]>
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 | |
e38f9817 | 22 | static int get_name(struct vfsmount *mnt, struct dentry *dentry, char *name, |
10f11c34 CH |
23 | struct dentry *child); |
24 | ||
25 | ||
e38f9817 CH |
26 | static 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 |
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; |
b3d9b7a3 | 47 | struct hlist_node *p; |
e2f99018 | 48 | |
fb66a198 CH |
49 | if (acceptable(context, result)) |
50 | return result; | |
51 | ||
873feea0 NP |
52 | inode = result->d_inode; |
53 | spin_lock(&inode->i_lock); | |
b3d9b7a3 | 54 | hlist_for_each_entry(dentry, p, &inode->i_dentry, d_alias) { |
dc0474be | 55 | dget(dentry); |
873feea0 | 56 | spin_unlock(&inode->i_lock); |
e2f99018 CH |
57 | if (toput) |
58 | dput(toput); | |
59 | if (dentry != result && acceptable(context, dentry)) { | |
60 | dput(result); | |
61 | return dentry; | |
62 | } | |
873feea0 | 63 | spin_lock(&inode->i_lock); |
e2f99018 CH |
64 | toput = dentry; |
65 | } | |
873feea0 | 66 | spin_unlock(&inode->i_lock); |
e2f99018 CH |
67 | |
68 | if (toput) | |
69 | dput(toput); | |
70 | return NULL; | |
71 | } | |
72 | ||
dd90b509 CH |
73 | /* |
74 | * Find root of a disconnected subtree and return a reference to it. | |
75 | */ | |
76 | static struct dentry * | |
77 | find_disconnected_root(struct dentry *dentry) | |
78 | { | |
79 | dget(dentry); | |
0461ee26 CH |
80 | while (!IS_ROOT(dentry)) { |
81 | struct dentry *parent = dget_parent(dentry); | |
82 | ||
83 | if (!(parent->d_flags & DCACHE_DISCONNECTED)) { | |
84 | dput(parent); | |
85 | break; | |
86 | } | |
87 | ||
dd90b509 CH |
88 | dput(dentry); |
89 | dentry = parent; | |
dd90b509 | 90 | } |
dd90b509 CH |
91 | return dentry; |
92 | } | |
93 | ||
019ab801 CH |
94 | /* |
95 | * Make sure target_dir is fully connected to the dentry tree. | |
1da177e4 | 96 | * |
019ab801 | 97 | * It may already be, as the flag isn't always updated when connection happens. |
1da177e4 | 98 | */ |
019ab801 | 99 | static int |
f3f8e175 | 100 | reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf) |
1da177e4 | 101 | { |
019ab801 CH |
102 | int noprogress = 0; |
103 | int err = -ESTALE; | |
1da177e4 LT |
104 | |
105 | /* | |
019ab801 | 106 | * It is possible that a confused file system might not let us complete |
1da177e4 LT |
107 | * the path to the root. For example, if get_parent returns a directory |
108 | * in which we cannot find a name for the child. While this implies a | |
109 | * very sick filesystem we don't want it to cause knfsd to spin. Hence | |
110 | * the noprogress counter. If we go through the loop 10 times (2 is | |
111 | * probably enough) without getting anywhere, we just give up | |
112 | */ | |
1da177e4 | 113 | while (target_dir->d_flags & DCACHE_DISCONNECTED && noprogress++ < 10) { |
dd90b509 | 114 | struct dentry *pd = find_disconnected_root(target_dir); |
1da177e4 LT |
115 | |
116 | if (!IS_ROOT(pd)) { | |
117 | /* must have found a connected parent - great */ | |
118 | spin_lock(&pd->d_lock); | |
119 | pd->d_flags &= ~DCACHE_DISCONNECTED; | |
120 | spin_unlock(&pd->d_lock); | |
121 | noprogress = 0; | |
e38f9817 | 122 | } else if (pd == mnt->mnt_sb->s_root) { |
1da177e4 LT |
123 | printk(KERN_ERR "export: Eeek filesystem root is not connected, impossible\n"); |
124 | spin_lock(&pd->d_lock); | |
125 | pd->d_flags &= ~DCACHE_DISCONNECTED; | |
126 | spin_unlock(&pd->d_lock); | |
127 | noprogress = 0; | |
128 | } else { | |
10f11c34 CH |
129 | /* |
130 | * We have hit the top of a disconnected path, try to | |
131 | * find parent and connect. | |
132 | * | |
133 | * Racing with some other process renaming a directory | |
134 | * isn't much of a problem here. If someone renames | |
135 | * the directory, it will end up properly connected, | |
136 | * which is what we want | |
137 | * | |
138 | * Getting the parent can't be supported generically, | |
139 | * the locking is too icky. | |
140 | * | |
141 | * Instead we just return EACCES. If server reboots | |
142 | * or inodes get flushed, you lose | |
1da177e4 | 143 | */ |
10f11c34 | 144 | struct dentry *ppd = ERR_PTR(-EACCES); |
1da177e4 LT |
145 | struct dentry *npd; |
146 | ||
1b1dcc1b | 147 | mutex_lock(&pd->d_inode->i_mutex); |
e38f9817 CH |
148 | if (mnt->mnt_sb->s_export_op->get_parent) |
149 | ppd = mnt->mnt_sb->s_export_op->get_parent(pd); | |
1b1dcc1b | 150 | mutex_unlock(&pd->d_inode->i_mutex); |
1da177e4 LT |
151 | |
152 | if (IS_ERR(ppd)) { | |
153 | err = PTR_ERR(ppd); | |
019ab801 | 154 | dprintk("%s: get_parent of %ld failed, err %d\n", |
8e24eea7 | 155 | __func__, pd->d_inode->i_ino, err); |
1da177e4 LT |
156 | dput(pd); |
157 | break; | |
158 | } | |
019ab801 | 159 | |
8e24eea7 | 160 | dprintk("%s: find name of %lu in %lu\n", __func__, |
019ab801 | 161 | pd->d_inode->i_ino, ppd->d_inode->i_ino); |
e38f9817 | 162 | err = exportfs_get_name(mnt, ppd, nbuf, pd); |
1da177e4 LT |
163 | if (err) { |
164 | dput(ppd); | |
165 | dput(pd); | |
166 | if (err == -ENOENT) | |
167 | /* some race between get_parent and | |
168 | * get_name? just try again | |
169 | */ | |
170 | continue; | |
171 | break; | |
172 | } | |
8e24eea7 | 173 | dprintk("%s: found name: %s\n", __func__, nbuf); |
1b1dcc1b | 174 | mutex_lock(&ppd->d_inode->i_mutex); |
1da177e4 | 175 | npd = lookup_one_len(nbuf, ppd, strlen(nbuf)); |
1b1dcc1b | 176 | mutex_unlock(&ppd->d_inode->i_mutex); |
1da177e4 LT |
177 | if (IS_ERR(npd)) { |
178 | err = PTR_ERR(npd); | |
019ab801 | 179 | dprintk("%s: lookup failed: %d\n", |
8e24eea7 | 180 | __func__, err); |
1da177e4 LT |
181 | dput(ppd); |
182 | dput(pd); | |
183 | break; | |
184 | } | |
185 | /* we didn't really want npd, we really wanted | |
186 | * a side-effect of the lookup. | |
187 | * hopefully, npd == pd, though it isn't really | |
188 | * a problem if it isn't | |
189 | */ | |
190 | if (npd == pd) | |
191 | noprogress = 0; | |
192 | else | |
8e24eea7 | 193 | printk("%s: npd != pd\n", __func__); |
1da177e4 LT |
194 | dput(npd); |
195 | dput(ppd); | |
196 | if (IS_ROOT(pd)) { | |
197 | /* something went wrong, we have to give up */ | |
198 | dput(pd); | |
199 | break; | |
200 | } | |
201 | } | |
202 | dput(pd); | |
203 | } | |
204 | ||
205 | if (target_dir->d_flags & DCACHE_DISCONNECTED) { | |
206 | /* something went wrong - oh-well */ | |
207 | if (!err) | |
208 | err = -ESTALE; | |
019ab801 | 209 | return err; |
1da177e4 | 210 | } |
019ab801 CH |
211 | |
212 | return 0; | |
213 | } | |
214 | ||
1da177e4 LT |
215 | struct getdents_callback { |
216 | char *name; /* name that was found. It already points to a | |
217 | buffer NAME_MAX+1 is size */ | |
218 | unsigned long ino; /* the inum we are looking for */ | |
219 | int found; /* inode matched? */ | |
