[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [dm-devel] [PATCH] device-mapper: Bad Block relocation target



On Sat, 2007-02-10 at 01:01 +0000, Daniel Drake wrote:

Nice patch Daniel, wondering about using it in 2.6.12. 

You code is very nicely commented, a feature I personally
find more important than the redundant brackets on one line
'if' statements. Over in Germany some of our best kernel programmers
considered the extra braces a cleaner coding style. I continue to
share that point of view. Feels like my eyes can parse the code a bit
faster. One of the very few things I like about Larry Wall's Perl
language it the required brackets on if statements.

If I was to add this to our dm components I was wondering how I'd
test it. No idea on how to get the hardware BBR replacement sectors
to be used up. 

I'll look around for the EVMS tar ball you mentioned.

Thanks for the nicely documented code. I'll re-read it this evening.

-piet


> The BBR target is designed to remap I/O write failures to another safe
> location on disk. Note that most disk drives have BBR built into them,
> this means that our software BBR will be only activated when all hardware
> BBR replacement sectors have been used.
> 
> This patch is included in the EVMS tarball, and Gentoo have been shipping
> it as part of the default kernel for a long time. I know that some users
> are running it, and haven't seen any reported bugs in a long time.
> 
> Please consider this for -mm and possible later inclusion in mainline.
> 
> I haven't been involved in the development of this target, but have brushed
> it up a little from the one included in the EVMS tarball.
> 
> Kevin, I see from mailing list archives that you were somehow involved in
> this project. Can you clarify who the main developers were so that we can
> include this in the history if it gets merged?
> 
> Signed-off-by: Daniel Drake <dsd gentoo org>
> 
> Index: linux-2.6.20-git4/drivers/md/Kconfig
> ===================================================================
> --- linux-2.6.20-git4.orig/drivers/md/Kconfig
> +++ linux-2.6.20-git4/drivers/md/Kconfig
> @@ -262,6 +262,17 @@ config DM_MULTIPATH_EMC
>  	---help---
>  	  Multipath support for EMC CX/AX series hardware.
>  
> +config BLK_DEV_DM_BBR
> +	tristate "Bad Block Relocation Device Target (EXPERIMENTAL)"
> +	depends on BLK_DEV_DM && EXPERIMENTAL
> +	---help---
> +	  Support for devices with software-based bad-block-relocation.
> +
> +	  To compile this as a module, choose M here: the module will be
> +	  called dm-bbr.
> +
> +	  If unsure, say N.
> +
>  endmenu
>  
>  endif
> Index: linux-2.6.20-git4/drivers/md/Makefile
> ===================================================================
> --- linux-2.6.20-git4.orig/drivers/md/Makefile
> +++ linux-2.6.20-git4/drivers/md/Makefile
> @@ -36,6 +36,7 @@ obj-$(CONFIG_DM_MULTIPATH_EMC)	+= dm-emc
>  obj-$(CONFIG_DM_SNAPSHOT)	+= dm-snapshot.o
>  obj-$(CONFIG_DM_MIRROR)		+= dm-mirror.o
>  obj-$(CONFIG_DM_ZERO)		+= dm-zero.o
> +obj-$(CONFIG_BLK_DEV_DM_BBR)	+= dm-bbr.o
>  
>  quiet_cmd_unroll = UNROLL  $@
>        cmd_unroll = $(PERL) $(srctree)/$(src)/unroll.pl $(UNROLL) \
> Index: linux-2.6.20-git4/drivers/md/dm-bbr.c
> ===================================================================
> --- /dev/null
> +++ linux-2.6.20-git4/drivers/md/dm-bbr.c
> @@ -0,0 +1,1004 @@
> +/*
> + *   (C) Copyright IBM Corp. 2002, 2004
> + *
> + *   This program is free software;  you can redistribute it and/or modify
> + *   it under the terms of the GNU General Public License as published by
> + *   the Free Software Foundation; either version 2 of the License, or
> + *   (at your option) any later version.
> + *
> + *   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;  if not, write to the Free Software
> + *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
> + *
> + * linux/drivers/md/dm-bbr.c
> + *
> + * Bad-block-relocation (BBR) target for device-mapper.
> + *
> + * The BBR target is designed to remap I/O write failures to another safe
> + * location on disk. Note that most disk drives have BBR built into them,
> + * this means that our software BBR will be only activated when all hardware
> + * BBR replacement sectors have been used.
> + */
> +
> +#include <linux/module.h>
> +#include <linux/init.h>
> +#include <linux/bio.h>
> +#include <linux/spinlock.h>
> +#include <linux/slab.h>
> +#include <linux/mempool.h>
> +#include <linux/workqueue.h>
> +#include <linux/vmalloc.h>
> +
> +#include "dm.h"
> +#include "dm-bio-list.h"
> +#include "dm-bio-record.h"
> +#include "dm-bbr.h"
> +#include "dm-io.h"
> +
> +#define DM_MSG_PREFIX "bbr"
> +#define SECTOR_SIZE (1 << SECTOR_SHIFT)
> +
> +static struct workqueue_struct *dm_bbr_wq = NULL;
> +static void bbr_remap_handler(void *data);
> +static kmem_cache_t *bbr_remap_cache;
> +static kmem_cache_t *bbr_io_cache;
> +static mempool_t *bbr_io_pool;
> +
> +/**
> + * bbr_binary_tree_destroy
> + *
> + * Destroy the binary tree.
> + **/
> +static void bbr_binary_tree_destroy(struct bbr_runtime_remap *root)
> +{
> +	struct bbr_runtime_remap **link = NULL;
> +	struct bbr_runtime_remap *node = root;
> +
> +	while (node) {
> +		if (node->left) {
> +			link = &(node->left);
> +			node = node->left;
> +			continue;
> +		}
> +		if (node->right) {
> +			link = &(node->right);
> +			node = node->right;
> +			continue;
> +		}
> +
> +		kmem_cache_free(bbr_remap_cache, node);
> +		if (node == root) {
> +			/* If root is deleted, we're done. */
> +			break;
> +		}
> +
> +		/* Back to root. */
> +		node = root;
> +		*link = NULL;
> +	}
> +}
> +
> +static void bbr_free_remap(struct bbr_private *bbr_id)
> +{
> +	spin_lock_irq(&bbr_id->remap_root_lock);
> +	bbr_binary_tree_destroy(bbr_id->remap_root);
> +	bbr_id->remap_root = NULL;
> +	spin_unlock_irq(&bbr_id->remap_root_lock);
> +}
> +
> +static struct bbr_private *bbr_alloc_private(void)
> +{
> +	struct bbr_private *bbr_id;
> +
> +	bbr_id = kmalloc(sizeof(*bbr_id), GFP_KERNEL);
> +	if (bbr_id) {
> +		memset(bbr_id, 0, sizeof(*bbr_id));
> +		INIT_WORK(&bbr_id->remap_work, bbr_remap_handler, bbr_id);
> +		bbr_id->remap_root_lock = SPIN_LOCK_UNLOCKED;
> +		bbr_id->remap_ios_lock = SPIN_LOCK_UNLOCKED;
> +		bbr_id->in_use_replacement_blks = (atomic_t)ATOMIC_INIT(0);
> +	}
> +
> +	return bbr_id;
> +}
> +
> +static void bbr_free_private(struct bbr_private *bbr_id)
> +{
> +	if (bbr_id->bbr_table) {
> +		vfree(bbr_id->bbr_table);
> +	}
> +	bbr_free_remap(bbr_id);
> +	kfree(bbr_id);
> +}
> +
> +static u32 crc_table[256];
> +static u32 crc_table_built = 0;
> +
> +static void build_crc_table(void)
> +{
> +	u32 i, j, crc;
> +
> +	for (i = 0; i <= 255; i++) {
> +		crc = i;
> +		for (j = 8; j > 0; j--) {
> +			if (crc & 1)
> +				crc = (crc >> 1) ^ CRC_POLYNOMIAL;
> +			else
> +				crc >>= 1;
> +		}
> +		crc_table[i] = crc;
> +	}
> +	crc_table_built = 1;
> +}
> +
> +static u32 calculate_crc(u32 crc, void *buffer, u32 buffersize)
> +{
> +	unsigned char *current_byte;
> +	u32 temp1, temp2, i;
> +
> +	current_byte = (unsigned char *) buffer;
> +	/* Make sure the crc table is available */
> +	if (!crc_table_built)
> +		build_crc_table();
> +	/* Process each byte in the buffer. */
> +	for (i = 0; i < buffersize; i++) {
> +		temp1 = (crc >> 8) & 0x00FFFFFF;
> +		temp2 = crc_table[(crc ^ (u32) * current_byte) &
> +				  (u32) 0xff];
> +		current_byte++;
> +		crc = temp1 ^ temp2;
> +	}
> +	return crc;
> +}
> +
> +/**
> + * le_bbr_table_sector_to_cpu
> + *
> + * Convert bbr meta data from on-disk (LE) format
> + * to the native cpu endian format.
> + **/
> +static void le_bbr_table_sector_to_cpu(struct bbr_table *p)
> +{
> +	int i;
> +	p->signature		= le32_to_cpup(&p->signature);
> +	p->crc			= le32_to_cpup(&p->crc);
> +	p->sequence_number	= le32_to_cpup(&p->sequence_number);
> +	p->in_use_cnt		= le32_to_cpup(&p->in_use_cnt);
> +	for (i = 0; i < BBR_ENTRIES_PER_SECT; i++) {
> +		p->entries[i].bad_sect =
> +			le64_to_cpup(&p->entries[i].bad_sect);
> +		p->entries[i].replacement_sect =
> +			le64_to_cpup(&p->entries[i].replacement_sect);
> +	}
> +}
> +
> +/**
> + * cpu_bbr_table_sector_to_le
> + *
> + * Convert bbr meta data from cpu endian format to on-disk (LE) format
> + **/
> +static void cpu_bbr_table_sector_to_le(struct bbr_table *p,
> +				       struct bbr_table *le)
> +{
> +	int i;
> +	le->signature		= cpu_to_le32p(&p->signature);
> +	le->crc			= cpu_to_le32p(&p->crc);
> +	le->sequence_number	= cpu_to_le32p(&p->sequence_number);
> +	le->in_use_cnt		= cpu_to_le32p(&p->in_use_cnt);
> +	for (i = 0; i < BBR_ENTRIES_PER_SECT; i++) {
> +		le->entries[i].bad_sect =
> +			cpu_to_le64p(&p->entries[i].bad_sect);
> +		le->entries[i].replacement_sect =
> +			cpu_to_le64p(&p->entries[i].replacement_sect);
> +	}
> +}
> +
> +/**
> + * validate_bbr_table_sector
> + *
> + * Check the specified BBR table sector for a valid signature and CRC. If it's
> + * valid, endian-convert the table sector.
> + **/
> +static int validate_bbr_table_sector(struct bbr_table *p)
> +{
> +	int rc = 0;
> +	int org_crc, final_crc;
> +
> +	if (le32_to_cpup(&p->signature) != BBR_TABLE_SIGNATURE) {
> +		DMERR("dm-bbr: BBR table signature doesn't match!");
> +		DMERR("dm-bbr: Found 0x%x. Expecting 0x%x",
> +		      le32_to_cpup(&p->signature), BBR_TABLE_SIGNATURE);
> +		rc = -EINVAL;
> +		goto out;
> +	}
> +
> +	if (!p->crc) {
> +		DMERR("dm-bbr: BBR table sector has no CRC!");
> +		rc = -EINVAL;
> +		goto out;
> +	}
> +
> +	org_crc = le32_to_cpup(&p->crc);
> +	p->crc = 0;
> +	final_crc = calculate_crc(INITIAL_CRC, (void *)p, sizeof(*p));
> +	if (final_crc != org_crc) {
> +		DMERR("dm-bbr: CRC failed!");
> +		DMERR("dm-bbr: Found 0x%x. Expecting 0x%x",
> +		      org_crc, final_crc);
> +		rc = -EINVAL;
> +		goto out;
> +	}
> +
> +	p->crc = cpu_to_le32p(&org_crc);
> +	le_bbr_table_sector_to_cpu(p);
> +
> +out:
> +	return rc;
> +}
> +
> +/**
> + * bbr_binary_tree_insert
> + *
> + * Insert a node into the binary tree.
> + **/
> +static void bbr_binary_tree_insert(struct bbr_runtime_remap **root,
> +				   struct bbr_runtime_remap *newnode)
> +{
> +	struct bbr_runtime_remap **node = root;
> +	while (node && *node) {
> +		if (newnode->remap.bad_sect > (*node)->remap.bad_sect) {
> +			node = &((*node)->right);
> +		} else {
> +			node = &((*node)->left);
> +		}
> +	}
> +
> +	newnode->left = newnode->right = NULL;
> +	*node = newnode;
> +}
> +
> +/**
> + * bbr_binary_search
> + *
> + * Search for a node that contains bad_sect == lsn.
> + **/
> +static struct bbr_runtime_remap *bbr_binary_search(
> +	struct bbr_runtime_remap *root,
> +	u64 lsn)
> +{
> +	struct bbr_runtime_remap *node = root;
> +	while (node) {
> +		if (node->remap.bad_sect == lsn) {
> +			break;
> +		}
> +		if (lsn > node->remap.bad_sect) {
> +			node = node->right;
> +		} else {
> +			node = node->left;
> +		}
> +	}
> +	return node;
> +}
> +
> +/**
> + * bbr_insert_remap_entry
> + *
> + * Create a new remap entry and add it to the binary tree for this node.
> + **/
> +static int bbr_insert_remap_entry(struct bbr_private *bbr_id,
> +				  struct bbr_table_entry *new_bbr_entry)
> +{
> +	struct bbr_runtime_remap *newnode;
> +
> +	newnode = kmem_cache_alloc(bbr_remap_cache, GFP_NOIO);
> +	if (!newnode) {
> +		DMERR("dm-bbr: Could not allocate from remap cache!");
> +		return -ENOMEM;
> +	}
> +	newnode->remap.bad_sect  = new_bbr_entry->bad_sect;
> +	newnode->remap.replacement_sect = new_bbr_entry->replacement_sect;
> +	spin_lock_irq(&bbr_id->remap_root_lock);
> +	bbr_binary_tree_insert(&bbr_id->remap_root, newnode);
> +	spin_unlock_irq(&bbr_id->remap_root_lock);
> +	return 0;
> +}
> +
> +/**
> + * bbr_table_to_remap_list
> + *
> + * The on-disk bbr table is sorted by the replacement sector LBA. In order to
> + * improve run time performance, the in memory remap list must be sorted by
> + * the bad sector LBA. This function is called at discovery time to initialize
> + * the remap list. This function assumes that at least one copy of meta data
> + * is valid.
> + **/
> +static u32 bbr_table_to_remap_list(struct bbr_private *bbr_id)
> +{
> +	u32 in_use_blks = 0;
> +	int i, j;
> +	struct bbr_table *p;
> +
> +	for (i = 0, p = bbr_id->bbr_table;
> +	     i < bbr_id->nr_sects_bbr_table;
> +	     i++, p++) {
> +		if (!p->in_use_cnt) {
> +			break;
> +		}
> +		in_use_blks += p->in_use_cnt;
> +		for (j = 0; j < p->in_use_cnt; j++) {
> +			bbr_insert_remap_entry(bbr_id, &p->entries[j]);
> +		}
> +	}
> +	if (in_use_blks) {
> +		char b[32];
> +		DMWARN("dm-bbr: There are %u BBR entries for device %s",
> +		       in_use_blks, format_dev_t(b, bbr_id->dev->bdev->bd_dev));
> +	}
> +
> +	return in_use_blks;
> +}
> +
> +/**
> + * bbr_search_remap_entry
> + *
> + * Search remap entry for the specified sector. If found, return a pointer to
> + * the table entry. Otherwise, return NULL.
> + **/
> +static struct bbr_table_entry *bbr_search_remap_entry(
> +	struct bbr_private *bbr_id,
> +	u64 lsn)
> +{
> +	struct bbr_runtime_remap *p;
> +
> +	spin_lock_irq(&bbr_id->remap_root_lock);
> +	p = bbr_binary_search(bbr_id->remap_root, lsn);
> +	spin_unlock_irq(&bbr_id->remap_root_lock);
> +	if (p) {
> +		return (&p->remap);
> +	} else {
> +		return NULL;
> +	}
> +}
> +
> +/**
> + * bbr_remap
> + *
> + * If *lsn is in the remap table, return TRUE and modify *lsn,
> + * else, return FALSE.
> + **/
> +static inline int bbr_remap(struct bbr_private *bbr_id,
> +			    u64 *lsn)
> +{
> +	struct bbr_table_entry *e;
> +
> +	if (atomic_read(&bbr_id->in_use_replacement_blks)) {
> +		e = bbr_search_remap_entry(bbr_id, *lsn);
> +		if (e) {
> +			*lsn = e->replacement_sect;
> +			return 1;
> +		}
> +	}
> +	return 0;
> +}
> +
> +/**
> + * bbr_remap_probe
> + *
> + * If any of the sectors in the range [lsn, lsn+nr_sects] are in the remap
> + * table return TRUE, Else, return FALSE.
> + **/
> +static inline int bbr_remap_probe(struct bbr_private *bbr_id,
> +				  u64 lsn, u64 nr_sects)
> +{
> +	u64 tmp, cnt;
> +
> +	if (atomic_read(&bbr_id->in_use_replacement_blks)) {
> +		for (cnt = 0, tmp = lsn;
> +		     cnt < nr_sects;
> +		     cnt += bbr_id->blksize_in_sects, tmp = lsn + cnt) {
> +			if (bbr_remap(bbr_id,&tmp)) {
> +				return 1;
> +			}
> +		}
> +	}
> +	return 0;
> +}
> +
> +/**
> + * bbr_setup
> + *
> + * Read the remap tables from disk and set up the initial remap tree.
> + **/
> +static int bbr_setup(struct bbr_private *bbr_id)
> +{
> +	struct bbr_table *table = bbr_id->bbr_table;
> +	struct io_region job;
> +	unsigned long error;
> +	int i, rc = 0;
> +
> +	job.bdev = bbr_id->dev->bdev;
> +	job.count = 1;
> +
> +	/* Read and verify each BBR table sector individually. */
> +	for (i = 0; i < bbr_id->nr_sects_bbr_table; i++, table++) {
> +		job.sector = bbr_id->lba_table1 + i;
> +		rc = dm_io_sync_vm(1, &job, READ, table, &error);
> +		if (rc && bbr_id->lba_table2) {
> +			job.sector = bbr_id->lba_table2 + i;
> +			rc = dm_io_sync_vm(1, &job, READ, table, &error);
> +		}
> +		if (rc) {
> +			goto out;
> +		}
> +
> +		rc = validate_bbr_table_sector(table);
> +		if (rc) {
> +			goto out;
> +		}
> +	}
> +	atomic_set(&bbr_id->in_use_replacement_blks,
> +		   bbr_table_to_remap_list(bbr_id));
> +
> +out:
> +	if (rc) {
> +		DMERR("dm-bbr: error during device setup: %d", rc);
> +	}
> +	return rc;
> +}
> +
> +/**
> + * bbr_io_remap_error
> + * @bbr_id:		Private data for the BBR node.
> + * @rw:			READ or WRITE.
> + * @starting_lsn:	Starting sector of request to remap.
> + * @count:		Number of sectors in the request.
> + * @page:		Page containing the data for the request.
> + * @offset:		Byte-offset of the data within the page.
> + *
> + * For the requested range, try to write each sector individually. For each
> + * sector that fails, find the next available remap location and write the
> + * data to that new location. Then update the table and write both copies
> + * of the table to disk. Finally, update the in-memory mapping and do any
> + * other necessary bookkeeping.
> + **/
> +static int bbr_io_remap_error(struct bbr_private *bbr_id,
> +			      int rw,
> +			      u64 starting_lsn,
> +			      u64 count,
> +			      struct page *page,
> +			      unsigned int offset)
> +{
> +	struct bbr_table *bbr_table;
> +	struct io_region job;
> +	struct page_list pl;
> +	unsigned long table_sector_index;
> +	unsigned long table_sector_offset;
> +	unsigned long index;
> +	unsigned long error;
> +	u64 lsn, new_lsn;
> +	char b[32];
> +	int rc;
> +
> +	job.bdev = bbr_id->dev->bdev;
> +	job.count = 1;
> +	pl.page = page;
> +	pl.next = NULL;
> +
> +	/* For each sector in the request. */
> +	for (lsn = 0; lsn < count; lsn++, offset += SECTOR_SIZE) {
> +		job.sector = starting_lsn + lsn;
> +		rc = dm_io_sync(1, &job, rw, &pl, offset, &error);
> +		while (rc) {
> +			/* Find the next available relocation sector. */
> +			new_lsn = atomic_read(&bbr_id->in_use_replacement_blks);
> +			if (new_lsn >= bbr_id->nr_replacement_blks) {
> +				/* No more replacement sectors available. */
> +				return -EIO;
> +			}
> +			new_lsn += bbr_id->start_replacement_sect;
> +
> +			/* Write the data to its new location. */
> +			DMWARN("dm-bbr: device %s: Trying to remap bad sector "PFU64" to sector "PFU64,
> +			       format_dev_t(b, bbr_id->dev->bdev->bd_dev),
> +			       starting_lsn + lsn, new_lsn);
> +			job.sector = new_lsn;
> +			rc = dm_io_sync(1, &job, rw, &pl, offset, &error);
> +			if (rc) {
> +				/* This replacement sector is bad.
> +				 * Try the next one.
> +				 */
> +				DMERR("dm-bbr: device %s: replacement sector "PFU64" is bad. Skipping.",
> +				      format_dev_t(b, bbr_id->dev->bdev->bd_dev), new_lsn);
> +				atomic_inc(&bbr_id->in_use_replacement_blks);
> +				continue;
> +			}
> +
> +			/* Add this new entry to the on-disk table. */
> +			table_sector_index = new_lsn -
> +					     bbr_id->start_replacement_sect;
> +			table_sector_offset = table_sector_index /
> +					      BBR_ENTRIES_PER_SECT;
> +			index = table_sector_index % BBR_ENTRIES_PER_SECT;
> +
> +			bbr_table = &bbr_id->bbr_table[table_sector_offset];
> +			bbr_table->entries[index].bad_sect = starting_lsn + lsn;
> +			bbr_table->entries[index].replacement_sect = new_lsn;
> +			bbr_table->in_use_cnt++;
> +			bbr_table->sequence_number++;
> +			bbr_table->crc = 0;
> +			bbr_table->crc = calculate_crc(INITIAL_CRC,
> +						       bbr_table,
> +						       sizeof(struct bbr_table));
> +
> +			/* Write the table to disk. */
> +			cpu_bbr_table_sector_to_le(bbr_table, bbr_table);
> +			if (bbr_id->lba_table1) {
> +				job.sector = bbr_id->lba_table1 + table_sector_offset;
> +				rc = dm_io_sync_vm(1, &job, WRITE, bbr_table, &error);
> +			}
> +			if (bbr_id->lba_table2) {
> +				job.sector = bbr_id->lba_table2 + table_sector_offset;
> +				rc |= dm_io_sync_vm(1, &job, WRITE, bbr_table, &error);
> +			}
> +			le_bbr_table_sector_to_cpu(bbr_table);
> +
> +			if (rc) {
> +				/* Error writing one of the tables to disk. */
> +				DMERR("dm-bbr: device %s: error updating BBR tables on disk.",
> +				      format_dev_t(b, bbr_id->dev->bdev->bd_dev));
> +				return rc;
> +			}
> +
> +			/* Insert a new entry in the remapping binary-tree. */
> +			rc = bbr_insert_remap_entry(bbr_id,
> +						    &bbr_table->entries[index]);
> +			if (rc) {
> +				DMERR("dm-bbr: device %s: error adding new entry to remap tree.",
> +				      format_dev_t(b, bbr_id->dev->bdev->bd_dev));
> +				return rc;
> +			}
> +
> +			atomic_inc(&bbr_id->in_use_replacement_blks);
> +		}
> +	}
> +
> +	return 0;
> +}
> +
> +/**
> + * bbr_io_process_request
> + *
> + * For each sector in this request, check if the sector has already
> + * been remapped. If so, process all previous sectors in the request,
> + * followed by the remapped sector. Then reset the starting lsn and
> + * count, and keep going with the rest of the request as if it were
> + * a whole new request. If any of the sync_io's return an error,
> + * call the remapper to relocate the bad sector(s).
> + *
> + * 2.5 Note: When switching over to bio's for the I/O path, we have made
> + * the assumption that the I/O request described by the bio is one
> + * virtually contiguous piece of memory (even though the bio vector
> + * describes it using a series of physical page addresses).
> + **/
> +static int bbr_io_process_request(struct bbr_private *bbr_id,
> +				  struct bio *bio)
> +{
> +	struct io_region job;
> +	u64 starting_lsn = bio->bi_sector;
> +	u64 count, lsn, remapped_lsn;
> +	struct page_list pl;
> +	unsigned int offset;
> +	unsigned long error;
> +	int i, rw = bio_data_dir(bio);
> +	int rc = 0;
> +
> +	job.bdev = bbr_id->dev->bdev;
> +	pl.next = NULL;
> +
> +	/* Each bio can contain multiple vectors, each with a different page.
> +	 * Treat each vector as a separate request.
> +	 */
> +	/* KMC: Is this the right way to walk the bvec list? */
> +	for (i = 0;
> +	     i < bio->bi_vcnt;
> +	     i++, bio->bi_idx++, starting_lsn += count) {
> +
> +		/* Bvec info: number of sectors, page,
> +		 * and byte-offset within page.
> +		 */
> +		count = bio_iovec(bio)->bv_len >> SECTOR_SHIFT;
> +		pl.page = bio_iovec(bio)->bv_page;
> +		offset = bio_iovec(bio)->bv_offset;
> +
> +		/* For each sector in this bvec, check if the sector has
> +		 * already been remapped. If so, process all previous sectors
> +		 * in this request, followed by the remapped sector. Then reset
> +		 * the starting lsn and count and keep going with the rest of
> +		 * the request as if it were a whole new request.
> +		 */
> +		for (lsn = 0; lsn < count; lsn++) {
> +			remapped_lsn = starting_lsn + lsn;
> +			rc = bbr_remap(bbr_id, &remapped_lsn);
> +			if (!rc) {
> +				/* This sector is fine. */
> +				continue;
> +			}
> +
> +			/* Process all sectors in the request up to this one. */
> +			if (lsn > 0) {
> +				job.sector = starting_lsn;
> +				job.count = lsn;
> +				rc = dm_io_sync(1, &job, rw, &pl,
> +						offset, &error);
> +				if (rc) {
> +					/* If this I/O failed, then one of the
> +					 * sectors in this request needs to be
> +					 * relocated.
> +					 */
> +					rc = bbr_io_remap_error(bbr_id, rw,
> +								starting_lsn,
> +								lsn, pl.page,
> +								offset);
> +					if (rc) {
> +						/* KMC: Return? Or continue to next bvec? */
> +						return rc;
> +					}
> +				}
> +				offset += (lsn << SECTOR_SHIFT);
> +			}
> +
> +			/* Process the remapped sector. */
> +			job.sector = remapped_lsn;
> +			job.count = 1;
> +			rc = dm_io_sync(1, &job, rw, &pl, offset, &error);
> +			if (rc) {
> +				/* BUGBUG - Need more processing if this caused
> +				 * an error. If this I/O failed, then the
> +				 * existing remap is now bad, and we need to
> +				 * find a new remap. Can't use
> +				 * bbr_io_remap_error(), because the existing
> +				 * map entry needs to be changed, not added
> +				 * again, and the original table entry also
> +				 * needs to be changed.
