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>
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>
25 #include "filecache.h"
27 #define NFSDDBG_FACILITY NFSDDBG_SVC
29 extern struct svc_program nfsd_program;
30 static int nfsd(void *vrqstp);
33 * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members
34 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
35 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
37 * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a
38 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
39 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
40 * entry of ->sv_pools[].
42 * Transitions of the thread count between zero and non-zero are of particular
43 * interest since the svc_serv needs to be created and initialized at that
46 * Finally, the nfsd_mutex also protects some of the global variables that are
47 * accessed when nfsd starts and that are settable via the write_* routines in
48 * nfsctl.c. In particular:
50 * user_recovery_dirname
54 DEFINE_MUTEX(nfsd_mutex);
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.
62 spinlock_t nfsd_drc_lock;
63 unsigned long nfsd_drc_max_mem;
64 unsigned long nfsd_drc_mem_used;
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,
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];
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,
83 .pg_stats = &nfsd_acl_svcstats,
84 .pg_authenticate = &svc_set_client,
87 static struct svc_stat nfsd_acl_svcstats = {
88 .program = &nfsd_acl_program,
90 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
92 static struct svc_version * nfsd_version[] = {
94 #if defined(CONFIG_NFSD_V3)
97 #if defined(CONFIG_NFSD_V4)
102 #define NFSD_MINVERS 2
103 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
104 static struct svc_version *nfsd_versions[NFSD_NRVERS];
106 struct svc_program nfsd_program = {
107 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
108 .pg_next = &nfsd_acl_program,
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 */
120 static bool nfsd_supported_minorversions[NFSD_SUPPORTED_MINOR_VERSION + 1] = {
126 int nfsd_vers(int vers, enum vers_op change)
128 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
132 nfsd_versions[vers] = nfsd_version[vers];
133 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
134 if (vers < NFSD_ACL_NRVERS)
135 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
139 nfsd_versions[vers] = NULL;
140 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
141 if (vers < NFSD_ACL_NRVERS)
142 nfsd_acl_versions[vers] = NULL;
146 return nfsd_versions[vers] != NULL;
148 return nfsd_version[vers] != NULL;
153 int nfsd_minorversion(u32 minorversion, enum vers_op change)
155 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
159 nfsd_supported_minorversions[minorversion] = true;
162 nfsd_supported_minorversions[minorversion] = false;
165 return nfsd_supported_minorversions[minorversion];
167 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
173 * Maximum number of nfsd processes
175 #define NFSD_MAXSERVS 8192
177 int nfsd_nrthreads(struct net *net)
180 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
182 mutex_lock(&nfsd_mutex);
184 rv = nn->nfsd_serv->sv_nrthreads;
185 mutex_unlock(&nfsd_mutex);
189 static int nfsd_init_socks(struct net *net)
192 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
194 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
197 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
202 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
210 static int nfsd_users = 0;
212 static int nfsd_startup_generic(int nrservs)
219 ret = nfsd_file_cache_init();
223 ret = nfs4_state_start();
229 nfsd_file_cache_shutdown();
235 static void nfsd_shutdown_generic(void)
240 nfs4_state_shutdown();
241 nfsd_file_cache_shutdown();
244 static bool nfsd_needs_lockd(void)
246 #if defined(CONFIG_NFSD_V3)
247 return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
249 return (nfsd_versions[2] != NULL);
253 static int nfsd_startup_net(int nrservs, struct net *net)
255 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
261 ret = nfsd_startup_generic(nrservs);
264 ret = nfsd_init_socks(net);
268 if (nfsd_needs_lockd() && !nn->lockd_up) {
275 ret = nfs4_state_start_net(net);
279 nn->nfsd_net_up = true;
288 nfsd_shutdown_generic();
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);
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)
317 nfsd_shutdown_net(net);
319 svc_rpcb_cleanup(serv, net);
321 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
323 nfsd_export_flush(net);
326 void nfsd_reset_versions(void)
331 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
332 if (nfsd_program.pg_vers[i])
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] =
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
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)
372 unsigned long long target;
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.
384 ret = NFSSVC_MAXBLKSIZE;
385 while (ret > target && ret >= 8*1024*2)
390 static struct svc_serv_ops nfsd_thread_sv_ops = {
391 .svo_shutdown = nfsd_last_thread,
392 .svo_function = nfsd,
393 .svo_enqueue_xprt = svc_xprt_do_enqueue,
394 .svo_setup = svc_set_num_threads,
395 .svo_module = THIS_MODULE,
398 int nfsd_create_serv(struct net *net)
401 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
403 WARN_ON(!mutex_is_locked(&nfsd_mutex));
405 svc_get(nn->nfsd_serv);
408 if (nfsd_max_blksize == 0)
409 nfsd_max_blksize = nfsd_get_default_max_blksize();
410 nfsd_reset_versions();
411 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
412 &nfsd_thread_sv_ops);
413 if (nn->nfsd_serv == NULL)
416 nn->nfsd_serv->sv_maxconn = nn->max_connections;
417 error = svc_bind(nn->nfsd_serv, net);
419 svc_destroy(nn->nfsd_serv);
424 do_gettimeofday(&nn->nfssvc_boot); /* record boot time */
428 int nfsd_nrpools(struct net *net)
430 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
432 if (nn->nfsd_serv == NULL)
435 return nn->nfsd_serv->sv_nrpools;
438 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
441 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
443 if (nn->nfsd_serv != NULL) {
444 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
445 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
451 void nfsd_destroy(struct net *net)
453 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
454 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
457 svc_shutdown_net(nn->nfsd_serv, net);
458 svc_destroy(nn->nfsd_serv);
460 nn->nfsd_serv = NULL;
463 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
468 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
470 WARN_ON(!mutex_is_locked(&nfsd_mutex));
472 if (nn->nfsd_serv == NULL || n <= 0)
475 if (n > nn->nfsd_serv->sv_nrpools)
476 n = nn->nfsd_serv->sv_nrpools;
478 /* enforce a global maximum number of threads */
480 for (i = 0; i < n; i++) {
481 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
484 if (tot > NFSD_MAXSERVS) {
485 /* total too large: scale down requested numbers */
486 for (i = 0; i < n && tot > 0; i++) {
487 int new = nthreads[i] * NFSD_MAXSERVS / tot;
488 tot -= (nthreads[i] - new);
491 for (i = 0; i < n && tot > 0; i++) {
498 * There must always be a thread in pool 0; the admin
499 * can't shut down NFS completely using pool_threads.
501 if (nthreads[0] == 0)
504 /* apply the new numbers */
505 svc_get(nn->nfsd_serv);
506 for (i = 0; i < n; i++) {
507 err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
508 &nn->nfsd_serv->sv_pools[i], nthreads[i]);
517 * Adjust the number of threads and return the new number of threads.
518 * This is also the function that starts the server if necessary, if
519 * this is the first time nrservs is nonzero.
522 nfsd_svc(int nrservs, struct net *net)
526 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
528 mutex_lock(&nfsd_mutex);
529 dprintk("nfsd: creating service\n");
531 nrservs = max(nrservs, 0);
532 nrservs = min(nrservs, NFSD_MAXSERVS);
535 if (nrservs == 0 && nn->nfsd_serv == NULL)
538 error = nfsd_create_serv(net);
542 nfsd_up_before = nn->nfsd_net_up;
544 error = nfsd_startup_net(nrservs, net);
547 error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
551 /* We are holding a reference to nn->nfsd_serv which
552 * we don't want to count in the return value,
555 error = nn->nfsd_serv->sv_nrthreads - 1;
557 if (error < 0 && !nfsd_up_before)
558 nfsd_shutdown_net(net);
560 nfsd_destroy(net); /* Release server */
562 mutex_unlock(&nfsd_mutex);
568 * This is the NFS server kernel thread
573 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
574 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
575 struct net *net = perm_sock->xpt_net;
576 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
579 /* Lock module and set up kernel thread */
580 mutex_lock(&nfsd_mutex);
582 /* At this point, the thread shares current->fs
583 * with the init process. We need to create files with a
584 * umask of 0 instead of init's umask. */
585 if (unshare_fs_struct() < 0) {
586 printk("Unable to start nfsd thread: out of memory\n");
590 current->fs->umask = 0;
593 * thread is spawned with all signals set to SIG_IGN, re-enable
594 * the ones that will bring down the thread
596 allow_signal(SIGKILL);
597 allow_signal(SIGHUP);
598 allow_signal(SIGINT);
599 allow_signal(SIGQUIT);
602 mutex_unlock(&nfsd_mutex);
607 * The main request loop
610 /* Update sv_maxconn if it has changed */
611 rqstp->rq_server->sv_maxconn = nn->max_connections;
614 * Find a socket with data available and call its
617 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
621 validate_process_creds();
623 validate_process_creds();
626 /* Clear signals before calling svc_exit_thread() */
627 flush_signals(current);
629 mutex_lock(&nfsd_mutex);
633 rqstp->rq_server = NULL;
635 /* Release the thread */
636 svc_exit_thread(rqstp);
641 mutex_unlock(&nfsd_mutex);
642 module_put_and_exit(0);
646 static __be32 map_new_errors(u32 vers, __be32 nfserr)
648 if (nfserr == nfserr_jukebox && vers == 2)
649 return nfserr_dropit;
650 if (nfserr == nfserr_wrongsec && vers < 4)
656 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
658 struct svc_procedure *proc;
663 dprintk("nfsd_dispatch: vers %d proc %d\n",
664 rqstp->rq_vers, rqstp->rq_proc);
665 proc = rqstp->rq_procinfo;
668 * Give the xdr decoder a chance to change this if it wants
669 * (necessary in the NFSv4.0 compound case)
671 rqstp->rq_cachetype = proc->pc_cachetype;
672 /* Decode arguments */
673 xdr = proc->pc_decode;
674 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
676 dprintk("nfsd: failed to decode arguments!\n");
677 *statp = rpc_garbage_args;
681 /* Check whether we have this call in the cache. */
682 switch (nfsd_cache_lookup(rqstp)) {
691 /* need to grab the location to store the status, as
692 * nfsv4 does some encoding while processing
694 nfserrp = rqstp->rq_res.head[0].iov_base
695 + rqstp->rq_res.head[0].iov_len;
696 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
698 /* Now call the procedure handler, and encode NFS status. */
699 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
700 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
701 if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
702 dprintk("nfsd: Dropping request; may be revisited later\n");
703 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
707 if (rqstp->rq_proc != 0)
711 * For NFSv2, additional info is never returned in case of an error.
713 if (!(nfserr && rqstp->rq_vers == 2)) {
714 xdr = proc->pc_encode;
715 if (xdr && !xdr(rqstp, nfserrp,
717 /* Failed to encode result. Release cache entry */
718 dprintk("nfsd: failed to encode result!\n");
719 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
720 *statp = rpc_system_err;
725 /* Store reply in cache. */
726 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
730 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
733 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
735 mutex_lock(&nfsd_mutex);
736 if (nn->nfsd_serv == NULL) {
737 mutex_unlock(&nfsd_mutex);
740 /* bump up the psudo refcount while traversing */
741 svc_get(nn->nfsd_serv);
742 ret = svc_pool_stats_open(nn->nfsd_serv, file);
743 mutex_unlock(&nfsd_mutex);
747 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
749 int ret = seq_release(inode, file);
750 struct net *net = inode->i_sb->s_fs_info;
752 mutex_lock(&nfsd_mutex);
753 /* this function really, really should have been called svc_put() */
755 mutex_unlock(&nfsd_mutex);