Loading...
1/*
2 * Central processing for nfsd.
3 *
4 * Authors: Olaf Kirch (okir@monad.swb.de)
5 *
6 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
7 */
8
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>
14
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"
25
26#define NFSDDBG_FACILITY NFSDDBG_SVC
27
28extern struct svc_program nfsd_program;
29static int nfsd(void *vrqstp);
30
31/*
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
35 *
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[].
40 *
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.
44 *
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:
48 *
49 * user_recovery_dirname
50 * user_lease_time
51 * nfsd_versions
52 */
53DEFINE_MUTEX(nfsd_mutex);
54
55/*
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.
60 */
61spinlock_t nfsd_drc_lock;
62unsigned long nfsd_drc_max_mem;
63unsigned long nfsd_drc_mem_used;
64
65#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
66static struct svc_stat nfsd_acl_svcstats;
67static struct svc_version * nfsd_acl_version[] = {
68 [2] = &nfsd_acl_version2,
69 [3] = &nfsd_acl_version3,
70};
71
72#define NFSD_ACL_MINVERS 2
73#define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
74static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
75
76static 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,
84};
85
86static struct svc_stat nfsd_acl_svcstats = {
87 .program = &nfsd_acl_program,
88};
89#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
90
91static 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
99};
100
101#define NFSD_MINVERS 2
102#define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
103static struct svc_version *nfsd_versions[NFSD_NRVERS];
104
105struct 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 */
116
117};
118
119static bool nfsd_supported_minorversions[NFSD_SUPPORTED_MINOR_VERSION + 1] = {
120 [0] = 1,
121 [1] = 1,
122};
123
124int nfsd_vers(int vers, enum vers_op change)
125{
126 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
127 return 0;
128 switch(change) {
129 case NFSD_SET:
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];
134#endif
135 break;
136 case NFSD_CLEAR:
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;
141#endif
142 break;
143 case NFSD_TEST:
144 return nfsd_versions[vers] != NULL;
145 case NFSD_AVAIL:
146 return nfsd_version[vers] != NULL;
147 }
148 return 0;
149}
150
151int nfsd_minorversion(u32 minorversion, enum vers_op change)
152{
153 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
154 return -1;
155 switch(change) {
156 case NFSD_SET:
157 nfsd_supported_minorversions[minorversion] = true;
158 break;
159 case NFSD_CLEAR:
160 nfsd_supported_minorversions[minorversion] = false;
161 break;
162 case NFSD_TEST:
163 return nfsd_supported_minorversions[minorversion];
164 case NFSD_AVAIL:
165 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
166 }
167 return 0;
168}
169
170/*
171 * Maximum number of nfsd processes
172 */
173#define NFSD_MAXSERVS 8192
174
175int nfsd_nrthreads(struct net *net)
176{
177 int rv = 0;
178 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
179
180 mutex_lock(&nfsd_mutex);
181 if (nn->nfsd_serv)
182 rv = nn->nfsd_serv->sv_nrthreads;
183 mutex_unlock(&nfsd_mutex);
184 return rv;
185}
186
187static int nfsd_init_socks(struct net *net)
188{
189 int error;
190 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
191
192 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
193 return 0;
194
195 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
196 SVC_SOCK_DEFAULTS);
197 if (error < 0)
198 return error;
199
200 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
201 SVC_SOCK_DEFAULTS);
202 if (error < 0)
203 return error;
204
205 return 0;
206}
207
208static int nfsd_users = 0;
209
210static int nfsd_startup_generic(int nrservs)
211{
212 int ret;
213
214 if (nfsd_users++)
215 return 0;
216
217 /*
218 * Readahead param cache - will no-op if it already exists.
219 * (Note therefore results will be suboptimal if number of
220 * threads is modified after nfsd start.)
