summaryrefslogtreecommitdiffstats
path: root/gluster/swift/obj/diskfile.py
blob: fdfabb53b1b3ead5a626637ae17d2b3ecf3f50a9 (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
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
# Copyright (c) 2012-2013 Red Hat, Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
# implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import os
import stat
import fcntl
import errno
try:
    from random import SystemRandom
    random = SystemRandom()
except ImportError:
    import random
import logging
from socket import gethostname
from hashlib import md5
from eventlet import sleep
from greenlet import getcurrent
from contextlib import contextmanager
from swift.common.utils import TRUE_VALUES, drop_buffer_cache, ThreadPool
from swift.common.exceptions import DiskFileNotExist, DiskFileError, \
    DiskFileNoSpace, DiskFileDeviceUnavailable

from gluster.swift.common.exceptions import GlusterFileSystemOSError, \
    GlusterFileSystemIOError
from gluster.swift.common.Glusterfs import mount
from gluster.swift.common.fs_utils import do_fstat, do_open, do_close, \
    do_unlink, do_chown, os_path, do_fsync, do_fchown, do_stat, do_write, \
    do_fdatasync, do_rename, Fake_file
from gluster.swift.common.utils import read_metadata, write_metadata, \
    validate_object, create_object_metadata, rmobjdir, dir_is_object, \
    get_object_metadata
from gluster.swift.common.utils import X_CONTENT_LENGTH, X_CONTENT_TYPE, \
    X_TIMESTAMP, X_TYPE, X_OBJECT_TYPE, FILE, OBJECT, DIR_TYPE, \
    FILE_TYPE, DEFAULT_UID, DEFAULT_GID, DIR_NON_OBJECT, DIR_OBJECT, \
    X_ETAG
from ConfigParser import ConfigParser, NoSectionError, NoOptionError

from swift.obj.diskfile import DiskFile as SwiftDiskFile

# FIXME: Hopefully we'll be able to move to Python 2.7+ where O_CLOEXEC will
# be back ported. See http://www.python.org/dev/peps/pep-0433/
O_CLOEXEC = 02000000

DEFAULT_DISK_CHUNK_SIZE = 65536
DEFAULT_BYTES_PER_SYNC = (512 * 1024 * 1024)
# keep these lower-case
DISALLOWED_HEADERS = set('content-length content-type deleted etag'.split())

MAX_RENAME_ATTEMPTS = 10
MAX_OPEN_ATTEMPTS = 10

_cur_pid = str(os.getpid())
_cur_host = str(gethostname())


def _random_sleep():
    sleep(random.uniform(0.5, 0.15))


def _lock_parent(full_path):
    parent_path, _ = full_path.rsplit(os.path.sep, 1)
    try:
        fd = os.open(parent_path, os.O_RDONLY | O_CLOEXEC)
    except OSError as err:
        if err.errno == errno.ENOENT:
            # Cannot lock the parent because it does not exist, let the caller
            # handle this situation.
            return False
        raise
    else:
        while True:
            # Spin sleeping for 1/10th of a second until we get the lock.
            # FIXME: Consider adding a final timeout just abort the operation.
            try:
                fcntl.flock(fd, fcntl.LOCK_EX | fcntl.LOCK_NB)
            except IOError as err:
                if err.errno == errno.EAGAIN:
                    _random_sleep()
                else:
                    # Don't leak an open file on an exception
                    os.close(fd)
                    raise
            except Exception:
                # Don't leak an open file for any other exception
                os.close(fd)
                raise
            else:
                break
        return fd


def _make_directory_locked(full_path, uid, gid, metadata=None):
    fd = _lock_parent(full_path)
    if fd is False:
        # Parent does not exist either, pass this situation on to the caller
        # to handle.
        return False, metadata
    try:
        # Check for directory existence
        stats = do_stat(full_path)
        if stats:
            # It now exists, having acquired the lock of its parent directory,
            # but verify it is actually a directory
            is_dir = stat.S_ISDIR(stats.st_mode)
            if not is_dir:
                # It is not a directory!
                raise DiskFileError("_make_directory_locked: non-directory"
                                    " found at path %s when expecting a"
                                    " directory", full_path)
            return True, metadata

