2 * Central processing for nfsd.
9 #include <linux/sched.h>
10 #include <linux/freezer.h>
11 #include <linux/module.h>
12 #include <linux/fs_struct.h>
13 #include <linux/swap.h>
15 #include <linux/sunrpc/stats.h>
16 #include <linux/sunrpc/svcsock.h>
17 #include <linux/lockd/bind.h>
18 #include <linux/nfsacl.h>
19 #include <linux/seq_file.h>
20 #include <net/net_namespace.h>
26 #define NFSDDBG_FACILITY NFSDDBG_SVC
28 extern struct svc_program nfsd_program;
29 static int nfsd(void *vrqstp);
32 * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members
33 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
34 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
36 * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a
37 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
38 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
39 * entry of ->sv_pools[].
41 * Transitions of the thread count between zero and non-zero are of particular
42 * interest since the svc_serv needs to be created and initialized at that
45 * Finally, the nfsd_mutex also protects some of the global variables that are
46 * accessed when nfsd starts and that are settable via the write_* routines in
47 * nfsctl.c. In particular:
49 * user_recovery_dirname
53 DEFINE_MUTEX(nfsd_mutex);
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(struct net *net)
177 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
179 mutex_lock(&nfsd_mutex);
181 rv = nn->nfsd_serv->sv_nrthreads;
182 mutex_unlock(&nfsd_mutex);
186 static int nfsd_init_socks(struct net *net)
189 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
191 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
194 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
199 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
207 static int nfsd_users = 0;
209 static int nfsd_startup_generic(int nrservs)
217 * Readahead param cache - will no-op if it already exists.
218 * (Note therefore results will be suboptimal if number of
219 * threads is modified after nfsd start.)
221 ret = nfsd_racache_init(2*nrservs);
224 ret = nfs4_state_start();
230 nfsd_racache_shutdown();
234 static void nfsd_shutdown_generic(void)
239 nfs4_state_shutdown();
240 nfsd_racache_shutdown();
243 static int nfsd_startup_net(int nrservs, struct net *net)
245 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
251 ret = nfsd_startup_generic(nrservs);
254 ret = nfsd_init_socks(net);
260 ret = nfs4_state_start_net(net);
264 nn->nfsd_net_up = true;
270 nfsd_shutdown_generic();
274 static void nfsd_shutdown_net(struct net *net)
276 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
278 nfs4_state_shutdown_net(net);
280 nn->nfsd_net_up = false;
281 nfsd_shutdown_generic();
284 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
286 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
289 * write_ports can create the server without actually starting
290 * any threads--if we get shut down before any threads are
291 * started, then nfsd_last_thread will be run before any of this
292 * other initialization has been done.
294 if (!nn->nfsd_net_up)
296 nfsd_shutdown_net(net);
298 svc_rpcb_cleanup(serv, net);
300 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
302 nfsd_export_flush(net);
305 void nfsd_reset_versions(void)
310 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
311 if (nfsd_program.pg_vers[i])
316 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
317 nfsd_program.pg_vers[i] = nfsd_version[i];
318 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
319 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
320 nfsd_acl_program.pg_vers[i] =
327 * Each session guarantees a negotiated per slot memory cache for replies
328 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
329 * NFSv4.1 server might want to use more memory for a DRC than a machine
330 * with mutiple services.
332 * Impose a hard limit on the number of pages for the DRC which varies
333 * according to the machines free pages. This is of course only a default.
335 * For now this is a #defined shift which could be under admin control
338 static void set_max_drc(void)
340 #define NFSD_DRC_SIZE_SHIFT 10
341 nfsd_drc_max_mem = (nr_free_buffer_pages()
342 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
343 nfsd_drc_mem_used = 0;
344 spin_lock_init(&nfsd_drc_lock);
345 dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
348 static int nfsd_get_default_max_blksize(void)
351 unsigned long long target;
355 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
357 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
358 * machines, but only uses 32K on 128M machines. Bottom out at
359 * 8K on 32M and smaller. Of course, this is only a default.
