[BBLISA] dump or Legato Networker?

[Edward Ned Harvey]

> > The issue I believe that I was remembering was something more along
> > the lines of a uniquely linux quirk where "dump" did not permit
> > correct restores. 

The issue at hand is:  Dump operates on the block-level device.  So any
files which have been changed in memory but the cache not yet flushed are at
risk.  Linus is sometimes overreactional, with a blanket statement like
"dump is stupid, leave it behind."

The same is not true of tools such as tar, because they just browse the
filesystem, and don't operate on the block level device.  However, tools
like tar introduce a whole new set of problems - First it's hard to get the
switches right to preserve all the idiosyncrasies of the filesystem, like
ownerships & permissions, broken softlinks, character special devices, and
hardlinks.  (At least the ownership/permission has an easy solution, as long
as you religiously remember to use -p).  And then tar will happily go off
backing up files from your network mounts too...  Surely not intented.  ;-)

I think Linus's overreaction is sort of an emotional issue; there are a few
ways that Linux is behind other OSes, including backups & snapshots.

Whenever you're backing up a mounted filesystem, you have to face the fact
that you don't get file coherence.  If some daemon (such as mysqld) has a
file permanently open, you simply can't back it up, end of story.  You have
to go read the mysqld manual for recommended backup procedure.  If some user
always has Cadence or Fluent or Synopsys or Microsoft Word open, and some
file always open, it too cannot be backed up.  The best thing you can do in
those situations is to use snapshotting, which doesn't exist in Linux.
(Although it exists, it's almost unusably garbage.)  

In the *true* ideal situation, you unmount the FS, and run dump.  That's the
only way you're guaranteed to get everything, coherently, including all
permissions, on any OS.  But since most people can't do this very often, you
have to live with the best you can do, which is to attempt snapshots (if you
have netapp or solaris), and if you can't do snapshots (if you have linux)
you just run dump or tar on the open filesystem and hope for the best.

Is there any way to signal the Linux kernel to flush its filesystem write
cache, directly before performing the backup?