2 * linux/fs/nfsd/nfssvc.c
4 * Central processing for nfsd.
6 * Authors: Olaf Kirch (okir@monad.swb.de)
8 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
11 #include <linux/module.h>
12 #include <linux/sched.h>
13 #include <linux/time.h>
14 #include <linux/errno.h>
15 #include <linux/nfs.h>
17 #include <linux/uio.h>
18 #include <linux/unistd.h>
19 #include <linux/slab.h>
20 #include <linux/smp.h>
21 #include <linux/smp_lock.h>
22 #include <linux/freezer.h>
23 #include <linux/fs_struct.h>
24 #include <linux/kthread.h>
25 #include <linux/swap.h>
27 #include <linux/sunrpc/types.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/sunrpc/svc.h>
30 #include <linux/sunrpc/svcsock.h>
31 #include <linux/sunrpc/cache.h>
32 #include <linux/nfsd/nfsd.h>
33 #include <linux/nfsd/stats.h>
34 #include <linux/nfsd/cache.h>
35 #include <linux/nfsd/syscall.h>
36 #include <linux/lockd/bind.h>
37 #include <linux/nfsacl.h>
39 #define NFSDDBG_FACILITY NFSDDBG_SVC
41 extern struct svc_program nfsd_program;
42 static int nfsd(void *vrqstp);
43 struct timeval nfssvc_boot;
46 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
47 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
48 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
50 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
51 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
52 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
53 * entry of ->sv_pools[].
55 * Transitions of the thread count between zero and non-zero are of particular
56 * interest since the svc_serv needs to be created and initialized at that
59 * Finally, the nfsd_mutex also protects some of the global variables that are
60 * accessed when nfsd starts and that are settable via the write_* routines in
61 * nfsctl.c. In particular:
63 * user_recovery_dirname
67 DEFINE_MUTEX(nfsd_mutex);
68 struct svc_serv *nfsd_serv;
71 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
72 * nfsd_drc_max_pages limits the total amount of memory available for
73 * version 4.1 DRC caches.
74 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
76 spinlock_t nfsd_drc_lock;
77 unsigned int nfsd_drc_max_pages;
78 unsigned int nfsd_drc_pages_used;
80 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
81 static struct svc_stat nfsd_acl_svcstats;
82 static struct svc_version * nfsd_acl_version[] = {
83 [2] = &nfsd_acl_version2,
84 [3] = &nfsd_acl_version3,
87 #define NFSD_ACL_MINVERS 2
88 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
89 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
91 static struct svc_program nfsd_acl_program = {
92 .pg_prog = NFS_ACL_PROGRAM,
93 .pg_nvers = NFSD_ACL_NRVERS,
94 .pg_vers = nfsd_acl_versions,
97 .pg_stats = &nfsd_acl_svcstats,
98 .pg_authenticate = &svc_set_client,
101 static struct svc_stat nfsd_acl_svcstats = {
102 .program = &nfsd_acl_program,
104 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
106 static struct svc_version * nfsd_version[] = {
107 [2] = &nfsd_version2,
108 #if defined(CONFIG_NFSD_V3)
109 [3] = &nfsd_version3,
111 #if defined(CONFIG_NFSD_V4)
112 [4] = &nfsd_version4,
116 #define NFSD_MINVERS 2
117 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
118 static struct svc_version *nfsd_versions[NFSD_NRVERS];
120 struct svc_program nfsd_program = {
121 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
122 .pg_next = &nfsd_acl_program,
124 .pg_prog = NFS_PROGRAM, /* program number */
125 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
126 .pg_vers = nfsd_versions, /* version table */
127 .pg_name = "nfsd", /* program name */
128 .pg_class = "nfsd", /* authentication class */
129 .pg_stats = &nfsd_svcstats, /* version table */
130 .pg_authenticate = &svc_set_client, /* export authentication */
134 u32 nfsd_supported_minorversion;
136 int nfsd_vers(int vers, enum vers_op change)
138 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
142 nfsd_versions[vers] = nfsd_version[vers];
143 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
144 if (vers < NFSD_ACL_NRVERS)
145 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
149 nfsd_versions[vers] = NULL;
150 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
151 if (vers < NFSD_ACL_NRVERS)
152 nfsd_acl_versions[vers] = NULL;
156 return nfsd_versions[vers] != NULL;
158 return nfsd_version[vers] != NULL;
163 int nfsd_minorversion(u32 minorversion, enum vers_op change)
165 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
169 nfsd_supported_minorversion = minorversion;
172 if (minorversion == 0)
174 nfsd_supported_minorversion = minorversion - 1;
177 return minorversion <= nfsd_supported_minorversion;
179 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
185 * Maximum number of nfsd processes
187 #define NFSD_MAXSERVS 8192
189 int nfsd_nrthreads(void)
192 mutex_lock(&nfsd_mutex);
194 rv = nfsd_serv->sv_nrthreads;
195 mutex_unlock(&nfsd_mutex);
199 static void nfsd_last_thread(struct svc_serv *serv)
201 /* When last nfsd thread exits we need to do some clean-up */
202 struct svc_xprt *xprt;
203 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list)
206 nfsd_racache_shutdown();
207 nfs4_state_shutdown();
209 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
214 void nfsd_reset_versions(void)
219 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
220 if (nfsd_program.pg_vers[i])
225 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
226 nfsd_program.pg_vers[i] = nfsd_version[i];
227 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
228 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
229 nfsd_acl_program.pg_vers[i] =
236 * Each session guarantees a negotiated per slot memory cache for replies
237 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
238 * NFSv4.1 server might want to use more memory for a DRC than a machine
239 * with mutiple services.
241 * Impose a hard limit on the number of pages for the DRC which varies
242 * according to the machines free pages. This is of course only a default.
244 * For now this is a #defined shift which could be under admin control
247 static void set_max_drc(void)
249 #define NFSD_DRC_SIZE_SHIFT 10
250 nfsd_drc_max_pages = nr_free_buffer_pages()
251 >> NFSD_DRC_SIZE_SHIFT;
252 nfsd_drc_pages_used = 0;
253 spin_lock_init(&nfsd_drc_lock);
254 dprintk("%s nfsd_drc_max_pages %u\n", __func__,
258 int nfsd_create_serv(void)
262 WARN_ON(!mutex_is_locked(&nfsd_mutex));
267 if (nfsd_max_blksize == 0) {
268 /* choose a suitable default */
271 /* Aim for 1/4096 of memory per thread
272 * This gives 1MB on 4Gig machines
273 * But only uses 32K on 128M machines.
274 * Bottom out at 8K on 32M and smaller.
275 * Of course, this is only a default.
277 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
278 i.totalram <<= PAGE_SHIFT - 12;
279 while (nfsd_max_blksize > i.totalram &&
280 nfsd_max_blksize >= 8*1024*2)
281 nfsd_max_blksize /= 2;
284 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
285 nfsd_last_thread, nfsd, THIS_MODULE);
286 if (nfsd_serv == NULL)
291 do_gettimeofday(&nfssvc_boot); /* record boot time */
295 static int nfsd_init_socks(int port)
298 if (!list_empty(&nfsd_serv->sv_permsocks))
301 error = svc_create_xprt(nfsd_serv, "udp", PF_INET, port,
310 error = svc_create_xprt(nfsd_serv, "tcp", PF_INET, port,
322 int nfsd_nrpools(void)
324 if (nfsd_serv == NULL)
327 return nfsd_serv->sv_nrpools;
330 int nfsd_get_nrthreads(int n, int *nthreads)
334 if (nfsd_serv != NULL) {
335 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
336 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
342 int nfsd_set_nrthreads(int n, int *nthreads)
348 WARN_ON(!mutex_is_locked(&nfsd_mutex));
350 if (nfsd_serv == NULL || n <= 0)
353 if (n > nfsd_serv->sv_nrpools)
354 n = nfsd_serv->sv_nrpools;
356 /* enforce a global maximum number of threads */
358 for (i = 0; i < n; i++) {
359 if (nthreads[i] > NFSD_MAXSERVS)
360 nthreads[i] = NFSD_MAXSERVS;
363 if (tot > NFSD_MAXSERVS) {
364 /* total too large: scale down requested numbers */
365 for (i = 0; i < n && tot > 0; i++) {
366 int new = nthreads[i] * NFSD_MAXSERVS / tot;
367 tot -= (nthreads[i] - new);
370 for (i = 0; i < n && tot > 0; i++) {
377 * There must always be a thread in pool 0; the admin
378 * can't shut down NFS completely using pool_threads.
380 if (nthreads[0] == 0)
383 /* apply the new numbers */
385 for (i = 0; i < n; i++) {
386 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
391 svc_destroy(nfsd_serv);
397 nfsd_svc(unsigned short port, int nrservs)
401 mutex_lock(&nfsd_mutex);
402 dprintk("nfsd: creating service\n");
405 if (nrservs > NFSD_MAXSERVS)
406 nrservs = NFSD_MAXSERVS;
408 if (nrservs == 0 && nfsd_serv == NULL)
411 /* Readahead param cache - will no-op if it already exists */
412 error = nfsd_racache_init(2*nrservs);
417 nfsd_reset_versions();
419 error = nfsd_create_serv();
423 error = nfsd_init_socks(port);
427 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
429 /* We are holding a reference to nfsd_serv which
430 * we don't want to count in the return value,
433 error = nfsd_serv->sv_nrthreads - 1;
435 svc_destroy(nfsd_serv); /* Release server */
437 mutex_unlock(&nfsd_mutex);
443 * This is the NFS server kernel thread
448 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
449 int err, preverr = 0;
451 /* Lock module and set up kernel thread */
452 mutex_lock(&nfsd_mutex);
454 /* At this point, the thread shares current->fs
455 * with the init process. We need to create files with a
456 * umask of 0 instead of init's umask. */
457 if (unshare_fs_struct() < 0) {
458 printk("Unable to start nfsd thread: out of memory\n");
462 current->fs->umask = 0;
465 * thread is spawned with all signals set to SIG_IGN, re-enable
466 * the ones that will bring down the thread
468 allow_signal(SIGKILL);
469 allow_signal(SIGHUP);
470 allow_signal(SIGINT);
471 allow_signal(SIGQUIT);
474 mutex_unlock(&nfsd_mutex);
477 * We want less throttling in balance_dirty_pages() so that nfs to
478 * localhost doesn't cause nfsd to lock up due to all the client's
481 current->flags |= PF_LESS_THROTTLE;
485 * The main request loop
489 * Find a socket with data available and call its
492 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
497 if (err != preverr) {
498 printk(KERN_WARNING "%s: unexpected error "
499 "from svc_recv (%d)\n", __func__, -err);
502 schedule_timeout_uninterruptible(HZ);
507 /* Lock the export hash tables for reading. */
512 /* Unlock export hash tables */
516 /* Clear signals before calling svc_exit_thread() */
517 flush_signals(current);
519 mutex_lock(&nfsd_mutex);
523 /* Release the thread */
524 svc_exit_thread(rqstp);
527 mutex_unlock(&nfsd_mutex);
528 module_put_and_exit(0);
532 static __be32 map_new_errors(u32 vers, __be32 nfserr)
534 if (nfserr == nfserr_jukebox && vers == 2)
535 return nfserr_dropit;
536 if (nfserr == nfserr_wrongsec && vers < 4)
542 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
544 struct svc_procedure *proc;
549 dprintk("nfsd_dispatch: vers %d proc %d\n",
550 rqstp->rq_vers, rqstp->rq_proc);
551 proc = rqstp->rq_procinfo;
553 /* Check whether we have this call in the cache. */
554 switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
564 /* Decode arguments */
565 xdr = proc->pc_decode;
566 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
568 dprintk("nfsd: failed to decode arguments!\n");
569 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
570 *statp = rpc_garbage_args;
574 /* need to grab the location to store the status, as
575 * nfsv4 does some encoding while processing
577 nfserrp = rqstp->rq_res.head[0].iov_base
578 + rqstp->rq_res.head[0].iov_len;
579 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
581 /* NFSv4.1 DRC requires statp */
582 if (rqstp->rq_vers == 4)
583 nfsd4_set_statp(rqstp, statp);
585 /* Now call the procedure handler, and encode NFS status. */
586 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
587 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
588 if (nfserr == nfserr_dropit) {
589 dprintk("nfsd: Dropping request; may be revisited later\n");
590 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
594 if (rqstp->rq_proc != 0)
598 * For NFSv2, additional info is never returned in case of an error.
600 if (!(nfserr && rqstp->rq_vers == 2)) {
601 xdr = proc->pc_encode;
602 if (xdr && !xdr(rqstp, nfserrp,
604 /* Failed to encode result. Release cache entry */
605 dprintk("nfsd: failed to encode result!\n");
606 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
607 *statp = rpc_system_err;
612 /* Store reply in cache. */
613 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
617 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
619 if (nfsd_serv == NULL)
621 return svc_pool_stats_open(nfsd_serv, file);