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 | |
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 | |
765927b2 | 22 | static int get_name(const struct path *path, char *name, struct dentry *child); |
10f11c34 CH |
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; |
765927b2 | 29 | struct path path = {.mnt = mnt, .dentry = dir}; |
10f11c34 CH |
30 | |
31 | if (nop->get_name) | |
32 | return nop->get_name(dir, name, child); | |
33 | else | |
765927b2 | 34 | return get_name(&path, 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 |
40 | static struct dentry * |
41 | find_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); | |
b67bfe0d | 53 | hlist_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 | */ | |
75 | static struct dentry * | |
76 | find_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 | 98 | static int |
f3f8e175 | 99 | reconnect_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 |
214 | struct 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 | */ | |
226 | static 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 | */ | |
765927b2 | 251 | static int get_name(const struct path *path, char *name, struct dentry *child) |
1da177e4 | 252 | { |
745ca247 | 253 | const struct cred *cred = current_cred(); |
765927b2 | 254 | struct inode *dir = path->dentry->d_inode; |
1da177e4 LT |
255 | int error; |
256 | struct file *file; | |
257 | struct getdents_callback buffer; | |
258 | ||
259 | error = -ENOTDIR; | |
260 | if (!dir || !S_ISDIR(dir->i_mode)) | |
261 | goto out; | |
262 | error = -EINVAL; | |
263 | if (!dir->i_fop) | |
264 | goto out; | |
265 | /* | |
266 | * Open the directory ... | |
267 | */ | |
765927b2 | 268 | file = dentry_open(path, O_RDONLY, cred); |
1da177e4 LT |
269 | error = PTR_ERR(file); |
270 | if (IS_ERR(file)) | |
271 | goto out; | |
272 | ||
273 | error = -EINVAL; | |
274 | if (!file->f_op->readdir) | |
275 | goto out_close; | |
276 | ||
277 | buffer.name = name; | |
278 | buffer.ino = child->d_inode->i_ino; | |
279 | buffer.found = 0; | |
280 | buffer.sequence = 0; | |
281 | while (1) { | |
282 | int old_seq = buffer.sequence; | |
283 | ||
284 | error = vfs_readdir(file, filldir_one, &buffer); | |
53c9c5c0 AV |
285 | if (buffer.found) { |
286 | error = 0; | |
287 | break; | |
288 | } | |
1da177e4 LT |
289 | |
290 | if (error < 0) | |
291 | break; | |
292 | ||
1da177e4 LT |
293 | error = -ENOENT; |
294 | if (old_seq == buffer.sequence) | |
295 | break; | |
296 | } | |
297 | ||
298 | out_close: | |
299 | fput(file); | |
300 | out: | |
301 | return error; | |
302 | } | |
303 | ||
1da177e4 LT |
304 | /** |
305 | * export_encode_fh - default export_operations->encode_fh function | |
b0b0382b | 306 | * @inode: the object to encode |
1da177e4 LT |
307 | * @fh: where to store the file handle fragment |
308 | * @max_len: maximum length to store there | |
b0b0382b | 309 | * @parent: parent directory inode, if wanted |
1da177e4 LT |
310 | * |
311 | * This default encode_fh function assumes that the 32 inode number | |
312 | * is suitable for locating an inode, and that the generation number | |
313 | * can be used to check that it is still valid. It places them in the | |
314 | * filehandle fragment where export_decode_fh expects to find them. | |
315 | */ | |
b0b0382b AV |
316 | static int export_encode_fh(struct inode *inode, struct fid *fid, |
317 | int *max_len, struct inode *parent) | |
1da177e4 | 318 | { |
1da177e4 | 319 | int len = *max_len; |
6e91ea2b | 320 | int type = FILEID_INO32_GEN; |
5fe0c237 | 321 | |
b0b0382b | 322 | if (parent && (len < 4)) { |
5fe0c237 | 323 | *max_len = 4; |
216b6cbd | 324 | return FILEID_INVALID; |
5fe0c237 AK |
325 | } else if (len < 2) { |
326 | *max_len = 2; | |
216b6cbd | 327 | return FILEID_INVALID; |
5fe0c237 | 328 | } |
1da177e4 LT |
329 | |
330 | len = 2; | |
6e91ea2b CH |
331 | fid->i32.ino = inode->i_ino; |
332 | fid->i32.gen = inode->i_generation; | |
b0b0382b | 333 | if (parent) { |
6e91ea2b CH |
334 | fid->i32.parent_ino = parent->i_ino; |
335 | fid->i32.parent_gen = parent->i_generation; | |
1da177e4 | 336 | len = 4; |
6e91ea2b | 337 | type = FILEID_INO32_GEN_PARENT; |
1da177e4 LT |
338 | } |
339 | *max_len = len; | |
340 | return type; | |
341 | } | |
342 | ||
711c7bf9 CG |
343 | int exportfs_encode_inode_fh(struct inode *inode, struct fid *fid, |
344 | int *max_len, struct inode *parent) | |
345 | { | |
346 | const struct export_operations *nop = inode->i_sb->s_export_op; | |
347 | ||
348 | if (nop && nop->encode_fh) | |
349 | return nop->encode_fh(inode, fid->raw, max_len, parent); | |
350 | ||
351 | return export_encode_fh(inode, fid, max_len, parent); | |
352 | } | |
353 | EXPORT_SYMBOL_GPL(exportfs_encode_inode_fh); | |
354 | ||
6e91ea2b | 355 | int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len, |
d37065cd CH |
356 | int connectable) |
357 | { | |
10f11c34 | 358 | int error; |
b0b0382b AV |
359 | struct dentry *p = NULL; |
360 | struct inode *inode = dentry->d_inode, *parent = NULL; | |
d37065cd | 361 | |
b0b0382b AV |
362 | if (connectable && !S_ISDIR(inode->i_mode)) { |
363 | p = dget_parent(dentry); | |
364 | /* | |
365 | * note that while p might've ceased to be our parent already, | |
366 | * it's still pinned by and still positive. | |
367 | */ | |
368 | parent = p->d_inode; | |
369 | } | |
711c7bf9 CG |
370 | |
371 | error = exportfs_encode_inode_fh(inode, fid, max_len, parent); | |
b0b0382b | 372 | dput(p); |
10f11c34 CH |
373 | |
374 | return error; | |
d37065cd CH |
375 | } |
376 | EXPORT_SYMBOL_GPL(exportfs_encode_fh); | |
377 | ||
6e91ea2b CH |
378 | struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid, |
379 | int fh_len, int fileid_type, | |
380 | int (*acceptable)(void *, struct dentry *), void *context) | |
d37065cd | 381 | { |
39655164 | 382 | const struct export_operations *nop = mnt->mnt_sb->s_export_op; |
2596110a | 383 | struct dentry *result, *alias; |
f3f8e175 | 384 | char nbuf[NAME_MAX+1]; |
2596110a | 385 | int err; |
d37065cd | 386 | |
2596110a CH |
387 | /* |
388 | * Try to get any dentry for the given file handle from the filesystem. | |
389 | */ | |
becfd1f3 AK |
390 | if (!nop || !nop->fh_to_dentry) |
391 | return ERR_PTR(-ESTALE); | |
2596110a | 392 | result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type); |
a4f4d6df BF |
393 | if (!result) |
394 | result = ERR_PTR(-ESTALE); | |
2596110a CH |
395 | if (IS_ERR(result)) |
396 | return result; | |
397 | ||
398 | if (S_ISDIR(result->d_inode->i_mode)) { | |
399 | /* | |
400 | * This request is for a directory. | |
401 | * | |
402 | * On the positive side there is only one dentry for each | |
403 | * directory inode. On the negative side this implies that we | |
404 | * to ensure our dentry is connected all the way up to the | |
405 | * filesystem root. | |
406 | */ | |
407 | if (result->d_flags & DCACHE_DISCONNECTED) { | |
