xref: /openbmc/linux/fs/overlayfs/readdir.c (revision 9cfc5c90)
1 /*
2  *
3  * Copyright (C) 2011 Novell Inc.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 as published by
7  * the Free Software Foundation.
8  */
9 
10 #include <linux/fs.h>
11 #include <linux/slab.h>
12 #include <linux/namei.h>
13 #include <linux/file.h>
14 #include <linux/xattr.h>
15 #include <linux/rbtree.h>
16 #include <linux/security.h>
17 #include <linux/cred.h>
18 #include "overlayfs.h"
19 
20 struct ovl_cache_entry {
21 	unsigned int len;
22 	unsigned int type;
23 	u64 ino;
24 	struct list_head l_node;
25 	struct rb_node node;
26 	struct ovl_cache_entry *next_maybe_whiteout;
27 	bool is_whiteout;
28 	char name[];
29 };
30 
31 struct ovl_dir_cache {
32 	long refcount;
33 	u64 version;
34 	struct list_head entries;
35 };
36 
37 struct ovl_readdir_data {
38 	struct dir_context ctx;
39 	bool is_merge;
40 	struct rb_root root;
41 	struct list_head *list;
42 	struct list_head middle;
43 	struct ovl_cache_entry *first_maybe_whiteout;
44 	int count;
45 	int err;
46 };
47 
48 struct ovl_dir_file {
49 	bool is_real;
50 	bool is_upper;
51 	struct ovl_dir_cache *cache;
52 	struct list_head *cursor;
53 	struct file *realfile;
54 	struct file *upperfile;
55 };
56 
57 static struct ovl_cache_entry *ovl_cache_entry_from_node(struct rb_node *n)
58 {
59 	return container_of(n, struct ovl_cache_entry, node);
60 }
61 
62 static struct ovl_cache_entry *ovl_cache_entry_find(struct rb_root *root,
63 						    const char *name, int len)
64 {
65 	struct rb_node *node = root->rb_node;
66 	int cmp;
67 
68 	while (node) {
69 		struct ovl_cache_entry *p = ovl_cache_entry_from_node(node);
70 
71 		cmp = strncmp(name, p->name, len);
72 		if (cmp > 0)
73 			node = p->node.rb_right;
74 		else if (cmp < 0 || len < p->len)
75 			node = p->node.rb_left;
76 		else
77 			return p;
78 	}
79 
80 	return NULL;
81 }
82 
83 static struct ovl_cache_entry *ovl_cache_entry_new(struct ovl_readdir_data *rdd,
84 						   const char *name, int len,
85 						   u64 ino, unsigned int d_type)
86 {
87 	struct ovl_cache_entry *p;
88 	size_t size = offsetof(struct ovl_cache_entry, name[len + 1]);
89 
90 	p = kmalloc(size, GFP_KERNEL);
91 	if (!p)
92 		return NULL;
93 
94 	memcpy(p->name, name, len);
95 	p->name[len] = '\0';
96 	p->len = len;
97 	p->type = d_type;
98 	p->ino = ino;
99 	p->is_whiteout = false;
100 
101 	if (d_type == DT_CHR) {
102 		p->next_maybe_whiteout = rdd->first_maybe_whiteout;
103 		rdd->first_maybe_whiteout = p;
104 	}
105 	return p;
106 }
107 
108 static int ovl_cache_entry_add_rb(struct ovl_readdir_data *rdd,
109 				  const char *name, int len, u64 ino,
110 				  unsigned int d_type)
111 {
112 	struct rb_node **newp = &rdd->root.rb_node;
113 	struct rb_node *parent = NULL;
114 	struct ovl_cache_entry *p;
115 
116 	while (*newp) {
117 		int cmp;
118 		struct ovl_cache_entry *tmp;
119 
120 		parent = *newp;
121 		tmp = ovl_cache_entry_from_node(*newp);
122 		cmp = strncmp(name, tmp->name, len);
123 		if (cmp > 0)
124 			newp = &tmp->node.rb_right;
125 		else if (cmp < 0 || len < tmp->len)
126 			newp = &tmp->node.rb_left;
127 		else
128 			return 0;
129 	}
130 
131 	p = ovl_cache_entry_new(rdd, name, len, ino, d_type);
132 	if (p == NULL)
133 		return -ENOMEM;
134 
135 	list_add_tail(&p->l_node, rdd->list);
136 	rb_link_node(&p->node, parent, newp);
137 	rb_insert_color(&p->node, &rdd->root);
138 
139 	return 0;
140 }
141 
142 static int ovl_fill_lower(struct ovl_readdir_data *rdd,
143 			  const char *name, int namelen,
144 			  loff_t offset, u64 ino, unsigned int d_type)
145 {
146 	struct ovl_cache_entry *p;
147 
148 	p = ovl_cache_entry_find(&rdd->root, name, namelen);
