mbox series

[v2,00/12] block/export: vhost-user-blk server tests and input validation

Message ID 20201207172030.251905-1-stefanha@redhat.com
Headers show
Series block/export: vhost-user-blk server tests and input validation | expand

Message

Stefan Hajnoczi Dec. 7, 2020, 5:20 p.m. UTC
v2:
 * Add abrt handler that terminates qemu-storage-daemon to
   vhost-user-blk-test. No more orphaned processes on test failure. [Peter]
 * Fix sector number calculation in vhost-user-blk-server.c
 * Introduce VIRTIO_BLK_SECTOR_BITS/SIZE to make code clearer [Max]
 * Fix vhost-user-blk-server.c blk_size double byteswap
 * Fix vhost-user-blk blkcfg->num_queues endianness [Peter]
 * Squashed cleanups into Coiby vhost-user-blk-test commit so the code is
   easier to review

The vhost-user-blk server test was already in Michael Tsirkin's recent vhost
pull request, but was dropped because it exposed vhost-user regressions
(b7c1bd9d7848 and the Based-on tag below). Now that the vhost-user regressions
are fixed we can re-introduce the test case.

This series adds missing input validation that led to a Coverity report. The
virtio-blk read, write, discard, and write zeroes commands need to check
sector/byte ranges and other inputs. This solves the issue Peter Maydell raised
in "[PATCH for-5.2] block/export/vhost-user-blk-server.c: Avoid potential
integer overflow".

Merging just the input validation patches would be possible too, but I prefer
to merge the corresponding tests so the code is exercised by the CI.

Coiby Xu (1):
  test: new qTest case to test the vhost-user-blk-server

Stefan Hajnoczi (11):
  vhost-user-blk: fix blkcfg->num_queues endianness
  libqtest: add qtest_socket_server()
  libqtest: add qtest_kill_qemu()
  libqtest: add qtest_remove_abrt_handler()
  tests/qtest: add multi-queue test case to vhost-user-blk-test
  block/export: fix blk_size double byteswap
  block/export: use VIRTIO_BLK_SECTOR_BITS
  block/export: fix vhost-user-blk export sector number calculation
  block/export: port virtio-blk discard/write zeroes input validation
  vhost-user-blk-test: test discard/write zeroes invalid inputs
  block/export: port virtio-blk read/write range check

 MAINTAINERS                          |   2 +
 tests/qtest/libqos/libqtest.h        |  37 +
 tests/qtest/libqos/vhost-user-blk.h  |  48 ++
 block/export/vhost-user-blk-server.c | 150 +++-
 hw/block/vhost-user-blk.c            |   7 +-
 tests/qtest/libqos/vhost-user-blk.c  | 130 ++++
 tests/qtest/libqtest.c               |  82 ++-
 tests/qtest/vhost-user-blk-test.c    | 983 +++++++++++++++++++++++++++
 tests/qtest/libqos/meson.build       |   1 +
 tests/qtest/meson.build              |   4 +
 10 files changed, 1385 insertions(+), 59 deletions(-)
 create mode 100644 tests/qtest/libqos/vhost-user-blk.h
 create mode 100644 tests/qtest/libqos/vhost-user-blk.c
 create mode 100644 tests/qtest/vhost-user-blk-test.c

Comments

Kevin Wolf Feb. 15, 2021, 10:41 a.m. UTC | #1
Am 07.12.2020 um 18:20 hat Stefan Hajnoczi geschrieben:
> v2:
>  * Add abrt handler that terminates qemu-storage-daemon to
>    vhost-user-blk-test. No more orphaned processes on test failure. [Peter]
>  * Fix sector number calculation in vhost-user-blk-server.c
>  * Introduce VIRTIO_BLK_SECTOR_BITS/SIZE to make code clearer [Max]
>  * Fix vhost-user-blk-server.c blk_size double byteswap
>  * Fix vhost-user-blk blkcfg->num_queues endianness [Peter]
>  * Squashed cleanups into Coiby vhost-user-blk-test commit so the code is
>    easier to review
> 
> The vhost-user-blk server test was already in Michael Tsirkin's recent vhost
> pull request, but was dropped because it exposed vhost-user regressions
> (b7c1bd9d7848 and the Based-on tag below). Now that the vhost-user regressions
> are fixed we can re-introduce the test case.
> 
> This series adds missing input validation that led to a Coverity report. The
> virtio-blk read, write, discard, and write zeroes commands need to check
> sector/byte ranges and other inputs. This solves the issue Peter Maydell raised
> in "[PATCH for-5.2] block/export/vhost-user-blk-server.c: Avoid potential
> integer overflow".
> 
> Merging just the input validation patches would be possible too, but I prefer
> to merge the corresponding tests so the code is exercised by the CI.

Is this series still open? I don't see it in master.

Kevin
Peter Maydell Feb. 19, 2021, 10:38 p.m. UTC | #2
On Mon, 15 Feb 2021 at 10:41, Kevin Wolf <kwolf@redhat.com> wrote:
>
> Am 07.12.2020 um 18:20 hat Stefan Hajnoczi geschrieben:
> > v2:
> >  * Add abrt handler that terminates qemu-storage-daemon to
> >    vhost-user-blk-test. No more orphaned processes on test failure. [Peter]
> >  * Fix sector number calculation in vhost-user-blk-server.c
> >  * Introduce VIRTIO_BLK_SECTOR_BITS/SIZE to make code clearer [Max]
> >  * Fix vhost-user-blk-server.c blk_size double byteswap
> >  * Fix vhost-user-blk blkcfg->num_queues endianness [Peter]
> >  * Squashed cleanups into Coiby vhost-user-blk-test commit so the code is
> >    easier to review
> >
> > The vhost-user-blk server test was already in Michael Tsirkin's recent vhost
> > pull request, but was dropped because it exposed vhost-user regressions
> > (b7c1bd9d7848 and the Based-on tag below). Now that the vhost-user regressions
> > are fixed we can re-introduce the test case.
> >
> > This series adds missing input validation that led to a Coverity report. The
> > virtio-blk read, write, discard, and write zeroes commands need to check
> > sector/byte ranges and other inputs. This solves the issue Peter Maydell raised
> > in "[PATCH for-5.2] block/export/vhost-user-blk-server.c: Avoid potential
> > integer overflow".
> >
> > Merging just the input validation patches would be possible too, but I prefer
> > to merge the corresponding tests so the code is exercised by the CI.
>
> Is this series still open? I don't see it in master.

The Coverity issue is still unfixed, at any rate...

-- PMM
Philippe Mathieu-Daudé Feb. 23, 2021, 11:06 a.m. UTC | #3
On 2/19/21 11:38 PM, Peter Maydell wrote:
> On Mon, 15 Feb 2021 at 10:41, Kevin Wolf <kwolf@redhat.com> wrote:
>>
>> Am 07.12.2020 um 18:20 hat Stefan Hajnoczi geschrieben:
>>> v2:
>>>  * Add abrt handler that terminates qemu-storage-daemon to
>>>    vhost-user-blk-test. No more orphaned processes on test failure. [Peter]
>>>  * Fix sector number calculation in vhost-user-blk-server.c
>>>  * Introduce VIRTIO_BLK_SECTOR_BITS/SIZE to make code clearer [Max]
>>>  * Fix vhost-user-blk-server.c blk_size double byteswap
>>>  * Fix vhost-user-blk blkcfg->num_queues endianness [Peter]
>>>  * Squashed cleanups into Coiby vhost-user-blk-test commit so the code is
>>>    easier to review
>>>
>>> The vhost-user-blk server test was already in Michael Tsirkin's recent vhost
>>> pull request, but was dropped because it exposed vhost-user regressions
>>> (b7c1bd9d7848 and the Based-on tag below). Now that the vhost-user regressions
>>> are fixed we can re-introduce the test case.
>>>
>>> This series adds missing input validation that led to a Coverity report. The
>>> virtio-blk read, write, discard, and write zeroes commands need to check
>>> sector/byte ranges and other inputs. This solves the issue Peter Maydell raised
>>> in "[PATCH for-5.2] block/export/vhost-user-blk-server.c: Avoid potential
>>> integer overflow".
>>>
>>> Merging just the input validation patches would be possible too, but I prefer
>>> to merge the corresponding tests so the code is exercised by the CI.
>>
>> Is this series still open? I don't see it in master.
> 
> The Coverity issue is still unfixed, at any rate...

Copying Coverity report here:

CID 1435956 Unintentional integer overflow

In vu_blk_discard_write_zeroes: An integer overflow occurs, with the
result converted to a wider integer type (CWE-190)

 61 static int coroutine_fn
 62 vu_blk_discard_write_zeroes(BlockBackend *blk, struct iovec *iov,
 63                             uint32_t iovcnt, uint32_t type)
 64 {
 65     struct virtio_blk_discard_write_zeroes desc;
 66     ssize_t size = iov_to_buf(iov, iovcnt, 0, &desc, sizeof(desc));
 67     if (unlikely(size != sizeof(desc))) {
 68         error_report("Invalid size %zd, expect %zu", size,
sizeof(desc));
 69         return -EINVAL;
 70     }
 71
 72     uint64_t range[2] = { le64_to_cpu(desc.sector) << 9,

CID 1435956 (#1 of 1): Unintentional integer overflow
(OVERFLOW_BEFORE_WIDEN)
overflow_before_widen: Potentially overflowing expression
le32_to_cpu(desc.num_sectors) << 9 with type uint32_t (32 bits,
unsigned) is evaluated using 32-bit arithmetic, and then used in a
context that expects an expression of type uint64_t (64 bits, unsigned).

 73                           le32_to_cpu(desc.num_sectors) << 9 };
Peter Maydell March 10, 2021, 3:51 p.m. UTC | #4
On Fri, 19 Feb 2021 at 22:38, Peter Maydell <peter.maydell@linaro.org> wrote:
>
> On Mon, 15 Feb 2021 at 10:41, Kevin Wolf <kwolf@redhat.com> wrote:
> >
> > Am 07.12.2020 um 18:20 hat Stefan Hajnoczi geschrieben:
> > > v2:
> > >  * Add abrt handler that terminates qemu-storage-daemon to
> > >    vhost-user-blk-test. No more orphaned processes on test failure. [Peter]
> > >  * Fix sector number calculation in vhost-user-blk-server.c
> > >  * Introduce VIRTIO_BLK_SECTOR_BITS/SIZE to make code clearer [Max]
> > >  * Fix vhost-user-blk-server.c blk_size double byteswap
> > >  * Fix vhost-user-blk blkcfg->num_queues endianness [Peter]
> > >  * Squashed cleanups into Coiby vhost-user-blk-test commit so the code is
> > >    easier to review
> > >
> > > The vhost-user-blk server test was already in Michael Tsirkin's recent vhost
> > > pull request, but was dropped because it exposed vhost-user regressions
> > > (b7c1bd9d7848 and the Based-on tag below). Now that the vhost-user regressions
> > > are fixed we can re-introduce the test case.
> > >
> > > This series adds missing input validation that led to a Coverity report. The
> > > virtio-blk read, write, discard, and write zeroes commands need to check
> > > sector/byte ranges and other inputs. This solves the issue Peter Maydell raised
> > > in "[PATCH for-5.2] block/export/vhost-user-blk-server.c: Avoid potential
> > > integer overflow".
> > >
> > > Merging just the input validation patches would be possible too, but I prefer
> > > to merge the corresponding tests so the code is exercised by the CI.
> >
> > Is this series still open? I don't see it in master.
>
> The Coverity issue is still unfixed, at any rate...

Ping^2 ! I'd like to get us down to no outstanding Coverity issues for the
6.0 release, and this (CID 1435956) is one of the handful still unfixed.

thanks
-- PMM
Kevin Wolf March 10, 2021, 4:26 p.m. UTC | #5
Am 10.03.2021 um 16:51 hat Peter Maydell geschrieben:
> On Fri, 19 Feb 2021 at 22:38, Peter Maydell <peter.maydell@linaro.org> wrote:
> >
> > On Mon, 15 Feb 2021 at 10:41, Kevin Wolf <kwolf@redhat.com> wrote:
> > >
> > > Am 07.12.2020 um 18:20 hat Stefan Hajnoczi geschrieben:
> > > > v2:
> > > >  * Add abrt handler that terminates qemu-storage-daemon to
> > > >    vhost-user-blk-test. No more orphaned processes on test failure. [Peter]
> > > >  * Fix sector number calculation in vhost-user-blk-server.c
> > > >  * Introduce VIRTIO_BLK_SECTOR_BITS/SIZE to make code clearer [Max]
> > > >  * Fix vhost-user-blk-server.c blk_size double byteswap
> > > >  * Fix vhost-user-blk blkcfg->num_queues endianness [Peter]
> > > >  * Squashed cleanups into Coiby vhost-user-blk-test commit so the code is
> > > >    easier to review
> > > >
> > > > The vhost-user-blk server test was already in Michael Tsirkin's recent vhost
> > > > pull request, but was dropped because it exposed vhost-user regressions
> > > > (b7c1bd9d7848 and the Based-on tag below). Now that the vhost-user regressions
> > > > are fixed we can re-introduce the test case.
> > > >
> > > > This series adds missing input validation that led to a Coverity report. The
> > > > virtio-blk read, write, discard, and write zeroes commands need to check
> > > > sector/byte ranges and other inputs. This solves the issue Peter Maydell raised
> > > > in "[PATCH for-5.2] block/export/vhost-user-blk-server.c: Avoid potential
> > > > integer overflow".
> > > >
> > > > Merging just the input validation patches would be possible too, but I prefer
> > > > to merge the corresponding tests so the code is exercised by the CI.
> > >
> > > Is this series still open? I don't see it in master.
> >
> > The Coverity issue is still unfixed, at any rate...
> 
> Ping^2 ! I'd like to get us down to no outstanding Coverity issues for the
> 6.0 release, and this (CID 1435956) is one of the handful still unfixed.

You pulled a newer version of this series (minus the tests that caused CI
failures on some older OSes) earlier today, so I assume this is fixed
now.

Kevin