); | |
1023 | return 0; | |
1024 | } | |
1025 | ||
1a047060 | 1026 | if (svsk->sk_sk->sk_state == TCP_LISTEN) { |
1da177e4 LT |
1027 | svc_tcp_accept(svsk); |
1028 | svc_sock_received(svsk); | |
1029 | return 0; | |
1030 | } | |
1031 | ||
1032 | if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags)) | |
1033 | /* sndbuf needs to have room for one request | |
1034 | * per thread, otherwise we can stall even when the | |
1035 | * network isn't a bottleneck. | |
3262c816 GB |
1036 | * |
1037 | * We count all threads rather than threads in a | |
1038 | * particular pool, which provides an upper bound | |
1039 | * on the number of threads which will access the socket. | |
1040 | * | |
1da177e4 | 1041 | * rcvbuf just needs to be able to hold a few requests. |
cca5172a | 1042 | * Normally they will be removed from the queue |
1da177e4 LT |
1043 | * as soon a a complete request arrives. |
1044 | */ | |
1045 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
1046 | (serv->sv_nrthreads+3) * serv->sv_max_mesg, |
1047 | 3 * serv->sv_max_mesg); | |
1da177e4 LT |
1048 | |
1049 | clear_bit(SK_DATA, &svsk->sk_flags); | |
1050 | ||
1051 | /* Receive data. If we haven't got the record length yet, get | |
1052 | * the next four bytes. Otherwise try to gobble up as much as | |
1053 | * possible up to the complete record length. | |
1054 | */ | |
1055 | if (svsk->sk_tcplen < 4) { | |
1056 | unsigned long want = 4 - svsk->sk_tcplen; | |
1057 | struct kvec iov; | |
1058 | ||
1059 | iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; | |
1060 | iov.iov_len = want; | |
1061 | if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) | |
1062 | goto error; | |
1063 | svsk->sk_tcplen += len; | |
1064 | ||
1065 | if (len < want) { | |
1066 | dprintk("svc: short recvfrom while reading record length (%d of %lu)\n", | |
cca5172a | 1067 | len, want); |
1da177e4 LT |
1068 | svc_sock_received(svsk); |
1069 | return -EAGAIN; /* record header not complete */ | |
1070 | } | |
1071 | ||
1072 | svsk->sk_reclen = ntohl(svsk->sk_reclen); | |
1073 | if (!(svsk->sk_reclen & 0x80000000)) { | |
1074 | /* FIXME: technically, a record can be fragmented, | |
1075 | * and non-terminal fragments will not have the top | |
1076 | * bit set in the fragment length header. | |
1077 | * But apparently no known nfs clients send fragmented | |
1078 | * records. */ | |
34e9a63b N |
1079 | if (net_ratelimit()) |
1080 | printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx" | |
1081 | " (non-terminal)\n", | |
1082 | (unsigned long) svsk->sk_reclen); | |
1da177e4 LT |
1083 | goto err_delete; |
1084 | } | |
1085 | svsk->sk_reclen &= 0x7fffffff; | |
1086 | dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); | |
c6b0a9f8 | 1087 | if (svsk->sk_reclen > serv->sv_max_mesg) { |
34e9a63b N |
1088 | if (net_ratelimit()) |
1089 | printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx" | |
1090 | " (large)\n", | |
1091 | (unsigned long) svsk->sk_reclen); | |
1da177e4 LT |
1092 | goto err_delete; |
1093 | } | |
1094 | } | |
1095 | ||
1096 | /* Check whether enough data is available */ | |
1097 | len = svc_recv_available(svsk); | |
1098 | if (len < 0) | |
1099 | goto error; | |
1100 | ||
1101 | if (len < svsk->sk_reclen) { | |
1102 | dprintk("svc: incomplete TCP record (%d of %d)\n", | |
1103 | len, svsk->sk_reclen); | |
1104 | svc_sock_received(svsk); | |
1105 | return -EAGAIN; /* record not complete */ | |
1106 | } | |
1107 | len = svsk->sk_reclen; | |
1108 | set_bit(SK_DATA, &svsk->sk_flags); | |
1109 | ||
3cc03b16 | 1110 | vec = rqstp->rq_vec; |
1da177e4 LT |
1111 | vec[0] = rqstp->rq_arg.head[0]; |
1112 | vlen = PAGE_SIZE; | |
