Received-SPF: pass (google.com: best guess record for domain of linux-ext4-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20200117214246.235591-1-drosen@google.com> <20200117214246.235591-6-drosen@google.com>
 <20200120013528.GY8904@ZenIV.linux.org.uk>
In-Reply-To: <20200120013528.GY8904@ZenIV.linux.org.uk>
From:   Daniel Rosenberg <drosen@google.com>
Date:   Thu, 23 Jan 2020 20:30:49 -0800
Message-ID: <CA+PiJmQPFG7OehStFfNQE_7MGwgozhaa0TxZd+aHL2cFLMFbsA@mail.gmail.com>
Subject: Re: [PATCH v3 5/9] vfs: Fold casefolding into vfs
To:     Al Viro <viro@zeniv.linux.org.uk>
Cc:     "Theodore Ts'o" <tytso@mit.edu>, linux-ext4@vger.kernel.org,
        Jaegeuk Kim <jaegeuk@kernel.org>, Chao Yu <chao@kernel.org>,
        linux-f2fs-devel@lists.sourceforge.net,
        Eric Biggers <ebiggers@kernel.org>,
        linux-fscrypt@vger.kernel.org,
        Andreas Dilger <adilger.kernel@dilger.ca>,
        Jonathan Corbet <corbet@lwn.net>, linux-doc@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org,
        Gabriel Krisman Bertazi <krisman@collabora.com>,
        kernel-team@android.com
Content-Type: text/plain; charset="UTF-8"
Sender: linux-ext4-owner@vger.kernel.org
Precedence: bulk

> ... buggering the filesystems (and boxen) that never planned to use
> that garbage.
>
I'm planning to rework this as dentry ops again. Your other comments
point out some issues that also exist in the old dentry_operations, so
that's a good opportunity to fix those up. How do you feel about just
having the two entries in struct super_block? With them there, I can
add the dentry_operations to fs/unicode where they won't bother anyone
else, while not making every filesystem that uses it have to carry
near identical code.

>
> Are you serious?
>         1) who said that ->d_inode is stable here?  If we are in RCU mode,
> it won't be.
>         2) page-sized kmalloc/kfree *ON* *COMPONENT* *AFTER* *COMPONENT*?
>

#2 is the part that made me the saddest in the patch. I'm planning to
move this to the unicode subsystem so it can just walk through the
name as it computes the hash without needing any allocation.

>
> ... and again, you are pulling in a lot of cachelines.
>

I probably should've just given it a DCACHE flag, like what fscrypt is
using. A simple flag there would've done everything that I'm doing
without making the cache super sad and making any attempts at making
it actually work with RCU much simpler.

> <understatement> IMO the whole thing is not a good idea. </understatement>