Re: [PATCH 1/2] quota: fix null-ptr-deref in ext4_acquire_dquot()

[Jan Kara <jack@suse.cz>]

On Sat 27-05-23 09:40:17, Ye Bin wrote:
> Syzbot found the following issue:
> Unable to handle kernel paging request at virtual address dfff800000000005
> KASAN: null-ptr-deref in range [0x0000000000000028-0x000000000000002f]

...
> CPU: 0 PID: 6080 Comm: syz-executor747 Not tainted 6.3.0-rc7-syzkaller-g14f8db1c0f9a #0
> Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 04/14/2023
> pstate: 80400005 (Nzcv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> pc : ext4_acquire_dquot+0x1d4/0x398 fs/ext4/super.c:6766
> lr : dquot_to_inode fs/ext4/super.c:6740 [inline]
> lr : ext4_acquire_dquot+0x1ac/0x398 fs/ext4/super.c:6766

OK, this is bad...

> Above issue may happens as follows:
> ProcessA              ProcessB                    ProcessC
> sys_fsconfig
>   vfs_fsconfig_locked
>    reconfigure_super
>      ext4_remount
>       dquot_suspend -> suspend all type quota
> 
>                  sys_fsconfig
> 		  vfs_fsconfig_locked
> 		    reconfigure_super
> 		     ext4_remount
> 		      dquot_resume
> 		       ret = dquot_load_quota_sb
>                         add_dquot_ref
> 		                           do_open  -> open file O_RDWR
> 					    vfs_open
> 					     do_dentry_open
> 					      get_write_access
> 					       atomic_inc_unless_negative(&inode->i_writecount)
>                                               ext4_file_open
> 					       dquot_file_open
> 					        dquot_initialize
> 						  __dquot_initialize
> 						   dqget
> 						    if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
> 
> 			  __dquot_initialize
> 			   __dquot_initialize
> 			    dqget
> 			     if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
> 	                       ext4_acquire_dquot -> Return error
> 		       if (ret < 0)
> 	                 vfs_cleanup_quota_inode
> 			  dqopt->files[type] = NULL;

But I don't see how this can happen. The code in dquot_load_quota_sb()
looks like:

        error = add_dquot_ref(sb, type);
        if (error)
                dquot_disable(sb, type, flags);

So if an error happens in add_dquot_ref(), we'll call dquot_disable().
dquot_disable() then does:

                drop_dquot_ref(sb, cnt);
                invalidate_dquots(sb, cnt);

and invalidate_dquots() waits for reference count of all dquots to drop to
0. Hence if dqget() returned a dquot pointer to ProcessC, then ProcessB
should wait until ProcessC drops the dquot reference (hence
ext4_acquire_dquot() is done).

What am I missing?

								Honza

> 
> 			                              ext4_acquire_dquot
> 						       -->dquot_to_inode(dquot) is NULL
> 
> To solve above issue, if quota has been loaded, there's unneed to cleaup quota
> inode if dquot_load_quota_sb() return failed when do dquot_resume();
> 
> Reported-by: syzbot+e633c79ceaecbf479854@syzkaller.appspotmail.com
> Signed-off-by: Ye Bin <yebin10@huawei.com>
> ---
>  fs/quota/dquot.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/quota/dquot.c b/fs/quota/dquot.c
> index ffd40dc3e4e9..4b913faa48ec 100644
> --- a/fs/quota/dquot.c
> +++ b/fs/quota/dquot.c
> @@ -2476,7 +2476,7 @@ int dquot_resume(struct super_block *sb, int type)
>  		flags = dquot_generic_flag(flags, cnt);
>  		ret = dquot_load_quota_sb(sb, cnt, dqopt->info[cnt].dqi_fmt_id,
>  					  flags);
> -		if (ret < 0)
> +		if (ret < 0 && !sb_has_quota_loaded(sb, cnt))
>  			vfs_cleanup_quota_inode(sb, cnt);
>  	}
>  
> -- 
> 2.31.1
> 
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR