usr/src/uts/common/inet diff (gate to joyent)

gistfile1.diff

Only in illumos-joyent/usr/src/uts/common/inet: inet_hash.h
diff -ur illumos-gate/usr/src/uts/common/inet/ip/ip_attr.c illumos-joyent/usr/src/uts/common/inet/ip/ip_attr.c
--- illumos-gate/usr/src/uts/common/inet/ip/ip_attr.c	2015-05-26 22:07:54.760515401 +0000
+++ illumos-joyent/usr/src/uts/common/inet/ip/ip_attr.c	2015-05-26 22:08:56.031054703 +0000
@@ -909,6 +909,11 @@
 	 */
 	if (ixa->ixa_free_flags & IXA_FREE_CRED)
 		crhold(ixa->ixa_cred);
+
+	/*
+	 * There is no cleanup in progress on this new copy.
+	 */
+	ixa->ixa_tcpcleanup = IXATC_IDLE;
 }
 
 /*
diff -ur illumos-gate/usr/src/uts/common/inet/ip/ip_if.c illumos-joyent/usr/src/uts/common/inet/ip/ip_if.c
--- illumos-gate/usr/src/uts/common/inet/ip/ip_if.c	2015-05-26 22:07:54.768694777 +0000
+++ illumos-joyent/usr/src/uts/common/inet/ip/ip_if.c	2015-05-26 22:08:56.041157550 +0000
@@ -25,6 +25,9 @@
  * Copyright 2013 Joyent, Inc.
  * Copyright (c) 2014, OmniTI Computer Consulting, Inc. All rights reserved.
  */
+/*
+ * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ */
 
 /*
  * This file contains the interface control functions for IP.
diff -ur illumos-gate/usr/src/uts/common/inet/ip/ip_squeue.c illumos-joyent/usr/src/uts/common/inet/ip/ip_squeue.c
--- illumos-gate/usr/src/uts/common/inet/ip/ip_squeue.c	2015-05-26 22:07:54.776765157 +0000
+++ illumos-joyent/usr/src/uts/common/inet/ip/ip_squeue.c	2015-05-26 22:08:56.050809608 +0000
@@ -163,7 +163,7 @@
 {
 	squeue_t *sqp;
 
-	sqp = squeue_create(ip_squeue_worker_wait, pri);
+	sqp = squeue_create(ip_squeue_worker_wait, pri, B_TRUE);
 	ASSERT(sqp != NULL);
 	if (ip_squeue_create_callback != NULL)
 		ip_squeue_create_callback(sqp);
diff -ur illumos-gate/usr/src/uts/common/inet/ip/ip_tunables.c illumos-joyent/usr/src/uts/common/inet/ip/ip_tunables.c
--- illumos-gate/usr/src/uts/common/inet/ip/ip_tunables.c	2015-05-26 22:07:54.777275286 +0000
+++ illumos-joyent/usr/src/uts/common/inet/ip/ip_tunables.c	2015-05-26 22:08:56.051415067 +0000
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  * Copyright (c) 2013 by Delphix. All rights reserved.
  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  */
diff -ur illumos-gate/usr/src/uts/common/inet/ip/ipsecesp.c illumos-joyent/usr/src/uts/common/inet/ip/ipsecesp.c
--- illumos-gate/usr/src/uts/common/inet/ip/ipsecesp.c	2015-05-26 22:07:54.782342290 +0000
+++ illumos-joyent/usr/src/uts/common/inet/ip/ipsecesp.c	2015-05-26 22:08:56.057572680 +0000
@@ -234,8 +234,7 @@
 {
 	espstack->esp_ksp = kstat_create_netstack("ipsecesp", 0, "esp_stat",
 	    "net", KSTAT_TYPE_NAMED,
-	    sizeof (esp_kstats_t) / sizeof (kstat_named_t),
-	    KSTAT_FLAG_PERSISTENT, stackid);
+	    sizeof (esp_kstats_t) / sizeof (kstat_named_t), 0, stackid);
 
 	if (espstack->esp_ksp == NULL || espstack->esp_ksp->ks_data == NULL)
 		return (B_FALSE);
diff -ur illumos-gate/usr/src/uts/common/inet/ipf/ip_fil_solaris.c illumos-joyent/usr/src/uts/common/inet/ipf/ip_fil_solaris.c
--- illumos-gate/usr/src/uts/common/inet/ipf/ip_fil_solaris.c	2015-05-26 22:07:54.799744331 +0000
+++ illumos-joyent/usr/src/uts/common/inet/ipf/ip_fil_solaris.c	2015-05-26 22:08:56.078426896 +0000
@@ -83,6 +83,14 @@
 static	int	ipf_hook6_loop_in __P((hook_event_token_t, hook_data_t,
     void *));
 static	int     ipf_hook6 __P((hook_data_t, int, int, void *));
+static	int	ipf_hookvndl3v4_in __P((hook_event_token_t, hook_data_t,
+    void *));
+static	int	ipf_hookvndl3v6_in __P((hook_event_token_t, hook_data_t,
+    void *));
+static	int	ipf_hookvndl3v4_out __P((hook_event_token_t, hook_data_t,
+    void *));
+static	int	ipf_hookvndl3v6_out __P((hook_event_token_t, hook_data_t,
+    void *));
 extern	int	ipf_geniter __P((ipftoken_t *, ipfgeniter_t *, ipf_stack_t *));
 extern	int	ipf_frruleiter __P((void *, int, void *, ipf_stack_t *));
 
