[PATCH 0 of 6] SCSI Data Integrity Support

[PATCH 0 of 6] SCSI Data Integrity Support

[Martin K. Petersen]

Another post of my SCSI DIF changes.  These patches require the block
layer integrity infrastructure in Jens' tree as well as the patches
that James took already.

The complete patch series against 2.6.26rc9 is available here:

    http://oss.oracle.com/~mkp/patches/

Changes since last post:

 - Make the distinction between DIF and DIX clearer in the scsi_host
   settings.

 - Revamp DIF/DIX negotiation to support all combinations.

 - Error handling: Correctly handle failures detected by HBA and disk
   respectively.

 - Integrate protection scsi_data_buffer allocation with the command
   pools.

 - scsi_device_protection() accessor.

 - Remove unnecessary printks at disk init time.

-- 
Martin K. Petersen	Oracle Linux Engineering

[PATCH 0 of 6] SCSI Data Integrity Support

[Martin K. Petersen]

These patches are against Linus' current git tree which now has all of
the block layer data integrity code + the preparatory SCSI bits.

Changes since last post:

 - Fixed a bug in error handling when HBA detected an integrity error.

 - Introduced prot_type in scsi_cmnd which informs the HBA of the
   target type.  This is necessary because the HBA needs to know which
   fields in the protection information to check and that depends on
   the target type.

-- 
Martin K. Petersen	Oracle Linux Engineering

[PATCH 1 of 6] scsi: Host protection capabilities

[Martin K. Petersen]

Controllers that support protection information must indicate this to
the SCSI midlayer so that the ULD can prepare scsi_cmnds accordingly.

This patch implements a host mask and various types of protection:

 - DIF Type 1-3 (between HBA and disk)
 - DIX Type 0-3 (between OS and HBA)

The patch also allows the HBA to set the guard type to something
different than the T10-mandated CRC.

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

---
2 files changed, 88 insertions(+)
drivers/scsi/scsi_sysfs.c |    4 ++
include/scsi/scsi_host.h  |   84 +++++++++++++++++++++++++++++++++++++++++++++



diff --git a/drivers/scsi/scsi_sysfs.c b/drivers/scsi/scsi_sysfs.c
--- a/drivers/scsi/scsi_sysfs.c
+++ b/drivers/scsi/scsi_sysfs.c
@@ -249,6 +249,8 @@ shost_rd_attr(can_queue, "%hd\n");
 shost_rd_attr(can_queue, "%hd\n");
 shost_rd_attr(sg_tablesize, "%hu\n");
 shost_rd_attr(unchecked_isa_dma, "%d\n");
+shost_rd_attr(prot_capabilities, "%u\n");
+shost_rd_attr(prot_guard_type, "%hd\n");
 shost_rd_attr2(proc_name, hostt->proc_name, "%s\n");
 
 static struct attribute *scsi_sysfs_shost_attrs[] = {
@@ -263,6 +265,8 @@ static struct attribute *scsi_sysfs_shos
 	&dev_attr_hstate.attr,
 	&dev_attr_supported_mode.attr,
 	&dev_attr_active_mode.attr,
+	&dev_attr_prot_capabilities.attr,
+	&dev_attr_prot_guard_type.attr,
 	NULL
 };
 
diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h
--- a/include/scsi/scsi_host.h
+++ b/include/scsi/scsi_host.h
@@ -636,6 +636,10 @@ struct Scsi_Host {
 	 */
 	unsigned int max_host_blocked;
 
+	/* Protection Information */
+	unsigned int prot_capabilities;
+	unsigned char prot_guard_type;
+
 	/*
 	 * q used for scsi_tgt msgs, async events or any other requests that
 	 * need to be processed in userspace
@@ -756,6 +760,86 @@ extern void scsi_free_host_dev(struct sc
 extern void scsi_free_host_dev(struct scsi_device *);
 extern struct scsi_device *scsi_get_host_dev(struct Scsi_Host *);
 
+/*
+ * DIF defines the exchange of protection information between
+ * initiator and SBC block device.
+ *
+ * DIX defines the exchange of protection information between OS and
+ * initiator.
+ */
+enum scsi_host_prot_capabilities {
+	SHOST_DIF_TYPE1_PROTECTION = 1 << 0, /* T10 DIF Type 1 */
+	SHOST_DIF_TYPE2_PROTECTION = 1 << 1, /* T10 DIF Type 2 */
+	SHOST_DIF_TYPE3_PROTECTION = 1 << 2, /* T10 DIF Type 3 */
+
+	SHOST_DIX_TYPE0_PROTECTION = 1 << 3, /* DIX between OS and HBA only */
+	SHOST_DIX_TYPE1_PROTECTION = 1 << 4, /* DIX with DIF Type 1 */
+	SHOST_DIX_TYPE2_PROTECTION = 1 << 5, /* DIX with DIF Type 2 */
+	SHOST_DIX_TYPE3_PROTECTION = 1 << 6, /* DIX with DIF Type 3 */
+};
+
+/*
+ * SCSI hosts which support the Data Integrity Extensions must
+ * indicate their capabilities by setting the prot_capabilities using
+ * this call.
+ */
+static inline void scsi_host_set_prot(struct Scsi_Host *shost, unsigned int mask)
+{
+	shost->prot_capabilities = mask;
+}
+
+static inline unsigned int scsi_host_get_prot(struct Scsi_Host *shost)
+{
+	return shost->prot_capabilities;
+}
+
+static inline unsigned int scsi_host_dif_capable(struct Scsi_Host *shost, unsigned int target_type)
+{
+	switch (target_type) {
+	case 1: return shost->prot_capabilities & SHOST_DIF_TYPE1_PROTECTION;
+	case 2: return shost->prot_capabilities & SHOST_DIF_TYPE2_PROTECTION;
+	case 3: return shost->prot_capabilities & SHOST_DIF_TYPE3_PROTECTION;
+	}
+
+	return 0;
+}
+
+static inline unsigned int scsi_host_dix_capable(struct Scsi_Host *shost, unsigned int target_type)
+{
+	switch (target_type) {
+	case 0: return shost->prot_capabilities & SHOST_DIX_TYPE0_PROTECTION;
+	case 1: return shost->prot_capabilities & SHOST_DIX_TYPE1_PROTECTION;
+	case 2: return shost->prot_capabilities & SHOST_DIX_TYPE2_PROTECTION;
+	case 3: return shost->prot_capabilities & SHOST_DIX_TYPE3_PROTECTION;
+	}
+
+	return 0;
+}
+
+/*
+ * All DIX-capable initiators must support the T10-mandated CRC
+ * checksum.  Controllers can optionally implement the IP checksum
+ * scheme which has much lower impact on system performance.  Note
+ * that the main rationale for the checksum is to match integrity
+ * metadata with data.  Detecting bit errors are a job for ECC memory
+ * and buses.
+ */
+
+enum scsi_host_guard_type {
+	SHOST_DIX_GUARD_CRC = 1 << 0,
+	SHOST_DIX_GUARD_IP  = 1 << 1,
+};
+
+static inline void scsi_host_set_guard(struct Scsi_Host *shost, unsigned char type)
+{
+	shost->prot_guard_type = type;
+}
+
+static inline unsigned char scsi_host_get_guard(struct Scsi_Host *shost)
+{
+	return shost->prot_guard_type;
+}
+
 /* legacy interfaces */
 extern struct Scsi_Host *scsi_register(struct scsi_host_template *, int);
 extern void scsi_unregister(struct Scsi_Host *);

[PATCH 2 of 6] scsi: Command protection operation

[Martin K. Petersen]

Controllers that support DMA of protection information must be told
explicitly how to handle the I/O.  The controller has no knowledge of
the protection capabilities of the target device so this information
must be passed in the scsi_cmnd.

 - The protection operation tells the HBA whether to generate, strip or
   verify protection information.

 - The protection type tells the HBA which layout the target is
   formatted with.  This is necessary because the controller must be
   able to correctly interpret the included protection information in
   order to verify it.

 - When a scsi_cmnd is reused for error handling the protection
   operation must be cleared and saved while error handling is in
   progress.

 - prot_op and prot_type are placed in an existing hole in scsi_cmnd
   and don't cause the structure to grow.

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

---
3 files changed, 70 insertions(+)
drivers/scsi/scsi_error.c |    3 ++
include/scsi/scsi_cmnd.h  |   66 +++++++++++++++++++++++++++++++++++++++++++++
include/scsi/scsi_eh.h    |    1 



diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -664,7 +664,9 @@ void scsi_eh_prep_cmnd(struct scsi_cmnd 
 	ses->sdb = scmd->sdb;
 	ses->next_rq = scmd->request->next_rq;
 	ses->result = scmd->result;
+	ses->prot_op = scmd->prot_op;
 
+	scmd->prot_op = SCSI_PROT_NORMAL;
 	scmd->cmnd = ses->eh_cmnd;
 	memset(scmd->cmnd, 0, BLK_MAX_CDB);
 	memset(&scmd->sdb, 0, sizeof(scmd->sdb));
@@ -722,6 +724,7 @@ void scsi_eh_restore_cmnd(struct scsi_cm
 	scmd->sdb = ses->sdb;
 	scmd->request->next_rq = ses->next_rq;
 	scmd->result = ses->result;
+	scmd->prot_op = ses->prot_op;
 }
 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
 
diff --git a/include/scsi/scsi_cmnd.h b/include/scsi/scsi_cmnd.h
--- a/include/scsi/scsi_cmnd.h
+++ b/include/scsi/scsi_cmnd.h
@@ -76,6 +76,9 @@ struct scsi_cmnd {
 	int retries;
 	int allowed;
 	int timeout_per_command;
+
+	unsigned char prot_op;
+	unsigned char prot_type;
 
 	unsigned short cmd_len;
 	enum dma_data_direction sc_data_direction;
@@ -208,4 +211,67 @@ static inline int scsi_sg_copy_to_buffer
 				 buf, buflen);
 }
 
+/*
+ * The operations below are hints that tell the controller driver how
+ * to handle I/Os with DIF or similar types of protection information.
+ */
+enum scsi_prot_operations {
+	/* Normal I/O */
+	SCSI_PROT_NORMAL = 0,
+
+	/* OS-HBA: Protected, HBA-Target: Unprotected */
+	SCSI_PROT_READ_INSERT,
+	SCSI_PROT_WRITE_STRIP,
+
+	/* OS-HBA: Unprotected, HBA-Target: Protected */
+	SCSI_PROT_READ_STRIP,
+	SCSI_PROT_WRITE_INSERT,
+
+	/* OS-HBA: Protected, HBA-Target: Protected */
+	SCSI_PROT_READ_PASS,
+	SCSI_PROT_WRITE_PASS,
+
+	/* OS-HBA: Protected, HBA-Target: Protected, checksum conversion */
+	SCSI_PROT_READ_CONVERT,
+	SCSI_PROT_WRITE_CONVERT,
+};
+
+static inline void scsi_set_prot_op(struct scsi_cmnd *scmd, unsigned char op)
+{
+	scmd->prot_op = op;
+}
+
+static inline unsigned char scsi_get_prot_op(struct scsi_cmnd *scmd)
+{
+	return scmd->prot_op;
+}
+
+/*
+ * The controller usually does not know anything about the target it
+ * is communicating with.  However, when DIX is enabled the controller
+ * must be know target type so it can verify the protection
+ * information passed along with the I/O.
+ */
+enum scsi_prot_target_type {
+	SCSI_PROT_DIF_TYPE0 = 0,
+	SCSI_PROT_DIF_TYPE1,
+	SCSI_PROT_DIF_TYPE2,
+	SCSI_PROT_DIF_TYPE3,
+};
+
+static inline void scsi_set_prot_type(struct scsi_cmnd *scmd, unsigned char type)
+{
+	scmd->prot_type = type;
+}
+
+static inline unsigned char scsi_get_prot_type(struct scsi_cmnd *scmd)
+{
+	return scmd->prot_type;
+}
+
+static inline sector_t scsi_get_lba(struct scsi_cmnd *scmd)
+{
+	return scmd->request->sector;
+}
+
 #endif /* _SCSI_SCSI_CMND_H */
diff --git a/include/scsi/scsi_eh.h b/include/scsi/scsi_eh.h
--- a/include/scsi/scsi_eh.h
+++ b/include/scsi/scsi_eh.h
@@ -75,6 +75,7 @@ struct scsi_eh_save {
 	int result;
 	enum dma_data_direction data_direction;
 	unsigned char cmd_len;
+	unsigned char prot_op;
 	unsigned char *cmnd;
 	struct scsi_data_buffer sdb;
 	struct request *next_rq;

[PATCH 3 of 6] scsi: Support devices with protection information

[Martin K. Petersen]

Implement support for DMA of protection information for devices that
are data integrity capable.

 - Add support for mapping an extra scatter-gather list containing
   the protection information.

 - Handle protection scsi_data_buffer in command allocation.

 - Preallocate protection scsi_data_buffer in command freelist if host
   is DIX (integrity DMA) capable.

 - Accessor function for checking whether a device has protection
   enabled.

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

---
7 files changed, 97 insertions(+), 11 deletions(-)
drivers/scsi/scsi.c         |   47 ++++++++++++++++++++++++++++++++++++++-----
drivers/scsi/scsi_error.c   |    2 -
drivers/scsi/scsi_lib.c     |   27 ++++++++++++++++++++++--
drivers/scsi/scsi_priv.h    |    1 
drivers/scsi/scsi_tgt_lib.c |    2 -
include/scsi/scsi_cmnd.h    |   24 ++++++++++++++++++++-
include/scsi/scsi_device.h  |    5 ++++



diff --git a/drivers/scsi/scsi.c b/drivers/scsi/scsi.c
--- a/drivers/scsi/scsi.c
+++ b/drivers/scsi/scsi.c
@@ -199,24 +199,60 @@ scsi_pool_free_command(struct scsi_host_
 {
 	kmem_cache_free(pool->sense_slab, cmd->sense_buffer);
 	kmem_cache_free(pool->cmd_slab, cmd);
+
+	if (cmd->prot_sdb)
+		kmem_cache_free(scsi_sdb_cache, cmd->prot_sdb);
+}
+
+/**
+ * scsi_host_alloc_command - internal function to allocate command
+ * @shost:	SCSI host whose pool to allocate from
+ * @gfp_mask:	mask for the allocation
+ *
+ * Returns a fully allocated command with sense buffer and protection
+ * data buffer (where applicable) or NULL on failure
+ */
+static struct scsi_cmnd *
+scsi_host_alloc_command(struct Scsi_Host *shost, gfp_t gfp_mask)
+{
+	struct scsi_cmnd *cmd;
+
+	cmd = scsi_pool_alloc_command(shost->cmd_pool, gfp_mask);
+	if (!cmd)
+		return NULL;
+
+	if (scsi_host_get_prot(shost) >= SHOST_DIX_TYPE0_PROTECTION) {
+		cmd->prot_sdb = kmem_cache_zalloc(scsi_sdb_cache, gfp_mask);
+
+		if (!cmd->prot_sdb) {
+			scsi_pool_free_command(shost->cmd_pool, cmd);
+			return NULL;
+		}
+	}
+
+	return cmd;
 }
 
 /**
  * __scsi_get_command - Allocate a struct scsi_cmnd
  * @shost: host to transmit command
  * @gfp_mask: allocation mask
+ * @protection: allocate protection sdb
  *
  * Description: allocate a struct scsi_cmd from host's slab, recycling from the
  *              host's free_list if necessary.
  */
-struct scsi_cmnd *__scsi_get_command(struct Scsi_Host *shost, gfp_t gfp_mask)
+struct scsi_cmnd *__scsi_get_command(struct Scsi_Host *shost, gfp_t gfp_mask, int protection)
 {
 	struct scsi_cmnd *cmd;
 	unsigned char *buf;
 
 	cmd = scsi_pool_alloc_command(shost->cmd_pool, gfp_mask);
 
-	if (unlikely(!cmd)) {
+	if (protection)
+		cmd->prot_sdb = kmem_cache_zalloc(scsi_sdb_cache, gfp_mask);
+
+	if (unlikely(!cmd) || unlikely(protection && cmd->prot_sdb == NULL)) {
 		unsigned long flags;
 
 		spin_lock_irqsave(&shost->free_list_lock, flags);
@@ -242,10 +278,11 @@ EXPORT_SYMBOL_GPL(__scsi_get_command);
  * scsi_get_command - Allocate and setup a scsi command block
  * @dev: parent scsi device
  * @gfp_mask: allocator flags
+ * @protection: allocate protection sdb
  *
  * Returns:	The allocated scsi command structure.
  */
-struct scsi_cmnd *scsi_get_command(struct scsi_device *dev, gfp_t gfp_mask)
+struct scsi_cmnd *scsi_get_command(struct scsi_device *dev, gfp_t gfp_mask, int protection)
 {
 	struct scsi_cmnd *cmd;
 
@@ -253,7 +290,7 @@ struct scsi_cmnd *scsi_get_command(struc
 	if (!get_device(&dev->sdev_gendev))
 		return NULL;
 
-	cmd = __scsi_get_command(dev->host, gfp_mask);
+	cmd = __scsi_get_command(dev->host, gfp_mask, protection);
 
 	if (likely(cmd != NULL)) {
 		unsigned long flags;
@@ -457,7 +494,7 @@ int scsi_setup_command_freelist(struct S
 	/*
 	 * Get one backup command for this host.
 	 */
-	cmd = scsi_pool_alloc_command(shost->cmd_pool, gfp_mask);
+	cmd = scsi_host_alloc_command(shost, gfp_mask);
 	if (!cmd) {
 		scsi_put_host_cmd_pool(gfp_mask);
 		shost->cmd_pool = NULL;
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -1780,7 +1780,7 @@ int
 int
 scsi_reset_provider(struct scsi_device *dev, int flag)
 {
-	struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
+	struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL, 0);
 	struct Scsi_Host *shost = dev->host;
 	struct request req;
 	unsigned long flags;
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -65,7 +65,7 @@ static struct scsi_host_sg_pool scsi_sg_
 };
 #undef SP
 
-static struct kmem_cache *scsi_sdb_cache;
+struct kmem_cache *scsi_sdb_cache;
 
 static void scsi_run_queue(struct request_queue *q);
 
@@ -787,6 +787,9 @@ void scsi_release_buffers(struct scsi_cm
 		kmem_cache_free(scsi_sdb_cache, bidi_sdb);
 		cmd->request->next_rq->special = NULL;
 	}
+
+	if (scsi_prot_sg_count(cmd))
+		scsi_free_sgtable(cmd->prot_sdb);
 }
 EXPORT_SYMBOL(scsi_release_buffers);
 
@@ -1072,6 +1075,26 @@ int scsi_init_io(struct scsi_cmnd *cmd, 
 			goto err_exit;
 	}
 
+	if (blk_integrity_rq(cmd->request)) {
+		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
+		int ivecs, count;
+
+		BUG_ON(prot_sdb == NULL);
+		ivecs = blk_rq_count_integrity_sg(cmd->request);
+
+		if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
+			error = BLKPREP_DEFER;
+			goto err_exit;
+		}
+
+		count = blk_rq_map_integrity_sg(cmd->request,
+						prot_sdb->table.sgl);
+		BUG_ON(unlikely(count > ivecs));
+
+		cmd->prot_sdb = prot_sdb;
+		cmd->prot_sdb->table.nents = count;
+	}
+
 	return BLKPREP_OK ;
 
 err_exit:
@@ -1091,7 +1114,7 @@ static struct scsi_cmnd *scsi_get_cmd_fr
 	struct scsi_cmnd *cmd;
 
 	if (!req->special) {
-		cmd = scsi_get_command(sdev, GFP_ATOMIC);
+		cmd = scsi_get_command(sdev, GFP_ATOMIC, blk_integrity_rq(req));
 		if (unlikely(!cmd))
 			return NULL;
 		req->special = cmd;
diff --git a/drivers/scsi/scsi_priv.h b/drivers/scsi/scsi_priv.h
--- a/drivers/scsi/scsi_priv.h
+++ b/drivers/scsi/scsi_priv.h
@@ -77,6 +77,7 @@ struct request_queue;
 struct request_queue;
 struct request;
 extern int scsi_prep_fn(struct request_queue *, struct request *);
+extern struct kmem_cache *scsi_sdb_cache;
 
 /* scsi_proc.c */
 #ifdef CONFIG_SCSI_PROC_FS
diff --git a/drivers/scsi/scsi_tgt_lib.c b/drivers/scsi/scsi_tgt_lib.c
--- a/drivers/scsi/scsi_tgt_lib.c
+++ b/drivers/scsi/scsi_tgt_lib.c
@@ -99,7 +99,7 @@ struct scsi_cmnd *scsi_host_get_command(
 	if (!rq)
 		goto free_tcmd;
 