        # We know the parent directory exists, and we have it locked, attempt
        # the creation of the target directory.
        return _make_directory_unlocked(full_path, uid, gid, metadata=metadata)
    finally:
        # We're done here, be sure to remove our lock and close our open FD.
        try:
            fcntl.flock(fd, fcntl.LOCK_UN)
        except:
            pass
        os.close(fd)


def _make_directory_unlocked(full_path, uid, gid, metadata=None):
    """
    Make a directory and change the owner ship as specified, and potentially
    creating the object metadata if requested.
    """
    try:
        os.mkdir(full_path)
    except OSError as err:
        if err.errno == errno.ENOENT:
            # Tell the caller some directory of the parent path does not
            # exist.
            return False, metadata
        elif err.errno == errno.EEXIST:
            # Possible race, in that the caller invoked this method when it
            # had previously determined the file did not exist.
            #
            # FIXME: When we are confident, remove this stat() call as it is
            # not necessary.
            try:
                stats = os.stat(full_path)
            except OSError as serr:
                # FIXME: Ideally we'd want to return an appropriate error
                # message and code in the PUT Object REST API response.
                raise DiskFileError("_make_directory_unlocked: os.mkdir failed"
                                    " because path %s already exists, and"
                                    " a subsequent os.stat on that same"
                                    " path failed (%s)" % (full_path,
                                                           str(serr)))
            else:
                is_dir = stat.S_ISDIR(stats.st_mode)
                if not is_dir:
                    # FIXME: Ideally we'd want to return an appropriate error
                    # message and code in the PUT Object REST API response.
                    raise DiskFileError("_make_directory_unlocked: os.mkdir"
                                        " failed on path %s because it already"
                                        " exists but not as a directory" % (
                                            full_path))
            return True, metadata
        elif err.errno == errno.ENOTDIR:
            # FIXME: Ideally we'd want to return an appropriate error
            # message and code in the PUT Object REST API response.
            raise DiskFileError("_make_directory_unlocked: os.mkdir failed"
                                " because some part of path %s is not in fact"
                                " a directory" % (full_path))
        elif err.errno == errno.EIO:
            # Sometimes Fuse will return an EIO error when it does not know
            # how to handle an unexpected, but transient situation. It is
            # possible the directory now exists, stat() it to find out after a
            # short period of time.
            _random_sleep()
            try:
                stats = os.stat(full_path)
            except OSError as serr:
                if serr.errno == errno.ENOENT:
                    errmsg = "_make_directory_unlocked: os.mkdir failed on" \
                             " path %s (EIO), and a subsequent os.stat on" \
                             " that same path did not find the file." % (
                                 full_path,)
                else:
                    errmsg = "_make_directory_unlocked: os.mkdir failed on" \
                             " path %s (%s), and a subsequent os.stat on" \
                             " that same path failed as well (%s)" % (
                                 full_path, str(err), str(serr))
                raise DiskFileError(errmsg)
            else:
                # The directory at least exists now
                is_dir = stat.S_ISDIR(stats.st_mode)
                if is_dir:
                    # Dump the stats to the log with the original exception.
                    logging.warn("_make_directory_unlocked: os.mkdir initially"
                                 " failed on path %s (%s) but a stat()"
                                 " following that succeeded: %r" % (full_path,
                                                                    str(err),
                                                                    stats))
                    # Assume another entity took care of the proper setup.
                    return True, metadata
                else:
                    raise DiskFileError("_make_directory_unlocked: os.mkdir"
                                        " initially failed on path %s (%s) but"
                                        " now we see that it exists but is not"
                                        " a directory (%r)" % (full_path,
                                                               str(err),
                                                               stats))
        else:
            # Some other potentially rare exception occurred that does not
            # currently warrant a special log entry to help diagnose.
            raise DiskFileError("_make_directory_unlocked: os.mkdir failed on"
                                " path %s (%s)" % (full_path, str(err)))
    else:
        if metadata:
            # We were asked to set the initial metadata for this object.
            metadata_orig = get_object_metadata(full_path)
            metadata_orig.update(metadata)
            write_metadata(full_path, metadata_orig)
            metadata = metadata_orig

        # We created it, so we are reponsible for always setting the proper
        # ownership.
        do_chown(full_path, uid, gid)
        return True, metadata


_fs_conf = ConfigParser()
if _fs_conf.read(os.path.join('/etc/swift', 'fs.conf')):
    try:
        _mkdir_locking = _fs_conf.get('DEFAULT', 'mkdir_locking', "no") \
            in TRUE_VALUES
    except (NoSectionError, NoOptionError):
        _mkdir_locking = False
    try:
        _use_put_mount = _fs_conf.get('DEFAULT', 'use_put_mount', "no") \
            in TRUE_VALUES
    except (NoSectionError, NoOptionError):
        _use_put_mount = False
else:
    _mkdir_locking = False
    _use_put_mount = False

if _mkdir_locking:
    make_directory = _make_directory_locked
else:
    make_directory = _make_directory_unlocked


def _adjust_metadata(metadata):
    # Fix up the metadata to ensure it has a proper value for the
    # Content-Type metadata, as well as an X_TYPE and X_OBJECT_TYPE
    # metadata values.
    content_type = metadata.get(X_CONTENT_TYPE, '')

    if not content_type:
        # FIXME: How can this be that our caller supplied us with metadata
        # that has a content type that evaluates to False?
        #
        # FIXME: If the file exists, we would already know it is a
        # directory. So why are we assuming it is a file object?
        metadata[X_CONTENT_TYPE] = FILE_TYPE
        metadata[X_OBJECT_TYPE] = FILE
    else:
        if content_type.lower() == DIR_TYPE:
            metadata[X_OBJECT_TYPE] = DIR_OBJECT
        else:
            metadata[X_OBJECT_TYPE] = FILE

    metadata[X_TYPE] = OBJECT
    return metadata


class DiskWriter(object):
    """
    Encapsulation of the write context for servicing PUT REST API
    requests. Serves as the context manager object for DiskFile's writer()
    method.

    We just override the put() method for Gluster.
    """
    def __init__(self, disk_file, fd, tmppath, threadpool):
        self.disk_file = disk_file
        self.fd = fd
        self.tmppath = tmppath
        self.upload_size = 0
        self.last_sync = 0
        self.threadpool = threadpool

    def close(self):
        """
        Close the file descriptor
        """
        if self.fd:
            do_close(self.fd)
            self.fd = None

    def write(self, chunk):
        """
        Write a chunk of data into the temporary file.

        :param chunk: the chunk of data to write as a string object
        """

        def _write_entire_chunk(chunk):
            while chunk:
                written = do_write(self.fd, chunk)
                self.upload_size += written
                chunk = chunk[written:]

        self.threadpool.run_in_thread(_write_entire_chunk, chunk)

        # For large files sync every 512MB (by default) written
        diff = self.upload_size - self.last_sync
        if diff >= self.disk_file.bytes_per_sync:
            self.threadpool.force_run_in_thread(do_fdatasync, self.fd)
            drop_buffer_cache(self.fd, self.last_sync, diff)
            self.last_sync = self.upload_size

    def _finalize_put(self, metadata):
        # Write out metadata before fsync() to ensure it is also forced to
        # disk.
        write_metadata(self.fd, metadata)

        # We call fsync() before calling drop_cache() to lower the
        # amount of redundant work the drop cache code will perform on
        # the pages (now that after fsync the pages will be all
        # clean).
        do_fsync(self.fd)

        # From the Department of the Redundancy Department, make sure
        # we call drop_cache() after fsync() to avoid redundant work
        # (pages all clean).
        drop_buffer_cache(self.fd, 0, self.upload_size)

