NFS: fix subtle change in COMMIT behavior
[linux/fpc-iii.git] / fs / nfsd / nfssvc.c
blob752d56bbe0ba47cda7ed87cb724130944879d269
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 goto dec_users;
226 ret = nfs4_state_start();
227 if (ret)
228 goto out_racache;
229 return 0;
231 out_racache:
232 nfsd_racache_shutdown();
233 dec_users:
234 nfsd_users--;
235 return ret;
238 static void nfsd_shutdown_generic(void)
240 if (--nfsd_users)
241 return;
243 nfs4_state_shutdown();
244 nfsd_racache_shutdown();
247 static bool nfsd_needs_lockd(void)
249 #if defined(CONFIG_NFSD_V3)
250 return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
251 #else
252 return (nfsd_versions[2] != NULL);
253 #endif
256 static int nfsd_startup_net(int nrservs, struct net *net)
258 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
259 int ret;
261 if (nn->nfsd_net_up)
262 return 0;
264 ret = nfsd_startup_generic(nrservs);
265 if (ret)
266 return ret;
267 ret = nfsd_init_socks(net);
268 if (ret)
269 goto out_socks;
271 if (nfsd_needs_lockd() && !nn->lockd_up) {
272 ret = lockd_up(net);
273 if (ret)
274 goto out_socks;
275 nn->lockd_up = 1;
278 ret = nfs4_state_start_net(net);
279 if (ret)
280 goto out_lockd;
282 nn->nfsd_net_up = true;
283 return 0;
285 out_lockd:
286 if (nn->lockd_up) {
287 lockd_down(net);
288 nn->lockd_up = 0;
290 out_socks:
291 nfsd_shutdown_generic();
292 return ret;
295 static void nfsd_shutdown_net(struct net *net)
297 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
299 nfs4_state_shutdown_net(net);
300 if (nn->lockd_up) {
301 lockd_down(net);
302 nn->lockd_up = 0;
304 nn->nfsd_net_up = false;
305 nfsd_shutdown_generic();
308 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
310 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
313 * write_ports can create the server without actually starting
314 * any threads--if we get shut down before any threads are
315 * started, then nfsd_last_thread will be run before any of this
316 * other initialization has been done.
318 if (!nn->nfsd_net_up)
319 return;
320 nfsd_shutdown_net(net);
322 svc_rpcb_cleanup(serv, net);
324 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
325 "cache\n");
326 nfsd_export_flush(net);
329 void nfsd_reset_versions(void)
331 int found_one = 0;
332 int i;
334 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
335 if (nfsd_program.pg_vers[i])
336 found_one = 1;
339 if (!found_one) {
340 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
341 nfsd_program.pg_vers[i] = nfsd_version[i];
342 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
343 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
344 nfsd_acl_program.pg_vers[i] =
345 nfsd_acl_version[i];
346 #endif
351 * Each session guarantees a negotiated per slot memory cache for replies
352 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
353 * NFSv4.1 server might want to use more memory for a DRC than a machine
354 * with mutiple services.
356 * Impose a hard limit on the number of pages for the DRC which varies
357 * according to the machines free pages. This is of course only a default.
359 * For now this is a #defined shift which could be under admin control
360 * in the future.
362 static void set_max_drc(void)
364 #define NFSD_DRC_SIZE_SHIFT 10
365 nfsd_drc_max_mem = (nr_free_buffer_pages()
366 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
367 nfsd_drc_mem_used = 0;
368 spin_lock_init(&nfsd_drc_lock);
369 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
372 static int nfsd_get_default_max_blksize(void)
374 struct sysinfo i;
375 unsigned long long target;
376 unsigned long ret;
378 si_meminfo(&i);
379 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
381 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
382 * machines, but only uses 32K on 128M machines. Bottom out at
383 * 8K on 32M and smaller. Of course, this is only a default.