-	cmd = __scsi_get_command(shost, gfp_mask);
+	cmd = __scsi_get_command(shost, gfp_mask, 0);
 	if (!cmd)
 		goto release_rq;
 
diff --git a/include/scsi/scsi_cmnd.h b/include/scsi/scsi_cmnd.h
--- a/include/scsi/scsi_cmnd.h
+++ b/include/scsi/scsi_cmnd.h
@@ -90,6 +90,8 @@ struct scsi_cmnd {
 
 	/* These elements define the operation we ultimately want to perform */
 	struct scsi_data_buffer sdb;
+	struct scsi_data_buffer *prot_sdb;
+
 	unsigned underflow;	/* Return error if less than
 				   this amount is transferred */
 
@@ -131,8 +133,8 @@ struct scsi_cmnd {
 	unsigned char tag;	/* SCSI-II queued command tag */
 };
 
-extern struct scsi_cmnd *scsi_get_command(struct scsi_device *, gfp_t);
-extern struct scsi_cmnd *__scsi_get_command(struct Scsi_Host *, gfp_t);
+extern struct scsi_cmnd *scsi_get_command(struct scsi_device *, gfp_t, int);
+extern struct scsi_cmnd *__scsi_get_command(struct Scsi_Host *, gfp_t, int);
 extern void scsi_put_command(struct scsi_cmnd *);
 extern void __scsi_put_command(struct Scsi_Host *, struct scsi_cmnd *,
 			       struct device *);
@@ -274,4 +276,22 @@ static inline sector_t scsi_get_lba(stru
 	return scmd->request->sector;
 }
 
+static inline unsigned scsi_prot_sg_count(struct scsi_cmnd *cmd)
+{
+	return cmd->prot_sdb ? cmd->prot_sdb->table.nents : 0;
+}
+
+static inline struct scatterlist *scsi_prot_sglist(struct scsi_cmnd *cmd)
+{
+	return cmd->prot_sdb ? cmd->prot_sdb->table.sgl : NULL;
+}
+
+static inline struct scsi_data_buffer *scsi_prot(struct scsi_cmnd *cmd)
+{
+	return cmd->prot_sdb;
+}
+
+#define scsi_for_each_prot_sg(cmd, sg, nseg, __i)		\
+	for_each_sg(scsi_prot_sglist(cmd), sg, nseg, __i)
+
 #endif /* _SCSI_SCSI_CMND_H */
diff --git a/include/scsi/scsi_device.h b/include/scsi/scsi_device.h
--- a/include/scsi/scsi_device.h
+++ b/include/scsi/scsi_device.h
@@ -416,6 +416,11 @@ static inline int scsi_device_enclosure(
 	return sdev->inquiry[6] & (1<<6);
 }
 
+static inline int scsi_device_protection(struct scsi_device *sdev)
+{
+	return sdev->inquiry[5] & (1<<0);
+}
+
 #define MODULE_ALIAS_SCSI_DEVICE(type) \
 	MODULE_ALIAS("scsi:t-" __stringify(type) "*")
 #define SCSI_DEVICE_MODALIAS_FMT "scsi:t-0x%02x"

[PATCH 4 of 6] scsi: Do not retry a request whose data integrity check failed

[Martin K. Petersen]

If initiator or target reject the I/O due to DIF errors there is no
point in retrying.

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

---
2 files changed, 10 insertions(+), 2 deletions(-)
drivers/scsi/scsi_error.c |    3 +++
drivers/scsi/scsi_lib.c   |    9 +++++++--



diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -344,6 +344,9 @@ static int scsi_check_sense(struct scsi_
 		return /* soft_error */ SUCCESS;
 
 	case ABORTED_COMMAND:
+		if (sshdr.asc == 0x10) /* DIF */
+			return SUCCESS;
+
 		return NEEDS_RETRY;
 	case NOT_READY:
 	case UNIT_ATTENTION:
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -950,9 +950,14 @@ void scsi_io_completion(struct scsi_cmnd
 				 * 6-byte command.
 				 */
 				scsi_requeue_command(q, cmd);
-				return;
-			} else {
+			} else if (sshdr.asc == 0x10) /* DIX */
+				scsi_end_request(cmd, -EIO, this_count, 0);
+			else
 				scsi_end_request(cmd, -EIO, this_count, 1);
+			return;
+		case ABORTED_COMMAND:
+			if (sshdr.asc == 0x10) { /* DIF */
+				scsi_end_request(cmd, -EIO, this_count, 0);
 				return;
 			}
 			break;

[PATCH 5 of 6] sd: Identify DIF protection type and application tag ownership

[Martin K. Petersen]

If a disk is formatted with protection information (Inquiry bit
PROTECT=1) it is required to support Read Capacity(16).  Force use of
the 16-bit command in this case and extract the P_TYPE field which
indicates whether the disk is formatted using DIF Type 1, 2 or 3.

The ATO (App Tag Own) bit in the Control Mode Page indicates whether
the storage device or the initiator own the contents of the
DIF application tag.

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

---
2 files changed, 141 insertions(+), 4 deletions(-)
drivers/scsi/sd.c |  122 +++++++++++++++++++++++++++++++++++++++++++++++++++--
drivers/scsi/sd.h |   23 +++++++++



diff --git a/drivers/scsi/sd.c b/drivers/scsi/sd.c
--- a/drivers/scsi/sd.c
+++ b/drivers/scsi/sd.c
@@ -234,6 +234,24 @@ sd_show_allow_restart(struct device *dev
 	return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
 }
 
+static ssize_t
+sd_show_protection_type(struct device *dev, struct device_attribute *attr,
+			char *buf)
+{
+	struct scsi_disk *sdkp = to_scsi_disk(dev);
+
+	return snprintf(buf, 20, "%u\n", sdkp->protection_type);
+}
+
+static ssize_t
+sd_show_app_tag_own(struct device *dev, struct device_attribute *attr,
+		    char *buf)
+{
+	struct scsi_disk *sdkp = to_scsi_disk(dev);
+
+	return snprintf(buf, 20, "%u\n", sdkp->ATO);
+}
+
 static struct device_attribute sd_disk_attrs[] = {
 	__ATTR(cache_type, S_IRUGO|S_IWUSR, sd_show_cache_type,
 	       sd_store_cache_type),
@@ -242,6 +260,8 @@ static struct device_attribute sd_disk_a
 	       sd_store_allow_restart),
 	__ATTR(manage_start_stop, S_IRUGO|S_IWUSR, sd_show_manage_start_stop,
 	       sd_store_manage_start_stop),
+	__ATTR(protection_type, S_IRUGO, sd_show_protection_type, NULL),
+	__ATTR(app_tag_own, S_IRUGO, sd_show_app_tag_own, NULL),
 	__ATTR_NULL,
 };
 
@@ -1165,6 +1185,49 @@ sd_spinup_disk(struct scsi_disk *sdkp)
 	}
 }
 
+
+/*
+ * Determine whether disk supports Data Integrity Field.
+ */
+void sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
+{
+	struct scsi_device *sdp = sdkp->device;
+	u8 type;
+
+	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
+		type = 0;
+	else
+		type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
+
+	switch (type) {
+	case SD_DIF_TYPE0_PROTECTION:
+		sdkp->protection_type = 0;
+		break;
+
+	case SD_DIF_TYPE1_PROTECTION:
+	case SD_DIF_TYPE3_PROTECTION:
+		sdkp->protection_type = type;
+		break;
+
+	case SD_DIF_TYPE2_PROTECTION:
+		sd_printk(KERN_ERR, sdkp, "formatted with DIF Type 2 "	\
+			  "protection which is currently unsupported. "	\
+			  "Disabling disk!\n");
+		goto disable;
+
+	default:
+		sd_printk(KERN_ERR, sdkp, "formatted with unknown "	\
+			  "protection type %d. Disabling disk!\n", type);
+		goto disable;
+	}
+
+	return;
+
+disable:
+	sdkp->protection_type = 0;
+	sdkp->capacity = 0;
+}
+
 /*
  * read disk capacity
  */
@@ -1174,7 +1237,8 @@ sd_read_capacity(struct scsi_disk *sdkp,
 	unsigned char cmd[16];
 	int the_result, retries;
 	int sector_size = 0;
-	int longrc = 0;
+	/* Force READ CAPACITY(16) when PROTECT=1 */
+	int longrc = scsi_device_protection(sdkp->device) ? 1 : 0;
 	struct scsi_sense_hdr sshdr;
 	int sense_valid = 0;
 	struct scsi_device *sdp = sdkp->device;
@@ -1186,8 +1250,8 @@ repeat:
 			memset((void *) cmd, 0, 16);
 			cmd[0] = SERVICE_ACTION_IN;
 			cmd[1] = SAI_READ_CAPACITY_16;
-			cmd[13] = 12;
-			memset((void *) buffer, 0, 12);
+			cmd[13] = 13;
+			memset((void *) buffer, 0, 13);
 		} else {
 			cmd[0] = READ_CAPACITY;
 			memset((void *) &cmd[1], 0, 9);
@@ -1195,7 +1259,7 @@ repeat:
 		}
 		
 		the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
-					      buffer, longrc ? 12 : 8, &sshdr,
+					      buffer, longrc ? 13 : 8, &sshdr,
 					      SD_TIMEOUT, SD_MAX_RETRIES);
 
 		if (media_not_present(sdkp, &sshdr))
@@ -1270,6 +1334,8 @@ repeat:
 			
 		sector_size = (buffer[8] << 24) |
 			(buffer[9] << 16) | (buffer[10] << 8) | buffer[11];
+
+		sd_read_protection_type(sdkp, buffer);
 	}	
 
 	/* Some devices return the total number of sectors, not the
@@ -1531,6 +1597,52 @@ defaults:
 	sdkp->DPOFUA = 0;
 }
 
+/*
+ * The ATO bit indicates whether the DIF application tag is available
+ * for use by the operating system.
+ */
+void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
+{
+	int res, offset;
+	struct scsi_device *sdp = sdkp->device;
+	struct scsi_mode_data data;
+	struct scsi_sense_hdr sshdr;
+
+	if (sdp->type != TYPE_DISK)
+		return;
+
+	if (sdkp->protection_type == 0)
+		return;
+
+	res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
+			      SD_MAX_RETRIES, &data, &sshdr);
+
+	if (!scsi_status_is_good(res) || !data.header_length ||
+	    data.length < 6) {
+		sd_printk(KERN_WARNING, sdkp,
+			  "getting Control mode page failed, assume no ATO\n");
+
+		if (scsi_sense_valid(&sshdr))
+			sd_print_sense_hdr(sdkp, &sshdr);
+
+		return;
+	}
+
+	offset = data.header_length + data.block_descriptor_length;
+
+	if ((buffer[offset] & 0x3f) != 0x0a) {
+		sd_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
+		return;
+	}
+
+	if ((buffer[offset + 5] & 0x80) == 0)
+		return;
+
+	sdkp->ATO = 1;
+
+	return;
+}
+
 /**
  *	sd_revalidate_disk - called the first time a new disk is seen,
  *	performs disk spin up, read_capacity, etc.
@@ -1567,6 +1679,7 @@ static int sd_revalidate_disk(struct gen
 	sdkp->write_prot = 0;
 	sdkp->WCE = 0;
 	sdkp->RCD = 0;
+	sdkp->ATO = 0;
 
 	sd_spinup_disk(sdkp);
 
