The patch below does not apply to the 6.1-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to stable@vger.kernel.org.
To reproduce the conflict and resubmit, you may use the following commands:
git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.1.y git checkout FETCH_HEAD git cherry-pick -x f50733b45d865f91db90919f8311e2127ce5a0cb # <resolve conflicts, build, test, etc.> git commit -s git send-email --to 'stable@vger.kernel.org' --in-reply-to '2024081450-exploring-lego-5070@gregkh' --subject-prefix 'PATCH 6.1.y' HEAD^..
Possible dependencies:
f50733b45d86 ("exec: Fix ToCToU between perm check and set-uid/gid usage") e67fe63341b8 ("fs: port i_{g,u}id_into_vfs{g,u}id() to mnt_idmap") 9452e93e6dae ("fs: port privilege checking helpers to mnt_idmap") f2d40141d5d9 ("fs: port inode_init_owner() to mnt_idmap") 4609e1f18e19 ("fs: port ->permission() to pass mnt_idmap") 13e83a4923be ("fs: port ->set_acl() to pass mnt_idmap") 77435322777d ("fs: port ->get_acl() to pass mnt_idmap") 011e2b717b1b ("fs: port ->tmpfile() to pass mnt_idmap") 5ebb29bee8d5 ("fs: port ->mknod() to pass mnt_idmap") c54bd91e9eab ("fs: port ->mkdir() to pass mnt_idmap") 7a77db95511c ("fs: port ->symlink() to pass mnt_idmap") 6c960e68aaed ("fs: port ->create() to pass mnt_idmap") b74d24f7a74f ("fs: port ->getattr() to pass mnt_idmap") c1632a0f1120 ("fs: port ->setattr() to pass mnt_idmap") abf08576afe3 ("fs: port vfs_*() helpers to struct mnt_idmap") 6022ec6ee2c3 ("Merge tag 'ntfs3_for_6.2' of https://github.com/Paragon-Software-Group/linux-ntfs3")
thanks,
greg k-h
------------------ original commit in Linus's tree ------------------
From f50733b45d865f91db90919f8311e2127ce5a0cb Mon Sep 17 00:00:00 2001 From: Kees Cook kees@kernel.org Date: Thu, 8 Aug 2024 11:39:08 -0700 Subject: [PATCH] exec: Fix ToCToU between perm check and set-uid/gid usage
When opening a file for exec via do_filp_open(), permission checking is done against the file's metadata at that moment, and on success, a file pointer is passed back. Much later in the execve() code path, the file metadata (specifically mode, uid, and gid) is used to determine if/how to set the uid and gid. However, those values may have changed since the permissions check, meaning the execution may gain unintended privileges.
For example, if a file could change permissions from executable and not set-id:
---------x 1 root root 16048 Aug 7 13:16 target
to set-id and non-executable:
---S------ 1 root root 16048 Aug 7 13:16 target
it is possible to gain root privileges when execution should have been disallowed.
While this race condition is rare in real-world scenarios, it has been observed (and proven exploitable) when package managers are updating the setuid bits of installed programs. Such files start with being world-executable but then are adjusted to be group-exec with a set-uid bit. For example, "chmod o-x,u+s target" makes "target" executable only by uid "root" and gid "cdrom", while also becoming setuid-root:
-rwxr-xr-x 1 root cdrom 16048 Aug 7 13:16 target
becomes:
-rwsr-xr-- 1 root cdrom 16048 Aug 7 13:16 target
But racing the chmod means users without group "cdrom" membership can get the permission to execute "target" just before the chmod, and when the chmod finishes, the exec reaches brpm_fill_uid(), and performs the setuid to root, violating the expressed authorization of "only cdrom group members can setuid to root".
Re-check that we still have execute permissions in case the metadata has changed. It would be better to keep a copy from the perm-check time, but until we can do that refactoring, the least-bad option is to do a full inode_permission() call (under inode lock). It is understood that this is safe against dead-locks, but hardly optimal.
Reported-by: Marco Vanotti mvanotti@google.com Tested-by: Marco Vanotti mvanotti@google.com Suggested-by: Linus Torvalds torvalds@linux-foundation.org Cc: stable@vger.kernel.org Cc: Eric Biederman ebiederm@xmission.com Cc: Alexander Viro viro@zeniv.linux.org.uk Cc: Christian Brauner brauner@kernel.org Signed-off-by: Kees Cook kees@kernel.org
diff --git a/fs/exec.c b/fs/exec.c index a126e3d1cacb..50e76cc633c4 100644 --- a/fs/exec.c +++ b/fs/exec.c @@ -1692,6 +1692,7 @@ static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file) unsigned int mode; vfsuid_t vfsuid; vfsgid_t vfsgid; + int err;
if (!mnt_may_suid(file->f_path.mnt)) return; @@ -1708,12 +1709,17 @@ static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file) /* Be careful if suid/sgid is set */ inode_lock(inode);
- /* reload atomically mode/uid/gid now that lock held */ + /* Atomically reload and check mode/uid/gid now that lock held. */ mode = inode->i_mode; vfsuid = i_uid_into_vfsuid(idmap, inode); vfsgid = i_gid_into_vfsgid(idmap, inode); + err = inode_permission(idmap, inode, MAY_EXEC); inode_unlock(inode);
+ /* Did the exec bit vanish out from under us? Give up. */ + if (err) + return; + /* We ignore suid/sgid if there are no mappings for them in the ns */ if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) || !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid))