[linux-lvm] Re: [PATCH] [CFT] - LVM snapshotting support for XFS

[Steve Lord]

> This message is in MIME format. Since your mail reader does not understand
> this format, some or all of this message may not be legible.
> 
> ------_=_NextPart_000_01C0DC9D.5B6166A0
> Content-Type: text/plain;
> 	charset="iso-8859-1"
> 
> I'll be glad to give this a try.  I've been using the attached patches (in a
> tgz) to use snapshots with xfs in the past.  Basically they are some short
> LVM patches to allow writable snapshots.  Using this I was able to take a
> snapshot and mount the snapshot read-write, allowing XFS to straighten
> things out (without the benefit of an xfs_freeze-like command, it was like
> turning the power off on the original of the system).  Then I would unmount
> the snapshot, change it back to read-only, and remount it for normal
> snapshot use.
> 
> I did have one other problem using XFS snapshots.  As long as the original
> was mounted, I couldn't mount the snapshot as well because it had the same
> uuid.  So before I could mount the snapshot read-write I had to change the
> UUID using xfs_db (which also required writeable snapshots).  Actually, I
> had to change it twice.  My usual sequence would be something like this:

I forgot about the uuid change step in there, yes this will need changing
to make the filesystem mountable - I will think about this. The reason for
the uuid code in XFS is to deal with hardware which has alternate paths to
disks - it prevents you from mounting the same filesystem from what appear
to be two devices, but are actually alternate routes to the same disks. It
may be reasonable to override it for readonly mounts - although that too
is really dangerous in the alternate path case.


> 
> # sync
> # lvcreate --size 100 --chunksize 64  --snapshot --name snap0
> /dev/volgr1/lvol0
> # lvchange -p rw /dev/volgr1/snap0
> # xfs_db -x -p xfs_admin -c 'uuid generate'
> # xfs_db -x -p xfs_admin -c 'uuid generate'
> # mount /dev/volgr1/snap0 /hd/vol_mnt1
> # umount /dev/volgr1/snap0
> # lvchange -p r /dev/volgr1/snap0
> # mount /dev/volgr1/snap0 /hd/vol_mnt1
> 
> With a way to freeze the filesystem, I should be safer than just calling
> sync and I don't think I'll need to do the read-write mount to replay.  Is
> there a way around changing the UUID?

Using this could you should be able to do a readonly mount of the xfs
snapshotted filesystem, modulo the uuid issue.

Steve

> 
> CCed to linux-lvm, since somebody over there may be interested in testing
> XFS snapshots.
> 
> Dale Stephenson
> steph at connex.com
> 
> > -----Original Message-----
> > From: Steve Lord [mailto:lord at sgi.com]
> > Sent: Monday, May 14, 2001 9:50 AM
> > To: linux-xfs at oss.sgi.com
> > Subject: [PATCH] [CFT] - LVM snapshotting support for XFS
> > 
> > 
> > 
> > Here is some experimental code for XFS which should allow use 
> > of the LVM
> > snapshotting capability. There are two attachments to this message, a
> > kernel patch which will apply to the current development tree 
> > (2.4.4 base)
> > and a tar file which adds a new command to the xfs commands. 
> > You will also
> > need to apply the VFS-lock.patch from the lvm sources to 
> > enable snapshotting.
> > 
> > The new command added is xfs_freeze which can be used to stop 
> > and flush
> > an xfs filesystem and then restart it. If you apply the VFS-lock patch
> > from the lvm sources then you will not need this.
> > 
> > I am looking for people to test this code, previous 
> > experience with lvm 
> > and its snapshotting facility would probably be a pre-requisit.
> > 
> > Steve
> > 
> > 
> > 
> 
> 
> ------_=_NextPart_000_01C0DC9D.5B6166A0
> Content-Type: application/octet-stream;
> 	name="writesnap.tgz"
> Content-Transfer-Encoding: base64
> Content-Disposition: attachment;
> 	filename="writesnap.tgz"
> 
> H4sIAAEsADsAA+1YbU/bSBDmK/yKKdLlkjpObCexSXIgUMvpqqNwAtp+KMgy9iZx8ZtsxyAd+e+3
> L363EziOUl3lR2qzLzOzM7PjfXaxIpu/880QaTcW4gNH83qeFuqLrReEIAqCLA+3BAxFHhV+MSSJ
> tJUBaY4UScDyQ3k02gLhJZ1Yh2UQaj7AVhAib1PUj83/T8HzPFims7zvG74ZIT/o20bfiuye3nN9
> c759uVjCe6SDOARBmQz3JoMRSHindjiOW6OY1xHFyWg8GQ6YzuEh8KI8HnZFETjaGChweLgD2xgr
> +kP/XwGyAgR/73Db2wCGe+dAu2VF/IEVqaREg4UbqgGyO1MmYc6g7d/B/j58Of9weQwPD5DrHnUS
> SwD9t0CrHUIXEkuAUxAPLnx3OV8Qez6yNc9DBrztJ7pkldiJG8vVb1V0ryMvNF2HLHX66eQktxCA
> 55tOeNv+8/j8VH2H/SAzbGr3lwBw1iNbxWtM4ORz5gqemS0t68rZ7WbyRNLRbNSFeHnSSWInWHqb
> 80Pgo3DpOzj/6dgqafgGwnpaBPusGdpeKuQHSA9dP5mOe3kJkpY3xMV0ZZo7lqN2K7HI4qEhtXJm
> csMeUsm3GOaGrKhTyCnePt1HGt4qDRx0B2SPTGee26TYnzYOF/tbcOvd2Zd2jTdQ702dO/ls5IeJ
> mx3az5alFUXWVENytMbpiOZquDCDLtYo70P8E1c++RZwtDSVWXlY7tzUNQsi11raiITNYznygazb
> f2JnQ+G+iQuXfYJP2MXi5mVJAhoR8WbpbazFOMwNJQvsFCD/bhb8wY1KFi/U5o8+NH8ikC0XeuOe
> 8S3oWZG+0Jw5eukbAOF/RRmt5X9hqGT8L4iU/5WG/18FhP9PPn9UBXXcD13XCvppFVRvAJjNpfFk
> IGQ3gHWqJa2BPBHl7A4wkMZdBbjBQOjuUf5PobtOiC8VaJqMrWhjh8/O9oQENV1HQQAt7IN6cXr0
> 18UfZ5eULHKijwi3Wok0BZZ3PcwTHXKBoE0ta/odSkWZ9IwSPFklCA3k+13G6zywHAA+XSvn9m5O
> v4rd+Fi/woaudsHBippluXfIIPcB0G2DHLMJ+9cljSZXEWhyFakrSkl2c9I0MYZrhnAAQjEkADqh
> Bkub46ZJ0hmKiSQcF7enhRwSijZc59cQdAvhDyuWn1naPADGOzBzfbIP7K6mOQbpnB8fvac0nreV
> I678BlLFTlG04iG/n4pOn2SVePA0o0SyZLMo9lBIT8EezT6pp4ilPy2ifPGQK01sykO+bQYB5urg
> kerJsOvOyjcFVlL0Wvk0G6Vi+9HHVIPvhDz/R/NA1xx87r8u/2PiF8v8P1KUhv9fA1X+T6uAsv+F
> FlIehzGI0mQoTYS9deyfKBZ0hpPBeCLt5d7/0qgr49e/JHeVhJ3wEQg65hRhmnbNYtdiXS7pWsVp
> G+8ifoi4PsLjvx+dXBxnc+ywLYizV5rA+FKSh/gWwkmy3JXk4m0E48a6VdnzQyid5ITF2rFj+Oc3
> iOaUA2ztHvc5rkKtyenP5L5a1+nrC0vyZdGyMLVNPMFKsVPqzEfoa9y5rnLXM2zU+MxQyPDl+acy
> qZbTkuSlkpg4M/XKZY+LOVqrU1VLI93fEOkmexh11orG2kl1cCB2rtdlhGG1cZa++utiSN/Gtung
> zOJw2N8dqNbHD6dn51inPsJ/H2J5sWSt2nWKwXf+S/TPDrxS2d876DS5zw533VTt+Krua6x+zvU1
> znHX04p+yeKqudk1aNCgQYMGDRo0aNCgQYMGPz/+AWeG5d0AKAAA
> 
> ------_=_NextPart_000_01C0DC9D.5B6166A0--