385 target >>= 12;
387 ret = NFSSVC_MAXBLKSIZE;
388 while (ret > target && ret >= 8*1024*2)
389 ret /= 2;
390 return ret;
393 int nfsd_create_serv(struct net *net)
395 int error;
396 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
398 WARN_ON(!mutex_is_locked(&nfsd_mutex));
399 if (nn->nfsd_serv) {
400 svc_get(nn->nfsd_serv);
401 return 0;
403 if (nfsd_max_blksize == 0)
404 nfsd_max_blksize = nfsd_get_default_max_blksize();
405 nfsd_reset_versions();
406 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
407 nfsd_last_thread, nfsd, THIS_MODULE);
408 if (nn->nfsd_serv == NULL)
409 return -ENOMEM;
411 nn->nfsd_serv->sv_maxconn = nn->max_connections;
412 error = svc_bind(nn->nfsd_serv, net);
413 if (error < 0) {
414 svc_destroy(nn->nfsd_serv);
415 return error;
418 set_max_drc();
419 do_gettimeofday(&nn->nfssvc_boot); /* record boot time */
420 return 0;
423 int nfsd_nrpools(struct net *net)
425 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
427 if (nn->nfsd_serv == NULL)
428 return 0;
429 else
430 return nn->nfsd_serv->sv_nrpools;
433 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
435 int i = 0;
436 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
438 if (nn->nfsd_serv != NULL) {
439 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
440 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
443 return 0;
446 void nfsd_destroy(struct net *net)
448 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
449 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
451 if (destroy)
452 svc_shutdown_net(nn->nfsd_serv, net);
453 svc_destroy(nn->nfsd_serv);
454 if (destroy)
455 nn->nfsd_serv = NULL;
458 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
460 int i = 0;
461 int tot = 0;
462 int err = 0;
463 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
465 WARN_ON(!mutex_is_locked(&nfsd_mutex));
467 if (nn->nfsd_serv == NULL || n <= 0)
468 return 0;
470 if (n > nn->nfsd_serv->sv_nrpools)
471 n = nn->nfsd_serv->sv_nrpools;
473 /* enforce a global maximum number of threads */
474 tot = 0;
475 for (i = 0; i < n; i++) {
476 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
477 tot += nthreads[i];
479 if (tot > NFSD_MAXSERVS) {
480 /* total too large: scale down requested numbers */
481 for (i = 0; i < n && tot > 0; i++) {
482 int new = nthreads[i] * NFSD_MAXSERVS / tot;
483 tot -= (nthreads[i] - new);
484 nthreads[i] = new;
486 for (i = 0; i < n && tot > 0; i++) {
487 nthreads[i]--;
488 tot--;
493 * There must always be a thread in pool 0; the admin
494 * can't shut down NFS completely using pool_threads.
496 if (nthreads[0] == 0)
497 nthreads[0] = 1;
499 /* apply the new numbers */
500 svc_get(nn->nfsd_serv);
501 for (i = 0; i < n; i++) {
502 err = svc_set_num_threads(nn->nfsd_serv, &nn->nfsd_serv->sv_pools[i],
503 nthreads[i]);
504 if (err)
505 break;
507 nfsd_destroy(net);
508 return err;
512 * Adjust the number of threads and return the new number of threads.
513 * This is also the function that starts the server if necessary, if
514 * this is the first time nrservs is nonzero.
517 nfsd_svc(int nrservs, struct net *net)
519 int error;
520 bool nfsd_up_before;
521 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
523 mutex_lock(&nfsd_mutex);
524 dprintk("nfsd: creating service\n");
526 nrservs = max(nrservs, 0);
527 nrservs = min(nrservs, NFSD_MAXSERVS);
528 error = 0;
530 if (nrservs == 0 && nn->nfsd_serv == NULL)
531 goto out;
533 error = nfsd_create_serv(net);
534 if (error)
535 goto out;
537 nfsd_up_before = nn->nfsd_net_up;
539 error = nfsd_startup_net(nrservs, net);
540 if (error)
541 goto out_destroy;
542 error = svc_set_num_threads(nn->nfsd_serv, NULL, nrservs);
543 if (error)
544 goto out_shutdown;
545 /* We are holding a reference to nn->nfsd_serv which
546 * we don't want to count in the return value,
547 * so subtract 1
549 error = nn->nfsd_serv->sv_nrthreads - 1;
550 out_shutdown:
551 if (error < 0 && !nfsd_up_before)
552 nfsd_shutdown_net(net);
553 out_destroy:
554 nfsd_destroy(net); /* Release server */
555 out:
556 mutex_unlock(&nfsd_mutex);
557 return error;
562 * This is the NFS server kernel thread
564 static int
565 nfsd(void *vrqstp)
567 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
568 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
569 struct net *net = perm_sock->xpt_net;
570 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
571 int err;
573 /* Lock module and set up kernel thread */
574 mutex_lock(&nfsd_mutex);
576 /* At this point, the thread shares current->fs
577 * with the init process. We need to create files with a
578 * umask of 0 instead of init's umask. */
579 if (unshare_fs_struct() < 0) {
580 printk("Unable to start nfsd thread: out of memory\n");
581 goto out;
584 current->fs->umask = 0;
587 * thread is spawned with all signals set to SIG_IGN, re-enable
588 * the ones that will bring down the thread
590 allow_signal(SIGKILL);
591 allow_signal(SIGHUP);
592 allow_signal(SIGINT);
593 allow_signal(SIGQUIT);
595 nfsdstats.th_cnt++;
596 mutex_unlock(&nfsd_mutex);
598 set_freezable();
601 * The main request loop
603 for (;;) {
604 /* Update sv_maxconn if it has changed */
605 rqstp->rq_server->sv_maxconn = nn->max_connections;
608 * Find a socket with data available and call its
609 * recvfrom routine.
