Skip to contentSkip to conversation

Add support for block disk resize notification

PATCH RESEND Superseded Balbir Singh 7 participants 38 messages Jan 2, 2020 – Jan 21, 2020

Subsystems (7)
SCSI Block layer virtio core NVMe +3 more subsystems

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

cover Add support for block disk resize notification

Allow block/genhd to notify user space about disk size changes using a new helper disk_set_capacity(), which is a wrapper on top of set_capacity(). disk_set_capacity() will only notify if the current capacity or the target capacity is not zero.

Background:

As a part of a patch to allow sending the RESIZE event on disk capacity change, Christoph ([email protected]) requested that the patch be made generic and the hacks for virtio block and xen block devices be removed and merged via a generic helper.

This series consists of 5 changes. The first one adds the basic support for changing the size and notifying. The follow up patches are per block subsystem changes. Other block drivers can add their changes as necessary on top of this series.

Testing:
1. I did some basic testing with an NVME device, by resizing it in the backend and ensured that udevd received the event.

NOTE: After these changes, the notification might happen before revalidate disk, where as it occured later before.

Balbir Singh (5):
block/genhd: Notify udev about capacity change drivers/block/virtio_blk.c: Convert to use disk_set_capacity drivers/block/xen-blkfront.c: Convert to use disk_set_capacity drivers/nvme/host/core.c: Convert to use disk_set_capacity drivers/scsi/sd.c: Convert to use disk_set_capacity

 block/genhd.c                | 19 +++++++++++++++++++
 drivers/block/virtio_blk.c   |  4 +---
 drivers/block/xen-blkfront.c |  5 +----
 drivers/nvme/host/core.c     |  2 +-
 drivers/scsi/sd.c            |  2 +-
 include/linux/genhd.h        |  1 +
 6 files changed, 24 insertions(+), 9 deletions(-)
Signature · 1 line
2.16.5
#1 lore raw

PATCH RESEND 1/5 block/genhd: Notify udev about capacity change

Allow block/genhd to notify user space (via udev) about disk size changes using a new helper disk_set_capacity(), which is a wrapper on top of set_capacity(). disk_set_capacity() will only notify via udev if the current capacity or the target capacity is not zero.

---
 block/genhd.c         | 19 +++++++++++++++++++
 include/linux/genhd.h |  1 +
 2 files changed, 20 insertions(+)
block/genhd.c +19 -0
index ff6268970ddc..94faec98607b 100644
@@ -46,6 +46,25 @@ static void disk_add_events(struct gendisk *disk);
 static void disk_del_events(struct gendisk *disk);
 static void disk_release_events(struct gendisk *disk);
 
+/*
+ * Set disk capacity and notify if the size is not currently
+ * zero and will not be set to zero
+ */
+void disk_set_capacity(struct gendisk *disk, sector_t size)
+{
+	sector_t capacity = get_capacity(disk);
+
+	set_capacity(disk, size);
+	if (capacity != 0 && size != 0) {
+		char *envp[] = { "RESIZE=1", NULL };
+
+		kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
+	}
+}
+
+EXPORT_SYMBOL_GPL(disk_set_capacity);
+
+
 void part_inc_in_flight(struct request_queue *q, struct hd_struct *part, int rw)
 {
 	if (queue_is_mq(q))
include/linux/genhd.h +1 -0
index a927829bb73a..f1a5ddcc781d 100644
@@ -449,6 +449,7 @@ static inline int get_disk_ro(struct gendisk *disk)
 extern void disk_block_events(struct gendisk *disk);
 extern void disk_unblock_events(struct gendisk *disk);
 extern void disk_flush_events(struct gendisk *disk, unsigned int mask);
+extern void disk_set_capacity(struct gendisk *disk, sector_t size);
 extern unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask);
 
 /* drivers/char/random.c */
Signature · 7 lines
2.16.5


_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#2 lore raw

Re: [resend v1 1/5] block/genhd: Notify udev about capacity change

Quoted 9 lines · Balbir Singh

Allow block/genhd to notify user space (via udev) about disk size changes using a new helper disk_set_capacity(), which is a wrapper on top of set_capacity(). disk_set_capacity() will only notify via udev if the current capacity or the target capacity is not zero.

---

Since disk_set_capacity(() is on the same line as set_capacity() we should follow the same convention, which is :-

1. Avoid exporting symbol.
2. Mark new function in-line.

Unless there is a very specific reason for breaking the pattern.

Why not this (totally untested but easy convey above comment) on the top of this patch ?

# git diff
block/genhd.c +0 -13
index 94faec98607b..ff6268970ddc 100644
@@ -46,25 +46,6 @@ static void disk_add_events(struct gendisk *disk);
  static void disk_del_events(struct gendisk *disk);
  static void disk_release_events(struct gendisk *disk);
 
-/*
- * Set disk capacity and notify if the size is not currently
- * zero and will not be set to zero
- */
-void disk_set_capacity(struct gendisk *disk, sector_t size)
-{
-       sector_t capacity = get_capacity(disk);
-
-       set_capacity(disk, size);
-       if (capacity != 0 && size != 0) {
-               char *envp[] = { "RESIZE=1", NULL };
-
-               kobject_uevent_env(&disk_to_dev(disk)->kobj, 

KOBJ_CHANGE, envp);
- }
-}
-
-EXPORT_SYMBOL_GPL(disk_set_capacity);
-
-
void part_inc_in_flight(struct request_queue *q, struct hd_struct
*part, int rw)

  {
         if (queue_is_mq(q))
include/linux/genhd.h +0 -0
index d5e87d7cc357..5e595a28f893 100644
@@ -469,6 +469,22 @@ static inline void set_capacity(struct gendisk 

*disk, sector_t size)

         disk->part0.nr_sects = size;

}

+/*
+ * Set disk capacity and notify if the size is not currently
+ * zero and will not be set to zero
+ */
+static inline void disk_set_capacity(struct gendisk *disk, sector_t size) +{
+ sector_t capacity = get_capacity(disk);
+
+ set_capacity(disk, size);
+ if (capacity != 0 && size != 0) {
+ char *envp[] = { "RESIZE=1", NULL };
+
+ kobject_uevent_env(&disk_to_dev(disk)->kobj,
KOBJ_CHANGE, envp);
+ }
+}
+
#ifdef CONFIG_SOLARIS_X86_PARTITION

#3 lore raw

Re: [resend v1 1/5] block/genhd: Notify udev about capacity change

Quoted 11 lines · Chaitanya Kulkarni
Quoted 9 lines · Balbir Singh

Allow block/genhd to notify user space (via udev) about disk size changes using a new helper disk_set_capacity(), which is a wrapper on top of set_capacity(). disk_set_capacity() will only notify via udev if the current capacity or the target capacity is not zero.

---

Quoted 8 lines

Since disk_set_capacity(() is on the same line as set_capacity() we should follow the same convention, which is :-

1. Avoid exporting symbol.
2. Mark new function in-line.

Unless there is a very specific reason for breaking the pattern.

I don't see the benefit of marking it as inline. I expect this code to be potentially expanded to provide callbacks into file system code for expansion (something you brought up), marking it as inline might be a limitation.

I'd rather not clutter the headers with this code, but I am open to what the maintainers think as well.

Balbir Singh.

#4 lore raw

Re: [resend v1 1/5] block/genhd: Notify udev about capacity change

Quoted 12 lines · Balbir
Quoted 8 lines
Quoted 8 lines

Since disk_set_capacity(() is on the same line as set_capacity() we should follow the same convention, which is :-

1. Avoid exporting symbol.
2. Mark new function in-line.

Unless there is a very specific reason for breaking the pattern.

I don't see the benefit of marking it as inline. I expect this code to be potentially expanded to provide callbacks into file system code for expansion (something you brought up), marking it as inline might be a limitation.

That can be done as a prep patch when framework is ready.

Quoted 3 lines

I'd rather not clutter the headers with this code, but I am open to what the maintainers think as well.

Okay.

Quoted 3 lines

Balbir Singh.

#5 lore raw

Re: [resend v1 1/5] block/genhd: Notify udev about capacity change

In reply to Chaitanya Kulkarni
Quoted 7 lines · Chaitanya Kulkarni

Since disk_set_capacity(() is on the same line as set_capacity() we should follow the same convention, which is :-

1. Avoid exporting symbol.
2. Mark new function in-line.

Unless there is a very specific reason for breaking the pattern.

Why would we mark it as inline? It isn't by any means in the fast path, and there are no easy opportunities for constant propagation, so the only thing that would do is increase the code size.

#6 lore raw

Re: [resend v1 1/5] block/genhd: Notify udev about capacity change

In reply to Balbir Singh

Hi Balbir,

Quoted 4 lines

Allow block/genhd to notify user space (via udev) about disk size changes using a new helper disk_set_capacity(), which is a wrapper on top of set_capacity(). disk_set_capacity() will only notify via udev if the current capacity or the target capacity is not zero.

I know set_capacity() is called all over the place making it a bit of a pain to audit. Is that the reason you introduced a new function instead of just emitting the event in set_capacity()?

Signature · 1 line
Martin K. Petersen	Oracle Linux Engineering
#7 lore raw

Re: [resend v1 1/5] block/genhd: Notify udev about capacity change

Quoted 11 lines · Martin K. Petersen

Hi Balbir,

Quoted 4 lines

Allow block/genhd to notify user space (via udev) about disk size changes using a new helper disk_set_capacity(), which is a wrapper on top of set_capacity(). disk_set_capacity() will only notify via udev if the current capacity or the target capacity is not zero.

I know set_capacity() is called all over the place making it a bit of a pain to audit. Is that the reason you introduced a new function instead of just emitting the event in set_capacity()?

I did this to avoid having to enforce that set_capacity() implied a notification. Largely to control the impact of the change by default.

Balbir Singh.

#8 lore raw

Re: [resend v1 1/5] block/genhd: Notify udev about capacity change

Balbir,

Quoted 2 lines

I did this to avoid having to enforce that set_capacity() implied a notification. Largely to control the impact of the change by default.

What I thought. I'm OK with set_capacity_and_notify(), btw.

Signature · 6 lines
Martin K. Petersen	Oracle Linux Engineering

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#9 lore raw

Re: [resend v1 1/5] block/genhd: Notify udev about capacity change

Quoted 7 lines · Martin K. Petersen

Balbir,

Quoted 2 lines

I did this to avoid having to enforce that set_capacity() implied a notification. Largely to control the impact of the change by default.

What I thought. I'm OK with set_capacity_and_notify(), btw.

To some extent it might make sense to always notify from set_capacity and have a set_capacity_nonotify if we don't want to notify, as in general we probably should notify unless we have a good reason not to.

#10 lore raw

Re: [resend v1 1/5] block/genhd: Notify udev about capacity change

Quoted 4 lines · On Wed, 2020-01-08 at 16:04 +0100, [email protected] wrote:
Quoted 3 lines · Martin K. Petersen

Balbir,

Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

I did this to avoid having to enforce that set_capacity() implied a notification. Largely to control the impact of the change by default.

What I thought. I'm OK with set_capacity_and_notify(), btw.

To some extent it might make sense to always notify from set_capacity and have a set_capacity_nonotify if we don't want to notify, as in general we probably should notify unless we have a good reason not to.

I am not sure if this is the right path, since with the new interface, we'll now have revalidate disk bits built into the function (see the comments from Bob earlier). I'd be happier converting interfaces a few at a time, A quick grep found over 100 references and I am not even sure how many of them are resizable without hotplug in/out on the fly. Hence, I limited myself to the small subset, I could consider expanding it later based on the need and their ability to resize on the fly

Balbir Singh.

#11 lore raw

PATCH RESEND 2/5 drivers/block/virtio_blk.c: Convert to use disk_set_capacity

In reply to Balbir Singh

block/genhd provides disk_set_capacity() for sending RESIZE notifications via uevents.

---
 drivers/block/virtio_blk.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/block/virtio_blk.c +1 -3
index fbbf18ac1d5d..9848c94a7eb4 100644
@@ -479,18 +479,16 @@ static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
 		   cap_str_10,
 		   cap_str_2);
 
-	set_capacity(vblk->disk, capacity);
+	disk_set_capacity(vblk->disk, capacity);
 }
 
 static void virtblk_config_changed_work(struct work_struct *work)
 {
 	struct virtio_blk *vblk =
 		container_of(work, struct virtio_blk, config_work);
-	char *envp[] = { "RESIZE=1", NULL };
 
 	virtblk_update_capacity(vblk, true);
 	revalidate_disk(vblk->disk);
-	kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
 }
 
 static void virtblk_config_changed(struct virtio_device *vdev)
Signature · 1 line
2.16.5
#12 lore raw

PATCH RESEND 3/5 drivers/block/xen-blkfront.c: Convert to use disk_set_capacity

In reply to Balbir Singh

block/genhd provides disk_set_capacity() for sending RESIZE notifications via uevents.

---
 drivers/block/xen-blkfront.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
drivers/block/xen-blkfront.c +1 -4
index 23c86350a5ab..bed7050a697d 100644
@@ -2337,7 +2337,6 @@ static void blkfront_connect(struct blkfront_info *info)
 	unsigned long sector_size;
 	unsigned int physical_sector_size;
 	unsigned int binfo;
-	char *envp[] = { "RESIZE=1", NULL };
 	int err, i;
 
 	switch (info->connected) {
@@ -2352,10 +2351,8 @@ static void blkfront_connect(struct blkfront_info *info)
 			return;
 		printk(KERN_INFO "Setting capacity to %Lu\n",
 		       sectors);
-		set_capacity(info->gd, sectors);
+		disk_set_capacity(info->gd, sectors);
 		revalidate_disk(info->gd);
-		kobject_uevent_env(&disk_to_dev(info->gd)->kobj,
-				   KOBJ_CHANGE, envp);
 
 		return;
 	case BLKIF_STATE_SUSPENDED:
Signature · 1 line
2.16.5
#13 lore raw

PATCH RESEND 4/5 drivers/nvme/host/core.c: Convert to use disk_set_capacity

In reply to Balbir Singh

block/genhd provides disk_set_capacity() for sending RESIZE notifications via uevents. This notification is newly added to NVME devices

---
 drivers/nvme/host/core.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/nvme/host/core.c +1 -1
index 667f18f465be..cb214e914fc2 100644
@@ -1808,7 +1808,7 @@ static void nvme_update_disk_info(struct gendisk *disk,
 	    ns->lba_shift > PAGE_SHIFT)
 		capacity = 0;
 
-	set_capacity(disk, capacity);
+	disk_set_capacity(disk, capacity);
 
 	nvme_config_discard(disk, ns);
 	nvme_config_write_zeroes(disk, ns);
Signature · 1 line
2.16.5
#14 lore raw

Re: [resend v1 4/5] drivers/nvme/host/core.c: Convert to use disk_set_capacity

Quick question here if user executes nvme ns-rescan /dev/nvme1 will following code result in triggering uevent(s) for the namespace(s( for which there is no change in the size ?

If so is that an expected behavior ?

Quoted 23 lines · Balbir Singh

block/genhd provides disk_set_capacity() for sending RESIZE notifications via uevents. This notification is newly added to NVME devices

---
  drivers/nvme/host/core.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/nvme/host/core.c +1 -1
index 667f18f465be..cb214e914fc2 100644
@@ -1808,7 +1808,7 @@ static void nvme_update_disk_info(struct gendisk *disk,
  	    ns->lba_shift > PAGE_SHIFT)
  		capacity = 0;
 
-	set_capacity(disk, capacity);
+	disk_set_capacity(disk, capacity);
 
  	nvme_config_discard(disk, ns);
  	nvme_config_write_zeroes(disk, ns);
#15 lore raw

Re: [resend v1 4/5] drivers/nvme/host/core.c: Convert to use disk_set_capacity

Quoted 6 lines · Chaitanya Kulkarni

Quick question here if user executes nvme ns-rescan /dev/nvme1 will following code result in triggering uevent(s) for the namespace(s( for which there is no change in the size ?

If so is that an expected behavior ?

My old code had a check to see if old_capacity != new_capacity as well. I can redo those bits if needed.

The expected behaviour is not clear, but the functionality is not broken, user space should be able to deal with a resize event where the previous capacity == new capacity IMHO.

Balbir Singh.

Quoted 29 lines · Balbir Singh
Quoted 26 lines

block/genhd provides disk_set_capacity() for sending RESIZE notifications via uevents. This notification is newly added to NVME devices

---
  drivers/nvme/host/core.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/nvme/host/core.c +1 -1
index 667f18f465be..cb214e914fc2 100644
@@ -1808,7 +1808,7 @@ static void nvme_update_disk_info(struct gendisk
 *disk,
  	    ns->lba_shift > PAGE_SHIFT)
  		capacity = 0;
 
-	set_capacity(disk, capacity);
+	disk_set_capacity(disk, capacity);
 
 
  	nvme_config_discard(disk, ns);
  	nvme_config_write_zeroes(disk, ns);
#16 lore raw

Re: [resend v1 4/5] drivers/nvme/host/core.c: Convert to use disk_set_capacity

Quoted 8 lines · Balbir
Quoted 6 lines · Chaitanya Kulkarni

Quick question here if user executes nvme ns-rescan /dev/nvme1 will following code result in triggering uevent(s) for the namespace(s( for which there is no change in the size ?

If so is that an expected behavior ?

Quoted 6 lines

My old code had a check to see if old_capacity != new_capacity as well. I can redo those bits if needed.

The expected behaviour is not clear, but the functionality is not broken, user space should be able to deal with a resize event where the previous capacity == new capacity IMHO.

I think it makes sense to not bother with a notification unless there is an actual change.

#17 lore raw

Re: [resend v1 4/5] drivers/nvme/host/core.c: Convert to use disk_set_capacity

Christoph,

Quoted 6 lines
Quoted 3 lines

The expected behaviour is not clear, but the functionality is not broken, user space should be able to deal with a resize event where the previous capacity == new capacity IMHO.

I think it makes sense to not bother with a notification unless there is an actual change.

I agree.

Signature · 6 lines
Martin K. Petersen	Oracle Linux Engineering

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#18 lore raw

Re: [resend v1 4/5] drivers/nvme/host/core.c: Convert to use disk_set_capacity

Quoted 11 lines · Martin K. Petersen

Christoph,

Quoted 6 lines
Quoted 3 lines

The expected behaviour is not clear, but the functionality is not broken, user space should be able to deal with a resize event where the previous capacity == new capacity IMHO.

I think it makes sense to not bother with a notification unless there is an actual change.

I agree.

Yes, absolutely.

#19 lore raw

Re: [resend v1 4/5] drivers/nvme/host/core.c: Convert to use disk_set_capacity

In reply to [email protected]
Quoted 12 lines · On Wed, 2020-01-08 at 16:04 +0100, [email protected] wrote:
Quoted 11 lines · Balbir
Quoted 6 lines · Chaitanya Kulkarni

Quick question here if user executes nvme ns-rescan /dev/nvme1 will following code result in triggering uevent(s) for the namespace(s( for which there is no change in the size ?

If so is that an expected behavior ?

My old code had a check to see if old_capacity != new_capacity as well. I can redo those bits if needed.

Quoted 8 lines
Quoted 5 lines

The expected behaviour is not clear, but the functionality is not broken, user
space should be able to deal with a resize event where the previous capacity
== new capacity IMHO.

I think it makes sense to not bother with a notification unless there is an actual change.

[Sorry for the delayed response, just back from LCA]

Agreed!

Balbir Singh

#20 lore raw

PATCH RESEND 5/5 drivers/scsi/sd.c: Convert to use disk_set_capacity

In reply to Balbir Singh

block/genhd provides disk_set_capacity() for sending RESIZE notifications via uevents. This notification is newly added to scsi sd.

---
 drivers/scsi/sd.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/scsi/sd.c +1 -1
index 5afb0046b12a..1a3be30b6b78 100644
@@ -3184,7 +3184,7 @@ static int sd_revalidate_disk(struct gendisk *disk)
 
 	sdkp->first_scan = 0;
 
-	set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
+	disk_set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
 	sd_config_write_same(sdkp);
 	kfree(buffer);
 
Signature · 1 line
2.16.5
#21 lore raw

Re: [resend v1 5/5] drivers/scsi/sd.c: Convert to use disk_set_capacity

Quoted 3 lines · Balbir Singh

block/genhd provides disk_set_capacity() for sending RESIZE notifications via uevents. This notification is newly added to scsi sd.

nit :-

The above commit messages in this series are looking odd from the ones we have in the tree and is it possible to fold it in two lines ?

[Can be done at the time of applying series.]

#22 lore raw

Re: [resend v1 5/5] drivers/scsi/sd.c: Convert to use disk_set_capacity

Quoted 5 lines · Chaitanya Kulkarni
Quoted 3 lines · Balbir Singh

block/genhd provides disk_set_capacity() for sending RESIZE notifications via uevents. This notification is newly added to scsi sd.

Quoted 8 lines

nit :-

The above commit messages in this series are looking odd from the ones we have in the tree and is it possible to fold it in two lines ?

[Can be done at the time of applying series.]

Sounds good! I'll request the maintainer or repost if needed

Balbir Singh.

#23 lore raw

Re: [resend v1 5/5] drivers/scsi/sd.c: Convert to use disk_set_capacity

In reply to Balbir Singh

Balbir,

drivers/scsi/sd.c static int sd_revalidate_disk(struct gendisk *disk) · line 3189

Quoted 12 lines
drivers/scsi/sd.c +1 -1
index 5afb0046b12a..1a3be30b6b78 100644
@@ -3184,7 +3184,7 @@ static int sd_revalidate_disk(struct gendisk *disk)
 
 	sdkp->first_scan = 0;
 
-	set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
+	disk_set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
 	sd_config_write_same(sdkp);
 	kfree(buffer);

We already emit an SDEV_EVT_CAPACITY_CHANGE_REPORTED event if device capacity changes. However, this event does not automatically cause revalidation.

Signature · 6 lines
Martin K. Petersen	Oracle Linux Engineering

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#24 lore raw

Re: [resend v1 5/5] drivers/scsi/sd.c: Convert to use disk_set_capacity

Quoted 13 lines · Martin K. Petersen

Balbir,

Quoted 11 lines
drivers/scsi/sd.c +0 -1
index 5afb0046b12a..1a3be30b6b78 100644
@@ -3184,7 +3184,7 @@ static int sd_revalidate_disk(struct gendisk
 *disk)
 
 	sdkp->first_scan = 0;
 
-	set_capacity(disk, logical_to_sectors(sdp, sdkp-
 >capacity));
Quoted 8 lines

drivers/scsi/sd.c static int sd_revalidate_disk(struct gendisk *disk) · line 3189

Quoted 4 lines
drivers/scsi/sd.c +1 -0
+	disk_set_capacity(disk, logical_to_sectors(sdp, sdkp-
 >capacity));
drivers/scsi/sd.c +0 -0
 	sd_config_write_same(sdkp);
 	kfree(buffer);

We already emit an SDEV_EVT_CAPACITY_CHANGE_REPORTED event if device capacity changes. However, this event does not automatically cause revalidation.

Which I seem to remember was a deliberate choice: some change capacities occur because the path goes passive and default values get installed.

James

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#25 lore raw

Re: [resend v1 5/5] drivers/scsi/sd.c: Convert to use disk_set_capacity

James,

Quoted 7 lines
Quoted 3 lines

We already emit an SDEV_EVT_CAPACITY_CHANGE_REPORTED event if device capacity changes. However, this event does not automatically cause revalidation.