1113 | pnum = 1; | |
1114 | while (vlen < len) { | |
44524359 | 1115 | vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]); |
1da177e4 LT |
1116 | vec[pnum].iov_len = PAGE_SIZE; |
1117 | pnum++; | |
1118 | vlen += PAGE_SIZE; | |
1119 | } | |
44524359 | 1120 | rqstp->rq_respages = &rqstp->rq_pages[pnum]; |
1da177e4 LT |
1121 | |
1122 | /* Now receive data */ | |
1123 | len = svc_recvfrom(rqstp, vec, pnum, len); | |
1124 | if (len < 0) | |
1125 | goto error; | |
1126 | ||
1127 | dprintk("svc: TCP complete record (%d bytes)\n", len); | |
1128 | rqstp->rq_arg.len = len; | |
1129 | rqstp->rq_arg.page_base = 0; | |
1130 | if (len <= rqstp->rq_arg.head[0].iov_len) { | |
1131 | rqstp->rq_arg.head[0].iov_len = len; | |
1132 | rqstp->rq_arg.page_len = 0; | |
1133 | } else { | |
1134 | rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; | |
1135 | } | |
1136 | ||
1137 | rqstp->rq_skbuff = NULL; | |
1138 | rqstp->rq_prot = IPPROTO_TCP; | |
1139 | ||
1140 | /* Reset TCP read info */ | |
1141 | svsk->sk_reclen = 0; | |
1142 | svsk->sk_tcplen = 0; | |
1143 | ||
1144 | svc_sock_received(svsk); | |
1145 | if (serv->sv_stats) | |
1146 | serv->sv_stats->nettcpcnt++; | |
1147 | ||
1148 | return len; | |
1149 | ||
1150 | err_delete: | |
1151 | svc_delete_socket(svsk); | |
1152 | return -EAGAIN; | |
1153 | ||
1154 | error: | |
1155 | if (len == -EAGAIN) { | |
1156 | dprintk("RPC: TCP recvfrom got EAGAIN\n"); | |
1157 | svc_sock_received(svsk); | |
1158 | } else { | |
1159 | printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", | |
1160 | svsk->sk_server->sv_name, -len); | |
93fbf1a5 | 1161 | goto err_delete; |
1da177e4 LT |
1162 | } |
1163 | ||
1164 | return len; | |
1165 | } | |
1166 | ||
1167 | /* | |
1168 | * Send out data on TCP socket. | |
1169 | */ | |
1170 | static int | |
1171 | svc_tcp_sendto(struct svc_rqst *rqstp) | |
1172 | { | |
1173 | struct xdr_buf *xbufp = &rqstp->rq_res; | |
1174 | int sent; | |
d8ed029d | 1175 | __be32 reclen; |
1da177e4 LT |
1176 | |
1177 | /* Set up the first element of the reply kvec. | |
1178 | * Any other kvecs that may be in use have been taken | |
1179 | * care of by the server implementation itself. | |
1180 | */ | |
1181 | reclen = htonl(0x80000000|((xbufp->len ) - 4)); | |
1182 | memcpy(xbufp->head[0].iov_base, &reclen, 4); | |
1183 | ||
1184 | if (test_bit(SK_DEAD, &rqstp->rq_sock->sk_flags)) | |
1185 | return -ENOTCONN; | |
1186 | ||
1187 | sent = svc_sendto(rqstp, &rqstp->rq_res); | |
1188 | if (sent != xbufp->len) { | |
1189 | printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n", | |
1190 | rqstp->rq_sock->sk_server->sv_name, | |
1191 | (sent<0)?"got error":"sent only", | |
1192 | sent, xbufp->len); | |
aaf68cfb N |
1193 | set_bit(SK_CLOSE, &rqstp->rq_sock->sk_flags); |
1194 | svc_sock_enqueue(rqstp->rq_sock); | |
1da177e4 LT |
1195 | sent = -EAGAIN; |
1196 | } | |
1197 | return sent; | |
1198 | } | |
1199 | ||
1200 | static void | |
1201 | svc_tcp_init(struct svc_sock *svsk) | |
1202 | { | |
1203 | struct sock *sk = svsk->sk_sk; | |
1204 | struct tcp_sock *tp = tcp_sk(sk); | |
1205 | ||
1206 | svsk->sk_recvfrom = svc_tcp_recvfrom; | |
1207 | svsk->sk_sendto = svc_tcp_sendto; | |
1208 | ||
1209 | if (sk->sk_state == TCP_LISTEN) { | |
1210 | dprintk("setting up TCP socket for listening\n"); | |
1211 | sk->sk_data_ready = svc_tcp_listen_data_ready; | |
1212 | set_bit(SK_CONN, &svsk->sk_flags); | |
1213 | } else { | |
1214 | dprintk("setting up TCP socket for reading\n"); | |
1215 | sk->sk_state_change = svc_tcp_state_change; | |
1216 | sk->sk_data_ready = svc_tcp_data_ready; | |
1217 | sk->sk_write_space = svc_write_space; | |
1218 | ||
1219 | svsk->sk_reclen = 0; | |
1220 | svsk->sk_tcplen = 0; | |
1221 | ||
1222 | tp->nonagle = 1; /* disable Nagle's algorithm */ | |
1223 | ||
1224 | /* initialise setting must have enough space to | |
cca5172a | 1225 | * receive and respond to one request. |
1da177e4 LT |
1226 | * svc_tcp_recvfrom will re-adjust if necessary |
1227 | */ | |
1228 | svc_sock_setbufsize(svsk->sk_sock, | |
c6b0a9f8 N |
1229 | 3 * svsk->sk_server->sv_max_mesg, |
1230 | 3 * svsk->sk_server->sv_max_mesg); | |
1da177e4 LT |
1231 | |
1232 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1233 | set_bit(SK_DATA, &svsk->sk_flags); | |
cca5172a | 1234 | if (sk->sk_state != TCP_ESTABLISHED) |
1da177e4 LT |
1235 | set_bit(SK_CLOSE, &svsk->sk_flags); |
1236 | } | |
1237 | } | |
1238 | ||
1239 | void | |
1240 | svc_sock_update_bufs(struct svc_serv *serv) | |
1241 | { | |
1242 | /* | |
1243 | * The number of server threads has changed. Update | |
1244 | * rcvbuf and sndbuf accordingly on all sockets | |
1245 | */ | |
1246 | struct list_head *le; | |
1247 | ||
1248 | spin_lock_bh(&serv->sv_lock); | |
1249 | list_for_each(le, &serv->sv_permsocks) { | |
cca5172a | 1250 | struct svc_sock *svsk = |
1da177e4 LT |
1251 | list_entry(le, struct svc_sock, sk_list); |
1252 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1253 | } | |
1254 | list_for_each(le, &serv->sv_tempsocks) { | |
1255 | struct svc_sock *svsk = | |
1256 | list_entry(le, struct svc_sock, sk_list); | |
1257 | set_bit(SK_CHNGBUF, &svsk->sk_flags); | |
1258 | } | |
1259 | spin_unlock_bh(&serv->sv_lock); | |
1260 | } | |
1261 | ||
1262 | /* | |
3262c816 GB |
1263 | * Receive the next request on any socket. This code is carefully |
1264 | * organised not to touch any cachelines in the shared svc_serv | |
1265 | * structure, only cachelines in the local svc_pool. | |
1da177e4 LT |
1266 | */ |
1267 | int | |
6fb2b47f | 1268 | svc_recv(struct svc_rqst *rqstp, long timeout) |
1da177e4 LT |
1269 | { |
1270 | struct svc_sock *svsk =NULL; | |
6fb2b47f | 1271 | struct svc_serv *serv = rqstp->rq_server; |
3262c816 | 1272 | struct svc_pool *pool = rqstp->rq_pool; |
44524359 | 1273 | int len, i; |
1da177e4 LT |
1274 | int pages; |
1275 | struct xdr_buf *arg; | |
1276 | DECLARE_WAITQUEUE(wait, current); | |
1277 | ||
1278 | dprintk("svc: server %p waiting for data (to = %ld)\n", | |
1279 | rqstp, timeout); | |
1280 | ||
1281 | if (rqstp->rq_sock) | |
cca5172a | 1282 | printk(KERN_ERR |
1da177e4 LT |
1283 | "svc_recv: service %p, socket not NULL!\n", |
1284 | rqstp); | |
1285 | if (waitqueue_active(&rqstp->rq_wait)) | |
cca5172a | 1286 | printk(KERN_ERR |
1da177e4 LT |
1287 | "svc_recv: service %p, wait queue active!\n", |
1288 | rqstp); | |
1289 | ||
1da177e4 LT |
1290 | |
1291 | /* now allocate needed pages. If we get a failure, sleep briefly */ | |
c6b0a9f8 | 1292 | pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE; |
44524359 N |
1293 | for (i=0; i < pages ; i++) |
1294 | while (rqstp->rq_pages[i] == NULL) { | |
1295 | struct page *p = alloc_page(GFP_KERNEL); | |
1296 | if (!p) | |
1297 | schedule_timeout_uninterruptible(msecs_to_jiffies(500)); | |
1298 | rqstp->rq_pages[i] = p; | |
1da177e4 | 1299 | } |
250f3915 N |
1300 | rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */ |
1301 | BUG_ON(pages >= RPCSVC_MAXPAGES); | |
1da177e4 LT |
1302 | |
1303 | /* Make arg->head point to first page and arg->pages point to rest */ | |
1304 | arg = &rqstp->rq_arg; | |
44524359 | 1305 | arg->head[0].iov_base = page_address(rqstp->rq_pages[0]); |
1da177e4 | 1306 | arg->head[0].iov_len = PAGE_SIZE; |
44524359 | 1307 | arg->pages = rqstp->rq_pages + 1; |
1da177e4 LT |
1308 | arg->page_base = 0; |
1309 | /* save at least one page for response */ | |
1310 | arg->page_len = (pages-2)*PAGE_SIZE; | |
1311 | arg->len = (pages-1)*PAGE_SIZE; | |
1312 | arg->tail[0].iov_len = 0; | |
3e1d1d28 CL |
1313 | |
1314 | try_to_freeze(); | |
1887b935 | 1315 | cond_resched(); |
1da177e4 LT |
1316 | if (signalled()) |
1317 | return -EINTR; | |
1318 | ||
3262c816 GB |
1319 | spin_lock_bh(&pool->sp_lock); |
1320 | if ((svsk = svc_sock_dequeue(pool)) != NULL) { | |
1da177e4 | 1321 | rqstp->rq_sock = svsk; |
c45c357d | 1322 | atomic_inc(&svsk->sk_inuse); |
c6b0a9f8 | 1323 | rqstp->rq_reserved = serv->sv_max_mesg; |
5685f0fa | 1324 | atomic_add(rqstp->rq_reserved, &svsk->sk_reserved); |
1da177e4 LT |
1325 | } else { |
1326 | /* No data pending. Go to sleep */ | |
3262c816 | 1327 | svc_thread_enqueue(pool, rqstp); |
1da177e4 LT |
1328 | |
1329 | /* | |
1330 | * We have to be able to interrupt this wait | |
1331 | * to bring down the daemons ... | |
1332 | */ | |
1333 | set_current_state(TASK_INTERRUPTIBLE); | |
1334 | add_wait_queue(&rqstp->rq_wait, &wait); | |
3262c816 | 1335 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 LT |
1336 | |
1337 | schedule_timeout(timeout); | |
1338 | ||
3e1d1d28 | 1339 | try_to_freeze(); |
1da177e4 | 1340 | |
3262c816 | 1341 | spin_lock_bh(&pool->sp_lock); |
1da177e4 LT |
1342 | remove_wait_queue(&rqstp->rq_wait, &wait); |
1343 | ||
1344 | if (!(svsk = rqstp->rq_sock)) { | |
3262c816 GB |
1345 | svc_thread_dequeue(pool, rqstp); |
1346 | spin_unlock_bh(&pool->sp_lock); | |
1da177e4 LT |
1347 | dprintk("svc: server %p, no data yet\n", rqstp); |
1348 | return signalled()? -EINTR : -EAGAIN; | |
1349 | } | |
1350 | } | |
3262c816 | 1351 | spin_unlock_bh(&pool->sp_lock); |
1da177e4 | 1352 | |
3262c816 GB |
1353 | dprintk("svc: server %p, pool %u, socket %p, inuse=%d\n", |
1354 | rqstp, pool->sp_id, svsk, atomic_read(&svsk->sk_inuse)); | |
1da177e4 LT |
1355 | len = svsk->sk_recvfrom(rqstp); |
1356 | dprintk("svc: got len=%d\n", len); | |
1357 | ||
1358 | /* No data, incomplete (TCP) read, or accept() */ | |
1359 | if (len == 0 || len == -EAGAIN) { | |
1360 | rqstp->rq_res.len = 0; | |
1361 | svc_sock_release(rqstp); | |
1362 | return -EAGAIN; | |
1363 | } | |
1364 | svsk->sk_lastrecv = get_seconds(); | |
36bdfc8b | 1365 | clear_bit(SK_OLD, &svsk->sk_flags); |
1da177e4 LT |
1366 | |
1367 | rqstp->rq_secure = ntohs(rqstp->rq_addr.sin_port) < 1024; | |
1368 | rqstp->rq_chandle.defer = svc_defer; | |
1369 | ||
1370 | if (serv->sv_stats) | |
1371 | serv->sv_stats->netcnt++; | |
1372 | return len; | |
1373 | } | |
1374 | ||
cca5172a | 1375 | /* |
1da177e4 LT |
1376 | * Drop request |
1377 | */ | |
1378 | void | |
1379 | svc_drop(struct svc_rqst *rqstp) | |
1380 | { | |
1381 | dprintk("svc: socket %p dropped request\n", rqstp->rq_sock); | |
1382 | svc_sock_release(rqstp); | |
1383 | } | |
1384 | ||
1385 | /* | |
1386 | * Return reply to client. | |
1387 | */ | |
1388 | int | |
1389 | svc_send(struct svc_rqst *rqstp) | |
1390 | { | |
1391 | struct svc_sock *svsk; | |
1392 | int len; | |
1393 | struct xdr_buf *xb; | |
1394 | ||
1395 | if ((svsk = rqstp->rq_sock) == NULL) { | |
1396 | printk(KERN_WARNING "NULL socket pointer in %s:%d\n", | |
1397 | __FILE__, __LINE__); | |
1398 | return -EFAULT; | |
1399 | } | |
1400 | ||
1401 | /* release the receive skb before sending the reply */ | |
1402 | svc_release_skb(rqstp); | |
1403 | ||
1404 | /* calculate over-all length */ | |
1405 | xb = & rqstp->rq_res; | |
1406 | xb->len = xb->head[0].iov_len + | |
1407 | xb->page_len + | |
1408 | xb->tail[0].iov_len; | |
1409 | ||
57b47a53 IM |
