From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755939Ab0CITLd (ORCPT ); Tue, 9 Mar 2010 14:11:33 -0500 Received: from quartz.orcorp.ca ([139.142.54.143]:57429 "EHLO quartz.orcorp.ca" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755359Ab0CITLb (ORCPT ); Tue, 9 Mar 2010 14:11:31 -0500 Date: Tue, 9 Mar 2010 12:11:27 -0700 From: Jason Gunthorpe To: linux-kernel@vger.kernel.org Cc: "David P. Quigley" , Greg Kroah-Hartman , James Morris , "Eric W. Biederman" , Tejun Heo Subject: [PATCH] sysfs: Fixup broken chown tracking Message-ID: <20100309191127.GA30462@obsidianresearch.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.5.18 (2008-05-17) X-Broken-Reverse-DNS: no host name found for IP address 10.0.0.11 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org This patch: commit ddd29ec6597125c830f7badb608a86c98b936b64 Author: David P. Quigley Date: Wed Sep 9 14:25:37 2009 -0400 sysfs: Add labeling support for sysfs Introduced a bug into the way sysfs handles attribute changes from user space. Booting a kernel straight to busy box: Linux version 2.6.33-rc8-00033-g5e96a56-dirty (jgg@bertha1) (gcc version 3.4.4) #37 Tue Mar 9 11: [..] Freeing unused kernel memory: 1184k init BusyBox v1.1.0 (2006.12.14-23:35+0000) Built-in shell (ash) /empty # chown 0.301 /sys/devices/system/fpga/fpga0/led_0b /empty # chmod 0664 /sys/devices/system/fpga/fpga0/led_0b /empty # ls -l /sys/devices/system/fpga/fpga0/led_0b -rw-rw-r-- 1 0 0 4096 Mar 9 18:44 /sys/devices/system/fpga/fpga0/led_0b The chown was lost. The problem seems to stem from moving the check of iattr to be conditional on the first allocation of the sd_iattr. The chown call path looks like: [c308be50] [c00b1db4] sysfs_sd_setattr+0x34/0xf0 (unreliable) [c308be70] [c00b1ee8] sysfs_setattr+0x78/0xa0 [c308be90] [c0086338] notify_change+0x19c/0x2d4 [c308bec0] [c00707d4] chown_common+0x84/0xa8 [c308bf20] [c007084c] sys_chown+0x54/0x80 [c308bf40] [c000dcf0] ret_from_syscall+0x0/0x3c And sysfs_sd_setattr goes through the sysfs_init_inode_attrs allocation path and completely ignores the passed in iattr that contains the alteration from the chown. This patch makes the run through the iattr mandatory in all cases - this seems to solve the problem for me. Signed-off-by: Jason Gunthorpe --- fs/sysfs/inode.c | 36 ++++++++++++++++++------------------ 1 files changed, 18 insertions(+), 18 deletions(-) I've no idea if this is correct for any cases other than this single one, but there is definitely a nasty bug here. Hopefully someone more knowledgeable can sort this out :) diff --git a/fs/sysfs/inode.c b/fs/sysfs/inode.c index 220b758..b36f81c 100644 --- a/fs/sysfs/inode.c +++ b/fs/sysfs/inode.c @@ -81,24 +81,24 @@ int sysfs_sd_setattr(struct sysfs_dirent *sd, struct iattr * iattr) if (!sd_attrs) return -ENOMEM; sd->s_iattr = sd_attrs; - } else { - /* attributes were changed at least once in past */ - iattrs = &sd_attrs->ia_iattr; - - if (ia_valid & ATTR_UID) - iattrs->ia_uid = iattr->ia_uid; - if (ia_valid & ATTR_GID) - iattrs->ia_gid = iattr->ia_gid; - if (ia_valid & ATTR_ATIME) - iattrs->ia_atime = iattr->ia_atime; - if (ia_valid & ATTR_MTIME) - iattrs->ia_mtime = iattr->ia_mtime; - if (ia_valid & ATTR_CTIME) - iattrs->ia_ctime = iattr->ia_ctime; - if (ia_valid & ATTR_MODE) { - umode_t mode = iattr->ia_mode; - iattrs->ia_mode = sd->s_mode = mode; - } + } + + /* attributes were changed at least once in past */ + iattrs = &sd_attrs->ia_iattr; + + if (ia_valid & ATTR_UID) + iattrs->ia_uid = iattr->ia_uid; + if (ia_valid & ATTR_GID) + iattrs->ia_gid = iattr->ia_gid; + if (ia_valid & ATTR_ATIME) + iattrs->ia_atime = iattr->ia_atime; + if (ia_valid & ATTR_MTIME) + iattrs->ia_mtime = iattr->ia_mtime; + if (ia_valid & ATTR_CTIME) + iattrs->ia_ctime = iattr->ia_ctime; + if (ia_valid & ATTR_MODE) { + umode_t mode = iattr->ia_mode; + iattrs->ia_mode = sd->s_mode = mode; } return 0; } -- 1.5.4.2