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/sunrpc/svc_xprt.h>
18#include <linux/lockd/bind.h>
19#include <linux/nfsacl.h>
20#include <linux/seq_file.h>
21#include <linux/inetdevice.h>
22#include <net/addrconf.h>
23#include <net/ipv6.h>
24#include <net/net_namespace.h>
25#include "nfsd.h"
26#include "cache.h"
27#include "vfs.h"
28#include "netns.h"
29
30#define NFSDDBG_FACILITY NFSDDBG_SVC
31
32extern struct svc_program nfsd_program;
33static int nfsd(void *vrqstp);
34
35/*
36 * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members
37 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
38 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
39 *
40 * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a
41 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
42 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
43 * entry of ->sv_pools[].
44 *
45 * Transitions of the thread count between zero and non-zero are of particular
46 * interest since the svc_serv needs to be created and initialized at that
47 * point, or freed.
48 *
49 * Finally, the nfsd_mutex also protects some of the global variables that are
50 * accessed when nfsd starts and that are settable via the write_* routines in
51 * nfsctl.c. In particular:
52 *
53 * user_recovery_dirname
54 * user_lease_time
55 * nfsd_versions
56 */
57DEFINE_MUTEX(nfsd_mutex);
58
59/*
60 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
61 * nfsd_drc_max_pages limits the total amount of memory available for
62 * version 4.1 DRC caches.
63 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
64 */
65spinlock_t nfsd_drc_lock;
66unsigned long nfsd_drc_max_mem;
67unsigned long nfsd_drc_mem_used;
68
69#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
70static struct svc_stat nfsd_acl_svcstats;
71static struct svc_version * nfsd_acl_version[] = {
72 [2] = &nfsd_acl_version2,
73 [3] = &nfsd_acl_version3,
74};
75
76#define NFSD_ACL_MINVERS 2
77#define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
78static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
79
80static struct svc_program nfsd_acl_program = {
81 .pg_prog = NFS_ACL_PROGRAM,
82 .pg_nvers = NFSD_ACL_NRVERS,
83 .pg_vers = nfsd_acl_versions,
84 .pg_name = "nfsacl",
85 .pg_class = "nfsd",
86 .pg_stats = &nfsd_acl_svcstats,
87 .pg_authenticate = &svc_set_client,
88};
89
90static struct svc_stat nfsd_acl_svcstats = {
91 .program = &nfsd_acl_program,
92};
93#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
94
95static struct svc_version * nfsd_version[] = {
96 [2] = &nfsd_version2,
97#if defined(CONFIG_NFSD_V3)
98 [3] = &nfsd_version3,
99#endif
100#if defined(CONFIG_NFSD_V4)
101 [4] = &nfsd_version4,
102#endif
103};
104
105#define NFSD_MINVERS 2
106#define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
107static struct svc_version *nfsd_versions[NFSD_NRVERS];
108
109struct svc_program nfsd_program = {
110#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
111 .pg_next = &nfsd_acl_program,
112#endif
113 .pg_prog = NFS_PROGRAM, /* program number */
114 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
115 .pg_vers = nfsd_versions, /* version table */
116 .pg_name = "nfsd", /* program name */
117 .pg_class = "nfsd", /* authentication class */
118 .pg_stats = &nfsd_svcstats, /* version table */
119 .pg_authenticate = &svc_set_client, /* export authentication */
120
121};
122
123static bool nfsd_supported_minorversions[NFSD_SUPPORTED_MINOR_VERSION + 1] = {
124 [0] = 1,
125 [1] = 1,
126 [2] = 1,
127};
128
129int nfsd_vers(int vers, enum vers_op change)
130{
131 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
132 return 0;
133 switch(change) {
134 case NFSD_SET:
135 nfsd_versions[vers] = nfsd_version[vers];
136#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
137 if (vers < NFSD_ACL_NRVERS)
138 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
139#endif
140 break;
141 case NFSD_CLEAR:
142 nfsd_versions[vers] = NULL;
143#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
144 if (vers < NFSD_ACL_NRVERS)
145 nfsd_acl_versions[vers] = NULL;
146#endif
147 break;
148 case NFSD_TEST:
149 return nfsd_versions[vers] != NULL;
150 case NFSD_AVAIL:
151 return nfsd_version[vers] != NULL;
152 }
153 return 0;
154}
155
156int nfsd_minorversion(u32 minorversion, enum vers_op change)
157{
158 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
159 return -1;
160 switch(change) {
161 case NFSD_SET:
162 nfsd_supported_minorversions[minorversion] = true;
163 break;
164 case NFSD_CLEAR:
165 nfsd_supported_minorversions[minorversion] = false;
166 break;
167 case NFSD_TEST:
168 return nfsd_supported_minorversions[minorversion];
169 case NFSD_AVAIL:
170 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
171 }
172 return 0;
173}
174
175/*
176 * Maximum number of nfsd processes
177 */
178#define NFSD_MAXSERVS 8192
179
180int nfsd_nrthreads(struct net *net)
181{
182 int rv = 0;
183 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
184
185 mutex_lock(&nfsd_mutex);
186 if (nn->nfsd_serv)
187 rv = nn->nfsd_serv->sv_nrthreads;
188 mutex_unlock(&nfsd_mutex);
189 return rv;
190}
191
192static int nfsd_init_socks(struct net *net)
193{
194 int error;
195 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
196
197 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
198 return 0;
199
200 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
201 SVC_SOCK_DEFAULTS);
202 if (error < 0)
203 return error;
204
205 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
206 SVC_SOCK_DEFAULTS);
207 if (error < 0)
208 return error;
209
210 return 0;
211}
212
213static int nfsd_users = 0;
214
215static int nfsd_startup_generic(int nrservs)
216{
217 int ret;
218
219 if (nfsd_users++)
220 return 0;
221
222 /*
223 * Readahead param cache - will no-op if it already exists.
224 * (Note therefore results will be suboptimal if number of
225 * threads is modified after nfsd start.)
226 */
227 ret = nfsd_racache_init(2*nrservs);
228 if (ret)
229 goto dec_users;
230
231 ret = nfs4_state_start();
232 if (ret)
233 goto out_racache;
234 return 0;
235
236out_racache:
237 nfsd_racache_shutdown();
238dec_users:
239 nfsd_users--;
240 return ret;
241}
242
243static void nfsd_shutdown_generic(void)
244{
245 if (--nfsd_users)
246 return;
247
248 nfs4_state_shutdown();
249 nfsd_racache_shutdown();
250}
251
252static bool nfsd_needs_lockd(void)
253{
254#if defined(CONFIG_NFSD_V3)
255 return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
256#else
257 return (nfsd_versions[2] != NULL);
258#endif
259}
260
261static int nfsd_startup_net(int nrservs, struct net *net)
262{
263 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
264 int ret;
265
266 if (nn->nfsd_net_up)
267 return 0;
268
269 ret = nfsd_startup_generic(nrservs);
270 if (ret)
271 return ret;
272 ret = nfsd_init_socks(net);
273 if (ret)
274 goto out_socks;
275
276 if (nfsd_needs_lockd() && !nn->lockd_up) {
277 ret = lockd_up(net);
278 if (ret)
279 goto out_socks;
280 nn->lockd_up = 1;
281 }
282
283 ret = nfs4_state_start_net(net);
284 if (ret)
285 goto out_lockd;
286
287 nn->nfsd_net_up = true;
288 return 0;
289
290out_lockd:
291 if (nn->lockd_up) {
292 lockd_down(net);
293 nn->lockd_up = 0;
294 }
295out_socks:
296 nfsd_shutdown_generic();
297 return ret;
298}
299
300static void nfsd_shutdown_net(struct net *net)
301{
302 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
303
304 nfs4_state_shutdown_net(net);
305 if (nn->lockd_up) {
306 lockd_down(net);
307 nn->lockd_up = 0;
308 }
309 nn->nfsd_net_up = false;
310 nfsd_shutdown_generic();
311}
312
313static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
314 void *ptr)
315{
316 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
317 struct net_device *dev = ifa->ifa_dev->dev;
318 struct net *net = dev_net(dev);
319 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
320 struct sockaddr_in sin;
321
322 if (event != NETDEV_DOWN)
323 goto out;
324
325 if (nn->nfsd_serv) {
326 dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
327 sin.sin_family = AF_INET;
328 sin.sin_addr.s_addr = ifa->ifa_local;
329 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
330 }
331
332out:
333 return NOTIFY_DONE;
334}
335
336static struct notifier_block nfsd_inetaddr_notifier = {
337 .notifier_call = nfsd_inetaddr_event,
338};
339
340#if IS_ENABLED(CONFIG_IPV6)
341static int nfsd_inet6addr_event(struct notifier_block *this,
342 unsigned long event, void *ptr)
343{
344 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
345 struct net_device *dev = ifa->idev->dev;
346 struct net *net = dev_net(dev);
347 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
348 struct sockaddr_in6 sin6;
349
350 if (event != NETDEV_DOWN)
351 goto out;
352
353 if (nn->nfsd_serv) {
354 dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
355 sin6.sin6_family = AF_INET6;
356 sin6.sin6_addr = ifa->addr;
357 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
358 }
359
360out:
361 return NOTIFY_DONE;
362}
363
364static struct notifier_block nfsd_inet6addr_notifier = {
365 .notifier_call = nfsd_inet6addr_event,
366};
367#endif
368
369/* Only used under nfsd_mutex, so this atomic may be overkill: */
370static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
371
372static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
373{
374 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
375
376 /* check if the notifier still has clients */
377 if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
378 unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
379#if IS_ENABLED(CONFIG_IPV6)
380 unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
381#endif
382 }
383
384 /*
385 * write_ports can create the server without actually starting
386 * any threads--if we get shut down before any threads are
387 * started, then nfsd_last_thread will be run before any of this
388 * other initialization has been done except the rpcb information.
389 */
390 svc_rpcb_cleanup(serv, net);
391 if (!nn->nfsd_net_up)
392 return;
393
394 nfsd_shutdown_net(net);
395 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
396 "cache\n");
397 nfsd_export_flush(net);
398}
399
400void nfsd_reset_versions(void)
401{
402 int found_one = 0;
403 int i;
404
405 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
406 if (nfsd_program.pg_vers[i])
407 found_one = 1;
408 }
409
410 if (!found_one) {
411 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
412 nfsd_program.pg_vers[i] = nfsd_version[i];
413#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
414 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
415 nfsd_acl_program.pg_vers[i] =
416 nfsd_acl_version[i];
417#endif
418 }
419}
420
421/*
422 * Each session guarantees a negotiated per slot memory cache for replies
423 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
424 * NFSv4.1 server might want to use more memory for a DRC than a machine
425 * with mutiple services.
426 *
427 * Impose a hard limit on the number of pages for the DRC which varies
428 * according to the machines free pages. This is of course only a default.
429 *
430 * For now this is a #defined shift which could be under admin control
431 * in the future.
432 */
433static void set_max_drc(void)
434{
435 #define NFSD_DRC_SIZE_SHIFT 10
436 nfsd_drc_max_mem = (nr_free_buffer_pages()
437 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
438 nfsd_drc_mem_used = 0;
439 spin_lock_init(&nfsd_drc_lock);
440 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
441}
442
443static int nfsd_get_default_max_blksize(void)
444{
445 struct sysinfo i;
446 unsigned long long target;
447 unsigned long ret;
448
449 si_meminfo(&i);
450 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
451 /*
452 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
453 * machines, but only uses 32K on 128M machines. Bottom out at
454 * 8K on 32M and smaller. Of course, this is only a default.
455 */
456 target >>= 12;
457
458 ret = NFSSVC_MAXBLKSIZE;
459 while (ret > target && ret >= 8*1024*2)
460 ret /= 2;
461 return ret;
462}
463
464static struct svc_serv_ops nfsd_thread_sv_ops = {
465 .svo_shutdown = nfsd_last_thread,
466 .svo_function = nfsd,
467 .svo_enqueue_xprt = svc_xprt_do_enqueue,
468 .svo_setup = svc_set_num_threads,
469 .svo_module = THIS_MODULE,
470};
471
472int nfsd_create_serv(struct net *net)
473{
474 int error;
475 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
476
477 WARN_ON(!mutex_is_locked(&nfsd_mutex));
478 if (nn->nfsd_serv) {
479 svc_get(nn->nfsd_serv);
480 return 0;
481 }
482 if (nfsd_max_blksize == 0)
483 nfsd_max_blksize = nfsd_get_default_max_blksize();
484 nfsd_reset_versions();
485 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
486 &nfsd_thread_sv_ops);
487 if (nn->nfsd_serv == NULL)
488 return -ENOMEM;
489
490 nn->nfsd_serv->sv_maxconn = nn->max_connections;
491 error = svc_bind(nn->nfsd_serv, net);
492 if (error < 0) {
493 svc_destroy(nn->nfsd_serv);
494 return error;
495 }
496
497 set_max_drc();
498 /* check if the notifier is already set */
499 if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
500 register_inetaddr_notifier(&nfsd_inetaddr_notifier);
501#if IS_ENABLED(CONFIG_IPV6)
502 register_inet6addr_notifier(&nfsd_inet6addr_notifier);
503#endif
504 }
505 do_gettimeofday(&nn->nfssvc_boot); /* record boot time */
506 return 0;
507}
508
509int nfsd_nrpools(struct net *net)
510{
511 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
512
513 if (nn->nfsd_serv == NULL)
514 return 0;
515 else
516 return nn->nfsd_serv->sv_nrpools;
517}
518
519int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
520{
521 int i = 0;
522 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
523
524 if (nn->nfsd_serv != NULL) {
525 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
526 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
527 }
528
529 return 0;
530}
531
532void nfsd_destroy(struct net *net)
533{
534 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
535 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
536
537 if (destroy)
538 svc_shutdown_net(nn->nfsd_serv, net);
539 svc_destroy(nn->nfsd_serv);
540 if (destroy)
541 nn->nfsd_serv = NULL;
542}
543
544int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
545{
546 int i = 0;
547 int tot = 0;
548 int err = 0;
549 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
550
551 WARN_ON(!mutex_is_locked(&nfsd_mutex));
552
553 if (nn->nfsd_serv == NULL || n <= 0)
554 return 0;
555
556 if (n > nn->nfsd_serv->sv_nrpools)
557 n = nn->nfsd_serv->sv_nrpools;
558
559 /* enforce a global maximum number of threads */
560 tot = 0;
561 for (i = 0; i < n; i++) {
562 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
563 tot += nthreads[i];
564 }
565 if (tot > NFSD_MAXSERVS) {
566 /* total too large: scale down requested numbers */
567 for (i = 0; i < n && tot > 0; i++) {
568 int new = nthreads[i] * NFSD_MAXSERVS / tot;
569 tot -= (nthreads[i] - new);
570 nthreads[i] = new;
571 }
572 for (i = 0; i < n && tot > 0; i++) {
573 nthreads[i]--;
574 tot--;
575 }
576 }
577
578 /*
579 * There must always be a thread in pool 0; the admin
580 * can't shut down NFS completely using pool_threads.
581 */
582 if (nthreads[0] == 0)
583 nthreads[0] = 1;
584
585 /* apply the new numbers */
586 svc_get(nn->nfsd_serv);
587 for (i = 0; i < n; i++) {
588 err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
589 &nn->nfsd_serv->sv_pools[i], nthreads[i]);
590 if (err)
591 break;
592 }
593 nfsd_destroy(net);
594 return err;
595}
596
597/*
598 * Adjust the number of threads and return the new number of threads.
599 * This is also the function that starts the server if necessary, if
600 * this is the first time nrservs is nonzero.
601 */
602int
603nfsd_svc(int nrservs, struct net *net)
604{
605 int error;
606 bool nfsd_up_before;
607 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
608
609 mutex_lock(&nfsd_mutex);
610 dprintk("nfsd: creating service\n");
611
612 nrservs = max(nrservs, 0);
613 nrservs = min(nrservs, NFSD_MAXSERVS);
614 error = 0;
615
616 if (nrservs == 0 && nn->nfsd_serv == NULL)
617 goto out;
618
619 error = nfsd_create_serv(net);
620 if (error)
621 goto out;
622
623 nfsd_up_before = nn->nfsd_net_up;
624
625 error = nfsd_startup_net(nrservs, net);
626 if (error)
627 goto out_destroy;
628 error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
629 NULL, nrservs);
630 if (error)
631 goto out_shutdown;
632 /* We are holding a reference to nn->nfsd_serv which
633 * we don't want to count in the return value,
634 * so subtract 1
635 */
636 error = nn->nfsd_serv->sv_nrthreads - 1;
637out_shutdown:
638 if (error < 0 && !nfsd_up_before)
639 nfsd_shutdown_net(net);
640out_destroy:
641 nfsd_destroy(net); /* Release server */
642out:
643 mutex_unlock(&nfsd_mutex);
644 return error;
645}
646
647
648/*
649 * This is the NFS server kernel thread
650 */
651static int
652nfsd(void *vrqstp)
653{
654 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
655 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
656 struct net *net = perm_sock->xpt_net;
657 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
658 int err;
659
660 /* Lock module and set up kernel thread */
661 mutex_lock(&nfsd_mutex);
662
663 /* At this point, the thread shares current->fs
664 * with the init process. We need to create files with the
665 * umask as defined by the client instead of init's umask. */
666 if (unshare_fs_struct() < 0) {
667 printk("Unable to start nfsd thread: out of memory\n");
668 goto out;
669 }
670
671 current->fs->umask = 0;
672
673 /*
674 * thread is spawned with all signals set to SIG_IGN, re-enable
675 * the ones that will bring down the thread
676 */
677 allow_signal(SIGKILL);
678 allow_signal(SIGHUP);
679 allow_signal(SIGINT);
680 allow_signal(SIGQUIT);
681
682 nfsdstats.th_cnt++;
683 mutex_unlock(&nfsd_mutex);
684
685 set_freezable();
686
687 /*
688 * The main request loop
689 */
690 for (;;) {
691 /* Update sv_maxconn if it has changed */
692 rqstp->rq_server->sv_maxconn = nn->max_connections;
693
694 /*
695 * Find a socket with data available and call its
696 * recvfrom routine.
697 */
698 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
699 ;
700 if (err == -EINTR)
701 break;
702 validate_process_creds();
703 svc_process(rqstp);
704 validate_process_creds();
705 }
706
707 /* Clear signals before calling svc_exit_thread() */
708 flush_signals(current);
709
710 mutex_lock(&nfsd_mutex);
711 nfsdstats.th_cnt --;
712
713out:
714 rqstp->rq_server = NULL;
715
716 /* Release the thread */
717 svc_exit_thread(rqstp);
718
719 nfsd_destroy(net);
720
721 /* Release module */
722 mutex_unlock(&nfsd_mutex);
723 module_put_and_exit(0);
724 return 0;
725}
726
727static __be32 map_new_errors(u32 vers, __be32 nfserr)
728{
729 if (nfserr == nfserr_jukebox && vers == 2)
730 return nfserr_dropit;
731 if (nfserr == nfserr_wrongsec && vers < 4)
732 return nfserr_acces;
733 return nfserr;
734}
735
736int
737nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
738{
739 struct svc_procedure *proc;
740 kxdrproc_t xdr;
741 __be32 nfserr;
742 __be32 *nfserrp;
743
744 dprintk("nfsd_dispatch: vers %d proc %d\n",
745 rqstp->rq_vers, rqstp->rq_proc);
746 proc = rqstp->rq_procinfo;
747
748 /*
749 * Give the xdr decoder a chance to change this if it wants
750 * (necessary in the NFSv4.0 compound case)
751 */
752 rqstp->rq_cachetype = proc->pc_cachetype;
753 /* Decode arguments */
754 xdr = proc->pc_decode;
755 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
756 rqstp->rq_argp)) {
757 dprintk("nfsd: failed to decode arguments!\n");
758 *statp = rpc_garbage_args;
759 return 1;
760 }
761
762 /* Check whether we have this call in the cache. */
763 switch (nfsd_cache_lookup(rqstp)) {
764 case RC_DROPIT:
765 return 0;
766 case RC_REPLY:
767 return 1;
768 case RC_DOIT:;
769 /* do it */
770 }
771
772 /* need to grab the location to store the status, as
773 * nfsv4 does some encoding while processing
774 */
775 nfserrp = rqstp->rq_res.head[0].iov_base
776 + rqstp->rq_res.head[0].iov_len;
777 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
778
779 /* Now call the procedure handler, and encode NFS status. */
780 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
781 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
782 if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
783 dprintk("nfsd: Dropping request; may be revisited later\n");
784 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
785 return 0;
786 }
787
788 if (rqstp->rq_proc != 0)
789 *nfserrp++ = nfserr;
790
791 /* Encode result.
792 * For NFSv2, additional info is never returned in case of an error.
793 */
794 if (!(nfserr && rqstp->rq_vers == 2)) {
795 xdr = proc->pc_encode;
796 if (xdr && !xdr(rqstp, nfserrp,
797 rqstp->rq_resp)) {
798 /* Failed to encode result. Release cache entry */
799 dprintk("nfsd: failed to encode result!\n");
800 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
801 *statp = rpc_system_err;
802 return 1;
803 }
804 }
805
806 /* Store reply in cache. */
807 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
808 return 1;
809}
810
811int nfsd_pool_stats_open(struct inode *inode, struct file *file)
812{
813 int ret;
814 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
815
816 mutex_lock(&nfsd_mutex);
817 if (nn->nfsd_serv == NULL) {
818 mutex_unlock(&nfsd_mutex);
819 return -ENODEV;
820 }
821 /* bump up the psudo refcount while traversing */
822 svc_get(nn->nfsd_serv);
823 ret = svc_pool_stats_open(nn->nfsd_serv, file);
824 mutex_unlock(&nfsd_mutex);
825 return ret;
826}
827
828int nfsd_pool_stats_release(struct inode *inode, struct file *file)
829{
830 int ret = seq_release(inode, file);
831 struct net *net = inode->i_sb->s_fs_info;
832
833 mutex_lock(&nfsd_mutex);
834 /* this function really, really should have been called svc_put() */
835 nfsd_destroy(net);
836 mutex_unlock(&nfsd_mutex);
837 return ret;
838}
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#include <linux/siphash.h>
16
17#include <linux/sunrpc/stats.h>
18#include <linux/sunrpc/svcsock.h>
19#include <linux/sunrpc/svc_xprt.h>
20#include <linux/lockd/bind.h>
21#include <linux/nfsacl.h>
22#include <linux/seq_file.h>
23#include <linux/inetdevice.h>
24#include <net/addrconf.h>
25#include <net/ipv6.h>
26#include <net/net_namespace.h>
27#include "nfsd.h"
28#include "cache.h"
29#include "vfs.h"
30#include "netns.h"
31#include "filecache.h"
32
33#include "trace.h"
34
35#define NFSDDBG_FACILITY NFSDDBG_SVC
36
37extern struct svc_program nfsd_program;
38static int nfsd(void *vrqstp);
39#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
40static int nfsd_acl_rpcbind_set(struct net *,
41 const struct svc_program *,
42 u32, int,
43 unsigned short,
44 unsigned short);
45static __be32 nfsd_acl_init_request(struct svc_rqst *,
46 const struct svc_program *,
47 struct svc_process_info *);
48#endif
49static int nfsd_rpcbind_set(struct net *,
50 const struct svc_program *,
51 u32, int,
52 unsigned short,
53 unsigned short);
54static __be32 nfsd_init_request(struct svc_rqst *,
55 const struct svc_program *,
56 struct svc_process_info *);
57
58/*
59 * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and some members
60 * of the svc_serv struct such as ->sv_temp_socks and ->sv_permsocks.
61 *
62 * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a
63 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0 (unless
64 * nn->keep_active is set). That number of nfsd threads must
65 * exist and each must be listed in ->sp_all_threads in some entry of
66 * ->sv_pools[].
67 *
68 * Each active thread holds a counted reference on nn->nfsd_serv, as does
69 * the nn->keep_active flag and various transient calls to svc_get().
70 *
71 * Finally, the nfsd_mutex also protects some of the global variables that are
72 * accessed when nfsd starts and that are settable via the write_* routines in
73 * nfsctl.c. In particular:
74 *
75 * user_recovery_dirname
76 * user_lease_time
77 * nfsd_versions
78 */
79DEFINE_MUTEX(nfsd_mutex);
80
81/*
82 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
83 * nfsd_drc_max_pages limits the total amount of memory available for
84 * version 4.1 DRC caches.
85 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
86 */
87DEFINE_SPINLOCK(nfsd_drc_lock);
88unsigned long nfsd_drc_max_mem;
89unsigned long nfsd_drc_mem_used;
90
91#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
92static struct svc_stat nfsd_acl_svcstats;
93static const struct svc_version *nfsd_acl_version[] = {
94# if defined(CONFIG_NFSD_V2_ACL)
95 [2] = &nfsd_acl_version2,
96# endif
97# if defined(CONFIG_NFSD_V3_ACL)
98 [3] = &nfsd_acl_version3,
99# endif
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#if defined(CONFIG_NFSD_V2)
124 [2] = &nfsd_version2,
125#endif
126 [3] = &nfsd_version3,
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_xprt_create(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_xprt_create(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(void)
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
352/**
353 * nfsd_copy_write_verifier - Atomically copy a write verifier
354 * @verf: buffer in which to receive the verifier cookie
355 * @nn: NFS net namespace
356 *
357 * This function provides a wait-free mechanism for copying the
358 * namespace's write verifier without tearing it.
359 */
360void nfsd_copy_write_verifier(__be32 verf[2], struct nfsd_net *nn)
361{
362 int seq = 0;
363
364 do {
365 read_seqbegin_or_lock(&nn->writeverf_lock, &seq);
366 memcpy(verf, nn->writeverf, sizeof(*verf));
367 } while (need_seqretry(&nn->writeverf_lock, seq));
368 done_seqretry(&nn->writeverf_lock, seq);
369}
370
371static void nfsd_reset_write_verifier_locked(struct nfsd_net *nn)
372{
373 struct timespec64 now;
374 u64 verf;
375
376 /*
377 * Because the time value is hashed, y2038 time_t overflow
378 * is irrelevant in this usage.
379 */
380 ktime_get_raw_ts64(&now);
381 verf = siphash_2u64(now.tv_sec, now.tv_nsec, &nn->siphash_key);
382 memcpy(nn->writeverf, &verf, sizeof(nn->writeverf));
383}
384
385/**
386 * nfsd_reset_write_verifier - Generate a new write verifier
387 * @nn: NFS net namespace
388 *
389 * This function updates the ->writeverf field of @nn. This field
390 * contains an opaque cookie that, according to Section 18.32.3 of
391 * RFC 8881, "the client can use to determine whether a server has
392 * changed instance state (e.g., server restart) between a call to
393 * WRITE and a subsequent call to either WRITE or COMMIT. This
394 * cookie MUST be unchanged during a single instance of the NFSv4.1
395 * server and MUST be unique between instances of the NFSv4.1
396 * server."
397 */
398void nfsd_reset_write_verifier(struct nfsd_net *nn)
399{
400 write_seqlock(&nn->writeverf_lock);
401 nfsd_reset_write_verifier_locked(nn);
402 write_sequnlock(&nn->writeverf_lock);
403}
404
405static int nfsd_startup_net(struct net *net, const struct cred *cred)
406{
407 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
408 int ret;
409
410 if (nn->nfsd_net_up)
411 return 0;
412
413 ret = nfsd_startup_generic();
414 if (ret)
415 return ret;
416 ret = nfsd_init_socks(net, cred);
417 if (ret)
418 goto out_socks;
419
420 if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
421 ret = lockd_up(net, cred);
422 if (ret)
423 goto out_socks;
424 nn->lockd_up = true;
425 }
426
427 ret = nfsd_file_cache_start_net(net);
428 if (ret)
429 goto out_lockd;
430 ret = nfs4_state_start_net(net);
431 if (ret)
432 goto out_filecache;
433
434#ifdef CONFIG_NFSD_V4_2_INTER_SSC
435 nfsd4_ssc_init_umount_work(nn);
436#endif
437 nn->nfsd_net_up = true;
438 return 0;
439
440out_filecache:
441 nfsd_file_cache_shutdown_net(net);
442out_lockd:
443 if (nn->lockd_up) {
444 lockd_down(net);
445 nn->lockd_up = false;
446 }
447out_socks:
448 nfsd_shutdown_generic();
449 return ret;
450}
451
452static void nfsd_shutdown_net(struct net *net)
453{
454 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
455
456 nfs4_state_shutdown_net(net);
457 nfsd_file_cache_shutdown_net(net);
458 if (nn->lockd_up) {
459 lockd_down(net);
460 nn->lockd_up = false;
461 }
462 nn->nfsd_net_up = false;
463 nfsd_shutdown_generic();
464}
465
466static DEFINE_SPINLOCK(nfsd_notifier_lock);
467static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
468 void *ptr)
469{
470 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
471 struct net_device *dev = ifa->ifa_dev->dev;
472 struct net *net = dev_net(dev);
473 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
474 struct sockaddr_in sin;
475
476 if (event != NETDEV_DOWN || !nn->nfsd_serv)
477 goto out;
478
479 spin_lock(&nfsd_notifier_lock);
480 if (nn->nfsd_serv) {
481 dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
482 sin.sin_family = AF_INET;
483 sin.sin_addr.s_addr = ifa->ifa_local;
484 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
485 }
486 spin_unlock(&nfsd_notifier_lock);
487
488out:
489 return NOTIFY_DONE;
490}
491
492static struct notifier_block nfsd_inetaddr_notifier = {
493 .notifier_call = nfsd_inetaddr_event,
494};
495
496#if IS_ENABLED(CONFIG_IPV6)
497static int nfsd_inet6addr_event(struct notifier_block *this,
498 unsigned long event, void *ptr)
499{
500 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
501 struct net_device *dev = ifa->idev->dev;
502 struct net *net = dev_net(dev);
503 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
504 struct sockaddr_in6 sin6;
505
506 if (event != NETDEV_DOWN || !nn->nfsd_serv)
507 goto out;
508
509 spin_lock(&nfsd_notifier_lock);
510 if (nn->nfsd_serv) {
511 dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
512 sin6.sin6_family = AF_INET6;
513 sin6.sin6_addr = ifa->addr;
514 if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
515 sin6.sin6_scope_id = ifa->idev->dev->ifindex;
516 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
517 }
518 spin_unlock(&nfsd_notifier_lock);
519
520out:
521 return NOTIFY_DONE;
522}
523
524static struct notifier_block nfsd_inet6addr_notifier = {
525 .notifier_call = nfsd_inet6addr_event,
526};
527#endif
528
529/* Only used under nfsd_mutex, so this atomic may be overkill: */
530static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
531
532static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
533{
534 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
535
536 /* check if the notifier still has clients */
537 if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
538 unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
539#if IS_ENABLED(CONFIG_IPV6)
540 unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
541#endif
542 }
543
544 /*
545 * write_ports can create the server without actually starting
546 * any threads--if we get shut down before any threads are
547 * started, then nfsd_last_thread will be run before any of this
548 * other initialization has been done except the rpcb information.
549 */
550 svc_rpcb_cleanup(serv, net);
551 if (!nn->nfsd_net_up)
552 return;
553
554 nfsd_shutdown_net(net);
555 pr_info("nfsd: last server has exited, flushing export cache\n");
556 nfsd_export_flush(net);
557}
558
559void nfsd_reset_versions(struct nfsd_net *nn)
560{
561 int i;
562
563 for (i = 0; i < NFSD_NRVERS; i++)
564 if (nfsd_vers(nn, i, NFSD_TEST))
565 return;
566
567 for (i = 0; i < NFSD_NRVERS; i++)
568 if (i != 4)
569 nfsd_vers(nn, i, NFSD_SET);
570 else {
571 int minor = 0;
572 while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
573 minor++;
574 }
575}
576
577/*
578 * Each session guarantees a negotiated per slot memory cache for replies
579 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
580 * NFSv4.1 server might want to use more memory for a DRC than a machine
581 * with mutiple services.
582 *
583 * Impose a hard limit on the number of pages for the DRC which varies
584 * according to the machines free pages. This is of course only a default.
585 *
586 * For now this is a #defined shift which could be under admin control
587 * in the future.
588 */
589static void set_max_drc(void)
590{
591 #define NFSD_DRC_SIZE_SHIFT 7
592 nfsd_drc_max_mem = (nr_free_buffer_pages()
593 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
594 nfsd_drc_mem_used = 0;
595 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
596}
597
598static int nfsd_get_default_max_blksize(void)
599{
600 struct sysinfo i;
601 unsigned long long target;
602 unsigned long ret;
603
604 si_meminfo(&i);
605 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
606 /*
607 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
608 * machines, but only uses 32K on 128M machines. Bottom out at
609 * 8K on 32M and smaller. Of course, this is only a default.
610 */
611 target >>= 12;
612
613 ret = NFSSVC_MAXBLKSIZE;
614 while (ret > target && ret >= 8*1024*2)
615 ret /= 2;
616 return ret;
617}
618
619void nfsd_shutdown_threads(struct net *net)
620{
621 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
622 struct svc_serv *serv;
623
624 mutex_lock(&nfsd_mutex);
625 serv = nn->nfsd_serv;
626 if (serv == NULL) {
627 mutex_unlock(&nfsd_mutex);
628 return;
629 }
630
631 svc_get(serv);
632 /* Kill outstanding nfsd threads */
633 svc_set_num_threads(serv, NULL, 0);
634 nfsd_put(net);
635 mutex_unlock(&nfsd_mutex);
636}
637
638bool i_am_nfsd(void)
639{
640 return kthread_func(current) == nfsd;
641}
642
643int nfsd_create_serv(struct net *net)
644{
645 int error;
646 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
647 struct svc_serv *serv;
648
649 WARN_ON(!mutex_is_locked(&nfsd_mutex));
650 if (nn->nfsd_serv) {
651 svc_get(nn->nfsd_serv);
652 return 0;
653 }
654 if (nfsd_max_blksize == 0)
655 nfsd_max_blksize = nfsd_get_default_max_blksize();
656 nfsd_reset_versions(nn);
657 serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize, nfsd);
658 if (serv == NULL)
659 return -ENOMEM;
660
661 serv->sv_maxconn = nn->max_connections;
662 error = svc_bind(serv, net);
663 if (error < 0) {
664 /* NOT nfsd_put() as notifiers (see below) haven't
665 * been set up yet.
666 */
667 svc_put(serv);
668 return error;
669 }
670 spin_lock(&nfsd_notifier_lock);
671 nn->nfsd_serv = serv;
672 spin_unlock(&nfsd_notifier_lock);
673
674 set_max_drc();
675 /* check if the notifier is already set */
676 if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
677 register_inetaddr_notifier(&nfsd_inetaddr_notifier);
678#if IS_ENABLED(CONFIG_IPV6)
679 register_inet6addr_notifier(&nfsd_inet6addr_notifier);
680#endif
681 }
682 nfsd_reset_write_verifier(nn);
683 return 0;
684}
685
686int nfsd_nrpools(struct net *net)
687{
688 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
689
690 if (nn->nfsd_serv == NULL)
691 return 0;
692 else
693 return nn->nfsd_serv->sv_nrpools;
694}
695
696int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
697{
698 int i = 0;
699 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
700
701 if (nn->nfsd_serv != NULL) {
702 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
703 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
704 }
705
706 return 0;
707}
708
709/* This is the callback for kref_put() below.
710 * There is no code here as the first thing to be done is
711 * call svc_shutdown_net(), but we cannot get the 'net' from
712 * the kref. So do all the work when kref_put returns true.
713 */
714static void nfsd_noop(struct kref *ref)
715{
716}
717
718void nfsd_put(struct net *net)
719{
720 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
721
722 if (kref_put(&nn->nfsd_serv->sv_refcnt, nfsd_noop)) {
723 svc_xprt_destroy_all(nn->nfsd_serv, net);
724 nfsd_last_thread(nn->nfsd_serv, net);
725 svc_destroy(&nn->nfsd_serv->sv_refcnt);
726 spin_lock(&nfsd_notifier_lock);
727 nn->nfsd_serv = NULL;
728 spin_unlock(&nfsd_notifier_lock);
729 }
730}
731
732int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
733{
734 int i = 0;
735 int tot = 0;
736 int err = 0;
737 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
738
739 WARN_ON(!mutex_is_locked(&nfsd_mutex));
740
741 if (nn->nfsd_serv == NULL || n <= 0)
742 return 0;
743
744 if (n > nn->nfsd_serv->sv_nrpools)
745 n = nn->nfsd_serv->sv_nrpools;
746
747 /* enforce a global maximum number of threads */
748 tot = 0;
749 for (i = 0; i < n; i++) {
750 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
751 tot += nthreads[i];
752 }
753 if (tot > NFSD_MAXSERVS) {
754 /* total too large: scale down requested numbers */
755 for (i = 0; i < n && tot > 0; i++) {
756 int new = nthreads[i] * NFSD_MAXSERVS / tot;
757 tot -= (nthreads[i] - new);
758 nthreads[i] = new;
759 }
760 for (i = 0; i < n && tot > 0; i++) {
761 nthreads[i]--;
762 tot--;
763 }
764 }
765
766 /*
767 * There must always be a thread in pool 0; the admin
768 * can't shut down NFS completely using pool_threads.
769 */
770 if (nthreads[0] == 0)
771 nthreads[0] = 1;
772
773 /* apply the new numbers */
774 svc_get(nn->nfsd_serv);
775 for (i = 0; i < n; i++) {
776 err = svc_set_num_threads(nn->nfsd_serv,
777 &nn->nfsd_serv->sv_pools[i],
778 nthreads[i]);
779 if (err)
780 break;
781 }
782 nfsd_put(net);
783 return err;
784}
785
786/*
787 * Adjust the number of threads and return the new number of threads.
788 * This is also the function that starts the server if necessary, if
789 * this is the first time nrservs is nonzero.
790 */
791int
792nfsd_svc(int nrservs, struct net *net, const struct cred *cred)
793{
794 int error;
795 bool nfsd_up_before;
796 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
797
798 mutex_lock(&nfsd_mutex);
799 dprintk("nfsd: creating service\n");
800
801 nrservs = max(nrservs, 0);
802 nrservs = min(nrservs, NFSD_MAXSERVS);
803 error = 0;
804
805 if (nrservs == 0 && nn->nfsd_serv == NULL)
806 goto out;
807
808 strscpy(nn->nfsd_name, utsname()->nodename,
809 sizeof(nn->nfsd_name));
810
811 error = nfsd_create_serv(net);
812 if (error)
813 goto out;
814
815 nfsd_up_before = nn->nfsd_net_up;
816
817 error = nfsd_startup_net(net, cred);
818 if (error)
819 goto out_put;
820 error = svc_set_num_threads(nn->nfsd_serv, NULL, nrservs);
821 if (error)
822 goto out_shutdown;
823 error = nn->nfsd_serv->sv_nrthreads;
824out_shutdown:
825 if (error < 0 && !nfsd_up_before)
826 nfsd_shutdown_net(net);
827out_put:
828 /* Threads now hold service active */
829 if (xchg(&nn->keep_active, 0))
830 nfsd_put(net);
831 nfsd_put(net);
832out:
833 mutex_unlock(&nfsd_mutex);
834 return error;
835}
836
837#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
838static bool
839nfsd_support_acl_version(int vers)
840{
841 if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
842 return nfsd_acl_version[vers] != NULL;
843 return false;
844}
845
846static int
847nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
848 u32 version, int family, unsigned short proto,
849 unsigned short port)
850{
851 if (!nfsd_support_acl_version(version) ||
852 !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
853 return 0;
854 return svc_generic_rpcbind_set(net, progp, version, family,
855 proto, port);
856}
857
858static __be32
859nfsd_acl_init_request(struct svc_rqst *rqstp,
860 const struct svc_program *progp,
861 struct svc_process_info *ret)
862{
863 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
864 int i;
865
866 if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
867 nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
868 return svc_generic_init_request(rqstp, progp, ret);
869
870 ret->mismatch.lovers = NFSD_ACL_NRVERS;
871 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
872 if (nfsd_support_acl_version(rqstp->rq_vers) &&
873 nfsd_vers(nn, i, NFSD_TEST)) {
874 ret->mismatch.lovers = i;
875 break;
876 }
877 }
878 if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
879 return rpc_prog_unavail;
880 ret->mismatch.hivers = NFSD_ACL_MINVERS;
881 for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
882 if (nfsd_support_acl_version(rqstp->rq_vers) &&
883 nfsd_vers(nn, i, NFSD_TEST)) {
884 ret->mismatch.hivers = i;
885 break;
886 }
887 }
888 return rpc_prog_mismatch;
889}
890#endif
891
892static int
893nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
894 u32 version, int family, unsigned short proto,
895 unsigned short port)
896{
897 if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
898 return 0;
899 return svc_generic_rpcbind_set(net, progp, version, family,
900 proto, port);
901}
902
903static __be32
904nfsd_init_request(struct svc_rqst *rqstp,
905 const struct svc_program *progp,
906 struct svc_process_info *ret)
907{
908 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
909 int i;
910
911 if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
912 return svc_generic_init_request(rqstp, progp, ret);
913
914 ret->mismatch.lovers = NFSD_NRVERS;
915 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
916 if (nfsd_vers(nn, i, NFSD_TEST)) {
917 ret->mismatch.lovers = i;
918 break;
919 }
920 }
921 if (ret->mismatch.lovers == NFSD_NRVERS)
922 return rpc_prog_unavail;
923 ret->mismatch.hivers = NFSD_MINVERS;
924 for (i = NFSD_NRVERS - 1; i >= NFSD_MINVERS; i--) {
925 if (nfsd_vers(nn, i, NFSD_TEST)) {
926 ret->mismatch.hivers = i;
927 break;
928 }
929 }
930 return rpc_prog_mismatch;
931}
932
933/*
934 * This is the NFS server kernel thread
935 */
936static int
937nfsd(void *vrqstp)
938{
939 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
940 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
941 struct net *net = perm_sock->xpt_net;
942 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
943 int err;
944
945 /* At this point, the thread shares current->fs
946 * with the init process. We need to create files with the
947 * umask as defined by the client instead of init's umask. */
948 if (unshare_fs_struct() < 0) {
949 printk("Unable to start nfsd thread: out of memory\n");
950 goto out;
951 }
952
953 current->fs->umask = 0;
954
955 /*
956 * thread is spawned with all signals set to SIG_IGN, re-enable
957 * the ones that will bring down the thread
958 */
959 allow_signal(SIGKILL);
960 allow_signal(SIGHUP);
961 allow_signal(SIGINT);
962 allow_signal(SIGQUIT);
963
964 atomic_inc(&nfsdstats.th_cnt);
965
966 set_freezable();
967
968 /*
969 * The main request loop
970 */
971 for (;;) {
972 /* Update sv_maxconn if it has changed */
973 rqstp->rq_server->sv_maxconn = nn->max_connections;
974
975 /*
976 * Find a socket with data available and call its
977 * recvfrom routine.
978 */
979 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
980 ;
981 if (err == -EINTR)
982 break;
983 validate_process_creds();
984 svc_process(rqstp);
985 validate_process_creds();
986 }
987
988 /* Clear signals before calling svc_exit_thread() */
989 flush_signals(current);
990
991 atomic_dec(&nfsdstats.th_cnt);
992
993out:
994 /* Take an extra ref so that the svc_put in svc_exit_thread()
995 * doesn't call svc_destroy()
996 */
997 svc_get(nn->nfsd_serv);
998
999 /* Release the thread */
1000 svc_exit_thread(rqstp);
1001
1002 /* We need to drop a ref, but may not drop the last reference
1003 * without holding nfsd_mutex, and we cannot wait for nfsd_mutex as that
1004 * could deadlock with nfsd_shutdown_threads() waiting for us.
1005 * So three options are:
1006 * - drop a non-final reference,
1007 * - get the mutex without waiting
1008 * - sleep briefly andd try the above again
1009 */
1010 while (!svc_put_not_last(nn->nfsd_serv)) {
1011 if (mutex_trylock(&nfsd_mutex)) {
1012 nfsd_put(net);
1013 mutex_unlock(&nfsd_mutex);
1014 break;
1015 }
1016 msleep(20);
1017 }
1018
1019 return 0;
1020}
1021
1022/**
1023 * nfsd_dispatch - Process an NFS or NFSACL Request
1024 * @rqstp: incoming request
1025 * @statp: pointer to location of accept_stat field in RPC Reply buffer
1026 *
1027 * This RPC dispatcher integrates the NFS server's duplicate reply cache.
1028 *
1029 * Return values:
1030 * %0: Processing complete; do not send a Reply
1031 * %1: Processing complete; send Reply in rqstp->rq_res
1032 */
1033int nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1034{
1035 const struct svc_procedure *proc = rqstp->rq_procinfo;
1036
1037 /*
1038 * Give the xdr decoder a chance to change this if it wants
1039 * (necessary in the NFSv4.0 compound case)
1040 */
1041 rqstp->rq_cachetype = proc->pc_cachetype;
1042
1043 svcxdr_init_decode(rqstp);
1044 if (!proc->pc_decode(rqstp, &rqstp->rq_arg_stream))
1045 goto out_decode_err;
1046
1047 switch (nfsd_cache_lookup(rqstp)) {
1048 case RC_DOIT:
1049 break;
1050 case RC_REPLY:
1051 goto out_cached_reply;
1052 case RC_DROPIT:
1053 goto out_dropit;
1054 }
1055
1056 /*
1057 * Need to grab the location to store the status, as
1058 * NFSv4 does some encoding while processing
1059 */
1060 svcxdr_init_encode(rqstp);
1061
1062 *statp = proc->pc_func(rqstp);
1063 if (test_bit(RQ_DROPME, &rqstp->rq_flags))
1064 goto out_update_drop;
1065
1066 if (!proc->pc_encode(rqstp, &rqstp->rq_res_stream))
1067 goto out_encode_err;
1068
1069 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
1070out_cached_reply:
1071 return 1;
1072
1073out_decode_err:
1074 trace_nfsd_garbage_args_err(rqstp);
1075 *statp = rpc_garbage_args;
1076 return 1;
1077
1078out_update_drop:
1079 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1080out_dropit:
1081 return 0;
1082
1083out_encode_err:
1084 trace_nfsd_cant_encode_err(rqstp);
1085 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1086 *statp = rpc_system_err;
1087 return 1;
1088}
1089
1090/**
1091 * nfssvc_decode_voidarg - Decode void arguments
1092 * @rqstp: Server RPC transaction context
1093 * @xdr: XDR stream positioned at arguments to decode
1094 *
1095 * Return values:
1096 * %false: Arguments were not valid
1097 * %true: Decoding was successful
1098 */
1099bool nfssvc_decode_voidarg(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1100{
1101 return true;
1102}
1103
1104/**
1105 * nfssvc_encode_voidres - Encode void results
1106 * @rqstp: Server RPC transaction context
1107 * @xdr: XDR stream into which to encode results
1108 *
1109 * Return values:
1110 * %false: Local error while encoding
1111 * %true: Encoding was successful
1112 */
1113bool nfssvc_encode_voidres(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1114{
1115 return true;
1116}
1117
1118int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1119{
1120 int ret;
1121 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1122
1123 mutex_lock(&nfsd_mutex);
1124 if (nn->nfsd_serv == NULL) {
1125 mutex_unlock(&nfsd_mutex);
1126 return -ENODEV;
1127 }
1128 svc_get(nn->nfsd_serv);
1129 ret = svc_pool_stats_open(nn->nfsd_serv, file);
1130 mutex_unlock(&nfsd_mutex);
1131 return ret;
1132}
1133
1134int nfsd_pool_stats_release(struct inode *inode, struct file *file)
1135{
1136 int ret = seq_release(inode, file);
1137 struct net *net = inode->i_sb->s_fs_info;
1138
1139 mutex_lock(&nfsd_mutex);
1140 nfsd_put(net);
1141 mutex_unlock(&nfsd_mutex);
1142 return ret;
1143}