cleanup btrfs bio handling, part 2 v3

cleanup btrfs bio handling, part 2 v3

[Christoph Hellwig]

Hi all,

this series removes the need to allocate a separate object for I/O
completions for all read and some write I/Os, and reduced the memory
usage of the low-level bios cloned by btrfs_map_bio by using plain bios
instead of the much larger btrfs_bio.

Changes since v2:
 - rebased to the latests misc-next tree
 - fixed an incorrectly assigned bi_end_io handler in the raid56 code

Changes since v1:
 - rebased to the latests misc-next tree
 - don't call destroy_workqueue on NULL workqueues
 - drop a marginal and slightly controversial cleanup to btrfs_map_bio

Diffstat:
 compression.c |   41 +++++-------
 compression.h |    7 +-
 ctree.h       |   14 ++--
 disk-io.c     |  148 +++++-----------------------------------------
 disk-io.h     |   11 ---
 extent_io.c   |   35 ++++-------
 extent_io.h   |    1 
 inode.c       |  162 +++++++++++++++++++--------------------------------
 raid56.c      |  109 +++++++++++++---------------------
 super.c       |   13 ----
 volumes.c     |  184 +++++++++++++++++++++++++++-------------------------------
 volumes.h     |    8 ++
 12 files changed, 262 insertions(+), 471 deletions(-)

[PATCH 01/10] btrfs: don't double-defer bio completions for compressed reads

[Christoph Hellwig]

The bio completion handler of the bio used for the compressed data is
already run in a workqueue using btrfs_bio_wq_end_io, so don't schedule
the completion of the original bio to the same workqueue again but just
execute it directly.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/btrfs/inode.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index b42d6e7e4049f..39da19645e890 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -2598,10 +2598,6 @@ void btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
 	}
 
 	if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
-		ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
-		if (ret)
-			goto out;
-
 		if (compress_type != BTRFS_COMPRESS_NONE) {
 			/*
 			 * btrfs_submit_compressed_read will handle completing
@@ -2611,6 +2607,10 @@ void btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
 			btrfs_submit_compressed_read(inode, bio, mirror_num);
 			return;
 		} else {
+			ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
+			if (ret)
+				goto out;
+
 			/*
 			 * Lookup bio sums does extra checks around whether we
 			 * need to csum or not, which is why we ignore skip_sum
-- 
2.30.2

[PATCH 02/10] btrfs: move more work into btrfs_end_bioc

[Christoph Hellwig]

Assign ->mirror_num and ->bi_status in btrfs_end_bioc instead of
duplicating the logic in the callers.  Also remove the bio argument as
it always must be bioc->orig_bio and the now pointless bioc_error that
did nothing but assign bi_sector to the same value just sampled in the
caller.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
---
 fs/btrfs/volumes.c | 72 ++++++++++++++--------------------------------
 1 file changed, 22 insertions(+), 50 deletions(-)

diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 748614b00ffa2..aeacb87457687 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -6621,19 +6621,29 @@ int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
 	return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 0, 1);
 }
 
-static inline void btrfs_end_bioc(struct btrfs_io_context *bioc, struct bio *bio)
+static inline void btrfs_end_bioc(struct btrfs_io_context *bioc)
 {
-	bio->bi_private = bioc->private;
-	bio->bi_end_io = bioc->end_io;
-	bio_endio(bio);
+	struct bio *orig_bio = bioc->orig_bio;
 
+	btrfs_bio(orig_bio)->mirror_num = bioc->mirror_num;
+	orig_bio->bi_private = bioc->private;
+	orig_bio->bi_end_io = bioc->end_io;
+
+	/*
+	 * Only send an error to the higher layers if it is beyond the tolerance
+	 * threshold.
+	 */
+	if (atomic_read(&bioc->error) > bioc->max_errors)
+		orig_bio->bi_status = BLK_STS_IOERR;
+	else
+		orig_bio->bi_status = BLK_STS_OK;
+	bio_endio(orig_bio);
 	btrfs_put_bioc(bioc);
 }
 
 static void btrfs_end_bio(struct bio *bio)
 {
 	struct btrfs_io_context *bioc = bio->bi_private;
-	int is_orig_bio = 0;
 
 	if (bio->bi_status) {
 		atomic_inc(&bioc->error);
@@ -6654,35 +6664,12 @@ static void btrfs_end_bio(struct bio *bio)
 		}
 	}
 
-	if (bio == bioc->orig_bio)
-		is_orig_bio = 1;
+	if (bio != bioc->orig_bio)
+		bio_put(bio);
 
 	btrfs_bio_counter_dec(bioc->fs_info);
-
-	if (atomic_dec_and_test(&bioc->stripes_pending)) {
-		if (!is_orig_bio) {
-			bio_put(bio);
-			bio = bioc->orig_bio;
-		}
-
-		btrfs_bio(bio)->mirror_num = bioc->mirror_num;
-		/* only send an error to the higher layers if it is
-		 * beyond the tolerance of the btrfs bio
-		 */
-		if (atomic_read(&bioc->error) > bioc->max_errors) {
-			bio->bi_status = BLK_STS_IOERR;
-		} else {
-			/*
-			 * this bio is actually up to date, we didn't
-			 * go over the max number of errors
-			 */
-			bio->bi_status = BLK_STS_OK;
-		}
-
-		btrfs_end_bioc(bioc, bio);
-	} else if (!is_orig_bio) {
-		bio_put(bio);
-	}
+	if (atomic_dec_and_test(&bioc->stripes_pending))
+		btrfs_end_bioc(bioc);
 }
 
 static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio,
@@ -6720,23 +6707,6 @@ static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio,
 	submit_bio(bio);
 }
 
-static void bioc_error(struct btrfs_io_context *bioc, struct bio *bio, u64 logical)
-{
-	atomic_inc(&bioc->error);
-	if (atomic_dec_and_test(&bioc->stripes_pending)) {
-		/* Should be the original bio. */
-		WARN_ON(bio != bioc->orig_bio);
-
-		btrfs_bio(bio)->mirror_num = bioc->mirror_num;
-		bio->bi_iter.bi_sector = logical >> 9;
-		if (atomic_read(&bioc->error) > bioc->max_errors)
-			bio->bi_status = BLK_STS_IOERR;
-		else
-			bio->bi_status = BLK_STS_OK;
-		btrfs_end_bioc(bioc, bio);
-	}
-}
-
 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 			   int mirror_num)
 {
@@ -6795,7 +6765,9 @@ blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 						   &dev->dev_state) ||
 		    (btrfs_op(first_bio) == BTRFS_MAP_WRITE &&
 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
-			bioc_error(bioc, first_bio, logical);
+			atomic_inc(&bioc->error);
+			if (atomic_dec_and_test(&bioc->stripes_pending))
+				btrfs_end_bioc(bioc);
 			continue;
 		}
 
-- 
2.30.2

[PATCH 03/10] btrfs: cleanup btrfs_submit_dio_bio

[Christoph Hellwig]

Remove the pointless goto just to return err and clean up the code flow
to be a little more straight forward.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
---
 fs/btrfs/inode.c | 22 ++++++++--------------
 1 file changed, 8 insertions(+), 14 deletions(-)

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 39da19645e890..b98f5291e941c 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -7948,31 +7948,27 @@ static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
 	bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
 	blk_status_t ret;
 
-	/* Check btrfs_submit_bio_hook() for rules about async submit. */
-	if (async_submit)
-		async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
-
 	if (!write) {
 		ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
 		if (ret)
-			goto err;
+			return ret;
 	}
 
 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
 		goto map;
 
-	if (write && async_submit) {
-		ret = btrfs_wq_submit_bio(inode, bio, 0, file_offset,
-					  btrfs_submit_bio_start_direct_io);
-		goto err;
-	} else if (write) {
+	if (write) {
+		/* check btrfs_submit_data_bio() for async submit rules */
+		if (async_submit && !atomic_read(&BTRFS_I(inode)->sync_writers))
+			return btrfs_wq_submit_bio(inode, bio, 0, file_offset,
+					btrfs_submit_bio_start_direct_io);
 		/*
 		 * If we aren't doing async submit, calculate the csum of the
 		 * bio now.
 		 */
 		ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, false);
 		if (ret)
-			goto err;
+			return ret;
 	} else {
 		u64 csum_offset;
 
@@ -7982,9 +7978,7 @@ static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
 		btrfs_bio(bio)->csum = dip->csums + csum_offset;
 	}
 map:
-	ret = btrfs_map_bio(fs_info, bio, 0);
-err:
-	return ret;
+	return btrfs_map_bio(fs_info, bio, 0);
 }
 
 /*
-- 
2.30.2

[PATCH 04/10] btrfs: split btrfs_submit_data_bio

[Christoph Hellwig]

Split btrfs_submit_data_bio into one helper for reads and one for writes.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/btrfs/ctree.h     |   6 +-
 fs/btrfs/extent_io.c |  14 +++--
 fs/btrfs/inode.c     | 134 ++++++++++++++++++++-----------------------
 3 files changed, 76 insertions(+), 78 deletions(-)

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 6e939bf01dcc3..6f539ddf7467a 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -3251,8 +3251,10 @@ void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_siz
 u64 btrfs_file_extent_end(const struct btrfs_path *path);
 
 /* inode.c */
-void btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
-			   int mirror_num, enum btrfs_compression_type compress_type);
+void btrfs_submit_data_write_bio(struct inode *inode, struct bio *bio,
+		int mirror_num);
+void btrfs_submit_data_read_bio(struct inode *inode, struct bio *bio,
+		int mirror_num, enum btrfs_compression_type compress_type);
 unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio,
 				    u32 bio_offset, struct page *page,
 				    u64 start, u64 end);
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 1b1baeb0d76bc..1394a6f80d285 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -182,17 +182,21 @@ static void submit_one_bio(struct bio *bio, int mirror_num,
 			   enum btrfs_compression_type compress_type)
 {
 	struct extent_io_tree *tree = bio->bi_private;
+	struct inode *inode = tree->private_data;
 
 	bio->bi_private = NULL;
 
 	/* Caller should ensure the bio has at least some range added */
 	ASSERT(bio->bi_iter.bi_size);
 
-	if (is_data_inode(tree->private_data))
-		btrfs_submit_data_bio(tree->private_data, bio, mirror_num,
-					    compress_type);
+	if (!is_data_inode(tree->private_data))
+		btrfs_submit_metadata_bio(inode, bio, mirror_num);
+	else if (btrfs_op(bio) == BTRFS_MAP_WRITE)
+		btrfs_submit_data_write_bio(inode, bio, mirror_num);
 	else
-		btrfs_submit_metadata_bio(tree->private_data, bio, mirror_num);
+		btrfs_submit_data_read_bio(inode, bio, mirror_num,
+					   compress_type);
+
 	/*
 	 * Above submission hooks will handle the error by ending the bio,
 	 * which will do the cleanup properly.  So here we should not return
@@ -2774,7 +2778,7 @@ static blk_status_t submit_data_read_repair(struct inode *inode,
 		ret = btrfs_repair_one_sector(inode, failed_bio,
 				bio_offset + offset,
 				page, pgoff + offset, start + offset,
-				failed_mirror, btrfs_submit_data_bio);
+				failed_mirror, btrfs_submit_data_read_bio);
 		if (!ret) {
 			/*
 			 * We have submitted the read repair, the page release
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index b98f5291e941c..5499b39cab61b 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -2554,90 +2554,82 @@ static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
 	return errno_to_blk_status(ret);
 }
 
-/*
- * extent_io.c submission hook. This does the right thing for csum calculation
- * on write, or reading the csums from the tree before a read.
- *
- * Rules about async/sync submit,
- * a) read:				sync submit
- *
- * b) write without checksum:		sync submit
- *
- * c) write with checksum:
- *    c-1) if bio is issued by fsync:	sync submit
- *         (sync_writers != 0)
- *
- *    c-2) if root is reloc root:	sync submit
- *         (only in case of buffered IO)
- *
- *    c-3) otherwise:			async submit
- */
-void btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
-			   int mirror_num, enum btrfs_compression_type compress_type)
+void btrfs_submit_data_write_bio(struct inode *inode, struct bio *bio,
+		int mirror_num)
 {
 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
-	struct btrfs_root *root = BTRFS_I(inode)->root;
-	enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
-	blk_status_t ret = 0;
-	int skip_sum;
-	int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
-
-	skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
-		test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
-
-	if (btrfs_is_free_space_inode(BTRFS_I(inode)))
-		metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
+	struct btrfs_inode *bi = BTRFS_I(inode);
+	blk_status_t ret;
 
 	if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
-		struct page *page = bio_first_bvec_all(bio)->bv_page;
-		loff_t file_offset = page_offset(page);
-
-		ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
+		ret = extract_ordered_extent(bi, bio,
+				page_offset(bio_first_bvec_all(bio)->bv_page));
 		if (ret)
 			goto out;
 	}
 
-	if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
-		if (compress_type != BTRFS_COMPRESS_NONE) {
-			/*
-			 * btrfs_submit_compressed_read will handle completing
-			 * the bio if there were any errors, so just return
-			 * here.
-			 */
-			btrfs_submit_compressed_read(inode, bio, mirror_num);
-			return;
-		} else {
-			ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
-			if (ret)
-				goto out;
-
-			/*
-			 * Lookup bio sums does extra checks around whether we
-			 * need to csum or not, which is why we ignore skip_sum
-			 * here.
-			 */
-			ret = btrfs_lookup_bio_sums(inode, bio, NULL);
+	/*
+	 * Rules for async/sync submit:
+	 *   a) write without checksum:			sync submit
+	 *   b) write with checksum:
+	 *      b-1) if bio is issued by fsync:		sync submit
+	 *           (sync_writers != 0)
+	 *      b-2) if root is reloc root:		sync submit
+	 *           (only in case of buffered IO)
+	 *      b-3) otherwise:				async submit
+	 */
+	if (!(bi->flags & BTRFS_INODE_NODATASUM) &&
+	    !test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)) {
+		if (atomic_read(&bi->sync_writers)) {
+			ret = btrfs_csum_one_bio(bi, bio, (u64)-1, false);
 			if (ret)
 				goto out;
-		}
-		goto mapit;
-	} else if (async && !skip_sum) {
-		/* csum items have already been cloned */
-		if (btrfs_is_data_reloc_root(root))
-			goto mapit;
-		/* we're doing a write, do the async checksumming */
-		ret = btrfs_wq_submit_bio(inode, bio, mirror_num,
-					  0, btrfs_submit_bio_start);
-		goto out;
-	} else if (!skip_sum) {
-		ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
-		if (ret)
+		} else if (btrfs_is_data_reloc_root(bi->root)) {
+			; /* csum items have already been cloned */
+		} else {
+			ret = btrfs_wq_submit_bio(inode, bio,
+					mirror_num, 0,
+					btrfs_submit_bio_start);
 			goto out;
+		}
 	}
-
-mapit:
 	ret = btrfs_map_bio(fs_info, bio, mirror_num);
+out:
+	if (ret) {
+		bio->bi_status = ret;
+		bio_endio(bio);
+	}
+}
 
+void btrfs_submit_data_read_bio(struct inode *inode, struct bio *bio,
+		int mirror_num, enum btrfs_compression_type compress_type)
+{
+	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
+	blk_status_t ret;
+
+	if (compress_type != BTRFS_COMPRESS_NONE) {
+		/*
+		 * btrfs_submit_compressed_read will handle completing the bio
+		 * if there were any errors, so just return here.
+		 */
+		btrfs_submit_compressed_read(inode, bio, mirror_num);
+		return;
+	}
+
+	ret = btrfs_bio_wq_end_io(fs_info, bio,
+			btrfs_is_free_space_inode(BTRFS_I(inode)) ?
+			BTRFS_WQ_ENDIO_FREE_SPACE : BTRFS_WQ_ENDIO_DATA);
+	if (ret)
+		goto out;
+
+	/*
+	 * Lookup bio sums does extra checks around whether we need to csum or
+	 * not, which is why we ignore skip_sum here.
+	 */
+	ret = btrfs_lookup_bio_sums(inode, bio, NULL);
+	if (ret)
+		goto out;
+	ret = btrfs_map_bio(fs_info, bio, mirror_num);
 out:
 	if (ret) {
 		bio->bi_status = ret;
@@ -7958,7 +7950,7 @@ static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
 		goto map;
 
 	if (write) {
-		/* check btrfs_submit_data_bio() for async submit rules */
+		/* check btrfs_submit_data_write_bio() for async submit rules */
 		if (async_submit && !atomic_read(&BTRFS_I(inode)->sync_writers))
 			return btrfs_wq_submit_bio(inode, bio, 0, file_offset,
 					btrfs_submit_bio_start_direct_io);
-- 
2.30.2

Re: [PATCH 04/10] btrfs: split btrfs_submit_data_bio

[Qu Wenruo]

On 2022/5/4 20:25, Christoph Hellwig wrote:
> Split btrfs_submit_data_bio into one helper for reads and one for writes.
>
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>   fs/btrfs/ctree.h     |   6 +-
>   fs/btrfs/extent_io.c |  14 +++--
>   fs/btrfs/inode.c     | 134 ++++++++++++++++++++-----------------------
>   3 files changed, 76 insertions(+), 78 deletions(-)
>
> diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
> index 6e939bf01dcc3..6f539ddf7467a 100644
> --- a/fs/btrfs/ctree.h
> +++ b/fs/btrfs/ctree.h
> @@ -3251,8 +3251,10 @@ void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_siz
>   u64 btrfs_file_extent_end(const struct btrfs_path *path);
>
>   /* inode.c */
> -void btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
> -			   int mirror_num, enum btrfs_compression_type compress_type);
> +void btrfs_submit_data_write_bio(struct inode *inode, struct bio *bio,
> +		int mirror_num);
> +void btrfs_submit_data_read_bio(struct inode *inode, struct bio *bio,
> +		int mirror_num, enum btrfs_compression_type compress_type);
>   unsigned int btrfs_verify_data_csum(struct btrfs_bio *bbio,
>   				    u32 bio_offset, struct page *page,
>   				    u64 start, u64 end);
> diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
> index 1b1baeb0d76bc..1394a6f80d285 100644
> --- a/fs/btrfs/extent_io.c
> +++ b/fs/btrfs/extent_io.c
> @@ -182,17 +182,21 @@ static void submit_one_bio(struct bio *bio, int mirror_num,
>   			   enum btrfs_compression_type compress_type)
>   {
>   	struct extent_io_tree *tree = bio->bi_private;
> +	struct inode *inode = tree->private_data;

I guess we shouldn't use the extent_io_tree for bio->bi_private at all
if we're just tring to grab an inode.

In fact, for all submit_one_bio() callers, we are handling buffered
read/write, thus we can grab inode using
bio_first_page_all(bio)->mapping->host.

No need for such weird io_tree based dance.

>
>   	bio->bi_private = NULL;
>
>   	/* Caller should ensure the bio has at least some range added */
>   	ASSERT(bio->bi_iter.bi_size);
>
> -	if (is_data_inode(tree->private_data))
> -		btrfs_submit_data_bio(tree->private_data, bio, mirror_num,
> -					    compress_type);
> +	if (!is_data_inode(tree->private_data))
> +		btrfs_submit_metadata_bio(inode, bio, mirror_num);

Can we just call btrfs_submit_metadata_bio() and return?

Every time I see an "if () else if ()", I can't stop thinking if we have
some corner cases not taken into consideration.

Thanks,
Qu

> +	else if (btrfs_op(bio) == BTRFS_MAP_WRITE)
> +		btrfs_submit_data_write_bio(inode, bio, mirror_num);
>   	else
> -		btrfs_submit_metadata_bio(tree->private_data, bio, mirror_num);
> +		btrfs_submit_data_read_bio(inode, bio, mirror_num,
> +					   compress_type);
> +
>   	/*
>   	 * Above submission hooks will handle the error by ending the bio,
>   	 * which will do the cleanup properly.  So here we should not return
> @@ -2774,7 +2778,7 @@ static blk_status_t submit_data_read_repair(struct inode *inode,
>   		ret = btrfs_repair_one_sector(inode, failed_bio,
>   				bio_offset + offset,
>   				page, pgoff + offset, start + offset,
> -				failed_mirror, btrfs_submit_data_bio);
> +				failed_mirror, btrfs_submit_data_read_bio);
>   		if (!ret) {
>   			/*
>   			 * We have submitted the read repair, the page release
> diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
> index b98f5291e941c..5499b39cab61b 100644
> --- a/fs/btrfs/inode.c
> +++ b/fs/btrfs/inode.c
> @@ -2554,90 +2554,82 @@ static blk_status_t extract_ordered_extent(struct btrfs_inode *inode,
>   	return errno_to_blk_status(ret);
>   }
>
> -/*
> - * extent_io.c submission hook. This does the right thing for csum calculation
> - * on write, or reading the csums from the tree before a read.
> - *
> - * Rules about async/sync submit,
> - * a) read:				sync submit
> - *
> - * b) write without checksum:		sync submit
> - *
> - * c) write with checksum:
> - *    c-1) if bio is issued by fsync:	sync submit
> - *         (sync_writers != 0)
> - *
> - *    c-2) if root is reloc root:	sync submit
> - *         (only in case of buffered IO)
> - *
> - *    c-3) otherwise:			async submit
> - */
> -void btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
> -			   int mirror_num, enum btrfs_compression_type compress_type)
> +void btrfs_submit_data_write_bio(struct inode *inode, struct bio *bio,
> +		int mirror_num)
>   {
>   	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
> -	struct btrfs_root *root = BTRFS_I(inode)->root;
> -	enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
> -	blk_status_t ret = 0;
> -	int skip_sum;
> -	int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
> -
> -	skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
> -		test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
> -
> -	if (btrfs_is_free_space_inode(BTRFS_I(inode)))
> -		metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
> +	struct btrfs_inode *bi = BTRFS_I(inode);
> +	blk_status_t ret;
>
>   	if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
> -		struct page *page = bio_first_bvec_all(bio)->bv_page;
> -		loff_t file_offset = page_offset(page);
> -
> -		ret = extract_ordered_extent(BTRFS_I(inode), bio, file_offset);
> +		ret = extract_ordered_extent(bi, bio,
> +				page_offset(bio_first_bvec_all(bio)->bv_page));
>   		if (ret)
>   			goto out;
>   	}
>
> -	if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
> -		if (compress_type != BTRFS_COMPRESS_NONE) {
> -			/*
> -			 * btrfs_submit_compressed_read will handle completing
> -			 * the bio if there were any errors, so just return
> -			 * here.
> -			 */
> -			btrfs_submit_compressed_read(inode, bio, mirror_num);
> -			return;
> -		} else {
> -			ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
> -			if (ret)
> -				goto out;
> -
> -			/*
> -			 * Lookup bio sums does extra checks around whether we
> -			 * need to csum or not, which is why we ignore skip_sum
> -			 * here.
> -			 */
> -			ret = btrfs_lookup_bio_sums(inode, bio, NULL);
> +	/*
> +	 * Rules for async/sync submit:
> +	 *   a) write without checksum:			sync submit
> +	 *   b) write with checksum:
> +	 *      b-1) if bio is issued by fsync:		sync submit
> +	 *           (sync_writers != 0)
> +	 *      b-2) if root is reloc root:		sync submit
> +	 *           (only in case of buffered IO)
> +	 *      b-3) otherwise:				async submit
> +	 */
> +	if (!(bi->flags & BTRFS_INODE_NODATASUM) &&
> +	    !test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state)) {
> +		if (atomic_read(&bi->sync_writers)) {
> +			ret = btrfs_csum_one_bio(bi, bio, (u64)-1, false);
>   			if (ret)
>   				goto out;
> -		}
> -		goto mapit;
> -	} else if (async && !skip_sum) {
> -		/* csum items have already been cloned */
> -		if (btrfs_is_data_reloc_root(root))
> -			goto mapit;
> -		/* we're doing a write, do the async checksumming */
> -		ret = btrfs_wq_submit_bio(inode, bio, mirror_num,
> -					  0, btrfs_submit_bio_start);
> -		goto out;
> -	} else if (!skip_sum) {
> -		ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, (u64)-1, false);
> -		if (ret)
> +		} else if (btrfs_is_data_reloc_root(bi->root)) {
> +			; /* csum items have already been cloned */
> +		} else {
> +			ret = btrfs_wq_submit_bio(inode, bio,
> +					mirror_num, 0,
> +					btrfs_submit_bio_start);
>   			goto out;
> +		}
>   	}
> -
> -mapit:
>   	ret = btrfs_map_bio(fs_info, bio, mirror_num);
> +out:
> +	if (ret) {
> +		bio->bi_status = ret;
> +		bio_endio(bio);
> +	}
> +}
>
> +void btrfs_submit_data_read_bio(struct inode *inode, struct bio *bio,
> +		int mirror_num, enum btrfs_compression_type compress_type)
> +{
> +	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
> +	blk_status_t ret;
> +
> +	if (compress_type != BTRFS_COMPRESS_NONE) {
> +		/*
> +		 * btrfs_submit_compressed_read will handle completing the bio
> +		 * if there were any errors, so just return here.
> +		 */
> +		btrfs_submit_compressed_read(inode, bio, mirror_num);
> +		return;
> +	}
> +
> +	ret = btrfs_bio_wq_end_io(fs_info, bio,
> +			btrfs_is_free_space_inode(BTRFS_I(inode)) ?
> +			BTRFS_WQ_ENDIO_FREE_SPACE : BTRFS_WQ_ENDIO_DATA);
> +	if (ret)
> +		goto out;
> +
> +	/*
> +	 * Lookup bio sums does extra checks around whether we need to csum or
> +	 * not, which is why we ignore skip_sum here.
> +	 */
> +	ret = btrfs_lookup_bio_sums(inode, bio, NULL);
> +	if (ret)
> +		goto out;
> +	ret = btrfs_map_bio(fs_info, bio, mirror_num);
>   out:
>   	if (ret) {
>   		bio->bi_status = ret;
> @@ -7958,7 +7950,7 @@ static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
>   		goto map;
>
>   	if (write) {
> -		/* check btrfs_submit_data_bio() for async submit rules */
> +		/* check btrfs_submit_data_write_bio() for async submit rules */
>   		if (async_submit && !atomic_read(&BTRFS_I(inode)->sync_writers))
>   			return btrfs_wq_submit_bio(inode, bio, 0, file_offset,
>   					btrfs_submit_bio_start_direct_io);

Re: [PATCH 04/10] btrfs: split btrfs_submit_data_bio

[Christoph Hellwig]

On Wed, May 04, 2022 at 08:38:23PM +0800, Qu Wenruo wrote:
>> +	struct inode *inode = tree->private_data;
>
> I guess we shouldn't use the extent_io_tree for bio->bi_private at all
> if we're just tring to grab an inode.
>
> In fact, for all submit_one_bio() callers, we are handling buffered
> read/write, thus we can grab inode using
> bio_first_page_all(bio)->mapping->host.
>
> No need for such weird io_tree based dance.

Yes, we can eventully.  Not for this series, though.

>> -	if (is_data_inode(tree->private_data))
>> -		btrfs_submit_data_bio(tree->private_data, bio, mirror_num,
>> -					    compress_type);
>> +	if (!is_data_inode(tree->private_data))
>> +		btrfs_submit_metadata_bio(inode, bio, mirror_num);
>
> Can we just call btrfs_submit_metadata_bio() and return?
>
> Every time I see an "if () else if ()", I can't stop thinking if we have
> some corner cases not taken into consideration.

I generally agree with you, but for this case I think it is pretty
simple.  But a few more series down the road these helpers will change
a bit anyway, so we can revisit it.

Re: [PATCH 04/10] btrfs: split btrfs_submit_data_bio

[Qu Wenruo]

On 2022/5/4 22:08, Christoph Hellwig wrote:
> On Wed, May 04, 2022 at 08:38:23PM +0800, Qu Wenruo wrote:
>>> +	struct inode *inode = tree->private_data;
>>
>> I guess we shouldn't use the extent_io_tree for bio->bi_private at all
>> if we're just tring to grab an inode.
>>
>> In fact, for all submit_one_bio() callers, we are handling buffered
>> read/write, thus we can grab inode using
>> bio_first_page_all(bio)->mapping->host.
>>
>> No need for such weird io_tree based dance.
>
> Yes, we can eventully.  Not for this series, though.

Looking forward to the new cleanups on various weird private member usage.
>
>>> -	if (is_data_inode(tree->private_data))
>>> -		btrfs_submit_data_bio(tree->private_data, bio, mirror_num,
>>> -					    compress_type);
>>> +	if (!is_data_inode(tree->private_data))
>>> +		btrfs_submit_metadata_bio(inode, bio, mirror_num);
>>
>> Can we just call btrfs_submit_metadata_bio() and return?
>>
>> Every time I see an "if () else if ()", I can't stop thinking if we have
>> some corner cases not taken into consideration.
>
> I generally agree with you, but for this case I think it is pretty
> simple.  But a few more series down the road these helpers will change
> a bit anyway, so we can revisit it.
>
OK, that sounds good.

Just a little worried about how many series there still are...

Thanks,
Qu

Re: [PATCH 04/10] btrfs: split btrfs_submit_data_bio

[Qu Wenruo]

On 2022/5/5 06:41, Qu Wenruo wrote:
> 
> 
> On 2022/5/4 22:08, Christoph Hellwig wrote:
>> On Wed, May 04, 2022 at 08:38:23PM +0800, Qu Wenruo wrote:
>>>> +    struct inode *inode = tree->private_data;
>>>
>>> I guess we shouldn't use the extent_io_tree for bio->bi_private at all
>>> if we're just tring to grab an inode.
>>>
>>> In fact, for all submit_one_bio() callers, we are handling buffered
>>> read/write, thus we can grab inode using
>>> bio_first_page_all(bio)->mapping->host.
>>>
>>> No need for such weird io_tree based dance.
>>
>> Yes, we can eventully.  Not for this series, though.
> 
> Looking forward to the new cleanups on various weird private member usage.
>>
>>>> -    if (is_data_inode(tree->private_data))
>>>> -        btrfs_submit_data_bio(tree->private_data, bio, mirror_num,
>>>> -                        compress_type);
>>>> +    if (!is_data_inode(tree->private_data))
>>>> +        btrfs_submit_metadata_bio(inode, bio, mirror_num);
>>>
>>> Can we just call btrfs_submit_metadata_bio() and return?
>>>
>>> Every time I see an "if () else if ()", I can't stop thinking if we have
>>> some corner cases not taken into consideration.
>>
>> I generally agree with you, but for this case I think it is pretty
>> simple.  But a few more series down the road these helpers will change
>> a bit anyway, so we can revisit it.
>>
> OK, that sounds good.
> 
> Just a little worried about how many series there still are...
> 
> Thanks,
> Qu
> 

Oh, forgot the tag.

Reviewed-by: Qu Wenruo <wqu@suse.com>

Thanks,
Qu

Re: [PATCH 04/10] btrfs: split btrfs_submit_data_bio

[Christoph Hellwig]

On Thu, May 05, 2022 at 06:41:50AM +0800, Qu Wenruo wrote:
> OK, that sounds good.
>
> Just a little worried about how many series there still are...

A lot.  There is so much fairly low hanging fruit in the btrfs I/O path,
and with Dav preferring series of about 10 patches that means it will
be quite a few.

[PATCH 05/10] btrfs: defer I/O completion based on the btrfs_raid_bio

[Christoph Hellwig]

Instead of attaching a an extra allocation an indirect call to each
low-level bio issued by the RAID code, add a work_struct to struct
btrfs_raid_bio and only defer the per-rbio completion action.  The
per-bio action for all the I/Os are trivial and can be safely done
from interrupt context.

