Otherwise, directory with zero access mode cannot be removed.
This is a revised version of 808c8b25ee,
- dropped O_NOFOLLOW from fd_reopen()
- fixed error handling on opening path in rm_rf().
To remove directory structures, we need to remove the innermost items first,
and then recursively remove higher-level directories. We would recursively
descend into directories and invoke rm_rf_children and rm_rm_children_inner.
This is problematic when too many directories are nested.
Instead, let's create a "TODO" queue. In the the queue, for each level we
hold the DIR* object we were working on, and the name of the directory. This
allows us to leave a partially-processed directory, and restart the removal
loop one level down. When done with the inner directory, we use the name to
unlinkat() it from the parent, and proceed with the removal of other items.
Because the nesting is increased by one level, it is best to view this patch
with -b/--ignore-space-change.
This fixes CVE-2021-3997, https://bugzilla.redhat.com/show_bug.cgi?id=2024639.
The issue was reported and patches reviewed by Qualys Team.
Mauro Matteo Cascella and Riccardo Schirone from Red Hat handled the disclosure.
The variable is not useful outside of the loop (it'll always be null
after the loop is finished), so we can declare it inline in the loop.
This saves one variable declaration and reduces the chances that somebody
tries to use the variable outside of the loop.
For consistency, 'de' is used everywhere for the var name.
The goal is to move everything that requires selinux or smack
away from src/basic/. This means that src/basic/label.[ch] must move,
which implies btrfs-util.[ch], copy.[ch], and a bunch of other files
which form a cluster of internal use.
This is just moving text around, so there should be no functional difference.
test-blockdev-util is new, because path_is_encrypted() is moved to
blockdev-util.c, and so far we didn't have any tests for code there.
Commits 27cc6f166 and f25afeb broke path_is_mount_point() for files (such as
/etc/machine-id → /run/machine-id bind mounts) as with the factorization of
fd_is_mount_point() we lost the parent directory. We cannot determine that from
an fd only as openat(fd, "..") only works for directory fds.
Change fd_is_mount_point() to behave like openat(): It now takes a file
descriptor of the containing directory, a file name in it, and flags (which can
be 0 or AT_SYMLINK_FOLLOW). Unlike name_to_handle_at() or openat(), fstatat()
only accepts the inverse flag AT_SYMLINK_NOFOLLOW and complains with EINVAL
about AT_SYMLINK_FOLLOW; so we need to transform the flags for that fallback.
Adjust rm_rf_children() accordingly (only other caller of fd_is_mount_point()
aside from path_is_mount_point()).
Add test cases for files, links, and file bind mounts (the latter will only
work when running as root). Split out a new test_path_is_mount_point() test
case function as it got significantly larger now.
Yu Watanabe
Luca Boccassi
Zbigniew Jędrzejewski-Szmek
Lennart Poettering
Kay Sievers
Martin Pitt