221 */
222 ret = nfsd_racache_init(2*nrservs);
223 if (ret)
224 return ret;
225 ret = nfs4_state_start();
226 if (ret)
227 goto out_racache;
228 return 0;
229
230out_racache:
231 nfsd_racache_shutdown();
232 return ret;
233}
234
235static void nfsd_shutdown_generic(void)
236{
237 if (--nfsd_users)
238 return;
239
240 nfs4_state_shutdown();
241 nfsd_racache_shutdown();
242}
243
244static bool nfsd_needs_lockd(void)
245{
246#if defined(CONFIG_NFSD_V3)
247 return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
248#else
249 return (nfsd_versions[2] != NULL);
250#endif
251}
252
253static int nfsd_startup_net(int nrservs, struct net *net)
254{
255 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
256 int ret;
257
258 if (nn->nfsd_net_up)
259 return 0;
260
261 ret = nfsd_startup_generic(nrservs);
262 if (ret)
263 return ret;
264 ret = nfsd_init_socks(net);
265 if (ret)
266 goto out_socks;
267
268 if (nfsd_needs_lockd() && !nn->lockd_up) {
269 ret = lockd_up(net);
270 if (ret)
271 goto out_socks;
272 nn->lockd_up = 1;
273 }
274
275 ret = nfs4_state_start_net(net);
276 if (ret)
277 goto out_lockd;
278
279 nn->nfsd_net_up = true;
280 return 0;
281
282out_lockd:
283 if (nn->lockd_up) {
284 lockd_down(net);
285 nn->lockd_up = 0;
286 }
287out_socks:
288 nfsd_shutdown_generic();
289 return ret;
290}
291
292static void nfsd_shutdown_net(struct net *net)
293{
294 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
295
296 nfs4_state_shutdown_net(net);
297 if (nn->lockd_up) {
298 lockd_down(net);
299 nn->lockd_up = 0;
300 }
301 nn->nfsd_net_up = false;
302 nfsd_shutdown_generic();
303}
304
305static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
306{
307 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
308
309 /*
310 * write_ports can create the server without actually starting
311 * any threads--if we get shut down before any threads are
312 * started, then nfsd_last_thread will be run before any of this
313 * other initialization has been done.
314 */
315 if (!nn->nfsd_net_up)
316 return;
317 nfsd_shutdown_net(net);
318
319 svc_rpcb_cleanup(serv, net);
320
321 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
322 "cache\n");
323 nfsd_export_flush(net);
324}
325
326void nfsd_reset_versions(void)
327{
328 int found_one = 0;
329 int i;
330
331 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
332 if (nfsd_program.pg_vers[i])
333 found_one = 1;
334 }
335
336 if (!found_one) {
337 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
338 nfsd_program.pg_vers[i] = nfsd_version[i];
339#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
340 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
341 nfsd_acl_program.pg_vers[i] =
342 nfsd_acl_version[i];
343#endif
344 }
345}
346
347/*
348 * Each session guarantees a negotiated per slot memory cache for replies
349 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
350 * NFSv4.1 server might want to use more memory for a DRC than a machine
351 * with mutiple services.
352 *
353 * Impose a hard limit on the number of pages for the DRC which varies
354 * according to the machines free pages. This is of course only a default.
355 *
356 * For now this is a #defined shift which could be under admin control
357 * in the future.
358 */
359static void set_max_drc(void)
360{
361 #define NFSD_DRC_SIZE_SHIFT 10
362 nfsd_drc_max_mem = (nr_free_buffer_pages()
363 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
364 nfsd_drc_mem_used = 0;
365 spin_lock_init(&nfsd_drc_lock);
366 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
367}
368
369static int nfsd_get_default_max_blksize(void)
370{
371 struct sysinfo i;
372 unsigned long long target;
373 unsigned long ret;
374
375 si_meminfo(&i);
376 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
377 /*
378 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
379 * machines, but only uses 32K on 128M machines. Bottom out at
380 * 8K on 32M and smaller. Of course, this is only a default.
381 */
382 target >>= 12;
383
384 ret = NFSSVC_MAXBLKSIZE;
385 while (ret > target && ret >= 8*1024*2)
386 ret /= 2;
387 return ret;
388}
389
390int nfsd_create_serv(struct net *net)
391{
392 int error;
393 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
394
395 WARN_ON(!mutex_is_locked(&nfsd_mutex));
396 if (nn->nfsd_serv) {
397 svc_get(nn->nfsd_serv);
398 return 0;
399 }
400 if (nfsd_max_blksize == 0)
401 nfsd_max_blksize = nfsd_get_default_max_blksize();
402 nfsd_reset_versions();
403 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
404 nfsd_last_thread, nfsd, THIS_MODULE);
405 if (nn->nfsd_serv == NULL)
406 return -ENOMEM;
407
408 error = svc_bind(nn->nfsd_serv, net);
409 if (error < 0) {
410 svc_destroy(nn->nfsd_serv);
411 return error;
412 }
413
414 set_max_drc();
415 do_gettimeofday(&nn->nfssvc_boot); /* record boot time */
416 return 0;
417}
418
419int nfsd_nrpools(struct net *net)
420{
421 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
422
423 if (nn->nfsd_serv == NULL)
424 return 0;
425 else
426 return nn->nfsd_serv->sv_nrpools;
427}
428
429int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
430{
431 int i = 0;
432 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
433
434 if (nn->nfsd_serv != NULL) {
435 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
436 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
437 }
438
439 return 0;
440}
441
442void nfsd_destroy(struct net *net)
443{
444 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
445 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
446
447 if (destroy)
448 svc_shutdown_net(nn->nfsd_serv, net);
449 svc_destroy(nn->nfsd_serv);
450 if (destroy)
451 nn->nfsd_serv = NULL;
452}
453
454int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
455{
456 int i = 0;
457 int tot = 0;
458 int err = 0;
459 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
460
461 WARN_ON(!mutex_is_locked(&nfsd_mutex));
462
463 if (nn->nfsd_serv == NULL || n <= 0)
464 return 0;
465
466 if (n > nn->nfsd_serv->sv_nrpools)
467 n = nn->nfsd_serv->sv_nrpools;
468
469 /* enforce a global maximum number of threads */
470 tot = 0;
471 for (i = 0; i < n; i++) {
472 if (nthreads[i] > NFSD_MAXSERVS)
473 nthreads[i] = NFSD_MAXSERVS;
474 tot += nthreads[i];
475 }
476 if (tot > NFSD_MAXSERVS) {
477 /* total too large: scale down requested numbers */
478 for (i = 0; i < n && tot > 0; i++) {
479 int new = nthreads[i] * NFSD_MAXSERVS / tot;
480 tot -= (nthreads[i] - new);
481 nthreads[i] = new;
482 }
483 for (i = 0; i < n && tot > 0; i++) {
484 nthreads[i]--;
485 tot--;
486 }
487 }
488
489 /*
490 * There must always be a thread in pool 0; the admin
491 * can't shut down NFS completely using pool_threads.
492 */
493 if (nthreads[0] == 0)
494 nthreads[0] = 1;
495
496 /* apply the new numbers */
497 svc_get(nn->nfsd_serv);
498 for (i = 0; i < n; i++) {
499 err = svc_set_num_threads(nn->nfsd_serv, &nn->nfsd_serv->sv_pools[i],
500 nthreads[i]);
501 if (err)
502 break;
503 }
504 nfsd_destroy(net);
505 return err;
506}
507
508/*
509 * Adjust the number of threads and return the new number of threads.
510 * This is also the function that starts the server if necessary, if
511 * this is the first time nrservs is nonzero.
512 */
513int
514nfsd_svc(int nrservs, struct net *net)
515{
516 int error;
517 bool nfsd_up_before;
518 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
519
520 mutex_lock(&nfsd_mutex);
521 dprintk("nfsd: creating service\n");
522 if (nrservs <= 0)
523 nrservs = 0;
524 if (nrservs > NFSD_MAXSERVS)
525 nrservs = NFSD_MAXSERVS;
526 error = 0;
527 if (nrservs == 0 && nn->nfsd_serv == NULL)
528 goto out;
529
530 error = nfsd_create_serv(net);
531 if (error)
532 goto out;
533
534 nfsd_up_before = nn->nfsd_net_up;
535
536 error = nfsd_startup_net(nrservs, net);
537 if (error)
538 goto out_destroy;
539 error = svc_set_num_threads(nn->nfsd_serv, NULL, nrservs);
540 if (error)
541 goto out_shutdown;
542 /* We are holding a reference to nn->nfsd_serv which
543 * we don't want to count in the return value,
544 * so subtract 1
545 */
546 error = nn->nfsd_serv->sv_nrthreads - 1;
547out_shutdown:
548 if (error < 0 && !nfsd_up_before)
549 nfsd_shutdown_net(net);
550out_destroy:
551 nfsd_destroy(net); /* Release server */
552out:
553 mutex_unlock(&nfsd_mutex);
554 return error;
555}
556
557
558/*
559 * This is the NFS server kernel thread
560 */
561static int
562nfsd(void *vrqstp)
563{
564 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
565 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
566 struct net *net = perm_sock->xpt_net;
567 int err;
568
569 /* Lock module and set up kernel thread */
570 mutex_lock(&nfsd_mutex);
571
572 /* At this point, the thread shares current->fs
573 * with the init process. We need to create files with a
574 * umask of 0 instead of init's umask. */
575 if (unshare_fs_struct() < 0) {
576 printk("Unable to start nfsd thread: out of memory\n");
577 goto out;
578 }
579
580 current->fs->umask = 0;
581
582 /*
583 * thread is spawned with all signals set to SIG_IGN, re-enable
584 * the ones that will bring down the thread
585 */
586 allow_signal(SIGKILL);
587 allow_signal(SIGHUP);
588 allow_signal(SIGINT);
589 allow_signal(SIGQUIT);
590
591 nfsdstats.th_cnt++;
592 mutex_unlock(&nfsd_mutex);
593
594 /*
595 * We want less throttling in balance_dirty_pages() so that nfs to
596 * localhost doesn't cause nfsd to lock up due to all the client's
597 * dirty pages.
