Synnefo » snf-image-creator

# Copyright 2012 GRNET S.A. All rights reserved.

#

# Redistribution and use in source and binary forms, with or

# without modification, are permitted provided that the following

# conditions are met:

#

#   1. Redistributions of source code must retain the above

#      copyright notice, this list of conditions and the following

#      disclaimer.

#

#   2. Redistributions in binary form must reproduce the above

#      copyright notice, this list of conditions and the following

#      disclaimer in the documentation and/or other materials

#      provided with the distribution.

#

# THIS SOFTWARE IS PROVIDED BY GRNET S.A. ``AS IS'' AND ANY EXPRESS

# OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED

# WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR

# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL GRNET S.A OR

# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,

# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT

# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF

# USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED

# AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT

# LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN

# ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE

# POSSIBILITY OF SUCH DAMAGE.

#

# The views and conclusions contained in the software and

# documentation are those of the authors and should not be

# interpreted as representing official policies, either expressed

# or implied, of GRNET S.A.

from image_creator.util import get_command

from image_creator.util import FatalError

from image_creator.util import try_fail_repeat

from image_creator.util import free_space

from image_creator.gpt import GPTPartitionTable

from image_creator.bundle_volume import BundleVolume

import stat

import os

import tempfile

import uuid

import re

import sys

import guestfs

import shutil

from sendfile import sendfile

dd = get_command('dd')

dmsetup = get_command('dmsetup')

losetup = get_command('losetup')

blockdev = get_command('blockdev')

TMP_CANDIDATES = ['/var/tmp', os.path.expanduser('~'), '/mnt']

class Disk(object):

    """This class represents a hard disk hosting an Operating System

    A Disk instance never alters the source media it is created from.

    Any change is done on a snapshot created by the device-mapper of

    the Linux kernel.

    """

    def __init__(self, source, output, tmp=None):

        """Create a new Disk instance out of a source media. The source

        media can be an image file, a block device or a directory.

        """

        self._cleanup_jobs = []

        self._devices = []

        self.source = source

        self.out = output

        self.meta = {}

        self.tmp = tempfile.mkdtemp(prefix='.snf_image_creator.',

                                    dir=self._get_tmp_dir(tmp))

        self._add_cleanup(shutil.rmtree, self.tmp)

    def _get_tmp_dir(self, default=None):

        if default is not None:

            return default

        space = map(free_space, TMP_CANDIDATES)

        max_idx = 0

        max_val = space[0]

        for i, val in zip(range(len(space)), space):

            if val > max_val:

                max_val = val

                max_idx = i

        # Return the candidate path with more available space

        return TMP_CANDIDATES[max_idx]

    def _add_cleanup(self, job, *args):

        self._cleanup_jobs.append((job, args))

    def _losetup(self, fname):

        loop = losetup('-f', '--show', fname)

        loop = loop.strip()  # remove the new-line char

        self._add_cleanup(try_fail_repeat, losetup, '-d', loop)

        return loop

    def _dir_to_disk(self):

        if self.source == '/':

            bundle = BundleVolume(self.out, self.meta)

            image = '%s/%s.diskdump' % (self.tmp, uuid.uuid4().hex)

            def check_unlink(path):

                if os.path.exists(path):

                    os.unlink(path)

            self._add_cleanup(check_unlink, image)

            bundle.create_image(image)

            return self._losetup(image)

        raise FatalError("Using a directory as media source is supported")

    def cleanup(self):

        """Cleanup internal data. This needs to be called before the

        program ends.

        """

        try:

            while len(self._devices):

                device = self._devices.pop()

                device.destroy()

        finally:

            # Make sure those are executed even if one of the device.destroy

            # methods throws exeptions.

            while len(self._cleanup_jobs):

                job, args = self._cleanup_jobs.pop()

                job(*args)

    def snapshot(self):

        """Creates a snapshot of the original source media of the Disk

        instance.

        """

        self.out.output("Examining source media `%s' ..." % self.source, False)

        sourcedev = self.source

        mode = os.stat(self.source).st_mode

        if stat.S_ISDIR(mode):

            self.out.success('looks like a directory')

            return self._dir_to_disk()

        elif stat.S_ISREG(mode):

            self.out.success('looks like an image file')

            sourcedev = self._losetup(self.source)

        elif not stat.S_ISBLK(mode):

            raise ValueError("Invalid media source. Only block devices, "

                             "regular files and directories are supported.")

        else:

            self.out.success('looks like a block device')

        # Take a snapshot and return it to the user

        self.out.output("Snapshotting media source...", False)

        size = blockdev('--getsz', sourcedev)

        cowfd, cow = tempfile.mkstemp(dir=self.tmp)

        os.close(cowfd)

        self._add_cleanup(os.unlink, cow)

        # Create cow sparse file

        dd('if=/dev/null', 'of=%s' % cow, 'bs=512', 'seek=%d' % int(size))

        cowdev = self._losetup(cow)

        snapshot = uuid.uuid4().hex

        tablefd, table = tempfile.mkstemp()

        try:

            os.write(tablefd, "0 %d snapshot %s %s n 8" %

                              (int(size), sourcedev, cowdev))

            dmsetup('create', snapshot, table)

            self._add_cleanup(try_fail_repeat, dmsetup, 'remove', snapshot)

        finally:

            os.unlink(table)

        self.out.success('done')

        return "/dev/mapper/%s" % snapshot

    def get_device(self, media):

        """Returns a newly created DiskDevice instance."""

        new_device = DiskDevice(media, self.out)

        self._devices.append(new_device)

        new_device.enable()

        return new_device

    def destroy_device(self, device):

        """Destroys a DiskDevice instance previously created by

        get_device method.

        """

        self._devices.remove(device)

        device.destroy()

class DiskDevice(object):

    """This class represents a block device hosting an Operating System

    as created by the device-mapper.

    """

    def __init__(self, device, output, bootable=True, meta={}):

        """Create a new DiskDevice."""

        self.real_device = device

        self.out = output

        self.bootable = bootable

        self.meta = meta

        self.progress_bar = None

        self.guestfs_device = None

        self.size = 0

        self.g = guestfs.GuestFS()

        self.g.add_drive_opts(self.real_device, readonly=0, format="raw")

        # Before version 1.17.14 the recovery process, which is a fork of the

        # original process that called libguestfs, did not close its inherited

        # file descriptors. This can cause problems especially if the parent

        # process has opened pipes. Since the recovery process is an optional

        # feature of libguestfs, it's better to disable it.

        self.g.set_recovery_proc(0)

        version = self.g.version()

        if version['major'] > 1 or \

            (version['major'] == 1 and (version['minor'] >= 18 or

                                        (version['minor'] == 17 and

                                         version['release'] >= 14))):

            self.g.set_recovery_proc(1)

            self.out.output("Enabling recovery proc")

        #self.g.set_trace(1)

        #self.g.set_verbose(1)

        self.guestfs_enabled = False

    def enable(self):

        """Enable a newly created DiskDevice"""

        self.out.output('Launching helper VM (may take a while) ...', False)

        # self.progressbar = self.out.Progress(100, "Launching helper VM",

        #                                     "percent")

        # eh = self.g.set_event_callback(self.progress_callback,

        #                               guestfs.EVENT_PROGRESS)

        self.g.launch()

        self.guestfs_enabled = True

        # self.g.delete_event_callback(eh)

        # self.progressbar.success('done')

        # self.progressbar = None

        self.out.success('done')

        self.out.output('Inspecting Operating System ...', False)

        roots = self.g.inspect_os()

        if len(roots) == 0:

            raise FatalError("No operating system found")

        if len(roots) > 1:

            raise FatalError("Multiple operating systems found."

                             "We only support images with one OS.")

        self.root = roots[0]

        self.guestfs_device = self.g.part_to_dev(self.root)

        self.size = self.g.blockdev_getsize64(self.guestfs_device)

        self.meta['PARTITION_TABLE'] = \

            self.g.part_get_parttype(self.guestfs_device)

        self.ostype = self.g.inspect_get_type(self.root)

        self.distro = self.g.inspect_get_distro(self.root)

        self.out.success('found a(n) %s system' % self.distro)

    def destroy(self):

        """Destroy this DiskDevice instance."""

        # In new guestfs versions, there is a handy shutdown method for this

        try:

            if self.guestfs_enabled:

                self.g.umount_all()

                self.g.sync()

        finally:

            # Close the guestfs handler if open

            self.g.close()

#    def progress_callback(self, ev, eh, buf, array):

#        position = array[2]

#        total = array[3]

#

#        self.progressbar.goto((position * 100) // total)

    def mount(self, readonly=False):

        """Mount all disk partitions in a correct order."""

        mount = self.g.mount_ro if readonly else self.g.mount

        msg = " read-only" if readonly else ""

        self.out.output("Mounting the media%s ..." % msg, False)

        mps = self.g.inspect_get_mountpoints(self.root)

        # Sort the keys to mount the fs in a correct order.

        # / should be mounted befor /boot, etc

        def compare(a, b):

            if len(a[0]) > len(b[0]):

                return 1

            elif len(a[0]) == len(b[0]):

                return 0

            else:

                return -1

        mps.sort(compare)

        for mp, dev in mps:

            try:

                mount(dev, mp)

            except RuntimeError as msg:

                self.out.warn("%s (ignored)" % msg)

        self.out.success("done")

    def umount(self):

        """Umount all mounted filesystems."""

        self.g.umount_all()

    def _last_partition(self):

        if self.meta['PARTITION_TABLE'] not in 'msdos' 'gpt':

            msg = "Unsupported partition table: %s. Only msdos and gpt " \

                "partition tables are supported" % self.meta['PARTITION_TABLE']

            raise FatalError(msg)

        is_extended = lambda p: \

            self.g.part_get_mbr_id(self.guestfs_device, p['part_num']) \

            in (0x5, 0xf)

        is_logical = lambda p: \

            self.meta['PARTITION_TABLE'] == 'msdos' and p['part_num'] > 4

        partitions = self.g.part_list(self.guestfs_device)

        last_partition = partitions[-1]

        if is_logical(last_partition):

            # The disk contains extended and logical partitions....

            extended = filter(is_extended, partitions)[0]

            last_primary = [p for p in partitions if p['part_num'] <= 4][-1]

            # check if extended is the last primary partition

            if last_primary['part_num'] > extended['part_num']:

                last_partition = last_primary

        return last_partition

    def shrink(self):

        """Shrink the disk.

        This is accomplished by shrinking the last filesystem in the

        disk and then updating the partition table. The new disk size

        (in bytes) is returned.

        ATTENTION: make sure unmount is called before shrink

        """

        get_fstype = lambda p: \

            self.g.vfs_type("%s%d" % (self.guestfs_device, p['part_num']))

        is_logical = lambda p: \

            self.meta['PARTITION_TABLE'] == 'msdos' and p['part_num'] > 4

        is_extended = lambda p: \

            self.meta['PARTITION_TABLE'] == 'msdos' and \

            self.g.part_get_mbr_id(self.guestfs_device, p['part_num']) \

            in (0x5, 0xf)

        part_add = lambda ptype, start, stop: \

            self.g.part_add(self.guestfs_device, ptype, start, stop)

        part_del = lambda p: self.g.part_del(self.guestfs_device, p)

        part_get_id = lambda p: self.g.part_get_mbr_id(self.guestfs_device, p)

        part_set_id = lambda p, id: \

            self.g.part_set_mbr_id(self.guestfs_device, p, id)

        part_get_bootable = lambda p: \

            self.g.part_get_bootable(self.guestfs_device, p)

        part_set_bootable = lambda p, bootable: \

            self.g.part_set_bootable(self.guestfs_device, p, bootable)

        MB = 2 ** 20

        self.out.output("Shrinking image (this may take a while) ...", False)

        sector_size = self.g.blockdev_getss(self.guestfs_device)

        last_part = None

        fstype = None

        while True:

            last_part = self._last_partition()

            fstype = get_fstype(last_part)

            if fstype == 'swap':

                self.meta['SWAP'] = "%d:%s" % \

                    (last_part['part_num'],

                     (last_part['part_size'] + MB - 1) // MB)

                part_del(last_part['part_num'])

                continue

            elif is_extended(last_part):

                part_del(last_part['part_num'])

                continue

            # Most disk manipulation programs leave 2048 sectors after the last

            # partition

            new_size = last_part['part_end'] + 1 + 2048 * sector_size

            self.size = min(self.size, new_size)

            break

        if not re.match("ext[234]", fstype):

            self.out.warn("Don't know how to resize %s partitions." % fstype)

            return self.size

        part_dev = "%s%d" % (self.guestfs_device, last_part['part_num'])

        self.g.e2fsck_f(part_dev)

        self.g.resize2fs_M(part_dev)

        out = self.g.tune2fs_l(part_dev)

        block_size = int(filter(lambda x: x[0] == 'Block size', out)[0][1])

        block_cnt = int(filter(lambda x: x[0] == 'Block count', out)[0][1])

        start = last_part['part_start'] / sector_size

        end = start + (block_size * block_cnt) / sector_size - 1

        if is_logical(last_part):

            partitions = self.g.part_list(self.guestfs_device)

            logical = []  # logical partitions

            for partition in partitions:

                if partition['part_num'] < 4:

                    continue

                logical.append({

                    'num': partition['part_num'],

                    'start': partition['part_start'] / sector_size,

                    'end': partition['part_end'] / sector_size,

                    'id': part_get_id(partition['part_num']),

                    'bootable': part_get_bootable(partition['part_num'])

                })

            logical[-1]['end'] = end  # new end after resize

            # Recreate the extended partition

            extended = filter(is_extended, partitions)[0]

            part_del(extended['part_num'])

            part_add('e', extended['part_start'] / sector_size, end)

            # Create all the logical partitions back

            for l in logical:

                part_add('l', l['start'], l['end'])

                part_set_id(l['num'], l['id'])

                part_set_bootable(l['num'], l['bootable'])

        else:

            # Recreate the last partition

            if self.meta['PARTITION_TABLE'] == 'msdos':

                last_part['id'] = part_get_id(last_part['part_num'])

            last_part['bootable'] = part_get_bootable(last_part['part_num'])

            part_del(last_part['part_num'])

            part_add('p', start, end)

            part_set_bootable(last_part['part_num'], last_part['bootable'])

            if self.meta['PARTITION_TABLE'] == 'msdos':

                part_set_id(last_part['part_num'], last_part['id'])

        new_size = (end + 1) * sector_size

        assert (new_size <= self.size)

        if self.meta['PARTITION_TABLE'] == 'gpt':

            ptable = GPTPartitionTable(self.real_device)

            self.size = ptable.shrink(new_size, self.size)

        else:

            self.size = min(new_size + 2048 * sector_size, self.size)

        self.out.success("new size is %dMB" % ((self.size + MB - 1) // MB))

        return self.size

    def dump(self, outfile):

        """Dumps the content of device into a file.

        This method will only dump the actual payload, found by reading the

        partition table. Empty space in the end of the device will be ignored.

        """

        MB = 2 ** 20

        blocksize = 4 * MB  # 4MB

        size = self.size

        progr_size = (size + MB - 1) // MB  # in MB

        progressbar = self.out.Progress(progr_size, "Dumping image file", 'mb')

        with open(self.real_device, 'r') as src:

            with open(outfile, "w") as dst:

                left = size

                offset = 0

                progressbar.next()

                while left > 0:

                    length = min(left, blocksize)

                    sent = sendfile(dst.fileno(), src.fileno(), offset, length)

                    # Workaround for python-sendfile API change. In

                    # python-sendfile 1.2.x (py-sendfile) the returning value

                    # of sendfile is a tuple, where in version 2.x (pysendfile)

                    # it is just a sigle integer.

                    if isinstance(sent, tuple):

                        sent = sent[1]

                    offset += sent

                    left -= sent

                    progressbar.goto((size - left) // MB)

        progressbar.success('image file %s was successfully created' % outfile)

# vim: set sta sts=4 shiftwidth=4 sw=4 et ai :