Skip to contentSkip to conversation

Emulate BLKRRPART on device-mapper

PATCH In review Nikanth Karthikesan 4 participants 6 messages Jul 8, 2009 – Jul 22, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Emulate BLKRRPART on device-mapper

From: Hannes Reinecke <[email protected]>
Subject: Emulate BLKRRPART on device-mapper

Partitions on device-mapper devices are managed by kpartx (if at all). So if we were just to send out a 'change' event if someone called BLKRRPART on these devices, kpartx will be triggered via udev and can manage the partitions accordingly.

---

drivers/md/dm.c +14 -8
Index: linux-2.6-dm/drivers/md/dm.c
===================================================================
@@ -394,19 +394,25 @@ static int dm_blk_ioctl(struct block_dev
 	if (!map || !dm_table_get_size(map))
 		goto out;
 
-	/* We only support devices that have a single target */
-	if (dm_table_get_num_targets(map) != 1)
-		goto out;
-
-	tgt = dm_table_get_target(map, 0);
-
 	if (dm_suspended(md)) {
 		r = -EAGAIN;
 		goto out;
 	}
 
-	if (tgt->type->ioctl)
-		r = tgt->type->ioctl(tgt, cmd, arg);
+	if (cmd == BLKRRPART) {
+		/* Emulate Re-read partitions table */
+		kobject_uevent(&disk_to_dev(md->disk)->kobj, KOBJ_CHANGE);
+		r = 0;
+	} else {
+		/* We only support devices that have a single target */
+		if (dm_table_get_num_targets(map) != 1)
+			goto out;
+
+		tgt = dm_table_get_target(map, 0);
+
+		if (tgt->type->ioctl)
+			r = tgt->type->ioctl(tgt, cmd, arg);
+	}
 
 out:
 	dm_table_put(map);
#1 lore raw
Quoted 10 lines · Nikanth Karthikesan

From: Hannes Reinecke <[email protected]>
Subject: Emulate BLKRRPART on device-mapper

Partitions on device-mapper devices are managed by kpartx (if at all). So if we were just to send out a 'change' event if someone called BLKRRPART on these devices, kpartx will be triggered via udev and can manage the partitions accordingly.

Please could I have a 'Tested-by' for this one?

I agree with the idea, but I need some evidence that this specific patch works correctly. (The block layer is supposed to handle that class of ioctls itself and I didn't think dm saw them.)

Alasdair

#2 lore raw

RFC PATCH 1/1 Emulate BLKRRPART on device-mapper

Quoted 5 lines · Alasdair G Kergon
Quoted 3 lines · Nikanth Karthikesan

From: Hannes Reinecke <[email protected]>
Subject: Emulate BLKRRPART on device-mapper

Quoted 8 lines
Quoted 6 lines

Partitions on device-mapper devices are managed by kpartx (if at all). So if we were just to send out a 'change' event if someone called BLKRRPART on these devices, kpartx will be triggered via udev and can manage the partitions accordingly.

Please could I have a 'Tested-by' for this one?

I am afraid that this patch cannot work, BLRRPART never reach this code. I tried another idea - or is there better way how to achieve that?

Milan
---

From: Milan Broz <[email protected]>

Add genhd flag requesting notification of partition changes only.

This patch provides notification mechanism which allows handle partition code in userspace.

If the BLKRRPART ioctl arrives and GENHD_FL_PARTITION_CHANGE_NOTIFY is set, just send uevent and ignore in-kernel partitioning code.

This is useful e.g. for device-mapper devices, which can use kpartx or similar tool in udev rules.

---
 block/ioctl.c         |    3 ++-
 drivers/md/dm.c       |    1 +
 fs/partitions/check.c |    6 ++++++
 include/linux/genhd.h |    8 ++++++++
 4 files changed, 17 insertions(+), 1 deletions(-)
block/ioctl.c +2 -1
index 500e4c7..bce793f 100644
@@ -101,7 +101,8 @@ static int blkdev_reread_part(struct block_device *bdev)
 	struct gendisk *disk = bdev->bd_disk;
 	int res;
 
