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/freezer.h>
22 #include <linux/fs_struct.h>
23 #include <linux/kthread.h>
24 #include <linux/swap.h>
26 #include <linux/sunrpc/types.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/svc.h>
29 #include <linux/sunrpc/svcsock.h>
30 #include <linux/sunrpc/cache.h>
31 #include <linux/nfsd/nfsd.h>
32 #include <linux/nfsd/stats.h>
33 #include <linux/nfsd/cache.h>
34 #include <linux/nfsd/syscall.h>
35 #include <linux/lockd/bind.h>
36 #include <linux/nfsacl.h>
38 #define NFSDDBG_FACILITY NFSDDBG_SVC
40 extern struct svc_program nfsd_program;
41 static int nfsd(void *vrqstp);
42 struct timeval nfssvc_boot;
45 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
46 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
47 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
49 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
50 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
51 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
52 * entry of ->sv_pools[].
54 * Transitions of the thread count between zero and non-zero are of particular
55 * interest since the svc_serv needs to be created and initialized at that
58 * Finally, the nfsd_mutex also protects some of the global variables that are
59 * accessed when nfsd starts and that are settable via the write_* routines in
60 * nfsctl.c. In particular:
62 * user_recovery_dirname
66 DEFINE_MUTEX(nfsd_mutex);
67 struct svc_serv *nfsd_serv;
70 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
71 * nfsd_drc_max_pages limits the total amount of memory available for
72 * version 4.1 DRC caches.
73 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
75 spinlock_t nfsd_drc_lock;
76 unsigned int nfsd_drc_max_mem;
77 unsigned int nfsd_drc_mem_used;
79 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
80 static struct svc_stat nfsd_acl_svcstats;
81 static struct svc_version * nfsd_acl_version[] = {
82 [2] = &nfsd_acl_version2,
83 [3] = &nfsd_acl_version3,
86 #define NFSD_ACL_MINVERS 2
87 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
88 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
90 static struct svc_program nfsd_acl_program = {
91 .pg_prog = NFS_ACL_PROGRAM,
92 .pg_nvers = NFSD_ACL_NRVERS,
93 .pg_vers = nfsd_acl_versions,
96 .pg_stats = &nfsd_acl_svcstats,
97 .pg_authenticate = &svc_set_client,
100 static struct svc_stat nfsd_acl_svcstats = {
101 .program = &nfsd_acl_program,
103 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
105 static struct svc_version * nfsd_version[] = {
106 [2] = &nfsd_version2,
107 #if defined(CONFIG_NFSD_V3)
108 [3] = &nfsd_version3,
110 #if defined(CONFIG_NFSD_V4)
111 [4] = &nfsd_version4,
115 #define NFSD_MINVERS 2
116 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
117 static struct svc_version *nfsd_versions[NFSD_NRVERS];
119 struct svc_program nfsd_program = {
120 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
121 .pg_next = &nfsd_acl_program,
123 .pg_prog = NFS_PROGRAM, /* program number */
124 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
125 .pg_vers = nfsd_versions, /* version table */
126 .pg_name = "nfsd", /* program name */
127 .pg_class = "nfsd", /* authentication class */
128 .pg_stats = &nfsd_svcstats, /* version table */
129 .pg_authenticate = &svc_set_client, /* export authentication */
133 u32 nfsd_supported_minorversion;
135 int nfsd_vers(int vers, enum vers_op change)
137 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
141 nfsd_versions[vers] = nfsd_version[vers];
142 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
143 if (vers < NFSD_ACL_NRVERS)
144 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
148 nfsd_versions[vers] = NULL;
149 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
150 if (vers < NFSD_ACL_NRVERS)
151 nfsd_acl_versions[vers] = NULL;
155 return nfsd_versions[vers] != NULL;
157 return nfsd_version[vers] != NULL;
162 int nfsd_minorversion(u32 minorversion, enum vers_op change)
164 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
168 nfsd_supported_minorversion = minorversion;
171 if (minorversion == 0)
173 nfsd_supported_minorversion = minorversion - 1;
176 return minorversion <= nfsd_supported_minorversion;
178 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
184 * Maximum number of nfsd processes
186 #define NFSD_MAXSERVS 8192
188 int nfsd_nrthreads(void)
191 mutex_lock(&nfsd_mutex);
193 rv = nfsd_serv->sv_nrthreads;
194 mutex_unlock(&nfsd_mutex);
198 static void nfsd_last_thread(struct svc_serv *serv)
200 /* When last nfsd thread exits we need to do some clean-up */
201 struct svc_xprt *xprt;
202 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list)
205 nfsd_racache_shutdown();
206 nfs4_state_shutdown();
208 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
213 void nfsd_reset_versions(void)
218 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
219 if (nfsd_program.pg_vers[i])
224 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
225 nfsd_program.pg_vers[i] = nfsd_version[i];
226 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
227 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
228 nfsd_acl_program.pg_vers[i] =
235 * Each session guarantees a negotiated per slot memory cache for replies
236 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
237 * NFSv4.1 server might want to use more memory for a DRC than a machine
238 * with mutiple services.
240 * Impose a hard limit on the number of pages for the DRC which varies
241 * according to the machines free pages. This is of course only a default.
243 * For now this is a #defined shift which could be under admin control
246 static void set_max_drc(void)
248 #define NFSD_DRC_SIZE_SHIFT 10
249 nfsd_drc_max_mem = (nr_free_buffer_pages()
250 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
251 nfsd_drc_mem_used = 0;
252 spin_lock_init(&nfsd_drc_lock);
253 dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
256 int nfsd_create_serv(void)
260 WARN_ON(!mutex_is_locked(&nfsd_mutex));
265 if (nfsd_max_blksize == 0) {
266 /* choose a suitable default */
269 /* Aim for 1/4096 of memory per thread
270 * This gives 1MB on 4Gig machines
271 * But only uses 32K on 128M machines.
272 * Bottom out at 8K on 32M and smaller.
273 * Of course, this is only a default.
275 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
276 i.totalram <<= PAGE_SHIFT - 12;
277 while (nfsd_max_blksize > i.totalram &&
278 nfsd_max_blksize >= 8*1024*2)
279 nfsd_max_blksize /= 2;
282 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
283 nfsd_last_thread, nfsd, THIS_MODULE);
284 if (nfsd_serv == NULL)
289 do_gettimeofday(&nfssvc_boot); /* record boot time */
293 static int nfsd_init_socks(int port)
296 if (!list_empty(&nfsd_serv->sv_permsocks))
299 error = svc_create_xprt(nfsd_serv, "udp", PF_INET, port,
308 error = svc_create_xprt(nfsd_serv, "tcp", PF_INET, port,
320 int nfsd_nrpools(void)
322 if (nfsd_serv == NULL)
325 return nfsd_serv->sv_nrpools;
328 int nfsd_get_nrthreads(int n, int *nthreads)
332 if (nfsd_serv != NULL) {
333 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
334 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
340 int nfsd_set_nrthreads(int n, int *nthreads)
346 WARN_ON(!mutex_is_locked(&nfsd_mutex));
348 if (nfsd_serv == NULL || n <= 0)
351 if (n > nfsd_serv->sv_nrpools)
352 n = nfsd_serv->sv_nrpools;
354 /* enforce a global maximum number of threads */
356 for (i = 0; i < n; i++) {
357 if (nthreads[i] > NFSD_MAXSERVS)
358 nthreads[i] = NFSD_MAXSERVS;
361 if (tot > NFSD_MAXSERVS) {
362 /* total too large: scale down requested numbers */
363 for (i = 0; i < n && tot > 0; i++) {
364 int new = nthreads[i] * NFSD_MAXSERVS / tot;
365 tot -= (nthreads[i] - new);
368 for (i = 0; i < n && tot > 0; i++) {
375 * There must always be a thread in pool 0; the admin
376 * can't shut down NFS completely using pool_threads.
378 if (nthreads[0] == 0)
381 /* apply the new numbers */
383 for (i = 0; i < n; i++) {
384 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
389 svc_destroy(nfsd_serv);
395 nfsd_svc(unsigned short port, int nrservs)
399 mutex_lock(&nfsd_mutex);
400 dprintk("nfsd: creating service\n");
403 if (nrservs > NFSD_MAXSERVS)
404 nrservs = NFSD_MAXSERVS;
406 if (nrservs == 0 && nfsd_serv == NULL)
409 /* Readahead param cache - will no-op if it already exists */
410 error = nfsd_racache_init(2*nrservs);
415 nfsd_reset_versions();
417 error = nfsd_create_serv();
421 error = nfsd_init_socks(port);
425 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
427 /* We are holding a reference to nfsd_serv which
428 * we don't want to count in the return value,
431 error = nfsd_serv->sv_nrthreads - 1;
433 svc_destroy(nfsd_serv); /* Release server */
435 mutex_unlock(&nfsd_mutex);
441 * This is the NFS server kernel thread
446 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
447 int err, preverr = 0;
449 /* Lock module and set up kernel thread */
450 mutex_lock(&nfsd_mutex);
452 /* At this point, the thread shares current->fs
453 * with the init process. We need to create files with a
454 * umask of 0 instead of init's umask. */
455 if (unshare_fs_struct() < 0) {
456 printk("Unable to start nfsd thread: out of memory\n");
460 current->fs->umask = 0;
463 * thread is spawned with all signals set to SIG_IGN, re-enable
464 * the ones that will bring down the thread
466 allow_signal(SIGKILL);
467 allow_signal(SIGHUP);
468 allow_signal(SIGINT);
469 allow_signal(SIGQUIT);
472 mutex_unlock(&nfsd_mutex);
475 * We want less throttling in balance_dirty_pages() so that nfs to
476 * localhost doesn't cause nfsd to lock up due to all the client's
479 current->flags |= PF_LESS_THROTTLE;
483 * The main request loop
487 * Find a socket with data available and call its
490 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
495 if (err != preverr) {
496 printk(KERN_WARNING "%s: unexpected error "
497 "from svc_recv (%d)\n", __func__, -err);
500 schedule_timeout_uninterruptible(HZ);
505 /* Lock the export hash tables for reading. */
510 /* Unlock export hash tables */
514 /* Clear signals before calling svc_exit_thread() */
515 flush_signals(current);
517 mutex_lock(&nfsd_mutex);
521 /* Release the thread */
522 svc_exit_thread(rqstp);
525 mutex_unlock(&nfsd_mutex);
526 module_put_and_exit(0);
530 static __be32 map_new_errors(u32 vers, __be32 nfserr)
532 if (nfserr == nfserr_jukebox && vers == 2)
533 return nfserr_dropit;
534 if (nfserr == nfserr_wrongsec && vers < 4)
540 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
542 struct svc_procedure *proc;
547 dprintk("nfsd_dispatch: vers %d proc %d\n",
548 rqstp->rq_vers, rqstp->rq_proc);
549 proc = rqstp->rq_procinfo;
551 /* Check whether we have this call in the cache. */
552 switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
562 /* Decode arguments */
563 xdr = proc->pc_decode;
564 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
566 dprintk("nfsd: failed to decode arguments!\n");
567 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
568 *statp = rpc_garbage_args;
572 /* need to grab the location to store the status, as
573 * nfsv4 does some encoding while processing
575 nfserrp = rqstp->rq_res.head[0].iov_base
576 + rqstp->rq_res.head[0].iov_len;
577 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
579 /* NFSv4.1 DRC requires statp */
580 if (rqstp->rq_vers == 4)
581 nfsd4_set_statp(rqstp, statp);
583 /* Now call the procedure handler, and encode NFS status. */
584 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
585 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
586 if (nfserr == nfserr_dropit) {
587 dprintk("nfsd: Dropping request; may be revisited later\n");
588 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
592 if (rqstp->rq_proc != 0)
596 * For NFSv2, additional info is never returned in case of an error.
598 if (!(nfserr && rqstp->rq_vers == 2)) {
599 xdr = proc->pc_encode;
600 if (xdr && !xdr(rqstp, nfserrp,
602 /* Failed to encode result. Release cache entry */
603 dprintk("nfsd: failed to encode result!\n");
604 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
605 *statp = rpc_system_err;
610 /* Store reply in cache. */
611 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
615 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
617 if (nfsd_serv == NULL)
619 return svc_pool_stats_open(nfsd_serv, file);