From mboxrd@z Thu Jan 1 00:00:00 1970 From: Edward Shishkin Subject: Re: [PATCH] btrfs: implement FS_IOC_GETFLAGS/SETFLAGS/GETVERSION Date: Tue, 21 Apr 2009 23:30:54 +0200 Message-ID: <49EE3B0E.90706@gmail.com> References: <20090417083741.GA14933@lst.de> <49E855EB.5010001@gmail.com> <20090420154725.GA14284@lst.de> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Cc: linux-btrfs@vger.kernel.org To: Christoph Hellwig Return-path: In-Reply-To: <20090420154725.GA14284@lst.de> List-ID: Christoph Hellwig wrote: > On Fri, Apr 17, 2009 at 12:11:55PM +0200, Edward Shishkin wrote: > >> Christoph Hellwig wrote: >> >>> Add support for the standard attributes set via chattr and read vis >>> lsattr. Currently we store the attributes in the flags value in >>> the btrfs inode, but I wonder whether we should split it into two so >>> that we don't have to keep converting between the two formats. >>> >>> >> Imho, since inode items are of fixed size, is won't be possible >> to avoid such workarounds like conversion between formats. >> No? >> > > While the inode format is fixed it has 256 spare bits for expansion. > Ah, I meant the case when the spare bits are exhausted.. > But what I mean with the above is to split the current 64bit flags value > into a a 32 bit internal flags and a 32bit user visible flags value > and store the ioctl flags in the latter. > > OTOH every filesystem but extN seem to need some conversion so btrfs > wouldn't be unusual at that. Not sure about extN, but one of the techniques is to represent inode item as a set of (optional) "extensions", so that every such extension includes it's version number (think of it as release date). If file system driver is older, then some encountered extension, init_inode() returns error. OTOH, if some extension is missed, then some featured operations can be undefined (i.e. read(), or write(), etc.. will return error). No conversion is needed, however such approach requires more sophisticated fsck. > And the GETFLAGS/SETFLAGS flags value > are pretty ugly anyway as they mix up flags for user visible behaviour > with extN implementation details that shouldn't really need to be > exposed to userspace. > > >