License cleanup: add SPDX GPL-2.0 license identifier to files with no license
[linux-2.6-block.git] / fs / nfsd / nfssvc.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
1da177e4 2/*
1da177e4
LT
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
3f07c014 10#include <linux/sched/signal.h>
83144186 11#include <linux/freezer.h>
143cb494 12#include <linux/module.h>
1da177e4 13#include <linux/fs_struct.h>
c3d06f9c 14#include <linux/swap.h>
1da177e4 15
1da177e4 16#include <linux/sunrpc/stats.h>
1da177e4 17#include <linux/sunrpc/svcsock.h>
36684996 18#include <linux/sunrpc/svc_xprt.h>
1da177e4 19#include <linux/lockd/bind.h>
a257cdd0 20#include <linux/nfsacl.h>
ed2d8aed 21#include <linux/seq_file.h>
36684996
SM
22#include <linux/inetdevice.h>
23#include <net/addrconf.h>
24#include <net/ipv6.h>
fc5d00b0 25#include <net/net_namespace.h>
9a74af21
BH
26#include "nfsd.h"
27#include "cache.h"
0a3adade 28#include "vfs.h"
2c2fe290 29#include "netns.h"
1da177e4
LT
30
31#define NFSDDBG_FACILITY NFSDDBG_SVC
32
1da177e4 33extern struct svc_program nfsd_program;
9867d76c 34static int nfsd(void *vrqstp);
1da177e4 35
bedbdd8b 36/*
9dd9845f 37 * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members
bedbdd8b
NB
38 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
39 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
40 *
9dd9845f 41 * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a
bedbdd8b
NB
42 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
43 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
44 * entry of ->sv_pools[].
45 *
46 * Transitions of the thread count between zero and non-zero are of particular
47 * interest since the svc_serv needs to be created and initialized at that
48 * point, or freed.
3dd98a3b
JL
49 *
50 * Finally, the nfsd_mutex also protects some of the global variables that are
51 * accessed when nfsd starts and that are settable via the write_* routines in
52 * nfsctl.c. In particular:
53 *
54 * user_recovery_dirname
55 * user_lease_time
56 * nfsd_versions
bedbdd8b
NB
57 */
58DEFINE_MUTEX(nfsd_mutex);
bedbdd8b 59
4bd9b0f4
AA
60/*
61 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
62 * nfsd_drc_max_pages limits the total amount of memory available for
63 * version 4.1 DRC caches.
64 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
65 */
66spinlock_t nfsd_drc_lock;
697ce9be
ZY
67unsigned long nfsd_drc_max_mem;
68unsigned long nfsd_drc_mem_used;
4bd9b0f4 69
3fb803a9
AG
70#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
71static struct svc_stat nfsd_acl_svcstats;
e9679189 72static const struct svc_version *nfsd_acl_version[] = {
3fb803a9
AG
73 [2] = &nfsd_acl_version2,
74 [3] = &nfsd_acl_version3,
75};
76
77#define NFSD_ACL_MINVERS 2
e8c96f8c 78#define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
e9679189 79static const struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
3fb803a9
AG
80
81static struct svc_program nfsd_acl_program = {
82 .pg_prog = NFS_ACL_PROGRAM,
83 .pg_nvers = NFSD_ACL_NRVERS,
84 .pg_vers = nfsd_acl_versions,
1a8eff6d 85 .pg_name = "nfsacl",
3fb803a9
AG
86 .pg_class = "nfsd",
87 .pg_stats = &nfsd_acl_svcstats,
88 .pg_authenticate = &svc_set_client,
89};
90
91static struct svc_stat nfsd_acl_svcstats = {
92 .program = &nfsd_acl_program,
93};
94#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
95
e9679189 96static const struct svc_version *nfsd_version[] = {
70c3b76c
N
97 [2] = &nfsd_version2,
98#if defined(CONFIG_NFSD_V3)
99 [3] = &nfsd_version3,
100#endif
101#if defined(CONFIG_NFSD_V4)
102 [4] = &nfsd_version4,
103#endif
104};
105
106#define NFSD_MINVERS 2
e8c96f8c 107#define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
e9679189 108static const struct svc_version *nfsd_versions[NFSD_NRVERS];
70c3b76c
N
109
