Improve reporting of layout for raid10.
[thirdparty/mdadm.git] / super0.c
1 /*
2  * mdadm - manage Linux "md" devices aka RAID arrays.
3  *
4  * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de>
5  *
6  *
7  *    This program is free software; you can redistribute it and/or modify
8  *    it under the terms of the GNU General Public License as published by
9  *    the Free Software Foundation; either version 2 of the License, or
10  *    (at your option) any later version.
11  *
12  *    This program is distributed in the hope that it will be useful,
13  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
14  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  *    GNU General Public License for more details.
16  *
17  *    You should have received a copy of the GNU General Public License
18  *    along with this program; if not, write to the Free Software
19  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
20  *
21  *    Author: Neil Brown
22  *    Email: <neilb@cse.unsw.edu.au>
23  *    Paper: Neil Brown
24  *           School of Computer Science and Engineering
25  *           The University of New South Wales
26  *           Sydney, 2052
27  *           Australia
28  */
29
30 #define HAVE_STDINT_H 1
31 #include "mdadm.h"
32 #include "sha1.h"
33 /*
34  * All handling for the 0.90.0 version superblock is in
35  * this file.
36  * This includes:
37  *   - finding, loading, and writing the superblock.
38  *   - initialising a new superblock
39  *   - printing the superblock for --examine
40  *   - printing part of the superblock for --detail
41  * .. other stuff
42  */
43
44
45 static unsigned long calc_sb0_csum(mdp_super_t *super)
46 {
47         unsigned long csum = super->sb_csum;
48         unsigned long newcsum;
49         super->sb_csum= 0 ;
50         newcsum = calc_csum(super, MD_SB_BYTES);
51         super->sb_csum = csum;
52         return newcsum;
53 }
54
55
56 void super0_swap_endian(struct mdp_superblock_s *sb)
57 {
58         /* as super0 superblocks are host-endian, it is sometimes
59          * useful to be able to swap the endianness
60          * as (almost) everything is u32's we byte-swap every 4byte
61          * number.
62          * We then also have to swap the events_hi and events_lo
63          */
64         char *sbc = (char *)sb;
65         __u32 t32;
66         int i;
67
68         for (i=0; i < MD_SB_BYTES ; i+=4) {
69                 char t = sbc[i];
70                 sbc[i] = sbc[i+3];
71                 sbc[i+3] = t;
72                 t=sbc[i+1];
73                 sbc[i+1]=sbc[i+2];
74                 sbc[i+2]=t;
75         }
76         t32 = sb->events_hi;
77         sb->events_hi = sb->events_lo;
78         sb->events_lo = t32;
79
80         t32 = sb->cp_events_hi;
81         sb->cp_events_hi = sb->cp_events_lo;
82         sb->cp_events_lo = t32;
83
84 }
85
86 #ifndef MDASSEMBLE
87
88 static void examine_super0(struct supertype *st, char *homehost)
89 {
90         mdp_super_t *sb = st->sb;
91         time_t atime;
92         int d;
93         char *c;
94
95         printf("          Magic : %08x\n", sb->md_magic);
96         printf("        Version : %02d.%02d.%02d\n", sb->major_version, sb->minor_version,
97                sb->patch_version);
98         if (sb->minor_version >= 90) {
99                 printf("           UUID : %08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
100                        sb->set_uuid2, sb->set_uuid3);
101                 if (homehost) {
102                         char buf[20];
103                         void *hash = sha1_buffer(homehost,
104                                                  strlen(homehost),
105                                                  buf);
106                         if (memcmp(&sb->set_uuid2, hash, 8)==0)
107                                 printf(" (local to host %s)", homehost);
108                 }
109                 printf("\n");
110         } else
111                 printf("           UUID : %08x\n", sb->set_uuid0);
112
113         if (sb->not_persistent)
114                 printf("           Eedk : not persistent\n");
115
116         atime = sb->ctime;
117         printf("  Creation Time : %.24s\n", ctime(&atime));
118         c=map_num(pers, sb->level);
119         printf("     Raid Level : %s\n", c?c:"-unknown-");
120         if ((int)sb->level >= 0) {
121                 int ddsks=0;
122                 printf("  Used Dev Size : %d%s\n", sb->size,
123                        human_size((long long)sb->size<<10));
124                 switch(sb->level) {
125                 case 1: ddsks=1;break;
126                 case 4:
127                 case 5: ddsks = sb->raid_disks-1; break;
128                 case 6: ddsks = sb->raid_disks-2; break;
129                 case 10: ddsks = sb->raid_disks / (sb->layout&255) / ((sb->layout>>8)&255);
130                 }
131                 if (ddsks)
132                         printf("     Array Size : %llu%s\n", (unsigned long long)ddsks * sb->size,
133                                human_size(ddsks*(long long)sb->size<<10));
134         }
135         printf("   Raid Devices : %d\n", sb->raid_disks);
136         printf("  Total Devices : %d\n", sb->nr_disks);
137         printf("Preferred Minor : %d\n", sb->md_minor);
138         printf("\n");
139         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
140                 printf("  Reshape pos'n : %llu%s\n", (unsigned long long)sb->reshape_position/2, human_size((long long)sb->reshape_position<<9));
141                 if (sb->delta_disks) {
142                         printf("  Delta Devices : %d", sb->delta_disks);
143                         if (sb->delta_disks)
144                                 printf(" (%d->%d)\n", sb->raid_disks-sb->delta_disks, sb->raid_disks);
145                         else
146                                 printf(" (%d->%d)\n", sb->raid_disks, sb->raid_disks+sb->delta_disks);
147                 }
148                 if (sb->new_level != sb->level) {
149                         c = map_num(pers, sb->new_level);
150                         printf("      New Level : %s\n", c?c:"-unknown-");
151                 }
152                 if (sb->new_layout != sb->layout) {
153                         if (sb->level == 5) {
154                                 c = map_num(r5layout, sb->new_layout);
155                                 printf("     New Layout : %s\n", c?c:"-unknown-");
156                         }
157                         if (sb->level == 10) {
158                                 printf("     New Layout : near=%d, %s=%d\n",
159                                        sb->new_layout&255,
160                                        (sb->new_layout&0x10000)?"offset":"far",
161                                        (sb->new_layout>>8)&255);
162                         }
163                 }
164                 if (sb->new_chunk != sb->chunk_size)
165                         printf("  New Chunksize : %d\n", sb->new_chunk);
166                 printf("\n");
167         }
168         atime = sb->utime;
169         printf("    Update Time : %.24s\n", ctime(&atime));
170         printf("          State : %s\n",
171                (sb->state&(1<<MD_SB_CLEAN))?"clean":"active");
172         if (sb->state & (1<<MD_SB_BITMAP_PRESENT))
173                 printf("Internal Bitmap : present\n");
174         printf(" Active Devices : %d\n", sb->active_disks);
175         printf("Working Devices : %d\n", sb->working_disks);
176         printf(" Failed Devices : %d\n", sb->failed_disks);
177         printf("  Spare Devices : %d\n", sb->spare_disks);
178         if (calc_sb0_csum(sb) == sb->sb_csum)
179                 printf("       Checksum : %x - correct\n", sb->sb_csum);
180         else
181                 printf("       Checksum : %x - expected %lx\n", sb->sb_csum, calc_sb0_csum(sb));
182         printf("         Events : %llu\n",
183                ((unsigned long long)sb->events_hi << 32)
184                + sb->events_lo);
185         printf("\n");
186         if (sb->level == 5) {
187                 c = map_num(r5layout, sb->layout);
188                 printf("         Layout : %s\n", c?c:"-unknown-");
189         }
190         if (sb->level == 10) {
191                 printf("         Layout :");
192                 print_r10_layout(sb->layout);
193                 printf("\n");
194         }
195         switch(sb->level) {
196         case 0:
197         case 4:
198         case 5:
199         case 6:
200         case 10:
201                 printf("     Chunk Size : %dK\n", sb->chunk_size/1024);
202                 break;
203         case -1:
204                 printf("       Rounding : %dK\n", sb->chunk_size/1024);
205                 break;
206         default: break;
207         }
208         printf("\n");
209         printf("      Number   Major   Minor   RaidDevice State\n");
210         for (d= -1; d<(signed int)(sb->raid_disks+sb->spare_disks); d++) {
211                 mdp_disk_t *dp;
212                 char *dv;
213                 char nb[5];
214                 int wonly;
215                 if (d>=0) dp = &sb->disks[d];
216                 else dp = &sb->this_disk;
217                 snprintf(nb, sizeof(nb), "%4d", d);
218                 printf("%4s %5d   %5d    %5d    %5d     ", d < 0 ? "this" :  nb,
219                        dp->number, dp->major, dp->minor, dp->raid_disk);
220                 wonly = dp->state & (1<<MD_DISK_WRITEMOSTLY);
221                 dp->state &= ~(1<<MD_DISK_WRITEMOSTLY);
222                 if (dp->state & (1<<MD_DISK_FAULTY)) printf(" faulty");
223                 if (dp->state & (1<<MD_DISK_ACTIVE)) printf(" active");
224                 if (dp->state & (1<<MD_DISK_SYNC)) printf(" sync");
225                 if (dp->state & (1<<MD_DISK_REMOVED)) printf(" removed");
226                 if (wonly) printf(" write-mostly");
227                 if (dp->state == 0) printf(" spare");
228                 if ((dv=map_dev(dp->major, dp->minor, 0)))
229                         printf("   %s", dv);
230                 printf("\n");
231                 if (d == -1) printf("\n");
232         }
233 }
234
235 static void brief_examine_super0(struct supertype *st)
236 {
237         mdp_super_t *sb = st->sb;
238         char *c=map_num(pers, sb->level);
239         char devname[20];
240
241         sprintf(devname, "/dev/md%d", sb->md_minor);
242
243         printf("ARRAY %s level=%s num-devices=%d UUID=",
244                devname,
245                c?c:"-unknown-", sb->raid_disks);
246         if (sb->minor_version >= 90)
247                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
248                        sb->set_uuid2, sb->set_uuid3);
249         else
250                 printf("%08x", sb->set_uuid0);
251         printf("\n");
252 }
253
254 static void export_examine_super0(struct supertype *st)
255 {
256         mdp_super_t *sb = st->sb;
257
258         printf("MD_LEVEL=%s\n", map_num(pers, sb->level));
259         printf("MD_DEVICES=%d\n", sb->raid_disks);
260         if (sb->minor_version >= 90)
261                 printf("MD_UUID=%08x:%08x:%08x:%08x\n",
262                        sb->set_uuid0, sb->set_uuid1,
263                        sb->set_uuid2, sb->set_uuid3);
264         else
265                 printf("MD_UUID=%08x\n", sb->set_uuid0);
266         printf("MD_UPDATE_TIME=%llu\n",
267                __le64_to_cpu(sb->ctime) & 0xFFFFFFFFFFULL);
268         printf("MD_EVENTS=%llu\n",
269                ((unsigned long long)sb->events_hi << 32)
270                + sb->events_lo);
271 }
272
273 static void detail_super0(struct supertype *st, char *homehost)
274 {
275         mdp_super_t *sb = st->sb;
276         printf("           UUID : ");
277         if (sb->minor_version >= 90)
278                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
279                        sb->set_uuid2, sb->set_uuid3);
280         else
281                 printf("%08x", sb->set_uuid0);
282         if (homehost) {
283                 char buf[20];
284                 void *hash = sha1_buffer(homehost,
285                                          strlen(homehost),
286                                          buf);
287                 if (memcmp(&sb->set_uuid2, hash, 8)==0)
288                         printf(" (local to host %s)", homehost);
289         }
290         printf("\n         Events : %d.%d\n\n", sb->events_hi, sb->events_lo);
291 }
292
293 static void brief_detail_super0(struct supertype *st)
294 {
295         mdp_super_t *sb = st->sb;
296         printf(" UUID=");
297         if (sb->minor_version >= 90)
298                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
299                        sb->set_uuid2, sb->set_uuid3);
300         else
301                 printf("%08x", sb->set_uuid0);
302 }
303
304 static void export_detail_super0(struct supertype *st)
305 {
306         mdp_super_t *sb = st->sb;
307         printf("MD_UUID=");
308         if (sb->minor_version >= 90)
309                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
310                        sb->set_uuid2, sb->set_uuid3);
311         else
312                 printf("%08x", sb->set_uuid0);
313         printf("\n");
314 }
315 #endif
316
317 static int match_home0(struct supertype *st, char *homehost)
318 {
319         mdp_super_t *sb = st->sb;
320         char buf[20];
321         char *hash = sha1_buffer(homehost,
322                                  strlen(homehost),
323                                  buf);
324
325         return (memcmp(&sb->set_uuid2, hash, 8)==0);
326 }
327
328 static void uuid_from_super0(struct supertype *st, int uuid[4])
329 {
330         mdp_super_t *super = st->sb;
331         uuid[0] = super->set_uuid0;
332         if (super->minor_version >= 90) {
333                 uuid[1] = super->set_uuid1;
334                 uuid[2] = super->set_uuid2;
335                 uuid[3] = super->set_uuid3;
336         } else {
337                 uuid[1] = 0;
338                 uuid[2] = 0;
339                 uuid[3] = 0;
340         }
341 }
342
343 static void getinfo_super0(struct supertype *st, struct mdinfo *info)
344 {
345         mdp_super_t *sb = st->sb;
346         int working = 0;
347         int i;
348
349         info->array.major_version = sb->major_version;
350         info->array.minor_version = sb->minor_version;
351         info->array.patch_version = sb->patch_version;
352         info->array.raid_disks = sb->raid_disks;
353         info->array.level = sb->level;
354         info->array.layout = sb->layout;
355         info->array.md_minor = sb->md_minor;
356         info->array.ctime = sb->ctime;
357         info->array.utime = sb->utime;
358         info->array.chunk_size = sb->chunk_size;
359         info->array.state = sb->state;
360         info->component_size = sb->size*2;
361
362         info->disk.state = sb->this_disk.state;
363         info->disk.major = sb->this_disk.major;
364         info->disk.minor = sb->this_disk.minor;
365         info->disk.raid_disk = sb->this_disk.raid_disk;
366         info->disk.number = sb->this_disk.number;
367
368         info->events = md_event(sb);
369         info->data_offset = 0;
370
371         uuid_from_super0(st, info->uuid);
372
373         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
374                 info->reshape_active = 1;
375                 info->reshape_progress = sb->reshape_position;
376                 info->new_level = sb->new_level;
377                 info->delta_disks = sb->delta_disks;
378                 info->new_layout = sb->new_layout;
379                 info->new_chunk = sb->new_chunk;
380         } else
381                 info->reshape_active = 0;
382
383         sprintf(info->name, "%d", sb->md_minor);
384         /* work_disks is calculated rather than read directly */
385         for (i=0; i < MD_SB_DISKS; i++)
386                 if ((sb->disks[i].state & (1<<MD_DISK_SYNC)) &&
387                     (sb->disks[i].raid_disk < info->array.raid_disks) &&
388                     (sb->disks[i].state & (1<<MD_DISK_ACTIVE)) &&
389                     !(sb->disks[i].state & (1<<MD_DISK_FAULTY)))
390                         working ++;
391         info->array.working_disks = working;
392 }
393
394
395 static int update_super0(struct supertype *st, struct mdinfo *info,
396                          char *update,
397                          char *devname, int verbose,
398                          int uuid_set, char *homehost)
399 {
400         /* NOTE: for 'assemble' and 'force' we need to return non-zero if any change was made.
401          * For others, the return value is ignored.
402          */
403         int rv = 0;
404         mdp_super_t *sb = st->sb;
405         if (strcmp(update, "sparc2.2")==0 ) {
406                 /* 2.2 sparc put the events in the wrong place
407                  * So we copy the tail of the superblock
408                  * up 4 bytes before continuing
409                  */
410                 __u32 *sb32 = (__u32*)sb;
411                 memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
412                        sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
413                        (MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
414                 if (verbose >= 0)
415                         fprintf (stderr, Name ": adjusting superblock of %s for 2.2/sparc compatability.\n",
416                                  devname);
417         }
418         if (strcmp(update, "super-minor") ==0) {
419                 sb->md_minor = info->array.md_minor;
420                 if (verbose > 0)
421                         fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
422                                 devname, info->array.md_minor);
423         }
424         if (strcmp(update, "summaries") == 0) {
425                 int i;
426                 /* set nr_disks, active_disks, working_disks,
427                  * failed_disks, spare_disks based on disks[]
428                  * array in superblock.
429                  * Also make sure extra slots aren't 'failed'
430                  */
431                 sb->nr_disks = sb->active_disks =
432                         sb->working_disks = sb->failed_disks =
433                         sb->spare_disks = 0;
434                 for (i=0; i < MD_SB_DISKS ; i++)
435                         if (sb->disks[i].major ||
436                             sb->disks[i].minor) {
437                                 int state = sb->disks[i].state;
438                                 if (state & (1<<MD_DISK_REMOVED))
439                                         continue;
440                                 sb->nr_disks++;
441                                 if (state & (1<<MD_DISK_ACTIVE))
442                                         sb->active_disks++;
443                                 if (state & (1<<MD_DISK_FAULTY))
444                                         sb->failed_disks++;
445                                 else
446                                         sb->working_disks++;
447                                 if (state == 0)
448                                         sb->spare_disks++;
449                         } else if (i >= sb->raid_disks && sb->disks[i].number == 0)
450                                 sb->disks[i].state = 0;
451         }
452         if (strcmp(update, "force-one")==0) {
453                 /* Not enough devices for a working array, so
454                  * bring this one up-to-date.
455                  */
456                 __u32 ehi = sb->events_hi, elo = sb->events_lo;
457                 sb->events_hi = (info->events>>32) & 0xFFFFFFFF;
458                 sb->events_lo = (info->events) & 0xFFFFFFFF;
459                 if (sb->events_hi != ehi ||
460                     sb->events_lo != elo)
461                         rv = 1;
462         }
463         if (strcmp(update, "force-array")==0) {
464                 /* degraded array and 'force' requested, so
465                  * maybe need to mark it 'clean'
466                  */
467                 if ((sb->level == 5 || sb->level == 4 || sb->level == 6) &&
468                     (sb->state & (1 << MD_SB_CLEAN)) == 0) {
469                         /* need to force clean */
470                         sb->state |= (1 << MD_SB_CLEAN);
471                         rv = 1;
472                 }
473         }
474         if (strcmp(update, "assemble")==0) {
475                 int d = info->disk.number;
476                 int wonly = sb->disks[d].state & (1<<MD_DISK_WRITEMOSTLY);
477                 if ((sb->disks[d].state & ~(1<<MD_DISK_WRITEMOSTLY))
478                     != info->disk.state) {
479                         sb->disks[d].state = info->disk.state | wonly;
480                         rv = 1;
481                 }
482         }
483         if (strcmp(update, "linear-grow-new") == 0) {
484                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
485                 sb->disks[info->disk.number].number = info->disk.number;
486                 sb->disks[info->disk.number].major = info->disk.major;
487                 sb->disks[info->disk.number].minor = info->disk.minor;
488                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
489                 sb->disks[info->disk.number].state = info->disk.state;
490                 sb->this_disk = sb->disks[info->disk.number];
491         }
492         if (strcmp(update, "linear-grow-update") == 0) {
493                 sb->raid_disks = info->array.raid_disks;
494                 sb->nr_disks = info->array.nr_disks;
495                 sb->active_disks = info->array.active_disks;
496                 sb->working_disks = info->array.working_disks;
497                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
498                 sb->disks[info->disk.number].number = info->disk.number;
499                 sb->disks[info->disk.number].major = info->disk.major;
500                 sb->disks[info->disk.number].minor = info->disk.minor;
501                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
502                 sb->disks[info->disk.number].state = info->disk.state;
503         }
504         if (strcmp(update, "resync") == 0) {
505                 /* make sure resync happens */
506                 sb->state &= ~(1<<MD_SB_CLEAN);
507                 sb->recovery_cp = 0;
508         }
509         if (strcmp(update, "homehost") == 0 &&
510             homehost) {
511                 uuid_set = 0;
512                 update = "uuid";
513                 info->uuid[0] = sb->set_uuid0;
514                 info->uuid[1] = sb->set_uuid1;
515         }
516         if (strcmp(update, "uuid") == 0) {
517                 if (!uuid_set && homehost) {
518                         char buf[20];
519                         char *hash = sha1_buffer(homehost,
520                                                  strlen(homehost),
521                                                  buf);
522                         memcpy(info->uuid+2, hash, 8);
523                 }
524                 sb->set_uuid0 = info->uuid[0];
525                 sb->set_uuid1 = info->uuid[1];
526                 sb->set_uuid2 = info->uuid[2];
527                 sb->set_uuid3 = info->uuid[3];
528                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
529                         struct bitmap_super_s *bm;
530                         bm = (struct bitmap_super_s*)(sb+1);
531                         uuid_from_super0(st, (int*)bm->uuid);
532                 }
533         }
534         if (strcmp(update, "_reshape_progress")==0)
535                 sb->reshape_position = info->reshape_progress;
536
537         sb->sb_csum = calc_sb0_csum(sb);
538         return rv;
539 }
540
541 /*
542  * For verion-0 superblock, the homehost is 'stored' in the
543  * uuid.  8 bytes for a hash of the host leaving 8 bytes
544  * of random material.
545  * We use the first 8 bytes (64bits) of the sha1 of the
546  * host name
547  */
548
549
550 static int init_super0(struct supertype *st, mdu_array_info_t *info,
551                        unsigned long long size, char *ignored_name, char *homehost,
552                        int *uuid)
553 {
554         mdp_super_t *sb = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
555         int spares;
556         memset(sb, 0, MD_SB_BYTES + sizeof(bitmap_super_t));
557
558         st->sb = sb;
559         if (info->major_version == -1) {
560                 /* zeroing the superblock */
561                 return 0;
562         }
563
564         spares = info->working_disks - info->active_disks;
565         if (info->raid_disks + spares  > MD_SB_DISKS) {
566                 fprintf(stderr, Name ": too many devices requested: %d+%d > %d\n",
567                         info->raid_disks , spares, MD_SB_DISKS);
568                 return 0;
569         }
570
571         sb->md_magic = MD_SB_MAGIC;
572         sb->major_version = 0;
573         sb->minor_version = 90;
574         sb->patch_version = 0;
575         sb->gvalid_words = 0; /* ignored */
576         sb->ctime = time(0);
577         sb->level = info->level;
578         if (size != info->size)
579                 return 0;
580         sb->size = info->size;
581         sb->nr_disks = info->nr_disks;
582         sb->raid_disks = info->raid_disks;
583         sb->md_minor = info->md_minor;
584         sb->not_persistent = 0;
585         if (uuid) {
586                 sb->set_uuid0 = uuid[0];
587                 sb->set_uuid1 = uuid[1];
588                 sb->set_uuid2 = uuid[2];
589                 sb->set_uuid3 = uuid[3];
590         } else {
591                 int rfd = open("/dev/urandom", O_RDONLY);
592                 if (rfd < 0 || read(rfd, &sb->set_uuid0, 4) != 4)
593                         sb->set_uuid0 = random();
594                 if (rfd < 0 || read(rfd, &sb->set_uuid1, 12) != 12) {
595                         sb->set_uuid1 = random();
596                         sb->set_uuid2 = random();
597                         sb->set_uuid3 = random();
598                 }
599                 if (rfd >= 0)
600                         close(rfd);
601         }
602         if (homehost) {
603                 char buf[20];
604                 char *hash = sha1_buffer(homehost,
605                                          strlen(homehost),
606                                          buf);
607                 memcpy(&sb->set_uuid2, hash, 8);
608         }
609
610         sb->utime = sb->ctime;
611         sb->state = info->state;
612         sb->active_disks = info->active_disks;
613         sb->working_disks = info->working_disks;
614         sb->failed_disks = info->failed_disks;
615         sb->spare_disks = info->spare_disks;
616         sb->events_hi = 0;
617         sb->events_lo = 1;
618
619         sb->layout = info->layout;
620         sb->chunk_size = info->chunk_size;
621
622         return 1;
623 }
624
625 /* Add a device to the superblock being created */
626 static void add_to_super0(struct supertype *st, mdu_disk_info_t *dinfo)
627 {
628         mdp_super_t *sb = st->sb;
629         mdp_disk_t *dk = &sb->disks[dinfo->number];
630
631         dk->number = dinfo->number;
632         dk->major = dinfo->major;
633         dk->minor = dinfo->minor;
634         dk->raid_disk = dinfo->raid_disk;
635         dk->state = dinfo->state;
636 }
637
638 static int store_super0(struct supertype *st, int fd)
639 {
640         unsigned long long dsize;
641         unsigned long long offset;
642         mdp_super_t *super = st->sb;
643
644         if (!get_dev_size(fd, NULL, &dsize))
645                 return 1;
646
647         if (dsize < MD_RESERVED_SECTORS*512)
648                 return 2;
649
650         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
651
652         offset *= 512;
653
654         if (lseek64(fd, offset, 0)< 0LL)
655                 return 3;
656
657         if (write(fd, super, sizeof(*super)) != sizeof(*super))
658                 return 4;
659
660         if (super->state & (1<<MD_SB_BITMAP_PRESENT)) {
661                 struct bitmap_super_s * bm = (struct bitmap_super_s*)(super+1);
662                 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC)
663                         if (write(fd, bm, sizeof(*bm)) != sizeof(*bm))
664                             return 5;
665         }
666
667         fsync(fd);
668         return 0;
669 }
670
671 static int write_init_super0(struct supertype *st,
672                              mdu_disk_info_t *dinfo, char *devname)
673 {
674         mdp_super_t *sb = st->sb;
675         int fd = open(devname, O_RDWR|O_EXCL);
676         int rv;
677
678         if (fd < 0) {
679                 fprintf(stderr, Name ": Failed to open %s to write superblock\n", devname);
680                 return -1;
681         }
682
683         sb->disks[dinfo->number].state &= ~(1<<MD_DISK_FAULTY);
684
685         sb->this_disk = sb->disks[dinfo->number];
686         sb->sb_csum = calc_sb0_csum(sb);
687         rv = store_super0(st, fd);
688
689         if (rv == 0 && (sb->state & (1<<MD_SB_BITMAP_PRESENT)))
690                 rv = st->ss->write_bitmap(st, fd);
691
692         close(fd);
693         if (rv)
694                 fprintf(stderr, Name ": failed to write superblock to %s\n", devname);
695         return rv;
696 }
697
698 static int compare_super0(struct supertype *st, struct supertype *tst)
699 {
700         /*
701          * return:
702          *  0 same, or first was empty, and second was copied
703          *  1 second had wrong number
704          *  2 wrong uuid
705          *  3 wrong other info
706          */
707         mdp_super_t *first = st->sb;
708         mdp_super_t *second = tst->sb;
709         int uuid1[4], uuid2[4];
710
711         if (second->md_magic != MD_SB_MAGIC)
712                 return 1;
713         if (!first) {
714                 first = malloc(MD_SB_BYTES + sizeof(struct bitmap_super_s));
715                 memcpy(first, second, MD_SB_BYTES + sizeof(struct bitmap_super_s));
716                 st->sb = first;
717                 return 0;
718         }
719
720         uuid_from_super0(st, uuid1);
721         uuid_from_super0(tst, uuid2);
722         if (!same_uuid(uuid1, uuid2, 0))
723                 return 2;
724         if (first->major_version != second->major_version ||
725             first->minor_version != second->minor_version ||
726             first->patch_version != second->patch_version ||
727             first->gvalid_words  != second->gvalid_words  ||
728             first->ctime         != second->ctime         ||
729             first->level         != second->level         ||
730             first->size          != second->size          ||
731             first->raid_disks    != second->raid_disks    )
732                 return 3;
733
734         return 0;
735 }
736
737
738 static void free_super0(struct supertype *st);
739
740 static int load_super0(struct supertype *st, int fd, char *devname)
741 {
742         /* try to read in the superblock
743          * Return:
744          *  0 on success
745          *  1 on cannot get superblock
746          *  2 on superblock meaningless
747          */
748         unsigned long long dsize;
749         unsigned long long offset;
750         mdp_super_t *super;
751         int uuid[4];
752         struct bitmap_super_s *bsb;
753
754         free_super0(st);
755
756         if (!get_dev_size(fd, devname, &dsize))
757                 return 1;
758
759         if (dsize < MD_RESERVED_SECTORS*512) {
760                 if (devname)
761                         fprintf(stderr, Name
762                             ": %s is too small for md: size is %llu sectors.\n",
763                                 devname, dsize);
764                 return 1;
765         }
766
767         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
768
769         offset *= 512;
770
771         ioctl(fd, BLKFLSBUF, 0); /* make sure we read current data */
772
773         if (lseek64(fd, offset, 0)< 0LL) {
774                 if (devname)
775                         fprintf(stderr, Name ": Cannot seek to superblock on %s: %s\n",
776                                 devname, strerror(errno));
777                 return 1;
778         }
779
780         super = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
781
782         if (read(fd, super, sizeof(*super)) != MD_SB_BYTES) {
783                 if (devname)
784                         fprintf(stderr, Name ": Cannot read superblock on %s\n",
785                                 devname);
786                 free(super);
787                 return 1;
788         }
789
790         if (st->ss && st->minor_version == 9)
791                 super0_swap_endian(super);
792
793         if (super->md_magic != MD_SB_MAGIC) {
794                 if (devname)
795                         fprintf(stderr, Name ": No super block found on %s (Expected magic %08x, got %08x)\n",
796                                 devname, MD_SB_MAGIC, super->md_magic);
797                 free(super);
798                 return 2;
799         }
800
801         if (super->major_version != 0) {
802                 if (devname)
803                         fprintf(stderr, Name ": Cannot interpret superblock on %s - version is %d\n",
804                                 devname, super->major_version);
805                 free(super);
806                 return 2;
807         }
808         st->sb = super;
809
810         if (st->ss == NULL) {
811                 st->ss = &super0;
812                 st->minor_version = super->minor_version;
813                 st->max_devs = MD_SB_DISKS;
814         }
815
816         /* Now check on the bitmap superblock */
817         if ((super->state & (1<<MD_SB_BITMAP_PRESENT)) == 0)
818                 return 0;
819         /* Read the bitmap superblock and make sure it looks
820          * valid.  If it doesn't clear the bit.  An --assemble --force
821          * should get that written out.
822          */
823         if (read(fd, super+1, sizeof(struct bitmap_super_s))
824             != sizeof(struct bitmap_super_s))
825                 goto no_bitmap;
826
827         uuid_from_super0(st, uuid);
828         bsb = (struct bitmap_super_s *)(super+1);
829         if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
830             memcmp(bsb->uuid, uuid, 16) != 0)
831                 goto no_bitmap;
832         return 0;
833
834  no_bitmap:
835         super->state &= ~(1<<MD_SB_BITMAP_PRESENT);
836
837         return 0;
838 }
839
840 static struct supertype *match_metadata_desc0(char *arg)
841 {
842         struct supertype *st = malloc(sizeof(*st));
843         if (!st) return st;
844
845         st->ss = &super0;
846         st->minor_version = 90;
847         st->max_devs = MD_SB_DISKS;
848         st->sb = NULL;
849         if (strcmp(arg, "0") == 0 ||
850             strcmp(arg, "0.90") == 0 ||
851             strcmp(arg, "default") == 0 ||
852             strcmp(arg, "") == 0 /* no metadata */
853                 )
854                 return st;
855
856         st->minor_version = 91; /* reshape in progress */
857         if (strcmp(arg, "0.91") == 0) /* For dup_super support */
858                 return st;
859
860         st->minor_version = 9; /* flag for 'byte-swapped' */
861         if (strcmp(arg, "0.swap")==0 ||
862             strcmp(arg, "0.9") == 0) /* For dup_super support */
863                 return st;
864
865         free(st);
866         return NULL;
867 }
868
869 static __u64 avail_size0(struct supertype *st, __u64 devsize)
870 {
871         if (devsize < MD_RESERVED_SECTORS)
872                 return 0ULL;
873         return MD_NEW_SIZE_SECTORS(devsize);
874 }
875
876 static int add_internal_bitmap0(struct supertype *st, int *chunkp,
877                                 int delay, int write_behind,
878                                 unsigned long long size, int may_change,
879                                 int major)
880 {
881         /*
882          * The bitmap comes immediately after the superblock and must be 60K in size
883          * at most.  The default size is between 30K and 60K
884          *
885          * size is in sectors,  chunk is in bytes !!!
886          */
887         unsigned long long bits;
888         unsigned long long max_bits = 60*1024*8;
889         unsigned long long min_chunk;
890         int chunk = *chunkp;
891         mdp_super_t *sb = st->sb;
892         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MD_SB_BYTES);
893
894
895         min_chunk = 4096; /* sub-page chunks don't work yet.. */
896         bits = (size * 512) / min_chunk + 1;
897         while (bits > max_bits) {
898                 min_chunk *= 2;
899                 bits = (bits+1)/2;
900         }
901         if (chunk == UnSet)
902                 chunk = min_chunk;
903         else if (chunk < min_chunk)
904                 return 0; /* chunk size too small */
905
906         sb->state |= (1<<MD_SB_BITMAP_PRESENT);
907
908         memset(bms, 0, sizeof(*bms));
909         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
910         bms->version = __cpu_to_le32(major);
911         uuid_from_super0(st, (int*)bms->uuid);
912         bms->chunksize = __cpu_to_le32(chunk);
913         bms->daemon_sleep = __cpu_to_le32(delay);
914         bms->sync_size = __cpu_to_le64(size);
915         bms->write_behind = __cpu_to_le32(write_behind);
916         *chunkp = chunk;
917         return 1;
918 }
919
920
921 void locate_bitmap0(struct supertype *st, int fd)
922 {
923         unsigned long long dsize;
924         unsigned long long offset;
925
926         if (!get_dev_size(fd, NULL, &dsize))
927                 return;
928
929         if (dsize < MD_RESERVED_SECTORS*512)
930                 return;
931
932         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
933
934         offset *= 512;
935
936         offset += MD_SB_BYTES;
937
938         lseek64(fd, offset, 0);
939 }
940
941 int write_bitmap0(struct supertype *st, int fd)
942 {
943         unsigned long long dsize;
944         unsigned long long offset;
945         mdp_super_t *sb = st->sb;
946
947         int rv = 0;
948
949         int towrite, n;
950         char buf[4096];
951
952         if (!get_dev_size(fd, NULL, &dsize))
953                 return 1;
954
955
956         if (dsize < MD_RESERVED_SECTORS*512)
957                 return -1;
958
959         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
960
961         offset *= 512;
962
963         if (lseek64(fd, offset + 4096, 0)< 0LL)
964                 return 3;
965
966
967         if (write(fd, ((char*)sb)+MD_SB_BYTES, sizeof(bitmap_super_t)) !=
968             sizeof(bitmap_super_t))
969                 return -2;
970         towrite = 64*1024 - MD_SB_BYTES - sizeof(bitmap_super_t);
971         memset(buf, 0xff, sizeof(buf));
972         while (towrite > 0) {
973                 n = towrite;
974                 if (n > sizeof(buf))
975                         n = sizeof(buf);
976                 n = write(fd, buf, n);
977                 if (n > 0)
978                         towrite -= n;
979                 else
980                         break;
981         }
982         fsync(fd);
983         if (towrite)
984                 rv = -2;
985
986         return rv;
987 }
988
989 static void free_super0(struct supertype *st)
990 {
991         if (st->sb)
992                 free(st->sb);
993         st->sb = NULL;
994 }
995
996 struct superswitch super0 = {
997 #ifndef MDASSEMBLE
998         .examine_super = examine_super0,
999         .brief_examine_super = brief_examine_super0,
1000         .export_examine_super = export_examine_super0,
1001         .detail_super = detail_super0,
1002         .brief_detail_super = brief_detail_super0,
1003         .export_detail_super = export_detail_super0,
1004 #endif
1005         .match_home = match_home0,
1006         .uuid_from_super = uuid_from_super0,
1007         .getinfo_super = getinfo_super0,
1008         .update_super = update_super0,
1009         .init_super = init_super0,
1010         .add_to_super = add_to_super0,
1011         .store_super = store_super0,
1012         .write_init_super = write_init_super0,
1013         .compare_super = compare_super0,
1014         .load_super = load_super0,
1015         .match_metadata_desc = match_metadata_desc0,
1016         .avail_size = avail_size0,
1017         .add_internal_bitmap = add_internal_bitmap0,
1018         .locate_bitmap = locate_bitmap0,
1019         .write_bitmap = write_bitmap0,
1020         .free_super = free_super0,
1021         .major = 0,
1022         .swapuuid = 0,
1023 };