xref: /openbmc/linux/fs/exportfs/expfs.c (revision 7b6d864b)
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 
48 	if (acceptable(context, result))
49 		return result;
50 
51 	inode = result->d_inode;
52 	spin_lock(&inode->i_lock);
53 	hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
54 		dget(dentry);
55 		spin_unlock(&inode->i_lock);
56 		if (toput)
57 			dput(toput);
58 		if (dentry != result && acceptable(context, dentry)) {
59 			dput(result);
60 			return dentry;
61 		}
62 		spin_lock(&inode->i_lock);
63 		toput = dentry;
64 	}
65 	spin_unlock(&inode->i_lock);
66 
67 	if (toput)
68 		dput(toput);
69 	return NULL;
70 }
71 
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);
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 
87 		dput(dentry);
88 		dentry = parent;
89 	}
90 	return dentry;
91 }
92 
93 /*
94  * Make sure target_dir is fully connected to the dentry tree.
95  *
96  * It may already be, as the flag isn't always updated when connection happens.
97  */
98 static int
99 reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf)
100 {
101 	int noprogress = 0;
102 	int err = -ESTALE;
103 
104 	/*
105 	 * It is possible that a confused file system might not let us complete
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 	 */
112 	while (target_dir->d_flags & DCACHE_DISCONNECTED && noprogress++ < 10) {
113 		struct dentry *pd = find_disconnected_root(target_dir);
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;
121 		} else if (pd == mnt->mnt_sb->s_root) {
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 {
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
142 			 */
143 			struct dentry *ppd = ERR_PTR(-EACCES);
144 			struct dentry *npd;
145 
146 			mutex_lock(&pd->d_inode->i_mutex);
147 			if (mnt->mnt_sb->s_export_op->get_parent)
148 				ppd = mnt->mnt_sb->s_export_op->get_parent(pd);
149 			mutex_unlock(&pd->d_inode->i_mutex);
150 
151 			if (IS_ERR(ppd)) {
152 				err = PTR_ERR(ppd);
153 				dprintk("%s: get_parent of %ld failed, err %d\n",
154 					__func__, pd->d_inode->i_ino, err);
155 				dput(pd);
156 				break;
157 			}
158 
159 			dprintk("%s: find name of %lu in %lu\n", __func__,
160 				pd->d_inode->i_ino, ppd->d_inode->i_ino);
161 			err = exportfs_get_name(mnt, ppd, nbuf, pd);
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 			}
172 			dprintk("%s: found name: %s\n", __func__, nbuf);
173 			mutex_lock(&ppd->d_inode->i_mutex);
174 			npd = lookup_one_len(nbuf, ppd, strlen(nbuf));
175 			mutex_unlock(&ppd->d_inode->i_mutex);
176 			if (IS_ERR(npd)) {
177 				err = PTR_ERR(npd);
178 				dprintk("%s: lookup failed: %d\n",
179 					__func__, err);
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
192 				printk("%s: npd != pd\n", __func__);
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;
208 		return err;
209 	}
210 
211 	return 0;
212 }
213 
214 struct getdents_callback {
215 	struct dir_context ctx;
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 		.ctx.actor = filldir_one,
260 		.name = name,
261 		.ino = child->d_inode->i_ino
262 	};
263 
264 	error = -ENOTDIR;
265 	if (!dir || !S_ISDIR(dir->i_mode))
266 		goto out;
267 	error = -EINVAL;
268 	if (!dir->i_fop)
269 		goto out;
270 	/*
271 	 * Open the directory ...
272 	 */
273 	file = dentry_open(path, O_RDONLY, cred);
274 	error = PTR_ERR(file);
275 	if (IS_ERR(file))
276 		goto out;
277 
278 	error = -EINVAL;
279 	if (!file->f_op->iterate)
280 		goto out_close;
281 
282 	buffer.sequence = 0;
283 	while (1) {
284 		int old_seq = buffer.sequence;
285 
286 		error = iterate_dir(file, &buffer.ctx);
287 		if (buffer.found) {
288 			error = 0;
289 			break;
290 		}
291 
292 		if (error < 0)
293 			break;
294 
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 
306 /**
307  * export_encode_fh - default export_operations->encode_fh function
308  * @inode:   the object to encode
309  * @fh:      where to store the file handle fragment
310  * @max_len: maximum length to store there
311  * @parent:  parent directory inode, if wanted
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  */
318 static int export_encode_fh(struct inode *inode, struct fid *fid,
319 		int *max_len, struct inode *parent)
320 {
321 	int len = *max_len;
322 	int type = FILEID_INO32_GEN;
323 
324 	if (parent && (len < 4)) {
325 		*max_len = 4;
326 		return FILEID_INVALID;
327 	} else if (len < 2) {
328 		*max_len = 2;
329 		return FILEID_INVALID;
330 	}
331 
332 	len = 2;
333 	fid->i32.ino = inode->i_ino;
334 	fid->i32.gen = inode->i_generation;
335 	if (parent) {
336 		fid->i32.parent_ino = parent->i_ino;
337 		fid->i32.parent_gen = parent->i_generation;
338 		len = 4;
339 		type = FILEID_INO32_GEN_PARENT;
340 	}
341 	*max_len = len;
342 	return type;
343 }
344 
345 int exportfs_encode_inode_fh(struct inode *inode, struct fid *fid,
346 			     int *max_len, struct inode *parent)
347 {
348 	const struct export_operations *nop = inode->i_sb->s_export_op;
349 
350 	if (nop && nop->encode_fh)
351 		return nop->encode_fh(inode, fid->raw, max_len, parent);
352 
353 	return export_encode_fh(inode, fid, max_len, parent);
354 }
355 EXPORT_SYMBOL_GPL(exportfs_encode_inode_fh);
356 
357 int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
358 		int connectable)
359 {
360 	int error;
361 	struct dentry *p = NULL;
362 	struct inode *inode = dentry->d_inode, *parent = NULL;
363 
364 	if (connectable && !S_ISDIR(inode->i_mode)) {
365 		p = dget_parent(dentry);
366 		/*
367 		 * note that while p might've ceased to be our parent already,
368 		 * it's still pinned by and still positive.
369 		 */
370 		parent = p->d_inode;
371 	}
372 
373 	error = exportfs_encode_inode_fh(inode, fid, max_len, parent);
374 	dput(p);
375 
376 	return error;
377 }
378 EXPORT_SYMBOL_GPL(exportfs_encode_fh);
379 
380 struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid,
381 		int fh_len, int fileid_type,
382 		int (*acceptable)(void *, struct dentry *), void *context)
383 {
384 	const struct export_operations *nop = mnt->mnt_sb->s_export_op;
385 	struct dentry *result, *alias;
386 	char nbuf[NAME_MAX+1];
387 	int err;
388 
389 	/*
390 	 * Try to get any dentry for the given file handle from the filesystem.
391 	 */
392 	if (!nop || !nop->fh_to_dentry)
393 		return ERR_PTR(-ESTALE);
394 	result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
395 	if (!result)
396 		result = ERR_PTR(-ESTALE);
397 	if (IS_ERR(result))
398 		return result;
399 
400 	if (S_ISDIR(result->d_inode->i_mode)) {
401 		/*
402 		 * This request is for a directory.
403 		 *
404 		 * On the positive side there is only one dentry for each
405 		 * directory inode.  On the negative side this implies that we
406 		 * to ensure our dentry is connected all the way up to the
407 		 * filesystem root.
408 		 */
409 		if (result->d_flags & DCACHE_DISCONNECTED) {
410 			err = reconnect_path(mnt, result, nbuf);
411 			if (err)
412 				goto err_result;
413 		}
414 
415 		if (!acceptable(context, result)) {
416 			err = -EACCES;
417 			goto err_result;
418 		}
419 
420 		return result;
421 	} else {
422 		/*
423 		 * It's not a directory.  Life is a little more complicated.
424 		 */
425 		struct dentry *target_dir, *nresult;
426 
427 		/*
428 		 * See if either the dentry we just got from the filesystem
429 		 * or any alias for it is acceptable.  This is always true
430 		 * if this filesystem is exported without the subtreecheck
431 		 * option.  If the filesystem is exported with the subtree
432 		 * check option there's a fair chance we need to look at
433 		 * the parent directory in the file handle and make sure
434 		 * it's connected to the filesystem root.
435 		 */
436 		alias = find_acceptable_alias(result, acceptable, context);
437 		if (alias)
438 			return alias;
439 
440 		/*
441 		 * Try to extract a dentry for the parent directory from the
442 		 * file handle.  If this fails we'll have to give up.
443 		 */
444 		err = -ESTALE;
445 		if (!nop->fh_to_parent)
446 			goto err_result;
447 
448 		target_dir = nop->fh_to_parent(mnt->mnt_sb, fid,
449 				fh_len, fileid_type);
450 		if (!target_dir)
451 			goto err_result;
452 		err = PTR_ERR(target_dir);
453 		if (IS_ERR(target_dir))
454 			goto err_result;
455 
456 		/*
457 		 * And as usual we need to make sure the parent directory is
458 		 * connected to the filesystem root.  The VFS really doesn't
459 		 * like disconnected directories..
460 		 */
461 		err = reconnect_path(mnt, target_dir, nbuf);
462 		if (err) {
463 			dput(target_dir);
464 			goto err_result;
465 		}
466 
467 		/*
468 		 * Now that we've got both a well-connected parent and a
469 		 * dentry for the inode we're after, make sure that our
470 		 * inode is actually connected to the parent.
471 		 */
472 		err = exportfs_get_name(mnt, target_dir, nbuf, result);
473 		if (!err) {
474 			mutex_lock(&target_dir->d_inode->i_mutex);
475 			nresult = lookup_one_len(nbuf, target_dir,
476 						 strlen(nbuf));
477 			mutex_unlock(&target_dir->d_inode->i_mutex);
478 			if (!IS_ERR(nresult)) {
479 				if (nresult->d_inode) {
480 					dput(result);
481 					result = nresult;
482 				} else
483 					dput(nresult);
484 			}
485 		}
486 
487 		/*
488 		 * At this point we are done with the parent, but it's pinned
489 		 * by the child dentry anyway.
490 		 */
491 		dput(target_dir);
492 
493 		/*
494 		 * And finally make sure the dentry is actually acceptable
495 		 * to NFSD.
496 		 */
497 		alias = find_acceptable_alias(result, acceptable, context);
498 		if (!alias) {
499 			err = -EACCES;
500 			goto err_result;
501 		}
502 
503 		return alias;
504 	}
505 
506  err_result:
507 	dput(result);
508 	return ERR_PTR(err);
509 }
510 EXPORT_SYMBOL_GPL(exportfs_decode_fh);
511 
512 MODULE_LICENSE("GPL");
513