Move calls to SET_ARRAY_INFO to common helper.
[thirdparty/mdadm.git] / super-ddf.c
1 /*
2  * mdadm - manage Linux "md" devices aka RAID arrays.
3  *
4  * Copyright (C) 2006-2007 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: <neil@brown.name>
23  *
24  * Specifications for DDF takes from Common RAID DDF Specification Revision 1.2
25  * (July 28 2006).  Reused by permission of SNIA.
26  */
27
28 #define HAVE_STDINT_H 1
29 #include "mdadm.h"
30 #include "mdmon.h"
31 #include "sha1.h"
32 #include <values.h>
33
34 /* a non-official T10 name for creation GUIDs */
35 static char T10[] = "Linux-MD";
36
37 /* DDF timestamps are 1980 based, so we need to add
38  * second-in-decade-of-seventies to convert to linux timestamps.
39  * 10 years with 2 leap years.
40  */
41 #define DECADE (3600*24*(365*10+2))
42 unsigned long crc32(
43         unsigned long crc,
44         const unsigned char *buf,
45         unsigned len);
46
47 /* The DDF metadata handling.
48  * DDF metadata lives at the end of the device.
49  * The last 512 byte block provides an 'anchor' which is used to locate
50  * the rest of the metadata which usually lives immediately behind the anchor.
51  *
52  * Note:
53  *  - all multibyte numeric fields are bigendian.
54  *  - all strings are space padded.
55  *
56  */
57
58 /* Primary Raid Level (PRL) */
59 #define DDF_RAID0       0x00
60 #define DDF_RAID1       0x01
61 #define DDF_RAID3       0x03
62 #define DDF_RAID4       0x04
63 #define DDF_RAID5       0x05
64 #define DDF_RAID1E      0x11
65 #define DDF_JBOD        0x0f
66 #define DDF_CONCAT      0x1f
67 #define DDF_RAID5E      0x15
68 #define DDF_RAID5EE     0x25
69 #define DDF_RAID6       0x06
70
71 /* Raid Level Qualifier (RLQ) */
72 #define DDF_RAID0_SIMPLE        0x00
73 #define DDF_RAID1_SIMPLE        0x00 /* just 2 devices in this plex */
74 #define DDF_RAID1_MULTI         0x01 /* exactly 3 devices in this plex */
75 #define DDF_RAID3_0             0x00 /* parity in first extent */
76 #define DDF_RAID3_N             0x01 /* parity in last extent */
77 #define DDF_RAID4_0             0x00 /* parity in first extent */
78 #define DDF_RAID4_N             0x01 /* parity in last extent */
79 /* these apply to raid5e and raid5ee as well */
80 #define DDF_RAID5_0_RESTART     0x00 /* same as 'right asymmetric' - layout 1 */
81 #define DDF_RAID6_0_RESTART     0x01 /* raid6 different from raid5 here!!! */
82 #define DDF_RAID5_N_RESTART     0x02 /* same as 'left asymmetric' - layout 0 */
83 #define DDF_RAID5_N_CONTINUE    0x03 /* same as 'left symmetric' - layout 2 */
84
85 #define DDF_RAID1E_ADJACENT     0x00 /* raid10 nearcopies==2 */
86 #define DDF_RAID1E_OFFSET       0x01 /* raid10 offsetcopies==2 */
87
88 /* Secondary RAID Level (SRL) */
89 #define DDF_2STRIPED    0x00    /* This is weirder than RAID0 !! */
90 #define DDF_2MIRRORED   0x01
91 #define DDF_2CONCAT     0x02
92 #define DDF_2SPANNED    0x03    /* This is also weird - be careful */
93
94 /* Magic numbers */
95 #define DDF_HEADER_MAGIC        __cpu_to_be32(0xDE11DE11)
96 #define DDF_CONTROLLER_MAGIC    __cpu_to_be32(0xAD111111)
97 #define DDF_PHYS_RECORDS_MAGIC  __cpu_to_be32(0x22222222)
98 #define DDF_PHYS_DATA_MAGIC     __cpu_to_be32(0x33333333)
99 #define DDF_VIRT_RECORDS_MAGIC  __cpu_to_be32(0xDDDDDDDD)
100 #define DDF_VD_CONF_MAGIC       __cpu_to_be32(0xEEEEEEEE)
101 #define DDF_SPARE_ASSIGN_MAGIC  __cpu_to_be32(0x55555555)
102 #define DDF_VU_CONF_MAGIC       __cpu_to_be32(0x88888888)
103 #define DDF_VENDOR_LOG_MAGIC    __cpu_to_be32(0x01dBEEF0)
104 #define DDF_BBM_LOG_MAGIC       __cpu_to_be32(0xABADB10C)
105
106 #define DDF_GUID_LEN    24
107 #define DDF_REVISION_0  "01.00.00"
108 #define DDF_REVISION_2  "01.02.00"
109
110 struct ddf_header {
111         __u32   magic;          /* DDF_HEADER_MAGIC */
112         __u32   crc;
113         char    guid[DDF_GUID_LEN];
114         char    revision[8];    /* 01.02.00 */
115         __u32   seq;            /* starts at '1' */
116         __u32   timestamp;
117         __u8    openflag;
118         __u8    foreignflag;
119         __u8    enforcegroups;
120         __u8    pad0;           /* 0xff */
121         __u8    pad1[12];       /* 12 * 0xff */
122         /* 64 bytes so far */
123         __u8    header_ext[32]; /* reserved: fill with 0xff */
124         __u64   primary_lba;
125         __u64   secondary_lba;
126         __u8    type;
127         __u8    pad2[3];        /* 0xff */
128         __u32   workspace_len;  /* sectors for vendor space -
129                                  * at least 32768(sectors) */
130         __u64   workspace_lba;
131         __u16   max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */
132         __u16   max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */
133         __u16   max_partitions; /* i.e. max num of configuration
134                                    record entries per disk */
135         __u16   config_record_len; /* 1 +ROUNDUP(max_primary_element_entries
136                                                  *12/512) */
137         __u16   max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */
138         __u8    pad3[54];       /* 0xff */
139         /* 192 bytes so far */
140         __u32   controller_section_offset;
141         __u32   controller_section_length;
142         __u32   phys_section_offset;
143         __u32   phys_section_length;
144         __u32   virt_section_offset;
145         __u32   virt_section_length;
146         __u32   config_section_offset;
147         __u32   config_section_length;
148         __u32   data_section_offset;
149         __u32   data_section_length;
150         __u32   bbm_section_offset;
151         __u32   bbm_section_length;
152         __u32   diag_space_offset;
153         __u32   diag_space_length;
154         __u32   vendor_offset;
155         __u32   vendor_length;
156         /* 256 bytes so far */
157         __u8    pad4[256];      /* 0xff */
158 };
159
160 /* type field */
161 #define DDF_HEADER_ANCHOR       0x00
162 #define DDF_HEADER_PRIMARY      0x01
163 #define DDF_HEADER_SECONDARY    0x02
164
165 /* The content of the 'controller section' - global scope */
166 struct ddf_controller_data {
167         __u32   magic;                  /* DDF_CONTROLLER_MAGIC */
168         __u32   crc;
169         char    guid[DDF_GUID_LEN];
170         struct controller_type {
171                 __u16 vendor_id;
172                 __u16 device_id;
173                 __u16 sub_vendor_id;
174                 __u16 sub_device_id;
175         } type;
176         char    product_id[16];
177         __u8    pad[8]; /* 0xff */
178         __u8    vendor_data[448];
179 };
180
181 /* The content of phys_section - global scope */
182 struct phys_disk {
183         __u32   magic;          /* DDF_PHYS_RECORDS_MAGIC */
184         __u32   crc;
185         __u16   used_pdes;
186         __u16   max_pdes;
187         __u8    pad[52];
188         struct phys_disk_entry {
189                 char    guid[DDF_GUID_LEN];
190                 __u32   refnum;
191                 __u16   type;
192                 __u16   state;
193                 __u64   config_size; /* DDF structures must be after here */
194                 char    path[18];       /* another horrible structure really */
195                 __u8    pad[6];
196         } entries[0];
197 };
198
199 /* phys_disk_entry.type is a bitmap - bigendian remember */
200 #define DDF_Forced_PD_GUID              1
201 #define DDF_Active_in_VD                2
202 #define DDF_Global_Spare                4 /* VD_CONF records are ignored */
203 #define DDF_Spare                       8 /* overrides Global_spare */
204 #define DDF_Foreign                     16
205 #define DDF_Legacy                      32 /* no DDF on this device */
206
207 #define DDF_Interface_mask              0xf00
208 #define DDF_Interface_SCSI              0x100
209 #define DDF_Interface_SAS               0x200
210 #define DDF_Interface_SATA              0x300
211 #define DDF_Interface_FC                0x400
212
213 /* phys_disk_entry.state is a bigendian bitmap */
214 #define DDF_Online                      1
215 #define DDF_Failed                      2 /* overrides  1,4,8 */
216 #define DDF_Rebuilding                  4
217 #define DDF_Transition                  8
218 #define DDF_SMART                       16
219 #define DDF_ReadErrors                  32
220 #define DDF_Missing                     64
221
222 /* The content of the virt_section global scope */
223 struct virtual_disk {
224         __u32   magic;          /* DDF_VIRT_RECORDS_MAGIC */
225         __u32   crc;
226         __u16   populated_vdes;
227         __u16   max_vdes;
228         __u8    pad[52];
229         struct virtual_entry {
230                 char    guid[DDF_GUID_LEN];
231                 __u16   unit;
232                 __u16   pad0;   /* 0xffff */
233                 __u16   guid_crc;
234                 __u16   type;
235                 __u8    state;
236                 __u8    init_state;
237                 __u8    pad1[14];
238                 char    name[16];
239         } entries[0];
240 };
241
242 /* virtual_entry.type is a bitmap - bigendian */
243 #define DDF_Shared              1
244 #define DDF_Enforce_Groups      2
245 #define DDF_Unicode             4
246 #define DDF_Owner_Valid         8
247
248 /* virtual_entry.state is a bigendian bitmap */
249 #define DDF_state_mask          0x7
250 #define DDF_state_optimal       0x0
251 #define DDF_state_degraded      0x1
252 #define DDF_state_deleted       0x2
253 #define DDF_state_missing       0x3
254 #define DDF_state_failed        0x4
255 #define DDF_state_part_optimal  0x5
256
257 #define DDF_state_morphing      0x8
258 #define DDF_state_inconsistent  0x10
259
260 /* virtual_entry.init_state is a bigendian bitmap */
261 #define DDF_initstate_mask      0x03
262 #define DDF_init_not            0x00
263 #define DDF_init_quick          0x01 /* initialisation is progress.
264                                       * i.e. 'state_inconsistent' */
265 #define DDF_init_full           0x02
266
267 #define DDF_access_mask         0xc0
268 #define DDF_access_rw           0x00
269 #define DDF_access_ro           0x80
270 #define DDF_access_blocked      0xc0
271
272 /* The content of the config_section - local scope
273  * It has multiple records each config_record_len sectors
274  * They can be vd_config or spare_assign
275  */
276
277 struct vd_config {
278         __u32   magic;          /* DDF_VD_CONF_MAGIC */
279         __u32   crc;
280         char    guid[DDF_GUID_LEN];
281         __u32   timestamp;
282         __u32   seqnum;
283         __u8    pad0[24];
284         __u16   prim_elmnt_count;
285         __u8    chunk_shift;    /* 0 == 512, 1==1024 etc */
286         __u8    prl;
287         __u8    rlq;
288         __u8    sec_elmnt_count;
289         __u8    sec_elmnt_seq;
290         __u8    srl;
291         __u64   blocks;         /* blocks per component could be different
292                                  * on different component devices...(only
293                                  * for concat I hope) */
294         __u64   array_blocks;   /* blocks in array */
295         __u8    pad1[8];
296         __u32   spare_refs[8];
297         __u8    cache_pol[8];
298         __u8    bg_rate;
299         __u8    pad2[3];
300         __u8    pad3[52];
301         __u8    pad4[192];
302         __u8    v0[32]; /* reserved- 0xff */
303         __u8    v1[32]; /* reserved- 0xff */
304         __u8    v2[16]; /* reserved- 0xff */
305         __u8    v3[16]; /* reserved- 0xff */
306         __u8    vendor[32];
307         __u32   phys_refnum[0]; /* refnum of each disk in sequence */
308       /*__u64   lba_offset[0];  LBA offset in each phys.  Note extents in a
309                                 bvd are always the same size */
310 };
311
312 /* vd_config.cache_pol[7] is a bitmap */
313 #define DDF_cache_writeback     1       /* else writethrough */
314 #define DDF_cache_wadaptive     2       /* only applies if writeback */
315 #define DDF_cache_readahead     4
316 #define DDF_cache_radaptive     8       /* only if doing read-ahead */
317 #define DDF_cache_ifnobatt      16      /* even to write cache if battery is poor */
318 #define DDF_cache_wallowed      32      /* enable write caching */
319 #define DDF_cache_rallowed      64      /* enable read caching */
320
321 struct spare_assign {
322         __u32   magic;          /* DDF_SPARE_ASSIGN_MAGIC */
323         __u32   crc;
324         __u32   timestamp;
325         __u8    reserved[7];
326         __u8    type;
327         __u16   populated;      /* SAEs used */
328         __u16   max;            /* max SAEs */
329         __u8    pad[8];
330         struct spare_assign_entry {
331                 char    guid[DDF_GUID_LEN];
332                 __u16   secondary_element;
333                 __u8    pad[6];
334         } spare_ents[0];
335 };
336 /* spare_assign.type is a bitmap */
337 #define DDF_spare_dedicated     0x1     /* else global */
338 #define DDF_spare_revertible    0x2     /* else committable */
339 #define DDF_spare_active        0x4     /* else not active */
340 #define DDF_spare_affinity      0x8     /* enclosure affinity */
341
342 /* The data_section contents - local scope */
343 struct disk_data {
344         __u32   magic;          /* DDF_PHYS_DATA_MAGIC */
345         __u32   crc;
346         char    guid[DDF_GUID_LEN];
347         __u32   refnum;         /* crc of some magic drive data ... */
348         __u8    forced_ref;     /* set when above was not result of magic */
349         __u8    forced_guid;    /* set if guid was forced rather than magic */
350         __u8    vendor[32];
351         __u8    pad[442];
352 };
353
354 /* bbm_section content */
355 struct bad_block_log {
356         __u32   magic;
357         __u32   crc;
358         __u16   entry_count;
359         __u32   spare_count;
360         __u8    pad[10];
361         __u64   first_spare;
362         struct mapped_block {
363                 __u64   defective_start;
364                 __u32   replacement_start;
365                 __u16   remap_count;
366                 __u8    pad[2];
367         } entries[0];
368 };
369
370 /* Struct for internally holding ddf structures */
371 /* The DDF structure stored on each device is potentially
372  * quite different, as some data is global and some is local.
373  * The global data is:
374  *   - ddf header
375  *   - controller_data
376  *   - Physical disk records
377  *   - Virtual disk records
378  * The local data is:
379  *   - Configuration records
380  *   - Physical Disk data section
381  *  (  and Bad block and vendor which I don't care about yet).
382  *
383  * The local data is parsed into separate lists as it is read
384  * and reconstructed for writing.  This means that we only need
385  * to make config changes once and they are automatically
386  * propagated to all devices.
387  * Note that the ddf_super has space of the conf and disk data
388  * for this disk and also for a list of all such data.
389  * The list is only used for the superblock that is being
390  * built in Create or Assemble to describe the whole array.
391  */
392 struct ddf_super {
393         struct ddf_header anchor, primary, secondary;
394         struct ddf_controller_data controller;
395         struct ddf_header *active;
396         struct phys_disk        *phys;
397         struct virtual_disk     *virt;
398         int pdsize, vdsize;
399         int max_part, mppe, conf_rec_len;
400         int currentdev;
401         int updates_pending;
402         struct vcl {
403                 union {
404                         char space[512];
405                         struct {
406                                 struct vcl      *next;
407                                 __u64           *lba_offset; /* location in 'conf' of
408                                                               * the lba table */
409                                 int     vcnum; /* index into ->virt */
410                                 __u64           *block_sizes; /* NULL if all the same */
411                         };
412                 };
413                 struct vd_config conf;
414         } *conflist, *currentconf;
415         struct dl {
416                 union {
417                         char space[512];
418                         struct {
419                                 struct dl       *next;
420                                 int major, minor;
421                                 char *devname;
422                                 int fd;
423                                 unsigned long long size; /* sectors */
424                                 int pdnum;      /* index in ->phys */
425                                 struct spare_assign *spare;
426                         };
427                 };
428                 struct disk_data disk;
429                 void *mdupdate; /* hold metadata update */
430                 struct vcl *vlist[0]; /* max_part in size */
431         } *dlist, *add_list;
432 };
433
434 #ifndef offsetof
435 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
436 #endif
437
438
439 static int calc_crc(void *buf, int len)
440 {
441         /* crcs are always at the same place as in the ddf_header */
442         struct ddf_header *ddf = buf;
443         __u32 oldcrc = ddf->crc;
444         __u32 newcrc;
445         ddf->crc = 0xffffffff;
446
447         newcrc = crc32(0, buf, len);
448         ddf->crc = oldcrc;
449         return newcrc;
450 }
451
452 static int load_ddf_header(int fd, unsigned long long lba,
453                            unsigned long long size,
454                            int type,
455                            struct ddf_header *hdr, struct ddf_header *anchor)
456 {
457         /* read a ddf header (primary or secondary) from fd/lba
458          * and check that it is consistent with anchor
459          * Need to check:
460          *   magic, crc, guid, rev, and LBA's header_type, and
461          *  everything after header_type must be the same
462          */
463         if (lba >= size-1)
464                 return 0;
465
466         if (lseek64(fd, lba<<9, 0) < 0)
467                 return 0;
468
469         if (read(fd, hdr, 512) != 512)
470                 return 0;
471
472         if (hdr->magic != DDF_HEADER_MAGIC)
473                 return 0;
474         if (calc_crc(hdr, 512) != hdr->crc)
475                 return 0;
476         if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
477             memcmp(anchor->revision, hdr->revision, 8) != 0 ||
478             anchor->primary_lba != hdr->primary_lba ||
479             anchor->secondary_lba != hdr->secondary_lba ||
480             hdr->type != type ||
481             memcmp(anchor->pad2, hdr->pad2, 512 -
482                    offsetof(struct ddf_header, pad2)) != 0)
483                 return 0;
484
485         /* Looks good enough to me... */
486         return 1;
487 }
488
489 static void *load_section(int fd, struct ddf_super *super, void *buf,
490                           __u32 offset_be, __u32 len_be, int check)
491 {
492         unsigned long long offset = __be32_to_cpu(offset_be);
493         unsigned long long len = __be32_to_cpu(len_be);
494         int dofree = (buf == NULL);
495
496         if (check)
497                 if (len != 2 && len != 8 && len != 32
498                     && len != 128 && len != 512)
499                         return NULL;
500
501         if (len > 1024)
502                 return NULL;
503         if (buf) {
504                 /* All pre-allocated sections are a single block */
505                 if (len != 1)
506                         return NULL;
507         } else {
508                 posix_memalign(&buf, 512, len<<9);
509         }
510
511         if (!buf)
512                 return NULL;
513
514         if (super->active->type == 1)
515                 offset += __be64_to_cpu(super->active->primary_lba);
516         else
517                 offset += __be64_to_cpu(super->active->secondary_lba);
518
519         if (lseek64(fd, offset<<9, 0) != (offset<<9)) {
520                 if (dofree)
521                         free(buf);
522                 return NULL;
523         }
524         if (read(fd, buf, len<<9) != (len<<9)) {
525                 if (dofree)
526                         free(buf);
527                 return NULL;
528         }
529         return buf;
530 }
531
532 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
533 {
534         unsigned long long dsize;
535
536         get_dev_size(fd, NULL, &dsize);
537
538         if (lseek64(fd, dsize-512, 0) < 0) {
539                 if (devname)
540                         fprintf(stderr,
541                                 Name": Cannot seek to anchor block on %s: %s\n",
542                                 devname, strerror(errno));
543                 return 1;
544         }
545         if (read(fd, &super->anchor, 512) != 512) {
546                 if (devname)
547                         fprintf(stderr,
548                                 Name ": Cannot read anchor block on %s: %s\n",
549                                 devname, strerror(errno));
550                 return 1;
551         }
552         if (super->anchor.magic != DDF_HEADER_MAGIC) {
553                 if (devname)
554                         fprintf(stderr, Name ": no DDF anchor found on %s\n",
555                                 devname);
556                 return 2;
557         }
558         if (calc_crc(&super->anchor, 512) != super->anchor.crc) {
559                 if (devname)
560                         fprintf(stderr, Name ": bad CRC on anchor on %s\n",
561                                 devname);
562                 return 2;
563         }
564         if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
565             memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
566                 if (devname)
567                         fprintf(stderr, Name ": can only support super revision"
568                                 " %.8s and earlier, not %.8s on %s\n",
569                                 DDF_REVISION_2, super->anchor.revision,devname);
570                 return 2;
571         }
572         if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba),
573                             dsize >> 9,  1,
574                             &super->primary, &super->anchor) == 0) {
575                 if (devname)
576                         fprintf(stderr,
577                                 Name ": Failed to load primary DDF header "
578                                 "on %s\n", devname);
579                 return 2;
580         }
581         super->active = &super->primary;
582         if (load_ddf_header(fd, __be64_to_cpu(super->anchor.secondary_lba),
583                             dsize >> 9,  2,
584                             &super->secondary, &super->anchor)) {
585                 if ((__be32_to_cpu(super->primary.seq)
586                      < __be32_to_cpu(super->secondary.seq) &&
587                      !super->secondary.openflag)
588                     || (__be32_to_cpu(super->primary.seq)
589                         == __be32_to_cpu(super->secondary.seq) &&
590                         super->primary.openflag && !super->secondary.openflag)
591                         )
592                         super->active = &super->secondary;
593         }
594         return 0;
595 }
596
597 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
598 {
599         void *ok;
600         ok = load_section(fd, super, &super->controller,
601                           super->active->controller_section_offset,
602                           super->active->controller_section_length,
603                           0);
604         super->phys = load_section(fd, super, NULL,
605                                    super->active->phys_section_offset,
606                                    super->active->phys_section_length,
607                                    1);
608         super->pdsize = __be32_to_cpu(super->active->phys_section_length) * 512;
609
610         super->virt = load_section(fd, super, NULL,
611                                    super->active->virt_section_offset,
612                                    super->active->virt_section_length,
613                                    1);
614         super->vdsize = __be32_to_cpu(super->active->virt_section_length) * 512;
615         if (!ok ||
616             !super->phys ||
617             !super->virt) {
618                 free(super->phys);
619                 free(super->virt);
620                 super->phys = NULL;
621                 super->virt = NULL;
622                 return 2;
623         }
624         super->conflist = NULL;
625         super->dlist = NULL;
626
627         super->max_part = __be16_to_cpu(super->active->max_partitions);
628         super->mppe = __be16_to_cpu(super->active->max_primary_element_entries);
629         super->conf_rec_len = __be16_to_cpu(super->active->config_record_len);
630         return 0;
631 }
632
633 static int load_ddf_local(int fd, struct ddf_super *super,
634                           char *devname, int keep)
635 {
636         struct dl *dl;
637         struct stat stb;
638         char *conf;
639         int i;
640         int vnum;
641         int max_virt_disks = __be16_to_cpu(super->active->max_vd_entries);
642         unsigned long long dsize;
643
644         /* First the local disk info */
645         posix_memalign((void**)&dl, 512,
646                        sizeof(*dl) +
647                        (super->max_part) * sizeof(dl->vlist[0]));
648
649         load_section(fd, super, &dl->disk,
650                      super->active->data_section_offset,
651                      super->active->data_section_length,
652                      0);
653         dl->devname = devname ? strdup(devname) : NULL;
654
655         fstat(fd, &stb);
656         dl->major = major(stb.st_rdev);
657         dl->minor = minor(stb.st_rdev);
658         dl->next = super->dlist;
659         dl->fd = keep ? fd : -1;
660
661         dl->size = 0;
662         if (get_dev_size(fd, devname, &dsize))
663                 dl->size = dsize >> 9;
664         dl->spare = NULL;
665         for (i=0 ; i < super->max_part ; i++)
666                 dl->vlist[i] = NULL;
667         super->dlist = dl;
668         dl->pdnum = -1;
669         for (i=0; i < __be16_to_cpu(super->active->max_pd_entries); i++)
670                 if (memcmp(super->phys->entries[i].guid,
671                            dl->disk.guid, DDF_GUID_LEN) == 0)
672                         dl->pdnum = i;
673
674         /* Now the config list. */
675         /* 'conf' is an array of config entries, some of which are
676          * probably invalid.  Those which are good need to be copied into
677          * the conflist
678          */
679
680         conf = load_section(fd, super, NULL,
681                             super->active->config_section_offset,
682                             super->active->config_section_length,
683                             0);
684
685         vnum = 0;
686         for (i = 0;
687              i < __be32_to_cpu(super->active->config_section_length);
688              i += super->conf_rec_len) {
689                 struct vd_config *vd =
690                         (struct vd_config *)((char*)conf + i*512);
691                 struct vcl *vcl;
692
693                 if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) {
694                         if (dl->spare)
695                                 continue;
696                         posix_memalign((void**)&dl->spare, 512,
697                                        super->conf_rec_len*512);
698                         memcpy(dl->spare, vd, super->conf_rec_len*512);
699                         continue;
700                 }
701                 if (vd->magic != DDF_VD_CONF_MAGIC)
702                         continue;
703                 for (vcl = super->conflist; vcl; vcl = vcl->next) {
704                         if (memcmp(vcl->conf.guid,
705                                    vd->guid, DDF_GUID_LEN) == 0)
706                                 break;
707                 }
708
709                 if (vcl) {
710                         dl->vlist[vnum++] = vcl;
711                         if (__be32_to_cpu(vd->seqnum) <=
712                             __be32_to_cpu(vcl->conf.seqnum))
713                                 continue;
714                 } else {
715                         posix_memalign((void**)&vcl, 512,
716                                        (super->conf_rec_len*512 +
717                                         offsetof(struct vcl, conf)));
718                         vcl->next = super->conflist;
719                         vcl->block_sizes = NULL; /* FIXME not for CONCAT */
720                         super->conflist = vcl;
721                         dl->vlist[vnum++] = vcl;
722                 }
723                 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
724                 vcl->lba_offset = (__u64*)
725                         &vcl->conf.phys_refnum[super->mppe];
726
727                 for (i=0; i < max_virt_disks ; i++)
728                         if (memcmp(super->virt->entries[i].guid,
729                                    vcl->conf.guid, DDF_GUID_LEN)==0)
730                                 break;
731                 if (i < max_virt_disks)
732                         vcl->vcnum = i;
733         }
734         free(conf);
735
736         return 0;
737 }
738
739 #ifndef MDASSEMBLE
740 static int load_super_ddf_all(struct supertype *st, int fd,
741                               void **sbp, char *devname, int keep_fd);
742 #endif
743 static int load_super_ddf(struct supertype *st, int fd,
744                           char *devname)
745 {
746         unsigned long long dsize;
747         struct ddf_super *super;
748         int rv;
749
750 #ifndef MDASSEMBLE
751         /* if 'fd' is a container, load metadata from all the devices */
752         if (load_super_ddf_all(st, fd, &st->sb, devname, 1) == 0)
753                 return 0;
754 #endif
755         if (st->subarray[0])
756                 return 1; /* FIXME Is this correct */
757
758         if (get_dev_size(fd, devname, &dsize) == 0)
759                 return 1;
760
761         /* 32M is a lower bound */
762         if (dsize <= 32*1024*1024) {
763                 if (devname) {
764                         fprintf(stderr,
765                                 Name ": %s is too small for ddf: "
766                                 "size is %llu sectors.\n",
767                                 devname, dsize>>9);
768                         return 1;
769                 }
770         }
771         if (dsize & 511) {
772                 if (devname) {
773                         fprintf(stderr,
774                                 Name ": %s is an odd size for ddf: "
775                                 "size is %llu bytes.\n",
776                                 devname, dsize);
777                         return 1;
778                 }
779         }
780
781         if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) {
782                 fprintf(stderr, Name ": malloc of %zu failed.\n",
783                         sizeof(*super));
784                 return 1;
785         }
786         memset(super, 0, sizeof(*super));
787
788         rv = load_ddf_headers(fd, super, devname);
789         if (rv) {
790                 free(super);
791                 return rv;
792         }
793
794         /* Have valid headers and have chosen the best. Let's read in the rest*/
795
796         rv = load_ddf_global(fd, super, devname);
797
798         if (rv) {
799                 if (devname)
800                         fprintf(stderr,
801                                 Name ": Failed to load all information "
802                                 "sections on %s\n", devname);
803                 free(super);
804                 return rv;
805         }
806
807         load_ddf_local(fd, super, devname, 0);
808
809         /* Should possibly check the sections .... */
810
811         st->sb = super;
812         if (st->ss == NULL) {
813                 st->ss = &super_ddf;
814                 st->minor_version = 0;
815                 st->max_devs = 512;
816         }
817         return 0;
818
819 }
820
821 static void free_super_ddf(struct supertype *st)
822 {
823         struct ddf_super *ddf = st->sb;
824         if (ddf == NULL)
825                 return;
826         free(ddf->phys);
827         free(ddf->virt);
828         while (ddf->conflist) {
829                 struct vcl *v = ddf->conflist;
830                 ddf->conflist = v->next;
831                 if (v->block_sizes)
832                         free(v->block_sizes);
833                 free(v);
834         }
835         while (ddf->dlist) {
836                 struct dl *d = ddf->dlist;
837                 ddf->dlist = d->next;
838                 if (d->fd >= 0)
839                         close(d->fd);
840                 if (d->spare)
841                         free(d->spare);
842                 free(d);
843         }
844         free(ddf);
845         st->sb = NULL;
846 }
847
848 static struct supertype *match_metadata_desc_ddf(char *arg)
849 {
850         /* 'ddf' only support containers */
851         struct supertype *st;
852         if (strcmp(arg, "ddf") != 0 &&
853             strcmp(arg, "default") != 0
854                 )
855                 return NULL;
856
857         st = malloc(sizeof(*st));
858         memset(st, 0, sizeof(*st));
859         st->ss = &super_ddf;
860         st->max_devs = 512;
861         st->minor_version = 0;
862         st->sb = NULL;
863         return st;
864 }
865
866
867 #ifndef MDASSEMBLE
868
869 static mapping_t ddf_state[] = {
870         { "Optimal", 0},
871         { "Degraded", 1},
872         { "Deleted", 2},
873         { "Missing", 3},
874         { "Failed", 4},
875         { "Partially Optimal", 5},
876         { "-reserved-", 6},
877         { "-reserved-", 7},
878         { NULL, 0}
879 };
880
881 static mapping_t ddf_init_state[] = {
882         { "Not Initialised", 0},
883         { "QuickInit in Progress", 1},
884         { "Fully Initialised", 2},
885         { "*UNKNOWN*", 3},
886         { NULL, 0}
887 };
888 static mapping_t ddf_access[] = {
889         { "Read/Write", 0},
890         { "Reserved", 1},
891         { "Read Only", 2},
892         { "Blocked (no access)", 3},
893         { NULL ,0}
894 };
895
896 static mapping_t ddf_level[] = {
897         { "RAID0", DDF_RAID0},
898         { "RAID1", DDF_RAID1},
899         { "RAID3", DDF_RAID3},
900         { "RAID4", DDF_RAID4},
901         { "RAID5", DDF_RAID5},
902         { "RAID1E",DDF_RAID1E},
903         { "JBOD",  DDF_JBOD},
904         { "CONCAT",DDF_CONCAT},
905         { "RAID5E",DDF_RAID5E},
906         { "RAID5EE",DDF_RAID5EE},
907         { "RAID6", DDF_RAID6},
908         { NULL, 0}
909 };
910 static mapping_t ddf_sec_level[] = {
911         { "Striped", DDF_2STRIPED},
912         { "Mirrored", DDF_2MIRRORED},
913         { "Concat", DDF_2CONCAT},
914         { "Spanned", DDF_2SPANNED},
915         { NULL, 0}
916 };
917 #endif
918
919 struct num_mapping {
920         int num1, num2;
921 };
922 static struct num_mapping ddf_level_num[] = {
923         { DDF_RAID0, 0 },
924         { DDF_RAID1, 1 },
925         { DDF_RAID3, LEVEL_UNSUPPORTED },
926         { DDF_RAID4, 4 },
927         { DDF_RAID5, 5 },
928         { DDF_RAID1E, LEVEL_UNSUPPORTED },
929         { DDF_JBOD, LEVEL_UNSUPPORTED },
930         { DDF_CONCAT, LEVEL_LINEAR },
931         { DDF_RAID5E, LEVEL_UNSUPPORTED },
932         { DDF_RAID5EE, LEVEL_UNSUPPORTED },
933         { DDF_RAID6, 6},
934         { MAXINT, MAXINT }
935 };
936
937 static int map_num1(struct num_mapping *map, int num)
938 {
939         int i;
940         for (i=0 ; map[i].num1 != MAXINT; i++)
941                 if (map[i].num1 == num)
942                         break;
943         return map[i].num2;
944 }
945
946 #ifndef MDASSEMBLE
947 static void print_guid(char *guid, int tstamp)
948 {
949         /* A GUIDs are part (or all) ASCII and part binary.
