From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mo-p00-ob.rzone.de ([81.169.146.162]:59575 "EHLO mo-p00-ob.rzone.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757428Ab3BSRxO (ORCPT ); Tue, 19 Feb 2013 12:53:14 -0500 Message-ID: <5123BC0D.9020304@giantdisaster.de> Date: Tue, 19 Feb 2013 18:53:17 +0100 From: Stefan Behrens MIME-Version: 1.0 To: bo.li.liu@oracle.com CC: linux-btrfs@vger.kernel.org Subject: Re: [PATCH V5] Btrfs: snapshot-aware defrag References: <20130123075155.GE17162@liubo.jp.oracle.com> <20130124005221.GA28406@liubo> <5102A76C.5050706@giantdisaster.de> <20130127131952.GB16722@liubo> <5106AD9D.5020906@giantdisaster.de> <20130216064743.GA3124@liubo.jp.oracle.com> <51225C9E.5000603@giantdisaster.de> <20130219042912.GA25330@liubo.jp.oracle.com> In-Reply-To: <20130219042912.GA25330@liubo.jp.oracle.com> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Sender: linux-btrfs-owner@vger.kernel.org List-ID: On 19.02.2013 05:29, Liu Bo wrote: > On Mon, Feb 18, 2013 at 05:53:50PM +0100, Stefan Behrens wrote: >> On Sat, 16 Feb 2013 14:47:45 +0800, Liu Bo wrote: >>> What about this patch(UNTESTED)? >>> >>> thanks, >>> liubo >>> >>> diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c >>> index ca7ace7..dac9d4b 100644 >>> --- a/fs/btrfs/inode.c >>> +++ b/fs/btrfs/inode.c >>> @@ -4142,9 +4142,14 @@ static void inode_tree_del(struct inode *inode) >>> * root_refs of 0, so this could end up dropping the tree root as a >>> * snapshot, so we need the extra !root->fs_info->tree_root check to >>> * make sure we don't drop it. >>> + * >>> + * Inode cache's inodes may be iput and add root back to dead roots >>> + * list during killing super, which leads to use-after-free, so >>> + * we need to check fs_info->closing to keep us from use-after-free. >>> */ >>> if (empty && btrfs_root_refs(&root->root_item) == 0 && >>> - root != root->fs_info->tree_root) { >>> + root != root->fs_info->tree_root && >>> + btrfs_fs_closing(root->fs_info) > 1) { >>> synchronize_srcu(&root->fs_info->subvol_srcu); >>> spin_lock(&root->inode_lock); >>> empty = RB_EMPTY_ROOT(&root->inode_tree); >> >> No improvement with this patch. The inode_cache causes a crash in __list_add. >> I tested it on the latest cmason/for-linus with and without your patch. > > Ahh, I think I made a finger error, > > + btrfs_fs_closing(root->fs_info) > 1) { > SHOULD be > + btrfs_fs_closing(root->fs_info) < 2) { Yes, this eliminates the crash. Thanks! >> This script is an 100% reproducer on my test box: >> mkfs.btrfs -d single -m raid1 /dev/sdc /dev/sdj /dev/sds /dev/sdt /dev/sdu /dev/sdv >> mount /dev/sdc /mnt -o compress=lzo,space_cache,inode_cache >> btrfs subv create /mnt/src >> (cd ~/git/btrfs/fs/btrfs && tar cf - .) | (cd /mnt/src && tar xf -) >> for i in `seq 2000`; do btrfs subv create /mnt/${i}; (cd /mnt/src && tar cf - .) | (cd /mnt/${i} && tar xf -); done >> for i in /mnt/[0-9]*; do btrfs subv dele ${i}; done >> sleep 45 >> umount /mnt > > With the latest cmason/for-linus(commit 6f60cbd3ae442cb35861bb522f388db123d42ec1 > btrfs: access superblock via pagecache in scan_one_device), > I ran this script several times with all good, I used two 40G disks, > others remains same. 6f60cbd is my HEAD too. Maybe the DEBUG options in the .config make the difference although I seem to recall that the issue was also always there with a plain RHEL config. > I'm wondering which line does 'del_fs_roots+0xaf/0xf0 [btrfs]' refer to? del_fs_roots+0xaf/0xf0 [btrfs]: 0x2c2af is in del_fs_roots (fs/btrfs/disk-io.c:3243). 3238 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3239 (void **)gang, 0, 3240 ARRAY_SIZE(gang)); 3241 if (!ret) 3242 break; 3243 for (i = 0; i < ret; i++) 3244 btrfs_free_fs_root(fs_info, gang[i]); 3245 } 3246 } 3247 Thanks for the patch :) >> BUG: unable to handle kernel paging request at ffff88023517d830 >> IP: [] __list_add+0x17/0xd0 >> PGD 1e0c063 PUD bf58e067 PMD bf737067 PTE 800000023517d160 >> Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC >> Modules linked in: btrfs raid1 mpt2sas scsi_transport_sas raid_class >> CPU 2 >> Pid: 18503, comm: umount Not tainted 3.7.0+ #44 Supermicro X8SIL/X8SIL >> RIP: 0010:[] [] __list_add+0x17/0xd0 >> RSP: 0018:ffff88019e1abbd8 EFLAGS: 00010286 >> RAX: ffff8802353aa290 RBX: ffff880229e38828 RCX: 0000000000000001 >> RDX: ffff88023517d828 RSI: ffff8802327214c0 RDI: ffff880229e38828 >> RBP: ffff88019e1abbf8 R08: 000000000006e130 R09: 0000000000000000 >> R10: 0000000000000000 R11: 0000000000000001 R12: ffff880229e38000 >> R13: ffff880229e38898 R14: 0000000000000000 R15: ffff88019e1abd30 >> FS: 00007f75eabc4740(0000) GS:ffff880236a00000(0000) knlGS:0000000000000000 >> CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b >> CR2: ffff88023517d830 CR3: 000000019e17e000 CR4: 00000000000007e0 >> DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 >> DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400 >> Process umount (pid: 18503, threadinfo ffff88019e1aa000, task ffff8802353aa290) >> Stack: >> ffffffffa008619f ffff880229e38000 ffff880229e38000 ffff880229e38898 >> ffff88019e1abc18 ffffffffa00861c0 ffff88012760dc38 ffff88012760dc38 >> ffff88019e1abc48 ffffffffa0095358 ffff88012760dc38 ffff88012760dcc0 >> Call Trace: >> [] ? btrfs_add_dead_root+0x1f/0x60 [btrfs] >> [] btrfs_add_dead_root+0x40/0x60 [btrfs] >> [] btrfs_destroy_inode+0x1d8/0x2d0 [btrfs] >> [] destroy_inode+0x37/0x60 >> [] evict+0x10d/0x1a0 >> [] iput+0x105/0x190 >> [] free_fs_root+0x18/0x90 [btrfs] >> [] btrfs_free_fs_root+0x7b/0x90 [btrfs] >> [] del_fs_roots+0xaf/0xf0 [btrfs] >> [] close_ctree+0x1c6/0x300 [btrfs] >> [] ? evict_inodes+0xec/0x100 >> [] btrfs_put_super+0x14/0x20 [btrfs] >> [] generic_shutdown_super+0x5c/0xe0 >> [] kill_anon_super+0x11/0x20 >> [] btrfs_kill_super+0x15/0x90 [btrfs] >> [] ? deactivate_super+0x41/0x70 >> [] deactivate_locked_super+0x3d/0x70 >> [] deactivate_super+0x49/0x70 >> [] mntput_no_expire+0xd2/0x130 >> [] sys_umount+0x71/0x390 >> [] system_call_fastpath+0x16/0x1b >>