Message-id: <004501d05ad6$6cd7af40$46870dc0$@samsung.com>
Date:	Tue, 10 Mar 2015 10:02:46 +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 12:16 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 Mon, Mar 09, 2015 at 04:30:32PM +0800, Chao Yu wrote:
> > Hi Jaegeuk,
> >
> > > -----Original Message-----
> > > From: Jaegeuk Kim [mailto:jaegeuk@...nel.org]
> > > Sent: Monday, March 09, 2015 11:49 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 Mon, Mar 09, 2015 at 10:24:35AM +0800, Chao Yu wrote:
> > > > Hi Jaegeuk,
> > > >
> > > > > -----Original Message-----
> > > > > From: Jaegeuk Kim [mailto:jaegeuk@...nel.org]
> > > > > Sent: Friday, March 06, 2015 5:34 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 Thu, Mar 05, 2015 at 05:41:32PM +0800, Chao Yu wrote:
> > > > > > Previously if inode is with inline data, we will try to invalid partial inline
> > > > > > data in page #0 when we truncate size of inode in truncate_partial_data_page().
> > > > > > And then we set page #0 to dirty, after this we can synchronize inode page with
> > > > > > page #0 at ->writepage().
> > > > > >
> > > > > > But sometimes we will fail to operate page #0 in truncate_partial_data_page()
> > > > > > due to below reason:
> > > > > > a) if offset is zero, we will skip setting page #0 to dirty.
> > > > > > b) if page #0 is not uptodate, we will fail to update it as it has no mapping
> > > > > > data.
> > > > > >
> > > > > > So with following operations, we will meet recent data which should be
> > > > > > truncated.
> > > > > >
> > > > > > 1.write inline data to file
> > > > > > 2.sync first data page to inode page
> > > > > > 3.truncate file size to 0
> > > > > > 4.truncate file size to max_inline_size
> > > > > > 5.echo 1 > /proc/sys/vm/drop_caches
> > > > > > 6.read file --> meet original inline data which is remained in inode page.
> > > > > >
> > > > > > This patch renames truncate_inline_data() to truncate_inline_inode() for code
> > > > > > readability, then intruduces new truncate_inline_data() and use it to truncate
> > > > > > inline data in page #0 and inode page in truncate_partial_data_page() for
> > > > > > fixing.
> > > > > >
> > > > > > Signed-off-by: Chao Yu <chao2.yu@...sung.com>
> > > > > > ---
> > > > > >  fs/f2fs/f2fs.h   |  1 +
> > > > > >  fs/f2fs/file.c   | 13 ++++++++-----
> > > > > >  fs/f2fs/inline.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++++++-----
> > > > > >  3 files changed, 61 insertions(+), 10 deletions(-)
> > > > > >
> > > > > > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> > > > > > index 511d6cd..55dbbe7 100644
> > > > > > --- a/fs/f2fs/f2fs.h
> > > > > > +++ b/fs/f2fs/f2fs.h
> > > > > > @@ -1754,6 +1754,7 @@ extern struct kmem_cache *inode_entry_slab;
> > > > > >   */
> > > > > >  bool f2fs_may_inline(struct inode *);
> > > > > >  void read_inline_data(struct page *, struct page *);
> > > > > > +int truncate_inline_data(struct inode *, u64 from);
> > > > > >  int f2fs_read_inline_data(struct inode *, struct page *);
> > > > > >  int f2fs_convert_inline_page(struct dnode_of_data *, struct page *);
> > > > > >  int f2fs_convert_inline_inode(struct inode *);
> > > > > > diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> > > > > > index f1341c7..6c4443e 100644
> > > > > > --- a/fs/f2fs/file.c
> > > > > > +++ b/fs/f2fs/file.c
> > > > > > @@ -461,6 +461,9 @@ static int truncate_partial_data_page(struct inode *inode, u64
> from)
> > > > > >  	unsigned offset = from & (PAGE_CACHE_SIZE - 1);
> > > > > >  	struct page *page;
> > > > > >
> > > > > > +	if (f2fs_has_inline_data(inode))
> > > > > > +		return 0;
> > > > > > +
> > > > > >  	if (!offset)
> > > > > >  		return 0;
> > > > > >
> > > > > > @@ -497,14 +500,14 @@ int truncate_blocks(struct inode *inode, u64 from, bool lock)
> > > > > >  	if (lock)
> > > > > >  		f2fs_lock_op(sbi);
> > > > > >
> > > > > > -	ipage = get_node_page(sbi, inode->i_ino);
> > > > > > -	if (IS_ERR(ipage)) {
> > > > > > -		err = PTR_ERR(ipage);
> > > > > > +	if (f2fs_has_inline_data(inode)) {
> > > > > > +		err = truncate_inline_data(inode, from);
> > > > > >  		goto out;
> > > > > >  	}
> > > > > >
> > > > > > -	if (f2fs_has_inline_data(inode)) {
> > > > > > -		f2fs_put_page(ipage, 1);
> > > > > > +	ipage = get_node_page(sbi, inode->i_ino);
> > > > > > +	if (IS_ERR(ipage)) {
> > > > > > +		err = PTR_ERR(ipage);
> > > > > >  		goto out;
> > > > > >  	}
> > > > > >
> > > > > > diff --git a/fs/f2fs/inline.c b/fs/f2fs/inline.c
> > > > > > index 4ba9732..a8189c92 100644
> > > > > > --- a/fs/f2fs/inline.c
> > > > > > +++ b/fs/f2fs/inline.c
> > > > > > @@ -50,10 +50,57 @@ void read_inline_data(struct page *page, struct page *ipage)
> > > > > >  	SetPageUptodate(page);
> > > > > >  }
> > > > > >
> > > > > > -static void truncate_inline_data(struct page *ipage)
> > > > > > +static void truncate_inline_inode(struct page *ipage, u64 from)
> > > > > >  {
> > > > > > +	void *addr = inline_data_addr(ipage);
> > > > > > +
> > > > > >  	f2fs_wait_on_page_writeback(ipage, NODE);
> > > > > > -	memset(inline_data_addr(ipage), 0, MAX_INLINE_DATA);
> > > > > > +	memset(addr + from, 0, MAX_INLINE_DATA - from);
> > > > > > +}
> > > > > > +
> > > > > > +int truncate_inline_data(struct inode *inode, u64 from)
> > > > > > +{
> > > > > > +	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
> > > > > > +	struct page *page, *ipage;
> > > > > > +	void *addr;
> > > > > > +	int err = 0;
> > > > > > +
> > > > > > +	/*
> > > > > > +	 * 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(sbi, from > MAX_INLINE_DATA);
> > > > > > +
> > > > > > +	if (from == MAX_INLINE_DATA)
> > > > > > +		return 0;
> > > > > > +
> > > > > > +	page = grab_cache_page(inode->i_mapping, 0);
> > > > > > +	if (!page)
> > > > > > +		return -ENOMEM;
> > > > >
> > > > > Do we need to grab and read #0 page?
> > > >
> > > > No need to read #0 page, sorry for the mistake.
> > > >
> > > > > How about checking its cached page after on-disk truncation?
> > > > >
> > > > > 1. page = get_node_page()
> > > > > 2. check_inline
> > > > > 3. truncate_inline_inode
> > > > > 4. f2fs_put_page(ipage)
> > > > >
> > > > > 5. truncate_partial_data_page(inode, from, force)
> > > > > 	if (!offset && !force)
> > > > >  		return 0;
> > > > >
> > > > > > +
> > > > > > +	ipage = get_node_page(sbi, inode->i_ino);
> > > > > > +	if (IS_ERR(ipage)) {
> > > > > > +		err = PTR_ERR(ipage);
> > > > > > +		goto out;
> > > > > > +	}
> > > > > > +
> > > > > > +	if (!f2fs_has_inline_data(inode)) {
> > > > > > +		f2fs_put_page(ipage, 1);
> > > > > > +		goto out;
> > > > > > +	}
> > > > > > +
> > > > > > +	truncate_inline_inode(ipage, from);
> > > > > > +	set_page_dirty(ipage);
> > > > > > +
> > > > > > +	read_inline_data(page, ipage);
> > > > > > +	f2fs_put_page(ipage, 1);
> > > > > > +
> > > > > > +	addr = inline_data_addr(page);
> > > > > > +	memset(addr + from, 0, MAX_INLINE_DATA - from);
> > > > >
> > > > > It needs to do likewise truncate_partial_data_page.
> > > >
> > > > Do you mean like this?
> > >
> > > I meant this.
> > >
> > > Anyway, I also wrote a patch about your testing scenario wrt inline_data.
> > > So, let me submit the patch sooner or later to xfstests mailing list.
> >
> > Sounds good! :)
> >
> > >
> > > Thanks,
> > >
> > > ---
> > >  fs/f2fs/f2fs.h   |  1 +
> > >  fs/f2fs/file.c   | 10 +++++++---
> > >  fs/f2fs/inline.c | 12 +++++++-----
> > >  3 files changed, 15 insertions(+), 8 deletions(-)
> > >
> > > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> > > index 31b440e..b222b3a 100644
> > > --- a/fs/f2fs/f2fs.h
> > > +++ b/fs/f2fs/f2fs.h
> > > @@ -1760,6 +1760,7 @@ extern struct kmem_cache *inode_entry_slab;
> > >   */
> > >  bool f2fs_may_inline(struct inode *);
> > >  void read_inline_data(struct page *, struct page *);
> > > +void truncate_inline_inode(struct page *, u64);
> > >  int f2fs_read_inline_data(struct inode *, struct page *);
> > >  int f2fs_convert_inline_page(struct dnode_of_data *, struct page *);
> > >  int f2fs_convert_inline_inode(struct inode *);
> > > diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> > > index f1341c7..ab249dd 100644
> > > --- a/fs/f2fs/file.c
> > > +++ b/fs/f2fs/file.c
> > > @@ -456,12 +456,12 @@ void truncate_data_blocks(struct dnode_of_data *dn)
> > >  	truncate_data_blocks_range(dn, ADDRS_PER_BLOCK);
> > >  }
> > >
> > > -static int truncate_partial_data_page(struct inode *inode, u64 from)
> > > +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.

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)
		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.

How do you think of this?

> 
> Thanks,
> 
> Signed-off-by: Jaegeuk Kim <jaegeuk@...nel.org>
> ---
>  fs/f2fs/f2fs.h   |  1 +
>  fs/f2fs/file.c   |  1 +
>  fs/f2fs/inline.c | 16 +++++++++++-----
>  3 files changed, 13 insertions(+), 5 deletions(-)
> 
> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> index 511d6cd..5cd3a2f 100644
> --- a/fs/f2fs/f2fs.h
> +++ b/fs/f2fs/f2fs.h
> @@ -1754,6 +1754,7 @@ extern struct kmem_cache *inode_entry_slab;
>   */
>  bool f2fs_may_inline(struct inode *);
>  void read_inline_data(struct page *, struct page *);
> +void truncate_inline_inode(struct page *, u64);
>  int f2fs_read_inline_data(struct inode *, struct page *);
>  int f2fs_convert_inline_page(struct dnode_of_data *, struct page *);
>  int f2fs_convert_inline_inode(struct inode *);
> diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> index f1341c7..b6b6a04 100644
> --- a/fs/f2fs/file.c
> +++ b/fs/f2fs/file.c
> @@ -504,6 +504,7 @@ int truncate_blocks(struct inode *inode, u64 from, bool lock)
>  	}
> 
>  	if (f2fs_has_inline_data(inode)) {
> +		truncate_inline_inode(ipage, from);
>  		f2fs_put_page(ipage, 1);
>  		goto out;
>  	}
> diff --git a/fs/f2fs/inline.c b/fs/f2fs/inline.c
> index 4ba9732..40bbf96 100644
> --- a/fs/f2fs/inline.c
> +++ b/fs/f2fs/inline.c
> @@ -50,10 +50,16 @@ void read_inline_data(struct page *page, struct page *ipage)
>  	SetPageUptodate(page);
>  }
> 
> -static void truncate_inline_data(struct page *ipage)
> +void truncate_inline_inode(struct page *ipage, u64 from)
>  {
> +	void *addr = inline_data_addr(ipage);
> +
> +	if (from >= MAX_INLIE_DATA)
> +		return;
> +
>  	f2fs_wait_on_page_writeback(ipage, NODE);
> -	memset(inline_data_addr(ipage), 0, MAX_INLINE_DATA);
> +	memset(addr + from, 0, MAX_INLINE_DATA - from);
> +	set_page_dirty(ipage);
>  }
> 
>  int f2fs_read_inline_data(struct inode *inode, struct page *page)
> @@ -131,7 +137,7 @@ no_update:
>  	set_inode_flag(F2FS_I(dn->inode), FI_APPEND_WRITE);
> 
>  	/* clear inline data and flag after data writeback */
> -	truncate_inline_data(dn->inode_page);
> +	truncate_inline_inode(dn->inode_page, 0);
>  clear_out:
>  	stat_dec_inline_inode(dn->inode);
>  	f2fs_clear_inline_inode(dn->inode);
> @@ -245,7 +251,7 @@ process_inline:
>  	if (f2fs_has_inline_data(inode)) {
>  		ipage = get_node_page(sbi, inode->i_ino);
>  		f2fs_bug_on(sbi, IS_ERR(ipage));
> -		truncate_inline_data(ipage);
> +		truncate_inline_inode(ipage, 0);
>  		f2fs_clear_inline_inode(inode);
>  		update_inode(inode, ipage);
>  		f2fs_put_page(ipage, 1);
> @@ -363,7 +369,7 @@ static int f2fs_convert_inline_dir(struct inode *dir, struct page *ipage,
>  	set_page_dirty(page);
> 
>  	/* clear inline dir and flag after data writeback */
> -	truncate_inline_data(ipage);
> +	truncate_inline_inode(ipage, 0);
> 
>  	stat_dec_inline_dir(dir);
>  	clear_inode_flag(F2FS_I(dir), FI_INLINE_DENTRY);
> --
> 2.1.1

--
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