Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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> | |
a257cdd0 | 34 | #include <linux/nfsacl.h> |
1da177e4 LT |
35 | |
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 | ||
52 | extern struct svc_program nfsd_program; | |
53 | static void nfsd(struct svc_rqst *rqstp); | |
54 | struct timeval nfssvc_boot; | |
55 | static struct svc_serv *nfsd_serv; | |
56 | static atomic_t nfsd_busy; | |
57 | static unsigned long nfsd_last_call; | |
58 | static DEFINE_SPINLOCK(nfsd_call_lock); | |
59 | ||
60 | struct nfsd_list { | |
61 | struct list_head list; | |
62 | struct task_struct *task; | |
63 | }; | |
64 | static 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 | ||
71 | int nfsd_nrthreads(void) | |
72 | { | |
73 | if (nfsd_serv == NULL) | |
74 | return 0; | |
75 | else | |
76 | return nfsd_serv->sv_nrthreads; | |
77 | } | |
78 | ||
79 | int | |
80 | nfsd_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; | |
76a3550e | 98 | error = nfs4_state_start(); |
1da177e4 LT |
99 | 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 | ||
150 | static inline void | |
151 | update_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 | */ | |
174 | static void | |
175 | nfsd(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(¤t->pending.signal, signo) && | |
258 | !sigismember(¤t->blocked, signo)) | |
259 | break; | |
260 | err = signo; | |
261 | } | |
9e416052 N |
262 | /* Clear signals before calling lockd_down() and svc_exit_thread() */ |
263 | flush_signals(current); | |
1da177e4 LT |
264 | |
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 | ||
285 | out: | |
286 | /* Release the thread */ | |
287 | svc_exit_thread(rqstp); | |
288 | ||
289 | /* Release module */ | |
290 | module_put_and_exit(0); | |
291 | } | |
292 | ||
293 | int | |
294 | nfsd_dispatch(struct svc_rqst *rqstp, u32 *statp) | |
295 | { | |
296 | struct svc_procedure *proc; | |
297 | kxdrproc_t xdr; | |
298 | u32 nfserr; | |
299 | u32 *nfserrp; | |
300 | ||
301 | dprintk("nfsd_dispatch: vers %d proc %d\n", | |
302 | rqstp->rq_vers, rqstp->rq_proc); | |
303 | proc = rqstp->rq_procinfo; | |
304 | ||
305 | /* Check whether we have this call in the cache. */ | |
306 | switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) { | |
307 | case RC_INTR: | |
308 | case RC_DROPIT: | |
309 | return 0; | |
310 | case RC_REPLY: | |
311 | return 1; | |
312 | case RC_DOIT:; | |
313 | /* do it */ | |
314 | } | |
315 | ||
316 | /* Decode arguments */ | |
317 | xdr = proc->pc_decode; | |
318 | if (xdr && !xdr(rqstp, (u32*)rqstp->rq_arg.head[0].iov_base, | |
319 | rqstp->rq_argp)) { | |
320 | dprintk("nfsd: failed to decode arguments!\n"); | |
321 | nfsd_cache_update(rqstp, RC_NOCACHE, NULL); | |
322 | *statp = rpc_garbage_args; | |
323 | return 1; | |
324 | } | |
325 | ||
326 | /* need to grab the location to store the status, as | |
327 | * nfsv4 does some encoding while processing | |
328 | */ | |
329 | nfserrp = rqstp->rq_res.head[0].iov_base | |
330 | + rqstp->rq_res.head[0].iov_len; | |
331 | rqstp->rq_res.head[0].iov_len += sizeof(u32); | |
332 | ||
333 | /* Now call the procedure handler, and encode NFS status. */ | |
334 | nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp); | |
335 | if (nfserr == nfserr_jukebox && rqstp->rq_vers == 2) | |
336 | nfserr = nfserr_dropit; | |
337 | if (nfserr == nfserr_dropit) { | |
338 | dprintk("nfsd: Dropping request due to malloc failure!\n"); | |
339 | nfsd_cache_update(rqstp, RC_NOCACHE, NULL); | |
340 | return 0; | |
341 | } | |
342 | ||
343 | if (rqstp->rq_proc != 0) | |
344 | *nfserrp++ = nfserr; | |
345 | ||
346 | /* Encode result. | |
347 | * For NFSv2, additional info is never returned in case of an error. | |
348 | */ | |
349 | if (!(nfserr && rqstp->rq_vers == 2)) { | |
350 | xdr = proc->pc_encode; | |
351 | if (xdr && !xdr(rqstp, nfserrp, | |
352 | rqstp->rq_resp)) { | |
353 | /* Failed to encode result. Release cache entry */ | |
354 | dprintk("nfsd: failed to encode result!\n"); | |
355 | nfsd_cache_update(rqstp, RC_NOCACHE, NULL); | |
356 | *statp = rpc_system_err; | |
357 | return 1; | |
358 | } | |
359 | } | |
360 | ||
361 | /* Store reply in cache. */ | |
362 | nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1); | |
363 | return 1; | |
364 | } | |
365 | ||
a257cdd0 AG |
366 | #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) |
367 | static struct svc_stat nfsd_acl_svcstats; | |
368 | static struct svc_version * nfsd_acl_version[] = { | |
369 | [2] = &nfsd_acl_version2, | |
370 | [3] = &nfsd_acl_version3, | |
371 | }; | |
372 | ||
373 | #define NFSD_ACL_NRVERS (sizeof(nfsd_acl_version)/sizeof(nfsd_acl_version[0])) | |
374 | static struct svc_program nfsd_acl_program = { | |
375 | .pg_prog = NFS_ACL_PROGRAM, | |
376 | .pg_nvers = NFSD_ACL_NRVERS, | |
377 | .pg_vers = nfsd_acl_version, | |
378 | .pg_name = "nfsd", | |
21348425 | 379 | .pg_class = "nfsd", |
a257cdd0 AG |
380 | .pg_stats = &nfsd_acl_svcstats, |
381 | }; | |
382 | ||
383 | static struct svc_stat nfsd_acl_svcstats = { | |
384 | .program = &nfsd_acl_program, | |
385 | }; | |
386 | ||
387 | #define nfsd_acl_program_p &nfsd_acl_program | |
388 | #else | |
389 | #define nfsd_acl_program_p NULL | |
390 | #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ | |
391 | ||
1da177e4 LT |
392 | extern struct svc_version nfsd_version2, nfsd_version3, nfsd_version4; |
393 | ||
394 | static struct svc_version * nfsd_version[] = { | |
395 | [2] = &nfsd_version2, | |
396 | #if defined(CONFIG_NFSD_V3) | |
397 | [3] = &nfsd_version3, | |
398 | #endif | |
399 | #if defined(CONFIG_NFSD_V4) | |
400 | [4] = &nfsd_version4, | |
401 | #endif | |
402 | }; | |
403 | ||
404 | #define NFSD_NRVERS (sizeof(nfsd_version)/sizeof(nfsd_version[0])) | |
405 | struct svc_program nfsd_program = { | |
a257cdd0 | 406 | .pg_next = nfsd_acl_program_p, |
1da177e4 LT |
407 | .pg_prog = NFS_PROGRAM, /* program number */ |
408 | .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */ | |
409 | .pg_vers = nfsd_version, /* version table */ | |
410 | .pg_name = "nfsd", /* program name */ | |
411 | .pg_class = "nfsd", /* authentication class */ | |
412 | .pg_stats = &nfsd_svcstats, /* version table */ | |
413 | .pg_authenticate = &svc_set_client, /* export authentication */ | |
414 | ||
415 | }; |