3 * Copyright (C) 2011 Novell Inc.
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.
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"
20 struct ovl_cache_entry {
25 struct list_head l_node;
30 struct ovl_dir_cache {
33 struct list_head entries;
36 struct ovl_readdir_data {
37 struct dir_context ctx;
40 struct list_head *list;
41 struct list_head middle;
49 struct ovl_dir_cache *cache;
50 struct ovl_cache_entry cursor;
51 struct file *realfile;
52 struct file *upperfile;
55 static struct ovl_cache_entry *ovl_cache_entry_from_node(struct rb_node *n)
57 return container_of(n, struct ovl_cache_entry, node);
60 static struct ovl_cache_entry *ovl_cache_entry_find(struct rb_root *root,
61 const char *name, int len)
63 struct rb_node *node = root->rb_node;
67 struct ovl_cache_entry *p = ovl_cache_entry_from_node(node);
69 cmp = strncmp(name, p->name, len);
71 node = p->node.rb_right;
72 else if (cmp < 0 || len < p->len)
73 node = p->node.rb_left;
81 static struct ovl_cache_entry *ovl_cache_entry_new(const char *name, int len,
82 u64 ino, unsigned int d_type)
84 struct ovl_cache_entry *p;
85 size_t size = offsetof(struct ovl_cache_entry, name[len + 1]);
87 p = kmalloc(size, GFP_KERNEL);
89 memcpy(p->name, name, len);
94 p->is_whiteout = false;
100 static int ovl_cache_entry_add_rb(struct ovl_readdir_data *rdd,
101 const char *name, int len, u64 ino,
104 struct rb_node **newp = &rdd->root.rb_node;
105 struct rb_node *parent = NULL;
106 struct ovl_cache_entry *p;
110 struct ovl_cache_entry *tmp;
113 tmp = ovl_cache_entry_from_node(*newp);
114 cmp = strncmp(name, tmp->name, len);
116 newp = &tmp->node.rb_right;
117 else if (cmp < 0 || len < tmp->len)
118 newp = &tmp->node.rb_left;
123 p = ovl_cache_entry_new(name, len, ino, d_type);
127 list_add_tail(&p->l_node, rdd->list);
128 rb_link_node(&p->node, parent, newp);
129 rb_insert_color(&p->node, &rdd->root);
134 static int ovl_fill_lower(struct ovl_readdir_data *rdd,
135 const char *name, int namelen,
136 loff_t offset, u64 ino, unsigned int d_type)
138 struct ovl_cache_entry *p;
140 p = ovl_cache_entry_find(&rdd->root, name, namelen);
142 list_move_tail(&p->l_node, &rdd->middle);
144 p = ovl_cache_entry_new(name, namelen, ino, d_type);
148 list_add_tail(&p->l_node, &rdd->middle);
154 void ovl_cache_free(struct list_head *list)
156 struct ovl_cache_entry *p;
157 struct ovl_cache_entry *n;
159 list_for_each_entry_safe(p, n, list, l_node)
162 INIT_LIST_HEAD(list);
165 static void ovl_cache_put(struct ovl_dir_file *od, struct dentry *dentry)
167 struct ovl_dir_cache *cache = od->cache;
169 list_del(&od->cursor.l_node);
170 WARN_ON(cache->refcount <= 0);
172 if (!cache->refcount) {
173 if (ovl_dir_cache(dentry) == cache)
174 ovl_set_dir_cache(dentry, NULL);
176 ovl_cache_free(&cache->entries);
181 static int ovl_fill_merge(void *buf, const char *name, int namelen,
182 loff_t offset, u64 ino, unsigned int d_type)
184 struct ovl_readdir_data *rdd = buf;
188 return ovl_cache_entry_add_rb(rdd, name, namelen, ino, d_type);
190 return ovl_fill_lower(rdd, name, namelen, offset, ino, d_type);
193 static inline int ovl_dir_read(struct path *realpath,
194 struct ovl_readdir_data *rdd)
196 struct file *realfile;
199 realfile = ovl_path_open(realpath, O_RDONLY | O_DIRECTORY);
200 if (IS_ERR(realfile))
201 return PTR_ERR(realfile);
207 err = iterate_dir(realfile, &rdd->ctx);
210 } while (!err && rdd->count);
216 static void ovl_dir_reset(struct file *file)
218 struct ovl_dir_file *od = file->private_data;
219 struct ovl_dir_cache *cache = od->cache;
220 struct dentry *dentry = file->f_path.dentry;
221 enum ovl_path_type type = ovl_path_type(dentry);
223 if (cache && ovl_dentry_version_get(dentry) != cache->version) {
224 ovl_cache_put(od, dentry);
227 WARN_ON(!od->is_real && type != OVL_PATH_MERGE);
228 if (od->is_real && type == OVL_PATH_MERGE)
232 static int ovl_dir_mark_whiteouts(struct dentry *dir,
233 struct ovl_readdir_data *rdd)
235 struct ovl_cache_entry *p;
236 struct dentry *dentry;
237 const struct cred *old_cred;
238 struct cred *override_cred;
240 override_cred = prepare_creds();
241 if (!override_cred) {
242 ovl_cache_free(rdd->list);
247 * CAP_DAC_OVERRIDE for lookup
249 cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
250 old_cred = override_creds(override_cred);
252 mutex_lock(&dir->d_inode->i_mutex);
253 list_for_each_entry(p, rdd->list, l_node) {
257 if (p->type != DT_CHR)
260 dentry = lookup_one_len(p->name, dir, p->len);
264 p->is_whiteout = ovl_is_whiteout(dentry);
267 mutex_unlock(&dir->d_inode->i_mutex);
269 revert_creds(old_cred);
270 put_cred(override_cred);
275 static inline int ovl_dir_read_merged(struct path *upperpath,
276 struct path *lowerpath,
277 struct list_head *list)
280 struct ovl_readdir_data rdd = {
281 .ctx.actor = ovl_fill_merge,
287 if (upperpath->dentry) {
288 err = ovl_dir_read(upperpath, &rdd);
292 if (lowerpath->dentry) {
293 err = ovl_dir_mark_whiteouts(upperpath->dentry, &rdd);
298 if (lowerpath->dentry) {
300 * Insert lowerpath entries before upperpath ones, this allows
301 * offsets to be reasonably constant
303 list_add(&rdd.middle, rdd.list);
305 err = ovl_dir_read(lowerpath, &rdd);
306 list_del(&rdd.middle);
313 static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
315 struct ovl_cache_entry *p;
318 list_for_each_entry(p, &od->cache->entries, l_node) {
325 list_move_tail(&od->cursor.l_node, &p->l_node);
328 static struct ovl_dir_cache *ovl_cache_get(struct dentry *dentry)
331 struct path lowerpath;
332 struct path upperpath;
333 struct ovl_dir_cache *cache;
335 cache = ovl_dir_cache(dentry);
336 if (cache && ovl_dentry_version_get(dentry) == cache->version) {
340 ovl_set_dir_cache(dentry, NULL);
342 cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
344 return ERR_PTR(-ENOMEM);
347 INIT_LIST_HEAD(&cache->entries);
349 ovl_path_lower(dentry, &lowerpath);
350 ovl_path_upper(dentry, &upperpath);
352 res = ovl_dir_read_merged(&upperpath, &lowerpath, &cache->entries);
354 ovl_cache_free(&cache->entries);
359 cache->version = ovl_dentry_version_get(dentry);
360 ovl_set_dir_cache(dentry, cache);
365 static int ovl_iterate(struct file *file, struct dir_context *ctx)
367 struct ovl_dir_file *od = file->private_data;
368 struct dentry *dentry = file->f_path.dentry;
374 return iterate_dir(od->realfile, ctx);
377 struct ovl_dir_cache *cache;
379 cache = ovl_cache_get(dentry);
381 return PTR_ERR(cache);
384 ovl_seek_cursor(od, ctx->pos);
387 while (od->cursor.l_node.next != &od->cache->entries) {
388 struct ovl_cache_entry *p;
390 p = list_entry(od->cursor.l_node.next, struct ovl_cache_entry, l_node);
393 if (!p->is_whiteout) {
394 if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
