Received-SPF: pass (google.com: domain of linux-kernel+bounces-94940-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45d1:ec00::1 as permitted sender) client-ip=2604:1380:45d1:ec00::1;
Message-ID: <7e262242-d90d-4f61-a217-f156219eaa4d@linux.alibaba.com>
Date: Thu, 7 Mar 2024 11:41:52 +0800
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH] erofs: fix lockdep false positives on initializing
 erofs_pseudo_mnt
Content-Language: en-US
To: Gao Xiang <hsiangkao@linux.alibaba.com>, Baokun Li
 <libaokun1@huawei.com>, linux-erofs@lists.ozlabs.org
Cc: xiang@kernel.org, chao@kernel.org, huyue2@coolpad.com,
 linux-kernel@vger.kernel.org, yangerkun@huawei.com, houtao1@huawei.com,
 yukuai3@huawei.com, chengzhihao1@huawei.com
References: <20240307024459.883044-1-libaokun1@huawei.com>
 <f9e30004-6691-4171-abc5-7e286d9ccec6@linux.alibaba.com>
From: Jingbo Xu <jefflexu@linux.alibaba.com>
In-Reply-To: <f9e30004-6691-4171-abc5-7e286d9ccec6@linux.alibaba.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Hi Baokun,

Thanks for catching this!


On 3/7/24 10:52 AM, Gao Xiang wrote:
> Hi Baokun,
> 
> On 2024/3/7 10:44, Baokun Li wrote:
>> Lockdep reported the following issue when mounting erofs with a
>> domain_id:
>>
>> ============================================
>> WARNING: possible recursive locking detected
>> 6.8.0-rc7-xfstests #521 Not tainted
>> --------------------------------------------
>> mount/396 is trying to acquire lock:
>> ffff907a8aaaa0e0 (&type->s_umount_key#50/1){+.+.}-{3:3},
>>                         at: alloc_super+0xe3/0x3d0
>>
>> but task is already holding lock:
>> ffff907a8aaa90e0 (&type->s_umount_key#50/1){+.+.}-{3:3},
>>                         at: alloc_super+0xe3/0x3d0
>>
>> other info that might help us debug this:
>>   Possible unsafe locking scenario:
>>
>>         CPU0
>>         ----
>>    lock(&type->s_umount_key#50/1);
>>    lock(&type->s_umount_key#50/1);
>>
>>   *** DEADLOCK ***
>>
>>   May be due to missing lock nesting notation
>>
>> 2 locks held by mount/396:
>>   #0: ffff907a8aaa90e0 (&type->s_umount_key#50/1){+.+.}-{3:3},
>>             at: alloc_super+0xe3/0x3d0
>>   #1: ffffffffc00e6f28 (erofs_domain_list_lock){+.+.}-{3:3},
>>             at: erofs_fscache_register_fs+0x3d/0x270 [erofs]
>>
>> stack backtrace:
>> CPU: 1 PID: 396 Comm: mount Not tainted 6.8.0-rc7-xfstests #521
>> Call Trace:
>>   <TASK>
>>   dump_stack_lvl+0x64/0xb0
>>   validate_chain+0x5c4/0xa00
>>   __lock_acquire+0x6a9/0xd50
>>   lock_acquire+0xcd/0x2b0
>>   down_write_nested+0x45/0xd0
>>   alloc_super+0xe3/0x3d0
>>   sget_fc+0x62/0x2f0
>>   vfs_get_super+0x21/0x90
>>   vfs_get_tree+0x2c/0xf0
>>   fc_mount+0x12/0x40
>>   vfs_kern_mount.part.0+0x75/0x90
>>   kern_mount+0x24/0x40
>>   erofs_fscache_register_fs+0x1ef/0x270 [erofs]
>>   erofs_fc_fill_super+0x213/0x380 [erofs]
>>
>> This is because the file_system_type of both erofs and the pseudo-mount
>> point of domain_id is erofs_fs_type, so two successive calls to
>> alloc_super() are considered to be using the same lock and trigger the
>> warning above.
>>
>> Therefore add a nodev file_system_type named erofs_anon_fs_type to
>> silence this complaint. In addition, to reduce code coupling, refactor
>> out the erofs_anon_init_fs_context() and erofs_kill_pseudo_sb() functions
>> and move the erofs_pseudo_mnt related code to fscache.c.
>>
>> Signed-off-by: Baokun Li <libaokun1@huawei.com>
> 
> IMHO, in the beginning, I'd like to avoid introducing another fs type
> for erofs to share (meta)data between filesystems since it will cause
> churn, could we use some alternative way to resolve this?

Yeah as Gao Xiang said, this is initially intended to avoid introducing
anothoer file_system_type, say erofs_anon_fs_type.

What we need is actually a method of allocating anonymous inode as a
sentinel identifying each blob.  There is indeed a global mount, i.e.
anon_inode_mnt, for allocating anonymous inode/file specifically.  At
the time the share domain feature is introduced, there's only one
anonymous inode, i.e. anon_inode_inode, and all the allocated anonymous
files are bound to this single anon_inode_inode.  Thus we decided to
implement a erofs internal pseudo mount for this usage.

But I noticed that we can now allocate unique anonymous inodes from
anon_inode_mnt since commit e7e832c ("fs: add LSM-supporting anon-inode
interface"), though the new interface is initially for LSM usage.

-- 
Thanks,
Jingbo