Re: SCSI eats error from flush failure during hot plug

[Christoph Hellwig <hch@infradead.org>]

On Thu, Jul 03, 2014 at 09:57:32AM -0700, Steven Haber wrote:
> Both patches work fine -- zero uncommitted writes over several hundred
> power cycles.

Thanks for the testing.

Below is James' patch with a trivial comment fix and a proper changelog.
James, can I get your signoff for this one?  I'll throw it into the
core-for-3.16 tree then.

---
>From 92510736caa3b14a964af981ba288950042e79e3 Mon Sep 17 00:00:00 2001
From: James Bottomley <James.Bottomley@HansenPartnership.com>
Date: Thu, 3 Jul 2014 19:17:34 +0200
Subject: scsi: handle flush errors properly

Flush commands don't transfer data and thus need to be special cased
in the I/O completion handler so that we can propagate errors to
the block layer and filesystem.

Reported-by: Steven Haber <steven@qumulo.com>
Tested-by: Steven Haber <steven@qumulo.com>
---
 drivers/scsi/scsi_lib.c |    8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index f7e3163..143d0e2 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -733,6 +733,14 @@ void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
 			scsi_next_command(cmd);
 			return;
 		}
+	} else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
+		/*
+		 * Certain non BLOCK_PC requests are commands that don't
+		 * actually transfer anything (FLUSH), so cannot use
+		 * good_bytes == 0 as the signal for an error.  This sets the
+		 * error explicitly for the problem case.
+		 */
+		error = __scsi_error_from_host_byte(cmd, result);
 	}
 
 	/* no bidi support for !REQ_TYPE_BLOCK_PC yet */
-- 
1.7.10.4