As a nice side effect this also allows sharing the boilerplate code
for the per-bio completions

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/btrfs/ctree.h   |   2 +-
 fs/btrfs/disk-io.c |  12 ++---
 fs/btrfs/disk-io.h |   1 -
 fs/btrfs/raid56.c  | 109 ++++++++++++++++++---------------------------
 4 files changed, 48 insertions(+), 76 deletions(-)

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 6f539ddf7467a..a3739143bf44c 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -853,7 +853,7 @@ struct btrfs_fs_info {
 	struct btrfs_workqueue *flush_workers;
 	struct btrfs_workqueue *endio_workers;
 	struct btrfs_workqueue *endio_meta_workers;
-	struct btrfs_workqueue *endio_raid56_workers;
+	struct workqueue_struct *endio_raid56_workers;
 	struct workqueue_struct *rmw_workers;
 	struct btrfs_workqueue *endio_meta_write_workers;
 	struct btrfs_workqueue *endio_write_workers;
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 56d4d4db976bd..ef117eaab468c 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -753,14 +753,10 @@ static void end_workqueue_bio(struct bio *bio)
 			wq = fs_info->endio_meta_write_workers;
 		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
 			wq = fs_info->endio_freespace_worker;
-		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
-			wq = fs_info->endio_raid56_workers;
 		else
 			wq = fs_info->endio_write_workers;
 	} else {
-		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
-			wq = fs_info->endio_raid56_workers;
-		else if (end_io_wq->metadata)
+		if (end_io_wq->metadata)
 			wq = fs_info->endio_meta_workers;
 		else
 			wq = fs_info->endio_workers;
@@ -2273,7 +2269,8 @@ static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
 	btrfs_destroy_workqueue(fs_info->hipri_workers);
 	btrfs_destroy_workqueue(fs_info->workers);
 	btrfs_destroy_workqueue(fs_info->endio_workers);
-	btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
+	if (fs_info->endio_raid56_workers)
+		destroy_workqueue(fs_info->endio_raid56_workers);
 	if (fs_info->rmw_workers)
 		destroy_workqueue(fs_info->rmw_workers);
 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
@@ -2476,8 +2473,7 @@ static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
 		btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
 				      max_active, 2);
 	fs_info->endio_raid56_workers =
-		btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
-				      max_active, 4);
+		alloc_workqueue("btrfs-endio-raid56", flags, max_active);
 	fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
 	fs_info->endio_write_workers =
 		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h
index 4ee8c42c9f783..809ef065f1666 100644
--- a/fs/btrfs/disk-io.h
+++ b/fs/btrfs/disk-io.h
@@ -21,7 +21,6 @@ enum btrfs_wq_endio_type {
 	BTRFS_WQ_ENDIO_DATA,
 	BTRFS_WQ_ENDIO_METADATA,
 	BTRFS_WQ_ENDIO_FREE_SPACE,
-	BTRFS_WQ_ENDIO_RAID56,
 };
 
 static inline u64 btrfs_sb_offset(int mirror)
diff --git a/fs/btrfs/raid56.c b/fs/btrfs/raid56.c
index a5b623ee6facd..8e089fb4212ff 100644
--- a/fs/btrfs/raid56.c
+++ b/fs/btrfs/raid56.c
@@ -164,6 +164,9 @@ struct btrfs_raid_bio {
 	atomic_t stripes_pending;
 
 	atomic_t error;
+
+	struct work_struct end_io_work;
+
 	/*
 	 * these are two arrays of pointers.  We allocate the
 	 * rbio big enough to hold them both and setup their
@@ -1552,15 +1555,7 @@ static void set_bio_pages_uptodate(struct btrfs_raid_bio *rbio, struct bio *bio)
 	}
 }
 
-/*
- * end io for the read phase of the rmw cycle.  All the bios here are physical
- * stripe bios we've read from the disk so we can recalculate the parity of the
- * stripe.
- *
- * This will usually kick off finish_rmw once all the bios are read in, but it
- * may trigger parity reconstruction if we had any errors along the way
- */
-static void raid_rmw_end_io(struct bio *bio)
+static void raid56_bio_end_io(struct bio *bio)
 {
 	struct btrfs_raid_bio *rbio = bio->bi_private;
 
@@ -1571,23 +1566,34 @@ static void raid_rmw_end_io(struct bio *bio)
 
 	bio_put(bio);
 
-	if (!atomic_dec_and_test(&rbio->stripes_pending))
-		return;
+	if (atomic_dec_and_test(&rbio->stripes_pending))
+		queue_work(rbio->bioc->fs_info->endio_raid56_workers,
+			   &rbio->end_io_work);
+}
 
-	if (atomic_read(&rbio->error) > rbio->bioc->max_errors)
-		goto cleanup;
+/*
+ * End io handler for the read phase of the rmw cycle.  All the bios here are
+ * physical stripe bios we've read from the disk so we can recalculate the
+ * parity of the stripe.
+ *
+ * This will usually kick off finish_rmw once all the bios are read in, but it
+ * may trigger parity reconstruction if we had any errors along the way
+ */
+static void raid56_rmw_end_io_work(struct work_struct *work)
+{
+	struct btrfs_raid_bio *rbio =
+		container_of(work, struct btrfs_raid_bio, end_io_work);
+
+	if (atomic_read(&rbio->error) > rbio->bioc->max_errors) {
+		rbio_orig_end_io(rbio, BLK_STS_IOERR);
+		return;
+	}
 
 	/*
-	 * this will normally call finish_rmw to start our write
-	 * but if there are any failed stripes we'll reconstruct
-	 * from parity first
+	 * This will normally call finish_rmw to start our write but if there
+	 * are any failed stripes we'll reconstruct from parity first.
 	 */
 	validate_rbio_for_rmw(rbio);
-	return;
-
-cleanup:
-
-	rbio_orig_end_io(rbio, BLK_STS_IOERR);
 }
 
 /*
@@ -1662,11 +1668,9 @@ static int raid56_rmw_stripe(struct btrfs_raid_bio *rbio)
 	 * touch it after that.
 	 */
 	atomic_set(&rbio->stripes_pending, bios_to_read);
+	INIT_WORK(&rbio->end_io_work, raid56_rmw_end_io_work);
 	while ((bio = bio_list_pop(&bio_list))) {
-		bio->bi_end_io = raid_rmw_end_io;
-
-		btrfs_bio_wq_end_io(rbio->bioc->fs_info, bio, BTRFS_WQ_ENDIO_RAID56);
-
+		bio->bi_end_io = raid56_bio_end_io;
 		submit_bio(bio);
 	}
 	/* the actual write will happen once the reads are done */
@@ -2108,25 +2112,13 @@ static void __raid_recover_end_io(struct btrfs_raid_bio *rbio)
 }
 
 /*
- * This is called only for stripes we've read from disk to
- * reconstruct the parity.
+ * This is called only for stripes we've read from disk to reconstruct the
+ * parity.
  */
-static void raid_recover_end_io(struct bio *bio)
+static void raid_recover_end_io_work(struct work_struct *work)
 {
-	struct btrfs_raid_bio *rbio = bio->bi_private;
-
-	/*
-	 * we only read stripe pages off the disk, set them
-	 * up to date if there were no errors
-	 */
-	if (bio->bi_status)
-		fail_bio_stripe(rbio, bio);
-	else
-		set_bio_pages_uptodate(rbio, bio);
-	bio_put(bio);
-
-	if (!atomic_dec_and_test(&rbio->stripes_pending))
-		return;
+	struct btrfs_raid_bio *rbio =
+		container_of(work, struct btrfs_raid_bio, end_io_work);
 
 	if (atomic_read(&rbio->error) > rbio->bioc->max_errors)
 		rbio_orig_end_io(rbio, BLK_STS_IOERR);
@@ -2209,11 +2201,9 @@ static int __raid56_parity_recover(struct btrfs_raid_bio *rbio)
 	 * touch it after that.
 	 */
 	atomic_set(&rbio->stripes_pending, bios_to_read);
+	INIT_WORK(&rbio->end_io_work, raid_recover_end_io_work);
 	while ((bio = bio_list_pop(&bio_list))) {
-		bio->bi_end_io = raid_recover_end_io;
-
-		btrfs_bio_wq_end_io(rbio->bioc->fs_info, bio, BTRFS_WQ_ENDIO_RAID56);
-
+		bio->bi_end_io = raid56_bio_end_io;
 		submit_bio(bio);
 	}
 
@@ -2583,7 +2573,6 @@ static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio,
 
 	while ((bio = bio_list_pop(&bio_list))) {
 		bio->bi_end_io = raid_write_end_io;
-
 		submit_bio(bio);
 	}
 	return;
@@ -2671,24 +2660,14 @@ static void validate_rbio_for_parity_scrub(struct btrfs_raid_bio *rbio)
  * This will usually kick off finish_rmw once all the bios are read in, but it
  * may trigger parity reconstruction if we had any errors along the way
  */
-static void raid56_parity_scrub_end_io(struct bio *bio)
+static void raid56_parity_scrub_end_io_work(struct work_struct *work)
 {
-	struct btrfs_raid_bio *rbio = bio->bi_private;
-
-	if (bio->bi_status)
-		fail_bio_stripe(rbio, bio);
-	else
-		set_bio_pages_uptodate(rbio, bio);
-
-	bio_put(bio);
-
-	if (!atomic_dec_and_test(&rbio->stripes_pending))
-		return;
+	struct btrfs_raid_bio *rbio =
+		container_of(work, struct btrfs_raid_bio, end_io_work);
 
 	/*
-	 * this will normally call finish_rmw to start our write
-	 * but if there are any failed stripes we'll reconstruct
-	 * from parity first
+	 * This will normally call finish_rmw to start our write, but if there
+	 * are any failed stripes we'll reconstruct from parity first
 	 */
 	validate_rbio_for_parity_scrub(rbio);
 }
@@ -2758,11 +2737,9 @@ static void raid56_parity_scrub_stripe(struct btrfs_raid_bio *rbio)
 	 * touch it after that.
 	 */
 	atomic_set(&rbio->stripes_pending, bios_to_read);
+	INIT_WORK(&rbio->end_io_work, raid56_parity_scrub_end_io_work);
 	while ((bio = bio_list_pop(&bio_list))) {
-		bio->bi_end_io = raid56_parity_scrub_end_io;
-
-		btrfs_bio_wq_end_io(rbio->bioc->fs_info, bio, BTRFS_WQ_ENDIO_RAID56);
-
+		bio->bi_end_io = raid56_bio_end_io;
 		submit_bio(bio);
 	}
 	/* the actual write will happen once the reads are done */
-- 
2.30.2

[PATCH 06/10] btrfs: don't use btrfs_bio_wq_end_io for compressed writes

[Christoph Hellwig]

Compressed write bio completion is the only user of btrfs_bio_wq_end_io
for writes, and the use of btrfs_bio_wq_end_io is a little suboptimal
here as we only real need user context for the final completion of a
compressed_bio structure, and not every single bio completion.

Add a work_struct to struct compressed_bio instead and use that to call
finish_compressed_bio_write.  This allows to remove all handling of
write bios in the btrfs_bio_wq_end_io infrastructure.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/btrfs/compression.c | 45 +++++++++++++++++++++---------------------
 fs/btrfs/compression.h |  7 +++++--
 fs/btrfs/ctree.h       |  2 +-
 fs/btrfs/disk-io.c     | 30 +++++++++++-----------------
 fs/btrfs/super.c       |  2 --
 5 files changed, 41 insertions(+), 45 deletions(-)

diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c
index f4564f32f6d93..9d5986a30a4a2 100644
--- a/fs/btrfs/compression.c
+++ b/fs/btrfs/compression.c
@@ -403,6 +403,14 @@ static void finish_compressed_bio_write(struct compressed_bio *cb)
 	kfree(cb);
 }
 
+static void btrfs_finish_compressed_write_work(struct work_struct *work)
+{
+	struct compressed_bio *cb =
+		container_of(work, struct compressed_bio, write_end_work);
+
+	finish_compressed_bio_write(cb);
+}
+
 /*
  * Do the cleanup once all the compressed pages hit the disk.  This will clear
  * writeback on the file pages and free the compressed pages.
@@ -414,29 +422,16 @@ static void end_compressed_bio_write(struct bio *bio)
 {
 	struct compressed_bio *cb = bio->bi_private;
 
-	if (!dec_and_test_compressed_bio(cb, bio))
-		goto out;
-
-	btrfs_record_physical_zoned(cb->inode, cb->start, bio);
+	if (dec_and_test_compressed_bio(cb, bio)) {
+		struct btrfs_fs_info *fs_info = btrfs_sb(cb->inode->i_sb);
 
-	finish_compressed_bio_write(cb);
-out:
+		btrfs_record_physical_zoned(cb->inode, cb->start, bio);
+		queue_work(fs_info->compressed_write_workers,
+			   &cb->write_end_work);
+	}
 	bio_put(bio);
 }
 
-static blk_status_t submit_compressed_bio(struct btrfs_fs_info *fs_info,
-					  struct bio *bio, int mirror_num)
-{
-	blk_status_t ret;
-
-	ASSERT(bio->bi_iter.bi_size);
-	ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
-	if (ret)
-		return ret;
-	ret = btrfs_map_bio(fs_info, bio, mirror_num);
-	return ret;
-}
-
 /*
  * Allocate a compressed_bio, which will be used to read/write on-disk
  * (aka, compressed) * data.
@@ -533,7 +528,7 @@ blk_status_t btrfs_submit_compressed_write(struct btrfs_inode *inode, u64 start,
 	cb->compressed_pages = compressed_pages;
 	cb->compressed_len = compressed_len;
 	cb->writeback = writeback;
-	cb->orig_bio = NULL;
+	INIT_WORK(&cb->write_end_work, btrfs_finish_compressed_write_work);
 	cb->nr_pages = nr_pages;
 
 	if (blkcg_css)
@@ -603,7 +598,8 @@ blk_status_t btrfs_submit_compressed_write(struct btrfs_inode *inode, u64 start,
 					goto finish_cb;
 			}
 
-			ret = submit_compressed_bio(fs_info, bio, 0);
+			ASSERT(bio->bi_iter.bi_size);
+			ret = btrfs_map_bio(fs_info, bio, 0);
 			if (ret)
 				goto finish_cb;
 			bio = NULL;
@@ -941,7 +937,12 @@ void btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
 						  fs_info->sectorsize);
 			sums += fs_info->csum_size * nr_sectors;
 
-			ret = submit_compressed_bio(fs_info, comp_bio, mirror_num);
+			ASSERT(comp_bio->bi_iter.bi_size);
+			ret = btrfs_bio_wq_end_io(fs_info, comp_bio,
+						  BTRFS_WQ_ENDIO_DATA);
+			if (ret)
+				goto finish_cb;
+			ret = btrfs_map_bio(fs_info, comp_bio, mirror_num);
 			if (ret)
 				goto finish_cb;
 			comp_bio = NULL;
diff --git a/fs/btrfs/compression.h b/fs/btrfs/compression.h
index 2707404389a5d..4a40725cbf1db 100644
--- a/fs/btrfs/compression.h
+++ b/fs/btrfs/compression.h
@@ -61,8 +61,11 @@ struct compressed_bio {
 	blk_status_t status;
 	int mirror_num;
 
-	/* for reads, this is the bio we are copying the data into */
-	struct bio *orig_bio;
+	union {
+		/* for reads, this is the bio we are copying the data into */
+		struct bio *orig_bio;
+		struct work_struct write_end_work;
+	};
 
 	/*
 	 * the start of a variable length array of checksums only
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index a3739143bf44c..6cf699959286d 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -855,7 +855,7 @@ struct btrfs_fs_info {
 	struct btrfs_workqueue *endio_meta_workers;
 	struct workqueue_struct *endio_raid56_workers;
 	struct workqueue_struct *rmw_workers;
-	struct btrfs_workqueue *endio_meta_write_workers;
+	struct workqueue_struct *compressed_write_workers;
 	struct btrfs_workqueue *endio_write_workers;
 	struct btrfs_workqueue *endio_freespace_worker;
 	struct btrfs_workqueue *caching_workers;
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index ef117eaab468c..aa56ba9378e1f 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -748,19 +748,10 @@ static void end_workqueue_bio(struct bio *bio)
 	fs_info = end_io_wq->info;
 	end_io_wq->status = bio->bi_status;
 
-	if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
-		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
-			wq = fs_info->endio_meta_write_workers;
-		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
-			wq = fs_info->endio_freespace_worker;
-		else
-			wq = fs_info->endio_write_workers;
-	} else {
-		if (end_io_wq->metadata)
-			wq = fs_info->endio_meta_workers;
-		else
-			wq = fs_info->endio_workers;
-	}
+	if (end_io_wq->metadata)
+		wq = fs_info->endio_meta_workers;
+	else
+		wq = fs_info->endio_workers;
 
 	btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
 	btrfs_queue_work(wq, &end_io_wq->work);
@@ -771,6 +762,9 @@ blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 {
 	struct btrfs_end_io_wq *end_io_wq;
 
+	if (WARN_ON_ONCE(btrfs_op(bio) != BTRFS_MAP_WRITE))
+		return BLK_STS_IOERR;
+
 	end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 	if (!end_io_wq)
 		return BLK_STS_RESOURCE;
@@ -2273,6 +2267,8 @@ static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
 		destroy_workqueue(fs_info->endio_raid56_workers);
 	if (fs_info->rmw_workers)
 		destroy_workqueue(fs_info->rmw_workers);
+	if (fs_info->compressed_write_workers)
+		destroy_workqueue(fs_info->compressed_write_workers);
 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
 	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
 	btrfs_destroy_workqueue(fs_info->delayed_workers);
@@ -2287,7 +2283,6 @@ static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
 	 * queues can do metadata I/O operations.
 	 */
 	btrfs_destroy_workqueue(fs_info->endio_meta_workers);
-	btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
 }
 
 static void free_root_extent_buffers(struct btrfs_root *root)
@@ -2469,15 +2464,14 @@ static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
 	fs_info->endio_meta_workers =
 		btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
 				      max_active, 4);
-	fs_info->endio_meta_write_workers =
-		btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
-				      max_active, 2);
 	fs_info->endio_raid56_workers =
 		alloc_workqueue("btrfs-endio-raid56", flags, max_active);
 	fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
 	fs_info->endio_write_workers =
 		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
 				      max_active, 2);
+	fs_info->compressed_write_workers =
+		alloc_workqueue("btrfs-compressed-write", flags, max_active);
 	fs_info->endio_freespace_worker =
 		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
 				      max_active, 0);
@@ -2492,7 +2486,7 @@ static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
 	if (!(fs_info->workers && fs_info->hipri_workers &&
 	      fs_info->delalloc_workers && fs_info->flush_workers &&
 	      fs_info->endio_workers && fs_info->endio_meta_workers &&
-	      fs_info->endio_meta_write_workers &&
+	      fs_info->compressed_write_workers &&
 	      fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
 	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
 	      fs_info->caching_workers && fs_info->fixup_workers &&
diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
index b1fdc6a26c76e..9c683c466d585 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -1908,8 +1908,6 @@ static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
 	btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
 	btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
 	btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
-	btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
-				new_pool_size);
 	btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
 	btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
 	btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
-- 
2.30.2

Re: [PATCH 06/10] btrfs: don't use btrfs_bio_wq_end_io for compressed writes

[Nikolay Borisov]

On 4.05.22 г. 15:25 ч., Christoph Hellwig wrote:
> Compressed write bio completion is the only user of btrfs_bio_wq_end_io
> for writes, and the use of btrfs_bio_wq_end_io is a little suboptimal
> here as we only real need user context for the final completion of a
> compressed_bio structure, and not every single bio completion.
> 
> Add a work_struct to struct compressed_bio instead and use that to call
> finish_compressed_bio_write.  This allows to remove all handling of
> write bios in the btrfs_bio_wq_end_io infrastructure.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>   fs/btrfs/compression.c | 45 +++++++++++++++++++++---------------------
>   fs/btrfs/compression.h |  7 +++++--
>   fs/btrfs/ctree.h       |  2 +-
>   fs/btrfs/disk-io.c     | 30 +++++++++++-----------------
>   fs/btrfs/super.c       |  2 --
>   5 files changed, 41 insertions(+), 45 deletions(-)
> 
> diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c
> index f4564f32f6d93..9d5986a30a4a2 100644
> --- a/fs/btrfs/compression.c
> +++ b/fs/btrfs/compression.c
> @@ -403,6 +403,14 @@ static void finish_compressed_bio_write(struct compressed_bio *cb)
>   	kfree(cb);
>   }
>   
> +static void btrfs_finish_compressed_write_work(struct work_struct *work)
> +{
> +	struct compressed_bio *cb =
> +		container_of(work, struct compressed_bio, write_end_work);
> +
> +	finish_compressed_bio_write(cb);
> +}
> +
>   /*
>    * Do the cleanup once all the compressed pages hit the disk.  This will clear
>    * writeback on the file pages and free the compressed pages.
> @@ -414,29 +422,16 @@ static void end_compressed_bio_write(struct bio *bio)
>   {
>   	struct compressed_bio *cb = bio->bi_private;
>   
> -	if (!dec_and_test_compressed_bio(cb, bio))
> -		goto out;
> -
> -	btrfs_record_physical_zoned(cb->inode, cb->start, bio);
> +	if (dec_and_test_compressed_bio(cb, bio)) {
> +		struct btrfs_fs_info *fs_info = btrfs_sb(cb->inode->i_sb);
>   
> -	finish_compressed_bio_write(cb);
> -out:
> +		btrfs_record_physical_zoned(cb->inode, cb->start, bio);
> +		queue_work(fs_info->compressed_write_workers,
> +			   &cb->write_end_work);
> +	}
>   	bio_put(bio);
>   }
>   
> -static blk_status_t submit_compressed_bio(struct btrfs_fs_info *fs_info,
> -					  struct bio *bio, int mirror_num)
> -{
> -	blk_status_t ret;
> -
> -	ASSERT(bio->bi_iter.bi_size);
> -	ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
> -	if (ret)
> -		return ret;
> -	ret = btrfs_map_bio(fs_info, bio, mirror_num);
> -	return ret;
> -}
> -
>   /*
>    * Allocate a compressed_bio, which will be used to read/write on-disk
>    * (aka, compressed) * data.
> @@ -533,7 +528,7 @@ blk_status_t btrfs_submit_compressed_write(struct btrfs_inode *inode, u64 start,
>   	cb->compressed_pages = compressed_pages;
>   	cb->compressed_len = compressed_len;
>   	cb->writeback = writeback;
> -	cb->orig_bio = NULL;
> +	INIT_WORK(&cb->write_end_work, btrfs_finish_compressed_write_work);
>   	cb->nr_pages = nr_pages;
>   
>   	if (blkcg_css)
> @@ -603,7 +598,8 @@ blk_status_t btrfs_submit_compressed_write(struct btrfs_inode *inode, u64 start,
>   					goto finish_cb;
>   			}
>   
> -			ret = submit_compressed_bio(fs_info, bio, 0);
> +			ASSERT(bio->bi_iter.bi_size);
> +			ret = btrfs_map_bio(fs_info, bio, 0);
>   			if (ret)
>   				goto finish_cb;
>   			bio = NULL;
> @@ -941,7 +937,12 @@ void btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
>   						  fs_info->sectorsize);
>   			sums += fs_info->csum_size * nr_sectors;
>   
> -			ret = submit_compressed_bio(fs_info, comp_bio, mirror_num);
> +			ASSERT(comp_bio->bi_iter.bi_size);
> +			ret = btrfs_bio_wq_end_io(fs_info, comp_bio,
> +						  BTRFS_WQ_ENDIO_DATA);
> +			if (ret)
> +				goto finish_cb;
> +			ret = btrfs_map_bio(fs_info, comp_bio, mirror_num);
>   			if (ret)
>   				goto finish_cb;
>   			comp_bio = NULL;
> diff --git a/fs/btrfs/compression.h b/fs/btrfs/compression.h
> index 2707404389a5d..4a40725cbf1db 100644
> --- a/fs/btrfs/compression.h
> +++ b/fs/btrfs/compression.h
> @@ -61,8 +61,11 @@ struct compressed_bio {
>   	blk_status_t status;
>   	int mirror_num;
>   
> -	/* for reads, this is the bio we are copying the data into */
> -	struct bio *orig_bio;
> +	union {
> +		/* for reads, this is the bio we are copying the data into */
> +		struct bio *orig_bio;
> +		struct work_struct write_end_work;
> +	};
>   
>   	/*
>   	 * the start of a variable length array of checksums only
> diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
> index a3739143bf44c..6cf699959286d 100644
> --- a/fs/btrfs/ctree.h
> +++ b/fs/btrfs/ctree.h
> @@ -855,7 +855,7 @@ struct btrfs_fs_info {
>   	struct btrfs_workqueue *endio_meta_workers;
>   	struct workqueue_struct *endio_raid56_workers;
>   	struct workqueue_struct *rmw_workers;
> -	struct btrfs_workqueue *endio_meta_write_workers;
> +	struct workqueue_struct *compressed_write_workers;
>   	struct btrfs_workqueue *endio_write_workers;
>   	struct btrfs_workqueue *endio_freespace_worker;
>   	struct btrfs_workqueue *caching_workers;
> diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
> index ef117eaab468c..aa56ba9378e1f 100644
> --- a/fs/btrfs/disk-io.c
> +++ b/fs/btrfs/disk-io.c
> @@ -748,19 +748,10 @@ static void end_workqueue_bio(struct bio *bio)
>   	fs_info = end_io_wq->info;
>   	end_io_wq->status = bio->bi_status;
>   
> -	if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
> -		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
> -			wq = fs_info->endio_meta_write_workers;
> -		else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
> -			wq = fs_info->endio_freespace_worker;
> -		else
> -			wq = fs_info->endio_write_workers;
> -	} else {
> -		if (end_io_wq->metadata)
> -			wq = fs_info->endio_meta_workers;
> -		else
> -			wq = fs_info->endio_workers;
> -	}
> +	if (end_io_wq->metadata)
> +		wq = fs_info->endio_meta_workers;
> +	else
> +		wq = fs_info->endio_workers;
>   
>   	btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
>   	btrfs_queue_work(wq, &end_io_wq->work);
> @@ -771,6 +762,9 @@ blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
>   {
>   	struct btrfs_end_io_wq *end_io_wq;
>   
> +	if (WARN_ON_ONCE(btrfs_op(bio) != BTRFS_MAP_WRITE))
> +		return BLK_STS_IOERR;

This is broken w.r.t to btrfs_submit_metadata_bio since in
that function the code is:

  if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
                       /*
                        * called for a read, do the setup so that checksum validation
                        * can happen in the async kernel threads
                        */
                     ret = btrfs_bio_wq_end_io(fs_info, bio,
                                                 BTRFS_WQ_ENDIO_METADATA);
                       if (!ret)
                               ret = btrfs_map_bio(fs_info, bio, mirror_num);



So metadata reads  will end up in function which disallows reads..
This results in failure to mount when reading various metadata.

I guess the correct fix is to include the following hunk from patch 8 in this series:

@@ -916,14 +839,7 @@ void btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, int mirror_
  	bio->bi_opf |= REQ_META;
  
  	if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
-		/*
-		 * called for a read, do the setup so that checksum validation
-		 * can happen in the async kernel threads
-		 */
-		ret = btrfs_bio_wq_end_io(fs_info, bio,
-					  BTRFS_WQ_ENDIO_METADATA);
-		if (!ret)
-			ret = btrfs_map_bio(fs_info, bio, mirror_num);
+		ret = btrfs_map_bio(fs_info, bio, mirror_num);





> +
>   	end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
>   	if (!end_io_wq)
>   		return BLK_STS_RESOURCE;
> @@ -2273,6 +2267,8 @@ static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
>   		destroy_workqueue(fs_info->endio_raid56_workers);
>   	if (fs_info->rmw_workers)
>   		destroy_workqueue(fs_info->rmw_workers);
> +	if (fs_info->compressed_write_workers)
> +		destroy_workqueue(fs_info->compressed_write_workers);
>   	btrfs_destroy_workqueue(fs_info->endio_write_workers);
>   	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
>   	btrfs_destroy_workqueue(fs_info->delayed_workers);
> @@ -2287,7 +2283,6 @@ static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
>   	 * queues can do metadata I/O operations.
>   	 */
>   	btrfs_destroy_workqueue(fs_info->endio_meta_workers);
> -	btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
>   }
>   
>   static void free_root_extent_buffers(struct btrfs_root *root)
> @@ -2469,15 +2464,14 @@ static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
>   	fs_info->endio_meta_workers =
>   		btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
>   				      max_active, 4);
> -	fs_info->endio_meta_write_workers =
> -		btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
> -				      max_active, 2);
>   	fs_info->endio_raid56_workers =
>   		alloc_workqueue("btrfs-endio-raid56", flags, max_active);
>   	fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
>   	fs_info->endio_write_workers =
>   		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
>   				      max_active, 2);
> +	fs_info->compressed_write_workers =
> +		alloc_workqueue("btrfs-compressed-write", flags, max_active);
>   	fs_info->endio_freespace_worker =
>   		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
>   				      max_active, 0);
> @@ -2492,7 +2486,7 @@ static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
>   	if (!(fs_info->workers && fs_info->hipri_workers &&
>   	      fs_info->delalloc_workers && fs_info->flush_workers &&
>   	      fs_info->endio_workers && fs_info->endio_meta_workers &&
> -	      fs_info->endio_meta_write_workers &&
> +	      fs_info->compressed_write_workers &&
>   	      fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
>   	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
>   	      fs_info->caching_workers && fs_info->fixup_workers &&
> diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
> index b1fdc6a26c76e..9c683c466d585 100644
> --- a/fs/btrfs/super.c
> +++ b/fs/btrfs/super.c
> @@ -1908,8 +1908,6 @@ static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
>   	btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
>   	btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
>   	btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
> -	btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
> -				new_pool_size);
>   	btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
>   	btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
>   	btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);

Re: [PATCH 06/10] btrfs: don't use btrfs_bio_wq_end_io for compressed writes

[Nikolay Borisov]

On 11.05.22 г. 22:20 ч., Nikolay Borisov wrote:
> 
> 
> On 4.05.22 г. 15:25 ч., Christoph Hellwig wrote:
>> Compressed write bio completion is the only user of btrfs_bio_wq_end_io
>> for writes, and the use of btrfs_bio_wq_end_io is a little suboptimal
>> here as we only real need user context for the final completion of a
>> compressed_bio structure, and not every single bio completion.
>>
>> Add a work_struct to struct compressed_bio instead and use that to call
>> finish_compressed_bio_write.  This allows to remove all handling of
>> write bios in the btrfs_bio_wq_end_io infrastructure.
>>
>> Signed-off-by: Christoph Hellwig <hch@lst.de>

<snip>


>> @@ -771,6 +762,9 @@ blk_status_t btrfs_bio_wq_end_io(struct 
>> btrfs_fs_info *info, struct bio *bio,
>>   {
>>       struct btrfs_end_io_wq *end_io_wq;
>> +    if (WARN_ON_ONCE(btrfs_op(bio) != BTRFS_MAP_WRITE))
>> +        return BLK_STS_IOERR;
> 
> This is broken w.r.t to btrfs_submit_metadata_bio since in
> that function the code is:
> 
>   if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
>                        /*
>                         * called for a read, do the setup so that 
> checksum validation
>                         * can happen in the async kernel threads
>                         */
>                      ret = btrfs_bio_wq_end_io(fs_info, bio,
>                                                  BTRFS_WQ_ENDIO_METADATA);
>                        if (!ret)
>                                ret = btrfs_map_bio(fs_info, bio, 
> mirror_num);
> 
> 
> 
> So metadata reads  will end up in function which disallows reads..
> This results in failure to mount when reading various metadata.
> 
> I guess the correct fix is to include the following hunk from patch 8 in 
> this series:
> 
> @@ -916,14 +839,7 @@ void btrfs_submit_metadata_bio(struct inode *inode, 
> struct bio *bio, int mirror_
>       bio->bi_opf |= REQ_META;
> 
>       if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
> -        /*
> -         * called for a read, do the setup so that checksum validation
> -         * can happen in the async kernel threads
> -         */
> -        ret = btrfs_bio_wq_end_io(fs_info, bio,
> -                      BTRFS_WQ_ENDIO_METADATA);
> -        if (!ret)
> -            ret = btrfs_map_bio(fs_info, bio, mirror_num);
> +        ret = btrfs_map_bio(fs_info, bio, mirror_num);
> 
> 
> 
> 
> 

Even if I apply this hunk to patch 6 I get more io failures, this time 
from btrfs_submit_data_read_bio, again there is relevant hunk in patch 8:

-	ret = btrfs_bio_wq_end_io(fs_info, bio,
-			btrfs_is_free_space_inode(BTRFS_I(inode)) ?
-			BTRFS_WQ_ENDIO_FREE_SPACE : BTRFS_WQ_ENDIO_DATA);
-	if (ret)
-		goto out;
-

And then there is another due to DIO in btrfs_submit_dio_bio's call of 
btrfs_bio_wq_end_io if we want to read.


Please fix those to ensure the series is actually bisectable.


<Snip>

Re: [PATCH 06/10] btrfs: don't use btrfs_bio_wq_end_io for compressed writes

[Christoph Hellwig]

On Wed, May 11, 2022 at 10:28:02PM +0300, Nikolay Borisov wrote:
> Please fix those to ensure the series is actually bisectable.

Yeah.  Funny thing is this how I had it until the last repost,
just to reshuffle it for some reason.  I should have just left it as-is..

