blob: 8f0f1fb3dc5259419a1e823f522d9e5792538831 [file] [log] [blame]
Tom Tucker1d8206b92007-12-30 21:07:15 -06001/*
2 * linux/net/sunrpc/svc_xprt.c
3 *
4 * Author: Tom Tucker <tom@opengridcomputing.com>
5 */
6
7#include <linux/sched.h>
Alexey Dobriyan405f5572009-07-11 22:08:37 +04008#include <linux/smp_lock.h>
Tom Tucker1d8206b92007-12-30 21:07:15 -06009#include <linux/errno.h>
Tom Tucker1d8206b92007-12-30 21:07:15 -060010#include <linux/freezer.h>
Jeff Layton70867212008-02-07 16:34:54 -050011#include <linux/kthread.h>
Tom Tucker1d8206b92007-12-30 21:07:15 -060012#include <net/sock.h>
Tom Tucker1d8206b92007-12-30 21:07:15 -060013#include <linux/sunrpc/stats.h>
14#include <linux/sunrpc/svc_xprt.h>
H Hartley Sweetendcf1a352009-04-22 20:18:19 -040015#include <linux/sunrpc/svcsock.h>
Tom Tucker1d8206b92007-12-30 21:07:15 -060016
17#define RPCDBG_FACILITY RPCDBG_SVCXPRT
18
Tom Tucker0f0257e2007-12-30 21:08:27 -060019static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt);
20static int svc_deferred_recv(struct svc_rqst *rqstp);
21static struct cache_deferred_req *svc_defer(struct cache_req *req);
22static void svc_age_temp_xprts(unsigned long closure);
23
24/* apparently the "standard" is that clients close
25 * idle connections after 5 minutes, servers after
26 * 6 minutes
27 * http://www.connectathon.org/talks96/nfstcp.pdf
28 */
29static int svc_conn_age_period = 6*60;
30
Tom Tucker1d8206b92007-12-30 21:07:15 -060031/* List of registered transport classes */
32static DEFINE_SPINLOCK(svc_xprt_class_lock);
33static LIST_HEAD(svc_xprt_class_list);
34
Tom Tucker0f0257e2007-12-30 21:08:27 -060035/* SMP locking strategy:
36 *
37 * svc_pool->sp_lock protects most of the fields of that pool.
38 * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
39 * when both need to be taken (rare), svc_serv->sv_lock is first.
40 * BKL protects svc_serv->sv_nrthread.
41 * svc_sock->sk_lock protects the svc_sock->sk_deferred list
42 * and the ->sk_info_authunix cache.
43 *
44 * The XPT_BUSY bit in xprt->xpt_flags prevents a transport being
45 * enqueued multiply. During normal transport processing this bit
46 * is set by svc_xprt_enqueue and cleared by svc_xprt_received.
47 * Providers should not manipulate this bit directly.
48 *
49 * Some flags can be set to certain values at any time
50 * providing that certain rules are followed:
51 *
52 * XPT_CONN, XPT_DATA:
53 * - Can be set or cleared at any time.
54 * - After a set, svc_xprt_enqueue must be called to enqueue
55 * the transport for processing.
56 * - After a clear, the transport must be read/accepted.
57 * If this succeeds, it must be set again.
58 * XPT_CLOSE:
59 * - Can set at any time. It is never cleared.
60 * XPT_DEAD:
61 * - Can only be set while XPT_BUSY is held which ensures
62 * that no other thread will be using the transport or will
63 * try to set XPT_DEAD.
64 */
65
Tom Tucker1d8206b92007-12-30 21:07:15 -060066int svc_reg_xprt_class(struct svc_xprt_class *xcl)
67{
68 struct svc_xprt_class *cl;
69 int res = -EEXIST;
70
71 dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name);
72
73 INIT_LIST_HEAD(&xcl->xcl_list);
74 spin_lock(&svc_xprt_class_lock);
75 /* Make sure there isn't already a class with the same name */
76 list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) {
77 if (strcmp(xcl->xcl_name, cl->xcl_name) == 0)
78 goto out;
79 }
80 list_add_tail(&xcl->xcl_list, &svc_xprt_class_list);
81 res = 0;
82out:
83 spin_unlock(&svc_xprt_class_lock);
84 return res;
85}
86EXPORT_SYMBOL_GPL(svc_reg_xprt_class);
87
88void svc_unreg_xprt_class(struct svc_xprt_class *xcl)
89{
90 dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name);
91 spin_lock(&svc_xprt_class_lock);
92 list_del_init(&xcl->xcl_list);
93 spin_unlock(&svc_xprt_class_lock);
94}
95EXPORT_SYMBOL_GPL(svc_unreg_xprt_class);
96
Tom Tuckerdc9a16e2007-12-30 21:08:31 -060097/*
98 * Format the transport list for printing
99 */
100int svc_print_xprts(char *buf, int maxlen)
101{
102 struct list_head *le;
103 char tmpstr[80];
104 int len = 0;
105 buf[0] = '\0';
106
107 spin_lock(&svc_xprt_class_lock);
108 list_for_each(le, &svc_xprt_class_list) {
109 int slen;
110 struct svc_xprt_class *xcl =
111 list_entry(le, struct svc_xprt_class, xcl_list);
112
113 sprintf(tmpstr, "%s %d\n", xcl->xcl_name, xcl->xcl_max_payload);
114 slen = strlen(tmpstr);
115 if (len + slen > maxlen)
116 break;
117 len += slen;
118 strcat(buf, tmpstr);
119 }
120 spin_unlock(&svc_xprt_class_lock);
121
122 return len;
123}
124
Tom Tuckere1b31572007-12-30 21:07:46 -0600125static void svc_xprt_free(struct kref *kref)
126{
127 struct svc_xprt *xprt =
128 container_of(kref, struct svc_xprt, xpt_ref);
129 struct module *owner = xprt->xpt_class->xcl_owner;
Joe Perchesf64f9e72009-11-29 16:55:45 -0800130 if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags) &&
131 xprt->xpt_auth_cache != NULL)
Tom Tuckerdef13d72007-12-30 21:08:08 -0600132 svcauth_unix_info_release(xprt->xpt_auth_cache);
Tom Tuckere1b31572007-12-30 21:07:46 -0600133 xprt->xpt_ops->xpo_free(xprt);
134 module_put(owner);
135}
136
137void svc_xprt_put(struct svc_xprt *xprt)
138{
139 kref_put(&xprt->xpt_ref, svc_xprt_free);
140}
141EXPORT_SYMBOL_GPL(svc_xprt_put);
142
Tom Tucker1d8206b92007-12-30 21:07:15 -0600143/*
144 * Called by transport drivers to initialize the transport independent
145 * portion of the transport instance.
