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=-8.5 required=3.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH,MAILING_LIST_MULTI,SIGNED_OFF_BY,SPF_PASS,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 26F54C04EB8 for ; Wed, 5 Dec 2018 03:03:11 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id DA7C02081C for ; Wed, 5 Dec 2018 03:03:10 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org DA7C02081C Authentication-Results: mail.kernel.org; dmarc=fail (p=none dis=none) header.from=redhat.com Authentication-Results: mail.kernel.org; spf=none smtp.mailfrom=linux-block-owner@vger.kernel.org Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1725865AbeLEDDK (ORCPT ); Tue, 4 Dec 2018 22:03:10 -0500 Received: from mx1.redhat.com ([209.132.183.28]:44628 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1725864AbeLEDDK (ORCPT ); Tue, 4 Dec 2018 22:03:10 -0500 Received: from smtp.corp.redhat.com (int-mx04.intmail.prod.int.phx2.redhat.com [10.5.11.14]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mx1.redhat.com (Postfix) with ESMTPS id 96845307DAB6; Wed, 5 Dec 2018 03:03:09 +0000 (UTC) Received: from ming.t460p (ovpn-8-22.pek2.redhat.com [10.72.8.22]) by smtp.corp.redhat.com (Postfix) with ESMTPS id 74CC418E30; Wed, 5 Dec 2018 03:03:05 +0000 (UTC) Date: Wed, 5 Dec 2018 11:03:01 +0800 From: Ming Lei To: Jens Axboe Cc: "linux-block@vger.kernel.org" Subject: Re: [PATCH] blk-mq: fix corruption with direct issue Message-ID: <20181205030300.GG17845@ming.t460p> References: <1d359819-5410-7af2-d02b-f0ecca39d2c9@kernel.dk> <20181205013736.GD17845@ming.t460p> <37bf8821-c205-717a-df0d-96ecfb0f75aa@kernel.dk> <20181205022716.GE17845@ming.t460p> <227a40a3-6599-9fc0-ab58-674f063e9c3a@kernel.dk> <20181205025801.GF17845@ming.t460p> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20181205025801.GF17845@ming.t460p> User-Agent: Mutt/1.9.1 (2017-09-22) X-Scanned-By: MIMEDefang 2.79 on 10.5.11.14 X-Greylist: Sender IP whitelisted, not delayed by milter-greylist-4.5.16 (mx1.redhat.com [10.5.110.42]); Wed, 05 Dec 2018 03:03:09 +0000 (UTC) Sender: linux-block-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org On Wed, Dec 05, 2018 at 10:58:02AM +0800, Ming Lei wrote: > On Tue, Dec 04, 2018 at 07:30:24PM -0700, Jens Axboe wrote: > > On 12/4/18 7:27 PM, Ming Lei wrote: > > > On Tue, Dec 04, 2018 at 07:16:11PM -0700, Jens Axboe wrote: > > >> On 12/4/18 6:37 PM, Ming Lei wrote: > > >>> On Tue, Dec 04, 2018 at 03:47:46PM -0700, Jens Axboe wrote: > > >>>> If we attempt a direct issue to a SCSI device, and it returns BUSY, then > > >>>> we queue the request up normally. However, the SCSI layer may have > > >>>> already setup SG tables etc for this particular command. If we later > > >>>> merge with this request, then the old tables are no longer valid. Once > > >>>> we issue the IO, we only read/write the original part of the request, > > >>>> not the new state of it. > > >>>> > > >>>> This causes data corruption, and is most often noticed with the file > > >>>> system complaining about the just read data being invalid: > > >>>> > > >>>> [ 235.934465] EXT4-fs error (device sda1): ext4_iget:4831: inode #7142: comm dpkg-query: bad extra_isize 24937 (inode size 256) > > >>>> > > >>>> because most of it is garbage... > > >>>> > > >>>> This doesn't happen from the normal issue path, as we will simply defer > > >>>> the request to the hardware queue dispatch list if we fail. Once it's on > > >>>> the dispatch list, we never merge with it. > > >>>> > > >>>> Fix this from the direct issue path by flagging the request as > > >>>> REQ_NOMERGE so we don't change the size of it before issue. > > >>>> > > >>>> See also: > > >>>> https://bugzilla.kernel.org/show_bug.cgi?id=201685 > > >>>> > > >>>> Fixes: 6ce3dd6eec1 ("blk-mq: issue directly if hw queue isn't busy in case of 'none'") > > >>>> Signed-off-by: Jens Axboe > > >>>> > > >>>> --- > > >>>> > > >>>> diff --git a/block/blk-mq.c b/block/blk-mq.c > > >>>> index 3f91c6e5b17a..d8f518c6ea38 100644 > > >>>> --- a/block/blk-mq.c > > >>>> +++ b/block/blk-mq.c > > >>>> @@ -1715,6 +1715,15 @@ static blk_status_t __blk_mq_issue_directly(struct blk_mq_hw_ctx *hctx, > > >>>> break; > > >>>> case BLK_STS_RESOURCE: > > >>>> case BLK_STS_DEV_RESOURCE: > > >>>> + /* > > >>>> + * If direct dispatch fails, we cannot allow any merging on > > >>>> + * this IO. Drivers (like SCSI) may have set up permanent state > > >>>> + * for this request, like SG tables and mappings, and if we > > >>>> + * merge to it later on then we'll still only do IO to the > > >>>> + * original part. > > >>>> + */ > > >>>> + rq->cmd_flags |= REQ_NOMERGE; > > >>>> + > > >>>> blk_mq_update_dispatch_busy(hctx, true); > > >>>> __blk_mq_requeue_request(rq); > > >>>> break; > > >>>> > > >>> > > >>> Not sure it is enough to just mark it as NOMERGE, for example, driver > > >>> may have setup the .special_vec for discard, and NOMERGE may not prevent > > >>> request from entering elevator queue completely. Cause 'rq.rb_node' and > > >>> 'rq.special_vec' share same space. > > >> > > >> We should rather limit the scope of the direct dispatch instead. It > > >> doesn't make sense to do for anything but read/write anyway. > > > > > > discard is kind of write, and it isn't treated very specially in make > > > request path, except for multi-range discard. > > > > The point of direct dispatch is to reduce latencies for requests, > > discards are so damn slow on ALL devices anyway that it doesn't make any > > sense to try direct dispatch to begin with, regardless of whether it > > possible or not. > > SCSI MQ device may benefit from direct dispatch from reduced lock contention. > > > > > >>> So how about inserting this request via blk_mq_request_bypass_insert() > > >>> in case that direct issue returns BUSY? Then it is invariant that > > >>> any request queued via .queue_rq() won't enter scheduler queue. > > >> > > >> I did consider this, but I didn't want to experiment with exercising > > >> a new path for an important bug fix. You do realize that your original > > >> patch has been corrupting data for months? I think a little caution > > >> is in order here. > > > > > > But marking NOMERGE still may have a hole on re-insert discard request as > > > mentioned above. > > > > What I said was further limit the scope of direct dispatch, which means > > not allowing anything that isn't a read/write. > > IMO, the conservative approach is to take the one used in legacy io > path, in which it is never allowed to re-insert queued request to > scheduler queue except for requeue, however RQF_DONTPREP is cleared > before requeuing request to scheduler. > > > > > > Given we never allow to re-insert queued request to scheduler queue > > > except for 6ce3dd6eec1, I think it is the correct thing to do, and the > > > fix is simple too. > > > > As I said, it's not the time to experiment. This issue has been there > > since 4.19-rc1. The alternative is yanking both those patches, and then > > looking at it later when the direct issue path has been cleaned up > > first. > > The issue should have been there from v4.1, especially after commit > f984df1f0f7 ("blk-mq: do limited block plug for multiple queue case"), > which is the 1st one to re-insert the queued request into scheduler > queue. But at that time, there isn't io scheduler for MQ, so in theory the issue should be there since v4.11, especially 945ffb60c11d ("mq-deadline: add blk-mq adaptation of the deadline IO scheduler"). Thanks, Ming