[PATCH 07/10] btrfs: centralize setting REQ_META

[Christoph Hellwig]

Set REQ_META in btrfs_submit_metadata_bio instead of the various callers.
We'll start relying on this flag inside of btrfs in a bit, and this
ensures it is always set correctly.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Qu Wenruo <wqu@suse.com>
---
 fs/btrfs/disk-io.c   | 2 ++
 fs/btrfs/extent_io.c | 8 ++++----
 2 files changed, 6 insertions(+), 4 deletions(-)

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index aa56ba9378e1f..8fed801423504 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -913,6 +913,8 @@ void btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, int mirror_
 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 	blk_status_t ret;
 
+	bio->bi_opf |= REQ_META;
+
 	if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
 		/*
 		 * called for a read, do the setup so that checksum validation
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 1394a6f80d285..93191771e5bf1 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -4581,7 +4581,7 @@ static int write_one_subpage_eb(struct extent_buffer *eb,
 {
 	struct btrfs_fs_info *fs_info = eb->fs_info;
 	struct page *page = eb->pages[0];
-	unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
+	unsigned int write_flags = wbc_to_write_flags(wbc);
 	bool no_dirty_ebs = false;
 	int ret;
 
@@ -4626,7 +4626,7 @@ static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
 {
 	u64 disk_bytenr = eb->start;
 	int i, num_pages;
-	unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
+	unsigned int write_flags = wbc_to_write_flags(wbc);
 	int ret = 0;
 
 	prepare_eb_write(eb);
@@ -6637,7 +6637,7 @@ static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait,
 	btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len);
 
 	btrfs_subpage_start_reader(fs_info, page, eb->start, eb->len);
-	ret = submit_extent_page(REQ_OP_READ | REQ_META, NULL, &bio_ctrl,
+	ret = submit_extent_page(REQ_OP_READ, NULL, &bio_ctrl,
 				 page, eb->start, eb->len,
 				 eb->start - page_offset(page),
 				 end_bio_extent_readpage, mirror_num, 0,
@@ -6744,7 +6744,7 @@ int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num)
 			}
 
 			ClearPageError(page);
-			err = submit_extent_page(REQ_OP_READ | REQ_META, NULL,
+			err = submit_extent_page(REQ_OP_READ, NULL,
 					 &bio_ctrl, page, page_offset(page),
 					 PAGE_SIZE, 0, end_bio_extent_readpage,
 					 mirror_num, 0, false);
-- 
2.30.2

[PATCH 08/10] btrfs: remove btrfs_end_io_wq

[Christoph Hellwig]

All reads bio that go through btrfs_map_bio need to be completed in
user context.  And read I/Os are the most common and timing critical
in almost any file system workloads.

Embedd a work_struct into struct btrfs_bio and use it to complete all
read bios submitted through btrfs_map, using the REQ_META flag to decide
which workqueue they are placed on.

This removes the need for a separate 128 byte allocation (typically
rounded up to 192 bytes by slab) for all reads with a size increase
of 24 bytes for struct btrfs_bio.  Future patches will reorgnize
struct btrfs_bio to make use of this extra space for writes as well.

(all sizes are based a on typical 64-bit non-debug build)

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/btrfs/compression.c |   4 --
 fs/btrfs/ctree.h       |   4 +-
 fs/btrfs/disk-io.c     | 120 +++--------------------------------------
 fs/btrfs/disk-io.h     |  10 ----
 fs/btrfs/inode.c       |  24 +--------
 fs/btrfs/super.c       |  11 +---
 fs/btrfs/volumes.c     |  33 ++++++++++--
 fs/btrfs/volumes.h     |   3 ++
 8 files changed, 42 insertions(+), 167 deletions(-)

diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c
index 9d5986a30a4a2..c73ee8ae070d7 100644
--- a/fs/btrfs/compression.c
+++ b/fs/btrfs/compression.c
@@ -938,10 +938,6 @@ void btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
 			sums += fs_info->csum_size * nr_sectors;
 
 			ASSERT(comp_bio->bi_iter.bi_size);
-			ret = btrfs_bio_wq_end_io(fs_info, comp_bio,
-						  BTRFS_WQ_ENDIO_DATA);
-			if (ret)
-				goto finish_cb;
 			ret = btrfs_map_bio(fs_info, comp_bio, mirror_num);
 			if (ret)
 				goto finish_cb;
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 6cf699959286d..c839c3e77c21e 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -851,8 +851,8 @@ struct btrfs_fs_info {
 	struct btrfs_workqueue *hipri_workers;
 	struct btrfs_workqueue *delalloc_workers;
 	struct btrfs_workqueue *flush_workers;
-	struct btrfs_workqueue *endio_workers;
-	struct btrfs_workqueue *endio_meta_workers;
+	struct workqueue_struct *endio_workers;
+	struct workqueue_struct *endio_meta_workers;
 	struct workqueue_struct *endio_raid56_workers;
 	struct workqueue_struct *rmw_workers;
 	struct workqueue_struct *compressed_write_workers;
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 8fed801423504..1d9330443c716 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -50,7 +50,6 @@
 				 BTRFS_SUPER_FLAG_METADUMP |\
 				 BTRFS_SUPER_FLAG_METADUMP_V2)
 
-static void end_workqueue_fn(struct btrfs_work *work);
 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
 				      struct btrfs_fs_info *fs_info);
@@ -63,40 +62,6 @@ static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
 
-/*
- * btrfs_end_io_wq structs are used to do processing in task context when an IO
- * is complete.  This is used during reads to verify checksums, and it is used
- * by writes to insert metadata for new file extents after IO is complete.
- */
-struct btrfs_end_io_wq {
-	struct bio *bio;
-	bio_end_io_t *end_io;
-	void *private;
-	struct btrfs_fs_info *info;
-	blk_status_t status;
-	enum btrfs_wq_endio_type metadata;
-	struct btrfs_work work;
-};
-
-static struct kmem_cache *btrfs_end_io_wq_cache;
-
-int __init btrfs_end_io_wq_init(void)
-{
-	btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
-					sizeof(struct btrfs_end_io_wq),
-					0,
-					SLAB_MEM_SPREAD,
-					NULL);
-	if (!btrfs_end_io_wq_cache)
-		return -ENOMEM;
-	return 0;
-}
-
-void __cold btrfs_end_io_wq_exit(void)
-{
-	kmem_cache_destroy(btrfs_end_io_wq_cache);
-}
-
 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
 {
 	if (fs_info->csum_shash)
@@ -739,48 +704,6 @@ int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
 	return ret;
 }
 
-static void end_workqueue_bio(struct bio *bio)
-{
-	struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
-	struct btrfs_fs_info *fs_info;
-	struct btrfs_workqueue *wq;
-
-	fs_info = end_io_wq->info;
-	end_io_wq->status = bio->bi_status;
-
-	if (end_io_wq->metadata)
-		wq = fs_info->endio_meta_workers;
-	else
-		wq = fs_info->endio_workers;
-
-	btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
-	btrfs_queue_work(wq, &end_io_wq->work);
-}
-
-blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
-			enum btrfs_wq_endio_type metadata)
-{
-	struct btrfs_end_io_wq *end_io_wq;
-
-	if (WARN_ON_ONCE(btrfs_op(bio) != BTRFS_MAP_WRITE))
-		return BLK_STS_IOERR;
-
-	end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
-	if (!end_io_wq)
-		return BLK_STS_RESOURCE;
-
-	end_io_wq->private = bio->bi_private;
-	end_io_wq->end_io = bio->bi_end_io;
-	end_io_wq->info = info;
-	end_io_wq->status = 0;
-	end_io_wq->bio = bio;
-	end_io_wq->metadata = metadata;
-
-	bio->bi_private = end_io_wq;
-	bio->bi_end_io = end_workqueue_bio;
-	return 0;
-}
-
 static void run_one_async_start(struct btrfs_work *work)
 {
 	struct async_submit_bio *async;
@@ -916,14 +839,7 @@ void btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, int mirror_
 	bio->bi_opf |= REQ_META;
 
 	if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
-		/*
-		 * called for a read, do the setup so that checksum validation
-		 * can happen in the async kernel threads
-		 */
-		ret = btrfs_bio_wq_end_io(fs_info, bio,
-					  BTRFS_WQ_ENDIO_METADATA);
-		if (!ret)
-			ret = btrfs_map_bio(fs_info, bio, mirror_num);
+		ret = btrfs_map_bio(fs_info, bio, mirror_num);
 	} else if (!should_async_write(fs_info, BTRFS_I(inode))) {
 		ret = btree_csum_one_bio(bio);
 		if (!ret)
@@ -1939,25 +1855,6 @@ struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
 	return root;
 }
 
-/*
- * called by the kthread helper functions to finally call the bio end_io
- * functions.  This is where read checksum verification actually happens
- */
-static void end_workqueue_fn(struct btrfs_work *work)
-{
-	struct bio *bio;
-	struct btrfs_end_io_wq *end_io_wq;
-
-	end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
-	bio = end_io_wq->bio;
-
-	bio->bi_status = end_io_wq->status;
-	bio->bi_private = end_io_wq->private;
-	bio->bi_end_io = end_io_wq->end_io;
-	bio_endio(bio);
-	kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
-}
-
 static int cleaner_kthread(void *arg)
 {
 	struct btrfs_fs_info *fs_info = arg;
@@ -2264,7 +2161,8 @@ static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
 	btrfs_destroy_workqueue(fs_info->delalloc_workers);
 	btrfs_destroy_workqueue(fs_info->hipri_workers);
 	btrfs_destroy_workqueue(fs_info->workers);
-	btrfs_destroy_workqueue(fs_info->endio_workers);
+	if (fs_info->endio_workers)
+		destroy_workqueue(fs_info->endio_workers);
 	if (fs_info->endio_raid56_workers)
 		destroy_workqueue(fs_info->endio_raid56_workers);
 	if (fs_info->rmw_workers)
@@ -2284,7 +2182,8 @@ static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
 	 * the queues used for metadata I/O, since tasks from those other work
 	 * queues can do metadata I/O operations.
 	 */
-	btrfs_destroy_workqueue(fs_info->endio_meta_workers);
+	if (fs_info->endio_meta_workers)
+		destroy_workqueue(fs_info->endio_meta_workers);
 }
 
 static void free_root_extent_buffers(struct btrfs_root *root)
@@ -2457,15 +2356,10 @@ static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
 	fs_info->fixup_workers =
 		btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
 
-	/*
-	 * endios are largely parallel and should have a very
-	 * low idle thresh
-	 */
 	fs_info->endio_workers =
-		btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
+		alloc_workqueue("btrfs-endio", flags, max_active);
 	fs_info->endio_meta_workers =
-		btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
-				      max_active, 4);
+		alloc_workqueue("btrfs-endio-meta", flags, max_active);
 	fs_info->endio_raid56_workers =
 		alloc_workqueue("btrfs-endio-raid56", flags, max_active);
 	fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h
index 809ef065f1666..05e779a41a997 100644
--- a/fs/btrfs/disk-io.h
+++ b/fs/btrfs/disk-io.h
@@ -17,12 +17,6 @@
  */
 #define BTRFS_BDEV_BLOCKSIZE	(4096)
 
-enum btrfs_wq_endio_type {
-	BTRFS_WQ_ENDIO_DATA,
-	BTRFS_WQ_ENDIO_METADATA,
-	BTRFS_WQ_ENDIO_FREE_SPACE,
-};
-
 static inline u64 btrfs_sb_offset(int mirror)
 {
 	u64 start = SZ_16K;
@@ -120,8 +114,6 @@ int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
 			  int atomic);
 int btrfs_read_extent_buffer(struct extent_buffer *buf, u64 parent_transid,
 			     int level, struct btrfs_key *first_key);
-blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
-			enum btrfs_wq_endio_type metadata);
 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
 				 int mirror_num, u64 dio_file_offset,
 				 extent_submit_bio_start_t *submit_bio_start);
@@ -144,8 +136,6 @@ int btree_lock_page_hook(struct page *page, void *data,
 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags);
 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid);
 int btrfs_init_root_free_objectid(struct btrfs_root *root);
-int __init btrfs_end_io_wq_init(void);
-void __cold btrfs_end_io_wq_exit(void);
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 void btrfs_set_buffer_lockdep_class(u64 objectid,
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 5499b39cab61b..8195fd70633ab 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -2616,12 +2616,6 @@ void btrfs_submit_data_read_bio(struct inode *inode, struct bio *bio,
 		return;
 	}
 
-	ret = btrfs_bio_wq_end_io(fs_info, bio,
-			btrfs_is_free_space_inode(BTRFS_I(inode)) ?
-			BTRFS_WQ_ENDIO_FREE_SPACE : BTRFS_WQ_ENDIO_DATA);
-	if (ret)
-		goto out;
-
 	/*
 	 * Lookup bio sums does extra checks around whether we need to csum or
 	 * not, which is why we ignore skip_sum here.
@@ -7834,9 +7828,6 @@ static void submit_dio_repair_bio(struct inode *inode, struct bio *bio,
 
 	BUG_ON(bio_op(bio) == REQ_OP_WRITE);
 
-	if (btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA))
-		return;
-
 	refcount_inc(&dip->refs);
 	if (btrfs_map_bio(fs_info, bio, mirror_num))
 		refcount_dec(&dip->refs);
@@ -7937,19 +7928,12 @@ static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
 {
 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 	struct btrfs_dio_private *dip = bio->bi_private;
-	bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
 	blk_status_t ret;
 
-	if (!write) {
-		ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
-		if (ret)
-			return ret;
-	}
-
 	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
 		goto map;
 
-	if (write) {
+	if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
 		/* check btrfs_submit_data_write_bio() for async submit rules */
 		if (async_submit && !atomic_read(&BTRFS_I(inode)->sync_writers))
 			return btrfs_wq_submit_bio(inode, bio, 0, file_offset,
@@ -10298,12 +10282,6 @@ static blk_status_t submit_encoded_read_bio(struct btrfs_inode *inode,
 			return ret;
 	}
 
-	ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
-	if (ret) {
-		btrfs_bio_free_csum(bbio);
-		return ret;
-	}
-
 	atomic_inc(&priv->pending);
 	ret = btrfs_map_bio(fs_info, bio, mirror_num);
 	if (ret) {
diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
index 9c683c466d585..64eb8aeed156f 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -1906,8 +1906,6 @@ static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
 	btrfs_workqueue_set_max(fs_info->hipri_workers, new_pool_size);
 	btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
 	btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
-	btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
-	btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
 	btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
 	btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
 	btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
@@ -2668,13 +2666,9 @@ static int __init init_btrfs_fs(void)
 	if (err)
 		goto free_delayed_ref;
 
-	err = btrfs_end_io_wq_init();
-	if (err)
-		goto free_prelim_ref;
-
 	err = btrfs_interface_init();
 	if (err)
-		goto free_end_io_wq;
+		goto free_prelim_ref;
 
 	btrfs_print_mod_info();
 
@@ -2690,8 +2684,6 @@ static int __init init_btrfs_fs(void)
 
 unregister_ioctl:
 	btrfs_interface_exit();
-free_end_io_wq:
-	btrfs_end_io_wq_exit();
 free_prelim_ref:
 	btrfs_prelim_ref_exit();
 free_delayed_ref:
@@ -2729,7 +2721,6 @@ static void __exit exit_btrfs_fs(void)
 	extent_state_cache_exit();
 	extent_io_exit();
 	btrfs_interface_exit();
-	btrfs_end_io_wq_exit();
 	unregister_filesystem(&btrfs_fs_type);
 	btrfs_exit_sysfs();
 	btrfs_cleanup_fs_uuids();
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index aeacb87457687..5f18e9105fe08 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -6621,11 +6621,27 @@ int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
 	return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 0, 1);
 }
 
-static inline void btrfs_end_bioc(struct btrfs_io_context *bioc)
+static struct workqueue_struct *btrfs_end_io_wq(struct btrfs_io_context *bioc)
+{
+	if (bioc->orig_bio->bi_opf & REQ_META)
+		return bioc->fs_info->endio_meta_workers;
+	return bioc->fs_info->endio_workers;
+}
+
+static void btrfs_end_bio_work(struct work_struct *work)
+{
+	struct btrfs_bio *bbio =
+		container_of(work, struct btrfs_bio, end_io_work);
+
+	bio_endio(&bbio->bio);
+}
+
+static void btrfs_end_bioc(struct btrfs_io_context *bioc, bool async)
 {
 	struct bio *orig_bio = bioc->orig_bio;
+	struct btrfs_bio *bbio = btrfs_bio(orig_bio);
 
-	btrfs_bio(orig_bio)->mirror_num = bioc->mirror_num;
+	bbio->mirror_num = bioc->mirror_num;
 	orig_bio->bi_private = bioc->private;
 	orig_bio->bi_end_io = bioc->end_io;
 
@@ -6637,7 +6653,14 @@ static inline void btrfs_end_bioc(struct btrfs_io_context *bioc)
 		orig_bio->bi_status = BLK_STS_IOERR;
 	else
 		orig_bio->bi_status = BLK_STS_OK;
-	bio_endio(orig_bio);
+
+	if (btrfs_op(orig_bio) == BTRFS_MAP_READ && async) {
+		INIT_WORK(&bbio->end_io_work, btrfs_end_bio_work);
+		queue_work(btrfs_end_io_wq(bioc), &bbio->end_io_work);
+	} else {
+		bio_endio(orig_bio);
+	}
+
 	btrfs_put_bioc(bioc);
 }
 
@@ -6669,7 +6692,7 @@ static void btrfs_end_bio(struct bio *bio)
 
 	btrfs_bio_counter_dec(bioc->fs_info);
 	if (atomic_dec_and_test(&bioc->stripes_pending))
-		btrfs_end_bioc(bioc);
+		btrfs_end_bioc(bioc, true);
 }
 
 static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio,
@@ -6767,7 +6790,7 @@ blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
 			atomic_inc(&bioc->error);
 			if (atomic_dec_and_test(&bioc->stripes_pending))
-				btrfs_end_bioc(bioc);
+				btrfs_end_bioc(bioc, false);
 			continue;
 		}
 
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
index 12b2af9260e92..28e28b7c48649 100644
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -371,6 +371,9 @@ struct btrfs_bio {
 	u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
 	struct bvec_iter iter;
 
+	/* for read end I/O handling */
+	struct work_struct end_io_work;
+
 	/*
 	 * This member must come last, bio_alloc_bioset will allocate enough
 	 * bytes for entire btrfs_bio but relies on bio being last.
-- 
2.30.2

Re: [PATCH 08/10] btrfs: remove btrfs_end_io_wq

[Qu Wenruo]

On 2022/5/4 20:25, Christoph Hellwig wrote:
> All reads bio that go through btrfs_map_bio need to be completed in
> user context.  And read I/Os are the most common and timing critical
> in almost any file system workloads.

I did a quick search on other fses, it looks like it's a common pratice
to queue the work into the workqueue when the last one finished, other
than making everything happen in workqueue.
(One example, ext4 uses the same pattern for decrypt/verify work).

So I'm fine with the idea.
This specially speeds up most endios where they are just doing some
small atomic operations, which don't really need to happen in workqueue
context.

Reviewed-by: Qu Wenruo <wqu@suse.com>

Thanks,
Qu
>
> Embedd a work_struct into struct btrfs_bio and use it to complete all
> read bios submitted through btrfs_map, using the REQ_META flag to decide
> which workqueue they are placed on.
>
> This removes the need for a separate 128 byte allocation (typically
> rounded up to 192 bytes by slab) for all reads with a size increase
> of 24 bytes for struct btrfs_bio.  Future patches will reorgnize
> struct btrfs_bio to make use of this extra space for writes as well.
>
> (all sizes are based a on typical 64-bit non-debug build)
>
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>   fs/btrfs/compression.c |   4 --
>   fs/btrfs/ctree.h       |   4 +-
>   fs/btrfs/disk-io.c     | 120 +++--------------------------------------
>   fs/btrfs/disk-io.h     |  10 ----
>   fs/btrfs/inode.c       |  24 +--------
>   fs/btrfs/super.c       |  11 +---
>   fs/btrfs/volumes.c     |  33 ++++++++++--
>   fs/btrfs/volumes.h     |   3 ++
>   8 files changed, 42 insertions(+), 167 deletions(-)
>
> diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c
> index 9d5986a30a4a2..c73ee8ae070d7 100644
> --- a/fs/btrfs/compression.c
> +++ b/fs/btrfs/compression.c
> @@ -938,10 +938,6 @@ void btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
>   			sums += fs_info->csum_size * nr_sectors;
>
>   			ASSERT(comp_bio->bi_iter.bi_size);
> -			ret = btrfs_bio_wq_end_io(fs_info, comp_bio,
> -						  BTRFS_WQ_ENDIO_DATA);
> -			if (ret)
> -				goto finish_cb;
>   			ret = btrfs_map_bio(fs_info, comp_bio, mirror_num);
>   			if (ret)
>   				goto finish_cb;
> diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
> index 6cf699959286d..c839c3e77c21e 100644
> --- a/fs/btrfs/ctree.h
> +++ b/fs/btrfs/ctree.h
> @@ -851,8 +851,8 @@ struct btrfs_fs_info {
>   	struct btrfs_workqueue *hipri_workers;
>   	struct btrfs_workqueue *delalloc_workers;
>   	struct btrfs_workqueue *flush_workers;
> -	struct btrfs_workqueue *endio_workers;
> -	struct btrfs_workqueue *endio_meta_workers;
> +	struct workqueue_struct *endio_workers;
> +	struct workqueue_struct *endio_meta_workers;
>   	struct workqueue_struct *endio_raid56_workers;
>   	struct workqueue_struct *rmw_workers;
>   	struct workqueue_struct *compressed_write_workers;
> diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
> index 8fed801423504..1d9330443c716 100644
> --- a/fs/btrfs/disk-io.c
> +++ b/fs/btrfs/disk-io.c
> @@ -50,7 +50,6 @@
>   				 BTRFS_SUPER_FLAG_METADUMP |\
>   				 BTRFS_SUPER_FLAG_METADUMP_V2)
>
> -static void end_workqueue_fn(struct btrfs_work *work);
>   static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
>   static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
>   				      struct btrfs_fs_info *fs_info);
> @@ -63,40 +62,6 @@ static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
>   static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
>   static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
>
> -/*
> - * btrfs_end_io_wq structs are used to do processing in task context when an IO
> - * is complete.  This is used during reads to verify checksums, and it is used
> - * by writes to insert metadata for new file extents after IO is complete.
> - */
> -struct btrfs_end_io_wq {
> -	struct bio *bio;
> -	bio_end_io_t *end_io;
> -	void *private;
> -	struct btrfs_fs_info *info;
> -	blk_status_t status;
> -	enum btrfs_wq_endio_type metadata;
> -	struct btrfs_work work;
> -};
> -
> -static struct kmem_cache *btrfs_end_io_wq_cache;
> -
> -int __init btrfs_end_io_wq_init(void)
> -{
> -	btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
> -					sizeof(struct btrfs_end_io_wq),
> -					0,
> -					SLAB_MEM_SPREAD,
> -					NULL);
> -	if (!btrfs_end_io_wq_cache)
> -		return -ENOMEM;
> -	return 0;
> -}
> -
> -void __cold btrfs_end_io_wq_exit(void)
> -{
> -	kmem_cache_destroy(btrfs_end_io_wq_cache);
> -}
> -
>   static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
>   {
>   	if (fs_info->csum_shash)
> @@ -739,48 +704,6 @@ int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
>   	return ret;
>   }
>
> -static void end_workqueue_bio(struct bio *bio)
> -{
> -	struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
> -	struct btrfs_fs_info *fs_info;
> -	struct btrfs_workqueue *wq;
> -
> -	fs_info = end_io_wq->info;
> -	end_io_wq->status = bio->bi_status;
> -
> -	if (end_io_wq->metadata)
> -		wq = fs_info->endio_meta_workers;
> -	else
> -		wq = fs_info->endio_workers;
> -
> -	btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
> -	btrfs_queue_work(wq, &end_io_wq->work);
> -}
> -
> -blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
> -			enum btrfs_wq_endio_type metadata)
> -{
> -	struct btrfs_end_io_wq *end_io_wq;
> -
> -	if (WARN_ON_ONCE(btrfs_op(bio) != BTRFS_MAP_WRITE))
> -		return BLK_STS_IOERR;
> -
> -	end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
> -	if (!end_io_wq)
> -		return BLK_STS_RESOURCE;
> -
> -	end_io_wq->private = bio->bi_private;
> -	end_io_wq->end_io = bio->bi_end_io;
> -	end_io_wq->info = info;
> -	end_io_wq->status = 0;
> -	end_io_wq->bio = bio;
> -	end_io_wq->metadata = metadata;
> -
> -	bio->bi_private = end_io_wq;
> -	bio->bi_end_io = end_workqueue_bio;
> -	return 0;
> -}
> -
>   static void run_one_async_start(struct btrfs_work *work)
>   {
>   	struct async_submit_bio *async;
> @@ -916,14 +839,7 @@ void btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, int mirror_
>   	bio->bi_opf |= REQ_META;
>
>   	if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
> -		/*
> -		 * called for a read, do the setup so that checksum validation
> -		 * can happen in the async kernel threads
> -		 */
> -		ret = btrfs_bio_wq_end_io(fs_info, bio,
> -					  BTRFS_WQ_ENDIO_METADATA);
> -		if (!ret)
> -			ret = btrfs_map_bio(fs_info, bio, mirror_num);
> +		ret = btrfs_map_bio(fs_info, bio, mirror_num);
>   	} else if (!should_async_write(fs_info, BTRFS_I(inode))) {
>   		ret = btree_csum_one_bio(bio);
>   		if (!ret)
> @@ -1939,25 +1855,6 @@ struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
>   	return root;
>   }
>
> -/*
> - * called by the kthread helper functions to finally call the bio end_io
> - * functions.  This is where read checksum verification actually happens
> - */
> -static void end_workqueue_fn(struct btrfs_work *work)
> -{
> -	struct bio *bio;
> -	struct btrfs_end_io_wq *end_io_wq;
> -
> -	end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
> -	bio = end_io_wq->bio;
> -
> -	bio->bi_status = end_io_wq->status;
> -	bio->bi_private = end_io_wq->private;
> -	bio->bi_end_io = end_io_wq->end_io;
> -	bio_endio(bio);
> -	kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
> -}
> -
>   static int cleaner_kthread(void *arg)
>   {
>   	struct btrfs_fs_info *fs_info = arg;
> @@ -2264,7 +2161,8 @@ static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
>   	btrfs_destroy_workqueue(fs_info->delalloc_workers);
>   	btrfs_destroy_workqueue(fs_info->hipri_workers);
>   	btrfs_destroy_workqueue(fs_info->workers);
> -	btrfs_destroy_workqueue(fs_info->endio_workers);
> +	if (fs_info->endio_workers)
> +		destroy_workqueue(fs_info->endio_workers);
>   	if (fs_info->endio_raid56_workers)
>   		destroy_workqueue(fs_info->endio_raid56_workers);
>   	if (fs_info->rmw_workers)
> @@ -2284,7 +2182,8 @@ static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
>   	 * the queues used for metadata I/O, since tasks from those other work
>   	 * queues can do metadata I/O operations.
>   	 */
> -	btrfs_destroy_workqueue(fs_info->endio_meta_workers);
> +	if (fs_info->endio_meta_workers)
> +		destroy_workqueue(fs_info->endio_meta_workers);
>   }
>
>   static void free_root_extent_buffers(struct btrfs_root *root)
> @@ -2457,15 +2356,10 @@ static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
>   	fs_info->fixup_workers =
>   		btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
>
> -	/*
> -	 * endios are largely parallel and should have a very
> -	 * low idle thresh
> -	 */
>   	fs_info->endio_workers =
> -		btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
> +		alloc_workqueue("btrfs-endio", flags, max_active);
>   	fs_info->endio_meta_workers =
> -		btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
> -				      max_active, 4);
> +		alloc_workqueue("btrfs-endio-meta", flags, max_active);
>   	fs_info->endio_raid56_workers =
>   		alloc_workqueue("btrfs-endio-raid56", flags, max_active);
>   	fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
> diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h
> index 809ef065f1666..05e779a41a997 100644
> --- a/fs/btrfs/disk-io.h
> +++ b/fs/btrfs/disk-io.h
> @@ -17,12 +17,6 @@
>    */
>   #define BTRFS_BDEV_BLOCKSIZE	(4096)
>
> -enum btrfs_wq_endio_type {
> -	BTRFS_WQ_ENDIO_DATA,
> -	BTRFS_WQ_ENDIO_METADATA,
> -	BTRFS_WQ_ENDIO_FREE_SPACE,
> -};
> -
>   static inline u64 btrfs_sb_offset(int mirror)
>   {
>   	u64 start = SZ_16K;
> @@ -120,8 +114,6 @@ int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
>   			  int atomic);
>   int btrfs_read_extent_buffer(struct extent_buffer *buf, u64 parent_transid,
>   			     int level, struct btrfs_key *first_key);
> -blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
> -			enum btrfs_wq_endio_type metadata);
>   blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
>   				 int mirror_num, u64 dio_file_offset,
>   				 extent_submit_bio_start_t *submit_bio_start);
> @@ -144,8 +136,6 @@ int btree_lock_page_hook(struct page *page, void *data,
>   int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags);
>   int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid);
>   int btrfs_init_root_free_objectid(struct btrfs_root *root);
> -int __init btrfs_end_io_wq_init(void);
> -void __cold btrfs_end_io_wq_exit(void);
>
>   #ifdef CONFIG_DEBUG_LOCK_ALLOC
>   void btrfs_set_buffer_lockdep_class(u64 objectid,
> diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
> index 5499b39cab61b..8195fd70633ab 100644
> --- a/fs/btrfs/inode.c
> +++ b/fs/btrfs/inode.c
> @@ -2616,12 +2616,6 @@ void btrfs_submit_data_read_bio(struct inode *inode, struct bio *bio,
>   		return;
>   	}
>
> -	ret = btrfs_bio_wq_end_io(fs_info, bio,
> -			btrfs_is_free_space_inode(BTRFS_I(inode)) ?
> -			BTRFS_WQ_ENDIO_FREE_SPACE : BTRFS_WQ_ENDIO_DATA);
> -	if (ret)
> -		goto out;
> -
>   	/*
>   	 * Lookup bio sums does extra checks around whether we need to csum or
>   	 * not, which is why we ignore skip_sum here.
> @@ -7834,9 +7828,6 @@ static void submit_dio_repair_bio(struct inode *inode, struct bio *bio,
>
>   	BUG_ON(bio_op(bio) == REQ_OP_WRITE);
>
> -	if (btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA))
> -		return;
> -
>   	refcount_inc(&dip->refs);
>   	if (btrfs_map_bio(fs_info, bio, mirror_num))
>   		refcount_dec(&dip->refs);
> @@ -7937,19 +7928,12 @@ static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
>   {
>   	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
>   	struct btrfs_dio_private *dip = bio->bi_private;
> -	bool write = btrfs_op(bio) == BTRFS_MAP_WRITE;
>   	blk_status_t ret;
>
> -	if (!write) {
> -		ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
> -		if (ret)
> -			return ret;
> -	}
> -
>   	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
>   		goto map;
>
> -	if (write) {
> +	if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
>   		/* check btrfs_submit_data_write_bio() for async submit rules */
>   		if (async_submit && !atomic_read(&BTRFS_I(inode)->sync_writers))
>   			return btrfs_wq_submit_bio(inode, bio, 0, file_offset,
> @@ -10298,12 +10282,6 @@ static blk_status_t submit_encoded_read_bio(struct btrfs_inode *inode,
>   			return ret;
>   	}
>
> -	ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
> -	if (ret) {
> -		btrfs_bio_free_csum(bbio);
> -		return ret;
> -	}
> -
>   	atomic_inc(&priv->pending);
>   	ret = btrfs_map_bio(fs_info, bio, mirror_num);
>   	if (ret) {
> diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
> index 9c683c466d585..64eb8aeed156f 100644
> --- a/fs/btrfs/super.c
> +++ b/fs/btrfs/super.c
> @@ -1906,8 +1906,6 @@ static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
>   	btrfs_workqueue_set_max(fs_info->hipri_workers, new_pool_size);
>   	btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
>   	btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
> -	btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
> -	btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
>   	btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
>   	btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
>   	btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
> @@ -2668,13 +2666,9 @@ static int __init init_btrfs_fs(void)
>   	if (err)
>   		goto free_delayed_ref;
>
> -	err = btrfs_end_io_wq_init();
> -	if (err)
> -		goto free_prelim_ref;
> -
>   	err = btrfs_interface_init();
>   	if (err)
> -		goto free_end_io_wq;
> +		goto free_prelim_ref;
>
>   	btrfs_print_mod_info();
>
> @@ -2690,8 +2684,6 @@ static int __init init_btrfs_fs(void)
>
>   unregister_ioctl:
>   	btrfs_interface_exit();
> -free_end_io_wq:
> -	btrfs_end_io_wq_exit();
>   free_prelim_ref:
>   	btrfs_prelim_ref_exit();
>   free_delayed_ref:
> @@ -2729,7 +2721,6 @@ static void __exit exit_btrfs_fs(void)
>   	extent_state_cache_exit();
>   	extent_io_exit();
>   	btrfs_interface_exit();
> -	btrfs_end_io_wq_exit();
>   	unregister_filesystem(&btrfs_fs_type);
>   	btrfs_exit_sysfs();
>   	btrfs_cleanup_fs_uuids();
> diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
> index aeacb87457687..5f18e9105fe08 100644
> --- a/fs/btrfs/volumes.c
> +++ b/fs/btrfs/volumes.c
> @@ -6621,11 +6621,27 @@ int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
>   	return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 0, 1);
>   }
>
> -static inline void btrfs_end_bioc(struct btrfs_io_context *bioc)
> +static struct workqueue_struct *btrfs_end_io_wq(struct btrfs_io_context *bioc)
> +{
> +	if (bioc->orig_bio->bi_opf & REQ_META)
> +		return bioc->fs_info->endio_meta_workers;
> +	return bioc->fs_info->endio_workers;
> +}
> +
> +static void btrfs_end_bio_work(struct work_struct *work)
> +{
> +	struct btrfs_bio *bbio =
> +		container_of(work, struct btrfs_bio, end_io_work);
> +
> +	bio_endio(&bbio->bio);
> +}
> +
> +static void btrfs_end_bioc(struct btrfs_io_context *bioc, bool async)
>   {
>   	struct bio *orig_bio = bioc->orig_bio;
> +	struct btrfs_bio *bbio = btrfs_bio(orig_bio);
>
> -	btrfs_bio(orig_bio)->mirror_num = bioc->mirror_num;
> +	bbio->mirror_num = bioc->mirror_num;
>   	orig_bio->bi_private = bioc->private;
>   	orig_bio->bi_end_io = bioc->end_io;
>
> @@ -6637,7 +6653,14 @@ static inline void btrfs_end_bioc(struct btrfs_io_context *bioc)
>   		orig_bio->bi_status = BLK_STS_IOERR;
>   	else
>   		orig_bio->bi_status = BLK_STS_OK;
> -	bio_endio(orig_bio);
> +
> +	if (btrfs_op(orig_bio) == BTRFS_MAP_READ && async) {
> +		INIT_WORK(&bbio->end_io_work, btrfs_end_bio_work);
> +		queue_work(btrfs_end_io_wq(bioc), &bbio->end_io_work);
> +	} else {
> +		bio_endio(orig_bio);
> +	}
> +
>   	btrfs_put_bioc(bioc);
>   }
>
> @@ -6669,7 +6692,7 @@ static void btrfs_end_bio(struct bio *bio)
>
>   	btrfs_bio_counter_dec(bioc->fs_info);
>   	if (atomic_dec_and_test(&bioc->stripes_pending))
> -		btrfs_end_bioc(bioc);
> +		btrfs_end_bioc(bioc, true);
>   }
>
>   static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio,
> @@ -6767,7 +6790,7 @@ blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
>   		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
>   			atomic_inc(&bioc->error);
>   			if (atomic_dec_and_test(&bioc->stripes_pending))
> -				btrfs_end_bioc(bioc);
> +				btrfs_end_bioc(bioc, false);
>   			continue;
>   		}
>
> diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
> index 12b2af9260e92..28e28b7c48649 100644
> --- a/fs/btrfs/volumes.h
> +++ b/fs/btrfs/volumes.h
> @@ -371,6 +371,9 @@ struct btrfs_bio {
>   	u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE];
>   	struct bvec_iter iter;
>
> +	/* for read end I/O handling */
> +	struct work_struct end_io_work;
> +
>   	/*
>   	 * This member must come last, bio_alloc_bioset will allocate enough
>   	 * bytes for entire btrfs_bio but relies on bio being last.