f3f8e175 | 408 | err = reconnect_path(mnt, result, nbuf); |
2596110a CH |
409 | if (err) |
410 | goto err_result; | |
411 | } | |
412 | ||
413 | if (!acceptable(context, result)) { | |
414 | err = -EACCES; | |
415 | goto err_result; | |
416 | } | |
417 | ||
418 | return result; | |
10f11c34 | 419 | } else { |
2596110a CH |
420 | /* |
421 | * It's not a directory. Life is a little more complicated. | |
422 | */ | |
423 | struct dentry *target_dir, *nresult; | |
2596110a CH |
424 | |
425 | /* | |
426 | * See if either the dentry we just got from the filesystem | |
427 | * or any alias for it is acceptable. This is always true | |
428 | * if this filesystem is exported without the subtreecheck | |
429 | * option. If the filesystem is exported with the subtree | |
430 | * check option there's a fair chance we need to look at | |
431 | * the parent directory in the file handle and make sure | |
432 | * it's connected to the filesystem root. | |
433 | */ | |
434 | alias = find_acceptable_alias(result, acceptable, context); | |
435 | if (alias) | |
436 | return alias; | |
437 | ||
438 | /* | |
439 | * Try to extract a dentry for the parent directory from the | |
440 | * file handle. If this fails we'll have to give up. | |
441 | */ | |
442 | err = -ESTALE; | |
443 | if (!nop->fh_to_parent) | |
444 | goto err_result; | |
445 | ||
446 | target_dir = nop->fh_to_parent(mnt->mnt_sb, fid, | |
447 | fh_len, fileid_type); | |
a4f4d6df BF |
448 | if (!target_dir) |
449 | goto err_result; | |
2596110a CH |
450 | err = PTR_ERR(target_dir); |
451 | if (IS_ERR(target_dir)) | |
452 | goto err_result; | |
453 | ||
454 | /* | |
455 | * And as usual we need to make sure the parent directory is | |
456 | * connected to the filesystem root. The VFS really doesn't | |
457 | * like disconnected directories.. | |
458 | */ | |
f3f8e175 | 459 | err = reconnect_path(mnt, target_dir, nbuf); |
2596110a CH |
460 | if (err) { |
461 | dput(target_dir); | |
462 | goto err_result; | |
463 | } | |
464 | ||
465 | /* | |
466 | * Now that we've got both a well-connected parent and a | |
467 | * dentry for the inode we're after, make sure that our | |
468 | * inode is actually connected to the parent. | |
469 | */ | |
e38f9817 | 470 | err = exportfs_get_name(mnt, target_dir, nbuf, result); |
2596110a CH |
471 | if (!err) { |
472 | mutex_lock(&target_dir->d_inode->i_mutex); | |
473 | nresult = lookup_one_len(nbuf, target_dir, | |
474 | strlen(nbuf)); | |
475 | mutex_unlock(&target_dir->d_inode->i_mutex); | |
476 | if (!IS_ERR(nresult)) { | |
477 | if (nresult->d_inode) { | |
478 | dput(result); | |
479 | result = nresult; | |
480 | } else | |
481 | dput(nresult); | |
482 | } | |
483 | } | |
484 | ||
485 | /* | |
486 | * At this point we are done with the parent, but it's pinned | |
487 | * by the child dentry anyway. | |
488 | */ | |
489 | dput(target_dir); | |
490 | ||
491 | /* | |
492 | * And finally make sure the dentry is actually acceptable | |
493 | * to NFSD. | |
494 | */ | |
495 | alias = find_acceptable_alias(result, acceptable, context); | |
496 | if (!alias) { | |
497 | err = -EACCES; | |
498 | goto err_result; | |
499 | } | |
500 | ||
501 | return alias; | |
10f11c34 CH |
502 | } |
503 | ||
2596110a CH |
504 | err_result: |
505 | dput(result); | |
506 | return ERR_PTR(err); | |
d37065cd CH |
507 | } |
508 | EXPORT_SYMBOL_GPL(exportfs_decode_fh); | |
509 | ||
1da177e4 | 510 | MODULE_LICENSE("GPL"); |