linux-kernel - Re: [PATCH] fs: ufs: Remove switch statement from ufs_set_de

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]

Date:   Sun, 21 Oct 2018 08:30:35 +0300
From:   Amir Goldstein <amir73il@...il.com>
To:     Matthew Wilcox <willy@...radead.org>
Cc:     phil@...lpotter.co.uk, Evgeniy Dushistov <dushistov@...l.ru>,
        Al Viro <viro@...iv.linux.org.uk>, David.Laight@...lab.com,
        linux-kernel <linux-kernel@...r.kernel.org>,
        linux-fsdevel <linux-fsdevel@...r.kernel.org>
Subject: Re: [PATCH] fs: ufs: Remove switch statement from ufs_set_de_type function

On Sun, Oct 21, 2018 at 1:27 AM Matthew Wilcox <willy@...radead.org> wrote:
>
> On Sat, Oct 20, 2018 at 11:09:57PM +0100, Phillip Potter wrote:
> > Remove switch statement from ufs_set_de_type function in fs/ufs/util.h
> > header and replace with simple assignment. For each case, S_IFx >> 12
> > is equal to DT_x, so in valid cases (mode & S_IFMT) >> 12 should give
> > us the correct file type. It is expected that for *nix compatibility
> > reasons, the relation between S_IFx and DT_x will not change. For
> > cases where the mode is invalid, upper layer validation catches this
> > anyway, so this improves readability and arguably performance by
> > assigning (mode & S_IFMT) >> 12 directly without any jump table
> > or conditional logic.
>
> Shouldn't we also do this for other filesystems?  A quick scan suggests
> someone should at least look at xfs, ubifs, squashfs, romfs, ocfs2,
> nilfs2, hfsplus, f2fs, ext4, ext2, exofs, coda, cifs & btrfs.
>

I've tried to do that 2 years ago, not even for readability, but to
fix a lurking
bug in conversion table implementation that was copy&pasted from ext2 to
many other fs:
https://marc.info/?l=linux-fsdevel&m=148217829301701&w=2
https://marc.info/?l=linux-fsdevel&w=2&r=1&s=amir73il+file+type+conversion&q=b

The push back from ext4/xfs maintainers was that while all fs use common
values to encode d_type in on disk format, those on-disk format values
are private to the fs.

Eventually, the fix that got merged (only to xfs) did the opposite, it converted
the conversion table lookup to a switch statement:
https://marc.info/?l=linux-fsdevel&m=148243866204444&w=2

Some fs maintainers were happy about the initial version, but I did not
pursue pushing that cleanup:
https://marc.info/?l=linux-fsdevel&m=148243866204444&w=2

Phillip, you are welcome to re-spin those patches if you like.
Note that the generic conversion helpers do not rely on upper layer
to catch invalid mode values.

Cheers,
Amir.