X-Git-Url: http://git.ipfire.org/?p=thirdparty%2Fmdadm.git;a=blobdiff_plain;f=Create.c;h=b7f0914329b79940db80577b615e3a73e6ea4052;hp=66a88d2008bd03a950971048437c753207df4bb6;hb=159c3a1a7785693d21f5226182f9c066882bef66;hpb=d7eaf49f65ecb99e24255026331b0af7a9245cc2

diff --git a/Create.c b/Create.c
index 66a88d20..b7f09143 100644
--- a/Create.c
+++ b/Create.c
@@ -1,7 +1,7 @@
 /*
  * mdadm - manage Linux "md" devices aka RAID arrays.
  *
- * Copyright (C) 2001-2002 Neil Brown <neilb@cse.unsw.edu.au>
+ * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de>
  *
  *
  *    This program is free software; you can redistribute it and/or modify
@@ -30,11 +30,14 @@
 #include "mdadm.h"
 #include	"md_u.h"
 #include	"md_p.h"
+#include	<ctype.h>
 
-int Create(char *mddev, int mdfd,
-	   int chunk, int level, int layout, unsigned long size, int raiddisks, int sparedisks,
+int Create(struct supertype *st, char *mddev, int mdfd,
+	   int chunk, int level, int layout, unsigned long long size, int raiddisks, int sparedisks,
+	   char *name, char *homehost, int *uuid,
 	   int subdevs, mddev_dev_t devlist,
-	   int runstop, int verbose, int force)
+	   int runstop, int verbose, int force, int assume_clean,
+	   char *bitmap_file, int bitmap_chunk, int write_behind, int delay)
 {
 	/*
 	 * Create a new raid array.
@@ -49,7 +52,7 @@ int Create(char *mddev, int mdfd,
 	 * abort.
 	 *
 	 * SET_ARRAY_INFO and ADD_NEW_DISK, and
-	 * if runstop==run, or raiddisks diskswere used,
+	 * if runstop==run, or raiddisks disks were used,
 	 * RUN_ARRAY
 	 */
 	unsigned long long minsize=0, maxsize=0;
@@ -59,26 +62,50 @@ int Create(char *mddev, int mdfd,
 	mddev_dev_t dv;
 	int fail=0, warn=0;
 	struct stat stb;
-	int first_missing = MD_SB_DISKS*2;
+	int first_missing = subdevs * 2;
+	int second_missing = subdevs * 2;
 	int missing_disks = 0;
-	int insert_point = MD_SB_DISKS*2; /* where to insert a missing drive */
-	mddev_dev_t moved_disk = NULL; /* the disk that was moved out of the insert point */
+	int insert_point = subdevs * 2; /* where to insert a missing drive */
+	int pass;
+	int vers;
+	int rv;
+	int bitmap_fd;
+	int have_container = 0;
+	int container_fd;
+	int need_mdmon = 0;
+	unsigned long long bitmapsize;
+	struct mdinfo *sra;
+	struct mdinfo info;
 
-	mdu_array_info_t array;
-	
+	int major_num = BITMAP_MAJOR_HI;
 
