Now, that we have gjournal in the tree add possibility to configure

gmirror and graid3 in a way that it is not resynchronized after a
power failure or system crash.
It is safe when gjournal is running on top of gmirror/graid3.

6 files changed, 98 insertions, 15 deletions

diff --git a/sys/geom/mirror/g_mirror.c b/sys/geom/mirror/g_mirror.c
index f03da3b..419611e 100644
--- a/sys/geom/mirror/g_mirror.c
+++ b/sys/geom/mirror/g_mirror.c
@@ -802,6 +802,8 @@ g_mirror_idle(struct g_mirror_softc *sc, int acw)
 
 	if (sc->sc_provider == NULL)
 		return (0);
+	if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0)
+		return (0);
 	if (sc->sc_idle)
 		return (0);
 	if (sc->sc_writes > 0)
@@ -831,6 +833,8 @@ g_mirror_unidle(struct g_mirror_softc *sc)
 	g_topology_assert_not();
 	sx_assert(&sc->sc_lock, SX_XLOCKED);
 
+	if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0)
+		return;
 	sc->sc_idle = 0;
 	sc->sc_last_write = time_uptime;
 	LIST_FOREACH(disk, &sc->sc_disks, d_next) {
@@ -1884,6 +1888,8 @@ g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk)
 
 	sx_assert(&sc->sc_lock, SX_LOCKED);
 
+	if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0)
+		return;
 	if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) {
 		G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.",
 		    g_mirror_get_diskname(disk), sc->sc_name);
@@ -1928,7 +1934,8 @@ g_mirror_sync_start(struct g_mirror_disk *disk)
 
 	G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name,
 	    g_mirror_get_diskname(disk));
-	disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY;
+	if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0)
+		disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY;
 	KASSERT(disk->d_sync.ds_consumer == NULL,
 	    ("Sync consumer already exists (device=%s, disk=%s).",
 	    sc->sc_name, g_mirror_get_diskname(disk)));
@@ -3180,6 +3187,7 @@ g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
 		sbuf_printf(sb, name);					\
 	}								\
 } while (0)
+			ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC");
 			ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC");
 #undef	ADD_FLAG
 		}
diff --git a/sys/geom/mirror/g_mirror.h b/sys/geom/mirror/g_mirror.h
index a6ddc51..eb67b6e0 100644
--- a/sys/geom/mirror/g_mirror.h
+++ b/sys/geom/mirror/g_mirror.h
@@ -41,8 +41,9 @@
  * 1 - Added 'prefer' balance algorithm.
  * 2 - Added md_genid field to metadata.
  * 3 - Added md_provsize field to metadata.
+ * 4 - Added 'no failure synchronization' flag.
  */
-#define	G_MIRROR_VERSION	3
+#define	G_MIRROR_VERSION	4
 
 #define	G_MIRROR_BALANCE_NONE		0
 #define	G_MIRROR_BALANCE_ROUND_ROBIN	1
@@ -64,7 +65,9 @@
 					 G_MIRROR_DISK_FLAG_INACTIVE)
 
 #define	G_MIRROR_DEVICE_FLAG_NOAUTOSYNC	0x0000000000000001ULL
-#define	G_MIRROR_DEVICE_FLAG_MASK	(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC)
+#define	G_MIRROR_DEVICE_FLAG_NOFAILSYNC	0x0000000000000002ULL
+#define	G_MIRROR_DEVICE_FLAG_MASK	(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC | \
+					 G_MIRROR_DEVICE_FLAG_NOFAILSYNC)
 
 #ifdef _KERNEL
 extern u_int g_mirror_debug;
@@ -341,7 +344,7 @@ mirror_metadata_decode_v2(const u_char *data, struct g_mirror_metadata *md)
 	return (0);
 }
 static __inline int
-mirror_metadata_decode_v3(const u_char *data, struct g_mirror_metadata *md)
+mirror_metadata_decode_v3v4(const u_char *data, struct g_mirror_metadata *md)
 {
 	MD5_CTX ctx;
 
@@ -385,7 +388,8 @@ mirror_metadata_decode(const u_char *data, struct g_mirror_metadata *md)
 		error = mirror_metadata_decode_v2(data, md);
 		break;
 	case 3:
-		error = mirror_metadata_decode_v3(data, md);
+	case 4:
+		error = mirror_metadata_decode_v3v4(data, md);
 		break;
 	default:
 		error = EINVAL;
@@ -456,6 +460,8 @@ mirror_metadata_dump(const struct g_mirror_metadata *md)
 	if (md->md_mflags == 0)
 		printf(" NONE");
 	else {
+		if ((md->md_mflags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0)
+			printf(" NOFAILSYNC");
 		if ((md->md_mflags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0)
 			printf(" NOAUTOSYNC");
 	}
diff --git a/sys/geom/mirror/g_mirror_ctl.c b/sys/geom/mirror/g_mirror_ctl.c
index 9578782..27e58ce 100644
--- a/sys/geom/mirror/g_mirror_ctl.c
+++ b/sys/geom/mirror/g_mirror_ctl.c
@@ -97,7 +97,8 @@ g_mirror_ctl_configure(struct gctl_req *req, struct g_class *mp)
 	intmax_t *slicep;
 	uint32_t slice;
 	uint8_t balance;
-	int *nargs, *autosync, *noautosync, *hardcode, *dynamic, do_sync = 0;
+	int *autosync, *noautosync, *failsync, *nofailsync, *hardcode, *dynamic;
+	int *nargs, do_sync = 0, dirty = 1;
 
 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
 	if (nargs == NULL) {
@@ -128,6 +129,16 @@ g_mirror_ctl_configure(struct gctl_req *req, struct g_class *mp)
 		gctl_error(req, "No '%s' argument.", "noautosync");
 		return;
 	}
+	failsync = gctl_get_paraml(req, "failsync", sizeof(*failsync));
+	if (failsync == NULL) {
+		gctl_error(req, "No '%s' argument.", "failsync");
+		return;
+	}
+	nofailsync = gctl_get_paraml(req, "nofailsync", sizeof(*nofailsync));
+	if (nofailsync == NULL) {
+		gctl_error(req, "No '%s' argument.", "nofailsync");
+		return;
+	}
 	hardcode = gctl_get_paraml(req, "hardcode", sizeof(*hardcode));
 	if (hardcode == NULL) {
 		gctl_error(req, "No '%s' argument.", "hardcode");
@@ -143,6 +154,11 @@ g_mirror_ctl_configure(struct gctl_req *req, struct g_class *mp)
 		    "noautosync");
 		return;
 	}
+	if (*failsync && *nofailsync) {
+		gctl_error(req, "'%s' and '%s' specified.", "failsync",
+		    "nofailsync");
+		return;
+	}
 	if (*hardcode && *dynamic) {
 		gctl_error(req, "'%s' and '%s' specified.", "hardcode",
 		    "dynamic");
@@ -180,7 +196,8 @@ g_mirror_ctl_configure(struct gctl_req *req, struct g_class *mp)
 		return;
 	}
 	if (sc->sc_balance == balance && sc->sc_slice == slice && !*autosync &&
-	    !*noautosync && !*hardcode && !*dynamic) {
+	    !*noautosync && !*failsync && !*nofailsync && !*hardcode &&
+	    !*dynamic) {
 		sx_xunlock(&sc->sc_lock);
 		gctl_error(req, "Nothing has changed.");
 		return;
@@ -196,6 +213,15 @@ g_mirror_ctl_configure(struct gctl_req *req, struct g_class *mp)
 		if (*noautosync)
 			sc->sc_flags |= G_MIRROR_DEVICE_FLAG_NOAUTOSYNC;
 	}
+	if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) {
+		if (*failsync)
+			sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_NOFAILSYNC;
+	} else {
+		if (*nofailsync) {
+			sc->sc_flags |= G_MIRROR_DEVICE_FLAG_NOFAILSYNC;
+			dirty = 0;
+		}
+	}
 	LIST_FOREACH(disk, &sc->sc_disks, d_next) {
 		if (do_sync) {
 			if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING)
@@ -205,6 +231,8 @@ g_mirror_ctl_configure(struct gctl_req *req, struct g_class *mp)
 			disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED;
 		else if (*dynamic)
 			disk->d_flags &= ~G_MIRROR_DISK_FLAG_HARDCODED;
+		if (!dirty)
+			disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
 		g_mirror_update_metadata(disk);
 		if (do_sync) {
 			if (disk->d_state == G_MIRROR_DISK_STATE_STALE) {
diff --git a/sys/geom/raid3/g_raid3.c b/sys/geom/raid3/g_raid3.c
index e08e331..776b6bd 100644
--- a/sys/geom/raid3/g_raid3.c
+++ b/sys/geom/raid3/g_raid3.c
@@ -861,6 +861,8 @@ g_raid3_idle(struct g_raid3_softc *sc, int acw)
 
 	if (sc->sc_provider == NULL)
 		return (0);
+	if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0)
+		return (0);
 	if (sc->sc_idle)
 		return (0);
 	if (sc->sc_writes > 0)
@@ -892,6 +894,8 @@ g_raid3_unidle(struct g_raid3_softc *sc)
 	g_topology_assert_not();
 	sx_assert(&sc->sc_lock, SX_XLOCKED);
 
+	if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0)
+		return;
 	sc->sc_idle = 0;
 	sc->sc_last_write = time_uptime;
 	for (i = 0; i < sc->sc_ndisks; i++) {
@@ -2154,6 +2158,8 @@ g_raid3_update_idle(struct g_raid3_softc *sc, struct g_raid3_disk *disk)
 {
 
 	sx_assert(&sc->sc_lock, SX_LOCKED);
+	if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0)
+		return;
 	if (!sc->sc_idle && (disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) == 0) {
 		G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.",
 		    g_raid3_get_diskname(disk), sc->sc_name);
@@ -2206,7 +2212,8 @@ g_raid3_sync_start(struct g_raid3_softc *sc)
 
 	G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name,
 	    g_raid3_get_diskname(disk));
-	disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY;
+	if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) == 0)
+		disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY;
 	KASSERT(disk->d_sync.ds_consumer == NULL,
 	    ("Sync consumer already exists (device=%s, disk=%s).",
 	    sc->sc_name, g_raid3_get_diskname(disk)));
@@ -3481,6 +3488,7 @@ g_raid3_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
 		sbuf_printf(sb, name);					\
 	}								\
 } while (0)
+			ADD_FLAG(G_RAID3_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC");
 			ADD_FLAG(G_RAID3_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC");
 			ADD_FLAG(G_RAID3_DEVICE_FLAG_ROUND_ROBIN,
 			    "ROUND-ROBIN");
diff --git a/sys/geom/raid3/g_raid3.h b/sys/geom/raid3/g_raid3.h
index 1032142..a375999 100644
--- a/sys/geom/raid3/g_raid3.h
+++ b/sys/geom/raid3/g_raid3.h
@@ -42,8 +42,9 @@
  * 2 - Added 'verify reading' algorithm.
  * 3 - Added md_genid field to metadata.
  * 4 - Added md_provsize field to metadata.
+ * 5 - Added 'no failure synchronization' flag.
  */
-#define	G_RAID3_VERSION		4
+#define	G_RAID3_VERSION		5
 
 #define	G_RAID3_DISK_FLAG_DIRTY		0x0000000000000001ULL
 #define	G_RAID3_DISK_FLAG_SYNCHRONIZING	0x0000000000000002ULL
@@ -57,9 +58,11 @@
 #define	G_RAID3_DEVICE_FLAG_NOAUTOSYNC	0x0000000000000001ULL
 #define	G_RAID3_DEVICE_FLAG_ROUND_ROBIN	0x0000000000000002ULL
 #define	G_RAID3_DEVICE_FLAG_VERIFY	0x0000000000000004ULL
+#define	G_RAID3_DEVICE_FLAG_NOFAILSYNC	0x0000000000000008ULL
 #define	G_RAID3_DEVICE_FLAG_MASK	(G_RAID3_DEVICE_FLAG_NOAUTOSYNC | \
 					 G_RAID3_DEVICE_FLAG_ROUND_ROBIN | \
-					 G_RAID3_DEVICE_FLAG_VERIFY)
+					 G_RAID3_DEVICE_FLAG_VERIFY | \
+					 G_RAID3_DEVICE_FLAG_NOFAILSYNC)
 
 #ifdef _KERNEL
 extern u_int g_raid3_debug;
@@ -363,7 +366,7 @@ raid3_metadata_decode_v3(const u_char *data, struct g_raid3_metadata *md)
 	return (0);
 }
 static __inline int
-raid3_metadata_decode_v4(const u_char *data, struct g_raid3_metadata *md)
+raid3_metadata_decode_v4v5(const u_char *data, struct g_raid3_metadata *md)
 {
 	MD5_CTX ctx;
 
@@ -405,7 +408,8 @@ raid3_metadata_decode(const u_char *data, struct g_raid3_metadata *md)
 		error = raid3_metadata_decode_v3(data, md);
 		break;
 	case 4:
-		error = raid3_metadata_decode_v4(data, md);
+	case 5:
+		error = raid3_metadata_decode_v4v5(data, md);
 		break;
 	default:
 		error = EINVAL;
@@ -442,6 +446,8 @@ raid3_metadata_dump(const struct g_raid3_metadata *md)
 			printf(" ROUND-ROBIN");
 		if ((md->md_mflags & G_RAID3_DEVICE_FLAG_VERIFY) != 0)
 			printf(" VERIFY");
+		if ((md->md_mflags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0)
+			printf(" NOFAILSYNC");
 	}
 	printf("\n");
 	printf("    dflags:");
diff --git a/sys/geom/raid3/g_raid3_ctl.c b/sys/geom/raid3/g_raid3_ctl.c
index 67d1f44..952ac2b 100644
--- a/sys/geom/raid3/g_raid3_ctl.c
+++ b/sys/geom/raid3/g_raid3_ctl.c
@@ -98,8 +98,9 @@ g_raid3_ctl_configure(struct gctl_req *req, struct g_class *mp)
 	struct g_raid3_softc *sc;
 	struct g_raid3_disk *disk;
 	const char *name;
-	int *nargs, do_sync = 0;
+	int *nargs, do_sync = 0, dirty = 1;
 	int *autosync, *noautosync;
+	int *failsync, *nofailsync;
 	int *round_robin, *noround_robin;
 	int *verify, *noverify;
 	u_int n;
@@ -128,6 +129,21 @@ g_raid3_ctl_configure(struct gctl_req *req, struct g_class *mp)
 		    "noautosync");
 		return;
 	}
+	failsync = gctl_get_paraml(req, "failsync", sizeof(*failsync));
+	if (failsync == NULL) {
+		gctl_error(req, "No '%s' argument.", "failsync");
+		return;
+	}
+	nofailsync = gctl_get_paraml(req, "nofailsync", sizeof(*nofailsync));
+	if (nofailsync == NULL) {
+		gctl_error(req, "No '%s' argument.", "nofailsync");
+		return;
+	}
+	if (*failsync && *nofailsync) {
+		gctl_error(req, "'%s' and '%s' specified.", "failsync",
+		    "nofailsync");
+		return;
+	}
 	round_robin = gctl_get_paraml(req, "round_robin", sizeof(*round_robin));
 	if (round_robin == NULL) {
 		gctl_error(req, "No '%s' argument.", "round_robin");
@@ -159,8 +175,8 @@ g_raid3_ctl_configure(struct gctl_req *req, struct g_class *mp)
 		    "noverify");
 		return;
 	}
-	if (!*autosync && !*noautosync && !*round_robin && !*noround_robin &&
-	    !*verify && !*noverify) {
+	if (!*autosync && !*noautosync && !*failsync && !*nofailsync &&
+	    !*round_robin && !*noround_robin && !*verify && !*noverify) {
 		gctl_error(req, "Nothing has changed.");
 		return;
 	}
@@ -188,6 +204,15 @@ g_raid3_ctl_configure(struct gctl_req *req, struct g_class *mp)
 		if (*noautosync)
 			sc->sc_flags |= G_RAID3_DEVICE_FLAG_NOAUTOSYNC;
 	}
+	if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0) {
+		if (*failsync)
+			sc->sc_flags &= ~G_RAID3_DEVICE_FLAG_NOFAILSYNC;
+	} else {
+		if (*nofailsync) {
+			sc->sc_flags |= G_RAID3_DEVICE_FLAG_NOFAILSYNC;
+			dirty = 0;
+		}
+	}
 	if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_VERIFY) != 0) {
 		if (*noverify)
 			sc->sc_flags &= ~G_RAID3_DEVICE_FLAG_VERIFY;
@@ -215,6 +240,8 @@ g_raid3_ctl_configure(struct gctl_req *req, struct g_class *mp)
 			if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING)
 				disk->d_flags &= ~G_RAID3_DISK_FLAG_FORCE_SYNC;
 		}
+		if (!dirty)
+			disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY;
 		g_raid3_update_metadata(disk);
 		if (do_sync) {
 			if (disk->d_state == G_RAID3_DISK_STATE_STALE) {