Skip to contentSkip to conversation

qcow2: Forbid discard in qcow2 v2 images with backing files

PATCH Superseded Alberto Garcia 2 participants 6 messages Mar 27, 2020 Block layer

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 qcow2: Forbid discard in qcow2 v2 images with backing files

A discard request deallocates the selected clusters so they read back as zeroes. This is done by clearing the cluster offset field and setting QCOW_OFLAG_ZERO in the L2 entry.

This flag is however only supported when qcow_version >= 3. In older images the cluster is simply deallocated, exposing any possible stale data from the backing file.

Since discard is an advisory operation it's safer to simply forbid it in this scenario.

Note that we are adding this check to qcow2_co_pdiscard() and not to qcow2_cluster_discard() or discard_in_l2_slice() because the last two are also used by qcow2_snapshot_create() to discard the clusters used by the VM state. In this case there's no risk of exposing stale data to the guest and we really want that the clusters are always discarded.

---
v3:
- Rebase and change iotest number
- Show output of qemu-img map in iotest 290 [Kevin]
- Use the l2_offset and rb_offset variables in iotest 060

v2:

- Don't create the image with compat=0.10 in iotest 060 [Max]
- Use $TEST_IMG.base for the backing image name in iotest 289 [Max]
- Add list of unsupported options to iotest 289 [Max]

 block/qcow2.c              |  6 +++
 tests/qemu-iotests/060     | 12 ++---
 tests/qemu-iotests/060.out |  2 -
 tests/qemu-iotests/290     | 94 ++++++++++++++++++++++++++++++++++++++
 tests/qemu-iotests/290.out | 57 +++++++++++++++++++++++
 tests/qemu-iotests/group   |  1 +
 6 files changed, 163 insertions(+), 9 deletions(-)
 create mode 100755 tests/qemu-iotests/290
 create mode 100644 tests/qemu-iotests/290.out
block/qcow2.c +6 -0
index 2bb536b014..e8cbcc1ec1 100644
@@ -3784,6 +3784,12 @@ static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
     int ret;
     BDRVQcow2State *s = bs->opaque;
 
+    /* If the image does not support QCOW_OFLAG_ZERO then discarding
+     * clusters could expose stale data from the backing file. */
+    if (s->qcow_version < 3 && bs->backing) {
+        return -ENOTSUP;
+    }
+
     if (!QEMU_IS_ALIGNED(offset | bytes, s->cluster_size)) {
         assert(bytes < s->cluster_size);
         /* Ignore partial clusters, except for the special case of the
tests/qemu-iotests/060 +5 -7
index 043f12904a..32c0ecce9e 100755
@@ -160,18 +160,16 @@ TEST_IMG=$BACKING_IMG _make_test_img 1G
 
 $QEMU_IO -c 'write 0k 64k' "$BACKING_IMG" | _filter_qemu_io
 
-# compat=0.10 is required in order to make the following discard actually
-# unallocate the sector rather than make it a zero sector - we want COW, after
-# all.
-_make_test_img -o 'compat=0.10' -b "$BACKING_IMG" 1G
+_make_test_img -b "$BACKING_IMG" 1G
 # Write two clusters, the second one enforces creation of an L2 table after
 # the first data cluster.
 $QEMU_IO -c 'write 0k 64k' -c 'write 512M 64k' "$TEST_IMG" | _filter_qemu_io
-# Discard the first cluster. This cluster will soon enough be reallocated and
+# Free the first cluster. This cluster will soon enough be reallocated and
 # used for COW.
-$QEMU_IO -c 'discard 0k 64k' "$TEST_IMG" | _filter_qemu_io
+poke_file "$TEST_IMG" "$l2_offset" "\x00\x00\x00\x00\x00\x00\x00\x00"
+poke_file "$TEST_IMG" "$(($rb_offset+10))" "\x00\x00"
 # Now, corrupt the image by marking the second L2 table cluster as free.
-poke_file "$TEST_IMG" '131084' "\x00\x00" # 0x2000c
+poke_file "$TEST_IMG" "$(($rb_offset+12))" "\x00\x00"
 # Start a write operation requiring COW on the image stopping it right before
 # doing the read; then, trigger the corruption prevention by writing anything to
 # any unallocated cluster, leading to an attempt to overwrite the second L2
tests/qemu-iotests/060.out +0 -2
index d27692a33c..09caaea865 100644
@@ -105,8 +105,6 @@ wrote 65536/65536 bytes at offset 0
 64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
 wrote 65536/65536 bytes at offset 536870912
 64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
-discard 65536/65536 bytes at offset 0
-64 KiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
 qcow2: Marking image as corrupt: Preventing invalid write on metadata (overlaps with active L2 table); further corruption events will be suppressed
 blkdebug: Suspended request '0'
 write failed: Input/output error
tests/qemu-iotests/290 new file +94 -0

94 lines not on this page · raw message

tests/qemu-iotests/290.out new file +57 -0

57 lines not on this page · raw message

tests/qemu-iotests/group +1 -0
index 79c6dfc85d..435dccd5af 100644
@@ -296,3 +296,4 @@
 286 rw quick
 288 quick
 289 rw quick
+290 rw auto quick
Signature · 1 line
2.20.1
Reviewed-by on v3 1/1
#1 lore raw
Quoted 28 lines · Alberto Garcia

A discard request deallocates the selected clusters so they read back as zeroes. This is done by clearing the cluster offset field and setting QCOW_OFLAG_ZERO in the L2 entry.

This flag is however only supported when qcow_version >= 3. In older images the cluster is simply deallocated, exposing any possible stale data from the backing file.

Since discard is an advisory operation it's safer to simply forbid it in this scenario.

Note that we are adding this check to qcow2_co_pdiscard() and not to qcow2_cluster_discard() or discard_in_l2_slice() because the last two are also used by qcow2_snapshot_create() to discard the clusters used by the VM state. In this case there's no risk of exposing stale data to the guest and we really want that the clusters are always discarded.

---

+++ b/tests/qemu-iotests/290

tests/qemu-iotests/290 new file +5 -0
+
+echo
+echo "### Test 'qemu-io -c discard' on a QCOW2 image without a backing file"
+echo
+for qcow2_compat in 0.10 1.1; do

tests/qemu-iotests/290 line 58

Quoted 10 lines
tests/qemu-iotests/290 new file +10 -0
+    echo "# Create an image with compat=$qcow2_compat without a backing file"
+    _make_test_img -o "compat=$qcow2_compat" 128k
+
+    echo "# Fill all clusters with data and then discard them"
+    $QEMU_IO -c 'write -P 0x01 0 128k' "$TEST_IMG" | _filter_qemu_io
+    $QEMU_IO -c 'discard 0 128k' "$TEST_IMG" | _filter_qemu_io
+
+    echo "# Read the data from the discarded clusters"
+    $QEMU_IO -c 'read -P 0x00 0 128k' "$TEST_IMG" | _filter_qemu_io
+done

Should this loop also inspect qemu-img map output?

Quoted 21 lines
tests/qemu-iotests/290 new file +21 -0
+
+echo
+echo "### Test 'qemu-io -c discard' on a QCOW2 image with a backing file"
+echo
+
+echo "# Create a backing image and fill it with data"
+BACKING_IMG="$TEST_IMG.base"
+TEST_IMG="$BACKING_IMG" _make_test_img 128k
+$QEMU_IO -c 'write -P 0xff 0 128k' "$BACKING_IMG" | _filter_qemu_io
+
+for qcow2_compat in 0.10 1.1; do
+    echo "# Create an image with compat=$qcow2_compat and a backing file"
+    _make_test_img -o "compat=$qcow2_compat" -b "$BACKING_IMG"
+
+    echo "# Fill all clusters with data and then discard them"
+    $QEMU_IO -c 'write -P 0x01 0 128k' "$TEST_IMG" | _filter_qemu_io
+    $QEMU_IO -c 'discard 0 128k' "$TEST_IMG" | _filter_qemu_io
+
+    echo "# Read the data from the discarded clusters"
+    if [ "$qcow2_compat" = "1.1" ]; then
+        # In qcow2 v3 clusters are zeroed (with QCOW_OFLAG_ZERO)

tests/qemu-iotests/290 line 89

Quoted 10 lines
tests/qemu-iotests/290 new file +10 -0
+        $QEMU_IO -c 'read -P 0x00 0 128k' "$TEST_IMG" | _filter_qemu_io
+    else
+        # In qcow2 v2 if there's a backing image we cannot zero the clusters
+        # without exposing the backing file data so discard does nothing
+        $QEMU_IO -c 'read -P 0x01 0 128k' "$TEST_IMG" | _filter_qemu_io
+    fi
+
+    echo "# Output of qemu-img map"
+    $QEMU_IMG map "$TEST_IMG" | _filter_testdir
+done

But I agree this was the more interesting one, so we at least have decent coverage of the change itself.

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org
#2 lore raw
Quoted 5 lines · Eric Blake
Quoted 5 lines
tests/qemu-iotests/290 new file +5 -0
+for qcow2_compat in 0.10 1.1; do
+    echo "# Create an image with compat=$qcow2_compat without a backing file"
+    _make_test_img -o "compat=$qcow2_compat" 128k
+
+    echo "# Fill all clusters with data and then discard them"
Quoted 8 lines

tests/qemu-iotests/290 line 58

Quoted 6 lines
tests/qemu-iotests/290 new file +6 -0
+    $QEMU_IO -c 'write -P 0x01 0 128k' "$TEST_IMG" | _filter_qemu_io
+    $QEMU_IO -c 'discard 0 128k' "$TEST_IMG" | _filter_qemu_io
+
+    echo "# Read the data from the discarded clusters"
+    $QEMU_IO -c 'read -P 0x00 0 128k' "$TEST_IMG" | _filter_qemu_io
+done

Should this loop also inspect qemu-img map output?

I guess we can, although here the image is completely unallocated in both cases.

Berto

#3 lore raw
Quoted 15 lines · Alberto Garcia
Quoted 5 lines · Eric Blake
Quoted 5 lines
tests/qemu-iotests/290 new file +5 -0
+for qcow2_compat in 0.10 1.1; do
+    echo "# Create an image with compat=$qcow2_compat without a backing file"
+    _make_test_img -o "compat=$qcow2_compat" 128k
+
+    echo "# Fill all clusters with data and then discard them"
Quoted 8 lines

tests/qemu-iotests/290 line 58

Quoted 6 lines
tests/qemu-iotests/290 new file +6 -0
+    $QEMU_IO -c 'write -P 0x01 0 128k' "$TEST_IMG" | _filter_qemu_io
+    $QEMU_IO -c 'discard 0 128k' "$TEST_IMG" | _filter_qemu_io
+
+    echo "# Read the data from the discarded clusters"
+    $QEMU_IO -c 'read -P 0x00 0 128k' "$TEST_IMG" | _filter_qemu_io
+done

Should this loop also inspect qemu-img map output?

Quoted 2 lines

I guess we can, although here the image is completely unallocated in both cases.

Which shows that even for v2 images, discard DOES do something when there is no backing file (even if it is now a no-op when there is a backing file after this patch).

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org
#4 lore raw
In reply to Alberto Garcia
Quoted 23 lines · Alberto Garcia

A discard request deallocates the selected clusters so they read back as zeroes. This is done by clearing the cluster offset field and setting QCOW_OFLAG_ZERO in the L2 entry.

This flag is however only supported when qcow_version >= 3. In older images the cluster is simply deallocated, exposing any possible stale data from the backing file.

Since discard is an advisory operation it's safer to simply forbid it in this scenario.

Note that we are adding this check to qcow2_co_pdiscard() and not to qcow2_cluster_discard() or discard_in_l2_slice() because the last two are also used by qcow2_snapshot_create() to discard the clusters used by the VM state. In this case there's no risk of exposing stale data to the guest and we really want that the clusters are always discarded.

---

+++ b/block/qcow2.c

block/qcow2.c static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs, · line 3792

Quoted 9 lines
block/qcow2.c +6 -0
@@ -3784,6 +3784,12 @@ static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
      int ret;
      BDRVQcow2State *s = bs->opaque;
 
+    /* If the image does not support QCOW_OFLAG_ZERO then discarding
+     * clusters could expose stale data from the backing file. */
+    if (s->qcow_version < 3 && bs->backing) {
+        return -ENOTSUP;
+    }
+

Hmm. Should we blindly always fail for v2, or can we be a little bit smarter and still discard a cluster in the top layer if the backing layer does not also have it allocated? In other words, is it also worth checking bdrv_is_allocated(), and avoiding the -ENOTSUP failure in the case where the backing chain does not have any allocation of the same range (at which point, discarding the cluster in the top layer will read as zeroes rather than as stale data)?

But that's a minor optimization for an older file format; it's just as easy to tell users that if they want maximum discarding power on their active layer, then they should be using v3 images.

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org
#5 lore raw
Quoted 9 lines · Eric Blake

block/qcow2.c static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs, · line 3791

Quoted 5 lines
block/qcow2.c +5 -0
+    /* If the image does not support QCOW_OFLAG_ZERO then discarding
+     * clusters could expose stale data from the backing file. */
+    if (s->qcow_version < 3 && bs->backing) {
+        return -ENOTSUP;
+    }

Hmm. Should we blindly always fail for v2, or can we be a little bit smarter and still discard a cluster in the top layer if the backing layer does not also have it allocated?

Not sure if that's worth it. I only wanted to fix what looks like a potential security bug so I prefer to keep it simple. qcow2 v3 has been out for many years already.

Berto