~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/sunrpc/svc_xprt.c

Version: ~ [ linux-5.9-rc6 ] ~ [ linux-5.8.10 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.66 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.146 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.198 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.236 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.236 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * linux/net/sunrpc/svc_xprt.c
  3  *
  4  * Author: Tom Tucker <tom@opengridcomputing.com>
  5  */
  6 
  7 #include <linux/sched.h>
  8 #include <linux/errno.h>
  9 #include <linux/freezer.h>
 10 #include <linux/kthread.h>
 11 #include <linux/slab.h>
 12 #include <net/sock.h>
 13 #include <linux/sunrpc/addr.h>
 14 #include <linux/sunrpc/stats.h>
 15 #include <linux/sunrpc/svc_xprt.h>
 16 #include <linux/sunrpc/svcsock.h>
 17 #include <linux/sunrpc/xprt.h>
 18 #include <linux/module.h>
 19 #include <linux/netdevice.h>
 20 #include <trace/events/sunrpc.h>
 21 
 22 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
 23 
 24 static unsigned int svc_rpc_per_connection_limit __read_mostly;
 25 module_param(svc_rpc_per_connection_limit, uint, 0644);
 26 
 27 
 28 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt);
 29 static int svc_deferred_recv(struct svc_rqst *rqstp);
 30 static struct cache_deferred_req *svc_defer(struct cache_req *req);
 31 static void svc_age_temp_xprts(struct timer_list *t);
 32 static void svc_delete_xprt(struct svc_xprt *xprt);
 33 
 34 /* apparently the "standard" is that clients close
 35  * idle connections after 5 minutes, servers after
 36  * 6 minutes
 37  *   http://www.connectathon.org/talks96/nfstcp.pdf
 38  */
 39 static int svc_conn_age_period = 6*60;
 40 
 41 /* List of registered transport classes */
 42 static DEFINE_SPINLOCK(svc_xprt_class_lock);
 43 static LIST_HEAD(svc_xprt_class_list);
 44 
 45 /* SMP locking strategy:
 46  *
 47  *      svc_pool->sp_lock protects most of the fields of that pool.
 48  *      svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
 49  *      when both need to be taken (rare), svc_serv->sv_lock is first.
 50  *      The "service mutex" protects svc_serv->sv_nrthread.
 51  *      svc_sock->sk_lock protects the svc_sock->sk_deferred list
 52  *             and the ->sk_info_authunix cache.
 53  *
 54  *      The XPT_BUSY bit in xprt->xpt_flags prevents a transport being
 55  *      enqueued multiply. During normal transport processing this bit
 56  *      is set by svc_xprt_enqueue and cleared by svc_xprt_received.
 57  *      Providers should not manipulate this bit directly.
 58  *
 59  *      Some flags can be set to certain values at any time
 60  *      providing that certain rules are followed:
 61  *
 62  *      XPT_CONN, XPT_DATA:
 63  *              - Can be set or cleared at any time.
 64  *              - After a set, svc_xprt_enqueue must be called to enqueue
 65  *                the transport for processing.
 66  *              - After a clear, the transport must be read/accepted.
 67  *                If this succeeds, it must be set again.
 68  *      XPT_CLOSE:
 69  *              - Can set at any time. It is never cleared.
 70  *      XPT_DEAD:
 71  *              - Can only be set while XPT_BUSY is held which ensures
 72  *                that no other thread will be using the transport or will
 73  *                try to set XPT_DEAD.
 74  */
 75 int svc_reg_xprt_class(struct svc_xprt_class *xcl)
 76 {
 77         struct svc_xprt_class *cl;
 78         int res = -EEXIST;
 79 
 80         dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name);
 81 
 82         INIT_LIST_HEAD(&xcl->xcl_list);
 83         spin_lock(&svc_xprt_class_lock);
 84         /* Make sure there isn't already a class with the same name */
 85         list_for_each_entry(cl, &svc_xprt_class_list, xcl_list) {
 86                 if (strcmp(xcl->xcl_name, cl->xcl_name) == 0)
 87                         goto out;
 88         }
 89         list_add_tail(&xcl->xcl_list, &svc_xprt_class_list);
 90         res = 0;
 91 out:
 92         spin_unlock(&svc_xprt_class_lock);
 93         return res;
 94 }
 95 EXPORT_SYMBOL_GPL(svc_reg_xprt_class);
 96 
 97 void svc_unreg_xprt_class(struct svc_xprt_class *xcl)
 98 {
 99         dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name);
100         spin_lock(&svc_xprt_class_lock);
101         list_del_init(&xcl->xcl_list);
102         spin_unlock(&svc_xprt_class_lock);
103 }
104 EXPORT_SYMBOL_GPL(svc_unreg_xprt_class);
105 
106 /*
107  * Format the transport list for printing
108  */
109 int svc_print_xprts(char *buf, int maxlen)
110 {
111         struct svc_xprt_class *xcl;
112         char tmpstr[80];
113         int len = 0;
114         buf[0] = '\0';
115 
116         spin_lock(&svc_xprt_class_lock);
117         list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
118                 int slen;
119 
120                 sprintf(tmpstr, "%s %d\n", xcl->xcl_name, xcl->xcl_max_payload);
121                 slen = strlen(tmpstr);
122                 if (len + slen > maxlen)
123                         break;
124                 len += slen;
125                 strcat(buf, tmpstr);
126         }
127         spin_unlock(&svc_xprt_class_lock);
128 
129         return len;
130 }
131 
132 static void svc_xprt_free(struct kref *kref)
133 {
134         struct svc_xprt *xprt =
135                 container_of(kref, struct svc_xprt, xpt_ref);
136         struct module *owner = xprt->xpt_class->xcl_owner;
137         if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags))
138                 svcauth_unix_info_release(xprt);
139         put_net(xprt->xpt_net);
140         /* See comment on corresponding get in xs_setup_bc_tcp(): */
141         if (xprt->xpt_bc_xprt)
142                 xprt_put(xprt->xpt_bc_xprt);
143         if (xprt->xpt_bc_xps)
144                 xprt_switch_put(xprt->xpt_bc_xps);
145         xprt->xpt_ops->xpo_free(xprt);
146         module_put(owner);
147 }
148 
149 void svc_xprt_put(struct svc_xprt *xprt)
150 {
151         kref_put(&xprt->xpt_ref, svc_xprt_free);
152 }
153 EXPORT_SYMBOL_GPL(svc_xprt_put);
154 
155 /*
156  * Called by transport drivers to initialize the transport independent
157  * portion of the transport instance.
158  */
159 void svc_xprt_init(struct net *net, struct svc_xprt_class *xcl,
160                    struct svc_xprt *xprt, struct svc_serv *serv)
161 {
162         memset(xprt, 0, sizeof(*xprt));
163         xprt->xpt_class = xcl;
164         xprt->xpt_ops = xcl->xcl_ops;
165         kref_init(&xprt->xpt_ref);
166         xprt->xpt_server = serv;
167         INIT_LIST_HEAD(&xprt->xpt_list);
168         INIT_LIST_HEAD(&xprt->xpt_ready);
169         INIT_LIST_HEAD(&xprt->xpt_deferred);
170         INIT_LIST_HEAD(&xprt->xpt_users);
171         mutex_init(&xprt->xpt_mutex);
172         spin_lock_init(&xprt->xpt_lock);
173         set_bit(XPT_BUSY, &xprt->xpt_flags);
174         rpc_init_wait_queue(&xprt->xpt_bc_pending, "xpt_bc_pending");
175         xprt->xpt_net = get_net(net);
176 }
177 EXPORT_SYMBOL_GPL(svc_xprt_init);
178 
179 static struct svc_xprt *__svc_xpo_create(struct svc_xprt_class *xcl,
180                                          struct svc_serv *serv,
181                                          struct net *net,
182                                          const int family,
183                                          const unsigned short port,
184                                          int flags)
185 {
186         struct sockaddr_in sin = {
187                 .sin_family             = AF_INET,
188                 .sin_addr.s_addr        = htonl(INADDR_ANY),
189                 .sin_port               = htons(port),
190         };
191 #if IS_ENABLED(CONFIG_IPV6)
192         struct sockaddr_in6 sin6 = {
193                 .sin6_family            = AF_INET6,
194                 .sin6_addr              = IN6ADDR_ANY_INIT,
195                 .sin6_port              = htons(port),
196         };
197 #endif
198         struct sockaddr *sap;
199         size_t len;
200 
201         switch (family) {
202         case PF_INET:
203                 sap = (struct sockaddr *)&sin;
204                 len = sizeof(sin);
205                 break;
206 #if IS_ENABLED(CONFIG_IPV6)
207         case PF_INET6:
208                 sap = (struct sockaddr *)&sin6;
209                 len = sizeof(sin6);
210                 break;
211 #endif
212         default:
213                 return ERR_PTR(-EAFNOSUPPORT);
214         }
215 
216         return xcl->xcl_ops->xpo_create(serv, net, sap, len, flags);
217 }
218 
219 /*
220  * svc_xprt_received conditionally queues the transport for processing
221  * by another thread. The caller must hold the XPT_BUSY bit and must
222  * not thereafter touch transport data.
223  *
224  * Note: XPT_DATA only gets cleared when a read-attempt finds no (or
225  * insufficient) data.
226  */
227 static void svc_xprt_received(struct svc_xprt *xprt)
228 {
229         if (!test_bit(XPT_BUSY, &xprt->xpt_flags)) {
230                 WARN_ONCE(1, "xprt=0x%p already busy!", xprt);
231                 return;
232         }
233 
234         /* As soon as we clear busy, the xprt could be closed and
235          * 'put', so we need a reference to call svc_enqueue_xprt with:
236          */
237         svc_xprt_get(xprt);
238         smp_mb__before_atomic();
239         clear_bit(XPT_BUSY, &xprt->xpt_flags);
240         xprt->xpt_server->sv_ops->svo_enqueue_xprt(xprt);
241         svc_xprt_put(xprt);
242 }
243 
244 void svc_add_new_perm_xprt(struct svc_serv *serv, struct svc_xprt *new)
245 {
246         clear_bit(XPT_TEMP, &new->xpt_flags);
247         spin_lock_bh(&serv->sv_lock);
248         list_add(&new->xpt_list, &serv->sv_permsocks);
249         spin_unlock_bh(&serv->sv_lock);
250         svc_xprt_received(new);
251 }
252 
253 static int _svc_create_xprt(struct svc_serv *serv, const char *xprt_name,
254                             struct net *net, const int family,
255                             const unsigned short port, int flags)
256 {
257         struct svc_xprt_class *xcl;
258 
259         spin_lock(&svc_xprt_class_lock);
260         list_for_each_entry(xcl, &svc_xprt_class_list, xcl_list) {
261                 struct svc_xprt *newxprt;
262                 unsigned short newport;
263 
264                 if (strcmp(xprt_name, xcl->xcl_name))
265                         continue;
266 
267                 if (!try_module_get(xcl->xcl_owner))
268                         goto err;
269 
270                 spin_unlock(&svc_xprt_class_lock);
271                 newxprt = __svc_xpo_create(xcl, serv, net, family, port, flags);
272                 if (IS_ERR(newxprt)) {
273                         module_put(xcl->xcl_owner);
274                         return PTR_ERR(newxprt);
275                 }
276                 svc_add_new_perm_xprt(serv, newxprt);
277                 newport = svc_xprt_local_port(newxprt);
278                 return newport;
279         }
280  err:
281         spin_unlock(&svc_xprt_class_lock);
282         /* This errno is exposed to user space.  Provide a reasonable
283          * perror msg for a bad transport. */
284         return -EPROTONOSUPPORT;
285 }
286 
287 int svc_create_xprt(struct svc_serv *serv, const char *xprt_name,
288                     struct net *net, const int family,
289                     const unsigned short port, int flags)
290 {
291         int err;
292 
293         dprintk("svc: creating transport %s[%d]\n", xprt_name, port);
294         err = _svc_create_xprt(serv, xprt_name, net, family, port, flags);
295         if (err == -EPROTONOSUPPORT) {
296                 request_module("svc%s", xprt_name);
297                 err = _svc_create_xprt(serv, xprt_name, net, family, port, flags);
298         }
299         if (err)
300                 dprintk("svc: transport %s not found, err %d\n",
301                         xprt_name, err);
302         return err;
303 }
304 EXPORT_SYMBOL_GPL(svc_create_xprt);
305 
306 /*
307  * Copy the local and remote xprt addresses to the rqstp structure
308  */
309 void svc_xprt_copy_addrs(struct svc_rqst *rqstp, struct svc_xprt *xprt)
310 {
311         memcpy(&rqstp->rq_addr, &xprt->xpt_remote, xprt->xpt_remotelen);
312         rqstp->rq_addrlen = xprt->xpt_remotelen;
313 
314         /*
315          * Destination address in request is needed for binding the
316          * source address in RPC replies/callbacks later.
317          */
318         memcpy(&rqstp->rq_daddr, &xprt->xpt_local, xprt->xpt_locallen);
319         rqstp->rq_daddrlen = xprt->xpt_locallen;
320 }
321 EXPORT_SYMBOL_GPL(svc_xprt_copy_addrs);
322 
323 /**
324  * svc_print_addr - Format rq_addr field for printing
325  * @rqstp: svc_rqst struct containing address to print
326  * @buf: target buffer for formatted address
327  * @len: length of target buffer
328  *
329  */
330 char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
331 {
332         return __svc_print_addr(svc_addr(rqstp), buf, len);
333 }
334 EXPORT_SYMBOL_GPL(svc_print_addr);
335 
336 static bool svc_xprt_slots_in_range(struct svc_xprt *xprt)
337 {
338         unsigned int limit = svc_rpc_per_connection_limit;
339         int nrqsts = atomic_read(&xprt->xpt_nr_rqsts);
340 
341         return limit == 0 || (nrqsts >= 0 && nrqsts < limit);
342 }
343 
344 static bool svc_xprt_reserve_slot(struct svc_rqst *rqstp, struct svc_xprt *xprt)
345 {
346         if (!test_bit(RQ_DATA, &rqstp->rq_flags)) {
347                 if (!svc_xprt_slots_in_range(xprt))
348                         return false;
349                 atomic_inc(&xprt->xpt_nr_rqsts);
350                 set_bit(RQ_DATA, &rqstp->rq_flags);
351         }
352         return true;
353 }
354 
355 static void svc_xprt_release_slot(struct svc_rqst *rqstp)
356 {
357         struct svc_xprt *xprt = rqstp->rq_xprt;
358         if (test_and_clear_bit(RQ_DATA, &rqstp->rq_flags)) {
359                 atomic_dec(&xprt->xpt_nr_rqsts);
360                 svc_xprt_enqueue(xprt);
361         }
362 }
363 
364 static bool svc_xprt_has_something_to_do(struct svc_xprt *xprt)
365 {
366         if (xprt->xpt_flags & ((1<<XPT_CONN)|(1<<XPT_CLOSE)))
367                 return true;
368         if (xprt->xpt_flags & ((1<<XPT_DATA)|(1<<XPT_DEFERRED))) {
369                 if (xprt->xpt_ops->xpo_has_wspace(xprt) &&
370                     svc_xprt_slots_in_range(xprt))
371                         return true;
372                 trace_svc_xprt_no_write_space(xprt);
373                 return false;
374         }
375         return false;
376 }
377 
378 void svc_xprt_do_enqueue(struct svc_xprt *xprt)
379 {
380         struct svc_pool *pool;
381         struct svc_rqst *rqstp = NULL;
382         int cpu;
383 
384         if (!svc_xprt_has_something_to_do(xprt))
385                 goto out;
386 
387         /* Mark transport as busy. It will remain in this state until
388          * the provider calls svc_xprt_received. We update XPT_BUSY
389          * atomically because it also guards against trying to enqueue
390          * the transport twice.
391          */
392         if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags)) {
393                 /* Don't enqueue transport while already enqueued */
394                 dprintk("svc: transport %p busy, not enqueued\n", xprt);
395                 goto out;
396         }
397 
398         cpu = get_cpu();
399         pool = svc_pool_for_cpu(xprt->xpt_server, cpu);
400 
401         atomic_long_inc(&pool->sp_stats.packets);
402 
403         dprintk("svc: transport %p put into queue\n", xprt);
404         spin_lock_bh(&pool->sp_lock);
405         list_add_tail(&xprt->xpt_ready, &pool->sp_sockets);
406         pool->sp_stats.sockets_queued++;
407         spin_unlock_bh(&pool->sp_lock);
408 
409         /* find a thread for this xprt */
410         rcu_read_lock();
411         list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) {
412                 if (test_and_set_bit(RQ_BUSY, &rqstp->rq_flags))
413                         continue;
414                 atomic_long_inc(&pool->sp_stats.threads_woken);
415                 wake_up_process(rqstp->rq_task);
416                 goto out_unlock;
417         }
418         set_bit(SP_CONGESTED, &pool->sp_flags);
419         rqstp = NULL;
420 out_unlock:
421         rcu_read_unlock();
422         put_cpu();
423 out:
424         trace_svc_xprt_do_enqueue(xprt, rqstp);
425 }
426 EXPORT_SYMBOL_GPL(svc_xprt_do_enqueue);
427 
428 /*
429  * Queue up a transport with data pending. If there are idle nfsd
430  * processes, wake 'em up.
431  *
432  */
433 void svc_xprt_enqueue(struct svc_xprt *xprt)
434 {
435         if (test_bit(XPT_BUSY, &xprt->xpt_flags))
436                 return;
437         xprt->xpt_server->sv_ops->svo_enqueue_xprt(xprt);
438 }
439 EXPORT_SYMBOL_GPL(svc_xprt_enqueue);
440 
441 /*
442  * Dequeue the first transport, if there is one.
443  */
444 static struct svc_xprt *svc_xprt_dequeue(struct svc_pool *pool)
445 {
446         struct svc_xprt *xprt = NULL;
447 
448         if (list_empty(&pool->sp_sockets))
449                 goto out;
450 
451         spin_lock_bh(&pool->sp_lock);
452         if (likely(!list_empty(&pool->sp_sockets))) {
453                 xprt = list_first_entry(&pool->sp_sockets,
454                                         struct svc_xprt, xpt_ready);
455                 list_del_init(&xprt->xpt_ready);
456                 svc_xprt_get(xprt);
457 
458                 dprintk("svc: transport %p dequeued, inuse=%d\n",
459                         xprt, kref_read(&xprt->xpt_ref));
460         }
461         spin_unlock_bh(&pool->sp_lock);
462 out:
463         trace_svc_xprt_dequeue(xprt);
464         return xprt;
465 }
466 
467 /**
468  * svc_reserve - change the space reserved for the reply to a request.
469  * @rqstp:  The request in question
470  * @space: new max space to reserve
471  *
472  * Each request reserves some space on the output queue of the transport
473  * to make sure the reply fits.  This function reduces that reserved
474  * space to be the amount of space used already, plus @space.
475  *
476  */
477 void svc_reserve(struct svc_rqst *rqstp, int space)
478 {
479         space += rqstp->rq_res.head[0].iov_len;
480 
481         if (space < rqstp->rq_reserved) {
482                 struct svc_xprt *xprt = rqstp->rq_xprt;
483                 atomic_sub((rqstp->rq_reserved - space), &xprt->xpt_reserved);
484                 rqstp->rq_reserved = space;
485 
486                 svc_xprt_enqueue(xprt);
487         }
488 }
489 EXPORT_SYMBOL_GPL(svc_reserve);
490 
491 static void svc_xprt_release(struct svc_rqst *rqstp)
492 {
493         struct svc_xprt *xprt = rqstp->rq_xprt;
494 
495         rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
496 
497         kfree(rqstp->rq_deferred);
498         rqstp->rq_deferred = NULL;
499 
500         svc_free_res_pages(rqstp);
501         rqstp->rq_res.page_len = 0;
502         rqstp->rq_res.page_base = 0;
503 
504         /* Reset response buffer and release
505          * the reservation.
506          * But first, check that enough space was reserved
507          * for the reply, otherwise we have a bug!
508          */
509         if ((rqstp->rq_res.len) >  rqstp->rq_reserved)
510                 printk(KERN_ERR "RPC request reserved %d but used %d\n",
511                        rqstp->rq_reserved,
512                        rqstp->rq_res.len);
513 
514         rqstp->rq_res.head[0].iov_len = 0;
515         svc_reserve(rqstp, 0);
516         svc_xprt_release_slot(rqstp);
517         rqstp->rq_xprt = NULL;
518         svc_xprt_put(xprt);
519 }
520 
521 /*
522  * Some svc_serv's will have occasional work to do, even when a xprt is not
523  * waiting to be serviced. This function is there to "kick" a task in one of
524  * those services so that it can wake up and do that work. Note that we only
525  * bother with pool 0 as we don't need to wake up more than one thread for
526  * this purpose.
527  */
528 void svc_wake_up(struct svc_serv *serv)
529 {
530         struct svc_rqst *rqstp;
531         struct svc_pool *pool;
532 
533         pool = &serv->sv_pools[0];
534 
535         rcu_read_lock();
536         list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) {
537                 /* skip any that aren't queued */
538                 if (test_bit(RQ_BUSY, &rqstp->rq_flags))
539                         continue;
540                 rcu_read_unlock();
541                 dprintk("svc: daemon %p woken up.\n", rqstp);
542                 wake_up_process(rqstp->rq_task);
543                 trace_svc_wake_up(rqstp->rq_task->pid);
544                 return;
545         }
546         rcu_read_unlock();
547 
548         /* No free entries available */
549         set_bit(SP_TASK_PENDING, &pool->sp_flags);
550         smp_wmb();
551         trace_svc_wake_up(0);
552 }
553 EXPORT_SYMBOL_GPL(svc_wake_up);
554 
555 int svc_port_is_privileged(struct sockaddr *sin)
556 {
557         switch (sin->sa_family) {
558         case AF_INET:
559                 return ntohs(((struct sockaddr_in *)sin)->sin_port)
560                         < PROT_SOCK;
561         case AF_INET6:
562                 return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
563                         < PROT_SOCK;
564         default:
565                 return 0;
566         }
567 }
568 
569 /*
570  * Make sure that we don't have too many active connections. If we have,
571  * something must be dropped. It's not clear what will happen if we allow
572  * "too many" connections, but when dealing with network-facing software,
573  * we have to code defensively. Here we do that by imposing hard limits.
574  *
575  * There's no point in trying to do random drop here for DoS
576  * prevention. The NFS clients does 1 reconnect in 15 seconds. An
577  * attacker can easily beat that.
578  *
579  * The only somewhat efficient mechanism would be if drop old
580  * connections from the same IP first. But right now we don't even
581  * record the client IP in svc_sock.
582  *
583  * single-threaded services that expect a lot of clients will probably
584  * need to set sv_maxconn to override the default value which is based
585  * on the number of threads
586  */
587 static void svc_check_conn_limits(struct svc_serv *serv)
588 {
589         unsigned int limit = serv->sv_maxconn ? serv->sv_maxconn :
590                                 (serv->sv_nrthreads+3) * 20;
591 
592         if (serv->sv_tmpcnt > limit) {
593                 struct svc_xprt *xprt = NULL;
594                 spin_lock_bh(&serv->sv_lock);
595                 if (!list_empty(&serv->sv_tempsocks)) {
596                         /* Try to help the admin */
597                         net_notice_ratelimited("%s: too many open connections, consider increasing the %s\n",
598                                                serv->sv_name, serv->sv_maxconn ?
599                                                "max number of connections" :
600                                                "number of threads");
601                         /*
602                          * Always select the oldest connection. It's not fair,
603                          * but so is life
604                          */
605                         xprt = list_entry(serv->sv_tempsocks.prev,
606                                           struct svc_xprt,
607                                           xpt_list);
608                         set_bit(XPT_CLOSE, &xprt->xpt_flags);
609                         svc_xprt_get(xprt);
610                 }
611                 spin_unlock_bh(&serv->sv_lock);
612 
613                 if (xprt) {
614                         svc_xprt_enqueue(xprt);
615                         svc_xprt_put(xprt);
616                 }
617         }
618 }
619 
620 static int svc_alloc_arg(struct svc_rqst *rqstp)
621 {
622         struct svc_serv *serv = rqstp->rq_server;
623         struct xdr_buf *arg;
624         int pages;
625         int i;
626 
627         /* now allocate needed pages.  If we get a failure, sleep briefly */
628         pages = (serv->sv_max_mesg + 2 * PAGE_SIZE) >> PAGE_SHIFT;
629         if (pages > RPCSVC_MAXPAGES) {
630                 pr_warn_once("svc: warning: pages=%u > RPCSVC_MAXPAGES=%lu\n",
631                              pages, RPCSVC_MAXPAGES);
632                 /* use as many pages as possible */
633                 pages = RPCSVC_MAXPAGES;
634         }
635         for (i = 0; i < pages ; i++)
636                 while (rqstp->rq_pages[i] == NULL) {
637                         struct page *p = alloc_page(GFP_KERNEL);
638                         if (!p) {
639                                 set_current_state(TASK_INTERRUPTIBLE);
640                                 if (signalled() || kthread_should_stop()) {
641                                         set_current_state(TASK_RUNNING);
642                                         return -EINTR;
643                                 }
644                                 schedule_timeout(msecs_to_jiffies(500));
645                         }
646                         rqstp->rq_pages[i] = p;
647                 }
648         rqstp->rq_page_end = &rqstp->rq_pages[i];
649         rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */
650 
651         /* Make arg->head point to first page and arg->pages point to rest */
652         arg = &rqstp->rq_arg;
653         arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
654         arg->head[0].iov_len = PAGE_SIZE;
655         arg->pages = rqstp->rq_pages + 1;
656         arg->page_base = 0;
657         /* save at least one page for response */
658         arg->page_len = (pages-2)*PAGE_SIZE;
659         arg->len = (pages-1)*PAGE_SIZE;
660         arg->tail[0].iov_len = 0;
661         return 0;
662 }
663 
664 static bool
665 rqst_should_sleep(struct svc_rqst *rqstp)
666 {
667         struct svc_pool         *pool = rqstp->rq_pool;
668 
669         /* did someone call svc_wake_up? */
670         if (test_and_clear_bit(SP_TASK_PENDING, &pool->sp_flags))
671                 return false;
672 
673         /* was a socket queued? */
674         if (!list_empty(&pool->sp_sockets))
675                 return false;
676 
677         /* are we shutting down? */
678         if (signalled() || kthread_should_stop())
679                 return false;
680 
681         /* are we freezing? */
682         if (freezing(current))
683                 return false;
684 
685         return true;
686 }
687 
688 static struct svc_xprt *svc_get_next_xprt(struct svc_rqst *rqstp, long timeout)
689 {
690         struct svc_pool         *pool = rqstp->rq_pool;
691         long                    time_left = 0;
692 
693         /* rq_xprt should be clear on entry */
694         WARN_ON_ONCE(rqstp->rq_xprt);
695 
696         rqstp->rq_xprt = svc_xprt_dequeue(pool);
697         if (rqstp->rq_xprt)
698                 goto out_found;
699 
700         /*
701          * We have to be able to interrupt this wait
702          * to bring down the daemons ...
703          */
704         set_current_state(TASK_INTERRUPTIBLE);
705         smp_mb__before_atomic();
706         clear_bit(SP_CONGESTED, &pool->sp_flags);
707         clear_bit(RQ_BUSY, &rqstp->rq_flags);
708         smp_mb__after_atomic();
709 
710         if (likely(rqst_should_sleep(rqstp)))
711                 time_left = schedule_timeout(timeout);
712         else
713                 __set_current_state(TASK_RUNNING);
714 
715         try_to_freeze();
716 
717         set_bit(RQ_BUSY, &rqstp->rq_flags);
718         smp_mb__after_atomic();
719         rqstp->rq_xprt = svc_xprt_dequeue(pool);
720         if (rqstp->rq_xprt)
721                 goto out_found;
722 
723         if (!time_left)
724                 atomic_long_inc(&pool->sp_stats.threads_timedout);
725 
726         if (signalled() || kthread_should_stop())
727                 return ERR_PTR(-EINTR);
728         return ERR_PTR(-EAGAIN);
729 out_found:
730         /* Normally we will wait up to 5 seconds for any required
731          * cache information to be provided.
732          */
733         if (!test_bit(SP_CONGESTED, &pool->sp_flags))
734                 rqstp->rq_chandle.thread_wait = 5*HZ;
735         else
736                 rqstp->rq_chandle.thread_wait = 1*HZ;
737         return rqstp->rq_xprt;
738 }
739 
740 static void svc_add_new_temp_xprt(struct svc_serv *serv, struct svc_xprt *newxpt)
741 {
742         spin_lock_bh(&serv->sv_lock);
743         set_bit(XPT_TEMP, &newxpt->xpt_flags);
744         list_add(&newxpt->xpt_list, &serv->sv_tempsocks);
745         serv->sv_tmpcnt++;
746         if (serv->sv_temptimer.function == NULL) {
747                 /* setup timer to age temp transports */
748                 serv->sv_temptimer.function = svc_age_temp_xprts;
749                 mod_timer(&serv->sv_temptimer,
750                           jiffies + svc_conn_age_period * HZ);
751         }
752         spin_unlock_bh(&serv->sv_lock);
753         svc_xprt_received(newxpt);
754 }
755 
756 static int svc_handle_xprt(struct svc_rqst *rqstp, struct svc_xprt *xprt)
757 {
758         struct svc_serv *serv = rqstp->rq_server;
759         int len = 0;
760 
761         if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) {
762                 dprintk("svc_recv: found XPT_CLOSE\n");
763                 if (test_and_clear_bit(XPT_KILL_TEMP, &xprt->xpt_flags))
764                         xprt->xpt_ops->xpo_kill_temp_xprt(xprt);
765                 svc_delete_xprt(xprt);
766                 /* Leave XPT_BUSY set on the dead xprt: */
767                 goto out;
768         }
769         if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
770                 struct svc_xprt *newxpt;
771                 /*
772                  * We know this module_get will succeed because the
773                  * listener holds a reference too
774                  */
775                 __module_get(xprt->xpt_class->xcl_owner);
776                 svc_check_conn_limits(xprt->xpt_server);
777                 newxpt = xprt->xpt_ops->xpo_accept(xprt);
778                 if (newxpt)
779                         svc_add_new_temp_xprt(serv, newxpt);
780                 else
781                         module_put(xprt->xpt_class->xcl_owner);
782         } else if (svc_xprt_reserve_slot(rqstp, xprt)) {
783                 /* XPT_DATA|XPT_DEFERRED case: */
784                 dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n",
785                         rqstp, rqstp->rq_pool->sp_id, xprt,
786                         kref_read(&xprt->xpt_ref));
787                 rqstp->rq_deferred = svc_deferred_dequeue(xprt);
788                 if (rqstp->rq_deferred)
789                         len = svc_deferred_recv(rqstp);
790                 else
791                         len = xprt->xpt_ops->xpo_recvfrom(rqstp);
792                 dprintk("svc: got len=%d\n", len);
793                 rqstp->rq_reserved = serv->sv_max_mesg;
794                 atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
795         }
796         /* clear XPT_BUSY: */
797         svc_xprt_received(xprt);
798 out:
799         trace_svc_handle_xprt(xprt, len);
800         return len;
801 }
802 
803 /*
804  * Receive the next request on any transport.  This code is carefully
805  * organised not to touch any cachelines in the shared svc_serv
806  * structure, only cachelines in the local svc_pool.
807  */
808 int svc_recv(struct svc_rqst *rqstp, long timeout)
809 {
810         struct svc_xprt         *xprt = NULL;
811         struct svc_serv         *serv = rqstp->rq_server;
812         int                     len, err;
813 
814         dprintk("svc: server %p waiting for data (to = %ld)\n",
815                 rqstp, timeout);
816 
817         if (rqstp->rq_xprt)
818                 printk(KERN_ERR
819                         "svc_recv: service %p, transport not NULL!\n",
820                          rqstp);
821 
822         err = svc_alloc_arg(rqstp);
823         if (err)
824                 goto out;
825 
826         try_to_freeze();
827         cond_resched();
828         err = -EINTR;
829         if (signalled() || kthread_should_stop())
830                 goto out;
831 
832         xprt = svc_get_next_xprt(rqstp, timeout);
833         if (IS_ERR(xprt)) {
834                 err = PTR_ERR(xprt);
835                 goto out;
836         }
837 
838         len = svc_handle_xprt(rqstp, xprt);
839 
840         /* No data, incomplete (TCP) read, or accept() */
841         err = -EAGAIN;
842         if (len <= 0)
843                 goto out_release;
844 
845         clear_bit(XPT_OLD, &xprt->xpt_flags);
846 
847         if (xprt->xpt_ops->xpo_secure_port(rqstp))
848                 set_bit(RQ_SECURE, &rqstp->rq_flags);
849         else
850                 clear_bit(RQ_SECURE, &rqstp->rq_flags);
851         rqstp->rq_chandle.defer = svc_defer;
852         rqstp->rq_xid = svc_getu32(&rqstp->rq_arg.head[0]);
853 
854         if (serv->sv_stats)
855                 serv->sv_stats->netcnt++;
856         trace_svc_recv(rqstp, len);
857         return len;
858 out_release:
859         rqstp->rq_res.len = 0;
860         svc_xprt_release(rqstp);
861 out:
862         trace_svc_recv(rqstp, err);
863         return err;
864 }
865 EXPORT_SYMBOL_GPL(svc_recv);
866 
867 /*
868  * Drop request
869  */
870 void svc_drop(struct svc_rqst *rqstp)
871 {
872         trace_svc_drop(rqstp);
873         dprintk("svc: xprt %p dropped request\n", rqstp->rq_xprt);
874         svc_xprt_release(rqstp);
875 }
876 EXPORT_SYMBOL_GPL(svc_drop);
877 
878 /*
879  * Return reply to client.
880  */
881 int svc_send(struct svc_rqst *rqstp)
882 {
883         struct svc_xprt *xprt;
884         int             len = -EFAULT;
885         struct xdr_buf  *xb;
886 
887         xprt = rqstp->rq_xprt;
888         if (!xprt)
889                 goto out;
890 
891         /* release the receive skb before sending the reply */
892         rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
893 
894         /* calculate over-all length */
895         xb = &rqstp->rq_res;
896         xb->len = xb->head[0].iov_len +
897                 xb->page_len +
898                 xb->tail[0].iov_len;
899 
900         /* Grab mutex to serialize outgoing data. */
901         mutex_lock(&xprt->xpt_mutex);
902         if (test_bit(XPT_DEAD, &xprt->xpt_flags)
903                         || test_bit(XPT_CLOSE, &xprt->xpt_flags))
904                 len = -ENOTCONN;
905         else
906                 len = xprt->xpt_ops->xpo_sendto(rqstp);
907         mutex_unlock(&xprt->xpt_mutex);
908         rpc_wake_up(&xprt->xpt_bc_pending);
909         svc_xprt_release(rqstp);
910 
911         if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
912                 len = 0;
913 out:
914         trace_svc_send(rqstp, len);
915         return len;
916 }
917 
918 /*
919  * Timer function to close old temporary transports, using
920  * a mark-and-sweep algorithm.
921  */
922 static void svc_age_temp_xprts(struct timer_list *t)
923 {
924         struct svc_serv *serv = from_timer(serv, t, sv_temptimer);
925         struct svc_xprt *xprt;
926         struct list_head *le, *next;
927 
928         dprintk("svc_age_temp_xprts\n");
929 
930         if (!spin_trylock_bh(&serv->sv_lock)) {
931                 /* busy, try again 1 sec later */
932                 dprintk("svc_age_temp_xprts: busy\n");
933                 mod_timer(&serv->sv_temptimer, jiffies + HZ);
934                 return;
935         }
936 
937         list_for_each_safe(le, next, &serv->sv_tempsocks) {
938                 xprt = list_entry(le, struct svc_xprt, xpt_list);
939 
940                 /* First time through, just mark it OLD. Second time
941                  * through, close it. */
942                 if (!test_and_set_bit(XPT_OLD, &xprt->xpt_flags))
943                         continue;
944                 if (kref_read(&xprt->xpt_ref) > 1 ||
945                     test_bit(XPT_BUSY, &xprt->xpt_flags))
946                         continue;
947                 list_del_init(le);
948                 set_bit(XPT_CLOSE, &xprt->xpt_flags);
949                 dprintk("queuing xprt %p for closing\n", xprt);
950 
951                 /* a thread will dequeue and close it soon */
952                 svc_xprt_enqueue(xprt);
953         }
954         spin_unlock_bh(&serv->sv_lock);
955 
956         mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
957 }
958 
959 /* Close temporary transports whose xpt_local matches server_addr immediately
960  * instead of waiting for them to be picked up by the timer.
961  *
962  * This is meant to be called from a notifier_block that runs when an ip
963  * address is deleted.
964  */
965 void svc_age_temp_xprts_now(struct svc_serv *serv, struct sockaddr *server_addr)
966 {
967         struct svc_xprt *xprt;
968         struct list_head *le, *next;
969         LIST_HEAD(to_be_closed);
970 
971         spin_lock_bh(&serv->sv_lock);
972         list_for_each_safe(le, next, &serv->sv_tempsocks) {
973                 xprt = list_entry(le, struct svc_xprt, xpt_list);
974                 if (rpc_cmp_addr(server_addr, (struct sockaddr *)
975                                 &xprt->xpt_local)) {
976                         dprintk("svc_age_temp_xprts_now: found %p\n", xprt);
977                         list_move(le, &to_be_closed);
978                 }
979         }
980         spin_unlock_bh(&serv->sv_lock);
981 
982         while (!list_empty(&to_be_closed)) {
983                 le = to_be_closed.next;
984                 list_del_init(le);
985                 xprt = list_entry(le, struct svc_xprt, xpt_list);
986                 set_bit(XPT_CLOSE, &xprt->xpt_flags);
987                 set_bit(XPT_KILL_TEMP, &xprt->xpt_flags);
988                 dprintk("svc_age_temp_xprts_now: queuing xprt %p for closing\n",
989                                 xprt);
990                 svc_xprt_enqueue(xprt);
991         }
992 }
993 EXPORT_SYMBOL_GPL(svc_age_temp_xprts_now);
994 
995 static void call_xpt_users(struct svc_xprt *xprt)
996 {
997         struct svc_xpt_user *u;
998 
999         spin_lock(&xprt->xpt_lock);
1000         while (!list_empty(&xprt->xpt_users)) {
1001                 u = list_first_entry(&xprt->xpt_users, struct svc_xpt_user, list);
1002                 list_del(&u->list);
1003                 u->callback(u);
1004         }
1005         spin_unlock(&xprt->xpt_lock);
1006 }
1007 
1008 /*
1009  * Remove a dead transport
1010  */
1011 static void svc_delete_xprt(struct svc_xprt *xprt)
1012 {
1013         struct svc_serv *serv = xprt->xpt_server;
1014         struct svc_deferred_req *dr;
1015 
1016         /* Only do this once */
1017         if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags))
1018                 BUG();
1019 
1020         dprintk("svc: svc_delete_xprt(%p)\n", xprt);
1021         xprt->xpt_ops->xpo_detach(xprt);
1022 
1023         spin_lock_bh(&serv->sv_lock);
1024         list_del_init(&xprt->xpt_list);
1025         WARN_ON_ONCE(!list_empty(&xprt->xpt_ready));
1026         if (test_bit(XPT_TEMP, &xprt->xpt_flags))
1027                 serv->sv_tmpcnt--;
1028         spin_unlock_bh(&serv->sv_lock);
1029 
1030         while ((dr = svc_deferred_dequeue(xprt)) != NULL)
1031                 kfree(dr);
1032 
1033         call_xpt_users(xprt);
1034         svc_xprt_put(xprt);
1035 }
1036 
1037 void svc_close_xprt(struct svc_xprt *xprt)
1038 {
1039         set_bit(XPT_CLOSE, &xprt->xpt_flags);
1040         if (test_and_set_bit(XPT_BUSY, &xprt->xpt_flags))
1041                 /* someone else will have to effect the close */
1042                 return;
1043         /*
1044          * We expect svc_close_xprt() to work even when no threads are
1045          * running (e.g., while configuring the server before starting
1046          * any threads), so if the transport isn't busy, we delete
1047          * it ourself:
1048          */
1049         svc_delete_xprt(xprt);
1050 }
1051 EXPORT_SYMBOL_GPL(svc_close_xprt);
1052 
1053 static int svc_close_list(struct svc_serv *serv, struct list_head *xprt_list, struct net *net)
1054 {
1055         struct svc_xprt *xprt;
1056         int ret = 0;
1057 
1058         spin_lock(&serv->sv_lock);
1059         list_for_each_entry(xprt, xprt_list, xpt_list) {
1060                 if (xprt->xpt_net != net)
1061                         continue;
1062                 ret++;
1063                 set_bit(XPT_CLOSE, &xprt->xpt_flags);
1064                 svc_xprt_enqueue(xprt);
1065         }
1066         spin_unlock(&serv->sv_lock);
1067         return ret;
1068 }
1069 
1070 static struct svc_xprt *svc_dequeue_net(struct svc_serv *serv, struct net *net)
1071 {
1072         struct svc_pool *pool;
1073         struct svc_xprt *xprt;
1074         struct svc_xprt *tmp;
1075         int i;
1076 
1077         for (i = 0; i < serv->sv_nrpools; i++) {
1078                 pool = &serv->sv_pools[i];
1079 
1080                 spin_lock_bh(&pool->sp_lock);
1081                 list_for_each_entry_safe(xprt, tmp, &pool->sp_sockets, xpt_ready) {
1082                         if (xprt->xpt_net != net)
1083                                 continue;
1084                         list_del_init(&xprt->xpt_ready);
1085                         spin_unlock_bh(&pool->sp_lock);
1086                         return xprt;
1087                 }
1088                 spin_unlock_bh(&pool->sp_lock);
1089         }
1090         return NULL;
1091 }
1092 
1093 static void svc_clean_up_xprts(struct svc_serv *serv, struct net *net)
1094 {
1095         struct svc_xprt *xprt;
1096 
1097         while ((xprt = svc_dequeue_net(serv, net))) {
1098                 set_bit(XPT_CLOSE, &xprt->xpt_flags);
1099                 svc_delete_xprt(xprt);
1100         }
1101 }
1102 
1103 /*
1104  * Server threads may still be running (especially in the case where the
1105  * service is still running in other network namespaces).
1106  *
1107  * So we shut down sockets the same way we would on a running server, by
1108  * setting XPT_CLOSE, enqueuing, and letting a thread pick it up to do
1109  * the close.  In the case there are no such other threads,
1110  * threads running, svc_clean_up_xprts() does a simple version of a
1111  * server's main event loop, and in the case where there are other
1112  * threads, we may need to wait a little while and then check again to
1113  * see if they're done.
1114  */
1115 void svc_close_net(struct svc_serv *serv, struct net *net)
1116 {
1117         int delay = 0;
1118 
1119         while (svc_close_list(serv, &serv->sv_permsocks, net) +
1120                svc_close_list(serv, &serv->sv_tempsocks, net)) {
1121 
1122                 svc_clean_up_xprts(serv, net);
1123                 msleep(delay++);
1124         }
1125 }
1126 
1127 /*
1128  * Handle defer and revisit of requests
1129  */
1130 
1131 static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1132 {
1133         struct svc_deferred_req *dr =
1134                 container_of(dreq, struct svc_deferred_req, handle);
1135         struct svc_xprt *xprt = dr->xprt;
1136 
1137         spin_lock(&xprt->xpt_lock);
1138         set_bit(XPT_DEFERRED, &xprt->xpt_flags);
1139         if (too_many || test_bit(XPT_DEAD, &xprt->xpt_flags)) {
1140                 spin_unlock(&xprt->xpt_lock);
1141                 dprintk("revisit canceled\n");
1142                 svc_xprt_put(xprt);
1143                 trace_svc_drop_deferred(dr);
1144                 kfree(dr);
1145                 return;
1146         }
1147         dprintk("revisit queued\n");
1148         dr->xprt = NULL;
1149         list_add(&dr->handle.recent, &xprt->xpt_deferred);
1150         spin_unlock(&xprt->xpt_lock);
1151         svc_xprt_enqueue(xprt);
1152         svc_xprt_put(xprt);
1153 }
1154 
1155 /*
1156  * Save the request off for later processing. The request buffer looks
1157  * like this:
1158  *
1159  * <xprt-header><rpc-header><rpc-pagelist><rpc-tail>
1160  *
1161  * This code can only handle requests that consist of an xprt-header
1162  * and rpc-header.
1163  */
1164 static struct cache_deferred_req *svc_defer(struct cache_req *req)
1165 {
1166         struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
1167         struct svc_deferred_req *dr;
1168 
1169         if (rqstp->rq_arg.page_len || !test_bit(RQ_USEDEFERRAL, &rqstp->rq_flags))
1170                 return NULL; /* if more than a page, give up FIXME */
1171         if (rqstp->rq_deferred) {
1172                 dr = rqstp->rq_deferred;
1173                 rqstp->rq_deferred = NULL;
1174         } else {
1175                 size_t skip;
1176                 size_t size;
1177                 /* FIXME maybe discard if size too large */
1178                 size = sizeof(struct svc_deferred_req) + rqstp->rq_arg.len;
1179                 dr = kmalloc(size, GFP_KERNEL);
1180                 if (dr == NULL)
1181                         return NULL;
1182 
1183                 dr->handle.owner = rqstp->rq_server;
1184                 dr->prot = rqstp->rq_prot;
1185                 memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
1186                 dr->addrlen = rqstp->rq_addrlen;
1187                 dr->daddr = rqstp->rq_daddr;
1188                 dr->argslen = rqstp->rq_arg.len >> 2;
1189                 dr->xprt_hlen = rqstp->rq_xprt_hlen;
1190 
1191                 /* back up head to the start of the buffer and copy */
1192                 skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1193                 memcpy(dr->args, rqstp->rq_arg.head[0].iov_base - skip,
1194                        dr->argslen << 2);
1195         }
1196         svc_xprt_get(rqstp->rq_xprt);
1197         dr->xprt = rqstp->rq_xprt;
1198         set_bit(RQ_DROPME, &rqstp->rq_flags);
1199 
1200         dr->handle.revisit = svc_revisit;
1201         trace_svc_defer(rqstp);
1202         return &dr->handle;
1203 }
1204 
1205 /*
1206  * recv data from a deferred request into an active one
1207  */
1208 static int svc_deferred_recv(struct svc_rqst *rqstp)
1209 {
1210         struct svc_deferred_req *dr = rqstp->rq_deferred;
1211 
1212         /* setup iov_base past transport header */
1213         rqstp->rq_arg.head[0].iov_base = dr->args + (dr->xprt_hlen>>2);
1214         /* The iov_len does not include the transport header bytes */
1215         rqstp->rq_arg.head[0].iov_len = (dr->argslen<<2) - dr->xprt_hlen;
1216         rqstp->rq_arg.page_len = 0;
1217         /* The rq_arg.len includes the transport header bytes */
1218         rqstp->rq_arg.len     = dr->argslen<<2;
1219         rqstp->rq_prot        = dr->prot;
1220         memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
1221         rqstp->rq_addrlen     = dr->addrlen;
1222         /* Save off transport header len in case we get deferred again */
1223         rqstp->rq_xprt_hlen   = dr->xprt_hlen;
1224         rqstp->rq_daddr       = dr->daddr;
1225         rqstp->rq_respages    = rqstp->rq_pages;
1226         return (dr->argslen<<2) - dr->xprt_hlen;
1227 }
1228 
1229 
1230 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_xprt *xprt)
1231 {
1232         struct svc_deferred_req *dr = NULL;
1233 
1234         if (!test_bit(XPT_DEFERRED, &xprt->xpt_flags))
1235                 return NULL;
1236         spin_lock(&xprt->xpt_lock);
1237         if (!list_empty(&xprt->xpt_deferred)) {
1238                 dr = list_entry(xprt->xpt_deferred.next,
1239                                 struct svc_deferred_req,
1240                                 handle.recent);
1241                 list_del_init(&dr->handle.recent);
1242                 trace_svc_revisit_deferred(dr);
1243         } else
1244                 clear_bit(XPT_DEFERRED, &xprt->xpt_flags);
1245         spin_unlock(&xprt->xpt_lock);
1246         return dr;
1247 }
1248 
1249 /**
1250  * svc_find_xprt - find an RPC transport instance
1251  * @serv: pointer to svc_serv to search
1252  * @xcl_name: C string containing transport's class name
1253  * @net: owner net pointer
1254  * @af: Address family of transport's local address
1255  * @port: transport's IP port number
1256  *
1257  * Return the transport instance pointer for the endpoint accepting
1258  * connections/peer traffic from the specified transport class,
1259  * address family and port.
1260  *
1261  * Specifying 0 for the address family or port is effectively a
1262  * wild-card, and will result in matching the first transport in the
1263  * service's list that has a matching class name.
1264  */
1265 struct svc_xprt *svc_find_xprt(struct svc_serv *serv, const char *xcl_name,
1266                                struct net *net, const sa_family_t af,
1267                                const unsigned short port)
1268 {
1269         struct svc_xprt *xprt;
1270         struct svc_xprt *found = NULL;
1271 
1272         /* Sanity check the args */
1273         if (serv == NULL || xcl_name == NULL)
1274                 return found;
1275 
1276         spin_lock_bh(&serv->sv_lock);
1277         list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1278                 if (xprt->xpt_net != net)
1279                         continue;
1280                 if (strcmp(xprt->xpt_class->xcl_name, xcl_name))
1281                         continue;
1282                 if (af != AF_UNSPEC && af != xprt->xpt_local.ss_family)
1283                         continue;
1284                 if (port != 0 && port != svc_xprt_local_port(xprt))
1285                         continue;
1286                 found = xprt;
1287                 svc_xprt_get(xprt);
1288                 break;
1289         }
1290         spin_unlock_bh(&serv->sv_lock);
1291         return found;
1292 }
1293 EXPORT_SYMBOL_GPL(svc_find_xprt);
1294 
1295 static int svc_one_xprt_name(const struct svc_xprt *xprt,
1296                              char *pos, int remaining)
1297 {
1298         int len;
1299 
1300         len = snprintf(pos, remaining, "%s %u\n",
1301                         xprt->xpt_class->xcl_name,
1302                         svc_xprt_local_port(xprt));
1303         if (len >= remaining)
1304                 return -ENAMETOOLONG;
1305         return len;
1306 }
1307 
1308 /**
1309  * svc_xprt_names - format a buffer with a list of transport names
1310  * @serv: pointer to an RPC service
1311  * @buf: pointer to a buffer to be filled in
1312  * @buflen: length of buffer to be filled in
1313  *
1314  * Fills in @buf with a string containing a list of transport names,
1315  * each name terminated with '\n'.
1316  *
1317  * Returns positive length of the filled-in string on success; otherwise
1318  * a negative errno value is returned if an error occurs.
1319  */
1320 int svc_xprt_names(struct svc_serv *serv, char *buf, const int buflen)
1321 {
1322         struct svc_xprt *xprt;
1323         int len, totlen;
1324         char *pos;
1325 
1326         /* Sanity check args */
1327         if (!serv)
1328                 return 0;
1329 
1330         spin_lock_bh(&serv->sv_lock);
1331 
1332         pos = buf;
1333         totlen = 0;
1334         list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
1335                 len = svc_one_xprt_name(xprt, pos, buflen - totlen);
1336                 if (len < 0) {
1337                         *buf = '\0';
1338                         totlen = len;
1339                 }
1340                 if (len <= 0)
1341                         break;
1342 
1343                 pos += len;
1344                 totlen += len;
1345         }
1346 
1347         spin_unlock_bh(&serv->sv_lock);
1348         return totlen;
1349 }
1350 EXPORT_SYMBOL_GPL(svc_xprt_names);
1351 
1352 
1353 /*----------------------------------------------------------------------------*/
1354 
1355 static void *svc_pool_stats_start(struct seq_file *m, loff_t *pos)
1356 {
1357         unsigned int pidx = (unsigned int)*pos;
1358         struct svc_serv *serv = m->private;
1359 
1360         dprintk("svc_pool_stats_start, *pidx=%u\n", pidx);
1361 
1362         if (!pidx)
1363                 return SEQ_START_TOKEN;
1364         return (pidx > serv->sv_nrpools ? NULL : &serv->sv_pools[pidx-1]);
1365 }
1366 
1367 static void *svc_pool_stats_next(struct seq_file *m, void *p, loff_t *pos)
1368 {
1369         struct svc_pool *pool = p;
1370         struct svc_serv *serv = m->private;
1371 
1372         dprintk("svc_pool_stats_next, *pos=%llu\n", *pos);
1373 
1374         if (p == SEQ_START_TOKEN) {
1375                 pool = &serv->sv_pools[0];
1376         } else {
1377                 unsigned int pidx = (pool - &serv->sv_pools[0]);
1378                 if (pidx < serv->sv_nrpools-1)
1379                         pool = &serv->sv_pools[pidx+1];
1380                 else
1381                         pool = NULL;
1382         }
1383         ++*pos;
1384         return pool;
1385 }
1386 
1387 static void svc_pool_stats_stop(struct seq_file *m, void *p)
1388 {
1389 }
1390 
1391 static int svc_pool_stats_show(struct seq_file *m, void *p)
1392 {
1393         struct svc_pool *pool = p;
1394 
1395         if (p == SEQ_START_TOKEN) {
1396                 seq_puts(m, "# pool packets-arrived sockets-enqueued threads-woken threads-timedout\n");
1397                 return 0;
1398         }
1399 
1400         seq_printf(m, "%u %lu %lu %lu %lu\n",
1401                 pool->sp_id,
1402                 (unsigned long)atomic_long_read(&pool->sp_stats.packets),
1403                 pool->sp_stats.sockets_queued,
1404                 (unsigned long)atomic_long_read(&pool->sp_stats.threads_woken),
1405                 (unsigned long)atomic_long_read(&pool->sp_stats.threads_timedout));
1406 
1407         return 0;
1408 }
1409 
1410 static const struct seq_operations svc_pool_stats_seq_ops = {
1411         .start  = svc_pool_stats_start,
1412         .next   = svc_pool_stats_next,
1413         .stop   = svc_pool_stats_stop,
1414         .show   = svc_pool_stats_show,
1415 };
1416 
1417 int svc_pool_stats_open(struct svc_serv *serv, struct file *file)
1418 {
1419         int err;
1420 
1421         err = seq_open(file, &svc_pool_stats_seq_ops);
1422         if (!err)
1423                 ((struct seq_file *) file->private_data)->private = serv;
1424         return err;
1425 }
1426 EXPORT_SYMBOL(svc_pool_stats_open);
1427 
1428 /*----------------------------------------------------------------------------*/
1429 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp