Skip to contentSkip to conversation

nvme/core: Notify udev about capacity change

RFC PATCH In review Balbir Singh 4 participants 8 messages Nov 29, 2019 – Dec 3, 2019 NVMe

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 nvme/core: Notify udev about capacity change

From: Someswarudu Sangaraju <[email protected]>

NVME capacity can change asynchronously via NVME_AER_NOTICE_NS_CHANGED The current stack does the right thing and resizes the disk via set_capacity() called from nvme_update_disk_info(). This patch adds a user space notification via uevent to let udev know that a resize occurred if the capacity changes.

[Balbir - send the notification only on capacity change and wrote the changelog]

---
 drivers/nvme/host/core.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/nvme/host/core.c +6 -0
index dfe37a525f3a..a413aee3107d 100644
@@ -1757,6 +1757,7 @@ static void nvme_update_disk_info(struct gendisk *disk,
 		struct nvme_ns *ns, struct nvme_id_ns *id)
 {
 	sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
+	sector_t old_capacity = get_capacity(disk);
 	unsigned short bs = 1 << ns->lba_shift;
 	u32 atomic_bs, phys_bs, io_opt;
 
@@ -1807,6 +1808,11 @@ static void nvme_update_disk_info(struct gendisk *disk,
 		capacity = 0;
 
 	set_capacity(disk, capacity);
+	if (old_capacity != capacity) {
+		char *envp[] = { "RESIZE=1", NULL };
+
+		kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
+	}
 
 	nvme_config_discard(disk, ns);
 	nvme_config_write_zeroes(disk, ns);
Signature · 7 lines
2.16.5


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

I don't think this makes sense, and they way it has been done in virtio_blk and xen-blkfront is an utter hack. Please first add a properly documented block helper, and then make sure it doesn't send the even for the first call that moves from a 0 capacity. SCSI also will need updating to use the helper.

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

#2 lore raw
Quoted 5 lines · Christoph Hellwig

I don't think this makes sense, and they way it has been done in virtio_blk and xen-blkfront is an utter hack. Please first add a properly documented block helper, and then make sure it doesn't send the even for the first call that moves from a 0 capacity. SCSI also will need updating to use the helper.

Makes sense, I am wondering if I should refactor set_capacity to __set_capacity and have set_capacity do notifcations for

oldsize vs newsize where oldsize is not 0

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

#3 lore raw
Quoted 7 lines · Balbir
Quoted 5 lines · Christoph Hellwig

I don't think this makes sense, and they way it has been done in virtio_blk and xen-blkfront is an utter hack. Please first add a properly documented block helper, and then make sure it doesn't send the even for the first call that moves from a 0 capacity. SCSI also will need updating to use the helper.

Quoted 5 lines

Makes sense, I am wondering if I should refactor set_capacity to __set_capacity and have set_capacity do notifcations for

oldsize vs newsize where oldsize is not 0

I pretty much hate the set_capacity name to start with, so what about a more descriptive name for new the helper? blk_set_capacity?

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

#4 lore raw
Quoted 2 lines · On 12/02/2019 10:54 PM, [email protected] wrote:

I pretty much hate the set_capacity name to start with, so what about a more descriptive name for new the helper? blk_set_capacity?

Do we also need a blk_mq_ops or appropriate block layer callback capacity change function, so that file systems can register a callback function to take an appropriate action, e.g. rebuilding the metadata with new size ?

If this make sense I'll be happy to take a look.

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

#5 lore raw
Quoted 10 lines · Chaitanya Kulkarni
Quoted 2 lines · On 12/02/2019 10:54 PM, [email protected] wrote:

I pretty much hate the set_capacity name to start with, so what about a more descriptive name for new the helper? blk_set_capacity?

Do we also need a blk_mq_ops or appropriate block layer callback capacity change function, so that file systems can register a callback function to take an appropriate action, e.g. rebuilding the metadata with new size ?

If this make sense I'll be happy to take a look.

Well, and op isn't going to help with file system notifications as that goes the other way around. We'd probably need some way how file systems can register a new callback into the existing block device events framework that is used for media changes.

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

#6 lore raw
Quoted 10 lines · On Tue, Dec 03, 2019 at 08:34:47AM +0100, [email protected] wrote:
Quoted 9 lines · Chaitanya Kulkarni
Quoted 2 lines · On 12/02/2019 10:54 PM, [email protected] wrote:

I pretty much hate the set_capacity name to start with, so what about a more descriptive name for new the helper? blk_set_capacity?

Do we also need a blk_mq_ops or appropriate block layer callback capacity change function, so that file systems can register a callback function to take an appropriate action, e.g. rebuilding the metadata with new size ?

Quoted 6 lines
Quoted 1 line

If this make sense I'll be happy to take a look.

Well, and op isn't going to help with file system notifications as that goes the other way around. We'd probably need some way how file systems can register a new callback into the existing block device events framework that is used for media changes.

I had this on my TODO List but one thing stopped my endeavors, having a "notifier" for resize is easy when we only grow the block-device, but what should the FS do when we shrink it?

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

#7 lore raw
Quoted 3 lines · Johannes Thumshirn

I had this on my TODO List but one thing stopped my endeavors, having a "notifier" for resize is easy when we only grow the block-device, but what should the FS do when we shrink it?

Assuming this is a "normal" fs and not something fancy with built in redundancy - shut the file system down and stop any new access to it as that would just make things worse.

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