Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Mon, 23 Mar 2020 02:50:36 +0100
From:   =?utf-8?Q?Ond=C5=99ej?= Jirman <megi@xff.cz>
To:     Chao Yu <yuchao0@huawei.com>
Cc:     jaegeuk@kernel.org, linux-f2fs-devel@lists.sourceforge.net,
        linux-kernel@vger.kernel.org, chao@kernel.org
Subject: Re: [PATCH v3] f2fs: fix potential .flags overflow on 32bit
 architecture
Message-ID: <20200323015036.pniupuucfl3dug4m@core.my.home>
Mail-Followup-To: =?utf-8?Q?Ond=C5=99ej?= Jirman <megi@xff.cz>,
        Chao Yu <yuchao0@huawei.com>, jaegeuk@kernel.org,
        linux-f2fs-devel@lists.sourceforge.net,
        linux-kernel@vger.kernel.org, chao@kernel.org
References: <20200323012519.41536-1-yuchao0@huawei.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200323012519.41536-1-yuchao0@huawei.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hello Chao Yu,

On Mon, Mar 23, 2020 at 09:25:19AM +0800, Chao Yu wrote:
> [snip]
>  
> +static inline void __set_inode_flag(struct inode *inode, int flag)
> +{
> +	test_and_set_bit(flag % BITS_PER_LONG,
> +			&F2FS_I(inode)->flags[BIT_WORD(flag)]);

This can simply be:

    test_and_set_bit(flag, F2FS_I(inode)->flags);

all of these bitmap manipulation functions already will do the
right thing to access the correct location in the flags array:

  https://elixir.bootlin.com/linux/latest/source/include/asm-generic/bitops/atomic.h#L32

see BIT_MASK and BIT_WORD use in that function.

> +}
> +
>  static inline void set_inode_flag(struct inode *inode, int flag)
>  {
> -	if (!test_bit(flag, &F2FS_I(inode)->flags))
> -		set_bit(flag, &F2FS_I(inode)->flags);
> +	__set_inode_flag(inode, flag);
>  	__mark_inode_dirty_flag(inode, flag, true);
>  }
>  
>  static inline int is_inode_flag_set(struct inode *inode, int flag)
>  {
> -	return test_bit(flag, &F2FS_I(inode)->flags);
> +	return test_bit(flag % BITS_PER_LONG,
> +				&F2FS_I(inode)->flags[BIT_WORD(flag)]);

ditto

>  }
>  
>  static inline void clear_inode_flag(struct inode *inode, int flag)
>  {
> -	if (test_bit(flag, &F2FS_I(inode)->flags))
> -		clear_bit(flag, &F2FS_I(inode)->flags);
> +	test_and_clear_bit(flag % BITS_PER_LONG,
> +				&F2FS_I(inode)->flags[BIT_WORD(flag)]);

ditto

I'm going to test the patch. It looks like that this was really
the root cause of all those locking issues I was seeing on my
32-bit tablet. It seems to explain why my 64-bit systems were
not affected, and why reverting compession fixed it too.
Great job figuring this out.

I'll let you know soon.

thank you and regards,
	o.

>  	__mark_inode_dirty_flag(inode, flag, false);
>  }
>