Create: add support for RAID0 layouts.
[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 0:
55                         layout = RAID0_ORIG_LAYOUT;
56                         break;
57                 case 10:
58                         layout = 0x102; /* near=2, far=1 */
59                         if (verbose > 0)
60                                 pr_err("layout defaults to n2\n");
61                         break;
62                 case 5:
63                 case 6:
64                         layout = map_name(r5layout, "default");
65                         if (verbose > 0)
66                                 pr_err("layout defaults to %s\n", map_num(r5layout, layout));
67                         break;
68                 case LEVEL_FAULTY:
69                         layout = map_name(faultylayout, "default");
70
71                         if (verbose > 0)
72                                 pr_err("layout defaults to %s\n", map_num(faultylayout, layout));
73                         break;
74                 }
75
76         return layout;
77 }
78
79 int Create(struct supertype *st, char *mddev,
80            char *name, int *uuid,
81            int subdevs, struct mddev_dev *devlist,
82            struct shape *s,
83            struct context *c, unsigned long long data_offset)
84 {
85         /*
86          * Create a new raid array.
87          *
88          * First check that necessary details are available
89          * (i.e. level, raid-disks)
90          *
91          * Then check each disk to see what might be on it
92          * and report anything interesting.
93          *
94          * If anything looks odd, and runstop not set,
95          * abort.
96          *
97          * SET_ARRAY_INFO and ADD_NEW_DISK, and
98          * if runstop==run, or raiddisks disks were used,
99          * RUN_ARRAY
100          */
101         int mdfd;
102         unsigned long long minsize = 0, maxsize = 0;
103         char *mindisc = NULL;
104         char *maxdisc = NULL;
105         int dnum, raid_disk_num;
106         struct mddev_dev *dv;
107         dev_t rdev;
108         int fail = 0, warn = 0;
109         int first_missing = subdevs * 2;
110         int second_missing = subdevs * 2;
111         int missing_disks = 0;
112         int insert_point = subdevs * 2; /* where to insert a missing drive */
113         int total_slots;
114         int pass;
115         int rv;
116         int bitmap_fd;
117         int have_container = 0;
118         int container_fd = -1;
119         int need_mdmon = 0;
120         unsigned long long bitmapsize;
121         struct mdinfo info, *infos;
122         int did_default = 0;
123         int do_default_layout = 0;
124         int do_default_chunk = 0;
125         unsigned long safe_mode_delay = 0;
126         char chosen_name[1024];
127         struct map_ent *map = NULL;
128         unsigned long long newsize;
129         mdu_array_info_t inf;
130
131         int major_num = BITMAP_MAJOR_HI;
132         if (s->bitmap_file && strcmp(s->bitmap_file, "clustered") == 0) {
133                 major_num = BITMAP_MAJOR_CLUSTERED;
134                 if (c->nodes <= 1) {
135                         pr_err("At least 2 nodes are needed for cluster-md\n");
136                         return 1;
137                 }
138         }
139
140         memset(&info, 0, sizeof(info));
141         if (s->level == UnSet && st && st->ss->default_geometry)
142                 st->ss->default_geometry(st, &s->level, NULL, NULL);
143         if (s->level == UnSet) {
144                 pr_err("a RAID level is needed to create an array.\n");
145                 return 1;
146         }
147         if (s->raiddisks < 4 && s->level == 6) {
148                 pr_err("at least 4 raid-devices needed for level 6\n");
149                 return 1;
150         }
151         if (s->raiddisks > 256 && s->level == 6) {
152                 pr_err("no more than 256 raid-devices supported for level 6\n");
153                 return 1;
154         }
155         if (s->raiddisks < 2 && s->level >= 4) {
156                 pr_err("at least 2 raid-devices needed for level 4 or 5\n");
157                 return 1;
158         }
159         if (s->level <= 0 && s->sparedisks) {
160                 pr_err("This level does not support spare devices\n");
161                 return 1;
162         }
163
164         if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) {
165                 /* If given a single device, it might be a container, and we can
166                  * extract a device list from there
167                  */
168                 int fd;
169
170                 memset(&inf, 0, sizeof(inf));
171                 fd = open(devlist->devname, O_RDONLY);
172                 if (fd >= 0 &&
173                     md_get_array_info(fd, &inf) == 0 && inf.raid_disks == 0) {
174                         /* yep, looks like a container */
175                         if (st) {
176                                 rv = st->ss->load_container(st, fd,
177                                                             devlist->devname);
178                                 if (rv == 0)
179                                         have_container = 1;
180                         } else {
181                                 st = super_by_fd(fd, NULL);
182                                 if (st && !(rv = st->ss->
183                                             load_container(st, fd,
184                                                            devlist->devname)))
185                                         have_container = 1;
186                                 else
187                                         st = NULL;
188                         }
189                         if (have_container) {
190                                 subdevs = s->raiddisks;
191                                 first_missing = subdevs * 2;
192                                 second_missing = subdevs * 2;
193                                 insert_point = subdevs * 2;
194                         }
195                 }
196                 if (fd >= 0)
197                         close(fd);
198         }
199         if (st && st->ss->external && s->sparedisks) {
200                 pr_err("This metadata type does not support spare disks at create time\n");
201                 return 1;
202         }
203         if (subdevs > s->raiddisks+s->sparedisks+s->journaldisks) {
204                 pr_err("You have listed more devices (%d) than are in the array(%d)!\n", subdevs, s->raiddisks+s->sparedisks);
205                 return 1;
206         }
207         if (!have_container && subdevs < s->raiddisks+s->sparedisks+s->journaldisks) {
