[thirdparty/mdadm.git] / Assemble.c

/*
 * mdadm - manage Linux "md" devices aka RAID arrays.
 *
 * Copyright (C) 2001-2002 Neil Brown <neilb@cse.unsw.edu.au>
 *
 *
 *    This program is free software; you can redistribute it and/or modify
 *    it under the terms of the GNU General Public License as published by
 *    the Free Software Foundation; either version 2 of the License, or
 *    (at your option) any later version.
 *
 *    This program is distributed in the hope that it will be useful,
 *    but WITHOUT ANY WARRANTY; without even the implied warranty of
 *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *    GNU General Public License for more details.
 *
 *    You should have received a copy of the GNU General Public License
 *    along with this program; if not, write to the Free Software
 *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 *
 *    Author: Neil Brown
 *    Email: <neilb@cse.unsw.edu.au>
 *    Paper: Neil Brown
 *           School of Computer Science and Engineering
 *           The University of New South Wales
 *           Sydney, 2052
 *           Australia
 */

#include	"mdadm.h"
#include	"md_u.h"
#include	"md_p.h"

int Assemble(char *mddev, int mdfd,
	     mddev_ident_t ident, char *conffile,
	     mddev_dev_t devlist,
	     int readonly, int runstop,
	     int verbose, int force)
{
	/*
	 * The task of Assemble is to find a collection of
	 * devices that should (according to their superblocks)
	 * form an array, and to give this collection to the MD driver.
	 * In Linux-2.4 and later, this involves submitting a
	 * SET_ARRAY_INFO ioctl with no arg - to prepare
	 * the array - and then submit a number of
	 * ADD_NEW_DISK ioctls to add disks into
	 * the array.  Finally RUN_ARRAY might
	 * be submitted to start the array.
	 *
	 * Much of the work of Assemble is in finding and/or
	 * checking the disks to make sure they look right.
	 *
	 * If mddev is not set, then scan must be and we
	 *  read through the config file for dev+uuid mapping
	 *  We recurse, setting mddev, for each device that
	 *    - isn't running
	 *    - has a valid uuid (or any uuid if !uuidset
	 *
	 * If mddev is set, we try to determine state of md.
	 *   check version - must be at least 0.90.0
	 *   check kernel version.  must be at least 2.4.
	 *    If not, we can possibly fall back on START_ARRAY
	 *   Try to GET_ARRAY_INFO.
	 *     If possible, give up
	 *     If not, try to STOP_ARRAY just to make sure
	 *
	 * If !uuidset and scan, look in conf-file for uuid
	 *       If not found, give up
	 * If !devlist and scan and uuidset, get list of devs from conf-file 
	 *
	 * For each device:
	 *   Check superblock - discard if bad
	 *   Check uuid (set if we don't have one) - discard if no match
	 *   Check superblock similarity if we have a superbloc - discard if different
	 *   Record events, devicenum, utime
	 * This should give us a list of devices for the array
	 * We should collect the most recent event and utime numbers
	 *
	 * Count disks with recent enough event count
	 * While force && !enough disks
	 *    Choose newest rejected disks, update event count
	 *     mark clean and rewrite superblock
	 * If recent kernel:
	 *    SET_ARRAY_INFO
	 *    foreach device with recent events : ADD_NEW_DISK
	 *    if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
	 * If old kernel:
	 *    Check the device numbers in superblock are right
	 *    update superblock if any changes
	 *    START_ARRAY
	 *
	 */
	int old_linux = 0;
	int vers;
	mdu_array_info_t array;
	mdp_super_t first_super, super;
	struct {
		char *devname;
		int major, minor;
		int oldmajor, oldminor;
		long long events;
		time_t utime;
		int uptodate;
		int raid_disk;
	} devices[MD_SB_DISKS];
	int best[MD_SB_DISKS]; /* indexed by raid_disk */
	int devcnt = 0, okcnt, sparecnt;
	int i;
	int most_recent = 0;
	int chosen_drive;
	int change = 0;
	int inargv = 0;
	int start_partial_ok = force || devlist==NULL;
	
	vers = md_get_version(mdfd);
	if (vers <= 0) {
		fprintf(stderr, Name ": %s appears not to be an md device.\n");
		return 1;
	}
	if (vers < 9000) {
		fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
			"    Upgrade your kernel or try --build\n");
		return 1;
	}
	if (get_linux_version() < 2004000)
		old_linux = 1;

	if (ioctl(mdfd, GET_ARRAY_INFO, &array)>=0) {
		fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
			mddev);
		return 1;
	}
	ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */

	/*
	 * If any subdevs are listed, then any that don't
	 * match ident are discarded.  Remainder must all match and
	 * become the array.
	 * If no subdevs, then we scan all devices in the config file, but
	 * there must be something in the identity
	 */

	if (!devlist &&
	    ident->uuid_set == 0 &&
	    ident->super_minor < 0 &&
	    ident->devices == NULL) {
		fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
			mddev);
		return 1;
	}
	if (devlist == NULL)
		devlist = conf_get_devs(conffile);
	else inargv = 1;

	first_super.md_magic = 0;
	for (i=0; i<MD_SB_DISKS; i++)
		best[i] = -1;

	if (verbose)
	    fprintf(stderr, Name ": looking for devices for %s\n",
		    mddev);

	while ( devlist) {
		char *devname;
		int this_uuid[4];
		int dfd;
		struct stat stb;
		int havesuper=0;

		devname = devlist->devname;
		devlist = devlist->next;

		if (ident->devices &&
		    !match_oneof(ident->devices, devname))
			continue;
		
		dfd = open(devname, O_RDONLY, 0);
		if (dfd < 0) {
			if (inargv || verbose)
				fprintf(stderr, Name ": cannot open device %s: %s\n",
					devname, strerror(errno));
		} else if (fstat(dfd, &stb)< 0) {
			/* Impossible! */
			fprintf(stderr, Name ": fstat failed for %s: %s\n",
				devname, strerror(errno));
			close(dfd);
		} else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
			fprintf(stderr, Name ": %s is not a block device.\n",
				devname);
			close(dfd);
		} else if (load_super(dfd, &super)) {
			if (inargv || verbose)
				fprintf( stderr, Name ": no RAID superblock on %s\n",
					 devname);
			close(dfd);
		} else {
			havesuper =1;
			uuid_from_super(this_uuid, &super);
			close(dfd);
		}

		if (ident->uuid_set &&
		    (!havesuper || same_uuid(this_uuid, ident->uuid)==0)) {
			if (inargv || verbose)
				fprintf(stderr, Name ": %s has wrong uuid.\n",
					devname);
			continue;
		}
		if (ident->super_minor >= 0 &&
		    (!havesuper || ident->super_minor != super.md_minor)) {
			if (inargv || verbose)
				fprintf(stderr, Name ": %s has wrong super-minor.\n",
					devname);
			continue;
		}
		if (ident->level != -10 &&
		    (!havesuper|| ident->level != super.level)) {
			if (inargv || verbose)
				fprintf(stderr, Name ": %s has wrong raid level.\n",
					devname);
			continue;
		}
		if (ident->raid_disks != -1 &&
		    (!havesuper || ident->raid_disks!= super.raid_disks)) {
			if (inargv || verbose)
				fprintf(stderr, Name ": %s requires wrong number of drives.\n",
					devname);
			continue;
		}

		/* If we are this far, then we are commited to this device.
		 * If the super_block doesn't exist, or doesn't match others,
		 * then we cannot continue
		 */

		if (!havesuper) {
			fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
				devname);
			return 1;
		}
		if (compare_super(&first_super, &super)) {
			fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
				devname);
			return 1;
		}

		if (devcnt >= MD_SB_DISKS) {
		    fprintf(stderr, Name ": ouch - too many devices appear to be in this array. Ignoring %s\n",
			    devname);
		    continue;
		}
		if (verbose)
			fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
				devname, mddev, super.this_disk.raid_disk);
		devices[devcnt].devname = devname;
		devices[devcnt].major = MAJOR(stb.st_rdev);
		devices[devcnt].minor = MINOR(stb.st_rdev);
		devices[devcnt].oldmajor = super.this_disk.major;
		devices[devcnt].oldminor = super.this_disk.minor;
		devices[devcnt].events = md_event(&super);
		devices[devcnt].utime = super.utime;
		devices[devcnt].raid_disk = super.this_disk.raid_disk;
		devices[devcnt].uptodate = 0;
		if (most_recent < devcnt) {
			if (devices[devcnt].events
			    > devices[most_recent].events)
				most_recent = devcnt;
		}
		i = devices[devcnt].raid_disk;
		if (i>=0 && i < MD_SB_DISKS)
			if (best[i] == -1
			    || devices[best[i]].events < devices[devcnt].events)
				best[i] = devcnt;

		devcnt++;
	}

	if (devcnt == 0) {
		fprintf(stderr, Name ": no devices found for %s\n",
			mddev);
		return 1;
	}
	/* now we have some devices that might be suitable.
	 * I wonder how many
	 */
	okcnt = 0;
	sparecnt=0;
	for (i=0; i< MD_SB_DISKS;i++) {
		int j = best[i];
		int event_margin = !force;
		if (j < 0) continue;
		if (devices[j].events+event_margin >=
		    devices[most_recent].events) {
			devices[j].uptodate = 1;
			if (i < first_super.raid_disks)
				okcnt++;
			else
				sparecnt++;
		}
	}
	while (force && !enough(first_super.level, first_super.raid_disks, okcnt)) {
		/* Choose the newest best drive which is
		 * not up-to-date, update the superblock
		 * and add it.
		 */
		int fd;
		chosen_drive = -1;
		for (i=0; i<first_super.raid_disks; i++) {
			int j = best[i];
			if (j>=0 &&
			    !devices[j].uptodate &&
			    devices[j].events > 0 &&
			    (chosen_drive < 0 ||
			     devices[j].events > devices[chosen_drive].events))
				chosen_drive = j;
		}
		if (chosen_drive < 0)
			break;
		fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
			devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
			(int)(devices[chosen_drive].events),
			(int)(devices[most_recent].events));
		fd = open(devices[chosen_drive].devname, O_RDWR);
		if (fd < 0) {
			fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
				devices[chosen_drive].devname);
			devices[chosen_drive].events = 0;
			continue;
		}
		if (load_super(fd, &super)) {
			close(fd);
			fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
				devices[chosen_drive].devname);
			devices[chosen_drive].events = 0;
			continue;
		}
		super.events_hi = (devices[most_recent].events>>32)&0xFFFFFFFF;
		super.events_lo = (devices[most_recent].events)&0xFFFFFFFF;
		super.sb_csum = calc_sb_csum(&super);
/*DRYRUN*/	if (store_super(fd, &super)) {
			close(fd);
			fprintf(stderr, Name ": Could not re-write superblock on %s\n",
				devices[chosen_drive].devname);
			devices[chosen_drive].events = 0;
			continue;
		}
		close(fd);
		devices[chosen_drive].events = devices[most_recent].events;
		devices[chosen_drive].uptodate = 1;
		okcnt++;
	}

	/* Now we want to look at the superblock which the kernel will base things on
	 * and compare the devices that we think are working with the devices that the
	 * superblock thinks are working.
	 * If there are differences and --force is given, then update this chosen
	 * superblock.
	 */
	chosen_drive = -1;
	for (i=0; chosen_drive < 0 && i<MD_SB_DISKS; i++) {
		int j = best[i];
		int fd;
		if (j<0)
			continue;
		if (!devices[j].uptodate)
			continue;
		chosen_drive = j;
		if ((fd=open(devices[j].devname, O_RDONLY))< 0) {
			fprintf(stderr, Name ": Cannot open %s: %s\n",
				devices[j].devname, strerror(errno));
			return 1;
		}
		if (load_super(fd, &super)) {
			close(fd);
			fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
				devices[j].devname);
			return 1;
		}
		close(fd);
	}

	for (i=0; i<MD_SB_DISKS; i++) {
		int j = best[i];
		int desired_state;

		if (i < super.raid_disks)
			desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
		else
			desired_state = 0;

		if (j<0)
			continue;
		if (!devices[j].uptodate)
			continue;
#if 0
This doesnt work yet 
		if (devices[j].major != super.disks[i].major ||
		    devices[j].minor != super.disks[i].minor) {
			change |= 1;
			super.disks[i].major = devices[j].major;
			super.disks[i].minor = devices[j].minor;
		}
#endif 
		if (devices[j].oldmajor != super.disks[i].major ||
		    devices[j].oldminor != super.disks[i].minor) {
			change |= 2;
			super.disks[i].major = devices[i].oldmajor;
			super.disks[i].minor = devices[i].oldminor;
		}
		if (devices[j].uptodate &&
		    (super.disks[i].state != desired_state)) {
			if (force) {
				fprintf(stderr, Name ": "
					"clearing FAULTY flag for device %d in %s for %s\n",
					j, mddev, devices[j].devname);
				super.disks[i].state = desired_state;
				change |= 2;
			} else {
				fprintf(stderr, Name ": "
					"device %d in %s has wrong state in superblock, but %s seems ok\n",
					i, mddev, devices[j].devname);
			}
		}
		if (!devices[j].uptodate &&
		    !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
			fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
				i, mddev);
		}
	}

	if ((force && (change & 2))
	    || (old_linux && (change & 1))) {
		int fd;
		super.sb_csum = calc_sb_csum(&super);
		fd = open(devices[chosen_drive].devname, O_RDWR);
		if (fd < 0) {
			fprintf(stderr, Name ": Could open %s for write - cannot Assemble array.\n",
				devices[chosen_drive].devname);
			return 1;
		}
		if (store_super(fd, &super)) {
			close(fd);
			fprintf(stderr, Name ": Could not re-write superblock on %s\n",
				devices[chosen_drive].devname);
			return 1;
		}
		close(fd);
		change = 0;
	}

	/* Almost ready to actually *do* something */
	if (!old_linux) {
		if (ioctl(mdfd, SET_ARRAY_INFO, NULL) != 0) {
			fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
				mddev, strerror(errno));
			return 1;
		}
		/* First, add the raid disks, but add the chosen one last */
		for (i=0; i<=MD_SB_DISKS; i++) {
			int j;
			if (i < MD_SB_DISKS) {
				j = best[i];
				if (j == chosen_drive)
					continue;
			} else
				j = chosen_drive;

			if (j >= 0 && devices[j].uptodate) {
				mdu_disk_info_t disk;
				memset(&disk, 0, sizeof(disk));
				disk.major = devices[j].major;
				disk.minor = devices[j].minor;
				if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
					fprintf(stderr, Name ": failed to add %s to %s: %s\n",
						devices[j].devname,
						mddev,
						strerror(errno));
					if (i < first_super.raid_disks)
						okcnt--;
					else
						sparecnt--;
				} else if (verbose)
					fprintf(stderr, Name ": added %s to %s as %d\n",
						devices[j].devname, mddev, devices[j].raid_disk);
			} else if (verbose && i < first_super.raid_disks)
				fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
					i, mddev);
		}
		
		if (runstop == 1 ||
		    (runstop == 0 && 
		     ( first_super.raid_disks == okcnt
		       || start_partial_ok && enough(first_super.level, first_super.raid_disks, okcnt))
			    )) {
			if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
				fprintf(stderr, Name ": %s has been started with %d drive%s",
					mddev, okcnt, okcnt==1?"":"s");
				if (sparecnt)
					fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
				fprintf(stderr, ".\n");
				return 0;
			}
			fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
				mddev, strerror(errno));
			return 1;
		}
		if (runstop == -1) {
			fprintf(stderr, Name ": %s assembled from %d drive%s, but not started.\n",
				mddev, okcnt, okcnt==1?"":"s");
			return 0;
		}
		fprintf(stderr, Name ": %s assembled from %d drive%s - not enough to start it (use --run to insist).\n",
			mddev, okcnt, okcnt==1?"":"s");
		return 1;
	} else {
		/* The "chosen_drive" is a good choice, and if necessary, the superblock has
		 * been updated to point to the current locations of devices.
		 * so we can just start the array
		 */
		unsigned long dev;
		dev = MKDEV(devices[chosen_drive].major,
			    devices[chosen_drive].minor);
		if (ioctl(mdfd, START_ARRAY, dev)) {
		    fprintf(stderr, Name ": Cannot start array: %s\n",
			    strerror(errno));
		}
		
	}
}
Commit	Line	Data
64c4757e	1	/*
9a9dab36	2	* mdadm - manage Linux "md" devices aka RAID arrays.
64c4757e	3	*
cd29a5c8	4	* Copyright (C) 2001-2002 Neil Brown <neilb@cse.unsw.edu.au>
64c4757e NB	5	*
	6	*
	7	* This program is free software; you can redistribute it and/or modify
	8	* it under the terms of the GNU General Public License as published by
	9	* the Free Software Foundation; either version 2 of the License, or
	10	* (at your option) any later version.
	11	*
	12	* This program is distributed in the hope that it will be useful,
	13	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	14	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
	15	* GNU General Public License for more details.
	16	*
	17	* You should have received a copy of the GNU General Public License
	18	* along with this program; if not, write to the Free Software
	19	* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
	20	*
	21	* Author: Neil Brown
	22	* Email: <neilb@cse.unsw.edu.au>
	23	* Paper: Neil Brown
	24	* School of Computer Science and Engineering
	25	* The University of New South Wales
	26	* Sydney, 2052
	27	* Australia
	28	*/
	29
9a9dab36	30	#include "mdadm.h"
64c4757e	31	#include "md_u.h"
52826846	32	#include "md_p.h"
64c4757e NB	33
64c4757e NB	34	int Assemble(char *mddev, int mdfd,
52826846	35	mddev_ident_t ident, char *conffile,
cd29a5c8	36	mddev_dev_t devlist,
64c4757e NB	37	int readonly, int runstop,
	38	int verbose, int force)
	39	{
	40	/*
52826846 NB	41	* The task of Assemble is to find a collection of
	42	* devices that should (according to their superblocks)
	43	* form an array, and to give this collection to the MD driver.
	44	* In Linux-2.4 and later, this involves submitting a
64c4757e NB	45	* SET_ARRAY_INFO ioctl with no arg - to prepare
	46	* the array - and then submit a number of
	47	* ADD_NEW_DISK ioctls to add disks into
	48	* the array. Finally RUN_ARRAY might
	49	* be submitted to start the array.
	50	*
	51	* Much of the work of Assemble is in finding and/or
	52	* checking the disks to make sure they look right.
	53	*
	54	* If mddev is not set, then scan must be and we
	55	* read through the config file for dev+uuid mapping
	56	* We recurse, setting mddev, for each device that
	57	* - isn't running
	58	* - has a valid uuid (or any uuid if !uuidset
	59	*
	60	* If mddev is set, we try to determine state of md.
	61	* check version - must be at least 0.90.0
	62	* check kernel version. must be at least 2.4.
	63	* If not, we can possibly fall back on START_ARRAY
	64	* Try to GET_ARRAY_INFO.
	65	* If possible, give up
	66	* If not, try to STOP_ARRAY just to make sure
	67	*
	68	* If !uuidset and scan, look in conf-file for uuid
	69	* If not found, give up
cd29a5c8	70	* If !devlist and scan and uuidset, get list of devs from conf-file
64c4757e NB	71	*
	72	* For each device:
	73	* Check superblock - discard if bad
	74	* Check uuid (set if we don't have one) - discard if no match
	75	* Check superblock similarity if we have a superbloc - discard if different
	76	* Record events, devicenum, utime
	77	* This should give us a list of devices for the array
	78	* We should collect the most recent event and utime numbers
	79	*
	80	* Count disks with recent enough event count
	81	* While force && !enough disks
	82	* Choose newest rejected disks, update event count
	83	* mark clean and rewrite superblock
	84	* If recent kernel:
	85	* SET_ARRAY_INFO
	86	* foreach device with recent events : ADD_NEW_DISK
	87	* if runstop == 1 \|\| "enough" disks and runstop==0 -> RUN_ARRAY
	88	* If old kernel:
	89	* Check the device numbers in superblock are right
	90	* update superblock if any changes
	91	* START_ARRAY
	92	*
	93	*/
	94	int old_linux = 0;
	95	int vers;
	96	mdu_array_info_t array;
64c4757e NB	97	mdp_super_t first_super, super;
	98	struct {
	99	char *devname;
	100	int major, minor;
11a3e71d	101	int oldmajor, oldminor;
64c4757e NB	102	long long events;
	103	time_t utime;
	104	int uptodate;
52826846	105	int raid_disk;
64c4757e NB	106	} devices[MD_SB_DISKS];
64c4757e NB	107	int best[MD_SB_DISKS]; /* indexed by raid_disk */
52826846	108	int devcnt = 0, okcnt, sparecnt;
64c4757e NB	109	int i;
64c4757e NB	110	int most_recent = 0;
cd29a5c8	111	int chosen_drive;
52826846	112	int change = 0;
cd29a5c8 NB	113	int inargv = 0;
cd29a5c8 NB	114	int start_partial_ok = force \|\| devlist==NULL;
64c4757e	115
64c4757e NB	116	vers = md_get_version(mdfd);
64c4757e NB	117	if (vers <= 0) {
682c7051	118	fprintf(stderr, Name ": %s appears not to be an md device.\n");
64c4757e NB	119	return 1;
64c4757e NB	120	}
682c7051 NB	121	if (vers < 9000) {
682c7051 NB	122	fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
52826846	123	" Upgrade your kernel or try --build\n");
64c4757e NB	124	return 1;
64c4757e NB	125	}
682c7051	126	if (get_linux_version() < 2004000)
64c4757e NB	127	old_linux = 1;
	128
	129	if (ioctl(mdfd, GET_ARRAY_INFO, &array)>=0) {
682c7051	130	fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
64c4757e NB	131	mddev);
	132	return 1;
	133	}
	134	ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
	135
	136	/*
52826846 NB	137	* If any subdevs are listed, then any that don't
	138	* match ident are discarded. Remainder must all match and
	139	* become the array.
	140	* If no subdevs, then we scan all devices in the config file, but
	141	* there must be something in the identity
64c4757e	142	*/
64c4757e	143
cd29a5c8	144	if (!devlist &&
52826846 NB	145	ident->uuid_set == 0 &&
	146	ident->super_minor < 0 &&
	147	ident->devices == NULL) {
	148	fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
	149	mddev);
	150	return 1;
64c4757e	151	}
cd29a5c8	152	if (devlist == NULL)
64c4757e	153	devlist = conf_get_devs(conffile);
cd29a5c8	154	else inargv = 1;
64c4757e	155
64c4757e NB	156	first_super.md_magic = 0;
	157	for (i=0; i<MD_SB_DISKS; i++)
	158	best[i] = -1;
	159
82b27616 NB	160	if (verbose)
	161	fprintf(stderr, Name ": looking for devices for %s\n",
	162	mddev);
	163
cd29a5c8	164	while ( devlist) {
64c4757e NB	165	char *devname;
	166	int this_uuid[4];
	167	int dfd;
	168	struct stat stb;
52826846 NB	169	int havesuper=0;
52826846 NB	170
cd29a5c8 NB	171	devname = devlist->devname;
cd29a5c8 NB	172	devlist = devlist->next;
64c4757e	173
52826846 NB	174	if (ident->devices &&
	175	!match_oneof(ident->devices, devname))
	176	continue;
	177
64c4757e NB	178	dfd = open(devname, O_RDONLY, 0);
	179	if (dfd < 0) {
	180	if (inargv \|\| verbose)
682c7051	181	fprintf(stderr, Name ": cannot open device %s: %s\n",
64c4757e	182	devname, strerror(errno));
52826846 NB	183	} else if (fstat(dfd, &stb)< 0) {
	184	/* Impossible! */
	185	fprintf(stderr, Name ": fstat failed for %s: %s\n",
	186	devname, strerror(errno));
	187	close(dfd);
cd29a5c8 NB	188	} else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
cd29a5c8 NB	189	fprintf(stderr, Name ": %s is not a block device.\n",
52826846 NB	190	devname);
52826846 NB	191	close(dfd);
cd29a5c8	192	} else if (load_super(dfd, &super)) {
64c4757e	193	if (inargv \|\| verbose)
682c7051	194	fprintf( stderr, Name ": no RAID superblock on %s\n",
64c4757e NB	195	devname);
64c4757e NB	196	close(dfd);
52826846 NB	197	} else {
	198	havesuper =1;
	199	uuid_from_super(this_uuid, &super);
	200	close(dfd);
64c4757e	201	}
52826846 NB	202
	203	if (ident->uuid_set &&
	204	(!havesuper \|\| same_uuid(this_uuid, ident->uuid)==0)) {
	205	if (inargv \|\| verbose)
	206	fprintf(stderr, Name ": %s has wrong uuid.\n",
	207	devname);
	208	continue;
82b27616	209	}
52826846 NB	210	if (ident->super_minor >= 0 &&
52826846 NB	211	(!havesuper \|\| ident->super_minor != super.md_minor)) {
64c4757e	212	if (inargv \|\| verbose)
52826846	213	fprintf(stderr, Name ": %s has wrong super-minor.\n",
64c4757e NB	214	devname);
	215	continue;
	216	}
cd29a5c8 NB	217	if (ident->level != -10 &&
	218	(!havesuper\|\| ident->level != super.level)) {
	219	if (inargv \|\| verbose)
	220	fprintf(stderr, Name ": %s has wrong raid level.\n",
	221	devname);
	222	continue;
	223	}
	224	if (ident->raid_disks != -1 &&
	225	(!havesuper \|\| ident->raid_disks!= super.raid_disks)) {
	226	if (inargv \|\| verbose)
	227	fprintf(stderr, Name ": %s requires wrong number of drives.\n",
	228	devname);
	229	continue;
	230	}
52826846 NB	231
	232	/* If we are this far, then we are commited to this device.
	233	* If the super_block doesn't exist, or doesn't match others,
	234	* then we cannot continue
	235	*/
52826846 NB	236
	237	if (!havesuper) {
	238	fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
	239	devname);
	240	return 1;
	241	}
	242	if (compare_super(&first_super, &super)) {
	243	fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
	244	devname);
	245	return 1;
64c4757e NB	246	}
64c4757e NB	247
64c4757e	248	if (devcnt >= MD_SB_DISKS) {
682c7051	249	fprintf(stderr, Name ": ouch - too many devices appear to be in this array. Ignoring %s\n",
64c4757e NB	250	devname);
	251	continue;
	252	}
cd29a5c8 NB	253	if (verbose)
	254	fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
	255	devname, mddev, super.this_disk.raid_disk);
64c4757e NB	256	devices[devcnt].devname = devname;
	257	devices[devcnt].major = MAJOR(stb.st_rdev);
	258	devices[devcnt].minor = MINOR(stb.st_rdev);
11a3e71d NB	259	devices[devcnt].oldmajor = super.this_disk.major;
11a3e71d NB	260	devices[devcnt].oldminor = super.this_disk.minor;
64c4757e NB	261	devices[devcnt].events = md_event(&super);
64c4757e NB	262	devices[devcnt].utime = super.utime;
52826846	263	devices[devcnt].raid_disk = super.this_disk.raid_disk;
64c4757e NB	264	devices[devcnt].uptodate = 0;
	265	if (most_recent < devcnt) {
	266	if (devices[devcnt].events
	267	> devices[most_recent].events)
	268	most_recent = devcnt;
	269	}
52826846 NB	270	i = devices[devcnt].raid_disk;
	271	if (i>=0 && i < MD_SB_DISKS)
	272	if (best[i] == -1
	273	\|\| devices[best[i]].events < devices[devcnt].events)
	274	best[i] = devcnt;
	275
64c4757e NB	276	devcnt++;
	277	}
	278
	279	if (devcnt == 0) {
682c7051	280	fprintf(stderr, Name ": no devices found for %s\n",
64c4757e NB	281	mddev);
	282	return 1;
	283	}
	284	/* now we have some devices that might be suitable.
	285	* I wonder how many
	286	*/
	287	okcnt = 0;
52826846 NB	288	sparecnt=0;
52826846 NB	289	for (i=0; i< MD_SB_DISKS;i++) {
64c4757e	290	int j = best[i];
cd29a5c8	291	int event_margin = !force;
64c4757e	292	if (j < 0) continue;
cd29a5c8	293	if (devices[j].events+event_margin >=
64c4757e NB	294	devices[most_recent].events) {
64c4757e NB	295	devices[j].uptodate = 1;
52826846 NB	296	if (i < first_super.raid_disks)
	297	okcnt++;
	298	else
	299	sparecnt++;
64c4757e NB	300	}
	301	}
	302	while (force && !enough(first_super.level, first_super.raid_disks, okcnt)) {
	303	/* Choose the newest best drive which is
	304	* not up-to-date, update the superblock
	305	* and add it.
	306	*/
52826846	307	int fd;
cd29a5c8	308	chosen_drive = -1;
52826846 NB	309	for (i=0; i<first_super.raid_disks; i++) {
	310	int j = best[i];
	311	if (j>=0 &&
	312	!devices[j].uptodate &&
	313	devices[j].events > 0 &&
	314	(chosen_drive < 0 \|\|
	315	devices[j].events > devices[chosen_drive].events))
	316	chosen_drive = j;
	317	}
	318	if (chosen_drive < 0)
	319	break;
	320	fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
	321	devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
	322	(int)(devices[chosen_drive].events),
	323	(int)(devices[most_recent].events));
	324	fd = open(devices[chosen_drive].devname, O_RDWR);
	325	if (fd < 0) {
	326	fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
	327	devices[chosen_drive].devname);
	328	devices[chosen_drive].events = 0;
	329	continue;
	330	}
	331	if (load_super(fd, &super)) {
	332	close(fd);
	333	fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
	334	devices[chosen_drive].devname);
	335	devices[chosen_drive].events = 0;
	336	continue;
	337	}
	338	super.events_hi = (devices[most_recent].events>>32)&0xFFFFFFFF;
	339	super.events_lo = (devices[most_recent].events)&0xFFFFFFFF;
	340	super.sb_csum = calc_sb_csum(&super);
	341	/DRYRUN/ if (store_super(fd, &super)) {
	342	close(fd);
	343	fprintf(stderr, Name ": Could not re-write superblock on %s\n",
	344	devices[chosen_drive].devname);
	345	devices[chosen_drive].events = 0;
	346	continue;
	347	}
	348	close(fd);
	349	devices[chosen_drive].events = devices[most_recent].events;
	350	devices[chosen_drive].uptodate = 1;
	351	okcnt++;
64c4757e	352	}
52826846 NB	353
	354	/* Now we want to look at the superblock which the kernel will base things on
	355	* and compare the devices that we think are working with the devices that the
	356	* superblock thinks are working.
	357	* If there are differences and --force is given, then update this chosen
	358	* superblock.
	359	*/
cd29a5c8	360	chosen_drive = -1;
52826846 NB	361	for (i=0; chosen_drive < 0 && i<MD_SB_DISKS; i++) {
	362	int j = best[i];
	363	int fd;
	364	if (j<0)
	365	continue;
	366	if (!devices[j].uptodate)
	367	continue;
	368	chosen_drive = j;
	369	if ((fd=open(devices[j].devname, O_RDONLY))< 0) {
	370	fprintf(stderr, Name ": Cannot open %s: %s\n",
	371	devices[j].devname, strerror(errno));
	372	return 1;
	373	}
	374	if (load_super(fd, &super)) {
	375	close(fd);
	376	fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
	377	devices[j].devname);
	378	return 1;
	379	}
	380	close(fd);
	381	}
	382
	383	for (i=0; i<MD_SB_DISKS; i++) {
	384	int j = best[i];
11a3e71d NB	385	int desired_state;
	386
	387	if (i < super.raid_disks)
	388	desired_state = (1<<MD_DISK_ACTIVE) \| (1<<MD_DISK_SYNC);
	389	else
	390	desired_state = 0;
	391
52826846 NB	392	if (j<0)
	393	continue;
	394	if (!devices[j].uptodate)
	395	continue;
11a3e71d NB	396	#if 0
	397	This doesnt work yet
	398	if (devices[j].major != super.disks[i].major \|\|
	399	devices[j].minor != super.disks[i].minor) {
52826846	400	change \|= 1;
11a3e71d NB	401	super.disks[i].major = devices[j].major;
	402	super.disks[i].minor = devices[j].minor;
	403	}
	404	#endif
	405	if (devices[j].oldmajor != super.disks[i].major \|\|
	406	devices[j].oldminor != super.disks[i].minor) {
	407	change \|= 2;
	408	super.disks[i].major = devices[i].oldmajor;
	409	super.disks[i].minor = devices[i].oldminor;
52826846 NB	410	}
52826846 NB	411	if (devices[j].uptodate &&
11a3e71d	412	(super.disks[i].state != desired_state)) {
52826846 NB	413	if (force) {
52826846 NB	414	fprintf(stderr, Name ": "
cd29a5c8	415	"clearing FAULTY flag for device %d in %s for %s\n",
52826846	416	j, mddev, devices[j].devname);
11a3e71d	417	super.disks[i].state = desired_state;
52826846 NB	418	change \|= 2;
	419	} else {
	420	fprintf(stderr, Name ": "
11a3e71d	421	"device %d in %s has wrong state in superblock, but %s seems ok\n",
52826846 NB	422	i, mddev, devices[j].devname);
	423	}
	424	}
	425	if (!devices[j].uptodate &&
	426	!(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
	427	fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
	428	i, mddev);
	429	}
	430	}
	431
	432	if ((force && (change & 2))
	433	\|\| (old_linux && (change & 1))) {
	434	int fd;
	435	super.sb_csum = calc_sb_csum(&super);
	436	fd = open(devices[chosen_drive].devname, O_RDWR);
	437	if (fd < 0) {
	438	fprintf(stderr, Name ": Could open %s for write - cannot Assemble array.\n",
	439	devices[chosen_drive].devname);
	440	return 1;
	441	}
	442	if (store_super(fd, &super)) {
	443	close(fd);
	444	fprintf(stderr, Name ": Could not re-write superblock on %s\n",
	445	devices[chosen_drive].devname);
	446	return 1;
	447	}
	448	close(fd);
	449	change = 0;
	450	}
	451
64c4757e NB	452	/* Almost ready to actually do something */
	453	if (!old_linux) {
	454	if (ioctl(mdfd, SET_ARRAY_INFO, NULL) != 0) {
682c7051	455	fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
64c4757e NB	456	mddev, strerror(errno));
	457	return 1;
	458	}
52826846 NB	459	/* First, add the raid disks, but add the chosen one last */
	460	for (i=0; i<=MD_SB_DISKS; i++) {
	461	int j;
	462	if (i < MD_SB_DISKS) {
	463	j = best[i];
	464	if (j == chosen_drive)
	465	continue;
	466	} else
	467	j = chosen_drive;
	468
0db17fcb	469	if (j >= 0 && devices[j].uptodate) {
64c4757e NB	470	mdu_disk_info_t disk;
	471	memset(&disk, 0, sizeof(disk));
	472	disk.major = devices[j].major;
	473	disk.minor = devices[j].minor;
	474	if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
682c7051	475	fprintf(stderr, Name ": failed to add %s to %s: %s\n",
64c4757e NB	476	devices[j].devname,
	477	mddev,
	478	strerror(errno));
52826846 NB	479	if (i < first_super.raid_disks)
	480	okcnt--;
	481	else
	482	sparecnt--;
	483	} else if (verbose)
	484	fprintf(stderr, Name ": added %s to %s as %d\n",
	485	devices[j].devname, mddev, devices[j].raid_disk);
	486	} else if (verbose && i < first_super.raid_disks)
682c7051	487	fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
64c4757e NB	488	i, mddev);
64c4757e NB	489	}
52826846	490
64c4757e NB	491	if (runstop == 1 \|\|
64c4757e NB	492	(runstop == 0 &&
cd29a5c8 NB	493	( first_super.raid_disks == okcnt
	494	\|\| start_partial_ok && enough(first_super.level, first_super.raid_disks, okcnt))
	495	)) {
82b27616	496	if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
52826846 NB	497	fprintf(stderr, Name ": %s has been started with %d drive%s",
	498	mddev, okcnt, okcnt==1?"":"s");
	499	if (sparecnt)
	500	fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
	501	fprintf(stderr, ".\n");
64c4757e	502	return 0;
82b27616	503	}
682c7051	504	fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
64c4757e NB	505	mddev, strerror(errno));
	506	return 1;
	507	}
82b27616	508	if (runstop == -1) {
52826846 NB	509	fprintf(stderr, Name ": %s assembled from %d drive%s, but not started.\n",
52826846 NB	510	mddev, okcnt, okcnt==1?"":"s");
64c4757e	511	return 0;
82b27616	512	}
cd29a5c8	513	fprintf(stderr, Name ": %s assembled from %d drive%s - not enough to start it (use --run to insist).\n",
52826846	514	mddev, okcnt, okcnt==1?"":"s");
64c4757e	515	return 1;
82b27616	516	} else {
52826846 NB	517	/* The "chosen_drive" is a good choice, and if necessary, the superblock has
	518	* been updated to point to the current locations of devices.
	519	* so we can just start the array
82b27616	520	*/
cd29a5c8	521	unsigned long dev;
82b27616 NB	522	dev = MKDEV(devices[chosen_drive].major,
	523	devices[chosen_drive].minor);
	524	if (ioctl(mdfd, START_ARRAY, dev)) {
	525	fprintf(stderr, Name ": Cannot start array: %s\n",
	526	strerror(errno));
	527	}
	528
64c4757e NB	529	}
64c4757e NB	530	}