@@ -1578,6 +1691,7 @@ static int sd_revalidate_disk(struct gen
 		sd_read_capacity(sdkp, buffer);
 		sd_read_write_protect_flag(sdkp, buffer);
 		sd_read_cache_type(sdkp, buffer);
+		sd_read_app_tag_own(sdkp, buffer);
 	}
 
 	/*
diff --git a/drivers/scsi/sd.h b/drivers/scsi/sd.h
--- a/drivers/scsi/sd.h
+++ b/drivers/scsi/sd.h
@@ -41,7 +41,9 @@ struct scsi_disk {
 	u32		index;
 	u8		media_present;
 	u8		write_prot;
+	u8		protection_type;/* Data Integrity Field */
 	unsigned	previous_state : 1;
+	unsigned	ATO : 1;	/* state of disk ATO bit */
 	unsigned	WCE : 1;	/* state of disk WCE bit */
 	unsigned	RCD : 1;	/* state of disk RCD bit, unused */
 	unsigned	DPOFUA : 1;	/* state of disk DPOFUA bit */
@@ -59,4 +61,25 @@ static inline struct scsi_disk *scsi_dis
 		    (sdsk)->disk->disk_name, ##a) :			\
 	sdev_printk(prefix, (sdsk)->device, fmt, ##a)
 
+/*
+ * A DIF-capable target device can be formatted with different
+ * protection schemes.  Currently 0 through 3 are defined:
+ *
+ * Type 0 is regular (unprotected) I/O
+ *
+ * Type 1 defines the contents of the guard and reference tags
+ *
+ * Type 2 defines the contents of the guard and reference tags and
+ * uses 32-byte commands to seed the latter
+ *
+ * Type 3 defines the contents of the guard tag only
+ */
+
+enum sd_dif_target_protection_types {
+	SD_DIF_TYPE0_PROTECTION = 0x0,
+	SD_DIF_TYPE1_PROTECTION = 0x1,
+	SD_DIF_TYPE2_PROTECTION = 0x2,
+	SD_DIF_TYPE3_PROTECTION = 0x3,
+};
+
 #endif /* _SCSI_DISK_H */

[PATCH 6 of 6] sd: Support for SCSI disk (SBC) Data Integrity Field

[Martin K. Petersen]

Support for controllers and disks that implement DIF protection
information:

 - During command preparation the RDPROTECT/WRPROTECT must be set
   correctly if the target has DIF enabled.

 - READ(6) and WRITE(6) are not supported when DIF is on.

 - The controller must be told how to handle the I/O via the
   protection operation field in scsi_cmnd.

 - Refactor the I/O completion code that extracts failed LBA from the
   returned sense data and handle DIF failures correctly.

 - sd_dif.c implements the functions required to prepare and complete
   requests with protection information attached.

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

---
5 files changed, 647 insertions(+), 40 deletions(-)
drivers/scsi/Kconfig  |    1 
drivers/scsi/Makefile |    2 
drivers/scsi/sd.c     |  121 +++++++----
drivers/scsi/sd.h     |   25 ++
drivers/scsi/sd_dif.c |  538 +++++++++++++++++++++++++++++++++++++++++++++++++



diff --git a/drivers/scsi/Kconfig b/drivers/scsi/Kconfig
--- a/drivers/scsi/Kconfig
+++ b/drivers/scsi/Kconfig
@@ -63,6 +63,7 @@ config BLK_DEV_SD
 config BLK_DEV_SD
 	tristate "SCSI disk support"
 	depends on SCSI
+	select CRC_T10DIF
 	---help---
 	  If you want to use SCSI hard disks, Fibre Channel disks,
 	  Serial ATA (SATA) or Parallel ATA (PATA) hard disks,
diff --git a/drivers/scsi/Makefile b/drivers/scsi/Makefile
--- a/drivers/scsi/Makefile
+++ b/drivers/scsi/Makefile
@@ -151,6 +151,8 @@ scsi_tgt-y			+= scsi_tgt_lib.o scsi_tgt_
 scsi_tgt-y			+= scsi_tgt_lib.o scsi_tgt_if.o
 
 sd_mod-objs	:= sd.o
+sd_mod-$(CONFIG_BLK_DEV_INTEGRITY) += sd_dif.o
+
 sr_mod-objs	:= sr.o sr_ioctl.o sr_vendor.o
 ncr53c8xx-flags-$(CONFIG_SCSI_ZALON) \
 		:= -DCONFIG_NCR53C8XX_PREFETCH -DSCSI_NCR_BIG_ENDIAN \
diff --git a/drivers/scsi/sd.c b/drivers/scsi/sd.c
--- a/drivers/scsi/sd.c
+++ b/drivers/scsi/sd.c
@@ -374,6 +374,7 @@ static int sd_prep_fn(struct request_que
 	struct scsi_cmnd *SCpnt;
 	struct scsi_device *sdp = q->queuedata;
 	struct gendisk *disk = rq->rq_disk;
+	struct scsi_disk *sdkp;
 	sector_t block = rq->sector;
 	unsigned int this_count = rq->nr_sectors;
 	unsigned int timeout = sdp->timeout;
@@ -390,6 +391,7 @@ static int sd_prep_fn(struct request_que
 	if (ret != BLKPREP_OK)
 		goto out;
 	SCpnt = rq->special;
+	sdkp = scsi_disk(disk);
 
 	/* from here on until we're complete, any goto out
 	 * is used for a killable error condition */
@@ -479,6 +481,11 @@ static int sd_prep_fn(struct request_que
 		}
 		SCpnt->cmnd[0] = WRITE_6;
 		SCpnt->sc_data_direction = DMA_TO_DEVICE;
+
+		if (blk_integrity_rq(rq) &&
+		    sd_dif_prepare(rq, block, sdp->sector_size) == -EIO)
+			goto out;
+
 	} else if (rq_data_dir(rq) == READ) {
 		SCpnt->cmnd[0] = READ_6;
 		SCpnt->sc_data_direction = DMA_FROM_DEVICE;
@@ -493,8 +500,12 @@ static int sd_prep_fn(struct request_que
 					"writing" : "reading", this_count,
 					rq->nr_sectors));
 
-	SCpnt->cmnd[1] = 0;
-	
+	/* Set RDPROTECT/WRPROTECT if disk is formatted with DIF */
+	if (scsi_host_dif_capable(sdp->host, sdkp->protection_type))
+		SCpnt->cmnd[1] = 1 << 5;
+	else
+		SCpnt->cmnd[1] = 0;
+
 	if (block > 0xffffffff) {
 		SCpnt->cmnd[0] += READ_16 - READ_6;
 		SCpnt->cmnd[1] |= blk_fua_rq(rq) ? 0x8 : 0;
@@ -512,6 +523,7 @@ static int sd_prep_fn(struct request_que
 		SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
 		SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
 	} else if ((this_count > 0xff) || (block > 0x1fffff) ||
+		   scsi_device_protection(SCpnt->device) ||
 		   SCpnt->device->use_10_for_rw) {
 		if (this_count > 0xffff)
 			this_count = 0xffff;
@@ -545,6 +557,10 @@ static int sd_prep_fn(struct request_que
 		SCpnt->cmnd[5] = 0;
 	}
 	SCpnt->sdb.length = this_count * sdp->sector_size;
+
+	/* If DIF or DIX is enabled, tell HBA how to handle request */
+	if (sdkp->protection_type || scsi_prot_sg_count(SCpnt))
+		sd_dif_op(SCpnt, sdkp->protection_type, scsi_prot_sg_count(SCpnt));
 
 	/*
 	 * We shouldn't disconnect in the middle of a sector, so with a dumb
@@ -940,6 +956,48 @@ static struct block_device_operations sd
 	.revalidate_disk	= sd_revalidate_disk,
 };
 
+static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
+{
+	u64 start_lba = scmd->request->sector;
+	u64 end_lba = scmd->request->sector + (scsi_bufflen(scmd) / 512);
+	u64 bad_lba;
+	int info_valid;
+
+	if (!blk_fs_request(scmd->request))
+		return 0;
+
+	info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
+					     SCSI_SENSE_BUFFERSIZE,
+					     &bad_lba);
+	if (!info_valid)
+		return 0;
+
+	if (scsi_bufflen(scmd) <= scmd->device->sector_size)
+		return 0;
+
+	if (scmd->device->sector_size < 512) {
+		/* only legitimate sector_size here is 256 */
+		start_lba <<= 1;
+		end_lba <<= 1;
+	} else {
+		/* be careful ... don't want any overflows */
+		u64 factor = scmd->device->sector_size / 512;
+		do_div(start_lba, factor);
+		do_div(end_lba, factor);
+	}
+
+	/* The bad lba was reported incorrectly, we have no idea where
+	 * the error is.
+	 */
+	if (bad_lba < start_lba  || bad_lba >= end_lba)
+		return 0;
+
+	/* This computation should always be done in terms of
+	 * the resolution of the device's medium.
+	 */
+	return (bad_lba - start_lba) * scmd->device->sector_size;
+}
+
 /**
  *	sd_done - bottom half handler: called when the lower level
  *	driver has completed (successfully or otherwise) a scsi command.
@@ -950,15 +1008,10 @@ static int sd_done(struct scsi_cmnd *SCp
 static int sd_done(struct scsi_cmnd *SCpnt)
 {
 	int result = SCpnt->result;
-	unsigned int xfer_size = scsi_bufflen(SCpnt);
- 	unsigned int good_bytes = result ? 0 : xfer_size;
- 	u64 start_lba = SCpnt->request->sector;
-	u64 end_lba = SCpnt->request->sector + (xfer_size / 512);
- 	u64 bad_lba;
+	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
 	struct scsi_sense_hdr sshdr;
 	int sense_valid = 0;
 	int sense_deferred = 0;
-	int info_valid;
 
 	if (result) {
 		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
@@ -983,36 +1036,7 @@ static int sd_done(struct scsi_cmnd *SCp
 	switch (sshdr.sense_key) {
 	case HARDWARE_ERROR:
 	case MEDIUM_ERROR:
-		if (!blk_fs_request(SCpnt->request))
-			goto out;
-		info_valid = scsi_get_sense_info_fld(SCpnt->sense_buffer,
-						     SCSI_SENSE_BUFFERSIZE,
-						     &bad_lba);
-		if (!info_valid)
-			goto out;
-		if (xfer_size <= SCpnt->device->sector_size)
-			goto out;
-		if (SCpnt->device->sector_size < 512) {
-			/* only legitimate sector_size here is 256 */
-			start_lba <<= 1;
-			end_lba <<= 1;
-		} else {
-			/* be careful ... don't want any overflows */
-			u64 factor = SCpnt->device->sector_size / 512;
-			do_div(start_lba, factor);
-			do_div(end_lba, factor);
-		}
-
-		if (bad_lba < start_lba  || bad_lba >= end_lba)
-			/* the bad lba was reported incorrectly, we have
-			 * no idea where the error is
-			 */
-			goto out;
-
-		/* This computation should always be done in terms of
-		 * the resolution of the device's medium.
-		 */
-		good_bytes = (bad_lba - start_lba)*SCpnt->device->sector_size;
+		good_bytes = sd_completed_bytes(SCpnt);
 		break;
 	case RECOVERED_ERROR:
 	case NO_SENSE:
@@ -1022,10 +1046,23 @@ static int sd_done(struct scsi_cmnd *SCp
 		scsi_print_sense("sd", SCpnt);
 		SCpnt->result = 0;
 		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
-		good_bytes = xfer_size;
+		good_bytes = scsi_bufflen(SCpnt);
+		break;
+	case ABORTED_COMMAND:
+		if (sshdr.asc == 0x10) { /* DIF: Disk detected corruption */
+			scsi_print_result(SCpnt);
+			scsi_print_sense("sd", SCpnt);
+			good_bytes = sd_completed_bytes(SCpnt);
+		}
 		break;
 	case ILLEGAL_REQUEST:
-		if (SCpnt->device->use_10_for_rw &&
+		if (sshdr.asc == 0x10) { /* DIX: HBA detected corruption */
+			scsi_print_result(SCpnt);
+			scsi_print_sense("sd", SCpnt);
+			good_bytes = sd_completed_bytes(SCpnt);
+		}
+		if (!scsi_device_protection(SCpnt->device) &&
+		    SCpnt->device->use_10_for_rw &&
 		    (SCpnt->cmnd[0] == READ_10 ||
 		     SCpnt->cmnd[0] == WRITE_10))
 			SCpnt->device->use_10_for_rw = 0;
@@ -1038,6 +1075,9 @@ static int sd_done(struct scsi_cmnd *SCp
 		break;
 	}
  out:
+	if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
+		sd_dif_complete(SCpnt, good_bytes);
+
 	return good_bytes;
 }
 
@@ -1825,6 +1865,7 @@ static int sd_probe(struct device *dev)
 
 	dev_set_drvdata(dev, sdkp);
 	add_disk(gd);
+	sd_dif_config_host(sdkp);
 
 	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
 		  sdp->removable ? "removable " : "");
diff --git a/drivers/scsi/sd.h b/drivers/scsi/sd.h
--- a/drivers/scsi/sd.h
+++ b/drivers/scsi/sd.h
@@ -82,4 +82,29 @@ enum sd_dif_target_protection_types {
 	SD_DIF_TYPE3_PROTECTION = 0x3,
 };
 
+/*
+ * Data Integrity Field tuple.
+ */
+struct sd_dif_tuple {
+       __be16 guard_tag;	/* Checksum */
+       __be16 app_tag;		/* Opaque storage */
+       __be32 ref_tag;		/* Target LBA or indirect LBA */
+};
+
+#if defined(CONFIG_BLK_DEV_INTEGRITY)
+
+extern void sd_dif_op(struct scsi_cmnd *, unsigned int, unsigned int);
+extern void sd_dif_config_host(struct scsi_disk *);
+extern int sd_dif_prepare(struct request *rq, sector_t, unsigned int);
+extern void sd_dif_complete(struct scsi_cmnd *, unsigned int);
+
+#else /* CONFIG_BLK_DEV_INTEGRITY */
+
+#define sd_dif_op(a, b, c)			do { } while (0)
+#define sd_dif_config_host(a)			do { } while (0)
+#define sd_dif_prepare(a, b, c)			(0)
+#define sd_dif_complete(a, b)			(0)
+
+#endif /* CONFIG_BLK_DEV_INTEGRITY */
+
 #endif /* _SCSI_DISK_H */
diff --git a/drivers/scsi/sd_dif.c b/drivers/scsi/sd_dif.c
new file mode 100644
--- /dev/null
+++ b/drivers/scsi/sd_dif.c
@@ -0,0 +1,538 @@
+/*
+ * sd_dif.c - SCSI Data Integrity Field
+ *
+ * Copyright (C) 2007, 2008 Oracle Corporation
+ * Written by: Martin K. Petersen <martin.petersen@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License version
+ * 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; see the file COPYING.  If not, write to
+ * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
+ * USA.
+ *
+ */
+
+#include <linux/blkdev.h>
+#include <linux/crc-t10dif.h>
+
+#include <scsi/scsi.h>
+#include <scsi/scsi_cmnd.h>
+#include <scsi/scsi_dbg.h>
+#include <scsi/scsi_device.h>
+#include <scsi/scsi_driver.h>
+#include <scsi/scsi_eh.h>
+#include <scsi/scsi_host.h>
+#include <scsi/scsi_ioctl.h>
+#include <scsi/scsicam.h>
+
+#include <net/checksum.h>
+
+#include "sd.h"
+
+typedef __u16 (csum_fn) (void *, unsigned int);
+
+static __u16 sd_dif_crc_fn(void *data, unsigned int len)
+{
+	return cpu_to_be16(crc_t10dif(data, len));
+}
+
+static __u16 sd_dif_ip_fn(void *data, unsigned int len)
+{
+	return ip_compute_csum(data, len);
+}
+
+/*
+ * Type 1 and Type 2 protection use the same format: 16 bit guard tag,
+ * 16 bit app tag, 32 bit reference tag.
+ */
+static void sd_dif_type1_generate(struct blk_integrity_exchg *bix, csum_fn *fn)
+{
+	void *buf = bix->data_buf;
+	struct sd_dif_tuple *sdt = bix->prot_buf;
+	sector_t sector = bix->sector;
+	unsigned int i;
+
+	for (i = 0 ; i < bix->data_size ; i += bix->sector_size, sdt++) {
+		sdt->guard_tag = fn(buf, bix->sector_size);
+		sdt->ref_tag = cpu_to_be32(sector & 0xffffffff);
+		sdt->app_tag = 0;
+
+		buf += bix->sector_size;
+		sector++;
+	}
+}
+
+static void sd_dif_type1_generate_crc(struct blk_integrity_exchg *bix)
+{
+	sd_dif_type1_generate(bix, sd_dif_crc_fn);
+}
+
+static void sd_dif_type1_generate_ip(struct blk_integrity_exchg *bix)
+{
+	sd_dif_type1_generate(bix, sd_dif_ip_fn);
+}
+
+static int sd_dif_type1_verify(struct blk_integrity_exchg *bix, csum_fn *fn)
+{
+	void *buf = bix->data_buf;
+	struct sd_dif_tuple *sdt = bix->prot_buf;
+	sector_t sector = bix->sector;
+	unsigned int i;
+	__u16 csum;
+
+	for (i = 0 ; i < bix->data_size ; i += bix->sector_size, sdt++) {
+		/* Unwritten sectors */
+		if (sdt->app_tag == 0xffff)
+			return 0;
+
+		/* Bad ref tag received from disk */
+		if (sdt->ref_tag == 0xffffffff) {
+			printk(KERN_ERR
+			       "%s: bad phys ref tag on sector %lu\n",
+			       bix->disk_name, (unsigned long)sector);
+			return -EIO;
+		}
+
+		if (be32_to_cpu(sdt->ref_tag) != (sector & 0xffffffff)) {
+			printk(KERN_ERR
+			       "%s: ref tag error on sector %lu (rcvd %u)\n",
+			       bix->disk_name, (unsigned long)sector,
+			       be32_to_cpu(sdt->ref_tag));
+			return -EIO;
+		}
+
+		csum = fn(buf, bix->sector_size);
+
+		if (sdt->guard_tag != csum) {
+			printk(KERN_ERR "%s: guard tag error on sector %lu " \
+			       "(rcvd %04x, data %04x)\n", bix->disk_name,
+			       (unsigned long)sector,
+			       be16_to_cpu(sdt->guard_tag), be16_to_cpu(csum));
+			return -EIO;
+		}
+
+		buf += bix->sector_size;
+		sector++;
+	}
+
+	return 0;
+}
+
+static int sd_dif_type1_verify_crc(struct blk_integrity_exchg *bix)
+{
+	return sd_dif_type1_verify(bix, sd_dif_crc_fn);
+}
+
+static int sd_dif_type1_verify_ip(struct blk_integrity_exchg *bix)
+{
+	return sd_dif_type1_verify(bix, sd_dif_ip_fn);
+}
+
+/*
+ * Functions for interleaving and deinterleaving application tags
+ */
+static void sd_dif_type1_set_tag(void *prot, void *tag_buf, unsigned int sectors)
+{
+	struct sd_dif_tuple *sdt = prot;
+	char *tag = tag_buf;
+	unsigned int i, j;
+
+	for (i = 0, j = 0 ; i < sectors ; i++, j += 2, sdt++) {
+		sdt->app_tag = tag[j] << 8 | tag[j+1];
+		BUG_ON(sdt->app_tag == 0xffff);
+	}
+}
+
+static void sd_dif_type1_get_tag(void *prot, void *tag_buf, unsigned int sectors)
+{
+	struct sd_dif_tuple *sdt = prot;
+	char *tag = tag_buf;
+	unsigned int i, j;
+
+	for (i = 0, j = 0 ; i < sectors ; i++, j += 2, sdt++) {
+		tag[j] = (sdt->app_tag & 0xff00) >> 8;
+		tag[j+1] = sdt->app_tag & 0xff;
+	}
+}
+
+static struct blk_integrity dif_type1_integrity_crc = {
+	.name			= "T10-DIF-TYPE1-CRC",
+	.generate_fn		= sd_dif_type1_generate_crc,
+	.verify_fn		= sd_dif_type1_verify_crc,
+	.get_tag_fn		= sd_dif_type1_get_tag,
+	.set_tag_fn		= sd_dif_type1_set_tag,
+	.tuple_size		= sizeof(struct sd_dif_tuple),
+	.tag_size		= 0,
+};
+
+static struct blk_integrity dif_type1_integrity_ip = {
+	.name			= "T10-DIF-TYPE1-IP",
+	.generate_fn		= sd_dif_type1_generate_ip,
+	.verify_fn		= sd_dif_type1_verify_ip,
+	.get_tag_fn		= sd_dif_type1_get_tag,
+	.set_tag_fn		= sd_dif_type1_set_tag,
+	.tuple_size		= sizeof(struct sd_dif_tuple),
+	.tag_size		= 0,
+};
+
+
+/*
+ * Type 3 protection has a 16-bit guard tag and 16 + 32 bits of opaque
+ * tag space.
+ */
+static void sd_dif_type3_generate(struct blk_integrity_exchg *bix, csum_fn *fn)
+{
+	void *buf = bix->data_buf;
+	struct sd_dif_tuple *sdt = bix->prot_buf;
+	unsigned int i;
+
+	for (i = 0 ; i < bix->data_size ; i += bix->sector_size, sdt++) {
+		sdt->guard_tag = fn(buf, bix->sector_size);
+		sdt->ref_tag = 0;
+		sdt->app_tag = 0;
+
+		buf += bix->sector_size;
+	}
+}
+
+static void sd_dif_type3_generate_crc(struct blk_integrity_exchg *bix)
+{
+	sd_dif_type3_generate(bix, sd_dif_crc_fn);
+}
+
+static void sd_dif_type3_generate_ip(struct blk_integrity_exchg *bix)
+{
+	sd_dif_type3_generate(bix, sd_dif_ip_fn);
+}
+
+static int sd_dif_type3_verify(struct blk_integrity_exchg *bix, csum_fn *fn)
+{
+	void *buf = bix->data_buf;
+	struct sd_dif_tuple *sdt = bix->prot_buf;
+	sector_t sector = bix->sector;
+	unsigned int i;
+	__u16 csum;
+
+	for (i = 0 ; i < bix->data_size ; i += bix->sector_size, sdt++) {
+		/* Unwritten sectors */
+		if (sdt->app_tag == 0xffff && sdt->ref_tag == 0xffffffff)
+			return 0;
+
+		csum = fn(buf, bix->sector_size);
+
+		if (sdt->guard_tag != csum) {
+			printk(KERN_ERR "%s: guard tag error on sector %lu " \
+			       "(rcvd %04x, data %04x)\n", bix->disk_name,
+			       (unsigned long)sector,
+			       be16_to_cpu(sdt->guard_tag), be16_to_cpu(csum));
+			return -EIO;
+		}
+
+		buf += bix->sector_size;
+		sector++;
+	}
+
+	return 0;
+}
+
+static int sd_dif_type3_verify_crc(struct blk_integrity_exchg *bix)
+{
+	return sd_dif_type3_verify(bix, sd_dif_crc_fn);
+}
+
+static int sd_dif_type3_verify_ip(struct blk_integrity_exchg *bix)
+{
+	return sd_dif_type3_verify(bix, sd_dif_ip_fn);
+}
+
+static void sd_dif_type3_set_tag(void *prot, void *tag_buf, unsigned int sectors)
+{
+	struct sd_dif_tuple *sdt = prot;
+	char *tag = tag_buf;
+	unsigned int i, j;
+
+	for (i = 0, j = 0 ; i < sectors ; i++, j += 6, sdt++) {
+		sdt->app_tag = tag[j] << 8 | tag[j+1];
+		sdt->ref_tag = tag[j+2] << 24 | tag[j+3] << 16 |
+			tag[j+4] << 8 | tag[j+5];
+	}
+}
+
+static void sd_dif_type3_get_tag(void *prot, void *tag_buf, unsigned int sectors)
+{
+	struct sd_dif_tuple *sdt = prot;
+	char *tag = tag_buf;
+	unsigned int i, j;
+
+	for (i = 0, j = 0 ; i < sectors ; i++, j += 2, sdt++) {
+		tag[j] = (sdt->app_tag & 0xff00) >> 8;
+		tag[j+1] = sdt->app_tag & 0xff;
+		tag[j+2] = (sdt->ref_tag & 0xff000000) >> 24;
+		tag[j+3] = (sdt->ref_tag & 0xff0000) >> 16;
+		tag[j+4] = (sdt->ref_tag & 0xff00) >> 8;
+		tag[j+5] = sdt->ref_tag & 0xff;
+		BUG_ON(sdt->app_tag == 0xffff || sdt->ref_tag == 0xffffffff);
+	}
+}
+
+static struct blk_integrity dif_type3_integrity_crc = {
+	.name			= "T10-DIF-TYPE3-CRC",
+	.generate_fn		= sd_dif_type3_generate_crc,
+	.verify_fn		= sd_dif_type3_verify_crc,
+	.get_tag_fn		= sd_dif_type3_get_tag,
+	.set_tag_fn		= sd_dif_type3_set_tag,
+	.tuple_size		= sizeof(struct sd_dif_tuple),
+	.tag_size		= 0,
+};
+
+static struct blk_integrity dif_type3_integrity_ip = {
+	.name			= "T10-DIF-TYPE3-IP",
+	.generate_fn		= sd_dif_type3_generate_ip,
+	.verify_fn		= sd_dif_type3_verify_ip,
+	.get_tag_fn		= sd_dif_type3_get_tag,
+	.set_tag_fn		= sd_dif_type3_set_tag,
+	.tuple_size		= sizeof(struct sd_dif_tuple),
+	.tag_size		= 0,
+};
+
+/*
+ * Configure exchange of protection information between OS and HBA.
+ */
+void sd_dif_config_host(struct scsi_disk *sdkp)
+{
+	struct scsi_device *sdp = sdkp->device;
+	struct gendisk *disk = sdkp->disk;
+	u8 type = sdkp->protection_type;
+
+	/* If this HBA doesn't support DIX, resort to normal I/O or DIF */
+	if (scsi_host_dix_capable(sdp->host, type) == 0) {
+
+		if (type == SD_DIF_TYPE0_PROTECTION)
+			return;
+
+		if (scsi_host_dif_capable(sdp->host, type) == 0) {
+			sd_printk(KERN_INFO, sdkp, "Type %d protection " \
+				  "unsupported by HBA. Disabling DIF.\n", type);
+			sdkp->protection_type = 0;
+			return;
+		}
+
+		sd_printk(KERN_INFO, sdkp, "Enabling DIF Type %d protection\n",
+			  type);
+
+		return;
+	}
+
+	/* Enable DMA of protection information */
+	if (scsi_host_get_guard(sdkp->device->host) & SHOST_DIX_GUARD_IP)
+		if (type == SD_DIF_TYPE3_PROTECTION)
+			blk_integrity_register(disk, &dif_type3_integrity_ip);
+		else
+			blk_integrity_register(disk, &dif_type1_integrity_ip);
+	else
+		if (type == SD_DIF_TYPE3_PROTECTION)
+			blk_integrity_register(disk, &dif_type3_integrity_crc);
+		else
+			blk_integrity_register(disk, &dif_type1_integrity_crc);
+
+	sd_printk(KERN_INFO, sdkp,
+		  "Enabling %s integrity protection\n", disk->integrity->name);
+
+	/* Signal to block layer that we support sector tagging */
+	if (type && sdkp->ATO) {
+		if (type == SD_DIF_TYPE3_PROTECTION)
+			disk->integrity->tag_size = sizeof(u16) + sizeof(u32);
+		else
+			disk->integrity->tag_size = sizeof(u16);
+
+		sd_printk(KERN_INFO, sdkp, "DIF application tag size %u\n",
+			  disk->integrity->tag_size);
+	}
+}
+
+/*
+ * DIF DMA operation magic decoder ring.
+ */
+void sd_dif_op(struct scsi_cmnd *scmd, unsigned int dif, unsigned int dix)
+{
+	int csum_convert, prot_op;
+
+	prot_op = 0;
+
+	/* Convert checksum? */
+	if (scsi_host_get_guard(scmd->device->host) != SHOST_DIX_GUARD_CRC)
+		csum_convert = 1;
+	else
+		csum_convert = 0;
+
+	switch (scmd->cmnd[0]) {
+	case READ_10:
+	case READ_12:
+	case READ_16:
+		if (dif && dix)
+			if (csum_convert)
+				prot_op = SCSI_PROT_READ_CONVERT;
+			else
+				prot_op = SCSI_PROT_READ_PASS;
+		else if (dif && !dix)
+			prot_op = SCSI_PROT_READ_STRIP;
+		else if (!dif && dix)
+			prot_op = SCSI_PROT_READ_INSERT;
+
+		break;
+
+	case WRITE_10:
+	case WRITE_12:
+	case WRITE_16:
+		if (dif && dix)
+			if (csum_convert)
+				prot_op = SCSI_PROT_WRITE_CONVERT;
+			else
+				prot_op = SCSI_PROT_WRITE_PASS;
+		else if (dif && !dix)
+			prot_op = SCSI_PROT_WRITE_INSERT;
+		else if (!dif && dix)
+			prot_op = SCSI_PROT_WRITE_STRIP;
+
+		break;
+	}
+
+	scsi_set_prot_op(scmd, prot_op);
+	scsi_set_prot_type(scmd, dif);
+}
+
+/*
+ * The virtual start sector is the one that was originally submitted
+ * by the block layer.	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.
+ *
+ * From a protocol perspective there's a slight difference between
+ * Type 1 and 2.  The latter uses 32-byte CDBs exclusively, and the
+ * reference tag is seeded in the CDB.  This gives us the potential to
+ * avoid virt->phys remapping during write.  However, at read time we
+ * don't know whether the virt sector is the same as when we wrote it
+ * (we could be reading from real disk as opposed to MD/DM device.  So
+ * we always remap Type 2 making it identical to Type 1.
+ *
+ * Type 3 does not have a reference tag so no remapping is required.
+ */
+int sd_dif_prepare(struct request *rq, sector_t hw_sector, unsigned int sector_sz)
+{
+	const int tuple_sz = sizeof(struct sd_dif_tuple);
+	struct bio *bio;
+	struct scsi_disk *sdkp;
+	struct sd_dif_tuple *sdt;
+	unsigned int i, j;
+	u32 phys, virt;
+
+	/* Already remapped? */
+	if (rq->cmd_flags & REQ_INTEGRITY)
+		return 0;
+
+	sdkp = rq->bio->bi_bdev->bd_disk->private_data;
+
+	if (sdkp->protection_type == SD_DIF_TYPE3_PROTECTION)
+		return 0;
+
+	rq->cmd_flags |= REQ_INTEGRITY;
+	phys = hw_sector & 0xffffffff;
+
+	__rq_for_each_bio(bio, rq) {
+		struct bio_vec *iv;
+
+		virt = bio->bi_integrity->bip_sector & 0xffffffff;
+
+		bip_for_each_vec(iv, bio->bi_integrity, i) {
+			sdt = kmap_atomic(iv->bv_page, KM_USER0)
+				+ iv->bv_offset;
+
+			for (j = 0 ; j < iv->bv_len ; j += tuple_sz, sdt++) {
+
+				if (be32_to_cpu(sdt->ref_tag) != virt)
+					goto error;
+
+				sdt->ref_tag = cpu_to_be32(phys);
+				virt++;
+				phys++;
+			}
+
+			kunmap_atomic(sdt, KM_USER0);
+		}
+	}
+
+	return 0;
+
+error:
+	kunmap_atomic(sdt, KM_USER0);
+	sd_printk(KERN_ERR, sdkp, "%s: virt %u, phys %u, ref %u\n",
+		  __func__, virt, phys, be32_to_cpu(sdt->ref_tag));
+
+	return -EIO;
+}
+
+/*
+ * Remap physical sector values in the reference tag to the virtual
+ * values expected by the block layer.
+ */
+void sd_dif_complete(struct scsi_cmnd *scmd, unsigned int good_bytes)
+{
+	const int tuple_sz = sizeof(struct sd_dif_tuple);
+	struct scsi_disk *sdkp;
+	struct bio *bio;
+	struct sd_dif_tuple *sdt;
+	unsigned int i, j, sectors, sector_sz;
+	u32 phys, virt;
+
+	sdkp = scsi_disk(scmd->request->rq_disk);
+
+	if (sdkp->protection_type == SD_DIF_TYPE3_PROTECTION || good_bytes == 0)
+		return;
+
+	sector_sz = scmd->device->sector_size;
+	sectors = good_bytes / sector_sz;
+
+	phys = scmd->request->sector & 0xffffffff;
+	if (sector_sz == 4096)
+		phys >>= 3;
+
+	__rq_for_each_bio(bio, scmd->request) {
+		struct bio_vec *iv;
+
+		virt = bio->bi_integrity->bip_sector & 0xffffffff;
+
+		bip_for_each_vec(iv, bio->bi_integrity, i) {
+			sdt = kmap_atomic(iv->bv_page, KM_USER0)
+				+ iv->bv_offset;
+
+			for (j = 0 ; j < iv->bv_len ; j += tuple_sz, sdt++) {
+
+				if (sectors == 0) {
+					kunmap_atomic(sdt, KM_USER0);
+					return;
+				}
+
+				if (be32_to_cpu(sdt->ref_tag) != phys &&
+				    sdt->app_tag != 0xffff)
+					sdt->ref_tag = 0xffffffff; /* Bad ref */
+				else
+					sdt->ref_tag = cpu_to_be32(virt);
+
+				virt++;
+				phys++;
+				sectors--;
+			}
+
+			kunmap_atomic(sdt, KM_USER0);
+		}
+	}
+}
+

