292f92a9ae639ecaa981e273e30a9f3686345460
[thirdparty/mdadm.git] / Create.c
1 /*
2  * mdadm - manage Linux "md" devices aka RAID arrays.
3  *
4  * Copyright (C) 2001-2013 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@suse.de>
23  */
24
25 #include        "mdadm.h"
26 #include        "md_u.h"
27 #include        "md_p.h"
28 #include        <ctype.h>
29
30 static int round_size_and_verify(unsigned long long *size, int chunk)
31 {
32         if (*size == 0)
33                 return 0;
34         *size &= ~(unsigned long long)(chunk - 1);
35         if (*size == 0) {
36                 pr_err("Size cannot be smaller than chunk.\n");
37                 return 1;
38         }
39         return 0;
40 }
41
42 static int default_layout(struct supertype *st, int level, int verbose)
43 {
44         int layout = UnSet;
45
46         if (st && st->ss->default_geometry)
47                 st->ss->default_geometry(st, &level, &layout, NULL);
48
49         if (layout == UnSet)
50                 switch(level) {
51                 default: /* no layout */
52                         layout = 0;
53                         break;
54                 case 10:
55                         layout = 0x102; /* near=2, far=1 */
56                         if (verbose > 0)
57                                 pr_err("layout defaults to n2\n");
58                         break;
59                 case 5:
60                 case 6:
61                         layout = map_name(r5layout, "default");
62                         if (verbose > 0)
63                                 pr_err("layout defaults to %s\n", map_num(r5layout, layout));
64                         break;
65                 case LEVEL_FAULTY:
66                         layout = map_name(faultylayout, "default");
67
68                         if (verbose > 0)
69                                 pr_err("layout defaults to %s\n", map_num(faultylayout, layout));
70                         break;
71                 }
72
73         return layout;
74 }
75
76 int Create(struct supertype *st, char *mddev,
77            char *name, int *uuid,
78            int subdevs, struct mddev_dev *devlist,
79            struct shape *s,
80            struct context *c, unsigned long long data_offset)
81 {
82         /*
83          * Create a new raid array.
84          *
85          * First check that necessary details are available
86          * (i.e. level, raid-disks)
87          *
88          * Then check each disk to see what might be on it
89          * and report anything interesting.
90          *
91          * If anything looks odd, and runstop not set,
92          * abort.
93          *
94          * SET_ARRAY_INFO and ADD_NEW_DISK, and
95          * if runstop==run, or raiddisks disks were used,
96          * RUN_ARRAY
97          */
98         int mdfd;
99         unsigned long long minsize = 0, maxsize = 0;
100         char *mindisc = NULL;
101         char *maxdisc = NULL;
102         int dnum, raid_disk_num;
103         struct mddev_dev *dv;
104         dev_t rdev;
105         int fail = 0, warn = 0;
106         int first_missing = subdevs * 2;
107         int second_missing = subdevs * 2;
108         int missing_disks = 0;
109         int insert_point = subdevs * 2; /* where to insert a missing drive */
110         int total_slots;
111         int pass;
112         int rv;
113         int bitmap_fd;
114         int have_container = 0;
115         int container_fd = -1;
116         int need_mdmon = 0;
117         unsigned long long bitmapsize;
118         struct mdinfo info, *infos;
119         int did_default = 0;
120         int do_default_layout = 0;
121         int do_default_chunk = 0;
122         unsigned long safe_mode_delay = 0;
123         char chosen_name[1024];
124         struct map_ent *map = NULL;
125         unsigned long long newsize;
126         mdu_array_info_t inf;
127
128         int major_num = BITMAP_MAJOR_HI;
129         if (s->bitmap_file && strcmp(s->bitmap_file, "clustered") == 0) {
130                 major_num = BITMAP_MAJOR_CLUSTERED;
131                 if (c->nodes <= 1) {
132                         pr_err("At least 2 nodes are needed for cluster-md\n");
133                         return 1;
134                 }
135         }
136
137         memset(&info, 0, sizeof(info));
138         if (s->level == UnSet && st && st->ss->default_geometry)
139                 st->ss->default_geometry(st, &s->level, NULL, NULL);
140         if (s->level == UnSet) {
141                 pr_err("a RAID level is needed to create an array.\n");
142                 return 1;
143         }
144         if (s->raiddisks < 4 && s->level == 6) {
145                 pr_err("at least 4 raid-devices needed for level 6\n");
146                 return 1;
147         }
148         if (s->raiddisks > 256 && s->level == 6) {
149                 pr_err("no more than 256 raid-devices supported for level 6\n");
150                 return 1;
151         }
152         if (s->raiddisks < 2 && s->level >= 4) {
153                 pr_err("at least 2 raid-devices needed for level 4 or 5\n");
154                 return 1;
155         }
156         if (s->level <= 0 && s->sparedisks) {
157                 pr_err("This level does not support spare devices\n");
158                 return 1;
159         }
160
161         if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) {
162                 /* If given a single device, it might be a container, and we can
163                  * extract a device list from there
164                  */
165                 int fd;
166
167                 memset(&inf, 0, sizeof(inf));
168                 fd = open(devlist->devname, O_RDONLY);
169                 if (fd >= 0 &&
170                     md_get_array_info(fd, &inf) == 0 && inf.raid_disks == 0) {
171                         /* yep, looks like a container */
172                         if (st) {
173                                 rv = st->ss->load_container(st, fd,
174                                                             devlist->devname);
175                                 if (rv == 0)
176                                         have_container = 1;
177                         } else {
178                                 st = super_by_fd(fd, NULL);
179                                 if (st && !(rv = st->ss->
180                                             load_container(st, fd,
181                                                            devlist->devname)))
182                                         have_container = 1;
183                                 else
184                                         st = NULL;
