net: Clone skb before setting peeked flag
[linux/fpc-iii.git] / fs / nfsd / nfssvc.c
blob9277cc91c21b051011c4dbfe4d677b32d37ad37f
1 /*
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>
7 */
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>
21 #include "nfsd.h"
22 #include "cache.h"
23 #include "vfs.h"
24 #include "netns.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
43 * point, or freed.
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
50 * user_lease_time
51 * nfsd_versions
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 long nfsd_drc_max_mem;
63 unsigned long 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,
80 .pg_name = "nfsacl",
81 .pg_class = "nfsd",
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[] = {
92 [2] = &nfsd_version2,
93 #if defined(CONFIG_NFSD_V3)
94 [3] = &nfsd_version3,
95 #endif
96 #if defined(CONFIG_NFSD_V4)
97 [4] = &nfsd_version4,
98 #endif
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,
108 #endif
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 static bool nfsd_supported_minorversions[NFSD_SUPPORTED_MINOR_VERSION + 1] = {
120 [0] = 1,
121 [1] = 1,
122 [2] = 1,
125 int nfsd_vers(int vers, enum vers_op change)
127 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
128 return 0;
129 switch(change) {
130 case NFSD_SET:
131 nfsd_versions[vers] = nfsd_version[vers];
132 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
133 if (vers < NFSD_ACL_NRVERS)
134 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
135 #endif
136 break;
137 case NFSD_CLEAR:
138 nfsd_versions[vers] = NULL;
139 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
140 if (vers < NFSD_ACL_NRVERS)
141 nfsd_acl_versions[vers] = NULL;
142 #endif
143 break;
144 case NFSD_TEST:
145 return nfsd_versions[vers] != NULL;
146 case NFSD_AVAIL:
147 return nfsd_version[vers] != NULL;
149 return 0;
152 int nfsd_minorversion(u32 minorversion, enum vers_op change)
154 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
155 return -1;
156 switch(change) {
157 case NFSD_SET:
158 nfsd_supported_minorversions[minorversion] = true;
159 break;
160 case NFSD_CLEAR:
161 nfsd_supported_minorversions[minorversion] = false;
162 break;
163 case NFSD_TEST:
164 return nfsd_supported_minorversions[minorversion];
165 case NFSD_AVAIL:
166 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
168 return 0;
172 * Maximum number of nfsd processes
174 #define NFSD_MAXSERVS 8192
176 int nfsd_nrthreads(struct net *net)
178 int rv = 0;
179 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
181 mutex_lock(&nfsd_mutex);
182 if (nn->nfsd_serv)
183 rv = nn->nfsd_serv->sv_nrthreads;
184 mutex_unlock(&nfsd_mutex);
185 return rv;
188 static int nfsd_init_socks(struct net *net)
190 int error;
191 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
193 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
194 return 0;
196 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
197 SVC_SOCK_DEFAULTS);
198 if (error < 0)
199 return error;
201 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
202 SVC_SOCK_DEFAULTS);
203 if (error < 0)
204 return error;
206 return 0;
209 static int nfsd_users = 0;
211 static int nfsd_startup_generic(int nrservs)
213 int ret;
215 if (nfsd_users++)
216 return 0;
219 * Readahead param cache - will no-op if it already exists.
220 * (Note therefore results will be suboptimal if number of
221 * threads is modified after nfsd start.)
