linux-kernel - Re: kernel crash in mknod

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20240325204607.GX538574@ZenIV>
Date: Mon, 25 Mar 2024 20:46:07 +0000
From: Al Viro <viro@...iv.linux.org.uk>
To: Steve French <smfrench@...il.com>
Cc: Christian Brauner <brauner@...nel.org>,
	Roberto Sassu <roberto.sassu@...wei.com>,
	LKML <linux-kernel@...r.kernel.org>,
	linux-fsdevel <linux-fsdevel@...r.kernel.org>,
	CIFS <linux-cifs@...r.kernel.org>,
	Paulo Alcantara <pc@...guebit.com>,
	Christian Brauner <christian@...uner.io>,
	Mimi Zohar <zohar@...ux.ibm.com>, Paul Moore <paul@...l-moore.com>,
	"linux-integrity@...r.kernel.org" <linux-integrity@...r.kernel.org>,
	"linux-security-module@...r.kernel.org" <linux-security-module@...r.kernel.org>
Subject: Re: kernel crash in mknod

On Mon, Mar 25, 2024 at 07:54:13PM +0000, Al Viro wrote:

> Note that cifs_sfu_make_node() is the only case in CIFS where that happens -
> other codepaths (both in cifs_make_node() and in smb2_make_node()) will
> instantiate.  How painful would it be for cifs_sfu_make_node()?
> AFAICS, you do open/sync_write/close there; would it be hard to do
> an eqiuvalent of fstat and set the inode up?  No need to reread the
> file contents (as cifs_sfu_type() does), and you do have full path
> anyway, so it's less work than for full ->lookup() even if you need
> a path-based protocol operations...
> 
> Does that thing have an equivalent of fstat() that would return the
> metadata of opened file?

You do have a FID there, so doing ->query_file_info() just before close,
using the result to build inode (with type and ->i_rdev taken from what
you've been given by the caller) and passing it to d_instantiate() looks
not entirely implausible, but I'm really not familiar with the codebase,
so take that with a cartload of salt.

mknod() usually is followed by lookup of some sort pretty soon, and your
lookup would have to do at least open/sync_read/close just to decode the
device number.  So if anything, *not* setting an inode up during mknod()
is likely to be a pessimization...

If we did it in vfs_mknod() callers, that would be something along the
lines of
	err = vfs_mknod(..., dir, dentry, ...)
	if (err)
		fuck off
	if (unlikely(!dentry->d_inode)) {
		if (d_unhashed(dentry)) {
			struct dentry *d = dir->i_op->lookup(dir, dentry, 0);
			if (unlikely(d)) {
				if (IS_ERR(d)) {
					fuck off, lookup failed
				} else {
					// ->lookup returns a pointer to existing
					// alias *ONLY* for directories; WTF is
					// going on?
					dput(d);
					fuck off, wrong thing created there
				}
			}
			if (!dentry->d_inode)
				fuck off, it hasn't been created
			if (wrong type of dentry->d_inode))
				fuck off, wrong thing created there
			OK, there we go
		} else {
			complain about bogus ->mknod() behavior
			fuck off - it hasn't been created, apparently
		}
	}
at least in net/unix/af_unix.c:unix_bind().  So the minimal change
would be to have your d_drop(dentry) in that codepath followed by
cifs_lookup(<parent inode>, dentry, 0) and checking the result.

But I would very much suspect that fetching metadata by fid before
you close the file would be cheaper than full-blown cifs_lookup()
there.