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.
10 #include <linux/module.h>
12 #include <linux/slab.h>
13 #include <linux/file.h>
14 #include <linux/splice.h>
15 #include <linux/xattr.h>
16 #include <linux/security.h>
17 #include <linux/uaccess.h>
18 #include <linux/sched.h>
19 #include <linux/namei.h>
20 #include <linux/fdtable.h>
21 #include <linux/ratelimit.h>
22 #include "overlayfs.h"
24 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
26 static bool __read_mostly ovl_check_copy_up;
27 module_param_named(check_copy_up, ovl_check_copy_up, bool,
29 MODULE_PARM_DESC(ovl_check_copy_up,
30 "Warn on copy-up when causing process also has a R/O fd open");
32 static int ovl_check_fd(const void *data, struct file *f, unsigned int fd)
34 const struct dentry *dentry = data;
36 if (f->f_inode == d_inode(dentry))
37 pr_warn_ratelimited("overlayfs: Warning: Copying up %pD, but open R/O on fd %u which will cease to be coherent [pid=%d %s]\n",
38 f, fd, current->pid, current->comm);
43 * Check the fds open by this process and warn if something like the following
44 * scenario is about to occur:
46 * fd1 = open("foo", O_RDONLY);
47 * fd2 = open("foo", O_RDWR);
49 static void ovl_do_check_copy_up(struct dentry *dentry)
51 if (ovl_check_copy_up)
52 iterate_fd(current->files, 0, ovl_check_fd, dentry);
55 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
57 ssize_t list_size, size, value_size = 0;
58 char *buf, *name, *value = NULL;
62 if (!(old->d_inode->i_opflags & IOP_XATTR) ||
63 !(new->d_inode->i_opflags & IOP_XATTR))
66 list_size = vfs_listxattr(old, NULL, 0);
68 if (list_size == -EOPNOTSUPP)
73 buf = kzalloc(list_size, GFP_KERNEL);
77 list_size = vfs_listxattr(old, buf, list_size);
83 for (name = buf; list_size; name += slen) {
84 slen = strnlen(name, list_size) + 1;
86 /* underlying fs providing us with an broken xattr list? */
87 if (WARN_ON(slen > list_size)) {
93 if (ovl_is_private_xattr(name))
96 error = security_inode_copy_up_xattr(name);
97 if (error < 0 && error != -EOPNOTSUPP)
101 continue; /* Discard */
104 size = vfs_getxattr(old, name, value, value_size);
106 size = vfs_getxattr(old, name, NULL, 0);
113 if (size > value_size) {
116 new = krealloc(value, size, GFP_KERNEL);
126 error = vfs_setxattr(new, name, value, size, 0);
136 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
138 struct file *old_file;
139 struct file *new_file;
147 old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
148 if (IS_ERR(old_file))
149 return PTR_ERR(old_file);
151 new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
152 if (IS_ERR(new_file)) {
153 error = PTR_ERR(new_file);
157 /* FIXME: copy up sparse files efficiently */
159 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
165 if (signal_pending_state(TASK_KILLABLE, current)) {
170 bytes = do_splice_direct(old_file, &old_pos,
172 this_len, SPLICE_F_MOVE);
177 WARN_ON(old_pos != new_pos);
183 error = vfs_fsync(new_file, 0);
190 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
192 struct iattr attr = {
194 ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
195 .ia_atime = stat->atime,
196 .ia_mtime = stat->mtime,
199 return notify_change(upperdentry, &attr, NULL);
202 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
206 if (!S_ISLNK(stat->mode)) {
207 struct iattr attr = {
208 .ia_valid = ATTR_MODE,
209 .ia_mode = stat->mode,
211 err = notify_change(upperdentry, &attr, NULL);
214 struct iattr attr = {
215 .ia_valid = ATTR_UID | ATTR_GID,
219 err = notify_change(upperdentry, &attr, NULL);
222 ovl_set_timestamps(upperdentry, stat);
227 static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
228 struct dentry *dentry, struct path *lowerpath,
229 struct kstat *stat, const char *link)
231 struct inode *wdir = workdir->d_inode;
232 struct inode *udir = upperdir->d_inode;
233 struct dentry *newdentry = NULL;
234 struct dentry *upper = NULL;
235 umode_t mode = stat->mode;
237 const struct cred *old_creds = NULL;
238 struct cred *new_creds = NULL;
240 newdentry = ovl_lookup_temp(workdir, dentry);
241 err = PTR_ERR(newdentry);
242 if (IS_ERR(newdentry))
245 upper = lookup_one_len(dentry->d_name.name, upperdir,
247 err = PTR_ERR(upper);
251 err = security_inode_copy_up(dentry, &new_creds);
256 old_creds = override_creds(new_creds);
258 /* Can't properly set mode on creation because of the umask */
259 stat->mode &= S_IFMT;
260 err = ovl_create_real(wdir, newdentry, stat, link, NULL, true);
264 revert_creds(old_creds);
271 if (S_ISREG(stat->mode)) {
272 struct path upperpath;
274 ovl_path_upper(dentry, &upperpath);
275 BUG_ON(upperpath.dentry != NULL);
276 upperpath.dentry = newdentry;
278 err = ovl_copy_up_data(lowerpath, &upperpath, stat->size);
283 err = ovl_copy_xattr(lowerpath->dentry, newdentry);
287 inode_lock(newdentry->d_inode);
288 err = ovl_set_attr(newdentry, stat);
289 inode_unlock(newdentry->d_inode);
293 err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
297 ovl_dentry_update(dentry, newdentry);
298 ovl_inode_update(d_inode(dentry), d_inode(newdentry));
302 * Non-directores become opaque when copied up.
304 if (!S_ISDIR(stat->mode))
305 ovl_dentry_set_opaque(dentry, true);
314 ovl_cleanup(wdir, newdentry);
319 * Copy up a single dentry
321 * Directory renames only allowed on "pure upper" (already created on
322 * upper filesystem, never copied up). Directories which are on lower or
323 * are merged may not be renamed. For these -EXDEV is returned and
324 * userspace has to deal with it. This means, when copying up a
325 * directory we can rely on it and ancestors being stable.
327 * Non-directory renames start with copy up of source if necessary. The
328 * actual rename will only proceed once the copy up was successful. Copy
329 * up uses upper parent i_mutex for exclusion. Since rename can change
330 * d_parent it is possible that the copy up will lock the old parent. At
331 * that point the file will have already been copied up anyway.
333 int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
334 struct path *lowerpath, struct kstat *stat)
336 DEFINE_DELAYED_CALL(done);
337 struct dentry *workdir = ovl_workdir(dentry);
340 struct path parentpath;
341 struct dentry *lowerdentry = lowerpath->dentry;
342 struct dentry *upperdir;
343 struct dentry *upperdentry;
344 const char *link = NULL;
346 if (WARN_ON(!workdir))
349 ovl_do_check_copy_up(lowerdentry);
351 ovl_path_upper(parent, &parentpath);
352 upperdir = parentpath.dentry;
354 err = vfs_getattr(&parentpath, &pstat);
358 if (S_ISLNK(stat->mode)) {
359 link = vfs_get_link(lowerdentry, &done);
361 return PTR_ERR(link);
365 if (lock_rename(workdir, upperdir) != NULL) {
366 pr_err("overlayfs: failed to lock workdir+upperdir\n");
369 upperdentry = ovl_dentry_upper(dentry);
371 /* Raced with another copy-up? Nothing to do, then... */
376 err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
379 /* Restore timestamps on parent (best effort) */
380 ovl_set_timestamps(upperdir, &pstat);
383 unlock_rename(workdir, upperdir);
384 do_delayed_call(&done);
389 int ovl_copy_up(struct dentry *dentry)
392 const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
396 struct dentry *parent;
397 struct path lowerpath;
399 enum ovl_path_type type = ovl_path_type(dentry);
401 if (OVL_TYPE_UPPER(type))
405 /* find the topmost dentry not yet copied up */
407 parent = dget_parent(next);
409 type = ovl_path_type(parent);
410 if (OVL_TYPE_UPPER(type))
417 ovl_path_lower(next, &lowerpath);
418 err = vfs_getattr(&lowerpath, &stat);
420 err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
425 revert_creds(old_cred);