Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that colo-compare is active
* Zhang, Chen (chen.zh...@intel.com) wrote: > > > > -Original Message- > > From: Lukas Straub > > Sent: Friday, May 8, 2020 2:11 PM > > To: Zhang, Chen > > Cc: qemu-devel ; Li Zhijian > > ; Jason Wang ; Marc- > > André Lureau ; Paolo Bonzini > > > > Subject: Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that > > colo-compare is active > > > > On Fri, 8 May 2020 02:26:21 + > > "Zhang, Chen" wrote: > > > > > > -Original Message- > > > > From: Lukas Straub > > > > Sent: Thursday, May 7, 2020 11:54 PM > > > > To: Zhang, Chen > > > > Cc: qemu-devel ; Li Zhijian > > > > ; Jason Wang ; Marc- > > > > André Lureau ; Paolo Bonzini > > > > > > > > Subject: Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check > > > > that colo-compare is active > > > > > > > > On Thu, 7 May 2020 11:38:04 + > > > > "Zhang, Chen" wrote: > > > > > > > > > > -----Original Message- > > > > > > From: Lukas Straub > > > > > > Sent: Monday, May 4, 2020 6:28 PM > > > > > > To: qemu-devel > > > > > > Cc: Zhang, Chen ; Li Zhijian > > > > > > ; Jason Wang ; > > > > > > Marc- André Lureau ; Paolo Bonzini > > > > > > > > > > > > Subject: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check > > > > > > that > > > > > > colo- compare is active > > > > > > > > > > > > If the colo-compare object is removed before failover and a > > > > > > checkpoint happens, qemu crashes because it tries to lock the > > > > > > destroyed event_mtx in colo_notify_compares_event. > > > > > > > > > > > > Fix this by checking if everything is initialized by introducing > > > > > > a new variable colo_compare_active which is protected by a new > > > > > > mutex colo_compare_mutex. The new mutex also protects against > > > > > > concurrent access of the net_compares list and makes sure that > > > > > > colo_notify_compares_event isn't active while we destroy > > > > > > event_mtx and event_complete_cond. > > > > > > > > > > > > With this it also is again possible to use colo without > > > > > > colo-compare (periodic > > > > > > mode) and to use multiple colo-compare for multiple network > > interfaces. > > > > > > > > > > > > > > > > Hi Lukas, > > > > > > > > > > For this case I think we don't need to touch vl.c code, we can > > > > > solve this > > > > issue from another perspective: > > > > > How to remove colo-compare? > > > > > User will use qemu-monitor or QMP command to disable an object, so > > > > > we just need return operation failed When user try to remove > > > > > colo-compare > > > > object while COLO is running. > > > > > > > > Yeah, but that still leaves the other problem that colo can't be > > > > used without colo-compare (qemu crashes then). > > > > > > Yes, the COLO-compare is necessary module in COLO original design. > > > At most cases, user need it do dynamic sync. > > > For rare cases, maybe we can add a new colo-compare parameter to > > bypass all the network workload. > > > > I think such an parameter would only be a workaround instead of a real > > solution like this patch. > > The root problem is why COLO-compare is necessary. > Yes, maybe someone want to use pure periodic synchronization mode, > But it means it will lost all guest network support(without > colo-compare/filter-mirror/filter-redirector/filter-rewriter). > The secondary guest just a solid backup for the primary guest, when occur > failover the new build stateful connection (like TCP) > will crashed, need userspace to handle this status. It lost the original > meaning for COLO FT/HA solution, no need use do HA in application layer. > it looks like normal/remote periodic VM snapshot here. > Dave or Jason have any comments here? People have done fixed-rate VM synchronisation in the past; and there are workloads where the packet stream is particularly random so you almost always get comparison miscompares. But in those setups, the rest of the configuration is very
RE: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that colo-compare is active
> -Original Message- > From: Lukas Straub > Sent: Saturday, May 9, 2020 8:21 PM > To: Zhang, Chen > Cc: qemu-devel ; Li Zhijian > ; Jason Wang ; Marc- > André Lureau ; Paolo Bonzini > ; Dr . David Alan Gilbert > Subject: Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that > colo-compare is active > > On Fri, 8 May 2020 06:50:39 + > "Zhang, Chen" wrote: > > > > -Original Message- > > > From: Lukas Straub > > > Sent: Friday, May 8, 2020 2:11 PM > > > To: Zhang, Chen > > > Cc: qemu-devel ; Li Zhijian > > > ; Jason Wang ; Marc- > > > André Lureau ; Paolo Bonzini > > > > > > Subject: Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check > > > that colo-compare is active > > > > > > On Fri, 8 May 2020 02:26:21 + > > > "Zhang, Chen" wrote: > > > > > > > > -Original Message- > > > > > From: Lukas Straub > > > > > Sent: Thursday, May 7, 2020 11:54 PM > > > > > To: Zhang, Chen > > > > > Cc: qemu-devel ; Li Zhijian > > > > > ; Jason Wang ; > > > > > Marc- André Lureau ; Paolo Bonzini > > > > > > > > > > Subject: Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: > > > > > Check that colo-compare is active > > > > > > > > > > On Thu, 7 May 2020 11:38:04 + "Zhang, Chen" > > > > > wrote: > > > > > > > > > > > > -Original Message- > > > > > > > From: Lukas Straub > > > > > > > Sent: Monday, May 4, 2020 6:28 PM > > > > > > > To: qemu-devel > > > > > > > Cc: Zhang, Chen ; Li Zhijian > > > > > > > ; Jason Wang > > > > > > > ; > > > > > > > Marc- André Lureau ; Paolo > > > > > > > Bonzini > > > > > > > Subject: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: > > > > > > > Check that > > > > > > > colo- compare is active > > > > > > > > > > > > > > If the colo-compare object is removed before failover and a > > > > > > > checkpoint happens, qemu crashes because it tries to lock > > > > > > > the destroyed event_mtx in colo_notify_compares_event. > > > > > > > > > > > > > > Fix this by checking if everything is initialized by > > > > > > > introducing a new variable colo_compare_active which is > > > > > > > protected by a new mutex colo_compare_mutex. The new mutex > > > > > > > also protects against concurrent access of the net_compares > > > > > > > list and makes sure that colo_notify_compares_event isn't > > > > > > > active while we destroy event_mtx and event_complete_cond. > > > > > > > > > > > > > > With this it also is again possible to use colo without > > > > > > > colo-compare (periodic > > > > > > > mode) and to use multiple colo-compare for multiple network > > > interfaces. > > > > > > > > > > > > > > > > > > > Hi Lukas, > > > > > > > > > > > > For this case I think we don't need to touch vl.c code, we can > > > > > > solve this > > > > > issue from another perspective: > > > > > > How to remove colo-compare? > > > > > > User will use qemu-monitor or QMP command to disable an > > > > > > object, so we just need return operation failed When user try > > > > > > to remove colo-compare > > > > > object while COLO is running. > > > > > > > > > > Yeah, but that still leaves the other problem that colo can't be > > > > > used without colo-compare (qemu crashes then). > > > > > > > > Yes, the COLO-compare is necessary module in COLO original design. > > > > At most cases, user need it do dynamic sync. > > > > For rare cases, maybe we can add a new colo-compare parameter to > > > bypass all the network workload. > > > > > > I think such an parameter would only be a workaround instead of a > > > real solution like this patch. > > > > The root problem is why COLO-compare is necessary. > > Yes, maybe someone want to use
Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that colo-compare is active
On Fri, 8 May 2020 06:50:39 + "Zhang, Chen" wrote: > > -Original Message- > > From: Lukas Straub > > Sent: Friday, May 8, 2020 2:11 PM > > To: Zhang, Chen > > Cc: qemu-devel ; Li Zhijian > > ; Jason Wang ; Marc- > > André Lureau ; Paolo Bonzini > > > > Subject: Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that > > colo-compare is active > > > > On Fri, 8 May 2020 02:26:21 + > > "Zhang, Chen" wrote: > > > > > > -Original Message- > > > > From: Lukas Straub > > > > Sent: Thursday, May 7, 2020 11:54 PM > > > > To: Zhang, Chen > > > > Cc: qemu-devel ; Li Zhijian > > > > ; Jason Wang ; Marc- > > > > André Lureau ; Paolo Bonzini > > > > > > > > Subject: Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check > > > > that colo-compare is active > > > > > > > > On Thu, 7 May 2020 11:38:04 + > > > > "Zhang, Chen" wrote: > > > > > > > > > > -Original Message----- > > > > > > From: Lukas Straub > > > > > > Sent: Monday, May 4, 2020 6:28 PM > > > > > > To: qemu-devel > > > > > > Cc: Zhang, Chen ; Li Zhijian > > > > > > ; Jason Wang ; > > > > > > Marc- André Lureau ; Paolo Bonzini > > > > > > > > > > > > Subject: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check > > > > > > that > > > > > > colo- compare is active > > > > > > > > > > > > If the colo-compare object is removed before failover and a > > > > > > checkpoint happens, qemu crashes because it tries to lock the > > > > > > destroyed event_mtx in colo_notify_compares_event. > > > > > > > > > > > > Fix this by checking if everything is initialized by introducing > > > > > > a new variable colo_compare_active which is protected by a new > > > > > > mutex colo_compare_mutex. The new mutex also protects against > > > > > > concurrent access of the net_compares list and makes sure that > > > > > > colo_notify_compares_event isn't active while we destroy > > > > > > event_mtx and event_complete_cond. > > > > > > > > > > > > With this it also is again possible to use colo without > > > > > > colo-compare (periodic > > > > > > mode) and to use multiple colo-compare for multiple network > > interfaces. > > > > > > > > > > > > > > > > Hi Lukas, > > > > > > > > > > For this case I think we don't need to touch vl.c code, we can > > > > > solve this > > > > issue from another perspective: > > > > > How to remove colo-compare? > > > > > User will use qemu-monitor or QMP command to disable an object, so > > > > > we just