2 * linux/net/sunrpc/svc_xprt.c
4 * Author: Tom Tucker <tom@opengridcomputing.com>
7 #include <linux/sched.h>
8 #include <linux/smp_lock.h>
9 #include <linux/errno.h>
10 #include <linux/freezer.h>
11 #include <linux/kthread.h>
13 #include <linux/sunrpc/stats.h>
14 #include <linux/sunrpc/svc_xprt.h>
15 #include <linux/sunrpc/svcsock.h>
17 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
19 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt);
20 static int svc_deferred_recv(struct svc_rqst *rqstp);
21 static struct cache_deferred_req *svc_defer(struct cache_req *req);
22 static void svc_age_temp_xprts(unsigned long closure);
24 /* apparently the "standard" is that clients close
25 * idle connections after 5 minutes, servers after
27 * http://www.connectathon.org/talks96/nfstcp.pdf
29 static int svc_conn_age_period = 6*60;
31 /* List of registered transport classes */
32 static DEFINE_SPINLOCK(svc_xprt_class_lock);
33 static LIST_HEAD(svc_xprt_class_list);
35 /* SMP locking strategy:
37 * svc_pool->sp_lock protects most of the fields of that pool.
38 * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
39 * when both need to be taken (rare), svc_serv->sv_lock is first.
40 * BKL protects svc_serv->sv_nrthread.
41 * svc_sock->sk_lock protects the svc_sock->sk_deferred list
42 * and the ->sk_info_authunix cache.
44 * The XPT_BUSY bit in xprt->xpt_flags prevents a transport being
45 * enqueued multiply. During normal transport processing this bit
46 * is set by svc_xprt_enqueue and cleared by svc_xprt_received.
47 * Providers should not manipulate this bit directly.
49 * Some flags can be set to certain values at any time
50 * providing that certain rules are followed:
53 * - Can be set or cleared at any time.
54 * - After a set, svc_xprt_enqueue must be called to enqueue
55 * the transport for processing.
56 * - After a clear, the transport must be read/accepted.
57 * If this succeeds, it must be set again.
59 * - Can set at any time. It is never cleared.
61 * - Can only be set while XPT_BUSY is held which ensures
62 * that no other thread will be using the transport or will
63 * try to set XPT_DEAD.
66 int svc_reg_xprt_class(struct svc_xprt_class *xcl)
68 struct svc_xprt_class *cl;
71 dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name);
73 INIT_LIST_HEAD(&xcl->xcl_list);
74 spin_lock(&svc_xprt_class_lock);
75 /* Make sure there isn't already a class with the same name */
76 list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) {
77 if (strcmp(xcl->xcl_name, cl->xcl_name) == 0)
80 list_add_tail(&xcl->xcl_list, &svc_xprt_class_list);
83 spin_unlock(&svc_xprt_class_lock);
86 EXPORT_SYMBOL_GPL(svc_reg_xprt_class);
88 void svc_unreg_xprt_class(struct svc_xprt_class *xcl)
90 dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name);
91 spin_lock(&svc_xprt_class_lock);
92 list_del_init(&xcl->xcl_list);
93 spin_unlock(&svc_xprt_class_lock);
95 EXPORT_SYMBOL_GPL(svc_unreg_xprt_class);
98 * Format the transport list for printing
100 int svc_print_xprts(char *buf, int maxlen)
102 struct list_head *le;
107 spin_lock(&svc_xprt_class_lock);
108 list_for_each(le, &svc_xprt_class_list) {
110 struct svc_xprt_class *xcl =
111 list_entry(le, struct svc_xprt_class, xcl_list);
113 sprintf(tmpstr, "%s %d\n", xcl->xcl_name, xcl->xcl_max_payload);
114 slen = strlen(tmpstr);
115 if (len + slen > maxlen)
120 spin_unlock(&svc_xprt_class_lock);
125 static void svc_xprt_free(struct kref *kref)
127 struct svc_xprt *xprt =
128 container_of(kref, struct svc_xprt, xpt_ref);
129 struct module *owner = xprt->xpt_class->xcl_owner;
130 if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags) &&
131 xprt->xpt_auth_cache != NULL)
132 svcauth_unix_info_release(xprt->xpt_auth_cache);
133 xprt->xpt_ops->xpo_free(xprt);
137 void svc_xprt_put(struct svc_xprt *xprt)
139 kref_put(&xprt->xpt_ref, svc_xprt_free);
141 EXPORT_SYMBOL_GPL(svc_xprt_put);
144 * Called by transport drivers to initialize the transport independent
145 * portion of the transport instance.
147 void svc_xprt_init(struct svc_xprt_class *xcl, struct svc_xprt *xprt,
148 struct svc_serv *serv)
150 memset(xprt, 0, sizeof(*xprt));
151 xprt->xpt_class = xcl;
152 xprt->xpt_ops = xcl->xcl_ops;
153 kref_init(&xprt->xpt_ref);
154 xprt->xpt_server = serv;
155 INIT_LIST_HEAD(&xprt->xpt_list);
156 INIT_LIST_HEAD(&xprt->xpt_ready);
157 INIT_LIST_HEAD(&xprt->xpt_deferred);
158 mutex_init(&xprt->xpt_mutex);
159 spin_lock_init(&xprt->xpt_lock);
160 set_bit(XPT_BUSY, &xprt->xpt_flags);
161 rpc_init_wait_queue(&xprt->xpt_bc_pending, "xpt_bc_pending");
163 EXPORT_SYMBOL_GPL(svc_xprt_init);
165 static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl,
166 struct svc_serv *serv,
168 const unsigned short port,
171 struct sockaddr_in sin = {
172 .sin_family = AF_INET,
173 .sin_addr.s_addr = htonl(INADDR_ANY),
174 .sin_port = htons(port),
176 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
177 struct sockaddr_in6 sin6 = {
178 .sin6_family = AF_INET6,
179 .sin6_addr = IN6ADDR_ANY_INIT,
180 .sin6_port = htons(port),
182 #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
183 struct sockaddr *sap;
188 sap = (struct sockaddr *)&sin;
191 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
193 sap = (struct sockaddr *)&sin6;
196 #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
198 return ERR_PTR(-EAFNOSUPPORT);
201 return xcl->xcl_ops->xpo_create(serv, sap, len, flags);
204 int svc_create_xprt(struct svc_serv *serv, const char *xprt_name,
205 const int family, const unsigned short port,
208 struct svc_xprt_class *xcl;
210 dprintk("svc: creating transport %s[%d]\n", xprt_name, port);
211 spin_lock(&svc_xprt_class_lock);
212 list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
213 struct svc_xprt *newxprt;
215 if (strcmp(xprt_name, xcl->xcl_name))
218 if (!try_module_get(xcl->xcl_owner))
221 spin_unlock(&svc_xprt_class_lock);
222 newxprt = __svc_xpo_create(xcl, serv, family, port, flags);
223 if (IS_ERR(newxprt)) {
224 module_put(xcl->xcl_owner);
225 return PTR_ERR(newxprt);
228 clear_bit(XPT_TEMP, &newxprt->xpt_flags);
229 spin_lock_bh(&serv->sv_lock);
230 list_add(&newxprt->xpt_list, &serv->sv_permsocks);
231 spin_unlock_bh(&serv->sv_lock);
232 clear_bit(XPT_BUSY, &newxprt->xpt_flags);
233 return svc_xprt_local_port(newxprt);
236 spin_unlock(&svc_xprt_class_lock);
237 dprintk("svc: transport %s not found\n", xprt_name);
239 /* This errno is exposed to user space. Provide a reasonable
240 * perror msg for a bad transport. */
241 return -EPROTONOSUPPORT;
243 EXPORT_SYMBOL_GPL(svc_create_xprt);
246 * Copy the local and remote xprt addresses to the rqstp structure
248 void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt)
250 struct sockaddr *sin;
252 memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen);
253 rqstp->rq_addrlen = xprt->xpt_remotelen;
256 * Destination address in request is needed for binding the
257 * source address in RPC replies/callbacks later.
259 sin = (struct sockaddr *)&xprt->xpt_local;
260 switch (sin->sa_family) {
262 rqstp->rq_daddr.addr = ((struct sockaddr_in *)sin)->sin_addr;
265 rqstp->rq_daddr.addr6 = ((struct sockaddr_in6 *)sin)->sin6_addr;
269 EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs);
272 * svc_print_addr - Format rq_addr field for printing
273 * @rqstp: svc_rqst struct containing address to print
274 * @buf: target buffer for formatted address
275 * @len: length of target buffer
278 char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
280 return __svc_print_addr(svc_addr(rqstp), buf, len);
282 EXPORT_SYMBOL_GPL(svc_print_addr);
285 * Queue up an idle server thread. Must have pool->sp_lock held.
286 * Note: this is really a stack rather than a queue, so that we only
287 * use as many different threads as we need, and the rest don't pollute
290 static void svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp)
292 list_add(&rqstp->rq_list, &pool->sp_threads);
296 * Dequeue an nfsd thread. Must have pool->sp_lock held.
