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"
26 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
28 static bool __read_mostly ovl_check_copy_up;
29 module_param_named(check_copy_up, ovl_check_copy_up, bool,
31 MODULE_PARM_DESC(ovl_check_copy_up,
32 "Warn on copy-up when causing process also has a R/O fd open");
34 static int ovl_check_fd(const void *data, struct file *f, unsigned int fd)
36 const struct dentry *dentry = data;
38 if (file_inode(f) == d_inode(dentry))
39 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",
40 f, fd, current->pid, current->comm);
45 * Check the fds open by this process and warn if something like the following
46 * scenario is about to occur:
48 * fd1 = open("foo", O_RDONLY);
49 * fd2 = open("foo", O_RDWR);
51 static void ovl_do_check_copy_up(struct dentry *dentry)
53 if (ovl_check_copy_up)
54 iterate_fd(current->files, 0, ovl_check_fd, dentry);
57 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
59 ssize_t list_size, size, value_size = 0;
60 char *buf, *name, *value = NULL;
61 int uninitialized_var(error);
64 if (!(old->d_inode->i_opflags & IOP_XATTR) ||
65 !(new->d_inode->i_opflags & IOP_XATTR))
68 list_size = vfs_listxattr(old, NULL, 0);
70 if (list_size == -EOPNOTSUPP)
75 buf = kzalloc(list_size, GFP_KERNEL);
79 list_size = vfs_listxattr(old, buf, list_size);
85 for (name = buf; list_size; name += slen) {
86 slen = strnlen(name, list_size) + 1;
88 /* underlying fs providing us with an broken xattr list? */
89 if (WARN_ON(slen > list_size)) {
95 if (ovl_is_private_xattr(name))
98 size = vfs_getxattr(old, name, value, value_size);
100 size = vfs_getxattr(old, name, NULL, 0);
107 if (size > value_size) {
110 new = krealloc(value, size, GFP_KERNEL);
120 error = security_inode_copy_up_xattr(name);
121 if (error < 0 && error != -EOPNOTSUPP)
125 continue; /* Discard */
127 error = vfs_setxattr(new, name, value, size, 0);
137 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
139 struct file *old_file;
140 struct file *new_file;
148 old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
149 if (IS_ERR(old_file))
150 return PTR_ERR(old_file);
152 new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
153 if (IS_ERR(new_file)) {
154 error = PTR_ERR(new_file);
158 /* Try to use clone_file_range to clone up within the same fs */
159 error = vfs_clone_file_range(old_file, 0, new_file, 0, len);
162 /* Couldn't clone, so now we try to copy the data */
165 /* FIXME: copy up sparse files efficiently */
167 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
173 if (signal_pending_state(TASK_KILLABLE, current)) {
178 bytes = do_splice_direct(old_file, &old_pos,
180 this_len, SPLICE_F_MOVE);
185 WARN_ON(old_pos != new_pos);
191 error = vfs_fsync(new_file, 0);
198 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
200 struct iattr attr = {
202 ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
203 .ia_atime = stat->atime,
204 .ia_mtime = stat->mtime,
207 return notify_change(upperdentry, &attr, NULL);
210 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
214 if (!S_ISLNK(stat->mode)) {
215 struct iattr attr = {
216 .ia_valid = ATTR_MODE,
217 .ia_mode = stat->mode,
219 err = notify_change(upperdentry, &attr, NULL);
222 struct iattr attr = {
223 .ia_valid = ATTR_UID | ATTR_GID,
227 err = notify_change(upperdentry, &attr, NULL);
230 ovl_set_timestamps(upperdentry, stat);
235 struct ovl_fh *ovl_encode_real_fh(struct dentry *real, bool is_upper)
238 int fh_type, fh_len, dwords;
240 int buflen = MAX_HANDLE_SZ;
241 uuid_t *uuid = &real->d_sb->s_uuid;
243 buf = kmalloc(buflen, GFP_KERNEL);
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(real, 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;
275 * When we will want to decode an overlay dentry from this handle
276 * and all layers are on the same fs, if we get a disconncted real
277 * dentry when we decode fid, the only way to tell if we should assign
278 * it to upperdentry or to lowerstack is by checking this flag.
281 fh->flags |= OVL_FH_FLAG_PATH_UPPER;
284 memcpy(fh->fid, buf, buflen);
291 int ovl_set_origin(struct dentry *dentry, struct dentry *lower,
292 struct dentry *upper)
294 const struct ovl_fh *fh = NULL;
298 * When lower layer doesn't support export operations store a 'null' fh,
299 * so we can use the overlay.origin xattr to distignuish between a copy
300 * up and a pure upper inode.
302 if (ovl_can_decode_fh(lower->d_sb)) {
303 fh = ovl_encode_real_fh(lower, false);
309 * Do not fail when upper doesn't support xattrs.
311 err = ovl_check_setxattr(dentry, upper, OVL_XATTR_ORIGIN, fh,
312 fh ? fh->len : 0, 0);
318 /* Store file handle of @upper dir in @index dir entry */
319 static int ovl_set_upper_fh(struct dentry *upper, struct dentry *index)
321 const struct ovl_fh *fh;
324 fh = ovl_encode_real_fh(upper, true);
328 err = ovl_do_setxattr(index, OVL_XATTR_UPPER, fh, fh->len, 0);
335 * Create and install index entry.
337 * Caller must hold i_mutex on indexdir.
339 static int ovl_create_index(struct dentry *dentry, struct dentry *origin,
340 struct dentry *upper)
342 struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
343 struct inode *dir = d_inode(indexdir);
344 struct dentry *index = NULL;
345 struct dentry *temp = NULL;
346 struct qstr name = { };
350 * For now this is only used for creating index entry for directories,
351 * because non-dir are copied up directly to index and then hardlinked
354 * TODO: implement create index for non-dir, so we can call it when
355 * encoding file handle for non-dir in case index does not exist.
357 if (WARN_ON(!d_is_dir(dentry)))
360 /* Directory not expected to be indexed before copy up */
361 if (WARN_ON(ovl_test_flag(OVL_INDEX, d_inode(dentry))))
364 err = ovl_get_index_name(origin, &name);
368 temp = ovl_create_temp(indexdir, OVL_CATTR(S_IFDIR | 0));
373 err = ovl_set_upper_fh(upper, temp);
377 index = lookup_one_len(name.name, indexdir, name.len);
379 err = PTR_ERR(index);
381 err = ovl_do_rename(dir, temp, dir, index, 0);
386 ovl_cleanup(dir, temp);
393 struct ovl_copy_up_ctx {
394 struct dentry *parent;
395 struct dentry *dentry;
396 struct path lowerpath;
400 struct dentry *destdir;
401 struct qstr destname;
402 struct dentry *workdir;
408 static int ovl_link_up(struct ovl_copy_up_ctx *c)
411 struct dentry *upper;
412 struct dentry *upperdir = ovl_dentry_upper(c->parent);
413 struct inode *udir = d_inode(upperdir);
415 /* Mark parent "impure" because it may now contain non-pure upper */
416 err = ovl_set_impure(c->parent, upperdir);
420 err = ovl_set_nlink_lower(c->dentry);
424 inode_lock_nested(udir, I_MUTEX_PARENT);
425 upper = lookup_one_len(c->dentry->d_name.name, upperdir,
426 c->dentry->d_name.len);
427 err = PTR_ERR(upper);
428 if (!IS_ERR(upper)) {
429 err = ovl_do_link(ovl_dentry_upper(c->dentry), udir, upper);
433 /* Restore timestamps on parent (best effort) */
434 ovl_set_timestamps(upperdir, &c->pstat);
435 ovl_dentry_set_upper_alias(c->dentry);
442 err = ovl_set_nlink_upper(c->dentry);
447 static int ovl_install_temp(struct ovl_copy_up_ctx *c, struct dentry *temp,
448 struct dentry **newdentry)
451 struct dentry *upper;
452 struct inode *udir = d_inode(c->destdir);
454 upper = lookup_one_len(c->destname.name, c->destdir, c->destname.len);
456 return PTR_ERR(upper);
459 err = ovl_do_link(temp, udir, upper);
461 err = ovl_do_rename(d_inode(c->workdir), temp, udir, upper, 0);
464 *newdentry = dget(c->tmpfile ? upper : temp);
470 static struct dentry *ovl_get_tmpfile(struct ovl_copy_up_ctx *c)
474 const struct cred *old_creds = NULL;
475 struct cred *new_creds = NULL;
476 struct ovl_cattr cattr = {
477 /* Can't properly set mode on creation because of the umask */
478 .mode = c->stat.mode & S_IFMT,
479 .rdev = c->stat.rdev,
483 err = security_inode_copy_up(c->dentry, &new_creds);
489 old_creds = override_creds(new_creds);
492 temp = ovl_do_tmpfile(c->workdir, c->stat.mode);
494 temp = ovl_create_temp(c->workdir, &cattr);
497 revert_creds(old_creds);
504 static int ovl_copy_up_inode(struct ovl_copy_up_ctx *c, struct dentry *temp)
508 if (S_ISREG(c->stat.mode)) {
509 struct path upperpath;
511 ovl_path_upper(c->dentry, &upperpath);
512 BUG_ON(upperpath.dentry != NULL);
513 upperpath.dentry = temp;
515 err = ovl_copy_up_data(&c->lowerpath, &upperpath, c->stat.size);
520 err = ovl_copy_xattr(c->lowerpath.dentry, temp);
524 inode_lock(temp->d_inode);
525 err = ovl_set_attr(temp, &c->stat);
526 inode_unlock(temp->d_inode);
531 * Store identifier of lower inode in upper inode xattr to
532 * allow lookup of the copy up origin inode.
534 * Don't set origin when we are breaking the association with a lower
538 err = ovl_set_origin(c->dentry, c->lowerpath.dentry, temp);
546 static int ovl_copy_up_locked(struct ovl_copy_up_ctx *c)
548 struct inode *udir = c->destdir->d_inode;
550 struct dentry *newdentry = NULL;
554 temp = ovl_get_tmpfile(c);
556 return PTR_ERR(temp);
558 err = ovl_copy_up_inode(c, temp);
562 if (S_ISDIR(c->stat.mode) && c->indexed) {
563 err = ovl_create_index(c->dentry, c->lowerpath.dentry, temp);
569 inode_lock_nested(udir, I_MUTEX_PARENT);
570 err = ovl_install_temp(c, temp, &newdentry);
573 err = ovl_install_temp(c, temp, &newdentry);
578 inode = d_inode(c->dentry);
579 ovl_inode_update(inode, newdentry);
580 if (S_ISDIR(inode->i_mode))
581 ovl_set_flag(OVL_WHITEOUTS, inode);
584 if (err && !c->tmpfile)
585 ovl_cleanup(d_inode(c->workdir), temp);
592 * Copy up a single dentry
594 * All renames start with copy up of source if necessary. The actual
595 * rename will only proceed once the copy up was successful. Copy up uses
596 * upper parent i_mutex for exclusion. Since rename can change d_parent it
597 * is possible that the copy up will lock the old parent. At that point
598 * the file will have already been copied up anyway.
600 static int ovl_do_copy_up(struct ovl_copy_up_ctx *c)
603 struct ovl_fs *ofs = c->dentry->d_sb->s_fs_info;
604 bool to_index = false;
607 * Indexed non-dir is copied up directly to the index entry and then
608 * hardlinked to upper dir. Indexed dir is copied up to indexdir,
609 * then index entry is created and then copied up dir installed.
610 * Copying dir up to indexdir instead of workdir simplifies locking.
612 if (ovl_need_index(c->dentry)) {
614 if (S_ISDIR(c->stat.mode))
615 c->workdir = ovl_indexdir(c->dentry->d_sb);
620 if (S_ISDIR(c->stat.mode) || c->stat.nlink == 1 || to_index)
624 c->destdir = ovl_indexdir(c->dentry->d_sb);
625 err = ovl_get_index_name(c->lowerpath.dentry, &c->destname);
628 } else if (WARN_ON(!c->parent)) {
629 /* Disconnected dentry must be copied up to index dir */
633 * Mark parent "impure" because it may now contain non-pure
636 err = ovl_set_impure(c->parent, c->destdir);
641 /* Should we copyup with O_TMPFILE or with workdir? */
642 if (S_ISREG(c->stat.mode) && ofs->tmpfile) {
644 err = ovl_copy_up_locked(c);
646 err = ovl_lock_rename_workdir(c->workdir, c->destdir);
648 err = ovl_copy_up_locked(c);
649 unlock_rename(c->workdir, c->destdir);
658 ovl_set_flag(OVL_INDEX, d_inode(c->dentry));
661 /* Initialize nlink for copy up of disconnected dentry */
662 err = ovl_set_nlink_upper(c->dentry);
664 struct inode *udir = d_inode(c->destdir);
666 /* Restore timestamps on parent (best effort) */
668 ovl_set_timestamps(c->destdir, &c->pstat);
671 ovl_dentry_set_upper_alias(c->dentry);
676 kfree(c->destname.name);
680 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
684 DEFINE_DELAYED_CALL(done);
685 struct path parentpath;
686 struct ovl_copy_up_ctx ctx = {
689 .workdir = ovl_workdir(dentry),
692 if (WARN_ON(!ctx.workdir))
695 ovl_path_lower(dentry, &ctx.lowerpath);
696 err = vfs_getattr(&ctx.lowerpath, &ctx.stat,
697 STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
702 ovl_path_upper(parent, &parentpath);
703 ctx.destdir = parentpath.dentry;
704 ctx.destname = dentry->d_name;
706 err = vfs_getattr(&parentpath, &ctx.pstat,
707 STATX_ATIME | STATX_MTIME,
708 AT_STATX_SYNC_AS_STAT);
713 /* maybe truncate regular file. this has no effect on dirs */
717 if (S_ISLNK(ctx.stat.mode)) {
718 ctx.link = vfs_get_link(ctx.lowerpath.dentry, &done);
719 if (IS_ERR(ctx.link))
720 return PTR_ERR(ctx.link);
722 ovl_do_check_copy_up(ctx.lowerpath.dentry);
724 err = ovl_copy_up_start(dentry);
725 /* err < 0: interrupted, err > 0: raced with another copy-up */
730 if (!ovl_dentry_upper(dentry))
731 err = ovl_do_copy_up(&ctx);
732 if (!err && parent && !ovl_dentry_has_upper_alias(dentry))
733 err = ovl_link_up(&ctx);
734 ovl_copy_up_end(dentry);
736 do_delayed_call(&done);
741 int ovl_copy_up_flags(struct dentry *dentry, int flags)
744 const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
745 bool disconnected = (dentry->d_flags & DCACHE_DISCONNECTED);
748 * With NFS export, copy up can get called for a disconnected non-dir.
749 * In this case, we will copy up lower inode to index dir without
750 * linking it to upper dir.
752 if (WARN_ON(disconnected && d_is_dir(dentry)))
757 struct dentry *parent = NULL;
760 * Check if copy-up has happened as well as for upper alias (in
761 * case of hard links) is there.
763 * Both checks are lockless:
764 * - false negatives: will recheck under oi->lock
766 * + ovl_dentry_upper() uses memory barriers to ensure the
767 * upper dentry is up-to-date
768 * + ovl_dentry_has_upper_alias() relies on locking of
769 * upper parent i_rwsem to prevent reordering copy-up
772 if (ovl_dentry_upper(dentry) &&
773 (ovl_dentry_has_upper_alias(dentry) || disconnected))
777 /* find the topmost dentry not yet copied up */
778 for (; !disconnected;) {
779 parent = dget_parent(next);
781 if (ovl_dentry_upper(parent))
788 err = ovl_copy_up_one(parent, next, flags);
793 revert_creds(old_cred);
798 int ovl_copy_up(struct dentry *dentry)
800 return ovl_copy_up_flags(dentry, 0);