Re: [PATCH 3 of 6] scsi: Support devices with protection information

[James Bottomley]

On Thu, 2008-07-17 at 04:28 -0400, Martin K. Petersen wrote:
> -struct scsi_cmnd *__scsi_get_command(struct Scsi_Host *shost, gfp_t gfp_mask)
> +struct scsi_cmnd *__scsi_get_command(struct Scsi_Host *shost, gfp_t gfp_mask, int protection)
>  {
>  	struct scsi_cmnd *cmd;
>  	unsigned char *buf;
>  
>  	cmd = scsi_pool_alloc_command(shost->cmd_pool, gfp_mask);
>  
> -	if (unlikely(!cmd)) {
> +	if (protection)
> +		cmd->prot_sdb = kmem_cache_zalloc(scsi_sdb_cache, gfp_mask);

Really, no.  if cmd is NULL, this will cause an immediate oops.

> +
> +	if (unlikely(!cmd) || unlikely(protection && cmd->prot_sdb == NULL)) {

And here, too ...

>  		unsigned long flags;
>  		spin_lock_irqsave(&shost->free_list_lock, flags);

In general, you're on the right lines, the problem is with this
protection flag.  To use the poll like this, the protection is going to
have to be always on or always off.  If you try and gate it by a flag,
the command on the free list is going to lose the protection buffer
eventually (because it can get allocated on low memory, but it won't
necessarily be the command returned to the free list, and if the
returned command is allocated without protection, we lose the buffer).

James

Re: [PATCH 3 of 6] scsi: Support devices with protection information

[Martin K. Petersen]

>>>>> "James" == James Bottomley <James.Bottomley@HansenPartnership.com> writes:

James> Really, no.  if cmd is NULL, this will cause an immediate oops.

*blush*


James> In general, you're on the right lines, the problem is with this
James> protection flag.  To use the poll like this, the protection is
James> going to have to be always on or always off.  If you try and
James> gate it by a flag, the command on the free list is going to
James> lose the protection buffer eventually (because it can get
James> allocated on low memory, but it won't necessarily be the
James> command returned to the free list, and if the returned command
James> is allocated without protection, we lose the buffer).

Good point.  The per-command allocation flag was a poorly executed
optimisation afterthought.  The following patch is what I did
originally...

---8<---

scsi: Support devices with protection information

Implement support for DMA of protection information for devices that
are data integrity capable.

 - Add support for mapping an extra scatter-gather list containing
   the protection information.

 - Allocate protection scsi_data_buffer if host is DIX (integrity DMA)
   capable.

 - Accessor function for checking whether a device has protection
   enabled.

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

---

diff --git a/drivers/scsi/scsi.c b/drivers/scsi/scsi.c
--- a/drivers/scsi/scsi.c
+++ b/drivers/scsi/scsi.c
@@ -199,6 +199,38 @@ scsi_pool_free_command(struct scsi_host_
 {
 	kmem_cache_free(pool->sense_slab, cmd->sense_buffer);
 	kmem_cache_free(pool->cmd_slab, cmd);
+
+	if (cmd->prot_sdb)
+		kmem_cache_free(scsi_sdb_cache, cmd->prot_sdb);
+}
+
+/**
+ * scsi_host_alloc_command - internal function to allocate command
+ * @shost:	SCSI host whose pool to allocate from
+ * @gfp_mask:	mask for the allocation
+ *
+ * Returns a fully allocated command with sense buffer and protection
+ * data buffer (where applicable) or NULL on failure
+ */
+static struct scsi_cmnd *
+scsi_host_alloc_command(struct Scsi_Host *shost, gfp_t gfp_mask)
+{
+	struct scsi_cmnd *cmd;
+
+	cmd = scsi_pool_alloc_command(shost->cmd_pool, gfp_mask);
+	if (!cmd)
+		return NULL;
+
+	if (scsi_host_get_prot(shost) >= SHOST_DIX_TYPE0_PROTECTION) {
+		cmd->prot_sdb = kmem_cache_zalloc(scsi_sdb_cache, gfp_mask);
+
+		if (!cmd->prot_sdb) {
+			scsi_pool_free_command(shost->cmd_pool, cmd);
+			return NULL;
+		}
+	}
+
+	return cmd;
 }
 
 /**
@@ -214,7 +246,7 @@ struct scsi_cmnd *__scsi_get_command(str
 	struct scsi_cmnd *cmd;
 	unsigned char *buf;
 
-	cmd = scsi_pool_alloc_command(shost->cmd_pool, gfp_mask);
+	cmd = scsi_host_alloc_command(shost, gfp_mask);
 
 	if (unlikely(!cmd)) {
 		unsigned long flags;
@@ -457,7 +489,7 @@ int scsi_setup_command_freelist(struct S
 	/*
 	 * Get one backup command for this host.
 	 */
-	cmd = scsi_pool_alloc_command(shost->cmd_pool, gfp_mask);
+	cmd = scsi_host_alloc_command(shost, gfp_mask);
 	if (!cmd) {
 		scsi_put_host_cmd_pool(gfp_mask);
 		shost->cmd_pool = NULL;
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -65,7 +65,7 @@ static struct scsi_host_sg_pool scsi_sg_
 };
 #undef SP
 
-static struct kmem_cache *scsi_sdb_cache;
+struct kmem_cache *scsi_sdb_cache;
 
 static void scsi_run_queue(struct request_queue *q);
 
@@ -787,6 +787,9 @@ void scsi_release_buffers(struct scsi_cm
 		kmem_cache_free(scsi_sdb_cache, bidi_sdb);
 		cmd->request->next_rq->special = NULL;
 	}
+
+	if (scsi_prot_sg_count(cmd))
+		scsi_free_sgtable(cmd->prot_sdb);
 }
 EXPORT_SYMBOL(scsi_release_buffers);
 
@@ -1070,6 +1073,26 @@ int scsi_init_io(struct scsi_cmnd *cmd, 
 								    GFP_ATOMIC);
 		if (error)
 			goto err_exit;
+	}
+
+	if (blk_integrity_rq(cmd->request)) {
+		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
+		int ivecs, count;
+
+		BUG_ON(prot_sdb == NULL);
+		ivecs = blk_rq_count_integrity_sg(cmd->request);
+
+		if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
+			error = BLKPREP_DEFER;
+			goto err_exit;
+		}
+
+		count = blk_rq_map_integrity_sg(cmd->request,
+						prot_sdb->table.sgl);
+		BUG_ON(unlikely(count > ivecs));
+
+		cmd->prot_sdb = prot_sdb;
+		cmd->prot_sdb->table.nents = count;
 	}
 
 	return BLKPREP_OK ;
diff --git a/drivers/scsi/scsi_priv.h b/drivers/scsi/scsi_priv.h
--- a/drivers/scsi/scsi_priv.h
+++ b/drivers/scsi/scsi_priv.h
@@ -77,6 +77,7 @@ struct request_queue;
 struct request_queue;
 struct request;
 extern int scsi_prep_fn(struct request_queue *, struct request *);
+extern struct kmem_cache *scsi_sdb_cache;
 
 /* scsi_proc.c */
 #ifdef CONFIG_SCSI_PROC_FS
diff --git a/include/scsi/scsi_cmnd.h b/include/scsi/scsi_cmnd.h
--- a/include/scsi/scsi_cmnd.h
+++ b/include/scsi/scsi_cmnd.h
@@ -90,6 +90,8 @@ struct scsi_cmnd {
 
 	/* These elements define the operation we ultimately want to perform */
 	struct scsi_data_buffer sdb;
+	struct scsi_data_buffer *prot_sdb;
+
 	unsigned underflow;	/* Return error if less than
 				   this amount is transferred */
 
@@ -274,4 +276,22 @@ static inline sector_t scsi_get_lba(stru
 	return scmd->request->sector;
 }
 
+static inline unsigned scsi_prot_sg_count(struct scsi_cmnd *cmd)
+{
+	return cmd->prot_sdb ? cmd->prot_sdb->table.nents : 0;
+}
+
+static inline struct scatterlist *scsi_prot_sglist(struct scsi_cmnd *cmd)
+{
+	return cmd->prot_sdb ? cmd->prot_sdb->table.sgl : NULL;
+}
+
+static inline struct scsi_data_buffer *scsi_prot(struct scsi_cmnd *cmd)
+{
+	return cmd->prot_sdb;
+}
+
+#define scsi_for_each_prot_sg(cmd, sg, nseg, __i)		\
+	for_each_sg(scsi_prot_sglist(cmd), sg, nseg, __i)
+
 #endif /* _SCSI_SCSI_CMND_H */
diff --git a/include/scsi/scsi_device.h b/include/scsi/scsi_device.h
--- a/include/scsi/scsi_device.h
+++ b/include/scsi/scsi_device.h
@@ -416,6 +416,11 @@ static inline int scsi_device_enclosure(
 	return sdev->inquiry[6] & (1<<6);
 }
 
+static inline int scsi_device_protection(struct scsi_device *sdev)
+{
+	return sdev->inquiry[5] & (1<<0);
+}
+
 #define MODULE_ALIAS_SCSI_DEVICE(type) \
 	MODULE_ALIAS("scsi:t-" __stringify(type) "*")
 #define SCSI_DEVICE_MODALIAS_FMT "scsi:t-0x%02x"

Re: [PATCH 3 of 6] scsi: Support devices with protection information

[Michael Hernandez]

On 7/17/08 1:28 AM, "Martin K. Petersen" <martin.petersen@oracle.com> wrote:

> diff --git a/drivers/scsi/scsi.c b/drivers/scsi/scsi.c
> --- a/drivers/scsi/scsi.c
> +++ b/drivers/scsi/scsi.c
> @@ -199,24 +199,60 @@ scsi_pool_free_command(struct scsi_host_
>  {
> kmem_cache_free(pool->sense_slab, cmd->sense_buffer);
> kmem_cache_free(pool->cmd_slab, cmd);
> +
> + if (cmd->prot_sdb)
> +  kmem_cache_free(scsi_sdb_cache, cmd->prot_sdb);
> +}
> +

Shouldn't this check (and potential kmem_cache_free call) get done before
freeing the cmd to the cmd_slab?

Mike

Re: [PATCH 3 of 6] scsi: Support devices with protection information

[Martin K. Petersen]

>>>>> "Mike" == Michael Hernandez <michael.hernandez@qlogic.com> writes:

Mike> Shouldn't this check (and potential kmem_cache_free call) get
Mike> done before freeing the cmd to the cmd_slab?

*sigh*  I think today is Brown Paper Bag Day for me.

But thanks - good spotting!

---8<---

scsi: Support devices with protection information

Implement support for DMA of protection information for devices that
are data integrity capable.

 - Add support for mapping an extra scatter-gather list containing
   the protection information.

 - Allocate protection scsi_data_buffer if host is DIX (integrity DMA)
   capable.

 - Accessor function for checking whether a device has protection
   enabled.

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

---

diff --git a/drivers/scsi/scsi.c b/drivers/scsi/scsi.c
--- a/drivers/scsi/scsi.c
+++ b/drivers/scsi/scsi.c
@@ -197,8 +197,40 @@ scsi_pool_free_command(struct scsi_host_
 scsi_pool_free_command(struct scsi_host_cmd_pool *pool,
 			 struct scsi_cmnd *cmd)
 {
+	if (cmd->prot_sdb)
+		kmem_cache_free(scsi_sdb_cache, cmd->prot_sdb);
+
 	kmem_cache_free(pool->sense_slab, cmd->sense_buffer);
 	kmem_cache_free(pool->cmd_slab, cmd);
+}
+
+/**
+ * scsi_host_alloc_command - internal function to allocate command
+ * @shost:	SCSI host whose pool to allocate from
+ * @gfp_mask:	mask for the allocation
+ *
+ * Returns a fully allocated command with sense buffer and protection
+ * data buffer (where applicable) or NULL on failure
+ */
+static struct scsi_cmnd *
+scsi_host_alloc_command(struct Scsi_Host *shost, gfp_t gfp_mask)
+{
+	struct scsi_cmnd *cmd;
+
+	cmd = scsi_pool_alloc_command(shost->cmd_pool, gfp_mask);
+	if (!cmd)
+		return NULL;
+
+	if (scsi_host_get_prot(shost) >= SHOST_DIX_TYPE0_PROTECTION) {
+		cmd->prot_sdb = kmem_cache_zalloc(scsi_sdb_cache, gfp_mask);
+
+		if (!cmd->prot_sdb) {
+			scsi_pool_free_command(shost->cmd_pool, cmd);
+			return NULL;
+		}
+	}
+
+	return cmd;
 }
 
 /**
@@ -214,7 +246,7 @@ struct scsi_cmnd *__scsi_get_command(str
 	struct scsi_cmnd *cmd;
 	unsigned char *buf;
 
-	cmd = scsi_pool_alloc_command(shost->cmd_pool, gfp_mask);
+	cmd = scsi_host_alloc_command(shost, gfp_mask);
 
 	if (unlikely(!cmd)) {
 		unsigned long flags;
@@ -457,7 +489,7 @@ int scsi_setup_command_freelist(struct S
 	/*
 	 * Get one backup command for this host.
 	 */
-	cmd = scsi_pool_alloc_command(shost->cmd_pool, gfp_mask);
+	cmd = scsi_host_alloc_command(shost, gfp_mask);
 	if (!cmd) {
 		scsi_put_host_cmd_pool(gfp_mask);
 		shost->cmd_pool = NULL;
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -65,7 +65,7 @@ static struct scsi_host_sg_pool scsi_sg_
 };
 #undef SP
 
-static struct kmem_cache *scsi_sdb_cache;
+struct kmem_cache *scsi_sdb_cache;
 
 static void scsi_run_queue(struct request_queue *q);
 
@@ -787,6 +787,9 @@ void scsi_release_buffers(struct scsi_cm
 		kmem_cache_free(scsi_sdb_cache, bidi_sdb);
 		cmd->request->next_rq->special = NULL;
 	}
+
+	if (scsi_prot_sg_count(cmd))
+		scsi_free_sgtable(cmd->prot_sdb);
 }
 EXPORT_SYMBOL(scsi_release_buffers);
 
@@ -1070,6 +1073,26 @@ int scsi_init_io(struct scsi_cmnd *cmd, 
 								    GFP_ATOMIC);
 		if (error)
 			goto err_exit;
+	}
+
+	if (blk_integrity_rq(cmd->request)) {
+		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
+		int ivecs, count;
+
+		BUG_ON(prot_sdb == NULL);
+		ivecs = blk_rq_count_integrity_sg(cmd->request);
+
+		if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
+			error = BLKPREP_DEFER;
+			goto err_exit;
+		}
+
+		count = blk_rq_map_integrity_sg(cmd->request,
+						prot_sdb->table.sgl);
+		BUG_ON(unlikely(count > ivecs));
+
+		cmd->prot_sdb = prot_sdb;
+		cmd->prot_sdb->table.nents = count;
 	}
 
 	return BLKPREP_OK ;
diff --git a/drivers/scsi/scsi_priv.h b/drivers/scsi/scsi_priv.h
--- a/drivers/scsi/scsi_priv.h
+++ b/drivers/scsi/scsi_priv.h
@@ -77,6 +77,7 @@ struct request_queue;
 struct request_queue;
 struct request;
 extern int scsi_prep_fn(struct request_queue *, struct request *);
+extern struct kmem_cache *scsi_sdb_cache;
 
 /* scsi_proc.c */
 #ifdef CONFIG_SCSI_PROC_FS
diff --git a/include/scsi/scsi_cmnd.h b/include/scsi/scsi_cmnd.h
--- a/include/scsi/scsi_cmnd.h
+++ b/include/scsi/scsi_cmnd.h
@@ -90,6 +90,8 @@ struct scsi_cmnd {
 
 	/* These elements define the operation we ultimately want to perform */
 	struct scsi_data_buffer sdb;
+	struct scsi_data_buffer *prot_sdb;
+
 	unsigned underflow;	/* Return error if less than
 				   this amount is transferred */
 
@@ -274,4 +276,22 @@ static inline sector_t scsi_get_lba(stru
 	return scmd->request->sector;
 }
 
+static inline unsigned scsi_prot_sg_count(struct scsi_cmnd *cmd)
+{
+	return cmd->prot_sdb ? cmd->prot_sdb->table.nents : 0;
+}
+
+static inline struct scatterlist *scsi_prot_sglist(struct scsi_cmnd *cmd)
+{
+	return cmd->prot_sdb ? cmd->prot_sdb->table.sgl : NULL;
+}
+
+static inline struct scsi_data_buffer *scsi_prot(struct scsi_cmnd *cmd)
+{
+	return cmd->prot_sdb;
+}
+
+#define scsi_for_each_prot_sg(cmd, sg, nseg, __i)		\
+	for_each_sg(scsi_prot_sglist(cmd), sg, nseg, __i)
+
 #endif /* _SCSI_SCSI_CMND_H */
diff --git a/include/scsi/scsi_device.h b/include/scsi/scsi_device.h
--- a/include/scsi/scsi_device.h
+++ b/include/scsi/scsi_device.h
@@ -416,6 +416,11 @@ static inline int scsi_device_enclosure(
 	return sdev->inquiry[6] & (1<<6);
 }
 
+static inline int scsi_device_protection(struct scsi_device *sdev)
+{
+	return sdev->inquiry[5] & (1<<0);
+}
+
 #define MODULE_ALIAS_SCSI_DEVICE(type) \
 	MODULE_ALIAS("scsi:t-" __stringify(type) "*")
 #define SCSI_DEVICE_MODALIAS_FMT "scsi:t-0x%02x"