blob: 51885b5f744e938eaa71ffcb8617759469a68f7b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
Andrew Morton91483c42005-08-09 20:20:07 -070029#include <linux/tcp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/unistd.h>
31#include <linux/slab.h>
32#include <linux/netdevice.h>
33#include <linux/skbuff.h>
34#include <net/sock.h>
35#include <net/checksum.h>
36#include <net/ip.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070037#include <net/tcp_states.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <asm/uaccess.h>
39#include <asm/ioctls.h>
40
41#include <linux/sunrpc/types.h>
42#include <linux/sunrpc/xdr.h>
43#include <linux/sunrpc/svcsock.h>
44#include <linux/sunrpc/stats.h>
45
46/* SMP locking strategy:
47 *
48 * svc_serv->sv_lock protects most stuff for that service.
49 *
50 * Some flags can be set to certain values at any time
51 * providing that certain rules are followed:
52 *
53 * SK_BUSY can be set to 0 at any time.
54 * svc_sock_enqueue must be called afterwards
55 * SK_CONN, SK_DATA, can be set or cleared at any time.
56 * after a set, svc_sock_enqueue must be called.
57 * after a clear, the socket must be read/accepted
58 * if this succeeds, it must be set again.
59 * SK_CLOSE can set at any time. It is never cleared.
60 *
61 */
62
63#define RPCDBG_FACILITY RPCDBG_SVCSOCK
64
65
66static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
67 int *errp, int pmap_reg);
68static void svc_udp_data_ready(struct sock *, int);
69static int svc_udp_recvfrom(struct svc_rqst *);
70static int svc_udp_sendto(struct svc_rqst *);
71
72static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk);
73static int svc_deferred_recv(struct svc_rqst *rqstp);
74static struct cache_deferred_req *svc_defer(struct cache_req *req);
75
76/*
77 * Queue up an idle server thread. Must have serv->sv_lock held.
78 * Note: this is really a stack rather than a queue, so that we only
79 * use as many different threads as we need, and the rest don't polute
80 * the cache.
81 */
82static inline void
83svc_serv_enqueue(struct svc_serv *serv, struct svc_rqst *rqstp)
84{
85 list_add(&rqstp->rq_list, &serv->sv_threads);
86}
87
88/*
89 * Dequeue an nfsd thread. Must have serv->sv_lock held.
90 */
91static inline void
92svc_serv_dequeue(struct svc_serv *serv, struct svc_rqst *rqstp)
93{
94 list_del(&rqstp->rq_list);
95}
96
97/*
98 * Release an skbuff after use
99 */
100static inline void
101svc_release_skb(struct svc_rqst *rqstp)
102{
103 struct sk_buff *skb = rqstp->rq_skbuff;
104 struct svc_deferred_req *dr = rqstp->rq_deferred;
105
106 if (skb) {
107 rqstp->rq_skbuff = NULL;
108
109 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
110 skb_free_datagram(rqstp->rq_sock->sk_sk, skb);
111 }
112 if (dr) {
113 rqstp->rq_deferred = NULL;
114 kfree(dr);
115 }
116}
117
118/*
119 * Any space to write?
120 */
121static inline unsigned long
122svc_sock_wspace(struct svc_sock *svsk)
123{
124 int wspace;
125
126 if (svsk->sk_sock->type == SOCK_STREAM)
127 wspace = sk_stream_wspace(svsk->sk_sk);
128 else
129 wspace = sock_wspace(svsk->sk_sk);
130
131 return wspace;
132}
133
134/*
135 * Queue up a socket with data pending. If there are idle nfsd
136 * processes, wake 'em up.
137 *
138 */
139static void
140svc_sock_enqueue(struct svc_sock *svsk)
141{
142 struct svc_serv *serv = svsk->sk_server;
143 struct svc_rqst *rqstp;
144
145 if (!(svsk->sk_flags &
146 ( (1<<SK_CONN)|(1<<SK_DATA)|(1<<SK_CLOSE)|(1<<SK_DEFERRED)) ))
147 return;
148 if (test_bit(SK_DEAD, &svsk->sk_flags))
149 return;
150
151 spin_lock_bh(&serv->sv_lock);
152
153 if (!list_empty(&serv->sv_threads) &&
154 !list_empty(&serv->sv_sockets))
155 printk(KERN_ERR
156 "svc_sock_enqueue: threads and sockets both waiting??\n");
157
158 if (test_bit(SK_DEAD, &svsk->sk_flags)) {
159 /* Don't enqueue dead sockets */
160 dprintk("svc: socket %p is dead, not enqueued\n", svsk->sk_sk);
161 goto out_unlock;
162 }
163
164 if (test_bit(SK_BUSY, &svsk->sk_flags)) {
165 /* Don't enqueue socket while daemon is receiving */
166 dprintk("svc: socket %p busy, not enqueued\n", svsk->sk_sk);
167 goto out_unlock;
168 }
169
170 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
171 if (((svsk->sk_reserved + serv->sv_bufsz)*2
172 > svc_sock_wspace(svsk))
173 && !test_bit(SK_CLOSE, &svsk->sk_flags)
174 && !test_bit(SK_CONN, &svsk->sk_flags)) {
175 /* Don't enqueue while not enough space for reply */
176 dprintk("svc: socket %p no space, %d*2 > %ld, not enqueued\n",
177 svsk->sk_sk, svsk->sk_reserved+serv->sv_bufsz,
178 svc_sock_wspace(svsk));
179 goto out_unlock;
180 }
181 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
182
183 /* Mark socket as busy. It will remain in this state until the
184 * server has processed all pending data and put the socket back
185 * on the idle list.
186 */
187 set_bit(SK_BUSY, &svsk->sk_flags);
188
189 if (!list_empty(&serv->sv_threads)) {
190 rqstp = list_entry(serv->sv_threads.next,
191 struct svc_rqst,
192 rq_list);
193 dprintk("svc: socket %p served by daemon %p\n",
194 svsk->sk_sk, rqstp);
195 svc_serv_dequeue(serv, rqstp);
196 if (rqstp->rq_sock)
197 printk(KERN_ERR
198 "svc_sock_enqueue: server %p, rq_sock=%p!\n",
199 rqstp, rqstp->rq_sock);
200 rqstp->rq_sock = svsk;
201 svsk->sk_inuse++;
202 rqstp->rq_reserved = serv->sv_bufsz;
203 svsk->sk_reserved += rqstp->rq_reserved;
204 wake_up(&rqstp->rq_wait);
205 } else {
206 dprintk("svc: socket %p put into queue\n", svsk->sk_sk);
207 list_add_tail(&svsk->sk_ready, &serv->sv_sockets);
208 }
209
210out_unlock:
211 spin_unlock_bh(&serv->sv_lock);
212}
213
214/*
215 * Dequeue the first socket. Must be called with the serv->sv_lock held.
