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/fs_struct.h>
12#include <linux/swap.h>
13
14#include <linux/sunrpc/stats.h>
15#include <linux/sunrpc/svcsock.h>
16#include <linux/lockd/bind.h>
17#include <linux/nfsacl.h>
18#include <linux/seq_file.h>
19#include <net/net_namespace.h>
20#include "nfsd.h"
21#include "cache.h"
22#include "vfs.h"
23
24#define NFSDDBG_FACILITY NFSDDBG_SVC
25
26extern struct svc_program nfsd_program;
27static int nfsd(void *vrqstp);
28struct timeval nfssvc_boot;
29
30/*
31 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
32 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
33 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
34 *
35 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
36 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
37 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
38 * entry of ->sv_pools[].
39 *
40 * Transitions of the thread count between zero and non-zero are of particular
41 * interest since the svc_serv needs to be created and initialized at that
42 * point, or freed.
43 *
44 * Finally, the nfsd_mutex also protects some of the global variables that are
45 * accessed when nfsd starts and that are settable via the write_* routines in
46 * nfsctl.c. In particular:
47 *
48 * user_recovery_dirname
49 * user_lease_time
50 * nfsd_versions
51 */
52DEFINE_MUTEX(nfsd_mutex);
53struct svc_serv *nfsd_serv;
54
55/*
56 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
57 * nfsd_drc_max_pages limits the total amount of memory available for
58 * version 4.1 DRC caches.
59 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
60 */
61spinlock_t nfsd_drc_lock;
62unsigned int nfsd_drc_max_mem;
63unsigned int nfsd_drc_mem_used;
64
65#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
66static struct svc_stat nfsd_acl_svcstats;
67static struct svc_version * nfsd_acl_version[] = {
68 [2] = &nfsd_acl_version2,
69 [3] = &nfsd_acl_version3,
70};
71
72#define NFSD_ACL_MINVERS 2
73#define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
74static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
75
76static struct svc_program nfsd_acl_program = {
77 .pg_prog = NFS_ACL_PROGRAM,
78 .pg_nvers = NFSD_ACL_NRVERS,
79 .pg_vers = nfsd_acl_versions,
80 .pg_name = "nfsacl",
81 .pg_class = "nfsd",
82 .pg_stats = &nfsd_acl_svcstats,
83 .pg_authenticate = &svc_set_client,
84};
85
86static struct svc_stat nfsd_acl_svcstats = {
87 .program = &nfsd_acl_program,
88};
89#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
90
91static struct svc_version * nfsd_version[] = {
92 [2] = &nfsd_version2,
93#if defined(CONFIG_NFSD_V3)
94 [3] = &nfsd_version3,
95#endif
96#if defined(CONFIG_NFSD_V4)
97 [4] = &nfsd_version4,
98#endif
99};
100
101#define NFSD_MINVERS 2
102#define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
103static struct svc_version *nfsd_versions[NFSD_NRVERS];
104
105struct svc_program nfsd_program = {
106#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
107 .pg_next = &nfsd_acl_program,
108#endif
109 .pg_prog = NFS_PROGRAM, /* program number */
110 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
111 .pg_vers = nfsd_versions, /* version table */
112 .pg_name = "nfsd", /* program name */
113 .pg_class = "nfsd", /* authentication class */
114 .pg_stats = &nfsd_svcstats, /* version table */
115 .pg_authenticate = &svc_set_client, /* export authentication */
116
117};
118
119u32 nfsd_supported_minorversion;
120
121int nfsd_vers(int vers, enum vers_op change)
122{
123 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
124 return 0;
125 switch(change) {
126 case NFSD_SET:
127 nfsd_versions[vers] = nfsd_version[vers];
128#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
129 if (vers < NFSD_ACL_NRVERS)
130 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
131#endif
132 break;
133 case NFSD_CLEAR:
134 nfsd_versions[vers] = NULL;
135#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
136 if (vers < NFSD_ACL_NRVERS)
137 nfsd_acl_versions[vers] = NULL;
138#endif
139 break;
140 case NFSD_TEST:
141 return nfsd_versions[vers] != NULL;
142 case NFSD_AVAIL:
143 return nfsd_version[vers] != NULL;
144 }
145 return 0;
146}
147
148int nfsd_minorversion(u32 minorversion, enum vers_op change)
149{
150 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
151 return -1;
152 switch(change) {
153 case NFSD_SET:
154 nfsd_supported_minorversion = minorversion;
155 break;
156 case NFSD_CLEAR:
157 if (minorversion == 0)
158 return -1;
159 nfsd_supported_minorversion = minorversion - 1;
160 break;
161 case NFSD_TEST:
162 return minorversion <= nfsd_supported_minorversion;
163 case NFSD_AVAIL:
164 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
165 }
166 return 0;
167}
168
169/*
170 * Maximum number of nfsd processes
171 */
172#define NFSD_MAXSERVS 8192
173
174int nfsd_nrthreads(void)
175{
176 int rv = 0;
177 mutex_lock(&nfsd_mutex);
178 if (nfsd_serv)
179 rv = nfsd_serv->sv_nrthreads;
180 mutex_unlock(&nfsd_mutex);
181 return rv;
182}
183
184static int nfsd_init_socks(int port)
185{
186 int error;
187 if (!list_empty(&nfsd_serv->sv_permsocks))
188 return 0;
189
190 error = svc_create_xprt(nfsd_serv, "udp", &init_net, PF_INET, port,
191 SVC_SOCK_DEFAULTS);
192 if (error < 0)
193 return error;
194
195 error = svc_create_xprt(nfsd_serv, "tcp", &init_net, PF_INET, port,
196 SVC_SOCK_DEFAULTS);
197 if (error < 0)
198 return error;
199
200 return 0;
201}
202
203static bool nfsd_up = false;
204
205static int nfsd_startup(unsigned short port, int nrservs)
206{
207 int ret;
208
209 if (nfsd_up)
210 return 0;
211 /*
212 * Readahead param cache - will no-op if it already exists.
213 * (Note therefore results will be suboptimal if number of
214 * threads is modified after nfsd start.)
215 */
216 ret = nfsd_racache_init(2*nrservs);
217 if (ret)
218 return ret;
219 ret = nfsd_init_socks(port);
220 if (ret)
221 goto out_racache;
222 ret = lockd_up();
223 if (ret)
224 goto out_racache;
225 ret = nfs4_state_start();
226 if (ret)
227 goto out_lockd;
228 nfsd_up = true;
229 return 0;
230out_lockd:
231 lockd_down();
232out_racache:
233 nfsd_racache_shutdown();
234 return ret;
235}
236
237static void nfsd_shutdown(void)
238{
239 /*
240 * write_ports can create the server without actually starting
241 * any threads--if we get shut down before any threads are
242 * started, then nfsd_last_thread will be run before any of this
243 * other initialization has been done.
244 */
245 if (!nfsd_up)
246 return;
247 nfs4_state_shutdown();
248 lockd_down();
249 nfsd_racache_shutdown();
250 nfsd_up = false;
251}
252
253static void nfsd_last_thread(struct svc_serv *serv)
254{
255 /* When last nfsd thread exits we need to do some clean-up */
256 nfsd_serv = NULL;
257 nfsd_shutdown();
258
259 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
260 "cache\n");
261 nfsd_export_flush();
262}
263
264void nfsd_reset_versions(void)
265{
266 int found_one = 0;
267 int i;
268
269 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
270 if (nfsd_program.pg_vers[i])
271 found_one = 1;
272 }
273
274 if (!found_one) {
275 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
276 nfsd_program.pg_vers[i] = nfsd_version[i];
277#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
278 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
279 nfsd_acl_program.pg_vers[i] =
280 nfsd_acl_version[i];
281#endif
282 }
283}
284
285/*
286 * Each session guarantees a negotiated per slot memory cache for replies
287 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
288 * NFSv4.1 server might want to use more memory for a DRC than a machine
289 * with mutiple services.
290 *
291 * Impose a hard limit on the number of pages for the DRC which varies
292 * according to the machines free pages. This is of course only a default.
293 *
294 * For now this is a #defined shift which could be under admin control
295 * in the future.
296 */
297static void set_max_drc(void)
298{
299 #define NFSD_DRC_SIZE_SHIFT 10
300 nfsd_drc_max_mem = (nr_free_buffer_pages()
301 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
302 nfsd_drc_mem_used = 0;
303 spin_lock_init(&nfsd_drc_lock);
304 dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
305}
306
307int nfsd_create_serv(void)
308{
309 int err = 0;
310
311 WARN_ON(!mutex_is_locked(&nfsd_mutex));
312 if (nfsd_serv) {
313 svc_get(nfsd_serv);
314 return 0;
315 }
316 if (nfsd_max_blksize == 0) {
317 /* choose a suitable default */
318 struct sysinfo i;
319 si_meminfo(&i);
320 /* Aim for 1/4096 of memory per thread
321 * This gives 1MB on 4Gig machines
322 * But only uses 32K on 128M machines.
323 * Bottom out at 8K on 32M and smaller.
324 * Of course, this is only a default.
325 */
326 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
327 i.totalram <<= PAGE_SHIFT - 12;
328 while (nfsd_max_blksize > i.totalram &&
329 nfsd_max_blksize >= 8*1024*2)
330 nfsd_max_blksize /= 2;
331 }
332 nfsd_reset_versions();
333
334 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
335 nfsd_last_thread, nfsd, THIS_MODULE);
336 if (nfsd_serv == NULL)
337 return -ENOMEM;
338
339 set_max_drc();
340 do_gettimeofday(&nfssvc_boot); /* record boot time */
341 return err;
342}
343
344int nfsd_nrpools(void)
345{
346 if (nfsd_serv == NULL)
347 return 0;
348 else
349 return nfsd_serv->sv_nrpools;
350}
351
352int nfsd_get_nrthreads(int n, int *nthreads)
353{
354 int i = 0;
355
356 if (nfsd_serv != NULL) {
357 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
358 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
359 }
360
361 return 0;
362}
363
364int nfsd_set_nrthreads(int n, int *nthreads)
365{
366 int i = 0;
367 int tot = 0;
368 int err = 0;
369
370 WARN_ON(!mutex_is_locked(&nfsd_mutex));
371
372 if (nfsd_serv == NULL || n <= 0)
373 return 0;
374
375 if (n > nfsd_serv->sv_nrpools)
376 n = nfsd_serv->sv_nrpools;
377
378 /* enforce a global maximum number of threads */
379 tot = 0;
380 for (i = 0; i < n; i++) {
381 if (nthreads[i] > NFSD_MAXSERVS)
382 nthreads[i] = NFSD_MAXSERVS;
383 tot += nthreads[i];
384 }
385 if (tot > NFSD_MAXSERVS) {
386 /* total too large: scale down requested numbers */
387 for (i = 0; i < n && tot > 0; i++) {
388 int new = nthreads[i] * NFSD_MAXSERVS / tot;
389 tot -= (nthreads[i] - new);
390 nthreads[i] = new;
391 }
392 for (i = 0; i < n && tot > 0; i++) {
393 nthreads[i]--;
394 tot--;
395 }
396 }
397
398 /*
399 * There must always be a thread in pool 0; the admin
400 * can't shut down NFS completely using pool_threads.
401 */
402 if (nthreads[0] == 0)
403 nthreads[0] = 1;
404
405 /* apply the new numbers */
406 svc_get(nfsd_serv);
407 for (i = 0; i < n; i++) {
408 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
409 nthreads[i]);
410 if (err)
411 break;
412 }
413 svc_destroy(nfsd_serv);
414
415 return err;
416}
417
418/*
419 * Adjust the number of threads and return the new number of threads.
420 * This is also the function that starts the server if necessary, if
421 * this is the first time nrservs is nonzero.
422 */
423int
424nfsd_svc(unsigned short port, int nrservs)
425{
426 int error;
427 bool nfsd_up_before;
428
429 mutex_lock(&nfsd_mutex);
430 dprintk("nfsd: creating service\n");
431 if (nrservs <= 0)
432 nrservs = 0;
433 if (nrservs > NFSD_MAXSERVS)
434 nrservs = NFSD_MAXSERVS;
435 error = 0;
436 if (nrservs == 0 && nfsd_serv == NULL)
437 goto out;
438
439 error = nfsd_create_serv();
440 if (error)
441 goto out;
442
443 nfsd_up_before = nfsd_up;
444
445 error = nfsd_startup(port, nrservs);
446 if (error)
447 goto out_destroy;
448 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
449 if (error)
450 goto out_shutdown;
451 /* We are holding a reference to nfsd_serv which
452 * we don't want to count in the return value,
453 * so subtract 1
454 */
455 error = nfsd_serv->sv_nrthreads - 1;
456out_shutdown:
457 if (error < 0 && !nfsd_up_before)
458 nfsd_shutdown();
459out_destroy:
460 svc_destroy(nfsd_serv); /* Release server */
461out:
462 mutex_unlock(&nfsd_mutex);
463 return error;
464}
465
466
467/*
468 * This is the NFS server kernel thread
469 */
470static int
471nfsd(void *vrqstp)
472{
473 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
474 int err, preverr = 0;
475
476 /* Lock module and set up kernel thread */
477 mutex_lock(&nfsd_mutex);
478
479 /* At this point, the thread shares current->fs
480 * with the init process. We need to create files with a
481 * umask of 0 instead of init's umask. */
482 if (unshare_fs_struct() < 0) {
483 printk("Unable to start nfsd thread: out of memory\n");
484 goto out;
485 }
486
487 current->fs->umask = 0;
488
489 /*
490 * thread is spawned with all signals set to SIG_IGN, re-enable
491 * the ones that will bring down the thread
492 */
493 allow_signal(SIGKILL);
494 allow_signal(SIGHUP);
495 allow_signal(SIGINT);
496 allow_signal(SIGQUIT);
497
498 nfsdstats.th_cnt++;
499 mutex_unlock(&nfsd_mutex);
500
501 /*
502 * We want less throttling in balance_dirty_pages() so that nfs to
503 * localhost doesn't cause nfsd to lock up due to all the client's
504 * dirty pages.
505 */
506 current->flags |= PF_LESS_THROTTLE;
507 set_freezable();
508
509 /*
510 * The main request loop
511 */
512 for (;;) {
513 /*
514 * Find a socket with data available and call its
515 * recvfrom routine.
516 */
517 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
518 ;
519 if (err == -EINTR)
520 break;
521 else if (err < 0) {
522 if (err != preverr) {
523 printk(KERN_WARNING "%s: unexpected error "
524 "from svc_recv (%d)\n", __func__, -err);
525 preverr = err;
526 }
527 schedule_timeout_uninterruptible(HZ);
528 continue;
529 }
530
531 validate_process_creds();
532 svc_process(rqstp);
533 validate_process_creds();
534 }
535
536 /* Clear signals before calling svc_exit_thread() */
537 flush_signals(current);
538
539 mutex_lock(&nfsd_mutex);
540 nfsdstats.th_cnt --;
541
542out:
543 /* Release the thread */
544 svc_exit_thread(rqstp);
545
546 /* Release module */
547 mutex_unlock(&nfsd_mutex);
548 module_put_and_exit(0);
549 return 0;
550}
551
552static __be32 map_new_errors(u32 vers, __be32 nfserr)
553{
554 if (nfserr == nfserr_jukebox && vers == 2)
555 return nfserr_dropit;
556 if (nfserr == nfserr_wrongsec && vers < 4)
557 return nfserr_acces;
558 return nfserr;
559}
560
561int
562nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
563{
564 struct svc_procedure *proc;
565 kxdrproc_t xdr;
566 __be32 nfserr;
567 __be32 *nfserrp;
568
569 dprintk("nfsd_dispatch: vers %d proc %d\n",
570 rqstp->rq_vers, rqstp->rq_proc);
571 proc = rqstp->rq_procinfo;
572
573 /*
574 * Give the xdr decoder a chance to change this if it wants
575 * (necessary in the NFSv4.0 compound case)
576 */
577 rqstp->rq_cachetype = proc->pc_cachetype;
578 /* Decode arguments */
579 xdr = proc->pc_decode;
580 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
581 rqstp->rq_argp)) {
582 dprintk("nfsd: failed to decode arguments!\n");
583 *statp = rpc_garbage_args;
584 return 1;
585 }
586
587 /* Check whether we have this call in the cache. */
588 switch (nfsd_cache_lookup(rqstp)) {
589 case RC_INTR:
590 case RC_DROPIT:
591 return 0;
592 case RC_REPLY:
593 return 1;
594 case RC_DOIT:;
595 /* do it */
596 }
597
598 /* need to grab the location to store the status, as
599 * nfsv4 does some encoding while processing
600 */
601 nfserrp = rqstp->rq_res.head[0].iov_base
602 + rqstp->rq_res.head[0].iov_len;
603 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
604
605 /* Now call the procedure handler, and encode NFS status. */
606 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
607 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
608 if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
609 dprintk("nfsd: Dropping request; may be revisited later\n");
610 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
611 return 0;
612 }
613
614 if (rqstp->rq_proc != 0)
615 *nfserrp++ = nfserr;
616
617 /* Encode result.
618 * For NFSv2, additional info is never returned in case of an error.
619 */
620 if (!(nfserr && rqstp->rq_vers == 2)) {
621 xdr = proc->pc_encode;
622 if (xdr && !xdr(rqstp, nfserrp,
623 rqstp->rq_resp)) {
624 /* Failed to encode result. Release cache entry */
625 dprintk("nfsd: failed to encode result!\n");
626 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
627 *statp = rpc_system_err;
628 return 1;
629 }
630 }
631
632 /* Store reply in cache. */
633 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
634 return 1;
635}
636
637int nfsd_pool_stats_open(struct inode *inode, struct file *file)
638{
639 int ret;
640 mutex_lock(&nfsd_mutex);
641 if (nfsd_serv == NULL) {
642 mutex_unlock(&nfsd_mutex);
643 return -ENODEV;
644 }
645 /* bump up the psudo refcount while traversing */
646 svc_get(nfsd_serv);
647 ret = svc_pool_stats_open(nfsd_serv, file);
648 mutex_unlock(&nfsd_mutex);
649 return ret;
650}
651
652int nfsd_pool_stats_release(struct inode *inode, struct file *file)
653{
654 int ret = seq_release(inode, file);
655 mutex_lock(&nfsd_mutex);
656 /* this function really, really should have been called svc_put() */
657 svc_destroy(nfsd_serv);
658 mutex_unlock(&nfsd_mutex);
659 return ret;
660}
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}