        # At this point we know that the object's full directory path
        # exists, so we can just rename it directly without using Swift's
        # swift.common.utils.renamer(), which makes the directory path and
        # adds extra stat() calls.
        df = self.disk_file
        data_file = os.path.join(df.put_datadir, df._obj)
        attempts = 1
        while True:
            try:
                do_rename(self.tmppath, data_file)
            except OSError as err:
                if err.errno in (errno.ENOENT, errno.EIO) \
                        and attempts < MAX_RENAME_ATTEMPTS:
                    # FIXME: Why either of these two error conditions is
                    # happening is unknown at this point. This might be a
                    # FUSE issue of some sort or a possible race
                    # condition. So let's sleep on it, and double check
                    # the environment after a good nap.
                    _random_sleep()
                    # Tease out why this error occurred. The man page for
                    # rename reads:
                    #   "The link named by tmppath does not exist; or, a
                    #    directory component in data_file does not exist;
                    #    or, tmppath or data_file is an empty string."
                    assert len(self.tmppath) > 0 and len(data_file) > 0
                    tpstats = do_stat(self.tmppath)
                    tfstats = do_fstat(self.fd)
                    assert tfstats
                    if not tpstats or tfstats.st_ino != tpstats.st_ino:
                        # Temporary file name conflict
                        raise DiskFileError(
                            'DiskFile.put(): temporary file, %s, was'
                            ' already renamed (targeted for %s)' % (
                                self.tmppath, data_file))
                    else:
                        # Data file target name now has a bad path!
                        dfstats = do_stat(df.put_datadir)
                        if not dfstats:
                            raise DiskFileError(
                                'DiskFile.put(): path to object, %s, no'
                                ' longer exists (targeted for %s)' % (
                                    df.put_datadir,
                                    data_file))
                        else:
                            is_dir = stat.S_ISDIR(dfstats.st_mode)
                            if not is_dir:
                                raise DiskFileError(
                                    'DiskFile.put(): path to object, %s,'
                                    ' no longer a directory (targeted for'
                                    ' %s)' % (df.put_datadir,
                                              data_file))
                            else:
                                # Let's retry since everything looks okay
                                logging.warn(
                                    "DiskFile.put(): os.rename('%s','%s')"
                                    " initially failed (%s) but a"
                                    " stat('%s') following that succeeded:"
                                    " %r" % (
                                        self.tmppath, data_file,
                                        str(err), df.put_datadir,
                                        dfstats))
                                attempts += 1
                                continue
                else:
                    raise GlusterFileSystemOSError(
                        err.errno, "%s, os.rename('%s', '%s')" % (
                            err.strerror, self.tmppath, data_file))
            else:
                # Success!
                break
        # Close here so the calling context does not have to perform this
        # in a thread.
        self.close()

    def put(self, metadata, extension='.data'):
        """
        Finalize writing the file on disk, and renames it from the temp file
        to the real location.  This should be called after the data has been
        written to the temp file.

        :param metadata: dictionary of metadata to be written
        :param extension: extension to be used when making the file
        """
        # Our caller will use '.data' here; we just ignore it since we map the
        # URL directly to the file system.

        assert self.tmppath is not None
        metadata = _adjust_metadata(metadata)
        df = self.disk_file

        if dir_is_object(metadata):
            if not df.data_file:
                # Does not exist, create it
                data_file = os.path.join(df._obj_path, df._obj)
                _, df._metadata = self.threadpool.force_run_in_thread(
                    df._create_dir_object, data_file, metadata)
                df.data_file = os.path.join(df._container_path, data_file)
            elif not df._is_dir:
                # Exists, but as a file
                raise DiskFileError('DiskFile.put(): directory creation failed'
                                    ' since the target, %s, already exists as'
                                    ' a file' % df.data_file)
            return

        try:
            self.threadpool.force_run_in_thread(self._finalize_put, metadata)
        except GlusterFileSystemOSError as err:
            if err.errno == errno.EISDIR:
                # A pre-existing directory already exists on the file
                # system, perhaps gratuitously created when another
                # object was created, or created externally to Swift
                # REST API servicing (UFO use case).
                raise DiskFileError('DiskFile.put(): file creation failed'
                                    ' since the target, %s, already exists as'
                                    ' a directory' % df.data_file)
            raise

        # Avoid the unlink() system call as part of the mkstemp context
        # cleanup
        self.tmppath = None

        df._metadata = metadata
        df._filter_metadata()

        # Mark that it actually exists now
        df.data_file = os.path.join(df.datadir, df._obj)


class DiskFile(SwiftDiskFile):
    """
    Manage object files on disk.

