Clarify disks/volumes above 2TiB support in geom_raid:

- add support for volumes above 2TiB with Promise metadata format;
 - enforse and document other limitations:
   - Intel and Promise metadata formats do not support disks above 2TiB;
   - NVIDIA metadata format does not support volumes above 2TiB.

Sponsored by:	iXsystems, Inc.
MFC after:	2 weeks

sbin/geom/class/raid/graid.8

@@ -24,7 +24,7 @@
 .\"
 .\" $FreeBSD$
 .\"
-.Dd March 22, 2011
+.Dd October 26, 2011
 .Dt GRAID 8
 .Os
 .Sh NAME
@@ -250,6 +250,9 @@ If you started migration using BIOS or in some other way, make sure to
 complete it there.
 Do not run GEOM RAID class on migrating volumes under pain of possible data
 corruption!
+.Sh 2TiB BARRIERS
+Intel and Promise metadata formats do not support disks above 2TiB.
+NVIDIA metadata format does not support volumes above 2TiB.
 .Sh EXIT STATUS
 Exit status is 0 on success, and non-zero if the command fails.
 .Sh SEE ALSO

sys/geom/raid/md_intel.c

@@ -1172,15 +1172,18 @@ g_raid_md_taste_intel(struct g_raid_md_object *md, struct g_class *mp,
 	g_access(cp, -1, 0, 0);
 	if (meta == NULL) {
 		if (g_raid_aggressive_spare) {
-			if (vendor == 0x8086) {
+			if (vendor != 0x8086) {
+				G_RAID_DEBUG(1,
+				    "Intel vendor mismatch 0x%04x != 0x8086",
+				    vendor);
+			} else if (pp->mediasize / pp->sectorsize > UINT32_MAX) {
+				G_RAID_DEBUG(1,
+				    "Intel disk '%s' is too big.", pp->name);
+			} else {
 				G_RAID_DEBUG(1,
 				    "No Intel metadata, forcing spare.");
 				spare = 2;
 				goto search;
-			} else {
-				G_RAID_DEBUG(1,
-				    "Intel vendor mismatch 0x%04x != 0x8086",
-				    vendor);
 			}
 		}
 		return (G_RAID_MD_TASTE_FAIL);
@@ -1194,9 +1197,9 @@ g_raid_md_taste_intel(struct g_raid_md_object *md, struct g_class *mp,
 	}
 	if (meta->disk[disk_pos].sectors !=
 	    (pp->mediasize / pp->sectorsize)) {
-		G_RAID_DEBUG(1, "Intel size mismatch %u != %u",
+		G_RAID_DEBUG(1, "Intel size mismatch %ju != %ju",
-		    meta->disk[disk_pos].sectors,
+		    (off_t)meta->disk[disk_pos].sectors,
-		    (u_int)(pp->mediasize / pp->sectorsize));
+		    (off_t)(pp->mediasize / pp->sectorsize));
 		goto fail1;
 	}
 
@@ -1449,6 +1452,13 @@ g_raid_md_ctl_intel(struct g_raid_md_object *md,
 			cp->private = disk;
 			g_topology_unlock();
 
+			if (pp->mediasize / pp->sectorsize > UINT32_MAX) {
+				gctl_error(req,
+				    "Disk '%s' is too big.", diskname);
+				error = -8;
+				break;
+			}
+
 			error = g_raid_md_get_label(cp,
 			    &pd->pd_disk_meta.serial[0], INTEL_SERIAL_LEN);
 			if (error != 0) {
@@ -1940,6 +1950,14 @@ g_raid_md_ctl_intel(struct g_raid_md_object *md,
 			pp = cp->provider;
 			g_topology_unlock();
 
+			if (pp->mediasize / pp->sectorsize > UINT32_MAX) {
+				gctl_error(req,
+				    "Disk '%s' is too big.", diskname);
+				g_raid_kill_consumer(sc, cp);
+				error = -8;
+				break;
+			}
+
 			/* Read disk serial. */
 			error = g_raid_md_get_label(cp,
 			    &serial[0], INTEL_SERIAL_LEN);

