tests/qemu-iotests/271

#!/usr/bin/env bash
# group: rw auto
#
# Test qcow2 images with extended L2 entries
#
# Copyright (C) 2019-2020 Igalia, S.L.
# Author: Alberto Garcia <berto@igalia.com>
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program.  If not, see <http://www.gnu.org/licenses/>.
#

# creator
owner=berto@igalia.com

seq="$(basename $0)"
echo "QA output created by $seq"

here="$PWD"
status=1	# failure is the default!

_cleanup()
{
        _cleanup_test_img
        rm -f "$TEST_IMG.raw"
}
trap "_cleanup; exit \$status" 0 1 2 3 15

# get standard environment, filters and checks
. ./common.rc
. ./common.filter

_supported_fmt qcow2
_supported_proto file nfs
_supported_os Linux
_unsupported_imgopts extended_l2 compat=0.10 cluster_size data_file refcount_bits=1[^0-9]

l2_offset=$((0x40000))

_verify_img()
{
    $QEMU_IMG compare "$TEST_IMG" "$TEST_IMG.raw" | grep -v 'Images are identical'
    $QEMU_IMG check "$TEST_IMG" | _filter_qemu_img_check | \
        grep -v 'No errors were found on the image'
}

# Compare the bitmap of an extended L2 entry against an expected value
_verify_l2_bitmap()
{
    entry_no="$1"            # L2 entry number, starting from 0
    expected_alloc="$alloc"  # Space-separated list of allocated subcluster indexes
    expected_zero="$zero"    # Space-separated list of zero subcluster indexes

    offset=$(($l2_offset + $entry_no * 16))
    entry=$(peek_file_be "$TEST_IMG" $offset 8)
    offset=$(($offset + 8))
    bitmap=$(peek_file_be "$TEST_IMG" $offset 8)

    expected_bitmap=0
    for bit in $expected_alloc; do
        expected_bitmap=$(($expected_bitmap | (1 << $bit)))
    done
    for bit in $expected_zero; do
        expected_bitmap=$(($expected_bitmap | (1 << (32 + $bit))))
    done
    printf -v expected_bitmap "%u" $expected_bitmap # Convert to unsigned

    printf "L2 entry #%d: 0x%016x %016x\n" "$entry_no" "$entry" "$bitmap"
    if [ "$bitmap" != "$expected_bitmap" ]; then
        printf "ERROR: expecting bitmap       0x%016x\n" "$expected_bitmap"
    fi
}

# This should be called as _run_test c=XXX sc=XXX off=XXX len=XXX cmd=XXX
# c:   cluster number (0 if unset)
# sc:  subcluster number inside cluster @c (0 if unset)
# off: offset inside subcluster @sc, in kilobytes (0 if unset)
# len: request length, passed directly to qemu-io (e.g: 256, 4k, 1M, ...)
# cmd: the command to pass to qemu-io, must be one of
#      write    -> write
#      zero     -> write -z
#      unmap    -> write -z -u
#      compress -> write -c
#      discard  -> discard
_run_test()
{
    unset c sc off len cmd
    for var in "$@"; do eval "$var"; done
    case "${cmd:-write}" in
        zero)
            cmd="write -q -z";;
        unmap)
            cmd="write -q -z -u";;
        compress)
            pat=$((${pat:-0} + 1))
            cmd="write -q -c -P ${pat}";;
        write)
            pat=$((${pat:-0} + 1))
            cmd="write -q -P ${pat}";;
        discard)
            cmd="discard -q";;
        *)
            echo "Unknown option $cmd"
            exit 1;;
    esac
    c="${c:-0}"
    sc="${sc:-0}"
    off="${off:-0}"
    offset="$(($c * 64 + $sc * 2 + $off))"
    [ "$offset" != 0 ] && offset="${offset}k"
    cmd="$cmd ${offset} ${len}"
    raw_cmd=$(echo $cmd | sed s/-c//) # Raw images don't support -c
    echo $cmd | sed 's/-P [0-9][0-9]\?/-P PATTERN/'
    $QEMU_IO -c "$cmd" "$TEST_IMG" | _filter_qemu_io
    $QEMU_IO -c "$raw_cmd" -f raw "$TEST_IMG.raw" | _filter_qemu_io
    _verify_img
    _verify_l2_bitmap "$c"
}

_reset_img()
{
    size="$1"
    $QEMU_IMG create -f raw "$TEST_IMG.raw" "$size" | _filter_img_create
    if [ "$use_backing_file" = "yes" ]; then
        $QEMU_IMG create -f raw "$TEST_IMG.base" "$size" | _filter_img_create
        $QEMU_IO -c "write -q -P 0xFF 0 $size" -f raw "$TEST_IMG.base" | _filter_qemu_io
        $QEMU_IO -c "write -q -P 0xFF 0 $size" -f raw "$TEST_IMG.raw" | _filter_qemu_io
        _make_test_img -o extended_l2=on -F raw -b "$TEST_IMG.base" "$size"
    else
        _make_test_img -o extended_l2=on "$size"
    fi
}

############################################################
############################################################
############################################################

# Test that writing to an image with subclusters produces the expected
# results, in images with and without backing files
for use_backing_file in yes no; do
    echo
    echo "### Standard write tests (backing file: $use_backing_file) ###"
    echo
    _reset_img 1M
    ### Write subcluster #0 (beginning of subcluster) ###
    alloc="0"; zero=""
    _run_test sc=0 len=1k

    ### Write subcluster #1 (middle of subcluster) ###
    alloc="0 1"; zero=""
    _run_test sc=1 off=1 len=512

    ### Write subcluster #2 (end of subcluster) ###
    alloc="0 1 2"; zero=""
    _run_test sc=2 off=1 len=1k

    ### Write subcluster #3 (full subcluster) ###
    alloc="0 1 2 3"; zero=""
    _run_test sc=3 len=2k

    ### Write subclusters #4-6 (full subclusters) ###
    alloc="$(seq 0 6)"; zero=""
    _run_test sc=4 len=6k

    ### Write subclusters #7-9 (partial subclusters) ###
    alloc="$(seq 0 9)"; zero=""
    _run_test sc=7 off=1 len=4k

    ### Write subcluster #16 (partial subcluster) ###
    alloc="$(seq 0 9) 16"; zero=""
    _run_test sc=16 len=1k

    ### Write subcluster #31-#33 (cluster overlap) ###
    alloc="$(seq 0 9) 16 31"; zero=""
    _run_test sc=31 off=1 len=4k
    alloc="0 1" ; zero=""
    _verify_l2_bitmap 1

    ### Zero subcluster #1
    alloc="0 $(seq 2 9) 16 31"; zero="1"
    _run_test sc=1 len=2k cmd=zero

    ### Zero cluster #0
    alloc=""; zero="$(seq 0 31)"
    _run_test sc=0 len=64k cmd=zero

    ### Fill cluster #0 with data
    alloc="$(seq 0 31)"; zero=""
    _run_test sc=0 len=64k

    ### Zero and unmap half of cluster #0 (this won't unmap it)
    alloc="$(seq 16 31)"; zero="$(seq 0 15)"
    _run_test sc=0 len=32k cmd=unmap

    ### Zero and unmap cluster #0
    alloc=""; zero="$(seq 0 31)"
    _run_test sc=0 len=64k cmd=unmap

    ### Write subcluster #1 (middle of subcluster)
    alloc="1"; zero="0 $(seq 2 31)"
    _run_test sc=1 off=1 len=512

    ### Fill cluster #0 with data
    alloc="$(seq 0 31)"; zero=""
    _run_test sc=0 len=64k

    ### Discard cluster #0
    alloc=""; zero="$(seq 0 31)"
    _run_test sc=0 len=64k cmd=discard

    ### Write compressed data to cluster #0
    alloc=""; zero=""
    _run_test sc=0 len=64k cmd=compress

    ### Write subcluster #1 (middle of subcluster)
    alloc="$(seq 0 31)"; zero=""
    _run_test sc=1 off=1 len=512
done

############################################################
############################################################
############################################################

# calculate_l2_meta() checks if none of the clusters affected by a
# write operation need COW or changes to their L2 metadata and simply
# returns when they don't. This is a test for that optimization.
# Here clusters #0-#3 are overwritten but only #1 and #2 need changes.
echo
echo '### Overwriting several clusters without COW ###'
echo
use_backing_file="no" _reset_img 1M
# Write cluster #0, subclusters #12-#31
alloc="$(seq 12 31)"; zero=""
_run_test sc=12 len=40k

# Write cluster #1, subcluster #13
alloc="13"; zero=""
_run_test c=1 sc=13 len=2k

# Zeroize cluster #2, subcluster #14
alloc="14"; zero=""
_run_test c=2 sc=14 len=2k
alloc=""; zero="14"
_run_test c=2 sc=14 len=2k cmd=zero

# Write cluster #3, subclusters #0-#16
alloc="$(seq 0 16)"; zero=""
_run_test c=3 sc=0 len=34k

# Write from cluster #0, subcluster #12 to cluster #3, subcluster #11
alloc="$(seq 12 31)"; zero=""
_run_test sc=12 len=192k
alloc="$(seq 0 31)"; zero=""
_verify_l2_bitmap 1
_verify_l2_bitmap 2

alloc="$(seq 0 16)"; zero=""
_verify_l2_bitmap 3

############################################################
############################################################
############################################################

# Test different patterns of writing zeroes
for use_backing_file in yes no; do
    echo
    echo "### Writing zeroes 1: unallocated clusters (backing file: $use_backing_file) ###"
    echo
    # Note that the image size is not a multiple of the cluster size
    _reset_img 2083k

    # Cluster-aligned request from clusters #0 to #2
    alloc=""; zero="$(seq 0 31)"
    _run_test c=0 sc=0 len=192k cmd=zero
    _verify_l2_bitmap 1
    _verify_l2_bitmap 2

    # Subcluster-aligned request from clusters #3 to #5
    alloc=""; zero="$(seq 16 31)"
    _run_test c=3 sc=16 len=128k cmd=zero
    alloc=""; zero="$(seq 0 31)"
    _verify_l2_bitmap 4
    alloc=""; zero="$(seq 0 15)"
    _verify_l2_bitmap 5

    # Unaligned request from clusters #6 to #8
    if [ "$use_backing_file" = "yes" ]; then
        alloc="15"; zero="$(seq 16 31)" # copy-on-write happening here
    else
        alloc=""; zero="$(seq 15 31)"
    fi
    _run_test c=6 sc=15 off=1 len=128k cmd=zero
    alloc=""; zero="$(seq 0 31)"
    _verify_l2_bitmap 7
    if [ "$use_backing_file" = "yes" ]; then
        alloc="15"; zero="$(seq 0 14)" # copy-on-write happening here
    else
        alloc=""; zero="$(seq 0 15)"
    fi
    _verify_l2_bitmap 8

    echo
    echo "### Writing zeroes 2: allocated clusters (backing file: $use_backing_file) ###"
    echo
    alloc="$(seq 0 31)"; zero=""
    _run_test c=9 sc=0 len=576k
    _verify_l2_bitmap 10
    _verify_l2_bitmap 11
    _verify_l2_bitmap 12
    _verify_l2_bitmap 13
    _verify_l2_bitmap 14
    _verify_l2_bitmap 15
    _verify_l2_bitmap 16
    _verify_l2_bitmap 17

    # Cluster-aligned request from clusters #9 to #11
    alloc=""; zero="$(seq 0 31)"
    _run_test c=9 sc=0 len=192k cmd=zero
    _verify_l2_bitmap 10
    _verify_l2_bitmap 11

    # Subcluster-aligned request from clusters #12 to #14
    alloc="$(seq 0 15)"; zero="$(seq 16 31)"
    _run_test c=12 sc=16 len=128k cmd=zero
    alloc=""; zero="$(seq 0 31)"
    _verify_l2_bitmap 13
    alloc="$(seq 16 31)"; zero="$(seq 0 15)"
    _verify_l2_bitmap 14

    # Unaligned request from clusters #15 to #17
    alloc="$(seq 0 15)"; zero="$(seq 16 31)"
    _run_test c=15 sc=15 off=1 len=128k cmd=zero
    alloc=""; zero="$(seq 0 31)"
    _verify_l2_bitmap 16
    alloc="$(seq 15 31)"; zero="$(seq 0 14)"
    _verify_l2_bitmap 17

    echo
    echo "### Writing zeroes 3: compressed clusters (backing file: $use_backing_file) ###"
    echo
    alloc=""; zero=""
    for c in $(seq 18 28); do
        _run_test c=$c sc=0 len=64k cmd=compress
    done

    # Cluster-aligned request from clusters #18 to #20
    alloc=""; zero="$(seq 0 31)"
    _run_test c=18 sc=0 len=192k cmd=zero
    _verify_l2_bitmap 19
    _verify_l2_bitmap 20

    # Subcluster-aligned request from clusters #21 to #23.
    # We cannot partially zero a compressed cluster so the code
    # returns -ENOTSUP, which means copy-on-write of the compressed
    # data and fill the rest with actual zeroes on disk.
    # TODO: cluster #22 should use the 'all zeroes' bits.
    alloc="$(seq 0 31)"; zero=""
    _run_test c=21 sc=16 len=128k cmd=zero
    _verify_l2_bitmap 22
    _verify_l2_bitmap 23

    # Unaligned request from clusters #24 to #26
    # In this case QEMU internally sends a 1k request followed by a
    # subcluster-aligned 128k request. The first request decompresses
    # cluster #24, but that's not enough to perform the second request
    # efficiently because it partially writes to cluster #26 (which is
    # compressed) so we hit the same problem as before.
    alloc="$(seq 0 31)"; zero=""
    _run_test c=24 sc=15 off=1 len=129k cmd=zero
    _verify_l2_bitmap 25
    _verify_l2_bitmap 26

    # Unaligned request from clusters #27 to #29
    # Similar to the previous case, but this time the tail of the
    # request does not correspond to a compressed cluster, so it can
    # be zeroed efficiently.
    # Note that the very last subcluster is partially written, so if
    # there's a backing file we need to perform cow.
    alloc="$(seq 0 15)"; zero="$(seq 16 31)"
    _run_test c=27 sc=15 off=1 len=128k cmd=zero
    alloc=""; zero="$(seq 0 31)"
    _verify_l2_bitmap 28
    if [ "$use_backing_file" = "yes" ]; then
        alloc="15"; zero="$(seq 0 14)" # copy-on-write happening here
    else
        alloc=""; zero="$(seq 0 15)"
    fi
    _verify_l2_bitmap 29

    echo
    echo "### Writing zeroes 4: other tests (backing file: $use_backing_file) ###"
    echo
    # Unaligned request in the middle of cluster #30.
    # If there's a backing file we need to allocate and do
    # copy-on-write on the partially zeroed subclusters.
    # If not we can set the 'all zeroes' bit on them.
    if [ "$use_backing_file" = "yes" ]; then
        alloc="15 19"; zero="$(seq 16 18)" # copy-on-write happening here
    else
        alloc=""; zero="$(seq 15 19)"
    fi
    _run_test c=30 sc=15 off=1 len=8k cmd=zero

    # Fill the last cluster with zeroes, up to the end of the image
    # (the image size is not a multiple of the cluster or subcluster size).
    alloc=""; zero="$(seq 0 17)"
    _run_test c=32 sc=0 len=35k cmd=zero
done

############################################################
############################################################
############################################################

# Zero + unmap
for use_backing_file in yes no; do
    echo
    echo "### Zero + unmap 1: allocated clusters (backing file: $use_backing_file) ###"
    echo
    # Note that the image size is not a multiple of the cluster size
    _reset_img 2083k
    alloc="$(seq 0 31)"; zero=""
    _run_test c=9 sc=0 len=576k
    _verify_l2_bitmap 10
    _verify_l2_bitmap 11
    _verify_l2_bitmap 12
    _verify_l2_bitmap 13
    _verify_l2_bitmap 14
    _verify_l2_bitmap 15
    _verify_l2_bitmap 16
    _verify_l2_bitmap 17

    # Cluster-aligned request from clusters #9 to #11
    alloc=""; zero="$(seq 0 31)"
    _run_test c=9 sc=0 len=192k cmd=unmap
    _verify_l2_bitmap 10
    _verify_l2_bitmap 11

    # Subcluster-aligned request from clusters #12 to #14
    alloc="$(seq 0 15)"; zero="$(seq 16 31)"
    _run_test c=12 sc=16 len=128k cmd=unmap
    alloc=""; zero="$(seq 0 31)"
    _verify_l2_bitmap 13
    alloc="$(seq 16 31)"; zero="$(seq 0 15)"
    _verify_l2_bitmap 14

    # Unaligned request from clusters #15 to #17
    alloc="$(seq 0 15)"; zero="$(seq 16 31)"
    _run_test c=15 sc=15 off=1 len=128k cmd=unmap
    alloc=""; zero="$(seq 0 31)"
    _verify_l2_bitmap 16
    alloc="$(seq 15 31)"; zero="$(seq 0 14)"
    _verify_l2_bitmap 17

    echo
    echo "### Zero + unmap 2: compressed clusters (backing file: $use_backing_file) ###"
    echo
    alloc=""; zero=""
    for c in $(seq 18 28); do
        _run_test c=$c sc=0 len=64k cmd=compress
    done

    # Cluster-aligned request from clusters #18 to #20
    alloc=""; zero="$(seq 0 31)"
    _run_test c=18 sc=0 len=192k cmd=unmap
    _verify_l2_bitmap 19
    _verify_l2_bitmap 20

    # Subcluster-aligned request from clusters #21 to #23.
    # We cannot partially zero a compressed cluster so the code
    # returns -ENOTSUP, which means copy-on-write of the compressed
    # data and fill the rest with actual zeroes on disk.
    # TODO: cluster #22 should use the 'all zeroes' bits.
    alloc="$(seq 0 31)"; zero=""
    _run_test c=21 sc=16 len=128k cmd=unmap
    _verify_l2_bitmap 22
    _verify_l2_bitmap 23

    # Unaligned request from clusters #24 to #26
    # In this case QEMU internally sends a 1k request followed by a
    # subcluster-aligned 128k request. The first request decompresses
    # cluster #24, but that's not enough to perform the second request
    # efficiently because it partially writes to cluster #26 (which is
    # compressed) so we hit the same problem as before.
    alloc="$(seq 0 31)"; zero=""
    _run_test c=24 sc=15 off=1 len=129k cmd=unmap
    _verify_l2_bitmap 25
    _verify_l2_bitmap 26

    # Unaligned request from clusters #27 to #29
    # Similar to the previous case, but this time the tail of the
    # request does not correspond to a compressed cluster, so it can
    # be zeroed efficiently.
    # Note that the very last subcluster is partially written, so if
    # there's a backing file we need to perform cow.
    alloc="$(seq 0 15)"; zero="$(seq 16 31)"
    _run_test c=27 sc=15 off=1 len=128k cmd=unmap
    alloc=""; zero="$(seq 0 31)"
    _verify_l2_bitmap 28
    if [ "$use_backing_file" = "yes" ]; then
        alloc="15"; zero="$(seq 0 14)" # copy-on-write happening here
    else
        alloc=""; zero="$(seq 0 15)"
    fi
    _verify_l2_bitmap 29
done

############################################################
############################################################
############################################################

# Test qcow2_cluster_discard() with full and normal discards
for use_backing_file in yes no; do
    echo
    echo "### Discarding clusters with non-zero bitmaps (backing file: $use_backing_file) ###"
    echo
    if [ "$use_backing_file" = "yes" ]; then
        _make_test_img -o extended_l2=on -F raw -b "$TEST_IMG.base" 1M
    else
        _make_test_img -o extended_l2=on 1M
    fi
    # Write clusters #0-#2 and then discard them
    $QEMU_IO -c 'write -q 0 128k' "$TEST_IMG"
    $QEMU_IO -c 'discard -q 0 128k' "$TEST_IMG"
    # 'qemu-io discard' doesn't do a full discard, it zeroizes the
    # cluster, so both clusters have all zero bits set now
    alloc=""; zero="$(seq 0 31)"
    _verify_l2_bitmap 0
    _verify_l2_bitmap 1
    # Now mark the 2nd half of the subclusters from cluster #0 as unallocated
    poke_file "$TEST_IMG" $(($l2_offset+8)) "\x00\x00"
    # Discard cluster #0 again to see how the zero bits have changed
    $QEMU_IO -c 'discard -q 0 64k' "$TEST_IMG"
    # And do a full discard of cluster #1 by shrinking and growing the image
    $QEMU_IMG resize --shrink "$TEST_IMG" 64k
    $QEMU_IMG resize "$TEST_IMG" 1M
    # A normal discard sets all 'zero' bits only if the image has a
    # backing file, otherwise it won't touch them.
    if [ "$use_backing_file" = "yes" ]; then
        alloc=""; zero="$(seq 0 31)"
    else
        alloc=""; zero="$(seq 0 15)"
    fi
    _verify_l2_bitmap 0
    # A full discard should clear the L2 entry completely. However
    # when growing an image with a backing file the new clusters are
    # zeroized to hide the stale data from the backing file
    if [ "$use_backing_file" = "yes" ]; then
        alloc=""; zero="$(seq 0 31)"
    else
        alloc=""; zero=""
    fi
    _verify_l2_bitmap 1
done

############################################################
############################################################
############################################################

# Test that corrupted L2 entries are detected in both read and write
# operations
for corruption_test_cmd in read write; do
    echo
    echo "### Corrupted L2 entries - $corruption_test_cmd test (allocated) ###"
    echo
    echo "# 'cluster is zero' bit set on the standard cluster descriptor"
    echo
    # We actually don't consider this a corrupted image.
    # The 'cluster is zero' bit is unused in extended L2 entries so
    # QEMU ignores it.
    # TODO: maybe treat the image as corrupted and make qemu-img check fix it?
    _make_test_img -o extended_l2=on 1M
    $QEMU_IO -c 'write -q -P 0x11 0 2k' "$TEST_IMG"
    poke_file "$TEST_IMG" $(($l2_offset+7)) "\x01"
    alloc="0"; zero=""
    _verify_l2_bitmap 0
    $QEMU_IO -c "$corruption_test_cmd -q -P 0x11 0 1k" "$TEST_IMG"
    if [ "$corruption_test_cmd" = "write" ]; then
        alloc="0"; zero=""
    fi
    _verify_l2_bitmap 0

    echo
    echo "# Both 'subcluster is zero' and 'subcluster is allocated' bits set"
    echo
    _make_test_img -o extended_l2=on 1M
    # Write from the middle of cluster #0 to the middle of cluster #2
    $QEMU_IO -c 'write -q 32k 128k' "$TEST_IMG"
    # Corrupt the L2 entry from cluster #1
    poke_file_be "$TEST_IMG" $(($l2_offset+24)) 4 1
    alloc="$(seq 0 31)"; zero="0"
    _verify_l2_bitmap 1
    $QEMU_IO -c "$corruption_test_cmd 0 192k" "$TEST_IMG"

    echo
    echo "### Corrupted L2 entries - $corruption_test_cmd test (unallocated) ###"
    echo
    echo "# 'cluster is zero' bit set on the standard cluster descriptor"
    echo
    # We actually don't consider this a corrupted image.
    # The 'cluster is zero' bit is unused in extended L2 entries so
    # QEMU ignores it.
    # TODO: maybe treat the image as corrupted and make qemu-img check fix it?
    _make_test_img -o extended_l2=on 1M
    # We want to modify the (empty) L2 entry from cluster #0,
    # but we write to #4 in order to initialize the L2 table first
    $QEMU_IO -c 'write -q 256k 1k' "$TEST_IMG"
    poke_file "$TEST_IMG" $(($l2_offset+7)) "\x01"
    alloc=""; zero=""
    _verify_l2_bitmap 0
    $QEMU_IO -c "$corruption_test_cmd -q 0 1k" "$TEST_IMG"
    if [ "$corruption_test_cmd" = "write" ]; then
        alloc="0"; zero=""
    fi
    _verify_l2_bitmap 0

    echo
    echo "# 'subcluster is allocated' bit set"
    echo
    _make_test_img -o extended_l2=on 1M
    # We want to corrupt the (empty) L2 entry from cluster #0,
    # but we write to #4 in order to initialize the L2 table first
    $QEMU_IO -c 'write -q 256k 1k' "$TEST_IMG"
    poke_file "$TEST_IMG" $(($l2_offset+15)) "\x01"
    alloc="0"; zero=""
    _verify_l2_bitmap 0
    $QEMU_IO -c "$corruption_test_cmd 0 1k" "$TEST_IMG"

    echo
    echo "# Both 'subcluster is zero' and 'subcluster is allocated' bits set"
    echo
    _make_test_img -o extended_l2=on 1M
    # We want to corrupt the (empty) L2 entry from cluster #1,
    # but we write to #4 in order to initialize the L2 table first
    $QEMU_IO -c 'write -q 256k 1k' "$TEST_IMG"
    # Corrupt the L2 entry from cluster #1
    poke_file_be "$TEST_IMG" $(($l2_offset+24)) 8 $(((1 << 32) | 1))
    alloc="0"; zero="0"
    _verify_l2_bitmap 1
    $QEMU_IO -c "$corruption_test_cmd 0 192k" "$TEST_IMG"

    echo
    echo "### Compressed cluster with subcluster bitmap != 0 - $corruption_test_cmd test ###"
    echo
    # We actually don't consider this a corrupted image.
    # The bitmap in compressed clusters is unused so QEMU should just ignore it.
    _make_test_img -o extended_l2=on 1M
    $QEMU_IO -c 'write -q -P 11 -c 0 64k' "$TEST_IMG"
    # Change the L2 bitmap to allocate subcluster #31 and zeroize subcluster #0
    poke_file "$TEST_IMG" $(($l2_offset+11)) "\x01\x80"
    alloc="31"; zero="0"
    _verify_l2_bitmap 0
    $QEMU_IO -c "$corruption_test_cmd -P 11 0 64k" "$TEST_IMG" | _filter_qemu_io
    # Writing allocates a new uncompressed cluster so we get a new bitmap
    if [ "$corruption_test_cmd" = "write" ]; then
        alloc="$(seq 0 31)"; zero=""
    fi
    _verify_l2_bitmap 0
done

############################################################
############################################################
############################################################

echo
echo "### Detect and repair unaligned clusters ###"
echo
# Create a backing file and fill it with data
$QEMU_IMG create -f raw "$TEST_IMG.base" 128k | _filter_img_create
$QEMU_IO -c "write -q -P 0xff 0 128k" -f raw "$TEST_IMG.base" | _filter_qemu_io

echo "# Corrupted L2 entry, allocated subcluster #"
# Create a new image, allocate a cluster and write some data to it
_make_test_img -o extended_l2=on -F raw -b "$TEST_IMG.base"
$QEMU_IO -c 'write -q -P 1 4k 2k' "$TEST_IMG"
# Corrupt the L2 entry by making the offset unaligned
poke_file "$TEST_IMG" "$(($l2_offset+6))" "\x02"
# This cannot be repaired, qemu-img check will fail to fix it
_check_test_img -r all
# Attempting to read the image will still show that it's corrupted
$QEMU_IO -c 'read -q 0 2k' "$TEST_IMG"

echo "# Corrupted L2 entry, no allocated subclusters #"
# Create a new image, allocate a cluster and zeroize subcluster #2
_make_test_img -o extended_l2=on -F raw -b "$TEST_IMG.base"
$QEMU_IO -c 'write -q -P 1 4k 2k' "$TEST_IMG"
$QEMU_IO -c 'write -q -z   4k 2k' "$TEST_IMG"
# Corrupt the L2 entry by making the offset unaligned
poke_file "$TEST_IMG" "$(($l2_offset+6))" "\x02"
# This time none of the subclusters are allocated so we can repair the image
_check_test_img -r all
# And the data can be read normally
$QEMU_IO -c 'read -q -P 0xff  0   4k' "$TEST_IMG"
$QEMU_IO -c 'read -q -P 0x00 4k   2k' "$TEST_IMG"
$QEMU_IO -c 'read -q -P 0xff 6k 122k' "$TEST_IMG"

############################################################
############################################################
############################################################

echo
echo "### Image creation options ###"
echo
echo "# cluster_size < 16k"
_make_test_img -o extended_l2=on,cluster_size=8k 1M

echo "# backing file and preallocation=metadata"
# For preallocation with backing files, create a backing file first
$QEMU_IMG create -f raw "$TEST_IMG.base" 1M | _filter_img_create
$QEMU_IO -c "write -q -P 0xff 0 1M" -f raw "$TEST_IMG.base" | _filter_qemu_io

_make_test_img -o extended_l2=on,preallocation=metadata -F raw -b "$TEST_IMG.base" 512k
$QEMU_IMG resize "$TEST_IMG" 1M
$QEMU_IO -c 'read -P 0xff    0 512k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IO -c 'read -P 0x00 512k 512k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IMG map "$TEST_IMG" | _filter_testdir

echo "# backing file and preallocation=falloc"
_make_test_img -o extended_l2=on,preallocation=falloc -F raw -b "$TEST_IMG.base" 512k
$QEMU_IMG resize "$TEST_IMG" 1M
$QEMU_IO -c 'read -P 0xff    0 512k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IO -c 'read -P 0x00 512k 512k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IMG map "$TEST_IMG" | _filter_testdir

echo "# backing file and preallocation=full"
_make_test_img -o extended_l2=on,preallocation=full -F raw -b "$TEST_IMG.base" 512k
$QEMU_IMG resize "$TEST_IMG" 1M
$QEMU_IO -c 'read -P 0xff    0 512k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IO -c 'read -P 0x00 512k 512k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IMG map "$TEST_IMG" | _filter_testdir

echo
echo "### Image resizing with preallocation and backing files ###"
echo
# In this case the new subclusters must have the 'all zeroes' bit set
echo "# resize --preallocation=metadata"
_make_test_img -o extended_l2=on -F raw -b "$TEST_IMG.base" 503k
$QEMU_IMG resize --preallocation=metadata "$TEST_IMG" 1013k
$QEMU_IO -c 'read -P 0xff    0 503k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IO -c 'read -P 0x00 503k 510k' "$TEST_IMG" | _filter_qemu_io

# In this case and the next one the new subclusters must be allocated
echo "# resize --preallocation=falloc"
_make_test_img -o extended_l2=on -F raw -b "$TEST_IMG.base" 503k
$QEMU_IMG resize --preallocation=falloc "$TEST_IMG" 1013k
$QEMU_IO -c 'read -P 0xff    0 503k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IO -c 'read -P 0x00 503k 510k' "$TEST_IMG" | _filter_qemu_io

echo "# resize --preallocation=full"
_make_test_img -o extended_l2=on -F raw -b "$TEST_IMG.base" 503k
$QEMU_IMG resize --preallocation=full "$TEST_IMG" 1013k
$QEMU_IO -c 'read -P 0xff    0 503k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IO -c 'read -P 0x00 503k 510k' "$TEST_IMG" | _filter_qemu_io

echo
echo "### Image resizing with preallocation without backing files ###"
echo
# In this case the new subclusters must have the 'all zeroes' bit set
echo "# resize --preallocation=metadata"
_make_test_img -o extended_l2=on 503k
$QEMU_IO -c 'write -P 0xff    0 503k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IMG resize --preallocation=metadata "$TEST_IMG" 1013k
$QEMU_IO -c 'read -P 0xff    0 503k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IO -c 'read -P 0x00 503k 510k' "$TEST_IMG" | _filter_qemu_io

# In this case and the next one the new subclusters must be allocated
echo "# resize --preallocation=falloc"
_make_test_img -o extended_l2=on 503k
$QEMU_IO -c 'write -P 0xff    0 503k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IMG resize --preallocation=falloc "$TEST_IMG" 1013k
$QEMU_IO -c 'read -P 0xff    0 503k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IO -c 'read -P 0x00 503k 510k' "$TEST_IMG" | _filter_qemu_io

echo "# resize --preallocation=full"
_make_test_img -o extended_l2=on 503k
$QEMU_IO -c 'write -P 0xff    0 503k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IMG resize --preallocation=full "$TEST_IMG" 1013k
$QEMU_IO -c 'read -P 0xff    0 503k' "$TEST_IMG" | _filter_qemu_io
$QEMU_IO -c 'read -P 0x00 503k 510k' "$TEST_IMG" | _filter_qemu_io

echo
echo "### qemu-img measure ###"
echo
echo "# 512MB, extended_l2=off" # This needs one L2 table
$QEMU_IMG measure --size 512M -O qcow2 -o extended_l2=off
echo "# 512MB, extended_l2=on"  # This needs two L2 tables
$QEMU_IMG measure --size 512M -O qcow2 -o extended_l2=on

echo "# 16K clusters, 64GB, extended_l2=off" # This needs one full L1 table cluster
$QEMU_IMG measure --size 64G -O qcow2 -o cluster_size=16k,extended_l2=off
echo "# 16K clusters, 64GB, extended_l2=on"  # This needs two full L2 table clusters
$QEMU_IMG measure --size 64G -O qcow2 -o cluster_size=16k,extended_l2=on

echo "# 8k clusters" # This should fail
$QEMU_IMG measure --size 1M -O qcow2 -o cluster_size=8k,extended_l2=on

echo "# 1024 TB" # Maximum allowed size with extended_l2=on and 64K clusters
$QEMU_IMG measure --size 1024T -O qcow2 -o extended_l2=on
echo "# 1025 TB" # This should fail
$QEMU_IMG measure --size 1025T -O qcow2 -o extended_l2=on

echo
echo "### qemu-img amend ###"
echo
_make_test_img -o extended_l2=on 1M
$QEMU_IMG amend -o extended_l2=off "$TEST_IMG" && echo "Unexpected pass"

_make_test_img -o extended_l2=off 1M
$QEMU_IMG amend -o extended_l2=on "$TEST_IMG" && echo "Unexpected pass"

echo
echo "### Test copy-on-write on an image with snapshots ###"
echo
_make_test_img -o extended_l2=on 1M

# For each cluster from #0 to #9 this loop zeroes subcluster #7
# and allocates subclusters #13 and #18.
alloc="13 18"; zero="7"
for c in $(seq 0 9); do
    $QEMU_IO -c "write -q -z $((64*$c+14))k 2k" \
             -c "write -q -P $((0xd0+$c)) $((64*$c+26))k 2k" \
             -c "write -q -P $((0xe0+$c)) $((64*$c+36))k 2k" "$TEST_IMG"
    _verify_l2_bitmap "$c"
done

# Create a snapshot and set l2_offset to the new L2 table
$QEMU_IMG snapshot -c snap1 "$TEST_IMG"
l2_offset=$((0x110000))

# Write different patterns to each one of the clusters
# in order to see how copy-on-write behaves in each case.
$QEMU_IO -c "write -q -P 0xf0 $((64*0+30))k 1k" \
         -c "write -q -P 0xf1 $((64*1+20))k 1k" \
         -c "write -q -P 0xf2 $((64*2+40))k 1k" \
         -c "write -q -P 0xf3 $((64*3+26))k 1k" \
         -c "write -q -P 0xf4 $((64*4+14))k 1k" \
         -c "write -q -P 0xf5 $((64*5+1))k  1k" \
         -c "write -q -z      $((64*6+30))k 3k" \
         -c "write -q -z      $((64*7+26))k 2k" \
         -c "write -q -z      $((64*8+26))k 1k" \
         -c "write -q -z      $((64*9+12))k 1k" \
         "$TEST_IMG"
alloc="$(seq 13 18)"; zero="7" _verify_l2_bitmap 0
alloc="$(seq 10 18)"; zero="7" _verify_l2_bitmap 1
alloc="$(seq 13 20)"; zero="7" _verify_l2_bitmap 2
alloc="$(seq 13 18)"; zero="7" _verify_l2_bitmap 3
alloc="$(seq 7 18)";  zero=""  _verify_l2_bitmap 4
alloc="$(seq 0 18)";  zero=""  _verify_l2_bitmap 5
alloc="13 18";  zero="7 15 16" _verify_l2_bitmap 6
alloc="18";        zero="7 13" _verify_l2_bitmap 7
alloc="$(seq 13 18)"; zero="7" _verify_l2_bitmap 8
alloc="13 18";      zero="6 7" _verify_l2_bitmap 9

echo
echo "### Test concurrent requests ###"
echo

_concurrent_io()
{
# Allocate three subclusters in the same cluster.
# This works because handle_dependencies() checks whether the requests
# allocate the same cluster, even if the COW regions don't overlap (in
# this case they don't).
cat <<EOF
open -o driver=$IMGFMT blkdebug::$TEST_IMG
break write_aio A
aio_write -P 10 30k 2k
wait_break A
aio_write -P 11 20k 2k
aio_write -P 12 40k 2k
resume A
aio_flush
EOF
}

_concurrent_verify()
{
cat <<EOF
open -o driver=$IMGFMT $TEST_IMG
read -q -P 10 30k 2k
read -q -P 11 20k 2k
read -q -P 12 40k 2k
EOF
}

_make_test_img -o extended_l2=on 1M
# Second and third writes in _concurrent_io() are independent and may finish in
# different order. So, filter offset out to match both possible variants.
_concurrent_io     | $QEMU_IO | _filter_qemu_io | \
    sed -e 's/\(20480\|40960\)/OFFSET/'
_concurrent_verify | $QEMU_IO | _filter_qemu_io

############################################################
############################################################
############################################################

echo
echo "### Rebase of qcow2 images with subclusters ###"
echo

l2_offset=$((0x400000))

# Check that rebase operation preserve holes between allocated subclusters
# within one cluster (i.e. does not allocate extra space).  Check that the
# data is preserved as well.
#
# Base (new backing): -- -- -- ... -- -- --
# Mid (old backing):  -- 11 -- ... -- 22 --
# Top:                -- -- -- ... -- -- --

echo "### Preservation of unallocated holes after rebase ###"
echo

echo "# create backing chain"
echo

TEST_IMG="$TEST_IMG.base" _make_test_img -o cluster_size=1M,extended_l2=on 1M
TEST_IMG="$TEST_IMG.mid" _make_test_img -o cluster_size=1M,extended_l2=on \
    -b "$TEST_IMG.base" -F qcow2 1M
TEST_IMG="$TEST_IMG.top" _make_test_img -o cluster_size=1M,extended_l2=on \
    -b "$TEST_IMG.mid" -F qcow2 1M

echo
echo "# fill old backing with data (separate subclusters within cluster)"
echo

$QEMU_IO -c "write -P 0x11 32k 32k" \
         -c "write -P 0x22 $(( 30 * 32 ))k 32k" \
         "$TEST_IMG.mid" | _filter_qemu_io

echo
echo "# rebase topmost image onto the new backing"
echo

$QEMU_IMG rebase -b "$TEST_IMG.base" -F qcow2 "$TEST_IMG.top"

echo "# verify that data is read the same before and after rebase"
echo

$QEMU_IO -c "read -P 0x00 0 32k" \
         -c "read -P 0x11 32k 32k" \
         -c "read -P 0x00 64k $(( 28 * 32 ))k" \
         -c "read -P 0x22 $(( 30 * 32 ))k 32k" \
         -c "read -P 0x00 $(( 31 * 32 ))k 32k" \
         "$TEST_IMG.top" | _filter_qemu_io

echo
echo "# verify that only selected subclusters remain allocated"
echo

$QEMU_IMG map "$TEST_IMG.top" | _filter_testdir

echo
echo "# verify image bitmap"
echo

TEST_IMG="$TEST_IMG.top" alloc="1 30" zero="" _verify_l2_bitmap 0

# Check that rebase with compression works correctly with images containing
# subclusters.  When compression is enabled and we allocate a new
# subcluster within the target (overlay) image, we expect the entire cluster
# containing that subcluster to become compressed.
#
# Here we expect 1st and 3rd clusters of the top (overlay) image to become
# compressed after the rebase, while cluster 2 to remain unallocated and
# be read from the base (new backing) image.
#
# Base (new backing): |-- -- .. -- --|11 11 .. 11 11|-- -- .. -- --|
# Mid (old backing):  |-- -- .. -- 22|-- -- .. -- --|33 -- .. -- --|
# Top:                |-- -- .. -- --|-- -- -- -- --|-- -- .. -- --|

echo
echo "### Rebase with compression for images with subclusters ###"
echo

echo "# create backing chain"
echo

TEST_IMG="$TEST_IMG.base" _make_test_img -o cluster_size=1M,extended_l2=on 3M
TEST_IMG="$TEST_IMG.mid" _make_test_img -o cluster_size=1M,extended_l2=on \
    -b "$TEST_IMG.base" -F qcow2 3M
TEST_IMG="$TEST_IMG.top" _make_test_img -o cluster_size=1M,extended_l2=on \
    -b "$TEST_IMG.mid" -F qcow2 3M

echo
echo "# fill old and new backing with data"
echo

$QEMU_IO -c "write -P 0x11 1M 1M" "$TEST_IMG.base" | _filter_qemu_io
$QEMU_IO -c "write -P 0x22 $(( 31 * 32 ))k 32k" \
         -c "write -P 0x33 $(( 64 * 32 ))k 32k" \
         "$TEST_IMG.mid" | _filter_qemu_io

echo
echo "# rebase topmost image onto the new backing, with compression"
echo

$QEMU_IMG rebase -c -b "$TEST_IMG.base" -F qcow2 "$TEST_IMG.top"

echo "# verify that the 1st and 3rd clusters've become compressed"
echo

$QEMU_IMG map --output=json "$TEST_IMG.top" | _filter_testdir

echo
echo "# verify that data is read the same before and after rebase"
echo

$QEMU_IO -c "read -P 0x22 $(( 31 * 32 ))k 32k" \
         -c "read -P 0x11 1M 1M" \
         -c "read -P 0x33 $(( 64 * 32 ))k 32k" \
         "$TEST_IMG.top" | _filter_qemu_io

echo
echo "# verify image bitmap"
echo

# For compressed clusters bitmap is always 0.  For unallocated cluster
# there should be no entry at all, thus bitmap is also 0.
TEST_IMG="$TEST_IMG.top" alloc="" zero="" _verify_l2_bitmap 0
TEST_IMG="$TEST_IMG.top" alloc="" zero="" _verify_l2_bitmap 1
TEST_IMG="$TEST_IMG.top" alloc="" zero="" _verify_l2_bitmap 2

# success, all done
echo "*** done"
rm -f $seq.full
status=0