xref: /qemu/tests/qemu-iotests/108 (revision c4b8ffcb)
1#!/usr/bin/env bash
2# group: rw auto quick
3#
4# Test case for repairing qcow2 images which cannot be repaired using
5# the on-disk refcount structures
6#
7# Copyright (C) 2014 Red Hat, Inc.
8#
9# This program is free software; you can redistribute it and/or modify
10# it under the terms of the GNU General Public License as published by
11# the Free Software Foundation; either version 2 of the License, or
12# (at your option) any later version.
13#
14# This program is distributed in the hope that it will be useful,
15# but WITHOUT ANY WARRANTY; without even the implied warranty of
16# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17# GNU General Public License for more details.
18#
19# You should have received a copy of the GNU General Public License
20# along with this program.  If not, see <http://www.gnu.org/licenses/>.
21#
22
23# creator
24owner=hreitz@redhat.com
25
26seq="$(basename $0)"
27echo "QA output created by $seq"
28
29status=1	# failure is the default!
30
31_cleanup()
32{
33    _cleanup_test_img
34    if [ -f "$TEST_DIR/qsd.pid" ]; then
35        qsd_pid=$(cat "$TEST_DIR/qsd.pid")
36        kill -KILL "$qsd_pid"
37        fusermount -u "$TEST_DIR/fuse-export" &>/dev/null
38    fi
39    rm -f "$TEST_DIR/fuse-export"
40}
41trap "_cleanup; exit \$status" 0 1 2 3 15
42
43# get standard environment, filters and checks
44. ./common.rc
45. ./common.filter
46. ./common.qemu
47
48# This tests qcow2-specific low-level functionality
49_supported_fmt qcow2
50_supported_proto file fuse
51_supported_os Linux
52# This test directly modifies a refblock so it relies on refcount_bits being 16;
53# and the low-level modification it performs are not tuned for external data
54# files
55_unsupported_imgopts 'refcount_bits=\([^1]\|.\([^6]\|$\)\)' data_file
56
57# This test either needs sudo -n losetup or FUSE exports to work
58if sudo -n losetup &>/dev/null; then
59    loopdev=true
60else
61    loopdev=false
62
63    # QSD --export fuse will either yield "Parameter 'id' is missing"
64    # or "Invalid parameter 'fuse'", depending on whether there is
65    # FUSE support or not.
66    error=$($QSD --export fuse 2>&1)
67    if [[ $error = *"'fuse'"* ]]; then
68        _notrun 'Passwordless sudo for losetup or FUSE support required, but' \
69                'neither is available'
70    fi
71fi
72
73echo
74echo '=== Repairing an image without any refcount table ==='
75echo
76
77_make_test_img 64M
78# just write some data
79$QEMU_IO -c 'write -P 42 0 64k' "$TEST_IMG" | _filter_qemu_io
80
81# refcount_table_offset
82poke_file "$TEST_IMG" $((0x30)) "\x00\x00\x00\x00\x00\x00\x00\x00"
83# refcount_table_clusters
84poke_file "$TEST_IMG" $((0x38)) "\x00\x00\x00\x00"
85
86_check_test_img -r all
87
88$QEMU_IO -c 'read -P 42 0 64k' "$TEST_IMG" | _filter_qemu_io
89
90echo
91echo '=== Repairing unreferenced data cluster in new refblock area ==='
92echo
93
94_make_test_img -o 'cluster_size=512' 64M
95# Allocate the first 128 kB in the image (first refblock)
96$QEMU_IO -c 'write 0 0x1b200' "$TEST_IMG" | _filter_qemu_io
97# should be 131072 == 0x20000
98stat -c '%s' "$TEST_IMG"
99
100# Enter a cluster at 128 kB (0x20000)
101# XXX: This should be the first free entry in the last L2 table, but we cannot
102# be certain
103poke_file "$TEST_IMG" $((0x1ccc8)) "\x80\x00\x00\x00\x00\x02\x00\x00"
104
105# Fill the cluster
106truncate -s $((0x20200)) "$TEST_IMG"
107$QEMU_IO -c "open -o driver=raw $TEST_IMG" -c 'write -P 42 128k 512' \
108    | _filter_qemu_io
109
110# The data should now appear at this guest offset
111$QEMU_IO -c 'read -P 42 0x1b200 512' "$TEST_IMG" | _filter_qemu_io
112
113# This cluster is unallocated; fix it
114_check_test_img -r all
115
116# This repair operation must have allocated a new refblock; and that refblock
117# should not overlap with the unallocated data cluster. If it does, the data
118# will be damaged, so check it.
