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/file.h>
13 #include <linux/splice.h>
14 #include <linux/xattr.h>
15 #include <linux/security.h>
16 #include <linux/uaccess.h>
17 #include <linux/sched.h>
18 #include <linux/namei.h>
19 #include "overlayfs.h"
21 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
23 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
25 ssize_t list_size, size, value_size = 0;
26 char *buf, *name, *value = NULL;
27 int uninitialized_var(error);
29 if (!old->d_inode->i_op->getxattr ||
30 !new->d_inode->i_op->getxattr)
33 list_size = vfs_listxattr(old, NULL, 0);
35 if (list_size == -EOPNOTSUPP)
40 buf = kzalloc(list_size, GFP_KERNEL);
44 list_size = vfs_listxattr(old, buf, list_size);
50 for (name = buf; name < (buf + list_size); name += strlen(name) + 1) {
52 size = vfs_getxattr(old, name, value, value_size);
54 size = vfs_getxattr(old, name, NULL, 0);
61 if (size > value_size) {
64 new = krealloc(value, size, GFP_KERNEL);
74 error = vfs_setxattr(new, name, value, size, 0);
84 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
86 struct file *old_file;
87 struct file *new_file;
95 old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
97 return PTR_ERR(old_file);
99 new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
100 if (IS_ERR(new_file)) {
101 error = PTR_ERR(new_file);
105 /* FIXME: copy up sparse files efficiently */
107 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
113 if (signal_pending_state(TASK_KILLABLE, current)) {
118 bytes = do_splice_direct(old_file, &old_pos,
120 this_len, SPLICE_F_MOVE);
125 WARN_ON(old_pos != new_pos);
136 static char *ovl_read_symlink(struct dentry *realdentry)
140 struct inode *inode = realdentry->d_inode;
144 if (!inode->i_op->readlink)
148 buf = (char *) __get_free_page(GFP_KERNEL);
154 /* The cast to a user pointer is valid due to the set_fs() */
155 res = inode->i_op->readlink(realdentry,
156 (char __user *)buf, PAGE_SIZE - 1);
159 free_page((unsigned long) buf);
170 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
172 struct iattr attr = {
174 ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
175 .ia_atime = stat->atime,
176 .ia_mtime = stat->mtime,
179 return notify_change(upperdentry, &attr, NULL);
182 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
186 if (!S_ISLNK(stat->mode)) {
187 struct iattr attr = {
188 .ia_valid = ATTR_MODE,
189 .ia_mode = stat->mode,
191 err = notify_change(upperdentry, &attr, NULL);
194 struct iattr attr = {
195 .ia_valid = ATTR_UID | ATTR_GID,
199 err = notify_change(upperdentry, &attr, NULL);
202 ovl_set_timestamps(upperdentry, stat);
207 static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
208 struct dentry *dentry, struct path *lowerpath,
209 struct kstat *stat, const char *link)
211 struct inode *wdir = workdir->d_inode;
212 struct inode *udir = upperdir->d_inode;
213 struct dentry *newdentry = NULL;
214 struct dentry *upper = NULL;
215 umode_t mode = stat->mode;
218 newdentry = ovl_lookup_temp(workdir, dentry);
219 err = PTR_ERR(newdentry);
220 if (IS_ERR(newdentry))
223 upper = lookup_one_len(dentry->d_name.name, upperdir,
225 err = PTR_ERR(upper);
229 /* Can't properly set mode on creation because of the umask */
230 stat->mode &= S_IFMT;
231 err = ovl_create_real(wdir, newdentry, stat, link, NULL, true);
236 if (S_ISREG(stat->mode)) {
237 struct path upperpath;
238 ovl_path_upper(dentry, &upperpath);
239 BUG_ON(upperpath.dentry != NULL);
240 upperpath.dentry = newdentry;
242 err = ovl_copy_up_data(lowerpath, &upperpath, stat->size);
247 err = ovl_copy_xattr(lowerpath->dentry, newdentry);
251 mutex_lock(&newdentry->d_inode->i_mutex);
252 err = ovl_set_attr(newdentry, stat);
253 mutex_unlock(&newdentry->d_inode->i_mutex);
257 err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
261 ovl_dentry_update(dentry, newdentry);
265 * Non-directores become opaque when copied up.
267 if (!S_ISDIR(stat->mode))
268 ovl_dentry_set_opaque(dentry, true);
277 ovl_cleanup(wdir, newdentry);
282 * Copy up a single dentry
284 * Directory renames only allowed on "pure upper" (already created on
285 * upper filesystem, never copied up). Directories which are on lower or
286 * are merged may not be renamed. For these -EXDEV is returned and
287 * userspace has to deal with it. This means, when copying up a
288 * directory we can rely on it and ancestors being stable.
290 * Non-directory renames start with copy up of source if necessary. The
291 * actual rename will only proceed once the copy up was successful. Copy
292 * up uses upper parent i_mutex for exclusion. Since rename can change
293 * d_parent it is possible that the copy up will lock the old parent. At
294 * that point the file will have already been copied up anyway.
296 int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
297 struct path *lowerpath, struct kstat *stat)
299 struct dentry *workdir = ovl_workdir(dentry);
302 struct path parentpath;
303 struct dentry *upperdir;
304 struct dentry *upperdentry;
305 const struct cred *old_cred;
306 struct cred *override_cred;
309 if (WARN_ON(!workdir))
312 ovl_path_upper(parent, &parentpath);
313 upperdir = parentpath.dentry;
315 err = vfs_getattr(&parentpath, &pstat);
319 if (S_ISLNK(stat->mode)) {
320 link = ovl_read_symlink(lowerpath->dentry);
322 return PTR_ERR(link);
326 override_cred = prepare_creds();
330 override_cred->fsuid = stat->uid;
331 override_cred->fsgid = stat->gid;
333 * CAP_SYS_ADMIN for copying up extended attributes
334 * CAP_DAC_OVERRIDE for create
335 * CAP_FOWNER for chmod, timestamp update
336 * CAP_FSETID for chmod
337 * CAP_CHOWN for chown
338 * CAP_MKNOD for mknod
340 cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
341 cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
342 cap_raise(override_cred->cap_effective, CAP_FOWNER);
343 cap_raise(override_cred->cap_effective, CAP_FSETID);
344 cap_raise(override_cred->cap_effective, CAP_CHOWN);
345 cap_raise(override_cred->cap_effective, CAP_MKNOD);
346 old_cred = override_creds(override_cred);
349 if (lock_rename(workdir, upperdir) != NULL) {
350 pr_err("overlayfs: failed to lock workdir+upperdir\n");
353 upperdentry = ovl_dentry_upper(dentry);
355 /* Raced with another copy-up? Nothing to do, then... */
360 err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
363 /* Restore timestamps on parent (best effort) */
364 ovl_set_timestamps(upperdir, &pstat);
367 unlock_rename(workdir, upperdir);
368 revert_creds(old_cred);
369 put_cred(override_cred);
373 free_page((unsigned long) link);
378 int ovl_copy_up(struct dentry *dentry)
385 struct dentry *parent;
386 struct path lowerpath;
388 enum ovl_path_type type = ovl_path_type(dentry);
390 if (OVL_TYPE_UPPER(type))
394 /* find the topmost dentry not yet copied up */
396 parent = dget_parent(next);
398 type = ovl_path_type(parent);
399 if (OVL_TYPE_UPPER(type))
406 ovl_path_lower(next, &lowerpath);
407 err = vfs_getattr(&lowerpath, &stat);
409 err = ovl_copy_up_one(parent, next, &lowerpath, &stat);