blob: bef1e8844a2c028673fdd2594b681fd757461626 [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090012#include <linux/slab.h>
Tom Tucker1d8206b92007-12-30 21:07:15 -060013#include <net/sock.h>
Tom Tucker1d8206b92007-12-30 21:07:15 -060014#include <linux/sunrpc/stats.h>
15#include <linux/sunrpc/svc_xprt.h>
H Hartley Sweetendcf1a352009-04-22 20:18:19 -040016#include <linux/sunrpc/svcsock.h>
Tom Tucker1d8206b92007-12-30 21:07:15 -060017
18#define RPCDBG_FACILITY RPCDBG_SVCXPRT
19
Tom Tucker0f0257e2007-12-30 21:08:27 -060020static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt);
21static int svc_deferred_recv(struct svc_rqst *rqstp);
22static struct cache_deferred_req *svc_defer(struct cache_req *req);
23static void svc_age_temp_xprts(unsigned long closure);
24
25/* apparently the "standard" is that clients close
26 * idle connections after 5 minutes, servers after
27 * 6 minutes
28 * http://www.connectathon.org/talks96/nfstcp.pdf
29 */
30static int svc_conn_age_period = 6*60;
31
Tom Tucker1d8206b92007-12-30 21:07:15 -060032/* List of registered transport classes */
33static DEFINE_SPINLOCK(svc_xprt_class_lock);
34static LIST_HEAD(svc_xprt_class_list);
35
Tom Tucker0f0257e2007-12-30 21:08:27 -060036/* SMP locking strategy:
37 *
38 * svc_pool->sp_lock protects most of the fields of that pool.
39 * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
40 * when both need to be taken (rare), svc_serv->sv_lock is first.
41 * BKL protects svc_serv->sv_nrthread.
42 * svc_sock->sk_lock protects the svc_sock->sk_deferred list
43 * and the ->sk_info_authunix cache.
44 *
45 * The XPT_BUSY bit in xprt->xpt_flags prevents a transport being
46 * enqueued multiply. During normal transport processing this bit
47 * is set by svc_xprt_enqueue and cleared by svc_xprt_received.
48 * Providers should not manipulate this bit directly.
49 *
50 * Some flags can be set to certain values at any time
51 * providing that certain rules are followed:
52 *
53 * XPT_CONN, XPT_DATA:
54 * - Can be set or cleared at any time.
55 * - After a set, svc_xprt_enqueue must be called to enqueue
56 * the transport for processing.
57 * - After a clear, the transport must be read/accepted.
58 * If this succeeds, it must be set again.
59 * XPT_CLOSE:
60 * - Can set at any time. It is never cleared.
61 * XPT_DEAD:
62 * - Can only be set while XPT_BUSY is held which ensures
63 * that no other thread will be using the transport or will
64 * try to set XPT_DEAD.
65 */
66
Tom Tucker1d8206b92007-12-30 21:07:15 -060067int svc_reg_xprt_class(struct svc_xprt_class *xcl)
68{
69 struct svc_xprt_class *cl;
70 int res = -EEXIST;
71
72 dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name);
73
74 INIT_LIST_HEAD(&xcl->xcl_list);
75 spin_lock(&svc_xprt_class_lock);
76 /* Make sure there isn't already a class with the same name */
77 list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) {
78 if (strcmp(xcl->xcl_name, cl->xcl_name) == 0)
79 goto out;
80 }
81 list_add_tail(&xcl->xcl_list, &svc_xprt_class_list);
82 res = 0;
83out:
84 spin_unlock(&svc_xprt_class_lock);
85 return res;
86}
87EXPORT_SYMBOL_GPL(svc_reg_xprt_class);
88
89void svc_unreg_xprt_class(struct svc_xprt_class *xcl)
90{
91 dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name);
92 spin_lock(&svc_xprt_class_lock);
93 list_del_init(&xcl->xcl_list);
94 spin_unlock(&svc_xprt_class_lock);
95}
96EXPORT_SYMBOL_GPL(svc_unreg_xprt_class);
97
Tom Tuckerdc9a16e2007-12-30 21:08:31 -060098/*
99 * Format the transport list for printing
100 */
101int svc_print_xprts(char *buf, int maxlen)
102{
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +0400103 struct svc_xprt_class *xcl;
Tom Tuckerdc9a16e2007-12-30 21:08:31 -0600104 char tmpstr[80];
105 int len = 0;
106 buf[0] = '\0';
107
108 spin_lock(&svc_xprt_class_lock);
Pavel Emelyanov8f3a6de2010-10-05 23:30:19 +0400109 list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
Tom Tuckerdc9a16e2007-12-30 21:08:31 -0600110 int slen;
Tom Tuckerdc9a16e2007-12-30 21:08:31 -0600111
112 sprintf(tmpstr, "%s %d\n", xcl->xcl_name, xcl->xcl_max_payload);
113 slen = strlen(tmpstr);
114 if (len + slen > maxlen)
115 break;
116 len += slen;
117 strcat(buf, tmpstr);
118 }
119 spin_unlock(&svc_xprt_class_lock);
120
121 return len;
122}
123
Tom Tuckere1b31572007-12-30 21:07:46 -0600124static void svc_xprt_free(struct kref *kref)
125{
126 struct svc_xprt *xprt =
127 container_of(kref, struct svc_xprt, xpt_ref);
128 struct module *owner = xprt->xpt_class->xcl_owner;
Pavel Emelyanove3bfca02010-09-27 13:58:42 +0400129 if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags))
130 svcauth_unix_info_release(xprt);
Pavel Emelyanov4fb85182010-09-27 14:00:49 +0400131 put_net(xprt->xpt_net);
Tom Tuckere1b31572007-12-30 21:07:46 -0600132 xprt->xpt_ops->xpo_free(xprt);
133 module_put(owner);
134}
135
136void svc_xprt_put(struct svc_xprt *xprt)
137{
138 kref_put(&xprt->xpt_ref, svc_xprt_free);
139}
140EXPORT_SYMBOL_GPL(svc_xprt_put);
141
Tom Tucker1d8206b92007-12-30 21:07:15 -0600142/*
143 * Called by transport drivers to initialize the transport independent
144 * portion of the transport instance.
145 */
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600146void svc_xprt_init(struct svc_xprt_class *xcl, struct svc_xprt *xprt,
147 struct svc_serv *serv)
Tom Tucker1d8206b92007-12-30 21:07:15 -0600148{
149 memset(xprt, 0, sizeof(*xprt));
150 xprt->xpt_class = xcl;
151 xprt->xpt_ops = xcl->xcl_ops;
Tom Tuckere1b31572007-12-30 21:07:46 -0600152 kref_init(&xprt->xpt_ref);
Tom Tuckerbb5cf162007-12-30 21:07:50 -0600153 xprt->xpt_server = serv;
Tom Tucker7a182082007-12-30 21:07:53 -0600154 INIT_LIST_HEAD(&xprt->xpt_list);
155 INIT_LIST_HEAD(&xprt->xpt_ready);
Tom Tucker8c7b0172007-12-30 21:08:10 -0600156 INIT_LIST_HEAD(&xprt->xpt_deferred);
J. Bruce Fieldsedc7a892010-03-22 15:37:17 -0400157 INIT_LIST_HEAD(&xprt->xpt_users);
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");
Pavel Emelyanov4fb85182010-09-27 14:00:49 +0400162 xprt->xpt_net = get_net(&init_net);
Tom Tucker1d8206b92007-12-30 21:07:15 -0600163}
164EXPORT_SYMBOL_GPL(svc_xprt_init);
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600165
Chuck Lever5dd248f2008-06-30 18:45:37 -0400166static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl,
167 struct svc_serv *serv,
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400168 struct net *net,
Chuck Lever9652ada2009-03-18 20:46:21 -0400169 const int family,
170 const unsigned short port,
171 int flags)
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600172{
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600173 struct sockaddr_in sin = {
174 .sin_family = AF_INET,
Al Viroe6f1ceb2008-03-17 22:44:53 -0700175 .sin_addr.s_addr = htonl(INADDR_ANY),
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600176 .sin_port = htons(port),
177 };
Chuck Leverd6783b22010-01-26 14:04:04 -0500178#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
Chuck Lever5dd248f2008-06-30 18:45:37 -0400179 struct sockaddr_in6 sin6 = {
180 .sin6_family = AF_INET6,
181 .sin6_addr = IN6ADDR_ANY_INIT,
182 .sin6_port = htons(port),
183 };
Chuck Leverd6783b22010-01-26 14:04:04 -0500184#endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
Chuck Lever5dd248f2008-06-30 18:45:37 -0400185 struct sockaddr *sap;
186 size_t len;
187
Chuck Lever9652ada2009-03-18 20:46:21 -0400188 switch (family) {
189 case PF_INET:
Chuck Lever5dd248f2008-06-30 18:45:37 -0400190 sap = (struct sockaddr *)&sin;
191 len = sizeof(sin);
192 break;
Chuck Leverd6783b22010-01-26 14:04:04 -0500193#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
Chuck Lever9652ada2009-03-18 20:46:21 -0400194 case PF_INET6:
Chuck Lever5dd248f2008-06-30 18:45:37 -0400195 sap = (struct sockaddr *)&sin6;
196 len = sizeof(sin6);
197 break;
Chuck Leverd6783b22010-01-26 14:04:04 -0500198#endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
Chuck Lever5dd248f2008-06-30 18:45:37 -0400199 default:
200 return ERR_PTR(-EAFNOSUPPORT);
201 }
202
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400203 return xcl->xcl_ops->xpo_create(serv, net, sap, len, flags);
Chuck Lever5dd248f2008-06-30 18:45:37 -0400204}
205
Chuck Lever9652ada2009-03-18 20:46:21 -0400206int svc_create_xprt(struct svc_serv *serv, const char *xprt_name,
Pavel Emelyanovfc5d00b2010-09-29 16:03:50 +0400207 struct net *net, const int family,
208 const unsigned short port, int flags)
Chuck Lever5dd248f2008-06-30 18:45:37 -0400209{
210 struct svc_xprt_class *xcl;
211
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600212 dprintk("svc: creating transport %s[%d]\n", xprt_name, port);
213 spin_lock(&svc_xprt_class_lock);
214 list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
Tom Tucker4e5caaa2007-12-30 21:08:20 -0600215 struct svc_xprt *newxprt;
216
217 if (strcmp(xprt_name, xcl->xcl_name))
218 continue;
219
220 if (!try_module_get(xcl->xcl_owner))
221 goto err;
222
223 spin_unlock(&svc_xprt_class_lock);
Pavel Emelyanov62832c02010-09-29 16:04:18 +0400224 newxprt = __svc_xpo_create(xcl, serv, net, family, port, flags);
Tom Tucker4e5caaa2007-12-30 21:08:20 -0600225 if (IS_ERR(newxprt)) {
226 module_put(xcl->xcl_owner);
227 return PTR_ERR(newxprt);
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600228 }
Tom Tucker4e5caaa2007-12-30 21:08:20 -0600229
230 clear_bit(XPT_TEMP, &newxprt->xpt_flags);
231 spin_lock_bh(&serv->sv_lock);
232 list_add(&newxprt->xpt_list, &serv->sv_permsocks);
233 spin_unlock_bh(&serv->sv_lock);
234 clear_bit(XPT_BUSY, &newxprt->xpt_flags);
235 return svc_xprt_local_port(newxprt);
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600236 }
Tom Tucker4e5caaa2007-12-30 21:08:20 -0600237 err:
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600238 spin_unlock(&svc_xprt_class_lock);
239 dprintk("svc: transport %s not found\n", xprt_name);
Chuck Lever68717902010-01-26 14:04:13 -0500240
241 /* This errno is exposed to user space. Provide a reasonable
242 * perror msg for a bad transport. */
243 return -EPROTONOSUPPORT;
Tom Tuckerb700cbb2007-12-30 21:07:42 -0600244}
245EXPORT_SYMBOL_GPL(svc_create_xprt);
Tom Tucker9dbc2402007-12-30 21:08:12 -0600246
247/*
248 * Copy the local and remote xprt addresses to the rqstp structure
249 */
250void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt)
251{
252 struct sockaddr *sin;
253
254 memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen);
255 rqstp->rq_addrlen = xprt->xpt_remotelen;
256
257 /*
258 * Destination address in request is needed for binding the
259 * source address in RPC replies/callbacks later.
260 */
261 sin = (struct sockaddr *)&xprt->xpt_local;
262 switch (sin->sa_family) {
263 case AF_INET:
264 rqstp->rq_daddr.addr = ((struct sockaddr_in *)sin)->sin_addr;
265 break;
266 case AF_INET6:
267 rqstp->rq_daddr.addr6 = ((struct sockaddr_in6 *)sin)->sin6_addr;
268 break;
269 }
270}
271EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs);
272
Tom Tucker0f0257e2007-12-30 21:08:27 -0600273/**
274 * svc_print_addr - Format rq_addr field for printing
275 * @rqstp: svc_rqst struct containing address to print
276 * @buf: target buffer for formatted address
277 * @len: length of target buffer
278 *
279 */
280char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
281{
282 return __svc_print_addr(svc_addr(rqstp), buf, len);
283}
284EXPORT_SYMBOL_GPL(svc_print_addr);
285
286/*
287 * Queue up an idle server thread. Must have pool->sp_lock held.
288 * Note: this is really a stack rather than a queue, so that we only
289 * use as many different threads as we need, and the rest don't pollute
290 * the cache.
291 */
292static void svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp)
293{
294 list_add(&rqstp->rq_list, &pool->sp_threads);
295}
296
297/*
298 * Dequeue an nfsd thread. Must have pool->sp_lock held.
299 */
300static void svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp)
301{
302 list_del(&rqstp->rq_list);
303}
304
305/*
306 * Queue up a transport with data pending. If there are idle nfsd
307 * processes, wake 'em up.
308 *
309 */
310void svc_xprt_enqueue(struct svc_xprt *xprt)
311{
312 struct svc_serv *serv = xprt->xpt_server;
313 struct svc_pool *pool;
314 struct svc_rqst *rqstp;
315 int cpu;
316
317 if (!(xprt->xpt_flags &
318 ((1<<XPT_CONN)|(1<<XPT_DATA)|(1<<XPT_CLOSE)|(1<<XPT_DEFERRED))))
319 return;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600320
321 cpu = get_cpu();
322 pool = svc_pool_for_cpu(xprt->xpt_server, cpu);
323 put_cpu();
324
325 spin_lock_bh(&pool->sp_lock);
326
J. Bruce Fields78c210e2009-08-06 15:41:34 -0400327 if (!list_empty(&pool->sp_threads) &&
328 !list_empty(&pool->sp_sockets))
329 printk(KERN_ERR
330 "svc_xprt_enqueue: "
331 "threads and transports both waiting??\n");
332
Tom Tucker0f0257e2007-12-30 21:08:27 -0600333 if (test_bit(XPT_DEAD, &xprt->xpt_flags)) {
334 /* Don't enqueue dead transports */
335 dprintk("svc: transport %p is dead, not enqueued\n", xprt);
336 goto out_unlock;
337 }
338
Greg Banks03cf6c92009-01-13 21:26:36 +1100339 pool->sp_stats.packets++;
340
Tom Tucker0f0257e2007-12-30 21:08:27 -0600341 /* Mark transport as busy. It will remain in this state until
342 * the provider calls svc_xprt_received. We update XPT_BUSY
343 * atomically because it also guards against trying to enqueue
344 * the transport twice.
345 */
346 if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) {
347 /* Don't enqueue transport while already enqueued */
348 dprintk("svc: transport %p busy, not enqueued\n", xprt);
349 goto out_unlock;
350 }
351 BUG_ON(xprt->xpt_pool != NULL);
352 xprt->xpt_pool = pool;
353
354 /* Handle pending connection */
355 if (test_bit(XPT_CONN, &xprt->xpt_flags))
356 goto process;
357
358 /* Handle close in-progress */
359 if (test_bit(XPT_CLOSE, &xprt->xpt_flags))
360 goto process;
361
362 /* Check if we have space to reply to a request */
363 if (!xprt->xpt_ops->xpo_has_wspace(xprt)) {
364 /* Don't enqueue while not enough space for reply */
365 dprintk("svc: no write space, transport %p not enqueued\n",
366 xprt);
367 xprt->xpt_pool = NULL;
368 clear_bit(XPT_BUSY, &xprt->xpt_flags);
369 goto out_unlock;
370 }
371
372 process:
J. Bruce Fields78c210e2009-08-06 15:41:34 -0400373 if (!list_empty(&pool->sp_threads)) {
Tom Tucker0f0257e2007-12-30 21:08:27 -0600374 rqstp = list_entry(pool->sp_threads.next,
375 struct svc_rqst,
376 rq_list);
377 dprintk("svc: transport %p served by daemon %p\n",
378 xprt, rqstp);
379 svc_thread_dequeue(pool, rqstp);
380 if (rqstp->rq_xprt)
381 printk(KERN_ERR
382 "svc_xprt_enqueue: server %p, rq_xprt=%p!\n",
383 rqstp, rqstp->rq_xprt);
384 rqstp->rq_xprt = xprt;
385 svc_xprt_get(xprt);
386 rqstp->rq_reserved = serv->sv_max_mesg;
387 atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
Greg Banks03cf6c92009-01-13 21:26:36 +1100388 pool->sp_stats.threads_woken++;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600389 BUG_ON(xprt->xpt_pool != pool);
390 wake_up(&rqstp->rq_wait);
391 } else {
392 dprintk("svc: transport %p put into queue\n", xprt);
393 list_add_tail(&xprt->xpt_ready, &pool->sp_sockets);
Greg Banks03cf6c92009-01-13 21:26:36 +1100394 pool->sp_stats.sockets_queued++;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600395 BUG_ON(xprt->xpt_pool != pool);
396 }
397
398out_unlock:
399 spin_unlock_bh(&pool->sp_lock);
400}
401EXPORT_SYMBOL_GPL(svc_xprt_enqueue);
402
403/*
404 * Dequeue the first transport. Must be called with the pool->sp_lock held.
405 */
406static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool)
407{
408 struct svc_xprt *xprt;
409
410 if (list_empty(&pool->sp_sockets))
411 return NULL;
412
413 xprt = list_entry(pool->sp_sockets.next,
414 struct svc_xprt, xpt_ready);
415 list_del_init(&xprt->xpt_ready);
416
417 dprintk("svc: transport %p dequeued, inuse=%d\n",
418 xprt, atomic_read(&xprt->xpt_ref.refcount));
419
420 return xprt;
421}
422
423/*
424 * svc_xprt_received conditionally queues the transport for processing
425 * by another thread. The caller must hold the XPT_BUSY bit and must
426 * not thereafter touch transport data.
427 *
428 * Note: XPT_DATA only gets cleared when a read-attempt finds no (or
429 * insufficient) data.
430 */
431void svc_xprt_received(struct svc_xprt *xprt)
432{
433 BUG_ON(!test_bit(XPT_BUSY, &xprt->xpt_flags));
434 xprt->xpt_pool = NULL;
435 clear_bit(XPT_BUSY, &xprt->xpt_flags);
436 svc_xprt_enqueue(xprt);
437}
438EXPORT_SYMBOL_GPL(svc_xprt_received);
439
440/**
441 * svc_reserve - change the space reserved for the reply to a request.
442 * @rqstp: The request in question
443 * @space: new max space to reserve
444 *
445 * Each request reserves some space on the output queue of the transport
446 * to make sure the reply fits. This function reduces that reserved
447 * space to be the amount of space used already, plus @space.
448 *
449 */
450void svc_reserve(struct svc_rqst *rqstp, int space)
451{
452 space += rqstp->rq_res.head[0].iov_len;
453
454 if (space < rqstp->rq_reserved) {
455 struct svc_xprt *xprt = rqstp->rq_xprt;
456 atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved);
457 rqstp->rq_reserved = space;
458
459 svc_xprt_enqueue(xprt);
460 }
461}
Trond Myklebust24c37672008-12-23 16:30:12 -0500462EXPORT_SYMBOL_GPL(svc_reserve);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600463
464static void svc_xprt_release(struct svc_rqst *rqstp)
465{
466 struct svc_xprt *xprt = rqstp->rq_xprt;
467
468 rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
469
Tom Tucker2779e3a2009-01-05 11:12:52 -0600470 kfree(rqstp->rq_deferred);
471 rqstp->rq_deferred = NULL;
472
Tom Tucker0f0257e2007-12-30 21:08:27 -0600473 svc_free_res_pages(rqstp);
474 rqstp->rq_res.page_len = 0;
475 rqstp->rq_res.page_base = 0;
476
477 /* Reset response buffer and release
478 * the reservation.
479 * But first, check that enough space was reserved
480 * for the reply, otherwise we have a bug!
481 */
482 if ((rqstp->rq_res.len) > rqstp->rq_reserved)
483 printk(KERN_ERR "RPC request reserved %d but used %d\n",
484 rqstp->rq_reserved,
485 rqstp->rq_res.len);
486
487 rqstp->rq_res.head[0].iov_len = 0;
488 svc_reserve(rqstp, 0);
489 rqstp->rq_xprt = NULL;
490
491 svc_xprt_put(xprt);
492}
493
494/*
495 * External function to wake up a server waiting for data
496 * This really only makes sense for services like lockd
497 * which have exactly one thread anyway.
498 */
499void svc_wake_up(struct svc_serv *serv)
500{
501 struct svc_rqst *rqstp;
502 unsigned int i;
503 struct svc_pool *pool;
504
505 for (i = 0; i < serv->sv_nrpools; i++) {
506 pool = &serv->sv_pools[i];
507
508 spin_lock_bh(&pool->sp_lock);
509 if (!list_empty(&pool->sp_threads)) {
510 rqstp = list_entry(pool->sp_threads.next,
511 struct svc_rqst,
512 rq_list);
513 dprintk("svc: daemon %p woken up.\n", rqstp);
514 /*
515 svc_thread_dequeue(pool, rqstp);
516 rqstp->rq_xprt = NULL;
517 */
518 wake_up(&rqstp->rq_wait);
519 }
520 spin_unlock_bh(&pool->sp_lock);
521 }
522}
Trond Myklebust24c37672008-12-23 16:30:12 -0500523EXPORT_SYMBOL_GPL(svc_wake_up);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600524
525int svc_port_is_privileged(struct sockaddr *sin)
526{
527 switch (sin->sa_family) {
528 case AF_INET:
529 return ntohs(((struct sockaddr_in *)sin)->sin_port)
530 < PROT_SOCK;
531 case AF_INET6:
532 return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
533 < PROT_SOCK;
534 default:
535 return 0;
536 }
537}
538
539/*
Jeff Laytonc9233eb2008-10-20 11:51:57 -0400540 * Make sure that we don't have too many active connections. If we have,
541 * something must be dropped. It's not clear what will happen if we allow
542 * "too many" connections, but when dealing with network-facing software,
543 * we have to code defensively. Here we do that by imposing hard limits.
Tom Tucker0f0257e2007-12-30 21:08:27 -0600544 *
545 * There's no point in trying to do random drop here for DoS
546 * prevention. The NFS clients does 1 reconnect in 15 seconds. An
547 * attacker can easily beat that.
548 *
549 * The only somewhat efficient mechanism would be if drop old
550 * connections from the same IP first. But right now we don't even
551 * record the client IP in svc_sock.
Jeff Laytonc9233eb2008-10-20 11:51:57 -0400552 *
553 * single-threaded services that expect a lot of clients will probably
554 * need to set sv_maxconn to override the default value which is based
555 * on the number of threads
Tom Tucker0f0257e2007-12-30 21:08:27 -0600556 */
557static void svc_check_conn_limits(struct svc_serv *serv)
558{
Jeff Laytonc9233eb2008-10-20 11:51:57 -0400559 unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn :
560 (serv->sv_nrthreads+3) * 20;
561
562 if (serv->sv_tmpcnt > limit) {
Tom Tucker0f0257e2007-12-30 21:08:27 -0600563 struct svc_xprt *xprt = NULL;
564 spin_lock_bh(&serv->sv_lock);
565 if (!list_empty(&serv->sv_tempsocks)) {
566 if (net_ratelimit()) {
567 /* Try to help the admin */
568 printk(KERN_NOTICE "%s: too many open "
Jeff Laytonc9233eb2008-10-20 11:51:57 -0400569 "connections, consider increasing %s\n",
570 serv->sv_name, serv->sv_maxconn ?
571 "the max number of connections." :
572 "the number of threads.");
Tom Tucker0f0257e2007-12-30 21:08:27 -0600573 }
574 /*
575 * Always select the oldest connection. It's not fair,
576 * but so is life
577 */
578 xprt = list_entry(serv->sv_tempsocks.prev,
579 struct svc_xprt,
580 xpt_list);
581 set_bit(XPT_CLOSE, &xprt->xpt_flags);
582 svc_xprt_get(xprt);
583 }
584 spin_unlock_bh(&serv->sv_lock);
585
586 if (xprt) {
587 svc_xprt_enqueue(xprt);
588 svc_xprt_put(xprt);
589 }
590 }
591}
592
593/*
594 * Receive the next request on any transport. This code is carefully
595 * organised not to touch any cachelines in the shared svc_serv
596 * structure, only cachelines in the local svc_pool.
597 */
598int svc_recv(struct svc_rqst *rqstp, long timeout)
599{
600 struct svc_xprt *xprt = NULL;
601 struct svc_serv *serv = rqstp->rq_server;
602 struct svc_pool *pool = rqstp->rq_pool;
603 int len, i;
604 int pages;
605 struct xdr_buf *arg;
606 DECLARE_WAITQUEUE(wait, current);
Greg Banks03cf6c92009-01-13 21:26:36 +1100607 long time_left;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600608
609 dprintk("svc: server %p waiting for data (to = %ld)\n",
610 rqstp, timeout);
611
612 if (rqstp->rq_xprt)
613 printk(KERN_ERR
614 "svc_recv: service %p, transport not NULL!\n",
615 rqstp);
616 if (waitqueue_active(&rqstp->rq_wait))
617 printk(KERN_ERR
618 "svc_recv: service %p, wait queue active!\n",
619 rqstp);
620
621 /* now allocate needed pages. If we get a failure, sleep briefly */
622 pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE;
623 for (i = 0; i < pages ; i++)
624 while (rqstp->rq_pages[i] == NULL) {
625 struct page *p = alloc_page(GFP_KERNEL);
626 if (!p) {
Jeff Layton7b54fe62008-02-12 11:47:24 -0500627 set_current_state(TASK_INTERRUPTIBLE);
628 if (signalled() || kthread_should_stop()) {
629 set_current_state(TASK_RUNNING);
Jeff Layton70867212008-02-07 16:34:54 -0500630 return -EINTR;
Jeff Layton7b54fe62008-02-12 11:47:24 -0500631 }
632 schedule_timeout(msecs_to_jiffies(500));
Tom Tucker0f0257e2007-12-30 21:08:27 -0600633 }
634 rqstp->rq_pages[i] = p;
635 }
636 rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */
637 BUG_ON(pages >= RPCSVC_MAXPAGES);
638
639 /* Make arg->head point to first page and arg->pages point to rest */
640 arg = &rqstp->rq_arg;
641 arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
642 arg->head[0].iov_len = PAGE_SIZE;
643 arg->pages = rqstp->rq_pages + 1;
644 arg->page_base = 0;
645 /* save at least one page for response */
646 arg->page_len = (pages-2)*PAGE_SIZE;
647 arg->len = (pages-1)*PAGE_SIZE;
648 arg->tail[0].iov_len = 0;
649
650 try_to_freeze();
651 cond_resched();
Jeff Layton70867212008-02-07 16:34:54 -0500652 if (signalled() || kthread_should_stop())
Tom Tucker0f0257e2007-12-30 21:08:27 -0600653 return -EINTR;
654
NeilBrownf16b6e82010-08-12 17:04:06 +1000655 /* Normally we will wait up to 5 seconds for any required
656 * cache information to be provided.
657 */
658 rqstp->rq_chandle.thread_wait = 5*HZ;
659
Tom Tucker0f0257e2007-12-30 21:08:27 -0600660 spin_lock_bh(&pool->sp_lock);
661 xprt = svc_xprt_dequeue(pool);
662 if (xprt) {
663 rqstp->rq_xprt = xprt;
664 svc_xprt_get(xprt);
665 rqstp->rq_reserved = serv->sv_max_mesg;
666 atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
NeilBrownf16b6e82010-08-12 17:04:06 +1000667
668 /* As there is a shortage of threads and this request
J. Bruce Fields6610f722010-08-26 13:19:52 -0400669 * had to be queued, don't allow the thread to wait so
NeilBrownf16b6e82010-08-12 17:04:06 +1000670 * long for cache updates.
671 */
672 rqstp->rq_chandle.thread_wait = 1*HZ;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600673 } else {
674 /* No data pending. Go to sleep */
675 svc_thread_enqueue(pool, rqstp);
676
677 /*
678 * We have to be able to interrupt this wait
679 * to bring down the daemons ...
680 */
681 set_current_state(TASK_INTERRUPTIBLE);
Jeff Layton70867212008-02-07 16:34:54 -0500682
683 /*
684 * checking kthread_should_stop() here allows us to avoid
685 * locking and signalling when stopping kthreads that call
686 * svc_recv. If the thread has already been woken up, then
687 * we can exit here without sleeping. If not, then it
688 * it'll be woken up quickly during the schedule_timeout
689 */
690 if (kthread_should_stop()) {
691 set_current_state(TASK_RUNNING);
692 spin_unlock_bh(&pool->sp_lock);
693 return -EINTR;
694 }
695
Tom Tucker0f0257e2007-12-30 21:08:27 -0600696 add_wait_queue(&rqstp->rq_wait, &wait);
697 spin_unlock_bh(&pool->sp_lock);
698
Greg Banks03cf6c92009-01-13 21:26:36 +1100699 time_left = schedule_timeout(timeout);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600700
701 try_to_freeze();
702
703 spin_lock_bh(&pool->sp_lock);
704 remove_wait_queue(&rqstp->rq_wait, &wait);
Greg Banks03cf6c92009-01-13 21:26:36 +1100705 if (!time_left)
706 pool->sp_stats.threads_timedout++;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600707
708 xprt = rqstp->rq_xprt;
709 if (!xprt) {
710 svc_thread_dequeue(pool, rqstp);
711 spin_unlock_bh(&pool->sp_lock);
712 dprintk("svc: server %p, no data yet\n", rqstp);
Jeff Layton70867212008-02-07 16:34:54 -0500713 if (signalled() || kthread_should_stop())
714 return -EINTR;
715 else
716 return -EAGAIN;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600717 }
718 }
719 spin_unlock_bh(&pool->sp_lock);
720
721 len = 0;
J. Bruce Fields1b644b62010-02-28 16:33:31 -0500722 if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) {
723 dprintk("svc_recv: found XPT_CLOSE\n");
724 svc_delete_xprt(xprt);
725 } else if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
Tom Tucker0f0257e2007-12-30 21:08:27 -0600726 struct svc_xprt *newxpt;
727 newxpt = xprt->xpt_ops->xpo_accept(xprt);
728 if (newxpt) {
729 /*
730 * We know this module_get will succeed because the
731 * listener holds a reference too
732 */
733 __module_get(newxpt->xpt_class->xcl_owner);
734 svc_check_conn_limits(xprt->xpt_server);
735 spin_lock_bh(&serv->sv_lock);
736 set_bit(XPT_TEMP, &newxpt->xpt_flags);
737 list_add(&newxpt->xpt_list, &serv->sv_tempsocks);
738 serv->sv_tmpcnt++;
739 if (serv->sv_temptimer.function == NULL) {
740 /* setup timer to age temp transports */
741 setup_timer(&serv->sv_temptimer,
742 svc_age_temp_xprts,
743 (unsigned long)serv);
744 mod_timer(&serv->sv_temptimer,
745 jiffies + svc_conn_age_period * HZ);
746 }
747 spin_unlock_bh(&serv->sv_lock);
748 svc_xprt_received(newxpt);
749 }
750 svc_xprt_received(xprt);
J. Bruce Fields1b644b62010-02-28 16:33:31 -0500751 } else {
Tom Tucker0f0257e2007-12-30 21:08:27 -0600752 dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n",
753 rqstp, pool->sp_id, xprt,
754 atomic_read(&xprt->xpt_ref.refcount));
755 rqstp->rq_deferred = svc_deferred_dequeue(xprt);
756 if (rqstp->rq_deferred) {
757 svc_xprt_received(xprt);
758 len = svc_deferred_recv(rqstp);
Neil Brownb48fa6b2010-03-01 16:51:14 +1100759 } else {
Tom Tucker0f0257e2007-12-30 21:08:27 -0600760 len = xprt->xpt_ops->xpo_recvfrom(rqstp);
Neil Brownb48fa6b2010-03-01 16:51:14 +1100761 svc_xprt_received(xprt);
762 }
Tom Tucker0f0257e2007-12-30 21:08:27 -0600763 dprintk("svc: got len=%d\n", len);
764 }
765
766 /* No data, incomplete (TCP) read, or accept() */
767 if (len == 0 || len == -EAGAIN) {
768 rqstp->rq_res.len = 0;
769 svc_xprt_release(rqstp);
770 return -EAGAIN;
771 }
772 clear_bit(XPT_OLD, &xprt->xpt_flags);
773
774 rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp));
775 rqstp->rq_chandle.defer = svc_defer;
776
777 if (serv->sv_stats)
778 serv->sv_stats->netcnt++;
779 return len;
780}
Trond Myklebust24c37672008-12-23 16:30:12 -0500781EXPORT_SYMBOL_GPL(svc_recv);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600782
783/*
784 * Drop request
785 */
786void svc_drop(struct svc_rqst *rqstp)
787{
788 dprintk("svc: xprt %p dropped request\n", rqstp->rq_xprt);
789 svc_xprt_release(rqstp);
790}
Trond Myklebust24c37672008-12-23 16:30:12 -0500791EXPORT_SYMBOL_GPL(svc_drop);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600792
793/*
794 * Return reply to client.
795 */
796int svc_send(struct svc_rqst *rqstp)
797{
798 struct svc_xprt *xprt;
799 int len;
800 struct xdr_buf *xb;
801
802 xprt = rqstp->rq_xprt;
803 if (!xprt)
804 return -EFAULT;
805
806 /* release the receive skb before sending the reply */
807 rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
808
809 /* calculate over-all length */
810 xb = &rqstp->rq_res;
811 xb->len = xb->head[0].iov_len +
812 xb->page_len +
813 xb->tail[0].iov_len;
814
815 /* Grab mutex to serialize outgoing data. */
816 mutex_lock(&xprt->xpt_mutex);
817 if (test_bit(XPT_DEAD, &xprt->xpt_flags))
818 len = -ENOTCONN;
819 else
820 len = xprt->xpt_ops->xpo_sendto(rqstp);
821 mutex_unlock(&xprt->xpt_mutex);
Rahul Iyer4cfc7e62009-09-10 17:32:28 +0300822 rpc_wake_up(&xprt->xpt_bc_pending);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600823 svc_xprt_release(rqstp);
824
825 if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
826 return 0;
827 return len;
828}
829
830/*
831 * Timer function to close old temporary transports, using
832 * a mark-and-sweep algorithm.
833 */
834static void svc_age_temp_xprts(unsigned long closure)
835{
836 struct svc_serv *serv = (struct svc_serv *)closure;
837 struct svc_xprt *xprt;
838 struct list_head *le, *next;
839 LIST_HEAD(to_be_aged);
840
841 dprintk("svc_age_temp_xprts\n");
842
843 if (!spin_trylock_bh(&serv->sv_lock)) {
844 /* busy, try again 1 sec later */
845 dprintk("svc_age_temp_xprts: busy\n");
846 mod_timer(&serv->sv_temptimer, jiffies + HZ);
847 return;
848 }
849
850 list_for_each_safe(le, next, &serv->sv_tempsocks) {
851 xprt = list_entry(le, struct svc_xprt, xpt_list);
852
853 /* First time through, just mark it OLD. Second time
854 * through, close it. */
855 if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags))
856 continue;
Joe Perchesf64f9e72009-11-29 16:55:45 -0800857 if (atomic_read(&xprt->xpt_ref.refcount) > 1 ||
858 test_bit(XPT_BUSY, &xprt->xpt_flags))
Tom Tucker0f0257e2007-12-30 21:08:27 -0600859 continue;
860 svc_xprt_get(xprt);
861 list_move(le, &to_be_aged);
862 set_bit(XPT_CLOSE, &xprt->xpt_flags);
863 set_bit(XPT_DETACHED, &xprt->xpt_flags);
864 }
865 spin_unlock_bh(&serv->sv_lock);
866
867 while (!list_empty(&to_be_aged)) {
868 le = to_be_aged.next;
869 /* fiddling the xpt_list node is safe 'cos we're XPT_DETACHED */
870 list_del_init(le);
871 xprt = list_entry(le, struct svc_xprt, xpt_list);
872
873 dprintk("queuing xprt %p for closing\n", xprt);
874
875 /* a thread will dequeue and close it soon */
876 svc_xprt_enqueue(xprt);
877 svc_xprt_put(xprt);
878 }
879
880 mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
881}
882
J. Bruce Fieldsedc7a892010-03-22 15:37:17 -0400883static void call_xpt_users(struct svc_xprt *xprt)
884{
885 struct svc_xpt_user *u;
886
887 spin_lock(&xprt->xpt_lock);
888 while (!list_empty(&xprt->xpt_users)) {
889 u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list);
890 list_del(&u->list);
891 u->callback(u);
892 }
893 spin_unlock(&xprt->xpt_lock);
894}
895
Tom Tucker0f0257e2007-12-30 21:08:27 -0600896/*
897 * Remove a dead transport
898 */
899void svc_delete_xprt(struct svc_xprt *xprt)
900{
901 struct svc_serv *serv = xprt->xpt_server;
Tom Tucker22945e42009-01-05 15:21:19 -0600902 struct svc_deferred_req *dr;
903
904 /* Only do this once */
905 if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags))
906 return;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600907
908 dprintk("svc: svc_delete_xprt(%p)\n", xprt);
909 xprt->xpt_ops->xpo_detach(xprt);
910
911 spin_lock_bh(&serv->sv_lock);
912 if (!test_and_set_bit(XPT_DETACHED, &xprt->xpt_flags))
913 list_del_init(&xprt->xpt_list);
914 /*
915 * We used to delete the transport from whichever list
916 * it's sk_xprt.xpt_ready node was on, but we don't actually
917 * need to. This is because the only time we're called
918 * while still attached to a queue, the queue itself
919 * is about to be destroyed (in svc_destroy).
920 */
Tom Tucker22945e42009-01-05 15:21:19 -0600921 if (test_bit(XPT_TEMP, &xprt->xpt_flags))
922 serv->sv_tmpcnt--;
J. Bruce Fields788e69e2010-03-29 21:02:31 -0400923 spin_unlock_bh(&serv->sv_lock);
Tom Tucker22945e42009-01-05 15:21:19 -0600924
Neil Brownab1b18f2010-02-27 09:33:40 +1100925 while ((dr = svc_deferred_dequeue(xprt)) != NULL)
Tom Tucker22945e42009-01-05 15:21:19 -0600926 kfree(dr);
Tom Tucker22945e42009-01-05 15:21:19 -0600927
J. Bruce Fieldsedc7a892010-03-22 15:37:17 -0400928 call_xpt_users(xprt);
Tom Tucker22945e42009-01-05 15:21:19 -0600929 svc_xprt_put(xprt);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600930}
931
932void svc_close_xprt(struct svc_xprt *xprt)
933{
934 set_bit(XPT_CLOSE, &xprt->xpt_flags);
935 if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
936 /* someone else will have to effect the close */
937 return;
938
939 svc_xprt_get(xprt);
940 svc_delete_xprt(xprt);
941 clear_bit(XPT_BUSY, &xprt->xpt_flags);
942 svc_xprt_put(xprt);
943}
Tom Tuckera2178132007-12-30 21:08:35 -0600944EXPORT_SYMBOL_GPL(svc_close_xprt);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600945
946void svc_close_all(struct list_head *xprt_list)
947{
948 struct svc_xprt *xprt;
949 struct svc_xprt *tmp;
950
951 list_for_each_entry_safe(xprt, tmp, xprt_list, xpt_list) {
952 set_bit(XPT_CLOSE, &xprt->xpt_flags);
953 if (test_bit(XPT_BUSY, &xprt->xpt_flags)) {
954 /* Waiting to be processed, but no threads left,
955 * So just remove it from the waiting list
956 */
957 list_del_init(&xprt->xpt_ready);
958 clear_bit(XPT_BUSY, &xprt->xpt_flags);
959 }
960 svc_close_xprt(xprt);
961 }
962}
963
964/*
965 * Handle defer and revisit of requests
966 */
967
968static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
969{
970 struct svc_deferred_req *dr =
971 container_of(dreq, struct svc_deferred_req, handle);
972 struct svc_xprt *xprt = dr->xprt;
973
Tom Tucker22945e42009-01-05 15:21:19 -0600974 spin_lock(&xprt->xpt_lock);
975 set_bit(XPT_DEFERRED, &xprt->xpt_flags);
976 if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) {
977 spin_unlock(&xprt->xpt_lock);
978 dprintk("revisit canceled\n");
Tom Tucker0f0257e2007-12-30 21:08:27 -0600979 svc_xprt_put(xprt);
980 kfree(dr);
981 return;
982 }
983 dprintk("revisit queued\n");
984 dr->xprt = NULL;
Tom Tucker0f0257e2007-12-30 21:08:27 -0600985 list_add(&dr->handle.recent, &xprt->xpt_deferred);
986 spin_unlock(&xprt->xpt_lock);
Tom Tucker0f0257e2007-12-30 21:08:27 -0600987 svc_xprt_enqueue(xprt);
988 svc_xprt_put(xprt);
989}
990
Tom Tucker260c1d12007-12-30 21:08:29 -0600991/*
992 * Save the request off for later processing. The request buffer looks
993 * like this:
994 *
995 * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
996 *
997 * This code can only handle requests that consist of an xprt-header
998 * and rpc-header.
999 */
Tom Tucker0f0257e2007-12-30 21:08:27 -06001000static struct cache_deferred_req *svc_defer(struct cache_req *req)
1001{
1002 struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
Tom Tucker0f0257e2007-12-30 21:08:27 -06001003 struct svc_deferred_req *dr;
1004
Andy Adamson2f425872009-04-03 08:27:32 +03001005 if (rqstp->rq_arg.page_len || !rqstp->rq_usedeferral)
Tom Tucker0f0257e2007-12-30 21:08:27 -06001006 return NULL; /* if more than a page, give up FIXME */
1007 if (rqstp->rq_deferred) {
1008 dr = rqstp->rq_deferred;
1009 rqstp->rq_deferred = NULL;
1010 } else {
Tom Tucker260c1d12007-12-30 21:08:29 -06001011 size_t skip;
1012 size_t size;
Tom Tucker0f0257e2007-12-30 21:08:27 -06001013 /* FIXME maybe discard if size too large */
Tom Tucker260c1d12007-12-30 21:08:29 -06001014 size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
Tom Tucker0f0257e2007-12-30 21:08:27 -06001015 dr = kmalloc(size, GFP_KERNEL);
1016 if (dr == NULL)
1017 return NULL;
1018
1019 dr->handle.owner = rqstp->rq_server;
1020 dr->prot = rqstp->rq_prot;
1021 memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
1022 dr->addrlen = rqstp->rq_addrlen;
1023 dr->daddr = rqstp->rq_daddr;
1024 dr->argslen = rqstp->rq_arg.len >> 2;
Tom Tucker260c1d12007-12-30 21:08:29 -06001025 dr->xprt_hlen = rqstp->rq_xprt_hlen;
1026
1027 /* back up head to the start of the buffer and copy */
1028 skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1029 memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
1030 dr->argslen << 2);
Tom Tucker0f0257e2007-12-30 21:08:27 -06001031 }
1032 svc_xprt_get(rqstp->rq_xprt);
1033 dr->xprt = rqstp->rq_xprt;
1034
1035 dr->handle.revisit = svc_revisit;
1036 return &dr->handle;
1037}
1038
1039/*
1040 * recv data from a deferred request into an active one
1041 */
1042static int svc_deferred_recv(struct svc_rqst *rqstp)
1043{
1044 struct svc_deferred_req *dr = rqstp->rq_deferred;
1045
Tom Tucker260c1d12007-12-30 21:08:29 -06001046 /* setup iov_base past transport header */
1047 rqstp->rq_arg.head[0].iov_base = dr->args + (dr->xprt_hlen>>2);
1048 /* The iov_len does not include the transport header bytes */
1049 rqstp->rq_arg.head[0].iov_len = (dr->argslen<<2) - dr->xprt_hlen;
Tom Tucker0f0257e2007-12-30 21:08:27 -06001050 rqstp->rq_arg.page_len = 0;
Tom Tucker260c1d12007-12-30 21:08:29 -06001051 /* The rq_arg.len includes the transport header bytes */
1052 rqstp->rq_arg.len = dr->argslen<<2;
Tom Tucker0f0257e2007-12-30 21:08:27 -06001053 rqstp->rq_prot = dr->prot;
1054 memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
1055 rqstp->rq_addrlen = dr->addrlen;
Tom Tucker260c1d12007-12-30 21:08:29 -06001056 /* Save off transport header len in case we get deferred again */
1057 rqstp->rq_xprt_hlen = dr->xprt_hlen;
Tom Tucker0f0257e2007-12-30 21:08:27 -06001058 rqstp->rq_daddr = dr->daddr;
1059 rqstp->rq_respages = rqstp->rq_pages;
Tom Tucker260c1d12007-12-30 21:08:29 -06001060 return (dr->argslen<<2) - dr->xprt_hlen;
Tom Tucker0f0257e2007-12-30 21:08:27 -06001061}
1062
1063
1064static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt)
1065{
1066 struct svc_deferred_req *dr = NULL;
1067
1068 if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags))
1069 return NULL;
1070 spin_lock(&xprt->xpt_lock);
1071 clear_bit(XPT_DEFERRED, &xprt->xpt_flags);
1072 if (!list_empty(&xprt->xpt_deferred)) {
1073 dr = list_entry(xprt->xpt_deferred.next,
1074 struct svc_deferred_req,
1075 handle.recent);
1076 list_del_init(&dr->handle.recent);
1077 set_bit(XPT_DEFERRED, &xprt->xpt_flags);
1078 }
1079 spin_unlock(&xprt->xpt_lock);
1080 return dr;
1081}
Tom Tucker7fcb98d2007-12-30 21:08:33 -06001082
Chuck Lever156e6202009-03-18 20:45:58 -04001083/**
1084 * svc_find_xprt - find an RPC transport instance
1085 * @serv: pointer to svc_serv to search
1086 * @xcl_name: C string containing transport's class name
1087 * @af: Address family of transport's local address
1088 * @port: transport's IP port number
1089 *
Tom Tucker7fcb98d2007-12-30 21:08:33 -06001090 * Return the transport instance pointer for the endpoint accepting
1091 * connections/peer traffic from the specified transport class,
1092 * address family and port.
1093 *
1094 * Specifying 0 for the address family or port is effectively a
1095 * wild-card, and will result in matching the first transport in the
1096 * service's list that has a matching class name.
1097 */
Chuck Lever156e6202009-03-18 20:45:58 -04001098struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name,
1099 const sa_family_t af, const unsigned short port)
Tom Tucker7fcb98d2007-12-30 21:08:33 -06001100{
1101 struct svc_xprt *xprt;
1102 struct svc_xprt *found = NULL;
1103
1104 /* Sanity check the args */
Chuck Lever156e6202009-03-18 20:45:58 -04001105 if (serv == NULL || xcl_name == NULL)
Tom Tucker7fcb98d2007-12-30 21:08:33 -06001106 return found;
1107
1108 spin_lock_bh(&serv->sv_lock);
1109 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1110 if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1111 continue;
1112 if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family)
1113 continue;
Chuck Lever156e6202009-03-18 20:45:58 -04001114 if (port != 0 && port != svc_xprt_local_port(xprt))
Tom Tucker7fcb98d2007-12-30 21:08:33 -06001115 continue;
1116 found = xprt;
Tom Tuckera2178132007-12-30 21:08:35 -06001117 svc_xprt_get(xprt);
Tom Tucker7fcb98d2007-12-30 21:08:33 -06001118 break;
1119 }
1120 spin_unlock_bh(&serv->sv_lock);
1121 return found;
1122}
1123EXPORT_SYMBOL_GPL(svc_find_xprt);
Tom Tucker9571af12007-12-30 21:08:37 -06001124
Chuck Lever335c54b2009-04-23 19:32:25 -04001125static int svc_one_xprt_name(const struct svc_xprt *xprt,
1126 char *pos, int remaining)
1127{
1128 int len;
1129
1130 len = snprintf(pos, remaining, "%s %u\n",
1131 xprt->xpt_class->xcl_name,
1132 svc_xprt_local_port(xprt));
1133 if (len >= remaining)
1134 return -ENAMETOOLONG;
1135 return len;
1136}
1137
1138/**
1139 * svc_xprt_names - format a buffer with a list of transport names
1140 * @serv: pointer to an RPC service
1141 * @buf: pointer to a buffer to be filled in
1142 * @buflen: length of buffer to be filled in
1143 *
1144 * Fills in @buf with a string containing a list of transport names,
1145 * each name terminated with '\n'.
1146 *
1147 * Returns positive length of the filled-in string on success; otherwise
1148 * a negative errno value is returned if an error occurs.
Tom Tucker9571af12007-12-30 21:08:37 -06001149 */
Chuck Lever335c54b2009-04-23 19:32:25 -04001150int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen)
Tom Tucker9571af12007-12-30 21:08:37 -06001151{
1152 struct svc_xprt *xprt;
Chuck Lever335c54b2009-04-23 19:32:25 -04001153 int len, totlen;
1154 char *pos;
Tom Tucker9571af12007-12-30 21:08:37 -06001155
1156 /* Sanity check args */
1157 if (!serv)
1158 return 0;
1159
1160 spin_lock_bh(&serv->sv_lock);
Chuck Lever335c54b2009-04-23 19:32:25 -04001161
1162 pos = buf;
1163 totlen = 0;
Tom Tucker9571af12007-12-30 21:08:37 -06001164 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
Chuck Lever335c54b2009-04-23 19:32:25 -04001165 len = svc_one_xprt_name(xprt, pos, buflen - totlen);
1166 if (len < 0) {
1167 *buf = '\0';
1168 totlen = len;
1169 }
1170 if (len <= 0)
Tom Tucker9571af12007-12-30 21:08:37 -06001171 break;
Chuck Lever335c54b2009-04-23 19:32:25 -04001172
1173 pos += len;
Tom Tucker9571af12007-12-30 21:08:37 -06001174 totlen += len;
1175 }
Chuck Lever335c54b2009-04-23 19:32:25 -04001176
Tom Tucker9571af12007-12-30 21:08:37 -06001177 spin_unlock_bh(&serv->sv_lock);
1178 return totlen;
1179}
1180EXPORT_SYMBOL_GPL(svc_xprt_names);
Greg Banks03cf6c92009-01-13 21:26:36 +11001181
1182
1183/*----------------------------------------------------------------------------*/
1184
1185static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos)
1186{
1187 unsigned int pidx = (unsigned int)*pos;
1188 struct svc_serv *serv = m->private;
1189
1190 dprintk("svc_pool_stats_start, *pidx=%u\n", pidx);
1191
Greg Banks03cf6c92009-01-13 21:26:36 +11001192 if (!pidx)
1193 return SEQ_START_TOKEN;
1194 return (pidx > serv->sv_nrpools ? NULL : &serv->sv_pools[pidx-1]);
1195}
1196
1197static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos)
1198{
1199 struct svc_pool *pool = p;
1200 struct svc_serv *serv = m->private;
1201
1202 dprintk("svc_pool_stats_next, *pos=%llu\n", *pos);
1203
1204 if (p == SEQ_START_TOKEN) {
1205 pool = &serv->sv_pools[0];
1206 } else {
1207 unsigned int pidx = (pool - &serv->sv_pools[0]);
1208 if (pidx < serv->sv_nrpools-1)
1209 pool = &serv->sv_pools[pidx+1];
1210 else
1211 pool = NULL;
1212 }
1213 ++*pos;
1214 return pool;
1215}
1216
1217static void svc_pool_stats_stop(struct seq_file *m, void *p)
1218{
Greg Banks03cf6c92009-01-13 21:26:36 +11001219}
1220
1221static int svc_pool_stats_show(struct seq_file *m, void *p)
1222{
1223 struct svc_pool *pool = p;
1224
1225 if (p == SEQ_START_TOKEN) {
J. Bruce Fields78c210e2009-08-06 15:41:34 -04001226 seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n");
Greg Banks03cf6c92009-01-13 21:26:36 +11001227 return 0;
1228 }
1229
J. Bruce Fields78c210e2009-08-06 15:41:34 -04001230 seq_printf(m, "%u %lu %lu %lu %lu\n",
Greg Banks03cf6c92009-01-13 21:26:36 +11001231 pool->sp_id,
1232 pool->sp_stats.packets,
1233 pool->sp_stats.sockets_queued,
1234 pool->sp_stats.threads_woken,
Greg Banks03cf6c92009-01-13 21:26:36 +11001235 pool->sp_stats.threads_timedout);
1236
1237 return 0;
1238}
1239
1240static const struct seq_operations svc_pool_stats_seq_ops = {
1241 .start = svc_pool_stats_start,
1242 .next = svc_pool_stats_next,
1243 .stop = svc_pool_stats_stop,
1244 .show = svc_pool_stats_show,
1245};
1246
1247int svc_pool_stats_open(struct svc_serv *serv, struct file *file)
1248{
1249 int err;
1250
1251 err = seq_open(file, &svc_pool_stats_seq_ops);
1252 if (!err)
1253 ((struct seq_file *) file->private_data)->private = serv;
1254 return err;
1255}
1256EXPORT_SYMBOL(svc_pool_stats_open);
1257
1258/*----------------------------------------------------------------------------*/