Merge remote-tracking branch 'moduleh/module.h-split'
[linux-2.6/next.git] / fs / nfsd / nfssvc.c
blob4b8e828ae15f78264b79c25ea2f9b4e78a982350
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"
25 #define NFSDDBG_FACILITY NFSDDBG_SVC
27 extern struct svc_program nfsd_program;
28 static int nfsd(void *vrqstp);
29 struct timeval nfssvc_boot;
32 * nfsd_mutex protects 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) 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);
54 struct svc_serv *nfsd_serv;
57 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
58 * nfsd_drc_max_pages limits the total amount of memory available for
59 * version 4.1 DRC caches.
60 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
62 spinlock_t nfsd_drc_lock;
63 unsigned int nfsd_drc_max_mem;
64 unsigned int nfsd_drc_mem_used;
66 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
67 static struct svc_stat nfsd_acl_svcstats;
68 static struct svc_version * nfsd_acl_version[] = {
69 [2] = &nfsd_acl_version2,
70 [3] = &nfsd_acl_version3,
73 #define NFSD_ACL_MINVERS 2
74 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
75 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
77 static struct svc_program nfsd_acl_program = {
78 .pg_prog = NFS_ACL_PROGRAM,
79 .pg_nvers = NFSD_ACL_NRVERS,
80 .pg_vers = nfsd_acl_versions,
81 .pg_name = "nfsacl",
82 .pg_class = "nfsd",
83 .pg_stats = &nfsd_acl_svcstats,
84 .pg_authenticate = &svc_set_client,
87 static struct svc_stat nfsd_acl_svcstats = {
88 .program = &nfsd_acl_program,
90 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
92 static struct svc_version * nfsd_version[] = {
93 [2] = &nfsd_version2,
94 #if defined(CONFIG_NFSD_V3)
95 [3] = &nfsd_version3,
96 #endif
97 #if defined(CONFIG_NFSD_V4)
98 [4] = &nfsd_version4,
99 #endif
102 #define NFSD_MINVERS 2
103 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
104 static struct svc_version *nfsd_versions[NFSD_NRVERS];
106 struct svc_program nfsd_program = {
107 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
108 .pg_next = &nfsd_acl_program,
109 #endif
110 .pg_prog = NFS_PROGRAM, /* program number */
111 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
112 .pg_vers = nfsd_versions, /* version table */
113 .pg_name = "nfsd", /* program name */
114 .pg_class = "nfsd", /* authentication class */
115 .pg_stats = &nfsd_svcstats, /* version table */
116 .pg_authenticate = &svc_set_client, /* export authentication */
120 u32 nfsd_supported_minorversion;
122 int nfsd_vers(int vers, enum vers_op change)
124 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
125 return 0;
126 switch(change) {
127 case NFSD_SET:
128 nfsd_versions[vers] = nfsd_version[vers];
129 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
130 if (vers < NFSD_ACL_NRVERS)
131 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
132 #endif
133 break;
134 case NFSD_CLEAR:
135 nfsd_versions[vers] = NULL;
136 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
137 if (vers < NFSD_ACL_NRVERS)
138 nfsd_acl_versions[vers] = NULL;
139 #endif
140 break;
141 case NFSD_TEST:
142 return nfsd_versions[vers] != NULL;
143 case NFSD_AVAIL:
144 return nfsd_version[vers] != NULL;
146 return 0;
149 int nfsd_minorversion(u32 minorversion, enum vers_op change)
151 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
152 return -1;
153 switch(change) {
154 case NFSD_SET:
155 nfsd_supported_minorversion = minorversion;
156 break;
157 case NFSD_CLEAR:
158 if (minorversion == 0)
159 return -1;
160 nfsd_supported_minorversion = minorversion - 1;
161 break;
162 case NFSD_TEST:
163 return minorversion <= nfsd_supported_minorversion;
164 case NFSD_AVAIL:
165 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
167 return 0;
171 * Maximum number of nfsd processes
173 #define NFSD_MAXSERVS 8192
175 int nfsd_nrthreads(void)
177 int rv = 0;
178 mutex_lock(&nfsd_mutex);
179 if (nfsd_serv)
180 rv = nfsd_serv->sv_nrthreads;
181 mutex_unlock(&nfsd_mutex);
182 return rv;
185 static int nfsd_init_socks(int port)
187 int error;
188 if (!list_empty(&nfsd_serv->sv_permsocks))
189 return 0;
191 error = svc_create_xprt(nfsd_serv, "udp", &init_net, PF_INET, port,
192 SVC_SOCK_DEFAULTS);
193 if (error < 0)
194 return error;
196 error = svc_create_xprt(nfsd_serv, "tcp", &init_net, PF_INET, port,
197 SVC_SOCK_DEFAULTS);
198 if (error < 0)
199 return error;
201 return 0;
204 static bool nfsd_up = false;
206 static int nfsd_startup(unsigned short port, int nrservs)
208 int ret;
210 if (nfsd_up)
211 return 0;
213 * Readahead param cache - will no-op if it already exists.
214 * (Note therefore results will be suboptimal if number of
215 * threads is modified after nfsd start.)
217 ret = nfsd_racache_init(2*nrservs);
218 if (ret)
219 return ret;
220 ret = nfsd_init_socks(port);
221 if (ret)
222 goto out_racache;
223 ret = lockd_up();
224 if (ret)
225 goto out_racache;
226 ret = nfs4_state_start();
227 if (ret)
228 goto out_lockd;
229 nfsd_up = true;
230 return 0;
231 out_lockd:
232 lockd_down();
233 out_racache:
234 nfsd_racache_shutdown();
235 return ret;
238 static void nfsd_shutdown(void)
241 * write_ports can create the server without actually starting
242 * any threads--if we get shut down before any threads are
243 * started, then nfsd_last_thread will be run before any of this
244 * other initialization has been done.
246 if (!nfsd_up)
247 return;
248 nfs4_state_shutdown();
249 lockd_down();
250 nfsd_racache_shutdown();
251 nfsd_up = false;
254 static void nfsd_last_thread(struct svc_serv *serv)
256 /* When last nfsd thread exits we need to do some clean-up */
257 nfsd_serv = NULL;
258 nfsd_shutdown();
260 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
261 "cache\n");
262 nfsd_export_flush();
265 void nfsd_reset_versions(void)
267 int found_one = 0;
268 int i;
270 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
271 if (nfsd_program.pg_vers[i])
272 found_one = 1;
275 if (!found_one) {
276 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
277 nfsd_program.pg_vers[i] = nfsd_version[i];
278 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
279 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
280 nfsd_acl_program.pg_vers[i] =
281 nfsd_acl_version[i];
282 #endif
287 * Each session guarantees a negotiated per slot memory cache for replies
288 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
289 * NFSv4.1 server might want to use more memory for a DRC than a machine
290 * with mutiple services.
292 * Impose a hard limit on the number of pages for the DRC which varies
293 * according to the machines free pages. This is of course only a default.
295 * For now this is a #defined shift which could be under admin control
296 * in the future.
298 static void set_max_drc(void)
300 #define NFSD_DRC_SIZE_SHIFT 10
301 nfsd_drc_max_mem = (nr_free_buffer_pages()
302 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
303 nfsd_drc_mem_used = 0;
304 spin_lock_init(&nfsd_drc_lock);
305 dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
308 int nfsd_create_serv(void)
310 int err = 0;
312 WARN_ON(!mutex_is_locked(&nfsd_mutex));
313 if (nfsd_serv) {
314 svc_get(nfsd_serv);
315 return 0;
317 if (nfsd_max_blksize == 0) {
318 /* choose a suitable default */
319 struct sysinfo i;
320 si_meminfo(&i);
321 /* Aim for 1/4096 of memory per thread
322 * This gives 1MB on 4Gig machines
323 * But only uses 32K on 128M machines.
324 * Bottom out at 8K on 32M and smaller.
325 * Of course, this is only a default.
327 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
328 i.totalram <<= PAGE_SHIFT - 12;
329 while (nfsd_max_blksize > i.totalram &&
330 nfsd_max_blksize >= 8*1024*2)
331 nfsd_max_blksize /= 2;
333 nfsd_reset_versions();
335 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
336 nfsd_last_thread, nfsd, THIS_MODULE);
337 if (nfsd_serv == NULL)
338 return -ENOMEM;
340 set_max_drc();
341 do_gettimeofday(&nfssvc_boot); /* record boot time */
342 return err;
345 int nfsd_nrpools(void)
347 if (nfsd_serv == NULL)
348 return 0;
349 else
350 return nfsd_serv->sv_nrpools;
353 int nfsd_get_nrthreads(int n, int *nthreads)
355 int i = 0;
357 if (nfsd_serv != NULL) {
358 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
359 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
362 return 0;
365 int nfsd_set_nrthreads(int n, int *nthreads)
367 int i = 0;
368 int tot = 0;
369 int err = 0;
371 WARN_ON(!mutex_is_locked(&nfsd_mutex));
373 if (nfsd_serv == NULL || n <= 0)
374 return 0;
376 if (n > nfsd_serv->sv_nrpools)
377 n = nfsd_serv->sv_nrpools;
379 /* enforce a global maximum number of threads */
380 tot = 0;
381 for (i = 0; i < n; i++) {
382 if (nthreads[i] > NFSD_MAXSERVS)
383 nthreads[i] = NFSD_MAXSERVS;
384 tot += nthreads[i];
386 if (tot > NFSD_MAXSERVS) {
387 /* total too large: scale down requested numbers */
388 for (i = 0; i < n && tot > 0; i++) {
389 int new = nthreads[i] * NFSD_MAXSERVS / tot;
390 tot -= (nthreads[i] - new);
391 nthreads[i] = new;
393 for (i = 0; i < n && tot > 0; i++) {
394 nthreads[i]--;
395 tot--;
400 * There must always be a thread in pool 0; the admin
401 * can't shut down NFS completely using pool_threads.
403 if (nthreads[0] == 0)
404 nthreads[0] = 1;
406 /* apply the new numbers */
407 svc_get(nfsd_serv);
408 for (i = 0; i < n; i++) {
409 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
410 nthreads[i]);
411 if (err)
412 break;
414 svc_destroy(nfsd_serv);
416 return err;
420 * Adjust the number of threads and return the new number of threads.
421 * This is also the function that starts the server if necessary, if
422 * this is the first time nrservs is nonzero.
425 nfsd_svc(unsigned short port, int nrservs)
427 int error;
428 bool nfsd_up_before;
430 mutex_lock(&nfsd_mutex);
431 dprintk("nfsd: creating service\n");
432 if (nrservs <= 0)
433 nrservs = 0;
434 if (nrservs > NFSD_MAXSERVS)
435 nrservs = NFSD_MAXSERVS;
436 error = 0;
437 if (nrservs == 0 && nfsd_serv == NULL)
438 goto out;
440 error = nfsd_create_serv();
441 if (error)
442 goto out;
444 nfsd_up_before = nfsd_up;
446 error = nfsd_startup(port, nrservs);
447 if (error)
448 goto out_destroy;
449 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
450 if (error)
451 goto out_shutdown;
452 /* We are holding a reference to nfsd_serv which
453 * we don't want to count in the return value,
454 * so subtract 1
456 error = nfsd_serv->sv_nrthreads - 1;
457 out_shutdown:
458 if (error < 0 && !nfsd_up_before)
459 nfsd_shutdown();
460 out_destroy:
461 svc_destroy(nfsd_serv); /* Release server */
462 out:
463 mutex_unlock(&nfsd_mutex);
464 return error;
469 * This is the NFS server kernel thread
471 static int
472 nfsd(void *vrqstp)
474 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
475 int err, preverr = 0;
477 /* Lock module and set up kernel thread */
478 mutex_lock(&nfsd_mutex);
480 /* At this point, the thread shares current->fs
481 * with the init process. We need to create files with a
482 * umask of 0 instead of init's umask. */
483 if (unshare_fs_struct() < 0) {
484 printk("Unable to start nfsd thread: out of memory\n");
485 goto out;
488 current->fs->umask = 0;
491 * thread is spawned with all signals set to SIG_IGN, re-enable
492 * the ones that will bring down the thread
494 allow_signal(SIGKILL);
495 allow_signal(SIGHUP);
496 allow_signal(SIGINT);
497 allow_signal(SIGQUIT);
499 nfsdstats.th_cnt++;
500 mutex_unlock(&nfsd_mutex);
503 * We want less throttling in balance_dirty_pages() so that nfs to
504 * localhost doesn't cause nfsd to lock up due to all the client's
505 * dirty pages.
507 current->flags |= PF_LESS_THROTTLE;
508 set_freezable();
511 * The main request loop
513 for (;;) {
515 * Find a socket with data available and call its
516 * recvfrom routine.
518 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
520 if (err == -EINTR)
521 break;
522 else if (err < 0) {
523 if (err != preverr) {
524 printk(KERN_WARNING "%s: unexpected error "
525 "from svc_recv (%d)\n", __func__, -err);
526 preverr = err;
528 schedule_timeout_uninterruptible(HZ);
529 continue;
532 validate_process_creds();
533 svc_process(rqstp);
534 validate_process_creds();
537 /* Clear signals before calling svc_exit_thread() */
538 flush_signals(current);
540 mutex_lock(&nfsd_mutex);
541 nfsdstats.th_cnt --;
543 out:
544 /* Release the thread */
545 svc_exit_thread(rqstp);
547 /* Release module */
548 mutex_unlock(&nfsd_mutex);
549 module_put_and_exit(0);
550 return 0;
553 static __be32 map_new_errors(u32 vers, __be32 nfserr)
555 if (nfserr == nfserr_jukebox && vers == 2)
556 return nfserr_dropit;
557 if (nfserr == nfserr_wrongsec && vers < 4)
558 return nfserr_acces;
559 return nfserr;
563 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
565 struct svc_procedure *proc;
566 kxdrproc_t xdr;
567 __be32 nfserr;
568 __be32 *nfserrp;
570 dprintk("nfsd_dispatch: vers %d proc %d\n",
571 rqstp->rq_vers, rqstp->rq_proc);
572 proc = rqstp->rq_procinfo;
575 * Give the xdr decoder a chance to change this if it wants
576 * (necessary in the NFSv4.0 compound case)
578 rqstp->rq_cachetype = proc->pc_cachetype;
579 /* Decode arguments */
580 xdr = proc->pc_decode;
581 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
582 rqstp->rq_argp)) {
583 dprintk("nfsd: failed to decode arguments!\n");
584 *statp = rpc_garbage_args;
585 return 1;
588 /* Check whether we have this call in the cache. */
589 switch (nfsd_cache_lookup(rqstp)) {
590 case RC_INTR:
591 case RC_DROPIT:
592 return 0;
593 case RC_REPLY:
594 return 1;
595 case RC_DOIT:;
596 /* do it */
599 /* need to grab the location to store the status, as
600 * nfsv4 does some encoding while processing
602 nfserrp = rqstp->rq_res.head[0].iov_base
603 + rqstp->rq_res.head[0].iov_len;
604 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
606 /* Now call the procedure handler, and encode NFS status. */
607 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
608 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
609 if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
610 dprintk("nfsd: Dropping request; may be revisited later\n");
611 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
612 return 0;
615 if (rqstp->rq_proc != 0)
616 *nfserrp++ = nfserr;
618 /* Encode result.
619 * For NFSv2, additional info is never returned in case of an error.
621 if (!(nfserr && rqstp->rq_vers == 2)) {
622 xdr = proc->pc_encode;
623 if (xdr && !xdr(rqstp, nfserrp,
624 rqstp->rq_resp)) {
625 /* Failed to encode result. Release cache entry */
626 dprintk("nfsd: failed to encode result!\n");
627 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
628 *statp = rpc_system_err;
629 return 1;
633 /* Store reply in cache. */
634 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
635 return 1;
638 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
640 int ret;
641 mutex_lock(&nfsd_mutex);
642 if (nfsd_serv == NULL) {
643 mutex_unlock(&nfsd_mutex);
644 return -ENODEV;
646 /* bump up the psudo refcount while traversing */
647 svc_get(nfsd_serv);
648 ret = svc_pool_stats_open(nfsd_serv, file);
649 mutex_unlock(&nfsd_mutex);
650 return ret;
653 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
655 int ret = seq_release(inode, file);
656 mutex_lock(&nfsd_mutex);
657 /* this function really, really should have been called svc_put() */
658 svc_destroy(nfsd_serv);
659 mutex_unlock(&nfsd_mutex);
660 return ret;