Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754834Ab0D1DBi (ORCPT ); Tue, 27 Apr 2010 23:01:38 -0400 Received: from rcsinet10.oracle.com ([148.87.113.121]:27912 "EHLO rcsinet10.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754118Ab0D1DBh (ORCPT ); Tue, 27 Apr 2010 23:01:37 -0400 From: Tao Ma To: linux-kernel@vger.kernel.org Cc: Tao Ma , xfs@oss.sgi.com, Dave Chinner , Eric Sandeen , Christoph Hellwig , Alex Elder Subject: [PATCH v2] XFS: Let the broken fiemap work in query mode. Date: Wed, 28 Apr 2010 11:00:25 +0800 Message-Id: <1272423625-20142-1-git-send-email-tao.ma@oracle.com> X-Mailer: git-send-email 1.6.3.3 In-Reply-To: <20100428023058.GD9783@dastard> References: <20100428023058.GD9783@dastard> X-Auth-Type: Internal IP X-Source-IP: acsinet15.oracle.com [141.146.126.227] X-CT-RefId: str=0001.0A090209.4BD7A4FB.004B:SCFMA922111,ss=1,fgs=0 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 4833 Lines: 112 Dave Chinner wrote: > On Wed, Apr 28, 2010 at 10:00:01AM +0800, Tao Ma wrote: >> Hi Dave, >> >> Dave Chinner wrote: >>> On Tue, Apr 27, 2010 at 02:17:45PM +0800, Tao Ma wrote: >>>> According to Documentation/filesystems/fiemap.txt, If fm_extent_count >>>> is zero, then the fm_extents[] array is ignored (no extents will be >>>> returned), and the fm_mapped_extents count will hold the number of >>>> extents needed. >>>> >>>> But as the commit 97db39a1f6f69e906e98118392400de5217aa33a has changed >>>> bmv_count to the caller's input buffer, this number query function can't >>>> work any more. As this commit is written to change bmv_count from >>>> MAXEXTNUM because of ENOMEM, we can't find a really suitable number to >>>> set bmv_count now in xfs_vn_fiemap. Since we really have no idea of how >>>> much extents the file has, a big number may cause ENOMEM, while a small >>>> one will mask the real extent no. >>>> >>>> So this patch try to resolve this problem by adding a temporary getbmapx >>>> in xfs_getbmap. If the caller didn't give bmv_count, we don't allocate >>>> the "out" either. Instead, every time we want to use 'out', use '&tmp' >>>> instead. >>>> >>>> I know this solution is a bit ugly, but I can't find a way to resolve >>>> this issue while not changing the codes too much. So any good suggestion >>>> is welcomed. >>> I don't see a need to change xfs_getbmap() to fix this. We can limit >>> the maximum allocation size to something realistic just by setting >>> bm.bmv.count to something sane. e.g, in xfs_vn_fiemap: >>> >>> - bm.bmv_count = fieinfo->fi_extents_max + 1; >>> + bm.bmv.count = !fieinfo->fi_extents_max ? MAXEXTNUM : >>> + fieinfo->fi_extents_max - 1; >>> + bm.bmv_count = MIN(bm.bmv_count, >>> (PAGE_SIZE * 16 / sizeof(struct getbmapx))); >>> >>> Unless I'm missing something, that should also prevent the case of >>> an application providing a really large fi_extents_max from >>> triggering ENOMEM in most cases as well. >> I just worry about one thing: What if the real extent number is >> larger than the PAGE_SIZE * 16 / sizeof(struct getbmapx)? In this >> case, we will give up the wrong extent number to the user space. > > Applications need to handle mappings changing from query to getting > the mapping, so this should not be a major issue. Especially as the > method of fiemap indicating there are more extents to be extracted > from the inode in the case the kernel can't allocate a buffer large > enough is already documented. > > Realistically though, xfs_getbmap() needs a complete rewrite so > right now I'd prefer just to do the minimum to fix the reported > problem that continue to make it into even more of a mess than it is > now... Fair enough. Here is the updated patch. btw, I am working on adding the test cases in xfstests. Regards, Tao >From e5d32636c907be106d55d63c253d1750a4a898d7 Mon Sep 17 00:00:00 2001 From: Tao Ma Date: Wed, 28 Apr 2010 10:25:33 +0800 Subject: [PATCH v2] XFS: Let the broken fiemap work in query mode. According to Documentation/filesystems/fiemap.txt, If fm_extent_count is zero, then the fm_extents[] array is ignored (no extents will be returned), and the fm_mapped_extents count will hold the number of extents needed. But as the commit 97db39a1f6f69e906e98118392400de5217aa33a has changed bmv_count to the caller's input buffer, this number query function can't work any more. As this commit is written to change bmv_count from MAXEXTNUM because of ENOMEM. This patch just try to set bm.bmv.count to something sane. Thanks to Dave Chinner for the suggestion. Cc: Dave Chinner Cc: Eric Sandeen Cc: Christoph Hellwig Cc: Alex Elder Signed-off-by: Tao Ma --- fs/xfs/linux-2.6/xfs_iops.c | 5 ++++- 1 files changed, 4 insertions(+), 1 deletions(-) diff --git a/fs/xfs/linux-2.6/xfs_iops.c b/fs/xfs/linux-2.6/xfs_iops.c index 2259460..24ccad9 100644 --- a/fs/xfs/linux-2.6/xfs_iops.c +++ b/fs/xfs/linux-2.6/xfs_iops.c @@ -662,7 +662,10 @@ xfs_vn_fiemap( bm.bmv_length = BTOBB(length); /* We add one because in getbmap world count includes the header */ - bm.bmv_count = fieinfo->fi_extents_max + 1; + bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM : + fieinfo->fi_extents_max + 1; + bm.bmv_count = MIN(bm.bmv_count, + (__s32)(PAGE_SIZE * 16 / sizeof(struct getbmapx))); bm.bmv_iflags = BMV_IF_PREALLOC; if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) bm.bmv_iflags |= BMV_IF_ATTRFORK; -- 1.6.3.3.334.g916e1.dirty -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/