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>
25 #include <linux/sunrpc/types.h>
26 #include <linux/sunrpc/stats.h>
27 #include <linux/sunrpc/svc.h>
28 #include <linux/sunrpc/svcsock.h>
29 #include <linux/sunrpc/cache.h>
30 #include <linux/nfsd/nfsd.h>
31 #include <linux/nfsd/stats.h>
32 #include <linux/nfsd/cache.h>
33 #include <linux/nfsd/syscall.h>
34 #include <linux/lockd/bind.h>
35 #include <linux/nfsacl.h>
37 #define NFSDDBG_FACILITY NFSDDBG_SVC
39 /* these signals will be delivered to an nfsd thread
40 * when handling a request
42 #define ALLOWED_SIGS (sigmask(SIGKILL))
43 /* these signals will be delivered to an nfsd thread
44 * when not handling a request. i.e. when waiting
46 #define SHUTDOWN_SIGS (sigmask(SIGKILL) | sigmask(SIGHUP) | sigmask(SIGINT) | sigmask(SIGQUIT))
47 /* if the last thread dies with SIGHUP, then the exports table is
48 * left unchanged ( like 2.4-{0-9} ). Any other signal will clear
49 * the exports table (like 2.2).
51 #define SIG_NOCLEAN SIGHUP
53 extern struct svc_program nfsd_program;
54 static void nfsd(struct svc_rqst *rqstp);
55 struct timeval nfssvc_boot;
56 static atomic_t nfsd_busy;
57 static unsigned long nfsd_last_call;
58 static DEFINE_SPINLOCK(nfsd_call_lock);
61 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
62 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
63 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
65 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
66 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
67 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
68 * entry of ->sv_pools[].
70 * Transitions of the thread count between zero and non-zero are of particular
71 * interest since the svc_serv needs to be created and initialized at that
74 * Finally, the nfsd_mutex also protects some of the global variables that are
75 * accessed when nfsd starts and that are settable via the write_* routines in
76 * nfsctl.c. In particular:
78 * user_recovery_dirname
82 DEFINE_MUTEX(nfsd_mutex);
83 struct svc_serv *nfsd_serv;
85 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
86 static struct svc_stat nfsd_acl_svcstats;
87 static struct svc_version * nfsd_acl_version[] = {
88 [2] = &nfsd_acl_version2,
89 [3] = &nfsd_acl_version3,
92 #define NFSD_ACL_MINVERS 2
93 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
94 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
96 static struct svc_program nfsd_acl_program = {
97 .pg_prog = NFS_ACL_PROGRAM,
98 .pg_nvers = NFSD_ACL_NRVERS,
99 .pg_vers = nfsd_acl_versions,
102 .pg_stats = &nfsd_acl_svcstats,
103 .pg_authenticate = &svc_set_client,
106 static struct svc_stat nfsd_acl_svcstats = {
107 .program = &nfsd_acl_program,
109 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
111 static struct svc_version * nfsd_version[] = {
112 [2] = &nfsd_version2,
113 #if defined(CONFIG_NFSD_V3)
114 [3] = &nfsd_version3,
116 #if defined(CONFIG_NFSD_V4)
117 [4] = &nfsd_version4,
121 #define NFSD_MINVERS 2
122 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
123 static struct svc_version *nfsd_versions[NFSD_NRVERS];
125 struct svc_program nfsd_program = {
126 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
127 .pg_next = &nfsd_acl_program,
129 .pg_prog = NFS_PROGRAM, /* program number */
130 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
131 .pg_vers = nfsd_versions, /* version table */
132 .pg_name = "nfsd", /* program name */
133 .pg_class = "nfsd", /* authentication class */
134 .pg_stats = &nfsd_svcstats, /* version table */
135 .pg_authenticate = &svc_set_client, /* export authentication */
139 int nfsd_vers(int vers, enum vers_op change)
141 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
145 nfsd_versions[vers] = nfsd_version[vers];
146 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
147 if (vers < NFSD_ACL_NRVERS)
148 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
152 nfsd_versions[vers] = NULL;
153 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
154 if (vers < NFSD_ACL_NRVERS)
155 nfsd_acl_versions[vers] = NULL;
159 return nfsd_versions[vers] != NULL;
161 return nfsd_version[vers] != NULL;
166 * Maximum number of nfsd processes
168 #define NFSD_MAXSERVS 8192
170 int nfsd_nrthreads(void)
172 if (nfsd_serv == NULL)
175 return nfsd_serv->sv_nrthreads;
178 static int killsig; /* signal that was used to kill last nfsd */
179 static void nfsd_last_thread(struct svc_serv *serv)
181 /* When last nfsd thread exits we need to do some clean-up */
182 struct svc_xprt *xprt;
183 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list)
186 nfsd_racache_shutdown();
187 nfs4_state_shutdown();
189 printk(KERN_WARNING "nfsd: last server has exited\n");
190 if (killsig != SIG_NOCLEAN) {
191 printk(KERN_WARNING "nfsd: unexporting all filesystems\n");
196 void nfsd_reset_versions(void)
201 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
202 if (nfsd_program.pg_vers[i])
207 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
208 nfsd_program.pg_vers[i] = nfsd_version[i];
209 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
210 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
211 nfsd_acl_program.pg_vers[i] =
218 int nfsd_create_serv(void)
222 WARN_ON(!mutex_is_locked(&nfsd_mutex));
227 if (nfsd_max_blksize == 0) {
228 /* choose a suitable default */
231 /* Aim for 1/4096 of memory per thread
232 * This gives 1MB on 4Gig machines
233 * But only uses 32K on 128M machines.
234 * Bottom out at 8K on 32M and smaller.
235 * Of course, this is only a default.
237 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
238 i.totalram <<= PAGE_SHIFT - 12;
239 while (nfsd_max_blksize > i.totalram &&
240 nfsd_max_blksize >= 8*1024*2)
241 nfsd_max_blksize /= 2;
244 atomic_set(&nfsd_busy, 0);
245 nfsd_serv = svc_create_pooled(&nfsd_program,
248 nfsd, SIG_NOCLEAN, THIS_MODULE);
249 if (nfsd_serv == NULL)
252 do_gettimeofday(&nfssvc_boot); /* record boot time */
256 static int nfsd_init_socks(int port)
259 if (!list_empty(&nfsd_serv->sv_permsocks))
262 error = lockd_up(IPPROTO_UDP);
264 error = svc_create_xprt(nfsd_serv, "udp", port,
272 error = lockd_up(IPPROTO_TCP);
274 error = svc_create_xprt(nfsd_serv, "tcp", port,
284 int nfsd_nrpools(void)
286 if (nfsd_serv == NULL)
289 return nfsd_serv->sv_nrpools;
292 int nfsd_get_nrthreads(int n, int *nthreads)
296 if (nfsd_serv != NULL) {
297 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
298 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
304 int nfsd_set_nrthreads(int n, int *nthreads)
310 WARN_ON(!mutex_is_locked(&nfsd_mutex));
312 if (nfsd_serv == NULL || n <= 0)
315 if (n > nfsd_serv->sv_nrpools)
316 n = nfsd_serv->sv_nrpools;
318 /* enforce a global maximum number of threads */
320 for (i = 0; i < n; i++) {
321 if (nthreads[i] > NFSD_MAXSERVS)
322 nthreads[i] = NFSD_MAXSERVS;
325 if (tot > NFSD_MAXSERVS) {
326 /* total too large: scale down requested numbers */
327 for (i = 0; i < n && tot > 0; i++) {
328 int new = nthreads[i] * NFSD_MAXSERVS / tot;
329 tot -= (nthreads[i] - new);
332 for (i = 0; i < n && tot > 0; i++) {
339 * There must always be a thread in pool 0; the admin
340 * can't shut down NFS completely using pool_threads.
342 if (nthreads[0] == 0)
345 /* apply the new numbers */
347 for (i = 0; i < n; i++) {
348 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
353 svc_destroy(nfsd_serv);
359 nfsd_svc(unsigned short port, int nrservs)
363 mutex_lock(&nfsd_mutex);
364 dprintk("nfsd: creating service\n");
368 if (nrservs > NFSD_MAXSERVS)
369 nrservs = NFSD_MAXSERVS;
371 /* Readahead param cache - will no-op if it already exists */
372 error = nfsd_racache_init(2*nrservs);
377 nfsd_reset_versions();
379 error = nfsd_create_serv();
383 error = nfsd_init_socks(port);
387 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
389 svc_destroy(nfsd_serv); /* Release server */
391 mutex_unlock(&nfsd_mutex);
396 update_thread_usage(int busy_threads)
398 unsigned long prev_call;
402 spin_lock(&nfsd_call_lock);
403 prev_call = nfsd_last_call;
404 nfsd_last_call = jiffies;
405 decile = busy_threads*10/nfsdstats.th_cnt;
406 if (decile>0 && decile <= 10) {
407 diff = nfsd_last_call - prev_call;
408 if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
409 nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
411 nfsdstats.th_fullcnt++;
413 spin_unlock(&nfsd_call_lock);
417 * This is the NFS server kernel thread
420 nfsd(struct svc_rqst *rqstp)
422 struct fs_struct *fsp;
424 sigset_t shutdown_mask, allowed_mask;
426 /* Lock module and set up kernel thread */
427 mutex_lock(&nfsd_mutex);
430 /* After daemonize() this kernel thread shares current->fs
431 * with the init process. We need to create files with a
432 * umask of 0 instead of init's umask. */
433 fsp = copy_fs_struct(current->fs);
435 printk("Unable to start nfsd thread: out of memory\n");
440 current->fs->umask = 0;
442 siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS);
443 siginitsetinv(&allowed_mask, ALLOWED_SIGS);
448 rqstp->rq_task = current;
450 mutex_unlock(&nfsd_mutex);
454 * We want less throttling in balance_dirty_pages() so that nfs to
455 * localhost doesn't cause nfsd to lock up due to all the client's
458 current->flags |= PF_LESS_THROTTLE;
462 * The main request loop
465 /* Block all but the shutdown signals */
466 sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
469 * Find a socket with data available and call its
472 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
476 update_thread_usage(atomic_read(&nfsd_busy));
477 atomic_inc(&nfsd_busy);
479 /* Lock the export hash tables for reading. */
482 /* Process request with signals blocked. */
483 sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
487 /* Unlock export hash tables */
489 update_thread_usage(atomic_read(&nfsd_busy));
490 atomic_dec(&nfsd_busy);
494 printk(KERN_WARNING "nfsd: terminating on error %d\n", -err);
498 for (signo = 1; signo <= _NSIG; signo++)
499 if (sigismember(¤t->pending.signal, signo) &&
500 !sigismember(¤t->blocked, signo))
504 /* Clear signals before calling svc_exit_thread() */
505 flush_signals(current);
507 mutex_lock(&nfsd_mutex);
512 /* Release the thread */
513 svc_exit_thread(rqstp);
516 mutex_unlock(&nfsd_mutex);
517 module_put_and_exit(0);
520 static __be32 map_new_errors(u32 vers, __be32 nfserr)
522 if (nfserr == nfserr_jukebox && vers == 2)
523 return nfserr_dropit;
524 if (nfserr == nfserr_wrongsec && vers < 4)
530 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
532 struct svc_procedure *proc;
537 dprintk("nfsd_dispatch: vers %d proc %d\n",
538 rqstp->rq_vers, rqstp->rq_proc);
539 proc = rqstp->rq_procinfo;
541 /* Check whether we have this call in the cache. */
542 switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
552 /* Decode arguments */
553 xdr = proc->pc_decode;
554 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
556 dprintk("nfsd: failed to decode arguments!\n");
557 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
558 *statp = rpc_garbage_args;
562 /* need to grab the location to store the status, as
563 * nfsv4 does some encoding while processing
565 nfserrp = rqstp->rq_res.head[0].iov_base
566 + rqstp->rq_res.head[0].iov_len;
567 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
569 /* Now call the procedure handler, and encode NFS status. */
570 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
571 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
572 if (nfserr == nfserr_dropit) {
573 dprintk("nfsd: Dropping request; may be revisited later\n");
574 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
578 if (rqstp->rq_proc != 0)
582 * For NFSv2, additional info is never returned in case of an error.
584 if (!(nfserr && rqstp->rq_vers == 2)) {
585 xdr = proc->pc_encode;
586 if (xdr && !xdr(rqstp, nfserrp,
588 /* Failed to encode result. Release cache entry */
589 dprintk("nfsd: failed to encode result!\n");
590 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
591 *statp = rpc_system_err;
596 /* Store reply in cache. */
597 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);