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>
25 #include <linux/swap.h>
27 #include <linux/sunrpc/types.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/sunrpc/svc.h>
30 #include <linux/sunrpc/svcsock.h>
31 #include <linux/sunrpc/cache.h>
32 #include <linux/nfsd/nfsd.h>
33 #include <linux/nfsd/stats.h>
34 #include <linux/nfsd/cache.h>
35 #include <linux/nfsd/syscall.h>
36 #include <linux/lockd/bind.h>
37 #include <linux/nfsacl.h>
39 #define NFSDDBG_FACILITY NFSDDBG_SVC
41 extern struct svc_program nfsd_program;
42 static int nfsd(void *vrqstp);
43 struct timeval nfssvc_boot;
46 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
47 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
48 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
50 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
51 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
52 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
53 * entry of ->sv_pools[].
55 * Transitions of the thread count between zero and non-zero are of particular
56 * interest since the svc_serv needs to be created and initialized at that
59 * Finally, the nfsd_mutex also protects some of the global variables that are
60 * accessed when nfsd starts and that are settable via the write_* routines in
61 * nfsctl.c. In particular:
63 * user_recovery_dirname
67 DEFINE_MUTEX(nfsd_mutex);
68 struct svc_serv *nfsd_serv;
70 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
71 static struct svc_stat nfsd_acl_svcstats;
72 static struct svc_version * nfsd_acl_version[] = {
73 [2] = &nfsd_acl_version2,
74 [3] = &nfsd_acl_version3,
77 #define NFSD_ACL_MINVERS 2
78 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
79 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
81 static struct svc_program nfsd_acl_program = {
82 .pg_prog = NFS_ACL_PROGRAM,
83 .pg_nvers = NFSD_ACL_NRVERS,
84 .pg_vers = nfsd_acl_versions,
87 .pg_stats = &nfsd_acl_svcstats,
88 .pg_authenticate = &svc_set_client,
91 static struct svc_stat nfsd_acl_svcstats = {
92 .program = &nfsd_acl_program,
94 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
96 static struct svc_version * nfsd_version[] = {
98 #if defined(CONFIG_NFSD_V3)
101 #if defined(CONFIG_NFSD_V4)
102 [4] = &nfsd_version4,
106 #define NFSD_MINVERS 2
107 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
108 static struct svc_version *nfsd_versions[NFSD_NRVERS];
110 struct svc_program nfsd_program = {
111 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
112 .pg_next = &nfsd_acl_program,
114 .pg_prog = NFS_PROGRAM, /* program number */
115 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
116 .pg_vers = nfsd_versions, /* version table */
117 .pg_name = "nfsd", /* program name */
118 .pg_class = "nfsd", /* authentication class */
119 .pg_stats = &nfsd_svcstats, /* version table */
120 .pg_authenticate = &svc_set_client, /* export authentication */
124 int nfsd_vers(int vers, enum vers_op change)
126 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
130 nfsd_versions[vers] = nfsd_version[vers];
131 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
132 if (vers < NFSD_ACL_NRVERS)
133 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
137 nfsd_versions[vers] = NULL;
138 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
139 if (vers < NFSD_ACL_NRVERS)
140 nfsd_acl_versions[vers] = NULL;
144 return nfsd_versions[vers] != NULL;
146 return nfsd_version[vers] != NULL;
151 * Maximum number of nfsd processes
153 #define NFSD_MAXSERVS 8192
155 int nfsd_nrthreads(void)
158 mutex_lock(&nfsd_mutex);
160 rv = nfsd_serv->sv_nrthreads;
161 mutex_unlock(&nfsd_mutex);
165 static void nfsd_last_thread(struct svc_serv *serv)
167 /* When last nfsd thread exits we need to do some clean-up */
168 struct svc_xprt *xprt;
169 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list)
172 nfsd_racache_shutdown();
173 nfs4_state_shutdown();
175 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
180 void nfsd_reset_versions(void)
185 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
186 if (nfsd_program.pg_vers[i])
191 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
192 nfsd_program.pg_vers[i] = nfsd_version[i];
193 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
194 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
195 nfsd_acl_program.pg_vers[i] =
202 * Each session guarantees a negotiated per slot memory cache for replies
203 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
204 * NFSv4.1 server might want to use more memory for a DRC than a machine
205 * with mutiple services.
207 * Impose a hard limit on the number of pages for the DRC which varies
208 * according to the machines free pages. This is of course only a default.
210 * For now this is a #defined shift which could be under admin control
213 static void set_max_drc(void)
215 nfsd_serv->sv_drc_max_pages = nr_free_buffer_pages()
216 >> NFSD_DRC_SIZE_SHIFT;
217 nfsd_serv->sv_drc_pages_used = 0;
218 dprintk("%s svc_drc_max_pages %u\n", __func__,
219 nfsd_serv->sv_drc_max_pages);
222 int nfsd_create_serv(void)
226 WARN_ON(!mutex_is_locked(&nfsd_mutex));
231 if (nfsd_max_blksize == 0) {
232 /* choose a suitable default */
235 /* Aim for 1/4096 of memory per thread
236 * This gives 1MB on 4Gig machines
237 * But only uses 32K on 128M machines.
238 * Bottom out at 8K on 32M and smaller.
239 * Of course, this is only a default.
241 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
242 i.totalram <<= PAGE_SHIFT - 12;
243 while (nfsd_max_blksize > i.totalram &&
244 nfsd_max_blksize >= 8*1024*2)
245 nfsd_max_blksize /= 2;
248 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
250 nfsd_last_thread, nfsd, THIS_MODULE);
251 if (nfsd_serv == NULL)
256 do_gettimeofday(&nfssvc_boot); /* record boot time */
260 static int nfsd_init_socks(int port)
263 if (!list_empty(&nfsd_serv->sv_permsocks))
266 error = svc_create_xprt(nfsd_serv, "udp", port,
275 error = svc_create_xprt(nfsd_serv, "tcp", port,
287 int nfsd_nrpools(void)
289 if (nfsd_serv == NULL)
292 return nfsd_serv->sv_nrpools;
295 int nfsd_get_nrthreads(int n, int *nthreads)
299 if (nfsd_serv != NULL) {
300 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
301 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
307 int nfsd_set_nrthreads(int n, int *nthreads)
313 WARN_ON(!mutex_is_locked(&nfsd_mutex));
315 if (nfsd_serv == NULL || n <= 0)
318 if (n > nfsd_serv->sv_nrpools)
319 n = nfsd_serv->sv_nrpools;
321 /* enforce a global maximum number of threads */
323 for (i = 0; i < n; i++) {
324 if (nthreads[i] > NFSD_MAXSERVS)
325 nthreads[i] = NFSD_MAXSERVS;
328 if (tot > NFSD_MAXSERVS) {
329 /* total too large: scale down requested numbers */
330 for (i = 0; i < n && tot > 0; i++) {
331 int new = nthreads[i] * NFSD_MAXSERVS / tot;
332 tot -= (nthreads[i] - new);
335 for (i = 0; i < n && tot > 0; i++) {
342 * There must always be a thread in pool 0; the admin
343 * can't shut down NFS completely using pool_threads.
345 if (nthreads[0] == 0)
348 /* apply the new numbers */
350 for (i = 0; i < n; i++) {
351 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
356 svc_destroy(nfsd_serv);
362 nfsd_svc(unsigned short port, int nrservs)
366 mutex_lock(&nfsd_mutex);
367 dprintk("nfsd: creating service\n");
371 if (nrservs > NFSD_MAXSERVS)
372 nrservs = NFSD_MAXSERVS;
374 /* Readahead param cache - will no-op if it already exists */
375 error = nfsd_racache_init(2*nrservs);
380 nfsd_reset_versions();
382 error = nfsd_create_serv();
386 error = nfsd_init_socks(port);
390 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
392 svc_destroy(nfsd_serv); /* Release server */
394 mutex_unlock(&nfsd_mutex);
400 * This is the NFS server kernel thread
405 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
406 struct fs_struct *fsp;
407 int err, preverr = 0;
409 /* Lock module and set up kernel thread */
410 mutex_lock(&nfsd_mutex);
412 /* At this point, the thread shares current->fs
413 * with the init process. We need to create files with a
414 * umask of 0 instead of init's umask. */
415 fsp = copy_fs_struct(current->fs);
417 printk("Unable to start nfsd thread: out of memory\n");
422 current->fs->umask = 0;
425 * thread is spawned with all signals set to SIG_IGN, re-enable
426 * the ones that will bring down the thread
428 allow_signal(SIGKILL);
429 allow_signal(SIGHUP);
430 allow_signal(SIGINT);
431 allow_signal(SIGQUIT);
434 mutex_unlock(&nfsd_mutex);
437 * We want less throttling in balance_dirty_pages() so that nfs to
438 * localhost doesn't cause nfsd to lock up due to all the client's
441 current->flags |= PF_LESS_THROTTLE;
445 * The main request loop
449 * Find a socket with data available and call its
452 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
457 if (err != preverr) {
458 printk(KERN_WARNING "%s: unexpected error "
459 "from svc_recv (%d)\n", __func__, -err);
462 schedule_timeout_uninterruptible(HZ);
467 /* Lock the export hash tables for reading. */
472 /* Unlock export hash tables */
476 /* Clear signals before calling svc_exit_thread() */
477 flush_signals(current);
479 mutex_lock(&nfsd_mutex);
483 /* Release the thread */
484 svc_exit_thread(rqstp);
487 mutex_unlock(&nfsd_mutex);
488 module_put_and_exit(0);
492 static __be32 map_new_errors(u32 vers, __be32 nfserr)
494 if (nfserr == nfserr_jukebox && vers == 2)
495 return nfserr_dropit;
496 if (nfserr == nfserr_wrongsec && vers < 4)
502 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
504 struct svc_procedure *proc;
509 dprintk("nfsd_dispatch: vers %d proc %d\n",
510 rqstp->rq_vers, rqstp->rq_proc);
511 proc = rqstp->rq_procinfo;
513 /* Check whether we have this call in the cache. */
514 switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
524 /* Decode arguments */
525 xdr = proc->pc_decode;
526 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
528 dprintk("nfsd: failed to decode arguments!\n");
529 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
530 *statp = rpc_garbage_args;
534 /* need to grab the location to store the status, as
535 * nfsv4 does some encoding while processing
537 nfserrp = rqstp->rq_res.head[0].iov_base
538 + rqstp->rq_res.head[0].iov_len;
539 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
541 /* NFSv4.1 DRC requires statp */
542 if (rqstp->rq_vers == 4)
543 nfsd4_set_statp(rqstp, statp);
545 /* Now call the procedure handler, and encode NFS status. */
546 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
547 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
548 if (nfserr == nfserr_dropit) {
549 dprintk("nfsd: Dropping request; may be revisited later\n");
550 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
554 if (rqstp->rq_proc != 0)
558 * For NFSv2, additional info is never returned in case of an error.
560 if (!(nfserr && rqstp->rq_vers == 2)) {
561 xdr = proc->pc_encode;
562 if (xdr && !xdr(rqstp, nfserrp,
564 /* Failed to encode result. Release cache entry */
565 dprintk("nfsd: failed to encode result!\n");
566 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
567 *statp = rpc_system_err;
572 /* Store reply in cache. */
573 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
577 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
579 if (nfsd_serv == NULL)
581 return svc_pool_stats_open(nfsd_serv, file);