950          * They tend to be space padded.
951          * We print the GUID in HEX, then in parentheses add
952          * any initial ASCII sequence, and a possible
953          * time stamp from bytes 16-19
954          */
955         int l = DDF_GUID_LEN;
956         int i;
957
958         for (i=0 ; i<DDF_GUID_LEN ; i++) {
959                 if ((i&3)==0 && i != 0) printf(":");
960                 printf("%02X", guid[i]&255);
961         }
962
963         printf(" (");
964         while (l && guid[l-1] == ' ')
965                 l--;
966         for (i=0 ; i<l ; i++) {
967                 if (guid[i] >= 0x20 && guid[i] < 0x7f)
968                         fputc(guid[i], stdout);
969                 else
970                         break;
971         }
972         if (tstamp) {
973                 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
974                 char tbuf[100];
975                 struct tm *tm;
976                 tm = localtime(&then);
977                 strftime(tbuf, 100, " %D %T",tm);
978                 fputs(tbuf, stdout);
979         }
980         printf(")");
981 }
982
983 static void examine_vd(int n, struct ddf_super *sb, char *guid)
984 {
985         int crl = sb->conf_rec_len;
986         struct vcl *vcl;
987
988         for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
989                 struct vd_config *vc = &vcl->conf;
990
991                 if (calc_crc(vc, crl*512) != vc->crc)
992                         continue;
993                 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
994                         continue;
995
996                 /* Ok, we know about this VD, let's give more details */
997                 printf(" Raid Devices[%d] : %d\n", n,
998                        __be16_to_cpu(vc->prim_elmnt_count));
999                 printf("   Chunk Size[%d] : %d sectors\n", n,
1000                        1 << vc->chunk_shift);
1001                 printf("   Raid Level[%d] : %s\n", n,
1002                        map_num(ddf_level, vc->prl)?:"-unknown-");
1003                 if (vc->sec_elmnt_count != 1) {
1004                         printf("  Secondary Position[%d] : %d of %d\n", n,
1005                                vc->sec_elmnt_seq, vc->sec_elmnt_count);
1006                         printf("  Secondary Level[%d] : %s\n", n,
1007                                map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1008                 }
1009                 printf("  Device Size[%d] : %llu\n", n,
1010                        __be64_to_cpu(vc->blocks)/2);
1011                 printf("   Array Size[%d] : %llu\n", n,
1012                        __be64_to_cpu(vc->array_blocks)/2);
1013         }
1014 }
1015
1016 static void examine_vds(struct ddf_super *sb)
1017 {
1018         int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1019         int i;
1020         printf("  Virtual Disks : %d\n", cnt);
1021
1022         for (i=0; i<cnt; i++) {
1023                 struct virtual_entry *ve = &sb->virt->entries[i];
1024                 printf("      VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1025                 printf("\n");
1026                 printf("         unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1027                 printf("        state[%d] : %s, %s%s\n", i,
1028                        map_num(ddf_state, ve->state & 7),
1029                        (ve->state & 8) ? "Morphing, ": "",
1030                        (ve->state & 16)? "Not Consistent" : "Consistent");
1031                 printf("   init state[%d] : %s\n", i,
1032                        map_num(ddf_init_state, ve->init_state&3));
1033                 printf("       access[%d] : %s\n", i,
1034                        map_num(ddf_access, (ve->init_state>>6) & 3));
1035                 printf("         Name[%d] : %.16s\n", i, ve->name);
1036                 examine_vd(i, sb, ve->guid);
1037         }
1038         if (cnt) printf("\n");
1039 }
1040
1041 static void examine_pds(struct ddf_super *sb)
1042 {
1043         int cnt = __be16_to_cpu(sb->phys->used_pdes);
1044         int i;
1045         struct dl *dl;
1046         printf(" Physical Disks : %d\n", cnt);
1047
1048         for (i=0 ; i<cnt ; i++) {
1049                 struct phys_disk_entry *pd = &sb->phys->entries[i];
1050                 int type = __be16_to_cpu(pd->type);
1051                 int state = __be16_to_cpu(pd->state);
1052
1053                 printf("      PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1054                 printf("\n");
1055                 printf("          ref[%d] : %08x\n", i,
1056                        __be32_to_cpu(pd->refnum));
1057                 printf("         mode[%d] : %s%s%s%s%s\n", i,
1058                        (type&2) ? "active":"",
1059                        (type&4) ? "Global Spare":"",
1060                        (type&8) ? "spare" : "",
1061                        (type&16)? ", foreign" : "",
1062                        (type&32)? "pass-through" : "");
1063                 printf("        state[%d] : %s%s%s%s%s%s%s\n", i,
1064                        (state&1)? "Online": "Offline",
1065                        (state&2)? ", Failed": "",
1066                        (state&4)? ", Rebuilding": "",
1067                        (state&8)? ", in-transition": "",
1068                        (state&16)? ", SMART errors": "",
1069                        (state&32)? ", Unrecovered Read Errors": "",
1070                        (state&64)? ", Missing" : "");
1071                 printf("   Avail Size[%d] : %llu K\n", i,
1072                        __be64_to_cpu(pd->config_size)>>1);
1073                 for (dl = sb->dlist; dl ; dl = dl->next) {
1074                         if (dl->disk.refnum == pd->refnum) {
1075                                 char *dv = map_dev(dl->major, dl->minor, 0);
1076                                 if (dv)
1077                                         printf("       Device[%d] : %s\n",
1078                                                i, dv);
1079                         }
1080                 }
1081                 printf("\n");
1082         }
1083 }
1084
1085 static void examine_super_ddf(struct supertype *st, char *homehost)
1086 {
1087         struct ddf_super *sb = st->sb;
1088
1089         printf("          Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1090         printf("        Version : %.8s\n", sb->anchor.revision);
1091         printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1092         printf("\n");
1093         printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1094         printf("\n");
1095         printf("            Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1096         printf("  Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1097                ?"yes" : "no");
1098         examine_vds(sb);
1099         examine_pds(sb);
1100 }
1101
1102 static void brief_examine_super_ddf(struct supertype *st)
1103 {
1104         /* We just write a generic DDF ARRAY entry
1105          * The uuid is all hex, 6 groups of 4 bytes
1106          */
1107         struct ddf_super *ddf = st->sb;
1108         int i;
1109         printf("ARRAY /dev/ddf metadata=ddf UUID=");
1110         for (i = 0; i < DDF_GUID_LEN; i++) {
1111                 if ((i&3) == 0 && i != 0)
1112                         printf(":");
1113                 printf("%02X", 255&ddf->anchor.guid[i]);
1114         }
1115         printf("\n");
1116 }
1117
1118 static void detail_super_ddf(struct supertype *st, char *homehost)
1119 {
1120         /* FIXME later
1121          * Could print DDF GUID
1122          * Need to find which array
1123          *  If whole, briefly list all arrays
1124          *  If one, give name
1125          */
1126 }
1127
1128 static void brief_detail_super_ddf(struct supertype *st)
1129 {
1130         /* FIXME I really need to know which array we are detailing.
1131          * Can that be stored in ddf_super??
1132          */
1133 //      struct ddf_super *ddf = st->sb;
1134 }
1135 #endif
1136
1137 static int match_home_ddf(struct supertype *st, char *homehost)
1138 {
1139         /* It matches 'this' host if the controller is a
1140          * Linux-MD controller with vendor_data matching
1141          * the hostname
1142          */
1143         struct ddf_super *ddf = st->sb;
1144         int len = strlen(homehost);
1145
1146         return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1147                 len < sizeof(ddf->controller.vendor_data) &&
1148                 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1149                 ddf->controller.vendor_data[len] == 0);
1150 }
1151
1152 #ifndef MDASSEMBLE
1153 static struct vd_config *find_vdcr(struct ddf_super *ddf, int inst)
1154 {
1155         struct vcl *v;
1156
1157         for (v = ddf->conflist; v; v = v->next)
1158                 if (inst == v->vcnum)
1159                         return &v->conf;
1160         return NULL;
1161 }
1162 #endif
1163
1164 static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1165 {
1166         /* Find the entry in phys_disk which has the given refnum
1167          * and return it's index
1168          */
1169         int i;
1170         for (i=0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1171                 if (ddf->phys->entries[i].refnum == phys_refnum)
1172                         return i;
1173         return -1;
1174 }
1175
1176 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1177 {
1178         /* The uuid returned here is used for:
1179          *  uuid to put into bitmap file (Create, Grow)
1180          *  uuid for backup header when saving critical section (Grow)
1181          *  comparing uuids when re-adding a device into an array
1182          * For each of these we can make do with a truncated
1183          * or hashed uuid rather than the original, as long as
1184          * everyone agrees.
1185          * In each case the uuid required is that of the data-array,
1186          * not the device-set.
1187          * In the case of SVD we assume the BVD is of interest,
1188          * though that might be the case if a bitmap were made for
1189          * a mirrored SVD - worry about that later.
1190          * So we need to find the VD configuration record for the
1191          * relevant BVD and extract the GUID and Secondary_Element_Seq.
1192          * The first 16 bytes of the sha1 of these is used.
1193          */
1194         struct ddf_super *ddf = st->sb;
1195         struct vcl *vcl = ddf->currentconf;
1196
1197         if (!vcl)
1198                 memset(uuid, 0, sizeof (uuid));
1199         else {
1200                 char buf[20];
1201                 struct sha1_ctx ctx;
1202                 sha1_init_ctx(&ctx);
1203                 sha1_process_bytes(&vcl->conf.guid, DDF_GUID_LEN, &ctx);
1204                 if (vcl->conf.sec_elmnt_count > 1)
1205                         sha1_process_bytes(&vcl->conf.sec_elmnt_seq, 1, &ctx);
1206                 sha1_finish_ctx(&ctx, buf);
1207                 memcpy(uuid, buf, sizeof(uuid));
1208         }
1209 }
1210
1211 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info);
1212
1213 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info)
1214 {
1215         struct ddf_super *ddf = st->sb;
1216
1217         if (ddf->currentconf) {
1218                 getinfo_super_ddf_bvd(st, info);
1219                 return;
1220         }
1221
1222         info->array.raid_disks    = __be16_to_cpu(ddf->phys->used_pdes);
1223         info->array.level         = LEVEL_CONTAINER;
1224         info->array.layout        = 0;
1225         info->array.md_minor      = -1;
1226         info->array.ctime         = DECADE + __be32_to_cpu(*(__u32*)
1227                                                          (ddf->anchor.guid+16));
1228         info->array.utime         = 0;
1229         info->array.chunk_size    = 0;
1230
1231
1232         info->disk.major = 0;
1233         info->disk.minor = 0;
1234         if (ddf->dlist) {
1235                 info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1236                 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1237
1238                 info->data_offset = __be64_to_cpu(ddf->phys->
1239                                           entries[info->disk.raid_disk].
