xref: /linux/net/sunrpc/svc_xprt.c (revision 46e6acfe3501fa938af9c5bd730f0020235b08a2)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * linux/net/sunrpc/svc_xprt.c
4  *
5  * Author: Tom Tucker <tom@opengridcomputing.com>
6  */
7 
8 #include <linux/sched.h>
9 #include <linux/sched/mm.h>
10 #include <linux/errno.h>
11 #include <linux/freezer.h>
12 #include <linux/slab.h>
13 #include <net/sock.h>
14 #include <linux/sunrpc/addr.h>
15 #include <linux/sunrpc/stats.h>
16 #include <linux/sunrpc/svc_xprt.h>
17 #include <linux/sunrpc/svcsock.h>
18 #include <linux/sunrpc/xprt.h>
19 #include <linux/sunrpc/bc_xprt.h>
20 #include <linux/module.h>
21 #include <linux/netdevice.h>
22 #include <trace/events/sunrpc.h>
23 
24 #define RPCDBG_FACILITY	RPCDBG_SVCXPRT
25 
26 static unsigned int svc_rpc_per_connection_limit __read_mostly;
27 module_param(svc_rpc_per_connection_limit, uint, 0644);
28 
29 
30 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt);
31 static int svc_deferred_recv(struct svc_rqst *rqstp);
32 static struct cache_deferred_req *svc_defer(struct cache_req *req);
33 static void svc_age_temp_xprts(struct timer_list *t);
34 static void svc_delete_xprt(struct svc_xprt *xprt);
35 
36 /* apparently the "standard" is that clients close
37  * idle connections after 5 minutes, servers after
38  * 6 minutes
39  *   http://nfsv4bat.org/Documents/ConnectAThon/1996/nfstcp.pdf
40  */
41 static int svc_conn_age_period = 6*60;
42 
43 /* List of registered transport classes */
44 static DEFINE_SPINLOCK(svc_xprt_class_lock);
45 static LIST_HEAD(svc_xprt_class_list);
46 
47 /* SMP locking strategy:
48  *
49  *	svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
50  *	when both need to be taken (rare), svc_serv->sv_lock is first.
51  *	The "service mutex" protects svc_serv->sv_nrthread.
52  *	svc_sock->sk_lock protects the svc_sock->sk_deferred list
53  *             and the ->sk_info_authunix cache.
54  *
55  *	The XPT_BUSY bit in xprt->xpt_flags prevents a transport being
56  *	enqueued multiply. During normal transport processing this bit
57  *	is set by svc_xprt_enqueue and cleared by svc_xprt_received.
58  *	Providers should not manipulate this bit directly.
59  *
60  *	Some flags can be set to certain values at any time
61  *	providing that certain rules are followed:
62  *
63  *	XPT_CONN, XPT_DATA:
64  *		- Can be set or cleared at any time.
65  *		- After a set, svc_xprt_enqueue must be called to enqueue
66  *		  the transport for processing.
67  *		- After a clear, the transport must be read/accepted.
68  *		  If this succeeds, it must be set again.
69  *	XPT_CLOSE:
70  *		- Can set at any time. It is never cleared.
71  *      XPT_DEAD:
72  *		- Can only be set while XPT_BUSY is held which ensures
73  *		  that no other thread will be using the transport or will
74  *		  try to set XPT_DEAD.
75  */
76 
77 /**
78  * svc_reg_xprt_class - Register a server-side RPC transport class
79  * @xcl: New transport class to be registered
80  *
81  * Returns zero on success; otherwise a negative errno is returned.
82  */
83 int svc_reg_xprt_class(struct svc_xprt_class *xcl)
84 {
85 	struct svc_xprt_class *cl;
86 	int res = -EEXIST;
87 
88 	INIT_LIST_HEAD(&xcl->xcl_list);
89 	spin_lock(&svc_xprt_class_lock);
90 	/* Make sure there isn't already a class with the same name */
91 	list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) {
92 		if (strcmp(xcl->xcl_name, cl->xcl_name) == 0)
93 			goto out;
94 	}
95 	list_add_tail(&xcl->xcl_list, &svc_xprt_class_list);
96 	res = 0;
97 out:
98 	spin_unlock(&svc_xprt_class_lock);
99 	return res;
100 }
101 EXPORT_SYMBOL_GPL(svc_reg_xprt_class);
102 
103 /**
104  * svc_unreg_xprt_class - Unregister a server-side RPC transport class
105  * @xcl: Transport class to be unregistered
106  *
107  */
108 void svc_unreg_xprt_class(struct svc_xprt_class *xcl)
109 {
110 	spin_lock(&svc_xprt_class_lock);
111 	list_del_init(&xcl->xcl_list);
112 	spin_unlock(&svc_xprt_class_lock);
113 }
114 EXPORT_SYMBOL_GPL(svc_unreg_xprt_class);
115 
116 /**
117  * svc_print_xprts - Format the transport list for printing
118  * @buf: target buffer for formatted address
119  * @maxlen: length of target buffer
120  *
121  * Fills in @buf with a string containing a list of transport names, each name
122  * terminated with '\n'. If the buffer is too small, some entries may be
123  * missing, but it is guaranteed that all lines in the output buffer are
124  * complete.
125  *
126  * Returns positive length of the filled-in string.
127  */
128 int svc_print_xprts(char *buf, int maxlen)
129 {
130 	struct svc_xprt_class *xcl;
131 	char tmpstr[80];
132 	int len = 0;
133 	buf[0] = '\0';
134 
135 	spin_lock(&svc_xprt_class_lock);
136 	list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
137 		int slen;
138 
139 		slen = snprintf(tmpstr, sizeof(tmpstr), "%s %d\n",
140 				xcl->xcl_name, xcl->xcl_max_payload);
141 		if (slen >= sizeof(tmpstr) || len + slen >= maxlen)
142 			break;
143 		len += slen;
144 		strcat(buf, tmpstr);
145 	}
146 	spin_unlock(&svc_xprt_class_lock);
147 
148 	return len;
149 }
150 
151 /**
152  * svc_xprt_deferred_close - Close a transport
153  * @xprt: transport instance
154  *
155  * Used in contexts that need to defer the work of shutting down
156  * the transport to an nfsd thread.
157  */
158 void svc_xprt_deferred_close(struct svc_xprt *xprt)
159 {
160 	trace_svc_xprt_close(xprt);
161 	if (!test_and_set_bit(XPT_CLOSE, &xprt->xpt_flags))
162 		svc_xprt_enqueue(xprt);
163 }
164 EXPORT_SYMBOL_GPL(svc_xprt_deferred_close);
165 
166 static void svc_xprt_free(struct kref *kref)
167 {
168 	struct svc_xprt *xprt =
169 		container_of(kref, struct svc_xprt, xpt_ref);
170 	struct module *owner = xprt->xpt_class->xcl_owner;
171 	if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags))
172 		svcauth_unix_info_release(xprt);
173 	put_cred(xprt->xpt_cred);
174 	put_net_track(xprt->xpt_net, &xprt->ns_tracker);
175 	/* See comment on corresponding get in xs_setup_bc_tcp(): */
176 	if (xprt->xpt_bc_xprt)
177 		xprt_put(xprt->xpt_bc_xprt);
178 	if (xprt->xpt_bc_xps)
179 		xprt_switch_put(xprt->xpt_bc_xps);
180 	trace_svc_xprt_free(xprt);
181 	xprt->xpt_ops->xpo_free(xprt);
182 	module_put(owner);
183 }
184 
185 void svc_xprt_put(struct svc_xprt *xprt)
186 {
187 	kref_put(&xprt->xpt_ref, svc_xprt_free);
188 }
189 EXPORT_SYMBOL_GPL(svc_xprt_put);
190 
191 /*
192  * Called by transport drivers to initialize the transport independent
193  * portion of the transport instance.
194  */
195 void svc_xprt_init(struct net *net, struct svc_xprt_class *xcl,
196 		   struct svc_xprt *xprt, struct svc_serv *serv)
197 {
198 	memset(xprt, 0, sizeof(*xprt));
199 	xprt->xpt_class = xcl;
200 	xprt->xpt_ops = xcl->xcl_ops;
201 	kref_init(&xprt->xpt_ref);
202 	xprt->xpt_server = serv;
203 	INIT_LIST_HEAD(&xprt->xpt_list);
204 	INIT_LIST_HEAD(&xprt->xpt_deferred);
205 	INIT_LIST_HEAD(&xprt->xpt_users);
206 	mutex_init(&xprt->xpt_mutex);
207 	spin_lock_init(&xprt->xpt_lock);
208 	set_bit(XPT_BUSY, &xprt->xpt_flags);
209 	xprt->xpt_net = get_net_track(net, &xprt->ns_tracker, GFP_ATOMIC);
210 	strcpy(xprt->xpt_remotebuf, "uninitialized");
211 }
212 EXPORT_SYMBOL_GPL(svc_xprt_init);
213 
214 /**
215  * svc_xprt_received - start next receiver thread
216  * @xprt: controlling transport
217  *
218  * The caller must hold the XPT_BUSY bit and must
219  * not thereafter touch transport data.
220  *
221  * Note: XPT_DATA only gets cleared when a read-attempt finds no (or
222  * insufficient) data.
223  */
224 void svc_xprt_received(struct svc_xprt *xprt)
225 {
226 	if (!test_bit(XPT_BUSY, &xprt->xpt_flags)) {
227 		WARN_ONCE(1, "xprt=0x%p already busy!", xprt);
228 		return;
229 	}
230 
231 	/* As soon as we clear busy, the xprt could be closed and
232 	 * 'put', so we need a reference to call svc_xprt_enqueue with:
233 	 */
234 	svc_xprt_get(xprt);
235 	smp_mb__before_atomic();
236 	clear_bit(XPT_BUSY, &xprt->xpt_flags);
237 	svc_xprt_enqueue(xprt);
238 	svc_xprt_put(xprt);
239 }
240 EXPORT_SYMBOL_GPL(svc_xprt_received);
241 
242 void svc_add_new_perm_xprt(struct svc_serv *serv, struct svc_xprt *new)
243 {
244 	clear_bit(XPT_TEMP, &new->xpt_flags);
245 	spin_lock_bh(&serv->sv_lock);
246 	list_add(&new->xpt_list, &serv->sv_permsocks);
247 	spin_unlock_bh(&serv->sv_lock);
248 	svc_xprt_received(new);
249 }
250 
251 static int _svc_xprt_create(struct svc_serv *serv, const char *xprt_name,
252 			    struct net *net, struct sockaddr *sap,
253 			    size_t len, int flags, const struct cred *cred)
254 {
255 	struct svc_xprt_class *xcl;
256 
257 	spin_lock(&svc_xprt_class_lock);
258 	list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
259 		struct svc_xprt *newxprt;
260 		unsigned short newport;
261 
262 		if (strcmp(xprt_name, xcl->xcl_name))
263 			continue;
264 
265 		if (!try_module_get(xcl->xcl_owner))
266 			goto err;
267 
268 		spin_unlock(&svc_xprt_class_lock);
269 		newxprt = xcl->xcl_ops->xpo_create(serv, net, sap, len, flags);
270 		if (IS_ERR(newxprt)) {
271 			trace_svc_xprt_create_err(serv->sv_program->pg_name,
272 						  xcl->xcl_name, sap, len,
273 						  newxprt);
274 			module_put(xcl->xcl_owner);
275 			return PTR_ERR(newxprt);
276 		}
277 		newxprt->xpt_cred = get_cred(cred);
278 		svc_add_new_perm_xprt(serv, newxprt);
279 		newport = svc_xprt_local_port(newxprt);
280 		return newport;
281 	}
282  err:
283 	spin_unlock(&svc_xprt_class_lock);
284 	/* This errno is exposed to user space.  Provide a reasonable
285 	 * perror msg for a bad transport. */
286 	return -EPROTONOSUPPORT;
287 }
288 
289 /**
290  * svc_xprt_create_from_sa - Add a new listener to @serv from socket address
291  * @serv: target RPC service
292  * @xprt_name: transport class name
293  * @net: network namespace
294  * @sap: socket address pointer
295  * @flags: SVC_SOCK flags
296  * @cred: credential to bind to this transport
297  *
298  * Return local xprt port on success or %-EPROTONOSUPPORT on failure
299  */
300 int svc_xprt_create_from_sa(struct svc_serv *serv, const char *xprt_name,
301 			    struct net *net, struct sockaddr *sap,
302 			    int flags, const struct cred *cred)
303 {
304 	size_t len;
305 	int err;
306 
307 	switch (sap->sa_family) {
308 	case AF_INET:
309 		len = sizeof(struct sockaddr_in);
310 		break;
311 #if IS_ENABLED(CONFIG_IPV6)
312 	case AF_INET6:
313 		len = sizeof(struct sockaddr_in6);
314 		break;
315 #endif
316 	default:
317 		return -EAFNOSUPPORT;
318 	}
319 
320 	err = _svc_xprt_create(serv, xprt_name, net, sap, len, flags, cred);
321 	if (err == -EPROTONOSUPPORT) {
322 		request_module("svc%s", xprt_name);
323 		err = _svc_xprt_create(serv, xprt_name, net, sap, len, flags,
324 				       cred);
325 	}
326 
327 	return err;
328 }
329 EXPORT_SYMBOL_GPL(svc_xprt_create_from_sa);
330 
331 /**
332  * svc_xprt_create - Add a new listener to @serv
333  * @serv: target RPC service
334  * @xprt_name: transport class name
335  * @net: network namespace
336  * @family: network address family
337  * @port: listener port
338  * @flags: SVC_SOCK flags
339  * @cred: credential to bind to this transport
340  *
341  * Return local xprt port on success or %-EPROTONOSUPPORT on failure
342  */
343 int svc_xprt_create(struct svc_serv *serv, const char *xprt_name,
344 		    struct net *net, const int family,
345 		    const unsigned short port, int flags,
346 		    const struct cred *cred)
347 {
348 	struct sockaddr_in sin = {
349 		.sin_family		= AF_INET,
350 		.sin_addr.s_addr	= htonl(INADDR_ANY),
351 		.sin_port		= htons(port),
352 	};
353 #if IS_ENABLED(CONFIG_IPV6)
354 	struct sockaddr_in6 sin6 = {
355 		.sin6_family		= AF_INET6,
356 		.sin6_addr		= IN6ADDR_ANY_INIT,
357 		.sin6_port		= htons(port),
358 	};
359 #endif
360 	struct sockaddr *sap;
361 
362 	switch (family) {
363 	case PF_INET:
364 		sap = (struct sockaddr *)&sin;
365 		break;
366 #if IS_ENABLED(CONFIG_IPV6)
367 	case PF_INET6:
368 		sap = (struct sockaddr *)&sin6;
369 		break;
370 #endif
371 	default:
372 		return -EAFNOSUPPORT;
373 	}
374 
375 	return svc_xprt_create_from_sa(serv, xprt_name, net, sap, flags, cred);
376 }
377 EXPORT_SYMBOL_GPL(svc_xprt_create);
378 
379 /*
380  * Copy the local and remote xprt addresses to the rqstp structure
381  */
382 void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt)
383 {
384 	memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen);
385 	rqstp->rq_addrlen = xprt->xpt_remotelen;
386 
387 	/*
388 	 * Destination address in request is needed for binding the
389 	 * source address in RPC replies/callbacks later.
390 	 */
391 	memcpy(&rqstp->rq_daddr, &xprt->xpt_local, xprt->xpt_locallen);
392 	rqstp->rq_daddrlen = xprt->xpt_locallen;
393 }
394 EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs);
395 
396 /**
397  * svc_print_addr - Format rq_addr field for printing
398  * @rqstp: svc_rqst struct containing address to print
399  * @buf: target buffer for formatted address
400  * @len: length of target buffer
401  *
402  */
403 char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
404 {
405 	return __svc_print_addr(svc_addr(rqstp), buf, len);
406 }
407 EXPORT_SYMBOL_GPL(svc_print_addr);
408 
409 static bool svc_xprt_slots_in_range(struct svc_xprt *xprt)
410 {
411 	unsigned int limit = svc_rpc_per_connection_limit;
412 	int nrqsts = atomic_read(&xprt->xpt_nr_rqsts);
413 
414 	return limit == 0 || (nrqsts >= 0 && nrqsts < limit);
415 }
416 
417 static bool svc_xprt_reserve_slot(struct svc_rqst *rqstp, struct svc_xprt *xprt)
418 {
419 	if (!test_bit(RQ_DATA, &rqstp->rq_flags)) {
420 		if (!svc_xprt_slots_in_range(xprt))
421 			return false;
422 		atomic_inc(&xprt->xpt_nr_rqsts);
423 		set_bit(RQ_DATA, &rqstp->rq_flags);
424 	}
425 	return true;
426 }
427 
428 static void svc_xprt_release_slot(struct svc_rqst *rqstp)
429 {
430 	struct svc_xprt	*xprt = rqstp->rq_xprt;
431 	if (test_and_clear_bit(RQ_DATA, &rqstp->rq_flags)) {
432 		atomic_dec(&xprt->xpt_nr_rqsts);
433 		smp_wmb(); /* See smp_rmb() in svc_xprt_ready() */
434 		svc_xprt_enqueue(xprt);
435 	}
436 }
437 
438 static bool svc_xprt_ready(struct svc_xprt *xprt)
439 {
440 	unsigned long xpt_flags;
441 
442 	/*
443 	 * If another cpu has recently updated xpt_flags,
444 	 * sk_sock->flags, xpt_reserved, or xpt_nr_rqsts, we need to
445 	 * know about it; otherwise it's possible that both that cpu and
446 	 * this one could call svc_xprt_enqueue() without either
447 	 * svc_xprt_enqueue() recognizing that the conditions below
448 	 * are satisfied, and we could stall indefinitely:
449 	 */
450 	smp_rmb();
451 	xpt_flags = READ_ONCE(xprt->xpt_flags);
452 
453 	trace_svc_xprt_enqueue(xprt, xpt_flags);
454 	if (xpt_flags & BIT(XPT_BUSY))
455 		return false;
456 	if (xpt_flags & (BIT(XPT_CONN) | BIT(XPT_CLOSE) | BIT(XPT_HANDSHAKE)))
457 		return true;
458 	if (xpt_flags & (BIT(XPT_DATA) | BIT(XPT_DEFERRED))) {
459 		if (xprt->xpt_ops->xpo_has_wspace(xprt) &&
460 		    svc_xprt_slots_in_range(xprt))
461 			return true;
462 		trace_svc_xprt_no_write_space(xprt);
463 		return false;
464 	}
465 	return false;
466 }
467 
468 /**
469  * svc_xprt_enqueue - Queue a transport on an idle nfsd thread
470  * @xprt: transport with data pending
471  *
472  */
473 void svc_xprt_enqueue(struct svc_xprt *xprt)
474 {
475 	struct svc_pool *pool;
476 
477 	if (!svc_xprt_ready(xprt))
478 		return;
479 
480 	/* Mark transport as busy. It will remain in this state until
481 	 * the provider calls svc_xprt_received. We update XPT_BUSY
482 	 * atomically because it also guards against trying to enqueue
483 	 * the transport twice.
484 	 */
485 	if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
486 		return;
487 
488 	pool = svc_pool_for_cpu(xprt->xpt_server);
489 
490 	percpu_counter_inc(&pool->sp_sockets_queued);
491 	lwq_enqueue(&xprt->xpt_ready, &pool->sp_xprts);
492 
493 	svc_pool_wake_idle_thread(pool);
494 }
495 EXPORT_SYMBOL_GPL(svc_xprt_enqueue);
496 
497 /*
498  * Dequeue the first transport, if there is one.
499  */
500 static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool)
501 {
502 	struct svc_xprt	*xprt = NULL;
503 
504 	xprt = lwq_dequeue(&pool->sp_xprts, struct svc_xprt, xpt_ready);
505 	if (xprt)
506 		svc_xprt_get(xprt);
507 	return xprt;
508 }
509 
510 /**
511  * svc_reserve - change the space reserved for the reply to a request.
512  * @rqstp:  The request in question
513  * @space: new max space to reserve
514  *
515  * Each request reserves some space on the output queue of the transport
516  * to make sure the reply fits.  This function reduces that reserved
517  * space to be the amount of space used already, plus @space.
518  *
519  */
520 void svc_reserve(struct svc_rqst *rqstp, int space)
521 {
522 	struct svc_xprt *xprt = rqstp->rq_xprt;
523 
524 	space += rqstp->rq_res.head[0].iov_len;
525 
526 	if (xprt && space < rqstp->rq_reserved) {
527 		atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved);
528 		rqstp->rq_reserved = space;
529 		smp_wmb(); /* See smp_rmb() in svc_xprt_ready() */
530 		svc_xprt_enqueue(xprt);
531 	}
532 }
533 EXPORT_SYMBOL_GPL(svc_reserve);
534 
535 static void free_deferred(struct svc_xprt *xprt, struct svc_deferred_req *dr)
536 {
537 	if (!dr)
538 		return;
539 
540 	xprt->xpt_ops->xpo_release_ctxt(xprt, dr->xprt_ctxt);
541 	kfree(dr);
542 }
543 
544 static void svc_xprt_release(struct svc_rqst *rqstp)
545 {
546 	struct svc_xprt	*xprt = rqstp->rq_xprt;
547 
548 	xprt->xpt_ops->xpo_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
549 	rqstp->rq_xprt_ctxt = NULL;
550 
551 	free_deferred(xprt, rqstp->rq_deferred);
552 	rqstp->rq_deferred = NULL;
553 
554 	svc_rqst_release_pages(rqstp);
555 	rqstp->rq_res.page_len = 0;
556 	rqstp->rq_res.page_base = 0;
557 
558 	/* Reset response buffer and release
559 	 * the reservation.
560 	 * But first, check that enough space was reserved
561 	 * for the reply, otherwise we have a bug!
562 	 */
563 	if ((rqstp->rq_res.len) >  rqstp->rq_reserved)
564 		printk(KERN_ERR "RPC request reserved %d but used %d\n",
565 		       rqstp->rq_reserved,
566 		       rqstp->rq_res.len);
567 
568 	rqstp->rq_res.head[0].iov_len = 0;
569 	svc_reserve(rqstp, 0);
570 	svc_xprt_release_slot(rqstp);
571 	rqstp->rq_xprt = NULL;
572 	svc_xprt_put(xprt);
573 }
574 
575 /**
576  * svc_wake_up - Wake up a service thread for non-transport work
577  * @serv: RPC service
578  *
579  * Some svc_serv's will have occasional work to do, even when a xprt is not
580  * waiting to be serviced. This function is there to "kick" a task in one of
581  * those services so that it can wake up and do that work. Note that we only
582  * bother with pool 0 as we don't need to wake up more than one thread for
583  * this purpose.
584  */
585 void svc_wake_up(struct svc_serv *serv)
586 {
587 	struct svc_pool *pool = &serv->sv_pools[0];
588 
589 	set_bit(SP_TASK_PENDING, &pool->sp_flags);
590 	svc_pool_wake_idle_thread(pool);
591 }
592 EXPORT_SYMBOL_GPL(svc_wake_up);
593 
594 int svc_port_is_privileged(struct sockaddr *sin)
595 {
596 	switch (sin->sa_family) {
597 	case AF_INET:
598 		return ntohs(((struct sockaddr_in *)sin)->sin_port)
599 			< PROT_SOCK;
600 	case AF_INET6:
601 		return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
602 			< PROT_SOCK;
603 	default:
604 		return 0;
605 	}
606 }
607 
608 /*
609  * Make sure that we don't have too many active connections. If we have,
610  * something must be dropped. It's not clear what will happen if we allow
611  * "too many" connections, but when dealing with network-facing software,
612  * we have to code defensively. Here we do that by imposing hard limits.
613  *
614  * There's no point in trying to do random drop here for DoS
615  * prevention. The NFS clients does 1 reconnect in 15 seconds. An
616  * attacker can easily beat that.
617  *
618  * The only somewhat efficient mechanism would be if drop old
619  * connections from the same IP first. But right now we don't even
620  * record the client IP in svc_sock.
621  *
622  * single-threaded services that expect a lot of clients will probably
623  * need to set sv_maxconn to override the default value which is based
624  * on the number of threads
625  */
626 static void svc_check_conn_limits(struct svc_serv *serv)
627 {
628 	unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn :
629 				(serv->sv_nrthreads+3) * 20;
630 
631 	if (serv->sv_tmpcnt > limit) {
632 		struct svc_xprt *xprt = NULL;
633 		spin_lock_bh(&serv->sv_lock);
634 		if (!list_empty(&serv->sv_tempsocks)) {
635 			/* Try to help the admin */
636 			net_notice_ratelimited("%s: too many open connections, consider increasing the %s\n",
637 					       serv->sv_name, serv->sv_maxconn ?
638 					       "max number of connections" :
639 					       "number of threads");
640 			/*
641 			 * Always select the oldest connection. It's not fair,
642 			 * but so is life
643 			 */
644 			xprt = list_entry(serv->sv_tempsocks.prev,
645 					  struct svc_xprt,
646 					  xpt_list);
647 			set_bit(XPT_CLOSE, &xprt->xpt_flags);
648 			svc_xprt_get(xprt);
649 		}
650 		spin_unlock_bh(&serv->sv_lock);
651 
652 		if (xprt) {
653 			svc_xprt_enqueue(xprt);
654 			svc_xprt_put(xprt);
655 		}
656 	}
657 }
658 
659 static bool svc_alloc_arg(struct svc_rqst *rqstp)
660 {
661 	struct svc_serv *serv = rqstp->rq_server;
662 	struct xdr_buf *arg = &rqstp->rq_arg;
663 	unsigned long pages, filled, ret;
664 
665 	pages = (serv->sv_max_mesg + 2 * PAGE_SIZE) >> PAGE_SHIFT;
666 	if (pages > RPCSVC_MAXPAGES) {
667 		pr_warn_once("svc: warning: pages=%lu > RPCSVC_MAXPAGES=%lu\n",
668 			     pages, RPCSVC_MAXPAGES);
669 		/* use as many pages as possible */
670 		pages = RPCSVC_MAXPAGES;
671 	}
672 
673 	for (filled = 0; filled < pages; filled = ret) {
674 		ret = alloc_pages_bulk_array(GFP_KERNEL, pages,
675 					     rqstp->rq_pages);
676 		if (ret > filled)
677 			/* Made progress, don't sleep yet */
678 			continue;
679 
680 		set_current_state(TASK_IDLE);
681 		if (svc_thread_should_stop(rqstp)) {
682 			set_current_state(TASK_RUNNING);
683 			return false;
684 		}
685 		trace_svc_alloc_arg_err(pages, ret);
686 		memalloc_retry_wait(GFP_KERNEL);
687 	}
688 	rqstp->rq_page_end = &rqstp->rq_pages[pages];
689 	rqstp->rq_pages[pages] = NULL; /* this might be seen in nfsd_splice_actor() */
690 
691 	/* Make arg->head point to first page and arg->pages point to rest */
692 	arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
693 	arg->head[0].iov_len = PAGE_SIZE;
694 	arg->pages = rqstp->rq_pages + 1;
695 	arg->page_base = 0;
696 	/* save at least one page for response */
697 	arg->page_len = (pages-2)*PAGE_SIZE;
698 	arg->len = (pages-1)*PAGE_SIZE;
699 	arg->tail[0].iov_len = 0;
700 
701 	rqstp->rq_xid = xdr_zero;
702 	return true;
703 }
704 
705 static bool
706 svc_thread_should_sleep(struct svc_rqst *rqstp)
707 {
708 	struct svc_pool		*pool = rqstp->rq_pool;
709 
710 	/* did someone call svc_wake_up? */
711 	if (test_bit(SP_TASK_PENDING, &pool->sp_flags))
712 		return false;
713 
714 	/* was a socket queued? */
715 	if (!lwq_empty(&pool->sp_xprts))
716 		return false;
717 
718 	/* are we shutting down? */
719 	if (svc_thread_should_stop(rqstp))
720 		return false;
721 
722 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
723 	if (svc_is_backchannel(rqstp)) {
724 		if (!lwq_empty(&rqstp->rq_server->sv_cb_list))
725 			return false;
726 	}
727 #endif
728 
729 	return true;
730 }
731 
732 static void svc_thread_wait_for_work(struct svc_rqst *rqstp)
733 {
734 	struct svc_pool *pool = rqstp->rq_pool;
735 
736 	if (svc_thread_should_sleep(rqstp)) {
737 		set_current_state(TASK_IDLE | TASK_FREEZABLE);
738 		llist_add(&rqstp->rq_idle, &pool->sp_idle_threads);
739 		if (likely(svc_thread_should_sleep(rqstp)))
740 			schedule();
741 
742 		while (!llist_del_first_this(&pool->sp_idle_threads,
743 					     &rqstp->rq_idle)) {
744 			/* Work just became available.  This thread can only
745 			 * handle it after removing rqstp from the idle
746 			 * list. If that attempt failed, some other thread
747 			 * must have queued itself after finding no
748 			 * work to do, so that thread has taken responsibly
749 			 * for this new work.  This thread can safely sleep
750 			 * until woken again.
751 			 */
752 			schedule();
753 			set_current_state(TASK_IDLE | TASK_FREEZABLE);
754 		}
755 		__set_current_state(TASK_RUNNING);
756 	} else {
757 		cond_resched();
758 	}
759 	try_to_freeze();
760 }
761 
762 static void svc_add_new_temp_xprt(struct svc_serv *serv, struct svc_xprt *newxpt)
763 {
764 	spin_lock_bh(&serv->sv_lock);
765 	set_bit(XPT_TEMP, &newxpt->xpt_flags);
766 	list_add(&newxpt->xpt_list, &serv->sv_tempsocks);
767 	serv->sv_tmpcnt++;
768 	if (serv->sv_temptimer.function == NULL) {
769 		/* setup timer to age temp transports */
770 		serv->sv_temptimer.function = svc_age_temp_xprts;
771 		mod_timer(&serv->sv_temptimer,
772 			  jiffies + svc_conn_age_period * HZ);
773 	}
774 	spin_unlock_bh(&serv->sv_lock);
775 	svc_xprt_received(newxpt);
776 }
777 
778 static void svc_handle_xprt(struct svc_rqst *rqstp, struct svc_xprt *xprt)
779 {
780 	struct svc_serv *serv = rqstp->rq_server;
781 	int len = 0;
782 
783 	if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) {
784 		if (test_and_clear_bit(XPT_KILL_TEMP, &xprt->xpt_flags))
785 			xprt->xpt_ops->xpo_kill_temp_xprt(xprt);
786 		svc_delete_xprt(xprt);
787 		/* Leave XPT_BUSY set on the dead xprt: */
788 		goto out;
789 	}
790 	if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
791 		struct svc_xprt *newxpt;
792 		/*
793 		 * We know this module_get will succeed because the
794 		 * listener holds a reference too
795 		 */
796 		__module_get(xprt->xpt_class->xcl_owner);
797 		svc_check_conn_limits(xprt->xpt_server);
798 		newxpt = xprt->xpt_ops->xpo_accept(xprt);
799 		if (newxpt) {
800 			newxpt->xpt_cred = get_cred(xprt->xpt_cred);
801 			svc_add_new_temp_xprt(serv, newxpt);
802 			trace_svc_xprt_accept(newxpt, serv->sv_name);
803 		} else {
804 			module_put(xprt->xpt_class->xcl_owner);
805 		}
806 		svc_xprt_received(xprt);
807 	} else if (test_bit(XPT_HANDSHAKE, &xprt->xpt_flags)) {
808 		xprt->xpt_ops->xpo_handshake(xprt);
809 		svc_xprt_received(xprt);
810 	} else if (svc_xprt_reserve_slot(rqstp, xprt)) {
811 		/* XPT_DATA|XPT_DEFERRED case: */
812 		rqstp->rq_deferred = svc_deferred_dequeue(xprt);
813 		if (rqstp->rq_deferred)
814 			len = svc_deferred_recv(rqstp);
815 		else
816 			len = xprt->xpt_ops->xpo_recvfrom(rqstp);
817 		rqstp->rq_reserved = serv->sv_max_mesg;
818 		atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
819 		if (len <= 0)
820 			goto out;
821 
822 		trace_svc_xdr_recvfrom(&rqstp->rq_arg);
823 
824 		clear_bit(XPT_OLD, &xprt->xpt_flags);
825 
826 		rqstp->rq_chandle.defer = svc_defer;
827 
828 		if (serv->sv_stats)
829 			serv->sv_stats->netcnt++;
830 		percpu_counter_inc(&rqstp->rq_pool->sp_messages_arrived);
831 		rqstp->rq_stime = ktime_get();
832 		svc_process(rqstp);
833 	} else
834 		svc_xprt_received(xprt);
835 
836 out:
837 	rqstp->rq_res.len = 0;
838 	svc_xprt_release(rqstp);
839 }
840 
841 static void svc_thread_wake_next(struct svc_rqst *rqstp)
842 {
843 	if (!svc_thread_should_sleep(rqstp))
844 		/* More work pending after I dequeued some,
845 		 * wake another worker
846 		 */
847 		svc_pool_wake_idle_thread(rqstp->rq_pool);
848 }
849 
850 /**
851  * svc_recv - Receive and process the next request on any transport
852  * @rqstp: an idle RPC service thread
853  *
854  * This code is carefully organised not to touch any cachelines in
855  * the shared svc_serv structure, only cachelines in the local
856  * svc_pool.
857  */
858 void svc_recv(struct svc_rqst *rqstp)
859 {
860 	struct svc_pool *pool = rqstp->rq_pool;
861 
862 	if (!svc_alloc_arg(rqstp))
863 		return;
864 
865 	svc_thread_wait_for_work(rqstp);
866 
867 	clear_bit(SP_TASK_PENDING, &pool->sp_flags);
868 
869 	if (svc_thread_should_stop(rqstp)) {
870 		svc_thread_wake_next(rqstp);
871 		return;
872 	}
873 
874 	rqstp->rq_xprt = svc_xprt_dequeue(pool);
875 	if (rqstp->rq_xprt) {
876 		struct svc_xprt *xprt = rqstp->rq_xprt;
877 
878 		svc_thread_wake_next(rqstp);
879 		/* Normally we will wait up to 5 seconds for any required
880 		 * cache information to be provided.  When there are no
881 		 * idle threads, we reduce the wait time.
882 		 */
883 		if (pool->sp_idle_threads.first)
884 			rqstp->rq_chandle.thread_wait = 5 * HZ;
885 		else
886 			rqstp->rq_chandle.thread_wait = 1 * HZ;
887 
888 		trace_svc_xprt_dequeue(rqstp);
889 		svc_handle_xprt(rqstp, xprt);
890 	}
891 
892 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
893 	if (svc_is_backchannel(rqstp)) {
894 		struct svc_serv *serv = rqstp->rq_server;
895 		struct rpc_rqst *req;
896 
897 		req = lwq_dequeue(&serv->sv_cb_list,
898 				  struct rpc_rqst, rq_bc_list);
899 		if (req) {
900 			svc_thread_wake_next(rqstp);
901 			svc_process_bc(req, rqstp);
902 		}
903 	}
904 #endif
905 }
906 EXPORT_SYMBOL_GPL(svc_recv);
907 
908 /*
909  * Drop request
910  */
911 void svc_drop(struct svc_rqst *rqstp)
912 {
913 	trace_svc_drop(rqstp);
914 }
915 EXPORT_SYMBOL_GPL(svc_drop);
916 
917 /**
918  * svc_send - Return reply to client
919  * @rqstp: RPC transaction context
920  *
921  */
922 void svc_send(struct svc_rqst *rqstp)
923 {
924 	struct svc_xprt	*xprt;
925 	struct xdr_buf	*xb;
926 	int status;
927 
928 	xprt = rqstp->rq_xprt;
929 
930 	/* calculate over-all length */
931 	xb = &rqstp->rq_res;
932 	xb->len = xb->head[0].iov_len +
933 		xb->page_len +
934 		xb->tail[0].iov_len;
935 	trace_svc_xdr_sendto(rqstp->rq_xid, xb);
936 	trace_svc_stats_latency(rqstp);
937 
938 	status = xprt->xpt_ops->xpo_sendto(rqstp);
939 
940 	trace_svc_send(rqstp, status);
941 }
942 
943 /*
944  * Timer function to close old temporary transports, using
945  * a mark-and-sweep algorithm.
946  */
947 static void svc_age_temp_xprts(struct timer_list *t)
948 {
949 	struct svc_serv *serv = from_timer(serv, t, sv_temptimer);
950 	struct svc_xprt *xprt;
951 	struct list_head *le, *next;
952 
953 	dprintk("svc_age_temp_xprts\n");
954 
955 	if (!spin_trylock_bh(&serv->sv_lock)) {
956 		/* busy, try again 1 sec later */
957 		dprintk("svc_age_temp_xprts: busy\n");
958 		mod_timer(&serv->sv_temptimer, jiffies + HZ);
959 		return;
960 	}
961 
962 	list_for_each_safe(le, next, &serv->sv_tempsocks) {
963 		xprt = list_entry(le, struct svc_xprt, xpt_list);
964 
965 		/* First time through, just mark it OLD. Second time
966 		 * through, close it. */
967 		if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags))
968 			continue;
969 		if (kref_read(&xprt->xpt_ref) > 1 ||
970 		    test_bit(XPT_BUSY, &xprt->xpt_flags))
971 			continue;
972 		list_del_init(le);
973 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
974 		dprintk("queuing xprt %p for closing\n", xprt);
975 
976 		/* a thread will dequeue and close it soon */
977 		svc_xprt_enqueue(xprt);
978 	}
979 	spin_unlock_bh(&serv->sv_lock);
980 
981 	mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
982 }
983 
984 /* Close temporary transports whose xpt_local matches server_addr immediately
985  * instead of waiting for them to be picked up by the timer.
986  *
987  * This is meant to be called from a notifier_block that runs when an ip
988  * address is deleted.
989  */
990 void svc_age_temp_xprts_now(struct svc_serv *serv, struct sockaddr *server_addr)
991 {
992 	struct svc_xprt *xprt;
993 	struct list_head *le, *next;
994 	LIST_HEAD(to_be_closed);
995 
996 	spin_lock_bh(&serv->sv_lock);
997 	list_for_each_safe(le, next, &serv->sv_tempsocks) {
998 		xprt = list_entry(le, struct svc_xprt, xpt_list);
999 		if (rpc_cmp_addr(server_addr, (struct sockaddr *)
1000 				&xprt->xpt_local)) {
1001 			dprintk("svc_age_temp_xprts_now: found %p\n", xprt);
1002 			list_move(le, &to_be_closed);
1003 		}
1004 	}
1005 	spin_unlock_bh(&serv->sv_lock);
1006 
1007 	while (!list_empty(&to_be_closed)) {
1008 		le = to_be_closed.next;
1009 		list_del_init(le);
1010 		xprt = list_entry(le, struct svc_xprt, xpt_list);
1011 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
1012 		set_bit(XPT_KILL_TEMP, &xprt->xpt_flags);
1013 		dprintk("svc_age_temp_xprts_now: queuing xprt %p for closing\n",
1014 				xprt);
1015 		svc_xprt_enqueue(xprt);
1016 	}
1017 }
1018 EXPORT_SYMBOL_GPL(svc_age_temp_xprts_now);
1019 
1020 static void call_xpt_users(struct svc_xprt *xprt)
1021 {
1022 	struct svc_xpt_user *u;
1023 
1024 	spin_lock(&xprt->xpt_lock);
1025 	while (!list_empty(&xprt->xpt_users)) {
1026 		u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list);
1027 		list_del_init(&u->list);
1028 		u->callback(u);
1029 	}
1030 	spin_unlock(&xprt->xpt_lock);
1031 }
1032 
1033 /*
1034  * Remove a dead transport
1035  */
1036 static void svc_delete_xprt(struct svc_xprt *xprt)
1037 {
1038 	struct svc_serv	*serv = xprt->xpt_server;
1039 	struct svc_deferred_req *dr;
1040 
1041 	if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags))
1042 		return;
1043 
1044 	trace_svc_xprt_detach(xprt);
1045 	xprt->xpt_ops->xpo_detach(xprt);
1046 	if (xprt->xpt_bc_xprt)
1047 		xprt->xpt_bc_xprt->ops->close(xprt->xpt_bc_xprt);
1048 
1049 	spin_lock_bh(&serv->sv_lock);
1050 	list_del_init(&xprt->xpt_list);
1051 	if (test_bit(XPT_TEMP, &xprt->xpt_flags))
1052 		serv->sv_tmpcnt--;
1053 	spin_unlock_bh(&serv->sv_lock);
1054 
1055 	while ((dr = svc_deferred_dequeue(xprt)) != NULL)
1056 		free_deferred(xprt, dr);
1057 
1058 	call_xpt_users(xprt);
1059 	svc_xprt_put(xprt);
1060 }
1061 
1062 /**
1063  * svc_xprt_close - Close a client connection
1064  * @xprt: transport to disconnect
1065  *
1066  */
1067 void svc_xprt_close(struct svc_xprt *xprt)
1068 {
1069 	trace_svc_xprt_close(xprt);
1070 	set_bit(XPT_CLOSE, &xprt->xpt_flags);
1071 	if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
1072 		/* someone else will have to effect the close */
1073 		return;
1074 	/*
1075 	 * We expect svc_close_xprt() to work even when no threads are
1076 	 * running (e.g., while configuring the server before starting
1077 	 * any threads), so if the transport isn't busy, we delete
1078 	 * it ourself:
1079 	 */
1080 	svc_delete_xprt(xprt);
1081 }
1082 EXPORT_SYMBOL_GPL(svc_xprt_close);
1083 
1084 static int svc_close_list(struct svc_serv *serv, struct list_head *xprt_list, struct net *net)
1085 {
1086 	struct svc_xprt *xprt;
1087 	int ret = 0;
1088 
1089 	spin_lock_bh(&serv->sv_lock);
1090 	list_for_each_entry(xprt, xprt_list, xpt_list) {
1091 		if (xprt->xpt_net != net)
1092 			continue;
1093 		ret++;
1094 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
1095 		svc_xprt_enqueue(xprt);
1096 	}
1097 	spin_unlock_bh(&serv->sv_lock);
1098 	return ret;
1099 }
1100 
1101 static void svc_clean_up_xprts(struct svc_serv *serv, struct net *net)
1102 {
1103 	struct svc_xprt *xprt;
1104 	int i;
1105 
1106 	for (i = 0; i < serv->sv_nrpools; i++) {
1107 		struct svc_pool *pool = &serv->sv_pools[i];
1108 		struct llist_node *q, **t1, *t2;
1109 
1110 		q = lwq_dequeue_all(&pool->sp_xprts);
1111 		lwq_for_each_safe(xprt, t1, t2, &q, xpt_ready) {
1112 			if (xprt->xpt_net == net) {
1113 				set_bit(XPT_CLOSE, &xprt->xpt_flags);
1114 				svc_delete_xprt(xprt);
1115 				xprt = NULL;
1116 			}
1117 		}
1118 
1119 		if (q)
1120 			lwq_enqueue_batch(q, &pool->sp_xprts);
1121 	}
1122 }
1123 
1124 /**
1125  * svc_xprt_destroy_all - Destroy transports associated with @serv
1126  * @serv: RPC service to be shut down
1127  * @net: target network namespace
1128  *
1129  * Server threads may still be running (especially in the case where the
1130  * service is still running in other network namespaces).
1131  *
1132  * So we shut down sockets the same way we would on a running server, by
1133  * setting XPT_CLOSE, enqueuing, and letting a thread pick it up to do
1134  * the close.  In the case there are no such other threads,
1135  * threads running, svc_clean_up_xprts() does a simple version of a
1136  * server's main event loop, and in the case where there are other
1137  * threads, we may need to wait a little while and then check again to
1138  * see if they're done.
1139  */
1140 void svc_xprt_destroy_all(struct svc_serv *serv, struct net *net)
1141 {
1142 	int delay = 0;
1143 
1144 	while (svc_close_list(serv, &serv->sv_permsocks, net) +
1145 	       svc_close_list(serv, &serv->sv_tempsocks, net)) {
1146 
1147 		svc_clean_up_xprts(serv, net);
1148 		msleep(delay++);
1149 	}
1150 }
1151 EXPORT_SYMBOL_GPL(svc_xprt_destroy_all);
1152 
1153 /*
1154  * Handle defer and revisit of requests
1155  */
1156 
1157 static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1158 {
1159 	struct svc_deferred_req *dr =
1160 		container_of(dreq, struct svc_deferred_req, handle);
1161 	struct svc_xprt *xprt = dr->xprt;
1162 
1163 	spin_lock(&xprt->xpt_lock);
1164 	set_bit(XPT_DEFERRED, &xprt->xpt_flags);
1165 	if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) {
1166 		spin_unlock(&xprt->xpt_lock);
1167 		trace_svc_defer_drop(dr);
1168 		free_deferred(xprt, dr);
1169 		svc_xprt_put(xprt);
1170 		return;
1171 	}
1172 	dr->xprt = NULL;
1173 	list_add(&dr->handle.recent, &xprt->xpt_deferred);
1174 	spin_unlock(&xprt->xpt_lock);
1175 	trace_svc_defer_queue(dr);
1176 	svc_xprt_enqueue(xprt);
1177 	svc_xprt_put(xprt);
1178 }
1179 
1180 /*
1181  * Save the request off for later processing. The request buffer looks
1182  * like this:
1183  *
1184  * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
1185  *
1186  * This code can only handle requests that consist of an xprt-header
1187  * and rpc-header.
1188  */
1189 static struct cache_deferred_req *svc_defer(struct cache_req *req)
1190 {
1191 	struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
1192 	struct svc_deferred_req *dr;
1193 
1194 	if (rqstp->rq_arg.page_len || !test_bit(RQ_USEDEFERRAL, &rqstp->rq_flags))
1195 		return NULL; /* if more than a page, give up FIXME */
1196 	if (rqstp->rq_deferred) {
1197 		dr = rqstp->rq_deferred;
1198 		rqstp->rq_deferred = NULL;
1199 	} else {
1200 		size_t skip;
1201 		size_t size;
1202 		/* FIXME maybe discard if size too large */
1203 		size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
1204 		dr = kmalloc(size, GFP_KERNEL);
1205 		if (dr == NULL)
1206 			return NULL;
1207 
1208 		dr->handle.owner = rqstp->rq_server;
1209 		dr->prot = rqstp->rq_prot;
1210 		memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
1211 		dr->addrlen = rqstp->rq_addrlen;
1212 		dr->daddr = rqstp->rq_daddr;
1213 		dr->argslen = rqstp->rq_arg.len >> 2;
1214 
1215 		/* back up head to the start of the buffer and copy */
1216 		skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1217 		memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
1218 		       dr->argslen << 2);
1219 	}
1220 	dr->xprt_ctxt = rqstp->rq_xprt_ctxt;
1221 	rqstp->rq_xprt_ctxt = NULL;
1222 	trace_svc_defer(rqstp);
1223 	svc_xprt_get(rqstp->rq_xprt);
1224 	dr->xprt = rqstp->rq_xprt;
1225 	set_bit(RQ_DROPME, &rqstp->rq_flags);
1226 
1227 	dr->handle.revisit = svc_revisit;
1228 	return &dr->handle;
1229 }
1230 
1231 /*
1232  * recv data from a deferred request into an active one
1233  */
1234 static noinline int svc_deferred_recv(struct svc_rqst *rqstp)
1235 {
1236 	struct svc_deferred_req *dr = rqstp->rq_deferred;
1237 
1238 	trace_svc_defer_recv(dr);
1239 
1240 	/* setup iov_base past transport header */
1241 	rqstp->rq_arg.head[0].iov_base = dr->args;
1242 	/* The iov_len does not include the transport header bytes */
1243 	rqstp->rq_arg.head[0].iov_len = dr->argslen << 2;
1244 	rqstp->rq_arg.page_len = 0;
1245 	/* The rq_arg.len includes the transport header bytes */
1246 	rqstp->rq_arg.len     = dr->argslen << 2;
1247 	rqstp->rq_prot        = dr->prot;
1248 	memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
1249 	rqstp->rq_addrlen     = dr->addrlen;
1250 	/* Save off transport header len in case we get deferred again */
1251 	rqstp->rq_daddr       = dr->daddr;
1252 	rqstp->rq_respages    = rqstp->rq_pages;
1253 	rqstp->rq_xprt_ctxt   = dr->xprt_ctxt;
1254 
1255 	dr->xprt_ctxt = NULL;
1256 	svc_xprt_received(rqstp->rq_xprt);
1257 	return dr->argslen << 2;
1258 }
1259 
1260 
1261 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt)
1262 {
1263 	struct svc_deferred_req *dr = NULL;
1264 
1265 	if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags))
1266 		return NULL;
1267 	spin_lock(&xprt->xpt_lock);
1268 	if (!list_empty(&xprt->xpt_deferred)) {
1269 		dr = list_entry(xprt->xpt_deferred.next,
1270 				struct svc_deferred_req,
1271 				handle.recent);
1272 		list_del_init(&dr->handle.recent);
1273 	} else
1274 		clear_bit(XPT_DEFERRED, &xprt->xpt_flags);
1275 	spin_unlock(&xprt->xpt_lock);
1276 	return dr;
1277 }
1278 
1279 /**
1280  * svc_find_listener - find an RPC transport instance
1281  * @serv: pointer to svc_serv to search
1282  * @xcl_name: C string containing transport's class name
1283  * @net: owner net pointer
1284  * @sa: sockaddr containing address
1285  *
1286  * Return the transport instance pointer for the endpoint accepting
1287  * connections/peer traffic from the specified transport class,
1288  * and matching sockaddr.