119$QEMU_IO -c 'read -P 42 0x1b200 512' "$TEST_IMG" | _filter_qemu_io
120
121echo
122echo '=== Repairing refblock beyond the image end ==='
123echo
124
125echo
126echo '--- Otherwise clean ---'
127echo
128
129_make_test_img 64M
130# Normally, qemu doesn't create empty refblocks, so we just have to do it by
131# hand
132# XXX: This should be the entry for the second refblock
133poke_file "$TEST_IMG" $((0x10008)) "\x00\x00\x00\x00\x00\x10\x00\x00"
134# Mark that refblock as used
135# XXX: This should be the 17th entry (cluster 16) of the first
136# refblock
137poke_file "$TEST_IMG" $((0x20020)) "\x00\x01"
138_check_test_img -r all
139
140echo
141echo '--- Refblock is unallocated ---'
142echo
143
144_make_test_img 64M
145poke_file "$TEST_IMG" $((0x10008)) "\x00\x00\x00\x00\x00\x10\x00\x00"
146_check_test_img -r all
147
148echo
149echo '--- Signed overflow after the refblock ---'
150echo
151
152_make_test_img 64M
153poke_file "$TEST_IMG" $((0x10008)) "\x7f\xff\xff\xff\xff\xff\x00\x00"
154_check_test_img -r all
155
156echo
157echo '--- Unsigned overflow after the refblock ---'
158echo
159
160_make_test_img 64M
161poke_file "$TEST_IMG" $((0x10008)) "\xff\xff\xff\xff\xff\xff\x00\x00"
162_check_test_img -r all
163
164echo
165echo '=== Check rebuilt reftable location ==='
166
167# In an earlier version of the refcount rebuild algorithm, the
168# reftable was generally placed at the image end (unless something was
169# allocated in the area covered by the refblock right before the image
170# file end, then we would try to place the reftable in that refblock).
171# This was later changed so the reftable would be placed in the
172# earliest possible location.  Test this.
173
174echo
175echo '--- Does the image size increase? ---'
176echo
177
178# First test: Just create some image, write some data to it, and
179# resize it so there is free space at the end of the image (enough
180# that it spans at least one full refblock, which for cluster_size=512
181# images, spans 128k).  With the old algorithm, the reftable would
182# have then been placed at the end of the image file, but with the new
183# one, it will be put in that free space.
184# We want to check whether the size of the image file increases due to
185# rebuilding the refcount structures (it should not).
186
187_make_test_img -o 'cluster_size=512' 1M
188# Write something
189$QEMU_IO -c 'write 0 64k' "$TEST_IMG" | _filter_qemu_io
190
191# Add free space
192file_len=$(stat -c '%s' "$TEST_IMG")
193truncate -s $((file_len + 256 * 1024)) "$TEST_IMG"
194
195# Corrupt the image by saying the image header was not allocated
196rt_offset=$(peek_file_be "$TEST_IMG" 48 8)
197rb_offset=$(peek_file_be "$TEST_IMG" $rt_offset 8)
198poke_file "$TEST_IMG" $rb_offset "\x00\x00"
199
200# Check whether rebuilding the refcount structures increases the image
201# file size
202file_len=$(stat -c '%s' "$TEST_IMG")
203echo
204# The only leaks there can be are the old refcount structures that are
205# leaked during rebuilding, no need to clutter the output with them
206_check_test_img -r all | grep -v '^Repairing cluster.*refcount=1 reference=0'
207echo
208post_repair_file_len=$(stat -c '%s' "$TEST_IMG")
209
210if [[ $file_len -eq $post_repair_file_len ]]; then
211    echo 'OK: Image size did not change'
212else
213    echo 'ERROR: Image size differs' \
214        "($file_len before, $post_repair_file_len after)"
215fi
216
217echo
218echo '--- Will the reftable occupy a hole specifically left for it?  ---'
219echo
220
221# Note: With cluster_size=512, every refblock covers 128k.
222# The reftable covers 8M per reftable cluster.
223
224# Create an image that requires two reftable clusters (just because
225# this is more interesting than a single-clustered reftable).
226_make_test_img -o 'cluster_size=512' 9M
227$QEMU_IO -c 'write 0 8M' "$TEST_IMG" | _filter_qemu_io
228
229# Writing 8M will have resized the reftable.  Unfortunately, doing so
230# will leave holes in the file, so we need to fill them up so we can
231# be sure the whole file is allocated.  Do that by writing
232# consecutively smaller chunks starting from 8 MB, until the file
233# length increases even with a chunk size of 512.  Then we must have
234# filled all holes.
235ofs=$((8 * 1024 * 1024))
236block_len=$((16 * 1024))
237while [[ $block_len -ge 512 ]]; do
238    file_len=$(stat -c '%s' "$TEST_IMG")
239    while [[ $(stat -c '%s' "$TEST_IMG") -eq $file_len ]]; do
240        # Do not include this in the reference output, it does not
241        # really matter which qemu-io calls we do here exactly
242        $QEMU_IO -c "write $ofs $block_len" "$TEST_IMG" >/dev/null
243        ofs=$((ofs + block_len))
244    done
245    block_len=$((block_len / 2))
246done
247
248# Fill up to 9M (do not include this in the reference output either,
249# $ofs is random for all we know)
250$QEMU_IO -c "write $ofs $((9 * 1024 * 1024 - ofs))" "$TEST_IMG" >/dev/null
251
252# Make space as follows:
253# - For the first refblock: Right at the beginning of the image (this
254#   refblock is placed in the first place possible),
255# - For the reftable somewhere soon afterwards, still near the
256#   beginning of the image (i.e. covered by the first refblock); the
257#   reftable too is placed in the first place possible, but only after
258#   all refblocks have been placed)
259# No space is needed for the other refblocks, because no refblock is
260# put before the space it covers.  In this test case, we do not mind
261# if they are placed at the image file's end.
262
263# Before we make that space, we have to find out the host offset of
264# the area that belonged to the two data clusters at guest offset 4k,
265# because we expect the reftable to be placed there, and we will have
266# to verify that it is.
267
268l1_offset=$(peek_file_be "$TEST_IMG" 40 8)
269l2_offset=$(peek_file_be "$TEST_IMG" $l1_offset 8)
270l2_offset=$((l2_offset & 0x00fffffffffffe00))
271data_4k_offset=$(peek_file_be "$TEST_IMG" \
272                 $((l2_offset + 4096 / 512 * 8)) 8)
273data_4k_offset=$((data_4k_offset & 0x00fffffffffffe00))
274
275$QEMU_IO -c "discard 0 512" -c "discard 4k 1k" "$TEST_IMG" | _filter_qemu_io
276
277# Corrupt the image by saying the image header was not allocated
278rt_offset=$(peek_file_be "$TEST_IMG" 48 8)
279rb_offset=$(peek_file_be "$TEST_IMG" $rt_offset 8)
280poke_file "$TEST_IMG" $rb_offset "\x00\x00"
281
282echo
283# The only leaks there can be are the old refcount structures that are
284# leaked during rebuilding, no need to clutter the output with them
285_check_test_img -r all | grep -v '^Repairing cluster.*refcount=1 reference=0'
286echo
287
288# Check whether the reftable was put where we expected
289rt_offset=$(peek_file_be "$TEST_IMG" 48 8)
290if [[ $rt_offset -eq $data_4k_offset ]]; then
291    echo 'OK: Reftable is where we expect it'
292else
293    echo "ERROR: Reftable is at $rt_offset, but was expected at $data_4k_offset"
294fi
295
296echo
297echo '--- Rebuilding refcount structures on block devices ---'
298echo
299
300# A block device cannot really grow, at least not during qemu-img
301# check.  As mentioned in the above cases, rebuilding the refcount
302# structure may lead to new refcount structures being written after
303# the end of the image, and in the past that happened even if there
304# was more than sufficient space in the image.  Such post-EOF writes
305# will not work on block devices, so test that the new algorithm
306# avoids it.
307
308# If we have passwordless sudo and losetup, we can use those to create
309# a block device.  Otherwise, we can resort to qemu's FUSE export to
310# create a file that isn't growable, which effectively tests the same
311# thing.
312
313_cleanup_test_img
314truncate -s $((64 * 1024 * 1024)) "$TEST_IMG"
315
316if $loopdev; then
317    export_mp=$(sudo -n losetup --show -f "$TEST_IMG")
318    export_mp_driver=host_device
319    sudo -n chmod go+rw "$export_mp"
320else
321    # Create non-growable FUSE export that is a bit like an empty
322    # block device
323    export_mp="$TEST_DIR/fuse-export"
324    export_mp_driver=file
325    touch "$export_mp"
326
327    $QSD \
328        --blockdev file,node-name=export-node,filename="$TEST_IMG" \
329        --export fuse,id=fuse-export,node-name=export-node,mountpoint="$export_mp",writable=on,growable=off,allow-other=off \
330        --pidfile "$TEST_DIR/qsd.pid" \
331        --daemonize
332fi
333
334# Now create a qcow2 image on the device -- unfortunately, qemu-img
335# create force-creates the file, so we have to resort to the
336# blockdev-create job.
337_launch_qemu \
338    --blockdev $export_mp_driver,node-name=file,filename="$export_mp"
339
340_send_qemu_cmd \
341    $QEMU_HANDLE \
342    '{ "execute": "qmp_capabilities" }' \
343    'return'
344
345# Small cluster size again, so the image needs multiple refblocks
346_send_qemu_cmd \
347    $QEMU_HANDLE \
348    '{ "execute": "blockdev-create",
349       "arguments": {
350           "job-id": "create",
351           "options": {
352               "driver": "qcow2",
353               "file": "file",
354               "size": '$((64 * 1024 * 1024))',
355               "cluster-size": 512
356           } } }' \
357    '"concluded"'
358
359_send_qemu_cmd \
360    $QEMU_HANDLE \
361    '{ "execute": "job-dismiss", "arguments": { "id": "create" } }' \
362    'return'
363
364_send_qemu_cmd \
365    $QEMU_HANDLE \
366    '{ "execute": "quit" }' \
367    'return'
368
369wait=y _cleanup_qemu
370echo
371
372# Write some data
373$QEMU_IO -c 'write 0 64k' "$export_mp" | _filter_qemu_io
374
375# Corrupt the image by saying the image header was not allocated
376rt_offset=$(peek_file_be "$export_mp" 48 8)
377rb_offset=$(peek_file_be "$export_mp" $rt_offset 8)
378poke_file "$export_mp" $rb_offset "\x00\x00"
379
380# Repairing such a simple case should just work
381# (We used to put the reftable at the end of the image file, which can
382# never work for non-growable devices.)
383echo
384TEST_IMG="$export_mp" _check_test_img -r all \
385    | grep -v '^Repairing cluster.*refcount=1 reference=0'
386
387if $loopdev; then
388    sudo -n losetup -d "$export_mp"
389else
390    qsd_pid=$(cat "$TEST_DIR/qsd.pid")
391    kill -TERM "$qsd_pid"
392    # Wait for process to exit (cannot `wait` because the QSD is daemonized)
393    while [ -f "$TEST_DIR/qsd.pid" ]; do
394        true
395    done
396fi
397
398# success, all done
399echo '*** done'
400rm -f $seq.full
401status=0
402