208                 pr_err("You haven't given enough devices (real or missing) to create this array\n");
209                 return 1;
210         }
211         if (s->bitmap_file && s->level <= 0) {
212                 pr_err("bitmaps not meaningful with level %s\n",
213                         map_num(pers, s->level)?:"given");
214                 return 1;
215         }
216
217         /* now set some defaults */
218
219         if (s->layout == UnSet) {
220                 do_default_layout = 1;
221                 s->layout = default_layout(st, s->level, c->verbose);
222         }
223
224         if (s->level == 10)
225                 /* check layout fits in array*/
226                 if ((s->layout&255) * ((s->layout>>8)&255) > s->raiddisks) {
227                         pr_err("that layout requires at least %d devices\n",
228                                 (s->layout&255) * ((s->layout>>8)&255));
229                         return 1;
230                 }
231
232         switch(s->level) {
233         case 4:
234         case 5:
235         case 10:
236         case 6:
237         case 0:
238                 if (s->chunk == 0 || s->chunk == UnSet) {
239                         s->chunk = UnSet;
240                         do_default_chunk = 1;
241                         /* chunk will be set later */
242                 }
243                 break;
244         case LEVEL_LINEAR:
245                 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
246                 if (get_linux_version() < 2006016 && s->chunk == 0) {
247                         s->chunk = 64;
248                         if (c->verbose > 0)
249                                 pr_err("chunk size defaults to 64K\n");
250                 }
251                 break;
252         case 1:
253         case LEVEL_FAULTY:
254         case LEVEL_MULTIPATH:
255         case LEVEL_CONTAINER:
256                 if (s->chunk) {
257                         s->chunk = 0;
258                         if (c->verbose > 0)
259                                 pr_err("chunk size ignored for this level\n");
260                 }
261                 break;
262         default:
263                 pr_err("unknown level %d\n", s->level);
264                 return 1;
265         }
266
267         if (s->size == MAX_SIZE)
268                 /* use '0' to mean 'max' now... */
269                 s->size = 0;
270         if (s->size && s->chunk && s->chunk != UnSet)
271                 if (round_size_and_verify(&s->size, s->chunk))
272                         return 1;
273
274         newsize = s->size * 2;
275         if (st && ! st->ss->validate_geometry(st, s->level, s->layout, s->raiddisks,
276                                               &s->chunk, s->size*2,
277                                               data_offset, NULL,
278                                               &newsize, s->consistency_policy,
279                                               c->verbose >= 0))
280                 return 1;
281
282         if (s->chunk && s->chunk != UnSet) {
283                 newsize &= ~(unsigned long long)(s->chunk*2 - 1);
284                 if (do_default_chunk) {
285                         /* default chunk was just set */
286                         if (c->verbose > 0)
287                                 pr_err("chunk size defaults to %dK\n", s->chunk);
288                         if (round_size_and_verify(&s->size, s->chunk))
289                                 return 1;
290                         do_default_chunk = 0;
291                 }
292         }
293
294         if (s->size == 0) {
295                 s->size = newsize / 2;
296                 if (s->level == 1)
297                         /* If this is ever reshaped to RAID5, we will
298                          * need a chunksize.  So round it off a bit
299                          * now just to be safe
300                          */
301                         s->size &= ~(64ULL-1);
302
303                 if (s->size && c->verbose > 0)
304                         pr_err("setting size to %lluK\n", s->size);
305         }
306
307         /* now look at the subdevs */
308         info.array.active_disks = 0;
309         info.array.working_disks = 0;
310         dnum = 0;
311         for (dv = devlist; dv; dv = dv->next)
312                 if (data_offset == VARIABLE_OFFSET)
313                         dv->data_offset = INVALID_SECTORS;
314                 else
315                         dv->data_offset = data_offset;
316
317         for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) {
318                 char *dname = dv->devname;
319                 unsigned long long freesize;
320                 int dfd;
321                 char *doff;
322
323                 if (strcasecmp(dname, "missing") == 0) {
324                         if (first_missing > dnum)
325                                 first_missing = dnum;
326                         if (second_missing > dnum && dnum > first_missing)
327                                 second_missing = dnum;
328                         missing_disks ++;
329                         continue;
330                 }
331                 if (data_offset == VARIABLE_OFFSET) {
332                         doff = strchr(dname, ':');
333                         if (doff) {
334                                 *doff++ = 0;
335                                 dv->data_offset = parse_size(doff);
336                         } else
337                                 dv->data_offset = INVALID_SECTORS;
338                 } else
339                         dv->data_offset = data_offset;
340
341                 dfd = open(dname, O_RDONLY);
342                 if (dfd < 0) {
343                         pr_err("cannot open %s: %s\n",
344                                 dname, strerror(errno));
345                         exit(2);
346                 }
347                 if (!fstat_is_blkdev(dfd, dname, NULL)) {
348                         close(dfd);
349                         exit(2);
350                 }
351                 close(dfd);
352                 info.array.working_disks++;
353                 if (dnum < s->raiddisks && dv->disposition != 'j')
354                         info.array.active_disks++;
355                 if (st == NULL) {
356                         struct createinfo *ci = conf_get_create_info();
357                         if (ci)
358                                 st = ci->supertype;
359                 }
360                 if (st == NULL) {
361                         /* Need to choose a default metadata, which is different
362                          * depending on geometry of array.
363                          */
364                         int i;
365                         char *name = "default";
366                         for(i = 0; !st && superlist[i]; i++) {
367                                 st = superlist[i]->match_metadata_desc(name);
368                                 if (!st)
369                                         continue;
370                                 if (do_default_layout)
371                                         s->layout = default_layout(st, s->level, c->verbose);
372                                 switch (st->ss->validate_geometry(
373                                                 st, s->level, s->layout, s->raiddisks,
374                                                 &s->chunk, s->size*2,
375                                                 dv->data_offset, dname,
376                                                 &freesize, s->consistency_policy,
377                                                 c->verbose > 0)) {
378                                 case -1: /* Not valid, message printed, and not
379                                           * worth checking any further */
380                                         exit(2);
381                                         break;
382                                 case 0: /* Geometry not valid */
383                                         free(st);
384                                         st = NULL;
385                                         s->chunk = do_default_chunk ? UnSet : s->chunk;
386                                         break;
387                                 case 1: /* All happy */
388                                         break;
389                                 }
390                         }
391
392                         if (!st) {
393                                 int dfd = open(dname, O_RDONLY|O_EXCL);
394                                 if (dfd < 0) {
395                                         pr_err("cannot open %s: %s\n",
396                                                 dname, strerror(errno));
397                                         exit(2);
398                                 }
399                                 pr_err("device %s not suitable for any style of array\n",
400                                         dname);
401                                 exit(2);
402                         }
403                         if (st->ss != &super0 ||
404                             st->minor_version != 90)
405                                 did_default = 1;
406                 } else {
407                         if (do_default_layout)
408                                 s->layout = default_layout(st, s->level, 0);
409                         if (!st->ss->validate_geometry(st, s->level, s->layout,
410                                                        s->raiddisks,
411                                                        &s->chunk, s->size*2,
412                                                        dv->data_offset,
413                                                        dname, &freesize,
414                                                        s->consistency_policy,
415                                                        c->verbose >= 0)) {
416
417                                 pr_err("%s is not suitable for this array.\n",
418                                        dname);
419                                 fail = 1;
420                                 continue;
421                         }
422                 }
423
424                 if (dv->disposition == 'j')
425                         goto skip_size_check;  /* skip write journal for size check */
426
427                 freesize /= 2; /* convert to K */
428                 if (s->chunk && s->chunk != UnSet) {
429                         /* round to chunk size */
430                         freesize = freesize & ~(s->chunk-1);
431                         if (do_default_chunk) {
432                                 /* default chunk was just set */
433                                 if (c->verbose > 0)
434                                         pr_err("chunk size defaults to %dK\n", s->chunk);
435                                 if (round_size_and_verify(&s->size, s->chunk))
436                                         return 1;
437                                 do_default_chunk = 0;
438                         }
439                 }
440                 if (!freesize) {
441                         pr_err("no free space left on %s\n", dname);
442                         fail = 1;
443                         continue;
444                 }
445
446                 if (s->size && freesize < s->size) {
447                         pr_err("%s is smaller than given size. %lluK < %lluK + metadata\n",
448                                 dname, freesize, s->size);
449                         fail = 1;
450                         continue;
451                 }
452                 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
453                         maxdisc = dname;
454                         maxsize = freesize;
455                 }
456                 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
457                         mindisc = dname;
458                         minsize = freesize;
459                 }
460         skip_size_check:
461                 if (c->runstop != 1 || c->verbose >= 0) {
462                         int fd = open(dname, O_RDONLY);
463                         if (fd < 0) {
464                                 pr_err("Cannot open %s: %s\n",
465                                         dname, strerror(errno));
466                                 fail = 1;
467                                 continue;
468                         }
469                         warn |= check_ext2(fd, dname);
470                         warn |= check_reiser(fd, dname);
471                         warn |= check_raid(fd, dname);
472                         if (strcmp(st->ss->name, "1.x") == 0 &&
473                             st->minor_version >= 1)
474                                 /* metadata at front */