need return operation failed When user try to remove > > > > > colo-compare > > > > object while COLO is running. > > > > > > > > Yeah, but that still leaves the other problem that colo can't be > > > > used without colo-compare (qemu crashes then). > > > > > > Yes, the COLO-compare is necessary module in COLO original design. > > > At most cases, user need it do dynamic sync. > > > For rare cases, maybe we can add a new colo-compare parameter to > > bypass all the network workload. > > > > I think such an parameter would only be a workaround instead of a real > > solution like this patch. > > The root problem is why COLO-compare is necessary. > Yes, maybe someone want to use pure periodic synchronization mode, > But it means it will lost all guest network support(without > colo-compare/filter-mirror/filter-redirector/filter-rewriter). > The secondary guest just a solid backup for the primary guest, when occur > failover the new build stateful connection (like TCP) > will crashed, need userspace to handle this status. It lost the original > meaning for COLO FT/HA solution, no need use do HA in application layer. > it looks like normal/remote periodic VM snapshot here. Sure, but maybe the user doesn't need (reliable) network on failover. Also proper network support with periodic mode can easily be implemented by modifying filter-buffer to buffer packets until checkpoint. Being able to use colo wit
RE: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that colo-compare is active
> -Original Message- > From: Lukas Straub > Sent: Friday, May 8, 2020 2:11 PM > To: Zhang, Chen > Cc: qemu-devel ; Li Zhijian > ; Jason Wang ; Marc- > André Lureau ; Paolo Bonzini > > Subject: Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that > colo-compare is active > > On Fri, 8 May 2020 02:26:21 + > "Zhang, Chen" wrote: > > > > -Original Message- > > > From: Lukas Straub > > > Sent: Thursday, May 7, 2020 11:54 PM > > > To: Zhang, Chen > > > Cc: qemu-devel ; Li Zhijian > > > ; Jason Wang ; Marc- > > > André Lureau ; Paolo Bonzini > > > > > > Subject: Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check > > > that colo-compare is active > > > > > > On Thu, 7 May 2020 11:38:04 + > > > "Zhang, Chen" wrote: > > > > > > > > -Original Message- > > > > > From: Lukas Straub > > > > > Sent: Monday, May 4, 2020 6:28 PM > > > > > To: qemu-devel > > > > > Cc: Zhang, Chen ; Li Zhijian > > > > > ; Jason Wang ; > > > > > Marc- André Lureau ; Paolo Bonzini > > > > > > > > > > Subject: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check > > > > > that > > > > > colo- compare is active > > > > > > > > > > If the colo-compare object is removed before failover and a > > > > > checkpoint happens, qemu crashes because it tries to lock the > > > > > destroyed event_mtx in colo_notify_compares_event. > > > > > > > > > > Fix this by checking if everything is initialized by introducing > > > > > a new variable colo_compare_active which is protected by a new > > > > > mutex colo_compare_mutex. The new mutex also protects against > > > > > concurrent access of the net_compares list and makes sure that > > > > > colo_notify_compares_event isn't active while we destroy > > > > > event_mtx and event_complete_cond. > > > > > > > > > > With this it also is again possible to use colo without > > > > > colo-compare (periodic > > > > > mode) and to use multiple colo-compare for multiple network > interfaces. > > > > > > > > > > > > > Hi Lukas, > > > > > > > > For this case I think we don't need to touch vl.c code, we can > > > > solve this > > > issue from another perspective: > > > > How to remove colo-compare? > > > > User will use qemu-monitor or QMP command to disable an object, so > > > > we just need return operation failed When user try to remove > > > > colo-compare > > > object while COLO is running. > > > > > > Yeah, but that still leaves the other problem that colo can't be > > > used without colo-compare (qemu crashes then). > > > > Yes, the COLO-compare is necessary module in COLO original design. > > At most cases, user need it do dynamic sync. > > For rare cases, maybe we can add a new colo-compare parameter to > bypass all the network workload. > > I think such an parameter would only be a workaround instead of a real > solution like this patch. The root problem is why COLO-compare is necessary. Yes, maybe someone want to use pure periodic synchronization mode, But it means it will lost all guest network support(without colo-compare/filter-mirror/filter-redirector/filter-rewriter). The secondary guest just a solid backup for the primary guest, when occur failover the new build stateful connection (like TCP) will crashed, need userspace to handle this status. It lost the original meaning for COLO FT/HA solution, no need use do HA in application layer. it looks like normal/remote periodic VM snapshot here. Dave or Jason have any comments here? Thanks Zhang Chen > > Regards, > Lukas Straub > > > Thanks > > Zhang Chen > > > > > > > > Regards, > > > Lukas Straub > > > > > > > Thanks > > > > Zhang Chen > > > > > > > > > Signed-off-by: Lukas Straub > > > > > --- > > > > > net/colo-compare.c | 35 +-- > > > > > net/colo-compare.h | 1 + > > > > > softmmu/vl.c | 2 ++ > > > > > 3 files changed, 32 insertions(+), 6 deletions(-) > > > > > > > > > > diff --git a/net/colo-compare.c b/net/colo-compare.c index > > > &
Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that colo-compare is active
On Fri, 8 May 2020 02:26:21 + "Zhang, Chen" wrote: > > -Original Message- > > From: Lukas Straub > > Sent: Thursday, May 7, 2020 11:54 PM > > To: Zhang, Chen > > Cc: qemu-devel ; Li Zhijian > > ; Jason Wang ; Marc- > > André Lureau ; Paolo Bonzini > > > > Subject: Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that > > colo-compare is active > > > > On Thu, 7 May 2020 11:38:04 + > > "Zhang, Chen" wrote: > > > > > > -Original Message- > > > > From: Lukas Straub > > > > Sent: Monday, May 4, 2020 6:28 PM > > > > To: qemu-devel > > > > Cc: Zhang, Chen ; Li Zhijian > > > > ; Jason Wang ; Marc- > > > > André Lureau ; Paolo Bonzini > > > > > > > > Subject: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that > > > > colo- compare is active > > > > > > > > If the colo-compare object is removed before failover and a > > > > checkpoint happens, qemu crashes because it tries to lock the > > > > destroyed event_mtx in colo_notify_compares_event. > > > > > > > > Fix this by checking if everything is initialized by introducing a > > > > new variable colo_compare_active which is protected by a new mutex > > > > colo_compare_mutex. The new mutex also protects against concurrent > > > > access of the net_compares list and makes sure that > > > > colo_notify_compares_event isn't active while we destroy event_mtx > > > > and event_complete_cond. > > > > > > > > With this it also is again possible to use colo without colo-compare > > > > (periodic > > > > mode) and to use multiple colo-compare for multiple network interfaces. > > > > > > > > > > Hi Lukas, > > > > > > For this case I think we don't need to touch vl.c code, we can solve this > > > > > issue from another perspective: > > > How to remove colo-compare? > > > User will use qemu-monitor or QMP command to disable an object, so we > > > just need return operation failed When user try to remove colo-compare > > object while COLO is running. > > > > Yeah, but that still leaves the other problem that colo can't be used > > without > > colo-compare (qemu crashes then). > > Yes, the COLO-compare is necessary module in COLO original design. > At most cases, user need it do dynamic sync. > For rare cases, maybe we can add a new colo-compare parameter to bypass all > the network workload. I think such an parameter would only be a workaround instead of a real solution like this patch. Regards, Lukas Straub > Thanks > Zhang Chen > > > > > Regards, > > Lukas Straub > > > > > Thanks > > > Zhang Chen > > > > > > > Signed-off-by: Lukas Straub > > > > --- > > > > net/colo-compare.c | 35 +-- > > > > net/colo-compare.h | 1 + > > > > softmmu/vl.c | 2 ++ > > > > 3 files changed, 32 insertions(+), 6 deletions(-) > > > > > > > > diff --git a/net/colo-compare.c b/net/colo-compare.c index > > > > 56db3d3bfc..c7572d75e9 100644 > > > > --- a/net/colo-compare.c > > > > +++ b/net/colo-compare.c > > > > @@ -54,6 +54,8 @@ static NotifierList colo_compare_notifiers = > > > > #define REGULAR_PACKET_CHECK_MS 3000 #define > > DEFAULT_TIME_OUT_MS > > > > 3000 > > > > > > > > +static QemuMutex colo_compare_mutex; static bool > > > > +colo_compare_active; > > > > static QemuMutex event_mtx; > > > > static QemuCond event_complete_cond; static int > > > > event_unhandled_count; @@ -906,6 +908,12 @@ static void > > > > check_old_packet_regular(void *opaque) void > > > > colo_notify_compares_event(void *opaque, int event, Error **errp) { > > > > CompareState *s; > > > > +qemu_mutex_lock(_compare_mutex); > > > > + > > > > +if (!colo_compare_active) { > > > > +qemu_mutex_unlock(_compare_mutex); > > > > +return; > > > > +} > > > > > > > > qemu_mutex_lock(_mtx); > > > > QTAILQ_FOREACH(s, _compares, next) { @@ -919,6 +927,7 @@ > > > > void colo_notify_compares_event(void *opaque, int event, Error **errp) > > > > } &
Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that colo-compare is active
On 2020/5/8 上午10:26, Zhang, Chen wrote: -Original Message- From: Lukas Straub Sent: Thursday, May 7, 2020 11:54 PM To: Zhang, Chen Cc: qemu-devel ; Li Zhijian ; Jason Wang ; Marc- André Lureau ; Paolo Bonzini Subject: Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that colo-compare is active On Thu, 7 May 2020 11:38:04 + "Zhang, Chen" wrote: -Original Message- From: Lukas Straub Sent: Monday, May 4, 2020 6:28 PM To: qemu-devel Cc: Zhang, Chen ; Li Zhijian ; Jason Wang ; Marc- André Lureau ; Paolo Bonzini Subject: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that colo- compare is active If the colo-compare object is removed before failover and a checkpoint happens, qemu crashes because it tries to lock the destroyed event_mtx in colo_notify_compares_event. Fix this by checking if everything is initialized by introducing a new variable colo_compare_active which is protected by a new mutex colo_compare_mutex. The new mutex also protects against concurrent access of the net_compares list and makes sure that colo_notify_compares_event isn't active while we destroy event_mtx and event_complete_cond. With this it also is again possible to use colo without colo-compare (periodic mode) and to use multiple colo-compare for multiple network interfaces. Hi Lukas, For this case I think we don't need to touch vl.c code, we can solve this issue from another perspective: How to remove colo-compare? User will use qemu-monitor or QMP command to disable an object, so we just need return operation failed When user try to remove colo-compare object while COLO is running. Yeah, but that still leaves the other problem that colo can't be used without colo-compare (qemu crashes then). Yes, the COLO-compare is necessary module in COLO original design. At most cases, user need it do dynamic sync. For rare cases, maybe we can add a new colo-compare parameter to bypass all the network workload. Hi, Chen In our application, we only need "periodical mode" because of the performance issue, and have internal patch now. Is it OK to send the internal patch for review? Thanks. Regards, Derek Thanks Zhang Chen Regards, Lukas Straub Thanks Zhang Chen Signed-off-by: Lukas Straub --- net/colo-compare.c | 35 +-- net/colo-compare.h | 1 + softmmu/vl.c | 2 ++ 3 files changed, 32 insertions(+), 6 deletions(-) diff --git a/net/colo-compare.c b/net/colo-compare.c index 56db3d3bfc..c7572d75e9 100644 --- a/net/colo-compare.c +++ b/net/colo-compare.c @@ -54,6 +54,8 @@ static NotifierList colo_compare_notifiers = #define REGULAR_PACKET_CHECK_MS 3000 #define DEFAULT_TIME_OUT_MS 3000 +static QemuMutex colo_compare_mutex; static bool +colo_compare_active; static QemuMutex event_mtx; static QemuCond event_complete_cond; static int event_unhandled_count; @@ -906,6 +908,12 @@ static void check_old_packet_regular(void *opaque) void colo_notify_compares_event(void *opaque, int event, Error **errp) { CompareState *s; +qemu_mutex_lock(_compare_mutex); + +if (!colo_compare_active) { +qemu_mutex_unlock(_compare_mutex); +return; +} qemu_mutex_lock(_mtx); QTAILQ_FOREACH(s, _compares, next) { @@ -919,6 +927,7 @@ void colo_notify_compares_event(void *opaque, int event, Error **errp) } qemu_mutex_unlock(_mtx); +qemu_mutex_unlock(_compare_mutex); } static void colo_compare_timer_init(CompareState *s) @@ -1274,7 +1283,14 @@ static void colo_compare_complete(UserCreatable *uc, Error **errp) s->vnet_hdr); } +qemu_mutex_lock(_compare_mutex); +if (!colo_compare_active) { +qemu_mutex_init(_mtx); +qemu_cond_init(_complete_cond); +colo_compare_active = true; +} QTAILQ_INSERT_TAIL(_compares, s, next); +qemu_mutex_unlock(_compare_mutex); s->out_sendco.s = s; s->out_sendco.chr = >chr_out; @@ -1290,9 +1306,6 @@ static void colo_compare_complete(UserCreatable *uc, Error **errp) g_queue_init(>conn_list); -qemu_mutex_init(_mtx); -qemu_cond_init(_complete_cond); - s->connection_track_table = g_hash_table_new_full(connection_key_hash, connection_key_equal, g_free, @@ -1384,12 +1397,19 @@ static void colo_compare_finalize(Object *obj) qemu_bh_delete(s->event_bh); +qemu_mutex_lock(_compare_mutex); QTAILQ_FOREACH(tmp, _compares, next) { if (tmp == s) { QTAILQ_REMOVE(_compares, s, next); break; } } +if (QTAILQ_EMPTY(_compares)) { +colo_compare_active = false; +qemu_mutex_destroy(_mtx); +qemu_cond_destroy(_complete_cond); +} +qemu_mutex_unlock(_compare_mutex); AioContext *ctx = iothread_get_aio_contex
RE: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that colo-compare is active
> -Original Message- > From: Lukas Straub > Sent: Thursday, May 7, 2020 11:54 PM > To: Zhang, Chen > Cc: qemu-devel ; Li Zhijian > ; Jason Wang ; Marc- > André Lureau ; Paolo Bonzini > > Subject: Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that > colo-compare is active > > On Thu, 7 May 2020 11:38:04 + > "Zhang, Chen" wrote: > > > > -Original Message- > > > From: Lukas Straub > > > Sent: Monday, May 4, 2020 6:28 PM > > > To: qemu-devel > > > Cc: Zhang, Chen ; Li Zhijian > > > ; Jason Wang ; Marc- > > > André Lureau ; Paolo Bonzini > > > > > > Subject: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that > > > colo- compare is active > > > > > > If the colo-compare object is removed before failover and a > > > checkpoint happens, qemu crashes because it tries to lock the > > > destroyed event_mtx in colo_notify_compares_event. > > > > > > Fix this by checking if everything is initialized by introducing a > > > new variable colo_compare_active which is protected by a new mutex > > > colo_compare_mutex. The new mutex also protects against concurrent > > > access of the net_compares list and makes sure that > > > colo_notify_compares_event isn't active while we destroy event_mtx > > > and event_complete_cond. > > > > > > With this it also is again possible to use colo without colo-compare > > > (periodic > > > mode) and to use multiple colo-compare for multiple network interfaces. > > > > > > > Hi Lukas, > > > > For this case I think we don't need to touch vl.c code, we can solve this > issue from another perspective: > > How to remove colo-compare? > > User will use qemu-monitor or QMP command to disable an object, so we > > just need return operation failed When user try to remove colo-compare > object while COLO is running. > > Yeah, but that still leaves the other problem that colo can't be used without > colo-compare (qemu crashes then). Yes, the COLO-compare is necessary module in COLO original design. At most cases, user need it do dynamic sync. For rare cases, maybe we can add a new colo-compare parameter to bypass all the network workload. Thanks Zhang Chen > > Regards, > Lukas Straub > > > Thanks > > Zhang Chen > > > > > Signed-off-by: Lukas Straub > > > --- > > > net/colo-compare.c | 35 +-- > > > net/colo-compare.h | 1 + > > > softmmu/vl.c | 2 ++ > > > 3 files changed, 32 insertions(+), 6 deletions(-) > > > > > > diff --git a/net/colo-compare.c b/net/colo-compare.c index > > > 56db3d3bfc..c7572d75e9 100644 > > > --- a/net/colo-compare.c > > > +++ b/net/colo-compare.c > > > @@ -54,6 +54,8 @@ static NotifierList colo_compare_notifiers = > > > #define REGULAR_PACKET_CHECK_MS 3000 #define > DEFAULT_TIME_OUT_MS > > > 3000 > > > > > > +static QemuMutex colo_compare_mutex; static bool > > > +colo_compare_active; > > > static QemuMutex event_mtx; > > > static QemuCond event_complete_cond; static int > > > event_unhandled_count; @@ -906,6 +908,12 @@ static void > > > check_old_packet_regular(void *opaque) void > > > colo_notify_compares_event(void *opaque, int event, Error **errp) { > > > CompareState *s; > > > +qemu_mutex_lock(_compare_mutex); > > > + > > > +if (!colo_compare_active) { > > > +qemu_mutex_unlock(_compare_mutex); > > > +return; > > > +} > > > > > > qemu_mutex_lock(_mtx); > > > QTAILQ_FOREACH(s, _compares, next) { @@ -919,6 +927,7 @@ > > > void colo_notify_compares_event(void *opaque, int event, Error **errp) > > > } > > > > > > qemu_mutex_unlock(_mtx); > > > +qemu_mutex_unlock(_compare_mutex); > > > } > > > > > > static void colo_compare_timer_init(CompareState *s) @@ -1274,7 > > > +1283,14 @@ static void colo_compare_complete(UserCreatable *uc, > Error **errp) > > > s->vnet_hdr); > > > } > > > > > > +qemu_mutex_lock(_compare_mutex); > > > +if (!colo_compare_active) { > > > +qemu_mutex_init(_mtx); > > > +qemu_cond_init(_complete_cond); > > > +colo_compare_active = true; > > > +} > > > QTAILQ_INSERT_TAIL(_compa
Re: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that colo-compare is active
On Thu, 7 May 2020 11:38:04 + "Zhang, Chen" wrote: > > -Original Message- > > From: Lukas Straub > > Sent: Monday, May 4, 2020 6:28 PM > > To: qemu-devel > > Cc: Zhang, Chen ; Li Zhijian > > ; Jason Wang ; Marc- > > André Lureau ; Paolo Bonzini > > > > Subject: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that colo- > > compare is active > > > > If the colo-compare object is removed before failover and a checkpoint > > happens, qemu crashes because it tries to lock the destroyed event_mtx in > > colo_notify_compares_event. > > > > Fix this by checking if everything is initialized by introducing a new > > variable > > colo_compare_active which is protected by a new mutex > > colo_compare_mutex. The new mutex also protects against concurrent > > access of the net_compares list and makes sure that > > colo_notify_compares_event isn't active while we destroy event_mtx and > > event_complete_cond. > > > > With this it also is again possible to use colo without colo-compare > > (periodic > > mode) and to use multiple colo-compare for multiple network interfaces. > > > > Hi Lukas, > > For this case I think we don't need to touch vl.c code, we can solve this > issue from another perspective: > How to remove colo-compare? > User will use qemu-monitor or QMP command to disable an object, so we just > need return operation failed > When user try to remove colo-compare object while COLO is running. Yeah, but that still leaves the other problem that colo can't be used without colo-compare (qemu crashes then). Regards, Lukas Straub > Thanks > Zhang Chen > > > Signed-off-by: Lukas Straub > > --- > > net/colo-compare.c | 35 +-- > > net/colo-compare.h | 1 + > > softmmu/vl.c | 2 ++ > > 3 files changed, 32 insertions(+), 6 deletions(-) > > > > diff --git a/net/colo-compare.c b/net/colo-compare.c index > > 56db3d3bfc..c7572d75e9 100644 > > --- a/net/colo-compare.c > > +++ b/net/colo-compare.c > > @@ -54,6 +54,8 @@ static NotifierList colo_compare_notifiers = #define > > REGULAR_PACKET_CHECK_MS 3000 #define DEFAULT_TIME_OUT_MS 3000 > > > > +static QemuMutex colo_compare_mutex; > > +static bool colo_compare_active; > > static QemuMutex event_mtx; > > static QemuCond event_complete_cond; > > static int event_unhandled_count; > > @@ -906,6 +908,12 @@ static void check_old_packet_regular(void *opaque) > > void colo_notify_compares_event(void *opaque, int event, Error **errp) { > > CompareState *s; > > +qemu_mutex_lock(_compare_mutex); > > + > > +if (!colo_compare_active) { > > +qemu_mutex_unlock(_compare_mutex); > > +return; > > +} > > > > qemu_mutex_lock(_mtx); > > QTAILQ_FOREACH(s, _compares, next) { @@ -919,6 +927,7 @@ void > > colo_notify_compares_event(void *opaque, int event, Error **errp) > > } > > > > qemu_mutex_unlock(_mtx); > > +qemu_mutex_unlock(_compare_mutex); > > } > > > > static void colo_compare_timer_init(CompareState *s) @@ -1274,7 +1283,14 > > @@ static void colo_compare_complete(UserCreatable *uc, Error **errp) > > s->vnet_hdr); > > } > > > > +qemu_mutex_lock(_compare_mutex); > > +if (!colo_compare_active) { > > +qemu_mutex_init(_mtx); > > +qemu_cond_init(_complete_cond); > > +colo_compare_active = true; > > +} > > QTAILQ_INSERT_TAIL(_compares, s, next); > > +qemu_mutex_unlock(_compare_mutex); > > > > s->out_sendco.s = s; > > s->out_sendco.chr = >chr_out; > > @@ -1290,9 +1306,6 @@ static void colo_compare_complete(UserCreatable > > *uc, Error **errp) > > > > g_queue_init(>conn_list); > > > > -qemu_mutex_init(_mtx); > > -qemu_cond_init(_complete_cond); > > - > > s->connection_track_table = > > g_hash_table_new_full(connection_key_hash, > >connection_key_equal, > >g_free, @@ -1384,12 > > +1397,19 @@ static void > > colo_compare_finalize(Object *obj) > > > > qemu_bh_delete(s->event_bh); > > > > +qemu_mutex_lock(_compare_mutex); > > QTAILQ_FOREACH(tmp, _compares, next) { > > if (tmp == s) { > > QTAILQ_REMOVE(_compares, s, next); >
RE: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that colo-compare is active
> -Original Message- > From: Lukas Straub > Sent: Monday, May 4, 2020 6:28 PM > To: qemu-devel > Cc: Zhang, Chen ; Li Zhijian > ; Jason Wang ; Marc- > André Lureau ; Paolo Bonzini > > Subject: [PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that colo- > compare is active > > If the colo-compare object is removed before failover and a checkpoint > happens, qemu crashes because it tries to lock the destroyed event_mtx in > colo_notify_compares_event. > > Fix this by checking if everything is initialized by introducing a new > variable > colo_compare_active which is protected by a new mutex > colo_compare_mutex. The new mutex also protects against concurrent > access of the net_compares list and makes sure that > colo_notify_compares_event isn't active while we destroy event_mtx and > event_complete_cond. > > With this it also is again possible to use colo without colo-compare (periodic > mode) and to use multiple colo-compare for multiple network interfaces. > Hi Lukas, For this case I think we don't need to touch vl.c code, we can solve this issue from another perspective: How to remove colo-compare? User will use qemu-monitor or QMP command to disable an object, so we just need return operation failed When user try to remove colo-compare object while COLO is running. Thanks Zhang Chen > Signed-off-by: Lukas Straub > --- > net/colo-compare.c | 35 +-- > net/colo-compare.h | 1 + > softmmu/vl.c | 2 ++ > 3 files changed, 32 insertions(+), 6 deletions(-) > > diff --git a/net/colo-compare.c b/net/colo-compare.c index > 56db3d3bfc..c7572d75e9 100644 > --- a/net/colo-compare.c > +++ b/net/colo-compare.c > @@ -54,6 +54,8 @@ static NotifierList colo_compare_notifiers = #define > REGULAR_PACKET_CHECK_MS 3000 #define DEFAULT_TIME_OUT_MS 3000 > > +static QemuMutex colo_compare_mutex; > +static bool colo_compare_active; > static QemuMutex event_mtx; > static QemuCond event_complete_cond; > static int event_unhandled_count; > @@ -906,6 +908,12 @@ static void check_old_packet_regular(void *opaque) > void colo_notify_compares_event(void *opaque, int event, Error **errp) { > CompareState *s; > +qemu_mutex_lock(_compare_mutex); > + > +if (!colo_compare_active) { > +qemu_mutex_unlock(_compare_mutex); > +return; > +} > > qemu_mutex_lock(_mtx); > QTAILQ_FOREACH(s, _compares, next) { @@ -919,6 +927,7 @@ void > colo_notify_compares_event(void *opaque, int event, Error **errp) > } > > qemu_mutex_unlock(_mtx); > +qemu_mutex_unlock(_compare_mutex); > } > > static void colo_compare_timer_init(CompareState *s) @@ -1274,7 +1283,14 > @@ static void colo_compare_complete(UserCreatable *uc, Error **errp) > s->vnet_hdr); > } > > +qemu_mutex_lock(_compare_mutex); > +if (!colo_compare_active) { > +qemu_mutex_init(_mtx); > +qemu_cond_init(_complete_cond); > +colo_compare_active = true; > +} > QTAILQ_INSERT_TAIL(_compares, s, next); > +qemu_mutex_unlock(_compare_mutex); > > s->out_sendco.s = s; > s->out_sendco.chr = >chr_out; > @@ -1290,9 +1306,6 @@ static void colo_compare_complete(UserCreatable > *uc, Error **errp) > > g_queue_init(>conn_list); > > -qemu_mutex_init(_mtx); > -qemu_cond_init(_complete_cond); > - > s->connection_track_table = > g_hash_table_new_full(connection_key_hash, >connection_key_equal, >g_free, @@ -1384,12 > +1397,19 @@ static void > colo_compare_finalize(Object *obj) > > qemu_bh_delete(s->event_bh); > > +qemu_mutex_lock(_compare_mutex); > QTAILQ_FOREACH(tmp, _compares, next) { > if (tmp == s) { > QTAILQ_REMOVE(_compares, s, next); > break; > } > } > +if (QTAILQ_EMPTY(_compares)) { > +colo_compare_active = false; > +qemu_mutex_destroy(_mtx); > +qemu_cond_destroy(_complete_cond); > +} > +qemu_mutex_unlock(_compare_mutex); > > AioContext *ctx = iothread_get_aio_context(s->iothread); > aio_context_acquire(ctx); > @@ -1413,15 +1433,18 @@ static void colo_compare_finalize(Object *obj) > object_unref(OBJECT(s->iothread)); > } > > -qemu_mutex_destroy(_mtx); > -qemu_cond_destroy(_complete_cond); > - > g_free(s->pri_indev); > g_free(s->sec_indev); > g_free(s->outdev); > g_free(s->notify_dev); > } > &