    Object names ending or beginning with a '/' as in /a, a/, /a/b/,
    etc, or object names with multiple consecutive slahes, like a//b,
    are not supported.  The proxy server's contraints filter
    gluster.common.constrains.gluster_check_object_creation() should
    reject such requests.

    :param path: path to devices on the node/mount path for UFO.
    :param device: device name/account_name for UFO.
    :param partition: partition on the device the object lives in
    :param account: account name for the object
    :param container: container name for the object
    :param obj: object name for the object
    :param logger: logger object for writing out log file messages
    :param disk_chunk_Size: size of chunks on file reads
    :param bytes_per_sync: number of bytes between fdatasync calls
    :param iter_hook: called when __iter__ returns a chunk
    :param threadpool: thread pool in which to do blocking operations
    :param obj_dir: ignored
    :param mount_check: check the target device is a mount point and not on the
                        root volume
    :param uid: user ID disk object should assume (file or directory)
    :param gid: group ID disk object should assume (file or directory)
    """

    def __init__(self, path, device, partition, account, container, obj,
                 logger, disk_chunk_size=DEFAULT_DISK_CHUNK_SIZE,
                 bytes_per_sync=DEFAULT_BYTES_PER_SYNC, iter_hook=None,
                 threadpool=None, obj_dir='objects', mount_check=False,
                 uid=DEFAULT_UID, gid=DEFAULT_GID):
        if mount_check and not mount(path, device):
            raise DiskFileDeviceUnavailable()
        self.disk_chunk_size = disk_chunk_size
        self.bytes_per_sync = bytes_per_sync
        self.iter_hook = iter_hook
        obj = obj.strip(os.path.sep)

        if os.path.sep in obj:
            self._obj_path, self._obj = os.path.split(obj)
        else:
            self._obj_path = ''
            self._obj = obj

        if self._obj_path:
            self.name = os.path.join(container, self._obj_path)
        else:
            self.name = container
        # Absolute path for object directory.
        self.datadir = os.path.join(path, device, self.name)
        self.device_path = os.path.join(path, device)
        self._container_path = os.path.join(path, device, container)
        if _use_put_mount:
            self.put_datadir = os.path.join(self.device_path + '_PUT',
                                            self.name)
        else:
            self.put_datadir = self.datadir
        self._is_dir = False
        self.logger = logger
        self._metadata = None
        # Don't store a value for data_file until we know it exists.
        self.data_file = None
        self._data_file_size = None
        self.fp = None
        self.iter_etag = None
        self.started_at_0 = False
        self.read_to_eof = False
        self.quarantined_dir = None
        self.suppress_file_closing = False
        self._verify_close = False
        self.threadpool = threadpool or ThreadPool(nthreads=0)
        # FIXME(portante): this attribute is set after open and affects the
        # behavior of the class (i.e. public interface)
        self.keep_cache = False
        self.uid = int(uid)
        self.gid = int(gid)

    def open(self, verify_close=False):
        """
        Open the file and read the metadata.

        This method must populate the _metadata attribute.

        :param verify_close: force implicit close to verify_file, no effect on
                             explicit close.

        :raises DiskFileCollision: on md5 collision
        """
        data_file = os.path.join(self.put_datadir, self._obj)
        try:
            fd = do_open(data_file, os.O_RDONLY | os.O_EXCL)
        except GlusterFileSystemOSError as err:
            if err.errno != errno.ENOENT:
                self.logger.exception(
                    "Error opening file, %s :: %s", data_file, err)
        else:
            try:
                stats = do_fstat(fd)
            except GlusterFileSystemOSError as err:
                self.logger.exception(
                    "Error stat'ing open file, %s :: %s", data_file, err)
            else:
                self._is_dir = stat.S_ISDIR(stats.st_mode)

                self.data_file = data_file

                self._metadata = read_metadata(fd)
                if not self._metadata:
                    create_object_metadata(fd)
                    self._metadata = read_metadata(fd)

                if not validate_object(self._metadata):
                    create_object_metadata(fd)
                    self._metadata = read_metadata(fd)

                self._filter_metadata()

                if self._is_dir:
                    # Use a fake file handle to satisfy the super class's
                    # __iter__ method requirement when dealing with
                    # directories as objects.
                    os.close(fd)
                    self.fp = Fake_file(data_file)
                else:
                    self.fp = os.fdopen(fd, 'rb')
        self._verify_close = verify_close
        self._metadata = self._metadata or {}
        return self

    def _drop_cache(self, fd, offset, length):
        if fd >= 0:
            super(DiskFile, self)._drop_cache(fd, offset, length)

    def close(self, verify_file=True):
        """
        Close the file. Will handle quarantining file if necessary.

        :param verify_file: Defaults to True. If false, will not check
                            file to see if it needs quarantining.
        """
        if self.fp:
            do_close(self.fp)
            self.fp = None
        self._metadata = None
        self._data_file_size = None
        self._verify_close = False

    def _filter_metadata(self):
        if self._metadata is None:
            return
        if X_TYPE in self._metadata:
            self._metadata.pop(X_TYPE)
        if X_OBJECT_TYPE in self._metadata:
            self._metadata.pop(X_OBJECT_TYPE)

    def _create_dir_object(self, dir_path, metadata=None):
        """
        Create a directory object at the specified path. No check is made to
        see if the directory object already exists, that is left to the caller
        (this avoids a potentially duplicate stat() system call).

        The "dir_path" must be relative to its container, self._container_path.

        The "metadata" object is an optional set of metadata to apply to the
        newly created directory object. If not present, no initial metadata is
        applied.

        The algorithm used is as follows:

          1. An attempt is made to create the directory, assuming the parent
             directory already exists

             * Directory creation races are detected, returning success in
               those cases

          2. If the directory creation fails because some part of the path to
             the directory does not exist, then a search back up the path is
             performed to find the first existing ancestor directory, and then
             the missing parents are successively created, finally creating
             the target directory
        """
        full_path = os.path.join(self._container_path, dir_path)
        cur_path = full_path
        stack = []
        while True:
            md = None if cur_path != full_path else metadata
            ret, newmd = make_directory(cur_path, self.uid, self.gid, md)
            if ret:
                break
            # Some path of the parent did not exist, so loop around and
            # create that, pushing this parent on the stack.
            if os.path.sep not in cur_path:
                raise DiskFileError("DiskFile._create_dir_object(): failed to"
                                    " create directory path while exhausting"
                                    " path elements to create: %s" % full_path)
            cur_path, child = cur_path.rsplit(os.path.sep, 1)
            assert child
            stack.append(child)

        child = stack.pop() if stack else None
        while child:
            cur_path = os.path.join(cur_path, child)
            md = None if cur_path != full_path else metadata
            ret, newmd = make_directory(cur_path, self.uid, self.gid, md)
            if not ret:
                raise DiskFileError("DiskFile._create_dir_object(): failed to"
                                    " create directory path to target, %s,"
                                    " on subpath: %s" % (full_path, cur_path))
            child = stack.pop() if stack else None
        return True, newmd

    @contextmanager
    def create(self, size=None):
        """
        Contextmanager to make a temporary file, optionally of a specified
        initial size.

        For Gluster, we first optimistically create the temporary file using
        the "rsync-friendly" .NAME.random naming. If we find that some path to
        the file does not exist, we then create that path and then create the
        temporary file again. If we get file name conflict, we'll retry using
        different random suffixes 1,000 times before giving up.
        """
        data_file = os.path.join(self.put_datadir, self._obj)

        # Assume the full directory path exists to the file already, and
        # construct the proper name for the temporary file.
        attempts = 1
        cur_thread = str(getcurrent())
        while True:
            postfix = md5(self._obj + _cur_host + _cur_pid + cur_thread
                          + str(random.random())).hexdigest()
            tmpfile = '.' + self._obj + '.' + postfix
            tmppath = os.path.join(self.put_datadir, tmpfile)
            try:
                fd = do_open(tmppath,
                             os.O_WRONLY | os.O_CREAT | os.O_EXCL | O_CLOEXEC)
            except GlusterFileSystemOSError as gerr:
                if gerr.errno in (errno.ENOSPC, errno.EDQUOT):
                    # Raise DiskFileNoSpace to be handled by upper layers when
                    # there is no space on disk OR when quota is exceeded
                    raise DiskFileNoSpace()
                if gerr.errno not in (errno.ENOENT, errno.EEXIST, errno.EIO):
                    # FIXME: Other cases we should handle?
                    raise
                if attempts >= MAX_OPEN_ATTEMPTS:
                    # We failed after N attempts to create the temporary
                    # file.
                    raise DiskFileError('DiskFile.mkstemp(): failed to'
                                        ' successfully create a temporary file'
                                        ' without running into a name conflict'
                                        ' after %d of %d attempts for: %s' % (
                                            attempts, MAX_OPEN_ATTEMPTS,
                                            data_file))
                if gerr.errno == errno.EEXIST:
                    # Retry with a different random number.
                    attempts += 1
                elif gerr.errno == errno.EIO:
                    # FIXME: Possible FUSE issue or race condition, let's
                    # sleep on it and retry the operation.
                    _random_sleep()
                    logging.warn("DiskFile.mkstemp(): %s ... retrying in"
                                 " 0.1 secs", gerr)
                    attempts += 1
                elif not self._obj_path:
                    # No directory hierarchy and the create failed telling us
                    # the container or volume directory does not exist. This
                    # could be a FUSE issue or some race condition, so let's
                    # sleep a bit and retry.
                    _random_sleep()
                    logging.warn("DiskFile.mkstemp(): %s ... retrying in"
                                 " 0.1 secs", gerr)
                    attempts += 1
                elif attempts > 1:
                    # Got ENOENT after previously making the path. This could
                    # also be a FUSE issue or some race condition, nap and
                    # retry.
                    _random_sleep()
                    logging.warn("DiskFile.mkstemp(): %s ... retrying in"
                                 " 0.1 secs" % gerr)
                    attempts += 1
                else:
                    # It looks like the path to the object does not already
                    # exist; don't count this as an attempt, though, since
                    # we perform the open() system call optimistically.
                    self._create_dir_object(self._obj_path)
            else:
                break
        dw = None
        try:
            # Ensure it is properly owned before we make it available.
            do_fchown(fd, self.uid, self.gid)
            # NOTE: we do not perform the fallocate() call at all. We ignore
            # it completely.
            dw = DiskWriter(self, fd, tmppath, self.threadpool)
            yield dw
        finally:
            dw.close()
            if dw.tmppath:
                do_unlink(dw.tmppath)

    def put_metadata(self, metadata, tombstone=False):
        """
        Short hand for putting metadata to .meta and .ts files.

        :param metadata: dictionary of metadata to be written
        :param tombstone: whether or not we are writing a tombstone
        """
        if tombstone:
            # We don't write tombstone files. So do nothing.
            return
        metadata = self._keep_sys_metadata(metadata)
        data_file = os.path.join(self.put_datadir, self._obj)
        self.threadpool.run_in_thread(write_metadata, data_file, metadata)

    def _keep_sys_metadata(self, metadata):
        """
        Make sure system metadata is not lost when writing new user metadata

        This method will read the existing metadata and check for system
        metadata. If there are any, it should be appended to the metadata obj
        the user is trying to write.
        """
        if self._metadata:
            orig_metadata = self._metadata
        else:
            data_file = os.path.join(self.put_datadir, self._obj)
            orig_metadata = read_metadata(data_file)

        sys_keys = [X_CONTENT_TYPE, X_ETAG, 'name', X_CONTENT_LENGTH,
                    X_OBJECT_TYPE, X_TYPE]

        for key in sys_keys:
            if key in orig_metadata:
                metadata[key] = orig_metadata[key]

        if X_OBJECT_TYPE not in orig_metadata:
            if metadata[X_CONTENT_TYPE].lower() == DIR_TYPE:
                metadata[X_OBJECT_TYPE] = DIR_OBJECT
            else:
                metadata[X_OBJECT_TYPE] = FILE

        if X_TYPE not in orig_metadata:
            metadata[X_TYPE] = OBJECT

        return metadata

    def _delete(self):
        if self._is_dir:
            # Marker, or object, directory.
            #
            # Delete from the filesystem only if it contains no objects.
            # If it does contain objects, then just remove the object
            # metadata tag which will make this directory a
            # fake-filesystem-only directory and will be deleted when the
            # container or parent directory is deleted.
            metadata = read_metadata(self.data_file)
            if dir_is_object(metadata):
                metadata[X_OBJECT_TYPE] = DIR_NON_OBJECT
                write_metadata(self.data_file, metadata)
            rmobjdir(self.data_file)
        else:
            # Delete file object
            do_unlink(self.data_file)

        # Garbage collection of non-object directories.  Now that we
        # deleted the file, determine if the current directory and any
        # parent directory may be deleted.
        dirname = os.path.dirname(self.data_file)
        while dirname and dirname != self._container_path:
            # Try to remove any directories that are not objects.
            if not rmobjdir(dirname):
                # If a directory with objects has been found, we can stop
                # garabe collection
                break
            else:
                dirname = os.path.dirname(dirname)

    def delete(self, timestamp):
        """
        Remove any older versions of the object file.  Any file that has an
        older timestamp than timestamp will be deleted.

        :param timestamp: timestamp to compare with each file
        """
        timestamp_fl = float(timestamp)
        data_file = os.path.join(self.put_datadir, self._obj)
        try:
            metadata = read_metadata(data_file)
        except (GlusterFileSystemIOError, GlusterFileSystemOSError) as err:
            if err.errno != errno.ENOENT:
                raise
        else:
            try:
                old_ts = float(metadata[X_TIMESTAMP]) >= timestamp_fl
            except (KeyError, ValueError):
                # If no X-Timestamp to compare against, or the timestamp is
                # not a valid float, we'll just delete the object anyways.
                old_ts = False
            if not old_ts:
                self.threadpool.run_in_thread(self._delete)
        self._metadata = {}
        self.data_file = None

    def _get_data_file_size(self):
        """
        Returns the os_path.getsize for the file.  Raises an exception if this
        file does not match the Content-Length stored in the metadata, or if
        self.data_file does not exist.

        :returns: file size as an int
        :raises DiskFileError: on file size mismatch.
        :raises DiskFileNotExist: on file not existing (including deleted)
        """
        if self._is_dir:
            # Directories have no size.
            return 0
        try:
            file_size = 0
            if self.data_file:
                def _old_getsize():
                    file_size = os_path.getsize(self.data_file)
                    if X_CONTENT_LENGTH in self._metadata:
                        metadata_size = int(self._metadata[X_CONTENT_LENGTH])
                        if file_size != metadata_size:
                            # FIXME - bit rot detection?
                            self._metadata[X_CONTENT_LENGTH] = file_size
                            write_metadata(self.data_file, self._metadata)
                    return file_size
                file_size = self.threadpool.run_in_thread(_old_getsize)
                return file_size
        except OSError as err:
            if err.errno != errno.ENOENT:
                raise
        raise DiskFileNotExist('Data File does not exist.')