Return-Path: Received: from sonic301-28.consmr.mail.gq1.yahoo.com ([98.137.64.154]:35511 "EHLO sonic301-28.consmr.mail.gq1.yahoo.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751068AbeEaVIT (ORCPT ); Thu, 31 May 2018 17:08:19 -0400 Subject: Re: [PATCH 1/1] Fix memory leak in kernfs_security_xattr_set and kernfs_security_xattr_set To: "Eric W. Biederman" Cc: CHANDAN VN , gregkh@linuxfoundation.org, tj@kernel.org, bfields@fieldses.org, jlayton@kernel.org, linux-kernel@vger.kernel.org, linux-nfs@vger.kernel.org, cpgs@samsung.com, sireesha.t@samsung.com, Casey Schaufler References: <1527758911-18610-1-git-send-email-chandan.vn@samsung.com> <87po1ba6hv.fsf@xmission.com> From: Casey Schaufler Message-ID: <9fb8de9d-3a9a-706b-50a9-11e768f72851@schaufler-ca.com> Date: Thu, 31 May 2018 14:08:14 -0700 MIME-Version: 1.0 In-Reply-To: <87po1ba6hv.fsf@xmission.com> Content-Type: text/plain; charset=utf-8 Sender: linux-nfs-owner@vger.kernel.org List-ID: On 5/31/2018 1:57 PM, Eric W. Biederman wrote: > Casey Schaufler writes: > >> On 5/31/2018 2:28 AM, CHANDAN VN wrote: >>> From: "sireesha.t" >>> >>> Leak is caused because smack_inode_getsecurity() is allocating memory >>> using kstrdup(). Though the security_release_secctx() is called, it >>> would not free the allocated memory. Calling security_release_secctx is >>> not relevant for this scenario as inode_getsecurity() does not provide a >>> "secctx". >>> >>> Similar fix has been mainlined: >>> https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/commit/?id=57e7ba04d422c3d41c8426380303ec9b7533ded9 >>> >>> The fix is to replace the security_release_secctx() with a kfree() >>> >>> Below is the KMEMLEAK dump: >>> unreferenced object 0xffffffc025e11c80 (size 64): >>> comm "systemd-tmpfile", pid 2452, jiffies 4294894464 (age 235587.492s) >>> hex dump (first 32 bytes): >>> 53 79 73 74 65 6d 3a 3a 53 68 61 72 65 64 00 00 System::Shared.. >>> 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ >>> backtrace: >>> [] __save_stack_trace+0x28/0x34 >>> [] create_object+0x130/0x25c >>> [] kmemleak_alloc+0x30/0x5c >>> [] __kmalloc_track_caller+0x1cc/0x2a8 >>> [] kstrdup+0x3c/0x6c >>> [] smack_inode_getsecurity+0xcc/0xec >>> [] smack_inode_getsecctx+0x24/0x44 >>> [] security_inode_getsecctx+0x50/0x70 >>> [] kernfs_security_xattr_set+0x74/0xe0 >>> [] __vfs_setxattr+0x74/0x90 >>> [] __vfs_setxattr_noperm+0x80/0x1ac >>> [] vfs_setxattr+0x84/0xac >>> [] setxattr+0x114/0x178 >>> [] path_setxattr+0x74/0xb8 >>> [] SyS_lsetxattr+0x10/0x1c >>> [] __sys_trace_return+0x0/0x4 >>> >>> Signed-off-by: sireesha.t >>> Signed-off-by: CHANDAN VN >> Why not: >> >> static int smack_inode_getsecctx(struct inode *inode, void **ctx, u32 *ctxlen) >> { >> - int len = 0; >> - len = smack_inode_getsecurity(inode, XATTR_SMACK_SUFFIX, ctx, true); >> + int len = smack_inode_getsecurity(inode, XATTR_SMACK_SUFFIX, ctx, false); >> > The practical difference here is the true vs the false in the call > to smack_inode_getsecurity? That is correct. The author of smack_inode_getsecctx() has a SELinux background and appears to have missed that Smack is careful not to allocate memory and make copies of labels when it doesn't need to. > >> if (len < 0) >> return len; >> > Eric >