support file system generated / verified integrity information

support file system generated / verified integrity information

[Christoph Hellwig]

Hi all,

this series adds support to generate and verify integrity information
(aka T10 PI) in the file system, instead of the automatic below the
covers support that is currently used.

There two reasons for this:

  a) to increase the protection enveloped.  Right now this is just a
     minor step from the bottom of the block layer to the file system,
     but it is required to support io_uring integrity data passthrough in
     the file system similar to the currently existing support for block
     devices, which will follow next.  It also allows the file system to
     directly see the integrity error and act upon in, e.g. when using
     RAID either integrated (as in btrfs) or by supporting reading
     redundant copies through the block layer.
  b) to make the PI processing more efficient.  This is primarily a
     concern for reads, where the block layer auto PI has to schedule a
     work item for each bio, and the file system them has to do it again
     for bounce buffering.  Additionally the current iomap post-I/O
     workqueue handling is a lot more efficient by supporting merging and
     avoiding workqueue scheduling storms.

The implementation is based on refactoring the existing block layer PI
code to be reusable for this use case, and then adding relatively small
wrappers for the file system use case.  These are then used in iomap
to implement the semantics, and wired up in XFS with a small amount of
glue code.

Compared to the baseline (iomap-bounce branch), this does not change
performance for writes, but increases read performance up to 15% for 4k
I/O, with the benefit decreasing with larger I/O sizes as even the
baseline maxes out the device quickly.  But even for larger I/O sizes,
the number of context switches decreases and cpu usage sees a minor
decrease.

The patches both require the recent integrity changes in the block tree
and the direct I/O bounce buffering series I previously sent out, so
you're probably better off just using the git tree instead of applying
the patches.

Git tree:

    git://git.infradead.org/users/hch/misc.git iomap-pi

Gitweb:

    https://git.infradead.org/?p=users/hch/misc.git;a=shortlog;h=refs/heads/iomap-pi

Diffstat:
 block/Makefile                |    2 
 block/bio-integrity-auto.c    |   80 +++---------------------
 block/bio-integrity-fs.c      |   81 +++++++++++++++++++++++++
 block/bio-integrity.c         |   64 +++++++++++++++++++
 block/bio.c                   |   17 +++--
 block/blk-mq.c                |    6 +
 block/blk-settings.c          |   13 ----
 block/blk.h                   |    6 +
 block/t10-pi.c                |   12 +--
 drivers/nvdimm/btt.c          |    6 +
 fs/fuse/file.c                |    5 -
 fs/iomap/bio.c                |  135 ++++++++++++++++++++++++++++--------------
 fs/iomap/buffered-io.c        |    8 +-
 fs/iomap/direct-io.c          |   15 ++++
 fs/iomap/internal.h           |   14 ++++
 fs/iomap/ioend.c              |   28 +++++++-
 fs/xfs/xfs_aops.c             |   47 +++++++++++++-
 fs/xfs/xfs_iomap.c            |    9 +-
 include/linux/bio-integrity.h |   12 ++-
 include/linux/bio.h           |    2 
 include/linux/blk-integrity.h |   21 ++++--
 include/linux/blkdev.h        |   34 ++++++++--
 include/linux/iomap.h         |   20 +++++-
 23 files changed, 456 insertions(+), 181 deletions(-)

[PATCH 01/15] block: factor out a bio_integrity_action helper

[Christoph Hellwig]

Split the logic to see if a bio needs integrity metadata from
bio_integrity_prep into a reusable helper than can be called from
file system code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 block/bio-integrity-auto.c    | 64 +++++------------------------------
 block/bio-integrity.c         | 48 ++++++++++++++++++++++++++
 block/blk-mq.c                |  6 ++--
 drivers/nvdimm/btt.c          |  6 ++--
 include/linux/bio-integrity.h |  5 ++-
 include/linux/blk-integrity.h | 16 +++++++++
 6 files changed, 82 insertions(+), 63 deletions(-)

diff --git a/block/bio-integrity-auto.c b/block/bio-integrity-auto.c
index 44dcdf7520c5..3a4141a9de0c 100644
--- a/block/bio-integrity-auto.c
+++ b/block/bio-integrity-auto.c
@@ -50,11 +50,6 @@ static bool bip_should_check(struct bio_integrity_payload *bip)
 	return bip->bip_flags & BIP_CHECK_FLAGS;
 }
 
-static bool bi_offload_capable(struct blk_integrity *bi)
-{
-	return bi->metadata_size == bi->pi_tuple_size;
-}
-
 /**
  * __bio_integrity_endio - Integrity I/O completion function
  * @bio:	Protected bio
@@ -84,69 +79,27 @@ bool __bio_integrity_endio(struct bio *bio)
 /**
  * bio_integrity_prep - Prepare bio for integrity I/O
  * @bio:	bio to prepare
+ * @action:	preparation action needed
+ *
+ * Allocate the integrity payload.  For writes, generate the integrity metadata
+ * and for reads, setup the completion handler to verify the metadata.
  *
- * Checks if the bio already has an integrity payload attached.  If it does, the
- * payload has been generated by another kernel subsystem, and we just pass it
- * through.
- * Otherwise allocates integrity payload and for writes the integrity metadata
- * will be generated.  For reads, the completion handler will verify the
- * metadata.
+ * This is used for bios that do not have user integrity payloads attached.
  */
-bool bio_integrity_prep(struct bio *bio)
+void bio_integrity_prep(struct bio *bio, unsigned int action)
 {
 	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
 	struct bio_integrity_data *bid;
-	bool set_flags = true;
-	gfp_t gfp = GFP_NOIO;
-
-	if (!bi)
-		return true;
-
-	if (!bio_sectors(bio))
-		return true;
-
-	/* Already protected? */
-	if (bio_integrity(bio))
-		return true;
-
-	switch (bio_op(bio)) {
-	case REQ_OP_READ:
-		if (bi->flags & BLK_INTEGRITY_NOVERIFY) {
-			if (bi_offload_capable(bi))
-				return true;
-			set_flags = false;
-		}
-		break;
-	case REQ_OP_WRITE:
-		/*
-		 * Zero the memory allocated to not leak uninitialized kernel
-		 * memory to disk for non-integrity metadata where nothing else
-		 * initializes the memory.
-		 */
-		if (bi->flags & BLK_INTEGRITY_NOGENERATE) {
-			if (bi_offload_capable(bi))
-				return true;
-			set_flags = false;
-			gfp |= __GFP_ZERO;
-		} else if (bi->metadata_size > bi->pi_tuple_size)
-			gfp |= __GFP_ZERO;
-		break;
-	default:
-		return true;
-	}
-
-	if (WARN_ON_ONCE(bio_has_crypt_ctx(bio)))
-		return true;
 
 	bid = mempool_alloc(&bid_pool, GFP_NOIO);
 	bio_integrity_init(bio, &bid->bip, &bid->bvec, 1);
 	bid->bio = bio;
 	bid->bip.bip_flags |= BIP_BLOCK_INTEGRITY;
-	bio_integrity_alloc_buf(bio, gfp & __GFP_ZERO);
+	bio_integrity_alloc_buf(bio, action & BI_ACT_ZERO);
 
 	bip_set_seed(&bid->bip, bio->bi_iter.bi_sector);
 
-	if (set_flags) {
+	if (action & BI_ACT_CHECK) {
 		if (bi->csum_type == BLK_INTEGRITY_CSUM_IP)
 			bid->bip.bip_flags |= BIP_IP_CHECKSUM;
 		if (bi->csum_type)
@@ -160,7 +113,6 @@ bool bio_integrity_prep(struct bio *bio)
 		blk_integrity_generate(bio);
 	else
 		bid->saved_bio_iter = bio->bi_iter;
-	return true;
 }
 EXPORT_SYMBOL(bio_integrity_prep);
 
diff --git a/block/bio-integrity.c b/block/bio-integrity.c
index 09eeaf6e74b8..6bdbb4ed2d1a 100644
--- a/block/bio-integrity.c
+++ b/block/bio-integrity.c
@@ -7,6 +7,7 @@
  */
 
 #include <linux/blk-integrity.h>
+#include <linux/t10-pi.h>
 #include "blk.h"
 
 struct bio_integrity_alloc {
@@ -16,6 +17,53 @@ struct bio_integrity_alloc {
 
 static mempool_t integrity_buf_pool;
 
+static bool bi_offload_capable(struct blk_integrity *bi)
+{
+	return bi->metadata_size == bi->pi_tuple_size;
+}
+
+unsigned int __bio_integrity_action(struct bio *bio)
+{
+	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
+
+	if (WARN_ON_ONCE(bio_has_crypt_ctx(bio)))
+		return 0;
+
+	switch (bio_op(bio)) {
+	case REQ_OP_READ:
+		if (bi->flags & BLK_INTEGRITY_NOVERIFY) {
+			if (bi_offload_capable(bi))
+				return 0;
+			return BI_ACT_BUFFER;
+		}
+		return BI_ACT_BUFFER | BI_ACT_CHECK;
+	case REQ_OP_WRITE:
+		/*
+		 * Flush masquerading as write?
+		 */
+		if (!bio_sectors(bio))
+			return 0;
+
+		/*
+		 * Zero the memory allocated to not leak uninitialized kernel
+		 * memory to disk for non-integrity metadata where nothing else
+		 * initializes the memory.
+		 */
+		if (bi->flags & BLK_INTEGRITY_NOGENERATE) {
+			if (bi_offload_capable(bi))
+				return 0;
+			return BI_ACT_BUFFER | BI_ACT_ZERO;
+		}
+
+		if (bi->metadata_size > bi->pi_tuple_size)
+			return BI_ACT_BUFFER | BI_ACT_CHECK | BI_ACT_ZERO;
+		return BI_ACT_BUFFER | BI_ACT_CHECK;
+	default:
+		return 0;
+	}
+}
+EXPORT_SYMBOL_GPL(__bio_integrity_action);
+
 void bio_integrity_alloc_buf(struct bio *bio, bool zero_buffer)
 {
 	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
diff --git a/block/blk-mq.c b/block/blk-mq.c
index a29d8ac9d3e3..3e58f6d50a1a 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -3133,6 +3133,7 @@ void blk_mq_submit_bio(struct bio *bio)
 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
 	struct blk_plug *plug = current->plug;
 	const int is_sync = op_is_sync(bio->bi_opf);
+	unsigned int integrity_action;
 	struct blk_mq_hw_ctx *hctx;
 	unsigned int nr_segs;
 	struct request *rq;
@@ -3185,8 +3186,9 @@ void blk_mq_submit_bio(struct bio *bio)
 	if (!bio)
 		goto queue_exit;
 
-	if (!bio_integrity_prep(bio))
-		goto queue_exit;
+	integrity_action = bio_integrity_action(bio);
+	if (integrity_action)
+		bio_integrity_prep(bio, integrity_action);
 
 	blk_mq_bio_issue_init(q, bio);
 	if (blk_mq_attempt_bio_merge(q, bio, nr_segs))
diff --git a/drivers/nvdimm/btt.c b/drivers/nvdimm/btt.c
index a933db961ed7..9cc4b659de1a 100644
--- a/drivers/nvdimm/btt.c
+++ b/drivers/nvdimm/btt.c
@@ -1437,14 +1437,16 @@ static void btt_submit_bio(struct bio *bio)
 {
 	struct bio_integrity_payload *bip = bio_integrity(bio);
 	struct btt *btt = bio->bi_bdev->bd_disk->private_data;
+	unsigned int integrity_action;
 	struct bvec_iter iter;
 	unsigned long start;
 	struct bio_vec bvec;
 	int err = 0;
 	bool do_acct;
 
-	if (!bio_integrity_prep(bio))
-		return;
+	integrity_action = bio_integrity_action(bio);
+	if (integrity_action)
+		bio_integrity_prep(bio, integrity_action);
 
 	do_acct = blk_queue_io_stat(bio->bi_bdev->bd_disk->queue);
 	if (do_acct)
diff --git a/include/linux/bio-integrity.h b/include/linux/bio-integrity.h
index 21e4652dcfd2..276cbbdd2c9d 100644
--- a/include/linux/bio-integrity.h
+++ b/include/linux/bio-integrity.h
@@ -78,7 +78,7 @@ int bio_integrity_add_page(struct bio *bio, struct page *page, unsigned int len,
 int bio_integrity_map_user(struct bio *bio, struct iov_iter *iter);
 int bio_integrity_map_iter(struct bio *bio, struct uio_meta *meta);
 void bio_integrity_unmap_user(struct bio *bio);
-bool bio_integrity_prep(struct bio *bio);
+void bio_integrity_prep(struct bio *bio, unsigned int action);
 void bio_integrity_advance(struct bio *bio, unsigned int bytes_done);
 void bio_integrity_trim(struct bio *bio);
 int bio_integrity_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp_mask);
@@ -104,9 +104,8 @@ static inline void bio_integrity_unmap_user(struct bio *bio)
 {
 }
 
-static inline bool bio_integrity_prep(struct bio *bio)
+static inline void bio_integrity_prep(struct bio *bio, unsigned int action)
 {
-	return true;
 }
 
 static inline int bio_integrity_clone(struct bio *bio, struct bio *bio_src,
diff --git a/include/linux/blk-integrity.h b/include/linux/blk-integrity.h
index c15b1ac62765..91d12610d252 100644
--- a/include/linux/blk-integrity.h
+++ b/include/linux/blk-integrity.h
@@ -180,4 +180,20 @@ static inline struct bio_vec rq_integrity_vec(struct request *rq)
 }
 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 
+enum bio_integrity_action {
+	BI_ACT_BUFFER		= (1u << 0),	/* allocate buffer */
+	BI_ACT_CHECK		= (1u << 1),	/* generate / verify PI */
+	BI_ACT_ZERO		= (1u << 2),	/* zero buffer */
+};
+
+unsigned int __bio_integrity_action(struct bio *bio);
+static inline unsigned int bio_integrity_action(struct bio *bio)
+{
+	if (!blk_get_integrity(bio->bi_bdev->bd_disk))
+		return 0;
+	if (bio_integrity(bio))
+		return 0;
+	return __bio_integrity_action(bio);
+}
+
 #endif /* _LINUX_BLK_INTEGRITY_H */
-- 
2.47.3

Re: [PATCH 01/15] block: factor out a bio_integrity_action helper

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:09AM +0100, Christoph Hellwig wrote:
> Split the logic to see if a bio needs integrity metadata from
> bio_integrity_prep into a reusable helper than can be called from
> file system code.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  block/bio-integrity-auto.c    | 64 +++++------------------------------
>  block/bio-integrity.c         | 48 ++++++++++++++++++++++++++
>  block/blk-mq.c                |  6 ++--
>  drivers/nvdimm/btt.c          |  6 ++--
>  include/linux/bio-integrity.h |  5 ++-
>  include/linux/blk-integrity.h | 16 +++++++++
>  6 files changed, 82 insertions(+), 63 deletions(-)
> 
> diff --git a/block/bio-integrity-auto.c b/block/bio-integrity-auto.c
> index 44dcdf7520c5..3a4141a9de0c 100644
> --- a/block/bio-integrity-auto.c
> +++ b/block/bio-integrity-auto.c
> @@ -50,11 +50,6 @@ static bool bip_should_check(struct bio_integrity_payload *bip)
>  	return bip->bip_flags & BIP_CHECK_FLAGS;
>  }
>  
> -static bool bi_offload_capable(struct blk_integrity *bi)
> -{
> -	return bi->metadata_size == bi->pi_tuple_size;
> -}
> -
>  /**
>   * __bio_integrity_endio - Integrity I/O completion function
>   * @bio:	Protected bio
> @@ -84,69 +79,27 @@ bool __bio_integrity_endio(struct bio *bio)
>  /**
>   * bio_integrity_prep - Prepare bio for integrity I/O
>   * @bio:	bio to prepare
> + * @action:	preparation action needed

What is @action?  Is it a bitset of BI_ACT_* values?  If yes, then can
the comment please say that explicitly?

> + *
> + * Allocate the integrity payload.  For writes, generate the integrity metadata
> + * and for reads, setup the completion handler to verify the metadata.
>   *
> - * Checks if the bio already has an integrity payload attached.  If it does, the
> - * payload has been generated by another kernel subsystem, and we just pass it
> - * through.
> - * Otherwise allocates integrity payload and for writes the integrity metadata
> - * will be generated.  For reads, the completion handler will verify the
> - * metadata.
> + * This is used for bios that do not have user integrity payloads attached.
>   */
> -bool bio_integrity_prep(struct bio *bio)
> +void bio_integrity_prep(struct bio *bio, unsigned int action)
>  {
>  	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
>  	struct bio_integrity_data *bid;
> -	bool set_flags = true;
> -	gfp_t gfp = GFP_NOIO;
> -
> -	if (!bi)
> -		return true;
> -
> -	if (!bio_sectors(bio))
> -		return true;
> -
> -	/* Already protected? */
> -	if (bio_integrity(bio))
> -		return true;
> -
> -	switch (bio_op(bio)) {
> -	case REQ_OP_READ:
> -		if (bi->flags & BLK_INTEGRITY_NOVERIFY) {
> -			if (bi_offload_capable(bi))
> -				return true;
> -			set_flags = false;
> -		}
> -		break;
> -	case REQ_OP_WRITE:
> -		/*
> -		 * Zero the memory allocated to not leak uninitialized kernel
> -		 * memory to disk for non-integrity metadata where nothing else
> -		 * initializes the memory.
> -		 */
> -		if (bi->flags & BLK_INTEGRITY_NOGENERATE) {
> -			if (bi_offload_capable(bi))
> -				return true;
> -			set_flags = false;
> -			gfp |= __GFP_ZERO;
> -		} else if (bi->metadata_size > bi->pi_tuple_size)
> -			gfp |= __GFP_ZERO;
> -		break;
> -	default:
> -		return true;
> -	}
> -
> -	if (WARN_ON_ONCE(bio_has_crypt_ctx(bio)))
> -		return true;
>  
>  	bid = mempool_alloc(&bid_pool, GFP_NOIO);
>  	bio_integrity_init(bio, &bid->bip, &bid->bvec, 1);
>  	bid->bio = bio;
>  	bid->bip.bip_flags |= BIP_BLOCK_INTEGRITY;
> -	bio_integrity_alloc_buf(bio, gfp & __GFP_ZERO);
> +	bio_integrity_alloc_buf(bio, action & BI_ACT_ZERO);
>  
>  	bip_set_seed(&bid->bip, bio->bi_iter.bi_sector);
>  
> -	if (set_flags) {
> +	if (action & BI_ACT_CHECK) {

Yes, I suppose it is a bitset.

--D

>  		if (bi->csum_type == BLK_INTEGRITY_CSUM_IP)
>  			bid->bip.bip_flags |= BIP_IP_CHECKSUM;
>  		if (bi->csum_type)
> @@ -160,7 +113,6 @@ bool bio_integrity_prep(struct bio *bio)
>  		blk_integrity_generate(bio);
>  	else
>  		bid->saved_bio_iter = bio->bi_iter;
> -	return true;
>  }
>  EXPORT_SYMBOL(bio_integrity_prep);
>  
> diff --git a/block/bio-integrity.c b/block/bio-integrity.c
> index 09eeaf6e74b8..6bdbb4ed2d1a 100644
> --- a/block/bio-integrity.c
> +++ b/block/bio-integrity.c
> @@ -7,6 +7,7 @@
>   */
>  
>  #include <linux/blk-integrity.h>
> +#include <linux/t10-pi.h>
>  #include "blk.h"
>  
>  struct bio_integrity_alloc {
> @@ -16,6 +17,53 @@ struct bio_integrity_alloc {
>  
>  static mempool_t integrity_buf_pool;
>  
> +static bool bi_offload_capable(struct blk_integrity *bi)
> +{
> +	return bi->metadata_size == bi->pi_tuple_size;
> +}

Just out of curiosity, what happens if metadata_size > pi_tuple_size?

Can it be the case that metadata_size < pi_tuple_size?

> +unsigned int __bio_integrity_action(struct bio *bio)

Hrm, this function returns a bitset of BI_ACT_* flags, doesn't it?

Would be kinda nice if a comment could say that.

> +{
> +	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
> +
> +	if (WARN_ON_ONCE(bio_has_crypt_ctx(bio)))
> +		return 0;
> +
> +	switch (bio_op(bio)) {
> +	case REQ_OP_READ:
> +		if (bi->flags & BLK_INTEGRITY_NOVERIFY) {
> +			if (bi_offload_capable(bi))
> +				return 0;
> +			return BI_ACT_BUFFER;
> +		}
> +		return BI_ACT_BUFFER | BI_ACT_CHECK;
> +	case REQ_OP_WRITE:
> +		/*
> +		 * Flush masquerading as write?
> +		 */
> +		if (!bio_sectors(bio))
> +			return 0;
> +
> +		/*
> +		 * Zero the memory allocated to not leak uninitialized kernel
> +		 * memory to disk for non-integrity metadata where nothing else
> +		 * initializes the memory.

Er... does someone initialize it eventually?  Such as the filesystem?
Or maybe an io_uring caller?

> +		 */
> +		if (bi->flags & BLK_INTEGRITY_NOGENERATE) {
> +			if (bi_offload_capable(bi))
> +				return 0;
> +			return BI_ACT_BUFFER | BI_ACT_ZERO;
> +		}
> +
> +		if (bi->metadata_size > bi->pi_tuple_size)
> +			return BI_ACT_BUFFER | BI_ACT_CHECK | BI_ACT_ZERO;
> +		return BI_ACT_BUFFER | BI_ACT_CHECK;

"check" feels like a weird name for a write, where we're generating the
PI information.  It really means "block layer takes care of PI
generation and validation", right?  As opposed to whichever upper layer
is using the block device?

BI_ACT_YOUDOIT <snerk>

How about BI_ACT_BDEV /* block layer checks/validates PI */

Everything else in here looks reasonable.

--D

> +	default:
> +		return 0;
> +	}
> +}
> +EXPORT_SYMBOL_GPL(__bio_integrity_action);
> +
>  void bio_integrity_alloc_buf(struct bio *bio, bool zero_buffer)
>  {
>  	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index a29d8ac9d3e3..3e58f6d50a1a 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -3133,6 +3133,7 @@ void blk_mq_submit_bio(struct bio *bio)
>  	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
>  	struct blk_plug *plug = current->plug;
>  	const int is_sync = op_is_sync(bio->bi_opf);
> +	unsigned int integrity_action;
>  	struct blk_mq_hw_ctx *hctx;
>  	unsigned int nr_segs;
>  	struct request *rq;
> @@ -3185,8 +3186,9 @@ void blk_mq_submit_bio(struct bio *bio)
>  	if (!bio)
>  		goto queue_exit;
>  
> -	if (!bio_integrity_prep(bio))
> -		goto queue_exit;
> +	integrity_action = bio_integrity_action(bio);
> +	if (integrity_action)
> +		bio_integrity_prep(bio, integrity_action);
>  
>  	blk_mq_bio_issue_init(q, bio);
>  	if (blk_mq_attempt_bio_merge(q, bio, nr_segs))
> diff --git a/drivers/nvdimm/btt.c b/drivers/nvdimm/btt.c
> index a933db961ed7..9cc4b659de1a 100644
> --- a/drivers/nvdimm/btt.c
> +++ b/drivers/nvdimm/btt.c
> @@ -1437,14 +1437,16 @@ static void btt_submit_bio(struct bio *bio)
>  {
>  	struct bio_integrity_payload *bip = bio_integrity(bio);
>  	struct btt *btt = bio->bi_bdev->bd_disk->private_data;
> +	unsigned int integrity_action;
>  	struct bvec_iter iter;
>  	unsigned long start;
>  	struct bio_vec bvec;
>  	int err = 0;
>  	bool do_acct;
>  
> -	if (!bio_integrity_prep(bio))
> -		return;
> +	integrity_action = bio_integrity_action(bio);
> +	if (integrity_action)
> +		bio_integrity_prep(bio, integrity_action);
>  
>  	do_acct = blk_queue_io_stat(bio->bi_bdev->bd_disk->queue);
>  	if (do_acct)
> diff --git a/include/linux/bio-integrity.h b/include/linux/bio-integrity.h
> index 21e4652dcfd2..276cbbdd2c9d 100644
> --- a/include/linux/bio-integrity.h
> +++ b/include/linux/bio-integrity.h
> @@ -78,7 +78,7 @@ int bio_integrity_add_page(struct bio *bio, struct page *page, unsigned int len,
>  int bio_integrity_map_user(struct bio *bio, struct iov_iter *iter);
>  int bio_integrity_map_iter(struct bio *bio, struct uio_meta *meta);
>  void bio_integrity_unmap_user(struct bio *bio);
> -bool bio_integrity_prep(struct bio *bio);
> +void bio_integrity_prep(struct bio *bio, unsigned int action);
>  void bio_integrity_advance(struct bio *bio, unsigned int bytes_done);
>  void bio_integrity_trim(struct bio *bio);
>  int bio_integrity_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp_mask);
> @@ -104,9 +104,8 @@ static inline void bio_integrity_unmap_user(struct bio *bio)
>  {
>  }
>  
> -static inline bool bio_integrity_prep(struct bio *bio)
> +static inline void bio_integrity_prep(struct bio *bio, unsigned int action)
>  {
> -	return true;
>  }
>  
>  static inline int bio_integrity_clone(struct bio *bio, struct bio *bio_src,
> diff --git a/include/linux/blk-integrity.h b/include/linux/blk-integrity.h
> index c15b1ac62765..91d12610d252 100644
> --- a/include/linux/blk-integrity.h
> +++ b/include/linux/blk-integrity.h
> @@ -180,4 +180,20 @@ static inline struct bio_vec rq_integrity_vec(struct request *rq)
>  }
>  #endif /* CONFIG_BLK_DEV_INTEGRITY */
>  
> +enum bio_integrity_action {
> +	BI_ACT_BUFFER		= (1u << 0),	/* allocate buffer */
> +	BI_ACT_CHECK		= (1u << 1),	/* generate / verify PI */
> +	BI_ACT_ZERO		= (1u << 2),	/* zero buffer */
> +};
> +
> +unsigned int __bio_integrity_action(struct bio *bio);
> +static inline unsigned int bio_integrity_action(struct bio *bio)
> +{
> +	if (!blk_get_integrity(bio->bi_bdev->bd_disk))
> +		return 0;
> +	if (bio_integrity(bio))
> +		return 0;
> +	return __bio_integrity_action(bio);
> +}
> +
>  #endif /* _LINUX_BLK_INTEGRITY_H */
> -- 
> 2.47.3
> 
> 

Re: [PATCH 01/15] block: factor out a bio_integrity_action helper

[Christoph Hellwig]

On Thu, Jan 22, 2026 at 04:01:13PM -0800, Darrick J. Wong wrote:
> >  /**
> >   * bio_integrity_prep - Prepare bio for integrity I/O
> >   * @bio:	bio to prepare
> > + * @action:	preparation action needed
> 
> What is @action?

Yes.

> Is it a bitset of BI_ACT_* values?  If yes, then can
> the comment please say that explicitly?

Is this good enough?

 * @action:     preparation action needed (BI_ACT_*)

> > +static bool bi_offload_capable(struct blk_integrity *bi)
> > +{
> > +	return bi->metadata_size == bi->pi_tuple_size;
> > +}
> 
> Just out of curiosity, what happens if metadata_size > pi_tuple_size?

Then we still have to provide a buffer as the automatic insert/strip
doesn't work. (I find the offload name rather confusing for this)

> Can it be the case that metadata_size < pi_tuple_size?

No.  See blk_validate_integrity_limits:

	if (bi->pi_offset + bi->pi_tuple_size > bi->metadata_size) {
		pr_warn("pi_offset (%u) + pi_tuple_size (%u) exceeds metadata_size (%u)\n",
			bi->pi_offset, bi->pi_tuple_size,
			bi->metadata_size);
		return -EINVAL;
	}


> 
> > +unsigned int __bio_integrity_action(struct bio *bio)
> 
> Hrm, this function returns a bitset of BI_ACT_* flags, doesn't it?
> 
> Would be kinda nice if a comment could say that.

Is this ok?

/**
 * bio_integrity_action - return the integrity action needed for a bio
 * @bio:        bio to operate on
 *
 * Returns the mask of integrity actions (BI_ACT_*) that need to be performed
 * for @bio.
 */


> > +		/*
> > +		 * Zero the memory allocated to not leak uninitialized kernel
> > +		 * memory to disk for non-integrity metadata where nothing else
> > +		 * initializes the memory.
> 
> Er... does someone initialize it eventually?  Such as the filesystem?
> Or maybe an io_uring caller?

For integrity metadata?  The code called later fills it out.  But it
doesn't fill non-integrity metadata, so we need to zero it.

> > +		 */
> > +		if (bi->flags & BLK_INTEGRITY_NOGENERATE) {
> > +			if (bi_offload_capable(bi))
> > +				return 0;
> > +			return BI_ACT_BUFFER | BI_ACT_ZERO;
> > +		}
> > +
> > +		if (bi->metadata_size > bi->pi_tuple_size)
> > +			return BI_ACT_BUFFER | BI_ACT_CHECK | BI_ACT_ZERO;
> > +		return BI_ACT_BUFFER | BI_ACT_CHECK;
> 
> "check" feels like a weird name for a write, where we're generating the
> PI information.  It really means "block layer takes care of PI
> generation and validation", right?  As opposed to whichever upper layer
> is using the block device?
> 
> BI_ACT_YOUDOIT <snerk>
> 
> How about BI_ACT_BDEV /* block layer checks/validates PI */

I think BI_ACT_BDEV is not very useful.  Check is supposed to
include generate and verify, but I'm not sure how we could word this
in a nice way.

Re: [PATCH 01/15] block: factor out a bio_integrity_action helper

[Darrick J. Wong]

On Fri, Jan 23, 2026 at 07:03:24AM +0100, Christoph Hellwig wrote:
> On Thu, Jan 22, 2026 at 04:01:13PM -0800, Darrick J. Wong wrote:
> > >  /**
> > >   * bio_integrity_prep - Prepare bio for integrity I/O
> > >   * @bio:	bio to prepare
> > > + * @action:	preparation action needed
> > 
> > What is @action?
> 
> Yes.
> 
> > Is it a bitset of BI_ACT_* values?  If yes, then can
> > the comment please say that explicitly?
> 
> Is this good enough?
> 
>  * @action:     preparation action needed (BI_ACT_*)

Yes.

> > > +static bool bi_offload_capable(struct blk_integrity *bi)
> > > +{
> > > +	return bi->metadata_size == bi->pi_tuple_size;
> > > +}
> > 
> > Just out of curiosity, what happens if metadata_size > pi_tuple_size?
> 
> Then we still have to provide a buffer as the automatic insert/strip
> doesn't work. (I find the offload name rather confusing for this)
> 
> > Can it be the case that metadata_size < pi_tuple_size?
> 
> No.  See blk_validate_integrity_limits:
> 
> 	if (bi->pi_offset + bi->pi_tuple_size > bi->metadata_size) {
> 		pr_warn("pi_offset (%u) + pi_tuple_size (%u) exceeds metadata_size (%u)\n",
> 			bi->pi_offset, bi->pi_tuple_size,
> 			bi->metadata_size);
> 		return -EINVAL;
> 	}
> 
> 
> > 
> > > +unsigned int __bio_integrity_action(struct bio *bio)
> > 
> > Hrm, this function returns a bitset of BI_ACT_* flags, doesn't it?
> > 
> > Would be kinda nice if a comment could say that.
> 
> Is this ok?
> 
> /**
>  * bio_integrity_action - return the integrity action needed for a bio
>  * @bio:        bio to operate on
>  *
>  * Returns the mask of integrity actions (BI_ACT_*) that need to be performed
>  * for @bio.
>  */

Excellent!

> 
> > > +		/*
> > > +		 * Zero the memory allocated to not leak uninitialized kernel
> > > +		 * memory to disk for non-integrity metadata where nothing else
> > > +		 * initializes the memory.
> > 
> > Er... does someone initialize it eventually?  Such as the filesystem?
> > Or maybe an io_uring caller?
> 
> For integrity metadata?  The code called later fills it out.  But it
> doesn't fill non-integrity metadata, so we need to zero it.

Ahhh, right, the app tag or whatever?

> > > +		 */
> > > +		if (bi->flags & BLK_INTEGRITY_NOGENERATE) {
> > > +			if (bi_offload_capable(bi))
> > > +				return 0;
> > > +			return BI_ACT_BUFFER | BI_ACT_ZERO;
> > > +		}
> > > +
> > > +		if (bi->metadata_size > bi->pi_tuple_size)
> > > +			return BI_ACT_BUFFER | BI_ACT_CHECK | BI_ACT_ZERO;
> > > +		return BI_ACT_BUFFER | BI_ACT_CHECK;
> > 
> > "check" feels like a weird name for a write, where we're generating the
> > PI information.  It really means "block layer takes care of PI
> > generation and validation", right?  As opposed to whichever upper layer
> > is using the block device?
> > 
> > BI_ACT_YOUDOIT <snerk>
> > 
> > How about BI_ACT_BDEV /* block layer checks/validates PI */
> 
> I think BI_ACT_BDEV is not very useful.  Check is supposed to
> include generate and verify, but I'm not sure how we could word this
> in a nice way.

Me neither.  If nobody else here comes up with a better suggestion then
I guess BI_ACT_CHECK + its comment wins.

--D

Re: [PATCH 01/15] block: factor out a bio_integrity_action helper

[Christoph Hellwig]

On Thu, Jan 22, 2026 at 11:13:23PM -0800, Darrick J. Wong wrote:
> > > Er... does someone initialize it eventually?  Such as the filesystem?
> > > Or maybe an io_uring caller?
> > 
> > For integrity metadata?  The code called later fills it out.  But it
> > doesn't fill non-integrity metadata, so we need to zero it.
> 
> Ahhh, right, the app tag or whatever?

No, the app_tag is zeroed by the PI code itself.  NVMe added the concept
of generic user specified metdata.  That metadata could either be only
the PI tuple, contain the PI tuple inside of it, or not contain any PI
tuple at all.  This non-PI part is what needs explicit zeroing here.

Re: [PATCH 01/15] block: factor out a bio_integrity_action helper

[Kanchan Joshi]

On 1/21/2026 12:13 PM, Christoph Hellwig wrote:
> +	case REQ_OP_WRITE:
> +		/*
> +		 * Flush masquerading as write?
> +		 */
> +		if (!bio_sectors(bio))
> +			return 0;
> +

Earlier this check was happening for REQ_OP_READ too but not sure 
whether that was superfluous or actually filtering out zero-sector 
reads. For everything else,

Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>

Re: [PATCH 01/15] block: factor out a bio_integrity_action helper

[Martin K. Petersen]

Christoph,

> Split the logic to see if a bio needs integrity metadata from
> bio_integrity_prep into a reusable helper than can be called from
> file system code.

Looks fine with Darrick's requested clarifications.

Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>

-- 
Martin K. Petersen

Re: [PATCH 01/15] block: factor out a bio_integrity_action helper

[Anuj gupta]

Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>

[PATCH 02/15] block: factor out a bio_integrity_setup_default helper

[Christoph Hellwig]

Add a helper to set the seed and check flag based on useful defaults
from the profile.

Note that this includes a small behavior change, as we ow only sets the
seed if any action is set, which is fine as nothing will look at it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 block/bio-integrity-auto.c    | 14 ++------------
 block/bio-integrity.c         | 16 ++++++++++++++++
 include/linux/bio-integrity.h |  1 +
 3 files changed, 19 insertions(+), 12 deletions(-)

diff --git a/block/bio-integrity-auto.c b/block/bio-integrity-auto.c
index 3a4141a9de0c..5345d55b9998 100644
--- a/block/bio-integrity-auto.c
+++ b/block/bio-integrity-auto.c
@@ -88,7 +88,6 @@ bool __bio_integrity_endio(struct bio *bio)
  */
 void bio_integrity_prep(struct bio *bio, unsigned int action)
 {
-	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
 	struct bio_integrity_data *bid;
 
 	bid = mempool_alloc(&bid_pool, GFP_NOIO);
@@ -96,17 +95,8 @@ void bio_integrity_prep(struct bio *bio, unsigned int action)
 	bid->bio = bio;
 	bid->bip.bip_flags |= BIP_BLOCK_INTEGRITY;
 	bio_integrity_alloc_buf(bio, action & BI_ACT_ZERO);
-
-	bip_set_seed(&bid->bip, bio->bi_iter.bi_sector);
-
-	if (action & BI_ACT_CHECK) {
-		if (bi->csum_type == BLK_INTEGRITY_CSUM_IP)
-			bid->bip.bip_flags |= BIP_IP_CHECKSUM;
-		if (bi->csum_type)
-			bid->bip.bip_flags |= BIP_CHECK_GUARD;
-		if (bi->flags & BLK_INTEGRITY_REF_TAG)
-			bid->bip.bip_flags |= BIP_CHECK_REFTAG;
-	}
+	if (action & BI_ACT_CHECK)
+		bio_integrity_setup_default(bio);
 
 	/* Auto-generate integrity metadata if this is a write */
 	if (bio_data_dir(bio) == WRITE && bip_should_check(&bid->bip))
diff --git a/block/bio-integrity.c b/block/bio-integrity.c
index 6bdbb4ed2d1a..0e8ebe84846e 100644
--- a/block/bio-integrity.c
+++ b/block/bio-integrity.c
@@ -101,6 +101,22 @@ void bio_integrity_free_buf(struct bio_integrity_payload *bip)
 		kfree(bvec_virt(bv));
 }
 
+void bio_integrity_setup_default(struct bio *bio)
+{
+	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
+	struct bio_integrity_payload *bip = bio_integrity(bio);
+
+	bip_set_seed(bip, bio->bi_iter.bi_sector);
+
+	if (bi->csum_type) {
+		bip->bip_flags |= BIP_CHECK_GUARD;
+		if (bi->csum_type == BLK_INTEGRITY_CSUM_IP)
+			bip->bip_flags |= BIP_IP_CHECKSUM;
+	}
+	if (bi->flags & BLK_INTEGRITY_REF_TAG)
+		bip->bip_flags |= BIP_CHECK_REFTAG;
+}
+
 /**
  * bio_integrity_free - Free bio integrity payload
  * @bio:	bio containing bip to be freed
diff --git a/include/linux/bio-integrity.h b/include/linux/bio-integrity.h
index 276cbbdd2c9d..232b86b9bbcb 100644
--- a/include/linux/bio-integrity.h
+++ b/include/linux/bio-integrity.h
@@ -143,5 +143,6 @@ static inline int bio_integrity_add_page(struct bio *bio, struct page *page,
 
 void bio_integrity_alloc_buf(struct bio *bio, bool zero_buffer);
 void bio_integrity_free_buf(struct bio_integrity_payload *bip);
+void bio_integrity_setup_default(struct bio *bio);
 
 #endif /* _LINUX_BIO_INTEGRITY_H */
-- 
2.47.3

Re: [PATCH 02/15] block: factor out a bio_integrity_setup_default helper

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:10AM +0100, Christoph Hellwig wrote:
> Add a helper to set the seed and check flag based on useful defaults
> from the profile.
> 
> Note that this includes a small behavior change, as we ow only sets the

"...as we only set the seed if..." ?

> seed if any action is set, which is fine as nothing will look at it.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  block/bio-integrity-auto.c    | 14 ++------------
>  block/bio-integrity.c         | 16 ++++++++++++++++
>  include/linux/bio-integrity.h |  1 +
>  3 files changed, 19 insertions(+), 12 deletions(-)
> 
> diff --git a/block/bio-integrity-auto.c b/block/bio-integrity-auto.c
> index 3a4141a9de0c..5345d55b9998 100644
> --- a/block/bio-integrity-auto.c
> +++ b/block/bio-integrity-auto.c
> @@ -88,7 +88,6 @@ bool __bio_integrity_endio(struct bio *bio)
>   */
>  void bio_integrity_prep(struct bio *bio, unsigned int action)
>  {
> -	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
>  	struct bio_integrity_data *bid;
>  
>  	bid = mempool_alloc(&bid_pool, GFP_NOIO);
> @@ -96,17 +95,8 @@ void bio_integrity_prep(struct bio *bio, unsigned int action)
>  	bid->bio = bio;
>  	bid->bip.bip_flags |= BIP_BLOCK_INTEGRITY;
>  	bio_integrity_alloc_buf(bio, action & BI_ACT_ZERO);
> -
> -	bip_set_seed(&bid->bip, bio->bi_iter.bi_sector);
> -
> -	if (action & BI_ACT_CHECK) {
> -		if (bi->csum_type == BLK_INTEGRITY_CSUM_IP)
> -			bid->bip.bip_flags |= BIP_IP_CHECKSUM;
> -		if (bi->csum_type)
> -			bid->bip.bip_flags |= BIP_CHECK_GUARD;
> -		if (bi->flags & BLK_INTEGRITY_REF_TAG)
> -			bid->bip.bip_flags |= BIP_CHECK_REFTAG;
> -	}
> +	if (action & BI_ACT_CHECK)
> +		bio_integrity_setup_default(bio);
>  
>  	/* Auto-generate integrity metadata if this is a write */
>  	if (bio_data_dir(bio) == WRITE && bip_should_check(&bid->bip))
> diff --git a/block/bio-integrity.c b/block/bio-integrity.c
> index 6bdbb4ed2d1a..0e8ebe84846e 100644
> --- a/block/bio-integrity.c
> +++ b/block/bio-integrity.c
> @@ -101,6 +101,22 @@ void bio_integrity_free_buf(struct bio_integrity_payload *bip)
>  		kfree(bvec_virt(bv));
>  }
>  
> +void bio_integrity_setup_default(struct bio *bio)
> +{
> +	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
> +	struct bio_integrity_payload *bip = bio_integrity(bio);
> +
> +	bip_set_seed(bip, bio->bi_iter.bi_sector);
> +
> +	if (bi->csum_type) {
> +		bip->bip_flags |= BIP_CHECK_GUARD;
> +		if (bi->csum_type == BLK_INTEGRITY_CSUM_IP)

/me wonders if this should be a switch, but it'd be a pretty lame one.

	switch (bi->csum_type) {
	case BLK_INTEGRITY_CSUM_NONE:
		break;
	case BLK_INTEGRITY_CSUM_IP:
		bip->bip_flags |= BIP_IP_CHECKSUM;
		fallthrough;
	case BLK_INTEGRITY_CSUM_CRC:
	case BLK_INTEGRITY_CSUM_CRC64:
		bip->bip_flags |= BIP_CHECK_GUARD;
		break;
	}

<shrug> I'll let you decide if you want slightly better typechecking.

With the commit message fixed,
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D


> +			bip->bip_flags |= BIP_IP_CHECKSUM;
> +	}
> +	if (bi->flags & BLK_INTEGRITY_REF_TAG)
> +		bip->bip_flags |= BIP_CHECK_REFTAG;
> +}
> +
>  /**
>   * bio_integrity_free - Free bio integrity payload
>   * @bio:	bio containing bip to be freed
> diff --git a/include/linux/bio-integrity.h b/include/linux/bio-integrity.h
> index 276cbbdd2c9d..232b86b9bbcb 100644
> --- a/include/linux/bio-integrity.h
> +++ b/include/linux/bio-integrity.h
> @@ -143,5 +143,6 @@ static inline int bio_integrity_add_page(struct bio *bio, struct page *page,
>  
>  void bio_integrity_alloc_buf(struct bio *bio, bool zero_buffer);
>  void bio_integrity_free_buf(struct bio_integrity_payload *bip);
> +void bio_integrity_setup_default(struct bio *bio);
>  
>  #endif /* _LINUX_BIO_INTEGRITY_H */
> -- 
> 2.47.3
> 
> 

Re: [PATCH 02/15] block: factor out a bio_integrity_setup_default helper

[Christoph Hellwig]

On Thu, Jan 22, 2026 at 04:05:37PM -0800, Darrick J. Wong wrote:
> On Wed, Jan 21, 2026 at 07:43:10AM +0100, Christoph Hellwig wrote:
> > Add a helper to set the seed and check flag based on useful defaults
> > from the profile.
> > 
> > Note that this includes a small behavior change, as we ow only sets the
> 
> "...as we only set the seed if..." ?

Yes.

> > +void bio_integrity_setup_default(struct bio *bio)
> > +{
> > +	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
> > +	struct bio_integrity_payload *bip = bio_integrity(bio);
> > +
> > +	bip_set_seed(bip, bio->bi_iter.bi_sector);
> > +
> > +	if (bi->csum_type) {
> > +		bip->bip_flags |= BIP_CHECK_GUARD;
> > +		if (bi->csum_type == BLK_INTEGRITY_CSUM_IP)
> 
> /me wonders if this should be a switch, but it'd be a pretty lame one.
> 
> 	switch (bi->csum_type) {
> 	case BLK_INTEGRITY_CSUM_NONE:
> 		break;
> 	case BLK_INTEGRITY_CSUM_IP:
> 		bip->bip_flags |= BIP_IP_CHECKSUM;
> 		fallthrough;
> 	case BLK_INTEGRITY_CSUM_CRC:
> 	case BLK_INTEGRITY_CSUM_CRC64:
> 		bip->bip_flags |= BIP_CHECK_GUARD;
> 		break;
> 	}

I don't really think that's a good idea here.  BIP_IP_CHECKSUM is a
really a special snowflake for SCSI HBA (and not even actual device)
usage, so it should be treated like a special snowflake with the if.
I sincerely hope no new special snowflakes will show up for the checksums
in the future.

Re: [PATCH 02/15] block: factor out a bio_integrity_setup_default helper

[Darrick J. Wong]

On Fri, Jan 23, 2026 at 07:08:33AM +0100, Christoph Hellwig wrote:
> On Thu, Jan 22, 2026 at 04:05:37PM -0800, Darrick J. Wong wrote:
> > On Wed, Jan 21, 2026 at 07:43:10AM +0100, Christoph Hellwig wrote:
> > > Add a helper to set the seed and check flag based on useful defaults
> > > from the profile.
> > > 
> > > Note that this includes a small behavior change, as we ow only sets the
> > 
> > "...as we only set the seed if..." ?
> 
> Yes.
> 
> > > +void bio_integrity_setup_default(struct bio *bio)
> > > +{
> > > +	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
> > > +	struct bio_integrity_payload *bip = bio_integrity(bio);
> > > +
> > > +	bip_set_seed(bip, bio->bi_iter.bi_sector);
> > > +
> > > +	if (bi->csum_type) {
> > > +		bip->bip_flags |= BIP_CHECK_GUARD;
> > > +		if (bi->csum_type == BLK_INTEGRITY_CSUM_IP)
> > 
> > /me wonders if this should be a switch, but it'd be a pretty lame one.
> > 
> > 	switch (bi->csum_type) {
> > 	case BLK_INTEGRITY_CSUM_NONE:
> > 		break;
> > 	case BLK_INTEGRITY_CSUM_IP:
> > 		bip->bip_flags |= BIP_IP_CHECKSUM;
> > 		fallthrough;
> > 	case BLK_INTEGRITY_CSUM_CRC:
> > 	case BLK_INTEGRITY_CSUM_CRC64:
> > 		bip->bip_flags |= BIP_CHECK_GUARD;
> > 		break;
> > 	}
> 
> I don't really think that's a good idea here.  BIP_IP_CHECKSUM is a
> really a special snowflake for SCSI HBA (and not even actual device)
> usage, so it should be treated like a special snowflake with the if.
> I sincerely hope no new special snowflakes will show up for the checksums
> in the future.

Fair enough.  The BIP_* flags encoding is ... odd. :)

--D

Re: [PATCH 02/15] block: factor out a bio_integrity_setup_default helper

[Kanchan Joshi]

On 1/21/2026 12:13 PM, Christoph Hellwig wrote:
> Note that this includes a small behavior change, as we ow only sets the

with commit wording fixed,

Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>

Re: [PATCH 02/15] block: factor out a bio_integrity_setup_default helper

[Martin K. Petersen]

Christoph,

> Add a helper to set the seed and check flag based on useful defaults
> from the profile.
>
> Note that this includes a small behavior change, as we ow only sets
> the seed if any action is set, which is fine as nothing will look at
> it.

Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>

-- 
Martin K. Petersen

Re: [PATCH 02/15] block: factor out a bio_integrity_setup_default helper

[Anuj gupta]

Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>

[PATCH 03/15] block: add a bdev_has_integrity_csum helper

[Christoph Hellwig]

Factor out a helper to see if the block device has an integrity checksum
from bdev_stable_writes so that it can be reused for other checks.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 include/linux/blkdev.h | 16 ++++++++++------
 1 file changed, 10 insertions(+), 6 deletions(-)

diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 438c4946b6e5..c1f3e6bcc217 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -1472,14 +1472,18 @@ static inline bool bdev_synchronous(struct block_device *bdev)
 	return bdev->bd_disk->queue->limits.features & BLK_FEAT_SYNCHRONOUS;
 }
 
-static inline bool bdev_stable_writes(struct block_device *bdev)
+static inline bool bdev_has_integrity_csum(struct block_device *bdev)
 {
-	struct request_queue *q = bdev_get_queue(bdev);
+	struct queue_limits *lim = bdev_limits(bdev);
 
-	if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
-	    q->limits.integrity.csum_type != BLK_INTEGRITY_CSUM_NONE)
-		return true;
-	return q->limits.features & BLK_FEAT_STABLE_WRITES;
+	return IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
+		lim->integrity.csum_type != BLK_INTEGRITY_CSUM_NONE;
+}
+
+static inline bool bdev_stable_writes(struct block_device *bdev)
+{
+	return bdev_has_integrity_csum(bdev) ||
+		(bdev_limits(bdev)->features & BLK_FEAT_STABLE_WRITES);
 }
 
 static inline bool blk_queue_write_cache(struct request_queue *q)
-- 
2.47.3

Re: [PATCH 03/15] block: add a bdev_has_integrity_csum helper

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:11AM +0100, Christoph Hellwig wrote:
> Factor out a helper to see if the block device has an integrity checksum
> from bdev_stable_writes so that it can be reused for other checks.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

This looks like a straightforward splitting of responsibilities, so
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D

> ---
>  include/linux/blkdev.h | 16 ++++++++++------
>  1 file changed, 10 insertions(+), 6 deletions(-)
> 
> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
> index 438c4946b6e5..c1f3e6bcc217 100644
> --- a/include/linux/blkdev.h
> +++ b/include/linux/blkdev.h
> @@ -1472,14 +1472,18 @@ static inline bool bdev_synchronous(struct block_device *bdev)
>  	return bdev->bd_disk->queue->limits.features & BLK_FEAT_SYNCHRONOUS;
>  }
>  
> -static inline bool bdev_stable_writes(struct block_device *bdev)
> +static inline bool bdev_has_integrity_csum(struct block_device *bdev)
>  {
> -	struct request_queue *q = bdev_get_queue(bdev);
> +	struct queue_limits *lim = bdev_limits(bdev);
>  
> -	if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
> -	    q->limits.integrity.csum_type != BLK_INTEGRITY_CSUM_NONE)
> -		return true;
> -	return q->limits.features & BLK_FEAT_STABLE_WRITES;
> +	return IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
> +		lim->integrity.csum_type != BLK_INTEGRITY_CSUM_NONE;
> +}
> +
> +static inline bool bdev_stable_writes(struct block_device *bdev)
> +{
> +	return bdev_has_integrity_csum(bdev) ||
> +		(bdev_limits(bdev)->features & BLK_FEAT_STABLE_WRITES);
>  }
>  
>  static inline bool blk_queue_write_cache(struct request_queue *q)
> -- 
> 2.47.3
> 
> 

Re: [PATCH 03/15] block: add a bdev_has_integrity_csum helper

[Kanchan Joshi]

Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>

Re: [PATCH 03/15] block: add a bdev_has_integrity_csum helper

[Martin K. Petersen]

Christoph,

> Factor out a helper to see if the block device has an integrity checksum
> from bdev_stable_writes so that it can be reused for other checks.

Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>

-- 
Martin K. Petersen

Re: [PATCH 03/15] block: add a bdev_has_integrity_csum helper

[Anuj gupta]

Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>

[PATCH 04/15] block: prepare generation / verification helpers for fs usage

[Christoph Hellwig]

Return the status from verify instead of directly stashing it in the bio,
and rename the helpers to use the usual bio_ prefix for things operating
on a bio.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 block/bio-integrity-auto.c |  4 ++--
 block/blk.h                |  6 ++++--
 block/t10-pi.c             | 12 ++++++------
 3 files changed, 12 insertions(+), 10 deletions(-)

diff --git a/block/bio-integrity-auto.c b/block/bio-integrity-auto.c
index 5345d55b9998..f68a17a8dbc2 100644
--- a/block/bio-integrity-auto.c
+++ b/block/bio-integrity-auto.c
@@ -39,7 +39,7 @@ static void bio_integrity_verify_fn(struct work_struct *work)
 		container_of(work, struct bio_integrity_data, work);
 	struct bio *bio = bid->bio;
 
-	blk_integrity_verify_iter(bio, &bid->saved_bio_iter);
+	bio->bi_status = bio_integrity_verify(bio, &bid->saved_bio_iter);
 	bio_integrity_finish(bid);
 	bio_endio(bio);
 }
@@ -100,7 +100,7 @@ void bio_integrity_prep(struct bio *bio, unsigned int action)
 
 	/* Auto-generate integrity metadata if this is a write */
 	if (bio_data_dir(bio) == WRITE && bip_should_check(&bid->bip))
-		blk_integrity_generate(bio);
+		bio_integrity_generate(bio);
 	else
 		bid->saved_bio_iter = bio->bi_iter;
 }
diff --git a/block/blk.h b/block/blk.h
index 886238cae5f1..d222ce4b6dfc 100644
--- a/block/blk.h
+++ b/block/blk.h
@@ -695,8 +695,10 @@ int bdev_open(struct block_device *bdev, blk_mode_t mode, void *holder,
 	      const struct blk_holder_ops *hops, struct file *bdev_file);
 int bdev_permission(dev_t dev, blk_mode_t mode, void *holder);
 
-void blk_integrity_generate(struct bio *bio);
-void blk_integrity_verify_iter(struct bio *bio, struct bvec_iter *saved_iter);
+void bio_integrity_generate(struct bio *bio);
+blk_status_t bio_integrity_verify(struct bio *bio,
+		struct bvec_iter *saved_iter);
+
 void blk_integrity_prepare(struct request *rq);
 void blk_integrity_complete(struct request *rq, unsigned int nr_bytes);
 
diff --git a/block/t10-pi.c b/block/t10-pi.c
index 0c4ed9702146..d27be6041fd3 100644
--- a/block/t10-pi.c
+++ b/block/t10-pi.c
@@ -372,7 +372,7 @@ static void ext_pi_type1_complete(struct request *rq, unsigned int nr_bytes)
 	}
 }
 
-void blk_integrity_generate(struct bio *bio)
+void bio_integrity_generate(struct bio *bio)
 {
 	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
 	struct bio_integrity_payload *bip = bio_integrity(bio);
@@ -404,7 +404,7 @@ void blk_integrity_generate(struct bio *bio)
 	}
 }
 
-void blk_integrity_verify_iter(struct bio *bio, struct bvec_iter *saved_iter)
+blk_status_t bio_integrity_verify(struct bio *bio, struct bvec_iter *saved_iter)
 {
 	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
 	struct bio_integrity_payload *bip = bio_integrity(bio);
@@ -439,11 +439,11 @@ void blk_integrity_verify_iter(struct bio *bio, struct bvec_iter *saved_iter)
 		}
 		kunmap_local(kaddr);
 
-		if (ret) {
-			bio->bi_status = ret;
-			return;
-		}
+		if (ret)
+			return ret;
 	}
+
+	return BLK_STS_OK;
 }
 
 void blk_integrity_prepare(struct request *rq)
-- 
2.47.3

Re: [PATCH 04/15] block: prepare generation / verification helpers for fs usage

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:12AM +0100, Christoph Hellwig wrote:
> Return the status from verify instead of directly stashing it in the bio,
> and rename the helpers to use the usual bio_ prefix for things operating
> on a bio.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

LGTM,
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D

> ---
>  block/bio-integrity-auto.c |  4 ++--
>  block/blk.h                |  6 ++++--
>  block/t10-pi.c             | 12 ++++++------
>  3 files changed, 12 insertions(+), 10 deletions(-)
> 
> diff --git a/block/bio-integrity-auto.c b/block/bio-integrity-auto.c
> index 5345d55b9998..f68a17a8dbc2 100644
> --- a/block/bio-integrity-auto.c
> +++ b/block/bio-integrity-auto.c
> @@ -39,7 +39,7 @@ static void bio_integrity_verify_fn(struct work_struct *work)
>  		container_of(work, struct bio_integrity_data, work);
>  	struct bio *bio = bid->bio;
>  
> -	blk_integrity_verify_iter(bio, &bid->saved_bio_iter);
> +	bio->bi_status = bio_integrity_verify(bio, &bid->saved_bio_iter);
>  	bio_integrity_finish(bid);
>  	bio_endio(bio);
>  }
> @@ -100,7 +100,7 @@ void bio_integrity_prep(struct bio *bio, unsigned int action)
>  
>  	/* Auto-generate integrity metadata if this is a write */
>  	if (bio_data_dir(bio) == WRITE && bip_should_check(&bid->bip))
> -		blk_integrity_generate(bio);
> +		bio_integrity_generate(bio);
>  	else
>  		bid->saved_bio_iter = bio->bi_iter;
>  }
> diff --git a/block/blk.h b/block/blk.h
> index 886238cae5f1..d222ce4b6dfc 100644
> --- a/block/blk.h
> +++ b/block/blk.h
> @@ -695,8 +695,10 @@ int bdev_open(struct block_device *bdev, blk_mode_t mode, void *holder,
>  	      const struct blk_holder_ops *hops, struct file *bdev_file);
>  int bdev_permission(dev_t dev, blk_mode_t mode, void *holder);
>  
> -void blk_integrity_generate(struct bio *bio);
> -void blk_integrity_verify_iter(struct bio *bio, struct bvec_iter *saved_iter);
> +void bio_integrity_generate(struct bio *bio);
> +blk_status_t bio_integrity_verify(struct bio *bio,
> +		struct bvec_iter *saved_iter);
> +
>  void blk_integrity_prepare(struct request *rq);
>  void blk_integrity_complete(struct request *rq, unsigned int nr_bytes);
>  
> diff --git a/block/t10-pi.c b/block/t10-pi.c
> index 0c4ed9702146..d27be6041fd3 100644
> --- a/block/t10-pi.c
> +++ b/block/t10-pi.c
> @@ -372,7 +372,7 @@ static void ext_pi_type1_complete(struct request *rq, unsigned int nr_bytes)
>  	}
>  }
>  
> -void blk_integrity_generate(struct bio *bio)
> +void bio_integrity_generate(struct bio *bio)
>  {
>  	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
>  	struct bio_integrity_payload *bip = bio_integrity(bio);
> @@ -404,7 +404,7 @@ void blk_integrity_generate(struct bio *bio)
>  	}
>  }
>  
> -void blk_integrity_verify_iter(struct bio *bio, struct bvec_iter *saved_iter)
> +blk_status_t bio_integrity_verify(struct bio *bio, struct bvec_iter *saved_iter)
>  {
>  	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
>  	struct bio_integrity_payload *bip = bio_integrity(bio);
> @@ -439,11 +439,11 @@ void blk_integrity_verify_iter(struct bio *bio, struct bvec_iter *saved_iter)
>  		}
>  		kunmap_local(kaddr);
>  
> -		if (ret) {
> -			bio->bi_status = ret;
> -			return;
> -		}
> +		if (ret)
> +			return ret;
>  	}
> +
> +	return BLK_STS_OK;
>  }
>  
>  void blk_integrity_prepare(struct request *rq)
> -- 
> 2.47.3
> 
> 

Re: [PATCH 04/15] block: prepare generation / verification helpers for fs usage

[Kanchan Joshi]

Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>

Re: [PATCH 04/15] block: prepare generation / verification helpers for fs usage

[Martin K. Petersen]

Christoph,

> Return the status from verify instead of directly stashing it in the bio,
> and rename the helpers to use the usual bio_ prefix for things operating
> on a bio.

Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>

-- 
Martin K. Petersen

Re: [PATCH 04/15] block: prepare generation / verification helpers for fs usage

[Anuj gupta]

Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>

[PATCH 05/15] block: make max_integrity_io_size public

[Christoph Hellwig]

File systems that generate integrity will need this, so move it out
of the block private or blk-mq specific headers.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 block/blk-settings.c          | 13 -------------
 include/linux/blk-integrity.h |  5 -----
 include/linux/blkdev.h        | 18 ++++++++++++++++++
 3 files changed, 18 insertions(+), 18 deletions(-)

diff --git a/block/blk-settings.c b/block/blk-settings.c
index a9e65dc090da..dabfab97fbab 100644
--- a/block/blk-settings.c
+++ b/block/blk-settings.c
@@ -123,19 +123,6 @@ static int blk_validate_zoned_limits(struct queue_limits *lim)
 	return 0;
 }
 
-/*
- * Maximum size of I/O that needs a block layer integrity buffer.  Limited
- * by the number of intervals for which we can fit the integrity buffer into
- * the buffer size.  Because the buffer is a single segment it is also limited
- * by the maximum segment size.
- */
-static inline unsigned int max_integrity_io_size(struct queue_limits *lim)
-{
-	return min_t(unsigned int, lim->max_segment_size,
-		(BLK_INTEGRITY_MAX_SIZE / lim->integrity.metadata_size) <<
-			lim->integrity.interval_exp);
-}
-
 static int blk_validate_integrity_limits(struct queue_limits *lim)
 {
 	struct blk_integrity *bi = &lim->integrity;
diff --git a/include/linux/blk-integrity.h b/include/linux/blk-integrity.h
index 91d12610d252..b2f34f696a4f 100644
--- a/include/linux/blk-integrity.h
+++ b/include/linux/blk-integrity.h
@@ -8,11 +8,6 @@
 
 struct request;
 
-/*
- * Maximum contiguous integrity buffer allocation.
- */
-#define BLK_INTEGRITY_MAX_SIZE		SZ_2M
-
 enum blk_integrity_flags {
 	BLK_INTEGRITY_NOVERIFY		= 1 << 0,
 	BLK_INTEGRITY_NOGENERATE	= 1 << 1,
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index c1f3e6bcc217..8d4e622c1f98 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -1875,6 +1875,24 @@ static inline int bio_split_rw_at(struct bio *bio,
 	return bio_split_io_at(bio, lim, segs, max_bytes, lim->dma_alignment);
 }
 
+/*
+ * Maximum contiguous integrity buffer allocation.
+ */
+#define BLK_INTEGRITY_MAX_SIZE		SZ_2M
+
+/*
+ * Maximum size of I/O that needs a block layer integrity buffer.  Limited
+ * by the number of intervals for which we can fit the integrity buffer into
+ * the buffer size.  Because the buffer is a single segment it is also limited
+ * by the maximum segment size.
+ */
+static inline unsigned int max_integrity_io_size(struct queue_limits *lim)
+{
+	return min_t(unsigned int, lim->max_segment_size,
+		(BLK_INTEGRITY_MAX_SIZE / lim->integrity.metadata_size) <<
+			lim->integrity.interval_exp);
+}
+
 #define DEFINE_IO_COMP_BATCH(name)	struct io_comp_batch name = { }
 
 #endif /* _LINUX_BLKDEV_H */
-- 
2.47.3

Re: [PATCH 05/15] block: make max_integrity_io_size public

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:13AM +0100, Christoph Hellwig wrote:
> File systems that generate integrity will need this, so move it out
> of the block private or blk-mq specific headers.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Seems reasonable to me given what I saw in the rest of the patchset...
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D

> ---
>  block/blk-settings.c          | 13 -------------
>  include/linux/blk-integrity.h |  5 -----
>  include/linux/blkdev.h        | 18 ++++++++++++++++++
>  3 files changed, 18 insertions(+), 18 deletions(-)
> 
> diff --git a/block/blk-settings.c b/block/blk-settings.c
> index a9e65dc090da..dabfab97fbab 100644
> --- a/block/blk-settings.c
> +++ b/block/blk-settings.c
> @@ -123,19 +123,6 @@ static int blk_validate_zoned_limits(struct queue_limits *lim)
>  	return 0;
>  }
>  
> -/*
> - * Maximum size of I/O that needs a block layer integrity buffer.  Limited
> - * by the number of intervals for which we can fit the integrity buffer into
> - * the buffer size.  Because the buffer is a single segment it is also limited
> - * by the maximum segment size.
> - */
> -static inline unsigned int max_integrity_io_size(struct queue_limits *lim)
> -{
> -	return min_t(unsigned int, lim->max_segment_size,
> -		(BLK_INTEGRITY_MAX_SIZE / lim->integrity.metadata_size) <<
> -			lim->integrity.interval_exp);
> -}
> -
>  static int blk_validate_integrity_limits(struct queue_limits *lim)
>  {
>  	struct blk_integrity *bi = &lim->integrity;
> diff --git a/include/linux/blk-integrity.h b/include/linux/blk-integrity.h
> index 91d12610d252..b2f34f696a4f 100644
> --- a/include/linux/blk-integrity.h
> +++ b/include/linux/blk-integrity.h
> @@ -8,11 +8,6 @@
>  
>  struct request;
>  
> -/*
> - * Maximum contiguous integrity buffer allocation.
> - */
> -#define BLK_INTEGRITY_MAX_SIZE		SZ_2M
> -
>  enum blk_integrity_flags {
>  	BLK_INTEGRITY_NOVERIFY		= 1 << 0,
>  	BLK_INTEGRITY_NOGENERATE	= 1 << 1,
> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
> index c1f3e6bcc217..8d4e622c1f98 100644
> --- a/include/linux/blkdev.h
> +++ b/include/linux/blkdev.h
> @@ -1875,6 +1875,24 @@ static inline int bio_split_rw_at(struct bio *bio,
>  	return bio_split_io_at(bio, lim, segs, max_bytes, lim->dma_alignment);
>  }
>  
> +/*
> + * Maximum contiguous integrity buffer allocation.
> + */
> +#define BLK_INTEGRITY_MAX_SIZE		SZ_2M
> +
> +/*
> + * Maximum size of I/O that needs a block layer integrity buffer.  Limited
> + * by the number of intervals for which we can fit the integrity buffer into
> + * the buffer size.  Because the buffer is a single segment it is also limited
> + * by the maximum segment size.
> + */
> +static inline unsigned int max_integrity_io_size(struct queue_limits *lim)
> +{
> +	return min_t(unsigned int, lim->max_segment_size,
> +		(BLK_INTEGRITY_MAX_SIZE / lim->integrity.metadata_size) <<
> +			lim->integrity.interval_exp);
> +}
> +
>  #define DEFINE_IO_COMP_BATCH(name)	struct io_comp_batch name = { }
>  
>  #endif /* _LINUX_BLKDEV_H */
> -- 
> 2.47.3
> 
> 

Re: [PATCH 05/15] block: make max_integrity_io_size public

[Kanchan Joshi]

Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>

Re: [PATCH 05/15] block: make max_integrity_io_size public

[Martin K. Petersen]

Christoph,

> File systems that generate integrity will need this, so move it out
> of the block private or blk-mq specific headers.

Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>

-- 
Martin K. Petersen

Re: [PATCH 05/15] block: make max_integrity_io_size public

[Anuj gupta]

Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>

[PATCH 06/15] block: add fs_bio_integrity helpers

[Christoph Hellwig]

Add a set of helpers for file system initiated integrity information.
These include mempool backed allocations and verifying based on a passed
in sector and size which is often available from file system completion
routines.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 block/Makefile                |  2 +-
 block/bio-integrity-fs.c      | 81 +++++++++++++++++++++++++++++++++++
 include/linux/bio-integrity.h |  6 +++
 3 files changed, 88 insertions(+), 1 deletion(-)
 create mode 100644 block/bio-integrity-fs.c

diff --git a/block/Makefile b/block/Makefile
index c65f4da93702..7dce2e44276c 100644
--- a/block/Makefile
+++ b/block/Makefile
@@ -26,7 +26,7 @@ bfq-y				:= bfq-iosched.o bfq-wf2q.o bfq-cgroup.o
 obj-$(CONFIG_IOSCHED_BFQ)	+= bfq.o
 
 obj-$(CONFIG_BLK_DEV_INTEGRITY) += bio-integrity.o blk-integrity.o t10-pi.o \
-				   bio-integrity-auto.o
+				   bio-integrity-auto.o bio-integrity-fs.o
 obj-$(CONFIG_BLK_DEV_ZONED)	+= blk-zoned.o
 obj-$(CONFIG_BLK_WBT)		+= blk-wbt.o
 obj-$(CONFIG_BLK_DEBUG_FS)	+= blk-mq-debugfs.o
diff --git a/block/bio-integrity-fs.c b/block/bio-integrity-fs.c
new file mode 100644
index 000000000000..c8b3c753965d
--- /dev/null
+++ b/block/bio-integrity-fs.c
@@ -0,0 +1,81 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2025 Christoph Hellwig.
+ */
+#include <linux/blk-integrity.h>
+#include <linux/bio-integrity.h>
+#include "blk.h"
+
+struct fs_bio_integrity_buf {
+	struct bio_integrity_payload	bip;
+	struct bio_vec			bvec;
+};
+
+static struct kmem_cache *fs_bio_integrity_cache;
+static mempool_t fs_bio_integrity_pool;
+
+void fs_bio_integrity_alloc(struct bio *bio)
+{
+	struct fs_bio_integrity_buf *iib;
+	unsigned int action;
+
+	action = bio_integrity_action(bio);
+	if (!action)
+		return;
+
+	iib = mempool_alloc(&fs_bio_integrity_pool, GFP_NOIO);
+	bio_integrity_init(bio, &iib->bip, &iib->bvec, 1);
+
+	bio_integrity_alloc_buf(bio, action & BI_ACT_ZERO);
+	if (action & BI_ACT_CHECK)
+		bio_integrity_setup_default(bio);
+}
+
+void fs_bio_integrity_free(struct bio *bio)
+{
+	struct bio_integrity_payload *bip = bio_integrity(bio);
+
+	bio_integrity_free_buf(bip);
+	mempool_free(container_of(bip, struct fs_bio_integrity_buf, bip),
+			&fs_bio_integrity_pool);
+
+	bio->bi_integrity = NULL;
+	bio->bi_opf &= ~REQ_INTEGRITY;
+}
+
+void fs_bio_integrity_generate(struct bio *bio)
+{
+	fs_bio_integrity_alloc(bio);
+	bio_integrity_generate(bio);
+}
+EXPORT_SYMBOL_GPL(fs_bio_integrity_generate);
+
+int fs_bio_integrity_verify(struct bio *bio, sector_t sector, unsigned int size)
+{
+	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
+	struct bio_integrity_payload *bip = bio_integrity(bio);
+
+	/*
+	 * Reinitialize bip->bit_iter.
+	 *
+	 * This is for use in the submitter after the driver is done with the
+	 * bio. Requires the submitter to remember the sector and the size.
+	 */
+
+	memset(&bip->bip_iter, 0, sizeof(bip->bip_iter));
+	bip->bip_iter.bi_sector = sector;
+	bip->bip_iter.bi_size = bio_integrity_bytes(bi, size >> SECTOR_SHIFT);
+	return blk_status_to_errno(bio_integrity_verify(bio, &bip->bip_iter));
+}
+
+static int __init fs_bio_integrity_init(void)
+{
+	fs_bio_integrity_cache = kmem_cache_create("fs_bio_integrity",
+			sizeof(struct fs_bio_integrity_buf), 0,
+			SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
+	if (mempool_init_slab_pool(&fs_bio_integrity_pool, BIO_POOL_SIZE,
+			fs_bio_integrity_cache))
+		panic("fs_bio_integrity: can't create pool\n");
+	return 0;
+}
+fs_initcall(fs_bio_integrity_init);
diff --git a/include/linux/bio-integrity.h b/include/linux/bio-integrity.h
index 232b86b9bbcb..503dc9bc655d 100644
--- a/include/linux/bio-integrity.h
+++ b/include/linux/bio-integrity.h
@@ -145,4 +145,10 @@ void bio_integrity_alloc_buf(struct bio *bio, bool zero_buffer);
 void bio_integrity_free_buf(struct bio_integrity_payload *bip);
 void bio_integrity_setup_default(struct bio *bio);
 
+void fs_bio_integrity_alloc(struct bio *bio);
+void fs_bio_integrity_free(struct bio *bio);
+void fs_bio_integrity_generate(struct bio *bio);
+int fs_bio_integrity_verify(struct bio *bio, sector_t sector,
+		unsigned int size);
+
 #endif /* _LINUX_BIO_INTEGRITY_H */
-- 
2.47.3

Re: [PATCH 06/15] block: add fs_bio_integrity helpers

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:14AM +0100, Christoph Hellwig wrote:
> Add a set of helpers for file system initiated integrity information.
> These include mempool backed allocations and verifying based on a passed
> in sector and size which is often available from file system completion
> routines.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  block/Makefile                |  2 +-
>  block/bio-integrity-fs.c      | 81 +++++++++++++++++++++++++++++++++++
>  include/linux/bio-integrity.h |  6 +++
>  3 files changed, 88 insertions(+), 1 deletion(-)
>  create mode 100644 block/bio-integrity-fs.c
> 
> diff --git a/block/Makefile b/block/Makefile
> index c65f4da93702..7dce2e44276c 100644
> --- a/block/Makefile
> +++ b/block/Makefile
> @@ -26,7 +26,7 @@ bfq-y				:= bfq-iosched.o bfq-wf2q.o bfq-cgroup.o
>  obj-$(CONFIG_IOSCHED_BFQ)	+= bfq.o
>  
>  obj-$(CONFIG_BLK_DEV_INTEGRITY) += bio-integrity.o blk-integrity.o t10-pi.o \
> -				   bio-integrity-auto.o
> +				   bio-integrity-auto.o bio-integrity-fs.o
>  obj-$(CONFIG_BLK_DEV_ZONED)	+= blk-zoned.o
>  obj-$(CONFIG_BLK_WBT)		+= blk-wbt.o
>  obj-$(CONFIG_BLK_DEBUG_FS)	+= blk-mq-debugfs.o
> diff --git a/block/bio-integrity-fs.c b/block/bio-integrity-fs.c
> new file mode 100644
> index 000000000000..c8b3c753965d
> --- /dev/null
> +++ b/block/bio-integrity-fs.c
> @@ -0,0 +1,81 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright (c) 2025 Christoph Hellwig.
> + */
> +#include <linux/blk-integrity.h>
> +#include <linux/bio-integrity.h>
> +#include "blk.h"
> +
> +struct fs_bio_integrity_buf {
> +	struct bio_integrity_payload	bip;
> +	struct bio_vec			bvec;
> +};
> +
> +static struct kmem_cache *fs_bio_integrity_cache;
> +static mempool_t fs_bio_integrity_pool;
> +
> +void fs_bio_integrity_alloc(struct bio *bio)
> +{
> +	struct fs_bio_integrity_buf *iib;
> +	unsigned int action;
> +
> +	action = bio_integrity_action(bio);
> +	if (!action)
> +		return;
> +
> +	iib = mempool_alloc(&fs_bio_integrity_pool, GFP_NOIO);
> +	bio_integrity_init(bio, &iib->bip, &iib->bvec, 1);
> +
> +	bio_integrity_alloc_buf(bio, action & BI_ACT_ZERO);
> +	if (action & BI_ACT_CHECK)
> +		bio_integrity_setup_default(bio);
> +}
> +
> +void fs_bio_integrity_free(struct bio *bio)
> +{
> +	struct bio_integrity_payload *bip = bio_integrity(bio);
> +
> +	bio_integrity_free_buf(bip);
> +	mempool_free(container_of(bip, struct fs_bio_integrity_buf, bip),
> +			&fs_bio_integrity_pool);
> +
> +	bio->bi_integrity = NULL;
> +	bio->bi_opf &= ~REQ_INTEGRITY;
> +}
> +
> +void fs_bio_integrity_generate(struct bio *bio)
> +{
> +	fs_bio_integrity_alloc(bio);
> +	bio_integrity_generate(bio);
> +}
> +EXPORT_SYMBOL_GPL(fs_bio_integrity_generate);
> +
> +int fs_bio_integrity_verify(struct bio *bio, sector_t sector, unsigned int size)
> +{
> +	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
> +	struct bio_integrity_payload *bip = bio_integrity(bio);
> +
> +	/*
> +	 * Reinitialize bip->bit_iter.

s/bit_iter/bip_iter/ ?

With that fixed, this looks fine to me;
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D

> +	 *
> +	 * This is for use in the submitter after the driver is done with the
> +	 * bio. Requires the submitter to remember the sector and the size.
> +	 */
> +
> +	memset(&bip->bip_iter, 0, sizeof(bip->bip_iter));
> +	bip->bip_iter.bi_sector = sector;
> +	bip->bip_iter.bi_size = bio_integrity_bytes(bi, size >> SECTOR_SHIFT);
> +	return blk_status_to_errno(bio_integrity_verify(bio, &bip->bip_iter));
> +}
> +
> +static int __init fs_bio_integrity_init(void)
> +{
> +	fs_bio_integrity_cache = kmem_cache_create("fs_bio_integrity",
> +			sizeof(struct fs_bio_integrity_buf), 0,
> +			SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
> +	if (mempool_init_slab_pool(&fs_bio_integrity_pool, BIO_POOL_SIZE,
> +			fs_bio_integrity_cache))
> +		panic("fs_bio_integrity: can't create pool\n");
> +	return 0;
> +}
> +fs_initcall(fs_bio_integrity_init);
> diff --git a/include/linux/bio-integrity.h b/include/linux/bio-integrity.h
> index 232b86b9bbcb..503dc9bc655d 100644
> --- a/include/linux/bio-integrity.h
> +++ b/include/linux/bio-integrity.h
> @@ -145,4 +145,10 @@ void bio_integrity_alloc_buf(struct bio *bio, bool zero_buffer);
>  void bio_integrity_free_buf(struct bio_integrity_payload *bip);
>  void bio_integrity_setup_default(struct bio *bio);
>  
> +void fs_bio_integrity_alloc(struct bio *bio);
> +void fs_bio_integrity_free(struct bio *bio);
> +void fs_bio_integrity_generate(struct bio *bio);
> +int fs_bio_integrity_verify(struct bio *bio, sector_t sector,
> +		unsigned int size);
> +
>  #endif /* _LINUX_BIO_INTEGRITY_H */
> -- 
> 2.47.3
> 
> 

Re: [PATCH 06/15] block: add fs_bio_integrity helpers

[Kanchan Joshi]

On 1/21/2026 12:13 PM, Christoph Hellwig wrote:
> +void fs_bio_integrity_alloc(struct bio *bio)
> +{
> +	struct fs_bio_integrity_buf *iib;
> +	unsigned int action;
> +
> +	action = bio_integrity_action(bio);
> +	if (!action)
> +		return;

So this may return from here, but <below>

> +
> +	iib = mempool_alloc(&fs_bio_integrity_pool, GFP_NOIO);
> +	bio_integrity_init(bio, &iib->bip, &iib->bvec, 1);
> +
> +	bio_integrity_alloc_buf(bio, action & BI_ACT_ZERO);
> +	if (action & BI_ACT_CHECK)
> +		bio_integrity_setup_default(bio);
> +}
> +
> +void fs_bio_integrity_free(struct bio *bio)
> +{
> +	struct bio_integrity_payload *bip = bio_integrity(bio);
> +
> +	bio_integrity_free_buf(bip);
> +	mempool_free(container_of(bip, struct fs_bio_integrity_buf, bip),
> +			&fs_bio_integrity_pool);
> +
> +	bio->bi_integrity = NULL;
> +	bio->bi_opf &= ~REQ_INTEGRITY;
> +}
> +
> +void fs_bio_integrity_generate(struct bio *bio)
> +{
> +	fs_bio_integrity_alloc(bio);

no check here. A potential null pointer deference in the next line as 
bio has no bip?
> +	bio_integrity_generate(bio);
> +}

Re: [PATCH 06/15] block: add fs_bio_integrity helpers

[Christoph Hellwig]

On Mon, Jan 26, 2026 at 11:42:10PM +0530, Kanchan Joshi wrote:
> On 1/21/2026 12:13 PM, Christoph Hellwig wrote:
> > +void fs_bio_integrity_alloc(struct bio *bio)
> > +{
> > +	struct fs_bio_integrity_buf *iib;
> > +	unsigned int action;
> > +
> > +	action = bio_integrity_action(bio);
> > +	if (!action)
> > +		return;
> 
> So this may return from here, but <below>

> > +void fs_bio_integrity_generate(struct bio *bio)
> > +{
> > +	fs_bio_integrity_alloc(bio);
> 
> no check here. A potential null pointer deference in the next line as 
> bio has no bip?
> > +	bio_integrity_generate(bio);
> > +}

fs_bio_integrity_alloc is only called when the device has PI metadata
with checksums.  So this case can't really happen.  That being said,
handling it in one case and not the other seems suboptimal and confusing.

Re: [PATCH 06/15] block: add fs_bio_integrity helpers

[Martin K. Petersen]

Christoph,

> Add a set of helpers for file system initiated integrity information.
> These include mempool backed allocations and verifying based on a
> passed in sector and size which is often available from file system
> completion routines.

Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>

-- 
Martin K. Petersen

Re: [PATCH 06/15] block: add fs_bio_integrity helpers

[Anuj gupta]

Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>

[PATCH 07/15] block: pass a maxlen argument to bio_iov_iter_bounce

[Christoph Hellwig]

Allow the file system to limit the size processed in a single
bounce operation.  This is needed when generating integrity data
so that the size of a single integrity segment can't overflow.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 block/bio.c          | 17 ++++++++++-------
 fs/iomap/direct-io.c |  2 +-
 include/linux/bio.h  |  2 +-
 3 files changed, 12 insertions(+), 9 deletions(-)

diff --git a/block/bio.c b/block/bio.c
index da795b1df52a..e89b24dc0283 100644
--- a/block/bio.c
+++ b/block/bio.c
@@ -1293,9 +1293,10 @@ static void bio_free_folios(struct bio *bio)
 	}
 }
 
-static int bio_iov_iter_bounce_write(struct bio *bio, struct iov_iter *iter)
+static int bio_iov_iter_bounce_write(struct bio *bio, struct iov_iter *iter,
+		size_t maxlen)
 {
-	size_t total_len = iov_iter_count(iter);
+	size_t total_len = min(maxlen, iov_iter_count(iter));
 
 	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
 		return -EINVAL;
@@ -1333,9 +1334,10 @@ static int bio_iov_iter_bounce_write(struct bio *bio, struct iov_iter *iter)
 	return 0;
 }
 
-static int bio_iov_iter_bounce_read(struct bio *bio, struct iov_iter *iter)
+static int bio_iov_iter_bounce_read(struct bio *bio, struct iov_iter *iter,
+		size_t maxlen)
 {
-	size_t len = min(iov_iter_count(iter), SZ_1M);
+	size_t len = min3(iov_iter_count(iter), maxlen, SZ_1M);
 	struct folio *folio;
 
 	folio = folio_alloc_greedy(GFP_KERNEL, &len);
@@ -1372,6 +1374,7 @@ static int bio_iov_iter_bounce_read(struct bio *bio, struct iov_iter *iter)
  * bio_iov_iter_bounce - bounce buffer data from an iter into a bio
  * @bio:	bio to send
  * @iter:	iter to read from / write into
+ * @maxlen:	maximum size to bounce
  *
  * Helper for direct I/O implementations that need to bounce buffer because
  * we need to checksum the data or perform other operations that require
@@ -1379,11 +1382,11 @@ static int bio_iov_iter_bounce_read(struct bio *bio, struct iov_iter *iter)
  * copies the data into it.  Needs to be paired with bio_iov_iter_unbounce()
  * called on completion.
  */
-int bio_iov_iter_bounce(struct bio *bio, struct iov_iter *iter)
+int bio_iov_iter_bounce(struct bio *bio, struct iov_iter *iter, size_t maxlen)
 {
 	if (op_is_write(bio_op(bio)))
-		return bio_iov_iter_bounce_write(bio, iter);
-	return bio_iov_iter_bounce_read(bio, iter);
+		return bio_iov_iter_bounce_write(bio, iter, maxlen);
+	return bio_iov_iter_bounce_read(bio, iter, maxlen);
 }
 
 static void bvec_unpin(struct bio_vec *bv, bool mark_dirty)
diff --git a/fs/iomap/direct-io.c b/fs/iomap/direct-io.c
index 9c572de0d596..842fc7fecb2d 100644
--- a/fs/iomap/direct-io.c
+++ b/fs/iomap/direct-io.c
@@ -326,7 +326,7 @@ static ssize_t iomap_dio_bio_iter_one(struct iomap_iter *iter,
 	bio->bi_end_io = iomap_dio_bio_end_io;
 
 	if (dio->flags & IOMAP_DIO_BOUNCE)
-		ret = bio_iov_iter_bounce(bio, dio->submit.iter);
+		ret = bio_iov_iter_bounce(bio, dio->submit.iter, UINT_MAX);
 	else
 		ret = bio_iov_iter_get_pages(bio, dio->submit.iter,
 					     alignment - 1);
diff --git a/include/linux/bio.h b/include/linux/bio.h
index 95cfc79b88b8..df0d7e71372a 100644
--- a/include/linux/bio.h
+++ b/include/linux/bio.h
@@ -479,7 +479,7 @@ void __bio_release_pages(struct bio *bio, bool mark_dirty);
 extern void bio_set_pages_dirty(struct bio *bio);
 extern void bio_check_pages_dirty(struct bio *bio);
 
-int bio_iov_iter_bounce(struct bio *bio, struct iov_iter *iter);
+int bio_iov_iter_bounce(struct bio *bio, struct iov_iter *iter, size_t maxlen);
 void bio_iov_iter_unbounce(struct bio *bio, bool is_error, bool mark_dirty);
 
 extern void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
-- 
2.47.3

Re: [PATCH 07/15] block: pass a maxlen argument to bio_iov_iter_bounce

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:15AM +0100, Christoph Hellwig wrote:
> Allow the file system to limit the size processed in a single
> bounce operation.  This is needed when generating integrity data
> so that the size of a single integrity segment can't overflow.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  block/bio.c          | 17 ++++++++++-------
>  fs/iomap/direct-io.c |  2 +-
>  include/linux/bio.h  |  2 +-
>  3 files changed, 12 insertions(+), 9 deletions(-)
> 
> diff --git a/block/bio.c b/block/bio.c
> index da795b1df52a..e89b24dc0283 100644
> --- a/block/bio.c
> +++ b/block/bio.c
> @@ -1293,9 +1293,10 @@ static void bio_free_folios(struct bio *bio)
>  	}
>  }
>  
> -static int bio_iov_iter_bounce_write(struct bio *bio, struct iov_iter *iter)
> +static int bio_iov_iter_bounce_write(struct bio *bio, struct iov_iter *iter,
> +		size_t maxlen)
>  {
> -	size_t total_len = iov_iter_count(iter);
> +	size_t total_len = min(maxlen, iov_iter_count(iter));
>  
>  	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
>  		return -EINVAL;
> @@ -1333,9 +1334,10 @@ static int bio_iov_iter_bounce_write(struct bio *bio, struct iov_iter *iter)
>  	return 0;
>  }
>  
> -static int bio_iov_iter_bounce_read(struct bio *bio, struct iov_iter *iter)
> +static int bio_iov_iter_bounce_read(struct bio *bio, struct iov_iter *iter,
> +		size_t maxlen)
>  {
> -	size_t len = min(iov_iter_count(iter), SZ_1M);
> +	size_t len = min3(iov_iter_count(iter), maxlen, SZ_1M);
>  	struct folio *folio;
>  
>  	folio = folio_alloc_greedy(GFP_KERNEL, &len);
> @@ -1372,6 +1374,7 @@ static int bio_iov_iter_bounce_read(struct bio *bio, struct iov_iter *iter)
>   * bio_iov_iter_bounce - bounce buffer data from an iter into a bio
>   * @bio:	bio to send
>   * @iter:	iter to read from / write into
> + * @maxlen:	maximum size to bounce
>   *
>   * Helper for direct I/O implementations that need to bounce buffer because
>   * we need to checksum the data or perform other operations that require
> @@ -1379,11 +1382,11 @@ static int bio_iov_iter_bounce_read(struct bio *bio, struct iov_iter *iter)
>   * copies the data into it.  Needs to be paired with bio_iov_iter_unbounce()
>   * called on completion.
>   */
> -int bio_iov_iter_bounce(struct bio *bio, struct iov_iter *iter)
> +int bio_iov_iter_bounce(struct bio *bio, struct iov_iter *iter, size_t maxlen)
>  {
>  	if (op_is_write(bio_op(bio)))
> -		return bio_iov_iter_bounce_write(bio, iter);
> -	return bio_iov_iter_bounce_read(bio, iter);
> +		return bio_iov_iter_bounce_write(bio, iter, maxlen);
> +	return bio_iov_iter_bounce_read(bio, iter, maxlen);
>  }
>  
>  static void bvec_unpin(struct bio_vec *bv, bool mark_dirty)
> diff --git a/fs/iomap/direct-io.c b/fs/iomap/direct-io.c
> index 9c572de0d596..842fc7fecb2d 100644
> --- a/fs/iomap/direct-io.c
> +++ b/fs/iomap/direct-io.c
> @@ -326,7 +326,7 @@ static ssize_t iomap_dio_bio_iter_one(struct iomap_iter *iter,
>  	bio->bi_end_io = iomap_dio_bio_end_io;
>  
>  	if (dio->flags & IOMAP_DIO_BOUNCE)
> -		ret = bio_iov_iter_bounce(bio, dio->submit.iter);
> +		ret = bio_iov_iter_bounce(bio, dio->submit.iter, UINT_MAX);

Nitpicking here, but shouldn't this be SIZE_MAX?

--D

>  	else
>  		ret = bio_iov_iter_get_pages(bio, dio->submit.iter,
>  					     alignment - 1);
> diff --git a/include/linux/bio.h b/include/linux/bio.h
> index 95cfc79b88b8..df0d7e71372a 100644
> --- a/include/linux/bio.h
> +++ b/include/linux/bio.h
> @@ -479,7 +479,7 @@ void __bio_release_pages(struct bio *bio, bool mark_dirty);
>  extern void bio_set_pages_dirty(struct bio *bio);
>  extern void bio_check_pages_dirty(struct bio *bio);
>  
> -int bio_iov_iter_bounce(struct bio *bio, struct iov_iter *iter);
> +int bio_iov_iter_bounce(struct bio *bio, struct iov_iter *iter, size_t maxlen);
>  void bio_iov_iter_unbounce(struct bio *bio, bool is_error, bool mark_dirty);
>  
>  extern void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
> -- 
> 2.47.3
> 
> 

Re: [PATCH 07/15] block: pass a maxlen argument to bio_iov_iter_bounce

[Christoph Hellwig]

On Wed, Jan 21, 2026 at 05:04:40PM -0800, Darrick J. Wong wrote:
> >  	if (dio->flags & IOMAP_DIO_BOUNCE)
> > -		ret = bio_iov_iter_bounce(bio, dio->submit.iter);
> > +		ret = bio_iov_iter_bounce(bio, dio->submit.iter, UINT_MAX);
> 
> Nitpicking here, but shouldn't this be SIZE_MAX?

bi_size can't store more than UINT_MAX, so I think this is correct.

Re: [PATCH 07/15] block: pass a maxlen argument to bio_iov_iter_bounce

[Darrick J. Wong]

On Thu, Jan 22, 2026 at 07:04:00AM +0100, Christoph Hellwig wrote:
> On Wed, Jan 21, 2026 at 05:04:40PM -0800, Darrick J. Wong wrote:
> > >  	if (dio->flags & IOMAP_DIO_BOUNCE)
> > > -		ret = bio_iov_iter_bounce(bio, dio->submit.iter);
> > > +		ret = bio_iov_iter_bounce(bio, dio->submit.iter, UINT_MAX);
> > 
> > Nitpicking here, but shouldn't this be SIZE_MAX?
> 
> bi_size can't store more than UINT_MAX, so I think this is correct.

Fair enough, I was just surprised that this later gets turned into a
min() involving SIZE_MAX.

Practically speaking, they're the same value so it doesn't matter much.
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D

Re: [PATCH 07/15] block: pass a maxlen argument to bio_iov_iter_bounce

[Martin K. Petersen]

Christoph,

> Allow the file system to limit the size processed in a single bounce
> operation. This is needed when generating integrity data so that the
> size of a single integrity segment can't overflow.

Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>

-- 
Martin K. Petersen

Re: [PATCH 07/15] block: pass a maxlen argument to bio_iov_iter_bounce

[Anuj gupta]

Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>

[PATCH 08/15] iomap: refactor iomap_bio_read_folio_range

[Christoph Hellwig]

Split out the logic to allocate a new bio and only keep the fast path
that adds more data to an existing bio in iomap_bio_read_folio_range.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/iomap/bio.c | 69 +++++++++++++++++++++++++++-----------------------
 1 file changed, 37 insertions(+), 32 deletions(-)

diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
index fc045f2e4c45..578b1202e037 100644
--- a/fs/iomap/bio.c
+++ b/fs/iomap/bio.c
@@ -26,45 +26,50 @@ static void iomap_bio_submit_read(struct iomap_read_folio_ctx *ctx)
 		submit_bio(bio);
 }
 
-static int iomap_bio_read_folio_range(const struct iomap_iter *iter,
+static void iomap_read_alloc_bio(const struct iomap_iter *iter,
 		struct iomap_read_folio_ctx *ctx, size_t plen)
 {
-	struct folio *folio = ctx->cur_folio;
 	const struct iomap *iomap = &iter->iomap;
-	loff_t pos = iter->pos;
-	size_t poff = offset_in_folio(folio, pos);
-	loff_t length = iomap_length(iter);
-	sector_t sector;
+	unsigned int nr_vecs = DIV_ROUND_UP(iomap_length(iter), PAGE_SIZE);
+	struct folio *folio = ctx->cur_folio;
+	gfp_t gfp = mapping_gfp_constraint(folio->mapping, GFP_KERNEL);
+	gfp_t orig_gfp = gfp;
 	struct bio *bio = ctx->read_ctx;
 
-	sector = iomap_sector(iomap, pos);
-	if (!bio || bio_end_sector(bio) != sector ||
-	    !bio_add_folio(bio, folio, plen, poff)) {
-		gfp_t gfp = mapping_gfp_constraint(folio->mapping, GFP_KERNEL);
-		gfp_t orig_gfp = gfp;
-		unsigned int nr_vecs = DIV_ROUND_UP(length, PAGE_SIZE);
+	if (bio)
+		submit_bio(bio);
+
+	/* Same as readahead_gfp_mask: */
+	if (ctx->rac)
+		gfp |= __GFP_NORETRY | __GFP_NOWARN;
 
-		if (bio)
-			submit_bio(bio);
+	/*
+	 * If the bio_alloc fails, try it again for a single page to avoid
+	 * having to deal with partial page reads.  This emulates what
+	 * do_mpage_read_folio does.
+	 */
+	bio = bio_alloc(iomap->bdev, bio_max_segs(nr_vecs), REQ_OP_READ, gfp);
+	if (!bio)
+		bio = bio_alloc(iomap->bdev, 1, REQ_OP_READ, orig_gfp);
+	if (ctx->rac)
+		bio->bi_opf |= REQ_RAHEAD;
+	bio->bi_iter.bi_sector = iomap_sector(iomap, iter->pos);
+	bio->bi_end_io = iomap_read_end_io;
+	bio_add_folio_nofail(bio, folio, plen,
+			offset_in_folio(folio, iter->pos));
+	ctx->read_ctx = bio;
+}
+
+static int iomap_bio_read_folio_range(const struct iomap_iter *iter,
+		struct iomap_read_folio_ctx *ctx, size_t plen)
+{
+	struct folio *folio = ctx->cur_folio;
+	struct bio *bio = ctx->read_ctx;
 
-		if (ctx->rac) /* same as readahead_gfp_mask */
-			gfp |= __GFP_NORETRY | __GFP_NOWARN;
-		bio = bio_alloc(iomap->bdev, bio_max_segs(nr_vecs), REQ_OP_READ,
-				     gfp);
-		/*
-		 * If the bio_alloc fails, try it again for a single page to
-		 * avoid having to deal with partial page reads.  This emulates
-		 * what do_mpage_read_folio does.
-		 */
-		if (!bio)
-			bio = bio_alloc(iomap->bdev, 1, REQ_OP_READ, orig_gfp);
-		if (ctx->rac)
-			bio->bi_opf |= REQ_RAHEAD;
-		bio->bi_iter.bi_sector = sector;
-		bio->bi_end_io = iomap_read_end_io;
-		bio_add_folio_nofail(bio, folio, plen, poff);
-		ctx->read_ctx = bio;
-	}
+	if (!bio ||
+	    bio_end_sector(bio) != iomap_sector(&iter->iomap, iter->pos) ||
+	    !bio_add_folio(bio, folio, plen, offset_in_folio(folio, iter->pos)))
+		iomap_read_alloc_bio(iter, ctx, plen);
 	return 0;
 }
 
-- 
2.47.3

Re: [PATCH 08/15] iomap: refactor iomap_bio_read_folio_range

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:16AM +0100, Christoph Hellwig wrote:
> Split out the logic to allocate a new bio and only keep the fast path
> that adds more data to an existing bio in iomap_bio_read_folio_range.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Looks like a simple function split to me...
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D

> ---
>  fs/iomap/bio.c | 69 +++++++++++++++++++++++++++-----------------------
>  1 file changed, 37 insertions(+), 32 deletions(-)
> 
> diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
> index fc045f2e4c45..578b1202e037 100644
> --- a/fs/iomap/bio.c
> +++ b/fs/iomap/bio.c
> @@ -26,45 +26,50 @@ static void iomap_bio_submit_read(struct iomap_read_folio_ctx *ctx)
>  		submit_bio(bio);
>  }
>  
> -static int iomap_bio_read_folio_range(const struct iomap_iter *iter,
> +static void iomap_read_alloc_bio(const struct iomap_iter *iter,
>  		struct iomap_read_folio_ctx *ctx, size_t plen)
>  {
> -	struct folio *folio = ctx->cur_folio;
>  	const struct iomap *iomap = &iter->iomap;
> -	loff_t pos = iter->pos;
> -	size_t poff = offset_in_folio(folio, pos);
> -	loff_t length = iomap_length(iter);
> -	sector_t sector;
> +	unsigned int nr_vecs = DIV_ROUND_UP(iomap_length(iter), PAGE_SIZE);
> +	struct folio *folio = ctx->cur_folio;
> +	gfp_t gfp = mapping_gfp_constraint(folio->mapping, GFP_KERNEL);
> +	gfp_t orig_gfp = gfp;
>  	struct bio *bio = ctx->read_ctx;
>  
> -	sector = iomap_sector(iomap, pos);
> -	if (!bio || bio_end_sector(bio) != sector ||
> -	    !bio_add_folio(bio, folio, plen, poff)) {
> -		gfp_t gfp = mapping_gfp_constraint(folio->mapping, GFP_KERNEL);
> -		gfp_t orig_gfp = gfp;
> -		unsigned int nr_vecs = DIV_ROUND_UP(length, PAGE_SIZE);
> +	if (bio)
> +		submit_bio(bio);
> +
> +	/* Same as readahead_gfp_mask: */
> +	if (ctx->rac)
> +		gfp |= __GFP_NORETRY | __GFP_NOWARN;
>  
> -		if (bio)
> -			submit_bio(bio);
> +	/*
> +	 * If the bio_alloc fails, try it again for a single page to avoid
> +	 * having to deal with partial page reads.  This emulates what
> +	 * do_mpage_read_folio does.
> +	 */
> +	bio = bio_alloc(iomap->bdev, bio_max_segs(nr_vecs), REQ_OP_READ, gfp);
> +	if (!bio)
> +		bio = bio_alloc(iomap->bdev, 1, REQ_OP_READ, orig_gfp);
> +	if (ctx->rac)
> +		bio->bi_opf |= REQ_RAHEAD;
> +	bio->bi_iter.bi_sector = iomap_sector(iomap, iter->pos);
> +	bio->bi_end_io = iomap_read_end_io;
> +	bio_add_folio_nofail(bio, folio, plen,
> +			offset_in_folio(folio, iter->pos));
> +	ctx->read_ctx = bio;
> +}
> +
> +static int iomap_bio_read_folio_range(const struct iomap_iter *iter,
> +		struct iomap_read_folio_ctx *ctx, size_t plen)
> +{
> +	struct folio *folio = ctx->cur_folio;
> +	struct bio *bio = ctx->read_ctx;
>  
> -		if (ctx->rac) /* same as readahead_gfp_mask */
> -			gfp |= __GFP_NORETRY | __GFP_NOWARN;
> -		bio = bio_alloc(iomap->bdev, bio_max_segs(nr_vecs), REQ_OP_READ,
> -				     gfp);
> -		/*
> -		 * If the bio_alloc fails, try it again for a single page to
> -		 * avoid having to deal with partial page reads.  This emulates
> -		 * what do_mpage_read_folio does.
> -		 */
> -		if (!bio)
> -			bio = bio_alloc(iomap->bdev, 1, REQ_OP_READ, orig_gfp);
> -		if (ctx->rac)
> -			bio->bi_opf |= REQ_RAHEAD;
> -		bio->bi_iter.bi_sector = sector;
> -		bio->bi_end_io = iomap_read_end_io;
> -		bio_add_folio_nofail(bio, folio, plen, poff);
> -		ctx->read_ctx = bio;
> -	}
> +	if (!bio ||
> +	    bio_end_sector(bio) != iomap_sector(&iter->iomap, iter->pos) ||
> +	    !bio_add_folio(bio, folio, plen, offset_in_folio(folio, iter->pos)))
> +		iomap_read_alloc_bio(iter, ctx, plen);
>  	return 0;
>  }
>  
> -- 
> 2.47.3
> 
> 

[PATCH 09/15] iomap: pass the iomap_iter to ->submit_read

[Christoph Hellwig]

This provides additional context for file systems.