598 */
599 current->flags |= PF_LESS_THROTTLE;
600 set_freezable();
601
602 /*
603 * The main request loop
604 */
605 for (;;) {
606 /*
607 * Find a socket with data available and call its
608 * recvfrom routine.
609 */
610 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
611 ;
612 if (err == -EINTR)
613 break;
614 validate_process_creds();
615 svc_process(rqstp);
616 validate_process_creds();
617 }
618
619 /* Clear signals before calling svc_exit_thread() */
620 flush_signals(current);
621
622 mutex_lock(&nfsd_mutex);
623 nfsdstats.th_cnt --;
624
625out:
626 rqstp->rq_server = NULL;
627
628 /* Release the thread */
629 svc_exit_thread(rqstp);
630
631 nfsd_destroy(net);
632
633 /* Release module */
634 mutex_unlock(&nfsd_mutex);
635 module_put_and_exit(0);
636 return 0;
637}
638
639static __be32 map_new_errors(u32 vers, __be32 nfserr)
640{
641 if (nfserr == nfserr_jukebox && vers == 2)
642 return nfserr_dropit;
643 if (nfserr == nfserr_wrongsec && vers < 4)
644 return nfserr_acces;
645 return nfserr;
646}
647
648int
649nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
650{
651 struct svc_procedure *proc;
652 kxdrproc_t xdr;
653 __be32 nfserr;
654 __be32 *nfserrp;
655
656 dprintk("nfsd_dispatch: vers %d proc %d\n",
657 rqstp->rq_vers, rqstp->rq_proc);
658 proc = rqstp->rq_procinfo;
659
660 /*
661 * Give the xdr decoder a chance to change this if it wants
662 * (necessary in the NFSv4.0 compound case)
663 */
664 rqstp->rq_cachetype = proc->pc_cachetype;
665 /* Decode arguments */
666 xdr = proc->pc_decode;
667 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
668 rqstp->rq_argp)) {
669 dprintk("nfsd: failed to decode arguments!\n");
670 *statp = rpc_garbage_args;
671 return 1;
672 }
673
674 /* Check whether we have this call in the cache. */
675 switch (nfsd_cache_lookup(rqstp)) {
676 case RC_DROPIT:
677 return 0;
678 case RC_REPLY:
679 return 1;
680 case RC_DOIT:;
681 /* do it */
682 }
683
684 /* need to grab the location to store the status, as
685 * nfsv4 does some encoding while processing
686 */
687 nfserrp = rqstp->rq_res.head[0].iov_base
688 + rqstp->rq_res.head[0].iov_len;
689 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
690
691 /* Now call the procedure handler, and encode NFS status. */
692 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
693 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
694 if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
695 dprintk("nfsd: Dropping request; may be revisited later\n");
696 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
697 return 0;
698 }
699
700 if (rqstp->rq_proc != 0)
701 *nfserrp++ = nfserr;
702
703 /* Encode result.
704 * For NFSv2, additional info is never returned in case of an error.
705 */
706 if (!(nfserr && rqstp->rq_vers == 2)) {
707 xdr = proc->pc_encode;
708 if (xdr && !xdr(rqstp, nfserrp,
709 rqstp->rq_resp)) {
710 /* Failed to encode result. Release cache entry */
711 dprintk("nfsd: failed to encode result!\n");
712 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
713 *statp = rpc_system_err;
714 return 1;
715 }
716 }
717
718 /* Store reply in cache. */
719 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
720 return 1;
721}
722
723int nfsd_pool_stats_open(struct inode *inode, struct file *file)
724{
725 int ret;
726 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
727
728 mutex_lock(&nfsd_mutex);
729 if (nn->nfsd_serv == NULL) {
730 mutex_unlock(&nfsd_mutex);
731 return -ENODEV;
732 }
733 /* bump up the psudo refcount while traversing */
734 svc_get(nn->nfsd_serv);
735 ret = svc_pool_stats_open(nn->nfsd_serv, file);
736 mutex_unlock(&nfsd_mutex);
737 return ret;
738}
739
740int nfsd_pool_stats_release(struct inode *inode, struct file *file)
741{
742 int ret = seq_release(inode, file);
743 struct net *net = inode->i_sb->s_fs_info;
744
745 mutex_lock(&nfsd_mutex);
746 /* this function really, really should have been called svc_put() */
747 nfsd_destroy(net);
748 mutex_unlock(&nfsd_mutex);
749 return ret;
750}
1/*
2 * Central processing for nfsd.
3 *
4 * Authors: Olaf Kirch (okir@monad.swb.de)
5 *
6 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
7 */
8
9#include <linux/sched.h>
10#include <linux/freezer.h>
11#include <linux/fs_struct.h>
12#include <linux/swap.h>
13
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>
20#include "nfsd.h"
21#include "cache.h"
22#include "vfs.h"
23
24#define NFSDDBG_FACILITY NFSDDBG_SVC
25
26extern struct svc_program nfsd_program;
27static int nfsd(void *vrqstp);
28struct timeval nfssvc_boot;
29
30/*
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
34 *
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[].
39 *
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
42 * point, or freed.
43 *
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:
47 *
48 * user_recovery_dirname
49 * user_lease_time
50 * nfsd_versions
51 */
52DEFINE_MUTEX(nfsd_mutex);
53struct svc_serv *nfsd_serv;
54
55/*
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.
60 */
61spinlock_t nfsd_drc_lock;
62unsigned int nfsd_drc_max_mem;
63unsigned int nfsd_drc_mem_used;
64
65#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
66static struct svc_stat nfsd_acl_svcstats;
67static struct svc_version * nfsd_acl_version[] = {
68 [2] = &nfsd_acl_version2,
69 [3] = &nfsd_acl_version3,
70};
71
72#define NFSD_ACL_MINVERS 2
73#define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
74static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
75
76static 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,
84};
85
86static struct svc_stat nfsd_acl_svcstats = {
87 .program = &nfsd_acl_program,
88};
89#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
90
91static 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
99};
100
101#define NFSD_MINVERS 2
102#define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
103static struct svc_version *nfsd_versions[NFSD_NRVERS];
104
105struct 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 */
116
117};
118
119u32 nfsd_supported_minorversion;
120
121int nfsd_vers(int vers, enum vers_op change)
122{
123 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
124 return 0;
125 switch(change) {
126 case NFSD_SET:
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];
131#endif
132 break;
133 case NFSD_CLEAR:
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;
138#endif
139 break;
140 case NFSD_TEST:
141 return nfsd_versions[vers] != NULL;
142 case NFSD_AVAIL:
143 return nfsd_version[vers] != NULL;
144 }
145 return 0;
146}
147
148int nfsd_minorversion(u32 minorversion, enum vers_op change)
149{
150 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
151 return -1;
152 switch(change) {
153 case NFSD_SET:
154 nfsd_supported_minorversion = minorversion;
155 break;
156 case NFSD_CLEAR:
157 if (minorversion == 0)
158 return -1;
159 nfsd_supported_minorversion = minorversion - 1;
160 break;
161 case NFSD_TEST:
162 return minorversion <= nfsd_supported_minorversion;
163 case NFSD_AVAIL:
164 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
165 }
166 return 0;
167}
168
169/*
170 * Maximum number of nfsd processes
171 */
172#define NFSD_MAXSERVS 8192
173
174int nfsd_nrthreads(void)
175{
176 int rv = 0;
177 mutex_lock(&nfsd_mutex);
178 if (nfsd_serv)
179 rv = nfsd_serv->sv_nrthreads;
180 mutex_unlock(&nfsd_mutex);
181 return rv;
182}
183
184static int nfsd_init_socks(int port)
185{
186 int error;
187 if (!list_empty(&nfsd_serv->sv_permsocks))
188 return 0;
189
190 error = svc_create_xprt(nfsd_serv, "udp", &init_net, PF_INET, port,
191 SVC_SOCK_DEFAULTS);
192 if (error < 0)
193 return error;
194
195 error = svc_create_xprt(nfsd_serv, "tcp", &init_net, PF_INET, port,
196 SVC_SOCK_DEFAULTS);
197 if (error < 0)
198 return error;
199
200 return 0;
201}
202
203static bool nfsd_up = false;
204
205static int nfsd_startup(unsigned short port, int nrservs)
206{
207 int ret;
208
209 if (nfsd_up)
210 return 0;
211 /*
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.)
215 */
216 ret = nfsd_racache_init(2*nrservs);
217 if (ret)
218 return ret;
219 ret = nfsd_init_socks(port);
220 if (ret)
221 goto out_racache;
222 ret = lockd_up();
223 if (ret)
224 goto out_racache;
225 ret = nfs4_state_start();
226 if (ret)
227 goto out_lockd;
228 nfsd_up = true;
229 return 0;
230out_lockd:
231 lockd_down();
232out_racache:
233 nfsd_racache_shutdown();
234 return ret;
235}
236
237static void nfsd_shutdown(void)
238{
239 /*
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.
244 */
245 if (!nfsd_up)
246 return;
247 nfs4_state_shutdown();
248 lockd_down();
249 nfsd_racache_shutdown();
250 nfsd_up = false;
251}
252
253static void nfsd_last_thread(struct svc_serv *serv)
254{
255 /* When last nfsd thread exits we need to do some clean-up */
256 nfsd_serv = NULL;
257 nfsd_shutdown();
258
259 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
260 "cache\n");
261 nfsd_export_flush();
262}
263
264void nfsd_reset_versions(void)
265{
266 int found_one = 0;
267 int i;
268
269 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
270 if (nfsd_program.pg_vers[i])
271 found_one = 1;
272 }
273
274 if (!found_one) {
275 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
276 nfsd_program.pg_vers[i] = nfsd_version[i];
277#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
278 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
279 nfsd_acl_program.pg_vers[i] =
280 nfsd_acl_version[i];
281#endif
282 }
283}
284
285/*
286 * Each session guarantees a negotiated per slot memory cache for replies
287 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
288 * NFSv4.1 server might want to use more memory for a DRC than a machine
289 * with mutiple services.
290 *
291 * Impose a hard limit on the number of pages for the DRC which varies
292 * according to the machines free pages. This is of course only a default.
293 *
294 * For now this is a #defined shift which could be under admin control
295 * in the future.
296 */
297static void set_max_drc(void)
298{
299 #define NFSD_DRC_SIZE_SHIFT 10
300 nfsd_drc_max_mem = (nr_free_buffer_pages()
301 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
302 nfsd_drc_mem_used = 0;
303 spin_lock_init(&nfsd_drc_lock);
304 dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
305}
306
307int nfsd_create_serv(void)
308{
309 int err = 0;
310
311 WARN_ON(!mutex_is_locked(&nfsd_mutex));
312 if (nfsd_serv) {
313 svc_get(nfsd_serv);
314 return 0;
315 }
316 if (nfsd_max_blksize == 0) {
317 /* choose a suitable default */
318 struct sysinfo i;
319 si_meminfo(&i);
320 /* Aim for 1/4096 of memory per thread
321 * This gives 1MB on 4Gig machines
322 * But only uses 32K on 128M machines.
323 * Bottom out at 8K on 32M and smaller.
324 * Of course, this is only a default.