475                                 warn |= check_partitions(fd, dname, 0, 0);
476                         else if (s->level == 1 || s->level == LEVEL_CONTAINER ||
477                                  (s->level == 0 && s->raiddisks == 1))
478                                 /* partitions could be meaningful */
479                                 warn |= check_partitions(fd, dname, freesize*2, s->size*2);
480                         else
481                                 /* partitions cannot be meaningful */
482                                 warn |= check_partitions(fd, dname, 0, 0);
483                         if (strcmp(st->ss->name, "1.x") == 0 &&
484                             st->minor_version >= 1 &&
485                             did_default &&
486                             s->level == 1 &&
487                             (warn & 1024) == 0) {
488                                 warn |= 1024;
489                                 pr_err("Note: this array has metadata at the start and\n"
490                                         "    may not be suitable as a boot device.  If you plan to\n"
491                                         "    store '/boot' on this device please ensure that\n"
492                                         "    your boot-loader understands md/v1.x metadata, or use\n"
493                                         "    --metadata=0.90\n");
494                         }
495                         close(fd);
496                 }
497         }
498         if (missing_disks == dnum && !have_container) {
499                 pr_err("Subdevs can't be all missing\n");
500                 return 1;
501         }
502         if (s->raiddisks + s->sparedisks > st->max_devs) {
503                 pr_err("Too many devices: %s metadata only supports %d\n",
504                         st->ss->name, st->max_devs);
505                 return 1;
506         }
507         if (have_container)
508                 info.array.working_disks = s->raiddisks;
509         if (fail) {
510                 pr_err("create aborted\n");
511                 return 1;
512         }
513         if (s->size == 0) {
514                 if (mindisc == NULL && !have_container) {
515                         pr_err("no size and no drives given - aborting create.\n");
516                         return 1;
517                 }
518                 if (s->level > 0 || s->level == LEVEL_MULTIPATH ||
519                     s->level == LEVEL_FAULTY || st->ss->external) {
520                         /* size is meaningful */
521                         if (!st->ss->validate_geometry(st, s->level, s->layout,
522                                                        s->raiddisks,
523                                                        &s->chunk, minsize*2,
524                                                        data_offset,
525                                                        NULL, NULL,
526                                                        s->consistency_policy, 0)) {
527                                 pr_err("devices too large for RAID level %d\n", s->level);
528                                 return 1;
529                         }
530                         s->size = minsize;
531                         if (s->level == 1)
532                                 /* If this is ever reshaped to RAID5, we will
533                                  * need a chunksize.  So round it off a bit
534                                  * now just to be safe
535                                  */
536                                 s->size &= ~(64ULL-1);
537                         if (c->verbose > 0)
538                                 pr_err("size set to %lluK\n", s->size);
539                 }
540         }
541
542         if (!s->bitmap_file &&
543             s->level >= 1 &&
544             st->ss->add_internal_bitmap &&
545             (s->consistency_policy != CONSISTENCY_POLICY_RESYNC &&
546              s->consistency_policy != CONSISTENCY_POLICY_PPL) &&
547             (s->write_behind || s->size > 100*1024*1024ULL)) {
548                 if (c->verbose > 0)
549                         pr_err("automatically enabling write-intent bitmap on large array\n");
550                 s->bitmap_file = "internal";
551         }
552         if (s->bitmap_file && strcmp(s->bitmap_file, "none") == 0)
553                 s->bitmap_file = NULL;
554
555         if (s->consistency_policy == CONSISTENCY_POLICY_PPL &&
556             !st->ss->write_init_ppl) {
557                 pr_err("%s metadata does not support PPL\n", st->ss->name);
558                 return 1;
559         }
560
561         if (!have_container && s->level > 0 && ((maxsize-s->size)*100 > maxsize)) {
562                 if (c->runstop != 1 || c->verbose >= 0)
563                         pr_err("largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
564                                 maxdisc, s->size);
565                 warn = 1;
566         }
567
568         if (st->ss->detail_platform && st->ss->detail_platform(0, 1, NULL) != 0) {
569                 if (c->runstop != 1 || c->verbose >= 0)
570                         pr_err("%s unable to enumerate platform support\n"
571                                 "    array may not be compatible with hardware/firmware\n",
572                                 st->ss->name);
573                 warn = 1;
574         }
575         st->nodes = c->nodes;
576         st->cluster_name = c->homecluster;
577
578         if (warn) {
579                 if (c->runstop!= 1) {
580                         if (!ask("Continue creating array? ")) {
581                                 pr_err("create aborted.\n");
582                                 return 1;
583                         }
584                 } else {
585                         if (c->verbose > 0)
586                                 pr_err("creation continuing despite oddities due to --run\n");
587                 }
588         }
589
590         /* If this is raid4/5, we want to configure the last active slot
591          * as missing, so that a reconstruct happens (faster than re-parity)
592          * FIX: Can we do this for raid6 as well?
593          */
594         if (st->ss->external == 0 && s->assume_clean == 0 &&
595             c->force == 0 && first_missing >= s->raiddisks) {
596                 switch (s->level) {
597                 case 4:
598                 case 5:
599                         insert_point = s->raiddisks-1;
600                         s->sparedisks++;
601                         info.array.active_disks--;
602                         missing_disks++;
603                         break;
604                 default:
605                         break;
606                 }
607         }
608         /* For raid6, if creating with 1 missing drive, make a good drive
609          * into a spare, else the create will fail
610          */
611         if (s->assume_clean == 0 && c->force == 0 && first_missing < s->raiddisks &&
612             st->ss->external == 0 &&
613             second_missing >= s->raiddisks && s->level == 6) {
614                 insert_point = s->raiddisks - 1;
615                 if (insert_point == first_missing)
616                         insert_point--;
617                 s->sparedisks ++;
618                 info.array.active_disks--;
619                 missing_disks++;
620         }
621
622         if (s->level <= 0 && first_missing < subdevs * 2) {
623                 pr_err("This level does not support missing devices\n");
624                 return 1;
625         }
626
627         /* We need to create the device */
628         map_lock(&map);
629         mdfd = create_mddev(mddev, name, c->autof, LOCAL, chosen_name, 1);
630         if (mdfd < 0) {
631                 map_unlock(&map);
632                 return 1;
633         }
634         /* verify if chosen_name is not in use,
635          * it could be in conflict with already existing device
636          * e.g. container, array
637          */
638         if (strncmp(chosen_name, "/dev/md/", 8) == 0 &&
639             map_by_name(&map, chosen_name+8) != NULL) {
640                 pr_err("Array name %s is in use already.\n",
641                         chosen_name);
642                 close(mdfd);
643                 map_unlock(&map);
644                 udev_unblock();
645                 return 1;
646         }
647         mddev = chosen_name;
648
649         memset(&inf, 0, sizeof(inf));
650         md_get_array_info(mdfd, &inf);
651         if (inf.working_disks != 0) {
652                 pr_err("another array by this name is already running.\n");
653                 goto abort_locked;
654         }
655
656         /* Ok, lets try some ioctls */
657
658         info.array.level = s->level;
659         info.array.size = s->size;
660         info.array.raid_disks = s->raiddisks;
661         /* The kernel should *know* what md_minor we are dealing
662          * with, but it chooses to trust me instead. Sigh
663          */
664         info.array.md_minor = 0;
665         if (fstat_is_blkdev(mdfd, mddev, &rdev))
666                 info.array.md_minor = minor(rdev);
667         info.array.not_persistent = 0;
668
669         if (((s->level == 4 || s->level == 5) &&
670              (insert_point < s->raiddisks || first_missing < s->raiddisks)) ||
671             (s->level == 6 && (insert_point < s->raiddisks ||
672                                second_missing < s->raiddisks)) ||
673             (s->level <= 0) || s->assume_clean) {
674                 info.array.state = 1; /* clean, but one+ drive will be missing*/
675                 info.resync_start = MaxSector;
676         } else {
677                 info.array.state = 0; /* not clean, but no errors */
678                 info.resync_start = 0;
679         }
680         if (s->level == 10) {
681                 /* for raid10, the bitmap size is the capacity of the array,
682                  * which is array.size * raid_disks / ncopies;
683                  * .. but convert to sectors.
684                  */
685                 int ncopies = ((s->layout>>8) & 255) * (s->layout & 255);
686                 bitmapsize = s->size * s->raiddisks / ncopies * 2;
687 /*              printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, s->size, s->raiddisks, ncopies);*/
688         } else
689                 bitmapsize = s->size * 2;
690
691         /* There is lots of redundancy in these disk counts,
692          * raid_disks is the most meaningful value
693          *          it describes the geometry of the array
694          *          it is constant
695          * nr_disks is total number of used slots.
696          *          it should be raid_disks+spare_disks
697          * spare_disks is the number of extra disks present
698          *          see above
699          * active_disks is the number of working disks in
700          *          active slots. (With raid_disks)
701          * working_disks is the total number of working disks,
702          *          including spares
703          * failed_disks is the number of disks marked failed
704          *
705          * Ideally, the kernel would keep these (except raid_disks)
706          * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
707          * So for now, we assume that all raid and spare
708          * devices will be given.
709          */
710         info.array.spare_disks=s->sparedisks;
711         info.array.failed_disks=missing_disks;
712         info.array.nr_disks = info.array.working_disks
713                 + info.array.failed_disks;
714         info.array.layout = s->layout;
715         info.array.chunk_size = s->chunk*1024;
716
717         if (name == NULL || *name == 0) {
718                 /* base name on mddev */
719                 /*  /dev/md0 -> 0
720                  *  /dev/md_d0 -> d0
721                  *  /dev/md_foo -> foo
722                  *  /dev/md/1 -> 1
723                  *  /dev/md/d1 -> d1
724                  *  /dev/md/home -> home
725                  *  /dev/mdhome -> home
726                  */
727                 /* FIXME compare this with rules in create_mddev */
728                 name = strrchr(mddev, '/');
729                 if (name) {
730                         name++;
731                         if (strncmp(name, "md_", 3) == 0 &&
732                             strlen(name) > 3 && (name-mddev) == 5 /* /dev/ */)
733                                 name += 3;
734                         else if (strncmp(name, "md", 2) == 0 &&
735                                  strlen(name) > 2 && isdigit(name[2]) &&
736                                  (name-mddev) == 5 /* /dev/ */)
737                                 name += 2;
738                 }
739         }
740         if (!st->ss->init_super(st, &info.array, s, name, c->homehost, uuid,
741                                 data_offset))
742                 goto abort_locked;
743
744         total_slots = info.array.nr_disks;
745         st->ss->getinfo_super(st, &info, NULL);
746         if (sysfs_init(&info, mdfd, NULL)) {
747                 pr_err("unable to initialize sysfs\n");
748                 goto abort_locked;
749         }
750
751         if (did_default && c->verbose >= 0) {
752                 if (is_subarray(info.text_version)) {
753                         char devnm[32];
754                         char *ep;
755                         struct mdinfo *mdi;
756
757                         strncpy(devnm, info.text_version+1, 32);
758                         devnm[31] = 0;
759                         ep = strchr(devnm, '/');
760                         if (ep)
761                                 *ep = 0;
762
763                         mdi = sysfs_read(-1, devnm, GET_VERSION);
764
765                         pr_err("Creating array inside %s container %s\n",
766                                 mdi?mdi->text_version:"managed", devnm);
767                         sysfs_free(mdi);
768                 } else
769                         pr_err("Defaulting to version %s metadata\n", info.text_version);
770         }
771
772         map_update(&map, fd2devnm(mdfd), info.text_version,
773                    info.uuid, chosen_name);
774         /* Keep map locked until devices have been added to array
775          * to stop another mdadm from finding and using those devices.
776          */
777
778         if (s->bitmap_file && (strcmp(s->bitmap_file, "internal") == 0 ||
779                                strcmp(s->bitmap_file, "clustered") == 0)) {
780                 if (!st->ss->add_internal_bitmap) {
781                         pr_err("internal bitmaps not supported with %s metadata\n",
782                                 st->ss->name);
783                         goto abort_locked;
784                 }
785                 if (st->ss->add_internal_bitmap(st, &s->bitmap_chunk,
786                                                 c->delay, s->write_behind,
787                                                 bitmapsize, 1, major_num)) {
788                         pr_err("Given bitmap chunk size not supported.\n");
789                         goto abort_locked;
790                 }
791                 s->bitmap_file = NULL;
792         }
793
794         if (sysfs_init(&info, mdfd, NULL)) {
795                 pr_err("unable to initialize sysfs\n");
796                 goto abort_locked;
797         }
798
799         if (st->ss->external && st->container_devnm[0]) {
800                 /* member */
801
802                 /* When creating a member, we need to be careful
803                  * to negotiate with mdmon properly.
804                  * If it is already running, we cannot write to
805                  * the devices and must ask it to do that part.
806                  * If it isn't running, we write to the devices,
807                  * and then start it.
808                  * We hold an exclusive open on the container
809                  * device to make sure mdmon doesn't exit after
810                  * we checked that it is running.
811                  *
812                  * For now, fail if it is already running.
813                  */
814                 container_fd = open_dev_excl(st->container_devnm);
815                 if (container_fd < 0) {
816                         pr_err("Cannot get exclusive open on container - weird.\n");
817                         goto abort_locked;
818                 }
819                 if (mdmon_running(st->container_devnm)) {
820                         if (c->verbose)
821                                 pr_err("reusing mdmon for %s.\n",
822                                         st->container_devnm);
823                         st->update_tail = &st->updates;
824                 } else
825                         need_mdmon = 1;
826         }
827         rv = set_array_info(mdfd, st, &info);
828         if (rv) {
829                 pr_err("failed to set array info for %s: %s\n",
830                         mddev, strerror(errno));
831                 goto abort_locked;
832         }
833
834         if (s->bitmap_file) {
835                 int uuid[4];
836
837                 st->ss->uuid_from_super(st, uuid);
838                 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
839                                  c->delay, s->write_behind,
840                                  bitmapsize,
841                                  major_num)) {
842                         goto abort_locked;
843                 }
844                 bitmap_fd = open(s->bitmap_file, O_RDWR);
845                 if (bitmap_fd < 0) {
846                         pr_err("weird: %s cannot be opened\n",
847                                 s->bitmap_file);
848                         goto abort_locked;
849                 }
850                 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
851                         pr_err("Cannot set bitmap file for %s: %s\n",
852                                 mddev, strerror(errno));
853                         goto abort_locked;
854                 }
855         }
856
857         infos = xmalloc(sizeof(*infos) * total_slots);
858         enable_fds(total_slots);
859         for (pass = 1; pass <= 2; pass++) {
860                 struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */
861
862                 for (dnum = 0, raid_disk_num = 0, dv = devlist; dv;
863                      dv = (dv->next) ? (dv->next) : moved_disk, dnum++) {
864                         int fd;
865                         struct mdinfo *inf = &infos[dnum];
866
867                         if (dnum >= total_slots)
868                                 abort();
869                         if (dnum == insert_point) {
870                                 raid_disk_num += 1;
871                                 moved_disk = dv;
872                                 continue;
873                         }
874                         if (strcasecmp(dv->devname, "missing") == 0) {
875                                 raid_disk_num += 1;
876                                 continue;
877                         }
878                         if (have_container)
879                                 moved_disk = NULL;
880                         if (have_container && dnum < info.array.raid_disks - 1)
881                                 /* repeatedly use the container */
882                                 moved_disk = dv;
883
884                         switch(pass) {
885                         case 1:
886                                 *inf = info;
887
888                                 inf->disk.number = dnum;
889                                 inf->disk.raid_disk = raid_disk_num++;
890
891                                 if (dv->disposition == 'j') {
892                                         inf->disk.raid_disk = MD_DISK_ROLE_JOURNAL;
893                                         inf->disk.state = (1<<MD_DISK_JOURNAL);
894                                         raid_disk_num--;
895                                 } else if (inf->disk.raid_disk < s->raiddisks)
896                                         inf->disk.state = (1<<MD_DISK_ACTIVE) |
897                                                 (1<<MD_DISK_SYNC);
898                                 else
899                                         inf->disk.state = 0;
900
901                                 if (dv->writemostly == FlagSet)
902                                         inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY);
903                                 if (dv->failfast == FlagSet)
904                                         inf->disk.state |= (1<<MD_DISK_FAILFAST);
905
906                                 if (have_container)
907                                         fd = -1;
908                                 else {
909                                         if (st->ss->external &&
910                                             st->container_devnm[0])
911                                                 fd = open(dv->devname, O_RDWR);
912                                         else
913                                                 fd = open(dv->devname, O_RDWR|O_EXCL);
914
915                                         if (fd < 0) {
916                                                 pr_err("failed to open %s after earlier success - aborting\n",
917                                                         dv->devname);
918                                                 goto abort_locked;
919                                         }
920                                         if (!fstat_is_blkdev(fd, dv->devname, &rdev))
921                                                 return 1;
922                                         inf->disk.major = major(rdev);
923                                         inf->disk.minor = minor(rdev);
924                                 }
925                                 if (fd >= 0)
926                                         remove_partitions(fd);
927                                 if (st->ss->add_to_super(st, &inf->disk,
928                                                          fd, dv->devname,
929                                                          dv->data_offset)) {
930                                         ioctl(mdfd, STOP_ARRAY, NULL);
931                                         goto abort_locked;
932                                 }
933                                 st->ss->getinfo_super(st, inf, NULL);
934                                 safe_mode_delay = inf->safe_mode_delay;
935
936                                 if (have_container && c->verbose > 0)
937                                         pr_err("Using %s for device %d\n",
938                                                 map_dev(inf->disk.major,
939                                                         inf->disk.minor,
940                                                         0), dnum);
941
942                                 if (!have_container) {
943                                         /* getinfo_super might have lost these ... */
944                                         inf->disk.major = major(rdev);
945                                         inf->disk.minor = minor(rdev);
946                                 }
947                                 break;
948                         case 2:
949                                 inf->errors = 0;
950
951                                 rv = add_disk(mdfd, st, &info, inf);
952
953                                 if (rv) {
954                                         pr_err("ADD_NEW_DISK for %s failed: %s\n",
955                                                dv->devname, strerror(errno));
956                                         if (errno == EINVAL &&
957                                             info.array.level == 0) {
958                                                 pr_err("Possibly your kernel doesn't support RAID0 layouts.\n");
959                                                 pr_err("Either upgrade, or use --layout=dangerous\n");
960                                         }
961                                         goto abort_locked;
962                                 }
963                                 break;
964                         }
965                         if (!have_container &&
966                             dv == moved_disk && dnum != insert_point) break;
967                 }
968                 if (pass == 1) {
969                         struct mdinfo info_new;
970                         struct map_ent *me = NULL;
971
972                         /* check to see if the uuid has changed due to these
973                          * metadata changes, and if so update the member array
974                          * and container uuid.  Note ->write_init_super clears
975                          * the subarray cursor such that ->getinfo_super once
976                          * again returns container info.
977                          */
978                         st->ss->getinfo_super(st, &info_new, NULL);
979                         if (st->ss->external && s->level != LEVEL_CONTAINER &&
980                             !same_uuid(info_new.uuid, info.uuid, 0)) {
981                                 map_update(&map, fd2devnm(mdfd),
982                                            info_new.text_version,
983                                            info_new.uuid, chosen_name);
984                                 me = map_by_devnm(&map, st->container_devnm);
985                         }
986
987                         if (st->ss->write_init_super(st)) {
988                                 st->ss->free_super(st);
989                                 goto abort_locked;
990                         }
991
992                         /* update parent container uuid */
993                         if (me) {
994                                 char *path = xstrdup(me->path);
995
996                                 st->ss->getinfo_super(st, &info_new, NULL);
997                                 map_update(&map, st->container_devnm,
998                                            info_new.text_version,
999                                            info_new.uuid, path);
1000                                 free(path);
1001                         }
1002
1003                         flush_metadata_updates(st);
1004                         st->ss->free_super(st);
1005                 }
1006         }
1007         map_unlock(&map);
1008         free(infos);
1009
1010         if (s->level == LEVEL_CONTAINER) {
1011                 /* No need to start.  But we should signal udev to
1012                  * create links */
1013                 sysfs_uevent(&info, "change");
1014                 if (c->verbose >= 0)
1015                         pr_err("container %s prepared.\n", mddev);
1016                 wait_for(chosen_name, mdfd);
1017         } else if (c->runstop == 1 || subdevs >= s->raiddisks) {
1018                 if (st->ss->external) {
1019                         int err;
1020                         switch(s->level) {
1021                         case LEVEL_LINEAR:
1022                         case LEVEL_MULTIPATH:
1023                         case 0:
1024                                 err = sysfs_set_str(&info, NULL, "array_state",
1025                                                     c->readonly
1026                                                     ? "readonly"
1027                                                     : "active");
1028                                 need_mdmon = 0;
1029                                 break;
1030                         default:
1031                                 err = sysfs_set_str(&info, NULL, "array_state",
1032                                                     "readonly");
1033                                 break;
1034                         }
1035                         sysfs_set_safemode(&info, safe_mode_delay);
1036                         if (err) {
1037                                 pr_err("failed to activate array.\n");
1038                                 ioctl(mdfd, STOP_ARRAY, NULL);
1039                                 goto abort;
1040                         }
1041                 } else if (c->readonly &&
1042                            sysfs_attribute_available(
1043                                    &info, NULL, "array_state")) {
1044                         if (sysfs_set_str(&info, NULL,
1045                                           "array_state", "readonly") < 0) {
1046                                 pr_err("Failed to start array: %s\n",
1047                                        strerror(errno));
1048                                 ioctl(mdfd, STOP_ARRAY, NULL);
1049                                 goto abort;
1050                         }
1051                 } else {
1052                         /* param is not actually used */
1053                         mdu_param_t param;
1054                         if (ioctl(mdfd, RUN_ARRAY, &param)) {
1055                                 pr_err("RUN_ARRAY failed: %s\n",
1056                                        strerror(errno));
1057                                 if (errno == 524 /* ENOTSUP */ &&
1058                                     info.array.level == 0)
1059                                         cont_err("Please use --layout=original or --layout=alternate\n");
1060                                 if (info.array.chunk_size & (info.array.chunk_size-1)) {
1061                                         cont_err("Problem may be that chunk size is not a power of 2\n");
1062                                 }
1063                                 ioctl(mdfd, STOP_ARRAY, NULL);
1064                                 goto abort;
1065                         }
1066                         /* if start_ro module parameter is set, array is
1067                          * auto-read-only, which is bad as the resync won't
1068                          * start.  So lets make it read-write now.
1069                          */
1070                         ioctl(mdfd, RESTART_ARRAY_RW, NULL);
1071                 }
1072                 if (c->verbose >= 0)
1073                         pr_err("array %s started.\n", mddev);
1074                 if (st->ss->external && st->container_devnm[0]) {
1075                         if (need_mdmon)
1076                                 start_mdmon(st->container_devnm);
1077
1078                         ping_monitor(st->container_devnm);
1079                         close(container_fd);
1080                 }
1081                 wait_for(chosen_name, mdfd);
1082         } else {
1083                 pr_err("not starting array - not enough devices.\n");
1084         }
1085         close(mdfd);
1086         /* Give udev a moment to process the Change event caused
1087          * by the close.
1088          */
1089         usleep(100*1000);
1090         udev_unblock();
1091         return 0;
1092
1093  abort:
1094         udev_unblock();
1095         map_lock(&map);
1096  abort_locked:
1097         map_remove(&map, fd2devnm(mdfd));
1098         map_unlock(&map);
1099
1100         if (mdfd >= 0)
1101                 close(mdfd);
1102         return 1;
1103 }