Re: [PATCH 2/4] f2fs: build stat_info before orphan inode recovery

[Jaegeuk Kim <jaegeuk@kernel.org>]

On 03/09, Chao Yu wrote:
> On 2017/3/8 6:13, Jaegeuk Kim wrote:
> > f2fs_sync_fs() -> write_checkpoint() calls stat_inc_cp_count(sbi->stat_info),
> > which needs stat_info allocation.
> > Otherwise, we can hit:
> > 
> > [254042.598623]  ? count_shadow_nodes+0xa0/0xa0
> > [254042.598633]  f2fs_sync_fs+0x65/0xd0 [f2fs]
> > [254042.598645]  f2fs_balance_fs_bg+0xe4/0x1c0 [f2fs]
> > [254042.598657]  f2fs_write_node_pages+0x34/0x1a0 [f2fs]
> > [254042.598664]  ? pagevec_lookup_entries+0x1e/0x30
> > [254042.598673]  do_writepages+0x1e/0x30
> > [254042.598682]  __writeback_single_inode+0x45/0x330
> > [254042.598688]  writeback_single_inode+0xd7/0x190
> > [254042.598694]  write_inode_now+0x86/0xa0
> > [254042.598699]  iput+0x122/0x200
> > [254042.598709]  f2fs_fill_super+0xd4a/0x14d0 [f2fs]
> > [254042.598717]  mount_bdev+0x184/0x1c0
> > [254042.598934]  ? f2fs_commit_super+0x100/0x100 [f2fs]
> > [254042.599142]  f2fs_mount+0x15/0x20 [f2fs]
> > [254042.599349]  mount_fs+0x39/0x160
> > [254042.599554]  ? __alloc_percpu+0x15/0x20
> > [254042.599759]  vfs_kern_mount+0x67/0x110
> > [254042.599972]  do_mount+0x1bb/0xc80
> > [254042.600175]  ? memdup_user+0x42/0x60
> > [254042.600380]  SyS_mount+0x83/0xd0
> > [254042.600583]  entry_SYSCALL_64_fastpath+0x1e/0xad
> > 
> > Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
> > ---
> >  fs/f2fs/super.c | 10 +++++-----
> >  1 file changed, 5 insertions(+), 5 deletions(-)
> > 
> > diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
> > index 96fe8ed73100..cfb40d3fd875 100644
> > --- a/fs/f2fs/super.c
> > +++ b/fs/f2fs/super.c
> > @@ -2022,6 +2022,10 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
> >  
> >  	f2fs_join_shrinker(sbi);
> >  
> > +	err = f2fs_build_stats(sbi);
> > +	if (err)
> > +		goto free_nm;
> 
> Both node_inode and shrinker need to be released and detached.
> 
> How about relocate f2fs_build_stats in front of build_segment_manager?

I just got another panic to try this, since sys/../f2fs/status needs them.

Thanks,

> 
> Thanks,
> 
> > +
> >  	/* if there are nt orphan nodes free them */
> >  	err = recover_orphan_inodes(sbi);
> >  	if (err)
> > @@ -2046,10 +2050,6 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
> >  		goto free_root_inode;
> >  	}
> >  
> > -	err = f2fs_build_stats(sbi);
> > -	if (err)
> > -		goto free_root_inode;
> > -
> >  	if (f2fs_proc_root)
> >  		sbi->s_proc = proc_mkdir(sb->s_id, f2fs_proc_root);
> >  
> > @@ -2143,7 +2143,6 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
> >  		remove_proc_entry("segment_bits", sbi->s_proc);
> >  		remove_proc_entry(sb->s_id, f2fs_proc_root);
> >  	}
> > -	f2fs_destroy_stats(sbi);
> >  free_root_inode:
> >  	dput(sb->s_root);
> >  	sb->s_root = NULL;
> > @@ -2161,6 +2160,7 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
> >  	truncate_inode_pages_final(META_MAPPING(sbi));
> >  	iput(sbi->node_inode);
> >  	mutex_unlock(&sbi->umount_mutex);
> > +	f2fs_destroy_stats(sbi);
> >  free_nm:
> >  	destroy_node_manager(sbi);
> >  free_sm:
> > 

------------------------------------------------------------------------------
Announcing the Oxford Dictionaries API! The API offers world-renowned
dictionary content that is easy and intuitive to access. Sign up for an
account today to start using our lexical data to power your apps and
projects. Get started today and enter our developer competition.
http://sdm.link/oxford