223 ret = nfsd_racache_init(2*nrservs);
224 if (ret)
225 goto dec_users;
227 ret = nfs4_state_start();
228 if (ret)
229 goto out_racache;
230 return 0;
232 out_racache:
233 nfsd_racache_shutdown();
234 dec_users:
235 nfsd_users--;
236 return ret;
239 static void nfsd_shutdown_generic(void)
241 if (--nfsd_users)
242 return;
244 nfs4_state_shutdown();
245 nfsd_racache_shutdown();
248 static bool nfsd_needs_lockd(void)
250 #if defined(CONFIG_NFSD_V3)
251 return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
252 #else
253 return (nfsd_versions[2] != NULL);
254 #endif
257 static int nfsd_startup_net(int nrservs, struct net *net)
259 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
260 int ret;
262 if (nn->nfsd_net_up)
263 return 0;
265 ret = nfsd_startup_generic(nrservs);
266 if (ret)
267 return ret;
268 ret = nfsd_init_socks(net);
269 if (ret)
270 goto out_socks;
272 if (nfsd_needs_lockd() && !nn->lockd_up) {
273 ret = lockd_up(net);
274 if (ret)
275 goto out_socks;
276 nn->lockd_up = 1;
279 ret = nfs4_state_start_net(net);
280 if (ret)
281 goto out_lockd;
283 nn->nfsd_net_up = true;
284 return 0;
286 out_lockd:
287 if (nn->lockd_up) {
288 lockd_down(net);
289 nn->lockd_up = 0;
291 out_socks:
292 nfsd_shutdown_generic();
293 return ret;
296 static void nfsd_shutdown_net(struct net *net)
298 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
300 nfs4_state_shutdown_net(net);
301 if (nn->lockd_up) {
302 lockd_down(net);
303 nn->lockd_up = 0;
305 nn->nfsd_net_up = false;
306 nfsd_shutdown_generic();
309 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
311 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
314 * write_ports can create the server without actually starting
315 * any threads--if we get shut down before any threads are
316 * started, then nfsd_last_thread will be run before any of this
317 * other initialization has been done.
319 if (!nn->nfsd_net_up)
320 return;
321 nfsd_shutdown_net(net);
323 svc_rpcb_cleanup(serv, net);
325 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
326 "cache\n");
327 nfsd_export_flush(net);
330 void nfsd_reset_versions(void)
332 int found_one = 0;
333 int i;
335 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
336 if (nfsd_program.pg_vers[i])
337 found_one = 1;
340 if (!found_one) {
341 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
342 nfsd_program.pg_vers[i] = nfsd_version[i];
343 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
344 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
345 nfsd_acl_program.pg_vers[i] =
346 nfsd_acl_version[i];
347 #endif
352 * Each session guarantees a negotiated per slot memory cache for replies
353 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
354 * NFSv4.1 server might want to use more memory for a DRC than a machine
355 * with mutiple services.
357 * Impose a hard limit on the number of pages for the DRC which varies
358 * according to the machines free pages. This is of course only a default.
360 * For now this is a #defined shift which could be under admin control
361 * in the future.
363 static void set_max_drc(void)
365 #define NFSD_DRC_SIZE_SHIFT 10
366 nfsd_drc_max_mem = (nr_free_buffer_pages()
367 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
368 nfsd_drc_mem_used = 0;
369 spin_lock_init(&nfsd_drc_lock);
370 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
373 static int nfsd_get_default_max_blksize(void)
375 struct sysinfo i;
376 unsigned long long target;
377 unsigned long ret;
379 si_meminfo(&i);
380 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
382 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
383 * machines, but only uses 32K on 128M machines. Bottom out at
384 * 8K on 32M and smaller. Of course, this is only a default.
386 target >>= 12;
388 ret = NFSSVC_MAXBLKSIZE;
389 while (ret > target && ret >= 8*1024*2)
390 ret /= 2;
391 return ret;
394 int nfsd_create_serv(struct net *net)
396 int error;
397 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
399 WARN_ON(!mutex_is_locked(&nfsd_mutex));
400 if (nn->nfsd_serv) {
401 svc_get(nn->nfsd_serv);
402 return 0;
404 if (nfsd_max_blksize == 0)
405 nfsd_max_blksize = nfsd_get_default_max_blksize();
406 nfsd_reset_versions();
407 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
408 nfsd_last_thread, nfsd, THIS_MODULE);
409 if (nn->nfsd_serv == NULL)
410 return -ENOMEM;
412 nn->nfsd_serv->sv_maxconn = nn->max_connections;
413 error = svc_bind(nn->nfsd_serv, net);
414 if (error < 0) {
415 svc_destroy(nn->nfsd_serv);
416 return error;
419 set_max_drc();
420 do_gettimeofday(&nn->nfssvc_boot); /* record boot time */
421 return 0;
424 int nfsd_nrpools(struct net *net)
426 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
428 if (nn->nfsd_serv == NULL)
429 return 0;
430 else
431 return nn->nfsd_serv->sv_nrpools;
434 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
436 int i = 0;
437 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
439 if (nn->nfsd_serv != NULL) {
440 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
441 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
444 return 0;
447 void nfsd_destroy(struct net *net)
449 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
450 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
452 if (destroy)
453 svc_shutdown_net(nn->nfsd_serv, net);
454 svc_destroy(nn->nfsd_serv);
455 if (destroy)
456 nn->nfsd_serv = NULL;
459 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
461 int i = 0;
462 int tot = 0;
463 int err = 0;
464 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
466 WARN_ON(!mutex_is_locked(&nfsd_mutex));
468 if (nn->nfsd_serv == NULL || n <= 0)
469 return 0;
471 if (n > nn->nfsd_serv->sv_nrpools)
472 n = nn->nfsd_serv->sv_nrpools;
474 /* enforce a global maximum number of threads */
475 tot = 0;
476 for (i = 0; i < n; i++) {
477 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
478 tot += nthreads[i];
480 if (tot > NFSD_MAXSERVS) {
481 /* total too large: scale down requested numbers */
482 for (i = 0; i < n && tot > 0; i++) {
483 int new = nthreads[i] * NFSD_MAXSERVS / tot;
484 tot -= (nthreads[i] - new);
485 nthreads[i] = new;
487 for (i = 0; i < n && tot > 0; i++) {
488 nthreads[i]--;
489 tot--;
494 * There must always be a thread in pool 0; the admin
495 * can't shut down NFS completely using pool_threads.
497 if (nthreads[0] == 0)
498 nthreads[0] = 1;
500 /* apply the new numbers */
501 svc_get(nn->nfsd_serv);
502 for (i = 0; i < n; i++) {
503 err = svc_set_num_threads(nn->nfsd_serv, &nn->nfsd_serv->sv_pools[i],
504 nthreads[i]);
505 if (err)
506 break;
508 nfsd_destroy(net);
509 return err;
513 * Adjust the number of threads and return the new number of threads.
514 * This is also the function that starts the server if necessary, if
515 * this is the first time nrservs is nonzero.
518 nfsd_svc(int nrservs, struct net *net)
520 int error;
521 bool nfsd_up_before;
522 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
524 mutex_lock(&nfsd_mutex);
525 dprintk("nfsd: creating service\n");
527 nrservs = max(nrservs, 0);
528 nrservs = min(nrservs, NFSD_MAXSERVS);
529 error = 0;
531 if (nrservs == 0 && nn->nfsd_serv == NULL)
532 goto out;
534 error = nfsd_create_serv(net);
535 if (error)
536 goto out;
538 nfsd_up_before = nn->nfsd_net_up;
540 error = nfsd_startup_net(nrservs, net);
541 if (error)
542 goto out_destroy;
543 error = svc_set_num_threads(nn->nfsd_serv, NULL, nrservs);
544 if (error)
545 goto out_shutdown;
546 /* We are holding a reference to nn->nfsd_serv which
547 * we don't want to count in the return value,
548 * so subtract 1
550 error = nn->nfsd_serv->sv_nrthreads - 1;
551 out_shutdown:
552 if (error < 0 && !nfsd_up_before)
553 nfsd_shutdown_net(net);
554 out_destroy:
555 nfsd_destroy(net); /* Release server */
556 out:
557 mutex_unlock(&nfsd_mutex);
558 return error;
563 * This is the NFS server kernel thread
565 static int
566 nfsd(void *vrqstp)
568 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
569 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
570 struct net *net = perm_sock->xpt_net;
571 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
572 int err;
574 /* Lock module and set up kernel thread */
575 mutex_lock(&nfsd_mutex);
577 /* At this point, the thread shares current->fs
578 * with the init process. We need to create files with a
579 * umask of 0 instead of init's umask. */
580 if (unshare_fs_struct() < 0) {
581 printk("Unable to start nfsd thread: out of memory\n");
582 goto out;
585 current->fs->umask = 0;
588 * thread is spawned with all signals set to SIG_IGN, re-enable
589 * the ones that will bring down the thread
591 allow_signal(SIGKILL);
592 allow_signal(SIGHUP);
593 allow_signal(SIGINT);
594 allow_signal(SIGQUIT);
596 nfsdstats.th_cnt++;
597 mutex_unlock(&nfsd_mutex);
599 set_freezable();
602 * The main request loop
604 for (;;) {
605 /* Update sv_maxconn if it has changed */
606 rqstp->rq_server->sv_maxconn = nn->max_connections;
609 * Find a socket with data available and call its
610 * recvfrom routine.