363 ret = NFSSVC_MAXBLKSIZE;
364 while (ret > target && ret >= 8*1024*2)
369 int nfsd_create_serv(struct net *net)
372 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
374 WARN_ON(!mutex_is_locked(&nfsd_mutex));
376 svc_get(nn->nfsd_serv);
379 if (nfsd_max_blksize == 0)
380 nfsd_max_blksize = nfsd_get_default_max_blksize();
381 nfsd_reset_versions();
382 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
383 nfsd_last_thread, nfsd, THIS_MODULE);
384 if (nn->nfsd_serv == NULL)
387 error = svc_bind(nn->nfsd_serv, net);
389 svc_destroy(nn->nfsd_serv);
394 do_gettimeofday(&nn->nfssvc_boot); /* record boot time */
398 int nfsd_nrpools(struct net *net)
400 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
402 if (nn->nfsd_serv == NULL)
405 return nn->nfsd_serv->sv_nrpools;
408 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
411 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
413 if (nn->nfsd_serv != NULL) {
414 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
415 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
421 void nfsd_destroy(struct net *net)
423 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
424 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
427 svc_shutdown_net(nn->nfsd_serv, net);
428 svc_destroy(nn->nfsd_serv);
430 nn->nfsd_serv = NULL;
433 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
438 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
440 WARN_ON(!mutex_is_locked(&nfsd_mutex));
442 if (nn->nfsd_serv == NULL || n <= 0)
445 if (n > nn->nfsd_serv->sv_nrpools)
446 n = nn->nfsd_serv->sv_nrpools;
448 /* enforce a global maximum number of threads */
450 for (i = 0; i < n; i++) {
451 if (nthreads[i] > NFSD_MAXSERVS)
452 nthreads[i] = NFSD_MAXSERVS;
455 if (tot > NFSD_MAXSERVS) {
456 /* total too large: scale down requested numbers */
457 for (i = 0; i < n && tot > 0; i++) {
458 int new = nthreads[i] * NFSD_MAXSERVS / tot;
459 tot -= (nthreads[i] - new);
462 for (i = 0; i < n && tot > 0; i++) {
469 * There must always be a thread in pool 0; the admin
470 * can't shut down NFS completely using pool_threads.
472 if (nthreads[0] == 0)
475 /* apply the new numbers */
476 svc_get(nn->nfsd_serv);
477 for (i = 0; i < n; i++) {
478 err = svc_set_num_threads(nn->nfsd_serv, &nn->nfsd_serv->sv_pools[i],
488 * Adjust the number of threads and return the new number of threads.
489 * This is also the function that starts the server if necessary, if
490 * this is the first time nrservs is nonzero.
493 nfsd_svc(int nrservs, struct net *net)
497 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
499 mutex_lock(&nfsd_mutex);
500 dprintk("nfsd: creating service\n");
503 if (nrservs > NFSD_MAXSERVS)
504 nrservs = NFSD_MAXSERVS;
506 if (nrservs == 0 && nn->nfsd_serv == NULL)
509 error = nfsd_create_serv(net);
513 nfsd_up_before = nn->nfsd_net_up;
515 error = nfsd_startup_net(nrservs, net);
518 error = svc_set_num_threads(nn->nfsd_serv, NULL, nrservs);
521 /* We are holding a reference to nn->nfsd_serv which
522 * we don't want to count in the return value,
525 error = nn->nfsd_serv->sv_nrthreads - 1;
527 if (error < 0 && !nfsd_up_before)
528 nfsd_shutdown_net(net);
530 nfsd_destroy(net); /* Release server */
532 mutex_unlock(&nfsd_mutex);
538 * This is the NFS server kernel thread
543 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
544 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
545 struct net *net = perm_sock->xpt_net;
548 /* Lock module and set up kernel thread */
549 mutex_lock(&nfsd_mutex);
551 /* At this point, the thread shares current->fs
552 * with the init process. We need to create files with a
553 * umask of 0 instead of init's umask. */
554 if (unshare_fs_struct() < 0) {
555 printk("Unable to start nfsd thread: out of memory\n");
559 current->fs->umask = 0;
562 * thread is spawned with all signals set to SIG_IGN, re-enable
563 * the ones that will bring down the thread
565 allow_signal(SIGKILL);
566 allow_signal(SIGHUP);
567 allow_signal(SIGINT);
568 allow_signal(SIGQUIT);
571 mutex_unlock(&nfsd_mutex);
574 * We want less throttling in balance_dirty_pages() so that nfs to
575 * localhost doesn't cause nfsd to lock up due to all the client's
578 current->flags |= PF_LESS_THROTTLE;
582 * The main request loop
586 * Find a socket with data available and call its
589 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
593 validate_process_creds();
595 validate_process_creds();
598 /* Clear signals before calling svc_exit_thread() */
599 flush_signals(current);
601 mutex_lock(&nfsd_mutex);
605 rqstp->rq_server = NULL;
607 /* Release the thread */
608 svc_exit_thread(rqstp);
613 mutex_unlock(&nfsd_mutex);
614 module_put_and_exit(0);
618 static __be32 map_new_errors(u32 vers, __be32 nfserr)
620 if (nfserr == nfserr_jukebox && vers == 2)
621 return nfserr_dropit;
622 if (nfserr == nfserr_wrongsec && vers < 4)
628 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
630 struct svc_procedure *proc;
635 dprintk("nfsd_dispatch: vers %d proc %d\n",
636 rqstp->rq_vers, rqstp->rq_proc);
637 proc = rqstp->rq_procinfo;
640 * Give the xdr decoder a chance to change this if it wants
641 * (necessary in the NFSv4.0 compound case)
643 rqstp->rq_cachetype = proc->pc_cachetype;
644 /* Decode arguments */
645 xdr = proc->pc_decode;
646 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
648 dprintk("nfsd: failed to decode arguments!\n");
649 *statp = rpc_garbage_args;
653 /* Check whether we have this call in the cache. */
654 switch (nfsd_cache_lookup(rqstp)) {
664 /* need to grab the location to store the status, as
665 * nfsv4 does some encoding while processing
667 nfserrp = rqstp->rq_res.head[0].iov_base
668 + rqstp->rq_res.head[0].iov_len;
669 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
671 /* Now call the procedure handler, and encode NFS status. */
672 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
673 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
674 if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
675 dprintk("nfsd: Dropping request; may be revisited later\n");
676 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
680 if (rqstp->rq_proc != 0)
684 * For NFSv2, additional info is never returned in case of an error.
686 if (!(nfserr && rqstp->rq_vers == 2)) {
687 xdr = proc->pc_encode;
688 if (xdr && !xdr(rqstp, nfserrp,
690 /* Failed to encode result. Release cache entry */
691 dprintk("nfsd: failed to encode result!\n");
692 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
693 *statp = rpc_system_err;
698 /* Store reply in cache. */
699 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
703 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
706 struct net *net = &init_net;
707 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
709 mutex_lock(&nfsd_mutex);
710 if (nn->nfsd_serv == NULL) {
711 mutex_unlock(&nfsd_mutex);
714 /* bump up the psudo refcount while traversing */
715 svc_get(nn->nfsd_serv);
716 ret = svc_pool_stats_open(nn->nfsd_serv, file);
717 mutex_unlock(&nfsd_mutex);
721 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
723 int ret = seq_release(inode, file);
724 struct net *net = &init_net;
726 mutex_lock(&nfsd_mutex);
727 /* this function really, really should have been called svc_put() */
729 mutex_unlock(&nfsd_mutex);