blob: 1697539a7171777815cf6e54bd7738e3e3fdacde [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/nfsd/nfssvc.c
3 *
4 * Central processing for nfsd.
5 *
6 * Authors: Olaf Kirch (okir@monad.swb.de)
7 *
8 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
9 */
10
11#include <linux/config.h>
12#include <linux/module.h>
13
14#include <linux/time.h>
15#include <linux/errno.h>
16#include <linux/nfs.h>
17#include <linux/in.h>
18#include <linux/uio.h>
19#include <linux/unistd.h>
20#include <linux/slab.h>
21#include <linux/smp.h>
22#include <linux/smp_lock.h>
23#include <linux/fs_struct.h>
24
25#include <linux/sunrpc/types.h>
26#include <linux/sunrpc/stats.h>
27#include <linux/sunrpc/svc.h>
28#include <linux/sunrpc/svcsock.h>
29#include <linux/sunrpc/cache.h>
30#include <linux/nfsd/nfsd.h>
31#include <linux/nfsd/stats.h>
32#include <linux/nfsd/cache.h>
33#include <linux/lockd/bind.h>
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +000034#include <linux/nfsacl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
36#define NFSDDBG_FACILITY NFSDDBG_SVC
37
38/* these signals will be delivered to an nfsd thread
39 * when handling a request
40 */
41#define ALLOWED_SIGS (sigmask(SIGKILL))
42/* these signals will be delivered to an nfsd thread
43 * when not handling a request. i.e. when waiting
44 */
45#define SHUTDOWN_SIGS (sigmask(SIGKILL) | sigmask(SIGHUP) | sigmask(SIGINT) | sigmask(SIGQUIT))
46/* if the last thread dies with SIGHUP, then the exports table is
47 * left unchanged ( like 2.4-{0-9} ). Any other signal will clear
48 * the exports table (like 2.2).
49 */
50#define SIG_NOCLEAN SIGHUP
51
52extern struct svc_program nfsd_program;
53static void nfsd(struct svc_rqst *rqstp);
54struct timeval nfssvc_boot;
55static struct svc_serv *nfsd_serv;
56static atomic_t nfsd_busy;
57static unsigned long nfsd_last_call;
58static DEFINE_SPINLOCK(nfsd_call_lock);
59
60struct nfsd_list {
61 struct list_head list;
62 struct task_struct *task;
63};
64static struct list_head nfsd_list = LIST_HEAD_INIT(nfsd_list);
65
66/*
67 * Maximum number of nfsd processes
68 */
69#define NFSD_MAXSERVS 8192
70
71int nfsd_nrthreads(void)
72{
73 if (nfsd_serv == NULL)
74 return 0;
75 else
76 return nfsd_serv->sv_nrthreads;
77}
78
79int
80nfsd_svc(unsigned short port, int nrservs)
81{
82 int error;
83 int none_left;
84 struct list_head *victim;
85
86 lock_kernel();
87 dprintk("nfsd: creating service\n");
88 error = -EINVAL;
89 if (nrservs <= 0)
90 nrservs = 0;
91 if (nrservs > NFSD_MAXSERVS)
92 nrservs = NFSD_MAXSERVS;
93
94 /* Readahead param cache - will no-op if it already exists */
95 error = nfsd_racache_init(2*nrservs);
96 if (error<0)
97 goto out;
NeilBrown76a35502005-06-23 22:03:26 -070098 error = nfs4_state_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 if (error<0)
100 goto out;
101 if (!nfsd_serv) {
102 atomic_set(&nfsd_busy, 0);
103 error = -ENOMEM;
104 nfsd_serv = svc_create(&nfsd_program, NFSD_BUFSIZE);
105 if (nfsd_serv == NULL)
106 goto out;
107 error = svc_makesock(nfsd_serv, IPPROTO_UDP, port);
108 if (error < 0)
109 goto failure;
110
111#ifdef CONFIG_NFSD_TCP
112 error = svc_makesock(nfsd_serv, IPPROTO_TCP, port);
113 if (error < 0)
114 goto failure;
115#endif
116 do_gettimeofday(&nfssvc_boot); /* record boot time */
117 } else
118 nfsd_serv->sv_nrthreads++;
119 nrservs -= (nfsd_serv->sv_nrthreads-1);
120 while (nrservs > 0) {
121 nrservs--;
122 __module_get(THIS_MODULE);
123 error = svc_create_thread(nfsd, nfsd_serv);
124 if (error < 0) {
125 module_put(THIS_MODULE);
126 break;
127 }
128 }
129 victim = nfsd_list.next;
130 while (nrservs < 0 && victim != &nfsd_list) {
131 struct nfsd_list *nl =
132 list_entry(victim,struct nfsd_list, list);
133 victim = victim->next;
134 send_sig(SIG_NOCLEAN, nl->task, 1);
135 nrservs++;
136 }
137 failure:
138 none_left = (nfsd_serv->sv_nrthreads == 1);
139 svc_destroy(nfsd_serv); /* Release server */
140 if (none_left) {
141 nfsd_serv = NULL;
142 nfsd_racache_shutdown();
