| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-id: <004f01d05aef$599b2dd0$0cd18970$@samsung.com>
Date: Tue, 10 Mar 2015 13:00:59 +0800
From: Chao Yu <chao2.yu@...sung.com>
To: 'Jaegeuk Kim' <jaegeuk@...nel.org>
Cc: 'Changman Lee' <cm224.lee@...sung.com>,
linux-f2fs-devel@...ts.sourceforge.net,
linux-kernel@...r.kernel.org
Subject: RE: [PATCH] f2fs: fix to truncate inline data past EOF
Hi Jaegeuk,
> -----Original Message-----
> From: Jaegeuk Kim [mailto:jaegeuk@...nel.org]
> Sent: Tuesday, March 10, 2015 11:00 AM
> To: Chao Yu
> Cc: 'Changman Lee'; linux-f2fs-devel@...ts.sourceforge.net; linux-kernel@...r.kernel.org
> Subject: Re: [PATCH] f2fs: fix to truncate inline data past EOF
>
> Hi Chao,
>
> On Tue, Mar 10, 2015 at 10:02:46AM +0800, Chao Yu wrote:
> > Hi Jaegeuk,
> >
>
> [snip]
>
> > > > > +static int truncate_partial_data_page(struct inode *inode, u64 from, bool force)
> > > > > {
> > > > > unsigned offset = from & (PAGE_CACHE_SIZE - 1);
> > > > > struct page *page;
> > > > >
> > > > > - if (!offset)
> > > > > + if (!offset && !force)
> > > >
> > > > We truncate on-disk inode page and #0 page separately, so IMO there is a very
> > > > small chance the following issue will occur.
> > > >
> > > > (0 < truncated_size < MAX_INLINE_SIZE)
> > > > ->f2fs_setattr
> > > > ->truncate_setsize #0 page has been truncated partially
> > > > ->f2fs_truncate
> > > > ->truncate_blocks
> > > > invalidate #0 page by reclaimer
> > > > update #0 page with original data in inode page by reader
> > >
> > > truncate_setsize called truncate_pagesize.
> > > so #0 page was truncated successfully.
> >
> > Yeah, but it's partially truncating as 0 < truncated_size < MAX_INLINE_SIZE,
> > After truncating, we still keep [0, truncated_size) valid data in #0 page, and
> > our #0 page status is uptodate | !dirty.
> >
> > So what I mean is that mm can reclaim this #0 page, then if someone else read
> > the #0 page again, we will update the #0 page with original data in inode page
> > since inode page haven't truncated yet.
>
> The truncate_blocks dropped inline_data in inode page, which is modified by
> this patch.
> And, the cached #0 page was truncated by truncate_setsize.
> Even if this page was evicted and reloaded again, the data would be filled with
> the inode page having truncated inline_data.
> So, it seems there is no problem.
>
> BTW, do you mean like this scenario?
Yeah, actually it is. ;-)
>
> -> f2fs_setattr
> ->truncate_setsize: #0 page has been truncated partially
> ->f2fs_truncate
> invalidate #0 page by reclaimer
> update #0 page with original data in inode page by reader
>
> ->truncate_blocks
> (*)-> truncate_inline_inode
> (*)-> truncate_partial_data_page(,, force)
> find_data_page(,, force) <- we can use *force* here.
>
> Then, I agree that two functions as noted (*) above would be necessary.
>
> >
> > If this happened, if we don't truncate #0 page in following code of truncate_blocks,
> > we will remain the original data for user, it's wrong.
> >
> > IMO, it's better to truncate inode page and #0 page together, or truncate #0 page
> > in truncate_partial_data_page.
> >
> > How do you think?
> >
> > >
> > > > ->truncate_inline_inode
> > > > ->truncate_partial_data_page
> > > > we will fail to truncate #0 page because we can't find valid blkaddr for
> > > > #0 page in find_data_page(,,false)
> > > >
> > > > How about using find_data_page(,,true) to check weather this page is update or not
> > > > for this condition.
> > >
> > > Oh, I realized that we don't need to call truncate_partial_data_page, since the
> > > cached #0 page was truncated already. We don't care about that.
> >
> > IMO, we should care about this #0 page if above example can happen.
> >
> > > So, do we need to add just truncate_inline_inode() like below?
> > >
> > > >
> > > > > return 0;
> > > > >
> > > > > page = find_data_page(inode, from >> PAGE_CACHE_SHIFT, false);
> > > > > @@ -489,6 +489,7 @@ int truncate_blocks(struct inode *inode, u64 from, bool lock)
> > > > > pgoff_t free_from;
> > > > > int count = 0, err = 0;
> > > > > struct page *ipage;
> > > > > + bool truncate_page = false;
> > > > >
> > > > > trace_f2fs_truncate_blocks_enter(inode, from);
> > > > >
> > > > > @@ -504,6 +505,9 @@ int truncate_blocks(struct inode *inode, u64 from, bool lock)
> > > > > }
> > > > >
> > > > > if (f2fs_has_inline_data(inode)) {
> > > > > + truncate_inline_inode(ipage, from);
> > > > > + set_page_dirty(ipage);
> > > >
> > > > If @from is the same as MAX_INLINE_DATA, we will change nothing in inode page,
> > > > How about skipping set_page_dirty for inode page in this condition?
> > >
> > > Agreed.
> > > How about adding this in truncate_inline_inode?
> > >
> > > if (from >= MAX_INLINE_DATA)
> > > return;
> > > ...
> > > set_page_dirty(ipage);
> >
> > Yeah, that's good.
> >
> > And What I suggest is:
> >
> > bool truncate_inline_inode(struct page *ipage, u64 from)
> > {
> > /*
> > * we should never truncate inline data past max inline data size,
> > * because we always convert inline inode to normal one before
> > * truncating real data if new size is past max inline data size.
> > */
> > f2fs_bug_on(F2FS_P_SB(ipage), from > MAX_INLINE_DATA);
> >
> > if (from == MAX_INLINE_DATA)
>
> if (from >= MAX_INLINE_DATA) to handle when f2fs_bug_on is bypassed.
Agreed, we should handle this condition.
>
> > return false;
> > ...
> > return true;
> > }
> >
> > in truncate_blocks()
> >
> > if (f2fs_has_inline_data(inode)) {
> > if (truncate_inline_inode(ipage, from))
> > set_page_dirty(ipage);
> >
> > So by this way, we can distinguish between a bug and invalid truncating in
> > truncate_inline_inode, and also we can avoid unneeded set_dirty_page for inode
> > page in other call path.
>
> Ok, it's not a big deal.
Thanks for your help! Let me send a v2 patch.
Regards,
Chao
[sinp]
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists