tests/qemu-iotests/060

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