Skip to contentSkip to conversation

updates to UAS

PATCH Superseded Oliver Neukum 2 participants 6 messages Sep 16, 2020

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Oliver Neukum

author maintainer
#0 lore raw

PATCH 1/1 updates to UAS

The first patch just modernizes to new helpers. No functional change intended

The second patch fixes a corner case where a whole tree of devices is removed and I got a report of a live lock

author maintainer
#1 lore raw

PATCH 1/2 UAS: use macro for reporting results

The SCSI layer has introduced a new macro for recording the result of a command. Use it.

---
 drivers/usb/storage/uas.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
drivers/usb/storage/uas.c +5 -5
index d592071119ba..842ca5c82091 100644
@@ -279,17 +279,17 @@ static bool uas_evaluate_response_iu(struct response_iu *riu, struct scsi_cmnd *
 
 	switch (response_code) {
 	case RC_INCORRECT_LUN:
-		cmnd->result = DID_BAD_TARGET << 16;
+		set_host_byte(cmnd, DID_BAD_TARGET);
 		break;
 	case RC_TMF_SUCCEEDED:
-		cmnd->result = DID_OK << 16;
+		set_host_byte(cmnd, DID_OK);
 		break;
 	case RC_TMF_NOT_SUPPORTED:
-		cmnd->result = DID_TARGET_FAILURE << 16;
+		set_host_byte(cmnd, DID_TARGET_FAILURE);
 		break;
 	default:
 		uas_log_cmd_state(cmnd, "response iu", response_code);
-		cmnd->result = DID_ERROR << 16;
+		set_host_byte(cmnd, DID_ERROR);
 		break;
 	}
 
@@ -660,7 +660,7 @@ static int uas_queuecommand_lck(struct scsi_cmnd *cmnd,
 	spin_lock_irqsave(&devinfo->lock, flags);
 
 	if (devinfo->resetting) {
-		cmnd->result = DID_ERROR << 16;
+		set_host_byte(cmnd, DID_ERROR);
 		cmnd->scsi_done(cmnd);
 		spin_unlock_irqrestore(&devinfo->lock, flags);
 		return 0;
Signature · 1 line
2.16.4
author maintainer
#2 lore raw

PATCH 2/2 UAS: fix disconnect by unplugging a hub

In reply to Oliver Neukum

The SCSI layer can go into an ugly loop if you ignore that a device is gone. You need to report an error in the command rather than in the return value of the queue method.
We need to specifically check for ENODEV..

---
 drivers/usb/storage/uas.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
drivers/usb/storage/uas.c +12 -2
index 842ca5c82091..19e730f9ad19 100644
@@ -662,8 +662,7 @@ static int uas_queuecommand_lck(struct scsi_cmnd *cmnd,
 	if (devinfo->resetting) {
 		set_host_byte(cmnd, DID_ERROR);
 		cmnd->scsi_done(cmnd);
-		spin_unlock_irqrestore(&devinfo->lock, flags);
-		return 0;
+		goto zombie;
 	}
 
 	/* Find a free uas-tag */
@@ -699,6 +698,16 @@ static int uas_queuecommand_lck(struct scsi_cmnd *cmnd,
 		cmdinfo->state &= ~(SUBMIT_DATA_IN_URB | SUBMIT_DATA_OUT_URB);
 
 	err = uas_submit_urbs(cmnd, devinfo);
+	/*
+	 * in case of fatal errors the SCSI layer is peculiar
+	 * a command that has finished is a success for the purpose
+	 * of queueing, no matter how fatal the error
+	 */
+	if (err == -ENODEV) {
+		set_host_byte(cmnd, DID_ERROR);
+		cmnd->scsi_done(cmnd);
+		goto zombie;
+	}
 	if (err) {
 		/* If we did nothing, give up now */
 		if (cmdinfo->state & SUBMIT_STATUS_URB) {
@@ -709,6 +718,7 @@ static int uas_queuecommand_lck(struct scsi_cmnd *cmnd,
 	}
 
 	devinfo->cmnd[idx] = cmnd;
+zombie:
 	spin_unlock_irqrestore(&devinfo->lock, flags);
 	return 0;
 }
Signature · 1 line
2.16.4
maintainer
#3 lore raw

Re: [PATCH 2/2] UAS: fix disconnect by unplugging a hub

Quoted 9 lines · Oliver Neukum

The SCSI layer can go into an ugly loop if you ignore that a device is gone. You need to report an error in the command rather than in the return value of the queue method.
We need to specifically check for ENODEV..

---
 drivers/usb/storage/uas.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

Should this one go to the stable kernels?

thanks,

greg k-h

author maintainer
#4 lore raw

Re: [PATCH 2/2] UAS: fix disconnect by unplugging a hub

Am Dienstag, den 15.09.2020, 16:00 +0200 schrieb Greg KH:

Quoted 6 lines · Oliver Neukum
Quoted 5 lines

The SCSI layer can go into an ugly loop if you ignore that a device is gone. You need to report an error in the command rather than in the return value of the queue method.
We need to specifically check for ENODEV..

Quoted 6 lines
Quoted 4 lines
---
 drivers/usb/storage/uas.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

Should this one go to the stable kernels?

Hi,

in theory yes, but it depends on the previous patch. Should I submit a separate patch?

Regards
Oliver

maintainer
#5 lore raw

Re: [PATCH 2/2] UAS: fix disconnect by unplugging a hub

Quoted 12 lines · Oliver Neukum

Am Dienstag, den 15.09.2020, 16:00 +0200 schrieb Greg KH:

Quoted 11 lines · Oliver Neukum
Quoted 9 lines

The SCSI layer can go into an ugly loop if you ignore that a device is gone. You need to report an error in the command rather than in the return value of the queue method.
We need to specifically check for ENODEV..

---
 drivers/usb/storage/uas.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
Quoted 6 lines
Quoted 1 line

Should this one go to the stable kernels?

Hi,

in theory yes, but it depends on the previous patch. Should I submit a separate patch?

Reordering these would be best, thanks!

greg k-h