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 Received: from lists.gnu.org (lists.gnu.org [209.51.188.17]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id A983DE81A3B for ; Mon, 16 Feb 2026 15:31:33 +0000 (UTC) Received: from localhost ([::1] helo=lists1p.gnu.org) by lists.gnu.org with esmtp (Exim 4.90_1) (envelope-from ) id 1vs0Z1-0001Tw-Lj; Mon, 16 Feb 2026 10:31:07 -0500 Received: from eggs.gnu.org ([2001:470:142:3::10]) by lists.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1vs0Z0-0001TZ-2Q for qemu-devel@nongnu.org; Mon, 16 Feb 2026 10:31:06 -0500 Received: from us-smtp-delivery-124.mimecast.com ([170.10.129.124]) by eggs.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1vs0Yw-00069y-VK for qemu-devel@nongnu.org; Mon, 16 Feb 2026 10:31:05 -0500 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=redhat.com; s=mimecast20190719; t=1771255861; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version:content-type:content-type: in-reply-to:in-reply-to:references:references; bh=RCNcDgCEWgn87j8TFIPECxqRK40V1qcHgDASHOLoCtA=; b=NQM3RmgyCiZLCkzRTgMsr62+FbAmLI6MyOL0H3BB7PcoPoRTIyhQ4pgYe2fw3yqU6Jeuxf mfXtSetjhSFP4iTn/3t33WsSGM8MT3aU2ThvxGYGDE8WCeoTVCHVAPj4IMBzhzfii8CzHD PJwAIbQXoOnBPz56bThCEim1PL/DQMk= Received: from mx-prod-mc-03.mail-002.prod.us-west-2.aws.redhat.com (ec2-54-186-198-63.us-west-2.compute.amazonaws.com [54.186.198.63]) by relay.mimecast.com with ESMTP with STARTTLS (version=TLSv1.3, cipher=TLS_AES_256_GCM_SHA384) id us-mta-323-Ww3tiHd7NL-BWsd-9F9WwA-1; Mon, 16 Feb 2026 10:30:59 -0500 X-MC-Unique: Ww3tiHd7NL-BWsd-9F9WwA-1 X-Mimecast-MFC-AGG-ID: Ww3tiHd7NL-BWsd-9F9WwA_1771255858 Received: from mx-prod-int-06.mail-002.prod.us-west-2.aws.redhat.com (mx-prod-int-06.mail-002.prod.us-west-2.aws.redhat.com [10.30.177.93]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (2048 bits) server-digest SHA256) (No client certificate requested) by mx-prod-mc-03.mail-002.prod.us-west-2.aws.redhat.com (Postfix) with ESMTPS id 53F451956054; Mon, 16 Feb 2026 15:30:57 +0000 (UTC) Received: from redhat.com (unknown [10.44.33.255]) by mx-prod-int-06.mail-002.prod.us-west-2.aws.redhat.com (Postfix) with ESMTPS id 3E3131800666; Mon, 16 Feb 2026 15:30:52 +0000 (UTC) Date: Mon, 16 Feb 2026 16:30:50 +0100 From: Kevin Wolf To: Fiona Ebner Cc: qemu-devel@nongnu.org, qemu-stable@nongnu.org, qemu-block@nongnu.org, hreitz@redhat.com, vsementsov@yandex-team.ru, jsnow@redhat.com, jean-louis@dupond.be, dionbosschieter@gmail.com Subject: Re: [PATCH] block/mirror: disable dirty bitmap only after job init Message-ID: References: <20260212120411.369498-1-f.ebner@proxmox.com> <66c3fca8-0f64-4cec-a2a9-5959ef67a251@proxmox.com> <5f336c06-c675-4d21-8c1c-e1a1ebe8e09f@proxmox.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <5f336c06-c675-4d21-8c1c-e1a1ebe8e09f@proxmox.com> X-Scanned-By: MIMEDefang 3.4.1 on 10.30.177.93 Received-SPF: pass client-ip=170.10.129.124; envelope-from=kwolf@redhat.com; helo=us-smtp-delivery-124.mimecast.com X-Spam_score_int: -20 X-Spam_score: -2.1 X-Spam_bar: -- X-Spam_report: (-2.1 / 5.0 requ) BAYES_00=-1.9, DKIMWL_WL_HIGH=-0.001, DKIM_SIGNED=0.1, DKIM_VALID=-0.1, DKIM_VALID_AU=-0.1, DKIM_VALID_EF=-0.1, RCVD_IN_DNSWL_NONE=-0.0001, RCVD_IN_MSPIKE_H4=0.001, RCVD_IN_MSPIKE_WL=0.001, RCVD_IN_VALIDITY_RPBL_BLOCKED=0.001, RCVD_IN_VALIDITY_SAFE_BLOCKED=0.001, SPF_HELO_PASS=-0.001, SPF_PASS=-0.001 autolearn=ham autolearn_force=no X-Spam_action: no action X-BeenThere: qemu-devel@nongnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: qemu development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: qemu-devel-bounces+qemu-devel=archiver.kernel.org@nongnu.org Sender: qemu-devel-bounces+qemu-devel=archiver.kernel.org@nongnu.org Am 16.02.2026 um 14:01 hat Fiona Ebner geschrieben: > Am 16.02.26 um 1:48 PM schrieb Fiona Ebner: > > Am 16.02.26 um 11:25 AM schrieb Kevin Wolf: > >> Am 12.02.2026 um 13:02 hat Fiona Ebner geschrieben: > >>> Currently, the dirty bitmap is disabled too early and the following > >>> bad scenario is possible: > >>> > >>> 1. Dirty bitmap is disabled in mirror_start_job() > >>> 2. Some request are started in mirror_top_bs while s->job == NULL > >>> 3. mirror_dirty_init() -> bdrv_co_is_allocated_above() runs and because > >>> the request hasn't completed yet, the block isn't allocated > >>> 4. The request completes, still sees s->job == NULL and skips the > >>> bitmap, and nothing else will mark it dirty either > >>> > >>> One ingredient is that mirror_top_opaque->job is only set after the > >>> job is fully initialized. For the rationale, see commit 32125b1460 > >>> ("mirror: Fix access of uninitialised fields during start"). > >>> > >>> Disabling the dirty bitmap is safe once bdrv_mirror_top_do_write() > >>> sees that the job is set, because then: > >>> > >>> 1. When not using MIRROR_COPY_MODE_WRITE_BLOCKING, the dirty bitmap > >>> will be set by bdrv_mirror_top_do_write(). > >>> > >>> 2. When using MIRROR_COPY_MODE_WRITE_BLOCKING, writes will be done > >>> synchronously to the target. > >>> > >>> At least with virtio-blk using iothread-vq-mapping, mirror_run() and > >>> bdrv_mirror_top_do_write() might be called in different threads. > >>> bdrv_disable_dirty_bitmap() acquires and releases the dirty bitmap > >>> mutex, so the memory is synchronized between threads. > >>> > >>> Many thanks to Kevin Wolf for discussing the issue and solutions with > >>> me! [0] > >>> > >>> [0]: https://lore.kernel.org/qemu-devel/4853b0e5-8ec3-41e9-9a53-b1912b8e4449@dupond.be/T/ > >>> > >>> Cc: qemu-stable@nongnu.org > >>> Closes: https://gitlab.com/qemu-project/qemu/-/issues/3273 > >>> Signed-off-by: Fiona Ebner > >>> --- > >>> block/mirror.c | 21 +++++++++++++++------ > >>> 1 file changed, 15 insertions(+), 6 deletions(-) > >>> > >>> diff --git a/block/mirror.c b/block/mirror.c > >>> index b344182c74..eadd4501e8 100644 > >>> --- a/block/mirror.c > >>> +++ b/block/mirror.c > >>> @@ -1123,6 +1123,21 @@ static int coroutine_fn mirror_run(Job *job, Error **errp) > >>> */ > >>> mirror_top_opaque->job = s; > >>> > >>> + /* > >>> + * Disabling the dirty bitmap is safe once bdrv_mirror_top_do_write() sees > >>> + * that the job is set, because then: > >>> + * > >>> + * 1. When not using MIRROR_COPY_MODE_WRITE_BLOCKING, the dirty bitmap will > >>> + * be set by bdrv_mirror_top_do_write(). > >>> + * > >>> + * 2. When using MIRROR_COPY_MODE_WRITE_BLOCKING, writes will be done > >>> + * synchronously to the target. > >>> + * > >>> + * bdrv_disable_dirty_bitmap() acquires and releases the dirty bitmap mutex, > >>> + * so the memory is synchronized between threads. > >>> + */ > >>> + bdrv_disable_dirty_bitmap(s->dirty_bitmap); > >>> + > >>> assert(!s->dbi); > >>> s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap); > >>> for (;;) { > >>> @@ -2014,12 +2029,6 @@ static BlockJob *mirror_start_job( > >>> goto fail; > >>> } > >>> > >>> - /* > >>> - * The dirty bitmap is set by bdrv_mirror_top_do_write() when not in active > >>> - * mode. > >>> - */ > >>> - bdrv_disable_dirty_bitmap(s->dirty_bitmap); > >>> - > >>> bdrv_graph_wrlock_drained(); > >>> ret = block_job_add_bdrv(&s->common, "source", bs, 0, > >>> BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE | > >> > >> The thing I meant in the other thread is if we don't need something like > >> this additionally: > >> > >> diff --git a/block/mirror.c b/block/mirror.c > >> index b344182c747..159954158ba 100644 > >> --- a/block/mirror.c > >> +++ b/block/mirror.c > >> @@ -1672,9 +1672,17 @@ bdrv_mirror_top_do_write(BlockDriverState *bs, MirrorMethod method, > >> abort(); > >> } > >> > >> - if (!copy_to_target && s->job && s->job->dirty_bitmap) { > >> + if (!copy_to_target) { > >> qatomic_set(&s->job->actively_synced, false); > >> - bdrv_set_dirty_bitmap(s->job->dirty_bitmap, offset, bytes); > >> + if (s->job && s->job->dirty_bitmap) { > >> + bdrv_set_dirty_bitmap(s->job->dirty_bitmap, offset, bytes); > >> + } else { > >> + /* > >> + * Avoid race in the case that mirror_run() disables the bitmap > >> + * between here and bdrv_co_write_req_finish(). > >> + */ > >> + bdrv_set_dirty(bs, offset, bytes); > >> + } > >> } > >> > >> if (ret < 0) { > > > > You're right! It's not enough to ensure that the job is set before the > > bitmap is disabled, because both could happen after the check for job here. > > > > Isn't this change enough by itself? After the change, > > bdrv_mirror_top_do_write() ensures that: > > > > 1. When copy_to_target == true, the write is done synchronously to the > > target, no need to set the dirty bitmap. > > > > 2. When copy_to_target == false, the dirty bitmap is set. > > > > So it doesn't really matter at which point the dirty bitmap is disabled? > > Or am I missing something again? > > Ah right, bdrv_set_dirty() only sets it if still enabled :) Yes, exactly. If that's still too confusing, how about just giving mirror_top_bs access to its dirty bitmap right from the start while it's drained? I didn't try to reproduce the bug with it yet, though. Kevin diff --git a/block/mirror.c b/block/mirror.c index b344182c747..c11aca1366c 100644 --- a/block/mirror.c +++ b/block/mirror.c @@ -99,6 +99,7 @@ typedef struct MirrorBlockJob { typedef struct MirrorBDSOpaque { MirrorBlockJob *job; + BdrvDirtyBitmap *dirty_bitmap; bool stop; bool is_commit; } MirrorBDSOpaque; @@ -1672,9 +1673,9 @@ bdrv_mirror_top_do_write(BlockDriverState *bs, MirrorMethod method, abort(); } - if (!copy_to_target && s->job && s->job->dirty_bitmap) { + if (!copy_to_target) { qatomic_set(&s->job->actively_synced, false); - bdrv_set_dirty_bitmap(s->job->dirty_bitmap, offset, bytes); + bdrv_set_dirty_bitmap(s->dirty_bitmap, offset, bytes); } if (ret < 0) { @@ -1901,13 +1902,28 @@ static BlockJob *mirror_start_job( bdrv_drained_begin(bs); ret = bdrv_append(mirror_top_bs, bs, errp); - bdrv_drained_end(bs); - if (ret < 0) { + bdrv_drained_end(bs); bdrv_unref(mirror_top_bs); return NULL; } + bs_opaque->dirty_bitmap = bdrv_create_dirty_bitmap(mirror_top_bs, + granularity, + NULL, errp); + if (!bs_opaque->dirty_bitmap) { + bdrv_drained_end(bs); + bdrv_unref(mirror_top_bs); + return NULL; + } + + /* + * The dirty bitmap is set by bdrv_mirror_top_do_write() when not in active + * mode. + */ + bdrv_disable_dirty_bitmap(bs_opaque->dirty_bitmap); + bdrv_drained_end(bs); + /* Make sure that the source is not resized while the job is running */ s = block_job_create(job_id, driver, NULL, mirror_top_bs, BLK_PERM_CONSISTENT_READ, @@ -2002,24 +2018,13 @@ static BlockJob *mirror_start_job( s->base_overlay = bdrv_find_overlay(bs, base); s->granularity = granularity; s->buf_size = ROUND_UP(buf_size, granularity); + s->dirty_bitmap = bs_opaque->dirty_bitmap; s->unmap = unmap; if (auto_complete) { s->should_complete = true; } bdrv_graph_rdunlock_main_loop(); - s->dirty_bitmap = bdrv_create_dirty_bitmap(s->mirror_top_bs, granularity, - NULL, errp); - if (!s->dirty_bitmap) { - goto fail; - } - - /* - * The dirty bitmap is set by bdrv_mirror_top_do_write() when not in active - * mode. - */ - bdrv_disable_dirty_bitmap(s->dirty_bitmap); - bdrv_graph_wrlock_drained(); ret = block_job_add_bdrv(&s->common, "source", bs, 0, BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE | @@ -2099,9 +2104,6 @@ fail: g_free(s->replaces); blk_unref(s->target); bs_opaque->job = NULL; - if (s->dirty_bitmap) { - bdrv_release_dirty_bitmap(s->dirty_bitmap); - } job_early_fail(&s->common.job); } @@ -2115,6 +2117,7 @@ fail: bdrv_graph_wrunlock(); bdrv_drained_end(bs); + bdrv_release_dirty_bitmap(bs_opaque->dirty_bitmap); bdrv_unref(mirror_top_bs); return NULL;