Is it a problem that after virEventRegisterDefaultImpl we have handlers leaked

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hello,

When I ran programme event-test compiled from event-test.c, I found a problem that, after virEventRegisterDefaultImpl I do virConnectOpenAuth and virConnectClose, there will be handlers of socket and pipe opened by virConnectOpenAuth leaked after virConnectClose. So I did some analysis, and I found the fact following:

 

In the condition that we only have one connection here

 

int virNetSocketAddIOCallback(virNetSocketPtr sock,

                              int events,

                              virNetSocketIOFunc func,

                              void *opaque,

                              virFreeCallback ff)

{

    int ret = -1;

 

    virObjectRef(sock); //Here we add sock refers once, then we will get refers equal 2 after

    virObjectLock(sock);

    if (sock->watch > 0) {

        VIR_DEBUG("Watch already registered on socket %p", sock);

        goto cleanup;

    }

 

    if ((sock->watch = virEventAddHandle(sock->fd, //If we have called virEventRegisterDefaultImpl, then here we’ll get a sock watch non negative and will not go to cleanup.

                                         events,

                                         virNetSocketEventHandle,

                                         sock,

                                         virNetSocketEventFree)) < 0) {

        VIR_DEBUG("Failed to register watch on socket %p", sock);

        goto cleanup;

    }

    sock->func = func;

    sock->opaque = opaque;

    sock->ff = ff;

 

    ret = 0;

 

cleanup:

    virObjectUnlock(sock);

    if (ret != 0)

        virObjectUnref(sock); //If we haven’t called virEventRegisterDefaultImpl, we’ll be here after virEventAddHandle, and sock refers will decrease to 1

    return ret;

}

 

Condition with virEventRegisterDefaultImpl, we’ll do unrefer action in two places:

1.       virEventRunDefaultImpl  ->virEventPollRunOnce  ->virEventRunDefaultImpl  ->virEventPollRunOnce  ->virEventPollCleanupHandles -> virNetSocketEventFree -> virObjectUnref

2.       virConnectClose ->virObjectUnref ->virConnectDispose ->remoteConnectClose  ->doRemoteClose  ->virNetClientClose  ->virNetClientCloseInternal -> virNetClientIOEventLoopPassTheBuck -> virNetClientCloseLocked -> virObjectUnref

 

When event dealing loop is alive, everything work fine, we’ll get sock refers 2 after virConnectOpenAuth and unrefer twice in two places above after virConnectClose. But If some accidents happened like we quit event dealing loop or virEventRunDefaultImpl suspended, then sock refers will never be zero, and handlers will never be freed.

In a particular condition we have two threads, thread one doing event dealing loop, thread two doing something like virConnectOpenAuth and virConnectClose ceaselessly, then we’ll get a big trouble that handlers available will be ate up soon.

 

I consider to add something like virEventDeregisterDefaultImpl to set addHandleImpl and his buddy NULL. Apparently it is far away from fixing it completely.

 

At last I have two questions here:

 

1.       Is it a problem that after virEventRegisterDefaultImpl we have handlers leaked?

2.       If it is a problem, is there anybody any good idea to help me out?

 

 

Best Regards,

-WangYufei

 

--
libvir-list mailing list
libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list

[Index of Archives]     [Virt Tools]     [Libvirt Users]     [Lib OS Info]     [Fedora Users]     [Fedora Desktop]     [Fedora SELinux]     [Big List of Linux Books]     [Yosemite News]     [KDE Users]     [Fedora Tools]