xref: /openbmc/linux/net/sunrpc/svcsock.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
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_states.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 
66 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
67 					 int *errp, int pmap_reg);
68 static void		svc_udp_data_ready(struct sock *, int);
69 static int		svc_udp_recvfrom(struct svc_rqst *);
70 static int		svc_udp_sendto(struct svc_rqst *);
71 
72 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk);
73 static int svc_deferred_recv(struct svc_rqst *rqstp);
74 static 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  */
82 static inline void
83 svc_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  */
91 static inline void
92 svc_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  */
100 static inline void
101 svc_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  */
121 static inline unsigned long
122 svc_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  */
139 static void
140 svc_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 
210 out_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  */
217 static inline struct svc_sock *
218 svc_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  */
241 static inline void
242 svc_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  */
259 void 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  */
277 static inline void
278 svc_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 
293 static void
294 svc_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  */
325 void
326 svc_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  */
348 static int
349 svc_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 	}
423 out:
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  */
434 static int
435 svc_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  */
451 static int
452 svc_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  */
486 static inline void
487 svc_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  */
512 static void
513 svc_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 		dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
519 			svsk, sk, count, test_bit(SK_BUSY, &svsk->sk_flags));
520 		set_bit(SK_DATA, &svsk->sk_flags);
521 		svc_sock_enqueue(svsk);
522 	}
523 	if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
524 		wake_up_interruptible(sk->sk_sleep);
525 }
526 
527 /*
528  * INET callback when space is newly available on the socket.
529  */
530 static void
531 svc_write_space(struct sock *sk)
532 {
533 	struct svc_sock	*svsk = (struct svc_sock *)(sk->sk_user_data);
534 
535 	if (svsk) {
536 		dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
537 			svsk, sk, test_bit(SK_BUSY, &svsk->sk_flags));
538 		svc_sock_enqueue(svsk);
539 	}
540 
541 	if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) {
542 		dprintk("RPC svc_write_space: someone sleeping on %p\n",
543 		       svsk);
544 		wake_up_interruptible(sk->sk_sleep);
545 	}
546 }
547 
548 /*
549  * Receive a datagram from a UDP socket.
550  */
551 static int
552 svc_udp_recvfrom(struct svc_rqst *rqstp)
553 {
554 	struct svc_sock	*svsk = rqstp->rq_sock;
555 	struct svc_serv	*serv = svsk->sk_server;
556 	struct sk_buff	*skb;
557 	int		err, len;
558 
559 	if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags))
560 	    /* udp sockets need large rcvbuf as all pending
561 	     * requests are still in that buffer.  sndbuf must
562 	     * also be large enough that there is enough space
563 	     * for one reply per thread.
564 	     */
565 	    svc_sock_setbufsize(svsk->sk_sock,
566 				(serv->sv_nrthreads+3) * serv->sv_bufsz,
567 				(serv->sv_nrthreads+3) * serv->sv_bufsz);
568 
569 	if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
570 		svc_sock_received(svsk);
571 		return svc_deferred_recv(rqstp);
572 	}
573 
574 	clear_bit(SK_DATA, &svsk->sk_flags);
575 	while ((skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err)) == NULL) {
576 		if (err == -EAGAIN) {
577 			svc_sock_received(svsk);
578 			return err;
579 		}
580 		/* possibly an icmp error */
581 		dprintk("svc: recvfrom returned error %d\n", -err);
582 	}
583 	if (skb->tstamp.off_sec == 0) {
584 		struct timeval tv;
585 
586 		tv.tv_sec = xtime.tv_sec;
587 		tv.tv_usec = xtime.tv_nsec / NSEC_PER_USEC;
588 		skb_set_timestamp(skb, &tv);
589 		/* Don't enable netstamp, sunrpc doesn't
590 		   need that much accuracy */
591 	}
592 	skb_get_timestamp(skb, &svsk->sk_sk->sk_stamp);
593 	set_bit(SK_DATA, &svsk->sk_flags); /* there may be more data... */
594 
595 	/*
596 	 * Maybe more packets - kick another thread ASAP.
597 	 */
598 	svc_sock_received(svsk);
599 
600 	len  = skb->len - sizeof(struct udphdr);
601 	rqstp->rq_arg.len = len;
602 
603 	rqstp->rq_prot        = IPPROTO_UDP;
604 
605 	/* Get sender address */
606 	rqstp->rq_addr.sin_family = AF_INET;
607 	rqstp->rq_addr.sin_port = skb->h.uh->source;
608 	rqstp->rq_addr.sin_addr.s_addr = skb->nh.iph->saddr;
609 	rqstp->rq_daddr = skb->nh.iph->daddr;
610 
611 	if (skb_is_nonlinear(skb)) {
612 		/* we have to copy */
613 		local_bh_disable();
614 		if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
615 			local_bh_enable();
616 			/* checksum error */
617 			skb_free_datagram(svsk->sk_sk, skb);
618 			return 0;
619 		}
620 		local_bh_enable();
621 		skb_free_datagram(svsk->sk_sk, skb);
622 	} else {
623 		/* we can use it in-place */
624 		rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr);
625 		rqstp->rq_arg.head[0].iov_len = len;
626 		if (skb->ip_summed != CHECKSUM_UNNECESSARY) {
627 			if ((unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum))) {
628 				skb_free_datagram(svsk->sk_sk, skb);
629 				return 0;
630 			}
631 			skb->ip_summed = CHECKSUM_UNNECESSARY;
632 		}
633 		rqstp->rq_skbuff = skb;
634 	}
635 
636 	rqstp->rq_arg.page_base = 0;
637 	if (len <= rqstp->rq_arg.head[0].iov_len) {
638 		rqstp->rq_arg.head[0].iov_len = len;
639 		rqstp->rq_arg.page_len = 0;
640 	} else {
641 		rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
642 		rqstp->rq_argused += (rqstp->rq_arg.page_len + PAGE_SIZE - 1)/ PAGE_SIZE;
643 	}
644 
645 	if (serv->sv_stats)
646 		serv->sv_stats->netudpcnt++;
647 
648 	return len;
649 }
650 
651 static int
652 svc_udp_sendto(struct svc_rqst *rqstp)
653 {
654 	int		error;
655 
656 	error = svc_sendto(rqstp, &rqstp->rq_res);
657 	if (error == -ECONNREFUSED)
658 		/* ICMP error on earlier request. */
659 		error = svc_sendto(rqstp, &rqstp->rq_res);
660 
661 	return error;
662 }
663 
664 static void
665 svc_udp_init(struct svc_sock *svsk)
666 {
667 	svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
668 	svsk->sk_sk->sk_write_space = svc_write_space;
669 	svsk->sk_recvfrom = svc_udp_recvfrom;
670 	svsk->sk_sendto = svc_udp_sendto;
671 
672 	/* initialise setting must have enough space to
673 	 * receive and respond to one request.
674 	 * svc_udp_recvfrom will re-adjust if necessary
675 	 */
676 	svc_sock_setbufsize(svsk->sk_sock,
677 			    3 * svsk->sk_server->sv_bufsz,
678 			    3 * svsk->sk_server->sv_bufsz);
679 
680 	set_bit(SK_DATA, &svsk->sk_flags); /* might have come in before data_ready set up */
681 	set_bit(SK_CHNGBUF, &svsk->sk_flags);
682 }
683 
684 /*
685  * A data_ready event on a listening socket means there's a connection
686  * pending. Do not use state_change as a substitute for it.
687  */
688 static void
689 svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
690 {
691 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
692 
693 	dprintk("svc: socket %p TCP (listen) state change %d\n",
694 		sk, sk->sk_state);
695 
696 	/*
697 	 * This callback may called twice when a new connection
698 	 * is established as a child socket inherits everything
699 	 * from a parent LISTEN socket.
700 	 * 1) data_ready method of the parent socket will be called
701 	 *    when one of child sockets become ESTABLISHED.
702 	 * 2) data_ready method of the child socket may be called
703 	 *    when it receives data before the socket is accepted.
704 	 * In case of 2, we should ignore it silently.
705 	 */
706 	if (sk->sk_state == TCP_LISTEN) {
707 		if (svsk) {
708 			set_bit(SK_CONN, &svsk->sk_flags);
709 			svc_sock_enqueue(svsk);
710 		} else
711 			printk("svc: socket %p: no user data\n", sk);
712 	}
713 
714 	if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
715 		wake_up_interruptible_all(sk->sk_sleep);
716 }
717 
718 /*
719  * A state change on a connected socket means it's dying or dead.
720  */
721 static void
722 svc_tcp_state_change(struct sock *sk)
723 {
724 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
725 
726 	dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
727 		sk, sk->sk_state, sk->sk_user_data);
728 
729 	if (!svsk)
730 		printk("svc: socket %p: no user data\n", sk);
731 	else {
732 		set_bit(SK_CLOSE, &svsk->sk_flags);
733 		svc_sock_enqueue(svsk);
734 	}
735 	if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
736 		wake_up_interruptible_all(sk->sk_sleep);
737 }
738 
739 static void
740 svc_tcp_data_ready(struct sock *sk, int count)
741 {
742 	struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
743 
744 	dprintk("svc: socket %p TCP data ready (svsk %p)\n",
745 		sk, sk->sk_user_data);
746 	if (svsk) {
747 		set_bit(SK_DATA, &svsk->sk_flags);
748 		svc_sock_enqueue(svsk);
749 	}
750 	if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
751 		wake_up_interruptible(sk->sk_sleep);
752 }
753 
754 /*
755  * Accept a TCP connection
756  */
757 static void
758 svc_tcp_accept(struct svc_sock *svsk)
759 {
760 	struct sockaddr_in sin;
761 	struct svc_serv	*serv = svsk->sk_server;
762 	struct socket	*sock = svsk->sk_sock;
763 	struct socket	*newsock;
764 	struct proto_ops *ops;
765 	struct svc_sock	*newsvsk;
766 	int		err, slen;
767 
768 	dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
769 	if (!sock)
770 		return;
771 
772 	err = sock_create_lite(PF_INET, SOCK_STREAM, IPPROTO_TCP, &newsock);
773 	if (err) {
774 		if (err == -ENOMEM)
775 			printk(KERN_WARNING "%s: no more sockets!\n",
776 			       serv->sv_name);
777 		return;
778 	}
779 
780 	dprintk("svc: tcp_accept %p allocated\n", newsock);
781 	newsock->ops = ops = sock->ops;
782 
783 	clear_bit(SK_CONN, &svsk->sk_flags);
784 	if ((err = ops->accept(sock, newsock, O_NONBLOCK)) < 0) {
785 		if (err != -EAGAIN && net_ratelimit())
786 			printk(KERN_WARNING "%s: accept failed (err %d)!\n",
787 				   serv->sv_name, -err);
788 		goto failed;		/* aborted connection or whatever */
789 	}
790 	set_bit(SK_CONN, &svsk->sk_flags);
791 	svc_sock_enqueue(svsk);
792 
793 	slen = sizeof(sin);
794 	err = ops->getname(newsock, (struct sockaddr *) &sin, &slen, 1);
795 	if (err < 0) {
796 		if (net_ratelimit())
797 			printk(KERN_WARNING "%s: peername failed (err %d)!\n",
798 				   serv->sv_name, -err);
799 		goto failed;		/* aborted connection or whatever */
800 	}
801 
802 	/* Ideally, we would want to reject connections from unauthorized
803 	 * hosts here, but when we get encription, the IP of the host won't
804 	 * tell us anything. For now just warn about unpriv connections.
805 	 */
806 	if (ntohs(sin.sin_port) >= 1024) {
807 		dprintk(KERN_WARNING
808 			"%s: connect from unprivileged port: %u.%u.%u.%u:%d\n",
809 			serv->sv_name,
810 			NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port));
811 	}
812 
813 	dprintk("%s: connect from %u.%u.%u.%u:%04x\n", serv->sv_name,
814 			NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port));
815 
816 	/* make sure that a write doesn't block forever when
817 	 * low on memory
818 	 */
819 	newsock->sk->sk_sndtimeo = HZ*30;
820 
821 	if (!(newsvsk = svc_setup_socket(serv, newsock, &err, 0)))
822 		goto failed;
823 
824 
825 	/* make sure that we don't have too many active connections.
826 	 * If we have, something must be dropped.
827 	 *
828 	 * There's no point in trying to do random drop here for
829 	 * DoS prevention. The NFS clients does 1 reconnect in 15
830 	 * seconds. An attacker can easily beat that.
831 	 *
832 	 * The only somewhat efficient mechanism would be if drop
833 	 * old connections from the same IP first. But right now
834 	 * we don't even record the client IP in svc_sock.
835 	 */
836 	if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) {
837 		struct svc_sock *svsk = NULL;
838 		spin_lock_bh(&serv->sv_lock);
839 		if (!list_empty(&serv->sv_tempsocks)) {
840 			if (net_ratelimit()) {
841 				/* Try to help the admin */
842 				printk(KERN_NOTICE "%s: too many open TCP "
843 					"sockets, consider increasing the "
844 					"number of nfsd threads\n",
845 						   serv->sv_name);
846 				printk(KERN_NOTICE "%s: last TCP connect from "
847 					"%u.%u.%u.%u:%d\n",
848 					serv->sv_name,
849 					NIPQUAD(sin.sin_addr.s_addr),
850 					ntohs(sin.sin_port));
851 			}
852 			/*
853 			 * Always select the oldest socket. It's not fair,
854 			 * but so is life
855 			 */
856 			svsk = list_entry(serv->sv_tempsocks.prev,
857 					  struct svc_sock,
858 					  sk_list);
859 			set_bit(SK_CLOSE, &svsk->sk_flags);
860 			svsk->sk_inuse ++;
861 		}
862 		spin_unlock_bh(&serv->sv_lock);
863 
864 		if (svsk) {
865 			svc_sock_enqueue(svsk);
866 			svc_sock_put(svsk);
867 		}
868 
869 	}
870 
871 	if (serv->sv_stats)
872 		serv->sv_stats->nettcpconn++;
873 
874 	return;
875 
876 failed:
877 	sock_release(newsock);
878 	return;
879 }
880 
881 /*
882  * Receive data from a TCP socket.
883  */
884 static int
885 svc_tcp_recvfrom(struct svc_rqst *rqstp)
886 {
887 	struct svc_sock	*svsk = rqstp->rq_sock;
888 	struct svc_serv	*serv = svsk->sk_server;
889 	int		len;
890 	struct kvec vec[RPCSVC_MAXPAGES];
891 	int pnum, vlen;
892 
893 	dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
894 		svsk, test_bit(SK_DATA, &svsk->sk_flags),
895 		test_bit(SK_CONN, &svsk->sk_flags),
896 		test_bit(SK_CLOSE, &svsk->sk_flags));
897 
898 	if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
899 		svc_sock_received(svsk);
900 		return svc_deferred_recv(rqstp);
901 	}
902 
903 	if (test_bit(SK_CLOSE, &svsk->sk_flags)) {
904 		svc_delete_socket(svsk);
905 		return 0;
906 	}
907 
908 	if (test_bit(SK_CONN, &svsk->sk_flags)) {
909 		svc_tcp_accept(svsk);
910 		svc_sock_received(svsk);
911 		return 0;
912 	}
913 
914 	if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags))
915 		/* sndbuf needs to have room for one request
916 		 * per thread, otherwise we can stall even when the
917 		 * network isn't a bottleneck.
918 		 * rcvbuf just needs to be able to hold a few requests.
919 		 * Normally they will be removed from the queue
920 		 * as soon a a complete request arrives.
921 		 */
922 		svc_sock_setbufsize(svsk->sk_sock,
923 				    (serv->sv_nrthreads+3) * serv->sv_bufsz,
924 				    3 * serv->sv_bufsz);
925 
926 	clear_bit(SK_DATA, &svsk->sk_flags);
927 
928 	/* Receive data. If we haven't got the record length yet, get
929 	 * the next four bytes. Otherwise try to gobble up as much as
930 	 * possible up to the complete record length.
931 	 */
932 	if (svsk->sk_tcplen < 4) {
933 		unsigned long	want = 4 - svsk->sk_tcplen;
934 		struct kvec	iov;
935 
936 		iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
937 		iov.iov_len  = want;
938 		if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
939 			goto error;
940 		svsk->sk_tcplen += len;
941 
942 		if (len < want) {
943 			dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
944 			        len, want);
945 			svc_sock_received(svsk);
946 			return -EAGAIN; /* record header not complete */
947 		}
948 
949 		svsk->sk_reclen = ntohl(svsk->sk_reclen);
950 		if (!(svsk->sk_reclen & 0x80000000)) {
951 			/* FIXME: technically, a record can be fragmented,
952 			 *  and non-terminal fragments will not have the top
953 			 *  bit set in the fragment length header.
954 			 *  But apparently no known nfs clients send fragmented
955 			 *  records. */
956 			printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx (non-terminal)\n",
957 			       (unsigned long) svsk->sk_reclen);
958 			goto err_delete;
959 		}
960 		svsk->sk_reclen &= 0x7fffffff;
961 		dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
962 		if (svsk->sk_reclen > serv->sv_bufsz) {
963 			printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx (large)\n",
964 			       (unsigned long) svsk->sk_reclen);
965 			goto err_delete;
966 		}
967 	}
968 
969 	/* Check whether enough data is available */
970 	len = svc_recv_available(svsk);
971 	if (len < 0)
972 		goto error;
973 
974 	if (len < svsk->sk_reclen) {
975 		dprintk("svc: incomplete TCP record (%d of %d)\n",
976 			len, svsk->sk_reclen);
977 		svc_sock_received(svsk);
978 		return -EAGAIN;	/* record not complete */
979 	}
980 	len = svsk->sk_reclen;
981 	set_bit(SK_DATA, &svsk->sk_flags);
982 
983 	vec[0] = rqstp->rq_arg.head[0];
984 	vlen = PAGE_SIZE;
985 	pnum = 1;
986 	while (vlen < len) {
987 		vec[pnum].iov_base = page_address(rqstp->rq_argpages[rqstp->rq_argused++]);
988 		vec[pnum].iov_len = PAGE_SIZE;
989 		pnum++;
990 		vlen += PAGE_SIZE;
991 	}
992 
993 	/* Now receive data */
994 	len = svc_recvfrom(rqstp, vec, pnum, len);
995 	if (len < 0)
996 		goto error;
997 
998 	dprintk("svc: TCP complete record (%d bytes)\n", len);
999 	rqstp->rq_arg.len = len;
1000 	rqstp->rq_arg.page_base = 0;
1001 	if (len <= rqstp->rq_arg.head[0].iov_len) {
1002 		rqstp->rq_arg.head[0].iov_len = len;
1003 		rqstp->rq_arg.page_len = 0;
1004 	} else {
1005 		rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
1006 	}
1007 
1008 	rqstp->rq_skbuff      = NULL;
1009 	rqstp->rq_prot	      = IPPROTO_TCP;
1010 
1011 	/* Reset TCP read info */
1012 	svsk->sk_reclen = 0;
1013 	svsk->sk_tcplen = 0;
1014 
1015 	svc_sock_received(svsk);
1016 	if (serv->sv_stats)
1017 		serv->sv_stats->nettcpcnt++;
1018 
1019 	return len;
1020 
1021  err_delete:
1022 	svc_delete_socket(svsk);
1023 	return -EAGAIN;
1024 
1025  error:
1026 	if (len == -EAGAIN) {
1027 		dprintk("RPC: TCP recvfrom got EAGAIN\n");
1028 		svc_sock_received(svsk);
1029 	} else {
1030 		printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1031 					svsk->sk_server->sv_name, -len);
1032 		svc_sock_received(svsk);
1033 	}
1034 
1035 	return len;
1036 }
1037 
1038 /*
1039  * Send out data on TCP socket.
1040  */
1041 static int
1042 svc_tcp_sendto(struct svc_rqst *rqstp)
1043 {
1044 	struct xdr_buf	*xbufp = &rqstp->rq_res;
1045 	int sent;
1046 	u32 reclen;
1047 
1048 	/* Set up the first element of the reply kvec.
1049 	 * Any other kvecs that may be in use have been taken
1050 	 * care of by the server implementation itself.