143 nfs4_state_shutdown();
144 }
145 out:
146 unlock_kernel();
147 return error;
148}
149
150static inline void
151update_thread_usage(int busy_threads)
152{
153 unsigned long prev_call;
154 unsigned long diff;
155 int decile;
156
157 spin_lock(&nfsd_call_lock);
158 prev_call = nfsd_last_call;
159 nfsd_last_call = jiffies;
160 decile = busy_threads*10/nfsdstats.th_cnt;
161 if (decile>0 && decile <= 10) {
162 diff = nfsd_last_call - prev_call;
163 if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
164 nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
165 if (decile == 10)
166 nfsdstats.th_fullcnt++;
167 }
168 spin_unlock(&nfsd_call_lock);
169}
170
171/*
172 * This is the NFS server kernel thread
173 */
174static void
175nfsd(struct svc_rqst *rqstp)
176{
177 struct svc_serv *serv = rqstp->rq_server;
178 struct fs_struct *fsp;
179 int err;
180 struct nfsd_list me;
181 sigset_t shutdown_mask, allowed_mask;
182
183 /* Lock module and set up kernel thread */
184 lock_kernel();
185 daemonize("nfsd");
186
187 /* After daemonize() this kernel thread shares current->fs
188 * with the init process. We need to create files with a
189 * umask of 0 instead of init's umask. */
190 fsp = copy_fs_struct(current->fs);
191 if (!fsp) {
192 printk("Unable to start nfsd thread: out of memory\n");
193 goto out;
194 }
195 exit_fs(current);
196 current->fs = fsp;
197 current->fs->umask = 0;
198
199 siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS);
200 siginitsetinv(&allowed_mask, ALLOWED_SIGS);
201
202 nfsdstats.th_cnt++;
203
204 lockd_up(); /* start lockd */
205
206 me.task = current;
207 list_add(&me.list, &nfsd_list);
208
209 unlock_kernel();
210
211 /*
212 * We want less throttling in balance_dirty_pages() so that nfs to
213 * localhost doesn't cause nfsd to lock up due to all the client's
214 * dirty pages.
215 */
216 current->flags |= PF_LESS_THROTTLE;
217
218 /*
219 * The main request loop
220 */
221 for (;;) {
222 /* Block all but the shutdown signals */
223 sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
224
225 /*
226 * Find a socket with data available and call its
227 * recvfrom routine.
228 */
229 while ((err = svc_recv(serv, rqstp,
230 60*60*HZ)) == -EAGAIN)
231 ;
232 if (err < 0)
233 break;
234 update_thread_usage(atomic_read(&nfsd_busy));
235 atomic_inc(&nfsd_busy);
236
237 /* Lock the export hash tables for reading. */
238 exp_readlock();
239
240 /* Process request with signals blocked. */
241 sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
242
243 svc_process(serv, rqstp);
244
245 /* Unlock export hash tables */
246 exp_readunlock();
247 update_thread_usage(atomic_read(&nfsd_busy));
248 atomic_dec(&nfsd_busy);
249 }
250
251 if (err != -EINTR) {
252 printk(KERN_WARNING "nfsd: terminating on error %d\n", -err);
253 } else {
254 unsigned int signo;
255
256 for (signo = 1; signo <= _NSIG; signo++)
257 if (sigismember(&current->pending.signal, signo) &&
258 !sigismember(&current->blocked, signo))
259 break;
260 err = signo;
261 }
NeilBrown9e4160522005-04-16 15:26:37 -0700262 /* Clear signals before calling lockd_down() and svc_exit_thread() */
263 flush_signals(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
265 lock_kernel();
266
267 /* Release lockd */
268 lockd_down();
269
270 /* Check if this is last thread */
271 if (serv->sv_nrthreads==1) {
272
273 printk(KERN_WARNING "nfsd: last server has exited\n");
274 if (err != SIG_NOCLEAN) {
275 printk(KERN_WARNING "nfsd: unexporting all filesystems\n");
276 nfsd_export_flush();
277 }
278 nfsd_serv = NULL;
279 nfsd_racache_shutdown(); /* release read-ahead cache */
280 nfs4_state_shutdown();
281 }
282 list_del(&me.list);
283 nfsdstats.th_cnt --;
284
285out:
286 /* Release the thread */
287 svc_exit_thread(rqstp);
288
289 /* Release module */
Steven Rostedtc4f92db2005-08-17 14:25:23 -0400290 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 module_put_and_exit(0);
292}
293
294int
295nfsd_dispatch(struct svc_rqst *rqstp, u32 *statp)
