linux-ext4 - Re: More KCSAN data-race Reports

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] [day] [month] [year] [list]

Date:   Mon, 12 Apr 2021 13:31:58 +0200
From:   Jan Kara <jack@...e.cz>
To:     Hao Sun <sunhao.th@...il.com>
Cc:     Jan Kara <jack@...e.cz>, linux-kernel@...r.kernel.org,
        jack@...e.com, linux-ext4@...r.kernel.org, tytso@....edu
Subject: Re: More KCSAN data-race Reports

On Mon 12-04-21 18:42:58, Hao Sun wrote:
> Jan Kara <jack@...e.cz> 于2021年4月12日周一 下午5:02写道：
> >
> > Hello,
> >
> > On Sun 11-04-21 11:42:05, Hao Sun wrote:
> > > Since the last KCSAN report[1], I found two more KCSAN reports that
> > > Syzbot had not reported.
> > > Not sure if they are valid bugs, I hope the stack information in
> > > reports can help you locate the problem.
> > > Kernel config can be found in the attachment.
> >
> > Do we have symbolic decoding of the traces below? Because involved
> > functions are really big so it's difficult to guess what KCSAN is
> > complaining about... At least I wasn't able to guess it after looking into
> > the stacktraces for a while.
> >
> Sorry, the log processing module of Fuzzer still has some logic bugs,
> only some of the symbolized reports are stored in the disk.
> Interestingly, however, the read-write end that causes data racing in
> both reports are in the same location (fs/jbd2/commit.c:443), and this
> information should help locate the problem.
> 
> Partial symbolized report 1:
> ==================================================================
> BUG: KCSAN: data-race in ext4_mark_iloc_dirty / jbd2_journal_commit_transaction
> read-write to 0xffff88804451d800 of 8 bytes by task 4821 on cpu 1:
>  jbd2_journal_commit_transaction+0x222/0x3200 fs/jbd2/commit.c:443
>  kjournald2+0x253/0x470 fs/jbd2/journal.c:213
>  kthread+0x1f0/0x220 kernel/kthread.c:292
>  ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:294

OK, that is:
	journal->j_flags |= JBD2_FULL_COMMIT_ONGOING;

So likely this is a complaint about j_flags update vs j_flags check race
(we check for JBD2_ABORT flag) all around the code.

So again this is harmless unless the compiler plays some devilish tricks
and doesn't store bogus intermediate values in j_flags during RMW
operations. Not sure how to deal with this one. Just putting data_race()
here doesn't seem right - if the compiler does something unexpected, we are
indeed in trouble. Maybe using bitops for j_flags would be beneficial for
other reasons as well as silencing KCSAN. But it needs more thought.

								Honza
-- 
Jan Kara <jack@...e.com>
SUSE Labs, CR