1410 | /* Grab svsk->sk_mutex to serialize outgoing data. */ |
1411 | mutex_lock(&svsk->sk_mutex); | |
1da177e4 LT |
1412 | if (test_bit(SK_DEAD, &svsk->sk_flags)) |
1413 | len = -ENOTCONN; | |
1414 | else | |
1415 | len = svsk->sk_sendto(rqstp); | |
57b47a53 | 1416 | mutex_unlock(&svsk->sk_mutex); |
1da177e4 LT |
1417 | svc_sock_release(rqstp); |
1418 | ||
1419 | if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN) | |
1420 | return 0; | |
1421 | return len; | |
1422 | } | |
1423 | ||
36bdfc8b GB |
1424 | /* |
1425 | * Timer function to close old temporary sockets, using | |
1426 | * a mark-and-sweep algorithm. | |
1427 | */ | |
1428 | static void | |
1429 | svc_age_temp_sockets(unsigned long closure) | |
1430 | { | |
1431 | struct svc_serv *serv = (struct svc_serv *)closure; | |
1432 | struct svc_sock *svsk; | |
1433 | struct list_head *le, *next; | |
1434 | LIST_HEAD(to_be_aged); | |
1435 | ||
1436 | dprintk("svc_age_temp_sockets\n"); | |
1437 | ||
1438 | if (!spin_trylock_bh(&serv->sv_lock)) { | |
1439 | /* busy, try again 1 sec later */ | |
1440 | dprintk("svc_age_temp_sockets: busy\n"); | |
1441 | mod_timer(&serv->sv_temptimer, jiffies + HZ); | |
1442 | return; | |
1443 | } | |
1444 | ||
1445 | list_for_each_safe(le, next, &serv->sv_tempsocks) { | |
1446 | svsk = list_entry(le, struct svc_sock, sk_list); | |
1447 | ||
1448 | if (!test_and_set_bit(SK_OLD, &svsk->sk_flags)) | |
1449 | continue; | |
c45c357d | 1450 | if (atomic_read(&svsk->sk_inuse) || test_bit(SK_BUSY, &svsk->sk_flags)) |
36bdfc8b | 1451 | continue; |
c45c357d | 1452 | atomic_inc(&svsk->sk_inuse); |
36bdfc8b GB |
1453 | list_move(le, &to_be_aged); |
1454 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1455 | set_bit(SK_DETACHED, &svsk->sk_flags); | |
1456 | } | |
1457 | spin_unlock_bh(&serv->sv_lock); | |
1458 | ||
1459 | while (!list_empty(&to_be_aged)) { | |
1460 | le = to_be_aged.next; | |
1461 | /* fiddling the sk_list node is safe 'cos we're SK_DETACHED */ | |
1462 | list_del_init(le); | |
1463 | svsk = list_entry(le, struct svc_sock, sk_list); | |
1464 | ||
1465 | dprintk("queuing svsk %p for closing, %lu seconds old\n", | |
1466 | svsk, get_seconds() - svsk->sk_lastrecv); | |
1467 | ||
1468 | /* a thread will dequeue and close it soon */ | |
1469 | svc_sock_enqueue(svsk); | |
1470 | svc_sock_put(svsk); | |
1471 | } | |
1472 | ||
1473 | mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ); | |
1474 | } | |
1475 | ||
1da177e4 LT |
1476 | /* |
1477 | * Initialize socket for RPC use and create svc_sock struct | |
1478 | * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. | |
1479 | */ | |
6b174337 CL |
1480 | static struct svc_sock *svc_setup_socket(struct svc_serv *serv, |
1481 | struct socket *sock, | |
1482 | int *errp, int flags) | |
1da177e4 LT |
1483 | { |
1484 | struct svc_sock *svsk; | |
1485 | struct sock *inet; | |
6b174337 CL |
1486 | int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); |
1487 | int is_temporary = flags & SVC_SOCK_TEMPORARY; | |
1da177e4 LT |
1488 | |
1489 | dprintk("svc: svc_setup_socket %p\n", sock); | |
0da974f4 | 1490 | if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { |
1da177e4 LT |
1491 | *errp = -ENOMEM; |
1492 | return NULL; | |
1493 | } | |
1da177e4 LT |
1494 | |
1495 | inet = sock->sk; | |
1496 | ||
1497 | /* Register socket with portmapper */ | |
1498 | if (*errp >= 0 && pmap_register) | |
1499 | *errp = svc_register(serv, inet->sk_protocol, | |
1500 | ntohs(inet_sk(inet)->sport)); | |
1501 | ||
1502 | if (*errp < 0) { | |
1503 | kfree(svsk); | |
1504 | return NULL; | |
1505 | } | |
1506 | ||
1507 | set_bit(SK_BUSY, &svsk->sk_flags); | |
1508 | inet->sk_user_data = svsk; | |
