OS-7389 restart_init() needs to be observable

Reviewed by: Patrick Mooney <patrick.mooney@joyent.com>
Reviewed by: Hans Rosenfeld <hans.rosenfeld@joyent.com>
Approved by: Hans Rosenfeld <hans.rosenfeld@joyent.com>

3 files changed, 71 insertions, 0 deletions

diff --git a/usr/src/uts/common/os/exit.c b/usr/src/uts/common/os/exit.c
index 62176e76f0..06e0117cd6 100644
--- a/usr/src/uts/common/os/exit.c
+++ b/usr/src/uts/common/os/exit.c
@@ -138,6 +138,27 @@ rexit(int rval)
 }
 
 /*
+ * Bump the init_restarts kstat and let interested parties know about the
+ * restart.
+ */
+static void
+restart_init_notify(zone_t *zone)
+{
+	nvlist_t *nvl = NULL;
+
+	zone->zone_proc_init_restarts++;
+
+	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0 &&
+	    nvlist_add_uint32(nvl, ZONE_CB_RESTARTS,
+	    zone->zone_proc_init_restarts) == 0) {
+		zone_sysevent_publish(zone, ZONE_EVENT_INIT_CLASS,
+		    ZONE_EVENT_INIT_RESTART_SC, nvl);
+	}
+
+	nvlist_free(nvl);
+}
+
+/*
  * Called by proc_exit() when a zone's init exits, presumably because
  * it failed.  As long as the given zone is still in the "running"
  * state, we will re-exec() init, but first we need to reset things
@@ -260,6 +281,8 @@ restart_init(int what, int why)
 	ASSERT(p == curproc);
 	(void) freectty(B_TRUE);
 
+	restart_init_notify(p->p_zone);
+
 	/*
 	 * Now exec() the new init(1M) on top of the current process.  If we
 	 * succeed, the caller will treat this like a successful system call.
diff --git a/usr/src/uts/common/os/zone.c b/usr/src/uts/common/os/zone.c
index 1f5cdbc1c1..cf869025bd 100644
--- a/usr/src/uts/common/os/zone.c
+++ b/usr/src/uts/common/os/zone.c
@@ -2399,6 +2399,7 @@ zone_misc_kstat_update(kstat_t *ksp, int rw)
 	zmp->zm_nested_intp.value.ui32 = zone->zone_nested_intp;
 
 	zmp->zm_init_pid.value.ui32 = zone->zone_proc_initpid;
+	zmp->zm_init_restarts.value.ui32 = zone->zone_proc_init_restarts;
 	zmp->zm_boot_time.value.ui64 = (uint64_t)zone->zone_boot_time;
 
 	return (0);
@@ -2444,6 +2445,8 @@ zone_misc_kstat_create(zone_t *zone)
 	kstat_named_init(&zmp->zm_nested_intp, "nested_interp",
 	    KSTAT_DATA_UINT32);
 	kstat_named_init(&zmp->zm_init_pid, "init_pid", KSTAT_DATA_UINT32);
+	kstat_named_init(&zmp->zm_init_restarts, "init_restarts",
+	    KSTAT_DATA_UINT32);
 	kstat_named_init(&zmp->zm_boot_time, "boot_time", KSTAT_DATA_UINT64);
 
 	ksp->ks_update = zone_misc_kstat_update;
@@ -3004,6 +3007,8 @@ zone_status_set(zone_t *zone, zone_status_t status)
 #ifdef DEBUG
 		(void) printf(
 		    "Failed to allocate and send zone state change event.\n");
+#else
+		/* EMPTY */
 #endif
 	}
 	nvlist_free(nvl);
@@ -3023,6 +3028,38 @@ zone_status_get(zone_t *zone)
 	return (zone->zone_status);
 }
 
+/*
+ * Publish a zones-related sysevent for purposes other than zone state changes.
+ * While it is unfortunate that zone_event_chan is associated with
+ * "com.sun:zones:status" (rather than "com.sun:zones") state changes should be
+ * the only ones with class "status" and subclass "change".
+ */
+void
+zone_sysevent_publish(zone_t *zone, const char *class, const char *subclass,
+    nvlist_t *ev_nvl)
+{
+	nvlist_t *nvl = NULL;
+	timestruc_t now;
+	uint64_t t;
+
+	gethrestime(&now);
+	t = (now.tv_sec * NANOSEC) + now.tv_nsec;
+
+	if (nvlist_dup(ev_nvl, &nvl, KM_SLEEP) != 0 ||
+	    nvlist_add_string(nvl, ZONE_CB_NAME, zone->zone_name) != 0 ||
+	    nvlist_add_uint64(nvl, ZONE_CB_ZONEID, zone->zone_id) != 0 ||
+	    nvlist_add_uint64(nvl, ZONE_CB_TIMESTAMP, t) != 0 ||
+	    sysevent_evc_publish(zone_event_chan, class, subclass, "sun.com",
+	    "kernel", nvl, EVCH_SLEEP) != 0) {
+#ifdef DEBUG
+		(void) printf("Failed to allocate and send zone misc event.\n");
+#else
+		/* EMPTY */
+#endif
+	}
+	nvlist_free(nvl);
+}
+
 static int
 zone_set_bootargs(zone_t *zone, const char *zone_bootargs)
 {
diff --git a/usr/src/uts/common/sys/zone.h b/usr/src/uts/common/sys/zone.h
index 845f20a46f..a4ec347ce4 100644
--- a/usr/src/uts/common/sys/zone.h
+++ b/usr/src/uts/common/sys/zone.h
@@ -43,6 +43,7 @@
 #include <sys/secflags.h>
 #include <netinet/in.h>
 #include <sys/cpu_uarray.h>
+#include <sys/nvpair.h>
 
 #ifdef	__cplusplus
 extern "C" {
@@ -143,9 +144,13 @@ extern "C" {
 #define	ZONE_CB_NAME		"zonename"
 #define	ZONE_CB_NEWSTATE	"newstate"
 #define	ZONE_CB_OLDSTATE	"oldstate"
+#define	ZONE_CB_RESTARTS	"restarts"
 #define	ZONE_CB_TIMESTAMP	"when"
 #define	ZONE_CB_ZONEID		"zoneid"
 
+#define	ZONE_EVENT_INIT_CLASS		"init"
+#define	ZONE_EVENT_INIT_RESTART_SC	"restart"
+
 /*
  * Exit values that may be returned by scripts or programs invoked by various
  * zone commands.
@@ -486,6 +491,7 @@ typedef struct {
 	kstat_named_t	zm_mfseglim;
 	kstat_named_t	zm_nested_intp;
 	kstat_named_t	zm_init_pid;
+	kstat_named_t	zm_init_restarts;
 	kstat_named_t	zm_boot_time;
 } zone_misc_kstat_t;
 
@@ -581,6 +587,7 @@ typedef struct zone {
 	kcondvar_t	zone_cv;	/* used to signal state changes */
 	struct proc	*zone_zsched;	/* Dummy kernel "zsched" process */
 	pid_t		zone_proc_initpid; /* pid of "init" for this zone */
+	uint_t		zone_proc_init_restarts; /* times init restarted */
 	char		*zone_initname;		/* fs path to 'init' */
 	int		zone_init_status;	/* init's exit status */
 	int		zone_boot_err;  /* for zone_boot() if boot fails */
@@ -1014,6 +1021,10 @@ extern rctl_hndl_t rc_zone_max_swap;
 extern rctl_hndl_t rc_zone_phys_mem;
 extern rctl_hndl_t rc_zone_max_lofi;
 
+/* For publishing sysevents related to a particular zone */
+extern void zone_sysevent_publish(zone_t *, const char *, const char *,
+    nvlist_t *);
+
 #endif	/* _KERNEL */
 
 #ifdef	__cplusplus