tests/qemu-iotests/257

   1 #!/usr/bin/env python3
   2 # group: rw
   3 #
   4 # Test bitmap-sync backups (incremental, differential, and partials)
   5 #
   6 # Copyright (c) 2019 John Snow for 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 # owner=jsnow@redhat.com
  22
  23 import math
  24 import os
  25
  26 import iotests
  27 from iotests import log, qemu_img
  28
  29 SIZE = 64 * 1024 * 1024
  30 GRANULARITY = 64 * 1024
  31
  32
  33 class Pattern:
  34     def __init__(self, byte, offset, size=GRANULARITY):
  35         self.byte = byte
  36         self.offset = offset
  37         self.size = size
  38
  39     def bits(self, granularity):
  40         lower = self.offset // granularity
  41         upper = (self.offset + self.size - 1) // granularity
  42         return set(range(lower, upper + 1))
  43
  44
  45 class PatternGroup:
  46     """Grouping of Pattern objects. Initialize with an iterable of Patterns."""
  47     def __init__(self, patterns):
  48         self.patterns = patterns
  49
  50     def bits(self, granularity):
  51         """Calculate the unique bits dirtied by this pattern grouping"""
  52         res = set()
  53         for pattern in self.patterns:
  54             res |= pattern.bits(granularity)
  55         return res
  56
  57
  58 GROUPS = [
  59     PatternGroup([
  60         # Batch 0: 4 clusters
  61         Pattern('0x49', 0x0000000),
  62         Pattern('0x6c', 0x0100000),   # 1M
  63         Pattern('0x6f', 0x2000000),   # 32M
  64         Pattern('0x76', 0x3ff0000)]), # 64M - 64K
  65     PatternGroup([
  66         # Batch 1: 6 clusters (3 new)
  67         Pattern('0x65', 0x0000000),   # Full overwrite
  68         Pattern('0x77', 0x00f8000),   # Partial-left (1M-32K)
  69         Pattern('0x72', 0x2008000),   # Partial-right (32M+32K)
  70         Pattern('0x69', 0x3fe0000)]), # Adjacent-left (64M - 128K)
  71     PatternGroup([
  72         # Batch 2: 7 clusters (3 new)
  73         Pattern('0x74', 0x0010000),   # Adjacent-right
  74         Pattern('0x69', 0x00e8000),   # Partial-left  (1M-96K)
  75         Pattern('0x6e', 0x2018000),   # Partial-right (32M+96K)
  76         Pattern('0x67', 0x3fe0000,
  77                 2*GRANULARITY)]),     # Overwrite [(64M-128K)-64M)
  78     PatternGroup([
  79         # Batch 3: 8 clusters (5 new)
  80         # Carefully chosen such that nothing re-dirties the one cluster
  81         # that copies out successfully before failure in Group #1.
  82         Pattern('0xaa', 0x0010000,
  83                 3*GRANULARITY),       # Overwrite and 2x Adjacent-right
  84         Pattern('0xbb', 0x00d8000),   # Partial-left (1M-160K)
  85         Pattern('0xcc', 0x2028000),   # Partial-right (32M+160K)
  86         Pattern('0xdd', 0x3fc0000)]), # New; leaving a gap to the right
  87 ]
  88
  89
  90 class EmulatedBitmap:
  91     def __init__(self, granularity=GRANULARITY):
  92         self._bits = set()
  93         self.granularity = granularity
  94
  95     def dirty_bits(self, bits):
  96         self._bits |= set(bits)
  97
  98     def dirty_group(self, n):
  99         self.dirty_bits(GROUPS[n].bits(self.granularity))
 100
 101     def clear(self):
 102         self._bits = set()
 103
 104     def clear_bits(self, bits):
 105         self._bits -= set(bits)
 106
 107     def clear_bit(self, bit):
 108         self.clear_bits({bit})
 109
 110     def clear_group(self, n):
 111         self.clear_bits(GROUPS[n].bits(self.granularity))
 112
 113     @property
 114     def first_bit(self):
 115         return sorted(self.bits)[0]
 116
 117     @property
 118     def bits(self):
 119         return self._bits
 120
 121     @property
 122     def count(self):
 123         return len(self.bits)
 124
 125     def compare(self, qmp_bitmap):
 126         """
 127         Print a nice human-readable message checking that a bitmap as reported
 128         by the QMP interface has as many bits set as we expect it to.
