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#include <linux/nsproxy.h>
15
16#include <linux/sunrpc/stats.h>
17#include <linux/sunrpc/svcsock.h>
18#include <linux/lockd/bind.h>
19#include <linux/nfsacl.h>
20#include <linux/seq_file.h>
21#include <net/net_namespace.h>
22#include "nfsd.h"
23#include "cache.h"
24#include "vfs.h"
25
26#define NFSDDBG_FACILITY NFSDDBG_SVC
27
28extern struct svc_program nfsd_program;
29static int nfsd(void *vrqstp);
30struct timeval nfssvc_boot;
31
32/*
33 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
34 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
35 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
36 *
37 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
38 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
39 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
40 * entry of ->sv_pools[].
41 *
42 * Transitions of the thread count between zero and non-zero are of particular
43 * interest since the svc_serv needs to be created and initialized at that
44 * point, or freed.
45 *
46 * Finally, the nfsd_mutex also protects some of the global variables that are
47 * accessed when nfsd starts and that are settable via the write_* routines in
48 * nfsctl.c. In particular:
49 *
50 * user_recovery_dirname
51 * user_lease_time
52 * nfsd_versions
53 */
54DEFINE_MUTEX(nfsd_mutex);
55struct svc_serv *nfsd_serv;
56
57/*
58 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
59 * nfsd_drc_max_pages limits the total amount of memory available for
60 * version 4.1 DRC caches.
61 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
62 */
63spinlock_t nfsd_drc_lock;
64unsigned int nfsd_drc_max_mem;
65unsigned int nfsd_drc_mem_used;
66
67#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
68static struct svc_stat nfsd_acl_svcstats;
69static struct svc_version * nfsd_acl_version[] = {
70 [2] = &nfsd_acl_version2,
71 [3] = &nfsd_acl_version3,
72};
73
74#define NFSD_ACL_MINVERS 2
75#define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
76static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
77
78static struct svc_program nfsd_acl_program = {
79 .pg_prog = NFS_ACL_PROGRAM,
80 .pg_nvers = NFSD_ACL_NRVERS,
81 .pg_vers = nfsd_acl_versions,
82 .pg_name = "nfsacl",
83 .pg_class = "nfsd",
84 .pg_stats = &nfsd_acl_svcstats,
85 .pg_authenticate = &svc_set_client,
86};
87
88static struct svc_stat nfsd_acl_svcstats = {
89 .program = &nfsd_acl_program,
90};
91#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
92
93static struct svc_version * nfsd_version[] = {
94 [2] = &nfsd_version2,
95#if defined(CONFIG_NFSD_V3)
96 [3] = &nfsd_version3,
97#endif
98#if defined(CONFIG_NFSD_V4)
99 [4] = &nfsd_version4,
100#endif
101};
102
103#define NFSD_MINVERS 2
104#define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
105static struct svc_version *nfsd_versions[NFSD_NRVERS];
106
107struct svc_program nfsd_program = {
108#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
109 .pg_next = &nfsd_acl_program,
110#endif
111 .pg_prog = NFS_PROGRAM, /* program number */
112 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
113 .pg_vers = nfsd_versions, /* version table */
114 .pg_name = "nfsd", /* program name */
115 .pg_class = "nfsd", /* authentication class */
116 .pg_stats = &nfsd_svcstats, /* version table */
117 .pg_authenticate = &svc_set_client, /* export authentication */
118
119};
120
121u32 nfsd_supported_minorversion;
122
123int nfsd_vers(int vers, enum vers_op change)
124{
125 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
126 return 0;
127 switch(change) {
128 case NFSD_SET:
129 nfsd_versions[vers] = nfsd_version[vers];
130#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
131 if (vers < NFSD_ACL_NRVERS)
132 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
133#endif
134 break;
135 case NFSD_CLEAR:
136 nfsd_versions[vers] = NULL;
137#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
138 if (vers < NFSD_ACL_NRVERS)
139 nfsd_acl_versions[vers] = NULL;
140#endif
141 break;
142 case NFSD_TEST:
143 return nfsd_versions[vers] != NULL;
144 case NFSD_AVAIL:
145 return nfsd_version[vers] != NULL;
146 }
147 return 0;
148}
149
150int nfsd_minorversion(u32 minorversion, enum vers_op change)
151{
152 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
153 return -1;
154 switch(change) {
155 case NFSD_SET:
156 nfsd_supported_minorversion = minorversion;
157 break;
158 case NFSD_CLEAR:
159 if (minorversion == 0)
160 return -1;
161 nfsd_supported_minorversion = minorversion - 1;
162 break;
163 case NFSD_TEST:
164 return minorversion <= nfsd_supported_minorversion;
165 case NFSD_AVAIL:
166 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
167 }
168 return 0;
169}
170
171/*
172 * Maximum number of nfsd processes
173 */
174#define NFSD_MAXSERVS 8192
175
176int nfsd_nrthreads(void)
177{
178 int rv = 0;
179 mutex_lock(&nfsd_mutex);
180 if (nfsd_serv)
181 rv = nfsd_serv->sv_nrthreads;
182 mutex_unlock(&nfsd_mutex);
183 return rv;
184}
185
186static int nfsd_init_socks(int port)
187{
188 int error;
189 if (!list_empty(&nfsd_serv->sv_permsocks))
190 return 0;
191
192 error = svc_create_xprt(nfsd_serv, "udp", &init_net, PF_INET, port,
193 SVC_SOCK_DEFAULTS);
194 if (error < 0)
195 return error;
196
197 error = svc_create_xprt(nfsd_serv, "tcp", &init_net, PF_INET, port,
198 SVC_SOCK_DEFAULTS);
199 if (error < 0)
200 return error;
201
202 return 0;
203}
204
205static bool nfsd_up = false;
206
207static int nfsd_startup(unsigned short port, int nrservs)
208{
209 int ret;
210
211 if (nfsd_up)
212 return 0;
213 /*
214 * Readahead param cache - will no-op if it already exists.
215 * (Note therefore results will be suboptimal if number of
216 * threads is modified after nfsd start.)
217 */
218 ret = nfsd_racache_init(2*nrservs);
219 if (ret)
220 return ret;
221 ret = nfsd_init_socks(port);
222 if (ret)
223 goto out_racache;
224 ret = lockd_up(&init_net);
225 if (ret)
226 goto out_racache;
227 ret = nfs4_state_start();
228 if (ret)
229 goto out_lockd;
230 nfsd_up = true;
231 return 0;
232out_lockd:
233 lockd_down(&init_net);
234out_racache:
235 nfsd_racache_shutdown();
236 return ret;
237}
238
239static void nfsd_shutdown(void)
240{
241 /*
242 * write_ports can create the server without actually starting
243 * any threads--if we get shut down before any threads are
244 * started, then nfsd_last_thread will be run before any of this
245 * other initialization has been done.
246 */
247 if (!nfsd_up)
248 return;
249 nfs4_state_shutdown();
250 lockd_down(&init_net);
251 nfsd_racache_shutdown();
252 nfsd_up = false;
253}
254
255static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
256{
257 nfsd_shutdown();
258
259 svc_rpcb_cleanup(serv, net);
260
261 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
262 "cache\n");
263 nfsd_export_flush(net);
264}
265
266void nfsd_reset_versions(void)
267{
268 int found_one = 0;
269 int i;
270
271 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
272 if (nfsd_program.pg_vers[i])
273 found_one = 1;
274 }
275
276 if (!found_one) {
277 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
278 nfsd_program.pg_vers[i] = nfsd_version[i];
279#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
280 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
281 nfsd_acl_program.pg_vers[i] =
282 nfsd_acl_version[i];
283#endif
284 }
285}
286
287/*
288 * Each session guarantees a negotiated per slot memory cache for replies
289 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
290 * NFSv4.1 server might want to use more memory for a DRC than a machine
291 * with mutiple services.
292 *
293 * Impose a hard limit on the number of pages for the DRC which varies
294 * according to the machines free pages. This is of course only a default.
295 *
296 * For now this is a #defined shift which could be under admin control
297 * in the future.
298 */
299static void set_max_drc(void)
300{
301 #define NFSD_DRC_SIZE_SHIFT 10
302 nfsd_drc_max_mem = (nr_free_buffer_pages()
303 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
304 nfsd_drc_mem_used = 0;
305 spin_lock_init(&nfsd_drc_lock);
306 dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
307}
308
309static int nfsd_get_default_max_blksize(void)
310{
311 struct sysinfo i;
312 unsigned long long target;
313 unsigned long ret;
314
315 si_meminfo(&i);
316 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
317 /*
318 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
319 * machines, but only uses 32K on 128M machines. Bottom out at
320 * 8K on 32M and smaller. Of course, this is only a default.
321 */
322 target >>= 12;
323
324 ret = NFSSVC_MAXBLKSIZE;
325 while (ret > target && ret >= 8*1024*2)
326 ret /= 2;
327 return ret;
328}
329
330int nfsd_create_serv(void)
331{
332 int error;
333 struct net *net = current->nsproxy->net_ns;
334
335 WARN_ON(!mutex_is_locked(&nfsd_mutex));
336 if (nfsd_serv) {
337 svc_get(nfsd_serv);
338 return 0;
339 }
340 if (nfsd_max_blksize == 0)
341 nfsd_max_blksize = nfsd_get_default_max_blksize();
342 nfsd_reset_versions();
343 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
344 nfsd_last_thread, nfsd, THIS_MODULE);
345 if (nfsd_serv == NULL)
346 return -ENOMEM;
347
348 error = svc_bind(nfsd_serv, net);
349 if (error < 0) {
350 svc_destroy(nfsd_serv);
351 return error;
352 }
353
354 set_max_drc();
355 do_gettimeofday(&nfssvc_boot); /* record boot time */
356 return 0;
357}
358
359int nfsd_nrpools(void)
360{
361 if (nfsd_serv == NULL)
362 return 0;
363 else
364 return nfsd_serv->sv_nrpools;
365}
366
367int nfsd_get_nrthreads(int n, int *nthreads)
368{
369 int i = 0;
370
371 if (nfsd_serv != NULL) {
372 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
373 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
374 }
375
376 return 0;
377}
378
379int nfsd_set_nrthreads(int n, int *nthreads)
380{
381 int i = 0;
382 int tot = 0;
383 int err = 0;
384 struct net *net = &init_net;
385
386 WARN_ON(!mutex_is_locked(&nfsd_mutex));
387
388 if (nfsd_serv == NULL || n <= 0)
389 return 0;
390
391 if (n > nfsd_serv->sv_nrpools)
392 n = nfsd_serv->sv_nrpools;
393
394 /* enforce a global maximum number of threads */
395 tot = 0;
396 for (i = 0; i < n; i++) {
397 if (nthreads[i] > NFSD_MAXSERVS)
398 nthreads[i] = NFSD_MAXSERVS;
399 tot += nthreads[i];
400 }
401 if (tot > NFSD_MAXSERVS) {
402 /* total too large: scale down requested numbers */
403 for (i = 0; i < n && tot > 0; i++) {
404 int new = nthreads[i] * NFSD_MAXSERVS / tot;
405 tot -= (nthreads[i] - new);
406 nthreads[i] = new;
407 }
408 for (i = 0; i < n && tot > 0; i++) {
409 nthreads[i]--;
410 tot--;
411 }
412 }
413
414 /*
415 * There must always be a thread in pool 0; the admin
416 * can't shut down NFS completely using pool_threads.
417 */
418 if (nthreads[0] == 0)
419 nthreads[0] = 1;
420
421 /* apply the new numbers */
422 svc_get(nfsd_serv);
423 for (i = 0; i < n; i++) {
424 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
425 nthreads[i]);
426 if (err)
427 break;
428 }
429 nfsd_destroy(net);
430 return err;
431}
432
433/*
434 * Adjust the number of threads and return the new number of threads.
435 * This is also the function that starts the server if necessary, if
436 * this is the first time nrservs is nonzero.
437 */
438int
439nfsd_svc(unsigned short port, int nrservs)
440{
441 int error;
442 bool nfsd_up_before;
443 struct net *net = &init_net;
444
445 mutex_lock(&nfsd_mutex);
446 dprintk("nfsd: creating service\n");
447 if (nrservs <= 0)
448 nrservs = 0;
449 if (nrservs > NFSD_MAXSERVS)
450 nrservs = NFSD_MAXSERVS;
451 error = 0;
452 if (nrservs == 0 && nfsd_serv == NULL)
453 goto out;
454
455 error = nfsd_create_serv();
456 if (error)
457 goto out;
458
459 nfsd_up_before = nfsd_up;
460
461 error = nfsd_startup(port, nrservs);
462 if (error)
463 goto out_destroy;
464 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
465 if (error)
466 goto out_shutdown;
467 /* We are holding a reference to nfsd_serv which
468 * we don't want to count in the return value,
469 * so subtract 1
470 */
471 error = nfsd_serv->sv_nrthreads - 1;
472out_shutdown:
473 if (error < 0 && !nfsd_up_before)
474 nfsd_shutdown();
475out_destroy:
476 nfsd_destroy(net); /* Release server */
477out:
478 mutex_unlock(&nfsd_mutex);
479 return error;
480}
481
482
483/*
484 * This is the NFS server kernel thread
485 */
486static int
487nfsd(void *vrqstp)
488{
489 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
490 int err, preverr = 0;
491
492 /* Lock module and set up kernel thread */
493 mutex_lock(&nfsd_mutex);
494
495 /* At this point, the thread shares current->fs
496 * with the init process. We need to create files with a
497 * umask of 0 instead of init's umask. */
498 if (unshare_fs_struct() < 0) {
499 printk("Unable to start nfsd thread: out of memory\n");
500 goto out;
501 }
502
503 current->fs->umask = 0;
504
505 /*
506 * thread is spawned with all signals set to SIG_IGN, re-enable
507 * the ones that will bring down the thread
508 */
509 allow_signal(SIGKILL);
510 allow_signal(SIGHUP);
511 allow_signal(SIGINT);
512 allow_signal(SIGQUIT);
513
514 nfsdstats.th_cnt++;
515 mutex_unlock(&nfsd_mutex);
516
517 /*
518 * We want less throttling in balance_dirty_pages() so that nfs to
519 * localhost doesn't cause nfsd to lock up due to all the client's
520 * dirty pages.
521 */
522 current->flags |= PF_LESS_THROTTLE;
523 set_freezable();
524
525 /*
526 * The main request loop
527 */
528 for (;;) {
529 /*
530 * Find a socket with data available and call its
531 * recvfrom routine.
532 */
533 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
534 ;
535 if (err == -EINTR)
536 break;
537 else if (err < 0) {
538 if (err != preverr) {
539 printk(KERN_WARNING "%s: unexpected error "
540 "from svc_recv (%d)\n", __func__, -err);
541 preverr = err;
542 }
543 schedule_timeout_uninterruptible(HZ);
544 continue;
545 }
546
547 validate_process_creds();
548 svc_process(rqstp);
549 validate_process_creds();
550 }
551
552 /* Clear signals before calling svc_exit_thread() */
553 flush_signals(current);
554
555 mutex_lock(&nfsd_mutex);
556 nfsdstats.th_cnt --;
557
558out:
559 rqstp->rq_server = NULL;
560
561 /* Release the thread */
562 svc_exit_thread(rqstp);
563
564 nfsd_destroy(&init_net);
565
566 /* Release module */
567 mutex_unlock(&nfsd_mutex);
568 module_put_and_exit(0);
569 return 0;
570}
571
572static __be32 map_new_errors(u32 vers, __be32 nfserr)
573{
574 if (nfserr == nfserr_jukebox && vers == 2)
575 return nfserr_dropit;
576 if (nfserr == nfserr_wrongsec && vers < 4)
577 return nfserr_acces;
578 return nfserr;
579}
580
581int
582nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
583{
584 struct svc_procedure *proc;
585 kxdrproc_t xdr;
586 __be32 nfserr;
587 __be32 *nfserrp;
588
589 dprintk("nfsd_dispatch: vers %d proc %d\n",
590 rqstp->rq_vers, rqstp->rq_proc);
591 proc = rqstp->rq_procinfo;
592
593 /*
594 * Give the xdr decoder a chance to change this if it wants
595 * (necessary in the NFSv4.0 compound case)
596 */
597 rqstp->rq_cachetype = proc->pc_cachetype;
598 /* Decode arguments */
599 xdr = proc->pc_decode;
600 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
601 rqstp->rq_argp)) {
602 dprintk("nfsd: failed to decode arguments!\n");
603 *statp = rpc_garbage_args;
604 return 1;
605 }
606
607 /* Check whether we have this call in the cache. */
608 switch (nfsd_cache_lookup(rqstp)) {
609 case RC_INTR:
610 case RC_DROPIT:
611 return 0;
612 case RC_REPLY:
613 return 1;
614 case RC_DOIT:;
615 /* do it */
616 }
617
618 /* need to grab the location to store the status, as
619 * nfsv4 does some encoding while processing
620 */
621 nfserrp = rqstp->rq_res.head[0].iov_base
622 + rqstp->rq_res.head[0].iov_len;
623 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
624
625 /* Now call the procedure handler, and encode NFS status. */
626 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
627 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
628 if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
629 dprintk("nfsd: Dropping request; may be revisited later\n");
630 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
631 return 0;
632 }
633
634 if (rqstp->rq_proc != 0)
635 *nfserrp++ = nfserr;
636
637 /* Encode result.
638 * For NFSv2, additional info is never returned in case of an error.
639 */
640 if (!(nfserr && rqstp->rq_vers == 2)) {
641 xdr = proc->pc_encode;
642 if (xdr && !xdr(rqstp, nfserrp,
643 rqstp->rq_resp)) {
644 /* Failed to encode result. Release cache entry */
645 dprintk("nfsd: failed to encode result!\n");
646 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
647 *statp = rpc_system_err;
648 return 1;
649 }
650 }
651
652 /* Store reply in cache. */
653 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
654 return 1;
655}
656
657int nfsd_pool_stats_open(struct inode *inode, struct file *file)
658{
659 int ret;
660 mutex_lock(&nfsd_mutex);
661 if (nfsd_serv == NULL) {
662 mutex_unlock(&nfsd_mutex);
663 return -ENODEV;
664 }
665 /* bump up the psudo refcount while traversing */
666 svc_get(nfsd_serv);
667 ret = svc_pool_stats_open(nfsd_serv, file);
668 mutex_unlock(&nfsd_mutex);
669 return ret;
670}
671
672int nfsd_pool_stats_release(struct inode *inode, struct file *file)
673{
674 int ret = seq_release(inode, file);
675 struct net *net = &init_net;
676
677 mutex_lock(&nfsd_mutex);
678 /* this function really, really should have been called svc_put() */
679 nfsd_destroy(net);
680 mutex_unlock(&nfsd_mutex);
681 return ret;
682}
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/nfslocalio.h>
23#include <linux/seq_file.h>
24#include <linux/inetdevice.h>
25#include <net/addrconf.h>
26#include <net/ipv6.h>
27#include <net/net_namespace.h>
28#include "nfsd.h"
29#include "cache.h"
30#include "vfs.h"
31#include "netns.h"
32#include "filecache.h"
33
34#include "trace.h"
35
36#define NFSDDBG_FACILITY NFSDDBG_SVC
37
38atomic_t nfsd_th_cnt = ATOMIC_INIT(0);
39static int nfsd(void *vrqstp);
40#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
41static int nfsd_acl_rpcbind_set(struct net *,
42 const struct svc_program *,
43 u32, int,
44 unsigned short,
45 unsigned short);
46static __be32 nfsd_acl_init_request(struct svc_rqst *,
47 const struct svc_program *,
48 struct svc_process_info *);
49#endif
50static int nfsd_rpcbind_set(struct net *,
51 const struct svc_program *,
52 u32, int,
53 unsigned short,
54 unsigned short);
55static __be32 nfsd_init_request(struct svc_rqst *,
56 const struct svc_program *,
57 struct svc_process_info *);
58
59/*
60 * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and some members
61 * of the svc_serv struct such as ->sv_temp_socks and ->sv_permsocks.
62 *
63 * Finally, the nfsd_mutex also protects some of the global variables that are
64 * accessed when nfsd starts and that are settable via the write_* routines in
65 * nfsctl.c. In particular:
66 *
67 * user_recovery_dirname
68 * user_lease_time
69 * nfsd_versions
70 */
71DEFINE_MUTEX(nfsd_mutex);
72
73/*
74 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
75 * nfsd_drc_max_pages limits the total amount of memory available for
76 * version 4.1 DRC caches.
77 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
78 */
79DEFINE_SPINLOCK(nfsd_drc_lock);
80unsigned long nfsd_drc_max_mem;
81unsigned long nfsd_drc_mem_used;
82
83#if IS_ENABLED(CONFIG_NFS_LOCALIO)
84static const struct svc_version *localio_versions[] = {
85 [1] = &localio_version1,
86};
87
88#define NFSD_LOCALIO_NRVERS ARRAY_SIZE(localio_versions)
89
90#endif /* CONFIG_NFS_LOCALIO */
91
92#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
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
105#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
106
107static const struct svc_version *nfsd_version[NFSD_MAXVERS+1] = {
108#if defined(CONFIG_NFSD_V2)
109 [2] = &nfsd_version2,
110#endif
111 [3] = &nfsd_version3,
112#if defined(CONFIG_NFSD_V4)
113 [4] = &nfsd_version4,
114#endif
115};
116
117struct svc_program nfsd_programs[] = {
118 {
119 .pg_prog = NFS_PROGRAM, /* program number */
120 .pg_nvers = NFSD_MAXVERS+1, /* nr of entries in nfsd_version */
121 .pg_vers = nfsd_version, /* version table */
122 .pg_name = "nfsd", /* program name */
123 .pg_class = "nfsd", /* authentication class */
124 .pg_authenticate = svc_set_client, /* export authentication */
125 .pg_init_request = nfsd_init_request,
126 .pg_rpcbind_set = nfsd_rpcbind_set,
127 },
128#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
129 {
130 .pg_prog = NFS_ACL_PROGRAM,
131 .pg_nvers = NFSD_ACL_NRVERS,
132 .pg_vers = nfsd_acl_version,
133 .pg_name = "nfsacl",
134 .pg_class = "nfsd",
135 .pg_authenticate = svc_set_client,
136 .pg_init_request = nfsd_acl_init_request,
137 .pg_rpcbind_set = nfsd_acl_rpcbind_set,
138 },
139#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
140#if IS_ENABLED(CONFIG_NFS_LOCALIO)
141 {
142 .pg_prog = NFS_LOCALIO_PROGRAM,
143 .pg_nvers = NFSD_LOCALIO_NRVERS,
144 .pg_vers = localio_versions,
145 .pg_name = "nfslocalio",
146 .pg_class = "nfsd",
147 .pg_authenticate = svc_set_client,
148 .pg_init_request = svc_generic_init_request,
149 .pg_rpcbind_set = svc_generic_rpcbind_set,
150 }
151#endif /* CONFIG_NFS_LOCALIO */
152};
153
154bool nfsd_support_version(int vers)
155{
156 if (vers >= NFSD_MINVERS && vers <= NFSD_MAXVERS)
157 return nfsd_version[vers] != NULL;
158 return false;
159}
160
161int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change)
162{
163 if (vers < NFSD_MINVERS || vers > NFSD_MAXVERS)
164 return 0;
165 switch(change) {
166 case NFSD_SET:
167 nn->nfsd_versions[vers] = nfsd_support_version(vers);
168 break;
169 case NFSD_CLEAR:
170 nn->nfsd_versions[vers] = false;
171 break;
172 case NFSD_TEST:
173 return nn->nfsd_versions[vers];
174 case NFSD_AVAIL:
175 return nfsd_support_version(vers);
176 }
177 return 0;
178}
179
180static void
181nfsd_adjust_nfsd_versions4(struct nfsd_net *nn)
182{
183 unsigned i;
184
185 for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
186 if (nn->nfsd4_minorversions[i])
187 return;
188 }
189 nfsd_vers(nn, 4, NFSD_CLEAR);
190}
191
192int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change)
193{
194 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
195 change != NFSD_AVAIL)
196 return -1;
197
198 switch(change) {
199 case NFSD_SET:
200 nfsd_vers(nn, 4, NFSD_SET);
201 nn->nfsd4_minorversions[minorversion] =
202 nfsd_vers(nn, 4, NFSD_TEST);
203 break;
204 case NFSD_CLEAR:
205 nn->nfsd4_minorversions[minorversion] = false;
206 nfsd_adjust_nfsd_versions4(nn);
207 break;
208 case NFSD_TEST:
209 return nn->nfsd4_minorversions[minorversion];
210 case NFSD_AVAIL:
211 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION &&
212 nfsd_vers(nn, 4, NFSD_AVAIL);
213 }
214 return 0;
215}
216
217bool nfsd_serv_try_get(struct net *net) __must_hold(rcu)
218{
219 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
220
221 return (nn && percpu_ref_tryget_live(&nn->nfsd_serv_ref));
222}
223
224void nfsd_serv_put(struct net *net) __must_hold(rcu)
225{
226 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
227
228 percpu_ref_put(&nn->nfsd_serv_ref);
229}
230
231static void nfsd_serv_done(struct percpu_ref *ref)
232{
233 struct nfsd_net *nn = container_of(ref, struct nfsd_net, nfsd_serv_ref);
234
235 complete(&nn->nfsd_serv_confirm_done);
236}
237
238static void nfsd_serv_free(struct percpu_ref *ref)
239{
240 struct nfsd_net *nn = container_of(ref, struct nfsd_net, nfsd_serv_ref);
241
242 complete(&nn->nfsd_serv_free_done);
243}
244
245/*
246 * Maximum number of nfsd processes
247 */
248#define NFSD_MAXSERVS 8192
249
250int nfsd_nrthreads(struct net *net)
251{
252 int rv = 0;
253 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
254
255 mutex_lock(&nfsd_mutex);
256 if (nn->nfsd_serv)
257 rv = nn->nfsd_serv->sv_nrthreads;
258 mutex_unlock(&nfsd_mutex);
259 return rv;
260}
261
262static int nfsd_init_socks(struct net *net, const struct cred *cred)
263{
264 int error;
265 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
266
267 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
268 return 0;
269
270 error = svc_xprt_create(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
271 SVC_SOCK_DEFAULTS, cred);
272 if (error < 0)
273 return error;
274
275 error = svc_xprt_create(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
276 SVC_SOCK_DEFAULTS, cred);
277 if (error < 0)
278 return error;
279
280 return 0;
281}
282
283static int nfsd_users = 0;
284
285static int nfsd_startup_generic(void)
286{
287 int ret;
288
289 if (nfsd_users++)
290 return 0;
291
292 ret = nfsd_file_cache_init();
293 if (ret)
294 goto dec_users;
295
296 ret = nfs4_state_start();
297 if (ret)
298 goto out_file_cache;
299 return 0;
