Linux 3.16-rc2
[linux/fpc-iii.git] / fs / nfsd / nfssvc.c
blob1879e43f2868c7ffbf21405f41412d2f540f19f7
1 /*
2 * Central processing for nfsd.
4 * Authors: Olaf Kirch (okir@monad.swb.de)
6 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
7 */
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>
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 #include "netns.h"
26 #define NFSDDBG_FACILITY NFSDDBG_SVC
28 extern struct svc_program nfsd_program;
29 static int nfsd(void *vrqstp);
32 * nfsd_mutex protects nn->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
36 * If (out side the lock) nn->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[].
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.
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:
49 * user_recovery_dirname
50 * user_lease_time
51 * nfsd_versions
53 DEFINE_MUTEX(nfsd_mutex);
56 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
57 * nfsd_drc_max_pages limits the total amount of memory available for
58 * version 4.1 DRC caches.
59 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
61 spinlock_t nfsd_drc_lock;
62 unsigned long nfsd_drc_max_mem;
63 unsigned long nfsd_drc_mem_used;
65 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
66 static struct svc_stat nfsd_acl_svcstats;
67 static struct svc_version * nfsd_acl_version[] = {
68 [2] = &nfsd_acl_version2,
69 [3] = &nfsd_acl_version3,
72 #define NFSD_ACL_MINVERS 2
73 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
74 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
76 static struct svc_program nfsd_acl_program = {
77 .pg_prog = NFS_ACL_PROGRAM,
78 .pg_nvers = NFSD_ACL_NRVERS,
79 .pg_vers = nfsd_acl_versions,
80 .pg_name = "nfsacl",
81 .pg_class = "nfsd",
82 .pg_stats = &nfsd_acl_svcstats,
83 .pg_authenticate = &svc_set_client,
86 static struct svc_stat nfsd_acl_svcstats = {
87 .program = &nfsd_acl_program,
89 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
91 static struct svc_version * nfsd_version[] = {
92 [2] = &nfsd_version2,
93 #if defined(CONFIG_NFSD_V3)
94 [3] = &nfsd_version3,
95 #endif
96 #if defined(CONFIG_NFSD_V4)
97 [4] = &nfsd_version4,
98 #endif
101 #define NFSD_MINVERS 2
102 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
103 static struct svc_version *nfsd_versions[NFSD_NRVERS];
105 struct svc_program nfsd_program = {
106 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
107 .pg_next = &nfsd_acl_program,
108 #endif
109 .pg_prog = NFS_PROGRAM, /* program number */
110 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
111 .pg_vers = nfsd_versions, /* version table */
112 .pg_name = "nfsd", /* program name */
113 .pg_class = "nfsd", /* authentication class */
114 .pg_stats = &nfsd_svcstats, /* version table */
115 .pg_authenticate = &svc_set_client, /* export authentication */
119 static bool nfsd_supported_minorversions[NFSD_SUPPORTED_MINOR_VERSION + 1] = {
120 [0] = 1,
121 [1] = 1,
124 int nfsd_vers(int vers, enum vers_op change)
126 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
127 return 0;
128 switch(change) {
129 case NFSD_SET:
130 nfsd_versions[vers] = nfsd_version[vers];
131 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
132 if (vers < NFSD_ACL_NRVERS)
133 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
134 #endif
135 break;
136 case NFSD_CLEAR:
137 nfsd_versions[vers] = NULL;
138 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
139 if (vers < NFSD_ACL_NRVERS)
140 nfsd_acl_versions[vers] = NULL;
141 #endif
142 break;
143 case NFSD_TEST:
144 return nfsd_versions[vers] != NULL;
145 case NFSD_AVAIL:
146 return nfsd_version[vers] != NULL;
148 return 0;
151 int nfsd_minorversion(u32 minorversion, enum vers_op change)
153 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
154 return -1;
155 switch(change) {
156 case NFSD_SET:
157 nfsd_supported_minorversions[minorversion] = true;
158 break;
159 case NFSD_CLEAR:
160 nfsd_supported_minorversions[minorversion] = false;
161 break;
162 case NFSD_TEST:
163 return nfsd_supported_minorversions[minorversion];
164 case NFSD_AVAIL:
165 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
167 return 0;
171 * Maximum number of nfsd processes
173 #define NFSD_MAXSERVS 8192
175 int nfsd_nrthreads(struct net *net)
177 int rv = 0;
178 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
180 mutex_lock(&nfsd_mutex);
181 if (nn->nfsd_serv)
182 rv = nn->nfsd_serv->sv_nrthreads;
183 mutex_unlock(&nfsd_mutex);
184 return rv;
187 static int nfsd_init_socks(struct net *net)
189 int error;
190 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
192 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
193 return 0;
195 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
196 SVC_SOCK_DEFAULTS);
197 if (error < 0)
198 return error;
200 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
201 SVC_SOCK_DEFAULTS);
202 if (error < 0)
203 return error;
205 return 0;
208 static int nfsd_users = 0;
210 static int nfsd_startup_generic(int nrservs)
212 int ret;
214 if (nfsd_users++)
215 return 0;
218 * Readahead param cache - will no-op if it already exists.
219 * (Note therefore results will be suboptimal if number of
220 * threads is modified after nfsd start.)
222 ret = nfsd_racache_init(2*nrservs);
223 if (ret)
224 return ret;
225 ret = nfs4_state_start();
226 if (ret)
227 goto out_racache;
228 return 0;
230 out_racache:
231 nfsd_racache_shutdown();
232 return ret;
235 static void nfsd_shutdown_generic(void)
237 if (--nfsd_users)
238 return;
240 nfs4_state_shutdown();
241 nfsd_racache_shutdown();
244 static bool nfsd_needs_lockd(void)
246 #if defined(CONFIG_NFSD_V3)
247 return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
248 #else
249 return (nfsd_versions[2] != NULL);
250 #endif
253 static int nfsd_startup_net(int nrservs, struct net *net)
255 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
256 int ret;
258 if (nn->nfsd_net_up)
259 return 0;
261 ret = nfsd_startup_generic(nrservs);
262 if (ret)
263 return ret;
264 ret = nfsd_init_socks(net);
265 if (ret)
266 goto out_socks;
268 if (nfsd_needs_lockd() && !nn->lockd_up) {
269 ret = lockd_up(net);
270 if (ret)
271 goto out_socks;
272 nn->lockd_up = 1;
275 ret = nfs4_state_start_net(net);
276 if (ret)
277 goto out_lockd;
279 nn->nfsd_net_up = true;
280 return 0;
282 out_lockd:
283 if (nn->lockd_up) {
284 lockd_down(net);
285 nn->lockd_up = 0;
287 out_socks:
288 nfsd_shutdown_generic();
289 return ret;
292 static void nfsd_shutdown_net(struct net *net)
294 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
296 nfs4_state_shutdown_net(net);
297 if (nn->lockd_up) {
298 lockd_down(net);
299 nn->lockd_up = 0;
301 nn->nfsd_net_up = false;
302 nfsd_shutdown_generic();
305 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
307 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
310 * write_ports can create the server without actually starting
311 * any threads--if we get shut down before any threads are
312 * started, then nfsd_last_thread will be run before any of this
313 * other initialization has been done.
315 if (!nn->nfsd_net_up)
316 return;
317 nfsd_shutdown_net(net);
319 svc_rpcb_cleanup(serv, net);
321 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
322 "cache\n");
323 nfsd_export_flush(net);
326 void nfsd_reset_versions(void)
328 int found_one = 0;
329 int i;
331 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
332 if (nfsd_program.pg_vers[i])
333 found_one = 1;
336 if (!found_one) {
337 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
338 nfsd_program.pg_vers[i] = nfsd_version[i];
339 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
340 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
341 nfsd_acl_program.pg_vers[i] =
342 nfsd_acl_version[i];
343 #endif
348 * Each session guarantees a negotiated per slot memory cache for replies
349 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
350 * NFSv4.1 server might want to use more memory for a DRC than a machine
351 * with mutiple services.
353 * Impose a hard limit on the number of pages for the DRC which varies
354 * according to the machines free pages. This is of course only a default.
356 * For now this is a #defined shift which could be under admin control
357 * in the future.
359 static void set_max_drc(void)
361 #define NFSD_DRC_SIZE_SHIFT 10
362 nfsd_drc_max_mem = (nr_free_buffer_pages()
363 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
364 nfsd_drc_mem_used = 0;
365 spin_lock_init(&nfsd_drc_lock);
366 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
369 static int nfsd_get_default_max_blksize(void)
371 struct sysinfo i;
372 unsigned long long target;
373 unsigned long ret;
375 si_meminfo(&i);
376 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
378 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
379 * machines, but only uses 32K on 128M machines. Bottom out at
380 * 8K on 32M and smaller. Of course, this is only a default.
