2 * linux/net/sunrpc/svcsock.c
4 * These are the RPC server socket internals.
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_xprt_enqueue procedure...
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.
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/errno.h>
25 #include <linux/fcntl.h>
26 #include <linux/net.h>
28 #include <linux/inet.h>
29 #include <linux/udp.h>
30 #include <linux/tcp.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/netdevice.h>
34 #include <linux/skbuff.h>
35 #include <linux/file.h>
36 #include <linux/freezer.h>
38 #include <net/checksum.h>
41 #include <net/tcp_states.h>
42 #include <asm/uaccess.h>
43 #include <asm/ioctls.h>
45 #include <linux/sunrpc/types.h>
46 #include <linux/sunrpc/clnt.h>
47 #include <linux/sunrpc/xdr.h>
48 #include <linux/sunrpc/svcsock.h>
49 #include <linux/sunrpc/stats.h>
51 /* SMP locking strategy:
53 * svc_pool->sp_lock protects most of the fields of that pool.
54 * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
55 * when both need to be taken (rare), svc_serv->sv_lock is first.
56 * BKL protects svc_serv->sv_nrthread.
57 * svc_sock->sk_lock protects the svc_sock->sk_deferred list
58 * and the ->sk_info_authunix cache.
59 * svc_sock->sk_xprt.xpt_flags.XPT_BUSY prevents a svc_sock being
62 * Some flags can be set to certain values at any time
63 * providing that certain rules are followed:
65 * XPT_CONN, XPT_DATA, can be set or cleared at any time.
66 * after a set, svc_xprt_enqueue must be called.
67 * after a clear, the socket must be read/accepted
68 * if this succeeds, it must be set again.
69 * XPT_CLOSE can set at any time. It is never cleared.
70 * xpt_ref contains a bias of '1' until XPT_DEAD is set.
71 * so when xprt_ref hits zero, we know the transport is dead
72 * and no-one is using it.
73 * XPT_DEAD can only be set while XPT_BUSY is held which ensures
74 * no other thread will be using the socket or will try to
79 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
82 static struct svc_sock
*svc_setup_socket(struct svc_serv
*, struct socket
*,
83 int *errp
, int flags
);
84 static void svc_delete_xprt(struct svc_xprt
*xprt
);
85 static void svc_udp_data_ready(struct sock
*, int);
86 static int svc_udp_recvfrom(struct svc_rqst
*);
87 static int svc_udp_sendto(struct svc_rqst
*);
88 static void svc_close_xprt(struct svc_xprt
*xprt
);
89 static void svc_sock_detach(struct svc_xprt
*);
90 static void svc_sock_free(struct svc_xprt
*);
92 static struct svc_deferred_req
*svc_deferred_dequeue(struct svc_xprt
*xprt
);
93 static int svc_deferred_recv(struct svc_rqst
*rqstp
);
94 static struct cache_deferred_req
*svc_defer(struct cache_req
*req
);
95 static struct svc_xprt
*svc_create_socket(struct svc_serv
*, int,
96 struct sockaddr
*, int, int);
97 static void svc_age_temp_xprts(unsigned long closure
);
99 /* apparently the "standard" is that clients close
100 * idle connections after 5 minutes, servers after
102 * http://www.connectathon.org/talks96/nfstcp.pdf
104 static int svc_conn_age_period
= 6*60;
106 #ifdef CONFIG_DEBUG_LOCK_ALLOC
107 static struct lock_class_key svc_key
[2];
108 static struct lock_class_key svc_slock_key
[2];
110 static inline void svc_reclassify_socket(struct socket
*sock
)
112 struct sock
*sk
= sock
->sk
;
113 BUG_ON(sock_owned_by_user(sk
));
114 switch (sk
->sk_family
) {
116 sock_lock_init_class_and_name(sk
, "slock-AF_INET-NFSD",
118 "sk_xprt.xpt_lock-AF_INET-NFSD",
123 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-NFSD",
125 "sk_xprt.xpt_lock-AF_INET6-NFSD",
134 static inline void svc_reclassify_socket(struct socket
*sock
)
139 static char *__svc_print_addr(struct sockaddr
*addr
, char *buf
, size_t len
)
141 switch (addr
->sa_family
) {
143 snprintf(buf
, len
, "%u.%u.%u.%u, port=%u",
144 NIPQUAD(((struct sockaddr_in
*) addr
)->sin_addr
),
145 ntohs(((struct sockaddr_in
*) addr
)->sin_port
));
149 snprintf(buf
, len
, "%x:%x:%x:%x:%x:%x:%x:%x, port=%u",
150 NIP6(((struct sockaddr_in6
*) addr
)->sin6_addr
),
151 ntohs(((struct sockaddr_in6
*) addr
)->sin6_port
));
155 snprintf(buf
, len
, "unknown address type: %d", addr
->sa_family
);
162 * svc_print_addr - Format rq_addr field for printing
163 * @rqstp: svc_rqst struct containing address to print
164 * @buf: target buffer for formatted address
165 * @len: length of target buffer
168 char *svc_print_addr(struct svc_rqst
*rqstp
, char *buf
, size_t len
)
170 return __svc_print_addr(svc_addr(rqstp
), buf
, len
);
172 EXPORT_SYMBOL_GPL(svc_print_addr
);
175 * Queue up an idle server thread. Must have pool->sp_lock held.
176 * Note: this is really a stack rather than a queue, so that we only
177 * use as many different threads as we need, and the rest don't pollute
181 svc_thread_enqueue(struct svc_pool
*pool
, struct svc_rqst
*rqstp
)
183 list_add(&rqstp
->rq_list
, &pool
->sp_threads
);
187 * Dequeue an nfsd thread. Must have pool->sp_lock held.
190 svc_thread_dequeue(struct svc_pool
*pool
, struct svc_rqst
*rqstp
)
192 list_del(&rqstp
->rq_list
);
196 * Release an skbuff after use
198 static void svc_release_skb(struct svc_rqst
*rqstp
)
200 struct sk_buff
*skb
= rqstp
->rq_xprt_ctxt
;
201 struct svc_deferred_req
*dr
= rqstp
->rq_deferred
;
204 struct svc_sock
*svsk
=
205 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
206 rqstp
->rq_xprt_ctxt
= NULL
;
208 dprintk("svc: service %p, releasing skb %p\n", rqstp
, skb
);
209 skb_free_datagram(svsk
->sk_sk
, skb
);
212 rqstp
->rq_deferred
= NULL
;
218 * Queue up a socket with data pending. If there are idle nfsd
219 * processes, wake 'em up.
222 void svc_xprt_enqueue(struct svc_xprt
*xprt
)
224 struct svc_serv
*serv
= xprt
->xpt_server
;
225 struct svc_pool
*pool
;
226 struct svc_rqst
*rqstp
;
229 if (!(xprt
->xpt_flags
&
230 ((1<<XPT_CONN
)|(1<<XPT_DATA
)|(1<<XPT_CLOSE
)|(1<<XPT_DEFERRED
))))
232 if (test_bit(XPT_DEAD
, &xprt
->xpt_flags
))
236 pool
= svc_pool_for_cpu(xprt
->xpt_server
, cpu
);
239 spin_lock_bh(&pool
->sp_lock
);
241 if (!list_empty(&pool
->sp_threads
) &&
242 !list_empty(&pool
->sp_sockets
))
245 "threads and transports both waiting??\n");
247 if (test_bit(XPT_DEAD
, &xprt
->xpt_flags
)) {
248 /* Don't enqueue dead sockets */
249 dprintk("svc: transport %p is dead, not enqueued\n", xprt
);
253 /* Mark socket as busy. It will remain in this state until the
254 * server has processed all pending data and put the socket back
255 * on the idle list. We update XPT_BUSY atomically because
256 * it also guards against trying to enqueue the svc_sock twice.
258 if (test_and_set_bit(XPT_BUSY
, &xprt
->xpt_flags
)) {
259 /* Don't enqueue socket while already enqueued */
260 dprintk("svc: transport %p busy, not enqueued\n", xprt
);
263 BUG_ON(xprt
->xpt_pool
!= NULL
);
264 xprt
->xpt_pool
= pool
;
266 /* Handle pending connection */
267 if (test_bit(XPT_CONN
, &xprt
->xpt_flags
))
270 /* Handle close in-progress */
271 if (test_bit(XPT_CLOSE
, &xprt
->xpt_flags
))
274 /* Check if we have space to reply to a request */
275 if (!xprt
->xpt_ops
->xpo_has_wspace(xprt
)) {
276 /* Don't enqueue while not enough space for reply */
277 dprintk("svc: no write space, transport %p not enqueued\n",
279 xprt
->xpt_pool
= NULL
;
280 clear_bit(XPT_BUSY
, &xprt
->xpt_flags
);
285 if (!list_empty(&pool
->sp_threads
)) {
286 rqstp
= list_entry(pool
->sp_threads
.next
,
289 dprintk("svc: transport %p served by daemon %p\n",
291 svc_thread_dequeue(pool
, rqstp
);
294 "svc_xprt_enqueue: server %p, rq_xprt=%p!\n",
295 rqstp
, rqstp
->rq_xprt
);
296 rqstp
->rq_xprt
= xprt
;
298 rqstp
->rq_reserved
= serv
->sv_max_mesg
;
299 atomic_add(rqstp
->rq_reserved
, &xprt
->xpt_reserved
);
300 BUG_ON(xprt
->xpt_pool
!= pool
);
301 wake_up(&rqstp
->rq_wait
);
303 dprintk("svc: transport %p put into queue\n", xprt
);
304 list_add_tail(&xprt
->xpt_ready
, &pool
->sp_sockets
);
305 BUG_ON(xprt
->xpt_pool
!= pool
);
309 spin_unlock_bh(&pool
->sp_lock
);
311 EXPORT_SYMBOL_GPL(svc_xprt_enqueue
);
314 * Dequeue the first socket. Must be called with the pool->sp_lock held.
316 static struct svc_xprt
*svc_xprt_dequeue(struct svc_pool
*pool
)
318 struct svc_xprt
*xprt
;
320 if (list_empty(&pool
->sp_sockets
))
323 xprt
= list_entry(pool
->sp_sockets
.next
,
324 struct svc_xprt
, xpt_ready
);
325 list_del_init(&xprt
->xpt_ready
);
327 dprintk("svc: transport %p dequeued, inuse=%d\n",
328 xprt
, atomic_read(&xprt
->xpt_ref
.refcount
));
334 * svc_xprt_received conditionally queues the transport for processing
335 * by another thread. The caller must hold the XPT_BUSY bit and must
336 * not thereafter touch transport data.
338 * Note: XPT_DATA only gets cleared when a read-attempt finds no (or
339 * insufficient) data.
341 void svc_xprt_received(struct svc_xprt
*xprt
)
343 BUG_ON(!test_bit(XPT_BUSY
, &xprt
->xpt_flags
));
344 xprt
->xpt_pool
= NULL
;
345 clear_bit(XPT_BUSY
, &xprt
->xpt_flags
);
346 svc_xprt_enqueue(xprt
);
348 EXPORT_SYMBOL_GPL(svc_xprt_received
);
351 * svc_reserve - change the space reserved for the reply to a request.
352 * @rqstp: The request in question
353 * @space: new max space to reserve
355 * Each request reserves some space on the output queue of the socket
356 * to make sure the reply fits. This function reduces that reserved
357 * space to be the amount of space used already, plus @space.
360 void svc_reserve(struct svc_rqst
*rqstp
, int space
)
362 space
+= rqstp
->rq_res
.head
[0].iov_len
;
364 if (space
< rqstp
->rq_reserved
) {
365 struct svc_xprt
*xprt
= rqstp
->rq_xprt
;
366 atomic_sub((rqstp
->rq_reserved
- space
), &xprt
->xpt_reserved
);
367 rqstp
->rq_reserved
= space
;
369 svc_xprt_enqueue(xprt
);
373 static void svc_xprt_release(struct svc_rqst
*rqstp
)
375 struct svc_xprt
*xprt
= rqstp
->rq_xprt
;
377 rqstp
->rq_xprt
->xpt_ops
->xpo_release_rqst(rqstp
);
379 svc_free_res_pages(rqstp
);
380 rqstp
->rq_res
.page_len
= 0;
381 rqstp
->rq_res
.page_base
= 0;
383 /* Reset response buffer and release
385 * But first, check that enough space was reserved
386 * for the reply, otherwise we have a bug!
388 if ((rqstp
->rq_res
.len
) > rqstp
->rq_reserved
)
389 printk(KERN_ERR
"RPC request reserved %d but used %d\n",
393 rqstp
->rq_res
.head
[0].iov_len
= 0;
394 svc_reserve(rqstp
, 0);
395 rqstp
->rq_xprt
= NULL
;
401 * External function to wake up a server waiting for data
402 * This really only makes sense for services like lockd
403 * which have exactly one thread anyway.
406 svc_wake_up(struct svc_serv
*serv
)
408 struct svc_rqst
*rqstp
;
410 struct svc_pool
*pool
;
412 for (i
= 0; i
< serv
->sv_nrpools
; i
++) {
413 pool
= &serv
->sv_pools
[i
];
415 spin_lock_bh(&pool
->sp_lock
);
416 if (!list_empty(&pool
->sp_threads
)) {
417 rqstp
= list_entry(pool
->sp_threads
.next
,
420 dprintk("svc: daemon %p woken up.\n", rqstp
);
422 svc_thread_dequeue(pool, rqstp);
423 rqstp->rq_xprt = NULL;
425 wake_up(&rqstp
->rq_wait
);
427 spin_unlock_bh(&pool
->sp_lock
);
431 union svc_pktinfo_u
{
432 struct in_pktinfo pkti
;
433 struct in6_pktinfo pkti6
;
435 #define SVC_PKTINFO_SPACE \
436 CMSG_SPACE(sizeof(union svc_pktinfo_u))
438 static void svc_set_cmsg_data(struct svc_rqst
*rqstp
, struct cmsghdr
*cmh
)
440 struct svc_sock
*svsk
=
441 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
442 switch (svsk
->sk_sk
->sk_family
) {
444 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
446 cmh
->cmsg_level
= SOL_IP
;
447 cmh
->cmsg_type
= IP_PKTINFO
;
448 pki
->ipi_ifindex
= 0;
449 pki
->ipi_spec_dst
.s_addr
= rqstp
->rq_daddr
.addr
.s_addr
;
450 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
455 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
457 cmh
->cmsg_level
= SOL_IPV6
;
458 cmh
->cmsg_type
= IPV6_PKTINFO
;
459 pki
->ipi6_ifindex
= 0;
460 ipv6_addr_copy(&pki
->ipi6_addr
,
461 &rqstp
->rq_daddr
.addr6
);
462 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
470 * Generic sendto routine
473 svc_sendto(struct svc_rqst
*rqstp
, struct xdr_buf
*xdr
)
475 struct svc_sock
*svsk
=
476 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
477 struct socket
*sock
= svsk
->sk_sock
;
481 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
483 struct cmsghdr
*cmh
= &buffer
.hdr
;
487 struct page
**ppage
= xdr
->pages
;
488 size_t base
= xdr
->page_base
;
489 unsigned int pglen
= xdr
->page_len
;
490 unsigned int flags
= MSG_MORE
;
491 char buf
[RPC_MAX_ADDRBUFLEN
];
495 if (rqstp
->rq_prot
== IPPROTO_UDP
) {
496 struct msghdr msg
= {
497 .msg_name
= &rqstp
->rq_addr
,
498 .msg_namelen
= rqstp
->rq_addrlen
,
500 .msg_controllen
= sizeof(buffer
),
501 .msg_flags
= MSG_MORE
,
504 svc_set_cmsg_data(rqstp
, cmh
);
506 if (sock_sendmsg(sock
, &msg
, 0) < 0)
511 if (slen
== xdr
->head
[0].iov_len
)
513 len
= kernel_sendpage(sock
, rqstp
->rq_respages
[0], 0,
514 xdr
->head
[0].iov_len
, flags
);
515 if (len
!= xdr
->head
[0].iov_len
)
517 slen
-= xdr
->head
[0].iov_len
;
522 size
= PAGE_SIZE
- base
< pglen
? PAGE_SIZE
- base
: pglen
;
526 result
= kernel_sendpage(sock
, *ppage
, base
, size
, flags
);
533 size
= PAGE_SIZE
< pglen
? PAGE_SIZE
: pglen
;
538 if (xdr
->tail
[0].iov_len
) {
539 result
= kernel_sendpage(sock
, rqstp
->rq_respages
[0],
540 ((unsigned long)xdr
->tail
[0].iov_base
)
542 xdr
->tail
[0].iov_len
, 0);
548 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
549 svsk
, xdr
->head
[0].iov_base
, xdr
->head
[0].iov_len
,
550 xdr
->len
, len
, svc_print_addr(rqstp
, buf
, sizeof(buf
)));
556 * Report socket names for nfsdfs
558 static int one_sock_name(char *buf
, struct svc_sock
*svsk
)
562 switch(svsk
->sk_sk
->sk_family
) {
564 len
= sprintf(buf
, "ipv4 %s %u.%u.%u.%u %d\n",
565 svsk
->sk_sk
->sk_protocol
==IPPROTO_UDP
?
567 NIPQUAD(inet_sk(svsk
->sk_sk
)->rcv_saddr
),
568 inet_sk(svsk
->sk_sk
)->num
);
571 len
= sprintf(buf
, "*unknown-%d*\n",
572 svsk
->sk_sk
->sk_family
);
578 svc_sock_names(char *buf
, struct svc_serv
*serv
, char *toclose
)
580 struct svc_sock
*svsk
, *closesk
= NULL
;
585 spin_lock_bh(&serv
->sv_lock
);
586 list_for_each_entry(svsk
, &serv
->sv_permsocks
, sk_xprt
.xpt_list
) {
587 int onelen
= one_sock_name(buf
+len
, svsk
);
588 if (toclose
&& strcmp(toclose
, buf
+len
) == 0)
593 spin_unlock_bh(&serv
->sv_lock
);
595 /* Should unregister with portmap, but you cannot
596 * unregister just one protocol...
598 svc_close_xprt(&closesk
->sk_xprt
);
603 EXPORT_SYMBOL(svc_sock_names
);
606 * Check input queue length
609 svc_recv_available(struct svc_sock
*svsk
)
611 struct socket
*sock
= svsk
->sk_sock
;
614 err
= kernel_sock_ioctl(sock
, TIOCINQ
, (unsigned long) &avail
);
616 return (err
>= 0)? avail
: err
;
620 * Generic recvfrom routine.
623 svc_recvfrom(struct svc_rqst
*rqstp
, struct kvec
*iov
, int nr
, int buflen
)
625 struct svc_sock
*svsk
=
626 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
627 struct msghdr msg
= {
628 .msg_flags
= MSG_DONTWAIT
,
632 len
= kernel_recvmsg(svsk
->sk_sock
, &msg
, iov
, nr
, buflen
,
635 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
636 svsk
, iov
[0].iov_base
, iov
[0].iov_len
, len
);
641 * Set socket snd and rcv buffer lengths
644 svc_sock_setbufsize(struct socket
*sock
, unsigned int snd
, unsigned int rcv
)
648 oldfs
= get_fs(); set_fs(KERNEL_DS
);
649 sock_setsockopt(sock
, SOL_SOCKET
, SO_SNDBUF
,
650 (char*)&snd
, sizeof(snd
));
651 sock_setsockopt(sock
, SOL_SOCKET
, SO_RCVBUF
,
652 (char*)&rcv
, sizeof(rcv
));
654 /* sock_setsockopt limits use to sysctl_?mem_max,
655 * which isn't acceptable. Until that is made conditional
656 * on not having CAP_SYS_RESOURCE or similar, we go direct...
657 * DaveM said I could!
660 sock
->sk
->sk_sndbuf
= snd
* 2;
661 sock
->sk
->sk_rcvbuf
= rcv
* 2;
662 sock
->sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
|SOCK_RCVBUF_LOCK
;
663 release_sock(sock
->sk
);
667 * INET callback when data has been received on the socket.
670 svc_udp_data_ready(struct sock
*sk
, int count
)
672 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
675 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
677 test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
678 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
679 svc_xprt_enqueue(&svsk
->sk_xprt
);
681 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
682 wake_up_interruptible(sk
->sk_sleep
);
686 * INET callback when space is newly available on the socket.
689 svc_write_space(struct sock
*sk
)
691 struct svc_sock
*svsk
= (struct svc_sock
*)(sk
->sk_user_data
);
694 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
695 svsk
, sk
, test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
696 svc_xprt_enqueue(&svsk
->sk_xprt
);
699 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
)) {
700 dprintk("RPC svc_write_space: someone sleeping on %p\n",
702 wake_up_interruptible(sk
->sk_sleep
);
707 * Copy the UDP datagram's destination address to the rqstp structure.
708 * The 'destination' address in this case is the address to which the
709 * peer sent the datagram, i.e. our local address. For multihomed
710 * hosts, this can change from msg to msg. Note that only the IP
711 * address changes, the port number should remain the same.
713 static void svc_udp_get_dest_address(struct svc_rqst
*rqstp
,
716 struct svc_sock
*svsk
=
717 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
718 switch (svsk
->sk_sk
->sk_family
) {
720 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
721 rqstp
->rq_daddr
.addr
.s_addr
= pki
->ipi_spec_dst
.s_addr
;
725 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
726 ipv6_addr_copy(&rqstp
->rq_daddr
.addr6
, &pki
->ipi6_addr
);
733 * Receive a datagram from a UDP socket.
736 svc_udp_recvfrom(struct svc_rqst
*rqstp
)
738 struct svc_sock
*svsk
=
739 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
740 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
744 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
746 struct cmsghdr
*cmh
= &buffer
.hdr
;
748 struct msghdr msg
= {
749 .msg_name
= svc_addr(rqstp
),
751 .msg_controllen
= sizeof(buffer
),
752 .msg_flags
= MSG_DONTWAIT
,
755 if (test_and_clear_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
))
756 /* udp sockets need large rcvbuf as all pending
757 * requests are still in that buffer. sndbuf must
758 * also be large enough that there is enough space
759 * for one reply per thread. We count all threads
760 * rather than threads in a particular pool, which
761 * provides an upper bound on the number of threads
762 * which will access the socket.
764 svc_sock_setbufsize(svsk
->sk_sock
,
765 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
,
766 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
);
768 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
770 err
= kernel_recvmsg(svsk
->sk_sock
, &msg
, NULL
,
771 0, 0, MSG_PEEK
| MSG_DONTWAIT
);
773 skb
= skb_recv_datagram(svsk
->sk_sk
, 0, 1, &err
);
776 if (err
!= -EAGAIN
) {
777 /* possibly an icmp error */
778 dprintk("svc: recvfrom returned error %d\n", -err
);
779 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
781 svc_xprt_received(&svsk
->sk_xprt
);
784 len
= svc_addr_len(svc_addr(rqstp
));
787 rqstp
->rq_addrlen
= len
;
788 if (skb
->tstamp
.tv64
== 0) {
789 skb
->tstamp
= ktime_get_real();
790 /* Don't enable netstamp, sunrpc doesn't
791 need that much accuracy */
793 svsk
->sk_sk
->sk_stamp
= skb
->tstamp
;
794 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
); /* there may be more data... */
797 * Maybe more packets - kick another thread ASAP.
799 svc_xprt_received(&svsk
->sk_xprt
);
801 len
= skb
->len
- sizeof(struct udphdr
);
802 rqstp
->rq_arg
.len
= len
;
804 rqstp
->rq_prot
= IPPROTO_UDP
;
806 if (cmh
->cmsg_level
!= IPPROTO_IP
||
807 cmh
->cmsg_type
!= IP_PKTINFO
) {
809 printk("rpcsvc: received unknown control message:"
811 cmh
->cmsg_level
, cmh
->cmsg_type
);
812 skb_free_datagram(svsk
->sk_sk
, skb
);
815 svc_udp_get_dest_address(rqstp
, cmh
);
817 if (skb_is_nonlinear(skb
)) {
818 /* we have to copy */
820 if (csum_partial_copy_to_xdr(&rqstp
->rq_arg
, skb
)) {
823 skb_free_datagram(svsk
->sk_sk
, skb
);
827 skb_free_datagram(svsk
->sk_sk
, skb
);
829 /* we can use it in-place */
830 rqstp
->rq_arg
.head
[0].iov_base
= skb
->data
+ sizeof(struct udphdr
);
831 rqstp
->rq_arg
.head
[0].iov_len
= len
;
832 if (skb_checksum_complete(skb
)) {
833 skb_free_datagram(svsk
->sk_sk
, skb
);
836 rqstp
->rq_xprt_ctxt
= skb
;
839 rqstp
->rq_arg
.page_base
= 0;
840 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
841 rqstp
->rq_arg
.head
[0].iov_len
= len
;
842 rqstp
->rq_arg
.page_len
= 0;
843 rqstp
->rq_respages
= rqstp
->rq_pages
+1;
845 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
846 rqstp
->rq_respages
= rqstp
->rq_pages
+ 1 +
847 DIV_ROUND_UP(rqstp
->rq_arg
.page_len
, PAGE_SIZE
);
851 serv
->sv_stats
->netudpcnt
++;
857 svc_udp_sendto(struct svc_rqst
*rqstp
)
861 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
862 if (error
== -ECONNREFUSED
)
863 /* ICMP error on earlier request. */
864 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
869 static void svc_udp_prep_reply_hdr(struct svc_rqst
*rqstp
)
873 static int svc_udp_has_wspace(struct svc_xprt
*xprt
)
875 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
876 struct svc_serv
*serv
= xprt
->xpt_server
;
877 unsigned long required
;
880 * Set the SOCK_NOSPACE flag before checking the available
883 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
884 required
= atomic_read(&svsk
->sk_xprt
.xpt_reserved
) + serv
->sv_max_mesg
;
885 if (required
*2 > sock_wspace(svsk
->sk_sk
))
887 clear_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
891 static struct svc_xprt
*svc_udp_accept(struct svc_xprt
*xprt
)
897 static struct svc_xprt
*svc_udp_create(struct svc_serv
*serv
,
898 struct sockaddr
*sa
, int salen
,
901 return svc_create_socket(serv
, IPPROTO_UDP
, sa
, salen
, flags
);
904 static struct svc_xprt_ops svc_udp_ops
= {
905 .xpo_create
= svc_udp_create
,
906 .xpo_recvfrom
= svc_udp_recvfrom
,
907 .xpo_sendto
= svc_udp_sendto
,
908 .xpo_release_rqst
= svc_release_skb
,
909 .xpo_detach
= svc_sock_detach
,
910 .xpo_free
= svc_sock_free
,
911 .xpo_prep_reply_hdr
= svc_udp_prep_reply_hdr
,
912 .xpo_has_wspace
= svc_udp_has_wspace
,
913 .xpo_accept
= svc_udp_accept
,
916 static struct svc_xprt_class svc_udp_class
= {
918 .xcl_owner
= THIS_MODULE
,
919 .xcl_ops
= &svc_udp_ops
,
920 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_UDP
,
923 static void svc_udp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
928 svc_xprt_init(&svc_udp_class
, &svsk
->sk_xprt
, serv
);
929 clear_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
930 svsk
->sk_sk
->sk_data_ready
= svc_udp_data_ready
;
931 svsk
->sk_sk
->sk_write_space
= svc_write_space
;
933 /* initialise setting must have enough space to
934 * receive and respond to one request.
935 * svc_udp_recvfrom will re-adjust if necessary
937 svc_sock_setbufsize(svsk
->sk_sock
,
938 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
,
939 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
);
941 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
); /* might have come in before data_ready set up */
942 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
946 /* make sure we get destination address info */
947 svsk
->sk_sock
->ops
->setsockopt(svsk
->sk_sock
, IPPROTO_IP
, IP_PKTINFO
,
948 (char __user
*)&one
, sizeof(one
));
953 * A data_ready event on a listening socket means there's a connection
954 * pending. Do not use state_change as a substitute for it.
957 svc_tcp_listen_data_ready(struct sock
*sk
, int count_unused
)
959 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
961 dprintk("svc: socket %p TCP (listen) state change %d\n",
965 * This callback may called twice when a new connection
966 * is established as a child socket inherits everything
967 * from a parent LISTEN socket.
968 * 1) data_ready method of the parent socket will be called
969 * when one of child sockets become ESTABLISHED.
970 * 2) data_ready method of the child socket may be called
971 * when it receives data before the socket is accepted.
972 * In case of 2, we should ignore it silently.
974 if (sk
->sk_state
== TCP_LISTEN
) {
976 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
977 svc_xprt_enqueue(&svsk
->sk_xprt
);
979 printk("svc: socket %p: no user data\n", sk
);
982 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
983 wake_up_interruptible_all(sk
->sk_sleep
);
987 * A state change on a connected socket means it's dying or dead.
990 svc_tcp_state_change(struct sock
*sk
)
992 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
994 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
995 sk
, sk
->sk_state
, sk
->sk_user_data
);
998 printk("svc: socket %p: no user data\n", sk
);
1000 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1001 svc_xprt_enqueue(&svsk
->sk_xprt
);
1003 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1004 wake_up_interruptible_all(sk
->sk_sleep
);
1008 svc_tcp_data_ready(struct sock
*sk
, int count
)
1010 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
1012 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
1013 sk
, sk
->sk_user_data
);
1015 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
1016 svc_xprt_enqueue(&svsk
->sk_xprt
);
1018 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1019 wake_up_interruptible(sk
->sk_sleep
);
1022 static inline int svc_port_is_privileged(struct sockaddr
*sin
)
1024 switch (sin
->sa_family
) {
1026 return ntohs(((struct sockaddr_in
*)sin
)->sin_port
)
1029 return ntohs(((struct sockaddr_in6
*)sin
)->sin6_port
)
1037 * Accept a TCP connection
1039 static struct svc_xprt
*svc_tcp_accept(struct svc_xprt
*xprt
)
1041 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1042 struct sockaddr_storage addr
;
1043 struct sockaddr
*sin
= (struct sockaddr
*) &addr
;
1044 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
1045 struct socket
*sock
= svsk
->sk_sock
;
1046 struct socket
*newsock
;
1047 struct svc_sock
*newsvsk
;
1049 char buf
[RPC_MAX_ADDRBUFLEN
];
1051 dprintk("svc: tcp_accept %p sock %p\n", svsk
, sock
);
1055 clear_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
1056 err
= kernel_accept(sock
, &newsock
, O_NONBLOCK
);
1059 printk(KERN_WARNING
"%s: no more sockets!\n",
1061 else if (err
!= -EAGAIN
&& net_ratelimit())
1062 printk(KERN_WARNING
"%s: accept failed (err %d)!\n",
1063 serv
->sv_name
, -err
);
1066 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
1068 err
= kernel_getpeername(newsock
, sin
, &slen
);
1070 if (net_ratelimit())
1071 printk(KERN_WARNING
"%s: peername failed (err %d)!\n",
1072 serv
->sv_name
, -err
);
1073 goto failed
; /* aborted connection or whatever */
1076 /* Ideally, we would want to reject connections from unauthorized
1077 * hosts here, but when we get encryption, the IP of the host won't
1078 * tell us anything. For now just warn about unpriv connections.
1080 if (!svc_port_is_privileged(sin
)) {
1081 dprintk(KERN_WARNING
1082 "%s: connect from unprivileged port: %s\n",
1084 __svc_print_addr(sin
, buf
, sizeof(buf
)));
1086 dprintk("%s: connect from %s\n", serv
->sv_name
,
1087 __svc_print_addr(sin
, buf
, sizeof(buf
)));
1089 /* make sure that a write doesn't block forever when
1092 newsock
->sk
->sk_sndtimeo
= HZ
*30;
1094 if (!(newsvsk
= svc_setup_socket(serv
, newsock
, &err
,
1095 (SVC_SOCK_ANONYMOUS
| SVC_SOCK_TEMPORARY
))))
1097 svc_xprt_set_remote(&newsvsk
->sk_xprt
, sin
, slen
);
1098 err
= kernel_getsockname(newsock
, sin
, &slen
);
1099 if (unlikely(err
< 0)) {
1100 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err
);
1101 slen
= offsetof(struct sockaddr
, sa_data
);
1103 svc_xprt_set_local(&newsvsk
->sk_xprt
, sin
, slen
);
1106 serv
->sv_stats
->nettcpconn
++;
1108 return &newsvsk
->sk_xprt
;
1111 sock_release(newsock
);
1116 * Receive data from a TCP socket.
1119 svc_tcp_recvfrom(struct svc_rqst
*rqstp
)
1121 struct svc_sock
*svsk
=
1122 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
1123 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
1128 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1129 svsk
, test_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
),
1130 test_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
),
1131 test_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
));
1133 if (test_and_clear_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
))
1134 /* sndbuf needs to have room for one request
1135 * per thread, otherwise we can stall even when the
1136 * network isn't a bottleneck.
1138 * We count all threads rather than threads in a
1139 * particular pool, which provides an upper bound
1140 * on the number of threads which will access the socket.
1142 * rcvbuf just needs to be able to hold a few requests.
1143 * Normally they will be removed from the queue
1144 * as soon a a complete request arrives.
1146 svc_sock_setbufsize(svsk
->sk_sock
,
1147 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
,
1148 3 * serv
->sv_max_mesg
);
1150 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
1152 /* Receive data. If we haven't got the record length yet, get
1153 * the next four bytes. Otherwise try to gobble up as much as
1154 * possible up to the complete record length.
1156 if (svsk
->sk_tcplen
< 4) {
1157 unsigned long want
= 4 - svsk
->sk_tcplen
;
1160 iov
.iov_base
= ((char *) &svsk
->sk_reclen
) + svsk
->sk_tcplen
;
1162 if ((len
= svc_recvfrom(rqstp
, &iov
, 1, want
)) < 0)
1164 svsk
->sk_tcplen
+= len
;
1167 dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
1169 svc_xprt_received(&svsk
->sk_xprt
);
1170 return -EAGAIN
; /* record header not complete */
1173 svsk
->sk_reclen
= ntohl(svsk
->sk_reclen
);
1174 if (!(svsk
->sk_reclen
& 0x80000000)) {
1175 /* FIXME: technically, a record can be fragmented,
1176 * and non-terminal fragments will not have the top
1177 * bit set in the fragment length header.
1178 * But apparently no known nfs clients send fragmented
1180 if (net_ratelimit())
1181 printk(KERN_NOTICE
"RPC: bad TCP reclen 0x%08lx"
1182 " (non-terminal)\n",
1183 (unsigned long) svsk
->sk_reclen
);
1186 svsk
->sk_reclen
&= 0x7fffffff;
1187 dprintk("svc: TCP record, %d bytes\n", svsk
->sk_reclen
);
1188 if (svsk
->sk_reclen
> serv
->sv_max_mesg
) {
1189 if (net_ratelimit())
1190 printk(KERN_NOTICE
"RPC: bad TCP reclen 0x%08lx"
1192 (unsigned long) svsk
->sk_reclen
);
1197 /* Check whether enough data is available */
1198 len
= svc_recv_available(svsk
);
1202 if (len
< svsk
->sk_reclen
) {
1203 dprintk("svc: incomplete TCP record (%d of %d)\n",
1204 len
, svsk
->sk_reclen
);
1205 svc_xprt_received(&svsk
->sk_xprt
);
1206 return -EAGAIN
; /* record not complete */
1208 len
= svsk
->sk_reclen
;
1209 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
1211 vec
= rqstp
->rq_vec
;
1212 vec
[0] = rqstp
->rq_arg
.head
[0];
1215 while (vlen
< len
) {
1216 vec
[pnum
].iov_base
= page_address(rqstp
->rq_pages
[pnum
]);
1217 vec
[pnum
].iov_len
= PAGE_SIZE
;
1221 rqstp
->rq_respages
= &rqstp
->rq_pages
[pnum
];
1223 /* Now receive data */
1224 len
= svc_recvfrom(rqstp
, vec
, pnum
, len
);
1228 dprintk("svc: TCP complete record (%d bytes)\n", len
);
1229 rqstp
->rq_arg
.len
= len
;
1230 rqstp
->rq_arg
.page_base
= 0;
1231 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
1232 rqstp
->rq_arg
.head
[0].iov_len
= len
;
1233 rqstp
->rq_arg
.page_len
= 0;
1235 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
1238 rqstp
->rq_xprt_ctxt
= NULL
;
1239 rqstp
->rq_prot
= IPPROTO_TCP
;
1241 /* Reset TCP read info */
1242 svsk
->sk_reclen
= 0;
1243 svsk
->sk_tcplen
= 0;
1245 svc_xprt_copy_addrs(rqstp
, &svsk
->sk_xprt
);
1246 svc_xprt_received(&svsk
->sk_xprt
);
1248 serv
->sv_stats
->nettcpcnt
++;
1253 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1257 if (len
== -EAGAIN
) {
1258 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1259 svc_xprt_received(&svsk
->sk_xprt
);
1261 printk(KERN_NOTICE
"%s: recvfrom returned errno %d\n",
1262 svsk
->sk_xprt
.xpt_server
->sv_name
, -len
);
1270 * Send out data on TCP socket.
1273 svc_tcp_sendto(struct svc_rqst
*rqstp
)
1275 struct xdr_buf
*xbufp
= &rqstp
->rq_res
;
1279 /* Set up the first element of the reply kvec.
1280 * Any other kvecs that may be in use have been taken
1281 * care of by the server implementation itself.
1283 reclen
= htonl(0x80000000|((xbufp
->len
) - 4));
1284 memcpy(xbufp
->head
[0].iov_base
, &reclen
, 4);
1286 if (test_bit(XPT_DEAD
, &rqstp
->rq_xprt
->xpt_flags
))
1289 sent
= svc_sendto(rqstp
, &rqstp
->rq_res
);
1290 if (sent
!= xbufp
->len
) {
1291 printk(KERN_NOTICE
"rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1292 rqstp
->rq_xprt
->xpt_server
->sv_name
,
1293 (sent
<0)?"got error":"sent only",
1295 set_bit(XPT_CLOSE
, &rqstp
->rq_xprt
->xpt_flags
);
1296 svc_xprt_enqueue(rqstp
->rq_xprt
);
1303 * Setup response header. TCP has a 4B record length field.
1305 static void svc_tcp_prep_reply_hdr(struct svc_rqst
*rqstp
)
1307 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
1309 /* tcp needs a space for the record length... */
1313 static int svc_tcp_has_wspace(struct svc_xprt
*xprt
)
1315 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1316 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
1321 * Set the SOCK_NOSPACE flag before checking the available
1324 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
1325 required
= atomic_read(&svsk
->sk_xprt
.xpt_reserved
) + serv
->sv_max_mesg
;
1326 wspace
= sk_stream_wspace(svsk
->sk_sk
);
1328 if (wspace
< sk_stream_min_wspace(svsk
->sk_sk
))
1330 if (required
* 2 > wspace
)
1333 clear_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
1337 static struct svc_xprt
*svc_tcp_create(struct svc_serv
*serv
,
1338 struct sockaddr
*sa
, int salen
,
1341 return svc_create_socket(serv
, IPPROTO_TCP
, sa
, salen
, flags
);
1344 static struct svc_xprt_ops svc_tcp_ops
= {
1345 .xpo_create
= svc_tcp_create
,
1346 .xpo_recvfrom
= svc_tcp_recvfrom
,
1347 .xpo_sendto
= svc_tcp_sendto
,
1348 .xpo_release_rqst
= svc_release_skb
,
1349 .xpo_detach
= svc_sock_detach
,
1350 .xpo_free
= svc_sock_free
,
1351 .xpo_prep_reply_hdr
= svc_tcp_prep_reply_hdr
,
1352 .xpo_has_wspace
= svc_tcp_has_wspace
,
1353 .xpo_accept
= svc_tcp_accept
,
1356 static struct svc_xprt_class svc_tcp_class
= {
1358 .xcl_owner
= THIS_MODULE
,
1359 .xcl_ops
= &svc_tcp_ops
,
1360 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_TCP
,
1363 void svc_init_xprt_sock(void)
1365 svc_reg_xprt_class(&svc_tcp_class
);
1366 svc_reg_xprt_class(&svc_udp_class
);
1369 void svc_cleanup_xprt_sock(void)
1371 svc_unreg_xprt_class(&svc_tcp_class
);
1372 svc_unreg_xprt_class(&svc_udp_class
);
1375 static void svc_tcp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
1377 struct sock
*sk
= svsk
->sk_sk
;
1378 struct tcp_sock
*tp
= tcp_sk(sk
);
1380 svc_xprt_init(&svc_tcp_class
, &svsk
->sk_xprt
, serv
);
1381 set_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
1382 if (sk
->sk_state
== TCP_LISTEN
) {
1383 dprintk("setting up TCP socket for listening\n");
1384 set_bit(XPT_LISTENER
, &svsk
->sk_xprt
.xpt_flags
);
1385 sk
->sk_data_ready
= svc_tcp_listen_data_ready
;
1386 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
1388 dprintk("setting up TCP socket for reading\n");
1389 sk
->sk_state_change
= svc_tcp_state_change
;
1390 sk
->sk_data_ready
= svc_tcp_data_ready
;
1391 sk
->sk_write_space
= svc_write_space
;
1393 svsk
->sk_reclen
= 0;
1394 svsk
->sk_tcplen
= 0;
1396 tp
->nonagle
= 1; /* disable Nagle's algorithm */
1398 /* initialise setting must have enough space to
1399 * receive and respond to one request.
1400 * svc_tcp_recvfrom will re-adjust if necessary
1402 svc_sock_setbufsize(svsk
->sk_sock
,
1403 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
,
1404 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
);
1406 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
1407 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
1408 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1409 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1414 svc_sock_update_bufs(struct svc_serv
*serv
)
1417 * The number of server threads has changed. Update
1418 * rcvbuf and sndbuf accordingly on all sockets
1420 struct list_head
*le
;
1422 spin_lock_bh(&serv
->sv_lock
);
1423 list_for_each(le
, &serv
->sv_permsocks
) {
1424 struct svc_sock
*svsk
=
1425 list_entry(le
, struct svc_sock
, sk_xprt
.xpt_list
);
1426 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
1428 list_for_each(le
, &serv
->sv_tempsocks
) {
1429 struct svc_sock
*svsk
=
1430 list_entry(le
, struct svc_sock
, sk_xprt
.xpt_list
);
1431 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
1433 spin_unlock_bh(&serv
->sv_lock
);
1437 * Make sure that we don't have too many active connections. If we
1438 * have, something must be dropped.
1440 * There's no point in trying to do random drop here for DoS
1441 * prevention. The NFS clients does 1 reconnect in 15 seconds. An
1442 * attacker can easily beat that.
1444 * The only somewhat efficient mechanism would be if drop old
1445 * connections from the same IP first. But right now we don't even
1446 * record the client IP in svc_sock.
1448 static void svc_check_conn_limits(struct svc_serv
*serv
)
1450 if (serv
->sv_tmpcnt
> (serv
->sv_nrthreads
+3)*20) {
1451 struct svc_xprt
*xprt
= NULL
;
1452 spin_lock_bh(&serv
->sv_lock
);
1453 if (!list_empty(&serv
->sv_tempsocks
)) {
1454 if (net_ratelimit()) {
1455 /* Try to help the admin */
1456 printk(KERN_NOTICE
"%s: too many open "
1457 "connections, consider increasing the "
1458 "number of nfsd threads\n",
1462 * Always select the oldest connection. It's not fair,
1465 xprt
= list_entry(serv
->sv_tempsocks
.prev
,
1468 set_bit(XPT_CLOSE
, &xprt
->xpt_flags
);
1471 spin_unlock_bh(&serv
->sv_lock
);
1474 svc_xprt_enqueue(xprt
);
1481 * Receive the next request on any socket. This code is carefully
1482 * organised not to touch any cachelines in the shared svc_serv
1483 * structure, only cachelines in the local svc_pool.
1486 svc_recv(struct svc_rqst
*rqstp
, long timeout
)
1488 struct svc_xprt
*xprt
= NULL
;
1489 struct svc_serv
*serv
= rqstp
->rq_server
;
1490 struct svc_pool
*pool
= rqstp
->rq_pool
;
1493 struct xdr_buf
*arg
;
1494 DECLARE_WAITQUEUE(wait
, current
);
1496 dprintk("svc: server %p waiting for data (to = %ld)\n",
1501 "svc_recv: service %p, transport not NULL!\n",
1503 if (waitqueue_active(&rqstp
->rq_wait
))
1505 "svc_recv: service %p, wait queue active!\n",
1509 /* now allocate needed pages. If we get a failure, sleep briefly */
1510 pages
= (serv
->sv_max_mesg
+ PAGE_SIZE
) / PAGE_SIZE
;
1511 for (i
=0; i
< pages
; i
++)
1512 while (rqstp
->rq_pages
[i
] == NULL
) {
1513 struct page
*p
= alloc_page(GFP_KERNEL
);
1515 schedule_timeout_uninterruptible(msecs_to_jiffies(500));
1516 rqstp
->rq_pages
[i
] = p
;
1518 rqstp
->rq_pages
[i
++] = NULL
; /* this might be seen in nfs_read_actor */
1519 BUG_ON(pages
>= RPCSVC_MAXPAGES
);
1521 /* Make arg->head point to first page and arg->pages point to rest */
1522 arg
= &rqstp
->rq_arg
;
1523 arg
->head
[0].iov_base
= page_address(rqstp
->rq_pages
[0]);
1524 arg
->head
[0].iov_len
= PAGE_SIZE
;
1525 arg
->pages
= rqstp
->rq_pages
+ 1;
1527 /* save at least one page for response */
1528 arg
->page_len
= (pages
-2)*PAGE_SIZE
;
1529 arg
->len
= (pages
-1)*PAGE_SIZE
;
1530 arg
->tail
[0].iov_len
= 0;
1537 spin_lock_bh(&pool
->sp_lock
);
1538 xprt
= svc_xprt_dequeue(pool
);
1540 rqstp
->rq_xprt
= xprt
;
1542 rqstp
->rq_reserved
= serv
->sv_max_mesg
;
1543 atomic_add(rqstp
->rq_reserved
, &xprt
->xpt_reserved
);
1545 /* No data pending. Go to sleep */
1546 svc_thread_enqueue(pool
, rqstp
);
1549 * We have to be able to interrupt this wait
1550 * to bring down the daemons ...
1552 set_current_state(TASK_INTERRUPTIBLE
);
1553 add_wait_queue(&rqstp
->rq_wait
, &wait
);
1554 spin_unlock_bh(&pool
->sp_lock
);
1556 schedule_timeout(timeout
);
1560 spin_lock_bh(&pool
->sp_lock
);
1561 remove_wait_queue(&rqstp
->rq_wait
, &wait
);
1563 xprt
= rqstp
->rq_xprt
;
1565 svc_thread_dequeue(pool
, rqstp
);
1566 spin_unlock_bh(&pool
->sp_lock
);
1567 dprintk("svc: server %p, no data yet\n", rqstp
);
1568 return signalled()? -EINTR
: -EAGAIN
;
1571 spin_unlock_bh(&pool
->sp_lock
);
1574 if (test_bit(XPT_CLOSE
, &xprt
->xpt_flags
)) {
1575 dprintk("svc_recv: found XPT_CLOSE\n");
1576 svc_delete_xprt(xprt
);
1577 } else if (test_bit(XPT_LISTENER
, &xprt
->xpt_flags
)) {
1578 struct svc_xprt
*newxpt
;
1579 newxpt
= xprt
->xpt_ops
->xpo_accept(xprt
);
1582 * We know this module_get will succeed because the
1583 * listener holds a reference too
1585 __module_get(newxpt
->xpt_class
->xcl_owner
);
1586 svc_check_conn_limits(xprt
->xpt_server
);
1587 spin_lock_bh(&serv
->sv_lock
);
1588 set_bit(XPT_TEMP
, &newxpt
->xpt_flags
);
1589 list_add(&newxpt
->xpt_list
, &serv
->sv_tempsocks
);
1591 if (serv
->sv_temptimer
.function
== NULL
) {
1592 /* setup timer to age temp sockets */
1593 setup_timer(&serv
->sv_temptimer
,
1595 (unsigned long)serv
);
1596 mod_timer(&serv
->sv_temptimer
,
1597 jiffies
+ svc_conn_age_period
* HZ
);
1599 spin_unlock_bh(&serv
->sv_lock
);
1600 svc_xprt_received(newxpt
);
1602 svc_xprt_received(xprt
);
1604 dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n",
1605 rqstp
, pool
->sp_id
, xprt
,
1606 atomic_read(&xprt
->xpt_ref
.refcount
));
1607 rqstp
->rq_deferred
= svc_deferred_dequeue(xprt
);
1608 if (rqstp
->rq_deferred
) {
1609 svc_xprt_received(xprt
);
1610 len
= svc_deferred_recv(rqstp
);
1612 len
= xprt
->xpt_ops
->xpo_recvfrom(rqstp
);
1613 dprintk("svc: got len=%d\n", len
);
1616 /* No data, incomplete (TCP) read, or accept() */
1617 if (len
== 0 || len
== -EAGAIN
) {
1618 rqstp
->rq_res
.len
= 0;
1619 svc_xprt_release(rqstp
);
1622 clear_bit(XPT_OLD
, &xprt
->xpt_flags
);
1624 rqstp
->rq_secure
= svc_port_is_privileged(svc_addr(rqstp
));
1625 rqstp
->rq_chandle
.defer
= svc_defer
;
1628 serv
->sv_stats
->netcnt
++;
1636 svc_drop(struct svc_rqst
*rqstp
)
1638 dprintk("svc: xprt %p dropped request\n", rqstp
->rq_xprt
);
1639 svc_xprt_release(rqstp
);
1643 * Return reply to client.
1646 svc_send(struct svc_rqst
*rqstp
)
1648 struct svc_xprt
*xprt
;
1652 xprt
= rqstp
->rq_xprt
;
1656 /* release the receive skb before sending the reply */
1657 rqstp
->rq_xprt
->xpt_ops
->xpo_release_rqst(rqstp
);
1659 /* calculate over-all length */
1660 xb
= & rqstp
->rq_res
;
1661 xb
->len
= xb
->head
[0].iov_len
+
1663 xb
->tail
[0].iov_len
;
1665 /* Grab mutex to serialize outgoing data. */
1666 mutex_lock(&xprt
->xpt_mutex
);
1667 if (test_bit(XPT_DEAD
, &xprt
->xpt_flags
))
1670 len
= xprt
->xpt_ops
->xpo_sendto(rqstp
);
1671 mutex_unlock(&xprt
->xpt_mutex
);
1672 svc_xprt_release(rqstp
);
1674 if (len
== -ECONNREFUSED
|| len
== -ENOTCONN
|| len
== -EAGAIN
)
1680 * Timer function to close old temporary sockets, using
1681 * a mark-and-sweep algorithm.
1683 static void svc_age_temp_xprts(unsigned long closure
)
1685 struct svc_serv
*serv
= (struct svc_serv
*)closure
;
1686 struct svc_xprt
*xprt
;
1687 struct list_head
*le
, *next
;
1688 LIST_HEAD(to_be_aged
);
1690 dprintk("svc_age_temp_xprts\n");
1692 if (!spin_trylock_bh(&serv
->sv_lock
)) {
1693 /* busy, try again 1 sec later */
1694 dprintk("svc_age_temp_xprts: busy\n");
1695 mod_timer(&serv
->sv_temptimer
, jiffies
+ HZ
);
1699 list_for_each_safe(le
, next
, &serv
->sv_tempsocks
) {
1700 xprt
= list_entry(le
, struct svc_xprt
, xpt_list
);
1702 /* First time through, just mark it OLD. Second time
1703 * through, close it. */
1704 if (!test_and_set_bit(XPT_OLD
, &xprt
->xpt_flags
))
1706 if (atomic_read(&xprt
->xpt_ref
.refcount
) > 1
1707 || test_bit(XPT_BUSY
, &xprt
->xpt_flags
))
1710 list_move(le
, &to_be_aged
);
1711 set_bit(XPT_CLOSE
, &xprt
->xpt_flags
);
1712 set_bit(XPT_DETACHED
, &xprt
->xpt_flags
);
1714 spin_unlock_bh(&serv
->sv_lock
);
1716 while (!list_empty(&to_be_aged
)) {
1717 le
= to_be_aged
.next
;
1718 /* fiddling the xpt_list node is safe 'cos we're XPT_DETACHED */
1720 xprt
= list_entry(le
, struct svc_xprt
, xpt_list
);
1722 dprintk("queuing xprt %p for closing\n", xprt
);
1724 /* a thread will dequeue and close it soon */
1725 svc_xprt_enqueue(xprt
);
1729 mod_timer(&serv
->sv_temptimer
, jiffies
+ svc_conn_age_period
* HZ
);
1733 * Initialize socket for RPC use and create svc_sock struct
1734 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1736 static struct svc_sock
*svc_setup_socket(struct svc_serv
*serv
,
1737 struct socket
*sock
,
1738 int *errp
, int flags
)
1740 struct svc_sock
*svsk
;
1742 int pmap_register
= !(flags
& SVC_SOCK_ANONYMOUS
);
1744 dprintk("svc: svc_setup_socket %p\n", sock
);
1745 if (!(svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
))) {
1752 /* Register socket with portmapper */
1753 if (*errp
>= 0 && pmap_register
)
1754 *errp
= svc_register(serv
, inet
->sk_protocol
,
1755 ntohs(inet_sk(inet
)->sport
));
1762 inet
->sk_user_data
= svsk
;
1763 svsk
->sk_sock
= sock
;
1765 svsk
->sk_ostate
= inet
->sk_state_change
;
1766 svsk
->sk_odata
= inet
->sk_data_ready
;
1767 svsk
->sk_owspace
= inet
->sk_write_space
;
1769 /* Initialize the socket */
1770 if (sock
->type
== SOCK_DGRAM
)
1771 svc_udp_init(svsk
, serv
);
1773 svc_tcp_init(svsk
, serv
);
1775 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1781 int svc_addsock(struct svc_serv
*serv
,
1787 struct socket
*so
= sockfd_lookup(fd
, &err
);
1788 struct svc_sock
*svsk
= NULL
;
1792 if (so
->sk
->sk_family
!= AF_INET
)
1793 err
= -EAFNOSUPPORT
;
1794 else if (so
->sk
->sk_protocol
!= IPPROTO_TCP
&&
1795 so
->sk
->sk_protocol
!= IPPROTO_UDP
)
1796 err
= -EPROTONOSUPPORT
;
1797 else if (so
->state
> SS_UNCONNECTED
)
1800 svsk
= svc_setup_socket(serv
, so
, &err
, SVC_SOCK_DEFAULTS
);
1802 struct sockaddr_storage addr
;
1803 struct sockaddr
*sin
= (struct sockaddr
*)&addr
;
1805 if (kernel_getsockname(svsk
->sk_sock
, sin
, &salen
) == 0)
1806 svc_xprt_set_local(&svsk
->sk_xprt
, sin
, salen
);
1807 clear_bit(XPT_TEMP
, &svsk
->sk_xprt
.xpt_flags
);
1808 spin_lock_bh(&serv
->sv_lock
);
1809 list_add(&svsk
->sk_xprt
.xpt_list
, &serv
->sv_permsocks
);
1810 spin_unlock_bh(&serv
->sv_lock
);
1811 svc_xprt_received(&svsk
->sk_xprt
);
1819 if (proto
) *proto
= so
->sk
->sk_protocol
;
1820 return one_sock_name(name_return
, svsk
);
1822 EXPORT_SYMBOL_GPL(svc_addsock
);
1825 * Create socket for RPC service.
1827 static struct svc_xprt
*svc_create_socket(struct svc_serv
*serv
,
1829 struct sockaddr
*sin
, int len
,
1832 struct svc_sock
*svsk
;
1833 struct socket
*sock
;
1836 char buf
[RPC_MAX_ADDRBUFLEN
];
1837 struct sockaddr_storage addr
;
1838 struct sockaddr
*newsin
= (struct sockaddr
*)&addr
;
1841 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1842 serv
->sv_program
->pg_name
, protocol
,
1843 __svc_print_addr(sin
, buf
, sizeof(buf
)));
1845 if (protocol
!= IPPROTO_UDP
&& protocol
!= IPPROTO_TCP
) {
1846 printk(KERN_WARNING
"svc: only UDP and TCP "
1847 "sockets supported\n");
1848 return ERR_PTR(-EINVAL
);
1850 type
= (protocol
== IPPROTO_UDP
)? SOCK_DGRAM
: SOCK_STREAM
;
1852 error
= sock_create_kern(sin
->sa_family
, type
, protocol
, &sock
);
1854 return ERR_PTR(error
);
1856 svc_reclassify_socket(sock
);
1858 if (type
== SOCK_STREAM
)
1859 sock
->sk
->sk_reuse
= 1; /* allow address reuse */
1860 error
= kernel_bind(sock
, sin
, len
);
1865 error
= kernel_getsockname(sock
, newsin
, &newlen
);
1869 if (protocol
== IPPROTO_TCP
) {
1870 if ((error
= kernel_listen(sock
, 64)) < 0)
1874 if ((svsk
= svc_setup_socket(serv
, sock
, &error
, flags
)) != NULL
) {
1875 svc_xprt_set_local(&svsk
->sk_xprt
, newsin
, newlen
);
1876 return (struct svc_xprt
*)svsk
;
1880 dprintk("svc: svc_create_socket error = %d\n", -error
);
1882 return ERR_PTR(error
);
1886 * Detach the svc_sock from the socket so that no
1887 * more callbacks occur.
1889 static void svc_sock_detach(struct svc_xprt
*xprt
)
1891 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1892 struct sock
*sk
= svsk
->sk_sk
;
1894 dprintk("svc: svc_sock_detach(%p)\n", svsk
);
1896 /* put back the old socket callbacks */
1897 sk
->sk_state_change
= svsk
->sk_ostate
;
1898 sk
->sk_data_ready
= svsk
->sk_odata
;
1899 sk
->sk_write_space
= svsk
->sk_owspace
;
1903 * Free the svc_sock's socket resources and the svc_sock itself.
1905 static void svc_sock_free(struct svc_xprt
*xprt
)
1907 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1908 dprintk("svc: svc_sock_free(%p)\n", svsk
);
1910 if (svsk
->sk_sock
->file
)
1911 sockfd_put(svsk
->sk_sock
);
1913 sock_release(svsk
->sk_sock
);
1918 * Remove a dead transport
1920 static void svc_delete_xprt(struct svc_xprt
*xprt
)
1922 struct svc_serv
*serv
= xprt
->xpt_server
;
1924 dprintk("svc: svc_delete_xprt(%p)\n", xprt
);
1925 xprt
->xpt_ops
->xpo_detach(xprt
);
1927 spin_lock_bh(&serv
->sv_lock
);
1928 if (!test_and_set_bit(XPT_DETACHED
, &xprt
->xpt_flags
))
1929 list_del_init(&xprt
->xpt_list
);
1931 * We used to delete the transport from whichever list
1932 * it's sk_xprt.xpt_ready node was on, but we don't actually
1933 * need to. This is because the only time we're called
1934 * while still attached to a queue, the queue itself
1935 * is about to be destroyed (in svc_destroy).
1937 if (!test_and_set_bit(XPT_DEAD
, &xprt
->xpt_flags
)) {
1938 BUG_ON(atomic_read(&xprt
->xpt_ref
.refcount
) < 2);
1939 if (test_bit(XPT_TEMP
, &xprt
->xpt_flags
))
1943 spin_unlock_bh(&serv
->sv_lock
);
1946 static void svc_close_xprt(struct svc_xprt
*xprt
)
1948 set_bit(XPT_CLOSE
, &xprt
->xpt_flags
);
1949 if (test_and_set_bit(XPT_BUSY
, &xprt
->xpt_flags
))
1950 /* someone else will have to effect the close */
1954 svc_delete_xprt(xprt
);
1955 clear_bit(XPT_BUSY
, &xprt
->xpt_flags
);
1959 void svc_close_all(struct list_head
*xprt_list
)
1961 struct svc_xprt
*xprt
;
1962 struct svc_xprt
*tmp
;
1964 list_for_each_entry_safe(xprt
, tmp
, xprt_list
, xpt_list
) {
1965 set_bit(XPT_CLOSE
, &xprt
->xpt_flags
);
1966 if (test_bit(XPT_BUSY
, &xprt
->xpt_flags
)) {
1967 /* Waiting to be processed, but no threads left,
1968 * So just remove it from the waiting list
1970 list_del_init(&xprt
->xpt_ready
);
1971 clear_bit(XPT_BUSY
, &xprt
->xpt_flags
);
1973 svc_close_xprt(xprt
);
1978 * Handle defer and revisit of requests
1981 static void svc_revisit(struct cache_deferred_req
*dreq
, int too_many
)
1983 struct svc_deferred_req
*dr
= container_of(dreq
, struct svc_deferred_req
, handle
);
1984 struct svc_xprt
*xprt
= dr
->xprt
;
1991 dprintk("revisit queued\n");
1993 spin_lock(&xprt
->xpt_lock
);
1994 list_add(&dr
->handle
.recent
, &xprt
->xpt_deferred
);
1995 spin_unlock(&xprt
->xpt_lock
);
1996 set_bit(XPT_DEFERRED
, &xprt
->xpt_flags
);
1997 svc_xprt_enqueue(xprt
);
2001 static struct cache_deferred_req
*
2002 svc_defer(struct cache_req
*req
)
2004 struct svc_rqst
*rqstp
= container_of(req
, struct svc_rqst
, rq_chandle
);
2005 int size
= sizeof(struct svc_deferred_req
) + (rqstp
->rq_arg
.len
);
2006 struct svc_deferred_req
*dr
;
2008 if (rqstp
->rq_arg
.page_len
)
2009 return NULL
; /* if more than a page, give up FIXME */
2010 if (rqstp
->rq_deferred
) {
2011 dr
= rqstp
->rq_deferred
;
2012 rqstp
->rq_deferred
= NULL
;
2014 int skip
= rqstp
->rq_arg
.len
- rqstp
->rq_arg
.head
[0].iov_len
;
2015 /* FIXME maybe discard if size too large */
2016 dr
= kmalloc(size
, GFP_KERNEL
);
2020 dr
->handle
.owner
= rqstp
->rq_server
;
2021 dr
->prot
= rqstp
->rq_prot
;
2022 memcpy(&dr
->addr
, &rqstp
->rq_addr
, rqstp
->rq_addrlen
);
2023 dr
->addrlen
= rqstp
->rq_addrlen
;
2024 dr
->daddr
= rqstp
->rq_daddr
;
2025 dr
->argslen
= rqstp
->rq_arg
.len
>> 2;
2026 memcpy(dr
->args
, rqstp
->rq_arg
.head
[0].iov_base
-skip
, dr
->argslen
<<2);
2028 svc_xprt_get(rqstp
->rq_xprt
);
2029 dr
->xprt
= rqstp
->rq_xprt
;
2031 dr
->handle
.revisit
= svc_revisit
;
2036 * recv data from a deferred request into an active one
2038 static int svc_deferred_recv(struct svc_rqst
*rqstp
)
2040 struct svc_deferred_req
*dr
= rqstp
->rq_deferred
;
2042 rqstp
->rq_arg
.head
[0].iov_base
= dr
->args
;
2043 rqstp
->rq_arg
.head
[0].iov_len
= dr
->argslen
<<2;
2044 rqstp
->rq_arg
.page_len
= 0;
2045 rqstp
->rq_arg
.len
= dr
->argslen
<<2;
2046 rqstp
->rq_prot
= dr
->prot
;
2047 memcpy(&rqstp
->rq_addr
, &dr
->addr
, dr
->addrlen
);
2048 rqstp
->rq_addrlen
= dr
->addrlen
;
2049 rqstp
->rq_daddr
= dr
->daddr
;
2050 rqstp
->rq_respages
= rqstp
->rq_pages
;
2051 return dr
->argslen
<<2;
2055 static struct svc_deferred_req
*svc_deferred_dequeue(struct svc_xprt
*xprt
)
2057 struct svc_deferred_req
*dr
= NULL
;
2059 if (!test_bit(XPT_DEFERRED
, &xprt
->xpt_flags
))
2061 spin_lock(&xprt
->xpt_lock
);
2062 clear_bit(XPT_DEFERRED
, &xprt
->xpt_flags
);
2063 if (!list_empty(&xprt
->xpt_deferred
)) {
2064 dr
= list_entry(xprt
->xpt_deferred
.next
,
2065 struct svc_deferred_req
,
2067 list_del_init(&dr
->handle
.recent
);
2068 set_bit(XPT_DEFERRED
, &xprt
->xpt_flags
);
2070 spin_unlock(&xprt
->xpt_lock
);