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/signal.h>
19 #include <linux/cred.h>
20 #include <linux/namei.h>
21 #include <linux/fdtable.h>
22 #include <linux/ratelimit.h>
23 #include <linux/exportfs.h>
24 #include "overlayfs.h"
25 #include "ovl_entry.h"
27 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
29 static bool __read_mostly ovl_check_copy_up;
30 module_param_named(check_copy_up, ovl_check_copy_up, bool,
32 MODULE_PARM_DESC(ovl_check_copy_up,
33 "Warn on copy-up when causing process also has a R/O fd open");
35 static int ovl_check_fd(const void *data, struct file *f, unsigned int fd)
37 const struct dentry *dentry = data;
39 if (file_inode(f) == d_inode(dentry))
40 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",
41 f, fd, current->pid, current->comm);
46 * Check the fds open by this process and warn if something like the following
47 * scenario is about to occur:
49 * fd1 = open("foo", O_RDONLY);
50 * fd2 = open("foo", O_RDWR);
52 static void ovl_do_check_copy_up(struct dentry *dentry)
54 if (ovl_check_copy_up)
55 iterate_fd(current->files, 0, ovl_check_fd, dentry);
58 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
60 ssize_t list_size, size, value_size = 0;
61 char *buf, *name, *value = NULL;
62 int uninitialized_var(error);
65 if (!(old->d_inode->i_opflags & IOP_XATTR) ||
66 !(new->d_inode->i_opflags & IOP_XATTR))
69 list_size = vfs_listxattr(old, NULL, 0);
71 if (list_size == -EOPNOTSUPP)
76 buf = kzalloc(list_size, GFP_KERNEL);
80 list_size = vfs_listxattr(old, buf, list_size);
86 for (name = buf; list_size; name += slen) {
87 slen = strnlen(name, list_size) + 1;
89 /* underlying fs providing us with an broken xattr list? */
90 if (WARN_ON(slen > list_size)) {
96 if (ovl_is_private_xattr(name))
99 size = vfs_getxattr(old, name, value, value_size);
101 size = vfs_getxattr(old, name, NULL, 0);
108 if (size > value_size) {
111 new = krealloc(value, size, GFP_KERNEL);
121 error = security_inode_copy_up_xattr(name);
122 if (error < 0 && error != -EOPNOTSUPP)
126 continue; /* Discard */
128 error = vfs_setxattr(new, name, value, size, 0);
138 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
140 struct file *old_file;
141 struct file *new_file;
149 old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
150 if (IS_ERR(old_file))
151 return PTR_ERR(old_file);
153 new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
154 if (IS_ERR(new_file)) {
155 error = PTR_ERR(new_file);
159 /* Try to use clone_file_range to clone up within the same fs */
160 error = vfs_clone_file_range(old_file, 0, new_file, 0, len);
163 /* Couldn't clone, so now we try to copy the data */
166 /* FIXME: copy up sparse files efficiently */
168 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
174 if (signal_pending_state(TASK_KILLABLE, current)) {
179 bytes = do_splice_direct(old_file, &old_pos,
181 this_len, SPLICE_F_MOVE);
186 WARN_ON(old_pos != new_pos);
192 error = vfs_fsync(new_file, 0);
199 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
201 struct iattr attr = {
203 ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
204 .ia_atime = stat->atime,
205 .ia_mtime = stat->mtime,
208 return notify_change(upperdentry, &attr, NULL);
211 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
215 if (!S_ISLNK(stat->mode)) {
216 struct iattr attr = {
217 .ia_valid = ATTR_MODE,
218 .ia_mode = stat->mode,
220 err = notify_change(upperdentry, &attr, NULL);
223 struct iattr attr = {
224 .ia_valid = ATTR_UID | ATTR_GID,
228 err = notify_change(upperdentry, &attr, NULL);
231 ovl_set_timestamps(upperdentry, stat);
236 static struct ovl_fh *ovl_encode_fh(struct dentry *lower, uuid_t *uuid)
239 int fh_type, fh_len, dwords;
241 int buflen = MAX_HANDLE_SZ;
243 buf = kmalloc(buflen, GFP_TEMPORARY);
245 return ERR_PTR(-ENOMEM);
248 * We encode a non-connectable file handle for non-dir, because we
249 * only need to find the lower inode number and we don't want to pay
250 * the price or reconnecting the dentry.
252 dwords = buflen >> 2;
253 fh_type = exportfs_encode_fh(lower, buf, &dwords, 0);
254 buflen = (dwords << 2);
257 if (WARN_ON(fh_type < 0) ||
258 WARN_ON(buflen > MAX_HANDLE_SZ) ||
259 WARN_ON(fh_type == FILEID_INVALID))
262 BUILD_BUG_ON(MAX_HANDLE_SZ + offsetof(struct ovl_fh, fid) > 255);
263 fh_len = offsetof(struct ovl_fh, fid) + buflen;
264 fh = kmalloc(fh_len, GFP_KERNEL);
266 fh = ERR_PTR(-ENOMEM);
270 fh->version = OVL_FH_VERSION;
271 fh->magic = OVL_FH_MAGIC;
273 fh->flags = OVL_FH_FLAG_CPU_ENDIAN;
276 memcpy(fh->fid, buf, buflen);
283 static int ovl_set_origin(struct dentry *dentry, struct dentry *lower,
284 struct dentry *upper)
286 struct super_block *sb = lower->d_sb;
287 const struct ovl_fh *fh = NULL;
291 * When lower layer doesn't support export operations store a 'null' fh,
292 * so we can use the overlay.origin xattr to distignuish between a copy
293 * up and a pure upper inode.
295 if (sb->s_export_op && sb->s_export_op->fh_to_dentry &&
296 !uuid_is_null(&sb->s_uuid)) {
297 fh = ovl_encode_fh(lower, &sb->s_uuid);
303 * Do not fail when upper doesn't support xattrs.
305 err = ovl_check_setxattr(dentry, upper, OVL_XATTR_ORIGIN, fh,
306 fh ? fh->len : 0, 0);
312 static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
313 struct dentry *dentry, struct path *lowerpath,
314 struct kstat *stat, const char *link,
315 struct kstat *pstat, bool tmpfile)
317 struct inode *wdir = workdir->d_inode;
318 struct inode *udir = upperdir->d_inode;
319 struct dentry *newdentry = NULL;
320 struct dentry *upper = NULL;
321 struct dentry *temp = NULL;
323 const struct cred *old_creds = NULL;
324 struct cred *new_creds = NULL;
325 struct cattr cattr = {
326 /* Can't properly set mode on creation because of the umask */
327 .mode = stat->mode & S_IFMT,
332 err = security_inode_copy_up(dentry, &new_creds);
337 old_creds = override_creds(new_creds);
340 temp = ovl_do_tmpfile(upperdir, stat->mode);
342 temp = ovl_lookup_temp(workdir);
349 if (!err && !tmpfile)
350 err = ovl_create_real(wdir, temp, &cattr, NULL, true);
353 revert_creds(old_creds);
360 if (S_ISREG(stat->mode)) {
361 struct path upperpath;
363 ovl_path_upper(dentry, &upperpath);
364 BUG_ON(upperpath.dentry != NULL);
365 upperpath.dentry = temp;
369 err = ovl_copy_up_data(lowerpath, &upperpath,
371 inode_lock_nested(udir, I_MUTEX_PARENT);
373 err = ovl_copy_up_data(lowerpath, &upperpath,
381 err = ovl_copy_xattr(lowerpath->dentry, temp);
385 inode_lock(temp->d_inode);
386 err = ovl_set_attr(temp, stat);
387 inode_unlock(temp->d_inode);
392 * Store identifier of lower inode in upper inode xattr to
393 * allow lookup of the copy up origin inode.
395 * Don't set origin when we are breaking the association with a lower
398 if (S_ISDIR(stat->mode) || stat->nlink == 1) {
399 err = ovl_set_origin(dentry, lowerpath->dentry, temp);
404 upper = lookup_one_len(dentry->d_name.name, upperdir,
407 err = PTR_ERR(upper);
413 err = ovl_do_link(temp, udir, upper, true);
415 err = ovl_do_rename(wdir, temp, udir, upper, 0);
419 newdentry = dget(tmpfile ? upper : temp);
420 ovl_dentry_update(dentry, newdentry);
421 ovl_inode_update(d_inode(dentry), d_inode(newdentry));
423 /* Restore timestamps on parent (best effort) */
424 ovl_set_timestamps(upperdir, pstat);
432 ovl_cleanup(wdir, temp);
437 * Copy up a single dentry
439 * All renames start with copy up of source if necessary. The actual
440 * rename will only proceed once the copy up was successful. Copy up uses
441 * upper parent i_mutex for exclusion. Since rename can change d_parent it
442 * is possible that the copy up will lock the old parent. At that point
443 * the file will have already been copied up anyway.
445 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
446 struct path *lowerpath, struct kstat *stat)
448 DEFINE_DELAYED_CALL(done);
449 struct dentry *workdir = ovl_workdir(dentry);
452 struct path parentpath;
453 struct dentry *lowerdentry = lowerpath->dentry;
454 struct dentry *upperdir;
455 const char *link = NULL;
456 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
458 if (WARN_ON(!workdir))
461 ovl_do_check_copy_up(lowerdentry);
463 ovl_path_upper(parent, &parentpath);
464 upperdir = parentpath.dentry;
466 /* Mark parent "impure" because it may now contain non-pure upper */
467 err = ovl_set_impure(parent, upperdir);
471 err = vfs_getattr(&parentpath, &pstat,
472 STATX_ATIME | STATX_MTIME, AT_STATX_SYNC_AS_STAT);
476 if (S_ISLNK(stat->mode)) {
477 link = vfs_get_link(lowerdentry, &done);
479 return PTR_ERR(link);
482 /* Should we copyup with O_TMPFILE or with workdir? */
483 if (S_ISREG(stat->mode) && ofs->tmpfile) {
484 err = ovl_copy_up_start(dentry);
485 /* err < 0: interrupted, err > 0: raced with another copy-up */
487 pr_debug("ovl_copy_up_start(%pd2) = %i\n", dentry, err);
493 inode_lock_nested(upperdir->d_inode, I_MUTEX_PARENT);
494 err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
495 stat, link, &pstat, true);
496 inode_unlock(upperdir->d_inode);
497 ovl_copy_up_end(dentry);
502 if (lock_rename(workdir, upperdir) != NULL) {
503 pr_err("overlayfs: failed to lock workdir+upperdir\n");
506 if (ovl_dentry_upper(dentry)) {
507 /* Raced with another copy-up? Nothing to do, then... */
512 err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
513 stat, link, &pstat, false);
515 unlock_rename(workdir, upperdir);
517 do_delayed_call(&done);
522 int ovl_copy_up_flags(struct dentry *dentry, int flags)
525 const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
529 struct dentry *parent;
530 struct path lowerpath;
532 enum ovl_path_type type = ovl_path_type(dentry);
534 if (OVL_TYPE_UPPER(type))
538 /* find the topmost dentry not yet copied up */
540 parent = dget_parent(next);
542 type = ovl_path_type(parent);
543 if (OVL_TYPE_UPPER(type))
550 ovl_path_lower(next, &lowerpath);
551 err = vfs_getattr(&lowerpath, &stat,
552 STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
553 /* maybe truncate regular file. this has no effect on dirs */
557 err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
562 revert_creds(old_cred);
567 int ovl_copy_up(struct dentry *dentry)
569 return ovl_copy_up_flags(dentry, 0);