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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 0A674C43217 for ; Sat, 15 Oct 2022 01:26:02 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S229550AbiJOB0A (ORCPT ); Fri, 14 Oct 2022 21:26:00 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:50274 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S229513AbiJOBZ6 (ORCPT ); Fri, 14 Oct 2022 21:25:58 -0400 Received: from szxga08-in.huawei.com (szxga08-in.huawei.com [45.249.212.255]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 8004C5D70F for ; Fri, 14 Oct 2022 18:25:51 -0700 (PDT) Received: from canpemm500002.china.huawei.com (unknown [172.30.72.56]) by szxga08-in.huawei.com (SkyGuard) with ESMTP id 4Mq57B6HxWz1P7Lp; Sat, 15 Oct 2022 09:21:10 +0800 (CST) Received: from [10.174.151.185] (10.174.151.185) by canpemm500002.china.huawei.com (7.192.104.244) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256) id 15.1.2375.31; Sat, 15 Oct 2022 09:25:47 +0800 Subject: Re: [PATCH 1/3] hugetlb: fix vma lock handling during split vma and range unmapping To: Mike Kravetz , , CC: Muchun Song , David Hildenbrand , Sven Schnelle , Michal Hocko , Peter Xu , Naoya Horiguchi , "Aneesh Kumar K . V" , Andrea Arcangeli , "Kirill A . Shutemov" , Davidlohr Bueso , Prakash Sangappa , James Houghton , Mina Almasry , Pasha Tatashin , Axel Rasmussen , Ray Fucillo , Andrew Morton References: <20221005011707.514612-1-mike.kravetz@oracle.com> <20221005011707.514612-2-mike.kravetz@oracle.com> From: Miaohe Lin Message-ID: <5154292a-4c55-28cd-0935-82441e512fc3@huawei.com> Date: Sat, 15 Oct 2022 09:25:47 +0800 User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:78.0) Gecko/20100101 Thunderbird/78.6.0 MIME-Version: 1.0 In-Reply-To: <20221005011707.514612-2-mike.kravetz@oracle.com> Content-Type: text/plain; charset="utf-8" Content-Language: en-US Content-Transfer-Encoding: 7bit X-Originating-IP: [10.174.151.185] X-ClientProxiedBy: dggems701-chm.china.huawei.com (10.3.19.178) To canpemm500002.china.huawei.com (7.192.104.244) X-CFilter-Loop: Reflected Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Sorry for late respond. It's a really busy week. :) On 2022/10/5 9:17, Mike Kravetz wrote: > The hugetlb vma lock hangs off the vm_private_data field and is specific > to the vma. When vm_area_dup() is called as part of vma splitting, the Oh, I checked vm_area_dup() from callsite of copy_vma and dup_mmap but split_vma is missed... And yes, vma splitting can occur but vma merging won't for hugetlb vma. Thanks for catching this, Mike. > vma lock pointer is copied to the new vma. This will result in issues > such as double freeing of the structure. Update the hugetlb open vm_ops > to allocate a new vma lock for the new vma. > > The routine __unmap_hugepage_range_final unconditionally unset > VM_MAYSHARE to prevent subsequent pmd sharing. hugetlb_vma_lock_free > attempted to anticipate this by checking both VM_MAYSHARE and VM_SHARED. > However, if only VM_MAYSHARE was set we would miss the free. With the > introduction of the vma lock, a vma can not participate in pmd sharing > if vm_private_data is NULL. Instead of clearing VM_MAYSHARE in > __unmap_hugepage_range_final, free the vma lock to prevent sharing. Also, > update the sharing code to make sure vma lock is indeed a condition for > pmd sharing. hugetlb_vma_lock_free can then key off VM_MAYSHARE and not > miss any vmas. > > Fixes: "hugetlb: add vma based lock for pmd sharing" > Signed-off-by: Mike Kravetz > --- > mm/hugetlb.c | 43 +++++++++++++++++++++++++++---------------- > mm/memory.c | 4 ---- > 2 files changed, 27 insertions(+), 20 deletions(-) > > diff --git a/mm/hugetlb.c b/mm/hugetlb.c > index 4443e87e814b..0129d371800c 100644 > --- a/mm/hugetlb.c > +++ b/mm/hugetlb.c > @@ -4612,7 +4612,14 @@ static void hugetlb_vm_op_open(struct vm_area_struct *vma) > kref_get(&resv->refs); > } > > - hugetlb_vma_lock_alloc(vma); > + /* > + * vma_lock structure for sharable mappings is vma specific. > + * Clear old pointer (if copied via vm_area_dup) and create new. > + */ > + if (vma->vm_flags & VM_MAYSHARE) { > + vma->vm_private_data = NULL; > + hugetlb_vma_lock_alloc(vma); > + } IMHO this would lead to memoryleak. Think about the below move_vma() flow: move_vma copy_vma new_vma = vm_area_dup(vma); new_vma->vm_ops->open(new_vma); --> new_vma has its own vma lock. is_vm_hugetlb_page(vma) clear_vma_resv_huge_pages hugetlb_dup_vma_private --> vma->vm_private_data is set to NULL without put ref. So vma lock is *leaked*? Other part looks good to me. Thanks for your work. Thanks, Miaohe Lin