X-Git-Url: http://git.ipfire.org/?p=thirdparty%2Fmdadm.git;a=blobdiff_plain;f=Create.c;h=c1d607fa7c20d224aeb522e241e722c9baaaa872;hp=2a4335a891cf882c7b42ff3367245518aa6887e5;hb=2f6079dc96180335fdb56d0b89e462e903abbb50;hpb=5b28bd56725371d4a72988d4e72bb2fd852c6fa7

diff --git a/Create.c b/Create.c
index 2a4335a8..c1d607fa 100644
--- a/Create.c
+++ b/Create.c
@@ -63,25 +63,44 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 	int fail=0, warn=0;
 	struct stat stb;
 	int first_missing = subdevs * 2;
+	int second_missing = subdevs * 2;
 	int missing_disks = 0;
 	int insert_point = subdevs * 2; /* where to insert a missing drive */
-	void *super;
 	int pass;
 	int vers;
 	int rv;
 	int bitmap_fd;
+	int have_container = 0;
 	unsigned long long bitmapsize;
+	struct mdinfo *sra;
+	struct mdinfo info;
 
-	mdu_array_info_t array;
-	int major = BITMAP_MAJOR_HI;
+	int major_num = BITMAP_MAJOR_HI;
 
-	memset(&array, 0, sizeof(array));
+	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) {
+		/* "ddf" metadata only supports one level - should possibly
+		 * push this into metadata handler??
+		 */
+		if (st && st->ss == &super_ddf)
+			level = LEVEL_CONTAINER;
+	}
+
 	if (level == UnSet) {
 		fprintf(stderr,
 			Name ": a RAID level is needed to create an array.\n");
@@ -102,11 +121,52 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 			Name ": at least 2 raid-devices needed for level 4 or 5\n");
 		return 1;
 	}
+	if (level <= 0 && sparedisks) {
+		fprintf(stderr,
+			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;
 	}
@@ -158,51 +218,50 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 	case 10:
 	case 6:
 	case 0:
-	case -1: /* linear */
+	case LEVEL_LINEAR: /* linear */
 		if (chunk == 0) {
 			chunk = 64;
 			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 > 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 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;
-		}
-		if (!get_dev_size(fd, dname, &ldsize)) {
-			fail = 1;
-			close(fd);
-			continue;
-		}
+			info.array.active_disks++;
 		if (st == NULL) {
 			struct createinfo *ci = conf_get_create_info();
 			if (ci)
@@ -210,32 +269,43 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 		}
 		if (st == NULL) {
 			/* Need to choose a default metadata, which is different
-			 * depending on the sizes of devices
+			 * depending on geometry of array.
 			 */
 			int i;
 			char *name = "default";
-			if (level >= 1 && ldsize > (0x7fffffffULL<<10))
-				name = "default/large";
-			for(i=0; !st && superlist[i]; i++)
+			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 ": internal error - no default metadata style\n");
+				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 verion %d.%d metadata\n",
+				fprintf(stderr, Name ": Defaulting to version"
+					" %d.%d metadata\n",
 					st->ss->major,
 					st->minor_version);
-		}
-		freesize = st->ss->avail_size(st, ldsize >> 9);
-		if (freesize == 0) {
-			fprintf(stderr, Name ": %s is too small: %luK\n",
-				dname, (unsigned long)(ldsize>>10));
-			fail = 1;
-			close(fd);
-			continue;
+		} 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;
+			}
 		}
 
 		freesize /= 2; /* convert to K */
@@ -246,9 +316,9 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 
 		if (size && freesize < size) {
 			fprintf(stderr, Name ": %s is smaller that given size."
-				" %lluK < %lluK + superblock\n", dname, freesize, size);
+				" %lluK < %lluK + metadata\n",
+				dname, freesize, size);
 			fail = 1;
-			close(fd);
 			continue;
 		}
 		if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
@@ -260,25 +330,34 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 			minsize = freesize;
 		}
 		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);
 		}
-		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 || level == LEVEL_MULTIPATH || level == LEVEL_FAULTY) {
+		if (level > 0 || level == LEVEL_MULTIPATH
+		    || level == LEVEL_FAULTY
+		    || (st && st->ss->external) ) {
 			/* size is meaningful */
 			if (minsize > 0x100000000ULL && st->ss->major == 0) {
-				fprintf(stderr, Name ": devices too large for RAID level %d\n", level);	
+				fprintf(stderr, Name ": devices too large for RAID level %d\n", level);
 				return 1;
 			}
 			size = minsize;
@@ -305,46 +384,66 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 		}
 	}
 
-	/* 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 (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,
@@ -376,12 +475,13 @@ int Create(struct supertype *st, 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;
-	array.major_version = st->ss->major;
+	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 */
@@ -407,11 +507,11 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 				name += 2;
 		}
 	}
-	if (!st->ss->init_super(st, &super, &array, size, name, homehost, uuid))
+	if (!st->ss->init_super(st, &info.array, size, name, homehost, uuid))
 		return 1;
 
 	if (bitmap_file && vers < 9003) {
-		major = BITMAP_MAJOR_HOSTENDIAN;
+		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");
@@ -423,9 +523,9 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 			fprintf(stderr, Name ": internal bitmaps not supported by this kernel.\n");
 			return 1;
 		}
-		if (!st->ss->add_internal_bitmap(st, super, &bitmap_chunk,
+		if (!st->ss->add_internal_bitmap(st, &bitmap_chunk,
 						 delay, write_behind,
-						 bitmapsize, 1, major)) {
+						 bitmapsize, 1, major_num)) {
 			fprintf(stderr, Name ": Given bitmap chunk size not supported.\n");
 			return 1;
 		}
@@ -433,14 +533,35 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 	}
 
 
+	sra = sysfs_read(mdfd, 0, 0);
 
-	if ((vers % 100) >= 1) { /* can use different versions */
+	if (st->ss->external) {
+		char ver[100];
+		if (st->ss->external == 1)
+			/* container */
+			strcat(strcpy(ver, "external:"), st->ss->text_version);
+		else {
+			/* member */
+			sprintf(ver, "external:/%s/%d",
+				devnum2devname(st->container_dev),
+				st->container_member);
+		}
+		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 
+	} else
 		rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
 	if (rv) {
 		fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
@@ -451,11 +572,11 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 	if (bitmap_file) {
 		int uuid[4];
 
-		st->ss->uuid_from_super(uuid, super);
+		st->ss->uuid_from_super(st, uuid);
 		if (CreateBitmap(bitmap_file, force, (char*)uuid, bitmap_chunk,
 				 delay, write_behind,
 				 bitmapsize,
-				 major)) {
+				 major_num)) {
 			return 1;
 		}
 		bitmap_fd = open(bitmap_file, O_RDWR);
@@ -472,7 +593,6 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 	}
 
 
-
 	for (pass=1; pass <=2 ; pass++) {
 		mddev_dev_t moved_disk = NULL; /* the disk that was moved out of the insert point */
 
@@ -480,71 +600,102 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 		     dv=(dv->next)?(dv->next):moved_disk, dnum++) {
 			int fd;
 			struct stat stb;
-			mdu_disk_info_t disk;
 
-			disk.number = dnum;
+			info.disk.number = dnum;
 			if (dnum == insert_point) {
 				moved_disk = dv;
 			}
-			disk.raid_disk = disk.number;
-			if (disk.raid_disk < raiddisks)
-				disk.state = (1<<MD_DISK_ACTIVE) |
+			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
-				disk.state = 0;
+				info.disk.state = 0;
 			if (dv->writemostly)
-				disk.state |= (1<<MD_DISK_WRITEMOSTLY);
+				info.disk.state |= (1<<MD_DISK_WRITEMOSTLY);
 
 			if (dnum == insert_point ||
-			    strcasecmp(dv->devname, "missing")==0) {
-				disk.major = 0;
-				disk.minor = 0;
-				disk.state = (1<<MD_DISK_FAULTY);
-			} else {
-				fd = open(dv->devname, O_RDONLY|O_EXCL, 0);
-				if (fd < 0) {
-					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);
-				remove_partitions(fd);
-				close(fd);
+			    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",
+					dv->devname);
+				return 1;
 			}
+			fstat(fd, &stb);
+			info.disk.major = major(stb.st_rdev);
+			info.disk.minor = minor(stb.st_rdev);
+
 			switch(pass){
 			case 1:
-				st->ss->add_to_super(super, &disk);
+				remove_partitions(fd);
+				st->ss->add_to_super(st, &info.disk,
+						     fd, dv->devname);
 				break;
 			case 2:
-				if (disk.state == 1) break;
-				Kill(dv->devname, 0, 1); /* Just be sure it is clean */
-				Kill(dv->devname, 0, 1); /* and again, there could be two superblocks */
-				st->ss->write_init_super(st, super, &disk, dv->devname);
-
-				if (ioctl(mdfd, ADD_NEW_DISK, &disk)) {
-					fprintf(stderr, Name ": ADD_NEW_DISK for %s failed: %s\n",
+				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, fd, &info);
+					close(fd);
+				} else {
+					close(fd);
+					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));
-					free(super);
+					st->ss->free_super(st);
 					return 1;
 				}
-
 				break;
 			}
 			if (dv == moved_disk && dnum != insert_point) break;
 		}
+		if (pass == 1)
+			st->ss->write_init_super(st);
 	}
-	free(super);
+	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, 0);
-			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");
+				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);