> +				 */
> +				return rc;
> +			}
> +
> +			starting_lsn	+= (lsn + 1);
> +			count		-= (lsn + 1);
> +			lsn		= -1;
> +			offset		+= SECTOR_SIZE;
> +		}
> +
> +		/* Check for any remaining sectors after the last split. This
> +		 * could potentially be the whole request, but that should be a
> +		 * rare case because requests should only be processed by the
> +		 * thread if we know an error occurred or they contained one or
> +		 * more remapped sectors.
> +		 */
> +		if (count) {
> +			job.sector = starting_lsn;
> +			job.count = count;
> +			rc = dm_io_sync(1, &job, rw, &pl, offset, &error);
> +			if (rc) {
> +				/* If this I/O failed, then one of the sectors
> +				 * in this request needs to be relocated.
> +				 */
> +				rc = bbr_io_remap_error(bbr_id, rw, starting_lsn,
> +							count, pl.page, offset);
> +				if (rc) {
> +					/* KMC: Return? Or continue to next bvec? */
> +					return rc;
> +				}
> +			}
> +		}
> +	}
> +
> +	return 0;
> +}
> +
> +static void bbr_io_process_requests(struct bbr_private *bbr_id,
> +				    struct bio *bio)
> +{
> +	struct bio *next;
> +	int rc;
> +
> +	while (bio) {
> +		next = bio->bi_next;
> +		bio->bi_next = NULL;
> +
> +		rc = bbr_io_process_request(bbr_id, bio);
> +
> +		bio_endio(bio, bio->bi_size, rc);
> +
> +		bio = next;
> +	}
> +}
> +
> +/**
> + * bbr_remap_handler
> + *
> + * This is the handler for the bbr work-queue.
> + *
> + * I/O requests should only be sent to this handler if we know that:
> + * a) the request contains at least one remapped sector.
> + *   or
> + * b) the request caused an error on the normal I/O path.
> + *
> + * This function uses synchronous I/O, so sending a request to this
> + * thread that doesn't need special processing will cause severe
> + * performance degredation.
> + **/
> +static void bbr_remap_handler(void *data)
> +{
> +	struct bbr_private *bbr_id = data;
> +	struct bio *bio;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&bbr_id->remap_ios_lock, flags);
> +	bio = bio_list_get(&bbr_id->remap_ios);
> +	spin_unlock_irqrestore(&bbr_id->remap_ios_lock, flags);
> +
> +	bbr_io_process_requests(bbr_id, bio);
> +}
> +
> +/**
> + * bbr_endio
> + *
> + * This is the callback for normal write requests. Check for an error
> + * during the I/O, and send to the thread for processing if necessary.
> + **/
> +static int bbr_endio(struct dm_target *ti, struct bio *bio,
> +		     int error, union map_info *map_context)
> +{
> +	struct bbr_private *bbr_id = ti->private;
> +	struct dm_bio_details *bbr_io = map_context->ptr;
> +
> +	if (error && bbr_io) {
> +		unsigned long flags;
> +		char b[32];
> +
> +		dm_bio_restore(bbr_io, bio);
> +		map_context->ptr = NULL;
> +
> +		DMERR("dm-bbr: device %s: I/O failure on sector %lu. "
> +		      "Scheduling for retry.",
> +		      format_dev_t(b, bbr_id->dev->bdev->bd_dev),
> +		      (unsigned long)bio->bi_sector);
> +
> +		spin_lock_irqsave(&bbr_id->remap_ios_lock, flags);
> +		bio_list_add(&bbr_id->remap_ios, bio);
> +		spin_unlock_irqrestore(&bbr_id->remap_ios_lock, flags);
> +
> +		queue_work(dm_bbr_wq, &bbr_id->remap_work);
> +
> +		error = 1;
> +	}
> +
> +	if (bbr_io)
> +		mempool_free(bbr_io, bbr_io_pool);
> +
> +	return error;
> +}
> +
> +/**
> + * Construct a bbr mapping
> + **/
> +static int bbr_ctr(struct dm_target *ti, unsigned int argc, char **argv)
> +{
> +	struct bbr_private *bbr_id;
> +	unsigned long block_size;
> +	char *end;
> +	int rc = -EINVAL;
> +
> +	if (argc != 8) {
> +		ti->error = "dm-bbr requires exactly 8 arguments: "
> +			    "device offset table1_lsn table2_lsn table_size start_replacement nr_replacement_blks block_size";
> +		goto out1;
> +	}
> +
> +	bbr_id = bbr_alloc_private();
> +	if (!bbr_id) {
> +		ti->error = "dm-bbr: Error allocating bbr private data.";
> +		goto out1;
> +	}
> +
> +	bbr_id->offset = simple_strtoull(argv[1], &end, 10);
> +	bbr_id->lba_table1 = simple_strtoull(argv[2], &end, 10);
> +	bbr_id->lba_table2 = simple_strtoull(argv[3], &end, 10);
> +	bbr_id->nr_sects_bbr_table = simple_strtoull(argv[4], &end, 10);
> +	bbr_id->start_replacement_sect = simple_strtoull(argv[5], &end, 10);
> +	bbr_id->nr_replacement_blks = simple_strtoull(argv[6], &end, 10);
> +	block_size = simple_strtoul(argv[7], &end, 10);
> +	bbr_id->blksize_in_sects = (block_size >> SECTOR_SHIFT);
> +
> +	bbr_id->bbr_table = vmalloc(bbr_id->nr_sects_bbr_table << SECTOR_SHIFT);
> +	if (!bbr_id->bbr_table) {
> +		ti->error = "dm-bbr: Error allocating bbr table.";
> +		goto out2;
> +	}
> +
> +	if (dm_get_device(ti, argv[0], 0, ti->len,
> +			  dm_table_get_mode(ti->table), &bbr_id->dev)) {
> +		ti->error = "dm-bbr: Device lookup failed";
> +		goto out2;
> +	}
> +
> +	rc = bbr_setup(bbr_id);
> +	if (rc) {
> +		ti->error = "dm-bbr: Device setup failed";
> +		goto out3;
> +	}
> +
> +	ti->private = bbr_id;
> +	return 0;
> +
> +out3:
> +	dm_put_device(ti, bbr_id->dev);
> +out2:
> +	bbr_free_private(bbr_id);
> +out1:
> +	return rc;
> +}
> +
> +static void bbr_dtr(struct dm_target *ti)
> +{
> +	struct bbr_private *bbr_id = ti->private;
> +
> +	dm_put_device(ti, bbr_id->dev);
> +	bbr_free_private(bbr_id);
> +}
> +
> +static int bbr_map(struct dm_target *ti, struct bio *bio,
> +		   union map_info *map_context)
> +{
> +	struct bbr_private *bbr_id = ti->private;
> +	struct dm_bio_details *bbr_io;
> +	unsigned long flags;
> +	int rc = 1;
> +
> +	bio->bi_sector += bbr_id->offset;
> +
> +	if (atomic_read(&bbr_id->in_use_replacement_blks) == 0 ||
> +	    !bbr_remap_probe(bbr_id, bio->bi_sector, bio_sectors(bio))) {
> +		/* No existing remaps or this request doesn't
> +		 * contain any remapped sectors.
> +		 */
> +		bio->bi_bdev = bbr_id->dev->bdev;
> +
> +		bbr_io = mempool_alloc(bbr_io_pool, GFP_NOIO);
> +		dm_bio_record(bbr_io, bio);
> +		map_context->ptr = bbr_io;
> +	} else {
> +		/* This request has at least one remapped sector.
> +		 * Give it to the work-queue for processing.
> +		 */
> +		map_context->ptr = NULL;
> +		spin_lock_irqsave(&bbr_id->remap_ios_lock, flags);
> +		bio_list_add(&bbr_id->remap_ios, bio);
> +		spin_unlock_irqrestore(&bbr_id->remap_ios_lock, flags);
> +
> +		queue_work(dm_bbr_wq, &bbr_id->remap_work);
> +		rc = 0;
> +	}
> +
> +	return rc;
> +}
> +
> +static int bbr_status(struct dm_target *ti, status_type_t type,
> +		      char *result, unsigned int maxlen)
> +{
> +	struct bbr_private *bbr_id = ti->private;
> +	char b[BDEVNAME_SIZE];
> +
> +	switch (type) {
> +	case STATUSTYPE_INFO:
> +		result[0] = '\0';
> +		break;
> +
> +	case STATUSTYPE_TABLE:
> +		snprintf(result, maxlen, "%s "PFU64" "PFU64" "PFU64" "PFU64" "PFU64" "PFU64" %u",
> +			 format_dev_t(b, bbr_id->dev->bdev->bd_dev),
> +			 bbr_id->offset, bbr_id->lba_table1, bbr_id->lba_table2,
> +			 bbr_id->nr_sects_bbr_table,
> +			 bbr_id->start_replacement_sect,
> +			 bbr_id->nr_replacement_blks,
> +			 bbr_id->blksize_in_sects << SECTOR_SHIFT);
> +		 break;
> +	}
> +	return 0;
> +}
> +
> +static struct target_type bbr_target = {
> +	.name	= "bbr",
> +	.version= {1, 0, 1},
> +	.module	= THIS_MODULE,
> +	.ctr	= bbr_ctr,
> +	.dtr	= bbr_dtr,
> +	.map	= bbr_map,
> +	.end_io	= bbr_endio,
> +	.status	= bbr_status,
> +};
> +
> +int __init dm_bbr_init(void)
> +{
> +	int rc;
> +
> +	rc = dm_register_target(&bbr_target);
> +	if (rc) {
> +		DMERR("dm-bbr: error registering target.");
> +		goto err1;
> +	}
> +
> +	bbr_remap_cache = kmem_cache_create("bbr-remap",
> +					    sizeof(struct bbr_runtime_remap),
> +					    0, SLAB_HWCACHE_ALIGN, NULL, NULL);
> +	if (!bbr_remap_cache) {
> +		DMERR("dm-bbr: error creating remap cache.");
> +		rc = ENOMEM;
> +		goto err2;
> +	}
> +
> +	bbr_io_cache = kmem_cache_create("bbr-io", sizeof(struct dm_bio_details),
> +					 0, SLAB_HWCACHE_ALIGN, NULL, NULL);
> +	if (!bbr_io_cache) {
> +		DMERR("dm-bbr: error creating io cache.");
> +		rc = ENOMEM;
> +		goto err3;
> +	}
> +
> +	bbr_io_pool = mempool_create(256, mempool_alloc_slab,
> +				     mempool_free_slab, bbr_io_cache);
> +	if (!bbr_io_pool) {
> +		DMERR("dm-bbr: error creating io mempool.");
> +		rc = ENOMEM;
> +		goto err4;
> +	}
> +
> +	dm_bbr_wq = create_workqueue("dm-bbr");
> +	if (!dm_bbr_wq) {
> +		DMERR("dm-bbr: error creating work-queue.");
> +		rc = ENOMEM;
> +		goto err5;
> +	}
> +
> +	rc = dm_io_get(1);
> +	if (rc) {
> +		DMERR("dm-bbr: error initializing I/O service.");
> +		goto err6;
> +	}
> +
> +	return 0;
> +
> +err6:
> +	destroy_workqueue(dm_bbr_wq);
> +err5:
> +	mempool_destroy(bbr_io_pool);
> +err4:
> +	kmem_cache_destroy(bbr_io_cache);
> +err3:
> +	kmem_cache_destroy(bbr_remap_cache);
> +err2:
> +	dm_unregister_target(&bbr_target);
> +err1:
> +	return rc;
> +}
> +
> +void __exit dm_bbr_exit(void)
> +{
> +	dm_io_put(1);
> +	destroy_workqueue(dm_bbr_wq);
> +	mempool_destroy(bbr_io_pool);
> +	kmem_cache_destroy(bbr_io_cache);
> +	kmem_cache_destroy(bbr_remap_cache);
> +	dm_unregister_target(&bbr_target);
> +}
> +
> +module_init(dm_bbr_init);
> +module_exit(dm_bbr_exit);
> +MODULE_LICENSE("GPL");
> Index: linux-2.6.20-git4/drivers/md/dm-bbr.h
> ===================================================================
> --- /dev/null
> +++ linux-2.6.20-git4/drivers/md/dm-bbr.h
> @@ -0,0 +1,125 @@
> +/*
> + *   (C) Copyright IBM Corp. 2002, 2004
> + *
> + *   This program is free software;  you can redistribute it and/or modify
> + *   it under the terms of the GNU General Public License as published by
> + *   the Free Software Foundation; either version 2 of the License, or
> + *   (at your option) any later version.
> + *
> + *   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;  if not, write to the Free Software
> + *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
> + *
> + * linux/drivers/md/dm-bbr.h
> + *
> + * Bad-block-relocation (BBR) target for device-mapper.
> + *
> + * The BBR target is designed to remap I/O write failures to another safe
> + * location on disk. Note that most disk drives have BBR built into them,
> + * this means that our software BBR will be only activated when all hardware
> + * BBR replacement sectors have been used.
> + */
> +
> +#define BBR_TABLE_SIGNATURE		0x42627254 /* BbrT */
> +#define BBR_ENTRIES_PER_SECT		31
> +#define INITIAL_CRC			0xFFFFFFFF
> +#define CRC_POLYNOMIAL			0xEDB88320L
> +
> +/**
> + * Macros to cleanly print 64-bit numbers on both 32-bit and 64-bit machines.
> + * Use these in place of %Ld, %Lu, and %Lx.
> + **/
> +#if BITS_PER_LONG > 32
> +#define PFU64 "%lu"
> +#else
> +#define PFU64 "%Lu"
> +#endif
> +
> +/**
> + * struct bbr_table_entry
> + * @bad_sect:		LBA of bad location.
> + * @replacement_sect:	LBA of new location.
> + *
> + * Structure to describe one BBR remap.
> + **/
> +struct bbr_table_entry {
> +	u64 bad_sect;
> +	u64 replacement_sect;
> +};
> +
> +/**
> + * struct bbr_table
> + * @signature:		Signature on each BBR table sector.
> + * @crc:		CRC for this table sector.
> + * @sequence_number:	Used to resolve conflicts when primary and secondary
> + *			tables do not match.
> + * @in_use_cnt:		Number of in-use table entries.
> + * @entries:		Actual table of remaps.
> + *
> + * Structure to describe each sector of the metadata table. Each sector in this
> + * table can describe 31 remapped sectors.
> + **/
> +struct bbr_table {
> +	u32			signature;
> +	u32			crc;
> +	u32			sequence_number;
> +	u32			in_use_cnt;
> +	struct bbr_table_entry	entries[BBR_ENTRIES_PER_SECT];
> +};
> +
> +/**
> + * struct bbr_runtime_remap
> + *
> + * Node in the binary tree used to keep track of remaps.
> + **/
> +struct bbr_runtime_remap {
> +	struct bbr_table_entry		remap;
> +	struct bbr_runtime_remap	*left;
> +	struct bbr_runtime_remap	*right;
> +};
> +
> +/**
> + * struct bbr_private
> + * @dev:			Info about underlying device.
> + * @bbr_table:			Copy of metadata table.
> + * @remap_root:			Binary tree containing all remaps.
> + * @remap_root_lock:		Lock for the binary tree.
> + * @remap_work:			For adding work items to the work-queue.
> + * @remap_ios:			List of I/Os for the work-queue to handle.
> + * @remap_ios_lock:		Lock for the remap_ios list.
> + * @offset:			LBA of data area.
> + * @lba_table1:			LBA of primary BBR table.
> + * @lba_table2:			LBA of secondary BBR table.
> + * @nr_sects_bbr_table:		Size of each BBR table.
> + * @nr_replacement_blks:	Number of replacement blocks.
> + * @start_replacement_sect:	LBA of start of replacement blocks.
> + * @blksize_in_sects:		Size of each block.
> + * @in_use_replacement_blks:	Current number of remapped blocks.
> + *
> + * Private data for each BBR target.
> + **/
> +struct bbr_private {
> +	struct dm_dev			*dev;
> +	struct bbr_table		*bbr_table;
> +	struct bbr_runtime_remap	*remap_root;
> +	spinlock_t			remap_root_lock;
> +
> +	struct work_struct		remap_work;
> +	struct bio_list			remap_ios;
> +	spinlock_t			remap_ios_lock;
> +
> +	u64				offset;
> +	u64				lba_table1;
> +	u64				lba_table2;
> +	u64				nr_sects_bbr_table;
> +	u64				start_replacement_sect;
> +	u64				nr_replacement_blks;
> +	u32				blksize_in_sects;
> +	atomic_t			in_use_replacement_blks;
> +};
> +
> 
> --
> dm-devel mailing list
> dm-devel redhat com
> https://www.redhat.com/mailman/listinfo/dm-devel
-- 
Piet Delaney                                    Phone: (408) 200-5256
Blue Lane Technologies                          Fax:   (408) 200-5299
10450 Bubb Rd.
Cupertino, Ca. 95014                            Email: piet bluelane com

Attachment: signature.asc
Description: This is a digitally signed message part


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]