From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([209.51.188.92]:35272) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1hHASZ-0007AH-6r for qemu-devel@nongnu.org; Thu, 18 Apr 2019 13:04:28 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1hHAPi-0006AL-Jh for qemu-devel@nongnu.org; Thu, 18 Apr 2019 13:01:33 -0400 Received: from mx1.redhat.com ([209.132.183.28]:13536) by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1hHAPc-00066g-Uq for qemu-devel@nongnu.org; Thu, 18 Apr 2019 13:01:27 -0400 Date: Thu, 18 Apr 2019 18:01:19 +0100 From: "Dr. David Alan Gilbert" Message-ID: <20190418170118.GK2984@work-vm> References: <259551555324396@vla1-1374b6242101.qloud-c.yandex.net> <20190415104725.GI5718@redhat.com> <20190415111511.GG2852@work-vm> <20190415111911.GK5718@redhat.com> <20190415113045.GI2852@work-vm> <612681555412465@iva6-8a76e93b6298.qloud-c.yandex.net> <20190418141956.GF2984@work-vm> <1285061555601789@iva6-8a76e93b6298.qloud-c.yandex.net> <20190418160300.GI2984@work-vm> <1201831555604730@vla1-1374b6242101.qloud-c.yandex.net> MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Disposition: inline In-Reply-To: <1201831555604730@vla1-1374b6242101.qloud-c.yandex.net> Content-Transfer-Encoding: quoted-printable Subject: Re: [Qemu-devel] [PATCH] migration: Fix handling fd protocol List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Yury Kotov Cc: Daniel =?iso-8859-1?Q?P=2E_Berrang=E9?= , "qemu-devel@nongnu.org" , "yc-core@yandex-team.ru" , Juan Quintela * Yury Kotov (yury-kotov@yandex-team.ru) wrote: > 18.04.2019, 19:03, "Dr. David Alan Gilbert" : > > * Yury Kotov (yury-kotov@yandex-team.ru) wrote: > >> =A018.04.2019, 17:20, "Dr. David Alan Gilbert" = : > >> =A0> * Yury Kotov (yury-kotov@yandex-team.ru) wrote: > >> =A0>> =A015.04.2019, 14:30, "Dr. David Alan Gilbert" : > >> =A0>> =A0> * Daniel P. Berrang=E9 (berrange@redhat.com) wrote: > >> =A0>> =A0>> =A0On Mon, Apr 15, 2019 at 12:15:12PM +0100, Dr. David A= lan Gilbert wrote: > >> =A0>> =A0>> =A0> * Daniel P. Berrang=E9 (berrange@redhat.com) wrote: > >> =A0>> =A0>> =A0> > On Mon, Apr 15, 2019 at 01:33:21PM +0300, Yury Ko= tov wrote: > >> =A0>> =A0>> =A0> > > 15.04.2019, 13:25, "Daniel P. Berrang=E9" : > >> =A0>> =A0>> =A0> > > > On Mon, Apr 15, 2019 at 01:17:06PM +0300, Yur= y Kotov wrote: > >> =A0>> =A0>> =A0> > > >> =A015.04.2019, 13:11, "Daniel P. Berrang=E9"= : > >> =A0>> =A0>> =A0> > > >> =A0> On Mon, Apr 15, 2019 at 12:50:08PM +030= 0, Yury Kotov wrote: > >> =A0>> =A0>> =A0> > > >> =A0>> =A0Hi, > >> =A0>> =A0>> =A0> > > >> =A0>> > >> =A0>> =A0>> =A0> > > >> =A0>> =A0Just to clarify. I see two possible= solutions: > >> =A0>> =A0>> =A0> > > >> =A0>> > >> =A0>> =A0>> =A0> > > >> =A0>> =A01) Since the migration code doesn't= receive fd, it isn't responsible for > >> =A0>> =A0>> =A0> > > >> =A0>> =A0closing it. So, it may be better to= use migrate_fd_param for both > >> =A0>> =A0>> =A0> > > >> =A0>> =A0incoming/outgoing and add dupping f= or migrate_fd_param. Thus, clients must > >> =A0>> =A0>> =A0> > > >> =A0>> =A0close the fd themselves. But existi= ng clients will have a leak. > >> =A0>> =A0>> =A0> > > >> =A0> > >> =A0>> =A0>> =A0> > > >> =A0> We