1289  */
1290 struct svc_xprt *svc_find_listener(struct svc_serv *serv, const char *xcl_name,
1291 				   struct net *net, const struct sockaddr *sa)
1292 {
1293 	struct svc_xprt *xprt;
1294 	struct svc_xprt *found = NULL;
1295 
1296 	spin_lock_bh(&serv->sv_lock);
1297 	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1298 		if (xprt->xpt_net != net)
1299 			continue;
1300 		if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1301 			continue;
1302 		if (!rpc_cmp_addr_port(sa, (struct sockaddr *)&xprt->xpt_local))
1303 			continue;
1304 		found = xprt;
1305 		svc_xprt_get(xprt);
1306 		break;
1307 	}
1308 	spin_unlock_bh(&serv->sv_lock);
1309 	return found;
1310 }
1311 EXPORT_SYMBOL_GPL(svc_find_listener);
1312 
1313 /**
1314  * svc_find_xprt - find an RPC transport instance
1315  * @serv: pointer to svc_serv to search
1316  * @xcl_name: C string containing transport's class name
1317  * @net: owner net pointer
1318  * @af: Address family of transport's local address
1319  * @port: transport's IP port number
1320  *
1321  * Return the transport instance pointer for the endpoint accepting
1322  * connections/peer traffic from the specified transport class,
1323  * address family and port.
1324  *
1325  * Specifying 0 for the address family or port is effectively a
1326  * wild-card, and will result in matching the first transport in the
1327  * service's list that has a matching class name.
1328  */
1329 struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name,
1330 			       struct net *net, const sa_family_t af,
1331 			       const unsigned short port)
1332 {
1333 	struct svc_xprt *xprt;
1334 	struct svc_xprt *found = NULL;
1335 
1336 	/* Sanity check the args */
1337 	if (serv == NULL || xcl_name == NULL)
1338 		return found;
1339 
1340 	spin_lock_bh(&serv->sv_lock);
1341 	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1342 		if (xprt->xpt_net != net)
1343 			continue;
1344 		if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1345 			continue;
1346 		if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family)
1347 			continue;
1348 		if (port != 0 && port != svc_xprt_local_port(xprt))
1349 			continue;
1350 		found = xprt;
1351 		svc_xprt_get(xprt);
1352 		break;
1353 	}
1354 	spin_unlock_bh(&serv->sv_lock);
1355 	return found;
1356 }
1357 EXPORT_SYMBOL_GPL(svc_find_xprt);
1358 
1359 static int svc_one_xprt_name(const struct svc_xprt *xprt,
1360 			     char *pos, int remaining)
1361 {
1362 	int len;
1363 
1364 	len = snprintf(pos, remaining, "%s %u\n",
1365 			xprt->xpt_class->xcl_name,
1366 			svc_xprt_local_port(xprt));
1367 	if (len >= remaining)
1368 		return -ENAMETOOLONG;
1369 	return len;
1370 }
1371 
1372 /**
1373  * svc_xprt_names - format a buffer with a list of transport names
1374  * @serv: pointer to an RPC service
1375  * @buf: pointer to a buffer to be filled in
1376  * @buflen: length of buffer to be filled in
1377  *
1378  * Fills in @buf with a string containing a list of transport names,
1379  * each name terminated with '\n'.
1380  *
1381  * Returns positive length of the filled-in string on success; otherwise
1382  * a negative errno value is returned if an error occurs.
1383  */
1384 int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen)
1385 {
1386 	struct svc_xprt *xprt;
1387 	int len, totlen;
1388 	char *pos;
1389 
1390 	/* Sanity check args */
1391 	if (!serv)
1392 		return 0;
1393 
1394 	spin_lock_bh(&serv->sv_lock);
1395 
1396 	pos = buf;
1397 	totlen = 0;
1398 	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1399 		len = svc_one_xprt_name(xprt, pos, buflen - totlen);
1400 		if (len < 0) {
1401 			*buf = '\0';
1402 			totlen = len;
1403 		}
1404 		if (len <= 0)
1405 			break;
1406 
1407 		pos += len;
1408 		totlen += len;
1409 	}
1410 
1411 	spin_unlock_bh(&serv->sv_lock);
1412 	return totlen;
1413 }
1414 EXPORT_SYMBOL_GPL(svc_xprt_names);
1415 
1416 /*----------------------------------------------------------------------------*/
1417 
1418 static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos)
1419 {
1420 	unsigned int pidx = (unsigned int)*pos;
1421 	struct svc_info *si = m->private;
1422 
1423 	dprintk("svc_pool_stats_start, *pidx=%u\n", pidx);
1424 
1425 	mutex_lock(si->mutex);
1426 
1427 	if (!pidx)
1428 		return SEQ_START_TOKEN;
1429 	if (!si->serv)
1430 		return NULL;
1431 	return pidx > si->serv->sv_nrpools ? NULL
1432 		: &si->serv->sv_pools[pidx - 1];
1433 }
1434 
1435 static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos)
1436 {
1437 	struct svc_pool *pool = p;
1438 	struct svc_info *si = m->private;
1439 	struct svc_serv *serv = si->serv;
1440 
1441 	dprintk("svc_pool_stats_next, *pos=%llu\n", *pos);
1442 
1443 	if (!serv) {
1444 		pool = NULL;
1445 	} else if (p == SEQ_START_TOKEN) {
1446 		pool = &serv->sv_pools[0];
1447 	} else {
1448 		unsigned int pidx = (pool - &serv->sv_pools[0]);
1449 		if (pidx < serv->sv_nrpools-1)
1450 			pool = &serv->sv_pools[pidx+1];
1451 		else
1452 			pool = NULL;
1453 	}
1454 	++*pos;
1455 	return pool;
1456 }
1457 
1458 static void svc_pool_stats_stop(struct seq_file *m, void *p)
1459 {
1460 	struct svc_info *si = m->private;
1461 
1462 	mutex_unlock(si->mutex);
1463 }
1464 
1465 static int svc_pool_stats_show(struct seq_file *m, void *p)
1466 {
1467 	struct svc_pool *pool = p;
1468 
1469 	if (p == SEQ_START_TOKEN) {
1470 		seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n");
1471 		return 0;
1472 	}
1473 
1474 	seq_printf(m, "%u %llu %llu %llu 0\n",
1475 		   pool->sp_id,
1476 		   percpu_counter_sum_positive(&pool->sp_messages_arrived),
1477 		   percpu_counter_sum_positive(&pool->sp_sockets_queued),
1478 		   percpu_counter_sum_positive(&pool->sp_threads_woken));
1479 
1480 	return 0;
1481 }
1482 
1483 static const struct seq_operations svc_pool_stats_seq_ops = {
1484 	.start	= svc_pool_stats_start,
1485 	.next	= svc_pool_stats_next,
1486 	.stop	= svc_pool_stats_stop,
1487 	.show	= svc_pool_stats_show,
1488 };
1489 
1490 int svc_pool_stats_open(struct svc_info *info, struct file *file)
1491 {
1492 	struct seq_file *seq;
1493 	int err;
1494 
1495 	err = seq_open(file, &svc_pool_stats_seq_ops);
1496 	if (err)
1497 		return err;
1498 	seq = file->private_data;
1499 	seq->private = info;
1500 
1501 	return 0;
1502 }
1503 EXPORT_SYMBOL(svc_pool_stats_open);
1504 
1505 /*----------------------------------------------------------------------------*/
1506