xref: /openbmc/linux/net/sunrpc/svc_xprt.c (revision 97d5f2e9)
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/kthread.h>
13 #include <linux/slab.h>
14 #include <net/sock.h>
15 #include <linux/sunrpc/addr.h>
16 #include <linux/sunrpc/stats.h>
17 #include <linux/sunrpc/svc_xprt.h>
18 #include <linux/sunrpc/svcsock.h>
19 #include <linux/sunrpc/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_pool->sp_lock protects most of the fields of that pool.
50  *	svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
51  *	when both need to be taken (rare), svc_serv->sv_lock is first.
52  *	The "service mutex" protects svc_serv->sv_nrthread.
53  *	svc_sock->sk_lock protects the svc_sock->sk_deferred list
54  *             and the ->sk_info_authunix cache.
55  *
56  *	The XPT_BUSY bit in xprt->xpt_flags prevents a transport being
57  *	enqueued multiply. During normal transport processing this bit
58  *	is set by svc_xprt_enqueue and cleared by svc_xprt_received.
59  *	Providers should not manipulate this bit directly.
60  *
61  *	Some flags can be set to certain values at any time
62  *	providing that certain rules are followed:
63  *
64  *	XPT_CONN, XPT_DATA:
65  *		- Can be set or cleared at any time.
66  *		- After a set, svc_xprt_enqueue must be called to enqueue
67  *		  the transport for processing.
68  *		- After a clear, the transport must be read/accepted.
69  *		  If this succeeds, it must be set again.
70  *	XPT_CLOSE:
71  *		- Can set at any time. It is never cleared.
72  *      XPT_DEAD:
73  *		- Can only be set while XPT_BUSY is held which ensures
74  *		  that no other thread will be using the transport or will
75  *		  try to set XPT_DEAD.
76  */
77 int svc_reg_xprt_class(struct svc_xprt_class *xcl)
78 {
79 	struct svc_xprt_class *cl;
80 	int res = -EEXIST;
81 
82 	dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name);
83 
84 	INIT_LIST_HEAD(&xcl->xcl_list);
85 	spin_lock(&svc_xprt_class_lock);
86 	/* Make sure there isn't already a class with the same name */
87 	list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) {
88 		if (strcmp(xcl->xcl_name, cl->xcl_name) == 0)
89 			goto out;
90 	}
91 	list_add_tail(&xcl->xcl_list, &svc_xprt_class_list);
92 	res = 0;
93 out:
94 	spin_unlock(&svc_xprt_class_lock);
95 	return res;
96 }
97 EXPORT_SYMBOL_GPL(svc_reg_xprt_class);
98 
99 void svc_unreg_xprt_class(struct svc_xprt_class *xcl)
100 {
101 	dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name);
102 	spin_lock(&svc_xprt_class_lock);
103 	list_del_init(&xcl->xcl_list);
104 	spin_unlock(&svc_xprt_class_lock);
105 }
106 EXPORT_SYMBOL_GPL(svc_unreg_xprt_class);
107 
108 /**
109  * svc_print_xprts - Format the transport list for printing
110  * @buf: target buffer for formatted address
111  * @maxlen: length of target buffer
112  *
113  * Fills in @buf with a string containing a list of transport names, each name
114  * terminated with '\n'. If the buffer is too small, some entries may be
115  * missing, but it is guaranteed that all lines in the output buffer are
116  * complete.
117  *
118  * Returns positive length of the filled-in string.
119  */
120 int svc_print_xprts(char *buf, int maxlen)
121 {
122 	struct svc_xprt_class *xcl;
123 	char tmpstr[80];
124 	int len = 0;
125 	buf[0] = '\0';
126 
127 	spin_lock(&svc_xprt_class_lock);
128 	list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
129 		int slen;
130 
131 		slen = snprintf(tmpstr, sizeof(tmpstr), "%s %d\n",
132 				xcl->xcl_name, xcl->xcl_max_payload);
133 		if (slen >= sizeof(tmpstr) || len + slen >= maxlen)
134 			break;
135 		len += slen;
136 		strcat(buf, tmpstr);
137 	}
138 	spin_unlock(&svc_xprt_class_lock);
139 
140 	return len;
141 }
142 
143 /**
144  * svc_xprt_deferred_close - Close a transport
145  * @xprt: transport instance
146  *
147  * Used in contexts that need to defer the work of shutting down
148  * the transport to an nfsd thread.
149  */
150 void svc_xprt_deferred_close(struct svc_xprt *xprt)
151 {
152 	if (!test_and_set_bit(XPT_CLOSE, &xprt->xpt_flags))
153 		svc_xprt_enqueue(xprt);
154 }
155 EXPORT_SYMBOL_GPL(svc_xprt_deferred_close);
156 
157 static void svc_xprt_free(struct kref *kref)
158 {
159 	struct svc_xprt *xprt =
160 		container_of(kref, struct svc_xprt, xpt_ref);
161 	struct module *owner = xprt->xpt_class->xcl_owner;
162 	if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags))
163 		svcauth_unix_info_release(xprt);
164 	put_cred(xprt->xpt_cred);
165 	put_net_track(xprt->xpt_net, &xprt->ns_tracker);
166 	/* See comment on corresponding get in xs_setup_bc_tcp(): */
167 	if (xprt->xpt_bc_xprt)
168 		xprt_put(xprt->xpt_bc_xprt);
169 	if (xprt->xpt_bc_xps)
170 		xprt_switch_put(xprt->xpt_bc_xps);
171 	trace_svc_xprt_free(xprt);
172 	xprt->xpt_ops->xpo_free(xprt);
173 	module_put(owner);
174 }
175 
176 void svc_xprt_put(struct svc_xprt *xprt)
177 {
178 	kref_put(&xprt->xpt_ref, svc_xprt_free);
179 }
180 EXPORT_SYMBOL_GPL(svc_xprt_put);
181 
182 /*
183  * Called by transport drivers to initialize the transport independent
184  * portion of the transport instance.
185  */
186 void svc_xprt_init(struct net *net, struct svc_xprt_class *xcl,
187 		   struct svc_xprt *xprt, struct svc_serv *serv)
188 {
189 	memset(xprt, 0, sizeof(*xprt));
190 	xprt->xpt_class = xcl;
191 	xprt->xpt_ops = xcl->xcl_ops;
192 	kref_init(&xprt->xpt_ref);
193 	xprt->xpt_server = serv;
194 	INIT_LIST_HEAD(&xprt->xpt_list);
195 	INIT_LIST_HEAD(&xprt->xpt_ready);
196 	INIT_LIST_HEAD(&xprt->xpt_deferred);
197 	INIT_LIST_HEAD(&xprt->xpt_users);
198 	mutex_init(&xprt->xpt_mutex);
199 	spin_lock_init(&xprt->xpt_lock);
200 	set_bit(XPT_BUSY, &xprt->xpt_flags);
201 	xprt->xpt_net = get_net_track(net, &xprt->ns_tracker, GFP_ATOMIC);
202 	strcpy(xprt->xpt_remotebuf, "uninitialized");
203 }
204 EXPORT_SYMBOL_GPL(svc_xprt_init);
205 
206 static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl,
207 					 struct svc_serv *serv,
208 					 struct net *net,
209 					 const int family,
210 					 const unsigned short port,
211 					 int flags)
212 {
213 	struct sockaddr_in sin = {
214 		.sin_family		= AF_INET,
215 		.sin_addr.s_addr	= htonl(INADDR_ANY),
216 		.sin_port		= htons(port),
217 	};
218 #if IS_ENABLED(CONFIG_IPV6)
219 	struct sockaddr_in6 sin6 = {
220 		.sin6_family		= AF_INET6,
221 		.sin6_addr		= IN6ADDR_ANY_INIT,
222 		.sin6_port		= htons(port),
223 	};
224 #endif
225 	struct svc_xprt *xprt;
226 	struct sockaddr *sap;
227 	size_t len;
228 
229 	switch (family) {
230 	case PF_INET:
231 		sap = (struct sockaddr *)&sin;
232 		len = sizeof(sin);
233 		break;
234 #if IS_ENABLED(CONFIG_IPV6)
235 	case PF_INET6:
236 		sap = (struct sockaddr *)&sin6;
237 		len = sizeof(sin6);
238 		break;
239 #endif
240 	default:
241 		return ERR_PTR(-EAFNOSUPPORT);
242 	}
243 
244 	xprt = xcl->xcl_ops->xpo_create(serv, net, sap, len, flags);
245 	if (IS_ERR(xprt))
246 		trace_svc_xprt_create_err(serv->sv_program->pg_name,
247 					  xcl->xcl_name, sap, len, xprt);
248 	return xprt;
249 }
250 
251 /**
252  * svc_xprt_received - start next receiver thread
253  * @xprt: controlling transport
254  *
255  * The caller must hold the XPT_BUSY bit and must
256  * not thereafter touch transport data.
257  *
258  * Note: XPT_DATA only gets cleared when a read-attempt finds no (or
259  * insufficient) data.
260  */
261 void svc_xprt_received(struct svc_xprt *xprt)
262 {
263 	if (!test_bit(XPT_BUSY, &xprt->xpt_flags)) {
264 		WARN_ONCE(1, "xprt=0x%p already busy!", xprt);
265 		return;
266 	}
267 
268 	/* As soon as we clear busy, the xprt could be closed and
269 	 * 'put', so we need a reference to call svc_xprt_enqueue with:
270 	 */
271 	svc_xprt_get(xprt);
272 	smp_mb__before_atomic();
273 	clear_bit(XPT_BUSY, &xprt->xpt_flags);
274 	svc_xprt_enqueue(xprt);
275 	svc_xprt_put(xprt);
276 }
277 EXPORT_SYMBOL_GPL(svc_xprt_received);
278 
279 void svc_add_new_perm_xprt(struct svc_serv *serv, struct svc_xprt *new)
280 {
281 	clear_bit(XPT_TEMP, &new->xpt_flags);
282 	spin_lock_bh(&serv->sv_lock);
283 	list_add(&new->xpt_list, &serv->sv_permsocks);
284 	spin_unlock_bh(&serv->sv_lock);
285 	svc_xprt_received(new);
286 }
287 
288 static int _svc_xprt_create(struct svc_serv *serv, const char *xprt_name,
289 			    struct net *net, const int family,
290 			    const unsigned short port, int flags,
291 			    const struct cred *cred)
292 {
293 	struct svc_xprt_class *xcl;
294 
295 	spin_lock(&svc_xprt_class_lock);
296 	list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
297 		struct svc_xprt *newxprt;
298 		unsigned short newport;
299 
300 		if (strcmp(xprt_name, xcl->xcl_name))
301 			continue;
302 
303 		if (!try_module_get(xcl->xcl_owner))
304 			goto err;
305 
306 		spin_unlock(&svc_xprt_class_lock);
307 		newxprt = __svc_xpo_create(xcl, serv, net, family, port, flags);
308 		if (IS_ERR(newxprt)) {
309 			module_put(xcl->xcl_owner);
310 			return PTR_ERR(newxprt);
311 		}
312 		newxprt->xpt_cred = get_cred(cred);
313 		svc_add_new_perm_xprt(serv, newxprt);
314 		newport = svc_xprt_local_port(newxprt);
315 		return newport;
316 	}
317  err:
318 	spin_unlock(&svc_xprt_class_lock);
319 	/* This errno is exposed to user space.  Provide a reasonable
320 	 * perror msg for a bad transport. */
321 	return -EPROTONOSUPPORT;
322 }
323 
324 /**
325  * svc_xprt_create - Add a new listener to @serv
326  * @serv: target RPC service
327  * @xprt_name: transport class name
328  * @net: network namespace
329  * @family: network address family
330  * @port: listener port
331  * @flags: SVC_SOCK flags
332  * @cred: credential to bind to this transport
333  *
334  * Return values:
335  *   %0: New listener added successfully
336  *   %-EPROTONOSUPPORT: Requested transport type not supported
337  */
338 int svc_xprt_create(struct svc_serv *serv, const char *xprt_name,
339 		    struct net *net, const int family,
340 		    const unsigned short port, int flags,
341 		    const struct cred *cred)
342 {
343 	int err;
344 
345 	err = _svc_xprt_create(serv, xprt_name, net, family, port, flags, cred);
346 	if (err == -EPROTONOSUPPORT) {
347 		request_module("svc%s", xprt_name);
348 		err = _svc_xprt_create(serv, xprt_name, net, family, port, flags, cred);
349 	}
350 	return err;
351 }
352 EXPORT_SYMBOL_GPL(svc_xprt_create);
353 
354 /*
355  * Copy the local and remote xprt addresses to the rqstp structure
356  */
357 void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt)
358 {
359 	memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen);
360 	rqstp->rq_addrlen = xprt->xpt_remotelen;
361 
362 	/*
363 	 * Destination address in request is needed for binding the
364 	 * source address in RPC replies/callbacks later.
365 	 */
366 	memcpy(&rqstp->rq_daddr, &xprt->xpt_local, xprt->xpt_locallen);
367 	rqstp->rq_daddrlen = xprt->xpt_locallen;
368 }
369 EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs);
370 
371 /**
372  * svc_print_addr - Format rq_addr field for printing
373  * @rqstp: svc_rqst struct containing address to print
374  * @buf: target buffer for formatted address
375  * @len: length of target buffer
376  *
377  */
378 char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
379 {
380 	return __svc_print_addr(svc_addr(rqstp), buf, len);
381 }
382 EXPORT_SYMBOL_GPL(svc_print_addr);
383 
384 static bool svc_xprt_slots_in_range(struct svc_xprt *xprt)
385 {
386 	unsigned int limit = svc_rpc_per_connection_limit;
387 	int nrqsts = atomic_read(&xprt->xpt_nr_rqsts);
388 
389 	return limit == 0 || (nrqsts >= 0 && nrqsts < limit);
390 }
391 
392 static bool svc_xprt_reserve_slot(struct svc_rqst *rqstp, struct svc_xprt *xprt)
393 {
394 	if (!test_bit(RQ_DATA, &rqstp->rq_flags)) {
395 		if (!svc_xprt_slots_in_range(xprt))
396 			return false;
397 		atomic_inc(&xprt->xpt_nr_rqsts);
398 		set_bit(RQ_DATA, &rqstp->rq_flags);
399 	}
400 	return true;
401 }
402 
403 static void svc_xprt_release_slot(struct svc_rqst *rqstp)
404 {
405 	struct svc_xprt	*xprt = rqstp->rq_xprt;
406 	if (test_and_clear_bit(RQ_DATA, &rqstp->rq_flags)) {
407 		atomic_dec(&xprt->xpt_nr_rqsts);
408 		smp_wmb(); /* See smp_rmb() in svc_xprt_ready() */
409 		svc_xprt_enqueue(xprt);
410 	}
411 }
412 
413 static bool svc_xprt_ready(struct svc_xprt *xprt)
414 {
415 	unsigned long xpt_flags;
416 
417 	/*
418 	 * If another cpu has recently updated xpt_flags,
419 	 * sk_sock->flags, xpt_reserved, or xpt_nr_rqsts, we need to
420 	 * know about it; otherwise it's possible that both that cpu and
421 	 * this one could call svc_xprt_enqueue() without either
422 	 * svc_xprt_enqueue() recognizing that the conditions below
423 	 * are satisfied, and we could stall indefinitely:
424 	 */
425 	smp_rmb();
426 	xpt_flags = READ_ONCE(xprt->xpt_flags);
427 
428 	if (xpt_flags & BIT(XPT_BUSY))
429 		return false;
430 	if (xpt_flags & (BIT(XPT_CONN) | BIT(XPT_CLOSE) | BIT(XPT_HANDSHAKE)))
431 		return true;
432 	if (xpt_flags & (BIT(XPT_DATA) | BIT(XPT_DEFERRED))) {
433 		if (xprt->xpt_ops->xpo_has_wspace(xprt) &&
434 		    svc_xprt_slots_in_range(xprt))
435 			return true;
436 		trace_svc_xprt_no_write_space(xprt);
437 		return false;
438 	}
439 	return false;
440 }
441 
442 /**
443  * svc_xprt_enqueue - Queue a transport on an idle nfsd thread
444  * @xprt: transport with data pending
445  *
446  */
447 void svc_xprt_enqueue(struct svc_xprt *xprt)
448 {
449 	struct svc_pool *pool;
450 	struct svc_rqst	*rqstp = NULL;
451 
452 	if (!svc_xprt_ready(xprt))
453 		return;
454 
455 	/* Mark transport as busy. It will remain in this state until
456 	 * the provider calls svc_xprt_received. We update XPT_BUSY
457 	 * atomically because it also guards against trying to enqueue
458 	 * the transport twice.
459 	 */
460 	if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
461 		return;
462 
463 	pool = svc_pool_for_cpu(xprt->xpt_server);
464 
465 	percpu_counter_inc(&pool->sp_sockets_queued);
466 	spin_lock_bh(&pool->sp_lock);
467 	list_add_tail(&xprt->xpt_ready, &pool->sp_sockets);
468 	spin_unlock_bh(&pool->sp_lock);
469 
470 	/* find a thread for this xprt */
471 	rcu_read_lock();
472 	list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) {
473 		if (test_and_set_bit(RQ_BUSY, &rqstp->rq_flags))
474 			continue;
475 		percpu_counter_inc(&pool->sp_threads_woken);
476 		rqstp->rq_qtime = ktime_get();
477 		wake_up_process(rqstp->rq_task);
478 		goto out_unlock;
479 	}
480 	set_bit(SP_CONGESTED, &pool->sp_flags);
481 	rqstp = NULL;
482 out_unlock:
483 	rcu_read_unlock();
484 	trace_svc_xprt_enqueue(xprt, rqstp);
485 }
486 EXPORT_SYMBOL_GPL(svc_xprt_enqueue);
487 
488 /*
489  * Dequeue the first transport, if there is one.
490  */
491 static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool)
492 {
493 	struct svc_xprt	*xprt = NULL;
494 
495 	if (list_empty(&pool->sp_sockets))
496 		goto out;
497 
498 	spin_lock_bh(&pool->sp_lock);
499 	if (likely(!list_empty(&pool->sp_sockets))) {
500 		xprt = list_first_entry(&pool->sp_sockets,
501 					struct svc_xprt, xpt_ready);
502 		list_del_init(&xprt->xpt_ready);
503 		svc_xprt_get(xprt);
504 	}
505 	spin_unlock_bh(&pool->sp_lock);
506 out:
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  * Some svc_serv's will have occasional work to do, even when a xprt is not
577  * waiting to be serviced. This function is there to "kick" a task in one of
578  * those services so that it can wake up and do that work. Note that we only
579  * bother with pool 0 as we don't need to wake up more than one thread for
580  * this purpose.
581  */
582 void svc_wake_up(struct svc_serv *serv)
583 {
584 	struct svc_rqst	*rqstp;
585 	struct svc_pool *pool;
586 
587 	pool = &serv->sv_pools[0];
588 
589 	rcu_read_lock();
590 	list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) {
591 		/* skip any that aren't queued */
592 		if (test_bit(RQ_BUSY, &rqstp->rq_flags))
593 			continue;
594 		rcu_read_unlock();
595 		wake_up_process(rqstp->rq_task);
596 		trace_svc_wake_up(rqstp->rq_task->pid);
597 		return;
598 	}
599 	rcu_read_unlock();
600 
601 	/* No free entries available */
602 	set_bit(SP_TASK_PENDING, &pool->sp_flags);
603 	smp_wmb();
604 	trace_svc_wake_up(0);
605 }
606 EXPORT_SYMBOL_GPL(svc_wake_up);
607 
608 int svc_port_is_privileged(struct sockaddr *sin)
609 {
610 	switch (sin->sa_family) {
611 	case AF_INET:
612 		return ntohs(((struct sockaddr_in *)sin)->sin_port)
613 			< PROT_SOCK;
614 	case AF_INET6:
615 		return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
616 			< PROT_SOCK;
617 	default:
618 		return 0;
619 	}
620 }
621 
622 /*
623  * Make sure that we don't have too many active connections. If we have,
624  * something must be dropped. It's not clear what will happen if we allow
625  * "too many" connections, but when dealing with network-facing software,
626  * we have to code defensively. Here we do that by imposing hard limits.
627  *
628  * There's no point in trying to do random drop here for DoS
629  * prevention. The NFS clients does 1 reconnect in 15 seconds. An
630  * attacker can easily beat that.
631  *
632  * The only somewhat efficient mechanism would be if drop old
633  * connections from the same IP first. But right now we don't even
634  * record the client IP in svc_sock.
635  *
636  * single-threaded services that expect a lot of clients will probably
637  * need to set sv_maxconn to override the default value which is based
638  * on the number of threads
639  */
640 static void svc_check_conn_limits(struct svc_serv *serv)
641 {
642 	unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn :
643 				(serv->sv_nrthreads+3) * 20;
644 
645 	if (serv->sv_tmpcnt > limit) {
646 		struct svc_xprt *xprt = NULL;
647 		spin_lock_bh(&serv->sv_lock);
648 		if (!list_empty(&serv->sv_tempsocks)) {
649 			/* Try to help the admin */
650 			net_notice_ratelimited("%s: too many open connections, consider increasing the %s\n",
651 					       serv->sv_name, serv->sv_maxconn ?
652 					       "max number of connections" :
653 					       "number of threads");
654 			/*
655 			 * Always select the oldest connection. It's not fair,
656 			 * but so is life
657 			 */
658 			xprt = list_entry(serv->sv_tempsocks.prev,
659 					  struct svc_xprt,
660 					  xpt_list);
661 			set_bit(XPT_CLOSE, &xprt->xpt_flags);
662 			svc_xprt_get(xprt);
663 		}
664 		spin_unlock_bh(&serv->sv_lock);
665 
666 		if (xprt) {
667 			svc_xprt_enqueue(xprt);
668 			svc_xprt_put(xprt);
669 		}
670 	}
671 }
672 
673 static int svc_alloc_arg(struct svc_rqst *rqstp)
674 {
675 	struct svc_serv *serv = rqstp->rq_server;
676 	struct xdr_buf *arg = &rqstp->rq_arg;
677 	unsigned long pages, filled, ret;
678 
679 	pages = (serv->sv_max_mesg + 2 * PAGE_SIZE) >> PAGE_SHIFT;
680 	if (pages > RPCSVC_MAXPAGES) {
681 		pr_warn_once("svc: warning: pages=%lu > RPCSVC_MAXPAGES=%lu\n",
682 			     pages, RPCSVC_MAXPAGES);
683 		/* use as many pages as possible */
684 		pages = RPCSVC_MAXPAGES;
685 	}
686 
687 	for (filled = 0; filled < pages; filled = ret) {
688 		ret = alloc_pages_bulk_array(GFP_KERNEL, pages,
689 					     rqstp->rq_pages);
690 		if (ret > filled)
691 			/* Made progress, don't sleep yet */
692 			continue;
693 
694 		set_current_state(TASK_INTERRUPTIBLE);
695 		if (signalled() || kthread_should_stop()) {
696 			set_current_state(TASK_RUNNING);
697 			return -EINTR;
698 		}
699 		trace_svc_alloc_arg_err(pages, ret);
700 		memalloc_retry_wait(GFP_KERNEL);
701 	}
702 	rqstp->rq_page_end = &rqstp->rq_pages[pages];
703 	rqstp->rq_pages[pages] = NULL; /* this might be seen in nfsd_splice_actor() */
704 
705 	/* Make arg->head point to first page and arg->pages point to rest */
706 	arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
707 	arg->head[0].iov_len = PAGE_SIZE;
708 	arg->pages = rqstp->rq_pages + 1;
709 	arg->page_base = 0;
710 	/* save at least one page for response */
711 	arg->page_len = (pages-2)*PAGE_SIZE;
712 	arg->len = (pages-1)*PAGE_SIZE;
713 	arg->tail[0].iov_len = 0;
714 
715 	rqstp->rq_xid = xdr_zero;
716 	return 0;
717 }
718 
719 static bool
720 rqst_should_sleep(struct svc_rqst *rqstp)
721 {
722 	struct svc_pool		*pool = rqstp->rq_pool;
723 
724 	/* did someone call svc_wake_up? */
725 	if (test_and_clear_bit(SP_TASK_PENDING, &pool->sp_flags))
726 		return false;
727 
728 	/* was a socket queued? */
729 	if (!list_empty(&pool->sp_sockets))
730 		return false;
731 
732 	/* are we shutting down? */
733 	if (signalled() || kthread_should_stop())
734 		return false;
735 
736 	/* are we freezing? */
737 	if (freezing(current))
738 		return false;
739 
740 	return true;
741 }
742 
743 static struct svc_xprt *svc_get_next_xprt(struct svc_rqst *rqstp, long timeout)
744 {
745 	struct svc_pool		*pool = rqstp->rq_pool;
746 	long			time_left = 0;
747 
748 	/* rq_xprt should be clear on entry */
749 	WARN_ON_ONCE(rqstp->rq_xprt);
750 
751 	rqstp->rq_xprt = svc_xprt_dequeue(pool);
752 	if (rqstp->rq_xprt)
753 		goto out_found;
754 
755 	/*
756 	 * We have to be able to interrupt this wait
757 	 * to bring down the daemons ...
758 	 */
759 	set_current_state(TASK_INTERRUPTIBLE);
760 	smp_mb__before_atomic();
761 	clear_bit(SP_CONGESTED, &pool->sp_flags);
762 	clear_bit(RQ_BUSY, &rqstp->rq_flags);
763 	smp_mb__after_atomic();
764 
765 	if (likely(rqst_should_sleep(rqstp)))
766 		time_left = schedule_timeout(timeout);
767 	else
768 		__set_current_state(TASK_RUNNING);
769 
770 	try_to_freeze();
771 
772 	set_bit(RQ_BUSY, &rqstp->rq_flags);
773 	smp_mb__after_atomic();
774 	rqstp->rq_xprt = svc_xprt_dequeue(pool);
775 	if (rqstp->rq_xprt)
776 		goto out_found;
777 
778 	if (!time_left)
779 		percpu_counter_inc(&pool->sp_threads_timedout);
780 
781 	if (signalled() || kthread_should_stop())
782 		return ERR_PTR(-EINTR);
783 	return ERR_PTR(-EAGAIN);
784 out_found:
785 	/* Normally we will wait up to 5 seconds for any required
786 	 * cache information to be provided.
787 	 */
788 	if (!test_bit(SP_CONGESTED, &pool->sp_flags))
789 		rqstp->rq_chandle.thread_wait = 5*HZ;
790 	else
791 		rqstp->rq_chandle.thread_wait = 1*HZ;
792 	trace_svc_xprt_dequeue(rqstp);
793 	return rqstp->rq_xprt;
794 }
795 
796 static void svc_add_new_temp_xprt(struct svc_serv *serv, struct svc_xprt *newxpt)
797 {
798 	spin_lock_bh(&serv->sv_lock);
799 	set_bit(XPT_TEMP, &newxpt->xpt_flags);
800 	list_add(&newxpt->xpt_list, &serv->sv_tempsocks);
801 	serv->sv_tmpcnt++;
802 	if (serv->sv_temptimer.function == NULL) {
803 		/* setup timer to age temp transports */
804 		serv->sv_temptimer.function = svc_age_temp_xprts;
805 		mod_timer(&serv->sv_temptimer,
806 			  jiffies + svc_conn_age_period * HZ);
807 	}
808 	spin_unlock_bh(&serv->sv_lock);
809 	svc_xprt_received(newxpt);
810 }
811 
812 static int svc_handle_xprt(struct svc_rqst *rqstp, struct svc_xprt *xprt)
813 {
814 	struct svc_serv *serv = rqstp->rq_server;
815 	int len = 0;
816 
817 	if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) {
818 		if (test_and_clear_bit(XPT_KILL_TEMP, &xprt->xpt_flags))
819 			xprt->xpt_ops->xpo_kill_temp_xprt(xprt);
820 		svc_delete_xprt(xprt);
821 		/* Leave XPT_BUSY set on the dead xprt: */
822 		goto out;
823 	}
824 	if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
825 		struct svc_xprt *newxpt;
826 		/*
827 		 * We know this module_get will succeed because the
828 		 * listener holds a reference too
829 		 */
830 		__module_get(xprt->xpt_class->xcl_owner);
831 		svc_check_conn_limits(xprt->xpt_server);
832 		newxpt = xprt->xpt_ops->xpo_accept(xprt);
833 		if (newxpt) {
834 			newxpt->xpt_cred = get_cred(xprt->xpt_cred);
835 			svc_add_new_temp_xprt(serv, newxpt);
836 			trace_svc_xprt_accept(newxpt, serv->sv_name);
837 		} else {
838 			module_put(xprt->xpt_class->xcl_owner);
839 		}
840 		svc_xprt_received(xprt);
841 	} else if (test_bit(XPT_HANDSHAKE, &xprt->xpt_flags)) {
842 		xprt->xpt_ops->xpo_handshake(xprt);
843 		svc_xprt_received(xprt);
844 	} else if (svc_xprt_reserve_slot(rqstp, xprt)) {
845 		/* XPT_DATA|XPT_DEFERRED case: */
846 		dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n",
847 			rqstp, rqstp->rq_pool->sp_id, xprt,
848 			kref_read(&xprt->xpt_ref));
849 		rqstp->rq_deferred = svc_deferred_dequeue(xprt);
850 		if (rqstp->rq_deferred)
851 			len = svc_deferred_recv(rqstp);
852 		else
853 			len = xprt->xpt_ops->xpo_recvfrom(rqstp);
854 		rqstp->rq_stime = ktime_get();
855 		rqstp->rq_reserved = serv->sv_max_mesg;
856 		atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
857 	} else
858 		svc_xprt_received(xprt);
859 
860 out:
861 	return len;
862 }
863 
864 /*
865  * Receive the next request on any transport.  This code is carefully
866  * organised not to touch any cachelines in the shared svc_serv
867  * structure, only cachelines in the local svc_pool.
868  */
869 int svc_recv(struct svc_rqst *rqstp, long timeout)
870 {
871 	struct svc_xprt		*xprt = NULL;
872 	struct svc_serv		*serv = rqstp->rq_server;
873 	int			len, err;
874 
875 	err = svc_alloc_arg(rqstp);
876 	if (err)
877 		goto out;
878 
879 	try_to_freeze();
880 	cond_resched();
881 	err = -EINTR;
882 	if (signalled() || kthread_should_stop())
883 		goto out;
884 
885 	xprt = svc_get_next_xprt(rqstp, timeout);
886 	if (IS_ERR(xprt)) {
887 		err = PTR_ERR(xprt);
888 		goto out;
889 	}
890 
891 	len = svc_handle_xprt(rqstp, xprt);
892 
893 	/* No data, incomplete (TCP) read, or accept() */
894 	err = -EAGAIN;
895 	if (len <= 0)
896 		goto out_release;
897 	trace_svc_xdr_recvfrom(&rqstp->rq_arg);
898 
899 	clear_bit(XPT_OLD, &xprt->xpt_flags);
900 
901 	rqstp->rq_chandle.defer = svc_defer;
902 
903 	if (serv->sv_stats)
904 		serv->sv_stats->netcnt++;
905 	return len;
906 out_release:
907 	rqstp->rq_res.len = 0;
908 	svc_xprt_release(rqstp);
909 out:
910 	return err;
911 }
912 EXPORT_SYMBOL_GPL(svc_recv);
913 
914 /*
915  * Drop request
916  */
917 void svc_drop(struct svc_rqst *rqstp)
918 {
919 	trace_svc_drop(rqstp);
920 	svc_xprt_release(rqstp);
921 }
922 EXPORT_SYMBOL_GPL(svc_drop);
923 
924 /**
925  * svc_send - Return reply to client
926  * @rqstp: RPC transaction context
927  *
928  */
929 void svc_send(struct svc_rqst *rqstp)
930 {
931 	struct svc_xprt	*xprt;
932 	struct xdr_buf	*xb;
933 	int status;
934 
935 	xprt = rqstp->rq_xprt;
936 	if (!xprt)
937 		return;
938 
939 	/* calculate over-all length */
940 	xb = &rqstp->rq_res;
941 	xb->len = xb->head[0].iov_len +
942 		xb->page_len +
943 		xb->tail[0].iov_len;
944 	trace_svc_xdr_sendto(rqstp->rq_xid, xb);
945 	trace_svc_stats_latency(rqstp);
946 
947 	status = xprt->xpt_ops->xpo_sendto(rqstp);
948 
949 	trace_svc_send(rqstp, status);
950 	svc_xprt_release(rqstp);
951 }
952 
953 /*
954  * Timer function to close old temporary transports, using
955  * a mark-and-sweep algorithm.
956  */
957 static void svc_age_temp_xprts(struct timer_list *t)
958 {
959 	struct svc_serv *serv = from_timer(serv, t, sv_temptimer);
960 	struct svc_xprt *xprt;
961 	struct list_head *le, *next;
962 
963 	dprintk("svc_age_temp_xprts\n");
964 
965 	if (!spin_trylock_bh(&serv->sv_lock)) {
966 		/* busy, try again 1 sec later */
967 		dprintk("svc_age_temp_xprts: busy\n");
968 		mod_timer(&serv->sv_temptimer, jiffies + HZ);
969 		return;
970 	}
971 
972 	list_for_each_safe(le, next, &serv->sv_tempsocks) {
973 		xprt = list_entry(le, struct svc_xprt, xpt_list);
974 
975 		/* First time through, just mark it OLD. Second time
976 		 * through, close it. */
977 		if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags))
978 			continue;
979 		if (kref_read(&xprt->xpt_ref) > 1 ||
980 		    test_bit(XPT_BUSY, &xprt->xpt_flags))
981 			continue;
982 		list_del_init(le);
983 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
984 		dprintk("queuing xprt %p for closing\n", xprt);
985 
986 		/* a thread will dequeue and close it soon */
987 		svc_xprt_enqueue(xprt);
988 	}
989 	spin_unlock_bh(&serv->sv_lock);
990 
991 	mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
992 }
993 
994 /* Close temporary transports whose xpt_local matches server_addr immediately
995  * instead of waiting for them to be picked up by the timer.
996  *
997  * This is meant to be called from a notifier_block that runs when an ip
998  * address is deleted.
999  */
1000 void svc_age_temp_xprts_now(struct svc_serv *serv, struct sockaddr *server_addr)
1001 {
1002 	struct svc_xprt *xprt;
1003 	struct list_head *le, *next;
1004 	LIST_HEAD(to_be_closed);
1005 
1006 	spin_lock_bh(&serv->sv_lock);
1007 	list_for_each_safe(le, next, &serv->sv_tempsocks) {
1008 		xprt = list_entry(le, struct svc_xprt, xpt_list);
1009 		if (rpc_cmp_addr(server_addr, (struct sockaddr *)
1010 				&xprt->xpt_local)) {
1011 			dprintk("svc_age_temp_xprts_now: found %p\n", xprt);
1012 			list_move(le, &to_be_closed);
1013 		}
1014 	}
1015 	spin_unlock_bh(&serv->sv_lock);
1016 
1017 	while (!list_empty(&to_be_closed)) {
1018 		le = to_be_closed.next;
1019 		list_del_init(le);
1020 		xprt = list_entry(le, struct svc_xprt, xpt_list);
1021 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
1022 		set_bit(XPT_KILL_TEMP, &xprt->xpt_flags);
1023 		dprintk("svc_age_temp_xprts_now: queuing xprt %p for closing\n",
1024 				xprt);
1025 		svc_xprt_enqueue(xprt);
1026 	}
1027 }
1028 EXPORT_SYMBOL_GPL(svc_age_temp_xprts_now);
1029 
1030 static void call_xpt_users(struct svc_xprt *xprt)
1031 {
1032 	struct svc_xpt_user *u;
1033 
1034 	spin_lock(&xprt->xpt_lock);
1035 	while (!list_empty(&xprt->xpt_users)) {
1036 		u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list);
1037 		list_del_init(&u->list);
1038 		u->callback(u);
1039 	}
1040 	spin_unlock(&xprt->xpt_lock);
1041 }
1042 
1043 /*
1044  * Remove a dead transport
1045  */
1046 static void svc_delete_xprt(struct svc_xprt *xprt)
1047 {
1048 	struct svc_serv	*serv = xprt->xpt_server;
1049 	struct svc_deferred_req *dr;
1050 
1051 	if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags))
1052 		return;
1053 
1054 	trace_svc_xprt_detach(xprt);
1055 	xprt->xpt_ops->xpo_detach(xprt);
1056 	if (xprt->xpt_bc_xprt)
1057 		xprt->xpt_bc_xprt->ops->close(xprt->xpt_bc_xprt);
1058 
1059 	spin_lock_bh(&serv->sv_lock);
1060 	list_del_init(&xprt->xpt_list);
1061 	WARN_ON_ONCE(!list_empty(&xprt->xpt_ready));
1062 	if (test_bit(XPT_TEMP, &xprt->xpt_flags))
1063 		serv->sv_tmpcnt--;
1064 	spin_unlock_bh(&serv->sv_lock);
1065 
1066 	while ((dr = svc_deferred_dequeue(xprt)) != NULL)
1067 		free_deferred(xprt, dr);
1068 
1069 	call_xpt_users(xprt);
1070 	svc_xprt_put(xprt);
1071 }
1072 
1073 /**
1074  * svc_xprt_close - Close a client connection
1075  * @xprt: transport to disconnect
1076  *
1077  */
1078 void svc_xprt_close(struct svc_xprt *xprt)
1079 {
1080 	trace_svc_xprt_close(xprt);
1081 	set_bit(XPT_CLOSE, &xprt->xpt_flags);
1082 	if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
1083 		/* someone else will have to effect the close */
1084 		return;
1085 	/*
1086 	 * We expect svc_close_xprt() to work even when no threads are
1087 	 * running (e.g., while configuring the server before starting
1088 	 * any threads), so if the transport isn't busy, we delete
1089 	 * it ourself:
1090 	 */
1091 	svc_delete_xprt(xprt);
1092 }
1093 EXPORT_SYMBOL_GPL(svc_xprt_close);
1094 
1095 static int svc_close_list(struct svc_serv *serv, struct list_head *xprt_list, struct net *net)
1096 {
1097 	struct svc_xprt *xprt;
1098 	int ret = 0;
1099 
1100 	spin_lock_bh(&serv->sv_lock);
1101 	list_for_each_entry(xprt, xprt_list, xpt_list) {
1102 		if (xprt->xpt_net != net)
1103 			continue;
1104 		ret++;
1105 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
1106 		svc_xprt_enqueue(xprt);
1107 	}
1108 	spin_unlock_bh(&serv->sv_lock);
1109 	return ret;
1110 }
1111 
1112 static struct svc_xprt *svc_dequeue_net(struct svc_serv *serv, struct net *net)
1113 {
1114 	struct svc_pool *pool;
1115 	struct svc_xprt *xprt;
1116 	struct svc_xprt *tmp;
1117 	int i;
1118 
1119 	for (i = 0; i < serv->sv_nrpools; i++) {
1120 		pool = &serv->sv_pools[i];
1121 
1122 		spin_lock_bh(&pool->sp_lock);
1123 		list_for_each_entry_safe(xprt, tmp, &pool->sp_sockets, xpt_ready) {
1124 			if (xprt->xpt_net != net)
1125 				continue;
1126 			list_del_init(&xprt->xpt_ready);
1127 			spin_unlock_bh(&pool->sp_lock);
1128 			return xprt;
1129 		}
1130 		spin_unlock_bh(&pool->sp_lock);
1131 	}
1132 	return NULL;
1133 }
1134 
1135 static void svc_clean_up_xprts(struct svc_serv *serv, struct net *net)
1136 {
1137 	struct svc_xprt *xprt;
1138 
1139 	while ((xprt = svc_dequeue_net(serv, net))) {
1140 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
1141 		svc_delete_xprt(xprt);
1142 	}
1143 }
1144 
1145 /**
1146  * svc_xprt_destroy_all - Destroy transports associated with @serv
1147  * @serv: RPC service to be shut down
1148  * @net: target network namespace
1149  *
1150  * Server threads may still be running (especially in the case where the
1151  * service is still running in other network namespaces).
1152  *
1153  * So we shut down sockets the same way we would on a running server, by
1154  * setting XPT_CLOSE, enqueuing, and letting a thread pick it up to do
1155  * the close.  In the case there are no such other threads,
1156  * threads running, svc_clean_up_xprts() does a simple version of a
1157  * server's main event loop, and in the case where there are other
1158  * threads, we may need to wait a little while and then check again to
1159  * see if they're done.
1160  */
1161 void svc_xprt_destroy_all(struct svc_serv *serv, struct net *net)
1162 {
1163 	int delay = 0;
1164 
1165 	while (svc_close_list(serv, &serv->sv_permsocks, net) +
1166 	       svc_close_list(serv, &serv->sv_tempsocks, net)) {
1167 
1168 		svc_clean_up_xprts(serv, net);
1169 		msleep(delay++);
1170 	}
1171 }
1172 EXPORT_SYMBOL_GPL(svc_xprt_destroy_all);
1173 
1174 /*
1175  * Handle defer and revisit of requests
1176  */
1177 
1178 static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1179 {
1180 	struct svc_deferred_req *dr =
1181 		container_of(dreq, struct svc_deferred_req, handle);
1182 	struct svc_xprt *xprt = dr->xprt;
1183 
1184 	spin_lock(&xprt->xpt_lock);
1185 	set_bit(XPT_DEFERRED, &xprt->xpt_flags);
1186 	if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) {
1187 		spin_unlock(&xprt->xpt_lock);
1188 		trace_svc_defer_drop(dr);
1189 		free_deferred(xprt, dr);
1190 		svc_xprt_put(xprt);
1191 		return;
1192 	}
1193 	dr->xprt = NULL;
1194 	list_add(&dr->handle.recent, &xprt->xpt_deferred);
1195 	spin_unlock(&xprt->xpt_lock);
1196 	trace_svc_defer_queue(dr);
1197 	svc_xprt_enqueue(xprt);
1198 	svc_xprt_put(xprt);
1199 }
1200 
1201 /*
1202  * Save the request off for later processing. The request buffer looks
1203  * like this:
1204  *
1205  * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
1206  *
1207  * This code can only handle requests that consist of an xprt-header
1208  * and rpc-header.
1209  */
1210 static struct cache_deferred_req *svc_defer(struct cache_req *req)
1211 {
1212 	struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
1213 	struct svc_deferred_req *dr;
1214 
1215 	if (rqstp->rq_arg.page_len || !test_bit(RQ_USEDEFERRAL, &rqstp->rq_flags))
1216 		return NULL; /* if more than a page, give up FIXME */
1217 	if (rqstp->rq_deferred) {
1218 		dr = rqstp->rq_deferred;
1219 		rqstp->rq_deferred = NULL;
1220 	} else {
1221 		size_t skip;
1222 		size_t size;
1223 		/* FIXME maybe discard if size too large */
1224 		size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
1225 		dr = kmalloc(size, GFP_KERNEL);
1226 		if (dr == NULL)
1227 			return NULL;
1228 
1229 		dr->handle.owner = rqstp->rq_server;
1230 		dr->prot = rqstp->rq_prot;
1231 		memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
1232 		dr->addrlen = rqstp->rq_addrlen;
1233 		dr->daddr = rqstp->rq_daddr;
1234 		dr->argslen = rqstp->rq_arg.len >> 2;
1235 
1236 		/* back up head to the start of the buffer and copy */
1237 		skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1238 		memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
1239 		       dr->argslen << 2);
1240 	}
1241 	dr->xprt_ctxt = rqstp->rq_xprt_ctxt;
1242 	rqstp->rq_xprt_ctxt = NULL;
1243 	trace_svc_defer(rqstp);
1244 	svc_xprt_get(rqstp->rq_xprt);
1245 	dr->xprt = rqstp->rq_xprt;
1246 	set_bit(RQ_DROPME, &rqstp->rq_flags);
1247 
1248 	dr->handle.revisit = svc_revisit;
1249 	return &dr->handle;
1250 }
1251 
1252 /*
1253  * recv data from a deferred request into an active one
1254  */
1255 static noinline int svc_deferred_recv(struct svc_rqst *rqstp)
1256 {
1257 	struct svc_deferred_req *dr = rqstp->rq_deferred;
1258 
1259 	trace_svc_defer_recv(dr);
1260 
1261 	/* setup iov_base past transport header */
1262 	rqstp->rq_arg.head[0].iov_base = dr->args;
1263 	/* The iov_len does not include the transport header bytes */
1264 	rqstp->rq_arg.head[0].iov_len = dr->argslen << 2;
1265 	rqstp->rq_arg.page_len = 0;
1266 	/* The rq_arg.len includes the transport header bytes */
1267 	rqstp->rq_arg.len     = dr->argslen << 2;
1268 	rqstp->rq_prot        = dr->prot;
1269 	memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
1270 	rqstp->rq_addrlen     = dr->addrlen;
1271 	/* Save off transport header len in case we get deferred again */
1272 	rqstp->rq_daddr       = dr->daddr;
1273 	rqstp->rq_respages    = rqstp->rq_pages;
1274 	rqstp->rq_xprt_ctxt   = dr->xprt_ctxt;
1275 
1276 	dr->xprt_ctxt = NULL;
1277 	svc_xprt_received(rqstp->rq_xprt);
1278 	return dr->argslen << 2;
1279 }
1280 
1281 
1282 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt)
1283 {
1284 	struct svc_deferred_req *dr = NULL;
1285 
1286 	if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags))
1287 		return NULL;
1288 	spin_lock(&xprt->xpt_lock);
1289 	if (!list_empty(&xprt->xpt_deferred)) {
1290 		dr = list_entry(xprt->xpt_deferred.next,
1291 				struct svc_deferred_req,
1292 				handle.recent);
1293 		list_del_init(&dr->handle.recent);
1294 	} else
1295 		clear_bit(XPT_DEFERRED, &xprt->xpt_flags);
1296 	spin_unlock(&xprt->xpt_lock);
1297 	return dr;
1298 }
1299 
1300 /**
1301  * svc_find_xprt - find an RPC transport instance
1302  * @serv: pointer to svc_serv to search
1303  * @xcl_name: C string containing transport's class name
1304  * @net: owner net pointer
1305  * @af: Address family of transport's local address
1306  * @port: transport's IP port number
1307  *
1308  * Return the transport instance pointer for the endpoint accepting
1309  * connections/peer traffic from the specified transport class,
1310  * address family and port.
1311  *
1312  * Specifying 0 for the address family or port is effectively a
1313  * wild-card, and will result in matching the first transport in the
1314  * service's list that has a matching class name.
1315  */
1316 struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name,
1317 			       struct net *net, const sa_family_t af,
1318 			       const unsigned short port)
1319 {
1320 	struct svc_xprt *xprt;
1321 	struct svc_xprt *found = NULL;
1322 
1323 	/* Sanity check the args */
1324 	if (serv == NULL || xcl_name == NULL)
1325 		return found;
1326 
1327 	spin_lock_bh(&serv->sv_lock);
1328 	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1329 		if (xprt->xpt_net != net)
1330 			continue;
1331 		if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1332 			continue;
1333 		if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family)
1334 			continue;
1335 		if (port != 0 && port != svc_xprt_local_port(xprt))
1336 			continue;
1337 		found = xprt;
1338 		svc_xprt_get(xprt);
1339 		break;
1340 	}
1341 	spin_unlock_bh(&serv->sv_lock);
1342 	return found;
1343 }
1344 EXPORT_SYMBOL_GPL(svc_find_xprt);
1345 
1346 static int svc_one_xprt_name(const struct svc_xprt *xprt,
1347 			     char *pos, int remaining)
1348 {
1349 	int len;
1350 
1351 	len = snprintf(pos, remaining, "%s %u\n",
1352 			xprt->xpt_class->xcl_name,
1353 			svc_xprt_local_port(xprt));
1354 	if (len >= remaining)
1355 		return -ENAMETOOLONG;
1356 	return len;
1357 }
1358 
1359 /**
1360  * svc_xprt_names - format a buffer with a list of transport names
1361  * @serv: pointer to an RPC service
1362  * @buf: pointer to a buffer to be filled in
1363  * @buflen: length of buffer to be filled in
1364  *
1365  * Fills in @buf with a string containing a list of transport names,
1366  * each name terminated with '\n'.
1367  *
1368  * Returns positive length of the filled-in string on success; otherwise
1369  * a negative errno value is returned if an error occurs.
1370  */
1371 int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen)
1372 {
1373 	struct svc_xprt *xprt;
1374 	int len, totlen;
1375 	char *pos;
1376 
1377 	/* Sanity check args */
1378 	if (!serv)
1379 		return 0;
1380 
1381 	spin_lock_bh(&serv->sv_lock);
1382 
1383 	pos = buf;
1384 	totlen = 0;
1385 	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1386 		len = svc_one_xprt_name(xprt, pos, buflen - totlen);
1387 		if (len < 0) {
1388 			*buf = '\0';
1389 			totlen = len;
1390 		}
1391 		if (len <= 0)
1392 			break;
1393 
1394 		pos += len;
1395 		totlen += len;
1396 	}
1397 
1398 	spin_unlock_bh(&serv->sv_lock);
1399 	return totlen;
1400 }
1401 EXPORT_SYMBOL_GPL(svc_xprt_names);
1402 
1403 
1404 /*----------------------------------------------------------------------------*/
1405 
1406 static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos)
1407 {
1408 	unsigned int pidx = (unsigned int)*pos;
1409 	struct svc_serv *serv = m->private;
1410 
1411 	dprintk("svc_pool_stats_start, *pidx=%u\n", pidx);
1412 
1413 	if (!pidx)
1414 		return SEQ_START_TOKEN;
1415 	return (pidx > serv->sv_nrpools ? NULL : &serv->sv_pools[pidx-1]);
1416 }
1417 
1418 static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos)
1419 {
1420 	struct svc_pool *pool = p;
1421 	struct svc_serv *serv = m->private;
1422 
1423 	dprintk("svc_pool_stats_next, *pos=%llu\n", *pos);
1424 
1425 	if (p == SEQ_START_TOKEN) {
1426 		pool = &serv->sv_pools[0];
1427 	} else {
1428 		unsigned int pidx = (pool - &serv->sv_pools[0]);
1429 		if (pidx < serv->sv_nrpools-1)
1430 			pool = &serv->sv_pools[pidx+1];
1431 		else
1432 			pool = NULL;
1433 	}
1434 	++*pos;
1435 	return pool;
1436 }
1437 
1438 static void svc_pool_stats_stop(struct seq_file *m, void *p)
1439 {
1440 }
1441 
1442 static int svc_pool_stats_show(struct seq_file *m, void *p)
1443 {
1444 	struct svc_pool *pool = p;
1445 
1446 	if (p == SEQ_START_TOKEN) {
1447 		seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n");
1448 		return 0;
1449 	}
1450 
1451 	seq_printf(m, "%u %llu %llu %llu %llu\n",
1452 		pool->sp_id,
1453 		percpu_counter_sum_positive(&pool->sp_sockets_queued),
1454 		percpu_counter_sum_positive(&pool->sp_sockets_queued),
1455 		percpu_counter_sum_positive(&pool->sp_threads_woken),
1456 		percpu_counter_sum_positive(&pool->sp_threads_timedout));
1457 
1458 	return 0;
1459 }
1460 
1461 static const struct seq_operations svc_pool_stats_seq_ops = {
1462 	.start	= svc_pool_stats_start,
1463 	.next	= svc_pool_stats_next,
1464 	.stop	= svc_pool_stats_stop,
1465 	.show	= svc_pool_stats_show,
1466 };
1467 
1468 int svc_pool_stats_open(struct svc_serv *serv, struct file *file)
1469 {
1470 	int err;
1471 
1472 	err = seq_open(file, &svc_pool_stats_seq_ops);
1473 	if (!err)
1474 		((struct seq_file *) file->private_data)->private = serv;
1475 	return err;
1476 }
1477 EXPORT_SYMBOL(svc_pool_stats_open);
1478 
1479 /*----------------------------------------------------------------------------*/
1480