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 DEFINE_MUTEX(nfsd_mutex);
75 struct svc_serv *nfsd_serv;
77 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
78 static struct svc_stat nfsd_acl_svcstats;
79 static struct svc_version * nfsd_acl_version[] = {
80 [2] = &nfsd_acl_version2,
81 [3] = &nfsd_acl_version3,
84 #define NFSD_ACL_MINVERS 2
85 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
86 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
88 static struct svc_program nfsd_acl_program = {
89 .pg_prog = NFS_ACL_PROGRAM,
90 .pg_nvers = NFSD_ACL_NRVERS,
91 .pg_vers = nfsd_acl_versions,
94 .pg_stats = &nfsd_acl_svcstats,
95 .pg_authenticate = &svc_set_client,
98 static struct svc_stat nfsd_acl_svcstats = {
99 .program = &nfsd_acl_program,
101 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
103 static struct svc_version * nfsd_version[] = {
104 [2] = &nfsd_version2,
105 #if defined(CONFIG_NFSD_V3)
106 [3] = &nfsd_version3,
108 #if defined(CONFIG_NFSD_V4)
109 [4] = &nfsd_version4,
113 #define NFSD_MINVERS 2
114 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
115 static struct svc_version *nfsd_versions[NFSD_NRVERS];
117 struct svc_program nfsd_program = {
118 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
119 .pg_next = &nfsd_acl_program,
121 .pg_prog = NFS_PROGRAM, /* program number */
122 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
123 .pg_vers = nfsd_versions, /* version table */
124 .pg_name = "nfsd", /* program name */
125 .pg_class = "nfsd", /* authentication class */
126 .pg_stats = &nfsd_svcstats, /* version table */
127 .pg_authenticate = &svc_set_client, /* export authentication */
131 int nfsd_vers(int vers, enum vers_op change)
133 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
137 nfsd_versions[vers] = nfsd_version[vers];
138 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
139 if (vers < NFSD_ACL_NRVERS)
140 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
144 nfsd_versions[vers] = NULL;
145 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
146 if (vers < NFSD_ACL_NRVERS)
147 nfsd_acl_versions[vers] = NULL;
151 return nfsd_versions[vers] != NULL;
153 return nfsd_version[vers] != NULL;
158 * Maximum number of nfsd processes
160 #define NFSD_MAXSERVS 8192
162 int nfsd_nrthreads(void)
164 if (nfsd_serv == NULL)
167 return nfsd_serv->sv_nrthreads;
170 static int killsig; /* signal that was used to kill last nfsd */
171 static void nfsd_last_thread(struct svc_serv *serv)
173 /* When last nfsd thread exits we need to do some clean-up */
174 struct svc_xprt *xprt;
175 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list)
178 nfsd_racache_shutdown();
179 nfs4_state_shutdown();
181 printk(KERN_WARNING "nfsd: last server has exited\n");
182 if (killsig != SIG_NOCLEAN) {
183 printk(KERN_WARNING "nfsd: unexporting all filesystems\n");
188 void nfsd_reset_versions(void)
193 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
194 if (nfsd_program.pg_vers[i])
199 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
200 nfsd_program.pg_vers[i] = nfsd_version[i];
201 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
202 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
203 nfsd_acl_program.pg_vers[i] =
210 int nfsd_create_serv(void)
214 WARN_ON(!mutex_is_locked(&nfsd_mutex));
219 if (nfsd_max_blksize == 0) {
220 /* choose a suitable default */
223 /* Aim for 1/4096 of memory per thread
224 * This gives 1MB on 4Gig machines
225 * But only uses 32K on 128M machines.
226 * Bottom out at 8K on 32M and smaller.
227 * Of course, this is only a default.
229 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
230 i.totalram <<= PAGE_SHIFT - 12;
231 while (nfsd_max_blksize > i.totalram &&
232 nfsd_max_blksize >= 8*1024*2)
233 nfsd_max_blksize /= 2;
236 atomic_set(&nfsd_busy, 0);
237 nfsd_serv = svc_create_pooled(&nfsd_program,
240 nfsd, SIG_NOCLEAN, THIS_MODULE);
241 if (nfsd_serv == NULL)
244 do_gettimeofday(&nfssvc_boot); /* record boot time */
248 static int nfsd_init_socks(int port)
251 if (!list_empty(&nfsd_serv->sv_permsocks))
254 error = lockd_up(IPPROTO_UDP);
256 error = svc_create_xprt(nfsd_serv, "udp", port,
264 error = lockd_up(IPPROTO_TCP);
266 error = svc_create_xprt(nfsd_serv, "tcp", port,
276 int nfsd_nrpools(void)
278 if (nfsd_serv == NULL)
281 return nfsd_serv->sv_nrpools;
284 int nfsd_get_nrthreads(int n, int *nthreads)
288 if (nfsd_serv != NULL) {
289 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
290 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
296 int nfsd_set_nrthreads(int n, int *nthreads)
302 WARN_ON(!mutex_is_locked(&nfsd_mutex));
304 if (nfsd_serv == NULL || n <= 0)
307 if (n > nfsd_serv->sv_nrpools)
308 n = nfsd_serv->sv_nrpools;
310 /* enforce a global maximum number of threads */
312 for (i = 0; i < n; i++) {
313 if (nthreads[i] > NFSD_MAXSERVS)
314 nthreads[i] = NFSD_MAXSERVS;
317 if (tot > NFSD_MAXSERVS) {
318 /* total too large: scale down requested numbers */
319 for (i = 0; i < n && tot > 0; i++) {
320 int new = nthreads[i] * NFSD_MAXSERVS / tot;
321 tot -= (nthreads[i] - new);
324 for (i = 0; i < n && tot > 0; i++) {
331 * There must always be a thread in pool 0; the admin
332 * can't shut down NFS completely using pool_threads.
334 if (nthreads[0] == 0)
337 /* apply the new numbers */
339 for (i = 0; i < n; i++) {
340 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
345 svc_destroy(nfsd_serv);
351 nfsd_svc(unsigned short port, int nrservs)
355 mutex_lock(&nfsd_mutex);
356 dprintk("nfsd: creating service\n");
360 if (nrservs > NFSD_MAXSERVS)
361 nrservs = NFSD_MAXSERVS;
363 /* Readahead param cache - will no-op if it already exists */
364 error = nfsd_racache_init(2*nrservs);
369 nfsd_reset_versions();
371 error = nfsd_create_serv();
375 error = nfsd_init_socks(port);
379 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
381 svc_destroy(nfsd_serv); /* Release server */
383 mutex_unlock(&nfsd_mutex);
388 update_thread_usage(int busy_threads)
390 unsigned long prev_call;
394 spin_lock(&nfsd_call_lock);
395 prev_call = nfsd_last_call;
396 nfsd_last_call = jiffies;
397 decile = busy_threads*10/nfsdstats.th_cnt;
398 if (decile>0 && decile <= 10) {
399 diff = nfsd_last_call - prev_call;
400 if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
401 nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
403 nfsdstats.th_fullcnt++;
405 spin_unlock(&nfsd_call_lock);
409 * This is the NFS server kernel thread
412 nfsd(struct svc_rqst *rqstp)
414 struct fs_struct *fsp;
416 sigset_t shutdown_mask, allowed_mask;
418 /* Lock module and set up kernel thread */
419 mutex_lock(&nfsd_mutex);
422 /* After daemonize() this kernel 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);
440 rqstp->rq_task = current;
442 mutex_unlock(&nfsd_mutex);
446 * We want less throttling in balance_dirty_pages() so that nfs to
447 * localhost doesn't cause nfsd to lock up due to all the client's
450 current->flags |= PF_LESS_THROTTLE;
454 * The main request loop
457 /* Block all but the shutdown signals */
458 sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
461 * Find a socket with data available and call its
464 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
468 update_thread_usage(atomic_read(&nfsd_busy));
469 atomic_inc(&nfsd_busy);
471 /* Lock the export hash tables for reading. */
474 /* Process request with signals blocked. */
475 sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
479 /* Unlock export hash tables */
481 update_thread_usage(atomic_read(&nfsd_busy));
482 atomic_dec(&nfsd_busy);
486 printk(KERN_WARNING "nfsd: terminating on error %d\n", -err);
490 for (signo = 1; signo <= _NSIG; signo++)
491 if (sigismember(¤t->pending.signal, signo) &&
492 !sigismember(¤t->blocked, signo))
496 /* Clear signals before calling svc_exit_thread() */
497 flush_signals(current);
499 mutex_lock(&nfsd_mutex);
504 /* Release the thread */
505 svc_exit_thread(rqstp);
508 mutex_unlock(&nfsd_mutex);
509 module_put_and_exit(0);
512 static __be32 map_new_errors(u32 vers, __be32 nfserr)
514 if (nfserr == nfserr_jukebox && vers == 2)
515 return nfserr_dropit;
516 if (nfserr == nfserr_wrongsec && vers < 4)
522 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
524 struct svc_procedure *proc;
529 dprintk("nfsd_dispatch: vers %d proc %d\n",
530 rqstp->rq_vers, rqstp->rq_proc);
531 proc = rqstp->rq_procinfo;
533 /* Check whether we have this call in the cache. */
534 switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
544 /* Decode arguments */
545 xdr = proc->pc_decode;
546 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
548 dprintk("nfsd: failed to decode arguments!\n");
549 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
550 *statp = rpc_garbage_args;
554 /* need to grab the location to store the status, as
555 * nfsv4 does some encoding while processing
557 nfserrp = rqstp->rq_res.head[0].iov_base
558 + rqstp->rq_res.head[0].iov_len;
559 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
561 /* Now call the procedure handler, and encode NFS status. */
562 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
563 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
564 if (nfserr == nfserr_dropit) {
565 dprintk("nfsd: Dropping request; may be revisited later\n");
566 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
570 if (rqstp->rq_proc != 0)
574 * For NFSv2, additional info is never returned in case of an error.
576 if (!(nfserr && rqstp->rq_vers == 2)) {
577 xdr = proc->pc_encode;
578 if (xdr && !xdr(rqstp, nfserrp,
580 /* Failed to encode result. Release cache entry */
581 dprintk("nfsd: failed to encode result!\n");
582 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
583 *statp = rpc_system_err;
588 /* Store reply in cache. */
589 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);