xref: /linux/fs/exportfs/expfs.c (revision d8327c784b51b57dac2c26cfad87dce0d68dfd98)
1 
2 #include <linux/fs.h>
3 #include <linux/file.h>
4 #include <linux/module.h>
5 #include <linux/smp_lock.h>
6 #include <linux/namei.h>
7 
8 struct export_operations export_op_default;
9 
10 #define	CALL(ops,fun) ((ops->fun)?(ops->fun):export_op_default.fun)
11 
12 #define dprintk(fmt, args...) do{}while(0)
13 
14 static struct dentry *
15 find_acceptable_alias(struct dentry *result,
16 		int (*acceptable)(void *context, struct dentry *dentry),
17 		void *context)
18 {
19 	struct dentry *dentry, *toput = NULL;
20 
21 	spin_lock(&dcache_lock);
22 	list_for_each_entry(dentry, &result->d_inode->i_dentry, d_alias) {
23 		dget_locked(dentry);
24 		spin_unlock(&dcache_lock);
25 		if (toput)
26 			dput(toput);
27 		if (dentry != result && acceptable(context, dentry)) {
28 			dput(result);
29 			return dentry;
30 		}
31 		spin_lock(&dcache_lock);
32 		toput = dentry;
33 	}
34 	spin_unlock(&dcache_lock);
35 
36 	if (toput)
37 		dput(toput);
38 	return NULL;
39 }
40 
41 /**
42  * find_exported_dentry - helper routine to implement export_operations->decode_fh
43  * @sb:		The &super_block identifying the filesystem
44  * @obj:	An opaque identifier of the object to be found - passed to
45  *		get_inode
46  * @parent:	An optional opqaue identifier of the parent of the object.
47  * @acceptable:	A function used to test possible &dentries to see if they are
48  *		acceptable
49  * @context:	A parameter to @acceptable so that it knows on what basis to
50  *		judge.
51  *
52  * find_exported_dentry is the central helper routine to enable file systems
53  * to provide the decode_fh() export_operation.  It's main task is to take
54  * an &inode, find or create an appropriate &dentry structure, and possibly
55  * splice this into the dcache in the correct place.
56  *
57  * The decode_fh() operation provided by the filesystem should call
58  * find_exported_dentry() with the same parameters that it received except
59  * that instead of the file handle fragment, pointers to opaque identifiers
60  * for the object and optionally its parent are passed.  The default decode_fh
61  * routine passes one pointer to the start of the filehandle fragment, and
62  * one 8 bytes into the fragment.  It is expected that most filesystems will
63  * take this approach, though the offset to the parent identifier may well be
64  * different.
65  *
66  * find_exported_dentry() will call get_dentry to get an dentry pointer from
67  * the file system.  If any &dentry in the d_alias list is acceptable, it will
68  * be returned.  Otherwise find_exported_dentry() will attempt to splice a new
69  * &dentry into the dcache using get_name() and get_parent() to find the
70  * appropriate place.
71  */
72 
73 struct dentry *
74 find_exported_dentry(struct super_block *sb, void *obj, void *parent,
75 		     int (*acceptable)(void *context, struct dentry *de),
76 		     void *context)
77 {
78 	struct dentry *result = NULL;
79 	struct dentry *target_dir;
80 	int err;
81 	struct export_operations *nops = sb->s_export_op;
82 	struct dentry *alias;
83 	int noprogress;
84 	char nbuf[NAME_MAX+1];
85 
86 	/*
87 	 * Attempt to find the inode.
88 	 */
89 	result = CALL(sb->s_export_op,get_dentry)(sb,obj);
90 	err = -ESTALE;
91 	if (result == NULL)
92 		goto err_out;
93 	if (IS_ERR(result)) {
94 		err = PTR_ERR(result);
95 		goto err_out;
96 	}
97 	if (S_ISDIR(result->d_inode->i_mode) &&
98 	    (result->d_flags & DCACHE_DISCONNECTED)) {
99 		/* it is an unconnected directory, we must connect it */
100 		;
101 	} else {
102 		if (acceptable(context, result))
103 			return result;
104 		if (S_ISDIR(result->d_inode->i_mode)) {
105 			/* there is no other dentry, so fail */
106 			goto err_result;
107 		}
108 
109 		alias = find_acceptable_alias(result, acceptable, context);
110 		if (alias)
111 			return alias;
112 	}
113 
114 	/* It's a directory, or we are required to confirm the file's
115 	 * location in the tree based on the parent information
116  	 */
117 	dprintk("find_exported_dentry: need to look harder for %s/%d\n",sb->s_id,*(int*)obj);
118 	if (S_ISDIR(result->d_inode->i_mode))
119 		target_dir = dget(result);
120 	else {
121 		if (parent == NULL)
122 			goto err_result;
123 
124 		target_dir = CALL(sb->s_export_op,get_dentry)(sb,parent);
125 		if (IS_ERR(target_dir))
126 			err = PTR_ERR(target_dir);
127 		if (target_dir == NULL || IS_ERR(target_dir))
128 			goto err_result;
129 	}
130 	/*
131 	 * Now we need to make sure that target_dir is properly connected.
132 	 * It may already be, as the flag isn't always updated when connection
133 	 * happens.
134 	 * So, we walk up parent links until we find a connected directory,
135 	 * or we run out of directories.  Then we find the parent, find
136 	 * the name of the child in that parent, and do a lookup.
137 	 * This should connect the child into the parent
138 	 * We then repeat.
139 	 */
140 
141 	/* it is possible that a confused file system might not let us complete
142 	 * the path to the root.  For example, if get_parent returns a directory
143 	 * in which we cannot find a name for the child.  While this implies a
144 	 * very sick filesystem we don't want it to cause knfsd to spin.  Hence
145 	 * the noprogress counter.  If we go through the loop 10 times (2 is
146 	 * probably enough) without getting anywhere, we just give up
147 	 */
148 	noprogress= 0;
149 	while (target_dir->d_flags & DCACHE_DISCONNECTED && noprogress++ < 10) {
150 		struct dentry *pd = target_dir;
151 
152 		dget(pd);
153 		spin_lock(&pd->d_lock);
154 		while (!IS_ROOT(pd) &&
155 				(pd->d_parent->d_flags&DCACHE_DISCONNECTED)) {
156 			struct dentry *parent = pd->d_parent;
157 
158 			dget(parent);
159 			spin_unlock(&pd->d_lock);
160 			dput(pd);
161 			pd = parent;
162 			spin_lock(&pd->d_lock);
163 		}
164 		spin_unlock(&pd->d_lock);
165 
166 		if (!IS_ROOT(pd)) {
167 			/* must have found a connected parent - great */
168 			spin_lock(&pd->d_lock);
169 			pd->d_flags &= ~DCACHE_DISCONNECTED;
170 			spin_unlock(&pd->d_lock);
171 			noprogress = 0;
172 		} else if (pd == sb->s_root) {
173 			printk(KERN_ERR "export: Eeek filesystem root is not connected, impossible\n");
174 			spin_lock(&pd->d_lock);
175 			pd->d_flags &= ~DCACHE_DISCONNECTED;
176 			spin_unlock(&pd->d_lock);
177 			noprogress = 0;
178 		} else {
179 			/* we have hit the top of a disconnected path.  Try
180 			 * to find parent and connect
181 			 * note: racing with some other process renaming a
182 			 * directory isn't much of a problem here.  If someone
183 			 * renames the directory, it will end up properly
184 			 * connected, which is what we want
185 			 */
186 			struct dentry *ppd;
187 			struct dentry *npd;
188 
189 			mutex_lock(&pd->d_inode->i_mutex);
190 			ppd = CALL(nops,get_parent)(pd);
191 			mutex_unlock(&pd->d_inode->i_mutex);
192 
193 			if (IS_ERR(ppd)) {
194 				err = PTR_ERR(ppd);
195 				dprintk("find_exported_dentry: get_parent of %ld failed, err %d\n",
196 					pd->d_inode->i_ino, err);
197 				dput(pd);
198 				break;
199 			}
200 			dprintk("find_exported_dentry: find name of %lu in %lu\n", pd->d_inode->i_ino, ppd->d_inode->i_ino);
201 			err = CALL(nops,get_name)(ppd, nbuf, pd);
202 			if (err) {
203 				dput(ppd);
204 				dput(pd);
205 				if (err == -ENOENT)
206 					/* some race between get_parent and
207 					 * get_name?  just try again
208 					 */
209 					continue;
210 				break;
211 			}
212 			dprintk("find_exported_dentry: found name: %s\n", nbuf);
213 			mutex_lock(&ppd->d_inode->i_mutex);
214 			npd = lookup_one_len(nbuf, ppd, strlen(nbuf));
215 			mutex_unlock(&ppd->d_inode->i_mutex);
216 			if (IS_ERR(npd)) {
217 				err = PTR_ERR(npd);
218 				dprintk("find_exported_dentry: lookup failed: %d\n", err);
219 				dput(ppd);
220 				dput(pd);
221 				break;
222 			}
223 			/* we didn't really want npd, we really wanted
224 			 * a side-effect of the lookup.
225 			 * hopefully, npd == pd, though it isn't really
226 			 * a problem if it isn't
227 			 */
228 			if (npd == pd)
229 				noprogress = 0;
230 			else
231 				printk("find_exported_dentry: npd != pd\n");
232 			dput(npd);
233 			dput(ppd);
234 			if (IS_ROOT(pd)) {
235 				/* something went wrong, we have to give up */
236 				dput(pd);
237 				break;
238 			}
239 		}
240 		dput(pd);
241 	}
242 
243 	if (target_dir->d_flags & DCACHE_DISCONNECTED) {
244 		/* something went wrong - oh-well */
245 		if (!err)
246 			err = -ESTALE;
247 		goto err_target;
248 	}
249 	/* if we weren't after a directory, have one more step to go */
250 	if (result != target_dir) {
251 		struct dentry *nresult;
252 		err = CALL(nops,get_name)(target_dir, nbuf, result);
253 		if (!err) {
254 			mutex_lock(&target_dir->d_inode->i_mutex);
255 			nresult = lookup_one_len(nbuf, target_dir, strlen(nbuf));
256 			mutex_unlock(&target_dir->d_inode->i_mutex);
257 			if (!IS_ERR(nresult)) {
258 				if (nresult->d_inode) {
259 					dput(result);
260 					result = nresult;
261 				} else
262 					dput(nresult);
263 			}
264 		}
265 	}
266 	dput(target_dir);
267 	/* now result is properly connected, it is our best bet */
268 	if (acceptable(context, result))
269 		return result;
270 
271 	alias = find_acceptable_alias(result, acceptable, context);
272 	if (alias)
273 		return alias;
274 
275 	/* drat - I just cannot find anything acceptable */
276 	dput(result);
277 	/* It might be justifiable to return ESTALE here,
278 	 * but the filehandle at-least looks reasonable good
279 	 * and it just be a permission problem, so returning
280 	 * -EACCESS is safer
281 	 */
282 	return ERR_PTR(-EACCES);
283 
284  err_target:
285 	dput(target_dir);
286  err_result:
287 	dput(result);
288  err_out:
289 	return ERR_PTR(err);
290 }
291 
292 
293 
294 static struct dentry *get_parent(struct dentry *child)
295 {
296 	/* get_parent cannot be supported generically, the locking
297 	 * is too icky.
298 	 * instead, we just return EACCES.  If server reboots or inodes
299 	 * get flushed, you lose
300 	 */
301 	return ERR_PTR(-EACCES);
302 }
303 
304 
305 struct getdents_callback {
306 	char *name;		/* name that was found. It already points to a
307 				   buffer NAME_MAX+1 is size */
308 	unsigned long ino;	/* the inum we are looking for */
309 	int found;		/* inode matched? */
310 	int sequence;		/* sequence counter */
311 };
312 
313 /*
314  * A rather strange filldir function to capture
315  * the name matching the specified inode number.
316  */
317 static int filldir_one(void * __buf, const char * name, int len,
318 			loff_t pos, ino_t ino, unsigned int d_type)
319 {
320 	struct getdents_callback *buf = __buf;
321 	int result = 0;
322 
323 	buf->sequence++;
324 	if (buf->ino == ino) {
325 		memcpy(buf->name, name, len);
326 		buf->name[len] = '\0';
327 		buf->found = 1;
328 		result = -1;
329 	}
330 	return result;
331 }
332 
333 /**
334  * get_name - default export_operations->get_name function
335  * @dentry: the directory in which to find a name
336  * @name:   a pointer to a %NAME_MAX+1 char buffer to store the name
337  * @child:  the dentry for the child directory.
338  *
339  * calls readdir on the parent until it finds an entry with
340  * the same inode number as the child, and returns that.
341  */
342 static int get_name(struct dentry *dentry, char *name,
343 			struct dentry *child)
344 {
345 	struct inode *dir = dentry->d_inode;
346 	int error;
347 	struct file *file;
348 	struct getdents_callback buffer;
349 
350 	error = -ENOTDIR;
351 	if (!dir || !S_ISDIR(dir->i_mode))
352 		goto out;
353 	error = -EINVAL;
354 	if (!dir->i_fop)
355 		goto out;
356 	/*
357 	 * Open the directory ...
358 	 */
359 	file = dentry_open(dget(dentry), NULL, O_RDONLY);
360 	error = PTR_ERR(file);
361 	if (IS_ERR(file))
362 		goto out;
363 
364 	error = -EINVAL;
365 	if (!file->f_op->readdir)
366 		goto out_close;
367 
368 	buffer.name = name;
369 	buffer.ino = child->d_inode->i_ino;
370 	buffer.found = 0;
371 	buffer.sequence = 0;
372 	while (1) {
373 		int old_seq = buffer.sequence;
374 
375 		error = vfs_readdir(file, filldir_one, &buffer);
376 
377 		if (error < 0)
378 			break;
379 
380 		error = 0;
381 		if (buffer.found)
382 			break;
383 		error = -ENOENT;
384 		if (old_seq == buffer.sequence)
385 			break;
386 	}
387 
388 out_close:
389 	fput(file);
390 out:
391 	return error;
392 }
393 
394 
395 static struct dentry *export_iget(struct super_block *sb, unsigned long ino, __u32 generation)
396 {
397 
398 	/* iget isn't really right if the inode is currently unallocated!!
399 	 * This should really all be done inside each filesystem
400 	 *
401 	 * ext2fs' read_inode has been strengthed to return a bad_inode if
402 	 * the inode had been deleted.
403 	 *
404 	 * Currently we don't know the generation for parent directory, so
405 	 * a generation of 0 means "accept any"
406 	 */
407 	struct inode *inode;
408 	struct dentry *result;
409 	if (ino == 0)
410 		return ERR_PTR(-ESTALE);
411 	inode = iget(sb, ino);
412 	if (inode == NULL)
413 		return ERR_PTR(-ENOMEM);
414 	if (is_bad_inode(inode)
415 	    || (generation && inode->i_generation != generation)
416 		) {
417 		/* we didn't find the right inode.. */
418 		dprintk("fh_verify: Inode %lu, Bad count: %d %d or version  %u %u\n",
419 			inode->i_ino,
420 			inode->i_nlink, atomic_read(&inode->i_count),
421 			inode->i_generation,
422 			generation);
423 
424 		iput(inode);
425 		return ERR_PTR(-ESTALE);
426 	}
427 	/* now to find a dentry.
