From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:52994) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1bUctz-0008K3-ML for qemu-devel@nongnu.org; Tue, 02 Aug 2016 12:50:48 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1bUctx-0000Wk-Kr for qemu-devel@nongnu.org; Tue, 02 Aug 2016 12:50:46 -0400 References: <1469616590-38683-1-git-send-email-vsementsov@virtuozzo.com> <1469616590-38683-2-git-send-email-vsementsov@virtuozzo.com> <9586a77d-3656-94e9-b94c-edb45ea10816@redhat.com> <57A07E90.10803@virtuozzo.com> From: John Snow Message-ID: <0e4e9795-f0c7-7a6a-ed9d-ad85c5c02f51@redhat.com> Date: Tue, 2 Aug 2016 12:50:37 -0400 MIME-Version: 1.0 In-Reply-To: <57A07E90.10803@virtuozzo.com> Content-Type: text/plain; charset=windows-1252; format=flowed Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [Qemu-block] [PATCH 1/3] blockjob: fix dead pointer in txn list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Vladimir Sementsov-Ogievskiy , qemu-devel@nongnu.org, qemu-block@nongnu.org Cc: kwolf@redhat.com, den@openvz.org, famz@redhat.com, stefanha@redhat.com, mreitz@redhat.com On 08/02/2016 07:05 AM, Vladimir Sementsov-Ogievskiy wrote: > On 02.08.2016 01:39, John Snow wrote: >> On 07/27/2016 06:49 AM, Vladimir Sementsov-Ogievskiy wrote: >>> Job may be freed in block_job_unref and in this case this would break >>> transaction QLIST. >>> >>> Fix this by removing job from this list before unref. >>> >>> Signed-off-by: Vladimir Sementsov-Ogievskiy >>> --- >>> blockjob.c | 1 + >>> 1 file changed, 1 insertion(+) >>> >>> diff --git a/blockjob.c b/blockjob.c >>> index a5ba3be..e045091 100644 >>> --- a/blockjob.c >>> +++ b/blockjob.c >>> @@ -216,6 +216,7 @@ static void block_job_completed_single(BlockJob >>> *job) >>> } >>> job->cb(job->opaque, job->ret); >>> if (job->txn) { >>> + QLIST_REMOVE(job, txn_list); >>> block_job_txn_unref(job->txn); >>> } >>> block_job_unref(job); >>> >> >> Has this caused actual problems for you? > > Yes, with the same changed test 124 (my parallel thread). Backup job can > finish too early (if dirty bitmap is empty) and then we use this > transaction job list with dead pointer. > >> >> This function is only ever called in a transactional context if the >> transaction is over -- so we're not likely to use the pointers ever >> again anyway. > > Backup job may finish even earlier than all jobs are added to the list. > (same case, empty dirty bitmap for one of drives). > AHA, I get it now. I think the right solution will be a general mechanism at the transactional level, not backup-specific hacks, but thank you for explaining this to me. >> >> Still, it's good practice, and the caller uses a safe iteration of the >> list, so I think this should be safe. >> >> But I don't think this SHOULD fix an actual bug. If it does, I think >> something else is wrong. >> >> Tested-by: John Snow >> Reviewed-by: John Snow > >