612 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
614 if (err == -EINTR)
615 break;
616 validate_process_creds();
617 svc_process(rqstp);
618 validate_process_creds();
621 /* Clear signals before calling svc_exit_thread() */
622 flush_signals(current);
624 mutex_lock(&nfsd_mutex);
625 nfsdstats.th_cnt --;
627 out:
628 rqstp->rq_server = NULL;
630 /* Release the thread */
631 svc_exit_thread(rqstp);
633 nfsd_destroy(net);
635 /* Release module */
636 mutex_unlock(&nfsd_mutex);
637 module_put_and_exit(0);
638 return 0;
641 static __be32 map_new_errors(u32 vers, __be32 nfserr)
643 if (nfserr == nfserr_jukebox && vers == 2)
644 return nfserr_dropit;
645 if (nfserr == nfserr_wrongsec && vers < 4)
646 return nfserr_acces;
647 return nfserr;
651 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
653 struct svc_procedure *proc;
654 kxdrproc_t xdr;
655 __be32 nfserr;
656 __be32 *nfserrp;
658 dprintk("nfsd_dispatch: vers %d proc %d\n",
659 rqstp->rq_vers, rqstp->rq_proc);
660 proc = rqstp->rq_procinfo;
663 * Give the xdr decoder a chance to change this if it wants
664 * (necessary in the NFSv4.0 compound case)
666 rqstp->rq_cachetype = proc->pc_cachetype;
667 /* Decode arguments */
668 xdr = proc->pc_decode;
669 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
670 rqstp->rq_argp)) {
671 dprintk("nfsd: failed to decode arguments!\n");
672 *statp = rpc_garbage_args;
673 return 1;
676 /* Check whether we have this call in the cache. */
677 switch (nfsd_cache_lookup(rqstp)) {
678 case RC_DROPIT:
679 return 0;
680 case RC_REPLY:
681 return 1;
682 case RC_DOIT:;
683 /* do it */
686 /* need to grab the location to store the status, as
687 * nfsv4 does some encoding while processing
689 nfserrp = rqstp->rq_res.head[0].iov_base
690 + rqstp->rq_res.head[0].iov_len;
691 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
693 /* Now call the procedure handler, and encode NFS status. */
694 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
695 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
696 if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
697 dprintk("nfsd: Dropping request; may be revisited later\n");
698 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
699 return 0;
702 if (rqstp->rq_proc != 0)
703 *nfserrp++ = nfserr;
705 /* Encode result.
706 * For NFSv2, additional info is never returned in case of an error.
708 if (!(nfserr && rqstp->rq_vers == 2)) {
709 xdr = proc->pc_encode;
710 if (xdr && !xdr(rqstp, nfserrp,
711 rqstp->rq_resp)) {
712 /* Failed to encode result. Release cache entry */
713 dprintk("nfsd: failed to encode result!\n");
714 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
715 *statp = rpc_system_err;
716 return 1;
720 /* Store reply in cache. */
721 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
722 return 1;
725 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
727 int ret;
728 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
730 mutex_lock(&nfsd_mutex);
731 if (nn->nfsd_serv == NULL) {
732 mutex_unlock(&nfsd_mutex);
733 return -ENODEV;
735 /* bump up the psudo refcount while traversing */
736 svc_get(nn->nfsd_serv);
737 ret = svc_pool_stats_open(nn->nfsd_serv, file);
738 mutex_unlock(&nfsd_mutex);
739 return ret;
742 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
744 int ret = seq_release(inode, file);
745 struct net *net = inode->i_sb->s_fs_info;
747 mutex_lock(&nfsd_mutex);
748 /* this function really, really should have been called svc_put() */
749 nfsd_destroy(net);
750 mutex_unlock(&nfsd_mutex);
751 return ret;