300
301out_file_cache:
302 nfsd_file_cache_shutdown();
303dec_users:
304 nfsd_users--;
305 return ret;
306}
307
308static void nfsd_shutdown_generic(void)
309{
310 if (--nfsd_users)
311 return;
312
313 nfs4_state_shutdown();
314 nfsd_file_cache_shutdown();
315}
316
317static bool nfsd_needs_lockd(struct nfsd_net *nn)
318{
319 return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST);
320}
321
322/**
323 * nfsd_copy_write_verifier - Atomically copy a write verifier
324 * @verf: buffer in which to receive the verifier cookie
325 * @nn: NFS net namespace
326 *
327 * This function provides a wait-free mechanism for copying the
328 * namespace's write verifier without tearing it.
329 */
330void nfsd_copy_write_verifier(__be32 verf[2], struct nfsd_net *nn)
331{
332 unsigned int seq;
333
334 do {
335 seq = read_seqbegin(&nn->writeverf_lock);
336 memcpy(verf, nn->writeverf, sizeof(nn->writeverf));
337 } while (read_seqretry(&nn->writeverf_lock, seq));
338}
339
340static void nfsd_reset_write_verifier_locked(struct nfsd_net *nn)
341{
342 struct timespec64 now;
343 u64 verf;
344
345 /*
346 * Because the time value is hashed, y2038 time_t overflow
347 * is irrelevant in this usage.
348 */
349 ktime_get_raw_ts64(&now);
350 verf = siphash_2u64(now.tv_sec, now.tv_nsec, &nn->siphash_key);
351 memcpy(nn->writeverf, &verf, sizeof(nn->writeverf));
352}
353
354/**
355 * nfsd_reset_write_verifier - Generate a new write verifier
356 * @nn: NFS net namespace
357 *
358 * This function updates the ->writeverf field of @nn. This field
359 * contains an opaque cookie that, according to Section 18.32.3 of
360 * RFC 8881, "the client can use to determine whether a server has
361 * changed instance state (e.g., server restart) between a call to
362 * WRITE and a subsequent call to either WRITE or COMMIT. This
363 * cookie MUST be unchanged during a single instance of the NFSv4.1
364 * server and MUST be unique between instances of the NFSv4.1
365 * server."
366 */
367void nfsd_reset_write_verifier(struct nfsd_net *nn)
368{
369 write_seqlock(&nn->writeverf_lock);
370 nfsd_reset_write_verifier_locked(nn);
371 write_sequnlock(&nn->writeverf_lock);
372}
373
374/*
375 * Crank up a set of per-namespace resources for a new NFSD instance,
376 * including lockd, a duplicate reply cache, an open file cache
377 * instance, and a cache of NFSv4 state objects.
378 */
379static int nfsd_startup_net(struct net *net, const struct cred *cred)
380{
381 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
382 int ret;
383
384 if (nn->nfsd_net_up)
385 return 0;
386
387 ret = nfsd_startup_generic();
388 if (ret)
389 return ret;
390 ret = nfsd_init_socks(net, cred);
391 if (ret)
392 goto out_socks;
393
394 if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
395 ret = lockd_up(net, cred);
396 if (ret)
397 goto out_socks;
398 nn->lockd_up = true;
399 }
400
401 ret = nfsd_file_cache_start_net(net);
402 if (ret)
403 goto out_lockd;
404
405 ret = nfsd_reply_cache_init(nn);
406 if (ret)
407 goto out_filecache;
408
409 ret = nfs4_state_start_net(net);
410 if (ret)
411 goto out_reply_cache;
412
413#ifdef CONFIG_NFSD_V4_2_INTER_SSC
414 nfsd4_ssc_init_umount_work(nn);
415#endif
416 nn->nfsd_net_up = true;
417 return 0;
418
419out_reply_cache:
420 nfsd_reply_cache_shutdown(nn);
421out_filecache:
422 nfsd_file_cache_shutdown_net(net);
423out_lockd:
424 if (nn->lockd_up) {
425 lockd_down(net);
426 nn->lockd_up = false;
427 }
428out_socks:
429 nfsd_shutdown_generic();
430 return ret;
431}
432
433static void nfsd_shutdown_net(struct net *net)
434{
435 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
436
437 if (!nn->nfsd_net_up)
438 return;
439 nfsd_export_flush(net);
440 nfs4_state_shutdown_net(net);
441 nfsd_reply_cache_shutdown(nn);
442 nfsd_file_cache_shutdown_net(net);
443 if (nn->lockd_up) {
444 lockd_down(net);
445 nn->lockd_up = false;
446 }
447 percpu_ref_exit(&nn->nfsd_serv_ref);
448 nn->nfsd_net_up = false;
449 nfsd_shutdown_generic();
450}
451
452static DEFINE_SPINLOCK(nfsd_notifier_lock);
453static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
454 void *ptr)
455{
456 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
457 struct net_device *dev = ifa->ifa_dev->dev;
458 struct net *net = dev_net(dev);
459 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
460 struct sockaddr_in sin;
461
462 if (event != NETDEV_DOWN || !nn->nfsd_serv)
463 goto out;
464
465 spin_lock(&nfsd_notifier_lock);
466 if (nn->nfsd_serv) {
467 dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
468 sin.sin_family = AF_INET;
469 sin.sin_addr.s_addr = ifa->ifa_local;
470 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
471 }
472 spin_unlock(&nfsd_notifier_lock);
473
474out:
475 return NOTIFY_DONE;
476}
477
478static struct notifier_block nfsd_inetaddr_notifier = {
479 .notifier_call = nfsd_inetaddr_event,
480};
481
482#if IS_ENABLED(CONFIG_IPV6)
483static int nfsd_inet6addr_event(struct notifier_block *this,
484 unsigned long event, void *ptr)
485{
486 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
487 struct net_device *dev = ifa->idev->dev;
488 struct net *net = dev_net(dev);
489 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
490 struct sockaddr_in6 sin6;
491
492 if (event != NETDEV_DOWN || !nn->nfsd_serv)
493 goto out;
494
495 spin_lock(&nfsd_notifier_lock);
496 if (nn->nfsd_serv) {
497 dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
498 sin6.sin6_family = AF_INET6;
499 sin6.sin6_addr = ifa->addr;
500 if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
501 sin6.sin6_scope_id = ifa->idev->dev->ifindex;
502 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
503 }
504 spin_unlock(&nfsd_notifier_lock);
505
506out:
507 return NOTIFY_DONE;
508}
509
510static struct notifier_block nfsd_inet6addr_notifier = {
511 .notifier_call = nfsd_inet6addr_event,
512};
513#endif
514
515/* Only used under nfsd_mutex, so this atomic may be overkill: */
516static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
517
518/**
519 * nfsd_destroy_serv - tear down NFSD's svc_serv for a namespace
520 * @net: network namespace the NFS service is associated with
521 */
522void nfsd_destroy_serv(struct net *net)
523{
524 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
525 struct svc_serv *serv = nn->nfsd_serv;
526
527 lockdep_assert_held(&nfsd_mutex);
528
529 percpu_ref_kill_and_confirm(&nn->nfsd_serv_ref, nfsd_serv_done);
530 wait_for_completion(&nn->nfsd_serv_confirm_done);
531 wait_for_completion(&nn->nfsd_serv_free_done);
532 /* percpu_ref_exit is called in nfsd_shutdown_net */
533
534 spin_lock(&nfsd_notifier_lock);
535 nn->nfsd_serv = NULL;
536 spin_unlock(&nfsd_notifier_lock);
537
538 /* check if the notifier still has clients */
539 if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
540 unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
541#if IS_ENABLED(CONFIG_IPV6)
542 unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
543#endif
544 }
545
546 svc_xprt_destroy_all(serv, net);
547
548 /*
549 * write_ports can create the server without actually starting
550 * any threads--if we get shut down before any threads are
551 * started, then nfsd_destroy_serv will be run before any of this
552 * other initialization has been done except the rpcb information.
553 */
554 svc_rpcb_cleanup(serv, net);
555
556 nfsd_shutdown_net(net);
557 svc_destroy(&serv);
558}
559
560void nfsd_reset_versions(struct nfsd_net *nn)
561{
562 int i;
563
564 for (i = 0; i <= NFSD_MAXVERS; i++)
565 if (nfsd_vers(nn, i, NFSD_TEST))
566 return;
567
568 for (i = 0; i <= NFSD_MAXVERS; i++)
569 if (i != 4)
570 nfsd_vers(nn, i, NFSD_SET);
571 else {
572 int minor = 0;
573 while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
574 minor++;
575 }
576}
577
578/*
579 * Each session guarantees a negotiated per slot memory cache for replies
580 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
581 * NFSv4.1 server might want to use more memory for a DRC than a machine
582 * with mutiple services.
583 *
584 * Impose a hard limit on the number of pages for the DRC which varies
585 * according to the machines free pages. This is of course only a default.
586 *
587 * For now this is a #defined shift which could be under admin control
588 * in the future.
589 */
590static void set_max_drc(void)
591{
592 #define NFSD_DRC_SIZE_SHIFT 7
593 nfsd_drc_max_mem = (nr_free_buffer_pages()
594 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
595 nfsd_drc_mem_used = 0;
596 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
597}
598
599static int nfsd_get_default_max_blksize(void)
600{
601 struct sysinfo i;
602 unsigned long long target;
603 unsigned long ret;
604
605 si_meminfo(&i);
606 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
607 /*
608 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
609 * machines, but only uses 32K on 128M machines. Bottom out at
610 * 8K on 32M and smaller. Of course, this is only a default.
611 */
612 target >>= 12;
613
614 ret = NFSSVC_MAXBLKSIZE;
615 while (ret > target && ret >= 8*1024*2)
616 ret /= 2;
617 return ret;
618}
619
620void nfsd_shutdown_threads(struct net *net)
621{
622 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
623 struct svc_serv *serv;
624
625 mutex_lock(&nfsd_mutex);
626 serv = nn->nfsd_serv;
627 if (serv == NULL) {
628 mutex_unlock(&nfsd_mutex);
629 return;
630 }
631
632 /* Kill outstanding nfsd threads */
633 svc_set_num_threads(serv, NULL, 0);
634 nfsd_destroy_serv(net);
635 mutex_unlock(&nfsd_mutex);
636}
637
638struct svc_rqst *nfsd_current_rqst(void)
639{
640 if (kthread_func(current) == nfsd)
641 return kthread_data(current);
642 return NULL;
643}
644
645int nfsd_create_serv(struct net *net)
646{
647 int error;
648 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
649 struct svc_serv *serv;
650
651 WARN_ON(!mutex_is_locked(&nfsd_mutex));
652 if (nn->nfsd_serv)
653 return 0;
654
655 error = percpu_ref_init(&nn->nfsd_serv_ref, nfsd_serv_free,
656 0, GFP_KERNEL);
657 if (error)
658 return error;
659 init_completion(&nn->nfsd_serv_free_done);
660 init_completion(&nn->nfsd_serv_confirm_done);
661
662 if (nfsd_max_blksize == 0)
663 nfsd_max_blksize = nfsd_get_default_max_blksize();
664 nfsd_reset_versions(nn);
665 serv = svc_create_pooled(nfsd_programs, ARRAY_SIZE(nfsd_programs),
666 &nn->nfsd_svcstats,
667 nfsd_max_blksize, nfsd);
668 if (serv == NULL)
669 return -ENOMEM;
670
671 serv->sv_maxconn = nn->max_connections;
672 error = svc_bind(serv, net);
673 if (error < 0) {
674 svc_destroy(&serv);
675 return error;
676 }
677 spin_lock(&nfsd_notifier_lock);
678 nn->nfsd_serv = serv;
679 spin_unlock(&nfsd_notifier_lock);
680
681 set_max_drc();
682 /* check if the notifier is already set */
683 if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
684 register_inetaddr_notifier(&nfsd_inetaddr_notifier);
685#if IS_ENABLED(CONFIG_IPV6)
686 register_inet6addr_notifier(&nfsd_inet6addr_notifier);
687#endif
688 }
689 nfsd_reset_write_verifier(nn);
690 return 0;
691}
692
693int nfsd_nrpools(struct net *net)
694{
695 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
696
697 if (nn->nfsd_serv == NULL)
698 return 0;
699 else
700 return nn->nfsd_serv->sv_nrpools;
701}
702
703int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
704{
705 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
706 struct svc_serv *serv = nn->nfsd_serv;
707 int i;
708
709 if (serv)
710 for (i = 0; i < serv->sv_nrpools && i < n; i++)
711 nthreads[i] = serv->sv_pools[i].sp_nrthreads;
712 return 0;
713}
714
715/**
716 * nfsd_set_nrthreads - set the number of running threads in the net's service
717 * @n: number of array members in @nthreads
718 * @nthreads: array of thread counts for each pool
719 * @net: network namespace to operate within
720 *
721 * This function alters the number of running threads for the given network
722 * namespace in each pool. If passed an array longer then the number of pools
723 * the extra pool settings are ignored. If passed an array shorter than the
724 * number of pools, the missing values are interpreted as 0's.
