Hi Jaegeuk,
> -----Original Message-----
> From: Jaegeuk Kim [mailto:jaeg...@kernel.org]
> Sent: Tuesday, January 26, 2016 3:18 AM
> To: Chao Yu
> Cc: linux-kernel@vger.kernel.org; linux-fsde...@vger.kernel.org;
> linux-f2fs-de...@lists.sourceforge.net
> Subject: Re: [f2fs-dev] [PATCH 1/2] f2fs: avoid multiple node page writes due
> to inline_data
>
> Hi Chao,
>
> On Mon, Jan 25, 2016 at 05:42:40PM +0800, Chao Yu wrote:
> > Hi Jaegeuk,
> >
> > > -----Original Message-----
> > > From: Jaegeuk Kim [mailto:jaeg...@kernel.org]
> > > Sent: Sunday, January 24, 2016 4:16 AM
> > > To: linux-kernel@vger.kernel.org; linux-fsde...@vger.kernel.org;
> > > linux-f2fs-de...@lists.sourceforge.net
> > > Cc: Jaegeuk Kim
> > > Subject: [f2fs-dev] [PATCH 1/2] f2fs: avoid multiple node page writes due
> > > to inline_data
> > >
> > > The sceanrio is:
> > > 1. create fully node blocks
> > > 2. flush node blocks
> > > 3. write inline_data for all the node blocks again
> > > 4. flush node blocks redundantly
> > >
> > > Signed-off-by: Jaegeuk Kim <jaeg...@kernel.org>
> > > ---
> > > fs/f2fs/data.c | 14 +++++++++++---
> > > 1 file changed, 11 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> > > index 8d0d9ec..011456e 100644
> > > --- a/fs/f2fs/data.c
> > > +++ b/fs/f2fs/data.c
> > > @@ -1622,14 +1622,22 @@ static int f2fs_write_end(struct file *file,
> > >
> > > trace_f2fs_write_end(inode, pos, len, copied);
> > >
> > > - set_page_dirty(page);
> > > -
> > > if (pos + copied > i_size_read(inode)) {
> > > i_size_write(inode, pos + copied);
> > > mark_inode_dirty(inode);
> > > - update_inode_page(inode);
> > > }
> > >
> > > + if (f2fs_has_inline_data(inode) &&
> > > + is_inode_flag_set(F2FS_I(inode), FI_DATA_EXIST)) {
> > > + int err = f2fs_write_inline_data(inode, page);
> >
> > Oh, I'm sure this can fix that issue, but IMO:
> > a) this implementation has side-effect, it triggers inline data copying
> > between data page and node page whenever user write inline datas, so if
> > user updates inline data frequently, write-through approach would cause
> > memory copy overhead.
>
> Agreed.
>
> > b) inline storm should be a rare case, as we didn't get any report about
> > problem for long time until Dave's, and write_end is a hot path, I think
> > it's better to be cautious to change our inline data cache policy for
> > fixing a rare issue in hot path.
> >
> > What about delaying the merge operation? like:
> > 1) as I proposed before, merging inline page into inode page when
> > detecting free_sections <= (node_secs + 2 * dent_secs + inline_secs).
> > 2) merge inline page into inode page before writeback inode page in
> > sync_node_pages.
>
> Okay, I'm thinking more general way where we can get rid of every inlien_data
> write when we flush node pages.
I encountered deadlock issue, could you have a look at it?
======================================================
[ INFO: possible circular locking dependency detected ]
4.5.0-rc1 #45 Tainted: G O
-------------------------------------------------------
fstrim/15301 is trying to acquire lock:
(sb_internal#2){++++..}, at: [<ffffffff81216fca>] __sb_start_write+0xda/0xf0
but task is already holding lock:
(&sbi->cp_rwsem){++++..}, at: [<ffffffffa07d06d2>]
block_operations+0x82/0x130 [f2fs]
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #1 (&sbi->cp_rwsem){++++..}:
[<ffffffff810bf827>] lock_acquire+0xb7/0x130
[<ffffffff817de829>] down_read+0x39/0x50
[<ffffffffa07c27af>] f2fs_evict_inode+0x26f/0x370 [f2fs]
[<ffffffff812326cd>] evict+0xdd/0x1d0
[<ffffffff8123323f>] iput+0x19f/0x250
[<ffffffff81224d9d>] do_unlinkat+0x20d/0x310
[<ffffffff81224ee2>] SyS_unlinkat+0x22/0x40
[<ffffffff817e0957>] entry_SYSCALL_64_fastpath+0x12/0x6f
-> #0 (sb_internal#2){++++..}:
[<ffffffff810bf32b>] __lock_acquire+0x132b/0x1770
[<ffffffff810bf827>] lock_acquire+0xb7/0x130
[<ffffffff810b8fac>] percpu_down_read+0x3c/0x80
[<ffffffff81216fca>] __sb_start_write+0xda/0xf0
[<ffffffffa07c2761>] f2fs_evict_inode+0x221/0x370 [f2fs]
[<ffffffff812326cd>] evict+0xdd/0x1d0
[<ffffffff8123323f>] iput+0x19f/0x250
[<ffffffffa07dd4d3>] sync_node_pages+0x703/0x900 [f2fs]
[<ffffffffa07d075a>] block_operations+0x10a/0x130 [f2fs]
[<ffffffffa07d13e4>] write_checkpoint+0xc4/0xb80 [f2fs]
[<ffffffffa07e0af2>] f2fs_trim_fs+0x122/0x1d0 [f2fs]
[<ffffffffa07c07da>] f2fs_ioctl+0x7fa/0x9d0 [f2fs]
[<ffffffff81228448>] vfs_ioctl+0x18/0x40
[<ffffffff81228b96>] do_vfs_ioctl+0x96/0x680
[<ffffffff81229212>] SyS_ioctl+0x92/0xa0
[<ffffffff817e0957>] entry_SYSCALL_64_fastpath+0x12/0x6f
other info that might help us debug this:
Possible unsafe locking scenario:
CPU0 CPU1
---- ----
lock(&sbi->cp_rwsem);
lock(sb_internal#2);
lock(&sbi->cp_rwsem);
lock(sb_internal#2);
*** DEADLOCK ***
Thanks,
>
> I've been testing this patch.
>
> From ebddf607c64da691fef08cf68a8ecadafd5d896b Mon Sep 17 00:00:00 2001
> From: Jaegeuk Kim <jaeg...@kernel.org>
> Date: Mon, 25 Jan 2016 05:57:05 -0800
> Subject: [PATCH] f2fs: avoid multiple node page writes due to inline_data
>
> The sceanrio is:
> 1. create fully node blocks
> 2. flush node blocks
> 3. write inline_data for all the node blocks again
> 4. flush node blocks redundantly
>
> So, this patch tries to flush inline_data when flushing node blocks.
>
> Signed-off-by: Jaegeuk Kim <jaeg...@kernel.org>
> ---
> fs/f2fs/data.c | 1 +
> fs/f2fs/inline.c | 2 ++
> fs/f2fs/node.c | 35 +++++++++++++++++++++++++++++++++++
> fs/f2fs/node.h | 15 +++++++++++++++
> 4 files changed, 53 insertions(+)
>
> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> index 6925c10..9043ecf 100644
> --- a/fs/f2fs/data.c
> +++ b/fs/f2fs/data.c
> @@ -1464,6 +1464,7 @@ restart:
> if (pos + len <= MAX_INLINE_DATA) {
> read_inline_data(page, ipage);
> set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
> + set_inline_node(ipage);
> sync_inode_page(&dn);
> } else {
> err = f2fs_convert_inline_page(&dn, page);
> diff --git a/fs/f2fs/inline.c b/fs/f2fs/inline.c
> index 8df13e5..fc4d298 100644
> --- a/fs/f2fs/inline.c
> +++ b/fs/f2fs/inline.c
> @@ -159,6 +159,7 @@ no_update:
>
> /* clear inline data and flag after data writeback */
> truncate_inline_inode(dn->inode_page, 0);
> + clear_inline_node(dn->inode_page);
> clear_out:
> stat_dec_inline_inode(dn->inode);
> f2fs_clear_inline_inode(dn->inode);
> @@ -233,6 +234,7 @@ int f2fs_write_inline_data(struct inode *inode, struct
> page *page)
> set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
>
> sync_inode_page(&dn);
> + clear_inline_node(dn.inode_page);
> f2fs_put_dnode(&dn);
> return 0;
> }
> diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
> index 23b800d..1c5023e 100644
> --- a/fs/f2fs/node.c
> +++ b/fs/f2fs/node.c
> @@ -1154,6 +1154,33 @@ void sync_inode_page(struct dnode_of_data *dn)
> dn->node_changed = ret ? true: false;
> }
>
> +static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
> +{
> + struct inode *inode;
> + struct page *page;
> +
> + inode = ilookup(sbi->sb, ino);
> + if (!inode)
> + return;
> +
> + page = find_lock_page(inode->i_mapping, 0);
> + if (!page)
> + goto iput_out;
> +
> + if (!PageDirty(page))
> + goto put_page_out;
> +
> + if (!clear_page_dirty_for_io(page))
> + goto put_page_out;
> +
> + if (!f2fs_write_inline_data(inode, page))
> + inode_dec_dirty_pages(inode);
> +put_page_out:
> + f2fs_put_page(page, 1);
> +iput_out:
> + iput(inode);
> +}
> +
> int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
> struct writeback_control *wbc)
> {
> @@ -1221,6 +1248,14 @@ continue_unlock:
> goto continue_unlock;
> }
>
> + /* flush inline_data */
> + if (!ino && is_inline_node(page)) {
> + clear_inline_node(page);
> + unlock_page(page);
> + flush_inline_data(sbi, ino_of_node(page));
> + continue;
> + }
> +
> if (!clear_page_dirty_for_io(page))
> goto continue_unlock;
>
> diff --git a/fs/f2fs/node.h b/fs/f2fs/node.h
> index 23bd992..1f4f9d4 100644
> --- a/fs/f2fs/node.h
> +++ b/fs/f2fs/node.h
> @@ -379,6 +379,21 @@ static inline int is_node(struct page *page, int type)
> #define is_fsync_dnode(page) is_node(page, FSYNC_BIT_SHIFT)
> #define is_dent_dnode(page) is_node(page, DENT_BIT_SHIFT)
>
> +static inline int is_inline_node(struct page *page)
> +{
> + return PageChecked(page);
> +}
> +
> +static inline void set_inline_node(struct page *page)
> +{
> + SetPageChecked(page);
> +}
> +
> +static inline void clear_inline_node(struct page *page)
> +{
> + ClearPageChecked(page);
> +}
> +
> static inline void set_cold_node(struct inode *inode, struct page *page)
> {
> struct f2fs_node *rn = F2FS_NODE(page);
> --
> 2.6.3
>