Which I seem to remember was a deliberate choice: some change capacities occur because the path goes passive and default values get installed.

Yep, it's very tricky territory.

Signature · 1 line
Martin K. Petersen	Oracle Linux Engineering
#26 lore raw

Re: [resend v1 5/5] drivers/scsi/sd.c: Convert to use disk_set_capacity

Quoted 6 lines · Martin K. Petersen

James,

Quoted 4 lines
Quoted 3 lines

We already emit an SDEV_EVT_CAPACITY_CHANGE_REPORTED event if device capacity changes. However, this event does not automatically cause revalidation.

Quoted 6 lines
Quoted 3 lines

Which I seem to remember was a deliberate choice: some change capacities occur because the path goes passive and default values get installed.

Yep, it's very tricky territory.

Yes, there are some storage arrays that refuse a READ CAPACITY command in certain ALUA states so you can't get the new capacity anyway.

It might be nice to improve this, though, there are some cases now where we set the capacity to zero when we revalidate and can't get the value.

-Ewan

#27 lore raw

Re: [resend v1 5/5] drivers/scsi/sd.c: Convert to use disk_set_capacity

Ewan,

Quoted 3 lines

Yes, there are some storage arrays that refuse a READ CAPACITY command in certain ALUA states so you can't get the new capacity anyway.

Yep. And some devices will temporarily return a capacity of 0xFFFFFFFF... If we were to trigger a filesystem resize, the results would be disastrous.

Quoted 3 lines

It might be nice to improve this, though, there are some cases now where we set the capacity to zero when we revalidate and can't get the value.

If you have a test case, let's fix it.

Signature · 6 lines
Martin K. Petersen	Oracle Linux Engineering

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#28 lore raw

Re: [resend v1 5/5] drivers/scsi/sd.c: Convert to use disk_set_capacity

Quoted 10 lines · Martin K. Petersen

Ewan,

Quoted 3 lines

Yes, there are some storage arrays that refuse a READ CAPACITY command in certain ALUA states so you can't get the new capacity anyway.

Yep. And some devices will temporarily return a capacity of 0xFFFFFFFF... If we were to trigger a filesystem resize, the results would be disastrous.

Quoted 6 lines
Quoted 3 lines

It might be nice to improve this, though, there are some cases now where we set the capacity to zero when we revalidate and can't get the value.

If you have a test case, let's fix it.

This happens with NVMe fabric devices, I thought. I'll check.

#29 lore raw

Re: [resend v1 5/5] drivers/scsi/sd.c: Convert to use disk_set_capacity

In reply to Martin K. Petersen
Quoted 10 lines · Martin K. Petersen

Balbir,

Quoted 8 lines
drivers/scsi/sd.c +0 -0
index 5afb0046b12a..1a3be30b6b78 100644
@@ -3184,7 +3184,7 @@ static int sd_revalidate_disk(struct gendisk *disk)
 
 	sdkp->first_scan = 0;
 
Quoted 9 lines

drivers/scsi/sd.c static int sd_revalidate_disk(struct gendisk *disk) · line 3189

Quoted 4 lines
drivers/scsi/sd.c +1 -1
-	set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
+	disk_set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
 	sd_config_write_same(sdkp);
 	kfree(buffer);

We already emit an SDEV_EVT_CAPACITY_CHANGE_REPORTED event if device capacity changes. However, this event does not automatically cause revalidation.

The proposed idea is to not reinforce revalidation, unless explictly specified (in the thread before Bob Liu had suggestions). The goal is to notify user space of changes via RESIZE. SCSI sd can opt out of this IOW, I can remove this if you feel SDEV_EVT_CAPACITY_CHANGE_REPORTED is sufficient for current use cases.

Balbir Singh.

#30 lore raw

Re: [resend v1 5/5] drivers/scsi/sd.c: Convert to use disk_set_capacity

Balbir,

Quoted 9 lines
Quoted 3 lines

We already emit an SDEV_EVT_CAPACITY_CHANGE_REPORTED event if device capacity changes. However, this event does not automatically cause revalidation.

The proposed idea is to not reinforce revalidation, unless explictly specified (in the thread before Bob Liu had suggestions). The goal is to notify user space of changes via RESIZE. SCSI sd can opt out of this IOW, I can remove this if you feel SDEV_EVT_CAPACITY_CHANGE_REPORTED is sufficient for current use cases.

I have no particular objection to the code change. I was just observing that in the context of sd.c, RESIZE=1 is more of a "your request to resize was successful" notification due to the requirement of an explicit userland action in case a device reports a capacity change.

Signature · 6 lines
Martin K. Petersen	Oracle Linux Engineering

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#31 lore raw

Re: [resend v1 5/5] drivers/scsi/sd.c: Convert to use disk_set_capacity

Quoted 12 lines · Martin K. Petersen

Balbir,

Quoted 9 lines
Quoted 3 lines

We already emit an SDEV_EVT_CAPACITY_CHANGE_REPORTED event if device capacity changes. However, this event does not automatically cause revalidation.

The proposed idea is to not reinforce revalidation, unless explictly specified (in the thread before Bob Liu had suggestions). The goal is to notify user space of changes via RESIZE. SCSI sd can opt out of this IOW, I can remove this if you feel SDEV_EVT_CAPACITY_CHANGE_REPORTED is sufficient for current use cases.

Quoted 5 lines

I have no particular objection to the code change. I was just observing that in the context of sd.c, RESIZE=1 is more of a "your request to resize was successful" notification due to the requirement of an explicit userland action in case a device reports a capacity change.

That is true, yes I agree with your observation.

Balbir Singh.

#32 lore raw

Re: [resend v1 5/5] drivers/scsi/sd.c: Convert to use disk_set_capacity

In reply to Martin K. Petersen
Quoted 11 lines · Martin K. Petersen

Balbir,

Quoted 9 lines
Quoted 3 lines

We already emit an SDEV_EVT_CAPACITY_CHANGE_REPORTED event if device capacity changes. However, this event does not automatically cause revalidation.

The proposed idea is to not reinforce revalidation, unless explictly specified (in the thread before Bob Liu had suggestions). The goal is to notify user space of changes via RESIZE. SCSI sd can opt out of this IOW, I can remove this if you feel SDEV_EVT_CAPACITY_CHANGE_REPORTED is sufficient for current use cases.

Remember that this event is generated because of a Unit Attention from the device. We are only passing on this indication to udev. It basically allows automation without having to scrape the log file. We don't proactively look. e.g. in the case of SCSI unless you have commands being sent to the device to return the UA status you won't hear about it.

Quoted 6 lines

I have no particular objection to the code change. I was just observing that in the context of sd.c, RESIZE=1 is more of a "your request to resize was successful" notification due to the requirement of an explicit userland action in case a device reports a capacity change.

#33 lore raw

Re: [resend v1 5/5] drivers/scsi/sd.c: Convert to use disk_set_capacity

In reply to Martin K. Petersen
Quoted 11 lines · Martin K. Petersen

Balbir,

Quoted 8 lines
drivers/scsi/sd.c +0 -0
index 5afb0046b12a..1a3be30b6b78 100644
@@ -3184,7 +3184,7 @@ static int sd_revalidate_disk(struct gendisk *disk)
 
 	sdkp->first_scan = 0;
 
Quoted 8 lines

drivers/scsi/sd.c static int sd_revalidate_disk(struct gendisk *disk) · line 3189

Quoted 4 lines
drivers/scsi/sd.c +1 -1
-	set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
+	disk_set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
 	sd_config_write_same(sdkp);
 	kfree(buffer);

We already emit an SDEV_EVT_CAPACITY_CHANGE_REPORTED event if device capacity changes. However, this event does not automatically cause revalidation.

Who is looking at these sdev specific events? (And who is looking at the virtio/xenblk ones?) It makes a lot of sense to have one event supported by all devices. But don't ask me which one would be the best..

#34 lore raw

Re: [resend v1 5/5] drivers/scsi/sd.c: Convert to use disk_set_capacity

Christoph,

Quoted 7 lines
Quoted 3 lines

We already emit an SDEV_EVT_CAPACITY_CHANGE_REPORTED event if device capacity changes. However, this event does not automatically cause revalidation.

Who is looking at these sdev specific events? (And who is looking at the virtio/xenblk ones?) It makes a lot of sense to have one event supported by all devices. But don't ask me which one would be the best..

Users typically have site-specific udev rules or similar. There is no standard entity handling these events. Sadly.

I'm all for standardizing on RESIZE=1. However, we can't really get rid of the emitting existing SDEV* event without breaking existing setups.

Signature · 6 lines
Martin K. Petersen	Oracle Linux Engineering

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#35 lore raw
In reply to Balbir Singh
Quoted 5 lines · Balbir Singh

Allow block/genhd to notify user space about disk size changes using a new helper disk_set_capacity(), which is a wrapper on top of set_capacity(). disk_set_capacity() will only notify if the current capacity or the target capacity is not zero.

set_capacity_and_notify() may be a more straightforward name.

Quoted 12 lines

Background:

As a part of a patch to allow sending the RESIZE event on disk capacity change, Christoph ([email protected]) requested that the patch be made generic and the hacks for virtio block and xen block devices be removed and merged via a generic helper.

This series consists of 5 changes. The first one adds the basic support for changing the size and notifying. The follow up patches are per block subsystem changes. Other block drivers can add their changes as necessary on top of this series.

Quoted 7 lines

Testing:
1. I did some basic testing with an NVME device, by resizing it in the backend and ensured that udevd received the event.

NOTE: After these changes, the notification might happen before revalidate disk, where as it occured later before.

It's better not to change original behavior.
How about something like:

+void set_capacity_and_notify(struct gendisk *disk, sector_t size, bool revalidate)

{
	sector_t capacity = get_capacity(disk);

	set_capacity(disk, size);

+ if (revalidate)

+		revalidate_disk(disk);
	if (capacity != 0 && size != 0) {
		char *envp[] = { "RESIZE=1", NULL };

		kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
	}
}
Quoted 15 lines

Balbir Singh (5):
block/genhd: Notify udev about capacity change drivers/block/virtio_blk.c: Convert to use disk_set_capacity drivers/block/xen-blkfront.c: Convert to use disk_set_capacity drivers/nvme/host/core.c: Convert to use disk_set_capacity drivers/scsi/sd.c: Convert to use disk_set_capacity

 block/genhd.c                | 19 +++++++++++++++++++
 drivers/block/virtio_blk.c   |  4 +---
 drivers/block/xen-blkfront.c |  5 +----
 drivers/nvme/host/core.c     |  2 +-
 drivers/scsi/sd.c            |  2 +-
 include/linux/genhd.h        |  1 +
 6 files changed, 24 insertions(+), 9 deletions(-)
#36 lore raw
Quoted 9 lines · Bob Liu
Quoted 5 lines · Balbir Singh

Allow block/genhd to notify user space about disk size changes using a new helper disk_set_capacity(), which is a wrapper on top of set_capacity(). disk_set_capacity() will only notify if the current capacity or the target capacity is not zero.

set_capacity_and_notify() may be a more straightforward name.

Yes, agreed.

Quoted 30 lines
Quoted 12 lines

Background:

As a part of a patch to allow sending the RESIZE event on disk capacity change, Christoph ([email protected]) requested that the patch be made generic and the hacks for virtio block and xen block devices be removed and merged via a generic helper.

This series consists of 5 changes. The first one adds the basic support for changing the size and notifying. The follow up patches are per block subsystem changes. Other block drivers can add their changes as necessary on top of this series.

Quoted 7 lines

Testing:
1. I did some basic testing with an NVME device, by resizing it in the backend and ensured that udevd received the event.

NOTE: After these changes, the notification might happen before revalidate disk, where as it occured later before.

It's better not to change original behavior.
How about something like:

+void set_capacity_and_notify(struct gendisk *disk, sector_t size, bool revalidate)

{
	sector_t capacity = get_capacity(disk);

	set_capacity(disk, size);

Review

Quoted 2 lines
(quoted diff) +2 -0
+        if (revalidate)
+		revalidate_disk(disk);

Do you see a concern with the notification going out before revalidate_disk? I could keep the behaviour and come up with a suitable name

revalidate_disk_and_notify() (set_capacity is a part of the revalidation process), or feel free to suggest a better name

Thanks,
Balbir Singh

Reviewed-by
#37 lore raw
Quoted 41 lines · Balbir
Quoted 9 lines · Bob Liu
Quoted 5 lines · Balbir Singh

Allow block/genhd to notify user space about disk size changes using a new helper disk_set_capacity(), which is a wrapper on top of set_capacity(). disk_set_capacity() will only notify if the current capacity or the target capacity is not zero.

set_capacity_and_notify() may be a more straightforward name.

Yes, agreed.

Quoted 28 lines
Quoted 12 lines

Background:

As a part of a patch to allow sending the RESIZE event on disk capacity change, Christoph ([email protected]) requested that the patch be made generic and the hacks for virtio block and xen block devices be removed and merged via a generic helper.

This series consists of 5 changes. The first one adds the basic support for changing the size and notifying. The follow up patches are per block subsystem changes. Other block drivers can add their changes as necessary on top of this series.

Quoted 7 lines

Testing:
1. I did some basic testing with an NVME device, by resizing it in the backend and ensured that udevd received the event.

NOTE: After these changes, the notification might happen before revalidate disk, where as it occured later before.

It's better not to change original behavior.
How about something like:

+void set_capacity_and_notify(struct gendisk *disk, sector_t size, bool revalidate)

{
	sector_t capacity = get_capacity(disk);
Quoted 6 lines

Review

Quoted 4 lines
	set_capacity(disk, size);
(quoted diff) +2 -0
+        if (revalidate)
+		revalidate_disk(disk);

Do you see a concern with the notification going out before revalidate_disk?

Not aware any, but keep the original behavior is safer especial when not must change..

Quoted 5 lines

I could keep the behaviour and come up with a suitable name

revalidate_disk_and_notify() (set_capacity is a part of the revalidation process), or feel free to suggest a better name

Feel free to add my reviewed-by next version in this series.

Quoted 3 lines

Thanks,
Balbir Singh

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme