Merge branch 'gpio/merge' of git://git.secretlab.ca/git/linux-2.6 into grant
[linux-2.6-block.git] / fs / exportfs / expfs.c
CommitLineData
e38f9817
CH
1/*
2 * Copyright (C) Neil Brown 2002
3 * Copyright (C) Christoph Hellwig 2007
4 *
5 * This file contains the code mapping from inodes to NFS file handles,
6 * and for mapping back from file handles to dentries.
7 *
8 * For details on why we do all the strange and hairy things in here
dc7a0816 9 * take a look at Documentation/filesystems/nfs/Exporting.
e38f9817 10 */
a5694255 11#include <linux/exportfs.h>
1da177e4
LT
12#include <linux/fs.h>
13#include <linux/file.h>
14#include <linux/module.h>
d37065cd 15#include <linux/mount.h>
1da177e4 16#include <linux/namei.h>
745ca247 17#include <linux/sched.h>
1da177e4 18
10f11c34 19#define dprintk(fmt, args...) do{}while(0)
1da177e4 20
1da177e4 21
e38f9817 22static int get_name(struct vfsmount *mnt, struct dentry *dentry, char *name,
10f11c34
CH
23 struct dentry *child);
24
25
e38f9817
CH
26static int exportfs_get_name(struct vfsmount *mnt, struct dentry *dir,
27 char *name, struct dentry *child)
10f11c34 28{
39655164 29 const struct export_operations *nop = dir->d_sb->s_export_op;
10f11c34
CH
30
31 if (nop->get_name)
32 return nop->get_name(dir, name, child);
33 else
e38f9817 34 return get_name(mnt, dir, name, child);
10f11c34 35}
1da177e4 36
fb66a198
CH
37/*
38 * Check if the dentry or any of it's aliases is acceptable.
39 */
e2f99018
CH
40static struct dentry *
41find_acceptable_alias(struct dentry *result,
42 int (*acceptable)(void *context, struct dentry *dentry),
43 void *context)
44{
45 struct dentry *dentry, *toput = NULL;
873feea0 46 struct inode *inode;
e2f99018 47
fb66a198
CH
48 if (acceptable(context, result))
49 return result;
50
873feea0
NP
51 inode = result->d_inode;
52 spin_lock(&inode->i_lock);
53 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
dc0474be 54 dget(dentry);
873feea0 55 spin_unlock(&inode->i_lock);
e2f99018
CH
56 if (toput)
57 dput(toput);
58 if (dentry != result && acceptable(context, dentry)) {
59 dput(result);
60 return dentry;
61 }
873feea0 62 spin_lock(&inode->i_lock);
e2f99018
CH
63 toput = dentry;
64 }
873feea0 65 spin_unlock(&inode->i_lock);
e2f99018
CH
66
67 if (toput)
68 dput(toput);
69 return NULL;
70}
71
dd90b509
CH
72/*
73 * Find root of a disconnected subtree and return a reference to it.
74 */
75static struct dentry *
76find_disconnected_root(struct dentry *dentry)
77{
78 dget(dentry);
0461ee26
CH
79 while (!IS_ROOT(dentry)) {
80 struct dentry *parent = dget_parent(dentry);
81
82 if (!(parent->d_flags & DCACHE_DISCONNECTED)) {
83 dput(parent);
84 break;
85 }
86
dd90b509
CH
87 dput(dentry);
88 dentry = parent;
dd90b509 89 }
dd90b509
CH
90 return dentry;
91}
92
019ab801
CH
93/*
94 * Make sure target_dir is fully connected to the dentry tree.
1da177e4 95 *
019ab801 96 * It may already be, as the flag isn't always updated when connection happens.
1da177e4 97 */
019ab801 98static int
f3f8e175 99reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf)
1da177e4 100{
019ab801
CH
101 int noprogress = 0;
102 int err = -ESTALE;
1da177e4
LT
103
104 /*
019ab801 105 * It is possible that a confused file system might not let us complete
1da177e4
LT
106 * the path to the root. For example, if get_parent returns a directory
107 * in which we cannot find a name for the child. While this implies a
108 * very sick filesystem we don't want it to cause knfsd to spin. Hence
109 * the noprogress counter. If we go through the loop 10 times (2 is
110 * probably enough) without getting anywhere, we just give up
111 */
1da177e4 112 while (target_dir->d_flags & DCACHE_DISCONNECTED && noprogress++ < 10) {
dd90b509 113 struct dentry *pd = find_disconnected_root(target_dir);
1da177e4
LT
114
115 if (!IS_ROOT(pd)) {
116 /* must have found a connected parent - great */
117 spin_lock(&pd->d_lock);
118 pd->d_flags &= ~DCACHE_DISCONNECTED;
119 spin_unlock(&pd->d_lock);
120 noprogress = 0;
e38f9817 121 } else if (pd == mnt->mnt_sb->s_root) {
1da177e4
LT
122 printk(KERN_ERR "export: Eeek filesystem root is not connected, impossible\n");
123 spin_lock(&pd->d_lock);
124 pd->d_flags &= ~DCACHE_DISCONNECTED;
125 spin_unlock(&pd->d_lock);
126 noprogress = 0;
127 } else {
10f11c34
CH
128 /*
129 * We have hit the top of a disconnected path, try to
130 * find parent and connect.
131 *
132 * Racing with some other process renaming a directory
133 * isn't much of a problem here. If someone renames
134 * the directory, it will end up properly connected,
135 * which is what we want
136 *
137 * Getting the parent can't be supported generically,
138 * the locking is too icky.
139 *
140 * Instead we just return EACCES. If server reboots
141 * or inodes get flushed, you lose
1da177e4 142 */
10f11c34 143 struct dentry *ppd = ERR_PTR(-EACCES);
1da177e4
LT
144 struct dentry *npd;
145
1b1dcc1b 146 mutex_lock(&pd->d_inode->i_mutex);
e38f9817
CH
147 if (mnt->mnt_sb->s_export_op->get_parent)
148 ppd = mnt->mnt_sb->s_export_op->get_parent(pd);
1b1dcc1b 149 mutex_unlock(&pd->d_inode->i_mutex);
1da177e4
LT
150
151 if (IS_ERR(ppd)) {
152 err = PTR_ERR(ppd);
019ab801 153 dprintk("%s: get_parent of %ld failed, err %d\n",
8e24eea7 154 __func__, pd->d_inode->i_ino, err);
1da177e4
LT
155 dput(pd);
156 break;
157 }
019ab801 158
8e24eea7 159 dprintk("%s: find name of %lu in %lu\n", __func__,
019ab801 160 pd->d_inode->i_ino, ppd->d_inode->i_ino);
e38f9817 161 err = exportfs_get_name(mnt, ppd, nbuf, pd);
1da177e4
LT
162 if (err) {
163 dput(ppd);
164 dput(pd);
165 if (err == -ENOENT)
166 /* some race between get_parent and
167 * get_name? just try again
168 */
169 continue;
170 break;
171 }
8e24eea7 172 dprintk("%s: found name: %s\n", __func__, nbuf);
1b1dcc1b 173 mutex_lock(&ppd->d_inode->i_mutex);
1da177e4 174 npd = lookup_one_len(nbuf, ppd, strlen(nbuf));
1b1dcc1b 175 mutex_unlock(&ppd->d_inode->i_mutex);
1da177e4
LT
176 if (IS_ERR(npd)) {
177 err = PTR_ERR(npd);
019ab801 178 dprintk("%s: lookup failed: %d\n",
8e24eea7 179 __func__, err);
1da177e4
LT
180 dput(ppd);
181 dput(pd);
182 break;
183 }
184 /* we didn't really want npd, we really wanted
185 * a side-effect of the lookup.
186 * hopefully, npd == pd, though it isn't really
187 * a problem if it isn't
188 */
189 if (npd == pd)
190 noprogress = 0;
191 else
8e24eea7 192 printk("%s: npd != pd\n", __func__);
1da177e4
LT
193 dput(npd);
194 dput(ppd);
195 if (IS_ROOT(pd)) {
196 /* something went wrong, we have to give up */
197 dput(pd);
198 break;
199 }
200 }
201 dput(pd);
202 }
203
204 if (target_dir->d_flags & DCACHE_DISCONNECTED) {
205 /* something went wrong - oh-well */
206 if (!err)
207 err = -ESTALE;
019ab801 208 return err;
1da177e4 209 }
019ab801
CH
210
211 return 0;
212}
213
1da177e4
LT
214struct getdents_callback {
215 char *name; /* name that was found. It already points to a
216 buffer NAME_MAX+1 is size */
217 unsigned long ino; /* the inum we are looking for */
218 int found; /* inode matched? */
219 int sequence; /* sequence counter */
220};
221
222/*
223 * A rather strange filldir function to capture
224 * the name matching the specified inode number.
225 */
226static int filldir_one(void * __buf, const char * name, int len,
afefdbb2 227 loff_t pos, u64 ino, unsigned int d_type)
1da177e4
LT
228{
229 struct getdents_callback *buf = __buf;
230 int result = 0;
231
232 buf->sequence++;
233 if (buf->ino == ino) {
234 memcpy(buf->name, name, len);
235 buf->name[len] = '\0';
236 buf->found = 1;
237 result = -1;
238 }
239 return result;
240}
241
242/**
243 * get_name - default export_operations->get_name function
244 * @dentry: the directory in which to find a name
245 * @name: a pointer to a %NAME_MAX+1 char buffer to store the name
246 * @child: the dentry for the child directory.
247 *
248 * calls readdir on the parent until it finds an entry with
249 * the same inode number as the child, and returns that.
250 */
e38f9817
CH
251static int get_name(struct vfsmount *mnt, struct dentry *dentry,
252 char *name, struct dentry *child)
1da177e4 253{
745ca247 254 const struct cred *cred = current_cred();
1da177e4
LT
255 struct inode *dir = dentry->d_inode;
256 int error;
257 struct file *file;
258 struct getdents_callback buffer;
259
260 error = -ENOTDIR;
261 if (!dir || !S_ISDIR(dir->i_mode))
262 goto out;
263 error = -EINVAL;
264 if (!dir->i_fop)
265 goto out;
266 /*
267 * Open the directory ...
268 */
745ca247 269 file = dentry_open(dget(dentry), mntget(mnt), O_RDONLY, cred);
1da177e4
LT
270 error = PTR_ERR(file);
271 if (IS_ERR(file))
272 goto out;
273
274 error = -EINVAL;
275 if (!file->f_op->readdir)
276 goto out_close;
277
278 buffer.name = name;
279 buffer.ino = child->d_inode->i_ino;
280 buffer.found = 0;
281 buffer.sequence = 0;
282 while (1) {
283 int old_seq = buffer.sequence;
284
285 error = vfs_readdir(file, filldir_one, &buffer);
53c9c5c0
AV
286 if (buffer.found) {
287 error = 0;
288 break;
289 }
1da177e4
LT
290
291 if (error < 0)
292 break;
293
1da177e4
LT
294 error = -ENOENT;
295 if (old_seq == buffer.sequence)
296 break;
297 }
298
299out_close:
300 fput(file);
301out:
302 return error;
303}
304
1da177e4
LT
305/**
306 * export_encode_fh - default export_operations->encode_fh function
b0b0382b 307 * @inode: the object to encode
1da177e4
LT
308 * @fh: where to store the file handle fragment
309 * @max_len: maximum length to store there
b0b0382b 310 * @parent: parent directory inode, if wanted
1da177e4
LT
311 *
312 * This default encode_fh function assumes that the 32 inode number
313 * is suitable for locating an inode, and that the generation number
314 * can be used to check that it is still valid. It places them in the
315 * filehandle fragment where export_decode_fh expects to find them.
316 */
b0b0382b
AV
317static int export_encode_fh(struct inode *inode, struct fid *fid,
318 int *max_len, struct inode *parent)
1da177e4 319{
1da177e4 320 int len = *max_len;
6e91ea2b 321 int type = FILEID_INO32_GEN;
5fe0c237 322
b0b0382b 323 if (parent && (len < 4)) {
5fe0c237
AK
324 *max_len = 4;
325 return 255;
326 } else if (len < 2) {
327 *max_len = 2;
1da177e4 328 return 255;
5fe0c237 329 }
1da177e4
LT
330
331 len = 2;
6e91ea2b
CH
332 fid->i32.ino = inode->i_ino;
333 fid->i32.gen = inode->i_generation;
b0b0382b 334 if (parent) {
6e91ea2b
CH
335 fid->i32.parent_ino = parent->i_ino;
336 fid->i32.parent_gen = parent->i_generation;
1da177e4 337 len = 4;
6e91ea2b 338 type = FILEID_INO32_GEN_PARENT;
1da177e4
LT
339 }
340 *max_len = len;
341 return type;
342}
343
6e91ea2b 344int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
d37065cd
CH
345 int connectable)
346{
39655164 347 const struct export_operations *nop = dentry->d_sb->s_export_op;
10f11c34 348 int error;
b0b0382b
AV
349 struct dentry *p = NULL;
350 struct inode *inode = dentry->d_inode, *parent = NULL;
d37065cd 351
b0b0382b
AV
352 if (connectable && !S_ISDIR(inode->i_mode)) {
353 p = dget_parent(dentry);
354 /*
355 * note that while p might've ceased to be our parent already,
356 * it's still pinned by and still positive.
357 */
358 parent = p->d_inode;
359 }
10f11c34 360 if (nop->encode_fh)
b0b0382b 361 error = nop->encode_fh(inode, fid->raw, max_len, parent);
10f11c34 362 else
b0b0382b
AV
363 error = export_encode_fh(inode, fid, max_len, parent);
364 dput(p);
10f11c34
CH
365
366 return error;
d37065cd
CH
367}
368EXPORT_SYMBOL_GPL(exportfs_encode_fh);
369
6e91ea2b
CH
370struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid,
371 int fh_len, int fileid_type,
372 int (*acceptable)(void *, struct dentry *), void *context)
d37065cd 373{
39655164 374 const struct export_operations *nop = mnt->mnt_sb->s_export_op;
2596110a 375 struct dentry *result, *alias;
f3f8e175 376 char nbuf[NAME_MAX+1];
2596110a 377 int err;
d37065cd 378
2596110a
CH
379 /*
380 * Try to get any dentry for the given file handle from the filesystem.
381 */
becfd1f3
AK
382 if (!nop || !nop->fh_to_dentry)
383 return ERR_PTR(-ESTALE);
2596110a 384 result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
a4f4d6df
BF
385 if (!result)
386 result = ERR_PTR(-ESTALE);
2596110a
CH
387 if (IS_ERR(result))
388 return result;
389
390 if (S_ISDIR(result->d_inode->i_mode)) {
391 /*
392 * This request is for a directory.
393 *
394 * On the positive side there is only one dentry for each
395 * directory inode. On the negative side this implies that we
396 * to ensure our dentry is connected all the way up to the
397 * filesystem root.
398 */
399 if (result->d_flags & DCACHE_DISCONNECTED) {
f3f8e175 400 err = reconnect_path(mnt, result, nbuf);
2596110a
CH
401 if (err)
402 goto err_result;
403 }
404
405 if (!acceptable(context, result)) {
406 err = -EACCES;
407 goto err_result;
408 }
409
410 return result;
10f11c34 411 } else {
2596110a
CH
412 /*
413 * It's not a directory. Life is a little more complicated.
414 */
415 struct dentry *target_dir, *nresult;
2596110a
CH
416
417 /*
418 * See if either the dentry we just got from the filesystem
419 * or any alias for it is acceptable. This is always true
420 * if this filesystem is exported without the subtreecheck
421 * option. If the filesystem is exported with the subtree
422 * check option there's a fair chance we need to look at
423 * the parent directory in the file handle and make sure
424 * it's connected to the filesystem root.
425 */
426 alias = find_acceptable_alias(result, acceptable, context);
427 if (alias)
428 return alias;
429
430 /*
431 * Try to extract a dentry for the parent directory from the
432 * file handle. If this fails we'll have to give up.
433 */
434 err = -ESTALE;
435 if (!nop->fh_to_parent)
436 goto err_result;
437
438 target_dir = nop->fh_to_parent(mnt->mnt_sb, fid,
439 fh_len, fileid_type);
a4f4d6df
BF
440 if (!target_dir)
441 goto err_result;
2596110a
CH
442 err = PTR_ERR(target_dir);
443 if (IS_ERR(target_dir))
444 goto err_result;
445
446 /*
447 * And as usual we need to make sure the parent directory is
448 * connected to the filesystem root. The VFS really doesn't
449 * like disconnected directories..
450 */
f3f8e175 451 err = reconnect_path(mnt, target_dir, nbuf);
2596110a
CH
452 if (err) {
453 dput(target_dir);
454 goto err_result;
455 }
456
457 /*
458 * Now that we've got both a well-connected parent and a
459 * dentry for the inode we're after, make sure that our
460 * inode is actually connected to the parent.
461 */
e38f9817 462 err = exportfs_get_name(mnt, target_dir, nbuf, result);
2596110a
CH
463 if (!err) {
464 mutex_lock(&target_dir->d_inode->i_mutex);
465 nresult = lookup_one_len(nbuf, target_dir,
466 strlen(nbuf));
467 mutex_unlock(&target_dir->d_inode->i_mutex);
468 if (!IS_ERR(nresult)) {
469 if (nresult->d_inode) {
470 dput(result);
471 result = nresult;
472 } else
473 dput(nresult);
474 }
475 }
476
477 /*
478 * At this point we are done with the parent, but it's pinned
479 * by the child dentry anyway.
480 */
481 dput(target_dir);
482
483 /*
484 * And finally make sure the dentry is actually acceptable
485 * to NFSD.
486 */
487 alias = find_acceptable_alias(result, acceptable, context);
488 if (!alias) {
489 err = -EACCES;
490 goto err_result;
491 }
492
493 return alias;
10f11c34
CH
494 }
495
2596110a
CH
496 err_result:
497 dput(result);
498 return ERR_PTR(err);
d37065cd
CH
499}
500EXPORT_SYMBOL_GPL(exportfs_decode_fh);
501
1da177e4 502MODULE_LICENSE("GPL");