]>
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 | * |
78cee9a8 BF |
96 | * On successful return, DCACHE_DISCONNECTED will be cleared on |
97 | * target_dir, and target_dir->d_parent->...->d_parent will reach the | |
98 | * root of the filesystem. | |
99 | * | |
100 | * Whenever DCACHE_DISCONNECTED is unset, target_dir is fully connected. | |
101 | * But the converse is not true: target_dir may have DCACHE_DISCONNECTED | |
102 | * set but already be connected. In that case we'll verify the | |
103 | * connection to root and then clear the flag. | |
104 | * | |
105 | * Note that target_dir could be removed by a concurrent operation. In | |
106 | * that case reconnect_path may still succeed with target_dir fully | |
107 | * connected, but further operations using the filehandle will fail when | |
108 | * necessary (due to S_DEAD being set on the directory). | |
1da177e4 | 109 | */ |
019ab801 | 110 | static int |
f3f8e175 | 111 | reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf) |
1da177e4 | 112 | { |
019ab801 CH |
113 | int noprogress = 0; |
114 | int err = -ESTALE; | |
1da177e4 LT |
115 | |
116 | /* | |
019ab801 | 117 | * It is possible that a confused file system might not let us complete |
1da177e4 LT |
118 | * the path to the root. For example, if get_parent returns a directory |
119 | * in which we cannot find a name for the child. While this implies a | |
120 | * very sick filesystem we don't want it to cause knfsd to spin. Hence | |
121 | * the noprogress counter. If we go through the loop 10 times (2 is | |
122 | * probably enough) without getting anywhere, we just give up | |
123 | */ | |
1da177e4 | 124 | while (target_dir->d_flags & DCACHE_DISCONNECTED && noprogress++ < 10) { |
dd90b509 | 125 | struct dentry *pd = find_disconnected_root(target_dir); |
1da177e4 | 126 | |
854ff5ca CH |
127 | BUG_ON(pd == mnt->mnt_sb->s_root); |
128 | ||
1da177e4 LT |
129 | if (!IS_ROOT(pd)) { |
130 | /* must have found a connected parent - great */ | |
131 | spin_lock(&pd->d_lock); | |
132 | pd->d_flags &= ~DCACHE_DISCONNECTED; | |
133 | spin_unlock(&pd->d_lock); | |
134 | noprogress = 0; | |
1da177e4 | 135 | } else { |
10f11c34 CH |
136 | /* |
137 | * We have hit the top of a disconnected path, try to | |
138 | * find parent and connect. | |
139 | * | |
140 | * Racing with some other process renaming a directory | |
141 | * isn't much of a problem here. If someone renames | |
142 | * the directory, it will end up properly connected, | |
143 | * which is what we want | |
144 | * | |
145 | * Getting the parent can't be supported generically, | |
146 | * the locking is too icky. | |
147 | * | |
148 | * Instead we just return EACCES. If server reboots | |
149 | * or inodes get flushed, you lose | |
1da177e4 | 150 | */ |
10f11c34 | 151 | struct dentry *ppd = ERR_PTR(-EACCES); |
1da177e4 LT |
152 | struct dentry *npd; |
153 | ||
1b1dcc1b | 154 | mutex_lock(&pd->d_inode->i_mutex); |
e38f9817 CH |
155 | if (mnt->mnt_sb->s_export_op->get_parent) |
156 | ppd = mnt->mnt_sb->s_export_op->get_parent(pd); | |
1b1dcc1b | 157 | mutex_unlock(&pd->d_inode->i_mutex); |
1da177e4 LT |
158 | |
159 | if (IS_ERR(ppd)) { | |
160 | err = PTR_ERR(ppd); | |
019ab801 | 161 | dprintk("%s: get_parent of %ld failed, err %d\n", |
8e24eea7 | 162 | __func__, pd->d_inode->i_ino, err); |
1da177e4 LT |
163 | dput(pd); |
164 | break; | |
165 | } | |
019ab801 | 166 | |
8e24eea7 | 167 | dprintk("%s: find name of %lu in %lu\n", __func__, |
019ab801 | 168 | pd->d_inode->i_ino, ppd->d_inode->i_ino); |
e38f9817 | 169 | err = exportfs_get_name(mnt, ppd, nbuf, pd); |
1da177e4 LT |
170 | if (err) { |
171 | dput(ppd); | |
172 | dput(pd); | |
173 | if (err == -ENOENT) | |
174 | /* some race between get_parent and | |
175 | * get_name? just try again | |
176 | */ | |
177 | continue; | |
178 | break; | |
179 | } | |
8e24eea7 | 180 | dprintk("%s: found name: %s\n", __func__, nbuf); |
1b1dcc1b | 181 | mutex_lock(&ppd->d_inode->i_mutex); |
1da177e4 | 182 | npd = lookup_one_len(nbuf, ppd, strlen(nbuf)); |
1b1dcc1b | 183 | mutex_unlock(&ppd->d_inode->i_mutex); |
1da177e4 LT |
184 | if (IS_ERR(npd)) { |
185 | err = PTR_ERR(npd); | |
019ab801 | 186 | dprintk("%s: lookup failed: %d\n", |
8e24eea7 | 187 | __func__, err); |
1da177e4 LT |
188 | dput(ppd); |
189 | dput(pd); | |
190 | break; | |
191 | } | |
192 | /* we didn't really want npd, we really wanted | |
193 | * a side-effect of the lookup. | |
194 | * hopefully, npd == pd, though it isn't really | |
195 | * a problem if it isn't | |
196 | */ | |
197 | if (npd == pd) | |
198 | noprogress = 0; | |
199 | else | |
8e24eea7 | 200 | printk("%s: npd != pd\n", __func__); |
1da177e4 LT |
201 | dput(npd); |
202 | dput(ppd); | |
203 | if (IS_ROOT(pd)) { | |
204 | /* something went wrong, we have to give up */ | |
205 | dput(pd); | |
206 | break; | |
207 | } | |
208 | } | |
209 | dput(pd); | |
210 | } | |
211 | ||
212 | if (target_dir->d_flags & DCACHE_DISCONNECTED) { | |
213 | /* something went wrong - oh-well */ | |
214 | if (!err) | |
215 | err = -ESTALE; | |
019ab801 | 216 | return err; |
1da177e4 | 217 | } |
019ab801 CH |
218 | |
219 | return 0; | |
220 | } | |
221 | ||
1da177e4 | 222 | struct getdents_callback { |
5c0ba4e0 | 223 | struct dir_context ctx; |
1da177e4 LT |
224 | char *name; /* name that was found. It already points to a |
225 | buffer NAME_MAX+1 is size */ | |
950ee956 | 226 | u64 ino; /* the inum we are looking for */ |
1da177e4 LT |
227 | int found; /* inode matched? */ |
228 | int sequence; /* sequence counter */ | |
229 | }; | |
230 | ||
231 | /* | |
232 | * A rather strange filldir function to capture | |
233 | * the name matching the specified inode number. | |
234 | */ | |
235 | static int filldir_one(void * __buf, const char * name, int len, | |
afefdbb2 | 236 | loff_t pos, u64 ino, unsigned int d_type) |
1da177e4 LT |
237 | { |
238 | struct getdents_callback *buf = __buf; | |
239 | int result = 0; | |
240 | ||
241 | buf->sequence++; | |
dfc59e2c | 242 | if (buf->ino == ino && len <= NAME_MAX) { |
1da177e4 LT |
243 | memcpy(buf->name, name, len); |
244 | buf->name[len] = '\0'; | |
245 | buf->found = 1; | |
246 | result = -1; | |
247 | } | |
248 | return result; | |
249 | } | |
250 | ||
251 | /** | |
252 | * get_name - default export_operations->get_name function | |
253 | * @dentry: the directory in which to find a name | |
254 | * @name: a pointer to a %NAME_MAX+1 char buffer to store the name | |
255 | * @child: the dentry for the child directory. | |
256 | * | |
257 | * calls readdir on the parent until it finds an entry with | |
258 | * the same inode number as the child, and returns that. | |
259 | */ | |
765927b2 | 260 | static int get_name(const struct path *path, char *name, struct dentry *child) |
1da177e4 | 261 | { |
745ca247 | 262 | const struct cred *cred = current_cred(); |
765927b2 | 263 | struct inode *dir = path->dentry->d_inode; |
1da177e4 LT |
264 | int error; |
265 | struct file *file; | |
950ee956 BF |
266 | struct kstat stat; |
267 | struct path child_path = { | |
268 | .mnt = path->mnt, | |
269 | .dentry = child, | |
270 | }; | |
ac6614b7 AV |
271 | struct getdents_callback buffer = { |
272 | .ctx.actor = filldir_one, | |
273 | .name = name, | |
ac6614b7 | 274 | }; |
1da177e4 LT |
275 | |
276 | error = -ENOTDIR; | |
277 | if (!dir || !S_ISDIR(dir->i_mode)) | |
278 | goto out; | |
279 | error = -EINVAL; | |
280 | if (!dir->i_fop) | |
281 | goto out; | |
950ee956 BF |
282 | /* |
283 | * inode->i_ino is unsigned long, kstat->ino is u64, so the | |
284 | * former would be insufficient on 32-bit hosts when the | |
285 | * filesystem supports 64-bit inode numbers. So we need to | |
286 | * actually call ->getattr, not just read i_ino: | |
287 | */ | |
288 | error = vfs_getattr_nosec(&child_path, &stat); | |
289 | if (error) | |
290 | return error; | |
291 | buffer.ino = stat.ino; | |
1da177e4 LT |
292 | /* |
293 | * Open the directory ... | |
294 | */ | |
765927b2 | 295 | file = dentry_open(path, O_RDONLY, cred); |
1da177e4 LT |
296 | error = PTR_ERR(file); |
297 | if (IS_ERR(file)) | |
298 | goto out; | |
299 | ||
300 | error = -EINVAL; | |
2233f31a | 301 | if (!file->f_op->iterate) |
1da177e4 LT |
302 | goto out_close; |
303 | ||
1da177e4 LT |
304 | buffer.sequence = 0; |
305 | while (1) { | |
306 | int old_seq = buffer.sequence; | |
307 | ||
5c0ba4e0 | 308 | error = iterate_dir(file, &buffer.ctx); |
53c9c5c0 AV |
309 | if (buffer.found) { |
310 | error = 0; | |
311 | break; | |
312 | } | |
1da177e4 LT |
313 | |
314 | if (error < 0) | |
315 | break; | |
316 | ||
1da177e4 LT |
317 | error = -ENOENT; |
318 | if (old_seq == buffer.sequence) | |
319 | break; | |
320 | } | |
321 | ||
322 | out_close: | |
323 | fput(file); | |
324 | out: | |
325 | return error; | |
326 | } | |
327 | ||
1da177e4 LT |
328 | /** |
329 | * export_encode_fh - default export_operations->encode_fh function | |
b0b0382b | 330 | * @inode: the object to encode |
1da177e4 LT |
331 | * @fh: where to store the file handle fragment |
332 | * @max_len: maximum length to store there | |
b0b0382b | 333 | * @parent: parent directory inode, if wanted |
1da177e4 LT |
334 | * |
335 | * This default encode_fh function assumes that the 32 inode number | |
336 | * is suitable for locating an inode, and that the generation number | |
337 | * can be used to check that it is still valid. It places them in the | |
338 | * filehandle fragment where export_decode_fh expects to find them. | |
339 | */ | |
b0b0382b AV |
340 | static int export_encode_fh(struct inode *inode, struct fid *fid, |
341 | int *max_len, struct inode *parent) | |
1da177e4 | 342 | { |
1da177e4 | 343 | int len = *max_len; |
6e91ea2b | 344 | int type = FILEID_INO32_GEN; |
5fe0c237 | 345 | |
b0b0382b | 346 | if (parent && (len < 4)) { |
5fe0c237 | 347 | *max_len = 4; |
216b6cbd | 348 | return FILEID_INVALID; |
5fe0c237 AK |
349 | } else if (len < 2) { |
350 | *max_len = 2; | |
216b6cbd | 351 | return FILEID_INVALID; |
5fe0c237 | 352 | } |
1da177e4 LT |
353 | |
354 | len = 2; | |
6e91ea2b CH |
355 | fid->i32.ino = inode->i_ino; |
356 | fid->i32.gen = inode->i_generation; | |
b0b0382b | 357 | if (parent) { |
6e91ea2b CH |
358 | fid->i32.parent_ino = parent->i_ino; |
359 | fid->i32.parent_gen = parent->i_generation; | |
1da177e4 | 360 | len = 4; |
6e91ea2b | 361 | type = FILEID_INO32_GEN_PARENT; |
1da177e4 LT |
362 | } |
363 | *max_len = len; | |
364 | return type; | |
365 | } | |
366 | ||
711c7bf9 CG |
367 | int exportfs_encode_inode_fh(struct inode *inode, struct fid *fid, |
368 | int *max_len, struct inode *parent) | |
369 | { | |
370 | const struct export_operations *nop = inode->i_sb->s_export_op; | |
371 | ||
372 | if (nop && nop->encode_fh) | |
373 | return nop->encode_fh(inode, fid->raw, max_len, parent); | |
374 | ||
375 | return export_encode_fh(inode, fid, max_len, parent); | |
376 | } | |
377 | EXPORT_SYMBOL_GPL(exportfs_encode_inode_fh); | |
378 | ||
6e91ea2b | 379 | int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len, |
d37065cd CH |
380 | int connectable) |
381 | { | |
10f11c34 | 382 | int error; |
b0b0382b AV |
383 | struct dentry *p = NULL; |
384 | struct inode *inode = dentry->d_inode, *parent = NULL; | |
d37065cd | 385 | |
b0b0382b AV |
386 | if (connectable && !S_ISDIR(inode->i_mode)) { |
387 | p = dget_parent(dentry); | |
388 | /* | |
389 | * note that while p might've ceased to be our parent already, | |
390 | * it's still pinned by and still positive. | |
391 | */ | |
392 | parent = p->d_inode; | |
393 | } | |
711c7bf9 CG |
394 | |
395 | error = exportfs_encode_inode_fh(inode, fid, max_len, parent); | |
b0b0382b | 396 | dput(p); |
10f11c34 CH |
397 | |
398 | return error; | |
d37065cd CH |
399 | } |
400 | EXPORT_SYMBOL_GPL(exportfs_encode_fh); | |
401 | ||
6e91ea2b CH |
402 | struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid, |
403 | int fh_len, int fileid_type, | |
404 | int (*acceptable)(void *, struct dentry *), void *context) | |
d37065cd | 405 | { |
39655164 | 406 | const struct export_operations *nop = mnt->mnt_sb->s_export_op; |
2596110a | 407 | struct dentry *result, *alias; |
f3f8e175 | 408 | char nbuf[NAME_MAX+1]; |
2596110a | 409 | int err; |
d37065cd | 410 | |
2596110a CH |
411 | /* |
412 | * Try to get any dentry for the given file handle from the filesystem. | |
413 | */ | |
becfd1f3 AK |
414 | if (!nop || !nop->fh_to_dentry) |
415 | return ERR_PTR(-ESTALE); | |
2596110a | 416 | result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type); |
a4f4d6df BF |
417 | if (!result) |
418 | result = ERR_PTR(-ESTALE); | |
2596110a CH |
419 | if (IS_ERR(result)) |
420 | return result; | |
421 | ||
422 | if (S_ISDIR(result->d_inode->i_mode)) { | |
423 | /* | |
424 | * This request is for a directory. | |
425 | * | |
426 | * On the positive side there is only one dentry for each | |
427 | * directory inode. On the negative side this implies that we | |
428 | * to ensure our dentry is connected all the way up to the | |
429 | * filesystem root. | |
430 | */ | |
431 | if (result->d_flags & DCACHE_DISCONNECTED) { | |
f3f8e175 | 432 | err = reconnect_path(mnt, result, nbuf); |
2596110a CH |
433 | if (err) |
434 | goto err_result; | |
435 | } | |
436 | ||
437 | if (!acceptable(context, result)) { | |
438 | err = -EACCES; | |
439 | goto err_result; | |
440 | } | |
441 | ||
442 | return result; | |
10f11c34 | 443 | } else { |
2596110a CH |
444 | /* |
445 | * It's not a directory. Life is a little more complicated. | |
446 | */ | |
447 | struct dentry *target_dir, *nresult; | |
2596110a CH |
448 | |
449 | /* | |
450 | * See if either the dentry we just got from the filesystem | |
451 | * or any alias for it is acceptable. This is always true | |
452 | * if this filesystem is exported without the subtreecheck | |
453 | * option. If the filesystem is exported with the subtree | |
454 | * check option there's a fair chance we need to look at | |
455 | * the parent directory in the file handle and make sure | |
456 | * it's connected to the filesystem root. | |
457 | */ | |
458 | alias = find_acceptable_alias(result, acceptable, context); | |
459 | if (alias) | |
460 | return alias; | |
461 | ||
462 | /* | |
463 | * Try to extract a dentry for the parent directory from the | |
464 | * file handle. If this fails we'll have to give up. | |
465 | */ | |
466 | err = -ESTALE; | |
467 | if (!nop->fh_to_parent) | |
468 | goto err_result; | |
469 | ||
470 | target_dir = nop->fh_to_parent(mnt->mnt_sb, fid, | |
471 | fh_len, fileid_type); | |
a4f4d6df BF |
472 | if (!target_dir) |
473 | goto err_result; | |
2596110a CH |
474 | err = PTR_ERR(target_dir); |
475 | if (IS_ERR(target_dir)) | |
476 | goto err_result; | |
477 | ||
478 | /* | |
479 | * And as usual we need to make sure the parent directory is | |
480 | * connected to the filesystem root. The VFS really doesn't | |
481 | * like disconnected directories.. | |
482 | */ | |
f3f8e175 | 483 | err = reconnect_path(mnt, target_dir, nbuf); |
2596110a CH |
484 | if (err) { |
485 | dput(target_dir); | |
486 | goto err_result; | |
487 | } | |
488 | ||
489 | /* | |
490 | * Now that we've got both a well-connected parent and a | |
491 | * dentry for the inode we're after, make sure that our | |
492 | * inode is actually connected to the parent. | |
493 | */ | |
e38f9817 | 494 | err = exportfs_get_name(mnt, target_dir, nbuf, result); |
2596110a CH |
495 | if (!err) { |
496 | mutex_lock(&target_dir->d_inode->i_mutex); | |
497 | nresult = lookup_one_len(nbuf, target_dir, | |
498 | strlen(nbuf)); | |
499 | mutex_unlock(&target_dir->d_inode->i_mutex); | |
500 | if (!IS_ERR(nresult)) { | |
501 | if (nresult->d_inode) { | |
502 | dput(result); | |
503 | result = nresult; | |
504 | } else | |
505 | dput(nresult); | |
506 | } | |
507 | } | |
508 | ||
509 | /* | |
510 | * At this point we are done with the parent, but it's pinned | |
511 | * by the child dentry anyway. | |
512 | */ | |
513 | dput(target_dir); | |
514 | ||
515 | /* | |
516 | * And finally make sure the dentry is actually acceptable | |
517 | * to NFSD. | |
518 | */ | |
519 | alias = find_acceptable_alias(result, acceptable, context); | |
520 | if (!alias) { | |
521 | err = -EACCES; | |
522 | goto err_result; | |
523 | } | |
524 | ||
525 | return alias; | |
10f11c34 CH |
526 | } |
527 | ||
2596110a CH |
528 | err_result: |
529 | dput(result); | |
530 | return ERR_PTR(err); | |
d37065cd CH |
531 | } |
532 | EXPORT_SYMBOL_GPL(exportfs_decode_fh); | |
533 | ||
1da177e4 | 534 | MODULE_LICENSE("GPL"); |