[PATCH v4 5/6] net/colo-compare.c, softmmu/vl.c: Check that colo-compare is active
If the colo-compare object is removed before failover and a checkpoint happens, qemu crashes because it tries to lock the destroyed event_mtx in colo_notify_compares_event. Fix this by checking if everything is initialized by introducing a new variable colo_compare_active which is protected by a new mutex colo_compare_mutex. The new mutex also protects against concurrent access of the net_compares list and makes sure that colo_notify_compares_event isn't active while we destroy event_mtx and event_complete_cond. With this it also is again possible to use colo without colo-compare (periodic mode) and to use multiple colo-compare for multiple network interfaces. Signed-off-by: Lukas Straub --- net/colo-compare.c | 35 +-- net/colo-compare.h | 1 + softmmu/vl.c | 2 ++ 3 files changed, 32 insertions(+), 6 deletions(-) diff --git a/net/colo-compare.c b/net/colo-compare.c index 56db3d3bfc..c7572d75e9 100644 --- a/net/colo-compare.c +++ b/net/colo-compare.c @@ -54,6 +54,8 @@ static NotifierList colo_compare_notifiers = #define REGULAR_PACKET_CHECK_MS 3000 #define DEFAULT_TIME_OUT_MS 3000 +static QemuMutex colo_compare_mutex; +static bool colo_compare_active; static QemuMutex event_mtx; static QemuCond event_complete_cond; static int event_unhandled_count; @@ -906,6 +908,12 @@ static void check_old_packet_regular(void *opaque) void colo_notify_compares_event(void *opaque, int event, Error **errp) { CompareState *s; +qemu_mutex_lock(_compare_mutex); + +if (!colo_compare_active) { +qemu_mutex_unlock(_compare_mutex); +return; +} qemu_mutex_lock(_mtx); QTAILQ_FOREACH(s, _compares, next) { @@ -919,6 +927,7 @@ void colo_notify_compares_event(void *opaque, int event, Error **errp) } qemu_mutex_unlock(_mtx); +qemu_mutex_unlock(_compare_mutex); } static void colo_compare_timer_init(CompareState *s) @@ -1274,7 +1283,14 @@ static void colo_compare_complete(UserCreatable *uc, Error **errp) s->vnet_hdr); } +qemu_mutex_lock(_compare_mutex); +if (!colo_compare_active) { +qemu_mutex_init(_mtx); +qemu_cond_init(_complete_cond); +colo_compare_active = true; +} QTAILQ_INSERT_TAIL(_compares, s, next); +qemu_mutex_unlock(_compare_mutex); s->out_sendco.s = s; s->out_sendco.chr = >chr_out; @@ -1290,9 +1306,6 @@ static void colo_compare_complete(UserCreatable *uc, Error **errp) g_queue_init(>conn_list); -qemu_mutex_init(_mtx); -qemu_cond_init(_complete_cond); - s->connection_track_table = g_hash_table_new_full(connection_key_hash, connection_key_equal, g_free, @@ -1384,12 +1397,19 @@ static void colo_compare_finalize(Object *obj) qemu_bh_delete(s->event_bh); +qemu_mutex_lock(_compare_mutex); QTAILQ_FOREACH(tmp, _compares, next) { if (tmp == s) { QTAILQ_REMOVE(_compares, s, next); break; } } +if (QTAILQ_EMPTY(_compares)) { +colo_compare_active = false; +qemu_mutex_destroy(_mtx); +qemu_cond_destroy(_complete_cond); +} +qemu_mutex_unlock(_compare_mutex); AioContext *ctx = iothread_get_aio_context(s->iothread); aio_context_acquire(ctx); @@ -1413,15 +1433,18 @@ static void colo_compare_finalize(Object *obj) object_unref(OBJECT(s->iothread)); } -qemu_mutex_destroy(_mtx); -qemu_cond_destroy(_complete_cond); - g_free(s->pri_indev); g_free(s->sec_indev); g_free(s->outdev); g_free(s->notify_dev); } +void colo_compare_init_globals(void) +{ +colo_compare_active = false; +qemu_mutex_init(_compare_mutex); +} + static const TypeInfo colo_compare_info = { .name = TYPE_COLO_COMPARE, .parent = TYPE_OBJECT, diff --git a/net/colo-compare.h b/net/colo-compare.h index 22ddd512e2..eb483ac586 100644 --- a/net/colo-compare.h +++ b/net/colo-compare.h @@ -17,6 +17,7 @@ #ifndef QEMU_COLO_COMPARE_H #define QEMU_COLO_COMPARE_H +void colo_compare_init_globals(void); void colo_notify_compares_event(void *opaque, int event, Error **errp); void colo_compare_register_notifier(Notifier *notify); void colo_compare_unregister_notifier(Notifier *notify); diff --git a/softmmu/vl.c b/softmmu/vl.c index 32c0047889..a913ed5469 100644 --- a/softmmu/vl.c +++ b/softmmu/vl.c @@ -112,6 +112,7 @@ #include "qapi/qmp/qerror.h" #include "sysemu/iothread.h" #include "qemu/guest-random.h" +#include "net/colo-compare.h" #define MAX_VIRTIO_CONSOLES 1 @@ -2906,6 +2907,7 @@ void qemu_init(int argc, char **argv, char **envp) precopy_infrastructure_init(); postcopy_infrastructure_init(); monitor_init_globals(); +colo_compare_init_globals(); if (qcrypto_init() < 0) { error_reportf_err(err, "cannot initialize crypto: "); --