[PATCHv3] blk-integrity: support arbitrary buffer alignment

[PATCHv3] blk-integrity: support arbitrary buffer alignment

[Keith Busch]

From: Keith Busch <kbusch@kernel.org>

A bio segment might have partial block data with the rest continuing
into the next segments.

At the same time, the protection information may also be split in
multiple segments. The most likely way that may happen is if two
requests merge, or if we're directly using the io_uring user metadata.
Further, it may be possible to have unalign the protection fields from
the user space buffer, or if there are odd additional opaque bytes in
front or in back of the protection information metadata region.

Change up the iteration to allow spanning multiple segments. This patch
is mostly a re-write of the protection information handling to allow any
arbitrary alignments, so it's probably easier to review the end result
rather than the diff.

Signed-off-by: Keith Busch <kbusch@kernel.org>
---
v2->v3:

  Folded in cleanups from Christoph

  A fix for the crc64 ref tag remapping

I ported the liburing's test to blktests here:

  https://lore.kernel.org/linux-block/20251107231836.1280881-1-kbusch@meta.com/

 block/blk-settings.c |  10 -
 block/t10-pi.c       | 791 +++++++++++++++++++++++--------------------
 2 files changed, 431 insertions(+), 370 deletions(-)

diff --git a/block/blk-settings.c b/block/blk-settings.c
index 78dfef1176231..e0d0b035f39d2 100644
--- a/block/blk-settings.c
+++ b/block/blk-settings.c
@@ -197,16 +197,6 @@ static int blk_validate_integrity_limits(struct queue_limits *lim)
 	if (!bi->interval_exp)
 		bi->interval_exp = ilog2(lim->logical_block_size);
 
-	/*
-	 * The PI generation / validation helpers do not expect intervals to
-	 * straddle multiple bio_vecs.  Enforce alignment so that those are
-	 * never generated, and that each buffer is aligned as expected.
-	 */
-	if (bi->csum_type) {
-		lim->dma_alignment = max(lim->dma_alignment,
-					(1U << bi->interval_exp) - 1);
-	}
-
 	/*
 	 * The block layer automatically adds integrity data for bios that don't
 	 * already have it.  Limit the I/O size so that a single maximum size
diff --git a/block/t10-pi.c b/block/t10-pi.c
index 0c4ed97021460..6266ce0043280 100644
--- a/block/t10-pi.c
+++ b/block/t10-pi.c
@@ -12,231 +12,136 @@
 #include <linux/unaligned.h>
 #include "blk.h"
 
+#define APP_TAG_ESCAPE 0xffff
+#define REF_TAG_ESCAPTE 0xffffffff
+
+/*
+ * This union is used for onstack allocations when the pi field is split across
+ * segments. blk_validate_integrity_limits() guarantees pi_tuple_size matches
+ * the sizeof one of these two types.
+ */
+union pi_tuple {
+	struct crc64_pi_tuple	crc64_pi;
+	struct t10_pi_tuple	t10_pi;
+};
+
 struct blk_integrity_iter {
-	void			*prot_buf;
-	void			*data_buf;
-	sector_t		seed;
-	unsigned int		data_size;
-	unsigned short		interval;
-	const char		*disk_name;
+	struct bio			*bio;
+	struct bio_integrity_payload	*bip;
+	struct blk_integrity		*bi;
+	struct bvec_iter		data_iter;
+	struct bvec_iter		prot_iter;
+	unsigned int			interval_remaining;
+	u64				seed;
+	u64				crc;
 };
 
-static __be16 t10_pi_csum(__be16 csum, void *data, unsigned int len,
-		unsigned char csum_type)
+static void blk_crc(struct blk_integrity_iter *iter, void *data,
+		    unsigned int len)
 {
-	if (csum_type == BLK_INTEGRITY_CSUM_IP)
-		return (__force __be16)ip_compute_csum(data, len);
-	return cpu_to_be16(crc_t10dif_update(be16_to_cpu(csum), data, len));
+	switch (iter->bi->csum_type) {
+	case BLK_INTEGRITY_CSUM_CRC64:
+		iter->crc = crc64_nvme(iter->crc, data, len);
+		break;
+	case BLK_INTEGRITY_CSUM_CRC:
+		iter->crc = crc_t10dif_update(iter->crc, data, len);
+		break;
+	case BLK_INTEGRITY_CSUM_IP:
+		iter->crc = (__force u32)csum_partial(data, len,
+						(__force __wsum)iter->crc);
+		break;
+	default:
+		WARN_ON_ONCE(1);
+		iter->crc = U64_MAX;
+		break;
+	}
 }
 
 /*
- * Type 1 and Type 2 protection use the same format: 16 bit guard tag,
- * 16 bit app tag, 32 bit reference tag. Type 3 does not define the ref
- * tag.
+ * Update the crc for formats that have metadata padding in front of the data
+ * integrity field
  */
-static void t10_pi_generate(struct blk_integrity_iter *iter,
-		struct blk_integrity *bi)
+static void blk_integrity_crc_offset(struct blk_integrity_iter *iter)
 {
-	u8 offset = bi->pi_offset;
-	unsigned int i;
-
-	for (i = 0 ; i < iter->data_size ; i += iter->interval) {
-		struct t10_pi_tuple *pi = iter->prot_buf + offset;
-
-		pi->guard_tag = t10_pi_csum(0, iter->data_buf, iter->interval,
-				bi->csum_type);
-		if (offset)
-			pi->guard_tag = t10_pi_csum(pi->guard_tag,
-					iter->prot_buf, offset, bi->csum_type);
-		pi->app_tag = 0;
-
-		if (bi->flags & BLK_INTEGRITY_REF_TAG)
-			pi->ref_tag = cpu_to_be32(lower_32_bits(iter->seed));
-		else
-			pi->ref_tag = 0;
-
-		iter->data_buf += iter->interval;
-		iter->prot_buf += bi->metadata_size;
-		iter->seed++;
+	unsigned int offset = iter->bi->pi_offset;
+	struct bio_vec *bvec = iter->bip->bip_vec;
+
+	while (offset > 0) {
+		struct bio_vec pbv = mp_bvec_iter_bvec(bvec, iter->prot_iter);
+		unsigned int len = min(pbv.bv_len, offset);
+		void *prot_buf = bvec_kmap_local(&pbv);
+
+		blk_crc(iter, prot_buf, len);
+		kunmap_local(prot_buf);
+		offset -= len;
+		bvec_iter_advance_single(bvec, &iter->prot_iter, len);
 	}
 }
 
-static blk_status_t t10_pi_verify(struct blk_integrity_iter *iter,
-		struct blk_integrity *bi)
+static void blk_integrity_copy_from_tuple(struct bio_integrity_payload *bip,
+					  struct bvec_iter *iter, void *tuple,
+					  unsigned int tuple_size)
 {
-	u8 offset = bi->pi_offset;
-	unsigned int i;
-
-	for (i = 0 ; i < iter->data_size ; i += iter->interval) {
-		struct t10_pi_tuple *pi = iter->prot_buf + offset;
-		__be16 csum;
-
-		if (bi->flags & BLK_INTEGRITY_REF_TAG) {
-			if (pi->app_tag == T10_PI_APP_ESCAPE)
-				goto next;
-
-			if (be32_to_cpu(pi->ref_tag) !=
-			    lower_32_bits(iter->seed)) {
-				pr_err("%s: ref tag error at location %llu " \
-				       "(rcvd %u)\n", iter->disk_name,
-				       (unsigned long long)
-				       iter->seed, be32_to_cpu(pi->ref_tag));
-				return BLK_STS_PROTECTION;
-			}
-		} else {
-			if (pi->app_tag == T10_PI_APP_ESCAPE &&
-			    pi->ref_tag == T10_PI_REF_ESCAPE)
-				goto next;
-		}
+	void *prot_buf;
 
-		csum = t10_pi_csum(0, iter->data_buf, iter->interval,
-				bi->csum_type);
-		if (offset)
-			csum = t10_pi_csum(csum, iter->prot_buf, offset,
-					bi->csum_type);
-
-		if (pi->guard_tag != csum) {
-			pr_err("%s: guard tag error at sector %llu " \
-			       "(rcvd %04x, want %04x)\n", iter->disk_name,
-			       (unsigned long long)iter->seed,
-			       be16_to_cpu(pi->guard_tag), be16_to_cpu(csum));
-			return BLK_STS_PROTECTION;
-		}
+	while (tuple_size) {
+		struct bio_vec pbv = mp_bvec_iter_bvec(bip->bip_vec, *iter);
+		unsigned int len = min(tuple_size, pbv.bv_len);
 
-next:
-		iter->data_buf += iter->interval;
-		iter->prot_buf += bi->metadata_size;
-		iter->seed++;
-	}
+		prot_buf = bvec_kmap_local(&pbv);
+		memcpy(prot_buf, tuple, len);
+		kunmap_local(prot_buf);
 
-	return BLK_STS_OK;
+		bvec_iter_advance_single(bip->bip_vec, iter, len);
+		tuple_size -= len;
+		tuple += len;
+	}
 }
 
-/**
- * t10_pi_type1_prepare - prepare PI prior submitting request to device
- * @rq:              request with PI that should be prepared
- *
- * For Type 1/Type 2, the virtual start sector is the one that was
- * originally submitted by the block layer for the ref_tag usage. Due to
- * partitioning, MD/DM cloning, etc. the actual physical start sector is
- * likely to be different. Remap protection information to match the
- * physical LBA.
- */
-static void t10_pi_type1_prepare(struct request *rq)
+static void blk_integrity_copy_to_tuple(struct bio_integrity_payload *bip,
+					struct bvec_iter *iter, void *tuple,
+					unsigned int tuple_size)
 {
-	struct blk_integrity *bi = &rq->q->limits.integrity;
-	const int tuple_sz = bi->metadata_size;
-	u32 ref_tag = t10_pi_ref_tag(rq);
-	u8 offset = bi->pi_offset;
-	struct bio *bio;
+	void *prot_buf;
 
-	__rq_for_each_bio(bio, rq) {
-		struct bio_integrity_payload *bip = bio_integrity(bio);
-		u32 virt = bip_get_seed(bip) & 0xffffffff;
-		struct bio_vec iv;
-		struct bvec_iter iter;
+	while (tuple_size) {
+		struct bio_vec pbv = mp_bvec_iter_bvec(bip->bip_vec, *iter);
+		unsigned int len = min(tuple_size, pbv.bv_len);
 
-		/* Already remapped? */
-		if (bip->bip_flags & BIP_MAPPED_INTEGRITY)
-			break;
-
-		bip_for_each_vec(iv, bip, iter) {
-			unsigned int j;
-			void *p;
-
-			p = bvec_kmap_local(&iv);
-			for (j = 0; j < iv.bv_len; j += tuple_sz) {
-				struct t10_pi_tuple *pi = p + offset;
-
-				if (be32_to_cpu(pi->ref_tag) == virt)
-					pi->ref_tag = cpu_to_be32(ref_tag);
-				virt++;
-				ref_tag++;
-				p += tuple_sz;
-			}
-			kunmap_local(p);
-		}
+		prot_buf = bvec_kmap_local(&pbv);
+		memcpy(tuple, prot_buf, len);
+		kunmap_local(prot_buf);
 
-		bip->bip_flags |= BIP_MAPPED_INTEGRITY;
+		bvec_iter_advance_single(bip->bip_vec, iter, len);
+		tuple_size -= len;
+		tuple += len;
 	}
 }
 
-/**
- * t10_pi_type1_complete - prepare PI prior returning request to the blk layer
- * @rq:              request with PI that should be prepared
- * @nr_bytes:        total bytes to prepare
- *
- * For Type 1/Type 2, the virtual start sector is the one that was
- * originally submitted by the block layer for the ref_tag usage. Due to
- * partitioning, MD/DM cloning, etc. the actual physical start sector is
- * likely to be different. Since the physical start sector was submitted
- * to the device, we should remap it back to virtual values expected by the
- * block layer.
- */
-static void t10_pi_type1_complete(struct request *rq, unsigned int nr_bytes)
+static void blk_set_ext_pi(struct crc64_pi_tuple *pi,
+			   struct blk_integrity_iter *iter)
 {
-	struct blk_integrity *bi = &rq->q->limits.integrity;
-	unsigned intervals = nr_bytes >> bi->interval_exp;
-	const int tuple_sz = bi->metadata_size;
-	u32 ref_tag = t10_pi_ref_tag(rq);
-	u8 offset = bi->pi_offset;
-	struct bio *bio;
-
-	__rq_for_each_bio(bio, rq) {
-		struct bio_integrity_payload *bip = bio_integrity(bio);
-		u32 virt = bip_get_seed(bip) & 0xffffffff;
-		struct bio_vec iv;
-		struct bvec_iter iter;
-
-		bip_for_each_vec(iv, bip, iter) {
-			unsigned int j;
-			void *p;
-
-			p = bvec_kmap_local(&iv);
-			for (j = 0; j < iv.bv_len && intervals; j += tuple_sz) {
-				struct t10_pi_tuple *pi = p + offset;
-
-				if (be32_to_cpu(pi->ref_tag) == ref_tag)
-					pi->ref_tag = cpu_to_be32(virt);
-				virt++;
-				ref_tag++;
-				intervals--;
-				p += tuple_sz;
-			}
-			kunmap_local(p);
-		}
-	}
+	put_unaligned_be64(iter->crc, &pi->guard_tag);
+	put_unaligned((__be16)0, &pi->app_tag);
+	put_unaligned_be48(iter->seed, &pi->ref_tag);
 }
 
-static __be64 ext_pi_crc64(u64 crc, void *data, unsigned int len)
+static void blk_set_t10_pi(struct t10_pi_tuple *pi,
+			   struct blk_integrity_iter *iter)
 {
-	return cpu_to_be64(crc64_nvme(crc, data, len));
+	put_unaligned_be16(iter->crc, &pi->guard_tag);
+	put_unaligned((__be16)0, &pi->app_tag);
+	put_unaligned_be32(iter->seed, &pi->ref_tag);
 }
 
-static void ext_pi_crc64_generate(struct blk_integrity_iter *iter,
-		struct blk_integrity *bi)
+static void blk_set_ip_pi(struct t10_pi_tuple *pi,
+			  struct blk_integrity_iter *iter)
 {
-	u8 offset = bi->pi_offset;
-	unsigned int i;
-
-	for (i = 0 ; i < iter->data_size ; i += iter->interval) {
-		struct crc64_pi_tuple *pi = iter->prot_buf + offset;
+	__be16 csum = (__force __be16)~(lower_16_bits(iter->crc));
 
