2 * Central processing for nfsd.
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] = {
125 int nfsd_vers(int vers, enum vers_op change)
127 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
131 nfsd_versions[vers] = nfsd_version[vers];
132 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
133 if (vers < NFSD_ACL_NRVERS)
134 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
138 nfsd_versions[vers] = NULL;
139 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
140 if (vers < NFSD_ACL_NRVERS)
141 nfsd_acl_versions[vers] = NULL;
145 return nfsd_versions[vers] != NULL;
147 return nfsd_version[vers] != NULL;
152 int nfsd_minorversion(u32 minorversion, enum vers_op change)
154 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
158 nfsd_supported_minorversions[minorversion] = true;
161 nfsd_supported_minorversions[minorversion] = false;
164 return nfsd_supported_minorversions[minorversion];
166 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
172 * Maximum number of nfsd processes
174 #define NFSD_MAXSERVS 8192
176 int nfsd_nrthreads(struct net *net)
179 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
181 mutex_lock(&nfsd_mutex);
183 rv = nn->nfsd_serv->sv_nrthreads;
184 mutex_unlock(&nfsd_mutex);
188 static int nfsd_init_socks(struct net *net)
191 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
193 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
196 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
201 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
209 static int nfsd_users = 0;
211 static int nfsd_startup_generic(int nrservs)
219 * Readahead param cache - will no-op if it already exists.
220 * (Note therefore results will be suboptimal if number of
221 * threads is modified after nfsd start.)
223 ret = nfsd_racache_init(2*nrservs);
227 ret = nfs4_state_start();
233 nfsd_racache_shutdown();
239 static void nfsd_shutdown_generic(void)
244 nfs4_state_shutdown();
245 nfsd_racache_shutdown();
248 static bool nfsd_needs_lockd(void)
250 #if defined(CONFIG_NFSD_V3)
251 return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
253 return (nfsd_versions[2] != NULL);
257 static int nfsd_startup_net(int nrservs, struct net *net)
259 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
265 ret = nfsd_startup_generic(nrservs);
268 ret = nfsd_init_socks(net);
272 if (nfsd_needs_lockd() && !nn->lockd_up) {
279 ret = nfs4_state_start_net(net);
283 nn->nfsd_net_up = true;
292 nfsd_shutdown_generic();
296 static void nfsd_shutdown_net(struct net *net)
298 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
300 nfs4_state_shutdown_net(net);
305 nn->nfsd_net_up = false;
306 nfsd_shutdown_generic();
309 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
311 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
314 * write_ports can create the server without actually starting
315 * any threads--if we get shut down before any threads are
316 * started, then nfsd_last_thread will be run before any of this
317 * other initialization has been done.
319 if (!nn->nfsd_net_up)
321 nfsd_shutdown_net(net);
323 svc_rpcb_cleanup(serv, net);
325 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
327 nfsd_export_flush(net);
330 void nfsd_reset_versions(void)
335 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
336 if (nfsd_program.pg_vers[i])
341 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
342 nfsd_program.pg_vers[i] = nfsd_version[i];
343 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
344 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
345 nfsd_acl_program.pg_vers[i] =
352 * Each session guarantees a negotiated per slot memory cache for replies
353 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
354 * NFSv4.1 server might want to use more memory for a DRC than a machine
355 * with mutiple services.
357 * Impose a hard limit on the number of pages for the DRC which varies
358 * according to the machines free pages. This is of course only a default.
360 * For now this is a #defined shift which could be under admin control
363 static void set_max_drc(void)
365 #define NFSD_DRC_SIZE_SHIFT 10
366 nfsd_drc_max_mem = (nr_free_buffer_pages()
367 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
368 nfsd_drc_mem_used = 0;
369 spin_lock_init(&nfsd_drc_lock);
370 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
373 static int nfsd_get_default_max_blksize(void)
376 unsigned long long target;
380 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
382 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
383 * machines, but only uses 32K on 128M machines. Bottom out at
384 * 8K on 32M and smaller. Of course, this is only a default.
388 ret = NFSSVC_MAXBLKSIZE;
389 while (ret > target && ret >= 8*1024*2)
394 static struct svc_serv_ops nfsd_thread_sv_ops = {
395 .svo_shutdown = nfsd_last_thread,
396 .svo_function = nfsd,
397 .svo_enqueue_xprt = svc_xprt_do_enqueue,
398 .svo_setup = svc_set_num_threads,
399 .svo_module = THIS_MODULE,
402 int nfsd_create_serv(struct net *net)
405 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
407 WARN_ON(!mutex_is_locked(&nfsd_mutex));
409 svc_get(nn->nfsd_serv);
412 if (nfsd_max_blksize == 0)
413 nfsd_max_blksize = nfsd_get_default_max_blksize();
414 nfsd_reset_versions();
415 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
416 &nfsd_thread_sv_ops);
417 if (nn->nfsd_serv == NULL)
420 nn->nfsd_serv->sv_maxconn = nn->max_connections;
421 error = svc_bind(nn->nfsd_serv, net);
423 svc_destroy(nn->nfsd_serv);
428 do_gettimeofday(&nn->nfssvc_boot); /* record boot time */
432 int nfsd_nrpools(struct net *net)
434 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
436 if (nn->nfsd_serv == NULL)
439 return nn->nfsd_serv->sv_nrpools;
442 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
445 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
447 if (nn->nfsd_serv != NULL) {
448 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
449 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
455 void nfsd_destroy(struct net *net)
457 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
458 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
461 svc_shutdown_net(nn->nfsd_serv, net);
462 svc_destroy(nn->nfsd_serv);
464 nn->nfsd_serv = NULL;
467 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
472 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
474 WARN_ON(!mutex_is_locked(&nfsd_mutex));
476 if (nn->nfsd_serv == NULL || n <= 0)
479 if (n > nn->nfsd_serv->sv_nrpools)
480 n = nn->nfsd_serv->sv_nrpools;
482 /* enforce a global maximum number of threads */
484 for (i = 0; i < n; i++) {
485 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
488 if (tot > NFSD_MAXSERVS) {
489 /* total too large: scale down requested numbers */
490 for (i = 0; i < n && tot > 0; i++) {
491 int new = nthreads[i] * NFSD_MAXSERVS / tot;
492 tot -= (nthreads[i] - new);
495 for (i = 0; i < n && tot > 0; i++) {
502 * There must always be a thread in pool 0; the admin
503 * can't shut down NFS completely using pool_threads.
505 if (nthreads[0] == 0)
508 /* apply the new numbers */
509 svc_get(nn->nfsd_serv);
510 for (i = 0; i < n; i++) {
511 err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
512 &nn->nfsd_serv->sv_pools[i], nthreads[i]);
521 * Adjust the number of threads and return the new number of threads.
522 * This is also the function that starts the server if necessary, if
523 * this is the first time nrservs is nonzero.
526 nfsd_svc(int nrservs, struct net *net)
530 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
532 mutex_lock(&nfsd_mutex);
533 dprintk("nfsd: creating service\n");
535 nrservs = max(nrservs, 0);
536 nrservs = min(nrservs, NFSD_MAXSERVS);
539 if (nrservs == 0 && nn->nfsd_serv == NULL)
542 error = nfsd_create_serv(net);
546 nfsd_up_before = nn->nfsd_net_up;
548 error = nfsd_startup_net(nrservs, net);
551 error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
555 /* We are holding a reference to nn->nfsd_serv which
556 * we don't want to count in the return value,
559 error = nn->nfsd_serv->sv_nrthreads - 1;
561 if (error < 0 && !nfsd_up_before)
562 nfsd_shutdown_net(net);
564 nfsd_destroy(net); /* Release server */
566 mutex_unlock(&nfsd_mutex);
572 * This is the NFS server kernel thread
577 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
578 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
579 struct net *net = perm_sock->xpt_net;
580 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
583 /* Lock module and set up kernel thread */
584 mutex_lock(&nfsd_mutex);
586 /* At this point, the thread shares current->fs
587 * with the init process. We need to create files with a
588 * umask of 0 instead of init's umask. */
589 if (unshare_fs_struct() < 0) {
590 printk("Unable to start nfsd thread: out of memory\n");
594 current->fs->umask = 0;
597 * thread is spawned with all signals set to SIG_IGN, re-enable
598 * the ones that will bring down the thread
600 allow_signal(SIGKILL);
601 allow_signal(SIGHUP);
602 allow_signal(SIGINT);
603 allow_signal(SIGQUIT);
606 mutex_unlock(&nfsd_mutex);
611 * The main request loop
614 /* Update sv_maxconn if it has changed */
615 rqstp->rq_server->sv_maxconn = nn->max_connections;
618 * Find a socket with data available and call its
621 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
625 validate_process_creds();
627 validate_process_creds();
630 /* Clear signals before calling svc_exit_thread() */
631 flush_signals(current);
633 mutex_lock(&nfsd_mutex);
637 rqstp->rq_server = NULL;
639 /* Release the thread */
640 svc_exit_thread(rqstp);
645 mutex_unlock(&nfsd_mutex);
646 module_put_and_exit(0);
650 static __be32 map_new_errors(u32 vers, __be32 nfserr)
652 if (nfserr == nfserr_jukebox && vers == 2)
653 return nfserr_dropit;
654 if (nfserr == nfserr_wrongsec && vers < 4)
660 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
662 struct svc_procedure *proc;
667 dprintk("nfsd_dispatch: vers %d proc %d\n",
668 rqstp->rq_vers, rqstp->rq_proc);
669 proc = rqstp->rq_procinfo;
672 * Give the xdr decoder a chance to change this if it wants
673 * (necessary in the NFSv4.0 compound case)
675 rqstp->rq_cachetype = proc->pc_cachetype;
676 /* Decode arguments */
677 xdr = proc->pc_decode;
678 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
680 dprintk("nfsd: failed to decode arguments!\n");
681 *statp = rpc_garbage_args;
685 /* Check whether we have this call in the cache. */
686 switch (nfsd_cache_lookup(rqstp)) {
695 /* need to grab the location to store the status, as
696 * nfsv4 does some encoding while processing
698 nfserrp = rqstp->rq_res.head[0].iov_base
699 + rqstp->rq_res.head[0].iov_len;
700 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
702 /* Now call the procedure handler, and encode NFS status. */
703 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
704 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
705 if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
706 dprintk("nfsd: Dropping request; may be revisited later\n");
707 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
711 if (rqstp->rq_proc != 0)
715 * For NFSv2, additional info is never returned in case of an error.
717 if (!(nfserr && rqstp->rq_vers == 2)) {
718 xdr = proc->pc_encode;
719 if (xdr && !xdr(rqstp, nfserrp,
721 /* Failed to encode result. Release cache entry */
722 dprintk("nfsd: failed to encode result!\n");
723 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
724 *statp = rpc_system_err;
729 /* Store reply in cache. */
730 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
734 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
737 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
739 mutex_lock(&nfsd_mutex);
740 if (nn->nfsd_serv == NULL) {
741 mutex_unlock(&nfsd_mutex);
744 /* bump up the psudo refcount while traversing */
745 svc_get(nn->nfsd_serv);
746 ret = svc_pool_stats_open(nn->nfsd_serv, file);
747 mutex_unlock(&nfsd_mutex);
751 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
753 int ret = seq_release(inode, file);
754 struct net *net = inode->i_sb->s_fs_info;
756 mutex_lock(&nfsd_mutex);
757 /* this function really, really should have been called svc_put() */
759 mutex_unlock(&nfsd_mutex);