xref: /openbmc/linux/fs/exportfs/expfs.c (revision 95e9fd10)
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/nfs/Exporting.
10  */
11 #include <linux/exportfs.h>
12 #include <linux/fs.h>
13 #include <linux/file.h>
14 #include <linux/module.h>
15 #include <linux/mount.h>
16 #include <linux/namei.h>
17 #include <linux/sched.h>
18 
19 #define dprintk(fmt, args...) do{}while(0)
20 
21 
22 static int get_name(const struct path *path, char *name, struct dentry *child);
23 
24 
25 static int exportfs_get_name(struct vfsmount *mnt, struct dentry *dir,
26 		char *name, struct dentry *child)
27 {
28 	const struct export_operations *nop = dir->d_sb->s_export_op;
29 	struct path path = {.mnt = mnt, .dentry = dir};
30 
31 	if (nop->get_name)
32 		return nop->get_name(dir, name, child);
33 	else
34 		return get_name(&path, name, child);
35 }
36 
37 /*
38  * Check if the dentry or any of it's aliases is acceptable.
39  */
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;
46 	struct inode *inode;
47 	struct hlist_node *p;
48 
49 	if (acceptable(context, result))
50 		return result;
51 
52 	inode = result->d_inode;
53 	spin_lock(&inode->i_lock);
54 	hlist_for_each_entry(dentry, p, &inode->i_dentry, d_alias) {
55 		dget(dentry);
56 		spin_unlock(&inode->i_lock);
57 		if (toput)
58 			dput(toput);
59 		if (dentry != result && acceptable(context, dentry)) {
60 			dput(result);
61 			return dentry;
62 		}
63 		spin_lock(&inode->i_lock);
64 		toput = dentry;
65 	}
66 	spin_unlock(&inode->i_lock);
67 
68 	if (toput)
69 		dput(toput);
70 	return NULL;
71 }
72 
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);
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 
88 		dput(dentry);
89 		dentry = parent;
90 	}
91 	return dentry;
92 }
93 
94 /*
95  * Make sure target_dir is fully connected to the dentry tree.
96  *
97  * It may already be, as the flag isn't always updated when connection happens.
98  */
99 static int
100 reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf)
101 {
102 	int noprogress = 0;
103 	int err = -ESTALE;
104 
105 	/*
106 	 * It is possible that a confused file system might not let us complete
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 	 */
113 	while (target_dir->d_flags & DCACHE_DISCONNECTED && noprogress++ < 10) {
114 		struct dentry *pd = find_disconnected_root(target_dir);
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;
122 		} else if (pd == mnt->mnt_sb->s_root) {
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 {
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
143 			 */
144 			struct dentry *ppd = ERR_PTR(-EACCES);
145 			struct dentry *npd;
146 
147 			mutex_lock(&pd->d_inode->i_mutex);
148 			if (mnt->mnt_sb->s_export_op->get_parent)
149 				ppd = mnt->mnt_sb->s_export_op->get_parent(pd);
150 			mutex_unlock(&pd->d_inode->i_mutex);
151 
152 			if (IS_ERR(ppd)) {
153 				err = PTR_ERR(ppd);
154 				dprintk("%s: get_parent of %ld failed, err %d\n",
155 					__func__, pd->d_inode->i_ino, err);
156 				dput(pd);
157 				break;
158 			}
159 
160 			dprintk("%s: find name of %lu in %lu\n", __func__,
161 				pd->d_inode->i_ino, ppd->d_inode->i_ino);
162 			err = exportfs_get_name(mnt, ppd, nbuf, pd);
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 			}
173 			dprintk("%s: found name: %s\n", __func__, nbuf);
174 			mutex_lock(&ppd->d_inode->i_mutex);
175 			npd = lookup_one_len(nbuf, ppd, strlen(nbuf));
176 			mutex_unlock(&ppd->d_inode->i_mutex);
177 			if (IS_ERR(npd)) {
178 				err = PTR_ERR(npd);
179 				dprintk("%s: lookup failed: %d\n",
180 					__func__, err);
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
193 				printk("%s: npd != pd\n", __func__);
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;
209 		return err;
210 	}
211 
212 	return 0;
213 }
214 
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,
228 			loff_t pos, u64 ino, unsigned int d_type)
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  */
252 static int get_name(const struct path *path, char *name, struct dentry *child)
253 {
254 	const struct cred *cred = current_cred();
255 	struct inode *dir = path->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 	 */
269 	file = dentry_open(path, O_RDONLY, cred);
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);
286 		if (buffer.found) {
287 			error = 0;
288 			break;
289 		}
290 
291 		if (error < 0)
292 			break;
293 
294 		error = -ENOENT;
295 		if (old_seq == buffer.sequence)
296 			break;
297 	}
298 
299 out_close:
300 	fput(file);
301 out:
302 	return error;
303 }
304 
305 /**
306  * export_encode_fh - default export_operations->encode_fh function
307  * @inode:   the object to encode
308  * @fh:      where to store the file handle fragment
309  * @max_len: maximum length to store there
310  * @parent:  parent directory inode, if wanted
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  */
317 static int export_encode_fh(struct inode *inode, struct fid *fid,
318 		int *max_len, struct inode *parent)
319 {
320 	int len = *max_len;
321 	int type = FILEID_INO32_GEN;
322 
323 	if (parent && (len < 4)) {
324 		*max_len = 4;
325 		return 255;
326 	} else if (len < 2) {
327 		*max_len = 2;
328 		return 255;
329 	}
330 
331 	len = 2;
332 	fid->i32.ino = inode->i_ino;
333 	fid->i32.gen = inode->i_generation;
334 	if (parent) {
335 		fid->i32.parent_ino = parent->i_ino;
336 		fid->i32.parent_gen = parent->i_generation;
337 		len = 4;
338 		type = FILEID_INO32_GEN_PARENT;
339 	}
340 	*max_len = len;
341 	return type;
342 }
343 
344 int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
345 		int connectable)
346 {
347 	const struct export_operations *nop = dentry->d_sb->s_export_op;
348 	int error;
349 	struct dentry *p = NULL;
350 	struct inode *inode = dentry->d_inode, *parent = NULL;
351 
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 	}
360 	if (nop->encode_fh)
361 		error = nop->encode_fh(inode, fid->raw, max_len, parent);
362 	else
363 		error = export_encode_fh(inode, fid, max_len, parent);
364 	dput(p);
365 
366 	return error;
367 }
368 EXPORT_SYMBOL_GPL(exportfs_encode_fh);
369 
370 struct 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)
373 {
374 	const struct export_operations *nop = mnt->mnt_sb->s_export_op;
375 	struct dentry *result, *alias;
376 	char nbuf[NAME_MAX+1];
377 	int err;
378 
379 	/*
380 	 * Try to get any dentry for the given file handle from the filesystem.
381 	 */
382 	if (!nop || !nop->fh_to_dentry)
383 		return ERR_PTR(-ESTALE);
384 	result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
385 	if (!result)
386 		result = ERR_PTR(-ESTALE);
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) {
400 			err = reconnect_path(mnt, result, nbuf);
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;
411 	} else {
412 		/*
413 		 * It's not a directory.  Life is a little more complicated.
414 		 */
415 		struct dentry *target_dir, *nresult;
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);
440 		if (!target_dir)
441 			goto err_result;
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 		 */
451 		err = reconnect_path(mnt, target_dir, nbuf);
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 		 */
462 		err = exportfs_get_name(mnt, target_dir, nbuf, result);
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;
494 	}
495 
496  err_result:
497 	dput(result);
498 	return ERR_PTR(err);
499 }
500 EXPORT_SYMBOL_GPL(exportfs_decode_fh);
501 
502 MODULE_LICENSE("GPL");
503