Re: [PATCH 08/10] btrfs: remove btrfs_end_io_wq

[Christoph Hellwig]

On Thu, May 05, 2022 at 10:34:34AM +0800, Qu Wenruo wrote:
> I did a quick search on other fses, it looks like it's a common pratice
> to queue the work into the workqueue when the last one finished, other
> than making everything happen in workqueue.

Yes.  It is a lot more efficient as queing up lots of items tends to
cause quite some overhead, also due to the dynamic threadpool of the
workqueue.  XFS goes even futher and adds many bios to list at
completion time to be able to batch them up for the workqueue based
completions.

[PATCH 09/10] btrfs: refactor btrfs_map_bio

[Christoph Hellwig]

Move all per-stripe handling into submit_stripe_bio and use a label to
cleanup instead of duplicating the logic.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/btrfs/volumes.c | 77 +++++++++++++++++++++-------------------------
 1 file changed, 35 insertions(+), 42 deletions(-)

diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 5f18e9105fe08..28ac2bfb5d296 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -6695,10 +6695,30 @@ static void btrfs_end_bio(struct bio *bio)
 		btrfs_end_bioc(bioc, true);
 }
 
-static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio,
-			      u64 physical, struct btrfs_device *dev)
+static void submit_stripe_bio(struct btrfs_io_context *bioc,
+		struct bio *orig_bio, int dev_nr, bool clone)
 {
 	struct btrfs_fs_info *fs_info = bioc->fs_info;
+	struct btrfs_device *dev = bioc->stripes[dev_nr].dev;
+	u64 physical = bioc->stripes[dev_nr].physical;
+	struct bio *bio;
+
+	if (!dev || !dev->bdev ||
+	    test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
+	    (btrfs_op(orig_bio) == BTRFS_MAP_WRITE &&
+	     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
+		atomic_inc(&bioc->error);
+		if (atomic_dec_and_test(&bioc->stripes_pending))
+			btrfs_end_bioc(bioc, false);
+		return;
+	}
+
+	if (clone) {
+		bio = btrfs_bio_clone(dev->bdev, orig_bio);
+	} else {
+		bio = orig_bio;
+		bio_set_dev(bio, dev->bdev);
+	}
 
 	bio->bi_private = bioc;
 	btrfs_bio(bio)->device = dev;
@@ -6733,32 +6753,25 @@ static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio,
 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 			   int mirror_num)
 {
-	struct btrfs_device *dev;
-	struct bio *first_bio = bio;
 	u64 logical = bio->bi_iter.bi_sector << 9;
-	u64 length = 0;
-	u64 map_length;
+	u64 length = bio->bi_iter.bi_size;
+	u64 map_length = length;
 	int ret;
 	int dev_nr;
 	int total_devs;
 	struct btrfs_io_context *bioc = NULL;
 
-	length = bio->bi_iter.bi_size;
-	map_length = length;
-
 	btrfs_bio_counter_inc_blocked(fs_info);
 	ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
 				&map_length, &bioc, mirror_num, 1);
-	if (ret) {
-		btrfs_bio_counter_dec(fs_info);
-		return errno_to_blk_status(ret);
-	}
+	if (ret)
+		goto out_dec;
 
 	total_devs = bioc->num_stripes;
-	bioc->orig_bio = first_bio;
-	bioc->private = first_bio->bi_private;
-	bioc->end_io = first_bio->bi_end_io;
-	atomic_set(&bioc->stripes_pending, bioc->num_stripes);
+	bioc->orig_bio = bio;
+	bioc->private = bio->bi_private;
+	bioc->end_io = bio->bi_end_io;
+	atomic_set(&bioc->stripes_pending, total_devs);
 
 	if ((bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
 	    ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) {
@@ -6770,9 +6783,7 @@ blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 			ret = raid56_parity_recover(bio, bioc, map_length,
 						    mirror_num, 1);
 		}
-
-		btrfs_bio_counter_dec(fs_info);
-		return errno_to_blk_status(ret);
+		goto out_dec;
 	}
 
 	if (map_length < length) {
@@ -6782,29 +6793,11 @@ blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 		BUG();
 	}
 
-	for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
-		dev = bioc->stripes[dev_nr].dev;
-		if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
-						   &dev->dev_state) ||
-		    (btrfs_op(first_bio) == BTRFS_MAP_WRITE &&
-		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
-			atomic_inc(&bioc->error);
-			if (atomic_dec_and_test(&bioc->stripes_pending))
-				btrfs_end_bioc(bioc, false);
-			continue;
-		}
-
-		if (dev_nr < total_devs - 1) {
-			bio = btrfs_bio_clone(dev->bdev, first_bio);
-		} else {
-			bio = first_bio;
-			bio_set_dev(bio, dev->bdev);
-		}
-
-		submit_stripe_bio(bioc, bio, bioc->stripes[dev_nr].physical, dev);
-	}
+	for (dev_nr = 0; dev_nr < total_devs; dev_nr++)
+		submit_stripe_bio(bioc, bio, dev_nr, dev_nr < total_devs - 1);
+out_dec:
 	btrfs_bio_counter_dec(fs_info);
-	return BLK_STS_OK;
+	return errno_to_blk_status(ret);
 }
 
 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
-- 
2.30.2

Re: [PATCH 09/10] btrfs: refactor btrfs_map_bio

[Qu Wenruo]

On 2022/5/4 20:25, Christoph Hellwig wrote:
> Move all per-stripe handling into submit_stripe_bio and use a label to
> cleanup instead of duplicating the logic.
>
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Reviewed-by: Qu Wenruo <wqu@suse.com>

Thanks,
Qu
> ---
>   fs/btrfs/volumes.c | 77 +++++++++++++++++++++-------------------------
>   1 file changed, 35 insertions(+), 42 deletions(-)
>
> diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
> index 5f18e9105fe08..28ac2bfb5d296 100644
> --- a/fs/btrfs/volumes.c
> +++ b/fs/btrfs/volumes.c
> @@ -6695,10 +6695,30 @@ static void btrfs_end_bio(struct bio *bio)
>   		btrfs_end_bioc(bioc, true);
>   }
>
> -static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio,
> -			      u64 physical, struct btrfs_device *dev)
> +static void submit_stripe_bio(struct btrfs_io_context *bioc,
> +		struct bio *orig_bio, int dev_nr, bool clone)
>   {
>   	struct btrfs_fs_info *fs_info = bioc->fs_info;
> +	struct btrfs_device *dev = bioc->stripes[dev_nr].dev;
> +	u64 physical = bioc->stripes[dev_nr].physical;
> +	struct bio *bio;
> +
> +	if (!dev || !dev->bdev ||
> +	    test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
> +	    (btrfs_op(orig_bio) == BTRFS_MAP_WRITE &&
> +	     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
> +		atomic_inc(&bioc->error);
> +		if (atomic_dec_and_test(&bioc->stripes_pending))
> +			btrfs_end_bioc(bioc, false);
> +		return;
> +	}
> +
> +	if (clone) {
> +		bio = btrfs_bio_clone(dev->bdev, orig_bio);
> +	} else {
> +		bio = orig_bio;
> +		bio_set_dev(bio, dev->bdev);
> +	}
>
>   	bio->bi_private = bioc;
>   	btrfs_bio(bio)->device = dev;
> @@ -6733,32 +6753,25 @@ static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio,
>   blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
>   			   int mirror_num)
>   {
> -	struct btrfs_device *dev;
> -	struct bio *first_bio = bio;
>   	u64 logical = bio->bi_iter.bi_sector << 9;
> -	u64 length = 0;
> -	u64 map_length;
> +	u64 length = bio->bi_iter.bi_size;
> +	u64 map_length = length;
>   	int ret;
>   	int dev_nr;
>   	int total_devs;
>   	struct btrfs_io_context *bioc = NULL;
>
> -	length = bio->bi_iter.bi_size;
> -	map_length = length;
> -
>   	btrfs_bio_counter_inc_blocked(fs_info);
>   	ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
>   				&map_length, &bioc, mirror_num, 1);
> -	if (ret) {
> -		btrfs_bio_counter_dec(fs_info);
> -		return errno_to_blk_status(ret);
> -	}
> +	if (ret)
> +		goto out_dec;
>
>   	total_devs = bioc->num_stripes;
> -	bioc->orig_bio = first_bio;
> -	bioc->private = first_bio->bi_private;
> -	bioc->end_io = first_bio->bi_end_io;
> -	atomic_set(&bioc->stripes_pending, bioc->num_stripes);
> +	bioc->orig_bio = bio;
> +	bioc->private = bio->bi_private;
> +	bioc->end_io = bio->bi_end_io;
> +	atomic_set(&bioc->stripes_pending, total_devs);
>
>   	if ((bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
>   	    ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) {
> @@ -6770,9 +6783,7 @@ blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
>   			ret = raid56_parity_recover(bio, bioc, map_length,
>   						    mirror_num, 1);
>   		}
> -
> -		btrfs_bio_counter_dec(fs_info);
> -		return errno_to_blk_status(ret);
> +		goto out_dec;
>   	}
>
>   	if (map_length < length) {
> @@ -6782,29 +6793,11 @@ blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
>   		BUG();
>   	}
>
> -	for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
> -		dev = bioc->stripes[dev_nr].dev;
> -		if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
> -						   &dev->dev_state) ||
> -		    (btrfs_op(first_bio) == BTRFS_MAP_WRITE &&
> -		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
> -			atomic_inc(&bioc->error);
> -			if (atomic_dec_and_test(&bioc->stripes_pending))
> -				btrfs_end_bioc(bioc, false);
> -			continue;
> -		}
> -
> -		if (dev_nr < total_devs - 1) {
> -			bio = btrfs_bio_clone(dev->bdev, first_bio);
> -		} else {
> -			bio = first_bio;
> -			bio_set_dev(bio, dev->bdev);
> -		}
> -
> -		submit_stripe_bio(bioc, bio, bioc->stripes[dev_nr].physical, dev);
> -	}
> +	for (dev_nr = 0; dev_nr < total_devs; dev_nr++)
> +		submit_stripe_bio(bioc, bio, dev_nr, dev_nr < total_devs - 1);
> +out_dec:
>   	btrfs_bio_counter_dec(fs_info);
> -	return BLK_STS_OK;
> +	return errno_to_blk_status(ret);
>   }
>
>   static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,

[PATCH 10/10] btrfs: do not allocate a btrfs_bio for low-level bios

[Christoph Hellwig]

The bios submitted from btrfs_map_bio don't really interact with the
rest of btrfs and the only btrfs_bio member actually used in the
low-level bios is the pointer to the btrfs_io_contex used for endio
handler.

