Merge branch 'kcm'

Tom Herbert says:

====================
kcm: Kernel Connection Multiplexor (KCM)

Kernel Connection Multiplexor (KCM) is a facility that provides a
message based interface over TCP for generic application protocols.
The motivation for this is based on the observation that although
TCP is byte stream transport protocol with no concept of message
boundaries, a common use case is to implement a framed application
layer protocol running over TCP. To date, most TCP stacks offer
byte stream API for applications, which places the burden of message
delineation, message I/O operation atomicity, and load balancing
in the application. With KCM an application can efficiently send
and receive application protocol messages over TCP using a
datagram interface.

In order to delineate message in a TCP stream for receive in KCM, the
kernel implements a message parser. For this we chose to employ BPF
which is applied to the TCP stream. BPF code parses application layer
messages and returns a message length. Nearly all binary application
protocols are parsable in this manner, so KCM should be applicable
across a wide range of applications. Other than message length
determination in receive, KCM does not require any other application
specific awareness. KCM does not implement any other application
protocol semantics-- these are are provided in userspace or could be
implemented in a kernel module layered above KCM.

KCM implements an NxM multiplexor in the kernel as diagrammed below:

+------------+   +------------+   +------------+   +------------+
| KCM socket |   | KCM socket |   | KCM socket |   | KCM socket |
+------------+   +------------+   +------------+   +------------+
      |                 |               |                |
      +-----------+     |               |     +----------+
                  |     |               |     |
               +----------------------------------+
               |           Multiplexor            |
               +----------------------------------+
                 |   |           |           |  |
       +---------+   |           |           |  ------------+
       |             |           |           |              |
+----------+  +----------+  +----------+  +----------+ +----------+
|  Psock   |  |  Psock   |  |  Psock   |  |  Psock   | |  Psock   |
+----------+  +----------+  +----------+  +----------+ +----------+
      |              |           |            |             |
+----------+  +----------+  +----------+  +----------+ +----------+
| TCP sock |  | TCP sock |  | TCP sock |  | TCP sock | | TCP sock |
+----------+  +----------+  +----------+  +----------+ +----------+

The KCM sockets provide the datagram interface to applications,
Psocks are the state for each attached TCP connection (i.e. where
message delineation is performed on receive).

A description of the APIs and design can be found in the included
Documentation/networking/kcm.txt.

In this patch set:

  - Add MSG_BATCH flag. This is used in sendmsg msg_hdr flags to
    indicate that more messages will be sent on the socket. The stack
    may batch messages up if it is beneficial for transmission.
  - In sendmmsg, set MSG_BATCH in all sub messages except for the last
    one.
  - In order to allow sendmmsg to contain multiple messages with
    SOCK_SEQPAKET we allow each msg_hdr in the sendmmsg to set MSG_EOR.
  - Add KCM module
    - This supports SOCK_DGRAM and SOCK_SEQPACKET.
  - KCM documentation

v2:
  - Added splice and page operations.
  - Assemble receive messages in place on TCP socket (don't have a
    separate assembly queue.
  - Based on above, enforce maxmimum receive message to be the size
    of the recceive socket buffer.
  - Support message assembly timeout. Use the timeout value in
    sk_rcvtimeo on the TCP socket.
  - Tested some with a couple of other production applications,
    see ~5% improvement in application latency.

Testing:

Dave Watson has integrated KCM into Thrift and we intend to put these
changes into open source. Example of this is in:

https://github.com/djwatson/fbthrift/commit/
dd7e0f9cf4e80912fdb90f6cd394db24e61a14cc

Some initial KCM Thrift benchmark numbers (comment from Dave)

Thrift by default ties a single connection to a single thread.  KCM is
instead able to load balance multiple connections across multiple epoll
loops easily.

A test sending ~5k bytes of data to a kcm thrift server, dropping the
bytes on recv:

QPS     Latency / std dev Latency
  without KCM
    70336     209/123
  with KCM
    70353     191/124

A test sending a small request, then doing work in the epoll thread,
before serving more requests:

QPS     Latency / std dev Latency
without KCM
    14282     559/602
with KCM
    23192     344/234

At the high end, there's definitely some additional kernel overhead:

Cranking the pipelining way up, with lots of small requests

QPS     Latency / std dev Latency
without KCM
   1863429     127/119
with KCM
   1337713     192/241

---

So for a "realistic" workload, KCM performs pretty well (second case).
Under extreme conditions of highest tps we still have some work to do.
In its nature a multiplexor will spread work between CPUs which is
logically good for load balancing but coan conflict with the goal
promoting affinity. Batching messages on both send and receive are
the means to recoup performance.

Future support:

 - Integration with TLS (TLS-in-kernel is a separate initiative).
 - Page operations/splice support
 - Unconnected KCM sockets. Will be able to attach sockets to different
   destinations, AF_KCM addresses with be used in sendmsg and recvmsg
   to indicate destination
 - Explore more utility in performing BPF inline with a TCP data stream
   (setting SO_MARK, rxhash for messages being sent received on
   KCM sockets).
 - Performance work
   - Diagnose performance issues under high message load

FAQ (Questions posted on LWN)

Q: Why do this in the kernel?

A: Because the kernel is good at scheduling threads and steering packets
   to threads. KCM fits well into this model since it allows the unit
   of work for scheduling and steering to be the application layer
   messages themselves. KCM should be thought of as generic application
   protocol acceleration. It to the philosophy that the kernel provides
   generic and extensible interfaces.

Q: How can adding code in the path yield better performance?

A: It is true that for just sending receiving a single message there
   would be some performance loss since the code path is longer (for
   instance comparing netperf to KCM). But for real production
   applications performance takes on many dynamics. Parallelism, context
   switching, affinity, granularity of locking, and load balancing are
   all relevant. The theory of KCM is that by an application-centric
   interface, the kernel can provide better support for these
   performance characteristics.

Q: Why not use an existing message-oriented protocol such as RUDP,
   DCCP, SCTP, RDS, and others?

A: Because that would entail using a completely new transport protocol.
   Deploying a new protocol at scale is either a huge undertaking or
   fundamentally infeasible. This is true in either the Internet and in
   the data center due in a large part to protocol ossification.
   Besides, KCM we want KCM to work existing, well deployed application
   protocols that we couldn't change even if we wanted to (e.g. http/2).

   KCM simply defines a new interface method, it does not redefine any
   aspect of the transport protocol nor application protocol, nor set
   any new requirements on these. Neither does KCM attempt to implement
   any application protocol logic other than message deliniation in the
   stream. These are fundamental requirement of KCM.

Q: How does this affect TCP?

A: It doesn't, not in the slightest. The use of KCM can be one-sided,
   KCM has no effect on the wire.

Q: Why force TCP into doing something it's not designed for?

A: TCP is defined as transport protocol and there is no standard that
   says the API into TCP must be stream based sockets, or for that
   matter sockets at all (or even that TCP needs to be implemented in a
   kernel). KCM is not inconsistent with the design of TCP just because
   to makes an message based interface over TCP, if it were then every
   application protocol sending messages over TCP would also be! :-)

Q: What about the problem of a connections with very slow rate of
   incoming data? As a result your application can get storms of very
   short reads. And it actually happens a lot with connection from
   mobile devices and it is a problem for servers handling a lot of
   connections.

A: The storm of short reads will occur regardless of whether KCM is used
   or not. KCM does have one advantage in this scenario though, it will
   only wake up the application when a full message has been received,
   not for each packet that makes up part of a bigger messages. If a
   bunch of small messages are received, the application can receive
   messages in batches using recvmmsg.

Q: Why not just use DPDK, or at least provide KCM like functionality in
   DPDK?

A: DPDK, or more generally OS bypass presumably with a TCP stack in
   userland, presents a different model of load balancing than that of
   KCM (and the kernel). KCM implements load balancing of messages
   across the threads of an application, whereas DPDK load balances
   based on queues which are more static and coarse-grained since
   multiple connections are bound to queues. DPDK works best when
   processing of packets is silo'ed in a thread on the CPU processing
   a queue, and packet processing (for both the stack and application)
   is fairly uniform. KCM works well for applications where the amount
   of work to process messages varies an application work is commonly
   delegated to worker threads often on different CPUs.

   The message based interface over TCP is something that could be
   provide by a DPDK or OS bypass library.

Q: I'm not quite seeing this for HTTP. Maybe for HTTP/2, I guess, or web
   sockets?

A: Yes. KCM is most appropriate for message based protocols over TCP
   where is easy to deduce the message length (e.g. a length field)
   and the protocol implements its own message ordering semantics.
   Fortunately this encompasses many modern protocols.

Q: How is memory limited and controlled?

A: In v2 all data for messages is now kept in socket buffers, either
   those for TCP or KCM, so socket buffer limits are applicable.
   This includes receive messages assembly which is now done ont teh
   TCP socket buffer instead of a separate queue-- this has the
   consequence that the TCP socket buffer limit provides an
   enforceable maxmimum message size.

   Additionally, a timeout may be set for messages assembly. The
   value used for this is taken from sk_rcvtimeo of the TCP socket.
====================

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

Documentation/networking/kcm.txt

@@ -0,0 +1,285 @@
+Kernel Connection Mulitplexor
+-----------------------------
+
+Kernel Connection Multiplexor (KCM) is a mechanism that provides a message based
+interface over TCP for generic application protocols. With KCM an application
+can efficiently send and receive application protocol messages over TCP using
+datagram sockets.
+
+KCM implements an NxM multiplexor in the kernel as diagrammed below:
+
++------------+   +------------+   +------------+   +------------+
+| KCM socket |   | KCM socket |   | KCM socket |   | KCM socket |
++------------+   +------------+   +------------+   +------------+
+      |                 |               |                |
+      +-----------+     |               |     +----------+
+                  |     |               |     |
+               +----------------------------------+
+               |           Multiplexor            |
+               +----------------------------------+
+                 |   |           |           |  |
+       +---------+   |           |           |  ------------+
+       |             |           |           |              |
++----------+  +----------+  +----------+  +----------+ +----------+
+|  Psock   |  |  Psock   |  |  Psock   |  |  Psock   | |  Psock   |
++----------+  +----------+  +----------+  +----------+ +----------+
+      |              |           |            |             |
++----------+  +----------+  +----------+  +----------+ +----------+
+| TCP sock |  | TCP sock |  | TCP sock |  | TCP sock | | TCP sock |
++----------+  +----------+  +----------+  +----------+ +----------+
+
+KCM sockets
+-----------
+
+The KCM sockets provide the user interface to the muliplexor. All the KCM sockets
+bound to a multiplexor are considered to have equivalent function, and I/O
+operations in different sockets may be done in parallel without the need for
+synchronization between threads in userspace.
+
+Multiplexor
+-----------
+
+The multiplexor provides the message steering. In the transmit path, messages
+written on a KCM socket are sent atomically on an appropriate TCP socket.
+Similarly, in the receive path, messages are constructed on each TCP socket
+(Psock) and complete messages are steered to a KCM socket.
+
+TCP sockets & Psocks
+--------------------
+
+TCP sockets may be bound to a KCM multiplexor. A Psock structure is allocated
+for each bound TCP socket, this structure holds the state for constructing
+messages on receive as well as other connection specific information for KCM.
+
+Connected mode semantics
+------------------------
+
+Each multiplexor assumes that all attached TCP connections are to the same
+destination and can use the different connections for load balancing when
+transmitting. The normal send and recv calls (include sendmmsg and recvmmsg)
+can be used to send and receive messages from the KCM socket.
+
+Socket types
+------------
+
+KCM supports SOCK_DGRAM and SOCK_SEQPACKET socket types.
+
+Message delineation
+-------------------
+
+Messages are sent over a TCP stream with some application protocol message
+format that typically includes a header which frames the messages. The length
+of a received message can be deduced from the application protocol header
+(often just a simple length field).
+
+A TCP stream must be parsed to determine message boundaries. Berkeley Packet
+Filter (BPF) is used for this. When attaching a TCP socket to a multiplexor a
+BPF program must be specified. The program is called at the start of receiving
+a new message and is given an skbuff that contains the bytes received so far.
+It parses the message header and returns the length of the message. Given this
+information, KCM will construct the message of the stated length and deliver it
+to a KCM socket.
+
+TCP socket management
+---------------------
+
+When a TCP socket is attached to a KCM multiplexor data ready (POLLIN) and
+write space available (POLLOUT) events are handled by the multiplexor. If there
+is a state change (disconnection) or other error on a TCP socket, an error is
+posted on the TCP socket so that a POLLERR event happens and KCM discontinues
+using the socket. When the application gets the error notification for a
+TCP socket, it should unattach the socket from KCM and then handle the error
+condition (the typical response is to close the socket and create a new
+connection if necessary).
+
+KCM limits the maximum receive message size to be the size of the receive
+socket buffer on the attached TCP socket (the socket buffer size can be set by
+SO_RCVBUF). If the length of a new message reported by the BPF program is
+greater than this limit a corresponding error (EMSGSIZE) is posted on the TCP
+socket. The BPF program may also enforce a maximum messages size and report an
+error when it is exceeded.
+
+A timeout may be set for assembling messages on a receive socket. The timeout
+value is taken from the receive timeout of the attached TCP socket (this is set
+by SO_RCVTIMEO). If the timer expires before assembly is complete an error
+(ETIMEDOUT) is posted on the socket.
+
+User interface
+==============
+
+Creating a multiplexor
+----------------------
+
+A new multiplexor and initial KCM socket is created by a socket call:
+
+  socket(AF_KCM, type, protocol)
+
+  - type is either SOCK_DGRAM or SOCK_SEQPACKET
+  - protocol is KCMPROTO_CONNECTED
+
+Cloning KCM sockets
+-------------------
+
+After the first KCM socket is created using the socket call as described
+above, additional sockets for the multiplexor can be created by cloning
+a KCM socket. This is accomplished by an ioctl on a KCM socket:
+
+  /* From linux/kcm.h */
+  struct kcm_clone {
+        int fd;
+  };
+
+  struct kcm_clone info;
+
+  memset(&info, 0, sizeof(info));
+
+  err = ioctl(kcmfd, SIOCKCMCLONE, &info);
+
+  if (!err)
+    newkcmfd = info.fd;
+
+Attach transport sockets
+------------------------
+
+Attaching of transport sockets to a multiplexor is performed by calling an
+ioctl on a KCM socket for the multiplexor. e.g.:
+
+  /* From linux/kcm.h */
+  struct kcm_attach {
+        int fd;
+	int bpf_fd;
+  };
+
+  struct kcm_attach info;
+
+  memset(&info, 0, sizeof(info));
+
+  info.fd = tcpfd;
+  info.bpf_fd = bpf_prog_fd;
+
+  ioctl(kcmfd, SIOCKCMATTACH, &info);
+
+The kcm_attach structure contains:
+  fd: file descriptor for TCP socket being attached
+  bpf_prog_fd: file descriptor for compiled BPF program downloaded
+
+Unattach transport sockets
+--------------------------
+
+Unattaching a transport socket from a multiplexor is straightforward. An
+"unattach" ioctl is done with the kcm_unattach structure as the argument:
+
+  /* From linux/kcm.h */
+  struct kcm_unattach {
+        int fd;
+  };
+
+  struct kcm_unattach info;
+
+  memset(&info, 0, sizeof(info));
+
+  info.fd = cfd;
+
+  ioctl(fd, SIOCKCMUNATTACH, &info);
+
+Disabling receive on KCM socket
+-------------------------------
+
+A setsockopt is used to disable or enable receiving on a KCM socket.
+When receive is disabled, any pending messages in the socket's
+receive buffer are moved to other sockets. This feature is useful
+if an application thread knows that it will be doing a lot of
+work on a request and won't be able to service new messages for a
+while. Example use:
+
+  int val = 1;
+
+  setsockopt(kcmfd, SOL_KCM, KCM_RECV_DISABLE, &val, sizeof(val))
+
+BFP programs for message delineation
+------------------------------------
+
+BPF programs can be compiled using the BPF LLVM backend. For exmple,
+the BPF program for parsing Thrift is:
+
+  #include "bpf.h" /* for __sk_buff */
+  #include "bpf_helpers.h" /* for load_word intrinsic */
+
+  SEC("socket_kcm")
+  int bpf_prog1(struct __sk_buff *skb)
+  {
+       return load_word(skb, 0) + 4;
+  }
+
+  char _license[] SEC("license") = "GPL";
+
+Use in applications
+===================
+
+KCM accelerates application layer protocols. Specifically, it allows
+applications to use a message based interface for sending and receiving
+messages. The kernel provides necessary assurances that messages are sent
+and received atomically. This relieves much of the burden applications have
+in mapping a message based protocol onto the TCP stream. KCM also make
+application layer messages a unit of work in the kernel for the purposes of
+steerng and scheduling, which in turn allows a simpler networking model in
+multithreaded applications.
+
+Configurations
+--------------
+
+In an Nx1 configuration, KCM logically provides multiple socket handles
+to the same TCP connection. This allows parallelism between in I/O
+operations on the TCP socket (for instance copyin and copyout of data is
+parallelized). In an application, a KCM socket can be opened for each
+processing thread and inserted into the epoll (similar to how SO_REUSEPORT
+is used to allow multiple listener sockets on the same port).
+
+In a MxN configuration, multiple connections are established to the
+same destination. These are used for simple load balancing.
+
+Message batching
+----------------
+
+The primary purpose of KCM is load balancing between KCM sockets and hence
+threads in a nominal use case. Perfect load balancing, that is steering
+each received message to a different KCM socket or steering each sent
+message to a different TCP socket, can negatively impact performance
+since this doesn't allow for affinities to be established. Balancing
+based on groups, or batches of messages, can be beneficial for performance.
+
+On transmit, there are three ways an application can batch (pipeline)
+messages on a KCM socket.
+  1) Send multiple messages in a single sendmmsg.
+  2) Send a group of messages each with a sendmsg call, where all messages
+     except the last have MSG_BATCH in the flags of sendmsg call.
+  3) Create "super message" composed of multiple messages and send this
+     with a single sendmsg.
+
+On receive, the KCM module attempts to queue messages received on the
+same KCM socket during each TCP ready callback. The targeted KCM socket
+changes at each receive ready callback on the KCM socket. The application
+does not need to configure this.
+
+Error handling
+--------------
+
+An application should include a thread to monitor errors raised on
+the TCP connection. Normally, this will be done by placing each
+TCP socket attached to a KCM multiplexor in epoll set for POLLERR
+event. If an error occurs on an attached TCP socket, KCM sets an EPIPE
+on the socket thus waking up the application thread. When the application
+sees the error (which may just be a disconnect) it should unattach the
+socket from KCM and then close it. It is assumed that once an error is
+posted on the TCP socket the data stream is unrecoverable (i.e. an error
+may have occurred in in the middle of receiving a messssge).
+
+TCP connection monitoring
+-------------------------
+
+In KCM there is no means to correlate a message to the TCP socket that
+was used to send or receive the message (except in the case there is
+only one attached TCP socket). However, the application does retain
+an open file descriptor to the socket so it will be able to get statistics
+from the socket which can be used in detecting issues (such as high
+retransmissions on the socket).

include/linux/net.h

@@ -215,6 +215,7 @@ int __sock_create(struct net *net, int family, int type, int proto,
 int sock_create(int family, int type, int proto, struct socket **res);
 int sock_create_kern(struct net *net, int family, int type, int proto, struct socket **res);
 int sock_create_lite(int family, int type, int proto, struct socket **res);
+struct socket *sock_alloc(void);
 void sock_release(struct socket *sock);
 int sock_sendmsg(struct socket *sock, struct msghdr *msg);
 int sock_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,

include/linux/rculist.h

@@ -318,6 +318,27 @@ static inline void list_splice_tail_init_rcu(struct list_head *list,
 	likely(__ptr != __next) ? list_entry_rcu(__next, type, member) : NULL; \
 })
 
+/**
+ * list_next_or_null_rcu - get the first element from a list
+ * @head:	the head for the list.
+ * @ptr:        the list head to take the next element from.
+ * @type:       the type of the struct this is embedded in.
+ * @member:     the name of the list_head within the struct.
+ *
+ * Note that if the ptr is at the end of the list, NULL is returned.
+ *
+ * This primitive may safely run concurrently with the _rcu list-mutation
+ * primitives such as list_add_rcu() as long as it's guarded by rcu_read_lock().
+ */
+#define list_next_or_null_rcu(head, ptr, type, member) \
+({ \
+	struct list_head *__head = (head); \
+	struct list_head *__ptr = (ptr); \
+	struct list_head *__next = READ_ONCE(__ptr->next); \
+	likely(__next != __head) ? list_entry_rcu(__next, type, \
+						  member) : NULL; \
+})
+
 /**
  * list_for_each_entry_rcu	-	iterate over rcu list of given type
  * @pos:	the type * to use as a loop cursor.

include/linux/socket.h

@@ -200,7 +200,9 @@ struct ucred {
 #define AF_ALG		38	/* Algorithm sockets		*/
 #define AF_NFC		39	/* NFC sockets			*/
 #define AF_VSOCK	40	/* vSockets			*/
-#define AF_MAX		41	/* For now.. */
+#define AF_KCM		41	/* Kernel Connection Multiplexor*/
+
+#define AF_MAX		42	/* For now.. */
 
 /* Protocol families, same as address families. */
 #define PF_UNSPEC	AF_UNSPEC
@@ -246,6 +248,7 @@ struct ucred {
 #define PF_ALG		AF_ALG
 #define PF_NFC		AF_NFC
 #define PF_VSOCK	AF_VSOCK
+#define PF_KCM		AF_KCM
 #define PF_MAX		AF_MAX
 
 /* Maximum queue length specifiable by listen.  */
@@ -274,6 +277,7 @@ struct ucred {
 #define MSG_MORE	0x8000	/* Sender will send more */
 #define MSG_WAITFORONE	0x10000	/* recvmmsg(): block until 1+ packets avail */
 #define MSG_SENDPAGE_NOTLAST 0x20000 /* sendpage() internal : not the last page */
+#define MSG_BATCH	0x40000 /* sendmmsg(): more messages coming */
 #define MSG_EOF         MSG_FIN
 
 #define MSG_FASTOPEN	0x20000000	/* Send data in TCP SYN */
@@ -322,6 +326,7 @@ struct ucred {
 #define SOL_CAIF	278
 #define SOL_ALG		279
 #define SOL_NFC		280
+#define SOL_KCM		281
 
 /* IPX options */
 #define IPX_TYPE	1