325 */
326 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
327 i.totalram <<= PAGE_SHIFT - 12;
328 while (nfsd_max_blksize > i.totalram &&
329 nfsd_max_blksize >= 8*1024*2)
330 nfsd_max_blksize /= 2;
331 }
332 nfsd_reset_versions();
333
334 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
335 nfsd_last_thread, nfsd, THIS_MODULE);
336 if (nfsd_serv == NULL)
337 return -ENOMEM;
338
339 set_max_drc();
340 do_gettimeofday(&nfssvc_boot); /* record boot time */
341 return err;
342}
343
344int nfsd_nrpools(void)
345{
346 if (nfsd_serv == NULL)
347 return 0;
348 else
349 return nfsd_serv->sv_nrpools;
350}
351
352int nfsd_get_nrthreads(int n, int *nthreads)
353{
354 int i = 0;
355
356 if (nfsd_serv != NULL) {
357 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
358 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
359 }
360
361 return 0;
362}
363
364int nfsd_set_nrthreads(int n, int *nthreads)
365{
366 int i = 0;
367 int tot = 0;
368 int err = 0;
369
370 WARN_ON(!mutex_is_locked(&nfsd_mutex));
371
372 if (nfsd_serv == NULL || n <= 0)
373 return 0;
374
375 if (n > nfsd_serv->sv_nrpools)
376 n = nfsd_serv->sv_nrpools;
377
378 /* enforce a global maximum number of threads */
379 tot = 0;
380 for (i = 0; i < n; i++) {
381 if (nthreads[i] > NFSD_MAXSERVS)
382 nthreads[i] = NFSD_MAXSERVS;
383 tot += nthreads[i];
384 }
385 if (tot > NFSD_MAXSERVS) {
386 /* total too large: scale down requested numbers */
387 for (i = 0; i < n && tot > 0; i++) {
388 int new = nthreads[i] * NFSD_MAXSERVS / tot;
389 tot -= (nthreads[i] - new);
390 nthreads[i] = new;
391 }
392 for (i = 0; i < n && tot > 0; i++) {
393 nthreads[i]--;
394 tot--;
395 }
396 }
397
398 /*
399 * There must always be a thread in pool 0; the admin
400 * can't shut down NFS completely using pool_threads.
401 */
402 if (nthreads[0] == 0)
403 nthreads[0] = 1;
404
405 /* apply the new numbers */
406 svc_get(nfsd_serv);
407 for (i = 0; i < n; i++) {
408 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
409 nthreads[i]);
410 if (err)
411 break;
412 }
413 svc_destroy(nfsd_serv);
414
415 return err;
416}
417
418/*
419 * Adjust the number of threads and return the new number of threads.
420 * This is also the function that starts the server if necessary, if
421 * this is the first time nrservs is nonzero.
422 */
423int
424nfsd_svc(unsigned short port, int nrservs)
425{
426 int error;
427 bool nfsd_up_before;
428
429 mutex_lock(&nfsd_mutex);
430 dprintk("nfsd: creating service\n");
431 if (nrservs <= 0)
432 nrservs = 0;
433 if (nrservs > NFSD_MAXSERVS)
434 nrservs = NFSD_MAXSERVS;
435 error = 0;
436 if (nrservs == 0 && nfsd_serv == NULL)
437 goto out;
438
439 error = nfsd_create_serv();
440 if (error)
441 goto out;
442
443 nfsd_up_before = nfsd_up;
444
445 error = nfsd_startup(port, nrservs);
446 if (error)
447 goto out_destroy;
448 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
449 if (error)
450 goto out_shutdown;
451 /* We are holding a reference to nfsd_serv which
452 * we don't want to count in the return value,
453 * so subtract 1
454 */
455 error = nfsd_serv->sv_nrthreads - 1;
456out_shutdown:
457 if (error < 0 && !nfsd_up_before)
458 nfsd_shutdown();
459out_destroy:
460 svc_destroy(nfsd_serv); /* Release server */
461out:
462 mutex_unlock(&nfsd_mutex);
463 return error;
464}
465
466
467/*
468 * This is the NFS server kernel thread
469 */
470static int
471nfsd(void *vrqstp)
472{
473 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
474 int err, preverr = 0;
475
476 /* Lock module and set up kernel thread */
477 mutex_lock(&nfsd_mutex);
478
479 /* At this point, the thread shares current->fs
480 * with the init process. We need to create files with a
481 * umask of 0 instead of init's umask. */
482 if (unshare_fs_struct() < 0) {
483 printk("Unable to start nfsd thread: out of memory\n");
484 goto out;
485 }
486
487 current->fs->umask = 0;
488
489 /*
490 * thread is spawned with all signals set to SIG_IGN, re-enable
491 * the ones that will bring down the thread
492 */
493 allow_signal(SIGKILL);
494 allow_signal(SIGHUP);
495 allow_signal(SIGINT);
496 allow_signal(SIGQUIT);
497
498 nfsdstats.th_cnt++;
499 mutex_unlock(&nfsd_mutex);
500
501 /*
502 * We want less throttling in balance_dirty_pages() so that nfs to
503 * localhost doesn't cause nfsd to lock up due to all the client's
504 * dirty pages.
505 */
506 current->flags |= PF_LESS_THROTTLE;
507 set_freezable();
508
509 /*
510 * The main request loop
511 */
512 for (;;) {
513 /*
514 * Find a socket with data available and call its
515 * recvfrom routine.
516 */
517 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
518 ;
519 if (err == -EINTR)
520 break;
521 else if (err < 0) {
522 if (err != preverr) {
523 printk(KERN_WARNING "%s: unexpected error "
524 "from svc_recv (%d)\n", __func__, -err);
525 preverr = err;
526 }
527 schedule_timeout_uninterruptible(HZ);
528 continue;
529 }
530
531 validate_process_creds();
532 svc_process(rqstp);
533 validate_process_creds();
534 }
535
536 /* Clear signals before calling svc_exit_thread() */
537 flush_signals(current);
538
539 mutex_lock(&nfsd_mutex);
540 nfsdstats.th_cnt --;
541
542out:
543 /* Release the thread */
544 svc_exit_thread(rqstp);
545
546 /* Release module */
547 mutex_unlock(&nfsd_mutex);
548 module_put_and_exit(0);
549 return 0;
550}
551
552static __be32 map_new_errors(u32 vers, __be32 nfserr)
553{
554 if (nfserr == nfserr_jukebox && vers == 2)
555 return nfserr_dropit;
556 if (nfserr == nfserr_wrongsec && vers < 4)
557 return nfserr_acces;
558 return nfserr;
559}
560
561int
562nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
563{
564 struct svc_procedure *proc;
565 kxdrproc_t xdr;
566 __be32 nfserr;
567 __be32 *nfserrp;
568
569 dprintk("nfsd_dispatch: vers %d proc %d\n",
570 rqstp->rq_vers, rqstp->rq_proc);
571 proc = rqstp->rq_procinfo;
572
573 /*
574 * Give the xdr decoder a chance to change this if it wants
575 * (necessary in the NFSv4.0 compound case)
576 */
577 rqstp->rq_cachetype = proc->pc_cachetype;
578 /* Decode arguments */
579 xdr = proc->pc_decode;
580 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
581 rqstp->rq_argp)) {
582 dprintk("nfsd: failed to decode arguments!\n");
583 *statp = rpc_garbage_args;
584 return 1;
585 }
586
587 /* Check whether we have this call in the cache. */
588 switch (nfsd_cache_lookup(rqstp)) {
589 case RC_INTR:
590 case RC_DROPIT:
591 return 0;
592 case RC_REPLY:
593 return 1;
594 case RC_DOIT:;
595 /* do it */
596 }
597
598 /* need to grab the location to store the status, as
599 * nfsv4 does some encoding while processing
600 */
601 nfserrp = rqstp->rq_res.head[0].iov_base
602 + rqstp->rq_res.head[0].iov_len;
603 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
604
605 /* Now call the procedure handler, and encode NFS status. */
606 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
607 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
608 if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
609 dprintk("nfsd: Dropping request; may be revisited later\n");
610 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
611 return 0;
612 }
613
614 if (rqstp->rq_proc != 0)
615 *nfserrp++ = nfserr;
616
617 /* Encode result.
618 * For NFSv2, additional info is never returned in case of an error.
619 */
620 if (!(nfserr && rqstp->rq_vers == 2)) {
621 xdr = proc->pc_encode;
622 if (xdr && !xdr(rqstp, nfserrp,
623 rqstp->rq_resp)) {
624 /* Failed to encode result. Release cache entry */
625 dprintk("nfsd: failed to encode result!\n");
626 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
627 *statp = rpc_system_err;
628 return 1;
629 }
630 }
631
632 /* Store reply in cache. */
633 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
634 return 1;
635}
636
637int nfsd_pool_stats_open(struct inode *inode, struct file *file)
638{
639 int ret;
640 mutex_lock(&nfsd_mutex);
641 if (nfsd_serv == NULL) {
642 mutex_unlock(&nfsd_mutex);
643 return -ENODEV;
644 }
645 /* bump up the psudo refcount while traversing */
646 svc_get(nfsd_serv);
647 ret = svc_pool_stats_open(nfsd_serv, file);
648 mutex_unlock(&nfsd_mutex);
649 return ret;
650}
651
652int nfsd_pool_stats_release(struct inode *inode, struct file *file)
653{
654 int ret = seq_release(inode, file);
655 mutex_lock(&nfsd_mutex);
656 /* this function really, really should have been called svc_put() */
657 svc_destroy(nfsd_serv);
658 mutex_unlock(&nfsd_mutex);
659 return ret;
660}