Re: [PATCH V2 2/2] NFSD: allow client to use write delegation stateid for READ

[Dai Ngo <dai.ngo@oracle.com>]

On 2/24/25 6:48 AM, Chuck Lever wrote:
> On 2/21/25 6:42 PM, Dai Ngo wrote:
>> Allow READ using write delegation stateid granted on OPENs with
>> OPEN4_SHARE_ACCESS_WRITE only, to accommodate clients whose WRITE
>> implementation may unavoidably do (e.g., due to buffer cache
>> constraints).
>>
>> When the server offers a write delegation for an OPEN with
>> OPEN4_SHARE_ACCESS_WRITE, the file access mode, the nfs4_file
>> and nfs4_ol_stateid are upgraded as if the OPEN was sent with
>> OPEN4_SHARE_ACCESS_BOTH.
>>
>> When this delegation is returned or revoked, the corresponding open
>> stateid is looked up and if it's found then the file access mode,
>> the nfs4_file and nfs4_ol_stateid are downgraded to remove the read
>> access.
> I probably don't understand something. The patch description seems to
> suggest that a WR_ONLY OPEN state ID is also granted read in this case?

Currently nfsd allows a WR_ONLY OPEN state ID to do READ. The access check
is done in access_permit_read:

static inline int
access_permit_read(struct nfs4_ol_stateid *stp)
{
         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);       <<====
}

Is this behavior intentional or is it a bug?

-Dai

>
>
>> Signed-off-by: Dai Ngo <dai.ngo@oracle.com>
>> ---
>>   fs/nfsd/nfs4state.c | 62 +++++++++++++++++++++++++++++++++++++++++++++
>>   fs/nfsd/state.h     |  2 ++
>>   2 files changed, 64 insertions(+)
>>
>> diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
>> index b533225e57cf..0c14f902c54c 100644
>> --- a/fs/nfsd/nfs4state.c
>> +++ b/fs/nfsd/nfs4state.c
>> @@ -6126,6 +6126,51 @@ nfs4_delegation_stat(struct nfs4_delegation *dp, struct svc_fh *currentfh,
>>   	return rc == 0;
>>   }
>>   
>> +/*
>> + * Upgrade file access mode to include FMODE_READ. This is called only when
>> + * a write delegation is granted for an OPEN with OPEN4_SHARE_ACCESS_WRITE.
>> + */
>> +static void
>> +nfs4_upgrade_rdwr_file_access(struct nfs4_ol_stateid *stp)
>> +{
>> +	struct nfs4_file *fp = stp->st_stid.sc_file;
>> +	struct nfsd_file *nflp;
>> +	struct file *file;
>> +
>> +	spin_lock(&fp->fi_lock);
>> +	nflp = fp->fi_fds[O_WRONLY];
>> +	file = nflp->nf_file;
>> +	file->f_mode |= FMODE_READ;
>> +	swap(fp->fi_fds[O_RDWR], fp->fi_fds[O_WRONLY]);
>> +	clear_access(NFS4_SHARE_ACCESS_WRITE, stp);
>> +	set_access(NFS4_SHARE_ACCESS_BOTH, stp);
>> +	__nfs4_file_get_access(fp, NFS4_SHARE_ACCESS_READ);	/* incr fi_access[O_RDONLY] */
>> +	spin_unlock(&fp->fi_lock);
>> +}
>> +
>> +/*
>> + * Downgrade file access mode to remove FMODE_READ. This is called when
>> + * a write delegation, granted for an OPEN with OPEN4_SHARE_ACCESS_WRITE,
>> + * is returned.
>> + */
>> +static void
>> +nfs4_downgrade_wronly_file_access(struct nfs4_ol_stateid *stp)
>> +{
>> +	struct nfs4_file *fp = stp->st_stid.sc_file;
>> +	struct nfsd_file *nflp;
>> +	struct file *file;
>> +
>> +	spin_lock(&fp->fi_lock);
>> +	nflp = fp->fi_fds[O_RDWR];
>> +	file = nflp->nf_file;
>> +	file->f_mode &= ~FMODE_READ;
>> +	swap(fp->fi_fds[O_WRONLY], fp->fi_fds[O_RDWR]);
>> +	clear_access(NFS4_SHARE_ACCESS_BOTH, stp);
>> +	set_access(NFS4_SHARE_ACCESS_WRITE, stp);
>> +	spin_unlock(&fp->fi_lock);
>> +	nfs4_file_put_access(fp, NFS4_SHARE_ACCESS_READ);	/* decr. fi_access[O_RDONLY] */
>> +}
>> +
>>   /*
>>    * The Linux NFS server does not offer write delegations to NFSv4.0
>>    * clients in order to avoid conflicts between write delegations and
>> @@ -6207,6 +6252,11 @@ nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
>>   		dp->dl_cb_fattr.ncf_cur_fsize = stat.size;
>>   		dp->dl_cb_fattr.ncf_initial_cinfo = nfsd4_change_attribute(&stat);
>>   		trace_nfsd_deleg_write(&dp->dl_stid.sc_stateid);
>> +
>> +		if ((open->op_share_access & NFS4_SHARE_ACCESS_BOTH) == NFS4_SHARE_ACCESS_WRITE) {
>> +			dp->dl_stateid = stp->st_stid.sc_stateid;
>> +			nfs4_upgrade_rdwr_file_access(stp);
>> +		}
>>   	} else {
>>   		open->op_delegate_type = deleg_ts ? OPEN_DELEGATE_READ_ATTRS_DELEG :
>>   						    OPEN_DELEGATE_READ;
>> @@ -7710,6 +7760,8 @@ nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
>>   	struct nfs4_stid *s;
>>   	__be32 status;
>>   	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
>> +	struct nfs4_ol_stateid *stp;
>> +	struct nfs4_stid *stid;
>>   
>>   	if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
>>   		return status;
>> @@ -7724,6 +7776,16 @@ nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
>>   
>>   	trace_nfsd_deleg_return(stateid);
>>   	destroy_delegation(dp);
>> +
>> +	if (dp->dl_stateid.si_generation && dp->dl_stateid.si_opaque.so_id) {
>> +		if (!nfsd4_lookup_stateid(cstate, &dp->dl_stateid,
>> +				SC_TYPE_OPEN, 0, &stid, nn)) {
>> +			stp = openlockstateid(stid);
>> +			nfs4_downgrade_wronly_file_access(stp);
>> +			nfs4_put_stid(stid);
>> +		}
>> +	}
>> +
>>   	smp_mb__after_atomic();
>>   	wake_up_var(d_inode(cstate->current_fh.fh_dentry));
>>   put_stateid:
>> diff --git a/fs/nfsd/state.h b/fs/nfsd/state.h
>> index 74d2d7b42676..3f2f1b92db66 100644
>> --- a/fs/nfsd/state.h
>> +++ b/fs/nfsd/state.h
>> @@ -207,6 +207,8 @@ struct nfs4_delegation {
>>   
>>   	/* for CB_GETATTR */
>>   	struct nfs4_cb_fattr    dl_cb_fattr;
>> +
>> +	stateid_t		dl_stateid;  /* open stateid */
>>   };
>>   
>>   static inline bool deleg_is_read(u32 dl_type)
>