-		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 (bi->flags & BLK_INTEGRITY_REF_TAG)
-			put_unaligned_be48(iter->seed, pi->ref_tag);
-		else
-			put_unaligned_be48(0ULL, pi->ref_tag);
-
-		iter->data_buf += iter->interval;
-		iter->prot_buf += bi->metadata_size;
-		iter->seed++;
-	}
+	__put_unaligned_t(__be16, csum, &pi->guard_tag);
+	put_unaligned_be16(0, &pi->app_tag);
+	put_unaligned_be32(iter->seed, &pi->ref_tag);
 }
 
 static bool ext_pi_ref_escape(const u8 ref_tag[6])
@@ -247,227 +152,393 @@ static bool ext_pi_ref_escape(const u8 ref_tag[6])
 }
 
 static blk_status_t ext_pi_crc64_verify(struct blk_integrity_iter *iter,
-		struct blk_integrity *bi)
+					struct crc64_pi_tuple *pi)
 {
-	u8 offset = bi->pi_offset;
-	unsigned int i;
-
-	for (i = 0; i < iter->data_size; i += iter->interval) {
-		struct crc64_pi_tuple *pi = iter->prot_buf + offset;
-		u64 ref, seed;
-		__be64 csum;
-
-		if (bi->flags & BLK_INTEGRITY_REF_TAG) {
-			if (pi->app_tag == T10_PI_APP_ESCAPE)
-				goto next;
-
-			ref = get_unaligned_be48(pi->ref_tag);
-			seed = lower_48_bits(iter->seed);
-			if (ref != seed) {
-				pr_err("%s: ref tag error at location %llu (rcvd %llu)\n",
-					iter->disk_name, seed, ref);
-				return BLK_STS_PROTECTION;
-			}
-		} else {
-			if (pi->app_tag == T10_PI_APP_ESCAPE &&
-			    ext_pi_ref_escape(pi->ref_tag))
-				goto next;
+	u64 seed = lower_48_bits(iter->seed);
+	u64 guard = get_unaligned_be64(&pi->guard_tag);
+	u64 ref = get_unaligned_be48(pi->ref_tag);
+	u16 app = get_unaligned_be16(&pi->app_tag);
+
+	if (iter->bi->flags & BLK_INTEGRITY_REF_TAG) {
+		if (app == APP_TAG_ESCAPE)
+			return BLK_STS_OK;
+		if (ref != seed) {
+			pr_err("%s: ref tag error at location %llu (rcvd %llu)\n",
+				iter->bio->bi_bdev->bd_disk->disk_name, seed,
+				ref);
+			return BLK_STS_PROTECTION;
 		}
+	} else if (app == APP_TAG_ESCAPE && ext_pi_ref_escape(pi->ref_tag)) {
+		return BLK_STS_OK;
+	}
+
+	if (guard != iter->crc) {
+		pr_err("%s: guard tag error at sector %llu (rcvd %016llx, want %016llx)\n",
+			iter->bio->bi_bdev->bd_disk->disk_name, iter->seed,
+			guard, iter->crc);
+		return BLK_STS_PROTECTION;
+	}
 
-		csum = ext_pi_crc64(0, iter->data_buf, iter->interval);
-		if (offset)
-			csum = ext_pi_crc64(be64_to_cpu(csum), iter->prot_buf,
-					    offset);
+	return BLK_STS_OK;
+}
 
-		if (pi->guard_tag != csum) {
-			pr_err("%s: guard tag error at sector %llu " \
-			       "(rcvd %016llx, want %016llx)\n",
-				iter->disk_name, (unsigned long long)iter->seed,
-				be64_to_cpu(pi->guard_tag), be64_to_cpu(csum));
+static blk_status_t t10_pi_verify(struct blk_integrity_iter *iter,
+				  struct t10_pi_tuple *pi)
+{
+	u32 seed = lower_32_bits(iter->seed);
+	u32 ref = get_unaligned_be32(&pi->ref_tag);
+	u16 guard = get_unaligned_be16(&pi->guard_tag);
+	u16 app = get_unaligned_be16(&pi->app_tag);
+
+	if (iter->bi->flags & BLK_INTEGRITY_REF_TAG) {
+		if (app == APP_TAG_ESCAPE)
+			return BLK_STS_OK;
+		if (ref != seed) {
+			pr_err("%s: ref tag error at location %u (rcvd %u)\n",
+				iter->bio->bi_bdev->bd_disk->disk_name, seed,
+				ref);
 			return BLK_STS_PROTECTION;
 		}
+	} else if (app == APP_TAG_ESCAPE && ref == REF_TAG_ESCAPTE) {
+		return BLK_STS_OK;
+	}
 
-next:
-		iter->data_buf += iter->interval;
-		iter->prot_buf += bi->metadata_size;
-		iter->seed++;
+	if (guard != (u16)iter->crc) {
+		pr_err("%s: guard tag error at sector %llu (rcvd %04x, want %04x)\n",
+			iter->bio->bi_bdev->bd_disk->disk_name, iter->seed,
+			guard, (u16)iter->crc);
+		return BLK_STS_PROTECTION;
 	}
 
 	return BLK_STS_OK;
 }
 
-static void ext_pi_type1_prepare(struct request *rq)
+static blk_status_t blk_integrity_verify(struct blk_integrity_iter *iter,
+					 union pi_tuple *tuple)
 {
-	struct blk_integrity *bi = &rq->q->limits.integrity;
-	const int tuple_sz = bi->metadata_size;
-	u64 ref_tag = ext_pi_ref_tag(rq);
-	u8 offset = bi->pi_offset;
-	struct bio *bio;
+	switch (iter->bi->csum_type) {
+	case BLK_INTEGRITY_CSUM_CRC64:
+		return ext_pi_crc64_verify(iter, &tuple->crc64_pi);
+	case BLK_INTEGRITY_CSUM_CRC:
+	case BLK_INTEGRITY_CSUM_IP:
+		return t10_pi_verify(iter, &tuple->t10_pi);
+	default:
+		return BLK_STS_OK;
+	}
+}
 
