Re: [PATCH 1/2] f2fs: fix sbi->extent_list corruption issue

[Sahitya Tummala <stummala@codeaurora.org>]

On Fri, Nov 23, 2018 at 05:52:16PM +0800, Chao Yu wrote:
> On 2018/11/23 11:42, Sahitya Tummala wrote:
> > On Thu, Nov 22, 2018 at 04:11:07AM -0800, Jaegeuk Kim wrote:
> >> On 11/22, Chao Yu wrote:
> >>> On 2018/11/22 18:59, Sahitya Tummala wrote:
> >>>> When there is a failure in f2fs_fill_super() after/during
> >>>> the recovery of fsync'd nodes, it frees the current sbi and
> >>>> retries again. This time the mount is successful, but the files
> >>>> that got recovered before retry, still holds the extent tree,
> >>>> whose extent nodes list is corrupted since sbi and sbi->extent_list
> >>>> is freed up. The list_del corruption issue is observed when the
> >>>> file system is getting unmounted and when those recoverd files extent
> >>>> node is being freed up in the below context.
> >>>>
> >>>> list_del corruption. prev->next should be fffffff1e1ef5480, but was (null)
> >>>> <...>
> >>>> kernel BUG at kernel/msm-4.14/lib/list_debug.c:53!
> >>>> task: fffffff1f46f2280 task.stack: ffffff8008068000
> >>>> lr : __list_del_entry_valid+0x94/0xb4
> >>>> pc : __list_del_entry_valid+0x94/0xb4
> >>>> <...>
> >>>> Call trace:
> >>>> __list_del_entry_valid+0x94/0xb4
> >>>> __release_extent_node+0xb0/0x114
> >>>> __free_extent_tree+0x58/0x7c
> >>>> f2fs_shrink_extent_tree+0xdc/0x3b0
> >>>> f2fs_leave_shrinker+0x28/0x7c
> >>>> f2fs_put_super+0xfc/0x1e0
> >>>> generic_shutdown_super+0x70/0xf4
> >>>> kill_block_super+0x2c/0x5c
> >>>> kill_f2fs_super+0x44/0x50
> >>>> deactivate_locked_super+0x60/0x8c
> >>>> deactivate_super+0x68/0x74
> >>>> cleanup_mnt+0x40/0x78
> >>>> __cleanup_mnt+0x1c/0x28
> >>>> task_work_run+0x48/0xd0
> >>>> do_notify_resume+0x678/0xe98
> >>>> work_pending+0x8/0x14
> >>>>
> >>>> Fix this by cleaning up the extent tree of those recovered files
> >>>> before freeing up sbi and before next retry.
> >>>
> >>> Would it be more clear to call shrink_dcache_sb earlier to invalid all
> >>> inodes and call f2fs_shrink_extent_tree release cached entries and trees in
> >>> error path?
> >>
> >> Agreed.
> >>
> > I have tried doing shrink_dcache_sb() earlier but that doesn't call
> > f2fs_shrink_extent_tree(). So I have moved f2fs_join_shrinker() earlier and 
> > tried calling f2fs_leave_shrinker() in the error path. That helps to clean up
> > the cached extent nodes. However, I see that extent tree is left intact for
> 
> I didn't get it, you mean, in error path, after we call shrink_dcache_sb &
> f2fs_leave_shrinker, for those recovered files, their extent nodes were
> evicted, but their extent trees are still in cache?
> 

Yes, only extent tree is present with zero extent nodes as
f2fs_leave_shrinker() is only clearing the exntent nodes from
sbi->extent_list.

> > those recovered files, which should not be a problem as it gets freed as part
> > of next umount/rm. Only one small problem I see with this is - during rm/umount when
> > those previoulsy recovered files are being evicted, extent tree memory gets
> > free'd but the counter sbi->total_ext_tree gets invalid as these recovered
> > files are not present as part of current sbi->extent_tree_root. So i have come
> > up with this patch below to fix this. Let me know if this looks good?
> > 
> > diff --git a/fs/f2fs/extent_cache.c b/fs/f2fs/extent_cache.c
> > index 1cb0fcc..3e4801e 100644
> > --- a/fs/f2fs/extent_cache.c
> > +++ b/fs/f2fs/extent_cache.c
> > @@ -654,9 +654,9 @@ unsigned int f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
> >  		}
> >  		f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
> >  		list_del_init(&et->list);
> > -		radix_tree_delete(&sbi->extent_tree_root, et->ino);
> > +		if (radix_tree_delete(&sbi->extent_tree_root, et->ino))
> > +			atomic_dec(&sbi->total_ext_tree);
> >  		kmem_cache_free(extent_tree_slab, et);
> > -		atomic_dec(&sbi->total_ext_tree);
> >  		atomic_dec(&sbi->total_zombie_tree);
> >  		tree_cnt++;
> >  
> > @@ -767,7 +767,8 @@ void f2fs_destroy_extent_tree(struct inode *inode)
> >  	/* delete extent tree entry in radix tree */
> >  	mutex_lock(&sbi->extent_tree_lock);
> >  	f2fs_bug_on(sbi, atomic_read(&et->node_cnt));
> > -	radix_tree_delete(&sbi->extent_tree_root, inode->i_ino);
> > +	if (radix_tree_delete(&sbi->extent_tree_root, inode->i_ino))
> > +		atomic_dec(&sbi->total_ext_tree);
> >  	kmem_cache_free(extent_tree_slab, et);
> >  	atomic_dec(&sbi->total_ext_tree);
> >  	mutex_unlock(&sbi->extent_tree_lock);
> > diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
> > index af58b2c..3e5588f 100644
> > --- a/fs/f2fs/super.c
> > +++ b/fs/f2fs/super.c
> > @@ -3295,6 +3295,7 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
> >  	if (err)
> >  		goto free_root_inode;
> >  
> > +	f2fs_join_shrinker(sbi);
> >  #ifdef CONFIG_QUOTA
> >  	/* Enable quota usage during mount */
> >  	if (f2fs_sb_has_quota_ino(sb) && !f2fs_readonly(sb)) {
> > @@ -3379,8 +3380,6 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
> >  			sbi->valid_super_block ? 1 : 2, err);
> >  	}
> >  
> > -	f2fs_join_shrinker(sbi);
> > -
> >  	f2fs_tuning_parameters(sbi);
> >  
> >  	f2fs_msg(sbi->sb, KERN_NOTICE, "Mounted with checkpoint version = %llx",
> > @@ -3402,6 +3401,8 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
> >  	 * falls into an infinite loop in f2fs_sync_meta_pages().
> >  	 */
> >  	truncate_inode_pages_final(META_MAPPING(sbi));
> > +	shrink_dcache_sb(sb);
> > +	f2fs_leave_shrinker(sbi);
> 
> Why not just calling f2fs_shrink_extent_tree(sbi, __count_extent_cache(sbi)); ?
> 

Sure, will update in the next patchset.

> Thanks,
> 
> >  	f2fs_unregister_sysfs(sbi);
> >  free_root_inode:
> >  	dput(sb->s_root);
> > @@ -3445,7 +3446,6 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
> >  	/* give only one another chance */
> >  	if (retry) {
> >  		retry = false;
> > -		shrink_dcache_sb(sb);
> >  		goto try_onemore;
> >  	}
> >  	return err;
> > 
> 

-- 
--
Sent by a consultant of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.