SCSI: fix regression in scsi_send_eh_cmnd()

Commit ac61d19 (scsi: set correct completion code in
scsi_send_eh_cmnd()) introduced a bug.  It changed the stored return
value from a queuecommand call, but it didn't take into account that
the return value was used again later on.  This patch fixes the bug by
changing the later usage.

There is a big comment in the middle of scsi_send_eh_cmnd() which
does a good job of explaining how the routine works.  But it mentions
a "rtn = FAILURE" value that doesn't exist in the code.  This patch
adjusts the code to match the comment (I assume the comment is right
and the code is wrong).

This fixes Bugzilla #88341.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Reported-by: Андрей Аладьев <aladjev.andrew@gmail.com>
Tested-by: Андрей Аладьев <aladjev.andrew@gmail.com>
Fixes: ac61d19
Acked-by: Hannes Reinecke <hare@suse.de>
Cc: <stable@vger.kernel.org>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>

drivers/scsi/scsi_error.c

@@ -1041,7 +1041,7 @@ static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
 		}
 		/* signal not to enter either branch of the if () below */
 		timeleft = 0;
-		rtn = NEEDS_RETRY;
+		rtn = FAILED;
 	} else {
 		timeleft = wait_for_completion_timeout(&done, timeout);
 		rtn = SUCCESS;
@@ -1081,7 +1081,7 @@ static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
 			rtn = FAILED;
 			break;
 		}
-	} else if (!rtn) {
+	} else if (rtn != FAILED) {
 		scsi_abort_eh_cmnd(scmd);
 		rtn = FAILED;
 	}