From: Long Li <[email protected]>
FC disks are usually setup in a multipath system, and they don't want to
unconditionaly reset I/O on timeout. I/O timeout is detected by multipath
as a good time to failover and recover.
Signed-off-by: Long Li <[email protected]>
---
drivers/scsi/storvsc_drv.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c
index 8d955db..d60b5ea 100644
--- a/drivers/scsi/storvsc_drv.c
+++ b/drivers/scsi/storvsc_drv.c
@@ -486,6 +486,7 @@ struct hv_host_device {
unsigned int port;
unsigned char path;
unsigned char target;
+ bool is_fc;
};
struct storvsc_scan_work {
@@ -1495,6 +1496,11 @@ static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
*/
static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
{
+ struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
+
+ if (host_dev->is_fc)
+ return BLK_EH_NOT_HANDLED;
+
return BLK_EH_RESET_TIMER;
}
@@ -1738,6 +1744,7 @@ static int storvsc_probe(struct hv_device *device,
host_dev->port = host->host_no;
host_dev->dev = device;
+ host_dev->is_fc = is_fc;
stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
--
2.7.4
On 06/13/2017 02:23 AM, Long Li wrote:
> From: Long Li <[email protected]>
>
> FC disks are usually setup in a multipath system, and they don't want to
> unconditionaly reset I/O on timeout. I/O timeout is detected by multipath
> as a good time to failover and recover.
>
> Signed-off-by: Long Li <[email protected]>
> ---
Looks good,
Reviewed-by: Johannes Thumshirn <[email protected]>
--
Johannes Thumshirn Storage
[email protected] +49 911 74053 689
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Felix Imendörffer, Jane Smithard, Graham Norton
HRB 21284 (AG Nürnberg)
Key fingerprint = EC38 9CAB C2C4 F25D 8600 D0D0 0393 969D 2D76 0850
On Mon, 2017-06-12 at 17:23 -0700, Long Li wrote:
> From: Long Li <[email protected]>
>
> FC disks are usually setup in a multipath system, and they don't want to
> unconditionaly reset I/O on timeout. I/O timeout is detected by multipath
> as a good time to failover and recover.
>
> Signed-off-by: Long Li <[email protected]>
> ---
> drivers/scsi/storvsc_drv.c | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c
> index 8d955db..d60b5ea 100644
> --- a/drivers/scsi/storvsc_drv.c
> +++ b/drivers/scsi/storvsc_drv.c
> @@ -486,6 +486,7 @@ struct hv_host_device {
> unsigned int port;
> unsigned char path;
> unsigned char target;
> + bool is_fc;
> };
>
> struct storvsc_scan_work {
> @@ -1495,6 +1496,11 @@ static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
> */
> static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
> {
> + struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
> +
> + if (host_dev->is_fc)
> + return BLK_EH_NOT_HANDLED;
> +
> return BLK_EH_RESET_TIMER;
> }
>
> @@ -1738,6 +1744,7 @@ static int storvsc_probe(struct hv_device *device,
>
> host_dev->port = host->host_no;
> host_dev->dev = device;
> + host_dev->is_fc = is_fc;
>
>
> stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
Hello Long,
As far as I know there is no other SCSI driver nor block driver in the Linux
kernel tree that returns BLK_EH_RESET_TIMER unconditionally. Would a valid
alternative fix be to remove storvsc_eh_timed_out() entirely? If not, what
would break if that function would be removed entirely?
Additionally, for FC, shouldn't that timeout handler handle the "port blocked"
state? Shouldn't fc_eh_timed_out() be used for FC instead of just returning
BLK_EH_NOT_HANDLED?
Thanks,
Bart.
> -----Original Message-----
> From: Bart Van Assche [mailto:[email protected]]
> Sent: Tuesday, June 13, 2017 8:38 AM
> To: [email protected]; KY Srinivasan <[email protected]>;
> [email protected]; [email protected]; Stephen Hemminger
> <[email protected]>; Long Li <[email protected]>;
> [email protected]; Haiyang Zhang <[email protected]>
> Cc: [email protected]; Long Li <[email protected]>
> Subject: Re: [Possible Phish Fraud][PATCH] storvsc: use default I/O timeout
> handler for FC devices
>
> On Mon, 2017-06-12 at 17:23 -0700, Long Li wrote:
> > From: Long Li <[email protected]>
> >
> > FC disks are usually setup in a multipath system, and they don't want
> > to unconditionaly reset I/O on timeout. I/O timeout is detected by
> > multipath as a good time to failover and recover.
> >
> > Signed-off-by: Long Li <[email protected]>
> > ---
> > drivers/scsi/storvsc_drv.c | 7 +++++++
> > 1 file changed, 7 insertions(+)
> >
> > diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c
> > index 8d955db..d60b5ea 100644
> > --- a/drivers/scsi/storvsc_drv.c
> > +++ b/drivers/scsi/storvsc_drv.c
> > @@ -486,6 +486,7 @@ struct hv_host_device {
> > unsigned int port;
> > unsigned char path;
> > unsigned char target;
> > + bool is_fc;
> > };
> >
> > struct storvsc_scan_work {
> > @@ -1495,6 +1496,11 @@ static int storvsc_host_reset_handler(struct
> scsi_cmnd *scmnd)
> > */
> > static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd
> > *scmnd) {
> > + struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
> > +
> > + if (host_dev->is_fc)
> > + return BLK_EH_NOT_HANDLED;
> > +
> > return BLK_EH_RESET_TIMER;
> > }
> >
> > @@ -1738,6 +1744,7 @@ static int storvsc_probe(struct hv_device
> > *device,
> >
> > host_dev->port = host->host_no;
> > host_dev->dev = device;
> > + host_dev->is_fc = is_fc;
> >
> >
> > stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
>
> Hello Long,
>
> As far as I know there is no other SCSI driver nor block driver in the Linux kernel
> tree that returns BLK_EH_RESET_TIMER unconditionally. Would a valid
> alternative fix be to remove storvsc_eh_timed_out() entirely? If not, what
> would break if that function would be removed entirely?
Storvsc handles virtualized devices for Hyper-V and Azure. The host decides on how to handle I/O timeout, and makes decision based on how long I/O has been in flight. This is why we use BLK_EH_RESET_TIMER and leave the I/O timeout handling to the host for most cases.
>
> Additionally, for FC, shouldn't that timeout handler handle the "port blocked"
> state? Shouldn't fc_eh_timed_out() be used for FC instead of just returning
> BLK_EH_NOT_HANDLED?
I agree with you. Storvsc didn't expose rport state. But this has changed with a recent commit daf0cd445a218314f9461d67d4f2b9c24cdd534b. I'll send a V2 to address this.
Thank you
Long
>
> Thanks,
>
> Bart.