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>
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
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
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,
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[] = {
93 #if defined(CONFIG_NFSD_V3)
96 #if defined(CONFIG_NFSD_V4)
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,
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] = {
124 int nfsd_vers(int vers, enum vers_op change)
126 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
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];
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;
144 return nfsd_versions[vers] != NULL;
146 return nfsd_version[vers] != NULL;
151 int nfsd_minorversion(u32 minorversion, enum vers_op change)
153 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
157 nfsd_supported_minorversions[minorversion] = true;
160 nfsd_supported_minorversions[minorversion] = false;
163 return nfsd_supported_minorversions[minorversion];
165 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
171 * Maximum number of nfsd processes
173 #define NFSD_MAXSERVS 8192
175 int nfsd_nrthreads(struct net *net)
178 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
180 mutex_lock(&nfsd_mutex);
182 rv = nn->nfsd_serv->sv_nrthreads;
183 mutex_unlock(&nfsd_mutex);
187 static int nfsd_init_socks(struct net *net)
190 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
192 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
195 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
200 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
208 static int nfsd_users = 0;
210 static int nfsd_startup_generic(int nrservs)
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);
225 ret = nfs4_state_start();
231 nfsd_racache_shutdown();
235 static void nfsd_shutdown_generic(void)
240 nfs4_state_shutdown();
241 nfsd_racache_shutdown();
244 static bool nfsd_needs_lockd(void)
246 return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
249 static int nfsd_startup_net(int nrservs, struct net *net)
251 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
257 ret = nfsd_startup_generic(nrservs);
260 ret = nfsd_init_socks(net);
264 if (nfsd_needs_lockd() && !nn->lockd_up) {
271 ret = nfs4_state_start_net(net);
275 nn->nfsd_net_up = true;
284 nfsd_shutdown_generic();
288 static void nfsd_shutdown_net(struct net *net)
290 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
292 nfs4_state_shutdown_net(net);
297 nn->nfsd_net_up = false;
298 nfsd_shutdown_generic();
301 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
303 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
306 * write_ports can create the server without actually starting
307 * any threads--if we get shut down before any threads are
308 * started, then nfsd_last_thread will be run before any of this
309 * other initialization has been done.
311 if (!nn->nfsd_net_up)
313 nfsd_shutdown_net(net);
315 svc_rpcb_cleanup(serv, net);
317 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
319 nfsd_export_flush(net);
322 void nfsd_reset_versions(void)
327 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
328 if (nfsd_program.pg_vers[i])
333 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
334 nfsd_program.pg_vers[i] = nfsd_version[i];
335 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
336 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
337 nfsd_acl_program.pg_vers[i] =
344 * Each session guarantees a negotiated per slot memory cache for replies
345 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
346 * NFSv4.1 server might want to use more memory for a DRC than a machine
347 * with mutiple services.
349 * Impose a hard limit on the number of pages for the DRC which varies
350 * according to the machines free pages. This is of course only a default.
352 * For now this is a #defined shift which could be under admin control
355 static void set_max_drc(void)
357 #define NFSD_DRC_SIZE_SHIFT 10
358 nfsd_drc_max_mem = (nr_free_buffer_pages()
359 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
360 nfsd_drc_mem_used = 0;
361 spin_lock_init(&nfsd_drc_lock);
362 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
365 static int nfsd_get_default_max_blksize(void)
368 unsigned long long target;
372 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
374 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
375 * machines, but only uses 32K on 128M machines. Bottom out at
376 * 8K on 32M and smaller. Of course, this is only a default.
380 ret = NFSSVC_MAXBLKSIZE;
381 while (ret > target && ret >= 8*1024*2)
386 int nfsd_create_serv(struct net *net)
389 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
391 WARN_ON(!mutex_is_locked(&nfsd_mutex));
393 svc_get(nn->nfsd_serv);
396 if (nfsd_max_blksize == 0)
397 nfsd_max_blksize = nfsd_get_default_max_blksize();
398 nfsd_reset_versions();
399 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
400 nfsd_last_thread, nfsd, THIS_MODULE);
401 if (nn->nfsd_serv == NULL)
404 error = svc_bind(nn->nfsd_serv, net);
406 svc_destroy(nn->nfsd_serv);
411 do_gettimeofday(&nn->nfssvc_boot); /* record boot time */
415 int nfsd_nrpools(struct net *net)
417 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
419 if (nn->nfsd_serv == NULL)
422 return nn->nfsd_serv->sv_nrpools;
425 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
428 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
430 if (nn->nfsd_serv != NULL) {
431 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
432 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
438 void nfsd_destroy(struct net *net)
440 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
441 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
444 svc_shutdown_net(nn->nfsd_serv, net);
445 svc_destroy(nn->nfsd_serv);
447 nn->nfsd_serv = NULL;
450 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
455 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
457 WARN_ON(!mutex_is_locked(&nfsd_mutex));
459 if (nn->nfsd_serv == NULL || n <= 0)
462 if (n > nn->nfsd_serv->sv_nrpools)
463 n = nn->nfsd_serv->sv_nrpools;
465 /* enforce a global maximum number of threads */
467 for (i = 0; i < n; i++) {
468 if (nthreads[i] > NFSD_MAXSERVS)
469 nthreads[i] = NFSD_MAXSERVS;
472 if (tot > NFSD_MAXSERVS) {
473 /* total too large: scale down requested numbers */
474 for (i = 0; i < n && tot > 0; i++) {
475 int new = nthreads[i] * NFSD_MAXSERVS / tot;
476 tot -= (nthreads[i] - new);
479 for (i = 0; i < n && tot > 0; i++) {
486 * There must always be a thread in pool 0; the admin
487 * can't shut down NFS completely using pool_threads.
489 if (nthreads[0] == 0)
492 /* apply the new numbers */
493 svc_get(nn->nfsd_serv);
494 for (i = 0; i < n; i++) {
495 err = svc_set_num_threads(nn->nfsd_serv, &nn->nfsd_serv->sv_pools[i],
505 * Adjust the number of threads and return the new number of threads.
506 * This is also the function that starts the server if necessary, if
507 * this is the first time nrservs is nonzero.
510 nfsd_svc(int nrservs, struct net *net)
514 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
516 mutex_lock(&nfsd_mutex);
517 dprintk("nfsd: creating service\n");
520 if (nrservs > NFSD_MAXSERVS)
521 nrservs = NFSD_MAXSERVS;
523 if (nrservs == 0 && nn->nfsd_serv == NULL)
526 error = nfsd_create_serv(net);
530 nfsd_up_before = nn->nfsd_net_up;
532 error = nfsd_startup_net(nrservs, net);
535 error = svc_set_num_threads(nn->nfsd_serv, NULL, nrservs);
538 /* We are holding a reference to nn->nfsd_serv which
539 * we don't want to count in the return value,
542 error = nn->nfsd_serv->sv_nrthreads - 1;
544 if (error < 0 && !nfsd_up_before)
545 nfsd_shutdown_net(net);
547 nfsd_destroy(net); /* Release server */
549 mutex_unlock(&nfsd_mutex);
555 * This is the NFS server kernel thread
560 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
561 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
562 struct net *net = perm_sock->xpt_net;
565 /* Lock module and set up kernel thread */
566 mutex_lock(&nfsd_mutex);
568 /* At this point, the thread shares current->fs
569 * with the init process. We need to create files with a
570 * umask of 0 instead of init's umask. */
571 if (unshare_fs_struct() < 0) {
572 printk("Unable to start nfsd thread: out of memory\n");
576 current->fs->umask = 0;
579 * thread is spawned with all signals set to SIG_IGN, re-enable
580 * the ones that will bring down the thread
582 allow_signal(SIGKILL);
583 allow_signal(SIGHUP);
584 allow_signal(SIGINT);
585 allow_signal(SIGQUIT);
588 mutex_unlock(&nfsd_mutex);
591 * We want less throttling in balance_dirty_pages() so that nfs to
592 * localhost doesn't cause nfsd to lock up due to all the client's
595 current->flags |= PF_LESS_THROTTLE;
599 * The main request loop
603 * Find a socket with data available and call its
606 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
610 validate_process_creds();
612 validate_process_creds();
615 /* Clear signals before calling svc_exit_thread() */
616 flush_signals(current);
618 mutex_lock(&nfsd_mutex);
622 rqstp->rq_server = NULL;
624 /* Release the thread */
625 svc_exit_thread(rqstp);
630 mutex_unlock(&nfsd_mutex);
631 module_put_and_exit(0);
635 static __be32 map_new_errors(u32 vers, __be32 nfserr)
637 if (nfserr == nfserr_jukebox && vers == 2)
638 return nfserr_dropit;
639 if (nfserr == nfserr_wrongsec && vers < 4)
645 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
647 struct svc_procedure *proc;
652 dprintk("nfsd_dispatch: vers %d proc %d\n",
653 rqstp->rq_vers, rqstp->rq_proc);
654 proc = rqstp->rq_procinfo;
657 * Give the xdr decoder a chance to change this if it wants
658 * (necessary in the NFSv4.0 compound case)
660 rqstp->rq_cachetype = proc->pc_cachetype;
661 /* Decode arguments */
662 xdr = proc->pc_decode;
663 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
665 dprintk("nfsd: failed to decode arguments!\n");
666 *statp = rpc_garbage_args;
670 /* Check whether we have this call in the cache. */
671 switch (nfsd_cache_lookup(rqstp)) {
680 /* need to grab the location to store the status, as
681 * nfsv4 does some encoding while processing
683 nfserrp = rqstp->rq_res.head[0].iov_base
684 + rqstp->rq_res.head[0].iov_len;
685 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
687 /* Now call the procedure handler, and encode NFS status. */
688 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
689 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
690 if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
691 dprintk("nfsd: Dropping request; may be revisited later\n");
692 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
696 if (rqstp->rq_proc != 0)
700 * For NFSv2, additional info is never returned in case of an error.
702 if (!(nfserr && rqstp->rq_vers == 2)) {
703 xdr = proc->pc_encode;
704 if (xdr && !xdr(rqstp, nfserrp,
706 /* Failed to encode result. Release cache entry */
707 dprintk("nfsd: failed to encode result!\n");
708 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
709 *statp = rpc_system_err;
714 /* Store reply in cache. */
715 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
719 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
722 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
724 mutex_lock(&nfsd_mutex);
725 if (nn->nfsd_serv == NULL) {
726 mutex_unlock(&nfsd_mutex);
729 /* bump up the psudo refcount while traversing */
730 svc_get(nn->nfsd_serv);
731 ret = svc_pool_stats_open(nn->nfsd_serv, file);
732 mutex_unlock(&nfsd_mutex);
736 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
738 int ret = seq_release(inode, file);
739 struct net *net = inode->i_sb->s_fs_info;
741 mutex_lock(&nfsd_mutex);
742 /* this function really, really should have been called svc_put() */
744 mutex_unlock(&nfsd_mutex);