This will allow us to use common functions soon.
Note this generates the following warnings from
scripts/checkpatch.pl --quiet:
WARNING: quoted string split across lines
#59: FILE: fs/smb/client/cifs_debug.c:481:
+ seq_printf(m, "\nDebug count_get_receive_buffer: %llu "
+ "count_put_receive_buffer: %llu count_send_empty: %llu",
WARNING: quoted string split across lines
#66: FILE: fs/smb/client/cifs_debug.c:486:
seq_printf(m, "\nRead Queue "
+ "count_enqueue_reassembly_queue: %llu "
WARNING: quoted string split across lines
#67: FILE: fs/smb/client/cifs_debug.c:487:
+ "count_enqueue_reassembly_queue: %llu "
+ "count_dequeue_reassembly_queue: %llu "
total: 0 errors, 3 warnings, 83 lines checked
scripts/checkpatch.pl: FAILED
But I left them in there, because it matches the code
arround it...
Cc: Steve French <smfrench@gmail.com>
Cc: Tom Talpey <tom@talpey.com>
Cc: Long Li <longli@microsoft.com>
Cc: linux-cifs@vger.kernel.org
Cc: samba-technical@lists.samba.org
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
This basically represents the same information as
sc->recv_io.reassembly.queue_length.
The only thing that's different is that
smbd_connection.count_reassembly_queue was updated in each
loop step, while sc->recv_io.reassembly.queue_length is only
updated once after the loop in smbd_recv.
Also sc->recv_io.reassembly.queue_length is updated under
a spinlock.
Cc: Steve French <smfrench@gmail.com>
Cc: Tom Talpey <tom@talpey.com>
Cc: Long Li <longli@microsoft.com>
Cc: linux-cifs@vger.kernel.org
Cc: samba-technical@lists.samba.org
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
We setup the first timer with the negotiate timeout and set
KEEP_ALIVE_PENDING, so that the expired timer disconnects.
On every incoming message we need to reset the timer to the keepalive
interval (120s).
On SMBDIRECT_FLAG_RESPONSE_REQUESTED we need to schedule a response
instead of setting KEEP_ALIVE_PENDING. Doing both would mean
we would also set SMBDIRECT_FLAG_RESPONSE_REQUESTED in that
response. If both ends would do that we'd play ping pong in
a busy loop.
If we move to KEEP_ALIVE_SENT and send the keepalive request
with SMBDIRECT_FLAG_RESPONSE_REQUESTED, we need to setup the
timer with keepalive timeout (5s) in order to disconnect
if no incoming message reset the timer.
The fired timer sets KEEP_ALIVE_PENDING and also
setup timer with keepalive timeout (5s) in order to disconnect
if no incoming message reset the timer.
We do that before queueing the send_immediate_work
and have that timer in case we didn't reach the send code
that typically sets the timer to keepalive timeout.
Cc: Steve French <smfrench@gmail.com>
Cc: Tom Talpey <tom@talpey.com>
Cc: Long Li <longli@microsoft.com>
Cc: linux-cifs@vger.kernel.org
Cc: samba-technical@lists.samba.org
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
(At least for me) the logic maintaining the count of posted
recv_io messages and the count of granted credits is much easier
to understand.
From there we can easily calculate the number of new_credits we'll
grant to the peer in outgoing send_io messages.
This will simplify the move to common logic that can be
shared between client and server in the following patches.
Cc: Steve French <smfrench@gmail.com>
Cc: Tom Talpey <tom@talpey.com>
Cc: Long Li <longli@microsoft.com>
Cc: linux-cifs@vger.kernel.org
Cc: samba-technical@lists.samba.org
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
We already call ib_drain_qp() before, which is an sync operation
that only returns in the queues are fully drained.
ib_drain_qp() completes pending requests with IB_WC_WR_FLUSH_ERR
so we have already called put_receive_buffer().
So all smbdirect_recv_io objects are either in the
smbdirect_socket.recv_io.free.list or
smbdirect_socket.recv_io.reassembly.list.
Then we explicitly iterate smbdirect_socket.recv_io.reassembly.list
and call put_receive_buffer(), so every object is
in smbdirect_socket.recv_io.free.list.
It means info->count_receive_queue == sp->recv_credit_max was
already true and calling wait_event(info->wait_receive_queues...
is pointless.
Cc: Steve French <smfrench@gmail.com>
Cc: Tom Talpey <tom@talpey.com>
Cc: Long Li <longli@microsoft.com>
Cc: linux-cifs@vger.kernel.org
Cc: samba-technical@lists.samba.org
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
If we're already disconnecting we don't need to queue the
disconnect_work again. disable_work() turns the next queue_work()
into a no-op.
Also let smbd_destroy() cancel(and disable) queued disconnect_work and
call smbd_disconnect_rdma_work() inline.
The makes it more obvious that disconnect_work is never queued
again after smbd_destroy() called smbd_disconnect_rdma_work().
It also means we have a single place to call rdma_disconnect().
While there we better also disable all other [delayed_]work.
Cc: Steve French <smfrench@gmail.com>
Cc: Tom Talpey <tom@talpey.com>
Cc: Long Li <longli@microsoft.com>
Cc: linux-cifs@vger.kernel.org
Cc: samba-technical@lists.samba.org
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
wake_up_interruptible[_all]() doesn't wake up tasks waiting
with wait_event().
So we better wake_up[_all]() in order to wake up all tasks in order
to simplify the logic.
As we currently don't use any wait_event_*_exclusive() it
doesn't really matter if we use wake_up() or wake_up_all().
But in this patch I try to use wake_up() for expected situations
and wake_up_all() for situations of a broken connection.
So don't need to adjust things in future when we
may use wait_event_*_exclusive() in order to wake up
only one process that should make progress.
Changing the wait_event_*() code in order to keep
wait_event(), wait_event_interruptible() and
wait_event_interruptible_timeout() or
changing them to wait_event_killable(),
wait_event_killable_timeout(),
wait_event_killable_exclusive()
is something to think about in a future patch.
The goal here is to avoid that some tasks are not
woken and freeze forever.
Also note that this patch only changes the existing
wake_up*() calls. Adding more wake_up*() calls for
other wait queues is also deferred to a future patch.
Link: https://lore.kernel.org/linux-cifs/13851363-0dc9-465c-9ced-3ede4904eef0@samba.org/T/#t
Cc: Steve French <smfrench@gmail.com>
Cc: Tom Talpey <tom@talpey.com>
Cc: Long Li <longli@microsoft.com>
Cc: linux-cifs@vger.kernel.org
Cc: samba-technical@lists.samba.org
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
For the reader it is not obvious that the counter values are displayed
in hex as there's no leading '0x' before '%x'.
So changed them to %u instead and added '0x' for non-counter values
and also a string for the status.
Note this generates some checkpatch warnings like this:
WARNING: quoted string split across lines
#45: FILE: fs/smb/client/cifs_debug.c:460:
+ seq_printf(m, "\nSMBDirect protocol version: 0x%x "
+ "transport status: %s (%u)",
But I'll leave this is the current style in the
related code...
Cc: Steve French <smfrench@gmail.com>
Cc: Tom Talpey <tom@talpey.com>
Cc: Long Li <longli@microsoft.com>
Cc: linux-cifs@vger.kernel.org
Cc: samba-technical@lists.samba.org
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
Do a real negotiation and check the servers initiator_depth and
responder_resources.
This should use big endian in order to be useful.
I have captures of windows clients showing this.
The fact that we used little endian up to now
means that we sent very large numbers and the
negotiation with the server truncated them to the
server limits.
Note the reason why this uses u8 for
initiator_depth and responder_resources is
that the rdma layer also uses it.
The inconsitency regarding the initiator_depth
and responder_resources values being reversed
for iwarp devices in RDMA_CM_EVENT_ESTABLISHED
should also be fixed later, but for now we should
fix it.
Cc: Steve French <smfrench@gmail.com>
Cc: Tom Talpey <tom@talpey.com>
Cc: Long Li <longli@microsoft.com>
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Cc: linux-cifs@vger.kernel.org
Cc: samba-technical@lists.samba.org
Cc: linux-rdma@vger.kernel.org
Fixes: c739858334 ("CIFS: SMBD: Implement RDMA memory registration")
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
This safer to start with and allows common code not care about if the
caller uses these or not. E.g. sc->mr_io.recovery_work is only used
on the client...
Note disable_[delayed_]work_sync() requires a valid function pointer
in INIT_[DELAYED_]WORK(). The non _sync() version don't require it,
but as we need to use the _sync() version on cleanup we better use
it here too, it won't block anyway here...
Cc: Steve French <smfrench@gmail.com>
Cc: Tom Talpey <tom@talpey.com>
Cc: Long Li <longli@microsoft.com>
Acked-by: Namjae Jeon <linkinjeon@kernel.org>
Cc: linux-cifs@vger.kernel.org
Cc: samba-technical@lists.samba.org
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
