[PATCH] nfs: pass flags to second superblock

[PATCH] nfs: pass flags to second superblock

[Li Lingfeng]

During the process of mounting an NFSv4 client, two superblocks will be
created in sequence. The first superblock corresponds to the root
directory exported by the server, and the second superblock corresponds to
the directory that will be actually mounted. The first superblock will
eventually be destroyed.
The flag passed from user mode will only be passed to the first
superblock, resulting in the actual used superblock not carrying the flag
passed from user mode(fs_context_for_submount() will set sb_flags as 0).

Since the superblock of NFS does not carry the ro tag, the file system
status displayed by /proc/self/mountstats shows that NFS is always in the
rw state, which may mislead users.

Pass sb_flags of the fc which carry flags passed by user to second
superblock to fix it.

Fixes: 281cad46b34d ("NFS: Create a submount rpc_op")
Signed-off-by: Li Lingfeng <lilingfeng3@huawei.com>
---
 fs/nfs/nfs4super.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/fs/nfs/nfs4super.c b/fs/nfs/nfs4super.c
index b29a26923ce0..9a3b73a33fbf 100644
--- a/fs/nfs/nfs4super.c
+++ b/fs/nfs/nfs4super.c
@@ -233,6 +233,7 @@ static int do_nfs4_mount(struct nfs_server *server,
 	if (IS_ERR(dentry))
 		return PTR_ERR(dentry);
 
+	dentry->d_sb->s_flags = fc->sb_flags;
 	fc->root = dentry;
 	return 0;
 }
-- 
2.31.1

Re: [PATCH] nfs: pass flags to second superblock

[Anna Schumaker]

Hi Li,

On Wed, Nov 13, 2024 at 11:33 PM Li Lingfeng <lilingfeng3@huawei.com> wrote:
>
> During the process of mounting an NFSv4 client, two superblocks will be
> created in sequence. The first superblock corresponds to the root
> directory exported by the server, and the second superblock corresponds to
> the directory that will be actually mounted. The first superblock will
> eventually be destroyed.
> The flag passed from user mode will only be passed to the first
> superblock, resulting in the actual used superblock not carrying the flag
> passed from user mode(fs_context_for_submount() will set sb_flags as 0).
>
> Since the superblock of NFS does not carry the ro tag, the file system
> status displayed by /proc/self/mountstats shows that NFS is always in the
> rw state, which may mislead users.
>
> Pass sb_flags of the fc which carry flags passed by user to second
> superblock to fix it.
>
> Fixes: 281cad46b34d ("NFS: Create a submount rpc_op")
> Signed-off-by: Li Lingfeng <lilingfeng3@huawei.com>
> ---
>  fs/nfs/nfs4super.c | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/fs/nfs/nfs4super.c b/fs/nfs/nfs4super.c
> index b29a26923ce0..9a3b73a33fbf 100644
> --- a/fs/nfs/nfs4super.c
> +++ b/fs/nfs/nfs4super.c
> @@ -233,6 +233,7 @@ static int do_nfs4_mount(struct nfs_server *server,
>         if (IS_ERR(dentry))
>                 return PTR_ERR(dentry);
>
> +       dentry->d_sb->s_flags = fc->sb_flags;

I'm seeing a handful of new xfstests failures that I bisected to this
patch: generic/157, generic/184, generic/306, generic/564, and
generic/598.

I'm seeing this on NFS v4.1 and v4.2, and it looks like each one of
these failures is due to a new -EIO error being generated. Any
thoughts about what could be causing this?

Thanks,
Anna


>         fc->root = dentry;
>         return 0;
>  }
> --
> 2.31.1
>

Re: [PATCH] nfs: pass flags to second superblock

[Li Lingfeng]

Hi Anna,

We found that this modification clears the flags (SB_BORN/SB_ACTIVE/
SB_POSIXACL) already set in the superblock, which would cause problems.
We will redesign a solution to address the original issue.
Sorry for the noise.

Thanks,
Li

在 2024/11/26 6:34, Anna Schumaker 写道:
> Hi Li,
>
> On Wed, Nov 13, 2024 at 11:33 PM Li Lingfeng <lilingfeng3@huawei.com> wrote:
>> During the process of mounting an NFSv4 client, two superblocks will be
>> created in sequence. The first superblock corresponds to the root
>> directory exported by the server, and the second superblock corresponds to
>> the directory that will be actually mounted. The first superblock will
>> eventually be destroyed.
>> The flag passed from user mode will only be passed to the first
>> superblock, resulting in the actual used superblock not carrying the flag
>> passed from user mode(fs_context_for_submount() will set sb_flags as 0).
>>
>> Since the superblock of NFS does not carry the ro tag, the file system
>> status displayed by /proc/self/mountstats shows that NFS is always in the
>> rw state, which may mislead users.
>>
>> Pass sb_flags of the fc which carry flags passed by user to second
>> superblock to fix it.
>>
>> Fixes: 281cad46b34d ("NFS: Create a submount rpc_op")
>> Signed-off-by: Li Lingfeng <lilingfeng3@huawei.com>
>> ---
>>   fs/nfs/nfs4super.c | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/fs/nfs/nfs4super.c b/fs/nfs/nfs4super.c
>> index b29a26923ce0..9a3b73a33fbf 100644
>> --- a/fs/nfs/nfs4super.c
>> +++ b/fs/nfs/nfs4super.c
>> @@ -233,6 +233,7 @@ static int do_nfs4_mount(struct nfs_server *server,
>>          if (IS_ERR(dentry))
>>                  return PTR_ERR(dentry);
>>
>> +       dentry->d_sb->s_flags = fc->sb_flags;
> I'm seeing a handful of new xfstests failures that I bisected to this
> patch: generic/157, generic/184, generic/306, generic/564, and
> generic/598.
>
> I'm seeing this on NFS v4.1 and v4.2, and it looks like each one of
> these failures is due to a new -EIO error being generated. Any
> thoughts about what could be causing this?
>
> Thanks,
> Anna
>
>
>>          fc->root = dentry;
>>          return 0;
>>   }
>> --
>> 2.31.1
>>
>