399 list_move(&od->cursor.l_node, &p->l_node);
404 static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
407 struct ovl_dir_file *od = file->private_data;
409 mutex_lock(&file_inode(file)->i_mutex);
414 res = vfs_llseek(od->realfile, offset, origin);
415 file->f_pos = od->realfile->f_pos;
421 offset += file->f_pos;
431 if (offset != file->f_pos) {
432 file->f_pos = offset;
434 ovl_seek_cursor(od, offset);
439 mutex_unlock(&file_inode(file)->i_mutex);
444 static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
447 struct ovl_dir_file *od = file->private_data;
448 struct dentry *dentry = file->f_path.dentry;
449 struct file *realfile = od->realfile;
452 * Need to check if we started out being a lower dir, but got copied up
454 if (!od->is_upper && ovl_path_type(dentry) == OVL_PATH_MERGE) {
455 struct inode *inode = file_inode(file);
457 realfile = od->upperfile;
459 struct path upperpath;
461 ovl_path_upper(dentry, &upperpath);
462 realfile = ovl_path_open(&upperpath, O_RDONLY);
463 mutex_lock(&inode->i_mutex);
464 if (!od->upperfile) {
465 if (IS_ERR(realfile)) {
466 mutex_unlock(&inode->i_mutex);
467 return PTR_ERR(realfile);
469 od->upperfile = realfile;
471 /* somebody has beaten us to it */
472 if (!IS_ERR(realfile))
474 realfile = od->upperfile;
476 mutex_unlock(&inode->i_mutex);
480 return vfs_fsync_range(realfile, start, end, datasync);
483 static int ovl_dir_release(struct inode *inode, struct file *file)
485 struct ovl_dir_file *od = file->private_data;
488 mutex_lock(&inode->i_mutex);
489 ovl_cache_put(od, file->f_path.dentry);
490 mutex_unlock(&inode->i_mutex);
500 static int ovl_dir_open(struct inode *inode, struct file *file)
502 struct path realpath;
503 struct file *realfile;
504 struct ovl_dir_file *od;
505 enum ovl_path_type type;
507 od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
511 type = ovl_path_real(file->f_path.dentry, &realpath);
512 realfile = ovl_path_open(&realpath, file->f_flags);
513 if (IS_ERR(realfile)) {
515 return PTR_ERR(realfile);
517 INIT_LIST_HEAD(&od->cursor.l_node);
518 od->realfile = realfile;
519 od->is_real = (type != OVL_PATH_MERGE);
520 od->is_upper = (type != OVL_PATH_LOWER);
521 file->private_data = od;
526 const struct file_operations ovl_dir_operations = {
527 .read = generic_read_dir,
528 .open = ovl_dir_open,
529 .iterate = ovl_iterate,
530 .llseek = ovl_dir_llseek,
531 .fsync = ovl_dir_fsync,
532 .release = ovl_dir_release,
535 int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
538 struct path lowerpath;
539 struct path upperpath;
540 struct ovl_cache_entry *p;
542 ovl_path_upper(dentry, &upperpath);
543 ovl_path_lower(dentry, &lowerpath);
545 err = ovl_dir_read_merged(&upperpath, &lowerpath, list);
551 list_for_each_entry(p, list, l_node) {
555 if (p->name[0] == '.') {
558 if (p->len == 2 && p->name[1] == '.')
568 void ovl_cleanup_whiteouts(struct dentry *upper, struct list_head *list)
570 struct ovl_cache_entry *p;
572 mutex_lock_nested(&upper->d_inode->i_mutex, I_MUTEX_PARENT);
573 list_for_each_entry(p, list, l_node) {
574 struct dentry *dentry;
579 dentry = lookup_one_len(p->name, upper, p->len);
580 if (IS_ERR(dentry)) {
581 pr_err("overlayfs: lookup '%s/%.*s' failed (%i)\n",
582 upper->d_name.name, p->len, p->name,
583 (int) PTR_ERR(dentry));
586 ovl_cleanup(upper->d_inode, dentry);
589 mutex_unlock(&upper->d_inode->i_mutex);