-	if (md_get_version(mdfd) < 9000) {
+	memset(&info, 0, sizeof(info));
+
+	vers = md_get_version(mdfd);
+	if (vers < 9000) {
 		fprintf(stderr, Name ": Create requires md driver version 0.90.0 or later\n");
 		return 1;
+	} else {
+		mdu_array_info_t inf;
+		memset(&inf, 0, sizeof(inf));
+		ioctl(mdfd, GET_ARRAY_INFO, &inf);
+		if (inf.working_disks != 0) {
+			fprintf(stderr, Name ": another array by this name"
+				" is already running.\n");
+			return 1;
+		}
 	}
 	if (level == UnSet) {
-		fprintf(stderr,
-			Name ": a RAID level is needed to create an array.\n");
-		return 1;
+		/* "ddf" and "imsm" metadata only supports one level - should possibly
+		 * push this into metadata handler??
+		 */
+		if (st && (st->ss == &super_ddf || st->ss == &super_imsm))
+			level = LEVEL_CONTAINER;
 	}
-	if (raiddisks < 1) {
+
+	if (level == UnSet) {
 		fprintf(stderr,
-			Name ": a number of --raid-devices must be given to create an array\n");
+			Name ": a RAID level is needed to create an array.\n");
 		return 1;
 	}
 	if (raiddisks < 4 && level == 6) {
@@ -96,20 +123,60 @@ int Create(char *mddev, int mdfd,
 			Name ": at least 2 raid-devices needed for level 4 or 5\n");
 		return 1;
 	}
-	if (raiddisks+sparedisks > MD_SB_DISKS) {
+	if (level <= 0 && sparedisks) {
 		fprintf(stderr,
-			Name ": too many devices requested: %d+%d > %d\n",
-			raiddisks, sparedisks, MD_SB_DISKS);
+			Name ": This level does not support spare devices\n");
 		return 1;
 	}
+
+	if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) {
+		/* If given a single device, it might be a container, and we can
+		 * extract a device list from there
+		 */
+		mdu_array_info_t inf;
+		int fd;
+
+		memset(&inf, 0, sizeof(inf));
+		fd = open(devlist->devname, O_RDONLY, 0);
+		if (fd >= 0 &&
+		    ioctl(fd, GET_ARRAY_INFO, &inf) == 0 &&
+		    inf.raid_disks == 0) {
+			/* yep, looks like a container */
+			if (st) {
+				rv = st->ss->load_super(st, fd,
+							devlist->devname);
+				if (rv == 0)
+					have_container = 1;
+			} else {
+				st = guess_super(fd);
+				if (st && !(rv = st->ss->
+					    load_super(st, fd,
+						       devlist->devname)))
+					have_container = 1;
+				else
+					st = NULL;
+			}
+		}
+		if (fd >= 0)
+			close(fd);
+		if (have_container) {
+			subdevs = 0;
+			devlist = NULL;
+		}
+	}
 	if (subdevs > raiddisks+sparedisks) {
 		fprintf(stderr, Name ": You have listed more devices (%d) than are in the array(%d)!\n", subdevs, raiddisks+sparedisks);
 		return 1;
 	}
-	if (subdevs < raiddisks+sparedisks) {
+	if (!have_container && subdevs < raiddisks+sparedisks) {
 		fprintf(stderr, Name ": You haven't given enough devices (real or missing) to create this array\n");
 		return 1;
 	}
+	if (bitmap_file && level <= 0) {
+		fprintf(stderr, Name ": bitmaps not meaningful with level %s\n",
+			map_num(pers, level)?:"given");
+		return 1;
+	}
 
 	/* now set some defaults */
 	if (layout == UnSet)
@@ -119,21 +186,21 @@ int Create(char *mddev, int mdfd,
 			break;
 		case 10:
 			layout = 0x102; /* near=2, far=1 */
-			if (verbose)
+			if (verbose > 0)
 				fprintf(stderr,
 					Name ": layout defaults to n1\n");
 			break;
 		case 5:
 		case 6:
 			layout = map_name(r5layout, "default");
-			if (verbose)
+			if (verbose > 0)
 				fprintf(stderr,
 					Name ": layout defaults to %s\n", map_num(r5layout, layout));
 			break;
 		case LEVEL_FAULTY:
 			layout = map_name(faultylayout, "default");
 
-			if (verbose)
+			if (verbose > 0)
 				fprintf(stderr,
 					Name ": layout defaults to %s\n", map_num(faultylayout, layout));
 			break;
@@ -153,78 +220,107 @@ int Create(char *mddev, int mdfd,
 	case 10:
 	case 6:
 	case 0:
-	case -1: /* linear */
+	case LEVEL_LINEAR: /* linear */
 		if (chunk == 0) {
 			chunk = 64;
-			if (verbose)
+			if (verbose > 0)
 				fprintf(stderr, Name ": chunk size defaults to 64K\n");
 		}
 		break;
-	default: /* raid1, multipath */
+	case 1:
+	case LEVEL_FAULTY:
+	case LEVEL_MULTIPATH:
+	case LEVEL_CONTAINER:
 		if (chunk) {
 			chunk = 0;
-			if (verbose)
+			if (verbose > 0)
 				fprintf(stderr, Name ": chunk size ignored for this level\n");
 		}
 		break;
+	default:
+		fprintf(stderr, Name ": unknown level %d\n", level);
+		return 1;
 	}
 
+	if (st && ! st->ss->validate_geometry(st, level, layout, raiddisks,
+					      chunk, size, NULL, NULL))
+		return 1;
+
 	/* now look at the subdevs */
-	array.active_disks = 0;
-	array.working_disks = 0;
+	info.array.active_disks = 0;
+	info.array.working_disks = 0;
 	dnum = 0;
 	for (dv=devlist; dv; dv=dv->next, dnum++) {
 		char *dname = dv->devname;
-		unsigned long dsize;
-		unsigned long long ldsize, freesize;
-		int fd;
+		unsigned long long freesize;
 		if (strcasecmp(dname, "missing")==0) {
 			if (first_missing > dnum)
 				first_missing = dnum;
+			if (second_missing > dnum && dnum > first_missing)
+				second_missing = dnum;
 			missing_disks ++;
 			continue;
 		}
-		array.working_disks++;
+		info.array.working_disks++;
 		if (dnum < raiddisks)
-			array.active_disks++;
-		fd = open(dname, O_RDONLY|O_EXCL, 0);
-		if (fd <0 ) {
-			fprintf(stderr, Name ": Cannot open %s: %s\n",
-				dname, strerror(errno));
-			fail=1;
-			continue;
+			info.array.active_disks++;
+		if (st == NULL) {
+			struct createinfo *ci = conf_get_create_info();
+			if (ci)
+				st = ci->supertype;
 		}
-#ifdef BLKGETSIZE64
-		if (ioctl(fd, BLKGETSIZE64, &ldsize)==0)
-			;
-		else
-#endif
-		if (ioctl(fd, BLKGETSIZE, &dsize)) {
-			fprintf(stderr, Name ": Cannot get size of %s: %s\n",
-				dname, strerror(errno));
-			fail = 1;
-			close(fd);
-			continue;
-		}
-		else {
-			ldsize = dsize;
-			dsize <<= 9;
+		if (st == NULL) {
+			/* Need to choose a default metadata, which is different
+			 * depending on geometry of array.
+			 */
+			int i;
+			char *name = "default";
+			for(i=0; !st && superlist[i]; i++) {
+				st = superlist[i]->match_metadata_desc(name);
+				if (st && !st->ss->validate_geometry
+					    	(st, level, layout, raiddisks,
+						 chunk, size, dname, &freesize))
+					st = NULL;
+			}
+
+			if (!st) {
+				fprintf(stderr, Name ": device %s not suitable "
+					"for any style of array\n",
+					dname);
+				exit(2);
+			}
+			if (st->ss->major != 0 ||
+			    st->minor_version != 90)
+				fprintf(stderr, Name ": Defaulting to version"
+					" %d.%d metadata\n",
+					st->ss->major,
+					st->minor_version);
+		} else {
+			if (!st->ss->validate_geometry(st, level, layout,
+						       raiddisks,
+						       chunk, size, dname,
+						       &freesize)) {
+
+				fprintf(stderr,
+					Name ": %s is not suitable for "
+					"this array.\n",
+					dname);
+				fail = 1;
+				continue;
+			}
 		}
-		if (ldsize < MD_RESERVED_SECTORS*2LL*512LL) {
-			fprintf(stderr, Name ": %s is too small: %luK\n",
-				dname, (unsigned long)(ldsize>>10));
-			fail = 1;
-			close(fd);
-			continue;
+
+		freesize /= 2; /* convert to K */
+		if (chunk) {
+			/* round to chunk size */
+			freesize = freesize & ~(chunk-1);
 		}
-		freesize = MD_NEW_SIZE_SECTORS((ldsize>>9));
-		freesize /= 2;
 
