Remove st->text_version in favour of info->text_version
[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 pass;
70         int vers;
71         int rv;
72         int bitmap_fd;
73         int have_container = 0;
74         int container_fd;
75         int need_mdmon = 0;
76         unsigned long long bitmapsize;
77         struct mdinfo *sra;
78         struct mdinfo info;
79
80         int major_num = BITMAP_MAJOR_HI;
81
82         memset(&info, 0, sizeof(info));
83
84         vers = md_get_version(mdfd);
85         if (vers < 9000) {
86                 fprintf(stderr, Name ": Create requires md driver version 0.90.0 or later\n");
87                 return 1;
88         } else {
89                 mdu_array_info_t inf;
90                 memset(&inf, 0, sizeof(inf));
91                 ioctl(mdfd, GET_ARRAY_INFO, &inf);
92                 if (inf.working_disks != 0) {
93                         fprintf(stderr, Name ": another array by this name"
94                                 " is already running.\n");
95                         return 1;
96                 }
97         }
98         if (level == UnSet) {
99                 /* "ddf" and "imsm" metadata only supports one level - should possibly
100                  * push this into metadata handler??
101                  */
102                 if (st && (st->ss == &super_ddf || st->ss == &super_imsm))
103                         level = LEVEL_CONTAINER;
104         }
105
106         if (level == UnSet) {
107                 fprintf(stderr,
108                         Name ": a RAID level is needed to create an array.\n");
109                 return 1;
110         }
111         if (raiddisks < 4 && level == 6) {
112                 fprintf(stderr,
113                         Name ": at least 4 raid-devices needed for level 6\n");
114                 return 1;
115         }
116         if (raiddisks > 256 && level == 6) {
117                 fprintf(stderr,
118                         Name ": no more than 256 raid-devices supported for level 6\n");
119                 return 1;
120         }
121         if (raiddisks < 2 && level >= 4) {
122                 fprintf(stderr,
123                         Name ": at least 2 raid-devices needed for level 4 or 5\n");
124                 return 1;
125         }
126         if (level <= 0 && sparedisks) {
127                 fprintf(stderr,
128                         Name ": This level does not support spare devices\n");
129                 return 1;
130         }
131
132         if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) {
133                 /* If given a single device, it might be a container, and we can
134                  * extract a device list from there
135                  */
136                 mdu_array_info_t inf;
137                 int fd;
138
139                 memset(&inf, 0, sizeof(inf));
140                 fd = open(devlist->devname, O_RDONLY, 0);
141                 if (fd >= 0 &&
142                     ioctl(fd, GET_ARRAY_INFO, &inf) == 0 &&
143                     inf.raid_disks == 0) {
144                         /* yep, looks like a container */
145                         if (st) {
146                                 rv = st->ss->load_super(st, fd,
147                                                         devlist->devname);
148                                 if (rv == 0)
149                                         have_container = 1;
150                         } else {
151                                 st = guess_super(fd);
152                                 if (st && !(rv = st->ss->
153                                             load_super(st, fd,
154                                                        devlist->devname)))
155                                         have_container = 1;
156                                 else
157                                         st = NULL;
158                         }
159                 }
160                 if (fd >= 0)
161                         close(fd);
162                 if (have_container) {
163                         subdevs = 0;
164                         devlist = NULL;
165                 }
166         }
167         if (subdevs > raiddisks+sparedisks) {
168                 fprintf(stderr, Name ": You have listed more devices (%d) than are in the array(%d)!\n", subdevs, raiddisks+sparedisks);
169                 return 1;
170         }
171         if (!have_container && subdevs < raiddisks+sparedisks) {
172                 fprintf(stderr, Name ": You haven't given enough devices (real or missing) to create this array\n");
173                 return 1;
174         }
175         if (bitmap_file && level <= 0) {
176                 fprintf(stderr, Name ": bitmaps not meaningful with level %s\n",
177                         map_num(pers, level)?:"given");
178                 return 1;
179         }
180
181         /* now set some defaults */
182         if (layout == UnSet)
183                 switch(level) {
184                 default: /* no layout */
185                         layout = 0;
186                         break;
187                 case 10:
188                         layout = 0x102; /* near=2, far=1 */
189                         if (verbose > 0)
190                                 fprintf(stderr,
191                                         Name ": layout defaults to n1\n");
192                         break;
193                 case 5:
194                 case 6:
195                         layout = map_name(r5layout, "default");
196                         if (verbose > 0)
197                                 fprintf(stderr,
198                                         Name ": layout defaults to %s\n", map_num(r5layout, layout));
199                         break;
200                 case LEVEL_FAULTY:
201                         layout = map_name(faultylayout, "default");
202
203                         if (verbose > 0)
204                                 fprintf(stderr,
205                                         Name ": layout defaults to %s\n", map_num(faultylayout, layout));
206                         break;
207                 }
208
209         if (level == 10)
210                 /* check layout fits in array*/
211                 if ((layout&255) * ((layout>>8)&255) > raiddisks) {
212                         fprintf(stderr, Name ": that layout requires at least %d devices\n",
213                                 (layout&255) * ((layout>>8)&255));
214                         return 1;
215                 }
216
217         switch(level) {
218         case 4:
219         case 5:
220         case 10:
221         case 6:
222         case 0:
223         case LEVEL_LINEAR: /* linear */
224                 if (chunk == 0) {
225                         chunk = 64;
226                         if (verbose > 0)
227                                 fprintf(stderr, Name ": chunk size defaults to 64K\n");
228                 }
229                 break;
230         case 1:
231         case LEVEL_FAULTY:
232         case LEVEL_MULTIPATH:
233         case LEVEL_CONTAINER:
234                 if (chunk) {
235                         chunk = 0;
236                         if (verbose > 0)
237                                 fprintf(stderr, Name ": chunk size ignored for this level\n");
238                 }
239                 break;
240         default:
241                 fprintf(stderr, Name ": unknown level %d\n", level);
242                 return 1;
243         }
244
245         if (st && ! st->ss->validate_geometry(st, level, layout, raiddisks,
246                                               chunk, size, NULL, NULL))
247                 return 1;
248
249         /* now look at the subdevs */
250         info.array.active_disks = 0;
251         info.array.working_disks = 0;
252         dnum = 0;
253         for (dv=devlist; dv; dv=dv->next, dnum++) {
254                 char *dname = dv->devname;
255                 unsigned long long freesize;
256                 if (strcasecmp(dname, "missing")==0) {
257                         if (first_missing > dnum)
258                                 first_missing = dnum;
259                         if (second_missing > dnum && dnum > first_missing)
260                                 second_missing = dnum;
261                         missing_disks ++;
262                         continue;
263                 }
264                 info.array.working_disks++;
265                 if (dnum < raiddisks)
266                         info.array.active_disks++;
267                 if (st == NULL) {
268                         struct createinfo *ci = conf_get_create_info();
269                         if (ci)
270                                 st = ci->supertype;
271                 }
272                 if (st == NULL) {
273                         /* Need to choose a default metadata, which is different
274                          * depending on geometry of array.
275                          */
276                         int i;
277                         char *name = "default";
278                         for(i=0; !st && superlist[i]; i++) {
279                                 st = superlist[i]->match_metadata_desc(name);
280                                 if (st && !st->ss->validate_geometry
281                                                 (st, level, layout, raiddisks,
282                                                  chunk, size, dname, &freesize))
283                                         st = NULL;
284                         }
285
286                         if (!st) {
287                                 fprintf(stderr, Name ": device %s not suitable "
288                                         "for any style of array\n",
289                                         dname);
290                                 exit(2);
291                         }
292                         if (st->ss->major != 0 ||
293                             st->minor_version != 90)
294                                 fprintf(stderr, Name ": Defaulting to version"
295                                         " %d.%d metadata\n",
296                                         st->ss->major,
297                                         st->minor_version);
298                 } else {
299                         if (!st->ss->validate_geometry(st, level, layout,
300                                                        raiddisks,
301                                                        chunk, size, dname,
302                                                        &freesize)) {
303
304                                 fprintf(stderr,
305                                         Name ": %s is not suitable for "
306                                         "this array.\n",
307                                         dname);
308                                 fail = 1;
309                                 continue;
310                         }
311                 }
312
313                 freesize /= 2; /* convert to K */
314                 if (chunk) {
315                         /* round to chunk size */
316                         freesize = freesize & ~(chunk-1);
317                 }
318
319                 if (size && freesize < size) {
320                         fprintf(stderr, Name ": %s is smaller that given size."
321                                 " %lluK < %lluK + metadata\n",
322                                 dname, freesize, size);
323                         fail = 1;
324                         continue;
325                 }
326                 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
327                         maxdisc = dname;
328                         maxsize = freesize;
329                 }
330                 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
331                         mindisc = dname;
332                         minsize = freesize;
333                 }
334                 if (runstop != 1 || verbose >= 0) {
335                         int fd = open(dname, O_RDONLY, 0);
336                         if (fd <0 ) {
337                                 fprintf(stderr, Name ": Cannot open %s: %s\n",
338                                         dname, strerror(errno));
339                                 fail=1;
340                                 continue;
341                         }
342                         warn |= check_ext2(fd, dname);
343                         warn |= check_reiser(fd, dname);
344                         warn |= check_raid(fd, dname);
345                         close(fd);
346                 }
347         }
348         if (fail) {
349                 fprintf(stderr, Name ": create aborted\n");
350                 return 1;
351         }
352         if (size == 0) {
353                 if (mindisc == NULL && !have_container) {
354                         fprintf(stderr, Name ": no size and no drives given - aborting create.\n");
355                         return 1;
356                 }
357                 if (level > 0 || level == LEVEL_MULTIPATH
358                     || level == LEVEL_FAULTY
359                     || (st && st->ss->external) ) {
360                         /* size is meaningful */
361                         if (minsize > 0x100000000ULL && st->ss->major == 0) {
362                                 fprintf(stderr, Name ": devices too large for RAID level %d\n", level);
363                                 return 1;
364                         }
365                         size = minsize;
366                         if (verbose > 0)
367                                 fprintf(stderr, Name ": size set to %lluK\n", size);
368                 }
369         }
370         if (level > 0 && ((maxsize-size)*100 > maxsize)) {
371                 if (runstop != 1 || verbose >= 0)
372                         fprintf(stderr, Name ": largest drive (%s) exceed size (%lluK) by more than 1%%\n",
373                                 maxdisc, size);
374                 warn = 1;
375         }
376
377         if (warn) {
378                 if (runstop!= 1) {
379                         if (!ask("Continue creating array? ")) {
380                                 fprintf(stderr, Name ": create aborted.\n");
381                                 return 1;
382                         }
383                 } else {
384                         if (verbose > 0)
385                                 fprintf(stderr, Name ": creation continuing despite oddities due to --run\n");
386                 }
387         }
388
389         /* If this is raid4/5, we want to configure the last active slot
390          * as missing, so that a reconstruct happens (faster than re-parity)
391          * FIX: Can we do this for raid6 as well?
