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// SPDX-License-Identifier: GPL-2.0
2/*
3 * Central processing for nfsd.
4 *
5 * Authors: Olaf Kirch (okir@monad.swb.de)
6 *
7 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
8 */
9
10#include <linux/sched/signal.h>
11#include <linux/freezer.h>
12#include <linux/module.h>
13#include <linux/fs_struct.h>
14#include <linux/swap.h>
15
16#include <linux/sunrpc/stats.h>
17#include <linux/sunrpc/svcsock.h>
18#include <linux/sunrpc/svc_xprt.h>
19#include <linux/lockd/bind.h>
20#include <linux/nfsacl.h>
21#include <linux/seq_file.h>
22#include <linux/inetdevice.h>
23#include <net/addrconf.h>
24#include <net/ipv6.h>
25#include <net/net_namespace.h>
26#include "nfsd.h"
27#include "cache.h"
28#include "vfs.h"
29#include "netns.h"
30#include "filecache.h"
31
32#define NFSDDBG_FACILITY NFSDDBG_SVC
33
34extern struct svc_program nfsd_program;
35static int nfsd(void *vrqstp);
36#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
37static int nfsd_acl_rpcbind_set(struct net *,
38 const struct svc_program *,
39 u32, int,
40 unsigned short,
41 unsigned short);
42static __be32 nfsd_acl_init_request(struct svc_rqst *,
43 const struct svc_program *,
44 struct svc_process_info *);
45#endif
46static int nfsd_rpcbind_set(struct net *,
47 const struct svc_program *,
48 u32, int,
49 unsigned short,
50 unsigned short);
51static __be32 nfsd_init_request(struct svc_rqst *,
52 const struct svc_program *,
53 struct svc_process_info *);
54
55/*
56 * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members
57 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
58 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
59 *
60 * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a
61 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
62 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
63 * entry of ->sv_pools[].
64 *
65 * Transitions of the thread count between zero and non-zero are of particular
66 * interest since the svc_serv needs to be created and initialized at that
67 * point, or freed.
68 *
69 * Finally, the nfsd_mutex also protects some of the global variables that are
70 * accessed when nfsd starts and that are settable via the write_* routines in
71 * nfsctl.c. In particular:
72 *
73 * user_recovery_dirname
74 * user_lease_time
75 * nfsd_versions
76 */
77DEFINE_MUTEX(nfsd_mutex);
78
79/*
80 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
81 * nfsd_drc_max_pages limits the total amount of memory available for
82 * version 4.1 DRC caches.
83 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
84 */
85spinlock_t nfsd_drc_lock;
86unsigned long nfsd_drc_max_mem;
87unsigned long nfsd_drc_mem_used;
88
89#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
90static struct svc_stat nfsd_acl_svcstats;
91static const struct svc_version *nfsd_acl_version[] = {
92 [2] = &nfsd_acl_version2,
93 [3] = &nfsd_acl_version3,
94};
95
96#define NFSD_ACL_MINVERS 2
97#define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
98static const struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
99
100static struct svc_program nfsd_acl_program = {
101 .pg_prog = NFS_ACL_PROGRAM,
102 .pg_nvers = NFSD_ACL_NRVERS,
103 .pg_vers = nfsd_acl_versions,
104 .pg_name = "nfsacl",
105 .pg_class = "nfsd",
106 .pg_stats = &nfsd_acl_svcstats,
107 .pg_authenticate = &svc_set_client,
108 .pg_init_request = nfsd_acl_init_request,
109 .pg_rpcbind_set = nfsd_acl_rpcbind_set,
110};
111
112static struct svc_stat nfsd_acl_svcstats = {
113 .program = &nfsd_acl_program,
114};
115#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
116
117static const struct svc_version *nfsd_version[] = {
118 [2] = &nfsd_version2,
119#if defined(CONFIG_NFSD_V3)
120 [3] = &nfsd_version3,
121#endif
122#if defined(CONFIG_NFSD_V4)
123 [4] = &nfsd_version4,
124#endif
125};
126
127#define NFSD_MINVERS 2
128#define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
129
130struct svc_program nfsd_program = {
131#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
132 .pg_next = &nfsd_acl_program,
133#endif
134 .pg_prog = NFS_PROGRAM, /* program number */
135 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
136 .pg_vers = nfsd_version, /* version table */
137 .pg_name = "nfsd", /* program name */
138 .pg_class = "nfsd", /* authentication class */
139 .pg_stats = &nfsd_svcstats, /* version table */
140 .pg_authenticate = &svc_set_client, /* export authentication */
141 .pg_init_request = nfsd_init_request,
142 .pg_rpcbind_set = nfsd_rpcbind_set,
143};
144
145static bool
146nfsd_support_version(int vers)
147{
148 if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
149 return nfsd_version[vers] != NULL;
150 return false;
151}
152
153static bool *
154nfsd_alloc_versions(void)
155{
156 bool *vers = kmalloc_array(NFSD_NRVERS, sizeof(bool), GFP_KERNEL);
157 unsigned i;
158
159 if (vers) {
160 /* All compiled versions are enabled by default */
161 for (i = 0; i < NFSD_NRVERS; i++)
162 vers[i] = nfsd_support_version(i);
163 }
164 return vers;
165}
166
167static bool *
168nfsd_alloc_minorversions(void)
169{
170 bool *vers = kmalloc_array(NFSD_SUPPORTED_MINOR_VERSION + 1,
171 sizeof(bool), GFP_KERNEL);
172 unsigned i;
173
174 if (vers) {
175 /* All minor versions are enabled by default */
176 for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
177 vers[i] = nfsd_support_version(4);
178 }
179 return vers;
180}
181
182void
183nfsd_netns_free_versions(struct nfsd_net *nn)
184{
185 kfree(nn->nfsd_versions);
186 kfree(nn->nfsd4_minorversions);
187 nn->nfsd_versions = NULL;
188 nn->nfsd4_minorversions = NULL;
189}
190
191static void
192nfsd_netns_init_versions(struct nfsd_net *nn)
193{
194 if (!nn->nfsd_versions) {
195 nn->nfsd_versions = nfsd_alloc_versions();
196 nn->nfsd4_minorversions = nfsd_alloc_minorversions();
197 if (!nn->nfsd_versions || !nn->nfsd4_minorversions)
198 nfsd_netns_free_versions(nn);
199 }
200}
201
202int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change)
203{
204 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
205 return 0;
206 switch(change) {
207 case NFSD_SET:
208 if (nn->nfsd_versions)
209 nn->nfsd_versions[vers] = nfsd_support_version(vers);
210 break;
211 case NFSD_CLEAR:
212 nfsd_netns_init_versions(nn);
213 if (nn->nfsd_versions)
214 nn->nfsd_versions[vers] = false;
215 break;
216 case NFSD_TEST:
217 if (nn->nfsd_versions)
218 return nn->nfsd_versions[vers];
219 /* Fallthrough */
220 case NFSD_AVAIL:
221 return nfsd_support_version(vers);
222 }
223 return 0;
224}
225
226static void
227nfsd_adjust_nfsd_versions4(struct nfsd_net *nn)
228{
229 unsigned i;
230
231 for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
232 if (nn->nfsd4_minorversions[i])
233 return;
234 }
235 nfsd_vers(nn, 4, NFSD_CLEAR);
236}
237
238int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change)
239{
240 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
241 change != NFSD_AVAIL)
242 return -1;
243
244 switch(change) {
245 case NFSD_SET:
246 if (nn->nfsd4_minorversions) {
247 nfsd_vers(nn, 4, NFSD_SET);
248 nn->nfsd4_minorversions[minorversion] =
249 nfsd_vers(nn, 4, NFSD_TEST);
250 }
251 break;
252 case NFSD_CLEAR:
253 nfsd_netns_init_versions(nn);
254 if (nn->nfsd4_minorversions) {
255 nn->nfsd4_minorversions[minorversion] = false;
256 nfsd_adjust_nfsd_versions4(nn);
257 }
258 break;
259 case NFSD_TEST:
260 if (nn->nfsd4_minorversions)
261 return nn->nfsd4_minorversions[minorversion];
262 return nfsd_vers(nn, 4, NFSD_TEST);
263 case NFSD_AVAIL:
264 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION &&
265 nfsd_vers(nn, 4, NFSD_AVAIL);
266 }
267 return 0;
268}
269
270/*
271 * Maximum number of nfsd processes
272 */
273#define NFSD_MAXSERVS 8192
274
275int nfsd_nrthreads(struct net *net)
276{
277 int rv = 0;
278 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
279
280 mutex_lock(&nfsd_mutex);
281 if (nn->nfsd_serv)
282 rv = nn->nfsd_serv->sv_nrthreads;
283 mutex_unlock(&nfsd_mutex);
284 return rv;
285}
286
287static int nfsd_init_socks(struct net *net, const struct cred *cred)
288{
289 int error;
290 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
291
292 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
293 return 0;
294
295 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
296 SVC_SOCK_DEFAULTS, cred);
297 if (error < 0)
298 return error;
299
300 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
301 SVC_SOCK_DEFAULTS, cred);
302 if (error < 0)
303 return error;
304
305 return 0;
306}
307
308static int nfsd_users = 0;
309
310static int nfsd_startup_generic(int nrservs)
311{
312 int ret;
313
314 if (nfsd_users++)
315 return 0;
316
317 ret = nfsd_file_cache_init();
318 if (ret)
319 goto dec_users;
320
321 ret = nfs4_state_start();
322 if (ret)
323 goto out_file_cache;
324 return 0;
325
326out_file_cache:
327 nfsd_file_cache_shutdown();
328dec_users:
329 nfsd_users--;
330 return ret;
331}
332
333static void nfsd_shutdown_generic(void)
334{
335 if (--nfsd_users)
336 return;
337
338 nfs4_state_shutdown();
339 nfsd_file_cache_shutdown();
340}
341
342static bool nfsd_needs_lockd(struct nfsd_net *nn)
343{
344 return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST);
345}
346
347void nfsd_copy_boot_verifier(__be32 verf[2], struct nfsd_net *nn)
348{
349 int seq = 0;
350
351 do {
352 read_seqbegin_or_lock(&nn->boot_lock, &seq);
353 /*
354 * This is opaque to client, so no need to byte-swap. Use
355 * __force to keep sparse happy. y2038 time_t overflow is
356 * irrelevant in this usage
357 */
358 verf[0] = (__force __be32)nn->nfssvc_boot.tv_sec;
359 verf[1] = (__force __be32)nn->nfssvc_boot.tv_nsec;
360 } while (need_seqretry(&nn->boot_lock, seq));
361 done_seqretry(&nn->boot_lock, seq);
362}
363
364static void nfsd_reset_boot_verifier_locked(struct nfsd_net *nn)
365{
366 ktime_get_real_ts64(&nn->nfssvc_boot);
367}
368
369void nfsd_reset_boot_verifier(struct nfsd_net *nn)
370{
371 write_seqlock(&nn->boot_lock);
372 nfsd_reset_boot_verifier_locked(nn);
373 write_sequnlock(&nn->boot_lock);
374}
375
376static int nfsd_startup_net(int nrservs, struct net *net, const struct cred *cred)
377{
378 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
379 int ret;
380
381 if (nn->nfsd_net_up)
382 return 0;
383
384 ret = nfsd_startup_generic(nrservs);
385 if (ret)
386 return ret;
387 ret = nfsd_init_socks(net, cred);
388 if (ret)
389 goto out_socks;
390
391 if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
392 ret = lockd_up(net, cred);
393 if (ret)
394 goto out_socks;
395 nn->lockd_up = 1;
396 }
397
398 ret = nfs4_state_start_net(net);
399 if (ret)
400 goto out_lockd;
401
402 nn->nfsd_net_up = true;
403 return 0;
404
405out_lockd:
406 if (nn->lockd_up) {
407 lockd_down(net);
408 nn->lockd_up = 0;
409 }
410out_socks:
411 nfsd_shutdown_generic();
412 return ret;
413}
414
415static void nfsd_shutdown_net(struct net *net)
416{
417 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
418
419 nfsd_file_cache_purge(net);
420 nfs4_state_shutdown_net(net);
421 if (nn->lockd_up) {
422 lockd_down(net);
423 nn->lockd_up = 0;
424 }
425 nn->nfsd_net_up = false;
426 nfsd_shutdown_generic();
427}
428
429static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
430 void *ptr)
431{
432 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
433 struct net_device *dev = ifa->ifa_dev->dev;
434 struct net *net = dev_net(dev);
435 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
436 struct sockaddr_in sin;
437
438 if ((event != NETDEV_DOWN) ||
439 !atomic_inc_not_zero(&nn->ntf_refcnt))
440 goto out;
441
442 if (nn->nfsd_serv) {
443 dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
444 sin.sin_family = AF_INET;
445 sin.sin_addr.s_addr = ifa->ifa_local;
446 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
447 }
448 atomic_dec(&nn->ntf_refcnt);
449 wake_up(&nn->ntf_wq);
450
451out:
452 return NOTIFY_DONE;
453}
454
455static struct notifier_block nfsd_inetaddr_notifier = {
456 .notifier_call = nfsd_inetaddr_event,
457};
458
459#if IS_ENABLED(CONFIG_IPV6)
460static int nfsd_inet6addr_event(struct notifier_block *this,
461 unsigned long event, void *ptr)
462{
463 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
464 struct net_device *dev = ifa->idev->dev;
465 struct net *net = dev_net(dev);
466 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
467 struct sockaddr_in6 sin6;
468
469 if ((event != NETDEV_DOWN) ||
470 !atomic_inc_not_zero(&nn->ntf_refcnt))
471 goto out;
472
473 if (nn->nfsd_serv) {
474 dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
475 sin6.sin6_family = AF_INET6;
476 sin6.sin6_addr = ifa->addr;
477 if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
478 sin6.sin6_scope_id = ifa->idev->dev->ifindex;
479 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
480 }
481 atomic_dec(&nn->ntf_refcnt);
482 wake_up(&nn->ntf_wq);
483out:
484 return NOTIFY_DONE;
485}
486
487static struct notifier_block nfsd_inet6addr_notifier = {
488 .notifier_call = nfsd_inet6addr_event,
489};
490#endif
491
492/* Only used under nfsd_mutex, so this atomic may be overkill: */
493static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
494
495static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
496{
497 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
498
499 atomic_dec(&nn->ntf_refcnt);
500 /* check if the notifier still has clients */
501 if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
502 unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
503#if IS_ENABLED(CONFIG_IPV6)
504 unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
505#endif
506 }
507 wait_event(nn->ntf_wq, atomic_read(&nn->ntf_refcnt) == 0);
508
509 /*
510 * write_ports can create the server without actually starting
511 * any threads--if we get shut down before any threads are
512 * started, then nfsd_last_thread will be run before any of this
513 * other initialization has been done except the rpcb information.
514 */
515 svc_rpcb_cleanup(serv, net);
516 if (!nn->nfsd_net_up)
517 return;
518
519 nfsd_shutdown_net(net);
520 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
521 "cache\n");
522 nfsd_export_flush(net);
523}
524
525void nfsd_reset_versions(struct nfsd_net *nn)
526{
527 int i;
528
529 for (i = 0; i < NFSD_NRVERS; i++)
530 if (nfsd_vers(nn, i, NFSD_TEST))
531 return;
532
533 for (i = 0; i < NFSD_NRVERS; i++)
534 if (i != 4)
535 nfsd_vers(nn, i, NFSD_SET);
536 else {
537 int minor = 0;
538 while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
539 minor++;
540 }
541}
542
543/*
544 * Each session guarantees a negotiated per slot memory cache for replies
545 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
546 * NFSv4.1 server might want to use more memory for a DRC than a machine
547 * with mutiple services.
548 *
549 * Impose a hard limit on the number of pages for the DRC which varies
550 * according to the machines free pages. This is of course only a default.
551 *
552 * For now this is a #defined shift which could be under admin control
553 * in the future.
554 */
555static void set_max_drc(void)
556{
557 #define NFSD_DRC_SIZE_SHIFT 7
558 nfsd_drc_max_mem = (nr_free_buffer_pages()
559 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
560 nfsd_drc_mem_used = 0;
561 spin_lock_init(&nfsd_drc_lock);
562 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
563}
564
565static int nfsd_get_default_max_blksize(void)
566{
567 struct sysinfo i;
568 unsigned long long target;
569 unsigned long ret;
570
571 si_meminfo(&i);
572 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
573 /*
574 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
575 * machines, but only uses 32K on 128M machines. Bottom out at
576 * 8K on 32M and smaller. Of course, this is only a default.
577 */
578 target >>= 12;
579
580 ret = NFSSVC_MAXBLKSIZE;
581 while (ret > target && ret >= 8*1024*2)
582 ret /= 2;
583 return ret;
584}
585
586static const struct svc_serv_ops nfsd_thread_sv_ops = {
587 .svo_shutdown = nfsd_last_thread,
588 .svo_function = nfsd,
589 .svo_enqueue_xprt = svc_xprt_do_enqueue,
590 .svo_setup = svc_set_num_threads,
591 .svo_module = THIS_MODULE,
592};
593
594int nfsd_create_serv(struct net *net)
595{
596 int error;
597 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
598
599 WARN_ON(!mutex_is_locked(&nfsd_mutex));
600 if (nn->nfsd_serv) {
601 svc_get(nn->nfsd_serv);
602 return 0;
603 }
604 if (nfsd_max_blksize == 0)
605 nfsd_max_blksize = nfsd_get_default_max_blksize();
606 nfsd_reset_versions(nn);
607 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
608 &nfsd_thread_sv_ops);
609 if (nn->nfsd_serv == NULL)
610 return -ENOMEM;
611
612 nn->nfsd_serv->sv_maxconn = nn->max_connections;
613 error = svc_bind(nn->nfsd_serv, net);
614 if (error < 0) {
615 svc_destroy(nn->nfsd_serv);
616 return error;
617 }
618
619 set_max_drc();
620 /* check if the notifier is already set */
621 if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
622 register_inetaddr_notifier(&nfsd_inetaddr_notifier);
623#if IS_ENABLED(CONFIG_IPV6)
624 register_inet6addr_notifier(&nfsd_inet6addr_notifier);
625#endif
626 }
627 atomic_inc(&nn->ntf_refcnt);
628 nfsd_reset_boot_verifier(nn);
629 return 0;
630}
631
632int nfsd_nrpools(struct net *net)
633{
634 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
635
636 if (nn->nfsd_serv == NULL)
637 return 0;
638 else
639 return nn->nfsd_serv->sv_nrpools;
640}
641
642int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
643{
644 int i = 0;
645 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
646
647 if (nn->nfsd_serv != NULL) {
648 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
649 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
650 }
651
652 return 0;
653}
654
655void nfsd_destroy(struct net *net)
656{
657 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
658 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
659
660 if (destroy)
661 svc_shutdown_net(nn->nfsd_serv, net);
662 svc_destroy(nn->nfsd_serv);
663 if (destroy)
664 nn->nfsd_serv = NULL;
665}
666
667int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
668{
669 int i = 0;
670 int tot = 0;
671 int err = 0;
672 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
673
674 WARN_ON(!mutex_is_locked(&nfsd_mutex));
675
676 if (nn->nfsd_serv == NULL || n <= 0)
677 return 0;
678
679 if (n > nn->nfsd_serv->sv_nrpools)
680 n = nn->nfsd_serv->sv_nrpools;
681
682 /* enforce a global maximum number of threads */
683 tot = 0;
684 for (i = 0; i < n; i++) {
685 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
686 tot += nthreads[i];
687 }
688 if (tot > NFSD_MAXSERVS) {
689 /* total too large: scale down requested numbers */
690 for (i = 0; i < n && tot > 0; i++) {
691 int new = nthreads[i] * NFSD_MAXSERVS / tot;
692 tot -= (nthreads[i] - new);
693 nthreads[i] = new;
694 }
695 for (i = 0; i < n && tot > 0; i++) {
696 nthreads[i]--;
697 tot--;
698 }
699 }
700
701 /*
702 * There must always be a thread in pool 0; the admin
703 * can't shut down NFS completely using pool_threads.
704 */
705 if (nthreads[0] == 0)
706 nthreads[0] = 1;
707
708 /* apply the new numbers */
709 svc_get(nn->nfsd_serv);
710 for (i = 0; i < n; i++) {
711 err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
712 &nn->nfsd_serv->sv_pools[i], nthreads[i]);
713 if (err)
714 break;
715 }
716 nfsd_destroy(net);
717 return err;
718}
719
720/*
721 * Adjust the number of threads and return the new number of threads.
722 * This is also the function that starts the server if necessary, if
723 * this is the first time nrservs is nonzero.
724 */
725int
726nfsd_svc(int nrservs, struct net *net, const struct cred *cred)
727{
728 int error;
729 bool nfsd_up_before;
730 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
731
732 mutex_lock(&nfsd_mutex);
733 dprintk("nfsd: creating service\n");
734
735 nrservs = max(nrservs, 0);
736 nrservs = min(nrservs, NFSD_MAXSERVS);
737 error = 0;
738
739 if (nrservs == 0 && nn->nfsd_serv == NULL)
740 goto out;
741
742 error = nfsd_create_serv(net);
743 if (error)
744 goto out;
745
746 nfsd_up_before = nn->nfsd_net_up;
747
748 error = nfsd_startup_net(nrservs, net, cred);
749 if (error)
750 goto out_destroy;
751 error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
752 NULL, nrservs);
753 if (error)
754 goto out_shutdown;
755 /* We are holding a reference to nn->nfsd_serv which
756 * we don't want to count in the return value,
757 * so subtract 1
758 */
759 error = nn->nfsd_serv->sv_nrthreads - 1;
760out_shutdown:
761 if (error < 0 && !nfsd_up_before)
762 nfsd_shutdown_net(net);
763out_destroy:
764 nfsd_destroy(net); /* Release server */
765out:
766 mutex_unlock(&nfsd_mutex);
767 return error;
768}
769
770#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
771static bool
772nfsd_support_acl_version(int vers)
773{
774 if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
775 return nfsd_acl_version[vers] != NULL;
776 return false;
777}
778
779static int
780nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
781 u32 version, int family, unsigned short proto,
782 unsigned short port)
783{
784 if (!nfsd_support_acl_version(version) ||
785 !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
786 return 0;
787 return svc_generic_rpcbind_set(net, progp, version, family,
788 proto, port);
789}
790
791static __be32
792nfsd_acl_init_request(struct svc_rqst *rqstp,
793 const struct svc_program *progp,
794 struct svc_process_info *ret)
795{
796 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
797 int i;
798
799 if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
800 nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
801 return svc_generic_init_request(rqstp, progp, ret);
802
803 ret->mismatch.lovers = NFSD_ACL_NRVERS;
804 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
805 if (nfsd_support_acl_version(rqstp->rq_vers) &&
806 nfsd_vers(nn, i, NFSD_TEST)) {
807 ret->mismatch.lovers = i;
808 break;
809 }
810 }
811 if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
812 return rpc_prog_unavail;
813 ret->mismatch.hivers = NFSD_ACL_MINVERS;
814 for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
815 if (nfsd_support_acl_version(rqstp->rq_vers) &&
816 nfsd_vers(nn, i, NFSD_TEST)) {
817 ret->mismatch.hivers = i;
818 break;
819 }
820 }
821 return rpc_prog_mismatch;
822}
823#endif
824
825static int
826nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
827 u32 version, int family, unsigned short proto,
828 unsigned short port)
829{
830 if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
831 return 0;
832 return svc_generic_rpcbind_set(net, progp, version, family,
833 proto, port);
834}
835
836static __be32
837nfsd_init_request(struct svc_rqst *rqstp,
838 const struct svc_program *progp,
839 struct svc_process_info *ret)
840{
841 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
842 int i;
843
844 if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
845 return svc_generic_init_request(rqstp, progp, ret);
846
847 ret->mismatch.lovers = NFSD_NRVERS;
848 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
849 if (nfsd_vers(nn, i, NFSD_TEST)) {
850 ret->mismatch.lovers = i;
851 break;
852 }
853 }
854 if (ret->mismatch.lovers == NFSD_NRVERS)
855 return rpc_prog_unavail;
856 ret->mismatch.hivers = NFSD_MINVERS;
857 for (i = NFSD_NRVERS - 1; i >= NFSD_MINVERS; i--) {
858 if (nfsd_vers(nn, i, NFSD_TEST)) {
859 ret->mismatch.hivers = i;
860 break;
861 }
862 }
863 return rpc_prog_mismatch;
864}
865
866/*
867 * This is the NFS server kernel thread
868 */
869static int
870nfsd(void *vrqstp)
871{
872 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
873 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
874 struct net *net = perm_sock->xpt_net;
875 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
876 int err;
877
878 /* Lock module and set up kernel thread */
879 mutex_lock(&nfsd_mutex);
880
881 /* At this point, the thread shares current->fs
882 * with the init process. We need to create files with the
883 * umask as defined by the client instead of init's umask. */
884 if (unshare_fs_struct() < 0) {
885 printk("Unable to start nfsd thread: out of memory\n");
886 goto out;
887 }
888
889 current->fs->umask = 0;
890
891 /*
892 * thread is spawned with all signals set to SIG_IGN, re-enable
893 * the ones that will bring down the thread
894 */
895 allow_signal(SIGKILL);
896 allow_signal(SIGHUP);
897 allow_signal(SIGINT);
898 allow_signal(SIGQUIT);
899
900 nfsdstats.th_cnt++;
901 mutex_unlock(&nfsd_mutex);
902
903 set_freezable();
904
905 /*
906 * The main request loop
907 */
908 for (;;) {
909 /* Update sv_maxconn if it has changed */
910 rqstp->rq_server->sv_maxconn = nn->max_connections;
911
912 /*
913 * Find a socket with data available and call its
914 * recvfrom routine.
915 */
916 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
917 ;
918 if (err == -EINTR)
919 break;
920 validate_process_creds();
921 svc_process(rqstp);
922 validate_process_creds();
923 }
924
925 /* Clear signals before calling svc_exit_thread() */
926 flush_signals(current);
927
928 mutex_lock(&nfsd_mutex);
929 nfsdstats.th_cnt --;
930
931out:
932 rqstp->rq_server = NULL;
933
934 /* Release the thread */
935 svc_exit_thread(rqstp);
936
937 nfsd_destroy(net);
938
939 /* Release module */
940 mutex_unlock(&nfsd_mutex);
941 module_put_and_exit(0);
942 return 0;
943}
944
945static __be32 map_new_errors(u32 vers, __be32 nfserr)
946{
947 if (nfserr == nfserr_jukebox && vers == 2)
948 return nfserr_dropit;
949 if (nfserr == nfserr_wrongsec && vers < 4)
950 return nfserr_acces;
951 return nfserr;
952}
953
954/*
955 * A write procedure can have a large argument, and a read procedure can
956 * have a large reply, but no NFSv2 or NFSv3 procedure has argument and
957 * reply that can both be larger than a page. The xdr code has taken
958 * advantage of this assumption to be a sloppy about bounds checking in
959 * some cases. Pending a rewrite of the NFSv2/v3 xdr code to fix that
960 * problem, we enforce these assumptions here:
961 */
962static bool nfs_request_too_big(struct svc_rqst *rqstp,
963 const struct svc_procedure *proc)
964{
965 /*
966 * The ACL code has more careful bounds-checking and is not
967 * susceptible to this problem:
968 */
969 if (rqstp->rq_prog != NFS_PROGRAM)
970 return false;
971 /*
972 * Ditto NFSv4 (which can in theory have argument and reply both
973 * more than a page):
974 */
975 if (rqstp->rq_vers >= 4)
976 return false;
977 /* The reply will be small, we're OK: */
978 if (proc->pc_xdrressize > 0 &&
979 proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE))
980 return false;
981
982 return rqstp->rq_arg.len > PAGE_SIZE;
983}
984
985int
986nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
987{
988 const struct svc_procedure *proc;
989 __be32 nfserr;
990 __be32 *nfserrp;
991
992 dprintk("nfsd_dispatch: vers %d proc %d\n",
993 rqstp->rq_vers, rqstp->rq_proc);
994 proc = rqstp->rq_procinfo;
995
996 if (nfs_request_too_big(rqstp, proc)) {
997 dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers);
998 *statp = rpc_garbage_args;
999 return 1;
1000 }
1001 /*
1002 * Give the xdr decoder a chance to change this if it wants
1003 * (necessary in the NFSv4.0 compound case)
1004 */
1005 rqstp->rq_cachetype = proc->pc_cachetype;
1006 /* Decode arguments */
1007 if (proc->pc_decode &&
1008 !proc->pc_decode(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base)) {
1009 dprintk("nfsd: failed to decode arguments!\n");
1010 *statp = rpc_garbage_args;
1011 return 1;
1012 }
1013
1014 /* Check whether we have this call in the cache. */
1015 switch (nfsd_cache_lookup(rqstp)) {
1016 case RC_DROPIT:
1017 return 0;
1018 case RC_REPLY:
1019 return 1;
1020 case RC_DOIT:;
1021 /* do it */
1022 }
1023
1024 /* need to grab the location to store the status, as
1025 * nfsv4 does some encoding while processing
1026 */
1027 nfserrp = rqstp->rq_res.head[0].iov_base
1028 + rqstp->rq_res.head[0].iov_len;
1029 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
1030
1031 /* Now call the procedure handler, and encode NFS status. */
1032 nfserr = proc->pc_func(rqstp);
1033 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
1034 if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
1035 dprintk("nfsd: Dropping request; may be revisited later\n");
1036 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1037 return 0;
1038 }
1039
1040 if (rqstp->rq_proc != 0)
1041 *nfserrp++ = nfserr;
1042
1043 /* Encode result.
1044 * For NFSv2, additional info is never returned in case of an error.
1045 */
1046 if (!(nfserr && rqstp->rq_vers == 2)) {
1047 if (proc->pc_encode && !proc->pc_encode(rqstp, nfserrp)) {
1048 /* Failed to encode result. Release cache entry */
1049 dprintk("nfsd: failed to encode result!\n");
1050 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1051 *statp = rpc_system_err;
1052 return 1;
1053 }
1054 }
1055
1056 /* Store reply in cache. */
1057 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
1058 return 1;
1059}
1060
1061int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1062{
1063 int ret;
1064 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1065
1066 mutex_lock(&nfsd_mutex);
1067 if (nn->nfsd_serv == NULL) {
1068 mutex_unlock(&nfsd_mutex);
1069 return -ENODEV;
1070 }
1071 /* bump up the psudo refcount while traversing */
1072 svc_get(nn->nfsd_serv);
1073 ret = svc_pool_stats_open(nn->nfsd_serv, file);
1074 mutex_unlock(&nfsd_mutex);
1075 return ret;
1076}
1077
1078int nfsd_pool_stats_release(struct inode *inode, struct file *file)
1079{
1080 int ret = seq_release(inode, file);
1081 struct net *net = inode->i_sb->s_fs_info;
1082
1083 mutex_lock(&nfsd_mutex);
1084 /* this function really, really should have been called svc_put() */
1085 nfsd_destroy(net);
1086 mutex_unlock(&nfsd_mutex);
1087 return ret;
1088}