Whamcloud - gitweb
LU-6010 lnet: prevent assert on LNet module unload 10/13110/16
authorAmir Shehata <amir.shehata@intel.com>
Wed, 17 Dec 2014 17:35:15 +0000 (09:35 -0800)
committerOleg Drokin <oleg.drokin@intel.com>
Sun, 8 Feb 2015 02:45:45 +0000 (02:45 +0000)
There is a use case where lnet can be unloaded while there are
no NIs configured.  Removing lnet in this case will cause
LNetFini() to be called without a prior call to LNetNIFini().
This will cause the LASSERT(the_lnet.ln_refcount == 0) to be
triggered.

To deal with this use case when LNet is configured a reference
count on the module is taken using try_module_get().  This way
LNet must be unconfigured before it could be removed; therefore
avoiding the above case.  When LNet is unconfigured module_put()
is called to return the reference count.

Signed-off-by: Amir Shehata <amir.shehata@intel.com>
Change-Id: I0f283eeb395fa9a076a4d65ab3edd5e7807fc169
Reviewed-on: http://review.whamcloud.com/13110
Tested-by: Jenkins
Reviewed-by: James Simmons <uja.ornl@gmail.com>
Reviewed-by: Doug Oucharek <doug.s.oucharek@intel.com>
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
lnet/lnet/module.c

index 197ff25..a8f39b3 100644 (file)
@@ -53,13 +53,21 @@ lnet_configure(void *arg)
        LNET_MUTEX_LOCK(&lnet_config_mutex);
 
        if (!the_lnet.ln_niinit_self) {
+               rc = try_module_get(THIS_MODULE);
+
+               if (rc != 1)
+                       goto out;
+
                rc = LNetNIInit(LNET_PID_LUSTRE);
                if (rc >= 0) {
                        the_lnet.ln_niinit_self = 1;
                        rc = 0;
+               } else {
+                       module_put(THIS_MODULE);
                }
        }
 
+out:
        LNET_MUTEX_UNLOCK(&lnet_config_mutex);
        return rc;
 }
@@ -67,21 +75,23 @@ lnet_configure(void *arg)
 static int
 lnet_unconfigure (void)
 {
-        int   refcount;
+       int   refcount;
+
+       LNET_MUTEX_LOCK(&lnet_config_mutex);
 
-        LNET_MUTEX_LOCK(&lnet_config_mutex);
+       if (the_lnet.ln_niinit_self) {
+               the_lnet.ln_niinit_self = 0;
+               LNetNIFini();
+               module_put(THIS_MODULE);
+       }
 
-        if (the_lnet.ln_niinit_self) {
-                the_lnet.ln_niinit_self = 0;
-                LNetNIFini();
-        }
+       LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex);
+       refcount = the_lnet.ln_refcount;
+       LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
 
-        LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex);
-        refcount = the_lnet.ln_refcount;
-        LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
+       LNET_MUTEX_UNLOCK(&lnet_config_mutex);
 
-        LNET_MUTEX_UNLOCK(&lnet_config_mutex);
-        return (refcount == 0) ? 0 : -EBUSY;
+       return (refcount == 0) ? 0 : -EBUSY;
 }
 
 static int