611 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
613 if (err == -EINTR)
614 break;
615 validate_process_creds();
616 svc_process(rqstp);
617 validate_process_creds();
620 /* Clear signals before calling svc_exit_thread() */
621 flush_signals(current);
623 mutex_lock(&nfsd_mutex);
624 nfsdstats.th_cnt --;
626 out:
627 rqstp->rq_server = NULL;
629 /* Release the thread */
630 svc_exit_thread(rqstp);
632 nfsd_destroy(net);
634 /* Release module */
635 mutex_unlock(&nfsd_mutex);
636 module_put_and_exit(0);
637 return 0;
640 static __be32 map_new_errors(u32 vers, __be32 nfserr)
642 if (nfserr == nfserr_jukebox && vers == 2)
643 return nfserr_dropit;
644 if (nfserr == nfserr_wrongsec && vers < 4)
645 return nfserr_acces;
646 return nfserr;
650 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
652 struct svc_procedure *proc;
653 kxdrproc_t xdr;
654 __be32 nfserr;
655 __be32 *nfserrp;
657 dprintk("nfsd_dispatch: vers %d proc %d\n",
658 rqstp->rq_vers, rqstp->rq_proc);
659 proc = rqstp->rq_procinfo;
662 * Give the xdr decoder a chance to change this if it wants
663 * (necessary in the NFSv4.0 compound case)
665 rqstp->rq_cachetype = proc->pc_cachetype;
666 /* Decode arguments */
667 xdr = proc->pc_decode;
668 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
669 rqstp->rq_argp)) {
670 dprintk("nfsd: failed to decode arguments!\n");
671 *statp = rpc_garbage_args;
672 return 1;
675 /* Check whether we have this call in the cache. */
676 switch (nfsd_cache_lookup(rqstp)) {
677 case RC_DROPIT:
678 return 0;
679 case RC_REPLY:
680 return 1;
681 case RC_DOIT:;
682 /* do it */
685 /* need to grab the location to store the status, as
686 * nfsv4 does some encoding while processing
688 nfserrp = rqstp->rq_res.head[0].iov_base
689 + rqstp->rq_res.head[0].iov_len;
690 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
692 /* Now call the procedure handler, and encode NFS status. */
693 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
694 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
695 if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
696 dprintk("nfsd: Dropping request; may be revisited later\n");
697 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
698 return 0;
701 if (rqstp->rq_proc != 0)
702 *nfserrp++ = nfserr;
704 /* Encode result.
705 * For NFSv2, additional info is never returned in case of an error.
707 if (!(nfserr && rqstp->rq_vers == 2)) {
708 xdr = proc->pc_encode;
709 if (xdr && !xdr(rqstp, nfserrp,
710 rqstp->rq_resp)) {
711 /* Failed to encode result. Release cache entry */
712 dprintk("nfsd: failed to encode result!\n");
713 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
714 *statp = rpc_system_err;
715 return 1;
719 /* Store reply in cache. */
720 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
721 return 1;
724 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
726 int ret;
727 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
729 mutex_lock(&nfsd_mutex);
730 if (nn->nfsd_serv == NULL) {
731 mutex_unlock(&nfsd_mutex);
732 return -ENODEV;
734 /* bump up the psudo refcount while traversing */
735 svc_get(nn->nfsd_serv);
736 ret = svc_pool_stats_open(nn->nfsd_serv, file);
737 mutex_unlock(&nfsd_mutex);
738 return ret;
741 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
743 int ret = seq_release(inode, file);
744 struct net *net = inode->i_sb->s_fs_info;
746 mutex_lock(&nfsd_mutex);
747 /* this function really, really should have been called svc_put() */
748 nfsd_destroy(net);
749 mutex_unlock(&nfsd_mutex);
750 return ret;