xref: /openbmc/linux/net/sunrpc/svc_xprt.c (revision 2a598d0b)
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 
78 /**
79  * svc_reg_xprt_class - Register a server-side RPC transport class
80  * @xcl: New transport class to be registered
81  *
82  * Returns zero on success; otherwise a negative errno is returned.
83  */
84 int svc_reg_xprt_class(struct svc_xprt_class *xcl)
85 {
86 	struct svc_xprt_class *cl;
87 	int res = -EEXIST;
88 
89 	INIT_LIST_HEAD(&xcl->xcl_list);
90 	spin_lock(&svc_xprt_class_lock);
91 	/* Make sure there isn't already a class with the same name */
92 	list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) {
93 		if (strcmp(xcl->xcl_name, cl->xcl_name) == 0)
94 			goto out;
95 	}
96 	list_add_tail(&xcl->xcl_list, &svc_xprt_class_list);
97 	res = 0;
98 out:
99 	spin_unlock(&svc_xprt_class_lock);
100 	return res;
101 }
102 EXPORT_SYMBOL_GPL(svc_reg_xprt_class);
103 
104 /**
105  * svc_unreg_xprt_class - Unregister a server-side RPC transport class
106  * @xcl: Transport class to be unregistered
107  *
108  */
109 void svc_unreg_xprt_class(struct svc_xprt_class *xcl)
110 {
111 	spin_lock(&svc_xprt_class_lock);
112 	list_del_init(&xcl->xcl_list);
113 	spin_unlock(&svc_xprt_class_lock);
114 }
115 EXPORT_SYMBOL_GPL(svc_unreg_xprt_class);
116 
117 /**
118  * svc_print_xprts - Format the transport list for printing
119  * @buf: target buffer for formatted address
120  * @maxlen: length of target buffer
121  *
122  * Fills in @buf with a string containing a list of transport names, each name
123  * terminated with '\n'. If the buffer is too small, some entries may be
124  * missing, but it is guaranteed that all lines in the output buffer are
125  * complete.
126  *
127  * Returns positive length of the filled-in string.
128  */
129 int svc_print_xprts(char *buf, int maxlen)
130 {
131 	struct svc_xprt_class *xcl;
132 	char tmpstr[80];
133 	int len = 0;
134 	buf[0] = '\0';
135 
136 	spin_lock(&svc_xprt_class_lock);
137 	list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
138 		int slen;
139 
140 		slen = snprintf(tmpstr, sizeof(tmpstr), "%s %d\n",
141 				xcl->xcl_name, xcl->xcl_max_payload);
142 		if (slen >= sizeof(tmpstr) || len + slen >= maxlen)
143 			break;
144 		len += slen;
145 		strcat(buf, tmpstr);
146 	}
147 	spin_unlock(&svc_xprt_class_lock);
148 
149 	return len;
150 }
151 
152 /**
153  * svc_xprt_deferred_close - Close a transport
154  * @xprt: transport instance
155  *
156  * Used in contexts that need to defer the work of shutting down
157  * the transport to an nfsd thread.
158  */
159 void svc_xprt_deferred_close(struct svc_xprt *xprt)
160 {
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_ready);
205 	INIT_LIST_HEAD(&xprt->xpt_deferred);
206 	INIT_LIST_HEAD(&xprt->xpt_users);
207 	mutex_init(&xprt->xpt_mutex);
208 	spin_lock_init(&xprt->xpt_lock);
209 	set_bit(XPT_BUSY, &xprt->xpt_flags);
210 	xprt->xpt_net = get_net_track(net, &xprt->ns_tracker, GFP_ATOMIC);
211 	strcpy(xprt->xpt_remotebuf, "uninitialized");
212 }
213 EXPORT_SYMBOL_GPL(svc_xprt_init);
214 
215 static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl,
216 					 struct svc_serv *serv,
217 					 struct net *net,
218 					 const int family,
219 					 const unsigned short port,
220 					 int flags)
221 {
222 	struct sockaddr_in sin = {
223 		.sin_family		= AF_INET,
224 		.sin_addr.s_addr	= htonl(INADDR_ANY),
225 		.sin_port		= htons(port),
226 	};
227 #if IS_ENABLED(CONFIG_IPV6)
228 	struct sockaddr_in6 sin6 = {
229 		.sin6_family		= AF_INET6,
230 		.sin6_addr		= IN6ADDR_ANY_INIT,
231 		.sin6_port		= htons(port),
232 	};
233 #endif
234 	struct svc_xprt *xprt;
235 	struct sockaddr *sap;
236 	size_t len;
237 
238 	switch (family) {
239 	case PF_INET:
240 		sap = (struct sockaddr *)&sin;
241 		len = sizeof(sin);
242 		break;
243 #if IS_ENABLED(CONFIG_IPV6)
244 	case PF_INET6:
245 		sap = (struct sockaddr *)&sin6;
246 		len = sizeof(sin6);
247 		break;
248 #endif
249 	default:
250 		return ERR_PTR(-EAFNOSUPPORT);
251 	}
252 
253 	xprt = xcl->xcl_ops->xpo_create(serv, net, sap, len, flags);
254 	if (IS_ERR(xprt))
255 		trace_svc_xprt_create_err(serv->sv_program->pg_name,
256 					  xcl->xcl_name, sap, len, xprt);
257 	return xprt;
258 }
259 
260 /**
261  * svc_xprt_received - start next receiver thread
262  * @xprt: controlling transport
263  *
264  * The caller must hold the XPT_BUSY bit and must
265  * not thereafter touch transport data.
266  *
267  * Note: XPT_DATA only gets cleared when a read-attempt finds no (or
268  * insufficient) data.
269  */
270 void svc_xprt_received(struct svc_xprt *xprt)
271 {
272 	if (!test_bit(XPT_BUSY, &xprt->xpt_flags)) {
273 		WARN_ONCE(1, "xprt=0x%p already busy!", xprt);
274 		return;
275 	}
276 
277 	/* As soon as we clear busy, the xprt could be closed and
278 	 * 'put', so we need a reference to call svc_xprt_enqueue with:
279 	 */
280 	svc_xprt_get(xprt);
281 	smp_mb__before_atomic();
282 	clear_bit(XPT_BUSY, &xprt->xpt_flags);
283 	svc_xprt_enqueue(xprt);
284 	svc_xprt_put(xprt);
285 }
286 EXPORT_SYMBOL_GPL(svc_xprt_received);
287 
288 void svc_add_new_perm_xprt(struct svc_serv *serv, struct svc_xprt *new)
289 {
290 	clear_bit(XPT_TEMP, &new->xpt_flags);
291 	spin_lock_bh(&serv->sv_lock);
292 	list_add(&new->xpt_list, &serv->sv_permsocks);
293 	spin_unlock_bh(&serv->sv_lock);
294 	svc_xprt_received(new);
295 }
296 
297 static int _svc_xprt_create(struct svc_serv *serv, const char *xprt_name,
298 			    struct net *net, const int family,
299 			    const unsigned short port, int flags,
300 			    const struct cred *cred)
301 {
302 	struct svc_xprt_class *xcl;
303 
304 	spin_lock(&svc_xprt_class_lock);
305 	list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
306 		struct svc_xprt *newxprt;
307 		unsigned short newport;
308 
309 		if (strcmp(xprt_name, xcl->xcl_name))
310 			continue;
311 
312 		if (!try_module_get(xcl->xcl_owner))
313 			goto err;
314 
315 		spin_unlock(&svc_xprt_class_lock);
316 		newxprt = __svc_xpo_create(xcl, serv, net, family, port, flags);
317 		if (IS_ERR(newxprt)) {
318 			module_put(xcl->xcl_owner);
319 			return PTR_ERR(newxprt);
320 		}
321 		newxprt->xpt_cred = get_cred(cred);
322 		svc_add_new_perm_xprt(serv, newxprt);
323 		newport = svc_xprt_local_port(newxprt);
324 		return newport;
325 	}
326  err:
327 	spin_unlock(&svc_xprt_class_lock);
328 	/* This errno is exposed to user space.  Provide a reasonable
329 	 * perror msg for a bad transport. */
330 	return -EPROTONOSUPPORT;
331 }
332 
333 /**
334  * svc_xprt_create - Add a new listener to @serv
335  * @serv: target RPC service
336  * @xprt_name: transport class name
337  * @net: network namespace
338  * @family: network address family
339  * @port: listener port
340  * @flags: SVC_SOCK flags
341  * @cred: credential to bind to this transport
342  *
343  * Return values:
344  *   %0: New listener added successfully
345  *   %-EPROTONOSUPPORT: Requested transport type not supported
346  */
347 int svc_xprt_create(struct svc_serv *serv, const char *xprt_name,
348 		    struct net *net, const int family,
349 		    const unsigned short port, int flags,
350 		    const struct cred *cred)
351 {
352 	int err;
353 
354 	err = _svc_xprt_create(serv, xprt_name, net, family, port, flags, cred);
355 	if (err == -EPROTONOSUPPORT) {
356 		request_module("svc%s", xprt_name);
357 		err = _svc_xprt_create(serv, xprt_name, net, family, port, flags, cred);
358 	}
359 	return err;
360 }
361 EXPORT_SYMBOL_GPL(svc_xprt_create);
362 
363 /*
364  * Copy the local and remote xprt addresses to the rqstp structure
365  */
366 void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt)
367 {
368 	memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen);
369 	rqstp->rq_addrlen = xprt->xpt_remotelen;
370 
371 	/*
372 	 * Destination address in request is needed for binding the
373 	 * source address in RPC replies/callbacks later.
374 	 */
375 	memcpy(&rqstp->rq_daddr, &xprt->xpt_local, xprt->xpt_locallen);
376 	rqstp->rq_daddrlen = xprt->xpt_locallen;
377 }
378 EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs);
379 
380 /**
381  * svc_print_addr - Format rq_addr field for printing
382  * @rqstp: svc_rqst struct containing address to print
383  * @buf: target buffer for formatted address
384  * @len: length of target buffer
385  *
386  */
387 char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
388 {
389 	return __svc_print_addr(svc_addr(rqstp), buf, len);
390 }
391 EXPORT_SYMBOL_GPL(svc_print_addr);
392 
393 static bool svc_xprt_slots_in_range(struct svc_xprt *xprt)
394 {
395 	unsigned int limit = svc_rpc_per_connection_limit;
396 	int nrqsts = atomic_read(&xprt->xpt_nr_rqsts);
397 
398 	return limit == 0 || (nrqsts >= 0 && nrqsts < limit);
399 }
400 
401 static bool svc_xprt_reserve_slot(struct svc_rqst *rqstp, struct svc_xprt *xprt)
402 {
403 	if (!test_bit(RQ_DATA, &rqstp->rq_flags)) {
404 		if (!svc_xprt_slots_in_range(xprt))
405 			return false;
406 		atomic_inc(&xprt->xpt_nr_rqsts);
407 		set_bit(RQ_DATA, &rqstp->rq_flags);
408 	}
409 	return true;
410 }
411 
412 static void svc_xprt_release_slot(struct svc_rqst *rqstp)
413 {
414 	struct svc_xprt	*xprt = rqstp->rq_xprt;
415 	if (test_and_clear_bit(RQ_DATA, &rqstp->rq_flags)) {
416 		atomic_dec(&xprt->xpt_nr_rqsts);
417 		smp_wmb(); /* See smp_rmb() in svc_xprt_ready() */
418 		svc_xprt_enqueue(xprt);
419 	}
420 }
421 
422 static bool svc_xprt_ready(struct svc_xprt *xprt)
423 {
424 	unsigned long xpt_flags;
425 
426 	/*
427 	 * If another cpu has recently updated xpt_flags,
428 	 * sk_sock->flags, xpt_reserved, or xpt_nr_rqsts, we need to
429 	 * know about it; otherwise it's possible that both that cpu and
430 	 * this one could call svc_xprt_enqueue() without either
431 	 * svc_xprt_enqueue() recognizing that the conditions below
432 	 * are satisfied, and we could stall indefinitely:
433 	 */
434 	smp_rmb();
435 	xpt_flags = READ_ONCE(xprt->xpt_flags);
436 
437 	if (xpt_flags & BIT(XPT_BUSY))
438 		return false;
439 	if (xpt_flags & (BIT(XPT_CONN) | BIT(XPT_CLOSE) | BIT(XPT_HANDSHAKE)))
440 		return true;
441 	if (xpt_flags & (BIT(XPT_DATA) | BIT(XPT_DEFERRED))) {
442 		if (xprt->xpt_ops->xpo_has_wspace(xprt) &&
443 		    svc_xprt_slots_in_range(xprt))
444 			return true;
445 		trace_svc_xprt_no_write_space(xprt);
446 		return false;
447 	}
448 	return false;
449 }
450 
451 /**
452  * svc_xprt_enqueue - Queue a transport on an idle nfsd thread
453  * @xprt: transport with data pending
454  *
455  */
456 void svc_xprt_enqueue(struct svc_xprt *xprt)
457 {
458 	struct svc_pool *pool;
459 	struct svc_rqst	*rqstp = NULL;
460 
461 	if (!svc_xprt_ready(xprt))
462 		return;
463 
464 	/* Mark transport as busy. It will remain in this state until
465 	 * the provider calls svc_xprt_received. We update XPT_BUSY
466 	 * atomically because it also guards against trying to enqueue
467 	 * the transport twice.
468 	 */
469 	if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
470 		return;
471 
472 	pool = svc_pool_for_cpu(xprt->xpt_server);
473 
474 	percpu_counter_inc(&pool->sp_sockets_queued);
475 	spin_lock_bh(&pool->sp_lock);
476 	list_add_tail(&xprt->xpt_ready, &pool->sp_sockets);
477 	spin_unlock_bh(&pool->sp_lock);
478 
479 	/* find a thread for this xprt */
480 	rcu_read_lock();
481 	list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) {
482 		if (test_and_set_bit(RQ_BUSY, &rqstp->rq_flags))
483 			continue;
484 		percpu_counter_inc(&pool->sp_threads_woken);
485 		rqstp->rq_qtime = ktime_get();
486 		wake_up_process(rqstp->rq_task);
487 		goto out_unlock;
488 	}
489 	set_bit(SP_CONGESTED, &pool->sp_flags);
490 	rqstp = NULL;
491 out_unlock:
492 	rcu_read_unlock();
493 	trace_svc_xprt_enqueue(xprt, rqstp);
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 	if (list_empty(&pool->sp_sockets))
505 		goto out;
506 
507 	spin_lock_bh(&pool->sp_lock);
508 	if (likely(!list_empty(&pool->sp_sockets))) {
509 		xprt = list_first_entry(&pool->sp_sockets,
510 					struct svc_xprt, xpt_ready);
511 		list_del_init(&xprt->xpt_ready);
512 		svc_xprt_get(xprt);
513 	}
514 	spin_unlock_bh(&pool->sp_lock);
515 out:
516 	return xprt;
517 }
518 
519 /**
520  * svc_reserve - change the space reserved for the reply to a request.
521  * @rqstp:  The request in question
522  * @space: new max space to reserve
523  *
524  * Each request reserves some space on the output queue of the transport
525  * to make sure the reply fits.  This function reduces that reserved
526  * space to be the amount of space used already, plus @space.
527  *
528  */
529 void svc_reserve(struct svc_rqst *rqstp, int space)
530 {
531 	struct svc_xprt *xprt = rqstp->rq_xprt;
532 
533 	space += rqstp->rq_res.head[0].iov_len;
534 
535 	if (xprt && space < rqstp->rq_reserved) {
536 		atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved);
537 		rqstp->rq_reserved = space;
538 		smp_wmb(); /* See smp_rmb() in svc_xprt_ready() */
539 		svc_xprt_enqueue(xprt);
540 	}
541 }
542 EXPORT_SYMBOL_GPL(svc_reserve);
543 
544 static void free_deferred(struct svc_xprt *xprt, struct svc_deferred_req *dr)
545 {
546 	if (!dr)
547 		return;
548 
549 	xprt->xpt_ops->xpo_release_ctxt(xprt, dr->xprt_ctxt);
550 	kfree(dr);
551 }
552 
553 static void svc_xprt_release(struct svc_rqst *rqstp)
554 {
555 	struct svc_xprt	*xprt = rqstp->rq_xprt;
556 
557 	xprt->xpt_ops->xpo_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
558 	rqstp->rq_xprt_ctxt = NULL;
559 
560 	free_deferred(xprt, rqstp->rq_deferred);
561 	rqstp->rq_deferred = NULL;
562 
563 	svc_rqst_release_pages(rqstp);
564 	rqstp->rq_res.page_len = 0;
565 	rqstp->rq_res.page_base = 0;
566 
567 	/* Reset response buffer and release
568 	 * the reservation.
569 	 * But first, check that enough space was reserved
570 	 * for the reply, otherwise we have a bug!
571 	 */
572 	if ((rqstp->rq_res.len) >  rqstp->rq_reserved)
573 		printk(KERN_ERR "RPC request reserved %d but used %d\n",
574 		       rqstp->rq_reserved,
575 		       rqstp->rq_res.len);
576 
577 	rqstp->rq_res.head[0].iov_len = 0;
578 	svc_reserve(rqstp, 0);
579 	svc_xprt_release_slot(rqstp);
580 	rqstp->rq_xprt = NULL;
581 	svc_xprt_put(xprt);
582 }
583 
584 /*
585  * Some svc_serv's will have occasional work to do, even when a xprt is not
586  * waiting to be serviced. This function is there to "kick" a task in one of
587  * those services so that it can wake up and do that work. Note that we only
588  * bother with pool 0 as we don't need to wake up more than one thread for
589  * this purpose.
590  */
591 void svc_wake_up(struct svc_serv *serv)
592 {
593 	struct svc_rqst	*rqstp;
594 	struct svc_pool *pool;
595 
596 	pool = &serv->sv_pools[0];
597 
598 	rcu_read_lock();
599 	list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) {
600 		/* skip any that aren't queued */
601 		if (test_bit(RQ_BUSY, &rqstp->rq_flags))
602 			continue;
603 		rcu_read_unlock();
604 		wake_up_process(rqstp->rq_task);
605 		trace_svc_wake_up(rqstp->rq_task->pid);
606 		return;
607 	}
608 	rcu_read_unlock();
609 
610 	/* No free entries available */
611 	set_bit(SP_TASK_PENDING, &pool->sp_flags);
612 	smp_wmb();
613 	trace_svc_wake_up(0);
614 }
615 EXPORT_SYMBOL_GPL(svc_wake_up);
616 
617 int svc_port_is_privileged(struct sockaddr *sin)
618 {
619 	switch (sin->sa_family) {
620 	case AF_INET:
621 		return ntohs(((struct sockaddr_in *)sin)->sin_port)
622 			< PROT_SOCK;
623 	case AF_INET6:
624 		return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
625 			< PROT_SOCK;
626 	default:
627 		return 0;
628 	}
629 }
630 
631 /*
632  * Make sure that we don't have too many active connections. If we have,
633  * something must be dropped. It's not clear what will happen if we allow
634  * "too many" connections, but when dealing with network-facing software,
635  * we have to code defensively. Here we do that by imposing hard limits.
636  *
637  * There's no point in trying to do random drop here for DoS
638  * prevention. The NFS clients does 1 reconnect in 15 seconds. An
639  * attacker can easily beat that.
640  *
641  * The only somewhat efficient mechanism would be if drop old
642  * connections from the same IP first. But right now we don't even
643  * record the client IP in svc_sock.
644  *
645  * single-threaded services that expect a lot of clients will probably
646  * need to set sv_maxconn to override the default value which is based
647  * on the number of threads
648  */
649 static void svc_check_conn_limits(struct svc_serv *serv)
650 {
651 	unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn :
652 				(serv->sv_nrthreads+3) * 20;
653 
654 	if (serv->sv_tmpcnt > limit) {
655 		struct svc_xprt *xprt = NULL;
656 		spin_lock_bh(&serv->sv_lock);
657 		if (!list_empty(&serv->sv_tempsocks)) {
658 			/* Try to help the admin */
659 			net_notice_ratelimited("%s: too many open connections, consider increasing the %s\n",
660 					       serv->sv_name, serv->sv_maxconn ?
661 					       "max number of connections" :
662 					       "number of threads");
663 			/*
664 			 * Always select the oldest connection. It's not fair,
665 			 * but so is life
666 			 */
667 			xprt = list_entry(serv->sv_tempsocks.prev,
668 					  struct svc_xprt,
669 					  xpt_list);
670 			set_bit(XPT_CLOSE, &xprt->xpt_flags);
671 			svc_xprt_get(xprt);
672 		}
673 		spin_unlock_bh(&serv->sv_lock);
674 
675 		if (xprt) {
676 			svc_xprt_enqueue(xprt);
677 			svc_xprt_put(xprt);
678 		}
679 	}
680 }
681 
682 static int svc_alloc_arg(struct svc_rqst *rqstp)
683 {
684 	struct svc_serv *serv = rqstp->rq_server;
685 	struct xdr_buf *arg = &rqstp->rq_arg;
686 	unsigned long pages, filled, ret;
687 
688 	pages = (serv->sv_max_mesg + 2 * PAGE_SIZE) >> PAGE_SHIFT;
689 	if (pages > RPCSVC_MAXPAGES) {
690 		pr_warn_once("svc: warning: pages=%lu > RPCSVC_MAXPAGES=%lu\n",
691 			     pages, RPCSVC_MAXPAGES);
692 		/* use as many pages as possible */
693 		pages = RPCSVC_MAXPAGES;
694 	}
695 
696 	for (filled = 0; filled < pages; filled = ret) {
697 		ret = alloc_pages_bulk_array_node(GFP_KERNEL,
698 						  rqstp->rq_pool->sp_id,
699 						  pages, rqstp->rq_pages);
700 		if (ret > filled)
701 			/* Made progress, don't sleep yet */
702 			continue;
703 
704 		set_current_state(TASK_INTERRUPTIBLE);
705 		if (signalled() || kthread_should_stop()) {
706 			set_current_state(TASK_RUNNING);
707 			return -EINTR;
708 		}
709 		trace_svc_alloc_arg_err(pages, ret);
710 		memalloc_retry_wait(GFP_KERNEL);
711 	}
712 	rqstp->rq_page_end = &rqstp->rq_pages[pages];
713 	rqstp->rq_pages[pages] = NULL; /* this might be seen in nfsd_splice_actor() */
714 
715 	/* Make arg->head point to first page and arg->pages point to rest */
716 	arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
717 	arg->head[0].iov_len = PAGE_SIZE;
718 	arg->pages = rqstp->rq_pages + 1;
719 	arg->page_base = 0;
720 	/* save at least one page for response */
721 	arg->page_len = (pages-2)*PAGE_SIZE;
722 	arg->len = (pages-1)*PAGE_SIZE;
723 	arg->tail[0].iov_len = 0;
724 
725 	rqstp->rq_xid = xdr_zero;
726 	return 0;
727 }
728 
729 static bool
730 rqst_should_sleep(struct svc_rqst *rqstp)
731 {
732 	struct svc_pool		*pool = rqstp->rq_pool;
733 
734 	/* did someone call svc_wake_up? */
735 	if (test_and_clear_bit(SP_TASK_PENDING, &pool->sp_flags))
736 		return false;
737 
738 	/* was a socket queued? */
739 	if (!list_empty(&pool->sp_sockets))
740 		return false;
741 
742 	/* are we shutting down? */
743 	if (signalled() || kthread_should_stop())
744 		return false;
745 
746 	/* are we freezing? */
747 	if (freezing(current))
748 		return false;
749 
750 	return true;
751 }
752 
753 static struct svc_xprt *svc_get_next_xprt(struct svc_rqst *rqstp, long timeout)
754 {
755 	struct svc_pool		*pool = rqstp->rq_pool;
756 	long			time_left = 0;
757 
758 	/* rq_xprt should be clear on entry */
759 	WARN_ON_ONCE(rqstp->rq_xprt);
760 
761 	rqstp->rq_xprt = svc_xprt_dequeue(pool);
762 	if (rqstp->rq_xprt)
763 		goto out_found;
764 
765 	/*
766 	 * We have to be able to interrupt this wait
767 	 * to bring down the daemons ...
768 	 */
769 	set_current_state(TASK_INTERRUPTIBLE);
770 	smp_mb__before_atomic();
771 	clear_bit(SP_CONGESTED, &pool->sp_flags);
772 	clear_bit(RQ_BUSY, &rqstp->rq_flags);
773 	smp_mb__after_atomic();
774 
775 	if (likely(rqst_should_sleep(rqstp)))
776 		time_left = schedule_timeout(timeout);
777 	else
778 		__set_current_state(TASK_RUNNING);
779 
780 	try_to_freeze();
781 
782 	set_bit(RQ_BUSY, &rqstp->rq_flags);
783 	smp_mb__after_atomic();
784 	rqstp->rq_xprt = svc_xprt_dequeue(pool);
785 	if (rqstp->rq_xprt)
786 		goto out_found;
787 
788 	if (!time_left)
789 		percpu_counter_inc(&pool->sp_threads_timedout);
790 
791 	if (signalled() || kthread_should_stop())
792 		return ERR_PTR(-EINTR);
793 	return ERR_PTR(-EAGAIN);
794 out_found:
795 	/* Normally we will wait up to 5 seconds for any required
796 	 * cache information to be provided.
797 	 */
798 	if (!test_bit(SP_CONGESTED, &pool->sp_flags))
799 		rqstp->rq_chandle.thread_wait = 5*HZ;
800 	else
801 		rqstp->rq_chandle.thread_wait = 1*HZ;
802 	trace_svc_xprt_dequeue(rqstp);
803 	return rqstp->rq_xprt;
804 }
805 
806 static void svc_add_new_temp_xprt(struct svc_serv *serv, struct svc_xprt *newxpt)
807 {
808 	spin_lock_bh(&serv->sv_lock);
809 	set_bit(XPT_TEMP, &newxpt->xpt_flags);
810 	list_add(&newxpt->xpt_list, &serv->sv_tempsocks);
811 	serv->sv_tmpcnt++;
812 	if (serv->sv_temptimer.function == NULL) {
813 		/* setup timer to age temp transports */
814 		serv->sv_temptimer.function = svc_age_temp_xprts;
815 		mod_timer(&serv->sv_temptimer,
816 			  jiffies + svc_conn_age_period * HZ);
817 	}
818 	spin_unlock_bh(&serv->sv_lock);
819 	svc_xprt_received(newxpt);
820 }
821 
822 static int svc_handle_xprt(struct svc_rqst *rqstp, struct svc_xprt *xprt)
823 {
824 	struct svc_serv *serv = rqstp->rq_server;
825 	int len = 0;
826 
827 	if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) {
828 		if (test_and_clear_bit(XPT_KILL_TEMP, &xprt->xpt_flags))
829 			xprt->xpt_ops->xpo_kill_temp_xprt(xprt);
830 		svc_delete_xprt(xprt);
831 		/* Leave XPT_BUSY set on the dead xprt: */
832 		goto out;
833 	}
834 	if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
835 		struct svc_xprt *newxpt;
836 		/*
837 		 * We know this module_get will succeed because the
838 		 * listener holds a reference too
839 		 */
840 		__module_get(xprt->xpt_class->xcl_owner);
841 		svc_check_conn_limits(xprt->xpt_server);
842 		newxpt = xprt->xpt_ops->xpo_accept(xprt);
843 		if (newxpt) {
844 			newxpt->xpt_cred = get_cred(xprt->xpt_cred);
845 			svc_add_new_temp_xprt(serv, newxpt);
846 			trace_svc_xprt_accept(newxpt, serv->sv_name);
847 		} else {
848 			module_put(xprt->xpt_class->xcl_owner);
849 		}
850 		svc_xprt_received(xprt);
851 	} else if (test_bit(XPT_HANDSHAKE, &xprt->xpt_flags)) {
852 		xprt->xpt_ops->xpo_handshake(xprt);
853 		svc_xprt_received(xprt);
854 	} else if (svc_xprt_reserve_slot(rqstp, xprt)) {
855 		/* XPT_DATA|XPT_DEFERRED case: */
856 		rqstp->rq_deferred = svc_deferred_dequeue(xprt);
857 		if (rqstp->rq_deferred)
858 			len = svc_deferred_recv(rqstp);
859 		else
860 			len = xprt->xpt_ops->xpo_recvfrom(rqstp);
861 		rqstp->rq_reserved = serv->sv_max_mesg;
862 		atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
863 	} else
864 		svc_xprt_received(xprt);
865 
866 out:
867 	return len;
868 }
869 
870 /*
871  * Receive the next request on any transport.  This code is carefully
872  * organised not to touch any cachelines in the shared svc_serv
873  * structure, only cachelines in the local svc_pool.
874  */
875 int svc_recv(struct svc_rqst *rqstp, long timeout)
876 {
877 	struct svc_xprt		*xprt = NULL;
878 	struct svc_serv		*serv = rqstp->rq_server;
879 	int			len, err;
880 
881 	err = svc_alloc_arg(rqstp);
882 	if (err)
883 		goto out;
884 
885 	try_to_freeze();
886 	cond_resched();
887 	err = -EINTR;
888 	if (signalled() || kthread_should_stop())
889 		goto out;
890 
891 	xprt = svc_get_next_xprt(rqstp, timeout);
892 	if (IS_ERR(xprt)) {
893 		err = PTR_ERR(xprt);
894 		goto out;
895 	}
896 
897 	len = svc_handle_xprt(rqstp, xprt);
898 
899 	/* No data, incomplete (TCP) read, or accept() */
900 	err = -EAGAIN;
901 	if (len <= 0)
902 		goto out_release;
903 
904 	trace_svc_xdr_recvfrom(&rqstp->rq_arg);
905 
906 	clear_bit(XPT_OLD, &xprt->xpt_flags);
907 
908 	rqstp->rq_chandle.defer = svc_defer;
909 
910 	if (serv->sv_stats)
911 		serv->sv_stats->netcnt++;
912 	rqstp->rq_stime = ktime_get();
913 	return len;
914 out_release:
915 	rqstp->rq_res.len = 0;
916 	svc_xprt_release(rqstp);
917 out:
918 	return err;
919 }
920 EXPORT_SYMBOL_GPL(svc_recv);
921 
922 /*
923  * Drop request
924  */
925 void svc_drop(struct svc_rqst *rqstp)
926 {
927 	trace_svc_drop(rqstp);
928 	svc_xprt_release(rqstp);
929 }
930 EXPORT_SYMBOL_GPL(svc_drop);
931 
932 /**
933  * svc_send - Return reply to client
934  * @rqstp: RPC transaction context
935  *
936  */
937 void svc_send(struct svc_rqst *rqstp)
938 {
939 	struct svc_xprt	*xprt;
940 	struct xdr_buf	*xb;
941 	int status;
942 
943 	xprt = rqstp->rq_xprt;
944 	if (!xprt)
945 		return;
946 
947 	/* calculate over-all length */
948 	xb = &rqstp->rq_res;
949 	xb->len = xb->head[0].iov_len +
950 		xb->page_len +
951 		xb->tail[0].iov_len;
952 	trace_svc_xdr_sendto(rqstp->rq_xid, xb);
953 	trace_svc_stats_latency(rqstp);
954 
955 	status = xprt->xpt_ops->xpo_sendto(rqstp);
956 
957 	trace_svc_send(rqstp, status);
958 	svc_xprt_release(rqstp);
959 }
960 
961 /*
962  * Timer function to close old temporary transports, using
963  * a mark-and-sweep algorithm.
964  */
965 static void svc_age_temp_xprts(struct timer_list *t)
966 {
967 	struct svc_serv *serv = from_timer(serv, t, sv_temptimer);
968 	struct svc_xprt *xprt;
969 	struct list_head *le, *next;
970 
971 	dprintk("svc_age_temp_xprts\n");
972 
973 	if (!spin_trylock_bh(&serv->sv_lock)) {
974 		/* busy, try again 1 sec later */
975 		dprintk("svc_age_temp_xprts: busy\n");
976 		mod_timer(&serv->sv_temptimer, jiffies + HZ);
977 		return;
978 	}
979 
980 	list_for_each_safe(le, next, &serv->sv_tempsocks) {
981 		xprt = list_entry(le, struct svc_xprt, xpt_list);
982 
983 		/* First time through, just mark it OLD. Second time
984 		 * through, close it. */
985 		if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags))
986 			continue;
987 		if (kref_read(&xprt->xpt_ref) > 1 ||
988 		    test_bit(XPT_BUSY, &xprt->xpt_flags))
989 			continue;
990 		list_del_init(le);
991 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
992 		dprintk("queuing xprt %p for closing\n", xprt);
993 
994 		/* a thread will dequeue and close it soon */
995 		svc_xprt_enqueue(xprt);
996 	}
997 	spin_unlock_bh(&serv->sv_lock);
998 
999 	mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
1000 }
1001 
1002 /* Close temporary transports whose xpt_local matches server_addr immediately
1003  * instead of waiting for them to be picked up by the timer.
1004  *
1005  * This is meant to be called from a notifier_block that runs when an ip
1006  * address is deleted.
1007  */
1008 void svc_age_temp_xprts_now(struct svc_serv *serv, struct sockaddr *server_addr)
1009 {
1010 	struct svc_xprt *xprt;
1011 	struct list_head *le, *next;
1012 	LIST_HEAD(to_be_closed);
1013 
1014 	spin_lock_bh(&serv->sv_lock);
1015 	list_for_each_safe(le, next, &serv->sv_tempsocks) {
1016 		xprt = list_entry(le, struct svc_xprt, xpt_list);
1017 		if (rpc_cmp_addr(server_addr, (struct sockaddr *)
1018 				&xprt->xpt_local)) {
1019 			dprintk("svc_age_temp_xprts_now: found %p\n", xprt);
1020 			list_move(le, &to_be_closed);
1021 		}
1022 	}
1023 	spin_unlock_bh(&serv->sv_lock);
1024 
1025 	while (!list_empty(&to_be_closed)) {
1026 		le = to_be_closed.next;
1027 		list_del_init(le);
1028 		xprt = list_entry(le, struct svc_xprt, xpt_list);
1029 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
1030 		set_bit(XPT_KILL_TEMP, &xprt->xpt_flags);
1031 		dprintk("svc_age_temp_xprts_now: queuing xprt %p for closing\n",
1032 				xprt);
1033 		svc_xprt_enqueue(xprt);
1034 	}
1035 }
1036 EXPORT_SYMBOL_GPL(svc_age_temp_xprts_now);
1037 
1038 static void call_xpt_users(struct svc_xprt *xprt)
1039 {
1040 	struct svc_xpt_user *u;
1041 
1042 	spin_lock(&xprt->xpt_lock);
1043 	while (!list_empty(&xprt->xpt_users)) {
1044 		u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list);
1045 		list_del_init(&u->list);
1046 		u->callback(u);
1047 	}
1048 	spin_unlock(&xprt->xpt_lock);
1049 }
1050 
1051 /*
1052  * Remove a dead transport
1053  */
1054 static void svc_delete_xprt(struct svc_xprt *xprt)
1055 {
1056 	struct svc_serv	*serv = xprt->xpt_server;
1057 	struct svc_deferred_req *dr;
1058 
1059 	if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags))
1060 		return;
1061 
1062 	trace_svc_xprt_detach(xprt);
1063 	xprt->xpt_ops->xpo_detach(xprt);
1064 	if (xprt->xpt_bc_xprt)
1065 		xprt->xpt_bc_xprt->ops->close(xprt->xpt_bc_xprt);
1066 
1067 	spin_lock_bh(&serv->sv_lock);
1068 	list_del_init(&xprt->xpt_list);
1069 	WARN_ON_ONCE(!list_empty(&xprt->xpt_ready));
1070 	if (test_bit(XPT_TEMP, &xprt->xpt_flags))
1071 		serv->sv_tmpcnt--;
1072 	spin_unlock_bh(&serv->sv_lock);
1073 
1074 	while ((dr = svc_deferred_dequeue(xprt)) != NULL)
1075 		free_deferred(xprt, dr);
1076 
1077 	call_xpt_users(xprt);
1078 	svc_xprt_put(xprt);
1079 }
1080 
1081 /**
1082  * svc_xprt_close - Close a client connection
1083  * @xprt: transport to disconnect
1084  *
1085  */
1086 void svc_xprt_close(struct svc_xprt *xprt)
1087 {
1088 	trace_svc_xprt_close(xprt);
1089 	set_bit(XPT_CLOSE, &xprt->xpt_flags);
1090 	if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
1091 		/* someone else will have to effect the close */
1092 		return;
1093 	/*
1094 	 * We expect svc_close_xprt() to work even when no threads are
1095 	 * running (e.g., while configuring the server before starting
1096 	 * any threads), so if the transport isn't busy, we delete
1097 	 * it ourself:
1098 	 */
1099 	svc_delete_xprt(xprt);
1100 }
1101 EXPORT_SYMBOL_GPL(svc_xprt_close);
1102 
1103 static int svc_close_list(struct svc_serv *serv, struct list_head *xprt_list, struct net *net)
1104 {
1105 	struct svc_xprt *xprt;
1106 	int ret = 0;
1107 
1108 	spin_lock_bh(&serv->sv_lock);
1109 	list_for_each_entry(xprt, xprt_list, xpt_list) {
1110 		if (xprt->xpt_net != net)
1111 			continue;
1112 		ret++;
1113 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
1114 		svc_xprt_enqueue(xprt);
1115 	}
1116 	spin_unlock_bh(&serv->sv_lock);
1117 	return ret;
1118 }
1119 
1120 static struct svc_xprt *svc_dequeue_net(struct svc_serv *serv, struct net *net)
1121 {
1122 	struct svc_pool *pool;
1123 	struct svc_xprt *xprt;
1124 	struct svc_xprt *tmp;
1125 	int i;
1126 
1127 	for (i = 0; i < serv->sv_nrpools; i++) {
1128 		pool = &serv->sv_pools[i];
1129 
1130 		spin_lock_bh(&pool->sp_lock);
1131 		list_for_each_entry_safe(xprt, tmp, &pool->sp_sockets, xpt_ready) {
1132 			if (xprt->xpt_net != net)
1133 				continue;
1134 			list_del_init(&xprt->xpt_ready);
1135 			spin_unlock_bh(&pool->sp_lock);
1136 			return xprt;
1137 		}
1138 		spin_unlock_bh(&pool->sp_lock);
1139 	}
1140 	return NULL;
1141 }
1142 
1143 static void svc_clean_up_xprts(struct svc_serv *serv, struct net *net)
1144 {
1145 	struct svc_xprt *xprt;
1146 
1147 	while ((xprt = svc_dequeue_net(serv, net))) {
1148 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
1149 		svc_delete_xprt(xprt);
1150 	}
1151 }
1152 
1153 /**
1154  * svc_xprt_destroy_all - Destroy transports associated with @serv
1155  * @serv: RPC service to be shut down
1156  * @net: target network namespace
1157  *
1158  * Server threads may still be running (especially in the case where the
1159  * service is still running in other network namespaces).
1160  *
1161  * So we shut down sockets the same way we would on a running server, by
1162  * setting XPT_CLOSE, enqueuing, and letting a thread pick it up to do
1163  * the close.  In the case there are no such other threads,
1164  * threads running, svc_clean_up_xprts() does a simple version of a
1165  * server's main event loop, and in the case where there are other
1166  * threads, we may need to wait a little while and then check again to
1167  * see if they're done.
1168  */
1169 void svc_xprt_destroy_all(struct svc_serv *serv, struct net *net)
1170 {
1171 	int delay = 0;
1172 
1173 	while (svc_close_list(serv, &serv->sv_permsocks, net) +
1174 	       svc_close_list(serv, &serv->sv_tempsocks, net)) {
1175 
1176 		svc_clean_up_xprts(serv, net);
1177 		msleep(delay++);
1178 	}
1179 }
1180 EXPORT_SYMBOL_GPL(svc_xprt_destroy_all);
1181 
1182 /*
1183  * Handle defer and revisit of requests
1184  */
1185 
1186 static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1187 {
1188 	struct svc_deferred_req *dr =
1189 		container_of(dreq, struct svc_deferred_req, handle);
1190 	struct svc_xprt *xprt = dr->xprt;
1191 
1192 	spin_lock(&xprt->xpt_lock);
1193 	set_bit(XPT_DEFERRED, &xprt->xpt_flags);
1194 	if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) {
1195 		spin_unlock(&xprt->xpt_lock);
1196 		trace_svc_defer_drop(dr);
1197 		free_deferred(xprt, dr);
1198 		svc_xprt_put(xprt);
1199 		return;
1200 	}
1201 	dr->xprt = NULL;
1202 	list_add(&dr->handle.recent, &xprt->xpt_deferred);
1203 	spin_unlock(&xprt->xpt_lock);
1204 	trace_svc_defer_queue(dr);
1205 	svc_xprt_enqueue(xprt);
1206 	svc_xprt_put(xprt);
1207 }
1208 
1209 /*
1210  * Save the request off for later processing. The request buffer looks
1211  * like this:
1212  *
1213  * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
1214  *
1215  * This code can only handle requests that consist of an xprt-header
1216  * and rpc-header.
1217  */
1218 static struct cache_deferred_req *svc_defer(struct cache_req *req)
1219 {
1220 	struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
1221 	struct svc_deferred_req *dr;
1222 
1223 	if (rqstp->rq_arg.page_len || !test_bit(RQ_USEDEFERRAL, &rqstp->rq_flags))
1224 		return NULL; /* if more than a page, give up FIXME */
1225 	if (rqstp->rq_deferred) {
1226 		dr = rqstp->rq_deferred;
1227 		rqstp->rq_deferred = NULL;
1228 	} else {
1229 		size_t skip;
1230 		size_t size;
1231 		/* FIXME maybe discard if size too large */
1232 		size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
1233 		dr = kmalloc(size, GFP_KERNEL);
1234 		if (dr == NULL)
1235 			return NULL;
1236 
1237 		dr->handle.owner = rqstp->rq_server;
1238 		dr->prot = rqstp->rq_prot;
1239 		memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
1240 		dr->addrlen = rqstp->rq_addrlen;
1241 		dr->daddr = rqstp->rq_daddr;
1242 		dr->argslen = rqstp->rq_arg.len >> 2;
1243 
1244 		/* back up head to the start of the buffer and copy */
1245 		skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1246 		memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
1247 		       dr->argslen << 2);
1248 	}
1249 	dr->xprt_ctxt = rqstp->rq_xprt_ctxt;
1250 	rqstp->rq_xprt_ctxt = NULL;
1251 	trace_svc_defer(rqstp);
1252 	svc_xprt_get(rqstp->rq_xprt);
1253 	dr->xprt = rqstp->rq_xprt;
1254 	set_bit(RQ_DROPME, &rqstp->rq_flags);
1255 
1256 	dr->handle.revisit = svc_revisit;
1257 	return &dr->handle;
1258 }
1259 
1260 /*
1261  * recv data from a deferred request into an active one
1262  */
1263 static noinline int svc_deferred_recv(struct svc_rqst *rqstp)
1264 {
1265 	struct svc_deferred_req *dr = rqstp->rq_deferred;
1266 
1267 	trace_svc_defer_recv(dr);
1268 
1269 	/* setup iov_base past transport header */
1270 	rqstp->rq_arg.head[0].iov_base = dr->args;
1271 	/* The iov_len does not include the transport header bytes */
1272 	rqstp->rq_arg.head[0].iov_len = dr->argslen << 2;
1273 	rqstp->rq_arg.page_len = 0;
1274 	/* The rq_arg.len includes the transport header bytes */
1275 	rqstp->rq_arg.len     = dr->argslen << 2;
1276 	rqstp->rq_prot        = dr->prot;
1277 	memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
1278 	rqstp->rq_addrlen     = dr->addrlen;
1279 	/* Save off transport header len in case we get deferred again */
1280 	rqstp->rq_daddr       = dr->daddr;
1281 	rqstp->rq_respages    = rqstp->rq_pages;
1282 	rqstp->rq_xprt_ctxt   = dr->xprt_ctxt;
1283 
1284 	dr->xprt_ctxt = NULL;
1285 	svc_xprt_received(rqstp->rq_xprt);
1286 	return dr->argslen << 2;
1287 }
1288 
1289 
1290 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt)
1291 {
1292 	struct svc_deferred_req *dr = NULL;
1293 
1294 	if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags))
1295 		return NULL;
1296 	spin_lock(&xprt->xpt_lock);
1297 	if (!list_empty(&xprt->xpt_deferred)) {
1298 		dr = list_entry(xprt->xpt_deferred.next,
1299 				struct svc_deferred_req,
1300 				handle.recent);
1301 		list_del_init(&dr->handle.recent);
1302 	} else
1303 		clear_bit(XPT_DEFERRED, &xprt->xpt_flags);
1304 	spin_unlock(&xprt->xpt_lock);
1305 	return dr;
1306 }
1307 
1308 /**
1309  * svc_find_xprt - find an RPC transport instance
1310  * @serv: pointer to svc_serv to search
1311  * @xcl_name: C string containing transport's class name
1312  * @net: owner net pointer
1313  * @af: Address family of transport's local address
1314  * @port: transport's IP port number
1315  *
1316  * Return the transport instance pointer for the endpoint accepting
1317  * connections/peer traffic from the specified transport class,
1318  * address family and port.
1319  *
1320  * Specifying 0 for the address family or port is effectively a
1321  * wild-card, and will result in matching the first transport in the
1322  * service's list that has a matching class name.
1323  */
1324 struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name,
1325 			       struct net *net, const sa_family_t af,
1326 			       const unsigned short port)
1327 {
1328 	struct svc_xprt *xprt;
1329 	struct svc_xprt *found = NULL;
1330 
1331 	/* Sanity check the args */
1332 	if (serv == NULL || xcl_name == NULL)
1333 		return found;
1334 
1335 	spin_lock_bh(&serv->sv_lock);
1336 	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1337 		if (xprt->xpt_net != net)
1338 			continue;
1339 		if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1340 			continue;
1341 		if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family)
1342 			continue;
1343 		if (port != 0 && port != svc_xprt_local_port(xprt))
1344 			continue;
1345 		found = xprt;
1346 		svc_xprt_get(xprt);
1347 		break;
1348 	}
1349 	spin_unlock_bh(&serv->sv_lock);
1350 	return found;
1351 }
1352 EXPORT_SYMBOL_GPL(svc_find_xprt);
1353 
1354 static int svc_one_xprt_name(const struct svc_xprt *xprt,
1355 			     char *pos, int remaining)
1356 {
1357 	int len;
1358 
1359 	len = snprintf(pos, remaining, "%s %u\n",
1360 			xprt->xpt_class->xcl_name,
1361 			svc_xprt_local_port(xprt));
1362 	if (len >= remaining)
1363 		return -ENAMETOOLONG;
1364 	return len;
1365 }
1366 
1367 /**
1368  * svc_xprt_names - format a buffer with a list of transport names
1369  * @serv: pointer to an RPC service
1370  * @buf: pointer to a buffer to be filled in
1371  * @buflen: length of buffer to be filled in
1372  *
1373  * Fills in @buf with a string containing a list of transport names,
1374  * each name terminated with '\n'.
1375  *
1376  * Returns positive length of the filled-in string on success; otherwise
1377  * a negative errno value is returned if an error occurs.
1378  */
1379 int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen)
1380 {
1381 	struct svc_xprt *xprt;
1382 	int len, totlen;
1383 	char *pos;
1384 
1385 	/* Sanity check args */
1386 	if (!serv)
1387 		return 0;
1388 
1389 	spin_lock_bh(&serv->sv_lock);
1390 
1391 	pos = buf;
1392 	totlen = 0;
1393 	list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1394 		len = svc_one_xprt_name(xprt, pos, buflen - totlen);
1395 		if (len < 0) {
1396 			*buf = '\0';
1397 			totlen = len;
1398 		}
1399 		if (len <= 0)
1400 			break;
1401 
1402 		pos += len;
1403 		totlen += len;
1404 	}
1405 
1406 	spin_unlock_bh(&serv->sv_lock);
1407 	return totlen;
1408 }
1409 EXPORT_SYMBOL_GPL(svc_xprt_names);
1410 
1411 
1412 /*----------------------------------------------------------------------------*/
1413 
1414 static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos)
1415 {
1416 	unsigned int pidx = (unsigned int)*pos;
1417 	struct svc_serv *serv = m->private;
1418 
1419 	dprintk("svc_pool_stats_start, *pidx=%u\n", pidx);
1420 
1421 	if (!pidx)
1422 		return SEQ_START_TOKEN;
1423 	return (pidx > serv->sv_nrpools ? NULL : &serv->sv_pools[pidx-1]);
1424 }
1425 
1426 static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos)
1427 {
1428 	struct svc_pool *pool = p;
1429 	struct svc_serv *serv = m->private;
1430 
1431 	dprintk("svc_pool_stats_next, *pos=%llu\n", *pos);
1432 
1433 	if (p == SEQ_START_TOKEN) {
1434 		pool = &serv->sv_pools[0];
1435 	} else {
1436 		unsigned int pidx = (pool - &serv->sv_pools[0]);
1437 		if (pidx < serv->sv_nrpools-1)
1438 			pool = &serv->sv_pools[pidx+1];
1439 		else
1440 			pool = NULL;
1441 	}
1442 	++*pos;
1443 	return pool;
1444 }
1445 
1446 static void svc_pool_stats_stop(struct seq_file *m, void *p)
1447 {
1448 }
1449 
1450 static int svc_pool_stats_show(struct seq_file *m, void *p)
1451 {
1452 	struct svc_pool *pool = p;
1453 
1454 	if (p == SEQ_START_TOKEN) {
1455 		seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n");
1456 		return 0;
1457 	}
1458 
1459 	seq_printf(m, "%u %llu %llu %llu %llu\n",
1460 		pool->sp_id,
1461 		percpu_counter_sum_positive(&pool->sp_sockets_queued),
1462 		percpu_counter_sum_positive(&pool->sp_sockets_queued),
1463 		percpu_counter_sum_positive(&pool->sp_threads_woken),
1464 		percpu_counter_sum_positive(&pool->sp_threads_timedout));
1465 
1466 	return 0;
1467 }
1468 
1469 static const struct seq_operations svc_pool_stats_seq_ops = {
1470 	.start	= svc_pool_stats_start,
1471 	.next	= svc_pool_stats_next,
1472 	.stop	= svc_pool_stats_stop,
1473 	.show	= svc_pool_stats_show,
1474 };
1475 
1476 int svc_pool_stats_open(struct svc_serv *serv, struct file *file)
1477 {
1478 	int err;
1479 
1480 	err = seq_open(file, &svc_pool_stats_seq_ops);
1481 	if (!err)
1482 		((struct seq_file *) file->private_data)->private = serv;
1483 	return err;
1484 }
1485 EXPORT_SYMBOL(svc_pool_stats_open);
1486 
1487 /*----------------------------------------------------------------------------*/
1488