aboutsummaryrefslogtreecommitdiffstats
path: root/net/sunrpc/svcsock.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
committerLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
commit1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch)
tree0bba044c4ce775e45a88a51686b5d9f90697ea9d /net/sunrpc/svcsock.c
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
Diffstat (limited to 'net/sunrpc/svcsock.c')
-rw-r--r--net/sunrpc/svcsock.c1585
1 files changed, 1585 insertions, 0 deletions
diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
new file mode 100644
index 000000000000..05907035bc96
--- /dev/null
+++ b/net/sunrpc/svcsock.c
@@ -0,0 +1,1585 @@
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>
29#include <linux/tcp.h>
30#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>
37#include <net/tcp.h>
38#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 }
587 if (skb->stamp.tv_sec == 0) {
588 skb->stamp.tv_sec = xtime.tv_sec;
589 skb->stamp.tv_usec = xtime.tv_nsec * 1000;
590 /* Don't enable netstamp, sunrpc doesn't
591 need that much accuracy */
592 }
593 svsk->sk_sk->sk_stamp = skb->stamp;
594 set_bit(SK_DATA, &svsk->sk_flags); /* there may be more data... */
595
596 /*
597 * Maybe more packets - kick another thread ASAP.
598 */
599 svc_sock_received(svsk);
600
601 len = skb->len - sizeof(struct udphdr);
602 rqstp->rq_arg.len = len;
603
604 rqstp->rq_prot = IPPROTO_UDP;
605
606 /* Get sender address */
607 rqstp->rq_addr.sin_family = AF_INET;
608 rqstp->rq_addr.sin_port = skb->h.uh->source;
609 rqstp->rq_addr.sin_addr.s_addr = skb->nh.iph->saddr;
610 rqstp->rq_daddr = skb->nh.iph->daddr;
611
612 if (skb_is_nonlinear(skb)) {
613 /* we have to copy */
614 local_bh_disable();
615 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
616 local_bh_enable();
617 /* checksum error */
618 skb_free_datagram(svsk->sk_sk, skb);
619 return 0;
620 }
621 local_bh_enable();
622 skb_free_datagram(svsk->sk_sk, skb);
623 } else {
624 /* we can use it in-place */
625 rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr);
626 rqstp->rq_arg.head[0].iov_len = len;
627 if (skb->ip_summed != CHECKSUM_UNNECESSARY) {
628 if ((unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum))) {
629 skb_free_datagram(svsk->sk_sk, skb);
630 return 0;
631 }
632 skb->ip_summed = CHECKSUM_UNNECESSARY;
633 }
634 rqstp->rq_skbuff = skb;
635 }
636
637 rqstp->rq_arg.page_base = 0;
638 if (len <= rqstp->rq_arg.head[0].iov_len) {
639 rqstp->rq_arg.head[0].iov_len = len;
640 rqstp->rq_arg.page_len = 0;
641 } else {
642 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
643 rqstp->rq_argused += (rqstp->rq_arg.page_len + PAGE_SIZE - 1)/ PAGE_SIZE;
644 }
645
646 if (serv->sv_stats)
647 serv->sv_stats->netudpcnt++;
648
649 return len;
650}
651
652static int
653svc_udp_sendto(struct svc_rqst *rqstp)
654{
655 int error;
656
657 error = svc_sendto(rqstp, &rqstp->rq_res);
658 if (error == -ECONNREFUSED)
659 /* ICMP error on earlier request. */
660 error = svc_sendto(rqstp, &rqstp->rq_res);
661
662 return error;
663}
664
665static void
666svc_udp_init(struct svc_sock *svsk)
667{
668 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
669 svsk->sk_sk->sk_write_space = svc_write_space;
670 svsk->sk_recvfrom = svc_udp_recvfrom;
671 svsk->sk_sendto = svc_udp_sendto;
672
673 /* initialise setting must have enough space to
674 * receive and respond to one request.
675 * svc_udp_recvfrom will re-adjust if necessary
676 */
677 svc_sock_setbufsize(svsk->sk_sock,
678 3 * svsk->sk_server->sv_bufsz,
679 3 * svsk->sk_server->sv_bufsz);
680
681 set_bit(SK_DATA, &svsk->sk_flags); /* might have come in before data_ready set up */
682 set_bit(SK_CHNGBUF, &svsk->sk_flags);
683}
684
685/*
686 * A data_ready event on a listening socket means there's a connection
687 * pending. Do not use state_change as a substitute for it.
688 */
689static void
690svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
691{
692 struct svc_sock *svsk;
693
694 dprintk("svc: socket %p TCP (listen) state change %d\n",
695 sk, sk->sk_state);
696
697 if (sk->sk_state != TCP_LISTEN) {
698 /*
699 * This callback may called twice when a new connection
700 * is established as a child socket inherits everything
701 * from a parent LISTEN socket.
702 * 1) data_ready method of the parent socket will be called
703 * when one of child sockets become ESTABLISHED.
704 * 2) data_ready method of the child socket may be called
705 * when it receives data before the socket is accepted.
706 * In case of 2, we should ignore it silently.
707 */
708 goto out;
709 }
710 if (!(svsk = (struct svc_sock *) sk->sk_user_data)) {
711 printk("svc: socket %p: no user data\n", sk);
712 goto out;
713 }
714 set_bit(SK_CONN, &svsk->sk_flags);
715 svc_sock_enqueue(svsk);
716 out:
717 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
718 wake_up_interruptible_all(sk->sk_sleep);
719}
720
721/*
722 * A state change on a connected socket means it's dying or dead.
723 */
724static void
725svc_tcp_state_change(struct sock *sk)
726{
727 struct svc_sock *svsk;
728
729 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
730 sk, sk->sk_state, sk->sk_user_data);
731
732 if (!(svsk = (struct svc_sock *) sk->sk_user_data)) {
733 printk("svc: socket %p: no user data\n", sk);
734 goto out;
735 }
736 set_bit(SK_CLOSE, &svsk->sk_flags);
737 svc_sock_enqueue(svsk);
738 out:
739 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
740 wake_up_interruptible_all(sk->sk_sleep);
741}
742
743static void
744svc_tcp_data_ready(struct sock *sk, int count)
745{
746 struct svc_sock * svsk;
747
748 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
749 sk, sk->sk_user_data);
750 if (!(svsk = (struct svc_sock *)(sk->sk_user_data)))
751 goto out;
752 set_bit(SK_DATA, &svsk->sk_flags);
753 svc_sock_enqueue(svsk);
754 out:
755 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
756 wake_up_interruptible(sk->sk_sleep);
757}
758
759/*
760 * Accept a TCP connection
761 */
762static void
763svc_tcp_accept(struct svc_sock *svsk)
764{
765 struct sockaddr_in sin;
766 struct svc_serv *serv = svsk->sk_server;
767 struct socket *sock = svsk->sk_sock;
768 struct socket *newsock;
769 struct proto_ops *ops;
770 struct svc_sock *newsvsk;
771 int err, slen;
772
773 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
774 if (!sock)
775 return;
776
777 err = sock_create_lite(PF_INET, SOCK_STREAM, IPPROTO_TCP, &newsock);
778 if (err) {
779 if (err == -ENOMEM)
780 printk(KERN_WARNING "%s: no more sockets!\n",
781 serv->sv_name);
782 return;
783 }
784
785 dprintk("svc: tcp_accept %p allocated\n", newsock);
786 newsock->ops = ops = sock->ops;
787
788 clear_bit(SK_CONN, &svsk->sk_flags);
789 if ((err = ops->accept(sock, newsock, O_NONBLOCK)) < 0) {
790 if (err != -EAGAIN && net_ratelimit())
791 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
792 serv->sv_name, -err);
793 goto failed; /* aborted connection or whatever */
794 }
795 set_bit(SK_CONN, &svsk->sk_flags);
796 svc_sock_enqueue(svsk);
797
798 slen = sizeof(sin);
799 err = ops->getname(newsock, (struct sockaddr *) &sin, &slen, 1);
800 if (err < 0) {
801 if (net_ratelimit())
802 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
803 serv->sv_name, -err);
804 goto failed; /* aborted connection or whatever */
805 }
806
807 /* Ideally, we would want to reject connections from unauthorized
808 * hosts here, but when we get encription, the IP of the host won't
809 * tell us anything. For now just warn about unpriv connections.
810 */
811 if (ntohs(sin.sin_port) >= 1024) {
812 dprintk(KERN_WARNING
813 "%s: connect from unprivileged port: %u.%u.%u.%u:%d\n",
814 serv->sv_name,
815 NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port));
816 }
817
818 dprintk("%s: connect from %u.%u.%u.%u:%04x\n", serv->sv_name,
819 NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port));
820
821 /* make sure that a write doesn't block forever when
822 * low on memory
823 */
824 newsock->sk->sk_sndtimeo = HZ*30;
825
826 if (!(newsvsk = svc_setup_socket(serv, newsock, &err, 0)))
827 goto failed;
828
829
830 /* make sure that we don't have too many active connections.
831 * If we have, something must be dropped.
832 *
833 * There's no point in trying to do random drop here for
834 * DoS prevention. The NFS clients does 1 reconnect in 15
835 * seconds. An attacker can easily beat that.
836 *
837 * The only somewhat efficient mechanism would be if drop
838 * old connections from the same IP first. But right now
839 * we don't even record the client IP in svc_sock.
840 */
841 if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) {
842 struct svc_sock *svsk = NULL;
843 spin_lock_bh(&serv->sv_lock);
844 if (!list_empty(&serv->sv_tempsocks)) {
845 if (net_ratelimit()) {
846 /* Try to help the admin */
847 printk(KERN_NOTICE "%s: too many open TCP "
848 "sockets, consider increasing the "
849 "number of nfsd threads\n",
850 serv->sv_name);
851 printk(KERN_NOTICE "%s: last TCP connect from "
852 "%u.%u.%u.%u:%d\n",
853 serv->sv_name,
854 NIPQUAD(sin.sin_addr.s_addr),
855 ntohs(sin.sin_port));
856 }
857 /*
858 * Always select the oldest socket. It's not fair,
859 * but so is life
860 */
861 svsk = list_entry(serv->sv_tempsocks.prev,
862 struct svc_sock,
863 sk_list);
864 set_bit(SK_CLOSE, &svsk->sk_flags);
865 svsk->sk_inuse ++;
866 }
867 spin_unlock_bh(&serv->sv_lock);
868
869 if (svsk) {
870 svc_sock_enqueue(svsk);
871 svc_sock_put(svsk);
872 }
873
874 }
875
876 if (serv->sv_stats)
877 serv->sv_stats->nettcpconn++;
878
879 return;
880
881failed:
882 sock_release(newsock);
883 return;
884}
885
886/*
887 * Receive data from a TCP socket.
888 */
889static int
890svc_tcp_recvfrom(struct svc_rqst *rqstp)
891{
892 struct svc_sock *svsk = rqstp->rq_sock;
893 struct svc_serv *serv = svsk->sk_server;
894 int len;
895 struct kvec vec[RPCSVC_MAXPAGES];
896 int pnum, vlen;
897
898 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
899 svsk, test_bit(SK_DATA, &svsk->sk_flags),
900 test_bit(SK_CONN, &svsk->sk_flags),
901 test_bit(SK_CLOSE, &svsk->sk_flags));
902
903 if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
904 svc_sock_received(svsk);
905 return svc_deferred_recv(rqstp);
906 }
907
908 if (test_bit(SK_CLOSE, &svsk->sk_flags)) {
909 svc_delete_socket(svsk);
910 return 0;
911 }
912
913 if (test_bit(SK_CONN, &svsk->sk_flags)) {
914 svc_tcp_accept(svsk);
915 svc_sock_received(svsk);
916 return 0;
917 }
918
919 if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags))
920 /* sndbuf needs to have room for one request
921 * per thread, otherwise we can stall even when the
922 * network isn't a bottleneck.
923 * rcvbuf just needs to be able to hold a few requests.
924 * Normally they will be removed from the queue
925 * as soon a a complete request arrives.
926 */
927 svc_sock_setbufsize(svsk->sk_sock,
928 (serv->sv_nrthreads+3) * serv->sv_bufsz,
929 3 * serv->sv_bufsz);
930
931 clear_bit(SK_DATA, &svsk->sk_flags);
932
933 /* Receive data. If we haven't got the record length yet, get
934 * the next four bytes. Otherwise try to gobble up as much as
935 * possible up to the complete record length.
936 */
937 if (svsk->sk_tcplen < 4) {
938 unsigned long want = 4 - svsk->sk_tcplen;
939 struct kvec iov;
940
941 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
942 iov.iov_len = want;
943 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
944 goto error;
945 svsk->sk_tcplen += len;
946
947 if (len < want) {
948 dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
949 len, want);
950 svc_sock_received(svsk);
951 return -EAGAIN; /* record header not complete */
952 }
953
954 svsk->sk_reclen = ntohl(svsk->sk_reclen);
955 if (!(svsk->sk_reclen & 0x80000000)) {
956 /* FIXME: technically, a record can be fragmented,
957 * and non-terminal fragments will not have the top
958 * bit set in the fragment length header.
959 * But apparently no known nfs clients send fragmented
960 * records. */
961 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx (non-terminal)\n",
962 (unsigned long) svsk->sk_reclen);
963 goto err_delete;
964 }
965 svsk->sk_reclen &= 0x7fffffff;
966 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
967 if (svsk->sk_reclen > serv->sv_bufsz) {
968 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx (large)\n",
969 (unsigned long) svsk->sk_reclen);
970 goto err_delete;
971 }
972 }
973
974 /* Check whether enough data is available */
975 len = svc_recv_available(svsk);
976 if (len < 0)
977 goto error;
978
979 if (len < svsk->sk_reclen) {
980 dprintk("svc: incomplete TCP record (%d of %d)\n",
981 len, svsk->sk_reclen);
982 svc_sock_received(svsk);
983 return -EAGAIN; /* record not complete */
984 }
985 len = svsk->sk_reclen;
986 set_bit(SK_DATA, &svsk->sk_flags);
987
988 vec[0] = rqstp->rq_arg.head[0];
989 vlen = PAGE_SIZE;
990 pnum = 1;
991 while (vlen < len) {
992 vec[pnum].iov_base = page_address(rqstp->rq_argpages[rqstp->rq_argused++]);
993 vec[pnum].iov_len = PAGE_SIZE;
994 pnum++;
995 vlen += PAGE_SIZE;
996 }
997
998 /* Now receive data */
999 len = svc_recvfrom(rqstp, vec, pnum, len);
1000 if (len < 0)
1001 goto error;
1002
1003 dprintk("svc: TCP complete record (%d bytes)\n", len);
1004 rqstp->rq_arg.len = len;
1005 rqstp->rq_arg.page_base = 0;
1006 if (len <= rqstp->rq_arg.head[0].iov_len) {
1007 rqstp->rq_arg.head[0].iov_len = len;
1008 rqstp->rq_arg.page_len = 0;
1009 } else {
1010 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
1011 }
1012
1013 rqstp->rq_skbuff = NULL;
1014 rqstp->rq_prot = IPPROTO_TCP;
1015
1016 /* Reset TCP read info */
1017 svsk->sk_reclen = 0;
1018 svsk->sk_tcplen = 0;
1019
1020 svc_sock_received(svsk);
1021 if (serv->sv_stats)
1022 serv->sv_stats->nettcpcnt++;
1023
1024 return len;
1025
1026 err_delete:
1027 svc_delete_socket(svsk);
1028 return -EAGAIN;
1029
1030 error:
1031 if (len == -EAGAIN) {
1032 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1033 svc_sock_received(svsk);
1034 } else {
1035 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1036 svsk->sk_server->sv_name, -len);
1037 svc_sock_received(svsk);
1038 }
1039
1040 return len;
1041}
1042
1043/*
1044 * Send out data on TCP socket.
1045 */
1046static int
1047svc_tcp_sendto(struct svc_rqst *rqstp)
1048{
1049 struct xdr_buf *xbufp = &rqstp->rq_res;
1050 int sent;
1051 u32 reclen;
1052
1053 /* Set up the first element of the reply kvec.
1054 * Any other kvecs that may be in use have been taken
1055 * care of by the server implementation itself.
1056 */
1057 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1058 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1059
1060 if (test_bit(SK_DEAD, &rqstp->rq_sock->sk_flags))
1061 return -ENOTCONN;
1062
1063 sent = svc_sendto(rqstp, &rqstp->rq_res);
1064 if (sent != xbufp->len) {
1065 printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1066 rqstp->rq_sock->sk_server->sv_name,
1067 (sent<0)?"got error":"sent only",
1068 sent, xbufp->len);
1069 svc_delete_socket(rqstp->rq_sock);
1070 sent = -EAGAIN;
1071 }
1072 return sent;
1073}
1074
1075static void
1076svc_tcp_init(struct svc_sock *svsk)
1077{
1078 struct sock *sk = svsk->sk_sk;
1079 struct tcp_sock *tp = tcp_sk(sk);
1080
1081 svsk->sk_recvfrom = svc_tcp_recvfrom;
1082 svsk->sk_sendto = svc_tcp_sendto;
1083
1084 if (sk->sk_state == TCP_LISTEN) {
1085 dprintk("setting up TCP socket for listening\n");
1086 sk->sk_data_ready = svc_tcp_listen_data_ready;
1087 set_bit(SK_CONN, &svsk->sk_flags);
1088 } else {
1089 dprintk("setting up TCP socket for reading\n");
1090 sk->sk_state_change = svc_tcp_state_change;
1091 sk->sk_data_ready = svc_tcp_data_ready;
1092 sk->sk_write_space = svc_write_space;
1093
1094 svsk->sk_reclen = 0;
1095 svsk->sk_tcplen = 0;
1096
1097 tp->nonagle = 1; /* disable Nagle's algorithm */
1098
1099 /* initialise setting must have enough space to
1100 * receive and respond to one request.
1101 * svc_tcp_recvfrom will re-adjust if necessary
1102 */
1103 svc_sock_setbufsize(svsk->sk_sock,
1104 3 * svsk->sk_server->sv_bufsz,
1105 3 * svsk->sk_server->sv_bufsz);
1106
1107 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1108 set_bit(SK_DATA, &svsk->sk_flags);
1109 if (sk->sk_state != TCP_ESTABLISHED)
1110 set_bit(SK_CLOSE, &svsk->sk_flags);
1111 }
1112}
1113
1114void
1115svc_sock_update_bufs(struct svc_serv *serv)
1116{
1117 /*
1118 * The number of server threads has changed. Update
1119 * rcvbuf and sndbuf accordingly on all sockets
1120 */
1121 struct list_head *le;
1122
1123 spin_lock_bh(&serv->sv_lock);
1124 list_for_each(le, &serv->sv_permsocks) {
1125 struct svc_sock *svsk =
1126 list_entry(le, struct svc_sock, sk_list);
1127 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1128 }
1129 list_for_each(le, &serv->sv_tempsocks) {
1130 struct svc_sock *svsk =
1131 list_entry(le, struct svc_sock, sk_list);
1132 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1133 }
1134 spin_unlock_bh(&serv->sv_lock);
1135}
1136
1137/*
1138 * Receive the next request on any socket.
1139 */
1140int
1141svc_recv(struct svc_serv *serv, struct svc_rqst *rqstp, long timeout)
1142{
1143 struct svc_sock *svsk =NULL;
1144 int len;
1145 int pages;
1146 struct xdr_buf *arg;
1147 DECLARE_WAITQUEUE(wait, current);
1148
1149 dprintk("svc: server %p waiting for data (to = %ld)\n",
1150 rqstp, timeout);
1151
1152 if (rqstp->rq_sock)
1153 printk(KERN_ERR
1154 "svc_recv: service %p, socket not NULL!\n",
1155 rqstp);
1156 if (waitqueue_active(&rqstp->rq_wait))
1157 printk(KERN_ERR
1158 "svc_recv: service %p, wait queue active!\n",
1159 rqstp);
1160
1161 /* Initialize the buffers */
1162 /* first reclaim pages that were moved to response list */
1163 svc_pushback_allpages(rqstp);
1164
1165 /* now allocate needed pages. If we get a failure, sleep briefly */
1166 pages = 2 + (serv->sv_bufsz + PAGE_SIZE -1) / PAGE_SIZE;
1167 while (rqstp->rq_arghi < pages) {
1168 struct page *p = alloc_page(GFP_KERNEL);
1169 if (!p) {
1170 set_current_state(TASK_UNINTERRUPTIBLE);
1171 schedule_timeout(HZ/2);
1172 continue;
1173 }
1174 rqstp->rq_argpages[rqstp->rq_arghi++] = p;
1175 }
1176
1177 /* Make arg->head point to first page and arg->pages point to rest */
1178 arg = &rqstp->rq_arg;
1179 arg->head[0].iov_base = page_address(rqstp->rq_argpages[0]);
1180 arg->head[0].iov_len = PAGE_SIZE;
1181 rqstp->rq_argused = 1;
1182 arg->pages = rqstp->rq_argpages + 1;
1183 arg->page_base = 0;
1184 /* save at least one page for response */
1185 arg->page_len = (pages-2)*PAGE_SIZE;
1186 arg->len = (pages-1)*PAGE_SIZE;
1187 arg->tail[0].iov_len = 0;
1188
1189 try_to_freeze(PF_FREEZE);
1190 if (signalled())
1191 return -EINTR;
1192
1193 spin_lock_bh(&serv->sv_lock);
1194 if (!list_empty(&serv->sv_tempsocks)) {
1195 svsk = list_entry(serv->sv_tempsocks.next,
1196 struct svc_sock, sk_list);
1197 /* apparently the "standard" is that clients close
1198 * idle connections after 5 minutes, servers after
1199 * 6 minutes
1200 * http://www.connectathon.org/talks96/nfstcp.pdf
1201 */
1202 if (get_seconds() - svsk->sk_lastrecv < 6*60
1203 || test_bit(SK_BUSY, &svsk->sk_flags))
1204 svsk = NULL;
1205 }
1206 if (svsk) {
1207 set_bit(SK_BUSY, &svsk->sk_flags);
1208 set_bit(SK_CLOSE, &svsk->sk_flags);
1209 rqstp->rq_sock = svsk;
1210 svsk->sk_inuse++;
1211 } else if ((svsk = svc_sock_dequeue(serv)) != NULL) {
1212 rqstp->rq_sock = svsk;
1213 svsk->sk_inuse++;
1214 rqstp->rq_reserved = serv->sv_bufsz;
1215 svsk->sk_reserved += rqstp->rq_reserved;
1216 } else {
1217 /* No data pending. Go to sleep */
1218 svc_serv_enqueue(serv, rqstp);
1219
1220 /*
1221 * We have to be able to interrupt this wait
1222 * to bring down the daemons ...
1223 */
1224 set_current_state(TASK_INTERRUPTIBLE);
1225 add_wait_queue(&rqstp->rq_wait, &wait);
1226 spin_unlock_bh(&serv->sv_lock);
1227
1228 schedule_timeout(timeout);
1229
1230 try_to_freeze(PF_FREEZE);
1231
1232 spin_lock_bh(&serv->sv_lock);
1233 remove_wait_queue(&rqstp->rq_wait, &wait);
1234
1235 if (!(svsk = rqstp->rq_sock)) {
1236 svc_serv_dequeue(serv, rqstp);
1237 spin_unlock_bh(&serv->sv_lock);
1238 dprintk("svc: server %p, no data yet\n", rqstp);
1239 return signalled()? -EINTR : -EAGAIN;
1240 }
1241 }
1242 spin_unlock_bh(&serv->sv_lock);
1243
1244 dprintk("svc: server %p, socket %p, inuse=%d\n",
1245 rqstp, svsk, svsk->sk_inuse);
1246 len = svsk->sk_recvfrom(rqstp);
1247 dprintk("svc: got len=%d\n", len);
1248
1249 /* No data, incomplete (TCP) read, or accept() */
1250 if (len == 0 || len == -EAGAIN) {
1251 rqstp->rq_res.len = 0;
1252 svc_sock_release(rqstp);
1253 return -EAGAIN;
1254 }
1255 svsk->sk_lastrecv = get_seconds();
1256 if (test_bit(SK_TEMP, &svsk->sk_flags)) {
1257 /* push active sockets to end of list */
1258 spin_lock_bh(&serv->sv_lock);
1259 if (!list_empty(&svsk->sk_list))
1260 list_move_tail(&svsk->sk_list, &serv->sv_tempsocks);
1261 spin_unlock_bh(&serv->sv_lock);
1262 }
1263
1264 rqstp->rq_secure = ntohs(rqstp->rq_addr.sin_port) < 1024;
1265 rqstp->rq_chandle.defer = svc_defer;
1266
1267 if (serv->sv_stats)
1268 serv->sv_stats->netcnt++;
1269 return len;
1270}
1271
1272/*
1273 * Drop request
1274 */
1275void
1276svc_drop(struct svc_rqst *rqstp)
1277{
1278 dprintk("svc: socket %p dropped request\n", rqstp->rq_sock);
1279 svc_sock_release(rqstp);
1280}
1281
1282/*
1283 * Return reply to client.
1284 */
1285int
1286svc_send(struct svc_rqst *rqstp)
1287{
1288 struct svc_sock *svsk;
1289 int len;
1290 struct xdr_buf *xb;
1291
1292 if ((svsk = rqstp->rq_sock) == NULL) {
1293 printk(KERN_WARNING "NULL socket pointer in %s:%d\n",
1294 __FILE__, __LINE__);
1295 return -EFAULT;
1296 }
1297
1298 /* release the receive skb before sending the reply */
1299 svc_release_skb(rqstp);
1300
1301 /* calculate over-all length */
1302 xb = & rqstp->rq_res;
1303 xb->len = xb->head[0].iov_len +
1304 xb->page_len +
1305 xb->tail[0].iov_len;
1306
1307 /* Grab svsk->sk_sem to serialize outgoing data. */
1308 down(&svsk->sk_sem);
1309 if (test_bit(SK_DEAD, &svsk->sk_flags))
1310 len = -ENOTCONN;
1311 else
1312 len = svsk->sk_sendto(rqstp);
1313 up(&svsk->sk_sem);
1314 svc_sock_release(rqstp);
1315
1316 if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
1317 return 0;
1318 return len;
1319}
1320
1321/*
1322 * Initialize socket for RPC use and create svc_sock struct
1323 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1324 */
1325static struct svc_sock *
1326svc_setup_socket(struct svc_serv *serv, struct socket *sock,
1327 int *errp, int pmap_register)
1328{
1329 struct svc_sock *svsk;
1330 struct sock *inet;
1331
1332 dprintk("svc: svc_setup_socket %p\n", sock);
1333 if (!(svsk = kmalloc(sizeof(*svsk), GFP_KERNEL))) {
1334 *errp = -ENOMEM;
1335 return NULL;
1336 }
1337 memset(svsk, 0, sizeof(*svsk));
1338
1339 inet = sock->sk;
1340
1341 /* Register socket with portmapper */
1342 if (*errp >= 0 && pmap_register)
1343 *errp = svc_register(serv, inet->sk_protocol,
1344 ntohs(inet_sk(inet)->sport));
1345
1346 if (*errp < 0) {
1347 kfree(svsk);
1348 return NULL;
1349 }
1350
1351 set_bit(SK_BUSY, &svsk->sk_flags);
1352 inet->sk_user_data = svsk;
1353 svsk->sk_sock = sock;
1354 svsk->sk_sk = inet;
1355 svsk->sk_ostate = inet->sk_state_change;
1356 svsk->sk_odata = inet->sk_data_ready;
1357 svsk->sk_owspace = inet->sk_write_space;
1358 svsk->sk_server = serv;
1359 svsk->sk_lastrecv = get_seconds();
1360 INIT_LIST_HEAD(&svsk->sk_deferred);
1361 INIT_LIST_HEAD(&svsk->sk_ready);
1362 sema_init(&svsk->sk_sem, 1);
1363
1364 /* Initialize the socket */
1365 if (sock->type == SOCK_DGRAM)
1366 svc_udp_init(svsk);
1367 else
1368 svc_tcp_init(svsk);
1369
1370 spin_lock_bh(&serv->sv_lock);
1371 if (!pmap_register) {
1372 set_bit(SK_TEMP, &svsk->sk_flags);
1373 list_add(&svsk->sk_list, &serv->sv_tempsocks);
1374 serv->sv_tmpcnt++;
1375 } else {
1376 clear_bit(SK_TEMP, &svsk->sk_flags);
1377 list_add(&svsk->sk_list, &serv->sv_permsocks);
1378 }
1379 spin_unlock_bh(&serv->sv_lock);
1380
1381 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1382 svsk, svsk->sk_sk);
1383
1384 clear_bit(SK_BUSY, &svsk->sk_flags);
1385 svc_sock_enqueue(svsk);
1386 return svsk;
1387}
1388
1389/*
1390 * Create socket for RPC service.
1391 */
1392static int
1393svc_create_socket(struct svc_serv *serv, int protocol, struct sockaddr_in *sin)
1394{
1395 struct svc_sock *svsk;
1396 struct socket *sock;
1397 int error;
1398 int type;
1399
1400 dprintk("svc: svc_create_socket(%s, %d, %u.%u.%u.%u:%d)\n",
1401 serv->sv_program->pg_name, protocol,
1402 NIPQUAD(sin->sin_addr.s_addr),
1403 ntohs(sin->sin_port));
1404
1405 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1406 printk(KERN_WARNING "svc: only UDP and TCP "
1407 "sockets supported\n");
1408 return -EINVAL;
1409 }
1410 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1411
1412 if ((error = sock_create_kern(PF_INET, type, protocol, &sock)) < 0)
1413 return error;
1414
1415 if (sin != NULL) {
1416 if (type == SOCK_STREAM)
1417 sock->sk->sk_reuse = 1; /* allow address reuse */
1418 error = sock->ops->bind(sock, (struct sockaddr *) sin,
1419 sizeof(*sin));
1420 if (error < 0)
1421 goto bummer;
1422 }
1423
1424 if (protocol == IPPROTO_TCP) {
1425 if ((error = sock->ops->listen(sock, 64)) < 0)
1426 goto bummer;
1427 }
1428
1429 if ((svsk = svc_setup_socket(serv, sock, &error, 1)) != NULL)
1430 return 0;
1431
1432bummer:
1433 dprintk("svc: svc_create_socket error = %d\n", -error);
1434 sock_release(sock);
1435 return error;
1436}
1437
1438/*
1439 * Remove a dead socket
1440 */
1441void
1442svc_delete_socket(struct svc_sock *svsk)
1443{
1444 struct svc_serv *serv;
1445 struct sock *sk;
1446
1447 dprintk("svc: svc_delete_socket(%p)\n", svsk);
1448
1449 serv = svsk->sk_server;
1450 sk = svsk->sk_sk;
1451
1452 sk->sk_state_change = svsk->sk_ostate;
1453 sk->sk_data_ready = svsk->sk_odata;
1454 sk->sk_write_space = svsk->sk_owspace;
1455
1456 spin_lock_bh(&serv->sv_lock);
1457
1458 list_del_init(&svsk->sk_list);
1459 list_del_init(&svsk->sk_ready);
1460 if (!test_and_set_bit(SK_DEAD, &svsk->sk_flags))
1461 if (test_bit(SK_TEMP, &svsk->sk_flags))
1462 serv->sv_tmpcnt--;
1463
1464 if (!svsk->sk_inuse) {
1465 spin_unlock_bh(&serv->sv_lock);
1466 sock_release(svsk->sk_sock);
1467 kfree(svsk);
1468 } else {
1469 spin_unlock_bh(&serv->sv_lock);
1470 dprintk(KERN_NOTICE "svc: server socket destroy delayed\n");
1471 /* svsk->sk_server = NULL; */
1472 }
1473}
1474
1475/*
1476 * Make a socket for nfsd and lockd
1477 */
1478int
1479svc_makesock(struct svc_serv *serv, int protocol, unsigned short port)
1480{
1481 struct sockaddr_in sin;
1482
1483 dprintk("svc: creating socket proto = %d\n", protocol);
1484 sin.sin_family = AF_INET;
1485 sin.sin_addr.s_addr = INADDR_ANY;
1486 sin.sin_port = htons(port);
1487 return svc_create_socket(serv, protocol, &sin);
1488}
1489
1490/*
1491 * Handle defer and revisit of requests
1492 */
1493
1494static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1495{
1496 struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle);
1497 struct svc_serv *serv = dreq->owner;
1498 struct svc_sock *svsk;
1499
1500 if (too_many) {
1501 svc_sock_put(dr->svsk);
1502 kfree(dr);
1503 return;
1504 }
1505 dprintk("revisit queued\n");
1506 svsk = dr->svsk;
1507 dr->svsk = NULL;
1508 spin_lock_bh(&serv->sv_lock);
1509 list_add(&dr->handle.recent, &svsk->sk_deferred);
1510 spin_unlock_bh(&serv->sv_lock);
1511 set_bit(SK_DEFERRED, &svsk->sk_flags);
1512 svc_sock_enqueue(svsk);
1513 svc_sock_put(svsk);
1514}
1515
1516static struct cache_deferred_req *
1517svc_defer(struct cache_req *req)
1518{
1519 struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
1520 int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len);
1521 struct svc_deferred_req *dr;
1522
1523 if (rqstp->rq_arg.page_len)
1524 return NULL; /* if more than a page, give up FIXME */
1525 if (rqstp->rq_deferred) {
1526 dr = rqstp->rq_deferred;
1527 rqstp->rq_deferred = NULL;
1528 } else {
1529 int skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1530 /* FIXME maybe discard if size too large */
1531 dr = kmalloc(size, GFP_KERNEL);
1532 if (dr == NULL)
1533 return NULL;
1534
1535 dr->handle.owner = rqstp->rq_server;
1536 dr->prot = rqstp->rq_prot;
1537 dr->addr = rqstp->rq_addr;
1538 dr->argslen = rqstp->rq_arg.len >> 2;
1539 memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2);
1540 }
1541 spin_lock_bh(&rqstp->rq_server->sv_lock);
1542 rqstp->rq_sock->sk_inuse++;
1543 dr->svsk = rqstp->rq_sock;
1544 spin_unlock_bh(&rqstp->rq_server->sv_lock);
1545
1546 dr->handle.revisit = svc_revisit;
1547 return &dr->handle;
1548}
1549
1550/*
1551 * recv data from a deferred request into an active one
1552 */
1553static int svc_deferred_recv(struct svc_rqst *rqstp)
1554{
1555 struct svc_deferred_req *dr = rqstp->rq_deferred;
1556
1557 rqstp->rq_arg.head[0].iov_base = dr->args;
1558 rqstp->rq_arg.head[0].iov_len = dr->argslen<<2;
1559 rqstp->rq_arg.page_len = 0;
1560 rqstp->rq_arg.len = dr->argslen<<2;
1561 rqstp->rq_prot = dr->prot;
1562 rqstp->rq_addr = dr->addr;
1563 return dr->argslen<<2;
1564}
1565
1566
1567static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk)
1568{
1569 struct svc_deferred_req *dr = NULL;
1570 struct svc_serv *serv = svsk->sk_server;
1571
1572 if (!test_bit(SK_DEFERRED, &svsk->sk_flags))
1573 return NULL;
1574 spin_lock_bh(&serv->sv_lock);
1575 clear_bit(SK_DEFERRED, &svsk->sk_flags);
1576 if (!list_empty(&svsk->sk_deferred)) {
1577 dr = list_entry(svsk->sk_deferred.next,
1578 struct svc_deferred_req,
1579 handle.recent);
1580 list_del_init(&dr->handle.recent);
1581 set_bit(SK_DEFERRED, &svsk->sk_flags);
1582 }
1583 spin_unlock_bh(&serv->sv_lock);
1584 return dr;
1585}