Merge branch 'for-3.10' of git://linux-nfs.org/~bfields/linux
[linux-3.10.git] / fs / nfsd / nfsctl.c
1 /*
2  * Syscall interface to knfsd.
3  *
4  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
5  */
6
7 #include <linux/slab.h>
8 #include <linux/namei.h>
9 #include <linux/ctype.h>
10
11 #include <linux/sunrpc/svcsock.h>
12 #include <linux/lockd/lockd.h>
13 #include <linux/sunrpc/addr.h>
14 #include <linux/sunrpc/gss_api.h>
15 #include <linux/sunrpc/gss_krb5_enctypes.h>
16 #include <linux/sunrpc/rpc_pipe_fs.h>
17 #include <linux/module.h>
18
19 #include "idmap.h"
20 #include "nfsd.h"
21 #include "cache.h"
22 #include "state.h"
23 #include "netns.h"
24
25 /*
26  *      We have a single directory with several nodes in it.
27  */
28 enum {
29         NFSD_Root = 1,
30         NFSD_List,
31         NFSD_Export_features,
32         NFSD_Fh,
33         NFSD_FO_UnlockIP,
34         NFSD_FO_UnlockFS,
35         NFSD_Threads,
36         NFSD_Pool_Threads,
37         NFSD_Pool_Stats,
38         NFSD_Reply_Cache_Stats,
39         NFSD_Versions,
40         NFSD_Ports,
41         NFSD_MaxBlkSize,
42         NFSD_SupportedEnctypes,
43         /*
44          * The below MUST come last.  Otherwise we leave a hole in nfsd_files[]
45          * with !CONFIG_NFSD_V4 and simple_fill_super() goes oops
46          */
47 #ifdef CONFIG_NFSD_V4
48         NFSD_Leasetime,
49         NFSD_Gracetime,
50         NFSD_RecoveryDir,
51 #endif
52 };
53
54 /*
55  * write() for these nodes.
56  */
57 static ssize_t write_filehandle(struct file *file, char *buf, size_t size);
58 static ssize_t write_unlock_ip(struct file *file, char *buf, size_t size);
59 static ssize_t write_unlock_fs(struct file *file, char *buf, size_t size);
60 static ssize_t write_threads(struct file *file, char *buf, size_t size);
61 static ssize_t write_pool_threads(struct file *file, char *buf, size_t size);
62 static ssize_t write_versions(struct file *file, char *buf, size_t size);
63 static ssize_t write_ports(struct file *file, char *buf, size_t size);
64 static ssize_t write_maxblksize(struct file *file, char *buf, size_t size);
65 #ifdef CONFIG_NFSD_V4
66 static ssize_t write_leasetime(struct file *file, char *buf, size_t size);
67 static ssize_t write_gracetime(struct file *file, char *buf, size_t size);
68 static ssize_t write_recoverydir(struct file *file, char *buf, size_t size);
69 #endif
70
71 static ssize_t (*write_op[])(struct file *, char *, size_t) = {
72         [NFSD_Fh] = write_filehandle,
73         [NFSD_FO_UnlockIP] = write_unlock_ip,
74         [NFSD_FO_UnlockFS] = write_unlock_fs,
75         [NFSD_Threads] = write_threads,
76         [NFSD_Pool_Threads] = write_pool_threads,
77         [NFSD_Versions] = write_versions,
78         [NFSD_Ports] = write_ports,
79         [NFSD_MaxBlkSize] = write_maxblksize,
80 #ifdef CONFIG_NFSD_V4
81         [NFSD_Leasetime] = write_leasetime,
82         [NFSD_Gracetime] = write_gracetime,
83         [NFSD_RecoveryDir] = write_recoverydir,
84 #endif
85 };
86
87 static ssize_t nfsctl_transaction_write(struct file *file, const char __user *buf, size_t size, loff_t *pos)
88 {
89         ino_t ino =  file_inode(file)->i_ino;
90         char *data;
91         ssize_t rv;
92
93         if (ino >= ARRAY_SIZE(write_op) || !write_op[ino])
94                 return -EINVAL;
95
96         data = simple_transaction_get(file, buf, size);
97         if (IS_ERR(data))
98                 return PTR_ERR(data);
99
100         rv =  write_op[ino](file, data, size);
101         if (rv >= 0) {
102                 simple_transaction_set(file, rv);
103                 rv = size;
104         }
105         return rv;
106 }
107
108 static ssize_t nfsctl_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
109 {
110         if (! file->private_data) {
111                 /* An attempt to read a transaction file without writing
112                  * causes a 0-byte write so that the file can return
113                  * state information
114                  */
115                 ssize_t rv = nfsctl_transaction_write(file, buf, 0, pos);
116                 if (rv < 0)
117                         return rv;
118         }
119         return simple_transaction_read(file, buf, size, pos);
120 }
121
122 static const struct file_operations transaction_ops = {
123         .write          = nfsctl_transaction_write,
124         .read           = nfsctl_transaction_read,
125         .release        = simple_transaction_release,
126         .llseek         = default_llseek,
127 };
128
129 static int exports_net_open(struct net *net, struct file *file)
130 {
131         int err;
132         struct seq_file *seq;
133         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
134
135         err = seq_open(file, &nfs_exports_op);
136         if (err)
137                 return err;
138
139         seq = file->private_data;
140         seq->private = nn->svc_export_cache;
141         return 0;
142 }
143
144 static int exports_proc_open(struct inode *inode, struct file *file)
145 {
146         return exports_net_open(current->nsproxy->net_ns, file);
147 }
148
149 static const struct file_operations exports_proc_operations = {
150         .open           = exports_proc_open,
151         .read           = seq_read,
152         .llseek         = seq_lseek,
153         .release        = seq_release,
154         .owner          = THIS_MODULE,
155 };
156
157 static int exports_nfsd_open(struct inode *inode, struct file *file)
158 {
159         return exports_net_open(inode->i_sb->s_fs_info, file);
160 }
161
162 static const struct file_operations exports_nfsd_operations = {
163         .open           = exports_nfsd_open,
164         .read           = seq_read,
165         .llseek         = seq_lseek,
166         .release        = seq_release,
167         .owner          = THIS_MODULE,
168 };
169
170 static int export_features_show(struct seq_file *m, void *v)
171 {
172         seq_printf(m, "0x%x 0x%x\n", NFSEXP_ALLFLAGS, NFSEXP_SECINFO_FLAGS);
173         return 0;
174 }
175
176 static int export_features_open(struct inode *inode, struct file *file)
177 {
178         return single_open(file, export_features_show, NULL);
179 }
180
181 static const struct file_operations export_features_operations = {
182         .open           = export_features_open,
183         .read           = seq_read,
184         .llseek         = seq_lseek,
185         .release        = single_release,
186 };
187
188 #if defined(CONFIG_SUNRPC_GSS) || defined(CONFIG_SUNRPC_GSS_MODULE)
189 static int supported_enctypes_show(struct seq_file *m, void *v)
190 {
191         seq_printf(m, KRB5_SUPPORTED_ENCTYPES);
192         return 0;
193 }
194
195 static int supported_enctypes_open(struct inode *inode, struct file *file)
196 {
197         return single_open(file, supported_enctypes_show, NULL);
198 }
199
200 static const struct file_operations supported_enctypes_ops = {
201         .open           = supported_enctypes_open,
202         .read           = seq_read,
203         .llseek         = seq_lseek,
204         .release        = single_release,
205 };
206 #endif /* CONFIG_SUNRPC_GSS or CONFIG_SUNRPC_GSS_MODULE */
207
208 static const struct file_operations pool_stats_operations = {
209         .open           = nfsd_pool_stats_open,
210         .read           = seq_read,
211         .llseek         = seq_lseek,
212         .release        = nfsd_pool_stats_release,
213         .owner          = THIS_MODULE,
214 };
215
216 static struct file_operations reply_cache_stats_operations = {
217         .open           = nfsd_reply_cache_stats_open,
218         .read           = seq_read,
219         .llseek         = seq_lseek,
220         .release        = single_release,
221 };
222
223 /*----------------------------------------------------------------------------*/
224 /*
225  * payload - write methods
226  */
227
228
229 /**
230  * write_unlock_ip - Release all locks used by a client
231  *
232  * Experimental.
233  *
234  * Input:
235  *                      buf:    '\n'-terminated C string containing a
236  *                              presentation format IP address
237  *                      size:   length of C string in @buf
238  * Output:
239  *      On success:     returns zero if all specified locks were released;
240  *                      returns one if one or more locks were not released
241  *      On error:       return code is negative errno value
242  */
243 static ssize_t write_unlock_ip(struct file *file, char *buf, size_t size)
244 {
245         struct sockaddr_storage address;
246         struct sockaddr *sap = (struct sockaddr *)&address;
247         size_t salen = sizeof(address);
248         char *fo_path;
249         struct net *net = file->f_dentry->d_sb->s_fs_info;
250
251         /* sanity check */
252         if (size == 0)
253                 return -EINVAL;
254
255         if (buf[size-1] != '\n')
256                 return -EINVAL;
257
258         fo_path = buf;
259         if (qword_get(&buf, fo_path, size) < 0)
260                 return -EINVAL;
261
262         if (rpc_pton(net, fo_path, size, sap, salen) == 0)
263                 return -EINVAL;
264
265         return nlmsvc_unlock_all_by_ip(sap);
266 }
267
268 /**
269  * write_unlock_fs - Release all locks on a local file system
270  *
271  * Experimental.
272  *
273  * Input:
274  *                      buf:    '\n'-terminated C string containing the
275  *                              absolute pathname of a local file system
276  *                      size:   length of C string in @buf
277  * Output:
278  *      On success:     returns zero if all specified locks were released;
279  *                      returns one if one or more locks were not released
280  *      On error:       return code is negative errno value
281  */
282 static ssize_t write_unlock_fs(struct file *file, char *buf, size_t size)
283 {
284         struct path path;
285         char *fo_path;
286         int error;
287
288         /* sanity check */
289         if (size == 0)
290                 return -EINVAL;
291
292         if (buf[size-1] != '\n')
293                 return -EINVAL;
294
295         fo_path = buf;
296         if (qword_get(&buf, fo_path, size) < 0)
297                 return -EINVAL;
298
299         error = kern_path(fo_path, 0, &path);
300         if (error)
301                 return error;
302
303         /*
304          * XXX: Needs better sanity checking.  Otherwise we could end up
305          * releasing locks on the wrong file system.
306          *
307          * For example:
308          * 1.  Does the path refer to a directory?
309          * 2.  Is that directory a mount point, or
310          * 3.  Is that directory the root of an exported file system?
311          */
312         error = nlmsvc_unlock_all_by_sb(path.dentry->d_sb);
313
314         path_put(&path);
315         return error;
316 }
317
318 /**
319  * write_filehandle - Get a variable-length NFS file handle by path
320  *
321  * On input, the buffer contains a '\n'-terminated C string comprised of
322  * three alphanumeric words separated by whitespace.  The string may
323  * contain escape sequences.
324  *
325  * Input:
326  *                      buf:
327  *                              domain:         client domain name
328  *                              path:           export pathname
329  *                              maxsize:        numeric maximum size of
330  *                                              @buf
331  *                      size:   length of C string in @buf
332  * Output:
333  *      On success:     passed-in buffer filled with '\n'-terminated C
334  *                      string containing a ASCII hex text version
335  *                      of the NFS file handle;
336  *                      return code is the size in bytes of the string
337  *      On error:       return code is negative errno value
338  */
339 static ssize_t write_filehandle(struct file *file, char *buf, size_t size)
340 {
341         char *dname, *path;
342         int uninitialized_var(maxsize);
343         char *mesg = buf;
344         int len;
345         struct auth_domain *dom;
346         struct knfsd_fh fh;
347         struct net *net = file->f_dentry->d_sb->s_fs_info;
348
349         if (size == 0)
350                 return -EINVAL;
351
352         if (buf[size-1] != '\n')
353                 return -EINVAL;
354         buf[size-1] = 0;
355
356         dname = mesg;
357         len = qword_get(&mesg, dname, size);
358         if (len <= 0)
359                 return -EINVAL;
360         
361         path = dname+len+1;
362         len = qword_get(&mesg, path, size);
363         if (len <= 0)
364                 return -EINVAL;
365
366         len = get_int(&mesg, &maxsize);
367         if (len)
368                 return len;
369
370         if (maxsize < NFS_FHSIZE)
371                 return -EINVAL;
372         if (maxsize > NFS3_FHSIZE)
373                 maxsize = NFS3_FHSIZE;
374
375         if (qword_get(&mesg, mesg, size)>0)
376                 return -EINVAL;
377
378         /* we have all the words, they are in buf.. */
379         dom = unix_domain_find(dname);
380         if (!dom)
381                 return -ENOMEM;
382
383         len = exp_rootfh(net, dom, path, &fh,  maxsize);
384         auth_domain_put(dom);
385         if (len)
386                 return len;
387         
388         mesg = buf;
389         len = SIMPLE_TRANSACTION_LIMIT;
390         qword_addhex(&mesg, &len, (char*)&fh.fh_base, fh.fh_size);
391         mesg[-1] = '\n';
392         return mesg - buf;      
393 }
394
395 /**
396  * write_threads - Start NFSD, or report the current number of running threads
397  *
398  * Input:
399  *                      buf:            ignored
400  *                      size:           zero
401  * Output:
402  *      On success:     passed-in buffer filled with '\n'-terminated C
403  *                      string numeric value representing the number of
404  *                      running NFSD threads;
405  *                      return code is the size in bytes of the string
406  *      On error:       return code is zero
407  *
408  * OR
409  *
410  * Input:
411  *                      buf:            C string containing an unsigned
412  *                                      integer value representing the
413  *                                      number of NFSD threads to start
414  *                      size:           non-zero length of C string in @buf
415  * Output:
416  *      On success:     NFS service is started;
417  *                      passed-in buffer filled with '\n'-terminated C
418  *                      string numeric value representing the number of
419  *                      running NFSD threads;
420  *                      return code is the size in bytes of the string
421  *      On error:       return code is zero or a negative errno value
422  */
423 static ssize_t write_threads(struct file *file, char *buf, size_t size)
424 {
425         char *mesg = buf;
426         int rv;
427         struct net *net = file->f_dentry->d_sb->s_fs_info;
428
429         if (size > 0) {
430                 int newthreads;
431                 rv = get_int(&mesg, &newthreads);
432                 if (rv)
433                         return rv;
434                 if (newthreads < 0)
435                         return -EINVAL;
436                 rv = nfsd_svc(newthreads, net);
437                 if (rv < 0)
438                         return rv;
439         } else
440                 rv = nfsd_nrthreads(net);
441
442         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%d\n", rv);
443 }
444
445 /**
446  * write_pool_threads - Set or report the current number of threads per pool
447  *
448  * Input:
449  *                      buf:            ignored
450  *                      size:           zero
451  *
452  * OR
453  *
454  * Input:
455  *                      buf:            C string containing whitespace-
456  *                                      separated unsigned integer values
457  *                                      representing the number of NFSD
458  *                                      threads to start in each pool
459  *                      size:           non-zero length of C string in @buf
460  * Output:
461  *      On success:     passed-in buffer filled with '\n'-terminated C
462  *                      string containing integer values representing the
463  *                      number of NFSD threads in each pool;
464  *                      return code is the size in bytes of the string
465  *      On error:       return code is zero or a negative errno value
466  */
467 static ssize_t write_pool_threads(struct file *file, char *buf, size_t size)
468 {
469         /* if size > 0, look for an array of number of threads per node
470          * and apply them  then write out number of threads per node as reply
471          */
472         char *mesg = buf;
473         int i;
474         int rv;
475         int len;
476         int npools;
477         int *nthreads;
478         struct net *net = file->f_dentry->d_sb->s_fs_info;
479
480         mutex_lock(&nfsd_mutex);
481         npools = nfsd_nrpools(net);
482         if (npools == 0) {
483                 /*
484                  * NFS is shut down.  The admin can start it by
485                  * writing to the threads file but NOT the pool_threads
486                  * file, sorry.  Report zero threads.
487                  */
488                 mutex_unlock(&nfsd_mutex);
489                 strcpy(buf, "0\n");
490                 return strlen(buf);
491         }
492
493         nthreads = kcalloc(npools, sizeof(int), GFP_KERNEL);
494         rv = -ENOMEM;
495         if (nthreads == NULL)
496                 goto out_free;
497
498         if (size > 0) {
499                 for (i = 0; i < npools; i++) {
500                         rv = get_int(&mesg, &nthreads[i]);
501                         if (rv == -ENOENT)
502                                 break;          /* fewer numbers than pools */
503                         if (rv)
504                                 goto out_free;  /* syntax error */
505                         rv = -EINVAL;
506                         if (nthreads[i] < 0)
507                                 goto out_free;
508                 }
509                 rv = nfsd_set_nrthreads(i, nthreads, net);
510                 if (rv)
511                         goto out_free;
512         }
513
514         rv = nfsd_get_nrthreads(npools, nthreads, net);
515         if (rv)
516                 goto out_free;
517
518         mesg = buf;
519         size = SIMPLE_TRANSACTION_LIMIT;
520         for (i = 0; i < npools && size > 0; i++) {
521                 snprintf(mesg, size, "%d%c", nthreads[i], (i == npools-1 ? '\n' : ' '));
522                 len = strlen(mesg);
523                 size -= len;
524                 mesg += len;
525         }
526         rv = mesg - buf;
527 out_free:
528         kfree(nthreads);
529         mutex_unlock(&nfsd_mutex);
530         return rv;
531 }
532
533 static ssize_t __write_versions(struct file *file, char *buf, size_t size)
534 {
535         char *mesg = buf;
536         char *vers, *minorp, sign;
537         int len, num, remaining;
538         unsigned minor;
539         ssize_t tlen = 0;
540         char *sep;
541         struct net *net = file->f_dentry->d_sb->s_fs_info;
542         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
543
544         if (size>0) {
545                 if (nn->nfsd_serv)
546                         /* Cannot change versions without updating
547                          * nn->nfsd_serv->sv_xdrsize, and reallocing
548                          * rq_argp and rq_resp
549                          */
550                         return -EBUSY;
551                 if (buf[size-1] != '\n')
552                         return -EINVAL;
553                 buf[size-1] = 0;
554
555                 vers = mesg;
556                 len = qword_get(&mesg, vers, size);
557                 if (len <= 0) return -EINVAL;
558                 do {
559                         sign = *vers;
560                         if (sign == '+' || sign == '-')
561                                 num = simple_strtol((vers+1), &minorp, 0);
562                         else
563                                 num = simple_strtol(vers, &minorp, 0);
564                         if (*minorp == '.') {
565                                 if (num != 4)
566                                         return -EINVAL;
567                                 minor = simple_strtoul(minorp+1, NULL, 0);
568                                 if (minor == 0)
569                                         return -EINVAL;
570                                 if (nfsd_minorversion(minor, sign == '-' ?
571                                                      NFSD_CLEAR : NFSD_SET) < 0)
572                                         return -EINVAL;
573                                 goto next;
574                         }
575                         switch(num) {
576                         case 2:
577                         case 3:
578                         case 4:
579                                 nfsd_vers(num, sign == '-' ? NFSD_CLEAR : NFSD_SET);
580                                 break;
581                         default:
582                                 return -EINVAL;
583                         }
584                 next:
585                         vers += len + 1;
586                 } while ((len = qword_get(&mesg, vers, size)) > 0);
587                 /* If all get turned off, turn them back on, as
588                  * having no versions is BAD
589                  */
590                 nfsd_reset_versions();
591         }
592
593         /* Now write current state into reply buffer */
594         len = 0;
595         sep = "";
596         remaining = SIMPLE_TRANSACTION_LIMIT;
597         for (num=2 ; num <= 4 ; num++)
598                 if (nfsd_vers(num, NFSD_AVAIL)) {
599                         len = snprintf(buf, remaining, "%s%c%d", sep,
600                                        nfsd_vers(num, NFSD_TEST)?'+':'-',
601                                        num);
602                         sep = " ";
603
604                         if (len > remaining)
605                                 break;
606                         remaining -= len;
607                         buf += len;
608                         tlen += len;
609                 }
610         if (nfsd_vers(4, NFSD_AVAIL))
611                 for (minor = 1; minor <= NFSD_SUPPORTED_MINOR_VERSION;
612                      minor++) {
613                         len = snprintf(buf, remaining, " %c4.%u",
614                                         (nfsd_vers(4, NFSD_TEST) &&
615                                          nfsd_minorversion(minor, NFSD_TEST)) ?
616                                                 '+' : '-',
617                                         minor);
618
619                         if (len > remaining)
620                                 break;
621                         remaining -= len;
622                         buf += len;
623                         tlen += len;
624                 }
625
626         len = snprintf(buf, remaining, "\n");
627         if (len > remaining)
628                 return -EINVAL;
629         return tlen + len;
630 }
631
632 /**
633  * write_versions - Set or report the available NFS protocol versions
634  *
635  * Input:
636  *                      buf:            ignored
637  *                      size:           zero
638  * Output:
639  *      On success:     passed-in buffer filled with '\n'-terminated C
640  *                      string containing positive or negative integer
641  *                      values representing the current status of each
642  *                      protocol version;
643  *                      return code is the size in bytes of the string
644  *      On error:       return code is zero or a negative errno value
645  *
646  * OR
647  *
648  * Input:
649  *                      buf:            C string containing whitespace-
650  *                                      separated positive or negative
651  *                                      integer values representing NFS
652  *                                      protocol versions to enable ("+n")
653  *                                      or disable ("-n")
654  *                      size:           non-zero length of C string in @buf
655  * Output:
656  *      On success:     status of zero or more protocol versions has
657  *                      been updated; passed-in buffer filled with
658  *                      '\n'-terminated C string containing positive
659  *                      or negative integer values representing the
660  *                      current status of each protocol version;
661  *                      return code is the size in bytes of the string
662  *      On error:       return code is zero or a negative errno value
663  */
664 static ssize_t write_versions(struct file *file, char *buf, size_t size)
665 {
666         ssize_t rv;
667
668         mutex_lock(&nfsd_mutex);
669         rv = __write_versions(file, buf, size);
670         mutex_unlock(&nfsd_mutex);
671         return rv;
672 }
673
674 /*
675  * Zero-length write.  Return a list of NFSD's current listener
676  * transports.
677  */
678 static ssize_t __write_ports_names(char *buf, struct net *net)
679 {
680         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
681
682         if (nn->nfsd_serv == NULL)
683                 return 0;
684         return svc_xprt_names(nn->nfsd_serv, buf, SIMPLE_TRANSACTION_LIMIT);
685 }
686
687 /*
688  * A single 'fd' number was written, in which case it must be for
689  * a socket of a supported family/protocol, and we use it as an
690  * nfsd listener.
691  */
692 static ssize_t __write_ports_addfd(char *buf, struct net *net)
693 {
694         char *mesg = buf;
695         int fd, err;
696         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
697
698         err = get_int(&mesg, &fd);
699         if (err != 0 || fd < 0)
700                 return -EINVAL;
701
702         err = nfsd_create_serv(net);
703         if (err != 0)
704                 return err;
705
706         err = svc_addsock(nn->nfsd_serv, fd, buf, SIMPLE_TRANSACTION_LIMIT);
707         if (err < 0) {
708                 nfsd_destroy(net);
709                 return err;
710         }
711
712         /* Decrease the count, but don't shut down the service */
713         nn->nfsd_serv->sv_nrthreads--;
714         return err;
715 }
716
717 /*
718  * A transport listener is added by writing it's transport name and
719  * a port number.
720  */
721 static ssize_t __write_ports_addxprt(char *buf, struct net *net)
722 {
723         char transport[16];
724         struct svc_xprt *xprt;
725         int port, err;
726         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
727
728         if (sscanf(buf, "%15s %5u", transport, &port) != 2)
729                 return -EINVAL;
730
731         if (port < 1 || port > USHRT_MAX)
732                 return -EINVAL;
733
734         err = nfsd_create_serv(net);
735         if (err != 0)
736                 return err;
737
738         err = svc_create_xprt(nn->nfsd_serv, transport, net,
739                                 PF_INET, port, SVC_SOCK_ANONYMOUS);
740         if (err < 0)
741                 goto out_err;
742
743         err = svc_create_xprt(nn->nfsd_serv, transport, net,
744                                 PF_INET6, port, SVC_SOCK_ANONYMOUS);
745         if (err < 0 && err != -EAFNOSUPPORT)
746                 goto out_close;
747
748         /* Decrease the count, but don't shut down the service */
749         nn->nfsd_serv->sv_nrthreads--;
750         return 0;
751 out_close:
752         xprt = svc_find_xprt(nn->nfsd_serv, transport, net, PF_INET, port);
753         if (xprt != NULL) {
754                 svc_close_xprt(xprt);
755                 svc_xprt_put(xprt);
756         }
757 out_err:
758         nfsd_destroy(net);
759         return err;
760 }
761
762 static ssize_t __write_ports(struct file *file, char *buf, size_t size,
763                              struct net *net)
764 {
765         if (size == 0)
766                 return __write_ports_names(buf, net);
767
768         if (isdigit(buf[0]))
769                 return __write_ports_addfd(buf, net);
770
771         if (isalpha(buf[0]))
772                 return __write_ports_addxprt(buf, net);
773
774         return -EINVAL;
775 }
776
777 /**
778  * write_ports - Pass a socket file descriptor or transport name to listen on
779  *
780  * Input:
781  *                      buf:            ignored
782  *                      size:           zero
783  * Output:
784  *      On success:     passed-in buffer filled with a '\n'-terminated C
785  *                      string containing a whitespace-separated list of
786  *                      named NFSD listeners;
787  *                      return code is the size in bytes of the string
788  *      On error:       return code is zero or a negative errno value
789  *
790  * OR
791  *
792  * Input:
793  *                      buf:            C string containing an unsigned
794  *                                      integer value representing a bound
795  *                                      but unconnected socket that is to be
796  *                                      used as an NFSD listener; listen(3)
797  *                                      must be called for a SOCK_STREAM
798  *                                      socket, otherwise it is ignored
799  *                      size:           non-zero length of C string in @buf
800  * Output:
801  *      On success:     NFS service is started;
802  *                      passed-in buffer filled with a '\n'-terminated C
803  *                      string containing a unique alphanumeric name of
804  *                      the listener;
805  *                      return code is the size in bytes of the string
806  *      On error:       return code is a negative errno value
807  *
808  * OR
809  *
810  * Input:
811  *                      buf:            C string containing a transport
812  *                                      name and an unsigned integer value
813  *                                      representing the port to listen on,
814  *                                      separated by whitespace
815  *                      size:           non-zero length of C string in @buf
816  * Output:
817  *      On success:     returns zero; NFS service is started
818  *      On error:       return code is a negative errno value
819  */
820 static ssize_t write_ports(struct file *file, char *buf, size_t size)
821 {
822         ssize_t rv;
823         struct net *net = file->f_dentry->d_sb->s_fs_info;
824
825         mutex_lock(&nfsd_mutex);
826         rv = __write_ports(file, buf, size, net);
827         mutex_unlock(&nfsd_mutex);
828         return rv;
829 }
830
831
832 int nfsd_max_blksize;
833
834 /**
835  * write_maxblksize - Set or report the current NFS blksize
836  *
837  * Input:
838  *                      buf:            ignored
839  *                      size:           zero
840  *
841  * OR
842  *
843  * Input:
844  *                      buf:            C string containing an unsigned
845  *                                      integer value representing the new
846  *                                      NFS blksize
847  *                      size:           non-zero length of C string in @buf
848  * Output:
849  *      On success:     passed-in buffer filled with '\n'-terminated C string
850  *                      containing numeric value of the current NFS blksize
851  *                      setting;
852  *                      return code is the size in bytes of the string
853  *      On error:       return code is zero or a negative errno value
854  */
855 static ssize_t write_maxblksize(struct file *file, char *buf, size_t size)
856 {
857         char *mesg = buf;
858         struct net *net = file->f_dentry->d_sb->s_fs_info;
859         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
860
861         if (size > 0) {
862                 int bsize;
863                 int rv = get_int(&mesg, &bsize);
864                 if (rv)
865                         return rv;
866                 /* force bsize into allowed range and
867                  * required alignment.
868                  */
869                 if (bsize < 1024)
870                         bsize = 1024;
871                 if (bsize > NFSSVC_MAXBLKSIZE)
872                         bsize = NFSSVC_MAXBLKSIZE;
873                 bsize &= ~(1024-1);
874                 mutex_lock(&nfsd_mutex);
875                 if (nn->nfsd_serv) {
876                         mutex_unlock(&nfsd_mutex);
877                         return -EBUSY;
878                 }
879                 nfsd_max_blksize = bsize;
880                 mutex_unlock(&nfsd_mutex);
881         }
882
883         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%d\n",
884                                                         nfsd_max_blksize);
885 }
886
887 #ifdef CONFIG_NFSD_V4
888 static ssize_t __nfsd4_write_time(struct file *file, char *buf, size_t size,
889                                   time_t *time, struct nfsd_net *nn)
890 {
891         char *mesg = buf;
892         int rv, i;
893
894         if (size > 0) {
895                 if (nn->nfsd_serv)
896                         return -EBUSY;
897                 rv = get_int(&mesg, &i);
898                 if (rv)
899                         return rv;
900                 /*
901                  * Some sanity checking.  We don't have a reason for
902                  * these particular numbers, but problems with the
903                  * extremes are:
904                  *      - Too short: the briefest network outage may
905                  *        cause clients to lose all their locks.  Also,
906                  *        the frequent polling may be wasteful.
907                  *      - Too long: do you really want reboot recovery
908                  *        to take more than an hour?  Or to make other
909                  *        clients wait an hour before being able to
910                  *        revoke a dead client's locks?
911                  */
912                 if (i < 10 || i > 3600)
913                         return -EINVAL;
914                 *time = i;
915         }
916
917         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%ld\n", *time);
918 }
919
920 static ssize_t nfsd4_write_time(struct file *file, char *buf, size_t size,
921                                 time_t *time, struct nfsd_net *nn)
922 {
923         ssize_t rv;
924
925         mutex_lock(&nfsd_mutex);
926         rv = __nfsd4_write_time(file, buf, size, time, nn);
927         mutex_unlock(&nfsd_mutex);
928         return rv;
929 }
930
931 /**
932  * write_leasetime - Set or report the current NFSv4 lease time
933  *
934  * Input:
935  *                      buf:            ignored
936  *                      size:           zero
937  *
938  * OR
939  *
940  * Input:
941  *                      buf:            C string containing an unsigned
942  *                                      integer value representing the new
943  *                                      NFSv4 lease expiry time
944  *                      size:           non-zero length of C string in @buf
945  * Output:
946  *      On success:     passed-in buffer filled with '\n'-terminated C
947  *                      string containing unsigned integer value of the
948  *                      current lease expiry time;
949  *                      return code is the size in bytes of the string
950  *      On error:       return code is zero or a negative errno value
951  */
952 static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
953 {
954         struct net *net = file->f_dentry->d_sb->s_fs_info;
955         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
956         return nfsd4_write_time(file, buf, size, &nn->nfsd4_lease, nn);
957 }
958
959 /**
960  * write_gracetime - Set or report current NFSv4 grace period time
961  *
962  * As above, but sets the time of the NFSv4 grace period.
963  *
964  * Note this should never be set to less than the *previous*
965  * lease-period time, but we don't try to enforce this.  (In the common
966  * case (a new boot), we don't know what the previous lease time was
967  * anyway.)
968  */
969 static ssize_t write_gracetime(struct file *file, char *buf, size_t size)
970 {
971         struct net *net = file->f_dentry->d_sb->s_fs_info;
972         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
973         return nfsd4_write_time(file, buf, size, &nn->nfsd4_grace, nn);
974 }
975
976 static ssize_t __write_recoverydir(struct file *file, char *buf, size_t size,
977                                    struct nfsd_net *nn)
978 {
979         char *mesg = buf;
980         char *recdir;
981         int len, status;
982
983         if (size > 0) {
984                 if (nn->nfsd_serv)
985                         return -EBUSY;
986                 if (size > PATH_MAX || buf[size-1] != '\n')
987                         return -EINVAL;
988                 buf[size-1] = 0;
989
990                 recdir = mesg;
991                 len = qword_get(&mesg, recdir, size);
992                 if (len <= 0)
993                         return -EINVAL;
994
995                 status = nfs4_reset_recoverydir(recdir);
996                 if (status)
997                         return status;
998         }
999
1000         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%s\n",
1001                                                         nfs4_recoverydir());
1002 }
1003
1004 /**
1005  * write_recoverydir - Set or report the pathname of the recovery directory
1006  *
1007  * Input:
1008  *                      buf:            ignored
1009  *                      size:           zero
1010  *
1011  * OR
1012  *
1013  * Input:
1014  *                      buf:            C string containing the pathname
1015  *                                      of the directory on a local file
1016  *                                      system containing permanent NFSv4
1017  *                                      recovery data
1018  *                      size:           non-zero length of C string in @buf
1019  * Output:
1020  *      On success:     passed-in buffer filled with '\n'-terminated C string
1021  *                      containing the current recovery pathname setting;
1022  *                      return code is the size in bytes of the string
1023  *      On error:       return code is zero or a negative errno value
1024  */
1025 static ssize_t write_recoverydir(struct file *file, char *buf, size_t size)
1026 {
1027         ssize_t rv;
1028         struct net *net = file->f_dentry->d_sb->s_fs_info;
1029         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1030
1031         mutex_lock(&nfsd_mutex);
1032         rv = __write_recoverydir(file, buf, size, nn);
1033         mutex_unlock(&nfsd_mutex);
1034         return rv;
1035 }
1036
1037 #endif
1038
1039 /*----------------------------------------------------------------------------*/
1040 /*
1041  *      populating the filesystem.
1042  */
1043
1044 static int nfsd_fill_super(struct super_block * sb, void * data, int silent)
1045 {
1046         static struct tree_descr nfsd_files[] = {
1047                 [NFSD_List] = {"exports", &exports_nfsd_operations, S_IRUGO},
1048                 [NFSD_Export_features] = {"export_features",
1049                                         &export_features_operations, S_IRUGO},
1050                 [NFSD_FO_UnlockIP] = {"unlock_ip",
1051                                         &transaction_ops, S_IWUSR|S_IRUSR},
1052                 [NFSD_FO_UnlockFS] = {"unlock_filesystem",
1053                                         &transaction_ops, S_IWUSR|S_IRUSR},
1054                 [NFSD_Fh] = {"filehandle", &transaction_ops, S_IWUSR|S_IRUSR},
1055                 [NFSD_Threads] = {"threads", &transaction_ops, S_IWUSR|S_IRUSR},
1056                 [NFSD_Pool_Threads] = {"pool_threads", &transaction_ops, S_IWUSR|S_IRUSR},
1057                 [NFSD_Pool_Stats] = {"pool_stats", &pool_stats_operations, S_IRUGO},
1058                 [NFSD_Reply_Cache_Stats] = {"reply_cache_stats", &reply_cache_stats_operations, S_IRUGO},
1059                 [NFSD_Versions] = {"versions", &transaction_ops, S_IWUSR|S_IRUSR},
1060                 [NFSD_Ports] = {"portlist", &transaction_ops, S_IWUSR|S_IRUGO},
1061                 [NFSD_MaxBlkSize] = {"max_block_size", &transaction_ops, S_IWUSR|S_IRUGO},
1062 #if defined(CONFIG_SUNRPC_GSS) || defined(CONFIG_SUNRPC_GSS_MODULE)
1063                 [NFSD_SupportedEnctypes] = {"supported_krb5_enctypes", &supported_enctypes_ops, S_IRUGO},
1064 #endif /* CONFIG_SUNRPC_GSS or CONFIG_SUNRPC_GSS_MODULE */
1065 #ifdef CONFIG_NFSD_V4
1066                 [NFSD_Leasetime] = {"nfsv4leasetime", &transaction_ops, S_IWUSR|S_IRUSR},
1067                 [NFSD_Gracetime] = {"nfsv4gracetime", &transaction_ops, S_IWUSR|S_IRUSR},
1068                 [NFSD_RecoveryDir] = {"nfsv4recoverydir", &transaction_ops, S_IWUSR|S_IRUSR},
1069 #endif
1070                 /* last one */ {""}
1071         };
1072         struct net *net = data;
1073         int ret;
1074
1075         ret = simple_fill_super(sb, 0x6e667364, nfsd_files);
1076         if (ret)
1077                 return ret;
1078         sb->s_fs_info = get_net(net);
1079         return 0;
1080 }
1081
1082 static struct dentry *nfsd_mount(struct file_system_type *fs_type,
1083         int flags, const char *dev_name, void *data)
1084 {
1085         return mount_ns(fs_type, flags, current->nsproxy->net_ns, nfsd_fill_super);
1086 }
1087
1088 static void nfsd_umount(struct super_block *sb)
1089 {
1090         struct net *net = sb->s_fs_info;
1091
1092         kill_litter_super(sb);
1093         put_net(net);
1094 }
1095
1096 static struct file_system_type nfsd_fs_type = {
1097         .owner          = THIS_MODULE,
1098         .name           = "nfsd",
1099         .mount          = nfsd_mount,
1100         .kill_sb        = nfsd_umount,
1101 };
1102 MODULE_ALIAS_FS("nfsd");
1103
1104 #ifdef CONFIG_PROC_FS
1105 static int create_proc_exports_entry(void)
1106 {
1107         struct proc_dir_entry *entry;
1108
1109         entry = proc_mkdir("fs/nfs", NULL);
1110         if (!entry)
1111                 return -ENOMEM;
1112         entry = proc_create("exports", 0, entry,
1113                                  &exports_proc_operations);
1114         if (!entry) {
1115                 remove_proc_entry("fs/nfs", NULL);
1116                 return -ENOMEM;
1117         }
1118         return 0;
1119 }
1120 #else /* CONFIG_PROC_FS */
1121 static int create_proc_exports_entry(void)
1122 {
1123         return 0;
1124 }
1125 #endif
1126
1127 int nfsd_net_id;
1128
1129 static __net_init int nfsd_init_net(struct net *net)
1130 {
1131         int retval;
1132         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1133
1134         retval = nfsd_export_init(net);
1135         if (retval)
1136                 goto out_export_error;
1137         retval = nfsd_idmap_init(net);
1138         if (retval)
1139                 goto out_idmap_error;
1140         nn->nfsd4_lease = 90;   /* default lease time */
1141         nn->nfsd4_grace = 90;
1142         return 0;
1143
1144 out_idmap_error:
1145         nfsd_export_shutdown(net);
1146 out_export_error:
1147         return retval;
1148 }
1149
1150 static __net_exit void nfsd_exit_net(struct net *net)
1151 {
1152         nfsd_idmap_shutdown(net);
1153         nfsd_export_shutdown(net);
1154 }
1155
1156 static struct pernet_operations nfsd_net_ops = {
1157         .init = nfsd_init_net,
1158         .exit = nfsd_exit_net,
1159         .id   = &nfsd_net_id,
1160         .size = sizeof(struct nfsd_net),
1161 };
1162
1163 static int __init init_nfsd(void)
1164 {
1165         int retval;
1166         printk(KERN_INFO "Installing knfsd (copyright (C) 1996 okir@monad.swb.de).\n");
1167
1168         retval = register_cld_notifier();
1169         if (retval)
1170                 return retval;
1171         retval = register_pernet_subsys(&nfsd_net_ops);
1172         if (retval < 0)
1173                 goto out_unregister_notifier;
1174         retval = nfsd4_init_slabs();
1175         if (retval)
1176                 goto out_unregister_pernet;
1177         nfs4_state_init();
1178         retval = nfsd_fault_inject_init(); /* nfsd fault injection controls */
1179         if (retval)
1180                 goto out_free_slabs;
1181         nfsd_stat_init();       /* Statistics */
1182         retval = nfsd_reply_cache_init();
1183         if (retval)
1184                 goto out_free_stat;
1185         nfsd_lockd_init();      /* lockd->nfsd callbacks */
1186         retval = create_proc_exports_entry();
1187         if (retval)
1188                 goto out_free_lockd;
1189         retval = register_filesystem(&nfsd_fs_type);
1190         if (retval)
1191                 goto out_free_all;
1192         return 0;
1193 out_free_all:
1194         remove_proc_entry("fs/nfs/exports", NULL);
1195         remove_proc_entry("fs/nfs", NULL);
1196 out_free_lockd:
1197         nfsd_lockd_shutdown();
1198         nfsd_reply_cache_shutdown();
1199 out_free_stat:
1200         nfsd_stat_shutdown();
1201         nfsd_fault_inject_cleanup();
1202 out_free_slabs:
1203         nfsd4_free_slabs();
1204 out_unregister_pernet:
1205         unregister_pernet_subsys(&nfsd_net_ops);
1206 out_unregister_notifier:
1207         unregister_cld_notifier();
1208         return retval;
1209 }
1210
1211 static void __exit exit_nfsd(void)
1212 {
1213         nfsd_reply_cache_shutdown();
1214         remove_proc_entry("fs/nfs/exports", NULL);
1215         remove_proc_entry("fs/nfs", NULL);
1216         nfsd_stat_shutdown();
1217         nfsd_lockd_shutdown();
1218         nfsd4_free_slabs();
1219         nfsd_fault_inject_cleanup();
1220         unregister_filesystem(&nfsd_fs_type);
1221         unregister_pernet_subsys(&nfsd_net_ops);
1222         unregister_cld_notifier();
1223 }
1224
1225 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1226 MODULE_LICENSE("GPL");
1227 module_init(init_nfsd)
1228 module_exit(exit_nfsd)