Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753443AbaBMIlU (ORCPT ); Thu, 13 Feb 2014 03:41:20 -0500 Received: from cn.fujitsu.com ([222.73.24.84]:61032 "EHLO song.cn.fujitsu.com" rhost-flags-OK-FAIL-OK-OK) by vger.kernel.org with ESMTP id S1751464AbaBMIlT (ORCPT ); Thu, 13 Feb 2014 03:41:19 -0500 X-IronPort-AV: E=Sophos;i="4.95,837,1384272000"; d="scan'208";a="9520096" Message-ID: <52FC832F.6000103@cn.fujitsu.com> Date: Thu, 13 Feb 2014 16:32:47 +0800 From: Gu Zheng User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:7.0.1) Gecko/20110930 Thunderbird/7.0.1 MIME-Version: 1.0 To: Andrey Tsyvarev CC: jaegeuk.kim@samsung.com, linux-kernel , linux-f2fs-devel@lists.sourceforge.net Subject: Re: [f2fs-dev] f2fs: f2fs unmount hangs if f2fs_init_acl() fails during mkdir syscall References: <52F320FC.50803@ispras.ru> <1391666564.25542.78.camel@kjgkr> <52F37D67.208@ispras.ru> <1391734185.25542.80.camel@kjgkr> <1391749933.25542.83.camel@kjgkr> <52F9DF85.7040402@ispras.ru> In-Reply-To: <52F9DF85.7040402@ispras.ru> X-MIMETrack: Itemize by SMTP Server on mailserver/fnst(Release 8.5.3|September 15, 2011) at 2014/02/13 16:39:04, Serialize by Router on mailserver/fnst(Release 8.5.3|September 15, 2011) at 2014/02/13 16:39:10, Serialize complete at 2014/02/13 16:39:10 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=UTF-8 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi Andrey, On 02/11/2014 04:29 PM, Andrey Tsyvarev wrote: > Hi, > >> It turns out that make_bad_inode prior to iput sets i_mode to a regular >> file, so that f2fs_evict_inode -> truncate_inode_pages -> >> f2fs_invalidate_data_page doesn't decrement dirty_dents. >> > It seems that remove_dirty_dir_inode() call should also be added to the error-path of > init_inode_metadata, because its functionality is also based on inode->i_mode field > which is changed by make_bad_inode(). It seems that your opinion is correct. remove_dirty_dir_inode() will not clean up the dir_inode_entry because make_bad_inode() sets i_mode to S_IFREG in the fail path of init_inode_metadata, and it leads to the following "memory leak". BTW, have you tested the case that added remove_dirty_dir_inode() into the fail path of init_inode_metadata? diff --git a/fs/f2fs/dir.c b/fs/f2fs/dir.c index e095a4f..d5a2c9e 100644 --- a/fs/f2fs/dir.c +++ b/fs/f2fs/dir.c @@ -375,6 +375,7 @@ put_error: /* once the failed inode becomes a bad inode, i_mode is S_IFREG */ truncate_inode_pages(&inode->i_data, 0); truncate_blocks(inode, 0); + remove_dirty_dir_inode(inode); error: remove_inode_page(inode); return ERR_PTR(err); Regards, Gu > > Otherwise memory leak is reported when f2fs module is unloaded: > > [ 231.378192] BUG f2fs_dirty_dir_entry (Tainted: GF O): Objects remaining in f2fs_dirty_dir_entry on kmem_cache_close() > [ 231.378193] ----------------------------------------------------------------------------- > > [ 231.378194] Disabling lock debugging due to kernel taint > [ 231.378195] INFO: Slab 0xffffea0000437200 objects=102 used=1 fp=0xffff880010dc8fc8 