[00/19,V2] orangefs: page cache

Commit Message

Martin Brandenburg Oct. 7, 2018, 11:27 p.m. UTC

V2... see https://marc.info/?l=linux-fsdevel&m=153721507330730&w=2

One important change is the following, without which an unaligned write
may end up written to the beginning of its page.  Surprisingly xfstests
did not catch this.  This was caught by an invalidate_inode_pages2 call
in read_iter (not part of this patch series) which exposed the bug.


The other big thing is an improved releasepage implementation and a
launder_page implementation.

We also size the writepages buffer based on the client core buffer.
However I'm not happy with the memory allocation in writepages.

Mike, I know you've had some trouble with the original series.  I'd like
to know if this fixes that.

Martin Brandenburg (19):
  orangefs: implement xattr cache
  orangefs: do not invalidate attributes on inode create
  orangefs: simplify orangefs_inode_getattr interface
  orangefs: update attributes rather than relying on server
  orangefs: hold i_lock during inode_getattr
  orangefs: set up and use backing_dev_info
  orangefs: let setattr write to cached inode
  orangefs: reorganize setattr functions to track attribute changes
  orangefs: remove orangefs_readpages
  orangefs: service ops done for writeback are not killable
  orangefs: migrate to generic_file_read_iter
  orangefs: implement writepage
  orangefs: skip inode writeout if nothing to write
  orangefs: write range tracking
  orangefs: avoid fsync service operation on flush
  orangefs: use kmem_cache for orangefs_write_request
  orangefs: implement writepages
  orangefs: use client-core buffer size to determine writepages count
  orangefs: do writepages_work if a single page must be written

 fs/orangefs/acl.c             |   4 +-
 fs/orangefs/file.c            | 194 +++--------
 fs/orangefs/inode.c           | 628 ++++++++++++++++++++++++++++------
 fs/orangefs/namei.c           |  41 +--
 fs/orangefs/orangefs-cache.c  |  24 +-
 fs/orangefs/orangefs-kernel.h |  56 ++-
 fs/orangefs/orangefs-mod.c    |  10 +-
 fs/orangefs/orangefs-utils.c  | 181 +++++-----
 fs/orangefs/super.c           |  38 +-
 fs/orangefs/waitqueue.c       |  18 +-
 fs/orangefs/xattr.c           | 104 ++++++
 11 files changed, 893 insertions(+), 405 deletions(-)

Comments

Mike Marshall Oct. 8, 2018, 8:43 p.m. UTC | #1

I still hit writepages_callback not private! on xfstests generic/127...

-Mike
On Sun, Oct 7, 2018 at 7:27 PM Martin Brandenburg <[e-mail]> wrote:
>
> V2... see https://marc.info/?l=linux-fsdevel&m=153721507330730&w=2
>
> One important change is the following, without which an unaligned write
> may end up written to the beginning of its page.  Surprisingly xfstests
> did not catch this.  This was caught by an invalidate_inode_pages2 call
> in read_iter (not part of this patch series) which exposed the bug.
>
> diff --git a/fs/orangefs/inode.c b/fs/orangefs/inode.c
> index 34b98d2ed377..cd1263c45bb2 100644
> --- a/fs/orangefs/inode.c
> +++ b/fs/orangefs/inode.c
> @@ -56,7 +54,7 @@ static int orangefs_writepage_locked(struct page *page,
>
>         bv.bv_page = page;
>         bv.bv_len = wlen;
> -       bv.bv_offset = 0;
> +       bv.bv_offset = off % PAGE_SIZE;
>         iov_iter_bvec(&iter, ITER_BVEC | WRITE, &bv, 1, wlen);
>
>         ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, wlen,
>
> The other big thing is an improved releasepage implementation and a
> launder_page implementation.
>
> We also size the writepages buffer based on the client core buffer.
> However I'm not happy with the memory allocation in writepages.
>
> Mike, I know you've had some trouble with the original series.  I'd like
> to know if this fixes that.
>
> Martin Brandenburg (19):
>   orangefs: implement xattr cache
>   orangefs: do not invalidate attributes on inode create
>   orangefs: simplify orangefs_inode_getattr interface
>   orangefs: update attributes rather than relying on server
>   orangefs: hold i_lock during inode_getattr
>   orangefs: set up and use backing_dev_info
>   orangefs: let setattr write to cached inode
>   orangefs: reorganize setattr functions to track attribute changes
>   orangefs: remove orangefs_readpages
>   orangefs: service ops done for writeback are not killable
>   orangefs: migrate to generic_file_read_iter
>   orangefs: implement writepage
>   orangefs: skip inode writeout if nothing to write
>   orangefs: write range tracking
>   orangefs: avoid fsync service operation on flush
>   orangefs: use kmem_cache for orangefs_write_request
>   orangefs: implement writepages
>   orangefs: use client-core buffer size to determine writepages count
>   orangefs: do writepages_work if a single page must be written
>
>  fs/orangefs/acl.c             |   4 +-
>  fs/orangefs/file.c            | 194 +++--------
>  fs/orangefs/inode.c           | 628 ++++++++++++++++++++++++++++------
>  fs/orangefs/namei.c           |  41 +--
>  fs/orangefs/orangefs-cache.c  |  24 +-
>  fs/orangefs/orangefs-kernel.h |  56 ++-
>  fs/orangefs/orangefs-mod.c    |  10 +-
>  fs/orangefs/orangefs-utils.c  | 181 +++++-----
>  fs/orangefs/super.c           |  38 +-
>  fs/orangefs/waitqueue.c       |  18 +-
>  fs/orangefs/xattr.c           | 104 ++++++
>  11 files changed, 893 insertions(+), 405 deletions(-)
>
> --
> 2.19.0
>

Martin Brandenburg Oct. 8, 2018, 11:16 p.m. UTC | #2

On Mon, Oct 08, 2018 at 04:43:36PM -0400, Mike Marshall wrote:
> I still hit writepages_callback not private! on xfstests generic/127...

There's six fsx runs in there.  Can you figure out which one fails?

Martin

Patch

diff --git a/fs/orangefs/inode.c b/fs/orangefs/inode.c
index 34b98d2ed377..cd1263c45bb2 100644
--- a/fs/orangefs/inode.c
+++ b/fs/orangefs/inode.c
@@ -56,7 +54,7 @@  static int orangefs_writepage_locked(struct page *page,
 
 	bv.bv_page = page;
 	bv.bv_len = wlen;
-	bv.bv_offset = 0;
+	bv.bv_offset = off % PAGE_SIZE;
 	iov_iter_bvec(&iter, ITER_BVEC | WRITE, &bv, 1, wlen);
 
 	ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, wlen,