216 */
217static inline struct svc_sock *
218svc_sock_dequeue(struct svc_serv *serv)
219{
220 struct svc_sock *svsk;
221
222 if (list_empty(&serv->sv_sockets))
223 return NULL;
224
225 svsk = list_entry(serv->sv_sockets.next,
226 struct svc_sock, sk_ready);
227 list_del_init(&svsk->sk_ready);
228
229 dprintk("svc: socket %p dequeued, inuse=%d\n",
230 svsk->sk_sk, svsk->sk_inuse);
231
232 return svsk;
233}
234
235/*
236 * Having read something from a socket, check whether it
237 * needs to be re-enqueued.
238 * Note: SK_DATA only gets cleared when a read-attempt finds
239 * no (or insufficient) data.
240 */
241static inline void
242svc_sock_received(struct svc_sock *svsk)
243{
244 clear_bit(SK_BUSY, &svsk->sk_flags);
245 svc_sock_enqueue(svsk);
246}
247
248
249/**
250 * svc_reserve - change the space reserved for the reply to a request.
251 * @rqstp: The request in question
252 * @space: new max space to reserve
253 *
254 * Each request reserves some space on the output queue of the socket
255 * to make sure the reply fits. This function reduces that reserved
256 * space to be the amount of space used already, plus @space.
257 *
258 */
259void svc_reserve(struct svc_rqst *rqstp, int space)
260{
261 space += rqstp->rq_res.head[0].iov_len;
262
263 if (space < rqstp->rq_reserved) {
264 struct svc_sock *svsk = rqstp->rq_sock;
265 spin_lock_bh(&svsk->sk_server->sv_lock);
266 svsk->sk_reserved -= (rqstp->rq_reserved - space);
267 rqstp->rq_reserved = space;
268 spin_unlock_bh(&svsk->sk_server->sv_lock);
269
270 svc_sock_enqueue(svsk);
271 }
272}
273
274/*
275 * Release a socket after use.
276 */
277static inline void
278svc_sock_put(struct svc_sock *svsk)
279{
280 struct svc_serv *serv = svsk->sk_server;
281
282 spin_lock_bh(&serv->sv_lock);
283 if (!--(svsk->sk_inuse) && test_bit(SK_DEAD, &svsk->sk_flags)) {
284 spin_unlock_bh(&serv->sv_lock);
285 dprintk("svc: releasing dead socket\n");
286 sock_release(svsk->sk_sock);
287 kfree(svsk);
288 }
289 else
290 spin_unlock_bh(&serv->sv_lock);
291}
292
293static void
294svc_sock_release(struct svc_rqst *rqstp)
295{
296 struct svc_sock *svsk = rqstp->rq_sock;
297
298 svc_release_skb(rqstp);
299
300 svc_free_allpages(rqstp);
301 rqstp->rq_res.page_len = 0;
302 rqstp->rq_res.page_base = 0;
303
304
305 /* Reset response buffer and release
306 * the reservation.
307 * But first, check that enough space was reserved
308 * for the reply, otherwise we have a bug!
309 */
310 if ((rqstp->rq_res.len) > rqstp->rq_reserved)
311 printk(KERN_ERR "RPC request reserved %d but used %d\n",
312 rqstp->rq_reserved,
313 rqstp->rq_res.len);
314
315 rqstp->rq_res.head[0].iov_len = 0;
316 svc_reserve(rqstp, 0);
317 rqstp->rq_sock = NULL;
318
319 svc_sock_put(svsk);
320}
321
322/*
323 * External function to wake up a server waiting for data
324 */
325void
326svc_wake_up(struct svc_serv *serv)
327{
328 struct svc_rqst *rqstp;
329
330 spin_lock_bh(&serv->sv_lock);
331 if (!list_empty(&serv->sv_threads)) {
332 rqstp = list_entry(serv->sv_threads.next,
333 struct svc_rqst,
334 rq_list);
335 dprintk("svc: daemon %p woken up.\n", rqstp);
336 /*
337 svc_serv_dequeue(serv, rqstp);
338 rqstp->rq_sock = NULL;
339 */
340 wake_up(&rqstp->rq_wait);
341 }
342 spin_unlock_bh(&serv->sv_lock);
343}
344
345/*
346 * Generic sendto routine
347 */
348static int
349svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
350{
351 struct svc_sock *svsk = rqstp->rq_sock;
352 struct socket *sock = svsk->sk_sock;
353 int slen;
354 char buffer[CMSG_SPACE(sizeof(struct in_pktinfo))];
355 struct cmsghdr *cmh = (struct cmsghdr *)buffer;
356 struct in_pktinfo *pki = (struct in_pktinfo *)CMSG_DATA(cmh);
357 int len = 0;
358 int result;
359 int size;
360 struct page **ppage = xdr->pages;
361 size_t base = xdr->page_base;
362 unsigned int pglen = xdr->page_len;
363 unsigned int flags = MSG_MORE;
364
365 slen = xdr->len;
366
367 if (rqstp->rq_prot == IPPROTO_UDP) {
368 /* set the source and destination */
369 struct msghdr msg;
370 msg.msg_name = &rqstp->rq_addr;
371 msg.msg_namelen = sizeof(rqstp->rq_addr);
372 msg.msg_iov = NULL;
373 msg.msg_iovlen = 0;
374 msg.msg_flags = MSG_MORE;
375
376 msg.msg_control = cmh;
377 msg.msg_controllen = sizeof(buffer);
378 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
379 cmh->cmsg_level = SOL_IP;
380 cmh->cmsg_type = IP_PKTINFO;
381 pki->ipi_ifindex = 0;
382 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr;
383
384 if (sock_sendmsg(sock, &msg, 0) < 0)
385 goto out;
386 }
387
388 /* send head */
389 if (slen == xdr->head[0].iov_len)
390 flags = 0;
391 len = sock->ops->sendpage(sock, rqstp->rq_respages[0], 0, xdr->head[0].iov_len, flags);
392 if (len != xdr->head[0].iov_len)
393 goto out;
394 slen -= xdr->head[0].iov_len;
395 if (slen == 0)
396 goto out;
397
398 /* send page data */
399 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
400 while (pglen > 0) {
401 if (slen == size)
402 flags = 0;
403 result = sock->ops->sendpage(sock, *ppage, base, size, flags);
404 if (result > 0)
405 len += result;
406 if (result != size)
407 goto out;
408 slen -= size;
409 pglen -= size;
410 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
411 base = 0;
412 ppage++;
413 }
414 /* send tail */
415 if (xdr->tail[0].iov_len) {
416 result = sock->ops->sendpage(sock, rqstp->rq_respages[rqstp->rq_restailpage],
417 ((unsigned long)xdr->tail[0].iov_base)& (PAGE_SIZE-1),
418 xdr->tail[0].iov_len, 0);
419
420 if (result > 0)
421 len += result;
422 }
423out:
424 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %x)\n",
425 rqstp->rq_sock, xdr->head[0].iov_base, xdr->head[0].iov_len, xdr->len, len,
426 rqstp->rq_addr.sin_addr.s_addr);
427
428 return len;
429}
430
431/*
432 * Check input queue length
433 */
434static int
435svc_recv_available(struct svc_sock *svsk)
436{
437 mm_segment_t oldfs;
438 struct socket *sock = svsk->sk_sock;
439 int avail, err;
440
441 oldfs = get_fs(); set_fs(KERNEL_DS);
442 err = sock->ops->ioctl(sock, TIOCINQ, (unsigned long) &avail);
443 set_fs(oldfs);
444
445 return (err >= 0)? avail : err;
446}
447
448/*
449 * Generic recvfrom routine.
450 */
451static int
452svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, int buflen)
453{
454 struct msghdr msg;
455 struct socket *sock;
456 int len, alen;
457
458 rqstp->rq_addrlen = sizeof(rqstp->rq_addr);
459 sock = rqstp->rq_sock->sk_sock;
460
461 msg.msg_name = &rqstp->rq_addr;
462 msg.msg_namelen = sizeof(rqstp->rq_addr);
463 msg.msg_control = NULL;
464 msg.msg_controllen = 0;
465
466 msg.msg_flags = MSG_DONTWAIT;
467
468 len = kernel_recvmsg(sock, &msg, iov, nr, buflen, MSG_DONTWAIT);
469
470 /* sock_recvmsg doesn't fill in the name/namelen, so we must..
471 * possibly we should cache this in the svc_sock structure
472 * at accept time. FIXME
473 */
474 alen = sizeof(rqstp->rq_addr);
475 sock->ops->getname(sock, (struct sockaddr *)&rqstp->rq_addr, &alen, 1);
476
477 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
478 rqstp->rq_sock, iov[0].iov_base, iov[0].iov_len, len);
479
480 return len;
481}
482
483/*
484 * Set socket snd and rcv buffer lengths
485 */
486static inline void
487svc_sock_setbufsize(struct socket *sock, unsigned int snd, unsigned int rcv)
488{
489#if 0
490 mm_segment_t oldfs;
491 oldfs = get_fs(); set_fs(KERNEL_DS);
492 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
493 (char*)&snd, sizeof(snd));
494 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
495 (char*)&rcv, sizeof(rcv));
496#else
497 /* sock_setsockopt limits use to sysctl_?mem_max,
498 * which isn't acceptable. Until that is made conditional
499 * on not having CAP_SYS_RESOURCE or similar, we go direct...
500 * DaveM said I could!
501 */
502 lock_sock(sock->sk);
503 sock->sk->sk_sndbuf = snd * 2;
504 sock->sk->sk_rcvbuf = rcv * 2;
505 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
506 release_sock(sock->sk);
507#endif
508}
509/*
510 * INET callback when data has been received on the socket.
511 */
512static void
513svc_udp_data_ready(struct sock *sk, int count)
514{
515 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
516
517 if (!svsk)
518 goto out;
519 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
520 svsk, sk, count, test_bit(SK_BUSY, &svsk->sk_flags));
521 set_bit(SK_DATA, &svsk->sk_flags);
522 svc_sock_enqueue(svsk);
523 out:
524 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
525 wake_up_interruptible(sk->sk_sleep);
526}
527
528/*
529 * INET callback when space is newly available on the socket.
530 */
531static void
532svc_write_space(struct sock *sk)
533{
534 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
535
536 if (svsk) {
537 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
538 svsk, sk, test_bit(SK_BUSY, &svsk->sk_flags));
539 svc_sock_enqueue(svsk);
540 }
541
542 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) {
543 printk(KERN_WARNING "RPC svc_write_space: some sleeping on %p\n",
544 svsk);
545 wake_up_interruptible(sk->sk_sleep);
546 }
547}
548
549/*
550 * Receive a datagram from a UDP socket.
551 */
552extern int
553csum_partial_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb);
554
555static int
556svc_udp_recvfrom(struct svc_rqst *rqstp)
557{
558 struct svc_sock *svsk = rqstp->rq_sock;
559 struct svc_serv *serv = svsk->sk_server;
560 struct sk_buff *skb;
561 int err, len;
562
563 if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags))
564 /* udp sockets need large rcvbuf as all pending
565 * requests are still in that buffer. sndbuf must
566 * also be large enough that there is enough space
567 * for one reply per thread.
568 */
569 svc_sock_setbufsize(svsk->sk_sock,
570 (serv->sv_nrthreads+3) * serv->sv_bufsz,
571 (serv->sv_nrthreads+3) * serv->sv_bufsz);
572
573 if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
574 svc_sock_received(svsk);
575 return svc_deferred_recv(rqstp);
576 }
577
578 clear_bit(SK_DATA, &svsk->sk_flags);
579 while ((skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err)) == NULL) {
580 if (err == -EAGAIN) {
581 svc_sock_received(svsk);
582 return err;
583 }
584 /* possibly an icmp error */
585 dprintk("svc: recvfrom returned error %d\n", -err);
586 }
Patrick McHardya61bbcf2005-08-14 17:24:31 -0700587 if (skb->tstamp.off_sec == 0) {
588 struct timeval tv;
589
590 tv.tv_sec = xtime.tv_sec;
591 tv.tv_usec = xtime.tv_nsec * 1000;
592 skb_set_timestamp(skb, &tv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 /* Don't enable netstamp, sunrpc doesn't
594 need that much accuracy */
595 }
Patrick McHardya61bbcf2005-08-14 17:24:31 -0700596 skb_get_timestamp(skb, &svsk->sk_sk->sk_stamp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 set_bit(SK_DATA, &svsk->sk_flags); /* there may be more data... */
598
599 /*
600 * Maybe more packets - kick another thread ASAP.
601 */
602 svc_sock_received(svsk);
603
604 len = skb->len - sizeof(struct udphdr);
605 rqstp->rq_arg.len = len;
606
607 rqstp->rq_prot = IPPROTO_UDP;
608
609 /* Get sender address */
610 rqstp->rq_addr.sin_family = AF_INET;
611 rqstp->rq_addr.sin_port = skb->h.uh->source;
612 rqstp->rq_addr.sin_addr.s_addr = skb->nh.iph->saddr;
613 rqstp->rq_daddr = skb->nh.iph->daddr;
614
615 if (skb_is_nonlinear(skb)) {
616 /* we have to copy */
617 local_bh_disable();
618 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
619 local_bh_enable();
620 /* checksum error */
621 skb_free_datagram(svsk->sk_sk, skb);
622 return 0;
623 }
624 local_bh_enable();
625 skb_free_datagram(svsk->sk_sk, skb);
626 } else {
627 /* we can use it in-place */
628 rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr);
629 rqstp->rq_arg.head[0].iov_len = len;
630 if (skb->ip_summed != CHECKSUM_UNNECESSARY) {
631 if ((unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum))) {
632 skb_free_datagram(svsk->sk_sk, skb);
633 return 0;
634 }
635 skb->ip_summed = CHECKSUM_UNNECESSARY;
636 }
637 rqstp->rq_skbuff = skb;
638 }
639
640 rqstp->rq_arg.page_base = 0;
641 if (len <= rqstp->rq_arg.head[0].iov_len) {
642 rqstp->rq_arg.head[0].iov_len = len;
643 rqstp->rq_arg.page_len = 0;
644 } else {
645 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
646 rqstp->rq_argused += (rqstp->rq_arg.page_len + PAGE_SIZE - 1)/ PAGE_SIZE;
647 }
648
649 if (serv->sv_stats)
650 serv->sv_stats->netudpcnt++;
651
652 return len;
653}
654
655static int
656svc_udp_sendto(struct svc_rqst *rqstp)
657{
658 int error;
659
660 error = svc_sendto(rqstp, &rqstp->rq_res);
661 if (error == -ECONNREFUSED)
662 /* ICMP error on earlier request. */
663 error = svc_sendto(rqstp, &rqstp->rq_res);
664
665 return error;
666}
667
668static void
669svc_udp_init(struct svc_sock *svsk)
670{
671 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
672 svsk->sk_sk->sk_write_space = svc_write_space;
673 svsk->sk_recvfrom = svc_udp_recvfrom;
674 svsk->sk_sendto = svc_udp_sendto;
675
676 /* initialise setting must have enough space to
677 * receive and respond to one request.
678 * svc_udp_recvfrom will re-adjust if necessary
679 */
680 svc_sock_setbufsize(svsk->sk_sock,
681 3 * svsk->sk_server->sv_bufsz,
682 3 * svsk->sk_server->sv_bufsz);
683
684 set_bit(SK_DATA, &svsk->sk_flags); /* might have come in before data_ready set up */
685 set_bit(SK_CHNGBUF, &svsk->sk_flags);
686}
687
688/*
689 * A data_ready event on a listening socket means there's a connection
690 * pending. Do not use state_change as a substitute for it.
691 */
692static void
693svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
694{
695 struct svc_sock *svsk;
696
697 dprintk("svc: socket %p TCP (listen) state change %d\n",
698 sk, sk->sk_state);
699
700 if (sk->sk_state != TCP_LISTEN) {
701 /*
702 * This callback may called twice when a new connection
703 * is established as a child socket inherits everything
704 * from a parent LISTEN socket.
705 * 1) data_ready method of the parent socket will be called
706 * when one of child sockets become ESTABLISHED.
707 * 2) data_ready method of the child socket may be called
708 * when it receives data before the socket is accepted.
709 * In case of 2, we should ignore it silently.
710 */
711 goto out;
712 }
713 if (!(svsk = (struct svc_sock *) sk->sk_user_data)) {
714 printk("svc: socket %p: no user data\n", sk);
715 goto out;
716 }
717 set_bit(SK_CONN, &svsk->sk_flags);
718 svc_sock_enqueue(svsk);
719 out:
720 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
721 wake_up_interruptible_all(sk->sk_sleep);
722}
723
724/*
725 * A state change on a connected socket means it's dying or dead.
726 */
727static void
728svc_tcp_state_change(struct sock *sk)
729{
730 struct svc_sock *svsk;
731
732 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
733 sk, sk->sk_state, sk->sk_user_data);
734
735 if (!(svsk = (struct svc_sock *) sk->sk_user_data)) {
736 printk("svc: socket %p: no user data\n", sk);
737 goto out;
738 }
739 set_bit(SK_CLOSE, &svsk->sk_flags);
740 svc_sock_enqueue(svsk);
741 out:
742 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
743 wake_up_interruptible_all(sk->sk_sleep);
744}
745
746static void
747svc_tcp_data_ready(struct sock *sk, int count)
748{
749 struct svc_sock * svsk;
750
751 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
752 sk, sk->sk_user_data);
753 if (!(svsk = (struct svc_sock *)(sk->sk_user_data)))
754 goto out;
755 set_bit(SK_DATA, &svsk->sk_flags);
756 svc_sock_enqueue(svsk);
757 out:
758 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
759 wake_up_interruptible(sk->sk_sleep);
760}
761
762/*
763 * Accept a TCP connection
764 */
765static void
766svc_tcp_accept(struct svc_sock *svsk)
767{
768 struct sockaddr_in sin;
769 struct svc_serv *serv = svsk->sk_server;
770 struct socket *sock = svsk->sk_sock;
771 struct socket *newsock;
772 struct proto_ops *ops;
773 struct svc_sock *newsvsk;
774 int err, slen;
775
776 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
777 if (!sock)
778 return;
779
780 err = sock_create_lite(PF_INET, SOCK_STREAM, IPPROTO_TCP, &newsock);
781 if (err) {
782 if (err == -ENOMEM)
783 printk(KERN_WARNING "%s: no more sockets!\n",
784 serv->sv_name);
785 return;
786 }
787
788 dprintk("svc: tcp_accept %p allocated\n", newsock);
789 newsock->ops = ops = sock->ops;
790
791 clear_bit(SK_CONN, &svsk->sk_flags);
792 if ((err = ops->accept(sock, newsock, O_NONBLOCK)) < 0) {
793 if (err != -EAGAIN && net_ratelimit())
794 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
795 serv->sv_name, -err);
796 goto failed; /* aborted connection or whatever */
797 }
798 set_bit(SK_CONN, &svsk->sk_flags);
799 svc_sock_enqueue(svsk);
800
801 slen = sizeof(sin);
802 err = ops->getname(newsock, (struct sockaddr *) &sin, &slen, 1);
803 if (err < 0) {
804 if (net_ratelimit())
805 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
806 serv->sv_name, -err);
807 goto failed; /* aborted connection or whatever */
808 }
809
810 /* Ideally, we would want to reject connections from unauthorized
811 * hosts here, but when we get encription, the IP of the host won't
812 * tell us anything. For now just warn about unpriv connections.
813 */
814 if (ntohs(sin.sin_port) >= 1024) {
815 dprintk(KERN_WARNING
816 "%s: connect from unprivileged port: %u.%u.%u.%u:%d\n",
817 serv->sv_name,
818 NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port));
819 }
820
821 dprintk("%s: connect from %u.%u.%u.%u:%04x\n", serv->sv_name,
822 NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port));
823
824 /* make sure that a write doesn't block forever when
825 * low on memory
826 */
827 newsock->sk->sk_sndtimeo = HZ*30;
828
829 if (!(newsvsk = svc_setup_socket(serv, newsock, &err, 0)))
830 goto failed;
831
832
833 /* make sure that we don't have too many active connections.
834 * If we have, something must be dropped.
835 *
836 * There's no point in trying to do random drop here for
837 * DoS prevention. The NFS clients does 1 reconnect in 15
838 * seconds. An attacker can easily beat that.
839 *
840 * The only somewhat efficient mechanism would be if drop
841 * old connections from the same IP first. But right now
842 * we don't even record the client IP in svc_sock.
843 */
844 if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) {
845 struct svc_sock *svsk = NULL;
846 spin_lock_bh(&serv->sv_lock);
847 if (!list_empty(&serv->sv_tempsocks)) {
848 if (net_ratelimit()) {
849 /* Try to help the admin */
850 printk(KERN_NOTICE "%s: too many open TCP "
851 "sockets, consider increasing the "
852 "number of nfsd threads\n",
853 serv->sv_name);
854 printk(KERN_NOTICE "%s: last TCP connect from "
855 "%u.%u.%u.%u:%d\n",
856 serv->sv_name,
857 NIPQUAD(sin.sin_addr.s_addr),
858 ntohs(sin.sin_port));
859 }
860 /*
861 * Always select the oldest socket. It's not fair,
862 * but so is life
863 */
864 svsk = list_entry(serv->sv_tempsocks.prev,
865 struct svc_sock,
866 sk_list);
867 set_bit(SK_CLOSE, &svsk->sk_flags);
868 svsk->sk_inuse ++;
869 }
870 spin_unlock_bh(&serv->sv_lock);
871
872 if (svsk) {
873 svc_sock_enqueue(svsk);
874 svc_sock_put(svsk);
875 }
876
877 }
878
879 if (serv->sv_stats)
880 serv->sv_stats->nettcpconn++;
881
882 return;
883
884failed:
885 sock_release(newsock);
886 return;
887}
888
889/*
890 * Receive data from a TCP socket.
891 */
892static int
893svc_tcp_recvfrom(struct svc_rqst *rqstp)
894{
895 struct svc_sock *svsk = rqstp->rq_sock;
896 struct svc_serv *serv = svsk->sk_server;
897 int len;
898 struct kvec vec[RPCSVC_MAXPAGES];
899 int pnum, vlen;
900
901 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
902 svsk, test_bit(SK_DATA, &svsk->sk_flags),
903 test_bit(SK_CONN, &svsk->sk_flags),
904 test_bit(SK_CLOSE, &svsk->sk_flags));
905
906 if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
907 svc_sock_received(svsk);
908 return svc_deferred_recv(rqstp);
909 }
910
911 if (test_bit(SK_CLOSE, &svsk->sk_flags)) {
912 svc_delete_socket(svsk);
913 return 0;
914 }
915
916 if (test_bit(SK_CONN, &svsk->sk_flags)) {
917 svc_tcp_accept(svsk);
918 svc_sock_received(svsk);
919 return 0;
920 }
921
922 if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags))
923 /* sndbuf needs to have room for one request
924 * per thread, otherwise we can stall even when the
925 * network isn't a bottleneck.
926 * rcvbuf just needs to be able to hold a few requests.
927 * Normally they will be removed from the queue
928 * as soon a a complete request arrives.
929 */
930 svc_sock_setbufsize(svsk->sk_sock,
931 (serv->sv_nrthreads+3) * serv->sv_bufsz,
932 3 * serv->sv_bufsz);
933
934 clear_bit(SK_DATA, &svsk->sk_flags);
935
936 /* Receive data. If we haven't got the record length yet, get
937 * the next four bytes. Otherwise try to gobble up as much as
938 * possible up to the complete record length.
939 */
940 if (svsk->sk_tcplen < 4) {
941 unsigned long want = 4 - svsk->sk_tcplen;
942 struct kvec iov;
943
944 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
945 iov.iov_len = want;
946 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
947 goto error;
948 svsk->sk_tcplen += len;
949
950 if (len < want) {
951 dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
952 len, want);
953 svc_sock_received(svsk);
954 return -EAGAIN; /* record header not complete */
955 }
956
957 svsk->sk_reclen = ntohl(svsk->sk_reclen);
958 if (!(svsk->sk_reclen & 0x80000000)) {
959 /* FIXME: technically, a record can be fragmented,
960 * and non-terminal fragments will not have the top
961 * bit set in the fragment length header.
962 * But apparently no known nfs clients send fragmented
963 * records. */
964 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx (non-terminal)\n",
965 (unsigned long) svsk->sk_reclen);
966 goto err_delete;
967 }
968 svsk->sk_reclen &= 0x7fffffff;
969 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
970 if (svsk->sk_reclen > serv->sv_bufsz) {
971 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx (large)\n",
972 (unsigned long) svsk->sk_reclen);
973 goto err_delete;
974 }
975 }
976
977 /* Check whether enough data is available */
978 len = svc_recv_available(svsk);
979 if (len < 0)
980 goto error;
981
982 if (len < svsk->sk_reclen) {
983 dprintk("svc: incomplete TCP record (%d of %d)\n",
984 len, svsk->sk_reclen);
985 svc_sock_received(svsk);
986 return -EAGAIN; /* record not complete */
987 }
988 len = svsk->sk_reclen;
989 set_bit(SK_DATA, &svsk->sk_flags);
990
991 vec[0] = rqstp->rq_arg.head[0];
992 vlen = PAGE_SIZE;
993 pnum = 1;
994 while (vlen < len) {
995 vec[pnum].iov_base = page_address(rqstp->rq_argpages[rqstp->rq_argused++]);
996 vec[pnum].iov_len = PAGE_SIZE;
997 pnum++;
998 vlen += PAGE_SIZE;
999 }
1000
1001 /* Now receive data */
1002 len = svc_recvfrom(rqstp, vec, pnum, len);
1003 if (len < 0)
1004 goto error;
1005
1006 dprintk("svc: TCP complete record (%d bytes)\n", len);
1007 rqstp->rq_arg.len = len;
1008 rqstp->rq_arg.page_base = 0;
1009 if (len <= rqstp->rq_arg.head[0].iov_len) {
1010 rqstp->rq_arg.head[0].iov_len = len;
1011 rqstp->rq_arg.page_len = 0;
1012 } else {
1013 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
1014 }
1015
1016 rqstp->rq_skbuff = NULL;
1017 rqstp->rq_prot = IPPROTO_TCP;
1018
1019 /* Reset TCP read info */
1020 svsk->sk_reclen = 0;
1021 svsk->sk_tcplen = 0;
1022
1023 svc_sock_received(svsk);
1024 if (serv->sv_stats)
1025 serv->sv_stats->nettcpcnt++;
1026
1027 return len;
1028
1029 err_delete:
1030 svc_delete_socket(svsk);
1031 return -EAGAIN;
1032
1033 error:
1034 if (len == -EAGAIN) {
1035 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1036 svc_sock_received(svsk);
1037 } else {
1038 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1039 svsk->sk_server->sv_name, -len);
1040 svc_sock_received(svsk);
1041 }
1042
1043 return len;
1044}
1045
1046/*
1047 * Send out data on TCP socket.
1048 */
1049static int
1050svc_tcp_sendto(struct svc_rqst *rqstp)
1051{
1052 struct xdr_buf *xbufp = &rqstp->rq_res;
1053 int sent;
1054 u32 reclen;
1055
1056 /* Set up the first element of the reply kvec.
1057 * Any other kvecs that may be in use have been taken
1058 * care of by the server implementation itself.
1059 */
1060 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1061 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1062
1063 if (test_bit(SK_DEAD, &rqstp->rq_sock->sk_flags))
1064 return -ENOTCONN;
1065
1066 sent = svc_sendto(rqstp, &rqstp->rq_res);
1067 if (sent != xbufp->len) {
1068 printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1069 rqstp->rq_sock->sk_server->sv_name,
1070 (sent<0)?"got error":"sent only",
1071 sent, xbufp->len);
1072 svc_delete_socket(rqstp->rq_sock);
1073 sent = -EAGAIN;
1074 }
1075 return sent;
1076}
1077
1078static void
1079svc_tcp_init(struct svc_sock *svsk)
1080{
1081 struct sock *sk = svsk->sk_sk;
1082 struct tcp_sock *tp = tcp_sk(sk);
1083
1084 svsk->sk_recvfrom = svc_tcp_recvfrom;
1085 svsk->sk_sendto = svc_tcp_sendto;
1086
1087 if (sk->sk_state == TCP_LISTEN) {
1088 dprintk("setting up TCP socket for listening\n");
1089 sk->sk_data_ready = svc_tcp_listen_data_ready;
1090 set_bit(SK_CONN, &svsk->sk_flags);
1091 } else {
1092 dprintk("setting up TCP socket for reading\n");
1093 sk->sk_state_change = svc_tcp_state_change;
1094 sk->sk_data_ready = svc_tcp_data_ready;
1095 sk->sk_write_space = svc_write_space;
1096
1097 svsk->sk_reclen = 0;
1098 svsk->sk_tcplen = 0;
1099
1100 tp->nonagle = 1; /* disable Nagle's algorithm */
1101
1102 /* initialise setting must have enough space to
1103 * receive and respond to one request.
1104 * svc_tcp_recvfrom will re-adjust if necessary
1105 */
1106 svc_sock_setbufsize(svsk->sk_sock,
1107 3 * svsk->sk_server->sv_bufsz,
1108 3 * svsk->sk_server->sv_bufsz);
1109
1110 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1111 set_bit(SK_DATA, &svsk->sk_flags);
1112 if (sk->sk_state != TCP_ESTABLISHED)
1113 set_bit(SK_CLOSE, &svsk->sk_flags);
1114 }
1115}
1116
1117void
1118svc_sock_update_bufs(struct svc_serv *serv)
1119{
1120 /*
1121 * The number of server threads has changed. Update
1122 * rcvbuf and sndbuf accordingly on all sockets
1123 */
1124 struct list_head *le;
1125
1126 spin_lock_bh(&serv->sv_lock);
1127 list_for_each(le, &serv->sv_permsocks) {
1128 struct svc_sock *svsk =
1129 list_entry(le, struct svc_sock, sk_list);
1130 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1131 }
1132 list_for_each(le, &serv->sv_tempsocks) {
1133 struct svc_sock *svsk =
1134 list_entry(le, struct svc_sock, sk_list);
1135 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1136 }
1137 spin_unlock_bh(&serv->sv_lock);
1138}
1139
1140/*
1141 * Receive the next request on any socket.
1142 */
1143int
1144svc_recv(struct svc_serv *serv, struct svc_rqst *rqstp, long timeout)
1145{
1146 struct svc_sock *svsk =NULL;
1147 int len;
1148 int pages;
1149 struct xdr_buf *arg;
1150 DECLARE_WAITQUEUE(wait, current);
1151
1152 dprintk("svc: server %p waiting for data (to = %ld)\n",
1153 rqstp, timeout);
1154
1155 if (rqstp->rq_sock)
1156 printk(KERN_ERR
1157 "svc_recv: service %p, socket not NULL!\n",
1158 rqstp);
1159 if (waitqueue_active(&rqstp->rq_wait))
1160 printk(KERN_ERR
1161 "svc_recv: service %p, wait queue active!\n",
1162 rqstp);
1163
1164 /* Initialize the buffers */
1165 /* first reclaim pages that were moved to response list */
1166 svc_pushback_allpages(rqstp);
1167
1168 /* now allocate needed pages. If we get a failure, sleep briefly */
1169 pages = 2 + (serv->sv_bufsz + PAGE_SIZE -1) / PAGE_SIZE;
1170 while (rqstp->rq_arghi < pages) {
1171 struct page *p = alloc_page(GFP_KERNEL);
1172 if (!p) {
Nishanth Aravamudan121caf52005-09-12 14:15:34 -07001173 schedule_timeout_uninterruptible(msecs_to_jiffies(500));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 continue;
1175 }
1176 rqstp->rq_argpages[rqstp->rq_arghi++] = p;
1177 }
1178
1179 /* Make arg->head point to first page and arg->pages point to rest */
1180 arg = &rqstp->rq_arg;
1181 arg->head[0].iov_base = page_address(rqstp->rq_argpages[0]);
1182 arg->head[0].iov_len = PAGE_SIZE;
1183 rqstp->rq_argused = 1;
1184 arg->pages = rqstp->rq_argpages + 1;
1185 arg->page_base = 0;
1186 /* save at least one page for response */
1187 arg->page_len = (pages-2)*PAGE_SIZE;
1188 arg->len = (pages-1)*PAGE_SIZE;
1189 arg->tail[0].iov_len = 0;
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07001190
1191 try_to_freeze();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 if (signalled())
1193 return -EINTR;
1194
1195 spin_lock_bh(&serv->sv_lock);
1196 if (!list_empty(&serv->sv_tempsocks)) {
1197 svsk = list_entry(serv->sv_tempsocks.next,
1198 struct svc_sock, sk_list);
1199 /* apparently the "standard" is that clients close
1200 * idle connections after 5 minutes, servers after
1201 * 6 minutes
1202 * http://www.connectathon.org/talks96/nfstcp.pdf
1203 */
1204 if (get_seconds() - svsk->sk_lastrecv < 6*60
1205 || test_bit(SK_BUSY, &svsk->sk_flags))
1206 svsk = NULL;
1207 }
1208 if (svsk) {
1209 set_bit(SK_BUSY, &svsk->sk_flags);
1210 set_bit(SK_CLOSE, &svsk->sk_flags);
1211 rqstp->rq_sock = svsk;
1212 svsk->sk_inuse++;
1213 } else if ((svsk = svc_sock_dequeue(serv)) != NULL) {
1214 rqstp->rq_sock = svsk;
1215 svsk->sk_inuse++;
1216 rqstp->rq_reserved = serv->sv_bufsz;
1217 svsk->sk_reserved += rqstp->rq_reserved;
1218 } else {
1219 /* No data pending. Go to sleep */
1220 svc_serv_enqueue(serv, rqstp);
1221
1222 /*
1223 * We have to be able to interrupt this wait
1224 * to bring down the daemons ...
