linux-kernel - Re: [f2fs-dev] [PATCH v3] f2fs: fix race of pending

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: <X88ZC4f2hZxAU3C0@gmail.com>
Date:   Mon, 7 Dec 2020 22:11:23 -0800
From:   Eric Biggers <ebiggers@...nel.org>
To:     Daeho Jeong <daeho43@...il.com>
Cc:     linux-kernel@...r.kernel.org,
        linux-f2fs-devel@...ts.sourceforge.net, kernel-team@...roid.com,
        Jaegeuk Kim <jaegeuk@...nel.org>,
        Daeho Jeong <daehojeong@...gle.com>
Subject: Re: [f2fs-dev] [PATCH v3] f2fs: fix race of pending_pages in
 decompression

On Tue, Dec 08, 2020 at 08:51:45AM +0900, Daeho Jeong wrote:
> > I am trying to review this but it is very hard, as the f2fs compression code is
> > very hard to understand.
> >
> > It looks like a 'struct decompress_io_ctx' represents the work to decompress a
> > particular cluster.  Since the compressed data of the cluster can be read using
> > multiple bios, there is a reference count of how many pages are remaining to be
> > read before all the cluster's pages have been read and decompression can start.
> >
> > What I don't understand is why that reference counting needs to work differently
> > depending on whether verity is enabled or not.  Shouldn't it be exactly the
> > same?
> >
> > There also seems to be some confusion about the scope of STEP_VERITY.  Before
> > f2fs compression was added, it was a per-bio thing.  But now in a compressed
> > file, it's really a per-cluster thing, since all decompressed pages in a
> > compressed cluster are verified (or not verified) at once.
> >
> > Wouldn't it make a lot more sense to, when a cluster needs both compression and
> > verity, *not* set STEP_VERITY on the bios, but rather set a similar flag in the
> > decompress_io_ctx?
> >
> 
> Eric,
> 
> Decompression and verity can be executed in different thread contexts
> in different timing, so we need separate counts for each.
> 
> We already use STEP_VERITY for non-compression case, so I think using
> this flag in here looks more making sense.
> 
> Thanks,

That didn't really answer my questions.

I gave up trying to review this patch as the compression post-read handling is
just way too weird and hard to understand.  I wrote a patch to clean it all up
instead, please take a look:
https://lkml.kernel.org/r/20201208060328.2237091-1-ebiggers@kernel.org

- Eric