PSARC 2008/575 ILB: Integrated L3/L4 Load balancer

6882718 in-kernel simple L3/L4 load balancing service should be provided in Solaris
6884202 ipobs_hook() in ip_input() invalidates DB_REF assumption

4 files changed, 47 insertions, 13 deletions

diff --git a/usr/src/lib/libinetutil/common/libinetutil.h b/usr/src/lib/libinetutil/common/libinetutil.h
index bacf64938a..a285103af6 100644
--- a/usr/src/lib/libinetutil/common/libinetutil.h
+++ b/usr/src/lib/libinetutil/common/libinetutil.h
@@ -158,7 +158,7 @@ typedef struct iu_timer_queue iu_tq_t;
 
 typedef int iu_timer_id_t;
 
-#define	IU_TIMER_ID_MAX	1024	/* max number of concurrent timers */
+#define	IU_TIMER_ID_MAX	4096	/* max number of concurrent timers */
 
 /*
  * a iu_tq_callback_t is a function that is called back in response to a
diff --git a/usr/src/lib/libinetutil/common/ofmt.c b/usr/src/lib/libinetutil/common/ofmt.c
index 2bfc4aafa4..c0fce1839e 100644
--- a/usr/src/lib/libinetutil/common/ofmt.c
+++ b/usr/src/lib/libinetutil/common/ofmt.c
@@ -66,6 +66,8 @@ typedef struct ofmt_state_s {
 	boolean_t	os_wrap;
 	int		os_nbad;
 	char		**os_badfields;
+	boolean_t	os_multiline;
+	int		os_maxnamelen;	/* longest name (f. multiline) */
 } ofmt_state_t;
 /*
  * A B_TRUE return value from the callback function will print out the contents
@@ -187,9 +189,12 @@ ofmt_open(const char *str, const ofmt_field_t *template, uint_t flags,
 	ofmt_status_t	err = OFMT_SUCCESS;
 	boolean_t	parsable = ((flags & OFMT_PARSABLE) != 0);
 	boolean_t	wrap = ((flags & OFMT_WRAP) != 0);
+	boolean_t	multiline = (flags & OFMT_MULTILINE);
 
 	*ofmt = NULL;
 	if (parsable) {
+		if (multiline)
+			return (OFMT_EPARSEMULTI);
 		/*
 		 * For parsable output mode, the caller always needs
 		 * to specify precisely which fields are to be selected,
@@ -229,6 +234,7 @@ ofmt_open(const char *str, const ofmt_field_t *template, uint_t flags,
 	os->os_parsable = parsable;
 	os->os_wrap = wrap;
 
+	os->os_multiline = multiline;
 	of = os->os_fields;
 	of_index = 0;
 	/*
@@ -260,6 +266,11 @@ ofmt_open(const char *str, const ofmt_field_t *template, uint_t flags,
 		of[of_index].of_name = strdup(template[j].of_name);
 		if (of[of_index].of_name == NULL)
 			goto nomem;
+		if (multiline) {
+			int n = strlen(of[of_index].of_name);
+
+			os->os_maxnamelen = MAX(n, os->os_maxnamelen);
+		}
 		of[of_index].of_width = template[j].of_width;
 		of[of_index].of_id = template[j].of_id;
 		of[of_index].of_cb = template[j].of_cb;
@@ -310,14 +321,13 @@ ofmt_print_field(ofmt_state_t *os, ofmt_field_t *ofp, const char *value,
 	uint_t	width = ofp->of_width;
 	uint_t	valwidth;
 	uint_t	compress;
-	boolean_t parsable = os->os_parsable;
 	char	c;
 
 	/*
 	 * Parsable fields are separated by ':'. If such a field contains
 	 * a ':' or '\', this character is prefixed by a '\'.
 	 */
-	if (parsable) {
+	if (os->os_parsable) {
 		if (os->os_nfields == 1) {
 			(void) printf("%s", value);
 			return;
@@ -329,7 +339,13 @@ ofmt_print_field(ofmt_state_t *os, ofmt_field_t *ofp, const char *value,
 		}
 		if (!os->os_lastfield)
 			(void) putchar(':');
-		return;
+	} else if (os->os_multiline) {
+		if (value[0] == '\0')
+			value = OFMT_VAL_UNDEF;
+		(void) printf("%*.*s: %s", os->os_maxnamelen,
+		    os->os_maxnamelen, ofp->of_name, value);
+		if (!os->os_lastfield)
+			(void) putchar('\n');
 	} else {
 		if (os->os_lastfield) {
 			(void) printf("%s", value);
@@ -408,11 +424,15 @@ ofmt_print(ofmt_handle_t ofmt, void *arg)
 			return;
 	}
 
-	if ((os->os_nrow++ % os->os_winsize.ws_row) == 0 && !os->os_parsable) {
+	if ((os->os_nrow++ % os->os_winsize.ws_row) == 0 && !os->os_parsable &&
+	    !os->os_multiline) {
 		ofmt_print_header(os);
 		os->os_nrow++;
 	}
 
+	if (os->os_multiline && os->os_nrow > 1)
+		(void) putchar('\n');
+
 	of = os->os_fields;
 	escsep = (os->os_nfields > 1);
 	more_rows = B_FALSE;
@@ -549,6 +569,9 @@ ofmt_strerror(ofmt_handle_t ofmt, ofmt_status_t err, char *buf, uint_t bufsize)
 	case OFMT_ENOFIELDS:
 		s = "no valid output fields";
 		break;
+	case OFMT_EPARSEMULTI:
+		s = "multiline mode incompatible with parsable mode";
+		break;
 	case OFMT_EPARSEALL:
 		s = "output field `all' invalid in parsable mode";
 		break;
diff --git a/usr/src/lib/libinetutil/common/ofmt.h b/usr/src/lib/libinetutil/common/ofmt.h
index d3e4142578..81693ae325 100644
--- a/usr/src/lib/libinetutil/common/ofmt.h
+++ b/usr/src/lib/libinetutil/common/ofmt.h
@@ -35,6 +35,8 @@
  * are separated by ':', with the ':' character itself escaped by a \
  * (e.g., IPv6 addresses  may be printed as "fe80\:\:1"); single field output
  * is printed as-is.
+ * In multiline mode, every [field,value] pair is printed in a line of
+ * its own, thus: "field: value".
  *
  * The caller must open a handle for each set of fields to be printed by
  * invoking ofmt_open(). The invocation to ofmt_open must provide the list of
@@ -57,6 +59,9 @@
  * (non machine-parsable) mode, a NULL fields_str, or a value of "all" for
  * fields_str, is treated as a request to print all allowable fields that fit
  * other applicable constraints.
+ * To achieve multiline mode, OFMT_MULTILINE needs to be specified in oflags.
+ * Specifying both OFMT_MULTILINE and OFMT_PARSABLE will result in
+ * OFMT_EPARSEMULTI.
  *
  * Thus a typical invocation to open the ofmt_handle would be:
  *
@@ -135,7 +140,8 @@ typedef enum {
 	OFMT_EPARSEALL,		/* 'all' invalid in parsable mode */
 	OFMT_EPARSENONE,	/* output fields missing in parsable mode */
 	OFMT_EPARSEWRAP,	/* parsable mode incompatible with wrap mode */
-	OFMT_ENOTEMPLATE	/* no template provided for fields */
+	OFMT_ENOTEMPLATE,	/* no template provided for fields */
+	OFMT_EPARSEMULTI	/* parsable and multiline don't mix */
 } ofmt_status_t;
 
 /*
@@ -171,6 +177,7 @@ extern ofmt_status_t ofmt_open(const char *, const ofmt_field_t *, uint_t,
 
 #define	OFMT_PARSABLE	0x00000001 /* machine parsable mode */
 #define	OFMT_WRAP	0x00000002 /* wrap output if field width is exceeded */
+#define	OFMT_MULTILINE	0x00000004 /* "long" output: "name: value" lines */
 
 /*
  * ofmt_close() must be called to free resources associated
diff --git a/usr/src/lib/libinetutil/common/tq.c b/usr/src/lib/libinetutil/common/tq.c
index 78505462bd..e809b3289d 100644
--- a/usr/src/lib/libinetutil/common/tq.c
+++ b/usr/src/lib/libinetutil/common/tq.c
@@ -2,9 +2,8 @@
  * CDDL HEADER START
  *
  * The contents of this file are subject to the terms of the
- * Common Development and Distribution License, Version 1.0 only
- * (the "License").  You may not use this file except in compliance
- * with the License.
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
  *
  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
  * or http://www.opensolaris.org/os/licensing.
@@ -20,12 +19,10 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2004 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include <stdlib.h>
 #include <limits.h>
 #include <sys/time.h>
@@ -342,7 +339,14 @@ iu_expire_timers(iu_tq_t *tq)
 	for (node = tq->iutq_head; node != NULL;
 	    node = node->iutn_expire_next) {
 
-		if (node->iutn_abs_timeout > current_time)
+		/*
+		 * If the timeout is within 1 millisec of current time,
+		 * consider it as expired already.  We do this because
+		 * iu_earliest_timer() only has millisec granularity.
+		 * So we should also use millisec grandularity in
+		 * comparing timeout values.
+		 */
+		if (node->iutn_abs_timeout - current_time > 1000000)
 			break;
 
 		/*