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

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

Version: ~ [ linux-5.10-rc5 ] ~ [ linux-5.9.10 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.79 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.159 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.208 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.245 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.245 ] ~ [ 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.c
  3  *
  4  * High-level RPC service routines
  5  *
  6  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
  7  *
  8  * Multiple threads pools and NUMAisation
  9  * Copyright (c) 2006 Silicon Graphics, Inc.
 10  * by Greg Banks <gnb@melbourne.sgi.com>
 11  */
 12 
 13 #include <linux/linkage.h>
 14 #include <linux/sched.h>
 15 #include <linux/errno.h>
 16 #include <linux/net.h>
 17 #include <linux/in.h>
 18 #include <linux/mm.h>
 19 #include <linux/interrupt.h>
 20 #include <linux/module.h>
 21 #include <linux/kthread.h>
 22 
 23 #include <linux/sunrpc/types.h>
 24 #include <linux/sunrpc/xdr.h>
 25 #include <linux/sunrpc/stats.h>
 26 #include <linux/sunrpc/svcsock.h>
 27 #include <linux/sunrpc/clnt.h>
 28 #include <linux/sunrpc/bc_xprt.h>
 29 
 30 #define RPCDBG_FACILITY RPCDBG_SVCDSP
 31 
 32 static void svc_unregister(const struct svc_serv *serv);
 33 
 34 #define svc_serv_is_pooled(serv)    ((serv)->sv_function)
 35 
 36 /*
 37  * Mode for mapping cpus to pools.
 38  */
 39 enum {
 40         SVC_POOL_AUTO = -1,     /* choose one of the others */
 41         SVC_POOL_GLOBAL,        /* no mapping, just a single global pool
 42                                  * (legacy & UP mode) */
 43         SVC_POOL_PERCPU,        /* one pool per cpu */
 44         SVC_POOL_PERNODE        /* one pool per numa node */
 45 };
 46 #define SVC_POOL_DEFAULT        SVC_POOL_GLOBAL
 47 
 48 /*
 49  * Structure for mapping cpus to pools and vice versa.
 50  * Setup once during sunrpc initialisation.
 51  */
 52 static struct svc_pool_map {
 53         int count;                      /* How many svc_servs use us */
 54         int mode;                       /* Note: int not enum to avoid
 55                                          * warnings about "enumeration value
 56                                          * not handled in switch" */
 57         unsigned int npools;
 58         unsigned int *pool_to;          /* maps pool id to cpu or node */
 59         unsigned int *to_pool;          /* maps cpu or node to pool id */
 60 } svc_pool_map = {
 61         .count = 0,
 62         .mode = SVC_POOL_DEFAULT
 63 };
 64 static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */
 65 
 66 static int
 67 param_set_pool_mode(const char *val, struct kernel_param *kp)
 68 {
 69         int *ip = (int *)kp->arg;
 70         struct svc_pool_map *m = &svc_pool_map;
 71         int err;
 72 
 73         mutex_lock(&svc_pool_map_mutex);
 74 
 75         err = -EBUSY;
 76         if (m->count)
 77                 goto out;
 78 
 79         err = 0;
 80         if (!strncmp(val, "auto", 4))
 81                 *ip = SVC_POOL_AUTO;
 82         else if (!strncmp(val, "global", 6))
 83                 *ip = SVC_POOL_GLOBAL;
 84         else if (!strncmp(val, "percpu", 6))
 85                 *ip = SVC_POOL_PERCPU;
 86         else if (!strncmp(val, "pernode", 7))
 87                 *ip = SVC_POOL_PERNODE;
 88         else
 89                 err = -EINVAL;
 90 
 91 out:
 92         mutex_unlock(&svc_pool_map_mutex);
 93         return err;
 94 }
 95 
 96 static int
 97 param_get_pool_mode(char *buf, struct kernel_param *kp)
 98 {
 99         int *ip = (int *)kp->arg;
100 
101         switch (*ip)
102         {
103         case SVC_POOL_AUTO:
104                 return strlcpy(buf, "auto", 20);
105         case SVC_POOL_GLOBAL:
106                 return strlcpy(buf, "global", 20);
107         case SVC_POOL_PERCPU:
108                 return strlcpy(buf, "percpu", 20);
109         case SVC_POOL_PERNODE:
110                 return strlcpy(buf, "pernode", 20);
111         default:
112                 return sprintf(buf, "%d", *ip);
113         }
114 }
115 
116 module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode,
117                  &svc_pool_map.mode, 0644);
118 
119 /*
120  * Detect best pool mapping mode heuristically,
121  * according to the machine's topology.
122  */
123 static int
124 svc_pool_map_choose_mode(void)
125 {
126         unsigned int node;
127 
128         if (nr_online_nodes > 1) {
129                 /*
130                  * Actually have multiple NUMA nodes,
131                  * so split pools on NUMA node boundaries
132                  */
133                 return SVC_POOL_PERNODE;
134         }
135 
136         node = any_online_node(node_online_map);
137         if (nr_cpus_node(node) > 2) {
138                 /*
139                  * Non-trivial SMP, or CONFIG_NUMA on
140                  * non-NUMA hardware, e.g. with a generic
141                  * x86_64 kernel on Xeons.  In this case we
142                  * want to divide the pools on cpu boundaries.
143                  */
144                 return SVC_POOL_PERCPU;
145         }
146 
147         /* default: one global pool */
148         return SVC_POOL_GLOBAL;
149 }
150 
151 /*
152  * Allocate the to_pool[] and pool_to[] arrays.
153  * Returns 0 on success or an errno.
154  */
155 static int
156 svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools)
157 {
158         m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
159         if (!m->to_pool)
160                 goto fail;
161         m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL);
162         if (!m->pool_to)
163                 goto fail_free;
164 
165         return 0;
166 
167 fail_free:
168         kfree(m->to_pool);
169         m->to_pool = NULL;
170 fail:
171         return -ENOMEM;
172 }
173 
174 /*
175  * Initialise the pool map for SVC_POOL_PERCPU mode.
176  * Returns number of pools or <0 on error.
177  */
178 static int
179 svc_pool_map_init_percpu(struct svc_pool_map *m)
180 {
181         unsigned int maxpools = nr_cpu_ids;
182         unsigned int pidx = 0;
183         unsigned int cpu;
184         int err;
185 
186         err = svc_pool_map_alloc_arrays(m, maxpools);
187         if (err)
188                 return err;
189 
190         for_each_online_cpu(cpu) {
191                 BUG_ON(pidx > maxpools);
192                 m->to_pool[cpu] = pidx;
193                 m->pool_to[pidx] = cpu;
194                 pidx++;
195         }
196         /* cpus brought online later all get mapped to pool0, sorry */
197 
198         return pidx;
199 };
200 
201 
202 /*
203  * Initialise the pool map for SVC_POOL_PERNODE mode.
204  * Returns number of pools or <0 on error.
205  */
206 static int
207 svc_pool_map_init_pernode(struct svc_pool_map *m)
208 {
209         unsigned int maxpools = nr_node_ids;
210         unsigned int pidx = 0;
211         unsigned int node;
212         int err;
213 
214         err = svc_pool_map_alloc_arrays(m, maxpools);
215         if (err)
216                 return err;
217 
218         for_each_node_with_cpus(node) {
219                 /* some architectures (e.g. SN2) have cpuless nodes */
220                 BUG_ON(pidx > maxpools);
221                 m->to_pool[node] = pidx;
222                 m->pool_to[pidx] = node;
223                 pidx++;
224         }
225         /* nodes brought online later all get mapped to pool0, sorry */
226 
227         return pidx;
228 }
229 
230 
231 /*
232  * Add a reference to the global map of cpus to pools (and
233  * vice versa).  Initialise the map if we're the first user.
234  * Returns the number of pools.
235  */
236 static unsigned int
237 svc_pool_map_get(void)
238 {
239         struct svc_pool_map *m = &svc_pool_map;
240         int npools = -1;
241 
242         mutex_lock(&svc_pool_map_mutex);
243 
244         if (m->count++) {
245                 mutex_unlock(&svc_pool_map_mutex);
246                 return m->npools;
247         }
248 
249         if (m->mode == SVC_POOL_AUTO)
250                 m->mode = svc_pool_map_choose_mode();
251 
252         switch (m->mode) {
253         case SVC_POOL_PERCPU:
254                 npools = svc_pool_map_init_percpu(m);
255                 break;
256         case SVC_POOL_PERNODE:
257                 npools = svc_pool_map_init_pernode(m);
258                 break;
259         }
260 
261         if (npools < 0) {
262                 /* default, or memory allocation failure */
263                 npools = 1;
264                 m->mode = SVC_POOL_GLOBAL;
265         }
266         m->npools = npools;
267 
268         mutex_unlock(&svc_pool_map_mutex);
269         return m->npools;
270 }
271 
272 
273 /*
274  * Drop a reference to the global map of cpus to pools.
275  * When the last reference is dropped, the map data is
276  * freed; this allows the sysadmin to change the pool
277  * mode using the pool_mode module option without
278  * rebooting or re-loading sunrpc.ko.
279  */
280 static void
281 svc_pool_map_put(void)
282 {
283         struct svc_pool_map *m = &svc_pool_map;
284 
285         mutex_lock(&svc_pool_map_mutex);
286 
287         if (!--m->count) {
288                 m->mode = SVC_POOL_DEFAULT;
289                 kfree(m->to_pool);
290                 m->to_pool = NULL;
291                 kfree(m->pool_to);
292                 m->pool_to = NULL;
293                 m->npools = 0;
294         }
295 
296         mutex_unlock(&svc_pool_map_mutex);
297 }
298 
299 
300 /*
301  * Set the given thread's cpus_allowed mask so that it
302  * will only run on cpus in the given pool.
303  */
304 static inline void
305 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx)
306 {
307         struct svc_pool_map *m = &svc_pool_map;
308         unsigned int node = m->pool_to[pidx];
309 
310         /*
311          * The caller checks for sv_nrpools > 1, which
312          * implies that we've been initialized.
313          */
314         BUG_ON(m->count == 0);
315 
316         switch (m->mode) {
317         case SVC_POOL_PERCPU:
318         {
319                 set_cpus_allowed_ptr(task, cpumask_of(node));
320                 break;
321         }
322         case SVC_POOL_PERNODE:
323         {
324                 set_cpus_allowed_ptr(task, cpumask_of_node(node));
325                 break;
326         }
327         }
328 }
329 
330 /*
331  * Use the mapping mode to choose a pool for a given CPU.
332  * Used when enqueueing an incoming RPC.  Always returns
333  * a non-NULL pool pointer.
334  */
335 struct svc_pool *
336 svc_pool_for_cpu(struct svc_serv *serv, int cpu)
337 {
338         struct svc_pool_map *m = &svc_pool_map;
339         unsigned int pidx = 0;
340 
341         /*
342          * An uninitialised map happens in a pure client when
343          * lockd is brought up, so silently treat it the
344          * same as SVC_POOL_GLOBAL.
345          */
346         if (svc_serv_is_pooled(serv)) {
347                 switch (m->mode) {
348                 case SVC_POOL_PERCPU:
349                         pidx = m->to_pool[cpu];
350                         break;
351                 case SVC_POOL_PERNODE:
352                         pidx = m->to_pool[cpu_to_node(cpu)];
353                         break;
354                 }
355         }
356         return &serv->sv_pools[pidx % serv->sv_nrpools];
357 }
358 
359 
360 /*
361  * Create an RPC service
362  */
363 static struct svc_serv *
364 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools,
365              void (*shutdown)(struct svc_serv *serv))
366 {
367         struct svc_serv *serv;
368         unsigned int vers;
369         unsigned int xdrsize;
370         unsigned int i;
371 
372         if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL)))
373                 return NULL;
374         serv->sv_name      = prog->pg_name;
375         serv->sv_program   = prog;
376         serv->sv_nrthreads = 1;
377         serv->sv_stats     = prog->pg_stats;
378         if (bufsize > RPCSVC_MAXPAYLOAD)
379                 bufsize = RPCSVC_MAXPAYLOAD;
380         serv->sv_max_payload = bufsize? bufsize : 4096;
381         serv->sv_max_mesg  = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE);
382         serv->sv_shutdown  = shutdown;
383         xdrsize = 0;
384         while (prog) {
385                 prog->pg_lovers = prog->pg_nvers-1;
386                 for (vers=0; vers<prog->pg_nvers ; vers++)
387                         if (prog->pg_vers[vers]) {
388                                 prog->pg_hivers = vers;
389                                 if (prog->pg_lovers > vers)
390                                         prog->pg_lovers = vers;
391                                 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize)
392                                         xdrsize = prog->pg_vers[vers]->vs_xdrsize;
393                         }
394                 prog = prog->pg_next;
395         }
396         serv->sv_xdrsize   = xdrsize;
397         INIT_LIST_HEAD(&serv->sv_tempsocks);
398         INIT_LIST_HEAD(&serv->sv_permsocks);
399         init_timer(&serv->sv_temptimer);
400         spin_lock_init(&serv->sv_lock);
401 
402         serv->sv_nrpools = npools;
403         serv->sv_pools =
404                 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool),
405                         GFP_KERNEL);
406         if (!serv->sv_pools) {
407                 kfree(serv);
408                 return NULL;
409         }
410 
411         for (i = 0; i < serv->sv_nrpools; i++) {
412                 struct svc_pool *pool = &serv->sv_pools[i];
413 
414                 dprintk("svc: initialising pool %u for %s\n",
415                                 i, serv->sv_name);
416 
417                 pool->sp_id = i;
418                 INIT_LIST_HEAD(&pool->sp_threads);
419                 INIT_LIST_HEAD(&pool->sp_sockets);
420                 INIT_LIST_HEAD(&pool->sp_all_threads);
421                 spin_lock_init(&pool->sp_lock);
422         }
423 
424         /* Remove any stale portmap registrations */
425         svc_unregister(serv);
426 
427         return serv;
428 }
429 
430 struct svc_serv *
431 svc_create(struct svc_program *prog, unsigned int bufsize,
432            void (*shutdown)(struct svc_serv *serv))
433 {
434         return __svc_create(prog, bufsize, /*npools*/1, shutdown);
435 }
436 EXPORT_SYMBOL_GPL(svc_create);
437 
438 struct svc_serv *
439 svc_create_pooled(struct svc_program *prog, unsigned int bufsize,
440                   void (*shutdown)(struct svc_serv *serv),
441                   svc_thread_fn func, struct module *mod)
442 {
443         struct svc_serv *serv;
444         unsigned int npools = svc_pool_map_get();
445 
446         serv = __svc_create(prog, bufsize, npools, shutdown);
447 
448         if (serv != NULL) {
449                 serv->sv_function = func;
450                 serv->sv_module = mod;
451         }
452 
453         return serv;
454 }
455 EXPORT_SYMBOL_GPL(svc_create_pooled);
456 
457 /*
458  * Destroy an RPC service. Should be called with appropriate locking to
459  * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
460  */
461 void
462 svc_destroy(struct svc_serv *serv)
463 {
464         dprintk("svc: svc_destroy(%s, %d)\n",
465                                 serv->sv_program->pg_name,
466                                 serv->sv_nrthreads);
467 
468         if (serv->sv_nrthreads) {
469                 if (--(serv->sv_nrthreads) != 0) {
470                         svc_sock_update_bufs(serv);
471                         return;
472                 }
473         } else
474                 printk("svc_destroy: no threads for serv=%p!\n", serv);
475 
476         del_timer_sync(&serv->sv_temptimer);
477 
478         svc_close_all(serv);
479 
480         if (serv->sv_shutdown)
481                 serv->sv_shutdown(serv);
482 
483         cache_clean_deferred(serv);
484 
485         if (svc_serv_is_pooled(serv))
486                 svc_pool_map_put();
487 
488 #if defined(CONFIG_NFS_V4_1)
489         svc_sock_destroy(serv->bc_xprt);
490 #endif /* CONFIG_NFS_V4_1 */
491 
492         svc_unregister(serv);
493         kfree(serv->sv_pools);
494         kfree(serv);
495 }
496 EXPORT_SYMBOL_GPL(svc_destroy);
497 
498 /*
499  * Allocate an RPC server's buffer space.
500  * We allocate pages and place them in rq_argpages.
501  */
502 static int
503 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size)
504 {
505         unsigned int pages, arghi;
506 
507         pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
508                                        * We assume one is at most one page
509                                        */
510         arghi = 0;
511         BUG_ON(pages > RPCSVC_MAXPAGES);
512         while (pages) {
513                 struct page *p = alloc_page(GFP_KERNEL);
514                 if (!p)
515                         break;
516                 rqstp->rq_pages[arghi++] = p;
517                 pages--;
518         }
519         return pages == 0;
520 }
521 
522 /*
523  * Release an RPC server buffer
524  */
525 static void
526 svc_release_buffer(struct svc_rqst *rqstp)
527 {
528         unsigned int i;
529 
530         for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++)
531                 if (rqstp->rq_pages[i])
532                         put_page(rqstp->rq_pages[i]);
533 }
534 
535 struct svc_rqst *
536 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool)
537 {
538         struct svc_rqst *rqstp;
539 
540         rqstp = kzalloc(sizeof(*rqstp), GFP_KERNEL);
541         if (!rqstp)
542                 goto out_enomem;
543 
544         init_waitqueue_head(&rqstp->rq_wait);
545 
546         serv->sv_nrthreads++;
547         spin_lock_bh(&pool->sp_lock);
548         pool->sp_nrthreads++;
549         list_add(&rqstp->rq_all, &pool->sp_all_threads);
550         spin_unlock_bh(&pool->sp_lock);
551         rqstp->rq_server = serv;
552         rqstp->rq_pool = pool;
553 
554         rqstp->rq_argp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
555         if (!rqstp->rq_argp)
556                 goto out_thread;
557 
558         rqstp->rq_resp = kmalloc(serv->sv_xdrsize, GFP_KERNEL);
559         if (!rqstp->rq_resp)
560                 goto out_thread;
561 
562         if (!svc_init_buffer(rqstp, serv->sv_max_mesg))
563                 goto out_thread;
564 
565         return rqstp;
566 out_thread:
567         svc_exit_thread(rqstp);
568 out_enomem:
569         return ERR_PTR(-ENOMEM);
570 }
571 EXPORT_SYMBOL_GPL(svc_prepare_thread);
572 
573 /*
574  * Choose a pool in which to create a new thread, for svc_set_num_threads
575  */
576 static inline struct svc_pool *
577 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
578 {
579         if (pool != NULL)
580                 return pool;
581 
582         return &serv->sv_pools[(*state)++ % serv->sv_nrpools];
583 }
584 
585 /*
586  * Choose a thread to kill, for svc_set_num_threads
587  */
588 static inline struct task_struct *
589 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state)
590 {
591         unsigned int i;
592         struct task_struct *task = NULL;
593 
594         if (pool != NULL) {
595                 spin_lock_bh(&pool->sp_lock);
596         } else {
597                 /* choose a pool in round-robin fashion */
598                 for (i = 0; i < serv->sv_nrpools; i++) {
599                         pool = &serv->sv_pools[--(*state) % serv->sv_nrpools];
600                         spin_lock_bh(&pool->sp_lock);
601                         if (!list_empty(&pool->sp_all_threads))
602                                 goto found_pool;
603                         spin_unlock_bh(&pool->sp_lock);
604                 }
605                 return NULL;
606         }
607 
608 found_pool:
609         if (!list_empty(&pool->sp_all_threads)) {
610                 struct svc_rqst *rqstp;
611 
612                 /*
613                  * Remove from the pool->sp_all_threads list
614                  * so we don't try to kill it again.
615                  */
616                 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all);
617                 list_del_init(&rqstp->rq_all);
618                 task = rqstp->rq_task;
619         }
620         spin_unlock_bh(&pool->sp_lock);
621 
622         return task;
623 }
624 
625 /*
626  * Create or destroy enough new threads to make the number
627  * of threads the given number.  If `pool' is non-NULL, applies
628  * only to threads in that pool, otherwise round-robins between
629  * all pools.  Must be called with a svc_get() reference and
630  * the BKL or another lock to protect access to svc_serv fields.
631  *
632  * Destroying threads relies on the service threads filling in
633  * rqstp->rq_task, which only the nfs ones do.  Assumes the serv
634  * has been created using svc_create_pooled().
635  *
636  * Based on code that used to be in nfsd_svc() but tweaked
637  * to be pool-aware.
638  */
639 int
640 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
641 {
642         struct svc_rqst *rqstp;
643         struct task_struct *task;
644         struct svc_pool *chosen_pool;
645         int error = 0;
646         unsigned int state = serv->sv_nrthreads-1;
647 
648         if (pool == NULL) {
649                 /* The -1 assumes caller has done a svc_get() */
650                 nrservs -= (serv->sv_nrthreads-1);
651         } else {
652                 spin_lock_bh(&pool->sp_lock);
653                 nrservs -= pool->sp_nrthreads;
654                 spin_unlock_bh(&pool->sp_lock);
655         }
656 
657         /* create new threads */
658         while (nrservs > 0) {
659                 nrservs--;
660                 chosen_pool = choose_pool(serv, pool, &state);
661 
662                 rqstp = svc_prepare_thread(serv, chosen_pool);
663                 if (IS_ERR(rqstp)) {
664                         error = PTR_ERR(rqstp);
665                         break;
666                 }
667 
668                 __module_get(serv->sv_module);
669                 task = kthread_create(serv->sv_function, rqstp, serv->sv_name);
670                 if (IS_ERR(task)) {
671                         error = PTR_ERR(task);
672                         module_put(serv->sv_module);
673                         svc_exit_thread(rqstp);
674                         break;
675                 }
676 
677                 rqstp->rq_task = task;
678                 if (serv->sv_nrpools > 1)
679                         svc_pool_map_set_cpumask(task, chosen_pool->sp_id);
680 
681                 svc_sock_update_bufs(serv);
682                 wake_up_process(task);
683         }
684         /* destroy old threads */
685         while (nrservs < 0 &&
686                (task = choose_victim(serv, pool, &state)) != NULL) {
687                 send_sig(SIGINT, task, 1);
688                 nrservs++;
689         }
690 
691         return error;
692 }
693 EXPORT_SYMBOL_GPL(svc_set_num_threads);
694 
695 /*
696  * Called from a server thread as it's exiting. Caller must hold the BKL or
697  * the "service mutex", whichever is appropriate for the service.
698  */
699 void
700 svc_exit_thread(struct svc_rqst *rqstp)
701 {
702         struct svc_serv *serv = rqstp->rq_server;
703         struct svc_pool *pool = rqstp->rq_pool;
704 
705         svc_release_buffer(rqstp);
706         kfree(rqstp->rq_resp);
707         kfree(rqstp->rq_argp);
708         kfree(rqstp->rq_auth_data);
709 
710         spin_lock_bh(&pool->sp_lock);
711         pool->sp_nrthreads--;
712         list_del(&rqstp->rq_all);
713         spin_unlock_bh(&pool->sp_lock);
714 
715         kfree(rqstp);
716 
717         /* Release the server */
718         if (serv)
719                 svc_destroy(serv);
720 }
721 EXPORT_SYMBOL_GPL(svc_exit_thread);
722 
723 /*
724  * Register an "inet" protocol family netid with the local
725  * rpcbind daemon via an rpcbind v4 SET request.
726  *
727  * No netconfig infrastructure is available in the kernel, so
728  * we map IP_ protocol numbers to netids by hand.
729  *
730  * Returns zero on success; a negative errno value is returned
731  * if any error occurs.
732  */
733 static int __svc_rpcb_register4(const u32 program, const u32 version,
734                                 const unsigned short protocol,
735                                 const unsigned short port)
736 {
737         const struct sockaddr_in sin = {
738                 .sin_family             = AF_INET,
739                 .sin_addr.s_addr        = htonl(INADDR_ANY),
740                 .sin_port               = htons(port),
741         };
742         const char *netid;
743         int error;
744 
745         switch (protocol) {
746         case IPPROTO_UDP:
747                 netid = RPCBIND_NETID_UDP;
748                 break;
749         case IPPROTO_TCP:
750                 netid = RPCBIND_NETID_TCP;
751                 break;
752         default:
753                 return -ENOPROTOOPT;
754         }
755 
756         error = rpcb_v4_register(program, version,
757                                         (const struct sockaddr *)&sin, netid);
758 
759         /*
760          * User space didn't support rpcbind v4, so retry this
761          * registration request with the legacy rpcbind v2 protocol.
762          */
763         if (error == -EPROTONOSUPPORT)
764                 error = rpcb_register(program, version, protocol, port);
765 
766         return error;
767 }
768 
769 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
770 /*
771  * Register an "inet6" protocol family netid with the local
772  * rpcbind daemon via an rpcbind v4 SET request.
773  *
774  * No netconfig infrastructure is available in the kernel, so
775  * we map IP_ protocol numbers to netids by hand.
776  *
777  * Returns zero on success; a negative errno value is returned
778  * if any error occurs.
779  */
780 static int __svc_rpcb_register6(const u32 program, const u32 version,
781                                 const unsigned short protocol,
782                                 const unsigned short port)
783 {
784         const struct sockaddr_in6 sin6 = {
785                 .sin6_family            = AF_INET6,
786                 .sin6_addr              = IN6ADDR_ANY_INIT,
787                 .sin6_port              = htons(port),
788         };
789         const char *netid;
790         int error;
791 
792         switch (protocol) {
793         case IPPROTO_UDP:
794                 netid = RPCBIND_NETID_UDP6;
795                 break;
796         case IPPROTO_TCP:
797                 netid = RPCBIND_NETID_TCP6;
798                 break;
799         default:
800                 return -ENOPROTOOPT;
801         }
802 
803         error = rpcb_v4_register(program, version,
804                                         (const struct sockaddr *)&sin6, netid);
805 
806         /*
807          * User space didn't support rpcbind version 4, so we won't
808          * use a PF_INET6 listener.
809          */
810         if (error == -EPROTONOSUPPORT)
811                 error = -EAFNOSUPPORT;
812 
813         return error;
814 }
815 #endif  /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
816 
817 /*
818  * Register a kernel RPC service via rpcbind version 4.
819  *
820  * Returns zero on success; a negative errno value is returned
821  * if any error occurs.
822  */
823 static int __svc_register(const char *progname,
824                           const u32 program, const u32 version,
825                           const int family,
826                           const unsigned short protocol,
827                           const unsigned short port)
828 {
829         int error = -EAFNOSUPPORT;
830 
831         switch (family) {
832         case PF_INET:
833                 error = __svc_rpcb_register4(program, version,
834                                                 protocol, port);
835                 break;
836 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
837         case PF_INET6:
838                 error = __svc_rpcb_register6(program, version,
839                                                 protocol, port);
840 #endif  /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
841         }
842 
843         if (error < 0)
844                 printk(KERN_WARNING "svc: failed to register %sv%u RPC "
845                         "service (errno %d).\n", progname, version, -error);
846         return error;
847 }
848 
849 /**
850  * svc_register - register an RPC service with the local portmapper
851  * @serv: svc_serv struct for the service to register
852  * @family: protocol family of service's listener socket
853  * @proto: transport protocol number to advertise
854  * @port: port to advertise
855  *
856  * Service is registered for any address in the passed-in protocol family
857  */
858 int svc_register(const struct svc_serv *serv, const int family,
859                  const unsigned short proto, const unsigned short port)
860 {
861         struct svc_program      *progp;
862         unsigned int            i;
863         int                     error = 0;
864 
865         BUG_ON(proto == 0 && port == 0);
866 
867         for (progp = serv->sv_program; progp; progp = progp->pg_next) {
868                 for (i = 0; i < progp->pg_nvers; i++) {
869                         if (progp->pg_vers[i] == NULL)
870                                 continue;
871 
872                         dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
873                                         progp->pg_name,
874                                         i,
875                                         proto == IPPROTO_UDP?  "udp" : "tcp",
876                                         port,
877                                         family,
878                                         progp->pg_vers[i]->vs_hidden?
879                                                 " (but not telling portmap)" : "");
880 
881                         if (progp->pg_vers[i]->vs_hidden)
882                                 continue;
883 
884                         error = __svc_register(progp->pg_name, progp->pg_prog,
885                                                 i, family, proto, port);
886                         if (error < 0)
887                                 break;
888                 }
889         }
890 
891         return error;
892 }
893 
894 /*
895  * If user space is running rpcbind, it should take the v4 UNSET
896  * and clear everything for this [program, version].  If user space
897  * is running portmap, it will reject the v4 UNSET, but won't have
898  * any "inet6" entries anyway.  So a PMAP_UNSET should be sufficient
899  * in this case to clear all existing entries for [program, version].
900  */
901 static void __svc_unregister(const u32 program, const u32 version,
902                              const char *progname)
903 {
904         int error;
905 
906         error = rpcb_v4_register(program, version, NULL, "");
907 
908         /*
909          * User space didn't support rpcbind v4, so retry this
910          * request with the legacy rpcbind v2 protocol.
911          */
912         if (error == -EPROTONOSUPPORT)
913                 error = rpcb_register(program, version, 0, 0);
914 
915         dprintk("svc: %s(%sv%u), error %d\n",
916                         __func__, progname, version, error);
917 }
918 
919 /*
920  * All netids, bind addresses and ports registered for [program, version]
921  * are removed from the local rpcbind database (if the service is not
922  * hidden) to make way for a new instance of the service.
923  *
924  * The result of unregistration is reported via dprintk for those who want
925  * verification of the result, but is otherwise not important.
926  */
927 static void svc_unregister(const struct svc_serv *serv)
928 {
929         struct svc_program *progp;
930         unsigned long flags;
931         unsigned int i;
932 
933         clear_thread_flag(TIF_SIGPENDING);
934 
935         for (progp = serv->sv_program; progp; progp = progp->pg_next) {
936                 for (i = 0; i < progp->pg_nvers; i++) {
937                         if (progp->pg_vers[i] == NULL)
938                                 continue;
939                         if (progp->pg_vers[i]->vs_hidden)
940                                 continue;
941 
942                         __svc_unregister(progp->pg_prog, i, progp->pg_name);
943                 }
944         }
945 
946         spin_lock_irqsave(&current->sighand->siglock, flags);
947         recalc_sigpending();
948         spin_unlock_irqrestore(&current->sighand->siglock, flags);
949 }
950 
951 /*
952  * Printk the given error with the address of the client that caused it.
953  */
954 static int
955 __attribute__ ((format (printf, 2, 3)))
956 svc_printk(struct svc_rqst *rqstp, const char *fmt, ...)
957 {
958         va_list args;
959         int     r;
960         char    buf[RPC_MAX_ADDRBUFLEN];
961 
962         if (!net_ratelimit())
963                 return 0;
964 
965         printk(KERN_WARNING "svc: %s: ",
966                 svc_print_addr(rqstp, buf, sizeof(buf)));
967 
968         va_start(args, fmt);
969         r = vprintk(fmt, args);
970         va_end(args);
971 
972         return r;
973 }
974 
975 /*
976  * Common routine for processing the RPC request.
977  */
978 static int
979 svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv)
980 {
981         struct svc_program      *progp;
982         struct svc_version      *versp = NULL;  /* compiler food */
983         struct svc_procedure    *procp = NULL;
984         struct svc_serv         *serv = rqstp->rq_server;
985         kxdrproc_t              xdr;
986         __be32                  *statp;
987         u32                     prog, vers, proc;
988         __be32                  auth_stat, rpc_stat;
989         int                     auth_res;
990         __be32                  *reply_statp;
991 
992         rpc_stat = rpc_success;
993 
994         if (argv->iov_len < 6*4)
995                 goto err_short_len;
996 
997         /* Will be turned off only in gss privacy case: */
998         rqstp->rq_splice_ok = 1;
999         /* Will be turned off only when NFSv4 Sessions are used */
1000         rqstp->rq_usedeferral = 1;
1001 
1002         /* Setup reply header */
1003         rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp);
1004 
1005         svc_putu32(resv, rqstp->rq_xid);
1006 
1007         vers = svc_getnl(argv);
1008 
1009         /* First words of reply: */
1010         svc_putnl(resv, 1);             /* REPLY */
1011 
1012         if (vers != 2)          /* RPC version number */
1013                 goto err_bad_rpc;
1014 
1015         /* Save position in case we later decide to reject: */
1016         reply_statp = resv->iov_base + resv->iov_len;
1017 
1018         svc_putnl(resv, 0);             /* ACCEPT */
1019 
1020         rqstp->rq_prog = prog = svc_getnl(argv);        /* program number */
1021         rqstp->rq_vers = vers = svc_getnl(argv);        /* version number */
1022         rqstp->rq_proc = proc = svc_getnl(argv);        /* procedure number */
1023 
1024         progp = serv->sv_program;
1025 
1026         for (progp = serv->sv_program; progp; progp = progp->pg_next)
1027                 if (prog == progp->pg_prog)
1028                         break;
1029 
1030         /*
1031          * Decode auth data, and add verifier to reply buffer.
1032          * We do this before anything else in order to get a decent
1033          * auth verifier.
1034          */
1035         auth_res = svc_authenticate(rqstp, &auth_stat);
1036         /* Also give the program a chance to reject this call: */
1037         if (auth_res == SVC_OK && progp) {
1038                 auth_stat = rpc_autherr_badcred;
1039                 auth_res = progp->pg_authenticate(rqstp);
1040         }
1041         switch (auth_res) {
1042         case SVC_OK:
1043                 break;
1044         case SVC_GARBAGE:
1045                 goto err_garbage;
1046         case SVC_SYSERR:
1047                 rpc_stat = rpc_system_err;
1048                 goto err_bad;
1049         case SVC_DENIED:
1050                 goto err_bad_auth;
1051         case SVC_DROP:
1052                 goto dropit;
1053         case SVC_COMPLETE:
1054                 goto sendit;
1055         }
1056 
1057         if (progp == NULL)
1058                 goto err_bad_prog;
1059 
1060         if (vers >= progp->pg_nvers ||
1061           !(versp = progp->pg_vers[vers]))
1062                 goto err_bad_vers;
1063 
1064         procp = versp->vs_proc + proc;
1065         if (proc >= versp->vs_nproc || !procp->pc_func)
1066                 goto err_bad_proc;
1067         rqstp->rq_procinfo = procp;
1068 
1069         /* Syntactic check complete */
1070         serv->sv_stats->rpccnt++;
1071 
1072         /* Build the reply header. */
1073         statp = resv->iov_base +resv->iov_len;
1074         svc_putnl(resv, RPC_SUCCESS);
1075 
1076         /* Bump per-procedure stats counter */
1077         procp->pc_count++;
1078 
1079         /* Initialize storage for argp and resp */
1080         memset(rqstp->rq_argp, 0, procp->pc_argsize);
1081         memset(rqstp->rq_resp, 0, procp->pc_ressize);
1082 
1083         /* un-reserve some of the out-queue now that we have a
1084          * better idea of reply size
1085          */
1086         if (procp->pc_xdrressize)
1087                 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2);
1088 
1089         /* Call the function that processes the request. */
1090         if (!versp->vs_dispatch) {
1091                 /* Decode arguments */
1092                 xdr = procp->pc_decode;
1093                 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp))
1094                         goto err_garbage;
1095 
1096                 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
1097 
1098                 /* Encode reply */
1099                 if (*statp == rpc_drop_reply) {
1100                         if (procp->pc_release)
1101                                 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1102                         goto dropit;
1103                 }
1104                 if (*statp == rpc_success && (xdr = procp->pc_encode)
1105                  && !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) {
1106                         dprintk("svc: failed to encode reply\n");
1107                         /* serv->sv_stats->rpcsystemerr++; */
1108                         *statp = rpc_system_err;
1109                 }
1110         } else {
1111                 dprintk("svc: calling dispatcher\n");
1112                 if (!versp->vs_dispatch(rqstp, statp)) {
1113                         /* Release reply info */
1114                         if (procp->pc_release)
1115                                 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1116                         goto dropit;
1117                 }
1118         }
1119 
1120         /* Check RPC status result */
1121         if (*statp != rpc_success)
1122                 resv->iov_len = ((void*)statp)  - resv->iov_base + 4;
1123 
1124         /* Release reply info */
1125         if (procp->pc_release)
1126                 procp->pc_release(rqstp, NULL, rqstp->rq_resp);
1127 
1128         if (procp->pc_encode == NULL)
1129                 goto dropit;
1130 
1131  sendit:
1132         if (svc_authorise(rqstp))
1133                 goto dropit;
1134         return 1;               /* Caller can now send it */
1135 
1136  dropit:
1137         svc_authorise(rqstp);   /* doesn't hurt to call this twice */
1138         dprintk("svc: svc_process dropit\n");
1139         svc_drop(rqstp);
1140         return 0;
1141 
1142 err_short_len:
1143         svc_printk(rqstp, "short len %Zd, dropping request\n",
1144                         argv->iov_len);
1145 
1146         goto dropit;                    /* drop request */
1147 
1148 err_bad_rpc:
1149         serv->sv_stats->rpcbadfmt++;
1150         svc_putnl(resv, 1);     /* REJECT */
1151         svc_putnl(resv, 0);     /* RPC_MISMATCH */
1152         svc_putnl(resv, 2);     /* Only RPCv2 supported */
1153         svc_putnl(resv, 2);
1154         goto sendit;
1155 
1156 err_bad_auth:
1157         dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat));
1158         serv->sv_stats->rpcbadauth++;
1159         /* Restore write pointer to location of accept status: */
1160         xdr_ressize_check(rqstp, reply_statp);
1161         svc_putnl(resv, 1);     /* REJECT */
1162         svc_putnl(resv, 1);     /* AUTH_ERROR */
1163         svc_putnl(resv, ntohl(auth_stat));      /* status */
1164         goto sendit;
1165 
1166 err_bad_prog:
1167         dprintk("svc: unknown program %d\n", prog);
1168         serv->sv_stats->rpcbadfmt++;
1169         svc_putnl(resv, RPC_PROG_UNAVAIL);
1170         goto sendit;
1171 
1172 err_bad_vers:
1173         svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n",
1174                        vers, prog, progp->pg_name);
1175 
1176         serv->sv_stats->rpcbadfmt++;
1177         svc_putnl(resv, RPC_PROG_MISMATCH);
1178         svc_putnl(resv, progp->pg_lovers);
1179         svc_putnl(resv, progp->pg_hivers);
1180         goto sendit;
1181 
1182 err_bad_proc:
1183         svc_printk(rqstp, "unknown procedure (%d)\n", proc);
1184 
1185         serv->sv_stats->rpcbadfmt++;
1186         svc_putnl(resv, RPC_PROC_UNAVAIL);
1187         goto sendit;
1188 
1189 err_garbage:
1190         svc_printk(rqstp, "failed to decode args\n");
1191 
1192         rpc_stat = rpc_garbage_args;
1193 err_bad:
1194         serv->sv_stats->rpcbadfmt++;
1195         svc_putnl(resv, ntohl(rpc_stat));
1196         goto sendit;
1197 }
1198 EXPORT_SYMBOL_GPL(svc_process);
1199 
1200 /*
1201  * Process the RPC request.
1202  */
1203 int
1204 svc_process(struct svc_rqst *rqstp)
1205 {
1206         struct kvec             *argv = &rqstp->rq_arg.head[0];
1207         struct kvec             *resv = &rqstp->rq_res.head[0];
1208         struct svc_serv         *serv = rqstp->rq_server;
1209         u32                     dir;
1210         int                     error;
1211 
1212         /*
1213          * Setup response xdr_buf.
1214          * Initially it has just one page
1215          */
1216         rqstp->rq_resused = 1;
1217         resv->iov_base = page_address(rqstp->rq_respages[0]);
1218         resv->iov_len = 0;
1219         rqstp->rq_res.pages = rqstp->rq_respages + 1;
1220         rqstp->rq_res.len = 0;
1221         rqstp->rq_res.page_base = 0;
1222         rqstp->rq_res.page_len = 0;
1223         rqstp->rq_res.buflen = PAGE_SIZE;
1224         rqstp->rq_res.tail[0].iov_base = NULL;
1225         rqstp->rq_res.tail[0].iov_len = 0;
1226 
1227         rqstp->rq_xid = svc_getu32(argv);
1228 
1229         dir  = svc_getnl(argv);
1230         if (dir != 0) {
1231                 /* direction != CALL */
1232                 svc_printk(rqstp, "bad direction %d, dropping request\n", dir);
1233                 serv->sv_stats->rpcbadfmt++;
1234                 svc_drop(rqstp);
1235                 return 0;
1236         }
1237 
1238         error = svc_process_common(rqstp, argv, resv);
1239         if (error <= 0)
1240                 return error;
1241 
1242         return svc_send(rqstp);
1243 }
1244 
1245 #if defined(CONFIG_NFS_V4_1)
1246 /*
1247  * Process a backchannel RPC request that arrived over an existing
1248  * outbound connection
1249  */
1250 int
1251 bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
1252                struct svc_rqst *rqstp)
1253 {
1254         struct kvec     *argv = &rqstp->rq_arg.head[0];
1255         struct kvec     *resv = &rqstp->rq_res.head[0];
1256         int             error;
1257 
1258         /* Build the svc_rqst used by the common processing routine */
1259         rqstp->rq_xprt = serv->bc_xprt;
1260         rqstp->rq_xid = req->rq_xid;
1261         rqstp->rq_prot = req->rq_xprt->prot;
1262         rqstp->rq_server = serv;
1263 
1264         rqstp->rq_addrlen = sizeof(req->rq_xprt->addr);
1265         memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen);
1266         memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg));
1267         memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res));
1268 
1269         /* reset result send buffer "put" position */
1270         resv->iov_len = 0;
1271 
1272         if (rqstp->rq_prot != IPPROTO_TCP) {
1273                 printk(KERN_ERR "No support for Non-TCP transports!\n");
1274                 BUG();
1275         }
1276 
1277         /*
1278          * Skip the next two words because they've already been
1279          * processed in the trasport
1280          */
1281         svc_getu32(argv);       /* XID */
1282         svc_getnl(argv);        /* CALLDIR */
1283 
1284         error = svc_process_common(rqstp, argv, resv);
1285         if (error <= 0)
1286                 return error;
1287 
1288         memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf));
1289         return bc_send(req);
1290 }
1291 EXPORT_SYMBOL(bc_svc_process);
1292 #endif /* CONFIG_NFS_V4_1 */
1293 
1294 /*
1295  * Return (transport-specific) limit on the rpc payload.
1296  */
1297 u32 svc_max_payload(const struct svc_rqst *rqstp)
1298 {
1299         u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload;
1300 
1301         if (rqstp->rq_server->sv_max_payload < max)
1302                 max = rqstp->rq_server->sv_max_payload;
1303         return max;
1304 }
1305 EXPORT_SYMBOL_GPL(svc_max_payload);
1306 

~ [ 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