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>
14 #include <linux/nsproxy.h>
16 #include <linux/sunrpc/stats.h>
17 #include <linux/sunrpc/svcsock.h>
18 #include <linux/lockd/bind.h>
19 #include <linux/nfsacl.h>
20 #include <linux/seq_file.h>
21 #include <net/net_namespace.h>
26 #define NFSDDBG_FACILITY NFSDDBG_SVC
28 extern struct svc_program nfsd_program;
29 static int nfsd(void *vrqstp);
30 struct timeval nfssvc_boot;
33 * nfsd_mutex protects 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) 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);
55 struct svc_serv *nfsd_serv;
58 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
59 * nfsd_drc_max_pages limits the total amount of memory available for
60 * version 4.1 DRC caches.
61 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
63 spinlock_t nfsd_drc_lock;
64 unsigned int nfsd_drc_max_mem;
65 unsigned int nfsd_drc_mem_used;
67 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
68 static struct svc_stat nfsd_acl_svcstats;
69 static struct svc_version * nfsd_acl_version[] = {
70 [2] = &nfsd_acl_version2,
71 [3] = &nfsd_acl_version3,
74 #define NFSD_ACL_MINVERS 2
75 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
76 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
78 static struct svc_program nfsd_acl_program = {
79 .pg_prog = NFS_ACL_PROGRAM,
80 .pg_nvers = NFSD_ACL_NRVERS,
81 .pg_vers = nfsd_acl_versions,
84 .pg_stats = &nfsd_acl_svcstats,
85 .pg_authenticate = &svc_set_client,
88 static struct svc_stat nfsd_acl_svcstats = {
89 .program = &nfsd_acl_program,
91 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
93 static struct svc_version * nfsd_version[] = {
95 #if defined(CONFIG_NFSD_V3)
98 #if defined(CONFIG_NFSD_V4)
103 #define NFSD_MINVERS 2
104 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
105 static struct svc_version *nfsd_versions[NFSD_NRVERS];
107 struct svc_program nfsd_program = {
108 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
109 .pg_next = &nfsd_acl_program,
111 .pg_prog = NFS_PROGRAM, /* program number */
112 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
113 .pg_vers = nfsd_versions, /* version table */
114 .pg_name = "nfsd", /* program name */
115 .pg_class = "nfsd", /* authentication class */
116 .pg_stats = &nfsd_svcstats, /* version table */
117 .pg_authenticate = &svc_set_client, /* export authentication */
121 u32 nfsd_supported_minorversion;
123 int nfsd_vers(int vers, enum vers_op change)
125 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
129 nfsd_versions[vers] = nfsd_version[vers];
130 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
131 if (vers < NFSD_ACL_NRVERS)
132 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
136 nfsd_versions[vers] = NULL;
137 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
138 if (vers < NFSD_ACL_NRVERS)
139 nfsd_acl_versions[vers] = NULL;
143 return nfsd_versions[vers] != NULL;
145 return nfsd_version[vers] != NULL;
150 int nfsd_minorversion(u32 minorversion, enum vers_op change)
152 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
156 nfsd_supported_minorversion = minorversion;
159 if (minorversion == 0)
161 nfsd_supported_minorversion = minorversion - 1;
164 return minorversion <= nfsd_supported_minorversion;
166 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
172 * Maximum number of nfsd processes
174 #define NFSD_MAXSERVS 8192
176 int nfsd_nrthreads(void)
179 mutex_lock(&nfsd_mutex);
181 rv = nfsd_serv->sv_nrthreads;
182 mutex_unlock(&nfsd_mutex);
186 static int nfsd_init_socks(int port)
189 if (!list_empty(&nfsd_serv->sv_permsocks))
192 error = svc_create_xprt(nfsd_serv, "udp", &init_net, PF_INET, port,
197 error = svc_create_xprt(nfsd_serv, "tcp", &init_net, PF_INET, port,
205 static bool nfsd_up = false;
207 static int nfsd_startup(unsigned short port, int nrservs)
214 * Readahead param cache - will no-op if it already exists.
215 * (Note therefore results will be suboptimal if number of
216 * threads is modified after nfsd start.)
218 ret = nfsd_racache_init(2*nrservs);
221 ret = nfsd_init_socks(port);
224 ret = lockd_up(&init_net);
227 ret = nfs4_state_start();
233 lockd_down(&init_net);
235 nfsd_racache_shutdown();
239 static void nfsd_shutdown(void)
242 * write_ports can create the server without actually starting
243 * any threads--if we get shut down before any threads are
244 * started, then nfsd_last_thread will be run before any of this
245 * other initialization has been done.
249 nfs4_state_shutdown();
250 lockd_down(&init_net);
251 nfsd_racache_shutdown();
255 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
257 /* When last nfsd thread exits we need to do some clean-up */
261 svc_rpcb_cleanup(serv, net);
263 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
265 nfsd_export_flush(net);
268 void nfsd_reset_versions(void)
273 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
274 if (nfsd_program.pg_vers[i])
279 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
280 nfsd_program.pg_vers[i] = nfsd_version[i];
281 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
282 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
283 nfsd_acl_program.pg_vers[i] =
290 * Each session guarantees a negotiated per slot memory cache for replies
291 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
292 * NFSv4.1 server might want to use more memory for a DRC than a machine
293 * with mutiple services.
295 * Impose a hard limit on the number of pages for the DRC which varies
296 * according to the machines free pages. This is of course only a default.
298 * For now this is a #defined shift which could be under admin control
301 static void set_max_drc(void)
303 #define NFSD_DRC_SIZE_SHIFT 10
304 nfsd_drc_max_mem = (nr_free_buffer_pages()
305 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
306 nfsd_drc_mem_used = 0;
307 spin_lock_init(&nfsd_drc_lock);
308 dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
311 static int nfsd_get_default_max_blksize(void)
314 unsigned long long target;
318 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
320 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
321 * machines, but only uses 32K on 128M machines. Bottom out at
322 * 8K on 32M and smaller. Of course, this is only a default.
326 ret = NFSSVC_MAXBLKSIZE;
327 while (ret > target && ret >= 8*1024*2)
332 int nfsd_create_serv(void)
336 WARN_ON(!mutex_is_locked(&nfsd_mutex));
341 if (nfsd_max_blksize == 0)
342 nfsd_max_blksize = nfsd_get_default_max_blksize();
343 nfsd_reset_versions();
344 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
345 nfsd_last_thread, nfsd, THIS_MODULE);
346 if (nfsd_serv == NULL)
349 error = svc_bind(nfsd_serv, current->nsproxy->net_ns);
351 svc_destroy(nfsd_serv);
356 do_gettimeofday(&nfssvc_boot); /* record boot time */
360 int nfsd_nrpools(void)
362 if (nfsd_serv == NULL)
365 return nfsd_serv->sv_nrpools;
368 int nfsd_get_nrthreads(int n, int *nthreads)
372 if (nfsd_serv != NULL) {
373 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
374 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
380 int nfsd_set_nrthreads(int n, int *nthreads)
385 struct net *net = &init_net;
387 WARN_ON(!mutex_is_locked(&nfsd_mutex));
389 if (nfsd_serv == NULL || n <= 0)
392 if (n > nfsd_serv->sv_nrpools)
393 n = nfsd_serv->sv_nrpools;
395 /* enforce a global maximum number of threads */
397 for (i = 0; i < n; i++) {
398 if (nthreads[i] > NFSD_MAXSERVS)
399 nthreads[i] = NFSD_MAXSERVS;
402 if (tot > NFSD_MAXSERVS) {
403 /* total too large: scale down requested numbers */
404 for (i = 0; i < n && tot > 0; i++) {
405 int new = nthreads[i] * NFSD_MAXSERVS / tot;
406 tot -= (nthreads[i] - new);
409 for (i = 0; i < n && tot > 0; i++) {
416 * There must always be a thread in pool 0; the admin
417 * can't shut down NFS completely using pool_threads.
419 if (nthreads[0] == 0)
422 /* apply the new numbers */
424 for (i = 0; i < n; i++) {
425 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
431 if (nfsd_serv->sv_nrthreads == 1)
432 svc_shutdown_net(nfsd_serv, net);
433 svc_destroy(nfsd_serv);
439 * Adjust the number of threads and return the new number of threads.
440 * This is also the function that starts the server if necessary, if
441 * this is the first time nrservs is nonzero.
444 nfsd_svc(unsigned short port, int nrservs)
448 struct net *net = &init_net;
450 mutex_lock(&nfsd_mutex);
451 dprintk("nfsd: creating service\n");
454 if (nrservs > NFSD_MAXSERVS)
455 nrservs = NFSD_MAXSERVS;
457 if (nrservs == 0 && nfsd_serv == NULL)
460 error = nfsd_create_serv();
464 nfsd_up_before = nfsd_up;
466 error = nfsd_startup(port, nrservs);
469 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
472 /* We are holding a reference to nfsd_serv which
473 * we don't want to count in the return value,
476 error = nfsd_serv->sv_nrthreads - 1;
478 if (error < 0 && !nfsd_up_before)
481 if (nfsd_serv->sv_nrthreads == 1)
482 svc_shutdown_net(nfsd_serv, net);
483 svc_destroy(nfsd_serv); /* Release server */
485 mutex_unlock(&nfsd_mutex);
491 * This is the NFS server kernel thread
496 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
497 int err, preverr = 0;
499 /* Lock module and set up kernel thread */
500 mutex_lock(&nfsd_mutex);
502 /* At this point, the thread shares current->fs
503 * with the init process. We need to create files with a
504 * umask of 0 instead of init's umask. */
505 if (unshare_fs_struct() < 0) {
506 printk("Unable to start nfsd thread: out of memory\n");
510 current->fs->umask = 0;
513 * thread is spawned with all signals set to SIG_IGN, re-enable
514 * the ones that will bring down the thread
516 allow_signal(SIGKILL);
517 allow_signal(SIGHUP);
518 allow_signal(SIGINT);
519 allow_signal(SIGQUIT);
522 mutex_unlock(&nfsd_mutex);
525 * We want less throttling in balance_dirty_pages() so that nfs to
526 * localhost doesn't cause nfsd to lock up due to all the client's
529 current->flags |= PF_LESS_THROTTLE;
533 * The main request loop
537 * Find a socket with data available and call its
540 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
545 if (err != preverr) {
546 printk(KERN_WARNING "%s: unexpected error "
547 "from svc_recv (%d)\n", __func__, -err);
550 schedule_timeout_uninterruptible(HZ);
554 validate_process_creds();
556 validate_process_creds();
559 /* Clear signals before calling svc_exit_thread() */
560 flush_signals(current);
562 mutex_lock(&nfsd_mutex);
566 if (rqstp->rq_server->sv_nrthreads == 1)
567 svc_shutdown_net(rqstp->rq_server, &init_net);
569 /* Release the thread */
570 svc_exit_thread(rqstp);
573 mutex_unlock(&nfsd_mutex);
574 module_put_and_exit(0);
578 static __be32 map_new_errors(u32 vers, __be32 nfserr)
580 if (nfserr == nfserr_jukebox && vers == 2)
581 return nfserr_dropit;
582 if (nfserr == nfserr_wrongsec && vers < 4)
588 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
590 struct svc_procedure *proc;
595 dprintk("nfsd_dispatch: vers %d proc %d\n",
596 rqstp->rq_vers, rqstp->rq_proc);
597 proc = rqstp->rq_procinfo;
600 * Give the xdr decoder a chance to change this if it wants
601 * (necessary in the NFSv4.0 compound case)
603 rqstp->rq_cachetype = proc->pc_cachetype;
604 /* Decode arguments */
605 xdr = proc->pc_decode;
606 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
608 dprintk("nfsd: failed to decode arguments!\n");
609 *statp = rpc_garbage_args;
613 /* Check whether we have this call in the cache. */
614 switch (nfsd_cache_lookup(rqstp)) {
624 /* need to grab the location to store the status, as
625 * nfsv4 does some encoding while processing
627 nfserrp = rqstp->rq_res.head[0].iov_base
628 + rqstp->rq_res.head[0].iov_len;
629 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
631 /* Now call the procedure handler, and encode NFS status. */
632 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
633 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
634 if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
635 dprintk("nfsd: Dropping request; may be revisited later\n");
636 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
640 if (rqstp->rq_proc != 0)
644 * For NFSv2, additional info is never returned in case of an error.
646 if (!(nfserr && rqstp->rq_vers == 2)) {
647 xdr = proc->pc_encode;
648 if (xdr && !xdr(rqstp, nfserrp,
650 /* Failed to encode result. Release cache entry */
651 dprintk("nfsd: failed to encode result!\n");
652 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
653 *statp = rpc_system_err;
658 /* Store reply in cache. */
659 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
663 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
666 mutex_lock(&nfsd_mutex);
667 if (nfsd_serv == NULL) {
668 mutex_unlock(&nfsd_mutex);
671 /* bump up the psudo refcount while traversing */
673 ret = svc_pool_stats_open(nfsd_serv, file);
674 mutex_unlock(&nfsd_mutex);
678 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
680 int ret = seq_release(inode, file);
681 struct net *net = &init_net;
683 mutex_lock(&nfsd_mutex);
684 /* this function really, really should have been called svc_put() */
685 if (nfsd_serv->sv_nrthreads == 1)
686 svc_shutdown_net(nfsd_serv, net);
687 svc_destroy(nfsd_serv);
688 mutex_unlock(&nfsd_mutex);