382 target >>= 12;
384 ret = NFSSVC_MAXBLKSIZE;
385 while (ret > target && ret >= 8*1024*2)
386 ret /= 2;
387 return ret;
390 int nfsd_create_serv(struct net *net)
392 int error;
393 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
395 WARN_ON(!mutex_is_locked(&nfsd_mutex));
396 if (nn->nfsd_serv) {
397 svc_get(nn->nfsd_serv);
398 return 0;
400 if (nfsd_max_blksize == 0)
401 nfsd_max_blksize = nfsd_get_default_max_blksize();
402 nfsd_reset_versions();
403 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
404 nfsd_last_thread, nfsd, THIS_MODULE);
405 if (nn->nfsd_serv == NULL)
406 return -ENOMEM;
408 error = svc_bind(nn->nfsd_serv, net);
409 if (error < 0) {
410 svc_destroy(nn->nfsd_serv);
411 return error;
414 set_max_drc();
415 do_gettimeofday(&nn->nfssvc_boot); /* record boot time */
416 return 0;
419 int nfsd_nrpools(struct net *net)
421 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
423 if (nn->nfsd_serv == NULL)
424 return 0;
425 else
426 return nn->nfsd_serv->sv_nrpools;
429 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
431 int i = 0;
432 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
434 if (nn->nfsd_serv != NULL) {
435 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
436 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
439 return 0;
442 void nfsd_destroy(struct net *net)
444 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
445 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
447 if (destroy)
448 svc_shutdown_net(nn->nfsd_serv, net);
449 svc_destroy(nn->nfsd_serv);
450 if (destroy)
451 nn->nfsd_serv = NULL;
454 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
456 int i = 0;
457 int tot = 0;
458 int err = 0;
459 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
461 WARN_ON(!mutex_is_locked(&nfsd_mutex));
463 if (nn->nfsd_serv == NULL || n <= 0)
464 return 0;
466 if (n > nn->nfsd_serv->sv_nrpools)
467 n = nn->nfsd_serv->sv_nrpools;
469 /* enforce a global maximum number of threads */
470 tot = 0;
471 for (i = 0; i < n; i++) {
472 if (nthreads[i] > NFSD_MAXSERVS)
473 nthreads[i] = NFSD_MAXSERVS;
474 tot += nthreads[i];
476 if (tot > NFSD_MAXSERVS) {
477 /* total too large: scale down requested numbers */
478 for (i = 0; i < n && tot > 0; i++) {
479 int new = nthreads[i] * NFSD_MAXSERVS / tot;
480 tot -= (nthreads[i] - new);
481 nthreads[i] = new;
483 for (i = 0; i < n && tot > 0; i++) {
484 nthreads[i]--;
485 tot--;
490 * There must always be a thread in pool 0; the admin
491 * can't shut down NFS completely using pool_threads.
493 if (nthreads[0] == 0)
494 nthreads[0] = 1;
496 /* apply the new numbers */
497 svc_get(nn->nfsd_serv);
498 for (i = 0; i < n; i++) {
499 err = svc_set_num_threads(nn->nfsd_serv, &nn->nfsd_serv->sv_pools[i],
500 nthreads[i]);
501 if (err)
502 break;
504 nfsd_destroy(net);
505 return err;
509 * Adjust the number of threads and return the new number of threads.
510 * This is also the function that starts the server if necessary, if
511 * this is the first time nrservs is nonzero.
514 nfsd_svc(int nrservs, struct net *net)
516 int error;
517 bool nfsd_up_before;
518 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
520 mutex_lock(&nfsd_mutex);
521 dprintk("nfsd: creating service\n");
522 if (nrservs <= 0)
523 nrservs = 0;
524 if (nrservs > NFSD_MAXSERVS)
525 nrservs = NFSD_MAXSERVS;
526 error = 0;
527 if (nrservs == 0 && nn->nfsd_serv == NULL)
528 goto out;
530 error = nfsd_create_serv(net);
531 if (error)
532 goto out;
534 nfsd_up_before = nn->nfsd_net_up;
536 error = nfsd_startup_net(nrservs, net);
537 if (error)
538 goto out_destroy;
539 error = svc_set_num_threads(nn->nfsd_serv, NULL, nrservs);
540 if (error)
541 goto out_shutdown;
542 /* We are holding a reference to nn->nfsd_serv which
543 * we don't want to count in the return value,
544 * so subtract 1
546 error = nn->nfsd_serv->sv_nrthreads - 1;
547 out_shutdown:
548 if (error < 0 && !nfsd_up_before)
549 nfsd_shutdown_net(net);
550 out_destroy:
551 nfsd_destroy(net); /* Release server */
552 out:
553 mutex_unlock(&nfsd_mutex);
554 return error;
559 * This is the NFS server kernel thread
561 static int
562 nfsd(void *vrqstp)
564 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
565 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
566 struct net *net = perm_sock->xpt_net;
567 int err;
569 /* Lock module and set up kernel thread */
570 mutex_lock(&nfsd_mutex);
572 /* At this point, the thread shares current->fs
573 * with the init process. We need to create files with a
574 * umask of 0 instead of init's umask. */
575 if (unshare_fs_struct() < 0) {
576 printk("Unable to start nfsd thread: out of memory\n");
577 goto out;
580 current->fs->umask = 0;
583 * thread is spawned with all signals set to SIG_IGN, re-enable
584 * the ones that will bring down the thread
586 allow_signal(SIGKILL);
587 allow_signal(SIGHUP);
588 allow_signal(SIGINT);
589 allow_signal(SIGQUIT);
591 nfsdstats.th_cnt++;
592 mutex_unlock(&nfsd_mutex);
594 set_freezable();
597 * The main request loop
599 for (;;) {
601 * Find a socket with data available and call its
602 * recvfrom routine.
604 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
606 if (err == -EINTR)
607 break;
608 validate_process_creds();
609 svc_process(rqstp);
610 validate_process_creds();
613 /* Clear signals before calling svc_exit_thread() */
614 flush_signals(current);
616 mutex_lock(&nfsd_mutex);
617 nfsdstats.th_cnt --;
619 out:
620 rqstp->rq_server = NULL;
622 /* Release the thread */
623 svc_exit_thread(rqstp);
625 nfsd_destroy(net);
627 /* Release module */
628 mutex_unlock(&nfsd_mutex);
629 module_put_and_exit(0);
630 return 0;
633 static __be32 map_new_errors(u32 vers, __be32 nfserr)
635 if (nfserr == nfserr_jukebox && vers == 2)
636 return nfserr_dropit;
637 if (nfserr == nfserr_wrongsec && vers < 4)
638 return nfserr_acces;
639 return nfserr;
643 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
645 struct svc_procedure *proc;
646 kxdrproc_t xdr;
647 __be32 nfserr;
648 __be32 *nfserrp;
650 dprintk("nfsd_dispatch: vers %d proc %d\n",
651 rqstp->rq_vers, rqstp->rq_proc);
652 proc = rqstp->rq_procinfo;
655 * Give the xdr decoder a chance to change this if it wants
656 * (necessary in the NFSv4.0 compound case)
658 rqstp->rq_cachetype = proc->pc_cachetype;
659 /* Decode arguments */
660 xdr = proc->pc_decode;
661 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
662 rqstp->rq_argp)) {
663 dprintk("nfsd: failed to decode arguments!\n");
664 *statp = rpc_garbage_args;
665 return 1;
668 /* Check whether we have this call in the cache. */
669 switch (nfsd_cache_lookup(rqstp)) {
670 case RC_DROPIT:
671 return 0;
672 case RC_REPLY:
673 return 1;
674 case RC_DOIT:;
675 /* do it */
678 /* need to grab the location to store the status, as
679 * nfsv4 does some encoding while processing
681 nfserrp = rqstp->rq_res.head[0].iov_base
682 + rqstp->rq_res.head[0].iov_len;
683 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
685 /* Now call the procedure handler, and encode NFS status. */
686 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
687 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
688 if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
689 dprintk("nfsd: Dropping request; may be revisited later\n");
690 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
691 return 0;
694 if (rqstp->rq_proc != 0)
695 *nfserrp++ = nfserr;
697 /* Encode result.
698 * For NFSv2, additional info is never returned in case of an error.
700 if (!(nfserr && rqstp->rq_vers == 2)) {
701 xdr = proc->pc_encode;
702 if (xdr && !xdr(rqstp, nfserrp,
703 rqstp->rq_resp)) {
704 /* Failed to encode result. Release cache entry */
705 dprintk("nfsd: failed to encode result!\n");
706 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
707 *statp = rpc_system_err;
708 return 1;
712 /* Store reply in cache. */
713 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
714 return 1;
717 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
719 int ret;
720 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
722 mutex_lock(&nfsd_mutex);
723 if (nn->nfsd_serv == NULL) {
724 mutex_unlock(&nfsd_mutex);
725 return -ENODEV;
727 /* bump up the psudo refcount while traversing */
728 svc_get(nn->nfsd_serv);
729 ret = svc_pool_stats_open(nn->nfsd_serv, file);
730 mutex_unlock(&nfsd_mutex);
731 return ret;
734 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
736 int ret = seq_release(inode, file);
737 struct net *net = inode->i_sb->s_fs_info;
739 mutex_lock(&nfsd_mutex);
740 /* this function really, really should have been called svc_put() */
741 nfsd_destroy(net);
742 mutex_unlock(&nfsd_mutex);
743 return ret;