1 /*
2 * Central processing for nfsd.
3 *
4 * Authors: Olaf Kirch (okir@monad.swb.de)
5 *
6 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
7 */
8
9 #include <linux/sched.h>
10 #include <linux/freezer.h>
11 #include <linux/module.h>
12 #include <linux/fs_struct.h>
13 #include <linux/swap.h>
14
15 #include <linux/sunrpc/stats.h>
16 #include <linux/sunrpc/svcsock.h>
17 #include <linux/lockd/bind.h>
18 #include <linux/nfsacl.h>
19 #include <linux/seq_file.h>
20 #include <net/net_namespace.h>
21 #include "nfsd.h"
22 #include "cache.h"
23 #include "vfs.h"
24
25 #define NFSDDBG_FACILITY NFSDDBG_SVC
26
27 extern struct svc_program nfsd_program;
28 static int nfsd(void *vrqstp);
29 struct timeval nfssvc_boot;
30
31 /*
32 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
33 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
34 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
35 *
36 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
37 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
38 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
39 * entry of ->sv_pools[].
40 *
41 * Transitions of the thread count between zero and non-zero are of particular
42 * interest since the svc_serv needs to be created and initialized at that
43 * point, or freed.
44 *
45 * Finally, the nfsd_mutex also protects some of the global variables that are
46 * accessed when nfsd starts and that are settable via the write_* routines in
47 * nfsctl.c. In particular:
48 *
49 * user_recovery_dirname
50 * user_lease_time
51 * nfsd_versions
52 */
53 DEFINE_MUTEX(nfsd_mutex);
54 struct svc_serv *nfsd_serv;
55
56 /*
57 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
58 * nfsd_drc_max_pages limits the total amount of memory available for
59 * version 4.1 DRC caches.
60 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
61 */
62 spinlock_t nfsd_drc_lock;
63 unsigned int nfsd_drc_max_mem;
64 unsigned int nfsd_drc_mem_used;
65
66 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
67 static struct svc_stat nfsd_acl_svcstats;
68 static struct svc_version * nfsd_acl_version[] = {
69 [2] = &nfsd_acl_version2,
70 [3] = &nfsd_acl_version3,
71 };
72
73 #define NFSD_ACL_MINVERS 2
74 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
75 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
76
77 static struct svc_program nfsd_acl_program = {
78 .pg_prog = NFS_ACL_PROGRAM,
79 .pg_nvers = NFSD_ACL_NRVERS,
80 .pg_vers = nfsd_acl_versions,
81 .pg_name = "nfsacl",
82 .pg_class = "nfsd",
83 .pg_stats = &nfsd_acl_svcstats,
84 .pg_authenticate = &svc_set_client,
85 };
86
87 static struct svc_stat nfsd_acl_svcstats = {
88 .program = &nfsd_acl_program,
89 };
90 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
91
92 static struct svc_version * nfsd_version[] = {
93 [2] = &nfsd_version2,
94 #if defined(CONFIG_NFSD_V3)
95 [3] = &nfsd_version3,
96 #endif
97 #if defined(CONFIG_NFSD_V4)
98 [4] = &nfsd_version4,
99 #endif
100 };
101
102 #define NFSD_MINVERS 2
103 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
104 static struct svc_version *nfsd_versions[NFSD_NRVERS];
105
106 struct svc_program nfsd_program = {
107 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
108 .pg_next = &nfsd_acl_program,
109 #endif
110 .pg_prog = NFS_PROGRAM, /* program number */
111 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
112 .pg_vers = nfsd_versions, /* version table */
113 .pg_name = "nfsd", /* program name */
114 .pg_class = "nfsd", /* authentication class */
115 .pg_stats = &nfsd_svcstats, /* version table */
116 .pg_authenticate = &svc_set_client, /* export authentication */
117
118 };
119
120 u32 nfsd_supported_minorversion;
121
nfsd_vers(int vers,enum vers_op change)122 int nfsd_vers(int vers, enum vers_op change)
123 {
124 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
125 return 0;
126 switch(change) {
127 case NFSD_SET:
128 nfsd_versions[vers] = nfsd_version[vers];
129 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
130 if (vers < NFSD_ACL_NRVERS)
131 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
132 #endif
133 break;
134 case NFSD_CLEAR:
135 nfsd_versions[vers] = NULL;
136 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
137 if (vers < NFSD_ACL_NRVERS)
138 nfsd_acl_versions[vers] = NULL;
139 #endif
140 break;
141 case NFSD_TEST:
142 return nfsd_versions[vers] != NULL;
143 case NFSD_AVAIL:
144 return nfsd_version[vers] != NULL;
145 }
146 return 0;
147 }
148
nfsd_minorversion(u32 minorversion,enum vers_op change)149 int nfsd_minorversion(u32 minorversion, enum vers_op change)
150 {
151 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
152 return -1;
153 switch(change) {
154 case NFSD_SET:
155 nfsd_supported_minorversion = minorversion;
156 break;
157 case NFSD_CLEAR:
158 if (minorversion == 0)
159 return -1;
160 nfsd_supported_minorversion = minorversion - 1;
161 break;
162 case NFSD_TEST:
163 return minorversion <= nfsd_supported_minorversion;
164 case NFSD_AVAIL:
165 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
166 }
167 return 0;
168 }
169
170 /*
171 * Maximum number of nfsd processes
172 */
173 #define NFSD_MAXSERVS 8192
174
nfsd_nrthreads(void)175 int nfsd_nrthreads(void)
176 {
177 int rv = 0;
178 mutex_lock(&nfsd_mutex);
179 if (nfsd_serv)
180 rv = nfsd_serv->sv_nrthreads;
181 mutex_unlock(&nfsd_mutex);
182 return rv;
183 }
184
nfsd_init_socks(int port,struct net * net)185 static int nfsd_init_socks(int port, struct net *net)
186 {
187 int error;
188 if (!list_empty(&nfsd_serv->sv_permsocks))
189 return 0;
190
191 error = svc_create_xprt(nfsd_serv, "udp", net, PF_INET, port,
192 SVC_SOCK_DEFAULTS);
193 if (error < 0)
194 return error;
195
196 error = svc_create_xprt(nfsd_serv, "tcp", net, PF_INET, port,
197 SVC_SOCK_DEFAULTS);
198 if (error < 0)
199 return error;
200
201 return 0;
202 }
203
204 static bool nfsd_up = false;
205
nfsd_startup(unsigned short port,int nrservs,struct net * net)206 static int nfsd_startup(unsigned short port, int nrservs, struct net *net)
207 {
208 int ret;
209
210 if (nfsd_up)
211 return 0;
212 /*
213 * Readahead param cache - will no-op if it already exists.
214 * (Note therefore results will be suboptimal if number of
215 * threads is modified after nfsd start.)
216 */
217 ret = nfsd_racache_init(2*nrservs);
218 if (ret)
219 return ret;
220 ret = nfsd_init_socks(port, net);
221 if (ret)
222 goto out_racache;
223 ret = lockd_up(net);
224 if (ret)
225 goto out_racache;
226 ret = nfs4_state_start();
227 if (ret)
228 goto out_lockd;
229 nfsd_up = true;
230 return 0;
231 out_lockd:
232 lockd_down(net);
233 out_racache:
234 nfsd_racache_shutdown();
235 return ret;
236 }
237
nfsd_shutdown(struct net * net)238 static void nfsd_shutdown(struct net *net)
239 {
240 /*
241 * write_ports can create the server without actually starting
242 * any threads--if we get shut down before any threads are
243 * started, then nfsd_last_thread will be run before any of this
244 * other initialization has been done.
245 */
246 if (!nfsd_up)
247 return;
248 nfs4_state_shutdown();
249 lockd_down(net);
250 nfsd_racache_shutdown();
251 nfsd_up = false;
252 }
253
nfsd_last_thread(struct svc_serv * serv,struct net * net)254 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
255 {
256 nfsd_shutdown(net);
257
258 svc_rpcb_cleanup(serv, net);
259
260 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
261 "cache\n");
262 nfsd_export_flush();
263 }
264
nfsd_reset_versions(void)265 void nfsd_reset_versions(void)
266 {
267 int found_one = 0;
268 int i;
269
270 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
271 if (nfsd_program.pg_vers[i])
272 found_one = 1;
273 }
274
275 if (!found_one) {
276 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
277 nfsd_program.pg_vers[i] = nfsd_version[i];
278 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
279 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
280 nfsd_acl_program.pg_vers[i] =
281 nfsd_acl_version[i];
282 #endif
283 }
284 }
285
286 /*
287 * Each session guarantees a negotiated per slot memory cache for replies
288 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
289 * NFSv4.1 server might want to use more memory for a DRC than a machine
290 * with mutiple services.
291 *
292 * Impose a hard limit on the number of pages for the DRC which varies
293 * according to the machines free pages. This is of course only a default.
294 *
295 * For now this is a #defined shift which could be under admin control
296 * in the future.
297 */
set_max_drc(void)298 static void set_max_drc(void)
299 {
300 #define NFSD_DRC_SIZE_SHIFT 10
301 nfsd_drc_max_mem = (nr_free_buffer_pages()
302 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
303 nfsd_drc_mem_used = 0;
304 spin_lock_init(&nfsd_drc_lock);
305 dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
306 }
307
nfsd_get_default_max_blksize(void)308 static int nfsd_get_default_max_blksize(void)
309 {
310 struct sysinfo i;
311 unsigned long long target;
312 unsigned long ret;
313
314 si_meminfo(&i);
315 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
316 /*
317 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
318 * machines, but only uses 32K on 128M machines. Bottom out at
319 * 8K on 32M and smaller. Of course, this is only a default.
320 */
321 target >>= 12;
322
323 ret = NFSSVC_MAXBLKSIZE;
324 while (ret > target && ret >= 8*1024*2)
325 ret /= 2;
326 return ret;
327 }
328
nfsd_create_serv(struct net * net)329 int nfsd_create_serv(struct net *net)
330 {
331 int error;
332
333 WARN_ON(!mutex_is_locked(&nfsd_mutex));
334 if (nfsd_serv) {
335 svc_get(nfsd_serv);
336 return 0;
337 }
338 if (nfsd_max_blksize == 0)
339 nfsd_max_blksize = nfsd_get_default_max_blksize();
340 nfsd_reset_versions();
341 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
342 nfsd_last_thread, nfsd, THIS_MODULE);
343 if (nfsd_serv == NULL)
344 return -ENOMEM;
345
346 error = svc_bind(nfsd_serv, net);
347 if (error < 0) {
348 svc_destroy(nfsd_serv);
349 return error;
350 }
351
352 set_max_drc();
353 do_gettimeofday(&nfssvc_boot); /* record boot time */
354 return 0;
355 }
356
nfsd_nrpools(void)357 int nfsd_nrpools(void)
358 {
359 if (nfsd_serv == NULL)
360 return 0;
361 else
362 return nfsd_serv->sv_nrpools;
363 }
364
nfsd_get_nrthreads(int n,int * nthreads)365 int nfsd_get_nrthreads(int n, int *nthreads)
366 {
367 int i = 0;
368
369 if (nfsd_serv != NULL) {
370 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
371 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
372 }
373
374 return 0;
375 }
376
nfsd_set_nrthreads(int n,int * nthreads,struct net * net)377 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
378 {
379 int i = 0;
380 int tot = 0;
381 int err = 0;
382
383 WARN_ON(!mutex_is_locked(&nfsd_mutex));
384
385 if (nfsd_serv == NULL || n <= 0)
386 return 0;
387
388 if (n > nfsd_serv->sv_nrpools)
389 n = nfsd_serv->sv_nrpools;
390
391 /* enforce a global maximum number of threads */
392 tot = 0;
393 for (i = 0; i < n; i++) {
394 if (nthreads[i] > NFSD_MAXSERVS)
395 nthreads[i] = NFSD_MAXSERVS;
396 tot += nthreads[i];
397 }
398 if (tot > NFSD_MAXSERVS) {
399 /* total too large: scale down requested numbers */
400 for (i = 0; i < n && tot > 0; i++) {
401 int new = nthreads[i] * NFSD_MAXSERVS / tot;
402 tot -= (nthreads[i] - new);
403 nthreads[i] = new;
404 }
405 for (i = 0; i < n && tot > 0; i++) {
406 nthreads[i]--;
407 tot--;
408 }
409 }
410
411 /*
412 * There must always be a thread in pool 0; the admin
413 * can't shut down NFS completely using pool_threads.
414 */
415 if (nthreads[0] == 0)
416 nthreads[0] = 1;
417
418 /* apply the new numbers */
419 svc_get(nfsd_serv);
420 for (i = 0; i < n; i++) {
421 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
422 nthreads[i]);
423 if (err)
424 break;
425 }
426 nfsd_destroy(net);
427 return err;
428 }
429
430 /*
431 * Adjust the number of threads and return the new number of threads.
432 * This is also the function that starts the server if necessary, if
433 * this is the first time nrservs is nonzero.
434 */
435 int
nfsd_svc(unsigned short port,int nrservs,struct net * net)436 nfsd_svc(unsigned short port, int nrservs, struct net *net)
437 {
438 int error;
439 bool nfsd_up_before;
440
441 mutex_lock(&nfsd_mutex);
442 dprintk("nfsd: creating service\n");
443 if (nrservs <= 0)
444 nrservs = 0;
445 if (nrservs > NFSD_MAXSERVS)
446 nrservs = NFSD_MAXSERVS;
447 error = 0;
448 if (nrservs == 0 && nfsd_serv == NULL)
449 goto out;
450
451 error = nfsd_create_serv(net);
452 if (error)
453 goto out;
454
455 nfsd_up_before = nfsd_up;
456
457 error = nfsd_startup(port, nrservs, net);
458 if (error)
459 goto out_destroy;
460 error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
461 if (error)
462 goto out_shutdown;
463 /* We are holding a reference to nfsd_serv which
464 * we don't want to count in the return value,
465 * so subtract 1
466 */
467 error = nfsd_serv->sv_nrthreads - 1;
468 out_shutdown:
469 if (error < 0 && !nfsd_up_before)
470 nfsd_shutdown(net);
471 out_destroy:
472 nfsd_destroy(net); /* Release server */
473 out:
474 mutex_unlock(&nfsd_mutex);
475 return error;
476 }
477
478
479 /*
480 * This is the NFS server kernel thread
481 */
482 static int
nfsd(void * vrqstp)483 nfsd(void *vrqstp)
484 {
485 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
486 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
487 struct net *net = perm_sock->xpt_net;
488 int err, preverr = 0;
489
490 /* Lock module and set up kernel thread */
491 mutex_lock(&nfsd_mutex);
492
493 /* At this point, the thread shares current->fs
494 * with the init process. We need to create files with a
495 * umask of 0 instead of init's umask. */
496 if (unshare_fs_struct() < 0) {
497 printk("Unable to start nfsd thread: out of memory\n");
498 goto out;
499 }
500
501 current->fs->umask = 0;
502
503 /*
504 * thread is spawned with all signals set to SIG_IGN, re-enable
505 * the ones that will bring down the thread
506 */
507 allow_signal(SIGKILL);
508 allow_signal(SIGHUP);
509 allow_signal(SIGINT);
510 allow_signal(SIGQUIT);
511
512 nfsdstats.th_cnt++;
513 mutex_unlock(&nfsd_mutex);
514
515 /*
516 * We want less throttling in balance_dirty_pages() so that nfs to
517 * localhost doesn't cause nfsd to lock up due to all the client's
518 * dirty pages.
519 */
520 current->flags |= PF_LESS_THROTTLE;
521 set_freezable();
522
523 /*
524 * The main request loop
525 */
526 for (;;) {
527 /*
528 * Find a socket with data available and call its
529 * recvfrom routine.
530 */
531 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
532 ;
533 if (err == -EINTR)
534 break;
535 else if (err < 0) {
536 if (err != preverr) {
537 printk(KERN_WARNING "%s: unexpected error "
538 "from svc_recv (%d)\n", __func__, -err);
539 preverr = err;
540 }
541 schedule_timeout_uninterruptible(HZ);
542 continue;
543 }
544
545 validate_process_creds();
546 svc_process(rqstp);
547 validate_process_creds();
548 }
549
550 /* Clear signals before calling svc_exit_thread() */
551 flush_signals(current);
552
553 mutex_lock(&nfsd_mutex);
554 nfsdstats.th_cnt --;
555
556 out:
557 rqstp->rq_server = NULL;
558
559 /* Release the thread */
560 svc_exit_thread(rqstp);
561
562 nfsd_destroy(net);
563
564 /* Release module */
565 mutex_unlock(&nfsd_mutex);
566 module_put_and_exit(0);
567 return 0;
568 }
569
map_new_errors(u32 vers,__be32 nfserr)570 static __be32 map_new_errors(u32 vers, __be32 nfserr)
571 {
572 if (nfserr == nfserr_jukebox && vers == 2)
573 return nfserr_dropit;
574 if (nfserr == nfserr_wrongsec && vers < 4)
575 return nfserr_acces;
576 return nfserr;
577 }
578
579 int
nfsd_dispatch(struct svc_rqst * rqstp,__be32 * statp)580 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
581 {
582 struct svc_procedure *proc;
583 kxdrproc_t xdr;
584 __be32 nfserr;
585 __be32 *nfserrp;
586
587 dprintk("nfsd_dispatch: vers %d proc %d\n",
588 rqstp->rq_vers, rqstp->rq_proc);
589 proc = rqstp->rq_procinfo;
590
591 /*
592 * Give the xdr decoder a chance to change this if it wants
593 * (necessary in the NFSv4.0 compound case)
594 */
595 rqstp->rq_cachetype = proc->pc_cachetype;
596 /* Decode arguments */
597 xdr = proc->pc_decode;
598 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
599 rqstp->rq_argp)) {
600 dprintk("nfsd: failed to decode arguments!\n");
601 *statp = rpc_garbage_args;
602 return 1;
603 }
604
605 /* Check whether we have this call in the cache. */
606 switch (nfsd_cache_lookup(rqstp)) {
607 case RC_INTR:
608 case RC_DROPIT:
609 return 0;
610 case RC_REPLY:
611 return 1;
612 case RC_DOIT:;
613 /* do it */
614 }
615
616 /* need to grab the location to store the status, as
617 * nfsv4 does some encoding while processing
618 */
619 nfserrp = rqstp->rq_res.head[0].iov_base
620 + rqstp->rq_res.head[0].iov_len;
621 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
622
623 /* Now call the procedure handler, and encode NFS status. */
624 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
625 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
626 if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
627 dprintk("nfsd: Dropping request; may be revisited later\n");
628 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
629 return 0;
630 }
631
632 if (rqstp->rq_proc != 0)
633 *nfserrp++ = nfserr;
634
635 /* Encode result.
636 * For NFSv2, additional info is never returned in case of an error.
637 */
638 if (!(nfserr && rqstp->rq_vers == 2)) {
639 xdr = proc->pc_encode;
640 if (xdr && !xdr(rqstp, nfserrp,
641 rqstp->rq_resp)) {
642 /* Failed to encode result. Release cache entry */
643 dprintk("nfsd: failed to encode result!\n");
644 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
645 *statp = rpc_system_err;
646 return 1;
647 }
648 }
649
650 /* Store reply in cache. */
651 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
652 return 1;
653 }
654
nfsd_pool_stats_open(struct inode * inode,struct file * file)655 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
656 {
657 int ret;
658 mutex_lock(&nfsd_mutex);
659 if (nfsd_serv == NULL) {
660 mutex_unlock(&nfsd_mutex);
661 return -ENODEV;
662 }
663 /* bump up the psudo refcount while traversing */
664 svc_get(nfsd_serv);
665 ret = svc_pool_stats_open(nfsd_serv, file);
666 mutex_unlock(&nfsd_mutex);
667 return ret;
668 }
669
nfsd_pool_stats_release(struct inode * inode,struct file * file)670 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
671 {
672 int ret = seq_release(inode, file);
673 struct net *net = inode->i_sb->s_fs_info;
674
675 mutex_lock(&nfsd_mutex);
676 /* this function really, really should have been called svc_put() */
677 nfsd_destroy(net);
678 mutex_unlock(&nfsd_mutex);
679 return ret;
680 }
681