[PATCH md 0 of 2] Introduction

[PATCH md 0 of 2] Introduction

[NeilBrown]

Following are two patches for md/linear.c in 2.6.10-rc1-mm5

The first fixes a bug that was recently introduced, that affects linear arrays
where the components aren't all the same size (I didn't test that case :-( ).
I would be good if this was in 2.6.10.

The second is a code cleanup and minor reduction in memory usage.
There is no particular hurry for this patch to go in.

NeilBrown

[PATCH md 2 of 2] Improve 'hash' code in linear.c

[NeilBrown]

The hashtable that linear uses to find the right device stores
two pointers for every entry.
The second is always one of:
   The first plus 1
   NULL
When NULL, it is never access, so any value can be stored.
Thus it could always be "first plus 1", and so we don't need to store
it as it is trivial to calculate.

This patch halves the size of this table, which results in some simpler
code as well.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>

### Diffstat output
 ./drivers/md/linear.c         |   38 ++++++++++++++------------------------
 ./include/linux/raid/linear.h |    7 +------
 2 files changed, 15 insertions(+), 30 deletions(-)

diff ./drivers/md/linear.c~current~ ./drivers/md/linear.c
--- ./drivers/md/linear.c~current~	2004-11-16 15:24:31.000000000 +1100
+++ ./drivers/md/linear.c	2004-11-16 15:24:37.000000000 +1100
@@ -31,7 +31,7 @@
  */
 static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
 {
-	struct linear_hash *hash;
+	dev_info_t *hash;
 	linear_conf_t *conf = mddev_to_conf(mddev);
 	sector_t block = sector >> 1;
 
@@ -39,12 +39,11 @@ static inline dev_info_t *which_dev(mdde
 	 * sector_div(a,b) returns the remainer and sets a to a/b
 	 */
 	(void)sector_div(block, conf->smallest->size);
-	hash = conf->hash_table + block;
+	hash = conf->hash_table[block];
 
-	if ((sector>>1) >= (hash->dev0->size + hash->dev0->offset))
-		return hash->dev1;
-	else
-		return hash->dev0;
+	while ((sector>>1) >= (hash->size + hash->offset))
+		hash++;
+	return hash;
 }
 
 /**
@@ -114,7 +113,7 @@ static int linear_issue_flush(request_qu
 static int linear_run (mddev_t *mddev)
 {
 	linear_conf_t *conf;
-	struct linear_hash *table;
+	dev_info_t **table;
 	mdk_rdev_t *rdev;
 	int i, nb_zone, cnt;
 	sector_t start;
@@ -184,7 +183,7 @@ static int linear_run (mddev_t *mddev)
 		nb_zone = conf->nr_zones = sz + (round ? 1 : 0);
 	}
 			
-	conf->hash_table = kmalloc (sizeof (struct linear_hash) * nb_zone,
+	conf->hash_table = kmalloc (sizeof (dev_info_t*) * nb_zone,
 					GFP_KERNEL);
 	if (!conf->hash_table)
 		goto out;
@@ -198,9 +197,6 @@ static int linear_run (mddev_t *mddev)
 	for (i = 0; i < cnt; i++) {
 		dev_info_t *disk = conf->disks + i;
 
-		if (start > curr_offset)
-			table[-1].dev1 = disk;
-
 		disk->offset = curr_offset;
 		curr_offset += disk->size;
 
@@ -208,10 +204,8 @@ static int linear_run (mddev_t *mddev)
 		 * 'start' is the start of table
 		 */
 		while (start < curr_offset) {
-			table->dev0 = disk;
-			table->dev1 = NULL;
+			*table++ = disk;
 			start += conf->smallest->size;
-			table++;
 		}
 	}
 	if (table-conf->hash_table != nb_zone)
@@ -255,13 +249,6 @@ static int linear_make_request (request_
 
 	tmp_dev = which_dev(mddev, bio->bi_sector);
 	block = bio->bi_sector >> 1;
-  
-	if (unlikely(!tmp_dev)) {
-		printk("linear_make_request: hash->dev1==NULL for block %llu\n",
-			(unsigned long long)block);
-		bio_io_error(bio, bio->bi_size);
-		return 0;
-	}
     
 	if (unlikely(block >= (tmp_dev->size + tmp_dev->offset)
 		     || block < tmp_dev->offset)) {
@@ -306,17 +293,20 @@ static void linear_status (struct seq_fi
 #ifdef MD_DEBUG
 	int j;
 	linear_conf_t *conf = mddev_to_conf(mddev);
+	sector_t s = 0;
   
 	seq_printf(seq, "      ");
 	for (j = 0; j < conf->nr_zones; j++)
 	{
 		char b[BDEVNAME_SIZE];
+		s += conf->smallest_size;
 		seq_printf(seq, "[%s",
-			   bdevname(conf->hash_table[j].dev0->rdev->bdev,b));
+			   bdevname(conf->hash_table[j][0].rdev->bdev,b));
 
-		if (conf->hash_table[j].dev1)
+		while (s > conf->hash_table[j][0].offset + 
+		           conf->hash_table[j][0].size)
 			seq_printf(seq, "/%s] ",
-				   bdevname(conf->hash_table[j].dev1->rdev->bdev,b));
+				   bdevname(conf->hash_table[j][1].rdev->bdev,b));
 		else
 			seq_printf(seq, "] ");
 	}

diff ./include/linux/raid/linear.h~current~ ./include/linux/raid/linear.h
--- ./include/linux/raid/linear.h~current~	2004-11-16 15:24:37.000000000 +1100
+++ ./include/linux/raid/linear.h	2004-11-16 15:24:37.000000000 +1100
@@ -11,14 +11,9 @@ struct dev_info {
 
 typedef struct dev_info dev_info_t;
 
-struct linear_hash
-{
-	dev_info_t *dev0, *dev1;
-};
-
 struct linear_private_data
 {
-	struct linear_hash	*hash_table;
+	dev_info_t		**hash_table;
 	dev_info_t		*smallest;
 	int			nr_zones;
 	dev_info_t		disks[0];

[PATCH md 1 of 2] Fix problem with unsigned variable going "negative" in linear.c

[NeilBrown]

We replace 'size' by 'start'.
'start' means exactly the same as 'curr_offset - size', and
the equivalence of the new code can be tested based on this.
The difference is that 'start' will never be negative and so can
fit in a 'sector_t' while 'size' could be negative.

Also make curr_offset sector_t, as it should have been.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>

### Diffstat output
 ./drivers/md/linear.c |   21 +++++++++++----------
 1 files changed, 11 insertions(+), 10 deletions(-)

diff ./drivers/md/linear.c~current~ ./drivers/md/linear.c
--- ./drivers/md/linear.c~current~	2004-11-16 15:24:31.000000000 +1100
+++ ./drivers/md/linear.c	2004-11-16 15:24:31.000000000 +1100
@@ -117,8 +117,8 @@ static int linear_run (mddev_t *mddev)
 	struct linear_hash *table;
 	mdk_rdev_t *rdev;
 	int i, nb_zone, cnt;
-	sector_t size;
-	unsigned int curr_offset;
+	sector_t start;
+	sector_t curr_offset;
 	struct list_head *tmp;
 
 	conf = kmalloc (sizeof (*conf) + mddev->raid_disks*sizeof(dev_info_t),
@@ -193,23 +193,24 @@ static int linear_run (mddev_t *mddev)
 	 * Here we generate the linear hash table
 	 */
 	table = conf->hash_table;
-	size = 0;
+	start = 0;
 	curr_offset = 0;
 	for (i = 0; i < cnt; i++) {
 		dev_info_t *disk = conf->disks + i;
 
+		if (start > curr_offset)
+			table[-1].dev1 = disk;
+
 		disk->offset = curr_offset;
 		curr_offset += disk->size;
 
-		if (size < 0) {
-			table[-1].dev1 = disk;
-		}
-		size += disk->size;
-
-		while (size>0) {
+		/* 'curr_offset' is the end of this disk
+		 * 'start' is the start of table
+		 */
+		while (start < curr_offset) {
 			table->dev0 = disk;
 			table->dev1 = NULL;
-			size -= conf->smallest->size;
+			start += conf->smallest->size;
 			table++;
 		}
 	}

[PATCH md 0 of 2] Introduction

[NeilBrown]

Two more little patches for md (sorry for the dribs-and-drabs).

Feel free to leave them until post-2.6.11, though the first is really trivial
and allows md/multipath arrays be assembled (without it they just won't work).

- Fix typo for multipat assembly with will
- Mark raid6 non-experimental and tidy up bits of Kconfig

NeilBrown

[PATCH md 0 of 2] Introduction

[NeilBrown]

Two patches for md/raid in 2.6.recent.

1/ Fix data corruption bug in raid10 when used on
   partitions.  raid10 is marked "experimental" but 
   it would be good to get this in 2.6.10
2/ Fix jiffies comparison during resync so it behaves
   correctly when jiffies wrap.  This bug has no
   performance, security, or data integrity impact.

NeilBrown

[PATCH md 0 of 2] Introduction

[NeilBrown]

Two patches for md in 2.6.9-rc1-mm2

The first adds support for notifying user-space of events in md.

The mechanism is very simple.  A reader of /proc/mdstat can select for
"exceptional" events.  When select/poll indicates one of these, the 
reader should re-read /proc/mdstat from the top looking for changes, and then
select again.

If the reader opens for write as well (O_RDWR) (only root can do
this), then it indicates that it is prepared to take remedial action.
This is currently only relevant for multipath.  On last-drive-failure,
if there is a reader of /proc/mdstat that has an open for write, then
multipath will wait for that reader to add a new drive or take other
action before resubmitting the failed requests.

The second patch just fixes a counter in raid5/6 which could get out-by-one, and so
produce confusing messages.

(I think I sent a single item of junk just before this, sorry about that.  Please
ignore it).

NeilBrown