392          */
393         if (assume_clean==0 && force == 0 && first_missing >= raiddisks) {
394                 switch ( level ) {
395                 case 4:
396                 case 5:
397                         insert_point = raiddisks-1;
398                         sparedisks++;
399                         info.array.active_disks--;
400                         missing_disks++;
401                         break;
402                 default:
403                         break;
404                 }
405         }
406         /* For raid6, if creating with 1 missing drive, make a good drive
407          * into a spare, else the create will fail
408          */
409         if (assume_clean == 0 && force == 0 && first_missing < raiddisks &&
410             second_missing >= raiddisks && level == 6) {
411                 insert_point = raiddisks - 1;
412                 if (insert_point == first_missing)
413                         insert_point--;
414                 sparedisks ++;
415                 info.array.active_disks--;
416                 missing_disks++;
417         }
418
419         if (level <= 0 && first_missing < subdevs * 2) {
420                 fprintf(stderr,
421                         Name ": This level does not support missing devices\n");
422                 return 1;
423         }
424
425         /* Ok, lets try some ioctls */
426
427         info.array.level = level;
428         info.array.size = size;
429         info.array.raid_disks = raiddisks;
430         /* The kernel should *know* what md_minor we are dealing
431          * with, but it chooses to trust me instead. Sigh
432          */
433         info.array.md_minor = 0;
434         if (fstat(mdfd, &stb)==0)
435                 info.array.md_minor = minor(stb.st_rdev);
436         info.array.not_persistent = 0;
437
438         if ( ( (level == 4 || level == 5) &&
439                (insert_point < raiddisks || first_missing < raiddisks) )
440              ||
441              ( level == 6 && (insert_point < raiddisks
442                               || second_missing < raiddisks))
443              ||
444              assume_clean
445                 )
446                 info.array.state = 1; /* clean, but one+ drive will be missing*/
447         else
448                 info.array.state = 0; /* not clean, but no errors */
449
450         if (level == 10) {
451                 /* for raid10, the bitmap size is the capacity of the array,
452                  * which is array.size * raid_disks / ncopies;
453                  * .. but convert to sectors.
454                  */
455                 int ncopies = ((layout>>8) & 255) * (layout & 255);
456                 bitmapsize = (unsigned long long)size * raiddisks / ncopies * 2;
457 /*              printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
458         } else
459                 bitmapsize = (unsigned long long)size * 2;
460
461         /* There is lots of redundancy in these disk counts,
462          * raid_disks is the most meaningful value
463          *          it describes the geometry of the array
464          *          it is constant
465          * nr_disks is total number of used slots.
466          *          it should be raid_disks+spare_disks
467          * spare_disks is the number of extra disks present
468          *          see above
469          * active_disks is the number of working disks in
470          *          active slots. (With raid_disks)
471          * working_disks is the total number of working disks,
472          *          including spares
473          * failed_disks is the number of disks marked failed
474          *
475          * Ideally, the kernel would keep these (except raid_disks)
476          * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
477          * So for now, we assume that all raid and spare
478          * devices will be given.
479          */
480         info.array.spare_disks=sparedisks;
481         info.array.failed_disks=missing_disks;
482         info.array.nr_disks = info.array.working_disks
483                 + info.array.failed_disks;
484         info.array.layout = layout;
485         info.array.chunk_size = chunk*1024;
486         info.array.major_version = st->ss->major;
487
488         if (name == NULL || *name == 0) {
489                 /* base name on mddev */
490                 /*  /dev/md0 -> 0
491                  *  /dev/md_d0 -> d0
492                  *  /dev/md/1 -> 1
493                  *  /dev/md/d1 -> d1
494                  *  /dev/md/home -> home
495                  *  /dev/mdhome -> home
496                  */
497                 name = strrchr(mddev, '/');
498                 if (name) {
499                         name++;
500                         if (strncmp(name, "md_d", 4)==0 &&
501                             strlen(name) > 4 &&
502                             isdigit(name[4]) &&
503                             (name-mddev) == 5 /* /dev/ */)
504                                 name += 3;
505                         else if (strncmp(name, "md", 2)==0 &&
506                                  strlen(name) > 2 &&
507                                  isdigit(name[2]) &&
508                                  (name-mddev) == 5 /* /dev/ */)
509                                 name += 2;
510                 }
511         }
512         if (!st->ss->init_super(st, &info.array, size, name, homehost, uuid))
513                 return 1;
514
515         st->ss->getinfo_super(st, &info);
516
517         if (bitmap_file && vers < 9003) {
518                 major_num = BITMAP_MAJOR_HOSTENDIAN;
519 #ifdef __BIG_ENDIAN
520                 fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n"
521                         "  between different architectured.  Consider upgrading the Linux kernel.\n");
522 #endif
523         }
524
525         if (bitmap_file && strcmp(bitmap_file, "internal")==0) {
526                 if ((vers%100) < 2) {
527                         fprintf(stderr, Name ": internal bitmaps not supported by this kernel.\n");
528                         return 1;
529                 }
530                 if (!st->ss->add_internal_bitmap(st, &bitmap_chunk,
531                                                  delay, write_behind,
532                                                  bitmapsize, 1, major_num)) {
533                         fprintf(stderr, Name ": Given bitmap chunk size not supported.\n");
534                         return 1;
535                 }
536                 bitmap_file = NULL;
537         }
538
539
540         sra = sysfs_read(mdfd, 0, 0);
541
542         if (st->ss->external) {
543                 char ver[100];
544                 if (st->ss->external == 1)
545                         /* container */
546                         strcat(strcpy(ver, "external:"),
547                                info.text_version);
548                 else {
549                         /* member */
550                         sprintf(ver, "external:/%s/%d",
551                                 devnum2devname(st->container_dev),
552                                 st->container_member);
553                         /* When creating a member, we need to be careful
554                          * to negotiate with mdmon properly.
555                          * If it is already running, we cannot write to
556                          * the devices and must ask it to do that part.
557                          * If it isn't running, we write to the devices,
558                          * and then start it.
559                          * We hold an exclusive open on the container
560                          * device to make sure mdmon doesn't exit after
561                          * we checked that it is running.
562                          *
563                          * For now, fail if it is already running.
564                          */
565                         container_fd = open_dev_excl(st->container_dev);
566                         if (container_fd < 0) {
567                                 fprintf(stderr, Name ": Cannot get exclusive "
568                                         "open on container - weird.\n");
569                                 return 1;
570                         }
571                         if (mdmon_running(st->container_dev)) {
572                                 fprintf(stderr, Name ": mdmon already running "
573                                         "for %s - sorry\n",
574                                         devnum2devname(st->container_dev));
575                                 return 1;
576                         }
577                         need_mdmon = 1;
578                 }
579                 if ((vers % 100) < 2 ||
580                     sra == NULL ||
581                     sysfs_set_str(sra, NULL, "metadata_version",
582                                   ver) < 0) {
583                         fprintf(stderr, Name ": This kernel does not "
584                                 "support external metadata.\n");
585                         return 1;
586                 }
587                 rv = sysfs_set_array(sra, &info);
588         } else  if ((vers % 100) >= 1) { /* can use different versions */
589                 mdu_array_info_t inf;
590                 memset(&inf, 0, sizeof(inf));
591                 inf.major_version = st->ss->major;
592                 inf.minor_version = st->minor_version;
593                 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
594         } else
595                 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
596         if (rv) {
597                 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
598                         mddev, strerror(errno));
599                 return 1;
600         }
601
602         if (bitmap_file) {
603                 int uuid[4];
604
605                 st->ss->uuid_from_super(st, uuid);
606                 if (CreateBitmap(bitmap_file, force, (char*)uuid, bitmap_chunk,
607                                  delay, write_behind,
608                                  bitmapsize,
609                                  major_num)) {
610                         return 1;
611                 }
612                 bitmap_fd = open(bitmap_file, O_RDWR);
613                 if (bitmap_fd < 0) {
614                         fprintf(stderr, Name ": weird: %s cannot be openned\n",
615                                 bitmap_file);
616                         return 1;
617                 }
618                 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
619                         fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
620                                 mddev, strerror(errno));
621                         return 1;
622                 }
623         }
624
625
626         for (pass=1; pass <=2 ; pass++) {
627                 mddev_dev_t moved_disk = NULL; /* the disk that was moved out of the insert point */
628
629                 for (dnum=0, dv = devlist ; dv ;
630                      dv=(dv->next)?(dv->next):moved_disk, dnum++) {
631                         int fd;
632                         struct stat stb;
633
634                         info.disk.number = dnum;
635                         if (dnum == insert_point) {
636                                 moved_disk = dv;
637                         }
638                         info.disk.raid_disk = info.disk.number;
639                         if (info.disk.raid_disk < raiddisks)
640                                 info.disk.state = (1<<MD_DISK_ACTIVE) |
641                                                 (1<<MD_DISK_SYNC);
642                         else
643                                 info.disk.state = 0;
644                         if (dv->writemostly)
645                                 info.disk.state |= (1<<MD_DISK_WRITEMOSTLY);
646
647                         if (dnum == insert_point ||
648                             strcasecmp(dv->devname, "missing")==0)
649                                 continue;
650
651                         if (st->ss->external == 2)
652                                 fd = open(dv->devname, O_RDWR, 0);
653                         else
654                                 fd = open(dv->devname, O_RDWR|O_EXCL,0);
655
656                         if (fd < 0) {
657                                 fprintf(stderr, Name ": failed to open %s "
658                                         "after earlier success - aborting\n",
659                                         dv->devname);
660                                 return 1;
661                         }
662                         fstat(fd, &stb);
663                         info.disk.major = major(stb.st_rdev);
664                         info.disk.minor = minor(stb.st_rdev);
665
666                         switch(pass){
667                         case 1:
668                                 remove_partitions(fd);
669                                 st->ss->add_to_super(st, &info.disk,
670                                                      fd, dv->devname);
671                                 break;
672                         case 2:
673                                 close(fd);
674                                 info.component_size = info.array.size * 2;
675                                 info.errors = 0;
676                                 rv = 0;
677
678                                 if (st->ss->external) {
679                                         st->ss->getinfo_super_n(st, &info);
680                                         rv = sysfs_add_disk(sra, &info);
681                                 } else {
682                                         rv = ioctl(mdfd, ADD_NEW_DISK,
683                                                  &info.disk);
684                                 }
685                                 if (rv) {
686                                         fprintf(stderr,
687                                                 Name ": ADD_NEW_DISK for %s "
688                                                 "failed: %s\n",
689                                                 dv->devname, strerror(errno));
690                                         st->ss->free_super(st);
691                                         return 1;
692                                 }
693                                 break;
694                         }
695                         if (dv == moved_disk && dnum != insert_point) break;
696                 }
697                 if (pass == 1)
698                         st->ss->write_init_super(st);
699         }
700         st->ss->free_super(st);
701
702         /* param is not actually used */
703         if (level == LEVEL_CONTAINER)
704                 /* No need to start */
705                 ;
706         else if (runstop == 1 || subdevs >= raiddisks) {
707                 if (st->ss->external) {
708                         switch(level) {
709                         case LEVEL_LINEAR:
710                         case LEVEL_MULTIPATH:
711                         case 0:
712                                 sysfs_set_str(sra, NULL, "array_state",
713                                               "active");
714                                 need_mdmon = 0;
715                                 break;
716                         default:
717                                 sysfs_set_str(sra, NULL, "array_state",
718                                               "readonly");
719                                 break;
720                         }
721                 } else {
722                         mdu_param_t param;
723                         if (ioctl(mdfd, RUN_ARRAY, &param)) {
724                                 fprintf(stderr, Name ": RUN_ARRAY failed: %s\n",
725                                         strerror(errno));
726                                 Manage_runstop(mddev, mdfd, -1, 0);
727                                 return 1;
728                         }
729                 }
730                 if (verbose >= 0)
731                         fprintf(stderr, Name ": array %s started.\n", mddev);
732                 if (st->ss->external == 2) {
733                         if (need_mdmon) {
734                                 int dn = st->container_dev;
735                                 int i;
736                                 switch(fork()) {
737                                 case 0:
738                                         /* FIXME yuk. CLOSE_EXEC?? */
739                                         for (i=3; i < 100; i++)
740                                                 close(i);
741                                         execl("./mdmon", "mdmon",
742                                               map_dev(dev2major(dn),
743                                                       dev2minor(dn),
744                                                       1), NULL);
745                                         exit(1);
746                                 case -1: fprintf(stderr, Name ": cannot fork. "
747                                                  "Array remains readonly\n");
748                                         return 1;
749                                 default: ; /* parent - good */
750                                 }
751                         } else
752                                 signal_mdmon(st->container_dev);
753                         /* FIXME wait for mdmon to set array to read-auto */
754                         sleep(1);
755                         close(container_fd);
756                 }
757         } else {
758                 fprintf(stderr, Name ": not starting array - not enough devices.\n");
759         }
760         return 0;
761 }