[PATCH v2 0/3] Block integrity with flexible-offset PI

[PATCH v2 0/3] Block integrity with flexible-offset PI

[Kanchan Joshi]

The block integrity subsystem can only work with PI placed in the first
bytes of the metadata buffer.

The series makes block-integrity support the flexible placement of PI.
And changes NVMe driver to make use of the new capability.

This helps to
(i) enable the more common case for NVMe (PI in last bytes is the norm)
(ii) reduce nop profile users (tried by Jens recently [1]).

/* For NS 4K+16b, 8b PI, last bytes */
Before:
# cat /sys/block/nvme0n1/integrity/format
nop

After:
# cat /sys/block/nvme0n1/integrity/format
T10-DIF-TYPE1-CRC

Changes since v1:
- Reworded commit messsage in patch 2 and 3 (hch)
- Variable initialization order change (hch)
- Collect reviewed-by

[1] https://lore.kernel.org/linux-block/20240111160226.1936351-1-axboe@kernel.dk/



Kanchan Joshi (3):
  block: refactor guard helpers
  block: support PI at non-zero offset within metadata
  nvme: allow integrity when PI is not in first bytes

 block/bio-integrity.c         |  1 +
 block/blk-integrity.c         |  1 +
 block/t10-pi.c                | 72 +++++++++++++++++++++++------------
 drivers/nvme/host/core.c      |  8 +++-
 drivers/nvme/host/nvme.h      |  1 +
 include/linux/blk-integrity.h |  1 +
 include/linux/blkdev.h        |  1 +
 7 files changed, 60 insertions(+), 25 deletions(-)

-- 
2.25.1

[PATCH v2 1/3] block: refactor guard helpers

[Kanchan Joshi]

Allow computation using the existing guard value.
This is a prep patch.

Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
---
 block/t10-pi.c | 20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

diff --git a/block/t10-pi.c b/block/t10-pi.c
index 914d8cddd43a..251a7b188963 100644
--- a/block/t10-pi.c
+++ b/block/t10-pi.c
@@ -12,14 +12,14 @@
 #include <net/checksum.h>
 #include <asm/unaligned.h>
 
-typedef __be16 (csum_fn) (void *, unsigned int);
+typedef __be16 (csum_fn) (__be16, void *, unsigned int);
 
-static __be16 t10_pi_crc_fn(void *data, unsigned int len)
+static __be16 t10_pi_crc_fn(__be16 crc, void *data, unsigned int len)
 {
-	return cpu_to_be16(crc_t10dif(data, len));
+	return cpu_to_be16(crc_t10dif_update(be16_to_cpu(crc), data, len));
 }
 
-static __be16 t10_pi_ip_fn(void *data, unsigned int len)
+static __be16 t10_pi_ip_fn(__be16 csum, void *data, unsigned int len)
 {
 	return (__force __be16)ip_compute_csum(data, len);
 }
@@ -37,7 +37,7 @@ static blk_status_t t10_pi_generate(struct blk_integrity_iter *iter,
 	for (i = 0 ; i < iter->data_size ; i += iter->interval) {
 		struct t10_pi_tuple *pi = iter->prot_buf;
 
-		pi->guard_tag = fn(iter->data_buf, iter->interval);
+		pi->guard_tag = fn(0, iter->data_buf, iter->interval);
 		pi->app_tag = 0;
 
 		if (type == T10_PI_TYPE1_PROTECTION)
@@ -83,7 +83,7 @@ static blk_status_t t10_pi_verify(struct blk_integrity_iter *iter,
 				goto next;
 		}
 
-		csum = fn(iter->data_buf, iter->interval);
+		csum = fn(0, iter->data_buf, iter->interval);
 
 		if (pi->guard_tag != csum) {
 			pr_err("%s: guard tag error at sector %llu " \
@@ -280,9 +280,9 @@ const struct blk_integrity_profile t10_pi_type3_ip = {
 };
 EXPORT_SYMBOL(t10_pi_type3_ip);
 
-static __be64 ext_pi_crc64(void *data, unsigned int len)
+static __be64 ext_pi_crc64(u64 crc, void *data, unsigned int len)
 {
-	return cpu_to_be64(crc64_rocksoft(data, len));
+	return cpu_to_be64(crc64_rocksoft_update(crc, data, len));
 }
 
 static blk_status_t ext_pi_crc64_generate(struct blk_integrity_iter *iter,
@@ -293,7 +293,7 @@ static blk_status_t ext_pi_crc64_generate(struct blk_integrity_iter *iter,
 	for (i = 0 ; i < iter->data_size ; i += iter->interval) {
 		struct crc64_pi_tuple *pi = iter->prot_buf;
 
-		pi->guard_tag = ext_pi_crc64(iter->data_buf, iter->interval);
+		pi->guard_tag = ext_pi_crc64(0, iter->data_buf, iter->interval);
 		pi->app_tag = 0;
 
 		if (type == T10_PI_TYPE1_PROTECTION)
@@ -343,7 +343,7 @@ static blk_status_t ext_pi_crc64_verify(struct blk_integrity_iter *iter,
 				goto next;
 		}
 
-		csum = ext_pi_crc64(iter->data_buf, iter->interval);
+		csum = ext_pi_crc64(0, iter->data_buf, iter->interval);
 		if (pi->guard_tag != csum) {
 			pr_err("%s: guard tag error at sector %llu " \
 			       "(rcvd %016llx, want %016llx)\n",
-- 
2.25.1

[PATCH v2 2/3] block: support PI at non-zero offset within metadata

[Kanchan Joshi]

Block layer integrity processing assumes that protection information
(PI) is placed in the first bytes of each metadata block.

Remove this limitation and include the metadata before the PI in the
calculation of the guard tag.

Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
Signed-off-by: Chinmay Gameti <c.gameti@samsung.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
---
 block/bio-integrity.c         |  1 +
 block/blk-integrity.c         |  1 +
 block/t10-pi.c                | 52 +++++++++++++++++++++++++----------
 include/linux/blk-integrity.h |  1 +
 include/linux/blkdev.h        |  1 +
 5 files changed, 42 insertions(+), 14 deletions(-)

diff --git a/block/bio-integrity.c b/block/bio-integrity.c
index c9a16fba58b9..2e3e8e04961e 100644
--- a/block/bio-integrity.c
+++ b/block/bio-integrity.c
@@ -395,6 +395,7 @@ static blk_status_t bio_integrity_process(struct bio *bio,
 	iter.tuple_size = bi->tuple_size;
 	iter.seed = proc_iter->bi_sector;
 	iter.prot_buf = bvec_virt(bip->bip_vec);
+	iter.pi_offset = bi->pi_offset;
 
 	__bio_for_each_segment(bv, bio, bviter, *proc_iter) {
 		void *kaddr = bvec_kmap_local(&bv);
diff --git a/block/blk-integrity.c b/block/blk-integrity.c
index d4e9b4556d14..ccbeb6dfa87a 100644
--- a/block/blk-integrity.c
+++ b/block/blk-integrity.c
@@ -370,6 +370,7 @@ void blk_integrity_register(struct gendisk *disk, struct blk_integrity *template
 	bi->profile = template->profile ? template->profile : &nop_profile;
 	bi->tuple_size = template->tuple_size;
 	bi->tag_size = template->tag_size;
+	bi->pi_offset = template->pi_offset;
 
 	blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, disk->queue);
 
diff --git a/block/t10-pi.c b/block/t10-pi.c
index 251a7b188963..d90892fd6f2a 100644
--- a/block/t10-pi.c
+++ b/block/t10-pi.c
@@ -32,12 +32,16 @@ static __be16 t10_pi_ip_fn(__be16 csum, void *data, unsigned int len)
 static blk_status_t t10_pi_generate(struct blk_integrity_iter *iter,
 		csum_fn *fn, enum t10_dif_type type)
 {
+	u8 offset = iter->pi_offset;
 	unsigned int i;
 
 	for (i = 0 ; i < iter->data_size ; i += iter->interval) {
-		struct t10_pi_tuple *pi = iter->prot_buf;
+		struct t10_pi_tuple *pi = iter->prot_buf + offset;
 
