Skip to contentSkip to conversation

sparc64: sunvdc: Complete I/O request if vdisk server error

PATCH Superseded chris hyser 2 participants 2 messages Nov 19, 2014 – Nov 21, 2014 Block layer

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 sparc64: sunvdc: Complete I/O request if vdisk server error

Hi All,

In vdc_end_one() of sunvdc.c a non-zero status indicates an error from Solaris vds. In the case of an actual disk as the vds backend, the error comes from the corresponding Solaris driver and may or may not represent an actual vdisk error and the I/O size may be 0. In this case __blk_end_request() does not complete the request and there may be no further actions to complete it. This results in an indefinite hang of the process initiating the I/O.

with patch:

chrish2> dd if=/dev/zero of=/dev/vdiskc count=1
sunvdc: virtual disk server I/O error 5
end_request: I/O error, dev vdiskc, sector 0 Buffer I/O error on device vdiskc, logical block 0 lost page write due to I/O error on vdiskc 1+0 records in
1+0 records out
512 bytes (512 B) copied, 0.000695392 s, 736 kB/s

Without this patch, dd will hang and cannot be killed. Note, 5 is a Solaris EIO. Still not sure what that is about. :-(

-chrish

-- >8 --
From 018ad8a7087a195f082dfdc2ae53288eae4a7487 Mon Sep 17 00:00:00 2001
From: chris hyser <[email protected]>
Date: Wed, 19 Nov 2014 12:41:56 -0800
Subject: [PATCH] sparc64: sunvdc: Complete I/O request if vdisk server error

If virtual disk server returns an I/O with an error and 0 data transferred, complete the I/O with EIO to ensure initiating process does not hang indefinitely.

drivers/block/sunvdc.c +6 -1
index 0ebadf9..17e9a50 100644
@@ -279,7 +279,12 @@ static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr,
 
  	rqe->req = NULL;
 
-	__blk_end_request(req, (desc->status ? -EIO : 0), desc->size);
+	if (desc->status && desc->size = 0) {
+		printk(KERN_ERR PFX "virtual disk server I/O error %d\n",
+				desc->status);
+		__blk_end_request_all(req, -EIO);
+	} else
+		__blk_end_request(req, (desc->status ? -EIO : 0) , desc->size);
 
  	/* restart blk queue when ring is half emptied */
  	if (blk_queue_stopped(port->disk->queue) &&
#1 lore raw

From: chris hyser <[email protected]>
Date: Wed, 19 Nov 2014 18:52:34 -0500

Quoted 4 lines

In vdc_end_one() of sunvdc.c a non-zero status indicates an error from Solaris
vds. In the case of an actual disk as the vds backend, the error comes from the

In plain text, this looks terrible.

Please properly format your commit message body to ~80 columns, but don't do it in a way where you get a full line of text, then a line with a few words, then another full line of text.

Look at the plain ascii representation of the result and think about what that will look like to other people trying to read it. Don't assume some GUI will magically fix the formatting for you.