nfsd: pass net to nfsd_startup() and nfsd_shutdown()
[deliverable/linux.git] / fs / nfsd / nfssvc.c
1 /*
2 * Central processing for nfsd.
3 *
4 * Authors: Olaf Kirch (okir@monad.swb.de)
5 *
6 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
7 */
8
9 #include <linux/sched.h>
10 #include <linux/freezer.h>
11 #include <linux/module.h>
12 #include <linux/fs_struct.h>
13 #include <linux/swap.h>
14
15 #include <linux/sunrpc/stats.h>
16 #include <linux/sunrpc/svcsock.h>
17 #include <linux/lockd/bind.h>
18 #include <linux/nfsacl.h>
19 #include <linux/seq_file.h>
20 #include <net/net_namespace.h>
21 #include "nfsd.h"
22 #include "cache.h"
23 #include "vfs.h"
24
25 #define NFSDDBG_FACILITY NFSDDBG_SVC
26
27 extern struct svc_program nfsd_program;
28 static int nfsd(void *vrqstp);
29 struct timeval nfssvc_boot;
30
31 /*
32 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
33 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
34 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
35 *
36 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
37 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
38 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
39 * entry of ->sv_pools[].
40 *
41 * Transitions of the thread count between zero and non-zero are of particular
42 * interest since the svc_serv needs to be created and initialized at that
43 * point, or freed.
44 *
45 * Finally, the nfsd_mutex also protects some of the global variables that are
46 * accessed when nfsd starts and that are settable via the write_* routines in
47 * nfsctl.c. In particular:
48 *
49 * user_recovery_dirname
50 * user_lease_time
51 * nfsd_versions
52 */
53 DEFINE_MUTEX(nfsd_mutex);
54 struct svc_serv *nfsd_serv;
55
56 /*
57 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
58 * nfsd_drc_max_pages limits the total amount of memory available for
59 * version 4.1 DRC caches.
60 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
61 */
62 spinlock_t nfsd_drc_lock;
63 unsigned int nfsd_drc_max_mem;
64 unsigned int nfsd_drc_mem_used;
65
66 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
67 static struct svc_stat nfsd_acl_svcstats;
68 static struct svc_version * nfsd_acl_version[] = {
69 [2] = &nfsd_acl_version2,
70 [3] = &nfsd_acl_version3,
71 };
72
73 #define NFSD_ACL_MINVERS 2
74 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
75 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
76
77 static struct svc_program nfsd_acl_program = {
78 .pg_prog = NFS_ACL_PROGRAM,
79 .pg_nvers = NFSD_ACL_NRVERS,
80 .pg_vers = nfsd_acl_versions,
81 .pg_name = "nfsacl",
82 .pg_class = "nfsd",
83 .pg_stats = &nfsd_acl_svcstats,
84 .pg_authenticate = &svc_set_client,
85 };
86
87 static struct svc_stat nfsd_acl_svcstats = {
88 .program = &nfsd_acl_program,
89 };
90 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
91
92 static struct svc_version * nfsd_version[] = {
93 [2] = &nfsd_version2,
94 #if defined(CONFIG_NFSD_V3)
95 [3] = &nfsd_version3,
96 #endif
97 #if defined(CONFIG_NFSD_V4)
98 [4] = &nfsd_version4,
99 #endif
100 };
101
102 #define NFSD_MINVERS 2
103 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
104 static struct svc_version *nfsd_versions[NFSD_NRVERS];
105
106 struct svc_program nfsd_program = {
107 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
108 .pg_next = &nfsd_acl_program,
109 #endif
110 .pg_prog = NFS_PROGRAM, /* program number */
111 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
