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