298 static void svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp)
300 list_del(&rqstp->rq_list);
304 * Queue up a transport with data pending. If there are idle nfsd
305 * processes, wake 'em up.
308 void svc_xprt_enqueue(struct svc_xprt *xprt)
310 struct svc_serv *serv = xprt->xpt_server;
311 struct svc_pool *pool;
312 struct svc_rqst *rqstp;
315 if (!(xprt->xpt_flags &
316 ((1<<XPT_CONN)|(1<<XPT_DATA)|(1<<XPT_CLOSE)|(1<<XPT_DEFERRED))))
320 pool = svc_pool_for_cpu(xprt->xpt_server, cpu);
323 spin_lock_bh(&pool->sp_lock);
325 if (!list_empty(&pool->sp_threads) &&
326 !list_empty(&pool->sp_sockets))
329 "threads and transports both waiting??\n");
331 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) {
332 /* Don't enqueue dead transports */
333 dprintk("svc: transport %p is dead, not enqueued\n", xprt);
337 pool->sp_stats.packets++;
339 /* Mark transport as busy. It will remain in this state until
340 * the provider calls svc_xprt_received. We update XPT_BUSY
341 * atomically because it also guards against trying to enqueue
342 * the transport twice.
344 if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) {
345 /* Don't enqueue transport while already enqueued */
346 dprintk("svc: transport %p busy, not enqueued\n", xprt);
349 BUG_ON(xprt->xpt_pool != NULL);
350 xprt->xpt_pool = pool;
352 /* Handle pending connection */
353 if (test_bit(XPT_CONN, &xprt->xpt_flags))
356 /* Handle close in-progress */
357 if (test_bit(XPT_CLOSE, &xprt->xpt_flags))
360 /* Check if we have space to reply to a request */
361 if (!xprt->xpt_ops->xpo_has_wspace(xprt)) {
362 /* Don't enqueue while not enough space for reply */
363 dprintk("svc: no write space, transport %p not enqueued\n",
365 xprt->xpt_pool = NULL;
366 clear_bit(XPT_BUSY, &xprt->xpt_flags);
371 if (!list_empty(&pool->sp_threads)) {
372 rqstp = list_entry(pool->sp_threads.next,
375 dprintk("svc: transport %p served by daemon %p\n",
377 svc_thread_dequeue(pool, rqstp);
380 "svc_xprt_enqueue: server %p, rq_xprt=%p!\n",
381 rqstp, rqstp->rq_xprt);
382 rqstp->rq_xprt = xprt;
384 rqstp->rq_reserved = serv->sv_max_mesg;
385 atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
386 pool->sp_stats.threads_woken++;
387 BUG_ON(xprt->xpt_pool != pool);
388 wake_up(&rqstp->rq_wait);
390 dprintk("svc: transport %p put into queue\n", xprt);
391 list_add_tail(&xprt->xpt_ready, &pool->sp_sockets);
392 pool->sp_stats.sockets_queued++;
393 BUG_ON(xprt->xpt_pool != pool);
397 spin_unlock_bh(&pool->sp_lock);
399 EXPORT_SYMBOL_GPL(svc_xprt_enqueue);
402 * Dequeue the first transport. Must be called with the pool->sp_lock held.
404 static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool)
406 struct svc_xprt *xprt;
408 if (list_empty(&pool->sp_sockets))
411 xprt = list_entry(pool->sp_sockets.next,
412 struct svc_xprt, xpt_ready);
413 list_del_init(&xprt->xpt_ready);
415 dprintk("svc: transport %p dequeued, inuse=%d\n",
416 xprt, atomic_read(&xprt->xpt_ref.refcount));
422 * svc_xprt_received conditionally queues the transport for processing
423 * by another thread. The caller must hold the XPT_BUSY bit and must
424 * not thereafter touch transport data.
426 * Note: XPT_DATA only gets cleared when a read-attempt finds no (or
427 * insufficient) data.
429 void svc_xprt_received(struct svc_xprt *xprt)
431 BUG_ON(!test_bit(XPT_BUSY, &xprt->xpt_flags));
432 xprt->xpt_pool = NULL;
433 clear_bit(XPT_BUSY, &xprt->xpt_flags);
434 svc_xprt_enqueue(xprt);
436 EXPORT_SYMBOL_GPL(svc_xprt_received);
439 * svc_reserve - change the space reserved for the reply to a request.
