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>
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/fs_struct.h>
24 #include <linux/sunrpc/types.h>
25 #include <linux/sunrpc/stats.h>
26 #include <linux/sunrpc/svc.h>
27 #include <linux/sunrpc/svcsock.h>
28 #include <linux/sunrpc/cache.h>
29 #include <linux/nfsd/nfsd.h>
30 #include <linux/nfsd/stats.h>
31 #include <linux/nfsd/cache.h>
32 #include <linux/nfsd/syscall.h>
33 #include <linux/lockd/bind.h>
34 #include <linux/nfsacl.h>
36 #define NFSDDBG_FACILITY NFSDDBG_SVC
38 /* these signals will be delivered to an nfsd thread
39 * when handling a request
41 #define ALLOWED_SIGS (sigmask(SIGKILL))
42 /* these signals will be delivered to an nfsd thread
43 * when not handling a request. i.e. when waiting
45 #define SHUTDOWN_SIGS (sigmask(SIGKILL) | sigmask(SIGHUP) | sigmask(SIGINT) | sigmask(SIGQUIT))
46 /* if the last thread dies with SIGHUP, then the exports table is
47 * left unchanged ( like 2.4-{0-9} ). Any other signal will clear
48 * the exports table (like 2.2).
50 #define SIG_NOCLEAN SIGHUP
52 extern struct svc_program nfsd_program;
53 static void nfsd(struct svc_rqst *rqstp);
54 struct timeval nfssvc_boot;
55 struct svc_serv *nfsd_serv;
56 static atomic_t nfsd_busy;
57 static unsigned long nfsd_last_call;
58 static DEFINE_SPINLOCK(nfsd_call_lock);
60 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
61 static struct svc_stat nfsd_acl_svcstats;
62 static struct svc_version * nfsd_acl_version[] = {
63 [2] = &nfsd_acl_version2,
64 [3] = &nfsd_acl_version3,
67 #define NFSD_ACL_MINVERS 2
68 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
69 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
71 static struct svc_program nfsd_acl_program = {
72 .pg_prog = NFS_ACL_PROGRAM,
73 .pg_nvers = NFSD_ACL_NRVERS,
74 .pg_vers = nfsd_acl_versions,
77 .pg_stats = &nfsd_acl_svcstats,
78 .pg_authenticate = &svc_set_client,
81 static struct svc_stat nfsd_acl_svcstats = {
82 .program = &nfsd_acl_program,
84 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
86 static struct svc_version * nfsd_version[] = {
88 #if defined(CONFIG_NFSD_V3)
91 #if defined(CONFIG_NFSD_V4)
96 #define NFSD_MINVERS 2
97 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
98 static struct svc_version *nfsd_versions[NFSD_NRVERS];
100 struct svc_program nfsd_program = {
101 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
102 .pg_next = &nfsd_acl_program,
104 .pg_prog = NFS_PROGRAM, /* program number */
105 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
106 .pg_vers = nfsd_versions, /* version table */
107 .pg_name = "nfsd", /* program name */
108 .pg_class = "nfsd", /* authentication class */
109 .pg_stats = &nfsd_svcstats, /* version table */
110 .pg_authenticate = &svc_set_client, /* export authentication */
114 int nfsd_vers(int vers, enum vers_op change)
116 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
120 nfsd_versions[vers] = nfsd_version[vers];
121 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
122 if (vers < NFSD_ACL_NRVERS)
123 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
127 nfsd_versions[vers] = NULL;
128 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
129 if (vers < NFSD_ACL_NRVERS)
130 nfsd_acl_versions[vers] = NULL;
134 return nfsd_versions[vers] != NULL;
136 return nfsd_version[vers] != NULL;
141 * Maximum number of nfsd processes
143 #define NFSD_MAXSERVS 8192
145 int nfsd_nrthreads(void)
147 if (nfsd_serv == NULL)
150 return nfsd_serv->sv_nrthreads;
153 static int killsig; /* signal that was used to kill last nfsd */
154 static void nfsd_last_thread(struct svc_serv *serv)
156 /* When last nfsd thread exits we need to do some clean-up */
157 struct svc_sock *svsk;
158 list_for_each_entry(svsk, &serv->sv_permsocks, sk_list)
161 nfsd_racache_shutdown();
162 nfs4_state_shutdown();
164 printk(KERN_WARNING "nfsd: last server has exited\n");
165 if (killsig != SIG_NOCLEAN) {
166 printk(KERN_WARNING "nfsd: unexporting all filesystems\n");
171 void nfsd_reset_versions(void)
176 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
177 if (nfsd_program.pg_vers[i])
182 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
183 nfsd_program.pg_vers[i] = nfsd_version[i];
184 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
185 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
186 nfsd_acl_program.pg_vers[i] =
192 int nfsd_create_serv(void)
201 if (nfsd_max_blksize == 0) {
202 /* choose a suitable default */
205 /* Aim for 1/4096 of memory per thread
206 * This gives 1MB on 4Gig machines
207 * But only uses 32K on 128M machines.
208 * Bottom out at 8K on 32M and smaller.
209 * Of course, this is only a default.
211 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
212 i.totalram <<= PAGE_SHIFT - 12;
213 while (nfsd_max_blksize > i.totalram &&
214 nfsd_max_blksize >= 8*1024*2)
215 nfsd_max_blksize /= 2;
218 atomic_set(&nfsd_busy, 0);
219 nfsd_serv = svc_create_pooled(&nfsd_program,
222 nfsd, SIG_NOCLEAN, THIS_MODULE);
223 if (nfsd_serv == NULL)
226 do_gettimeofday(&nfssvc_boot); /* record boot time */
230 static int nfsd_init_socks(int port)
233 if (!list_empty(&nfsd_serv->sv_permsocks))
236 error = lockd_up(IPPROTO_UDP);
238 error = svc_makesock(nfsd_serv, IPPROTO_UDP, port);
245 #ifdef CONFIG_NFSD_TCP
246 error = lockd_up(IPPROTO_TCP);
248 error = svc_makesock(nfsd_serv, IPPROTO_TCP, port);
258 int nfsd_nrpools(void)
260 if (nfsd_serv == NULL)
263 return nfsd_serv->sv_nrpools;
266 int nfsd_get_nrthreads(int n, int *nthreads)
270 if (nfsd_serv != NULL) {
271 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
272 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
278 int nfsd_set_nrthreads(int n, int *nthreads)
284 if (nfsd_serv == NULL || n <= 0)
287 if (n > nfsd_serv->sv_nrpools)
288 n = nfsd_serv->sv_nrpools;
290 /* enforce a global maximum number of threads */
292 for (i = 0; i < n; i++) {
293 if (nthreads[i] > NFSD_MAXSERVS)
294 nthreads[i] = NFSD_MAXSERVS;
297 if (tot > NFSD_MAXSERVS) {
298 /* total too large: scale down requested numbers */
299 for (i = 0; i < n && tot > 0; i++) {
300 int new = nthreads[i] * NFSD_MAXSERVS / tot;
301 tot -= (nthreads[i] - new);
304 for (i = 0; i < n && tot > 0; i++) {
311 * There must always be a thread in pool 0; the admin
312 * can't shut down NFS completely using pool_threads.
314 if (nthreads[0] == 0)
317 /* apply the new numbers */
320 for (i = 0; i < n; i++) {
321 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
326 svc_destroy(nfsd_serv);
333 nfsd_svc(unsigned short port, int nrservs)
338 dprintk("nfsd: creating service\n");
342 if (nrservs > NFSD_MAXSERVS)
343 nrservs = NFSD_MAXSERVS;
345 /* Readahead param cache - will no-op if it already exists */
346 error = nfsd_racache_init(2*nrservs);
349 error = nfs4_state_start();
353 nfsd_reset_versions();
355 error = nfsd_create_serv();
359 error = nfsd_init_socks(port);
363 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
365 svc_destroy(nfsd_serv); /* Release server */
372 update_thread_usage(int busy_threads)
374 unsigned long prev_call;
378 spin_lock(&nfsd_call_lock);
379 prev_call = nfsd_last_call;
380 nfsd_last_call = jiffies;
381 decile = busy_threads*10/nfsdstats.th_cnt;
382 if (decile>0 && decile <= 10) {
383 diff = nfsd_last_call - prev_call;
384 if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
385 nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
387 nfsdstats.th_fullcnt++;
389 spin_unlock(&nfsd_call_lock);
393 * This is the NFS server kernel thread
396 nfsd(struct svc_rqst *rqstp)
398 struct fs_struct *fsp;
400 sigset_t shutdown_mask, allowed_mask;
402 /* Lock module and set up kernel thread */
406 /* After daemonize() this kernel thread shares current->fs
407 * with the init process. We need to create files with a
408 * umask of 0 instead of init's umask. */
409 fsp = copy_fs_struct(current->fs);
411 printk("Unable to start nfsd thread: out of memory\n");
416 current->fs->umask = 0;
418 siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS);
419 siginitsetinv(&allowed_mask, ALLOWED_SIGS);
423 rqstp->rq_task = current;
428 * We want less throttling in balance_dirty_pages() so that nfs to
429 * localhost doesn't cause nfsd to lock up due to all the client's
432 current->flags |= PF_LESS_THROTTLE;
435 * The main request loop
438 /* Block all but the shutdown signals */
439 sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
442 * Find a socket with data available and call its
445 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
449 update_thread_usage(atomic_read(&nfsd_busy));
450 atomic_inc(&nfsd_busy);
452 /* Lock the export hash tables for reading. */
455 /* Process request with signals blocked. */
456 sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
460 /* Unlock export hash tables */
462 update_thread_usage(atomic_read(&nfsd_busy));
463 atomic_dec(&nfsd_busy);
467 printk(KERN_WARNING "nfsd: terminating on error %d\n", -err);
471 for (signo = 1; signo <= _NSIG; signo++)
472 if (sigismember(¤t->pending.signal, signo) &&
473 !sigismember(¤t->blocked, signo))
477 /* Clear signals before calling svc_exit_thread() */
478 flush_signals(current);
485 /* Release the thread */
486 svc_exit_thread(rqstp);
490 module_put_and_exit(0);
494 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
496 struct svc_procedure *proc;
501 dprintk("nfsd_dispatch: vers %d proc %d\n",
502 rqstp->rq_vers, rqstp->rq_proc);
503 proc = rqstp->rq_procinfo;
505 /* Check whether we have this call in the cache. */
506 switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
516 /* Decode arguments */
517 xdr = proc->pc_decode;
518 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
520 dprintk("nfsd: failed to decode arguments!\n");
521 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
522 *statp = rpc_garbage_args;
526 /* need to grab the location to store the status, as
527 * nfsv4 does some encoding while processing
529 nfserrp = rqstp->rq_res.head[0].iov_base
530 + rqstp->rq_res.head[0].iov_len;
531 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
533 /* Now call the procedure handler, and encode NFS status. */
534 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
535 if (nfserr == nfserr_jukebox && rqstp->rq_vers == 2)
536 nfserr = nfserr_dropit;
537 if (nfserr == nfserr_dropit) {
538 dprintk("nfsd: Dropping request due to malloc failure!\n");
539 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
543 if (rqstp->rq_proc != 0)
547 * For NFSv2, additional info is never returned in case of an error.
549 if (!(nfserr && rqstp->rq_vers == 2)) {
550 xdr = proc->pc_encode;
551 if (xdr && !xdr(rqstp, nfserrp,
553 /* Failed to encode result. Release cache entry */
554 dprintk("nfsd: failed to encode result!\n");
555 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
556 *statp = rpc_system_err;
561 /* Store reply in cache. */
562 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);