-	if (!disk_partitionable(disk) || bdev != bdev->bd_contains)
+	if (!disk_userspace_partitions(disk) &&
+	    (!disk_partitionable(disk) || bdev != bdev->bd_contains))
 		return -EINVAL;
 	if (!capable(CAP_SYS_ADMIN))
 		return -EACCES;
drivers/md/dm.c +1 -0
index 9acd54a..1186ce1 100644
@@ -1791,6 +1791,7 @@ static struct mapped_device *alloc_dev(int minor)
 	md->disk->queue = md->queue;
 	md->disk->private_data = md;
 	sprintf(md->disk->disk_name, "dm-%d", minor);
+	md->disk->flags |= GENHD_FL_PARTITION_CHANGE_NOTIFY;
 	add_disk(md->disk);
 	format_dev_t(md->name, MKDEV(_major, minor));
 
fs/partitions/check.c +6 -0
index ea4e6cb..bb42c44 100644
@@ -521,6 +521,12 @@ int rescan_partitions(struct gendisk *disk, struct block_device *bdev)
 	struct parsed_partitions *state;
 	int p, highest, res;
 
+	/* partitions handled in userspace, just send change event */
+	if (disk_userspace_partitions(disk)) {
+		kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
+		return 0;
+	}
+
 	if (bdev->bd_part_count)
 		return -EBUSY;
 	res = invalidate_partition(disk, 0);
include/linux/genhd.h +8 -0
index 45fc320..a241bd6 100644
@@ -116,6 +116,9 @@ struct hd_struct {
 #define GENHD_FL_EXT_DEVT			64 /* allow extended devt */
 #define GENHD_FL_NATIVE_CAPACITY		128
 
+/* notify udev instead of use in-kernel partitioning */
+#define GENHD_FL_PARTITION_CHANGE_NOTIFY	256
+
 #define BLK_SCSI_MAX_CMDS	(256)
 #define BLK_SCSI_CMD_PER_LONG	(BLK_SCSI_MAX_CMDS / (sizeof(long) * 8))
 
@@ -180,6 +183,11 @@ static inline struct gendisk *part_to_disk(struct hd_struct *part)
 	return NULL;
 }
 
+static inline bool disk_userspace_partitions(struct gendisk *disk)
+{
+	return (disk->flags & GENHD_FL_PARTITION_CHANGE_NOTIFY) ? 1 : 0;
+}
+
 static inline int disk_max_parts(struct gendisk *disk)
 {
 	if (disk->flags & GENHD_FL_EXT_DEVT)
#3 lore raw
Quoted 12 lines · Milan Broz
Quoted 11 lines · Alasdair G Kergon
Quoted 9 lines · Nikanth Karthikesan

From: Hannes Reinecke <[email protected]>
Subject: Emulate BLKRRPART on device-mapper

Partitions on device-mapper devices are managed by kpartx (if at all). So if we were just to send out a 'change' event if someone called BLKRRPART on these devices, kpartx will be triggered via udev and can manage the partitions accordingly.

Quoted 5 lines
Quoted 2 lines

Please could I have a 'Tested-by' for this one?

I am afraid that this patch cannot work, BLRRPART never reach this code. I tried another idea - or is there better way how to achieve that?

Hey, that is cool. And what's more, that's exactly what I need for another pet-project of mine, switching off the in-kernel partitioning code altogether.

But there are some comments, see below.

Quoted 55 lines

Milan
---

From: Milan Broz <[email protected]>

Add genhd flag requesting notification of partition changes only.

This patch provides notification mechanism which allows handle partition code in userspace.

If the BLKRRPART ioctl arrives and GENHD_FL_PARTITION_CHANGE_NOTIFY is set, just send uevent and ignore in-kernel partitioning code.

This is useful e.g. for device-mapper devices, which can use kpartx or similar tool in udev rules.

---
 block/ioctl.c         |    3 ++-
 drivers/md/dm.c       |    1 +
 fs/partitions/check.c |    6 ++++++
 include/linux/genhd.h |    8 ++++++++
 4 files changed, 17 insertions(+), 1 deletions(-)
block/ioctl.c +2 -1
index 500e4c7..bce793f 100644
@@ -101,7 +101,8 @@ static int blkdev_reread_part(struct block_device *bdev)
 	struct gendisk *disk = bdev->bd_disk;
 	int res;
 
