xref: /openbmc/qemu/tests/qemu-iotests/060 (revision 06d4c71f)
1#!/usr/bin/env bash
2#
3# Test case for image corruption (overlapping data structures) in qcow2
4#
5# Copyright (C) 2013 Red Hat, Inc.
6#
7# This program is free software; you can redistribute it and/or modify
8# it under the terms of the GNU General Public License as published by
9# the Free Software Foundation; either version 2 of the License, or
10# (at your option) any later version.
11#
12# This program is distributed in the hope that it will be useful,
13# but WITHOUT ANY WARRANTY; without even the implied warranty of
14# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15# GNU General Public License for more details.
16#
17# You should have received a copy of the GNU General Public License
18# along with this program.  If not, see <http://www.gnu.org/licenses/>.
19#
20
21# creator
22owner=mreitz@redhat.com
23
24seq="$(basename $0)"
25echo "QA output created by $seq"
26
27status=1	# failure is the default!
28
29_cleanup()
30{
31	_cleanup_test_img
32}
33trap "_cleanup; exit \$status" 0 1 2 3 15
34
35# Sometimes the error line might be dumped before/after an event
36# randomly.  Mask it out for specific test that may trigger this
37# uncertainty for current test for now.
38_filter_io_error()
39{
40    sed '/Input\/output error/d'
41}
42
43# get standard environment, filters and checks
44. ./common.rc
45. ./common.filter
46
47# This tests qcow2-specific low-level functionality
48_supported_fmt qcow2
49_supported_proto file
50_supported_os Linux
51# These tests only work for compat=1.1 images without an external
52# data file with refcount_bits=16
53_unsupported_imgopts 'compat=0.10' data_file \
54    'refcount_bits=\([^1]\|.\([^6]\|$\)\)'
55
56# The repair process will create a large file - so check for availability first
57_require_large_file 64G
58
59rt_offset=65536  # 0x10000 (XXX: just an assumption)
60rb_offset=131072 # 0x20000 (XXX: just an assumption)
61l1_offset=196608 # 0x30000 (XXX: just an assumption)
62l2_offset=262144 # 0x40000 (XXX: just an assumption)
63l2_offset_after_snapshot=524288 # 0x80000 (XXX: just an assumption)
64
65OPEN_RW="open -o overlap-check=all $TEST_IMG"
66# Overlap checks are done before write operations only, therefore opening an
67# image read-only makes the overlap-check option irrelevant
68OPEN_RO="open -r $TEST_IMG"
69
70echo
71echo "=== Testing L2 reference into L1 ==="
72echo
73_make_test_img 64M
74# Link first L1 entry (first L2 table) onto itself
75# (Note the MSb in the L1 entry is set, ensuring the refcount is one - else any
76# later write will result in a COW operation, effectively ruining this attempt
77# on image corruption)
78poke_file "$TEST_IMG" "$l1_offset" "\x80\x00\x00\x00\x00\x03\x00\x00"
79_check_test_img
80
81# The corrupt bit should not be set anyway
82$PYTHON qcow2.py "$TEST_IMG" dump-header | grep incompatible_features
83
84# Try to write something, thereby forcing the corrupt bit to be set
85$QEMU_IO -c "$OPEN_RW" -c "write -P 0x2a 0 512" | _filter_qemu_io
86
87# The corrupt bit must now be set
88$PYTHON qcow2.py "$TEST_IMG" dump-header | grep incompatible_features
89
90# This information should be available through qemu-img info
91_img_info --format-specific
92
93# Try to open the image R/W (which should fail)
94$QEMU_IO -c "$OPEN_RW" -c "read 0 512" 2>&1 | _filter_qemu_io \
95                                            | _filter_testdir \
96                                            | _filter_imgfmt
97
98# Try to open it RO (which should succeed)
99$QEMU_IO -c "$OPEN_RO" -c "read 0 512" | _filter_qemu_io
100
101# We could now try to fix the image, but this would probably fail (how should an
102# L2 table linked onto the L1 table be fixed?)
103
104echo
105echo "=== Testing cluster data reference into refcount block ==="
106echo
107_make_test_img 64M
108# Allocate L2 table
109truncate -s "$(($l2_offset+65536))" "$TEST_IMG"
110poke_file "$TEST_IMG" "$l1_offset" "\x80\x00\x00\x00\x00\x04\x00\x00"
111# Mark cluster as used
112poke_file "$TEST_IMG" "$(($rb_offset+8))" "\x00\x01"
113# Redirect new data cluster onto refcount block
114poke_file "$TEST_IMG" "$l2_offset" "\x80\x00\x00\x00\x00\x02\x00\x00"
115_check_test_img
116$PYTHON qcow2.py "$TEST_IMG" dump-header | grep incompatible_features
117$QEMU_IO -c "$OPEN_RW" -c "write -P 0x2a 0 512" | _filter_qemu_io
118$PYTHON qcow2.py "$TEST_IMG" dump-header | grep incompatible_features
119
120# Try to fix it
121_check_test_img -r all
122
123# The corrupt bit should be cleared
124$PYTHON qcow2.py "$TEST_IMG" dump-header | grep incompatible_features
125
126# Look if it's really really fixed
127$QEMU_IO -c "$OPEN_RW" -c "write -P 0x2a 0 512" | _filter_qemu_io
128$PYTHON qcow2.py "$TEST_IMG" dump-header | grep incompatible_features
129
130echo
131echo "=== Testing cluster data reference into inactive L2 table ==="
132echo
133_make_test_img 64M
134$QEMU_IO -c "$OPEN_RW" -c "write -P 1 0 512" | _filter_qemu_io
135$QEMU_IMG snapshot -c foo "$TEST_IMG"
136$QEMU_IO -c "$OPEN_RW" -c "write -P 2 0 512" | _filter_qemu_io
137# The inactive L2 table remains at its old offset
138poke_file "$TEST_IMG" "$l2_offset_after_snapshot" \
139                      "\x80\x00\x00\x00\x00\x04\x00\x00"
140_check_test_img
141$PYTHON qcow2.py "$TEST_IMG" dump-header | grep incompatible_features
142$QEMU_IO -c "$OPEN_RW" -c "write -P 3 0 512" | _filter_qemu_io
143$PYTHON qcow2.py "$TEST_IMG" dump-header | grep incompatible_features
144_check_test_img -r all
145$PYTHON qcow2.py "$TEST_IMG" dump-header | grep incompatible_features
146$QEMU_IO -c "$OPEN_RW" -c "write -P 4 0 512" | _filter_qemu_io
147$PYTHON qcow2.py "$TEST_IMG" dump-header | grep incompatible_features
148
149# Check data
150$QEMU_IO -c "$OPEN_RO" -c "read -P 4 0 512" | _filter_qemu_io
151$QEMU_IMG snapshot -a foo "$TEST_IMG"
152_check_test_img
153$QEMU_IO -c "$OPEN_RO" -c "read -P 1 0 512" | _filter_qemu_io
154
155echo
156echo "=== Testing overlap while COW is in flight ==="
157echo
158BACKING_IMG=$TEST_IMG.base
159TEST_IMG=$BACKING_IMG _make_test_img 1G
160
161$QEMU_IO -c 'write 0k 64k' "$BACKING_IMG" | _filter_qemu_io
162
163_make_test_img -b "$BACKING_IMG" 1G
164# Write two clusters, the second one enforces creation of an L2 table after
165# the first data cluster.
166$QEMU_IO -c 'write 0k 64k' -c 'write 512M 64k' "$TEST_IMG" | _filter_qemu_io
167# Free the first cluster. This cluster will soon enough be reallocated and
168# used for COW.
169poke_file "$TEST_IMG" "$l2_offset" "\x00\x00\x00\x00\x00\x00\x00\x00"
170poke_file "$TEST_IMG" "$(($rb_offset+10))" "\x00\x00"
171# Now, corrupt the image by marking the second L2 table cluster as free.
172poke_file "$TEST_IMG" "$(($rb_offset+12))" "\x00\x00"
173# Start a write operation requiring COW on the image stopping it right before
174# doing the read; then, trigger the corruption prevention by writing anything to
175# any unallocated cluster, leading to an attempt to overwrite the second L2
176# table. Finally, resume the COW write and see it fail (but not crash).
177echo "open -o file.driver=blkdebug $TEST_IMG
178break cow_read 0
179aio_write 0k 1k
180wait_break 0
181write 64k 64k
182resume 0" | $QEMU_IO | _filter_qemu_io
183
184echo
185echo "=== Testing unallocated image header ==="
186echo
187_make_test_img 64M
188# Create L1/L2
189$QEMU_IO -c "write 0 64k" "$TEST_IMG" | _filter_qemu_io
190poke_file "$TEST_IMG" "$rb_offset" "\x00\x00"
191$QEMU_IO -c "write 64k 64k" "$TEST_IMG" | _filter_qemu_io
192
193echo
194echo "=== Testing unaligned L1 entry ==="
195echo
196_make_test_img 64M
197$QEMU_IO -c "write 0 64k" "$TEST_IMG" | _filter_qemu_io
198# This will be masked with ~(512 - 1) = ~0x1ff, so whether the lower 9 bits are
199# aligned or not does not matter
200poke_file "$TEST_IMG" "$l1_offset" "\x80\x00\x00\x00\x00\x04\x2a\x00"
201$QEMU_IO -c "read 0 64k" "$TEST_IMG" | _filter_qemu_io
202
203# Test how well zero cluster expansion can cope with this
204_make_test_img 64M
205$QEMU_IO -c "write 0 64k" "$TEST_IMG" | _filter_qemu_io
206poke_file "$TEST_IMG" "$l1_offset" "\x80\x00\x00\x00\x00\x04\x2a\x00"
207$QEMU_IMG amend -o compat=0.10 "$TEST_IMG"
208
209echo
210echo "=== Testing unaligned L2 entry ==="
211echo
212_make_test_img 64M
213$QEMU_IO -c "write 0 64k" "$TEST_IMG" | _filter_qemu_io
214poke_file "$TEST_IMG" "$l2_offset" "\x80\x00\x00\x00\x00\x05\x2a\x00"
215$QEMU_IO -c "read 0 64k" "$TEST_IMG" | _filter_qemu_io
216
217echo
218echo "=== Testing unaligned pre-allocated zero cluster ==="
219echo
220_make_test_img 64M
221$QEMU_IO -c "write 0 64k" "$TEST_IMG" | _filter_qemu_io
222poke_file "$TEST_IMG" "$l2_offset" "\x80\x00\x00\x00\x00\x05\x2a\x01"
223# zero cluster expansion
224$QEMU_IMG amend -o compat=0.10 "$TEST_IMG"
225
226echo
227echo "=== Testing unaligned reftable entry ==="
228echo
229_make_test_img 64M
230poke_file "$TEST_IMG" "$rt_offset" "\x00\x00\x00\x00\x00\x02\x2a\x00"
231$QEMU_IO -c "write 0 64k" "$TEST_IMG" | _filter_qemu_io
232
233echo
234echo "=== Testing non-fatal corruption on freeing ==="
235echo
236_make_test_img 64M
237$QEMU_IO -c "write 0 64k" "$TEST_IMG" | _filter_qemu_io
238poke_file "$TEST_IMG" "$l2_offset" "\x80\x00\x00\x00\x00\x05\x2a\x00"
239$QEMU_IO -c "discard 0 64k" "$TEST_IMG" | _filter_qemu_io
240
241echo
242echo "=== Testing read-only corruption report ==="
243echo
244_make_test_img 64M
245$QEMU_IO -c "write 0 64k" "$TEST_IMG" | _filter_qemu_io
246poke_file "$TEST_IMG" "$l2_offset" "\x80\x00\x00\x00\x00\x05\x2a\x00"
247# Should only emit a single error message
248$QEMU_IO -c "$OPEN_RO" -c "read 0 64k" -c "read 0 64k" | _filter_qemu_io
249
250echo
251echo "=== Testing non-fatal and then fatal corruption report ==="
252echo
253_make_test_img 64M
254$QEMU_IO -c "write 0 128k" "$TEST_IMG" | _filter_qemu_io
255poke_file "$TEST_IMG" "$l2_offset"        "\x80\x00\x00\x00\x00\x05\x2a\x00"
256poke_file "$TEST_IMG" "$(($l2_offset+8))" "\x80\x00\x00\x00\x00\x06\x2a\x00"
257# Should emit two error messages
258$QEMU_IO -c "discard 0 64k" -c "read 64k 64k" "$TEST_IMG" | _filter_qemu_io
259
260echo
261echo "=== Testing empty refcount table ==="
262echo
263_make_test_img 64M
264poke_file "$TEST_IMG" "$rt_offset"        "\x00\x00\x00\x00\x00\x00\x00\x00"
265$QEMU_IO -c "write 0 64k" "$TEST_IMG" | _filter_qemu_io
266# Repair the image
267_check_test_img -r all
268
269echo
270echo "=== Testing empty refcount table with valid L1 and L2 tables ==="
271echo
272_make_test_img 64M
273$QEMU_IO -c "write 0 64k" "$TEST_IMG" | _filter_qemu_io
274poke_file "$TEST_IMG" "$rt_offset"        "\x00\x00\x00\x00\x00\x00\x00\x00"
275# Since the first data cluster is already allocated this triggers an
276# allocation with an explicit offset (using qcow2_alloc_clusters_at())
277# causing a refcount block to be allocated at offset 0
278$QEMU_IO -c "write 0 128k" "$TEST_IMG" | _filter_qemu_io
279# Repair the image
280_check_test_img -r all
281
282echo
283echo "=== Testing empty refcount block ==="
284echo
285_make_test_img 64M
286poke_file "$TEST_IMG" "$rb_offset"        "\x00\x00\x00\x00\x00\x00\x00\x00"
287$QEMU_IO -c "write 0 64k" "$TEST_IMG" | _filter_qemu_io
288# Repair the image
289_check_test_img -r all
290
291echo
292echo "=== Testing empty refcount block with compressed write ==="
293echo
294_make_test_img 64M
295$QEMU_IO -c "write 64k 64k" "$TEST_IMG" | _filter_qemu_io
296poke_file "$TEST_IMG" "$rb_offset"        "\x00\x00\x00\x00\x00\x00\x00\x00"
297# The previous write already allocated an L2 table, so now this new
298# write will try to allocate a compressed data cluster at offset 0.
299$QEMU_IO -c "write -c 0k 64k" "$TEST_IMG" | _filter_qemu_io
300# Repair the image
301_check_test_img -r all
302
303echo
304echo "=== Testing zero refcount table size ==="
305echo
306_make_test_img 64M
307poke_file "$TEST_IMG" "56"                "\x00\x00\x00\x00"
308$QEMU_IO -c "write 0 64k" "$TEST_IMG" 2>&1 | _filter_testdir | _filter_imgfmt
309# Repair the image
310_check_test_img -r all
311
312echo
313echo "=== Testing incorrect refcount table offset ==="
314echo
315_make_test_img 64M
316poke_file "$TEST_IMG" "48"                "\x00\x00\x00\x00\x00\x00\x00\x00"
317$QEMU_IO -c "write 0 64k" "$TEST_IMG" | _filter_qemu_io
318
319echo
320echo "=== Testing dirty corrupt image ==="
321echo
322
323_make_test_img 64M
324
325# Let the refblock appear unaligned
326poke_file "$TEST_IMG" "$rt_offset"        "\x00\x00\x00\x00\xff\xff\x2a\x00"
327# Mark the image dirty, thus forcing an automatic check when opening it
328poke_file "$TEST_IMG" 72 "\x00\x00\x00\x00\x00\x00\x00\x01"
329# Open the image (qemu should refuse to do so)
330$QEMU_IO -c close "$TEST_IMG" 2>&1 | _filter_testdir | _filter_imgfmt
331
332echo '--- Repairing ---'
333
334# The actual repair should have happened (because of the dirty bit),
335# but some cleanup may have failed (like freeing the old reftable)
336# because the image was already marked corrupt by that point
337_check_test_img -r all
338
339echo
340echo "=== Writing to an unaligned preallocated zero cluster ==="
341echo
342
343_make_test_img 64M
344
345# Allocate the L2 table
346$QEMU_IO -c "write 0 64k" -c "discard 0 64k" "$TEST_IMG" | _filter_qemu_io
347# Pretend there is a preallocated zero cluster somewhere inside the
348# image header
349poke_file "$TEST_IMG" "$l2_offset" "\x80\x00\x00\x00\x00\x00\x2a\x01"
350# Let's write to it!
351$QEMU_IO -c "write 0 64k" "$TEST_IMG" | _filter_qemu_io
352
353echo '--- Repairing ---'
354_check_test_img -r all
355
356echo
357echo '=== Discarding with an unaligned refblock ==='
358echo
359
360_make_test_img 64M
361
362$QEMU_IO -c "write 0 128k" "$TEST_IMG" | _filter_qemu_io
363# Make our refblock unaligned
364poke_file "$TEST_IMG" "$(($rt_offset))" "\x00\x00\x00\x00\x00\x00\x2a\x00"
365# Now try to discard something that will be submitted as two requests
366# (main part + tail)
367$QEMU_IO -c "discard 0 65537" "$TEST_IMG"
368
369echo '--- Repairing ---'
370# Fails the first repair because the corruption prevents the check
371# function from double-checking
372# (Using -q for the first invocation, because otherwise the
373#  double-check error message appears above the summary for some
374#  reason -- so let's just hide the summary)
375_check_test_img -q -r all
376_check_test_img -r all
377
378echo
379echo "=== Discarding an out-of-bounds refblock ==="
380echo
381
382_make_test_img 64M
383
384# Pretend there's a refblock really up high
385poke_file "$TEST_IMG" "$(($rt_offset+8))" "\x00\xff\xff\xff\x00\x00\x00\x00"
386# Let's try to shrink the qcow2 image so that the block driver tries
387# to discard that refblock (and see what happens!)
388$QEMU_IMG resize --shrink "$TEST_IMG" 32M
389
390echo '--- Checking and retrying ---'
391# Image should not be resized
392_img_info | grep 'virtual size'
393# But it should pass this check, because the "partial" resize has
394# already overwritten refblocks past the end
395_check_test_img -r all
396# So let's try again
397$QEMU_IMG resize --shrink "$TEST_IMG" 32M
398_img_info | grep 'virtual size'
399
400echo
401echo "=== Discarding a non-covered in-bounds refblock ==="
402echo
403
404_make_test_img -o 'refcount_bits=1' 64M
405
406# Pretend there's a refblock somewhere where there is no refblock to
407# cover it (but the covering refblock has a valid index in the
408# reftable)
409# Every refblock covers 65536 * 8 * 65536 = 32 GB, so we have to point
410# to 0x10_0000_0000 (64G) to point to the third refblock
411poke_file "$TEST_IMG" "$(($rt_offset+8))" "\x00\x00\x00\x10\x00\x00\x00\x00"
412$QEMU_IMG resize --shrink "$TEST_IMG" 32M
413
414echo '--- Checking and retrying ---'
415# Image should not be resized
416_img_info | grep 'virtual size'
417# But it should pass this check, because the "partial" resize has
418# already overwritten refblocks past the end
419_check_test_img -r all
420# So let's try again
421$QEMU_IMG resize --shrink "$TEST_IMG" 32M
422_img_info | grep 'virtual size'
423
424echo
425echo "=== Discarding a refblock covered by an unaligned refblock ==="
426echo
427
428_make_test_img -o 'refcount_bits=1' 64M
429
430# Same as above
431poke_file "$TEST_IMG" "$(($rt_offset+8))" "\x00\x00\x00\x10\x00\x00\x00\x00"
432# But now we actually "create" an unaligned third refblock
433poke_file "$TEST_IMG" "$(($rt_offset+16))" "\x00\x00\x00\x00\x00\x00\x02\x00"
434$QEMU_IMG resize --shrink "$TEST_IMG" 32M
435
436echo '--- Repairing ---'
437# Fails the first repair because the corruption prevents the check
438# function from double-checking
439# (Using -q for the first invocation, because otherwise the
440#  double-check error message appears above the summary for some
441#  reason -- so let's just hide the summary)
442_check_test_img -q -r all
443_check_test_img -r all
444
445echo
446echo "=== Testing the QEMU shutdown with a corrupted image ==="
447echo
448_make_test_img 64M
449poke_file "$TEST_IMG" "$rt_offset"        "\x00\x00\x00\x00\x00\x00\x00\x00"
450echo "{'execute': 'qmp_capabilities'}
451      {'execute': 'human-monitor-command',
452       'arguments': {'command-line': 'qemu-io drive \"write 0 512\"'}}
453      {'execute': 'quit'}" \
454    | $QEMU -qmp stdio -nographic -nodefaults \
455            -drive if=none,node-name=drive,file="$TEST_IMG",driver=qcow2 \
456    | _filter_qmp | _filter_qemu_io
457
458echo
459echo "=== Testing incoming inactive corrupted image ==="
460echo
461
462_make_test_img 64M
463# Create an unaligned L1 entry, so qemu will signal a corruption when
464# reading from the covered area
465poke_file "$TEST_IMG" "$l1_offset" "\x00\x00\x00\x00\x2a\x2a\x2a\x2a"
466
467# Inactive images are effectively read-only images, so this should be a
468# non-fatal corruption (which does not modify the image)
469echo "{'execute': 'qmp_capabilities'}
470      {'execute': 'human-monitor-command',
471       'arguments': {'command-line': 'qemu-io drive \"read 0 512\"'}}
472      {'execute': 'quit'}" \
473    | $QEMU -qmp stdio -nographic -nodefaults \
474            -blockdev "{'node-name': 'drive',
475                        'driver': 'qcow2',
476                        'file': {
477                            'driver': 'file',
478                            'filename': '$TEST_IMG'
479                        }}" \
480            -incoming exec:'cat /dev/null' \
481            2>&1 \
482    | _filter_qmp | _filter_qemu_io | _filter_io_error
483
484echo
485# Image should not have been marked corrupt
486_img_info --format-specific | grep 'corrupt:'
487
488# success, all done
489echo "*** done"
490rm -f $seq.full
491status=0
492