PSARC 2007/587 Volo -- Low Latency Socket Framework

PSARC 2008/694 Volo Interfaces Amendment
6765829 Integration of project Volo PSARC/2007/587
6644935 mblk cred_t reference counting limits scalability
6693633 TCP receive does not scale because of heavy refcounting of cred structures
4764841 connect/accept is slow on Solaris when compared to Linux
5105708 socket creation retains hold on accessvp
4764836 setsockopt is slow on Solaris when compared to Linux
4772191 socket close(2) is slow on Solaris when compared to Linux

--HG--
rename : usr/src/uts/common/fs/sockfs/socksctp.c => usr/src/uts/common/inet/sockmods/socksctp.c
rename : usr/src/uts/common/fs/sockfs/socksctp.h => usr/src/uts/common/inet/sockmods/socksctp.h
rename : usr/src/uts/common/fs/sockfs/socksctpsubr.c => usr/src/uts/common/inet/sockmods/socksctpsubr.c
rename : usr/src/uts/common/fs/sockfs/socksdp.c => usr/src/uts/common/inet/sockmods/socksdp.c
rename : usr/src/uts/common/fs/sockfs/socksdp.h => usr/src/uts/common/inet/sockmods/socksdp.h

1 files changed, 387 insertions, 345 deletions

diff --git a/usr/src/uts/common/sys/socketvar.h b/usr/src/uts/common/sys/socketvar.h
index 37a699345a..510d9445cf 100644
--- a/usr/src/uts/common/sys/socketvar.h
+++ b/usr/src/uts/common/sys/socketvar.h
@@ -48,25 +48,18 @@
 #include <sys/file.h>
 #include <sys/param.h>
 #include <sys/zone.h>
+#include <sys/sdt.h>
+#include <sys/modctl.h>
+#include <sys/atomic.h>
+#include <sys/socket.h>
+#include <sys/ksocket.h>
 #include <sys/sodirect.h>
-#include <inet/kssl/ksslapi.h>
 
 #ifdef	__cplusplus
 extern "C" {
 #endif
 
 /*
- * Internal representation used for addresses.
- */
-struct soaddr {
-	struct sockaddr	*soa_sa;	/* Actual address */
-	t_uscalar_t	soa_len;	/* Length in bytes for kmem_free */
-	t_uscalar_t	soa_maxlen;	/* Allocated length */
-};
-/* Maximum size address for transports that have ADDR_size == 1 */
-#define	SOA_DEFSIZE	128
-
-/*
  * Internal representation of the address used to represent addresses
  * in the loopback transport for AF_UNIX. While the sockaddr_un is used
  * as the sockfs layer address for AF_UNIX the pathnames contained in
@@ -97,6 +90,10 @@ struct sockaddr_ux {
 	struct so_ux_addr	sou_addr;
 };
 
+#if defined(_KERNEL) || defined(_KMEMUSER)
+
+#include <sys/socket_proto.h>
+
 typedef struct sonodeops sonodeops_t;
 typedef struct sonode sonode_t;
 
@@ -105,236 +102,149 @@ typedef struct sonode sonode_t;
  * name space and can not be opened using open() - only the socket, socketpair
  * and accept calls create sonodes.
  *
- * When an AF_UNIX socket is bound to a pathname the sockfs
- * creates a VSOCK vnode in the underlying file system. However, the vnodeops
- * etc in this VNODE remain those of the underlying file system.
- * Sockfs uses the v_stream pointer in the underlying file system VSOCK node
- * to find the sonode bound to the pathname. The bound pathname vnode
- * is accessed through so_ux_vp.
- *
- * A socket always corresponds to a VCHR stream representing the transport
- * provider (e.g. /dev/tcp). This information is retrieved from the kernel
- * socket configuration table and entered into so_accessvp. sockfs uses
- * this to perform VOP_ACCESS checks before allowing an open of the transport
- * provider.
+ * The locking of sockfs uses the so_lock mutex plus the SOLOCKED and
+ * SOREADLOCKED flags in so_flag. The mutex protects all the state in the
+ * sonode. It is expected that the underlying transport protocol serializes
+ * socket operations, so sockfs will not normally not single-thread
+ * operations. However, certain sockets, including TPI based ones, can only
+ * handle one control operation at a time. The SOLOCKED flag is used to
+ * single-thread operations from sockfs users to prevent e.g. multiple bind()
+ * calls to operate on the same sonode concurrently. The SOREADLOCKED flag is
+ * used to ensure that only one thread sleeps in kstrgetmsg for a given
+ * sonode. This is needed to ensure atomic operation for things like
+ * MSG_WAITALL.
  *
- * The locking of sockfs uses the so_lock mutex plus the SOLOCKED
- * and SOREADLOCKED flags in so_flag. The mutex protects all the state
- * in the sonode. The SOLOCKED flag is used to single-thread operations from
- * sockfs users to prevent e.g. multiple bind() calls to operate on the
- * same sonode concurrently. The SOREADLOCKED flag is used to ensure that
- * only one thread sleeps in kstrgetmsg for a given sonode. This is needed
- * to ensure atomic operation for things like MSG_WAITALL.
+ * The so_fallback_rwlock is used to ensure that for sockets that can
+ * fall back to TPI, the fallback is not initiated until all pending
+ * operations have completed.
  *
  * Note that so_lock is sometimes held across calls that might go to sleep
  * (kmem_alloc and soallocproto*). This implies that no other lock in
  * the system should be held when calling into sockfs; from the system call
- * side or from strrput. If locks are held while calling into sockfs
- * the system might hang when running low on memory.
+ * side or from strrput (in case of TPI based sockets). If locks are held
+ * while calling into sockfs the system might hang when running low on memory.
  */
 struct sonode {
 	struct	vnode	*so_vnode;	/* vnode associated with this sonode */
 
-	sonodeops_t	*so_ops;	/* operations vector for this sonode */
-
-	/*
-	 * These fields are initialized once.
-	 */
-	dev_t		so_dev;		/* device the sonode represents */
-	struct	vnode	*so_accessvp;	/* vnode for the /dev entry */
+	sonodeops_t 	*so_ops;	/* operations vector for this sonode */
+	void		*so_priv;	/* sonode private data */
 
-	/* The locks themselves */
+	krwlock_t	so_fallback_rwlock;
 	kmutex_t	so_lock;	/* protects sonode fields */
-	kmutex_t	so_plumb_lock;	/* serializes plumbs, and the related */
-					/* fields so_version and so_pushcnt */
+
 	kcondvar_t	so_state_cv;	/* synchronize state changes */
-	kcondvar_t	so_ack_cv;	/* wait for TPI acks */
-	kcondvar_t	so_connind_cv;	/* wait for T_CONN_IND */
 	kcondvar_t	so_want_cv;	/* wait due to SOLOCKED */
 
 	/* These fields are protected by so_lock */
-	uint_t	so_state;		/* internal state flags SS_*, below */
-	uint_t	so_mode;		/* characteristics on socket. SM_* */
 
-	mblk_t	*so_ack_mp;		/* TPI ack received from below */
-	mblk_t	*so_conn_ind_head;	/* b_next list of T_CONN_IND */
-	mblk_t	*so_conn_ind_tail;
-	mblk_t	*so_unbind_mp;		/* Preallocated T_UNBIND_REQ message */
+	uint_t		so_state;	/* internal state flags SS_*, below */
+	uint_t		so_mode;	/* characteristics on socket. SM_* */
+	ushort_t 	so_flag;	/* flags, see below */
+	int		so_count;	/* count of opened references */
+
+	sock_connid_t	so_proto_connid; /* protocol generation number */
 
-	ushort_t so_flag;		/* flags, see below */
-	dev_t	so_fsid;		/* file system identifier */
-	time_t  so_atime;		/* time of last access */
-	time_t  so_mtime;		/* time of last modification */
-	time_t  so_ctime;		/* time of last attributes change */
-	int	so_count;		/* count of opened references */
+	ushort_t 	so_error;	/* error affecting connection */
 
+	struct sockparams *so_sockparams;	/* vnode or socket module */
 	/* Needed to recreate the same socket for accept */
 	short	so_family;
 	short	so_type;
 	short	so_protocol;
 	short	so_version;		/* From so_socket call */
-	short	so_pushcnt;		/* Number of modules above "sockmod" */
+
+	/* Accept queue */
+	kmutex_t	so_acceptq_lock;	/* protects accept queue */
+	struct sonode	*so_acceptq_next;	/* acceptq list node */
+	struct sonode 	*so_acceptq_head;
+	struct sonode	**so_acceptq_tail;
+	unsigned int	so_acceptq_len;
+	unsigned int	so_backlog;		/* Listen backlog */
+	kcondvar_t	so_acceptq_cv;		/* wait for new conn. */
 
 	/* Options */
 	short	so_options;		/* From socket call, see socket.h */
 	struct linger	so_linger;	/* SO_LINGER value */
-	int	so_sndbuf;		/* SO_SNDBUF value */
-	int	so_rcvbuf;		/* SO_RCVBUF value */
-	int	so_sndlowat;		/* send low water mark */
-	int	so_rcvlowat;		/* receive low water mark */
-#ifdef notyet
-	int	so_sndtimeo;		/* Not yet implemented */
-	int	so_rcvtimeo;		/* Not yet implemented */
-#endif /* notyet */
-	ushort_t so_error;		/* error affecting connection */
-	ushort_t so_delayed_error;	/* From T_uderror_ind */
-	int	so_backlog;		/* Listen backlog */
+#define	so_sndbuf	so_proto_props.sopp_txhiwat	/* SO_SNDBUF value */
+#define	so_sndlowat	so_proto_props.sopp_txlowat	/* tx low water mark */
+#define	so_rcvbuf	so_proto_props.sopp_rxhiwat	/* SO_RCVBUF value */
+#define	so_rcvlowat	so_proto_props.sopp_rxlowat	/* rx low water mark */
+#define	so_max_addr_len	so_proto_props.sopp_maxaddrlen
+#define	so_minpsz	so_proto_props.sopp_minpsz
+#define	so_maxpsz	so_proto_props.sopp_maxpsz
+
+	clock_t	so_sndtimeo;		/* send timeout */
+	clock_t	so_rcvtimeo;		/* recv timeout */
 
-	/*
-	 * The counts (so_oobcnt and so_oobsigcnt) track the number of
-	 * urgent indicates that are (logically) queued on the stream head
-	 * read queue. The urgent data is queued on the stream head
-	 * as follows.
-	 *
-	 * In the normal case the SIGURG is not generated until
-	 * the T_EXDATA_IND arrives at the stream head. However, transports
-	 * that have an early indication that urgent data is pending
-	 * (e.g. TCP receiving a "new" urgent pointer value) can send up
-	 * an M_PCPROTO/SIGURG message to generate the signal early.
-	 *
-	 * The mark is indicated by either:
-	 *  - a T_EXDATA_IND (with no M_DATA b_cont) with MSGMARK set.
-	 *    When this message is consumed by sorecvmsg the socket layer
-	 *    sets SS_RCVATMARK until data has been consumed past the mark.
-	 *  - a message with MSGMARKNEXT set (indicating that the
-	 *    first byte of the next message constitutes the mark). When
-	 *    the last byte of the MSGMARKNEXT message is consumed in
-	 *    the stream head the stream head sets STRATMARK. This flag
-	 *    is cleared when at least one byte is read. (Note that
-	 *    the MSGMARKNEXT messages can be of zero length when there
-	 *    is no previous data to which the marknext can be attached.)
-	 *
-	 * While the T_EXDATA_IND method is the common case which is used
-	 * with all TPI transports, the MSGMARKNEXT method is needed to
-	 * indicate the mark when e.g. the TCP urgent byte has not been
-	 * received yet but the TCP urgent pointer has made TCP generate
-	 * the M_PCSIG/SIGURG.
-	 *
-	 * The signal (the M_PCSIG carrying the SIGURG) and the mark
-	 * indication can not be delivered as a single message, since
-	 * the signal should be delivered as high priority and any mark
-	 * indication must flow with the data. This implies that immediately
-	 * when the SIGURG has been delivered if the stream head queue is
-	 * empty it is impossible to determine if this will be the position
-	 * of the mark. This race condition is resolved by using MSGNOTMARKNEXT
-	 * messages and the STRNOTATMARK flag in the stream head. The
-	 * SIOCATMARK code calls the stream head to wait for either a
-	 * non-empty queue or one of the STR*ATMARK flags being set.
-	 * This implies that any transport that is sending M_PCSIG(SIGURG)
-	 * should send the appropriate MSGNOTMARKNEXT message (which can be
-	 * zero length) after sending an M_PCSIG to prevent SIOCATMARK
-	 * from sleeping unnecessarily.
-	 */
 	mblk_t	*so_oobmsg;		/* outofline oob data */
-	uint_t	so_oobsigcnt;		/* Number of SIGURG generated */
-	uint_t	so_oobcnt;		/* Number of T_EXDATA_IND queued */
+	ssize_t	so_oobmark;		/* offset of the oob data */
+
 	pid_t	so_pgrp;		/* pgrp for signals */
 
-	/* From T_info_ack */
-	t_uscalar_t	so_tsdu_size;
-	t_uscalar_t	so_etsdu_size;
-	t_scalar_t	so_addr_size;
-	t_uscalar_t	so_opt_size;
-	t_uscalar_t	so_tidu_size;
-	t_scalar_t	so_serv_type;
+	cred_t		*so_peercred;	/* connected socket peer cred */
+	pid_t		so_cpid;	/* connected socket peer cached pid */
+	zoneid_t	so_zoneid;	/* opener's zoneid */
 
-	/* From T_capability_ack */
-	t_uscalar_t	so_acceptor_id;
+	struct pollhead	so_poll_list;	/* common pollhead */
+	short		so_pollev;	/* events that should be generated */
 
-	/* Internal provider information */
-	struct tpi_provinfo	*so_provinfo;
+	/* Receive */
+	unsigned int	so_rcv_queued;
+	mblk_t		*so_rcv_q_head;
+	mblk_t		*so_rcv_q_last_head;
+	mblk_t		*so_rcv_head;		/* 1st mblk in the list */
+	mblk_t		*so_rcv_last_head;	/* last mblk in b_next chain */
+	kcondvar_t	so_rcv_cv;
+	uint_t		so_rcv_wanted;	/* # of bytes wanted by app */
+	timeout_id_t	so_rcv_timer_tid;
 
-	/*
-	 * The local and remote addresses have multiple purposes
-	 * but one of the key reasons for their existence and careful
-	 * tracking in sockfs is to support getsockname and getpeername
-	 * when the transport does not handle the TI_GET*NAME ioctls
-	 * and caching when it does (signaled by valid bits in so_state).
-	 * When all transports support the new TPI (with T_ADDR_REQ)
-	 * we can revisit this code.
-	 * The other usage of so_faddr is to keep the "connected to"
-	 * address for datagram sockets.
-	 * Finally, for AF_UNIX both local and remote addresses are used
-	 * to record the sockaddr_un since we use a separate namespace
-	 * in the loopback transport.
-	 */
-	struct soaddr so_laddr;		/* Local address */
-	struct soaddr so_faddr;		/* Peer address */
-#define	so_laddr_sa	so_laddr.soa_sa
-#define	so_faddr_sa	so_faddr.soa_sa
-#define	so_laddr_len	so_laddr.soa_len
-#define	so_faddr_len	so_faddr.soa_len
-#define	so_laddr_maxlen	so_laddr.soa_maxlen
-#define	so_faddr_maxlen	so_faddr.soa_maxlen
-	mblk_t		*so_eaddr_mp;	/* for so_delayed_error */
+#define	so_rcv_thresh	so_proto_props.sopp_rcvthresh
+#define	so_rcv_timer_interval so_proto_props.sopp_rcvtimer
 
-	/*
-	 * For AF_UNIX sockets:
-	 * so_ux_laddr/faddr records the internal addresses used with the
-	 * transport.
-	 * so_ux_vp and v_stream->sd_vnode form the cross-
-	 * linkage between the underlying fs vnode corresponding to
-	 * the bound sockaddr_un and the socket node.
-	 */
-	struct so_ux_addr so_ux_laddr;	/* laddr bound with the transport */
-	struct so_ux_addr so_ux_faddr;	/* temporary peer address */
-	struct vnode	*so_ux_bound_vp; /* bound AF_UNIX file system vnode */
-	struct sonode	*so_next;	/* next sonode on socklist	*/
-	struct sonode	*so_prev;	/* previous sonode on socklist	*/
-	mblk_t	*so_discon_ind_mp;	/* T_DISCON_IND received from below */
-
-					/* put here for delayed processing  */
-	void		*so_priv;	/* sonode private data */
-	cred_t		*so_peercred;	/* connected socket peer cred */
-	pid_t		so_cpid;	/* connected socket peer cached pid */
-	zoneid_t	so_zoneid;	/* opener's zoneid */
+	/* Send */
+	boolean_t	so_snd_qfull;	/* Transmit full */
+	kcondvar_t	so_snd_cv;
 
-	kmem_cache_t	*so_cache;	/* object cache of this "sonode". */
-	void		*so_obj;	/* object to free */
+	boolean_t so_rcv_wakeup;
+	boolean_t so_snd_wakeup;
 
-	/*
-	 * For NL7C sockets:
-	 *
-	 * so_nl7c_flags	the NL7C state of URL processing.
-	 *
-	 * so_nl7c_rcv_mp	mblk_t chain of already received data to be
-	 *			passed up to the app after NL7C gives up on
-	 *			a socket.
-	 *
-	 * so_nl7c_rcv_rval	returned rval for last mblk_t from above.
-	 *
-	 * so_nl7c_uri		the URI currently being processed.
-	 *
-	 * so_nl7c_rtime	URI request gethrestime_sec().
-	 *
-	 * so_nl7c_addr		pointer returned by nl7c_addr_lookup().
-	 */
-	uint64_t	so_nl7c_flags;
-	mblk_t		*so_nl7c_rcv_mp;
-	int64_t		so_nl7c_rcv_rval;
-	void		*so_nl7c_uri;
-	time_t		so_nl7c_rtime;
-	void		*so_nl7c_addr;
-
-	/* For sockets acting as an in-kernel SSL proxy */
-	kssl_endpt_type_t	so_kssl_type;	/* is proxy/is proxied/none */
-	kssl_ent_t		so_kssl_ent;	/* SSL config entry */
-	kssl_ctx_t		so_kssl_ctx;	/* SSL session context */
+	/* Communication channel with protocol */
+	sock_lower_handle_t	so_proto_handle;
+	sock_downcalls_t 	*so_downcalls;
+
+	struct sock_proto_props	so_proto_props; /* protocol settings */
+	boolean_t		so_flowctrld;	/* Flow controlled */
+	uint_t			so_copyflag;	/* Copy related flag */
+	kcondvar_t		so_copy_cv;	/* Copy cond variable */
+
+	/* kernel sockets */
+	ksocket_callbacks_t 	so_ksock_callbacks;
+	void			*so_ksock_cb_arg;	/* callback argument */
+	kcondvar_t		so_closing_cv;
 
 	/* != NULL for sodirect_t enabled socket */
-	sodirect_t	*so_direct;
+	sodirect_t		*so_direct;
 };
 
+/*
+ * We do an initial check for events without holding locks. However,
+ * if there are no event available, then we redo the check for POLLIN
+ * events under the lock.
+ */
+#define	SO_HAVE_DATA(so)						\
+	((so)->so_rcv_timer_tid == 0 && (so->so_rcv_queued > 0)) ||	\
+	((so)->so_rcv_queued > (so)->so_rcv_thresh) ||			\
+	((so)->so_state & SS_CANTRCVMORE)
+
+/*
+ * Events handled by the protocol (in case sd_poll is set)
+ */
+#define	SO_PROTO_POLLEV		(POLLIN|POLLRDNORM|POLLRDBAND)
+
+
+#endif /* _KERNEL || _KMEMUSER */
+
 /* flags */
 #define	SOMOD		0x0001		/* update socket modification time */
 #define	SOACC		0x0002		/* update socket access time */
@@ -345,6 +255,8 @@ struct sonode {
 #define	SOCLONE		0x0080		/* child of clone driver */
 #define	SOASYNC_UNBIND	0x0100		/* wait for ACK of async unbind */
 
+#define	SOCK_IS_NONSTR(so)	((so)->so_vnode->v_stream == NULL)
+
 /*
  * Socket state bits.
  */
@@ -360,31 +272,59 @@ struct sonode {
 
 #define	SS_ASYNC		0x00000100 /* async i/o notify */
 #define	SS_ACCEPTCONN		0x00000200 /* listen done */
-#define	SS_HASCONNIND		0x00000400 /* T_CONN_IND for poll */
+/*	unused			0x00000400 */	/* was SS_HASCONNIND */
 #define	SS_SAVEDEOR		0x00000800 /* Saved MSG_EOR rcv side state */
 
 #define	SS_RCVATMARK		0x00001000 /* at mark on input */
 #define	SS_OOBPEND		0x00002000 /* OOB pending or present - poll */
 #define	SS_HAVEOOBDATA		0x00004000 /* OOB data present */
 #define	SS_HADOOBDATA		0x00008000 /* OOB data consumed */
+#define	SS_CLOSING		0x00010000 /* in process of closing */
 
-#define	SS_FADDR_NOXLATE	0x00020000 /* No xlation of faddr for AF_UNIX */
-
-#define	SS_HASDATA		0x00040000 /* NCAfs: data available */
-#define	SS_DONEREAD		0x00080000 /* NCAfs: all data read */
-#define	SS_MOREDATA		0x00100000 /* NCAfs: NCA has more data */
+/*	unused			0x00020000 */	/* was SS_FADDR_NOXLATE */
+/*	unused			0x00040000 */	/* was SS_HASDATA */
+/*	unused 			0x00080000 */	/* was SS_DONEREAD */
+/*	unused 			0x00100000 */	/* was SS_MOREDATA */
+/*	unused 			0x00200000 */	/* was SS_DIRECT */
 
-#define	SS_DIRECT		0x00200000 /* transport is directly below */
 #define	SS_SODIRECT		0x00400000 /* transport supports sodirect */
 
-#define	SS_LADDR_VALID		0x01000000	/* so_laddr valid for user */
-#define	SS_FADDR_VALID		0x02000000	/* so_faddr valid for user */
+/*	unused			0x01000000 */	/* was SS_LADDR_VALID */
+/*	unused			0x02000000 */	/* was SS_FADDR_VALID */
+
+#define	SS_SENTLASTREADSIG	0x10000000 /* last rx signal has been sent */
+#define	SS_SENTLASTWRITESIG	0x20000000 /* last tx signal has been sent */
+
+#define	SS_FALLBACK_PENDING	0x40000000
+#define	SS_FALLBACK_COMP	0x80000000
+
 
 /* Set of states when the socket can't be rebound */
 #define	SS_CANTREBIND	(SS_ISCONNECTED|SS_ISCONNECTING|SS_ISDISCONNECTING|\
 			    SS_CANTSENDMORE|SS_CANTRCVMORE|SS_ACCEPTCONN)
 
 /*
+ * Sockets that can fall back to TPI must ensure that fall back is not
+ * initiated while a thread is using a socket.
+ */
+#define	SO_BLOCK_FALLBACK(so, fn) {			\
+	ASSERT(MUTEX_NOT_HELD(&(so)->so_lock));		\
+	rw_enter(&(so)->so_fallback_rwlock, RW_READER);	\
+	if ((so)->so_state & SS_FALLBACK_COMP) {	\
+		rw_exit(&(so)->so_fallback_rwlock);	\
+		return (fn);				\
+	}						\
+}
+
+#define	SO_UNBLOCK_FALLBACK(so)	{			\
+	rw_exit(&(so)->so_fallback_rwlock);		\
+}
+
+/* Poll events */
+#define	SO_POLLEV_IN		0x1	/* POLLIN wakeup needed */
+#define	SO_POLLEV_ALWAYS	0x2	/* wakeups */
+
+/*
  * Characteristics of sockets. Not changed after the socket is created.
  */
 #define	SM_PRIV			0x001	/* privileged for broadcast, raw... */
@@ -399,6 +339,10 @@ struct sonode {
 
 #define	SM_ACCEPTOR_ID		0x100	/* so_acceptor_id is valid */
 
+#define	SM_KERNEL		0x200	/* kernel socket */
+
+#define	SM_ACCEPTSUPP		0x400	/* can handle accept() */
+
 /*
  * Socket versions. Used by the socket library when calling _so_socket().
  */
@@ -409,21 +353,177 @@ struct sonode {
 #define	SOV_XPG4_2	4	/* Xnet socket */
 
 #if defined(_KERNEL) || defined(_KMEMUSER)
+
+/*
+ * sonode create and destroy functions.
+ */
+typedef struct sonode *(*so_create_func_t)(struct sockparams *,
+    int, int, int, int, int, int *, cred_t *);
+typedef void (*so_destroy_func_t)(struct sonode *);
+
+/* STREAM device information */
+typedef struct sdev_info {
+	char	*sd_devpath;
+	int	sd_devpathlen; /* Is 0 if sp_devpath is a static string */
+	vnode_t	*sd_vnode;
+} sdev_info_t;
+
+#define	SOCKMOD_VERSION		1
+/* name of the TPI pseudo socket module */
+#define	SOTPI_SMOD_NAME		"socktpi"
+
+typedef struct __smod_priv_s {
+	so_create_func_t	smodp_sock_create_func;
+	so_destroy_func_t	smodp_sock_destroy_func;
+	so_proto_fallback_func_t smodp_proto_fallback_func;
+} __smod_priv_t;
+
 /*
- * Used for mapping family/type/protocol to vnode.
- * Defined here so that crash can use it.
+ * Socket module register information
+ */
+typedef struct smod_reg_s {
+	int		smod_version;
+	char		*smod_name;
+	size_t		smod_uc_version;
+	size_t		smod_dc_version;
+	so_proto_create_func_t	smod_proto_create_func;
+
+	/* __smod_priv_data must be NULL */
+	__smod_priv_t	*__smod_priv;
+} smod_reg_t;
+
+/*
+ * Socket module information
+ */
+typedef struct smod_info {
+	int		smod_version;
+	char		*smod_name;
+	uint_t		smod_refcnt;		/* # of entries */
+	size_t		smod_uc_version; 	/* upcall version */
+	size_t		smod_dc_version;	/* down call version */
+	so_proto_create_func_t	smod_proto_create_func;
+	so_proto_fallback_func_t smod_proto_fallback_func;
+	so_create_func_t	smod_sock_create_func;
+	so_destroy_func_t	smod_sock_destroy_func;
+	list_node_t	smod_node;
+} smod_info_t;
+
+/*
+ * sockparams
+ *
+ * Used for mapping family/type/protocol to module
  */
 struct sockparams {
-	int	sp_domain;
-	int	sp_type;
-	int	sp_protocol;
-	char	*sp_devpath;
-	int	sp_devpathlen;	/* Is 0 if sp_devpath is a static string */
-	vnode_t	*sp_vnode;
-	struct sockparams *sp_next;
+	/*
+	 * The family, type, protocol, sdev_info and smod_info are
+	 * set when the entry is created, and they will never change
+	 * thereafter.
+	 */
+	int		sp_family;
+	int		sp_type;
+	int		sp_protocol;
+
+	sdev_info_t	sp_sdev_info;	/* STREAM device */
+	char		*sp_smod_name;	/* socket module name */
+	smod_info_t	*sp_smod_info;	/* socket module */
+
+	kmutex_t	sp_lock;	/* lock for refcnt */
+	uint64_t	sp_refcnt;	/* entry reference count */
+
+	/*
+	 * The entries below are only modified while holding
+	 * splist_lock as a writer.
+	 */
+	int		sp_flags;	/* see below */
+	list_node_t	sp_node;
 };
 
-extern struct sockparams *sphead;
+
+/*
+ * sockparams flags
+ */
+#define	SOCKPARAMS_EPHEMERAL	0x1	/* temp. entry, not on global list */
+
+extern void sockparams_init(void);
+extern struct sockparams *sockparams_hold_ephemeral_bydev(int, int, int,
+    const char *, int, int *);
+extern struct sockparams *sockparams_hold_ephemeral_bymod(int, int, int,
+    const char *, int, int *);
+extern void sockparams_ephemeral_drop_last_ref(struct sockparams *);
+
+extern void smod_init(void);
+extern void smod_add(smod_info_t *);
+extern int smod_register(const smod_reg_t *);
+extern int smod_unregister(const char *);
+extern smod_info_t *smod_lookup_byname(const char *);
+
+#define	SOCKPARAMS_HAS_DEVICE(sp)					\
+	((sp)->sp_sdev_info.sd_devpath != NULL)
+
+/* Increase the smod_info_t reference count */
+#define	SMOD_INC_REF(smodp) {						\
+	ASSERT((smodp) != NULL);					\
+	DTRACE_PROBE1(smodinfo__inc__ref, struct smod_info *, (smodp));	\
+	atomic_inc_uint(&(smodp)->smod_refcnt);				\
+}
+
+/*
+ * Decreace the socket module entry reference count.
+ * When no one mapping to the entry, we try to unload the module from the
+ * kernel. If the module can't unload, just leave the module entry with
+ * a zero refcnt.
+ */
+#define	SMOD_DEC_REF(sp, smodp) {					\
+	ASSERT((smodp) != NULL);					\
+	ASSERT((smodp)->smod_refcnt != 0);				\
+	atomic_dec_uint(&(smodp)->smod_refcnt);				\
+	/*								\
+	 * No need to atomically check the return value because the	\
+	 * socket module framework will verify that no one is using	\
+	 * the module before unloading. Worst thing that can happen	\
+	 * here is multiple calls to mod_remove_by_name(), which is OK.	\
+	 */								\
+	if ((smodp)->smod_refcnt == 0)					\
+		(void) mod_remove_by_name((sp)->sp_smod_name);		\
+}
+
+/* Increase the reference count */
+#define	SOCKPARAMS_INC_REF(sp) {					\
+	ASSERT((sp) != NULL);						\
+	DTRACE_PROBE1(sockparams__inc__ref, struct sockparams *, (sp));	\
+	mutex_enter(&(sp)->sp_lock);					\
+	(sp)->sp_refcnt++;						\
+	ASSERT((sp)->sp_refcnt != 0);					\
+	mutex_exit(&(sp)->sp_lock);					\
+}
+
+/*
+ * Decrease the reference count.
+ *
+ * If the sockparams is ephemeral, then the thread dropping the last ref
+ * count will destroy the entry.
+ */
+#define	SOCKPARAMS_DEC_REF(sp) {					\
+	ASSERT((sp) != NULL);						\
+	DTRACE_PROBE1(sockparams__dec__ref, struct sockparams *, (sp));	\
+	mutex_enter(&(sp)->sp_lock);					\
+	ASSERT((sp)->sp_refcnt > 0);					\
+	if ((sp)->sp_refcnt == 1) {					\
+		if ((sp)->sp_flags & SOCKPARAMS_EPHEMERAL) {		\
+			mutex_exit(&(sp)->sp_lock);			\
+			sockparams_ephemeral_drop_last_ref((sp));	\
+		} else {						\
+			(sp)->sp_refcnt--;				\
+			if ((sp)->sp_smod_info != NULL)			\
+				SMOD_DEC_REF(sp, (sp)->sp_smod_info);	\
+			(sp)->sp_smod_info = NULL;			\
+			mutex_exit(&(sp)->sp_lock);			\
+		}							\
+	} else {							\
+		(sp)->sp_refcnt--;					\
+		mutex_exit(&(sp)->sp_lock);				\
+	}								\
+}
 
 /*
  * Used to traverse the list of AF_UNIX sockets to construct the kstat
@@ -490,49 +590,71 @@ struct sendfile_queue {
 
 /* Socket network operations switch */
 struct sonodeops {
-	int	(*sop_accept)(struct sonode *, int, struct sonode **);
-	int	(*sop_bind)(struct sonode *, struct sockaddr *, socklen_t,
+	int 	(*sop_init)(struct sonode *, struct sonode *, cred_t *,
 		    int);
-	int	(*sop_listen)(struct sonode *, int);
+	int	(*sop_accept)(struct sonode *, int, cred_t *, struct sonode **);
+	int	(*sop_bind)(struct sonode *, struct sockaddr *, socklen_t,
+		    int, cred_t *);
+	int	(*sop_listen)(struct sonode *, int, cred_t *);
 	int	(*sop_connect)(struct sonode *, const struct sockaddr *,
-		    socklen_t, int, int);
+		    socklen_t, int, int, cred_t *);
 	int	(*sop_recvmsg)(struct sonode *, struct msghdr *,
-		    struct uio *);
+		    struct uio *, cred_t *);
 	int	(*sop_sendmsg)(struct sonode *, struct msghdr *,
-		    struct uio *);
-	int	(*sop_getpeername)(struct sonode *);
-	int	(*sop_getsockname)(struct sonode *);
-	int	(*sop_shutdown)(struct sonode *, int);
+		    struct uio *, cred_t *);
+	int	(*sop_sendmblk)(struct sonode *, struct msghdr *, int,
+		    cred_t *, mblk_t **);
+	int	(*sop_getpeername)(struct sonode *, struct sockaddr *,
+		    socklen_t *, boolean_t, cred_t *);
+	int	(*sop_getsockname)(struct sonode *, struct sockaddr *,
+		    socklen_t *, cred_t *);
+	int	(*sop_shutdown)(struct sonode *, int, cred_t *);
 	int	(*sop_getsockopt)(struct sonode *, int, int, void *,
-		    socklen_t *, int);
+		    socklen_t *, int, cred_t *);
 	int 	(*sop_setsockopt)(struct sonode *, int, int, const void *,
-		    socklen_t);
+		    socklen_t, cred_t *);
+	int 	(*sop_ioctl)(struct sonode *, int, intptr_t, int,
+		    cred_t *, int32_t *);
+	int 	(*sop_poll)(struct sonode *, short, int, short *,
+		    struct pollhead **);
+	int 	(*sop_close)(struct sonode *, int, cred_t *);
 };
 
-#define	SOP_ACCEPT(so, fflag, nsop)	\
-	((so)->so_ops->sop_accept((so), (fflag), (nsop)))
-#define	SOP_BIND(so, name, namelen, flags)	\
-	((so)->so_ops->sop_bind((so), (name), (namelen), (flags)))
-#define	SOP_LISTEN(so, backlog)	\
-	((so)->so_ops->sop_listen((so), (backlog)))
-#define	SOP_CONNECT(so, name, namelen, fflag, flags)	\
-	((so)->so_ops->sop_connect((so), (name), (namelen), (fflag), (flags)))
-#define	SOP_RECVMSG(so, msg, uiop)	\
-	((so)->so_ops->sop_recvmsg((so), (msg), (uiop)))
-#define	SOP_SENDMSG(so, msg, uiop)	\
-	((so)->so_ops->sop_sendmsg((so), (msg), (uiop)))
-#define	SOP_GETPEERNAME(so)	\
-	((so)->so_ops->sop_getpeername((so)))
-#define	SOP_GETSOCKNAME(so)	\
-	((so)->so_ops->sop_getsockname((so)))
-#define	SOP_SHUTDOWN(so, how)	\
-	((so)->so_ops->sop_shutdown((so), (how)))
-#define	SOP_GETSOCKOPT(so, level, optionname, optval, optlenp, flags)	\
+#define	SOP_INIT(so, flag, cr, flags)	\
+	((so)->so_ops->sop_init((so), (flag), (cr), (flags)))
+#define	SOP_ACCEPT(so, fflag, cr, nsop)	\
+	((so)->so_ops->sop_accept((so), (fflag), (cr), (nsop)))
+#define	SOP_BIND(so, name, namelen, flags, cr)	\
+	((so)->so_ops->sop_bind((so), (name), (namelen), (flags), (cr)))
+#define	SOP_LISTEN(so, backlog, cr)	\
+	((so)->so_ops->sop_listen((so), (backlog), (cr)))
+#define	SOP_CONNECT(so, name, namelen, fflag, flags, cr)	\
+	((so)->so_ops->sop_connect((so), (name), (namelen), (fflag), (flags), \
+	(cr)))
+#define	SOP_RECVMSG(so, msg, uiop, cr)	\
+	((so)->so_ops->sop_recvmsg((so), (msg), (uiop), (cr)))
+#define	SOP_SENDMSG(so, msg, uiop, cr)	\
+	((so)->so_ops->sop_sendmsg((so), (msg), (uiop), (cr)))
+#define	SOP_SENDMBLK(so, msg, size, cr, mpp)	\
+	((so)->so_ops->sop_sendmblk((so), (msg), (size), (cr), (mpp)))
+#define	SOP_GETPEERNAME(so, addr, addrlen, accept, cr)	\
+	((so)->so_ops->sop_getpeername((so), (addr), (addrlen), (accept), (cr)))
+#define	SOP_GETSOCKNAME(so, addr, addrlen, cr)	\
+	((so)->so_ops->sop_getsockname((so), (addr), (addrlen), (cr)))
+#define	SOP_SHUTDOWN(so, how, cr)	\
+	((so)->so_ops->sop_shutdown((so), (how), (cr)))
+#define	SOP_GETSOCKOPT(so, level, optionname, optval, optlenp, flags, cr) \
 	((so)->so_ops->sop_getsockopt((so), (level), (optionname),	\
-	    (optval), (optlenp), (flags)))
-#define	SOP_SETSOCKOPT(so, level, optionname, optval, optlen)		\
+	    (optval), (optlenp), (flags), (cr)))
+#define	SOP_SETSOCKOPT(so, level, optionname, optval, optlen, cr)	\
 	((so)->so_ops->sop_setsockopt((so), (level), (optionname),	\
-	    (optval), (optlen)))
+	    (optval), (optlen), (cr)))
+#define	SOP_IOCTL(so, cmd, arg, mode, cr, rvalp)	\
+	((so)->so_ops->sop_ioctl((so), (cmd), (arg), (mode), (cr), (rvalp)))
+#define	SOP_POLL(so, events, anyyet, reventsp, phpp) \
+	((so)->so_ops->sop_poll((so), (events), (anyyet), (reventsp), (phpp)))
+#define	SOP_CLOSE(so, flag, cr)	\
+	((so)->so_ops->sop_close((so), (flag), (cr)))
 
 #endif /* defined(_KERNEL) || defined(_KMEMUSER) */
 
@@ -544,6 +666,8 @@ struct sonodeops {
 #define	ROUNDUP_cmsglen(len) \
 	(((len) + _CMSG_HDR_ALIGNMENT - 1) & ~(_CMSG_HDR_ALIGNMENT - 1))
 
+#define	IS_NON_STREAM_SOCK(vp) \
+	((vp)->v_type == VSOCK && (vp)->v_stream == NULL)
 /*
  * Macros that operate on struct cmsghdr.
  * Used in parsing msg_control.
@@ -686,10 +810,8 @@ extern int sockprinterr;
 #endif /* defined(DEBUG) */
 
 extern struct vfsops			sock_vfsops;
-extern struct vnodeops			*socktpi_vnodeops;
-extern const struct fs_operation_def	socktpi_vnodeops_template[];
-
-extern sonodeops_t			sotpi_sonodeops;
+extern struct vnodeops			*socket_vnodeops;
+extern const struct fs_operation_def	socket_vnodeops_template[];
 
 extern dev_t				sockdev;
 
@@ -700,20 +822,10 @@ extern int	sock_getmsg(vnode_t *, struct strbuf *, struct strbuf *,
 			uchar_t *, int *, int, rval_t *);
 extern int	sock_putmsg(vnode_t *, struct strbuf *, struct strbuf *,
 			uchar_t, int, int);
-struct sonode	*sotpi_create(vnode_t *, int, int, int, int, struct sonode *,
-			int *);
-extern int	socktpi_open(struct vnode **, int, struct cred *,
-			caller_context_t *);
-extern int	so_sock2stream(struct sonode *);
-extern void	so_stream2sock(struct sonode *);
+extern int	sogetvp(char *, vnode_t **, int);
 extern int	sockinit(int, char *);
-extern struct vnode
-		*makesockvp(struct vnode *, int, int, int);
-extern void	sockfree(struct sonode *);
-extern void	so_update_attrs(struct sonode *, int);
-extern int	soconfig(int, int, int,	char *, int);
-extern struct vnode
-		*solookup(int, int, int, char *, int *);
+extern int	soconfig(int, int, int,	char *, int, char *);
+extern int	solookup(int, int, int, struct sockparams **);
 extern void	so_lock_single(struct sonode *);
 extern void	so_unlock_single(struct sonode *, int);
 extern int	so_lock_read(struct sonode *, int);
@@ -723,10 +835,6 @@ extern void	*sogetoff(mblk_t *, t_uscalar_t, t_uscalar_t, uint_t);
 extern void	so_getopt_srcaddr(void *, t_uscalar_t,
 			void **, t_uscalar_t *);
 extern int	so_getopt_unix_close(void *, t_uscalar_t);
-extern int	so_addr_verify(struct sonode *, const struct sockaddr *,
-			socklen_t);
-extern int	so_ux_addr_xlate(struct sonode *, struct sockaddr *,
-			socklen_t, int, void **, socklen_t *);
 extern void	fdbuf_free(struct fdbuf *);
 extern mblk_t	*fdbuf_allocmsg(int, struct fdbuf *);
 extern int	fdbuf_create(void *, int, struct fdbuf **);
@@ -744,55 +852,13 @@ extern void	soisdisconnected(struct sonode *, int);
 extern void	socantsendmore(struct sonode *);
 extern void	socantrcvmore(struct sonode *);
 extern void	soseterror(struct sonode *, int);
-extern int	sogeterr(struct sonode *);
-extern int	sogetrderr(vnode_t *, int, int *);
-extern int	sogetwrerr(vnode_t *, int, int *);
-extern void	so_unix_close(struct sonode *);
-extern mblk_t	*soallocproto(size_t, int);
-extern mblk_t	*soallocproto1(const void *, ssize_t, ssize_t, int);
-extern void	soappendmsg(mblk_t *, const void *, ssize_t);
-extern mblk_t	*soallocproto2(const void *, ssize_t, const void *, ssize_t,
-			ssize_t, int);
-extern mblk_t	*soallocproto3(const void *, ssize_t, const void *, ssize_t,
-			const void *, ssize_t, ssize_t, int);
-extern int	sowaitprim(struct sonode *, t_scalar_t, t_scalar_t,
-			t_uscalar_t, mblk_t **, clock_t);
-extern int	sowaitokack(struct sonode *, t_scalar_t);
-extern int	sowaitack(struct sonode *, mblk_t **, clock_t);
-extern void	soqueueack(struct sonode *, mblk_t *);
-extern int	sowaitconnind(struct sonode *, int, mblk_t **);
-extern void	soqueueconnind(struct sonode *, mblk_t *);
-extern int	soflushconnind(struct sonode *, t_scalar_t);
-extern void	so_drain_discon_ind(struct sonode *);
-extern void	so_flush_discon_ind(struct sonode *);
+extern int	sogeterr(struct sonode *, boolean_t);
 extern int	sowaitconnected(struct sonode *, int, int);
 
-extern int	sostream_direct(struct sonode *, struct uio *,
-		    mblk_t *, cred_t *);
-extern int	sosend_dgram(struct sonode *, struct sockaddr *,
-		    socklen_t, struct uio *, int);
-extern int	sosend_svc(struct sonode *, struct uio *, t_scalar_t, int, int);
-extern void	so_installhooks(struct sonode *);
-extern int	so_strinit(struct sonode *, struct sonode *);
-extern int	sotpi_recvmsg(struct sonode *, struct nmsghdr *,
-		    struct uio *);
-extern int	sotpi_getpeername(struct sonode *);
-extern int	sotpi_getsockopt(struct sonode *, int, int, void *,
-		    socklen_t *, int);
-extern int	sotpi_setsockopt(struct sonode *, int, int, const void *,
-		    socklen_t);
-extern int	socktpi_ioctl(struct vnode *, int, intptr_t, int,
-		    struct cred *, int *, caller_context_t *);
-extern int	sodisconnect(struct sonode *, t_scalar_t, int);
 extern ssize_t	soreadfile(file_t *, uchar_t *, u_offset_t, int *, size_t);
-extern int	so_set_asyncsigs(vnode_t *, pid_t, int, int, cred_t *);
-extern int	so_set_events(struct sonode *, vnode_t *, cred_t *);
-extern int	so_flip_async(struct sonode *, vnode_t *, int, cred_t *);
-extern int	so_set_siggrp(struct sonode *, vnode_t *, pid_t, int, cred_t *);
 extern void	*sock_kstat_init(zoneid_t);
 extern void	sock_kstat_fini(zoneid_t, void *);
 extern struct sonode *getsonode(int, int *, file_t **);
-
 /*
  * Function wrappers (mostly around the sonode switch) for
  * backward compatibility.
@@ -805,44 +871,18 @@ extern int	soconnect(struct sonode *, const struct sockaddr *, socklen_t,
 		    int, int);
 extern int	sorecvmsg(struct sonode *, struct nmsghdr *, struct uio *);
 extern int	sosendmsg(struct sonode *, struct nmsghdr *, struct uio *);
-extern int	sogetpeername(struct sonode *);
-extern int	sogetsockname(struct sonode *);
 extern int	soshutdown(struct sonode *, int);
 extern int	sogetsockopt(struct sonode *, int, int, void *, socklen_t *,
 		    int);
 extern int	sosetsockopt(struct sonode *, int, int, const void *,
 		    t_uscalar_t);
 
-extern struct sonode	*socreate(vnode_t *, int, int, int, int,
-			    struct sonode *, int *);
+extern struct sonode	*socreate(struct sockparams *, int, int, int, int,
+			    int *);
 
 extern int	so_copyin(const void *, void *, size_t, int);
 extern int	so_copyout(const void *, void *, size_t, int);
 
-extern int	socktpi_access(struct vnode *, int, int, struct cred *,
-		    caller_context_t *);
-extern int	socktpi_fid(struct vnode *, struct fid *, caller_context_t *);
-extern int	socktpi_fsync(struct vnode *, int, struct cred *,
-		    caller_context_t *);
-extern int	socktpi_getattr(struct vnode *, struct vattr *, int,
-		    struct cred *, caller_context_t *);
-extern int	socktpi_seek(struct vnode *, offset_t, offset_t *,
-		    caller_context_t *);
-extern int	socktpi_setattr(struct vnode *, struct vattr *, int,
-		    struct cred *, caller_context_t *);
-extern int	socktpi_setfl(vnode_t *, int, int, cred_t *,
-		    caller_context_t *);
-
-/* SCTP sockfs */
-extern struct sonode	*sosctp_create(vnode_t *, int, int, int, int,
-			    struct sonode *, int *);
-extern int sosctp_init(void);
-
-/* SDP sockfs */
-extern struct sonode    *sosdp_create(vnode_t *, int, int, int, int,
-			    struct sonode *, int *);
-extern int sosdp_init(void);
-
 #endif
 
 /*
@@ -865,9 +905,11 @@ struct sockinfo {
 	uint16_t	si_faddr_family;
 	char		si_laddr_sun_path[MAXPATHLEN + 1]; /* NULL terminated */
 	char		si_faddr_sun_path[MAXPATHLEN + 1];
+	boolean_t	si_faddr_noxlate;
 	zoneid_t	si_szoneid;
 };
 
+#define	SOCKMOD_PATH	"socketmod"	/* dir where sockmods are stored */
 
 #ifdef	__cplusplus
 }