1509 | svsk->sk_sock = sock; | |
1510 | svsk->sk_sk = inet; | |
1511 | svsk->sk_ostate = inet->sk_state_change; | |
1512 | svsk->sk_odata = inet->sk_data_ready; | |
1513 | svsk->sk_owspace = inet->sk_write_space; | |
1514 | svsk->sk_server = serv; | |
aaf68cfb | 1515 | atomic_set(&svsk->sk_inuse, 1); |
1da177e4 | 1516 | svsk->sk_lastrecv = get_seconds(); |
1a68d952 | 1517 | spin_lock_init(&svsk->sk_defer_lock); |
1da177e4 LT |
1518 | INIT_LIST_HEAD(&svsk->sk_deferred); |
1519 | INIT_LIST_HEAD(&svsk->sk_ready); | |
57b47a53 | 1520 | mutex_init(&svsk->sk_mutex); |
1da177e4 LT |
1521 | |
1522 | /* Initialize the socket */ | |
1523 | if (sock->type == SOCK_DGRAM) | |
1524 | svc_udp_init(svsk); | |
1525 | else | |
1526 | svc_tcp_init(svsk); | |
1527 | ||
1528 | spin_lock_bh(&serv->sv_lock); | |
6b174337 | 1529 | if (is_temporary) { |
1da177e4 LT |
1530 | set_bit(SK_TEMP, &svsk->sk_flags); |
1531 | list_add(&svsk->sk_list, &serv->sv_tempsocks); | |
1532 | serv->sv_tmpcnt++; | |
36bdfc8b GB |
1533 | if (serv->sv_temptimer.function == NULL) { |
1534 | /* setup timer to age temp sockets */ | |
1535 | setup_timer(&serv->sv_temptimer, svc_age_temp_sockets, | |
1536 | (unsigned long)serv); | |
1537 | mod_timer(&serv->sv_temptimer, | |
1538 | jiffies + svc_conn_age_period * HZ); | |
1539 | } | |
1da177e4 LT |
1540 | } else { |
1541 | clear_bit(SK_TEMP, &svsk->sk_flags); | |
1542 | list_add(&svsk->sk_list, &serv->sv_permsocks); | |
1543 | } | |
1544 | spin_unlock_bh(&serv->sv_lock); | |
1545 | ||
1546 | dprintk("svc: svc_setup_socket created %p (inet %p)\n", | |
1547 | svsk, svsk->sk_sk); | |
1548 | ||
1549 | clear_bit(SK_BUSY, &svsk->sk_flags); | |
1550 | svc_sock_enqueue(svsk); | |
1551 | return svsk; | |
1552 | } | |
1553 | ||
b41b66d6 N |
1554 | int svc_addsock(struct svc_serv *serv, |
1555 | int fd, | |
1556 | char *name_return, | |
1557 | int *proto) | |
1558 | { | |
1559 | int err = 0; | |
1560 | struct socket *so = sockfd_lookup(fd, &err); | |
1561 | struct svc_sock *svsk = NULL; | |
1562 | ||
1563 | if (!so) | |
1564 | return err; | |
1565 | if (so->sk->sk_family != AF_INET) | |
1566 | err = -EAFNOSUPPORT; | |
1567 | else if (so->sk->sk_protocol != IPPROTO_TCP && | |
1568 | so->sk->sk_protocol != IPPROTO_UDP) | |
1569 | err = -EPROTONOSUPPORT; | |
1570 | else if (so->state > SS_UNCONNECTED) | |
1571 | err = -EISCONN; | |
1572 | else { | |
6b174337 | 1573 | svsk = svc_setup_socket(serv, so, &err, SVC_SOCK_DEFAULTS); |
b41b66d6 N |
1574 | if (svsk) |
1575 | err = 0; | |
1576 | } | |
1577 | if (err) { | |
1578 | sockfd_put(so); | |
1579 | return err; | |
1580 | } | |
1581 | if (proto) *proto = so->sk->sk_protocol; | |
1582 | return one_sock_name(name_return, svsk); | |
1583 | } | |
1584 | EXPORT_SYMBOL_GPL(svc_addsock); | |
1585 | ||
1da177e4 LT |
1586 | /* |
1587 | * Create socket for RPC service. | |
1588 | */ | |
6b174337 CL |
1589 | static int svc_create_socket(struct svc_serv *serv, int protocol, |
1590 | struct sockaddr_in *sin, int flags) | |
1da177e4 LT |
1591 | { |
1592 | struct svc_sock *svsk; | |
1593 | struct socket *sock; | |
1594 | int error; | |
1595 | int type; | |
1596 | ||
1597 | dprintk("svc: svc_create_socket(%s, %d, %u.%u.%u.%u:%d)\n", | |
1598 | serv->sv_program->pg_name, protocol, | |
1599 | NIPQUAD(sin->sin_addr.s_addr), | |
1600 | ntohs(sin->sin_port)); | |
1601 | ||
1602 | if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { | |
1603 | printk(KERN_WARNING "svc: only UDP and TCP " | |
1604 | "sockets supported\n"); | |
1605 | return -EINVAL; | |
1606 | } | |
1607 | type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; | |
1608 | ||
