path: root/net/rds/rds.h

#ifndef _RDS_RDS_H
#define _RDS_RDS_H

#include <net/sock.h>
#include <linux/scatterlist.h>
#include <linux/highmem.h>
#include <rdma/rdma_cm.h>
#include <linux/mutex.h>
#include <linux/rds.h>

#include "info.h"

/*
 * RDS Network protocol version
 */
#define RDS_PROTOCOL_3_0	0x0300
#define RDS_PROTOCOL_3_1	0x0301
#define RDS_PROTOCOL_VERSION	RDS_PROTOCOL_3_1
#define RDS_PROTOCOL_MAJOR(v)	((v) >> 8)
#define RDS_PROTOCOL_MINOR(v)	((v) & 255)
#define RDS_PROTOCOL(maj, min)	(((maj) << 8) | min)

/*
 * XXX randomly chosen, but at least seems to be unused:
 * #               18464-18768 Unassigned
 * We should do better.  We want a reserved port to discourage unpriv'ed
 * userspace from listening.
 */
#define RDS_PORT	18634

#ifdef ATOMIC64_INIT
#define KERNEL_HAS_ATOMIC64
#endif

#ifdef DEBUG
#define rdsdebug(fmt, args...) pr_debug("%s(): " fmt, __func__ , ##args)
#else
/* sigh, pr_debug() causes unused variable warnings */
static inline void __attribute__ ((format (printf, 1, 2)))
rdsdebug(char *fmt, ...)
{
}
#endif

/* XXX is there one of these somewhere? */
#define ceil(x, y) \
	({ unsigned long __x = (x), __y = (y); (__x + __y - 1) / __y; })

#define RDS_FRAG_SHIFT	12
#define RDS_FRAG_SIZE	((unsigned int)(1 << RDS_FRAG_SHIFT))

#define RDS_CONG_MAP_BYTES	(65536 / 8)
#define RDS_CONG_MAP_LONGS	(RDS_CONG_MAP_BYTES / sizeof(unsigned long))
#define RDS_CONG_MAP_PAGES	(PAGE_ALIGN(RDS_CONG_MAP_BYTES) / PAGE_SIZE)
#define RDS_CONG_MAP_PAGE_BITS	(PAGE_SIZE * 8)

struct rds_cong_map {
	struct rb_node		m_rb_node;
	__be32			m_addr;
	wait_queue_head_t	m_waitq;
	struct list_head	m_conn_list;
	unsigned long		m_page_addrs[RDS_CONG_MAP_PAGES];
};


/*
 * This is how we will track the connection state:
 * A connection is always in one of the following
 * states. Updates to the state are atomic and imply
 * a memory barrier.
 */
enum {
	RDS_CONN_DOWN = 0,
	RDS_CONN_CONNECTING,
	RDS_CONN_DISCONNECTING,
	RDS_CONN_UP,
	RDS_CONN_ERROR,
};

/* Bits for c_flags */
#define RDS_LL_SEND_FULL	0
#define RDS_RECONNECT_PENDING	1

struct rds_connection {
	struct hlist_node	c_hash_node;
	__be32			c_laddr;
	__be32			c_faddr;
	unsigned int		c_loopback:1;
	struct rds_connection	*c_passive;

	struct rds_cong_map	*c_lcong;
	struct rds_cong_map	*c_fcong;

	struct mutex		c_send_lock;	/* protect send ring */
	struct rds_message	*c_xmit_rm;
	unsigned long		c_xmit_sg;
	unsigned int		c_xmit_hdr_off;
	unsigned int		c_xmit_data_off;
	unsigned int		c_xmit_rdma_sent;

	spinlock_t		c_lock;		/* protect msg queues */
	u64			c_next_tx_seq;
	struct list_head	c_send_queue;
	struct list_head	c_retrans;

	u64			c_next_rx_seq;

	struct rds_transport	*c_trans;
	void			*c_transport_data;

	atomic_t		c_state;
	unsigned long		c_flags;
	unsigned long		c_reconnect_jiffies;
	struct delayed_work	c_send_w;
	struct delayed_work	c_recv_w;
	struct delayed_work	c_conn_w;
	struct work_struct	c_down_w;
	struct mutex		c_cm_lock;	/* protect conn state & cm */

	struct list_head	c_map_item;
	unsigned long		c_map_queued;
	unsigned long		c_map_offset;
	unsigned long		c_map_bytes;

	unsigned int		c_unacked_packets;
	unsigned int		c_unacked_bytes;

	/* Protocol version */
	unsigned int		c_version;
};

#define RDS_FLAG_CONG_BITMAP	0x01
#define RDS_FLAG_ACK_REQUIRED	0x02
#define RDS_FLAG_RETRANSMITTED	0x04
#define RDS_MAX_ADV_CREDIT	255

/*
 * Maximum space available for extension headers.
 */
#define RDS_HEADER_EXT_SPACE	16

struct rds_header {
	__be64	h_sequence;
	__be64	h_ack;
	__be32	h_len;
	__be16	h_sport;
	__be16	h_dport;
	u8	h_flags;
	u8	h_credit;
	u8	h_padding[4];
	__sum16	h_csum;

	u8	h_exthdr[RDS_HEADER_EXT_SPACE];
};

/*
 * Reserved - indicates end of extensions
 */
#define RDS_EXTHDR_NONE		0

/*
 * This extension header is included in the very
 * first message that is sent on a new connection,
 * and identifies the protocol level. This will help
 * rolling updates if a future change requires breaking
 * the protocol.
 * NB: This is no longer true for IB, where we do a version
 * negotiation during the connection setup phase (protocol
 * version information is included in the RDMA CM private data).
 */
#define RDS_EXTHDR_VERSION	1
struct rds_ext_header_version {
	__be32			h_version;
};

/*
 * This extension header is included in the RDS message
 * chasing an RDMA operation.
 */
#define RDS_EXTHDR_RDMA		2
struct rds_ext_header_rdma {
	__be32			h_rdma_rkey;
};

/*
 * This extension header tells the peer about the
 * destination <R_Key,offset> of the requested RDMA
 * operation.
 */
#define RDS_EXTHDR_RDMA_DEST	3
struct rds_ext_header_rdma_dest {
	__be32			h_rdma_rkey;
	__be32			h_rdma_offset;
};

#define __RDS_EXTHDR_MAX	16 /* for now */

struct rds_incoming {
	atomic_t		i_refcount;
	struct list_head	i_item;
	struct rds_connection	*i_conn;
	struct rds_header	i_hdr;
	unsigned long		i_rx_jiffies;
	__be32			i_saddr;

	rds_rdma_cookie_t	i_rdma_cookie;
};

/*
 * m_sock_item and m_conn_item are on lists that are serialized under
 * conn->c_lock.  m_sock_item has additional meaning in that once it is empty
 * the message will not be put back on the retransmit list after being sent.
 * messages that are canceled while being sent rely on this.
 *
 * m_inc is used by loopback so that it can pass an incoming message straight
 * back up into the rx path.  It embeds a wire header which is also used by
 * the send path, which is kind of awkward.
 *
 * m_sock_item indicates the message's presence on a socket's send or receive
 * queue.  m_rs will point to that socket.
 *
 * m_daddr is used by cancellation to prune messages to a given destination.
 *
 * The RDS_MSG_ON_SOCK and RDS_MSG_ON_CONN flags are used to avoid lock
 * nesting.  As paths iterate over messages on a sock, or conn, they must
 * also lock the conn, or sock, to remove the message from those lists too.
 * Testing the flag to determine if the message is still on the lists lets
 * us avoid testing the list_head directly.  That means each path can use
 * the message's list_head to keep it on a local list while juggling locks
 * without confusing the other path.
 *
 * m_ack_seq is an optional field set by transports who need a different
 * sequence number range to invalidate.  They can use this in a callback
 * that they pass to rds_send_drop_acked() to see if each message has been
 * acked.  The HAS_ACK_SEQ flag can be used to detect messages which haven't
 * had ack_seq set yet.
 */
#define RDS_MSG_ON_SOCK		1
#define RDS_MSG_ON_CONN		2
#define RDS_MSG_HAS_ACK_SEQ	3
#define RDS_MSG_ACK_REQUIRED	4
#define RDS_MSG_RETRANSMITTED	5
#define RDS_MSG_MAPPED		6
#define RDS_MSG_PAGEVEC		7

struct rds_message {
	atomic_t		m_refcount;
	struct list_head	m_sock_item;
	struct list_head	m_conn_item;
	struct rds_incoming	m_inc;
	u64			m_ack_seq;
	__be32			m_daddr;
	unsigned long		m_flags;

	/* Never access m_rs without holding m_rs_lock.
	 * Lock nesting is
	 *  rm->m_rs_lock
	 *   -> rs->rs_lock
	 */
	spinlock_t		m_rs_lock;
	struct rds_sock		*m_rs;
	struct rds_rdma_op	*m_rdma_op;
	rds_rdma_cookie_t	m_rdma_cookie;
	struct rds_mr		*m_rdma_mr;
	unsigned int		m_nents;
	unsigned int		m_count;
	struct scatterlist	m_sg[0];
};

/*
 * The RDS notifier is used (optionally) to tell the application about
 * completed RDMA operations. Rather than keeping the whole rds message
 * around on the queue, we allocate a small notifier that is put on the
 * socket's notifier_list. Notifications are delivered to the application
 * through control messages.
 */
struct rds_notifier {
	struct list_head	n_list;
	uint64_t		n_user_token;
	int			n_status;
};

/**
 * struct rds_transport -  transport specific behavioural hooks
 *
 * @xmit: .xmit is called by rds_send_xmit() to tell the transport to send
 *        part of a message.  The caller serializes on the send_sem so this
 *        doesn't need to be reentrant for a given conn.  The header must be
 *        sent before the data payload.  .xmit must be prepared to send a
 *        message with no data payload.  .xmit should return the number of
 *        bytes that were sent down the connection, including header bytes.
 *        Returning 0 tells the caller that it doesn't need to perform any
 *        additional work now.  This is usually the case when the transport has
 *        filled the sending queue for its connection and will handle
 *        triggering the rds thread to continue the send when space becomes
 *        available.  Returning -EAGAIN tells the caller to retry the send
 *        immediately.  Returning -ENOMEM tells the caller to retry the send at
 *        some point in the future.
 *
 * @conn_shutdown: conn_shutdown stops traffic on the given connection.  Once
 *                 it returns the connection can not call rds_recv_incoming().
 *                 This will only be called once after conn_connect returns
 *                 non-zero success and will The caller serializes this with
 *                 the send and connecting paths (xmit_* and conn_*).  The
 *                 transport is responsible for other serialization, including
 *                 rds_recv_incoming().  This is called in process context but
 *                 should try hard not to block.
 *
 * @xmit_cong_map: This asks the transport to send the local bitmap down the
 * 		   given connection.  XXX get a better story about the bitmap
 * 		   flag and header.
 */

#define RDS_TRANS_IB	0
#define RDS_TRANS_IWARP	1
#define RDS_TRANS_TCP	2
#define RDS_TRANS_COUNT	3

struct rds_transport {
	char			t_name[TRANSNAMSIZ];
	struct list_head	t_item;
	struct module		*t_owner;
	unsigned int		t_prefer_loopback:1;
	unsigned int		t_type;

	int (*laddr_check)(__be32 addr);
	int (*conn_alloc)(struct rds_connection *conn, gfp_t gfp);
	void (*conn_free)(void *data);
	int (*conn_connect)(struct rds_connection *conn);
	void (*conn_shutdown)(struct rds_connection *conn);
	void (*xmit_prepare)(struct rds_connection *conn);
	void (*xmit_complete)(struct rds_connection *conn);
	int (*xmit)(struct rds_connection *conn, struct rds_message *rm,
		    unsigned int hdr_off, unsigned int sg, unsigned int off);
	int (*xmit_cong_map</