149 	if (p) {
150 		list_move_tail(&p->l_node, &rdd->middle);
151 	} else {
152 		p = ovl_cache_entry_new(rdd, name, namelen, ino, d_type);
153 		if (p == NULL)
154 			rdd->err = -ENOMEM;
155 		else
156 			list_add_tail(&p->l_node, &rdd->middle);
157 	}
158 
159 	return rdd->err;
160 }
161 
162 void ovl_cache_free(struct list_head *list)
163 {
164 	struct ovl_cache_entry *p;
165 	struct ovl_cache_entry *n;
166 
167 	list_for_each_entry_safe(p, n, list, l_node)
168 		kfree(p);
169 
170 	INIT_LIST_HEAD(list);
171 }
172 
173 static void ovl_cache_put(struct ovl_dir_file *od, struct dentry *dentry)
174 {
175 	struct ovl_dir_cache *cache = od->cache;
176 
177 	WARN_ON(cache->refcount <= 0);
178 	cache->refcount--;
179 	if (!cache->refcount) {
180 		if (ovl_dir_cache(dentry) == cache)
181 			ovl_set_dir_cache(dentry, NULL);
182 
183 		ovl_cache_free(&cache->entries);
184 		kfree(cache);
185 	}
186 }
187 
188 static int ovl_fill_merge(struct dir_context *ctx, const char *name,
189 			  int namelen, loff_t offset, u64 ino,
190 			  unsigned int d_type)
191 {
192 	struct ovl_readdir_data *rdd =
193 		container_of(ctx, struct ovl_readdir_data, ctx);
194 
195 	rdd->count++;
196 	if (!rdd->is_merge)
197 		return ovl_cache_entry_add_rb(rdd, name, namelen, ino, d_type);
198 	else
199 		return ovl_fill_lower(rdd, name, namelen, offset, ino, d_type);
200 }
201 
202 static int ovl_check_whiteouts(struct dentry *dir, struct ovl_readdir_data *rdd)
203 {
204 	int err;
205 	struct ovl_cache_entry *p;
206 	struct dentry *dentry;
207 	const struct cred *old_cred;
208 	struct cred *override_cred;
209 
210 	override_cred = prepare_creds();
211 	if (!override_cred)
212 		return -ENOMEM;
213 
214 	/*
215 	 * CAP_DAC_OVERRIDE for lookup
216 	 */
217 	cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
218 	old_cred = override_creds(override_cred);
219 
220 	err = mutex_lock_killable(&dir->d_inode->i_mutex);
221 	if (!err) {
222 		while (rdd->first_maybe_whiteout) {
223 			p = rdd->first_maybe_whiteout;
224 			rdd->first_maybe_whiteout = p->next_maybe_whiteout;
225 			dentry = lookup_one_len(p->name, dir, p->len);
226 			if (!IS_ERR(dentry)) {
227 				p->is_whiteout = ovl_is_whiteout(dentry);
228 				dput(dentry);
229 			}
230 		}
231 		mutex_unlock(&dir->d_inode->i_mutex);
232 	}
233 	revert_creds(old_cred);
234 	put_cred(override_cred);
235 
236 	return err;
237 }
238 
239 static inline int ovl_dir_read(struct path *realpath,
240 			       struct ovl_readdir_data *rdd)
241 {
242 	struct file *realfile;
243 	int err;
244 
245 	realfile = ovl_path_open(realpath, O_RDONLY | O_DIRECTORY);
246 	if (IS_ERR(realfile))
247 		return PTR_ERR(realfile);
248 
249 	rdd->first_maybe_whiteout = NULL;
250 	rdd->ctx.pos = 0;
251 	do {
252 		rdd->count = 0;
253 		rdd->err = 0;
254 		err = iterate_dir(realfile, &rdd->ctx);
255 		if (err >= 0)
256 			err = rdd->err;
257 	} while (!err && rdd->count);
258 
259 	if (!err && rdd->first_maybe_whiteout)
260 		err = ovl_check_whiteouts(realpath->dentry, rdd);
261 
262 	fput(realfile);
263 
264 	return err;
265 }
266 
267 static void ovl_dir_reset(struct file *file)
268 {
269 	struct ovl_dir_file *od = file->private_data;
270 	struct ovl_dir_cache *cache = od->cache;
271 	struct dentry *dentry = file->f_path.dentry;
272 	enum ovl_path_type type = ovl_path_type(dentry);
273 
274 	if (cache && ovl_dentry_version_get(dentry) != cache->version) {
275 		ovl_cache_put(od, dentry);
276 		od->cache = NULL;
277 		od->cursor = NULL;
278 	}
279 	WARN_ON(!od->is_real && !OVL_TYPE_MERGE(type));
280 	if (od->is_real && OVL_TYPE_MERGE(type))
281 		od->is_real = false;
282 }
283 
284 static int ovl_dir_read_merged(struct dentry *dentry, struct list_head *list)
285 {
286 	int err;
287 	struct path realpath;
288 	struct ovl_readdir_data rdd = {
289 		.ctx.actor = ovl_fill_merge,
290 		.list = list,
291 		.root = RB_ROOT,
292 		.is_merge = false,
293 	};
294 	int idx, next;
295 
296 	for (idx = 0; idx != -1; idx = next) {
297 		next = ovl_path_next(idx, dentry, &realpath);
298 
299 		if (next != -1) {
300 			err = ovl_dir_read(&realpath, &rdd);
301 			if (err)
302 				break;
303 		} else {
304 			/*
305 			 * Insert lowest layer entries before upper ones, this
306 			 * allows offsets to be reasonably constant
307 			 */
308 			list_add(&rdd.middle, rdd.list);
309 			rdd.is_merge = true;
310 			err = ovl_dir_read(&realpath, &rdd);
311 			list_del(&rdd.middle);
312 		}
313 	}
314 	return err;
315 }
316 
317 static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
318 {
319 	struct list_head *p;
320 	loff_t off = 0;
321 
322 	list_for_each(p, &od->cache->entries) {
323 		if (off >= pos)
324 			break;
325 		off++;
326 	}
327 	/* Cursor is safe since the cache is stable */
328 	od->cursor = p;
329 }
330 
331 static struct ovl_dir_cache *ovl_cache_get(struct dentry *dentry)
332 {
333 	int res;
334 	struct ovl_dir_cache *cache;
335 
336 	cache = ovl_dir_cache(dentry);
337 	if (cache && ovl_dentry_version_get(dentry) == cache->version) {
338 		cache->refcount++;
339 		return cache;
340 	}
341 	ovl_set_dir_cache(dentry, NULL);
342 
343 	cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
344 	if (!cache)
345 		return ERR_PTR(-ENOMEM);
346 
347 	cache->refcount = 1;
348 	INIT_LIST_HEAD(&cache->entries);
349 
350 	res = ovl_dir_read_merged(dentry, &cache->entries);
351 	if (res) {
352 		ovl_cache_free(&cache->entries);
353 		kfree(cache);
354 		return ERR_PTR(res);
355 	}
356 
357 	cache->version = ovl_dentry_version_get(dentry);
358 	ovl_set_dir_cache(dentry, cache);
359 
360 	return cache;
361 }
362 
363 static int ovl_iterate(struct file *file, struct dir_context *ctx)
364 {
365 	struct ovl_dir_file *od = file->private_data;
366 	struct dentry *dentry = file->f_path.dentry;
367 	struct ovl_cache_entry *p;
368 
369 	if (!ctx->pos)
370 		ovl_dir_reset(file);
371 
372 	if (od->is_real)
373 		return iterate_dir(od->realfile, ctx);
374 
375 	if (!od->cache) {
376 		struct ovl_dir_cache *cache;
377 
378 		cache = ovl_cache_get(dentry);
379 		if (IS_ERR(cache))
380 			return PTR_ERR(cache);
381 
382 		od->cache = cache;
383 		ovl_seek_cursor(od, ctx->pos);
384 	}
385 
386 	while (od->cursor != &od->cache->entries) {
387 		p = list_entry(od->cursor, struct ovl_cache_entry, l_node);
388 		if (!p->is_whiteout)
389 			if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
390 				break;
391 		od->cursor = p->l_node.next;
392 		ctx->pos++;
393 	}
394 	return 0;
395 }
396 
397 static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
398 {
399 	loff_t res;
400 	struct ovl_dir_file *od = file->private_data;
401 
402 	mutex_lock(&file_inode(file)->i_mutex);
403 	if (!file->f_pos)
404 		ovl_dir_reset(file);
405 
406 	if (od->is_real) {
407 		res = vfs_llseek(od->realfile, offset, origin);
408 		file->f_pos = od->realfile->f_pos;
409 	} else {
410 		res = -EINVAL;
411 
412 		switch (origin) {
413 		case SEEK_CUR:
414 			offset += file->f_pos;
415 			break;
416 		case SEEK_SET:
417 			break;
418 		default:
419 			goto out_unlock;
420 		}
421 		if (offset < 0)
422 			goto out_unlock;
423 
424 		if (offset != file->f_pos) {
425 			file->f_pos = offset;
426 			if (od->cache)
427 				ovl_seek_cursor(od, offset);
428 		}
429 		res = offset;
430 	}
431 out_unlock:
432 	mutex_unlock(&file_inode(file)->i_mutex);
433 
434 	return res;
435 }
436 
437 static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
438 			 int datasync)
439 {
440 	struct ovl_dir_file *od = file->private_data;
441 	struct dentry *dentry = file->f_path.dentry;
442 	struct file *realfile = od->realfile;
443 
444 	/*
445 	 * Need to check if we started out being a lower dir, but got copied up
446 	 */
447 	if (!od->is_upper && OVL_TYPE_UPPER(ovl_path_type(dentry))) {
448 		struct inode *inode = file_inode(file);
449 
450 		realfile = lockless_dereference(od->upperfile);
451 		if (!realfile) {
452 			struct path upperpath;
453 
454 			ovl_path_upper(dentry, &upperpath);
455 			realfile = ovl_path_open(&upperpath, O_RDONLY);
456 			smp_mb__before_spinlock();
457 			mutex_lock(&inode->i_mutex);
458 			if (!od->upperfile) {
459 				if (IS_ERR(realfile)) {
460 					mutex_unlock(&inode->i_mutex);
461 					return PTR_ERR(realfile);
462 				}
463 				od->upperfile = realfile;
464 			} else {
465 				/* somebody has beaten us to it */
466 				if (!IS_ERR(realfile))
467 					fput(realfile);
468 				realfile = od->upperfile;
469 			}
470 			mutex_unlock(&inode->i_mutex);
471 		}
472 	}
473 
474 	return vfs_fsync_range(realfile, start, end, datasync);
475 }
476 
477 static int ovl_dir_release(struct inode *inode, struct file *file)
478 {
479 	struct ovl_dir_file *od = file->private_data;
480 
481 	if (od->cache) {
482 		mutex_lock(&inode->i_mutex);
483 		ovl_cache_put(od, file->f_path.dentry);
484 		mutex_unlock(&inode->i_mutex);
485 	}
486 	fput(od->realfile);
487 	if (od->upperfile)
488 		fput(od->upperfile);
489 	kfree(od);
490 
491 	return 0;
492 }
493 
494 static int ovl_dir_open(struct inode *inode, struct file *file)
495 {
496 	struct path realpath;
497 	struct file *realfile;
498 	struct ovl_dir_file *od;
499 	enum ovl_path_type type;
500 
501 	od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
502 	if (!od)
503 		return -ENOMEM;
504 
505 	type = ovl_path_real(file->f_path.dentry, &realpath);
506 	realfile = ovl_path_open(&realpath, file->f_flags);
507 	if (IS_ERR(realfile)) {
508 		kfree(od);
509 		return PTR_ERR(realfile);
510 	}
511 	od->realfile = realfile;
512 	od->is_real = !OVL_TYPE_MERGE(type);
513 	od->is_upper = OVL_TYPE_UPPER(type);
514 	file->private_data = od;
515 
516 	return 0;
517 }
518 
519 const struct file_operations ovl_dir_operations = {
520 	.read		= generic_read_dir,
521 	.open		= ovl_dir_open,
522 	.iterate	= ovl_iterate,
523 	.llseek		= ovl_dir_llseek,
524 	.fsync		= ovl_dir_fsync,
525 	.release	= ovl_dir_release,
526 };
527 
528 int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
529 {
530 	int err;
531 	struct ovl_cache_entry *p;
532 
533 	err = ovl_dir_read_merged(dentry, list);
534 	if (err)
535 		return err;
536 
537 	err = 0;
538 
539 	list_for_each_entry(p, list, l_node) {
540 		if (p->is_whiteout)
541 			continue;
542 
543 		if (p->name[0] == '.') {
544 			if (p->len == 1)
545 				continue;
546 			if (p->len == 2 && p->name[1] == '.')
547 				continue;
548 		}
549 		err = -ENOTEMPTY;
550 		break;
551 	}
552 
553 	return err;
554 }
555 
556 void ovl_cleanup_whiteouts(struct dentry *upper, struct list_head *list)
557 {
558 	struct ovl_cache_entry *p;
559 
560 	mutex_lock_nested(&upper->d_inode->i_mutex, I_MUTEX_CHILD);
561 	list_for_each_entry(p, list, l_node) {
562 		struct dentry *dentry;
563 
564 		if (!p->is_whiteout)
565 			continue;
566 
567 		dentry = lookup_one_len(p->name, upper, p->len);
568 		if (IS_ERR(dentry)) {
569 			pr_err("overlayfs: lookup '%s/%.*s' failed (%i)\n",
570 			       upper->d_name.name, p->len, p->name,
571 			       (int) PTR_ERR(dentry));
572 			continue;
573 		}
574 		ovl_cleanup(upper->d_inode, dentry);
575 		dput(dentry);
576 	}
577 	mutex_unlock(&upper->d_inode->i_mutex);
578 }
579