On Thu, Jul 15, 2010 at 10:44:46AM -0400, Chris Lalancette wrote:

If you try to execute two concurrent migrations p2p from A->B and B->A, the two libvirtd's will deadlock trying to perform the migrations. The reason for this is that in p2p migration, the libvirtd's are responsible for making the RPC Prepare, Migrate, and Finish calls. However, they are currently holding the driver lock while doing so, which basically guarantees deadlock in this scenario.

This patch fixes the situation by adding qemuDomainObjEnterRemoteWithDriver and qemuDomainObjExitRemoteWithDriver helper methods. The Enter take an additional object reference, then drops both the domain object lock and the driver lock. The Exit takes both the driver and domain object lock, then drops the reference. Adding calls to these Enter and Exit helpers around remote calls in the various migration methods seems to fix the problem for me in testing.

I *believe* that this makes the situation safe, though I'm not 100% certain. The additional domain object reference ensures that the domain object won't disappear while this operation is happening. The BeginJob that is called inside of qemudDomainMigratePerform ensures that we can't execute a second migrate (or shutdown, or save, etc) job while the migration is active. That being said, my understanding of the locking is still a bit shaky, so I'd like some additional confirmation that this is indeed safe.

Signed-off-by: Chris Lalancette <clalance@redhat.com> --- src/qemu/qemu_driver.c | 39 ++++++++++++++++++++++++++++++++------- 1 files changed, 32 insertions(+), 7 deletions(-)

diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c index 5f2607c..e81af84 100644 --- a/src/qemu/qemu_driver.c +++ b/src/qemu/qemu_driver.c @@ -531,6 +531,21 @@ static void qemuDomainObjExitMonitorWithDriver(struct qemud_driver *driver, virD } }

+static void qemuDomainObjEnterRemoteWithDriver(struct qemud_driver *driver, + virDomainObjPtr obj) +{ + virDomainObjRef(obj); + virDomainObjUnlock(obj); + qemuDriverUnlock(driver); +} + +static void qemuDomainObjExitRemoteWithDriver(struct qemud_driver *driver, + virDomainObjPtr obj) +{ + qemuDriverLock(driver); + virDomainObjLock(obj); + virDomainObjUnref(obj); +}

static int qemuCgroupControllerActive(struct qemud_driver *driver, int controller) @@ -10753,9 +10768,11 @@ static int doTunnelMigrate(virDomainPtr dom, /* virStreamNew only fails on OOM, and it reports the error itself */ goto cleanup;

+ qemuDomainObjEnterRemoteWithDriver(driver, vm); internalret = dconn->driver->domainMigratePrepareTunnel(dconn, st, flags, dname, resource, dom_xml); + qemuDomainObjExitRemoteWithDriver(driver, vm);

if (internalret < 0) /* domainMigratePrepareTunnel sets the error for us */ @@ -10833,8 +10850,10 @@ cancel:

finish: dname = dname ? dname : dom->name; + qemuDomainObjEnterRemoteWithDriver(driver, vm); ddomain = dconn->driver->domainMigrateFinish2 (dconn, dname, NULL, 0, uri, flags, retval); + qemuDomainObjExitRemoteWithDriver(driver, vm);

cleanup: if (client_sock != -1) @@ -10873,16 +10892,20 @@ static int doNonTunnelMigrate(virDomainPtr dom, virDomainPtr ddomain = NULL; int retval = -1; char *uri_out = NULL; + int rc;

+ qemuDomainObjEnterRemoteWithDriver(driver, vm); /* NB we don't pass 'uri' into this, since that's the libvirtd * URI in this context - so we let dest pick it */ - if (dconn->driver->domainMigratePrepare2(dconn, - NULL, /* cookie */ - 0, /* cookielen */ - NULL, /* uri */ - &uri_out, - flags, dname, - resource, dom_xml) < 0) + rc = dconn->driver->domainMigratePrepare2(dconn, + NULL, /* cookie */ + 0, /* cookielen */ + NULL, /* uri */ + &uri_out, + flags, dname, + resource, dom_xml); + qemuDomainObjExitRemoteWithDriver(driver, vm); + if (rc < 0) /* domainMigratePrepare2 sets the error for us */ goto cleanup;

@@ -10899,8 +10922,10 @@ static int doNonTunnelMigrate(virDomainPtr dom,

finish: dname = dname ? dname : dom->name; + qemuDomainObjEnterRemoteWithDriver(driver, vm); ddomain = dconn->driver->domainMigrateFinish2 (dconn, dname, NULL, 0, uri_out, flags, retval); + qemuDomainObjExitRemoteWithDriver(driver, vm);

if (ddomain) virUnrefDomain(ddomain);

Explanations and patch both looks fine to me, but I would feel more confident if Dan reviewed it too, the fine details of low level locking escape me I'm afraid. mini-ACK :-) Daniel -- Daniel Veillard | libxml Gnome XML XSLT toolkit http://xmlsoft.org/ daniel@veillard.com | Rpmfind RPM search engine http://rpmfind.net/ http://veillard.com/ | virtualization library http://libvirt.org/