can't break existing clients in this= way as they are correctly > >> =A0>> =A0>> =A0> > > >> =A0> using the monitor with its current sema= ntics. > >> =A0>> =A0>> =A0> > > >> =A0> > >> =A0>> =A0>> =A0> > > >> =A0>> =A02) If we don't duplicate fd, then a= t least we should remove fd from > >> =A0>> =A0>> =A0> > > >> =A0>> =A0the corresponding list. Therefore, = the solution is to fix qemu_close to find > >> =A0>> =A0>> =A0> > > >> =A0>> =A0the list and remove fd from it. But= qemu_close is currently consistent with > >> =A0>> =A0>> =A0> > > >> =A0>> =A0qemu_open (which opens/dups fd), so= adding additional logic might not be > >> =A0>> =A0>> =A0> > > >> =A0>> =A0a very good idea. > >> =A0>> =A0>> =A0> > > >> =A0> > >> =A0>> =A0>> =A0> > > >> =A0> qemu_close is not appropriate place to = deal with something speciifc > >> =A0>> =A0>> =A0> > > >> =A0> to the montor. > >> =A0>> =A0>> =A0> > > >> =A0> > >> =A0>> =A0>> =A0> > > >> =A0>> =A0I don't see any other solution, but= I might miss something. > >> =A0>> =A0>> =A0> > > >> =A0>> =A0What do you think? > >> =A0>> =A0>> =A0> > > >> =A0> > >> =A0>> =A0>> =A0> > > >> =A0> All callers of monitor_get_fd() will cl= ose() the FD they get back. > >> =A0>> =A0>> =A0> > > >> =A0> Thus monitor_get_fd() should remove it = from the list when it returns > >> =A0>> =A0>> =A0> > > >> =A0> it, and we should add API docs to monit= or_get_fd() to explain this. > >> =A0>> =A0>> =A0> > > >> =A0> > >> =A0>> =A0>> =A0> > > >> =A0Ok, it sounds reasonable. But monitor_get= _fd is only about outgoing migration. > >> =A0>> =A0>> =A0> > > >> =A0But what about the incoming migration? It= doesn't use monitor_get_fd but just > >> =A0>> =A0>> =A0> > > >> =A0converts input string to int and use it a= s fd. > >> =A0>> =A0>> =A0> > > > > >> =A0>> =A0>> =A0> > > > The incoming migration expects the FD to be p= assed into QEMU by the mgmt > >> =A0>> =A0>> =A0> > > > app when it is exec'ing the QEMU binary. It d= oesn't interact with the > >> =A0>> =A0>> =A0> > > > monitor at all AFAIR. > >> =A0>> =A0>> =A0> > > > > >> =A0>> =A0>> =A0> > > > >> =A0>> =A0>> =A0> > > Oh, sorry. This use case is not obvious. We use= d add-fd to pass fd for > >> =A0>> =A0>> =A0> > > migrate-incoming and such way has described pro= blems. > >> =A0>> =A0>> =A0> > > >> =A0>> =A0>> =A0> > That's a bug in your usage of QEMU IMHO, as the i= ncoming code is not > >> =A0>> =A0>> =A0> > designed to use add-fd. > >> =A0>> =A0>> =A0> > >> =A0>> =A0>> =A0> Hmm, that's true - although: > >> =A0>> =A0>> =A0> a) It's very non-obvious > >> =A0>> =A0>> =A0> b) Unfortunate, since it would go well with -incomi= ng defer > >> =A0>> =A0>> > >> =A0>> =A0>> =A0Yeah I think this is a screw up on QMEU's part when i= ntroducing 'defer'. > >> =A0>> =A0>> > >> =A0>> =A0>> =A0We should have mandated use of 'add-fd' when using 'd= efer', since FD > >> =A0>> =A0>> =A0inheritance-over-execve() should only be used for com= mand line args, > >> =A0>> =A0>> =A0not monitor commands. > >> =A0>> =A0>> > >> =A0>> =A0>> =A0Not sure how to best fix this is QEMU though without = breaking back > >> =A0>> =A0>> =A0compat for apps using 'defer' already. > >> =A0>> =A0> > >> =A0>> =A0> We