CA-165611: DM Multipath mishandles blacklisted devices if some paths …

…down

This bug may occur when [1 <= paths < total_num_paths] to a device are offline
and the device is blacklisted. Up until now, we checked only the first path in
the list, which might be offline. If this was the case, we incorrectly
determined that the device was not blacklisted (the lack of the
"wwid blacklisted" string doesn't mean it's whitelisted).

To tackle the issue, we check all paths to the device until we have a definite
answer.

Signed-off-by: Kostas Ladopoulos <konstantinos.ladopoulos@citrix.com>
Reviewed-by: Germano Percossi <germano.percossi@citrix.com>

GitHub: closes #254

drivers/mpath_dmp.py

@@ -181,19 +181,30 @@ def map_by_scsibus(sid,npaths=0):
         if(len(devices)>=npaths or npaths==0):
             # Enable this device's sid: it could be blacklisted
             # We expect devices to be blacklisted according to their
-            # wwid only. Checking the first one is sufficient
-            if wwid_conf.is_blacklisted(devices[0]):
+            # wwid only. We go through the list of paths until we have
+            # a definite answer about the device's blacklist status.
+            # If the path we are checking is down, we cannot tell.
+            for dev in devices:
                 try:
-                    wwid_conf.edit_wwid(sid)
-                except:
-                    util.SMlog("WARNING: exception raised while attempting to"
-                               " modify multipath.conf")
-                try:
-                    mpath_cli.reconfigure()
-                except:
-                    util.SMlog("WARNING: exception raised while attempting to"
-                               " reconfigure")
-                time.sleep(5)
+                    if wwid_conf.is_blacklisted(dev):
+                        try:
+                            wwid_conf.edit_wwid(sid)
+                        except:
+                            util.SMlog("WARNING: exception raised while "
+                                       "attempting to modify multipath.conf")
+                        try:
+                            mpath_cli.reconfigure()
+                        except:
+                            util.SMlog("WARNING: exception raised while "
+                                       "attempting to reconfigure")
+                        time.sleep(5)
+
+                    break
+                except wwid_conf.WWIDException as e:
+                    util.SMlog(e.errstr)
+            else:
+                util.SMlog("Device 'SCSI_id: {}' is inaccessible; "
+                           "All paths are down.".format(sid))
 
             __map_explicit(devices)
             return

drivers/wwid_conf.py

@@ -30,6 +30,10 @@
 CONF_FILE = "/etc/multipath.xenserver/multipath.conf"
 BELIST_TAG = "blacklist_exceptions"
 
+class WWIDException(Exception):
+    def __init__(self, errstr):
+        self.errstr = errstr
+
 
 def edit_wwid(wwid, remove=False):
     """Add a wwid to the list of exceptions or remove if
@@ -75,17 +79,33 @@ def is_blacklisted(dev):
     """This function returns True if the device is blacklisted according
     to multipath.conf rules.
 
+    There is also the possibility that the path to the device we are
+    checking is unavailable for some reason. In that case,
+    is_blacklisted() cannot tell if the device is blacklisted or not
+    and a WWIDException is raised.
+
     It cannot be used to check the current daemon rules because it
     could be running with an old configuration file in memory
 
-    dev -- it is any string accepted by "multipath -c". A full path
-    is sufficient
+    Input:
+        dev -- it is any string accepted by "multipath -c". A full path
+               is expected.
+
+    Return:
+        bool -- True or False, depending on whether the device is
+                blacklisted or not.
 
+    Raise:
+        WWIDException
     """
 
     (rc,stdout,stderr) = util.doexec(['/sbin/multipath','-v3','-c',dev])
 
-    # If the devices is blacklisted according to multipath.conf, the
+    if "scope is nul" in stdout:
+        raise WWIDException("WARNING: Could not retrieve device's "
+                            "WWID. Path {} is down".format(dev))
+
+    # If the device is blacklisted according to multipath.conf, the
     # string "wwid blacklisted" appears in the verbose output.
     # This is a very fragile mechanism and keeps changing.
     # What we want is a method to tell immediately if a device is
@@ -146,11 +166,14 @@ def usage():
         usage()
         sys.exit(1)
 
-    if force or xor(remove, is_blacklisted(device)):
-        try:
-            edit_wwid(wwid, remove)
-        except:
-            sys.exit(1)
+    try:
+        if force or xor(remove, is_blacklisted(device)):
+            try:
+                edit_wwid(wwid, remove)
+            except:
+                sys.exit(1)
+    except WWIDException as e:
+        util.SMlog(e.errstr)
 
     sys.exit(0)