Use a union in struct btrfs_io_stripe that allows the endio handler to
find the btrfs_io_context and remove the spurious ->device assignment
so that a plain fs_bio_set bio can be used for the low-level bios
allocated inside btrfs_map_bio.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/btrfs/extent_io.c | 13 -------------
 fs/btrfs/extent_io.h |  1 -
 fs/btrfs/volumes.c   | 20 ++++++++++----------
 fs/btrfs/volumes.h   |  5 ++++-
 4 files changed, 14 insertions(+), 25 deletions(-)

diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 93191771e5bf1..7efbc964585d3 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -3210,19 +3210,6 @@ struct bio *btrfs_bio_alloc(unsigned int nr_iovecs)
 	return bio;
 }
 
-struct bio *btrfs_bio_clone(struct block_device *bdev, struct bio *bio)
-{
-	struct btrfs_bio *bbio;
-	struct bio *new;
-
-	/* Bio allocation backed by a bioset does not fail */
-	new = bio_alloc_clone(bdev, bio, GFP_NOFS, &btrfs_bioset);
-	bbio = btrfs_bio(new);
-	btrfs_bio_init(bbio);
-	bbio->iter = bio->bi_iter;
-	return new;
-}
-
 struct bio *btrfs_bio_clone_partial(struct bio *orig, u64 offset, u64 size)
 {
 	struct bio *bio;
diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index 17674b7e699c6..ce69a4732f718 100644
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -248,7 +248,6 @@ void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
 
 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array);
 struct bio *btrfs_bio_alloc(unsigned int nr_iovecs);
-struct bio *btrfs_bio_clone(struct block_device *bdev, struct bio *bio);
 struct bio *btrfs_bio_clone_partial(struct bio *orig, u64 offset, u64 size);
 
 void end_extent_writepage(struct page *page, int err, u64 start, u64 end);
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 28ac2bfb5d296..8577893ccb4b9 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -6666,23 +6666,21 @@ static void btrfs_end_bioc(struct btrfs_io_context *bioc, bool async)
 
 static void btrfs_end_bio(struct bio *bio)
 {
-	struct btrfs_io_context *bioc = bio->bi_private;
+	struct btrfs_io_stripe *stripe = bio->bi_private;
+	struct btrfs_io_context *bioc = stripe->bioc;
 
 	if (bio->bi_status) {
 		atomic_inc(&bioc->error);
 		if (bio->bi_status == BLK_STS_IOERR ||
 		    bio->bi_status == BLK_STS_TARGET) {
-			struct btrfs_device *dev = btrfs_bio(bio)->device;
-
-			ASSERT(dev->bdev);
 			if (btrfs_op(bio) == BTRFS_MAP_WRITE)
-				btrfs_dev_stat_inc_and_print(dev,
+				btrfs_dev_stat_inc_and_print(stripe->dev,
 						BTRFS_DEV_STAT_WRITE_ERRS);
 			else if (!(bio->bi_opf & REQ_RAHEAD))
-				btrfs_dev_stat_inc_and_print(dev,
+				btrfs_dev_stat_inc_and_print(stripe->dev,
 						BTRFS_DEV_STAT_READ_ERRS);
 			if (bio->bi_opf & REQ_PREFLUSH)
-				btrfs_dev_stat_inc_and_print(dev,
+				btrfs_dev_stat_inc_and_print(stripe->dev,
 						BTRFS_DEV_STAT_FLUSH_ERRS);
 		}
 	}
@@ -6714,14 +6712,16 @@ static void submit_stripe_bio(struct btrfs_io_context *bioc,
 	}
 
 	if (clone) {
-		bio = btrfs_bio_clone(dev->bdev, orig_bio);
+		bio = bio_alloc_clone(dev->bdev, orig_bio, GFP_NOFS,
+				      &fs_bio_set);
 	} else {
 		bio = orig_bio;
 		bio_set_dev(bio, dev->bdev);
+		btrfs_bio(bio)->device = dev;
 	}
 
-	bio->bi_private = bioc;
-	btrfs_bio(bio)->device = dev;
+	bioc->stripes[dev_nr].bioc = bioc;
+	bio->bi_private = &bioc->stripes[dev_nr];
 	bio->bi_end_io = btrfs_end_bio;
 	bio->bi_iter.bi_sector = physical >> 9;
 	/*
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
index 28e28b7c48649..825e44c82f2b0 100644
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -396,7 +396,10 @@ static inline void btrfs_bio_free_csum(struct btrfs_bio *bbio)
 
 struct btrfs_io_stripe {
 	struct btrfs_device *dev;
-	u64 physical;
+	union {
+		u64 physical;			/* block mapping */
+		struct btrfs_io_context *bioc;	/* for the endio handler */
+	};
 	u64 length; /* only used for discard mappings */
 };
 
-- 
2.30.2

Re: [PATCH 10/10] btrfs: do not allocate a btrfs_bio for low-level bios

[Qu Wenruo]

On 2022/5/4 20:25, Christoph Hellwig wrote:
> The bios submitted from btrfs_map_bio don't really interact with the
> rest of btrfs and the only btrfs_bio member actually used in the
> low-level bios is the pointer to the btrfs_io_contex used for endio
> handler.
>
> Use a union in struct btrfs_io_stripe that allows the endio handler to
> find the btrfs_io_context and remove the spurious ->device assignment
> so that a plain fs_bio_set bio can be used for the low-level bios
> allocated inside btrfs_map_bio.
>
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Although such union is not that reader friendly, it should be safe enough.

Reviewed-by: Qu Wenruo <wqu@suse.com>


Although in the future, I may add a new bioset for such low-level bios
for other purposes (like read-repair at per-stripe level), thus may
allow a more readable solution.
But that would be another story in the future anyway.

Thanks,
Qu
> ---
>   fs/btrfs/extent_io.c | 13 -------------
>   fs/btrfs/extent_io.h |  1 -
>   fs/btrfs/volumes.c   | 20 ++++++++++----------
>   fs/btrfs/volumes.h   |  5 ++++-
>   4 files changed, 14 insertions(+), 25 deletions(-)
>
> diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
> index 93191771e5bf1..7efbc964585d3 100644
> --- a/fs/btrfs/extent_io.c
> +++ b/fs/btrfs/extent_io.c
> @@ -3210,19 +3210,6 @@ struct bio *btrfs_bio_alloc(unsigned int nr_iovecs)
>   	return bio;
>   }
>
> -struct bio *btrfs_bio_clone(struct block_device *bdev, struct bio *bio)
> -{
> -	struct btrfs_bio *bbio;
> -	struct bio *new;
> -
> -	/* Bio allocation backed by a bioset does not fail */
> -	new = bio_alloc_clone(bdev, bio, GFP_NOFS, &btrfs_bioset);
> -	bbio = btrfs_bio(new);
> -	btrfs_bio_init(bbio);
> -	bbio->iter = bio->bi_iter;
> -	return new;
> -}
> -
>   struct bio *btrfs_bio_clone_partial(struct bio *orig, u64 offset, u64 size)
>   {
>   	struct bio *bio;
> diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
> index 17674b7e699c6..ce69a4732f718 100644
> --- a/fs/btrfs/extent_io.h
> +++ b/fs/btrfs/extent_io.h
> @@ -248,7 +248,6 @@ void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
>
>   int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array);
>   struct bio *btrfs_bio_alloc(unsigned int nr_iovecs);
> -struct bio *btrfs_bio_clone(struct block_device *bdev, struct bio *bio);
>   struct bio *btrfs_bio_clone_partial(struct bio *orig, u64 offset, u64 size);
>
>   void end_extent_writepage(struct page *page, int err, u64 start, u64 end);
> diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
> index 28ac2bfb5d296..8577893ccb4b9 100644
> --- a/fs/btrfs/volumes.c
> +++ b/fs/btrfs/volumes.c
> @@ -6666,23 +6666,21 @@ static void btrfs_end_bioc(struct btrfs_io_context *bioc, bool async)
>
>   static void btrfs_end_bio(struct bio *bio)
>   {
> -	struct btrfs_io_context *bioc = bio->bi_private;
> +	struct btrfs_io_stripe *stripe = bio->bi_private;
> +	struct btrfs_io_context *bioc = stripe->bioc;
>
>   	if (bio->bi_status) {
>   		atomic_inc(&bioc->error);
>   		if (bio->bi_status == BLK_STS_IOERR ||
>   		    bio->bi_status == BLK_STS_TARGET) {
> -			struct btrfs_device *dev = btrfs_bio(bio)->device;
> -
> -			ASSERT(dev->bdev);
>   			if (btrfs_op(bio) == BTRFS_MAP_WRITE)
> -				btrfs_dev_stat_inc_and_print(dev,
> +				btrfs_dev_stat_inc_and_print(stripe->dev,
>   						BTRFS_DEV_STAT_WRITE_ERRS);
>   			else if (!(bio->bi_opf & REQ_RAHEAD))
> -				btrfs_dev_stat_inc_and_print(dev,
> +				btrfs_dev_stat_inc_and_print(stripe->dev,
>   						BTRFS_DEV_STAT_READ_ERRS);
>   			if (bio->bi_opf & REQ_PREFLUSH)
> -				btrfs_dev_stat_inc_and_print(dev,
> +				btrfs_dev_stat_inc_and_print(stripe->dev,
>   						BTRFS_DEV_STAT_FLUSH_ERRS);
>   		}
>   	}
> @@ -6714,14 +6712,16 @@ static void submit_stripe_bio(struct btrfs_io_context *bioc,
>   	}
>
>   	if (clone) {
> -		bio = btrfs_bio_clone(dev->bdev, orig_bio);
> +		bio = bio_alloc_clone(dev->bdev, orig_bio, GFP_NOFS,
> +				      &fs_bio_set);
>   	} else {
>   		bio = orig_bio;
>   		bio_set_dev(bio, dev->bdev);
> +		btrfs_bio(bio)->device = dev;
>   	}
>
> -	bio->bi_private = bioc;
> -	btrfs_bio(bio)->device = dev;
> +	bioc->stripes[dev_nr].bioc = bioc;
> +	bio->bi_private = &bioc->stripes[dev_nr];
>   	bio->bi_end_io = btrfs_end_bio;
>   	bio->bi_iter.bi_sector = physical >> 9;
>   	/*
> diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
> index 28e28b7c48649..825e44c82f2b0 100644
> --- a/fs/btrfs/volumes.h
> +++ b/fs/btrfs/volumes.h
> @@ -396,7 +396,10 @@ static inline void btrfs_bio_free_csum(struct btrfs_bio *bbio)
>
>   struct btrfs_io_stripe {
>   	struct btrfs_device *dev;
> -	u64 physical;
> +	union {
> +		u64 physical;			/* block mapping */
> +		struct btrfs_io_context *bioc;	/* for the endio handler */
> +	};
>   	u64 length; /* only used for discard mappings */
>   };
>

Re: cleanup btrfs bio handling, part 2 v3

[Qu Wenruo]

On 2022/5/4 20:25, Christoph Hellwig wrote:
> Hi all,
>
> this series removes the need to allocate a separate object for I/O
> completions for all read and some write I/Os, and reduced the memory
> usage of the low-level bios cloned by btrfs_map_bio by using plain bios
> instead of the much larger btrfs_bio.

The most important part of this series is the patch "btrfs: remove
btrfs_end_io_wq()" I guess.

As a guy growing up mostly reading btrfs code, the btrfs_end_io_wq()
function looks straightforward to me initially, just put everything into
wq context and call it a day.

But as I started to look into how other fses handle their delayed work,
the btrfs way turns more "unique" other than straightforward.


So yes, aligning us to other fses, by only delaying into wq context when
necessary, should be a good thing.
Not reducing the memory allocation for the btrfs_end_io_wq structure,
but also making the atomic part of the endio function executed with less
delay.

But here I'm not sure if the btrfs way is a legacy from the past, or has
some hidden intention (possible related to performance?).

So do you mind to do some basic benchmark for read and write and see how
the throughput and latency changed?
(I guess we can get slightly reduced latency without really damping
everything else).

With such benchmark the series would be more convincing, and I can learn
some tricks on how to do proper performance tests.


Also mentioning how other fses handling delayed work may help guys like
me to get an understanding of the more common practice used in other fses.

Thanks,
Qu
>
> Changes since v2:
>   - rebased to the latests misc-next tree
>   - fixed an incorrectly assigned bi_end_io handler in the raid56 code
>
> Changes since v1:
>   - rebased to the latests misc-next tree
>   - don't call destroy_workqueue on NULL workqueues
>   - drop a marginal and slightly controversial cleanup to btrfs_map_bio
>
> Diffstat:
>   compression.c |   41 +++++-------
>   compression.h |    7 +-
>   ctree.h       |   14 ++--
>   disk-io.c     |  148 +++++-----------------------------------------
>   disk-io.h     |   11 ---
>   extent_io.c   |   35 ++++-------
>   extent_io.h   |    1
>   inode.c       |  162 +++++++++++++++++++--------------------------------
>   raid56.c      |  109 +++++++++++++---------------------
>   super.c       |   13 ----
>   volumes.c     |  184 +++++++++++++++++++++++++++-------------------------------
>   volumes.h     |    8 ++
>   12 files changed, 262 insertions(+), 471 deletions(-)

Re: cleanup btrfs bio handling, part 2 v3

[Christoph Hellwig]

On Thu, May 05, 2022 at 02:56:49PM +0800, Qu Wenruo wrote:
> So do you mind to do some basic benchmark for read and write and see how
> the throughput and latency changed?

I'm not actually seeing any meaningful differences at all.  I think this
will help btrfs to better behave under memory pressure, though.

Re: cleanup btrfs bio handling, part 2 v3

[Anand Jain]

On 5/5/22 20:41, Christoph Hellwig wrote:
> On Thu, May 05, 2022 at 02:56:49PM +0800, Qu Wenruo wrote:
>> So do you mind to do some basic benchmark for read and write and see how
>> the throughput and latency changed?


> I'm not actually seeing any meaningful differences at all.  I think this
> will help btrfs to better behave under memory pressure, though.
This patch got me curious a couple of days back while I was tracing a
dio read performance issue on nvme.

Yep. No meaningful difference. I am sharing the results as below. [1].

[1]
Before:
  4971MB/s 4474GB iocounts: nvme3n1 545220968 nvme0n1 547007640 
single_d2/5.18.0-rc5+_misc-next_1

After:
  4968MB/s 4471GB iocounts: nvme3n1 544207371 nvme1n1 547458037 
single_d2/5.18.0-rc5+_dio_cleanup_hch_1

  readstat /btrfs fio --eta=auto --output=$CMDLOG \
--name fiotest --directory=/btrfs --rw=randread \
--bs=4k --size=4G --ioengine=libaio --iodepth=16 --direct=1 \
--time_based=1 --runtime=900 --randrepeat=1 --gtod_reduce=1 \
--group_reporting=1 --numjobs=64

Re: cleanup btrfs bio handling, part 2 v3

[Anand Jain]

On 5/12/22 11:52, Anand Jain wrote:
> On 5/5/22 20:41, Christoph Hellwig wrote:
>> On Thu, May 05, 2022 at 02:56:49PM +0800, Qu Wenruo wrote:
>>> So do you mind to do some basic benchmark for read and write and see how
>>> the throughput and latency changed?
> 
> 
>> I'm not actually seeing any meaningful differences at all.  I think this
>> will help btrfs to better behave under memory pressure, though.
> This patch got me curious a couple of days back while I was tracing a
> dio read performance issue on nvme.
> 
> Yep. No meaningful difference. I am sharing the results as below. [1].
> 
> [1]
> Before:
>   4971MB/s 4474GB iocounts: nvme3n1 545220968 nvme0n1 547007640 
> single_d2/5.18.0-rc5+_misc-next_1
> 
> After:
>   4968MB/s 4471GB iocounts: nvme3n1 544207371 nvme1n1 547458037 
> single_d2/5.18.0-rc5+_dio_cleanup_hch_1
> 
>   readstat /btrfs fio --eta=auto --output=$CMDLOG \
> --name fiotest --directory=/btrfs --rw=randread \
> --bs=4k --size=4G --ioengine=libaio --iodepth=16 --direct=1 \
> --time_based=1 --runtime=900 --randrepeat=1 --gtod_reduce=1 \
> --group_reporting=1 --numjobs=64

  Oops wrong thread pls ignore.

Re: cleanup btrfs bio handling, part 2 v3

[David Sterba]

On Thu, May 05, 2022 at 02:56:49PM +0800, Qu Wenruo wrote:
> On 2022/5/4 20:25, Christoph Hellwig wrote:
> > Hi all,
> >
> > this series removes the need to allocate a separate object for I/O
> > completions for all read and some write I/Os, and reduced the memory
> > usage of the low-level bios cloned by btrfs_map_bio by using plain bios
> > instead of the much larger btrfs_bio.
> 
> The most important part of this series is the patch "btrfs: remove
> btrfs_end_io_wq()" I guess.
> 
> As a guy growing up mostly reading btrfs code, the btrfs_end_io_wq()
> function looks straightforward to me initially, just put everything into
> wq context and call it a day.
> 
> But as I started to look into how other fses handle their delayed work,
> the btrfs way turns more "unique" other than straightforward.
> 
> So yes, aligning us to other fses, by only delaying into wq context when
> necessary, should be a good thing.
> Not reducing the memory allocation for the btrfs_end_io_wq structure,
> but also making the atomic part of the endio function executed with less
> delay.
> 
> But here I'm not sure if the btrfs way is a legacy from the past, or has
> some hidden intention (possible related to performance?).

Same here, I'm a bit worried about the implications of removing the
current end io way. I don't have a hard argument against, simplifying
the path is a good thing but as it's one of the fundamental
functionality I'd rather have it at least one cycle in the development
queue. As we're in rc5, there's not enough time to do evaluation on
various loads. Besides the test rig reports some performance degradation
on misc-next so it would be hard to evaluate this series as we don't
have a clean base line.

So current plan is to take what is safe from this patchset to 5.19
queue, the rest will appear in for-next. The misc-next will be frozen by
the end of the week, then 5.19 forked with selected fixes applied. In
parallel the for-next open is for any future development until 5.19-rc1.