Date:   Tue, 28 Aug 2018 18:43:47 -0700
From:   Jaegeuk Kim <jaegeuk@...nel.org>
To:     Chao Yu <yuchao0@...wei.com>
Cc:     Chao Yu <chao@...nel.org>, Eric Biggers <ebiggers@...nel.org>,
        Dmitry Kasatkin <dmitry.kasatkin@...il.com>,
        Michael Halcrow <mhalcrow@...gle.com>,
        linux-kernel@...r.kernel.org,
        linux-f2fs-devel@...ts.sourceforge.net,
        linux-fscrypt@...r.kernel.org, linux-fsdevel@...r.kernel.org,
        linux-integrity@...r.kernel.org, linux-ext4@...r.kernel.org,
        Mimi Zohar <zohar@...ux.vnet.ibm.com>,
        Victor Hsieh <victorhsieh@...gle.com>
Subject: Re: [f2fs-dev] [RFC PATCH 10/10] f2fs: fs-verity support

On 08/29, Chao Yu wrote:
> On 2018/8/29 1:01, Jaegeuk Kim wrote:
> > On 08/28, Chao Yu wrote:
> >> On 2018/8/28 15:27, Jaegeuk Kim wrote:
> >>> On 08/27, Chao Yu wrote:
> >>>> Hi Eric,
> >>>>
> >>>> On 2018/8/27 1:35, Eric Biggers wrote:
> >>>>> Hi Chao,
> >>>>>
> >>>>> On Sat, Aug 25, 2018 at 01:54:08PM +0800, Chao Yu wrote:
> >>>>>> On 2018/8/25 0:16, Eric Biggers wrote:
> >>>>>>> From: Eric Biggers <ebiggers@...gle.com>
> >>>>>>>  #ifdef CONFIG_F2FS_CHECK_FS
> >>>>>>>  #define f2fs_bug_on(sbi, condition)	BUG_ON(condition)
> >>>>>>>  #else
> >>>>>>> @@ -146,7 +149,7 @@ struct f2fs_mount_info {
> >>>>>>>  #define F2FS_FEATURE_QUOTA_INO		0x0080
> >>>>>>>  #define F2FS_FEATURE_INODE_CRTIME	0x0100
> >>>>>>>  #define F2FS_FEATURE_LOST_FOUND		0x0200
> >>>>>>> -#define F2FS_FEATURE_VERITY		0x0400	/* reserved */
> >>>>>>> +#define F2FS_FEATURE_VERITY		0x0400
> >>>>>>>  
> >>>>>>>  #define F2FS_HAS_FEATURE(sb, mask)					\
> >>>>>>>  	((F2FS_SB(sb)->raw_super->feature & cpu_to_le32(mask)) != 0)
> >>>>>>> @@ -598,7 +601,7 @@ enum {
> >>>>>>>  #define FADVISE_ENC_NAME_BIT	0x08
> >>>>>>>  #define FADVISE_KEEP_SIZE_BIT	0x10
> >>>>>>>  #define FADVISE_HOT_BIT		0x20
> >>>>>>> -#define FADVISE_VERITY_BIT	0x40	/* reserved */
> >>>>>>> +#define FADVISE_VERITY_BIT	0x40
> >>>>>>
> >>>>>> As I suggested before, how about moving f2fs' verity_bit from i_fadvise to more
> >>>>>> generic i_flags field like ext4, so we can a) remaining more bits for those
> >>>>>> demands which really need file advise fields. b) using i_flags bits keeping line
> >>>>>> with ext4. Not sure, if user want to know whether the file is verity one, it
> >>>>>> will be easy for f2fs to export the status through FS_IOC_SETFLAGS.
> >>>>>>
> >>>>>> #define EXT4_VERITY_FL			0x00100000 /* Verity protected inode */
> >>>>>>
> >>>>>> #define F2FS_VERITY_FL			0x00100000 /* Verity protected inode */
> >>>>>>
> >>>>>
> >>>>> I don't like using i_advise much either, but I actually don't see either
> >>>>> location being much better than the other at the moment.  The real problem is an
> >>>>> artificial one: the i_flags in f2fs's on-disk format are being assumed to use
> >>>>
> >>>> Yeah, but since most copied flags from vfs/ext4 are not actually used in f2fs,
> >>>> also 0x00100000 bit is not used now, so we can just define it now directly for
> >>>> verity bit.
> >>>>
> >>>> Cleanup and remapping in ioctl interface for those unused flags, we can do it
> >>>> latter?
> >>>
> >>> No, it was reserved by f2fs-tools, 
> >>
> >> That's not a problem, since we didn't use that reserved bit in any of images
> >> now, there is no backward compatibility issue.
> > 
> > We're using that.
> 
> Oops, if it was in production, I agree to keep it in i_advice, otherwise, we
> still can discuss its location.
> 
> > 
> >>
> >>> and I think this should be aligned to the encryption bit. 
> >>
> >> Alright, we could, but if so, i_advise will run out of space earlier, after that
> >> we have to add real advice bit into i_inline or i_flags, that would be a little
> >> weird.
> >>
> >> For encryption bit, as a common vfs feature flag, in the beginning of encryption
> >> development, it will be better to set it into i_flags, IMO, but now, we have to
> >> keep it as it was.
> >>
> >>> Moreover, we guarantee i_flags less strictly from power-cut than i_advise.
> >>
> >> IMO, in power-cut scenario, it needs to keep both i_flags and i_advise being
> >> recoverable strictly. Any condition that we can not recover i_flags?
> > 
> > In __f2fs_ioc_setflags, f2fs_mark_inode_dirty_sync(inode, false);
> 
> Ah, that's right, do you remember why we treat them with different recoverable
> level?

Since I thought that such the flags wouln't be critical on power cuts, but be
enough for us to guarantee by write_inode() or fsync().

> 
> Thanks,
> 
> > 
> >>
> >> Thanks,
> >>
> >>>
> >>>>
> >>>> Thanks,
> >>>>
> >>>>> the same numbering scheme as ext4's on-disk format, which makes it seem that
> >>>>> they have to be in sync, and that all new ext4 flags (say, EA_INODE) also
> >>>>> reserve bits in f2fs and vice versa, when they in fact do not.  Instead, f2fs
> >>>>> should use its own numbering for its i_flags, and it should map them to/from
> >>>>> whatever is needed for common APIs like FS_IOC_{GET,SET}FLAGS and
> >>>>> FS_IOC_FS{GET,SET}XATTR.
> >>>>>
> >>>>> So putting the verity flag in *either* location (i_advise or i_flags) is just
> >>>>> kicking the can down the road.  If I get around to it I will send a patch that
> >>>>> cleans up the f2fs flags properly...>
> >>>>> Thanks,
> >>>>>
> >>>>> - Eric
> >>>>>
> >>>>> ------------------------------------------------------------------------------
> >>>>> Check out the vibrant tech community on one of the world's most
> >>>>> engaging tech sites, Slashdot.org! http://sdm.link/slashdot
> >>>>> _______________________________________________
> >>>>> Linux-f2fs-devel mailing list
> >>>>> Linux-f2fs-devel@...ts.sourceforge.net
> >>>>> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
> >>>>>
> >>>
> >>> .
> >>>
> > 
> > .
> > 

Powered by blists - more mailing lists