296{
297 struct svc_procedure *proc;
298 kxdrproc_t xdr;
299 u32 nfserr;
300 u32 *nfserrp;
301
302 dprintk("nfsd_dispatch: vers %d proc %d\n",
303 rqstp->rq_vers, rqstp->rq_proc);
304 proc = rqstp->rq_procinfo;
305
306 /* Check whether we have this call in the cache. */
307 switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
308 case RC_INTR:
309 case RC_DROPIT:
310 return 0;
311 case RC_REPLY:
312 return 1;
313 case RC_DOIT:;
314 /* do it */
315 }
316
317 /* Decode arguments */
318 xdr = proc->pc_decode;
319 if (xdr && !xdr(rqstp, (u32*)rqstp->rq_arg.head[0].iov_base,
320 rqstp->rq_argp)) {
321 dprintk("nfsd: failed to decode arguments!\n");
322 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
323 *statp = rpc_garbage_args;
324 return 1;
325 }
326
327 /* need to grab the location to store the status, as
328 * nfsv4 does some encoding while processing
329 */
330 nfserrp = rqstp->rq_res.head[0].iov_base
331 + rqstp->rq_res.head[0].iov_len;
332 rqstp->rq_res.head[0].iov_len += sizeof(u32);
333
334 /* Now call the procedure handler, and encode NFS status. */
335 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
336 if (nfserr == nfserr_jukebox && rqstp->rq_vers == 2)
337 nfserr = nfserr_dropit;
338 if (nfserr == nfserr_dropit) {
339 dprintk("nfsd: Dropping request due to malloc failure!\n");
340 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
341 return 0;
342 }
343
344 if (rqstp->rq_proc != 0)
345 *nfserrp++ = nfserr;
346
347 /* Encode result.
348 * For NFSv2, additional info is never returned in case of an error.
349 */
350 if (!(nfserr && rqstp->rq_vers == 2)) {
351 xdr = proc->pc_encode;
352 if (xdr && !xdr(rqstp, nfserrp,
353 rqstp->rq_resp)) {
354 /* Failed to encode result. Release cache entry */
355 dprintk("nfsd: failed to encode result!\n");
356 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
357 *statp = rpc_system_err;
358 return 1;
359 }
360 }
361
362 /* Store reply in cache. */
363 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
364 return 1;
365}
366
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +0000367#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
368static struct svc_stat nfsd_acl_svcstats;
369static struct svc_version * nfsd_acl_version[] = {
370 [2] = &nfsd_acl_version2,
371 [3] = &nfsd_acl_version3,
372};
373
374#define NFSD_ACL_NRVERS (sizeof(nfsd_acl_version)/sizeof(nfsd_acl_version[0]))
375static struct svc_program nfsd_acl_program = {
376 .pg_prog = NFS_ACL_PROGRAM,
377 .pg_nvers = NFSD_ACL_NRVERS,
378 .pg_vers = nfsd_acl_version,
379 .pg_name = "nfsd",
Andreas Gruenbacher21348422005-06-22 17:16:28 +0000380 .pg_class = "nfsd",
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +0000381 .pg_stats = &nfsd_acl_svcstats,
382};
383
384static struct svc_stat nfsd_acl_svcstats = {
385 .program = &nfsd_acl_program,
386};
387
388#define nfsd_acl_program_p &nfsd_acl_program
389#else
390#define nfsd_acl_program_p NULL
391#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
392
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393extern struct svc_version nfsd_version2, nfsd_version3, nfsd_version4;
394
395static struct svc_version * nfsd_version[] = {
396 [2] = &nfsd_version2,
397#if defined(CONFIG_NFSD_V3)
398 [3] = &nfsd_version3,
399#endif
400#if defined(CONFIG_NFSD_V4)
401 [4] = &nfsd_version4,
402#endif
403};
404
405#define NFSD_NRVERS (sizeof(nfsd_version)/sizeof(nfsd_version[0]))
406struct svc_program nfsd_program = {
Andreas Gruenbachera257cdd2005-06-22 17:16:26 +0000407 .pg_next = nfsd_acl_program_p,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 .pg_prog = NFS_PROGRAM, /* program number */
409 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
410 .pg_vers = nfsd_version, /* version table */
411 .pg_name = "nfsd", /* program name */
412 .pg_class = "nfsd", /* authentication class */
413 .pg_stats = &nfsd_svcstats, /* version table */
414 .pg_authenticate = &svc_set_client, /* export authentication */
415
416};