Re: [PATCH 2/6] nfsd: release path refs on follow_down() error

[Chuck Lever]

On 6/1/26 11:57 AM, Jeff Layton wrote:
> On Mon, 2026-06-01 at 19:47 +0100, Al Viro wrote:
>> On Sun, May 31, 2026 at 08:06:59AM -0400, Jeff Layton wrote:
>>
>>
>>> Fix by calling path_put(&path) before the goto out in the err < 0
>>> arm so the entry-time refs are released on all follow_down() error
>>> returns.
>>
>> Might be better to unify all those path_put()...  Something like this,
>> perhaps?
>>
>> diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
>> index eafdf7b7890f..bdcd78f49112 100644
>> --- a/fs/nfsd/vfs.c
>> +++ b/fs/nfsd/vfs.c
>> @@ -141,13 +141,13 @@ nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
>>  	if (path.mnt == exp->ex_path.mnt && path.dentry == dentry &&
>>  	    nfsd_mountpoint(dentry, exp) == 2) {
>>  		/* This is only a mountpoint in some other namespace */
>> -		path_put(&path);
>>  		goto out;
>>  	}
>>  
>>  	exp2 = rqst_exp_get_by_name(rqstp, &path);
>>  	if (IS_ERR(exp2)) {
>>  		err = PTR_ERR(exp2);
>> +		exp2 = NULL;
>>  		/*
>>  		 * We normally allow NFS clients to continue
>>  		 * "underneath" a mountpoint that is not exported.
>> @@ -157,10 +157,7 @@ nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
>>  		 */
>>  		if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
>>  			err = 0;
>> -		path_put(&path);
>> -		goto out;
>> -	}
>> -	if (nfsd_v4client(rqstp) ||
>> +	} else if (nfsd_v4client(rqstp) ||
>>  		(exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
>>  		/* successfully crossed mount point */
>>  		/*
>> @@ -174,9 +171,10 @@ nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
>>  		*expp = exp2;
>>  		exp2 = exp;
>>  	}
>> -	path_put(&path);
>> -	exp_put(exp2);
>>  out:
>> +	path_put(&path);
>> +	if (exp2)
>> +		exp_put(exp2);
>>  	return err;
>>  }
>>  
> 
> Looks reasonable. Chuck has already taken this patch into his tree, but
> we could do a cleanup on top. Want to send us an "official" patch?

We can replace any of the patches already in nfsd-testing, it's a topic
branch.


-- 
Chuck Lever