xfs
[Top] [All Lists]

[PATCH v5 0/4] xfs: seek_data/seek_hole refinements

To: xfs@xxxxxxxxxxx
Subject: [PATCH v5 0/4] xfs: seek_data/seek_hole refinements
From: Jeff Liu <jeff.liu@xxxxxxxxxx>
Date: Thu, 26 Jul 2012 16:55:08 +0800
Organization: Oracle
Reply-to: jeff.liu@xxxxxxxxxx
User-agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.2.28) Gecko/20120313 Thunderbird/3.1.20
Hello,

This is the v5 patch set for xfs_seek_data/xfs_seek_hole refinements with lookup
the offset from page cache for desired type.

I split those changes to 4 patches this time, maybe it could facilitate the 
review process.

v4->v5:
* Introduce a new routine xfs_find_get_desired_pgoff() to search the desired 
offset from page cache
  for unwritten and delayed-allocation extents.  This function is mainly based 
on the original xfs_has_unwritten_buffer() but
  revised to be able to fetch a hole offset directly for the refinement of 
xfs_seek_hole().
* Remove type argument from xfs_seek_data()/xfs_seek_hole() as it is already 
indicated by those function naming explicitly.

v3->v4:
xfs_seek_hole() refinement, suggested by Mark and Christoph.
* refine xfs_seek_hole() with unwritten extents search, treat it as a hole if 
no data buffer was found from page cache.
* s/goto out/break/g, break out of the extent maps reading loop rather than 'go 
to', I must have got my head up in the clouds when writing v3.
* xfs_has_unwritten_buffer(), remove 'offset <= XFS_FSB_TO_B(mp, last))' from 
BH state checking branch.
  The page index offset might less than '*start', so we will miss a data extent 
if so.
* xfs_has_unwritten_buffer(), don't reset '*offset' to ZERO if no data buffer 
was found because of xfs_seek_hole() will call this function to
   examine an unwritten extent has data or not. If not, it will use the 
returned '*offset' as a hole offset.  So set '*offset' to zero in
   xfs_has_unwritten_buffer() will lead to wrong result.
* avoid re-starting the next round search in both xfs_seek_data() and 
xfs_seek_hole() if the end offset of the 2nd extent map is hit the EOF.
  So for SEEK_DATA, it means there is no data extent beyond the current offset 
and return ENXIO, for SEEK_HOLE, return the file size to
  indicate hitting EOF.   The comments were also changed(s/reading offset not 
beyond/reading offset not beyond or hit EOF/)accordingly.

v2->v3:
Tested by Mark, hit BUG() for continuous unwritten extents without data wrote.
* xfs_seek_data(), remove BUG() and having extents map search in loop.

v1->v2:
suggested by Mark.
* xfs_has_unwritten_buffer(), use the input offset instead of bmap->br_startoff 
to calculate page index for data buffer probing.

Thanks,
-Jeff

<Prev in Thread] Current Thread [Next in Thread>