[PATCH 0/6 EARLY RFC] Btrfs: Get rid of whole page I/O.

[PATCH 0/6 EARLY RFC] Btrfs: Get rid of whole page I/O.

[Chandan Rajendra]

This patchset describes a possible solution to solve whole page I/O issue
present in Btrfs. Currently whole page I/O is being performed because
bio_vec->[bv_len, bv_offset] can be modified by the Block I/O layer and hence
cannot be used by Btrfs's endio functions to operate on the original file
offset range that the bio was intended for. 

This patch is a precursor to get subpagesize-blocksize Btrfs feature to work
correctly as it would be required to do non-whole page I/O.

The patchset makes use of patches posted earlier by Chandra Seetharaman
(http://article.gmane.org/gmane.comp.file-systems.btrfs/30737). I believe that
getting non whole page I/O to work properly is the first issue to tackle
before working on the review comments posted for Chandra's patches. Please
correct me if my assumption is incorrect.

The patchset has been tested with only a handful of trivial I/O tests i.e. I/O
on different kinds of files (e.g. files with holes, etc). The tests were run
on 2k and 4k blocksized instances of Btrfs code (on x86_64) that has initial
support for mounting a 2k blocksized Btrfs filesystem. The corresponding git
trees can be found at
https://github.com/chandanr/btrfs-progs/tree/subpagesize-blocksize and
https://github.com/chandanr/linux/tree/btrfs/subpagesize-blocksize

The modified code is limited in feature i.e. the following are its known
limitations,
1. xfstests suite has not been executed.
2. Direct I/O has not been tested. 
3. Compression does not work with 2k blocksize filesystem instance.
4. fallocate does not work with 2k blocksize.
5. Data checksum does not work for 2k blocksize. Hence filesystem should be
   mounted by passing the nodatasum option.
6. Performance implications are unknown.

Chandan Rajendra (4):
  Btrfs: subpagesize-blocksize: Get rid of whole page reads.
  Btrfs: subpagesize-blocksize: Get rid of whole page writes.
  Btrfs: subpagesize-blocksize: Work with extents aligned to blocksize.
  Btrfs: subpagesize-blocksize: Hardcode MAX_EXTENT_BUFFERS_PER_PAGE to 2.

Chandra Seetharaman (2):
  Btrfs: subpagesize-blocksize: Define extent_buffer_head
  Btrfs: subpagesize-blocksize: Allow mounting filesystems where sectorsize != PAGE_SIZE

 fs/btrfs/backref.c           |   2 +-
 fs/btrfs/ctree.c             |   2 +-
 fs/btrfs/ctree.h             |   6 +-
 fs/btrfs/disk-io.c           | 117 ++++----
 fs/btrfs/extent-tree.c       |   6 +-
 fs/btrfs/extent_io.c         | 623 ++++++++++++++++++++++++++++---------------
 fs/btrfs/extent_io.h         |  60 ++++-
 fs/btrfs/file.c              |  13 +-
 fs/btrfs/volumes.c           |   2 +-
 fs/btrfs/volumes.h           |   3 +
 include/trace/events/btrfs.h |   2 +-
 11 files changed, 554 insertions(+), 282 deletions(-)

-- 
1.8.3.1

[PATCH 1/6] Btrfs: subpagesize-blocksize: Get rid of whole page reads.

[Chandan Rajendra]

bio_vec->{bv_offset, bv_len} cannot be relied upon by the end bio functions
to track the file offset range operated on by the bio. Hence this patch adds
two new members to 'struct btrfs_io_bio' to track the file offset range.

This patch also brings back check_page_locked() to reliably unlock pages in
readpage's end bio function.

Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
---
 fs/btrfs/extent_io.c | 122 +++++++++++++++++++++++++++++++++------------------
 fs/btrfs/volumes.h   |   3 ++
 2 files changed, 82 insertions(+), 43 deletions(-)

diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index fbe501d..5a65aee 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -1943,15 +1943,31 @@ int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
  * helper function to set a given page up to date if all the
  * extents in the tree for that page are up to date
  */
-static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
+static void check_page_uptodate(struct extent_io_tree *tree, struct page *page,
+				struct extent_state *cached)
 {
 	u64 start = page_offset(page);
 	u64 end = start + PAGE_CACHE_SIZE - 1;
-	if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
+	if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, cached))
 		SetPageUptodate(page);
 }
 
 /*
+ * helper function to unlock a page if all the extents in the tree
+ * for that page are unlocked
+ */
+static void check_page_locked(struct extent_io_tree *tree, struct page *page)
+{
+	u64 start = page_offset(page);
+	u64 end = start + PAGE_CACHE_SIZE - 1;
+
+	if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
+		unlock_page(page);
+	}
+}
+
+
+/*
  * When IO fails, either with EIO or csum verification fails, we
  * try other mirrors that might have a good copy of the data.  This
  * io_failure_record is used to record state as we go through all the
@@ -2414,16 +2430,33 @@ static void end_bio_extent_writepage(struct bio *bio, int err)
 	bio_put(bio);
 }
 
-static void
-endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
-			      int uptodate)
+static void unlock_extent_and_page(struct address_space *mapping,
+				   struct extent_io_tree *tree,
+				   struct btrfs_io_bio *io_bio)
 {
-	struct extent_state *cached = NULL;
-	u64 end = start + len - 1;
+	pgoff_t index;
+	u64 offset, len;
+	/*
+	 * This btrfs_io_bio may span multiple pages.
+	 * We need to unlock the pages convered by them
+	 * if we got endio callback for all the blocks in the page.
+	 * btrfs_io_bio also contain "contigous blocks of the file"
+	 * look at submit_extent_page for more details.
+	 */
 
-	if (uptodate && tree->track_uptodate)
-		set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
-	unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
+	offset = io_bio->start_offset;
+	len    = io_bio->len;
+	unlock_extent(tree, offset, offset + len - 1);
+
+	index = offset >> PAGE_CACHE_SHIFT;
+	while (offset < io_bio->start_offset + len) {
+		struct page *page;
+		page = find_get_page(mapping, index);
+		check_page_locked(tree, page);
+		page_cache_release(page);
+		index++;
+		offset += PAGE_CACHE_SIZE;
+	}
 }
 
 /*
@@ -2443,13 +2476,13 @@ static void end_bio_extent_readpage(struct bio *bio, int err)
 	struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
 	struct bio_vec *bvec = bio->bi_io_vec;
 	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
+	struct address_space *mapping = bio->bi_io_vec->bv_page->mapping;
 	struct extent_io_tree *tree;
+	struct extent_state *cached = NULL;
 	u64 offset = 0;
 	u64 start;
 	u64 end;
 	u64 len;
-	u64 extent_start = 0;
-	u64 extent_len = 0;
 	int mirror;
 	int ret;
 
@@ -2482,8 +2515,8 @@ static void end_bio_extent_readpage(struct bio *bio, int err)
 					bvec->bv_offset, bvec->bv_len);
 		}
 
-		start = page_offset(page);
-		end = start + bvec->bv_offset + bvec->bv_len - 1;
+		start = page_offset(page) + bvec->bv_offset;
+		end = start + bvec->bv_len - 1;
 		len = bvec->bv_len;
 
 		if (++bvec <= bvec_end)
@@ -2540,40 +2573,24 @@ readpage_ok:
 			offset = i_size & (PAGE_CACHE_SIZE-1);
 			if (page->index == end_index && offset)
 				zero_user_segment(page, offset, PAGE_CACHE_SIZE);
-			SetPageUptodate(page);
+			if (tree->track_uptodate)
+				set_extent_uptodate(tree, start, end, &cached,
+						    GFP_ATOMIC);
 		} else {
 			ClearPageUptodate(page);
 			SetPageError(page);
 		}
-		unlock_page(page);
-		offset += len;
 
-		if (unlikely(!uptodate)) {
-			if (extent_len) {
-				endio_readpage_release_extent(tree,
-							      extent_start,
-							      extent_len, 1);
-				extent_start = 0;
-				extent_len = 0;
-			}
-			endio_readpage_release_extent(tree, start,
-						      end - start + 1, 0);
-		} else if (!extent_len) {
-			extent_start = start;
-			extent_len = end + 1 - start;
-		} else if (extent_start + extent_len == start) {
-			extent_len += end + 1 - start;
-		} else {
-			endio_readpage_release_extent(tree, extent_start,
-						      extent_len, uptodate);
-			extent_start = start;
-			extent_len = end + 1 - start;
-		}
+		offset += len;
+		/*
+		 * Check whether the page in the bvec can be marked uptodate
+		 */
+		check_page_uptodate(tree, page, cached);
 	} while (bvec <= bvec_end);
-
-	if (extent_len)
-		endio_readpage_release_extent(tree, extent_start, extent_len,
-					      uptodate);
+	/*
+	 * Unlock the btrfs_bio and associated page
+	 */
+	unlock_extent_and_page(mapping, tree, io_bio);
 	if (io_bio->end_io)
 		io_bio->end_io(io_bio, err);
 	bio_put(bio);
@@ -2700,6 +2717,18 @@ static int submit_extent_page(int rw, struct extent_io_tree *tree,
 		else
 			contig = bio_end_sector(bio) == sector;
 
+		if (contig) {
+			/*
+			 * Check whether we are contig if file offsets.
+			 * We should mostly be for readpage/readpages
+			 * We need to do this because we use btrfs_io_bio
+			 * start_offset and len to unlock in endio routines.
+			 */
+			if ((page_offset(page) + offset) !=
+					(btrfs_io_bio(bio)->start_offset +
+					 btrfs_io_bio(bio)->len))
+				contig = 0;
+		}
 		if (prev_bio_flags != bio_flags || !contig ||
 		    merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
 		    bio_add_page(bio, page, page_size, offset) < page_size) {
@@ -2709,6 +2738,11 @@ static int submit_extent_page(int rw, struct extent_io_tree *tree,
 				return ret;
 			bio = NULL;
 		} else {
+			/*
+			 * update btrfs_io_bio len. So that we can unlock
+			 * correctly in end_io callback.
+			 */
+			btrfs_io_bio(bio)->len += page_size;
 			return 0;
 		}
 	}
@@ -2724,6 +2758,8 @@ static int submit_extent_page(int rw, struct extent_io_tree *tree,
 	bio_add_page(bio, page, page_size, offset);
 	bio->bi_end_io = end_io_func;
 	bio->bi_private = tree;
+	btrfs_io_bio(bio)->start_offset = page_offset(page) + offset;
+	btrfs_io_bio(bio)->len = page_size;
 
 	if (bio_ret)
 		*bio_ret = bio;
@@ -2914,7 +2950,7 @@ static int __do_readpage(struct extent_io_tree *tree,
 		/* the get_extent function already copied into the page */
 		if (test_range_bit(tree, cur, cur_end,
 				   EXTENT_UPTODATE, 1, NULL)) {
-			check_page_uptodate(tree, page);
+			check_page_uptodate(tree, page, NULL);
 			if (!parent_locked)
 				unlock_extent(tree, cur, cur + iosize - 1);
 			cur = cur + iosize;
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
index 80754f9..2e16f4d 100644
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -174,6 +174,9 @@ struct btrfs_io_bio {
 	u8 *csum_allocated;
 	btrfs_io_bio_end_io_t *end_io;
 	struct bio bio;
+	/* Track file offset range operated on by the bio.*/
+	u64 start_offset;
+	u64 len;
 };
 
 static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio)
-- 
1.8.3.1

Re: [PATCH 1/6] Btrfs: subpagesize-blocksize: Get rid of whole page reads.

[Liu Bo]

On Wed, Mar 12, 2014 at 07:50:28PM +0530, Chandan Rajendra wrote:
> bio_vec->{bv_offset, bv_len} cannot be relied upon by the end bio functions
> to track the file offset range operated on by the bio. Hence this patch adds
> two new members to 'struct btrfs_io_bio' to track the file offset range.
> 
> This patch also brings back check_page_locked() to reliably unlock pages in
> readpage's end bio function.
> 
> Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
> ---
>  fs/btrfs/extent_io.c | 122 +++++++++++++++++++++++++++++++++------------------
>  fs/btrfs/volumes.h   |   3 ++
>  2 files changed, 82 insertions(+), 43 deletions(-)
> 
> diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
> index fbe501d..5a65aee 100644
> --- a/fs/btrfs/extent_io.c
> +++ b/fs/btrfs/extent_io.c
> @@ -1943,15 +1943,31 @@ int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
>   * helper function to set a given page up to date if all the
>   * extents in the tree for that page are up to date
>   */
> -static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
> +static void check_page_uptodate(struct extent_io_tree *tree, struct page *page,
> +				struct extent_state *cached)
>  {
>  	u64 start = page_offset(page);
>  	u64 end = start + PAGE_CACHE_SIZE - 1;
> -	if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
> +	if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, cached))
>  		SetPageUptodate(page);
>  }
>  
>  /*
> + * helper function to unlock a page if all the extents in the tree
> + * for that page are unlocked
> + */
> +static void check_page_locked(struct extent_io_tree *tree, struct page *page)
> +{
> +	u64 start = page_offset(page);
> +	u64 end = start + PAGE_CACHE_SIZE - 1;
> +
> +	if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
> +		unlock_page(page);
> +	}
> +}
> +
> +
> +/*
>   * When IO fails, either with EIO or csum verification fails, we
>   * try other mirrors that might have a good copy of the data.  This
>   * io_failure_record is used to record state as we go through all the
> @@ -2414,16 +2430,33 @@ static void end_bio_extent_writepage(struct bio *bio, int err)
>  	bio_put(bio);
>  }
>  
> -static void
> -endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
> -			      int uptodate)
> +static void unlock_extent_and_page(struct address_space *mapping,
> +				   struct extent_io_tree *tree,
> +				   struct btrfs_io_bio *io_bio)
>  {
> -	struct extent_state *cached = NULL;
> -	u64 end = start + len - 1;
> +	pgoff_t index;
> +	u64 offset, len;
> +	/*
> +	 * This btrfs_io_bio may span multiple pages.
> +	 * We need to unlock the pages convered by them
> +	 * if we got endio callback for all the blocks in the page.
> +	 * btrfs_io_bio also contain "contigous blocks of the file"
> +	 * look at submit_extent_page for more details.
> +	 */
>  
> -	if (uptodate && tree->track_uptodate)
> -		set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
> -	unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
> +	offset = io_bio->start_offset;
> +	len    = io_bio->len;
> +	unlock_extent(tree, offset, offset + len - 1);
> +
> +	index = offset >> PAGE_CACHE_SHIFT;
> +	while (offset < io_bio->start_offset + len) {
> +		struct page *page;
> +		page = find_get_page(mapping, index);
> +		check_page_locked(tree, page);
> +		page_cache_release(page);
> +		index++;
> +		offset += PAGE_CACHE_SIZE;
> +	}
>  }
>  
>  /*
> @@ -2443,13 +2476,13 @@ static void end_bio_extent_readpage(struct bio *bio, int err)
>  	struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
>  	struct bio_vec *bvec = bio->bi_io_vec;
>  	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
> +	struct address_space *mapping = bio->bi_io_vec->bv_page->mapping;
>  	struct extent_io_tree *tree;
> +	struct extent_state *cached = NULL;
>  	u64 offset = 0;
>  	u64 start;
>  	u64 end;
>  	u64 len;
> -	u64 extent_start = 0;
> -	u64 extent_len = 0;
>  	int mirror;
>  	int ret;
>  
> @@ -2482,8 +2515,8 @@ static void end_bio_extent_readpage(struct bio *bio, int err)
>  					bvec->bv_offset, bvec->bv_len);
>  		}
>  
> -		start = page_offset(page);
> -		end = start + bvec->bv_offset + bvec->bv_len - 1;
> +		start = page_offset(page) + bvec->bv_offset;
> +		end = start + bvec->bv_len - 1;
>  		len = bvec->bv_len;
>  
>  		if (++bvec <= bvec_end)
> @@ -2540,40 +2573,24 @@ readpage_ok:
>  			offset = i_size & (PAGE_CACHE_SIZE-1);
>  			if (page->index == end_index && offset)
>  				zero_user_segment(page, offset, PAGE_CACHE_SIZE);
> -			SetPageUptodate(page);
> +			if (tree->track_uptodate)
> +				set_extent_uptodate(tree, start, end, &cached,
> +						    GFP_ATOMIC);
>  		} else {
>  			ClearPageUptodate(page);
>  			SetPageError(page);
>  		}
> -		unlock_page(page);
> -		offset += len;
>  
> -		if (unlikely(!uptodate)) {
> -			if (extent_len) {
> -				endio_readpage_release_extent(tree,
> -							      extent_start,
> -							      extent_len, 1);
> -				extent_start = 0;
> -				extent_len = 0;
> -			}
> -			endio_readpage_release_extent(tree, start,
> -						      end - start + 1, 0);
> -		} else if (!extent_len) {
> -			extent_start = start;
> -			extent_len = end + 1 - start;
> -		} else if (extent_start + extent_len == start) {
> -			extent_len += end + 1 - start;
> -		} else {
> -			endio_readpage_release_extent(tree, extent_start,
> -						      extent_len, uptodate);
> -			extent_start = start;
> -			extent_len = end + 1 - start;
> -		}
> +		offset += len;
> +		/*
> +		 * Check whether the page in the bvec can be marked uptodate
> +		 */
> +		check_page_uptodate(tree, page, cached);
>  	} while (bvec <= bvec_end);
> -
> -	if (extent_len)
> -		endio_readpage_release_extent(tree, extent_start, extent_len,
> -					      uptodate);
> +	/*
> +	 * Unlock the btrfs_bio and associated page
> +	 */
> +	unlock_extent_and_page(mapping, tree, io_bio);
>  	if (io_bio->end_io)
>  		io_bio->end_io(io_bio, err);
>  	bio_put(bio);
> @@ -2700,6 +2717,18 @@ static int submit_extent_page(int rw, struct extent_io_tree *tree,
>  		else
>  			contig = bio_end_sector(bio) == sector;
>  
> +		if (contig) {
> +			/*
> +			 * Check whether we are contig if file offsets.
> +			 * We should mostly be for readpage/readpages
> +			 * We need to do this because we use btrfs_io_bio
> +			 * start_offset and len to unlock in endio routines.
> +			 */
> +			if ((page_offset(page) + offset) !=
> +					(btrfs_io_bio(bio)->start_offset +
> +					 btrfs_io_bio(bio)->len))
> +				contig = 0;
> +		}
>  		if (prev_bio_flags != bio_flags || !contig ||
>  		    merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
>  		    bio_add_page(bio, page, page_size, offset) < page_size) {
> @@ -2709,6 +2738,11 @@ static int submit_extent_page(int rw, struct extent_io_tree *tree,
>  				return ret;
>  			bio = NULL;
>  		} else {
> +			/*
> +			 * update btrfs_io_bio len. So that we can unlock
> +			 * correctly in end_io callback.
> +			 */
> +			btrfs_io_bio(bio)->len += page_size;
>  			return 0;
>  		}
>  	}
> @@ -2724,6 +2758,8 @@ static int submit_extent_page(int rw, struct extent_io_tree *tree,
>  	bio_add_page(bio, page, page_size, offset);
>  	bio->bi_end_io = end_io_func;
>  	bio->bi_private = tree;
> +	btrfs_io_bio(bio)->start_offset = page_offset(page) + offset;
> +	btrfs_io_bio(bio)->len = page_size;
>  
>  	if (bio_ret)
>  		*bio_ret = bio;
> @@ -2914,7 +2950,7 @@ static int __do_readpage(struct extent_io_tree *tree,
>  		/* the get_extent function already copied into the page */
>  		if (test_range_bit(tree, cur, cur_end,
>  				   EXTENT_UPTODATE, 1, NULL)) {
> -			check_page_uptodate(tree, page);
> +			check_page_uptodate(tree, page, NULL);
>  			if (!parent_locked)
>  				unlock_extent(tree, cur, cur + iosize - 1);
>  			cur = cur + iosize;
> diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
> index 80754f9..2e16f4d 100644
> --- a/fs/btrfs/volumes.h
> +++ b/fs/btrfs/volumes.h
> @@ -174,6 +174,9 @@ struct btrfs_io_bio {
>  	u8 *csum_allocated;
>  	btrfs_io_bio_end_io_t *end_io;
>  	struct bio bio;
> +	/* Track file offset range operated on by the bio.*/
> +	u64 start_offset;
> +	u64 len;

These should be put in front of "struct bio bio",
otherwise, it might lead to errors, according to bioset_create()'s comments,

----------------------------------------------------------------------
"Note that the bio must be embedded at the END of that structure always,    
or things will break badly."
----------------------------------------------------------------------

-liubo

Re: [PATCH 1/6] Btrfs: subpagesize-blocksize: Get rid of whole page reads.

[chandan Rajendra]

> These should be put in front of "struct bio bio",
> otherwise, it might lead to errors, according to bioset_create()'s comments,
> 
> ----------------------------------------------------------------------
> "Note that the bio must be embedded at the END of that structure always,    
> or things will break badly."
> ----------------------------------------------------------------------
> 

Thank you for pointing that out. I will fix it.

Thanks,
chandan

[PATCH 2/6] Btrfs: subpagesize-blocksize: Get rid of whole page writes.

[Chandan Rajendra]

This commit brings back functions that set/clear EXTENT_WRITEBACK bits. These
are required to reliably clear PG_writeback page flag.

Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
---
 fs/btrfs/extent_io.c | 74 +++++++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 71 insertions(+), 3 deletions(-)

diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 5a65aee..9e77645 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -1293,6 +1293,20 @@ int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
 				cached_state, mask);
 }
 
+static int set_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
+				struct extent_state **cached_state, gfp_t mask)
+{
+	return set_extent_bit(tree, start, end, EXTENT_WRITEBACK, NULL,
+			cached_state, mask);
+}
+
+static int clear_extent_writeback(struct extent_io_tree *tree, u64 start, u64 end,
+				struct extent_state **cached_state, gfp_t mask)
+{
+	return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0,
+				cached_state, mask);
+}
+
 /*
  * either insert or lock state struct between start and end use mask to tell
  * us if waiting is desired.
@@ -1399,6 +1413,7 @@ static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
 		page_cache_release(page);
 		index++;
 	}
+	set_extent_writeback(tree, start, end, NULL, GFP_NOFS);
 	return 0;
 }
 
@@ -1966,6 +1981,14 @@ static void check_page_locked(struct extent_io_tree *tree, struct page *page)
 	}
 }
 
+static void check_page_writeback(struct extent_io_tree *tree, struct page *page)
+{
+	u64 start = page_offset(page);
+	u64 end = start + PAGE_CACHE_SIZE - 1;
+
+	if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0, NULL))
+		end_page_writeback(page);
+}
 
 /*
  * When IO fails, either with EIO or csum verification fails, we
@@ -2380,6 +2403,32 @@ int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
 	return 0;
 }
 
+static void clear_extent_and_page_writeback(struct address_space *mapping,
+					struct extent_io_tree *tree,
+					struct btrfs_io_bio *io_bio)
+{
+	struct page *page;
+	pgoff_t index;
+	u64 offset, len;
+
+	offset	= io_bio->start_offset;
+	len	= io_bio->len;
+
+	clear_extent_writeback(tree, offset, offset + len - 1, NULL,
+			GFP_ATOMIC);
+
+	index = offset >> PAGE_CACHE_SHIFT;
+	while (offset < io_bio->start_offset + len) {
+		page = find_get_page(mapping, index);
+		check_page_writeback(tree, page);
+		page_cache_release(page);
+		index++;
+		offset += PAGE_CACHE_SIZE;
+	}
+
+	unlock_extent(tree, io_bio->start_offset, io_bio->start_offset + len - 1);
+}
+
 /*
  * after a writepage IO is done, we need to:
  * clear the uptodate bits on error
@@ -2391,6 +2440,9 @@ int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
  */
 static void end_bio_extent_writepage(struct bio *bio, int err)
 {
+	struct address_space *mapping =  bio->bi_io_vec->bv_page->mapping;
+	struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
+	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
 	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 	u64 start;
 	u64 end;
@@ -2415,8 +2467,8 @@ static void end_bio_extent_writepage(struct bio *bio, int err)
 					bvec->bv_offset, bvec->bv_len);
 		}
 
-		start = page_offset(page);
-		end = start + bvec->bv_offset + bvec->bv_len - 1;
+		start = page_offset(page) + bvec->bv_offset;
+		end = start + bvec->bv_len - 1;
 
 		if (--bvec >= bio->bi_io_vec)
 			prefetchw(&bvec->bv_page->flags);
@@ -2424,9 +2476,10 @@ static void end_bio_extent_writepage(struct bio *bio, int err)
 		if (end_extent_writepage(page, err, start, end))
 			continue;
 
-		end_page_writeback(page);
 	} while (bvec >= bio->bi_io_vec);
 
+	clear_extent_and_page_writeback(mapping, tree, io_bio);
+
 	bio_put(bio);
 }
 
@@ -3153,6 +3206,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 	u64 last_byte = i_size_read(inode);
 	u64 block_start;
 	u64 iosize;
+	u64 unlock_start = start;
 	sector_t sector;
 	struct extent_state *cached_state = NULL;
 	struct extent_map *em;
@@ -3235,6 +3289,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 			/* File system has been set read-only */
 			if (ret) {
 				SetPageError(page);
+				unlock_start = page_end + 1;
 				goto done;
 			}
 			/*
@@ -3270,10 +3325,14 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 			goto done_unlocked;
 		}
 	}
+
+	lock_extent(tree, start, page_end);
+
 	if (tree->ops && tree->ops->writepage_start_hook) {
 		ret = tree->ops->writepage_start_hook(page, start,
 						      page_end);
 		if (ret) {
+			unlock_extent(tree, start, page_end);
 			/* Fixup worker will requeue */
 			if (ret == -EBUSY)
 				wbc->pages_skipped++;
@@ -3294,9 +3353,11 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 
 	end = page_end;
 	if (last_byte <= start) {
+		unlock_extent(tree, start, page_end);
 		if (tree->ops && tree->ops->writepage_end_io_hook)
 			tree->ops->writepage_end_io_hook(page, start,
 							 page_end, NULL, 1);
+		unlock_start = page_end + 1;
 		goto done;
 	}
 
@@ -3304,9 +3365,11 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 
 	while (cur <= end) {
 		if (cur >= last_byte) {
+			unlock_extent(tree, unlock_start, page_end);
 			if (tree->ops && tree->ops->writepage_end_io_hook)
 				tree->ops->writepage_end_io_hook(page, cur,
 							 page_end, NULL, 1);
+			unlock_start = page_end + 1;
 			break;
 		}
 		em = epd->get_extent(inode, page, pg_offset, cur,
@@ -3334,6 +3397,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 		 */
 		if (compressed || block_start == EXTENT_MAP_HOLE ||
 		    block_start == EXTENT_MAP_INLINE) {
+			unlock_extent(tree, unlock_start, cur + iosize - 1);
 			/*
 			 * end_io notification does not happen here for
 			 * compressed extents
@@ -3353,6 +3417,7 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 
 			cur += iosize;
 			pg_offset += iosize;
+			unlock_start = cur;
 			continue;
 		}
 		/* leave this out until we have a page_mkwrite call */
@@ -3399,6 +3464,9 @@ done:
 		set_page_writeback(page);
 		end_page_writeback(page);
 	}
+	if (unlock_start <= page_end)
+		unlock_extent(tree, unlock_start, page_end);
+
 	unlock_page(page);
 
 done_unlocked:
-- 
1.8.3.1

[PATCH 3/6] Btrfs: subpagesize-blocksize: Work with extents aligned to blocksize.

[Chandan Rajendra]

This is a small part of a bigger set of fixes needed to get
subpagesize-blocksize to function properly. This patch is required to get 2k
blocksize writes to work correctly.

Without this patch, In lock_and_cleanup_extent_if_need() two consecutive 2k
block writes that map to first and second 2k parts of a 4k page would result
in the second 2k write to *clear* out the EXTENT_DELALLOC bit of the first 2k
block. Fix this.

Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
---
 fs/btrfs/file.c | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 006af2f..1aa8fae 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -1339,18 +1339,21 @@ fail:
 static noinline int
 lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
 				size_t num_pages, loff_t pos,
+				size_t write_bytes,
 				u64 *lockstart, u64 *lockend,
 				struct extent_state **cached_state)
 {
+	struct btrfs_root *root = BTRFS_I(inode)->root;
 	u64 start_pos;
 	u64 last_pos;
 	int i;
 	int ret = 0;
 
-	start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
-	last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
+       start_pos = pos & ~((u64)root->sectorsize - 1);
+       last_pos = start_pos
+               + ALIGN(pos + write_bytes - start_pos, root->sectorsize) - 1;
 
-	if (start_pos < inode->i_size) {
+       if (start_pos < inode->i_size) {
 		struct btrfs_ordered_extent *ordered;
 		lock_extent_bits(&BTRFS_I(inode)->io_tree,
 				 start_pos, last_pos, 0, cached_state);
@@ -1536,8 +1539,8 @@ again:
 			break;
 
 		ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
-						      pos, &lockstart, &lockend,
-						      &cached_state);
+						pos, write_bytes, &lockstart, &lockend,
+						&cached_state);
 		if (ret < 0) {
 			if (ret == -EAGAIN)
 				goto again;
-- 
1.8.3.1

[PATCH 4/6] Btrfs: subpagesize-blocksize: Define extent_buffer_head

[Chandan Rajendra]

From: Chandra Seetharaman <sekharan@us.ibm.com>

In order to handle multiple extent buffers per page, first we
need to create a way to handle all the extent buffers that
are attached to a page.

This patch creates a new data structure eb_head, and moves
fields that are common to all extent buffers in a page from
extent buffer to eb_head.

This also adds changes that are needed to handle multiple
extent buffers per page case.

Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
---
 fs/btrfs/backref.c           |   2 +-
 fs/btrfs/ctree.c             |   2 +-
 fs/btrfs/ctree.h             |   6 +-
 fs/btrfs/disk-io.c           | 111 +++++++----
 fs/btrfs/extent-tree.c       |   6 +-
 fs/btrfs/extent_io.c         | 427 ++++++++++++++++++++++++++-----------------
 fs/btrfs/extent_io.h         |  55 ++++--
 fs/btrfs/volumes.c           |   2 +-
 include/trace/events/btrfs.h |   2 +-
 9 files changed, 388 insertions(+), 225 deletions(-)

diff --git a/fs/btrfs/backref.c b/fs/btrfs/backref.c
index a88da72..603ae44 100644
--- a/fs/btrfs/backref.c
+++ b/fs/btrfs/backref.c
@@ -1272,7 +1272,7 @@ char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
 		eb = path->nodes[0];
 		/* make sure we can use eb after releasing the path */
 		if (eb != eb_in) {
-			atomic_inc(&eb->refs);
+			atomic_inc(&eb_head(eb)->refs);
 			btrfs_tree_read_lock(eb);
 			btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 		}
diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index cbd3a7d..0d4ad91 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -169,7 +169,7 @@ struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
 		 * the inc_not_zero dance and if it doesn't work then
 		 * synchronize_rcu and try again.
 		 */
-		if (atomic_inc_not_zero(&eb->refs)) {
+		if (atomic_inc_not_zero(&eb_head(eb)->refs)) {
 			rcu_read_unlock();
 			break;
 		}
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index dac6653..901ada2 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -2138,14 +2138,16 @@ static inline void btrfs_set_token_##name(struct extent_buffer *eb,	\
 #define BTRFS_SETGET_HEADER_FUNCS(name, type, member, bits)		\
 static inline u##bits btrfs_##name(struct extent_buffer *eb)		\
 {									\
-	type *p = page_address(eb->pages[0]);				\
+	type *p = page_address(eb_head(eb)->pages[0]) +			\
+				(eb->start & (PAGE_CACHE_SIZE -1));	\
 	u##bits res = le##bits##_to_cpu(p->member);			\
 	return res;							\
 }									\
 static inline void btrfs_set_##name(struct extent_buffer *eb,		\
 				    u##bits val)			\
 {									\
-	type *p = page_address(eb->pages[0]);				\
+	type *p = page_address(eb_head(eb)->pages[0]) +			\
+				(eb->start & (PAGE_CACHE_SIZE -1));	\
 	p->member = cpu_to_le##bits(val);				\
 }
 
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index cc1b423..796d889 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -413,7 +413,7 @@ static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 	int mirror_num = 0;
 	int failed_mirror = 0;
 
-	clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
+	clear_bit(EXTENT_BUFFER_CORRUPT, &eb_head(eb)->bflags);
 	io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 	while (1) {
 		ret = read_extent_buffer_pages(io_tree, eb, start,
@@ -432,7 +432,7 @@ static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 		 * there is no reason to read the other copies, they won't be
 		 * any less wrong.
 		 */
-		if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
+		if (test_bit(EXTENT_BUFFER_CORRUPT, &eb_head(eb)->bflags))
 			break;
 
 		num_copies = btrfs_num_copies(root->fs_info,
@@ -442,7 +442,7 @@ static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 
 		if (!failed_mirror) {
 			failed = 1;
-			failed_mirror = eb->read_mirror;
+			failed_mirror = eb_head(eb)->read_mirror;
 		}
 
 		mirror_num++;
@@ -466,17 +466,20 @@ static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 
 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 {
-	u64 start = page_offset(page);
 	u64 found_start;
 	struct extent_buffer *eb;
+	struct extent_buffer_head *eb_head;
 
-	eb = (struct extent_buffer *)page->private;
-	if (page != eb->pages[0])
+	eb_head = (struct extent_buffer_head *)page->private;
+	if (page != eb_head->pages[0])
 		return 0;
-	found_start = btrfs_header_bytenr(eb);
-	if (WARN_ON(found_start != start || !PageUptodate(page)))
+	if (WARN_ON(!PageUptodate(page)))
 		return 0;
-	csum_tree_block(root, eb, 0);
+	for (eb = &eb_head->extent_buf[0]; eb->start; eb++) {
+		found_start = btrfs_header_bytenr(eb);
+		if (found_start == eb->start)
+			csum_tree_block(root, eb, 0);
+	}
 	return 0;
 }
 
@@ -573,24 +576,35 @@ static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 	struct extent_buffer *eb;
 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 	int ret = 0;
-	int reads_done;
+	int reads_done = 0;
+	struct extent_buffer_head *eb_head;
 
 	if (!page->private)
 		goto out;
 
-	eb = (struct extent_buffer *)page->private;
+
+	eb_head = (struct extent_buffer_head *)page->private;
+
+        /* Get the eb corresponding to this IO */
+	eb = eb_head->io_eb;
+	if (!eb) {
+		ret = -EIO;
+		goto err;
+	}
 
 	/* the pending IO might have been the only thing that kept this buffer
 	 * in memory.  Make sure we have a ref for all this other checks
 	 */
 	extent_buffer_get(eb);
 
-	reads_done = atomic_dec_and_test(&eb->io_pages);
+	reads_done = atomic_dec_and_test(&eb_head->io_pages);
 	if (!reads_done)
 		goto err;
 
-	eb->read_mirror = mirror;
-	if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
+	eb_head->io_eb = NULL;
+
+	eb_head->read_mirror = mirror;
+	if (test_bit(EXTENT_BUFFER_IOERR, &eb_head->bflags)) {
 		ret = -EIO;
 		goto err;
 	}
@@ -632,7 +646,7 @@ static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 	 * return -EIO.
 	 */
 	if (found_level == 0 && check_leaf(root, eb)) {
-		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
+		set_bit(EXTENT_BUFFER_CORRUPT, &eb_head->bflags);
 		ret = -EIO;
 	}
 
@@ -640,7 +654,7 @@ static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 		set_extent_buffer_uptodate(eb);
 err:
 	if (reads_done &&
-	    test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
+	    test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb_head->bflags))
 		btree_readahead_hook(root, eb, eb->start, ret);
 
 	if (ret) {
@@ -649,7 +663,7 @@ err:
 		 * again, we have to make sure it has something
 		 * to decrement
 		 */
-		atomic_inc(&eb->io_pages);
+		atomic_inc(&eb_head->io_pages);
 		clear_extent_buffer_uptodate(eb);
 	}
 	free_extent_buffer(eb);
@@ -659,15 +673,22 @@ out:
 
 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 {
+	struct extent_buffer_head *eb_head
+			=  (struct extent_buffer_head *)page->private;
 	struct extent_buffer *eb;
 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 
-	eb = (struct extent_buffer *)page->private;
-	set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
-	eb->read_mirror = failed_mirror;
-	atomic_dec(&eb->io_pages);
-	if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
+	set_bit(EXTENT_BUFFER_IOERR, &eb_head->bflags);
+	eb_head->read_mirror = failed_mirror;
+	atomic_dec(&eb_head->io_pages);
+	/* Get the eb corresponding to this IO */
+	eb = eb_head->io_eb;
+	if (!eb)
+		goto out;
+	eb_head->io_eb = NULL;
+	if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb_head->bflags))
 		btree_readahead_hook(root, eb, eb->start, -EIO);
+out:
 	return -EIO;	/* we fixed nothing */
 }
 
@@ -1017,14 +1038,20 @@ static void btree_invalidatepage(struct page *page, unsigned int offset,
 static int btree_set_page_dirty(struct page *page)
 {
 #ifdef DEBUG
+	struct extent_buffer_head *ebh;
 	struct extent_buffer *eb;
+	int i, dirty = 0;
 
 	BUG_ON(!PagePrivate(page));
-	eb = (struct extent_buffer *)page->private;
-	BUG_ON(!eb);
-	BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
-	BUG_ON(!atomic_read(&eb->refs));
-	btrfs_assert_tree_locked(eb);
+	ebh = (struct extent_buffer_head *)page->private;
+	BUG_ON(!ebh);
+	for (i = 0; i < MAX_EXTENT_BUFFERS_PER_PAGE && !dirty; i++) {
+		eb = &ebh->extent_buf[i];
+		dirty = test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->ebflags);
+	}
+	BUG_ON(dirty);
+	BUG_ON(!atomic_read(&ebh->refs));
+	btrfs_assert_tree_locked(&ebh->extent_buf[0]);
 #endif
 	return __set_page_dirty_nobuffers(page);
 }
@@ -1068,7 +1095,7 @@ int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 	if (!buf)
 		return 0;
 
-	set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
+	set_bit(EXTENT_BUFFER_READAHEAD, &eb_head(buf)->bflags);
 
 	ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
 				       btree_get_extent, mirror_num);
@@ -1077,7 +1104,7 @@ int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 		return ret;
 	}
 
-	if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
+	if (test_bit(EXTENT_BUFFER_CORRUPT, &eb_head(buf)->bflags)) {
 		free_extent_buffer(buf);
 		return -EIO;
 	} else if (extent_buffer_uptodate(buf)) {
@@ -1103,14 +1130,16 @@ struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 
 int btrfs_write_tree_block(struct extent_buffer *buf)
 {
-	return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
+	return filemap_fdatawrite_range(eb_head(buf)->pages[0]->mapping,
+					buf->start,
 					buf->start + buf->len - 1);
 }
 
 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 {
-	return filemap_fdatawait_range(buf->pages[0]->mapping,
-				       buf->start, buf->start + buf->len - 1);
+	return filemap_fdatawait_range(eb_head(buf)->pages[0]->mapping,
+					buf->start,
+					buf->start + buf->len - 1);
 }
 
 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
@@ -1141,7 +1170,8 @@ void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 	    fs_info->running_transaction->transid) {
 		btrfs_assert_tree_locked(buf);
 
-		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
+		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
+						&buf->ebflags)) {
 			__percpu_counter_add(&fs_info->dirty_metadata_bytes,
 					     -buf->len,
 					     fs_info->dirty_metadata_batch);
@@ -2613,7 +2643,8 @@ int open_ctree(struct super_block *sb,
 					   btrfs_super_chunk_root(disk_super),
 					   blocksize, generation);
 	if (!chunk_root->node ||
-	    !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
+		!test_bit(EXTENT_BUFFER_UPTODATE,
+			&eb_head(chunk_root->node)->bflags)) {
 		printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
 		       sb->s_id);
 		goto fail_tree_roots;
@@ -2652,7 +2683,8 @@ retry_root_backup:
 					  btrfs_super_root(disk_super),
 					  blocksize, generation);
 	if (!tree_root->node ||
-	    !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
+		!test_bit(EXTENT_BUFFER_UPTODATE,
+			&eb_head(tree_root->node)->bflags)) {
 		printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
 		       sb->s_id);
 
@@ -3642,7 +3674,7 @@ int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
 			  int atomic)
 {
 	int ret;
-	struct inode *btree_inode = buf->pages[0]->mapping->host;
+	struct inode *btree_inode = eb_head(buf)->pages[0]->mapping->host;
 
 	ret = extent_buffer_uptodate(buf);
 	if (!ret)
@@ -3675,7 +3707,7 @@ void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
 	if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
 		return;
 #endif
-	root = BTRFS_I(buf->pages[0]->mapping->host)->root;
+	root = BTRFS_I(eb_head(buf)->pages[0]->mapping->host)->root;
 	btrfs_assert_tree_locked(buf);
 	if (transid != root->fs_info->generation)
 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
@@ -3724,7 +3756,8 @@ void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
 
 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
 {
-	struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
+	struct btrfs_root *root =
+			BTRFS_I(eb_head(buf)->pages[0]->mapping->host)->root;
 	return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 }
 
@@ -3962,7 +3995,7 @@ static int btrfs_destroy_marked_extents(struct btrfs_root *root,
 			wait_on_extent_buffer_writeback(eb);
 
 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
-					       &eb->bflags))
+					       &eb->ebflags))
 				clear_extent_buffer_dirty(eb);
 			free_extent_buffer_stale(eb);
 		}
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 32312e0..0f677de 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -5984,7 +5984,7 @@ void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
 			goto out;
 		}
 
-		WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
+		WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->ebflags));
 
 		btrfs_add_free_space(cache, buf->start, buf->len);
 		btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
@@ -6001,7 +6001,7 @@ out:
 	 * Deleting the buffer, clear the corrupt flag since it doesn't matter
 	 * anymore.
 	 */
-	clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
+	clear_bit(EXTENT_BUFFER_CORRUPT, &eb_head(buf)->bflags);
 	btrfs_put_block_group(cache);
 }
 
@@ -6887,7 +6887,7 @@ btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 	btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
 	btrfs_tree_lock(buf);
 	clean_tree_block(trans, root, buf);
-	clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
+	clear_bit(EXTENT_BUFFER_STALE, &eb_head(buf)->bflags);
 
 	btrfs_set_lock_blocking(buf);
 	btrfs_set_buffer_uptodate(buf);
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 9e77645..25d34b0 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -54,8 +54,10 @@ void btrfs_leak_debug_del(struct list_head *entry)
 static inline
 void btrfs_leak_debug_check(void)
 {
+	int i;
 	struct extent_state *state;
 	struct extent_buffer *eb;
+	struct extent_buffer_head *ebh;
 
 	while (!list_empty(&states)) {
 		state = list_entry(states.next, struct extent_state, leak_list);
@@ -68,12 +70,17 @@ void btrfs_leak_debug_check(void)
 	}
 
 	while (!list_empty(&buffers)) {
-		eb = list_entry(buffers.next, struct extent_buffer, leak_list);
-		printk(KERN_ERR "BTRFS: buffer leak start %llu len %lu "
-		       "refs %d\n",
-		       eb->start, eb->len, atomic_read(&eb->refs));
-		list_del(&eb->leak_list);
-		kmem_cache_free(extent_buffer_cache, eb);
+		ebh = list_entry(buffers.next, struct extent_buffer_head, leak_list);
+		printk(KERN_ERR "btrfs buffer leak ");
+		for (i = 0; i < MAX_EXTENT_BUFFERS_PER_PAGE; i++) {
+			eb = &ebh->extent_buf[i];
+			if (!eb->start)
+				break;
+			printk(KERN_ERR "eb %p %llu:%lu ", eb, eb->start, eb->len);
+		}
+		printk(KERN_ERR "refs %d\n", atomic_read(&ebh->refs));
+		list_del(&ebh->leak_list);
+		kmem_cache_free(extent_buffer_cache, ebh);
 	}
 }
 
@@ -144,7 +151,7 @@ int __init extent_io_init(void)
 		return -ENOMEM;
 
 	extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
-			sizeof(struct extent_buffer), 0,
+			sizeof(struct extent_buffer_head), 0,
 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
 	if (!extent_buffer_cache)
 		goto free_state_cache;
@@ -2105,7 +2112,7 @@ int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start,
 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
 			 int mirror_num)
 {
-	u64 start = eb->start;
+	u64 start = eb_head(eb)->extent_buf[0].start;
 	unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
 	int ret = 0;
 
@@ -2822,15 +2829,15 @@ static int submit_extent_page(int rw, struct extent_io_tree *tree,
 	return ret;
 }
 
-static void attach_extent_buffer_page(struct extent_buffer *eb,
+static void attach_extent_buffer_page(struct extent_buffer_head *ebh,
 				      struct page *page)
 {
 	if (!PagePrivate(page)) {
 		SetPagePrivate(page);
 		page_cache_get(page);
-		set_page_private(page, (unsigned long)eb);
+		set_page_private(page, (unsigned long)ebh);
 	} else {
-		WARN_ON(page->private != (unsigned long)eb);
+		WARN_ON(page->private != (unsigned long)ebh);
 	}
 }
 
@@ -3484,17 +3491,19 @@ static int eb_wait(void *word)
 
 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
 {
-	wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
+	wait_on_bit(&eb_head(eb)->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
 		    TASK_UNINTERRUPTIBLE);
 }
 
-static int lock_extent_buffer_for_io(struct extent_buffer *eb,
+static int lock_extent_buffer_for_io(struct extent_buffer_head *ebh,
 				     struct btrfs_fs_info *fs_info,
 				     struct extent_page_data *epd)
 {
 	unsigned long i, num_pages;
 	int flush = 0;
+	bool dirty = false, dirty_arr[MAX_EXTENT_BUFFERS_PER_PAGE] = { 0 };
 	int ret = 0;
+	struct extent_buffer *eb = &ebh->extent_buf[0], *ebtemp;
 
 	if (!btrfs_try_tree_write_lock(eb)) {
 		flush = 1;
@@ -3502,7 +3511,7 @@ static int lock_extent_buffer_for_io(struct extent_buffer *eb,
 		btrfs_tree_lock(eb);
 	}
 
-	if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
+	if (test_bit(EXTENT_BUFFER_WRITEBACK, &ebh->bflags)) {
 		btrfs_tree_unlock(eb);
 		if (!epd->sync_io)
 			return 0;
@@ -3513,7 +3522,7 @@ static int lock_extent_buffer_for_io(struct extent_buffer *eb,
 		while (1) {
 			wait_on_extent_buffer_writeback(eb);
 			btrfs_tree_lock(eb);
-			if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
+			if (!test_bit(EXTENT_BUFFER_WRITEBACK, &ebh->bflags))
 				break;
 			btrfs_tree_unlock(eb);
 		}
@@ -3524,17 +3533,27 @@ static int lock_extent_buffer_for_io(struct extent_buffer *eb,
 	 * under IO since we can end up having no IO bits set for a short period
 	 * of time.
 	 */
-	spin_lock(&eb->refs_lock);
-	if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
-		set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
-		spin_unlock(&eb->refs_lock);
-		btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
-		__percpu_counter_add(&fs_info->dirty_metadata_bytes,
-				     -eb->len,
+	spin_lock(&ebh->refs_lock);
+	for (i = 0; i < MAX_EXTENT_BUFFERS_PER_PAGE; i++) {
+		ebtemp = &ebh->extent_buf[i];
+		dirty_arr[i] |= test_and_clear_bit(EXTENT_BUFFER_DIRTY, &ebtemp->ebflags);
+		dirty = dirty || dirty_arr[i];
+	}
+	if (dirty) {
+		set_bit(EXTENT_BUFFER_WRITEBACK, &ebh->bflags);
+		spin_unlock(&ebh->refs_lock);
+		for (i = 0; i < MAX_EXTENT_BUFFERS_PER_PAGE; i++) {
+			if (dirty_arr[i] == false)
+				continue;
+			ebtemp = &ebh->extent_buf[i];
+			btrfs_set_header_flag(ebtemp, BTRFS_HEADER_FLAG_WRITTEN);
+			__percpu_counter_add(&fs_info->dirty_metadata_bytes,
+				     -ebtemp->len,
 				     fs_info->dirty_metadata_batch);
+		}
 		ret = 1;
 	} else {
-		spin_unlock(&eb->refs_lock);
+		spin_unlock(&ebh->refs_lock);
 	}
 
 	btrfs_tree_unlock(eb);
@@ -3558,30 +3577,30 @@ static int lock_extent_buffer_for_io(struct extent_buffer *eb,
 	return ret;
 }
 
-static void end_extent_buffer_writeback(struct extent_buffer *eb)
+static void end_extent_buffer_writeback(struct extent_buffer_head *ebh)
 {
-	clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
+	clear_bit(EXTENT_BUFFER_WRITEBACK, &ebh->bflags);
 	smp_mb__after_clear_bit();
-	wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
+	wake_up_bit(&ebh->bflags, EXTENT_BUFFER_WRITEBACK);
 }
 
 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
 {
 	int uptodate = err == 0;
 	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
-	struct extent_buffer *eb;
+	struct extent_buffer_head *ebh;
 	int done;
 
 	do {
 		struct page *page = bvec->bv_page;
 
 		bvec--;
-		eb = (struct extent_buffer *)page->private;
-		BUG_ON(!eb);
-		done = atomic_dec_and_test(&eb->io_pages);
+		ebh = (struct extent_buffer_head *)page->private;
+		BUG_ON(!ebh);
+		done = atomic_dec_and_test(&ebh->io_pages);
 
-		if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
-			set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
+		if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &ebh->bflags)) {
+			set_bit(EXTENT_BUFFER_IOERR, &ebh->bflags);
 			ClearPageUptodate(page);
 			SetPageError(page);
 		}
@@ -3591,7 +3610,7 @@ static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
 		if (!done)
 			continue;
 
-		end_extent_buffer_writeback(eb);
+		end_extent_buffer_writeback(ebh);
 	} while (bvec >= bio->bi_io_vec);
 
 	bio_put(bio);
@@ -3605,15 +3624,15 @@ static int write_one_eb(struct extent_buffer *eb,
 {
 	struct block_device *bdev = fs_info->fs_devices->latest_bdev;
 	struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
-	u64 offset = eb->start;
+	u64 offset = eb->start & ~(PAGE_CACHE_SIZE - 1);
 	unsigned long i, num_pages;
 	unsigned long bio_flags = 0;
 	int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
 	int ret = 0;
 
-	clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
+	clear_bit(EXTENT_BUFFER_IOERR, &eb_head(eb)->bflags);
 	num_pages = num_extent_pages(eb->start, eb->len);
-	atomic_set(&eb->io_pages, num_pages);
+	atomic_set(&eb_head(eb)->io_pages, num_pages);
 	if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
 		bio_flags = EXTENT_BIO_TREE_LOG;
 
@@ -3628,10 +3647,11 @@ static int write_one_eb(struct extent_buffer *eb,
 					 0, epd->bio_flags, bio_flags);
 		epd->bio_flags = bio_flags;
 		if (ret) {
-			set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
+			set_bit(EXTENT_BUFFER_IOERR, &eb_head(eb)->bflags);
 			SetPageError(p);
-			if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
-				end_extent_buffer_writeback(eb);
+			if (atomic_sub_and_test(num_pages - i,
+							&eb_head(eb)->io_pages))
+				end_extent_buffer_writeback(eb_head(eb));
 			ret = -EIO;
 			break;
 		}
@@ -3655,7 +3675,8 @@ int btree_write_cache_pages(struct address_space *mapping,
 {
 	struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
 	struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
-	struct extent_buffer *eb, *prev_eb = NULL;
+	struct extent_buffer *eb;
+	struct extent_buffer_head *ebh, *prev_ebh = NULL;
 	struct extent_page_data epd = {
 		.bio = NULL,
 		.tree = tree,
@@ -3712,30 +3733,31 @@ retry:
 				continue;
 			}
 
-			eb = (struct extent_buffer *)page->private;
+			ebh = (struct extent_buffer_head *)page->private;
 
 			/*
 			 * Shouldn't happen and normally this would be a BUG_ON
 			 * but no sense in crashing the users box for something
 			 * we can survive anyway.
 			 */
-			if (WARN_ON(!eb)) {
+			if (WARN_ON(!ebh)) {
 				spin_unlock(&mapping->private_lock);
 				continue;
 			}
 
-			if (eb == prev_eb) {
+			if (ebh == prev_ebh) {
 				spin_unlock(&mapping->private_lock);
 				continue;
 			}
 
-			ret = atomic_inc_not_zero(&eb->refs);
+			ret = atomic_inc_not_zero(&ebh->refs);
 			spin_unlock(&mapping->private_lock);
 			if (!ret)
 				continue;
 
-			prev_eb = eb;
-			ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
+			eb = &ebh->extent_buf[0];
+			prev_ebh = ebh;
+			ret = lock_extent_buffer_for_io(ebh, fs_info, &epd);
 			if (!ret) {
 				free_extent_buffer(eb);
 				continue;
@@ -4410,17 +4432,23 @@ out:
 	return ret;
 }
 
-static void __free_extent_buffer(struct extent_buffer *eb)
+static void __free_extent_buffer(struct extent_buffer_head *ebh)
 {
-	btrfs_leak_debug_del(&eb->leak_list);
-	kmem_cache_free(extent_buffer_cache, eb);
+	btrfs_leak_debug_del(&ebh->leak_list);
+	kmem_cache_free(extent_buffer_cache, ebh);
 }
 
-static int extent_buffer_under_io(struct extent_buffer *eb)
+static int extent_buffer_under_io(struct extent_buffer_head *ebh)
 {
-	return (atomic_read(&eb->io_pages) ||
-		test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
-		test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
+	int i, dirty = 0;
+	struct extent_buffer *eb;
+
+	for (i = 0; i < MAX_EXTENT_BUFFERS_PER_PAGE && !dirty; i++) {
+		eb = &ebh->extent_buf[i];
+		dirty = test_bit(EXTENT_BUFFER_DIRTY, &eb->ebflags);
+	}
+	return (dirty || atomic_read(&ebh->io_pages) ||
+		test_bit(EXTENT_BUFFER_WRITEBACK, &ebh->bflags));
 }
 
 /*
@@ -4432,9 +4460,10 @@ static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
 	unsigned long index;
 	unsigned long num_pages;
 	struct page *page;
-	int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
+	struct extent_buffer_head *ebh = eb_head(eb);
+	int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &ebh->bflags);
 
-	BUG_ON(extent_buffer_under_io(eb));
+	BUG_ON(extent_buffer_under_io(ebh));
 
 	num_pages = num_extent_pages(eb->start, eb->len);
 	index = start_idx + num_pages;
@@ -4454,8 +4483,8 @@ static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
 			 * this eb.
 			 */
 			if (PagePrivate(page) &&
-			    page->private == (unsigned long)eb) {
-				BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
+			    page->private == (unsigned long)ebh) {
+				BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->ebflags));
 				BUG_ON(PageDirty(page));
 				BUG_ON(PageWriteback(page));
 				/*
@@ -4483,22 +4512,14 @@ static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
 {
 	btrfs_release_extent_buffer_page(eb, 0);
-	__free_extent_buffer(eb);
+	__free_extent_buffer(eb_head(eb));
 }
 
-static struct extent_buffer *
-__alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
-		      unsigned long len, gfp_t mask)
+static void __init_extent_buffer(struct extent_buffer *eb, u64 start,
+				unsigned long len)
 {
-	struct extent_buffer *eb = NULL;
-
-	eb = kmem_cache_zalloc(extent_buffer_cache, mask);
-	if (eb == NULL)
-		return NULL;
 	eb->start = start;
 	eb->len = len;
-	eb->fs_info = fs_info;
-	eb->bflags = 0;
 	rwlock_init(&eb->lock);
 	atomic_set(&eb->write_locks, 0);
 	atomic_set(&eb->read_locks, 0);
@@ -4509,12 +4530,26 @@ __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
 	eb->lock_nested = 0;
 	init_waitqueue_head(&eb->write_lock_wq);
 	init_waitqueue_head(&eb->read_lock_wq);
+}
 
-	btrfs_leak_debug_add(&eb->leak_list, &buffers);
+static struct extent_buffer *__alloc_extent_buffer(struct btrfs_fs_info *fs_info,
+						u64 start, unsigned long len,
+						gfp_t mask)
+{
+	struct extent_buffer_head *ebh = NULL;
+	struct extent_buffer *eb = NULL;
+	int i, index = -1;
+
+	ebh = kmem_cache_zalloc(extent_buffer_cache, mask);
+	if (ebh == NULL)
+		return NULL;
+	ebh->fs_info = fs_info;
+	ebh->bflags = 0;
+	btrfs_leak_debug_add(&ebh->leak_list, &buffers);
 
-	spin_lock_init(&eb->refs_lock);
-	atomic_set(&eb->refs, 1);
-	atomic_set(&eb->io_pages, 0);
+	spin_lock_init(&ebh->refs_lock);
+	atomic_set(&ebh->refs, 1);
+	atomic_set(&ebh->io_pages, 0);
 
 	/*
 	 * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
@@ -4523,6 +4558,34 @@ __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
 		> MAX_INLINE_EXTENT_BUFFER_SIZE);
 	BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
 
+	if (len < PAGE_CACHE_SIZE) {
+		u64 st = start & ~(PAGE_CACHE_SIZE - 1);
+		unsigned long totlen = 0;
+		/*
+		 * Make sure we have enough room to fit extent buffers
+		 * that belong a single page in a single extent_buffer_head.
+		 * If this BUG_ON is tripped, then it means either the
+		 * blocksize, i.e len, is too small or we need to increase
+		 * MAX_EXTENT_BUFFERS_PER_PAGE.
+		 */
+		BUG_ON(len * MAX_EXTENT_BUFFERS_PER_PAGE < PAGE_CACHE_SIZE);
+
+		for (i = 0; i < MAX_EXTENT_BUFFERS_PER_PAGE
+				&& totlen < PAGE_CACHE_SIZE ;
+				i++, st += len, totlen += len) {
+			__init_extent_buffer(&ebh->extent_buf[i], st, len);
+			if (st == start) {
+				index = i;
+				eb = &ebh->extent_buf[i];
+			}
+
+		}
+		BUG_ON(!eb);
+	} else {
+		eb = &ebh->extent_buf[0];
+		__init_extent_buffer(eb, start, len);
+	}
+
 	return eb;
 }
 
@@ -4543,15 +4606,15 @@ struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
 			btrfs_release_extent_buffer(new);
 			return NULL;
 		}
-		attach_extent_buffer_page(new, p);
+		attach_extent_buffer_page(eb_head(new), p);
 		WARN_ON(PageDirty(p));
 		SetPageUptodate(p);
-		new->pages[i] = p;
+		eb_head(new)->pages[i] = p;
 	}
 
 	copy_extent_buffer(new, src, 0, 0, src->len);
-	set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
-	set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
+	set_bit(EXTENT_BUFFER_UPTODATE, &eb_head(new)->bflags);
+	set_bit(EXTENT_BUFFER_DUMMY, &eb_head(new)->bflags);
 
 	return new;
 }
@@ -4567,19 +4630,19 @@ struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
 		return NULL;
 
 	for (i = 0; i < num_pages; i++) {
-		eb->pages[i] = alloc_page(GFP_NOFS);
-		if (!eb->pages[i])
+		eb_head(eb)->pages[i] = alloc_page(GFP_NOFS);
+		if (!eb_head(eb)->pages[i])
 			goto err;
 	}
 	set_extent_buffer_uptodate(eb);
 	btrfs_set_header_nritems(eb, 0);
-	set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
+	set_bit(EXTENT_BUFFER_DUMMY, &eb_head(eb)->bflags);
 
 	return eb;
 err:
 	for (; i > 0; i--)
-		__free_page(eb->pages[i - 1]);
-	__free_extent_buffer(eb);
+		__free_page(eb_head(eb)->pages[i - 1]);
+	__free_extent_buffer(eb_head(eb));
 	return NULL;
 }
 
@@ -4606,14 +4669,15 @@ static void check_buffer_tree_ref(struct extent_buffer *eb)
 	 * So bump the ref count first, then set the bit.  If someone
 	 * beat us to it, drop the ref we added.
 	 */
-	refs = atomic_read(&eb->refs);
-	if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
+	refs = atomic_read(&eb_head(eb)->refs);
+	if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF,
+						&eb_head(eb)->bflags))
 		return;
 
-	spin_lock(&eb->refs_lock);
-	if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
-		atomic_inc(&eb->refs);
-	spin_unlock(&eb->refs_lock);
+	spin_lock(&eb_head(eb)->refs_lock);
+	if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb_head(eb)->bflags))
+		atomic_inc(&eb_head(eb)->refs);
+	spin_unlock(&eb_head(eb)->refs_lock);
 }
 
 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
@@ -4633,14 +4697,22 @@ struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
 					 u64 start)
 {
 	struct extent_buffer *eb;
+	struct extent_buffer_head *ebh;
+	int i = 0;
 
 	rcu_read_lock();
-	eb = radix_tree_lookup(&fs_info->buffer_radix,
-			       start >> PAGE_CACHE_SHIFT);
-	if (eb && atomic_inc_not_zero(&eb->refs)) {
+	ebh = radix_tree_lookup(&fs_info->buffer_radix,
+				start >> PAGE_CACHE_SHIFT);
+	if (ebh && atomic_inc_not_zero(&ebh->refs)) {
 		rcu_read_unlock();
-		mark_extent_buffer_accessed(eb);
-		return eb;
+		do {
+			eb = &ebh->extent_buf[i++];
+			if (eb->start == start) {
+				mark_extent_buffer_accessed(eb);
+				return eb;
+			}
+		} while (i < MAX_EXTENT_BUFFERS_PER_PAGE);
+		BUG();
 	}
 	rcu_read_unlock();
 
@@ -4653,8 +4725,8 @@ struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
 	unsigned long num_pages = num_extent_pages(start, len);
 	unsigned long i;
 	unsigned long index = start >> PAGE_CACHE_SHIFT;
-	struct extent_buffer *eb;
-	struct extent_buffer *exists = NULL;
+	struct extent_buffer *eb, *old_eb = NULL;
+	struct extent_buffer_head *exists = NULL;
 	struct page *p;
 	struct address_space *mapping = fs_info->btree_inode->i_mapping;
 	int uptodate = 1;
@@ -4682,13 +4754,20 @@ struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
 			 * we can just return that one, else we know we can just
 			 * overwrite page->private.
 			 */
-			exists = (struct extent_buffer *)p->private;
+			exists = (struct extent_buffer_head *)p->private;
 			if (atomic_inc_not_zero(&exists->refs)) {
+				int j = 0;
 				spin_unlock(&mapping->private_lock);
 				unlock_page(p);
 				page_cache_release(p);
-				mark_extent_buffer_accessed(exists);
-				goto free_eb;
+				do {
+					old_eb = &exists->extent_buf[j++];
+					if (old_eb->start == start) {
+						mark_extent_buffer_accessed(old_eb);
+						goto free_eb;
+					}
+				} while (j < MAX_EXTENT_BUFFERS_PER_PAGE);
+				BUG();
 			}
 
 			/*
@@ -4699,11 +4778,11 @@ struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
 			WARN_ON(PageDirty(p));
 			page_cache_release(p);
 		}
-		attach_extent_buffer_page(eb, p);
+		attach_extent_buffer_page(eb_head(eb), p);
 		spin_unlock(&mapping->private_lock);
 		WARN_ON(PageDirty(p));
 		mark_page_accessed(p);
-		eb->pages[i] = p;
+		eb_head(eb)->pages[i] = p;
 		if (!PageUptodate(p))
 			uptodate = 0;
 
@@ -4713,7 +4792,7 @@ struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
 		 */
 	}
 	if (uptodate)
-		set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
+		set_bit(EXTENT_BUFFER_UPTODATE, &eb_head(eb)->bflags);
 again:
 	ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
 	if (ret)
@@ -4721,19 +4800,19 @@ again:
 
 	spin_lock(&fs_info->buffer_lock);
 	ret = radix_tree_insert(&fs_info->buffer_radix,
-				start >> PAGE_CACHE_SHIFT, eb);
+				start >> PAGE_CACHE_SHIFT, eb_head(eb));
 	spin_unlock(&fs_info->buffer_lock);
 	radix_tree_preload_end();
 	if (ret == -EEXIST) {
-		exists = find_extent_buffer(fs_info, start);
-		if (exists)
+		old_eb = find_extent_buffer(fs_info, start);
+		if (old_eb)
 			goto free_eb;
 		else
 			goto again;
 	}
 	/* add one reference for the tree */
 	check_buffer_tree_ref(eb);
-	set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
+	set_bit(EXTENT_BUFFER_IN_TREE, &eb_head(eb)->bflags);
 
 	/*
 	 * there is a race where release page may have
@@ -4744,58 +4823,58 @@ again:
 	 * after the extent buffer is in the radix tree so
 	 * it doesn't get lost
 	 */
-	SetPageChecked(eb->pages[0]);
+	SetPageChecked(eb_head(eb)->pages[0]);
 	for (i = 1; i < num_pages; i++) {
 		p = extent_buffer_page(eb, i);
 		ClearPageChecked(p);
 		unlock_page(p);
 	}
-	unlock_page(eb->pages[0]);
+	unlock_page(eb_head(eb)->pages[0]);
 	return eb;
 
 free_eb:
 	for (i = 0; i < num_pages; i++) {
-		if (eb->pages[i])
-			unlock_page(eb->pages[i]);
+		if (eb_head(eb)->pages[i])
+			unlock_page(eb_head(eb)->pages[i]);
 	}
 
-	WARN_ON(!atomic_dec_and_test(&eb->refs));
+	WARN_ON(!atomic_dec_and_test(&eb_head(eb)->refs));
 	btrfs_release_extent_buffer(eb);
-	return exists;
+	return old_eb;
 }
 
 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
 {
-	struct extent_buffer *eb =
-			container_of(head, struct extent_buffer, rcu_head);
+	struct extent_buffer_head *ebh =
+			container_of(head, struct extent_buffer_head, rcu_head);
 
-	__free_extent_buffer(eb);
+	__free_extent_buffer(ebh);
 }
 
 /* Expects to have eb->eb_lock already held */
-static int release_extent_buffer(struct extent_buffer *eb)
+static int release_extent_buffer(struct extent_buffer_head *ebh)
 {
-	WARN_ON(atomic_read(&eb->refs) == 0);
-	if (atomic_dec_and_test(&eb->refs)) {
-		if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
-			struct btrfs_fs_info *fs_info = eb->fs_info;
+	WARN_ON(atomic_read(&ebh->refs) == 0);
+	if (atomic_dec_and_test(&ebh->refs)) {
+		if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &ebh->bflags)) {
+			struct btrfs_fs_info *fs_info = ebh->fs_info;
 
-			spin_unlock(&eb->refs_lock);
+			spin_unlock(&ebh->refs_lock);
 
 			spin_lock(&fs_info->buffer_lock);
 			radix_tree_delete(&fs_info->buffer_radix,
-					  eb->start >> PAGE_CACHE_SHIFT);
+					ebh->extent_buf[0].start >> PAGE_CACHE_SHIFT);
 			spin_unlock(&fs_info->buffer_lock);
 		} else {
-			spin_unlock(&eb->refs_lock);
+			spin_unlock(&ebh->refs_lock);
 		}
 
 		/* Should be safe to release our pages at this point */
-		btrfs_release_extent_buffer_page(eb, 0);
-		call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
+		btrfs_release_extent_buffer_page(&ebh->extent_buf[0], 0);
+		call_rcu(&ebh->rcu_head, btrfs_release_extent_buffer_rcu);
 		return 1;
 	}
-	spin_unlock(&eb->refs_lock);
+	spin_unlock(&ebh->refs_lock);
 
 	return 0;
 }
@@ -4804,48 +4883,52 @@ void free_extent_buffer(struct extent_buffer *eb)
 {
 	int refs;
 	int old;
+	struct extent_buffer_head *ebh;
 	if (!eb)
 		return;
 
+	ebh = eb_head(eb);
 	while (1) {
-		refs = atomic_read(&eb->refs);
+		refs = atomic_read(&ebh->refs);
 		if (refs <= 3)
 			break;
-		old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
+		old = atomic_cmpxchg(&ebh->refs, refs, refs - 1);
 		if (old == refs)
 			return;
 	}
 
-	spin_lock(&eb->refs_lock);
-	if (atomic_read(&eb->refs) == 2 &&
-	    test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
-		atomic_dec(&eb->refs);
+	spin_lock(&ebh->refs_lock);
+	if (atomic_read(&ebh->refs) == 2 &&
+	    test_bit(EXTENT_BUFFER_DUMMY, &ebh->bflags))
+		atomic_dec(&ebh->refs);
 
-	if (atomic_read(&eb->refs) == 2 &&
-	    test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
-	    !extent_buffer_under_io(eb) &&
-	    test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
-		atomic_dec(&eb->refs);
+	if (atomic_read(&ebh->refs) == 2 &&
+	    test_bit(EXTENT_BUFFER_STALE, &ebh->bflags) &&
+	    !extent_buffer_under_io(ebh) &&
+	    test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &ebh->bflags))
+		atomic_dec(&ebh->refs);
 
 	/*
 	 * I know this is terrible, but it's temporary until we stop tracking
 	 * the uptodate bits and such for the extent buffers.
 	 */
-	release_extent_buffer(eb);
+	release_extent_buffer(ebh);
 }
 
 void free_extent_buffer_stale(struct extent_buffer *eb)
 {
+	struct extent_buffer_head *ebh;
 	if (!eb)
 		return;
 
-	spin_lock(&eb->refs_lock);
-	set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
+	ebh = eb_head(eb);
+	spin_lock(&ebh->refs_lock);
+	set_bit(EXTENT_BUFFER_STALE, &ebh->bflags);
 
-	if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
-	    test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
-		atomic_dec(&eb->refs);
-	release_extent_buffer(eb);
+	if (atomic_read(&ebh->refs) == 2 && !extent_buffer_under_io(ebh) &&
+	    test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &ebh->bflags))
+		atomic_dec(&ebh->refs);
+	release_extent_buffer(ebh);
 }
 
 void clear_extent_buffer_dirty(struct extent_buffer *eb)
@@ -4875,7 +4958,7 @@ void clear_extent_buffer_dirty(struct extent_buffer *eb)
 		ClearPageError(page);
 		unlock_page(page);
 	}
-	WARN_ON(atomic_read(&eb->refs) == 0);
+	WARN_ON(atomic_read(&eb_head(eb)->refs) == 0);
 }
 
 int set_extent_buffer_dirty(struct extent_buffer *eb)
@@ -4886,11 +4969,11 @@ int set_extent_buffer_dirty(struct extent_buffer *eb)
 
 	check_buffer_tree_ref(eb);
 
-	was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
+	was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->ebflags);
 
 	num_pages = num_extent_pages(eb->start, eb->len);
-	WARN_ON(atomic_read(&eb->refs) == 0);
-	WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
+	WARN_ON(atomic_read(&eb_head(eb)->refs) == 0);
+	WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb_head(eb)->bflags));
 
 	for (i = 0; i < num_pages; i++)
 		set_page_dirty(extent_buffer_page(eb, i));
@@ -4903,7 +4986,9 @@ int clear_extent_buffer_uptodate(struct extent_buffer *eb)
 	struct page *page;
 	unsigned long num_pages;
 
-	clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
+	if (!eb || !eb_head(eb))
+		return 0;
+	clear_bit(EXTENT_BUFFER_UPTODATE, &eb_head(eb)->bflags);
 	num_pages = num_extent_pages(eb->start, eb->len);
 	for (i = 0; i < num_pages; i++) {
 		page = extent_buffer_page(eb, i);
@@ -4919,7 +5004,7 @@ int set_extent_buffer_uptodate(struct extent_buffer *eb)
 	struct page *page;
 	unsigned long num_pages;
 
-	set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
+	set_bit(EXTENT_BUFFER_UPTODATE, &eb_head(eb)->bflags);
 	num_pages = num_extent_pages(eb->start, eb->len);
 	for (i = 0; i < num_pages; i++) {
 		page = extent_buffer_page(eb, i);
@@ -4930,7 +5015,7 @@ int set_extent_buffer_uptodate(struct extent_buffer *eb)
 
 int extent_buffer_uptodate(struct extent_buffer *eb)
 {
-	return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
+	return test_bit(EXTENT_BUFFER_UPTODATE, &eb_head(eb)->bflags);
 }
 
 int read_extent_buffer_pages(struct extent_io_tree *tree,
@@ -4948,8 +5033,9 @@ int read_extent_buffer_pages(struct extent_io_tree *tree,
 	unsigned long num_reads = 0;
 	struct bio *bio = NULL;
 	unsigned long bio_flags = 0;
+	struct extent_buffer_head *ebh = eb_head(eb);
 
-	if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
+	if (test_bit(EXTENT_BUFFER_UPTODATE, &ebh->bflags))
 		return 0;
 
 	if (start) {
@@ -4960,6 +5046,7 @@ int read_extent_buffer_pages(struct extent_io_tree *tree,
 		start_i = 0;
 	}
 
+recheck:
 	num_pages = num_extent_pages(eb->start, eb->len);
 	for (i = start_i; i < num_pages; i++) {
 		page = extent_buffer_page(eb, i);
@@ -4977,13 +5064,26 @@ int read_extent_buffer_pages(struct extent_io_tree *tree,
 	}
 	if (all_uptodate) {
 		if (start_i == 0)
-			set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
+			set_bit(EXTENT_BUFFER_UPTODATE, &ebh->bflags);
 		goto unlock_exit;
 	}
 
-	clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
-	eb->read_mirror = 0;
-	atomic_set(&eb->io_pages, num_reads);
+	if (eb_head(eb)->io_eb) {
+		all_uptodate = 1;
+		i = start_i;
+		while (locked_pages > 0) {
+			page = extent_buffer_page(eb, i);
+			i++;
+			unlock_page(page);
+			locked_pages--;
+		}
+		goto recheck;
+	}
+	BUG_ON(eb_head(eb)->io_eb);
+	eb_head(eb)->io_eb = eb;
+	clear_bit(EXTENT_BUFFER_IOERR, &ebh->bflags);
+	ebh->read_mirror = 0;
+	atomic_set(&ebh->io_pages, num_reads);
 	for (i = start_i; i < num_pages; i++) {
 		page = extent_buffer_page(eb, i);
 		if (!PageUptodate(page)) {
@@ -5350,7 +5450,7 @@ void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
 
 int try_release_extent_buffer(struct page *page)
 {
-	struct extent_buffer *eb;
+	struct extent_buffer_head *ebh;
 
 	/*
 	 * We need to make sure noboody is attaching this page to an eb right
@@ -5362,17 +5462,17 @@ int try_release_extent_buffer(struct page *page)
 		return 1;
 	}
 
-	eb = (struct extent_buffer *)page->private;
-	BUG_ON(!eb);
+	ebh = (struct extent_buffer_head *)page->private;
+	BUG_ON(!ebh);
 
 	/*
 	 * This is a little awful but should be ok, we need to make sure that
 	 * the eb doesn't disappear out from under us while we're looking at
 	 * this page.
 	 */
-	spin_lock(&eb->refs_lock);
-	if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
-		spin_unlock(&eb->refs_lock);
+	spin_lock(&ebh->refs_lock);
+	if (atomic_read(&ebh->refs) != 1 || extent_buffer_under_io(ebh)) {
+		spin_unlock(&ebh->refs_lock);
 		spin_unlock(&page->mapping->private_lock);
 		return 0;
 	}
@@ -5382,10 +5482,11 @@ int try_release_extent_buffer(struct page *page)
 	 * If tree ref isn't set then we know the ref on this eb is a real ref,
 	 * so just return, this page will likely be freed soon anyway.
 	 */
-	if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
-		spin_unlock(&eb->refs_lock);
+	if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &ebh->bflags)) {
+		spin_unlock(&ebh->refs_lock);
 		return 0;
 	}
 
-	return release_extent_buffer(eb);
+	return release_extent_buffer(ebh);
 }
+
diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index 58b27e5..71100ad 100644
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -123,19 +123,12 @@ struct extent_state {
 
 #define INLINE_EXTENT_BUFFER_PAGES 16
 #define MAX_INLINE_EXTENT_BUFFER_SIZE (INLINE_EXTENT_BUFFER_PAGES * PAGE_CACHE_SIZE)
+#define MAX_EXTENT_BUFFERS_PER_PAGE 16
+
 struct extent_buffer {
 	u64 start;
 	unsigned long len;
-	unsigned long map_start;
-	unsigned long map_len;
-	unsigned long bflags;
-	struct btrfs_fs_info *fs_info;
-	spinlock_t refs_lock;
-	atomic_t refs;
-	atomic_t io_pages;
-	int read_mirror;
-	struct rcu_head rcu_head;
-	pid_t lock_owner;
+	unsigned long ebflags;
 
 	/* count of read lock holders on the extent buffer */
 	atomic_t write_locks;
@@ -146,6 +139,8 @@ struct extent_buffer {
 	atomic_t spinning_writers;
 	int lock_nested;
 
+	pid_t lock_owner;
+
 	/* protects write locks */
 	rwlock_t lock;
 
@@ -159,7 +154,21 @@ struct extent_buffer {
 	 */
 	wait_queue_head_t read_lock_wq;
 	wait_queue_head_t lock_wq;
+};
+
+struct extent_buffer_head {
+	unsigned long bflags;
+	struct btrfs_fs_info *fs_info;
+	spinlock_t refs_lock;
+	atomic_t refs;
+	atomic_t io_pages;
+	int read_mirror;
+	struct rcu_head rcu_head;
+
 	struct page *pages[INLINE_EXTENT_BUFFER_PAGES];
+
+	struct extent_buffer extent_buf[MAX_EXTENT_BUFFERS_PER_PAGE];
+	struct extent_buffer *io_eb; /* eb that submitted the current I/O */
 #ifdef CONFIG_BTRFS_DEBUG
 	struct list_head leak_list;
 #endif
@@ -176,6 +185,24 @@ static inline int extent_compress_type(unsigned long bio_flags)
 	return bio_flags >> EXTENT_BIO_FLAG_SHIFT;
 }
 
+/*
+ * return the extent_buffer_head that contains the extent buffer provided.
+ */
+static inline struct extent_buffer_head *eb_head(struct extent_buffer *eb)
+{
+	int start, index;
+	struct extent_buffer_head *ebh;
+	struct extent_buffer *eb_base;
+
+	BUG_ON(!eb);
+	start = eb->start & (PAGE_CACHE_SIZE - 1);
+	index = start >> (ffs(eb->len) - 1);
+	eb_base = eb - index;
+	ebh = (struct extent_buffer_head *)
+		((char *) eb_base - offsetof(struct extent_buffer_head, extent_buf));
+	return ebh;
+
+}
 struct extent_map_tree;
 
 typedef struct extent_map *(get_extent_t)(struct inode *inode,
@@ -287,15 +314,15 @@ static inline unsigned long num_extent_pages(u64 start, u64 len)
 		(start >> PAGE_CACHE_SHIFT);
 }
 
-static inline struct page *extent_buffer_page(struct extent_buffer *eb,
-					      unsigned long i)
+static inline struct page *extent_buffer_page(
+			struct extent_buffer *eb, unsigned long i)
 {
-	return eb->pages[i];
+	return eb_head(eb)->pages[i];
 }
 
 static inline void extent_buffer_get(struct extent_buffer *eb)
 {
-	atomic_inc(&eb->refs);
+	atomic_inc(&eb_head(eb)->refs);
 }
 
 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 07629e9..1ec359b 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -5936,7 +5936,7 @@ int btrfs_read_sys_array(struct btrfs_root *root)
 	 * to silence the warning eg. on PowerPC 64.
 	 */
 	if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
-		SetPageUptodate(sb->pages[0]);
+		SetPageUptodate(eb_head(sb)->pages[0]);
 
 	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
 	array_size = btrfs_super_sys_array_size(super_copy);
diff --git a/include/trace/events/btrfs.h b/include/trace/events/btrfs.h
index 3176cdc..5b79ac2 100644
--- a/include/trace/events/btrfs.h
+++ b/include/trace/events/btrfs.h
@@ -695,7 +695,7 @@ TRACE_EVENT(btrfs_cow_block,
 	TP_fast_assign(
 		__entry->root_objectid	= root->root_key.objectid;
 		__entry->buf_start	= buf->start;
-		__entry->refs		= atomic_read(&buf->refs);
+		__entry->refs		= atomic_read(&eb_head(buf)->refs);
 		__entry->cow_start	= cow->start;
 		__entry->buf_level	= btrfs_header_level(buf);
 		__entry->cow_level	= btrfs_header_level(cow);
-- 
1.8.3.1

[PATCH 5/6] Btrfs: subpagesize-blocksize: Hardcode MAX_EXTENT_BUFFERS_PER_PAGE to 2.

[Chandan Rajendra]

With 2k blocksize the maximum number extent buffers per page will be 2. We
should be computing this value at runtime. For now, Let's work with the
hardcoded value.

Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
---
 fs/btrfs/extent_io.h | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index 71100ad..6cc5ed2 100644
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -123,7 +123,12 @@ struct extent_state {
 
 #define INLINE_EXTENT_BUFFER_PAGES 16
 #define MAX_INLINE_EXTENT_BUFFER_SIZE (INLINE_EXTENT_BUFFER_PAGES * PAGE_CACHE_SIZE)
-#define MAX_EXTENT_BUFFERS_PER_PAGE 16
+
+#if 0
+#define MAX_EXTENT_BUFFERS_PER_PAGE 1
+#else
+#define MAX_EXTENT_BUFFERS_PER_PAGE 2
+#endif
 
 struct extent_buffer {
 	u64 start;
-- 
1.8.3.1

[PATCH 6/6] Btrfs: subpagesize-blocksize: Allow mounting filesystems where sectorsize != PAGE_SIZE

[Chandan Rajendra]

From: Chandra Seetharaman <sekharan@us.ibm.com>

This patch allows mounting filesystems with blocksize smaller than the
PAGE_SIZE.

Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
---
 fs/btrfs/disk-io.c | 6 ------
 1 file changed, 6 deletions(-)

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 796d889..aa9e3bb 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -2617,12 +2617,6 @@ int open_ctree(struct super_block *sb,
 		goto fail_sb_buffer;
 	}
 
-	if (sectorsize != PAGE_SIZE) {
-		printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
-		       "found on %s\n", (unsigned long)sectorsize, sb->s_id);
-		goto fail_sb_buffer;
-	}
-
 	mutex_lock(&fs_info->chunk_mutex);
 	ret = btrfs_read_sys_array(tree_root);
 	mutex_unlock(&fs_info->chunk_mutex);
-- 
1.8.3.1

Re: [PATCH 0/6 EARLY RFC] Btrfs: Get rid of whole page I/O.

[David Sterba]

On Wed, Mar 12, 2014 at 07:50:27PM +0530, Chandan Rajendra wrote:
> The patchset has been tested with only a handful of trivial I/O tests i.e. I/O
> on different kinds of files (e.g. files with holes, etc). The tests were run
> on 2k and 4k blocksized instances of Btrfs code (on x86_64) that has initial
> support for mounting a 2k blocksized Btrfs filesystem.

> 3. Compression does not work with 2k blocksize filesystem instance.

I looked at previous postings of this patchset, but haven't found what
are the expected supported block sizes.

I assume powers of two starting with 512b, until 64k.

I'm asking because the compression container would need to take this
into account. (I originally had only >= 4k sizes in mind.)


thanks,
david

Re: [PATCH 0/6 EARLY RFC] Btrfs: Get rid of whole page I/O.

[chandan]

Hello David,

> I looked at previous postings of this patchset, but haven't found what
> are the expected supported block sizes.
> 
> I assume powers of two starting with 512b, until 64k.

The earlier patchset posted by Chandra Seethraman was to get 4k
blocksize to work with ppc64's 64k PAGE_SIZE. I chose to do 2k
blocksize on x86_64's 4k PAGE_SIZE since that would allow others in
the community to work/experiment with subpagesize-blocksize feature.

The root node of "tree root" tree has 1957 bytes being written by
make_btrfs() (in btrfs-progs).  Hence I chose to do 2k blocksize for
the initial subpagesize-blocksize work. So with this patchset the
supported blocksizes would be in the range 2k-64k.

Thanks,
chandan

Re: [PATCH 0/6 EARLY RFC] Btrfs: Get rid of whole page I/O.

[David Sterba]

On Tue, Mar 18, 2014 at 01:48:00PM +0630, chandan wrote:
> The earlier patchset posted by Chandra Seethraman was to get 4k
> blocksize to work with ppc64's 64k PAGE_SIZE.

Are we talking about metadata block sizes or data block sizes?

> The root node of "tree root" tree has 1957 bytes being written by
> make_btrfs() (in btrfs-progs).  Hence I chose to do 2k blocksize for
> the initial subpagesize-blocksize work. So with this patchset the
> supported blocksizes would be in the range 2k-64k.

So it's metadata blocks, and in this case 2k looks like the only
allowed size that's smaller than 4k, and thus can demonstrage sub-page
size allocations. I'm not sure if this is limiting for potential future
extensions of metadata structures that could be larger.

2k is ok for testing purposes, but I think a 4k-page machine will hardly
use a smaller page size. The more that 16k metadata blocks are now
default.

Re: [PATCH 0/6 EARLY RFC] Btrfs: Get rid of whole page I/O.

[Aneesh Kumar K.V]

David Sterba <dsterba@suse.cz> writes:

> On Tue, Mar 18, 2014 at 01:48:00PM +0630, chandan wrote:
>> The earlier patchset posted by Chandra Seethraman was to get 4k
>> blocksize to work with ppc64's 64k PAGE_SIZE.
>
> Are we talking about metadata block sizes or data block sizes?
>
>> The root node of "tree root" tree has 1957 bytes being written by
>> make_btrfs() (in btrfs-progs).  Hence I chose to do 2k blocksize for
>> the initial subpagesize-blocksize work. So with this patchset the
>> supported blocksizes would be in the range 2k-64k.
>
> So it's metadata blocks, and in this case 2k looks like the only
> allowed size that's smaller than 4k, and thus can demonstrage sub-page
> size allocations. I'm not sure if this is limiting for potential future
> extensions of metadata structures that could be larger.
>
> 2k is ok for testing purposes, but I think a 4k-page machine will hardly
> use a smaller page size. The more that 16k metadata blocks are now
> default.

The goal is to remove the assumption that supported blocks size is >= page
size. The primary reason to do that is to support migration of disk
devices across different architectures. If we have a btrfs disk created
on x86 box with data blocksize 4K and meta data block size 16K we should
make sure that, the disk can be read/written from ppc64 box (which have a page
size of 64K). To enable easy testing and community development we are
now focusing on achieving 2K data blocksize and 2K meata data block size
on x86. As you said this will never be used in production.

To achieve that we did the below
*) Add offset and len to btrfs_io_bio. These are file offsets and
len. This is later used to unlock extent io tree.

*) Now we also need to make sure that submit_extent_page only submit
 contiguous range in the file offset range. ie if we have holes in
 between we split them into two submit_extent_page.  This ensures that
 btrfs_io_bio offset and len represent a contiguous range.

Please let us know whether the above approach is acceptable.

 -aneesh
 

Re: [PATCH 0/6 EARLY RFC] Btrfs: Get rid of whole page I/O.

[David Sterba]

On Thu, Mar 20, 2014 at 08:50:27AM +0530, Aneesh Kumar K.V wrote:
> > On Tue, Mar 18, 2014 at 01:48:00PM +0630, chandan wrote:
> >> The earlier patchset posted by Chandra Seethraman was to get 4k
> >> blocksize to work with ppc64's 64k PAGE_SIZE.
> >
> > Are we talking about metadata block sizes or data block sizes?
> >
> >> The root node of "tree root" tree has 1957 bytes being written by
> >> make_btrfs() (in btrfs-progs).  Hence I chose to do 2k blocksize for
> >> the initial subpagesize-blocksize work. So with this patchset the
> >> supported blocksizes would be in the range 2k-64k.
> >
> > So it's metadata blocks, and in this case 2k looks like the only
> > allowed size that's smaller than 4k, and thus can demonstrage sub-page
> > size allocations. I'm not sure if this is limiting for potential future
> > extensions of metadata structures that could be larger.
> >
> > 2k is ok for testing purposes, but I think a 4k-page machine will hardly
> > use a smaller page size. The more that 16k metadata blocks are now
> > default.
> 
> The goal is to remove the assumption that supported blocks size is >= page
> size. The primary reason to do that is to support migration of disk
> devices across different architectures. If we have a btrfs disk created
> on x86 box with data blocksize 4K and meta data block size 16K we should
> make sure that, the disk can be read/written from ppc64 box (which have a page
> size of 64K). To enable easy testing and community development we are
> now focusing on achieving 2K data blocksize and 2K meata data block size
> on x86. As you said this will never be used in production.
> 
> To achieve that we did the below
> *) Add offset and len to btrfs_io_bio. These are file offsets and
> len. This is later used to unlock extent io tree.
> 
> *) Now we also need to make sure that submit_extent_page only submit
>  contiguous range in the file offset range. ie if we have holes in
>  between we split them into two submit_extent_page.  This ensures that
>  btrfs_io_bio offset and len represent a contiguous range.
> 
> Please let us know whether the above approach is acceptable.

I don't see any apparent problem with this approach.