1051 	 */
1052 	reclen = htonl(0x80000000|((xbufp->len ) - 4));
1053 	memcpy(xbufp->head[0].iov_base, &reclen, 4);
1054 
1055 	if (test_bit(SK_DEAD, &rqstp->rq_sock->sk_flags))
1056 		return -ENOTCONN;
1057 
1058 	sent = svc_sendto(rqstp, &rqstp->rq_res);
1059 	if (sent != xbufp->len) {
1060 		printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1061 		       rqstp->rq_sock->sk_server->sv_name,
1062 		       (sent<0)?"got error":"sent only",
1063 		       sent, xbufp->len);
1064 		svc_delete_socket(rqstp->rq_sock);
1065 		sent = -EAGAIN;
1066 	}
1067 	return sent;
1068 }
1069 
1070 static void
1071 svc_tcp_init(struct svc_sock *svsk)
1072 {
1073 	struct sock	*sk = svsk->sk_sk;
1074 	struct tcp_sock *tp = tcp_sk(sk);
1075 
1076 	svsk->sk_recvfrom = svc_tcp_recvfrom;
1077 	svsk->sk_sendto = svc_tcp_sendto;
1078 
1079 	if (sk->sk_state == TCP_LISTEN) {
1080 		dprintk("setting up TCP socket for listening\n");
1081 		sk->sk_data_ready = svc_tcp_listen_data_ready;
1082 		set_bit(SK_CONN, &svsk->sk_flags);
1083 	} else {
1084 		dprintk("setting up TCP socket for reading\n");
1085 		sk->sk_state_change = svc_tcp_state_change;
1086 		sk->sk_data_ready = svc_tcp_data_ready;
1087 		sk->sk_write_space = svc_write_space;
1088 
1089 		svsk->sk_reclen = 0;
1090 		svsk->sk_tcplen = 0;
1091 
1092 		tp->nonagle = 1;        /* disable Nagle's algorithm */
1093 
1094 		/* initialise setting must have enough space to
1095 		 * receive and respond to one request.
1096 		 * svc_tcp_recvfrom will re-adjust if necessary
1097 		 */
1098 		svc_sock_setbufsize(svsk->sk_sock,
1099 				    3 * svsk->sk_server->sv_bufsz,
1100 				    3 * svsk->sk_server->sv_bufsz);
1101 
1102 		set_bit(SK_CHNGBUF, &svsk->sk_flags);
1103 		set_bit(SK_DATA, &svsk->sk_flags);
1104 		if (sk->sk_state != TCP_ESTABLISHED)
1105 			set_bit(SK_CLOSE, &svsk->sk_flags);
1106 	}
1107 }
1108 
1109 void
1110 svc_sock_update_bufs(struct svc_serv *serv)
1111 {
1112 	/*
1113 	 * The number of server threads has changed. Update
1114 	 * rcvbuf and sndbuf accordingly on all sockets
1115 	 */
1116 	struct list_head *le;
1117 
1118 	spin_lock_bh(&serv->sv_lock);
1119 	list_for_each(le, &serv->sv_permsocks) {
1120 		struct svc_sock *svsk =
1121 			list_entry(le, struct svc_sock, sk_list);
1122 		set_bit(SK_CHNGBUF, &svsk->sk_flags);
1123 	}
1124 	list_for_each(le, &serv->sv_tempsocks) {
1125 		struct svc_sock *svsk =
1126 			list_entry(le, struct svc_sock, sk_list);
1127 		set_bit(SK_CHNGBUF, &svsk->sk_flags);
1128 	}
1129 	spin_unlock_bh(&serv->sv_lock);
1130 }
1131 
1132 /*
1133  * Receive the next request on any socket.
1134  */
1135 int
1136 svc_recv(struct svc_serv *serv, struct svc_rqst *rqstp, long timeout)
1137 {
1138 	struct svc_sock		*svsk =NULL;
1139 	int			len;
1140 	int 			pages;
1141 	struct xdr_buf		*arg;
1142 	DECLARE_WAITQUEUE(wait, current);
1143 
1144 	dprintk("svc: server %p waiting for data (to = %ld)\n",
1145 		rqstp, timeout);
1146 
1147 	if (rqstp->rq_sock)
1148 		printk(KERN_ERR
1149 			"svc_recv: service %p, socket not NULL!\n",
1150 			 rqstp);
1151 	if (waitqueue_active(&rqstp->rq_wait))
1152 		printk(KERN_ERR
1153 			"svc_recv: service %p, wait queue active!\n",
1154 			 rqstp);
1155 
1156 	/* Initialize the buffers */
1157 	/* first reclaim pages that were moved to response list */
1158 	svc_pushback_allpages(rqstp);
1159 
1160 	/* now allocate needed pages.  If we get a failure, sleep briefly */
1161 	pages = 2 + (serv->sv_bufsz + PAGE_SIZE -1) / PAGE_SIZE;
1162 	while (rqstp->rq_arghi < pages) {
1163 		struct page *p = alloc_page(GFP_KERNEL);
1164 		if (!p) {
1165 			schedule_timeout_uninterruptible(msecs_to_jiffies(500));
1166 			continue;
1167 		}
1168 		rqstp->rq_argpages[rqstp->rq_arghi++] = p;
1169 	}
1170 
1171 	/* Make arg->head point to first page and arg->pages point to rest */
1172 	arg = &rqstp->rq_arg;
1173 	arg->head[0].iov_base = page_address(rqstp->rq_argpages[0]);
1174 	arg->head[0].iov_len = PAGE_SIZE;
1175 	rqstp->rq_argused = 1;
1176 	arg->pages = rqstp->rq_argpages + 1;
1177 	arg->page_base = 0;
1178 	/* save at least one page for response */
1179 	arg->page_len = (pages-2)*PAGE_SIZE;
1180 	arg->len = (pages-1)*PAGE_SIZE;
1181 	arg->tail[0].iov_len = 0;
1182 
1183 	try_to_freeze();
1184 	if (signalled())
1185 		return -EINTR;
1186 
1187 	spin_lock_bh(&serv->sv_lock);
1188 	if (!list_empty(&serv->sv_tempsocks)) {
1189 		svsk = list_entry(serv->sv_tempsocks.next,
1190 				  struct svc_sock, sk_list);
1191 		/* apparently the "standard" is that clients close
1192 		 * idle connections after 5 minutes, servers after
1193 		 * 6 minutes
1194 		 *   http://www.connectathon.org/talks96/nfstcp.pdf
1195 		 */
1196 		if (get_seconds() - svsk->sk_lastrecv < 6*60
1197 		    || test_bit(SK_BUSY, &svsk->sk_flags))
1198 			svsk = NULL;
1199 	}
1200 	if (svsk) {
1201 		set_bit(SK_BUSY, &svsk->sk_flags);
1202 		set_bit(SK_CLOSE, &svsk->sk_flags);
1203 		rqstp->rq_sock = svsk;
1204 		svsk->sk_inuse++;
1205 	} else if ((svsk = svc_sock_dequeue(serv)) != NULL) {
1206 		rqstp->rq_sock = svsk;
1207 		svsk->sk_inuse++;
1208 		rqstp->rq_reserved = serv->sv_bufsz;
1209 		svsk->sk_reserved += rqstp->rq_reserved;
1210 	} else {
1211 		/* No data pending. Go to sleep */
1212 		svc_serv_enqueue(serv, rqstp);
1213 
1214 		/*
1215 		 * We have to be able to interrupt this wait
1216 		 * to bring down the daemons ...