1225 */
1226 set_current_state(TASK_INTERRUPTIBLE);
1227 add_wait_queue(&rqstp->rq_wait, &wait);
1228 spin_unlock_bh(&serv->sv_lock);
1229
1230 schedule_timeout(timeout);
1231
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07001232 try_to_freeze();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
1234 spin_lock_bh(&serv->sv_lock);
1235 remove_wait_queue(&rqstp->rq_wait, &wait);
1236
1237 if (!(svsk = rqstp->rq_sock)) {
1238 svc_serv_dequeue(serv, rqstp);
1239 spin_unlock_bh(&serv->sv_lock);
1240 dprintk("svc: server %p, no data yet\n", rqstp);
1241 return signalled()? -EINTR : -EAGAIN;
1242 }
1243 }
1244 spin_unlock_bh(&serv->sv_lock);
1245
1246 dprintk("svc: server %p, socket %p, inuse=%d\n",
1247 rqstp, svsk, svsk->sk_inuse);
1248 len = svsk->sk_recvfrom(rqstp);
1249 dprintk("svc: got len=%d\n", len);
1250
1251 /* No data, incomplete (TCP) read, or accept() */
1252 if (len == 0 || len == -EAGAIN) {
1253 rqstp->rq_res.len = 0;
1254 svc_sock_release(rqstp);
1255 return -EAGAIN;
1256 }
1257 svsk->sk_lastrecv = get_seconds();
1258 if (test_bit(SK_TEMP, &svsk->sk_flags)) {
1259 /* push active sockets to end of list */
1260 spin_lock_bh(&serv->sv_lock);
1261 if (!list_empty(&svsk->sk_list))
1262 list_move_tail(&svsk->sk_list, &serv->sv_tempsocks);
1263 spin_unlock_bh(&serv->sv_lock);
1264 }
1265
1266 rqstp->rq_secure = ntohs(rqstp->rq_addr.sin_port) < 1024;
1267 rqstp->rq_chandle.defer = svc_defer;
1268
1269 if (serv->sv_stats)
1270 serv->sv_stats->netcnt++;
1271 return len;
1272}
1273
1274/*
1275 * Drop request
1276 */
1277void
1278svc_drop(struct svc_rqst *rqstp)
1279{
1280 dprintk("svc: socket %p dropped request\n", rqstp->rq_sock);
1281 svc_sock_release(rqstp);
1282}
1283
1284/*
1285 * Return reply to client.
1286 */
1287int
1288svc_send(struct svc_rqst *rqstp)
1289{
1290 struct svc_sock *svsk;
1291 int len;
1292 struct xdr_buf *xb;
1293
1294 if ((svsk = rqstp->rq_sock) == NULL) {
1295 printk(KERN_WARNING "NULL socket pointer in %s:%d\n",
1296 __FILE__, __LINE__);
1297 return -EFAULT;
1298 }
1299
1300 /* release the receive skb before sending the reply */
1301 svc_release_skb(rqstp);
1302
1303 /* calculate over-all length */
1304 xb = & rqstp->rq_res;
1305 xb->len = xb->head[0].iov_len +
1306 xb->page_len +
1307 xb->tail[0].iov_len;
1308
1309 /* Grab svsk->sk_sem to serialize outgoing data. */
1310 down(&svsk->sk_sem);
1311 if (test_bit(SK_DEAD, &svsk->sk_flags))
1312 len = -ENOTCONN;
1313 else
1314 len = svsk->sk_sendto(rqstp);
1315 up(&svsk->sk_sem);
1316 svc_sock_release(rqstp);
1317
1318 if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
1319 return 0;
1320 return len;
1321}
1322
1323/*
1324 * Initialize socket for RPC use and create svc_sock struct
1325 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1326 */
1327static struct svc_sock *
1328svc_setup_socket(struct svc_serv *serv, struct socket *sock,
1329 int *errp, int pmap_register)
1330{
1331 struct svc_sock *svsk;
1332 struct sock *inet;
1333
1334 dprintk("svc: svc_setup_socket %p\n", sock);
1335 if (!(svsk = kmalloc(sizeof(*svsk), GFP_KERNEL))) {
1336 *errp = -ENOMEM;
1337 return NULL;
1338 }
1339 memset(svsk, 0, sizeof(*svsk));
1340
1341 inet = sock->sk;
1342
1343 /* Register socket with portmapper */
1344 if (*errp >= 0 && pmap_register)
1345 *errp = svc_register(serv, inet->sk_protocol,
1346 ntohs(inet_sk(inet)->sport));
1347
1348 if (*errp < 0) {
1349 kfree(svsk);
1350 return NULL;
1351 }
1352
1353 set_bit(SK_BUSY, &svsk->sk_flags);
1354 inet->sk_user_data = svsk;
1355 svsk->sk_sock = sock;
1356 svsk->sk_sk = inet;
1357 svsk->sk_ostate = inet->sk_state_change;
1358 svsk->sk_odata = inet->sk_data_ready;
1359 svsk->sk_owspace = inet->sk_write_space;
1360 svsk->sk_server = serv;
1361 svsk->sk_lastrecv = get_seconds();
1362 INIT_LIST_HEAD(&svsk->sk_deferred);
1363 INIT_LIST_HEAD(&svsk->sk_ready);
1364 sema_init(&svsk->sk_sem, 1);
1365
1366 /* Initialize the socket */
1367 if (sock->type == SOCK_DGRAM)
1368 svc_udp_init(svsk);
1369 else
1370 svc_tcp_init(svsk);
1371
1372 spin_lock_bh(&serv->sv_lock);
1373 if (!pmap_register) {
1374 set_bit(SK_TEMP, &svsk->sk_flags);
1375 list_add(&svsk->sk_list, &serv->sv_tempsocks);
1376 serv->sv_tmpcnt++;
1377 } else {
1378 clear_bit(SK_TEMP, &svsk->sk_flags);
1379 list_add(&svsk->sk_list, &serv->sv_permsocks);
1380 }
1381 spin_unlock_bh(&serv->sv_lock);
1382
1383 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1384 svsk, svsk->sk_sk);
1385
1386 clear_bit(SK_BUSY, &svsk->sk_flags);
1387 svc_sock_enqueue(svsk);
1388 return svsk;
1389}
1390
1391/*
1392 * Create socket for RPC service.
1393 */
1394static int
1395svc_create_socket(struct svc_serv *serv, int protocol, struct sockaddr_in *sin)
1396{
1397 struct svc_sock *svsk;
1398 struct socket *sock;
1399 int error;
1400 int type;
1401
1402 dprintk("svc: svc_create_socket(%s, %d, %u.%u.%u.%u:%d)\n",
1403 serv->sv_program->pg_name, protocol,
1404 NIPQUAD(sin->sin_addr.s_addr),
1405 ntohs(sin->sin_port));
1406
1407 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1408 printk(KERN_WARNING "svc: only UDP and TCP "
1409 "sockets supported\n");
1410 return -EINVAL;
1411 }
1412 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1413
1414 if ((error = sock_create_kern(PF_INET, type, protocol, &sock)) < 0)
1415 return error;
1416
1417 if (sin != NULL) {
1418 if (type == SOCK_STREAM)
1419 sock->sk->sk_reuse = 1; /* allow address reuse */
1420 error = sock->ops->bind(sock, (struct sockaddr *) sin,
1421 sizeof(*sin));
1422 if (error < 0)
1423 goto bummer;
1424 }
1425
1426 if (protocol == IPPROTO_TCP) {
1427 if ((error = sock->ops->listen(sock, 64)) < 0)
1428 goto bummer;
1429 }
1430
1431 if ((svsk = svc_setup_socket(serv, sock, &error, 1)) != NULL)
1432 return 0;
1433
1434bummer:
1435 dprintk("svc: svc_create_socket error = %d\n", -error);
1436 sock_release(sock);
1437 return error;
1438}
1439
1440/*
1441 * Remove a dead socket
1442 */
1443void
1444svc_delete_socket(struct svc_sock *svsk)
1445{
1446 struct svc_serv *serv;
1447 struct sock *sk;
1448
1449 dprintk("svc: svc_delete_socket(%p)\n", svsk);
1450
1451 serv = svsk->sk_server;
1452 sk = svsk->sk_sk;
1453
1454 sk->sk_state_change = svsk->sk_ostate;
1455 sk->sk_data_ready = svsk->sk_odata;
1456 sk->sk_write_space = svsk->sk_owspace;
1457
1458 spin_lock_bh(&serv->sv_lock);
1459
1460 list_del_init(&svsk->sk_list);
1461 list_del_init(&svsk->sk_ready);
1462 if (!test_and_set_bit(SK_DEAD, &svsk->sk_flags))
1463 if (test_bit(SK_TEMP, &svsk->sk_flags))
1464 serv->sv_tmpcnt--;
1465
1466 if (!svsk->sk_inuse) {
1467 spin_unlock_bh(&serv->sv_lock);
1468 sock_release(svsk->sk_sock);
1469 kfree(svsk);
1470 } else {
1471 spin_unlock_bh(&serv->sv_lock);
1472 dprintk(KERN_NOTICE "svc: server socket destroy delayed\n");
1473 /* svsk->sk_server = NULL; */
1474 }
1475}
1476
1477/*
1478 * Make a socket for nfsd and lockd
1479 */
1480int
1481svc_makesock(struct svc_serv *serv, int protocol, unsigned short port)
1482{
1483 struct sockaddr_in sin;
1484
1485 dprintk("svc: creating socket proto = %d\n", protocol);
1486 sin.sin_family = AF_INET;
1487 sin.sin_addr.s_addr = INADDR_ANY;
1488 sin.sin_port = htons(port);
1489 return svc_create_socket(serv, protocol, &sin);
1490}
1491
1492/*
1493 * Handle defer and revisit of requests
1494 */
1495
1496static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1497{
1498 struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle);
1499 struct svc_serv *serv = dreq->owner;
1500 struct svc_sock *svsk;
1501
1502 if (too_many) {
1503 svc_sock_put(dr->svsk);
1504 kfree(dr);
1505 return;
1506 }
1507 dprintk("revisit queued\n");
1508 svsk = dr->svsk;
1509 dr->svsk = NULL;
1510 spin_lock_bh(&serv->sv_lock);
1511 list_add(&dr->handle.recent, &svsk->sk_deferred);
1512 spin_unlock_bh(&serv->sv_lock);
1513 set_bit(SK_DEFERRED, &svsk->sk_flags);
1514 svc_sock_enqueue(svsk);
1515 svc_sock_put(svsk);
1516}
1517
1518static struct cache_deferred_req *
1519svc_defer(struct cache_req *req)
1520{
1521 struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
1522 int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len);
1523 struct svc_deferred_req *dr;
1524
1525 if (rqstp->rq_arg.page_len)
1526 return NULL; /* if more than a page, give up FIXME */
1527 if (rqstp->rq_deferred) {
1528 dr = rqstp->rq_deferred;
1529 rqstp->rq_deferred = NULL;
1530 } else {
1531 int skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1532 /* FIXME maybe discard if size too large */
1533 dr = kmalloc(size, GFP_KERNEL);
1534 if (dr == NULL)
1535 return NULL;
1536
1537 dr->handle.owner = rqstp->rq_server;
1538 dr->prot = rqstp->rq_prot;
1539 dr->addr = rqstp->rq_addr;
1540 dr->argslen = rqstp->rq_arg.len >> 2;
1541 memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2);
1542 }
1543 spin_lock_bh(&rqstp->rq_server->sv_lock);
1544 rqstp->rq_sock->sk_inuse++;
1545 dr->svsk = rqstp->rq_sock;
1546 spin_unlock_bh(&rqstp->rq_server->sv_lock);
1547
1548 dr->handle.revisit = svc_revisit;
1549 return &dr->handle;
1550}
1551
1552/*
1553 * recv data from a deferred request into an active one
1554 */
1555static int svc_deferred_recv(struct svc_rqst *rqstp)
1556{
1557 struct svc_deferred_req *dr = rqstp->rq_deferred;
1558
1559 rqstp->rq_arg.head[0].iov_base = dr->args;
1560 rqstp->rq_arg.head[0].iov_len = dr->argslen<<2;
1561 rqstp->rq_arg.page_len = 0;
1562 rqstp->rq_arg.len = dr->argslen<<2;
1563 rqstp->rq_prot = dr->prot;
1564 rqstp->rq_addr = dr->addr;
1565 return dr->argslen<<2;
1566}
1567
1568
1569static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk)
1570{
1571 struct svc_deferred_req *dr = NULL;
1572 struct svc_serv *serv = svsk->sk_server;
1573
1574 if (!test_bit(SK_DEFERRED, &svsk->sk_flags))
1575 return NULL;
1576 spin_lock_bh(&serv->sv_lock);
1577 clear_bit(SK_DEFERRED, &svsk->sk_flags);
1578 if (!list_empty(&svsk->sk_deferred)) {
1579 dr = list_entry(svsk->sk_deferred.next,
1580 struct svc_deferred_req,
1581 handle.recent);
1582 list_del_init(&dr->handle.recent);
1583 set_bit(SK_DEFERRED, &svsk->sk_flags);
1584 }
1585 spin_unlock_bh(&serv->sv_lock);
1586 return dr;
1587}