flags=0x3fffc000000080 > [ 231.378197] CPU: 0 PID: 2605 Comm: rmmod Tainted: GF B O 3.14.0-rc1fs #4 > [ 231.378198] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006 > [ 231.378199] ffff88000e5e3200 ffff88000cc9bd40 ffffffff8166fd7e ffffea0000437200 > [ 231.378202] ffff88000cc9be28 ffffffff811c3fdf ffff88003fc10066 ffffffff0cc9bda0 > [ 231.378203] ffffffff00000020 ffff88000cc9be38 ffff88000cc9bde0 656a624f00000296 > [ 231.378205] Call Trace: > [ 231.378210] [] dump_stack+0x45/0x56 > [ 231.378213] [] slab_err+0xaf/0xc0 > [ 231.378215] [] ? kmem_cache_close+0x133/0x340 > [ 231.378216] [] ? __kmalloc+0x1f5/0x250 > [ 231.378218] [] kmem_cache_close+0x153/0x340 > [ 231.378221] [] ? kmem_cache_destroy+0x27/0xf0 > [ 231.378223] [] __kmem_cache_shutdown+0x14/0x80 > [ 231.378224] [] kmem_cache_destroy+0x41/0xf0 > [ 231.378229] [] destroy_checkpoint_caches+0x21/0x30 [f2fs] > [ 231.378232] [] exit_f2fs_fs+0x28/0x34e [f2fs] > [ 231.378235] [] SyS_delete_module+0x152/0x1f0 > [ 231.378237] [] ? __audit_syscall_entry+0x9c/0xf0 > [ 231.378239] [] system_call_fastpath+0x16/0x1b > [ 231.378242] INFO: Object 0xffff880010dc8000 @offset=0 > [ 231.378245] kmem_cache_destroy f2fs_dirty_dir_entry: Slab cache still has objects > [ 231.378247] CPU: 0 PID: 2605 Comm: rmmod Tainted: GF B O 3.14.0-rc1fs #4 > [ 231.378247] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006 > [ 231.378248] ffff88000e5e3268 ffff88000cc9beb8 ffffffff8166fd7e ffff88000e5e3200 > [ 231.378250] ffff88000cc9bed8 ffffffff811934cf 0000000000000000 ffffffffa0204f60 > [ 231.378251] ffff88000cc9bee8 ffffffffa01eab91 ffff88000cc9bef8 ffffffffa01facda > [ 231.378253] Call Trace: > [ 231.378255] [] dump_stack+0x45/0x56 > [ 231.378256] [] kmem_cache_destroy+0xdf/0xf0 > [ 231.378259] [] destroy_checkpoint_caches+0x21/0x30 [f2fs] > [ 231.378262] [] exit_f2fs_fs+0x28/0x34e [f2fs] > [ 231.378263] [] SyS_delete_module+0x152/0x1f0 > [ 231.378265] [] ? __audit_syscall_entry+0x9c/0xf0 > [ 231.378266] [] system_call_fastpath+0x16/0x1b > > > Stack of allocation (obtained with KEDR, which is also used for fault simulation): > > [ 231.414875] [leak_check] Address: 0xffff880010dc8000, size: 24; stack trace of the allocation: > [ 231.414886] [leak_check] [] set_dirty_dir_page+0x62/0xe0 [f2fs] > [ 231.414893] [leak_check] [] f2fs_set_data_page_dirty+0x4e/0x90 [f2fs] > [ 231.414898] [leak_check] [] set_page_dirty+0x3a/0x60 > [ 231.414904] [leak_check] [] __f2fs_add_link+0x732/0x7d0 [f2fs] > [ 231.414909] [leak_check] [] f2fs_mkdir+0xbb/0x150 [f2fs] > [ 231.414914] [leak_check] [] vfs_mkdir+0xb7/0x160 > [ 231.414918] [leak_check] [] SyS_mkdir+0x5f/0xc0 > [ 231.414923] [leak_check] [] system_call_fastpath+0x16/0x1b > [ 231.414931] [leak_check] [] 0xffffffffffffffff > > > P.S. It was required to add 'slub_debug' kernel options for make SLUB output correct cache name, > otherwise cache "f2fs_dirty_dir_entry" was merged into "free_nid" one. It was surprise for me, > that's why patch investigation took so long time. > -- 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/