 		if (size && freesize < size) {
 			fprintf(stderr, Name ": %s is smaller that given size."
-				" %lluK < %luK + superblock\n", dname, freesize, size);
+				" %lluK < %lluK + metadata\n",
+				dname, freesize, size);
 			fail = 1;
-			close(fd);
 			continue;
 		}
 		if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
@@ -235,34 +331,46 @@ int Create(char *mddev, int mdfd,
 			mindisc = dname;
 			minsize = freesize;
 		}
-		warn |= check_ext2(fd, dname);
-		warn |= check_reiser(fd, dname);
-		warn |= check_raid(fd, dname);
-		close(fd);
+		if (runstop != 1 || verbose >= 0) {
+			int fd = open(dname, O_RDONLY, 0);
+			if (fd <0 ) {
+				fprintf(stderr, Name ": Cannot open %s: %s\n",
+					dname, strerror(errno));
+				fail=1;
+				continue;
+			}
+			warn |= check_ext2(fd, dname);
+			warn |= check_reiser(fd, dname);
+			warn |= check_raid(fd, dname);
+			close(fd);
+		}
 	}
 	if (fail) {
 		fprintf(stderr, Name ": create aborted\n");
 		return 1;
 	}
 	if (size == 0) {
-		if (mindisc == NULL) {
+		if (mindisc == NULL && !have_container) {
 			fprintf(stderr, Name ": no size and no drives given - aborting create.\n");
 			return 1;
 		}
-		if (level > 0) {
+		if (level > 0 || level == LEVEL_MULTIPATH
+		    || level == LEVEL_FAULTY
+		    || (st && st->ss->external) ) {
 			/* size is meaningful */
-			if (minsize > 0x100000000ULL) {
-				fprintf(stderr, Name ": devices too large for RAID level %d\n", level);	
+			if (minsize > 0x100000000ULL && st->ss->major == 0) {
+				fprintf(stderr, Name ": devices too large for RAID level %d\n", level);
 				return 1;
 			}
 			size = minsize;
-			if (verbose)
-				fprintf(stderr, Name ": size set to %luK\n", size);
+			if (verbose > 0)
+				fprintf(stderr, Name ": size set to %lluK\n", size);
 		}
 	}
 	if (level > 0 && ((maxsize-size)*100 > maxsize)) {
-		fprintf(stderr, Name ": largest drive (%s) exceed size (%luK) by more than 1%%\n",
-			maxdisc, size);
+		if (runstop != 1 || verbose >= 0)
+			fprintf(stderr, Name ": largest drive (%s) exceed size (%lluK) by more than 1%%\n",
+				maxdisc, size);
 		warn = 1;
 	}
 
@@ -273,49 +381,82 @@ int Create(char *mddev, int mdfd,
 				return 1;
 			}
 		} else {
-			if (verbose)
+			if (verbose > 0)
 				fprintf(stderr, Name ": creation continuing despite oddities due to --run\n");
 		}
 	}
 
-	/* If this is  raid5, we want to configure the last active slot
+	/* If this is raid4/5, we want to configure the last active slot
 	 * as missing, so that a reconstruct happens (faster than re-parity)
 	 * FIX: Can we do this for raid6 as well?
 	 */
-	if (force == 0 && first_missing >= raiddisks) {
+	if (assume_clean==0 && force == 0 && first_missing >= raiddisks) {
 		switch ( level ) {
+		case 4:
 		case 5:
 			insert_point = raiddisks-1;
 			sparedisks++;
-			array.active_disks--;
+			info.array.active_disks--;
 			missing_disks++;
 			break;
 		default:
 			break;
 		}
 	}
-	
+	/* For raid6, if creating with 1 missing drive, make a good drive
+	 * into a spare, else the create will fail
+	 */
+	if (assume_clean == 0 && force == 0 && first_missing < raiddisks &&
+	    second_missing >= raiddisks && level == 6) {
+		insert_point = raiddisks - 1;
+		if (insert_point == first_missing)
+			insert_point--;
+		sparedisks ++;
+		info.array.active_disks--;
+		missing_disks++;
+	}
+
+	if (level <= 0 && first_missing < subdevs * 2) {
+		fprintf(stderr,
+			Name ": This level does not support missing devices\n");
+		return 1;
+	}
+
 	/* Ok, lets try some ioctls */
 
-	array.level = level;
-	array.size = size;
-	array.raid_disks = raiddisks;
+	info.array.level = level;
+	info.array.size = size;
+	info.array.raid_disks = raiddisks;
 	/* The kernel should *know* what md_minor we are dealing
 	 * with, but it chooses to trust me instead. Sigh
 	 */
-	array.md_minor = 0;
+	info.array.md_minor = 0;
 	if (fstat(mdfd, &stb)==0)
-		array.md_minor = minor(stb.st_rdev);
-	array.not_persistent = 0;
-	/*** FIX: Need to do something about RAID-6 here ***/
-	if ( ( (level == 5) &&
+		info.array.md_minor = minor(stb.st_rdev);
+	info.array.not_persistent = 0;
+
+	if ( ( (level == 4 || level == 5) &&
 	       (insert_point < raiddisks || first_missing < raiddisks) )
 	     ||
-	     ( level == 6 && missing_disks == 2)
+	     ( level == 6 && (insert_point < raiddisks
+			      || second_missing < raiddisks))
+	     ||
+	     assume_clean
 		)
-		array.state = 1; /* clean, but one+ drive will be missing */
+		info.array.state = 1; /* clean, but one+ drive will be missing*/
 	else
-		array.state = 0; /* not clean, but no errors */
+		info.array.state = 0; /* not clean, but no errors */
+
+	if (level == 10) {
+		/* for raid10, the bitmap size is the capacity of the array,
+		 * which is array.size * raid_disks / ncopies;
+		 * .. but convert to sectors.
+		 */
+		int ncopies = ((layout>>8) & 255) * (layout & 255);
+		bitmapsize = (unsigned long long)size * raiddisks / ncopies * 2;
+/*		printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
+	} else
+		bitmapsize = (unsigned long long)size * 2;
 
 	/* There is lots of redundancy in these disk counts,
 	 * raid_disks is the most meaningful value
@@ -336,67 +477,283 @@ int Create(char *mddev, int mdfd,
 	 * So for now, we assume that all raid and spare
 	 * devices will be given.
 	 */
-	array.spare_disks=sparedisks;
-	array.failed_disks=missing_disks;
-	array.nr_disks = array.working_disks + array.failed_disks;
-	array.layout = layout;
-	array.chunk_size = chunk*1024;
+	info.array.spare_disks=sparedisks;
+	info.array.failed_disks=missing_disks;
+	info.array.nr_disks = info.array.working_disks
+		+ info.array.failed_disks;
+	info.array.layout = layout;
+	info.array.chunk_size = chunk*1024;
+	info.array.major_version = st->ss->major;
+
+	if (name == NULL || *name == 0) {
+		/* base name on mddev */
+		/*  /dev/md0 -> 0
+		 *  /dev/md_d0 -> d0
+		 *  /dev/md/1 -> 1
+		 *  /dev/md/d1 -> d1
+		 *  /dev/md/home -> home
+		 *  /dev/mdhome -> home
+		 */
+		name = strrchr(mddev, '/');
+		if (name) {
+			name++;
+			if (strncmp(name, "md_d", 4)==0 &&
+			    strlen(name) > 4 &&
+			    isdigit(name[4]) &&
+			    (name-mddev) == 5 /* /dev/ */)
+				name += 3;
+			else if (strncmp(name, "md", 2)==0 &&
+				 strlen(name) > 2 &&
+				 isdigit(name[2]) &&
+				 (name-mddev) == 5 /* /dev/ */)
+				name += 2;
+		}
+	}
+	if (!st->ss->init_super(st, &info.array, size, name, homehost, uuid))
+		return 1;
+
+	st->ss->getinfo_super(st, &info);
+
+	if (bitmap_file && vers < 9003) {
+		major_num = BITMAP_MAJOR_HOSTENDIAN;
+#ifdef __BIG_ENDIAN
+		fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n"
+			"  between different architectured.  Consider upgrading the Linux kernel.\n");
+#endif
+	}
+
+	if (bitmap_file && strcmp(bitmap_file, "internal")==0) {
+		if ((vers%100) < 2) {
+			fprintf(stderr, Name ": internal bitmaps not supported by this kernel.\n");
+			return 1;
+		}
+		if (!st->ss->add_internal_bitmap(st, &bitmap_chunk,
+						 delay, write_behind,
+						 bitmapsize, 1, major_num)) {
+			fprintf(stderr, Name ": Given bitmap chunk size not supported.\n");
+			return 1;
+		}
+		bitmap_file = NULL;
+	}
+
+
+	sra = sysfs_read(mdfd, 0, 0);
 
-	if (ioctl(mdfd, SET_ARRAY_INFO, &array)) {
+	if (st->ss->external) {
+		char ver[100];
+		if (st->ss->external == 1)
+			/* container */
+			strcat(strcpy(ver, "external:"),
+			       info.text_version);
+		else {
+			/* member */
+			sprintf(ver, "external:/%s/%d",
+				devnum2devname(st->container_dev),
+				st->container_member);
+			/* When creating a member, we need to be careful
+			 * to negotiate with mdmon properly.
+			 * If it is already running, we cannot write to
+			 * the devices and must ask it to do that part.
+			 * If it isn't running, we write to the devices,
+			 * and then start it.
+			 * We hold an exclusive open on the container
+			 * device to make sure mdmon doesn't exit after
+			 * we checked that it is running.
+			 *
+			 * For now, fail if it is already running.
+			 */
+			container_fd = open_dev_excl(st->container_dev);
+			if (container_fd < 0) {
+				fprintf(stderr, Name ": Cannot get exclusive "
+					"open on container - weird.\n");
+				return 1;
+			}
+			if (mdmon_running(st->container_dev)) {
+				fprintf(stderr, Name ": mdmon already running "
+					"for %s - sorry\n",
+					devnum2devname(st->container_dev));
+				return 1;
+			}
+			need_mdmon = 1;
+		}
+		if ((vers % 100) < 2 ||
+		    sra == NULL ||
+		    sysfs_set_str(sra, NULL, "metadata_version",
+				  ver) < 0) {
+			fprintf(stderr, Name ": This kernel does not "
+				"support external metadata.\n");
+			return 1;
+		}
+		rv = sysfs_set_array(sra, &info);
+	} else	if ((vers % 100) >= 1) { /* can use different versions */
+		mdu_array_info_t inf;
+		memset(&inf, 0, sizeof(inf));
+		inf.major_version = st->ss->major;
+		inf.minor_version = st->minor_version;
+		rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
+	} else
+		rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
+	if (rv) {
 		fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
 			mddev, strerror(errno));
 		return 1;
 	}
-	
-	for (dnum=0, dv = devlist ; dv ; dv=(dv->next)?(dv->next):moved_disk, dnum++) {
-		int fd;
-		struct stat stb;
-		mdu_disk_info_t disk;
 
-		disk.number = dnum;
-		if (dnum == insert_point) {
-			moved_disk = dv;
+	if (bitmap_file) {
+		int uuid[4];
+
+		st->ss->uuid_from_super(st, uuid);
+		if (CreateBitmap(bitmap_file, force, (char*)uuid, bitmap_chunk,
+				 delay, write_behind,
+				 bitmapsize,
+				 major_num)) {
+			return 1;
 		}
-		disk.raid_disk = disk.number;
-		if (disk.raid_disk < raiddisks)
-			disk.state = 6; /* active and in sync */
-		else
-			disk.state = 0;
-		if (dnum == insert_point ||
-		    strcasecmp(dv->devname, "missing")==0) {
-			disk.major = 0;
-			disk.minor = 0;
-			disk.state = 1; /* faulty */
-		} else {
-			fd = open(dv->devname, O_RDONLY|O_EXCL, 0);
+		bitmap_fd = open(bitmap_file, O_RDWR);
+		if (bitmap_fd < 0) {
+			fprintf(stderr, Name ": weird: %s cannot be openned\n",
+				bitmap_file);
+			return 1;
+		}
+		if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
+			fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
+				mddev, strerror(errno));
+			return 1;
+		}
+	}
+
+
+	for (pass=1; pass <=2 ; pass++) {
+		mddev_dev_t moved_disk = NULL; /* the disk that was moved out of the insert point */
+
+		for (dnum=0, dv = devlist ; dv ;
+		     dv=(dv->next)?(dv->next):moved_disk, dnum++) {
+			int fd;
+			struct stat stb;
+
+			info.disk.number = dnum;
+			if (dnum == insert_point) {
+				moved_disk = dv;
+			}
+			info.disk.raid_disk = info.disk.number;
+			if (info.disk.raid_disk < raiddisks)
+				info.disk.state = (1<<MD_DISK_ACTIVE) |
+						(1<<MD_DISK_SYNC);
+			else
+				info.disk.state = 0;
+			if (dv->writemostly)
+				info.disk.state |= (1<<MD_DISK_WRITEMOSTLY);
+
+			if (dnum == insert_point ||
+			    strcasecmp(dv->devname, "missing")==0)
+				continue;
+
+			if (st->ss->external == 2)
+				fd = open(dv->devname, O_RDWR, 0);
+			else
+				fd = open(dv->devname, O_RDWR|O_EXCL,0);
+
 			if (fd < 0) {
-				fprintf(stderr, Name ": failed to open %s after earlier success - aborting\n",
+				fprintf(stderr, Name ": failed to open %s "
+					"after earlier success - aborting\n",
 					dv->devname);
 				return 1;
 			}
 			fstat(fd, &stb);
-			disk.major = major(stb.st_rdev);
-			disk.minor = minor(stb.st_rdev);
-			close(fd);
-		}
-		if (ioctl(mdfd, ADD_NEW_DISK, &disk)) {
-			fprintf(stderr, Name ": ADD_NEW_DISK for %s failed: %s\n",
-				dv->devname, strerror(errno));
-			return 1;
+			info.disk.major = major(stb.st_rdev);
+			info.disk.minor = minor(stb.st_rdev);
+
+			switch(pass){
+			case 1:
+				remove_partitions(fd);
+				st->ss->add_to_super(st, &info.disk,
+						     fd, dv->devname);
+				break;
+			case 2:
+				close(fd);
+				info.component_size = info.array.size * 2;
+				info.errors = 0;
+				rv = 0;
+
+				if (st->ss->external) {
+					st->ss->getinfo_super_n(st, &info);
+					rv = sysfs_add_disk(sra, &info);
+				} else {
+					rv = ioctl(mdfd, ADD_NEW_DISK,
+						 &info.disk);
+				}
+				if (rv) {
+					fprintf(stderr,
+						Name ": ADD_NEW_DISK for %s "
+						"failed: %s\n",
+						dv->devname, strerror(errno));
+					st->ss->free_super(st);
+					return 1;
+				}
+				break;
+			}
+			if (dv == moved_disk && dnum != insert_point) break;
 		}
-		if (dv == moved_disk && dnum != insert_point) break;
+		if (pass == 1)
+			st->ss->write_init_super(st);
 	}
+	st->ss->free_super(st);
 
 	/* param is not actually used */
-	if (runstop == 1 || subdevs >= raiddisks) {
-		mdu_param_t param;
-		if (ioctl(mdfd, RUN_ARRAY, &param)) {
-			fprintf(stderr, Name ": RUN_ARRAY failed: %s\n",
-				strerror(errno));
-			Manage_runstop(mddev, mdfd, -1);
-			return 1;
+	if (level == LEVEL_CONTAINER)
+		/* No need to start */
+		;
+	else if (runstop == 1 || subdevs >= raiddisks) {
+		if (st->ss->external) {
+			switch(level) {
+			case LEVEL_LINEAR:
+			case LEVEL_MULTIPATH:
+			case 0:
+				sysfs_set_str(sra, NULL, "array_state",
+					      "active");
+				need_mdmon = 0;
+				break;
+			default:
+				sysfs_set_str(sra, NULL, "array_state",
+					      "readonly");
+				break;
+			}
+		} else {
+			mdu_param_t param;
+			if (ioctl(mdfd, RUN_ARRAY, &param)) {
+				fprintf(stderr, Name ": RUN_ARRAY failed: %s\n",
+					strerror(errno));
+				Manage_runstop(mddev, mdfd, -1, 0);
+				return 1;
+			}
+		}
+		if (verbose >= 0)
+			fprintf(stderr, Name ": array %s started.\n", mddev);
+		if (st->ss->external == 2) {
+			if (need_mdmon) {
+				int dn = st->container_dev;
+				int i;
+				switch(fork()) {
+				case 0:
+					/* FIXME yuk. CLOSE_EXEC?? */
+					for (i=3; i < 100; i++)
+						close(i);
+					execl("./mdmon", "mdmon",
+					      map_dev(dev2major(dn),
+						      dev2minor(dn),
+						      1), NULL);
+					exit(1);
+				case -1: fprintf(stderr, Name ": cannot fork. "
+						 "Array remains readonly\n");
+					return 1;
+				default: ; /* parent - good */
+				}
+			} else
+				signal_mdmon(st->container_dev);
+			/* FIXME wait for mdmon to set array to read-auto */
+			sleep(1);
+			close(container_fd);
 		}
-		fprintf(stderr, Name ": array %s started.\n", mddev);
 	} else {
 		fprintf(stderr, Name ": not starting array - not enough devices.\n");
 	}