sys/geom/raid/md_nvidia.c

@@ -1033,7 +1033,7 @@ g_raid_md_ctl_nvidia(struct g_raid_md_object *md,
 	char arg[16];
 	const char *verb, *volname, *levelname, *diskname;
 	int *nargs, *force;
-	off_t size, sectorsize, strip;
+	off_t size, sectorsize, strip, volsize;
 	intmax_t *sizearg, *striparg;
 	int numdisks, i, len, level, qual, update;
 	int error;
@@ -1182,7 +1182,20 @@ g_raid_md_ctl_nvidia(struct g_raid_md_object *md,
 			gctl_error(req, "Size too small.");
 			return (-13);
 		}
-		if (size > 0xffffffffffffllu * sectorsize) {
+
+		if (level == G_RAID_VOLUME_RL_RAID0 ||
+		    level == G_RAID_VOLUME_RL_CONCAT ||
+		    level == G_RAID_VOLUME_RL_SINGLE)
+			volsize = size * numdisks;
+		else if (level == G_RAID_VOLUME_RL_RAID1)
+			volsize = size;
+		else if (level == G_RAID_VOLUME_RL_RAID5)
+			volsize = size * (numdisks - 1);
+		else { /* RAID1E */
+			volsize = ((size * numdisks) / strip / 2) *
+			    strip;
+		}
+		if (volsize > 0xffffffffllu * sectorsize) {
 			gctl_error(req, "Size too big.");
 			return (-14);
 		}
@@ -1196,18 +1209,7 @@ g_raid_md_ctl_nvidia(struct g_raid_md_object *md,
 		vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
 		vol->v_strip_size = strip;
 		vol->v_disks_count = numdisks;
-		if (level == G_RAID_VOLUME_RL_RAID0 ||
+		vol->v_mediasize = volsize;
-		    level == G_RAID_VOLUME_RL_CONCAT ||
-		    level == G_RAID_VOLUME_RL_SINGLE)
-			vol->v_mediasize = size * numdisks;
-		else if (level == G_RAID_VOLUME_RL_RAID1)
-			vol->v_mediasize = size;
-		else if (level == G_RAID_VOLUME_RL_RAID5)
-			vol->v_mediasize = size * (numdisks - 1);
-		else { /* RAID1E */
-			vol->v_mediasize = ((size * numdisks) / strip / 2) *
-			    strip;
-		}
 		vol->v_sectorsize = sectorsize;
 		g_raid_start_volume(vol);
 

sys/geom/raid/md_promise.c

@@ -121,7 +121,8 @@ struct promise_raid_conf {
 	uint64_t	rebuild_lba64;	/* Per-volume rebuild position. */
 	uint32_t	magic_4;
 	uint32_t	magic_5;
-	uint32_t	filler3[325];
+	uint32_t	total_sectors_high;
+	uint32_t	filler3[324];
 	uint32_t	checksum;
 } __packed;
 
@@ -213,6 +214,7 @@ g_raid_md_promise_print(struct promise_raid_conf *meta)
 	printf("rebuild_lba64       %ju\n", meta->rebuild_lba64);
 	printf("magic_4             0x%08x\n", meta->magic_4);
 	printf("magic_5             0x%08x\n", meta->magic_5);
+	printf("total_sectors_high  0x%08x\n", meta->total_sectors_high);
 	printf("=================================================\n");
 }
 
@@ -867,6 +869,9 @@ g_raid_md_promise_start(struct g_raid_volume *vol)
 	vol->v_strip_size = 512 << meta->stripe_shift; //ZZZ
 	vol->v_disks_count = meta->total_disks;
 	vol->v_mediasize = (off_t)meta->total_sectors * 512; //ZZZ
+	if (meta->total_sectors_high < 256) /* If value looks sane. */
+		vol->v_mediasize |=
+		    ((off_t)meta->total_sectors_high << 32) * 512; //ZZZ
 	vol->v_sectorsize = 512; //ZZZ
 	for (i = 0; i < vol->v_disks_count; i++) {
 		sd = &vol->v_subdisks[i];
@@ -1318,6 +1323,13 @@ g_raid_md_ctl_promise(struct g_raid_md_object *md,
 			cp->private = disk;
 			g_topology_unlock();
 
+			if (pp->mediasize / pp->sectorsize > UINT32_MAX) {
+				gctl_error(req,
+				    "Disk '%s' is too big.", diskname);
+				error = -8;
+				break;
+			}
+
 			/* Read kernel dumping information. */
 			disk->d_kd.offset = 0;
 			disk->d_kd.length = OFF_MAX;
@@ -1609,8 +1621,17 @@ g_raid_md_ctl_promise(struct g_raid_md_object *md,
 				error = -4;
 				break;
 			}
+			pp = cp->provider;
 			g_topology_unlock();
 
+			if (pp->mediasize / pp->sectorsize > UINT32_MAX) {
+				gctl_error(req,
+				    "Disk '%s' is too big.", diskname);
+				g_raid_kill_consumer(sc, cp);
+				error = -8;
+				break;
+			}
+
 			pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
 
 			disk = g_raid_create_disk(sc);
@@ -1716,6 +1737,8 @@ g_raid_md_write_promise(struct g_raid_md_object *md, struct g_raid_volume *tvol,
 			meta->array_width /= 2;
 		meta->array_number = vol->v_global_id;
 		meta->total_sectors = vol->v_mediasize / vol->v_sectorsize;
+		meta->total_sectors_high =
+		    (vol->v_mediasize / vol->v_sectorsize) >> 32;
 		meta->cylinders = meta->total_sectors / (255 * 63) - 1;
 		meta->heads = 254;
 		meta->sectors = 63;