725 *
726 * Returns 0 on success or a negative errno on error.
727 */
728int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
729{
730 int i = 0;
731 int tot = 0;
732 int err = 0;
733 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
734
735 lockdep_assert_held(&nfsd_mutex);
736
737 if (nn->nfsd_serv == NULL || n <= 0)
738 return 0;
739
740 /*
741 * Special case: When n == 1, pass in NULL for the pool, so that the
742 * change is distributed equally among them.
743 */
744 if (n == 1)
745 return svc_set_num_threads(nn->nfsd_serv, NULL, nthreads[0]);
746
747 if (n > nn->nfsd_serv->sv_nrpools)
748 n = nn->nfsd_serv->sv_nrpools;
749
750 /* enforce a global maximum number of threads */
751 tot = 0;
752 for (i = 0; i < n; i++) {
753 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
754 tot += nthreads[i];
755 }
756 if (tot > NFSD_MAXSERVS) {
757 /* total too large: scale down requested numbers */
758 for (i = 0; i < n && tot > 0; i++) {
759 int new = nthreads[i] * NFSD_MAXSERVS / tot;
760 tot -= (nthreads[i] - new);
761 nthreads[i] = new;
762 }
763 for (i = 0; i < n && tot > 0; i++) {
764 nthreads[i]--;
765 tot--;
766 }
767 }
768
769 /* apply the new numbers */
770 for (i = 0; i < n; i++) {
771 err = svc_set_num_threads(nn->nfsd_serv,
772 &nn->nfsd_serv->sv_pools[i],
773 nthreads[i]);
774 if (err)
775 goto out;
776 }
777
778 /* Anything undefined in array is considered to be 0 */
779 for (i = n; i < nn->nfsd_serv->sv_nrpools; ++i) {
780 err = svc_set_num_threads(nn->nfsd_serv,
781 &nn->nfsd_serv->sv_pools[i],
782 0);
783 if (err)
784 goto out;
785 }
786out:
787 return err;
788}
789
790/**
791 * nfsd_svc: start up or shut down the nfsd server
792 * @n: number of array members in @nthreads
793 * @nthreads: array of thread counts for each pool
794 * @net: network namespace to operate within
795 * @cred: credentials to use for xprt creation
796 * @scope: server scope value (defaults to nodename)
797 *
798 * Adjust the number of threads in each pool and return the new
799 * total number of threads in the service.
800 */
801int
802nfsd_svc(int n, int *nthreads, struct net *net, const struct cred *cred, const char *scope)
803{
804 int error;
805 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
806 struct svc_serv *serv;
807
808 lockdep_assert_held(&nfsd_mutex);
809
810 dprintk("nfsd: creating service\n");
811
812 strscpy(nn->nfsd_name, scope ? scope : utsname()->nodename,
813 sizeof(nn->nfsd_name));
814
815 error = nfsd_create_serv(net);
816 if (error)
817 goto out;
818 serv = nn->nfsd_serv;
819
820 error = nfsd_startup_net(net, cred);
821 if (error)
822 goto out_put;
823 error = nfsd_set_nrthreads(n, nthreads, net);
824 if (error)
825 goto out_put;
826 error = serv->sv_nrthreads;
827out_put:
828 if (serv->sv_nrthreads == 0)
829 nfsd_destroy_serv(net);
830out:
831 return error;
832}
833
834#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
835static bool
836nfsd_support_acl_version(int vers)
837{
838 if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
839 return nfsd_acl_version[vers] != NULL;
840 return false;
841}
842
843static int
844nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
845 u32 version, int family, unsigned short proto,
846 unsigned short port)
847{
848 if (!nfsd_support_acl_version(version) ||
849 !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
850 return 0;
851 return svc_generic_rpcbind_set(net, progp, version, family,
852 proto, port);
853}
854
855static __be32
856nfsd_acl_init_request(struct svc_rqst *rqstp,
857 const struct svc_program *progp,
858 struct svc_process_info *ret)
859{
860 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
861 int i;
862
863 if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
864 nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
865 return svc_generic_init_request(rqstp, progp, ret);
866
867 ret->mismatch.lovers = NFSD_ACL_NRVERS;
868 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
869 if (nfsd_support_acl_version(rqstp->rq_vers) &&
870 nfsd_vers(nn, i, NFSD_TEST)) {
871 ret->mismatch.lovers = i;
872 break;
873 }
874 }
875 if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
876 return rpc_prog_unavail;
877 ret->mismatch.hivers = NFSD_ACL_MINVERS;
878 for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
879 if (nfsd_support_acl_version(rqstp->rq_vers) &&
880 nfsd_vers(nn, i, NFSD_TEST)) {
881 ret->mismatch.hivers = i;
882 break;
883 }
884 }
885 return rpc_prog_mismatch;
886}
887#endif
888
889static int
890nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
891 u32 version, int family, unsigned short proto,
892 unsigned short port)
893{
894 if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
895 return 0;
896 return svc_generic_rpcbind_set(net, progp, version, family,
897 proto, port);
898}
899
900static __be32
901nfsd_init_request(struct svc_rqst *rqstp,
902 const struct svc_program *progp,
903 struct svc_process_info *ret)
904{
905 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
906 int i;
907
908 if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
909 return svc_generic_init_request(rqstp, progp, ret);
910
911 ret->mismatch.lovers = NFSD_MAXVERS + 1;
912 for (i = NFSD_MINVERS; i <= NFSD_MAXVERS; i++) {
913 if (nfsd_vers(nn, i, NFSD_TEST)) {
914 ret->mismatch.lovers = i;
915 break;
916 }
917 }
918 if (ret->mismatch.lovers > NFSD_MAXVERS)
919 return rpc_prog_unavail;
920 ret->mismatch.hivers = NFSD_MINVERS;
921 for (i = NFSD_MAXVERS; i >= NFSD_MINVERS; i--) {
922 if (nfsd_vers(nn, i, NFSD_TEST)) {
923 ret->mismatch.hivers = i;
924 break;
925 }
926 }
927 return rpc_prog_mismatch;
928}
929
930/*
931 * This is the NFS server kernel thread
932 */
933static int
934nfsd(void *vrqstp)
935{
936 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
937 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
938 struct net *net = perm_sock->xpt_net;
939 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
940
941 /* At this point, the thread shares current->fs
942 * with the init process. We need to create files with the
943 * umask as defined by the client instead of init's umask.
944 */
945 svc_thread_init_status(rqstp, unshare_fs_struct());
946
947 current->fs->umask = 0;
948
949 atomic_inc(&nfsd_th_cnt);
950
951 set_freezable();
952
953 /*
954 * The main request loop
955 */
956 while (!svc_thread_should_stop(rqstp)) {
957 /* Update sv_maxconn if it has changed */
958 rqstp->rq_server->sv_maxconn = nn->max_connections;
959
960 svc_recv(rqstp);
961
962 nfsd_file_net_dispose(nn);
963 }
964
965 atomic_dec(&nfsd_th_cnt);
966
967 /* Release the thread */
968 svc_exit_thread(rqstp);
969 return 0;
970}
971
972/**
973 * nfsd_dispatch - Process an NFS or NFSACL or LOCALIO Request
974 * @rqstp: incoming request
975 *
976 * This RPC dispatcher integrates the NFS server's duplicate reply cache.
977 *
978 * Return values:
979 * %0: Processing complete; do not send a Reply
980 * %1: Processing complete; send Reply in rqstp->rq_res
981 */
982int nfsd_dispatch(struct svc_rqst *rqstp)
983{
984 const struct svc_procedure *proc = rqstp->rq_procinfo;
985 __be32 *statp = rqstp->rq_accept_statp;
986 struct nfsd_cacherep *rp;
987 unsigned int start, len;
988 __be32 *nfs_reply;
989
990 /*
991 * Give the xdr decoder a chance to change this if it wants
992 * (necessary in the NFSv4.0 compound case)
993 */
994 rqstp->rq_cachetype = proc->pc_cachetype;
995
996 /*
997 * ->pc_decode advances the argument stream past the NFS
998 * Call header, so grab the header's starting location and
999 * size now for the call to nfsd_cache_lookup().
1000 */
1001 start = xdr_stream_pos(&rqstp->rq_arg_stream);
1002 len = xdr_stream_remaining(&rqstp->rq_arg_stream);
1003 if (!proc->pc_decode(rqstp, &rqstp->rq_arg_stream))
1004 goto out_decode_err;
1005
1006 /*
1007 * Release rq_status_counter setting it to an odd value after the rpc
1008 * request has been properly parsed. rq_status_counter is used to
1009 * notify the consumers if the rqstp fields are stable
1010 * (rq_status_counter is odd) or not meaningful (rq_status_counter
1011 * is even).
1012 */
1013 smp_store_release(&rqstp->rq_status_counter, rqstp->rq_status_counter | 1);
1014
1015 rp = NULL;
1016 switch (nfsd_cache_lookup(rqstp, start, len, &rp)) {
1017 case RC_DOIT:
1018 break;
1019 case RC_REPLY:
1020 goto out_cached_reply;
1021 case RC_DROPIT:
1022 goto out_dropit;
1023 }
1024
1025 nfs_reply = xdr_inline_decode(&rqstp->rq_res_stream, 0);
1026 *statp = proc->pc_func(rqstp);
1027 if (test_bit(RQ_DROPME, &rqstp->rq_flags))
1028 goto out_update_drop;
1029
1030 if (!proc->pc_encode(rqstp, &rqstp->rq_res_stream))
1031 goto out_encode_err;
1032
1033 /*
1034 * Release rq_status_counter setting it to an even value after the rpc
1035 * request has been properly processed.
1036 */
1037 smp_store_release(&rqstp->rq_status_counter, rqstp->rq_status_counter + 1);
1038
1039 nfsd_cache_update(rqstp, rp, rqstp->rq_cachetype, nfs_reply);
1040out_cached_reply:
1041 return 1;
1042
1043out_decode_err:
1044 trace_nfsd_garbage_args_err(rqstp);
1045 *statp = rpc_garbage_args;
1046 return 1;
1047
1048out_update_drop:
1049 nfsd_cache_update(rqstp, rp, RC_NOCACHE, NULL);
1050out_dropit:
1051 return 0;
1052
1053out_encode_err:
1054 trace_nfsd_cant_encode_err(rqstp);
1055 nfsd_cache_update(rqstp, rp, RC_NOCACHE, NULL);
1056 *statp = rpc_system_err;
1057 return 1;
1058}
1059
1060/**
1061 * nfssvc_decode_voidarg - Decode void arguments
1062 * @rqstp: Server RPC transaction context
1063 * @xdr: XDR stream positioned at arguments to decode
1064 *
1065 * Return values:
1066 * %false: Arguments were not valid
1067 * %true: Decoding was successful
1068 */
1069bool nfssvc_decode_voidarg(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1070{
1071 return true;
1072}
1073
1074/**
1075 * nfssvc_encode_voidres - Encode void results
1076 * @rqstp: Server RPC transaction context
1077 * @xdr: XDR stream into which to encode results
1078 *
1079 * Return values:
1080 * %false: Local error while encoding
1081 * %true: Encoding was successful
1082 */
1083bool nfssvc_encode_voidres(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1084{
1085 return true;
1086}