Rename the fuse instance to match the method name while we're at it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/fuse/file.c         | 5 +++--
 fs/iomap/bio.c         | 3 ++-
 fs/iomap/buffered-io.c | 4 ++--
 include/linux/iomap.h  | 3 ++-
 4 files changed, 9 insertions(+), 6 deletions(-)

diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index 01bc894e9c2b..99b79dc876ea 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -947,7 +947,8 @@ static int fuse_iomap_read_folio_range_async(const struct iomap_iter *iter,
 	return ret;
 }
 
-static void fuse_iomap_read_submit(struct iomap_read_folio_ctx *ctx)
+static void fuse_iomap_submit_read(const struct iomap_iter *iter,
+		struct iomap_read_folio_ctx *ctx)
 {
 	struct fuse_fill_read_data *data = ctx->read_ctx;
 
@@ -958,7 +959,7 @@ static void fuse_iomap_read_submit(struct iomap_read_folio_ctx *ctx)
 
 static const struct iomap_read_ops fuse_iomap_read_ops = {
 	.read_folio_range = fuse_iomap_read_folio_range_async,
-	.submit_read = fuse_iomap_read_submit,
+	.submit_read = fuse_iomap_submit_read,
 };
 
 static int fuse_read_folio(struct file *file, struct folio *folio)
diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
index 578b1202e037..cb60d1facb5a 100644
--- a/fs/iomap/bio.c
+++ b/fs/iomap/bio.c
@@ -18,7 +18,8 @@ static void iomap_read_end_io(struct bio *bio)
 	bio_put(bio);
 }
 
-static void iomap_bio_submit_read(struct iomap_read_folio_ctx *ctx)
+static void iomap_bio_submit_read(const struct iomap_iter *iter,
+		struct iomap_read_folio_ctx *ctx)
 {
 	struct bio *bio = ctx->read_ctx;
 
diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c
index 6beb876658c0..4a15c6c153c4 100644
--- a/fs/iomap/buffered-io.c
+++ b/fs/iomap/buffered-io.c
@@ -573,7 +573,7 @@ void iomap_read_folio(const struct iomap_ops *ops,
 				&bytes_submitted);
 
 	if (ctx->ops->submit_read)
-		ctx->ops->submit_read(ctx);
+		ctx->ops->submit_read(&iter, ctx);
 
 	iomap_read_end(folio, bytes_submitted);
 }
@@ -637,7 +637,7 @@ void iomap_readahead(const struct iomap_ops *ops,
 					&cur_bytes_submitted);
 
 	if (ctx->ops->submit_read)
-		ctx->ops->submit_read(ctx);
+		ctx->ops->submit_read(&iter, ctx);
 
 	if (ctx->cur_folio)
 		iomap_read_end(ctx->cur_folio, cur_bytes_submitted);
diff --git a/include/linux/iomap.h b/include/linux/iomap.h
index ea79ca9c2d6b..bf6280fc51af 100644
--- a/include/linux/iomap.h
+++ b/include/linux/iomap.h
@@ -512,7 +512,8 @@ struct iomap_read_ops {
 	 *
 	 * This is optional.
 	 */
-	void (*submit_read)(struct iomap_read_folio_ctx *ctx);
+	void (*submit_read)(const struct iomap_iter *iter,
+			struct iomap_read_folio_ctx *ctx);
 };
 
 /*
-- 
2.47.3

Re: [PATCH 09/15] iomap: pass the iomap_iter to ->submit_read

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:17AM +0100, Christoph Hellwig wrote:
> This provides additional context for file systems.
> 
> Rename the fuse instance to match the method name while we're at it.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Looks fine, will probably have more comments when we start accessing the
iter...

Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D

> ---
>  fs/fuse/file.c         | 5 +++--
>  fs/iomap/bio.c         | 3 ++-
>  fs/iomap/buffered-io.c | 4 ++--
>  include/linux/iomap.h  | 3 ++-
>  4 files changed, 9 insertions(+), 6 deletions(-)
> 
> diff --git a/fs/fuse/file.c b/fs/fuse/file.c
> index 01bc894e9c2b..99b79dc876ea 100644
> --- a/fs/fuse/file.c
> +++ b/fs/fuse/file.c
> @@ -947,7 +947,8 @@ static int fuse_iomap_read_folio_range_async(const struct iomap_iter *iter,
>  	return ret;
>  }
>  
> -static void fuse_iomap_read_submit(struct iomap_read_folio_ctx *ctx)
> +static void fuse_iomap_submit_read(const struct iomap_iter *iter,
> +		struct iomap_read_folio_ctx *ctx)
>  {
>  	struct fuse_fill_read_data *data = ctx->read_ctx;
>  
> @@ -958,7 +959,7 @@ static void fuse_iomap_read_submit(struct iomap_read_folio_ctx *ctx)
>  
>  static const struct iomap_read_ops fuse_iomap_read_ops = {
>  	.read_folio_range = fuse_iomap_read_folio_range_async,
> -	.submit_read = fuse_iomap_read_submit,
> +	.submit_read = fuse_iomap_submit_read,
>  };
>  
>  static int fuse_read_folio(struct file *file, struct folio *folio)
> diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
> index 578b1202e037..cb60d1facb5a 100644
> --- a/fs/iomap/bio.c
> +++ b/fs/iomap/bio.c
> @@ -18,7 +18,8 @@ static void iomap_read_end_io(struct bio *bio)
>  	bio_put(bio);
>  }
>  
> -static void iomap_bio_submit_read(struct iomap_read_folio_ctx *ctx)
> +static void iomap_bio_submit_read(const struct iomap_iter *iter,
> +		struct iomap_read_folio_ctx *ctx)
>  {
>  	struct bio *bio = ctx->read_ctx;
>  
> diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c
> index 6beb876658c0..4a15c6c153c4 100644
> --- a/fs/iomap/buffered-io.c
> +++ b/fs/iomap/buffered-io.c
> @@ -573,7 +573,7 @@ void iomap_read_folio(const struct iomap_ops *ops,
>  				&bytes_submitted);
>  
>  	if (ctx->ops->submit_read)
> -		ctx->ops->submit_read(ctx);
> +		ctx->ops->submit_read(&iter, ctx);
>  
>  	iomap_read_end(folio, bytes_submitted);
>  }
> @@ -637,7 +637,7 @@ void iomap_readahead(const struct iomap_ops *ops,
>  					&cur_bytes_submitted);
>  
>  	if (ctx->ops->submit_read)
> -		ctx->ops->submit_read(ctx);
> +		ctx->ops->submit_read(&iter, ctx);
>  
>  	if (ctx->cur_folio)
>  		iomap_read_end(ctx->cur_folio, cur_bytes_submitted);
> diff --git a/include/linux/iomap.h b/include/linux/iomap.h
> index ea79ca9c2d6b..bf6280fc51af 100644
> --- a/include/linux/iomap.h
> +++ b/include/linux/iomap.h
> @@ -512,7 +512,8 @@ struct iomap_read_ops {
>  	 *
>  	 * This is optional.
>  	 */
> -	void (*submit_read)(struct iomap_read_folio_ctx *ctx);
> +	void (*submit_read)(const struct iomap_iter *iter,
> +			struct iomap_read_folio_ctx *ctx);
>  };
>  
>  /*
> -- 
> 2.47.3
> 
> 

[PATCH 10/15] iomap: only call into ->submit_read when there is a read_ctx

[Christoph Hellwig]

Move the NULL check into the callers to simplify the callees.  Not sure
how fuse worked before, given that it was missing the NULL check.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/iomap/bio.c         | 5 +----
 fs/iomap/buffered-io.c | 4 ++--
 2 files changed, 3 insertions(+), 6 deletions(-)

diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
index cb60d1facb5a..80bbd328bd3c 100644
--- a/fs/iomap/bio.c
+++ b/fs/iomap/bio.c
@@ -21,10 +21,7 @@ static void iomap_read_end_io(struct bio *bio)
 static void iomap_bio_submit_read(const struct iomap_iter *iter,
 		struct iomap_read_folio_ctx *ctx)
 {
-	struct bio *bio = ctx->read_ctx;
-
-	if (bio)
-		submit_bio(bio);
+	submit_bio(ctx->read_ctx);
 }
 
 static void iomap_read_alloc_bio(const struct iomap_iter *iter,
diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c
index 4a15c6c153c4..6367f7f38f0c 100644
--- a/fs/iomap/buffered-io.c
+++ b/fs/iomap/buffered-io.c
@@ -572,7 +572,7 @@ void iomap_read_folio(const struct iomap_ops *ops,
 		iter.status = iomap_read_folio_iter(&iter, ctx,
 				&bytes_submitted);
 
-	if (ctx->ops->submit_read)
+	if (ctx->read_ctx && ctx->ops->submit_read)
 		ctx->ops->submit_read(&iter, ctx);
 
 	iomap_read_end(folio, bytes_submitted);
@@ -636,7 +636,7 @@ void iomap_readahead(const struct iomap_ops *ops,
 		iter.status = iomap_readahead_iter(&iter, ctx,
 					&cur_bytes_submitted);
 
-	if (ctx->ops->submit_read)
+	if (ctx->read_ctx && ctx->ops->submit_read)
 		ctx->ops->submit_read(&iter, ctx);
 
 	if (ctx->cur_folio)
-- 
2.47.3

Re: [PATCH 10/15] iomap: only call into ->submit_read when there is a read_ctx

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:18AM +0100, Christoph Hellwig wrote:
> Move the NULL check into the callers to simplify the callees.  Not sure
> how fuse worked before, given that it was missing the NULL check.

Let's cc Joanne to find out then...? [done]

--D

> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  fs/iomap/bio.c         | 5 +----
>  fs/iomap/buffered-io.c | 4 ++--
>  2 files changed, 3 insertions(+), 6 deletions(-)
> 
> diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
> index cb60d1facb5a..80bbd328bd3c 100644
> --- a/fs/iomap/bio.c
> +++ b/fs/iomap/bio.c
> @@ -21,10 +21,7 @@ static void iomap_read_end_io(struct bio *bio)
>  static void iomap_bio_submit_read(const struct iomap_iter *iter,
>  		struct iomap_read_folio_ctx *ctx)
>  {
> -	struct bio *bio = ctx->read_ctx;
> -
> -	if (bio)
> -		submit_bio(bio);
> +	submit_bio(ctx->read_ctx);
>  }
>  
>  static void iomap_read_alloc_bio(const struct iomap_iter *iter,
> diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c
> index 4a15c6c153c4..6367f7f38f0c 100644
> --- a/fs/iomap/buffered-io.c
> +++ b/fs/iomap/buffered-io.c
> @@ -572,7 +572,7 @@ void iomap_read_folio(const struct iomap_ops *ops,
>  		iter.status = iomap_read_folio_iter(&iter, ctx,
>  				&bytes_submitted);
>  
> -	if (ctx->ops->submit_read)
> +	if (ctx->read_ctx && ctx->ops->submit_read)
>  		ctx->ops->submit_read(&iter, ctx);
>  
>  	iomap_read_end(folio, bytes_submitted);
> @@ -636,7 +636,7 @@ void iomap_readahead(const struct iomap_ops *ops,
>  		iter.status = iomap_readahead_iter(&iter, ctx,
>  					&cur_bytes_submitted);
>  
> -	if (ctx->ops->submit_read)
> +	if (ctx->read_ctx && ctx->ops->submit_read)
>  		ctx->ops->submit_read(&iter, ctx);
>  
>  	if (ctx->cur_folio)
> -- 
> 2.47.3
> 
> 

Re: [PATCH 10/15] iomap: only call into ->submit_read when there is a read_ctx

[Joanne Koong]

On Wed, Jan 21, 2026 at 4:44 PM Darrick J. Wong <djwong@kernel.org> wrote:
>
> On Wed, Jan 21, 2026 at 07:43:18AM +0100, Christoph Hellwig wrote:
> > Move the NULL check into the callers to simplify the callees.  Not sure
> > how fuse worked before, given that it was missing the NULL check.

In fuse, ctx->read_ctx is always valid. It gets initialized to point
to a local struct before it calls into
iomap_read_folio()/iomap_readahead()

>
> Let's cc Joanne to find out then...? [done]
>
> --D
>
> > Signed-off-by: Christoph Hellwig <hch@lst.de>
> > ---
> >  fs/iomap/bio.c         | 5 +----
> >  fs/iomap/buffered-io.c | 4 ++--
> >  2 files changed, 3 insertions(+), 6 deletions(-)
> >
> > diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
> > index cb60d1facb5a..80bbd328bd3c 100644
> > --- a/fs/iomap/bio.c
> > +++ b/fs/iomap/bio.c
> > @@ -21,10 +21,7 @@ static void iomap_read_end_io(struct bio *bio)
> >  static void iomap_bio_submit_read(const struct iomap_iter *iter,
> >               struct iomap_read_folio_ctx *ctx)
> >  {
> > -     struct bio *bio = ctx->read_ctx;
> > -
> > -     if (bio)
> > -             submit_bio(bio);
> > +     submit_bio(ctx->read_ctx);
> >  }
> >
> >  static void iomap_read_alloc_bio(const struct iomap_iter *iter,
> > diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c
> > index 4a15c6c153c4..6367f7f38f0c 100644
> > --- a/fs/iomap/buffered-io.c
> > +++ b/fs/iomap/buffered-io.c
> > @@ -572,7 +572,7 @@ void iomap_read_folio(const struct iomap_ops *ops,
> >               iter.status = iomap_read_folio_iter(&iter, ctx,
> >                               &bytes_submitted);
> >
> > -     if (ctx->ops->submit_read)
> > +     if (ctx->read_ctx && ctx->ops->submit_read)
> >               ctx->ops->submit_read(&iter, ctx);
> >
> >       iomap_read_end(folio, bytes_submitted);
> > @@ -636,7 +636,7 @@ void iomap_readahead(const struct iomap_ops *ops,
> >               iter.status = iomap_readahead_iter(&iter, ctx,
> >                                       &cur_bytes_submitted);
> >
> > -     if (ctx->ops->submit_read)
> > +     if (ctx->read_ctx && ctx->ops->submit_read)
> >               ctx->ops->submit_read(&iter, ctx);

I wonder if it makes sense to just have submit_read() take in the void
*read_ctx directly instead of it taking in the entire struct
iomap_read_folio_ctx. afaict, the other fields aren't used/necessary.

This patch as-is makes sense to me though.

Reviewed-by: Joanne Koong <joannelkoong@gmail.com>

Thanks,
Joanne

> >
> >       if (ctx->cur_folio)
> > --
> > 2.47.3
> >
> >

Re: [PATCH 10/15] iomap: only call into ->submit_read when there is a read_ctx

[Christoph Hellwig]

On Wed, Jan 21, 2026 at 06:44:12PM -0800, Joanne Koong wrote:
> On Wed, Jan 21, 2026 at 4:44 PM Darrick J. Wong <djwong@kernel.org> wrote:
> >
> > On Wed, Jan 21, 2026 at 07:43:18AM +0100, Christoph Hellwig wrote:
> > > Move the NULL check into the callers to simplify the callees.  Not sure
> > > how fuse worked before, given that it was missing the NULL check.
> 
> In fuse, ctx->read_ctx is always valid. It gets initialized to point
> to a local struct before it calls into
> iomap_read_folio()/iomap_readahead()

Ok, thanks.

Re: [PATCH 10/15] iomap: only call into ->submit_read when there is a read_ctx

[Darrick J. Wong]

On Thu, Jan 22, 2026 at 06:59:56AM +0100, Christoph Hellwig wrote:
> On Wed, Jan 21, 2026 at 06:44:12PM -0800, Joanne Koong wrote:
> > On Wed, Jan 21, 2026 at 4:44 PM Darrick J. Wong <djwong@kernel.org> wrote:
> > >
> > > On Wed, Jan 21, 2026 at 07:43:18AM +0100, Christoph Hellwig wrote:
> > > > Move the NULL check into the callers to simplify the callees.  Not sure
> > > > how fuse worked before, given that it was missing the NULL check.
> > 
> > In fuse, ctx->read_ctx is always valid. It gets initialized to point
> > to a local struct before it calls into
> > iomap_read_folio()/iomap_readahead()
> 
> Ok, thanks.

Looks good to me then,
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D

[PATCH 11/15] iomap: allow file systems to hook into buffered read bio submission

[Christoph Hellwig]

Files ystems such as btrfs have additional operations with bios such as
verifying data checksums.  Allow file systems to hook into submission
of the bio to allow for this processing by replacing the direct
submit_bio call in iomap_read_alloc_bio with a call into ->submit_read
and exporting iomap_read_alloc_bio.  Also add a new field to
struct iomap_read_folio_ctx to track the file logic offset of the current
read context.

Based on a patch from Goldwyn Rodrigues <rgoldwyn@suse.com>.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/iomap/bio.c        | 15 +++++++++------
 include/linux/iomap.h |  4 ++++
 2 files changed, 13 insertions(+), 6 deletions(-)

diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
index 80bbd328bd3c..903cb9fe759e 100644
--- a/fs/iomap/bio.c
+++ b/fs/iomap/bio.c
@@ -32,10 +32,11 @@ static void iomap_read_alloc_bio(const struct iomap_iter *iter,
 	struct folio *folio = ctx->cur_folio;
 	gfp_t gfp = mapping_gfp_constraint(folio->mapping, GFP_KERNEL);
 	gfp_t orig_gfp = gfp;
-	struct bio *bio = ctx->read_ctx;
+	struct bio *bio;
 
-	if (bio)
-		submit_bio(bio);
+	/* Submit the existing range if there was one. */
+	if (ctx->read_ctx)
+		ctx->ops->submit_read(iter, ctx);
 
 	/* Same as readahead_gfp_mask: */
 	if (ctx->rac)
@@ -56,9 +57,10 @@ static void iomap_read_alloc_bio(const struct iomap_iter *iter,
 	bio_add_folio_nofail(bio, folio, plen,
 			offset_in_folio(folio, iter->pos));
 	ctx->read_ctx = bio;
+	ctx->read_ctx_file_offset = iter->pos;
 }
 
-static int iomap_bio_read_folio_range(const struct iomap_iter *iter,
+int iomap_bio_read_folio_range(const struct iomap_iter *iter,
 		struct iomap_read_folio_ctx *ctx, size_t plen)
 {
 	struct folio *folio = ctx->cur_folio;
@@ -70,10 +72,11 @@ static int iomap_bio_read_folio_range(const struct iomap_iter *iter,
 		iomap_read_alloc_bio(iter, ctx, plen);
 	return 0;
 }
+EXPORT_SYMBOL_GPL(iomap_bio_read_folio_range);
 
 const struct iomap_read_ops iomap_bio_read_ops = {
-	.read_folio_range = iomap_bio_read_folio_range,
-	.submit_read = iomap_bio_submit_read,
+	.read_folio_range	= iomap_bio_read_folio_range,
+	.submit_read		= iomap_bio_submit_read,
 };
 EXPORT_SYMBOL_GPL(iomap_bio_read_ops);
 
diff --git a/include/linux/iomap.h b/include/linux/iomap.h
index bf6280fc51af..b3f545d41720 100644
--- a/include/linux/iomap.h
+++ b/include/linux/iomap.h
@@ -493,6 +493,7 @@ struct iomap_read_folio_ctx {
 	struct folio		*cur_folio;
 	struct readahead_control *rac;
 	void			*read_ctx;
+	loff_t			read_ctx_file_offset;
 };
 
 struct iomap_read_ops {
@@ -599,6 +600,9 @@ int iomap_swapfile_activate(struct swap_info_struct *sis,
 extern struct bio_set iomap_ioend_bioset;
 
 #ifdef CONFIG_BLOCK
+int iomap_bio_read_folio_range(const struct iomap_iter *iter,
+		struct iomap_read_folio_ctx *ctx, size_t plen);
+
 extern const struct iomap_read_ops iomap_bio_read_ops;
 
 static inline void iomap_bio_read_folio(struct folio *folio,
-- 
2.47.3

Re: [PATCH 11/15] iomap: allow file systems to hook into buffered read bio submission

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:19AM +0100, Christoph Hellwig wrote:
> Files ystems such as btrfs have additional operations with bios such as

  Filesystems

> verifying data checksums.  Allow file systems to hook into submission
> of the bio to allow for this processing by replacing the direct
> submit_bio call in iomap_read_alloc_bio with a call into ->submit_read
> and exporting iomap_read_alloc_bio.  Also add a new field to
> struct iomap_read_folio_ctx to track the file logic offset of the current
> read context.

Basically you're enabling filesystems to know what's the offset of a
read bio that iomap is about to start?  I guess that enables btrfs to
stash that info somewhere so that when the bio completes, it can go look
up the checksum or something, and compare?

Or for PI the filesystem can do the PI validation itself and if that
fails, decide if it's going to do something evil like ask another
replica to try reading the information?

(Not that iomap handles overlapping redundant mappings at all...)

--D

> Based on a patch from Goldwyn Rodrigues <rgoldwyn@suse.com>.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  fs/iomap/bio.c        | 15 +++++++++------
>  include/linux/iomap.h |  4 ++++
>  2 files changed, 13 insertions(+), 6 deletions(-)
> 
> diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
> index 80bbd328bd3c..903cb9fe759e 100644
> --- a/fs/iomap/bio.c
> +++ b/fs/iomap/bio.c
> @@ -32,10 +32,11 @@ static void iomap_read_alloc_bio(const struct iomap_iter *iter,
>  	struct folio *folio = ctx->cur_folio;
>  	gfp_t gfp = mapping_gfp_constraint(folio->mapping, GFP_KERNEL);
>  	gfp_t orig_gfp = gfp;
> -	struct bio *bio = ctx->read_ctx;
> +	struct bio *bio;
>  
> -	if (bio)
> -		submit_bio(bio);
> +	/* Submit the existing range if there was one. */
> +	if (ctx->read_ctx)
> +		ctx->ops->submit_read(iter, ctx);
>  
>  	/* Same as readahead_gfp_mask: */
>  	if (ctx->rac)
> @@ -56,9 +57,10 @@ static void iomap_read_alloc_bio(const struct iomap_iter *iter,
>  	bio_add_folio_nofail(bio, folio, plen,
>  			offset_in_folio(folio, iter->pos));
>  	ctx->read_ctx = bio;
> +	ctx->read_ctx_file_offset = iter->pos;
>  }
>  
> -static int iomap_bio_read_folio_range(const struct iomap_iter *iter,
> +int iomap_bio_read_folio_range(const struct iomap_iter *iter,
>  		struct iomap_read_folio_ctx *ctx, size_t plen)
>  {
>  	struct folio *folio = ctx->cur_folio;
> @@ -70,10 +72,11 @@ static int iomap_bio_read_folio_range(const struct iomap_iter *iter,
>  		iomap_read_alloc_bio(iter, ctx, plen);
>  	return 0;
>  }
> +EXPORT_SYMBOL_GPL(iomap_bio_read_folio_range);
>  
>  const struct iomap_read_ops iomap_bio_read_ops = {
> -	.read_folio_range = iomap_bio_read_folio_range,
> -	.submit_read = iomap_bio_submit_read,
> +	.read_folio_range	= iomap_bio_read_folio_range,
> +	.submit_read		= iomap_bio_submit_read,
>  };
>  EXPORT_SYMBOL_GPL(iomap_bio_read_ops);
>  
> diff --git a/include/linux/iomap.h b/include/linux/iomap.h
> index bf6280fc51af..b3f545d41720 100644
> --- a/include/linux/iomap.h
> +++ b/include/linux/iomap.h
> @@ -493,6 +493,7 @@ struct iomap_read_folio_ctx {
>  	struct folio		*cur_folio;
>  	struct readahead_control *rac;
>  	void			*read_ctx;
> +	loff_t			read_ctx_file_offset;
>  };
>  
>  struct iomap_read_ops {
> @@ -599,6 +600,9 @@ int iomap_swapfile_activate(struct swap_info_struct *sis,
>  extern struct bio_set iomap_ioend_bioset;
>  
>  #ifdef CONFIG_BLOCK
> +int iomap_bio_read_folio_range(const struct iomap_iter *iter,
> +		struct iomap_read_folio_ctx *ctx, size_t plen);
> +
>  extern const struct iomap_read_ops iomap_bio_read_ops;
>  
>  static inline void iomap_bio_read_folio(struct folio *folio,
> -- 
> 2.47.3
> 
> 

Re: [PATCH 11/15] iomap: allow file systems to hook into buffered read bio submission

[Christoph Hellwig]

On Wed, Jan 21, 2026 at 04:49:33PM -0800, Darrick J. Wong wrote:
> > verifying data checksums.  Allow file systems to hook into submission
> > of the bio to allow for this processing by replacing the direct
> > submit_bio call in iomap_read_alloc_bio with a call into ->submit_read
> > and exporting iomap_read_alloc_bio.  Also add a new field to
> > struct iomap_read_folio_ctx to track the file logic offset of the current
> > read context.
> 
> Basically you're enabling filesystems to know what's the offset of a
> read bio that iomap is about to start?

Yes.

> I guess that enables btrfs to
> stash that info somewhere so that when the bio completes, it can go look
> up the checksum or something, and compare?

Yes, where something in this series is the ioend that the bio is embedded
into.

> Or for PI the filesystem can do the PI validation itself and if that
> fails, decide if it's going to do something evil like ask another
> replica to try reading the information?

Yeah.  If your file system can do it.  Nothing in this series does
that, though.

Re: [PATCH 11/15] iomap: allow file systems to hook into buffered read bio submission

[Darrick J. Wong]

On Thu, Jan 22, 2026 at 07:01:14AM +0100, Christoph Hellwig wrote:
> On Wed, Jan 21, 2026 at 04:49:33PM -0800, Darrick J. Wong wrote:
> > > verifying data checksums.  Allow file systems to hook into submission
> > > of the bio to allow for this processing by replacing the direct
> > > submit_bio call in iomap_read_alloc_bio with a call into ->submit_read
> > > and exporting iomap_read_alloc_bio.  Also add a new field to
> > > struct iomap_read_folio_ctx to track the file logic offset of the current
> > > read context.
> > 
> > Basically you're enabling filesystems to know what's the offset of a
> > read bio that iomap is about to start?
> 
> Yes.
> 
> > I guess that enables btrfs to
> > stash that info somewhere so that when the bio completes, it can go look
> > up the checksum or something, and compare?
> 
> Yes, where something in this series is the ioend that the bio is embedded
> into.
> 
> > Or for PI the filesystem can do the PI validation itself and if that
> > fails, decide if it's going to do something evil like ask another
> > replica to try reading the information?
> 
> Yeah.  If your file system can do it.  Nothing in this series does
> that, though.

<nod> and at this point we're never going to circle back to that project
either.  :/

In the meantime I think I've understood this patch well enough to say
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D

[PATCH 12/15] iomap: add a bioset pointer to iomap_read_folio_ops

[Christoph Hellwig]

Optionally allocate the bio from the bioset provided in
iomap_read_folio_ops.  If no bioset is provided, fs_bio_set is still
used, which is the standard bioset for file systems.

Based on a patch from Goldwyn Rodrigues <rgoldwyn@suse.com>.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/iomap/bio.c        | 14 ++++++++++++--
 include/linux/iomap.h |  6 ++++++
 2 files changed, 18 insertions(+), 2 deletions(-)

diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
index 903cb9fe759e..259a2bf95a43 100644
--- a/fs/iomap/bio.c
+++ b/fs/iomap/bio.c
@@ -24,11 +24,19 @@ static void iomap_bio_submit_read(const struct iomap_iter *iter,
 	submit_bio(ctx->read_ctx);
 }
 
+static struct bio_set *iomap_read_bio_set(struct iomap_read_folio_ctx *ctx)
+{
+	if (ctx->ops && ctx->ops->bio_set)
+		return ctx->ops->bio_set;
+	return &fs_bio_set;
+}
+
 static void iomap_read_alloc_bio(const struct iomap_iter *iter,
 		struct iomap_read_folio_ctx *ctx, size_t plen)
 {
 	const struct iomap *iomap = &iter->iomap;
 	unsigned int nr_vecs = DIV_ROUND_UP(iomap_length(iter), PAGE_SIZE);
+	struct bio_set *bio_set = iomap_read_bio_set(ctx);
 	struct folio *folio = ctx->cur_folio;
 	gfp_t gfp = mapping_gfp_constraint(folio->mapping, GFP_KERNEL);
 	gfp_t orig_gfp = gfp;
@@ -47,9 +55,11 @@ static void iomap_read_alloc_bio(const struct iomap_iter *iter,
 	 * having to deal with partial page reads.  This emulates what
 	 * do_mpage_read_folio does.
 	 */
-	bio = bio_alloc(iomap->bdev, bio_max_segs(nr_vecs), REQ_OP_READ, gfp);
+	bio = bio_alloc_bioset(iomap->bdev, bio_max_segs(nr_vecs), REQ_OP_READ,
+			gfp, bio_set);
 	if (!bio)
-		bio = bio_alloc(iomap->bdev, 1, REQ_OP_READ, orig_gfp);
+		bio = bio_alloc_bioset(iomap->bdev, 1, REQ_OP_READ, orig_gfp,
+				bio_set);
 	if (ctx->rac)
 		bio->bi_opf |= REQ_RAHEAD;
 	bio->bi_iter.bi_sector = iomap_sector(iomap, iter->pos);
diff --git a/include/linux/iomap.h b/include/linux/iomap.h
index b3f545d41720..24f884b6b0c4 100644
--- a/include/linux/iomap.h
+++ b/include/linux/iomap.h
@@ -515,6 +515,12 @@ struct iomap_read_ops {
 	 */
 	void (*submit_read)(const struct iomap_iter *iter,
 			struct iomap_read_folio_ctx *ctx);
+
+	/*
+	 * Optional, allows filesystem to specify own bio_set, so new bio's
+	 * can be allocated from the provided bio_set.
+	 */
+	struct bio_set *bio_set;
 };
 
 /*
-- 
2.47.3

Re: [PATCH 12/15] iomap: add a bioset pointer to iomap_read_folio_ops

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:20AM +0100, Christoph Hellwig wrote:
> Optionally allocate the bio from the bioset provided in
> iomap_read_folio_ops.  If no bioset is provided, fs_bio_set is still
> used, which is the standard bioset for file systems.
> 
> Based on a patch from Goldwyn Rodrigues <rgoldwyn@suse.com>.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Looks reasonable,
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D

> ---
>  fs/iomap/bio.c        | 14 ++++++++++++--
>  include/linux/iomap.h |  6 ++++++
>  2 files changed, 18 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
> index 903cb9fe759e..259a2bf95a43 100644
> --- a/fs/iomap/bio.c
> +++ b/fs/iomap/bio.c
> @@ -24,11 +24,19 @@ static void iomap_bio_submit_read(const struct iomap_iter *iter,
>  	submit_bio(ctx->read_ctx);
>  }
>  
> +static struct bio_set *iomap_read_bio_set(struct iomap_read_folio_ctx *ctx)
> +{
> +	if (ctx->ops && ctx->ops->bio_set)
> +		return ctx->ops->bio_set;
> +	return &fs_bio_set;
> +}
> +
>  static void iomap_read_alloc_bio(const struct iomap_iter *iter,
>  		struct iomap_read_folio_ctx *ctx, size_t plen)
>  {
>  	const struct iomap *iomap = &iter->iomap;
>  	unsigned int nr_vecs = DIV_ROUND_UP(iomap_length(iter), PAGE_SIZE);
> +	struct bio_set *bio_set = iomap_read_bio_set(ctx);
>  	struct folio *folio = ctx->cur_folio;
>  	gfp_t gfp = mapping_gfp_constraint(folio->mapping, GFP_KERNEL);
>  	gfp_t orig_gfp = gfp;
> @@ -47,9 +55,11 @@ static void iomap_read_alloc_bio(const struct iomap_iter *iter,
>  	 * having to deal with partial page reads.  This emulates what
>  	 * do_mpage_read_folio does.
>  	 */
> -	bio = bio_alloc(iomap->bdev, bio_max_segs(nr_vecs), REQ_OP_READ, gfp);
> +	bio = bio_alloc_bioset(iomap->bdev, bio_max_segs(nr_vecs), REQ_OP_READ,
> +			gfp, bio_set);
>  	if (!bio)
> -		bio = bio_alloc(iomap->bdev, 1, REQ_OP_READ, orig_gfp);
> +		bio = bio_alloc_bioset(iomap->bdev, 1, REQ_OP_READ, orig_gfp,
> +				bio_set);
>  	if (ctx->rac)
>  		bio->bi_opf |= REQ_RAHEAD;
>  	bio->bi_iter.bi_sector = iomap_sector(iomap, iter->pos);
> diff --git a/include/linux/iomap.h b/include/linux/iomap.h
> index b3f545d41720..24f884b6b0c4 100644
> --- a/include/linux/iomap.h
> +++ b/include/linux/iomap.h
> @@ -515,6 +515,12 @@ struct iomap_read_ops {
>  	 */
>  	void (*submit_read)(const struct iomap_iter *iter,
>  			struct iomap_read_folio_ctx *ctx);
> +
> +	/*
> +	 * Optional, allows filesystem to specify own bio_set, so new bio's
> +	 * can be allocated from the provided bio_set.
> +	 */
> +	struct bio_set *bio_set;
>  };
>  
>  /*
> -- 
> 2.47.3
> 
> 

[PATCH 13/15] iomap: support ioends for buffered reads

[Christoph Hellwig]

Support using the ioend structure to defer I/O completion for
buffered reads by calling into the buffered read I/O completion
handler from iomap_finish_ioend.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/iomap/bio.c      | 19 ++++++++++++++++---
 fs/iomap/internal.h |  1 +
 fs/iomap/ioend.c    |  8 +++++---
 3 files changed, 22 insertions(+), 6 deletions(-)

diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
index 259a2bf95a43..b4de67bdd513 100644
--- a/fs/iomap/bio.c
+++ b/fs/iomap/bio.c
@@ -8,14 +8,27 @@
 #include "internal.h"
 #include "trace.h"
 
-static void iomap_read_end_io(struct bio *bio)
+static u32 __iomap_read_end_io(struct bio *bio, int error)
 {
-	int error = blk_status_to_errno(bio->bi_status);
 	struct folio_iter fi;
+	u32 folio_count = 0;
 
-	bio_for_each_folio_all(fi, bio)
+	bio_for_each_folio_all(fi, bio) {
 		iomap_finish_folio_read(fi.folio, fi.offset, fi.length, error);
+		folio_count++;
+	}
 	bio_put(bio);
+	return folio_count;
+}
+
+static void iomap_read_end_io(struct bio *bio)
+{
+	__iomap_read_end_io(bio, blk_status_to_errno(bio->bi_status));
+}
+
+u32 iomap_finish_ioend_buffered_read(struct iomap_ioend *ioend)
+{
+	return __iomap_read_end_io(&ioend->io_bio, ioend->io_error);
 }
 
 static void iomap_bio_submit_read(const struct iomap_iter *iter,
diff --git a/fs/iomap/internal.h b/fs/iomap/internal.h
index 3a4e4aad2bd1..b39dbc17e3f0 100644
--- a/fs/iomap/internal.h
+++ b/fs/iomap/internal.h
@@ -4,6 +4,7 @@
 
 #define IOEND_BATCH_SIZE	4096
 
+u32 iomap_finish_ioend_buffered_read(struct iomap_ioend *ioend);
 u32 iomap_finish_ioend_direct(struct iomap_ioend *ioend);
 
 #ifdef CONFIG_BLOCK
diff --git a/fs/iomap/ioend.c b/fs/iomap/ioend.c
index 800d12f45438..72f20e8c8893 100644
--- a/fs/iomap/ioend.c
+++ b/fs/iomap/ioend.c
@@ -36,7 +36,7 @@ EXPORT_SYMBOL_GPL(iomap_init_ioend);
  * state, release holds on bios, and finally free up memory.  Do not use the
  * ioend after this.
  */
-static u32 iomap_finish_ioend_buffered(struct iomap_ioend *ioend)
+static u32 iomap_finish_ioend_buffered_write(struct iomap_ioend *ioend)
 {
 	struct inode *inode = ioend->io_inode;
 	struct bio *bio = &ioend->io_bio;
@@ -68,7 +68,7 @@ static void ioend_writeback_end_bio(struct bio *bio)
 	struct iomap_ioend *ioend = iomap_ioend_from_bio(bio);
 
 	ioend->io_error = blk_status_to_errno(bio->bi_status);
-	iomap_finish_ioend_buffered(ioend);
+	iomap_finish_ioend_buffered_write(ioend);
 }
 
 /*
@@ -260,7 +260,9 @@ static u32 iomap_finish_ioend(struct iomap_ioend *ioend, int error)
 		return 0;
 	if (ioend->io_flags & IOMAP_IOEND_DIRECT)
 		return iomap_finish_ioend_direct(ioend);
-	return iomap_finish_ioend_buffered(ioend);
+	if (bio_op(&ioend->io_bio) == REQ_OP_READ)
+		return iomap_finish_ioend_buffered_read(ioend);
+	return iomap_finish_ioend_buffered_write(ioend);
 }
 
 /*
-- 
2.47.3

Re: [PATCH 13/15] iomap: support ioends for buffered reads

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:21AM +0100, Christoph Hellwig wrote:
> Support using the ioend structure to defer I/O completion for
> buffered reads by calling into the buffered read I/O completion
> handler from iomap_finish_ioend.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Seems simple enough...
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D

> ---
>  fs/iomap/bio.c      | 19 ++++++++++++++++---
>  fs/iomap/internal.h |  1 +
>  fs/iomap/ioend.c    |  8 +++++---
>  3 files changed, 22 insertions(+), 6 deletions(-)
> 
> diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
> index 259a2bf95a43..b4de67bdd513 100644
> --- a/fs/iomap/bio.c
> +++ b/fs/iomap/bio.c
> @@ -8,14 +8,27 @@
>  #include "internal.h"
>  #include "trace.h"
>  
> -static void iomap_read_end_io(struct bio *bio)
> +static u32 __iomap_read_end_io(struct bio *bio, int error)
>  {
> -	int error = blk_status_to_errno(bio->bi_status);
>  	struct folio_iter fi;
> +	u32 folio_count = 0;
>  
> -	bio_for_each_folio_all(fi, bio)
> +	bio_for_each_folio_all(fi, bio) {
>  		iomap_finish_folio_read(fi.folio, fi.offset, fi.length, error);
> +		folio_count++;
> +	}
>  	bio_put(bio);
> +	return folio_count;
> +}
> +
> +static void iomap_read_end_io(struct bio *bio)
> +{
> +	__iomap_read_end_io(bio, blk_status_to_errno(bio->bi_status));
> +}
> +
> +u32 iomap_finish_ioend_buffered_read(struct iomap_ioend *ioend)
> +{
> +	return __iomap_read_end_io(&ioend->io_bio, ioend->io_error);
>  }
>  
>  static void iomap_bio_submit_read(const struct iomap_iter *iter,
> diff --git a/fs/iomap/internal.h b/fs/iomap/internal.h
> index 3a4e4aad2bd1..b39dbc17e3f0 100644
> --- a/fs/iomap/internal.h
> +++ b/fs/iomap/internal.h
> @@ -4,6 +4,7 @@
>  
>  #define IOEND_BATCH_SIZE	4096
>  
> +u32 iomap_finish_ioend_buffered_read(struct iomap_ioend *ioend);
>  u32 iomap_finish_ioend_direct(struct iomap_ioend *ioend);
>  
>  #ifdef CONFIG_BLOCK
> diff --git a/fs/iomap/ioend.c b/fs/iomap/ioend.c
> index 800d12f45438..72f20e8c8893 100644
> --- a/fs/iomap/ioend.c
> +++ b/fs/iomap/ioend.c
> @@ -36,7 +36,7 @@ EXPORT_SYMBOL_GPL(iomap_init_ioend);
>   * state, release holds on bios, and finally free up memory.  Do not use the
>   * ioend after this.
>   */
> -static u32 iomap_finish_ioend_buffered(struct iomap_ioend *ioend)
> +static u32 iomap_finish_ioend_buffered_write(struct iomap_ioend *ioend)
>  {
>  	struct inode *inode = ioend->io_inode;
>  	struct bio *bio = &ioend->io_bio;
> @@ -68,7 +68,7 @@ static void ioend_writeback_end_bio(struct bio *bio)
>  	struct iomap_ioend *ioend = iomap_ioend_from_bio(bio);
>  
>  	ioend->io_error = blk_status_to_errno(bio->bi_status);
> -	iomap_finish_ioend_buffered(ioend);
> +	iomap_finish_ioend_buffered_write(ioend);
>  }
>  
>  /*
> @@ -260,7 +260,9 @@ static u32 iomap_finish_ioend(struct iomap_ioend *ioend, int error)
>  		return 0;
>  	if (ioend->io_flags & IOMAP_IOEND_DIRECT)
>  		return iomap_finish_ioend_direct(ioend);
> -	return iomap_finish_ioend_buffered(ioend);
> +	if (bio_op(&ioend->io_bio) == REQ_OP_READ)
> +		return iomap_finish_ioend_buffered_read(ioend);
> +	return iomap_finish_ioend_buffered_write(ioend);
>  }
>  
>  /*
> -- 
> 2.47.3
> 
> 

[PATCH 14/15] iomap: support T10 protection information

[Christoph Hellwig]

Add support for generating / verifying protection information in iomap.
This is done by hooking into the bio submission and then using the
generic PI helpers.  Compared to just using the block layer auto PI
this extends the protection envelope and also prepares for eventually
passing through PI from userspace at least for direct I/O.

To generate or verify PI, the file system needs to set the
IOMAP_F_INTEGRITY flag on the iomap for the request, and ensure the
ioends are used for all integrity I/O.  Additionally the file system
should defer read I/O completions to user context so that the guard
tag validation isn't run from interrupt context.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/iomap/bio.c        | 24 +++++++++++++++++++++---
 fs/iomap/direct-io.c  | 15 ++++++++++++++-
 fs/iomap/internal.h   | 13 +++++++++++++
 fs/iomap/ioend.c      | 20 ++++++++++++++++++--
 include/linux/iomap.h |  7 +++++++
 5 files changed, 73 insertions(+), 6 deletions(-)

diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
index b4de67bdd513..f989ffcaac96 100644
--- a/fs/iomap/bio.c
+++ b/fs/iomap/bio.c
@@ -3,6 +3,7 @@
  * Copyright (C) 2010 Red Hat, Inc.
  * Copyright (C) 2016-2023 Christoph Hellwig.
  */
+#include <linux/bio-integrity.h>
 #include <linux/iomap.h>
 #include <linux/pagemap.h>
 #include "internal.h"
@@ -17,6 +18,8 @@ static u32 __iomap_read_end_io(struct bio *bio, int error)
 		iomap_finish_folio_read(fi.folio, fi.offset, fi.length, error);
 		folio_count++;
 	}
+	if (bio_integrity(bio))
+		fs_bio_integrity_free(bio);
 	bio_put(bio);
 	return folio_count;
 }
@@ -34,7 +37,11 @@ u32 iomap_finish_ioend_buffered_read(struct iomap_ioend *ioend)
 static void iomap_bio_submit_read(const struct iomap_iter *iter,
 		struct iomap_read_folio_ctx *ctx)
 {
-	submit_bio(ctx->read_ctx);
+	struct bio *bio = ctx->read_ctx;
+
+	if (iter->iomap.flags & IOMAP_F_INTEGRITY)
+		fs_bio_integrity_alloc(bio);
+	submit_bio(bio);
 }
 
 static struct bio_set *iomap_read_bio_set(struct iomap_read_folio_ctx *ctx)
@@ -91,6 +98,7 @@ int iomap_bio_read_folio_range(const struct iomap_iter *iter,
 
 	if (!bio ||
 	    bio_end_sector(bio) != iomap_sector(&iter->iomap, iter->pos) ||
+	    bio->bi_iter.bi_size > iomap_max_bio_size(&iter->iomap) - plen ||
 	    !bio_add_folio(bio, folio, plen, offset_in_folio(folio, iter->pos)))
 		iomap_read_alloc_bio(iter, ctx, plen);
 	return 0;
@@ -107,11 +115,21 @@ int iomap_bio_read_folio_range_sync(const struct iomap_iter *iter,
 		struct folio *folio, loff_t pos, size_t len)
 {
 	const struct iomap *srcmap = iomap_iter_srcmap(iter);
+	sector_t sector = iomap_sector(srcmap, pos);
 	struct bio_vec bvec;
 	struct bio bio;
+	int error;
 
 	bio_init(&bio, srcmap->bdev, &bvec, 1, REQ_OP_READ);
-	bio.bi_iter.bi_sector = iomap_sector(srcmap, pos);
+	bio.bi_iter.bi_sector = sector;
 	bio_add_folio_nofail(&bio, folio, len, offset_in_folio(folio, pos));
-	return submit_bio_wait(&bio);
+	if (srcmap->flags & IOMAP_F_INTEGRITY)
+		fs_bio_integrity_alloc(&bio);
+	error = submit_bio_wait(&bio);
+	if (srcmap->flags & IOMAP_F_INTEGRITY) {
+		if (!error)
+			error = fs_bio_integrity_verify(&bio, sector, len);
+		fs_bio_integrity_free(&bio);
+	}
+	return error;
 }
diff --git a/fs/iomap/direct-io.c b/fs/iomap/direct-io.c
index 842fc7fecb2d..831378a6ced4 100644
--- a/fs/iomap/direct-io.c
+++ b/fs/iomap/direct-io.c
@@ -3,6 +3,7 @@
  * Copyright (C) 2010 Red Hat, Inc.
  * Copyright (c) 2016-2025 Christoph Hellwig.
  */
+#include <linux/bio-integrity.h>
 #include <linux/blk-crypto.h>
 #include <linux/fscrypt.h>
 #include <linux/pagemap.h>
@@ -215,6 +216,9 @@ static void __iomap_dio_bio_end_io(struct bio *bio, bool inline_completion)
 {
 	struct iomap_dio *dio = bio->bi_private;
 
+	if (bio_integrity(bio))
+		fs_bio_integrity_free(bio);
+
 	if (dio->flags & IOMAP_DIO_BOUNCE) {
 		bio_iov_iter_unbounce(bio, !!dio->error,
 				dio->flags & IOMAP_DIO_USER_BACKED);
@@ -325,8 +329,10 @@ static ssize_t iomap_dio_bio_iter_one(struct iomap_iter *iter,
 	bio->bi_private = dio;
 	bio->bi_end_io = iomap_dio_bio_end_io;
 
+
 	if (dio->flags & IOMAP_DIO_BOUNCE)
-		ret = bio_iov_iter_bounce(bio, dio->submit.iter, UINT_MAX);
+		ret = bio_iov_iter_bounce(bio, dio->submit.iter,
+				iomap_max_bio_size(&iter->iomap));
 	else
 		ret = bio_iov_iter_get_pages(bio, dio->submit.iter,
 					     alignment - 1);
@@ -343,6 +349,13 @@ static ssize_t iomap_dio_bio_iter_one(struct iomap_iter *iter,
 		goto out_put_bio;
 	}
 
+	if (iter->iomap.flags & IOMAP_F_INTEGRITY) {
+		if (dio->flags & IOMAP_DIO_WRITE)
+			fs_bio_integrity_generate(bio);
+		else
+			fs_bio_integrity_alloc(bio);
+	}
+
 	if (dio->flags & IOMAP_DIO_WRITE)
 		task_io_account_write(ret);
 	else if ((dio->flags & IOMAP_DIO_USER_BACKED) &&
diff --git a/fs/iomap/internal.h b/fs/iomap/internal.h
index b39dbc17e3f0..cfe63de9e5c7 100644
--- a/fs/iomap/internal.h
+++ b/fs/iomap/internal.h
@@ -4,6 +4,19 @@
 
 #define IOEND_BATCH_SIZE	4096
 
+/*
+ * Normally we can build bios as big as the data structure supports.
+ *
+ * But for integrity protected I/O we need to respect the maximum size of the
+ * single contiguous allocation for the integrity buffer.
+ */
+static inline size_t iomap_max_bio_size(const struct iomap *iomap)
+{
+	if (iomap->flags & IOMAP_F_INTEGRITY)
+		return max_integrity_io_size(bdev_limits(iomap->bdev));
+	return SIZE_MAX;
+}
+
 u32 iomap_finish_ioend_buffered_read(struct iomap_ioend *ioend);
 u32 iomap_finish_ioend_direct(struct iomap_ioend *ioend);
 
diff --git a/fs/iomap/ioend.c b/fs/iomap/ioend.c
index 72f20e8c8893..a2931f8c454c 100644
--- a/fs/iomap/ioend.c
+++ b/fs/iomap/ioend.c
@@ -2,6 +2,7 @@
 /*
  * Copyright (c) 2016-2025 Christoph Hellwig.
  */
+#include <linux/bio-integrity.h>
 #include <linux/iomap.h>
 #include <linux/list_sort.h>
 #include <linux/pagemap.h>
@@ -59,6 +60,8 @@ static u32 iomap_finish_ioend_buffered_write(struct iomap_ioend *ioend)
 		folio_count++;
 	}
 
+	if (bio_integrity(bio))
+		fs_bio_integrity_free(bio);
 	bio_put(bio);	/* frees the ioend */
 	return folio_count;
 }
@@ -92,6 +95,8 @@ int iomap_ioend_writeback_submit(struct iomap_writepage_ctx *wpc, int error)
 		return error;
 	}
 
+	if (wpc->iomap.flags & IOMAP_F_INTEGRITY)
+		fs_bio_integrity_generate(&ioend->io_bio);
 	submit_bio(&ioend->io_bio);
 	return 0;
 }
@@ -113,10 +118,13 @@ static struct iomap_ioend *iomap_alloc_ioend(struct iomap_writepage_ctx *wpc,
 }
 
 static bool iomap_can_add_to_ioend(struct iomap_writepage_ctx *wpc, loff_t pos,
-		u16 ioend_flags)
+		unsigned int map_len, u16 ioend_flags)
 {
 	struct iomap_ioend *ioend = wpc->wb_ctx;
 
+	if (ioend->io_bio.bi_iter.bi_size >
+	    iomap_max_bio_size(&wpc->iomap) - map_len)
+		return false;
 	if (ioend_flags & IOMAP_IOEND_BOUNDARY)
 		return false;
 	if ((ioend_flags & IOMAP_IOEND_NOMERGE_FLAGS) !=
@@ -181,7 +189,7 @@ ssize_t iomap_add_to_ioend(struct iomap_writepage_ctx *wpc, struct folio *folio,
 	if (pos == wpc->iomap.offset && (wpc->iomap.flags & IOMAP_F_BOUNDARY))
 		ioend_flags |= IOMAP_IOEND_BOUNDARY;
 
-	if (!ioend || !iomap_can_add_to_ioend(wpc, pos, ioend_flags)) {
+	if (!ioend || !iomap_can_add_to_ioend(wpc, pos, map_len, ioend_flags)) {
 new_ioend:
 		if (ioend) {
 			error = wpc->ops->writeback_submit(wpc, 0);
@@ -258,6 +266,14 @@ static u32 iomap_finish_ioend(struct iomap_ioend *ioend, int error)
 
 	if (!atomic_dec_and_test(&ioend->io_remaining))
 		return 0;
+
+	if (!ioend->io_error &&
+	    bio_integrity(&ioend->io_bio) &&
+	    bio_op(&ioend->io_bio) == REQ_OP_READ) {
+		ioend->io_error = fs_bio_integrity_verify(&ioend->io_bio,
+			ioend->io_sector, ioend->io_size);
+	}
+
 	if (ioend->io_flags & IOMAP_IOEND_DIRECT)
 		return iomap_finish_ioend_direct(ioend);
 	if (bio_op(&ioend->io_bio) == REQ_OP_READ)
diff --git a/include/linux/iomap.h b/include/linux/iomap.h
index 24f884b6b0c4..bde16d619654 100644
--- a/include/linux/iomap.h
+++ b/include/linux/iomap.h
@@ -65,6 +65,8 @@ struct vm_fault;
  *
  * IOMAP_F_ATOMIC_BIO indicates that (write) I/O will be issued as an atomic
  * bio, i.e. set REQ_ATOMIC.
+ *
+ * IOMAP_F_INTEGRITY indicates that the filesystems handles integrity metadata.
  */
 #define IOMAP_F_NEW		(1U << 0)
 #define IOMAP_F_DIRTY		(1U << 1)
@@ -79,6 +81,11 @@ struct vm_fault;
 #define IOMAP_F_BOUNDARY	(1U << 6)
 #define IOMAP_F_ANON_WRITE	(1U << 7)
 #define IOMAP_F_ATOMIC_BIO	(1U << 8)
+#ifdef CONFIG_BLK_DEV_INTEGRITY
+#define IOMAP_F_INTEGRITY	(1U << 9)
+#else
+#define IOMAP_F_INTEGRITY	0
+#endif /* CONFIG_BLK_DEV_INTEGRITY */
 
 /*
  * Flag reserved for file system specific usage
-- 
2.47.3

Re: [PATCH 14/15] iomap: support T10 protection information

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:22AM +0100, Christoph Hellwig wrote:
> Add support for generating / verifying protection information in iomap.
> This is done by hooking into the bio submission and then using the
> generic PI helpers.  Compared to just using the block layer auto PI
> this extends the protection envelope and also prepares for eventually
> passing through PI from userspace at least for direct I/O.
> 
> To generate or verify PI, the file system needs to set the
> IOMAP_F_INTEGRITY flag on the iomap for the request, and ensure the
> ioends are used for all integrity I/O.  Additionally the file system
> should defer read I/O completions to user context so that the guard

  must ?

> tag validation isn't run from interrupt context.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  fs/iomap/bio.c        | 24 +++++++++++++++++++++---
>  fs/iomap/direct-io.c  | 15 ++++++++++++++-
>  fs/iomap/internal.h   | 13 +++++++++++++
>  fs/iomap/ioend.c      | 20 ++++++++++++++++++--
>  include/linux/iomap.h |  7 +++++++
>  5 files changed, 73 insertions(+), 6 deletions(-)
> 
> diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
> index b4de67bdd513..f989ffcaac96 100644
> --- a/fs/iomap/bio.c
> +++ b/fs/iomap/bio.c
> @@ -3,6 +3,7 @@
>   * Copyright (C) 2010 Red Hat, Inc.
>   * Copyright (C) 2016-2023 Christoph Hellwig.
>   */
> +#include <linux/bio-integrity.h>
>  #include <linux/iomap.h>
>  #include <linux/pagemap.h>
>  #include "internal.h"
> @@ -17,6 +18,8 @@ static u32 __iomap_read_end_io(struct bio *bio, int error)
>  		iomap_finish_folio_read(fi.folio, fi.offset, fi.length, error);
>  		folio_count++;
>  	}
> +	if (bio_integrity(bio))
> +		fs_bio_integrity_free(bio);
>  	bio_put(bio);
>  	return folio_count;
>  }
> @@ -34,7 +37,11 @@ u32 iomap_finish_ioend_buffered_read(struct iomap_ioend *ioend)
>  static void iomap_bio_submit_read(const struct iomap_iter *iter,
>  		struct iomap_read_folio_ctx *ctx)
>  {
> -	submit_bio(ctx->read_ctx);
> +	struct bio *bio = ctx->read_ctx;
> +
> +	if (iter->iomap.flags & IOMAP_F_INTEGRITY)
> +		fs_bio_integrity_alloc(bio);
> +	submit_bio(bio);
>  }
>  
>  static struct bio_set *iomap_read_bio_set(struct iomap_read_folio_ctx *ctx)
> @@ -91,6 +98,7 @@ int iomap_bio_read_folio_range(const struct iomap_iter *iter,
>  
>  	if (!bio ||
>  	    bio_end_sector(bio) != iomap_sector(&iter->iomap, iter->pos) ||
> +	    bio->bi_iter.bi_size > iomap_max_bio_size(&iter->iomap) - plen ||
>  	    !bio_add_folio(bio, folio, plen, offset_in_folio(folio, iter->pos)))
>  		iomap_read_alloc_bio(iter, ctx, plen);
>  	return 0;
> @@ -107,11 +115,21 @@ int iomap_bio_read_folio_range_sync(const struct iomap_iter *iter,
>  		struct folio *folio, loff_t pos, size_t len)
>  {
>  	const struct iomap *srcmap = iomap_iter_srcmap(iter);
> +	sector_t sector = iomap_sector(srcmap, pos);
>  	struct bio_vec bvec;
>  	struct bio bio;
> +	int error;
>  
>  	bio_init(&bio, srcmap->bdev, &bvec, 1, REQ_OP_READ);
> -	bio.bi_iter.bi_sector = iomap_sector(srcmap, pos);
> +	bio.bi_iter.bi_sector = sector;
>  	bio_add_folio_nofail(&bio, folio, len, offset_in_folio(folio, pos));
> -	return submit_bio_wait(&bio);
> +	if (srcmap->flags & IOMAP_F_INTEGRITY)
> +		fs_bio_integrity_alloc(&bio);
> +	error = submit_bio_wait(&bio);
> +	if (srcmap->flags & IOMAP_F_INTEGRITY) {
> +		if (!error)
> +			error = fs_bio_integrity_verify(&bio, sector, len);
> +		fs_bio_integrity_free(&bio);
> +	}
> +	return error;
>  }
> diff --git a/fs/iomap/direct-io.c b/fs/iomap/direct-io.c
> index 842fc7fecb2d..831378a6ced4 100644
> --- a/fs/iomap/direct-io.c
> +++ b/fs/iomap/direct-io.c
> @@ -3,6 +3,7 @@
>   * Copyright (C) 2010 Red Hat, Inc.
>   * Copyright (c) 2016-2025 Christoph Hellwig.
>   */
> +#include <linux/bio-integrity.h>
>  #include <linux/blk-crypto.h>
>  #include <linux/fscrypt.h>
>  #include <linux/pagemap.h>
> @@ -215,6 +216,9 @@ static void __iomap_dio_bio_end_io(struct bio *bio, bool inline_completion)
>  {
>  	struct iomap_dio *dio = bio->bi_private;
>  
> +	if (bio_integrity(bio))
> +		fs_bio_integrity_free(bio);
> +
>  	if (dio->flags & IOMAP_DIO_BOUNCE) {
>  		bio_iov_iter_unbounce(bio, !!dio->error,
>  				dio->flags & IOMAP_DIO_USER_BACKED);
> @@ -325,8 +329,10 @@ static ssize_t iomap_dio_bio_iter_one(struct iomap_iter *iter,
>  	bio->bi_private = dio;
>  	bio->bi_end_io = iomap_dio_bio_end_io;
>  
> +
>  	if (dio->flags & IOMAP_DIO_BOUNCE)
> -		ret = bio_iov_iter_bounce(bio, dio->submit.iter, UINT_MAX);
> +		ret = bio_iov_iter_bounce(bio, dio->submit.iter,
> +				iomap_max_bio_size(&iter->iomap));
>  	else
>  		ret = bio_iov_iter_get_pages(bio, dio->submit.iter,
>  					     alignment - 1);
> @@ -343,6 +349,13 @@ static ssize_t iomap_dio_bio_iter_one(struct iomap_iter *iter,
>  		goto out_put_bio;
>  	}
>  
> +	if (iter->iomap.flags & IOMAP_F_INTEGRITY) {
> +		if (dio->flags & IOMAP_DIO_WRITE)
> +			fs_bio_integrity_generate(bio);
> +		else
> +			fs_bio_integrity_alloc(bio);
> +	}
> +
>  	if (dio->flags & IOMAP_DIO_WRITE)
>  		task_io_account_write(ret);
>  	else if ((dio->flags & IOMAP_DIO_USER_BACKED) &&
> diff --git a/fs/iomap/internal.h b/fs/iomap/internal.h
> index b39dbc17e3f0..cfe63de9e5c7 100644
> --- a/fs/iomap/internal.h
> +++ b/fs/iomap/internal.h
> @@ -4,6 +4,19 @@
>  
>  #define IOEND_BATCH_SIZE	4096
>  
> +/*
> + * Normally we can build bios as big as the data structure supports.
> + *
> + * But for integrity protected I/O we need to respect the maximum size of the
> + * single contiguous allocation for the integrity buffer.
> + */
> +static inline size_t iomap_max_bio_size(const struct iomap *iomap)
> +{
> +	if (iomap->flags & IOMAP_F_INTEGRITY)
> +		return max_integrity_io_size(bdev_limits(iomap->bdev));
> +	return SIZE_MAX;
> +}
> +
>  u32 iomap_finish_ioend_buffered_read(struct iomap_ioend *ioend);
>  u32 iomap_finish_ioend_direct(struct iomap_ioend *ioend);
>  
> diff --git a/fs/iomap/ioend.c b/fs/iomap/ioend.c
> index 72f20e8c8893..a2931f8c454c 100644
> --- a/fs/iomap/ioend.c
> +++ b/fs/iomap/ioend.c
> @@ -2,6 +2,7 @@
>  /*
>   * Copyright (c) 2016-2025 Christoph Hellwig.
>   */
> +#include <linux/bio-integrity.h>
>  #include <linux/iomap.h>
>  #include <linux/list_sort.h>
>  #include <linux/pagemap.h>
> @@ -59,6 +60,8 @@ static u32 iomap_finish_ioend_buffered_write(struct iomap_ioend *ioend)
>  		folio_count++;
>  	}
>  
> +	if (bio_integrity(bio))
> +		fs_bio_integrity_free(bio);
>  	bio_put(bio);	/* frees the ioend */
>  	return folio_count;
>  }
> @@ -92,6 +95,8 @@ int iomap_ioend_writeback_submit(struct iomap_writepage_ctx *wpc, int error)
>  		return error;
>  	}
>  
> +	if (wpc->iomap.flags & IOMAP_F_INTEGRITY)
> +		fs_bio_integrity_generate(&ioend->io_bio);
>  	submit_bio(&ioend->io_bio);
>  	return 0;
>  }
> @@ -113,10 +118,13 @@ static struct iomap_ioend *iomap_alloc_ioend(struct iomap_writepage_ctx *wpc,
>  }
>  
>  static bool iomap_can_add_to_ioend(struct iomap_writepage_ctx *wpc, loff_t pos,
> -		u16 ioend_flags)
> +		unsigned int map_len, u16 ioend_flags)
>  {
>  	struct iomap_ioend *ioend = wpc->wb_ctx;
>  
> +	if (ioend->io_bio.bi_iter.bi_size >
> +	    iomap_max_bio_size(&wpc->iomap) - map_len)
> +		return false;
>  	if (ioend_flags & IOMAP_IOEND_BOUNDARY)
>  		return false;
>  	if ((ioend_flags & IOMAP_IOEND_NOMERGE_FLAGS) !=

Unrelated question: should iomap_can_add_to_ioend return false if it did
an IOMAP_F_ANON_WRITE and the bdevs aren't the same, even if the sectors
match?  Currently not a problem for XFS, but some day we might want to
have a file that maps to zones on different devices.

The changes here otherwise look ok to me.

--D

> @@ -181,7 +189,7 @@ ssize_t iomap_add_to_ioend(struct iomap_writepage_ctx *wpc, struct folio *folio,
>  	if (pos == wpc->iomap.offset && (wpc->iomap.flags & IOMAP_F_BOUNDARY))
>  		ioend_flags |= IOMAP_IOEND_BOUNDARY;
>  
> -	if (!ioend || !iomap_can_add_to_ioend(wpc, pos, ioend_flags)) {
> +	if (!ioend || !iomap_can_add_to_ioend(wpc, pos, map_len, ioend_flags)) {
>  new_ioend:
>  		if (ioend) {
>  			error = wpc->ops->writeback_submit(wpc, 0);
> @@ -258,6 +266,14 @@ static u32 iomap_finish_ioend(struct iomap_ioend *ioend, int error)
>  
>  	if (!atomic_dec_and_test(&ioend->io_remaining))
>  		return 0;
> +
> +	if (!ioend->io_error &&
> +	    bio_integrity(&ioend->io_bio) &&
> +	    bio_op(&ioend->io_bio) == REQ_OP_READ) {
> +		ioend->io_error = fs_bio_integrity_verify(&ioend->io_bio,
> +			ioend->io_sector, ioend->io_size);
> +	}
> +
>  	if (ioend->io_flags & IOMAP_IOEND_DIRECT)
>  		return iomap_finish_ioend_direct(ioend);
>  	if (bio_op(&ioend->io_bio) == REQ_OP_READ)
> diff --git a/include/linux/iomap.h b/include/linux/iomap.h
> index 24f884b6b0c4..bde16d619654 100644
> --- a/include/linux/iomap.h
> +++ b/include/linux/iomap.h
> @@ -65,6 +65,8 @@ struct vm_fault;
>   *
>   * IOMAP_F_ATOMIC_BIO indicates that (write) I/O will be issued as an atomic
>   * bio, i.e. set REQ_ATOMIC.
> + *
> + * IOMAP_F_INTEGRITY indicates that the filesystems handles integrity metadata.
>   */
>  #define IOMAP_F_NEW		(1U << 0)
>  #define IOMAP_F_DIRTY		(1U << 1)
> @@ -79,6 +81,11 @@ struct vm_fault;
>  #define IOMAP_F_BOUNDARY	(1U << 6)
>  #define IOMAP_F_ANON_WRITE	(1U << 7)
>  #define IOMAP_F_ATOMIC_BIO	(1U << 8)
> +#ifdef CONFIG_BLK_DEV_INTEGRITY
> +#define IOMAP_F_INTEGRITY	(1U << 9)
> +#else
> +#define IOMAP_F_INTEGRITY	0
> +#endif /* CONFIG_BLK_DEV_INTEGRITY */
>  
>  /*
>   * Flag reserved for file system specific usage
> -- 
> 2.47.3
> 
> 

Re: [PATCH 14/15] iomap: support T10 protection information

[Christoph Hellwig]

On Wed, Jan 21, 2026 at 04:59:36PM -0800, Darrick J. Wong wrote:
> On Wed, Jan 21, 2026 at 07:43:22AM +0100, Christoph Hellwig wrote:
> > Add support for generating / verifying protection information in iomap.
> > This is done by hooking into the bio submission and then using the
> > generic PI helpers.  Compared to just using the block layer auto PI
> > this extends the protection envelope and also prepares for eventually
> > passing through PI from userspace at least for direct I/O.
> > 
> > To generate or verify PI, the file system needs to set the
> > IOMAP_F_INTEGRITY flag on the iomap for the request, and ensure the
> > ioends are used for all integrity I/O.  Additionally the file system
> > should defer read I/O completions to user context so that the guard
> 
>   must ?

Well, the copy isn't actually blocking.  So a small copy might actually
work from hardirq context, but you're not going to make friends with
anyone caring about latency.  I guess that means I should upgrade this
to a "must" :)

> >  {
> >  	struct iomap_ioend *ioend = wpc->wb_ctx;
> >  
> > +	if (ioend->io_bio.bi_iter.bi_size >
> > +	    iomap_max_bio_size(&wpc->iomap) - map_len)
> > +		return false;
> >  	if (ioend_flags & IOMAP_IOEND_BOUNDARY)
> >  		return false;
> >  	if ((ioend_flags & IOMAP_IOEND_NOMERGE_FLAGS) !=
> 
> Unrelated question: should iomap_can_add_to_ioend return false if it did
> an IOMAP_F_ANON_WRITE and the bdevs aren't the same, even if the sectors
> match?  Currently not a problem for XFS, but some day we might want to
> have a file that maps to zones on different devices.

For IOMAP_F_ANON_WRITE the bdev doesn't really matter at this level,
as it applies the actual mapping is done below.  So the bdev is really
just a placeholder here.

[PATCH 15/15] xfs: support T10 protection information

[Christoph Hellwig]

Add support for generating / verifying protection information in the file
system.  This is largely done by simply setting the IOMAP_F_INTEGRITY
flag and letting iomap do all of the work.  XFS just has to ensure that
the data read completions for integrity data are run from user context.

For zoned writeback, XFS also has to generate the integrity data itself
as the zoned writeback path is not using the generic writeback_submit
implementation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/xfs/xfs_aops.c  | 47 ++++++++++++++++++++++++++++++++++++++++++----
 fs/xfs/xfs_iomap.c |  9 ++++++---
 2 files changed, 49 insertions(+), 7 deletions(-)

diff --git a/fs/xfs/xfs_aops.c b/fs/xfs/xfs_aops.c
index c3c1e149fff4..4baf0a85271c 100644
--- a/fs/xfs/xfs_aops.c
+++ b/fs/xfs/xfs_aops.c
@@ -22,6 +22,7 @@
 #include "xfs_icache.h"
 #include "xfs_zone_alloc.h"
 #include "xfs_rtgroup.h"
+#include <linux/bio-integrity.h>
 
 struct xfs_writepage_ctx {
 	struct iomap_writepage_ctx ctx;
@@ -661,6 +662,8 @@ xfs_zoned_writeback_submit(
 		bio_endio(&ioend->io_bio);
 		return error;
 	}
+	if (wpc->iomap.flags & IOMAP_F_INTEGRITY)
+		fs_bio_integrity_generate(&ioend->io_bio);
 	xfs_zone_alloc_and_submit(ioend, &XFS_ZWPC(wpc)->open_zone);
 	return 0;
 }
@@ -741,12 +744,43 @@ xfs_vm_bmap(
 	return iomap_bmap(mapping, block, &xfs_read_iomap_ops);
 }
 
+static void
+xfs_bio_submit_read(
+	const struct iomap_iter		*iter,
+	struct iomap_read_folio_ctx	*ctx)
+{
+	struct bio			*bio = ctx->read_ctx;
+
+	/* delay read completions to the ioend workqueue */
+	iomap_init_ioend(iter->inode, bio, ctx->read_ctx_file_offset, 0);
+	bio->bi_end_io = xfs_end_bio;
+	submit_bio(bio);
+}
+
+static const struct iomap_read_ops xfs_bio_read_integrity_ops = {
+	.read_folio_range	= iomap_bio_read_folio_range,
+	.submit_read		= xfs_bio_submit_read,
+	.bio_set		= &iomap_ioend_bioset,
+};
+
+static inline const struct iomap_read_ops *bio_read_ops(struct xfs_inode *ip)
+{
+	if (bdev_has_integrity_csum(xfs_inode_buftarg(ip)->bt_bdev))
+		return &xfs_bio_read_integrity_ops;
+	return &iomap_bio_read_ops;
+}
+
 STATIC int
 xfs_vm_read_folio(
-	struct file		*unused,
-	struct folio		*folio)
+	struct file			*file,
+	struct folio			*folio)
 {
-	iomap_bio_read_folio(folio, &xfs_read_iomap_ops);
+	struct iomap_read_folio_ctx	ctx = {
+		.cur_folio	= folio,
+		.ops		= bio_read_ops(XFS_I(file->f_mapping->host)),
+	};
+
+	iomap_read_folio(&xfs_read_iomap_ops, &ctx);
 	return 0;
 }
 
@@ -754,7 +788,12 @@ STATIC void
 xfs_vm_readahead(
 	struct readahead_control	*rac)
 {
-	iomap_bio_readahead(rac, &xfs_read_iomap_ops);
+	struct iomap_read_folio_ctx	ctx = {
+		.rac		= rac,
+		.ops		= bio_read_ops(XFS_I(rac->mapping->host)),
+	};
+
+	iomap_readahead(&xfs_read_iomap_ops, &ctx);
 }
 
 static int
diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
index 37a1b33e9045..6ed784894b5a 100644
--- a/fs/xfs/xfs_iomap.c
+++ b/fs/xfs/xfs_iomap.c
@@ -143,11 +143,14 @@ xfs_bmbt_to_iomap(
 	}
 	iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
 	iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
-	if (mapping_flags & IOMAP_DAX)
+	iomap->flags = iomap_flags;
+	if (mapping_flags & IOMAP_DAX) {
 		iomap->dax_dev = target->bt_daxdev;
-	else
+	} else {
 		iomap->bdev = target->bt_bdev;
-	iomap->flags = iomap_flags;
+		if (bdev_has_integrity_csum(iomap->bdev))
+			iomap->flags |= IOMAP_F_INTEGRITY;
+	}
 
 	/*
 	 * If the inode is dirty for datasync purposes, let iomap know so it
-- 
2.47.3

Re: [PATCH 15/15] xfs: support T10 protection information

[Darrick J. Wong]

On Wed, Jan 21, 2026 at 07:43:23AM +0100, Christoph Hellwig wrote:
> Add support for generating / verifying protection information in the file
> system.  This is largely done by simply setting the IOMAP_F_INTEGRITY
> flag and letting iomap do all of the work.  XFS just has to ensure that
> the data read completions for integrity data are run from user context.
> 
> For zoned writeback, XFS also has to generate the integrity data itself
> as the zoned writeback path is not using the generic writeback_submit
> implementation.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  fs/xfs/xfs_aops.c  | 47 ++++++++++++++++++++++++++++++++++++++++++----
>  fs/xfs/xfs_iomap.c |  9 ++++++---
>  2 files changed, 49 insertions(+), 7 deletions(-)
> 
> diff --git a/fs/xfs/xfs_aops.c b/fs/xfs/xfs_aops.c
> index c3c1e149fff4..4baf0a85271c 100644
> --- a/fs/xfs/xfs_aops.c
> +++ b/fs/xfs/xfs_aops.c
> @@ -22,6 +22,7 @@
>  #include "xfs_icache.h"
>  #include "xfs_zone_alloc.h"
>  #include "xfs_rtgroup.h"
> +#include <linux/bio-integrity.h>
>  
>  struct xfs_writepage_ctx {
>  	struct iomap_writepage_ctx ctx;
> @@ -661,6 +662,8 @@ xfs_zoned_writeback_submit(
>  		bio_endio(&ioend->io_bio);
>  		return error;
>  	}
> +	if (wpc->iomap.flags & IOMAP_F_INTEGRITY)
> +		fs_bio_integrity_generate(&ioend->io_bio);
>  	xfs_zone_alloc_and_submit(ioend, &XFS_ZWPC(wpc)->open_zone);
>  	return 0;
>  }
> @@ -741,12 +744,43 @@ xfs_vm_bmap(
>  	return iomap_bmap(mapping, block, &xfs_read_iomap_ops);
>  }
>  
> +static void
> +xfs_bio_submit_read(
> +	const struct iomap_iter		*iter,
> +	struct iomap_read_folio_ctx	*ctx)
> +{
> +	struct bio			*bio = ctx->read_ctx;
> +
> +	/* delay read completions to the ioend workqueue */
> +	iomap_init_ioend(iter->inode, bio, ctx->read_ctx_file_offset, 0);
> +	bio->bi_end_io = xfs_end_bio;
> +	submit_bio(bio);
> +}
> +
> +static const struct iomap_read_ops xfs_bio_read_integrity_ops = {
> +	.read_folio_range	= iomap_bio_read_folio_range,
> +	.submit_read		= xfs_bio_submit_read,
> +	.bio_set		= &iomap_ioend_bioset,
> +};
> +
> +static inline const struct iomap_read_ops *bio_read_ops(struct xfs_inode *ip)

xfs_bio_read_ops to avoid conflicts with the block layer?

Otherwise this looks ok to me, so with that fixed,
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D


> +{
> +	if (bdev_has_integrity_csum(xfs_inode_buftarg(ip)->bt_bdev))
> +		return &xfs_bio_read_integrity_ops;
> +	return &iomap_bio_read_ops;
> +}
> +
>  STATIC int
>  xfs_vm_read_folio(
> -	struct file		*unused,
> -	struct folio		*folio)
> +	struct file			*file,
> +	struct folio			*folio)
>  {
> -	iomap_bio_read_folio(folio, &xfs_read_iomap_ops);
> +	struct iomap_read_folio_ctx	ctx = {
> +		.cur_folio	= folio,
> +		.ops		= bio_read_ops(XFS_I(file->f_mapping->host)),
> +	};
> +
> +	iomap_read_folio(&xfs_read_iomap_ops, &ctx);
>  	return 0;
>  }
>  
> @@ -754,7 +788,12 @@ STATIC void
>  xfs_vm_readahead(
>  	struct readahead_control	*rac)
>  {
> -	iomap_bio_readahead(rac, &xfs_read_iomap_ops);
> +	struct iomap_read_folio_ctx	ctx = {
> +		.rac		= rac,
> +		.ops		= bio_read_ops(XFS_I(rac->mapping->host)),
> +	};
> +
> +	iomap_readahead(&xfs_read_iomap_ops, &ctx);
>  }
>  
>  static int
> diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
> index 37a1b33e9045..6ed784894b5a 100644
> --- a/fs/xfs/xfs_iomap.c
> +++ b/fs/xfs/xfs_iomap.c
> @@ -143,11 +143,14 @@ xfs_bmbt_to_iomap(
>  	}
>  	iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
>  	iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
> -	if (mapping_flags & IOMAP_DAX)
> +	iomap->flags = iomap_flags;
> +	if (mapping_flags & IOMAP_DAX) {
>  		iomap->dax_dev = target->bt_daxdev;
> -	else
> +	} else {
>  		iomap->bdev = target->bt_bdev;
> -	iomap->flags = iomap_flags;
> +		if (bdev_has_integrity_csum(iomap->bdev))
> +			iomap->flags |= IOMAP_F_INTEGRITY;
> +	}
>  
>  	/*
>  	 * If the inode is dirty for datasync purposes, let iomap know so it
> -- 
> 2.47.3
> 
> 

Re: support file system generated / verified integrity information

[Anuj gupta]

Hi Christoph,

Here are the QD1 latency numbers (in usec)

Intel Optane:

Sequential read
  | size | xfs-bounce |  xfs-pi   |
  +------+------------+-----------+
  |   4k |    13.62   |    7.2    |
  |  64K |    99.66   |    34.16  |
  |   1M |    258.88  |    306.23 |
  +------+-------------------------+


Samsung PM1733

Sequential read
  | size | xfs-bounce |  xfs-pi   |
  +------+------------+-----------+
  |   4k |    118.92  |    91.6   |
  |  64K |    176.15  |    134.55 |
  |   1M |    612.67  |    584.84 |
  +------+-------------------------+


For sequential writes, I did not observe any noticeable difference,
which is expected.
The Optane numbers at bs=1M look a bit odd (xfs-pi slightly worse than
xfs-bounce), but they are reproducible across multiple runs.
Overall, the series shows a clear improvement in read performance,
which aligns well with the goal of making PI processing more efficient.

Feel free to add:
Tested-by: Anuj Gupta <anuj20.g@samsung.com>

Re: support file system generated / verified integrity information

[Christoph Hellwig]

On Tue, Jan 27, 2026 at 08:24:28PM +0530, Anuj gupta wrote:
> Hi Christoph,
> 
> Here are the QD1 latency numbers (in usec)

Thanks a lot!

Adding in the baseline numbers, as I wanted to compare those:

> Intel Optane:
> 
> Sequential read
>   | size | baseline | xfs-bounce |  xfs-pi  |
>   +------+----------+-----------+-----------+
>   |   4k |    7.18  |    13.62   |     7.20 |
>   |  64K |   36.40  |    99.66   |    34.16 |
>   |   1M |  206.38  |   258.88   |   306.23 |
>   +------+----------+------------+----------+

So for 4k and 64k reads we basically get back to the baseline.
The 1M numbers are puzzling, though.  I wonder if we need to
add WQ_CPU_INTENSIVE or do some other tweaks to the XFS I/O
completion workqueue so that we don't overload the scheduler.

Re: support file system generated / verified integrity information

[Anuj Gupta]

[-- Attachment #1: Type: text/plain, Size: 983 bytes --]

On 27/01/26 04:16PM, Christoph Hellwig wrote:
>On Tue, Jan 27, 2026 at 08:24:28PM +0530, Anuj gupta wrote:
>> Hi Christoph,
>>
>> Here are the QD1 latency numbers (in usec)
>
>Thanks a lot!
>
>Adding in the baseline numbers, as I wanted to compare those:
>
>> Intel Optane:
>>
>> Sequential read
>>   | size | baseline | xfs-bounce |  xfs-pi  |
>>   +------+----------+-----------+-----------+
>>   |   4k |    7.18  |    13.62   |     7.20 |
>>   |  64K |   36.40  |    99.66   |    34.16 |
>>   |   1M |  206.38  |   258.88   |   306.23 |
>>   +------+----------+------------+----------+
>
>So for 4k and 64k reads we basically get back to the baseline.
>The 1M numbers are puzzling, though.  I wonder if we need to
>add WQ_CPU_INTENSIVE or do some other tweaks to the XFS I/O
>completion workqueue so that we don't overload the scheduler.

Tried adding WQ_CPU_INTENSIVE[1] but didn't see any change in the 1M
latency. Looks like this needs something beyond just workqueue tuning.

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

support file system generated / verified integrity information v2

[Christoph Hellwig]

Hi all,

this series adds support to generate and verify integrity information
(aka T10 PI) in the file system, instead of the automatic below the
covers support that is currently used.

There two reasons for this:

  a) to increase the protection enveloped.  Right now this is just a
     minor step from the bottom of the block layer to the file system,
     but it is required to support io_uring integrity data passthrough in
     the file system similar to the currently existing support for block
     devices, which will follow next.  It also allows the file system to
     directly see the integrity error and act upon in, e.g. when using
     RAID either integrated (as in btrfs) or by supporting reading
     redundant copies through the block layer.
  b) to make the PI processing more efficient.  This is primarily a
     concern for reads, where the block layer auto PI has to schedule a
     work item for each bio, and the file system them has to do it again
     for bounce buffering.  Additionally the current iomap post-I/O
     workqueue handling is a lot more efficient by supporting merging and
     avoiding workqueue scheduling storms.

The implementation is based on refactoring the existing block layer PI
code to be reusable for this use case, and then adding relatively small
wrappers for the file system use case.  These are then used in iomap
to implement the semantics, and wired up in XFS with a small amount of
glue code.

Compared to the baseline (iomap-bounce branch), this does not change
performance for writes, but increases read performance up to 15% for 4k
I/O, with the benefit decreasing with larger I/O sizes as even the
baseline maxes out the device quickly on my older enterprise SSD.

Anuj gupta also measured a large decrease in QD1 latency on an
Intel optane device for small I/O sizes, but also an increase for
very large ones.

The series is based on the for-7.0/block-stable-pages in Jens's tee.

Git tree:

    git://git.infradead.org/users/hch/misc.git iomap-pi

Gitweb:

    https://git.infradead.org/?p=users/hch/misc.git;a=shortlog;h=refs/heads/iomap-pi

Changes since v1:
 - document usage of BI_ACT_*
 - return the action from fs_bio_integrity_alloc and use it in
   fs_bio_integrity_generation to be safe from misuse
 - use the newly added BIO_MAX_SIZE
 - rename bio_read_ops to xfs_bio_read_ops
 - fix commit message and comment typos

Diffstat:
 block/Makefile                |    2 
 block/bio-integrity-auto.c    |   80 +++---------------------
 block/bio-integrity-fs.c      |   81 +++++++++++++++++++++++++
 block/bio-integrity.c         |   64 +++++++++++++++++++
 block/bio.c                   |   17 +++--
 block/blk-mq.c                |    6 +
 block/blk-settings.c          |   13 ----
 block/blk.h                   |    6 +
 block/t10-pi.c                |   12 +--
 drivers/nvdimm/btt.c          |    6 +
 fs/fuse/file.c                |    5 -
 fs/iomap/bio.c                |  135 ++++++++++++++++++++++++++++--------------
 fs/iomap/buffered-io.c        |    8 +-
 fs/iomap/direct-io.c          |   15 ++++
 fs/iomap/internal.h           |   14 ++++
 fs/iomap/ioend.c              |   28 +++++++-
 fs/xfs/xfs_aops.c             |   49 ++++++++++++++-
 fs/xfs/xfs_iomap.c            |    9 +-
 include/linux/bio-integrity.h |   12 ++-
 include/linux/bio.h           |    2 
 include/linux/blk-integrity.h |   28 +++++++-
 include/linux/blkdev.h        |   34 ++++++++--
 include/linux/iomap.h         |   20 +++++-
 23 files changed, 465 insertions(+), 181 deletions(-)

[PATCH 11/15] iomap: allow file systems to hook into buffered read bio submission

[Christoph Hellwig]

File systems such as btrfs have additional operations with bios such as
verifying data checksums.  Allow file systems to hook into submission
of the bio to allow for this processing by replacing the direct
submit_bio call in iomap_read_alloc_bio with a call into ->submit_read
and exporting iomap_read_alloc_bio.  Also add a new field to
struct iomap_read_folio_ctx to track the file logic offset of the current
read context.

Based on a patch from Goldwyn Rodrigues <rgoldwyn@suse.com>.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Tested-by: Anuj Gupta <anuj20.g@samsung.com>
---
 fs/iomap/bio.c        | 15 +++++++++------
 include/linux/iomap.h |  4 ++++
 2 files changed, 13 insertions(+), 6 deletions(-)

diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
index 80bbd328bd3c..903cb9fe759e 100644
--- a/fs/iomap/bio.c
+++ b/fs/iomap/bio.c
@@ -32,10 +32,11 @@ static void iomap_read_alloc_bio(const struct iomap_iter *iter,
 	struct folio *folio = ctx->cur_folio;
 	gfp_t gfp = mapping_gfp_constraint(folio->mapping, GFP_KERNEL);
 	gfp_t orig_gfp = gfp;
-	struct bio *bio = ctx->read_ctx;
+	struct bio *bio;
 
-	if (bio)
-		submit_bio(bio);
+	/* Submit the existing range if there was one. */
+	if (ctx->read_ctx)
+		ctx->ops->submit_read(iter, ctx);
 
 	/* Same as readahead_gfp_mask: */
 	if (ctx->rac)
@@ -56,9 +57,10 @@ static void iomap_read_alloc_bio(const struct iomap_iter *iter,
 	bio_add_folio_nofail(bio, folio, plen,
 			offset_in_folio(folio, iter->pos));
 	ctx->read_ctx = bio;
+	ctx->read_ctx_file_offset = iter->pos;
 }
 
-static int iomap_bio_read_folio_range(const struct iomap_iter *iter,
+int iomap_bio_read_folio_range(const struct iomap_iter *iter,
 		struct iomap_read_folio_ctx *ctx, size_t plen)
 {
 	struct folio *folio = ctx->cur_folio;
@@ -70,10 +72,11 @@ static int iomap_bio_read_folio_range(const struct iomap_iter *iter,
 		iomap_read_alloc_bio(iter, ctx, plen);
 	return 0;
 }
+EXPORT_SYMBOL_GPL(iomap_bio_read_folio_range);
 
 const struct iomap_read_ops iomap_bio_read_ops = {
-	.read_folio_range = iomap_bio_read_folio_range,
-	.submit_read = iomap_bio_submit_read,
+	.read_folio_range	= iomap_bio_read_folio_range,
+	.submit_read		= iomap_bio_submit_read,
 };
 EXPORT_SYMBOL_GPL(iomap_bio_read_ops);
 
diff --git a/include/linux/iomap.h b/include/linux/iomap.h
index 673d7bf8803f..42d562e55d3d 100644
--- a/include/linux/iomap.h
+++ b/include/linux/iomap.h
@@ -489,6 +489,7 @@ struct iomap_read_folio_ctx {
 	struct folio		*cur_folio;
 	struct readahead_control *rac;
 	void			*read_ctx;
+	loff_t			read_ctx_file_offset;
 };
 
 struct iomap_read_ops {
@@ -595,6 +596,9 @@ int iomap_swapfile_activate(struct swap_info_struct *sis,
 extern struct bio_set iomap_ioend_bioset;
 
 #ifdef CONFIG_BLOCK
+int iomap_bio_read_folio_range(const struct iomap_iter *iter,
+		struct iomap_read_folio_ctx *ctx, size_t plen);
+
 extern const struct iomap_read_ops iomap_bio_read_ops;
 
 static inline void iomap_bio_read_folio(struct folio *folio,
-- 
2.47.3

support file system generated / verified integrity information v3

[Christoph Hellwig]

Hi all,

this series adds support to generate and verify integrity information
(aka T10 PI) in the file system, instead of the automatic below the
covers support that is currently used.

There two reasons for this:

  a) to increase the protection enveloped.  Right now this is just a
     minor step from the bottom of the block layer to the file system,
     but it is required to support io_uring integrity data passthrough in
     the file system similar to the currently existing support for block
     devices, which will follow next.  It also allows the file system to
     directly see the integrity error and act upon in, e.g. when using
     RAID either integrated (as in btrfs) or by supporting reading
     redundant copies through the block layer.
  b) to make the PI processing more efficient.  This is primarily a
     concern for reads, where the block layer auto PI has to schedule a
     work item for each bio, and the file system them has to do it again
     for bounce buffering.  Additionally the current iomap post-I/O
     workqueue handling is a lot more efficient by supporting merging and
     avoiding workqueue scheduling storms.

The implementation is based on refactoring the existing block layer PI
code to be reusable for this use case, and then adding relatively small
wrappers for the file system use case.  These are then used in iomap
to implement the semantics, and wired up in XFS with a small amount of
glue code.

Compared to the baseline (iomap-bounce branch), this does not change
performance for writes, but increases read performance up to 15% for 4k
I/O, with the benefit decreasing with larger I/O sizes as even the
baseline maxes out the device quickly on my older enterprise SSD.

Anuj Gupta also measured a large decrease in QD1 latency on an Intel
Optane device for small I/O sizes, but also an increase for very large
ones.

Note that the upcoming XFS fsverity support also depends on some
infrastructure in this series.

Git tree:

    git://git.infradead.org/users/hch/misc.git iomap-pi

Gitweb:

    https://git.infradead.org/?p=users/hch/misc.git;a=shortlog;h=refs/heads/iomap-pi

Changes since v2:
 - rebased to current linus tree
 - use folio->mapping instead of file->f_mapping to support the upcoming
   fsverity use case
 - rename and slightly refactor the xfs iomap read ops to better addres
   that this will be used by fsverity as well.

Changes since v1:
 - document usage of BI_ACT_*
 - return the action from fs_bio_integrity_alloc and use it in
   fs_bio_integrity_generation to be safe from misuse
 - use the newly added BIO_MAX_SIZE
 - rename bio_read_ops to xfs_bio_read_ops
 - fix commit message and comment typos

Diffstat:
 block/Makefile                |    2 
 block/bio-integrity-auto.c    |   80 +++---------------------
 block/bio-integrity-fs.c      |   81 +++++++++++++++++++++++++
 block/bio-integrity.c         |   64 +++++++++++++++++++
 block/bio.c                   |   17 +++--
 block/blk-mq.c                |    6 +
 block/blk-settings.c          |   13 ----
 block/blk.h                   |    6 +
 block/t10-pi.c                |   12 +--
 drivers/nvdimm/btt.c          |    6 +
 fs/fuse/file.c                |    5 -
 fs/iomap/bio.c                |  135 ++++++++++++++++++++++++++++--------------
 fs/iomap/buffered-io.c        |    8 +-
 fs/iomap/direct-io.c          |   15 ++++
 fs/iomap/internal.h           |   14 ++++
 fs/iomap/ioend.c              |   28 +++++++-
 fs/xfs/xfs_aops.c             |   47 +++++++++++++-
 fs/xfs/xfs_iomap.c            |    9 +-
 include/linux/bio-integrity.h |   12 ++-
 include/linux/bio.h           |    2 
 include/linux/blk-integrity.h |   28 +++++++-
 include/linux/blkdev.h        |   34 ++++++++--
 include/linux/iomap.h         |   20 +++++-
 23 files changed, 463 insertions(+), 181 deletions(-)

[PATCH 11/15] iomap: allow file systems to hook into buffered read bio submission

[Christoph Hellwig]

File systems such as btrfs have additional operations with bios such as
verifying data checksums.  Allow file systems to hook into submission
of the bio to allow for this processing by replacing the direct
submit_bio call in iomap_read_alloc_bio with a call into ->submit_read
and exporting iomap_read_alloc_bio.  Also add a new field to
struct iomap_read_folio_ctx to track the file logic offset of the current
read context.

Based on a patch from Goldwyn Rodrigues <rgoldwyn@suse.com>.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Tested-by: Anuj Gupta <anuj20.g@samsung.com>
---
 fs/iomap/bio.c        | 15 +++++++++------
 include/linux/iomap.h |  4 ++++
 2 files changed, 13 insertions(+), 6 deletions(-)

diff --git a/fs/iomap/bio.c b/fs/iomap/bio.c
index 80bbd328bd3c..903cb9fe759e 100644
--- a/fs/iomap/bio.c
+++ b/fs/iomap/bio.c
@@ -32,10 +32,11 @@ static void iomap_read_alloc_bio(const struct iomap_iter *iter,
 	struct folio *folio = ctx->cur_folio;
 	gfp_t gfp = mapping_gfp_constraint(folio->mapping, GFP_KERNEL);
 	gfp_t orig_gfp = gfp;
-	struct bio *bio = ctx->read_ctx;
+	struct bio *bio;
 
-	if (bio)
-		submit_bio(bio);
+	/* Submit the existing range if there was one. */
+	if (ctx->read_ctx)
+		ctx->ops->submit_read(iter, ctx);
 
 	/* Same as readahead_gfp_mask: */
 	if (ctx->rac)
@@ -56,9 +57,10 @@ static void iomap_read_alloc_bio(const struct iomap_iter *iter,
 	bio_add_folio_nofail(bio, folio, plen,
 			offset_in_folio(folio, iter->pos));
 	ctx->read_ctx = bio;
+	ctx->read_ctx_file_offset = iter->pos;
 }
 
-static int iomap_bio_read_folio_range(const struct iomap_iter *iter,
+int iomap_bio_read_folio_range(const struct iomap_iter *iter,
 		struct iomap_read_folio_ctx *ctx, size_t plen)
 {
 	struct folio *folio = ctx->cur_folio;
@@ -70,10 +72,11 @@ static int iomap_bio_read_folio_range(const struct iomap_iter *iter,
 		iomap_read_alloc_bio(iter, ctx, plen);
 	return 0;
 }
+EXPORT_SYMBOL_GPL(iomap_bio_read_folio_range);
 
 const struct iomap_read_ops iomap_bio_read_ops = {
-	.read_folio_range = iomap_bio_read_folio_range,
-	.submit_read = iomap_bio_submit_read,
+	.read_folio_range	= iomap_bio_read_folio_range,
+	.submit_read		= iomap_bio_submit_read,
 };
 EXPORT_SYMBOL_GPL(iomap_bio_read_ops);
 
diff --git a/include/linux/iomap.h b/include/linux/iomap.h
index 6fbe121e2adf..b2b9e649a3b8 100644
--- a/include/linux/iomap.h
+++ b/include/linux/iomap.h
@@ -493,6 +493,7 @@ struct iomap_read_folio_ctx {
 	struct folio		*cur_folio;
 	struct readahead_control *rac;
 	void			*read_ctx;
+	loff_t			read_ctx_file_offset;
 };
 
 struct iomap_read_ops {
@@ -599,6 +600,9 @@ int iomap_swapfile_activate(struct swap_info_struct *sis,
 extern struct bio_set iomap_ioend_bioset;
 
 #ifdef CONFIG_BLOCK
+int iomap_bio_read_folio_range(const struct iomap_iter *iter,
+		struct iomap_read_folio_ctx *ctx, size_t plen);
+
 extern const struct iomap_read_ops iomap_bio_read_ops;
 
 static inline void iomap_bio_read_folio(struct folio *folio,
-- 
2.47.3