440 * @rqstp: The request in question
441 * @space: new max space to reserve
443 * Each request reserves some space on the output queue of the transport
444 * to make sure the reply fits. This function reduces that reserved
445 * space to be the amount of space used already, plus @space.
448 void svc_reserve(struct svc_rqst *rqstp, int space)
450 space += rqstp->rq_res.head[0].iov_len;
452 if (space < rqstp->rq_reserved) {
453 struct svc_xprt *xprt = rqstp->rq_xprt;
454 atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved);
455 rqstp->rq_reserved = space;
457 svc_xprt_enqueue(xprt);
460 EXPORT_SYMBOL_GPL(svc_reserve);
462 static void svc_xprt_release(struct svc_rqst *rqstp)
464 struct svc_xprt *xprt = rqstp->rq_xprt;
466 rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
468 kfree(rqstp->rq_deferred);
469 rqstp->rq_deferred = NULL;
471 svc_free_res_pages(rqstp);
472 rqstp->rq_res.page_len = 0;
473 rqstp->rq_res.page_base = 0;
475 /* Reset response buffer and release
477 * But first, check that enough space was reserved
478 * for the reply, otherwise we have a bug!
480 if ((rqstp->rq_res.len) > rqstp->rq_reserved)
481 printk(KERN_ERR "RPC request reserved %d but used %d\n",
485 rqstp->rq_res.head[0].iov_len = 0;
486 svc_reserve(rqstp, 0);
487 rqstp->rq_xprt = NULL;
493 * External function to wake up a server waiting for data
494 * This really only makes sense for services like lockd
495 * which have exactly one thread anyway.
497 void svc_wake_up(struct svc_serv *serv)
499 struct svc_rqst *rqstp;
501 struct svc_pool *pool;
503 for (i = 0; i < serv->sv_nrpools; i++) {
504 pool = &serv->sv_pools[i];
506 spin_lock_bh(&pool->sp_lock);
507 if (!list_empty(&pool->sp_threads)) {
508 rqstp = list_entry(pool->sp_threads.next,
511 dprintk("svc: daemon %p woken up.\n", rqstp);
513 svc_thread_dequeue(pool, rqstp);
514 rqstp->rq_xprt = NULL;
516 wake_up(&rqstp->rq_wait);
518 spin_unlock_bh(&pool->sp_lock);
521 EXPORT_SYMBOL_GPL(svc_wake_up);
523 int svc_port_is_privileged(struct sockaddr *sin)
525 switch (sin->sa_family) {
527 return ntohs(((struct sockaddr_in *)sin)->sin_port)
530 return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
538 * Make sure that we don't have too many active connections. If we have,
539 * something must be dropped. It's not clear what will happen if we allow
540 * "too many" connections, but when dealing with network-facing software,
541 * we have to code defensively. Here we do that by imposing hard limits.
543 * There's no point in trying to do random drop here for DoS
544 * prevention. The NFS clients does 1 reconnect in 15 seconds. An
545 * attacker can easily beat that.
547 * The only somewhat efficient mechanism would be if drop old
548 * connections from the same IP first. But right now we don't even
549 * record the client IP in svc_sock.
551 * single-threaded services that expect a lot of clients will probably
552 * need to set sv_maxconn to override the default value which is based
553 * on the number of threads
555 static void svc_check_conn_limits(struct svc_serv *serv)
557 unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn :
558 (serv->sv_nrthreads+3) * 20;
560 if (serv->sv_tmpcnt > limit) {
561 struct svc_xprt *xprt = NULL;
562 spin_lock_bh(&serv->sv_lock);
563 if (!list_empty(&serv->sv_tempsocks)) {
564 if (net_ratelimit()) {
565 /* Try to help the admin */
566 printk(KERN_NOTICE "%s: too many open "
567 "connections, consider increasing %s\n",
568 serv->sv_name, serv->sv_maxconn ?
569 "the max number of connections." :
570 "the number of threads.");
573 * Always select the oldest connection. It's not fair,
576 xprt = list_entry(serv->sv_tempsocks.prev,
579 set_bit(XPT_CLOSE, &xprt->xpt_flags);
582 spin_unlock_bh(&serv->sv_lock);
585 svc_xprt_enqueue(xprt);
592 * Receive the next request on any transport. This code is carefully
593 * organised not to touch any cachelines in the shared svc_serv
594 * structure, only cachelines in the local svc_pool.