-	__rq_for_each_bio(bio, rq) {
-		struct bio_integrity_payload *bip = bio_integrity(bio);
-		u64 virt = lower_48_bits(bip_get_seed(bip));
-		struct bio_vec iv;
-		struct bvec_iter iter;
+static void blk_integrity_set(struct blk_integrity_iter *iter,
+			      union pi_tuple *tuple)
+{
+	switch (iter->bi->csum_type) {
+	case BLK_INTEGRITY_CSUM_CRC64:
+		return blk_set_ext_pi(&tuple->crc64_pi, iter);
+	case BLK_INTEGRITY_CSUM_CRC:
+		return blk_set_t10_pi(&tuple->t10_pi, iter);
+	case BLK_INTEGRITY_CSUM_IP:
+		return blk_set_ip_pi(&tuple->t10_pi, iter);
+	default:
+		WARN_ON_ONCE(1);
+		return;
+	}
+}
 
-		/* Already remapped? */
-		if (bip->bip_flags & BIP_MAPPED_INTEGRITY)
-			break;
+static blk_status_t blk_integrity_interval(struct blk_integrity_iter *iter,
+					   bool verify)
+{
+	blk_status_t ret = BLK_STS_OK;
+	union pi_tuple tuple;
+	void *ptuple = &tuple;
+	struct bio_vec pbv;
+
+	blk_integrity_crc_offset(iter);
+	pbv = mp_bvec_iter_bvec(iter->bip->bip_vec, iter->prot_iter);
+	if (pbv.bv_len >= iter->bi->pi_tuple_size) {
+		ptuple = bvec_kmap_local(&pbv);
+		bvec_iter_advance_single(iter->bip->bip_vec, &iter->prot_iter,
+				iter->bi->metadata_size - iter->bi->pi_offset);
+	} else if (verify) {
+		blk_integrity_copy_to_tuple(iter->bip, &iter->prot_iter,
+				ptuple, iter->bi->pi_tuple_size);
+	}
 
-		bip_for_each_vec(iv, bip, iter) {
-			unsigned int j;
-			void *p;
-
-			p = bvec_kmap_local(&iv);
-			for (j = 0; j < iv.bv_len; j += tuple_sz) {
-				struct crc64_pi_tuple *pi = p +  offset;
-				u64 ref = get_unaligned_be48(pi->ref_tag);
-
-				if (ref == virt)
-					put_unaligned_be48(ref_tag, pi->ref_tag);
-				virt++;
-				ref_tag++;
-				p += tuple_sz;
-			}
-			kunmap_local(p);
-		}
+	if (verify)
+		ret = blk_integrity_verify(iter, ptuple);
+	else
+		blk_integrity_set(iter, ptuple);
 
-		bip->bip_flags |= BIP_MAPPED_INTEGRITY;
+	if (likely(ptuple != &tuple)) {
+		kunmap_local(ptuple);
+	} else if (!verify) {
+		blk_integrity_copy_from_tuple(iter->bip, &iter->prot_iter,
+				ptuple, iter->bi->pi_tuple_size);
 	}
+
+
+	iter->interval_remaining = 1 << iter->bi->interval_exp;
+	iter->crc = 0;
+	iter->seed++;
+
+	return ret;
 }
 
-static void ext_pi_type1_complete(struct request *rq, unsigned int nr_bytes)
+static void blk_integrity_iterate(struct bio *bio, struct bvec_iter *data_iter,
+				  bool verify)
 {
-	struct blk_integrity *bi = &rq->q->limits.integrity;
-	unsigned intervals = nr_bytes >> bi->interval_exp;
-	const int tuple_sz = bi->metadata_size;
-	u64 ref_tag = ext_pi_ref_tag(rq);
-	u8 offset = bi->pi_offset;
-	struct bio *bio;
-
-	__rq_for_each_bio(bio, rq) {
-		struct bio_integrity_payload *bip = bio_integrity(bio);
-		u64 virt = lower_48_bits(bip_get_seed(bip));
-		struct bio_vec iv;
-		struct bvec_iter iter;
-
-		bip_for_each_vec(iv, bip, iter) {
-			unsigned int j;
-			void *p;
-
-			p = bvec_kmap_local(&iv);
-			for (j = 0; j < iv.bv_len && intervals; j += tuple_sz) {
-				struct crc64_pi_tuple *pi = p + offset;
-				u64 ref = get_unaligned_be48(pi->ref_tag);
-
-				if (ref == ref_tag)
-					put_unaligned_be48(virt, pi->ref_tag);
-				virt++;
-				ref_tag++;
-				intervals--;
-				p += tuple_sz;
-			}
-			kunmap_local(p);
+	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
+	struct bio_integrity_payload *bip = bio_integrity(bio);
+	struct blk_integrity_iter iter = {
+		.bio = bio,
+		.bip = bip,
+		.bi = bi,
+		.data_iter = *data_iter,
+		.prot_iter = bip->bip_iter,
+		.interval_remaining = 1 << bi->interval_exp,
+		.seed = data_iter->bi_sector,
+		.crc = 0,
+	};
+	blk_status_t ret = BLK_STS_OK;
+
+	while (iter.data_iter.bi_size && ret == BLK_STS_OK) {
+		struct bio_vec bv = mp_bvec_iter_bvec(iter.bio->bi_io_vec,
+						      iter.data_iter);
+		void *kaddr = bvec_kmap_local(&bv);
+		void *data = kaddr;
+		unsigned int len;
+
+		bvec_iter_advance_single(iter.bio->bi_io_vec, &iter.data_iter,
+					 bv.bv_len);
+		while (bv.bv_len && ret == BLK_STS_OK) {
+			len = min(iter.interval_remaining, bv.bv_len);
+			blk_crc(&iter, data, len);
+			bv.bv_len -= len;
+			data += len;
+			iter.interval_remaining -= len;
+			if (!iter.interval_remaining)
+				ret = blk_integrity_interval(&iter, verify);
 		}
+		kunmap_local(kaddr);
 	}
+
+	if (ret)
+		bio->bi_status = ret;
 }
 
 void blk_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);
-	struct blk_integrity_iter iter;
-	struct bvec_iter bviter;
-	struct bio_vec bv;
-
-	iter.disk_name = bio->bi_bdev->bd_disk->disk_name;
-	iter.interval = 1 << bi->interval_exp;
-	iter.seed = bio->bi_iter.bi_sector;
-	iter.prot_buf = bvec_virt(bip->bip_vec);
-	bio_for_each_segment(bv, bio, bviter) {
-		void *kaddr = bvec_kmap_local(&bv);
 
-		iter.data_buf = kaddr;
-		iter.data_size = bv.bv_len;
-		switch (bi->csum_type) {
-		case BLK_INTEGRITY_CSUM_CRC64:
-			ext_pi_crc64_generate(&iter, bi);
-			break;
-		case BLK_INTEGRITY_CSUM_CRC:
-		case BLK_INTEGRITY_CSUM_IP:
-			t10_pi_generate(&iter, bi);
-			break;
-		default:
-			break;
-		}
-		kunmap_local(kaddr);
+	switch (bi->csum_type) {
+	case BLK_INTEGRITY_CSUM_CRC64:
+	case BLK_INTEGRITY_CSUM_CRC:
+	case BLK_INTEGRITY_CSUM_IP:
+		blk_integrity_iterate(bio, &bio->bi_iter, false);
+		break;
+	default:
+		break;
 	}
 }
 
 void blk_integrity_verify_iter(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);
-	struct blk_integrity_iter iter;
-	struct bvec_iter bviter;
-	struct bio_vec bv;
+
+	switch (bi->csum_type) {
+	case BLK_INTEGRITY_CSUM_CRC64:
+	case BLK_INTEGRITY_CSUM_CRC:
+	case BLK_INTEGRITY_CSUM_IP:
+		blk_integrity_iterate(bio, saved_iter, true);
+		break;
+	default:
+		break;
+	}
+}
+
+/*
+ * Advance @iter past the protection offset for protection formats that
+ * contain front padding on the metadata region.
+ */
+static void blk_pi_advance_offset(struct blk_integrity *bi,
+				  struct bio_integrity_payload *bip,
+				  struct bvec_iter *iter)
+{
+	unsigned int offset = bi->pi_offset;
+
+	while (offset > 0) {
+		struct bio_vec bv = mp_bvec_iter_bvec(bip->bip_vec, *iter);
+		unsigned int len = min(bv.bv_len, offset);
+
+		bvec_iter_advance_single(bip->bip_vec, iter, len);
+		offset -= len;
+	}
+}
+
+static void *blk_tuple_remap_begin(union pi_tuple *tuple,
+				   struct blk_integrity *bi,
+				   struct bio_integrity_payload *bip,
+				   struct bvec_iter *iter)
+{
+	struct bvec_iter titer;
+	struct bio_vec pbv;
+
+	blk_pi_advance_offset(bi, bip, iter);
+	pbv = mp_bvec_iter_bvec(bip->bip_vec, *iter);
+	if (likely(pbv.bv_len >= bi->pi_tuple_size))
+		return bvec_kmap_local(&pbv);
 
 	/*
-	 * At the moment verify is called bi_iter has been advanced during split
-	 * and completion, so use the copy created during submission here.
+	 * We need to preserve the state of the original iter for the
+	 * copy_from_tuple at the end, so make a temp iter for here.
 	 */
-	iter.disk_name = bio->bi_bdev->bd_disk->disk_name;
-	iter.interval = 1 << bi->interval_exp;
-	iter.seed = saved_iter->bi_sector;
-	iter.prot_buf = bvec_virt(bip->bip_vec);
-	__bio_for_each_segment(bv, bio, bviter, *saved_iter) {
-		void *kaddr = bvec_kmap_local(&bv);
-		blk_status_t ret = BLK_STS_OK;
+	titer = *iter;
+	blk_integrity_copy_to_tuple(bip, &titer, tuple, bi->pi_tuple_size);
+	return tuple;
+}
 
-		iter.data_buf = kaddr;
-		iter.data_size = bv.bv_len;
-		switch (bi->csum_type) {
-		case BLK_INTEGRITY_CSUM_CRC64:
-			ret = ext_pi_crc64_verify(&iter, bi);
-			break;
-		case BLK_INTEGRITY_CSUM_CRC:
-		case BLK_INTEGRITY_CSUM_IP:
-			ret = t10_pi_verify(&iter, bi);
-			break;
-		default:
-			break;
-		}
-		kunmap_local(kaddr);
+static void blk_tuple_remap_end(union pi_tuple *tuple, void *ptuple,
+				struct blk_integrity *bi,
+				struct bio_integrity_payload *bip,
+				struct bvec_iter *iter)
+{
+	unsigned int len = bi->metadata_size - bi->pi_offset;
+
+	if (likely(ptuple != tuple)) {
+		kunmap_local(ptuple);
+	} else {
+		blk_integrity_copy_from_tuple(bip, iter, ptuple,
+				bi->pi_tuple_size);
+		len -= bi->pi_tuple_size;
+	}
 
-		if (ret) {
-			bio->bi_status = ret;
-			return;
-		}
+	bvec_iter_advance(bip->bip_vec, iter, len);
+}
+
+static void blk_set_ext_unmap_ref(struct crc64_pi_tuple *pi, u64 virt,
+				  u64 ref_tag)
+{
+	u64 ref = get_unaligned_be48(&pi->ref_tag);
+
+	if (ref == lower_48_bits(ref_tag) && ref != lower_48_bits(virt))
+		put_unaligned_be48(virt, pi->ref_tag);
+}
+
+static void blk_set_t10_unmap_ref(struct t10_pi_tuple *pi, u32 virt,
+				  u32 ref_tag)
+{
+	u32 ref = get_unaligned_be32(&pi->ref_tag);
+
+	if (ref == ref_tag && ref != virt)
+		put_unaligned_be32(virt, &pi->ref_tag);
+}
+
+static void blk_reftag_remap_complete(struct blk_integrity *bi,
+				      union pi_tuple *tuple, u64 virt, u64 ref)
+{
+	switch (bi->csum_type) {
+	case BLK_INTEGRITY_CSUM_CRC64:
+		blk_set_ext_unmap_ref(&tuple->crc64_pi, virt, ref);
+		break;
+	case BLK_INTEGRITY_CSUM_CRC:
+	case BLK_INTEGRITY_CSUM_IP:
+		blk_set_t10_unmap_ref(&tuple->t10_pi, virt, ref);
+		break;
+	default:
+		WARN_ON_ONCE(1);
+		break;
 	}
 }
 
-void blk_integrity_prepare(struct request *rq)
+static void blk_set_ext_map_ref(struct crc64_pi_tuple *pi, u64 virt,
+				u64 ref_tag)
 {
-	struct blk_integrity *bi = &rq->q->limits.integrity;
+	u64 ref = get_unaligned_be48(&pi->ref_tag);
 
-	if (!(bi->flags & BLK_INTEGRITY_REF_TAG))
+	if (ref == lower_48_bits(virt) && ref != ref_tag)
+		put_unaligned_be48(ref_tag, pi->ref_tag);
+}
+
+static void blk_set_t10_map_ref(struct t10_pi_tuple *pi, u32 virt, u32 ref_tag)
+{
+	u32 ref = get_unaligned_be32(&pi->ref_tag);
+
+	if (ref == virt && ref != ref_tag)
+		put_unaligned_be32(ref_tag, &pi->ref_tag);
+}
+
+static void blk_reftag_remap_prepare(struct blk_integrity *bi,
+				     union pi_tuple *tuple,
+				     u64 virt, u64 ref)
+{
+	switch (bi->csum_type) {
+	case BLK_INTEGRITY_CSUM_CRC64:
+		blk_set_ext_map_ref(&tuple->crc64_pi, virt, ref);
+		break;
+	case BLK_INTEGRITY_CSUM_CRC:
+	case BLK_INTEGRITY_CSUM_IP:
+		blk_set_t10_map_ref(&tuple->t10_pi, virt, ref);
+		break;
+	default:
+		WARN_ON_ONCE(1);
+		break;
+	}
+}
+
+static void __blk_reftag_remap(struct bio *bio, struct blk_integrity *bi,
+			       unsigned *intervals, u64 *ref, bool prep)
+{
+	struct bio_integrity_payload *bip = bio_integrity(bio);
+	struct bvec_iter iter = bip->bip_iter;
+	u64 virt = bip_get_seed(bip);
+	union pi_tuple tuple;
+	union pi_tuple *ptuple;
+
+	if (prep && bip->bip_flags & BIP_MAPPED_INTEGRITY) {
+		*ref += bio->bi_iter.bi_size >> bi->interval_exp;
 		return;
+	}
 
-	if (bi->csum_type == BLK_INTEGRITY_CSUM_CRC64)
-		ext_pi_type1_prepare(rq);
-	else
-		t10_pi_type1_prepare(rq);
+	while (iter.bi_size && *intervals) {
+		ptuple = blk_tuple_remap_begin(&tuple, bi, bip, &iter);
+
+		if (prep)
+			blk_reftag_remap_prepare(bi, ptuple, virt, *ref);
+		else
+			blk_reftag_remap_complete(bi, ptuple, virt, *ref);
+
+		blk_tuple_remap_end(&tuple, ptuple, bi, bip, &iter);
+		(*intervals)--;
+		(*ref)++;
+		virt++;
+	}
+
+	if (prep)
+		bip->bip_flags |= BIP_MAPPED_INTEGRITY;
 }
 
-void blk_integrity_complete(struct request *rq, unsigned int nr_bytes)
+static void blk_integrity_remap(struct request *rq, unsigned int nr_bytes,
+				bool prep)
 {
 	struct blk_integrity *bi = &rq->q->limits.integrity;
+	u64 ref = blk_rq_pos(rq) >> (bi->interval_exp - SECTOR_SHIFT);
+	unsigned intervals = nr_bytes >> bi->interval_exp;
+	struct bio *bio;
 
 	if (!(bi->flags & BLK_INTEGRITY_REF_TAG))
 		return;
 
-	if (bi->csum_type == BLK_INTEGRITY_CSUM_CRC64)
-		ext_pi_type1_complete(rq, nr_bytes);
-	else
-		t10_pi_type1_complete(rq, nr_bytes);
+	__rq_for_each_bio(bio, rq) {
+		__blk_reftag_remap(bio, bi, &intervals, &ref, prep);
+		if (!intervals)
+			break;
+	}
+}
+
+void blk_integrity_prepare(struct request *rq)
+{
+	blk_integrity_remap(rq, blk_rq_bytes(rq), true);
+}
+
+void blk_integrity_complete(struct request *rq, unsigned int nr_bytes)
+{
+	blk_integrity_remap(rq, nr_bytes, false);
 }
-- 
2.47.3

Re: [PATCHv3] blk-integrity: support arbitrary buffer alignment

[Martin K. Petersen]

Hi Keith!

> diff --git a/block/t10-pi.c b/block/t10-pi.c
> index 0c4ed97021460..6266ce0043280 100644
> --- a/block/t10-pi.c
> +++ b/block/t10-pi.c
> @@ -12,231 +12,136 @@
>  #include <linux/unaligned.h>
>  #include "blk.h"
>  
> +#define APP_TAG_ESCAPE 0xffff
> +#define REF_TAG_ESCAPTE 0xffffffff

ESCAPE

> +static void blk_crc(struct blk_integrity_iter *iter, void *data,
> +		    unsigned int len)

Maybe blk_calculate_checksum() or blk_calculate_guard()? It's a bit
weird to refer to the IP checksum as a CRC.

To my knowledge, no SCSI controller can handle PI split across segments
and the PI tuples are required to be naturally aligned in memory. SCSI
controllers probably can't handle split data segments either since the
design constraint is one PI segment per protection interval sized data
segment.

-- 
Martin K. Petersen

Re: [PATCHv3] blk-integrity: support arbitrary buffer alignment

[Keith Busch]

On Fri, Nov 07, 2025 at 10:52:14PM -0500, Martin K. Petersen wrote:
> > +static void blk_crc(struct blk_integrity_iter *iter, void *data,
> > +		    unsigned int len)
> 
> Maybe blk_calculate_checksum() or blk_calculate_guard()? It's a bit
> weird to refer to the IP checksum as a CRC.

Sounds good.
 
> To my knowledge, no SCSI controller can handle PI split across segments
> and the PI tuples are required to be naturally aligned in memory. SCSI
> controllers probably can't handle split data segments either since the
> design constraint is one PI segment per protection interval sized data
> segment.

There are not many nvme controllers that can handle split pi segments
either. Anything setting ID_CTRL SGLS.MSDS can do it, but it's a rarely
advertised capability. For everyone else, metadata buffers have a single
segment limit.

As for interval data split across segments, that's totally fine for any
nvme that supports PI formats. That kind of data was mainly what I was
initially trying to handle here since that's sufficient to reinstate the
lower dma_alignment limit. It wasn't much more work to support split pi
though, unlikely as it is to have hardware capable of it.

Re: [PATCHv3] blk-integrity: support arbitrary buffer alignment

[Christoph Hellwig]

On Sat, Nov 08, 2025 at 09:20:08AM -0700, Keith Busch wrote:
> > To my knowledge, no SCSI controller can handle PI split across segments
> > and the PI tuples are required to be naturally aligned in memory. SCSI
> > controllers probably can't handle split data segments either since the
> > design constraint is one PI segment per protection interval sized data
> > segment.
> 
> There are not many nvme controllers that can handle split pi segments
> either. Anything setting ID_CTRL SGLS.MSDS can do it, but it's a rarely
> advertised capability. For everyone else, metadata buffers have a single
> segment limit.
> 
> As for interval data split across segments, that's totally fine for any
> nvme that supports PI formats. That kind of data was mainly what I was
> initially trying to handle here since that's sufficient to reinstate the
> lower dma_alignment limit. It wasn't much more work to support split pi
> though, unlikely as it is to have hardware capable of it.

I guess we'll just need a flag to distinguish the SCSI vs NVMe
(or HBA vs integrated) semantics here.