RDMA recv function needs to place data to the correct place starting at
page offset.
Signed-off-by: NLong Li <longli@microsoft.com>
Signed-off-by: NSteve French <stfrench@microsoft.com>

The RDMA send function needs to look at offset in the request pages, and
send data starting from there.
Signed-off-by: NLong Li <longli@microsoft.com>
Signed-off-by: NSteve French <stfrench@microsoft.com>

It's not necessary to allocate another iov when going through the buffers
in smbd_send() through RDMA send.

Remove it to reduce stack size.

Thanks to Matt for spotting a printk typo in the earlier version of this.

CC: Matt Redfearn <matt.redfearn@mips.com>
Signed-off-by: NLong Li <longli@microsoft.com>
Acked-by: NRonnie Sahlberg <lsahlber@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: NSteve French <smfrench@gmail.com>

When sending through SMB Direct, also dump the packet in SMB send path.

Also fixed a typo in debug message.
Signed-off-by: NLong Li <longli@microsoft.com>
Cc: stable@vger.kernel.org
Signed-off-by: NSteve French <smfrench@gmail.com>
Reviewed-by: NRonnie Sahlberg <lsahlber@redhat.com>

When sending the last iov that breaks into smaller buffers to fit the
transfer size, it's necessary to check if this is the last iov.

If this is the latest iov, stop and proceed to send pages.
Signed-off-by: NLong Li <longli@microsoft.com>
Cc: stable@vger.kernel.org
Signed-off-by: NSteve French <stfrench@microsoft.com>
Reviewed-by: NRonnie Sahlberg <lsahlber@redhat.com>

On RDMA errors, transport should disconnect the RDMA CM connection. This
will notify the upper layer, and it will attempt transport reconnect.
Signed-off-by: NLong Li <longli@microsoft.com>
Signed-off-by: NSteve French <smfrench@gmail.com>
Reviewed-by: NRonnie Sahlberg <lsahlber@redhat.com>
CC: Stable <stable@vger.kernel.org>

During transport reconnect, other processes may have registered memory
and blocked on transport. This creates a deadlock situation because the
transport resources can't be freed, and reconnect is blocked.

Fix this by returning to upper layer on timeout. Before returning,
transport status is set to reconnecting so other processes will release
memory registration resources.

Upper layer will retry the reconnect. This is not in fast I/O path so
setting the timeout to 5 seconds.
Signed-off-by: NLong Li <longli@microsoft.com>
Signed-off-by: NSteve French <smfrench@gmail.com>
Reviewed-by: NRonnie Sahlberg <lsahlber@redhat.com>
CC: Stable <stable@vger.kernel.org>

Trivial fix to spelling mistake in log_rdma_send and log_rdma_mr
message text.
Signed-off-by: NColin Ian King <colin.king@canonical.com>
Signed-off-by: NSteve French <smfrench@gmail.com>
Reviewed-by: NAurelien Aptel <aaptel@suse.com>

Although at least one of these was an overly strict sparse warning
in the new smbdirect code, it is cleaner to fix - so no warnings.
Signed-off-by: NSteve French <smfrench@gmail.com>
Reviewed-by: NRonnie Sahlberg <lsahlber@redhat.com>

Minor cleanup of some sparse warnings (including a few misc
endian fixes for the new smb3 rdma code)
Signed-off-by: NSteve French <smfrench@gmail.com>
Reviewed-by: NRonnie Sahlberg <lsahlber@redhat.com>

GCC versions from 4.9 to 6.3 produce a false-positive warning when
dealing with a conditional spin_lock_irqsave():

fs/cifs/smbdirect.c: In function 'smbd_recv_buf':
include/linux/spinlock.h:260:3: warning: 'flags' may be used uninitialized in this function [-Wmaybe-uninitialized]

This function calls some sleeping interfaces, so it is clear that it
does not get called with interrupts disabled and there is no need
to save the irq state before taking the spinlock. This lets us
remove the variable, which makes the function slightly more efficient
and avoids the warning.

A further cleanup could do the same change for other functions in this
file, but I did not want to take this too far for now.

Fixes: ac69f66e54ca ("CIFS: SMBD: Implement function to receive data via RDMA receive")
Signed-off-by: NArnd Bergmann <arnd@arndb.de>
Signed-off-by: NSteve French <smfrench@gmail.com>

Fixes: 07495ff5d9bc ("CIFS: SMBD: Establish SMB Direct connection")
Signed-off-by: NFengguang Wu <fengguang.wu@intel.com>
Signed-off-by: NSteve French <smfrench@gmail.com>
Acked-by: NLong Li <longli@microsoft.com>

Memory registration is used for transferring payload via RDMA read or write.
After I/O is done, memory registrations are recovered and reused. This
process can be time consuming and is done in a work queue.
Signed-off-by: NLong Li <longli@microsoft.com>
Reviewed-by: NPavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: NRonnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: NSteve French <smfrench@gmail.com>

The transport doesn't maintain send buffers or send queue for transferring
payload via RDMA send. There is no data copy in the transport on send.
Signed-off-by: NLong Li <longli@microsoft.com>
Signed-off-by: NSteve French <smfrench@gmail.com>
Reviewed-by: NPavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: NRonnie Sahlberg <lsahlber@redhat.com>

On the receive path, the transport maintains receive buffers and a reassembly
queue for transferring payload via RDMA recv. There is data copy in the
transport on recv when it copies the payload to upper layer.

The transport recognizes the RFC1002 header length use in the SMB
upper layer payloads in CIFS. Because this length is mainly used for TCP and
not applicable to RDMA, it is handled as a out-of-band information and is
never sent over the wire, and the trasnport behaves like TCP to upper layer
by processing and exposing the length correctly on data payloads.
Signed-off-by: NLong Li <longli@microsoft.com>
Signed-off-by: NSteve French <smfrench@gmail.com>
Reviewed-by: NPavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: NRonnie Sahlberg <lsahlber@redhat.com>

Add function to tear down a SMB Direct connection. This is used by upper layer
to free all SMB Direct connection and transport resources.
Signed-off-by: NLong Li <longli@microsoft.com>
Signed-off-by: NSteve French <smfrench@gmail.com>
Reviewed-by: NPavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: NRonnie Sahlberg <lsahlber@redhat.com>

Add function to implement a reconnect to SMB Direct. This involves tearing down
the current connection and establishing/negotiating a new connection.
Signed-off-by: NLong Li <longli@microsoft.com>
Signed-off-by: NSteve French <smfrench@gmail.com>
Reviewed-by: NPavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: NRonnie Sahlberg <lsahlber@redhat.com>

The upper layer calls this function to connect to peer through SMB Direct.
Each SMB Direct connection is based on a RDMA RC Queue Pair.
Signed-off-by: NLong Li <longli@microsoft.com>
Signed-off-by: NSteve French <smfrench@gmail.com>
Reviewed-by: NPavel Shilovsky <pshilov@microsoft.com>

Add code to implement the core functions to establish a SMB Direct connection.

1. Establish an RDMA connection to SMB server.
2. Negotiate and setup SMB Direct protocol.
3. Implement idle connection timer and credit management.

SMB Direct is enabled by setting CONFIG_CIFS_SMB_DIRECT.

Add to Makefile to enable building SMB Direct.
Signed-off-by: NLong Li <longli@microsoft.com>
Signed-off-by: NSteve French <smfrench@gmail.com>
Reviewed-by: NPavel Shilovsky <pshilov@microsoft.com>

To prepare for protocol implementation, add constants and user-configurable
values for the SMB Direct protocol.
Signed-off-by: NLong Li <longli@microsoft.com>
Signed-off-by: NSteve French <smfrench@gmail.com>
Acked-by: Ronnie Sahlberg <lsahlber.redhat.com>
Reviewed-by: NPavel Shilovsky <pshilov@microsoft.com>