1240                                           config_size);
1241                 info->component_size = ddf->dlist->size - info->data_offset;
1242         } else {
1243                 info->disk.number = -1;
1244 //              info->disk.raid_disk = find refnum in the table and use index;
1245         }
1246         info->disk.state = (1 << MD_DISK_SYNC);
1247
1248
1249         info->reshape_active = 0;
1250
1251         info->array.major_version = -1;
1252         info->array.minor_version = -2;
1253         strcpy(info->text_version, "ddf");
1254         info->safe_mode_delay = 0;
1255
1256 //      uuid_from_super_ddf(info->uuid, sbv);
1257
1258 //      info->name[] ?? ;
1259 }
1260
1261 static int rlq_to_layout(int rlq, int prl, int raiddisks);
1262
1263 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info)
1264 {
1265         struct ddf_super *ddf = st->sb;
1266         struct vcl *vc = ddf->currentconf;
1267         int cd = ddf->currentdev;
1268
1269         /* FIXME this returns BVD info - what if we want SVD ?? */
1270
1271         info->array.raid_disks    = __be16_to_cpu(vc->conf.prim_elmnt_count);
1272         info->array.level         = map_num1(ddf_level_num, vc->conf.prl);
1273         info->array.layout        = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
1274                                                   info->array.raid_disks);
1275         info->array.md_minor      = -1;
1276         info->array.ctime         = DECADE +
1277                 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
1278         info->array.utime         = DECADE + __be32_to_cpu(vc->conf.timestamp);
1279         info->array.chunk_size    = 512 << vc->conf.chunk_shift;
1280
1281         if (cd >= 0 && cd < ddf->mppe) {
1282                 info->data_offset         = __be64_to_cpu(vc->lba_offset[cd]);
1283                 if (vc->block_sizes)
1284                         info->component_size = vc->block_sizes[cd];
1285                 else
1286                         info->component_size = __be64_to_cpu(vc->conf.blocks);
1287         }
1288
1289         info->disk.major = 0;
1290         info->disk.minor = 0;
1291 //      info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1292 //      info->disk.raid_disk = find refnum in the table and use index;
1293 //      info->disk.state = ???;
1294
1295         info->container_member = ddf->currentconf->vcnum;
1296
1297         info->resync_start = 0;
1298         if (!(ddf->virt->entries[info->container_member].state
1299               & DDF_state_inconsistent)  &&
1300             (ddf->virt->entries[info->container_member].init_state
1301              & DDF_initstate_mask)
1302             == DDF_init_full)
1303                 info->resync_start = ~0ULL;
1304
1305         uuid_from_super_ddf(st, info->uuid);
1306
1307         info->container_member = atoi(st->subarray);
1308         info->array.major_version = -1;
1309         info->array.minor_version = -2;
1310         sprintf(info->text_version, "/%s/%s",
1311                 devnum2devname(st->container_dev),
1312                 st->subarray);
1313         info->safe_mode_delay = 200;
1314
1315 //      info->name[] ?? ;
1316 }
1317
1318
1319 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1320                             char *update,
1321                             char *devname, int verbose,
1322                             int uuid_set, char *homehost)
1323 {
1324         /* For 'assemble' and 'force' we need to return non-zero if any
1325          * change was made.  For others, the return value is ignored.
1326          * Update options are:
1327          *  force-one : This device looks a bit old but needs to be included,
1328          *        update age info appropriately.
1329          *  assemble: clear any 'faulty' flag to allow this device to
1330          *              be assembled.
1331          *  force-array: Array is degraded but being forced, mark it clean
1332          *         if that will be needed to assemble it.
1333          *
1334          *  newdev:  not used ????
1335          *  grow:  Array has gained a new device - this is currently for
1336          *              linear only
1337          *  resync: mark as dirty so a resync will happen.
1338          *  uuid:  Change the uuid of the array to match what is given
1339          *  homehost:  update the recorded homehost
1340          *  name:  update the name - preserving the homehost
1341          *  _reshape_progress: record new reshape_progress position.
1342          *
1343          * Following are not relevant for this version:
1344          *  sparc2.2 : update from old dodgey metadata
1345          *  super-minor: change the preferred_minor number
1346          *  summaries:  update redundant counters.
1347          */
1348         int rv = 0;
1349 //      struct ddf_super *ddf = st->sb;
1350 //      struct vd_config *vd = find_vdcr(ddf, info->container_member);
1351 //      struct virtual_entry *ve = find_ve(ddf);
1352
1353         /* we don't need to handle "force-*" or "assemble" as
1354          * there is no need to 'trick' the kernel.  We the metadata is
1355          * first updated to activate the array, all the implied modifications
1356          * will just happen.
1357          */
1358
1359         if (strcmp(update, "grow") == 0) {
1360                 /* FIXME */
1361         }
1362         if (strcmp(update, "resync") == 0) {
1363 //              info->resync_checkpoint = 0;
1364         }
1365         /* We ignore UUID updates as they make even less sense
1366          * with DDF
1367          */
1368         if (strcmp(update, "homehost") == 0) {
1369                 /* homehost is stored in controller->vendor_data,
1370                  * or it is when we are the vendor
1371                  */
1372 //              if (info->vendor_is_local)
1373 //                      strcpy(ddf->controller.vendor_data, homehost);
1374         }
1375         if (strcmp(update, "name") == 0) {
1376                 /* name is stored in virtual_entry->name */
1377 //              memset(ve->name, ' ', 16);
1378 //              strncpy(ve->name, info->name, 16);
1379         }
1380         if (strcmp(update, "_reshape_progress") == 0) {
1381                 /* We don't support reshape yet */
1382         }
1383
1384 //      update_all_csum(ddf);
1385
1386         return rv;
1387 }
1388
1389 static void make_header_guid(char *guid)
1390 {
1391         __u32 stamp;
1392         int rfd;
1393         /* Create a DDF Header of Virtual Disk GUID */
1394
1395         /* 24 bytes of fiction required.
1396          * first 8 are a 'vendor-id'  - "Linux-MD"
1397          * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1398          * Remaining 8 random number plus timestamp
1399          */
1400         memcpy(guid, T10, sizeof(T10));
1401         stamp = __cpu_to_be32(0xdeadbeef);
1402         memcpy(guid+8, &stamp, 4);
1403         stamp = __cpu_to_be32(0);
1404         memcpy(guid+12, &stamp, 4);
1405         stamp = __cpu_to_be32(time(0) - DECADE);
1406         memcpy(guid+16, &stamp, 4);
1407         rfd = open("/dev/urandom", O_RDONLY);
1408         if (rfd < 0 || read(rfd, &stamp, 4) != 4)
1409                 stamp = random();
1410         memcpy(guid+20, &stamp, 4);
1411         if (rfd >= 0) close(rfd);
1412 }
1413
1414 static int init_super_ddf_bvd(struct supertype *st,
1415                               mdu_array_info_t *info,
1416                               unsigned long long size,
1417                               char *name, char *homehost,
1418                               int *uuid);
1419
1420 static int init_super_ddf(struct supertype *st,
1421                           mdu_array_info_t *info,
1422                           unsigned long long size, char *name, char *homehost,
1423                           int *uuid)
1424 {
1425         /* This is primarily called by Create when creating a new array.
1426          * We will then get add_to_super called for each component, and then
1427          * write_init_super called to write it out to each device.
1428          * For DDF, Create can create on fresh devices or on a pre-existing
1429          * array.
1430          * To create on a pre-existing array a different method will be called.
1431          * This one is just for fresh drives.
1432          *
1433          * We need to create the entire 'ddf' structure which includes:
1434          *  DDF headers - these are easy.
1435          *  Controller data - a Sector describing this controller .. not that
1436          *                  this is a controller exactly.
1437          *  Physical Disk Record - one entry per device, so
1438          *                      leave plenty of space.
1439          *  Virtual Disk Records - again, just leave plenty of space.
1440          *                   This just lists VDs, doesn't give details
1441          *  Config records - describes the VDs that use this disk
1442          *  DiskData  - describes 'this' device.
1443          *  BadBlockManagement - empty
1444          *  Diag Space - empty
1445          *  Vendor Logs - Could we put bitmaps here?
1446          *
1447          */
1448         struct ddf_super *ddf;
1449         char hostname[17];
1450         int hostlen;
1451         int max_phys_disks, max_virt_disks;
1452         unsigned long long sector;
1453         int clen;
1454         int i;
1455         int pdsize, vdsize;
1456         struct phys_disk *pd;
1457         struct virtual_disk *vd;
1458
1459         if (!info) {
1460                 st->sb = NULL;
1461                 return 0;
1462         }
1463         if (st->sb)
1464                 return init_super_ddf_bvd(st, info, size, name, homehost,
1465                                           uuid);
1466
1467         posix_memalign((void**)&ddf, 512, sizeof(*ddf));
1468         memset(ddf, 0, sizeof(*ddf));
1469         ddf->dlist = NULL; /* no physical disks yet */
1470         ddf->conflist = NULL; /* No virtual disks yet */
1471
1472         /* At least 32MB *must* be reserved for the ddf.  So let's just
1473          * start 32MB from the end, and put the primary header there.
1474          * Don't do secondary for now.
1475          * We don't know exactly where that will be yet as it could be
1476          * different on each device.  To just set up the lengths.
1477          *
1478          */
1479
1480         ddf->anchor.magic = DDF_HEADER_MAGIC;
1481         make_header_guid(ddf->anchor.guid);
1482
1483         memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
1484         ddf->anchor.seq = __cpu_to_be32(1);
1485         ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1486         ddf->anchor.openflag = 0xFF;
1487         ddf->anchor.foreignflag = 0;
1488         ddf->anchor.enforcegroups = 0; /* Is this best?? */
1489         ddf->anchor.pad0 = 0xff;
1490         memset(ddf->anchor.pad1, 0xff, 12);
1491         memset(ddf->anchor.header_ext, 0xff, 32);
1492         ddf->anchor.primary_lba = ~(__u64)0;
1493         ddf->anchor.secondary_lba = ~(__u64)0;
1494         ddf->anchor.type = DDF_HEADER_ANCHOR;
1495         memset(ddf->anchor.pad2, 0xff, 3);
1496         ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1497         ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1498                                                   of 32M reserved.. */
1499         max_phys_disks = 1023;   /* Should be enough */
1500         ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1501         max_virt_disks = 255;
1502         ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1503         ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1504         ddf->max_part = 64;
1505         ddf->mppe = 256;
1506         ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
1507         ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1508         ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe);
1509         memset(ddf->anchor.pad3, 0xff, 54);
1510         /* controller sections is one sector long immediately
1511          * after the ddf header */
1512         sector = 1;
1513         ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1514         ddf->anchor.controller_section_length = __cpu_to_be32(1);
1515         sector += 1;
1516
1517         /* phys is 8 sectors after that */
1518         pdsize = ROUND_UP(sizeof(struct phys_disk) +
1519                           sizeof(struct phys_disk_entry)*max_phys_disks,
1520                           512);
1521         switch(pdsize/512) {
1522         case 2: case 8: case 32: case 128: case 512: break;
1523         default: abort();
1524         }
1525         ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1526         ddf->anchor.phys_section_length =
1527                 __cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1528         sector += pdsize/512;
1529
1530         /* virt is another 32 sectors */
1531         vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1532                           sizeof(struct virtual_entry) * max_virt_disks,
1533                           512);
1534         switch(vdsize/512) {
1535         case 2: case 8: case 32: case 128: case 512: break;
1536         default: abort();
1537         }
1538         ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1539         ddf->anchor.virt_section_length =
1540                 __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1541         sector += vdsize/512;
1542
1543         clen = ddf->conf_rec_len * (ddf->max_part+1);
1544         ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1545         ddf->anchor.config_section_length = __cpu_to_be32(clen);
1546         sector += clen;
1547
1548         ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1549         ddf->anchor.data_section_length = __cpu_to_be32(1);
1550         sector += 1;
1551
1552         ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1553         ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1554         ddf->anchor.diag_space_length = __cpu_to_be32(0);
1555         ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1556         ddf->anchor.vendor_length = __cpu_to_be32(0);
1557         ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1558
1559         memset(ddf->anchor.pad4, 0xff, 256);
1560
1561         memcpy(&ddf->primary, &ddf->anchor, 512);
1562         memcpy(&ddf->secondary, &ddf->anchor, 512);
1563
1564         ddf->primary.openflag = 1; /* I guess.. */
1565         ddf->primary.type = DDF_HEADER_PRIMARY;
1566
1567         ddf->secondary.openflag = 1; /* I guess.. */
1568         ddf->secondary.type = DDF_HEADER_SECONDARY;
1569
1570         ddf->active = &ddf->primary;
1571
1572         ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1573
1574         /* 24 more bytes of fiction required.
1575          * first 8 are a 'vendor-id'  - "Linux-MD"
1576          * Remaining 16 are serial number.... maybe a hostname would do?
