Remove getinfo_super_n and do some other cleaning up.
[thirdparty/mdadm.git] / Create.c
1 /*
2  * mdadm - manage Linux "md" devices aka RAID arrays.
3  *
4  * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de>
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
30 #include "mdadm.h"
31 #include        "md_u.h"
32 #include        "md_p.h"
33 #include        <ctype.h>
34
35 int Create(struct supertype *st, char *mddev, int mdfd,
36            int chunk, int level, int layout, unsigned long long size, int raiddisks, int sparedisks,
37            char *name, char *homehost, int *uuid,
38            int subdevs, mddev_dev_t devlist,
39            int runstop, int verbose, int force, int assume_clean,
40            char *bitmap_file, int bitmap_chunk, int write_behind, int delay)
41 {
42         /*
43          * Create a new raid array.
44          *
45          * First check that necessary details are available
46          * (i.e. level, raid-disks)
47          *
48          * Then check each disk to see what might be on it
49          * and report anything interesting.
50          *
51          * If anything looks odd, and runstop not set,
52          * abort.
53          *
54          * SET_ARRAY_INFO and ADD_NEW_DISK, and
55          * if runstop==run, or raiddisks disks were used,
56          * RUN_ARRAY
57          */
58         unsigned long long minsize=0, maxsize=0;
59         char *mindisc = NULL;
60         char *maxdisc = NULL;
61         int dnum;
62         mddev_dev_t dv;
63         int fail=0, warn=0;
64         struct stat stb;
65         int first_missing = subdevs * 2;
66         int second_missing = subdevs * 2;
67         int missing_disks = 0;
68         int insert_point = subdevs * 2; /* where to insert a missing drive */
69         int total_slots;
70         int pass;
71         int vers;
72         int rv;
73         int bitmap_fd;
74         int have_container = 0;
75         int container_fd;
76         int need_mdmon = 0;
77         unsigned long long bitmapsize;
78         struct mdinfo *sra;
79         struct mdinfo info, *infos;
80         int did_default = 0;
81
82         int major_num = BITMAP_MAJOR_HI;
83
84         memset(&info, 0, sizeof(info));
85
86         vers = md_get_version(mdfd);
87         if (vers < 9000) {
88                 fprintf(stderr, Name ": Create requires md driver version 0.90.0 or later\n");
89                 return 1;
90         } else {
91                 mdu_array_info_t inf;
92                 memset(&inf, 0, sizeof(inf));
93                 ioctl(mdfd, GET_ARRAY_INFO, &inf);
94                 if (inf.working_disks != 0) {
95                         fprintf(stderr, Name ": another array by this name"
96                                 " is already running.\n");
97                         return 1;
98                 }
99         }
100         if (level == UnSet) {
101                 /* "ddf" and "imsm" metadata only supports one level - should possibly
102                  * push this into metadata handler??
103                  */
104                 if (st && (st->ss == &super_ddf || st->ss == &super_imsm))
105                         level = LEVEL_CONTAINER;
106         }
107
108         if (level == UnSet) {
109                 fprintf(stderr,
110                         Name ": a RAID level is needed to create an array.\n");
111                 return 1;
112         }
113         if (raiddisks < 4 && level == 6) {
114                 fprintf(stderr,
115                         Name ": at least 4 raid-devices needed for level 6\n");
116                 return 1;
117         }
118         if (raiddisks > 256 && level == 6) {
119                 fprintf(stderr,
120                         Name ": no more than 256 raid-devices supported for level 6\n");
121                 return 1;
122         }
123         if (raiddisks < 2 && level >= 4) {
124                 fprintf(stderr,
125                         Name ": at least 2 raid-devices needed for level 4 or 5\n");
126                 return 1;
127         }
128         if (level <= 0 && sparedisks) {
129                 fprintf(stderr,
130                         Name ": This level does not support spare devices\n");
131                 return 1;
132         }
133
134         if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) {
135                 /* If given a single device, it might be a container, and we can
136                  * extract a device list from there
137                  */
138                 mdu_array_info_t inf;
139                 int fd;
140
141                 memset(&inf, 0, sizeof(inf));
142                 fd = open(devlist->devname, O_RDONLY, 0);
143                 if (fd >= 0 &&
144                     ioctl(fd, GET_ARRAY_INFO, &inf) == 0 &&
145                     inf.raid_disks == 0) {
146                         /* yep, looks like a container */
147                         if (st) {
148                                 rv = st->ss->load_super(st, fd,
149                                                         devlist->devname);
150                                 if (rv == 0)
151                                         have_container = 1;
152                         } else {
153                                 st = guess_super(fd);
154                                 if (st && !(rv = st->ss->
155                                             load_super(st, fd,
156                                                        devlist->devname)))
157                                         have_container = 1;
158                                 else
159                                         st = NULL;
160                         }
161                 }
162                 if (fd >= 0)
163                         close(fd);
164                 if (have_container) {
165                         subdevs = 0;
166                         devlist = NULL;
167                 }
168         }
169         if (subdevs > raiddisks+sparedisks) {
170                 fprintf(stderr, Name ": You have listed more devices (%d) than are in the array(%d)!\n", subdevs, raiddisks+sparedisks);
171                 return 1;
172         }
173         if (!have_container && subdevs < raiddisks+sparedisks) {
174                 fprintf(stderr, Name ": You haven't given enough devices (real or missing) to create this array\n");
175                 return 1;
176         }
177         if (bitmap_file && level <= 0) {
178                 fprintf(stderr, Name ": bitmaps not meaningful with level %s\n",
179                         map_num(pers, level)?:"given");
180                 return 1;
181         }
182
183         /* now set some defaults */
184         if (layout == UnSet)
185                 switch(level) {
186                 default: /* no layout */
187                         layout = 0;
188                         break;
189                 case 10:
190                         layout = 0x102; /* near=2, far=1 */
191                         if (verbose > 0)
192                                 fprintf(stderr,
193                                         Name ": layout defaults to n1\n");
194                         break;
195                 case 5:
196                 case 6:
197                         layout = map_name(r5layout, "default");
198                         if (verbose > 0)
199                                 fprintf(stderr,
200                                         Name ": layout defaults to %s\n", map_num(r5layout, layout));
201                         break;
202                 case LEVEL_FAULTY:
203                         layout = map_name(faultylayout, "default");
204
205                         if (verbose > 0)
206                                 fprintf(stderr,
207                                         Name ": layout defaults to %s\n", map_num(faultylayout, layout));
208                         break;
209                 }
210
211         if (level == 10)
212                 /* check layout fits in array*/
213                 if ((layout&255) * ((layout>>8)&255) > raiddisks) {
214                         fprintf(stderr, Name ": that layout requires at least %d devices\n",
215                                 (layout&255) * ((layout>>8)&255));
216                         return 1;
217                 }
218
219         switch(level) {
220         case 4:
221         case 5:
222         case 10:
223         case 6:
224         case 0:
225         case LEVEL_LINEAR: /* linear */
226                 if (chunk == 0) {
227                         chunk = 64;
228                         if (verbose > 0)
229                                 fprintf(stderr, Name ": chunk size defaults to 64K\n");
230                 }
231                 break;
232         case 1:
233         case LEVEL_FAULTY:
234         case LEVEL_MULTIPATH:
235         case LEVEL_CONTAINER:
236                 if (chunk) {
237                         chunk = 0;
238                         if (verbose > 0)
239                                 fprintf(stderr, Name ": chunk size ignored for this level\n");
240                 }
241                 break;
242         default:
243                 fprintf(stderr, Name ": unknown level %d\n", level);
244                 return 1;
245         }
246
247         if (st && ! st->ss->validate_geometry(st, level, layout, raiddisks,
248                                               chunk, size, NULL, NULL))
249                 return 1;
250
251         /* now look at the subdevs */
252         info.array.active_disks = 0;
253         info.array.working_disks = 0;
254         dnum = 0;
255         for (dv=devlist; dv; dv=dv->next, dnum++) {
256                 char *dname = dv->devname;
257                 unsigned long long freesize;
258                 if (strcasecmp(dname, "missing")==0) {
259                         if (first_missing > dnum)
260                                 first_missing = dnum;
261                         if (second_missing > dnum && dnum > first_missing)
262                                 second_missing = dnum;
263                         missing_disks ++;
264                         continue;
265                 }
266                 info.array.working_disks++;
267                 if (dnum < raiddisks)
268                         info.array.active_disks++;
269                 if (st == NULL) {
270                         struct createinfo *ci = conf_get_create_info();
271                         if (ci)
272                                 st = ci->supertype;
273                 }
274                 if (st == NULL) {
275                         /* Need to choose a default metadata, which is different
276                          * depending on geometry of array.
277                          */
278                         int i;
279                         char *name = "default";
280                         for(i=0; !st && superlist[i]; i++) {
281                                 st = superlist[i]->match_metadata_desc(name);
282                                 if (st && !st->ss->validate_geometry
283                                                 (st, level, layout, raiddisks,
284                                                  chunk, size, dname, &freesize))
285                                         st = NULL;
286                         }
287
288                         if (!st) {
289                                 fprintf(stderr, Name ": device %s not suitable "
290                                         "for any style of array\n",
291                                         dname);
292                                 exit(2);
293                         }
294                         if (st->ss != &super0 ||
295                             st->minor_version != 90)
296                                 did_default = 1;
297                 } else {
298                         if (!st->ss->validate_geometry(st, level, layout,
299                                                        raiddisks,
300                                                        chunk, size, dname,
301                                                        &freesize)) {
302
303                                 fprintf(stderr,
304                                         Name ": %s is not suitable for "
305                                         "this array.\n",
306                                         dname);
307                                 fail = 1;
308                                 continue;
309                         }
310                 }
311
312                 freesize /= 2; /* convert to K */
313                 if (chunk) {
314                         /* round to chunk size */
315                         freesize = freesize & ~(chunk-1);
316                 }
317
318                 if (size && freesize < size) {
319                         fprintf(stderr, Name ": %s is smaller that given size."
320                                 " %lluK < %lluK + metadata\n",
321                                 dname, freesize, size);
322                         fail = 1;
323                         continue;
324                 }
325                 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
326                         maxdisc = dname;
327                         maxsize = freesize;
328                 }
329                 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
330                         mindisc = dname;
331                         minsize = freesize;
332                 }
333                 if (runstop != 1 || verbose >= 0) {
334                         int fd = open(dname, O_RDONLY, 0);
335                         if (fd <0 ) {
336                                 fprintf(stderr, Name ": Cannot open %s: %s\n",
337                                         dname, strerror(errno));
338                                 fail=1;
339                                 continue;
340                         }
341                         warn |= check_ext2(fd, dname);
342                         warn |= check_reiser(fd, dname);
343                         warn |= check_raid(fd, dname);
344                         close(fd);
345                 }
346         }
347         if (fail) {
348                 fprintf(stderr, Name ": create aborted\n");
349                 return 1;
350         }
351         if (size == 0) {
352                 if (mindisc == NULL && !have_container) {
353                         fprintf(stderr, Name ": no size and no drives given - aborting create.\n");
354                         return 1;
355                 }
356                 if (level > 0 || level == LEVEL_MULTIPATH
357                     || level == LEVEL_FAULTY
358                     || st->ss->external ) {
359                         /* size is meaningful */
360                         if (!st->ss->validate_geometry(st, level, layout,
361                                                        raiddisks,
362                                                        chunk, minsize,
363                                                        NULL, NULL)) {
364                                 fprintf(stderr, Name ": devices too large for RAID level %d\n", level);
365                                 return 1;
366                         }
367                         size = minsize;
368                         if (verbose > 0)
369                                 fprintf(stderr, Name ": size set to %lluK\n", size);
370                 }
371         }
372         if (level > 0 && ((maxsize-size)*100 > maxsize)) {
373                 if (runstop != 1 || verbose >= 0)
374                         fprintf(stderr, Name ": largest drive (%s) exceed size (%lluK) by more than 1%%\n",
375                                 maxdisc, size);
376                 warn = 1;
377         }
378
379         if (warn) {
380                 if (runstop!= 1) {
381                         if (!ask("Continue creating array? ")) {
382                                 fprintf(stderr, Name ": create aborted.\n");
383                                 return 1;
384                         }
385                 } else {
386                         if (verbose > 0)
387                                 fprintf(stderr, Name ": creation continuing despite oddities due to --run\n");
388                 }
389         }
390
391         /* If this is raid4/5, we want to configure the last active slot
392          * as missing, so that a reconstruct happens (faster than re-parity)
393          * FIX: Can we do this for raid6 as well?
394          */
395         if (assume_clean==0 && force == 0 && first_missing >= raiddisks) {
396                 switch ( level ) {
397                 case 4:
398                 case 5:
399                         insert_point = raiddisks-1;
400                         sparedisks++;
401                         info.array.active_disks--;
402                         missing_disks++;
403                         break;
404                 default:
405                         break;
406                 }
407         }
408         /* For raid6, if creating with 1 missing drive, make a good drive
409          * into a spare, else the create will fail
410          */
411         if (assume_clean == 0 && force == 0 && first_missing < raiddisks &&
412             second_missing >= raiddisks && level == 6) {
413                 insert_point = raiddisks - 1;
414                 if (insert_point == first_missing)
415                         insert_point--;
416                 sparedisks ++;
417                 info.array.active_disks--;
418                 missing_disks++;
419         }
420
421         if (level <= 0 && first_missing < subdevs * 2) {
422                 fprintf(stderr,
423                         Name ": This level does not support missing devices\n");
424                 return 1;
425         }
426
427         /* Ok, lets try some ioctls */
428
429         info.array.level = level;
430         info.array.size = size;
431         info.array.raid_disks = raiddisks;
432         /* The kernel should *know* what md_minor we are dealing
433          * with, but it chooses to trust me instead. Sigh
434          */
435         info.array.md_minor = 0;
436         if (fstat(mdfd, &stb)==0)
437                 info.array.md_minor = minor(stb.st_rdev);
438         info.array.not_persistent = 0;
439
440         if ( ( (level == 4 || level == 5) &&
441                (insert_point < raiddisks || first_missing < raiddisks) )
442              ||
443              ( level == 6 && (insert_point < raiddisks
444                               || second_missing < raiddisks))
445              ||
446              ( level <= 0 )
447              ||
448              assume_clean
449                 )
450                 info.array.state = 1; /* clean, but one+ drive will be missing*/
451         else
452                 info.array.state = 0; /* not clean, but no errors */
453
454         if (level == 10) {
455                 /* for raid10, the bitmap size is the capacity of the array,
456                  * which is array.size * raid_disks / ncopies;
457                  * .. but convert to sectors.
458                  */
459                 int ncopies = ((layout>>8) & 255) * (layout & 255);
460                 bitmapsize = (unsigned long long)size * raiddisks / ncopies * 2;
461 /*              printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
462         } else
463                 bitmapsize = (unsigned long long)size * 2;
464
465         /* There is lots of redundancy in these disk counts,
466          * raid_disks is the most meaningful value
467          *          it describes the geometry of the array
468          *          it is constant
469          * nr_disks is total number of used slots.
470          *          it should be raid_disks+spare_disks
471          * spare_disks is the number of extra disks present
472          *          see above
473          * active_disks is the number of working disks in
474          *          active slots. (With raid_disks)
475          * working_disks is the total number of working disks,
476          *          including spares
477          * failed_disks is the number of disks marked failed
478          *
479          * Ideally, the kernel would keep these (except raid_disks)
480          * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
481          * So for now, we assume that all raid and spare
482          * devices will be given.
483          */
484         info.array.spare_disks=sparedisks;
485         info.array.failed_disks=missing_disks;
486         info.array.nr_disks = info.array.working_disks
487                 + info.array.failed_disks;
488         info.array.layout = layout;
489         info.array.chunk_size = chunk*1024;
490         info.array.major_version = 0; /* Flag to say "not zeroing superblock" */
491
492         if (name == NULL || *name == 0) {
493                 /* base name on mddev */
494                 /*  /dev/md0 -> 0
495                  *  /dev/md_d0 -> d0
496                  *  /dev/md/1 -> 1
497                  *  /dev/md/d1 -> d1
498                  *  /dev/md/home -> home
499                  *  /dev/mdhome -> home
500                  */
501                 name = strrchr(mddev, '/');
502                 if (name) {
503                         name++;
504                         if (strncmp(name, "md_d", 4)==0 &&
505                             strlen(name) > 4 &&
506                             isdigit(name[4]) &&
507                             (name-mddev) == 5 /* /dev/ */)
508                                 name += 3;
509                         else if (strncmp(name, "md", 2)==0 &&
510                                  strlen(name) > 2 &&
511                                  isdigit(name[2]) &&
512                                  (name-mddev) == 5 /* /dev/ */)
513                                 name += 2;
514                 }
515         }
516         if (!st->ss->init_super(st, &info.array, size, name, homehost, uuid))
517                 return 1;
518
519         total_slots = info.array.nr_disks;
520         st->ss->getinfo_super(st, &info);
521
522         if (did_default)
523                 fprintf(stderr, Name ": Defaulting to version"
524                         " %s metadata\n", info.text_version);
525
526         if (bitmap_file && vers < 9003) {
527                 major_num = BITMAP_MAJOR_HOSTENDIAN;
528 #ifdef __BIG_ENDIAN
529                 fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n"
530                         "  between different architectured.  Consider upgrading the Linux kernel.\n");
531 #endif
532         }
533
534         if (bitmap_file && strcmp(bitmap_file, "internal")==0) {
535                 if ((vers%100) < 2) {
536                         fprintf(stderr, Name ": internal bitmaps not supported by this kernel.\n");
537                         return 1;
538                 }
539                 if (!st->ss->add_internal_bitmap(st, &bitmap_chunk,
540                                                  delay, write_behind,
541                                                  bitmapsize, 1, major_num)) {
542                         fprintf(stderr, Name ": Given bitmap chunk size not supported.\n");
543                         return 1;
544                 }
545                 bitmap_file = NULL;
546         }
547
548
549         sra = sysfs_read(mdfd, 0, 0);
550
551         if (st->ss->external) {
552                 char ver[100];
553                 strcat(strcpy(ver, "external:"),
554                        info.text_version);
555                 if (st->ss->external && st->subarray[0]) {
556                         /* member */
557
558                         /* When creating a member, we need to be careful
559                          * to negotiate with mdmon properly.
560                          * If it is already running, we cannot write to
561                          * the devices and must ask it to do that part.
562                          * If it isn't running, we write to the devices,
563                          * and then start it.
564                          * We hold an exclusive open on the container
565                          * device to make sure mdmon doesn't exit after
566                          * we checked that it is running.
567                          *
568                          * For now, fail if it is already running.
569                          */
570                         container_fd = open_dev_excl(st->container_dev);
571                         if (container_fd < 0) {
572                                 fprintf(stderr, Name ": Cannot get exclusive "
573                                         "open on container - weird.\n");
574                                 return 1;
575                         }
576                         if (mdmon_running(st->container_dev)) {
577                                 fprintf(stderr, Name ": mdmon already running "
578                                         "for %s - sorry\n",
579                                         devnum2devname(st->container_dev));
580                                 return 1;
581                         }
582                         need_mdmon = 1;
583                 }
584                 if ((vers % 100) < 2 ||
585                     sra == NULL ||
586                     sysfs_set_str(sra, NULL, "metadata_version",
587                                   ver) < 0) {
588                         fprintf(stderr, Name ": This kernel does not "
589                                 "support external metadata.\n");
590                         return 1;
591                 }
592                 rv = sysfs_set_array(sra, &info);
593         } else  if ((vers % 100) >= 1) { /* can use different versions */
594                 mdu_array_info_t inf;
595                 memset(&inf, 0, sizeof(inf));
596                 inf.major_version = info.array.major_version;
597                 inf.minor_version = info.array.minor_version;
598                 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
599         } else
600                 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
601         if (rv) {
602                 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
603                         mddev, strerror(errno));
604                 return 1;
605         }
606
607         if (bitmap_file) {
608                 int uuid[4];
609
610                 st->ss->uuid_from_super(st, uuid);
611                 if (CreateBitmap(bitmap_file, force, (char*)uuid, bitmap_chunk,
612                                  delay, write_behind,
613                                  bitmapsize,
614                                  major_num)) {
615                         return 1;
616                 }
617                 bitmap_fd = open(bitmap_file, O_RDWR);
618                 if (bitmap_fd < 0) {
619                         fprintf(stderr, Name ": weird: %s cannot be openned\n",
620                                 bitmap_file);
621                         return 1;
622                 }
623                 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
624                         fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
625                                 mddev, strerror(errno));
626                         return 1;
627                 }
628         }
629
630         infos = malloc(sizeof(*infos) * total_slots);
631
632         for (pass=1; pass <=2 ; pass++) {
633                 mddev_dev_t moved_disk = NULL; /* the disk that was moved out of the insert point */
634
635                 for (dnum=0, dv = devlist ; dv ;
636                      dv=(dv->next)?(dv->next):moved_disk, dnum++) {
637                         int fd;
638                         struct stat stb;
639                         struct mdinfo *inf = &infos[dnum];
640
641                         if (dnum >= total_slots)
642                                 abort();
643                         if (dnum == insert_point) {
644                                 moved_disk = dv;
645                         }
646                         if (dnum == insert_point ||
647                             strcasecmp(dv->devname, "missing")==0)
648                                 continue;
649
650                         switch(pass) {
651                         case 1:
652                                 *inf = info;
653
654                                 inf->disk.number = dnum;
655                                 inf->disk.raid_disk = dnum;
656                                 if (inf->disk.raid_disk < raiddisks)
657                                         inf->disk.state = (1<<MD_DISK_ACTIVE) |
658                                                 (1<<MD_DISK_SYNC);
659                                 else
660                                         inf->disk.state = 0;
661
662                                 if (dv->writemostly)
663                                         inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY);
664
665                                 if (st->ss->external && st->subarray[0])
666                                         fd = open(dv->devname, O_RDWR, 0);
667                                 else
668                                         fd = open(dv->devname, O_RDWR|O_EXCL,0);
669
670                                 if (fd < 0) {
671                                         fprintf(stderr, Name ": failed to open %s "
672                                                 "after earlier success - aborting\n",
673                                                 dv->devname);
674                                         return 1;
675                                 }
676                                 fstat(fd, &stb);
677                                 inf->disk.major = major(stb.st_rdev);
678                                 inf->disk.minor = minor(stb.st_rdev);
679
680                                 remove_partitions(fd);
681                                 st->ss->add_to_super(st, &inf->disk,
682                                                      fd, dv->devname);
683                                 st->ss->getinfo_super(st, inf);
684
685                                 /* getinfo_super might have lost these ... */
686                                 inf->disk.major = major(stb.st_rdev);
687                                 inf->disk.minor = minor(stb.st_rdev);
688                                 break;
689                         case 2:
690                                 inf->errors = 0;
691                                 rv = 0;
692
693                                 if (st->ss->external)
694                                         rv = sysfs_add_disk(sra, inf);
695                                 else
696                                         rv = ioctl(mdfd, ADD_NEW_DISK,
697                                                    &inf->disk);
698
699                                 if (rv) {
700                                         fprintf(stderr,
701                                                 Name ": ADD_NEW_DISK for %s "
702                                                 "failed: %s\n",
703                                                 dv->devname, strerror(errno));
704                                         st->ss->free_super(st);
705                                         return 1;
706                                 }
707                                 break;
708                         }
709                         if (dv == moved_disk && dnum != insert_point) break;
710                 }
711                 if (pass == 1)
712                         st->ss->write_init_super(st);
713         }
714         free(infos);
715         st->ss->free_super(st);
716
717         /* param is not actually used */
718         if (level == LEVEL_CONTAINER)
719                 /* No need to start */
720                 ;
721         else if (runstop == 1 || subdevs >= raiddisks) {
722                 if (st->ss->external) {
723                         switch(level) {
724                         case LEVEL_LINEAR:
725                         case LEVEL_MULTIPATH:
726                         case 0:
727                                 sysfs_set_str(sra, NULL, "array_state",
728                                               "active");
729                                 need_mdmon = 0;
730                                 break;
731                         default:
732                                 sysfs_set_str(sra, NULL, "array_state",
733                                               "readonly");
734                                 break;
735                         }
736                 } else {
737                         mdu_param_t param;
738                         if (ioctl(mdfd, RUN_ARRAY, &param)) {
739                                 fprintf(stderr, Name ": RUN_ARRAY failed: %s\n",
740                                         strerror(errno));
741                                 Manage_runstop(mddev, mdfd, -1, 0);
742                                 return 1;
743                         }
744                 }
745                 if (verbose >= 0)
746                         fprintf(stderr, Name ": array %s started.\n", mddev);
747                 if (st->ss->external && st->subarray[0]) {
748                         if (need_mdmon) {
749                                 int dn = st->container_dev;
750                                 int i;
751                                 switch(fork()) {
752                                 case 0:
753                                         /* FIXME yuk. CLOSE_EXEC?? */
754                                         for (i=3; i < 100; i++)
755                                                 close(i);
756                                         execl("./mdmon", "mdmon",
757                                               map_dev(dev2major(dn),
758                                                       dev2minor(dn),
759                                                       1), NULL);
760                                         exit(1);
761                                 case -1: fprintf(stderr, Name ": cannot fork. "
762                                                  "Array remains readonly\n");
763                                         return 1;
764                                 default: ; /* parent - good */
765                                 }
766                         } else
767                                 signal_mdmon(st->container_dev);
768                         /* FIXME wait for mdmon to set array to read-auto */
769                         sleep(1);
770                         close(container_fd);
771                 }
772         } else {
773                 fprintf(stderr, Name ": not starting array - not enough devices.\n");
774         }
775         return 0;
776 }