On Tue, 25 Mar 2003, Steven Dake wrote:
> The function write_full_page will unlock the page if delalloc_convert
> returns a negative value. I don't understand why delalloc_convert
> returns a negative value, and if the page should be unlocked, but it
> seems that it shouldn't. Could you give me some background on why the
> page is unlocked at this point? If I comment out this unlock, bonnie++
delalloc_convert can fail on some error, and we have to do something
with the buffer at that point. This case was put in because
if we ignored the error, we'd write out a buffer with block 0 (delalloc)
and clobber the superblock. Perhaps locking has changed such that
the unlock_page is no longer correct, I'd have to look more closely.
Are you hitting this case?
It would be very helpful if you could run this on a "stock" xfs
1.2 kernel to see if you still hit this problem.