From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:45154) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1d9Hv1-000833-Fi for qemu-devel@nongnu.org; Fri, 12 May 2017 17:16:12 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1d9Hux-00042j-DQ for qemu-devel@nongnu.org; Fri, 12 May 2017 17:16:11 -0400 Received: from mx1.redhat.com ([209.132.183.28]:56784) by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1d9Hux-00042R-4C for qemu-devel@nongnu.org; Fri, 12 May 2017 17:16:07 -0400 Received: from smtp.corp.redhat.com (int-mx01.intmail.prod.int.phx2.redhat.com [10.5.11.11]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mx1.redhat.com (Postfix) with ESMTPS id D0F0881138 for ; Fri, 12 May 2017 21:16:05 +0000 (UTC) Reply-To: vyasevic@redhat.com References: <20170512192428.GP2069@work-vm> From: Vlad Yasevich Message-ID: <5ee20153-8fc1-d4fc-4985-597263f93f6d@redhat.com> Date: Fri, 12 May 2017 17:16:00 -0400 MIME-Version: 1.0 In-Reply-To: <20170512192428.GP2069@work-vm> Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [PATCH V2] migration: expose qemu_announce_self() via qmp List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: "Dr. David Alan Gilbert" Cc: Germano Veit Michel , qemu-devel@nongnu.org, Juan Jose Quintela Carreira , Amit Shah On 05/12/2017 03:24 PM, Dr. David Alan Gilbert wrote: > * Vlad Yasevich (vyasevic@redhat.com) wrote: >> On 02/20/2017 07:16 PM, Germano Veit Michel wrote: >>> qemu_announce_self() is triggered by qemu at the end of migrations >>> to update the network regarding the path to the guest l2addr. >>> >>> however it is also useful when there is a network change such as >>> an active bond slave swap. Essentially, it's the same as a migration >>> from a network perspective - the guest moves to a different point >>> in the network topology. >>> >>> this exposes the function via qmp. >>> >>> Signed-off-by: Germano Veit Michel >>> --- >>> include/migration/vmstate.h | 5 +++++ >>> migration/savevm.c | 30 +++++++++++++++++++----------- >>> qapi-schema.json | 18 ++++++++++++++++++ >>> 3 files changed, 42 insertions(+), 11 deletions(-) >>> >>> diff --git a/include/migration/vmstate.h b/include/migration/vmstate.h >>> index 63e7b02..a08715c 100644 >>> --- a/include/migration/vmstate.h >>> +++ b/include/migration/vmstate.h >>> @@ -1042,6 +1042,11 @@ int64_t self_announce_delay(int round) >>> return 50 + (SELF_ANNOUNCE_ROUNDS - round - 1) * 100; >>> } >>> >>> +struct AnnounceRound { >>> + QEMUTimer *timer; >>> + int count; >>> +}; >>> + >>> void dump_vmstate_json_to_file(FILE *out_fp); >>> >>> #endif >>> diff --git a/migration/savevm.c b/migration/savevm.c >>> index 5ecd264..44e196b 100644 >>> --- a/migration/savevm.c >>> +++ b/migration/savevm.c >>> @@ -118,29 +118,37 @@ static void qemu_announce_self_iter(NICState >>> *nic, void *opaque) >>> qemu_send_packet_raw(qemu_get_queue(nic), buf, len); >>> } >>> >>> - >>> static void qemu_announce_self_once(void *opaque) >>> { >>> - static int count = SELF_ANNOUNCE_ROUNDS; >>> - QEMUTimer *timer = *(QEMUTimer **)opaque; >>> + struct AnnounceRound *round = opaque; >>> >>> qemu_foreach_nic(qemu_announce_self_iter, NULL); >>> >>> - if (--count) { >>> + round->count--; >>> + if (round->count) { >>> /* delay 50ms, 150ms, 250ms, ... */ >>> - timer_mod(timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + >>> - self_announce_delay(count)); >>> + timer_mod(round->timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + >>> + self_announce_delay(round->count)); >>> } else { >>> - timer_del(timer); >>> - timer_free(timer); >>> + timer_del(round->timer); >>> + timer_free(round->timer); >>> + g_free(round); >>> } >>> } >>> >>> void qemu_announce_self(void) >>> { >>> - static QEMUTimer *timer; >>> - timer = timer_new_ms(QEMU_CLOCK_REALTIME, qemu_announce_self_once, &timer); >>> - qemu_announce_self_once(&timer); >>> + struct AnnounceRound *round = g_malloc(sizeof(struct AnnounceRound)); >>> + if (!round) >>> + return; >>> + round->count = SELF_ANNOUNCE_ROUNDS; >>> + round->timer = timer_new_ms(QEMU_CLOCK_REALTIME, >>> qemu_announce_self_once, round); >>> + qemu_announce_self_once(round); >>> +} >> >> So, I've been looking and this code and have been playing with it and with David's >> patches and my patches to include virtio self announcements as well. What I've discovered >> is what I think is a possible packet amplification issue here. >> >> This creates a new timer every time we do do a announce_self. With just migration, >> this is not an issue since you only migrate once at a time, so there is only 1 timer. >> With exposing this as an API, a user can potentially call it in a tight loop >> and now you have a ton of timers being created. Add in David's patches allowing timeouts >> and retries to be configurable, and you may now have a ton of long lived timers. >> Add in the patches I am working on to let virtio do self announcements too (to really fix >> bonding issues), and now you add in a possibility of a lot of packets being sent for >> each timeout (RARP, GARP, NA, IGMPv4 Reports, IGMPv6 Reports [even worse if MLD1 is used]). >> >> As you can see, this can get rather ugly... >> >> I think we need timer user here. Migration and QMP being two to begin with. Each >> one would get a single timer to play with. If a given user already has a timer running, >> we could return an error or just not do anything. > > If you did have specific timers, then you could add to/reset the counts > rather than doing nothing. That way it's less racy; if you issue the > command just as you reconfigure your network, there's no chance the > command would fail, you will send the packets out. Yes. That's another possible way to handle this. -vlad > > Dave > >> -vlad >> >>> + >>> +void qmp_announce_self(Error **errp) >>> +{ >>> + qemu_announce_self(); >>> } >>> >>> /***********************************************************/ >>> diff --git a/qapi-schema.json b/qapi-schema.json >>> index baa0d26..0d9bffd 100644 >>> --- a/qapi-schema.json >>> +++ b/qapi-schema.json >>> @@ -6080,3 +6080,21 @@ >>> # >>> ## >>> { 'command': 'query-hotpluggable-cpus', 'returns': ['HotpluggableCPU'] } >>> + >>> +## >>> +# @announce-self: >>> +# >>> +# Trigger generation of broadcast RARP frames to update network switches. >>> +# This can be useful when network bonds fail-over the active slave. >>> +# >>> +# Arguments: None. >>> +# >>> +# Example: >>> +# >>> +# -> { "execute": "announce-self" } >>> +# <- { "return": {} } >>> +# >>> +# Since: 2.9 >>> +## >>> +{ 'command': 'announce-self' } >>> + >>> >> > -- > Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK >