Merge branch 'net-smc'

Ursula Braun says:

====================
net/smc: Shared Memory Communications - RDMA

here is now V4 of the SMC-R patches having processed your feedback from end
of November. The most important change is the replacement of sysfs by a
generic netlink solution in patch 04. And I tried to get rid of the __packed
attributes. There are still a few usages left due to SMC-R protocol defined
structures.

V4 changes:
The order of patches 03 and 04 for pnet table management and SMC IB-client
establishing has been exchanged, since pnet table management is now built on
top of smc_ib_devices.
Patch 01: Use EXPORT_SYMBOL_GPL().
Patch 02: Define "use_fallback" as bool.
          Get rid of useless smc_sock fields clearing in smc_sock_alloc(),
          since sk_alloc() clears out the memory.
Patch 03: Postpone smc_ib_remember_port_attr() call till ib_device is
          mentioned in the pnet table.
Patch 04: Replace sysfs-usage by a generic netlink approach for pnet table
          configuration.
          Change layout of pnet table entries to reference net_device and
          ib_device instead of dealing with names of net_devices and
          ib_devices.
Patch 05: Adapt "use_fallback" usages to new type bool.
          Get rid of useless smc_sock fields clearing in smc_sock_alloc()
          Avoid __packed where possible.
          Check if clc responses are not too big.
Patch 09: Postpone smc_setup_per_ibdev till the first connection with this
          ib_device is really created.
Patch 11: Get rid of __packed usage.

V3 changes:
Patch 05: Remove unneeded DEFINE_WAIT
Patch 06: Improve synchronization of link group creation
Patch 07: Rename peer_rmbe_len into peer_rmbe_size to be more consistent
Patch 09: Avoid calls of ib_get_memory_region with IB_ACCESS_LOCAL_WRITE,
          use new default local_dma_lkey from protection domain as lkey
          instead.
          Remove no longer needed function smc_ib_dereg_memory_region().
Patch 14: Switch to state ACTIVE only if still in state INIT.
          Return 0 for recvmsg invoked in a socket closing state.
          Allow getname call in state APPCLOSEWAIT1
          Do not trigger destruction of a socket-in-error queued in accept
          queue.
          During cleanup of accept queue, make sure sockets are destructed,
          and sockets in fallback mode are handled appropriately.
          When freeing sndbufs/rmbs, remove them from their list and free
          the entry.
          Use add_wait_queue() and remove_wait_queue() in close wait
          functions.
          If actively closing a socket in state for PEERFINCLOSEWAIT, keep
          this state.
          If passively closing a socket while bytes are to be received, move
          to state APPCLOSEWAIT1.
          If actively aborting a socket, skip sending the close_abort flag,
          since RDMA communication is no longer possible.
          When terminating a link group, do not schedule link group freeing a
          2nd time, since already done when unregistering the last remaining
          connection.
Patch 15: Introduce smc_diag module for monitoring SMC protocol sockets.
          This replaces the old patch 0015 dealing with procfs.

V2 changes:
Patch 0002: Add SMC versions for family key strings in net/core/sock.c.
Patch 0006: initialize rb_tree.
Patch 0007: Get rid of unneeded use of xchg() in smc_sndbuf_unuse() and
            smc_rmb_unuse().
Patch 0008: Correct error checking logic for ib_function calls.
            Define struct smc_link field wr_tx_id as atomic_long_t.
            Use "do_div" instead of "%" to be architecture-independent.
Patch 0009: Correct error checking logic for ib_function calls.
Patch 0011: Remove xchg() calls in cursor handling. Use atomic64_t for cursor
            overlays on 64-bit architectures. If not available, use plain u64
            and add locking for cursor reading and writing.
            Implement smc_curs_add() without modulo operator "%".
Patch 0012: Remove xchg() calls in cursor handling.
            Implement smc_tx_rdma_writes() without module operator "%".
Patch 0013: Remove xchg() calls in cursor handling.
Patch 0014: Return type bool in smc_wr_tx_has_pending().
            Remove unneeded semicolon in smc_close_shutdown_write().
            Call smc_close_active() in non-fallback case only.
            Get rid of duplicate schedule of sock_put_work().
            Take nested sock_lock in smc_listen_work().
            Start close stream_wait in case of prepared sends only.
Patch 0015: Remove unneeded socket ref_count in smc_proc_seq_show().
            Take lock before list_empty check in smc_proc_sock_list_del().

These patches are the initial part of the implementation of the
"Shared Memory Communications-RDMA" (SMC-R) protocol as defined in
RFC7609 [1]. While SMC-R does not aim to replace TCP,
it taps a wealth of existing data center TCP socket applications
to become more efficient without the need for rewriting them.
SMC-R uses RDMA over Converged Ethernet (RoCE) to save CPU consumption.
For instance, when running 10 parallel connections with uperf, we measured
a decrease of 60% in CPU consumption with SMC-R compared to TCP/IP
(with throughput and latency comparable;
measured on x86_64 with the same RoCE card and port).

SMC-R does not require an RDMA communication manager (RDMA CM).

SMC-R inherits TCP qualities such as reliable connections, host-based
firewall packet filtering (on connection establishment) and unmodified
application of communication encryption such as TLS (transport layer
security) or SSL (secure sockets layer). Since original TCP is used to
establish SMC-R connections, load balancers and packet inspection based
on TCP/IP connection establishment continue to work for SMC-R.

On the other hand, using SMC-R implies:
- either involving a preload library when invoking the unchanged TCP-application
  or slightly modifying the source by simply changing the socket family in
  the socket() call
- accepting extra overhead and latency in connection establishment due to
  SMC Connection Layer Control (CLC) handshake
- explicit coupling of RoCE ports with Ethernet ports
- not routable as currently built on RoCE V1
- bypassing of packet-based networking features
    - filtering (netfilter)
    - sniffing (libpcap, packet sockets, (E)BPF)
    - traffic control (scheduling, shaping)
- bypassing of IP-header based socket options
- bypassing of memory buffer (pressure) management
- unusable together with IPsec

Overview of the SMC-R Protocol described in informational RFC 7609

SMC-R is an open protocol that provides RDMA capabilities over RoCE
transparently for applications exploiting TCP sockets.
A new socket protocol family PF_SMC is introduced.
There are no changes required to applications using the sockets API for TCP
stream sockets other than the specification of the new socket family AF_SMC.
Unmodified applications can be used by means of a dynamic preload shared
library which rewrites the socket API call
socket(AF_INET, SOCK_STREAM, IPPROTO_TCP) into
socket(AF_SMC,  SOCK_STREAM, IPPROTO_TCP).
SMC-R re-uses the address family AF_INET for all addressing purposes around
struct sockaddr.

SMC-R system architecture layers:

+=============================================================================+
|                                      | unmodified TCP application           |
| native SMC application               +--------------------------------------+
|                                      | dynamic preload shared library       |
+=============================================================================+
|                                 SMC socket                                  |
+-----------------------------------------------------------------------------+
|                    | TCP socket (for connection establishment and fallback) |
| IB verbs           +--------------------------------------------------------+
|                    | IP                                                     |
+--------------------+--------------------------------------------------------+
| RoCE device driver | some network device driver                             |
+=============================================================================+

Terms:

A link group is determined by an ordered peer pair of TCP client and TCP server
(IP addresses and subnet). Reversed client server roles cause an own link group.
A link is a logical point-to-point connection based on an
infiniband reliable connected queue pair (RC-QP) between two RoCE ports
(MACs and GIDs) of a peer pair.
A link group can have 1..8 links for failover and load balancing.
This initial Linux implementation always has 1 link per link group.
Each link group on a peer can have 1..255 remote memory buffers (RMBs).
If more RMBs are needed, a peer can open another link group
(this initial Linux implementation) or fall back to TCP.
Each RMB has its own particular size and its own (R)DMA mapping and credentials
(rtoken consisting of rkey and RDMA "virtual address").
This initial Linux implementation uses physically contiguous memory for RMBs
but we are working towards scattered memory because of memory fragmentation.
Each RMB has 1..255 RMB elements (RMBEs) of equal size
to provide multiplexing of connections within an RMB.
An RMBE is the RDMA Write destination organized as wrapping ring buffer
for data transmit of a particular connection in one direction
(duplex by means of mirror symmetry as with TCP).
This initial Linux implementation always has 1 RMBE per RMB
and thus an individual RMB for each connection.

SMC-R connection establishment with subsequent data transfer:

   CLIENT                                                   SERVER

TCP three-way handshake:
                         regular TCP SYN
      -------------------------------------------------------->
                       regular TCP SYN ACK
      <--------------------------------------------------------
                         regular TCP ACK
      -------------------------------------------------------->

SMC Connection Layer Control (CLC) handshake
exchanges RDMA credentials between peers:
             via above TCP connection: SMC CLC Proposal
      -------------------------------------------------------->
              via above TCP connection: SMC CLC Accept
      <--------------------------------------------------------
             via above TCP connection: SMC CLC Confirm
      -------------------------------------------------------->

SMC Link Layer Control (LLC) (only once per link, i.e. 1st conn. of link group):
                 RoCE RC-QP: SMC LLC Confirm Link
      <========================================================
             RoCE RC-QP: SMC LLC Confirm Link response
      ========================================================>

SMC data transmission (incl. SMC Connection Data Control (CDC) message):
                       RoCE RC-QP: RDMA Write
      ========================================================>
             RoCE RC-QP: SMC CDC message (flow control)
      ========================================================>
                          ...

                       RoCE RC-QP: RDMA Write
      <========================================================
             RoCE RC-QP: SMC CDC message (flow control)
      <========================================================
                          ...

Data flow within an established connection:

+----------------------------------------------------------------------------
|            SENDER
| sendmsg()
|    |
|    | produces into sndbuf [sender's process context]
|    v
| +--------+
| | sndbuf | [ring buffer]
| +--------+
|    |
|    | consumes from sndbuf and produces into receiver's RMBE [any context]
|    | by sending RDMA Write followed by SMC CDC message over RoCE RC-QP
|    |
+----|-----------------------------------------------------------------------
     |
+----|-----------------------------------------------------------------------
|    v       RECEIVER
| +------+
| | RMBE | [ring buffer, can have size different from sender's sndbuf]
| |      | [RMBE represents rcvbuf, no further de-coupling as on sender side]
| +------+
|    |
|    | consumes from RMBE [receiver's process context]
|    v
| recvmsg()
+----------------------------------------------------------------------------

Flow control ("cursor" updates) by means of SMC CDC messages:

               SENDER                            RECEIVER

        sends updates via CDC-------------+   sends updates via CDC
        on consuming from sndbuf          |   on consuming from RMBE
        and producing into RMBE           |   by means of recvmsg()
                                          |            |
                                          |            |
      +-----------------------------------|------------+
      |                                   |
   +--v-------------------------+      +--v-----------------------+
   | receiver's consumer cursor |      | sender's producer cursor----+
   +----------------|-----------+      +--------------------------+  |
                    |                                                |
                    |                        receiver's RMBE         |
                    |                  +--------------------------+  |
                    |                  |                          |  |
                    +--------------------------------+            |  |
                                       |             |            |  |
                                       |             v            |  |
                                       |             +------------|  |
                                       |-------------+////////////|  |
                                       |//RDMA data written by////|  |
                                       |////sender that is////////|  |
                                       |/available to be consumed/|  |
                                       |///////// +---------------|  |
                                       |----------+^              |  |
                                       |           |              |  |
                                       |           +-----------------+
                                       |                          |
                                       +--------------------------+

Sending updates of the producer cursor is immediate for low latency;
something like Nagle's algorithm (absence of TCP_NODELAY) is optional and
currently not part of this initial Linux implementation.
Sending updates of the consumer cursor is conditional to avoid the
silly window syndrome.

Normal connection termination:

Normal connection termination starts transitioning from socket state
ACTIVE via either "Active Close" or "Passive Close".

shutdown rdwr               +-----------------+
or close,   +-------------->|  INIT / CLOSED  |<-------------+
send PeerCon|nClosed        +-----------------+              | PeerConnClosed
            |                       |                        | received
            |            connection | established            |
            |                       V                        |
    +----------------+     +-----------------+     +----------------+
    |AppFinCloseWait |     |     ACTIVE      |     |PeerFinCloseWait|
    +----------------+     +-----------------+     +----------------+
            |                   |         |                   |
            |     Active Close: |         |Passive Close:     |
            |     close or      |         |PeerConnClosed or  |
            |     shutdown wr or|         |PeerDoneWriting    |
            |     shutdown rdwr |         |received           |

    |                   V         V                   |
 PeerConnClo|sed    +--------------+   +-------------+        | close or
 received   +--<----|PeerCloseWait1|   |AppCloseWait1|--->----+ shutdown rdwr,
            |       +--------------+   +-------------+        | send
            |  PeerDoneWri|ting                | shutdown wr, | PeerConnClosed
            |  received   |            send Pee|rDoneWriting  |
            |             V                    V              |
            |       +--------------+   +-------------+        |
            +--<----|PeerCloseWait2|   |AppCloseWait2|--->----+
                    +--------------+   +-------------+

In state CLOSED, the socket can be destructed only, once the application has
issued a close().

Abnormal connection termination:

                            +-----------------+
            +-------------->|  INIT / CLOSED  |<-------------+
            |               +-----------------+              |
            |                                                |
            |           +-----------------------+            |
            |           |     Any state         |            |
 PeerConnAbo|rt         | (before setting       |            | send
 received   |           |  PeerConnClosed       |            | PeerConnAbort
            |           |  indicator in         |            |
            |           |  peer's RMBE)         |            |
            |           +-----------------------+            |
            |                   |         |                  |
            |     Active Abort: |         | Passive Abort:   |
            |     problem,      |         | PeerConnAbort    |
            |     send          |         | received,        |
            |     PeerConnAbort,|         | ECONNRESET       |
            |     ECONNABORTED  |         |                  |
            |                   V         V                  |
            |       +--------------+   +--------------+      |
            +-------|PeerAbortWait |   | ProcessAbort |------+
                    +--------------+   +--------------+

Implementation notes beyond RFC 7609:

A PNET table in sysfs provides the mapping between network device names and
RoCE Infiniband device names for the transparent switch of data communication.
A PNET table can contain an arbitrary number of PNETIDs.
Each PNETID contains exactly one (Ethernet) network device name
and one or more RoCE Infiniband device names.
Each device name can only exist in at most one PNETID (no overlapping).
This initial Linux implementation allows at most one RoCE Infiniband device
name per PNETID.
After a new TCP connection is established, the network device name
used for egress traffic with the TCP connection's local source IP address
is used as key to lookup the unique PNETID, and the RoCE Infiniband device
of this PNETID is used to switch data communication from TCP to RDMA
during SMC CLC handshake.

Problem determination:

A protocol dissector is available with upstream wireshark for formatting
SMC-R related RoCE LAN traffic.
[https://code.wireshark.org/review/gitweb?p=wireshark.git;a=blob;f=epan/dissectors/packet-smcr.c]

We are working on enhancing the Linux implementation to cover:

- Improve default socket closing asynchronicity
- Address corner cases with many parallel connections
- Tracing
- Integrated load balancing and fail-over within a link group
- Splice and sendpage support
- IPv6 addressing support
- Keepalive, Cork
- Namespaces support
- Urgent data
- More socket options
- Diagnostics
- Statistics support
- SNMP support

References:

[1] SMC-R Informational RFC: http://www.rfc-editor.org/info/rfc7609
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

MAINTAINERS

@@ -10850,6 +10850,13 @@ S:	Maintained
 F:	drivers/staging/media/st-cec/
 F:	Documentation/devicetree/bindings/media/stih-cec.txt
 
+SHARED MEMORY COMMUNICATIONS (SMC) SOCKETS
+M:	Ursula Braun <ubraun@linux.vnet.ibm.com>
+L:	linux-s390@vger.kernel.org
+W:	http://www.ibm.com/developerworks/linux/linux390/
+S:	Supported
+F:	net/smc/
+
 SYNOPSYS DESIGNWARE DMAC DRIVER
 M:	Viresh Kumar <vireshk@kernel.org>
 M:	Andy Shevchenko <andriy.shevchenko@linux.intel.com>

include/linux/socket.h

@@ -202,8 +202,12 @@ struct ucred {
 #define AF_VSOCK	40	/* vSockets			*/
 #define AF_KCM		41	/* Kernel Connection Multiplexor*/
 #define AF_QIPCRTR	42	/* Qualcomm IPC Router          */
+#define AF_SMC		43	/* smc sockets: reserve number for
+				 * PF_SMC protocol family that
+				 * reuses AF_INET address family
+				 */
 
-#define AF_MAX		43	/* For now.. */
+#define AF_MAX		44	/* For now.. */
 
 /* Protocol families, same as address families. */
 #define PF_UNSPEC	AF_UNSPEC
@@ -251,6 +255,7 @@ struct ucred {
 #define PF_VSOCK	AF_VSOCK
 #define PF_KCM		AF_KCM
 #define PF_QIPCRTR	AF_QIPCRTR
+#define PF_SMC		AF_SMC
 #define PF_MAX		AF_MAX
 
 /* Maximum queue length specifiable by listen.  */

include/net/smc.h

@@ -0,0 +1,20 @@
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Definitions for the SMC module (socket related)
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+#ifndef _SMC_H
+#define _SMC_H
+
+struct smc_hashinfo {
+	rwlock_t lock;
+	struct hlist_head ht;
+};
+
+int smc_hash_sk(struct sock *sk);
+void smc_unhash_sk(struct sock *sk);
+#endif	/* _SMC_H */

include/net/sock.h

@@ -70,6 +70,7 @@
 #include <net/checksum.h>
 #include <net/tcp_states.h>
 #include <linux/net_tstamp.h>
+#include <net/smc.h>
 
 /*
  * This structure really needs to be cleaned up.
@@ -986,6 +987,7 @@ struct request_sock_ops;
 struct timewait_sock_ops;
 struct inet_hashinfo;
 struct raw_hashinfo;
+struct smc_hashinfo;
 struct module;
 
 /*
@@ -1024,6 +1026,7 @@ struct proto {
 	int			(*getsockopt)(struct sock *sk, int level,
 					int optname, char __user *optval,
 					int __user *option);
+	void			(*keepalive)(struct sock *sk, int valbool);
 #ifdef CONFIG_COMPAT
 	int			(*compat_setsockopt)(struct sock *sk,
 					int level,
@@ -1093,6 +1096,7 @@ struct proto {
 		struct inet_hashinfo	*hashinfo;
 		struct udp_table	*udp_table;
 		struct raw_hashinfo	*raw_hash;
+		struct smc_hashinfo	*smc_hash;
 	} h;
 
 	struct module		*owner;

include/uapi/linux/netlink.h

@@ -27,6 +27,7 @@
 #define NETLINK_ECRYPTFS	19
 #define NETLINK_RDMA		20
 #define NETLINK_CRYPTO		21	/* Crypto layer */
+#define NETLINK_SMC		22	/* SMC monitoring */
 
 #define NETLINK_INET_DIAG	NETLINK_SOCK_DIAG
 

include/uapi/linux/smc.h

@@ -0,0 +1,35 @@
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Definitions for generic netlink based configuration of an SMC-R PNET table
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Thomas Richter <tmricht@linux.vnet.ibm.com>
+ */
+
+#ifndef _UAPI_LINUX_SMC_H_
+#define _UAPI_LINUX_SMC_H_
+
+/* Netlink SMC_PNETID attributes */
+enum {
+	SMC_PNETID_UNSPEC,
+	SMC_PNETID_NAME,
+	SMC_PNETID_ETHNAME,
+	SMC_PNETID_IBNAME,
+	SMC_PNETID_IBPORT,
+	__SMC_PNETID_MAX,
+	SMC_PNETID_MAX = __SMC_PNETID_MAX - 1
+};
+
+enum {				/* SMC PNET Table commands */
+	SMC_PNETID_GET = 1,
+	SMC_PNETID_ADD,
+	SMC_PNETID_DEL,
+	SMC_PNETID_FLUSH
+};
+
+#define SMCR_GENL_FAMILY_NAME		"SMC_PNETID"
+#define SMCR_GENL_FAMILY_VERSION	1
+
+#endif /* _UAPI_LINUX_SMC_H */

include/uapi/linux/smc_diag.h

@@ -0,0 +1,85 @@
+#ifndef _UAPI_SMC_DIAG_H_
+#define _UAPI_SMC_DIAG_H_
+
+#include <linux/types.h>
+#include <linux/inet_diag.h>
+#include <rdma/ib_verbs.h>
+
+/* Request structure */
+struct smc_diag_req {
+	__u8	diag_family;
+	__u8	pad[2];
+	__u8	diag_ext;		/* Query extended information */
+	struct inet_diag_sockid	id;
+};
+
+/* Base info structure. It contains socket identity (addrs/ports/cookie) based
+ * on the internal clcsock, and more SMC-related socket data
+ */
+struct smc_diag_msg {
+	__u8	diag_family;
+	__u8	diag_state;
+	__u8	diag_fallback;
+	__u8	diag_shutdown;
+	struct inet_diag_sockid id;
+
+	__u32	diag_uid;
+	__u64	diag_inode;
+};
+
+/* Extensions */
+
+enum {
+	SMC_DIAG_NONE,
+	SMC_DIAG_CONNINFO,
+	SMC_DIAG_LGRINFO,
+	SMC_DIAG_SHUTDOWN,
+	__SMC_DIAG_MAX,
+};
+
+#define SMC_DIAG_MAX (__SMC_DIAG_MAX - 1)
+
+/* SMC_DIAG_CONNINFO */
+
+struct smc_diag_cursor {
+	__u16	reserved;
+	__u16	wrap;
+	__u32	count;
+};
+
+struct smc_diag_conninfo {
+	__u32			token;		/* unique connection id */
+	__u32			sndbuf_size;	/* size of send buffer */
+	__u32			rmbe_size;	/* size of RMB element */
+	__u32			peer_rmbe_size;	/* size of peer RMB element */
+	/* local RMB element cursors */
+	struct smc_diag_cursor	rx_prod;	/* received producer cursor */
+	struct smc_diag_cursor	rx_cons;	/* received consumer cursor */
+	/* peer RMB element cursors */
+	struct smc_diag_cursor	tx_prod;	/* sent producer cursor */
+	struct smc_diag_cursor	tx_cons;	/* sent consumer cursor */
+	__u8			rx_prod_flags;	/* received producer flags */
+	__u8			rx_conn_state_flags; /* recvd connection flags*/
+	__u8			tx_prod_flags;	/* sent producer flags */
+	__u8			tx_conn_state_flags; /* sent connection flags*/
+	/* send buffer cursors */
+	struct smc_diag_cursor	tx_prep;	/* prepared to be sent cursor */
+	struct smc_diag_cursor	tx_sent;	/* sent cursor */
+	struct smc_diag_cursor	tx_fin;		/* confirmed sent cursor */
+};
+
+/* SMC_DIAG_LINKINFO */
+
+struct smc_diag_linkinfo {
+	__u8 link_id;			/* link identifier */
+	__u8 ibname[IB_DEVICE_NAME_MAX]; /* name of the RDMA device */
+	__u8 ibport;			/* RDMA device port number */
+	__u8 gid[40];			/* local GID */
+	__u8 peer_gid[40];		/* peer GID */
+};
+
+struct smc_diag_lgrinfo {
+	struct smc_diag_linkinfo	lnk[1];
+	__u8				role;
+};
+#endif /* _UAPI_SMC_DIAG_H_ */

net/Kconfig

@@ -57,6 +57,7 @@ source "net/packet/Kconfig"
 source "net/unix/Kconfig"
 source "net/xfrm/Kconfig"
 source "net/iucv/Kconfig"
+source "net/smc/Kconfig"
 
 config INET
 	bool "TCP/IP networking"

net/Makefile

@@ -51,6 +51,7 @@ obj-$(CONFIG_MAC80211)		+= mac80211/
 obj-$(CONFIG_TIPC)		+= tipc/
 obj-$(CONFIG_NETLABEL)		+= netlabel/
 obj-$(CONFIG_IUCV)		+= iucv/
+obj-$(CONFIG_SMC)		+= smc/
 obj-$(CONFIG_RFKILL)		+= rfkill/
 obj-$(CONFIG_NET_9P)		+= 9p/
 obj-$(CONFIG_CAIF)		+= caif/

net/core/sock.c

@@ -222,7 +222,7 @@ static const char *const af_family_key_strings[AF_MAX+1] = {
   "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN"     , "sk_lock-AF_PHONET"   ,
   "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG"      ,
   "sk_lock-AF_NFC"   , "sk_lock-AF_VSOCK"    , "sk_lock-AF_KCM"      ,
-  "sk_lock-AF_MAX"
+  "sk_lock-AF_SMC"   , "sk_lock-AF_MAX"
 };
 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
   "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
@@ -239,7 +239,7 @@ static const char *const af_family_slock_key_strings[AF_MAX+1] = {
   "slock-AF_RXRPC" , "slock-AF_ISDN"     , "slock-AF_PHONET"   ,
   "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG"      ,
   "slock-AF_NFC"   , "slock-AF_VSOCK"    ,"slock-AF_KCM"       ,
-  "slock-AF_MAX"
+  "slock-AF_SMC"   , "slock-AF_MAX"
 };
 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
   "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
@@ -256,7 +256,7 @@ static const char *const af_family_clock_key_strings[AF_MAX+1] = {
   "clock-AF_RXRPC" , "clock-AF_ISDN"     , "clock-AF_PHONET"   ,
   "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG"      ,
   "clock-AF_NFC"   , "clock-AF_VSOCK"    , "clock-AF_KCM"      ,
-  "clock-AF_MAX"
+  "closck-AF_smc"  , "clock-AF_MAX"
 };
 
 /*
@@ -762,11 +762,8 @@ set_rcvbuf:
 		goto set_rcvbuf;
 
 	case SO_KEEPALIVE:
-#ifdef CONFIG_INET
-		if (sk->sk_protocol == IPPROTO_TCP &&
-		    sk->sk_type == SOCK_STREAM)
-			tcp_set_keepalive(sk, valbool);
-#endif
+		if (sk->sk_prot->keepalive)
+			sk->sk_prot->keepalive(sk, valbool);
 		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 		break;
 

net/ipv4/tcp_ipv4.c

@@ -2376,6 +2376,7 @@ struct proto tcp_prot = {
 	.shutdown		= tcp_shutdown,
 	.setsockopt		= tcp_setsockopt,
 	.getsockopt		= tcp_getsockopt,
+	.keepalive		= tcp_set_keepalive,
 	.recvmsg		= tcp_recvmsg,
 	.sendmsg		= tcp_sendmsg,
 	.sendpage		= tcp_sendpage,

net/ipv4/tcp_timer.c

@@ -617,6 +617,7 @@ void tcp_set_keepalive(struct sock *sk, int val)
 	else if (!val)
 		inet_csk_delete_keepalive_timer(sk);
 }
+EXPORT_SYMBOL_GPL(tcp_set_keepalive);
 
 
 static void tcp_keepalive_timer (unsigned long data)

net/ipv6/tcp_ipv6.c

@@ -1889,6 +1889,7 @@ struct proto tcpv6_prot = {
 	.shutdown		= tcp_shutdown,
 	.setsockopt		= tcp_setsockopt,
 	.getsockopt		= tcp_getsockopt,
+	.keepalive		= tcp_set_keepalive,
 	.recvmsg		= tcp_recvmsg,
 	.sendmsg		= tcp_sendmsg,
 	.sendpage		= tcp_sendpage,

net/smc/Kconfig

@@ -0,0 +1,20 @@
+config SMC
+	tristate "SMC socket protocol family"
+	depends on INET && INFINIBAND
+	---help---
+	  SMC-R provides a "sockets over RDMA" solution making use of
+	  RDMA over Converged Ethernet (RoCE) technology to upgrade
+	  AF_INET TCP connections transparently.
+	  The Linux implementation of the SMC-R solution is designed as
+	  a separate socket family SMC.
+
+	  Select this option if you want to run SMC socket applications
+
+config SMC_DIAG
+	tristate "SMC: socket monitoring interface"
+	depends on SMC
+	---help---
+	  Support for SMC socket monitoring interface used by tools such as
+	  smcss.
+
+	  if unsure, say Y.

net/smc/Makefile

@@ -0,0 +1,4 @@
+obj-$(CONFIG_SMC)	+= smc.o
+obj-$(CONFIG_SMC_DIAG)	+= smc_diag.o
+smc-y := af_smc.o smc_pnet.o smc_ib.o smc_clc.o smc_core.o smc_wr.o smc_llc.o
+smc-y += smc_cdc.o smc_tx.o smc_rx.o smc_close.o

net/smc/af_smc.c

@@ -0,0 +1,1407 @@
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  AF_SMC protocol family socket handler keeping the AF_INET sock address type
+ *  applies to SOCK_STREAM sockets only
+ *  offers an alternative communication option for TCP-protocol sockets
+ *  applicable with RoCE-cards only
+ *
+ *  Initial restrictions:
+ *    - non-blocking connect postponed
+ *    - IPv6 support postponed
+ *    - support for alternate links postponed
+ *    - partial support for non-blocking sockets only
+ *    - support for urgent data postponed
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ *              based on prototype from Frank Blaschka
+ */
+
+#define KMSG_COMPONENT "smc"
+#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
+
+#include <linux/module.h>
+#include <linux/socket.h>
+#include <linux/inetdevice.h>
+#include <linux/workqueue.h>
+#include <linux/in.h>
+#include <net/sock.h>
+#include <net/tcp.h>
+#include <net/smc.h>
+
+#include "smc.h"
+#include "smc_clc.h"
+#include "smc_llc.h"
+#include "smc_cdc.h"
+#include "smc_core.h"
+#include "smc_ib.h"
+#include "smc_pnet.h"
+#include "smc_tx.h"
+#include "smc_rx.h"
+#include "smc_close.h"
+
+static DEFINE_MUTEX(smc_create_lgr_pending);	/* serialize link group
+						 * creation
+						 */
+
+struct smc_lgr_list smc_lgr_list = {		/* established link groups */
+	.lock = __SPIN_LOCK_UNLOCKED(smc_lgr_list.lock),
+	.list = LIST_HEAD_INIT(smc_lgr_list.list),
+};
+
+static void smc_tcp_listen_work(struct work_struct *);
+
+static void smc_set_keepalive(struct sock *sk, int val)
+{
+	struct smc_sock *smc = smc_sk(sk);
+
+	smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
+}
+
+static struct smc_hashinfo smc_v4_hashinfo = {
+	.lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
+};
+
+int smc_hash_sk(struct sock *sk)
+{
+	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
+	struct hlist_head *head;
+
+	head = &h->ht;
+
+	write_lock_bh(&h->lock);
+	sk_add_node(sk, head);
+	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
+	write_unlock_bh(&h->lock);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(smc_hash_sk);
+
+void smc_unhash_sk(struct sock *sk)
+{
+	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
+
+	write_lock_bh(&h->lock);
+	if (sk_del_node_init(sk))
+		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
+	write_unlock_bh(&h->lock);
+}
+EXPORT_SYMBOL_GPL(smc_unhash_sk);
+
+struct proto smc_proto = {
+	.name		= "SMC",
+	.owner		= THIS_MODULE,
+	.keepalive	= smc_set_keepalive,
+	.hash		= smc_hash_sk,
+	.unhash		= smc_unhash_sk,
+	.obj_size	= sizeof(struct smc_sock),
+	.h.smc_hash	= &smc_v4_hashinfo,
+	.slab_flags	= SLAB_DESTROY_BY_RCU,
+};
+EXPORT_SYMBOL_GPL(smc_proto);
+
+static int smc_release(struct socket *sock)
+{
+	struct sock *sk = sock->sk;
+	struct smc_sock *smc;
+	int rc = 0;
+
+	if (!sk)
+		goto out;
+
+	smc = smc_sk(sk);
+	sock_hold(sk);
+	if (sk->sk_state == SMC_LISTEN)
+		/* smc_close_non_accepted() is called and acquires
+		 * sock lock for child sockets again
+		 */
+		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
+	else
+		lock_sock(sk);
+
+	if (smc->use_fallback) {
+		sk->sk_state = SMC_CLOSED;
+		sk->sk_state_change(sk);
+	} else {
+		rc = smc_close_active(smc);
+		sock_set_flag(sk, SOCK_DEAD);
+		sk->sk_shutdown |= SHUTDOWN_MASK;
+	}
+	if (smc->clcsock) {
+		sock_release(smc->clcsock);
+		smc->clcsock = NULL;
+	}
+
+	/* detach socket */
+	sock_orphan(sk);
+	sock->sk = NULL;
+	if (smc->use_fallback) {
+		schedule_delayed_work(&smc->sock_put_work, TCP_TIMEWAIT_LEN);
+	} else if (sk->sk_state == SMC_CLOSED) {
+		smc_conn_free(&smc->conn);
+		schedule_delayed_work(&smc->sock_put_work,
+				      SMC_CLOSE_SOCK_PUT_DELAY);
+	}
+	sk->sk_prot->unhash(sk);
+	release_sock(sk);
+
+	sock_put(sk);
+out:
+	return rc;
+}
+
+static void smc_destruct(struct sock *sk)
+{
+	if (sk->sk_state != SMC_CLOSED)
+		return;
+	if (!sock_flag(sk, SOCK_DEAD))
+		return;
+
+	sk_refcnt_debug_dec(sk);
+}
+
+static struct sock *smc_sock_alloc(struct net *net, struct socket *sock)
+{
+	struct smc_sock *smc;
+	struct sock *sk;
+
+	sk = sk_alloc(net, PF_SMC, GFP_KERNEL, &smc_proto, 0);
+	if (!sk)
+		return NULL;
+
+	sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
+	sk->sk_state = SMC_INIT;
+	sk->sk_destruct = smc_destruct;
+	sk->sk_protocol = SMCPROTO_SMC;
+	smc = smc_sk(sk);
+	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
+	INIT_LIST_HEAD(&smc->accept_q);
+	spin_lock_init(&smc->accept_q_lock);
+	INIT_DELAYED_WORK(&smc->sock_put_work, smc_close_sock_put_work);
+	sk->sk_prot->hash(sk);
+	sk_refcnt_debug_inc(sk);
+
+	return sk;
+}
+
+static int smc_bind(struct socket *sock, struct sockaddr *uaddr,
+		    int addr_len)
+{
+	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
+	struct sock *sk = sock->sk;
+	struct smc_sock *smc;
+	int rc;
+
+	smc = smc_sk(sk);
+
+	/* replicate tests from inet_bind(), to be safe wrt. future changes */
+	rc = -EINVAL;
+	if (addr_len < sizeof(struct sockaddr_in))
+		goto out;
+
+	rc = -EAFNOSUPPORT;
+	/* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
+	if ((addr->sin_family != AF_INET) &&
+	    ((addr->sin_family != AF_UNSPEC) ||
+	     (addr->sin_addr.s_addr != htonl(INADDR_ANY))))
+		goto out;
+
+	lock_sock(sk);
+
+	/* Check if socket is already active */
+	rc = -EINVAL;
+	if (sk->sk_state != SMC_INIT)
+		goto out_rel;
+
+	smc->clcsock->sk->sk_reuse = sk->sk_reuse;
+	rc = kernel_bind(smc->clcsock, uaddr, addr_len);
+
+out_rel:
+	release_sock(sk);
+out:
+	return rc;
+}
+
+static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
+				   unsigned long mask)
+{
+	/* options we don't get control via setsockopt for */
+	nsk->sk_type = osk->sk_type;
+	nsk->sk_sndbuf = osk->sk_sndbuf;
+	nsk->sk_rcvbuf = osk->sk_rcvbuf;
+	nsk->sk_sndtimeo = osk->sk_sndtimeo;
+	nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
+	nsk->sk_mark = osk->sk_mark;
+	nsk->sk_priority = osk->sk_priority;
+	nsk->sk_rcvlowat = osk->sk_rcvlowat;
+	nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
+	nsk->sk_err = osk->sk_err;
+
+	nsk->sk_flags &= ~mask;
+	nsk->sk_flags |= osk->sk_flags & mask;
+}
+
+#define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
+			     (1UL << SOCK_KEEPOPEN) | \
+			     (1UL << SOCK_LINGER) | \
+			     (1UL << SOCK_BROADCAST) | \
+			     (1UL << SOCK_TIMESTAMP) | \
+			     (1UL << SOCK_DBG) | \
+			     (1UL << SOCK_RCVTSTAMP) | \
+			     (1UL << SOCK_RCVTSTAMPNS) | \
+			     (1UL << SOCK_LOCALROUTE) | \
+			     (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
+			     (1UL << SOCK_RXQ_OVFL) | \
+			     (1UL << SOCK_WIFI_STATUS) | \
+			     (1UL << SOCK_NOFCS) | \
+			     (1UL << SOCK_FILTER_LOCKED))
+/* copy only relevant settings and flags of SOL_SOCKET level from smc to
+ * clc socket (since smc is not called for these options from net/core)
+ */
+static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
+{
+	smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
+}
+
+#define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
+			     (1UL << SOCK_KEEPOPEN) | \
+			     (1UL << SOCK_LINGER) | \
+			     (1UL << SOCK_DBG))
+/* copy only settings and flags relevant for smc from clc to smc socket */
+static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
+{
+	smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
+}
+
+/* determine subnet and mask of internal TCP socket */
+int smc_netinfo_by_tcpsk(struct socket *clcsock,
+			 __be32 *subnet, u8 *prefix_len)
+{
+	struct dst_entry *dst = sk_dst_get(clcsock->sk);
+	struct sockaddr_in addr;
+	int rc = -ENOENT;
+	int len;
+
+	if (!dst) {
+		rc = -ENOTCONN;
+		goto out;
+	}
+	if (!dst->dev) {
+		rc = -ENODEV;
+		goto out_rel;
+	}
+
+	/* get address to which the internal TCP socket is bound */
+	kernel_getsockname(clcsock, (struct sockaddr *)&addr, &len);
+	/* analyze IPv4 specific data of net_device belonging to TCP socket */
+	for_ifa(dst->dev->ip_ptr) {
+		if (ifa->ifa_address != addr.sin_addr.s_addr)
+			continue;
+		*prefix_len = inet_mask_len(ifa->ifa_mask);
+		*subnet = ifa->ifa_address & ifa->ifa_mask;
+		rc = 0;
+		break;
+	} endfor_ifa(dst->dev->ip_ptr);
+
+out_rel:
+	dst_release(dst);
+out:
+	return rc;
+}
+
+static int smc_clnt_conf_first_link(struct smc_sock *smc, union ib_gid *gid)
+{
+	struct smc_link_group *lgr = smc->conn.lgr;
+	struct smc_link *link;
+	int rest;
+	int rc;
+
+	link = &lgr->lnk[SMC_SINGLE_LINK];
+	/* receive CONFIRM LINK request from server over RoCE fabric */
+	rest = wait_for_completion_interruptible_timeout(
+		&link->llc_confirm,
+		SMC_LLC_WAIT_FIRST_TIME);
+	if (rest <= 0) {
+		struct smc_clc_msg_decline dclc;
+
+		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
+				      SMC_CLC_DECLINE);
+		return rc;
+	}
+
+	rc = smc_ib_modify_qp_rts(link);
+	if (rc)
+		return SMC_CLC_DECL_INTERR;
+
+	smc_wr_remember_qp_attr(link);
+	/* send CONFIRM LINK response over RoCE fabric */
+	rc = smc_llc_send_confirm_link(link,
+				       link->smcibdev->mac[link->ibport - 1],
+				       gid, SMC_LLC_RESP);
+	if (rc < 0)
+		return SMC_CLC_DECL_TCL;
+
+	return rc;
+}
+
+static void smc_conn_save_peer_info(struct smc_sock *smc,
+				    struct smc_clc_msg_accept_confirm *clc)
+{
+	smc->conn.peer_conn_idx = clc->conn_idx;
+	smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token);
+	smc->conn.peer_rmbe_size = smc_uncompress_bufsize(clc->rmbe_size);
+	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
+}
+
+static void smc_link_save_peer_info(struct smc_link *link,
+				    struct smc_clc_msg_accept_confirm *clc)
+{
+	link->peer_qpn = ntoh24(clc->qpn);
+	memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE);
+	memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac));
+	link->peer_psn = ntoh24(clc->psn);
+	link->peer_mtu = clc->qp_mtu;
+}
+
+/* setup for RDMA connection of client */
+static int smc_connect_rdma(struct smc_sock *smc)
+{
+	struct sockaddr_in *inaddr = (struct sockaddr_in *)smc->addr;
+	struct smc_clc_msg_accept_confirm aclc;
+	int local_contact = SMC_FIRST_CONTACT;
+	struct smc_ib_device *smcibdev;
+	struct smc_link *link;
+	u8 srv_first_contact;
+	int reason_code = 0;
+	int rc = 0;
+	u8 ibport;
+
+	/* IPSec connections opt out of SMC-R optimizations */
+	if (using_ipsec(smc)) {
+		reason_code = SMC_CLC_DECL_IPSEC;
+		goto decline_rdma;
+	}
+
+	/* PNET table look up: search active ib_device and port
+	 * within same PNETID that also contains the ethernet device
+	 * used for the internal TCP socket
+	 */
+	smc_pnet_find_roce_resource(smc->clcsock->sk, &smcibdev, &ibport);
+	if (!smcibdev) {
+		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
+		goto decline_rdma;
+	}
+
+	/* do inband token exchange */
+	reason_code = smc_clc_send_proposal(smc, smcibdev, ibport);
+	if (reason_code < 0) {
+		rc = reason_code;
+		goto out_err;
+	}
+	if (reason_code > 0) /* configuration error */
+		goto decline_rdma;
+	/* receive SMC Accept CLC message */
+	reason_code = smc_clc_wait_msg(smc, &aclc, sizeof(aclc),
+				       SMC_CLC_ACCEPT);
+	if (reason_code < 0) {
+		rc = reason_code;
+		goto out_err;
+	}
+	if (reason_code > 0)
+		goto decline_rdma;
+
+	srv_first_contact = aclc.hdr.flag;
+	mutex_lock(&smc_create_lgr_pending);
+	local_contact = smc_conn_create(smc, inaddr->sin_addr.s_addr, smcibdev,
+					ibport, &aclc.lcl, srv_first_contact);
+	if (local_contact < 0) {
+		rc = local_contact;
+		if (rc == -ENOMEM)
+			reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
+		else if (rc == -ENOLINK)
+			reason_code = SMC_CLC_DECL_SYNCERR; /* synchr. error */
+		goto decline_rdma_unlock;
+	}
+	link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK];
+
+	smc_conn_save_peer_info(smc, &aclc);
+
+	rc = smc_sndbuf_create(smc);
+	if (rc) {
+		reason_code = SMC_CLC_DECL_MEM;
+		goto decline_rdma_unlock;
+	}
+	rc = smc_rmb_create(smc);
+	if (rc) {
+		reason_code = SMC_CLC_DECL_MEM;
+		goto decline_rdma_unlock;
+	}
+
+	if (local_contact == SMC_FIRST_CONTACT)
+		smc_link_save_peer_info(link, &aclc);
+
+	rc = smc_rmb_rtoken_handling(&smc->conn, &aclc);
+	if (rc) {
+		reason_code = SMC_CLC_DECL_INTERR;
+		goto decline_rdma_unlock;
+	}
+
+	if (local_contact == SMC_FIRST_CONTACT) {
+		rc = smc_ib_ready_link(link);
+		if (rc) {
+			reason_code = SMC_CLC_DECL_INTERR;
+			goto decline_rdma_unlock;
+		}
+	}
+
+	rc = smc_clc_send_confirm(smc);
+	if (rc)
+		goto out_err_unlock;
+
+	if (local_contact == SMC_FIRST_CONTACT) {
+		/* QP confirmation over RoCE fabric */
+		reason_code = smc_clnt_conf_first_link(
+			smc, &smcibdev->gid[ibport - 1]);
+		if (reason_code < 0) {
+			rc = reason_code;
+			goto out_err_unlock;
+		}
+		if (reason_code > 0)
+			goto decline_rdma_unlock;
+	}
+
+	mutex_unlock(&smc_create_lgr_pending);
+	smc_tx_init(smc);
+	smc_rx_init(smc);
+
+out_connected:
+	smc_copy_sock_settings_to_clc(smc);
+	if (smc->sk.sk_state == SMC_INIT)
+		smc->sk.sk_state = SMC_ACTIVE;
+
+	return rc ? rc : local_contact;
+
+decline_rdma_unlock:
+	mutex_unlock(&smc_create_lgr_pending);
+	smc_conn_free(&smc->conn);
+decline_rdma:
+	/* RDMA setup failed, switch back to TCP */
+	smc->use_fallback = true;
+	if (reason_code && (reason_code != SMC_CLC_DECL_REPLY)) {
+		rc = smc_clc_send_decline(smc, reason_code, 0);
+		if (rc < sizeof(struct smc_clc_msg_decline))
+			goto out_err;
+	}
+	goto out_connected;
+
+out_err_unlock:
+	mutex_unlock(&smc_create_lgr_pending);
+	smc_conn_free(&smc->conn);
+out_err:
+	return rc;
+}
+
+static int smc_connect(struct socket *sock, struct sockaddr *addr,
+		       int alen, int flags)
+{
+	struct sock *sk = sock->sk;
+	struct smc_sock *smc;
+	int rc = -EINVAL;
+
+	smc = smc_sk(sk);
+
+	/* separate smc parameter checking to be safe */
+	if (alen < sizeof(addr->sa_family))
+		goto out_err;
+	if (addr->sa_family != AF_INET)
+		goto out_err;
+	smc->addr = addr;	/* needed for nonblocking connect */
+
+	lock_sock(sk);
+	switch (sk->sk_state) {
+	default:
+		goto out;
+	case SMC_ACTIVE:
+		rc = -EISCONN;
+		goto out;
+	case SMC_INIT:
+		rc = 0;
+		break;
+	}
+
+	smc_copy_sock_settings_to_clc(smc);
+	rc = kernel_connect(smc->clcsock, addr, alen, flags);
+	if (rc)
+		goto out;
+
+	/* setup RDMA connection */
+	rc = smc_connect_rdma(smc);
+	if (rc < 0)
+		goto out;
+	else
+		rc = 0; /* success cases including fallback */
+
+out:
+	release_sock(sk);
+out_err:
+	return rc;
+}
+
+static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
+{
+	struct sock *sk = &lsmc->sk;
+	struct socket *new_clcsock;
+	struct sock *new_sk;
+	int rc;
+
+	release_sock(&lsmc->sk);
+	new_sk = smc_sock_alloc(sock_net(sk), NULL);
+	if (!new_sk) {
+		rc = -ENOMEM;
+		lsmc->sk.sk_err = ENOMEM;
+		*new_smc = NULL;
+		lock_sock(&lsmc->sk);
+		goto out;
+	}
+	*new_smc = smc_sk(new_sk);
+
+	rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0);
+	lock_sock(&lsmc->sk);
+	if  (rc < 0) {
+		lsmc->sk.sk_err = -rc;
+		new_sk->sk_state = SMC_CLOSED;
+		sock_set_flag(new_sk, SOCK_DEAD);
+		sk->sk_prot->unhash(new_sk);
+		sock_put(new_sk);
+		*new_smc = NULL;
+		goto out;
+	}
+	if (lsmc->sk.sk_state == SMC_CLOSED) {
+		if (new_clcsock)
+			sock_release(new_clcsock);
+		new_sk->sk_state = SMC_CLOSED;
+		sock_set_flag(new_sk, SOCK_DEAD);
+		sk->sk_prot->unhash(new_sk);
+		sock_put(new_sk);
+		*new_smc = NULL;
+		goto out;
+	}
+
+	(*new_smc)->clcsock = new_clcsock;
+out:
+	return rc;
+}
+
+/* add a just created sock to the accept queue of the listen sock as
+ * candidate for a following socket accept call from user space
+ */
+static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
+{
+	struct smc_sock *par = smc_sk(parent);
+
+	sock_hold(sk);
+	spin_lock(&par->accept_q_lock);
+	list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
+	spin_unlock(&par->accept_q_lock);
+	sk_acceptq_added(parent);
+}
+
+/* remove a socket from the accept queue of its parental listening socket */
+static void smc_accept_unlink(struct sock *sk)
+{
+	struct smc_sock *par = smc_sk(sk)->listen_smc;
+
+	spin_lock(&par->accept_q_lock);
+	list_del_init(&smc_sk(sk)->accept_q);
+	spin_unlock(&par->accept_q_lock);
+	sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
+	sock_put(sk);
+}
+
+/* remove a sock from the accept queue to bind it to a new socket created
+ * for a socket accept call from user space
+ */
+struct sock *smc_accept_dequeue(struct sock *parent,
+				struct socket *new_sock)
+{
+	struct smc_sock *isk, *n;
+	struct sock *new_sk;
+
+	list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
+		new_sk = (struct sock *)isk;
+
+		smc_accept_unlink(new_sk);
+		if (new_sk->sk_state == SMC_CLOSED) {
+			/* tbd in follow-on patch: close this sock */
+			continue;
+		}
+		if (new_sock)
+			sock_graft(new_sk, new_sock);
+		return new_sk;
+	}
+	return NULL;
+}
+
+/* clean up for a created but never accepted sock */
+void smc_close_non_accepted(struct sock *sk)
+{
+	struct smc_sock *smc = smc_sk(sk);
+
+	sock_hold(sk);
+	lock_sock(sk);
+	if (!sk->sk_lingertime)
+		/* wait for peer closing */
+		sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
+	if (!smc->use_fallback)
+		smc_close_active(smc);
+	if (smc->clcsock) {
+		struct socket *tcp;
+
+		tcp = smc->clcsock;
+		smc->clcsock = NULL;
+		sock_release(tcp);
+	}
+	sock_set_flag(sk, SOCK_DEAD);
+	sk->sk_shutdown |= SHUTDOWN_MASK;
+	if (smc->use_fallback) {
+		schedule_delayed_work(&smc->sock_put_work, TCP_TIMEWAIT_LEN);
+	} else {
+		smc_conn_free(&smc->conn);
+		schedule_delayed_work(&smc->sock_put_work,
+				      SMC_CLOSE_SOCK_PUT_DELAY);
+	}
+	release_sock(sk);
+	sock_put(sk);
+}
+
+static int smc_serv_conf_first_link(struct smc_sock *smc)
+{
+	struct smc_link_group *lgr = smc->conn.lgr;
+	struct smc_link *link;
+	int rest;
+	int rc;
+
+	link = &lgr->lnk[SMC_SINGLE_LINK];
+	/* send CONFIRM LINK request to client over the RoCE fabric */
+	rc = smc_llc_send_confirm_link(link,
+				       link->smcibdev->mac[link->ibport - 1],
+				       &link->smcibdev->gid[link->ibport - 1],
+				       SMC_LLC_REQ);
+	if (rc < 0)
+		return SMC_CLC_DECL_TCL;
+
+	/* receive CONFIRM LINK response from client over the RoCE fabric */
+	rest = wait_for_completion_interruptible_timeout(
+		&link->llc_confirm_resp,
+		SMC_LLC_WAIT_FIRST_TIME);
+	if (rest <= 0) {
+		struct smc_clc_msg_decline dclc;
+
+		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
+				      SMC_CLC_DECLINE);
+	}
+
+	return rc;
+}
+
+/* setup for RDMA connection of server */
+static void smc_listen_work(struct work_struct *work)
+{
+	struct smc_sock *new_smc = container_of(work, struct smc_sock,
+						smc_listen_work);
+	struct socket *newclcsock = new_smc->clcsock;
+	struct smc_sock *lsmc = new_smc->listen_smc;
+	struct smc_clc_msg_accept_confirm cclc;
+	int local_contact = SMC_REUSE_CONTACT;
+	struct sock *newsmcsk = &new_smc->sk;
+	struct smc_clc_msg_proposal pclc;
+	struct smc_ib_device *smcibdev;
+	struct sockaddr_in peeraddr;
+	struct smc_link *link;
+	int reason_code = 0;
+	int rc = 0, len;
+	__be32 subnet;
+	u8 prefix_len;
+	u8 ibport;
+
+	/* do inband token exchange -
+	 *wait for and receive SMC Proposal CLC message
+	 */
+	reason_code = smc_clc_wait_msg(new_smc, &pclc, sizeof(pclc),
+				       SMC_CLC_PROPOSAL);
+	if (reason_code < 0)
+		goto out_err;
+	if (reason_code > 0)
+		goto decline_rdma;
+
+	/* IPSec connections opt out of SMC-R optimizations */
+	if (using_ipsec(new_smc)) {
+		reason_code = SMC_CLC_DECL_IPSEC;
+		goto decline_rdma;
+	}
+
+	/* PNET table look up: search active ib_device and port
+	 * within same PNETID that also contains the ethernet device
+	 * used for the internal TCP socket
+	 */
+	smc_pnet_find_roce_resource(newclcsock->sk, &smcibdev, &ibport);
+	if (!smcibdev) {
+		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
+		goto decline_rdma;
+	}
+
+	/* determine subnet and mask from internal TCP socket */
+	rc = smc_netinfo_by_tcpsk(newclcsock, &subnet, &prefix_len);
+	if (rc) {
+		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
+		goto decline_rdma;
+	}
+	if ((pclc.outgoing_subnet != subnet) ||
+	    (pclc.prefix_len != prefix_len)) {
+		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
+		goto decline_rdma;
+	}
+
+	/* get address of the peer connected to the internal TCP socket */
+	kernel_getpeername(newclcsock, (struct sockaddr *)&peeraddr, &len);
+
+	/* allocate connection / link group */
+	mutex_lock(&smc_create_lgr_pending);
+	local_contact = smc_conn_create(new_smc, peeraddr.sin_addr.s_addr,
+					smcibdev, ibport, &pclc.lcl, 0);
+	if (local_contact == SMC_REUSE_CONTACT)
+		/* lock no longer needed, free it due to following
+		 * smc_clc_wait_msg() call
+		 */
+		mutex_unlock(&smc_create_lgr_pending);
+	if (local_contact < 0) {
+		rc = local_contact;
+		if (rc == -ENOMEM)
+			reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
+		else if (rc == -ENOLINK)
+			reason_code = SMC_CLC_DECL_SYNCERR; /* synchr. error */
+		goto decline_rdma;
+	}
+	link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
+
+	rc = smc_sndbuf_create(new_smc);
+	if (rc) {
+		reason_code = SMC_CLC_DECL_MEM;
+		goto decline_rdma;
+	}
+	rc = smc_rmb_create(new_smc);
+	if (rc) {
+		reason_code = SMC_CLC_DECL_MEM;
+		goto decline_rdma;
+	}
+
+	rc = smc_clc_send_accept(new_smc, local_contact);
+	if (rc)
+		goto out_err;
+
+	/* receive SMC Confirm CLC message */
+	reason_code = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
+				       SMC_CLC_CONFIRM);
+	if (reason_code < 0)
+		goto out_err;
+	if (reason_code > 0)
+		goto decline_rdma;
+	smc_conn_save_peer_info(new_smc, &cclc);
+	if (local_contact == SMC_FIRST_CONTACT)
+		smc_link_save_peer_info(link, &cclc);
+
+	rc = smc_rmb_rtoken_handling(&new_smc->conn, &cclc);
+	if (rc) {
+		reason_code = SMC_CLC_DECL_INTERR;
+		goto decline_rdma;
+	}
+
+	if (local_contact == SMC_FIRST_CONTACT) {
+		rc = smc_ib_ready_link(link);
+		if (rc) {
+			reason_code = SMC_CLC_DECL_INTERR;
+			goto decline_rdma;
+		}
+		/* QP confirmation over RoCE fabric */
+		reason_code = smc_serv_conf_first_link(new_smc);
+		if (reason_code < 0) {
+			/* peer is not aware of a problem */
+			rc = reason_code;
+			goto out_err;
+		}
+		if (reason_code > 0)
+			goto decline_rdma;
+	}
+
+	smc_tx_init(new_smc);
+	smc_rx_init(new_smc);
+
+out_connected:
+	sk_refcnt_debug_inc(newsmcsk);
+	if (newsmcsk->sk_state == SMC_INIT)
+		newsmcsk->sk_state = SMC_ACTIVE;
+enqueue:
+	if (local_contact == SMC_FIRST_CONTACT)
+		mutex_unlock(&smc_create_lgr_pending);
+	lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
+	if (lsmc->sk.sk_state == SMC_LISTEN) {
+		smc_accept_enqueue(&lsmc->sk, newsmcsk);
+	} else { /* no longer listening */
+		smc_close_non_accepted(newsmcsk);
+	}
+	release_sock(&lsmc->sk);
+
+	/* Wake up accept */
+	lsmc->sk.sk_data_ready(&lsmc->sk);
+	sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
+	return;
+
+decline_rdma:
+	/* RDMA setup failed, switch back to TCP */
+	smc_conn_free(&new_smc->conn);
+	new_smc->use_fallback = true;
+	if (reason_code && (reason_code != SMC_CLC_DECL_REPLY)) {
+		rc = smc_clc_send_decline(new_smc, reason_code, 0);
+		if (rc < sizeof(struct smc_clc_msg_decline))
+			goto out_err;
+	}
+	goto out_connected;
+
+out_err:
+	newsmcsk->sk_state = SMC_CLOSED;
+	smc_conn_free(&new_smc->conn);
+	goto enqueue; /* queue new sock with sk_err set */
+}
+
+static void smc_tcp_listen_work(struct work_struct *work)
+{
+	struct smc_sock *lsmc = container_of(work, struct smc_sock,
+					     tcp_listen_work);
+	struct smc_sock *new_smc;
+	int rc = 0;
+
+	lock_sock(&lsmc->sk);
+	while (lsmc->sk.sk_state == SMC_LISTEN) {
+		rc = smc_clcsock_accept(lsmc, &new_smc);
+		if (rc)
+			goto out;
+		if (!new_smc)
+			continue;
+
+		new_smc->listen_smc = lsmc;
+		new_smc->use_fallback = false; /* assume rdma capability first*/
+		sock_hold(&lsmc->sk); /* sock_put in smc_listen_work */
+		INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
+		smc_copy_sock_settings_to_smc(new_smc);
+		schedule_work(&new_smc->smc_listen_work);
+	}
+
+out:
+	release_sock(&lsmc->sk);
+	lsmc->sk.sk_data_ready(&lsmc->sk); /* no more listening, wake accept */
+}
+
+static int smc_listen(struct socket *sock, int backlog)
+{
+	struct sock *sk = sock->sk;
+	struct smc_sock *smc;
+	int rc;
+
+	smc = smc_sk(sk);
+	lock_sock(sk);
+
+	rc = -EINVAL;
+	if ((sk->sk_state != SMC_INIT) && (sk->sk_state != SMC_LISTEN))
+		goto out;
+
+	rc = 0;
+	if (sk->sk_state == SMC_LISTEN) {
+		sk->sk_max_ack_backlog = backlog;
+		goto out;
+	}
+	/* some socket options are handled in core, so we could not apply
+	 * them to the clc socket -- copy smc socket options to clc socket
+	 */
+	smc_copy_sock_settings_to_clc(smc);
+
+	rc = kernel_listen(smc->clcsock, backlog);
+	if (rc)
+		goto out;
+	sk->sk_max_ack_backlog = backlog;
+	sk->sk_ack_backlog = 0;
+	sk->sk_state = SMC_LISTEN;
+	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
+	schedule_work(&smc->tcp_listen_work);
+
+out:
+	release_sock(sk);
+	return rc;
+}
+
+static int smc_accept(struct socket *sock, struct socket *new_sock,
+		      int flags)
+{
+	struct sock *sk = sock->sk, *nsk;
+	DECLARE_WAITQUEUE(wait, current);
+	struct smc_sock *lsmc;
+	long timeo;
+	int rc = 0;
+
+	lsmc = smc_sk(sk);
+	lock_sock(sk);
+
+	if (lsmc->sk.sk_state != SMC_LISTEN) {
+		rc = -EINVAL;
+		goto out;
+	}
+
+	/* Wait for an incoming connection */
+	timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
+	add_wait_queue_exclusive(sk_sleep(sk), &wait);
+	while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
+		set_current_state(TASK_INTERRUPTIBLE);
+		if (!timeo) {
+			rc = -EAGAIN;
+			break;
+		}
+		release_sock(sk);
+		timeo = schedule_timeout(timeo);
+		/* wakeup by sk_data_ready in smc_listen_work() */
+		sched_annotate_sleep();
+		lock_sock(sk);
+		if (signal_pending(current)) {
+			rc = sock_intr_errno(timeo);
+			break;
+		}
+	}
+	set_current_state(TASK_RUNNING);
+	remove_wait_queue(sk_sleep(sk), &wait);
+
+	if (!rc)
+		rc = sock_error(nsk);
+
+out:
+	release_sock(sk);
+	return rc;
+}
+
+static int smc_getname(struct socket *sock, struct sockaddr *addr,
+		       int *len, int peer)
+{
+	struct smc_sock *smc;
+
+	if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
+	    (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
+		return -ENOTCONN;
+
+	smc = smc_sk(sock->sk);
+
+	return smc->clcsock->ops->getname(smc->clcsock, addr, len, peer);
+}
+
+static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
+{
+	struct sock *sk = sock->sk;
+	struct smc_sock *smc;
+	int rc = -EPIPE;
+
+	smc = smc_sk(sk);
+	lock_sock(sk);
+	if ((sk->sk_state != SMC_ACTIVE) &&
+	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
+	    (sk->sk_state != SMC_INIT))
+		goto out;
+	if (smc->use_fallback)
+		rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
+	else
+		rc = smc_tx_sendmsg(smc, msg, len);
+out:
+	release_sock(sk);
+	return rc;
+}
+
+static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
+		       int flags)
+{
+	struct sock *sk = sock->sk;
+	struct smc_sock *smc;
+	int rc = -ENOTCONN;
+
+	smc = smc_sk(sk);
+	lock_sock(sk);
+	if ((sk->sk_state == SMC_INIT) ||
+	    (sk->sk_state == SMC_LISTEN) ||
+	    (sk->sk_state == SMC_CLOSED))
+		goto out;
+
+	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
+		rc = 0;
+		goto out;
+	}
+
+	if (smc->use_fallback)
+		rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
+	else
+		rc = smc_rx_recvmsg(smc, msg, len, flags);
+
+out:
+	release_sock(sk);
+	return rc;
+}
+
+static unsigned int smc_accept_poll(struct sock *parent)
+{
+	struct smc_sock *isk;
+	struct sock *sk;
+
+	lock_sock(parent);
+	list_for_each_entry(isk, &smc_sk(parent)->accept_q, accept_q) {
+		sk = (struct sock *)isk;
+
+		if (sk->sk_state == SMC_ACTIVE) {
+			release_sock(parent);
+			return POLLIN | POLLRDNORM;
+		}
+	}
+	release_sock(parent);
+
+	return 0;
+}
+
+static unsigned int smc_poll(struct file *file, struct socket *sock,
+			     poll_table *wait)
+{
+	struct sock *sk = sock->sk;
+	unsigned int mask = 0;
+	struct smc_sock *smc;
+	int rc;
+
+	smc = smc_sk(sock->sk);
+	if ((sk->sk_state == SMC_INIT) || smc->use_fallback) {
+		/* delegate to CLC child sock */
+		mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
+		/* if non-blocking connect finished ... */
+		lock_sock(sk);
+		if ((sk->sk_state == SMC_INIT) && (mask & POLLOUT)) {
+			sk->sk_err = smc->clcsock->sk->sk_err;
+			if (sk->sk_err) {
+				mask |= POLLERR;
+			} else {
+				rc = smc_connect_rdma(smc);
+				if (rc < 0)
+					mask |= POLLERR;
+				else
+					/* success cases including fallback */
+					mask |= POLLOUT | POLLWRNORM;
+			}
+		}
+		release_sock(sk);
+	} else {
+		sock_poll_wait(file, sk_sleep(sk), wait);
+		if (sk->sk_state == SMC_LISTEN)
+			/* woken up by sk_data_ready in smc_listen_work() */
+			mask |= smc_accept_poll(sk);
+		if (sk->sk_err)
+			mask |= POLLERR;
+		if (atomic_read(&smc->conn.sndbuf_space) ||
+		    (sk->sk_shutdown & SEND_SHUTDOWN)) {
+			mask |= POLLOUT | POLLWRNORM;
+		} else {
+			sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
+			set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
+		}
+		if (atomic_read(&smc->conn.bytes_to_rcv))
+			mask |= POLLIN | POLLRDNORM;
+		if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
+		    (sk->sk_state == SMC_CLOSED))
+			mask |= POLLHUP;
+		if (sk->sk_shutdown & RCV_SHUTDOWN)
+			mask |= POLLIN | POLLRDNORM | POLLRDHUP;
+		if (sk->sk_state == SMC_APPCLOSEWAIT1)
+			mask |= POLLIN;
+
+	}
+
+	return mask;
+}
+
+static int smc_shutdown(struct socket *sock, int how)
+{
+	struct sock *sk = sock->sk;
+	struct smc_sock *smc;
+	int rc = -EINVAL;
+	int rc1 = 0;
+
+	smc = smc_sk(sk);
+
+	if ((how < SHUT_RD) || (how > SHUT_RDWR))
+		return rc;
+
+	lock_sock(sk);
+
+	rc = -ENOTCONN;
+	if ((sk->sk_state != SMC_LISTEN) &&
+	    (sk->sk_state != SMC_ACTIVE) &&
+	    (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
+	    (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
+	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
+	    (sk->sk_state != SMC_APPCLOSEWAIT2) &&
+	    (sk->sk_state != SMC_APPFINCLOSEWAIT))
+		goto out;
+	if (smc->use_fallback) {
+		rc = kernel_sock_shutdown(smc->clcsock, how);
+		sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
+		if (sk->sk_shutdown == SHUTDOWN_MASK)
+			sk->sk_state = SMC_CLOSED;
+		goto out;
+	}
+	switch (how) {
+	case SHUT_RDWR:		/* shutdown in both directions */
+		rc = smc_close_active(smc);
+		break;
+	case SHUT_WR:
+		rc = smc_close_shutdown_write(smc);
+		break;
+	case SHUT_RD:
+		if (sk->sk_state == SMC_LISTEN)
+			rc = smc_close_active(smc);
+		else
+			rc = 0;
+			/* nothing more to do because peer is not involved */
+		break;
+	}
+	rc1 = kernel_sock_shutdown(smc->clcsock, how);
+	/* map sock_shutdown_cmd constants to sk_shutdown value range */
+	sk->sk_shutdown |= how + 1;
+
+out:
+	release_sock(sk);
+	return rc ? rc : rc1;
+}
+
+static int smc_setsockopt(struct socket *sock, int level, int optname,
+			  char __user *optval, unsigned int optlen)
+{
+	struct sock *sk = sock->sk;
+	struct smc_sock *smc;
+
+	smc = smc_sk(sk);
+
+	/* generic setsockopts reaching us here always apply to the
+	 * CLC socket
+	 */
+	return smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
+					     optval, optlen);
+}
+
+static int smc_getsockopt(struct socket *sock, int level, int optname,
+			  char __user *optval, int __user *optlen)
+{
+	struct smc_sock *smc;
+
+	smc = smc_sk(sock->sk);
+	/* socket options apply to the CLC socket */
+	return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
+					     optval, optlen);
+}
+
+static int smc_ioctl(struct socket *sock, unsigned int cmd,
+		     unsigned long arg)
+{
+	struct smc_sock *smc;
+
+	smc = smc_sk(sock->sk);
+	if (smc->use_fallback)
+		return smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
+	else
+		return sock_no_ioctl(sock, cmd, arg);
+}
+
+static ssize_t smc_sendpage(struct socket *sock, struct page *page,
+			    int offset, size_t size, int flags)
+{
+	struct sock *sk = sock->sk;
+	struct smc_sock *smc;
+	int rc = -EPIPE;
+
+	smc = smc_sk(sk);
+	lock_sock(sk);
+	if (sk->sk_state != SMC_ACTIVE)
+		goto out;
+	if (smc->use_fallback)
+		rc = kernel_sendpage(smc->clcsock, page, offset,
+				     size, flags);
+	else
+		rc = sock_no_sendpage(sock, page, offset, size, flags);
+
+out:
+	release_sock(sk);
+	return rc;
+}
+
+static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
+			       struct pipe_inode_info *pipe, size_t len,
+				    unsigned int flags)
+{
+	struct sock *sk = sock->sk;
+	struct smc_sock *smc;
+	int rc = -ENOTCONN;
+
+	smc = smc_sk(sk);
+	lock_sock(sk);
+	if ((sk->sk_state != SMC_ACTIVE) && (sk->sk_state != SMC_CLOSED))
+		goto out;
+	if (smc->use_fallback) {
+		rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
+						    pipe, len, flags);
+	} else {
+		rc = -EOPNOTSUPP;
+	}
+out:
+	release_sock(sk);
+	return rc;
+}
+
+/* must look like tcp */
+static const struct proto_ops smc_sock_ops = {
+	.family		= PF_SMC,
+	.owner		= THIS_MODULE,
+	.release	= smc_release,
+	.bind		= smc_bind,
+	.connect	= smc_connect,
+	.socketpair	= sock_no_socketpair,
+	.accept		= smc_accept,
+	.getname	= smc_getname,
+	.poll		= smc_poll,
+	.ioctl		= smc_ioctl,
+	.listen		= smc_listen,
+	.shutdown	= smc_shutdown,
+	.setsockopt	= smc_setsockopt,
+	.getsockopt	= smc_getsockopt,
+	.sendmsg	= smc_sendmsg,
+	.recvmsg	= smc_recvmsg,
+	.mmap		= sock_no_mmap,
+	.sendpage	= smc_sendpage,
+	.splice_read	= smc_splice_read,
+};
+
+static int smc_create(struct net *net, struct socket *sock, int protocol,
+		      int kern)
+{
+	struct smc_sock *smc;
+	struct sock *sk;
+	int rc;
+
+	rc = -ESOCKTNOSUPPORT;
+	if (sock->type != SOCK_STREAM)
+		goto out;
+
+	rc = -EPROTONOSUPPORT;
+	if ((protocol != IPPROTO_IP) && (protocol != IPPROTO_TCP))
+		goto out;
+
+	rc = -ENOBUFS;
+	sock->ops = &smc_sock_ops;
+	sk = smc_sock_alloc(net, sock);
+	if (!sk)
+		goto out;
+
+	/* create internal TCP socket for CLC handshake and fallback */
+	smc = smc_sk(sk);
+	smc->use_fallback = false; /* assume rdma capability first */
+	rc = sock_create_kern(net, PF_INET, SOCK_STREAM,
+			      IPPROTO_TCP, &smc->clcsock);
+	if (rc)
+		sk_common_release(sk);
+	smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
+	smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
+
+out:
+	return rc;
+}
+
+static const struct net_proto_family smc_sock_family_ops = {
+	.family	= PF_SMC,
+	.owner	= THIS_MODULE,
+	.create	= smc_create,
+};
+
+static int __init smc_init(void)
+{
+	int rc;
+
+	rc = smc_pnet_init();
+	if (rc)
+		return rc;
+
+	rc = smc_llc_init();
+	if (rc) {
+		pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
+		goto out_pnet;
+	}
+
+	rc = smc_cdc_init();
+	if (rc) {
+		pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
+		goto out_pnet;
+	}
+
+	rc = proto_register(&smc_proto, 1);
+	if (rc) {
+		pr_err("%s: proto_register fails with %d\n", __func__, rc);
+		goto out_pnet;
+	}
+
+	rc = sock_register(&smc_sock_family_ops);
+	if (rc) {
+		pr_err("%s: sock_register fails with %d\n", __func__, rc);
+		goto out_proto;
+	}
+	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
+
+	rc = smc_ib_register_client();
+	if (rc) {
+		pr_err("%s: ib_register fails with %d\n", __func__, rc);
+		goto out_sock;
+	}
+
+	return 0;
+
+out_sock:
+	sock_unregister(PF_SMC);
+out_proto:
+	proto_unregister(&smc_proto);
+out_pnet:
+	smc_pnet_exit();
+	return rc;
+}
+
+static void __exit smc_exit(void)
+{
+	struct smc_link_group *lgr, *lg;
+	LIST_HEAD(lgr_freeing_list);
+
+	spin_lock_bh(&smc_lgr_list.lock);
+	if (!list_empty(&smc_lgr_list.list))
+		list_splice_init(&smc_lgr_list.list, &lgr_freeing_list);
+	spin_unlock_bh(&smc_lgr_list.lock);
+	list_for_each_entry_safe(lgr, lg, &lgr_freeing_list, list) {
+		list_del_init(&lgr->list);
+		smc_lgr_free(lgr); /* free link group */
+	}
+	smc_ib_unregister_client();
+	sock_unregister(PF_SMC);
+	proto_unregister(&smc_proto);
+	smc_pnet_exit();
+}
+
+module_init(smc_init);
+module_exit(smc_exit);
+
+MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
+MODULE_DESCRIPTION("smc socket address family");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS_NETPROTO(PF_SMC);

net/smc/smc.h

@@ -0,0 +1,274 @@
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Definitions for the SMC module (socket related)
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+#ifndef __SMC_H
+#define __SMC_H
+
+#include <linux/socket.h>
+#include <linux/types.h>
+#include <linux/compiler.h> /* __aligned */
+#include <net/sock.h>
+
+#include "smc_ib.h"
+
+#define SMCPROTO_SMC		0	/* SMC protocol */
+
+#define SMC_MAX_PORTS		2	/* Max # of ports */
+
+extern struct proto smc_proto;
+
+#ifdef ATOMIC64_INIT
+#define KERNEL_HAS_ATOMIC64
+#endif
+
+enum smc_state {		/* possible states of an SMC socket */
+	SMC_ACTIVE	= 1,
+	SMC_INIT	= 2,
+	SMC_CLOSED	= 7,
+	SMC_LISTEN	= 10,
+	/* normal close */
+	SMC_PEERCLOSEWAIT1	= 20,
+	SMC_PEERCLOSEWAIT2	= 21,
+	SMC_APPFINCLOSEWAIT	= 24,
+	SMC_APPCLOSEWAIT1	= 22,
+	SMC_APPCLOSEWAIT2	= 23,
+	SMC_PEERFINCLOSEWAIT	= 25,
+	/* abnormal close */
+	SMC_PEERABORTWAIT	= 26,
+	SMC_PROCESSABORT	= 27,
+};
+
+struct smc_link_group;
+
+struct smc_wr_rx_hdr {	/* common prefix part of LLC and CDC to demultiplex */
+	u8			type;
+} __aligned(1);
+
+struct smc_cdc_conn_state_flags {
+#if defined(__BIG_ENDIAN_BITFIELD)
+	u8	peer_done_writing : 1;	/* Sending done indicator */
+	u8	peer_conn_closed : 1;	/* Peer connection closed indicator */
+	u8	peer_conn_abort : 1;	/* Abnormal close indicator */
+	u8	reserved : 5;
+#elif defined(__LITTLE_ENDIAN_BITFIELD)
+	u8	reserved : 5;
+	u8	peer_conn_abort : 1;
+	u8	peer_conn_closed : 1;
+	u8	peer_done_writing : 1;
+#endif
+};
+
+struct smc_cdc_producer_flags {
+#if defined(__BIG_ENDIAN_BITFIELD)
+	u8	write_blocked : 1;	/* Writing Blocked, no rx buf space */
+	u8	urg_data_pending : 1;	/* Urgent Data Pending */
+	u8	urg_data_present : 1;	/* Urgent Data Present */
+	u8	cons_curs_upd_req : 1;	/* cursor update requested */
+	u8	failover_validation : 1;/* message replay due to failover */
+	u8	reserved : 3;
+#elif defined(__LITTLE_ENDIAN_BITFIELD)
+	u8	reserved : 3;
+	u8	failover_validation : 1;
+	u8	cons_curs_upd_req : 1;
+	u8	urg_data_present : 1;
+	u8	urg_data_pending : 1;
+	u8	write_blocked : 1;
+#endif
+};
+
+/* in host byte order */
+union smc_host_cursor {	/* SMC cursor - an offset in an RMBE */
+	struct {
+		u16	reserved;
+		u16	wrap;		/* window wrap sequence number */
+		u32	count;		/* cursor (= offset) part */
+	};
+#ifdef KERNEL_HAS_ATOMIC64
+	atomic64_t		acurs;	/* for atomic processing */
+#else
+	u64			acurs;	/* for atomic processing */
+#endif
+} __aligned(8);
+
+/* in host byte order, except for flag bitfields in network byte order */
+struct smc_host_cdc_msg {		/* Connection Data Control message */
+	struct smc_wr_rx_hdr		common; /* .type = 0xFE */
+	u8				len;	/* length = 44 */
+	u16				seqno;	/* connection seq # */
+	u32				token;	/* alert_token */
+	union smc_host_cursor		prod;		/* producer cursor */
+	union smc_host_cursor		cons;		/* consumer cursor,
+							 * piggy backed "ack"
+							 */
+	struct smc_cdc_producer_flags	prod_flags;	/* conn. tx/rx status */
+	struct smc_cdc_conn_state_flags	conn_state_flags; /* peer conn. status*/
+	u8				reserved[18];
+} __aligned(8);
+
+struct smc_connection {
+	struct rb_node		alert_node;
+	struct smc_link_group	*lgr;		/* link group of connection */
+	u32			alert_token_local; /* unique conn. id */
+	u8			peer_conn_idx;	/* from tcp handshake */
+	int			peer_rmbe_size;	/* size of peer rx buffer */
+	atomic_t		peer_rmbe_space;/* remaining free bytes in peer
+						 * rmbe
+						 */
+	int			rtoken_idx;	/* idx to peer RMB rkey/addr */
+
+	struct smc_buf_desc	*sndbuf_desc;	/* send buffer descriptor */
+	int			sndbuf_size;	/* sndbuf size <== sock wmem */
+	struct smc_buf_desc	*rmb_desc;	/* RMBE descriptor */
+	int			rmbe_size;	/* RMBE size <== sock rmem */
+	int			rmbe_size_short;/* compressed notation */
+	int			rmbe_update_limit;
+						/* lower limit for consumer
+						 * cursor update
+						 */
+
+	struct smc_host_cdc_msg	local_tx_ctrl;	/* host byte order staging
+						 * buffer for CDC msg send
+						 * .prod cf. TCP snd_nxt
+						 * .cons cf. TCP sends ack
+						 */
+	union smc_host_cursor	tx_curs_prep;	/* tx - prepared data
+						 * snd_max..wmem_alloc
+						 */
+	union smc_host_cursor	tx_curs_sent;	/* tx - sent data
+						 * snd_nxt ?
+						 */
+	union smc_host_cursor	tx_curs_fin;	/* tx - confirmed by peer
+						 * snd-wnd-begin ?
+						 */
+	atomic_t		sndbuf_space;	/* remaining space in sndbuf */
+	u16			tx_cdc_seq;	/* sequence # for CDC send */
+	spinlock_t		send_lock;	/* protect wr_sends */
+	struct work_struct	tx_work;	/* retry of smc_cdc_msg_send */
+
+	struct smc_host_cdc_msg	local_rx_ctrl;	/* filled during event_handl.
+						 * .prod cf. TCP rcv_nxt
+						 * .cons cf. TCP snd_una
+						 */
+	union smc_host_cursor	rx_curs_confirmed; /* confirmed to peer
+						    * source of snd_una ?
+						    */
+	atomic_t		bytes_to_rcv;	/* arrived data,
+						 * not yet received
+						 */
+#ifndef KERNEL_HAS_ATOMIC64
+	spinlock_t		acurs_lock;	/* protect cursors */
+#endif
+};
+
+struct smc_sock {				/* smc sock container */
+	struct sock		sk;
+	struct socket		*clcsock;	/* internal tcp socket */
+	struct smc_connection	conn;		/* smc connection */
+	struct sockaddr		*addr;		/* inet connect address */
+	struct smc_sock		*listen_smc;	/* listen parent */
+	struct work_struct	tcp_listen_work;/* handle tcp socket accepts */
+	struct work_struct	smc_listen_work;/* prepare new accept socket */
+	struct list_head	accept_q;	/* sockets to be accepted */
+	spinlock_t		accept_q_lock;	/* protects accept_q */
+	struct delayed_work	sock_put_work;	/* final socket freeing */
+	bool			use_fallback;	/* fallback to tcp */
+	u8			wait_close_tx_prepared : 1;
+						/* shutdown wr or close
+						 * started, waiting for unsent
+						 * data to be sent
+						 */
+};
+
+static inline struct smc_sock *smc_sk(const struct sock *sk)
+{
+	return (struct smc_sock *)sk;
+}
+
+#define SMC_SYSTEMID_LEN		8
+
+extern u8	local_systemid[SMC_SYSTEMID_LEN]; /* unique system identifier */
+
+/* convert an u32 value into network byte order, store it into a 3 byte field */
+static inline void hton24(u8 *net, u32 host)
+{
+	__be32 t;
+
+	t = cpu_to_be32(host);
+	memcpy(net, ((u8 *)&t) + 1, 3);
+}
+
+/* convert a received 3 byte field into host byte order*/
+static inline u32 ntoh24(u8 *net)
+{
+	__be32 t = 0;
+
+	memcpy(((u8 *)&t) + 1, net, 3);
+	return be32_to_cpu(t);
+}
+
+#define SMC_BUF_MIN_SIZE 16384		/* minimum size of an RMB */
+
+#define SMC_RMBE_SIZES	16	/* number of distinct sizes for an RMBE */
+/* theoretically, the RFC states that largest size would be 512K,
+ * i.e. compressed 5 and thus 6 sizes (0..5), despite
+ * struct smc_clc_msg_accept_confirm.rmbe_size being a 4 bit value (0..15)
+ */
+
+/* convert the RMB size into the compressed notation - minimum 16K.
+ * In contrast to plain ilog2, this rounds towards the next power of 2,
+ * so the socket application gets at least its desired sndbuf / rcvbuf size.
+ */
+static inline u8 smc_compress_bufsize(int size)
+{
+	u8 compressed;
+
+	if (size <= SMC_BUF_MIN_SIZE)
+		return 0;
+
+	size = (size - 1) >> 14;
+	compressed = ilog2(size) + 1;
+	if (compressed >= SMC_RMBE_SIZES)
+		compressed = SMC_RMBE_SIZES - 1;
+	return compressed;
+}
+
+/* convert the RMB size from compressed notation into integer */
+static inline int smc_uncompress_bufsize(u8 compressed)
+{
+	u32 size;
+
+	size = 0x00000001 << (((int)compressed) + 14);
+	return (int)size;
+}
+
+#ifdef CONFIG_XFRM
+static inline bool using_ipsec(struct smc_sock *smc)
+{
+	return (smc->clcsock->sk->sk_policy[0] ||
+		smc->clcsock->sk->sk_policy[1]) ? 1 : 0;
+}
+#else
+static inline bool using_ipsec(struct smc_sock *smc)
+{
+	return 0;
+}
+#endif
+
+struct smc_clc_msg_local;
+
+int smc_netinfo_by_tcpsk(struct socket *clcsock, __be32 *subnet,
+			 u8 *prefix_len);
+void smc_conn_free(struct smc_connection *conn);
+int smc_conn_create(struct smc_sock *smc, __be32 peer_in_addr,
+		    struct smc_ib_device *smcibdev, u8 ibport,
+		    struct smc_clc_msg_local *lcl, int srv_first_contact);
+struct sock *smc_accept_dequeue(struct sock *parent, struct socket *new_sock);
+void smc_close_non_accepted(struct sock *sk);
+
+#endif	/* __SMC_H */

net/smc/smc_cdc.c

@@ -0,0 +1,304 @@
+/*
+ * Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ * Connection Data Control (CDC)
+ * handles flow control
+ *
+ * Copyright IBM Corp. 2016
+ *
+ * Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#include <linux/spinlock.h>
+
+#include "smc.h"
+#include "smc_wr.h"
+#include "smc_cdc.h"
+#include "smc_tx.h"
+#include "smc_rx.h"
+#include "smc_close.h"
+
+/********************************** send *************************************/
+
+struct smc_cdc_tx_pend {
+	struct smc_connection	*conn;		/* socket connection */
+	union smc_host_cursor	cursor;	/* tx sndbuf cursor sent */
+	union smc_host_cursor	p_cursor;	/* rx RMBE cursor produced */
+	u16			ctrl_seq;	/* conn. tx sequence # */
+};
+
+/* handler for send/transmission completion of a CDC msg */
+static void smc_cdc_tx_handler(struct smc_wr_tx_pend_priv *pnd_snd,
+			       struct smc_link *link,
+			       enum ib_wc_status wc_status)
+{
+	struct smc_cdc_tx_pend *cdcpend = (struct smc_cdc_tx_pend *)pnd_snd;
+	struct smc_sock *smc;
+	int diff;
+
+	if (!cdcpend->conn)
+		/* already dismissed */
+		return;
+
+	smc = container_of(cdcpend->conn, struct smc_sock, conn);
+	bh_lock_sock(&smc->sk);
+	if (!wc_status) {
+		diff = smc_curs_diff(cdcpend->conn->sndbuf_size,
+				     &cdcpend->conn->tx_curs_fin,
+				     &cdcpend->cursor);
+		/* sndbuf_space is decreased in smc_sendmsg */
+		smp_mb__before_atomic();
+		atomic_add(diff, &cdcpend->conn->sndbuf_space);
+		/* guarantee 0 <= sndbuf_space <= sndbuf_size */
+		smp_mb__after_atomic();
+		smc_curs_write(&cdcpend->conn->tx_curs_fin,
+			       smc_curs_read(&cdcpend->cursor, cdcpend->conn),
+			       cdcpend->conn);
+	}
+	smc_tx_sndbuf_nonfull(smc);
+	if (smc->sk.sk_state != SMC_ACTIVE)
+		/* wake up smc_close_wait_tx_pends() */
+		smc->sk.sk_state_change(&smc->sk);
+	bh_unlock_sock(&smc->sk);
+}
+
+int smc_cdc_get_free_slot(struct smc_link *link,
+			  struct smc_wr_buf **wr_buf,
+			  struct smc_cdc_tx_pend **pend)
+{
+	return smc_wr_tx_get_free_slot(link, smc_cdc_tx_handler, wr_buf,
+				       (struct smc_wr_tx_pend_priv **)pend);
+}
+
+static inline void smc_cdc_add_pending_send(struct smc_connection *conn,
+					    struct smc_cdc_tx_pend *pend)
+{
+	BUILD_BUG_ON_MSG(
+		sizeof(struct smc_cdc_msg) > SMC_WR_BUF_SIZE,
+		"must increase SMC_WR_BUF_SIZE to at least sizeof(struct smc_cdc_msg)");
+	BUILD_BUG_ON_MSG(
+		offsetof(struct smc_cdc_msg, reserved) > SMC_WR_TX_SIZE,
+		"must adapt SMC_WR_TX_SIZE to sizeof(struct smc_cdc_msg); if not all smc_wr upper layer protocols use the same message size any more, must start to set link->wr_tx_sges[i].length on each individual smc_wr_tx_send()");
+	BUILD_BUG_ON_MSG(
+		sizeof(struct smc_cdc_tx_pend) > SMC_WR_TX_PEND_PRIV_SIZE,
+		"must increase SMC_WR_TX_PEND_PRIV_SIZE to at least sizeof(struct smc_cdc_tx_pend)");
+	pend->conn = conn;
+	pend->cursor = conn->tx_curs_sent;
+	pend->p_cursor = conn->local_tx_ctrl.prod;
+	pend->ctrl_seq = conn->tx_cdc_seq;
+}
+
+int smc_cdc_msg_send(struct smc_connection *conn,
+		     struct smc_wr_buf *wr_buf,
+		     struct smc_cdc_tx_pend *pend)
+{
+	struct smc_link *link;
+	int rc;
+
+	link = &conn->lgr->lnk[SMC_SINGLE_LINK];
+
+	smc_cdc_add_pending_send(conn, pend);
+
+	conn->tx_cdc_seq++;
+	conn->local_tx_ctrl.seqno = conn->tx_cdc_seq;
+	smc_host_msg_to_cdc((struct smc_cdc_msg *)wr_buf,
+			    &conn->local_tx_ctrl, conn);
+	rc = smc_wr_tx_send(link, (struct smc_wr_tx_pend_priv *)pend);
+	if (!rc)
+		smc_curs_write(&conn->rx_curs_confirmed,
+			       smc_curs_read(&conn->local_tx_ctrl.cons, conn),
+			       conn);
+
+	return rc;
+}
+
+int smc_cdc_get_slot_and_msg_send(struct smc_connection *conn)
+{
+	struct smc_cdc_tx_pend *pend;
+	struct smc_wr_buf *wr_buf;
+	int rc;
+
+	rc = smc_cdc_get_free_slot(&conn->lgr->lnk[SMC_SINGLE_LINK], &wr_buf,
+				   &pend);
+	if (rc)
+		return rc;
+
+	return smc_cdc_msg_send(conn, wr_buf, pend);
+}
+
+static bool smc_cdc_tx_filter(struct smc_wr_tx_pend_priv *tx_pend,
+			      unsigned long data)
+{
+	struct smc_connection *conn = (struct smc_connection *)data;
+	struct smc_cdc_tx_pend *cdc_pend =
+		(struct smc_cdc_tx_pend *)tx_pend;
+
+	return cdc_pend->conn == conn;
+}
+
+static void smc_cdc_tx_dismisser(struct smc_wr_tx_pend_priv *tx_pend)
+{
+	struct smc_cdc_tx_pend *cdc_pend =
+		(struct smc_cdc_tx_pend *)tx_pend;
+
+	cdc_pend->conn = NULL;
+}
+
+void smc_cdc_tx_dismiss_slots(struct smc_connection *conn)
+{
+	struct smc_link *link = &conn->lgr->lnk[SMC_SINGLE_LINK];
+
+	smc_wr_tx_dismiss_slots(link, SMC_CDC_MSG_TYPE,
+				smc_cdc_tx_filter, smc_cdc_tx_dismisser,
+				(unsigned long)conn);
+}
+
+bool smc_cdc_tx_has_pending(struct smc_connection *conn)
+{
+	struct smc_link *link = &conn->lgr->lnk[SMC_SINGLE_LINK];
+
+	return smc_wr_tx_has_pending(link, SMC_CDC_MSG_TYPE,
+				     smc_cdc_tx_filter, (unsigned long)conn);
+}
+
+/********************************* receive ***********************************/
+
+static inline bool smc_cdc_before(u16 seq1, u16 seq2)
+{
+	return (s16)(seq1 - seq2) < 0;
+}
+
+static void smc_cdc_msg_recv_action(struct smc_sock *smc,
+				    struct smc_link *link,
+				    struct smc_cdc_msg *cdc)
+{
+	union smc_host_cursor cons_old, prod_old;
+	struct smc_connection *conn = &smc->conn;
+	int diff_cons, diff_prod;
+
+	if (!cdc->prod_flags.failover_validation) {
+		if (smc_cdc_before(ntohs(cdc->seqno),
+				   conn->local_rx_ctrl.seqno))
+			/* received seqno is old */
+			return;
+	}
+	smc_curs_write(&prod_old,
+		       smc_curs_read(&conn->local_rx_ctrl.prod, conn),
+		       conn);
+	smc_curs_write(&cons_old,
+		       smc_curs_read(&conn->local_rx_ctrl.cons, conn),
+		       conn);
+	smc_cdc_msg_to_host(&conn->local_rx_ctrl, cdc, conn);
+
+	diff_cons = smc_curs_diff(conn->peer_rmbe_size, &cons_old,
+				  &conn->local_rx_ctrl.cons);
+	if (diff_cons) {
+		/* peer_rmbe_space is decreased during data transfer with RDMA
+		 * write
+		 */
+		smp_mb__before_atomic();
+		atomic_add(diff_cons, &conn->peer_rmbe_space);
+		/* guarantee 0 <= peer_rmbe_space <= peer_rmbe_size */
+		smp_mb__after_atomic();
+	}
+
+	diff_prod = smc_curs_diff(conn->rmbe_size, &prod_old,
+				  &conn->local_rx_ctrl.prod);
+	if (diff_prod) {
+		/* bytes_to_rcv is decreased in smc_recvmsg */
+		smp_mb__before_atomic();
+		atomic_add(diff_prod, &conn->bytes_to_rcv);
+		/* guarantee 0 <= bytes_to_rcv <= rmbe_size */
+		smp_mb__after_atomic();
+		smc->sk.sk_data_ready(&smc->sk);
+	}
+
+	if (conn->local_rx_ctrl.conn_state_flags.peer_conn_abort) {
+		smc->sk.sk_err = ECONNRESET;
+		conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
+	}
+	if (smc_cdc_rxed_any_close_or_senddone(conn))
+		smc_close_passive_received(smc);
+
+	/* piggy backed tx info */
+	/* trigger sndbuf consumer: RDMA write into peer RMBE and CDC */
+	if (diff_cons && smc_tx_prepared_sends(conn)) {
+		smc_tx_sndbuf_nonempty(conn);
+		/* trigger socket release if connection closed */
+		smc_close_wake_tx_prepared(smc);
+	}
+
+	/* subsequent patch: trigger socket release if connection closed */
+
+	/* socket connected but not accepted */
+	if (!smc->sk.sk_socket)
+		return;
+
+	/* data available */
+	if ((conn->local_rx_ctrl.prod_flags.write_blocked) ||
+	    (conn->local_rx_ctrl.prod_flags.cons_curs_upd_req))
+		smc_tx_consumer_update(conn);
+}
+
+/* called under tasklet context */
+static inline void smc_cdc_msg_recv(struct smc_cdc_msg *cdc,
+				    struct smc_link *link, u64 wr_id)
+{
+	struct smc_link_group *lgr = container_of(link, struct smc_link_group,
+						  lnk[SMC_SINGLE_LINK]);
+	struct smc_connection *connection;
+	struct smc_sock *smc;
+
+	/* lookup connection */
+	read_lock_bh(&lgr->conns_lock);
+	connection = smc_lgr_find_conn(ntohl(cdc->token), lgr);
+	if (!connection) {
+		read_unlock_bh(&lgr->conns_lock);
+		return;
+	}
+	smc = container_of(connection, struct smc_sock, conn);
+	sock_hold(&smc->sk);
+	read_unlock_bh(&lgr->conns_lock);
+	bh_lock_sock(&smc->sk);
+	smc_cdc_msg_recv_action(smc, link, cdc);
+	bh_unlock_sock(&smc->sk);
+	sock_put(&smc->sk); /* no free sk in softirq-context */
+}
+
+/***************************** init, exit, misc ******************************/
+
+static void smc_cdc_rx_handler(struct ib_wc *wc, void *buf)
+{
+	struct smc_link *link = (struct smc_link *)wc->qp->qp_context;
+	struct smc_cdc_msg *cdc = buf;
+
+	if (wc->byte_len < offsetof(struct smc_cdc_msg, reserved))
+		return; /* short message */
+	if (cdc->len != sizeof(*cdc))
+		return; /* invalid message */
+	smc_cdc_msg_recv(cdc, link, wc->wr_id);
+}
+
+static struct smc_wr_rx_handler smc_cdc_rx_handlers[] = {
+	{
+		.handler	= smc_cdc_rx_handler,
+		.type		= SMC_CDC_MSG_TYPE
+	},
+	{
+		.handler	= NULL,
+	}
+};
+
+int __init smc_cdc_init(void)
+{
+	struct smc_wr_rx_handler *handler;
+	int rc = 0;
+
+	for (handler = smc_cdc_rx_handlers; handler->handler; handler++) {
+		INIT_HLIST_NODE(&handler->list);
+		rc = smc_wr_rx_register_handler(handler);
+		if (rc)
+			break;
+	}
+	return rc;
+}

net/smc/smc_cdc.h

@@ -0,0 +1,218 @@
+/*
+ * Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ * Connection Data Control (CDC)
+ *
+ * Copyright IBM Corp. 2016
+ *
+ * Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#ifndef SMC_CDC_H
+#define SMC_CDC_H
+
+#include <linux/kernel.h> /* max_t */
+#include <linux/atomic.h>
+#include <linux/in.h>
+#include <linux/compiler.h>
+
+#include "smc.h"
+#include "smc_core.h"
+#include "smc_wr.h"
+
+#define	SMC_CDC_MSG_TYPE		0xFE
+
+/* in network byte order */
+union smc_cdc_cursor {		/* SMC cursor */
+	struct {
+		__be16	reserved;
+		__be16	wrap;
+		__be32	count;
+	};
+#ifdef KERNEL_HAS_ATOMIC64
+	atomic64_t	acurs;		/* for atomic processing */
+#else
+	u64		acurs;		/* for atomic processing */
+#endif
+} __aligned(8);
+
+/* in network byte order */
+struct smc_cdc_msg {
+	struct smc_wr_rx_hdr		common; /* .type = 0xFE */
+	u8				len;	/* 44 */
+	__be16				seqno;
+	__be32				token;
+	union smc_cdc_cursor		prod;
+	union smc_cdc_cursor		cons;	/* piggy backed "ack" */
+	struct smc_cdc_producer_flags	prod_flags;
+	struct smc_cdc_conn_state_flags	conn_state_flags;
+	u8				reserved[18];
+} __aligned(8);
+
+static inline bool smc_cdc_rxed_any_close(struct smc_connection *conn)
+{
+	return conn->local_rx_ctrl.conn_state_flags.peer_conn_abort ||
+	       conn->local_rx_ctrl.conn_state_flags.peer_conn_closed;
+}
+
+static inline bool smc_cdc_rxed_any_close_or_senddone(
+	struct smc_connection *conn)
+{
+	return smc_cdc_rxed_any_close(conn) ||
+	       conn->local_rx_ctrl.conn_state_flags.peer_done_writing;
+}
+
+static inline void smc_curs_add(int size, union smc_host_cursor *curs,
+				int value)
+{
+	curs->count += value;
+	if (curs->count >= size) {
+		curs->wrap++;
+		curs->count -= size;
+	}
+}
+
+/* SMC cursors are 8 bytes long and require atomic reading and writing */
+static inline u64 smc_curs_read(union smc_host_cursor *curs,
+				struct smc_connection *conn)
+{
+#ifndef KERNEL_HAS_ATOMIC64
+	unsigned long flags;
+	u64 ret;
+
+	spin_lock_irqsave(&conn->acurs_lock, flags);
+	ret = curs->acurs;
+	spin_unlock_irqrestore(&conn->acurs_lock, flags);
+	return ret;
+#else
+	return atomic64_read(&curs->acurs);
+#endif
+}
+
+static inline u64 smc_curs_read_net(union smc_cdc_cursor *curs,
+				    struct smc_connection *conn)
+{
+#ifndef KERNEL_HAS_ATOMIC64
+	unsigned long flags;
+	u64 ret;
+
+	spin_lock_irqsave(&conn->acurs_lock, flags);
+	ret = curs->acurs;
+	spin_unlock_irqrestore(&conn->acurs_lock, flags);
+	return ret;
+#else
+	return atomic64_read(&curs->acurs);
+#endif
+}
+
+static inline void smc_curs_write(union smc_host_cursor *curs, u64 val,
+				  struct smc_connection *conn)
+{
+#ifndef KERNEL_HAS_ATOMIC64
+	unsigned long flags;
+
+	spin_lock_irqsave(&conn->acurs_lock, flags);
+	curs->acurs = val;
+	spin_unlock_irqrestore(&conn->acurs_lock, flags);
+#else
+	atomic64_set(&curs->acurs, val);
+#endif
+}
+
+static inline void smc_curs_write_net(union smc_cdc_cursor *curs, u64 val,
+				      struct smc_connection *conn)
+{
+#ifndef KERNEL_HAS_ATOMIC64
+	unsigned long flags;
+
+	spin_lock_irqsave(&conn->acurs_lock, flags);
+	curs->acurs = val;
+	spin_unlock_irqrestore(&conn->acurs_lock, flags);
+#else
+	atomic64_set(&curs->acurs, val);
+#endif
+}
+
+/* calculate cursor difference between old and new, where old <= new */
+static inline int smc_curs_diff(unsigned int size,
+				union smc_host_cursor *old,
+				union smc_host_cursor *new)
+{
+	if (old->wrap != new->wrap)
+		return max_t(int, 0,
+			     ((size - old->count) + new->count));
+
+	return max_t(int, 0, (new->count - old->count));
+}
+
+static inline void smc_host_cursor_to_cdc(union smc_cdc_cursor *peer,
+					  union smc_host_cursor *local,
+					  struct smc_connection *conn)
+{
+	union smc_host_cursor temp;
+
+	smc_curs_write(&temp, smc_curs_read(local, conn), conn);
+	peer->count = htonl(temp.count);
+	peer->wrap = htons(temp.wrap);
+	/* peer->reserved = htons(0); must be ensured by caller */
+}
+
+static inline void smc_host_msg_to_cdc(struct smc_cdc_msg *peer,
+				       struct smc_host_cdc_msg *local,
+				       struct smc_connection *conn)
+{
+	peer->common.type = local->common.type;
+	peer->len = local->len;
+	peer->seqno = htons(local->seqno);
+	peer->token = htonl(local->token);
+	smc_host_cursor_to_cdc(&peer->prod, &local->prod, conn);
+	smc_host_cursor_to_cdc(&peer->cons, &local->cons, conn);
+	peer->prod_flags = local->prod_flags;
+	peer->conn_state_flags = local->conn_state_flags;
+}
+
+static inline void smc_cdc_cursor_to_host(union smc_host_cursor *local,
+					  union smc_cdc_cursor *peer,
+					  struct smc_connection *conn)
+{
+	union smc_host_cursor temp, old;
+	union smc_cdc_cursor net;
+
+	smc_curs_write(&old, smc_curs_read(local, conn), conn);
+	smc_curs_write_net(&net, smc_curs_read_net(peer, conn), conn);
+	temp.count = ntohl(net.count);
+	temp.wrap = ntohs(net.wrap);
+	if ((old.wrap > temp.wrap) && temp.wrap)
+		return;
+	if ((old.wrap == temp.wrap) &&
+	    (old.count > temp.count))
+		return;
+	smc_curs_write(local, smc_curs_read(&temp, conn), conn);
+}
+
+static inline void smc_cdc_msg_to_host(struct smc_host_cdc_msg *local,
+				       struct smc_cdc_msg *peer,
+				       struct smc_connection *conn)
+{
+	local->common.type = peer->common.type;
+	local->len = peer->len;
+	local->seqno = ntohs(peer->seqno);
+	local->token = ntohl(peer->token);
+	smc_cdc_cursor_to_host(&local->prod, &peer->prod, conn);
+	smc_cdc_cursor_to_host(&local->cons, &peer->cons, conn);
+	local->prod_flags = peer->prod_flags;
+	local->conn_state_flags = peer->conn_state_flags;
+}
+
+struct smc_cdc_tx_pend;
+
+int smc_cdc_get_free_slot(struct smc_link *link, struct smc_wr_buf **wr_buf,
+			  struct smc_cdc_tx_pend **pend);
+void smc_cdc_tx_dismiss_slots(struct smc_connection *conn);
+int smc_cdc_msg_send(struct smc_connection *conn, struct smc_wr_buf *wr_buf,
+		     struct smc_cdc_tx_pend *pend);
+int smc_cdc_get_slot_and_msg_send(struct smc_connection *conn);
+bool smc_cdc_tx_has_pending(struct smc_connection *conn);
+int smc_cdc_init(void) __init;
+
+#endif /* SMC_CDC_H */

net/smc/smc_clc.c

@@ -0,0 +1,282 @@
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  CLC (connection layer control) handshake over initial TCP socket to
+ *  prepare for RDMA traffic
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#include <linux/in.h>
+#include <net/sock.h>
+#include <net/tcp.h>
+
+#include "smc.h"
+#include "smc_core.h"
+#include "smc_clc.h"
+#include "smc_ib.h"
+
+/* Wait for data on the tcp-socket, analyze received data
+ * Returns:
+ * 0 if success and it was not a decline that we received.
+ * SMC_CLC_DECL_REPLY if decline received for fallback w/o another decl send.
+ * clcsock error, -EINTR, -ECONNRESET, -EPROTO otherwise.
+ */
+int smc_clc_wait_msg(struct smc_sock *smc, void *buf, int buflen,
+		     u8 expected_type)
+{
+	struct sock *clc_sk = smc->clcsock->sk;
+	struct smc_clc_msg_hdr *clcm = buf;
+	struct msghdr msg = {NULL, 0};
+	int reason_code = 0;
+	struct kvec vec;
+	int len, datlen;
+	int krflags;
+
+	/* peek the first few bytes to determine length of data to receive
+	 * so we don't consume any subsequent CLC message or payload data
+	 * in the TCP byte stream
+	 */
+	vec.iov_base = buf;
+	vec.iov_len = buflen;
+	krflags = MSG_PEEK | MSG_WAITALL;
+	smc->clcsock->sk->sk_rcvtimeo = CLC_WAIT_TIME;
+	len = kernel_recvmsg(smc->clcsock, &msg, &vec, 1,
+			     sizeof(struct smc_clc_msg_hdr), krflags);
+	if (signal_pending(current)) {
+		reason_code = -EINTR;
+		clc_sk->sk_err = EINTR;
+		smc->sk.sk_err = EINTR;
+		goto out;
+	}
+	if (clc_sk->sk_err) {
+		reason_code = -clc_sk->sk_err;
+		smc->sk.sk_err = clc_sk->sk_err;
+		goto out;
+	}
+	if (!len) { /* peer has performed orderly shutdown */
+		smc->sk.sk_err = ECONNRESET;
+		reason_code = -ECONNRESET;
+		goto out;
+	}
+	if (len < 0) {
+		smc->sk.sk_err = -len;
+		reason_code = len;
+		goto out;
+	}
+	datlen = ntohs(clcm->length);
+	if ((len < sizeof(struct smc_clc_msg_hdr)) ||
+	    (datlen < sizeof(struct smc_clc_msg_decline)) ||
+	    (datlen > sizeof(struct smc_clc_msg_accept_confirm)) ||
+	    memcmp(clcm->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER)) ||
+	    ((clcm->type != SMC_CLC_DECLINE) &&
+	     (clcm->type != expected_type))) {
+		smc->sk.sk_err = EPROTO;
+		reason_code = -EPROTO;
+		goto out;
+	}
+
+	/* receive the complete CLC message */
+	vec.iov_base = buf;
+	vec.iov_len = buflen;
+	memset(&msg, 0, sizeof(struct msghdr));
+	krflags = MSG_WAITALL;
+	smc->clcsock->sk->sk_rcvtimeo = CLC_WAIT_TIME;
+	len = kernel_recvmsg(smc->clcsock, &msg, &vec, 1, datlen, krflags);
+	if (len < datlen) {
+		smc->sk.sk_err = EPROTO;
+		reason_code = -EPROTO;
+		goto out;
+	}
+	if (clcm->type == SMC_CLC_DECLINE) {
+		reason_code = SMC_CLC_DECL_REPLY;
+		if (ntohl(((struct smc_clc_msg_decline *)buf)->peer_diagnosis)
+			== SMC_CLC_DECL_SYNCERR)
+			smc->conn.lgr->sync_err = true;
+	}
+
+out:
+	return reason_code;
+}
+
+/* send CLC DECLINE message across internal TCP socket */
+int smc_clc_send_decline(struct smc_sock *smc, u32 peer_diag_info,
+			 u8 out_of_sync)
+{
+	struct smc_clc_msg_decline dclc;
+	struct msghdr msg;
+	struct kvec vec;
+	int len;
+
+	memset(&dclc, 0, sizeof(dclc));
+	memcpy(dclc.hdr.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
+	dclc.hdr.type = SMC_CLC_DECLINE;
+	dclc.hdr.length = htons(sizeof(struct smc_clc_msg_decline));
+	dclc.hdr.version = SMC_CLC_V1;
+	dclc.hdr.flag = out_of_sync ? 1 : 0;
+	memcpy(dclc.id_for_peer, local_systemid, sizeof(local_systemid));
+	dclc.peer_diagnosis = htonl(peer_diag_info);
+	memcpy(dclc.trl.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
+
+	memset(&msg, 0, sizeof(msg));
+	vec.iov_base = &dclc;
+	vec.iov_len = sizeof(struct smc_clc_msg_decline);
+	len = kernel_sendmsg(smc->clcsock, &msg, &vec, 1,
+			     sizeof(struct smc_clc_msg_decline));
+	if (len < sizeof(struct smc_clc_msg_decline))
+		smc->sk.sk_err = EPROTO;
+	if (len < 0)
+		smc->sk.sk_err = -len;
+	return len;
+}
+
+/* send CLC PROPOSAL message across internal TCP socket */
+int smc_clc_send_proposal(struct smc_sock *smc,
+			  struct smc_ib_device *smcibdev,
+			  u8 ibport)
+{
+	struct smc_clc_msg_proposal pclc;
+	int reason_code = 0;
+	struct msghdr msg;
+	struct kvec vec;
+	int len, rc;
+
+	/* send SMC Proposal CLC message */
+	memset(&pclc, 0, sizeof(pclc));
+	memcpy(pclc.hdr.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
+	pclc.hdr.type = SMC_CLC_PROPOSAL;
+	pclc.hdr.length = htons(sizeof(pclc));
+	pclc.hdr.version = SMC_CLC_V1;		/* SMC version */
+	memcpy(pclc.lcl.id_for_peer, local_systemid, sizeof(local_systemid));
+	memcpy(&pclc.lcl.gid, &smcibdev->gid[ibport - 1], SMC_GID_SIZE);
+	memcpy(&pclc.lcl.mac, &smcibdev->mac[ibport - 1],
+	       sizeof(smcibdev->mac[ibport - 1]));
+
+	/* determine subnet and mask from internal TCP socket */
+	rc = smc_netinfo_by_tcpsk(smc->clcsock, &pclc.outgoing_subnet,
+				  &pclc.prefix_len);
+	if (rc)
+		return SMC_CLC_DECL_CNFERR; /* configuration error */
+	memcpy(pclc.trl.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
+	memset(&msg, 0, sizeof(msg));
+	vec.iov_base = &pclc;
+	vec.iov_len = sizeof(pclc);
+	/* due to the few bytes needed for clc-handshake this cannot block */
+	len = kernel_sendmsg(smc->clcsock, &msg, &vec, 1, sizeof(pclc));
+	if (len < sizeof(pclc)) {
+		if (len >= 0) {
+			reason_code = -ENETUNREACH;
+			smc->sk.sk_err = -reason_code;
+		} else {
+			smc->sk.sk_err = smc->clcsock->sk->sk_err;
+			reason_code = -smc->sk.sk_err;
+		}
+	}
+
+	return reason_code;
+}
+
+/* send CLC CONFIRM message across internal TCP socket */
+int smc_clc_send_confirm(struct smc_sock *smc)
+{
+	struct smc_connection *conn = &smc->conn;
+	struct smc_clc_msg_accept_confirm cclc;
+	struct smc_link *link;
+	int reason_code = 0;
+	struct msghdr msg;
+	struct kvec vec;
+	int len;
+
+	link = &conn->lgr->lnk[SMC_SINGLE_LINK];
+	/* send SMC Confirm CLC msg */
+	memset(&cclc, 0, sizeof(cclc));
+	memcpy(cclc.hdr.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
+	cclc.hdr.type = SMC_CLC_CONFIRM;
+	cclc.hdr.length = htons(sizeof(cclc));
+	cclc.hdr.version = SMC_CLC_V1;		/* SMC version */
+	memcpy(cclc.lcl.id_for_peer, local_systemid, sizeof(local_systemid));
+	memcpy(&cclc.lcl.gid, &link->smcibdev->gid[link->ibport - 1],
+	       SMC_GID_SIZE);
+	memcpy(&cclc.lcl.mac, &link->smcibdev->mac[link->ibport - 1],
+	       sizeof(link->smcibdev->mac));
+	hton24(cclc.qpn, link->roce_qp->qp_num);
+	cclc.rmb_rkey =
+		htonl(conn->rmb_desc->mr_rx[SMC_SINGLE_LINK]->rkey);
+	cclc.conn_idx = 1; /* for now: 1 RMB = 1 RMBE */
+	cclc.rmbe_alert_token = htonl(conn->alert_token_local);
+	cclc.qp_mtu = min(link->path_mtu, link->peer_mtu);
+	cclc.rmbe_size = conn->rmbe_size_short;
+	cclc.rmb_dma_addr =
+		cpu_to_be64((u64)conn->rmb_desc->dma_addr[SMC_SINGLE_LINK]);
+	hton24(cclc.psn, link->psn_initial);
+
+	memcpy(cclc.trl.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
+
+	memset(&msg, 0, sizeof(msg));
+	vec.iov_base = &cclc;
+	vec.iov_len = sizeof(cclc);
+	len = kernel_sendmsg(smc->clcsock, &msg, &vec, 1, sizeof(cclc));
+	if (len < sizeof(cclc)) {
+		if (len >= 0) {
+			reason_code = -ENETUNREACH;
+			smc->sk.sk_err = -reason_code;
+		} else {
+			smc->sk.sk_err = smc->clcsock->sk->sk_err;
+			reason_code = -smc->sk.sk_err;
+		}
+	}
+	return reason_code;
+}
+
+/* send CLC ACCEPT message across internal TCP socket */
+int smc_clc_send_accept(struct smc_sock *new_smc, int srv_first_contact)
+{
+	struct smc_connection *conn = &new_smc->conn;
+	struct smc_clc_msg_accept_confirm aclc;
+	struct smc_link *link;
+	struct msghdr msg;
+	struct kvec vec;
+	int rc = 0;
+	int len;
+
+	link = &conn->lgr->lnk[SMC_SINGLE_LINK];
+	memset(&aclc, 0, sizeof(aclc));
+	memcpy(aclc.hdr.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
+	aclc.hdr.type = SMC_CLC_ACCEPT;
+	aclc.hdr.length = htons(sizeof(aclc));
+	aclc.hdr.version = SMC_CLC_V1;		/* SMC version */
+	if (srv_first_contact)
+		aclc.hdr.flag = 1;
+	memcpy(aclc.lcl.id_for_peer, local_systemid, sizeof(local_systemid));
+	memcpy(&aclc.lcl.gid, &link->smcibdev->gid[link->ibport - 1],
+	       SMC_GID_SIZE);
+	memcpy(&aclc.lcl.mac, link->smcibdev->mac[link->ibport - 1],
+	       sizeof(link->smcibdev->mac[link->ibport - 1]));
+	hton24(aclc.qpn, link->roce_qp->qp_num);
+	aclc.rmb_rkey =
+		htonl(conn->rmb_desc->mr_rx[SMC_SINGLE_LINK]->rkey);
+	aclc.conn_idx = 1;			/* as long as 1 RMB = 1 RMBE */
+	aclc.rmbe_alert_token = htonl(conn->alert_token_local);
+	aclc.qp_mtu = link->path_mtu;
+	aclc.rmbe_size = conn->rmbe_size_short,
+	aclc.rmb_dma_addr =
+		cpu_to_be64((u64)conn->rmb_desc->dma_addr[SMC_SINGLE_LINK]);
+	hton24(aclc.psn, link->psn_initial);
+	memcpy(aclc.trl.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
+
+	memset(&msg, 0, sizeof(msg));
+	vec.iov_base = &aclc;
+	vec.iov_len = sizeof(aclc);
+	len = kernel_sendmsg(new_smc->clcsock, &msg, &vec, 1, sizeof(aclc));
+	if (len < sizeof(aclc)) {
+		if (len >= 0)
+			new_smc->sk.sk_err = EPROTO;
+		else
+			new_smc->sk.sk_err = new_smc->clcsock->sk->sk_err;
+		rc = sock_error(&new_smc->sk);
+	}
+
+	return rc;
+}

net/smc/smc_clc.h

@@ -0,0 +1,116 @@
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  CLC (connection layer control) handshake over initial TCP socket to
+ *  prepare for RDMA traffic
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#ifndef _SMC_CLC_H
+#define _SMC_CLC_H
+
+#include <rdma/ib_verbs.h>
+
+#include "smc.h"
+
+#define SMC_CLC_PROPOSAL	0x01
+#define SMC_CLC_ACCEPT		0x02
+#define SMC_CLC_CONFIRM		0x03
+#define SMC_CLC_DECLINE		0x04
+
+/* eye catcher "SMCR" EBCDIC for CLC messages */
+static const char SMC_EYECATCHER[4] = {'\xe2', '\xd4', '\xc3', '\xd9'};
+
+#define SMC_CLC_V1		0x1		/* SMC version                */
+#define CLC_WAIT_TIME		(6 * HZ)	/* max. wait time on clcsock  */
+#define SMC_CLC_DECL_MEM	0x01010000  /* insufficient memory resources  */
+#define SMC_CLC_DECL_TIMEOUT	0x02000000  /* timeout                        */
+#define SMC_CLC_DECL_CNFERR	0x03000000  /* configuration error            */
+#define SMC_CLC_DECL_IPSEC	0x03030000  /* IPsec usage                    */
+#define SMC_CLC_DECL_SYNCERR	0x04000000  /* synchronization error          */
+#define SMC_CLC_DECL_REPLY	0x06000000  /* reply to a received decline    */
+#define SMC_CLC_DECL_INTERR	0x99990000  /* internal error                 */
+#define SMC_CLC_DECL_TCL	0x02040000  /* timeout w4 QP confirm          */
+#define SMC_CLC_DECL_SEND	0x07000000  /* sending problem                */
+
+struct smc_clc_msg_hdr {	/* header1 of clc messages */
+	u8 eyecatcher[4];	/* eye catcher */
+	u8 type;		/* proposal / accept / confirm / decline */
+	__be16 length;
+#if defined(__BIG_ENDIAN_BITFIELD)
+	u8 version : 4,
+	   flag    : 1,
+	   rsvd	   : 3;
+#elif defined(__LITTLE_ENDIAN_BITFIELD)
+	u8 rsvd    : 3,
+	   flag    : 1,
+	   version : 4;
+#endif
+} __packed;			/* format defined in RFC7609 */
+
+struct smc_clc_msg_trail {	/* trailer of clc messages */
+	u8 eyecatcher[4];
+};
+
+struct smc_clc_msg_local {	/* header2 of clc messages */
+	u8 id_for_peer[SMC_SYSTEMID_LEN]; /* unique system id */
+	u8 gid[16];		/* gid of ib_device port */
+	u8 mac[6];		/* mac of ib_device port */
+};
+
+struct smc_clc_msg_proposal {	/* clc proposal message */
+	struct smc_clc_msg_hdr hdr;
+	struct smc_clc_msg_local lcl;
+	__be16 iparea_offset;	/* offset to IP address information area */
+	__be32 outgoing_subnet;	/* subnet mask */
+	u8 prefix_len;		/* number of significant bits in mask */
+	u8 reserved[2];
+	u8 ipv6_prefixes_cnt;	/* number of IPv6 prefixes in prefix array */
+	struct smc_clc_msg_trail trl; /* eye catcher "SMCR" EBCDIC */
+} __aligned(4);
+
+struct smc_clc_msg_accept_confirm {	/* clc accept / confirm message */
+	struct smc_clc_msg_hdr hdr;
+	struct smc_clc_msg_local lcl;
+	u8 qpn[3];		/* QP number */
+	__be32 rmb_rkey;	/* RMB rkey */
+	u8 conn_idx;		/* Connection index, which RMBE in RMB */
+	__be32 rmbe_alert_token;/* unique connection id */
+#if defined(__BIG_ENDIAN_BITFIELD)
+	u8 rmbe_size : 4,	/* RMBE buf size (compressed notation) */
+	   qp_mtu   : 4;	/* QP mtu */
+#elif defined(__LITTLE_ENDIAN_BITFIELD)
+	u8 qp_mtu   : 4,
+	   rmbe_size : 4;
+#endif
+	u8 reserved;
+	__be64 rmb_dma_addr;	/* RMB virtual address */
+	u8 reserved2;
+	u8 psn[3];		/* initial packet sequence number */
+	struct smc_clc_msg_trail trl; /* eye catcher "SMCR" EBCDIC */
+} __packed;			/* format defined in RFC7609 */
+
+struct smc_clc_msg_decline {	/* clc decline message */
+	struct smc_clc_msg_hdr hdr;
+	u8 id_for_peer[SMC_SYSTEMID_LEN]; /* sender peer_id */
+	__be32 peer_diagnosis;	/* diagnosis information */
+	u8 reserved2[4];
+	struct smc_clc_msg_trail trl; /* eye catcher "SMCR" EBCDIC */
+} __aligned(4);
+
+struct smc_sock;
+struct smc_ib_device;
+
+int smc_clc_wait_msg(struct smc_sock *smc, void *buf, int buflen,
+		     u8 expected_type);
+int smc_clc_send_decline(struct smc_sock *smc, u32 peer_diag_info,
+			 u8 out_of_sync);
+int smc_clc_send_proposal(struct smc_sock *smc, struct smc_ib_device *smcibdev,
+			  u8 ibport);
+int smc_clc_send_confirm(struct smc_sock *smc);
+int smc_clc_send_accept(struct smc_sock *smc, int srv_first_contact);
+
+#endif

net/smc/smc_close.c

@@ -0,0 +1,442 @@
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Socket Closing - normal and abnormal
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#include <linux/workqueue.h>
+#include <net/sock.h>
+
+#include "smc.h"
+#include "smc_tx.h"
+#include "smc_cdc.h"
+#include "smc_close.h"
+
+#define SMC_CLOSE_WAIT_TX_PENDS_TIME		(5 * HZ)
+
+static void smc_close_cleanup_listen(struct sock *parent)
+{
+	struct sock *sk;
+
+	/* Close non-accepted connections */
+	while ((sk = smc_accept_dequeue(parent, NULL)))
+		smc_close_non_accepted(sk);
+}
+
+static void smc_close_wait_tx_pends(struct smc_sock *smc)
+{
+	DEFINE_WAIT_FUNC(wait, woken_wake_function);
+	struct sock *sk = &smc->sk;
+	signed long timeout;
+
+	timeout = SMC_CLOSE_WAIT_TX_PENDS_TIME;
+	add_wait_queue(sk_sleep(sk), &wait);
+	while (!signal_pending(current) && timeout) {
+		int rc;
+
+		rc = sk_wait_event(sk, &timeout,
+				   !smc_cdc_tx_has_pending(&smc->conn),
+				   &wait);
+		if (rc)
+			break;
+	}
+	remove_wait_queue(sk_sleep(sk), &wait);
+}
+
+/* wait for sndbuf data being transmitted */
+static void smc_close_stream_wait(struct smc_sock *smc, long timeout)
+{
+	DEFINE_WAIT_FUNC(wait, woken_wake_function);
+	struct sock *sk = &smc->sk;
+
+	if (!timeout)
+		return;
+
+	if (!smc_tx_prepared_sends(&smc->conn))
+		return;
+
+	smc->wait_close_tx_prepared = 1;
+	add_wait_queue(sk_sleep(sk), &wait);
+	while (!signal_pending(current) && timeout) {
+		int rc;
+
+		rc = sk_wait_event(sk, &timeout,
+				   !smc_tx_prepared_sends(&smc->conn) ||
+				   (sk->sk_err == ECONNABORTED) ||
+				   (sk->sk_err == ECONNRESET),
+				   &wait);
+		if (rc)
+			break;
+	}
+	remove_wait_queue(sk_sleep(sk), &wait);
+	smc->wait_close_tx_prepared = 0;
+}
+
+void smc_close_wake_tx_prepared(struct smc_sock *smc)
+{
+	if (smc->wait_close_tx_prepared)
+		/* wake up socket closing */
+		smc->sk.sk_state_change(&smc->sk);
+}
+
+static int smc_close_wr(struct smc_connection *conn)
+{
+	conn->local_tx_ctrl.conn_state_flags.peer_done_writing = 1;
+
+	return smc_cdc_get_slot_and_msg_send(conn);
+}
+
+static int smc_close_final(struct smc_connection *conn)
+{
+	if (atomic_read(&conn->bytes_to_rcv))
+		conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
+	else
+		conn->local_tx_ctrl.conn_state_flags.peer_conn_closed = 1;
+
+	return smc_cdc_get_slot_and_msg_send(conn);
+}
+
+static int smc_close_abort(struct smc_connection *conn)
+{
+	conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
+
+	return smc_cdc_get_slot_and_msg_send(conn);
+}
+
+/* terminate smc socket abnormally - active abort
+ * RDMA communication no longer possible
+ */
+void smc_close_active_abort(struct smc_sock *smc)
+{
+	struct smc_cdc_conn_state_flags *txflags =
+		&smc->conn.local_tx_ctrl.conn_state_flags;
+
+	bh_lock_sock(&smc->sk);
+	smc->sk.sk_err = ECONNABORTED;
+	if (smc->clcsock && smc->clcsock->sk) {
+		smc->clcsock->sk->sk_err = ECONNABORTED;
+		smc->clcsock->sk->sk_state_change(smc->clcsock->sk);
+	}
+	switch (smc->sk.sk_state) {
+	case SMC_INIT:
+		smc->sk.sk_state = SMC_PEERABORTWAIT;
+		break;
+	case SMC_APPCLOSEWAIT1:
+	case SMC_APPCLOSEWAIT2:
+		txflags->peer_conn_abort = 1;
+		sock_release(smc->clcsock);
+		if (!smc_cdc_rxed_any_close(&smc->conn))
+			smc->sk.sk_state = SMC_PEERABORTWAIT;
+		else
+			smc->sk.sk_state = SMC_CLOSED;
+		break;
+	case SMC_PEERCLOSEWAIT1:
+	case SMC_PEERCLOSEWAIT2:
+		if (!txflags->peer_conn_closed) {
+			smc->sk.sk_state = SMC_PEERABORTWAIT;
+			txflags->peer_conn_abort = 1;
+			sock_release(smc->clcsock);
+		} else {
+			smc->sk.sk_state = SMC_CLOSED;
+		}
+		break;
+	case SMC_PROCESSABORT:
+	case SMC_APPFINCLOSEWAIT:
+		if (!txflags->peer_conn_closed) {
+			txflags->peer_conn_abort = 1;
+			sock_release(smc->clcsock);
+		}
+		smc->sk.sk_state = SMC_CLOSED;
+		break;
+	case SMC_PEERFINCLOSEWAIT:
+	case SMC_PEERABORTWAIT:
+	case SMC_CLOSED:
+		break;
+	}
+
+	sock_set_flag(&smc->sk, SOCK_DEAD);
+	bh_unlock_sock(&smc->sk);
+	smc->sk.sk_state_change(&smc->sk);
+}
+
+int smc_close_active(struct smc_sock *smc)
+{
+	struct smc_cdc_conn_state_flags *txflags =
+		&smc->conn.local_tx_ctrl.conn_state_flags;
+	long timeout = SMC_MAX_STREAM_WAIT_TIMEOUT;
+	struct smc_connection *conn = &smc->conn;
+	struct sock *sk = &smc->sk;
+	int old_state;
+	int rc = 0;
+
+	if (sock_flag(sk, SOCK_LINGER) &&
+	    !(current->flags & PF_EXITING))
+		timeout = sk->sk_lingertime;
+
+again:
+	old_state = sk->sk_state;
+	switch (old_state) {
+	case SMC_INIT:
+		sk->sk_state = SMC_CLOSED;
+		if (smc->smc_listen_work.func)
+			flush_work(&smc->smc_listen_work);
+		sock_put(sk);
+		break;
+	case SMC_LISTEN:
+		sk->sk_state = SMC_CLOSED;
+		sk->sk_state_change(sk); /* wake up accept */
+		if (smc->clcsock && smc->clcsock->sk) {
+			rc = kernel_sock_shutdown(smc->clcsock, SHUT_RDWR);
+			/* wake up kernel_accept of smc_tcp_listen_worker */
+			smc->clcsock->sk->sk_data_ready(smc->clcsock->sk);
+		}
+		release_sock(sk);
+		smc_close_cleanup_listen(sk);
+		flush_work(&smc->tcp_listen_work);
+		lock_sock(sk);
+		break;
+	case SMC_ACTIVE:
+		smc_close_stream_wait(smc, timeout);
+		release_sock(sk);
+		cancel_work_sync(&conn->tx_work);
+		lock_sock(sk);
+		if (sk->sk_state == SMC_ACTIVE) {
+			/* send close request */
+			rc = smc_close_final(conn);
+			sk->sk_state = SMC_PEERCLOSEWAIT1;
+		} else {
+			/* peer event has changed the state */
+			goto again;
+		}
+		break;
+	case SMC_APPFINCLOSEWAIT:
+		/* socket already shutdown wr or both (active close) */
+		if (txflags->peer_done_writing &&
+		    !txflags->peer_conn_closed) {
+			/* just shutdown wr done, send close request */
+			rc = smc_close_final(conn);
+		}
+		sk->sk_state = SMC_CLOSED;
+		smc_close_wait_tx_pends(smc);
+		break;
+	case SMC_APPCLOSEWAIT1:
+	case SMC_APPCLOSEWAIT2:
+		if (!smc_cdc_rxed_any_close(conn))
+			smc_close_stream_wait(smc, timeout);
+		release_sock(sk);
+		cancel_work_sync(&conn->tx_work);
+		lock_sock(sk);
+		if (sk->sk_err != ECONNABORTED) {
+			/* confirm close from peer */
+			rc = smc_close_final(conn);
+			if (rc)
+				break;
+		}
+		if (smc_cdc_rxed_any_close(conn))
+			/* peer has closed the socket already */
+			sk->sk_state = SMC_CLOSED;
+		else
+			/* peer has just issued a shutdown write */
+			sk->sk_state = SMC_PEERFINCLOSEWAIT;
+		smc_close_wait_tx_pends(smc);
+		break;
+	case SMC_PEERCLOSEWAIT1:
+	case SMC_PEERCLOSEWAIT2:
+	case SMC_PEERFINCLOSEWAIT:
+		/* peer sending PeerConnectionClosed will cause transition */
+		break;
+	case SMC_PROCESSABORT:
+		cancel_work_sync(&conn->tx_work);
+		smc_close_abort(conn);
+		sk->sk_state = SMC_CLOSED;
+		smc_close_wait_tx_pends(smc);
+		break;
+	case SMC_PEERABORTWAIT:
+	case SMC_CLOSED:
+		/* nothing to do, add tracing in future patch */
+		break;
+	}
+
+	if (old_state != sk->sk_state)
+		sk->sk_state_change(&smc->sk);
+	return rc;
+}
+
+static void smc_close_passive_abort_received(struct smc_sock *smc)
+{
+	struct smc_cdc_conn_state_flags *txflags =
+		&smc->conn.local_tx_ctrl.conn_state_flags;
+	struct sock *sk = &smc->sk;
+
+	switch (sk->sk_state) {
+	case SMC_ACTIVE:
+	case SMC_APPFINCLOSEWAIT:
+	case SMC_APPCLOSEWAIT1:
+	case SMC_APPCLOSEWAIT2:
+		smc_close_abort(&smc->conn);
+		sk->sk_state = SMC_PROCESSABORT;
+		break;
+	case SMC_PEERCLOSEWAIT1:
+	case SMC_PEERCLOSEWAIT2:
+		if (txflags->peer_done_writing &&
+		    !txflags->peer_conn_closed) {
+			/* just shutdown, but not yet closed locally */
+			smc_close_abort(&smc->conn);
+			sk->sk_state = SMC_PROCESSABORT;
+		} else {
+			sk->sk_state = SMC_CLOSED;
+		}
+		break;
+	case SMC_PEERFINCLOSEWAIT:
+	case SMC_PEERABORTWAIT:
+		sk->sk_state = SMC_CLOSED;
+		break;
+	case SMC_INIT:
+	case SMC_PROCESSABORT:
+	/* nothing to do, add tracing in future patch */
+		break;
+	}
+}
+
+/* Some kind of closing has been received: peer_conn_closed, peer_conn_abort,
+ * or peer_done_writing.
+ * Called under tasklet context.
+ */
+void smc_close_passive_received(struct smc_sock *smc)
+{
+	struct smc_cdc_conn_state_flags *rxflags =
+		&smc->conn.local_rx_ctrl.conn_state_flags;
+	struct sock *sk = &smc->sk;
+	int old_state;
+
+	sk->sk_shutdown |= RCV_SHUTDOWN;
+	if (smc->clcsock && smc->clcsock->sk)
+		smc->clcsock->sk->sk_shutdown |= RCV_SHUTDOWN;
+	sock_set_flag(&smc->sk, SOCK_DONE);
+
+	old_state = sk->sk_state;
+
+	if (rxflags->peer_conn_abort) {
+		smc_close_passive_abort_received(smc);
+		goto wakeup;
+	}
+
+	switch (sk->sk_state) {
+	case SMC_INIT:
+		if (atomic_read(&smc->conn.bytes_to_rcv) ||
+		    (rxflags->peer_done_writing &&
+		     !rxflags->peer_conn_closed))
+			sk->sk_state = SMC_APPCLOSEWAIT1;
+		else
+			sk->sk_state = SMC_CLOSED;
+		break;
+	case SMC_ACTIVE:
+		sk->sk_state = SMC_APPCLOSEWAIT1;
+		break;
+	case SMC_PEERCLOSEWAIT1:
+		if (rxflags->peer_done_writing)
+			sk->sk_state = SMC_PEERCLOSEWAIT2;
+		/* fall through to check for closing */
+	case SMC_PEERCLOSEWAIT2:
+	case SMC_PEERFINCLOSEWAIT:
+		if (!smc_cdc_rxed_any_close(&smc->conn))
+			break;
+		if (sock_flag(sk, SOCK_DEAD) &&
+		    (sk->sk_shutdown == SHUTDOWN_MASK)) {
+			/* smc_release has already been called locally */
+			sk->sk_state = SMC_CLOSED;
+		} else {
+			/* just shutdown, but not yet closed locally */
+			sk->sk_state = SMC_APPFINCLOSEWAIT;
+		}
+		break;
+	case SMC_APPCLOSEWAIT1:
+	case SMC_APPCLOSEWAIT2:
+	case SMC_APPFINCLOSEWAIT:
+	case SMC_PEERABORTWAIT:
+	case SMC_PROCESSABORT:
+	case SMC_CLOSED:
+		/* nothing to do, add tracing in future patch */
+		break;
+	}
+
+wakeup:
+	if (old_state != sk->sk_state)
+		sk->sk_state_change(sk);
+	sk->sk_data_ready(sk); /* wakeup blocked rcvbuf consumers */
+	sk->sk_write_space(sk); /* wakeup blocked sndbuf producers */
+
+	if ((sk->sk_state == SMC_CLOSED) &&
+	    (sock_flag(sk, SOCK_DEAD) || (old_state == SMC_INIT))) {
+		smc_conn_free(&smc->conn);
+		schedule_delayed_work(&smc->sock_put_work,
+				      SMC_CLOSE_SOCK_PUT_DELAY);
+	}
+}
+
+void smc_close_sock_put_work(struct work_struct *work)
+{
+	struct smc_sock *smc = container_of(to_delayed_work(work),
+					    struct smc_sock,
+					    sock_put_work);
+
+	smc->sk.sk_prot->unhash(&smc->sk);
+	sock_put(&smc->sk);
+}
+
+int smc_close_shutdown_write(struct smc_sock *smc)
+{
+	struct smc_connection *conn = &smc->conn;
+	long timeout = SMC_MAX_STREAM_WAIT_TIMEOUT;
+	struct sock *sk = &smc->sk;
+	int old_state;
+	int rc = 0;
+
+	if (sock_flag(sk, SOCK_LINGER))
+		timeout = sk->sk_lingertime;
+
+again:
+	old_state = sk->sk_state;
+	switch (old_state) {
+	case SMC_ACTIVE:
+		smc_close_stream_wait(smc, timeout);
+		release_sock(sk);
+		cancel_work_sync(&conn->tx_work);
+		lock_sock(sk);
+		/* send close wr request */
+		rc = smc_close_wr(conn);
+		if (sk->sk_state == SMC_ACTIVE)
+			sk->sk_state = SMC_PEERCLOSEWAIT1;
+		else
+			goto again;
+		break;
+	case SMC_APPCLOSEWAIT1:
+		/* passive close */
+		if (!smc_cdc_rxed_any_close(conn))
+			smc_close_stream_wait(smc, timeout);
+		release_sock(sk);
+		cancel_work_sync(&conn->tx_work);
+		lock_sock(sk);
+		/* confirm close from peer */
+		rc = smc_close_wr(conn);
+		sk->sk_state = SMC_APPCLOSEWAIT2;
+		break;
+	case SMC_APPCLOSEWAIT2:
+	case SMC_PEERFINCLOSEWAIT:
+	case SMC_PEERCLOSEWAIT1:
+	case SMC_PEERCLOSEWAIT2:
+	case SMC_APPFINCLOSEWAIT:
+	case SMC_PROCESSABORT:
+	case SMC_PEERABORTWAIT:
+		/* nothing to do, add tracing in future patch */
+		break;
+	}
+
+	if (old_state != sk->sk_state)
+		sk->sk_state_change(&smc->sk);
+	return rc;
+}

net/smc/smc_close.h

@@ -0,0 +1,28 @@
+/*
+ * Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ * Socket Closing
+ *
+ * Copyright IBM Corp. 2016
+ *
+ * Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#ifndef SMC_CLOSE_H
+#define SMC_CLOSE_H
+
+#include <linux/workqueue.h>
+
+#include "smc.h"
+
+#define SMC_MAX_STREAM_WAIT_TIMEOUT		(2 * HZ)
+#define SMC_CLOSE_SOCK_PUT_DELAY		HZ
+
+void smc_close_wake_tx_prepared(struct smc_sock *smc);
+void smc_close_active_abort(struct smc_sock *smc);
+int smc_close_active(struct smc_sock *smc);
+void smc_close_passive_received(struct smc_sock *smc);
+void smc_close_sock_put_work(struct work_struct *work);
+int smc_close_shutdown_write(struct smc_sock *smc);
+
+#endif /* SMC_CLOSE_H */

net/smc/smc_core.c

@@ -0,0 +1,677 @@
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Basic Transport Functions exploiting Infiniband API
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#include <linux/socket.h>
+#include <linux/if_vlan.h>
+#include <linux/random.h>
+#include <linux/workqueue.h>
+#include <net/tcp.h>
+#include <net/sock.h>
+#include <rdma/ib_verbs.h>
+
+#include "smc.h"
+#include "smc_clc.h"
+#include "smc_core.h"
+#include "smc_ib.h"
+#include "smc_wr.h"
+#include "smc_llc.h"
+#include "smc_cdc.h"
+#include "smc_close.h"
+
+#define SMC_LGR_NUM_INCR	256
+#define SMC_LGR_FREE_DELAY	(600 * HZ)
+
+static u32 smc_lgr_num;			/* unique link group number */
+
+/* Register connection's alert token in our lookup structure.
+ * To use rbtrees we have to implement our own insert core.
+ * Requires @conns_lock
+ * @smc		connection to register
+ * Returns 0 on success, != otherwise.
+ */
+static void smc_lgr_add_alert_token(struct smc_connection *conn)
+{
+	struct rb_node **link, *parent = NULL;
+	u32 token = conn->alert_token_local;
+
+	link = &conn->lgr->conns_all.rb_node;
+	while (*link) {
+		struct smc_connection *cur = rb_entry(*link,
+					struct smc_connection, alert_node);
+
+		parent = *link;
+		if (cur->alert_token_local > token)
+			link = &parent->rb_left;
+		else
+			link = &parent->rb_right;
+	}
+	/* Put the new node there */
+	rb_link_node(&conn->alert_node, parent, link);
+	rb_insert_color(&conn->alert_node, &conn->lgr->conns_all);
+}
+
+/* Register connection in link group by assigning an alert token
+ * registered in a search tree.
+ * Requires @conns_lock
+ * Note that '0' is a reserved value and not assigned.
+ */
+static void smc_lgr_register_conn(struct smc_connection *conn)
+{
+	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
+	static atomic_t nexttoken = ATOMIC_INIT(0);
+
+	/* find a new alert_token_local value not yet used by some connection
+	 * in this link group
+	 */
+	sock_hold(&smc->sk); /* sock_put in smc_lgr_unregister_conn() */
+	while (!conn->alert_token_local) {
+		conn->alert_token_local = atomic_inc_return(&nexttoken);
+		if (smc_lgr_find_conn(conn->alert_token_local, conn->lgr))
+			conn->alert_token_local = 0;
+	}
+	smc_lgr_add_alert_token(conn);
+	conn->lgr->conns_num++;
+}
+
+/* Unregister connection and reset the alert token of the given connection<
+ */
+static void __smc_lgr_unregister_conn(struct smc_connection *conn)
+{
+	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
+	struct smc_link_group *lgr = conn->lgr;
+
+	rb_erase(&conn->alert_node, &lgr->conns_all);
+	lgr->conns_num--;
+	conn->alert_token_local = 0;
+	conn->lgr = NULL;
+	sock_put(&smc->sk); /* sock_hold in smc_lgr_register_conn() */
+}
+
+/* Unregister connection and trigger lgr freeing if applicable
+ */
+static void smc_lgr_unregister_conn(struct smc_connection *conn)
+{
+	struct smc_link_group *lgr = conn->lgr;
+	int reduced = 0;
+
+	write_lock_bh(&lgr->conns_lock);
+	if (conn->alert_token_local) {
+		reduced = 1;
+		__smc_lgr_unregister_conn(conn);
+	}
+	write_unlock_bh(&lgr->conns_lock);
+	if (reduced && !lgr->conns_num)
+		schedule_delayed_work(&lgr->free_work, SMC_LGR_FREE_DELAY);
+}
+
+static void smc_lgr_free_work(struct work_struct *work)
+{
+	struct smc_link_group *lgr = container_of(to_delayed_work(work),
+						  struct smc_link_group,
+						  free_work);
+	bool conns;
+
+	spin_lock_bh(&smc_lgr_list.lock);
+	read_lock_bh(&lgr->conns_lock);
+	conns = RB_EMPTY_ROOT(&lgr->conns_all);
+	read_unlock_bh(&lgr->conns_lock);
+	if (!conns) { /* number of lgr connections is no longer zero */
+		spin_unlock_bh(&smc_lgr_list.lock);
+		return;
+	}
+	list_del_init(&lgr->list); /* remove from smc_lgr_list */
+	spin_unlock_bh(&smc_lgr_list.lock);
+	smc_lgr_free(lgr);
+}
+
+/* create a new SMC link group */
+static int smc_lgr_create(struct smc_sock *smc, __be32 peer_in_addr,
+			  struct smc_ib_device *smcibdev, u8 ibport,
+			  char *peer_systemid, unsigned short vlan_id)
+{
+	struct smc_link_group *lgr;
+	struct smc_link *lnk;
+	u8 rndvec[3];
+	int rc = 0;
+	int i;
+
+	lgr = kzalloc(sizeof(*lgr), GFP_KERNEL);
+	if (!lgr) {
+		rc = -ENOMEM;
+		goto out;
+	}
+	lgr->role = smc->listen_smc ? SMC_SERV : SMC_CLNT;
+	lgr->sync_err = false;
+	lgr->daddr = peer_in_addr;
+	memcpy(lgr->peer_systemid, peer_systemid, SMC_SYSTEMID_LEN);
+	lgr->vlan_id = vlan_id;
+	rwlock_init(&lgr->sndbufs_lock);
+	rwlock_init(&lgr->rmbs_lock);
+	for (i = 0; i < SMC_RMBE_SIZES; i++) {
+		INIT_LIST_HEAD(&lgr->sndbufs[i]);
+		INIT_LIST_HEAD(&lgr->rmbs[i]);
+	}
+	smc_lgr_num += SMC_LGR_NUM_INCR;
+	memcpy(&lgr->id, (u8 *)&smc_lgr_num, SMC_LGR_ID_SIZE);
+	INIT_DELAYED_WORK(&lgr->free_work, smc_lgr_free_work);
+	lgr->conns_all = RB_ROOT;
+
+	lnk = &lgr->lnk[SMC_SINGLE_LINK];
+	/* initialize link */
+	lnk->smcibdev = smcibdev;
+	lnk->ibport = ibport;
+	lnk->path_mtu = smcibdev->pattr[ibport - 1].active_mtu;
+	if (!smcibdev->initialized)
+		smc_ib_setup_per_ibdev(smcibdev);
+	get_random_bytes(rndvec, sizeof(rndvec));
+	lnk->psn_initial = rndvec[0] + (rndvec[1] << 8) + (rndvec[2] << 16);
+	rc = smc_wr_alloc_link_mem(lnk);
+	if (rc)
+		goto free_lgr;
+	init_waitqueue_head(&lnk->wr_tx_wait);
+	rc = smc_ib_create_protection_domain(lnk);
+	if (rc)
+		goto free_link_mem;
+	rc = smc_ib_create_queue_pair(lnk);
+	if (rc)
+		goto dealloc_pd;
+	rc = smc_wr_create_link(lnk);
+	if (rc)
+		goto destroy_qp;
+	init_completion(&lnk->llc_confirm);
+	init_completion(&lnk->llc_confirm_resp);
+
+	smc->conn.lgr = lgr;
+	rwlock_init(&lgr->conns_lock);
+	spin_lock_bh(&smc_lgr_list.lock);
+	list_add(&lgr->list, &smc_lgr_list.list);
+	spin_unlock_bh(&smc_lgr_list.lock);
+	return 0;
+
+destroy_qp:
+	smc_ib_destroy_queue_pair(lnk);
+dealloc_pd:
+	smc_ib_dealloc_protection_domain(lnk);
+free_link_mem:
+	smc_wr_free_link_mem(lnk);
+free_lgr:
+	kfree(lgr);
+out:
+	return rc;
+}
+
+static void smc_sndbuf_unuse(struct smc_connection *conn)
+{
+	if (conn->sndbuf_desc) {
+		conn->sndbuf_desc->used = 0;
+		conn->sndbuf_size = 0;
+	}
+}
+
+static void smc_rmb_unuse(struct smc_connection *conn)
+{
+	if (conn->rmb_desc) {
+		conn->rmb_desc->used = 0;
+		conn->rmbe_size = 0;
+	}
+}
+
+/* remove a finished connection from its link group */
+void smc_conn_free(struct smc_connection *conn)
+{
+	struct smc_link_group *lgr = conn->lgr;
+
+	if (!lgr)
+		return;
+	smc_cdc_tx_dismiss_slots(conn);
+	smc_lgr_unregister_conn(conn);
+	smc_rmb_unuse(conn);
+	smc_sndbuf_unuse(conn);
+}
+
+static void smc_link_clear(struct smc_link *lnk)
+{
+	lnk->peer_qpn = 0;
+	smc_ib_modify_qp_reset(lnk);
+	smc_wr_free_link(lnk);
+	smc_ib_destroy_queue_pair(lnk);
+	smc_ib_dealloc_protection_domain(lnk);
+	smc_wr_free_link_mem(lnk);
+}
+
+static void smc_lgr_free_sndbufs(struct smc_link_group *lgr)
+{
+	struct smc_buf_desc *sndbuf_desc, *bf_desc;
+	int i;
+
+	for (i = 0; i < SMC_RMBE_SIZES; i++) {
+		list_for_each_entry_safe(sndbuf_desc, bf_desc, &lgr->sndbufs[i],
+					 list) {
+			list_del(&sndbuf_desc->list);
+			smc_ib_buf_unmap(lgr->lnk[SMC_SINGLE_LINK].smcibdev,
+					 smc_uncompress_bufsize(i),
+					 sndbuf_desc, DMA_TO_DEVICE);
+			kfree(sndbuf_desc->cpu_addr);
+			kfree(sndbuf_desc);
+		}
+	}
+}
+
+static void smc_lgr_free_rmbs(struct smc_link_group *lgr)
+{
+	struct smc_buf_desc *rmb_desc, *bf_desc;
+	struct smc_link *lnk = &lgr->lnk[SMC_SINGLE_LINK];
+	int i;
+
+	for (i = 0; i < SMC_RMBE_SIZES; i++) {
+		list_for_each_entry_safe(rmb_desc, bf_desc, &lgr->rmbs[i],
+					 list) {
+			list_del(&rmb_desc->list);
+			smc_ib_buf_unmap(lnk->smcibdev,
+					 smc_uncompress_bufsize(i),
+					 rmb_desc, DMA_FROM_DEVICE);
+			kfree(rmb_desc->cpu_addr);
+			kfree(rmb_desc);
+		}
+	}
+}
+
+/* remove a link group */
+void smc_lgr_free(struct smc_link_group *lgr)
+{
+	smc_lgr_free_rmbs(lgr);
+	smc_lgr_free_sndbufs(lgr);
+	smc_link_clear(&lgr->lnk[SMC_SINGLE_LINK]);
+	kfree(lgr);
+}
+
+/* terminate linkgroup abnormally */
+void smc_lgr_terminate(struct smc_link_group *lgr)
+{
+	struct smc_connection *conn;
+	struct smc_sock *smc;
+	struct rb_node *node;
+
+	spin_lock_bh(&smc_lgr_list.lock);
+	if (list_empty(&lgr->list)) {
+		/* termination already triggered */
+		spin_unlock_bh(&smc_lgr_list.lock);
+		return;
+	}
+	/* do not use this link group for new connections */
+	list_del_init(&lgr->list);
+	spin_unlock_bh(&smc_lgr_list.lock);
+
+	write_lock_bh(&lgr->conns_lock);
+	node = rb_first(&lgr->conns_all);
+	while (node) {
+		conn = rb_entry(node, struct smc_connection, alert_node);
+		smc = container_of(conn, struct smc_sock, conn);
+		sock_hold(&smc->sk);
+		__smc_lgr_unregister_conn(conn);
+		smc_close_active_abort(smc);
+		sock_put(&smc->sk);
+		node = rb_first(&lgr->conns_all);
+	}
+	write_unlock_bh(&lgr->conns_lock);
+}
+
+/* Determine vlan of internal TCP socket.
+ * @vlan_id: address to store the determined vlan id into
+ */
+static int smc_vlan_by_tcpsk(struct socket *clcsock, unsigned short *vlan_id)
+{
+	struct dst_entry *dst = sk_dst_get(clcsock->sk);
+	int rc = 0;
+
+	*vlan_id = 0;
+	if (!dst) {
+		rc = -ENOTCONN;
+		goto out;
+	}
+	if (!dst->dev) {
+		rc = -ENODEV;
+		goto out_rel;
+	}
+
+	if (is_vlan_dev(dst->dev))
+		*vlan_id = vlan_dev_vlan_id(dst->dev);
+
+out_rel:
+	dst_release(dst);
+out:
+	return rc;
+}
+
+/* determine the link gid matching the vlan id of the link group */
+static int smc_link_determine_gid(struct smc_link_group *lgr)
+{
+	struct smc_link *lnk = &lgr->lnk[SMC_SINGLE_LINK];
+	struct ib_gid_attr gattr;
+	union ib_gid gid;
+	int i;
+
+	if (!lgr->vlan_id) {
+		lnk->gid = lnk->smcibdev->gid[lnk->ibport - 1];
+		return 0;
+	}
+
+	for (i = 0; i < lnk->smcibdev->pattr[lnk->ibport - 1].gid_tbl_len;
+	     i++) {
+		if (ib_query_gid(lnk->smcibdev->ibdev, lnk->ibport, i, &gid,
+				 &gattr))
+			continue;
+		if (gattr.ndev &&
+		    (vlan_dev_vlan_id(gattr.ndev) == lgr->vlan_id)) {
+			lnk->gid = gid;
+			return 0;
+		}
+	}
+	return -ENODEV;
+}
+
+/* create a new SMC connection (and a new link group if necessary) */
+int smc_conn_create(struct smc_sock *smc, __be32 peer_in_addr,
+		    struct smc_ib_device *smcibdev, u8 ibport,
+		    struct smc_clc_msg_local *lcl, int srv_first_contact)
+{
+	struct smc_connection *conn = &smc->conn;
+	struct smc_link_group *lgr;
+	unsigned short vlan_id;
+	enum smc_lgr_role role;
+	int local_contact = SMC_FIRST_CONTACT;
+	int rc = 0;
+
+	role = smc->listen_smc ? SMC_SERV : SMC_CLNT;
+	rc = smc_vlan_by_tcpsk(smc->clcsock, &vlan_id);
+	if (rc)
+		return rc;
+
+	if ((role == SMC_CLNT) && srv_first_contact)
+		/* create new link group as well */
+		goto create;
+
+	/* determine if an existing link group can be reused */
+	spin_lock_bh(&smc_lgr_list.lock);
+	list_for_each_entry(lgr, &smc_lgr_list.list, list) {
+		write_lock_bh(&lgr->conns_lock);
+		if (!memcmp(lgr->peer_systemid, lcl->id_for_peer,
+			    SMC_SYSTEMID_LEN) &&
+		    !memcmp(lgr->lnk[SMC_SINGLE_LINK].peer_gid, &lcl->gid,
+			    SMC_GID_SIZE) &&
+		    !memcmp(lgr->lnk[SMC_SINGLE_LINK].peer_mac, lcl->mac,
+			    sizeof(lcl->mac)) &&
+		    !lgr->sync_err &&
+		    (lgr->role == role) &&
+		    (lgr->vlan_id == vlan_id) &&
+		    ((role == SMC_CLNT) ||
+		     (lgr->conns_num < SMC_RMBS_PER_LGR_MAX))) {
+			/* link group found */
+			local_contact = SMC_REUSE_CONTACT;
+			conn->lgr = lgr;
+			smc_lgr_register_conn(conn); /* add smc conn to lgr */
+			write_unlock_bh(&lgr->conns_lock);
+			break;
+		}
+		write_unlock_bh(&lgr->conns_lock);
+	}
+	spin_unlock_bh(&smc_lgr_list.lock);
+
+	if (role == SMC_CLNT && !srv_first_contact &&
+	    (local_contact == SMC_FIRST_CONTACT)) {
+		/* Server reuses a link group, but Client wants to start
+		 * a new one
+		 * send out_of_sync decline, reason synchr. error
+		 */
+		return -ENOLINK;
+	}
+
+create:
+	if (local_contact == SMC_FIRST_CONTACT) {
+		rc = smc_lgr_create(smc, peer_in_addr, smcibdev, ibport,
+				    lcl->id_for_peer, vlan_id);
+		if (rc)
+			goto out;
+		smc_lgr_register_conn(conn); /* add smc conn to lgr */
+		rc = smc_link_determine_gid(conn->lgr);
+	}
+	conn->local_tx_ctrl.common.type = SMC_CDC_MSG_TYPE;
+	conn->local_tx_ctrl.len = sizeof(struct smc_cdc_msg);
+#ifndef KERNEL_HAS_ATOMIC64
+	spin_lock_init(&conn->acurs_lock);
+#endif
+
+out:
+	return rc ? rc : local_contact;
+}
+
+/* try to reuse a sndbuf description slot of the sndbufs list for a certain
+ * buf_size; if not available, return NULL
+ */
+static inline
+struct smc_buf_desc *smc_sndbuf_get_slot(struct smc_link_group *lgr,
+					 int compressed_bufsize)
+{
+	struct smc_buf_desc *sndbuf_slot;
+
+	read_lock_bh(&lgr->sndbufs_lock);
+	list_for_each_entry(sndbuf_slot, &lgr->sndbufs[compressed_bufsize],
+			    list) {
+		if (cmpxchg(&sndbuf_slot->used, 0, 1) == 0) {
+			read_unlock_bh(&lgr->sndbufs_lock);
+			return sndbuf_slot;
+		}
+	}
+	read_unlock_bh(&lgr->sndbufs_lock);
+	return NULL;
+}
+
+/* try to reuse an rmb description slot of the rmbs list for a certain
+ * rmbe_size; if not available, return NULL
+ */
+static inline
+struct smc_buf_desc *smc_rmb_get_slot(struct smc_link_group *lgr,
+				      int compressed_bufsize)
+{
+	struct smc_buf_desc *rmb_slot;
+
+	read_lock_bh(&lgr->rmbs_lock);
+	list_for_each_entry(rmb_slot, &lgr->rmbs[compressed_bufsize],
+			    list) {
+		if (cmpxchg(&rmb_slot->used, 0, 1) == 0) {
+			read_unlock_bh(&lgr->rmbs_lock);
+			return rmb_slot;
+		}
+	}
+	read_unlock_bh(&lgr->rmbs_lock);
+	return NULL;
+}
+
+/* one of the conditions for announcing a receiver's current window size is
+ * that it "results in a minimum increase in the window size of 10% of the
+ * receive buffer space" [RFC7609]
+ */
+static inline int smc_rmb_wnd_update_limit(int rmbe_size)
+{
+	return min_t(int, rmbe_size / 10, SOCK_MIN_SNDBUF / 2);
+}
+
+/* create the tx buffer for an SMC socket */
+int smc_sndbuf_create(struct smc_sock *smc)
+{
+	struct smc_connection *conn = &smc->conn;
+	struct smc_link_group *lgr = conn->lgr;
+	int tmp_bufsize, tmp_bufsize_short;
+	struct smc_buf_desc *sndbuf_desc;
+	int rc;
+
+	/* use socket send buffer size (w/o overhead) as start value */
+	for (tmp_bufsize_short = smc_compress_bufsize(smc->sk.sk_sndbuf / 2);
+	     tmp_bufsize_short >= 0; tmp_bufsize_short--) {
+		tmp_bufsize = smc_uncompress_bufsize(tmp_bufsize_short);
+		/* check for reusable sndbuf_slot in the link group */
+		sndbuf_desc = smc_sndbuf_get_slot(lgr, tmp_bufsize_short);
+		if (sndbuf_desc) {
+			memset(sndbuf_desc->cpu_addr, 0, tmp_bufsize);
+			break; /* found reusable slot */
+		}
+		/* try to alloc a new send buffer */
+		sndbuf_desc = kzalloc(sizeof(*sndbuf_desc), GFP_KERNEL);
+		if (!sndbuf_desc)
+			break; /* give up with -ENOMEM */
+		sndbuf_desc->cpu_addr = kzalloc(tmp_bufsize,
+						GFP_KERNEL | __GFP_NOWARN |
+						__GFP_NOMEMALLOC |
+						__GFP_NORETRY);
+		if (!sndbuf_desc->cpu_addr) {
+			kfree(sndbuf_desc);
+			/* if send buffer allocation has failed,
+			 * try a smaller one
+			 */
+			continue;
+		}
+		rc = smc_ib_buf_map(lgr->lnk[SMC_SINGLE_LINK].smcibdev,
+				    tmp_bufsize, sndbuf_desc,
+				    DMA_TO_DEVICE);
+		if (rc) {
+			kfree(sndbuf_desc->cpu_addr);
+			kfree(sndbuf_desc);
+			continue; /* if mapping failed, try smaller one */
+		}
+		sndbuf_desc->used = 1;
+		write_lock_bh(&lgr->sndbufs_lock);
+		list_add(&sndbuf_desc->list,
+			 &lgr->sndbufs[tmp_bufsize_short]);
+		write_unlock_bh(&lgr->sndbufs_lock);
+		break;
+	}
+	if (sndbuf_desc && sndbuf_desc->cpu_addr) {
+		conn->sndbuf_desc = sndbuf_desc;
+		conn->sndbuf_size = tmp_bufsize;
+		smc->sk.sk_sndbuf = tmp_bufsize * 2;
+		atomic_set(&conn->sndbuf_space, tmp_bufsize);
+		return 0;
+	} else {
+		return -ENOMEM;
+	}
+}
+
+/* create the RMB for an SMC socket (even though the SMC protocol
+ * allows more than one RMB-element per RMB, the Linux implementation
+ * uses just one RMB-element per RMB, i.e. uses an extra RMB for every
+ * connection in a link group
+ */
+int smc_rmb_create(struct smc_sock *smc)
+{
+	struct smc_connection *conn = &smc->conn;
+	struct smc_link_group *lgr = conn->lgr;
+	int tmp_bufsize, tmp_bufsize_short;
+	struct smc_buf_desc *rmb_desc;
+	int rc;
+
+	/* use socket recv buffer size (w/o overhead) as start value */
+	for (tmp_bufsize_short = smc_compress_bufsize(smc->sk.sk_rcvbuf / 2);
+	     tmp_bufsize_short >= 0; tmp_bufsize_short--) {
+		tmp_bufsize = smc_uncompress_bufsize(tmp_bufsize_short);
+		/* check for reusable rmb_slot in the link group */
+		rmb_desc = smc_rmb_get_slot(lgr, tmp_bufsize_short);
+		if (rmb_desc) {
+			memset(rmb_desc->cpu_addr, 0, tmp_bufsize);
+			break; /* found reusable slot */
+		}
+		/* try to alloc a new RMB */
+		rmb_desc = kzalloc(sizeof(*rmb_desc), GFP_KERNEL);
+		if (!rmb_desc)
+			break; /* give up with -ENOMEM */
+		rmb_desc->cpu_addr = kzalloc(tmp_bufsize,
+					     GFP_KERNEL | __GFP_NOWARN |
+					     __GFP_NOMEMALLOC |
+					     __GFP_NORETRY);
+		if (!rmb_desc->cpu_addr) {
+			kfree(rmb_desc);
+			/* if RMB allocation has failed,
+			 * try a smaller one
+			 */
+			continue;
+		}
+		rc = smc_ib_buf_map(lgr->lnk[SMC_SINGLE_LINK].smcibdev,
+				    tmp_bufsize, rmb_desc,
+				    DMA_FROM_DEVICE);
+		if (rc) {
+			kfree(rmb_desc->cpu_addr);
+			kfree(rmb_desc);
+			continue; /* if mapping failed, try smaller one */
+		}
+		rc = smc_ib_get_memory_region(lgr->lnk[SMC_SINGLE_LINK].roce_pd,
+					      IB_ACCESS_REMOTE_WRITE |
+					      IB_ACCESS_LOCAL_WRITE,
+					     &rmb_desc->mr_rx[SMC_SINGLE_LINK]);
+		if (rc) {
+			smc_ib_buf_unmap(lgr->lnk[SMC_SINGLE_LINK].smcibdev,
+					 tmp_bufsize, rmb_desc,
+					 DMA_FROM_DEVICE);
+			kfree(rmb_desc->cpu_addr);
+			kfree(rmb_desc);
+			continue;
+		}
+		rmb_desc->used = 1;
+		write_lock_bh(&lgr->rmbs_lock);
+		list_add(&rmb_desc->list,
+			 &lgr->rmbs[tmp_bufsize_short]);
+		write_unlock_bh(&lgr->rmbs_lock);
+		break;
+	}
+	if (rmb_desc && rmb_desc->cpu_addr) {
+		conn->rmb_desc = rmb_desc;
+		conn->rmbe_size = tmp_bufsize;
+		conn->rmbe_size_short = tmp_bufsize_short;
+		smc->sk.sk_rcvbuf = tmp_bufsize * 2;
+		atomic_set(&conn->bytes_to_rcv, 0);
+		conn->rmbe_update_limit = smc_rmb_wnd_update_limit(tmp_bufsize);
+		return 0;
+	} else {
+		return -ENOMEM;
+	}
+}
+
+static inline int smc_rmb_reserve_rtoken_idx(struct smc_link_group *lgr)
+{
+	int i;
+
+	for_each_clear_bit(i, lgr->rtokens_used_mask, SMC_RMBS_PER_LGR_MAX) {
+		if (!test_and_set_bit(i, lgr->rtokens_used_mask))
+			return i;
+	}
+	return -ENOSPC;
+}
+
+/* save rkey and dma_addr received from peer during clc handshake */
+int smc_rmb_rtoken_handling(struct smc_connection *conn,
+			    struct smc_clc_msg_accept_confirm *clc)
+{
+	u64 dma_addr = be64_to_cpu(clc->rmb_dma_addr);
+	struct smc_link_group *lgr = conn->lgr;
+	u32 rkey = ntohl(clc->rmb_rkey);
+	int i;
+
+	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
+		if ((lgr->rtokens[i][SMC_SINGLE_LINK].rkey == rkey) &&
+		    test_bit(i, lgr->rtokens_used_mask)) {
+			conn->rtoken_idx = i;
+			return 0;
+		}
+	}
+	conn->rtoken_idx = smc_rmb_reserve_rtoken_idx(lgr);
+	if (conn->rtoken_idx < 0)
+		return conn->rtoken_idx;
+	lgr->rtokens[conn->rtoken_idx][SMC_SINGLE_LINK].rkey = rkey;
+	lgr->rtokens[conn->rtoken_idx][SMC_SINGLE_LINK].dma_addr = dma_addr;
+	return 0;
+}

net/smc/smc_core.h

@@ -0,0 +1,181 @@
+/*
+ * Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Definitions for SMC Connections, Link Groups and Links
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#ifndef _SMC_CORE_H
+#define _SMC_CORE_H
+
+#include <linux/atomic.h>
+#include <rdma/ib_verbs.h>
+
+#include "smc.h"
+#include "smc_ib.h"
+
+#define SMC_RMBS_PER_LGR_MAX	255	/* max. # of RMBs per link group */
+
+struct smc_lgr_list {			/* list of link group definition */
+	struct list_head	list;
+	spinlock_t		lock;	/* protects list of link groups */
+};
+
+extern struct smc_lgr_list	smc_lgr_list; /* list of link groups */
+
+enum smc_lgr_role {		/* possible roles of a link group */
+	SMC_CLNT,	/* client */
+	SMC_SERV	/* server */
+};
+
+#define SMC_WR_BUF_SIZE		48	/* size of work request buffer */
+
+struct smc_wr_buf {
+	u8	raw[SMC_WR_BUF_SIZE];
+};
+
+struct smc_link {
+	struct smc_ib_device	*smcibdev;	/* ib-device */
+	u8			ibport;		/* port - values 1 | 2 */
+	struct ib_pd		*roce_pd;	/* IB protection domain,
+						 * unique for every RoCE QP
+						 */
+	struct ib_qp		*roce_qp;	/* IB queue pair */
+	struct ib_qp_attr	qp_attr;	/* IB queue pair attributes */
+
+	struct smc_wr_buf	*wr_tx_bufs;	/* WR send payload buffers */
+	struct ib_send_wr	*wr_tx_ibs;	/* WR send meta data */
+	struct ib_sge		*wr_tx_sges;	/* WR send gather meta data */
+	struct smc_wr_tx_pend	*wr_tx_pends;	/* WR send waiting for CQE */
+	/* above four vectors have wr_tx_cnt elements and use the same index */
+	dma_addr_t		wr_tx_dma_addr;	/* DMA address of wr_tx_bufs */
+	atomic_long_t		wr_tx_id;	/* seq # of last sent WR */
+	unsigned long		*wr_tx_mask;	/* bit mask of used indexes */
+	u32			wr_tx_cnt;	/* number of WR send buffers */
+	wait_queue_head_t	wr_tx_wait;	/* wait for free WR send buf */
+
+	struct smc_wr_buf	*wr_rx_bufs;	/* WR recv payload buffers */
+	struct ib_recv_wr	*wr_rx_ibs;	/* WR recv meta data */
+	struct ib_sge		*wr_rx_sges;	/* WR recv scatter meta data */
+	/* above three vectors have wr_rx_cnt elements and use the same index */
+	dma_addr_t		wr_rx_dma_addr;	/* DMA address of wr_rx_bufs */
+	u64			wr_rx_id;	/* seq # of last recv WR */
+	u32			wr_rx_cnt;	/* number of WR recv buffers */
+
+	union ib_gid		gid;		/* gid matching used vlan id */
+	u32			peer_qpn;	/* QP number of peer */
+	enum ib_mtu		path_mtu;	/* used mtu */
+	enum ib_mtu		peer_mtu;	/* mtu size of peer */
+	u32			psn_initial;	/* QP tx initial packet seqno */
+	u32			peer_psn;	/* QP rx initial packet seqno */
+	u8			peer_mac[ETH_ALEN];	/* = gid[8:10||13:15] */
+	u8			peer_gid[sizeof(union ib_gid)];	/* gid of peer*/
+	u8			link_id;	/* unique # within link group */
+	struct completion	llc_confirm;	/* wait for rx of conf link */
+	struct completion	llc_confirm_resp; /* wait 4 rx of cnf lnk rsp */
+};
+
+/* For now we just allow one parallel link per link group. The SMC protocol
+ * allows more (up to 8).
+ */
+#define SMC_LINKS_PER_LGR_MAX	1
+#define SMC_SINGLE_LINK		0
+
+#define SMC_FIRST_CONTACT	1		/* first contact to a peer */
+#define SMC_REUSE_CONTACT	0		/* follow-on contact to a peer*/
+
+/* tx/rx buffer list element for sndbufs list and rmbs list of a lgr */
+struct smc_buf_desc {
+	struct list_head	list;
+	u64			dma_addr[SMC_LINKS_PER_LGR_MAX];
+						/* mapped address of buffer */
+	void			*cpu_addr;	/* virtual address of buffer */
+	struct ib_mr		*mr_rx[SMC_LINKS_PER_LGR_MAX];
+						/* for rmb only:
+						 * rkey provided to peer
+						 */
+	u32			used;		/* currently used / unused */
+};
+
+struct smc_rtoken {				/* address/key of remote RMB */
+	u64			dma_addr;
+	u32			rkey;
+};
+
+#define SMC_LGR_ID_SIZE		4
+
+struct smc_link_group {
+	struct list_head	list;
+	enum smc_lgr_role	role;		/* client or server */
+	__be32			daddr;		/* destination ip address */
+	struct smc_link		lnk[SMC_LINKS_PER_LGR_MAX];	/* smc link */
+	char			peer_systemid[SMC_SYSTEMID_LEN];
+						/* unique system_id of peer */
+	struct rb_root		conns_all;	/* connection tree */
+	rwlock_t		conns_lock;	/* protects conns_all */
+	unsigned int		conns_num;	/* current # of connections */
+	unsigned short		vlan_id;	/* vlan id of link group */
+
+	struct list_head	sndbufs[SMC_RMBE_SIZES];/* tx buffers */
+	rwlock_t		sndbufs_lock;	/* protects tx buffers */
+	struct list_head	rmbs[SMC_RMBE_SIZES];	/* rx buffers */
+	rwlock_t		rmbs_lock;	/* protects rx buffers */
+	struct smc_rtoken	rtokens[SMC_RMBS_PER_LGR_MAX]
+				       [SMC_LINKS_PER_LGR_MAX];
+						/* remote addr/key pairs */
+	unsigned long		rtokens_used_mask[BITS_TO_LONGS(
+							SMC_RMBS_PER_LGR_MAX)];
+						/* used rtoken elements */
+
+	u8			id[SMC_LGR_ID_SIZE];	/* unique lgr id */
+	struct delayed_work	free_work;	/* delayed freeing of an lgr */
+	bool			sync_err;	/* lgr no longer fits to peer */
+};
+
+/* Find the connection associated with the given alert token in the link group.
+ * To use rbtrees we have to implement our own search core.
+ * Requires @conns_lock
+ * @token	alert token to search for
+ * @lgr		 link group to search in
+ * Returns connection associated with token if found, NULL otherwise.
+ */
+static inline struct smc_connection *smc_lgr_find_conn(
+	u32 token, struct smc_link_group *lgr)
+{
+	struct smc_connection *res = NULL;
+	struct rb_node *node;
+
+	node = lgr->conns_all.rb_node;
+	while (node) {
+		struct smc_connection *cur = rb_entry(node,
+					struct smc_connection, alert_node);
+
+		if (cur->alert_token_local > token) {
+			node = node->rb_left;
+		} else {
+			if (cur->alert_token_local < token) {
+				node = node->rb_right;
+			} else {
+				res = cur;
+				break;
+			}
+		}
+	}
+
+	return res;
+}
+
+struct smc_sock;
+struct smc_clc_msg_accept_confirm;
+
+void smc_lgr_free(struct smc_link_group *lgr);
+void smc_lgr_terminate(struct smc_link_group *lgr);
+int smc_sndbuf_create(struct smc_sock *smc);
+int smc_rmb_create(struct smc_sock *smc);
+int smc_rmb_rtoken_handling(struct smc_connection *conn,
+			    struct smc_clc_msg_accept_confirm *clc);
+
+#endif

net/smc/smc_diag.c

@@ -0,0 +1,215 @@
+/*
+ * Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ * Monitoring SMC transport protocol sockets
+ *
+ * Copyright IBM Corp. 2016
+ *
+ * Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/init.h>
+#include <linux/sock_diag.h>
+#include <linux/inet_diag.h>
+#include <linux/smc_diag.h>
+#include <net/netlink.h>
+#include <net/smc.h>
+
+#include "smc.h"
+#include "smc_core.h"
+
+static void smc_gid_be16_convert(__u8 *buf, u8 *gid_raw)
+{
+	sprintf(buf, "%04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x",
+		be16_to_cpu(((__be16 *)gid_raw)[0]),
+		be16_to_cpu(((__be16 *)gid_raw)[1]),
+		be16_to_cpu(((__be16 *)gid_raw)[2]),
+		be16_to_cpu(((__be16 *)gid_raw)[3]),
+		be16_to_cpu(((__be16 *)gid_raw)[4]),
+		be16_to_cpu(((__be16 *)gid_raw)[5]),
+		be16_to_cpu(((__be16 *)gid_raw)[6]),
+		be16_to_cpu(((__be16 *)gid_raw)[7]));
+}
+
+static void smc_diag_msg_common_fill(struct smc_diag_msg *r, struct sock *sk)
+{
+	struct smc_sock *smc = smc_sk(sk);
+
+	r->diag_family = sk->sk_family;
+	if (!smc->clcsock)
+		return;
+	r->id.idiag_sport = htons(smc->clcsock->sk->sk_num);
+	r->id.idiag_dport = smc->clcsock->sk->sk_dport;
+	r->id.idiag_if = smc->clcsock->sk->sk_bound_dev_if;
+	sock_diag_save_cookie(sk, r->id.idiag_cookie);
+	memset(&r->id.idiag_src, 0, sizeof(r->id.idiag_src));
+	memset(&r->id.idiag_dst, 0, sizeof(r->id.idiag_dst));
+	r->id.idiag_src[0] = smc->clcsock->sk->sk_rcv_saddr;
+	r->id.idiag_dst[0] = smc->clcsock->sk->sk_daddr;
+}
+
+static int smc_diag_msg_attrs_fill(struct sock *sk, struct sk_buff *skb,
+				   struct smc_diag_msg *r,
+				   struct user_namespace *user_ns)
+{
+	if (nla_put_u8(skb, SMC_DIAG_SHUTDOWN, sk->sk_shutdown))
+		return 1;
+
+	r->diag_uid = from_kuid_munged(user_ns, sock_i_uid(sk));
+	r->diag_inode = sock_i_ino(sk);
+	return 0;
+}
+
+static int __smc_diag_dump(struct sock *sk, struct sk_buff *skb,
+			   struct netlink_callback *cb,
+			   const struct smc_diag_req *req,
+			   struct nlattr *bc)
+{
+	struct smc_sock *smc = smc_sk(sk);
+	struct user_namespace *user_ns;
+	struct smc_diag_msg *r;
+	struct nlmsghdr *nlh;
+
+	nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
+			cb->nlh->nlmsg_type, sizeof(*r), NLM_F_MULTI);
+	if (!nlh)
+		return -EMSGSIZE;
+
+	r = nlmsg_data(nlh);
+	smc_diag_msg_common_fill(r, sk);
+	r->diag_state = sk->sk_state;
+	r->diag_fallback = smc->use_fallback;
+	user_ns = sk_user_ns(NETLINK_CB(cb->skb).sk);
+	if (smc_diag_msg_attrs_fill(sk, skb, r, user_ns))
+		goto errout;
+
+	if ((req->diag_ext & (1 << (SMC_DIAG_CONNINFO - 1))) && smc->conn.lgr) {
+		struct smc_connection *conn = &smc->conn;
+		struct smc_diag_conninfo cinfo = {
+			.token = conn->alert_token_local,
+			.sndbuf_size = conn->sndbuf_size,
+			.rmbe_size = conn->rmbe_size,
+			.peer_rmbe_size = conn->peer_rmbe_size,
+
+			.rx_prod.wrap = conn->local_rx_ctrl.prod.wrap,
+			.rx_prod.count = conn->local_rx_ctrl.prod.count,
+			.rx_cons.wrap = conn->local_rx_ctrl.cons.wrap,
+			.rx_cons.count = conn->local_rx_ctrl.cons.count,
+
+			.tx_prod.wrap = conn->local_tx_ctrl.prod.wrap,
+			.tx_prod.count = conn->local_tx_ctrl.prod.count,
+			.tx_cons.wrap = conn->local_tx_ctrl.cons.wrap,
+			.tx_cons.count = conn->local_tx_ctrl.cons.count,
+
+			.tx_prod_flags =
+				*(u8 *)&conn->local_tx_ctrl.prod_flags,
+			.tx_conn_state_flags =
+				*(u8 *)&conn->local_tx_ctrl.conn_state_flags,
+			.rx_prod_flags = *(u8 *)&conn->local_rx_ctrl.prod_flags,
+			.rx_conn_state_flags =
+				*(u8 *)&conn->local_rx_ctrl.conn_state_flags,
+
+			.tx_prep.wrap = conn->tx_curs_prep.wrap,
+			.tx_prep.count = conn->tx_curs_prep.count,
+			.tx_sent.wrap = conn->tx_curs_sent.wrap,
+			.tx_sent.count = conn->tx_curs_sent.count,
+			.tx_fin.wrap = conn->tx_curs_fin.wrap,
+			.tx_fin.count = conn->tx_curs_fin.count,
+		};
+
+		if (nla_put(skb, SMC_DIAG_CONNINFO, sizeof(cinfo), &cinfo) < 0)
+			goto errout;
+	}
+
+	if ((req->diag_ext & (1 << (SMC_DIAG_LGRINFO - 1))) && smc->conn.lgr) {
+		struct smc_diag_lgrinfo linfo = {
+			.role = smc->conn.lgr->role,
+			.lnk[0].ibport = smc->conn.lgr->lnk[0].ibport,
+			.lnk[0].link_id = smc->conn.lgr->lnk[0].link_id,
+		};
+
+		memcpy(linfo.lnk[0].ibname,
+		       smc->conn.lgr->lnk[0].smcibdev->ibdev->name,
+		       sizeof(smc->conn.lgr->lnk[0].smcibdev->ibdev->name));
+		smc_gid_be16_convert(linfo.lnk[0].gid,
+				     smc->conn.lgr->lnk[0].gid.raw);
+		smc_gid_be16_convert(linfo.lnk[0].peer_gid,
+				     smc->conn.lgr->lnk[0].peer_gid);
+
+		if (nla_put(skb, SMC_DIAG_LGRINFO, sizeof(linfo), &linfo) < 0)
+			goto errout;
+	}
+
+	nlmsg_end(skb, nlh);
+	return 0;
+
+errout:
+	nlmsg_cancel(skb, nlh);
+	return -EMSGSIZE;
+}
+
+static int smc_diag_dump(struct sk_buff *skb, struct netlink_callback *cb)
+{
+	struct net *net = sock_net(skb->sk);
+	struct nlattr *bc = NULL;
+	struct hlist_head *head;
+	struct sock *sk;
+	int rc = 0;
+
+	read_lock(&smc_proto.h.smc_hash->lock);
+	head = &smc_proto.h.smc_hash->ht;
+	if (hlist_empty(head))
+		goto out;
+
+	sk_for_each(sk, head) {
+		if (!net_eq(sock_net(sk), net))
+			continue;
+		rc = __smc_diag_dump(sk, skb, cb, nlmsg_data(cb->nlh), bc);
+		if (rc)
+			break;
+	}
+
+out:
+	read_unlock(&smc_proto.h.smc_hash->lock);
+	return rc;
+}
+
+static int smc_diag_handler_dump(struct sk_buff *skb, struct nlmsghdr *h)
+{
+	struct net *net = sock_net(skb->sk);
+
+	if (h->nlmsg_type == SOCK_DIAG_BY_FAMILY &&
+	    h->nlmsg_flags & NLM_F_DUMP) {
+		{
+			struct netlink_dump_control c = {
+				.dump = smc_diag_dump,
+				.min_dump_alloc = SKB_WITH_OVERHEAD(32768),
+			};
+			return netlink_dump_start(net->diag_nlsk, skb, h, &c);
+		}
+	}
+	return 0;
+}
+
+static const struct sock_diag_handler smc_diag_handler = {
+	.family = AF_SMC,
+	.dump = smc_diag_handler_dump,
+};
+
+static int __init smc_diag_init(void)
+{
+	return sock_diag_register(&smc_diag_handler);
+}
+
+static void __exit smc_diag_exit(void)
+{
+	sock_diag_unregister(&smc_diag_handler);
+}
+
+module_init(smc_diag_init);
+module_exit(smc_diag_exit);
+MODULE_LICENSE("GPL");
+MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_NETLINK, NETLINK_SOCK_DIAG, 43 /* AF_SMC */);

net/smc/smc_ib.c

@@ -0,0 +1,466 @@
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  IB infrastructure:
+ *  Establish SMC-R as an Infiniband Client to be notified about added and
+ *  removed IB devices of type RDMA.
+ *  Determine device and port characteristics for these IB devices.
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#include <linux/random.h>
+#include <linux/workqueue.h>
+#include <rdma/ib_verbs.h>
+
+#include "smc_pnet.h"
+#include "smc_ib.h"
+#include "smc_core.h"
+#include "smc_wr.h"
+#include "smc.h"
+
+#define SMC_QP_MIN_RNR_TIMER		5
+#define SMC_QP_TIMEOUT			15 /* 4096 * 2 ** timeout usec */
+#define SMC_QP_RETRY_CNT			7 /* 7: infinite */
+#define SMC_QP_RNR_RETRY			7 /* 7: infinite */
+
+struct smc_ib_devices smc_ib_devices = {	/* smc-registered ib devices */
+	.lock = __SPIN_LOCK_UNLOCKED(smc_ib_devices.lock),
+	.list = LIST_HEAD_INIT(smc_ib_devices.list),
+};
+
+#define SMC_LOCAL_SYSTEMID_RESET	"%%%%%%%"
+
+u8 local_systemid[SMC_SYSTEMID_LEN] = SMC_LOCAL_SYSTEMID_RESET;	/* unique system
+								 * identifier
+								 */
+
+int smc_ib_get_memory_region(struct ib_pd *pd, int access_flags,
+			     struct ib_mr **mr)
+{
+	int rc;
+
+	if (*mr)
+		return 0; /* already done */
+
+	/* obtain unique key -
+	 * next invocation of get_dma_mr returns a different key!
+	 */
+	*mr = pd->device->get_dma_mr(pd, access_flags);
+	rc = PTR_ERR_OR_ZERO(*mr);
+	if (IS_ERR(*mr))
+		*mr = NULL;
+	return rc;
+}
+
+static int smc_ib_modify_qp_init(struct smc_link *lnk)
+{
+	struct ib_qp_attr qp_attr;
+
+	memset(&qp_attr, 0, sizeof(qp_attr));
+	qp_attr.qp_state = IB_QPS_INIT;
+	qp_attr.pkey_index = 0;
+	qp_attr.port_num = lnk->ibport;
+	qp_attr.qp_access_flags = IB_ACCESS_LOCAL_WRITE
+				| IB_ACCESS_REMOTE_WRITE;
+	return ib_modify_qp(lnk->roce_qp, &qp_attr,
+			    IB_QP_STATE | IB_QP_PKEY_INDEX |
+			    IB_QP_ACCESS_FLAGS | IB_QP_PORT);
+}
+
+static int smc_ib_modify_qp_rtr(struct smc_link *lnk)
+{
+	enum ib_qp_attr_mask qp_attr_mask =
+		IB_QP_STATE | IB_QP_AV | IB_QP_PATH_MTU | IB_QP_DEST_QPN |
+		IB_QP_RQ_PSN | IB_QP_MAX_DEST_RD_ATOMIC | IB_QP_MIN_RNR_TIMER;
+	struct ib_qp_attr qp_attr;
+
+	memset(&qp_attr, 0, sizeof(qp_attr));
+	qp_attr.qp_state = IB_QPS_RTR;
+	qp_attr.path_mtu = min(lnk->path_mtu, lnk->peer_mtu);
+	qp_attr.ah_attr.port_num = lnk->ibport;
+	qp_attr.ah_attr.ah_flags = IB_AH_GRH;
+	qp_attr.ah_attr.grh.hop_limit = 1;
+	memcpy(&qp_attr.ah_attr.grh.dgid, lnk->peer_gid,
+	       sizeof(lnk->peer_gid));
+	memcpy(&qp_attr.ah_attr.dmac, lnk->peer_mac,
+	       sizeof(lnk->peer_mac));
+	qp_attr.dest_qp_num = lnk->peer_qpn;
+	qp_attr.rq_psn = lnk->peer_psn; /* starting receive packet seq # */
+	qp_attr.max_dest_rd_atomic = 1; /* max # of resources for incoming
+					 * requests
+					 */
+	qp_attr.min_rnr_timer = SMC_QP_MIN_RNR_TIMER;
+
+	return ib_modify_qp(lnk->roce_qp, &qp_attr, qp_attr_mask);
+}
+
+int smc_ib_modify_qp_rts(struct smc_link *lnk)
+{
+	struct ib_qp_attr qp_attr;
+
+	memset(&qp_attr, 0, sizeof(qp_attr));
+	qp_attr.qp_state = IB_QPS_RTS;
+	qp_attr.timeout = SMC_QP_TIMEOUT;	/* local ack timeout */
+	qp_attr.retry_cnt = SMC_QP_RETRY_CNT;	/* retry count */
+	qp_attr.rnr_retry = SMC_QP_RNR_RETRY;	/* RNR retries, 7=infinite */
+	qp_attr.sq_psn = lnk->psn_initial;	/* starting send packet seq # */
+	qp_attr.max_rd_atomic = 1;	/* # of outstanding RDMA reads and
+					 * atomic ops allowed
+					 */
+	return ib_modify_qp(lnk->roce_qp, &qp_attr,
+			    IB_QP_STATE | IB_QP_TIMEOUT | IB_QP_RETRY_CNT |
+			    IB_QP_SQ_PSN | IB_QP_RNR_RETRY |
+			    IB_QP_MAX_QP_RD_ATOMIC);
+}
+
+int smc_ib_modify_qp_reset(struct smc_link *lnk)
+{
+	struct ib_qp_attr qp_attr;
+
+	memset(&qp_attr, 0, sizeof(qp_attr));
+	qp_attr.qp_state = IB_QPS_RESET;
+	return ib_modify_qp(lnk->roce_qp, &qp_attr, IB_QP_STATE);
+}
+
+int smc_ib_ready_link(struct smc_link *lnk)
+{
+	struct smc_link_group *lgr =
+		container_of(lnk, struct smc_link_group, lnk[0]);
+	int rc = 0;
+
+	rc = smc_ib_modify_qp_init(lnk);
+	if (rc)
+		goto out;
+
+	rc = smc_ib_modify_qp_rtr(lnk);
+	if (rc)
+		goto out;
+	smc_wr_remember_qp_attr(lnk);
+	rc = ib_req_notify_cq(lnk->smcibdev->roce_cq_recv,
+			      IB_CQ_SOLICITED_MASK);
+	if (rc)
+		goto out;
+	rc = smc_wr_rx_post_init(lnk);
+	if (rc)
+		goto out;
+	smc_wr_remember_qp_attr(lnk);
+
+	if (lgr->role == SMC_SERV) {
+		rc = smc_ib_modify_qp_rts(lnk);
+		if (rc)
+			goto out;
+		smc_wr_remember_qp_attr(lnk);
+	}
+out:
+	return rc;
+}
+
+/* process context wrapper for might_sleep smc_ib_remember_port_attr */
+static void smc_ib_port_event_work(struct work_struct *work)
+{
+	struct smc_ib_device *smcibdev = container_of(
+		work, struct smc_ib_device, port_event_work);
+	u8 port_idx;
+
+	for_each_set_bit(port_idx, &smcibdev->port_event_mask, SMC_MAX_PORTS) {
+		smc_ib_remember_port_attr(smcibdev, port_idx + 1);
+		clear_bit(port_idx, &smcibdev->port_event_mask);
+	}
+}
+
+/* can be called in IRQ context */
+static void smc_ib_global_event_handler(struct ib_event_handler *handler,
+					struct ib_event *ibevent)
+{
+	struct smc_ib_device *smcibdev;
+	u8 port_idx;
+
+	smcibdev = container_of(handler, struct smc_ib_device, event_handler);
+	if (!smc_pnet_find_ib(smcibdev->ibdev->name))
+		return;
+
+	switch (ibevent->event) {
+	case IB_EVENT_PORT_ERR:
+		port_idx = ibevent->element.port_num - 1;
+		set_bit(port_idx, &smcibdev->port_event_mask);
+		schedule_work(&smcibdev->port_event_work);
+		/* fall through */
+	case IB_EVENT_DEVICE_FATAL:
+		/* tbd in follow-on patch:
+		 * abnormal close of corresponding connections
+		 */
+		break;
+	case IB_EVENT_PORT_ACTIVE:
+		port_idx = ibevent->element.port_num - 1;
+		set_bit(port_idx, &smcibdev->port_event_mask);
+		schedule_work(&smcibdev->port_event_work);
+		break;
+	default:
+		break;
+	}
+}
+
+void smc_ib_dealloc_protection_domain(struct smc_link *lnk)
+{
+	ib_dealloc_pd(lnk->roce_pd);
+	lnk->roce_pd = NULL;
+}
+
+int smc_ib_create_protection_domain(struct smc_link *lnk)
+{
+	int rc;
+
+	lnk->roce_pd = ib_alloc_pd(lnk->smcibdev->ibdev, 0);
+	rc = PTR_ERR_OR_ZERO(lnk->roce_pd);
+	if (IS_ERR(lnk->roce_pd))
+		lnk->roce_pd = NULL;
+	return rc;
+}
+
+static void smc_ib_qp_event_handler(struct ib_event *ibevent, void *priv)
+{
+	switch (ibevent->event) {
+	case IB_EVENT_DEVICE_FATAL:
+	case IB_EVENT_GID_CHANGE:
+	case IB_EVENT_PORT_ERR:
+	case IB_EVENT_QP_ACCESS_ERR:
+		/* tbd in follow-on patch:
+		 * abnormal close of corresponding connections
+		 */
+		break;
+	default:
+		break;
+	}
+}
+
+void smc_ib_destroy_queue_pair(struct smc_link *lnk)
+{
+	ib_destroy_qp(lnk->roce_qp);
+	lnk->roce_qp = NULL;
+}
+
+/* create a queue pair within the protection domain for a link */
+int smc_ib_create_queue_pair(struct smc_link *lnk)
+{
+	struct ib_qp_init_attr qp_attr = {
+		.event_handler = smc_ib_qp_event_handler,
+		.qp_context = lnk,
+		.send_cq = lnk->smcibdev->roce_cq_send,
+		.recv_cq = lnk->smcibdev->roce_cq_recv,
+		.srq = NULL,
+		.cap = {
+			.max_send_wr = SMC_WR_BUF_CNT,
+				/* include unsolicited rdma_writes as well,
+				 * there are max. 2 RDMA_WRITE per 1 WR_SEND
+				 */
+			.max_recv_wr = SMC_WR_BUF_CNT * 3,
+			.max_send_sge = SMC_IB_MAX_SEND_SGE,
+			.max_recv_sge = 1,
+			.max_inline_data = SMC_WR_TX_SIZE,
+		},
+		.sq_sig_type = IB_SIGNAL_REQ_WR,
+		.qp_type = IB_QPT_RC,
+	};
+	int rc;
+
+	lnk->roce_qp = ib_create_qp(lnk->roce_pd, &qp_attr);
+	rc = PTR_ERR_OR_ZERO(lnk->roce_qp);
+	if (IS_ERR(lnk->roce_qp))
+		lnk->roce_qp = NULL;
+	else
+		smc_wr_remember_qp_attr(lnk);
+	return rc;
+}
+
+/* map a new TX or RX buffer to DMA */
+int smc_ib_buf_map(struct smc_ib_device *smcibdev, int buf_size,
+		   struct smc_buf_desc *buf_slot,
+		   enum dma_data_direction data_direction)
+{
+	int rc = 0;
+
+	if (buf_slot->dma_addr[SMC_SINGLE_LINK])
+		return rc; /* already mapped */
+	buf_slot->dma_addr[SMC_SINGLE_LINK] =
+		ib_dma_map_single(smcibdev->ibdev, buf_slot->cpu_addr,
+				  buf_size, data_direction);
+	if (ib_dma_mapping_error(smcibdev->ibdev,
+				 buf_slot->dma_addr[SMC_SINGLE_LINK]))
+		rc = -EIO;
+	return rc;
+}
+
+void smc_ib_buf_unmap(struct smc_ib_device *smcibdev, int buf_size,
+		      struct smc_buf_desc *buf_slot,
+		      enum dma_data_direction data_direction)
+{
+	if (!buf_slot->dma_addr[SMC_SINGLE_LINK])
+		return; /* already unmapped */
+	ib_dma_unmap_single(smcibdev->ibdev, *buf_slot->dma_addr, buf_size,
+			    data_direction);
+	buf_slot->dma_addr[SMC_SINGLE_LINK] = 0;
+}
+
+static int smc_ib_fill_gid_and_mac(struct smc_ib_device *smcibdev, u8 ibport)
+{
+	struct net_device *ndev;
+	int rc;
+
+	rc = ib_query_gid(smcibdev->ibdev, ibport, 0,
+			  &smcibdev->gid[ibport - 1], NULL);
+	/* the SMC protocol requires specification of the roce MAC address;
+	 * if net_device cannot be determined, it can be derived from gid 0
+	 */
+	ndev = smcibdev->ibdev->get_netdev(smcibdev->ibdev, ibport);
+	if (ndev) {
+		memcpy(&smcibdev->mac, ndev->dev_addr, ETH_ALEN);
+	} else if (!rc) {
+		memcpy(&smcibdev->mac[ibport - 1][0],
+		       &smcibdev->gid[ibport - 1].raw[8], 3);
+		memcpy(&smcibdev->mac[ibport - 1][3],
+		       &smcibdev->gid[ibport - 1].raw[13], 3);
+		smcibdev->mac[ibport - 1][0] &= ~0x02;
+	}
+	return rc;
+}
+
+/* Create an identifier unique for this instance of SMC-R.
+ * The MAC-address of the first active registered IB device
+ * plus a random 2-byte number is used to create this identifier.
+ * This name is delivered to the peer during connection initialization.
+ */
+static inline void smc_ib_define_local_systemid(struct smc_ib_device *smcibdev,
+						u8 ibport)
+{
+	memcpy(&local_systemid[2], &smcibdev->mac[ibport - 1],
+	       sizeof(smcibdev->mac[ibport - 1]));
+	get_random_bytes(&local_systemid[0], 2);
+}
+
+bool smc_ib_port_active(struct smc_ib_device *smcibdev, u8 ibport)
+{
+	return smcibdev->pattr[ibport - 1].state == IB_PORT_ACTIVE;
+}
+
+int smc_ib_remember_port_attr(struct smc_ib_device *smcibdev, u8 ibport)
+{
+	int rc;
+
+	memset(&smcibdev->pattr[ibport - 1], 0,
+	       sizeof(smcibdev->pattr[ibport - 1]));
+	rc = ib_query_port(smcibdev->ibdev, ibport,
+			   &smcibdev->pattr[ibport - 1]);
+	if (rc)
+		goto out;
+	rc = smc_ib_fill_gid_and_mac(smcibdev, ibport);
+	if (rc)
+		goto out;
+	if (!strncmp(local_systemid, SMC_LOCAL_SYSTEMID_RESET,
+		     sizeof(local_systemid)) &&
+	    smc_ib_port_active(smcibdev, ibport))
+		/* create unique system identifier */
+		smc_ib_define_local_systemid(smcibdev, ibport);
+out:
+	return rc;
+}
+
+long smc_ib_setup_per_ibdev(struct smc_ib_device *smcibdev)
+{
+	struct ib_cq_init_attr cqattr =	{
+		.cqe = SMC_WR_MAX_CQE, .comp_vector = 0 };
+	long rc;
+
+	smcibdev->roce_cq_send = ib_create_cq(smcibdev->ibdev,
+					      smc_wr_tx_cq_handler, NULL,
+					      smcibdev, &cqattr);
+	rc = PTR_ERR_OR_ZERO(smcibdev->roce_cq_send);
+	if (IS_ERR(smcibdev->roce_cq_send)) {
+		smcibdev->roce_cq_send = NULL;
+		return rc;
+	}
+	smcibdev->roce_cq_recv = ib_create_cq(smcibdev->ibdev,
+					      smc_wr_rx_cq_handler, NULL,
+					      smcibdev, &cqattr);
+	rc = PTR_ERR_OR_ZERO(smcibdev->roce_cq_recv);
+	if (IS_ERR(smcibdev->roce_cq_recv)) {
+		smcibdev->roce_cq_recv = NULL;
+		goto err;
+	}
+	INIT_IB_EVENT_HANDLER(&smcibdev->event_handler, smcibdev->ibdev,
+			      smc_ib_global_event_handler);
+	ib_register_event_handler(&smcibdev->event_handler);
+	smc_wr_add_dev(smcibdev);
+	smcibdev->initialized = 1;
+	return rc;
+
+err:
+	ib_destroy_cq(smcibdev->roce_cq_send);
+	return rc;
+}
+
+static void smc_ib_cleanup_per_ibdev(struct smc_ib_device *smcibdev)
+{
+	if (!smcibdev->initialized)
+		return;
+	smc_wr_remove_dev(smcibdev);
+	ib_unregister_event_handler(&smcibdev->event_handler);
+	ib_destroy_cq(smcibdev->roce_cq_recv);
+	ib_destroy_cq(smcibdev->roce_cq_send);
+}
+
+static struct ib_client smc_ib_client;
+
+/* callback function for ib_register_client() */
+static void smc_ib_add_dev(struct ib_device *ibdev)
+{
+	struct smc_ib_device *smcibdev;
+
+	if (ibdev->node_type != RDMA_NODE_IB_CA)
+		return;
+
+	smcibdev = kzalloc(sizeof(*smcibdev), GFP_KERNEL);
+	if (!smcibdev)
+		return;
+
+	smcibdev->ibdev = ibdev;
+	INIT_WORK(&smcibdev->port_event_work, smc_ib_port_event_work);
+
+	spin_lock(&smc_ib_devices.lock);
+	list_add_tail(&smcibdev->list, &smc_ib_devices.list);
+	spin_unlock(&smc_ib_devices.lock);
+	ib_set_client_data(ibdev, &smc_ib_client, smcibdev);
+}
+
+/* callback function for ib_register_client() */
+static void smc_ib_remove_dev(struct ib_device *ibdev, void *client_data)
+{
+	struct smc_ib_device *smcibdev;
+
+	smcibdev = ib_get_client_data(ibdev, &smc_ib_client);
+	ib_set_client_data(ibdev, &smc_ib_client, NULL);
+	spin_lock(&smc_ib_devices.lock);
+	list_del_init(&smcibdev->list); /* remove from smc_ib_devices */
+	spin_unlock(&smc_ib_devices.lock);
+	smc_pnet_remove_by_ibdev(smcibdev);
+	smc_ib_cleanup_per_ibdev(smcibdev);
+	kfree(smcibdev);
+}
+
+static struct ib_client smc_ib_client = {
+	.name	= "smc_ib",
+	.add	= smc_ib_add_dev,
+	.remove = smc_ib_remove_dev,
+};
+
+int __init smc_ib_register_client(void)
+{
+	return ib_register_client(&smc_ib_client);
+}
+
+void smc_ib_unregister_client(void)
+{
+	ib_unregister_client(&smc_ib_client);
+}

net/smc/smc_ib.h

@@ -0,0 +1,69 @@
+/*
+ * Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Definitions for IB environment
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Ursula Braun <Ursula Braun@linux.vnet.ibm.com>
+ */
+
+#ifndef _SMC_IB_H
+#define _SMC_IB_H
+
+#include <rdma/ib_verbs.h>
+
+#define SMC_MAX_PORTS			2	/* Max # of ports */
+#define SMC_GID_SIZE			sizeof(union ib_gid)
+
+#define SMC_IB_MAX_SEND_SGE		2
+
+struct smc_ib_devices {			/* list of smc ib devices definition */
+	struct list_head	list;
+	spinlock_t		lock;	/* protects list of smc ib devices */
+};
+
+extern struct smc_ib_devices	smc_ib_devices; /* list of smc ib devices */
+
+struct smc_ib_device {				/* ib-device infos for smc */
+	struct list_head	list;
+	struct ib_device	*ibdev;
+	struct ib_port_attr	pattr[SMC_MAX_PORTS];	/* ib dev. port attrs */
+	struct ib_event_handler	event_handler;	/* global ib_event handler */
+	struct ib_cq		*roce_cq_send;	/* send completion queue */
+	struct ib_cq		*roce_cq_recv;	/* recv completion queue */
+	struct tasklet_struct	send_tasklet;	/* called by send cq handler */
+	struct tasklet_struct	recv_tasklet;	/* called by recv cq handler */
+	char			mac[SMC_MAX_PORTS][6]; /* mac address per port*/
+	union ib_gid		gid[SMC_MAX_PORTS]; /* gid per port */
+	u8			initialized : 1; /* ib dev CQ, evthdl done */
+	struct work_struct	port_event_work;
+	unsigned long		port_event_mask;
+};
+
+struct smc_buf_desc;
+struct smc_link;
+
+int smc_ib_register_client(void) __init;
+void smc_ib_unregister_client(void);
+bool smc_ib_port_active(struct smc_ib_device *smcibdev, u8 ibport);
+int smc_ib_remember_port_attr(struct smc_ib_device *smcibdev, u8 ibport);
+int smc_ib_buf_map(struct smc_ib_device *smcibdev, int buf_size,
+		   struct smc_buf_desc *buf_slot,
+		   enum dma_data_direction data_direction);
+void smc_ib_buf_unmap(struct smc_ib_device *smcibdev, int bufsize,
+		      struct smc_buf_desc *buf_slot,
+		      enum dma_data_direction data_direction);
+void smc_ib_dealloc_protection_domain(struct smc_link *lnk);
+int smc_ib_create_protection_domain(struct smc_link *lnk);
+void smc_ib_destroy_queue_pair(struct smc_link *lnk);
+int smc_ib_create_queue_pair(struct smc_link *lnk);
+int smc_ib_get_memory_region(struct ib_pd *pd, int access_flags,
+			     struct ib_mr **mr);
+int smc_ib_ready_link(struct smc_link *lnk);
+int smc_ib_modify_qp_rts(struct smc_link *lnk);
+int smc_ib_modify_qp_reset(struct smc_link *lnk);
+long smc_ib_setup_per_ibdev(struct smc_ib_device *smcibdev);
+
+
+#endif

net/smc/smc_llc.c

@@ -0,0 +1,158 @@
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Link Layer Control (LLC)
+ *
+ *  For now, we only support the necessary "confirm link" functionality
+ *  which happens for the first RoCE link after successful CLC handshake.
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Klaus Wacker <Klaus.Wacker@de.ibm.com>
+ *              Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#include <net/tcp.h>
+#include <rdma/ib_verbs.h>
+
+#include "smc.h"
+#include "smc_core.h"
+#include "smc_clc.h"
+#include "smc_llc.h"
+
+/********************************** send *************************************/
+
+struct smc_llc_tx_pend {
+};
+
+/* handler for send/transmission completion of an LLC msg */
+static void smc_llc_tx_handler(struct smc_wr_tx_pend_priv *pend,
+			       struct smc_link *link,
+			       enum ib_wc_status wc_status)
+{
+	/* future work: handle wc_status error for recovery and failover */
+}
+
+/**
+ * smc_llc_add_pending_send() - add LLC control message to pending WQE transmits
+ * @link: Pointer to SMC link used for sending LLC control message.
+ * @wr_buf: Out variable returning pointer to work request payload buffer.
+ * @pend: Out variable returning pointer to private pending WR tracking.
+ *	  It's the context the transmit complete handler will get.
+ *
+ * Reserves and pre-fills an entry for a pending work request send/tx.
+ * Used by mid-level smc_llc_send_msg() to prepare for later actual send/tx.
+ * Can sleep due to smc_get_ctrl_buf (if not in softirq context).
+ *
+ * Return: 0 on success, otherwise an error value.
+ */
+static int smc_llc_add_pending_send(struct smc_link *link,
+				    struct smc_wr_buf **wr_buf,
+				    struct smc_wr_tx_pend_priv **pend)
+{
+	int rc;
+
+	rc = smc_wr_tx_get_free_slot(link, smc_llc_tx_handler, wr_buf, pend);
+	if (rc < 0)
+		return rc;
+	BUILD_BUG_ON_MSG(
+		sizeof(union smc_llc_msg) > SMC_WR_BUF_SIZE,
+		"must increase SMC_WR_BUF_SIZE to at least sizeof(struct smc_llc_msg)");
+	BUILD_BUG_ON_MSG(
+		sizeof(union smc_llc_msg) != SMC_WR_TX_SIZE,
+		"must adapt SMC_WR_TX_SIZE to sizeof(struct smc_llc_msg); if not all smc_wr upper layer protocols use the same message size any more, must start to set link->wr_tx_sges[i].length on each individual smc_wr_tx_send()");
+	BUILD_BUG_ON_MSG(
+		sizeof(struct smc_llc_tx_pend) > SMC_WR_TX_PEND_PRIV_SIZE,
+		"must increase SMC_WR_TX_PEND_PRIV_SIZE to at least sizeof(struct smc_llc_tx_pend)");
+	return 0;
+}
+
+/* high-level API to send LLC confirm link */
+int smc_llc_send_confirm_link(struct smc_link *link, u8 mac[],
+			      union ib_gid *gid,
+			      enum smc_llc_reqresp reqresp)
+{
+	struct smc_link_group *lgr = container_of(link, struct smc_link_group,
+						  lnk[SMC_SINGLE_LINK]);
+	struct smc_llc_msg_confirm_link *confllc;
+	struct smc_wr_tx_pend_priv *pend;
+	struct smc_wr_buf *wr_buf;
+	int rc;
+
+	rc = smc_llc_add_pending_send(link, &wr_buf, &pend);
+	if (rc)
+		return rc;
+	confllc = (struct smc_llc_msg_confirm_link *)wr_buf;
+	memset(confllc, 0, sizeof(*confllc));
+	confllc->hd.common.type = SMC_LLC_CONFIRM_LINK;
+	confllc->hd.length = sizeof(struct smc_llc_msg_confirm_link);
+	if (reqresp == SMC_LLC_RESP)
+		confllc->hd.flags |= SMC_LLC_FLAG_RESP;
+	memcpy(confllc->sender_mac, mac, ETH_ALEN);
+	memcpy(confllc->sender_gid, gid, SMC_GID_SIZE);
+	hton24(confllc->sender_qp_num, link->roce_qp->qp_num);
+	/* confllc->link_num = SMC_SINGLE_LINK; already done by memset above */
+	memcpy(confllc->link_uid, lgr->id, SMC_LGR_ID_SIZE);
+	confllc->max_links = SMC_LINKS_PER_LGR_MAX;
+	/* send llc message */
+	rc = smc_wr_tx_send(link, pend);
+	return rc;
+}
+
+/********************************* receive ***********************************/
+
+static void smc_llc_rx_confirm_link(struct smc_link *link,
+				    struct smc_llc_msg_confirm_link *llc)
+{
+	struct smc_link_group *lgr;
+
+	lgr = container_of(link, struct smc_link_group, lnk[SMC_SINGLE_LINK]);
+	if (llc->hd.flags & SMC_LLC_FLAG_RESP) {
+		if (lgr->role == SMC_SERV)
+			complete(&link->llc_confirm_resp);
+	} else {
+		if (lgr->role == SMC_CLNT) {
+			link->link_id = llc->link_num;
+			complete(&link->llc_confirm);
+		}
+	}
+}
+
+static void smc_llc_rx_handler(struct ib_wc *wc, void *buf)
+{
+	struct smc_link *link = (struct smc_link *)wc->qp->qp_context;
+	union smc_llc_msg *llc = buf;
+
+	if (wc->byte_len < sizeof(*llc))
+		return; /* short message */
+	if (llc->raw.hdr.length != sizeof(*llc))
+		return; /* invalid message */
+	if (llc->raw.hdr.common.type == SMC_LLC_CONFIRM_LINK)
+		smc_llc_rx_confirm_link(link, &llc->confirm_link);
+}
+
+/***************************** init, exit, misc ******************************/
+
+static struct smc_wr_rx_handler smc_llc_rx_handlers[] = {
+	{
+		.handler	= smc_llc_rx_handler,
+		.type		= SMC_LLC_CONFIRM_LINK
+	},
+	{
+		.handler	= NULL,
+	}
+};
+
+int __init smc_llc_init(void)
+{
+	struct smc_wr_rx_handler *handler;
+	int rc = 0;
+
+	for (handler = smc_llc_rx_handlers; handler->handler; handler++) {
+		INIT_HLIST_NODE(&handler->list);
+		rc = smc_wr_rx_register_handler(handler);
+		if (rc)
+			break;
+	}
+	return rc;
+}

net/smc/smc_llc.h

@@ -0,0 +1,63 @@
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Definitions for LLC (link layer control) message handling
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Klaus Wacker <Klaus.Wacker@de.ibm.com>
+ *              Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#ifndef SMC_LLC_H
+#define SMC_LLC_H
+
+#include "smc_wr.h"
+
+#define SMC_LLC_FLAG_RESP		0x80
+
+#define SMC_LLC_WAIT_FIRST_TIME		(5 * HZ)
+
+enum smc_llc_reqresp {
+	SMC_LLC_REQ,
+	SMC_LLC_RESP
+};
+
+enum smc_llc_msg_type {
+	SMC_LLC_CONFIRM_LINK		= 0x01,
+};
+
+#define SMC_LLC_DATA_LEN		40
+
+struct smc_llc_hdr {
+	struct smc_wr_rx_hdr common;
+	u8 length;	/* 44 */
+	u8 reserved;
+	u8 flags;
+};
+
+struct smc_llc_msg_confirm_link {	/* type 0x01 */
+	struct smc_llc_hdr hd;
+	u8 sender_mac[ETH_ALEN];
+	u8 sender_gid[SMC_GID_SIZE];
+	u8 sender_qp_num[3];
+	u8 link_num;
+	u8 link_uid[SMC_LGR_ID_SIZE];
+	u8 max_links;
+	u8 reserved[9];
+};
+
+union smc_llc_msg {
+	struct smc_llc_msg_confirm_link confirm_link;
+	struct {
+		struct smc_llc_hdr hdr;
+		u8 data[SMC_LLC_DATA_LEN];
+	} raw;
+};
+
+/* transmit */
+int smc_llc_send_confirm_link(struct smc_link *lnk, u8 mac[], union ib_gid *gid,
+			      enum smc_llc_reqresp reqresp);
+int smc_llc_init(void) __init;
+
+#endif /* SMC_LLC_H */

net/smc/smc_pnet.c

@@ -0,0 +1,534 @@
+/*
+ *  Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  Generic netlink support functions to configure an SMC-R PNET table
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Thomas Richter <tmricht@linux.vnet.ibm.com>
+ */
+
+#include <linux/module.h>
+#include <linux/list.h>
+#include <linux/ctype.h>
+#include <net/netlink.h>
+#include <net/genetlink.h>
+
+#include <uapi/linux/if.h>
+#include <uapi/linux/smc.h>
+
+#include <rdma/ib_verbs.h>
+
+#include "smc_pnet.h"
+#include "smc_ib.h"
+
+#define SMC_MAX_PNET_ID_LEN	16	/* Max. length of PNET id */
+
+static struct nla_policy smc_pnet_policy[SMC_PNETID_MAX + 1] = {
+	[SMC_PNETID_NAME] = {
+		.type = NLA_NUL_STRING,
+		.len = SMC_MAX_PNET_ID_LEN - 1
+	},
+	[SMC_PNETID_ETHNAME] = {
+		.type = NLA_NUL_STRING,
+		.len = IFNAMSIZ - 1
+	},
+	[SMC_PNETID_IBNAME] = {
+		.type = NLA_NUL_STRING,
+		.len = IB_DEVICE_NAME_MAX - 1
+	},
+	[SMC_PNETID_IBPORT] = { .type = NLA_U8 }
+};
+
+static struct genl_family smc_pnet_nl_family;
+
+/**
+ * struct smc_pnettable - SMC PNET table anchor
+ * @lock: Lock for list action
+ * @pnetlist: List of PNETIDs
+ */
+static struct smc_pnettable {
+	rwlock_t lock;
+	struct list_head pnetlist;
+} smc_pnettable = {
+	.pnetlist = LIST_HEAD_INIT(smc_pnettable.pnetlist),
+	.lock = __RW_LOCK_UNLOCKED(smc_pnettable.lock)
+};
+
+/**
+ * struct smc_pnetentry - pnet identifier name entry
+ * @list: List node.
+ * @pnet_name: Pnet identifier name
+ * @ndev: pointer to network device.
+ * @smcibdev: Pointer to IB device.
+ */
+struct smc_pnetentry {
+	struct list_head list;
+	char pnet_name[SMC_MAX_PNET_ID_LEN + 1];
+	struct net_device *ndev;
+	struct smc_ib_device *smcibdev;
+	u8 ib_port;
+};
+
+/* Check if two RDMA device entries are identical. Use device name and port
+ * number for comparison.
+ */
+static bool smc_pnet_same_ibname(struct smc_pnetentry *pnetelem, char *ibname,
+				 u8 ibport)
+{
+	return pnetelem->ib_port == ibport &&
+	       !strncmp(pnetelem->smcibdev->ibdev->name, ibname,
+			sizeof(pnetelem->smcibdev->ibdev->name));
+}
+
+/* Find a pnetid in the pnet table.
+ */
+static struct smc_pnetentry *smc_pnet_find_pnetid(char *pnet_name)
+{
+	struct smc_pnetentry *pnetelem, *found_pnetelem = NULL;
+
+	read_lock(&smc_pnettable.lock);
+	list_for_each_entry(pnetelem, &smc_pnettable.pnetlist, list) {
+		if (!strncmp(pnetelem->pnet_name, pnet_name,
+			     sizeof(pnetelem->pnet_name))) {
+			found_pnetelem = pnetelem;
+			break;
+		}
+	}
+	read_unlock(&smc_pnettable.lock);
+	return found_pnetelem;
+}
+
+/* Remove a pnetid from the pnet table.
+ */
+static int smc_pnet_remove_by_pnetid(char *pnet_name)
+{
+	struct smc_pnetentry *pnetelem, *tmp_pe;
+	int rc = -ENOENT;
+
+	write_lock(&smc_pnettable.lock);
+	list_for_each_entry_safe(pnetelem, tmp_pe, &smc_pnettable.pnetlist,
+				 list) {
+		if (!strncmp(pnetelem->pnet_name, pnet_name,
+			     sizeof(pnetelem->pnet_name))) {
+			list_del(&pnetelem->list);
+			dev_put(pnetelem->ndev);
+			kfree(pnetelem);
+			rc = 0;
+			break;
+		}
+	}
+	write_unlock(&smc_pnettable.lock);
+	return rc;
+}
+
+/* Remove a pnet entry mentioning a given network device from the pnet table.
+ */
+static int smc_pnet_remove_by_ndev(struct net_device *ndev)
+{
+	struct smc_pnetentry *pnetelem, *tmp_pe;
+	int rc = -ENOENT;
+
+	write_lock(&smc_pnettable.lock);
+	list_for_each_entry_safe(pnetelem, tmp_pe, &smc_pnettable.pnetlist,
+				 list) {
+		if (pnetelem->ndev == ndev) {
+			list_del(&pnetelem->list);
+			dev_put(pnetelem->ndev);
+			kfree(pnetelem);
+			rc = 0;
+			break;
+		}
+	}
+	write_unlock(&smc_pnettable.lock);
+	return rc;
+}
+
+/* Remove a pnet entry mentioning a given ib device from the pnet table.
+ */
+int smc_pnet_remove_by_ibdev(struct smc_ib_device *ibdev)
+{
+	struct smc_pnetentry *pnetelem, *tmp_pe;
+	int rc = -ENOENT;
+
+	write_lock(&smc_pnettable.lock);
+	list_for_each_entry_safe(pnetelem, tmp_pe, &smc_pnettable.pnetlist,
+				 list) {
+		if (pnetelem->smcibdev == ibdev) {
+			list_del(&pnetelem->list);
+			dev_put(pnetelem->ndev);
+			kfree(pnetelem);
+			rc = 0;
+			break;
+		}
+	}
+	write_unlock(&smc_pnettable.lock);
+	return rc;
+}
+
+/* Append a pnetid to the end of the pnet table if not already on this list.
+ */
+static int smc_pnet_enter(struct smc_pnetentry *new_pnetelem)
+{
+	struct smc_pnetentry *pnetelem;
+	int rc = -EEXIST;
+
+	write_lock(&smc_pnettable.lock);
+	list_for_each_entry(pnetelem, &smc_pnettable.pnetlist, list) {
+		if (!strncmp(pnetelem->pnet_name, new_pnetelem->pnet_name,
+			     sizeof(new_pnetelem->pnet_name)) ||
+		    !strncmp(pnetelem->ndev->name, new_pnetelem->ndev->name,
+			     sizeof(new_pnetelem->ndev->name)) ||
+		    smc_pnet_same_ibname(pnetelem,
+					 new_pnetelem->smcibdev->ibdev->name,
+					 new_pnetelem->ib_port))
+			goto found;
+	}
+	list_add_tail(&new_pnetelem->list, &smc_pnettable.pnetlist);
+	rc = 0;
+found:
+	write_unlock(&smc_pnettable.lock);
+	return rc;
+}
+
+/* The limit for pnetid is 16 characters.
+ * Valid characters should be (single-byte character set) a-z, A-Z, 0-9.
+ * Lower case letters are converted to upper case.
+ * Interior blanks should not be used.
+ */
+static bool smc_pnetid_valid(const char *pnet_name, char *pnetid)
+{
+	char *bf = skip_spaces(pnet_name);
+	size_t len = strlen(bf);
+	char *end = bf + len;
+
+	if (!len)
+		return false;
+	while (--end >= bf && isspace(*end))
+		;
+	if (end - bf >= SMC_MAX_PNET_ID_LEN)
+		return false;
+	while (bf <= end) {
+		if (!isalnum(*bf))
+			return false;
+		*pnetid++ = islower(*bf) ? toupper(*bf) : *bf;
+		bf++;
+	}
+	*pnetid = '\0';
+	return true;
+}
+
+/* Find an infiniband device by a given name. The device might not exist. */
+struct smc_ib_device *smc_pnet_find_ib(char *ib_name)
+{
+	struct smc_ib_device *ibdev;
+
+	spin_lock(&smc_ib_devices.lock);
+	list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
+		if (!strncmp(ibdev->ibdev->name, ib_name,
+			     sizeof(ibdev->ibdev->name))) {
+			goto out;
+		}
+	}
+	ibdev = NULL;
+out:
+	spin_unlock(&smc_ib_devices.lock);
+	return ibdev;
+}
+
+/* Parse the supplied netlink attributes and fill a pnetentry structure.
+ * For ethernet and infiniband device names verify that the devices exist.
+ */
+static int smc_pnet_fill_entry(struct net *net, struct smc_pnetentry *pnetelem,
+			       struct nlattr *tb[])
+{
+	char *string, *ibname = NULL;
+	int rc = 0;
+
+	memset(pnetelem, 0, sizeof(*pnetelem));
+	INIT_LIST_HEAD(&pnetelem->list);
+	if (tb[SMC_PNETID_NAME]) {
+		string = (char *)nla_data(tb[SMC_PNETID_NAME]);
+		if (!smc_pnetid_valid(string, pnetelem->pnet_name)) {
+			rc = -EINVAL;
+			goto error;
+		}
+	}
+	if (tb[SMC_PNETID_ETHNAME]) {
+		string = (char *)nla_data(tb[SMC_PNETID_ETHNAME]);
+		pnetelem->ndev = dev_get_by_name(net, string);
+		if (!pnetelem->ndev)
+			return -ENOENT;
+	}
+	if (tb[SMC_PNETID_IBNAME]) {
+		ibname = (char *)nla_data(tb[SMC_PNETID_IBNAME]);
+		ibname = strim(ibname);
+		pnetelem->smcibdev = smc_pnet_find_ib(ibname);
+		if (!pnetelem->smcibdev) {
+			rc = -ENOENT;
+			goto error;
+		}
+	}
+	if (tb[SMC_PNETID_IBPORT]) {
+		pnetelem->ib_port = nla_get_u8(tb[SMC_PNETID_IBPORT]);
+		if (pnetelem->ib_port > SMC_MAX_PORTS) {
+			rc = -EINVAL;
+			goto error;
+		}
+	}
+	return 0;
+
+error:
+	if (pnetelem->ndev)
+		dev_put(pnetelem->ndev);
+	return rc;
+}
+
+/* Convert an smc_pnetentry to a netlink attribute sequence */
+static int smc_pnet_set_nla(struct sk_buff *msg, struct smc_pnetentry *pnetelem)
+{
+	if (nla_put_string(msg, SMC_PNETID_NAME, pnetelem->pnet_name) ||
+	    nla_put_string(msg, SMC_PNETID_ETHNAME, pnetelem->ndev->name) ||
+	    nla_put_string(msg, SMC_PNETID_IBNAME,
+			   pnetelem->smcibdev->ibdev->name) ||
+	    nla_put_u8(msg, SMC_PNETID_IBPORT, pnetelem->ib_port))
+		return -1;
+	return 0;
+}
+
+/* Retrieve one PNETID entry */
+static int smc_pnet_get(struct sk_buff *skb, struct genl_info *info)
+{
+	struct smc_pnetentry *pnetelem;
+	struct sk_buff *msg;
+	void *hdr;
+	int rc;
+
+	pnetelem = smc_pnet_find_pnetid(
+				(char *)nla_data(info->attrs[SMC_PNETID_NAME]));
+	if (!pnetelem)
+		return -ENOENT;
+	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
+	if (!msg)
+		return -ENOMEM;
+
+	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
+			  &smc_pnet_nl_family, 0, SMC_PNETID_GET);
+	if (!hdr) {
+		rc = -EMSGSIZE;
+		goto err_out;
+	}
+
+	if (smc_pnet_set_nla(msg, pnetelem)) {
+		rc = -ENOBUFS;
+		goto err_out;
+	}
+
+	genlmsg_end(msg, hdr);
+	return genlmsg_reply(msg, info);
+
+err_out:
+	nlmsg_free(msg);
+	return rc;
+}
+
+static int smc_pnet_add(struct sk_buff *skb, struct genl_info *info)
+{
+	struct net *net = genl_info_net(info);
+	struct smc_pnetentry *pnetelem;
+	int rc;
+
+	pnetelem = kzalloc(sizeof(*pnetelem), GFP_KERNEL);
+	if (!pnetelem)
+		return -ENOMEM;
+	rc = smc_pnet_fill_entry(net, pnetelem, info->attrs);
+	if (!rc)
+		rc = smc_pnet_enter(pnetelem);
+	if (rc) {
+		kfree(pnetelem);
+		return rc;
+	}
+	rc = smc_ib_remember_port_attr(pnetelem->smcibdev, pnetelem->ib_port);
+	if (rc)
+		smc_pnet_remove_by_pnetid(pnetelem->pnet_name);
+	return rc;
+}
+
+static int smc_pnet_del(struct sk_buff *skb, struct genl_info *info)
+{
+	return smc_pnet_remove_by_pnetid(
+				(char *)nla_data(info->attrs[SMC_PNETID_NAME]));
+}
+
+static int smc_pnet_dump_start(struct netlink_callback *cb)
+{
+	cb->args[0] = 0;
+	return 0;
+}
+
+static int smc_pnet_dumpinfo(struct sk_buff *skb,
+			     u32 portid, u32 seq, u32 flags,
+			     struct smc_pnetentry *pnetelem)
+{
+	void *hdr;
+
+	hdr = genlmsg_put(skb, portid, seq, &smc_pnet_nl_family,
+			  flags, SMC_PNETID_GET);
+	if (!hdr)
+		return -ENOMEM;
+	if (smc_pnet_set_nla(skb, pnetelem) < 0) {
+		genlmsg_cancel(skb, hdr);
+		return -EMSGSIZE;
+	}
+	genlmsg_end(skb, hdr);
+	return 0;
+}
+
+static int smc_pnet_dump(struct sk_buff *skb, struct netlink_callback *cb)
+{
+	struct smc_pnetentry *pnetelem;
+	int idx = 0;
+
+	read_lock(&smc_pnettable.lock);
+	list_for_each_entry(pnetelem, &smc_pnettable.pnetlist, list) {
+		if (idx++ < cb->args[0])
+			continue;
+		if (smc_pnet_dumpinfo(skb, NETLINK_CB(cb->skb).portid,
+				      cb->nlh->nlmsg_seq, NLM_F_MULTI,
+				      pnetelem)) {
+			--idx;
+			break;
+		}
+	}
+	cb->args[0] = idx;
+	read_unlock(&smc_pnettable.lock);
+	return skb->len;
+}
+
+/* Remove and delete all pnetids from pnet table.
+ */
+static int smc_pnet_flush(struct sk_buff *skb, struct genl_info *info)
+{
+	struct smc_pnetentry *pnetelem, *tmp_pe;
+
+	write_lock(&smc_pnettable.lock);
+	list_for_each_entry_safe(pnetelem, tmp_pe, &smc_pnettable.pnetlist,
+				 list) {
+		list_del(&pnetelem->list);
+		dev_put(pnetelem->ndev);
+		kfree(pnetelem);
+	}
+	write_unlock(&smc_pnettable.lock);
+	return 0;
+}
+
+/* SMC_PNETID generic netlink operation definition */
+static const struct genl_ops smc_pnet_ops[] = {
+	{
+		.cmd = SMC_PNETID_GET,
+		.flags = GENL_ADMIN_PERM,
+		.policy = smc_pnet_policy,
+		.doit = smc_pnet_get,
+		.dumpit = smc_pnet_dump,
+		.start = smc_pnet_dump_start
+	},
+	{
+		.cmd = SMC_PNETID_ADD,
+		.flags = GENL_ADMIN_PERM,
+		.policy = smc_pnet_policy,
+		.doit = smc_pnet_add
+	},
+	{
+		.cmd = SMC_PNETID_DEL,
+		.flags = GENL_ADMIN_PERM,
+		.policy = smc_pnet_policy,
+		.doit = smc_pnet_del
+	},
+	{
+		.cmd = SMC_PNETID_FLUSH,
+		.flags = GENL_ADMIN_PERM,
+		.policy = smc_pnet_policy,
+		.doit = smc_pnet_flush
+	}
+};
+
+/* SMC_PNETID family definition */
+static struct genl_family smc_pnet_nl_family = {
+	.hdrsize = 0,
+	.name = SMCR_GENL_FAMILY_NAME,
+	.version = SMCR_GENL_FAMILY_VERSION,
+	.maxattr = SMC_PNETID_MAX,
+	.netnsok = true,
+	.module = THIS_MODULE,
+	.ops = smc_pnet_ops,
+	.n_ops =  ARRAY_SIZE(smc_pnet_ops)
+};
+
+static int smc_pnet_netdev_event(struct notifier_block *this,
+				 unsigned long event, void *ptr)
+{
+	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
+
+	switch (event) {
+	case NETDEV_REBOOT:
+	case NETDEV_UNREGISTER:
+		smc_pnet_remove_by_ndev(event_dev);
+	default:
+		break;
+	}
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block smc_netdev_notifier = {
+	.notifier_call = smc_pnet_netdev_event
+};
+
+int __init smc_pnet_init(void)
+{
+	int rc;
+
+	rc = genl_register_family(&smc_pnet_nl_family);
+	if (rc)
+		return rc;
+	rc = register_netdevice_notifier(&smc_netdev_notifier);
+	if (rc)
+		genl_unregister_family(&smc_pnet_nl_family);
+	return rc;
+}
+
+void smc_pnet_exit(void)
+{
+	smc_pnet_flush(NULL, NULL);
+	unregister_netdevice_notifier(&smc_netdev_notifier);
+	genl_unregister_family(&smc_pnet_nl_family);
+}
+
+/* PNET table analysis for a given sock:
+ * determine ib_device and port belonging to used internal TCP socket
+ * ethernet interface.
+ */
+void smc_pnet_find_roce_resource(struct sock *sk,
+				 struct smc_ib_device **smcibdev, u8 *ibport)
+{
+	struct dst_entry *dst = sk_dst_get(sk);
+	struct smc_pnetentry *pnetelem;
+
+	*smcibdev = NULL;
+	*ibport = 0;
+
+	if (!dst)
+		return;
+	if (!dst->dev)
+		goto out_rel;
+	read_lock(&smc_pnettable.lock);
+	list_for_each_entry(pnetelem, &smc_pnettable.pnetlist, list) {
+		if (dst->dev == pnetelem->ndev) {
+			*smcibdev = pnetelem->smcibdev;
+			*ibport = pnetelem->ib_port;
+			break;
+		}
+	}
+	read_unlock(&smc_pnettable.lock);
+out_rel:
+	dst_release(dst);
+}

net/smc/smc_pnet.h

@@ -0,0 +1,23 @@
+/*
+ * Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ *  PNET table queries
+ *
+ *  Copyright IBM Corp. 2016
+ *
+ *  Author(s):  Thomas Richter <tmricht@linux.vnet.ibm.com>
+ */
+
+#ifndef _SMC_PNET_H
+#define _SMC_PNET_H
+
+struct smc_ib_device;
+
+int smc_pnet_init(void) __init;
+void smc_pnet_exit(void);
+int smc_pnet_remove_by_ibdev(struct smc_ib_device *ibdev);
+struct smc_ib_device *smc_pnet_find_ib(char *ib_name);
+void smc_pnet_find_roce_resource(struct sock *sk,
+				 struct smc_ib_device **smcibdev, u8 *ibport);
+
+#endif

net/smc/smc_rx.c

@@ -0,0 +1,217 @@
+/*
+ * Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ * Manage RMBE
+ * copy new RMBE data into user space
+ *
+ * Copyright IBM Corp. 2016
+ *
+ * Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#include <linux/net.h>
+#include <linux/rcupdate.h>
+#include <net/sock.h>
+
+#include "smc.h"
+#include "smc_core.h"
+#include "smc_cdc.h"
+#include "smc_tx.h" /* smc_tx_consumer_update() */
+#include "smc_rx.h"
+
+/* callback implementation for sk.sk_data_ready()
+ * to wakeup rcvbuf consumers that blocked with smc_rx_wait_data().
+ * indirectly called by smc_cdc_msg_recv_action().
+ */
+static void smc_rx_data_ready(struct sock *sk)
+{
+	struct socket_wq *wq;
+
+	/* derived from sock_def_readable() */
+	/* called already in smc_listen_work() */
+	rcu_read_lock();
+	wq = rcu_dereference(sk->sk_wq);
+	if (skwq_has_sleeper(wq))
+		wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
+						POLLRDNORM | POLLRDBAND);
+	if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
+	    (sk->sk_state == SMC_CLOSED))
+		sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
+	else
+		sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
+	rcu_read_unlock();
+}
+
+/* blocks rcvbuf consumer until >=len bytes available or timeout or interrupted
+ *   @smc    smc socket
+ *   @timeo  pointer to max seconds to wait, pointer to value 0 for no timeout
+ * Returns:
+ * 1 if at least 1 byte available in rcvbuf or if socket error/shutdown.
+ * 0 otherwise (nothing in rcvbuf nor timeout, e.g. interrupted).
+ */
+static int smc_rx_wait_data(struct smc_sock *smc, long *timeo)
+{
+	DEFINE_WAIT_FUNC(wait, woken_wake_function);
+	struct smc_connection *conn = &smc->conn;
+	struct sock *sk = &smc->sk;
+	int rc;
+
+	if (atomic_read(&conn->bytes_to_rcv))
+		return 1;
+	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
+	add_wait_queue(sk_sleep(sk), &wait);
+	rc = sk_wait_event(sk, timeo,
+			   sk->sk_err ||
+			   sk->sk_shutdown & RCV_SHUTDOWN ||
+			   sock_flag(sk, SOCK_DONE) ||
+			   atomic_read(&conn->bytes_to_rcv) ||
+			   smc_cdc_rxed_any_close_or_senddone(conn),
+			   &wait);
+	remove_wait_queue(sk_sleep(sk), &wait);
+	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
+	return rc;
+}
+
+/* rcvbuf consumer: main API called by socket layer.
+ * called under sk lock.
+ */
+int smc_rx_recvmsg(struct smc_sock *smc, struct msghdr *msg, size_t len,
+		   int flags)
+{
+	size_t copylen, read_done = 0, read_remaining = len;
+	size_t chunk_len, chunk_off, chunk_len_sum;
+	struct smc_connection *conn = &smc->conn;
+	union smc_host_cursor cons;
+	int readable, chunk;
+	char *rcvbuf_base;
+	struct sock *sk;
+	long timeo;
+	int target;		/* Read at least these many bytes */
+	int rc;
+
+	if (unlikely(flags & MSG_ERRQUEUE))
+		return -EINVAL; /* future work for sk.sk_family == AF_SMC */
+	if (flags & MSG_OOB)
+		return -EINVAL; /* future work */
+
+	sk = &smc->sk;
+	if (sk->sk_state == SMC_LISTEN)
+		return -ENOTCONN;
+	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
+	target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
+
+	msg->msg_namelen = 0;
+	/* we currently use 1 RMBE per RMB, so RMBE == RMB base addr */
+	rcvbuf_base = conn->rmb_desc->cpu_addr;
+
+	do { /* while (read_remaining) */
+		if (read_done >= target)
+			break;
+
+		if (atomic_read(&conn->bytes_to_rcv))
+			goto copy;
+
+		if (read_done) {
+			if (sk->sk_err ||
+			    sk->sk_state == SMC_CLOSED ||
+			    (sk->sk_shutdown & RCV_SHUTDOWN) ||
+			    !timeo ||
+			    signal_pending(current) ||
+			    smc_cdc_rxed_any_close_or_senddone(conn) ||
+			    conn->local_tx_ctrl.conn_state_flags.
+			    peer_conn_abort)
+				break;
+		} else {
+			if (sock_flag(sk, SOCK_DONE))
+				break;
+			if (sk->sk_err) {
+				read_done = sock_error(sk);
+				break;
+			}
+			if (sk->sk_shutdown & RCV_SHUTDOWN ||
+			    smc_cdc_rxed_any_close_or_senddone(conn) ||
+			    conn->local_tx_ctrl.conn_state_flags.
+			    peer_conn_abort)
+				break;
+			if (sk->sk_state == SMC_CLOSED) {
+				if (!sock_flag(sk, SOCK_DONE)) {
+					/* This occurs when user tries to read
+					 * from never connected socket.
+					 */
+					read_done = -ENOTCONN;
+					break;
+				}
+				break;
+			}
+			if (signal_pending(current)) {
+				read_done = sock_intr_errno(timeo);
+				break;
+			}
+		}
+
+		if (!atomic_read(&conn->bytes_to_rcv)) {
+			smc_rx_wait_data(smc, &timeo);
+			continue;
+		}
+
+copy:
+		/* initialize variables for 1st iteration of subsequent loop */
+		/* could be just 1 byte, even after smc_rx_wait_data above */
+		readable = atomic_read(&conn->bytes_to_rcv);
+		/* not more than what user space asked for */
+		copylen = min_t(size_t, read_remaining, readable);
+		smc_curs_write(&cons,
+			       smc_curs_read(&conn->local_tx_ctrl.cons, conn),
+			       conn);
+		/* determine chunks where to read from rcvbuf */
+		/* either unwrapped case, or 1st chunk of wrapped case */
+		chunk_len = min_t(size_t,
+				  copylen, conn->rmbe_size - cons.count);
+		chunk_len_sum = chunk_len;
+		chunk_off = cons.count;
+		for (chunk = 0; chunk < 2; chunk++) {
+			if (!(flags & MSG_TRUNC)) {
+				rc = memcpy_to_msg(msg, rcvbuf_base + chunk_off,
+						   chunk_len);
+				if (rc) {
+					if (!read_done)
+						read_done = -EFAULT;
+					goto out;
+				}
+			}
+			read_remaining -= chunk_len;
+			read_done += chunk_len;
+
+			if (chunk_len_sum == copylen)
+				break; /* either on 1st or 2nd iteration */
+			/* prepare next (== 2nd) iteration */
+			chunk_len = copylen - chunk_len; /* remainder */
+			chunk_len_sum += chunk_len;
+			chunk_off = 0; /* modulo offset in recv ring buffer */
+		}
+
+		/* update cursors */
+		if (!(flags & MSG_PEEK)) {
+			smc_curs_add(conn->rmbe_size, &cons, copylen);
+			/* increased in recv tasklet smc_cdc_msg_rcv() */
+			smp_mb__before_atomic();
+			atomic_sub(copylen, &conn->bytes_to_rcv);
+			/* guarantee 0 <= bytes_to_rcv <= rmbe_size */
+			smp_mb__after_atomic();
+			smc_curs_write(&conn->local_tx_ctrl.cons,
+				       smc_curs_read(&cons, conn),
+				       conn);
+			/* send consumer cursor update if required */
+			/* similar to advertising new TCP rcv_wnd if required */
+			smc_tx_consumer_update(conn);
+		}
+	} while (read_remaining);
+out:
+	return read_done;
+}
+
+/* Initialize receive properties on connection establishment. NB: not __init! */
+void smc_rx_init(struct smc_sock *smc)
+{
+	smc->sk.sk_data_ready = smc_rx_data_ready;
+}

net/smc/smc_rx.h

@@ -0,0 +1,23 @@
+/*
+ * Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ * Manage RMBE
+ *
+ * Copyright IBM Corp. 2016
+ *
+ * Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#ifndef SMC_RX_H
+#define SMC_RX_H
+
+#include <linux/socket.h>
+#include <linux/types.h>
+
+#include "smc.h"
+
+void smc_rx_init(struct smc_sock *smc);
+int smc_rx_recvmsg(struct smc_sock *smc, struct msghdr *msg, size_t len,
+		   int flags);
+
+#endif /* SMC_RX_H */

net/smc/smc_tx.c

@@ -0,0 +1,483 @@
+/*
+ * Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ * Manage send buffer.
+ * Producer:
+ * Copy user space data into send buffer, if send buffer space available.
+ * Consumer:
+ * Trigger RDMA write into RMBE of peer and send CDC, if RMBE space available.
+ *
+ * Copyright IBM Corp. 2016
+ *
+ * Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#include <linux/net.h>
+#include <linux/rcupdate.h>
+#include <linux/workqueue.h>
+#include <net/sock.h>
+
+#include "smc.h"
+#include "smc_wr.h"
+#include "smc_cdc.h"
+#include "smc_tx.h"
+
+/***************************** sndbuf producer *******************************/
+
+/* callback implementation for sk.sk_write_space()
+ * to wakeup sndbuf producers that blocked with smc_tx_wait_memory().
+ * called under sk_socket lock.
+ */
+static void smc_tx_write_space(struct sock *sk)
+{
+	struct socket *sock = sk->sk_socket;
+	struct smc_sock *smc = smc_sk(sk);
+	struct socket_wq *wq;
+
+	/* similar to sk_stream_write_space */
+	if (atomic_read(&smc->conn.sndbuf_space) && sock) {
+		clear_bit(SOCK_NOSPACE, &sock->flags);
+		rcu_read_lock();
+		wq = rcu_dereference(sk->sk_wq);
+		if (skwq_has_sleeper(wq))
+			wake_up_interruptible_poll(&wq->wait,
+						   POLLOUT | POLLWRNORM |
+						   POLLWRBAND);
+		if (wq && wq->fasync_list && !(sk->sk_shutdown & SEND_SHUTDOWN))
+			sock_wake_async(wq, SOCK_WAKE_SPACE, POLL_OUT);
+		rcu_read_unlock();
+	}
+}
+
+/* Wakeup sndbuf producers that blocked with smc_tx_wait_memory().
+ * Cf. tcp_data_snd_check()=>tcp_check_space()=>tcp_new_space().
+ */
+void smc_tx_sndbuf_nonfull(struct smc_sock *smc)
+{
+	if (smc->sk.sk_socket &&
+	    test_bit(SOCK_NOSPACE, &smc->sk.sk_socket->flags))
+		smc->sk.sk_write_space(&smc->sk);
+}
+
+/* blocks sndbuf producer until at least one byte of free space available */
+static int smc_tx_wait_memory(struct smc_sock *smc, int flags)
+{
+	DEFINE_WAIT_FUNC(wait, woken_wake_function);
+	struct smc_connection *conn = &smc->conn;
+	struct sock *sk = &smc->sk;
+	bool noblock;
+	long timeo;
+	int rc = 0;
+
+	/* similar to sk_stream_wait_memory */
+	timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
+	noblock = timeo ? false : true;
+	add_wait_queue(sk_sleep(sk), &wait);
+	while (1) {
+		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
+		if (sk->sk_err ||
+		    (sk->sk_shutdown & SEND_SHUTDOWN) ||
+		    conn->local_tx_ctrl.conn_state_flags.peer_done_writing) {
+			rc = -EPIPE;
+			break;
+		}
+		if (conn->local_rx_ctrl.conn_state_flags.peer_conn_abort) {
+			rc = -ECONNRESET;
+			break;
+		}
+		if (!timeo) {
+			if (noblock)
+				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
+			rc = -EAGAIN;
+			break;
+		}
+		if (signal_pending(current)) {
+			rc = sock_intr_errno(timeo);
+			break;
+		}
+		sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
+		if (atomic_read(&conn->sndbuf_space))
+			break; /* at least 1 byte of free space available */
+		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
+		sk->sk_write_pending++;
+		sk_wait_event(sk, &timeo,
+			      sk->sk_err ||
+			      (sk->sk_shutdown & SEND_SHUTDOWN) ||
+			      smc_cdc_rxed_any_close_or_senddone(conn) ||
+			      atomic_read(&conn->sndbuf_space),
+			      &wait);
+		sk->sk_write_pending--;
+	}
+	remove_wait_queue(sk_sleep(sk), &wait);
+	return rc;
+}
+
+/* sndbuf producer: main API called by socket layer.
+ * called under sock lock.
+ */
+int smc_tx_sendmsg(struct smc_sock *smc, struct msghdr *msg, size_t len)
+{
+	size_t copylen, send_done = 0, send_remaining = len;
+	size_t chunk_len, chunk_off, chunk_len_sum;
+	struct smc_connection *conn = &smc->conn;
+	union smc_host_cursor prep;
+	struct sock *sk = &smc->sk;
+	char *sndbuf_base;
+	int tx_cnt_prep;
+	int writespace;
+	int rc, chunk;
+
+	/* This should be in poll */
+	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
+
+	if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) {
+		rc = -EPIPE;
+		goto out_err;
+	}
+
+	while (msg_data_left(msg)) {
+		if (sk->sk_state == SMC_INIT)
+			return -ENOTCONN;
+		if (smc->sk.sk_shutdown & SEND_SHUTDOWN ||
+		    (smc->sk.sk_err == ECONNABORTED) ||
+		    conn->local_tx_ctrl.conn_state_flags.peer_conn_abort)
+			return -EPIPE;
+		if (smc_cdc_rxed_any_close(conn))
+			return send_done ?: -ECONNRESET;
+
+		if (!atomic_read(&conn->sndbuf_space)) {
+			rc = smc_tx_wait_memory(smc, msg->msg_flags);
+			if (rc) {
+				if (send_done)
+					return send_done;
+				goto out_err;
+			}
+			continue;
+		}
+
+		/* initialize variables for 1st iteration of subsequent loop */
+		/* could be just 1 byte, even after smc_tx_wait_memory above */
+		writespace = atomic_read(&conn->sndbuf_space);
+		/* not more than what user space asked for */
+		copylen = min_t(size_t, send_remaining, writespace);
+		/* determine start of sndbuf */
+		sndbuf_base = conn->sndbuf_desc->cpu_addr;
+		smc_curs_write(&prep,
+			       smc_curs_read(&conn->tx_curs_prep, conn),
+			       conn);
+		tx_cnt_prep = prep.count;
+		/* determine chunks where to write into sndbuf */
+		/* either unwrapped case, or 1st chunk of wrapped case */
+		chunk_len = min_t(size_t,
+				  copylen, conn->sndbuf_size - tx_cnt_prep);
+		chunk_len_sum = chunk_len;
+		chunk_off = tx_cnt_prep;
+		for (chunk = 0; chunk < 2; chunk++) {
+			rc = memcpy_from_msg(sndbuf_base + chunk_off,
+					     msg, chunk_len);
+			if (rc) {
+				if (send_done)
+					return send_done;
+				goto out_err;
+			}
+			send_done += chunk_len;
+			send_remaining -= chunk_len;
+
+			if (chunk_len_sum == copylen)
+				break; /* either on 1st or 2nd iteration */
+			/* prepare next (== 2nd) iteration */
+			chunk_len = copylen - chunk_len; /* remainder */
+			chunk_len_sum += chunk_len;
+			chunk_off = 0; /* modulo offset in send ring buffer */
+		}
+		/* update cursors */
+		smc_curs_add(conn->sndbuf_size, &prep, copylen);
+		smc_curs_write(&conn->tx_curs_prep,
+			       smc_curs_read(&prep, conn),
+			       conn);
+		/* increased in send tasklet smc_cdc_tx_handler() */
+		smp_mb__before_atomic();
+		atomic_sub(copylen, &conn->sndbuf_space);
+		/* guarantee 0 <= sndbuf_space <= sndbuf_size */
+		smp_mb__after_atomic();
+		/* since we just produced more new data into sndbuf,
+		 * trigger sndbuf consumer: RDMA write into peer RMBE and CDC
+		 */
+		smc_tx_sndbuf_nonempty(conn);
+	} /* while (msg_data_left(msg)) */
+
+	return send_done;
+
+out_err:
+	rc = sk_stream_error(sk, msg->msg_flags, rc);
+	/* make sure we wake any epoll edge trigger waiter */
+	if (unlikely(rc == -EAGAIN))
+		sk->sk_write_space(sk);
+	return rc;
+}
+
+/***************************** sndbuf consumer *******************************/
+
+/* sndbuf consumer: actual data transfer of one target chunk with RDMA write */
+static int smc_tx_rdma_write(struct smc_connection *conn, int peer_rmbe_offset,
+			     int num_sges, struct ib_sge sges[])
+{
+	struct smc_link_group *lgr = conn->lgr;
+	struct ib_send_wr *failed_wr = NULL;
+	struct ib_rdma_wr rdma_wr;
+	struct smc_link *link;
+	int rc;
+
+	memset(&rdma_wr, 0, sizeof(rdma_wr));
+	link = &lgr->lnk[SMC_SINGLE_LINK];
+	rdma_wr.wr.wr_id = smc_wr_tx_get_next_wr_id(link);
+	rdma_wr.wr.sg_list = sges;
+	rdma_wr.wr.num_sge = num_sges;
+	rdma_wr.wr.opcode = IB_WR_RDMA_WRITE;
+	rdma_wr.remote_addr =
+		lgr->rtokens[conn->rtoken_idx][SMC_SINGLE_LINK].dma_addr +
+		/* RMBE within RMB */
+		((conn->peer_conn_idx - 1) * conn->peer_rmbe_size) +
+		/* offset within RMBE */
+		peer_rmbe_offset;
+	rdma_wr.rkey = lgr->rtokens[conn->rtoken_idx][SMC_SINGLE_LINK].rkey;
+	rc = ib_post_send(link->roce_qp, &rdma_wr.wr, &failed_wr);
+	if (rc)
+		conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
+	return rc;
+}
+
+/* sndbuf consumer */
+static inline void smc_tx_advance_cursors(struct smc_connection *conn,
+					  union smc_host_cursor *prod,
+					  union smc_host_cursor *sent,
+					  size_t len)
+{
+	smc_curs_add(conn->peer_rmbe_size, prod, len);
+	/* increased in recv tasklet smc_cdc_msg_rcv() */
+	smp_mb__before_atomic();
+	/* data in flight reduces usable snd_wnd */
+	atomic_sub(len, &conn->peer_rmbe_space);
+	/* guarantee 0 <= peer_rmbe_space <= peer_rmbe_size */
+	smp_mb__after_atomic();
+	smc_curs_add(conn->sndbuf_size, sent, len);
+}
+
+/* sndbuf consumer: prepare all necessary (src&dst) chunks of data transmit;
+ * usable snd_wnd as max transmit
+ */
+static int smc_tx_rdma_writes(struct smc_connection *conn)
+{
+	size_t src_off, src_len, dst_off, dst_len; /* current chunk values */
+	size_t len, dst_len_sum, src_len_sum, dstchunk, srcchunk;
+	union smc_host_cursor sent, prep, prod, cons;
+	struct ib_sge sges[SMC_IB_MAX_SEND_SGE];
+	struct smc_link_group *lgr = conn->lgr;
+	int to_send, rmbespace;
+	struct smc_link *link;
+	int num_sges;
+	int rc;
+
+	/* source: sndbuf */
+	smc_curs_write(&sent, smc_curs_read(&conn->tx_curs_sent, conn), conn);
+	smc_curs_write(&prep, smc_curs_read(&conn->tx_curs_prep, conn), conn);
+	/* cf. wmem_alloc - (snd_max - snd_una) */
+	to_send = smc_curs_diff(conn->sndbuf_size, &sent, &prep);
+	if (to_send <= 0)
+		return 0;
+
+	/* destination: RMBE */
+	/* cf. snd_wnd */
+	rmbespace = atomic_read(&conn->peer_rmbe_space);
+	if (rmbespace <= 0)
+		return 0;
+	smc_curs_write(&prod,
+		       smc_curs_read(&conn->local_tx_ctrl.prod, conn),
+		       conn);
+	smc_curs_write(&cons,
+		       smc_curs_read(&conn->local_rx_ctrl.cons, conn),
+		       conn);
+
+	/* if usable snd_wnd closes ask peer to advertise once it opens again */
+	conn->local_tx_ctrl.prod_flags.write_blocked = (to_send >= rmbespace);
+	/* cf. usable snd_wnd */
+	len = min(to_send, rmbespace);
+
+	/* initialize variables for first iteration of subsequent nested loop */
+	link = &lgr->lnk[SMC_SINGLE_LINK];
+	dst_off = prod.count;
+	if (prod.wrap == cons.wrap) {
+		/* the filled destination area is unwrapped,
+		 * hence the available free destination space is wrapped
+		 * and we need 2 destination chunks of sum len; start with 1st
+		 * which is limited by what's available in sndbuf
+		 */
+		dst_len = min_t(size_t,
+				conn->peer_rmbe_size - prod.count, len);
+	} else {
+		/* the filled destination area is wrapped,
+		 * hence the available free destination space is unwrapped
+		 * and we need a single destination chunk of entire len
+		 */
+		dst_len = len;
+	}
+	dst_len_sum = dst_len;
+	src_off = sent.count;
+	/* dst_len determines the maximum src_len */
+	if (sent.count + dst_len <= conn->sndbuf_size) {
+		/* unwrapped src case: single chunk of entire dst_len */
+		src_len = dst_len;
+	} else {
+		/* wrapped src case: 2 chunks of sum dst_len; start with 1st: */
+		src_len = conn->sndbuf_size - sent.count;
+	}
+	src_len_sum = src_len;
+	for (dstchunk = 0; dstchunk < 2; dstchunk++) {
+		num_sges = 0;
+		for (srcchunk = 0; srcchunk < 2; srcchunk++) {
+			sges[srcchunk].addr =
+				conn->sndbuf_desc->dma_addr[SMC_SINGLE_LINK] +
+				src_off;
+			sges[srcchunk].length = src_len;
+			sges[srcchunk].lkey = link->roce_pd->local_dma_lkey;
+			num_sges++;
+			src_off += src_len;
+			if (src_off >= conn->sndbuf_size)
+				src_off -= conn->sndbuf_size;
+						/* modulo in send ring */
+			if (src_len_sum == dst_len)
+				break; /* either on 1st or 2nd iteration */
+			/* prepare next (== 2nd) iteration */
+			src_len = dst_len - src_len; /* remainder */
+			src_len_sum += src_len;
+		}
+		rc = smc_tx_rdma_write(conn, dst_off, num_sges, sges);
+		if (rc)
+			return rc;
+		if (dst_len_sum == len)
+			break; /* either on 1st or 2nd iteration */
+		/* prepare next (== 2nd) iteration */
+		dst_off = 0; /* modulo offset in RMBE ring buffer */
+		dst_len = len - dst_len; /* remainder */
+		dst_len_sum += dst_len;
+		src_len = min_t(int,
+				dst_len, conn->sndbuf_size - sent.count);
+		src_len_sum = src_len;
+	}
+
+	smc_tx_advance_cursors(conn, &prod, &sent, len);
+	/* update connection's cursors with advanced local cursors */
+	smc_curs_write(&conn->local_tx_ctrl.prod,
+		       smc_curs_read(&prod, conn),
+		       conn);
+							/* dst: peer RMBE */
+	smc_curs_write(&conn->tx_curs_sent,
+		       smc_curs_read(&sent, conn),
+		       conn);
+							/* src: local sndbuf */
+
+	return 0;
+}
+
+/* Wakeup sndbuf consumers from any context (IRQ or process)
+ * since there is more data to transmit; usable snd_wnd as max transmit
+ */
+int smc_tx_sndbuf_nonempty(struct smc_connection *conn)
+{
+	struct smc_cdc_tx_pend *pend;
+	struct smc_wr_buf *wr_buf;
+	int rc;
+
+	spin_lock_bh(&conn->send_lock);
+	rc = smc_cdc_get_free_slot(&conn->lgr->lnk[SMC_SINGLE_LINK], &wr_buf,
+				   &pend);
+	if (rc < 0) {
+		if (rc == -EBUSY) {
+			struct smc_sock *smc =
+				container_of(conn, struct smc_sock, conn);
+
+			if (smc->sk.sk_err == ECONNABORTED) {
+				rc = sock_error(&smc->sk);
+				goto out_unlock;
+			}
+			rc = 0;
+			schedule_work(&conn->tx_work);
+		}
+		goto out_unlock;
+	}
+
+	rc = smc_tx_rdma_writes(conn);
+	if (rc) {
+		smc_wr_tx_put_slot(&conn->lgr->lnk[SMC_SINGLE_LINK],
+				   (struct smc_wr_tx_pend_priv *)pend);
+		goto out_unlock;
+	}
+
+	rc = smc_cdc_msg_send(conn, wr_buf, pend);
+
+out_unlock:
+	spin_unlock_bh(&conn->send_lock);
+	return rc;
+}
+
+/* Wakeup sndbuf consumers from process context
+ * since there is more data to transmit
+ */
+static void smc_tx_work(struct work_struct *work)
+{
+	struct smc_connection *conn = container_of(work,
+						   struct smc_connection,
+						   tx_work);
+	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
+
+	lock_sock(&smc->sk);
+	smc_tx_sndbuf_nonempty(conn);
+	release_sock(&smc->sk);
+}
+
+void smc_tx_consumer_update(struct smc_connection *conn)
+{
+	union smc_host_cursor cfed, cons;
+	struct smc_cdc_tx_pend *pend;
+	struct smc_wr_buf *wr_buf;
+	int to_confirm, rc;
+
+	smc_curs_write(&cons,
+		       smc_curs_read(&conn->local_tx_ctrl.cons, conn),
+		       conn);
+	smc_curs_write(&cfed,
+		       smc_curs_read(&conn->rx_curs_confirmed, conn),
+		       conn);
+	to_confirm = smc_curs_diff(conn->rmbe_size, &cfed, &cons);
+
+	if (conn->local_rx_ctrl.prod_flags.cons_curs_upd_req ||
+	    ((to_confirm > conn->rmbe_update_limit) &&
+	     ((to_confirm > (conn->rmbe_size / 2)) ||
+	      conn->local_rx_ctrl.prod_flags.write_blocked))) {
+		rc = smc_cdc_get_free_slot(&conn->lgr->lnk[SMC_SINGLE_LINK],
+					   &wr_buf, &pend);
+		if (!rc)
+			rc = smc_cdc_msg_send(conn, wr_buf, pend);
+		if (rc < 0) {
+			schedule_work(&conn->tx_work);
+			return;
+		}
+		smc_curs_write(&conn->rx_curs_confirmed,
+			       smc_curs_read(&conn->local_tx_ctrl.cons, conn),
+			       conn);
+		conn->local_rx_ctrl.prod_flags.cons_curs_upd_req = 0;
+	}
+	if (conn->local_rx_ctrl.prod_flags.write_blocked &&
+	    !atomic_read(&conn->bytes_to_rcv))
+		conn->local_rx_ctrl.prod_flags.write_blocked = 0;
+}
+
+/***************************** send initialize *******************************/
+
+/* Initialize send properties on connection establishment. NB: not __init! */
+void smc_tx_init(struct smc_sock *smc)
+{
+	smc->sk.sk_write_space = smc_tx_write_space;
+	INIT_WORK(&smc->conn.tx_work, smc_tx_work);
+	spin_lock_init(&smc->conn.send_lock);
+}

net/smc/smc_tx.h

@@ -0,0 +1,35 @@
+/*
+ * Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ * Manage send buffer
+ *
+ * Copyright IBM Corp. 2016
+ *
+ * Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
+ */
+
+#ifndef SMC_TX_H
+#define SMC_TX_H
+
+#include <linux/socket.h>
+#include <linux/types.h>
+
+#include "smc.h"
+#include "smc_cdc.h"
+
+static inline int smc_tx_prepared_sends(struct smc_connection *conn)
+{
+	union smc_host_cursor sent, prep;
+
+	smc_curs_write(&sent, smc_curs_read(&conn->tx_curs_sent, conn), conn);
+	smc_curs_write(&prep, smc_curs_read(&conn->tx_curs_prep, conn), conn);
+	return smc_curs_diff(conn->sndbuf_size, &sent, &prep);
+}
+
+void smc_tx_init(struct smc_sock *smc);
+int smc_tx_sendmsg(struct smc_sock *smc, struct msghdr *msg, size_t len);
+int smc_tx_sndbuf_nonempty(struct smc_connection *conn);
+void smc_tx_sndbuf_nonfull(struct smc_sock *smc);
+void smc_tx_consumer_update(struct smc_connection *conn);
+
+#endif /* SMC_TX_H */

net/smc/smc_wr.c

@@ -0,0 +1,614 @@
+/*
+ * Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ * Work Requests exploiting Infiniband API
+ *
+ * Work requests (WR) of type ib_post_send or ib_post_recv respectively
+ * are submitted to either RC SQ or RC RQ respectively
+ * (reliably connected send/receive queue)
+ * and become work queue entries (WQEs).
+ * While an SQ WR/WQE is pending, we track it until transmission completion.
+ * Through a send or receive completion queue (CQ) respectively,
+ * we get completion queue entries (CQEs) [aka work completions (WCs)].
+ * Since the CQ callback is called from IRQ context, we split work by using
+ * bottom halves implemented by tasklets.
+ *
+ * SMC uses this to exchange LLC (link layer control)
+ * and CDC (connection data control) messages.
+ *
+ * Copyright IBM Corp. 2016
+ *
+ * Author(s):  Steffen Maier <maier@linux.vnet.ibm.com>
+ */
+
+#include <linux/atomic.h>
+#include <linux/hashtable.h>
+#include <linux/wait.h>
+#include <rdma/ib_verbs.h>
+#include <asm/div64.h>
+
+#include "smc.h"
+#include "smc_wr.h"
+
+#define SMC_WR_MAX_POLL_CQE 10	/* max. # of compl. queue elements in 1 poll */
+
+#define SMC_WR_RX_HASH_BITS 4
+static DEFINE_HASHTABLE(smc_wr_rx_hash, SMC_WR_RX_HASH_BITS);
+static DEFINE_SPINLOCK(smc_wr_rx_hash_lock);
+
+struct smc_wr_tx_pend {	/* control data for a pending send request */
+	u64			wr_id;		/* work request id sent */
+	smc_wr_tx_handler	handler;
+	enum ib_wc_status	wc_status;	/* CQE status */
+	struct smc_link		*link;
+	u32			idx;
+	struct smc_wr_tx_pend_priv priv;
+};
+
+/******************************** send queue *********************************/
+
+/*------------------------------- completion --------------------------------*/
+
+static inline int smc_wr_tx_find_pending_index(struct smc_link *link, u64 wr_id)
+{
+	u32 i;
+
+	for (i = 0; i < link->wr_tx_cnt; i++) {
+		if (link->wr_tx_pends[i].wr_id == wr_id)
+			return i;
+	}
+	return link->wr_tx_cnt;
+}
+
+static inline void smc_wr_tx_process_cqe(struct ib_wc *wc)
+{
+	struct smc_wr_tx_pend pnd_snd;
+	struct smc_link *link;
+	u32 pnd_snd_idx;
+	int i;
+
+	link = wc->qp->qp_context;
+	pnd_snd_idx = smc_wr_tx_find_pending_index(link, wc->wr_id);
+	if (pnd_snd_idx == link->wr_tx_cnt)
+		return;
+	link->wr_tx_pends[pnd_snd_idx].wc_status = wc->status;
+	memcpy(&pnd_snd, &link->wr_tx_pends[pnd_snd_idx], sizeof(pnd_snd));
+	/* clear the full struct smc_wr_tx_pend including .priv */
+	memset(&link->wr_tx_pends[pnd_snd_idx], 0,
+	       sizeof(link->wr_tx_pends[pnd_snd_idx]));
+	memset(&link->wr_tx_bufs[pnd_snd_idx], 0,
+	       sizeof(link->wr_tx_bufs[pnd_snd_idx]));
+	if (!test_and_clear_bit(pnd_snd_idx, link->wr_tx_mask))
+		return;
+	if (wc->status) {
+		struct smc_link_group *lgr;
+
+		for_each_set_bit(i, link->wr_tx_mask, link->wr_tx_cnt) {
+			/* clear full struct smc_wr_tx_pend including .priv */
+			memset(&link->wr_tx_pends[i], 0,
+			       sizeof(link->wr_tx_pends[i]));
+			memset(&link->wr_tx_bufs[i], 0,
+			       sizeof(link->wr_tx_bufs[i]));
+			clear_bit(i, link->wr_tx_mask);
+		}
+		/* terminate connections of this link group abnormally */
+		lgr = container_of(link, struct smc_link_group,
+				   lnk[SMC_SINGLE_LINK]);
+		smc_lgr_terminate(lgr);
+	}
+	if (pnd_snd.handler)
+		pnd_snd.handler(&pnd_snd.priv, link, wc->status);
+	wake_up(&link->wr_tx_wait);
+}
+
+static void smc_wr_tx_tasklet_fn(unsigned long data)
+{
+	struct smc_ib_device *dev = (struct smc_ib_device *)data;
+	struct ib_wc wc[SMC_WR_MAX_POLL_CQE];
+	int i = 0, rc;
+	int polled = 0;
+
+again:
+	polled++;
+	do {
+		rc = ib_poll_cq(dev->roce_cq_send, SMC_WR_MAX_POLL_CQE, wc);
+		if (polled == 1) {
+			ib_req_notify_cq(dev->roce_cq_send,
+					 IB_CQ_NEXT_COMP |
+					 IB_CQ_REPORT_MISSED_EVENTS);
+		}
+		if (!rc)
+			break;
+		for (i = 0; i < rc; i++)
+			smc_wr_tx_process_cqe(&wc[i]);
+	} while (rc > 0);
+	if (polled == 1)
+		goto again;
+}
+
+void smc_wr_tx_cq_handler(struct ib_cq *ib_cq, void *cq_context)
+{
+	struct smc_ib_device *dev = (struct smc_ib_device *)cq_context;
+
+	tasklet_schedule(&dev->send_tasklet);
+}
+
+/*---------------------------- request submission ---------------------------*/
+
+static inline int smc_wr_tx_get_free_slot_index(struct smc_link *link, u32 *idx)
+{
+	*idx = link->wr_tx_cnt;
+	for_each_clear_bit(*idx, link->wr_tx_mask, link->wr_tx_cnt) {
+		if (!test_and_set_bit(*idx, link->wr_tx_mask))
+			return 0;
+	}
+	*idx = link->wr_tx_cnt;
+	return -EBUSY;
+}
+
+/**
+ * smc_wr_tx_get_free_slot() - returns buffer for message assembly,
+ *			and sets info for pending transmit tracking
+ * @link:		Pointer to smc_link used to later send the message.
+ * @handler:		Send completion handler function pointer.
+ * @wr_buf:		Out value returns pointer to message buffer.
+ * @wr_pend_priv:	Out value returns pointer serving as handler context.
+ *
+ * Return: 0 on success, or -errno on error.
+ */
+int smc_wr_tx_get_free_slot(struct smc_link *link,
+			    smc_wr_tx_handler handler,
+			    struct smc_wr_buf **wr_buf,
+			    struct smc_wr_tx_pend_priv **wr_pend_priv)
+{
+	struct smc_wr_tx_pend *wr_pend;
+	struct ib_send_wr *wr_ib;
+	u64 wr_id;
+	u32 idx;
+	int rc;
+
+	*wr_buf = NULL;
+	*wr_pend_priv = NULL;
+	if (in_softirq()) {
+		rc = smc_wr_tx_get_free_slot_index(link, &idx);
+		if (rc)
+			return rc;
+	} else {
+		rc = wait_event_interruptible_timeout(
+			link->wr_tx_wait,
+			(smc_wr_tx_get_free_slot_index(link, &idx) != -EBUSY),
+			SMC_WR_TX_WAIT_FREE_SLOT_TIME);
+		if (!rc) {
+			/* timeout - terminate connections */
+			struct smc_link_group *lgr;
+
+			lgr = container_of(link, struct smc_link_group,
+					   lnk[SMC_SINGLE_LINK]);
+			smc_lgr_terminate(lgr);
+			return -EPIPE;
+		}
+		if (rc == -ERESTARTSYS)
+			return -EINTR;
+		if (idx == link->wr_tx_cnt)
+			return -EPIPE;
+	}
+	wr_id = smc_wr_tx_get_next_wr_id(link);
+	wr_pend = &link->wr_tx_pends[idx];
+	wr_pend->wr_id = wr_id;
+	wr_pend->handler = handler;
+	wr_pend->link = link;
+	wr_pend->idx = idx;
+	wr_ib = &link->wr_tx_ibs[idx];
+	wr_ib->wr_id = wr_id;
+	*wr_buf = &link->wr_tx_bufs[idx];
+	*wr_pend_priv = &wr_pend->priv;
+	return 0;
+}
+
+int smc_wr_tx_put_slot(struct smc_link *link,
+		       struct smc_wr_tx_pend_priv *wr_pend_priv)
+{
+	struct smc_wr_tx_pend *pend;
+
+	pend = container_of(wr_pend_priv, struct smc_wr_tx_pend, priv);
+	if (pend->idx < link->wr_tx_cnt) {
+		/* clear the full struct smc_wr_tx_pend including .priv */
+		memset(&link->wr_tx_pends[pend->idx], 0,
+		       sizeof(link->wr_tx_pends[pend->idx]));
+		memset(&link->wr_tx_bufs[pend->idx], 0,
+		       sizeof(link->wr_tx_bufs[pend->idx]));
+		test_and_clear_bit(pend->idx, link->wr_tx_mask);
+		return 1;
+	}
+
+	return 0;
+}
+
+/* Send prepared WR slot via ib_post_send.
+ * @priv: pointer to smc_wr_tx_pend_priv identifying prepared message buffer
+ */
+int smc_wr_tx_send(struct smc_link *link, struct smc_wr_tx_pend_priv *priv)
+{
+	struct ib_send_wr *failed_wr = NULL;
+	struct smc_wr_tx_pend *pend;
+	int rc;
+
+	ib_req_notify_cq(link->smcibdev->roce_cq_send,
+			 IB_CQ_SOLICITED_MASK | IB_CQ_REPORT_MISSED_EVENTS);
+	pend = container_of(priv, struct smc_wr_tx_pend, priv);
+	rc = ib_post_send(link->roce_qp, &link->wr_tx_ibs[pend->idx],
+			  &failed_wr);
+	if (rc)
+		smc_wr_tx_put_slot(link, priv);
+	return rc;
+}
+
+void smc_wr_tx_dismiss_slots(struct smc_link *link, u8 wr_rx_hdr_type,
+			     smc_wr_tx_filter filter,
+			     smc_wr_tx_dismisser dismisser,
+			     unsigned long data)
+{
+	struct smc_wr_tx_pend_priv *tx_pend;
+	struct smc_wr_rx_hdr *wr_rx;
+	int i;
+
+	for_each_set_bit(i, link->wr_tx_mask, link->wr_tx_cnt) {
+		wr_rx = (struct smc_wr_rx_hdr *)&link->wr_rx_bufs[i];
+		if (wr_rx->type != wr_rx_hdr_type)
+			continue;
+		tx_pend = &link->wr_tx_pends[i].priv;
+		if (filter(tx_pend, data))
+			dismisser(tx_pend);
+	}
+}
+
+bool smc_wr_tx_has_pending(struct smc_link *link, u8 wr_rx_hdr_type,
+			   smc_wr_tx_filter filter, unsigned long data)
+{
+	struct smc_wr_tx_pend_priv *tx_pend;
+	struct smc_wr_rx_hdr *wr_rx;
+	int i;
+
+	for_each_set_bit(i, link->wr_tx_mask, link->wr_tx_cnt) {
+		wr_rx = (struct smc_wr_rx_hdr *)&link->wr_rx_bufs[i];
+		if (wr_rx->type != wr_rx_hdr_type)
+			continue;
+		tx_pend = &link->wr_tx_pends[i].priv;
+		if (filter(tx_pend, data))
+			return true;
+	}
+	return false;
+}
+
+/****************************** receive queue ********************************/
+
+int smc_wr_rx_register_handler(struct smc_wr_rx_handler *handler)
+{
+	struct smc_wr_rx_handler *h_iter;
+	int rc = 0;
+
+	spin_lock(&smc_wr_rx_hash_lock);
+	hash_for_each_possible(smc_wr_rx_hash, h_iter, list, handler->type) {
+		if (h_iter->type == handler->type) {
+			rc = -EEXIST;
+			goto out_unlock;
+		}
+	}
+	hash_add(smc_wr_rx_hash, &handler->list, handler->type);
+out_unlock:
+	spin_unlock(&smc_wr_rx_hash_lock);
+	return rc;
+}
+
+/* Demultiplex a received work request based on the message type to its handler.
+ * Relies on smc_wr_rx_hash having been completely filled before any IB WRs,
+ * and not being modified any more afterwards so we don't need to lock it.
+ */
+static inline void smc_wr_rx_demultiplex(struct ib_wc *wc)
+{
+	struct smc_link *link = (struct smc_link *)wc->qp->qp_context;
+	struct smc_wr_rx_handler *handler;
+	struct smc_wr_rx_hdr *wr_rx;
+	u64 temp_wr_id;
+	u32 index;
+
+	if (wc->byte_len < sizeof(*wr_rx))
+		return; /* short message */
+	temp_wr_id = wc->wr_id;
+	index = do_div(temp_wr_id, link->wr_rx_cnt);
+	wr_rx = (struct smc_wr_rx_hdr *)&link->wr_rx_bufs[index];
+	hash_for_each_possible(smc_wr_rx_hash, handler, list, wr_rx->type) {
+		if (handler->type == wr_rx->type)
+			handler->handler(wc, wr_rx);
+	}
+}
+
+static inline void smc_wr_rx_process_cqes(struct ib_wc wc[], int num)
+{
+	struct smc_link *link;
+	int i;
+
+	for (i = 0; i < num; i++) {
+		link = wc[i].qp->qp_context;
+		if (wc[i].status == IB_WC_SUCCESS) {
+			smc_wr_rx_demultiplex(&wc[i]);
+			smc_wr_rx_post(link); /* refill WR RX */
+		} else {
+			struct smc_link_group *lgr;
+
+			/* handle status errors */
+			switch (wc[i].status) {
+			case IB_WC_RETRY_EXC_ERR:
+			case IB_WC_RNR_RETRY_EXC_ERR:
+			case IB_WC_WR_FLUSH_ERR:
+				/* terminate connections of this link group
+				 * abnormally
+				 */
+				lgr = container_of(link, struct smc_link_group,
+						   lnk[SMC_SINGLE_LINK]);
+				smc_lgr_terminate(lgr);
+				break;
+			default:
+				smc_wr_rx_post(link); /* refill WR RX */
+				break;
+			}
+		}
+	}
+}
+
+static void smc_wr_rx_tasklet_fn(unsigned long data)
+{
+	struct smc_ib_device *dev = (struct smc_ib_device *)data;
+	struct ib_wc wc[SMC_WR_MAX_POLL_CQE];
+	int polled = 0;
+	int rc;
+
+again:
+	polled++;
+	do {
+		memset(&wc, 0, sizeof(wc));
+		rc = ib_poll_cq(dev->roce_cq_recv, SMC_WR_MAX_POLL_CQE, wc);
+		if (polled == 1) {
+			ib_req_notify_cq(dev->roce_cq_recv,
+					 IB_CQ_SOLICITED_MASK
+					 | IB_CQ_REPORT_MISSED_EVENTS);
+		}
+		if (!rc)
+			break;
+		smc_wr_rx_process_cqes(&wc[0], rc);
+	} while (rc > 0);
+	if (polled == 1)
+		goto again;
+}
+
+void smc_wr_rx_cq_handler(struct ib_cq *ib_cq, void *cq_context)
+{
+	struct smc_ib_device *dev = (struct smc_ib_device *)cq_context;
+
+	tasklet_schedule(&dev->recv_tasklet);
+}
+
+int smc_wr_rx_post_init(struct smc_link *link)
+{
+	u32 i;
+	int rc = 0;
+
+	for (i = 0; i < link->wr_rx_cnt; i++)
+		rc = smc_wr_rx_post(link);
+	return rc;
+}
+
+/***************************** init, exit, misc ******************************/
+
+void smc_wr_remember_qp_attr(struct smc_link *lnk)
+{
+	struct ib_qp_attr *attr = &lnk->qp_attr;
+	struct ib_qp_init_attr init_attr;
+
+	memset(attr, 0, sizeof(*attr));
+	memset(&init_attr, 0, sizeof(init_attr));
+	ib_query_qp(lnk->roce_qp, attr,
+		    IB_QP_STATE |
+		    IB_QP_CUR_STATE |
+		    IB_QP_PKEY_INDEX |
+		    IB_QP_PORT |
+		    IB_QP_QKEY |
+		    IB_QP_AV |
+		    IB_QP_PATH_MTU |
+		    IB_QP_TIMEOUT |
+		    IB_QP_RETRY_CNT |
+		    IB_QP_RNR_RETRY |
+		    IB_QP_RQ_PSN |
+		    IB_QP_ALT_PATH |
+		    IB_QP_MIN_RNR_TIMER |
+		    IB_QP_SQ_PSN |
+		    IB_QP_PATH_MIG_STATE |
+		    IB_QP_CAP |
+		    IB_QP_DEST_QPN,
+		    &init_attr);
+
+	lnk->wr_tx_cnt = min_t(size_t, SMC_WR_BUF_CNT,
+			       lnk->qp_attr.cap.max_send_wr);
+	lnk->wr_rx_cnt = min_t(size_t, SMC_WR_BUF_CNT * 3,
+			       lnk->qp_attr.cap.max_recv_wr);
+}
+
+static void smc_wr_init_sge(struct smc_link *lnk)
+{
+	u32 i;
+
+	for (i = 0; i < lnk->wr_tx_cnt; i++) {
+		lnk->wr_tx_sges[i].addr =
+			lnk->wr_tx_dma_addr + i * SMC_WR_BUF_SIZE;
+		lnk->wr_tx_sges[i].length = SMC_WR_TX_SIZE;
+		lnk->wr_tx_sges[i].lkey = lnk->roce_pd->local_dma_lkey;
+		lnk->wr_tx_ibs[i].next = NULL;
+		lnk->wr_tx_ibs[i].sg_list = &lnk->wr_tx_sges[i];
+		lnk->wr_tx_ibs[i].num_sge = 1;
+		lnk->wr_tx_ibs[i].opcode = IB_WR_SEND;
+		lnk->wr_tx_ibs[i].send_flags =
+			IB_SEND_SIGNALED | IB_SEND_SOLICITED | IB_SEND_INLINE;
+	}
+	for (i = 0; i < lnk->wr_rx_cnt; i++) {
+		lnk->wr_rx_sges[i].addr =
+			lnk->wr_rx_dma_addr + i * SMC_WR_BUF_SIZE;
+		lnk->wr_rx_sges[i].length = SMC_WR_BUF_SIZE;
+		lnk->wr_rx_sges[i].lkey = lnk->roce_pd->local_dma_lkey;
+		lnk->wr_rx_ibs[i].next = NULL;
+		lnk->wr_rx_ibs[i].sg_list = &lnk->wr_rx_sges[i];
+		lnk->wr_rx_ibs[i].num_sge = 1;
+	}
+}
+
+void smc_wr_free_link(struct smc_link *lnk)
+{
+	struct ib_device *ibdev;
+
+	memset(lnk->wr_tx_mask, 0,
+	       BITS_TO_LONGS(SMC_WR_BUF_CNT) * sizeof(*lnk->wr_tx_mask));
+
+	if (!lnk->smcibdev)
+		return;
+	ibdev = lnk->smcibdev->ibdev;
+
+	if (lnk->wr_rx_dma_addr) {
+		ib_dma_unmap_single(ibdev, lnk->wr_rx_dma_addr,
+				    SMC_WR_BUF_SIZE * lnk->wr_rx_cnt,
+				    DMA_FROM_DEVICE);
+		lnk->wr_rx_dma_addr = 0;
+	}
+	if (lnk->wr_tx_dma_addr) {
+		ib_dma_unmap_single(ibdev, lnk->wr_tx_dma_addr,
+				    SMC_WR_BUF_SIZE * lnk->wr_tx_cnt,
+				    DMA_TO_DEVICE);
+		lnk->wr_tx_dma_addr = 0;
+	}
+}
+
+void smc_wr_free_link_mem(struct smc_link *lnk)
+{
+	kfree(lnk->wr_tx_pends);
+	lnk->wr_tx_pends = NULL;
+	kfree(lnk->wr_tx_mask);
+	lnk->wr_tx_mask = NULL;
+	kfree(lnk->wr_tx_sges);
+	lnk->wr_tx_sges = NULL;
+	kfree(lnk->wr_rx_sges);
+	lnk->wr_rx_sges = NULL;
+	kfree(lnk->wr_rx_ibs);
+	lnk->wr_rx_ibs = NULL;
+	kfree(lnk->wr_tx_ibs);
+	lnk->wr_tx_ibs = NULL;
+	kfree(lnk->wr_tx_bufs);
+	lnk->wr_tx_bufs = NULL;
+	kfree(lnk->wr_rx_bufs);
+	lnk->wr_rx_bufs = NULL;
+}
+
+int smc_wr_alloc_link_mem(struct smc_link *link)
+{
+	/* allocate link related memory */
+	link->wr_tx_bufs = kcalloc(SMC_WR_BUF_CNT, SMC_WR_BUF_SIZE, GFP_KERNEL);
+	if (!link->wr_tx_bufs)
+		goto no_mem;
+	link->wr_rx_bufs = kcalloc(SMC_WR_BUF_CNT * 3, SMC_WR_BUF_SIZE,
+				   GFP_KERNEL);
+	if (!link->wr_rx_bufs)
+		goto no_mem_wr_tx_bufs;
+	link->wr_tx_ibs = kcalloc(SMC_WR_BUF_CNT, sizeof(link->wr_tx_ibs[0]),
+				  GFP_KERNEL);
+	if (!link->wr_tx_ibs)
+		goto no_mem_wr_rx_bufs;
+	link->wr_rx_ibs = kcalloc(SMC_WR_BUF_CNT * 3,
+				  sizeof(link->wr_rx_ibs[0]),
+				  GFP_KERNEL);
+	if (!link->wr_rx_ibs)
+		goto no_mem_wr_tx_ibs;
+	link->wr_tx_sges = kcalloc(SMC_WR_BUF_CNT, sizeof(link->wr_tx_sges[0]),
+				   GFP_KERNEL);
+	if (!link->wr_tx_sges)
+		goto no_mem_wr_rx_ibs;
+	link->wr_rx_sges = kcalloc(SMC_WR_BUF_CNT * 3,
+				   sizeof(link->wr_rx_sges[0]),
+				   GFP_KERNEL);
+	if (!link->wr_rx_sges)
+		goto no_mem_wr_tx_sges;
+	link->wr_tx_mask = kzalloc(
+		BITS_TO_LONGS(SMC_WR_BUF_CNT) * sizeof(*link->wr_tx_mask),
+		GFP_KERNEL);
+	if (!link->wr_tx_mask)
+		goto no_mem_wr_rx_sges;
+	link->wr_tx_pends = kcalloc(SMC_WR_BUF_CNT,
+				    sizeof(link->wr_tx_pends[0]),
+				    GFP_KERNEL);
+	if (!link->wr_tx_pends)
+		goto no_mem_wr_tx_mask;
+	return 0;
+
+no_mem_wr_tx_mask:
+	kfree(link->wr_tx_mask);
+no_mem_wr_rx_sges:
+	kfree(link->wr_rx_sges);
+no_mem_wr_tx_sges:
+	kfree(link->wr_tx_sges);
+no_mem_wr_rx_ibs:
+	kfree(link->wr_rx_ibs);
+no_mem_wr_tx_ibs:
+	kfree(link->wr_tx_ibs);
+no_mem_wr_rx_bufs:
+	kfree(link->wr_rx_bufs);
+no_mem_wr_tx_bufs:
+	kfree(link->wr_tx_bufs);
+no_mem:
+	return -ENOMEM;
+}
+
+void smc_wr_remove_dev(struct smc_ib_device *smcibdev)
+{
+	tasklet_kill(&smcibdev->recv_tasklet);
+	tasklet_kill(&smcibdev->send_tasklet);
+}
+
+void smc_wr_add_dev(struct smc_ib_device *smcibdev)
+{
+	tasklet_init(&smcibdev->recv_tasklet, smc_wr_rx_tasklet_fn,
+		     (unsigned long)smcibdev);
+	tasklet_init(&smcibdev->send_tasklet, smc_wr_tx_tasklet_fn,
+		     (unsigned long)smcibdev);
+}
+
+int smc_wr_create_link(struct smc_link *lnk)
+{
+	struct ib_device *ibdev = lnk->smcibdev->ibdev;
+	int rc = 0;
+
+	smc_wr_tx_set_wr_id(&lnk->wr_tx_id, 0);
+	lnk->wr_rx_id = 0;
+	lnk->wr_rx_dma_addr = ib_dma_map_single(
+		ibdev, lnk->wr_rx_bufs,	SMC_WR_BUF_SIZE * lnk->wr_rx_cnt,
+		DMA_FROM_DEVICE);
+	if (ib_dma_mapping_error(ibdev, lnk->wr_rx_dma_addr)) {
+		lnk->wr_rx_dma_addr = 0;
+		rc = -EIO;
+		goto out;
+	}
+	lnk->wr_tx_dma_addr = ib_dma_map_single(
+		ibdev, lnk->wr_tx_bufs,	SMC_WR_BUF_SIZE * lnk->wr_tx_cnt,
+		DMA_TO_DEVICE);
+	if (ib_dma_mapping_error(ibdev, lnk->wr_tx_dma_addr)) {
+		rc = -EIO;
+		goto dma_unmap;
+	}
+	smc_wr_init_sge(lnk);
+	memset(lnk->wr_tx_mask, 0,
+	       BITS_TO_LONGS(SMC_WR_BUF_CNT) * sizeof(*lnk->wr_tx_mask));
+	return rc;
+
+dma_unmap:
+	ib_dma_unmap_single(ibdev, lnk->wr_rx_dma_addr,
+			    SMC_WR_BUF_SIZE * lnk->wr_rx_cnt,
+			    DMA_FROM_DEVICE);
+	lnk->wr_rx_dma_addr = 0;
+out:
+	return rc;
+}

net/smc/smc_wr.h

@@ -0,0 +1,106 @@
+/*
+ * Shared Memory Communications over RDMA (SMC-R) and RoCE
+ *
+ * Work Requests exploiting Infiniband API
+ *
+ * Copyright IBM Corp. 2016
+ *
+ * Author(s):  Steffen Maier <maier@linux.vnet.ibm.com>
+ */
+
+#ifndef SMC_WR_H
+#define SMC_WR_H
+
+#include <linux/atomic.h>
+#include <rdma/ib_verbs.h>
+#include <asm/div64.h>
+
+#include "smc.h"
+#include "smc_core.h"
+
+#define SMC_WR_MAX_CQE 32768	/* max. # of completion queue elements */
+#define SMC_WR_BUF_CNT 16	/* # of ctrl buffers per link */
+
+#define SMC_WR_TX_WAIT_FREE_SLOT_TIME	(10 * HZ)
+#define SMC_WR_TX_WAIT_PENDING_TIME	(5 * HZ)
+
+#define SMC_WR_TX_SIZE 44 /* actual size of wr_send data (<=SMC_WR_BUF_SIZE) */
+
+#define SMC_WR_TX_PEND_PRIV_SIZE 32
+
+struct smc_wr_tx_pend_priv {
+	u8			priv[SMC_WR_TX_PEND_PRIV_SIZE];
+};
+
+typedef void (*smc_wr_tx_handler)(struct smc_wr_tx_pend_priv *,
+				  struct smc_link *,
+				  enum ib_wc_status);
+
+typedef bool (*smc_wr_tx_filter)(struct smc_wr_tx_pend_priv *,
+				 unsigned long);
+
+typedef void (*smc_wr_tx_dismisser)(struct smc_wr_tx_pend_priv *);
+
+struct smc_wr_rx_handler {
+	struct hlist_node	list;	/* hash table collision resolution */
+	void			(*handler)(struct ib_wc *, void *);
+	u8			type;
+};
+
+/* Only used by RDMA write WRs.
+ * All other WRs (CDC/LLC) use smc_wr_tx_send handling WR_ID implicitly
+ */
+static inline long smc_wr_tx_get_next_wr_id(struct smc_link *link)
+{
+	return atomic_long_inc_return(&link->wr_tx_id);
+}
+
+static inline void smc_wr_tx_set_wr_id(atomic_long_t *wr_tx_id, long val)
+{
+	atomic_long_set(wr_tx_id, val);
+}
+
+/* post a new receive work request to fill a completed old work request entry */
+static inline int smc_wr_rx_post(struct smc_link *link)
+{
+	struct ib_recv_wr *bad_recv_wr = NULL;
+	int rc;
+	u64 wr_id, temp_wr_id;
+	u32 index;
+
+	wr_id = ++link->wr_rx_id; /* tasklet context, thus not atomic */
+	temp_wr_id = wr_id;
+	index = do_div(temp_wr_id, link->wr_rx_cnt);
+	link->wr_rx_ibs[index].wr_id = wr_id;
+	rc = ib_post_recv(link->roce_qp, &link->wr_rx_ibs[index], &bad_recv_wr);
+	return rc;
+}
+
+int smc_wr_create_link(struct smc_link *lnk);
+int smc_wr_alloc_link_mem(struct smc_link *lnk);
+void smc_wr_free_link(struct smc_link *lnk);
+void smc_wr_free_link_mem(struct smc_link *lnk);
+void smc_wr_remember_qp_attr(struct smc_link *lnk);
+void smc_wr_remove_dev(struct smc_ib_device *smcibdev);
+void smc_wr_add_dev(struct smc_ib_device *smcibdev);
+
+int smc_wr_tx_get_free_slot(struct smc_link *link, smc_wr_tx_handler handler,
+			    struct smc_wr_buf **wr_buf,
+			    struct smc_wr_tx_pend_priv **wr_pend_priv);
+int smc_wr_tx_put_slot(struct smc_link *link,
+		       struct smc_wr_tx_pend_priv *wr_pend_priv);
+int smc_wr_tx_send(struct smc_link *link,
+		   struct smc_wr_tx_pend_priv *wr_pend_priv);
+void smc_wr_tx_cq_handler(struct ib_cq *ib_cq, void *cq_context);
+bool smc_wr_tx_has_pending(struct smc_link *link, u8 wr_rx_hdr_type,
+			   smc_wr_tx_filter filter, unsigned long data);
+void smc_wr_tx_dismiss_slots(struct smc_link *lnk, u8 wr_rx_hdr_type,
+			     smc_wr_tx_filter filter,
+			     smc_wr_tx_dismisser dismisser,
+			     unsigned long data);
+
+int smc_wr_rx_register_handler(struct smc_wr_rx_handler *handler);
+int smc_wr_rx_post_init(struct smc_link *link);
+void smc_wr_rx_cq_handler(struct ib_cq *ib_cq, void *cq_context);
+
+#endif /* SMC_WR_H */