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 C3E25C47DAF for ; Thu, 18 Jan 2024 19:19:40 +0000 (UTC) Received: from localhost ([::1] helo=lists1p.gnu.org) by lists.gnu.org with esmtp (Exim 4.90_1) (envelope-from ) id 1rQXue-0002lI-Op; Thu, 18 Jan 2024 14:18:52 -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 1rQXuc-0002kY-4v for qemu-devel@nongnu.org; Thu, 18 Jan 2024 14:18:50 -0500 Received: from us-smtp-delivery-124.mimecast.com ([170.10.133.124]) by eggs.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1rQXua-0006Z2-5w for qemu-devel@nongnu.org; Thu, 18 Jan 2024 14:18:49 -0500 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=redhat.com; s=mimecast20190719; t=1705605525; 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=P0Xqmbst6oAXMoI/Xf5JcY391gphSyFh1bxi8ElP2QY=; b=hdGrbFVqgKpFDEFCj+88+DS7TsYu6IWPMfZacAHbbzSsuZcFI4FS7M7ikFDhXUg3fN0Ba0 QzyvKDF7KW5BEJMMW++/xAvErEzNHiudbU/9hIQ/fUqV7WPyWsdkemlR4omx7aaT7u40Y+ SWfaMaTO76ryL0MFXnfCRTHAXoYAY08= Received: from mimecast-mx02.redhat.com (mimecast-mx02.redhat.com [66.187.233.88]) by relay.mimecast.com with ESMTP with STARTTLS (version=TLSv1.3, cipher=TLS_AES_256_GCM_SHA384) id us-mta-656-3ZlrwpmgMJKiHHlXyTrpZg-1; Thu, 18 Jan 2024 14:18:42 -0500 X-MC-Unique: 3ZlrwpmgMJKiHHlXyTrpZg-1 Received: from smtp.corp.redhat.com (int-mx08.intmail.prod.int.rdu2.redhat.com [10.11.54.8]) (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 mimecast-mx02.redhat.com (Postfix) with ESMTPS id 3B311185A786; Thu, 18 Jan 2024 19:18:42 +0000 (UTC) Received: from redhat.com (unknown [10.39.194.209]) by smtp.corp.redhat.com (Postfix) with ESMTPS id 60DECC15587; Thu, 18 Jan 2024 19:18:41 +0000 (UTC) Date: Thu, 18 Jan 2024 20:18:40 +0100 From: Kevin Wolf To: Ari Sundholm Cc: qemu-devel@nongnu.org, Hanna Reitz , qemu-block@nongnu.org Subject: Re: [PATCH v2] block/blklogwrites: Protect mutable driver state with a mutex. Message-ID: References: <20240111163238.1346482-1-ari@tuxera.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20240111163238.1346482-1-ari@tuxera.com> X-Scanned-By: MIMEDefang 3.4.1 on 10.11.54.8 Received-SPF: pass client-ip=170.10.133.124; envelope-from=kwolf@redhat.com; helo=us-smtp-delivery-124.mimecast.com X-Spam_score_int: -38 X-Spam_score: -3.9 X-Spam_bar: --- X-Spam_report: (-3.9 / 5.0 requ) BAYES_00=-1.9, DKIMWL_WL_HIGH=-1.806, 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_H3=0.001, RCVD_IN_MSPIKE_WL=0.001, SPF_HELO_NONE=0.001, SPF_PASS=-0.001, T_SCC_BODY_TEXT_LINE=-0.01 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: 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 11.01.2024 um 17:32 hat Ari Sundholm geschrieben: > During the review of a fix for a concurrency issue in blklogwrites, > it was found that the driver needs an additional fix when enabling > multiqueue, which is a new feature introduced in QEMU 9.0, as the > driver state may be read and written by multiple threads at the same > time, which was not the case when the driver was originally written. > > Fix the multi-threaded scenario by introducing a mutex to protect the > mutable fields in the driver state, and always having the mutex locked > by the current thread when accessing them. Also use the mutex and a > condition variable to ensure that the super block is not being written > to by multiple threads concurrently. > > Additionally, add the const qualifier to a few BDRVBlkLogWritesState > pointer targets in contexts where the driver state is not written to. > > Signed-off-by: Ari Sundholm > > v1->v2: Ensure that the super block is not written to concurrently. > --- > block/blklogwrites.c | 77 +++++++++++++++++++++++++++++++++++++++----- > 1 file changed, 69 insertions(+), 8 deletions(-) > > diff --git a/block/blklogwrites.c b/block/blklogwrites.c > index ba717dab4d..f8bec7c863 100644 > --- a/block/blklogwrites.c > +++ b/block/blklogwrites.c > @@ -3,7 +3,7 @@ > * > * Copyright (c) 2017 Tuomas Tynkkynen > * Copyright (c) 2018 Aapo Vienamo > - * Copyright (c) 2018 Ari Sundholm > + * Copyright (c) 2018-2024 Ari Sundholm > * > * This work is licensed under the terms of the GNU GPL, version 2 or later. > * See the COPYING file in the top-level directory. > @@ -55,9 +55,34 @@ typedef struct { > BdrvChild *log_file; > uint32_t sectorsize; > uint32_t sectorbits; > + uint64_t update_interval; > + > + /* > + * The mutable state of the driver, consisting of the current log sector > + * and the number of log entries. > + * > + * May be read and/or written from multiple threads, and the mutex must be > + * held when accessing these fields. > + */ > uint64_t cur_log_sector; > uint64_t nr_entries; > - uint64_t update_interval; > + QemuMutex mutex; > + > + /* > + * The super block sequence number. Non-zero if a super block update is in > + * progress. > + * > + * The mutex must be held when accessing this field. > + */ > + uint64_t super_update_seq; > + > + /* > + * A condition variable to wait for and signal finished superblock updates. > + * > + * Used with the mutex to ensure that only one thread be updating the super > + * block at a time. > + */ > + QemuCond super_updated; > } BDRVBlkLogWritesState; > > static QemuOptsList runtime_opts = { > @@ -169,6 +194,9 @@ static int blk_log_writes_open(BlockDriverState *bs, QDict *options, int flags, > goto fail; > } > > + qemu_mutex_init(&s->mutex); > + qemu_cond_init(&s->super_updated); > + > log_append = qemu_opt_get_bool(opts, "log-append", false); > > if (log_append) { > @@ -231,6 +259,8 @@ static int blk_log_writes_open(BlockDriverState *bs, QDict *options, int flags, > s->nr_entries = 0; > } > > + s->super_update_seq = 0; > + > if (!blk_log_writes_sector_size_valid(log_sector_size)) { > ret = -EINVAL; > error_setg(errp, "Invalid log sector size %"PRIu64, log_sector_size); > @@ -255,6 +285,8 @@ fail_log: > bdrv_unref_child(bs, s->log_file); > bdrv_graph_wrunlock(); > s->log_file = NULL; > + qemu_cond_destroy(&s->super_updated); > + qemu_mutex_destroy(&s->mutex); > } > fail: > qemu_opts_del(opts); > @@ -269,6 +301,8 @@ static void blk_log_writes_close(BlockDriverState *bs) > bdrv_unref_child(bs, s->log_file); > s->log_file = NULL; > bdrv_graph_wrunlock(); > + qemu_cond_destroy(&s->super_updated); > + qemu_mutex_destroy(&s->mutex); > } > > static int64_t coroutine_fn GRAPH_RDLOCK > @@ -295,7 +329,7 @@ static void blk_log_writes_child_perm(BlockDriverState *bs, BdrvChild *c, > > static void blk_log_writes_refresh_limits(BlockDriverState *bs, Error **errp) > { > - BDRVBlkLogWritesState *s = bs->opaque; > + const BDRVBlkLogWritesState *s = bs->opaque; > bs->bl.request_alignment = s->sectorsize; > } > > @@ -338,15 +372,18 @@ blk_log_writes_co_do_log(BlkLogWritesLogReq *lr) > * driver may be modified by other driver operations while waiting for the > * I/O to complete. > */ > + qemu_mutex_lock(&s->mutex); > const uint64_t entry_start_sector = s->cur_log_sector; > const uint64_t entry_offset = entry_start_sector << s->sectorbits; > const uint64_t qiov_aligned_size = ROUND_UP(lr->qiov->size, s->sectorsize); > const uint64_t entry_aligned_size = qiov_aligned_size + > ROUND_UP(lr->zero_size, s->sectorsize); > const uint64_t entry_nr_sectors = entry_aligned_size >> s->sectorbits; > + const uint64_t entry_seq = s->nr_entries + 1; > > - s->nr_entries++; > + s->nr_entries = entry_seq; > s->cur_log_sector += entry_nr_sectors; > + qemu_mutex_unlock(&s->mutex); > > /* > * Write the log entry. Note that if this is a "write zeroes" operation, > @@ -366,17 +403,34 @@ blk_log_writes_co_do_log(BlkLogWritesLogReq *lr) > > /* Update super block on flush or every update interval */ > if (lr->log_ret == 0 && ((lr->entry.flags & LOG_FLUSH_FLAG) > - || (s->nr_entries % s->update_interval == 0))) > + || (entry_seq % s->update_interval == 0))) > { > struct log_write_super super = { > .magic = cpu_to_le64(WRITE_LOG_MAGIC), > .version = cpu_to_le64(WRITE_LOG_VERSION), > - .nr_entries = cpu_to_le64(s->nr_entries), > + .nr_entries = const_le64(0), > .sectorsize = cpu_to_le32(s->sectorsize), > }; > - void *zeroes = g_malloc0(s->sectorsize - sizeof(super)); > + void *zeroes; > QEMUIOVector qiov; > > + /* > + * Wait if a super block update is already in progress. > + * Bail out if a newer update got its turn before us. > + */ > + WITH_QEMU_LOCK_GUARD(&s->mutex) { > + while (s->super_update_seq) { > + if (entry_seq < s->super_update_seq) { > + return; > + } > + qemu_cond_wait(&s->super_updated, &s->mutex); This will block, which is exactly what you want if another thread is writing the super block. However, in a single-threaded case where it's just the previous request coroutine that is still writing its super block (i.e. bdrv_co_pwritev() below has yielded), this will deadlock because we'll never switch back and actually complete the previous super block write. So unless I'm missing a reason why this won't happen, I think you need a coroutine aware mechanism here. The obvious options would be using a CoMutex in the first place and holding it across the I/O operation or keeping the cheaper QemuMutex and replacing the condition variable with a CoQueue. > + } > + s->super_update_seq = entry_seq; > + super.nr_entries = cpu_to_le64(s->nr_entries); > + } > + > + zeroes = g_malloc0(s->sectorsize - sizeof(super)); > + > qemu_iovec_init(&qiov, 2); > qemu_iovec_add(&qiov, &super, sizeof(super)); > qemu_iovec_add(&qiov, zeroes, s->sectorsize - sizeof(super)); Kevin