Hi Al,
I like what you are planning here.
We walk the list of children for cache invalidations. This happens
both in InterMezzo and Coda.
Generally speaking cache invalidations are a reasonable thing to do,
but our implementation isn't that great. The purpose of a cache
invalidation is to
I think that's ok with the Coda directories, if not it's really easy to
fix in userland for us. Assume it.
>
> ObCODA: do you have any sort of alignment for cached directories? It would be
> ideal if you could guarantee that no entry crosses 4K-boundary in caching
> file...
>
Alexander Viro wrote:
>
> 6) unions in superblocks (and inodes) are nasty. Use separately-allocated
> private parts of either and keep pointer to them in ->u.generic_sbp and
> ->u.generic_ip, resp. It's less critical for superblocks, but for inodes
> it is pretty serious.
Is it expensive to all
The kernel lock is released the moment the process is put asleep in the
upcall. Keep that in mind.
- Peter -
On Wed, 9 Feb 2000, Jan Harkes wrote:
> On Wed, Feb 09, 2000 at 06:36:34PM -0500, Alexander Viro wrote:
> >
> >
> > On Wed, 9 Feb 2000, Tigran Aivazian wrote:
> >
> > > Hi Linus,
> >
Hi Stephen,
I have some trouble with using brw_kiovec when the vector contains more
than one kiobuf.
Suppose we pass in an array of 2 kiobufs, each for one page.
It seems to me that the invocation of do_kio in the inner (block) loop
will not happen, neither for iovec[0], nor for iovec[1], sin
Hi,
A friend lost her partition table...
Is there a scanning tool that can restore it?
Thanks!
- Peter -
700 (MST), "Peter J. Braam"
> <[EMAIL PROTECTED]> said:
>
> > So I'm now requesting to add a "d_put" dentry method, that a file system
> > can optionally declare. It would always be invoked by the dput method,
> > perhaps indicating also when the
> >
> > GFS gets it right, but solely by the grace that the VFS is not yet fully
> > threaded, and they have to do the O_EXCL tests twice, once in the VFS
> > while holding a readlock and then in GFS with a write lock. In
> > d_revalidate they take a readlock and validate it is still good and
>
Hi Al,
I agree with your reply and kept thinking more about this. We concluded
that lookup_dentry is _the_ point where passing a flag to the FS makes a
lot of sense if the file system maintains locks - and the flags you
suggest are better than my suggestions.
(For people who haven't followed t
I think I mean joining. What I need is:
braam starts trans
does A
calls reiser: hans starts
does B
hans commits; nothing goes to disk yet
braam does C
braam commits/aborts ABC now go or don't
- Peter -
On Wed, 5 Jan 2000, Hans Reiser wrote:
> Is nesting really the
Hi,
I have one request for the journal API for use by network file systems -
it is a request of a slightly different nature than the ones discussed so
far.
InterMezzo (www.inter-mezzo.org) exploits an existing disk file system as
a cache and wraps around it. (Any disk file system can be used,
so that we
can tell you where we are heading. We expect to release more software
frequently now - subscribe to the obd-announce mailing list if you
want to be informed.
Stelias wishes to thank many for support, advice and stimulating
discussions on this topic.
Peter J. Braam - [EMAIL PROTE
Let's just take one step back.
Al has successfully pointed out that one should not expect too much in
terms of security improvements for my hardlink suggestion.
Al additionally gave two reasons, totally unrelated to the security
issues, not to implement the link semantics as I suggested:
> On Wed, 1 Dec 1999, Peter J. Braam wrote:
>
> > 2. Hard links across directories are not permitted. Jan explained that
> > security is an issue here.
> >
> > I think there is wrong thinking in the way Unix does things normally and
> > the security a
Hi,
Thanks for your comments.
1. Coda's ctime not set on create is a bug -- I'll send a fix with the
other 2.3 fixes we will do over the next week or so.
2. Hard links across directories are not permitted. Jan explained that
security is an issue here.
I think there is wrong thinking in the
Hi,
Stephen wrote:
> Fixing this in raid seems far, far preferable to fixing it in the
> filesystems. The filesystem should be allowed to use the buffer cache
> for metadata and should be able to assume that there is a way to prevent
> those buffers from being written to disk until it is ready
Hi,
I'm working on a file system which talks to an "inode disk", the storage
industry calls these object based disks. A simulated object based disk
can be constructed from the lower half of ext2 (or any other file system
for that matter).
The file system has no knowledge of disk blocks, and so
Hi Andreas,
I only have experience with loop devices, but with these it should be
trivial to extend the size of those, but since the setup ioctl (losetup
executes this) finds the size of the device, you will need to change
that. From there on you'll be fine.
In loop.c add an ioctl and do someth
Whoops, I should have said I looked at 2.3.12 Ithink.
- Peter -
On Fri, 17 Sep 1999, Erez Zadok wrote:
> Heh, heh. Funny you should mention this Peter. I'm struggling with this
> question every time a new kernel release is made, b/c I have to updated my
> stackable f/s modules.
>
> If you
Hi
I wondered if someone could explain what is happening in
generic_file_read:
More generically, I'd like to understand how in a file system I can
"get" a page and use it to copy date into/out of it. How do I then put
the page away?
There seem to be several patterns, but I can't figure them o
s can _never_ migrate from one file to another like this.
- Peter -
> -Original Message-
> From: [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED]]On Behalf Of Matthew
> Kirkwood
> Sent: Wednesday, August 18, 1999 10:38 AM
> To: Peter J. Braam
> Cc: [EMAIL PROTECTED]
Whoops, of course, only if the power goes out. But then, this is still a
serious problem.
Peter
> -Original Message-
> From: [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED]]On Behalf Of Matthew
> Kirkwood
> Sent: Wednesday, August 18, 1999 10:20 AM
> To: Peter J.
Hi,
I've recently learned that disks sometimes reorder writes (I think the SCSI
models do it).
If we look at how the buffers are flushed for ext2 files, then the system is
(fs/buffer.c)
flush data buffers
write super block to buffers
write inodes to buffers
flush data buffers again
The bdflush
ed look at:
ftp://ftp.inter-mezzo.org/pub/obd
The code is GPL'd.
- Peter Braam -
Hi,
I'm working on a new distributed file system called InterMezzo (see
www.inter-mezzo.org), which has similar features to Coda, but is much
smaller and integrates better with existing disk file systems (for example
ext2 disk format is used for both caches and server data, and on servers
full ac
Hi,
I'm confused about reading and then writing files (say for ext2).
Suppose we first read block 1 in a file. Generic_readpage adds the data
to the inode page cache using brw_page. It uses some temporary
buffers that are not hashed.
Assume that next we want to write in block 1, then ext2_getb
26 matches
Mail list logo