On 1/24/25 9:50 AM, Jeff Layton wrote:
On Fri, 2025-01-24 at 09:43 -0500, Chuck Lever wrote:
On 1/23/25 3:25 PM, Jeff Layton wrote:
Add a new kerneldoc header, and clean up the comments a bit.
Usually I'm in favor of kdoc headers, but here, it's a static function
whose address is not shared outside of this source file. The only
documentation need is the meaning of the return code, IMO.
If you like. I figured it wouldn't hurt to do a full kdoc comment.
Signed-off-by: Jeff Layton <jlayton@xxxxxxxxxx>
---
fs/nfsd/nfs4callback.c | 26 ++++++++++++++++++++------
1 file changed, 20 insertions(+), 6 deletions(-)
diff --git a/fs/nfsd/nfs4callback.c b/fs/nfsd/nfs4callback.c
index 6e0561f3b21bd850b0387b5af7084eb05e818231..415fc8aae0f47c36f00b2384805c7a996fb1feb0 100644
--- a/fs/nfsd/nfs4callback.c
+++ b/fs/nfsd/nfs4callback.c
@@ -1325,6 +1325,17 @@ static void nfsd4_cb_prepare(struct rpc_task *task, void *calldata)
rpc_call_start(task);
}
+/**
+ * nfsd4_cb_sequence_done - process the result of a CB_SEQUENCE
+ * @task: rpc_task
+ * @cb: nfsd4_callback for this call
+ *
+ * For minorversion 0, there is no CB_SEQUENCE. Only restart the call
+ * if the callback RPC client was killed. For v4.1+ the error handling
+ * is more sophisticated.
It would be much clearer to pull the 4.0 error handling out of this
function, which is named "cb_/sequence/_done".
Perhaps the need_restart label can be hoisted into nfsd4_cb_done() ?
If we do that then we'll need to change this function to return
something other than a bool, and that's a larger change than I wanted
to make here.
I don't think that's needed. If you create a helper like so:
static bool nfsd4_cb_requeue(struct rpc_task *task,
struct nfsd4_callback *cb)
{
struct nfs4_client *clp = cb->cb_clp;
if (!test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags)) {
trace_nfsd_cb_restart(clp, cb);
task->tk_status = 0;
cb->cb_need_restart = true;
}
return false;
}
Then you can replace the "goto need_restart;" sites in both functions
with a tail call to this helper:
return nfsd4_cb_requeue(task, cb);
I really wanted to keep these as small, targeted patches
that can be backported easily.
Clean-ups are generally not back-portable, so I don't mind if you go to
a little extra trouble.
I wouldn't object to further cleanup here on top of that though.
+ *
+ * Returns true if reply processing should continue.
+ */
static bool nfsd4_cb_sequence_done(struct rpc_task *task, struct nfsd4_callback *cb)
{
struct nfs4_client *clp = cb->cb_clp;
@@ -1334,11 +1345,11 @@ static bool nfsd4_cb_sequence_done(struct rpc_task *task, struct nfsd4_callback
if (!clp->cl_minorversion) {
/*
* If the backchannel connection was shut down while this
- * task was queued, we need to resubmit it after setting up
- * a new backchannel connection.
+ * task was queued, resubmit it after setting up a new
+ * backchannel connection.
*
- * Note that if we lost our callback connection permanently
- * the submission code will error out, so we don't need to
+ * Note that if the callback connection is permanently lost,
+ * the submission code will error out. There is no need to
* handle that case here.
*/
if (RPC_SIGNALLED(task))
@@ -1355,8 +1366,6 @@ static bool nfsd4_cb_sequence_done(struct rpc_task *task, struct nfsd4_callback
switch (cb->cb_seq_status) {
case 0:
/*
- * No need for lock, access serialized in nfsd4_cb_prepare
- *
* RFC5661 20.9.3
* If CB_SEQUENCE returns an error, then the state of the slot
* (sequence ID, cached reply) MUST NOT change.
@@ -1365,6 +1374,11 @@ static bool nfsd4_cb_sequence_done(struct rpc_task *task, struct nfsd4_callback
ret = true;
break;
case -ESERVERFAULT:
+ /*
+ * Client returned NFS4_OK, but decoding failed. Mark the
+ * backchannel as faulty, but don't retransmit since the
+ * call was successful.
+ */
++session->se_cb_seq_nr[cb->cb_held_slot];
nfsd4_mark_cb_fault(cb->cb_clp);
break;
This old code abuses the meaning of ESERVERFAULT IMO. NFS4ERR_BADXDR is
a better choice. But why call mark_cb_fault in this case?
Maybe split this clean-up into a separate patch.
I'm only altering comments in this patch. Do you really want separate
patches for the different comments?
--
Chuck Lever