146 */
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600147void svc_xprt_init(struct svc_xprt_class *xcl, struct svc_xprt *xprt,
148 struct svc_serv *serv)
Tom Tucker1d8206b92007-12-30 21:07:15 -0600149{
150 memset(xprt, 0, sizeof(*xprt));
151 xprt->xpt_class = xcl;
152 xprt->xpt_ops = xcl->xcl_ops;
Tom Tuckere1b31572007-12-30 21:07:46 -0600153 kref_init(&xprt->xpt_ref);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600154 xprt->xpt_server = serv;
Tom Tucker7a182082007-12-30 21:07:53 -0600155 INIT_LIST_HEAD(&xprt->xpt_list);
156 INIT_LIST_HEAD(&xprt->xpt_ready);
Tom Tucker8c7b0172007-12-30 21:08:10 -0600157 INIT_LIST_HEAD(&xprt->xpt_deferred);
Tom Tuckera50fea22007-12-30 21:07:59 -0600158 mutex_init(&xprt->xpt_mutex);
Tom Tuckerdef13d72007-12-30 21:08:08 -0600159 spin_lock_init(&xprt->xpt_lock);
Tom Tucker4e5caaa2007-12-30 21:08:20 -0600160 set_bit(XPT_BUSY, &xprt->xpt_flags);
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300161 rpc_init_wait_queue(&xprt->xpt_bc_pending, "xpt_bc_pending");
Tom Tucker1d8206b92007-12-30 21:07:15 -0600162}
163EXPORT_SYMBOL_GPL(svc_xprt_init);
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600164
Chuck Lever5dd248f2008-06-30 18:45:37 -0400165static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl,
166 struct svc_serv *serv,
Chuck Lever9652ada2009-03-18 20:46:21 -0400167 const int family,
168 const unsigned short port,
169 int flags)
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600170{
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600171 struct sockaddr_in sin = {
172 .sin_family = AF_INET,
Al Viroe6f1ceb2008-03-17 22:44:53 -0700173 .sin_addr.s_addr = htonl(INADDR_ANY),
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600174 .sin_port = htons(port),
175 };
Chuck Leverd6783b22010-01-26 14:04:04 -0500176#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
Chuck Lever5dd248f2008-06-30 18:45:37 -0400177 struct sockaddr_in6 sin6 = {
178 .sin6_family = AF_INET6,
179 .sin6_addr = IN6ADDR_ANY_INIT,
180 .sin6_port = htons(port),
181 };
Chuck Leverd6783b22010-01-26 14:04:04 -0500182#endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
Chuck Lever5dd248f2008-06-30 18:45:37 -0400183 struct sockaddr *sap;
184 size_t len;
185
Chuck Lever9652ada2009-03-18 20:46:21 -0400186 switch (family) {
187 case PF_INET:
Chuck Lever5dd248f2008-06-30 18:45:37 -0400188 sap = (struct sockaddr *)&sin;
189 len = sizeof(sin);
190 break;
Chuck Leverd6783b22010-01-26 14:04:04 -0500191#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
Chuck Lever9652ada2009-03-18 20:46:21 -0400192 case PF_INET6:
Chuck Lever5dd248f2008-06-30 18:45:37 -0400193 sap = (struct sockaddr *)&sin6;
194 len = sizeof(sin6);
195 break;
Chuck Leverd6783b22010-01-26 14:04:04 -0500196#endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
Chuck Lever5dd248f2008-06-30 18:45:37 -0400197 default:
198 return ERR_PTR(-EAFNOSUPPORT);
199 }
200
201 return xcl->xcl_ops->xpo_create(serv, sap, len, flags);
202}
203
Chuck Lever9652ada2009-03-18 20:46:21 -0400204int svc_create_xprt(struct svc_serv *serv, const char *xprt_name,
205 const int family, const unsigned short port,
Chuck Lever5dd248f2008-06-30 18:45:37 -0400206 int flags)
207{
208 struct svc_xprt_class *xcl;
209
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600210 dprintk("svc: creating transport %s[%d]\n", xprt_name, port);
211 spin_lock(&svc_xprt_class_lock);
212 list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
Tom Tucker4e5caaa2007-12-30 21:08:20 -0600213 struct svc_xprt *newxprt;
214
215 if (strcmp(xprt_name, xcl->xcl_name))
216 continue;
217
218 if (!try_module_get(xcl->xcl_owner))
219 goto err;
220
221 spin_unlock(&svc_xprt_class_lock);
Chuck Lever9652ada2009-03-18 20:46:21 -0400222 newxprt = __svc_xpo_create(xcl, serv, family, port, flags);
Tom Tucker4e5caaa2007-12-30 21:08:20 -0600223 if (IS_ERR(newxprt)) {
224 module_put(xcl->xcl_owner);
225 return PTR_ERR(newxprt);
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600226 }
Tom Tucker4e5caaa2007-12-30 21:08:20 -0600227
228 clear_bit(XPT_TEMP, &newxprt->xpt_flags);
229 spin_lock_bh(&serv->sv_lock);
230 list_add(&newxprt->xpt_list, &serv->sv_permsocks);
231 spin_unlock_bh(&serv->sv_lock);
232 clear_bit(XPT_BUSY, &newxprt->xpt_flags);
233 return svc_xprt_local_port(newxprt);
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600234 }
Tom Tucker4e5caaa2007-12-30 21:08:20 -0600235 err:
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600236 spin_unlock(&svc_xprt_class_lock);
237 dprintk("svc: transport %s not found\n", xprt_name);
Chuck Lever68717902010-01-26 14:04:13 -0500238
239 /* This errno is exposed to user space. Provide a reasonable
240 * perror msg for a bad transport. */
241 return -EPROTONOSUPPORT;
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600242}
243EXPORT_SYMBOL_GPL(svc_create_xprt);
Tom Tucker9dbc2402007-12-30 21:08:12 -0600244
245/*
246 * Copy the local and remote xprt addresses to the rqstp structure
247 */
248void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt)
249{
250 struct sockaddr *sin;
251
252 memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen);
253 rqstp->rq_addrlen = xprt->xpt_remotelen;
254
255 /*
256 * Destination address in request is needed for binding the
257 * source address in RPC replies/callbacks later.
258 */
259 sin = (struct sockaddr *)&xprt->xpt_local;
260 switch (sin->sa_family) {
261 case AF_INET:
262 rqstp->rq_daddr.addr = ((struct sockaddr_in *)sin)->sin_addr;
263 break;
264 case AF_INET6:
265 rqstp->rq_daddr.addr6 = ((struct sockaddr_in6 *)sin)->sin6_addr;
266 break;
267 }
268}
269EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs);
270
Tom Tucker0f0257e2007-12-30 21:08:27 -0600271/**
272 * svc_print_addr - Format rq_addr field for printing
273 * @rqstp: svc_rqst struct containing address to print
274 * @buf: target buffer for formatted address
275 * @len: length of target buffer
276 *
277 */
278char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
279{
280 return __svc_print_addr(svc_addr(rqstp), buf, len);
281}
282EXPORT_SYMBOL_GPL(svc_print_addr);
283
284/*
285 * Queue up an idle server thread. Must have pool->sp_lock held.
286 * Note: this is really a stack rather than a queue, so that we only
287 * use as many different threads as we need, and the rest don't pollute
288 * the cache.
289 */
290static void svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp)
291{
292 list_add(&rqstp->rq_list, &pool->sp_threads);
293}
294
295/*
296 * Dequeue an nfsd thread. Must have pool->sp_lock held.
297 */
298static void svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp)
299{
300 list_del(&rqstp->rq_list);
301}
302
303/*
304 * Queue up a transport with data pending. If there are idle nfsd
305 * processes, wake 'em up.
306 *
307 */
308void svc_xprt_enqueue(struct svc_xprt *xprt)
309{
310 struct svc_serv *serv = xprt->xpt_server;
311 struct svc_pool *pool;
312 struct svc_rqst *rqstp;
313 int cpu;
314
315 if (!(xprt->xpt_flags &
316 ((1<<XPT_CONN)|(1<<XPT_DATA)|(1<<XPT_CLOSE)|(1<<XPT_DEFERRED))))
317 return;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600318
319 cpu = get_cpu();
320 pool = svc_pool_for_cpu(xprt->xpt_server, cpu);
321 put_cpu();
322
323 spin_lock_bh(&pool->sp_lock);
324
J. Bruce Fields78c210e2009-08-06 15:41:34 -0400325 if (!list_empty(&pool->sp_threads) &&
326 !list_empty(&pool->sp_sockets))
327 printk(KERN_ERR
328 "svc_xprt_enqueue: "
329 "threads and transports both waiting??\n");
330
Tom Tucker0f0257e2007-12-30 21:08:27 -0600331 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) {
332 /* Don't enqueue dead transports */
333 dprintk("svc: transport %p is dead, not enqueued\n", xprt);
334 goto out_unlock;
335 }
336
Greg Banks03cf6c92009-01-13 21:26:36 +1100337 pool->sp_stats.packets++;
338
Tom Tucker0f0257e2007-12-30 21:08:27 -0600339 /* Mark transport as busy. It will remain in this state until
340 * the provider calls svc_xprt_received. We update XPT_BUSY
341 * atomically because it also guards against trying to enqueue
342 * the transport twice.
343 */
344 if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) {
345 /* Don't enqueue transport while already enqueued */
346 dprintk("svc: transport %p busy, not enqueued\n", xprt);
347 goto out_unlock;
348 }
349 BUG_ON(xprt->xpt_pool != NULL);
350 xprt->xpt_pool = pool;
351
352 /* Handle pending connection */
353 if (test_bit(XPT_CONN, &xprt->xpt_flags))
354 goto process;
355
356 /* Handle close in-progress */
357 if (test_bit(XPT_CLOSE, &xprt->xpt_flags))
358 goto process;
359
360 /* Check if we have space to reply to a request */
361 if (!xprt->xpt_ops->xpo_has_wspace(xprt)) {
362 /* Don't enqueue while not enough space for reply */
363 dprintk("svc: no write space, transport %p not enqueued\n",
364 xprt);
365 xprt->xpt_pool = NULL;
366 clear_bit(XPT_BUSY, &xprt->xpt_flags);
367 goto out_unlock;
368 }
369
370 process:
J. Bruce Fields78c210e2009-08-06 15:41:34 -0400371 if (!list_empty(&pool->sp_threads)) {
Tom Tucker0f0257e2007-12-30 21:08:27 -0600372 rqstp = list_entry(pool->sp_threads.next,
373 struct svc_rqst,
374 rq_list);
375 dprintk("svc: transport %p served by daemon %p\n",
376 xprt, rqstp);
377 svc_thread_dequeue(pool, rqstp);
378 if (rqstp->rq_xprt)
379 printk(KERN_ERR
380 "svc_xprt_enqueue: server %p, rq_xprt=%p!\n",
381 rqstp, rqstp->rq_xprt);
382 rqstp->rq_xprt = xprt;
383 svc_xprt_get(xprt);
384 rqstp->rq_reserved = serv->sv_max_mesg;
385 atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
Greg Banks03cf6c92009-01-13 21:26:36 +1100386 pool->sp_stats.threads_woken++;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600387 BUG_ON(xprt->xpt_pool != pool);
388 wake_up(&rqstp->rq_wait);
389 } else {
390 dprintk("svc: transport %p put into queue\n", xprt);
391 list_add_tail(&xprt->xpt_ready, &pool->sp_sockets);
Greg Banks03cf6c92009-01-13 21:26:36 +1100392 pool->sp_stats.sockets_queued++;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600393 BUG_ON(xprt->xpt_pool != pool);
394 }
395
396out_unlock:
397 spin_unlock_bh(&pool->sp_lock);
398}
399EXPORT_SYMBOL_GPL(svc_xprt_enqueue);
400
401/*
402 * Dequeue the first transport. Must be called with the pool->sp_lock held.
403 */
404static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool)
405{
406 struct svc_xprt *xprt;
407
408 if (list_empty(&pool->sp_sockets))
409 return NULL;
410
411 xprt = list_entry(pool->sp_sockets.next,
412 struct svc_xprt, xpt_ready);
413 list_del_init(&xprt->xpt_ready);
414
415 dprintk("svc: transport %p dequeued, inuse=%d\n",
416 xprt, atomic_read(&xprt->xpt_ref.refcount));
417
418 return xprt;
419}
420
421/*
422 * svc_xprt_received conditionally queues the transport for processing
423 * by another thread. The caller must hold the XPT_BUSY bit and must
424 * not thereafter touch transport data.
425 *
426 * Note: XPT_DATA only gets cleared when a read-attempt finds no (or
427 * insufficient) data.
428 */
429void svc_xprt_received(struct svc_xprt *xprt)
430{
431 BUG_ON(!test_bit(XPT_BUSY, &xprt->xpt_flags));
432 xprt->xpt_pool = NULL;
433 clear_bit(XPT_BUSY, &xprt->xpt_flags);
434 svc_xprt_enqueue(xprt);
435}
436EXPORT_SYMBOL_GPL(svc_xprt_received);
437
438/**
439 * svc_reserve - change the space reserved for the reply to a request.
440 * @rqstp: The request in question
441 * @space: new max space to reserve
442 *
443 * Each request reserves some space on the output queue of the transport
444 * to make sure the reply fits. This function reduces that reserved
445 * space to be the amount of space used already, plus @space.
446 *
447 */
448void svc_reserve(struct svc_rqst *rqstp, int space)
449{
450 space += rqstp->rq_res.head[0].iov_len;
451
452 if (space < rqstp->rq_reserved) {
453 struct svc_xprt *xprt = rqstp->rq_xprt;
454 atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved);
455 rqstp->rq_reserved = space;
456
457 svc_xprt_enqueue(xprt);
458 }
459}
Trond Myklebust24c37672008-12-23 16:30:12 -0500460EXPORT_SYMBOL_GPL(svc_reserve);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600461
462static void svc_xprt_release(struct svc_rqst *rqstp)
463{
464 struct svc_xprt *xprt = rqstp->rq_xprt;
465
466 rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
467
Tom Tucker2779e3a2009-01-05 11:12:52 -0600468 kfree(rqstp->rq_deferred);
469 rqstp->rq_deferred = NULL;
470
Tom Tucker0f0257e2007-12-30 21:08:27 -0600471 svc_free_res_pages(rqstp);
472 rqstp->rq_res.page_len = 0;
473 rqstp->rq_res.page_base = 0;
474
475 /* Reset response buffer and release
476 * the reservation.
477 * But first, check that enough space was reserved
478 * for the reply, otherwise we have a bug!
479 */
480 if ((rqstp->rq_res.len) > rqstp->rq_reserved)
481 printk(KERN_ERR "RPC request reserved %d but used %d\n",
482 rqstp->rq_reserved,
483 rqstp->rq_res.len);
484
485 rqstp->rq_res.head[0].iov_len = 0;
486 svc_reserve(rqstp, 0);
487 rqstp->rq_xprt = NULL;
488
489 svc_xprt_put(xprt);
490}
491
492/*
493 * External function to wake up a server waiting for data
494 * This really only makes sense for services like lockd
495 * which have exactly one thread anyway.
496 */
497void svc_wake_up(struct svc_serv *serv)
498{
499 struct svc_rqst *rqstp;
500 unsigned int i;
501 struct svc_pool *pool;
502
503 for (i = 0; i < serv->sv_nrpools; i++) {
504 pool = &serv->sv_pools[i];
505
506 spin_lock_bh(&pool->sp_lock);
507 if (!list_empty(&pool->sp_threads)) {
508 rqstp = list_entry(pool->sp_threads.next,
509 struct svc_rqst,
510 rq_list);
511 dprintk("svc: daemon %p woken up.\n", rqstp);
512 /*
513 svc_thread_dequeue(pool, rqstp);
514 rqstp->rq_xprt = NULL;
515 */
516 wake_up(&rqstp->rq_wait);
517 }
518 spin_unlock_bh(&pool->sp_lock);
519 }
520}
Trond Myklebust24c37672008-12-23 16:30:12 -0500521EXPORT_SYMBOL_GPL(svc_wake_up);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600522
523int svc_port_is_privileged(struct sockaddr *sin)
524{
525 switch (sin->sa_family) {
526 case AF_INET:
527 return ntohs(((struct sockaddr_in *)sin)->sin_port)
528 < PROT_SOCK;
529 case AF_INET6:
530 return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
531 < PROT_SOCK;
532 default:
533 return 0;
534 }
535}
536
537/*
Jeff Laytonc9233eb2008-10-20 11:51:57 -0400538 * Make sure that we don't have too many active connections. If we have,
539 * something must be dropped. It's not clear what will happen if we allow
540 * "too many" connections, but when dealing with network-facing software,
541 * we have to code defensively. Here we do that by imposing hard limits.
Tom Tucker0f0257e2007-12-30 21:08:27 -0600542 *
543 * There's no point in trying to do random drop here for DoS
544 * prevention. The NFS clients does 1 reconnect in 15 seconds. An
545 * attacker can easily beat that.
546 *
547 * The only somewhat efficient mechanism would be if drop old
548 * connections from the same IP first. But right now we don't even
549 * record the client IP in svc_sock.
Jeff Laytonc9233eb2008-10-20 11:51:57 -0400550 *
551 * single-threaded services that expect a lot of clients will probably
552 * need to set sv_maxconn to override the default value which is based
553 * on the number of threads
Tom Tucker0f0257e2007-12-30 21:08:27 -0600554 */
555static void svc_check_conn_limits(struct svc_serv *serv)
556{
Jeff Laytonc9233eb2008-10-20 11:51:57 -0400557 unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn :
558 (serv->sv_nrthreads+3) * 20;
559
560 if (serv->sv_tmpcnt > limit) {
Tom Tucker0f0257e2007-12-30 21:08:27 -0600561 struct svc_xprt *xprt = NULL;
562 spin_lock_bh(&serv->sv_lock);
563 if (!list_empty(&serv->sv_tempsocks)) {
564 if (net_ratelimit()) {
565 /* Try to help the admin */
566 printk(KERN_NOTICE "%s: too many open "
Jeff Laytonc9233eb2008-10-20 11:51:57 -0400567 "connections, consider increasing %s\n",
568 serv->sv_name, serv->sv_maxconn ?
569 "the max number of connections." :
570 "the number of threads.");
Tom Tucker0f0257e2007-12-30 21:08:27 -0600571 }
572 /*
573 * Always select the oldest connection. It's not fair,
574 * but so is life
575 */
576 xprt = list_entry(serv->sv_tempsocks.prev,
577 struct svc_xprt,
578 xpt_list);
579 set_bit(XPT_CLOSE, &xprt->xpt_flags);
580 svc_xprt_get(xprt);
581 }
582 spin_unlock_bh(&serv->sv_lock);
583
584 if (xprt) {
585 svc_xprt_enqueue(xprt);
586 svc_xprt_put(xprt);
587 }
588 }
589}
590
591/*
592 * Receive the next request on any transport. This code is carefully
593 * organised not to touch any cachelines in the shared svc_serv
594 * structure, only cachelines in the local svc_pool.
595 */
596int svc_recv(struct svc_rqst *rqstp, long timeout)
597{
598 struct svc_xprt *xprt = NULL;
599 struct svc_serv *serv = rqstp->rq_server;
600 struct svc_pool *pool = rqstp->rq_pool;
601 int len, i;
602 int pages;
603 struct xdr_buf *arg;
604 DECLARE_WAITQUEUE(wait, current);
Greg Banks03cf6c92009-01-13 21:26:36 +1100605 long time_left;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600606
607 dprintk("svc: server %p waiting for data (to = %ld)\n",
608 rqstp, timeout);
609
610 if (rqstp->rq_xprt)
611 printk(KERN_ERR
612 "svc_recv: service %p, transport not NULL!\n",
613 rqstp);
614 if (waitqueue_active(&rqstp->rq_wait))
615 printk(KERN_ERR
616 "svc_recv: service %p, wait queue active!\n",
617 rqstp);
618
619 /* now allocate needed pages. If we get a failure, sleep briefly */
620 pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE;
621 for (i = 0; i < pages ; i++)
622 while (rqstp->rq_pages[i] == NULL) {
623 struct page *p = alloc_page(GFP_KERNEL);
624 if (!p) {
Jeff Layton7b54fe62008-02-12 11:47:24 -0500625 set_current_state(TASK_INTERRUPTIBLE);
626 if (signalled() || kthread_should_stop()) {
627 set_current_state(TASK_RUNNING);
Jeff Layton70867212008-02-07 16:34:54 -0500628 return -EINTR;
Jeff Layton7b54fe62008-02-12 11:47:24 -0500629 }
630 schedule_timeout(msecs_to_jiffies(500));
Tom Tucker0f0257e2007-12-30 21:08:27 -0600631 }
632 rqstp->rq_pages[i] = p;
633 }
634 rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */
635 BUG_ON(pages >= RPCSVC_MAXPAGES);
636
637 /* Make arg->head point to first page and arg->pages point to rest */
638 arg = &rqstp->rq_arg;
639 arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
640 arg->head[0].iov_len = PAGE_SIZE;
641 arg->pages = rqstp->rq_pages + 1;
642 arg->page_base = 0;
643 /* save at least one page for response */
644 arg->page_len = (pages-2)*PAGE_SIZE;
645 arg->len = (pages-1)*PAGE_SIZE;
646 arg->tail[0].iov_len = 0;
647
648 try_to_freeze();
649 cond_resched();
Jeff Layton70867212008-02-07 16:34:54 -0500650 if (signalled() || kthread_should_stop())
Tom Tucker0f0257e2007-12-30 21:08:27 -0600651 return -EINTR;
652
653 spin_lock_bh(&pool->sp_lock);
654 xprt = svc_xprt_dequeue(pool);
655 if (xprt) {
656 rqstp->rq_xprt = xprt;
657 svc_xprt_get(xprt);
658 rqstp->rq_reserved = serv->sv_max_mesg;
659 atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
660 } else {
661 /* No data pending. Go to sleep */
662 svc_thread_enqueue(pool, rqstp);
663
664 /*
665 * We have to be able to interrupt this wait
666 * to bring down the daemons ...
667 */
668 set_current_state(TASK_INTERRUPTIBLE);
Jeff Layton70867212008-02-07 16:34:54 -0500669
670 /*
671 * checking kthread_should_stop() here allows us to avoid
672 * locking and signalling when stopping kthreads that call
673 * svc_recv. If the thread has already been woken up, then
674 * we can exit here without sleeping. If not, then it
675 * it'll be woken up quickly during the schedule_timeout
676 */
677 if (kthread_should_stop()) {
678 set_current_state(TASK_RUNNING);
679 spin_unlock_bh(&pool->sp_lock);
680 return -EINTR;
681 }
682
Tom Tucker0f0257e2007-12-30 21:08:27 -0600683 add_wait_queue(&rqstp->rq_wait, &wait);
684 spin_unlock_bh(&pool->sp_lock);
685
Greg Banks03cf6c92009-01-13 21:26:36 +1100686 time_left = schedule_timeout(timeout);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600687
688 try_to_freeze();
689
690 spin_lock_bh(&pool->sp_lock);
691 remove_wait_queue(&rqstp->rq_wait, &wait);
Greg Banks03cf6c92009-01-13 21:26:36 +1100692 if (!time_left)
693 pool->sp_stats.threads_timedout++;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600694
695 xprt = rqstp->rq_xprt;
696 if (!xprt) {
697 svc_thread_dequeue(pool, rqstp);
698 spin_unlock_bh(&pool->sp_lock);
699 dprintk("svc: server %p, no data yet\n", rqstp);
Jeff Layton70867212008-02-07 16:34:54 -0500700 if (signalled() || kthread_should_stop())
701 return -EINTR;
702 else
703 return -EAGAIN;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600704 }
705 }
706 spin_unlock_bh(&pool->sp_lock);
707
708 len = 0;
J. Bruce Fields1b644b62010-02-28 16:33:31 -0500709 if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) {
710 dprintk("svc_recv: found XPT_CLOSE\n");
711 svc_delete_xprt(xprt);
712 } else if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
Tom Tucker0f0257e2007-12-30 21:08:27 -0600713 struct svc_xprt *newxpt;
714 newxpt = xprt->xpt_ops->xpo_accept(xprt);
715 if (newxpt) {
716 /*
717 * We know this module_get will succeed because the
718 * listener holds a reference too
719 */
720 __module_get(newxpt->xpt_class->xcl_owner);
721 svc_check_conn_limits(xprt->xpt_server);
722 spin_lock_bh(&serv->sv_lock);
723 set_bit(XPT_TEMP, &newxpt->xpt_flags);
724 list_add(&newxpt->xpt_list, &serv->sv_tempsocks);
725 serv->sv_tmpcnt++;
726 if (serv->sv_temptimer.function == NULL) {
727 /* setup timer to age temp transports */
728 setup_timer(&serv->sv_temptimer,
729 svc_age_temp_xprts,
730 (unsigned long)serv);
731 mod_timer(&serv->sv_temptimer,
732 jiffies + svc_conn_age_period * HZ);
733 }
734 spin_unlock_bh(&serv->sv_lock);
735 svc_xprt_received(newxpt);
736 }
737 svc_xprt_received(xprt);
J. Bruce Fields1b644b62010-02-28 16:33:31 -0500738 } else {
Tom Tucker0f0257e2007-12-30 21:08:27 -0600739 dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n",
740 rqstp, pool->sp_id, xprt,
741 atomic_read(&xprt->xpt_ref.refcount));
742 rqstp->rq_deferred = svc_deferred_dequeue(xprt);
743 if (rqstp->rq_deferred) {
744 svc_xprt_received(xprt);
745 len = svc_deferred_recv(rqstp);
746 } else
747 len = xprt->xpt_ops->xpo_recvfrom(rqstp);
748 dprintk("svc: got len=%d\n", len);
749 }
750
751 /* No data, incomplete (TCP) read, or accept() */
752 if (len == 0 || len == -EAGAIN) {
753 rqstp->rq_res.len = 0;
754 svc_xprt_release(rqstp);
755 return -EAGAIN;
756 }
757 clear_bit(XPT_OLD, &xprt->xpt_flags);
758
759 rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp));
760 rqstp->rq_chandle.defer = svc_defer;
761
762 if (serv->sv_stats)
763 serv->sv_stats->netcnt++;
764 return len;
765}
Trond Myklebust24c37672008-12-23 16:30:12 -0500766EXPORT_SYMBOL_GPL(svc_recv);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600767
768/*
769 * Drop request
770 */
771void svc_drop(struct svc_rqst *rqstp)
772{
773 dprintk("svc: xprt %p dropped request\n", rqstp->rq_xprt);
774 svc_xprt_release(rqstp);
775}
Trond Myklebust24c37672008-12-23 16:30:12 -0500776EXPORT_SYMBOL_GPL(svc_drop);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600777
778/*
779 * Return reply to client.
780 */
781int svc_send(struct svc_rqst *rqstp)
782{
783 struct svc_xprt *xprt;
784 int len;
785 struct xdr_buf *xb;
786
787 xprt = rqstp->rq_xprt;
788 if (!xprt)
789 return -EFAULT;
790
791 /* release the receive skb before sending the reply */
792 rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
793
794 /* calculate over-all length */
795 xb = &rqstp->rq_res;
796 xb->len = xb->head[0].iov_len +
797 xb->page_len +
798 xb->tail[0].iov_len;
799
800 /* Grab mutex to serialize outgoing data. */
801 mutex_lock(&xprt->xpt_mutex);
802 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
803 len = -ENOTCONN;
804 else
805 len = xprt->xpt_ops->xpo_sendto(rqstp);
806 mutex_unlock(&xprt->xpt_mutex);
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300807 rpc_wake_up(&xprt->xpt_bc_pending);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600808 svc_xprt_release(rqstp);
809
810 if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
811 return 0;
812 return len;
813}
814
815/*
816 * Timer function to close old temporary transports, using
817 * a mark-and-sweep algorithm.
818 */
819static void svc_age_temp_xprts(unsigned long closure)
820{
821 struct svc_serv *serv = (struct svc_serv *)closure;
822 struct svc_xprt *xprt;
823 struct list_head *le, *next;
824 LIST_HEAD(to_be_aged);
825
826 dprintk("svc_age_temp_xprts\n");
827
828 if (!spin_trylock_bh(&serv->sv_lock)) {
829 /* busy, try again 1 sec later */
830 dprintk("svc_age_temp_xprts: busy\n");
831 mod_timer(&serv->sv_temptimer, jiffies + HZ);
832 return;
833 }
834
835 list_for_each_safe(le, next, &serv->sv_tempsocks) {
836 xprt = list_entry(le, struct svc_xprt, xpt_list);
837
838 /* First time through, just mark it OLD. Second time
839 * through, close it. */
840 if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags))
841 continue;
Joe Perchesf64f9e72009-11-29 16:55:45 -0800842 if (atomic_read(&xprt->xpt_ref.refcount) > 1 ||
843 test_bit(XPT_BUSY, &xprt->xpt_flags))
Tom Tucker0f0257e2007-12-30 21:08:27 -0600844 continue;
845 svc_xprt_get(xprt);
846 list_move(le, &to_be_aged);
847 set_bit(XPT_CLOSE, &xprt->xpt_flags);
848 set_bit(XPT_DETACHED, &xprt->xpt_flags);
849 }
850 spin_unlock_bh(&serv->sv_lock);
851
852 while (!list_empty(&to_be_aged)) {
853 le = to_be_aged.next;
854 /* fiddling the xpt_list node is safe 'cos we're XPT_DETACHED */
855 list_del_init(le);
856 xprt = list_entry(le, struct svc_xprt, xpt_list);
857
858 dprintk("queuing xprt %p for closing\n", xprt);
859
860 /* a thread will dequeue and close it soon */
861 svc_xprt_enqueue(xprt);
862 svc_xprt_put(xprt);
863 }
864
865 mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
866}
867
868/*
869 * Remove a dead transport
870 */
871void svc_delete_xprt(struct svc_xprt *xprt)
872{
873 struct svc_serv *serv = xprt->xpt_server;
Tom Tucker22945e42009-01-05 15:21:19 -0600874 struct svc_deferred_req *dr;
875
876 /* Only do this once */
877 if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags))
878 return;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600879
880 dprintk("svc: svc_delete_xprt(%p)\n", xprt);
881 xprt->xpt_ops->xpo_detach(xprt);
882
883 spin_lock_bh(&serv->sv_lock);
884 if (!test_and_set_bit(XPT_DETACHED, &xprt->xpt_flags))
885 list_del_init(&xprt->xpt_list);
886 /*
887 * We used to delete the transport from whichever list
888 * it's sk_xprt.xpt_ready node was on, but we don't actually
889 * need to. This is because the only time we're called
890 * while still attached to a queue, the queue itself
891 * is about to be destroyed (in svc_destroy).
892 */
Tom Tucker22945e42009-01-05 15:21:19 -0600893 if (test_bit(XPT_TEMP, &xprt->xpt_flags))
894 serv->sv_tmpcnt--;
895
Neil Brownab1b18f2010-02-27 09:33:40 +1100896 while ((dr = svc_deferred_dequeue(xprt)) != NULL)
Tom Tucker22945e42009-01-05 15:21:19 -0600897 kfree(dr);
Tom Tucker22945e42009-01-05 15:21:19 -0600898
899 svc_xprt_put(xprt);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600900 spin_unlock_bh(&serv->sv_lock);
901}
902
903void svc_close_xprt(struct svc_xprt *xprt)
904{
905 set_bit(XPT_CLOSE, &xprt->xpt_flags);
906 if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
907 /* someone else will have to effect the close */
908 return;
909
910 svc_xprt_get(xprt);
911 svc_delete_xprt(xprt);
912 clear_bit(XPT_BUSY, &xprt->xpt_flags);
913 svc_xprt_put(xprt);
914}
Tom Tuckera2178132007-12-30 21:08:35 -0600915EXPORT_SYMBOL_GPL(svc_close_xprt);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600916
917void svc_close_all(struct list_head *xprt_list)
918{
919 struct svc_xprt *xprt;
920 struct svc_xprt *tmp;
921
922 list_for_each_entry_safe(xprt, tmp, xprt_list, xpt_list) {
923 set_bit(XPT_CLOSE, &xprt->xpt_flags);
924 if (test_bit(XPT_BUSY, &xprt->xpt_flags)) {
925 /* Waiting to be processed, but no threads left,
926 * So just remove it from the waiting list
927 */
928 list_del_init(&xprt->xpt_ready);
929 clear_bit(XPT_BUSY, &xprt->xpt_flags);
930 }
931 svc_close_xprt(xprt);
932 }
933}
934
935/*
936 * Handle defer and revisit of requests
937 */
938
939static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
940{
941 struct svc_deferred_req *dr =
942 container_of(dreq, struct svc_deferred_req, handle);
943 struct svc_xprt *xprt = dr->xprt;
944
Tom Tucker22945e42009-01-05 15:21:19 -0600945 spin_lock(&xprt->xpt_lock);
946 set_bit(XPT_DEFERRED, &xprt->xpt_flags);
947 if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) {
948 spin_unlock(&xprt->xpt_lock);
949 dprintk("revisit canceled\n");
Tom Tucker0f0257e2007-12-30 21:08:27 -0600950 svc_xprt_put(xprt);
951 kfree(dr);
952 return;
953 }
954 dprintk("revisit queued\n");
955 dr->xprt = NULL;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600956 list_add(&dr->handle.recent, &xprt->xpt_deferred);
957 spin_unlock(&xprt->xpt_lock);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600958 svc_xprt_enqueue(xprt);
959 svc_xprt_put(xprt);
960}
961
Tom Tucker260c1d12007-12-30 21:08:29 -0600962/*
963 * Save the request off for later processing. The request buffer looks
964 * like this:
965 *
966 * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
967 *
968 * This code can only handle requests that consist of an xprt-header
969 * and rpc-header.
970 */
Tom Tucker0f0257e2007-12-30 21:08:27 -0600971static struct cache_deferred_req *svc_defer(struct cache_req *req)
972{
973 struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600974 struct svc_deferred_req *dr;
975
Andy Adamson2f425872009-04-03 08:27:32 +0300976 if (rqstp->rq_arg.page_len || !rqstp->rq_usedeferral)
Tom Tucker0f0257e2007-12-30 21:08:27 -0600977 return NULL; /* if more than a page, give up FIXME */
978 if (rqstp->rq_deferred) {
979 dr = rqstp->rq_deferred;
980 rqstp->rq_deferred = NULL;
981 } else {
Tom Tucker260c1d12007-12-30 21:08:29 -0600982 size_t skip;
983 size_t size;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600984 /* FIXME maybe discard if size too large */
Tom Tucker260c1d12007-12-30 21:08:29 -0600985 size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600986 dr = kmalloc(size, GFP_KERNEL);
987 if (dr == NULL)
988 return NULL;
989
990 dr->handle.owner = rqstp->rq_server;
991 dr->prot = rqstp->rq_prot;
992 memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
993 dr->addrlen = rqstp->rq_addrlen;
994 dr->daddr = rqstp->rq_daddr;
995 dr->argslen = rqstp->rq_arg.len >> 2;
Tom Tucker260c1d12007-12-30 21:08:29 -0600996 dr->xprt_hlen = rqstp->rq_xprt_hlen;
997
998 /* back up head to the start of the buffer and copy */
999 skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1000 memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
1001 dr->argslen << 2);
Tom Tucker0f0257e2007-12-30 21:08:27 -06001002 }
1003 svc_xprt_get(rqstp->rq_xprt);
1004 dr->xprt = rqstp->rq_xprt;
1005
1006 dr->handle.revisit = svc_revisit;
1007 return &dr->handle;
1008}
1009
1010/*
1011 * recv data from a deferred request into an active one
1012 */
1013static int svc_deferred_recv(struct svc_rqst *rqstp)
1014{
1015 struct svc_deferred_req *dr = rqstp->rq_deferred;
1016
Tom Tucker260c1d12007-12-30 21:08:29 -06001017 /* setup iov_base past transport header */
1018 rqstp->rq_arg.head[0].iov_base = dr->args + (dr->xprt_hlen>>2);
1019 /* The iov_len does not include the transport header bytes */
1020 rqstp->rq_arg.head[0].iov_len = (dr->argslen<<2) - dr->xprt_hlen;
Tom Tucker0f0257e2007-12-30 21:08:27 -06001021 rqstp->rq_arg.page_len = 0;
Tom Tucker260c1d12007-12-30 21:08:29 -06001022 /* The rq_arg.len includes the transport header bytes */
1023 rqstp->rq_arg.len = dr->argslen<<2;
Tom Tucker0f0257e2007-12-30 21:08:27 -06001024 rqstp->rq_prot = dr->prot;
1025 memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
1026 rqstp->rq_addrlen = dr->addrlen;
Tom Tucker260c1d12007-12-30 21:08:29 -06001027 /* Save off transport header len in case we get deferred again */
1028 rqstp->rq_xprt_hlen = dr->xprt_hlen;
Tom Tucker0f0257e2007-12-30 21:08:27 -06001029 rqstp->rq_daddr = dr->daddr;
1030 rqstp->rq_respages = rqstp->rq_pages;
Tom Tucker260c1d12007-12-30 21:08:29 -06001031 return (dr->argslen<<2) - dr->xprt_hlen;
Tom Tucker0f0257e2007-12-30 21:08:27 -06001032}
1033
1034
1035static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt)
1036{
1037 struct svc_deferred_req *dr = NULL;
1038
1039 if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags))
1040 return NULL;
1041 spin_lock(&xprt->xpt_lock);
1042 clear_bit(XPT_DEFERRED, &xprt->xpt_flags);
1043 if (!list_empty(&xprt->xpt_deferred)) {
1044 dr = list_entry(xprt->xpt_deferred.next,
1045 struct svc_deferred_req,
1046 handle.recent);
1047 list_del_init(&dr->handle.recent);
1048 set_bit(XPT_DEFERRED, &xprt->xpt_flags);
1049 }
1050 spin_unlock(&xprt->xpt_lock);
1051 return dr;
1052}
Tom Tucker7fcb98d2007-12-30 21:08:33 -06001053
Chuck Lever156e6202009-03-18 20:45:58 -04001054/**
1055 * svc_find_xprt - find an RPC transport instance
1056 * @serv: pointer to svc_serv to search
1057 * @xcl_name: C string containing transport's class name
1058 * @af: Address family of transport's local address
1059 * @port: transport's IP port number
1060 *
Tom Tucker7fcb98d2007-12-30 21:08:33 -06001061 * Return the transport instance pointer for the endpoint accepting
1062 * connections/peer traffic from the specified transport class,
1063 * address family and port.
1064 *
1065 * Specifying 0 for the address family or port is effectively a
1066 * wild-card, and will result in matching the first transport in the
1067 * service's list that has a matching class name.
1068 */
Chuck Lever156e6202009-03-18 20:45:58 -04001069struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name,
1070 const sa_family_t af, const unsigned short port)
Tom Tucker7fcb98d2007-12-30 21:08:33 -06001071{
1072 struct svc_xprt *xprt;
1073 struct svc_xprt *found = NULL;
1074
1075 /* Sanity check the args */
Chuck Lever156e6202009-03-18 20:45:58 -04001076 if (serv == NULL || xcl_name == NULL)
Tom Tucker7fcb98d2007-12-30 21:08:33 -06001077 return found;
1078
1079 spin_lock_bh(&serv->sv_lock);
1080 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1081 if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1082 continue;
1083 if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family)
1084 continue;
Chuck Lever156e6202009-03-18 20:45:58 -04001085 if (port != 0 && port != svc_xprt_local_port(xprt))
Tom Tucker7fcb98d2007-12-30 21:08:33 -06001086 continue;
1087 found = xprt;
Tom Tuckera2178132007-12-30 21:08:35 -06001088 svc_xprt_get(xprt);
Tom Tucker7fcb98d2007-12-30 21:08:33 -06001089 break;
1090 }
1091 spin_unlock_bh(&serv->sv_lock);
1092 return found;
1093}
1094EXPORT_SYMBOL_GPL(svc_find_xprt);
Tom Tucker9571af12007-12-30 21:08:37 -06001095
Chuck Lever335c54b2009-04-23 19:32:25 -04001096static int svc_one_xprt_name(const struct svc_xprt *xprt,
1097 char *pos, int remaining)
1098{
1099 int len;
1100
1101 len = snprintf(pos, remaining, "%s %u\n",
1102 xprt->xpt_class->xcl_name,
1103 svc_xprt_local_port(xprt));
1104 if (len >= remaining)
1105 return -ENAMETOOLONG;
1106 return len;
1107}
1108
1109/**
1110 * svc_xprt_names - format a buffer with a list of transport names
1111 * @serv: pointer to an RPC service
1112 * @buf: pointer to a buffer to be filled in
1113 * @buflen: length of buffer to be filled in
1114 *
1115 * Fills in @buf with a string containing a list of transport names,
1116 * each name terminated with '\n'.
1117 *
1118 * Returns positive length of the filled-in string on success; otherwise
1119 * a negative errno value is returned if an error occurs.
Tom Tucker9571af12007-12-30 21:08:37 -06001120 */
Chuck Lever335c54b2009-04-23 19:32:25 -04001121int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen)
Tom Tucker9571af12007-12-30 21:08:37 -06001122{
1123 struct svc_xprt *xprt;
Chuck Lever335c54b2009-04-23 19:32:25 -04001124 int len, totlen;
1125 char *pos;
Tom Tucker9571af12007-12-30 21:08:37 -06001126
1127 /* Sanity check args */
1128 if (!serv)
1129 return 0;
1130
1131 spin_lock_bh(&serv->sv_lock);
Chuck Lever335c54b2009-04-23 19:32:25 -04001132
1133 pos = buf;
1134 totlen = 0;
Tom Tucker9571af12007-12-30 21:08:37 -06001135 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
Chuck Lever335c54b2009-04-23 19:32:25 -04001136 len = svc_one_xprt_name(xprt, pos, buflen - totlen);
1137 if (len < 0) {
1138 *buf = '\0';
1139 totlen = len;
1140 }
1141 if (len <= 0)
Tom Tucker9571af12007-12-30 21:08:37 -06001142 break;
Chuck Lever335c54b2009-04-23 19:32:25 -04001143
1144 pos += len;
Tom Tucker9571af12007-12-30 21:08:37 -06001145 totlen += len;
1146 }
Chuck Lever335c54b2009-04-23 19:32:25 -04001147
Tom Tucker9571af12007-12-30 21:08:37 -06001148 spin_unlock_bh(&serv->sv_lock);
1149 return totlen;
1150}
1151EXPORT_SYMBOL_GPL(svc_xprt_names);
Greg Banks03cf6c92009-01-13 21:26:36 +11001152
1153
1154/*----------------------------------------------------------------------------*/
1155
1156static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos)
1157{
1158 unsigned int pidx = (unsigned int)*pos;
1159 struct svc_serv *serv = m->private;
1160
1161 dprintk("svc_pool_stats_start, *pidx=%u\n", pidx);
1162
Greg Banks03cf6c92009-01-13 21:26:36 +11001163 if (!pidx)
1164 return SEQ_START_TOKEN;
1165 return (pidx > serv->sv_nrpools ? NULL : &serv->sv_pools[pidx-1]);
1166}
1167
1168static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos)
1169{
1170 struct svc_pool *pool = p;
1171 struct svc_serv *serv = m->private;
1172
1173 dprintk("svc_pool_stats_next, *pos=%llu\n", *pos);
1174
1175 if (p == SEQ_START_TOKEN) {
1176 pool = &serv->sv_pools[0];
1177 } else {
1178 unsigned int pidx = (pool - &serv->sv_pools[0]);
1179 if (pidx < serv->sv_nrpools-1)
1180 pool = &serv->sv_pools[pidx+1];
1181 else
1182 pool = NULL;
1183 }
1184 ++*pos;
1185 return pool;
1186}
1187
1188static void svc_pool_stats_stop(struct seq_file *m, void *p)
1189{
Greg Banks03cf6c92009-01-13 21:26:36 +11001190}
1191
1192static int svc_pool_stats_show(struct seq_file *m, void *p)
1193{
1194 struct svc_pool *pool = p;
1195
1196 if (p == SEQ_START_TOKEN) {
J. Bruce Fields78c210e2009-08-06 15:41:34 -04001197 seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n");
Greg Banks03cf6c92009-01-13 21:26:36 +11001198 return 0;
1199 }
1200
J. Bruce Fields78c210e2009-08-06 15:41:34 -04001201 seq_printf(m, "%u %lu %lu %lu %lu\n",
Greg Banks03cf6c92009-01-13 21:26:36 +11001202 pool->sp_id,
1203 pool->sp_stats.packets,
1204 pool->sp_stats.sockets_queued,
1205 pool->sp_stats.threads_woken,
Greg Banks03cf6c92009-01-13 21:26:36 +11001206 pool->sp_stats.threads_timedout);
1207
1208 return 0;
1209}
1210
1211static const struct seq_operations svc_pool_stats_seq_ops = {
1212 .start = svc_pool_stats_start,
1213 .next = svc_pool_stats_next,
1214 .stop = svc_pool_stats_stop,
1215 .show = svc_pool_stats_show,
1216};
1217
1218int svc_pool_stats_open(struct svc_serv *serv, struct file *file)
1219{
1220 int err;
1221
1222 err = seq_open(file, &svc_pool_stats_seq_ops);
1223 if (!err)
1224 ((struct seq_file *) file->private_data)->private = serv;
1225 return err;
1226}
1227EXPORT_SYMBOL(svc_pool_stats_open);
1228
1229/*----------------------------------------------------------------------------*/