This patch series contains a set of minor optimisations that are mainly
useful against pNFS systems that may incur a higher performance penalty
on unnecessary attribute requests by triggering an internal RPC call from
the MDS to the DS.
- Speed up state recovery after a server reboot by avoiding unnecessary
GETATTR and ACCESS checks.
- Avoid unnecessary attribute revalidations on operations such as
OPEN_DOWNGRADE, where we don't yet need to update close-to-open
cache consistency attributes.
- Avoid unnecessary attribute revalidations when we hold a delegation.
Trond Myklebust (5):
NFSv4: Don't check file access when reclaiming state
NFSv4: Don't ask for the change attribute when reclaiming state
NFSv4: Don't request a GETATTR on open_downgrade.
NFSv4: Don't request close-to-open attribute when holding a delegation
NFSv4: Optimise away forced revalidation when we know the attributes
are OK
fs/nfs/delegation.c | 4 ----
fs/nfs/inode.c | 2 +-
fs/nfs/nfs4proc.c | 25 +++++++++++++++++++------
fs/nfs/nfs4xdr.c | 13 ++++---------
4 files changed, 24 insertions(+), 20 deletions(-)
--
2.9.3
We don't need to ask for the change attribute when returning a delegation
or recovering from a server reboot, and it could actually cause us to
obtain an incorrect value if we're using a pNFS flavour that requires
LAYOUTCOMMIT.
Signed-off-by: Trond Myklebust <[email protected]>
---
fs/nfs/nfs4proc.c | 1 -
1 file changed, 1 deletion(-)
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index b582df89c083..c0628f78ed98 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -226,7 +226,6 @@ static const u32 nfs4_pnfs_open_bitmap[3] = {
static const u32 nfs4_open_noattr_bitmap[3] = {
FATTR4_WORD0_TYPE
- | FATTR4_WORD0_CHANGE
| FATTR4_WORD0_FILEID,
};
--
2.9.3
If we're reclaiming state after a reboot, or as part of returning a
delegation, we don't need to check access modes again.
Signed-off-by: Trond Myklebust <[email protected]>
---
fs/nfs/nfs4proc.c | 14 +++++++++++---
1 file changed, 11 insertions(+), 3 deletions(-)
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 241da19b7da4..b582df89c083 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -1221,6 +1221,7 @@ static struct nfs4_opendata *nfs4_opendata_alloc(struct dentry *dentry,
atomic_inc(&sp->so_count);
p->o_arg.open_flags = flags;
p->o_arg.fmode = fmode & (FMODE_READ|FMODE_WRITE);
+ p->o_arg.claim = nfs4_map_atomic_open_claim(server, claim);
p->o_arg.share_access = nfs4_map_atomic_open_share(server,
fmode, flags);
/* don't put an ACCESS op in OPEN compound if O_EXCL, because ACCESS
@@ -1228,8 +1229,16 @@ static struct nfs4_opendata *nfs4_opendata_alloc(struct dentry *dentry,
if (!(flags & O_EXCL)) {
/* ask server to check for all possible rights as results
* are cached */
- p->o_arg.access = NFS4_ACCESS_READ | NFS4_ACCESS_MODIFY |
- NFS4_ACCESS_EXTEND | NFS4_ACCESS_EXECUTE;
+ switch (p->o_arg.claim) {
+ default:
+ break;
+ case NFS4_OPEN_CLAIM_NULL:
+ case NFS4_OPEN_CLAIM_FH:
+ p->o_arg.access = NFS4_ACCESS_READ |
+ NFS4_ACCESS_MODIFY |
+ NFS4_ACCESS_EXTEND |
+ NFS4_ACCESS_EXECUTE;
+ }
}
p->o_arg.clientid = server->nfs_client->cl_clientid;
p->o_arg.id.create_time = ktime_to_ns(sp->so_seqid.create_time);
@@ -1239,7 +1248,6 @@ static struct nfs4_opendata *nfs4_opendata_alloc(struct dentry *dentry,
p->o_arg.bitmask = nfs4_bitmask(server, label);
p->o_arg.open_bitmap = &nfs4_fattr_bitmap[0];
p->o_arg.label = nfs4_label_copy(p->a_label, label);
- p->o_arg.claim = nfs4_map_atomic_open_claim(server, claim);
switch (p->o_arg.claim) {
case NFS4_OPEN_CLAIM_NULL:
case NFS4_OPEN_CLAIM_DELEGATE_CUR:
--
2.9.3
If we're not closing the file completely, there is no need to request
close-to-open attributes.
Signed-off-by: Trond Myklebust <[email protected]>
---
fs/nfs/nfs4xdr.c | 10 ++--------
1 file changed, 2 insertions(+), 8 deletions(-)
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index fc89e5ed07ee..c37473721230 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -499,14 +499,12 @@ static int nfs4_stat_to_errno(int);
(compound_encode_hdr_maxsz + \
encode_sequence_maxsz + \
encode_putfh_maxsz + \
- encode_open_downgrade_maxsz + \
- encode_getattr_maxsz)
+ encode_open_downgrade_maxsz)
#define NFS4_dec_open_downgrade_sz \
(compound_decode_hdr_maxsz + \
decode_sequence_maxsz + \
decode_putfh_maxsz + \
- decode_open_downgrade_maxsz + \
- decode_getattr_maxsz)
+ decode_open_downgrade_maxsz)
#define NFS4_enc_close_sz (compound_encode_hdr_maxsz + \
encode_sequence_maxsz + \
encode_putfh_maxsz + \
@@ -2328,7 +2326,6 @@ static void nfs4_xdr_enc_open_downgrade(struct rpc_rqst *req,
encode_sequence(xdr, &args->seq_args, &hdr);
encode_putfh(xdr, args->fh, &hdr);
encode_open_downgrade(xdr, args, &hdr);
- encode_getfattr(xdr, args->bitmask, &hdr);
encode_nops(&hdr);
}
@@ -6115,9 +6112,6 @@ static int nfs4_xdr_dec_open_downgrade(struct rpc_rqst *rqstp,
if (status)
goto out;
status = decode_open_downgrade(xdr, res);
- if (status != 0)
- goto out;
- decode_getfattr(xdr, res->fattr, res->server);
out:
return status;
}
--
2.9.3
If holding a delegation, we do not need to ask the server to return
close-to-open cache consistency attributes as part of the CLOSE
compound.
Signed-off-by: Trond Myklebust <[email protected]>
---
fs/nfs/nfs4proc.c | 10 ++++++++--
fs/nfs/nfs4xdr.c | 3 ++-
2 files changed, 10 insertions(+), 3 deletions(-)
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index c0628f78ed98..917a6db5c84f 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -3156,8 +3156,15 @@ static void nfs4_close_prepare(struct rpc_task *task, void *data)
goto out_wait;
}
- if (calldata->arg.fmode == 0)
+ if (calldata->arg.fmode == 0) {
task->tk_msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_CLOSE];
+
+ /* Close-to-open cache consistency revalidation */
+ if (!nfs4_have_delegation(inode, FMODE_READ))
+ calldata->arg.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
+ else
+ calldata->arg.bitmask = NULL;
+ }
if (calldata->roc)
pnfs_roc_get_barrier(inode, &calldata->roc_barrier);
@@ -3240,7 +3247,6 @@ int nfs4_do_close(struct nfs4_state *state, gfp_t gfp_mask, int wait)
if (IS_ERR(calldata->arg.seqid))
goto out_free_calldata;
calldata->arg.fmode = 0;
- calldata->arg.bitmask = server->cache_consistency_bitmask;
calldata->res.fattr = &calldata->fattr;
calldata->res.seqid = calldata->arg.seqid;
calldata->res.server = server;
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index c37473721230..b54931503872 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -2248,7 +2248,8 @@ static void nfs4_xdr_enc_close(struct rpc_rqst *req, struct xdr_stream *xdr,
encode_sequence(xdr, &args->seq_args, &hdr);
encode_putfh(xdr, args->fh, &hdr);
encode_close(xdr, args, &hdr);
- encode_getfattr(xdr, args->bitmask, &hdr);
+ if (args->bitmask != NULL)
+ encode_getfattr(xdr, args->bitmask, &hdr);
encode_nops(&hdr);
}
--
2.9.3
The NFS_INO_REVAL_FORCED flag needs to be set if we just got a delegation,
and we see that there might still be some ambiguity as to whether or not
our attribute or data cache are valid.
In practice, this means that a call to nfs_check_inode_attributes() will
have noticed a discrepancy between cached attributes and measured ones,
so let's move the setting of NFS_INO_REVAL_FORCED to there.
Signed-off-by: Trond Myklebust <[email protected]>
---
fs/nfs/delegation.c | 4 ----
fs/nfs/inode.c | 2 +-
2 files changed, 1 insertion(+), 5 deletions(-)
diff --git a/fs/nfs/delegation.c b/fs/nfs/delegation.c
index dff600ae0d74..d7df5e67b0c1 100644
--- a/fs/nfs/delegation.c
+++ b/fs/nfs/delegation.c
@@ -391,10 +391,6 @@ int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct
rcu_assign_pointer(nfsi->delegation, delegation);
delegation = NULL;
- /* Ensure we revalidate the attributes and page cache! */
- spin_lock(&inode->i_lock);
- nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
- spin_unlock(&inode->i_lock);
trace_nfs4_set_delegation(inode, res->delegation_type);
out:
diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
index 372bfe98f685..02453d76bfea 100644
--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -1317,7 +1317,7 @@ static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fat
invalid |= NFS_INO_INVALID_ATIME;
if (invalid != 0)
- nfs_set_cache_invalid(inode, invalid);
+ nfs_set_cache_invalid(inode, invalid | NFS_INO_REVAL_FORCED);
nfsi->read_cache_jiffies = fattr->time_start;
return 0;
--
2.9.3
Hi Trond,
On 12/01/2016 05:06 PM, Trond Myklebust wrote:
> We don't need to ask for the change attribute when returning a delegation
> or recovering from a server reboot, and it could actually cause us to
> obtain an incorrect value if we're using a pNFS flavour that requires
> LAYOUTCOMMIT.
>
> Signed-off-by: Trond Myklebust <[email protected]>
> ---
> fs/nfs/nfs4proc.c | 1 -
> 1 file changed, 1 deletion(-)
>
> diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
> index b582df89c083..c0628f78ed98 100644
> --- a/fs/nfs/nfs4proc.c
> +++ b/fs/nfs/nfs4proc.c
> @@ -226,7 +226,6 @@ static const u32 nfs4_pnfs_open_bitmap[3] = {
>
> static const u32 nfs4_open_noattr_bitmap[3] = {
> FATTR4_WORD0_TYPE
> - | FATTR4_WORD0_CHANGE
Do these patches depend on another patch series? I'm having trouble applying this patch since my tree doesn't have FATTR4_WORD0_CHANGE yet. I'm having trouble with patch 4, too.
Thanks,
Anna
> | FATTR4_WORD0_FILEID,
> };
>
>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On 12/02/2016 12:28 PM, Trond Myklebust wrote:
>
>> On Dec 2, 2016, at 12:05, Anna Schumaker <[email protected]> wrote:
>>
>> Hi Trond,
>>
>> On 12/01/2016 05:06 PM, Trond Myklebust wrote:
>>> We don't need to ask for the change attribute when returning a delegation
>>> or recovering from a server reboot, and it could actually cause us to
>>> obtain an incorrect value if we're using a pNFS flavour that requires
>>> LAYOUTCOMMIT.
>>>
>>> Signed-off-by: Trond Myklebust <[email protected]>
>>> ---
>>> fs/nfs/nfs4proc.c | 1 -
>>> 1 file changed, 1 deletion(-)
>>>
>>> diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
>>> index b582df89c083..c0628f78ed98 100644
>>> --- a/fs/nfs/nfs4proc.c
>>> +++ b/fs/nfs/nfs4proc.c
>>> @@ -226,7 +226,6 @@ static const u32 nfs4_pnfs_open_bitmap[3] = {
>>>
>>> static const u32 nfs4_open_noattr_bitmap[3] = {
>>> FATTR4_WORD0_TYPE
>>> - | FATTR4_WORD0_CHANGE
>>
>> Do these patches depend on another patch series? I'm having trouble applying this patch since my tree doesn't have FATTR4_WORD0_CHANGE yet. I'm having trouble with patch 4, too.
>>
>
> I’ve pushed out the linux-next branch http://git.linux-nfs.org/?p=trondmy/linux-nfs.git;a=shortlog;h=refs/heads/linux-next
I'll look there, then! Thanks!
>
>> Thanks,
>> Anna
>>
>>> | FATTR4_WORD0_FILEID,
>>> };
>