-	if (!disk_partitionable(disk) || bdev != bdev->bd_contains)
+	if (!disk_userspace_partitions(disk) &&
+	    (!disk_partitionable(disk) || bdev != bdev->bd_contains))
 		return -EINVAL;
 	if (!capable(CAP_SYS_ADMIN))
 		return -EACCES;
drivers/md/dm.c +1 -0
index 9acd54a..1186ce1 100644
@@ -1791,6 +1791,7 @@ static struct mapped_device *alloc_dev(int minor)
 	md->disk->queue = md->queue;
 	md->disk->private_data = md;
 	sprintf(md->disk->disk_name, "dm-%d", minor);
+	md->disk->flags |= GENHD_FL_PARTITION_CHANGE_NOTIFY;
 	add_disk(md->disk);
 	format_dev_t(md->name, MKDEV(_major, minor));
 
fs/partitions/check.c +0 -0
index ea4e6cb..bb42c44 100644

fs/partitions/check.c int rescan_partitions(struct gendisk *disk, struct block_device *bdev) · line 529

Quoted 9 lines
fs/partitions/check.c +6 -0
@@ -521,6 +521,12 @@ int rescan_partitions(struct gendisk *disk, struct block_device *bdev)
 	struct parsed_partitions *state;
 	int p, highest, res;
 
+	/* partitions handled in userspace, just send change event */
+	if (disk_userspace_partitions(disk)) {
+		kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
+		return 0;
+	}
+

Wrong. If you do it here, you'll never be able to catch any size changes of the disk. You'll have to move it to after the 'bdev->bd_invalidated = 0' line.

Quoted 8 lines
fs/partitions/check.c +0 -0
 	if (bdev->bd_part_count)
 		return -EBUSY;
 	res = invalidate_partition(disk, 0);
include/linux/genhd.h +0 -0
index 45fc320..a241bd6 100644

include/linux/genhd.h struct hd_struct { · line 121

Quoted 6 lines
include/linux/genhd.h +3 -0
@@ -116,6 +116,9 @@ struct hd_struct {
 #define GENHD_FL_EXT_DEVT			64 /* allow extended devt */
 #define GENHD_FL_NATIVE_CAPACITY		128
 
+/* notify udev instead of use in-kernel partitioning */
+#define GENHD_FL_PARTITION_CHANGE_NOTIFY	256
+

I would suggest renaming it to GENHD_FL_USERSPACE_PARTITIONS, as this is more in line with the function of the flag.
Plus I have a patch making use of it :-)

I'll send an adapted patchset.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke		      zSeries & Storage
[email protected]			      +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Markus Rex, HRB 16746 (AG Nürnberg)
#4 lore raw
Quoted 9 lines · Hannes Reinecke

fs/partitions/check.c int rescan_partitions(struct gendisk *disk, struct block_device *bdev) · line 529

Quoted 6 lines
fs/partitions/check.c +6 -0
+	/* partitions handled in userspace, just send change event */
+	if (disk_userspace_partitions(disk)) {
+		kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
+		return 0;
+	}
+

Wrong. If you do it here, you'll never be able to catch any size changes of the disk. You'll have to move it to after the 'bdev->bd_invalidated = 0' line.

Yes, for non-DM devices which want to use that flag.

For device-mapper device, size can change only by mapping table change which always generate change uevent.

Quoted 2 lines

I would suggest renaming it to GENHD_FL_USERSPACE_PARTITIONS, as this is more in line with the function of the flag.

no problem here:-)

Milan

#5 lore raw
Quoted 11 lines · Milan Broz
Quoted 9 lines · Hannes Reinecke

fs/partitions/check.c int rescan_partitions(struct gendisk *disk, struct block_device *bdev) · line 529

Quoted 6 lines
fs/partitions/check.c +6 -0
+	/* partitions handled in userspace, just send change event */
+	if (disk_userspace_partitions(disk)) {
+		kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
+		return 0;
+	}
+

Wrong. If you do it here, you'll never be able to catch any size changes of the disk. You'll have to move it to after the 'bdev->bd_invalidated = 0' line.

Quoted 5 lines

Yes, for non-DM devices which want to use that flag.

For device-mapper device, size can change only by mapping table change which always generate change uevent.

Okay. But the whole point of my objections is to make the flag useable for the general populace, so that my second patch can take advantage of it.

Quoted 4 lines
Quoted 2 lines

I would suggest renaming it to GENHD_FL_USERSPACE_PARTITIONS, as this is more in line with the function of the flag.

no problem here:-)

Hmm. I've send an updated patch, but it seems to be stuck in the mailqueue somewhere.
The second patch went through, though ... curious.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke		      zSeries & Storage
[email protected]			      +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Markus Rex, HRB 16746 (AG Nürnberg)