428 	 * If possible, get a well-connected one
429 	 */
430 	result = d_alloc_anon(inode);
431 	if (!result) {
432 		iput(inode);
433 		return ERR_PTR(-ENOMEM);
434 	}
435 	return result;
436 }
437 
438 
439 static struct dentry *get_object(struct super_block *sb, void *vobjp)
440 {
441 	__u32 *objp = vobjp;
442 	unsigned long ino = objp[0];
443 	__u32 generation = objp[1];
444 
445 	return export_iget(sb, ino, generation);
446 }
447 
448 
449 /**
450  * export_encode_fh - default export_operations->encode_fh function
451  * @dentry:  the dentry to encode
452  * @fh:      where to store the file handle fragment
453  * @max_len: maximum length to store there
454  * @connectable: whether to store parent information
455  *
456  * This default encode_fh function assumes that the 32 inode number
457  * is suitable for locating an inode, and that the generation number
458  * can be used to check that it is still valid.  It places them in the
459  * filehandle fragment where export_decode_fh expects to find them.
460  */
461 static int export_encode_fh(struct dentry *dentry, __u32 *fh, int *max_len,
462 		   int connectable)
463 {
464 	struct inode * inode = dentry->d_inode;
465 	int len = *max_len;
466 	int type = 1;
467 
468 	if (len < 2 || (connectable && len < 4))
469 		return 255;
470 
471 	len = 2;
472 	fh[0] = inode->i_ino;
473 	fh[1] = inode->i_generation;
474 	if (connectable && !S_ISDIR(inode->i_mode)) {
475 		struct inode *parent;
476 
477 		spin_lock(&dentry->d_lock);
478 		parent = dentry->d_parent->d_inode;
479 		fh[2] = parent->i_ino;
480 		fh[3] = parent->i_generation;
481 		spin_unlock(&dentry->d_lock);
482 		len = 4;
483 		type = 2;
484 	}
485 	*max_len = len;
486 	return type;
487 }
488 
489 
490 /**
491  * export_decode_fh - default export_operations->decode_fh function
492  * @sb:  The superblock
493  * @fh:  pointer to the file handle fragment
494  * @fh_len: length of file handle fragment
495  * @acceptable: function for testing acceptability of dentrys
496  * @context:   context for @acceptable
497  *
498  * This is the default decode_fh() function.
499  * a fileid_type of 1 indicates that the filehandlefragment
500  * just contains an object identifier understood by  get_dentry.
501  * a fileid_type of 2 says that there is also a directory
502  * identifier 8 bytes in to the filehandlefragement.
503  */
504 static struct dentry *export_decode_fh(struct super_block *sb, __u32 *fh, int fh_len,
505 			      int fileid_type,
506 			 int (*acceptable)(void *context, struct dentry *de),
507 			 void *context)
508 {
509 	__u32 parent[2];
510 	parent[0] = parent[1] = 0;
511 	if (fh_len < 2 || fileid_type > 2)
512 		return NULL;
513 	if (fileid_type == 2) {
514 		if (fh_len > 2) parent[0] = fh[2];
515 		if (fh_len > 3) parent[1] = fh[3];
516 	}
517 	return find_exported_dentry(sb, fh, parent,
518 				   acceptable, context);
519 }
520 
521 struct export_operations export_op_default = {
522 	.decode_fh	= export_decode_fh,
523 	.encode_fh	= export_encode_fh,
524 
525 	.get_name	= get_name,
526 	.get_parent	= get_parent,
527 	.get_dentry	= get_object,
528 };
529 
530 EXPORT_SYMBOL(export_op_default);
531 EXPORT_SYMBOL(find_exported_dentry);
532 
533 MODULE_LICENSE("GPL");
534