110struct svc_program nfsd_program = {
3fb803a9
AG
111#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
112 .pg_next = &nfsd_acl_program,
113#endif
70c3b76c
N
114 .pg_prog = NFS_PROGRAM, /* program number */
115 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
116 .pg_vers = nfsd_versions, /* version table */
117 .pg_name = "nfsd", /* program name */
118 .pg_class = "nfsd", /* authentication class */
119 .pg_stats = &nfsd_svcstats, /* version table */
120 .pg_authenticate = &svc_set_client, /* export authentication */
121
122};
123
35f7a14f
BF
124static bool nfsd_supported_minorversions[NFSD_SUPPORTED_MINOR_VERSION + 1] = {
125 [0] = 1,
126 [1] = 1,
c23ae601 127 [2] = 1,
35f7a14f 128};
8daf220a 129
6658d3a7
N
130int nfsd_vers(int vers, enum vers_op change)
131{
132 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
15ddb4ae 133 return 0;
6658d3a7
N
134 switch(change) {
135 case NFSD_SET:
136 nfsd_versions[vers] = nfsd_version[vers];
6658d3a7
N
137#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
138 if (vers < NFSD_ACL_NRVERS)
1a8eff6d 139 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
6658d3a7 140#endif
1a8eff6d 141 break;
6658d3a7
N
142 case NFSD_CLEAR:
143 nfsd_versions[vers] = NULL;
144#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
145 if (vers < NFSD_ACL_NRVERS)
1a8eff6d 146 nfsd_acl_versions[vers] = NULL;
6658d3a7
N
147#endif
148 break;
149 case NFSD_TEST:
150 return nfsd_versions[vers] != NULL;
151 case NFSD_AVAIL:
152 return nfsd_version[vers] != NULL;
153 }
154 return 0;
155}
8daf220a 156
d3635ff0
TM
157static void
158nfsd_adjust_nfsd_versions4(void)
159{
160 unsigned i;
161
162 for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
163 if (nfsd_supported_minorversions[i])
164 return;
165 }
166 nfsd_vers(4, NFSD_CLEAR);
167}
168
8daf220a
BH
169int nfsd_minorversion(u32 minorversion, enum vers_op change)
170{
928c6fb3
N
171 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
172 change != NFSD_AVAIL)
8daf220a
BH
173 return -1;
174 switch(change) {
175 case NFSD_SET:
35f7a14f 176 nfsd_supported_minorversions[minorversion] = true;
d3635ff0 177 nfsd_vers(4, NFSD_SET);
8daf220a
BH
178 break;
179 case NFSD_CLEAR:
35f7a14f 180 nfsd_supported_minorversions[minorversion] = false;
d3635ff0 181 nfsd_adjust_nfsd_versions4();
8daf220a
BH
182 break;
183 case NFSD_TEST:
35f7a14f 184 return nfsd_supported_minorversions[minorversion];
8daf220a
BH
185 case NFSD_AVAIL:
186 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
187 }
188 return 0;
189}
190
1da177e4
LT
191/*
192 * Maximum number of nfsd processes
193 */
194#define NFSD_MAXSERVS 8192
195
9dd9845f 196int nfsd_nrthreads(struct net *net)
1da177e4 197{
c7d106c9 198 int rv = 0;
9dd9845f
SK
199 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
200
c7d106c9 201 mutex_lock(&nfsd_mutex);
9dd9845f
SK
202 if (nn->nfsd_serv)
203 rv = nn->nfsd_serv->sv_nrthreads;
c7d106c9
NB
204 mutex_unlock(&nfsd_mutex);
205 return rv;
1da177e4
LT
206}
207
db6e182c 208static int nfsd_init_socks(struct net *net)
59db4a0c
BF
209{
210 int error;
9dd9845f
SK
211 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
212
213 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
59db4a0c
BF
214 return 0;
215
9dd9845f 216 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
59db4a0c
BF
217 SVC_SOCK_DEFAULTS);
218 if (error < 0)
219 return error;
220
9dd9845f 221 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
59db4a0c
BF
222 SVC_SOCK_DEFAULTS);
223 if (error < 0)
224 return error;
225
226 return 0;
227}
228
4539f149 229static int nfsd_users = 0;
4ad9a344 230
bda9cac1
SK
231static int nfsd_startup_generic(int nrservs)
232{
233 int ret;
234
4539f149 235 if (nfsd_users++)
bda9cac1
SK
236 return 0;
237
238 /*
239 * Readahead param cache - will no-op if it already exists.
240 * (Note therefore results will be suboptimal if number of
241 * threads is modified after nfsd start.)
242 */
243 ret = nfsd_racache_init(2*nrservs);
244 if (ret)
d9499a95
KM
245 goto dec_users;
246
bda9cac1
SK
247 ret = nfs4_state_start();
248 if (ret)
249 goto out_racache;
250 return 0;
251
252out_racache:
253 nfsd_racache_shutdown();
d9499a95
KM
254dec_users:
255 nfsd_users--;
bda9cac1
SK
256 return ret;
257}
258
259static void nfsd_shutdown_generic(void)
260{
4539f149
SK
261 if (--nfsd_users)
262 return;
263
bda9cac1
SK
264 nfs4_state_shutdown();
265 nfsd_racache_shutdown();
266}
267
8ef66714
KM
268static bool nfsd_needs_lockd(void)
269{
ff88825f 270#if defined(CONFIG_NFSD_V3)
8ef66714 271 return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
ff88825f
KM
272#else
273 return (nfsd_versions[2] != NULL);
274#endif
8ef66714
KM
275}
276
903d9bf0 277static int nfsd_startup_net(int nrservs, struct net *net)
6ff50b3d 278{
2c2fe290 279 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6ff50b3d
SK
280 int ret;
281
2c2fe290
SK
282 if (nn->nfsd_net_up)
283 return 0;
284
903d9bf0 285 ret = nfsd_startup_generic(nrservs);
6ff50b3d
SK
286 if (ret)
287 return ret;
903d9bf0
SK
288 ret = nfsd_init_socks(net);
289 if (ret)
290 goto out_socks;
8ef66714
KM
291
292 if (nfsd_needs_lockd() && !nn->lockd_up) {
293 ret = lockd_up(net);
294 if (ret)
295 goto out_socks;
296 nn->lockd_up = 1;
297 }
298
6ff50b3d
SK
299 ret = nfs4_state_start_net(net);
300 if (ret)
301 goto out_lockd;
302
2c2fe290 303 nn->nfsd_net_up = true;
6ff50b3d
SK
304 return 0;
305
306out_lockd:
8ef66714
KM
307 if (nn->lockd_up) {
308 lockd_down(net);
309 nn->lockd_up = 0;
310 }
903d9bf0 311out_socks:
bda9cac1 312 nfsd_shutdown_generic();
4ad9a344
JL
313 return ret;
314}
315
6ff50b3d
SK
316static void nfsd_shutdown_net(struct net *net)
317{
2c2fe290
SK
318 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
319
6ff50b3d 320 nfs4_state_shutdown_net(net);
8ef66714
KM
321 if (nn->lockd_up) {
322 lockd_down(net);
323 nn->lockd_up = 0;
324 }
2c2fe290 325 nn->nfsd_net_up = false;
903d9bf0 326 nfsd_shutdown_generic();
6ff50b3d
SK
327}
328
36684996
SM
329static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
330 void *ptr)
331{
332 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
333 struct net_device *dev = ifa->ifa_dev->dev;
334 struct net *net = dev_net(dev);
335 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
336 struct sockaddr_in sin;
337
338 if (event != NETDEV_DOWN)
339 goto out;
340
341 if (nn->nfsd_serv) {
342 dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
343 sin.sin_family = AF_INET;
344 sin.sin_addr.s_addr = ifa->ifa_local;
345 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
346 }
347
348out:
349 return NOTIFY_DONE;
350}
351
352static struct notifier_block nfsd_inetaddr_notifier = {
353 .notifier_call = nfsd_inetaddr_event,
354};
355
356#if IS_ENABLED(CONFIG_IPV6)
357static int nfsd_inet6addr_event(struct notifier_block *this,
358 unsigned long event, void *ptr)
359{
360 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
361 struct net_device *dev = ifa->idev->dev;
362 struct net *net = dev_net(dev);
363 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
364 struct sockaddr_in6 sin6;
365
366 if (event != NETDEV_DOWN)
367 goto out;
368
369 if (nn->nfsd_serv) {
370 dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
371 sin6.sin6_family = AF_INET6;
372 sin6.sin6_addr = ifa->addr;
7b19824d
SM
373 if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
374 sin6.sin6_scope_id = ifa->idev->dev->ifindex;
36684996
SM
375 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
376 }
377
378out:
379 return NOTIFY_DONE;
380}
381
382static struct notifier_block nfsd_inet6addr_notifier = {
383 .notifier_call = nfsd_inet6addr_event,
384};
385#endif
386
1eca45f8
VA
387/* Only used under nfsd_mutex, so this atomic may be overkill: */
388static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
389
541e864f 390static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
4ad9a344 391{
903d9bf0
SK
392 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
393
1eca45f8
VA
394 /* check if the notifier still has clients */
395 if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
396 unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
36684996 397#if IS_ENABLED(CONFIG_IPV6)
1eca45f8 398 unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
36684996 399#endif
1eca45f8
VA
400 }
401
4ad9a344
JL
402 /*
403 * write_ports can create the server without actually starting
404 * any threads--if we get shut down before any threads are
405 * started, then nfsd_last_thread will be run before any of this
691412b4 406 * other initialization has been done except the rpcb information.
4ad9a344 407 */
691412b4 408 svc_rpcb_cleanup(serv, net);
903d9bf0 409 if (!nn->nfsd_net_up)
4ad9a344 410 return;
16d05870 411
691412b4 412 nfsd_shutdown_net(net);
e096bbc6
JL
413 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
414 "cache\n");
b3853e0e 415 nfsd_export_flush(net);
bc591ccf 416}
6658d3a7
N
417
418void nfsd_reset_versions(void)
419{
6658d3a7
N
420 int i;
421
800a938f
N
422 for (i = 0; i < NFSD_NRVERS; i++)
423 if (nfsd_vers(i, NFSD_TEST))
424 return;
6658d3a7 425
800a938f
N
426 for (i = 0; i < NFSD_NRVERS; i++)
427 if (i != 4)
428 nfsd_vers(i, NFSD_SET);
429 else {
430 int minor = 0;
431 while (nfsd_minorversion(minor, NFSD_SET) >= 0)
432 minor++;
433 }
6658d3a7
N
434}
435
c3d06f9c
AA
436/*
437 * Each session guarantees a negotiated per slot memory cache for replies
438 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
439 * NFSv4.1 server might want to use more memory for a DRC than a machine
440 * with mutiple services.
441 *
442 * Impose a hard limit on the number of pages for the DRC which varies
443 * according to the machines free pages. This is of course only a default.
444 *
445 * For now this is a #defined shift which could be under admin control
446 * in the future.
447 */
448static void set_max_drc(void)
449{
6a14dd1a 450 #define NFSD_DRC_SIZE_SHIFT 10
0c193054
AA
451 nfsd_drc_max_mem = (nr_free_buffer_pages()
452 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
453 nfsd_drc_mem_used = 0;
4bd9b0f4 454 spin_lock_init(&nfsd_drc_lock);
697ce9be 455 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
c3d06f9c 456}
bedbdd8b 457
87b0fc7d 458static int nfsd_get_default_max_blksize(void)
02a375f0 459{
87b0fc7d
BF
460 struct sysinfo i;
461 unsigned long long target;
462 unsigned long ret;
bedbdd8b 463
87b0fc7d 464 si_meminfo(&i);
508f9227 465 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
87b0fc7d
BF
466 /*
467 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
468 * machines, but only uses 32K on 128M machines. Bottom out at
469 * 8K on 32M and smaller. Of course, this is only a default.
470 */
471 target >>= 12;
472
473 ret = NFSSVC_MAXBLKSIZE;
474 while (ret > target && ret >= 8*1024*2)
475 ret /= 2;
476 return ret;
477}
478
afea5657 479static const struct svc_serv_ops nfsd_thread_sv_ops = {
b9e13cdf
JL
480 .svo_shutdown = nfsd_last_thread,
481 .svo_function = nfsd,
482 .svo_enqueue_xprt = svc_xprt_do_enqueue,
598e2359 483 .svo_setup = svc_set_num_threads,
b9e13cdf 484 .svo_module = THIS_MODULE,
ea126e74
JL
485};
486
6777436b 487int nfsd_create_serv(struct net *net)
87b0fc7d 488{
9793f7c8 489 int error;
b9c0ef85 490 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
9793f7c8 491
bedbdd8b 492 WARN_ON(!mutex_is_locked(&nfsd_mutex));
9dd9845f
SK
493 if (nn->nfsd_serv) {
494 svc_get(nn->nfsd_serv);
02a375f0
N
495 return 0;
496 }
87b0fc7d
BF
497 if (nfsd_max_blksize == 0)
498 nfsd_max_blksize = nfsd_get_default_max_blksize();
e844a7b9 499 nfsd_reset_versions();
9dd9845f 500 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
b9e13cdf 501 &nfsd_thread_sv_ops);
9dd9845f 502 if (nn->nfsd_serv == NULL)
628b3687 503 return -ENOMEM;
bedbdd8b 504
5b8db00b 505 nn->nfsd_serv->sv_maxconn = nn->max_connections;
9dd9845f 506 error = svc_bind(nn->nfsd_serv, net);
9793f7c8 507 if (error < 0) {
9dd9845f 508 svc_destroy(nn->nfsd_serv);
9793f7c8
SK
509 return error;
510 }
511
628b3687 512 set_max_drc();
1eca45f8
VA
513 /* check if the notifier is already set */
514 if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
515 register_inetaddr_notifier(&nfsd_inetaddr_notifier);
36684996 516#if IS_ENABLED(CONFIG_IPV6)
1eca45f8 517 register_inet6addr_notifier(&nfsd_inet6addr_notifier);
36684996 518#endif
1eca45f8 519 }
b9c0ef85 520 do_gettimeofday(&nn->nfssvc_boot); /* record boot time */
87b0fc7d 521 return 0;
02a375f0
N
522}
523
9dd9845f 524int nfsd_nrpools(struct net *net)
eed2965a 525{
9dd9845f
SK
526 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
527
528 if (nn->nfsd_serv == NULL)
eed2965a
GB
529 return 0;
530 else
9dd9845f 531 return nn->nfsd_serv->sv_nrpools;
eed2965a
GB
532}
533
9dd9845f 534int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
eed2965a
GB
535{
536 int i = 0;
9dd9845f 537 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
eed2965a 538
9dd9845f
SK
539 if (nn->nfsd_serv != NULL) {
540 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
541 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
eed2965a
GB
542 }
543
544 return 0;
545}
546
9dd9845f
SK
547void nfsd_destroy(struct net *net)
548{
549 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
550 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
551
552 if (destroy)
553 svc_shutdown_net(nn->nfsd_serv, net);
554 svc_destroy(nn->nfsd_serv);
555 if (destroy)
556 nn->nfsd_serv = NULL;
557}
558
3938a0d5 559int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
eed2965a
GB
560{
561 int i = 0;
562 int tot = 0;
563 int err = 0;
9dd9845f 564 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
eed2965a 565
bedbdd8b
NB
566 WARN_ON(!mutex_is_locked(&nfsd_mutex));
567
9dd9845f 568 if (nn->nfsd_serv == NULL || n <= 0)
eed2965a
GB
569 return 0;
570
9dd9845f
SK
571 if (n > nn->nfsd_serv->sv_nrpools)
572 n = nn->nfsd_serv->sv_nrpools;
eed2965a
GB
573
574 /* enforce a global maximum number of threads */
575 tot = 0;
576 for (i = 0; i < n; i++) {
3c7aa15d 577 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
eed2965a
GB
578 tot += nthreads[i];
579 }
580 if (tot > NFSD_MAXSERVS) {
581 /* total too large: scale down requested numbers */
582 for (i = 0; i < n && tot > 0; i++) {
583 int new = nthreads[i] * NFSD_MAXSERVS / tot;
584 tot -= (nthreads[i] - new);
585 nthreads[i] = new;
586 }
587 for (i = 0; i < n && tot > 0; i++) {
588 nthreads[i]--;
589 tot--;
590 }
591 }
592
593 /*
594 * There must always be a thread in pool 0; the admin
595 * can't shut down NFS completely using pool_threads.
596 */
597 if (nthreads[0] == 0)
598 nthreads[0] = 1;
599
600 /* apply the new numbers */
9dd9845f 601 svc_get(nn->nfsd_serv);
eed2965a 602 for (i = 0; i < n; i++) {
598e2359
JL
603 err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
604 &nn->nfsd_serv->sv_pools[i], nthreads[i]);
eed2965a
GB
605 if (err)
606 break;
607 }
19f7e2ca 608 nfsd_destroy(net);
eed2965a
GB
609 return err;
610}
611
ac77efbe
JL
612/*
613 * Adjust the number of threads and return the new number of threads.
614 * This is also the function that starts the server if necessary, if
615 * this is the first time nrservs is nonzero.
616 */
1da177e4 617int
d41a9417 618nfsd_svc(int nrservs, struct net *net)
1da177e4
LT
619{
620 int error;
774f8bbd 621 bool nfsd_up_before;
9dd9845f 622 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
bedbdd8b
NB
623
624 mutex_lock(&nfsd_mutex);
6658d3a7 625 dprintk("nfsd: creating service\n");
3c7aa15d
KM
626
627 nrservs = max(nrservs, 0);
628 nrservs = min(nrservs, NFSD_MAXSERVS);
671e1fcf 629 error = 0;
3c7aa15d 630
9dd9845f 631 if (nrservs == 0 && nn->nfsd_serv == NULL)
671e1fcf
N
632 goto out;
633
6777436b 634 error = nfsd_create_serv(net);
02a375f0 635 if (error)
774f8bbd
BF
636 goto out;
637
903d9bf0 638 nfsd_up_before = nn->nfsd_net_up;
774f8bbd 639
903d9bf0 640 error = nfsd_startup_net(nrservs, net);
af4718f3
BF
641 if (error)
642 goto out_destroy;
598e2359
JL
643 error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
644 NULL, nrservs);
774f8bbd
BF
645 if (error)
646 goto out_shutdown;
9dd9845f 647 /* We are holding a reference to nn->nfsd_serv which
af4718f3
BF
648 * we don't want to count in the return value,
649 * so subtract 1
650 */
9dd9845f 651 error = nn->nfsd_serv->sv_nrthreads - 1;
4ad9a344 652out_shutdown:
774f8bbd 653 if (error < 0 && !nfsd_up_before)
541e864f 654 nfsd_shutdown_net(net);
774f8bbd 655out_destroy:
19f7e2ca 656 nfsd_destroy(net); /* Release server */
4ad9a344 657out:
bedbdd8b 658 mutex_unlock(&nfsd_mutex);
1da177e4
LT
659 return error;
660}
661
1da177e4
LT
662
663/*
664 * This is the NFS server kernel thread
665 */
9867d76c
JL
666static int
667nfsd(void *vrqstp)
1da177e4 668{
9867d76c 669 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
88c47666
SK
670 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
671 struct net *net = perm_sock->xpt_net;
5b8db00b 672 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5b444cc9 673 int err;
1da177e4
LT
674
675 /* Lock module and set up kernel thread */
bedbdd8b 676 mutex_lock(&nfsd_mutex);
1da177e4 677
9867d76c 678 /* At this point, the thread shares current->fs
47057abd
AG
679 * with the init process. We need to create files with the
680 * umask as defined by the client instead of init's umask. */
3e93cd67 681 if (unshare_fs_struct() < 0) {
1da177e4
LT
682 printk("Unable to start nfsd thread: out of memory\n");
683 goto out;
684 }
3e93cd67 685
1da177e4
LT
686 current->fs->umask = 0;
687
9867d76c
JL
688 /*
689 * thread is spawned with all signals set to SIG_IGN, re-enable
100766f8 690 * the ones that will bring down the thread
9867d76c 691 */
100766f8
JL
692 allow_signal(SIGKILL);
693 allow_signal(SIGHUP);
694 allow_signal(SIGINT);
695 allow_signal(SIGQUIT);
bedbdd8b 696
1da177e4 697 nfsdstats.th_cnt++;
bedbdd8b
NB
698 mutex_unlock(&nfsd_mutex);
699
83144186 700 set_freezable();
1da177e4
LT
701
702 /*
703 * The main request loop
704 */
705 for (;;) {
5b8db00b
JL
706 /* Update sv_maxconn if it has changed */
707 rqstp->rq_server->sv_maxconn = nn->max_connections;
708
1da177e4
LT
709 /*
710 * Find a socket with data available and call its
711 * recvfrom routine.
712 */
6fb2b47f 713 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
1da177e4 714 ;
9867d76c 715 if (err == -EINTR)
1da177e4 716 break;
e0e81739 717 validate_process_creds();
6fb2b47f 718 svc_process(rqstp);
e0e81739 719 validate_process_creds();
1da177e4
LT
720 }
721
24e36663 722 /* Clear signals before calling svc_exit_thread() */
9e416052 723 flush_signals(current);
1da177e4 724
bedbdd8b 725 mutex_lock(&nfsd_mutex);
1da177e4
LT
726 nfsdstats.th_cnt --;
727
728out:
57c8b13e 729 rqstp->rq_server = NULL;
786185b5 730
1da177e4
LT
731 /* Release the thread */
732 svc_exit_thread(rqstp);
733
88c47666 734 nfsd_destroy(net);
57c8b13e 735
1da177e4 736 /* Release module */
bedbdd8b 737 mutex_unlock(&nfsd_mutex);
1da177e4 738 module_put_and_exit(0);
9867d76c 739 return 0;
1da177e4
LT
740}
741
32c1eb0c
AA
742static __be32 map_new_errors(u32 vers, __be32 nfserr)
743{
744 if (nfserr == nfserr_jukebox && vers == 2)
745 return nfserr_dropit;
746 if (nfserr == nfserr_wrongsec && vers < 4)
747 return nfserr_acces;
748 return nfserr;
749}
750
e6838a29
BF
751/*
752 * A write procedure can have a large argument, and a read procedure can
753 * have a large reply, but no NFSv2 or NFSv3 procedure has argument and
754 * reply that can both be larger than a page. The xdr code has taken
755 * advantage of this assumption to be a sloppy about bounds checking in
756 * some cases. Pending a rewrite of the NFSv2/v3 xdr code to fix that
757 * problem, we enforce these assumptions here:
758 */
759static bool nfs_request_too_big(struct svc_rqst *rqstp,
860bda29 760 const struct svc_procedure *proc)
e6838a29
BF
761{
762 /*
763 * The ACL code has more careful bounds-checking and is not
764 * susceptible to this problem:
765 */
766 if (rqstp->rq_prog != NFS_PROGRAM)
767 return false;
768 /*
769 * Ditto NFSv4 (which can in theory have argument and reply both
770 * more than a page):
771 */
772 if (rqstp->rq_vers >= 4)
773 return false;
774 /* The reply will be small, we're OK: */
775 if (proc->pc_xdrressize > 0 &&
776 proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE))
777 return false;
778
779 return rqstp->rq_arg.len > PAGE_SIZE;
780}
781
1da177e4 782int
c7afef1f 783nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1da177e4 784{
860bda29 785 const struct svc_procedure *proc;
ad451d38
AV
786 __be32 nfserr;
787 __be32 *nfserrp;
1da177e4
LT
788
789 dprintk("nfsd_dispatch: vers %d proc %d\n",
790 rqstp->rq_vers, rqstp->rq_proc);
791 proc = rqstp->rq_procinfo;
792
e6838a29
BF
793 if (nfs_request_too_big(rqstp, proc)) {
794 dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers);
795 *statp = rpc_garbage_args;
796 return 1;
797 }
1091006c
BF
798 /*
799 * Give the xdr decoder a chance to change this if it wants
800 * (necessary in the NFSv4.0 compound case)
801 */
802 rqstp->rq_cachetype = proc->pc_cachetype;
803 /* Decode arguments */
026fec7e
CH
804 if (proc->pc_decode &&
805 !proc->pc_decode(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base)) {
1091006c
BF
806 dprintk("nfsd: failed to decode arguments!\n");
807 *statp = rpc_garbage_args;
808 return 1;
809 }
810
1da177e4 811 /* Check whether we have this call in the cache. */
1091006c 812 switch (nfsd_cache_lookup(rqstp)) {
1da177e4
LT
813 case RC_DROPIT:
814 return 0;
815 case RC_REPLY:
816 return 1;
817 case RC_DOIT:;
818 /* do it */
819 }
820
1da177e4
LT
821 /* need to grab the location to store the status, as
822 * nfsv4 does some encoding while processing
823 */
824 nfserrp = rqstp->rq_res.head[0].iov_base
825 + rqstp->rq_res.head[0].iov_len;
ad451d38 826 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
1da177e4
LT
827
828 /* Now call the procedure handler, and encode NFS status. */
a6beb732 829 nfserr = proc->pc_func(rqstp);
32c1eb0c 830 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
78b65eb3 831 if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
45457e09 832 dprintk("nfsd: Dropping request; may be revisited later\n");
1da177e4
LT
833 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
834 return 0;
835 }
836
837 if (rqstp->rq_proc != 0)
838 *nfserrp++ = nfserr;
839
840 /* Encode result.
841 * For NFSv2, additional info is never returned in case of an error.
842 */
843 if (!(nfserr && rqstp->rq_vers == 2)) {
63f8de37 844 if (proc->pc_encode && !proc->pc_encode(rqstp, nfserrp)) {
1da177e4
LT
845 /* Failed to encode result. Release cache entry */
846 dprintk("nfsd: failed to encode result!\n");
847 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
848 *statp = rpc_system_err;
849 return 1;
850 }
851 }
852
853 /* Store reply in cache. */
57d276d7 854 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
1da177e4
LT
855 return 1;
856}
03cf6c9f
GB
857
858int nfsd_pool_stats_open(struct inode *inode, struct file *file)
859{
ed2d8aed 860 int ret;
11f77942 861 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
9dd9845f 862
ed2d8aed 863 mutex_lock(&nfsd_mutex);
9dd9845f 864 if (nn->nfsd_serv == NULL) {
ed2d8aed 865 mutex_unlock(&nfsd_mutex);
03cf6c9f 866 return -ENODEV;
ed2d8aed
RY
867 }
868 /* bump up the psudo refcount while traversing */
9dd9845f
SK
869 svc_get(nn->nfsd_serv);
870 ret = svc_pool_stats_open(nn->nfsd_serv, file);
ed2d8aed
RY
871 mutex_unlock(&nfsd_mutex);
872 return ret;
873}
874
875int nfsd_pool_stats_release(struct inode *inode, struct file *file)
876{
877 int ret = seq_release(inode, file);
11f77942 878 struct net *net = inode->i_sb->s_fs_info;
786185b5 879
ed2d8aed
RY
880 mutex_lock(&nfsd_mutex);
881 /* this function really, really should have been called svc_put() */
19f7e2ca 882 nfsd_destroy(net);
ed2d8aed
RY
883 mutex_unlock(&nfsd_mutex);
884 return ret;
03cf6c9f 885}