Removes some code, and actually makes it faster. Problem reports,
especially from highmem systems (which I know how to crash in 30
seconds) appreciated - Hint, avoid direct I/O, buffered I/O on the
same file, do not run test 013 in the regression tests. I am
still chasing this bug.
Date: Fri Jun 8 20:58:42 PDT 2001
Workarea: jen.americas.sgi.com:/src/lord/xfs-linux.2.4
The following file(s) were checked into:
bonnie.engr.sgi.com:/isms/slinx/2.4.x-xfs
Modid: 2.4.x-xfs:slinx:96616a
linux/mm/vmscan.c - 1.57
- rework the launder code for delalloc pages, we issue a write on
them and let the next pass try to clean them up.
linux/include/linux/fs.h - 1.97
- remobe writepage_nounlock
linux/fs/buffer.c - 1.64
- Add a wait option to write_page, always grab the page when it is
set, used from write paths which always want to succeed. Remove
code from try_to_free_buffers and block_flushpage, the only places
which call these have already dealt with delalloc pages.
Finally, use regular writepage, the nounlock version is gone.
linux/fs/xfs/linux/xfs_iops.c - 1.107
- we now only have a writepage method, the nounlock case is gone
linux/fs/pagebuf/page_buf_io.c - 1.81
- Use code more similar to block_write_page in the write path, leave the
page locked for the duration of I/O.
|