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