From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from relay.sgi.com (relay2.corp.sgi.com [137.38.102.29]) by oss.sgi.com (Postfix) with ESMTP id 0AE577F37 for ; Thu, 4 Apr 2013 01:57:28 -0500 (CDT) Received: from cuda.sgi.com (cuda1.sgi.com [192.48.157.11]) by relay2.corp.sgi.com (Postfix) with ESMTP id B14F830404E for ; Wed, 3 Apr 2013 23:57:27 -0700 (PDT) Received: from Ishtar.hs.tlinx.org (ishtar.tlinx.org [173.164.175.65]) by cuda.sgi.com with ESMTP id kNrZcBJJR74TJc1H (version=TLSv1 cipher=AES256-SHA bits=256 verify=NO) for ; Wed, 03 Apr 2013 23:57:25 -0700 (PDT) Message-ID: <515D2447.8060809@tlinx.org> Date: Wed, 03 Apr 2013 23:57:11 -0700 From: Linda Walsh MIME-Version: 1.0 Subject: Re: [PATCH] xfs: xfs: fallback to vmalloc for large buffers in xfs_attrlist_by_handle References: <515CB2C1.1050109@tlinx.org> <20130404035237.GA12011@dastard> <515D0A09.5030808@sandeen.net> In-Reply-To: <515D0A09.5030808@sandeen.net> List-Id: XFS Filesystem from SGI List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Errors-To: xfs-bounces@oss.sgi.com Sender: xfs-bounces@oss.sgi.com To: Eric Sandeen Cc: xfs-oss Even if I could reboot immediately, the last dump took 5 hours to run, My system's boot situation has been complicated by opensuse moving to systemd and moving all of the early boot utils (udev, mount, etc). to /usr, expecting /usr to already be mounted by the time the rootfs init is called on disk. Dinosaur that I am, I still boot from disk, but with the utils needed to mount any file systems no longer on the root partition, my booting has become an exercise involving bailing wire and ductape, but the new kernel w/the patch is already on my /boot partition... its the coming back up and getting /usr mounted before the "official boot" starts that is tricky... i.e. will test as soon as practical... ;^) Eric Sandeen wrote: > On 4/3/13 10:52 PM, Dave Chinner wrote: >> On Wed, Apr 03, 2013 at 03:52:49PM -0700, Linda Walsh wrote: >>> Note -- I might have more attributes than some as the disks are shared via samba >>> w/windows >>> and samba stores ACL and xattr info from windows in xfs's attrs... >>> >>> >>> xfsdump: WARNING: could not get list of root attributes for nondir ino >>> 3412893285: Cannot allocate memory (12) >> .... >>> [1007383.689492] xfsdump: page allocation failure: order:4, mode:0x10c0d0 >> .... >>> [1007383.689531] [] kmalloc_order_trace+0x34/0xe0 >>> [1007383.689534] [] __kmalloc+0x186/0x190 >>> [1007383.689541] [] xfs_attrlist_by_handle+0x8c/0x110 >>> [1007383.689544] [] xfs_file_ioctl+0x437/0xb60 >> So it's the attribute handle interfaces that need the vmalloc >> treatment here, just like has been done for all the other attribute >> interfaces.... > > Like this? > > From: Eric Sandeen > > xfs: fallback to vmalloc for large buffers in xfs_attrlist_by_handle > > Shamelessly copied from dchinner's: > ad650f5b xfs: fallback to vmalloc for large buffers in xfs_attrmulti_attr_get > > xfsdump uses for a large buffer for extended attributes, which has a > kmalloc'd shadow buffer in the kernel. This can fail after the > system has been running for some time as it is a high order > allocation. Add a fallback to vmalloc so that it doesn't require > contiguous memory and so won't randomly fail while xfsdump is > running. > > Signed-off-by: Eric Sandeen > --- > > built & tested by running through xfstests ./check -g dump > > diff --git a/fs/xfs/xfs_ioctl.c b/fs/xfs/xfs_ioctl.c > index d681e34..5e99968 100644 > --- a/fs/xfs/xfs_ioctl.c > +++ b/fs/xfs/xfs_ioctl.c > @@ -422,9 +422,12 @@ xfs_attrlist_by_handle( > if (IS_ERR(dentry)) > return PTR_ERR(dentry); > > - kbuf = kzalloc(al_hreq.buflen, GFP_KERNEL); > - if (!kbuf) > - goto out_dput; > + kbuf = kmem_zalloc(al_hreq.buflen, KM_SLEEP | KM_MAYFAIL); > + if (!kbuf) { > + kbuf = kmem_zalloc_large(al_hreq.buflen); > + if (!kbuf) > + goto out_dput; > + } > > cursor = (attrlist_cursor_kern_t *)&al_hreq.pos; > error = -xfs_attr_list(XFS_I(dentry->d_inode), kbuf, al_hreq.buflen, > @@ -436,7 +439,10 @@ xfs_attrlist_by_handle( > error = -EFAULT; > > out_kfree: > - kfree(kbuf); > + if (is_vmalloc_addr(kbuf)) > + kmem_free_large(kbuf); > + else > + kmem_free(kbuf); > out_dput: > dput(dentry); > return error; > > > _______________________________________________ xfs mailing list xfs@oss.sgi.com http://oss.sgi.com/mailman/listinfo/xfs