Re: [PATCH 2/6] udf: Simplify handling of Volume Descriptor Pointers

From: Jan Kara <jack@suse.cz>
To: "Pali Rohár" <pali.rohar@gmail.com>
Cc: Jan Kara <jack@suse.cz>, linux-fsdevel@vger.kernel.org
Subject: Re: [PATCH 2/6] udf: Simplify handling of Volume Descriptor Pointers
Date: Thu, 15 Feb 2018 09:43:12 +0100	[thread overview]
Message-ID: <20180215084312.u4icv4ecdiflwpmy@quack2.suse.cz> (raw)
In-Reply-To: <20180214172603.z4sl5qhrxxsscltu@pali>

On Wed 14-02-18 18:26:03, Pali Rohï¿½r wrote:
> On Wednesday 14 February 2018 11:28:46 Jan Kara wrote:
> > According to ECMA-167 3/8.4.2 Volume Descriptor Pointer is terminating
> > current extent of Volume Descriptor Sequence. Also according to ECMA-167
> > 3/8.4.3 Volume Descriptor Sequence Number is not significant for Volume
> > Descriptor Pointers. Simplify the handling of Volume Descriptor Pointers
> > to take this into account.
> > 
> > Signed-off-by: Jan Kara <jack@suse.cz>
> > ---
> >  fs/udf/super.c | 41 ++++++++++++++++-------------------------
> >  1 file changed, 16 insertions(+), 25 deletions(-)
> > 
> > diff --git a/fs/udf/super.c b/fs/udf/super.c
> > index 5c5d5fd513cc..f80b97173acd 100644
> > --- a/fs/udf/super.c
> > +++ b/fs/udf/super.c
> > @@ -1615,7 +1615,6 @@ static noinline int udf_process_sequence(
> >  	bool done = false;
> >  	uint32_t vdsn;
> >  	uint16_t ident;
> > -	long next_s = 0, next_e = 0;
> >  	int ret;
> >  	unsigned int indirections = 0;
> >  
> > @@ -1647,19 +1646,22 @@ static noinline int udf_process_sequence(
> >  			}
> >  			break;
> >  		case TAG_IDENT_VDP: /* ISO 13346 3/10.3 */
> > -			curr = &vds[VDS_POS_VOL_DESC_PTR];
> > -			if (vdsn >= curr->volDescSeqNum) {
> > -				curr->volDescSeqNum = vdsn;
> > -				curr->block = block;
> > -
> > -				vdp = (struct volDescPtr *)bh->b_data;
> > -				next_s = le32_to_cpu(
> > -					vdp->nextVolDescSeqExt.extLocation);
> > -				next_e = le32_to_cpu(
> > -					vdp->nextVolDescSeqExt.extLength);
> > -				next_e = next_e >> sb->s_blocksize_bits;
> > -				next_e += next_s - 1;
> > +			if (++indirections > UDF_MAX_TD_NESTING) {
> > +				udf_err(sb, "too many Volume Descriptor "
> > +					"Pointers (max %u supported)\n",
> > +					UDF_MAX_TD_NESTING);
> > +				brelse(bh);
> > +				return -EIO;
> >  			}
> > +
> > +			vdp = (struct volDescPtr *)bh->b_data;
> > +			block = le32_to_cpu(vdp->nextVolDescSeqExt.extLocation);
> 
> Another pathological case: disc with two (or more) VDP descriptors and
> each points to another in cycle.
> 
> Seems that this would not cause infinite loop due to
> UDF_MAX_TD_NESTING, but probably can cause some troubles.

Yes. Such disk is invalid so our only goal is not to crash / livelock the
kernel and UDF_MAX_TD_NESTING protection is enough for that.

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