220 | int sequence; /* sequence counter */ | |
221 | }; | |
222 | ||
223 | /* | |
224 | * A rather strange filldir function to capture | |
225 | * the name matching the specified inode number. | |
226 | */ | |
227 | static int filldir_one(void * __buf, const char * name, int len, | |
afefdbb2 | 228 | loff_t pos, u64 ino, unsigned int d_type) |
1da177e4 LT |
229 | { |
230 | struct getdents_callback *buf = __buf; | |
231 | int result = 0; | |
232 | ||
233 | buf->sequence++; | |
234 | if (buf->ino == ino) { | |
235 | memcpy(buf->name, name, len); | |
236 | buf->name[len] = '\0'; | |
237 | buf->found = 1; | |
238 | result = -1; | |
239 | } | |
240 | return result; | |
241 | } | |
242 | ||
243 | /** | |
244 | * get_name - default export_operations->get_name function | |
245 | * @dentry: the directory in which to find a name | |
246 | * @name: a pointer to a %NAME_MAX+1 char buffer to store the name | |
247 | * @child: the dentry for the child directory. | |
248 | * | |
249 | * calls readdir on the parent until it finds an entry with | |
250 | * the same inode number as the child, and returns that. | |
251 | */ | |
e38f9817 CH |
252 | static int get_name(struct vfsmount *mnt, struct dentry *dentry, |
253 | char *name, struct dentry *child) | |
1da177e4 | 254 | { |
745ca247 | 255 | const struct cred *cred = current_cred(); |
1da177e4 LT |
256 | struct inode *dir = dentry->d_inode; |
257 | int error; | |
258 | struct file *file; | |
259 | struct getdents_callback buffer; | |
260 | ||
261 | error = -ENOTDIR; | |
262 | if (!dir || !S_ISDIR(dir->i_mode)) | |
263 | goto out; | |
264 | error = -EINVAL; | |
265 | if (!dir->i_fop) | |
266 | goto out; | |
267 | /* | |
268 | * Open the directory ... | |
269 | */ | |
745ca247 | 270 | file = dentry_open(dget(dentry), mntget(mnt), O_RDONLY, cred); |
1da177e4 LT |
271 | error = PTR_ERR(file); |
272 | if (IS_ERR(file)) | |
273 | goto out; | |
274 | ||
275 | error = -EINVAL; | |
276 | if (!file->f_op->readdir) | |
277 | goto out_close; | |
278 | ||
279 | buffer.name = name; | |
280 | buffer.ino = child->d_inode->i_ino; | |
281 | buffer.found = 0; | |
282 | buffer.sequence = 0; | |
283 | while (1) { | |
284 | int old_seq = buffer.sequence; | |
285 | ||
286 | error = vfs_readdir(file, filldir_one, &buffer); | |
53c9c5c0 AV |
287 | if (buffer.found) { |
288 | error = 0; | |
289 | break; | |
290 | } | |
1da177e4 LT |
291 | |
292 | if (error < 0) | |
293 | break; | |
294 | ||
1da177e4 LT |
295 | error = -ENOENT; |
296 | if (old_seq == buffer.sequence) | |
297 | break; | |
298 | } | |
299 | ||
300 | out_close: | |
301 | fput(file); | |
302 | out: | |
303 | return error; | |
304 | } | |
305 | ||
1da177e4 LT |
306 | /** |
307 | * export_encode_fh - default export_operations->encode_fh function | |
b0b0382b | 308 | * @inode: the object to encode |
1da177e4 LT |
309 | * @fh: where to store the file handle fragment |
310 | * @max_len: maximum length to store there | |
b0b0382b | 311 | * @parent: parent directory inode, if wanted |
1da177e4 LT |
312 | * |
313 | * This default encode_fh function assumes that the 32 inode number | |
314 | * is suitable for locating an inode, and that the generation number | |
315 | * can be used to check that it is still valid. It places them in the | |
316 | * filehandle fragment where export_decode_fh expects to find them. | |
317 | */ | |
b0b0382b AV |
318 | static int export_encode_fh(struct inode *inode, struct fid *fid, |
319 | int *max_len, struct inode *parent) | |
1da177e4 | 320 | { |
1da177e4 | 321 | int len = *max_len; |
6e91ea2b | 322 | int type = FILEID_INO32_GEN; |
5fe0c237 | 323 | |
b0b0382b | 324 | if (parent && (len < 4)) { |
5fe0c237 AK |
325 | *max_len = 4; |
326 | return 255; | |
327 | } else if (len < 2) { | |
328 | *max_len = 2; | |
1da177e4 | 329 | return 255; |
5fe0c237 | 330 | } |
1da177e4 LT |
331 | |
332 | len = 2; | |
6e91ea2b CH |
333 | fid->i32.ino = inode->i_ino; |
334 | fid->i32.gen = inode->i_generation; | |
b0b0382b | 335 | if (parent) { |
6e91ea2b CH |
336 | fid->i32.parent_ino = parent->i_ino; |
337 | fid->i32.parent_gen = parent->i_generation; | |
1da177e4 | 338 | len = 4; |
6e91ea2b | 339 | type = FILEID_INO32_GEN_PARENT; |
1da177e4 LT |
340 | } |
341 | *max_len = len; | |
342 | return type; | |
343 | } | |
344 | ||
6e91ea2b | 345 | int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len, |
d37065cd CH |
346 | int connectable) |
347 | { | |
39655164 | 348 | const struct export_operations *nop = dentry->d_sb->s_export_op; |
10f11c34 | 349 | int error; |
b0b0382b AV |
350 | struct dentry *p = NULL; |
351 | struct inode *inode = dentry->d_inode, *parent = NULL; | |
d37065cd | 352 | |
b0b0382b AV |
353 | if (connectable && !S_ISDIR(inode->i_mode)) { |
354 | p = dget_parent(dentry); | |
355 | /* | |
356 | * note that while p might've ceased to be our parent already, | |
357 | * it's still pinned by and still positive. | |
358 | */ | |
359 | parent = p->d_inode; | |
360 | } | |
10f11c34 | 361 | if (nop->encode_fh) |
b0b0382b | 362 | error = nop->encode_fh(inode, fid->raw, max_len, parent); |
10f11c34 | 363 | else |
b0b0382b AV |
364 | error = export_encode_fh(inode, fid, max_len, parent); |
365 | dput(p); | |
10f11c34 CH |
366 | |
367 | return error; | |
d37065cd CH |
368 | } |
369 | EXPORT_SYMBOL_GPL(exportfs_encode_fh); | |
370 | ||
6e91ea2b CH |
371 | struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid, |
372 | int fh_len, int fileid_type, | |
373 | int (*acceptable)(void *, struct dentry *), void *context) | |
d37065cd | 374 | { |
39655164 | 375 | const struct export_operations *nop = mnt->mnt_sb->s_export_op; |
2596110a | 376 | struct dentry *result, *alias; |
f3f8e175 | 377 | char nbuf[NAME_MAX+1]; |
2596110a | 378 | int err; |
d37065cd | 379 | |
2596110a CH |
380 | /* |
381 | * Try to get any dentry for the given file handle from the filesystem. | |
382 | */ | |
becfd1f3 AK |
383 | if (!nop || !nop->fh_to_dentry) |
384 | return ERR_PTR(-ESTALE); | |
2596110a | 385 | result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type); |
a4f4d6df BF |
386 | if (!result) |
387 | result = ERR_PTR(-ESTALE); | |
2596110a CH |
388 | if (IS_ERR(result)) |
389 | return result; | |
390 | ||
391 | if (S_ISDIR(result->d_inode->i_mode)) { | |
392 | /* | |
393 | * This request is for a directory. | |
394 | * | |
395 | * On the positive side there is only one dentry for each | |
396 | * directory inode. On the negative side this implies that we | |
397 | * to ensure our dentry is connected all the way up to the | |
398 | * filesystem root. | |
399 | */ | |
400 | if (result->d_flags & DCACHE_DISCONNECTED) { | |
f3f8e175 | 401 | err = reconnect_path(mnt, result, nbuf); |
2596110a CH |
402 | if (err) |
403 | goto err_result; | |
404 | } | |
405 | ||
406 | if (!acceptable(context, result)) { | |
407 | err = -EACCES; | |
408 | goto err_result; | |
409 | } | |
410 | ||
411 | return result; | |
10f11c34 | 412 | } else { |
2596110a CH |
413 | /* |
414 | * It's not a directory. Life is a little more complicated. | |
415 | */ | |
416 | struct dentry *target_dir, *nresult; | |
2596110a CH |
417 | |
418 | /* | |
419 | * See if either the dentry we just got from the filesystem | |
420 | * or any alias for it is acceptable. This is always true | |
421 | * if this filesystem is exported without the subtreecheck | |
422 | * option. If the filesystem is exported with the subtree | |
423 | * check option there's a fair chance we need to look at | |
424 | * the parent directory in the file handle and make sure | |
425 | * it's connected to the filesystem root. | |
426 | */ | |
427 | alias = find_acceptable_alias(result, acceptable, context); | |
428 | if (alias) | |
429 | return alias; | |
430 | ||
431 | /* | |
432 | * Try to extract a dentry for the parent directory from the | |
433 | * file handle. If this fails we'll have to give up. | |
434 | */ | |
435 | err = -ESTALE; | |
436 | if (!nop->fh_to_parent) | |
437 | goto err_result; | |
438 | ||
439 | target_dir = nop->fh_to_parent(mnt->mnt_sb, fid, | |
440 | fh_len, fileid_type); | |
a4f4d6df BF |
441 | if (!target_dir) |
442 | goto err_result; | |
2596110a CH |
443 | err = PTR_ERR(target_dir); |
444 | if (IS_ERR(target_dir)) | |
445 | goto err_result; | |
446 | ||
447 | /* | |
448 | * And as usual we need to make sure the parent directory is | |
449 | * connected to the filesystem root. The VFS really doesn't | |
450 | * like disconnected directories.. | |
451 | */ | |
f3f8e175 | 452 | err = reconnect_path(mnt, target_dir, nbuf); |
2596110a CH |
453 | if (err) { |
454 | dput(target_dir); | |
455 | goto err_result; | |
456 | } | |
457 | ||
458 | /* | |
459 | * Now that we've got both a well-connected parent and a | |
460 | * dentry for the inode we're after, make sure that our | |
461 | * inode is actually connected to the parent. | |
462 | */ | |
e38f9817 | 463 | err = exportfs_get_name(mnt, target_dir, nbuf, result); |
2596110a CH |
464 | if (!err) { |
465 | mutex_lock(&target_dir->d_inode->i_mutex); | |
466 | nresult = lookup_one_len(nbuf, target_dir, | |
467 | strlen(nbuf)); | |
468 | mutex_unlock(&target_dir->d_inode->i_mutex); | |
469 | if (!IS_ERR(nresult)) { | |
470 | if (nresult->d_inode) { | |
471 | dput(result); | |
472 | result = nresult; | |
473 | } else | |
474 | dput(nresult); | |
475 | } | |
476 | } | |
477 | ||
478 | /* | |
479 | * At this point we are done with the parent, but it's pinned | |
480 | * by the child dentry anyway. | |
481 | */ | |
482 | dput(target_dir); | |
483 | ||
484 | /* | |
485 | * And finally make sure the dentry is actually acceptable | |
486 | * to NFSD. | |
487 | */ | |
488 | alias = find_acceptable_alias(result, acceptable, context); | |
489 | if (!alias) { | |
490 | err = -EACCES; | |
491 | goto err_result; | |
492 | } | |
493 | ||
494 | return alias; | |
10f11c34 CH |
495 | } |
496 | ||
2596110a CH |
497 | err_result: |
498 | dput(result); | |
499 | return ERR_PTR(err); | |
d37065cd CH |
500 | } |
501 | EXPORT_SYMBOL_GPL(exportfs_decode_fh); | |
502 | ||
1da177e4 | 503 | MODULE_LICENSE("GPL"); |