2 * Central processing for nfsd.
4 * Authors: Olaf Kirch (okir@monad.swb.de)
6 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
9 #include <linux/sched.h>
10 #include <linux/freezer.h>
11 #include <linux/fs_struct.h>
12 #include <linux/swap.h>
14 #include <linux/sunrpc/stats.h>
15 #include <linux/sunrpc/svcsock.h>
16 #include <linux/lockd/bind.h>
17 #include <linux/nfsacl.h>
18 #include <linux/seq_file.h>
19 #include <net/net_namespace.h>
24 #define NFSDDBG_FACILITY NFSDDBG_SVC
26 extern struct svc_program nfsd_program;
27 static int nfsd(void *vrqstp);
28 struct timeval nfssvc_boot;
31 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
32 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
33 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
35 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
36 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
37 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
38 * entry of ->sv_pools[].
40 * Transitions of the thread count between zero and non-zero are of particular
41 * interest since the svc_serv needs to be created and initialized at that
44 * Finally, the nfsd_mutex also protects some of the global variables that are
45 * accessed when nfsd starts and that are settable via the write_* routines in
46 * nfsctl.c. In particular:
48 * user_recovery_dirname
52 DEFINE_MUTEX(nfsd_mutex);
53 struct svc_serv *nfsd_serv;
56 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
57 * nfsd_drc_max_pages limits the total amount of memory available for
58 * version 4.1 DRC caches.
59 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
61 spinlock_t nfsd_drc_lock;
62 unsigned int nfsd_drc_max_mem;
63 unsigned int nfsd_drc_mem_used;
65 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
66 static struct svc_stat nfsd_acl_svcstats;
67 static struct svc_version * nfsd_acl_version[] = {
68 [2] = &nfsd_acl_version2,
69 [3] = &nfsd_acl_version3,
72 #define NFSD_ACL_MINVERS 2
73 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
74 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
76 static struct svc_program nfsd_acl_program = {
77 .pg_prog = NFS_ACL_PROGRAM,
78 .pg_nvers = NFSD_ACL_NRVERS,
79 .pg_vers = nfsd_acl_versions,
82 .pg_stats = &nfsd_acl_svcstats,
83 .pg_authenticate = &svc_set_client,
86 static struct svc_stat nfsd_acl_svcstats = {
87 .program = &nfsd_acl_program,
89 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
91 static struct svc_version * nfsd_version[] = {
93 #if defined(CONFIG_NFSD_V3)
96 #if defined(CONFIG_NFSD_V4)
101 #define NFSD_MINVERS 2
102 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
103 static struct svc_version *nfsd_versions[NFSD_NRVERS];
105 struct svc_program nfsd_program = {
106 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
107 .pg_next = &nfsd_acl_program,
109 .pg_prog = NFS_PROGRAM, /* program number */
110 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
111 .pg_vers = nfsd_versions, /* version table */
112 .pg_name = "nfsd", /* program name */
113 .pg_class = "nfsd", /* authentication class */
114 .pg_stats = &nfsd_svcstats, /* version table */
115 .pg_authenticate = &svc_set_client, /* export authentication */
119 u32 nfsd_supported_minorversion;
121 int nfsd_vers(int vers, enum vers_op change)
123 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
127 nfsd_versions[vers] = nfsd_version[vers];
128 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
129 if (vers < NFSD_ACL_NRVERS)
130 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
134 nfsd_versions[vers] = NULL;
135 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
136 if (vers < NFSD_ACL_NRVERS)
137 nfsd_acl_versions[vers] = NULL;
141 return nfsd_versions[vers] != NULL;
143 return nfsd_version[vers] != NULL;
148 int nfsd_minorversion(u32 minorversion, enum vers_op change)
150 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
154 nfsd_supported_minorversion = minorversion;
157 if (minorversion == 0)
159 nfsd_supported_minorversion = minorversion - 1;
162 return minorversion <= nfsd_supported_minorversion;
164 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
170 * Maximum number of nfsd processes
172 #define NFSD_MAXSERVS 8192
174 int nfsd_nrthreads(void)
177 mutex_lock(&nfsd_mutex);
179 rv = nfsd_serv->sv_nrthreads;
180 mutex_unlock(&nfsd_mutex);
184 static int nfsd_init_socks(int port)
187 if (!list_empty(&nfsd_serv->sv_permsocks))
190 error = svc_create_xprt(nfsd_serv, "udp", &init_net, PF_INET, port,
195 error = svc_create_xprt(nfsd_serv, "tcp", &init_net, PF_INET, port,
203 static bool nfsd_up = false;
205 static int nfsd_startup(unsigned short port, int nrservs)
212 * Readahead param cache - will no-op if it already exists.
213 * (Note therefore results will be suboptimal if number of
214 * threads is modified after nfsd start.)
216 ret = nfsd_racache_init(2*nrservs);
219 ret = nfsd_init_socks(port);
225 ret = nfs4_state_start();
233 nfsd_racache_shutdown();
237 static void nfsd_shutdown(void)
240 * write_ports can create the server without actually starting
241 * any threads--if we get shut down before any threads are
242 * started, then nfsd_last_thread will be run before any of this
243 * other initialization has been done.
247 nfs4_state_shutdown();
249 nfsd_racache_shutdown();
253 static void nfsd_last_thread(struct svc_serv *serv)
255 /* When last nfsd thread exits we need to do some clean-up */
259 svc_rpcb_cleanup(serv);
261 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
266 void nfsd_reset_versions(void)
271 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
272 if (nfsd_program.pg_vers[i])
277 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
278 nfsd_program.pg_vers[i] = nfsd_version[i];
279 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
280 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
281 nfsd_acl_program.pg_vers[i] =
288 * Each session guarantees a negotiated per slot memory cache for replies
289 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
290 * NFSv4.1 server might want to use more memory for a DRC than a machine
291 * with mutiple services.
293 * Impose a hard limit on the number of pages for the DRC which varies
294 * according to the machines free pages. This is of course only a default.
296 * For now this is a #defined shift which could be under admin control
299 static void set_max_drc(void)
301 #define NFSD_DRC_SIZE_SHIFT 10
302 nfsd_drc_max_mem = (nr_free_buffer_pages()
303 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
304 nfsd_drc_mem_used = 0;
305 spin_lock_init(&nfsd_drc_lock);
306 dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
309 int nfsd_create_serv(void)
313 WARN_ON(!mutex_is_locked(&nfsd_mutex));
318 if (nfsd_max_blksize == 0) {
319 /* choose a suitable default */
322 /* Aim for 1/4096 of memory per thread
323 * This gives 1MB on 4Gig machines
324 * But only uses 32K on 128M machines.
325 * Bottom out at 8K on 32M and smaller.
326 * Of course, this is only a default.
328 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
329 i.totalram <<= PAGE_SHIFT - 12;
330 while (nfsd_max_blksize > i.totalram &&
331 nfsd_max_blksize >= 8*1024*2)
332 nfsd_max_blksize /= 2;
334 nfsd_reset_versions();
336 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
337 nfsd_last_thread, nfsd, THIS_MODULE);
338 if (nfsd_serv == NULL)
342 do_gettimeofday(&nfssvc_boot); /* record boot time */
346 int nfsd_nrpools(void)
348 if (nfsd_serv == NULL)
351 return nfsd_serv->sv_nrpools;
354 int nfsd_get_nrthreads(int n, int *nthreads)
358 if (nfsd_serv != NULL) {
359 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
360 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
366 int nfsd_set_nrthreads(int n, int *nthreads)
372 WARN_ON(!mutex_is_locked(&nfsd_mutex));
374 if (nfsd_serv == NULL || n <= 0)
377 if (n > nfsd_serv->sv_nrpools)
378 n = nfsd_serv->sv_nrpools;
380 /* enforce a global maximum number of threads */
382 for (i = 0; i < n; i++) {
383 if (nthreads[i] > NFSD_MAXSERVS)
384 nthreads[i] = NFSD_MAXSERVS;
387 if (tot > NFSD_MAXSERVS) {
388 /* total too large: scale down requested numbers */
389 for (i = 0; i < n && tot > 0; i++) {
390 int new = nthreads[i] * NFSD_MAXSERVS / tot;
391 tot -= (nthreads[i] - new);
394 for (i = 0; i < n && tot > 0; i++) {
401 * There must always be a thread in pool 0; the admin
402 * can't shut down NFS completely using pool_threads.
404 if (nthreads[0] == 0)
407 /* apply the new numbers */
409 for (i = 0; i < n; i++) {
410 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
415 svc_destroy(nfsd_serv);
421 * Adjust the number of threads and return the new number of threads.
422 * This is also the function that starts the server if necessary, if
423 * this is the first time nrservs is nonzero.
426 nfsd_svc(unsigned short port, int nrservs)
431 mutex_lock(&nfsd_mutex);
432 dprintk("nfsd: creating service\n");
435 if (nrservs > NFSD_MAXSERVS)
436 nrservs = NFSD_MAXSERVS;
438 if (nrservs == 0 && nfsd_serv == NULL)
441 error = nfsd_create_serv();
445 nfsd_up_before = nfsd_up;
447 error = nfsd_startup(port, nrservs);
450 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
453 /* We are holding a reference to nfsd_serv which
454 * we don't want to count in the return value,
457 error = nfsd_serv->sv_nrthreads - 1;
459 if (error < 0 && !nfsd_up_before)
462 svc_destroy(nfsd_serv); /* Release server */
464 mutex_unlock(&nfsd_mutex);
470 * This is the NFS server kernel thread
475 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
476 int err, preverr = 0;
478 /* Lock module and set up kernel thread */
479 mutex_lock(&nfsd_mutex);
481 /* At this point, the thread shares current->fs
482 * with the init process. We need to create files with a
483 * umask of 0 instead of init's umask. */
484 if (unshare_fs_struct() < 0) {
485 printk("Unable to start nfsd thread: out of memory\n");
489 current->fs->umask = 0;
492 * thread is spawned with all signals set to SIG_IGN, re-enable
493 * the ones that will bring down the thread
495 allow_signal(SIGKILL);
496 allow_signal(SIGHUP);
497 allow_signal(SIGINT);
498 allow_signal(SIGQUIT);
501 mutex_unlock(&nfsd_mutex);
504 * We want less throttling in balance_dirty_pages() so that nfs to
505 * localhost doesn't cause nfsd to lock up due to all the client's
508 current->flags |= PF_LESS_THROTTLE;
512 * The main request loop
516 * Find a socket with data available and call its
519 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
524 if (err != preverr) {
525 printk(KERN_WARNING "%s: unexpected error "
526 "from svc_recv (%d)\n", __func__, -err);
529 schedule_timeout_uninterruptible(HZ);
533 validate_process_creds();
535 validate_process_creds();
538 /* Clear signals before calling svc_exit_thread() */
539 flush_signals(current);
541 mutex_lock(&nfsd_mutex);
545 /* Release the thread */
546 svc_exit_thread(rqstp);
549 mutex_unlock(&nfsd_mutex);
550 module_put_and_exit(0);
554 static __be32 map_new_errors(u32 vers, __be32 nfserr)
556 if (nfserr == nfserr_jukebox && vers == 2)
557 return nfserr_dropit;
558 if (nfserr == nfserr_wrongsec && vers < 4)
564 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
566 struct svc_procedure *proc;
571 dprintk("nfsd_dispatch: vers %d proc %d\n",
572 rqstp->rq_vers, rqstp->rq_proc);
573 proc = rqstp->rq_procinfo;
576 * Give the xdr decoder a chance to change this if it wants
577 * (necessary in the NFSv4.0 compound case)
579 rqstp->rq_cachetype = proc->pc_cachetype;
580 /* Decode arguments */
581 xdr = proc->pc_decode;
582 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
584 dprintk("nfsd: failed to decode arguments!\n");
585 *statp = rpc_garbage_args;
589 /* Check whether we have this call in the cache. */
590 switch (nfsd_cache_lookup(rqstp)) {
600 /* need to grab the location to store the status, as
601 * nfsv4 does some encoding while processing
603 nfserrp = rqstp->rq_res.head[0].iov_base
604 + rqstp->rq_res.head[0].iov_len;
605 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
607 /* Now call the procedure handler, and encode NFS status. */
608 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
609 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
610 if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
611 dprintk("nfsd: Dropping request; may be revisited later\n");
612 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
616 if (rqstp->rq_proc != 0)
620 * For NFSv2, additional info is never returned in case of an error.
622 if (!(nfserr && rqstp->rq_vers == 2)) {
623 xdr = proc->pc_encode;
624 if (xdr && !xdr(rqstp, nfserrp,
626 /* Failed to encode result. Release cache entry */
627 dprintk("nfsd: failed to encode result!\n");
628 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
629 *statp = rpc_system_err;
634 /* Store reply in cache. */
635 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
639 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
642 mutex_lock(&nfsd_mutex);
643 if (nfsd_serv == NULL) {
644 mutex_unlock(&nfsd_mutex);
647 /* bump up the psudo refcount while traversing */
649 ret = svc_pool_stats_open(nfsd_serv, file);
650 mutex_unlock(&nfsd_mutex);
654 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
656 int ret = seq_release(inode, file);
657 mutex_lock(&nfsd_mutex);
658 /* this function really, really should have been called svc_put() */
659 svc_destroy(nfsd_serv);
660 mutex_unlock(&nfsd_mutex);