xfs
[Top] [All Lists]

[Bug 339] data loss problem

To: xfs-master@xxxxxxxxxxx
Subject: [Bug 339] data loss problem
From: bugzilla-daemon@xxxxxxxxxxx
Date: Sun, 20 Jun 2004 19:12:33 -0700
Sender: linux-xfs-bounce@xxxxxxxxxxx
http://oss.sgi.com/bugzilla/show_bug.cgi?id=339


nathans@xxxxxxx changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|xfs-master@xxxxxxxxxxx      |nathans@xxxxxxx




------- Additional Comments From nathans@xxxxxxx  2004-20-06 19:12 PDT -------
Hi there Tsuda!

Thanks for the analysis and great explanation again.  You know, I think
this can be more simply fixed by making mod xfs-linux:xfs-kern:166504a
happen on 2.4 as well (thats a change that was done only for 2.6) - this
was the rcs comment there: "make sure i_size_write is called under i_sem
Don't update i_size in vn_revalidate"

From a look through the 2.4 code, I believe a similar approach will work
on 2.4 as well, and should resolve this.  Could you try this alternate
patch? (see next attachment)

Looking through your patch, it'd probably have been simpler to add back
the setsize io op (exists on IRIX, woulda existed in some earlier Linux
XFS versions) rather than adding a vop.  I also think that a race will
exist in this patch where it calls generic_commit_write, then does the
XFS inode size update - if a revalidate was done between those two, the
size would be mishandled again, I think.  At the end of the day, if we
have anywhere that the revalidate is done without i_sem held, we will
get ourselves into strife.

cheers.




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.


<Prev in Thread] Current Thread [Next in Thread>