xref: /linux/net/sunrpc/svc_xprt.c (revision b85d45947951d23cb22d90caecf4c1eb81342c96)
1 /*
2  * linux/net/sunrpc/svc_xprt.c
3  *
4  * Author: Tom Tucker <tom@opengridcomputing.com>
5  */
6 
7 #include <linux/sched.h>
8 #include <linux/errno.h>
9 #include <linux/freezer.h>
10 #include <linux/kthread.h>
11 #include <linux/slab.h>
12 #include <net/sock.h>
13 #include <linux/sunrpc/stats.h>
14 #include <linux/sunrpc/svc_xprt.h>
15 #include <linux/sunrpc/svcsock.h>
16 #include <linux/sunrpc/xprt.h>
17 #include <linux/module.h>
18 #include <trace/events/sunrpc.h>
19 
20 #define RPCDBG_FACILITY	RPCDBG_SVCXPRT
21 
22 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt);
23 static int svc_deferred_recv(struct svc_rqst *rqstp);
24 static struct cache_deferred_req *svc_defer(struct cache_req *req);
25 static void svc_age_temp_xprts(unsigned long closure);
26 static void svc_delete_xprt(struct svc_xprt *xprt);
27 
28 /* apparently the "standard" is that clients close
29  * idle connections after 5 minutes, servers after
30  * 6 minutes
31  *   http://www.connectathon.org/talks96/nfstcp.pdf
32  */
33 static int svc_conn_age_period = 6*60;
34 
35 /* List of registered transport classes */
36 static DEFINE_SPINLOCK(svc_xprt_class_lock);
37 static LIST_HEAD(svc_xprt_class_list);
38 
39 /* SMP locking strategy:
40  *
41  *	svc_pool->sp_lock protects most of the fields of that pool.
42  *	svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
43  *	when both need to be taken (rare), svc_serv->sv_lock is first.
44  *	The "service mutex" protects svc_serv->sv_nrthread.
45  *	svc_sock->sk_lock protects the svc_sock->sk_deferred list
46  *             and the ->sk_info_authunix cache.
47  *
48  *	The XPT_BUSY bit in xprt->xpt_flags prevents a transport being
49  *	enqueued multiply. During normal transport processing this bit
50  *	is set by svc_xprt_enqueue and cleared by svc_xprt_received.
51  *	Providers should not manipulate this bit directly.
52  *
53  *	Some flags can be set to certain values at any time
54  *	providing that certain rules are followed:
55  *
56  *	XPT_CONN, XPT_DATA:
57  *		- Can be set or cleared at any time.
58  *		- After a set, svc_xprt_enqueue must be called to enqueue
59  *		  the transport for processing.
60  *		- After a clear, the transport must be read/accepted.
61  *		  If this succeeds, it must be set again.
62  *	XPT_CLOSE:
63  *		- Can set at any time. It is never cleared.
64  *      XPT_DEAD:
65  *		- Can only be set while XPT_BUSY is held which ensures
66  *		  that no other thread will be using the transport or will
67  *		  try to set XPT_DEAD.
68  */
69 int svc_reg_xprt_class(struct svc_xprt_class *xcl)
70 {
71 	struct svc_xprt_class *cl;
72 	int res = -EEXIST;
73 
74 	dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name);
75 
76 	INIT_LIST_HEAD(&xcl->xcl_list);
77 	spin_lock(&svc_xprt_class_lock);
78 	/* Make sure there isn't already a class with the same name */
79 	list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) {
80 		if (strcmp(xcl->xcl_name, cl->xcl_name) == 0)
81 			goto out;
82 	}
83 	list_add_tail(&xcl->xcl_list, &svc_xprt_class_list);
84 	res = 0;
85 out:
86 	spin_unlock(&svc_xprt_class_lock);
87 	return res;
88 }
89 EXPORT_SYMBOL_GPL(svc_reg_xprt_class);
90 
91 void svc_unreg_xprt_class(struct svc_xprt_class *xcl)
92 {
93 	dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name);
94 	spin_lock(&svc_xprt_class_lock);
95 	list_del_init(&xcl->xcl_list);
96 	spin_unlock(&svc_xprt_class_lock);
97 }
98 EXPORT_SYMBOL_GPL(svc_unreg_xprt_class);
99 
100 /*
101  * Format the transport list for printing
102  */
103 int svc_print_xprts(char *buf, int maxlen)
104 {
105 	struct svc_xprt_class *xcl;
106 	char tmpstr[80];
107 	int len = 0;
108 	buf[0] = '\0';
109 
110 	spin_lock(&svc_xprt_class_lock);
111 	list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
112 		int slen;
113 
114 		sprintf(tmpstr, "%s %d\n", xcl->xcl_name, xcl->xcl_max_payload);
115 		slen = strlen(tmpstr);
116 		if (len + slen > maxlen)
117 			break;
118 		len += slen;
119 		strcat(buf, tmpstr);
120 	}
121 	spin_unlock(&svc_xprt_class_lock);
122 
123 	return len;
124 }
125 
126 static void svc_xprt_free(struct kref *kref)
127 {
128 	struct svc_xprt *xprt =
129 		container_of(kref, struct svc_xprt, xpt_ref);
130 	struct module *owner = xprt->xpt_class->xcl_owner;
131 	if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags))
132 		svcauth_unix_info_release(xprt);
133 	put_net(xprt->xpt_net);
134 	/* See comment on corresponding get in xs_setup_bc_tcp(): */
135 	if (xprt->xpt_bc_xprt)
136 		xprt_put(xprt->xpt_bc_xprt);
137 	xprt->xpt_ops->xpo_free(xprt);
138 	module_put(owner);
139 }
140 
141 void svc_xprt_put(struct svc_xprt *xprt)
142 {
143 	kref_put(&xprt->xpt_ref, svc_xprt_free);
144 }
145 EXPORT_SYMBOL_GPL(svc_xprt_put);
146 
147 /*
148  * Called by transport drivers to initialize the transport independent
149  * portion of the transport instance.
150  */
151 void svc_xprt_init(struct net *net, struct svc_xprt_class *xcl,
152 		   struct svc_xprt *xprt, struct svc_serv *serv)
153 {
154 	memset(xprt, 0, sizeof(*xprt));
155 	xprt->xpt_class = xcl;
156 	xprt->xpt_ops = xcl->xcl_ops;
157 	kref_init(&xprt->xpt_ref);
158 	xprt->xpt_server = serv;
159 	INIT_LIST_HEAD(&xprt->xpt_list);
160 	INIT_LIST_HEAD(&xprt->xpt_ready);
161 	INIT_LIST_HEAD(&xprt->xpt_deferred);
162 	INIT_LIST_HEAD(&xprt->xpt_users);
163 	mutex_init(&xprt->xpt_mutex);
164 	spin_lock_init(&xprt->xpt_lock);
165 	set_bit(XPT_BUSY, &xprt->xpt_flags);
166 	rpc_init_wait_queue(&xprt->xpt_bc_pending, "xpt_bc_pending");
167 	xprt->xpt_net = get_net(net);
168 }
169 EXPORT_SYMBOL_GPL(svc_xprt_init);
170 
171 static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl,
172 					 struct svc_serv *serv,
173 					 struct net *net,
174 					 const int family,
175 					 const unsigned short port,
176 					 int flags)
177 {
178 	struct sockaddr_in sin = {
179 		.sin_family		= AF_INET,
180 		.sin_addr.s_addr	= htonl(INADDR_ANY),
181 		.sin_port		= htons(port),
182 	};
183 #if IS_ENABLED(CONFIG_IPV6)
184 	struct sockaddr_in6 sin6 = {
185 		.sin6_family		= AF_INET6,
186 		.sin6_addr		= IN6ADDR_ANY_INIT,
187 		.sin6_port		= htons(port),
188 	};
189 #endif
190 	struct sockaddr *sap;
191 	size_t len;
192 
193 	switch (family) {
194 	case PF_INET:
195 		sap = (struct sockaddr *)&sin;
196 		len = sizeof(sin);
197 		break;
198 #if IS_ENABLED(CONFIG_IPV6)
199 	case PF_INET6:
200 		sap = (struct sockaddr *)&sin6;
201 		len = sizeof(sin6);
202 		break;
203 #endif
204 	default:
205 		return ERR_PTR(-EAFNOSUPPORT);
206 	}
207 
208 	return xcl->xcl_ops->xpo_create(serv, net, sap, len, flags);
209 }
210 
211 /*
212  * svc_xprt_received conditionally queues the transport for processing
213  * by another thread. The caller must hold the XPT_BUSY bit and must
214  * not thereafter touch transport data.
215  *
216  * Note: XPT_DATA only gets cleared when a read-attempt finds no (or
217  * insufficient) data.
218  */
219 static void svc_xprt_received(struct svc_xprt *xprt)
220 {
221 	if (!test_bit(XPT_BUSY, &xprt->xpt_flags)) {
222 		WARN_ONCE(1, "xprt=0x%p already busy!", xprt);
223 		return;
224 	}
225 
226 	/* As soon as we clear busy, the xprt could be closed and
227 	 * 'put', so we need a reference to call svc_enqueue_xprt with:
228 	 */
229 	svc_xprt_get(xprt);
230 	smp_mb__before_atomic();
231 	clear_bit(XPT_BUSY, &xprt->xpt_flags);
232 	xprt->xpt_server->sv_ops->svo_enqueue_xprt(xprt);
233 	svc_xprt_put(xprt);
234 }
235 
236 void svc_add_new_perm_xprt(struct svc_serv *serv, struct svc_xprt *new)
237 {
238 	clear_bit(XPT_TEMP, &new->xpt_flags);
239 	spin_lock_bh(&serv->sv_lock);
240 	list_add(&new->xpt_list, &serv->sv_permsocks);
241 	spin_unlock_bh(&serv->sv_lock);
242 	svc_xprt_received(new);
243 }
244 
245 int svc_create_xprt(struct svc_serv *serv, const char *xprt_name,
246 		    struct net *net, const int family,
247 		    const unsigned short port, int flags)
248 {
249 	struct svc_xprt_class *xcl;
250 
251 	dprintk("svc: creating transport %s[%d]\n", xprt_name, port);
252 	spin_lock(&svc_xprt_class_lock);
253 	list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
254 		struct svc_xprt *newxprt;
255 		unsigned short newport;
256 
257 		if (strcmp(xprt_name, xcl->xcl_name))
258 			continue;
259 
260 		if (!try_module_get(xcl->xcl_owner))
261 			goto err;
262 
263 		spin_unlock(&svc_xprt_class_lock);
264 		newxprt = __svc_xpo_create(xcl, serv, net, family, port, flags);
265 		if (IS_ERR(newxprt)) {
266 			module_put(xcl->xcl_owner);
267 			return PTR_ERR(newxprt);
268 		}
269 		svc_add_new_perm_xprt(serv, newxprt);
270 		newport = svc_xprt_local_port(newxprt);
271 		return newport;
272 	}
273  err:
274 	spin_unlock(&svc_xprt_class_lock);
275 	dprintk("svc: transport %s not found\n", xprt_name);
276 
277 	/* This errno is exposed to user space.  Provide a reasonable
278 	 * perror msg for a bad transport. */
279 	return -EPROTONOSUPPORT;
280 }
281 EXPORT_SYMBOL_GPL(svc_create_xprt);
282 
283 /*
284  * Copy the local and remote xprt addresses to the rqstp structure
285  */
286 void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt)
287 {
288 	memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen);
289 	rqstp->rq_addrlen = xprt->xpt_remotelen;
290 
291 	/*
292 	 * Destination address in request is needed for binding the
293 	 * source address in RPC replies/callbacks later.
294 	 */
295 	memcpy(&rqstp->rq_daddr, &xprt->xpt_local, xprt->xpt_locallen);
296 	rqstp->rq_daddrlen = xprt->xpt_locallen;
297 }
298 EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs);
299 
300 /**
301  * svc_print_addr - Format rq_addr field for printing
302  * @rqstp: svc_rqst struct containing address to print
303  * @buf: target buffer for formatted address
304  * @len: length of target buffer
305  *
306  */
307 char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
308 {
309 	return __svc_print_addr(svc_addr(rqstp), buf, len);
310 }
311 EXPORT_SYMBOL_GPL(svc_print_addr);
312 
313 static bool svc_xprt_has_something_to_do(struct svc_xprt *xprt)
314 {
315 	if (xprt->xpt_flags & ((1<<XPT_CONN)|(1<<XPT_CLOSE)))
316 		return true;
317 	if (xprt->xpt_flags & ((1<<XPT_DATA)|(1<<XPT_DEFERRED)))
318 		return xprt->xpt_ops->xpo_has_wspace(xprt);
319 	return false;
320 }
321 
322 void svc_xprt_do_enqueue(struct svc_xprt *xprt)
323 {
324 	struct svc_pool *pool;
325 	struct svc_rqst	*rqstp = NULL;
326 	int cpu;
327 	bool queued = false;
328 
329 	if (!svc_xprt_has_something_to_do(xprt))
330 		goto out;
331 
332 	/* Mark transport as busy. It will remain in this state until
333 	 * the provider calls svc_xprt_received. We update XPT_BUSY
334 	 * atomically because it also guards against trying to enqueue
335 	 * the transport twice.
336 	 */
337 	if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) {
338 		/* Don't enqueue transport while already enqueued */
339 		dprintk("svc: transport %p busy, not enqueued\n", xprt);
340 		goto out;
341 	}
342 
343 	cpu = get_cpu();
344 	pool = svc_pool_for_cpu(xprt->xpt_server, cpu);
345 
346 	atomic_long_inc(&pool->sp_stats.packets);
347 
348 redo_search:
349 	/* find a thread for this xprt */
350 	rcu_read_lock();
351 	list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) {
352 		/* Do a lockless check first */
353 		if (test_bit(RQ_BUSY, &rqstp->rq_flags))
354 			continue;
355 
356 		/*
357 		 * Once the xprt has been queued, it can only be dequeued by
358 		 * the task that intends to service it. All we can do at that
359 		 * point is to try to wake this thread back up so that it can
360 		 * do so.
361 		 */
362 		if (!queued) {
363 			spin_lock_bh(&rqstp->rq_lock);
364 			if (test_and_set_bit(RQ_BUSY, &rqstp->rq_flags)) {
365 				/* already busy, move on... */
366 				spin_unlock_bh(&rqstp->rq_lock);
367 				continue;
368 			}
369 
370 			/* this one will do */
371 			rqstp->rq_xprt = xprt;
372 			svc_xprt_get(xprt);
373 			spin_unlock_bh(&rqstp->rq_lock);
374 		}
375 		rcu_read_unlock();
376 
377 		atomic_long_inc(&pool->sp_stats.threads_woken);
378 		wake_up_process(rqstp->rq_task);
379 		put_cpu();
380 		goto out;
381 	}
382 	rcu_read_unlock();
383 
384 	/*
385 	 * We didn't find an idle thread to use, so we need to queue the xprt.
386 	 * Do so and then search again. If we find one, we can't hook this one
387 	 * up to it directly but we can wake the thread up in the hopes that it
388 	 * will pick it up once it searches for a xprt to service.
389 	 */
390 	if (!queued) {
391 		queued = true;
392 		dprintk("svc: transport %p put into queue\n", xprt);
393 		spin_lock_bh(&pool->sp_lock);
394 		list_add_tail(&xprt->xpt_ready, &pool->sp_sockets);
395 		pool->sp_stats.sockets_queued++;
396 		spin_unlock_bh(&pool->sp_lock);
397 		goto redo_search;
398 	}
399 	rqstp = NULL;
400 	put_cpu();
401 out:
402 	trace_svc_xprt_do_enqueue(xprt, rqstp);
403 }
404 EXPORT_SYMBOL_GPL(svc_xprt_do_enqueue);
405 
406 /*
407  * Queue up a transport with data pending. If there are idle nfsd
408  * processes, wake 'em up.
409  *
410  */
411 void svc_xprt_enqueue(struct svc_xprt *xprt)
412 {
413 	if (test_bit(XPT_BUSY, &xprt->xpt_flags))
414 		return;
415 	xprt->xpt_server->sv_ops->svo_enqueue_xprt(xprt);
416 }
417 EXPORT_SYMBOL_GPL(svc_xprt_enqueue);
418 
419 /*
420  * Dequeue the first transport, if there is one.
421  */
422 static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool)
423 {
424 	struct svc_xprt	*xprt = NULL;
425 
426 	if (list_empty(&pool->sp_sockets))
427 		goto out;
428 
429 	spin_lock_bh(&pool->sp_lock);
430 	if (likely(!list_empty(&pool->sp_sockets))) {
431 		xprt = list_first_entry(&pool->sp_sockets,
432 					struct svc_xprt, xpt_ready);
433 		list_del_init(&xprt->xpt_ready);
434 		svc_xprt_get(xprt);
435 
436 		dprintk("svc: transport %p dequeued, inuse=%d\n",
437 			xprt, atomic_read(&xprt->xpt_ref.refcount));
438 	}
439 	spin_unlock_bh(&pool->sp_lock);
440 out:
441 	trace_svc_xprt_dequeue(xprt);
442 	return xprt;
443 }
444 
445 /**
446  * svc_reserve - change the space reserved for the reply to a request.
447  * @rqstp:  The request in question
448  * @space: new max space to reserve
449  *
450  * Each request reserves some space on the output queue of the transport
451  * to make sure the reply fits.  This function reduces that reserved
452  * space to be the amount of space used already, plus @space.
453  *
454  */
455 void svc_reserve(struct svc_rqst *rqstp, int space)
456 {
457 	space += rqstp->rq_res.head[0].iov_len;
458 
459 	if (space < rqstp->rq_reserved) {
460 		struct svc_xprt *xprt = rqstp->rq_xprt;
461 		atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved);
462 		rqstp->rq_reserved = space;
463 
464 		if (xprt->xpt_ops->xpo_adjust_wspace)
465 			xprt->xpt_ops->xpo_adjust_wspace(xprt);
466 		svc_xprt_enqueue(xprt);
467 	}
468 }
469 EXPORT_SYMBOL_GPL(svc_reserve);
470 
471 static void svc_xprt_release(struct svc_rqst *rqstp)
472 {
473 	struct svc_xprt	*xprt = rqstp->rq_xprt;
474 
475 	rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
476 
477 	kfree(rqstp->rq_deferred);
478 	rqstp->rq_deferred = NULL;
479 
480 	svc_free_res_pages(rqstp);
481 	rqstp->rq_res.page_len = 0;
482 	rqstp->rq_res.page_base = 0;
483 
484 	/* Reset response buffer and release
485 	 * the reservation.
486 	 * But first, check that enough space was reserved
487 	 * for the reply, otherwise we have a bug!
488 	 */
489 	if ((rqstp->rq_res.len) >  rqstp->rq_reserved)
490 		printk(KERN_ERR "RPC request reserved %d but used %d\n",
491 		       rqstp->rq_reserved,
492 		       rqstp->rq_res.len);
493 
494 	rqstp->rq_res.head[0].iov_len = 0;
495 	svc_reserve(rqstp, 0);
496 	rqstp->rq_xprt = NULL;
497 
498 	svc_xprt_put(xprt);
499 }
500 
501 /*
502  * Some svc_serv's will have occasional work to do, even when a xprt is not
503  * waiting to be serviced. This function is there to "kick" a task in one of
504  * those services so that it can wake up and do that work. Note that we only
505  * bother with pool 0 as we don't need to wake up more than one thread for
506  * this purpose.
507  */
508 void svc_wake_up(struct svc_serv *serv)
509 {
510 	struct svc_rqst	*rqstp;
511 	struct svc_pool *pool;
512 
513 	pool = &serv->sv_pools[0];
514 
515 	rcu_read_lock();
516 	list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) {
517 		/* skip any that aren't queued */
518 		if (test_bit(RQ_BUSY, &rqstp->rq_flags))
519 			continue;
520 		rcu_read_unlock();
521 		dprintk("svc: daemon %p woken up.\n", rqstp);
522 		wake_up_process(rqstp->rq_task);
523 		trace_svc_wake_up(rqstp->rq_task->pid);
524 		return;
525 	}
526 	rcu_read_unlock();
527 
528 	/* No free entries available */
529 	set_bit(SP_TASK_PENDING, &pool->sp_flags);
530 	smp_wmb();
531 	trace_svc_wake_up(0);
532 }
533 EXPORT_SYMBOL_GPL(svc_wake_up);
534 
535 int svc_port_is_privileged(struct sockaddr *sin)
536 {
537 	switch (sin->sa_family) {
538 	case AF_INET:
539 		return ntohs(((struct sockaddr_in *)sin)->sin_port)
540 			< PROT_SOCK;
541 	case AF_INET6:
542 		return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
543 			< PROT_SOCK;
544 	default:
545 		return 0;
546 	}
547 }
548 
549 /*
550  * Make sure that we don't have too many active connections. If we have,
551  * something must be dropped. It's not clear what will happen if we allow
552  * "too many" connections, but when dealing with network-facing software,
553  * we have to code defensively. Here we do that by imposing hard limits.
554  *
555  * There's no point in trying to do random drop here for DoS
556  * prevention. The NFS clients does 1 reconnect in 15 seconds. An
557  * attacker can easily beat that.
558  *
559  * The only somewhat efficient mechanism would be if drop old
560  * connections from the same IP first. But right now we don't even
561  * record the client IP in svc_sock.
562  *
563  * single-threaded services that expect a lot of clients will probably
564  * need to set sv_maxconn to override the default value which is based
565  * on the number of threads
566  */
567 static void svc_check_conn_limits(struct svc_serv *serv)
568 {
569 	unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn :
570 				(serv->sv_nrthreads+3) * 20;
571 
572 	if (serv->sv_tmpcnt > limit) {
573 		struct svc_xprt *xprt = NULL;
574 		spin_lock_bh(&serv->sv_lock);
575 		if (!list_empty(&serv->sv_tempsocks)) {
576 			/* Try to help the admin */
577 			net_notice_ratelimited("%s: too many open connections, consider increasing the %s\n",
578 					       serv->sv_name, serv->sv_maxconn ?
579 					       "max number of connections" :
580 					       "number of threads");
581 			/*
582 			 * Always select the oldest connection. It's not fair,
583 			 * but so is life
584 			 */
585 			xprt = list_entry(serv->sv_tempsocks.prev,
586 					  struct svc_xprt,
587 					  xpt_list);
588 			set_bit(XPT_CLOSE, &xprt->xpt_flags);
589 			svc_xprt_get(xprt);
590 		}
591 		spin_unlock_bh(&serv->sv_lock);
592 
593 		if (xprt) {
594 			svc_xprt_enqueue(xprt);
595 			svc_xprt_put(xprt);
596 		}
597 	}
598 }
599 
600 static int svc_alloc_arg(struct svc_rqst *rqstp)
601 {
602 	struct svc_serv *serv = rqstp->rq_server;
603 	struct xdr_buf *arg;
604 	int pages;
605 	int i;
606 
607 	/* now allocate needed pages.  If we get a failure, sleep briefly */
608 	pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE;
609 	WARN_ON_ONCE(pages >= RPCSVC_MAXPAGES);
610 	if (pages >= RPCSVC_MAXPAGES)
611 		/* use as many pages as possible */
612 		pages = RPCSVC_MAXPAGES - 1;
613 	for (i = 0; i < pages ; i++)
614 		while (rqstp->rq_pages[i] == NULL) {
615 			struct page *p = alloc_page(GFP_KERNEL);
616 			if (!p) {
617 				set_current_state(TASK_INTERRUPTIBLE);
618 				if (signalled() || kthread_should_stop()) {
619 					set_current_state(TASK_RUNNING);
620 					return -EINTR;
621 				}
622 				schedule_timeout(msecs_to_jiffies(500));
623 			}
624 			rqstp->rq_pages[i] = p;
625 		}
626 	rqstp->rq_page_end = &rqstp->rq_pages[i];
627 	rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */
628 
629 	/* Make arg->head point to first page and arg->pages point to rest */
630 	arg = &rqstp->rq_arg;
631 	arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
632 	arg->head[0].iov_len = PAGE_SIZE;
633 	arg->pages = rqstp->rq_pages + 1;
634 	arg->page_base = 0;
635 	/* save at least one page for response */
636 	arg->page_len = (pages-2)*PAGE_SIZE;
637 	arg->len = (pages-1)*PAGE_SIZE;
638 	arg->tail[0].iov_len = 0;
639 	return 0;
640 }
641 
642 static bool
643 rqst_should_sleep(struct svc_rqst *rqstp)
644 {
645 	struct svc_pool		*pool = rqstp->rq_pool;
646 
647 	/* did someone call svc_wake_up? */
648 	if (test_and_clear_bit(SP_TASK_PENDING, &pool->sp_flags))
649 		return false;
650 
651 	/* was a socket queued? */
652 	if (!list_empty(&pool->sp_sockets))
653 		return false;
654 
655 	/* are we shutting down? */
656 	if (signalled() || kthread_should_stop())
657 		return false;
658 
659 	/* are we freezing? */
660 	if (freezing(current))
661 		return false;
662 
663 	return true;
664 }
665 
666 static struct svc_xprt *svc_get_next_xprt(struct svc_rqst *rqstp, long timeout)
667 {
668 	struct svc_xprt *xprt;
669 	struct svc_pool		*pool = rqstp->rq_pool;
670 	long			time_left = 0;
671 
672 	/* rq_xprt should be clear on entry */
673 	WARN_ON_ONCE(rqstp->rq_xprt);
674 
675 	/* Normally we will wait up to 5 seconds for any required
676 	 * cache information to be provided.
677 	 */
678 	rqstp->rq_chandle.thread_wait = 5*HZ;
679 
680 	xprt = svc_xprt_dequeue(pool);
681 	if (xprt) {
682 		rqstp->rq_xprt = xprt;
683 
684 		/* As there is a shortage of threads and this request
685 		 * had to be queued, don't allow the thread to wait so
686 		 * long for cache updates.
687 		 */
688 		rqstp->rq_chandle.thread_wait = 1*HZ;
689 		clear_bit(SP_TASK_PENDING, &pool->sp_flags);
690 		return xprt;
691 	}
692 
693 	/*
694 	 * We have to be able to interrupt this wait
695 	 * to bring down the daemons ...
696 	 */
697 	set_current_state(TASK_INTERRUPTIBLE);
698 	clear_bit(RQ_BUSY, &rqstp->rq_flags);
699 	smp_mb();
700 
701 	if (likely(rqst_should_sleep(rqstp)))
702 		time_left = schedule_timeout(timeout);
703 	else
704 		__set_current_state(TASK_RUNNING);
705 
706 	try_to_freeze();
707 
708 	spin_lock_bh(&rqstp->rq_lock);
709 	set_bit(RQ_BUSY, &rqstp->rq_flags);
710 	spin_unlock_bh(&rqstp->rq_lock);
711 
712 	xprt = rqstp->rq_xprt;
713 	if (xprt != NULL)
714 		return xprt;
715 
716 	if (!time_left)
717 		atomic_long_inc(&pool->sp_stats.threads_timedout);
718 
719 	if (signalled() || kthread_should_stop())
720 		return ERR_PTR(-EINTR);
721 	return ERR_PTR(-EAGAIN);
722 }
723 
724 static void svc_add_new_temp_xprt(struct svc_serv *serv, struct svc_xprt *newxpt)
725 {
726 	spin_lock_bh(&serv->sv_lock);
727 	set_bit(XPT_TEMP, &newxpt->xpt_flags);
728 	list_add(&newxpt->xpt_list, &serv->sv_tempsocks);
729 	serv->sv_tmpcnt++;
730 	if (serv->sv_temptimer.function == NULL) {
731 		/* setup timer to age temp transports */
732 		setup_timer(&serv->sv_temptimer, svc_age_temp_xprts,
733 			    (unsigned long)serv);
734 		mod_timer(&serv->sv_temptimer,
735 			  jiffies + svc_conn_age_period * HZ);
736 	}
737 	spin_unlock_bh(&serv->sv_lock);
738 	svc_xprt_received(newxpt);
739 }
740 
741 static int svc_handle_xprt(struct svc_rqst *rqstp, struct svc_xprt *xprt)
742 {
743 	struct svc_serv *serv = rqstp->rq_server;
744 	int len = 0;
745 
746 	if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) {
747 		dprintk("svc_recv: found XPT_CLOSE\n");
748 		svc_delete_xprt(xprt);
749 		/* Leave XPT_BUSY set on the dead xprt: */
750 		goto out;
751 	}
752 	if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
753 		struct svc_xprt *newxpt;
754 		/*
755 		 * We know this module_get will succeed because the
756 		 * listener holds a reference too
757 		 */
758 		__module_get(xprt->xpt_class->xcl_owner);
759 		svc_check_conn_limits(xprt->xpt_server);
760 		newxpt = xprt->xpt_ops->xpo_accept(xprt);
761 		if (newxpt)
762 			svc_add_new_temp_xprt(serv, newxpt);
763 		else
764 			module_put(xprt->xpt_class->xcl_owner);
765 	} else {
766 		/* XPT_DATA|XPT_DEFERRED case: */
767 		dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n",
768 			rqstp, rqstp->rq_pool->sp_id, xprt,
769 			atomic_read(&xprt->xpt_ref.refcount));
770 		rqstp->rq_deferred = svc_deferred_dequeue(xprt);
771 		if (rqstp->rq_deferred)
772 			len = svc_deferred_recv(rqstp);
773 		else
774 			len = xprt->xpt_ops->xpo_recvfrom(rqstp);
775 		dprintk("svc: got len=%d\n", len);
776 		rqstp->rq_reserved = serv->sv_max_mesg;
777 		atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
778 	}
779 	/* clear XPT_BUSY: */
780 	svc_xprt_received(xprt);
781 out:
782 	trace_svc_handle_xprt(xprt, len);
783 	return len;
784 }
785 
786 /*
787  * Receive the next request on any transport.  This code is carefully
788  * organised not to touch any cachelines in the shared svc_serv
789  * structure, only cachelines in the local svc_pool.
790  */
791 int svc_recv(struct svc_rqst *rqstp, long timeout)
792 {
793 	struct svc_xprt		*xprt = NULL;
794 	struct svc_serv		*serv = rqstp->rq_server;
795 	int			len, err;
796 
797 	dprintk("svc: server %p waiting for data (to = %ld)\n",
798 		rqstp, timeout);
799 
800 	if (rqstp->rq_xprt)
801 		printk(KERN_ERR
802 			"svc_recv: service %p, transport not NULL!\n",
803 			 rqstp);
804 
805 	err = svc_alloc_arg(rqstp);
806 	if (err)
807 		goto out;
808 
809 	try_to_freeze();
810 	cond_resched();
811 	err = -EINTR;
812 	if (signalled() || kthread_should_stop())
813 		goto out;
814 
815 	xprt = svc_get_next_xprt(rqstp, timeout);
816 	if (IS_ERR(xprt)) {
817 		err = PTR_ERR(xprt);
818 		goto out;
819 	}
820 
821 	len = svc_handle_xprt(rqstp, xprt);
822 
823 	/* No data, incomplete (TCP) read, or accept() */
824 	err = -EAGAIN;
825 	if (len <= 0)
826 		goto out_release;
827 
828 	clear_bit(XPT_OLD, &xprt->xpt_flags);
829 
830 	if (xprt->xpt_ops->xpo_secure_port(rqstp))
831 		set_bit(RQ_SECURE, &rqstp->rq_flags);
832 	else
833 		clear_bit(RQ_SECURE, &rqstp->rq_flags);
834 	rqstp->rq_chandle.defer = svc_defer;
835 	rqstp->rq_xid = svc_getu32(&rqstp->rq_arg.head[0]);
836 
837 	if (serv->sv_stats)
838 		serv->sv_stats->netcnt++;
839 	trace_svc_recv(rqstp, len);
840 	return len;
841 out_release:
842 	rqstp->rq_res.len = 0;
843 	svc_xprt_release(rqstp);
844 out:
845 	trace_svc_recv(rqstp, err);
846 	return err;
847 }
848 EXPORT_SYMBOL_GPL(svc_recv);
849 
850 /*
851  * Drop request
852  */
853 void svc_drop(struct svc_rqst *rqstp)
854 {
855 	dprintk("svc: xprt %p dropped request\n", rqstp->rq_xprt);
856 	svc_xprt_release(rqstp);
857 }
858 EXPORT_SYMBOL_GPL(svc_drop);
859 
860 /*
861  * Return reply to client.
862  */
863 int svc_send(struct svc_rqst *rqstp)
864 {
865 	struct svc_xprt	*xprt;
866 	int		len = -EFAULT;
867 	struct xdr_buf	*xb;
868 
869 	xprt = rqstp->rq_xprt;
870 	if (!xprt)
871 		goto out;
872 
873 	/* release the receive skb before sending the reply */
874 	rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
875 
876 	/* calculate over-all length */
877 	xb = &rqstp->rq_res;
878 	xb->len = xb->head[0].iov_len +
879 		xb->page_len +
880 		xb->tail[0].iov_len;
881 
882 	/* Grab mutex to serialize outgoing data. */
883 	mutex_lock(&xprt->xpt_mutex);
884 	if (test_bit(XPT_DEAD, &xprt->xpt_flags)
885 			|| test_bit(XPT_CLOSE, &xprt->xpt_flags))
886 		len = -ENOTCONN;
887 	else
888 		len = xprt->xpt_ops->xpo_sendto(rqstp);
889 	mutex_unlock(&xprt->xpt_mutex);
890 	rpc_wake_up(&xprt->xpt_bc_pending);
891 	svc_xprt_release(rqstp);
892 
893 	if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
894 		len = 0;
895 out:
896 	trace_svc_send(rqstp, len);
897 	return len;
898 }
899 
900 /*
901  * Timer function to close old temporary transports, using
902  * a mark-and-sweep algorithm.
903  */
904 static void svc_age_temp_xprts(unsigned long closure)
905 {
906 	struct svc_serv *serv = (struct svc_serv *)closure;
907 	struct svc_xprt *xprt;
908 	struct list_head *le, *next;
909 
910 	dprintk("svc_age_temp_xprts\n");
911 
912 	if (!spin_trylock_bh(&serv->sv_lock)) {
913 		/* busy, try again 1 sec later */
914 		dprintk("svc_age_temp_xprts: busy\n");
915 		mod_timer(&serv->sv_temptimer, jiffies + HZ);
916 		return;
917 	}
918 
919 	list_for_each_safe(le, next, &serv->sv_tempsocks) {
920 		xprt = list_entry(le, struct svc_xprt, xpt_list);
921 
922 		/* First time through, just mark it OLD. Second time
923 		 * through, close it. */
924 		if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags))
925 			continue;
926 		if (atomic_read(&xprt->xpt_ref.refcount) > 1 ||
927 		    test_bit(XPT_BUSY, &xprt->xpt_flags))
928 			continue;
929 		list_del_init(le);
930 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
931 		dprintk("queuing xprt %p for closing\n", xprt);
932 
933 		/* a thread will dequeue and close it soon */
934 		svc_xprt_enqueue(xprt);
935 	}
936 	spin_unlock_bh(&serv->sv_lock);
937 
938 	mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
939 }
940 
941 static void call_xpt_users(struct svc_xprt *xprt)
942 {
943 	struct svc_xpt_user *u;
944 
945 	spin_lock(&xprt->xpt_lock);
946 	while (!list_empty(&xprt->xpt_users)) {
947 		u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list);
948 		list_del(&u->list);
949 		u->callback(u);
950 	}
951 	spin_unlock(&xprt->xpt_lock);
952 }
953 
954 /*
955  * Remove a dead transport
956  */
957 static void svc_delete_xprt(struct svc_xprt *xprt)
958 {
959 	struct svc_serv	*serv = xprt->xpt_server;
960 	struct svc_deferred_req *dr;
961 
962 	/* Only do this once */
963 	if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags))
964 		BUG();
965 
966 	dprintk("svc: svc_delete_xprt(%p)\n", xprt);
967 	xprt->xpt_ops->xpo_detach(xprt);
968 
969 	spin_lock_bh(&serv->sv_lock);
970 	list_del_init(&xprt->xpt_list);
971 	WARN_ON_ONCE(!list_empty(&xprt->xpt_ready));
972 	if (test_bit(XPT_TEMP, &xprt->xpt_flags))
973 		serv->sv_tmpcnt--;
974 	spin_unlock_bh(&serv->sv_lock);
975 
976 	while ((dr = svc_deferred_dequeue(xprt)) != NULL)
977 		kfree(dr);
978 
979 	call_xpt_users(xprt);
980 	svc_xprt_put(xprt);
981 }
982 
983 void svc_close_xprt(struct svc_xprt *xprt)
984 {
985 	set_bit(XPT_CLOSE, &xprt->xpt_flags);
986 	if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
987 		/* someone else will have to effect the close */
988 		return;
989 	/*
990 	 * We expect svc_close_xprt() to work even when no threads are
991 	 * running (e.g., while configuring the server before starting
992 	 * any threads), so if the transport isn't busy, we delete
993 	 * it ourself:
994 	 */
995 	svc_delete_xprt(xprt);
996 }
997 EXPORT_SYMBOL_GPL(svc_close_xprt);
998 
999 static int svc_close_list(struct svc_serv *serv, struct list_head *xprt_list, struct net *net)
1000 {
1001 	struct svc_xprt *xprt;
1002 	int ret = 0;
1003 
1004 	spin_lock(&serv->sv_lock);
1005 	list_for_each_entry(xprt, xprt_list, xpt_list) {
1006 		if (xprt->xpt_net != net)
1007 			continue;
1008 		ret++;
1009 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
1010 		svc_xprt_enqueue(xprt);
1011 	}
1012 	spin_unlock(&serv->sv_lock);
1013 	return ret;
1014 }
1015 
1016 static struct svc_xprt *svc_dequeue_net(struct svc_serv *serv, struct net *net)
1017 {
1018 	struct svc_pool *pool;
1019 	struct svc_xprt *xprt;
1020 	struct svc_xprt *tmp;
1021 	int i;
1022 
1023 	for (i = 0; i < serv->sv_nrpools; i++) {
1024 		pool = &serv->sv_pools[i];
1025 
1026 		spin_lock_bh(&pool->sp_lock);
1027 		list_for_each_entry_safe(xprt, tmp, &pool->sp_sockets, xpt_ready) {
1028 			if (xprt->xpt_net != net)
1029 				continue;
1030 			list_del_init(&xprt->xpt_ready);
1031 			spin_unlock_bh(&pool->sp_lock);
1032 			return xprt;
1033 		}
1034 		spin_unlock_bh(&pool->sp_lock);
1035 	}
1036 	return NULL;
1037 }
1038 
1039 static void svc_clean_up_xprts(struct svc_serv *serv, struct net *net)
1040 {
1041 	struct svc_xprt *xprt;
1042 
1043 	while ((xprt = svc_dequeue_net(serv, net))) {
1044 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
1045 		svc_delete_xprt(xprt);
1046 	}
1047 }
1048 
1049 /*
1050  * Server threads may still be running (especially in the case where the
1051  * service is still running in other network namespaces).
1052  *
1053  * So we shut down sockets the same way we would on a running server, by
1054  * setting XPT_CLOSE, enqueuing, and letting a thread pick it up to do
1055  * the close.  In the case there are no such other threads,
1056  * threads running, svc_clean_up_xprts() does a simple version of a
1057  * server's main event loop, and in the case where there are other
1058  * threads, we may need to wait a little while and then check again to
1059  * see if they're done.
1060  */
1061 void svc_close_net(struct svc_serv *serv, struct net *net)
1062 {
1063 	int delay = 0;
1064 
1065 	while (svc_close_list(serv, &serv->sv_permsocks, net) +
1066 	       svc_close_list(serv, &serv->sv_tempsocks, net)) {
1067 
1068 		svc_clean_up_xprts(serv, net);
1069 		msleep(delay++);
1070 	}
1071 }
1072 
1073 /*
1074  * Handle defer and revisit of requests
1075  */
1076 
1077 static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1078 {
1079 	struct svc_deferred_req *dr =
1080 		container_of(dreq, struct svc_deferred_req, handle);
1081 	struct svc_xprt *xprt = dr->xprt;
1082 
1083 	spin_lock(&xprt->xpt_lock);
1084 	set_bit(XPT_DEFERRED, &xprt->xpt_flags);
1085 	if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) {
1086 		spin_unlock(&xprt->xpt_lock);
1087 		dprintk("revisit canceled\n");
1088 		svc_xprt_put(xprt);
1089 		kfree(dr);
1090 		return;
1091 	}
1092 	dprintk("revisit queued\n");
1093 	dr->xprt = NULL;
1094 	list_add(&dr->handle.recent, &xprt->xpt_deferred);
1095 	spin_unlock(&xprt->xpt_lock);
1096 	svc_xprt_enqueue(xprt);
1097 	svc_xprt_put(xprt);
1098 }
1099 
1100 /*
1101  * Save the request off for later processing. The request buffer looks
1102  * like this:
1103  *
1104  * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
1105  *
1106  * This code can only handle requests that consist of an xprt-header
1107  * and rpc-header.
1108  */
1109 static struct cache_deferred_req *svc_defer(struct cache_req *req)
1110 {
1111 	struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
1112 	struct svc_deferred_req *dr;
1113 
1114 	if (rqstp->rq_arg.page_len || !test_bit(RQ_USEDEFERRAL, &rqstp->rq_flags))
1115 		return NULL; /* if more than a page, give up FIXME */
1116 	if (rqstp->rq_deferred) {
1117 		dr = rqstp->rq_deferred;
1118 		rqstp->rq_deferred = NULL;
1119 	} else {
1120 		size_t skip;
1121 		size_t size;
1122 		/* FIXME maybe discard if size too large */
1123 		size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
1124 		dr = kmalloc(size, GFP_KERNEL);
1125 		if (dr == NULL)
1126 			return NULL;
1127 
1128 		dr->handle.owner = rqstp->rq_server;
1129 		dr->prot = rqstp->rq_prot;
1130 		memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
1131 		dr->addrlen = rqstp->rq_addrlen;
1132 		dr->daddr = rqstp->rq_daddr;
1133 		dr->argslen = rqstp->rq_arg.len >> 2;
1134 		dr->xprt_hlen = rqstp->rq_xprt_hlen;
1135 
1136 		/* back up head to the start of the buffer and copy */
1137 		skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1138 		memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
1139 		       dr->argslen << 2);
1140 	}
1141 	svc_xprt_get(rqstp->rq_xprt);
1142 	dr->xprt = rqstp->rq_xprt;
1143 	set_bit(RQ_DROPME, &rqstp->rq_flags);
1144 
1145 	dr->handle.revisit = svc_revisit;
1146 	return &dr->handle;
1147 }
1148 
1149 /*
1150  * recv data from a deferred request into an active one
1151  */
1152 static int svc_deferred_recv(struct svc_rqst *rqstp)
1153 {
1154 	struct svc_deferred_req *dr = rqstp->rq_deferred;
1155 
1156 	/* setup iov_base past transport header */
1157 	rqstp->rq_arg.head[0].iov_base = dr->args + (dr->xprt_hlen>>2);
1158 	/* The iov_len does not include the transport header bytes */
1159 	rqstp->rq_arg.head[0].iov_len = (dr->argslen<<2) - dr->xprt_hlen;
1160 	rqstp->rq_arg.page_len = 0;
1161 	/* The rq_arg.len includes the transport header bytes */
1162 	rqstp->rq_arg.len     = dr->argslen<<2;
1163 	rqstp->rq_prot        = dr->prot;
1164 	memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
1165 	rqstp->rq_addrlen     = dr->addrlen;
1166 	/* Save off transport header len in case we get deferred again */
1167 	rqstp->rq_xprt_hlen   = dr->xprt_hlen;
1168 	rqstp->rq_daddr       = dr->daddr;
1169 	rqstp->rq_respages    = rqstp->rq_pages;
1170 	return (dr->argslen<<2) - dr->xprt_hlen;
1171 }
1172 
1173 
1174 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt)
1175 {
1176 	struct svc_deferred_req *dr = NULL;
1177 
1178 	if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags))
1179 		return NULL;
1180 	spin_lock(&xprt->xpt_lock);
1181 	if (!list_empty(&xprt->xpt_deferred)) {
1182 		dr = list_entry(xprt->xpt_deferred.next,
1183 				struct svc_deferred_req,
1184 				handle.recent);
1185 		list_del_init(&dr->handle.recent);
1186 	} else
1187 		clear_bit(XPT_DEFERRED, &xprt->xpt_flags);
1188 	spin_unlock(&xprt->xpt_lock);
1189 	return dr;
1190 }
1191 
1192 /**
1193  * svc_find_xprt - find an RPC transport instance
1194  * @serv: pointer to svc_serv to search
1195  * @xcl_name: C string containing transport's class name
1196  * @net: owner net pointer
1197  * @af: Address family of transport's local address
1198  * @port: transport's IP port number
1199  *
1200  * Return the transport instance pointer for the endpoint accepting
1201  * connections/peer traffic from the specified transport class,
1202  * address family and port.
1203  *
1204  * Specifying 0 for the address family or port is effectively a
1205  * wild-card, and will result in matching the first transport in the
1206  * service's list that has a matching class name.
1207  */
1208 struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name,
1209 			       struct net *net, const sa_family_t af,
1210 			       const unsigned short port)
1211 {
1212 	struct svc_xprt *xprt;
1213 	struct svc_xprt *found = NULL;
1214 
1215 	/* Sanity check the args */
1216 	if (serv == NULL || xcl_name == NULL)
1217 		return found;
1218 
1219 	spin_lock_bh(&serv->sv_lock);
1220 	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1221 		if (xprt->xpt_net != net)
1222 			continue;
1223 		if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1224 			continue;
1225 		if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family)
1226 			continue;
1227 		if (port != 0 && port != svc_xprt_local_port(xprt))
1228 			continue;
1229 		found = xprt;
1230 		svc_xprt_get(xprt);
1231 		break;
1232 	}
1233 	spin_unlock_bh(&serv->sv_lock);
1234 	return found;
1235 }
1236 EXPORT_SYMBOL_GPL(svc_find_xprt);
1237 
1238 static int svc_one_xprt_name(const struct svc_xprt *xprt,
1239 			     char *pos, int remaining)
1240 {
1241 	int len;
1242 
1243 	len = snprintf(pos, remaining, "%s %u\n",
1244 			xprt->xpt_class->xcl_name,
1245 			svc_xprt_local_port(xprt));
1246 	if (len >= remaining)
1247 		return -ENAMETOOLONG;
1248 	return len;
1249 }
1250 
1251 /**
1252  * svc_xprt_names - format a buffer with a list of transport names
1253  * @serv: pointer to an RPC service
1254  * @buf: pointer to a buffer to be filled in
1255  * @buflen: length of buffer to be filled in
1256  *
1257  * Fills in @buf with a string containing a list of transport names,
1258  * each name terminated with '\n'.
1259  *
1260  * Returns positive length of the filled-in string on success; otherwise
1261  * a negative errno value is returned if an error occurs.
1262  */
1263 int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen)
1264 {
1265 	struct svc_xprt *xprt;
1266 	int len, totlen;
1267 	char *pos;
1268 
1269 	/* Sanity check args */
1270 	if (!serv)
1271 		return 0;
1272 
1273 	spin_lock_bh(&serv->sv_lock);
1274 
1275 	pos = buf;
1276 	totlen = 0;
1277 	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1278 		len = svc_one_xprt_name(xprt, pos, buflen - totlen);
1279 		if (len < 0) {
1280 			*buf = '\0';
1281 			totlen = len;
1282 		}
1283 		if (len <= 0)
1284 			break;
1285 
1286 		pos += len;
1287 		totlen += len;
1288 	}
1289 
1290 	spin_unlock_bh(&serv->sv_lock);
1291 	return totlen;
1292 }
1293 EXPORT_SYMBOL_GPL(svc_xprt_names);
1294 
1295 
1296 /*----------------------------------------------------------------------------*/
1297 
1298 static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos)
1299 {
1300 	unsigned int pidx = (unsigned int)*pos;
1301 	struct svc_serv *serv = m->private;
1302 
1303 	dprintk("svc_pool_stats_start, *pidx=%u\n", pidx);
1304 
1305 	if (!pidx)
1306 		return SEQ_START_TOKEN;
1307 	return (pidx > serv->sv_nrpools ? NULL : &serv->sv_pools[pidx-1]);
1308 }
1309 
1310 static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos)
1311 {
1312 	struct svc_pool *pool = p;
1313 	struct svc_serv *serv = m->private;
1314 
1315 	dprintk("svc_pool_stats_next, *pos=%llu\n", *pos);
1316 
1317 	if (p == SEQ_START_TOKEN) {
1318 		pool = &serv->sv_pools[0];
1319 	} else {
1320 		unsigned int pidx = (pool - &serv->sv_pools[0]);
1321 		if (pidx < serv->sv_nrpools-1)
1322 			pool = &serv->sv_pools[pidx+1];
1323 		else
1324 			pool = NULL;
1325 	}
1326 	++*pos;
1327 	return pool;
1328 }
1329 
1330 static void svc_pool_stats_stop(struct seq_file *m, void *p)
1331 {
1332 }
1333 
1334 static int svc_pool_stats_show(struct seq_file *m, void *p)
1335 {
1336 	struct svc_pool *pool = p;
1337 
1338 	if (p == SEQ_START_TOKEN) {
1339 		seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n");
1340 		return 0;
1341 	}
1342 
1343 	seq_printf(m, "%u %lu %lu %lu %lu\n",
1344 		pool->sp_id,
1345 		(unsigned long)atomic_long_read(&pool->sp_stats.packets),
1346 		pool->sp_stats.sockets_queued,
1347 		(unsigned long)atomic_long_read(&pool->sp_stats.threads_woken),
1348 		(unsigned long)atomic_long_read(&pool->sp_stats.threads_timedout));
1349 
1350 	return 0;
1351 }
1352 
1353 static const struct seq_operations svc_pool_stats_seq_ops = {
1354 	.start	= svc_pool_stats_start,
1355 	.next	= svc_pool_stats_next,
1356 	.stop	= svc_pool_stats_stop,
1357 	.show	= svc_pool_stats_show,
1358 };
1359 
1360 int svc_pool_stats_open(struct svc_serv *serv, struct file *file)
1361 {
1362 	int err;
1363 
1364 	err = seq_open(file, &svc_pool_stats_seq_ops);
1365 	if (!err)
1366 		((struct seq_file *) file->private_data)->private = serv;
1367 	return err;
1368 }
1369 EXPORT_SYMBOL(svc_pool_stats_open);
1370 
1371 /*----------------------------------------------------------------------------*/
1372