[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAGudoHGui53Ryz1zunmd=G=Rr9cZOsWPFW7+GGBmxN4U_BNE4A@mail.gmail.com>
Date: Mon, 8 Sep 2025 17:39:22 +0200
From: Mateusz Guzik <mjguzik@...il.com>
To: Jan Kara <jack@...e.cz>
Cc: Joseph Qi <joseph.qi@...ux.alibaba.com>, Mark Tinguely <mark.tinguely@...cle.com>,
ocfs2-devel@...ts.linux.dev, viro@...iv.linux.org.uk,
linux-kernel@...r.kernel.org, linux-fsdevel@...r.kernel.org,
josef@...icpanda.com, jlbec@...lplan.org, mark@...heh.com, brauner@...nel.org,
willy@...radead.org, david@...morbit.com
Subject: Re: [External] : [PATCH] ocfs2: retire ocfs2_drop_inode() and
I_WILL_FREE usage
On Mon, Sep 8, 2025 at 3:54 PM Jan Kara <jack@...e.cz> wrote:
>
> On Mon 08-09-25 20:41:21, Joseph Qi wrote:
> >
> >
> > On 2025/9/8 18:23, Jan Kara wrote:
> > > On Mon 08-09-25 09:51:36, Joseph Qi wrote:
> > >> On 2025/9/5 00:22, Mateusz Guzik wrote:
> > >>> On Thu, Sep 4, 2025 at 6:15 PM Mark Tinguely <mark.tinguely@...cle.com> wrote:
> > >>>>
> > >>>> On 9/4/25 10:42 AM, Mateusz Guzik wrote:
> > >>>>> This postpones the writeout to ocfs2_evict_inode(), which I'm told is
> > >>>>> fine (tm).
> > >>>>>
> > >>>>> The intent is to retire the I_WILL_FREE flag.
> > >>>>>
> > >>>>> Signed-off-by: Mateusz Guzik <mjguzik@...il.com>
> > >>>>> ---
> > >>>>>
> > >>>>> ACHTUNG: only compile-time tested. Need an ocfs2 person to ack it.
> > >>>>>
> > >>>>> btw grep shows comments referencing ocfs2_drop_inode() which are already
> > >>>>> stale on the stock kernel, I opted to not touch them.
> > >>>>>
> > >>>>> This ties into an effort to remove the I_WILL_FREE flag, unblocking
> > >>>>> other work. If accepted would be probably best taken through vfs
> > >>>>> branches with said work, see https://urldefense.com/v3/__https://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git/log/?h=vfs-6.18.inode.refcount.preliminaries__;!!ACWV5N9M2RV99hQ!OLwk8DVo7uvC-Pd6XVTiUCgP6MUDMKBMEyuV27h_yPGXOjaq078-kMdC9ILFoYQh-4WX93yb0nMfBDFFY_0$
> > >>>>>
> > >>>>> fs/ocfs2/inode.c | 23 ++---------------------
> > >>>>> fs/ocfs2/inode.h | 1 -
> > >>>>> fs/ocfs2/ocfs2_trace.h | 2 --
> > >>>>> fs/ocfs2/super.c | 2 +-
> > >>>>> 4 files changed, 3 insertions(+), 25 deletions(-)
> > >>>>>
> > >>>>> diff --git a/fs/ocfs2/inode.c b/fs/ocfs2/inode.c
> > >>>>> index 6c4f78f473fb..5f4a2cbc505d 100644
> > >>>>> --- a/fs/ocfs2/inode.c
> > >>>>> +++ b/fs/ocfs2/inode.c
> > >>>>> @@ -1290,6 +1290,8 @@ static void ocfs2_clear_inode(struct inode *inode)
> > >>>>>
> > >>>>> void ocfs2_evict_inode(struct inode *inode)
> > >>>>> {
> > >>>>> + write_inode_now(inode, 1);
> > >>>>> +
> > >>>>> if (!inode->i_nlink ||
> > >>>>> (OCFS2_I(inode)->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)) {
> > >>>>> ocfs2_delete_inode(inode);
> > >>>>> @@ -1299,27 +1301,6 @@ void ocfs2_evict_inode(struct inode *inode)
> > >>>>> ocfs2_clear_inode(inode);
> > >>>>> }
> > >>>>>
> > >>>>> -/* Called under inode_lock, with no more references on the
> > >>>>> - * struct inode, so it's safe here to check the flags field
> > >>>>> - * and to manipulate i_nlink without any other locks. */
> > >>>>> -int ocfs2_drop_inode(struct inode *inode)
> > >>>>> -{
> > >>>>> - struct ocfs2_inode_info *oi = OCFS2_I(inode);
> > >>>>> -
> > >>>>> - trace_ocfs2_drop_inode((unsigned long long)oi->ip_blkno,
> > >>>>> - inode->i_nlink, oi->ip_flags);
> > >>>>> -
> > >>>>> - assert_spin_locked(&inode->i_lock);
> > >>>>> - inode->i_state |= I_WILL_FREE;
> > >>>>> - spin_unlock(&inode->i_lock);
> > >>>>> - write_inode_now(inode, 1);
> > >>>>> - spin_lock(&inode->i_lock);
> > >>>>> - WARN_ON(inode->i_state & I_NEW);
> > >>>>> - inode->i_state &= ~I_WILL_FREE;
> > >>>>> -
> > >>>>> - return 1;
> > >>>>> -}
> > >>>>> -
> > >>>>> /*
> > >>>>> * This is called from our getattr.
> > >>>>> */
> > >>>>> diff --git a/fs/ocfs2/inode.h b/fs/ocfs2/inode.h
> > >>>>> index accf03d4765e..07bd838e7843 100644
> > >>>>> --- a/fs/ocfs2/inode.h
> > >>>>> +++ b/fs/ocfs2/inode.h
> > >>>>> @@ -116,7 +116,6 @@ static inline struct ocfs2_caching_info *INODE_CACHE(struct inode *inode)
> > >>>>> }
> > >>>>>
> > >>>>> void ocfs2_evict_inode(struct inode *inode);
> > >>>>> -int ocfs2_drop_inode(struct inode *inode);
> > >>>>>
> > >>>>> /* Flags for ocfs2_iget() */
> > >>>>> #define OCFS2_FI_FLAG_SYSFILE 0x1
> > >>>>> diff --git a/fs/ocfs2/ocfs2_trace.h b/fs/ocfs2/ocfs2_trace.h
> > >>>>> index 54ed1495de9a..4b32fb5658ad 100644
> > >>>>> --- a/fs/ocfs2/ocfs2_trace.h
> > >>>>> +++ b/fs/ocfs2/ocfs2_trace.h
> > >>>>> @@ -1569,8 +1569,6 @@ DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_delete_inode);
> > >>>>>
> > >>>>> DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_clear_inode);
> > >>>>>
> > >>>>> -DEFINE_OCFS2_ULL_UINT_UINT_EVENT(ocfs2_drop_inode);
> > >>>>> -
> > >>>>> TRACE_EVENT(ocfs2_inode_revalidate,
> > >>>>> TP_PROTO(void *inode, unsigned long long ino,
> > >>>>> unsigned int flags),
> > >>>>> diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
> > >>>>> index 53daa4482406..e4b0d25f4869 100644
> > >>>>> --- a/fs/ocfs2/super.c
> > >>>>> +++ b/fs/ocfs2/super.c
> > >>>>> @@ -129,7 +129,7 @@ static const struct super_operations ocfs2_sops = {
> > >>>>> .statfs = ocfs2_statfs,
> > >>>>> .alloc_inode = ocfs2_alloc_inode,
> > >>>>> .free_inode = ocfs2_free_inode,
> > >>>>> - .drop_inode = ocfs2_drop_inode,
> > >>>>> + .drop_inode = generic_delete_inode,
> > >>>>> .evict_inode = ocfs2_evict_inode,
> > >>>>> .sync_fs = ocfs2_sync_fs,
> > >>>>> .put_super = ocfs2_put_super,
> > >>>>
> > >>>>
> > >>>> I agree, fileystems should not use I_FREEING/I_WILL_FREE.
> > >>>> Doing the sync write_inode_now() should be fine in ocfs_evict_inode().
> > >>>>
> > >>>> Question is ocfs_drop_inode. In commit 513e2dae9422:
> > >>>> ocfs2: flush inode data to disk and free inode when i_count becomes zero
> > >>>> the return of 1 drops immediate to fix a memory caching issue.
> > >>>> Shouldn't .drop_inode() still return 1?
> > >>>
> > >>> generic_delete_inode is a stub doing just that.
> > >>>
> > >> In case of "drop = 0", it may return directly without calling evict().
> > >> This seems break the expectation of commit 513e2dae9422.
> > >
> > > generic_delete_inode() always returns 1 so evict() will be called.
> > > ocfs2_drop_inode() always returns 1 as well after 513e2dae9422. So I'm not
> > > sure which case of "drop = 0" do you see...
> > >
> > I don't see a real case, just in theory.
> > As I described before, if we make sure write_inode_now() will be called
> > in iput_final(), it would be fine.
>
> I'm sorry but I still don't quite understand what you are proposing. If
> ->drop() returns 1, the filesystem wants to remove the inode from cache
> (perhaps because it was deleted). Hence iput_final() doesn't bother with
> writing out such inodes. This doesn't work well with ocfs2 wanting to
> always drop inodes hence ocfs2 needs to write the inode itself in
> ocfs2_evice_inode(). Perhaps you have some modification to iput_final() in
> mind but I'm not sure how that would work so can you perhaps suggest a
> patch if you think iput_final() should work differently? Thanks!
>
I think generic_delete_inode is a really bad name for what the routine
is doing and it perhaps contributes to the confusion in the thread.
Perhaps it could be renamed to inode_op_stub_always_drop or similar? I
don't for specifics, apart from explicitly stating that the return
value is to drop and bonus points for a prefix showing this is an
inode thing.
--
Mateusz Guzik <mjguzik gmail.com>
Powered by blists - more mailing lists