 		pi->guard_tag = fn(0, iter->data_buf, iter->interval);
+		if (offset)
+			pi->guard_tag = fn(pi->guard_tag, iter->prot_buf,
+					   offset);
 		pi->app_tag = 0;
 
 		if (type == T10_PI_TYPE1_PROTECTION)
@@ -56,12 +60,13 @@ static blk_status_t t10_pi_generate(struct blk_integrity_iter *iter,
 static blk_status_t t10_pi_verify(struct blk_integrity_iter *iter,
 		csum_fn *fn, enum t10_dif_type type)
 {
+	u8 offset = iter->pi_offset;
 	unsigned int i;
 
 	BUG_ON(type == T10_PI_TYPE0_PROTECTION);
 
 	for (i = 0 ; i < iter->data_size ; i += iter->interval) {
-		struct t10_pi_tuple *pi = iter->prot_buf;
+		struct t10_pi_tuple *pi = iter->prot_buf + offset;
 		__be16 csum;
 
 		if (type == T10_PI_TYPE1_PROTECTION ||
@@ -84,6 +89,8 @@ static blk_status_t t10_pi_verify(struct blk_integrity_iter *iter,
 		}
 
 		csum = fn(0, iter->data_buf, iter->interval);
+		if (offset)
+			csum = fn(csum, iter->prot_buf, offset);
 
 		if (pi->guard_tag != csum) {
 			pr_err("%s: guard tag error at sector %llu " \
@@ -134,8 +141,10 @@ static blk_status_t t10_pi_type1_verify_ip(struct blk_integrity_iter *iter)
  */
 static void t10_pi_type1_prepare(struct request *rq)
 {
-	const int tuple_sz = rq->q->integrity.tuple_size;
+	struct blk_integrity *bi = &rq->q->integrity;
+	const int tuple_sz = bi->tuple_size;
 	u32 ref_tag = t10_pi_ref_tag(rq);
+	u8 offset = bi->pi_offset;
 	struct bio *bio;
 
 	__rq_for_each_bio(bio, rq) {
@@ -154,7 +163,7 @@ static void t10_pi_type1_prepare(struct request *rq)
 
 			p = bvec_kmap_local(&iv);
 			for (j = 0; j < iv.bv_len; j += tuple_sz) {
-				struct t10_pi_tuple *pi = p;
+				struct t10_pi_tuple *pi = p + offset;
 
 				if (be32_to_cpu(pi->ref_tag) == virt)
 					pi->ref_tag = cpu_to_be32(ref_tag);
@@ -183,9 +192,11 @@ static void t10_pi_type1_prepare(struct request *rq)
  */
 static void t10_pi_type1_complete(struct request *rq, unsigned int nr_bytes)
 {
-	unsigned intervals = nr_bytes >> rq->q->integrity.interval_exp;
-	const int tuple_sz = rq->q->integrity.tuple_size;
+	struct blk_integrity *bi = &rq->q->integrity;
+	unsigned intervals = nr_bytes >> bi->interval_exp;
+	const int tuple_sz = bi->tuple_size;
 	u32 ref_tag = t10_pi_ref_tag(rq);
+	u8 offset = bi->pi_offset;
 	struct bio *bio;
 
 	__rq_for_each_bio(bio, rq) {
@@ -200,7 +211,7 @@ static void t10_pi_type1_complete(struct request *rq, unsigned int nr_bytes)
 
 			p = bvec_kmap_local(&iv);
 			for (j = 0; j < iv.bv_len && intervals; j += tuple_sz) {
-				struct t10_pi_tuple *pi = p;
+				struct t10_pi_tuple *pi = p + offset;
 
 				if (be32_to_cpu(pi->ref_tag) == ref_tag)
 					pi->ref_tag = cpu_to_be32(virt);
@@ -288,12 +299,16 @@ static __be64 ext_pi_crc64(u64 crc, void *data, unsigned int len)
 static blk_status_t ext_pi_crc64_generate(struct blk_integrity_iter *iter,
 					enum t10_dif_type type)
 {
+	u8 offset = iter->pi_offset;
 	unsigned int i;
 
 	for (i = 0 ; i < iter->data_size ; i += iter->interval) {
-		struct crc64_pi_tuple *pi = iter->prot_buf;
+		struct crc64_pi_tuple *pi = iter->prot_buf + offset;
 
 		pi->guard_tag = ext_pi_crc64(0, iter->data_buf, iter->interval);
+		if (offset)
+			pi->guard_tag = ext_pi_crc64(be64_to_cpu(pi->guard_tag),
+					iter->prot_buf, offset);
 		pi->app_tag = 0;
 
 		if (type == T10_PI_TYPE1_PROTECTION)
@@ -319,10 +334,11 @@ static bool ext_pi_ref_escape(u8 *ref_tag)
 static blk_status_t ext_pi_crc64_verify(struct blk_integrity_iter *iter,
 				      enum t10_dif_type type)
 {
+	u8 offset = iter->pi_offset;
 	unsigned int i;
 
 	for (i = 0; i < iter->data_size; i += iter->interval) {
-		struct crc64_pi_tuple *pi = iter->prot_buf;
+		struct crc64_pi_tuple *pi = iter->prot_buf + offset;
 		u64 ref, seed;
 		__be64 csum;
 
@@ -344,6 +360,10 @@ static blk_status_t ext_pi_crc64_verify(struct blk_integrity_iter *iter,
 		}
 
 		csum = ext_pi_crc64(0, iter->data_buf, iter->interval);
+		if (offset)
+			csum = ext_pi_crc64(be64_to_cpu(csum), iter->prot_buf,
+					    offset);
+
 		if (pi->guard_tag != csum) {
 			pr_err("%s: guard tag error at sector %llu " \
 			       "(rcvd %016llx, want %016llx)\n",
@@ -373,8 +393,10 @@ static blk_status_t ext_pi_type1_generate_crc64(struct blk_integrity_iter *iter)
 
 static void ext_pi_type1_prepare(struct request *rq)
 {
-	const int tuple_sz = rq->q->integrity.tuple_size;
+	struct blk_integrity *bi = &rq->q->integrity;
+	const int tuple_sz = bi->tuple_size;
 	u64 ref_tag = ext_pi_ref_tag(rq);
+	u8 offset = bi->pi_offset;
 	struct bio *bio;
 
 	__rq_for_each_bio(bio, rq) {
@@ -393,7 +415,7 @@ static void ext_pi_type1_prepare(struct request *rq)
 
 			p = bvec_kmap_local(&iv);
 			for (j = 0; j < iv.bv_len; j += tuple_sz) {
-				struct crc64_pi_tuple *pi = p;
+				struct crc64_pi_tuple *pi = p +  offset;
 				u64 ref = get_unaligned_be48(pi->ref_tag);
 
 				if (ref == virt)
@@ -411,9 +433,11 @@ static void ext_pi_type1_prepare(struct request *rq)
 
 static void ext_pi_type1_complete(struct request *rq, unsigned int nr_bytes)
 {
-	unsigned intervals = nr_bytes >> rq->q->integrity.interval_exp;
-	const int tuple_sz = rq->q->integrity.tuple_size;
+	struct blk_integrity *bi = &rq->q->integrity;
+	unsigned intervals = nr_bytes >> bi->interval_exp;
+	const int tuple_sz = bi->tuple_size;
 	u64 ref_tag = ext_pi_ref_tag(rq);
+	u8 offset = bi->pi_offset;
 	struct bio *bio;
 
 	__rq_for_each_bio(bio, rq) {
@@ -428,7 +452,7 @@ static void ext_pi_type1_complete(struct request *rq, unsigned int nr_bytes)
 
 			p = bvec_kmap_local(&iv);
 			for (j = 0; j < iv.bv_len && intervals; j += tuple_sz) {
-				struct crc64_pi_tuple *pi = p;
+				struct crc64_pi_tuple *pi = p + offset;
 				u64 ref = get_unaligned_be48(pi->ref_tag);
 
 				if (ref == ref_tag)
diff --git a/include/linux/blk-integrity.h b/include/linux/blk-integrity.h
index 378b2459efe2..e253e7bd0d17 100644
--- a/include/linux/blk-integrity.h
+++ b/include/linux/blk-integrity.h
@@ -20,6 +20,7 @@ struct blk_integrity_iter {
 	unsigned int		data_size;
 	unsigned short		interval;
 	unsigned char		tuple_size;
+	unsigned char		pi_offset;
 	const char		*disk_name;
 };
 
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index d7cac3de65b3..bb4d811fee46 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -108,6 +108,7 @@ struct blk_integrity {
 	const struct blk_integrity_profile	*profile;
 	unsigned char				flags;
 	unsigned char				tuple_size;
+	unsigned char				pi_offset;
 	unsigned char				interval_exp;
 	unsigned char				tag_size;
 };
-- 
2.25.1

[PATCH v2 3/3] nvme: allow integrity when PI is not in first bytes

[Kanchan Joshi]

NVM command set 1.0 (or later) mandates PI to be in the last bytes of
metadata. But this was not supported in the block-layer, and driver
registered a nop profile.

Since block-integrity can now handle flexible PI offset, change the
driver to support this configuration.

Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
---
 drivers/nvme/host/core.c | 8 +++++++-
 drivers/nvme/host/nvme.h | 1 +
 2 files changed, 8 insertions(+), 1 deletion(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 85ab0fcf9e88..4945d6259a13 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -1726,6 +1726,7 @@ static void nvme_init_integrity(struct gendisk *disk,
 	}
 
 	integrity.tuple_size = head->ms;
+	integrity.pi_offset = head->pi_offset;
 	blk_integrity_register(disk, &integrity);
 	blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
 }
@@ -1835,11 +1836,16 @@ static int nvme_init_ms(struct nvme_ctrl *ctrl, struct nvme_ns_head *head,
 free_data:
 	kfree(nvm);
 set_pi:
-	if (head->pi_size && (first || head->ms == head->pi_size))
+	if (head->pi_size && head->ms >= head->pi_size)
 		head->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
 	else
 		head->pi_type = 0;
 
+	if (first)
+		head->pi_offset = 0;
+	else
+		head->pi_offset = head->ms - head->pi_size;
+
 	return ret;
 }
 
diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
index 030c80818240..281657320c3a 100644
--- a/drivers/nvme/host/nvme.h
+++ b/drivers/nvme/host/nvme.h
@@ -462,6 +462,7 @@ struct nvme_ns_head {
 	u16			ms;
 	u16			pi_size;
 	u8			pi_type;
+	u8			pi_offset;
 	u8			guard_type;
 	u16			sgs;
 	u32			sws;
-- 
2.25.1

Re: [PATCH v2 0/3] Block integrity with flexible-offset PI

[Kanchan Joshi]

On Thu, Feb 1, 2024 at 6:39 PM Kanchan Joshi <joshi.k@samsung.com> wrote:
>
> The block integrity subsystem can only work with PI placed in the first
> bytes of the metadata buffer.
>
> The series makes block-integrity support the flexible placement of PI.
> And changes NVMe driver to make use of the new capability.

Hi Jens,

This has got the necessary reviews. Can this be picked up?
Or do you see anything missing?

Re: [PATCH v2 0/3] Block integrity with flexible-offset PI

[Jens Axboe]

On Thu, 01 Feb 2024 18:31:23 +0530, Kanchan Joshi wrote:
> The block integrity subsystem can only work with PI placed in the first
> bytes of the metadata buffer.
> 
> The series makes block-integrity support the flexible placement of PI.
> And changes NVMe driver to make use of the new capability.
> 
> This helps to
> (i) enable the more common case for NVMe (PI in last bytes is the norm)
> (ii) reduce nop profile users (tried by Jens recently [1]).
> 
> [...]

Applied, thanks!

[1/3] block: refactor guard helpers
      commit: 6b5c132a3f0d3b7c024ae98f0ace07c04d32cf73
[2/3] block: support PI at non-zero offset within metadata
      commit: 60d21aac52e26531affdadb7543fe5b93f58b450
[3/3] nvme: allow integrity when PI is not in first bytes
      commit: 921e81db524d17db683cc29aed7ff02f06ea3f96

Best regards,
-- 
Jens Axboe