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_fh(struct dentry *lower, bool is_upper)
238 int fh_type, fh_len, dwords;
240 int buflen = MAX_HANDLE_SZ;
241 uuid_t *uuid = &lower->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(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;
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 static 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_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 struct ovl_copy_up_ctx {
319 struct dentry *parent;
320 struct dentry *dentry;
321 struct path lowerpath;
325 struct dentry *destdir;
326 struct qstr destname;
327 struct dentry *workdir;
332 static int ovl_link_up(struct ovl_copy_up_ctx *c)
335 struct dentry *upper;
336 struct dentry *upperdir = ovl_dentry_upper(c->parent);
337 struct inode *udir = d_inode(upperdir);
339 /* Mark parent "impure" because it may now contain non-pure upper */
340 err = ovl_set_impure(c->parent, upperdir);
344 err = ovl_set_nlink_lower(c->dentry);
348 inode_lock_nested(udir, I_MUTEX_PARENT);
349 upper = lookup_one_len(c->dentry->d_name.name, upperdir,
350 c->dentry->d_name.len);
351 err = PTR_ERR(upper);
352 if (!IS_ERR(upper)) {
353 err = ovl_do_link(ovl_dentry_upper(c->dentry), udir, upper,
358 /* Restore timestamps on parent (best effort) */
359 ovl_set_timestamps(upperdir, &c->pstat);
360 ovl_dentry_set_upper_alias(c->dentry);
364 ovl_set_nlink_upper(c->dentry);
369 static int ovl_install_temp(struct ovl_copy_up_ctx *c, struct dentry *temp,
370 struct dentry **newdentry)
373 struct dentry *upper;
374 struct inode *udir = d_inode(c->destdir);
376 upper = lookup_one_len(c->destname.name, c->destdir, c->destname.len);
378 return PTR_ERR(upper);
381 err = ovl_do_link(temp, udir, upper, true);
383 err = ovl_do_rename(d_inode(c->workdir), temp, udir, upper, 0);
386 *newdentry = dget(c->tmpfile ? upper : temp);
392 static int ovl_get_tmpfile(struct ovl_copy_up_ctx *c, struct dentry **tempp)
396 const struct cred *old_creds = NULL;
397 struct cred *new_creds = NULL;
398 struct cattr cattr = {
399 /* Can't properly set mode on creation because of the umask */
400 .mode = c->stat.mode & S_IFMT,
401 .rdev = c->stat.rdev,
405 err = security_inode_copy_up(c->dentry, &new_creds);
410 old_creds = override_creds(new_creds);
413 temp = ovl_do_tmpfile(c->workdir, c->stat.mode);
417 temp = ovl_lookup_temp(c->workdir);
421 err = ovl_create_real(d_inode(c->workdir), temp, &cattr,
432 revert_creds(old_creds);
443 static int ovl_copy_up_inode(struct ovl_copy_up_ctx *c, struct dentry *temp)
447 if (S_ISREG(c->stat.mode)) {
448 struct path upperpath;
450 ovl_path_upper(c->dentry, &upperpath);
451 BUG_ON(upperpath.dentry != NULL);
452 upperpath.dentry = temp;
454 err = ovl_copy_up_data(&c->lowerpath, &upperpath, c->stat.size);
459 err = ovl_copy_xattr(c->lowerpath.dentry, temp);
463 inode_lock(temp->d_inode);
464 err = ovl_set_attr(temp, &c->stat);
465 inode_unlock(temp->d_inode);
470 * Store identifier of lower inode in upper inode xattr to
471 * allow lookup of the copy up origin inode.
473 * Don't set origin when we are breaking the association with a lower
477 err = ovl_set_origin(c->dentry, c->lowerpath.dentry, temp);
485 static int ovl_copy_up_locked(struct ovl_copy_up_ctx *c)
487 struct inode *udir = c->destdir->d_inode;
489 struct dentry *newdentry = NULL;
490 struct dentry *temp = NULL;
493 err = ovl_get_tmpfile(c, &temp);
497 err = ovl_copy_up_inode(c, temp);
502 inode_lock_nested(udir, I_MUTEX_PARENT);
503 err = ovl_install_temp(c, temp, &newdentry);
506 err = ovl_install_temp(c, temp, &newdentry);
511 inode = d_inode(c->dentry);
512 ovl_inode_update(inode, newdentry);
513 if (S_ISDIR(inode->i_mode))
514 ovl_set_flag(OVL_WHITEOUTS, inode);
522 ovl_cleanup(d_inode(c->workdir), temp);
527 * Copy up a single dentry
529 * All renames start with copy up of source if necessary. The actual
530 * rename will only proceed once the copy up was successful. Copy up uses
531 * upper parent i_mutex for exclusion. Since rename can change d_parent it
532 * is possible that the copy up will lock the old parent. At that point
533 * the file will have already been copied up anyway.
535 static int ovl_do_copy_up(struct ovl_copy_up_ctx *c)
538 struct ovl_fs *ofs = c->dentry->d_sb->s_fs_info;
539 bool indexed = false;
541 if (ovl_indexdir(c->dentry->d_sb) && !S_ISDIR(c->stat.mode) &&
545 if (S_ISDIR(c->stat.mode) || c->stat.nlink == 1 || indexed)
549 c->destdir = ovl_indexdir(c->dentry->d_sb);
550 err = ovl_get_index_name(c->lowerpath.dentry, &c->destname);
555 * Mark parent "impure" because it may now contain non-pure
558 err = ovl_set_impure(c->parent, c->destdir);
563 /* Should we copyup with O_TMPFILE or with workdir? */
564 if (S_ISREG(c->stat.mode) && ofs->tmpfile) {
566 err = ovl_copy_up_locked(c);
568 err = ovl_lock_rename_workdir(c->workdir, c->destdir);
570 err = ovl_copy_up_locked(c);
571 unlock_rename(c->workdir, c->destdir);
577 ovl_set_flag(OVL_INDEX, d_inode(c->dentry));
578 kfree(c->destname.name);
580 struct inode *udir = d_inode(c->destdir);
582 /* Restore timestamps on parent (best effort) */
584 ovl_set_timestamps(c->destdir, &c->pstat);
587 ovl_dentry_set_upper_alias(c->dentry);
593 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
597 DEFINE_DELAYED_CALL(done);
598 struct path parentpath;
599 struct ovl_copy_up_ctx ctx = {
602 .workdir = ovl_workdir(dentry),
605 if (WARN_ON(!ctx.workdir))
608 ovl_path_lower(dentry, &ctx.lowerpath);
609 err = vfs_getattr(&ctx.lowerpath, &ctx.stat,
610 STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
614 ovl_path_upper(parent, &parentpath);
615 ctx.destdir = parentpath.dentry;
616 ctx.destname = dentry->d_name;
618 err = vfs_getattr(&parentpath, &ctx.pstat,
619 STATX_ATIME | STATX_MTIME, AT_STATX_SYNC_AS_STAT);
623 /* maybe truncate regular file. this has no effect on dirs */
627 if (S_ISLNK(ctx.stat.mode)) {
628 ctx.link = vfs_get_link(ctx.lowerpath.dentry, &done);
629 if (IS_ERR(ctx.link))
630 return PTR_ERR(ctx.link);
632 ovl_do_check_copy_up(ctx.lowerpath.dentry);
634 err = ovl_copy_up_start(dentry);
635 /* err < 0: interrupted, err > 0: raced with another copy-up */
640 if (!ovl_dentry_upper(dentry))
641 err = ovl_do_copy_up(&ctx);
642 if (!err && !ovl_dentry_has_upper_alias(dentry))
643 err = ovl_link_up(&ctx);
644 ovl_copy_up_end(dentry);
646 do_delayed_call(&done);
651 int ovl_copy_up_flags(struct dentry *dentry, int flags)
654 const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
658 struct dentry *parent;
661 * Check if copy-up has happened as well as for upper alias (in
662 * case of hard links) is there.
664 * Both checks are lockless:
665 * - false negatives: will recheck under oi->lock
667 * + ovl_dentry_upper() uses memory barriers to ensure the
668 * upper dentry is up-to-date
669 * + ovl_dentry_has_upper_alias() relies on locking of
670 * upper parent i_rwsem to prevent reordering copy-up
673 if (ovl_dentry_upper(dentry) &&
674 ovl_dentry_has_upper_alias(dentry))
678 /* find the topmost dentry not yet copied up */
680 parent = dget_parent(next);
682 if (ovl_dentry_upper(parent))
689 err = ovl_copy_up_one(parent, next, flags);
694 revert_creds(old_cred);
699 int ovl_copy_up(struct dentry *dentry)
701 return ovl_copy_up_flags(dentry, 0);