From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:44791) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1XPuAq-000678-6W for qemu-devel@nongnu.org; Fri, 05 Sep 2014 10:07:41 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1XPuAl-0004jL-3s for qemu-devel@nongnu.org; Fri, 05 Sep 2014 10:07:36 -0400 Received: from mx1.redhat.com ([209.132.183.28]:14322) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1XPuAk-0004jB-RR for qemu-devel@nongnu.org; Fri, 05 Sep 2014 10:07:31 -0400 Received: from int-mx14.intmail.prod.int.phx2.redhat.com (int-mx14.intmail.prod.int.phx2.redhat.com [10.5.11.27]) by mx1.redhat.com (8.14.4/8.14.4) with ESMTP id s85E7TNg019879 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-GCM-SHA384 bits=256 verify=FAIL) for ; Fri, 5 Sep 2014 10:07:30 -0400 From: Max Reitz Date: Fri, 5 Sep 2014 16:07:14 +0200 Message-Id: <1409926039-29044-1-git-send-email-mreitz@redhat.com> Subject: [Qemu-devel] [PATCH v2 0/5] qcow2: Check L1/L2/reftable entries for alignment List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: qemu-devel@nongnu.org Cc: Kevin Wolf , Stefan Hajnoczi , Max Reitz The image fuzzer from Maria exposed a lot of assertions which might fail in qemu when fed with a broken qcow2 image. Some of them are related to qemu trusting the offsets given in the L1, L2 and refcount tables to always be properly aligned on cluster boundaries (e.g. https://bugs.launchpad.net/qemu/+bug/1354529). This series fixes this by verifying (hopefully) all data read from L1, L2 and refcount tables accordingly; if the offsets are not aligned on cluster boundaries, an error message is emitted and the image is marked corrupt unless it has been opened read-only. v2: - Added patch 1 which adds a new field to the BLOCK_IMAGE_CORRUPTED. I know this might be incompatible to existing users of that field, so I did an investigative search (I googled) but did not find any. So I'm confident this won't break anything. - Patch 2: - Suppress corruption events and messages after the first one; however, a fatal corruption should still be signaled even after a non-fatal one has already occured [Kevin] - Also, use the new "fatal" field - Patch 3: - Fix test 060 right in this patch (squashed the old patch 3 into patch 2, the result is now this patch 3) [Kevin] - Set the "fatal" field to true - Patch 4: - Emit more debugging information (like table indices etc.) [Kevin] - Generally set the "fatal" field to true - In case a cluster with an unaligned offset should be freed, call the corruption signalling function just like everywhere else; but set "fatal" to false [Kevin+Eric] - Patch 5: Added [Kevin] git-backport-diff against v1 (although not very useful): Key: [----] : patches are identical [####] : number of functional differences between upstream/downstream patch [down] : patch is downstream-only The flags [FC] indicate (F)unctional and (C)ontextual differences, respectively 001/5:[down] 'qapi/block: Add "fatal" to BLOCK_IMAGE_CORRUPTED' 002/5:[0045] [FC] 'qcow2: Add qcow2_signal_corruption()' 003/5:[0015] [FC] 'qcow2: Use qcow2_signal_corruption() for overlaps' 004/5:[0068] [FC] 'qcow2: Check L1/L2/reftable entries for alignment' 005/5:[down] 'iotests: Add more tests for qcow2 corruption' Max Reitz (5): qapi/block: Add "fatal" to BLOCK_IMAGE_CORRUPTED qcow2: Add qcow2_signal_corruption() qcow2: Use qcow2_signal_corruption() for overlaps qcow2: Check L1/L2/reftable entries for alignment iotests: Add more tests for qcow2 corruption block/qcow2-cluster.c | 43 ++++++++++++++++++++++++++--- block/qcow2-refcount.c | 67 +++++++++++++++++++++++++++++++--------------- block/qcow2.c | 48 +++++++++++++++++++++++++++++++++ block/qcow2.h | 5 ++++ qapi/block-core.json | 9 +++++-- tests/qemu-iotests/060 | 56 ++++++++++++++++++++++++++++++++++++-- tests/qemu-iotests/060.out | 61 +++++++++++++++++++++++++++++++++++++---- 7 files changed, 255 insertions(+), 34 deletions(-) -- 2.1.0