On Fri, Aug 08, 2014 at 12:04:40PM -0400, Chris Mason wrote:
> xfs is using truncate_pagecache_range to invalidate the page cache
> during DIO writes. The other filesystems are calling
> truncate_pagecache_range is meant to be used when we are freeing the
> underlying data structs from disk, so it will zero any partial ranges
> in the page. This means a DIO write can zero out part of the page cache
> page, and it is possible the page will stay in cache.
> This one is an RFC because it is untested and because I don't understand
> how XFS is dealing with pages the truncate was unable to clear away.
> I'm not able to actually trigger zeros by mixing DIO writes with
> buffered reads.
> Signed-off-by: Chris Mason <clm@xxxxxx>
> diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
> index 8d25d98..c30c112 100644
> --- a/fs/xfs/xfs_file.c
> +++ b/fs/xfs/xfs_file.c
> @@ -638,7 +638,10 @@ xfs_file_dio_aio_write(
> pos, -1);
> if (ret)
> goto out;
> - truncate_pagecache_range(VFS_I(ip), pos, -1);
> + /* what do we do if we can't invalidate the pages? */
> + invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
> + pos >> PAGE_CACHE_SHIFT, -1);
I don't think it can on XFS.
We're holding the XFS_IOLOCK_EXCL, so no other syscall based IO can
dirty pages, all the pages are clean, try_to_free_buffers() will
never fail, no-one can run a truncate operation concurently, and
The only thing that could cause an issue is a racing mmap page fault
dirtying the page. But if you are mixing mmap+direct IO on the same
file, you lost a long time ago so that's nothing new at all.
So, I'd just do:
ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
pos >> PAGE_CACHE_SHIFT, -1);
ret = 0;
That way we'll find out if it does ever fail, and if it does we can
take it from there.