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
34bool inter_copy_offload_enable;
35EXPORT_SYMBOL_GPL(inter_copy_offload_enable);
36module_param(inter_copy_offload_enable, bool, 0644);
37MODULE_PARM_DESC(inter_copy_offload_enable,
38 "Enable inter server to server copy offload. Default: false");
39
40extern struct svc_program nfsd_program;
41static int nfsd(void *vrqstp);
42#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
43static int nfsd_acl_rpcbind_set(struct net *,
44 const struct svc_program *,
45 u32, int,
46 unsigned short,
47 unsigned short);
48static __be32 nfsd_acl_init_request(struct svc_rqst *,
49 const struct svc_program *,
50 struct svc_process_info *);
51#endif
52static int nfsd_rpcbind_set(struct net *,
53 const struct svc_program *,
54 u32, int,
55 unsigned short,
56 unsigned short);
57static __be32 nfsd_init_request(struct svc_rqst *,
58 const struct svc_program *,
59 struct svc_process_info *);
60
61/*
62 * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members
63 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
64 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
65 *
66 * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a
67 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
68 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
69 * entry of ->sv_pools[].
70 *
71 * Transitions of the thread count between zero and non-zero are of particular
72 * interest since the svc_serv needs to be created and initialized at that
73 * point, or freed.
74 *
75 * Finally, the nfsd_mutex also protects some of the global variables that are
76 * accessed when nfsd starts and that are settable via the write_* routines in
77 * nfsctl.c. In particular:
78 *
79 * user_recovery_dirname
80 * user_lease_time
81 * nfsd_versions
82 */
83DEFINE_MUTEX(nfsd_mutex);
84
85/*
86 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
87 * nfsd_drc_max_pages limits the total amount of memory available for
88 * version 4.1 DRC caches.
89 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
90 */
91spinlock_t nfsd_drc_lock;
92unsigned long nfsd_drc_max_mem;
93unsigned long nfsd_drc_mem_used;
94
95#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
96static struct svc_stat nfsd_acl_svcstats;
97static const struct svc_version *nfsd_acl_version[] = {
98 [2] = &nfsd_acl_version2,
99 [3] = &nfsd_acl_version3,
100};
101
102#define NFSD_ACL_MINVERS 2
103#define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
104
105static struct svc_program nfsd_acl_program = {
106 .pg_prog = NFS_ACL_PROGRAM,
107 .pg_nvers = NFSD_ACL_NRVERS,
108 .pg_vers = nfsd_acl_version,
109 .pg_name = "nfsacl",
110 .pg_class = "nfsd",
111 .pg_stats = &nfsd_acl_svcstats,
112 .pg_authenticate = &svc_set_client,
113 .pg_init_request = nfsd_acl_init_request,
114 .pg_rpcbind_set = nfsd_acl_rpcbind_set,
115};
116
117static struct svc_stat nfsd_acl_svcstats = {
118 .program = &nfsd_acl_program,
119};
120#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
121
122static const struct svc_version *nfsd_version[] = {
123 [2] = &nfsd_version2,
124#if defined(CONFIG_NFSD_V3)
125 [3] = &nfsd_version3,
126#endif
127#if defined(CONFIG_NFSD_V4)
128 [4] = &nfsd_version4,
129#endif
130};
131
132#define NFSD_MINVERS 2
133#define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
134
135struct svc_program nfsd_program = {
136#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
137 .pg_next = &nfsd_acl_program,
138#endif
139 .pg_prog = NFS_PROGRAM, /* program number */
140 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
141 .pg_vers = nfsd_version, /* version table */
142 .pg_name = "nfsd", /* program name */
143 .pg_class = "nfsd", /* authentication class */
144 .pg_stats = &nfsd_svcstats, /* version table */
145 .pg_authenticate = &svc_set_client, /* export authentication */
146 .pg_init_request = nfsd_init_request,
147 .pg_rpcbind_set = nfsd_rpcbind_set,
148};
149
150static bool
151nfsd_support_version(int vers)
152{
153 if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
154 return nfsd_version[vers] != NULL;
155 return false;
156}
157
158static bool *
159nfsd_alloc_versions(void)
160{
161 bool *vers = kmalloc_array(NFSD_NRVERS, sizeof(bool), GFP_KERNEL);
162 unsigned i;
163
164 if (vers) {
165 /* All compiled versions are enabled by default */
166 for (i = 0; i < NFSD_NRVERS; i++)
167 vers[i] = nfsd_support_version(i);
168 }
169 return vers;
170}
171
172static bool *
173nfsd_alloc_minorversions(void)
174{
175 bool *vers = kmalloc_array(NFSD_SUPPORTED_MINOR_VERSION + 1,
176 sizeof(bool), GFP_KERNEL);
177 unsigned i;
178
179 if (vers) {
180 /* All minor versions are enabled by default */
181 for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
182 vers[i] = nfsd_support_version(4);
183 }
184 return vers;
185}
186
187void
188nfsd_netns_free_versions(struct nfsd_net *nn)
189{
190 kfree(nn->nfsd_versions);
191 kfree(nn->nfsd4_minorversions);
192 nn->nfsd_versions = NULL;
193 nn->nfsd4_minorversions = NULL;
194}
195
196static void
197nfsd_netns_init_versions(struct nfsd_net *nn)
198{
199 if (!nn->nfsd_versions) {
200 nn->nfsd_versions = nfsd_alloc_versions();
201 nn->nfsd4_minorversions = nfsd_alloc_minorversions();
202 if (!nn->nfsd_versions || !nn->nfsd4_minorversions)
203 nfsd_netns_free_versions(nn);
204 }
205}
206
207int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change)
208{
209 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
210 return 0;
211 switch(change) {
212 case NFSD_SET:
213 if (nn->nfsd_versions)
214 nn->nfsd_versions[vers] = nfsd_support_version(vers);
215 break;
216 case NFSD_CLEAR:
217 nfsd_netns_init_versions(nn);
218 if (nn->nfsd_versions)
219 nn->nfsd_versions[vers] = false;
220 break;
221 case NFSD_TEST:
222 if (nn->nfsd_versions)
223 return nn->nfsd_versions[vers];
224 fallthrough;
225 case NFSD_AVAIL:
226 return nfsd_support_version(vers);
227 }
228 return 0;
229}
230
231static void
232nfsd_adjust_nfsd_versions4(struct nfsd_net *nn)
233{
234 unsigned i;
235
236 for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
237 if (nn->nfsd4_minorversions[i])
238 return;
239 }
240 nfsd_vers(nn, 4, NFSD_CLEAR);
241}
242
243int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change)
244{
245 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
246 change != NFSD_AVAIL)
247 return -1;
248
249 switch(change) {
250 case NFSD_SET:
251 if (nn->nfsd4_minorversions) {
252 nfsd_vers(nn, 4, NFSD_SET);
253 nn->nfsd4_minorversions[minorversion] =
254 nfsd_vers(nn, 4, NFSD_TEST);
255 }
256 break;
257 case NFSD_CLEAR:
258 nfsd_netns_init_versions(nn);
259 if (nn->nfsd4_minorversions) {
260 nn->nfsd4_minorversions[minorversion] = false;
261 nfsd_adjust_nfsd_versions4(nn);
262 }
263 break;
264 case NFSD_TEST:
265 if (nn->nfsd4_minorversions)
266 return nn->nfsd4_minorversions[minorversion];
267 return nfsd_vers(nn, 4, NFSD_TEST);
268 case NFSD_AVAIL:
269 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION &&
270 nfsd_vers(nn, 4, NFSD_AVAIL);
271 }
272 return 0;
273}
274
275/*
276 * Maximum number of nfsd processes
277 */
278#define NFSD_MAXSERVS 8192
279
280int nfsd_nrthreads(struct net *net)
281{
282 int rv = 0;
283 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
284
285 mutex_lock(&nfsd_mutex);
286 if (nn->nfsd_serv)
287 rv = nn->nfsd_serv->sv_nrthreads;
288 mutex_unlock(&nfsd_mutex);
289 return rv;
290}
291
292static int nfsd_init_socks(struct net *net, const struct cred *cred)
293{
294 int error;
295 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
296
297 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
298 return 0;
299
300 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
301 SVC_SOCK_DEFAULTS, cred);
302 if (error < 0)
303 return error;
304
305 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
306 SVC_SOCK_DEFAULTS, cred);
307 if (error < 0)
308 return error;
309
310 return 0;
311}
312
313static int nfsd_users = 0;
314
315static int nfsd_startup_generic(int nrservs)
316{
317 int ret;
318
319 if (nfsd_users++)
320 return 0;
321
322 ret = nfsd_file_cache_init();
323 if (ret)
324 goto dec_users;
325
326 ret = nfs4_state_start();
327 if (ret)
328 goto out_file_cache;
329 return 0;
330
331out_file_cache:
332 nfsd_file_cache_shutdown();
333dec_users:
334 nfsd_users--;
335 return ret;
336}
337
338static void nfsd_shutdown_generic(void)
339{
340 if (--nfsd_users)
341 return;
342
343 nfs4_state_shutdown();
344 nfsd_file_cache_shutdown();
345}
346
347static bool nfsd_needs_lockd(struct nfsd_net *nn)
348{
349 return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST);
350}
351
352void nfsd_copy_boot_verifier(__be32 verf[2], struct nfsd_net *nn)
353{
354 int seq = 0;
355
356 do {
357 read_seqbegin_or_lock(&nn->boot_lock, &seq);
358 /*
359 * This is opaque to client, so no need to byte-swap. Use
360 * __force to keep sparse happy. y2038 time_t overflow is
361 * irrelevant in this usage
362 */
363 verf[0] = (__force __be32)nn->nfssvc_boot.tv_sec;
364 verf[1] = (__force __be32)nn->nfssvc_boot.tv_nsec;
365 } while (need_seqretry(&nn->boot_lock, seq));
366 done_seqretry(&nn->boot_lock, seq);
367}
368
369static void nfsd_reset_boot_verifier_locked(struct nfsd_net *nn)
370{
371 ktime_get_real_ts64(&nn->nfssvc_boot);
372}
373
374void nfsd_reset_boot_verifier(struct nfsd_net *nn)
375{
376 write_seqlock(&nn->boot_lock);
377 nfsd_reset_boot_verifier_locked(nn);
378 write_sequnlock(&nn->boot_lock);
379}
380
381static int nfsd_startup_net(int nrservs, struct net *net, const struct cred *cred)
382{
383 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
384 int ret;
385
386 if (nn->nfsd_net_up)
387 return 0;
388
389 ret = nfsd_startup_generic(nrservs);
390 if (ret)
391 return ret;
392 ret = nfsd_init_socks(net, cred);
393 if (ret)
394 goto out_socks;
395
396 if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
397 ret = lockd_up(net, cred);
398 if (ret)
399 goto out_socks;
400 nn->lockd_up = true;
401 }
402
403 ret = nfsd_file_cache_start_net(net);
404 if (ret)
405 goto out_lockd;
406 ret = nfs4_state_start_net(net);
407 if (ret)
408 goto out_filecache;
409
410 nn->nfsd_net_up = true;
411 return 0;
412
413out_filecache:
414 nfsd_file_cache_shutdown_net(net);
415out_lockd:
416 if (nn->lockd_up) {
417 lockd_down(net);
418 nn->lockd_up = false;
419 }
420out_socks:
421 nfsd_shutdown_generic();
422 return ret;
423}
424
425static void nfsd_shutdown_net(struct net *net)
426{
427 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
428
429 nfsd_file_cache_shutdown_net(net);
430 nfs4_state_shutdown_net(net);
431 if (nn->lockd_up) {
432 lockd_down(net);
433 nn->lockd_up = false;
434 }
435 nn->nfsd_net_up = false;
436 nfsd_shutdown_generic();
437}
438
439static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
440 void *ptr)
441{
442 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
443 struct net_device *dev = ifa->ifa_dev->dev;
444 struct net *net = dev_net(dev);
445 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
446 struct sockaddr_in sin;
447
448 if ((event != NETDEV_DOWN) ||
449 !atomic_inc_not_zero(&nn->ntf_refcnt))
450 goto out;
451
452 if (nn->nfsd_serv) {
453 dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
454 sin.sin_family = AF_INET;
455 sin.sin_addr.s_addr = ifa->ifa_local;
456 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
457 }
458 atomic_dec(&nn->ntf_refcnt);
459 wake_up(&nn->ntf_wq);
460
461out:
462 return NOTIFY_DONE;
463}
464
465static struct notifier_block nfsd_inetaddr_notifier = {
466 .notifier_call = nfsd_inetaddr_event,
467};
468
469#if IS_ENABLED(CONFIG_IPV6)
470static int nfsd_inet6addr_event(struct notifier_block *this,
471 unsigned long event, void *ptr)
472{
473 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
474 struct net_device *dev = ifa->idev->dev;
475 struct net *net = dev_net(dev);
476 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
477 struct sockaddr_in6 sin6;
478
479 if ((event != NETDEV_DOWN) ||
480 !atomic_inc_not_zero(&nn->ntf_refcnt))
481 goto out;
482
483 if (nn->nfsd_serv) {
484 dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
485 sin6.sin6_family = AF_INET6;
486 sin6.sin6_addr = ifa->addr;
487 if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
488 sin6.sin6_scope_id = ifa->idev->dev->ifindex;
489 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
490 }
491 atomic_dec(&nn->ntf_refcnt);
492 wake_up(&nn->ntf_wq);
493out:
494 return NOTIFY_DONE;
495}
496
497static struct notifier_block nfsd_inet6addr_notifier = {
498 .notifier_call = nfsd_inet6addr_event,
499};
500#endif
501
502/* Only used under nfsd_mutex, so this atomic may be overkill: */
503static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
504
505static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
506{
507 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
508
509 atomic_dec(&nn->ntf_refcnt);
510 /* check if the notifier still has clients */
511 if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
512 unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
513#if IS_ENABLED(CONFIG_IPV6)
514 unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
515#endif
516 }
517 wait_event(nn->ntf_wq, atomic_read(&nn->ntf_refcnt) == 0);
518
519 /*
520 * write_ports can create the server without actually starting
521 * any threads--if we get shut down before any threads are
522 * started, then nfsd_last_thread will be run before any of this
523 * other initialization has been done except the rpcb information.
524 */
525 svc_rpcb_cleanup(serv, net);
526 if (!nn->nfsd_net_up)
527 return;
528
529 nfsd_shutdown_net(net);
530 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
531 "cache\n");
532 nfsd_export_flush(net);
533}
534
535void nfsd_reset_versions(struct nfsd_net *nn)
536{
537 int i;
538
539 for (i = 0; i < NFSD_NRVERS; i++)
540 if (nfsd_vers(nn, i, NFSD_TEST))
541 return;
542
543 for (i = 0; i < NFSD_NRVERS; i++)
544 if (i != 4)
545 nfsd_vers(nn, i, NFSD_SET);
546 else {
547 int minor = 0;
548 while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
549 minor++;
550 }
551}
552
553/*
554 * Each session guarantees a negotiated per slot memory cache for replies
555 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
556 * NFSv4.1 server might want to use more memory for a DRC than a machine
557 * with mutiple services.
558 *
559 * Impose a hard limit on the number of pages for the DRC which varies
560 * according to the machines free pages. This is of course only a default.
561 *
562 * For now this is a #defined shift which could be under admin control
563 * in the future.
564 */
565static void set_max_drc(void)
566{
567 #define NFSD_DRC_SIZE_SHIFT 7
568 nfsd_drc_max_mem = (nr_free_buffer_pages()
569 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
570 nfsd_drc_mem_used = 0;
571 spin_lock_init(&nfsd_drc_lock);
572 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
573}
574
575static int nfsd_get_default_max_blksize(void)
576{
577 struct sysinfo i;
578 unsigned long long target;
579 unsigned long ret;
580
581 si_meminfo(&i);
582 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
583 /*
584 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
585 * machines, but only uses 32K on 128M machines. Bottom out at
586 * 8K on 32M and smaller. Of course, this is only a default.
587 */
588 target >>= 12;
589
590 ret = NFSSVC_MAXBLKSIZE;
591 while (ret > target && ret >= 8*1024*2)
592 ret /= 2;
593 return ret;
594}
595
596static const struct svc_serv_ops nfsd_thread_sv_ops = {
597 .svo_shutdown = nfsd_last_thread,
598 .svo_function = nfsd,
599 .svo_enqueue_xprt = svc_xprt_do_enqueue,
600 .svo_setup = svc_set_num_threads,
601 .svo_module = THIS_MODULE,
602};
603
604bool i_am_nfsd(void)
605{
606 return kthread_func(current) == nfsd;
607}
608
609int nfsd_create_serv(struct net *net)
610{
611 int error;
612 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
613
614 WARN_ON(!mutex_is_locked(&nfsd_mutex));
615 if (nn->nfsd_serv) {
616 svc_get(nn->nfsd_serv);
617 return 0;
618 }
619 if (nfsd_max_blksize == 0)
620 nfsd_max_blksize = nfsd_get_default_max_blksize();
621 nfsd_reset_versions(nn);
622 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
623 &nfsd_thread_sv_ops);
624 if (nn->nfsd_serv == NULL)
625 return -ENOMEM;
626
627 nn->nfsd_serv->sv_maxconn = nn->max_connections;
628 error = svc_bind(nn->nfsd_serv, net);
629 if (error < 0) {
630 svc_destroy(nn->nfsd_serv);
631 return error;
632 }
633
634 set_max_drc();
635 /* check if the notifier is already set */
636 if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
637 register_inetaddr_notifier(&nfsd_inetaddr_notifier);
638#if IS_ENABLED(CONFIG_IPV6)
639 register_inet6addr_notifier(&nfsd_inet6addr_notifier);
640#endif
641 }
642 atomic_inc(&nn->ntf_refcnt);
643 nfsd_reset_boot_verifier(nn);
644 return 0;
645}
646
647int nfsd_nrpools(struct net *net)
648{
649 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
650
651 if (nn->nfsd_serv == NULL)
652 return 0;
653 else
654 return nn->nfsd_serv->sv_nrpools;
655}
656
657int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
658{
659 int i = 0;
660 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
661
662 if (nn->nfsd_serv != NULL) {
663 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
664 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
665 }
666
667 return 0;
668}
669
670void nfsd_destroy(struct net *net)
671{
672 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
673 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
674
675 if (destroy)
676 svc_shutdown_net(nn->nfsd_serv, net);
677 svc_destroy(nn->nfsd_serv);
678 if (destroy)
679 nn->nfsd_serv = NULL;
680}
681
682int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
683{
684 int i = 0;
685 int tot = 0;
686 int err = 0;
687 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
688
689 WARN_ON(!mutex_is_locked(&nfsd_mutex));
690
691 if (nn->nfsd_serv == NULL || n <= 0)
692 return 0;
693
694 if (n > nn->nfsd_serv->sv_nrpools)
695 n = nn->nfsd_serv->sv_nrpools;
696
697 /* enforce a global maximum number of threads */
698 tot = 0;
699 for (i = 0; i < n; i++) {
700 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
701 tot += nthreads[i];
702 }
703 if (tot > NFSD_MAXSERVS) {
704 /* total too large: scale down requested numbers */
705 for (i = 0; i < n && tot > 0; i++) {
706 int new = nthreads[i] * NFSD_MAXSERVS / tot;
707 tot -= (nthreads[i] - new);
708 nthreads[i] = new;
709 }
710 for (i = 0; i < n && tot > 0; i++) {
711 nthreads[i]--;
712 tot--;
713 }
714 }
715
716 /*
717 * There must always be a thread in pool 0; the admin
718 * can't shut down NFS completely using pool_threads.
719 */
720 if (nthreads[0] == 0)
721 nthreads[0] = 1;
722
723 /* apply the new numbers */
724 svc_get(nn->nfsd_serv);
725 for (i = 0; i < n; i++) {
726 err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
727 &nn->nfsd_serv->sv_pools[i], nthreads[i]);
728 if (err)
729 break;
730 }
731 nfsd_destroy(net);
732 return err;
733}
734
735/*
736 * Adjust the number of threads and return the new number of threads.
737 * This is also the function that starts the server if necessary, if
738 * this is the first time nrservs is nonzero.
739 */
740int
741nfsd_svc(int nrservs, struct net *net, const struct cred *cred)
742{
743 int error;
744 bool nfsd_up_before;
745 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
746
747 mutex_lock(&nfsd_mutex);
748 dprintk("nfsd: creating service\n");
749
750 nrservs = max(nrservs, 0);
751 nrservs = min(nrservs, NFSD_MAXSERVS);
752 error = 0;
753
754 if (nrservs == 0 && nn->nfsd_serv == NULL)
755 goto out;
756
757 strlcpy(nn->nfsd_name, utsname()->nodename,
758 sizeof(nn->nfsd_name));
759
760 error = nfsd_create_serv(net);
761 if (error)
762 goto out;
763
764 nfsd_up_before = nn->nfsd_net_up;
765
766 error = nfsd_startup_net(nrservs, net, cred);
767 if (error)
768 goto out_destroy;
769 error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
770 NULL, nrservs);
771 if (error)
772 goto out_shutdown;
773 /* We are holding a reference to nn->nfsd_serv which
774 * we don't want to count in the return value,
775 * so subtract 1
776 */
777 error = nn->nfsd_serv->sv_nrthreads - 1;
778out_shutdown:
779 if (error < 0 && !nfsd_up_before)
780 nfsd_shutdown_net(net);
781out_destroy:
782 nfsd_destroy(net); /* Release server */
783out:
784 mutex_unlock(&nfsd_mutex);
785 return error;
786}
787
788#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
789static bool
790nfsd_support_acl_version(int vers)
791{
792 if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
793 return nfsd_acl_version[vers] != NULL;
794 return false;
795}
796
797static int
798nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
799 u32 version, int family, unsigned short proto,
800 unsigned short port)
801{
802 if (!nfsd_support_acl_version(version) ||
803 !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
804 return 0;
805 return svc_generic_rpcbind_set(net, progp, version, family,
806 proto, port);
807}
808
809static __be32
810nfsd_acl_init_request(struct svc_rqst *rqstp,
811 const struct svc_program *progp,
812 struct svc_process_info *ret)
813{
814 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
815 int i;
816
817 if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
818 nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
819 return svc_generic_init_request(rqstp, progp, ret);
820
821 ret->mismatch.lovers = NFSD_ACL_NRVERS;
822 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
823 if (nfsd_support_acl_version(rqstp->rq_vers) &&
824 nfsd_vers(nn, i, NFSD_TEST)) {
825 ret->mismatch.lovers = i;
826 break;
827 }
828 }
829 if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
830 return rpc_prog_unavail;
831 ret->mismatch.hivers = NFSD_ACL_MINVERS;
832 for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
833 if (nfsd_support_acl_version(rqstp->rq_vers) &&
834 nfsd_vers(nn, i, NFSD_TEST)) {
835 ret->mismatch.hivers = i;
836 break;
837 }
838 }
839 return rpc_prog_mismatch;
840}
841#endif
842
843static int
844nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
845 u32 version, int family, unsigned short proto,
846 unsigned short port)
847{
848 if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
849 return 0;
850 return svc_generic_rpcbind_set(net, progp, version, family,
851 proto, port);
852}
853
854static __be32
855nfsd_init_request(struct svc_rqst *rqstp,
856 const struct svc_program *progp,
857 struct svc_process_info *ret)
858{
859 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
860 int i;
861
862 if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
863 return svc_generic_init_request(rqstp, progp, ret);
864
865 ret->mismatch.lovers = NFSD_NRVERS;
866 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
867 if (nfsd_vers(nn, i, NFSD_TEST)) {
868 ret->mismatch.lovers = i;
869 break;
870 }
871 }
872 if (ret->mismatch.lovers == NFSD_NRVERS)
873 return rpc_prog_unavail;
874 ret->mismatch.hivers = NFSD_MINVERS;
875 for (i = NFSD_NRVERS - 1; i >= NFSD_MINVERS; i--) {
876 if (nfsd_vers(nn, i, NFSD_TEST)) {
877 ret->mismatch.hivers = i;
878 break;
879 }
880 }
881 return rpc_prog_mismatch;
882}
883
884/*
885 * This is the NFS server kernel thread
886 */
887static int
888nfsd(void *vrqstp)
889{
890 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
891 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
892 struct net *net = perm_sock->xpt_net;
893 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
894 int err;
895
896 /* Lock module and set up kernel thread */
897 mutex_lock(&nfsd_mutex);
898
899 /* At this point, the thread shares current->fs
900 * with the init process. We need to create files with the
901 * umask as defined by the client instead of init's umask. */
902 if (unshare_fs_struct() < 0) {
903 printk("Unable to start nfsd thread: out of memory\n");
904 goto out;
905 }
906
907 current->fs->umask = 0;
908
909 /*
910 * thread is spawned with all signals set to SIG_IGN, re-enable
911 * the ones that will bring down the thread
912 */
913 allow_signal(SIGKILL);
914 allow_signal(SIGHUP);
915 allow_signal(SIGINT);
916 allow_signal(SIGQUIT);
917
918 nfsdstats.th_cnt++;
919 mutex_unlock(&nfsd_mutex);
920
921 set_freezable();
922
923 /*
924 * The main request loop
925 */
926 for (;;) {
927 /* Update sv_maxconn if it has changed */
928 rqstp->rq_server->sv_maxconn = nn->max_connections;
929
930 /*
931 * Find a socket with data available and call its
932 * recvfrom routine.
933 */
934 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
935 ;
936 if (err == -EINTR)
937 break;
938 validate_process_creds();
939 svc_process(rqstp);
940 validate_process_creds();
941 }
942
943 /* Clear signals before calling svc_exit_thread() */
944 flush_signals(current);
945
946 mutex_lock(&nfsd_mutex);
947 nfsdstats.th_cnt --;
948
949out:
950 rqstp->rq_server = NULL;
951
952 /* Release the thread */
953 svc_exit_thread(rqstp);
954
955 nfsd_destroy(net);
956
957 /* Release module */
958 mutex_unlock(&nfsd_mutex);
959 module_put_and_exit(0);
960 return 0;
961}
962
963static __be32 map_new_errors(u32 vers, __be32 nfserr)
964{
965 if (nfserr == nfserr_jukebox && vers == 2)
966 return nfserr_dropit;
967 if (nfserr == nfserr_wrongsec && vers < 4)
968 return nfserr_acces;
969 return nfserr;
970}
971
972/*
973 * A write procedure can have a large argument, and a read procedure can
974 * have a large reply, but no NFSv2 or NFSv3 procedure has argument and
975 * reply that can both be larger than a page. The xdr code has taken
976 * advantage of this assumption to be a sloppy about bounds checking in
977 * some cases. Pending a rewrite of the NFSv2/v3 xdr code to fix that
978 * problem, we enforce these assumptions here:
979 */
980static bool nfs_request_too_big(struct svc_rqst *rqstp,
981 const struct svc_procedure *proc)
982{
983 /*
984 * The ACL code has more careful bounds-checking and is not
985 * susceptible to this problem:
986 */
987 if (rqstp->rq_prog != NFS_PROGRAM)
988 return false;
989 /*
990 * Ditto NFSv4 (which can in theory have argument and reply both
991 * more than a page):
992 */
993 if (rqstp->rq_vers >= 4)
994 return false;
995 /* The reply will be small, we're OK: */
996 if (proc->pc_xdrressize > 0 &&
997 proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE))
998 return false;
999
1000 return rqstp->rq_arg.len > PAGE_SIZE;
1001}
1002
1003int
1004nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1005{
1006 const struct svc_procedure *proc;
1007 __be32 nfserr;
1008 __be32 *nfserrp;
1009
1010 dprintk("nfsd_dispatch: vers %d proc %d\n",
1011 rqstp->rq_vers, rqstp->rq_proc);
1012 proc = rqstp->rq_procinfo;
1013
1014 if (nfs_request_too_big(rqstp, proc)) {
1015 dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers);
1016 *statp = rpc_garbage_args;
1017 return 1;
1018 }
1019 rqstp->rq_lease_breaker = NULL;
1020 /*
1021 * Give the xdr decoder a chance to change this if it wants
1022 * (necessary in the NFSv4.0 compound case)
1023 */
1024 rqstp->rq_cachetype = proc->pc_cachetype;
1025 /* Decode arguments */
1026 if (proc->pc_decode &&
1027 !proc->pc_decode(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base)) {
1028 dprintk("nfsd: failed to decode arguments!\n");
1029 *statp = rpc_garbage_args;
1030 return 1;
1031 }
1032
1033 /* Check whether we have this call in the cache. */
1034 switch (nfsd_cache_lookup(rqstp)) {
1035 case RC_DROPIT:
1036 return 0;
1037 case RC_REPLY:
1038 return 1;
1039 case RC_DOIT:;
1040 /* do it */
1041 }
1042
1043 /* need to grab the location to store the status, as
1044 * nfsv4 does some encoding while processing
1045 */
1046 nfserrp = rqstp->rq_res.head[0].iov_base
1047 + rqstp->rq_res.head[0].iov_len;
1048 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
1049
1050 /* Now call the procedure handler, and encode NFS status. */
1051 nfserr = proc->pc_func(rqstp);
1052 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
1053 if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
1054 dprintk("nfsd: Dropping request; may be revisited later\n");
1055 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1056 return 0;
1057 }
1058
1059 if (rqstp->rq_proc != 0)
1060 *nfserrp++ = nfserr;
1061
1062 /* Encode result.
1063 * For NFSv2, additional info is never returned in case of an error.
1064 */
1065 if (!(nfserr && rqstp->rq_vers == 2)) {
1066 if (proc->pc_encode && !proc->pc_encode(rqstp, nfserrp)) {
1067 /* Failed to encode result. Release cache entry */
1068 dprintk("nfsd: failed to encode result!\n");
1069 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1070 *statp = rpc_system_err;
1071 return 1;
1072 }
1073 }
1074
1075 /* Store reply in cache. */
1076 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
1077 return 1;
1078}
1079
1080int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1081{
1082 int ret;
1083 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1084
1085 mutex_lock(&nfsd_mutex);
1086 if (nn->nfsd_serv == NULL) {
1087 mutex_unlock(&nfsd_mutex);
1088 return -ENODEV;
1089 }
1090 /* bump up the psudo refcount while traversing */
1091 svc_get(nn->nfsd_serv);
1092 ret = svc_pool_stats_open(nn->nfsd_serv, file);
1093 mutex_unlock(&nfsd_mutex);
1094 return ret;
1095}
1096
1097int nfsd_pool_stats_release(struct inode *inode, struct file *file)
1098{
1099 int ret = seq_release(inode, file);
1100 struct net *net = inode->i_sb->s_fs_info;
1101
1102 mutex_lock(&nfsd_mutex);
1103 /* this function really, really should have been called svc_put() */
1104 nfsd_destroy(net);
1105 mutex_unlock(&nfsd_mutex);
1106 return ret;
1107}