596 int svc_recv(struct svc_rqst *rqstp, long timeout)
598 struct svc_xprt *xprt = NULL;
599 struct svc_serv *serv = rqstp->rq_server;
600 struct svc_pool *pool = rqstp->rq_pool;
604 DECLARE_WAITQUEUE(wait, current);
607 dprintk("svc: server %p waiting for data (to = %ld)\n",
612 "svc_recv: service %p, transport not NULL!\n",
614 if (waitqueue_active(&rqstp->rq_wait))
616 "svc_recv: service %p, wait queue active!\n",
619 /* now allocate needed pages. If we get a failure, sleep briefly */
620 pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE;
621 for (i = 0; i < pages ; i++)
622 while (rqstp->rq_pages[i] == NULL) {
623 struct page *p = alloc_page(GFP_KERNEL);
625 set_current_state(TASK_INTERRUPTIBLE);
626 if (signalled() || kthread_should_stop()) {
627 set_current_state(TASK_RUNNING);
630 schedule_timeout(msecs_to_jiffies(500));
632 rqstp->rq_pages[i] = p;
634 rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */
635 BUG_ON(pages >= RPCSVC_MAXPAGES);
637 /* Make arg->head point to first page and arg->pages point to rest */
638 arg = &rqstp->rq_arg;
639 arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
640 arg->head[0].iov_len = PAGE_SIZE;
641 arg->pages = rqstp->rq_pages + 1;
643 /* save at least one page for response */
644 arg->page_len = (pages-2)*PAGE_SIZE;
645 arg->len = (pages-1)*PAGE_SIZE;
646 arg->tail[0].iov_len = 0;
650 if (signalled() || kthread_should_stop())
653 spin_lock_bh(&pool->sp_lock);
654 xprt = svc_xprt_dequeue(pool);
656 rqstp->rq_xprt = xprt;
658 rqstp->rq_reserved = serv->sv_max_mesg;
659 atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
661 /* No data pending. Go to sleep */
662 svc_thread_enqueue(pool, rqstp);
665 * We have to be able to interrupt this wait
666 * to bring down the daemons ...
668 set_current_state(TASK_INTERRUPTIBLE);
671 * checking kthread_should_stop() here allows us to avoid
672 * locking and signalling when stopping kthreads that call
673 * svc_recv. If the thread has already been woken up, then
674 * we can exit here without sleeping. If not, then it
675 * it'll be woken up quickly during the schedule_timeout
677 if (kthread_should_stop()) {
678 set_current_state(TASK_RUNNING);
679 spin_unlock_bh(&pool->sp_lock);
683 add_wait_queue(&rqstp->rq_wait, &wait);
684 spin_unlock_bh(&pool->sp_lock);
686 time_left = schedule_timeout(timeout);
690 spin_lock_bh(&pool->sp_lock);
691 remove_wait_queue(&rqstp->rq_wait, &wait);
693 pool->sp_stats.threads_timedout++;
695 xprt = rqstp->rq_xprt;
697 svc_thread_dequeue(pool, rqstp);
698 spin_unlock_bh(&pool->sp_lock);
699 dprintk("svc: server %p, no data yet\n", rqstp);
700 if (signalled() || kthread_should_stop())
706 spin_unlock_bh(&pool->sp_lock);
709 if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
710 struct svc_xprt *newxpt;
711 newxpt = xprt->xpt_ops->xpo_accept(xprt);
714 * We know this module_get will succeed because the
715 * listener holds a reference too
717 __module_get(newxpt->xpt_class->xcl_owner);
718 svc_check_conn_limits(xprt->xpt_server);
719 spin_lock_bh(&serv->sv_lock);
720 set_bit(XPT_TEMP, &newxpt->xpt_flags);
721 list_add(&newxpt->xpt_list, &serv->sv_tempsocks);
723 if (serv->sv_temptimer.function == NULL) {
724 /* setup timer to age temp transports */
725 setup_timer(&serv->sv_temptimer,
727 (unsigned long)serv);
728 mod_timer(&serv->sv_temptimer,
729 jiffies + svc_conn_age_period * HZ);
731 spin_unlock_bh(&serv->sv_lock);
732 svc_xprt_received(newxpt);
734 svc_xprt_received(xprt);
735 } else if (!test_bit(XPT_CLOSE, &xprt->xpt_flags)) {
736 dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n",
737 rqstp, pool->sp_id, xprt,
738 atomic_read(&xprt->xpt_ref.refcount));
739 rqstp->rq_deferred = svc_deferred_dequeue(xprt);
740 if (rqstp->rq_deferred) {
741 svc_xprt_received(xprt);
742 len = svc_deferred_recv(rqstp);
744 len = xprt->xpt_ops->xpo_recvfrom(rqstp);
745 dprintk("svc: got len=%d\n", len);
748 if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) {
749 dprintk("svc_recv: found XPT_CLOSE\n");
750 svc_delete_xprt(xprt);
753 /* No data, incomplete (TCP) read, or accept() */
754 if (len == 0 || len == -EAGAIN) {
755 rqstp->rq_res.len = 0;
756 svc_xprt_release(rqstp);
759 clear_bit(XPT_OLD, &xprt->xpt_flags);
761 rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp));
762 rqstp->rq_chandle.defer = svc_defer;
765 serv->sv_stats->netcnt++;
768 EXPORT_SYMBOL_GPL(svc_recv);
773 void svc_drop(struct svc_rqst *rqstp)
775 dprintk("svc: xprt %p dropped request\n", rqstp->rq_xprt);
776 svc_xprt_release(rqstp);
778 EXPORT_SYMBOL_GPL(svc_drop);
781 * Return reply to client.