185                         }
186                         if (have_container) {
187                                 subdevs = s->raiddisks;
188                                 first_missing = subdevs * 2;
189                                 second_missing = subdevs * 2;
190                                 insert_point = subdevs * 2;
191                         }
192                 }
193                 if (fd >= 0)
194                         close(fd);
195         }
196         if (st && st->ss->external && s->sparedisks) {
197                 pr_err("This metadata type does not support spare disks at create time\n");
198                 return 1;
199         }
200         if (subdevs > s->raiddisks+s->sparedisks+s->journaldisks) {
201                 pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs, s->raiddisks+s->sparedisks);
202                 return 1;
203         }
204         if (!have_container && subdevs < s->raiddisks+s->sparedisks+s->journaldisks) {
205                 pr_err("You haven't given enough devices (real or missing) to create this array\n");
206                 return 1;
207         }
208         if (s->bitmap_file && s->level <= 0) {
209                 pr_err("bitmaps not meaningful with level %s\n",
210                         map_num(pers, s->level)?:"given");
211                 return 1;
212         }
213
214         /* now set some defaults */
215
216         if (s->layout == UnSet) {
217                 do_default_layout = 1;
218                 s->layout = default_layout(st, s->level, c->verbose);
219         }
220
221         if (s->level == 10)
222                 /* check layout fits in array*/
223                 if ((s->layout&255) * ((s->layout>>8)&255) > s->raiddisks) {
224                         pr_err("that layout requires at least %d devices\n",
225                                 (s->layout&255) * ((s->layout>>8)&255));
226                         return 1;
227                 }
228
229         switch(s->level) {
230         case 4:
231         case 5:
232         case 10:
233         case 6:
234         case 0:
235                 if (s->chunk == 0 || s->chunk == UnSet) {
236                         s->chunk = UnSet;
237                         do_default_chunk = 1;
238                         /* chunk will be set later */
239                 }
240                 break;
241         case LEVEL_LINEAR:
242                 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
243                 if (get_linux_version() < 2006016 && s->chunk == 0) {
244                         s->chunk = 64;
245                         if (c->verbose > 0)
246                                 pr_err("chunk size defaults to 64K\n");
247                 }
248                 break;
249         case 1:
250         case LEVEL_FAULTY:
251         case LEVEL_MULTIPATH:
252         case LEVEL_CONTAINER:
253                 if (s->chunk) {
254                         s->chunk = 0;
255                         if (c->verbose > 0)
256                                 pr_err("chunk size ignored for this level\n");
257                 }
258                 break;
259         default:
260                 pr_err("unknown level %d\n", s->level);
261                 return 1;
262         }
263
264         if (s->size == MAX_SIZE)
265                 /* use '0' to mean 'max' now... */
266                 s->size = 0;
267         if (s->size && s->chunk && s->chunk != UnSet)
268                 if (round_size_and_verify(&s->size, s->chunk))
269                         return 1;
270
271         newsize = s->size * 2;
272         if (st && ! st->ss->validate_geometry(st, s->level, s->layout, s->raiddisks,
273                                               &s->chunk, s->size*2,
274                                               data_offset, NULL,
275                                               &newsize, s->consistency_policy,
276                                               c->verbose >= 0))
277                 return 1;
278
279         if (s->chunk && s->chunk != UnSet) {
280                 newsize &= ~(unsigned long long)(s->chunk*2 - 1);
281                 if (do_default_chunk) {
282                         /* default chunk was just set */
283                         if (c->verbose > 0)
284                                 pr_err("chunk size defaults to %dK\n", s->chunk);
285                         if (round_size_and_verify(&s->size, s->chunk))
286                                 return 1;
287                         do_default_chunk = 0;
288                 }
289         }
290
291         if (s->size == 0) {
292                 s->size = newsize / 2;
293                 if (s->level == 1)
294                         /* If this is ever reshaped to RAID5, we will
295                          * need a chunksize.  So round it off a bit
296                          * now just to be safe
297                          */
298                         s->size &= ~(64ULL-1);
299
300                 if (s->size && c->verbose > 0)
301                         pr_err("setting size to %lluK\n", s->size);
302         }
303
304         /* now look at the subdevs */
305         info.array.active_disks = 0;
306         info.array.working_disks = 0;
307         dnum = 0;
308         for (dv = devlist; dv; dv = dv->next)
309                 if (data_offset == VARIABLE_OFFSET)
310                         dv->data_offset = INVALID_SECTORS;
311                 else
312                         dv->data_offset = data_offset;
313
314         for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) {
315                 char *dname = dv->devname;
316                 unsigned long long freesize;
317                 int dfd;
318                 char *doff;
319
320                 if (strcasecmp(dname, "missing") == 0) {
321                         if (first_missing > dnum)
322                                 first_missing = dnum;
323                         if (second_missing > dnum && dnum > first_missing)
324                                 second_missing = dnum;
325                         missing_disks ++;
326                         continue;
327                 }
328                 if (data_offset == VARIABLE_OFFSET) {
329                         doff = strchr(dname, ':');
330                         if (doff) {
331                                 *doff++ = 0;
332                                 dv->data_offset = parse_size(doff);
333                         } else
334                                 dv->data_offset = INVALID_SECTORS;
335                 } else
336                         dv->data_offset = data_offset;
337
338                 dfd = open(dname, O_RDONLY);
339                 if (dfd < 0) {
340                         pr_err("cannot open %s: %s\n",
341                                 dname, strerror(errno));
342                         exit(2);
343                 }
344                 if (!fstat_is_blkdev(dfd, dname, NULL)) {
345                         close(dfd);
346                         exit(2);
347                 }
348                 close(dfd);
349                 info.array.working_disks++;
350                 if (dnum < s->raiddisks && dv->disposition != 'j')
351                         info.array.active_disks++;
352                 if (st == NULL) {
353                         struct createinfo *ci = conf_get_create_info();
354                         if (ci)
355                                 st = ci->supertype;
356                 }
357                 if (st == NULL) {
358                         /* Need to choose a default metadata, which is different
359                          * depending on geometry of array.
360                          */
361                         int i;
362                         char *name = "default";
363                         for(i = 0; !st && superlist[i]; i++) {
364                                 st = superlist[i]->match_metadata_desc(name);
365                                 if (!st)
366                                         continue;
367                                 if (do_default_layout)
368                                         s->layout = default_layout(st, s->level, c->verbose);
369                                 switch (st->ss->validate_geometry(
370                                                 st, s->level, s->layout, s->raiddisks,
371                                                 &s->chunk, s->size*2,
372                                                 dv->data_offset, dname,
373                                                 &freesize, s->consistency_policy,
374                                                 c->verbose > 0)) {
375                                 case -1: /* Not valid, message printed, and not
376                                           * worth checking any further */
377                                         exit(2);
378                                         break;
379                                 case 0: /* Geometry not valid */
380                                         free(st);
381                                         st = NULL;
382                                         s->chunk = do_default_chunk ? UnSet : s->chunk;
383                                         break;
384                                 case 1: /* All happy */
385                                         break;
386                                 }
387                         }
388
389                         if (!st) {
390                                 int dfd = open(dname, O_RDONLY|O_EXCL);
391                                 if (dfd < 0) {
392                                         pr_err("cannot open %s: %s\n",
393                                                 dname, strerror(errno));
394                                         exit(2);
395                                 }
396                                 pr_err("device %s not suitable for any style of array\n",
397                                         dname);
398                                 exit(2);
399                         }
400                         if (st->ss != &super0 ||
401                             st->minor_version != 90)
402                                 did_default = 1;
403                 } else {
404                         if (do_default_layout)
405                                 s->layout = default_layout(st, s->level, 0);
406                         if (!st->ss->validate_geometry(st, s->level, s->layout,
407                                                        s->raiddisks,
408                                                        &s->chunk, s->size*2,
409                                                        dv->data_offset,
410                                                        dname, &freesize,
411                                                        s->consistency_policy,
412                                                        c->verbose >= 0)) {
413
414                                 pr_err("%s is not suitable for this array.\n",
415                                        dname);
416                                 fail = 1;
417                                 continue;
418                         }
419                 }
420
421                 if (dv->disposition == 'j')
422                         goto skip_size_check;  /* skip write journal for size check */
423
424                 freesize /= 2; /* convert to K */
425                 if (s->chunk && s->chunk != UnSet) {
426                         /* round to chunk size */
427                         freesize = freesize & ~(s->chunk-1);
428                         if (do_default_chunk) {
429                                 /* default chunk was just set */
430                                 if (c->verbose > 0)
431                                         pr_err("chunk size defaults to %dK\n", s->chunk);
432                                 if (round_size_and_verify(&s->size, s->chunk))
433                                         return 1;
434                                 do_default_chunk = 0;
435                         }
436                 }
437                 if (!freesize) {
438                         pr_err("no free space left on %s\n", dname);
439                         fail = 1;
440                         continue;
441                 }
442
443                 if (s->size && freesize < s->size) {
444                         pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n",
445                                 dname, freesize, s->size);
446                         fail = 1;
447                         continue;
448                 }
449                 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
450                         maxdisc = dname;
451                         maxsize = freesize;
452                 }
453                 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
454                         mindisc = dname;
455                         minsize = freesize;
456                 }
457         skip_size_check:
458                 if (c->runstop != 1 || c->verbose >= 0) {
459                         int fd = open(dname, O_RDONLY);
460                         if (fd < 0) {
461                                 pr_err("Cannot open %s: %s\n",
462                                         dname, strerror(errno));
463                                 fail = 1;
464                                 continue;
465                         }
466                         warn |= check_ext2(fd, dname);
467                         warn |= check_reiser(fd, dname);
468                         warn |= check_raid(fd, dname);
469                         if (strcmp(st->ss->name, "1.x") == 0 &&
470                             st->minor_version >= 1)
471                                 /* metadata at front */
472                                 warn |= check_partitions(fd, dname, 0, 0);
473                         else if (s->level == 1 || s->level == LEVEL_CONTAINER ||
474                                  (s->level == 0 && s->raiddisks == 1))
475                                 /* partitions could be meaningful */
476                                 warn |= check_partitions(fd, dname, freesize*2, s->size*2);
477                         else
478                                 /* partitions cannot be meaningful */
479                                 warn |= check_partitions(fd, dname, 0, 0);
480                         if (strcmp(st->ss->name, "1.x") == 0 &&
481                             st->minor_version >= 1 &&
482                             did_default &&
483                             s->level == 1 &&
484                             (warn & 1024) == 0) {
485                                 warn |= 1024;
486                                 pr_err("Note: this array has metadata at the start and\n"
487                                         "    may not be suitable as a boot device.  If you plan to\n"
488                                         "    store '/boot' on this device please ensure that\n"
489                                         "    your boot-loader understands md/v1.x metadata, or use\n"
490                                         "    --metadata=0.90\n");
491                         }
492                         close(fd);
493                 }
494         }
495         if (missing_disks == dnum && !have_container) {
496                 pr_err("Subdevs can't be all missing\n");
497                 return 1;
498         }
499         if (s->raiddisks + s->sparedisks > st->max_devs) {
500                 pr_err("Too many devices: %s metadata only supports %d\n",
501                         st->ss->name, st->max_devs);
502                 return 1;
503         }
504         if (have_container)
505                 info.array.working_disks = s->raiddisks;
506         if (fail) {
507                 pr_err("create aborted\n");
508                 return 1;
509         }
510         if (s->size == 0) {
511                 if (mindisc == NULL && !have_container) {
512                         pr_err("no size and no drives given - aborting create.\n");
513                         return 1;
514                 }
515                 if (s->level > 0 || s->level == LEVEL_MULTIPATH ||
516                     s->level == LEVEL_FAULTY || st->ss->external) {
517                         /* size is meaningful */
518                         if (!st->ss->validate_geometry(st, s->level, s->layout,
519                                                        s->raiddisks,
520                                                        &s->chunk, minsize*2,
521                                                        data_offset,
522                                                        NULL, NULL,
523                                                        s->consistency_policy, 0)) {
524                                 pr_err("devices too large for RAID level %d\n", s->level);
525                                 return 1;
526                         }
527                         s->size = minsize;
528                         if (s->level == 1)
529                                 /* If this is ever reshaped to RAID5, we will
530                                  * need a chunksize.  So round it off a bit
531                                  * now just to be safe
532                                  */
533                                 s->size &= ~(64ULL-1);
534                         if (c->verbose > 0)
535                                 pr_err("size set to %lluK\n", s->size);
536                 }
537         }
538
539         if (!s->bitmap_file &&
540             s->level >= 1 &&
541             st->ss->add_internal_bitmap &&
542             (s->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
543              s->consistency_policy != CONSISTENCY_POLICY_PPL) &&
544             (s->write_behind || s->size > 100*1024*1024ULL)) {
545                 if (c->verbose > 0)
546                         pr_err("automatically enabling write-intent bitmap on large array\n");
547                 s->bitmap_file = "internal";
548         }
549         if (s->bitmap_file && strcmp(s->bitmap_file, "none") == 0)
550                 s->bitmap_file = NULL;
551
552         if (s->consistency_policy == CONSISTENCY_POLICY_PPL &&
553             !st->ss->write_init_ppl) {
554                 pr_err("%s metadata does not support PPL\n", st->ss->name);
555                 return 1;
556         }
557
558         if (!have_container && s->level > 0 && ((maxsize-s->size)*100 > maxsize)) {
559                 if (c->runstop != 1 || c->verbose >= 0)
560                         pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
561                                 maxdisc, s->size);
562                 warn = 1;
563         }
564
565         if (st->ss->detail_platform && st->ss->detail_platform(0, 1, NULL) != 0) {
566                 if (c->runstop != 1 || c->verbose >= 0)
567                         pr_err("%s unable to enumerate platform support\n"
568                                 "    array may not be compatible with hardware/firmware\n",
569                                 st->ss->name);
570                 warn = 1;
571         }
572         st->nodes = c->nodes;
573         st->cluster_name = c->homecluster;
574
575         if (warn) {
576                 if (c->runstop!= 1) {
577                         if (!ask("Continue creating array? ")) {
578                                 pr_err("create aborted.\n");
579                                 return 1;
580                         }
581                 } else {
582                         if (c->verbose > 0)
583                                 pr_err("creation continuing despite oddities due to --run\n");
584                 }
585         }
586
587         /* If this is raid4/5, we want to configure the last active slot
588          * as missing, so that a reconstruct happens (faster than re-parity)
589          * FIX: Can we do this for raid6 as well?
590          */
591         if (st->ss->external == 0 && s->assume_clean == 0 &&
592             c->force == 0 && first_missing >= s->raiddisks) {
593                 switch (s->level) {
594                 case 4:
595                 case 5:
596                         insert_point = s->raiddisks-1;
597                         s->sparedisks++;
598                         info.array.active_disks--;
599                         missing_disks++;
600                         break;
601                 default:
602                         break;
603                 }
604         }
605         /* For raid6, if creating with 1 missing drive, make a good drive
606          * into a spare, else the create will fail
607          */
608         if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks &&
609             st->ss->external == 0 &&
610             second_missing >= s->raiddisks && s->level == 6) {
611                 insert_point = s->raiddisks - 1;
612                 if (insert_point == first_missing)
613                         insert_point--;
614                 s->sparedisks ++;
615                 info.array.active_disks--;
616                 missing_disks++;
617         }
618
619         if (s->level <= 0 && first_missing < subdevs * 2) {
620                 pr_err("This level does not support missing devices\n");
621                 return 1;
622         }
623
624         /* We need to create the device */
625         map_lock(&map);
626         mdfd = create_mddev(mddev, name, c->autof, LOCAL, chosen_name, 1);
627         if (mdfd < 0) {
628                 map_unlock(&map);
629                 return 1;
630         }
631         /* verify if chosen_name is not in use,
632          * it could be in conflict with already existing device
633          * e.g. container, array
634          */
635         if (strncmp(chosen_name, "/dev/md/", 8) == 0 &&
636             map_by_name(&map, chosen_name+8) != NULL) {
637                 pr_err("Array name %s is in use already.\n",
638                         chosen_name);
639                 close(mdfd);
640                 map_unlock(&map);
641                 udev_unblock();
642                 return 1;
643         }
644         mddev = chosen_name;
645
646         memset(&inf, 0, sizeof(inf));
647         md_get_array_info(mdfd, &inf);
648         if (inf.working_disks != 0) {
649                 pr_err("another array by this name is already running.\n");
650                 goto abort_locked;
651         }
652
653         /* Ok, lets try some ioctls */
654
655         info.array.level = s->level;
656         info.array.size = s->size;
657         info.array.raid_disks = s->raiddisks;
658         /* The kernel should *know* what md_minor we are dealing
659          * with, but it chooses to trust me instead. Sigh
660          */
661         info.array.md_minor = 0;
662         if (fstat_is_blkdev(mdfd, mddev, &rdev))
663                 info.array.md_minor = minor(rdev);
664         info.array.not_persistent = 0;
665
666         if (((s->level == 4 || s->level == 5) &&
667              (insert_point < s->raiddisks || first_missing < s->raiddisks)) ||
668             (s->level == 6 && (insert_point < s->raiddisks ||
669                                second_missing < s->raiddisks)) ||
670             (s->level <= 0) || s->assume_clean) {
671                 info.array.state = 1; /* clean, but one+ drive will be missing*/
672                 info.resync_start = MaxSector;
673         } else {
674                 info.array.state = 0; /* not clean, but no errors */
675                 info.resync_start = 0;
676         }
677         if (s->level == 10) {
678                 /* for raid10, the bitmap size is the capacity of the array,
679                  * which is array.size * raid_disks / ncopies;
680                  * .. but convert to sectors.
681                  */
682                 int ncopies = ((s->layout>>8) & 255) * (s->layout & 255);
683                 bitmapsize = s->size * s->raiddisks / ncopies * 2;
684 /*              printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
685         } else
686                 bitmapsize = s->size * 2;
687
688         /* There is lots of redundancy in these disk counts,
689          * raid_disks is the most meaningful value
690          *          it describes the geometry of the array
691          *          it is constant
692          * nr_disks is total number of used slots.
693          *          it should be raid_disks+spare_disks
694          * spare_disks is the number of extra disks present
695          *          see above
696          * active_disks is the number of working disks in
697          *          active slots. (With raid_disks)
698          * working_disks is the total number of working disks,
699          *          including spares
700          * failed_disks is the number of disks marked failed
701          *
702          * Ideally, the kernel would keep these (except raid_disks)
703          * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
704          * So for now, we assume that all raid and spare
705          * devices will be given.
706          */
707         info.array.spare_disks=s->sparedisks;
708         info.array.failed_disks=missing_disks;
709         info.array.nr_disks = info.array.working_disks
710                 + info.array.failed_disks;
711         info.array.layout = s->layout;
712         info.array.chunk_size = s->chunk*1024;
713
714         if (name == NULL || *name == 0) {
715                 /* base name on mddev */
716                 /*  /dev/md0 -> 0
717                  *  /dev/md_d0 -> d0
718                  *  /dev/md_foo -> foo
719                  *  /dev/md/1 -> 1
720                  *  /dev/md/d1 -> d1
721                  *  /dev/md/home -> home
722                  *  /dev/mdhome -> home
723                  */
724                 /* FIXME compare this with rules in create_mddev */
725                 name = strrchr(mddev, '/');
726                 if (name) {
727                         name++;
728                         if (strncmp(name, "md_", 3) == 0 &&
729                             strlen(name) > 3 && (name-mddev) == 5 /* /dev/ */)
730                                 name += 3;
731                         else if (strncmp(name, "md", 2) == 0 &&
732                                  strlen(name) > 2 && isdigit(name[2]) &&
733                                  (name-mddev) == 5 /* /dev/ */)
734                                 name += 2;
735                 }
736         }
737         if (!st->ss->init_super(st, &info.array, s, name, c->homehost, uuid,
738                                 data_offset))
739                 goto abort_locked;
740
741         total_slots = info.array.nr_disks;
742         st->ss->getinfo_super(st, &info, NULL);
743         if (sysfs_init(&info, mdfd, NULL)) {
744                 pr_err("unable to initialize sysfs\n");
745                 goto abort_locked;
746         }
747
748         if (did_default && c->verbose >= 0) {
749                 if (is_subarray(info.text_version)) {
750                         char devnm[32];
751                         char *ep;
752                         struct mdinfo *mdi;
753
754                         strncpy(devnm, info.text_version+1, 32);
755                         devnm[31] = 0;
756                         ep = strchr(devnm, '/');
757                         if (ep)
758                                 *ep = 0;
759
760                         mdi = sysfs_read(-1, devnm, GET_VERSION);
761
762                         pr_err("Creating array inside %s container %s\n",
763                                 mdi?mdi->text_version:"managed", devnm);
764                         sysfs_free(mdi);
765                 } else
766                         pr_err("Defaulting to version %s metadata\n", info.text_version);
767         }
768
769         map_update(&map, fd2devnm(mdfd), info.text_version,
770                    info.uuid, chosen_name);
771         /* Keep map locked until devices have been added to array
772          * to stop another mdadm from finding and using those devices.
773          */
774
775         if (s->bitmap_file && (strcmp(s->bitmap_file, "internal") == 0 ||
776                                strcmp(s->bitmap_file, "clustered") == 0)) {
777                 if (!st->ss->add_internal_bitmap) {
778                         pr_err("internal bitmaps not supported with %s metadata\n",
779                                 st->ss->name);
780                         goto abort_locked;
781                 }
782                 if (st->ss->add_internal_bitmap(st, &s->bitmap_chunk,
783                                                 c->delay, s->write_behind,
784                                                 bitmapsize, 1, major_num)) {
785                         pr_err("Given bitmap chunk size not supported.\n");
786                         goto abort_locked;
787                 }
788                 s->bitmap_file = NULL;
789         }
790
791         if (sysfs_init(&info, mdfd, NULL)) {
792                 pr_err("unable to initialize sysfs\n");
793                 goto abort_locked;
794         }
795
796         if (st->ss->external && st->container_devnm[0]) {
797                 /* member */
798
799                 /* When creating a member, we need to be careful
800                  * to negotiate with mdmon properly.
801                  * If it is already running, we cannot write to
802                  * the devices and must ask it to do that part.
803                  * If it isn't running, we write to the devices,
804                  * and then start it.
805                  * We hold an exclusive open on the container
806                  * device to make sure mdmon doesn't exit after
807                  * we checked that it is running.
808                  *
809                  * For now, fail if it is already running.
810                  */
811                 container_fd = open_dev_excl(st->container_devnm);
812                 if (container_fd < 0) {
813                         pr_err("Cannot get exclusive open on container - weird.\n");
814                         goto abort_locked;
815                 }
816                 if (mdmon_running(st->container_devnm)) {
817                         if (c->verbose)
818                                 pr_err("reusing mdmon for %s.\n",
819                                         st->container_devnm);
820                         st->update_tail = &st->updates;
821                 } else
822                         need_mdmon = 1;
823         }
824         rv = set_array_info(mdfd, st, &info);
825         if (rv) {
826                 pr_err("failed to set array info for %s: %s\n",
827                         mddev, strerror(errno));
828                 goto abort_locked;
829         }
830
831         if (s->bitmap_file) {
832                 int uuid[4];
833
834                 st->ss->uuid_from_super(st, uuid);
835                 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
836                                  c->delay, s->write_behind,
837                                  bitmapsize,
838                                  major_num)) {
839                         goto abort_locked;
840                 }
841                 bitmap_fd = open(s->bitmap_file, O_RDWR);
842                 if (bitmap_fd < 0) {
843                         pr_err("weird: %s cannot be opened\n",
844                                 s->bitmap_file);
845                         goto abort_locked;
846                 }
847                 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
848                         pr_err("Cannot set bitmap file for %s: %s\n",
849                                 mddev, strerror(errno));
850                         goto abort_locked;
851                 }
852         }
853
854         infos = xmalloc(sizeof(*infos) * total_slots);
855         enable_fds(total_slots);
856         for (pass = 1; pass <= 2; pass++) {
857                 struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */
858
859                 for (dnum = 0, raid_disk_num = 0, dv = devlist; dv;
860                      dv = (dv->next) ? (dv->next) : moved_disk, dnum++) {
861                         int fd;
862                         struct mdinfo *inf = &infos[dnum];
863
864                         if (dnum >= total_slots)
865                                 abort();
866                         if (dnum == insert_point) {
867                                 raid_disk_num += 1;
868                                 moved_disk = dv;
869                                 continue;
870                         }
871                         if (strcasecmp(dv->devname, "missing") == 0) {
872                                 raid_disk_num += 1;
873                                 continue;
874                         }
875                         if (have_container)
876                                 moved_disk = NULL;
877                         if (have_container && dnum < info.array.raid_disks - 1)
878                                 /* repeatedly use the container */
879                                 moved_disk = dv;
880
881                         switch(pass) {
882                         case 1:
883                                 *inf = info;
884
885                                 inf->disk.number = dnum;
886                                 inf->disk.raid_disk = raid_disk_num++;
887
888                                 if (dv->disposition == 'j') {
889                                         inf->disk.raid_disk = MD_DISK_ROLE_JOURNAL;
890                                         inf->disk.state = (1<<MD_DISK_JOURNAL);
891                                         raid_disk_num--;
892                                 } else if (inf->disk.raid_disk < s->raiddisks)
893                                         inf->disk.state = (1<<MD_DISK_ACTIVE) |
894                                                 (1<<MD_DISK_SYNC);
895                                 else
896                                         inf->disk.state = 0;
897
898                                 if (dv->writemostly == FlagSet)
899                                         inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY);
900                                 if (dv->failfast == FlagSet)
901                                         inf->disk.state |= (1<<MD_DISK_FAILFAST);
902
903                                 if (have_container)
904                                         fd = -1;
905                                 else {
906                                         if (st->ss->external &&
907                                             st->container_devnm[0])
908                                                 fd = open(dv->devname, O_RDWR);
909                                         else
910                                                 fd = open(dv->devname, O_RDWR|O_EXCL);
911
912                                         if (fd < 0) {
913                                                 pr_err("failed to open %s after earlier success - aborting\n",
914                                                         dv->devname);
915                                                 goto abort_locked;
916                                         }
917                                         if (!fstat_is_blkdev(fd, dv->devname, &rdev))
918                                                 return 1;
919                                         inf->disk.major = major(rdev);
920                                         inf->disk.minor = minor(rdev);
921                                 }
922                                 if (fd >= 0)
923                                         remove_partitions(fd);
924                                 if (st->ss->add_to_super(st, &inf->disk,
925                                                          fd, dv->devname,
926                                                          dv->data_offset)) {
927                                         ioctl(mdfd, STOP_ARRAY, NULL);
928                                         goto abort_locked;
929                                 }
930                                 st->ss->getinfo_super(st, inf, NULL);
931                                 safe_mode_delay = inf->safe_mode_delay;
932
933                                 if (have_container && c->verbose > 0)
934                                         pr_err("Using %s for device %d\n",
935                                                 map_dev(inf->disk.major,
936                                                         inf->disk.minor,
937                                                         0), dnum);
938
939                                 if (!have_container) {
940                                         /* getinfo_super might have lost these ... */
941                                         inf->disk.major = major(rdev);
942                                         inf->disk.minor = minor(rdev);
943                                 }
944                                 break;
945                         case 2:
946                                 inf->errors = 0;
947
948                                 rv = add_disk(mdfd, st, &info, inf);
949
950                                 if (rv) {
951                                         pr_err("ADD_NEW_DISK for %s failed: %s\n",
952                                                dv->devname, strerror(errno));
953                                         goto abort_locked;
954                                 }
955                                 break;
956                         }
957                         if (!have_container &&
958                             dv == moved_disk && dnum != insert_point) break;
959                 }
960                 if (pass == 1) {
961                         struct mdinfo info_new;
962                         struct map_ent *me = NULL;
963
964                         /* check to see if the uuid has changed due to these
965                          * metadata changes, and if so update the member array
966                          * and container uuid.  Note ->write_init_super clears
967                          * the subarray cursor such that ->getinfo_super once
968                          * again returns container info.
969                          */
970                         st->ss->getinfo_super(st, &info_new, NULL);
971                         if (st->ss->external && s->level != LEVEL_CONTAINER &&
972                             !same_uuid(info_new.uuid, info.uuid, 0)) {
973                                 map_update(&map, fd2devnm(mdfd),
974                                            info_new.text_version,
975                                            info_new.uuid, chosen_name);
976                                 me = map_by_devnm(&map, st->container_devnm);
977                         }
978
979                         if (st->ss->write_init_super(st)) {
980                                 st->ss->free_super(st);
981                                 goto abort_locked;
982                         }
983
984                         /* update parent container uuid */
985                         if (me) {
986                                 char *path = xstrdup(me->path);
987
988                                 st->ss->getinfo_super(st, &info_new, NULL);
989                                 map_update(&map, st->container_devnm,
990                                            info_new.text_version,
991                                            info_new.uuid, path);
992                                 free(path);
993                         }
994
995                         flush_metadata_updates(st);
996                         st->ss->free_super(st);
997                 }
998         }
999         map_unlock(&map);
1000         free(infos);
1001
1002         if (s->level == LEVEL_CONTAINER) {
1003                 /* No need to start.  But we should signal udev to
1004                  * create links */
1005                 sysfs_uevent(&info, "change");
1006                 if (c->verbose >= 0)
1007                         pr_err("container %s prepared.\n", mddev);
1008                 wait_for(chosen_name, mdfd);
1009         } else if (c->runstop == 1 || subdevs >= s->raiddisks) {
1010                 if (st->ss->external) {
1011                         int err;
1012                         switch(s->level) {
1013                         case LEVEL_LINEAR:
1014                         case LEVEL_MULTIPATH:
1015                         case 0:
1016                                 err = sysfs_set_str(&info, NULL, "array_state",
1017                                                     c->readonly
1018                                                     ? "readonly"
1019                                                     : "active");
1020                                 need_mdmon = 0;
1021                                 break;
1022                         default:
1023                                 err = sysfs_set_str(&info, NULL, "array_state",
1024                                                     "readonly");
1025                                 break;
1026                         }
1027                         sysfs_set_safemode(&info, safe_mode_delay);
1028                         if (err) {
1029                                 pr_err("failed to activate array.\n");
1030                                 ioctl(mdfd, STOP_ARRAY, NULL);
1031                                 goto abort;
1032                         }
1033                 } else if (c->readonly &&
1034                            sysfs_attribute_available(
1035                                    &info, NULL, "array_state")) {
1036                         if (sysfs_set_str(&info, NULL,
1037                                           "array_state", "readonly") < 0) {
1038                                 pr_err("Failed to start array: %s\n",
1039                                        strerror(errno));
1040                                 ioctl(mdfd, STOP_ARRAY, NULL);
1041                                 goto abort;
1042                         }
1043                 } else {
1044                         /* param is not actually used */
1045                         mdu_param_t param;
1046                         if (ioctl(mdfd, RUN_ARRAY, &param)) {
1047                                 pr_err("RUN_ARRAY failed: %s\n",
1048                                        strerror(errno));
1049                                 if (info.array.chunk_size & (info.array.chunk_size-1)) {
1050                                         cont_err("Problem may be that chunk size is not a power of 2\n");
1051                                 }
1052                                 ioctl(mdfd, STOP_ARRAY, NULL);
1053                                 goto abort;
1054                         }
1055                         /* if start_ro module parameter is set, array is
1056                          * auto-read-only, which is bad as the resync won't
1057                          * start.  So lets make it read-write now.
1058                          */
1059                         ioctl(mdfd, RESTART_ARRAY_RW, NULL);
1060                 }
1061                 if (c->verbose >= 0)
1062                         pr_err("array %s started.\n", mddev);
1063                 if (st->ss->external && st->container_devnm[0]) {
1064                         if (need_mdmon)
1065                                 start_mdmon(st->container_devnm);
1066
1067                         ping_monitor(st->container_devnm);
1068                         close(container_fd);
1069                 }
1070                 wait_for(chosen_name, mdfd);
1071         } else {
1072                 pr_err("not starting array - not enough devices.\n");
1073         }
1074         close(mdfd);
1075         /* Give udev a moment to process the Change event caused
1076          * by the close.
1077          */
1078         usleep(100*1000);
1079         udev_unblock();
1080         return 0;
1081
1082  abort:
1083         udev_unblock();
1084         map_lock(&map);
1085  abort_locked:
1086         map_remove(&map, fd2devnm(mdfd));
1087         map_unlock(&map);
1088
1089         if (mdfd >= 0)
1090                 close(mdfd);
1091         return 1;
1092 }