[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

[dm-devel] Re: New kernel shared snapshots

Thanks for the recent repost (18-Aug-2009) of your patches at:

The interface seems to be in-line with what we want. I noticed that you can't have dm-snapshot and dm-multisnapshot loaded in the kernel at the same time (kmem_cache_create: duplicate cache dm_snap_tracked_chunk -- in the module init function). You will also want to run your patches through 'linux/scripts/checkpatch.pl'. It mostly gets angry about the 80 characters / line breakage, but there is some other things it catches.


On Jul 27, 2009, at 6:24 AM, Mikulas Patocka wrote:


The new release of shared snapshots can be found here:

It has the interface we talked about on Linux Tag.
So look at it as a last chance to review the interface. And also test it.

To create a snapshot, you have to:
- send a message
- ask for status, get the snapshot id
- suspend (this creates the snapshot, when the filesystem is quiescent)
- resume

Then, you can attach the snapshots by loading "multisnap-snap" target.

If the table line contains argument "sync-snapshots", it synchronizes the
snapshot list against what is given on the table --- i.e. creates
snapshots that are on the table but are not in the store and deletes those
that are not in the store but are on the table.

Other changes:
- make more structures private (move them to dm-multisnap- private.h), so
 that ABI doesn't change if these are chaged.
- snapshot IDs are treated as strings, not numbers.
- metadata cache size can be specified as an argument to the "mikulas"
 exception store.
- an argument "preserve-on-error" that says that on error or overflow,
 writes to the origin should be disabled.
	- if you don't specify this, the origin continues operation, but
	snapshots are irreversibly damaged by this (use for non-important
	snapshots such as backups)
	- if you specify this, the data are always preserved, the whole
	thing just halts on error (user when snapshots contain some
	important data.
- fixed two bad bugs in dm-bufio.

Known bugs:
- that read-vs-realloc race (that I already fixed in the old snapshots a
year ago) is present there. I'll fix it, but it isn't so trivial as in
unshared snapshots.


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]