linux-kernel - Re: [PATCH V3] bcachefs: Add journal v2 entry nr value check

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: <bij226lxay4fr3bpxpu6kcgdynxrq4rnjhxvpsjgzukqjf7w4t@4ntovo66d32t>
Date: Tue, 20 Aug 2024 23:19:54 -0400
From: Kent Overstreet <kent.overstreet@...ux.dev>
To: Lizhi Xu <lizhi.xu@...driver.com>
Cc: linux-bcachefs@...r.kernel.org, linux-kernel@...r.kernel.org, 
	syzbot+47ecc948aadfb2ab3efc@...kaller.appspotmail.com, syzkaller-bugs@...glegroups.com
Subject: Re: [PATCH V3] bcachefs: Add journal v2 entry nr value check

On Tue, Aug 20, 2024 at 11:16:55PM GMT, Kent Overstreet wrote:
> On Wed, Aug 21, 2024 at 11:10:00AM GMT, Lizhi Xu wrote:
> > On Tue, 20 Aug 2024 23:00:05 -0400, Kent Overstreet wrote:
> > > > When the nr value of a signle entry or their sum overflows, it will
> > > > cause the value of ja->nr to be incorrect, this will result in the
> > > > allocated memory to ja->buckets being too small, leading to out of
> > > > bounds access in bch2_dev_journal_init.
> > > > 
> > > > Reported-by: syzbot+47ecc948aadfb2ab3efc@...kaller.appspotmail.com
> > > > Signed-off-by: Lizhi Xu <lizhi.xu@...driver.com>
> > > > ---
> > > >  fs/bcachefs/journal_sb.c | 16 +++++++++++++++-
> > > >  1 file changed, 15 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/fs/bcachefs/journal_sb.c b/fs/bcachefs/journal_sb.c
> > > > index db80e506e3ab..230ed99130e4 100644
> > > > --- a/fs/bcachefs/journal_sb.c
> > > > +++ b/fs/bcachefs/journal_sb.c
> > > > @@ -107,6 +107,7 @@ static int bch2_sb_journal_v2_validate(struct bch_sb *sb, struct bch_sb_field *f
> > > >  	unsigned nr;
> > > >  	unsigned i;
> > > >  	struct u64_range *b;
> > > > +	u64 total_nr = 0, entry_nr;
> > > >  
> > > >  	nr = bch2_sb_field_journal_v2_nr_entries(journal);
> > > >  	if (!nr)
> > > > @@ -117,8 +118,21 @@ static int bch2_sb_journal_v2_validate(struct bch_sb *sb, struct bch_sb_field *f
> > > >  		return -BCH_ERR_ENOMEM_sb_journal_v2_validate;
> > > >  
> > > >  	for (i = 0; i < nr; i++) {
> > > > +		entry_nr = le64_to_cpu(journal->d[i].nr);
> > > > +		if (entry_nr > UINT_MAX) {
> > > > +			prt_printf(err, "Journal v2 entry d[%u] nr %llu overflow\n",
> > > > +				i, entry_nr);
> > > > +			goto err;
> > > > +		}
> > > 
> > > This check is unnecessary; we know the sum can't overflow a u64 because
> > > we're also checking that the entries are nonoverlapping.
> > You didn't read my previous email carefully.
> > In this issue, journal->d[0] is 7, journal->d[1] is 18446744073709551615,
> > so the sum of their u64 type values will definitely overflow.
> 
> It doesn't matter. We're already checking that the entries are
> nonoverlapping, and within the range of [1, nbuckets), so total_nr can't
> overflow nbuckets, much less an s64 (not that that matters).

The check that's missing is that start + nr doesn't overflow, when we
convert to u64_ranges.