Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 30 Aug 2022 20:05:12 -0700
From:   Jaegeuk Kim <jaegeuk@kernel.org>
To:     Chao Yu <chao@kernel.org>
Cc:     syzbot+775a3440817f74fddb8c@syzkaller.appspotmail.com,
        linux-kernel@vger.kernel.org, stable@vger.kernel.org,
        linux-f2fs-devel@lists.sourceforge.net
Subject: Re: [f2fs-dev] [PATCH] f2fs: fix missing mapping caused by the
 mount/umount race
Message-ID: <Yw7P6BkNZmqxji3B@google.com>
References: <20220829215206.3082124-1-jaegeuk@kernel.org>
 <cbc4bfe5-14f9-a4e0-c9c5-6b6b06437d5d@kernel.org>
 <Yw55Ebk8zLIgBFfn@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Yw55Ebk8zLIgBFfn@google.com>
Precedence: bulk

On 08/30, Jaegeuk Kim wrote:
> On 08/30, Chao Yu wrote:
> > On 2022/8/30 5:52, Jaegeuk Kim wrote:
> > > Sometimes we can get a cached meta_inode which has no aops yet. Let's set it
> > > all the time to fix the below panic.
> > > 
> > > Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000
> > > Mem abort info:
> > >    ESR = 0x0000000086000004
> > >    EC = 0x21: IABT (current EL), IL = 32 bits
> > >    SET = 0, FnV = 0
> > >    EA = 0, S1PTW = 0
> > >    FSC = 0x04: level 0 translation fault
> > > user pgtable: 4k pages, 48-bit VAs, pgdp=0000000109ee4000
> > > [0000000000000000] pgd=0000000000000000, p4d=0000000000000000
> > > Internal error: Oops: 86000004 [#1] PREEMPT SMP
> > > Modules linked in:
> > > CPU: 1 PID: 3045 Comm: syz-executor330 Not tainted 6.0.0-rc2-syzkaller-16455-ga41a877bc12d #0
> > > Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 07/22/2022
> > > pstate: 80400005 (Nzcv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> > > pc : 0x0
> > > lr : folio_mark_dirty+0xbc/0x208 mm/page-writeback.c:2748
> > > sp : ffff800012783970
> > > x29: ffff800012783970 x28: 0000000000000000 x27: ffff800012783b08
> > > x26: 0000000000000001 x25: 0000000000000400 x24: 0000000000000001
> > > x23: ffff0000c736e000 x22: 0000000000000045 x21: 05ffc00000000015
> > > x20: ffff0000ca7403b8 x19: fffffc00032ec600 x18: 0000000000000181
> > > x17: ffff80000c04d6bc x16: ffff80000dbb8658 x15: 0000000000000000
> > > x14: 0000000000000000 x13: 0000000000000000 x12: 0000000000000000
> > > x11: ff808000083e9814 x10: 0000000000000000 x9 : ffff8000083e9814
> > > x8 : 0000000000000000 x7 : 0000000000000000 x6 : 0000000000000000
> > > x5 : ffff0000cbb19000 x4 : ffff0000cb3d2000 x3 : ffff0000cbb18f80
> > > x2 : fffffffffffffff0 x1 : fffffc00032ec600 x0 : ffff0000ca7403b8
> > > Call trace:
> > >   0x0
> > >   set_page_dirty+0x38/0xbc mm/folio-compat.c:62
> > >   f2fs_update_meta_page+0x80/0xa8 fs/f2fs/segment.c:2369
> > >   do_checkpoint+0x794/0xea8 fs/f2fs/checkpoint.c:1522
> > >   f2fs_write_checkpoint+0x3b8/0x568 fs/f2fs/checkpoint.c:1679
> > > 
> > > Cc: stable@vger.kernel.org
> > > Reported-by: syzbot+775a3440817f74fddb8c@syzkaller.appspotmail.com
> > > Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
> > > ---
> > >   fs/f2fs/inode.c | 13 ++++++++-----
> > >   1 file changed, 8 insertions(+), 5 deletions(-)
> > > 
> > > diff --git a/fs/f2fs/inode.c b/fs/f2fs/inode.c
> > > index 6d11c365d7b4..1feb0a8a699e 100644
> > > --- a/fs/f2fs/inode.c
> > > +++ b/fs/f2fs/inode.c
> > > @@ -490,10 +490,7 @@ struct inode *f2fs_iget(struct super_block *sb, unsigned long ino)
> > >   	if (!inode)
> > >   		return ERR_PTR(-ENOMEM);
> > > -	if (!(inode->i_state & I_NEW)) {
> > > -		trace_f2fs_iget(inode);
> > > -		return inode;
> > > -	}
> > > +	/* We can see an old cached inode. Let's set the aops all the time. */
> > 
> > Why an old cached inode (has no I_NEW flag) has NULL a_ops pointer? If it is a bad
> > inode, it should be unhashed before unlock_new_inode().
> 
> I'm trying to dig further tho, it's not a bad inode, nor I_FREEING | I_CLEAR.
> It's very werid that thie meta inode is found in newly created superblock by
> the global hash table. I've checked that the same superblock pointer was used
> in the previous tests, but inode was evictied all the time.

I'll drop this patch, since it turned out there is a bug in reiserfs which
doesn't free the root inode (ino=2). That leads f2fs to find an ino=2 with
the previous superblock point used by reiserfs. That stale inode has no valid
inode that f2fs can use. I tried to find where the root cause is in reiserfs,
but it seems quite hard to catch one.

- reiserfs_fill_super
 - reiserfs_xattr_init
  - create_privroot
   - xattr_mkdir
    - reiserfs_new_inode
     - reiserfs_get_unused_objectid returned 0 due to map crash

It seems the error path doesn't handle the root inode properly.

> 
> > 
> > Thanks,
> > 
> > >   	if (ino == F2FS_NODE_INO(sbi) || ino == F2FS_META_INO(sbi))
> > >   		goto make_now;
> > > @@ -502,6 +499,11 @@ struct inode *f2fs_iget(struct super_block *sb, unsigned long ino)
> > >   		goto make_now;
> > >   #endif
> > > +	if (!(inode->i_state & I_NEW)) {
> > > +		trace_f2fs_iget(inode);
> > > +		return inode;
> > > +	}
> > > +
> > >   	ret = do_read_inode(inode);
> > >   	if (ret)
> > >   		goto bad_inode;
> > > @@ -557,7 +559,8 @@ struct inode *f2fs_iget(struct super_block *sb, unsigned long ino)
> > >   		file_dont_truncate(inode);
> > >   	}
> > > -	unlock_new_inode(inode);
> > > +	if (inode->i_state & I_NEW)
> > > +		unlock_new_inode(inode);
> > >   	trace_f2fs_iget(inode);
> > >   	return inode;
> 
> 
> _______________________________________________
> Linux-f2fs-devel mailing list
> Linux-f2fs-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel