here are some issues that were raised during today's LVM2 lecture, and
some answers i found for them:

1. the word for "silencing" the file-system (or application) is
'quiesce'. see for a very
incomplete article.

2. Quiesce in reiserfs and ext3 - it appears there is integration
between snapshot creation and quiescing the file-system. this
integration is done in the kernel level as follows: before the snapshot
is created, the device mapper invokes a function of the block device
called "freeze_bdev". this function then will invoke a file-system
specific "sync_fs" function and later a function named
'write_super_lockfs' (if the file system supports these functions). in
reiserfs and ext3, this function commits any pending transactions,
putting the file-system in consistent state. since this operation also
blocks new I/O (until the snapshot creation was finished), we get a
consistent file-system.

3. the code that implements the old RAID levels (0, 1, 5, 6) indeed does
not use the new device mapper code-parts (like the dm-table). they just
placed all the source files in a single directory, to confuse the
reader ;)

4. when we create a read-write snapshot, and then mount the snapshot and
write to it, indeed the old (frozen) data is over-written on the
snapshot. there is no way to undo this, other then taking a snapshot of
the snapshot.

5. when a snapshot becomes full, it is switched to 'inactive' state -
it's not deleted from the system.

if there were other questions which i forgot, or new questions come to
mind, feel free to ask them here (or in private, if you prefer that).


Haifa Linux Club Mailing List (
To unsub send an empty message to [EMAIL PROTECTED]

Reply via email to