1609 | if ((error = sock_create_kern(PF_INET, type, protocol, &sock)) < 0) | |
1610 | return error; | |
1611 | ||
ed07536e PZ |
1612 | svc_reclassify_socket(sock); |
1613 | ||
18114746 ES |
1614 | if (type == SOCK_STREAM) |
1615 | sock->sk->sk_reuse = 1; /* allow address reuse */ | |
1616 | error = kernel_bind(sock, (struct sockaddr *) sin, | |
1617 | sizeof(*sin)); | |
1618 | if (error < 0) | |
1619 | goto bummer; | |
1da177e4 LT |
1620 | |
1621 | if (protocol == IPPROTO_TCP) { | |
e6242e92 | 1622 | if ((error = kernel_listen(sock, 64)) < 0) |
1da177e4 LT |
1623 | goto bummer; |
1624 | } | |
1625 | ||
6b174337 CL |
1626 | if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) |
1627 | return ntohs(inet_sk(svsk->sk_sk)->sport); | |
1da177e4 LT |
1628 | |
1629 | bummer: | |
1630 | dprintk("svc: svc_create_socket error = %d\n", -error); | |
1631 | sock_release(sock); | |
1632 | return error; | |
1633 | } | |
1634 | ||
1635 | /* | |
1636 | * Remove a dead socket | |
1637 | */ | |
aaf68cfb | 1638 | static void |
1da177e4 LT |
1639 | svc_delete_socket(struct svc_sock *svsk) |
1640 | { | |
1641 | struct svc_serv *serv; | |
1642 | struct sock *sk; | |
1643 | ||
1644 | dprintk("svc: svc_delete_socket(%p)\n", svsk); | |
1645 | ||
1646 | serv = svsk->sk_server; | |
1647 | sk = svsk->sk_sk; | |
1648 | ||
1649 | sk->sk_state_change = svsk->sk_ostate; | |
1650 | sk->sk_data_ready = svsk->sk_odata; | |
1651 | sk->sk_write_space = svsk->sk_owspace; | |
1652 | ||
1653 | spin_lock_bh(&serv->sv_lock); | |
1654 | ||
36bdfc8b GB |
1655 | if (!test_and_set_bit(SK_DETACHED, &svsk->sk_flags)) |
1656 | list_del_init(&svsk->sk_list); | |
cca5172a | 1657 | /* |
3262c816 GB |
1658 | * We used to delete the svc_sock from whichever list |
1659 | * it's sk_ready node was on, but we don't actually | |
1660 | * need to. This is because the only time we're called | |
1661 | * while still attached to a queue, the queue itself | |
1662 | * is about to be destroyed (in svc_destroy). | |
1663 | */ | |
aaf68cfb N |
1664 | if (!test_and_set_bit(SK_DEAD, &svsk->sk_flags)) { |
1665 | BUG_ON(atomic_read(&svsk->sk_inuse)<2); | |
1666 | atomic_dec(&svsk->sk_inuse); | |
1da177e4 LT |
1667 | if (test_bit(SK_TEMP, &svsk->sk_flags)) |
1668 | serv->sv_tmpcnt--; | |
aaf68cfb | 1669 | } |
1da177e4 | 1670 | |
d6740df9 | 1671 | spin_unlock_bh(&serv->sv_lock); |
aaf68cfb N |
1672 | } |
1673 | ||
1674 | void svc_close_socket(struct svc_sock *svsk) | |
1675 | { | |
1676 | set_bit(SK_CLOSE, &svsk->sk_flags); | |
1677 | if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) | |
1678 | /* someone else will have to effect the close */ | |
1679 | return; | |
1680 | ||
1681 | atomic_inc(&svsk->sk_inuse); | |
1682 | svc_delete_socket(svsk); | |
1683 | clear_bit(SK_BUSY, &svsk->sk_flags); | |
d6740df9 | 1684 | svc_sock_put(svsk); |
1da177e4 LT |
1685 | } |
1686 | ||
6b174337 CL |
1687 | /** |
1688 | * svc_makesock - Make a socket for nfsd and lockd | |
1689 | * @serv: RPC server structure | |
1690 | * @protocol: transport protocol to use | |
1691 | * @port: port to use | |
482fb94e | 1692 | * @flags: requested socket characteristics |
6b174337 | 1693 | * |
1da177e4 | 1694 | */ |
482fb94e CL |
1695 | int svc_makesock(struct svc_serv *serv, int protocol, unsigned short port, |
1696 | int flags) | |
1da177e4 | 1697 | { |
6b174337 CL |
1698 | struct sockaddr_in sin = { |
1699 | .sin_family = AF_INET, | |
1700 | .sin_addr.s_addr = INADDR_ANY, | |
1701 | .sin_port = htons(port), | |
1702 | }; | |
1da177e4 LT |
1703 | |
1704 | dprintk("svc: creating socket proto = %d\n", protocol); | |
482fb94e | 1705 | return svc_create_socket(serv, protocol, &sin, flags); |
1da177e4 LT |
1706 | } |
1707 | ||
1708 | /* | |
cca5172a | 1709 | * Handle defer and revisit of requests |
1da177e4 LT |
1710 | */ |
1711 | ||
1712 | static void svc_revisit(struct cache_deferred_req *dreq, int too_many) | |
1713 | { | |
1714 | struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle); | |
1da177e4 LT |
1715 | struct svc_sock *svsk; |
1716 | ||
1717 | if (too_many) { | |
1718 | svc_sock_put(dr->svsk); | |
1719 | kfree(dr); | |
1720 | return; | |
1721 | } | |
1722 | dprintk("revisit queued\n"); | |
1723 | svsk = dr->svsk; | |
1724 | dr->svsk = NULL; | |
1a68d952 | 1725 | spin_lock_bh(&svsk->sk_defer_lock); |
1da177e4 | 1726 | list_add(&dr->handle.recent, &svsk->sk_deferred); |
1a68d952 | 1727 | spin_unlock_bh(&svsk->sk_defer_lock); |
1da177e4 LT |
1728 | set_bit(SK_DEFERRED, &svsk->sk_flags); |
1729 | svc_sock_enqueue(svsk); | |
1730 | svc_sock_put(svsk); | |
1731 | } | |
1732 | ||
1733 | static struct cache_deferred_req * | |
1734 | svc_defer(struct cache_req *req) | |
1735 | { | |
1736 | struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle); | |
1737 | int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len); | |
1738 | struct svc_deferred_req *dr; | |
1739 | ||
1740 | if (rqstp->rq_arg.page_len) | |
1741 | return NULL; /* if more than a page, give up FIXME */ | |
1742 | if (rqstp->rq_deferred) { | |
1743 | dr = rqstp->rq_deferred; | |
1744 | rqstp->rq_deferred = NULL; | |
1745 | } else { | |
1746 | int skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; | |
1747 | /* FIXME maybe discard if size too large */ | |
1748 | dr = kmalloc(size, GFP_KERNEL); | |
1749 | if (dr == NULL) | |
1750 | return NULL; | |
1751 | ||
1752 | dr->handle.owner = rqstp->rq_server; | |
1753 | dr->prot = rqstp->rq_prot; | |
1754 | dr->addr = rqstp->rq_addr; | |
1918e341 | 1755 | dr->daddr = rqstp->rq_daddr; |
1da177e4 LT |
1756 | dr->argslen = rqstp->rq_arg.len >> 2; |
1757 | memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2); | |
1758 | } | |
c45c357d | 1759 | atomic_inc(&rqstp->rq_sock->sk_inuse); |
1da177e4 | 1760 | dr->svsk = rqstp->rq_sock; |
1da177e4 LT |
1761 | |
1762 | dr->handle.revisit = svc_revisit; | |
1763 | return &dr->handle; | |
1764 | } | |
1765 | ||
1766 | /* | |
1767 | * recv data from a deferred request into an active one | |
1768 | */ | |
1769 | static int svc_deferred_recv(struct svc_rqst *rqstp) | |
1770 | { | |
1771 | struct svc_deferred_req *dr = rqstp->rq_deferred; | |
1772 | ||
1773 | rqstp->rq_arg.head[0].iov_base = dr->args; | |
1774 | rqstp->rq_arg.head[0].iov_len = dr->argslen<<2; | |
1775 | rqstp->rq_arg.page_len = 0; | |
1776 | rqstp->rq_arg.len = dr->argslen<<2; | |
1777 | rqstp->rq_prot = dr->prot; | |
1778 | rqstp->rq_addr = dr->addr; | |
1918e341 | 1779 | rqstp->rq_daddr = dr->daddr; |
44524359 | 1780 | rqstp->rq_respages = rqstp->rq_pages; |
1da177e4 LT |
1781 | return dr->argslen<<2; |
1782 | } | |
1783 | ||
1784 | ||
1785 | static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk) | |
1786 | { | |
1787 | struct svc_deferred_req *dr = NULL; | |
cca5172a | 1788 | |
1da177e4 LT |
1789 | if (!test_bit(SK_DEFERRED, &svsk->sk_flags)) |
1790 | return NULL; | |
1a68d952 | 1791 | spin_lock_bh(&svsk->sk_defer_lock); |
1da177e4 LT |
1792 | clear_bit(SK_DEFERRED, &svsk->sk_flags); |
1793 | if (!list_empty(&svsk->sk_deferred)) { | |
1794 | dr = list_entry(svsk->sk_deferred.next, | |
1795 | struct svc_deferred_req, | |
1796 | handle.recent); | |
1797 | list_del_init(&dr->handle.recent); | |
1798 | set_bit(SK_DEFERRED, &svsk->sk_flags); | |
1799 | } | |
1a68d952 | 1800 | spin_unlock_bh(&svsk->sk_defer_lock); |
1da177e4 LT |
1801 | return dr; |
1802 | } |