1 // SPDX-License-Identifier: GPL-2.0-only
4 * Copyright (C) 2011 Novell Inc.
7 #include <linux/module.h>
9 #include <linux/slab.h>
10 #include <linux/file.h>
11 #include <linux/fileattr.h>
12 #include <linux/splice.h>
13 #include <linux/xattr.h>
14 #include <linux/security.h>
15 #include <linux/uaccess.h>
16 #include <linux/sched/signal.h>
17 #include <linux/cred.h>
18 #include <linux/namei.h>
19 #include <linux/fdtable.h>
20 #include <linux/ratelimit.h>
21 #include <linux/exportfs.h>
22 #include "overlayfs.h"
24 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
26 static int ovl_ccup_set(const char *buf, const struct kernel_param *param)
28 pr_warn("\"check_copy_up\" module option is obsolete\n");
32 static int ovl_ccup_get(char *buf, const struct kernel_param *param)
34 return sprintf(buf, "N\n");
37 module_param_call(check_copy_up, ovl_ccup_set, ovl_ccup_get, NULL, 0644);
38 MODULE_PARM_DESC(check_copy_up, "Obsolete; does nothing");
40 static bool ovl_must_copy_xattr(const char *name)
42 return !strcmp(name, XATTR_POSIX_ACL_ACCESS) ||
43 !strcmp(name, XATTR_POSIX_ACL_DEFAULT) ||
44 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN);
47 static int ovl_copy_acl(struct ovl_fs *ofs, const struct path *path,
48 struct dentry *dentry, const char *acl_name)
51 struct posix_acl *clone, *real_acl = NULL;
53 real_acl = ovl_get_acl_path(path, acl_name, false);
57 if (IS_ERR(real_acl)) {
58 err = PTR_ERR(real_acl);
59 if (err == -ENODATA || err == -EOPNOTSUPP)
64 clone = posix_acl_clone(real_acl, GFP_KERNEL);
65 posix_acl_release(real_acl); /* release original acl */
69 err = ovl_do_set_acl(ofs, dentry, acl_name, clone);
71 /* release cloned acl */
72 posix_acl_release(clone);
76 int ovl_copy_xattr(struct super_block *sb, const struct path *oldpath, struct dentry *new)
78 struct dentry *old = oldpath->dentry;
79 ssize_t list_size, size, value_size = 0;
80 char *buf, *name, *value = NULL;
84 if (!old->d_inode->i_op->listxattr || !new->d_inode->i_op->listxattr)
87 list_size = vfs_listxattr(old, NULL, 0);
89 if (list_size == -EOPNOTSUPP)
94 buf = kvzalloc(list_size, GFP_KERNEL);
98 list_size = vfs_listxattr(old, buf, list_size);
104 for (name = buf; list_size; name += slen) {
105 slen = strnlen(name, list_size) + 1;
107 /* underlying fs providing us with an broken xattr list? */
108 if (WARN_ON(slen > list_size)) {
114 if (ovl_is_private_xattr(sb, name))
117 error = security_inode_copy_up_xattr(name);
118 if (error < 0 && error != -EOPNOTSUPP)
122 continue; /* Discard */
125 if (is_posix_acl_xattr(name)) {
126 error = ovl_copy_acl(OVL_FS(sb), oldpath, new, name);
129 /* POSIX ACLs must be copied. */
134 size = ovl_do_getxattr(oldpath, name, value, value_size);
136 size = ovl_do_getxattr(oldpath, name, NULL, 0);
143 if (size > value_size) {
146 new = kvmalloc(size, GFP_KERNEL);
157 error = ovl_do_setxattr(OVL_FS(sb), new, name, value, size, 0);
159 if (error != -EOPNOTSUPP || ovl_must_copy_xattr(name))
162 /* Ignore failure to copy unknown xattrs */
172 static int ovl_copy_fileattr(struct inode *inode, const struct path *old,
173 const struct path *new)
175 struct fileattr oldfa = { .flags_valid = true };
176 struct fileattr newfa = { .flags_valid = true };
179 err = ovl_real_fileattr_get(old, &oldfa);
181 /* Ntfs-3g returns -EINVAL for "no fileattr support" */
182 if (err == -ENOTTY || err == -EINVAL)
184 pr_warn("failed to retrieve lower fileattr (%pd2, err=%i)\n",
190 * We cannot set immutable and append-only flags on upper inode,
191 * because we would not be able to link upper inode to upper dir
192 * not set overlay private xattr on upper inode.
193 * Store these flags in overlay.protattr xattr instead.
195 if (oldfa.flags & OVL_PROT_FS_FLAGS_MASK) {
196 err = ovl_set_protattr(inode, new->dentry, &oldfa);
198 pr_warn_once("copying fileattr: no xattr on upper\n");
203 /* Don't bother copying flags if none are set */
204 if (!(oldfa.flags & OVL_COPY_FS_FLAGS_MASK))
207 err = ovl_real_fileattr_get(new, &newfa);
210 * Returning an error if upper doesn't support fileattr will
211 * result in a regression, so revert to the old behavior.
213 if (err == -ENOTTY || err == -EINVAL) {
214 pr_warn_once("copying fileattr: no support on upper\n");
217 pr_warn("failed to retrieve upper fileattr (%pd2, err=%i)\n",
222 BUILD_BUG_ON(OVL_COPY_FS_FLAGS_MASK & ~FS_COMMON_FL);
223 newfa.flags &= ~OVL_COPY_FS_FLAGS_MASK;
224 newfa.flags |= (oldfa.flags & OVL_COPY_FS_FLAGS_MASK);
226 BUILD_BUG_ON(OVL_COPY_FSX_FLAGS_MASK & ~FS_XFLAG_COMMON);
227 newfa.fsx_xflags &= ~OVL_COPY_FSX_FLAGS_MASK;
228 newfa.fsx_xflags |= (oldfa.fsx_xflags & OVL_COPY_FSX_FLAGS_MASK);
230 return ovl_real_fileattr_set(new, &newfa);
233 static int ovl_copy_up_file(struct ovl_fs *ofs, struct dentry *dentry,
234 struct file *new_file, loff_t len)
236 struct path datapath;
237 struct file *old_file;
241 loff_t data_pos = -1;
243 bool skip_hole = false;
246 ovl_path_lowerdata(dentry, &datapath);
247 if (WARN_ON(datapath.dentry == NULL))
250 old_file = ovl_path_open(&datapath, O_LARGEFILE | O_RDONLY);
251 if (IS_ERR(old_file))
252 return PTR_ERR(old_file);
254 /* Try to use clone_file_range to clone up within the same fs */
255 ovl_start_write(dentry);
256 cloned = do_clone_file_range(old_file, 0, new_file, 0, len, 0);
257 ovl_end_write(dentry);
260 /* Couldn't clone, so now we try to copy the data */
262 /* Check if lower fs supports seek operation */
263 if (old_file->f_mode & FMODE_LSEEK)
267 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
273 if (signal_pending_state(TASK_KILLABLE, current)) {
279 * Fill zero for hole will cost unnecessary disk space
280 * and meanwhile slow down the copy-up speed, so we do
281 * an optimization for hole during copy-up, it relies
282 * on SEEK_DATA implementation in lower fs so if lower
283 * fs does not support it, copy-up will behave as before.
285 * Detail logic of hole detection as below:
286 * When we detect next data position is larger than current
287 * position we will skip that hole, otherwise we copy
288 * data in the size of OVL_COPY_UP_CHUNK_SIZE. Actually,
289 * it may not recognize all kind of holes and sometimes
290 * only skips partial of hole area. However, it will be
291 * enough for most of the use cases.
293 * We do not hold upper sb_writers throughout the loop to avert
294 * lockdep warning with llseek of lower file in nested overlay:
296 * -- lower ovl_inode_lock (ovl_llseek)
298 if (skip_hole && data_pos < old_pos) {
299 data_pos = vfs_llseek(old_file, old_pos, SEEK_DATA);
300 if (data_pos > old_pos) {
301 hole_len = data_pos - old_pos;
303 old_pos = new_pos = data_pos;
305 } else if (data_pos == -ENXIO) {
307 } else if (data_pos < 0) {
312 ovl_start_write(dentry);
313 bytes = do_splice_direct(old_file, &old_pos,
315 this_len, SPLICE_F_MOVE);
316 ovl_end_write(dentry);
321 WARN_ON(old_pos != new_pos);
325 if (!error && ovl_should_sync(ofs))
326 error = vfs_fsync(new_file, 0);
332 static int ovl_set_size(struct ovl_fs *ofs,
333 struct dentry *upperdentry, struct kstat *stat)
335 struct iattr attr = {
336 .ia_valid = ATTR_SIZE,
337 .ia_size = stat->size,
340 return ovl_do_notify_change(ofs, upperdentry, &attr);
343 static int ovl_set_timestamps(struct ovl_fs *ofs, struct dentry *upperdentry,
346 struct iattr attr = {
348 ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET | ATTR_CTIME,
349 .ia_atime = stat->atime,
350 .ia_mtime = stat->mtime,
353 return ovl_do_notify_change(ofs, upperdentry, &attr);
356 int ovl_set_attr(struct ovl_fs *ofs, struct dentry *upperdentry,
361 if (!S_ISLNK(stat->mode)) {
362 struct iattr attr = {
363 .ia_valid = ATTR_MODE,
364 .ia_mode = stat->mode,
366 err = ovl_do_notify_change(ofs, upperdentry, &attr);
369 struct iattr attr = {
370 .ia_valid = ATTR_UID | ATTR_GID,
371 .ia_vfsuid = VFSUIDT_INIT(stat->uid),
372 .ia_vfsgid = VFSGIDT_INIT(stat->gid),
374 err = ovl_do_notify_change(ofs, upperdentry, &attr);
377 ovl_set_timestamps(ofs, upperdentry, stat);
382 struct ovl_fh *ovl_encode_real_fh(struct ovl_fs *ofs, struct dentry *real,
387 int buflen = MAX_HANDLE_SZ;
388 uuid_t *uuid = &real->d_sb->s_uuid;
391 /* Make sure the real fid stays 32bit aligned */
392 BUILD_BUG_ON(OVL_FH_FID_OFFSET % 4);
393 BUILD_BUG_ON(MAX_HANDLE_SZ + OVL_FH_FID_OFFSET > 255);
395 fh = kzalloc(buflen + OVL_FH_FID_OFFSET, GFP_KERNEL);
397 return ERR_PTR(-ENOMEM);
400 * We encode a non-connectable file handle for non-dir, because we
401 * only need to find the lower inode number and we don't want to pay
402 * the price or reconnecting the dentry.
404 dwords = buflen >> 2;
405 fh_type = exportfs_encode_fh(real, (void *)fh->fb.fid, &dwords, 0);
406 buflen = (dwords << 2);
409 if (WARN_ON(fh_type < 0) ||
410 WARN_ON(buflen > MAX_HANDLE_SZ) ||
411 WARN_ON(fh_type == FILEID_INVALID))
414 fh->fb.version = OVL_FH_VERSION;
415 fh->fb.magic = OVL_FH_MAGIC;
416 fh->fb.type = fh_type;
417 fh->fb.flags = OVL_FH_FLAG_CPU_ENDIAN;
419 * When we will want to decode an overlay dentry from this handle
420 * and all layers are on the same fs, if we get a disconncted real
421 * dentry when we decode fid, the only way to tell if we should assign
422 * it to upperdentry or to lowerstack is by checking this flag.
425 fh->fb.flags |= OVL_FH_FLAG_PATH_UPPER;
426 fh->fb.len = sizeof(fh->fb) + buflen;
427 if (ovl_origin_uuid(ofs))
437 struct ovl_fh *ovl_get_origin_fh(struct ovl_fs *ofs, struct dentry *origin)
440 * When lower layer doesn't support export operations store a 'null' fh,
441 * so we can use the overlay.origin xattr to distignuish between a copy
442 * up and a pure upper inode.
444 if (!ovl_can_decode_fh(origin->d_sb))
447 return ovl_encode_real_fh(ofs, origin, false);
450 int ovl_set_origin_fh(struct ovl_fs *ofs, const struct ovl_fh *fh,
451 struct dentry *upper)
456 * Do not fail when upper doesn't support xattrs.
458 err = ovl_check_setxattr(ofs, upper, OVL_XATTR_ORIGIN, fh->buf,
459 fh ? fh->fb.len : 0, 0);
461 /* Ignore -EPERM from setting "user.*" on symlink/special */
462 return err == -EPERM ? 0 : err;
465 /* Store file handle of @upper dir in @index dir entry */
466 static int ovl_set_upper_fh(struct ovl_fs *ofs, struct dentry *upper,
467 struct dentry *index)
469 const struct ovl_fh *fh;
472 fh = ovl_encode_real_fh(ofs, upper, true);
476 err = ovl_setxattr(ofs, index, OVL_XATTR_UPPER, fh->buf, fh->fb.len);
483 * Create and install index entry.
485 * Caller must hold i_mutex on indexdir.
487 static int ovl_create_index(struct dentry *dentry, const struct ovl_fh *fh,
488 struct dentry *upper)
490 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
491 struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
492 struct inode *dir = d_inode(indexdir);
493 struct dentry *index = NULL;
494 struct dentry *temp = NULL;
495 struct qstr name = { };
499 * For now this is only used for creating index entry for directories,
500 * because non-dir are copied up directly to index and then hardlinked
503 * TODO: implement create index for non-dir, so we can call it when
504 * encoding file handle for non-dir in case index does not exist.
506 if (WARN_ON(!d_is_dir(dentry)))
509 /* Directory not expected to be indexed before copy up */
510 if (WARN_ON(ovl_test_flag(OVL_INDEX, d_inode(dentry))))
513 err = ovl_get_index_name_fh(fh, &name);
517 temp = ovl_create_temp(ofs, indexdir, OVL_CATTR(S_IFDIR | 0));
522 err = ovl_set_upper_fh(ofs, upper, temp);
526 index = ovl_lookup_upper(ofs, name.name, indexdir, name.len);
528 err = PTR_ERR(index);
530 err = ovl_do_rename(ofs, dir, temp, dir, index, 0);
535 ovl_cleanup(ofs, dir, temp);
542 struct ovl_copy_up_ctx {
543 struct dentry *parent;
544 struct dentry *dentry;
545 struct path lowerpath;
549 struct dentry *destdir;
550 struct qstr destname;
551 struct dentry *workdir;
552 const struct ovl_fh *origin_fh;
556 bool metacopy_digest;
559 static int ovl_link_up(struct ovl_copy_up_ctx *c)
562 struct dentry *upper;
563 struct dentry *upperdir = ovl_dentry_upper(c->parent);
564 struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
565 struct inode *udir = d_inode(upperdir);
567 ovl_start_write(c->dentry);
569 /* Mark parent "impure" because it may now contain non-pure upper */
570 err = ovl_set_impure(c->parent, upperdir);
574 err = ovl_set_nlink_lower(c->dentry);
578 inode_lock_nested(udir, I_MUTEX_PARENT);
579 upper = ovl_lookup_upper(ofs, c->dentry->d_name.name, upperdir,
580 c->dentry->d_name.len);
581 err = PTR_ERR(upper);
582 if (!IS_ERR(upper)) {
583 err = ovl_do_link(ofs, ovl_dentry_upper(c->dentry), udir, upper);
587 /* Restore timestamps on parent (best effort) */
588 ovl_set_timestamps(ofs, upperdir, &c->pstat);
589 ovl_dentry_set_upper_alias(c->dentry);
590 ovl_dentry_update_reval(c->dentry, upper);
597 err = ovl_set_nlink_upper(c->dentry);
600 ovl_end_write(c->dentry);
604 static int ovl_copy_up_data(struct ovl_copy_up_ctx *c, const struct path *temp)
606 struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
607 struct file *new_file;
610 if (!S_ISREG(c->stat.mode) || c->metacopy || !c->stat.size)
613 new_file = ovl_path_open(temp, O_LARGEFILE | O_WRONLY);
614 if (IS_ERR(new_file))
615 return PTR_ERR(new_file);
617 err = ovl_copy_up_file(ofs, c->dentry, new_file, c->stat.size);
623 static int ovl_copy_up_metadata(struct ovl_copy_up_ctx *c, struct dentry *temp)
625 struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
626 struct inode *inode = d_inode(c->dentry);
627 struct path upperpath = { .mnt = ovl_upper_mnt(ofs), .dentry = temp };
630 err = ovl_copy_xattr(c->dentry->d_sb, &c->lowerpath, temp);
634 if (inode->i_flags & OVL_COPY_I_FLAGS_MASK &&
635 (S_ISREG(c->stat.mode) || S_ISDIR(c->stat.mode))) {
637 * Copy the fileattr inode flags that are the source of already
640 err = ovl_copy_fileattr(inode, &c->lowerpath, &upperpath);
646 * Store identifier of lower inode in upper inode xattr to
647 * allow lookup of the copy up origin inode.
649 * Don't set origin when we are breaking the association with a lower
653 err = ovl_set_origin_fh(ofs, c->origin_fh, temp);
659 struct path lowerdatapath;
660 struct ovl_metacopy metacopy_data = OVL_METACOPY_INIT;
662 ovl_path_lowerdata(c->dentry, &lowerdatapath);
663 if (WARN_ON_ONCE(lowerdatapath.dentry == NULL))
665 err = ovl_get_verity_digest(ofs, &lowerdatapath, &metacopy_data);
669 if (metacopy_data.digest_algo)
670 c->metacopy_digest = true;
672 err = ovl_set_metacopy_xattr(ofs, temp, &metacopy_data);
677 inode_lock(temp->d_inode);
678 if (S_ISREG(c->stat.mode))
679 err = ovl_set_size(ofs, temp, &c->stat);
681 err = ovl_set_attr(ofs, temp, &c->stat);
682 inode_unlock(temp->d_inode);
687 struct ovl_cu_creds {
688 const struct cred *old;
692 static int ovl_prep_cu_creds(struct dentry *dentry, struct ovl_cu_creds *cc)
696 cc->old = cc->new = NULL;
697 err = security_inode_copy_up(dentry, &cc->new);
702 cc->old = override_creds(cc->new);
707 static void ovl_revert_cu_creds(struct ovl_cu_creds *cc)
710 revert_creds(cc->old);
716 * Copyup using workdir to prepare temp file. Used when copying up directories,
717 * special files or when upper fs doesn't support O_TMPFILE.
719 static int ovl_copy_up_workdir(struct ovl_copy_up_ctx *c)
721 struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
723 struct inode *udir = d_inode(c->destdir), *wdir = d_inode(c->workdir);
724 struct path path = { .mnt = ovl_upper_mnt(ofs) };
725 struct dentry *temp, *upper;
726 struct ovl_cu_creds cc;
728 struct ovl_cattr cattr = {
729 /* Can't properly set mode on creation because of the umask */
730 .mode = c->stat.mode & S_IFMT,
731 .rdev = c->stat.rdev,
735 err = ovl_prep_cu_creds(c->dentry, &cc);
739 ovl_start_write(c->dentry);
741 temp = ovl_create_temp(ofs, c->workdir, &cattr);
743 ovl_end_write(c->dentry);
744 ovl_revert_cu_creds(&cc);
747 return PTR_ERR(temp);
750 * Copy up data first and then xattrs. Writing data after
751 * xattrs will remove security.capability xattr automatically.
754 err = ovl_copy_up_data(c, &path);
756 * We cannot hold lock_rename() throughout this helper, because or
757 * lock ordering with sb_writers, which shouldn't be held when calling
758 * ovl_copy_up_data(), so lock workdir and destdir and make sure that
759 * temp wasn't moved before copy up completion or cleanup.
760 * If temp was moved, abort without the cleanup.
762 ovl_start_write(c->dentry);
763 if (lock_rename(c->workdir, c->destdir) != NULL ||
764 temp->d_parent != c->workdir) {
771 err = ovl_copy_up_metadata(c, temp);
775 if (S_ISDIR(c->stat.mode) && c->indexed) {
776 err = ovl_create_index(c->dentry, c->origin_fh, temp);
781 upper = ovl_lookup_upper(ofs, c->destname.name, c->destdir,
783 err = PTR_ERR(upper);
787 err = ovl_do_rename(ofs, wdir, temp, udir, upper, 0);
792 inode = d_inode(c->dentry);
793 if (c->metacopy_digest)
794 ovl_set_flag(OVL_HAS_DIGEST, inode);
796 ovl_clear_flag(OVL_HAS_DIGEST, inode);
797 ovl_clear_flag(OVL_VERIFIED_DIGEST, inode);
800 ovl_set_upperdata(inode);
801 ovl_inode_update(inode, temp);
802 if (S_ISDIR(inode->i_mode))
803 ovl_set_flag(OVL_WHITEOUTS, inode);
805 unlock_rename(c->workdir, c->destdir);
806 ovl_end_write(c->dentry);
811 ovl_cleanup(ofs, wdir, temp);
816 /* Copyup using O_TMPFILE which does not require cross dir locking */
817 static int ovl_copy_up_tmpfile(struct ovl_copy_up_ctx *c)
819 struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
820 struct inode *udir = d_inode(c->destdir);
821 struct dentry *temp, *upper;
822 struct file *tmpfile;
823 struct ovl_cu_creds cc;
826 err = ovl_prep_cu_creds(c->dentry, &cc);
830 ovl_start_write(c->dentry);
831 tmpfile = ovl_do_tmpfile(ofs, c->workdir, c->stat.mode);
832 ovl_end_write(c->dentry);
833 ovl_revert_cu_creds(&cc);
835 return PTR_ERR(tmpfile);
837 temp = tmpfile->f_path.dentry;
838 if (!c->metacopy && c->stat.size) {
839 err = ovl_copy_up_file(ofs, c->dentry, tmpfile, c->stat.size);
844 ovl_start_write(c->dentry);
846 err = ovl_copy_up_metadata(c, temp);
850 inode_lock_nested(udir, I_MUTEX_PARENT);
852 upper = ovl_lookup_upper(ofs, c->destname.name, c->destdir,
854 err = PTR_ERR(upper);
855 if (!IS_ERR(upper)) {
856 err = ovl_do_link(ofs, temp, udir, upper);
864 if (c->metacopy_digest)
865 ovl_set_flag(OVL_HAS_DIGEST, d_inode(c->dentry));
867 ovl_clear_flag(OVL_HAS_DIGEST, d_inode(c->dentry));
868 ovl_clear_flag(OVL_VERIFIED_DIGEST, d_inode(c->dentry));
871 ovl_set_upperdata(d_inode(c->dentry));
872 ovl_inode_update(d_inode(c->dentry), dget(temp));
875 ovl_end_write(c->dentry);
882 * Copy up a single dentry
884 * All renames start with copy up of source if necessary. The actual
885 * rename will only proceed once the copy up was successful. Copy up uses
886 * upper parent i_mutex for exclusion. Since rename can change d_parent it
887 * is possible that the copy up will lock the old parent. At that point
888 * the file will have already been copied up anyway.
890 static int ovl_do_copy_up(struct ovl_copy_up_ctx *c)
893 struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
894 struct dentry *origin = c->lowerpath.dentry;
895 struct ovl_fh *fh = NULL;
896 bool to_index = false;
899 * Indexed non-dir is copied up directly to the index entry and then
900 * hardlinked to upper dir. Indexed dir is copied up to indexdir,
901 * then index entry is created and then copied up dir installed.
902 * Copying dir up to indexdir instead of workdir simplifies locking.
904 if (ovl_need_index(c->dentry)) {
906 if (S_ISDIR(c->stat.mode))
907 c->workdir = ovl_indexdir(c->dentry->d_sb);
912 if (S_ISDIR(c->stat.mode) || c->stat.nlink == 1 || to_index) {
913 fh = ovl_get_origin_fh(ofs, origin);
917 /* origin_fh may be NULL */
923 c->destdir = ovl_indexdir(c->dentry->d_sb);
924 err = ovl_get_index_name(ofs, origin, &c->destname);
927 } else if (WARN_ON(!c->parent)) {
928 /* Disconnected dentry must be copied up to index dir */
933 * c->dentry->d_name is stabilzed by ovl_copy_up_start(),
934 * because if we got here, it means that c->dentry has no upper
935 * alias and changing ->d_name means going through ovl_rename()
936 * that will call ovl_copy_up() on source and target dentry.
938 c->destname = c->dentry->d_name;
940 * Mark parent "impure" because it may now contain non-pure
943 ovl_start_write(c->dentry);
944 err = ovl_set_impure(c->parent, c->destdir);
945 ovl_end_write(c->dentry);
950 /* Should we copyup with O_TMPFILE or with workdir? */
951 if (S_ISREG(c->stat.mode) && ofs->tmpfile)
952 err = ovl_copy_up_tmpfile(c);
954 err = ovl_copy_up_workdir(c);
959 ovl_set_flag(OVL_INDEX, d_inode(c->dentry));
961 ovl_start_write(c->dentry);
963 /* Initialize nlink for copy up of disconnected dentry */
964 err = ovl_set_nlink_upper(c->dentry);
966 struct inode *udir = d_inode(c->destdir);
968 /* Restore timestamps on parent (best effort) */
970 ovl_set_timestamps(ofs, c->destdir, &c->pstat);
973 ovl_dentry_set_upper_alias(c->dentry);
974 ovl_dentry_update_reval(c->dentry, ovl_dentry_upper(c->dentry));
976 ovl_end_write(c->dentry);
980 kfree(c->destname.name);
986 static bool ovl_need_meta_copy_up(struct dentry *dentry, umode_t mode,
989 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
991 if (!ofs->config.metacopy)
997 if (flags && ((OPEN_FMODE(flags) & FMODE_WRITE) || (flags & O_TRUNC)))
1000 /* Fall back to full copy if no fsverity on source data and we require verity */
1001 if (ofs->config.verity_mode == OVL_VERITY_REQUIRE) {
1002 struct path lowerdata;
1004 ovl_path_lowerdata(dentry, &lowerdata);
1006 if (WARN_ON_ONCE(lowerdata.dentry == NULL) ||
1007 ovl_ensure_verity_loaded(&lowerdata) ||
1008 !fsverity_active(d_inode(lowerdata.dentry))) {
1016 static ssize_t ovl_getxattr_value(const struct path *path, char *name, char **value)
1021 res = ovl_do_getxattr(path, name, NULL, 0);
1022 if (res == -ENODATA || res == -EOPNOTSUPP)
1026 buf = kzalloc(res, GFP_KERNEL);
1030 res = ovl_do_getxattr(path, name, buf, res);
1039 /* Copy up data of an inode which was copied up metadata only in the past. */
1040 static int ovl_copy_up_meta_inode_data(struct ovl_copy_up_ctx *c)
1042 struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
1043 struct path upperpath;
1045 char *capability = NULL;
1048 ovl_path_upper(c->dentry, &upperpath);
1049 if (WARN_ON(upperpath.dentry == NULL))
1053 err = cap_size = ovl_getxattr_value(&upperpath, XATTR_NAME_CAPS,
1059 err = ovl_copy_up_data(c, &upperpath);
1064 * Writing to upper file will clear security.capability xattr. We
1065 * don't want that to happen for normal copy-up operation.
1067 ovl_start_write(c->dentry);
1069 err = ovl_do_setxattr(ofs, upperpath.dentry, XATTR_NAME_CAPS,
1070 capability, cap_size, 0);
1073 err = ovl_removexattr(ofs, upperpath.dentry,
1074 OVL_XATTR_METACOPY);
1076 ovl_end_write(c->dentry);
1080 ovl_clear_flag(OVL_HAS_DIGEST, d_inode(c->dentry));
1081 ovl_clear_flag(OVL_VERIFIED_DIGEST, d_inode(c->dentry));
1082 ovl_set_upperdata(d_inode(c->dentry));
1089 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
1093 DEFINE_DELAYED_CALL(done);
1094 struct path parentpath;
1095 struct ovl_copy_up_ctx ctx = {
1098 .workdir = ovl_workdir(dentry),
1101 if (WARN_ON(!ctx.workdir))
1104 ovl_path_lower(dentry, &ctx.lowerpath);
1105 err = vfs_getattr(&ctx.lowerpath, &ctx.stat,
1106 STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
1110 if (!kuid_has_mapping(current_user_ns(), ctx.stat.uid) ||
1111 !kgid_has_mapping(current_user_ns(), ctx.stat.gid))
1114 ctx.metacopy = ovl_need_meta_copy_up(dentry, ctx.stat.mode, flags);
1117 ovl_path_upper(parent, &parentpath);
1118 ctx.destdir = parentpath.dentry;
1120 err = vfs_getattr(&parentpath, &ctx.pstat,
1121 STATX_ATIME | STATX_MTIME,
1122 AT_STATX_SYNC_AS_STAT);
1127 /* maybe truncate regular file. this has no effect on dirs */
1128 if (flags & O_TRUNC)
1131 if (S_ISLNK(ctx.stat.mode)) {
1132 ctx.link = vfs_get_link(ctx.lowerpath.dentry, &done);
1133 if (IS_ERR(ctx.link))
1134 return PTR_ERR(ctx.link);
1137 err = ovl_copy_up_start(dentry, flags);
1138 /* err < 0: interrupted, err > 0: raced with another copy-up */
1139 if (unlikely(err)) {
1143 if (!ovl_dentry_upper(dentry))
1144 err = ovl_do_copy_up(&ctx);
1145 if (!err && parent && !ovl_dentry_has_upper_alias(dentry))
1146 err = ovl_link_up(&ctx);
1147 if (!err && ovl_dentry_needs_data_copy_up_locked(dentry, flags))
1148 err = ovl_copy_up_meta_inode_data(&ctx);
1149 ovl_copy_up_end(dentry);
1151 do_delayed_call(&done);
1156 static int ovl_copy_up_flags(struct dentry *dentry, int flags)
1159 const struct cred *old_cred;
1160 bool disconnected = (dentry->d_flags & DCACHE_DISCONNECTED);
1163 * With NFS export, copy up can get called for a disconnected non-dir.
1164 * In this case, we will copy up lower inode to index dir without
1165 * linking it to upper dir.
1167 if (WARN_ON(disconnected && d_is_dir(dentry)))
1171 * We may not need lowerdata if we are only doing metacopy up, but it is
1172 * not very important to optimize this case, so do lazy lowerdata lookup
1173 * before any copy up, so we can do it before taking ovl_inode_lock().
1175 err = ovl_verify_lowerdata(dentry);
1179 old_cred = ovl_override_creds(dentry->d_sb);
1181 struct dentry *next;
1182 struct dentry *parent = NULL;
1184 if (ovl_already_copied_up(dentry, flags))
1187 next = dget(dentry);
1188 /* find the topmost dentry not yet copied up */
1189 for (; !disconnected;) {
1190 parent = dget_parent(next);
1192 if (ovl_dentry_upper(parent))
1199 err = ovl_copy_up_one(parent, next, flags);
1204 revert_creds(old_cred);
1209 static bool ovl_open_need_copy_up(struct dentry *dentry, int flags)
1211 /* Copy up of disconnected dentry does not set upper alias */
1212 if (ovl_already_copied_up(dentry, flags))
1215 if (special_file(d_inode(dentry)->i_mode))
1218 if (!ovl_open_flags_need_copy_up(flags))
1224 int ovl_maybe_copy_up(struct dentry *dentry, int flags)
1226 if (!ovl_open_need_copy_up(dentry, flags))
1229 return ovl_copy_up_flags(dentry, flags);
1232 int ovl_copy_up_with_data(struct dentry *dentry)
1234 return ovl_copy_up_flags(dentry, O_WRONLY);
1237 int ovl_copy_up(struct dentry *dentry)
1239 return ovl_copy_up_flags(dentry, 0);