From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from out02.mta.xmission.com ([166.70.13.232]:52949 "EHLO out02.mta.xmission.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752034AbcHIAlL (ORCPT ); Mon, 8 Aug 2016 20:41:11 -0400 From: ebiederm@xmission.com (Eric W. Biederman) To: Nikolaus Rath Cc: fuse-devel@lists.sourceforge.net, linux-fsdevel@vger.kernel.org, Miklos Szeredi , Michael j Theall , Jean-Pierre =?utf-8?Q?Andr=C3=A9?= , Seth Forshee References: <1470086846-19844-1-git-send-email-seth.forshee@canonical.com> <874m6u3j1p.fsf@thinkpad.rath.org> Date: Mon, 08 Aug 2016 19:27:50 -0500 In-Reply-To: <874m6u3j1p.fsf@thinkpad.rath.org> (Nikolaus Rath's message of "Mon, 08 Aug 2016 17:03:30 -0700") Message-ID: <87popilrax.fsf@x220.int.ebiederm.org> MIME-Version: 1.0 Content-Type: text/plain Subject: Re: [RFC v3 0/2] Support for posix acls in fuse Sender: linux-fsdevel-owner@vger.kernel.org List-ID: Nikolaus Rath writes: > On Aug 01 2016, Seth Forshee wrote: >> - Remove passthrough of acl xattrs when fuse acl support is disabled or >> default_permissions is not used. >> >> This last change is user visible, but as fuse filesystems cannot >> meaninfully support acls today it's not really a regression. > > Are you sure about that? I believe there are FUSE file systems out there > that are parsing/constructing the kernel's xattr representation and > (together with no_default_permissions) support ACLs. Or is there another > problem? fuse_permission does not have a mode where it always call into the filesystem. Without FUSE_DEFAULT_PERMISSIONS set the underlying filesystem is at most called when the syscalls chdir, access, and execve are called. (Basically Which means there is no way to enforce any kind of general acls in fuse without changes. That said I we seem to have figured out an implmenetation where passthrough is maintained for the time being when posix acl support is not enabled. And Miklos figures libfuse needs to parse the the xattr anyway so that the filesystems can have atomic mode changes instead of having two separate calls, one to setattr and another to setxattr. So I don't believe when the dust settles there is any danger of regression, despite the code not yet working in a way that enforces acls. Eric