112 .pg_vers = nfsd_versions, /* version table */
113 .pg_name = "nfsd", /* program name */
114 .pg_class = "nfsd", /* authentication class */
115 .pg_stats = &nfsd_svcstats, /* version table */
116 .pg_authenticate = &svc_set_client, /* export authentication */
117
118 };
119
120 u32 nfsd_supported_minorversion;
121
122 int nfsd_vers(int vers, enum vers_op change)
123 {
124 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
125 return 0;
126 switch(change) {
127 case NFSD_SET:
128 nfsd_versions[vers] = nfsd_version[vers];
129 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
130 if (vers < NFSD_ACL_NRVERS)
131 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
132 #endif
133 break;
134 case NFSD_CLEAR:
135 nfsd_versions[vers] = NULL;
136 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
137 if (vers < NFSD_ACL_NRVERS)
138 nfsd_acl_versions[vers] = NULL;
139 #endif
140 break;
141 case NFSD_TEST:
142 return nfsd_versions[vers] != NULL;
143 case NFSD_AVAIL:
144 return nfsd_version[vers] != NULL;
145 }
146 return 0;
147 }
148
149 int nfsd_minorversion(u32 minorversion, enum vers_op change)
150 {
151 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
152 return -1;
153 switch(change) {
154 case NFSD_SET:
155 nfsd_supported_minorversion = minorversion;
156 break;
157 case NFSD_CLEAR:
158 if (minorversion == 0)
159 return -1;
160 nfsd_supported_minorversion = minorversion - 1;
161 break;
162 case NFSD_TEST:
163 return minorversion <= nfsd_supported_minorversion;
164 case NFSD_AVAIL:
165 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
166 }
167 return 0;
168 }
169
170 /*
171 * Maximum number of nfsd processes
172 */
173 #define NFSD_MAXSERVS 8192
174
175 int nfsd_nrthreads(void)
176 {
177 int rv = 0;
178 mutex_lock(&nfsd_mutex);
179 if (nfsd_serv)
180 rv = nfsd_serv->sv_nrthreads;
181 mutex_unlock(&nfsd_mutex);
182 return rv;
183 }
184
185 static int nfsd_init_socks(struct net *net)
186 {
187 int error;
188 if (!list_empty(&nfsd_serv->sv_permsocks))
189 return 0;
190
191 error = svc_create_xprt(nfsd_serv, "udp", net, PF_INET, NFS_PORT,
192 SVC_SOCK_DEFAULTS);
193 if (error < 0)
194 return error;
195
196 error = svc_create_xprt(nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
197 SVC_SOCK_DEFAULTS);
198 if (error < 0)
199 return error;
200
201 return 0;
202 }
203
204 static bool nfsd_up = false;
205
206 static int nfsd_startup(int nrservs, struct net *net)
207 {
208 int ret;
209
210 if (nfsd_up)
211 return 0;
212 /*
213 * Readahead param cache - will no-op if it already exists.
214 * (Note therefore results will be suboptimal if number of
215 * threads is modified after nfsd start.)
216 */
217 ret = nfsd_racache_init(2*nrservs);
218 if (ret)
219 return ret;
220 ret = nfsd_init_socks(net);
221 if (ret)
222 goto out_racache;
223 ret = lockd_up(net);
224 if (ret)
225 goto out_racache;
226 ret = nfs4_state_start();
227 if (ret)
228 goto out_lockd;
229
230 ret = nfs4_state_start_net(net);
231 if (ret)
232 goto out_net_state;
233
234 nfsd_up = true;
235 return 0;
236 out_net_state:
237 nfs4_state_shutdown();
238 out_lockd:
239 lockd_down(net);
240 out_racache:
241 nfsd_racache_shutdown();
242 return ret;
243 }
244
245 static void nfsd_shutdown(struct net *net)
246 {
247 /*
248 * write_ports can create the server without actually starting
249 * any threads--if we get shut down before any threads are
250 * started, then nfsd_last_thread will be run before any of this
251 * other initialization has been done.
252 */
253 if (!nfsd_up)
254 return;
255 nfs4_state_shutdown_net(net);
256 nfs4_state_shutdown();
257 lockd_down(net);
258 nfsd_racache_shutdown();
259 nfsd_up = false;
260 }
261
262 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
263 {
264 nfsd_shutdown(net);
265
266 svc_rpcb_cleanup(serv, net);
267
268 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
269 "cache\n");
270 nfsd_export_flush(net);
271 }
272
273 void nfsd_reset_versions(void)
274 {
275 int found_one = 0;
276 int i;
277
278 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
279 if (nfsd_program.pg_vers[i])
280 found_one = 1;
281 }
282
283 if (!found_one) {
284 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
285 nfsd_program.pg_vers[i] = nfsd_version[i];
286 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
287 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
288 nfsd_acl_program.pg_vers[i] =
289 nfsd_acl_version[i];
290 #endif
291 }
292 }
293
294 /*
295 * Each session guarantees a negotiated per slot memory cache for replies
296 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
297 * NFSv4.1 server might want to use more memory for a DRC than a machine
298 * with mutiple services.
299 *
300 * Impose a hard limit on the number of pages for the DRC which varies
301 * according to the machines free pages. This is of course only a default.
302 *
303 * For now this is a #defined shift which could be under admin control
304 * in the future.
305 */
306 static void set_max_drc(void)
307 {
308 #define NFSD_DRC_SIZE_SHIFT 10
309 nfsd_drc_max_mem = (nr_free_buffer_pages()
310 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
311 nfsd_drc_mem_used = 0;
312 spin_lock_init(&nfsd_drc_lock);
313 dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
314 }
315
316 static int nfsd_get_default_max_blksize(void)
317 {
318 struct sysinfo i;
319 unsigned long long target;
320 unsigned long ret;
321
322 si_meminfo(&i);
323 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
324 /*
325 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
326 * machines, but only uses 32K on 128M machines. Bottom out at
327 * 8K on 32M and smaller. Of course, this is only a default.
328 */
329 target >>= 12;
330
331 ret = NFSSVC_MAXBLKSIZE;
332 while (ret > target && ret >= 8*1024*2)
333 ret /= 2;
334 return ret;
335 }
336
337 int nfsd_create_serv(void)
338 {
339 int error;
340 struct net *net = &init_net;
341
342 WARN_ON(!mutex_is_locked(&nfsd_mutex));
343 if (nfsd_serv) {
344 svc_get(nfsd_serv);
345 return 0;
346 }
347 if (nfsd_max_blksize == 0)
348 nfsd_max_blksize = nfsd_get_default_max_blksize();
349 nfsd_reset_versions();
350 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
351 nfsd_last_thread, nfsd, THIS_MODULE);
352 if (nfsd_serv == NULL)
353 return -ENOMEM;
354
355 error = svc_bind(nfsd_serv, net);
356 if (error < 0) {
357 svc_destroy(nfsd_serv);
358 return error;
359 }
360
361 set_max_drc();
362 do_gettimeofday(&nfssvc_boot); /* record boot time */
363 return 0;
364 }
365
366 int nfsd_nrpools(void)
367 {
368 if (nfsd_serv == NULL)
369 return 0;
370 else
371 return nfsd_serv->sv_nrpools;
372 }
373
374 int nfsd_get_nrthreads(int n, int *nthreads)
375 {
376 int i = 0;
377
378 if (nfsd_serv != NULL) {
379 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
380 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
381 }
382
383 return 0;
384 }
385
386 int nfsd_set_nrthreads(int n, int *nthreads)
387 {
388 int i = 0;
389 int tot = 0;
390 int err = 0;
391 struct net *net = &init_net;
392
393 WARN_ON(!mutex_is_locked(&nfsd_mutex));
394
395 if (nfsd_serv == NULL || n <= 0)
396 return 0;
397
398 if (n > nfsd_serv->sv_nrpools)
399 n = nfsd_serv->sv_nrpools;
400
401 /* enforce a global maximum number of threads */
402 tot = 0;
403 for (i = 0; i < n; i++) {
404 if (nthreads[i] > NFSD_MAXSERVS)
405 nthreads[i] = NFSD_MAXSERVS;
406 tot += nthreads[i];
407 }
408 if (tot > NFSD_MAXSERVS) {
409 /* total too large: scale down requested numbers */
410 for (i = 0; i < n && tot > 0; i++) {
411 int new = nthreads[i] * NFSD_MAXSERVS / tot;
412 tot -= (nthreads[i] - new);
413 nthreads[i] = new;
414 }
415 for (i = 0; i < n && tot > 0; i++) {
416 nthreads[i]--;
417 tot--;
418 }
419 }
420
421 /*
422 * There must always be a thread in pool 0; the admin
423 * can't shut down NFS completely using pool_threads.
424 */
425 if (nthreads[0] == 0)
426 nthreads[0] = 1;
427
428 /* apply the new numbers */
429 svc_get(nfsd_serv);
430 for (i = 0; i < n; i++) {
431 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
432 nthreads[i]);
433 if (err)
434 break;
435 }
436 nfsd_destroy(net);
437 return err;
438 }
439
440 /*
441 * Adjust the number of threads and return the new number of threads.
442 * This is also the function that starts the server if necessary, if
443 * this is the first time nrservs is nonzero.
444 */
445 int
446 nfsd_svc(int nrservs)
447 {
448 int error;
449 bool nfsd_up_before;
450 struct net *net = &init_net;
451
452 mutex_lock(&nfsd_mutex);
453 dprintk("nfsd: creating service\n");
454 if (nrservs <= 0)
455 nrservs = 0;
456 if (nrservs > NFSD_MAXSERVS)
457 nrservs = NFSD_MAXSERVS;
458 error = 0;
459 if (nrservs == 0 && nfsd_serv == NULL)
460 goto out;
461
462 error = nfsd_create_serv();
463 if (error)
464 goto out;
465
466 nfsd_up_before = nfsd_up;
467
468 error = nfsd_startup(nrservs, net);
469 if (error)
470 goto out_destroy;
471 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
472 if (error)
473 goto out_shutdown;
474 /* We are holding a reference to nfsd_serv which
475 * we don't want to count in the return value,
476 * so subtract 1
477 */
478 error = nfsd_serv->sv_nrthreads - 1;
479 out_shutdown:
480 if (error < 0 && !nfsd_up_before)
481 nfsd_shutdown(net);
482 out_destroy:
483 nfsd_destroy(net); /* Release server */
484 out:
485 mutex_unlock(&nfsd_mutex);
486 return error;
487 }
488
489
490 /*
491 * This is the NFS server kernel thread
492 */
493 static int
494 nfsd(void *vrqstp)
495 {
496 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
497 int err;
498
499 /* Lock module and set up kernel thread */
500 mutex_lock(&nfsd_mutex);
501
502 /* At this point, the thread shares current->fs
503 * with the init process. We need to create files with a
504 * umask of 0 instead of init's umask. */
505 if (unshare_fs_struct() < 0) {
506 printk("Unable to start nfsd thread: out of memory\n");
507 goto out;
508 }
509
510 current->fs->umask = 0;
511
512 /*
513 * thread is spawned with all signals set to SIG_IGN, re-enable
514 * the ones that will bring down the thread
515 */
516 allow_signal(SIGKILL);
517 allow_signal(SIGHUP);
518 allow_signal(SIGINT);
519 allow_signal(SIGQUIT);
520
521 nfsdstats.th_cnt++;
522 mutex_unlock(&nfsd_mutex);
523
524 /*
525 * We want less throttling in balance_dirty_pages() so that nfs to
526 * localhost doesn't cause nfsd to lock up due to all the client's
527 * dirty pages.
528 */
529 current->flags |= PF_LESS_THROTTLE;
530 set_freezable();
531
532 /*
533 * The main request loop
534 */
535 for (;;) {
536 /*
537 * Find a socket with data available and call its
538 * recvfrom routine.
539 */
540 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
541 ;
542 if (err == -EINTR)
543 break;
544 validate_process_creds();
545 svc_process(rqstp);
546 validate_process_creds();
547 }
548
549 /* Clear signals before calling svc_exit_thread() */
550 flush_signals(current);
551
552 mutex_lock(&nfsd_mutex);
553 nfsdstats.th_cnt --;
554
555 out:
556 rqstp->rq_server = NULL;
557
558 /* Release the thread */
559 svc_exit_thread(rqstp);
560
561 nfsd_destroy(&init_net);
562
563 /* Release module */
564 mutex_unlock(&nfsd_mutex);
565 module_put_and_exit(0);
566 return 0;
567 }
568
569 static __be32 map_new_errors(u32 vers, __be32 nfserr)
570 {
571 if (nfserr == nfserr_jukebox && vers == 2)
572 return nfserr_dropit;
573 if (nfserr == nfserr_wrongsec && vers < 4)
574 return nfserr_acces;
575 return nfserr;
576 }
577
578 int
579 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
580 {
581 struct svc_procedure *proc;
582 kxdrproc_t xdr;
583 __be32 nfserr;
584 __be32 *nfserrp;
585
586 dprintk("nfsd_dispatch: vers %d proc %d\n",
587 rqstp->rq_vers, rqstp->rq_proc);
588 proc = rqstp->rq_procinfo;
589
590 /*
591 * Give the xdr decoder a chance to change this if it wants
592 * (necessary in the NFSv4.0 compound case)
593 */
594 rqstp->rq_cachetype = proc->pc_cachetype;
595 /* Decode arguments */
596 xdr = proc->pc_decode;
597 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
598 rqstp->rq_argp)) {
599 dprintk("nfsd: failed to decode arguments!\n");
600 *statp = rpc_garbage_args;
601 return 1;
602 }
603
604 /* Check whether we have this call in the cache. */
605 switch (nfsd_cache_lookup(rqstp)) {
606 case RC_INTR:
607 case RC_DROPIT:
608 return 0;
609 case RC_REPLY:
610 return 1;
611 case RC_DOIT:;
612 /* do it */
613 }
614
615 /* need to grab the location to store the status, as
616 * nfsv4 does some encoding while processing
617 */
618 nfserrp = rqstp->rq_res.head[0].iov_base
619 + rqstp->rq_res.head[0].iov_len;
620 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
621
622 /* Now call the procedure handler, and encode NFS status. */
623 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
624 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
625 if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
626 dprintk("nfsd: Dropping request; may be revisited later\n");
627 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
628 return 0;
629 }
630
631 if (rqstp->rq_proc != 0)
632 *nfserrp++ = nfserr;
633
634 /* Encode result.
635 * For NFSv2, additional info is never returned in case of an error.
636 */
637 if (!(nfserr && rqstp->rq_vers == 2)) {
638 xdr = proc->pc_encode;
639 if (xdr && !xdr(rqstp, nfserrp,
640 rqstp->rq_resp)) {
641 /* Failed to encode result. Release cache entry */
642 dprintk("nfsd: failed to encode result!\n");
643 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
644 *statp = rpc_system_err;
645 return 1;
646 }
647 }
648
649 /* Store reply in cache. */
650 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
651 return 1;
652 }
653
654 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
655 {
656 int ret;
657 mutex_lock(&nfsd_mutex);
658 if (nfsd_serv == NULL) {
659 mutex_unlock(&nfsd_mutex);
660 return -ENODEV;
661 }
662 /* bump up the psudo refcount while traversing */
663 svc_get(nfsd_serv);
664 ret = svc_pool_stats_open(nfsd_serv, file);
665 mutex_unlock(&nfsd_mutex);
666 return ret;
667 }
668
669 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
670 {
671 int ret = seq_release(inode, file);
672 struct net *net = &init_net;
673
674 mutex_lock(&nfsd_mutex);
675 /* this function really, really should have been called svc_put() */
676 nfsd_destroy(net);
677 mutex_unlock(&nfsd_mutex);
678 return ret;
679 }
This page took 0.135695 seconds and 6 git commands to generate.