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 extern struct svc_program nfsd_program;
41 static int nfsd(void *vrqstp);
42 struct timeval nfssvc_boot;
43 static atomic_t nfsd_busy;
44 static unsigned long nfsd_last_call;
45 static DEFINE_SPINLOCK(nfsd_call_lock);
48 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
49 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
50 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
52 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
53 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
54 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
55 * entry of ->sv_pools[].
57 * Transitions of the thread count between zero and non-zero are of particular
58 * interest since the svc_serv needs to be created and initialized at that
61 * Finally, the nfsd_mutex also protects some of the global variables that are
62 * accessed when nfsd starts and that are settable via the write_* routines in
63 * nfsctl.c. In particular:
65 * user_recovery_dirname
69 DEFINE_MUTEX(nfsd_mutex);
70 struct svc_serv *nfsd_serv;
72 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
73 static struct svc_stat nfsd_acl_svcstats;
74 static struct svc_version * nfsd_acl_version[] = {
75 [2] = &nfsd_acl_version2,
76 [3] = &nfsd_acl_version3,
79 #define NFSD_ACL_MINVERS 2
80 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
81 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
83 static struct svc_program nfsd_acl_program = {
84 .pg_prog = NFS_ACL_PROGRAM,
85 .pg_nvers = NFSD_ACL_NRVERS,
86 .pg_vers = nfsd_acl_versions,
89 .pg_stats = &nfsd_acl_svcstats,
90 .pg_authenticate = &svc_set_client,
93 static struct svc_stat nfsd_acl_svcstats = {
94 .program = &nfsd_acl_program,
96 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
98 static struct svc_version * nfsd_version[] = {
100 #if defined(CONFIG_NFSD_V3)
101 [3] = &nfsd_version3,
103 #if defined(CONFIG_NFSD_V4)
104 [4] = &nfsd_version4,
108 #define NFSD_MINVERS 2
109 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
110 static struct svc_version *nfsd_versions[NFSD_NRVERS];
112 struct svc_program nfsd_program = {
113 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
114 .pg_next = &nfsd_acl_program,
116 .pg_prog = NFS_PROGRAM, /* program number */
117 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
118 .pg_vers = nfsd_versions, /* version table */
119 .pg_name = "nfsd", /* program name */
120 .pg_class = "nfsd", /* authentication class */
121 .pg_stats = &nfsd_svcstats, /* version table */
122 .pg_authenticate = &svc_set_client, /* export authentication */
126 int nfsd_vers(int vers, enum vers_op change)
128 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
132 nfsd_versions[vers] = nfsd_version[vers];
133 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
134 if (vers < NFSD_ACL_NRVERS)
135 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
139 nfsd_versions[vers] = NULL;
140 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
141 if (vers < NFSD_ACL_NRVERS)
142 nfsd_acl_versions[vers] = NULL;
146 return nfsd_versions[vers] != NULL;
148 return nfsd_version[vers] != NULL;
153 * Maximum number of nfsd processes
155 #define NFSD_MAXSERVS 8192
157 int nfsd_nrthreads(void)
160 mutex_lock(&nfsd_mutex);
162 rv = nfsd_serv->sv_nrthreads;
163 mutex_unlock(&nfsd_mutex);
167 static void nfsd_last_thread(struct svc_serv *serv)
169 /* When last nfsd thread exits we need to do some clean-up */
170 struct svc_xprt *xprt;
171 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list)
174 nfsd_racache_shutdown();
175 nfs4_state_shutdown();
177 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
182 void nfsd_reset_versions(void)
187 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
188 if (nfsd_program.pg_vers[i])
193 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
194 nfsd_program.pg_vers[i] = nfsd_version[i];
195 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
196 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
197 nfsd_acl_program.pg_vers[i] =
204 int nfsd_create_serv(void)
208 WARN_ON(!mutex_is_locked(&nfsd_mutex));
213 if (nfsd_max_blksize == 0) {
214 /* choose a suitable default */
217 /* Aim for 1/4096 of memory per thread
218 * This gives 1MB on 4Gig machines
219 * But only uses 32K on 128M machines.
220 * Bottom out at 8K on 32M and smaller.
221 * Of course, this is only a default.
223 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
224 i.totalram <<= PAGE_SHIFT - 12;
225 while (nfsd_max_blksize > i.totalram &&
226 nfsd_max_blksize >= 8*1024*2)
227 nfsd_max_blksize /= 2;
230 atomic_set(&nfsd_busy, 0);
231 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
232 nfsd_last_thread, nfsd, THIS_MODULE);
233 if (nfsd_serv == NULL)
236 do_gettimeofday(&nfssvc_boot); /* record boot time */
240 static int nfsd_init_socks(int port)
243 if (!list_empty(&nfsd_serv->sv_permsocks))
246 error = svc_create_xprt(nfsd_serv, "udp", PF_INET, port,
255 error = svc_create_xprt(nfsd_serv, "tcp", PF_INET, port,
267 int nfsd_nrpools(void)
269 if (nfsd_serv == NULL)
272 return nfsd_serv->sv_nrpools;
275 int nfsd_get_nrthreads(int n, int *nthreads)
279 if (nfsd_serv != NULL) {
280 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
281 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
287 int nfsd_set_nrthreads(int n, int *nthreads)
293 WARN_ON(!mutex_is_locked(&nfsd_mutex));
295 if (nfsd_serv == NULL || n <= 0)
298 if (n > nfsd_serv->sv_nrpools)
299 n = nfsd_serv->sv_nrpools;
301 /* enforce a global maximum number of threads */
303 for (i = 0; i < n; i++) {
304 if (nthreads[i] > NFSD_MAXSERVS)
305 nthreads[i] = NFSD_MAXSERVS;
308 if (tot > NFSD_MAXSERVS) {
309 /* total too large: scale down requested numbers */
310 for (i = 0; i < n && tot > 0; i++) {
311 int new = nthreads[i] * NFSD_MAXSERVS / tot;
312 tot -= (nthreads[i] - new);
315 for (i = 0; i < n && tot > 0; i++) {
322 * There must always be a thread in pool 0; the admin
323 * can't shut down NFS completely using pool_threads.
325 if (nthreads[0] == 0)
328 /* apply the new numbers */
330 for (i = 0; i < n; i++) {
331 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
336 svc_destroy(nfsd_serv);
342 nfsd_svc(unsigned short port, int nrservs)
346 mutex_lock(&nfsd_mutex);
347 dprintk("nfsd: creating service\n");
351 if (nrservs > NFSD_MAXSERVS)
352 nrservs = NFSD_MAXSERVS;
354 /* Readahead param cache - will no-op if it already exists */
355 error = nfsd_racache_init(2*nrservs);
360 nfsd_reset_versions();
362 error = nfsd_create_serv();
366 error = nfsd_init_socks(port);
370 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
372 svc_destroy(nfsd_serv); /* Release server */
374 mutex_unlock(&nfsd_mutex);
379 update_thread_usage(int busy_threads)
381 unsigned long prev_call;
385 spin_lock(&nfsd_call_lock);
386 prev_call = nfsd_last_call;
387 nfsd_last_call = jiffies;
388 decile = busy_threads*10/nfsdstats.th_cnt;
389 if (decile>0 && decile <= 10) {
390 diff = nfsd_last_call - prev_call;
391 if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
392 nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
394 nfsdstats.th_fullcnt++;
396 spin_unlock(&nfsd_call_lock);
400 * This is the NFS server kernel thread
405 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
406 int err, preverr = 0;
408 /* Lock module and set up kernel thread */
409 mutex_lock(&nfsd_mutex);
411 /* At this point, the thread shares current->fs
412 * with the init process. We need to create files with a
413 * umask of 0 instead of init's umask. */
414 if (unshare_fs_struct() < 0) {
415 printk("Unable to start nfsd thread: out of memory\n");
419 current->fs->umask = 0;
422 * thread is spawned with all signals set to SIG_IGN, re-enable
423 * the ones that will bring down the thread
425 allow_signal(SIGKILL);
426 allow_signal(SIGHUP);
427 allow_signal(SIGINT);
428 allow_signal(SIGQUIT);
431 mutex_unlock(&nfsd_mutex);
434 * We want less throttling in balance_dirty_pages() so that nfs to
435 * localhost doesn't cause nfsd to lock up due to all the client's
438 current->flags |= PF_LESS_THROTTLE;
442 * The main request loop
446 * Find a socket with data available and call its
449 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
454 if (err != preverr) {
455 printk(KERN_WARNING "%s: unexpected error "
456 "from svc_recv (%d)\n", __func__, -err);
459 schedule_timeout_uninterruptible(HZ);
463 update_thread_usage(atomic_read(&nfsd_busy));
464 atomic_inc(&nfsd_busy);
466 /* Lock the export hash tables for reading. */
471 /* Unlock export hash tables */
473 update_thread_usage(atomic_read(&nfsd_busy));
474 atomic_dec(&nfsd_busy);
477 /* Clear signals before calling svc_exit_thread() */
478 flush_signals(current);
480 mutex_lock(&nfsd_mutex);
484 /* Release the thread */
485 svc_exit_thread(rqstp);
488 mutex_unlock(&nfsd_mutex);
489 module_put_and_exit(0);
493 static __be32 map_new_errors(u32 vers, __be32 nfserr)
495 if (nfserr == nfserr_jukebox && vers == 2)
496 return nfserr_dropit;
497 if (nfserr == nfserr_wrongsec && vers < 4)
503 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
505 struct svc_procedure *proc;
510 dprintk("nfsd_dispatch: vers %d proc %d\n",
511 rqstp->rq_vers, rqstp->rq_proc);
512 proc = rqstp->rq_procinfo;
514 /* Check whether we have this call in the cache. */
515 switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
525 /* Decode arguments */
526 xdr = proc->pc_decode;
527 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
529 dprintk("nfsd: failed to decode arguments!\n");
530 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
531 *statp = rpc_garbage_args;
535 /* need to grab the location to store the status, as
536 * nfsv4 does some encoding while processing
538 nfserrp = rqstp->rq_res.head[0].iov_base
539 + rqstp->rq_res.head[0].iov_len;
540 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
542 /* Now call the procedure handler, and encode NFS status. */
543 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
544 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
545 if (nfserr == nfserr_dropit) {
546 dprintk("nfsd: Dropping request; may be revisited later\n");
547 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
551 if (rqstp->rq_proc != 0)
555 * For NFSv2, additional info is never returned in case of an error.
557 if (!(nfserr && rqstp->rq_vers == 2)) {
558 xdr = proc->pc_encode;
559 if (xdr && !xdr(rqstp, nfserrp,
561 /* Failed to encode result. Release cache entry */
562 dprintk("nfsd: failed to encode result!\n");
563 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
564 *statp = rpc_system_err;
569 /* Store reply in cache. */
570 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);