Re: [f2fs-dev] [PATCH] f2fs: prevent atomic file from being dirtied before commit

[Daeho Jeong]

On Wed, Sep 4, 2024 at 7:56 AM Daeho Jeong <daeho43@xxxxxxxxx> wrote:
>
> On Tue, Sep 3, 2024 at 8:35 PM Chao Yu <chao@xxxxxxxxxx> wrote:
> >
> > On 2024/9/4 10:52, Daeho Jeong wrote:
> > > On Tue, Sep 3, 2024 at 7:26 PM Chao Yu <chao@xxxxxxxxxx> wrote:
> > >>
> > >> On 2024/9/4 1:07, Daeho Jeong wrote:
> > >>> On Mon, Sep 2, 2024 at 3:08 AM Chao Yu <chao@xxxxxxxxxx> wrote:
> > >>>>
> > >>>> On 2024/8/27 4:23, Daeho Jeong wrote:
> > >>>>> From: Daeho Jeong <daehojeong@xxxxxxxxxx>
> > >>>>>
> > >>>>> Keep atomic file clean while updating and make it dirtied during commit
> > >>>>> in order to avoid unnecessary and excessive inode updates in the previous
> > >>>>> fix.
> > >>>>>
> > >>>>> Fixes: 4bf78322346f ("f2fs: mark inode dirty for FI_ATOMIC_COMMITTED flag")
> > >>>>> Signed-off-by: Daeho Jeong <daehojeong@xxxxxxxxxx>
> > >>>>> ---
> > >>>>>     fs/f2fs/f2fs.h    |  3 +--
> > >>>>>     fs/f2fs/inode.c   | 10 ++++++----
> > >>>>>     fs/f2fs/segment.c | 10 ++++++++--
> > >>>>>     3 files changed, 15 insertions(+), 8 deletions(-)
> > >>>>>
> > >>>>> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> > >>>>> index 465b2fd50c70..5a7f6fa8b585 100644
> > >>>>> --- a/fs/f2fs/f2fs.h
> > >>>>> +++ b/fs/f2fs/f2fs.h
> > >>>>> @@ -801,7 +801,7 @@ enum {
> > >>>>>         FI_COMPRESS_RELEASED,   /* compressed blocks were released */
> > >>>>>         FI_ALIGNED_WRITE,       /* enable aligned write */
> > >>>>>         FI_COW_FILE,            /* indicate COW file */
> > >>>>> -     FI_ATOMIC_COMMITTED,    /* indicate atomic commit completed except disk sync */
> > >>>>> +     FI_ATOMIC_DIRTIED,      /* indicate atomic file is dirtied */
> > >>>>>         FI_ATOMIC_REPLACE,      /* indicate atomic replace */
> > >>>>>         FI_OPENED_FILE,         /* indicate file has been opened */
> > >>>>>         FI_MAX,                 /* max flag, never be used */
> > >>>>> @@ -3042,7 +3042,6 @@ static inline void __mark_inode_dirty_flag(struct inode *inode,
> > >>>>>         case FI_INLINE_DOTS:
> > >>>>>         case FI_PIN_FILE:
> > >>>>>         case FI_COMPRESS_RELEASED:
> > >>>>> -     case FI_ATOMIC_COMMITTED:
> > >>>>>                 f2fs_mark_inode_dirty_sync(inode, true);
> > >>>>>         }
> > >>>>>     }
> > >>>>> diff --git a/fs/f2fs/inode.c b/fs/f2fs/inode.c
> > >>>>> index 1eb250c6b392..5dd3e55d2be2 100644
> > >>>>> --- a/fs/f2fs/inode.c
> > >>>>> +++ b/fs/f2fs/inode.c
> > >>>>> @@ -35,6 +35,11 @@ void f2fs_mark_inode_dirty_sync(struct inode *inode, bool sync)
> > >>>>>         if (f2fs_inode_dirtied(inode, sync))
> > >>>>
> > >>>> It will return directly here if inode was dirtied, so it may missed to set
> > >>>> FI_ATOMIC_DIRTIED flag?
> > >>>
> > >>> Is it possible for it to be already dirty, since we already made it
> > >>> clean with f2fs_write_inode() when we started the atomic write?
> > >>
> > >> Some ioctl interfaces may race w/ atomic write? e.g. set_pin_file won't
> > >> check atomic_file status, and may dirty inode after we started atomic
> > >> write, so we'd better detect such race condition and break ioctl to
> > >> avoid ruin atomic write? and maybe we can add f2fs_bug_on() in
> > >> f2fs_mark_inode_dirty_sync() to detect any other missing cases?
> > >>
> > >
> > > How about exchanging the positions of f2fs_write_inode() and
> > > set_inode_flag() in f2fs_ioc_start_atomic_write()?
> > >
> > > ...
> > >          f2fs_write_inode(inode, NULL);
> > >
> > >          stat_inc_atomic_inode(inode);
> > >
> > >          set_inode_flag(inode, FI_ATOMIC_FILE);
> > > ...
> >
> > Oh, I'm not sure I've got your point, after exchanging we still may suffer
> > below race condition, right?
> >
> > - f2fs_ioc_start_atomic_write
> >   - set_inode_flag(inode, FI_ATOMIC_FILE)
> >   - f2fs_write_inode(inode, NULL)
> >                                         - f2fs_ioc_set_pin_file
> >                                          - set_inode_flag(inode, FI_PIN_FILE)
> >                                           - __mark_inode_dirty_flag()
>                                                  => This attempt will
> be blocked by the below condition.
>
> +       if (f2fs_is_atomic_file(inode)) {
> +               set_inode_flag(inode, FI_ATOMIC_DIRTIED);
> +               return;
> +       }
>
> Plz, refer to the above comment.

By the way, I need to change this patch a little bit to prevent a
direct inode dirtying case by VFS layer.
Plz, wait for the 2nd one.

>
> > - f2fs_ioc_commit_atomic_write
> >
> > So that I proposed a fix for this:
> > https://lore.kernel.org/linux-f2fs-devel/20240904032047.1264706-1-chao@xxxxxxxxxx
> >
> > Thanks,
> >
> > >
> > >> Thanks,
> > >>
> > >>>
> > >>>>
> > >>>> Thanks,
> > >>>>
> > >>>>>                 return;
> > >>>>>
> > >>>>> +     if (f2fs_is_atomic_file(inode)) {
> > >>>>> +             set_inode_flag(inode, FI_ATOMIC_DIRTIED);
> > >>>>> +             return;
> > >>>>> +     }
> > >>>>> +
> > >>>>>         mark_inode_dirty_sync(inode);
> > >>>>>     }
> > >>>>>
> > >>>>> @@ -653,10 +658,7 @@ void f2fs_update_inode(struct inode *inode, struct page *node_page)
> > >>>>>         ri->i_gid = cpu_to_le32(i_gid_read(inode));
> > >>>>>         ri->i_links = cpu_to_le32(inode->i_nlink);
> > >>>>>         ri->i_blocks = cpu_to_le64(SECTOR_TO_BLOCK(inode->i_blocks) + 1);
> > >>>>> -
> > >>>>> -     if (!f2fs_is_atomic_file(inode) ||
> > >>>>> -                     is_inode_flag_set(inode, FI_ATOMIC_COMMITTED))
> > >>>>> -             ri->i_size = cpu_to_le64(i_size_read(inode));
> > >>>>> +     ri->i_size = cpu_to_le64(i_size_read(inode));
> > >>>>>
> > >>>>>         if (et) {
> > >>>>>                 read_lock(&et->lock);
> > >>>>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> > >>>>> index 78c3198a6308..2b5391b229a8 100644
> > >>>>> --- a/fs/f2fs/segment.c
> > >>>>> +++ b/fs/f2fs/segment.c
> > >>>>> @@ -196,9 +196,12 @@ void f2fs_abort_atomic_write(struct inode *inode, bool clean)
> > >>>>>                 truncate_inode_pages_final(inode->i_mapping);
> > >>>>>
> > >>>>>         release_atomic_write_cnt(inode);
> > >>>>> -     clear_inode_flag(inode, FI_ATOMIC_COMMITTED);
> > >>>>>         clear_inode_flag(inode, FI_ATOMIC_REPLACE);
> > >>>>>         clear_inode_flag(inode, FI_ATOMIC_FILE);
> > >>>>> +     if (is_inode_flag_set(inode, FI_ATOMIC_DIRTIED)) {
> > >>>>> +             clear_inode_flag(inode, FI_ATOMIC_DIRTIED);
> > >>>>> +             f2fs_mark_inode_dirty_sync(inode, true);
> > >>>>> +     }
> > >>>>>         stat_dec_atomic_inode(inode);
> > >>>>>
> > >>>>>         F2FS_I(inode)->atomic_write_task = NULL;
> > >>>>> @@ -365,7 +368,10 @@ static int __f2fs_commit_atomic_write(struct inode *inode)
> > >>>>>                 sbi->revoked_atomic_block += fi->atomic_write_cnt;
> > >>>>>         } else {
> > >>>>>                 sbi->committed_atomic_block += fi->atomic_write_cnt;
> > >>>>> -             set_inode_flag(inode, FI_ATOMIC_COMMITTED);
> > >>>>> +             if (is_inode_flag_set(inode, FI_ATOMIC_DIRTIED)) {
> > >>>>> +                     clear_inode_flag(inode, FI_ATOMIC_DIRTIED);
> > >>>>> +                     f2fs_mark_inode_dirty_sync(inode, true);
> > >>>>> +             }
> > >>>>>         }
> > >>>>>
> > >>>>>         __complete_revoke_list(inode, &revoke_list, ret ? true : false);
> > >>>>
> > >>
> >