Message-ID: <059C75EDB49641A49EF986D81EED4200@nsl.ad.nec.co.jp>
From: "Takashi Sato" <t-sato@yk.jp.nec.com>
To: "Christoph Hellwig" <hch@infradead.org>
Cc: "Andrew Morton" <akpm@linux-foundation.org>,
        "Christoph Hellwig" <hch@infradead.org>,
        "Oleg Nesterov" <oleg@tv-sign.ru>, <linux-fsdevel@vger.kernel.org>,
        <dm-devel@redhat.com>, <viro@ZenIV.linux.org.uk>,
        <linux-ext4@vger.kernel.org>, <xfs@oss.sgi.com>, <axboe@kernel.dk>,
        <mtk.manpages@googlemail.com>, <linux-kernel@vger.kernel.org>
References: <20080908205245t-sato@mail.jp.nec.com> <20080908171020.GA22521@infradead.org>
In-Reply-To: <20080908171020.GA22521@infradead.org>
Subject: Re: [PATCH 1/3] Implement generic freeze feature
Date: Thu, 11 Sep 2008 20:11:06 +0900
MIME-Version: 1.0
Content-Type: text/plain;
	format=flowed;
	charset="iso-8859-1";
	reply-type=original
Content-Transfer-Encoding: 7bit

Hi,

Christoph Hellwig wrote:
>> --- linux-2.6.27-rc5.org/fs/block_dev.c 2008-08-29 07:52:02.000000000 +0900
>> +++ linux-2.6.27-rc5-freeze/fs/block_dev.c 2008-09-05 20:00:29.000000000 +0900
>> @@ -285,6 +285,8 @@ static void init_once(void *foo)
>>      INIT_LIST_HEAD(&bdev->bd_holder_list);
>>  #endif
>>      inode_init_once(&ei->vfs_inode);
>> +     /* Initialize mutex for freeze. */
>> +     mutex_init(&bdev->bd_fsfreeze_mutex);
>
> Why not just freeze_mutex?

The Linux kernel has already had the name of "freezer" in the part of
power-management.  So I named the above mutex "fsfreeze"
(filesystem freeze) to distinguish it from the existent "freezer".
Andrew pointed it out.

>>  struct super_block *freeze_bdev(struct block_device *bdev)
>>  {
>>       struct super_block *sb;
>>
>> +    mutex_lock(&bdev->bd_fsfreeze_mutex);
>> +    if (bdev->bd_fsfreeze_count > 0) {
>> +        bdev->bd_fsfreeze_count++;
>> +        sb = get_super(bdev);
>> +        mutex_unlock(&bdev->bd_fsfreeze_mutex);
>> +        return sb;
>> +    }
>> +    bdev->bd_fsfreeze_count++;
>> +
>>      down(&bdev->bd_mount_sem);
>
> Note that we still have duplication with the bd_mount_sem.  I think
> you should look into getting rid of it and instead do a check of
> the freeze_count under proper freeze_mutex protection.

In the original implementation,
while the filesystem is frozen, subsequent mounts wait for unfreeze
with the semaphore (bd_mount_sem).
But if we replace the semphore with the check of the freeze_count,
subsequent mounts will abort.
I think the original behavior shouldn't be changed, so the existing bd_mount_sem
is better.

>> @@ -244,6 +274,8 @@ void thaw_bdev(struct block_device *bdev
>>  }
>>
>>        up(&bdev->bd_mount_sem);
>> +     mutex_unlock(&bdev->bd_fsfreeze_mutex);
>> +     return 0;
>
> Why do you add a return value here if we always return 0 anyway?

I forgot to remove the unneeded return value in above old patch.
But I need to implement a return value in the new patch
because thaw_bdev() needs to return an IO error which occurs
in unlockfs().
Eric pointed it out.

Cheers, Takashi