Received-SPF: pass (google.com: domain of linux-ext4+bounces-910-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Date: Wed, 24 Jan 2024 19:05:30 -0800
From: Eric Biggers <ebiggers@kernel.org>
To: Gabriel Krisman Bertazi <krisman@suse.de>
Cc: viro@zeniv.linux.org.uk, jaegeuk@kernel.org, tytso@mit.edu,
	linux-ext4@vger.kernel.org, linux-f2fs-devel@lists.sourceforge.net,
	linux-fsdevel@vger.kernel.org, amir73il@gmail.com
Subject: Re: [PATCH v3 02/10] fscrypt: Share code between functions that
 prepare lookup
Message-ID: <20240125030530.GB52073@sol.localdomain>
References: <20240119184742.31088-1-krisman@suse.de>
 <20240119184742.31088-3-krisman@suse.de>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20240119184742.31088-3-krisman@suse.de>

On Fri, Jan 19, 2024 at 03:47:34PM -0300, Gabriel Krisman Bertazi wrote:
> To make the patch simpler, we now call fscrypt_get_encryption_info twice
> for fscrypt_prepare_lookup, once inside fscrypt_setup_filename and once
> inside fscrypt_prepare_lookup_dentry.  It seems safe to do, and
> considering it will bail early in the second lookup and most lookups
> should go to the dcache anyway, it doesn't seem problematic for
> performance.  In addition, we add a function call for the unencrypted
> case, also during lookup.

Unfortunately I don't think it's correct.  This is basically undoing my fix
b01531db6cec ("fscrypt: fix race where ->lookup() marks plaintext dentry as
ciphertext") from several years ago.

When a lookup is done, the filesystem needs to either treat the name being
looked up as a no-key name *or* as a regular name, depending on whether the
directory's key is present.  We shouldn't enable race conditions where, due to
the key being concurrently added, the name is treated as a no-key name for
filename matching purposes but a regular name for dentry validation purposes.
That can result in an anomaly where a file that exists ends up with a negative
dentry that doesn't get invalidated.

Basically, the boolean fscrypt_name::is_nokey_name that's produced by
fscrypt_setup_filename() should continue to be propagated to DCACHE_NOKEY_NAME.

- Eric