1217 		 */
1218 		set_current_state(TASK_INTERRUPTIBLE);
1219 		add_wait_queue(&rqstp->rq_wait, &wait);
1220 		spin_unlock_bh(&serv->sv_lock);
1221 
1222 		schedule_timeout(timeout);
1223 
1224 		try_to_freeze();
1225 
1226 		spin_lock_bh(&serv->sv_lock);
1227 		remove_wait_queue(&rqstp->rq_wait, &wait);
1228 
1229 		if (!(svsk = rqstp->rq_sock)) {
1230 			svc_serv_dequeue(serv, rqstp);
1231 			spin_unlock_bh(&serv->sv_lock);
1232 			dprintk("svc: server %p, no data yet\n", rqstp);
1233 			return signalled()? -EINTR : -EAGAIN;
1234 		}
1235 	}
1236 	spin_unlock_bh(&serv->sv_lock);
1237 
1238 	dprintk("svc: server %p, socket %p, inuse=%d\n",
1239 		 rqstp, svsk, svsk->sk_inuse);
1240 	len = svsk->sk_recvfrom(rqstp);
1241 	dprintk("svc: got len=%d\n", len);
1242 
1243 	/* No data, incomplete (TCP) read, or accept() */
1244 	if (len == 0 || len == -EAGAIN) {
1245 		rqstp->rq_res.len = 0;
1246 		svc_sock_release(rqstp);
1247 		return -EAGAIN;
1248 	}
1249 	svsk->sk_lastrecv = get_seconds();
1250 	if (test_bit(SK_TEMP, &svsk->sk_flags)) {
1251 		/* push active sockets to end of list */
1252 		spin_lock_bh(&serv->sv_lock);
1253 		if (!list_empty(&svsk->sk_list))
1254 			list_move_tail(&svsk->sk_list, &serv->sv_tempsocks);
1255 		spin_unlock_bh(&serv->sv_lock);
1256 	}
1257 
1258 	rqstp->rq_secure  = ntohs(rqstp->rq_addr.sin_port) < 1024;
1259 	rqstp->rq_chandle.defer = svc_defer;
1260 
1261 	if (serv->sv_stats)
1262 		serv->sv_stats->netcnt++;
1263 	return len;
1264 }
1265 
1266 /*
1267  * Drop request
1268  */
1269 void
1270 svc_drop(struct svc_rqst *rqstp)
1271 {
1272 	dprintk("svc: socket %p dropped request\n", rqstp->rq_sock);
1273 	svc_sock_release(rqstp);
1274 }
1275 
1276 /*
1277  * Return reply to client.
1278  */
1279 int
1280 svc_send(struct svc_rqst *rqstp)
1281 {
1282 	struct svc_sock	*svsk;
1283 	int		len;
1284 	struct xdr_buf	*xb;
1285 
1286 	if ((svsk = rqstp->rq_sock) == NULL) {
1287 		printk(KERN_WARNING "NULL socket pointer in %s:%d\n",
1288 				__FILE__, __LINE__);
1289 		return -EFAULT;
1290 	}
1291 
1292 	/* release the receive skb before sending the reply */
1293 	svc_release_skb(rqstp);
1294 
1295 	/* calculate over-all length */
1296 	xb = & rqstp->rq_res;
1297 	xb->len = xb->head[0].iov_len +
1298 		xb->page_len +
1299 		xb->tail[0].iov_len;
1300 
1301 	/* Grab svsk->sk_sem to serialize outgoing data. */
1302 	down(&svsk->sk_sem);
1303 	if (test_bit(SK_DEAD, &svsk->sk_flags))
1304 		len = -ENOTCONN;
1305 	else
1306 		len = svsk->sk_sendto(rqstp);
1307 	up(&svsk->sk_sem);
1308 	svc_sock_release(rqstp);
1309 
1310 	if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
1311 		return 0;
1312 	return len;
1313 }
1314 
1315 /*
1316  * Initialize socket for RPC use and create svc_sock struct
1317  * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1318  */
1319 static struct svc_sock *
1320 svc_setup_socket(struct svc_serv *serv, struct socket *sock,
1321 					int *errp, int pmap_register)
1322 {
1323 	struct svc_sock	*svsk;
1324 	struct sock	*inet;
1325 
1326 	dprintk("svc: svc_setup_socket %p\n", sock);
1327 	if (!(svsk = kmalloc(sizeof(*svsk), GFP_KERNEL))) {
1328 		*errp = -ENOMEM;
1329 		return NULL;
1330 	}
1331 	memset(svsk, 0, sizeof(*svsk));
1332 
1333 	inet = sock->sk;
1334 
1335 	/* Register socket with portmapper */
1336 	if (*errp >= 0 && pmap_register)
1337 		*errp = svc_register(serv, inet->sk_protocol,
1338 				     ntohs(inet_sk(inet)->sport));
1339 
1340 	if (*errp < 0) {
1341 		kfree(svsk);
1342 		return NULL;
1343 	}
1344 
1345 	set_bit(SK_BUSY, &svsk->sk_flags);
1346 	inet->sk_user_data = svsk;
1347 	svsk->sk_sock = sock;
1348 	svsk->sk_sk = inet;
1349 	svsk->sk_ostate = inet->sk_state_change;
1350 	svsk->sk_odata = inet->sk_data_ready;
1351 	svsk->sk_owspace = inet->sk_write_space;
1352 	svsk->sk_server = serv;
1353 	svsk->sk_lastrecv = get_seconds();
1354 	INIT_LIST_HEAD(&svsk->sk_deferred);
1355 	INIT_LIST_HEAD(&svsk->sk_ready);
1356 	sema_init(&svsk->sk_sem, 1);
1357 
1358 	/* Initialize the socket */
1359 	if (sock->type == SOCK_DGRAM)
1360 		svc_udp_init(svsk);
1361 	else
1362 		svc_tcp_init(svsk);
1363 
1364 	spin_lock_bh(&serv->sv_lock);
1365 	if (!pmap_register) {
1366 		set_bit(SK_TEMP, &svsk->sk_flags);
1367 		list_add(&svsk->sk_list, &serv->sv_tempsocks);
1368 		serv->sv_tmpcnt++;
1369 	} else {
1370 		clear_bit(SK_TEMP, &svsk->sk_flags);
1371 		list_add(&svsk->sk_list, &serv->sv_permsocks);
1372 	}
1373 	spin_unlock_bh(&serv->sv_lock);
1374 
1375 	dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1376 				svsk, svsk->sk_sk);
1377 
1378 	clear_bit(SK_BUSY, &svsk->sk_flags);
1379 	svc_sock_enqueue(svsk);
1380 	return svsk;
1381 }
1382 
1383 /*
1384  * Create socket for RPC service.
1385  */
1386 static int
1387 svc_create_socket(struct svc_serv *serv, int protocol, struct sockaddr_in *sin)
1388 {
1389 	struct svc_sock	*svsk;
1390 	struct socket	*sock;
1391 	int		error;
1392 	int		type;
1393 
1394 	dprintk("svc: svc_create_socket(%s, %d, %u.%u.%u.%u:%d)\n",
1395 				serv->sv_program->pg_name, protocol,
1396 				NIPQUAD(sin->sin_addr.s_addr),
1397 				ntohs(sin->sin_port));
1398 
1399 	if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1400 		printk(KERN_WARNING "svc: only UDP and TCP "
1401 				"sockets supported\n");
1402 		return -EINVAL;
1403 	}
1404 	type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1405 
1406 	if ((error = sock_create_kern(PF_INET, type, protocol, &sock)) < 0)
1407 		return error;
1408 
1409 	if (sin != NULL) {
1410 		if (type == SOCK_STREAM)
1411 			sock->sk->sk_reuse = 1; /* allow address reuse */
1412 		error = sock->ops->bind(sock, (struct sockaddr *) sin,
1413 						sizeof(*sin));
1414 		if (error < 0)
1415 			goto bummer;
1416 	}
1417 
1418 	if (protocol == IPPROTO_TCP) {
1419 		if ((error = sock->ops->listen(sock, 64)) < 0)
1420 			goto bummer;
1421 	}
1422 
1423 	if ((svsk = svc_setup_socket(serv, sock, &error, 1)) != NULL)
1424 		return 0;
1425 
1426 bummer:
1427 	dprintk("svc: svc_create_socket error = %d\n", -error);
1428 	sock_release(sock);
1429 	return error;
1430 }
1431 
1432 /*
1433  * Remove a dead socket
1434  */
1435 void
1436 svc_delete_socket(struct svc_sock *svsk)
1437 {
1438 	struct svc_serv	*serv;
1439 	struct sock	*sk;
1440 
1441 	dprintk("svc: svc_delete_socket(%p)\n", svsk);
1442 
1443 	serv = svsk->sk_server;
1444 	sk = svsk->sk_sk;
1445 
1446 	sk->sk_state_change = svsk->sk_ostate;
1447 	sk->sk_data_ready = svsk->sk_odata;
1448 	sk->sk_write_space = svsk->sk_owspace;
1449 
1450 	spin_lock_bh(&serv->sv_lock);
1451 
1452 	list_del_init(&svsk->sk_list);
1453 	list_del_init(&svsk->sk_ready);
1454 	if (!test_and_set_bit(SK_DEAD, &svsk->sk_flags))
1455 		if (test_bit(SK_TEMP, &svsk->sk_flags))
1456 			serv->sv_tmpcnt--;
1457 
1458 	if (!svsk->sk_inuse) {
1459 		spin_unlock_bh(&serv->sv_lock);
1460 		sock_release(svsk->sk_sock);
1461 		kfree(svsk);
1462 	} else {
1463 		spin_unlock_bh(&serv->sv_lock);
1464 		dprintk(KERN_NOTICE "svc: server socket destroy delayed\n");
1465 		/* svsk->sk_server = NULL; */
1466 	}
1467 }
1468 
1469 /*
1470  * Make a socket for nfsd and lockd
1471  */
1472 int
1473 svc_makesock(struct svc_serv *serv, int protocol, unsigned short port)
1474 {
1475 	struct sockaddr_in	sin;
1476 
1477 	dprintk("svc: creating socket proto = %d\n", protocol);
1478 	sin.sin_family      = AF_INET;
1479 	sin.sin_addr.s_addr = INADDR_ANY;
1480 	sin.sin_port        = htons(port);
1481 	return svc_create_socket(serv, protocol, &sin);
1482 }
1483 
1484 /*
1485  * Handle defer and revisit of requests
1486  */
1487 
1488 static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1489 {
1490 	struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle);
1491 	struct svc_serv *serv = dreq->owner;
1492 	struct svc_sock *svsk;
1493 
1494 	if (too_many) {
1495 		svc_sock_put(dr->svsk);
1496 		kfree(dr);
1497 		return;
1498 	}
1499 	dprintk("revisit queued\n");
1500 	svsk = dr->svsk;
1501 	dr->svsk = NULL;
1502 	spin_lock_bh(&serv->sv_lock);
1503 	list_add(&dr->handle.recent, &svsk->sk_deferred);
1504 	spin_unlock_bh(&serv->sv_lock);
1505 	set_bit(SK_DEFERRED, &svsk->sk_flags);
1506 	svc_sock_enqueue(svsk);
1507 	svc_sock_put(svsk);
1508 }
1509 
1510 static struct cache_deferred_req *
1511 svc_defer(struct cache_req *req)
1512 {
1513 	struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
1514 	int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len);
1515 	struct svc_deferred_req *dr;
1516 
1517 	if (rqstp->rq_arg.page_len)
1518 		return NULL; /* if more than a page, give up FIXME */
1519 	if (rqstp->rq_deferred) {
1520 		dr = rqstp->rq_deferred;
1521 		rqstp->rq_deferred = NULL;
1522 	} else {
1523 		int skip  = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1524 		/* FIXME maybe discard if size too large */
1525 		dr = kmalloc(size, GFP_KERNEL);
1526 		if (dr == NULL)
1527 			return NULL;
1528 
1529 		dr->handle.owner = rqstp->rq_server;
1530 		dr->prot = rqstp->rq_prot;
1531 		dr->addr = rqstp->rq_addr;
1532 		dr->argslen = rqstp->rq_arg.len >> 2;
1533 		memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2);
1534 	}
1535 	spin_lock_bh(&rqstp->rq_server->sv_lock);
1536 	rqstp->rq_sock->sk_inuse++;
1537 	dr->svsk = rqstp->rq_sock;
1538 	spin_unlock_bh(&rqstp->rq_server->sv_lock);
1539 
1540 	dr->handle.revisit = svc_revisit;
1541 	return &dr->handle;
1542 }
1543 
1544 /*
1545  * recv data from a deferred request into an active one
1546  */
1547 static int svc_deferred_recv(struct svc_rqst *rqstp)
1548 {
1549 	struct svc_deferred_req *dr = rqstp->rq_deferred;
1550 
1551 	rqstp->rq_arg.head[0].iov_base = dr->args;
1552 	rqstp->rq_arg.head[0].iov_len = dr->argslen<<2;
1553 	rqstp->rq_arg.page_len = 0;
1554 	rqstp->rq_arg.len = dr->argslen<<2;
1555 	rqstp->rq_prot        = dr->prot;
1556 	rqstp->rq_addr        = dr->addr;
1557 	return dr->argslen<<2;
1558 }
1559 
1560 
1561 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk)
1562 {
1563 	struct svc_deferred_req *dr = NULL;
1564 	struct svc_serv	*serv = svsk->sk_server;
1565 
1566 	if (!test_bit(SK_DEFERRED, &svsk->sk_flags))
1567 		return NULL;
1568 	spin_lock_bh(&serv->sv_lock);
1569 	clear_bit(SK_DEFERRED, &svsk->sk_flags);
1570 	if (!list_empty(&svsk->sk_deferred)) {
1571 		dr = list_entry(svsk->sk_deferred.next,
1572 				struct svc_deferred_req,
1573 				handle.recent);
1574 		list_del_init(&dr->handle.recent);
1575 		set_bit(SK_DEFERRED, &svsk->sk_flags);
1576 	}
1577 	spin_unlock_bh(&serv->sv_lock);
1578 	return dr;
1579 }
1580