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>
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 /* these signals will be delivered to an nfsd thread
41 * when handling a request
43 #define ALLOWED_SIGS (sigmask(SIGKILL))
44 /* these signals will be delivered to an nfsd thread
45 * when not handling a request. i.e. when waiting
47 #define SHUTDOWN_SIGS (sigmask(SIGKILL) | sigmask(SIGHUP) | sigmask(SIGINT) | sigmask(SIGQUIT))
49 extern struct svc_program nfsd_program;
50 static int nfsd(void *vrqstp);
51 struct timeval nfssvc_boot;
52 static atomic_t nfsd_busy;
53 static unsigned long nfsd_last_call;
54 static DEFINE_SPINLOCK(nfsd_call_lock);
57 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
58 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
59 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
61 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
62 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
63 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
64 * entry of ->sv_pools[].
66 * Transitions of the thread count between zero and non-zero are of particular
67 * interest since the svc_serv needs to be created and initialized at that
70 * Finally, the nfsd_mutex also protects some of the global variables that are
71 * accessed when nfsd starts and that are settable via the write_* routines in
72 * nfsctl.c. In particular:
74 * user_recovery_dirname
78 DEFINE_MUTEX(nfsd_mutex);
79 struct svc_serv *nfsd_serv;
81 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
82 static struct svc_stat nfsd_acl_svcstats;
83 static struct svc_version * nfsd_acl_version[] = {
84 [2] = &nfsd_acl_version2,
85 [3] = &nfsd_acl_version3,
88 #define NFSD_ACL_MINVERS 2
89 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
90 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
92 static struct svc_program nfsd_acl_program = {
93 .pg_prog = NFS_ACL_PROGRAM,
94 .pg_nvers = NFSD_ACL_NRVERS,
95 .pg_vers = nfsd_acl_versions,
98 .pg_stats = &nfsd_acl_svcstats,
99 .pg_authenticate = &svc_set_client,
102 static struct svc_stat nfsd_acl_svcstats = {
103 .program = &nfsd_acl_program,
105 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
107 static struct svc_version * nfsd_version[] = {
108 [2] = &nfsd_version2,
109 #if defined(CONFIG_NFSD_V3)
110 [3] = &nfsd_version3,
112 #if defined(CONFIG_NFSD_V4)
113 [4] = &nfsd_version4,
117 #define NFSD_MINVERS 2
118 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
119 static struct svc_version *nfsd_versions[NFSD_NRVERS];
121 struct svc_program nfsd_program = {
122 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
123 .pg_next = &nfsd_acl_program,
125 .pg_prog = NFS_PROGRAM, /* program number */
126 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
127 .pg_vers = nfsd_versions, /* version table */
128 .pg_name = "nfsd", /* program name */
129 .pg_class = "nfsd", /* authentication class */
130 .pg_stats = &nfsd_svcstats, /* version table */
131 .pg_authenticate = &svc_set_client, /* export authentication */
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 * Maximum number of nfsd processes
164 #define NFSD_MAXSERVS 8192
166 int nfsd_nrthreads(void)
168 if (nfsd_serv == NULL)
171 return nfsd_serv->sv_nrthreads;
174 static void nfsd_last_thread(struct svc_serv *serv)
176 /* When last nfsd thread exits we need to do some clean-up */
177 struct svc_xprt *xprt;
178 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list)
181 nfsd_racache_shutdown();
182 nfs4_state_shutdown();
184 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
189 void nfsd_reset_versions(void)
194 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
195 if (nfsd_program.pg_vers[i])
200 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
201 nfsd_program.pg_vers[i] = nfsd_version[i];
202 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
203 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
204 nfsd_acl_program.pg_vers[i] =
211 int nfsd_create_serv(void)
215 WARN_ON(!mutex_is_locked(&nfsd_mutex));
220 if (nfsd_max_blksize == 0) {
221 /* choose a suitable default */
224 /* Aim for 1/4096 of memory per thread
225 * This gives 1MB on 4Gig machines
226 * But only uses 32K on 128M machines.
227 * Bottom out at 8K on 32M and smaller.
228 * Of course, this is only a default.
230 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
231 i.totalram <<= PAGE_SHIFT - 12;
232 while (nfsd_max_blksize > i.totalram &&
233 nfsd_max_blksize >= 8*1024*2)
234 nfsd_max_blksize /= 2;
237 atomic_set(&nfsd_busy, 0);
238 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
239 nfsd_last_thread, nfsd, THIS_MODULE);
240 if (nfsd_serv == NULL)
243 do_gettimeofday(&nfssvc_boot); /* record boot time */
247 static int nfsd_init_socks(int port)
250 if (!list_empty(&nfsd_serv->sv_permsocks))
253 error = lockd_up(IPPROTO_UDP);
255 error = svc_create_xprt(nfsd_serv, "udp", port,
263 error = lockd_up(IPPROTO_TCP);
265 error = svc_create_xprt(nfsd_serv, "tcp", port,
275 int nfsd_nrpools(void)
277 if (nfsd_serv == NULL)
280 return nfsd_serv->sv_nrpools;
283 int nfsd_get_nrthreads(int n, int *nthreads)
287 if (nfsd_serv != NULL) {
288 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
289 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
295 int nfsd_set_nrthreads(int n, int *nthreads)
301 WARN_ON(!mutex_is_locked(&nfsd_mutex));
303 if (nfsd_serv == NULL || n <= 0)
306 if (n > nfsd_serv->sv_nrpools)
307 n = nfsd_serv->sv_nrpools;
309 /* enforce a global maximum number of threads */
311 for (i = 0; i < n; i++) {
312 if (nthreads[i] > NFSD_MAXSERVS)
313 nthreads[i] = NFSD_MAXSERVS;
316 if (tot > NFSD_MAXSERVS) {
317 /* total too large: scale down requested numbers */
318 for (i = 0; i < n && tot > 0; i++) {
319 int new = nthreads[i] * NFSD_MAXSERVS / tot;
320 tot -= (nthreads[i] - new);
323 for (i = 0; i < n && tot > 0; i++) {
330 * There must always be a thread in pool 0; the admin
331 * can't shut down NFS completely using pool_threads.
333 if (nthreads[0] == 0)
336 /* apply the new numbers */
338 for (i = 0; i < n; i++) {
339 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
344 svc_destroy(nfsd_serv);
350 nfsd_svc(unsigned short port, int nrservs)
354 mutex_lock(&nfsd_mutex);
355 dprintk("nfsd: creating service\n");
359 if (nrservs > NFSD_MAXSERVS)
360 nrservs = NFSD_MAXSERVS;
362 /* Readahead param cache - will no-op if it already exists */
363 error = nfsd_racache_init(2*nrservs);
368 nfsd_reset_versions();
370 error = nfsd_create_serv();
374 error = nfsd_init_socks(port);
378 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
380 svc_destroy(nfsd_serv); /* Release server */
382 mutex_unlock(&nfsd_mutex);
387 update_thread_usage(int busy_threads)
389 unsigned long prev_call;
393 spin_lock(&nfsd_call_lock);
394 prev_call = nfsd_last_call;
395 nfsd_last_call = jiffies;
396 decile = busy_threads*10/nfsdstats.th_cnt;
397 if (decile>0 && decile <= 10) {
398 diff = nfsd_last_call - prev_call;
399 if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
400 nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
402 nfsdstats.th_fullcnt++;
404 spin_unlock(&nfsd_call_lock);
408 * This is the NFS server kernel thread
413 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
414 struct fs_struct *fsp;
415 sigset_t shutdown_mask, allowed_mask;
416 int err, preverr = 0;
419 /* Lock module and set up kernel thread */
420 mutex_lock(&nfsd_mutex);
422 /* At this point, the thread shares current->fs
423 * with the init process. We need to create files with a
424 * umask of 0 instead of init's umask. */
425 fsp = copy_fs_struct(current->fs);
427 printk("Unable to start nfsd thread: out of memory\n");
432 current->fs->umask = 0;
434 siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS);
435 siginitsetinv(&allowed_mask, ALLOWED_SIGS);
438 * thread is spawned with all signals set to SIG_IGN, re-enable
439 * the ones that matter
441 for (signo = 1; signo <= _NSIG; signo++) {
442 if (!sigismember(&shutdown_mask, signo))
447 mutex_unlock(&nfsd_mutex);
450 * We want less throttling in balance_dirty_pages() so that nfs to
451 * localhost doesn't cause nfsd to lock up due to all the client's
454 current->flags |= PF_LESS_THROTTLE;
458 * The main request loop
461 /* Block all but the shutdown signals */
462 sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
465 * Find a socket with data available and call its
468 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
473 if (err != preverr) {
474 printk(KERN_WARNING "%s: unexpected error "
475 "from svc_recv (%d)\n", __func__, -err);
478 schedule_timeout_uninterruptible(HZ);
482 update_thread_usage(atomic_read(&nfsd_busy));
483 atomic_inc(&nfsd_busy);
485 /* Lock the export hash tables for reading. */
488 /* Process request with signals blocked. */
489 sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
493 /* Unlock export hash tables */
495 update_thread_usage(atomic_read(&nfsd_busy));
496 atomic_dec(&nfsd_busy);
499 /* Clear signals before calling svc_exit_thread() */
500 flush_signals(current);
502 mutex_lock(&nfsd_mutex);
506 /* Release the thread */
507 svc_exit_thread(rqstp);
510 mutex_unlock(&nfsd_mutex);
511 module_put_and_exit(0);
515 static __be32 map_new_errors(u32 vers, __be32 nfserr)
517 if (nfserr == nfserr_jukebox && vers == 2)
518 return nfserr_dropit;
519 if (nfserr == nfserr_wrongsec && vers < 4)
525 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
527 struct svc_procedure *proc;
532 dprintk("nfsd_dispatch: vers %d proc %d\n",
533 rqstp->rq_vers, rqstp->rq_proc);
534 proc = rqstp->rq_procinfo;
536 /* Check whether we have this call in the cache. */
537 switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
547 /* Decode arguments */
548 xdr = proc->pc_decode;
549 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
551 dprintk("nfsd: failed to decode arguments!\n");
552 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
553 *statp = rpc_garbage_args;
557 /* need to grab the location to store the status, as
558 * nfsv4 does some encoding while processing
560 nfserrp = rqstp->rq_res.head[0].iov_base
561 + rqstp->rq_res.head[0].iov_len;
562 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
564 /* Now call the procedure handler, and encode NFS status. */
565 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
566 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
567 if (nfserr == nfserr_dropit) {
568 dprintk("nfsd: Dropping request; may be revisited later\n");
569 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
573 if (rqstp->rq_proc != 0)
577 * For NFSv2, additional info is never returned in case of an error.
579 if (!(nfserr && rqstp->rq_vers == 2)) {
580 xdr = proc->pc_encode;
581 if (xdr && !xdr(rqstp, nfserrp,
583 /* Failed to encode result. Release cache entry */
584 dprintk("nfsd: failed to encode result!\n");
585 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
586 *statp = rpc_system_err;
591 /* Store reply in cache. */
592 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);