783 int svc_send(struct svc_rqst *rqstp)
785 struct svc_xprt *xprt;
789 xprt = rqstp->rq_xprt;
793 /* release the receive skb before sending the reply */
794 rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
796 /* calculate over-all length */
798 xb->len = xb->head[0].iov_len +
802 /* Grab mutex to serialize outgoing data. */
803 mutex_lock(&xprt->xpt_mutex);
804 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
807 len = xprt->xpt_ops->xpo_sendto(rqstp);
808 mutex_unlock(&xprt->xpt_mutex);
809 rpc_wake_up(&xprt->xpt_bc_pending);
810 svc_xprt_release(rqstp);
812 if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
818 * Timer function to close old temporary transports, using
819 * a mark-and-sweep algorithm.
821 static void svc_age_temp_xprts(unsigned long closure)
823 struct svc_serv *serv = (struct svc_serv *)closure;
824 struct svc_xprt *xprt;
825 struct list_head *le, *next;
826 LIST_HEAD(to_be_aged);
828 dprintk("svc_age_temp_xprts\n");
830 if (!spin_trylock_bh(&serv->sv_lock)) {
831 /* busy, try again 1 sec later */
832 dprintk("svc_age_temp_xprts: busy\n");
833 mod_timer(&serv->sv_temptimer, jiffies + HZ);
837 list_for_each_safe(le, next, &serv->sv_tempsocks) {
838 xprt = list_entry(le, struct svc_xprt, xpt_list);
840 /* First time through, just mark it OLD. Second time
841 * through, close it. */
842 if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags))
844 if (atomic_read(&xprt->xpt_ref.refcount) > 1 ||
845 test_bit(XPT_BUSY, &xprt->xpt_flags))
848 list_move(le, &to_be_aged);
849 set_bit(XPT_CLOSE, &xprt->xpt_flags);
850 set_bit(XPT_DETACHED, &xprt->xpt_flags);
852 spin_unlock_bh(&serv->sv_lock);
854 while (!list_empty(&to_be_aged)) {
855 le = to_be_aged.next;
856 /* fiddling the xpt_list node is safe 'cos we're XPT_DETACHED */
858 xprt = list_entry(le, struct svc_xprt, xpt_list);
860 dprintk("queuing xprt %p for closing\n", xprt);
862 /* a thread will dequeue and close it soon */
863 svc_xprt_enqueue(xprt);
867 mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
871 * Remove a dead transport
873 void svc_delete_xprt(struct svc_xprt *xprt)
875 struct svc_serv *serv = xprt->xpt_server;
876 struct svc_deferred_req *dr;
878 /* Only do this once */
879 if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags))
882 dprintk("svc: svc_delete_xprt(%p)\n", xprt);
883 xprt->xpt_ops->xpo_detach(xprt);
885 spin_lock_bh(&serv->sv_lock);
886 if (!test_and_set_bit(XPT_DETACHED, &xprt->xpt_flags))
887 list_del_init(&xprt->xpt_list);
889 * We used to delete the transport from whichever list
890 * it's sk_xprt.xpt_ready node was on, but we don't actually
891 * need to. This is because the only time we're called
892 * while still attached to a queue, the queue itself
893 * is about to be destroyed (in svc_destroy).
895 if (test_bit(XPT_TEMP, &xprt->xpt_flags))
898 while ((dr = svc_deferred_dequeue(xprt)) != NULL)
902 spin_unlock_bh(&serv->sv_lock);
905 void svc_close_xprt(struct svc_xprt *xprt)
907 set_bit(XPT_CLOSE, &xprt->xpt_flags);
908 if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
909 /* someone else will have to effect the close */
913 svc_delete_xprt(xprt);
914 clear_bit(XPT_BUSY, &xprt->xpt_flags);
917 EXPORT_SYMBOL_GPL(svc_close_xprt);
919 void svc_close_all(struct list_head *xprt_list)
921 struct svc_xprt *xprt;
922 struct svc_xprt *tmp;
924 list_for_each_entry_safe(xprt, tmp, xprt_list, xpt_list) {
925 set_bit(XPT_CLOSE, &xprt->xpt_flags);
926 if (test_bit(XPT_BUSY, &xprt->xpt_flags)) {
927 /* Waiting to be processed, but no threads left,
928 * So just remove it from the waiting list
930 list_del_init(&xprt->xpt_ready);
931 clear_bit(XPT_BUSY, &xprt->xpt_flags);
933 svc_close_xprt(xprt);
938 * Handle defer and revisit of requests
941 static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
943 struct svc_deferred_req *dr =
944 container_of(dreq, struct svc_deferred_req, handle);
945 struct svc_xprt *xprt = dr->xprt;
947 spin_lock(&xprt->xpt_lock);
948 set_bit(XPT_DEFERRED, &xprt->xpt_flags);
949 if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) {
950 spin_unlock(&xprt->xpt_lock);
951 dprintk("revisit canceled\n");
956 dprintk("revisit queued\n");
958 list_add(&dr->handle.recent, &xprt->xpt_deferred);
959 spin_unlock(&xprt->xpt_lock);
960 svc_xprt_enqueue(xprt);
965 * Save the request off for later processing. The request buffer looks
968 * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
970 * This code can only handle requests that consist of an xprt-header
973 static struct cache_deferred_req *svc_defer(struct cache_req *req)
975 struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
976 struct svc_deferred_req *dr;
978 if (rqstp->rq_arg.page_len || !rqstp->rq_usedeferral)
979 return NULL; /* if more than a page, give up FIXME */
980 if (rqstp->rq_deferred) {
981 dr = rqstp->rq_deferred;
982 rqstp->rq_deferred = NULL;
986 /* FIXME maybe discard if size too large */
987 size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
988 dr = kmalloc(size, GFP_KERNEL);
992 dr->handle.owner = rqstp->rq_server;
993 dr->prot = rqstp->rq_prot;
994 memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
995 dr->addrlen = rqstp->rq_addrlen;
996 dr->daddr = rqstp->rq_daddr;
997 dr->argslen = rqstp->rq_arg.len >> 2;
998 dr->xprt_hlen = rqstp->rq_xprt_hlen;
1000 /* back up head to the start of the buffer and copy */
1001 skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1002 memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
1005 svc_xprt_get(rqstp->rq_xprt);
1006 dr->xprt = rqstp->rq_xprt;
1008 dr->handle.revisit = svc_revisit;
1013 * recv data from a deferred request into an active one
1015 static int svc_deferred_recv(struct svc_rqst *rqstp)
1017 struct svc_deferred_req *dr = rqstp->rq_deferred;
1019 /* setup iov_base past transport header */
1020 rqstp->rq_arg.head[0].iov_base = dr->args + (dr->xprt_hlen>>2);
1021 /* The iov_len does not include the transport header bytes */
1022 rqstp->rq_arg.head[0].iov_len = (dr->argslen<<2) - dr->xprt_hlen;
1023 rqstp->rq_arg.page_len = 0;
1024 /* The rq_arg.len includes the transport header bytes */
1025 rqstp->rq_arg.len = dr->argslen<<2;
1026 rqstp->rq_prot = dr->prot;
1027 memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
1028 rqstp->rq_addrlen = dr->addrlen;
1029 /* Save off transport header len in case we get deferred again */
1030 rqstp->rq_xprt_hlen = dr->xprt_hlen;
1031 rqstp->rq_daddr = dr->daddr;
1032 rqstp->rq_respages = rqstp->rq_pages;
1033 return (dr->argslen<<2) - dr->xprt_hlen;
1037 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt)
1039 struct svc_deferred_req *dr = NULL;
1041 if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags))
1043 spin_lock(&xprt->xpt_lock);
1044 clear_bit(XPT_DEFERRED, &xprt->xpt_flags);
1045 if (!list_empty(&xprt->xpt_deferred)) {
1046 dr = list_entry(xprt->xpt_deferred.next,
1047 struct svc_deferred_req,
1049 list_del_init(&dr->handle.recent);
1050 set_bit(XPT_DEFERRED, &xprt->xpt_flags);
1052 spin_unlock(&xprt->xpt_lock);
1057 * svc_find_xprt - find an RPC transport instance
1058 * @serv: pointer to svc_serv to search
1059 * @xcl_name: C string containing transport's class name
1060 * @af: Address family of transport's local address
1061 * @port: transport's IP port number
1063 * Return the transport instance pointer for the endpoint accepting
1064 * connections/peer traffic from the specified transport class,
1065 * address family and port.
1067 * Specifying 0 for the address family or port is effectively a
1068 * wild-card, and will result in matching the first transport in the
1069 * service's list that has a matching class name.
1071 struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name,
1072 const sa_family_t af, const unsigned short port)
1074 struct svc_xprt *xprt;
1075 struct svc_xprt *found = NULL;
1077 /* Sanity check the args */
1078 if (serv == NULL || xcl_name == NULL)
1081 spin_lock_bh(&serv->sv_lock);
1082 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1083 if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1085 if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family)
1087 if (port != 0 && port != svc_xprt_local_port(xprt))
1093 spin_unlock_bh(&serv->sv_lock);
1096 EXPORT_SYMBOL_GPL(svc_find_xprt);
1098 static int svc_one_xprt_name(const struct svc_xprt *xprt,
1099 char *pos, int remaining)
1103 len = snprintf(pos, remaining, "%s %u\n",
1104 xprt->xpt_class->xcl_name,
1105 svc_xprt_local_port(xprt));
1106 if (len >= remaining)
1107 return -ENAMETOOLONG;
1112 * svc_xprt_names - format a buffer with a list of transport names
1113 * @serv: pointer to an RPC service
1114 * @buf: pointer to a buffer to be filled in
1115 * @buflen: length of buffer to be filled in
1117 * Fills in @buf with a string containing a list of transport names,
1118 * each name terminated with '\n'.
1120 * Returns positive length of the filled-in string on success; otherwise
1121 * a negative errno value is returned if an error occurs.
1123 int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen)
1125 struct svc_xprt *xprt;
1129 /* Sanity check args */
1133 spin_lock_bh(&serv->sv_lock);
1137 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1138 len = svc_one_xprt_name(xprt, pos, buflen - totlen);
1150 spin_unlock_bh(&serv->sv_lock);
1153 EXPORT_SYMBOL_GPL(svc_xprt_names);
1156 /*----------------------------------------------------------------------------*/
1158 static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos)
1160 unsigned int pidx = (unsigned int)*pos;
1161 struct svc_serv *serv = m->private;
1163 dprintk("svc_pool_stats_start, *pidx=%u\n", pidx);
1166 return SEQ_START_TOKEN;
1167 return (pidx > serv->sv_nrpools ? NULL : &serv->sv_pools[pidx-1]);
1170 static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos)
1172 struct svc_pool *pool = p;
1173 struct svc_serv *serv = m->private;
1175 dprintk("svc_pool_stats_next, *pos=%llu\n", *pos);
1177 if (p == SEQ_START_TOKEN) {
1178 pool = &serv->sv_pools[0];
1180 unsigned int pidx = (pool - &serv->sv_pools[0]);
1181 if (pidx < serv->sv_nrpools-1)
1182 pool = &serv->sv_pools[pidx+1];
1190 static void svc_pool_stats_stop(struct seq_file *m, void *p)
1194 static int svc_pool_stats_show(struct seq_file *m, void *p)
1196 struct svc_pool *pool = p;
1198 if (p == SEQ_START_TOKEN) {
1199 seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n");
1203 seq_printf(m, "%u %lu %lu %lu %lu\n",
1205 pool->sp_stats.packets,
1206 pool->sp_stats.sockets_queued,
1207 pool->sp_stats.threads_woken,
1208 pool->sp_stats.threads_timedout);
1213 static const struct seq_operations svc_pool_stats_seq_ops = {
1214 .start = svc_pool_stats_start,
1215 .next = svc_pool_stats_next,
1216 .stop = svc_pool_stats_stop,
1217 .show = svc_pool_stats_show,
1220 int svc_pool_stats_open(struct svc_serv *serv, struct file *file)
1224 err = seq_open(file, &svc_pool_stats_seq_ops);
1226 ((struct seq_file *) file->private_data)->private = serv;
1229 EXPORT_SYMBOL(svc_pool_stats_open);
1231 /*----------------------------------------------------------------------------*/