1577          */
1578         memcpy(ddf->controller.guid, T10, sizeof(T10));
1579         gethostname(hostname, sizeof(hostname));
1580         hostname[sizeof(hostname) - 1] = 0;
1581         hostlen = strlen(hostname);
1582         memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1583         for (i = strlen(T10) ; i+hostlen < 24; i++)
1584                 ddf->controller.guid[i] = ' ';
1585
1586         ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1587         ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1588         ddf->controller.type.sub_vendor_id = 0;
1589         ddf->controller.type.sub_device_id = 0;
1590         memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1591         memset(ddf->controller.pad, 0xff, 8);
1592         memset(ddf->controller.vendor_data, 0xff, 448);
1593
1594         posix_memalign((void**)&pd, 512, pdsize);
1595         ddf->phys = pd;
1596         ddf->pdsize = pdsize;
1597
1598         memset(pd, 0xff, pdsize);
1599         memset(pd, 0, sizeof(*pd));
1600         pd->magic = DDF_PHYS_DATA_MAGIC;
1601         pd->used_pdes = __cpu_to_be16(0);
1602         pd->max_pdes = __cpu_to_be16(max_phys_disks);
1603         memset(pd->pad, 0xff, 52);
1604
1605         posix_memalign((void**)&vd, 512, vdsize);
1606         ddf->virt = vd;
1607         ddf->vdsize = vdsize;
1608         memset(vd, 0, vdsize);
1609         vd->magic = DDF_VIRT_RECORDS_MAGIC;
1610         vd->populated_vdes = __cpu_to_be16(0);
1611         vd->max_vdes = __cpu_to_be16(max_virt_disks);
1612         memset(vd->pad, 0xff, 52);
1613
1614         for (i=0; i<max_virt_disks; i++)
1615                 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1616
1617         st->sb = ddf;
1618         ddf->updates_pending = 1;
1619         return 1;
1620 }
1621
1622 static int all_ff(char *guid)
1623 {
1624         int i;
1625         for (i = 0; i < DDF_GUID_LEN; i++)
1626                 if (guid[i] != (char)0xff)
1627                         return 0;
1628         return 1;
1629 }
1630 static int chunk_to_shift(int chunksize)
1631 {
1632         return ffs(chunksize/512)-1;
1633 }
1634
1635 static int level_to_prl(int level)
1636 {
1637         switch (level) {
1638         case LEVEL_LINEAR: return DDF_CONCAT;
1639         case 0: return DDF_RAID0;
1640         case 1: return DDF_RAID1;
1641         case 4: return DDF_RAID4;
1642         case 5: return DDF_RAID5;
1643         case 6: return DDF_RAID6;
1644         default: return -1;
1645         }
1646 }
1647 static int layout_to_rlq(int level, int layout, int raiddisks)
1648 {
1649         switch(level) {
1650         case 0:
1651                 return DDF_RAID0_SIMPLE;
1652         case 1:
1653                 switch(raiddisks) {
1654                 case 2: return DDF_RAID1_SIMPLE;
1655                 case 3: return DDF_RAID1_MULTI;
1656                 default: return -1;
1657                 }
1658         case 4:
1659                 switch(layout) {
1660                 case 0: return DDF_RAID4_N;
1661                 }
1662                 break;
1663         case 5:
1664         case 6:
1665                 switch(layout) {
1666                 case ALGORITHM_LEFT_ASYMMETRIC:
1667                         return DDF_RAID5_N_RESTART;
1668                 case ALGORITHM_RIGHT_ASYMMETRIC:
1669                         if (level == 5)
1670                                 return DDF_RAID5_0_RESTART;
1671                         else
1672                                 return DDF_RAID6_0_RESTART;
1673                 case ALGORITHM_LEFT_SYMMETRIC:
1674                         return DDF_RAID5_N_CONTINUE;
1675                 case ALGORITHM_RIGHT_SYMMETRIC:
1676                         return -1; /* not mentioned in standard */
1677                 }
1678         }
1679         return -1;
1680 }
1681
1682 static int rlq_to_layout(int rlq, int prl, int raiddisks)
1683 {
1684         switch(prl) {
1685         case DDF_RAID0:
1686                 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1687         case DDF_RAID1:
1688                 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1689                              on raiddisks*/
1690         case DDF_RAID4:
1691                 switch(rlq) {
1692                 case DDF_RAID4_N:
1693                         return 0;
1694                 default:
1695                         /* not supported */
1696                         return -1; /* FIXME this isn't checked */
1697                 }
1698         case DDF_RAID5:
1699                 switch(rlq) {
1700                 case DDF_RAID5_N_RESTART:
1701                         return ALGORITHM_LEFT_ASYMMETRIC;
1702                 case DDF_RAID5_0_RESTART:
1703                         return ALGORITHM_RIGHT_ASYMMETRIC;
1704                 case DDF_RAID5_N_CONTINUE:
1705                         return ALGORITHM_LEFT_SYMMETRIC;
1706                 default:
1707                         return -1;
1708                 }
1709         case DDF_RAID6:
1710                 switch(rlq) {
1711                 case DDF_RAID5_N_RESTART:
1712                         return ALGORITHM_LEFT_ASYMMETRIC;
1713                 case DDF_RAID6_0_RESTART:
1714                         return ALGORITHM_RIGHT_ASYMMETRIC;
1715                 case DDF_RAID5_N_CONTINUE:
1716                         return ALGORITHM_LEFT_SYMMETRIC;
1717                 default:
1718                         return -1;
1719                 }
1720         }
1721         return -1;
1722 }
1723
1724 #ifndef MDASSEMBLE
1725 struct extent {
1726         unsigned long long start, size;
1727 };
1728 static int cmp_extent(const void *av, const void *bv)
1729 {
1730         const struct extent *a = av;
1731         const struct extent *b = bv;
1732         if (a->start < b->start)
1733                 return -1;
1734         if (a->start > b->start)
1735                 return 1;
1736         return 0;
1737 }
1738
1739 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
1740 {
1741         /* find a list of used extents on the give physical device
1742          * (dnum) of the given ddf.
1743          * Return a malloced array of 'struct extent'
1744
1745 FIXME ignore DDF_Legacy devices?
1746
1747          */
1748         struct extent *rv;
1749         int n = 0;
1750         int i, j;
1751
1752         rv = malloc(sizeof(struct extent) * (ddf->max_part + 2));
1753         if (!rv)
1754                 return NULL;
1755
1756         for (i = 0; i < ddf->max_part; i++) {
1757                 struct vcl *v = dl->vlist[i];
1758                 if (v == NULL)
1759                         continue;
1760                 for (j=0; j < v->conf.prim_elmnt_count; j++)
1761                         if (v->conf.phys_refnum[j] == dl->disk.refnum) {
1762                                 /* This device plays role 'j' in  'v'. */
1763                                 rv[n].start = __be64_to_cpu(v->lba_offset[j]);
1764                                 rv[n].size = __be64_to_cpu(v->conf.blocks);
1765                                 n++;
1766                                 break;
1767                         }
1768         }
1769         qsort(rv, n, sizeof(*rv), cmp_extent);
1770
1771         rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
1772         rv[n].size = 0;
1773         return rv;
1774 }
1775 #endif
1776
1777 static int init_super_ddf_bvd(struct supertype *st,
1778                               mdu_array_info_t *info,
1779                               unsigned long long size,
1780                               char *name, char *homehost,
1781                               int *uuid)
1782 {
1783         /* We are creating a BVD inside a pre-existing container.
1784          * so st->sb is already set.
1785          * We need to create a new vd_config and a new virtual_entry
1786          */
1787         struct ddf_super *ddf = st->sb;
1788         int venum;
1789         struct virtual_entry *ve;
1790         struct vcl *vcl;
1791         struct vd_config *vc;
1792
1793         if (__be16_to_cpu(ddf->virt->populated_vdes)
1794             >= __be16_to_cpu(ddf->virt->max_vdes)) {
1795                 fprintf(stderr, Name": This ddf already has the "
1796                         "maximum of %d virtual devices\n",
1797                         __be16_to_cpu(ddf->virt->max_vdes));
1798                 return 0;
1799         }
1800
1801         for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1802                 if (all_ff(ddf->virt->entries[venum].guid))
1803                         break;
1804         if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1805                 fprintf(stderr, Name ": Cannot find spare slot for "
1806                         "virtual disk - DDF is corrupt\n");
1807                 return 0;
1808         }
1809         ve = &ddf->virt->entries[venum];
1810
1811         /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1812          * timestamp, random number
1813          */
1814         make_header_guid(ve->guid);
1815         ve->unit = __cpu_to_be16(info->md_minor);
1816         ve->pad0 = 0xFFFF;
1817         ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1818         ve->type = 0;
1819         ve->state = DDF_state_degraded; /* Will be modified as devices are added */
1820         if (info->state & 1) /* clean */
1821                 ve->init_state = DDF_init_full;
1822         else
1823                 ve->init_state = DDF_init_not;
1824
1825         memset(ve->pad1, 0xff, 14);
1826         memset(ve->name, ' ', 16);
1827         if (name)
1828                 strncpy(ve->name, name, 16);
1829         ddf->virt->populated_vdes =
1830                 __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
1831
1832         /* Now create a new vd_config */
1833         posix_memalign((void**)&vcl, 512,
1834                        (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512));
1835         vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
1836         vcl->vcnum = venum;
1837         sprintf(st->subarray, "%d", venum);
1838         vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1839
1840         vc = &vcl->conf;
1841
1842         vc->magic = DDF_VD_CONF_MAGIC;
1843         memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
1844         vc->timestamp = __cpu_to_be32(time(0)-DECADE);
1845         vc->seqnum = __cpu_to_be32(1);
1846         memset(vc->pad0, 0xff, 24);
1847         vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
1848         vc->chunk_shift = chunk_to_shift(info->chunk_size);
1849         vc->prl = level_to_prl(info->level);
1850         vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
1851         vc->sec_elmnt_count = 1;
1852         vc->sec_elmnt_seq = 0;
1853         vc->srl = 0;
1854         vc->blocks = __cpu_to_be64(info->size * 2);
1855         vc->array_blocks = __cpu_to_be64(
1856                 calc_array_size(info->level, info->raid_disks, info->layout,
1857                                 info->chunk_size, info->size*2));
1858         memset(vc->pad1, 0xff, 8);
1859         vc->spare_refs[0] = 0xffffffff;
1860         vc->spare_refs[1] = 0xffffffff;
1861         vc->spare_refs[2] = 0xffffffff;
1862         vc->spare_refs[3] = 0xffffffff;
1863         vc->spare_refs[4] = 0xffffffff;
1864         vc->spare_refs[5] = 0xffffffff;
1865         vc->spare_refs[6] = 0xffffffff;
1866         vc->spare_refs[7] = 0xffffffff;
1867         memset(vc->cache_pol, 0, 8);
1868         vc->bg_rate = 0x80;
1869         memset(vc->pad2, 0xff, 3);
1870         memset(vc->pad3, 0xff, 52);
1871         memset(vc->pad4, 0xff, 192);
1872         memset(vc->v0, 0xff, 32);
1873         memset(vc->v1, 0xff, 32);
1874         memset(vc->v2, 0xff, 16);
1875         memset(vc->v3, 0xff, 16);
1876         memset(vc->vendor, 0xff, 32);
1877
1878         memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
1879         memset(vc->phys_refnum+(ddf->mppe * 4), 0x00, 8*ddf->mppe);
1880
1881         vcl->next = ddf->conflist;
1882         ddf->conflist = vcl;
1883         ddf->currentconf = vcl;
1884         ddf->updates_pending = 1;
1885         return 1;
1886 }
1887
1888 #ifndef MDASSEMBLE
1889 static void add_to_super_ddf_bvd(struct supertype *st,
1890                                  mdu_disk_info_t *dk, int fd, char *devname)
1891 {
1892         /* fd and devname identify a device with-in the ddf container (st).
1893          * dk identifies a location in the new BVD.
1894          * We need to find suitable free space in that device and update
1895          * the phys_refnum and lba_offset for the newly created vd_config.
1896          * We might also want to update the type in the phys_disk
1897          * section.
1898          */
1899         struct dl *dl;
1900         struct ddf_super *ddf = st->sb;
1901         struct vd_config *vc;
1902         __u64 *lba_offset;
1903         int working;
1904         int i;
1905         unsigned long long blocks, pos, esize;
1906         struct extent *ex;
1907
1908         for (dl = ddf->dlist; dl ; dl = dl->next)
1909                 if (dl->major == dk->major &&
1910                     dl->minor == dk->minor)
1911                         break;
1912         if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
1913                 return;
1914
1915         vc = &ddf->currentconf->conf;
1916         lba_offset = ddf->currentconf->lba_offset;
1917
1918         ex = get_extents(ddf, dl);
1919         if (!ex)
1920                 return;
1921
1922         i = 0; pos = 0;
1923         blocks = __be64_to_cpu(vc->blocks);
1924         if (ddf->currentconf->block_sizes)
1925                 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
1926
1927         do {
1928                 esize = ex[i].start - pos;
1929                 if (esize >= blocks)
1930                         break;
1931                 pos = ex[i].start + ex[i].size;
1932                 i++;
1933         } while (ex[i-1].size);
1934
1935         free(ex);
1936         if (esize < blocks)
1937                 return;
1938
1939         ddf->currentdev = dk->raid_disk;
1940         vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
1941         lba_offset[dk->raid_disk] = __cpu_to_be64(pos);
1942
1943         for (i=0; i < ddf->max_part ; i++)
1944                 if (dl->vlist[i] == NULL)
1945                         break;
1946         if (i == ddf->max_part)
1947                 return;
1948         dl->vlist[i] = ddf->currentconf;
1949
1950         dl->fd = fd;
1951         dl->devname = devname;
1952
1953         /* Check how many working raid_disks, and if we can mark
1954          * array as optimal yet
1955          */
1956         working = 0;
1957
1958         for (i=0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
1959                 if (vc->phys_refnum[i] != 0xffffffff)
1960                         working++;
1961
1962         /* Find which virtual_entry */
1963         i = ddf->currentconf->vcnum;
1964         if (working == __be16_to_cpu(vc->prim_elmnt_count))
1965                 ddf->virt->entries[i].state =
1966                         (ddf->virt->entries[i].state & ~DDF_state_mask)
1967                         | DDF_state_optimal;
1968
1969         if (vc->prl == DDF_RAID6 &&
1970             working+1 == __be16_to_cpu(vc->prim_elmnt_count))
1971                 ddf->virt->entries[i].state =
1972                         (ddf->virt->entries[i].state & ~DDF_state_mask)
1973                         | DDF_state_part_optimal;
1974
1975         ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
1976         ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
1977         ddf->updates_pending = 1;
1978 }
1979
1980 /* add a device to a container, either while creating it or while
1981  * expanding a pre-existing container
1982  */
1983 static void add_to_super_ddf(struct supertype *st,
1984                              mdu_disk_info_t *dk, int fd, char *devname)
1985 {
1986         struct ddf_super *ddf = st->sb;
1987         struct dl *dd;
1988         time_t now;
1989         struct tm *tm;
1990         unsigned long long size;
1991         struct phys_disk_entry *pde;
1992         int n, i;
1993         struct stat stb;
1994
1995         if (ddf->currentconf) {
1996                 add_to_super_ddf_bvd(st, dk, fd, devname);
1997                 return;
1998         }
1999
2000         /* This is device numbered dk->number.  We need to create
2001          * a phys_disk entry and a more detailed disk_data entry.
2002          */
2003         fstat(fd, &stb);
2004         posix_memalign((void**)&dd, 512,
2005                        sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part);
2006         dd->major = major(stb.st_rdev);
2007         dd->minor = minor(stb.st_rdev);
2008         dd->devname = devname;
2009         dd->fd = fd;
2010         dd->spare = NULL;
2011
2012         dd->disk.magic = DDF_PHYS_DATA_MAGIC;
2013         now = time(0);
2014         tm = localtime(&now);
2015         sprintf(dd->disk.guid, "%8s%04d%02d%02d",
2016                 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
2017         *(__u32*)(dd->disk.guid + 16) = random();
2018         *(__u32*)(dd->disk.guid + 20) = random();
2019
2020         do {
2021                 /* Cannot be bothered finding a CRC of some irrelevant details*/
2022                 dd->disk.refnum = random();
2023                 for (i = __be16_to_cpu(ddf->active->max_pd_entries) - 1;
2024                      i >= 0; i--)
2025                         if (ddf->phys->entries[i].refnum == dd->disk.refnum)
2026                                 break;
2027         } while (i >= 0);
2028
2029         dd->disk.forced_ref = 1;
2030         dd->disk.forced_guid = 1;
2031         memset(dd->disk.vendor, ' ', 32);
2032         memcpy(dd->disk.vendor, "Linux", 5);
2033         memset(dd->disk.pad, 0xff, 442);
2034         for (i = 0; i < ddf->max_part ; i++)
2035                 dd->vlist[i] = NULL;
2036
2037         n = __be16_to_cpu(ddf->phys->used_pdes);
2038         pde = &ddf->phys->entries[n];
2039         dd->pdnum = n;
2040
2041         if (st->update_tail) {
2042                 int len = (sizeof(struct phys_disk) +
2043                            sizeof(struct phys_disk_entry));
2044                 struct phys_disk *pd;
2045
2046                 pd = malloc(len);
2047                 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2048                 pd->used_pdes = __cpu_to_be16(n);
2049                 pde = &pd->entries[0];
2050                 dd->mdupdate = pd;
2051         } else {
2052                 n++;
2053                 ddf->phys->used_pdes = __cpu_to_be16(n);
2054         }
2055
2056         memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2057         pde->refnum = dd->disk.refnum;
2058         pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2059         pde->state = __cpu_to_be16(DDF_Online);
2060         get_dev_size(fd, NULL, &size);
2061         /* We are required to reserve 32Meg, and record the size in sectors */
2062         pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
2063         sprintf(pde->path, "%17.17s","Information: nil") ;
2064         memset(pde->pad, 0xff, 6);
2065
2066         dd->size = size >> 9;
2067         if (st->update_tail) {
2068                 dd->next = ddf->add_list;
2069                 ddf->add_list = dd;
2070         } else {
2071                 dd->next = ddf->dlist;
2072                 ddf->dlist = dd;
2073                 ddf->updates_pending = 1;
2074         }
2075 }
2076
2077 /*
2078  * This is the write_init_super method for a ddf container.  It is
2079  * called when creating a container or adding another device to a
2080  * container.
2081  */
2082
2083 static unsigned char null_conf[4096+512];
2084
2085 static int __write_init_super_ddf(struct supertype *st, int do_close)
2086 {
2087
2088         struct ddf_super *ddf = st->sb;
2089         int i;
2090         struct dl *d;
2091         int n_config;
2092         int conf_size;
2093
2094         unsigned long long size, sector;
2095
2096         for (d = ddf->dlist; d; d=d->next) {
2097                 int fd = d->fd;
2098
2099                 if (fd < 0)
2100                         continue;
2101
2102                 /* We need to fill in the primary, (secondary) and workspace
2103                  * lba's in the headers, set their checksums,
2104                  * Also checksum phys, virt....
2105                  *
2106                  * Then write everything out, finally the anchor is written.
2107                  */
2108                 get_dev_size(fd, NULL, &size);
2109                 size /= 512;
2110                 ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
2111                 ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
2112                 ddf->anchor.seq = __cpu_to_be32(1);
2113                 memcpy(&ddf->primary, &ddf->anchor, 512);
2114                 memcpy(&ddf->secondary, &ddf->anchor, 512);
2115
2116                 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
2117                 ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
2118                 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
2119
2120                 ddf->primary.openflag = 0;
2121                 ddf->primary.type = DDF_HEADER_PRIMARY;
2122
2123                 ddf->secondary.openflag = 0;
2124                 ddf->secondary.type = DDF_HEADER_SECONDARY;
2125
2126                 ddf->primary.crc = calc_crc(&ddf->primary, 512);
2127                 ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
2128
2129                 sector = size - 16*1024*2;
2130                 lseek64(fd, sector<<9, 0);
2131                 write(fd, &ddf->primary, 512);
2132
2133                 ddf->controller.crc = calc_crc(&ddf->controller, 512);
2134                 write(fd, &ddf->controller, 512);
2135
2136                 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
2137
2138                 write(fd, ddf->phys, ddf->pdsize);
2139
2140                 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
2141                 write(fd, ddf->virt, ddf->vdsize);
2142
2143                 /* Now write lots of config records. */
2144                 n_config = ddf->max_part;
2145                 conf_size = ddf->conf_rec_len * 512;
2146                 for (i = 0 ; i <= n_config ; i++) {
2147                         struct vcl *c = d->vlist[i];
2148                         if (i == n_config)
2149                                 c = (struct vcl*)d->spare;
2150
2151                         if (c) {
2152                                 c->conf.crc = calc_crc(&c->conf, conf_size);
2153                                 write(fd, &c->conf, conf_size);
2154                         } else {
2155                                 char *null_aligned = (char*)((((unsigned long)null_conf)+511)&~511UL);
2156                                 if (null_conf[0] != 0xff)
2157                                         memset(null_conf, 0xff, sizeof(null_conf));
2158                                 int togo = conf_size;
2159                                 while (togo > sizeof(null_conf)-512) {
2160                                         write(fd, null_aligned, sizeof(null_conf)-512);
2161                                         togo -= sizeof(null_conf)-512;
2162                                 }
2163                                 write(fd, null_aligned, togo);
2164                         }
2165                 }
2166                 d->disk.crc = calc_crc(&d->disk, 512);
2167                 write(fd, &d->disk, 512);
2168
2169                 /* Maybe do the same for secondary */
2170
2171                 lseek64(fd, (size-1)*512, SEEK_SET);
2172                 write(fd, &ddf->anchor, 512);
2173                 if (do_close) {
2174                         close(fd);
2175                         d->fd = -1;
2176                 }
2177         }
2178         return 1;
2179 }
2180
2181 static int write_init_super_ddf(struct supertype *st)
2182 {
2183
2184         if (st->update_tail) {
2185                 /* queue the virtual_disk and vd_config as metadata updates */
2186                 struct virtual_disk *vd;
2187                 struct vd_config *vc;
2188                 struct ddf_super *ddf = st->sb;
2189                 int len;
2190
2191                 if (!ddf->currentconf) {
2192                         int len = (sizeof(struct phys_disk) +
2193                                    sizeof(struct phys_disk_entry));
2194
2195                         /* adding a disk to the container. */
2196                         if (!ddf->add_list)
2197                                 return 0;
2198
2199                         append_metadata_update(st, ddf->add_list->mdupdate, len);
2200                         ddf->add_list->mdupdate = NULL;
2201                         return 0;
2202                 }
2203
2204                 /* Newly created VD */
2205
2206                 /* First the virtual disk.  We have a slightly fake header */
2207                 len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry);
2208                 vd = malloc(len);
2209                 *vd = *ddf->virt;
2210                 vd->entries[0] = ddf->virt->entries[ddf->currentconf->vcnum];
2211                 vd->populated_vdes = __cpu_to_be16(ddf->currentconf->vcnum);
2212                 append_metadata_update(st, vd, len);
2213
2214                 /* Then the vd_config */
2215                 len = ddf->conf_rec_len * 512;
2216                 vc = malloc(len);
2217                 memcpy(vc, &ddf->currentconf->conf, len);
2218                 append_metadata_update(st, vc, len);
2219
2220                 /* FIXME I need to close the fds! */
2221                 return 0;
2222         } else 
2223                 return __write_init_super_ddf(st, 1);
2224 }
2225
2226 #endif
2227
2228 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize)
2229 {
2230         /* We must reserve the last 32Meg */
2231         if (devsize <= 32*1024*2)
2232                 return 0;
2233         return devsize - 32*1024*2;
2234 }
2235
2236 #ifndef MDASSEMBLE
2237 static int
2238 validate_geometry_ddf_container(struct supertype *st,
2239                                 int level, int layout, int raiddisks,
2240                                 int chunk, unsigned long long size,
2241                                 char *dev, unsigned long long *freesize,
2242                                 int verbose);
2243
2244 static int validate_geometry_ddf_bvd(struct supertype *st,
2245                                      int level, int layout, int raiddisks,
2246                                      int chunk, unsigned long long size,
2247                                      char *dev, unsigned long long *freesize,
2248                                      int verbose);
2249
2250 static int validate_geometry_ddf(struct supertype *st,
2251                                  int level, int layout, int raiddisks,
2252                                  int chunk, unsigned long long size,
2253                                  char *dev, unsigned long long *freesize,
2254                                  int verbose)
2255 {
2256         int fd;
2257         struct mdinfo *sra;
2258         int cfd;
2259
2260         /* ddf potentially supports lots of things, but it depends on
2261          * what devices are offered (and maybe kernel version?)
2262          * If given unused devices, we will make a container.
2263          * If given devices in a container, we will make a BVD.
2264          * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2265          */
2266
2267         if (level == LEVEL_CONTAINER) {
2268                 /* Must be a fresh device to add to a container */
2269                 return validate_geometry_ddf_container(st, level, layout,
2270                                                        raiddisks, chunk,
2271                                                        size, dev, freesize,
2272                                                        verbose);
2273         }
2274
2275         if (st->sb) {
2276                 /* A container has already been opened, so we are
2277                  * creating in there.  Maybe a BVD, maybe an SVD.
2278                  * Should make a distinction one day.
2279                  */
2280                 return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
2281                                                  chunk, size, dev, freesize,
2282                                                  verbose);
2283         }
2284         if (!dev) {
2285                 /* Initial sanity check.  Exclude illegal levels. */
2286                 int i;
2287                 for (i=0; ddf_level_num[i].num1 != MAXINT; i++)
2288                         if (ddf_level_num[i].num2 == level)
2289                                 break;
2290                 if (ddf_level_num[i].num1 == MAXINT)
2291                         return 0;
2292                 /* Should check layout? etc */
2293                 return 1;
2294         }
2295
2296         /* This is the first device for the array.
2297          * If it is a container, we read it in and do automagic allocations,
2298          * no other devices should be given.
2299          * Otherwise it must be a member device of a container, and we
2300          * do manual allocation.
2301          * Later we should check for a BVD and make an SVD.
2302          */
2303         fd = open(dev, O_RDONLY|O_EXCL, 0);
2304         if (fd >= 0) {
2305                 sra = sysfs_read(fd, 0, GET_VERSION);
2306                 close(fd);
2307                 if (sra && sra->array.major_version == -1 &&
2308                     strcmp(sra->text_version, "ddf") == 0) {
2309
2310                         /* load super */
2311                         /* find space for 'n' devices. */
2312                         /* remember the devices */
2313                         /* Somehow return the fact that we have enough */
2314                 }
2315
2316                 if (verbose)
2317                         fprintf(stderr,
2318                                 Name ": ddf: Cannot create this array "
2319                                 "on device %s\n",
2320                                 dev);
2321                 return 0;
2322         }
2323         if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2324                 if (verbose)
2325                         fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2326                                 dev, strerror(errno));
2327                 return 0;
2328         }
2329         /* Well, it is in use by someone, maybe a 'ddf' container. */
2330         cfd = open_container(fd);
2331         if (cfd < 0) {
2332                 close(fd);
2333                 if (verbose)
2334                         fprintf(stderr, Name ": ddf: Cannot use %s: %s\n",
2335                                 dev, strerror(EBUSY));
2336                 return 0;
2337         }
2338         sra = sysfs_read(cfd, 0, GET_VERSION);
2339         close(fd);
2340         if (sra && sra->array.major_version == -1 &&
2341             strcmp(sra->text_version, "ddf") == 0) {
2342                 /* This is a member of a ddf container.  Load the container
2343                  * and try to create a bvd
2344                  */
2345                 struct ddf_super *ddf;
2346                 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL, 1) == 0) {
2347                         st->sb = ddf;
2348                         st->container_dev = fd2devnum(cfd);
2349                         close(cfd);
2350                         return validate_geometry_ddf_bvd(st, level, layout,
2351                                                          raiddisks, chunk, size,
2352                                                          dev, freesize,
2353                                                          verbose);
2354                 }
2355                 close(cfd);
2356         } else /* device may belong to a different container */
2357                 return 0;
2358
2359         return 1;
2360 }
2361
2362 static int
2363 validate_geometry_ddf_container(struct supertype *st,
2364                                 int level, int layout, int raiddisks,
2365                                 int chunk, unsigned long long size,
2366                                 char *dev, unsigned long long *freesize,
2367                                 int verbose)
2368 {
2369         int fd;
2370         unsigned long long ldsize;
2371
2372         if (level != LEVEL_CONTAINER)
2373                 return 0;
2374         if (!dev)
2375                 return 1;
2376
2377         fd = open(dev, O_RDONLY|O_EXCL, 0);
2378         if (fd < 0) {
2379                 if (verbose)
2380                         fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2381                                 dev, strerror(errno));
2382                 return 0;
2383         }
2384         if (!get_dev_size(fd, dev, &ldsize)) {
2385                 close(fd);
2386                 return 0;
2387         }
2388         close(fd);
2389
2390         *freesize = avail_size_ddf(st, ldsize >> 9);
2391
2392         return 1;
2393 }
2394
2395 static int validate_geometry_ddf_bvd(struct supertype *st,
2396                                      int level, int layout, int raiddisks,
2397                                      int chunk, unsigned long long size,
2398                                      char *dev, unsigned long long *freesize,
2399                                      int verbose)
2400 {
2401         struct stat stb;
2402         struct ddf_super *ddf = st->sb;
2403         struct dl *dl;
2404         unsigned long long pos = 0;
2405         unsigned long long maxsize;
2406         struct extent *e;
2407         int i;
2408         /* ddf/bvd supports lots of things, but not containers */
2409         if (level == LEVEL_CONTAINER)
2410                 return 0;
2411         /* We must have the container info already read in. */
2412         if (!ddf)
2413                 return 0;
2414
2415         if (!dev) {
2416                 /* General test:  make sure there is space for
2417                  * 'raiddisks' device extents of size 'size'.
2418                  */
2419                 unsigned long long minsize = size;
2420                 int dcnt = 0;
2421                 if (minsize == 0)
2422                         minsize = 8;
2423                 for (dl = ddf->dlist; dl ; dl = dl->next)
2424                 {
2425                         int found = 0;
2426                         pos = 0;
2427
2428                         i = 0;
2429                         e = get_extents(ddf, dl);
2430                         if (!e) continue;
2431                         do {
2432                                 unsigned long long esize;
2433                                 esize = e[i].start - pos;
2434                                 if (esize >= minsize)
2435                                         found = 1;
2436                                 pos = e[i].start + e[i].size;
2437                                 i++;
2438                         } while (e[i-1].size);
2439                         if (found)
2440                                 dcnt++;
2441                         free(e);
2442                 }
2443                 if (dcnt < raiddisks) {
2444                         if (verbose)
2445                                 fprintf(stderr,
2446                                         Name ": ddf: Not enough devices with "
2447                                         "space for this array (%d < %d)\n",
2448                                         dcnt, raiddisks);
2449                         return 0;
2450                 }
2451                 return 1;
2452         }
2453         /* This device must be a member of the set */
2454         if (stat(dev, &stb) < 0)
2455                 return 0;
2456         if ((S_IFMT & stb.st_mode) != S_IFBLK)
2457                 return 0;
2458         for (dl = ddf->dlist ; dl ; dl = dl->next) {
2459                 if (dl->major == major(stb.st_rdev) &&
2460                     dl->minor == minor(stb.st_rdev))
2461                         break;
2462         }
2463         if (!dl) {
2464                 if (verbose)
2465                         fprintf(stderr, Name ": ddf: %s is not in the "
2466                                 "same DDF set\n",
2467                                 dev);
2468                 return 0;
2469         }
2470         e = get_extents(ddf, dl);
2471         maxsize = 0;
2472         i = 0;
2473         if (e) do {
2474                 unsigned long long esize;
2475                 esize = e[i].start - pos;
2476                 if (esize >= maxsize)
2477                         maxsize = esize;
2478                 pos = e[i].start + e[i].size;
2479                 i++;
2480         } while (e[i-1].size);
2481         *freesize = maxsize;
2482         // FIXME here I am
2483
2484         return 1;
2485 }
2486
2487 static int load_super_ddf_all(struct supertype *st, int fd,
2488                               void **sbp, char *devname, int keep_fd)
2489 {
2490         struct mdinfo *sra;
2491         struct ddf_super *super;
2492         struct mdinfo *sd, *best = NULL;
2493         int bestseq = 0;
2494         int seq;
2495         char nm[20];
2496         int dfd;
2497
2498         sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
2499         if (!sra)
2500                 return 1;
2501         if (sra->array.major_version != -1 ||
2502             sra->array.minor_version != -2 ||
2503             strcmp(sra->text_version, "ddf") != 0)
2504                 return 1;
2505
2506         if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0)
2507                 return 1;
2508         memset(super, 0, sizeof(*super));
2509
2510         /* first, try each device, and choose the best ddf */
2511         for (sd = sra->devs ; sd ; sd = sd->next) {
2512                 int rv;
2513                 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2514                 dfd = dev_open(nm, O_RDONLY);
2515                 if (dfd < 0)
2516                         return 2;
2517                 rv = load_ddf_headers(dfd, super, NULL);
2518                 close(dfd);
2519                 if (rv == 0) {
2520                         seq = __be32_to_cpu(super->active->seq);
2521                         if (super->active->openflag)
2522                                 seq--;
2523                         if (!best || seq > bestseq) {
2524                                 bestseq = seq;
2525                                 best = sd;
2526                         }
2527                 }
2528         }
2529         if (!best)
2530                 return 1;
2531         /* OK, load this ddf */
2532         sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2533         dfd = dev_open(nm, O_RDONLY);
2534         if (dfd < 0)
2535                 return 1;
2536         load_ddf_headers(dfd, super, NULL);
2537         load_ddf_global(dfd, super, NULL);
2538         close(dfd);
2539         /* Now we need the device-local bits */
2540         for (sd = sra->devs ; sd ; sd = sd->next) {
2541                 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2542                 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2543                 if (dfd < 0)
2544                         return 2;
2545                 load_ddf_headers(dfd, super, NULL);
2546                 seq = load_ddf_local(dfd, super, NULL, keep_fd);
2547                 if (!keep_fd) close(dfd);
2548         }
2549         if (st->subarray[0]) {
2550                 struct vcl *v;
2551
2552                 for (v = super->conflist; v; v = v->next)
2553                         if (v->vcnum == atoi(st->subarray))
2554                                 super->currentconf = v;
2555                 if (!super->currentconf)
2556                         return 1;
2557         }
2558         *sbp = super;
2559         if (st->ss == NULL) {
2560                 st->ss = &super_ddf;
2561                 st->minor_version = 0;
2562                 st->max_devs = 512;
2563                 st->container_dev = fd2devnum(fd);
2564         }
2565         return 0;
2566 }
2567 #endif /* MDASSEMBLE */
2568
2569 static struct mdinfo *container_content_ddf(struct supertype *st)
2570 {
2571         /* Given a container loaded by load_super_ddf_all,
2572          * extract information about all the arrays into
2573          * an mdinfo tree.
2574          *
2575          * For each vcl in conflist: create an mdinfo, fill it in,
2576          *  then look for matching devices (phys_refnum) in dlist
2577          *  and create appropriate device mdinfo.
2578          */
2579         struct ddf_super *ddf = st->sb;
2580         struct mdinfo *rest = NULL;
2581         struct vcl *vc;
2582
2583         for (vc = ddf->conflist ; vc ; vc=vc->next)
2584         {
2585                 int i;
2586                 struct mdinfo *this;
2587                 this = malloc(sizeof(*this));
2588                 memset(this, 0, sizeof(*this));
2589                 this->next = rest;
2590                 rest = this;
2591
2592                 this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2593                 this->array.raid_disks =
2594                         __be16_to_cpu(vc->conf.prim_elmnt_count);
2595                 this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2596                                                    this->array.raid_disks);
2597                 this->array.md_minor      = -1;
2598                 this->array.major_version = -1;
2599                 this->array.minor_version = -2;
2600                 this->array.ctime         = DECADE +
2601                         __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
2602                 this->array.utime         = DECADE +
2603                         __be32_to_cpu(vc->conf.timestamp);
2604                 this->array.chunk_size    = 512 << vc->conf.chunk_shift;
2605
2606                 i = vc->vcnum;
2607                 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2608                     (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
2609                     DDF_init_full) {
2610                         this->array.state = 0;
2611                         this->resync_start = 0;
2612                 } else {
2613                         this->array.state = 1;
2614                         this->resync_start = ~0ULL;
2615                 }
2616                 memcpy(this->name, ddf->virt->entries[i].name, 32);
2617                 this->name[32]=0;
2618
2619                 memset(this->uuid, 0, sizeof(this->uuid));
2620                 this->component_size = __be64_to_cpu(vc->conf.blocks);
2621                 this->array.size = this->component_size / 2;
2622                 this->container_member = i;
2623
2624                 sprintf(this->text_version, "/%s/%d",
2625                         devnum2devname(st->container_dev),
2626                         this->container_member);
2627
2628                 for (i=0 ; i < ddf->mppe ; i++) {
2629                         struct mdinfo *dev;
2630                         struct dl *d;
2631
2632                         if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
2633                                 continue;
2634
2635                         this->array.working_disks++;
2636
2637                         for (d = ddf->dlist; d ; d=d->next)
2638                                 if (d->disk.refnum == vc->conf.phys_refnum[i])
2639                                         break;
2640                         if (d == NULL)
2641                                 break;
2642
2643                         dev = malloc(sizeof(*dev));
2644                         memset(dev, 0, sizeof(*dev));
2645                         dev->next = this->devs;
2646                         this->devs = dev;
2647
2648                         dev->disk.number = __be32_to_cpu(d->disk.refnum);
2649                         dev->disk.major = d->major;
2650                         dev->disk.minor = d->minor;
2651                         dev->disk.raid_disk = i;
2652                         dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2653
2654                         dev->events = __be32_to_cpu(ddf->primary.seq);
2655                         dev->data_offset = __be64_to_cpu(vc->lba_offset[i]);
2656                         dev->component_size = __be64_to_cpu(vc->conf.blocks);
2657                         if (d->devname)
2658                                 strcpy(dev->name, d->devname);
2659                 }
2660         }
2661         return rest;
2662 }
2663
2664 static int store_zero_ddf(struct supertype *st, int fd)
2665 {
2666         unsigned long long dsize;
2667         void *buf;
2668
2669         if (!get_dev_size(fd, NULL, &dsize))
2670                 return 1;
2671
2672         posix_memalign(&buf, 512, 512);
2673         memset(buf, 0, 512);
2674
2675         lseek64(fd, dsize-512, 0);
2676         write(fd, buf, 512);
2677         free(buf);
2678         return 0;
2679 }
2680
2681 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
2682 {
2683         /*
2684          * return:
2685          *  0 same, or first was empty, and second was copied
2686          *  1 second had wrong number
2687          *  2 wrong uuid
2688          *  3 wrong other info
2689          */
2690         struct ddf_super *first = st->sb;
2691         struct ddf_super *second = tst->sb;
2692
2693         if (!first) {
2694                 st->sb = tst->sb;
2695                 tst->sb = NULL;
2696                 return 0;
2697         }
2698
2699         if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
2700                 return 2;
2701
2702         /* FIXME should I look at anything else? */
2703         return 0;
2704 }
2705
2706 #ifndef MDASSEMBLE
2707 /*
2708  * A new array 'a' has been started which claims to be instance 'inst'
2709  * within container 'c'.
2710  * We need to confirm that the array matches the metadata in 'c' so
2711  * that we don't corrupt any metadata.
2712  */
2713 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
2714 {
2715         dprintf("ddf: open_new %s\n", inst);
2716         a->info.container_member = atoi(inst);
2717         return 0;
2718 }
2719
2720 /*
2721  * The array 'a' is to be marked clean in the metadata.
2722  * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2723  * clean up to the point (in sectors).  If that cannot be recorded in the
2724  * metadata, then leave it as dirty.
2725  *
2726  * For DDF, we need to clear the DDF_state_inconsistent bit in the
2727  * !global! virtual_disk.virtual_entry structure.
2728  */
2729 static int ddf_set_array_state(struct active_array *a, int consistent)
2730 {
2731         struct ddf_super *ddf = a->container->sb;
2732         int inst = a->info.container_member;
2733         int old = ddf->virt->entries[inst].state;
2734         if (consistent == 2) {
2735                 /* Should check if a recovery should be started FIXME */
2736                 consistent = 1;
2737                 if (a->resync_start != ~0ULL)
2738                         consistent = 0;
2739         }
2740         if (consistent)
2741                 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
2742         else
2743                 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
2744         if (old != ddf->virt->entries[inst].state)
2745                 ddf->updates_pending = 1;
2746
2747         old = ddf->virt->entries[inst].init_state;
2748         ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
2749         if (a->resync_start == ~0ULL)
2750                 ddf->virt->entries[inst].init_state |= DDF_init_full;
2751         else if (a->resync_start == 0)
2752                 ddf->virt->entries[inst].init_state |= DDF_init_not;
2753         else
2754                 ddf->virt->entries[inst].init_state |= DDF_init_quick;
2755         if (old != ddf->virt->entries[inst].init_state)
2756                 ddf->updates_pending = 1;
2757
2758         dprintf("ddf mark %d %s %llu\n", inst, consistent?"clean":"dirty",
2759                 a->resync_start);
2760         return consistent;
2761 }
2762
2763 /*
2764  * The state of each disk is stored in the global phys_disk structure
2765  * in phys_disk.entries[n].state.
2766  * This makes various combinations awkward.
2767  * - When a device fails in any array, it must be failed in all arrays
2768  *   that include a part of this device.
2769  * - When a component is rebuilding, we cannot include it officially in the
2770  *   array unless this is the only array that uses the device.
2771  *
2772  * So: when transitioning:
2773  *   Online -> failed,  just set failed flag.  monitor will propagate
2774  *   spare -> online,   the device might need to be added to the array.
2775  *   spare -> failed,   just set failed.  Don't worry if in array or not.
2776  */
2777 static void ddf_set_disk(struct active_array *a, int n, int state)
2778 {
2779         struct ddf_super *ddf = a->container->sb;
2780         int inst = a->info.container_member;
2781         struct vd_config *vc = find_vdcr(ddf, inst);
2782         int pd = find_phys(ddf, vc->phys_refnum[n]);
2783         int i, st, working;
2784
2785         if (vc == NULL) {
2786                 dprintf("ddf: cannot find instance %d!!\n", inst);
2787                 return;
2788         }
2789         if (pd < 0) {
2790                 /* disk doesn't currently exist. If it is now in_sync,
2791                  * insert it. */
2792                 if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
2793                         /* Find dev 'n' in a->info->devs, determine the
2794                          * ddf refnum, and set vc->phys_refnum and update
2795                          * phys->entries[]
2796                          */
2797                         /* FIXME */
2798                 }
2799         } else {
2800                 int old = ddf->phys->entries[pd].state;
2801                 if (state & DS_FAULTY)
2802                         ddf->phys->entries[pd].state  |= __cpu_to_be16(DDF_Failed);
2803                 if (state & DS_INSYNC) {
2804                         ddf->phys->entries[pd].state  |= __cpu_to_be16(DDF_Online);
2805                         ddf->phys->entries[pd].state  &= __cpu_to_be16(~DDF_Rebuilding);
2806                 }
2807                 if (old != ddf->phys->entries[pd].state)
2808                         ddf->updates_pending = 1;
2809         }
2810
2811         dprintf("ddf: set_disk %d to %x\n", n, state);
2812
2813         /* Now we need to check the state of the array and update
2814          * virtual_disk.entries[n].state.
2815          * It needs to be one of "optimal", "degraded", "failed".
2816          * I don't understand 'deleted' or 'missing'.
2817          */
2818         working = 0;
2819         for (i=0; i < a->info.array.raid_disks; i++) {
2820                 pd = find_phys(ddf, vc->phys_refnum[i]);
2821                 if (pd < 0)
2822                         continue;
2823                 st = __be16_to_cpu(ddf->phys->entries[pd].state);
2824                 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
2825                     == DDF_Online)
2826                         working++;
2827         }
2828         state = DDF_state_degraded;
2829         if (working == a->info.array.raid_disks)
2830                 state = DDF_state_optimal;
2831         else switch(vc->prl) {
2832         case DDF_RAID0:
2833         case DDF_CONCAT:
2834         case DDF_JBOD:
2835                 state = DDF_state_failed;
2836                 break;
2837         case DDF_RAID1:
2838                 if (working == 0)
2839                         state = DDF_state_failed;
2840                 break;
2841         case DDF_RAID4:
2842         case DDF_RAID5:
2843                 if (working < a->info.array.raid_disks-1)
2844                         state = DDF_state_failed;
2845                 break;
2846         case DDF_RAID6:
2847                 if (working < a->info.array.raid_disks-2)
2848                         state = DDF_state_failed;
2849                 else if (working == a->info.array.raid_disks-1)
2850                         state = DDF_state_part_optimal;
2851                 break;
2852         }
2853
2854         if (ddf->virt->entries[inst].state !=
2855             ((ddf->virt->entries[inst].state & ~DDF_state_mask)
2856              | state)) {
2857
2858                 ddf->virt->entries[inst].state =
2859                         (ddf->virt->entries[inst].state & ~DDF_state_mask)
2860                         | state;
2861                 ddf->updates_pending = 1;
2862         }
2863
2864 }
2865
2866 static void ddf_sync_metadata(struct supertype *st)
2867 {
2868
2869         /*
2870          * Write all data to all devices.
2871          * Later, we might be able to track whether only local changes
2872          * have been made, or whether any global data has been changed,
2873          * but ddf is sufficiently weird that it probably always
2874          * changes global data ....
2875          */
2876         struct ddf_super *ddf = st->sb;
2877         if (!ddf->updates_pending)
2878                 return;
2879         ddf->updates_pending = 0;
2880         __write_init_super_ddf(st, 0);
2881         dprintf("ddf: sync_metadata\n");
2882 }
2883
2884 static void ddf_process_update(struct supertype *st,
2885                                struct metadata_update *update)
2886 {
2887         /* Apply this update to the metadata.
2888          * The first 4 bytes are a DDF_*_MAGIC which guides
2889          * our actions.
2890          * Possible update are:
2891          *  DDF_PHYS_RECORDS_MAGIC
2892          *    Add a new physical device.  Changes to this record
2893          *    only happen implicitly.
2894          *    used_pdes is the device number.
2895          *  DDF_VIRT_RECORDS_MAGIC
2896          *    Add a new VD.  Possibly also change the 'access' bits.
2897          *    populated_vdes is the entry number.
2898          *  DDF_VD_CONF_MAGIC
2899          *    New or updated VD.  the VIRT_RECORD must already
2900          *    exist.  For an update, phys_refnum and lba_offset
2901          *    (at least) are updated, and the VD_CONF must
2902          *    be written to precisely those devices listed with
2903          *    a phys_refnum.
2904          *  DDF_SPARE_ASSIGN_MAGIC
2905          *    replacement Spare Assignment Record... but for which device?
2906          *
2907          * So, e.g.:
2908          *  - to create a new array, we send a VIRT_RECORD and
2909          *    a VD_CONF.  Then assemble and start the array.
2910          *  - to activate a spare we send a VD_CONF to add the phys_refnum
2911          *    and offset.  This will also mark the spare as active with
2912          *    a spare-assignment record.
2913          */
2914         struct ddf_super *ddf = st->sb;
2915         __u32 *magic = (__u32*)update->buf;
2916         struct phys_disk *pd;
2917         struct virtual_disk *vd;
2918         struct vd_config *vc;
2919         struct vcl *vcl;
2920         struct dl *dl;
2921         int mppe;
2922         int ent;
2923
2924         dprintf("Process update %x\n", *magic);
2925
2926         switch (*magic) {
2927         case DDF_PHYS_RECORDS_MAGIC:
2928
2929                 if (update->len != (sizeof(struct phys_disk) +
2930                                     sizeof(struct phys_disk_entry)))
2931                         return;
2932                 pd = (struct phys_disk*)update->buf;
2933
2934                 ent = __be16_to_cpu(pd->used_pdes);
2935                 if (ent >= __be16_to_cpu(ddf->phys->max_pdes))
2936                         return;
2937                 if (!all_ff(ddf->phys->entries[ent].guid))
2938                         return;
2939                 ddf->phys->entries[ent] = pd->entries[0];
2940                 ddf->phys->used_pdes = __cpu_to_be16(1 +
2941                                            __be16_to_cpu(ddf->phys->used_pdes));
2942                 ddf->updates_pending = 1;
2943                 if (ddf->add_list) {
2944                         struct active_array *a;
2945                         struct dl *al = ddf->add_list;
2946                         ddf->add_list = al->next;
2947
2948                         al->next = ddf->dlist;
2949                         ddf->dlist = al;
2950
2951                         /* As a device has been added, we should check
2952                          * for any degraded devices that might make
2953                          * use of this spare */
2954                         for (a = st->arrays ; a; a=a->next)
2955                                 a->check_degraded = 1;
2956                 }
2957                 break;
2958
2959         case DDF_VIRT_RECORDS_MAGIC:
2960
2961                 if (update->len != (sizeof(struct virtual_disk) +
2962                                     sizeof(struct virtual_entry)))
2963                         return;
2964                 vd = (struct virtual_disk*)update->buf;
2965
2966                 ent = __be16_to_cpu(vd->populated_vdes);
2967                 if (ent >= __be16_to_cpu(ddf->virt->max_vdes))
2968                         return;
2969                 if (!all_ff(ddf->virt->entries[ent].guid))
2970                         return;
2971                 ddf->virt->entries[ent] = vd->entries[0];
2972                 ddf->virt->populated_vdes = __cpu_to_be16(1 +
2973                               __be16_to_cpu(ddf->virt->populated_vdes));
2974                 ddf->updates_pending = 1;
2975                 break;
2976
2977         case DDF_VD_CONF_MAGIC:
2978                 dprintf("len %d %d\n", update->len, ddf->conf_rec_len);
2979
2980                 mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries);
2981                 if (update->len != ddf->conf_rec_len * 512)
2982                         return;
2983                 vc = (struct vd_config*)update->buf;
2984                 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2985                         if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
2986                                 break;
2987                 dprintf("vcl = %p\n", vcl);
2988                 if (vcl) {
2989                         /* An update, just copy the phys_refnum and lba_offset
2990                          * fields
2991                          */
2992                         memcpy(vcl->conf.phys_refnum, vc->phys_refnum,
2993                                mppe * (sizeof(__u32) + sizeof(__u64)));
2994                 } else {
2995                         /* A new VD_CONF */
2996                         vcl = update->space;
2997                         update->space = NULL;
2998                         vcl->next = ddf->conflist;
2999                         memcpy(&vcl->conf, vc, update->len);
3000                         vcl->lba_offset = (__u64*)
3001                                 &vcl->conf.phys_refnum[mppe];
3002                         ddf->conflist = vcl;
3003                 }
3004                 /* Now make sure vlist is correct for each dl. */
3005                 for (dl = ddf->dlist; dl; dl = dl->next) {
3006                         int dn;
3007                         int vn = 0;
3008                         for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
3009                                 for (dn=0; dn < ddf->mppe ; dn++)
3010                                         if (vcl->conf.phys_refnum[dn] ==
3011                                             dl->disk.refnum) {
3012                                                 dprintf("dev %d has %p at %d\n",
3013                                                         dl->pdnum, vcl, vn);
3014                                                 dl->vlist[vn++] = vcl;
3015                                                 break;
3016                                         }
3017                         while (vn < ddf->max_part)
3018                                 dl->vlist[vn++] = NULL;
3019                         if (dl->vlist[0]) {
3020                                 ddf->phys->entries[dl->pdnum].type &=
3021                                         ~__cpu_to_be16(DDF_Global_Spare);
3022                                 ddf->phys->entries[dl->pdnum].type |=
3023                                         __cpu_to_be16(DDF_Active_in_VD);
3024                         }
3025                         if (dl->spare) {
3026                                 ddf->phys->entries[dl->pdnum].type &=
3027                                         ~__cpu_to_be16(DDF_Global_Spare);
3028                                 ddf->phys->entries[dl->pdnum].type |=
3029                                         __cpu_to_be16(DDF_Spare);
3030                         }
3031                         if (!dl->vlist[0] && !dl->spare) {
3032                                 ddf->phys->entries[dl->pdnum].type |=
3033                                         __cpu_to_be16(DDF_Global_Spare);
3034                                 ddf->phys->entries[dl->pdnum].type &=
3035                                         ~__cpu_to_be16(DDF_Spare |
3036                                                        DDF_Active_in_VD);
3037                         }
3038                 }
3039                 ddf->updates_pending = 1;
3040                 break;
3041         case DDF_SPARE_ASSIGN_MAGIC:
3042         default: break;
3043         }
3044 }
3045
3046 static void ddf_prepare_update(struct supertype *st,
3047                                struct metadata_update *update)
3048 {
3049         /* This update arrived at managemon.
3050          * We are about to pass it to monitor.
3051          * If a malloc is needed, do it here.
3052          */
3053         struct ddf_super *ddf = st->sb;
3054         __u32 *magic = (__u32*)update->buf;
3055         if (*magic == DDF_VD_CONF_MAGIC)
3056                 posix_memalign(&update->space, 512,
3057                                offsetof(struct vcl, conf)
3058                                + ddf->conf_rec_len * 512);
3059 }
3060
3061 /*
3062  * Check if the array 'a' is degraded but not failed.
3063  * If it is, find as many spares as are available and needed and
3064  * arrange for their inclusion.
3065  * We only choose devices which are not already in the array,
3066  * and prefer those with a spare-assignment to this array.
3067  * otherwise we choose global spares - assuming always that
3068  * there is enough room.
3069  * For each spare that we assign, we return an 'mdinfo' which
3070  * describes the position for the device in the array.
3071  * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
3072  * the new phys_refnum and lba_offset values.
3073  *
3074  * Only worry about BVDs at the moment.
3075  */
3076 static struct mdinfo *ddf_activate_spare(struct active_array *a,
3077                                          struct metadata_update **updates)
3078 {
3079         int working = 0;
3080         struct mdinfo *d;
3081         struct ddf_super *ddf = a->container->sb;
3082         int global_ok = 0;
3083         struct mdinfo *rv = NULL;
3084         struct mdinfo *di;
3085         struct metadata_update *mu;
3086         struct dl *dl;
3087         int i;
3088         struct vd_config *vc;
3089         __u64 *lba;
3090
3091         for (d = a->info.devs ; d ; d = d->next) {
3092                 if ((d->curr_state & DS_FAULTY) &&
3093                         d->state_fd >= 0)
3094                         /* wait for Removal to happen */
3095                         return NULL;
3096                 if (d->state_fd >= 0)
3097                         working ++;
3098         }
3099
3100         dprintf("ddf_activate: working=%d (%d) level=%d\n", working, a->info.array.raid_disks,
3101                 a->info.array.level);
3102         if (working == a->info.array.raid_disks)
3103                 return NULL; /* array not degraded */
3104         switch (a->info.array.level) {
3105         case 1:
3106                 if (working == 0)
3107                         return NULL; /* failed */
3108                 break;
3109         case 4:
3110         case 5:
3111                 if (working < a->info.array.raid_disks - 1)
3112                         return NULL; /* failed */
3113                 break;
3114         case 6:
3115                 if (working < a->info.array.raid_disks - 2)
3116                         return NULL; /* failed */
3117                 break;
3118         default: /* concat or stripe */
3119                 return NULL; /* failed */
3120         }
3121
3122         /* For each slot, if it is not working, find a spare */
3123         dl = ddf->dlist;
3124         for (i = 0; i < a->info.array.raid_disks; i++) {
3125                 for (d = a->info.devs ; d ; d = d->next)
3126                         if (d->disk.raid_disk == i)
3127                                 break;
3128                 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
3129                 if (d && (d->state_fd >= 0))
3130                         continue;
3131
3132                 /* OK, this device needs recovery.  Find a spare */
3133         again:
3134                 for ( ; dl ; dl = dl->next) {
3135                         unsigned long long esize;
3136                         unsigned long long pos;
3137                         struct mdinfo *d2;
3138                         int is_global = 0;
3139                         int is_dedicated = 0;
3140                         struct extent *ex;
3141                         int j;
3142                         /* If in this array, skip */
3143                         for (d2 = a->info.devs ; d2 ; d2 = d2->next)
3144                                 if (d2->disk.major == dl->major &&
3145                                     d2->disk.minor == dl->minor) {
3146                                         dprintf("%x:%x already in array\n", dl->major, dl->minor);
3147                                         break;
3148                                 }
3149                         if (d2)
3150                                 continue;
3151                         if (ddf->phys->entries[dl->pdnum].type &
3152                             __cpu_to_be16(DDF_Spare)) {
3153                                 /* Check spare assign record */
3154                                 if (dl->spare) {
3155                                         if (dl->spare->type & DDF_spare_dedicated) {
3156                                                 /* check spare_ents for guid */
3157                                                 for (j = 0 ;
3158                                                      j < __be16_to_cpu(dl->spare->populated);
3159                                                      j++) {
3160                                                         if (memcmp(dl->spare->spare_ents[j].guid,
3161                                                                    ddf->virt->entries[a->info.container_member].guid,
3162                                                                    DDF_GUID_LEN) == 0)
3163                                                                 is_dedicated = 1;
3164                                                 }
3165                                         } else
3166                                                 is_global = 1;
3167                                 }
3168                         } else if (ddf->phys->entries[dl->pdnum].type &
3169                                    __cpu_to_be16(DDF_Global_Spare)) {
3170                                 is_global = 1;
3171                         }
3172                         if ( ! (is_dedicated ||
3173                                 (is_global && global_ok))) {
3174                                 dprintf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
3175                                        is_dedicated, is_global);
3176                                 continue;
3177                         }
3178
3179                         /* We are allowed to use this device - is there space?
3180                          * We need a->info.component_size sectors */
3181                         ex = get_extents(ddf, dl);
3182                         if (!ex) {
3183                                 dprintf("cannot get extents\n");
3184                                 continue;
3185                         }
3186                         j = 0; pos = 0;
3187                         esize = 0;
3188
3189                         do {
3190                                 esize = ex[j].start - pos;
3191                                 if (esize >= a->info.component_size)
3192                                         break;
3193                                 pos = ex[i].start + ex[i].size;
3194                                 i++;
3195                         } while (ex[i-1].size);
3196
3197                         free(ex);
3198                         if (esize < a->info.component_size) {
3199                                 dprintf("%x:%x has no room: %llu %llu\n", dl->major, dl->minor,
3200                                         esize, a->info.component_size);
3201                                 /* No room */
3202                                 continue;
3203                         }
3204
3205                         /* Cool, we have a device with some space at pos */
3206                         di = malloc(sizeof(*di));
3207                         memset(di, 0, sizeof(*di));
3208                         di->disk.number = i;
3209                         di->disk.raid_disk = i;
3210                         di->disk.major = dl->major;
3211                         di->disk.minor = dl->minor;
3212                         di->disk.state = 0;
3213                         di->data_offset = pos;
3214                         di->component_size = a->info.component_size;
3215                         di->container_member = dl->pdnum;
3216                         di->next = rv;
3217                         rv = di;
3218                         dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3219                                 i, pos);
3220
3221                         break;
3222                 }
3223                 if (!dl && ! global_ok) {
3224                         /* not enough dedicated spares, try global */
3225                         global_ok = 1;
3226                         dl = ddf->dlist;
3227                         goto again;
3228                 }
3229         }
3230
3231         if (!rv)
3232                 /* No spares found */
3233                 return rv;
3234         /* Now 'rv' has a list of devices to return.
3235          * Create a metadata_update record to update the
3236          * phys_refnum and lba_offset values
3237          */
3238         mu = malloc(sizeof(*mu));
3239         mu->buf = malloc(ddf->conf_rec_len * 512);
3240         posix_memalign(&mu->space, 512, sizeof(struct vcl));
3241         mu->len = ddf->conf_rec_len;
3242         mu->next = *updates;
3243         vc = find_vdcr(ddf, a->info.container_member);
3244         memcpy(mu->buf, vc, ddf->conf_rec_len * 512);
3245
3246         vc = (struct vd_config*)mu->buf;
3247         lba = (__u64*)&vc->phys_refnum[ddf->mppe];
3248         for (di = rv ; di ; di = di->next) {
3249                 vc->phys_refnum[di->disk.raid_disk] =
3250                         ddf->phys->entries[dl->pdnum].refnum;
3251                 lba[di->disk.raid_disk] = di->data_offset;
3252         }
3253         *updates = mu;
3254         return rv;
3255 }
3256 #endif /* MDASSEMBLE */
3257
3258 struct superswitch super_ddf = {
3259 #ifndef MDASSEMBLE
3260         .examine_super  = examine_super_ddf,
3261         .brief_examine_super = brief_examine_super_ddf,
3262         .detail_super   = detail_super_ddf,
3263         .brief_detail_super = brief_detail_super_ddf,
3264         .validate_geometry = validate_geometry_ddf,
3265         .write_init_super = write_init_super_ddf,
3266         .add_to_super   = add_to_super_ddf,
3267 #endif
3268         .match_home     = match_home_ddf,
3269         .uuid_from_super= uuid_from_super_ddf,
3270         .getinfo_super  = getinfo_super_ddf,
3271         .update_super   = update_super_ddf,
3272
3273         .avail_size     = avail_size_ddf,
3274
3275         .compare_super  = compare_super_ddf,
3276
3277         .load_super     = load_super