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 u32 nfsd_supported_minorversion;
121 int nfsd_vers(int vers, enum vers_op change)
123 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
127 nfsd_versions[vers] = nfsd_version[vers];
128 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
129 if (vers < NFSD_ACL_NRVERS)
130 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
134 nfsd_versions[vers] = NULL;
135 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
136 if (vers < NFSD_ACL_NRVERS)
137 nfsd_acl_versions[vers] = NULL;
141 return nfsd_versions[vers] != NULL;
143 return nfsd_version[vers] != NULL;
148 int nfsd_minorversion(u32 minorversion, enum vers_op change)
150 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
154 nfsd_supported_minorversion = minorversion;
157 if (minorversion == 0)
159 nfsd_supported_minorversion = minorversion - 1;
162 return minorversion <= nfsd_supported_minorversion;
164 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
170 * Maximum number of nfsd processes
172 #define NFSD_MAXSERVS 8192
174 int nfsd_nrthreads(struct net *net)
177 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
179 mutex_lock(&nfsd_mutex);
181 rv = nn->nfsd_serv->sv_nrthreads;
182 mutex_unlock(&nfsd_mutex);
186 static int nfsd_init_socks(struct net *net)
189 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
191 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
194 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
199 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
207 static int nfsd_users = 0;
209 static int nfsd_startup_generic(int nrservs)
217 * Readahead param cache - will no-op if it already exists.
218 * (Note therefore results will be suboptimal if number of
219 * threads is modified after nfsd start.)
221 ret = nfsd_racache_init(2*nrservs);
225 ret = nfs4_state_start();
231 nfsd_racache_shutdown();
237 static void nfsd_shutdown_generic(void)
242 nfs4_state_shutdown();
243 nfsd_racache_shutdown();
246 static int nfsd_startup_net(int nrservs, struct net *net)
248 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
254 ret = nfsd_startup_generic(nrservs);
257 ret = nfsd_init_socks(net);
263 ret = nfs4_state_start_net(net);
267 nn->nfsd_net_up = true;
273 nfsd_shutdown_generic();
277 static void nfsd_shutdown_net(struct net *net)
279 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
281 nfs4_state_shutdown_net(net);
283 nn->nfsd_net_up = false;
284 nfsd_shutdown_generic();
287 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
289 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
292 * write_ports can create the server without actually starting
293 * any threads--if we get shut down before any threads are
294 * started, then nfsd_last_thread will be run before any of this
295 * other initialization has been done.
297 if (!nn->nfsd_net_up)
299 nfsd_shutdown_net(net);
301 svc_rpcb_cleanup(serv, net);
303 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
305 nfsd_export_flush(net);
308 void nfsd_reset_versions(void)
313 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
314 if (nfsd_program.pg_vers[i])
319 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
320 nfsd_program.pg_vers[i] = nfsd_version[i];
321 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
322 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
323 nfsd_acl_program.pg_vers[i] =
330 * Each session guarantees a negotiated per slot memory cache for replies
331 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
332 * NFSv4.1 server might want to use more memory for a DRC than a machine
333 * with mutiple services.
335 * Impose a hard limit on the number of pages for the DRC which varies
336 * according to the machines free pages. This is of course only a default.
338 * For now this is a #defined shift which could be under admin control
341 static void set_max_drc(void)
343 #define NFSD_DRC_SIZE_SHIFT 10
344 nfsd_drc_max_mem = (nr_free_buffer_pages()
345 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
346 nfsd_drc_mem_used = 0;
347 spin_lock_init(&nfsd_drc_lock);
348 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
351 static int nfsd_get_default_max_blksize(void)
354 unsigned long long target;
358 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
360 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
361 * machines, but only uses 32K on 128M machines. Bottom out at
362 * 8K on 32M and smaller. Of course, this is only a default.
366 ret = NFSSVC_MAXBLKSIZE;
367 while (ret > target && ret >= 8*1024*2)
372 int nfsd_create_serv(struct net *net)
375 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
377 WARN_ON(!mutex_is_locked(&nfsd_mutex));
379 svc_get(nn->nfsd_serv);
382 if (nfsd_max_blksize == 0)
383 nfsd_max_blksize = nfsd_get_default_max_blksize();
384 nfsd_reset_versions();
385 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
386 nfsd_last_thread, nfsd, THIS_MODULE);
387 if (nn->nfsd_serv == NULL)
390 error = svc_bind(nn->nfsd_serv, net);
392 svc_destroy(nn->nfsd_serv);
397 do_gettimeofday(&nn->nfssvc_boot); /* record boot time */
401 int nfsd_nrpools(struct net *net)
403 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
405 if (nn->nfsd_serv == NULL)
408 return nn->nfsd_serv->sv_nrpools;
411 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
414 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
416 if (nn->nfsd_serv != NULL) {
417 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
418 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
424 void nfsd_destroy(struct net *net)
426 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
427 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
430 svc_shutdown_net(nn->nfsd_serv, net);
431 svc_destroy(nn->nfsd_serv);
433 nn->nfsd_serv = NULL;
436 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
441 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
443 WARN_ON(!mutex_is_locked(&nfsd_mutex));
445 if (nn->nfsd_serv == NULL || n <= 0)
448 if (n > nn->nfsd_serv->sv_nrpools)
449 n = nn->nfsd_serv->sv_nrpools;
451 /* enforce a global maximum number of threads */
453 for (i = 0; i < n; i++) {
454 if (nthreads[i] > NFSD_MAXSERVS)
455 nthreads[i] = NFSD_MAXSERVS;
458 if (tot > NFSD_MAXSERVS) {
459 /* total too large: scale down requested numbers */
460 for (i = 0; i < n && tot > 0; i++) {
461 int new = nthreads[i] * NFSD_MAXSERVS / tot;
462 tot -= (nthreads[i] - new);
465 for (i = 0; i < n && tot > 0; i++) {
472 * There must always be a thread in pool 0; the admin
473 * can't shut down NFS completely using pool_threads.
475 if (nthreads[0] == 0)
478 /* apply the new numbers */
479 svc_get(nn->nfsd_serv);
480 for (i = 0; i < n; i++) {
481 err = svc_set_num_threads(nn->nfsd_serv, &nn->nfsd_serv->sv_pools[i],
491 * Adjust the number of threads and return the new number of threads.
492 * This is also the function that starts the server if necessary, if
493 * this is the first time nrservs is nonzero.
496 nfsd_svc(int nrservs, struct net *net)
500 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
502 mutex_lock(&nfsd_mutex);
503 dprintk("nfsd: creating service\n");
506 if (nrservs > NFSD_MAXSERVS)
507 nrservs = NFSD_MAXSERVS;
509 if (nrservs == 0 && nn->nfsd_serv == NULL)
512 error = nfsd_create_serv(net);
516 nfsd_up_before = nn->nfsd_net_up;
518 error = nfsd_startup_net(nrservs, net);
521 error = svc_set_num_threads(nn->nfsd_serv, NULL, nrservs);
524 /* We are holding a reference to nn->nfsd_serv which
525 * we don't want to count in the return value,
528 error = nn->nfsd_serv->sv_nrthreads - 1;
530 if (error < 0 && !nfsd_up_before)
531 nfsd_shutdown_net(net);
533 nfsd_destroy(net); /* Release server */
535 mutex_unlock(&nfsd_mutex);
541 * This is the NFS server kernel thread
546 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
547 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
548 struct net *net = perm_sock->xpt_net;
551 /* Lock module and set up kernel thread */
552 mutex_lock(&nfsd_mutex);
554 /* At this point, the thread shares current->fs
555 * with the init process. We need to create files with a
556 * umask of 0 instead of init's umask. */
557 if (unshare_fs_struct() < 0) {
558 printk("Unable to start nfsd thread: out of memory\n");
562 current->fs->umask = 0;
565 * thread is spawned with all signals set to SIG_IGN, re-enable
566 * the ones that will bring down the thread
568 allow_signal(SIGKILL);
569 allow_signal(SIGHUP);
570 allow_signal(SIGINT);
571 allow_signal(SIGQUIT);
574 mutex_unlock(&nfsd_mutex);
577 * We want less throttling in balance_dirty_pages() so that nfs to
578 * localhost doesn't cause nfsd to lock up due to all the client's
581 current->flags |= PF_LESS_THROTTLE;
585 * The main request loop
589 * Find a socket with data available and call its
592 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
596 validate_process_creds();
598 validate_process_creds();
601 /* Clear signals before calling svc_exit_thread() */
602 flush_signals(current);
604 mutex_lock(&nfsd_mutex);
608 rqstp->rq_server = NULL;
610 /* Release the thread */
611 svc_exit_thread(rqstp);
616 mutex_unlock(&nfsd_mutex);
617 module_put_and_exit(0);
621 static __be32 map_new_errors(u32 vers, __be32 nfserr)
623 if (nfserr == nfserr_jukebox && vers == 2)
624 return nfserr_dropit;
625 if (nfserr == nfserr_wrongsec && vers < 4)
631 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
633 struct svc_procedure *proc;
638 dprintk("nfsd_dispatch: vers %d proc %d\n",
639 rqstp->rq_vers, rqstp->rq_proc);
640 proc = rqstp->rq_procinfo;
643 * Give the xdr decoder a chance to change this if it wants
644 * (necessary in the NFSv4.0 compound case)
646 rqstp->rq_cachetype = proc->pc_cachetype;
647 /* Decode arguments */
648 xdr = proc->pc_decode;
649 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
651 dprintk("nfsd: failed to decode arguments!\n");
652 *statp = rpc_garbage_args;
656 /* Check whether we have this call in the cache. */
657 switch (nfsd_cache_lookup(rqstp)) {
666 /* need to grab the location to store the status, as
667 * nfsv4 does some encoding while processing
669 nfserrp = rqstp->rq_res.head[0].iov_base
670 + rqstp->rq_res.head[0].iov_len;
671 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
673 /* Now call the procedure handler, and encode NFS status. */
674 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
675 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
676 if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
677 dprintk("nfsd: Dropping request; may be revisited later\n");
678 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
682 if (rqstp->rq_proc != 0)
686 * For NFSv2, additional info is never returned in case of an error.
688 if (!(nfserr && rqstp->rq_vers == 2)) {
689 xdr = proc->pc_encode;
690 if (xdr && !xdr(rqstp, nfserrp,
692 /* Failed to encode result. Release cache entry */
693 dprintk("nfsd: failed to encode result!\n");
694 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
695 *statp = rpc_system_err;
700 /* Store reply in cache. */
701 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
705 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
708 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
710 mutex_lock(&nfsd_mutex);
711 if (nn->nfsd_serv == NULL) {
712 mutex_unlock(&nfsd_mutex);
715 /* bump up the psudo refcount while traversing */
716 svc_get(nn->nfsd_serv);
717 ret = svc_pool_stats_open(nn->nfsd_serv, file);
718 mutex_unlock(&nfsd_mutex);
722 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
724 int ret = seq_release(inode, file);
725 struct net *net = inode->i_sb->s_fs_info;
727 mutex_lock(&nfsd_mutex);
728 /* this function really, really should have been called svc_put() */
730 mutex_unlock(&nfsd_mutex);