@@ -152,6 +160,16 @@
 char *hook6_loop_out = 		"ipfilter_hook6_loop_out";
 char *hook6_loop_out_gz = 	"ipfilter_hook6_loop_out_gz";
 
+/* vnd IPv4/v6 hook names */
+char *hook4_vnd_in =		"ipfilter_hookvndl3v4_in";
+char *hook4_vnd_in_gz =		"ipfilter_hookvndl3v4_in_gz";
+char *hook6_vnd_in =		"ipfilter_hookvndl3v6_in";
+char *hook6_vnd_in_gz =		"ipfilter_hookvndl3v6_in_gz";
+char *hook4_vnd_out =		"ipfilter_hookvndl3v4_out";
+char *hook4_vnd_out_gz =	"ipfilter_hookvndl3v4_out_gz";
+char *hook6_vnd_out =		"ipfilter_hookvndl3v6_out";
+char *hook6_vnd_out_gz =	"ipfilter_hookvndl3v6_out_gz";
+
 /* ------------------------------------------------------------------------ */
 /* Function:    ipldetach                                                   */
 /* Returns:     int - 0 == success, else error.                             */
@@ -248,6 +266,31 @@
 		ifs->ifs_ipf_ipv4 = NULL;
 	}
 
+	/*
+	 * Remove VND hooks
+	 */
+	if (ifs->ifs_ipf_vndl3v4 != NULL) {
+		UNDO_HOOK(ifs_ipf_vndl3v4, ifs_hookvndl3v4_physical_in,
+		    NH_PHYSICAL_IN, ifs_ipfhookvndl3v4_in);
+		UNDO_HOOK(ifs_ipf_vndl3v4, ifs_hookvndl3v4_physical_out,
+		    NH_PHYSICAL_OUT, ifs_ipfhookvndl3v4_out);
+
+		if (net_protocol_release(ifs->ifs_ipf_vndl3v4) != 0)
+			goto detach_failed;
+		ifs->ifs_ipf_vndl3v4 = NULL;
+	}
+
+	if (ifs->ifs_ipf_vndl3v6 != NULL) {
+		UNDO_HOOK(ifs_ipf_vndl3v6, ifs_hookvndl3v6_physical_in,
+		    NH_PHYSICAL_IN, ifs_ipfhookvndl3v6_in);
+		UNDO_HOOK(ifs_ipf_vndl3v6, ifs_hookvndl3v6_physical_out,
+		    NH_PHYSICAL_OUT, ifs_ipfhookvndl3v6_out);
+
+		if (net_protocol_release(ifs->ifs_ipf_vndl3v6) != 0)
+			goto detach_failed;
+		ifs->ifs_ipf_vndl3v6 = NULL;
+	}
+
 #undef UNDO_HOOK
 
 #ifdef	IPFDEBUG
@@ -445,6 +488,48 @@
 	}
 
 	/*
+	 * Add VND INET hooks
+	 */
+	ifs->ifs_ipf_vndl3v4 = net_protocol_lookup(id, NHF_VND_INET);
+	if (ifs->ifs_ipf_vndl3v4 == NULL)
+		goto hookup_failed;
+
+	HOOK_INIT_GZ_BEFORE(ifs->ifs_ipfhookvndl3v4_in, ipf_hookvndl3v4_in,
+	    hook4_vnd_in, hook4_vnd_in_gz, ifs);
+	HOOK_INIT_GZ_AFTER(ifs->ifs_ipfhookvndl3v4_out, ipf_hookvndl3v4_out,
+	    hook4_vnd_out, hook4_vnd_out_gz, ifs);
+	ifs->ifs_hookvndl3v4_physical_in = (net_hook_register(ifs->ifs_ipf_vndl3v4,
+	    NH_PHYSICAL_IN, ifs->ifs_ipfhookvndl3v4_in) == 0);
+	if (!ifs->ifs_hookvndl3v4_physical_in)
+		goto hookup_failed;
+
+	ifs->ifs_hookvndl3v4_physical_out = (net_hook_register(ifs->ifs_ipf_vndl3v4,
+	    NH_PHYSICAL_OUT, ifs->ifs_ipfhookvndl3v4_out) == 0);
+	if (!ifs->ifs_hookvndl3v4_physical_out)
+		goto hookup_failed;
+
+
+	/*
+	 * VND INET6 hooks
+	 */
+	ifs->ifs_ipf_vndl3v6 = net_protocol_lookup(id, NHF_VND_INET6);
+	if (ifs->ifs_ipf_vndl3v6 == NULL)
+		goto hookup_failed;
+
+	HOOK_INIT_GZ_BEFORE(ifs->ifs_ipfhookvndl3v6_in, ipf_hookvndl3v6_in,
+	    hook6_vnd_in, hook6_vnd_in_gz, ifs);
+	HOOK_INIT_GZ_AFTER(ifs->ifs_ipfhookvndl3v6_out, ipf_hookvndl3v6_out,
+	    hook6_vnd_out, hook6_vnd_out_gz, ifs);
+	ifs->ifs_hookvndl3v6_physical_in = (net_hook_register(ifs->ifs_ipf_vndl3v6,
+	    NH_PHYSICAL_IN, ifs->ifs_ipfhookvndl3v6_in) == 0);
+	if (!ifs->ifs_hookvndl3v6_physical_in)
+		goto hookup_failed;
+
+	ifs->ifs_hookvndl3v6_physical_out = (net_hook_register(ifs->ifs_ipf_vndl3v6,
+	    NH_PHYSICAL_OUT, ifs->ifs_ipfhookvndl3v6_out) == 0);
+	if (!ifs->ifs_hookvndl3v6_physical_out)
+		goto hookup_failed;
+	/*
 	 * Reacquire ipf_global, now it is safe.
 	 */
 	WRITE_ENTER(&ifs->ifs_ipf_global);
@@ -1011,7 +1096,6 @@
 		return ENXIO;
 	unit = isp->ipfs_minor;
 
-
         /*
 	 * ipf_find_stack returns with a read lock on ifs_ipf_global
 	 */
@@ -2045,6 +2129,42 @@
 }
 
 /* ------------------------------------------------------------------------ */
+/* Function:    ipf_hookvndl3_in					    */
+/* Returns:     int - 0 == packet ok, else problem, free packet if not done */
+/* Parameters:  event(I)     - pointer to event                             */
+/*              info(I)      - pointer to hook information for firewalling  */
+/*                                                                          */
+/* The vnd hooks are private hooks to ON. They represents a layer 2         */
+/* datapath generally used to implement virtual machines. The driver sends  */
+/* along L3 packets of either type IP or IPv6. The ethertype to distinguish */
+/* them is in the upper 16 bits while the remaining bits are the            */
+/* traditional packet hook flags.                                           */
+/*                                                                          */
+/* They end up calling the appropriate traditional ip hooks.                */
+/* ------------------------------------------------------------------------ */
+/*ARGSUSED*/
+int ipf_hookvndl3v4_in(hook_event_token_t token, hook_data_t info, void *arg)
+{
+	return ipf_hook4_in(token, info, arg);
+}
+
+int ipf_hookvndl3v6_in(hook_event_token_t token, hook_data_t info, void *arg)
+{
+	return ipf_hook6_in(token, info, arg);
+}
+
+/*ARGSUSED*/
+int ipf_hookvndl3v4_out(hook_event_token_t token, hook_data_t info, void *arg)
+{
+	return ipf_hook4_out(token, info, arg);
+}
+
+int ipf_hookvndl3v6_out(hook_event_token_t token, hook_data_t info, void *arg)
+{
+	return ipf_hook6_out(token, info, arg);
+}
+
+/* ------------------------------------------------------------------------ */
 /* Function:    ipf_hook4_loop_in                                           */
 /* Returns:     int - 0 == packet ok, else problem, free packet if not done */
 /* Parameters:  event(I)     - pointer to event                             */
diff -ur illumos-gate/usr/src/uts/common/inet/ipf/ipf.conf illumos-joyent/usr/src/uts/common/inet/ipf/ipf.conf
--- illumos-gate/usr/src/uts/common/inet/ipf/ipf.conf	2015-05-26 22:07:54.805361628 +0000
+++ illumos-joyent/usr/src/uts/common/inet/ipf/ipf.conf	2015-05-26 22:08:56.085109881 +0000
@@ -1,3 +1,8 @@
 #
 #
 name="ipf" parent="pseudo" instance=0;
+
+# Increase the state table limits. fr_statemax should be ~70% of fr_statesize,
+# and both should be prime numbers
+fr_statesize=151007;
+fr_statemax=113279;
diff -ur illumos-gate/usr/src/uts/common/inet/ipf/netinet/ipf_stack.h illumos-joyent/usr/src/uts/common/inet/ipf/netinet/ipf_stack.h
--- illumos-gate/usr/src/uts/common/inet/ipf/netinet/ipf_stack.h	2015-05-26 22:07:54.810795313 +0000
+++ illumos-joyent/usr/src/uts/common/inet/ipf/netinet/ipf_stack.h	2015-05-26 22:08:56.091596242 +0000
@@ -125,6 +125,10 @@
 	hook_t		*ifs_ipfhook6_loop_in;
 	hook_t		*ifs_ipfhook6_loop_out;
 	hook_t		*ifs_ipfhook6_nicevents;
+	hook_t		*ifs_ipfhookvndl3v4_in;
+	hook_t		*ifs_ipfhookvndl3v6_in;
+	hook_t		*ifs_ipfhookvndl3v4_out;
+	hook_t		*ifs_ipfhookvndl3v6_out;
 
 	/* flags to indicate whether hooks are registered. */
 	boolean_t	ifs_hook4_physical_in;
@@ -137,10 +141,16 @@
 	boolean_t	ifs_hook6_nic_events;
 	boolean_t	ifs_hook6_loopback_in;
 	boolean_t	ifs_hook6_loopback_out;
+	boolean_t	ifs_hookvndl3v4_physical_in;
+	boolean_t	ifs_hookvndl3v6_physical_in;
+	boolean_t	ifs_hookvndl3v4_physical_out;
+	boolean_t	ifs_hookvndl3v6_physical_out;
 
 	int		ifs_ipf_loopback;
 	net_handle_t	ifs_ipf_ipv4;
 	net_handle_t	ifs_ipf_ipv6;
+	net_handle_t	ifs_ipf_vndl3v4;
+	net_handle_t	ifs_ipf_vndl3v6;
 
 	/* ip_auth.c */
 	int			ifs_fr_authsize;
diff -ur illumos-gate/usr/src/uts/common/inet/ipf/solaris.c illumos-joyent/usr/src/uts/common/inet/ipf/solaris.c
--- illumos-gate/usr/src/uts/common/inet/ipf/solaris.c	2015-05-26 22:07:54.811581187 +0000
+++ illumos-joyent/usr/src/uts/common/inet/ipf/solaris.c	2015-05-26 22:08:56.092453707 +0000
@@ -625,7 +625,6 @@
 /*
  * Destroy things for ipf for one stack.
  */
-/* ARGSUSED */
 static void
 ipf_stack_destroy_one(const netid_t id, ipf_stack_t *ifs)
 {
Only in illumos-joyent/usr/src/uts/common/inet/sockmods: datafilt.c
diff -ur illumos-gate/usr/src/uts/common/inet/sockmods/socksctp.c illumos-joyent/usr/src/uts/common/inet/sockmods/socksctp.c
--- illumos-gate/usr/src/uts/common/inet/sockmods/socksctp.c	2015-05-26 22:07:54.836946288 +0000
+++ illumos-joyent/usr/src/uts/common/inet/sockmods/socksctp.c	2015-05-26 22:08:56.122541817 +0000
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015 Joyent, Inc.  All rights reserved.
  */
 
 #include <sys/types.h>
@@ -403,7 +404,6 @@
 	}
 
 	if (name == NULL || namelen == 0) {
-		mutex_exit(&so->so_lock);
 		error = EINVAL;
 		eprintsoline(so, error);
 		goto done;
diff -ur illumos-gate/usr/src/uts/common/inet/squeue.c illumos-joyent/usr/src/uts/common/inet/squeue.c
--- illumos-gate/usr/src/uts/common/inet/squeue.c	2015-05-26 22:07:54.838608054 +0000
+++ illumos-joyent/usr/src/uts/common/inet/squeue.c	2015-05-26 22:08:56.124547236 +0000
@@ -23,7 +23,7 @@
  */
 
 /*
- * Copyright 2012 Joyent, Inc.  All rights reserved.
+ * Copyright (c) 2014 Joyent, Inc.  All rights reserved.
  */
 
 /*
@@ -61,6 +61,10 @@
  * connection are processed on that squeue. The connection ("conn") to
  * squeue mapping is stored in "conn_t" member "conn_sqp".
  *
+ * If the squeue is not related to TCP/IP, then the value of sqp->sq_isip is
+ * false and it will not have an associated conn_t, which means many aspects of
+ * the system, such as polling and swtiching squeues will not be used.
+ *
  * Since the processing of the connection cuts across multiple layers
  * but still allows packets for different connnection to be processed on
  * other CPU/squeues, squeues are also termed as "Vertical Perimeter" or
@@ -244,7 +248,7 @@
 
 /* ARGSUSED */
 squeue_t *
-squeue_create(clock_t wait, pri_t pri)
+squeue_create(clock_t wait, pri_t pri, boolean_t isip)
 {
 	squeue_t *sqp = kmem_cache_alloc(squeue_cache, KM_SLEEP);
 
@@ -260,11 +264,36 @@
 
 	sqp->sq_enter = squeue_enter;
 	sqp->sq_drain = squeue_drain;
+	sqp->sq_isip = isip;
 
 	return (sqp);
 }
 
 /*
+ * We need to kill the threads and then clean up. We should VERIFY that
+ * polling is disabled so we don't have to worry about disassociating from
+ * MAC/IP/etc.
+ */
+void
+squeue_destroy(squeue_t *sqp)
+{
+	kt_did_t worker, poll;
+	mutex_enter(&sqp->sq_lock);
+	VERIFY(!(sqp->sq_state & (SQS_POLL_THR_QUIESCED |
+	    SQS_POLL_QUIESCE_DONE | SQS_PAUSE | SQS_EXIT)));
+	worker = sqp->sq_worker->t_did;
+	poll = sqp->sq_poll_thr->t_did;
+	sqp->sq_state |= SQS_EXIT;
+	cv_signal(&sqp->sq_poll_cv);
+	cv_signal(&sqp->sq_worker_cv);
+	mutex_exit(&sqp->sq_lock);
+
+	thread_join(poll);
+	thread_join(worker);
+	kmem_cache_free(squeue_cache, sqp);
+}
+
+/*
  * Bind squeue worker thread to the specified CPU, given by CPU id.
  * If the CPU id  value is -1, bind the worker thread to the value
  * specified in sq_bind field. If a thread is already bound to a
@@ -475,18 +504,21 @@
 			 * Handle squeue switching. More details in the
 			 * block comment at the top of the file
 			 */
-			if (connp->conn_sqp == sqp) {
+			if (sqp->sq_isip == B_FALSE || connp->conn_sqp == sqp) {
 				SQUEUE_DBG_SET(sqp, mp, proc, connp,
 				    tag);
-				connp->conn_on_sqp = B_TRUE;
+				if (sqp->sq_isip == B_TRUE)
+					connp->conn_on_sqp = B_TRUE;
 				DTRACE_PROBE3(squeue__proc__start, squeue_t *,
 				    sqp, mblk_t *, mp, conn_t *, connp);
 				(*proc)(connp, mp, sqp, ira);
 				DTRACE_PROBE2(squeue__proc__end, squeue_t *,
 				    sqp, conn_t *, connp);
-				connp->conn_on_sqp = B_FALSE;
+				if (sqp->sq_isip == B_TRUE) {
+					connp->conn_on_sqp = B_FALSE;
+					CONN_DEC_REF(connp);
+				}
 				SQUEUE_DBG_CLEAR(sqp);
-				CONN_DEC_REF(connp);
 			} else {
 				SQUEUE_ENTER_ONE(connp->conn_sqp, mp, proc,
 				    connp, ira, SQ_FILL, SQTAG_SQUEUE_CHANGE);
@@ -513,7 +545,7 @@
 				return;
 			}
 		} else {
-			if (ira != NULL) {
+			if (sqp->sq_isip == B_TRUE && ira != NULL) {
 				mblk_t	*attrmp;
 
 				ASSERT(cnt == 1);
@@ -587,7 +619,8 @@
 		if (!(sqp->sq_state & SQS_REENTER) &&
 		    (process_flag != SQ_FILL) && (sqp->sq_first == NULL) &&
 		    (sqp->sq_run == curthread) && (cnt == 1) &&
-		    (connp->conn_on_sqp == B_FALSE)) {
+		    (sqp->sq_isip == B_FALSE ||
+		    connp->conn_on_sqp == B_FALSE)) {
 			sqp->sq_state |= SQS_REENTER;
 			mutex_exit(&sqp->sq_lock);
 
@@ -602,15 +635,21 @@
 			 * Handle squeue switching. More details in the
 			 * block comment at the top of the file
 			 */
-			if (connp->conn_sqp == sqp) {
-				connp->conn_on_sqp = B_TRUE;
+			if (sqp->sq_isip == B_FALSE || connp->conn_sqp == sqp) {
+				SQUEUE_DBG_SET(sqp, mp, proc, connp,
+				    tag);
+				if (sqp->sq_isip == B_TRUE)
+					connp->conn_on_sqp = B_TRUE;
 				DTRACE_PROBE3(squeue__proc__start, squeue_t *,
 				    sqp, mblk_t *, mp, conn_t *, connp);
 				(*proc)(connp, mp, sqp, ira);
 				DTRACE_PROBE2(squeue__proc__end, squeue_t *,
 				    sqp, conn_t *, connp);
-				connp->conn_on_sqp = B_FALSE;
-				CONN_DEC_REF(connp);
+				if (sqp->sq_isip == B_TRUE) {
+					connp->conn_on_sqp = B_FALSE;
+					CONN_DEC_REF(connp);
+				}
+				SQUEUE_DBG_CLEAR(sqp);
 			} else {
 				SQUEUE_ENTER_ONE(connp->conn_sqp, mp, proc,
 				    connp, ira, SQ_FILL, SQTAG_SQUEUE_CHANGE);
@@ -631,7 +670,7 @@
 #ifdef DEBUG
 		mp->b_tag = tag;
 #endif
-		if (ira != NULL) {
+		if (sqp->sq_isip && ira != NULL) {
 			mblk_t	*attrmp;
 
 			ASSERT(cnt == 1);
@@ -779,7 +818,7 @@
 		mp->b_prev = NULL;
 
 		/* Is there an ip_recv_attr_t to handle? */
-		if (ip_recv_attr_is_mblk(mp)) {
+		if (sqp->sq_isip == B_TRUE && ip_recv_attr_is_mblk(mp)) {
 			mblk_t	*attrmp = mp;
 
 			ASSERT(attrmp->b_cont != NULL);
@@ -804,20 +843,25 @@
 
 
 		/*
-		 * Handle squeue switching. More details in the
-		 * block comment at the top of the file
+		 * Handle squeue switching. More details in the block comment at
+		 * the top of the file. non-IP squeues cannot switch, as there
+		 * is no conn_t.
 		 */
-		if (connp->conn_sqp == sqp) {
+		if (sqp->sq_isip == B_FALSE || connp->conn_sqp == sqp) {
 			SQUEUE_DBG_SET(sqp, mp, proc, connp,
 			    mp->b_tag);
-			connp->conn_on_sqp = B_TRUE;
+			if (sqp->sq_isip == B_TRUE)
+				connp->conn_on_sqp = B_TRUE;
 			DTRACE_PROBE3(squeue__proc__start, squeue_t *,
 			    sqp, mblk_t *, mp, conn_t *, connp);
 			(*proc)(connp, mp, sqp, ira);
 			DTRACE_PROBE2(squeue__proc__end, squeue_t *,
 			    sqp, conn_t *, connp);
-			connp->conn_on_sqp = B_FALSE;
-			CONN_DEC_REF(connp);
+			if (sqp->sq_isip == B_TRUE) {
+				connp->conn_on_sqp = B_FALSE;
+				CONN_DEC_REF(connp);
+			}
+			SQUEUE_DBG_CLEAR(sqp);
 		} else {
 			SQUEUE_ENTER_ONE(connp->conn_sqp, mp, proc, connp, ira,
 			    SQ_FILL, SQTAG_SQUEUE_CHANGE);
@@ -1051,6 +1095,11 @@
 		cv_wait(async, lock);
 		CALLB_CPR_SAFE_END(&cprinfo, lock);
 
+		if (sqp->sq_state & SQS_EXIT) {
+			mutex_exit(lock);
+			thread_exit();
+		}
+
 		ctl_state = sqp->sq_state & (SQS_POLL_THR_CONTROL |
 		    SQS_POLL_THR_QUIESCED);
 		if (ctl_state != 0) {
@@ -1076,6 +1125,9 @@
 		    (SQS_PROC|SQS_POLLING|SQS_GET_PKTS)) ==
 		    (SQS_PROC|SQS_POLLING|SQS_GET_PKTS));
 
+		/* Only IP related squeues should reach this point */
+		VERIFY(sqp->sq_isip == B_TRUE);
+
 poll_again:
 		sq_rx_ring = sqp->sq_rx_ring;
 		sq_get_pkts = sq_rx_ring->rr_rx;
@@ -1205,6 +1257,7 @@
 	ill_rx_ring_t	*rx_ring;
 
 	ASSERT(MUTEX_HELD(&sqp->sq_lock));
+	VERIFY(sqp->sq_isip == B_TRUE);
 
 	if (sqp->sq_state & SQS_POLL_RESTART) {
 		/* Restart implies a previous quiesce. */
@@ -1316,6 +1369,11 @@
 
 	for (;;) {
 		for (;;) {
+			if (sqp->sq_state & SQS_EXIT) {
+				mutex_exit(lock);
+				thread_exit();
+			}
+
 			/*
 			 * If the poll thread has handed control to us
 			 * we need to break out of the wait.
@@ -1412,6 +1470,7 @@
 
 again:
 	sqp = connp->conn_sqp;
+	VERIFY(sqp->sq_isip == B_TRUE);
 
 	mutex_enter(&sqp->sq_lock);
 	if (sqp->sq_first == NULL && !(sqp->sq_state & SQS_PROC)) {
@@ -1487,6 +1546,7 @@
 squeue_synch_exit(conn_t *connp)
 {
 	squeue_t *sqp = connp->conn_sqp;
+	VERIFY(sqp->sq_isip == B_TRUE);
 
 	mutex_enter(&sqp->sq_lock);
 	if (sqp->sq_run == curthread) {
diff -ur illumos-gate/usr/src/uts/common/inet/tcp_impl.h illumos-joyent/usr/src/uts/common/inet/tcp_impl.h
--- illumos-gate/usr/src/uts/common/inet/tcp_impl.h	2015-05-26 22:07:54.851683245 +0000
+++ illumos-joyent/usr/src/uts/common/inet/tcp_impl.h	2015-05-26 22:08:56.140696474 +0000
@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, Joyent Inc. All rights reserved.
+ * Copyright (c) 2013, Joyent Inc. All rights reserved.
  * Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
  * Copyright (c) 2013, 2014 by Delphix. All rights reserved.
  */
@@ -61,9 +61,9 @@
  * by setting it to 0.
  */
 #define	TCP_XMIT_LOWATER	4096
-#define	TCP_XMIT_HIWATER	49152
+#define	TCP_XMIT_HIWATER	128000
 #define	TCP_RECV_LOWATER	2048
-#define	TCP_RECV_HIWATER	128000
+#define	TCP_RECV_HIWATER	1048576
 
 /*
  * Bind hash list size and has function.  It has to be a power of 2 for
diff -ur illumos-gate/usr/src/uts/common/inet/udp/udp_opt_data.c illumos-joyent/usr/src/uts/common/inet/udp/udp_opt_data.c
--- illumos-gate/usr/src/uts/common/inet/udp/udp_opt_data.c	2015-05-26 22:07:54.855570156 +0000
+++ illumos-joyent/usr/src/uts/common/inet/udp/udp_opt_data.c	2015-05-26 22:08:56.145346851 +0000
@@ -21,6 +21,7 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #include <sys/types.h>
@@ -292,6 +293,7 @@
 	},
 { UDP_NAT_T_ENDPOINT, IPPROTO_UDP, OA_RW, OA_RW, OP_PRIVPORT, 0, sizeof (int),
 	0 },
+{ UDP_SRCPORT_HASH, IPPROTO_UDP, OA_R, OA_RW, OP_CONFIG, 0, sizeof (int), 0 }
 };
 
 /*
diff -ur illumos-gate/usr/src/uts/common/inet/udp/udp.c illumos-joyent/usr/src/uts/common/inet/udp/udp.c
--- illumos-gate/usr/src/uts/common/inet/udp/udp.c	2015-05-26 22:07:54.855029314 +0000
+++ illumos-joyent/usr/src/uts/common/inet/udp/udp.c	2015-05-26 22:08:56.144824894 +0000
@@ -22,6 +22,7 @@
  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
  * Copyright 2014, OmniTI Computer Consulting, Inc. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 /* Copyright (c) 1990 Mentat Inc. */
 
@@ -76,7 +77,8 @@
 #include <inet/ipclassifier.h>
 #include <sys/squeue_impl.h>
 #include <inet/ipnet.h>
-#include <sys/ethernet.h>
+#include <sys/vxlan.h>
+#include <inet/inet_hash.h>
 
 #include <sys/tsol/label.h>
 #include <sys/tsol/tnet.h>
@@ -346,6 +348,89 @@
 typedef union T_primitives *t_primp_t;
 
 /*
+ * Various protocols that encapsulate UDP have no real use for the source port.
+ * Instead, they want to vary the source port to provide better equal-cost
+ * multipathing and other systems that use fanout. Consider something like
+ * VXLAN. If you're actually sending multiple different streams to a single
+ * host, if you don't vary the source port, then the tuple of ( SRC IP, DST IP,
+ * SRC Port, DST Port) will always be the same.
+ *
+ * Here, we return a port to hash this to, if we know how to hash it. If for
+ * some reason we can't perform an L4 hash, then we just return the default
+ * value, usually the default port. After we determine the hash we transform it
+ * so that it's in the range of [ min, max ].
+ *
+ * We'd like to avoid a pull up for the sake of performing the hash. If the
+ * first mblk_t doesn't have the full protocol header, then we just send it to
+ * the default. If for some reason we have an encapsulated packet that has its
+ * protocol header in different parts of an mblk_t, then we'll go with the
+ * default port. This means that that if a driver isn't consistent about how it
+ * generates the frames for a given flow, it will not always be consistently
+ * hashed. That should be an uncommon event.
+ */
+uint16_t
+udp_srcport_hash(mblk_t *mp, int type, uint16_t min, uint16_t max,
+    uint16_t def)
+{
+	size_t szused = 0;
+	struct ether_header *ether;
+	struct ether_vlan_header *vether;
+	ip6_t *ip6h;
+	ipha_t *ipha;
+	uint16_t sap;
+	uint64_t hash;
+	uint32_t mod;
+
+	ASSERT(min <= max);
+
+	if (type != UDP_HASH_VXLAN)
+		return (def);
+
+	if (!IS_P2ALIGNED(mp->b_rptr, sizeof (uint16_t)))
+		return (def);
+
+	/*
+	 * The following logic is VXLAN specific to get at the header, if we
+	 * have formats, eg. GENEVE, then we should ignore this.
+	 *
+	 * The kernel overlay device often puts a first mblk_t for the data
+	 * which is just the encap. If so, then we're going to use that and try
+	 * to avoid a pull up.
+	 */
+	if (MBLKL(mp) == VXLAN_HDR_LEN) {
+		if (mp->b_cont == NULL)
+			return (def);
+		mp = mp->b_cont;
+		ether = (struct ether_header *)mp->b_rptr;
+	} else if (MBLKL(mp) < VXLAN_HDR_LEN) {
+		return (def);
+	} else {
+		szused = VXLAN_HDR_LEN;
+		ether = (struct ether_header *)((uintptr_t)mp->b_rptr + szused);
+	}
+
+	/* Can we hold a MAC header? */
+	if (MBLKL(mp) + szused < sizeof (struct ether_header))
+		return (def);
+
+	/*
+	 * We need to lie about the starting offset into the message block for
+	 * convenience. Undo it at the end. We know that inet_pkt_hash() won't
+	 * modify the mblk_t.
+	 */
+	mp->b_rptr += szused;
+	hash = inet_pkt_hash(DL_ETHER, mp, INET_PKT_HASH_L2 |
+	    INET_PKT_HASH_L3 | INET_PKT_HASH_L4);
+	mp->b_rptr -= szused;
+
+	if (hash == 0)
+		return (def);
+
+	mod = max - min + 1;
+	return ((hash % mod) + min);
+}
+
+/*
  * Return the next anonymous port in the privileged port range for
  * bind checking.
  *
@@ -1583,6 +1668,11 @@
 			*i1 = udp->udp_rcvhdr ? 1 : 0;
 			mutex_exit(&connp->conn_lock);
 			return (sizeof (int));
+		case UDP_SRCPORT_HASH:
+			mutex_enter(&connp->conn_lock);
+			*i1 = udp->udp_vxlanhash;
+			mutex_exit(&connp->conn_lock);
+			return (sizeof (int));
 		}
 	}
 	mutex_enter(&connp->conn_lock);
@@ -1718,6 +1808,26 @@
 			udp->udp_rcvhdr = onoff;
 			mutex_exit(&connp->conn_lock);
 			return (0);
+		case UDP_SRCPORT_HASH:
+			/*
+			 * This should have already been verified, but double
+			 * check.
+			 */
+			if ((error = secpolicy_ip_config(cr, B_FALSE)) != 0) {
+				return (error);
+			}
+
+			/* First see if the val is something we understand */
+			if (*i1 != UDP_HASH_DISABLE && *i1 != UDP_HASH_VXLAN)
+				return (EINVAL);
+
+			if (!checkonly) {
+				mutex_enter(&connp->conn_lock);
+				udp->udp_vxlanhash = *i1;
+				mutex_exit(&connp->conn_lock);
+			}
+			/* Fully handled this option. */
+			return (0);
 		}
 		break;
 	}
@@ -2001,13 +2111,25 @@
 	uint32_t	cksum;
 	udp_t		*udp = connp->conn_udp;
 	boolean_t	insert_spi = udp->udp_nat_t_endpoint;
+	boolean_t	hash_srcport = udp->udp_vxlanhash;
 	uint_t		ulp_hdr_len;
+	uint16_t	srcport;
 
 	data_len = msgdsize(data_mp);
 	ulp_hdr_len = UDPH_SIZE;
 	if (insert_spi)
 		ulp_hdr_len += sizeof (uint32_t);
 
+	/*
+	 * If we have source port hashing going on, determine the hash before
+	 * we modify the mblk_t.
+	 */
+	if (hash_srcport == B_TRUE) {
+		srcport = udp_srcport_hash(mp, UDP_HASH_VXLAN,
+		    IPPORT_DYNAMIC_MIN, IPPORT_DYNAMIC_MAX,
+		    ntohs(connp->conn_lport));
+	}
+
 	mp = conn_prepend_hdr(ixa, ipp, v6src, v6dst, IPPROTO_UDP, flowinfo,
 	    ulp_hdr_len, data_mp, data_len, us->us_wroff_extra, &cksum, errorp);
 	if (mp == NULL) {
@@ -2019,7 +2141,11 @@
 	ixa->ixa_pktlen = data_len + ixa->ixa_ip_hdr_length;
 
 	udpha = (udpha_t *)(mp->b_rptr + ixa->ixa_ip_hdr_length);
-	udpha->uha_src_port = connp->conn_lport;
+	if (hash_srcport == B_TRUE) {
+		udpha->uha_src_port = htons(srcport);
+	} else {
+		udpha->uha_src_port = connp->conn_lport;
+	}
 	udpha->uha_dst_port = dstport;
 	udpha->uha_checksum = 0;
 	udpha->uha_length = htons(data_len);
@@ -3194,6 +3320,7 @@
 	udp_t		*udp = connp->conn_udp;
 	udp_stack_t	*us = udp->udp_us;
 	boolean_t	insert_spi = udp->udp_nat_t_endpoint;
+	boolean_t	hash_srcport = udp->udp_vxlanhash;
 	uint_t		pktlen;
 	uint_t		alloclen;
 	uint_t		copylen;
@@ -3202,10 +3329,21 @@
 	udpha_t		*udpha;
 	uint32_t	cksum;
 	ip_pkt_t	*ipp;
+	uint16_t	srcport;
 
 	ASSERT(MUTEX_HELD(&connp->conn_lock));
 
 	/*
+	 * If we have source port hashing going on, determine the hash before
+	 * we modify the mblk_t.
+	 */
+	if (hash_srcport == B_TRUE) {
+		srcport = udp_srcport_hash(mp, UDP_HASH_VXLAN,
+		    IPPORT_DYNAMIC_MIN, IPPORT_DYNAMIC_MAX,
+		    ntohs(connp->conn_lport));
+	}
+
+	/*
 	 * Copy the header template and leave space for an SPI
 	 */
 	copylen = connp->conn_ht_iphc_len;
@@ -3303,6 +3441,9 @@
 		*((uint32_t *)(udpha + 1)) = 0;
 
 	udpha->uha_dst_port = dstport;
+	if (hash_srcport == B_TRUE)
+		udpha->uha_src_port = htons(srcport);
+
 	return (mp);
 }
 
diff -ur illumos-gate/usr/src/uts/common/inet/udp_impl.h illumos-joyent/usr/src/uts/common/inet/udp_impl.h
--- illumos-gate/usr/src/uts/common/inet/udp_impl.h	2015-05-26 22:07:54.856236567 +0000
+++ illumos-joyent/usr/src/uts/common/inet/udp_impl.h	2015-05-26 22:08:56.146163362 +0000
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2015, Joyent, Inc.
  */
 
 #ifndef	_UDP_IMPL_H
@@ -178,8 +179,11 @@
 		udp_issocket : 1,	/* socket mode; sockfs is on top */
 		udp_nat_t_endpoint : 1,	/* UDP_NAT_T_ENDPOINT option */
 		udp_rcvhdr : 1,		/* UDP_RCVHDR option */
+		udp_vxlanhash: 1,	/* UDP_SRCPORT_HASH option */
+					/* Because there's only VXLAN, cheat */
+					/* and only use a single bit */
 
-		udp_pad_to_bit_31 : 29;
+		udp_pad_to_bit_31 : 28;
 
 	/* Following 2 fields protected by the uf_lock */
 	struct udp_s	*udp_bind_hash; /* Bind hash chain */