I've finally nailed this one down...
Under certain circumstances during a clean remount, the log
recovery code clobbers a large area of disk after the end of
the log.
The reason this happens is because a particular piece of
recovery code uses xlog_get_bp to request a big buffer, then
uses XFS_BUF_SET_COUNT to set the pb_count_desired field
requesting that only a portion of the buffer be written.
pagebuf ignores the pb_count_desired field and writes
pb_buffer_length bytes anyway. The XFS_BUF_SET_COUNT isn't used
anywhere else in XFS.
Is pb_count_desired meant to be supported or is it an IRIX hang
over that needs to be removed?
If it's the latter, I'll happily remove it...
-----------------------------------------------------
Daniel Moore dxm@xxxxxxx
R&D Software Engineer Phone: +61-3-98348209
SGI Performance Tools Group Fax: +61-3-98132378
-----------------------------------------------------
|