• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 <net/sock.h>
12 #include <linux/sunrpc/stats.h>
13 #include <linux/sunrpc/svc_xprt.h>
14 
15 #define RPCDBG_FACILITY	RPCDBG_SVCXPRT
16 
17 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt);
18 static int svc_deferred_recv(struct svc_rqst *rqstp);
19 static struct cache_deferred_req *svc_defer(struct cache_req *req);
20 static void svc_age_temp_xprts(unsigned long closure);
21 
22 /* apparently the "standard" is that clients close
23  * idle connections after 5 minutes, servers after
24  * 6 minutes
25  *   http://www.connectathon.org/talks96/nfstcp.pdf
26  */
27 static int svc_conn_age_period = 6*60;
28 
29 /* List of registered transport classes */
30 static DEFINE_SPINLOCK(svc_xprt_class_lock);
31 static LIST_HEAD(svc_xprt_class_list);
32 
33 /* SMP locking strategy:
34  *
35  *	svc_pool->sp_lock protects most of the fields of that pool.
36  *	svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
37  *	when both need to be taken (rare), svc_serv->sv_lock is first.
38  *	BKL protects svc_serv->sv_nrthread.
39  *	svc_sock->sk_lock protects the svc_sock->sk_deferred list
40  *             and the ->sk_info_authunix cache.
41  *
42  *	The XPT_BUSY bit in xprt->xpt_flags prevents a transport being
43  *	enqueued multiply. During normal transport processing this bit
44  *	is set by svc_xprt_enqueue and cleared by svc_xprt_received.
45  *	Providers should not manipulate this bit directly.
46  *
47  *	Some flags can be set to certain values at any time
48  *	providing that certain rules are followed:
49  *
50  *	XPT_CONN, XPT_DATA:
51  *		- Can be set or cleared at any time.
52  *		- After a set, svc_xprt_enqueue must be called to enqueue
53  *		  the transport for processing.
54  *		- After a clear, the transport must be read/accepted.
55  *		  If this succeeds, it must be set again.
56  *	XPT_CLOSE:
57  *		- Can set at any time. It is never cleared.
58  *      XPT_DEAD:
59  *		- Can only be set while XPT_BUSY is held which ensures
60  *		  that no other thread will be using the transport or will
61  *		  try to set XPT_DEAD.
62  */
63 
svc_reg_xprt_class(struct svc_xprt_class * xcl)64 int svc_reg_xprt_class(struct svc_xprt_class *xcl)
65 {
66 	struct svc_xprt_class *cl;
67 	int res = -EEXIST;
68 
69 	dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name);
70 
71 	INIT_LIST_HEAD(&xcl->xcl_list);
72 	spin_lock(&svc_xprt_class_lock);
73 	/* Make sure there isn't already a class with the same name */
74 	list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) {
75 		if (strcmp(xcl->xcl_name, cl->xcl_name) == 0)
76 			goto out;
77 	}
78 	list_add_tail(&xcl->xcl_list, &svc_xprt_class_list);
79 	res = 0;
80 out:
81 	spin_unlock(&svc_xprt_class_lock);
82 	return res;
83 }
84 EXPORT_SYMBOL_GPL(svc_reg_xprt_class);
85 
svc_unreg_xprt_class(struct svc_xprt_class * xcl)86 void svc_unreg_xprt_class(struct svc_xprt_class *xcl)
87 {
88 	dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name);
89 	spin_lock(&svc_xprt_class_lock);
90 	list_del_init(&xcl->xcl_list);
91 	spin_unlock(&svc_xprt_class_lock);
92 }
93 EXPORT_SYMBOL_GPL(svc_unreg_xprt_class);
94 
95 /*
96  * Format the transport list for printing
97  */
svc_print_xprts(char * buf,int maxlen)98 int svc_print_xprts(char *buf, int maxlen)
99 {
100 	struct list_head *le;
101 	char tmpstr[80];
102 	int len = 0;
103 	buf[0] = '\0';
104 
105 	spin_lock(&svc_xprt_class_lock);
106 	list_for_each(le, &svc_xprt_class_list) {
107 		int slen;
108 		struct svc_xprt_class *xcl =
109 			list_entry(le, struct svc_xprt_class, xcl_list);
110 
111 		sprintf(tmpstr, "%s %d\n", xcl->xcl_name, xcl->xcl_max_payload);
112 		slen = strlen(tmpstr);
113 		if (len + slen > maxlen)
114 			break;
115 		len += slen;
116 		strcat(buf, tmpstr);
117 	}
118 	spin_unlock(&svc_xprt_class_lock);
119 
120 	return len;
121 }
122 
svc_xprt_free(struct kref * kref)123 static void svc_xprt_free(struct kref *kref)
124 {
125 	struct svc_xprt *xprt =
126 		container_of(kref, struct svc_xprt, xpt_ref);
127 	struct module *owner = xprt->xpt_class->xcl_owner;
128 	if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags)
129 	    && xprt->xpt_auth_cache != NULL)
130 		svcauth_unix_info_release(xprt->xpt_auth_cache);
131 	xprt->xpt_ops->xpo_free(xprt);
132 	module_put(owner);
133 }
134 
svc_xprt_put(struct svc_xprt * xprt)135 void svc_xprt_put(struct svc_xprt *xprt)
136 {
137 	kref_put(&xprt->xpt_ref, svc_xprt_free);
138 }
139 EXPORT_SYMBOL_GPL(svc_xprt_put);
140 
141 /*
142  * Called by transport drivers to initialize the transport independent
143  * portion of the transport instance.
144  */
svc_xprt_init(struct svc_xprt_class * xcl,struct svc_xprt * xprt,struct svc_serv * serv)145 void svc_xprt_init(struct svc_xprt_class *xcl, struct svc_xprt *xprt,
146 		   struct svc_serv *serv)
147 {
148 	memset(xprt, 0, sizeof(*xprt));
149 	xprt->xpt_class = xcl;
150 	xprt->xpt_ops = xcl->xcl_ops;
151 	kref_init(&xprt->xpt_ref);
152 	xprt->xpt_server = serv;
153 	INIT_LIST_HEAD(&xprt->xpt_list);
154 	INIT_LIST_HEAD(&xprt->xpt_ready);
155 	INIT_LIST_HEAD(&xprt->xpt_deferred);
156 	mutex_init(&xprt->xpt_mutex);
157 	spin_lock_init(&xprt->xpt_lock);
158 	set_bit(XPT_BUSY, &xprt->xpt_flags);
159 }
160 EXPORT_SYMBOL_GPL(svc_xprt_init);
161 
__svc_xpo_create(struct svc_xprt_class * xcl,struct svc_serv * serv,unsigned short port,int flags)162 static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl,
163 					 struct svc_serv *serv,
164 					 unsigned short port, int flags)
165 {
166 	struct sockaddr_in sin = {
167 		.sin_family		= AF_INET,
168 		.sin_addr.s_addr	= htonl(INADDR_ANY),
169 		.sin_port		= htons(port),
170 	};
171 	struct sockaddr_in6 sin6 = {
172 		.sin6_family		= AF_INET6,
173 		.sin6_addr		= IN6ADDR_ANY_INIT,
174 		.sin6_port		= htons(port),
175 	};
176 	struct sockaddr *sap;
177 	size_t len;
178 
179 	switch (serv->sv_family) {
180 	case AF_INET:
181 		sap = (struct sockaddr *)&sin;
182 		len = sizeof(sin);
183 		break;
184 	case AF_INET6:
185 		sap = (struct sockaddr *)&sin6;
186 		len = sizeof(sin6);
187 		break;
188 	default:
189 		return ERR_PTR(-EAFNOSUPPORT);
190 	}
191 
192 	return xcl->xcl_ops->xpo_create(serv, sap, len, flags);
193 }
194 
svc_create_xprt(struct svc_serv * serv,char * xprt_name,unsigned short port,int flags)195 int svc_create_xprt(struct svc_serv *serv, char *xprt_name, unsigned short port,
196 		    int flags)
197 {
198 	struct svc_xprt_class *xcl;
199 
200 	dprintk("svc: creating transport %s[%d]\n", xprt_name, port);
201 	spin_lock(&svc_xprt_class_lock);
202 	list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
203 		struct svc_xprt *newxprt;
204 
205 		if (strcmp(xprt_name, xcl->xcl_name))
206 			continue;
207 
208 		if (!try_module_get(xcl->xcl_owner))
209 			goto err;
210 
211 		spin_unlock(&svc_xprt_class_lock);
212 		newxprt = __svc_xpo_create(xcl, serv, port, flags);
213 		if (IS_ERR(newxprt)) {
214 			module_put(xcl->xcl_owner);
215 			return PTR_ERR(newxprt);
216 		}
217 
218 		clear_bit(XPT_TEMP, &newxprt->xpt_flags);
219 		spin_lock_bh(&serv->sv_lock);
220 		list_add(&newxprt->xpt_list, &serv->sv_permsocks);
221 		spin_unlock_bh(&serv->sv_lock);
222 		clear_bit(XPT_BUSY, &newxprt->xpt_flags);
223 		return svc_xprt_local_port(newxprt);
224 	}
225  err:
226 	spin_unlock(&svc_xprt_class_lock);
227 	dprintk("svc: transport %s not found\n", xprt_name);
228 	return -ENOENT;
229 }
230 EXPORT_SYMBOL_GPL(svc_create_xprt);
231 
232 /*
233  * Copy the local and remote xprt addresses to the rqstp structure
234  */
svc_xprt_copy_addrs(struct svc_rqst * rqstp,struct svc_xprt * xprt)235 void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt)
236 {
237 	struct sockaddr *sin;
238 
239 	memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen);
240 	rqstp->rq_addrlen = xprt->xpt_remotelen;
241 
242 	/*
243 	 * Destination address in request is needed for binding the
244 	 * source address in RPC replies/callbacks later.
245 	 */
246 	sin = (struct sockaddr *)&xprt->xpt_local;
247 	switch (sin->sa_family) {
248 	case AF_INET:
249 		rqstp->rq_daddr.addr = ((struct sockaddr_in *)sin)->sin_addr;
250 		break;
251 	case AF_INET6:
252 		rqstp->rq_daddr.addr6 = ((struct sockaddr_in6 *)sin)->sin6_addr;
253 		break;
254 	}
255 }
256 EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs);
257 
258 /**
259  * svc_print_addr - Format rq_addr field for printing
260  * @rqstp: svc_rqst struct containing address to print
261  * @buf: target buffer for formatted address
262  * @len: length of target buffer
263  *
264  */
svc_print_addr(struct svc_rqst * rqstp,char * buf,size_t len)265 char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
266 {
267 	return __svc_print_addr(svc_addr(rqstp), buf, len);
268 }
269 EXPORT_SYMBOL_GPL(svc_print_addr);
270 
271 /*
272  * Queue up an idle server thread.  Must have pool->sp_lock held.
273  * Note: this is really a stack rather than a queue, so that we only
274  * use as many different threads as we need, and the rest don't pollute
275  * the cache.
276  */
svc_thread_enqueue(struct svc_pool * pool,struct svc_rqst * rqstp)277 static void svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp)
278 {
279 	list_add(&rqstp->rq_list, &pool->sp_threads);
280 }
281 
282 /*
283  * Dequeue an nfsd thread.  Must have pool->sp_lock held.
284  */
svc_thread_dequeue(struct svc_pool * pool,struct svc_rqst * rqstp)285 static void svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp)
286 {
287 	list_del(&rqstp->rq_list);
288 }
289 
290 /*
291  * Queue up a transport with data pending. If there are idle nfsd
292  * processes, wake 'em up.
293  *
294  */
svc_xprt_enqueue(struct svc_xprt * xprt)295 void svc_xprt_enqueue(struct svc_xprt *xprt)
296 {
297 	struct svc_serv	*serv = xprt->xpt_server;
298 	struct svc_pool *pool;
299 	struct svc_rqst	*rqstp;
300 	int cpu;
301 
302 	if (!(xprt->xpt_flags &
303 	      ((1<<XPT_CONN)|(1<<XPT_DATA)|(1<<XPT_CLOSE)|(1<<XPT_DEFERRED))))
304 		return;
305 
306 	cpu = get_cpu();
307 	pool = svc_pool_for_cpu(xprt->xpt_server, cpu);
308 	put_cpu();
309 
310 	spin_lock_bh(&pool->sp_lock);
311 
312 	if (!list_empty(&pool->sp_threads) &&
313 	    !list_empty(&pool->sp_sockets))
314 		printk(KERN_ERR
315 		       "svc_xprt_enqueue: "
316 		       "threads and transports both waiting??\n");
317 
318 	if (test_bit(XPT_DEAD, &xprt->xpt_flags)) {
319 		/* Don't enqueue dead transports */
320 		dprintk("svc: transport %p is dead, not enqueued\n", xprt);
321 		goto out_unlock;
322 	}
323 
324 	/* Mark transport as busy. It will remain in this state until
325 	 * the provider calls svc_xprt_received. We update XPT_BUSY
326 	 * atomically because it also guards against trying to enqueue
327 	 * the transport twice.
328 	 */
329 	if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) {
330 		/* Don't enqueue transport while already enqueued */
331 		dprintk("svc: transport %p busy, not enqueued\n", xprt);
332 		goto out_unlock;
333 	}
334 	BUG_ON(xprt->xpt_pool != NULL);
335 	xprt->xpt_pool = pool;
336 
337 	/* Handle pending connection */
338 	if (test_bit(XPT_CONN, &xprt->xpt_flags))
339 		goto process;
340 
341 	/* Handle close in-progress */
342 	if (test_bit(XPT_CLOSE, &xprt->xpt_flags))
343 		goto process;
344 
345 	/* Check if we have space to reply to a request */
346 	if (!xprt->xpt_ops->xpo_has_wspace(xprt)) {
347 		/* Don't enqueue while not enough space for reply */
348 		dprintk("svc: no write space, transport %p  not enqueued\n",
349 			xprt);
350 		xprt->xpt_pool = NULL;
351 		clear_bit(XPT_BUSY, &xprt->xpt_flags);
352 		goto out_unlock;
353 	}
354 
355  process:
356 	if (!list_empty(&pool->sp_threads)) {
357 		rqstp = list_entry(pool->sp_threads.next,
358 				   struct svc_rqst,
359 				   rq_list);
360 		dprintk("svc: transport %p served by daemon %p\n",
361 			xprt, rqstp);
362 		svc_thread_dequeue(pool, rqstp);
363 		if (rqstp->rq_xprt)
364 			printk(KERN_ERR
365 				"svc_xprt_enqueue: server %p, rq_xprt=%p!\n",
366 				rqstp, rqstp->rq_xprt);
367 		rqstp->rq_xprt = xprt;
368 		svc_xprt_get(xprt);
369 		rqstp->rq_reserved = serv->sv_max_mesg;
370 		atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
371 		BUG_ON(xprt->xpt_pool != pool);
372 		wake_up(&rqstp->rq_wait);
373 	} else {
374 		dprintk("svc: transport %p put into queue\n", xprt);
375 		list_add_tail(&xprt->xpt_ready, &pool->sp_sockets);
376 		BUG_ON(xprt->xpt_pool != pool);
377 	}
378 
379 out_unlock:
380 	spin_unlock_bh(&pool->sp_lock);
381 }
382 EXPORT_SYMBOL_GPL(svc_xprt_enqueue);
383 
384 /*
385  * Dequeue the first transport.  Must be called with the pool->sp_lock held.
386  */
svc_xprt_dequeue(struct svc_pool * pool)387 static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool)
388 {
389 	struct svc_xprt	*xprt;
390 
391 	if (list_empty(&pool->sp_sockets))
392 		return NULL;
393 
394 	xprt = list_entry(pool->sp_sockets.next,
395 			  struct svc_xprt, xpt_ready);
396 	list_del_init(&xprt->xpt_ready);
397 
398 	dprintk("svc: transport %p dequeued, inuse=%d\n",
399 		xprt, atomic_read(&xprt->xpt_ref.refcount));
400 
401 	return xprt;
402 }
403 
404 /*
405  * svc_xprt_received conditionally queues the transport for processing
406  * by another thread. The caller must hold the XPT_BUSY bit and must
407  * not thereafter touch transport data.
408  *
409  * Note: XPT_DATA only gets cleared when a read-attempt finds no (or
410  * insufficient) data.
411  */
svc_xprt_received(struct svc_xprt * xprt)412 void svc_xprt_received(struct svc_xprt *xprt)
413 {
414 	BUG_ON(!test_bit(XPT_BUSY, &xprt->xpt_flags));
415 	xprt->xpt_pool = NULL;
416 	clear_bit(XPT_BUSY, &xprt->xpt_flags);
417 	svc_xprt_enqueue(xprt);
418 }
419 EXPORT_SYMBOL_GPL(svc_xprt_received);
420 
421 /**
422  * svc_reserve - change the space reserved for the reply to a request.
423  * @rqstp:  The request in question
424  * @space: new max space to reserve
425  *
426  * Each request reserves some space on the output queue of the transport
427  * to make sure the reply fits.  This function reduces that reserved
428  * space to be the amount of space used already, plus @space.
429  *
430  */
svc_reserve(struct svc_rqst * rqstp,int space)431 void svc_reserve(struct svc_rqst *rqstp, int space)
432 {
433 	space += rqstp->rq_res.head[0].iov_len;
434 
435 	if (space < rqstp->rq_reserved) {
436 		struct svc_xprt *xprt = rqstp->rq_xprt;
437 		atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved);
438 		rqstp->rq_reserved = space;
439 
440 		svc_xprt_enqueue(xprt);
441 	}
442 }
443 EXPORT_SYMBOL_GPL(svc_reserve);
444 
svc_xprt_release(struct svc_rqst * rqstp)445 static void svc_xprt_release(struct svc_rqst *rqstp)
446 {
447 	struct svc_xprt	*xprt = rqstp->rq_xprt;
448 
449 	rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
450 
451 	kfree(rqstp->rq_deferred);
452 	rqstp->rq_deferred = NULL;
453 
454 	svc_free_res_pages(rqstp);
455 	rqstp->rq_res.page_len = 0;
456 	rqstp->rq_res.page_base = 0;
457 
458 	/* Reset response buffer and release
459 	 * the reservation.
460 	 * But first, check that enough space was reserved
461 	 * for the reply, otherwise we have a bug!
462 	 */
463 	if ((rqstp->rq_res.len) >  rqstp->rq_reserved)
464 		printk(KERN_ERR "RPC request reserved %d but used %d\n",
465 		       rqstp->rq_reserved,
466 		       rqstp->rq_res.len);
467 
468 	rqstp->rq_res.head[0].iov_len = 0;
469 	svc_reserve(rqstp, 0);
470 	rqstp->rq_xprt = NULL;
471 
472 	svc_xprt_put(xprt);
473 }
474 
475 /*
476  * External function to wake up a server waiting for data
477  * This really only makes sense for services like lockd
478  * which have exactly one thread anyway.
479  */
svc_wake_up(struct svc_serv * serv)480 void svc_wake_up(struct svc_serv *serv)
481 {
482 	struct svc_rqst	*rqstp;
483 	unsigned int i;
484 	struct svc_pool *pool;
485 
486 	for (i = 0; i < serv->sv_nrpools; i++) {
487 		pool = &serv->sv_pools[i];
488 
489 		spin_lock_bh(&pool->sp_lock);
490 		if (!list_empty(&pool->sp_threads)) {
491 			rqstp = list_entry(pool->sp_threads.next,
492 					   struct svc_rqst,
493 					   rq_list);
494 			dprintk("svc: daemon %p woken up.\n", rqstp);
495 			/*
496 			svc_thread_dequeue(pool, rqstp);
497 			rqstp->rq_xprt = NULL;
498 			 */
499 			wake_up(&rqstp->rq_wait);
500 		}
501 		spin_unlock_bh(&pool->sp_lock);
502 	}
503 }
504 EXPORT_SYMBOL_GPL(svc_wake_up);
505 
svc_port_is_privileged(struct sockaddr * sin)506 int svc_port_is_privileged(struct sockaddr *sin)
507 {
508 	switch (sin->sa_family) {
509 	case AF_INET:
510 		return ntohs(((struct sockaddr_in *)sin)->sin_port)
511 			< PROT_SOCK;
512 	case AF_INET6:
513 		return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
514 			< PROT_SOCK;
515 	default:
516 		return 0;
517 	}
518 }
519 
520 /*
521  * Make sure that we don't have too many active connections. If we have,
522  * something must be dropped. It's not clear what will happen if we allow
523  * "too many" connections, but when dealing with network-facing software,
524  * we have to code defensively. Here we do that by imposing hard limits.
525  *
526  * There's no point in trying to do random drop here for DoS
527  * prevention. The NFS clients does 1 reconnect in 15 seconds. An
528  * attacker can easily beat that.
529  *
530  * The only somewhat efficient mechanism would be if drop old
531  * connections from the same IP first. But right now we don't even
532  * record the client IP in svc_sock.
533  *
534  * single-threaded services that expect a lot of clients will probably
535  * need to set sv_maxconn to override the default value which is based
536  * on the number of threads
537  */
svc_check_conn_limits(struct svc_serv * serv)538 static void svc_check_conn_limits(struct svc_serv *serv)
539 {
540 	unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn :
541 				(serv->sv_nrthreads+3) * 20;
542 
543 	if (serv->sv_tmpcnt > limit) {
544 		struct svc_xprt *xprt = NULL;
545 		spin_lock_bh(&serv->sv_lock);
546 		if (!list_empty(&serv->sv_tempsocks)) {
547 			if (net_ratelimit()) {
548 				/* Try to help the admin */
549 				printk(KERN_NOTICE "%s: too many open  "
550 				       "connections, consider increasing %s\n",
551 				       serv->sv_name, serv->sv_maxconn ?
552 				       "the max number of connections." :
553 				       "the number of threads.");
554 			}
555 			/*
556 			 * Always select the oldest connection. It's not fair,
557 			 * but so is life
558 			 */
559 			xprt = list_entry(serv->sv_tempsocks.prev,
560 					  struct svc_xprt,
561 					  xpt_list);
562 			set_bit(XPT_CLOSE, &xprt->xpt_flags);
563 			svc_xprt_get(xprt);
564 		}
565 		spin_unlock_bh(&serv->sv_lock);
566 
567 		if (xprt) {
568 			svc_xprt_enqueue(xprt);
569 			svc_xprt_put(xprt);
570 		}
571 	}
572 }
573 
574 /*
575  * Receive the next request on any transport.  This code is carefully
576  * organised not to touch any cachelines in the shared svc_serv
577  * structure, only cachelines in the local svc_pool.
578  */
svc_recv(struct svc_rqst * rqstp,long timeout)579 int svc_recv(struct svc_rqst *rqstp, long timeout)
580 {
581 	struct svc_xprt		*xprt = NULL;
582 	struct svc_serv		*serv = rqstp->rq_server;
583 	struct svc_pool		*pool = rqstp->rq_pool;
584 	int			len, i;
585 	int			pages;
586 	struct xdr_buf		*arg;
587 	DECLARE_WAITQUEUE(wait, current);
588 
589 	dprintk("svc: server %p waiting for data (to = %ld)\n",
590 		rqstp, timeout);
591 
592 	if (rqstp->rq_xprt)
593 		printk(KERN_ERR
594 			"svc_recv: service %p, transport not NULL!\n",
595 			 rqstp);
596 	if (waitqueue_active(&rqstp->rq_wait))
597 		printk(KERN_ERR
598 			"svc_recv: service %p, wait queue active!\n",
599 			 rqstp);
600 
601 	/* now allocate needed pages.  If we get a failure, sleep briefly */
602 	pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE;
603 	for (i = 0; i < pages ; i++)
604 		while (rqstp->rq_pages[i] == NULL) {
605 			struct page *p = alloc_page(GFP_KERNEL);
606 			if (!p) {
607 				set_current_state(TASK_INTERRUPTIBLE);
608 				if (signalled() || kthread_should_stop()) {
609 					set_current_state(TASK_RUNNING);
610 					return -EINTR;
611 				}
612 				schedule_timeout(msecs_to_jiffies(500));
613 			}
614 			rqstp->rq_pages[i] = p;
615 		}
616 	rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */
617 	BUG_ON(pages >= RPCSVC_MAXPAGES);
618 
619 	/* Make arg->head point to first page and arg->pages point to rest */
620 	arg = &rqstp->rq_arg;
621 	arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
622 	arg->head[0].iov_len = PAGE_SIZE;
623 	arg->pages = rqstp->rq_pages + 1;
624 	arg->page_base = 0;
625 	/* save at least one page for response */
626 	arg->page_len = (pages-2)*PAGE_SIZE;
627 	arg->len = (pages-1)*PAGE_SIZE;
628 	arg->tail[0].iov_len = 0;
629 
630 	try_to_freeze();
631 	cond_resched();
632 	if (signalled() || kthread_should_stop())
633 		return -EINTR;
634 
635 	spin_lock_bh(&pool->sp_lock);
636 	xprt = svc_xprt_dequeue(pool);
637 	if (xprt) {
638 		rqstp->rq_xprt = xprt;
639 		svc_xprt_get(xprt);
640 		rqstp->rq_reserved = serv->sv_max_mesg;
641 		atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
642 	} else {
643 		/* No data pending. Go to sleep */
644 		svc_thread_enqueue(pool, rqstp);
645 
646 		/*
647 		 * We have to be able to interrupt this wait
648 		 * to bring down the daemons ...
649 		 */
650 		set_current_state(TASK_INTERRUPTIBLE);
651 
652 		/*
653 		 * checking kthread_should_stop() here allows us to avoid
654 		 * locking and signalling when stopping kthreads that call
655 		 * svc_recv. If the thread has already been woken up, then
656 		 * we can exit here without sleeping. If not, then it
657 		 * it'll be woken up quickly during the schedule_timeout
658 		 */
659 		if (kthread_should_stop()) {
660 			set_current_state(TASK_RUNNING);
661 			spin_unlock_bh(&pool->sp_lock);
662 			return -EINTR;
663 		}
664 
665 		add_wait_queue(&rqstp->rq_wait, &wait);
666 		spin_unlock_bh(&pool->sp_lock);
667 
668 		schedule_timeout(timeout);
669 
670 		try_to_freeze();
671 
672 		spin_lock_bh(&pool->sp_lock);
673 		remove_wait_queue(&rqstp->rq_wait, &wait);
674 
675 		xprt = rqstp->rq_xprt;
676 		if (!xprt) {
677 			svc_thread_dequeue(pool, rqstp);
678 			spin_unlock_bh(&pool->sp_lock);
679 			dprintk("svc: server %p, no data yet\n", rqstp);
680 			if (signalled() || kthread_should_stop())
681 				return -EINTR;
682 			else
683 				return -EAGAIN;
684 		}
685 	}
686 	spin_unlock_bh(&pool->sp_lock);
687 
688 	len = 0;
689 	if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) {
690 		dprintk("svc_recv: found XPT_CLOSE\n");
691 		svc_delete_xprt(xprt);
692 	} else if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
693 		struct svc_xprt *newxpt;
694 		newxpt = xprt->xpt_ops->xpo_accept(xprt);
695 		if (newxpt) {
696 			/*
697 			 * We know this module_get will succeed because the
698 			 * listener holds a reference too
699 			 */
700 			__module_get(newxpt->xpt_class->xcl_owner);
701 			svc_check_conn_limits(xprt->xpt_server);
702 			spin_lock_bh(&serv->sv_lock);
703 			set_bit(XPT_TEMP, &newxpt->xpt_flags);
704 			list_add(&newxpt->xpt_list, &serv->sv_tempsocks);
705 			serv->sv_tmpcnt++;
706 			if (serv->sv_temptimer.function == NULL) {
707 				/* setup timer to age temp transports */
708 				setup_timer(&serv->sv_temptimer,
709 					    svc_age_temp_xprts,
710 					    (unsigned long)serv);
711 				mod_timer(&serv->sv_temptimer,
712 					  jiffies + svc_conn_age_period * HZ);
713 			}
714 			spin_unlock_bh(&serv->sv_lock);
715 			svc_xprt_received(newxpt);
716 		}
717 		svc_xprt_received(xprt);
718 	} else {
719 		dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n",
720 			rqstp, pool->sp_id, xprt,
721 			atomic_read(&xprt->xpt_ref.refcount));
722 		rqstp->rq_deferred = svc_deferred_dequeue(xprt);
723 		if (rqstp->rq_deferred) {
724 			svc_xprt_received(xprt);
725 			len = svc_deferred_recv(rqstp);
726 		} else
727 			len = xprt->xpt_ops->xpo_recvfrom(rqstp);
728 		dprintk("svc: got len=%d\n", len);
729 	}
730 
731 	/* No data, incomplete (TCP) read, or accept() */
732 	if (len == 0 || len == -EAGAIN) {
733 		rqstp->rq_res.len = 0;
734 		svc_xprt_release(rqstp);
735 		return -EAGAIN;
736 	}
737 	clear_bit(XPT_OLD, &xprt->xpt_flags);
738 
739 	rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp));
740 	rqstp->rq_chandle.defer = svc_defer;
741 
742 	if (serv->sv_stats)
743 		serv->sv_stats->netcnt++;
744 	return len;
745 }
746 EXPORT_SYMBOL_GPL(svc_recv);
747 
748 /*
749  * Drop request
750  */
svc_drop(struct svc_rqst * rqstp)751 void svc_drop(struct svc_rqst *rqstp)
752 {
753 	dprintk("svc: xprt %p dropped request\n", rqstp->rq_xprt);
754 	svc_xprt_release(rqstp);
755 }
756 EXPORT_SYMBOL_GPL(svc_drop);
757 
758 /*
759  * Return reply to client.
760  */
svc_send(struct svc_rqst * rqstp)761 int svc_send(struct svc_rqst *rqstp)
762 {
763 	struct svc_xprt	*xprt;
764 	int		len;
765 	struct xdr_buf	*xb;
766 
767 	xprt = rqstp->rq_xprt;
768 	if (!xprt)
769 		return -EFAULT;
770 
771 	/* release the receive skb before sending the reply */
772 	rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
773 
774 	/* calculate over-all length */
775 	xb = &rqstp->rq_res;
776 	xb->len = xb->head[0].iov_len +
777 		xb->page_len +
778 		xb->tail[0].iov_len;
779 
780 	/* Grab mutex to serialize outgoing data. */
781 	mutex_lock(&xprt->xpt_mutex);
782 	if (test_bit(XPT_DEAD, &xprt->xpt_flags))
783 		len = -ENOTCONN;
784 	else
785 		len = xprt->xpt_ops->xpo_sendto(rqstp);
786 	mutex_unlock(&xprt->xpt_mutex);
787 	svc_xprt_release(rqstp);
788 
789 	if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
790 		return 0;
791 	return len;
792 }
793 
794 /*
795  * Timer function to close old temporary transports, using
796  * a mark-and-sweep algorithm.
797  */
svc_age_temp_xprts(unsigned long closure)798 static void svc_age_temp_xprts(unsigned long closure)
799 {
800 	struct svc_serv *serv = (struct svc_serv *)closure;
801 	struct svc_xprt *xprt;
802 	struct list_head *le, *next;
803 	LIST_HEAD(to_be_aged);
804 
805 	dprintk("svc_age_temp_xprts\n");
806 
807 	if (!spin_trylock_bh(&serv->sv_lock)) {
808 		/* busy, try again 1 sec later */
809 		dprintk("svc_age_temp_xprts: busy\n");
810 		mod_timer(&serv->sv_temptimer, jiffies + HZ);
811 		return;
812 	}
813 
814 	list_for_each_safe(le, next, &serv->sv_tempsocks) {
815 		xprt = list_entry(le, struct svc_xprt, xpt_list);
816 
817 		/* First time through, just mark it OLD. Second time
818 		 * through, close it. */
819 		if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags))
820 			continue;
821 		if (atomic_read(&xprt->xpt_ref.refcount) > 1
822 		    || test_bit(XPT_BUSY, &xprt->xpt_flags))
823 			continue;
824 		svc_xprt_get(xprt);
825 		list_move(le, &to_be_aged);
826 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
827 		set_bit(XPT_DETACHED, &xprt->xpt_flags);
828 	}
829 	spin_unlock_bh(&serv->sv_lock);
830 
831 	while (!list_empty(&to_be_aged)) {
832 		le = to_be_aged.next;
833 		/* fiddling the xpt_list node is safe 'cos we're XPT_DETACHED */
834 		list_del_init(le);
835 		xprt = list_entry(le, struct svc_xprt, xpt_list);
836 
837 		dprintk("queuing xprt %p for closing\n", xprt);
838 
839 		/* a thread will dequeue and close it soon */
840 		svc_xprt_enqueue(xprt);
841 		svc_xprt_put(xprt);
842 	}
843 
844 	mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
845 }
846 
847 /*
848  * Remove a dead transport
849  */
svc_delete_xprt(struct svc_xprt * xprt)850 void svc_delete_xprt(struct svc_xprt *xprt)
851 {
852 	struct svc_serv	*serv = xprt->xpt_server;
853 	struct svc_deferred_req *dr;
854 
855 	/* Only do this once */
856 	if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags))
857 		return;
858 
859 	dprintk("svc: svc_delete_xprt(%p)\n", xprt);
860 	xprt->xpt_ops->xpo_detach(xprt);
861 
862 	spin_lock_bh(&serv->sv_lock);
863 	if (!test_and_set_bit(XPT_DETACHED, &xprt->xpt_flags))
864 		list_del_init(&xprt->xpt_list);
865 	/*
866 	 * We used to delete the transport from whichever list
867 	 * it's sk_xprt.xpt_ready node was on, but we don't actually
868 	 * need to.  This is because the only time we're called
869 	 * while still attached to a queue, the queue itself
870 	 * is about to be destroyed (in svc_destroy).
871 	 */
872 	if (test_bit(XPT_TEMP, &xprt->xpt_flags))
873 		serv->sv_tmpcnt--;
874 
875 	for (dr = svc_deferred_dequeue(xprt); dr;
876 	     dr = svc_deferred_dequeue(xprt)) {
877 		svc_xprt_put(xprt);
878 		kfree(dr);
879 	}
880 
881 	svc_xprt_put(xprt);
882 	spin_unlock_bh(&serv->sv_lock);
883 }
884 
svc_close_xprt(struct svc_xprt * xprt)885 void svc_close_xprt(struct svc_xprt *xprt)
886 {
887 	set_bit(XPT_CLOSE, &xprt->xpt_flags);
888 	if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
889 		/* someone else will have to effect the close */
890 		return;
891 
892 	svc_xprt_get(xprt);
893 	svc_delete_xprt(xprt);
894 	clear_bit(XPT_BUSY, &xprt->xpt_flags);
895 	svc_xprt_put(xprt);
896 }
897 EXPORT_SYMBOL_GPL(svc_close_xprt);
898 
svc_close_all(struct list_head * xprt_list)899 void svc_close_all(struct list_head *xprt_list)
900 {
901 	struct svc_xprt *xprt;
902 	struct svc_xprt *tmp;
903 
904 	list_for_each_entry_safe(xprt, tmp, xprt_list, xpt_list) {
905 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
906 		if (test_bit(XPT_BUSY, &xprt->xpt_flags)) {
907 			/* Waiting to be processed, but no threads left,
908 			 * So just remove it from the waiting list
909 			 */
910 			list_del_init(&xprt->xpt_ready);
911 			clear_bit(XPT_BUSY, &xprt->xpt_flags);
912 		}
913 		svc_close_xprt(xprt);
914 	}
915 }
916 
917 /*
918  * Handle defer and revisit of requests
919  */
920 
svc_revisit(struct cache_deferred_req * dreq,int too_many)921 static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
922 {
923 	struct svc_deferred_req *dr =
924 		container_of(dreq, struct svc_deferred_req, handle);
925 	struct svc_xprt *xprt = dr->xprt;
926 
927 	spin_lock(&xprt->xpt_lock);
928 	set_bit(XPT_DEFERRED, &xprt->xpt_flags);
929 	if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) {
930 		spin_unlock(&xprt->xpt_lock);
931 		dprintk("revisit canceled\n");
932 		svc_xprt_put(xprt);
933 		kfree(dr);
934 		return;
935 	}
936 	dprintk("revisit queued\n");
937 	dr->xprt = NULL;
938 	list_add(&dr->handle.recent, &xprt->xpt_deferred);
939 	spin_unlock(&xprt->xpt_lock);
940 	svc_xprt_enqueue(xprt);
941 	svc_xprt_put(xprt);
942 }
943 
944 /*
945  * Save the request off for later processing. The request buffer looks
946  * like this:
947  *
948  * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
949  *
950  * This code can only handle requests that consist of an xprt-header
951  * and rpc-header.
952  */
svc_defer(struct cache_req * req)953 static struct cache_deferred_req *svc_defer(struct cache_req *req)
954 {
955 	struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
956 	struct svc_deferred_req *dr;
957 
958 	if (rqstp->rq_arg.page_len)
959 		return NULL; /* if more than a page, give up FIXME */
960 	if (rqstp->rq_deferred) {
961 		dr = rqstp->rq_deferred;
962 		rqstp->rq_deferred = NULL;
963 	} else {
964 		size_t skip;
965 		size_t size;
966 		/* FIXME maybe discard if size too large */
967 		size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
968 		dr = kmalloc(size, GFP_KERNEL);
969 		if (dr == NULL)
970 			return NULL;
971 
972 		dr->handle.owner = rqstp->rq_server;
973 		dr->prot = rqstp->rq_prot;
974 		memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
975 		dr->addrlen = rqstp->rq_addrlen;
976 		dr->daddr = rqstp->rq_daddr;
977 		dr->argslen = rqstp->rq_arg.len >> 2;
978 		dr->xprt_hlen = rqstp->rq_xprt_hlen;
979 
980 		/* back up head to the start of the buffer and copy */
981 		skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
982 		memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
983 		       dr->argslen << 2);
984 	}
985 	svc_xprt_get(rqstp->rq_xprt);
986 	dr->xprt = rqstp->rq_xprt;
987 
988 	dr->handle.revisit = svc_revisit;
989 	return &dr->handle;
990 }
991 
992 /*
993  * recv data from a deferred request into an active one
994  */
svc_deferred_recv(struct svc_rqst * rqstp)995 static int svc_deferred_recv(struct svc_rqst *rqstp)
996 {
997 	struct svc_deferred_req *dr = rqstp->rq_deferred;
998 
999 	/* setup iov_base past transport header */
1000 	rqstp->rq_arg.head[0].iov_base = dr->args + (dr->xprt_hlen>>2);
1001 	/* The iov_len does not include the transport header bytes */
1002 	rqstp->rq_arg.head[0].iov_len = (dr->argslen<<2) - dr->xprt_hlen;
1003 	rqstp->rq_arg.page_len = 0;
1004 	/* The rq_arg.len includes the transport header bytes */
1005 	rqstp->rq_arg.len     = dr->argslen<<2;
1006 	rqstp->rq_prot        = dr->prot;
1007 	memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
1008 	rqstp->rq_addrlen     = dr->addrlen;
1009 	/* Save off transport header len in case we get deferred again */
1010 	rqstp->rq_xprt_hlen   = dr->xprt_hlen;
1011 	rqstp->rq_daddr       = dr->daddr;
1012 	rqstp->rq_respages    = rqstp->rq_pages;
1013 	return (dr->argslen<<2) - dr->xprt_hlen;
1014 }
1015 
1016 
svc_deferred_dequeue(struct svc_xprt * xprt)1017 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt)
1018 {
1019 	struct svc_deferred_req *dr = NULL;
1020 
1021 	if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags))
1022 		return NULL;
1023 	spin_lock(&xprt->xpt_lock);
1024 	clear_bit(XPT_DEFERRED, &xprt->xpt_flags);
1025 	if (!list_empty(&xprt->xpt_deferred)) {
1026 		dr = list_entry(xprt->xpt_deferred.next,
1027 				struct svc_deferred_req,
1028 				handle.recent);
1029 		list_del_init(&dr->handle.recent);
1030 		set_bit(XPT_DEFERRED, &xprt->xpt_flags);
1031 	}
1032 	spin_unlock(&xprt->xpt_lock);
1033 	return dr;
1034 }
1035 
1036 /*
1037  * Return the transport instance pointer for the endpoint accepting
1038  * connections/peer traffic from the specified transport class,
1039  * address family and port.
1040  *
1041  * Specifying 0 for the address family or port is effectively a
1042  * wild-card, and will result in matching the first transport in the
1043  * service's list that has a matching class name.
1044  */
svc_find_xprt(struct svc_serv * serv,char * xcl_name,int af,int port)1045 struct svc_xprt *svc_find_xprt(struct svc_serv *serv, char *xcl_name,
1046 			       int af, int port)
1047 {
1048 	struct svc_xprt *xprt;
1049 	struct svc_xprt *found = NULL;
1050 
1051 	/* Sanity check the args */
1052 	if (!serv || !xcl_name)
1053 		return found;
1054 
1055 	spin_lock_bh(&serv->sv_lock);
1056 	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1057 		if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1058 			continue;
1059 		if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family)
1060 			continue;
1061 		if (port && port != svc_xprt_local_port(xprt))
1062 			continue;
1063 		found = xprt;
1064 		svc_xprt_get(xprt);
1065 		break;
1066 	}
1067 	spin_unlock_bh(&serv->sv_lock);
1068 	return found;
1069 }
1070 EXPORT_SYMBOL_GPL(svc_find_xprt);
1071 
1072 /*
1073  * Format a buffer with a list of the active transports. A zero for
1074  * the buflen parameter disables target buffer overflow checking.
1075  */
svc_xprt_names(struct svc_serv * serv,char * buf,int buflen)1076 int svc_xprt_names(struct svc_serv *serv, char *buf, int buflen)
1077 {
1078 	struct svc_xprt *xprt;
1079 	char xprt_str[64];
1080 	int totlen = 0;
1081 	int len;
1082 
1083 	/* Sanity check args */
1084 	if (!serv)
1085 		return 0;
1086 
1087 	spin_lock_bh(&serv->sv_lock);
1088 	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1089 		len = snprintf(xprt_str, sizeof(xprt_str),
1090 			       "%s %d\n", xprt->xpt_class->xcl_name,
1091 			       svc_xprt_local_port(xprt));
1092 		/* If the string was truncated, replace with error string */
1093 		if (len >= sizeof(xprt_str))
1094 			strcpy(xprt_str, "name-too-long\n");
1095 		/* Don't overflow buffer */
1096 		len = strlen(xprt_str);
1097 		if (buflen && (len + totlen >= buflen))
1098 			break;
1099 		strcpy(buf+totlen, xprt_str);
1100 		totlen += len;
1101 	}
1102 	spin_unlock_bh(&serv->sv_lock);
1103 	return totlen;
1104 }
1105 EXPORT_SYMBOL_GPL(svc_xprt_names);
1106