 129         """
 130
 131         name = qmp_bitmap.get('name', '(anonymous)')
 132         log("= Checking Bitmap {:s} =".format(name))
 133
 134         want = self.count
 135         have = qmp_bitmap['count'] // qmp_bitmap['granularity']
 136
 137         log("expecting {:d} dirty sectors; have {:d}. {:s}".format(
 138             want, have, "OK!" if want == have else "ERROR!"))
 139         log('')
 140
 141
 142 class Drive:
 143     """Represents, vaguely, a drive attached to a VM.
 144     Includes format, graph, and device information."""
 145
 146     def __init__(self, path, vm=None):
 147         self.path = path
 148         self.vm = vm
 149         self.fmt = None
 150         self.size = None
 151         self.node = None
 152
 153     def img_create(self, fmt, size):
 154         self.fmt = fmt
 155         self.size = size
 156         iotests.qemu_img_create('-f', self.fmt, self.path, str(self.size))
 157
 158     def create_target(self, name, fmt, size):
 159         basename = os.path.basename(self.path)
 160         file_node_name = "file_{}".format(basename)
 161         vm = self.vm
 162
 163         log(vm.command('blockdev-create', job_id='bdc-file-job',
 164                        options={
 165                            'driver': 'file',
 166                            'filename': self.path,
 167                            'size': 0,
 168                        }))
 169         vm.run_job('bdc-file-job')
 170         log(vm.command('blockdev-add', driver='file',
 171                        node_name=file_node_name, filename=self.path))
 172
 173         log(vm.command('blockdev-create', job_id='bdc-fmt-job',
 174                        options={
 175                            'driver': fmt,
 176                            'file': file_node_name,
 177                            'size': size,
 178                        }))
 179         vm.run_job('bdc-fmt-job')
 180         log(vm.command('blockdev-add', driver=fmt,
 181                        node_name=name,
 182                        file=file_node_name))
 183         self.fmt = fmt
 184         self.size = size
 185         self.node = name
 186
 187 def blockdev_backup(vm, device, target, sync, **kwargs):
 188     # Strip any arguments explicitly nulled by the caller:
 189     kwargs = {key: val for key, val in kwargs.items() if val is not None}
 190     result = vm.qmp_log('blockdev-backup',
 191                         device=device,
 192                         target=target,
 193                         sync=sync,
 194                         filter_node_name='backup-top',
 195                         x_perf={'max-workers': 1},
 196                         **kwargs)
 197     return result
 198
 199 def blockdev_backup_mktarget(drive, target_id, filepath, sync, **kwargs):
 200     target_drive = Drive(filepath, vm=drive.vm)
 201     target_drive.create_target(target_id, drive.fmt, drive.size)
 202     blockdev_backup(drive.vm, drive.node, target_id, sync, **kwargs)
 203
 204 def reference_backup(drive, n, filepath):
 205     log("--- Reference Backup #{:d} ---\n".format(n))
 206     target_id = "ref_target_{:d}".format(n)
 207     job_id = "ref_backup_{:d}".format(n)
 208     blockdev_backup_mktarget(drive, target_id, filepath, "full",
 209                              job_id=job_id)
 210     drive.vm.run_job(job_id, auto_dismiss=True)
 211     log('')
 212
 213 def backup(drive, n, filepath, sync, **kwargs):
 214     log("--- Test Backup #{:d} ---\n".format(n))
 215     target_id = "backup_target_{:d}".format(n)
 216     job_id = "backup_{:d}".format(n)
 217     kwargs.setdefault('auto-finalize', False)
 218     blockdev_backup_mktarget(drive, target_id, filepath, sync,
 219                              job_id=job_id, **kwargs)
 220     return job_id
 221
 222 def perform_writes(drive, n, filter_node_name=None):
 223     log("--- Write #{:d} ---\n".format(n))
 224     for pattern in GROUPS[n].patterns:
 225         cmd = "write -P{:s} 0x{:07x} 0x{:x}".format(
 226             pattern.byte,
 227             pattern.offset,
 228             pattern.size)
 229         log(cmd)
 230         log(drive.vm.hmp_qemu_io(filter_node_name or drive.node, cmd))
 231     bitmaps = drive.vm.query_bitmaps()
 232     log({'bitmaps': bitmaps}, indent=2)
 233     log('')
 234     return bitmaps
 235
 236
 237 def compare_images(image, reference, baseimg=None, expected_match=True):
 238     """
 239     Print a nice human-readable message comparing these images.
 240     """
 241     expected_ret = 0 if expected_match else 1
 242     if baseimg:
 243         assert qemu_img("rebase", "-u", "-b", baseimg, '-F', iotests.imgfmt,
 244                         image) == 0
 245     ret = qemu_img("compare", image, reference)
 246     log('qemu_img compare "{:s}" "{:s}" ==> {:s}, {:s}'.format(
 247         image, reference,
 248         "Identical" if ret == 0 else "Mismatch",
 249         "OK!" if ret == expected_ret else "ERROR!"),
 250         filters=[iotests.filter_testfiles])
 251
 252 def test_bitmap_sync(bsync_mode, msync_mode='bitmap', failure=None):
 253     """
 254     Test bitmap backup routines.
 255
 256     :param bsync_mode: Is the Bitmap Sync mode, and can be any of:
 257         - on-success: This is the "incremental" style mode. Bitmaps are
 258                       synchronized to what was copied out only on success.
 259                       (Partial images must be discarded.)
 260         - never:      This is the "differential" style mode.
 261                       Bitmaps are never synchronized.
 262         - always:     This is a "best effort" style mode.
 263                       Bitmaps are always synchronized, regardless of failure.
 264                       (Partial images must be kept.)
 265
 266     :param msync_mode: The mirror sync mode to use for the first backup.
 267                        Can be any one of:
 268         - bitmap: Backups based on bitmap manifest.
 269         - full:   Full backups.
 270         - top:    Full backups of the top layer only.
 271
 272     :param failure: Is the (optional) failure mode, and can be any of:
 273         - None:         No failure. Test the normative path. Default.
 274         - simulated:    Cancel the job right before it completes.
 275                         This also tests writes "during" the job.
 276         - intermediate: This tests a job that fails mid-process and produces
 277                         an incomplete backup. Testing limitations prevent
 278                         testing competing writes.
 279     """
 280     with iotests.FilePath(
 281             'img', 'bsync1', 'bsync2', 'fbackup0', 'fbackup1', 'fbackup2') as \
 282             (img_path, bsync1, bsync2, fbackup0, fbackup1, fbackup2), \
 283          iotests.VM() as vm:
 284
 285         mode = "Mode {:s}; Bitmap Sync {:s}".format(msync_mode, bsync_mode)
 286         preposition = "with" if failure else "without"
 287         cond = "{:s} {:s}".format(preposition,
 288                                   "{:s} failure".format(failure) if failure
 289                                   else "failure")
 290         log("\n=== {:s} {:s} ===\n".format(mode, cond))
 291
 292         log('--- Preparing image & VM ---\n')
 293         drive0 = Drive(img_path, vm=vm)
 294         drive0.img_create(iotests.imgfmt, SIZE)
 295         vm.add_device("{},id=scsi0".format('virtio-scsi'))
 296         vm.launch()
 297
 298         file_config = {
 299             'driver': 'file',
 300             'filename': drive0.path
 301         }
 302
 303         if failure == 'intermediate':
 304             file_config = {
 305                 'driver': 'blkdebug',
 306                 'image': file_config,
 307                 'set-state': [{
 308                     'event': 'flush_to_disk',
 309                     'state': 1,
 310                     'new_state': 2
 311                 }, {
 312                     'event': 'read_aio',
 313                     'state': 2,
 314                     'new_state': 3
 315                 }],
 316                 'inject-error': [{
 317                     'event': 'read_aio',
 318                     'errno': 5,
 319                     'state': 3,
 320                     'immediately': False,
 321                     'once': True
 322                 }]
 323             }
 324
 325         drive0.node = 'drive0'
 326         vm.qmp_log('blockdev-add',
 327                    filters=[iotests.filter_qmp_testfiles],
 328                    node_name=drive0.node,
 329                    driver=drive0.fmt,
 330                    file=file_config)
 331         log('')
 332
 333         # 0 - Writes and Reference Backup
 334         perform_writes(drive0, 0)
 335         reference_backup(drive0, 0, fbackup0)
 336         log('--- Add Bitmap ---\n')
 337         vm.qmp_log("block-dirty-bitmap-add", node=drive0.node,
 338                    name="bitmap0", granularity=GRANULARITY)
 339         log('')
 340         ebitmap = EmulatedBitmap()
 341
 342         # 1 - Writes and Reference Backup
 343         bitmaps = perform_writes(drive0, 1)
 344         ebitmap.dirty_group(1)
 345         bitmap = vm.get_bitmap(drive0.node, 'bitmap0', bitmaps=bitmaps)
 346         ebitmap.compare(bitmap)
 347         reference_backup(drive0, 1, fbackup1)
 348
 349         # 1 - Test Backup (w/ Optional induced failure)
 350         if failure == 'intermediate':
 351             # Activate blkdebug induced failure for second-to-next read
 352             log(vm.hmp_qemu_io(drive0.node, 'flush'))
 353             log('')
 354         job = backup(drive0, 1, bsync1, msync_mode,
 355                      bitmap="bitmap0", bitmap_mode=bsync_mode)
 356
 357         def _callback():
 358             """Issue writes while the job is open to test bitmap divergence."""
 359             # Note: when `failure` is 'intermediate', this isn't called.
 360             log('')
 361             bitmaps = perform_writes(drive0, 2, filter_node_name='backup-top')
 362             # Named bitmap (static, should be unchanged)
 363             ebitmap.compare(vm.get_bitmap(drive0.node, 'bitmap0',
 364                                           bitmaps=bitmaps))
 365             # Anonymous bitmap (dynamic, shows new writes)
 366             anonymous = EmulatedBitmap()
 367             anonymous.dirty_group(2)
 368             anonymous.compare(vm.get_bitmap(drive0.node, '', recording=True,
 369                                             bitmaps=bitmaps))
 370
 371             # Simulate the order in which this will happen:
 372             # group 1 gets cleared first, then group two gets written.
 373             if ((bsync_mode == 'on-success' and not failure) or
 374                 (bsync_mode == 'always')):
 375                 ebitmap.clear()
 376             ebitmap.dirty_group(2)
 377
 378         vm.run_job(job, auto_dismiss=True, auto_finalize=False,
 379                    pre_finalize=_callback,
 380                    cancel=(failure == 'simulated'))
 381         bitmaps = vm.query_bitmaps()
 382         log({'bitmaps': bitmaps}, indent=2)
 383         log('')
 384
 385         if bsync_mode == 'always' and failure == 'intermediate':
 386             # TOP treats anything allocated as dirty, expect to see:
 387             if msync_mode == 'top':
 388                 ebitmap.dirty_group(0)
 389
 390             # We manage to copy one sector (one bit) before the error.
 391             ebitmap.clear_bit(ebitmap.first_bit)
 392
 393             # Full returns all bits set except what was copied/skipped
 394             if msync_mode == 'full':
 395                 fail_bit = ebitmap.first_bit
 396                 ebitmap.clear()
 397                 ebitmap.dirty_bits(range(fail_bit, SIZE // GRANULARITY))
 398
 399         ebitmap.compare(vm.get_bitmap(drive0.node, 'bitmap0', bitmaps=bitmaps))
 400
 401         # 2 - Writes and Reference Backup
 402         bitmaps = perform_writes(drive0, 3)
 403         ebitmap.dirty_group(3)
 404         ebitmap.compare(vm.get_bitmap(drive0.node, 'bitmap0', bitmaps=bitmaps))
 405         reference_backup(drive0, 2, fbackup2)
 406
 407         # 2 - Bitmap Backup (In failure modes, this is a recovery.)
 408         job = backup(drive0, 2, bsync2, "bitmap",
 409                      bitmap="bitmap0", bitmap_mode=bsync_mode)
 410         vm.run_job(job, auto_dismiss=True, auto_finalize=False)
 411         bitmaps = vm.query_bitmaps()
 412         log({'bitmaps': bitmaps}, indent=2)
 413         log('')
 414         if bsync_mode != 'never':
 415             ebitmap.clear()
 416         ebitmap.compare(vm.get_bitmap(drive0.node, 'bitmap0', bitmaps=bitmaps))
 417
 418         log('--- Cleanup ---\n')
 419         vm.qmp_log("block-dirty-bitmap-remove",
 420                    node=drive0.node, name="bitmap0")
 421         bitmaps = vm.query_bitmaps()
 422         log({'bitmaps': bitmaps}, indent=2)
 423         vm.shutdown()
 424         log('')
 425
 426         log('--- Verification ---\n')
 427         # 'simulated' failures will actually all pass here because we canceled
 428         # while "pending". This is actually undefined behavior,
 429         # don't rely on this to be true!
 430         compare_images(bsync1, fbackup1, baseimg=fbackup0,
 431                        expected_match=failure != 'intermediate')
 432         if not failure or bsync_mode == 'always':
 433             # Always keep the last backup on success or when using 'always'
 434             base = bsync1
 435         else:
 436             base = fbackup0
 437         compare_images(bsync2, fbackup2, baseimg=base)
 438         compare_images(img_path, fbackup2)
 439         log('')
 440
 441 def test_backup_api():
 442     """
 443     Test malformed and prohibited invocations of the backup API.
 444     """
 445     with iotests.FilePath('img', 'bsync1') as (img_path, backup_path), \
 446          iotests.VM() as vm:
 447
 448         log("\n=== API failure tests ===\n")
 449         log('--- Preparing image & VM ---\n')
 450         drive0 = Drive(img_path, vm=vm)
 451         drive0.img_create(iotests.imgfmt, SIZE)
 452         vm.add_device("{},id=scsi0".format('virtio-scsi'))
 453         vm.launch()
 454
 455         file_config = {
 456             'driver': 'file',
 457             'filename': drive0.path
 458         }
 459
 460         drive0.node = 'drive0'
 461         vm.qmp_log('blockdev-add',
 462                    filters=[iotests.filter_qmp_testfiles],
 463                    node_name=drive0.node,
 464                    driver=drive0.fmt,
 465                    file=file_config)
 466         log('')
 467
 468         target0 = Drive(backup_path, vm=vm)
 469         target0.create_target("backup_target", drive0.fmt, drive0.size)
 470         log('')
 471
 472         vm.qmp_log("block-dirty-bitmap-add", node=drive0.node,
 473                    name="bitmap0", granularity=GRANULARITY)
 474         log('')
 475
 476         log('-- Testing invalid QMP commands --\n')
 477
 478         error_cases = {
 479             'incremental': {
 480                 None:        ['on-success', 'always', 'never', None],
 481                 'bitmap404': ['on-success', 'always', 'never', None],
 482                 'bitmap0':   ['always', 'never']
 483             },
 484             'bitmap': {
 485                 None:        ['on-success', 'always', 'never', None],
 486                 'bitmap404': ['on-success', 'always', 'never', None],
 487                 'bitmap0':   [None],
 488             },
 489             'full': {
 490                 None:        ['on-success', 'always', 'never'],
 491                 'bitmap404': ['on-success', 'always', 'never', None],
 492                 'bitmap0':   ['never', None],
 493             },
 494             'top': {
 495                 None:        ['on-success', 'always', 'never'],
 496                 'bitmap404': ['on-success', 'always', 'never', None],
 497                 'bitmap0':   ['never', None],
 498             },
 499             'none': {
 500                 None:        ['on-success', 'always', 'never'],
 501                 'bitmap404': ['on-success', 'always', 'never', None],
 502                 'bitmap0':   ['on-success', 'always', 'never', None],
 503             }
 504         }
 505
 506         # Dicts, as always, are not stably-ordered prior to 3.7, so use tuples:
 507         for sync_mode in ('incremental', 'bitmap', 'full', 'top', 'none'):
 508             log("-- Sync mode {:s} tests --\n".format(sync_mode))
 509             for bitmap in (None, 'bitmap404', 'bitmap0'):
 510                 for policy in error_cases[sync_mode][bitmap]:
 511                     blockdev_backup(drive0.vm, drive0.node, "backup_target",
 512                                     sync_mode, job_id='api_job',
 513                                     bitmap=bitmap, bitmap_mode=policy)
 514                     log('')
 515
 516
 517 def main():
 518     for bsync_mode in ("never", "on-success", "always"):
 519         for failure in ("simulated", "intermediate", None):
 520             test_bitmap_sync(bsync_mode, "bitmap", failure)
 521
 522     for sync_mode in ('full', 'top'):
 523         for bsync_mode in ('on-success', 'always'):
 524             for failure in ('simulated', 'intermediate', None):
 525                 test_bitmap_sync(bsync_mode, sync_mode, failure)
 526
 527     test_backup_api()
 528
 529 if __name__ == '__main__':
 530     iotests.script_main(main, supported_fmts=['qcow2'],
 531                         supported_protocols=['file'])