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