Skip to contentSkip to conversation

scsi: core: Fix shost->cmd_per_lun calculation in scsi_add_host_with_dma()

PATCH Superseded Dexuan Cui 4 participants 5 messages Oct 7, 2021 – Oct 8, 2021

Subsystems (4)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Dexuan Cui

author maintainer
#0 lore raw

PATCH v2 1/1 scsi: core: Fix shost->cmd_per_lun calculation in scsi_add_host_with_dma()

After commit ea2f0f77538c, a 416-CPU VM running on Hyper-V hangs during boot because scsi_add_host_with_dma() sets shost->cmd_per_lun to a negative number (the below numbers may differ in different kernel versions):
in drivers/scsi/storvsc_drv.c, storvsc_drv_init() sets
'max_outstanding_req_per_channel' to 352, and storvsc_probe() sets 'max_sub_channels' to (416 - 1) / 4 = 103 and sets scsi_driver.can_queue to 352 * (103 + 1) * (100 - 10) / 100 = 32947, which exceeds SHRT_MAX.

Use min_t(int, ...) to fix the issue.

Fixes ea2f0f77538c ("scsi: core: Cap scsi_host cmd_per_lun at can_queue")

---

v1 tried to fix the issue by changing the storvsc driver:
https://lwn.net/ml/linux-kernel/BYAPR21MB1270BBC14D5F1AE69FC31A16BFB09@BYAPR21MB1270.namprd21.prod.outlook.com/

v2 directly fixes the scsi core change instead as Michael Kelley and John Garry suggested (refer to the above link).

 drivers/scsi/hosts.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/scsi/hosts.c +2 -1
index 3f6f14f0cafb..24b72ee4246f 100644
@@ -220,7 +220,8 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
 		goto fail;
 	}
 
-	shost->cmd_per_lun = min_t(short, shost->cmd_per_lun,
+	/* Use min_t(int, ...) in case shost->can_queue exceeds SHRT_MAX */
+	shost->cmd_per_lun = min_t(int, shost->cmd_per_lun,
 				   shost->can_queue);
 
 	error = scsi_init_sense_cache(shost);
Signature · 1 line
2.25.1
maintainer Reviewed-by on v2 1/1
#1 lore raw
Quoted 55 lines

-----Original Message-----
From: Dexuan Cui <[email protected]>
Sent: Thursday, October 7, 2021 1:50 PM
To: KY Srinivasan <[email protected]>; Stephen Hemminger <[email protected]>; [email protected]; [email protected]; [email protected]; Haiyang Zhang <[email protected]>; [email protected]; [email protected]; [email protected]; linux- [email protected]; [email protected]; Long Li <[email protected]>; Michael Kelley <[email protected]>
Cc: [email protected]; Dexuan Cui <[email protected]>; [email protected]
Subject: [PATCH v2] scsi: core: Fix shost->cmd_per_lun calculation in scsi_add_host_with_dma()

After commit ea2f0f77538c, a 416-CPU VM running on Hyper-V hangs during boot because scsi_add_host_with_dma() sets shost->cmd_per_lun to a negative number (the below numbers may differ in different kernel versions):
in drivers/scsi/storvsc_drv.c, storvsc_drv_init() sets
'max_outstanding_req_per_channel' to 352, and storvsc_probe() sets 'max_sub_channels' to (416 - 1) / 4 = 103 and sets scsi_driver.can_queue to
352 * (103 + 1) * (100 - 10) / 100 = 32947, which exceeds SHRT_MAX.

Use min_t(int, ...) to fix the issue.

Fixes ea2f0f77538c ("scsi: core: Cap scsi_host cmd_per_lun at

can_queue")

---

v1 tried to fix the issue by changing the storvsc driver:
https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Flwn.ne
t%2Fml%2Flinux- kernel%2FBYAPR21MB1270BBC14D5F1AE69FC31A16BFB09%40BYAPR21MB1270.namprd21 .prod.outlook.com%2F&amp;data=04%7C01%7Chaiyangz%40microsoft.com%7C366e6 d0bf755492c631c08d989baf4b9%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7 C637692258384408217%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoi V2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&amp;sdata=RbNgx1aBBBzfHC3p EdKyBZWaaQIQXS3U%2FItEQUe4NfQ%3D&amp;reserved=0

v2 directly fixes the scsi core change instead as Michael Kelley and John Garry suggested (refer to the above link).

 drivers/scsi/hosts.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/scsi/hosts.c +0 -0
index 3f6f14f0cafb..24b72ee4246f 100644

drivers/scsi/hosts.c int scsi_add_host_with_dma(struct Scsi_Host *shost, · line 226

Quoted 8 lines
drivers/scsi/hosts.c +2 -1
@@ -220,7 +220,8 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost,
 struct device *dev,
 		goto fail;
 	}
 
-	shost->cmd_per_lun = min_t(short, shost->cmd_per_lun,
+	/* Use min_t(int, ...) in case shost->can_queue exceeds SHRT_MAX */
+	shost->cmd_per_lun = min_t(int, shost->cmd_per_lun,
 				   shost->can_queue);

Since shost->can_queue is int, the min_t type should also be int (the longer type of the two vars).

#2 lore raw
In reply to Dexuan Cui
Quoted 7 lines · Dexuan Cui

After commit ea2f0f77538c, a 416-CPU VM running on Hyper-V hangs during boot because scsi_add_host_with_dma() sets shost->cmd_per_lun to a negative number (the below numbers may differ in different kernel versions):
in drivers/scsi/storvsc_drv.c, storvsc_drv_init() sets
'max_outstanding_req_per_channel' to 352, and storvsc_probe() sets 'max_sub_channels' to (416 - 1) / 4 = 103 and sets scsi_driver.can_queue to 352 * (103 + 1) * (100 - 10) / 100 = 32947, which exceeds SHRT_MAX.

I think that you just need to mention that if can_queue exceeds SHRT_MAX, then there is a data truncation issue.

Quoted 6 lines

Use min_t(int, ...) to fix the issue.

Fixes ea2f0f77538c ("scsi: core: Cap scsi_host cmd_per_lun at can_queue")

It looks ok, I'd just like to test it a bit more.

Thanks,
John

Quoted 7 lines

---

v1 tried to fix the issue by changing the storvsc driver:
https://lwn.net/ml/linux-kernel/BYAPR21MB1270BBC14D5F1AE69FC31A16BFB09@BYAPR21MB1270.namprd21.prod.outlook.com/

v2 directly fixes the scsi core change instead as Michael Kelley and John Garry suggested (refer to the above link).

To be fair, it was Michael's suggestion

Quoted 19 lines
  drivers/scsi/hosts.c | 3 ++-
  1 file changed, 2 insertions(+), 1 deletion(-)
drivers/scsi/hosts.c +2 -1
index 3f6f14f0cafb..24b72ee4246f 100644
@@ -220,7 +220,8 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
  		goto fail;
  	}
 
-	shost->cmd_per_lun = min_t(short, shost->cmd_per_lun,
+	/* Use min_t(int, ...) in case shost->can_queue exceeds SHRT_MAX */
+	shost->cmd_per_lun = min_t(int, shost->cmd_per_lun,
  				   shost->can_queue);
 
  	error = scsi_init_sense_cache(shost);
author maintainer
#3 lore raw
Quoted 12 lines

From: John Garry <[email protected]>
Sent: Thursday, October 7, 2021 2:42 PM

Quoted 7 lines · Dexuan Cui

After commit ea2f0f77538c, a 416-CPU VM running on Hyper-V hangs during boot because scsi_add_host_with_dma() sets shost->cmd_per_lun to a negative number (the below numbers may differ in different kernel versions):
in drivers/scsi/storvsc_drv.c, storvsc_drv_init() sets
'max_outstanding_req_per_channel' to 352, and storvsc_probe() sets 'max_sub_channels' to (416 - 1) / 4 = 103 and sets scsi_driver.can_queue to 352 * (103 + 1) * (100 - 10) / 100 = 32947, which exceeds SHRT_MAX.

Quoted 2 lines

I think that you just need to mention that if can_queue exceeds SHRT_MAX, then there is a data truncation issue.

I just hoped the explanation how the too big 'can_queue' value is generated is helpful.

OK, I think I can change the commit log to:

After commit ea2f0f77538c, a 416-CPU VM running on Hyper-V hangs during boot because the hv_storvsc driver sets scsi_driver.can_queue to an "int" value that exceeds SHRT_MAX, and hence scsi_add_host_with_dma() sets shost->cmd_per_lun to a negative "short" number.

Use min_t(int, ...) to fix the issue.

Quoted 4 lines

It looks ok, I'd just like to test it a bit more.

Thanks,
John

Thanks! I'll post v3 with the above commit log, and I look forward to your review/test.

Quoted 4 lines
Quoted 2 lines

v2 directly fixes the scsi core change instead as Michael Kelley and John Garry suggested (refer to the above link).

To be fair, it was Michael's suggestion

Yeah. Michael always gives good suggstions when reviewing patches. :-)

Reviewed-by on v2 1/1
#4 lore raw
In reply to Dexuan Cui
Quoted 29 lines · Dexuan Cui

After commit ea2f0f77538c, a 416-CPU VM running on Hyper-V hangs during boot because scsi_add_host_with_dma() sets shost->cmd_per_lun to a negative number (the below numbers may differ in different kernel versions):
in drivers/scsi/storvsc_drv.c, storvsc_drv_init() sets
'max_outstanding_req_per_channel' to 352, and storvsc_probe() sets 'max_sub_channels' to (416 - 1) / 4 = 103 and sets scsi_driver.can_queue to 352 * (103 + 1) * (100 - 10) / 100 = 32947, which exceeds SHRT_MAX.

Use min_t(int, ...) to fix the issue.

Fixes ea2f0f77538c ("scsi: core: Cap scsi_host cmd_per_lun at can_queue")

---

v1 tried to fix the issue by changing the storvsc driver:
https://lwn.net/ml/linux-kernel/BYAPR21MB1270BBC14D5F1AE69FC31A16BFB09@BYAPR21MB1270.namprd21.prod.outlook.com/

v2 directly fixes the scsi core change instead as Michael Kelley and John Garry suggested (refer to the above link).

 drivers/scsi/hosts.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/scsi/hosts.c +0 -0
index 3f6f14f0cafb..24b72ee4246f 100644

drivers/scsi/hosts.c int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev, · line 225

Quoted 7 lines
drivers/scsi/hosts.c +2 -1
@@ -220,7 +220,8 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
 		goto fail;
 	}
 
-	shost->cmd_per_lun = min_t(short, shost->cmd_per_lun,
+	/* Use min_t(int, ...) in case shost->can_queue exceeds SHRT_MAX */
+	shost->cmd_per_lun = min_t(int, shost->cmd_per_lun,
 				   shost->can_queue);

Looks fine:

Signature · 1 line
Ming