X-Git-Url: http://git.ipfire.org/?p=thirdparty%2Fmdadm.git;a=blobdiff_plain;f=Create.c;h=652bc1c761cbdd7e9a30839f33f29000e7d95a32;hp=d423585d9796f938e9de9102ec3c8559cdcb93ce;hb=f35f25259279573c6274e2783536c0b0a399bdd4;hpb=17f25ca6fb71221f4fa7891c0037f88d26dbc67c

diff --git a/Create.c b/Create.c
index d423585d..652bc1c7 100644
--- a/Create.c
+++ b/Create.c
@@ -66,12 +66,18 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 	int second_missing = subdevs * 2;
 	int missing_disks = 0;
 	int insert_point = subdevs * 2; /* where to insert a missing drive */
+	int total_slots;
 	int pass;
 	int vers;
 	int rv;
 	int bitmap_fd;
+	int have_container = 0;
+	int container_fd = -1;
+	int need_mdmon = 0;
 	unsigned long long bitmapsize;
-	struct mdinfo info;
+	struct mdinfo info, *infos;
+	int did_default = 0;
+	unsigned long safe_mode_delay = 0;
 
 	int major_num = BITMAP_MAJOR_HI;
 
@@ -91,6 +97,14 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 			return 1;
 		}
 	}
+	if (level == UnSet) {
+		/* "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 (level == UnSet) {
 		fprintf(stderr,
 			Name ": a RAID level is needed to create an array.\n");
@@ -116,11 +130,47 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 			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);
+		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;
 	}
@@ -195,7 +245,7 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 	}
 
 	if (st && ! st->ss->validate_geometry(st, level, layout, raiddisks,
-					      chunk, size, NULL, NULL))
+					      chunk, size, NULL, NULL, verbose>=0))
 		return 1;
 
 	/* now look at the subdevs */
@@ -205,7 +255,6 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 	for (dv=devlist; dv; dv=dv->next, dnum++) {
 		char *dname = dv->devname;
 		unsigned long long freesize;
-		int fd;
 		if (strcasecmp(dname, "missing")==0) {
 			if (first_missing > dnum)
 				first_missing = dnum;
@@ -217,13 +266,6 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 		info.array.working_disks++;
 		if (dnum < raiddisks)
 			info.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 (st == NULL) {
 			struct createinfo *ci = conf_get_create_info();
 			if (ci)
@@ -239,7 +281,8 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 				st = superlist[i]->match_metadata_desc(name);
 				if (st && !st->ss->validate_geometry
 					    	(st, level, layout, raiddisks,
-						 chunk, size, dname, &freesize))
+						 chunk, size, dname, &freesize,
+						 verbose > 0))
 					st = NULL;
 			}
 
@@ -249,17 +292,15 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 					dname);
 				exit(2);
 			}
-			if (st->ss->major != 0 ||
+			if (st->ss != &super0 ||
 			    st->minor_version != 90)
-				fprintf(stderr, Name ": Defaulting to version"
-					" %d.%d metadata\n",
-					st->ss->major,
-					st->minor_version);
+				did_default = 1;
 		} else {
 			if (!st->ss->validate_geometry(st, level, layout,
 						       raiddisks,
 						       chunk, size, dname,
-						       &freesize)) {
+						       &freesize,
+						       verbose > 0)) {
 
 				fprintf(stderr,
 					Name ": %s is not suitable for "
@@ -281,7 +322,6 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 				" %lluK < %lluK + metadata\n",
 				dname, freesize, size);
 			fail = 1;
-			close(fd);
 			continue;
 		}
 		if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
@@ -293,24 +333,36 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 			minsize = freesize;
 		}
 		if (runstop != 1 || verbose >= 0) {
+			int fd = open(dname, O_RDONLY);
+			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->ss->external ) {
 			/* size is meaningful */
-			if (minsize > 0x100000000ULL && st->ss->major == 0) {
+			if (!st->ss->validate_geometry(st, level, layout,
+						       raiddisks,
+						       chunk, minsize,
+						       NULL, NULL, 0)) {
 				fprintf(stderr, Name ": devices too large for RAID level %d\n", level);
 				return 1;
 			}
@@ -368,7 +420,7 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 		missing_disks++;
 	}
 
-	if (level <= 0 && first_missing != subdevs * 2) {
+	if (level <= 0 && first_missing < subdevs * 2) {
 		fprintf(stderr,
 			Name ": This level does not support missing devices\n");
 		return 1;
@@ -393,12 +445,16 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 	     ( level == 6 && (insert_point < raiddisks
 			      || second_missing < raiddisks))
 	     ||
+	     ( level <= 0 )
+	     ||
 	     assume_clean
-		)
+		) {
 		info.array.state = 1; /* clean, but one+ drive will be missing*/
-	else
+		info.resync_start = ~0ULL;
+	} else {
 		info.array.state = 0; /* not clean, but no errors */
-
+		info.resync_start = 0;
+	}
 	if (level == 10) {
 		/* for raid10, the bitmap size is the capacity of the array,
 		 * which is array.size * raid_disks / ncopies;
@@ -435,7 +491,6 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 		+ 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 */
@@ -464,6 +519,32 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 	if (!st->ss->init_super(st, &info.array, size, name, homehost, uuid))
 		return 1;
 
+	total_slots = info.array.nr_disks;
+	sysfs_init(&info, mdfd, 0);
+	st->ss->getinfo_super(st, &info);
+
+	if (did_default && verbose >= 0) {
+		if (is_subarray(info.text_version)) {
+			int dnum = devname2devnum(info.text_version+1);
+			char *path;
+			int mdp = get_mdp_major();
+			struct mdinfo *mdi;
+			if (dnum > 0)
+				path = map_dev(MD_MAJOR, dnum, 1);
+			else
+				path = map_dev(mdp, (-1-dnum)<< 6, 1);
+
+			mdi = sysfs_read(-1, dnum, GET_VERSION);
+
+			fprintf(stderr, Name ": Creating array inside "
+				"%s container %s\n", 
+				mdi?mdi->text_version:"managed", path);
+			sysfs_free(mdi);
+		} else
+			fprintf(stderr, Name ": Defaulting to version"
+				" %s metadata\n", info.text_version);
+	}
+
 	if (bitmap_file && vers < 9003) {
 		major_num = BITMAP_MAJOR_HOSTENDIAN;
 #ifdef __BIG_ENDIAN
@@ -487,17 +568,41 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 	}
 
 
+	sysfs_init(&info, mdfd, 0);
 
-	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 (st->ss->external && st->subarray[0]) {
+		/* 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)) {
+			if (verbose)
+				fprintf(stderr, Name ": reusing mdmon "
+					"for %s.\n",
+					devnum2devname(st->container_dev));
+			st->update_tail = &st->updates;
+		} else
+			need_mdmon = 1;
+	}
+	rv = set_array_info(mdfd, st, &info);
 	if (rv) {
-		fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
+		fprintf(stderr, Name ": failed to set array info for %s: %s\n",
 			mddev, strerror(errno));
 		return 1;
 	}
@@ -525,7 +630,7 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 		}
 	}
 
-
+	infos = malloc(sizeof(*infos) * total_slots);
 
 	for (pass=1; pass <=2 ; pass++) {
 		mddev_dev_t moved_disk = NULL; /* the disk that was moved out of the insert point */
@@ -534,61 +639,81 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 		     dv=(dv->next)?(dv->next):moved_disk, dnum++) {
 			int fd;
 			struct stat stb;
+			struct mdinfo *inf = &infos[dnum];
 
-			info.disk.number = dnum;
+			if (dnum >= total_slots)
+				abort();
 			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) |
+			if (dnum == insert_point ||
+			    strcasecmp(dv->devname, "missing")==0)
+				continue;
+
+			switch(pass) {
+			case 1:
+				*inf = info;
+
+				inf->disk.number = dnum;
+				inf->disk.raid_disk = dnum;
+				if (inf->disk.raid_disk < raiddisks)
+					inf->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);
+				else
+					inf->disk.state = 0;
+
+				if (dv->writemostly)
+					inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY);
+
+				if (st->ss->external && st->subarray[0])
+					fd = open(dv->devname, O_RDWR);
+				else
+					fd = open(dv->devname, O_RDWR|O_EXCL);
 
-			if (dnum == insert_point ||
-			    strcasecmp(dv->devname, "missing")==0) {
-				info.disk.major = 0;
-				info.disk.minor = 0;
-				info.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",
+					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);
+				inf->disk.major = major(stb.st_rdev);
+				inf->disk.minor = minor(stb.st_rdev);
+
 				remove_partitions(fd);
-				close(fd);
-			}
-			switch(pass){
-			case 1:
-				st->ss->add_to_super(st, &info.disk);
+				st->ss->add_to_super(st, &inf->disk,
+						     fd, dv->devname);
+				st->ss->getinfo_super(st, inf);
+				safe_mode_delay = inf->safe_mode_delay;
+
+				/* getinfo_super might have lost these ... */
+				inf->disk.major = major(stb.st_rdev);
+				inf->disk.minor = minor(stb.st_rdev);
 				break;
 			case 2:
-				if (info.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, &info.disk,
-							 dv->devname);
-
-				if (ioctl(mdfd, ADD_NEW_DISK, &info.disk)) {
-					fprintf(stderr, Name ": ADD_NEW_DISK for %s failed: %s\n",
+				inf->errors = 0;
+				rv = 0;
+
+				rv = add_disk(mdfd, st, &info, inf);
+
+				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 (pass == 1) {
+			st->ss->write_init_super(st);
+			flush_metadata_updates(st);
+		}
 	}
+	free(infos);
 	st->ss->free_super(st);
 
 	/* param is not actually used */
@@ -596,15 +721,39 @@ int Create(struct supertype *st, char *mddev, int mdfd,
 		/* No need to start */
 		;
 	else 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 (st->ss->external) {
+			switch(level) {
+			case LEVEL_LINEAR:
+			case LEVEL_MULTIPATH:
+			case 0:
+				sysfs_set_str(&info, NULL, "array_state",
+					      "active");
+				need_mdmon = 0;
+				break;
+			default:
+				sysfs_set_str(&info, NULL, "array_state",
+					      "readonly");
+				break;
+			}
+			sysfs_set_safemode(&info, safe_mode_delay);
+		} 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 && st->subarray[0]) {
+			if (need_mdmon)
+				start_mdmon(st->container_dev);
+
+			ping_monitor(devnum2devname(st->container_dev));
+			close(container_fd);
+		}
 	} else {
 		fprintf(stderr, Name ": not starting array - not enough devices.\n");
 	}