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