could add mon-fd: transports that has the same behavio= ur as now for > >> =A0>> =A0> outgoing, and for incoming uses the add-fd stash. > >> =A0>> =A0> > >> =A0>> > >> =A0>> =A0Oh, I'm sorry again. I think my suggestion about monitor_fd= _param wasn't > >> =A0>> =A0relevant to this issue. If migrate-incoming + "fd:" + add-f= d is an invalid use > >> =A0>> =A0case, should we disallow this? > >> =A0>> =A0I may add a check to fd_start_incoming_migration if fd is i= n mon fds list. > >> =A0>> =A0But I'm afraid there are users like me who are already usin= g this wrong use case. > >> =A0>> =A0Because currently nothing in QEMU's docs disallow this. > >> =A0>> > >> =A0>> =A0So which solution is better in your opinion? > >> =A0>> =A01) Disallow fd's from mon fds list in fd_start_incoming_mig= ration > >> =A0> > >> =A0> I'm surprised anything could be doing that - how would a user k= now what > >> =A0> the correct fd index was? > >> =A0> > >> > >> =A0Hmm, add-fd returns correct fd value. Maybe I din't catch you que= stion... > > > > I don't understand, where does it return it? > > >=20 > From misc.json: > # Example: > # > # -> { "execute": "add-fd", "arguments": { "fdset-id": 1 } } > # <- { "return": { "fdset-id": 1, "fd": 3 } } > # >=20 > "fd": 3 is a valid fd for migrate-incoming(uri =3D "fd:3") Ah OK. > >> =A0>> =A02) Allow these fds, but dup them or close them correctly > >> =A0> > >> =A0> I think I'd leave the current (confusing) fd: as it is, maybe p= ut a note > >> =A0> in the manual. > >> =A0> > >> > >> =A0So, using fd from fdset will be an undefined behavior, right? > > > > For incoming, yes. > > > >> =A0>> =A0And how to migrate-incoming defer through fd correctly? > >> =A0>> =A01) Add "mon-fd:" protocol to work with fds passed by "add-f= d/remove-fd" commands > >> =A0>> =A0as suggested by Dave > >> =A0> > >> =A0> That's my preference; it's explicitly named and consistent, and= it > >> =A0> doesn't touch the existing fd code. > >> =A0> > >> > >> =A0Ok, but please tell me what you think of my suggestion (2) about = using fd added > >> =A0by the "getfd" command for incoming migration. It doesn't require= s introducing > >> =A0new protocol and will be consistent with outgoing migration throu= gh fd. > > > > I worry how qemu knows whether the command means it comes from the ge= tfd > > command or is actually a normal fd like now? > > Can you give an example. > > >=20 > getfd manages naming fds list. > # -> { "execute": "getfd", "arguments": { "fdname": "fd1" } } > So, for migrate (not incoming) is now valid migrate(uri=3D"fd:fd1") >=20 > I want the same for migrate-incoming. If fdname is parseable int, then = it is > an old format. Otherwise - it is a name of fd added by addfd. >=20 > There is a function "monitor_fd_param" which do exactly what I mean: > int monitor_fd_param(Monitor *mon, const char *fdname, Error **errp) { > ... local vars ... > if (!qemu_isdigit(fdname[0]) && mon) { > fd =3D monitor_get_fd(mon, fdname, &local_err); > } else { > fd =3D qemu_parse_fd(fdname); > } > ... report err to errp ... > } OK, if we're already using monitor_fd_param everywhere then I think we're already down the rat-hole of guessing whether we're an add-fd or fd by whether it's an integer, and I agree with you that we should just fix incoming to use that. Now, that means I guess we need to modify monitor_fd_param to tell us which type of fd it got, so we know whether to close it later? Dave P.S. I'm out from tomorrow for a weekish. > >> =A0> > >> =A0>> =A02) My suggestion about monitor_fd_param and make "fd:" for > >> =A0>> =A0migrate/migrate-incoming consistent. So user will be able t= o use > >> =A0>> =A0getfd + migrate-incoming > >> =A0>> =A03) Both of them or something else > >> =A0>> > >> >=20 > Regards, > Yury -- Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-2.4 required=3.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, MAILING_LIST_MULTI,SPF_PASS,T_HK_NAME_DR,URIBL_BLOCKED,USER_AGENT_MUTT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 9E6B4C10F0E for ; Thu, 18 Apr 2019 17:05:52 +0000 (UTC) Received: from lists.gnu.org (lists.gnu.org [209.51.188.17]) (using TLSv1 with cipher AES256-SHA (256/256 bits)) (No client certificate requested) by mail.kernel.org (Postfix) with ESMTPS id 5F48F2171F for ; Thu, 18 Apr 2019 17:05:52 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org 5F48F2171F Authentication-Results: mail.kernel.org; dmarc=fail (p=none dis=none) header.from=redhat.com Authentication-Results: mail.kernel.org; spf=pass smtp.mailfrom=qemu-devel-bounces+qemu-devel=archiver.kernel.org@nongnu.org Received: from localhost ([127.0.0.1]:44511 helo=lists.gnu.org) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1hHATv-0007nZ-ET for qemu-devel@archiver.kernel.org; Thu, 18 Apr 2019 13:05:51 -0400 Received: from eggs.gnu.org ([209.51.188.92]:35272) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1hHASZ-0007AH-6r for qemu-devel@nongnu.org; Thu, 18 Apr 2019 13:04:28 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1hHAPi-0006AL-Jh for qemu-devel@nongnu.org; Thu, 18 Apr 2019 13:01:33 -0400 Received: from mx1.redhat.com ([209.132.183.28]:13536) by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1hHAPc-00066g-Uq for qemu-devel@nongnu.org; Thu, 18 Apr 2019 13:01:27 -0400 Received: from smtp.corp.redhat.com (int-mx02.intmail.prod.int.phx2.redhat.com [10.5.11.12]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mx1.redhat.com (Postfix) with ESMTPS id 635D588ABE; Thu, 18 Apr 2019 17:01:23 +0000 (UTC) Received: from work-vm (ovpn-117-54.ams2.redhat.com [10.36.117.54]) by smtp.corp.redhat.com (Postfix) with ESMTPS id C82AD60BE5; Thu, 18 Apr 2019 17:01:21 +0000 (UTC) Date: Thu, 18 Apr 2019 18:01:19 +0100 From: "Dr. David Alan Gilbert" To: Yury Kotov Message-ID: <20190418170118.GK2984@work-vm> References: <259551555324396@vla1-1374b6242101.qloud-c.yandex.net> <20190415104725.GI5718@redhat.com> <20190415111511.GG2852@work-vm> <20190415111911.GK5718@redhat.com> <20190415113045.GI2852@work-vm> <612681555412465@iva6-8a76e93b6298.qloud-c.yandex.net> <20190418141956.GF2984@work-vm> <1285061555601789@iva6-8a76e93b6298.qloud-c.yandex.net> <20190418160300.GI2984@work-vm> <1201831555604730@vla1-1374b6242101.qloud-c.yandex.net> MIME-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Disposition: inline In-Reply-To: <1201831555604730@vla1-1374b6242101.qloud-c.yandex.net> User-Agent: Mutt/1.11.4 (2019-03-13) X-Scanned-By: MIMEDefang 2.79 on 10.5.11.12 X-Greylist: Sender IP whitelisted, not delayed by milter-greylist-4.5.16 (mx1.redhat.com [10.5.110.26]); Thu, 18 Apr 2019 17:01:23 +0000 (UTC) Content-Transfer-Encoding: quoted-printable X-detected-operating-system: by eggs.gnu.org: GNU/Linux 2.2.x-3.x [generic] X-Received-From: 209.132.183.28 Subject: Re: [Qemu-devel] [PATCH] migration: Fix handling fd protocol X-BeenThere: qemu-devel@nongnu.org X-Mailman-Version: 2.1.21 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: "qemu-devel@nongnu.org" , "yc-core@yandex-team.ru" , Juan Quintela Errors-To: qemu-devel-bounces+qemu-devel=archiver.kernel.org@nongnu.org Sender: "Qemu-devel" Message-ID: <20190418170119.pWM8Xy4NUNa1AilHLMfVq_R_6GNHhHdEw-bbgnHVca8@z> * Yury Kotov (yury-kotov@yandex-team.ru) wrote: > 18.04.2019, 19:03, "Dr. David Alan Gilbert" : > > * Yury Kotov (yury-kotov@yandex-team.ru) wrote: > >> =A018.04.2019, 17:20, "Dr. David Alan Gilbert" = : > >> =A0> * Yury Kotov (yury-kotov@yandex-team.ru) wrote: > >> =A0>> =A015.04.2019, 14:30, "Dr. David Alan Gilbert" : > >> =A0>> =A0> * Daniel P. Berrang=E9 (berrange@redhat.com) wrote: > >> =A0>> =A0>> =A0On Mon, Apr 15, 2019 at 12:15:12PM +0100, Dr. David A= lan Gilbert wrote: > >> =A0>> =A0>> =A0> * Daniel P. Berrang=E9 (berrange@redhat.com) wrote: > >> =A0>> =A0>> =A0> > On Mon, Apr 15, 2019 at 01:33:21PM +0300, Yury Ko= tov wrote: > >> =A0>> =A0>> =A0> > > 15.04.2019, 13:25, "Daniel P. Berrang=E9" : > >> =A0>> =A0>> =A0> > > > On Mon, Apr 15, 2019 at 01:17:06PM +0300, Yur= y Kotov wrote: > >> =A0>> =A0>> =A0> > > >> =A015.04.2019, 13:11, "Daniel P. Berrang=E9"= : > >> =A0>> =A0>> =A0> > > >> =A0> On Mon, Apr 15, 2019 at 12:50:08PM +030= 0, Yury Kotov wrote: > >> =A0>> =A0>> =A0> > > >> =A0>> =A0Hi, > >> =A0>> =A0>> =A0> > > >> =A0>> > >> =A0>> =A0>> =A0> > > >> =A0>> =A0Just to clarify. I see two possible= solutions: > >> =A0>> =A0>> =A0> > > >> =A0>> > >> =A0>> =A0>> =A0> > > >> =A0>> =A01) Since the migration code doesn't= receive fd, it isn't responsible for > >> =A0>> =A0>> =A0> > > >> =A0>> =A0closing it. So, it may be better to= use migrate_fd_param for both > >> =A0>> =A0>> =A0> > > >> =A0>> =A0incoming/outgoing and add dupping f= or migrate_fd_param. Thus, clients must > >> =A0>> =A0>> =A0> > > >> =A0>> =A0close the fd themselves. But existi= ng clients will have a leak. > >> =A0>> =A0>> =A0> > > >> =A0> > >> =A0>> =A0>> =A0> > > >> =A0> We can't break existing clients in this= way as they are correctly > >> =A0>> =A0>> =A0> > > >> =A0> using the monitor with its current sema= ntics. > >> =A0>> =A0>> =A0> > > >> =A0> > >> =A0>> =A0>> =A0> > > >> =A0>> =A02) If we don't duplicate fd, then a= t least we should remove fd from > >> =A0>> =A0>> =A0> > > >> =A0>> =A0the corresponding list. Therefore, = the solution is to fix qemu_close to find > >> =A0>> =A0>> =A0> > > >> =A0>> =A0the list and remove fd from it. But= qemu_close is currently consistent with > >> =A0>> =A0>> =A0> > > >> =A0>> =A0qemu_open (which opens/dups fd), so= adding additional logic might not be > >> =A0>> =A0>> =A0> > > >> =A0>> =A0a very good idea. > >> =A0>> =A0>> =A0> > > >> =A0> > >> =A0>> =A0>> =A0> > > >> =A0> qemu_close is not appropriate place to = deal with something speciifc > >> =A0>> =A0>> =A0> > > >> =A0> to the montor. > >> =A0>> =A0>> =A0> > > >> =A0> > >> =A0>> =A0>> =A0> > > >> =A0>> =A0I don't see any other solution, but= I might miss something. > >> =A0>> =A0>> =A0> > > >> =A0>> =A0What do you think? > >> =A0>> =A0>> =A0> > > >> =A0> > >> =A0>> =A0>> =A0> > > >> =A0> All callers of monitor_get_fd() will cl= ose() the FD they get back. > >> =A0>> =A0>> =A0> > > >> =A0> Thus monitor_get_fd() should remove it = from the list when it returns > >> =A0>> =A0>> =A0> > > >> =A0> it, and we should add API docs to monit= or_get_fd() to explain this. > >> =A0>> =A0>> =A0> > > >> =A0> > >> =A0>> =A0>> =A0> > > >> =A0Ok, it sounds reasonable. But monitor_get= _fd is only about outgoing migration. > >> =A0>> =A0>> =A0> > > >> =A0But what about the incoming migration? It= doesn't use monitor_get_fd but just > >> =A0>> =A0>> =A0> > > >> =A0converts input string to int and use it a= s fd. > >> =A0>> =A0>> =A0> > > > > >> =A0>> =A0>> =A0> > > > The incoming migration expects the FD to be p= assed into QEMU by the mgmt > >> =A0>> =A0>> =A0> > > > app when it is exec'ing the QEMU binary. It d= oesn't interact with the > >> =A0>> =A0>> =A0> > > > monitor at all AFAIR. > >> =A0>> =A0>> =A0> > > > > >> =A0>> =A0>> =A0> > > > >> =A0>> =A0>> =A0> > > Oh, sorry. This use case is not obvious. We use= d add-fd to pass fd for > >> =A0>> =A0>> =A0> > > migrate-incoming and such way has described pro= blems. > >> =A0>> =A0>> =A0> > > >> =A0>> =A0>> =A0> > That's a bug in your usage of QEMU IMHO, as the i= ncoming code is not > >> =A0>> =A0>> =A0> > designed to use add-fd. > >> =A0>> =A0>> =A0> > >> =A0>> =A0>> =A0> Hmm, that's true - although: > >> =A0>> =A0>> =A0> a) It's very non-obvious > >> =A0>> =A0>> =A0> b) Unfortunate, since it would go well with -incomi= ng defer > >> =A0>> =A0>> > >> =A0>> =A0>> =A0Yeah I think this is a screw up on QMEU's part when i= ntroducing 'defer'. > >> =A0>> =A0>> > >> =A0>> =A0>> =A0We should have mandated use of 'add-fd' when using 'd= efer', since FD > >> =A0>> =A0>> =A0inheritance-over-execve() should only be used for com= mand line args, > >> =A0>> =A0>> =A0not monitor commands. > >> =A0>> =A0>> > >> =A0>> =A0>> =A0Not sure how to best fix this is QEMU though without = breaking back > >> =A0>> =A0>> =A0compat for apps using 'defer' already. > >> =A0>> =A0> > >> =A0>> =A0> We could add mon-fd: transports that has the same behavio= ur as now for > >> =A0>> =A0> outgoing, and for incoming uses the add-fd stash. > >> =A0>> =A0> > >> =A0>> > >> =A0>> =A0Oh, I'm sorry again. I think my suggestion about monitor_fd= _param wasn't > >> =A0>> =A0relevant to this issue. If migrate-incoming + "fd:" + add-f= d is an invalid use > >> =A0>> =A0case, should we disallow this? > >> =A0>> =A0I may add a check to fd_start_incoming_migration if fd is i= n mon fds list. > >> =A0>> =A0But I'm afraid there are users like me who are already usin= g this wrong use case. > >> =A0>> =A0Because currently nothing in QEMU's docs disallow this. > >> =A0>> > >> =A0>> =A0So which solution is better in your opinion? > >> =A0>> =A01) Disallow fd's from mon fds list in fd_start_incoming_mig= ration > >> =A0> > >> =A0> I'm surprised anything could be doing that - how would a user k= now what > >> =A0> the correct fd index was? > >> =A0> > >> > >> =A0Hmm, add-fd returns correct fd value. Maybe I din't catch you que= stion... > > > > I don't understand, where does it return it? > > >=20 > From misc.json: > # Example: > # > # -> { "execute": "add-fd", "arguments": { "fdset-id": 1 } } > # <- { "return": { "fdset-id": 1, "fd": 3 } } > # >=20 > "fd": 3 is a valid fd for migrate-incoming(uri =3D "fd:3") Ah OK. > >> =A0>> =A02) Allow these fds, but dup them or close them correctly > >> =A0> > >> =A0> I think I'd leave the current (confusing) fd: as it is, maybe p= ut a note > >> =A0> in the manual. > >> =A0> > >> > >> =A0So, using fd from fdset will be an undefined behavior, right? > > > > For incoming, yes. > > > >> =A0>> =A0And how to migrate-incoming defer through fd correctly? > >> =A0>> =A01) Add "mon-fd:" protocol to work with fds passed by "add-f= d/remove-fd" commands > >> =A0>> =A0as suggested by Dave > >> =A0> > >> =A0> That's my preference; it's explicitly named and consistent, and= it > >> =A0> doesn't touch the existing fd code. > >> =A0> > >> > >> =A0Ok, but please tell me what you think of my suggestion (2) about = using fd added > >> =A0by the "getfd" command for incoming migration. It doesn't require= s introducing > >> =A0new protocol and will be consistent with outgoing migration throu= gh fd. > > > > I worry how qemu knows whether the command means it comes from the ge= tfd > > command or is actually a normal fd like now? > > Can you give an example. > > >=20 > getfd manages naming fds list. > # -> { "execute": "getfd", "arguments": { "fdname": "fd1" } } > So, for migrate (not incoming) is now valid migrate(uri=3D"fd:fd1") >=20 > I want the same for migrate-incoming. If fdname is parseable int, then = it is > an old format. Otherwise - it is a name of fd added by addfd. >=20 > There is a function "monitor_fd_param" which do exactly what I mean: > int monitor_fd_param(Monitor *mon, const char *fdname, Error **errp) { > ... local vars ... > if (!qemu_isdigit(fdname[0]) && mon) { > fd =3D monitor_get_fd(mon, fdname, &local_err); > } else { > fd =3D qemu_parse_fd(fdname); > } > ... report err to errp ... > } OK, if we're already using monitor_fd_param everywhere then I think we're already down the rat-hole of guessing whether we're an add-fd or fd by whether it's an integer, and I agree with you that we should just fix incoming to use that. Now, that means I guess we need to modify monitor_fd_param to tell us which type of fd it got, so we know whether to close it later? Dave P.S. I'm out from tomorrow for a weekish. > >> =A0> > >> =A0>> =A02) My suggestion about monitor_fd_param and make "fd:" for > >> =A0>> =A0migrate/migrate-incoming consistent. So user will be able t= o use > >> =A0>> =A0getfd + migrate-incoming > >> =A0>> =A03) Both of them or something else > >> =A0>> > >> >=20 > Regards, > Yury -- Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK