summaryrefslogtreecommitdiff
path: root/bmaptools/BmapCopy.py
blob: dba24a5216ac9862f44ab6bd1e0d29d160eb31f5 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
# -*- coding: utf-8 -*-
# vim: ts=4 sw=4 et ai si
#
# Copyright (c) 2012-2014 Intel, Inc.
# License: GPLv2
# Author: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License, version 2,
# as published by the Free Software Foundation.
#
# 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.

"""
This module implements copying of images with bmap and provides the following
API.
  1. BmapCopy class - implements copying to any kind of file, be that a block
     device or a regular file.
  2. BmapBdevCopy class - based on BmapCopy and specializes on copying to block
     devices. It does some more sanity checks and some block device performance
     tuning.

The bmap file is an XML file which contains a list of mapped blocks of the
image. Mapped blocks are the blocks which have disk sectors associated with
them, as opposed to holes, which are blocks with no associated disk sectors. In
other words, the image is considered to be a sparse file, and bmap basically
contains a list of mapped blocks of this sparse file. The bmap additionally
contains some useful information like block size (usually 4KiB), image size,
mapped blocks count, etc.

The bmap is used for copying the image to a block device or to a regular file.
The idea is that we copy quickly with bmap because we copy only mapped blocks
and ignore the holes, because they are useless. And if the image is generated
properly (starting with a huge hole and writing all the data), it usually
contains only little mapped blocks, comparing to the overall image size. And
such an image compresses very well (because holes are read as all zeroes), so
it is beneficial to distributor them as compressed files along with the bmap.

Here is an example. Suppose you have a 4GiB image which contains only 100MiB of
user data and you need to flash it to a slow USB stick. With bmap you end up
copying only a little bit more than 100MiB of data from the image to the USB
stick (namely, you copy only mapped blocks). This is a lot faster than copying
all 4GiB of data. We say that it is a bit more than 100MiB because things like
file-system meta-data (inode tables, superblocks, etc), partition table, etc
also contribute to the mapped blocks and are also copied.
"""

# Disable the following pylint recommendations:
#   * Too many instance attributes (R0902)
#   * Too many arguments - R0913
#   * Too many statements (R0915)
# pylint: disable=R0902
# pylint: disable=R0913
# pylint: disable=R0915

import os
import stat
import sys
import hashlib
import logging
import datetime
from six import reraise
from six.moves import queue as Queue
from six.moves import _thread as thread
from xml.etree import ElementTree
from bmaptools.BmapHelpers import human_size

_log = logging.getLogger(__name__)  # pylint: disable=C0103

# The highest supported bmap format version
SUPPORTED_BMAP_VERSION = "2.0"


class Error(Exception):
    """
    A class for exceptions generated by the 'BmapCopy' module. We currently
    support only one type of exceptions, and we basically throw human-readable
    problem description in case of errors.
    """
    pass


class BmapCopy(object):
    """
    This class implements the bmap-based copying functionality. To copy an
    image with bmap you should create an instance of this class, which requires
    the following:

    * full path or a file-like object of the image to copy
    * full path or a file object of the destination file copy the image to
    * full path or a file object of the bmap file (optional)
    * image size in bytes (optional)

    Although the main purpose of this class is to use bmap, the bmap is not
    required, and if it was not provided then the entire image will be copied
    to the destination file.

    When the bmap is provided, it is not necessary to specify image size,
    because the size is contained in the bmap. Otherwise, it is benefitial to
    specify the size because it enables extra sanity checks and makes it
    possible to provide the progress bar.

    When the image size is known either from the bmap or the caller specified
    it to the class constructor, all the image geometry description attributes
    ('blocks_cnt', etc) are initialized by the class constructor and available
    for the user.

    However, when the size is not known, some of  the image geometry
    description attributes are not initialized by the class constructor.
    Instead, they are initialized only by the 'copy()' method.

    The 'copy()' method implements image copying. You may choose whether to
    verify the checksum while copying or not. Note, this is done only in case
    of bmap-based copying and only if bmap contains checksums (e.g., bmap
    version 1.0 did not have checksums support).

    You may choose whether to synchronize the destination file after writing or
    not. To explicitly synchronize it, use the 'sync()' method.

    This class supports all the bmap format versions up version
    'SUPPORTED_BMAP_VERSION'.

    It is possible to have a simple progress indicator while copying the image.
    Use the 'set_progress_indicator()' method.

    You can copy only once with an instance of this class. This means that in
    order to copy the image for the second time, you have to create a new class
    instance.
    """

    def __init__(self, image, dest, bmap=None, image_size=None):
        """
        The class constructor. The parameters are:
            image      - file-like object of the image which should be copied,
                         should only support 'read()' and 'seek()' methods,
                         and only seeking forward has to be supported.
            dest       - file object of the destination file to copy the image
                         to.
            bmap       - file object of the bmap file to use for copying.
            image_size - size of the image in bytes.
        """

        self._xml = None

        self._dest_fsync_watermark = None
        self._batch_blocks = None
        self._batch_queue = None
        self._batch_bytes = 1024 * 1024
        self._batch_queue_len = 6

        self.bmap_version = None
        self.bmap_version_major = None
        self.bmap_version_minor = None
        self.block_size = None
        self.blocks_cnt = None
        self.mapped_cnt = None
        self.image_size = None
        self.image_size_human = None
        self.mapped_size = None
        self.mapped_size_human = None
        self.mapped_percent = None

        self._f_bmap = None
        self._f_bmap_path = None

        self._progress_started = None
        self._progress_index = None
        self._progress_time = None
        self._progress_file = None
        self._progress_format = None
        self.set_progress_indicator(None, None)
        self._psplash_pipe = None

        self._f_image = image
        self._image_path = image.name

        self._f_dest = dest
        self._dest_path = dest.name
        st_data = os.fstat(self._f_dest.fileno())
        self._dest_is_regfile = stat.S_ISREG(st_data.st_mode)

        # The bmap file checksum type and length
        self._cs_type = None
        self._cs_len = None
        self._cs_attrib_name = None
        self._bmap_cs_attrib_name = None

        # Special quirk for /dev/null which does not support fsync()
        if stat.S_ISCHR(st_data.st_mode) and \
           os.major(st_data.st_rdev) == 1 and \
           os.minor(st_data.st_rdev) == 3:
            self._dest_supports_fsync = False
        else:
            self._dest_supports_fsync = True

        if bmap:
            self._f_bmap = bmap
            self._bmap_path = bmap.name
            self._parse_bmap()
        else:
            # There is no bmap. Initialize user-visible attributes to something
            # sensible with an assumption that we just have all blocks mapped.
            self.bmap_version = 0
            self.block_size = 4096
            self.mapped_percent = 100

        if image_size:
            self._set_image_size(image_size)

        self._batch_blocks = self._batch_bytes // self.block_size

    def set_psplash_pipe(self, path):
        """
        Set the psplash named pipe file path to be used when updating the
        progress - best effort.

        The 'path' argument is the named pipe used by the psplash process to get
        progress bar commands. When the path argument doesn't exist or is not a
        pipe, the function will ignore with a warning. This behavior is
        considered as such because the progress is considered a decoration
        functionality which might or might not be available even if requested.
        When used as a boot service, the unavailability of the psplash service
        (due to various reasons: no screen, racing issues etc.) should not
        break the writting process. This is why this implementation is done as
        a best effort.
        """

        if os.path.exists(path) and stat.S_ISFIFO(os.stat(path).st_mode):
            self._psplash_pipe = path
        else:
            _log.warning("'%s' is not a pipe, so psplash progress will not be "
                         "updated" % path)

    def set_progress_indicator(self, file_obj, format_string):
        """
        Setup the progress indicator which shows how much data has been copied
        in percent.

        The 'file_obj' argument is the console file object where the progress
        has to be printed to. Pass 'None' to disable the progress indicator.

        The 'format_string' argument is the format string for the progress
        indicator. It has to contain a single '%d' placeholder which will be
        substitutes with copied data in percent.
        """

        self._progress_file = file_obj
        if format_string:
            self._progress_format = format_string
        else:
            self._progress_format = "Copied %d%%"

    def _set_image_size(self, image_size):
        """
        Set image size and initialize various other geometry-related attributes.
        """

        if self.image_size is not None and self.image_size != image_size:
            raise Error("cannot set image size to %d bytes, it is known to "
                        "be %d bytes (%s)" % (image_size, self.image_size,
                                              self.image_size_human))

        self.image_size = image_size
        self.image_size_human = human_size(image_size)
        self.blocks_cnt = (self.image_size + self.block_size - 1) // self.block_size

        if self.mapped_cnt is None:
            self.mapped_cnt = self.blocks_cnt
            self.mapped_size = self.image_size
            self.mapped_size_human = self.image_size_human

    def _verify_bmap_checksum(self):
        """
        This is a helper function which verifies the bmap file checksum.
        """

        import mmap

        correct_chksum = self._xml.find(self._bmap_cs_attrib_name).text.strip()

        # Before verifying the shecksum, we have to substitute the checksum
        # value stored in the file with all zeroes. For these purposes we
        # create private memory mapping of the bmap file.
        mapped_bmap = mmap.mmap(self._f_bmap.fileno(), 0,
                                access=mmap.ACCESS_COPY)

        chksum_pos = mapped_bmap.find(correct_chksum.encode())
        assert chksum_pos != -1

        mapped_bmap[chksum_pos:chksum_pos + self._cs_len] = b'0' * self._cs_len

        hash_obj = hashlib.new(self._cs_type)
        hash_obj.update(mapped_bmap)
        calculated_chksum = hash_obj.hexdigest()

        mapped_bmap.close()

        if calculated_chksum != correct_chksum:
            raise Error("checksum mismatch for bmap file '%s': calculated "
                        "'%s', should be '%s'"
                        % (self._bmap_path, calculated_chksum, correct_chksum))

    def _parse_bmap(self):
        """
        Parse the bmap file and initialize corresponding class instance attributs.
        """

        try:
            self._xml = ElementTree.parse(self._f_bmap)
        except ElementTree.ParseError as err:
            # Extrace the erroneous line with some context
            self._f_bmap.seek(0)
            xml_extract = ""
            for num, line in enumerate(self._f_bmap):
                if num >= err.position[0] - 4 and num <= err.position[0] + 4:
                    xml_extract += "Line %d: %s" % (num, line)

            raise Error("cannot parse the bmap file '%s' which should be a "
                        "proper XML file: %s, the XML extract:\n%s" %
                        (self._bmap_path, err, xml_extract))

        xml = self._xml
        self.bmap_version = str(xml.getroot().attrib.get('version'))

        # Make sure we support this version
        self.bmap_version_major = int(self.bmap_version.split('.', 1)[0])
        self.bmap_version_minor = int(self.bmap_version.split('.', 1)[1])
        if self.bmap_version_major > int(SUPPORTED_BMAP_VERSION.split('.', 1)[0]):
            raise Error("only bmap format version up to %d is supported, "
                        "version %d is not supported"
                        % (SUPPORTED_BMAP_VERSION, self.bmap_version_major))

        # Fetch interesting data from the bmap XML file
        self.block_size = int(xml.find("BlockSize").text.strip())
        self.blocks_cnt = int(xml.find("BlocksCount").text.strip())
        self.mapped_cnt = int(xml.find("MappedBlocksCount").text.strip())
        self.image_size = int(xml.find("ImageSize").text.strip())
        self.image_size_human = human_size(self.image_size)
        self.mapped_size = self.mapped_cnt * self.block_size
        self.mapped_size_human = human_size(self.mapped_size)
        self.mapped_percent = (self.mapped_cnt * 100.0) / self.blocks_cnt

        blocks_cnt = (self.image_size + self.block_size - 1) // self.block_size
        if self.blocks_cnt != blocks_cnt:
            raise Error("Inconsistent bmap - image size does not match "
                        "blocks count (%d bytes != %d blocks * %d bytes)"
                        % (self.image_size, self.blocks_cnt, self.block_size))

        if self.bmap_version_major > 1 or \
           (self.bmap_version_major == 1 and self.bmap_version_minor == 4):
            # In bmap format version 1.0-1.3 the only supported checksum type
            # was SHA1. Version 2.0 started supporting arbitrary checksum
            # types. A new "ChecksumType" tag was introduce to specify the
            # checksum function name. And all XML tags which contained "sha1"
            # in their name were renamed to something more neutral. This was an
            # change incompatible with previous formats.
            #
            # There is a special format version 1.4, which should not have been
            # ever issued, but was released by a mistake. The mistake was that
            # when implementing version 2.0 support we mistakenly gave it
            # version number 1.4. This was later on fixed and format version
            # 1.4 became version 2.0. So 1.4 and 2.0 formats are identical.
            #
            # Note, bmap files did not contain checksums prior to version 1.3.
            self._cs_type = xml.find("ChecksumType").text.strip()
            self._cs_attrib_name = "chksum"
            self._bmap_cs_attrib_name = "BmapFileChecksum"
        elif self.bmap_version_minor == 3:
            self._cs_type = "sha1"
            self._cs_attrib_name = "sha1"
            self._bmap_cs_attrib_name = "BmapFileSHA1"

        if self._cs_type:
            try:
                self._cs_len = len(hashlib.new(self._cs_type).hexdigest())
            except ValueError as err:
                raise Error("cannot initialize hash function \"%s\": %s" %
                            (self._cs_type, err))
            self._verify_bmap_checksum()

    def _update_progress(self, blocks_written):
        """
        Print the progress indicator if the mapped area size is known and if
        the indicator has been enabled by assigning a console file object to
        the 'progress_file' attribute.
        """

        if self.mapped_cnt:
            assert blocks_written <= self.mapped_cnt
            percent = int((float(blocks_written) / self.mapped_cnt) * 100)
            _log.debug("wrote %d blocks out of %d (%d%%)" %
                       (blocks_written, self.mapped_cnt, percent))
        else:
            _log.debug("wrote %d blocks" % blocks_written)

        if not self._progress_file:
            return

        if self.mapped_cnt:
            progress = '\r' + self._progress_format % percent + '\n'
        else:
            # Do not rotate the wheel too fast
            now = datetime.datetime.now()
            min_delta = datetime.timedelta(milliseconds=250)
            if now - self._progress_time < min_delta:
                return
            self._progress_time = now

            progress_wheel = ('-', '\\', '|', '/')
            progress = '\r' + progress_wheel[self._progress_index % 4] + '\n'
            self._progress_index += 1

        # This is a little trick we do in order to make sure that the next
        # message will always start from a new line - we switch to the new
        # line after each progress update and move the cursor up. As an
        # example, this is useful when the copying is interrupted by an
        # exception - the error message will start form new line.
        if self._progress_started:
            # The "move cursor up" escape sequence
            self._progress_file.write('\033[1A')  # pylint: disable=W1401
        else:
            self._progress_started = True

        self._progress_file.write(progress)
        self._progress_file.flush()

        # Update psplash progress when configured. This is using a best effort
        # strategy to not affect the writing process when psplash breaks, is
        # not available early enough or screen is not available.
        if self._psplash_pipe and self.mapped_cnt:
            try:
                mode = os.O_WRONLY | os.O_NONBLOCK
                with os.fdopen(os.open(self._psplash_pipe, mode), 'w') as p_fo:
                    p_fo.write("PROGRESS %d\n" % percent)
            except:
                pass

    def _get_block_ranges(self):
        """
        This is a helper generator that parses the bmap XML file and for each
        block range in the XML file it yields ('first', 'last', 'chksum')
        tuples, where:
          * 'first' is the first block of the range;
          * 'last' is the last block of the range;
          * 'chksum' is the checksum of the range ('None' is used if it is
            missing).

        If there is no bmap file, the generator just yields a single range
        for entire image file. If the image size is unknown, the generator
        infinitely yields continuous ranges of size '_batch_blocks'.
        """

        if not self._f_bmap:
            # We do not have the bmap, yield a tuple with all blocks
            if self.blocks_cnt:
                yield (0, self.blocks_cnt - 1, None)
            else:
                # We do not know image size, keep yielding tuples with many
                # blocks infinitely.
                first = 0
                while True:
                    yield (first, first + self._batch_blocks - 1, None)
                    first += self._batch_blocks
            return

        # We have the bmap, just read it and yield block ranges
        xml = self._xml
        xml_bmap = xml.find("BlockMap")

        for xml_element in xml_bmap.findall("Range"):
            blocks_range = xml_element.text.strip()
            # The range of blocks has the "X - Y" format, or it can be just "X"
            # in old bmap format versions. First, split the blocks range string
            # and strip white-spaces.
            split = [x.strip() for x in blocks_range.split('-', 1)]

            first = int(split[0])
            if len(split) > 1:
                last = int(split[1])
                if first > last:
                    raise Error("bad range (first > last): '%s'" % blocks_range)
            else:
                last = first

            if self._cs_attrib_name in xml_element.attrib:
                chksum = xml_element.attrib[self._cs_attrib_name]
            else:
                chksum = None

            yield (first, last, chksum)

    def _get_batches(self, first, last):
        """
        This is a helper generator which splits block ranges from the bmap file
        to smaller batches. Indeed, we cannot read and write entire block
        ranges from the image file, because a range can be very large. So we
        perform the I/O in batches. Batch size is defined by the
        '_batch_blocks' attribute. Thus, for each (first, last) block range,
        the generator yields smaller (start, end, length) batch ranges, where:
          * 'start' is the starting batch block number;
          * 'last' is the ending batch block number;
          * 'length' is the batch length in blocks (same as
             'end' - 'start' + 1).
        """

        batch_blocks = self._batch_blocks

        while first + batch_blocks - 1 <= last:
            yield (first, first + batch_blocks - 1, batch_blocks)
            first += batch_blocks

        batch_blocks = last - first + 1
        if batch_blocks:
            yield (first, first + batch_blocks - 1, batch_blocks)

    def _get_data(self, verify):
        """
        This is generator  which reads the image file in '_batch_blocks' chunks
        and yields ('type', 'start', 'end',  'buf) tuples, where:
          * 'start' is the starting block number of the batch;
          * 'end' is the last block of the batch;
          * 'buf' a buffer containing the batch data.
        """

        _log.debug("the reader thread has started")
        try:
            for (first, last, chksum) in self._get_block_ranges():
                if verify and chksum:
                    hash_obj = hashlib.new(self._cs_type)

                self._f_image.seek(first * self.block_size)

                iterator = self._get_batches(first, last)
                for (start, end, length) in iterator:
                    try:
                        buf = self._f_image.read(length * self.block_size)
                    except IOError as err:
                        raise Error("error while reading blocks %d-%d of the "
                                    "image file '%s': %s"
                                    % (start, end, self._image_path, err))

                    if not buf:
                        _log.debug("no more data to read from file '%s'",
                                   self._image_path)
                        self._batch_queue.put(None)
                        return

                    if verify and chksum:
                        hash_obj.update(buf)

                    blocks = (len(buf) + self.block_size - 1) // self.block_size
                    _log.debug("queueing %d blocks, queue length is %d" %
                               (blocks, self._batch_queue.qsize()))

                    self._batch_queue.put(("range", start, start + blocks - 1,
                                           buf))

                if verify and chksum and hash_obj.hexdigest() != chksum:
                    raise Error("checksum mismatch for blocks range %d-%d: "
                                "calculated %s, should be %s (image file %s)"
                                % (first, last, hash_obj.hexdigest(),
                                   chksum, self._image_path))
        # Silence pylint warning about catching too general exception
        # pylint: disable=W0703
        except Exception:
            # pylint: enable=W0703
            # In case of any exception - just pass it to the main thread
            # through the queue.
            self._batch_queue.put(("error", sys.exc_info()))

        self._batch_queue.put(None)

    def copy(self, sync=True, verify=True):
        """
        Copy the image to the destination file using bmap. The 'sync' argument
        defines whether the destination file has to be synchronized upon
        return.  The 'verify' argument defines whether the checksum has to be
        verified while copying.
        """

        # Create the queue for block batches and start the reader thread, which
        # will read the image in batches and put the results to '_batch_queue'.
        self._batch_queue = Queue.Queue(self._batch_queue_len)
        thread.start_new_thread(self._get_data, (verify, ))

        blocks_written = 0
        bytes_written = 0
        fsync_last = 0

        self._progress_started = False
        self._progress_index = 0
        self._progress_time = datetime.datetime.now()

        if self.image_size and self._dest_is_regfile:
            # If we already know image size, make sure that destination file
            # has the same size as the image
            try:
                os.ftruncate(self._f_dest.fileno(), self.image_size)
            except OSError as err:
                raise Error("cannot truncate file '%s': %s"
                            % (self._dest_path, err))

        # Read the image in '_batch_blocks' chunks and write them to the
        # destination file
        while True:
            batch = self._batch_queue.get()
            if batch is None:
                # No more data, the image is written
                break
            elif batch[0] == "error":
                # The reader thread encountered an error and passed us the
                # exception.
                exc_info = batch[1]
                reraise(exc_info[0], exc_info[1], exc_info[2])

            (start, end, buf) = batch[1:4]

            assert len(buf) <= (end - start + 1) * self.block_size
            assert len(buf) > (end - start) * self.block_size

            self._f_dest.seek(start * self.block_size)

            # Synchronize the destination file if we reached the watermark
            if self._dest_fsync_watermark:
                if blocks_written >= fsync_last + self._dest_fsync_watermark:
                    fsync_last = blocks_written
                    self.sync()

            try:
                self._f_dest.write(buf)
            except IOError as err:
                raise Error("error while writing blocks %d-%d of '%s': %s"
                            % (start, end, self._dest_path, err))

            self._batch_queue.task_done()
            blocks_written += (end - start + 1)
            bytes_written += len(buf)

            self._update_progress(blocks_written)

        if not self.image_size:
            # The image size was unknown up until now, set it
            self._set_image_size(bytes_written)

        # This is just a sanity check - we should have written exactly
        # 'mapped_cnt' blocks.
        if blocks_written != self.mapped_cnt:
            raise Error("wrote %u blocks from image '%s' to '%s', but should "
                        "have %u - bmap file '%s' does not belong to this "
                        "image"
                        % (blocks_written, self._image_path, self._dest_path,
                           self.mapped_cnt, self._bmap_path))

        if self._dest_is_regfile:
            # Make sure the destination file has the same size as the image
            try:
                os.ftruncate(self._f_dest.fileno(), self.image_size)
            except OSError as err:
                raise Error("cannot truncate file '%s': %s"
                            % (self._dest_path, err))

        try:
            self._f_dest.flush()
        except IOError as err:
            raise Error("cannot flush '%s': %s" % (self._dest_path, err))

        if sync:
            self.sync()

    def sync(self):
        """
        Synchronize the destination file to make sure all the data are actually
        written to the disk.
        """

        if self._dest_supports_fsync:
            try:
                os.fsync(self._f_dest.fileno()),
            except OSError as err:
                raise Error("cannot synchronize '%s': %s "
                            % (self._dest_path, err.strerror))


class BmapBdevCopy(BmapCopy):
    """
    This class is a specialized version of 'BmapCopy' which copies the image to
    a block device. Unlike the base 'BmapCopy' class, this class does various
    optimizations specific to block devices, e.g., switching to the 'noop' I/O
    scheduler.
    """

    def __init__(self, image, dest, bmap=None, image_size=None):
        """
        The same as the constructor of the 'BmapCopy' base class, but adds
        useful guard-checks specific to block devices.
        """

        # Call the base class constructor first
        BmapCopy.__init__(self, image, dest, bmap, image_size)

        self._dest_fsync_watermark = (6 * 1024 * 1024) // self.block_size

        self._sysfs_base = None
        self._sysfs_scheduler_path = None
        self._sysfs_max_ratio_path = None
        self._old_scheduler_value = None
        self._old_max_ratio_value = None

        # If the image size is known, check that it fits the block device
        if self.image_size:
            try:
                bdev_size = os.lseek(self._f_dest.fileno(), 0, os.SEEK_END)
                os.lseek(self._f_dest.fileno(), 0, os.SEEK_SET)
            except OSError as err:
                raise Error("cannot seed block device '%s': %s "
                            % (self._dest_path, err.strerror))

            if bdev_size < self.image_size:
                raise Error("the image file '%s' has size %s and it will not "
                            "fit the block device '%s' which has %s capacity"
                            % (self._image_path, self.image_size_human,
                               self._dest_path, human_size(bdev_size)))

        # Construct the path to the sysfs directory of our block device
        st_rdev = os.fstat(self._f_dest.fileno()).st_rdev
        self._sysfs_base = "/sys/dev/block/%s:%s/" % \
                           (os.major(st_rdev), os.minor(st_rdev))

        # Check if the 'queue' sub-directory exists. If yes, then our block
        # device is entire disk. Otherwise, it is a partition, in which case we
        # need to go one level up in the sysfs hierarchy.
        if not os.path.exists(self._sysfs_base + "queue"):
            self._sysfs_base = self._sysfs_base + "../"

        self._sysfs_scheduler_path = self._sysfs_base + "queue/scheduler"
        self._sysfs_max_ratio_path = self._sysfs_base + "bdi/max_ratio"

    def _tune_block_device(self):
        """
        Tune the block device for better performance:
        1. Switch to the 'noop' I/O scheduler if it is available - sequential
           write to the block device becomes a lot faster comparing to CFQ.
        2. Limit the write buffering - we do not need the kernel to buffer a
           lot of the data we send to the block device, because we write
           sequentially. Limit the buffering.

        The old settings are saved in order to be able to restore them later.
        """
        # Switch to the 'noop' I/O scheduler
        try:
            with open(self._sysfs_scheduler_path, "r+") as f_scheduler:
                contents = f_scheduler.read()
                f_scheduler.seek(0)
                f_scheduler.write("noop")
        except IOError as err:
            _log.debug("failed to enable I/O optimization, expect "
                         "suboptimal speed (reason: cannot switch to the "
                         "'noop' I/O scheduler: %s or blk-mq in use)" % err)
        else:
            # The file contains a list of schedulers with the current
            # scheduler in square brackets, e.g., "noop deadline [cfq]".
            # Fetch the name of the current scheduler.
            import re

            match = re.match(r'.*\[(.+)\].*', contents)
            if match:
                self._old_scheduler_value = match.group(1)

        # Limit the write buffering, because we do not need too much of it when
        # writing sequntially. Excessive buffering makes some systems not very
        # responsive, e.g., this was observed in Fedora 17.
        try:
            with open(self._sysfs_max_ratio_path, "r+") as f_ratio:
                self._old_max_ratio_value = f_ratio.read()
                f_ratio.seek(0)
                f_ratio.write("1")
        except IOError as err:
            _log.warning("failed to disable excessive buffering, expect "
                         "worse system responsiveness (reason: cannot set "
                         "max. I/O ratio to 1: %s)" % err)

    def _restore_bdev_settings(self):
        """
        Restore old block device settings which we changed in
        '_tune_block_device()'.
        """

        if self._old_scheduler_value is not None:
            try:
                with open(self._sysfs_scheduler_path, "w") as f_scheduler:
                    f_scheduler.write(self._old_scheduler_value)
            except IOError as err:
                raise Error("cannot restore the '%s' I/O scheduler: %s"
                            % (self._old_scheduler_value, err))

        if self._old_max_ratio_value is not None:
            try:
                with open(self._sysfs_max_ratio_path, "w") as f_ratio:
                    f_ratio.write(self._old_max_ratio_value)
            except IOError as err:
                raise Error("cannot set the max. I/O ratio back to '%s': %s"
                            % (self._old_max_ratio_value, err))

    def copy(self, sync=True, verify=True):
        """
        The same as in the base class but tunes the block device for better
        performance before starting writing. Additionally, it forces block
        device synchronization from time to time in order to make sure we do
        not get stuck in 'fsync()' for too long time. The problem is that the
        kernel synchronizes block devices when the file is closed. And the
        result is that if the user interrupts us while we are copying the data,
        the program will be blocked in 'close()' waiting for the block device
        synchronization, which may last minutes for slow USB stick. This is
        very bad user experience, and we work around this effect by
        synchronizing from time to time.
        """

        self._tune_block_device()

        try:
            BmapCopy.copy(self, sync, verify)
        except:
            raise
        finally:
            self._restore_bdev_settings()