cb7f67c4fdeb9e00a12c4dee2b6109c9193fed76
[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                 struct vcl *vlist[0]; /* max_part in size */
430         } *dlist;
431 };
432
433 #ifndef offsetof
434 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
435 #endif
436
437
438 static int calc_crc(void *buf, int len)
439 {
440         /* crcs are always at the same place as in the ddf_header */
441         struct ddf_header *ddf = buf;
442         __u32 oldcrc = ddf->crc;
443         __u32 newcrc;
444         ddf->crc = 0xffffffff;
445
446         newcrc = crc32(0, buf, len);
447         ddf->crc = oldcrc;
448         return newcrc;
449 }
450
451 static int load_ddf_header(int fd, unsigned long long lba,
452                            unsigned long long size,
453                            int type,
454                            struct ddf_header *hdr, struct ddf_header *anchor)
455 {
456         /* read a ddf header (primary or secondary) from fd/lba
457          * and check that it is consistent with anchor
458          * Need to check:
459          *   magic, crc, guid, rev, and LBA's header_type, and
460          *  everything after header_type must be the same
461          */
462         if (lba >= size-1)
463                 return 0;
464
465         if (lseek64(fd, lba<<9, 0) < 0)
466                 return 0;
467
468         if (read(fd, hdr, 512) != 512)
469                 return 0;
470
471         if (hdr->magic != DDF_HEADER_MAGIC)
472                 return 0;
473         if (calc_crc(hdr, 512) != hdr->crc)
474                 return 0;
475         if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
476             memcmp(anchor->revision, hdr->revision, 8) != 0 ||
477             anchor->primary_lba != hdr->primary_lba ||
478             anchor->secondary_lba != hdr->secondary_lba ||
479             hdr->type != type ||
480             memcmp(anchor->pad2, hdr->pad2, 512 -
481                    offsetof(struct ddf_header, pad2)) != 0)
482                 return 0;
483
484         /* Looks good enough to me... */
485         return 1;
486 }
487
488 static void *load_section(int fd, struct ddf_super *super, void *buf,
489                           __u32 offset_be, __u32 len_be, int check)
490 {
491         unsigned long long offset = __be32_to_cpu(offset_be);
492         unsigned long long len = __be32_to_cpu(len_be);
493         int dofree = (buf == NULL);
494
495         if (check)
496                 if (len != 2 && len != 8 && len != 32
497                     && len != 128 && len != 512)
498                         return NULL;
499
500         if (len > 1024)
501                 return NULL;
502         if (buf) {
503                 /* All pre-allocated sections are a single block */
504                 if (len != 1)
505                         return NULL;
506         } else {
507                 posix_memalign(&buf, 512, len<<9);
508         }
509
510         if (!buf)
511                 return NULL;
512
513         if (super->active->type == 1)
514                 offset += __be64_to_cpu(super->active->primary_lba);
515         else
516                 offset += __be64_to_cpu(super->active->secondary_lba);
517
518         if (lseek64(fd, offset<<9, 0) != (offset<<9)) {
519                 if (dofree)
520                         free(buf);
521                 return NULL;
522         }
523         if (read(fd, buf, len<<9) != (len<<9)) {
524                 if (dofree)
525                         free(buf);
526                 return NULL;
527         }
528         return buf;
529 }
530
531 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
532 {
533         unsigned long long dsize;
534
535         get_dev_size(fd, NULL, &dsize);
536
537         if (lseek64(fd, dsize-512, 0) < 0) {
538                 if (devname)
539                         fprintf(stderr,
540                                 Name": Cannot seek to anchor block on %s: %s\n",
541                                 devname, strerror(errno));
542                 return 1;
543         }
544         if (read(fd, &super->anchor, 512) != 512) {
545                 if (devname)
546                         fprintf(stderr,
547                                 Name ": Cannot read anchor block on %s: %s\n",
548                                 devname, strerror(errno));
549                 return 1;
550         }
551         if (super->anchor.magic != DDF_HEADER_MAGIC) {
552                 if (devname)
553                         fprintf(stderr, Name ": no DDF anchor found on %s\n",
554                                 devname);
555                 return 2;
556         }
557         if (calc_crc(&super->anchor, 512) != super->anchor.crc) {
558                 if (devname)
559                         fprintf(stderr, Name ": bad CRC on anchor on %s\n",
560                                 devname);
561                 return 2;
562         }
563         if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
564             memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
565                 if (devname)
566                         fprintf(stderr, Name ": can only support super revision"
567                                 " %.8s and earlier, not %.8s on %s\n",
568                                 DDF_REVISION_2, super->anchor.revision,devname);
569                 return 2;
570         }
571         if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba),
572                             dsize >> 9,  1,
573                             &super->primary, &super->anchor) == 0) {
574                 if (devname)
575                         fprintf(stderr,
576                                 Name ": Failed to load primary DDF header "
577                                 "on %s\n", devname);
578                 return 2;
579         }
580         super->active = &super->primary;
581         if (load_ddf_header(fd, __be64_to_cpu(super->anchor.secondary_lba),
582                             dsize >> 9,  2,
583                             &super->secondary, &super->anchor)) {
584                 if ((__be32_to_cpu(super->primary.seq)
585                      < __be32_to_cpu(super->secondary.seq) &&
586                      !super->secondary.openflag)
587                     || (__be32_to_cpu(super->primary.seq)
588                         == __be32_to_cpu(super->secondary.seq) &&
589                         super->primary.openflag && !super->secondary.openflag)
590                         )
591                         super->active = &super->secondary;
592         }
593         return 0;
594 }
595
596 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
597 {
598         void *ok;
599         ok = load_section(fd, super, &super->controller,
600                           super->active->controller_section_offset,
601                           super->active->controller_section_length,
602                           0);
603         super->phys = load_section(fd, super, NULL,
604                                    super->active->phys_section_offset,
605                                    super->active->phys_section_length,
606                                    1);
607         super->pdsize = __be32_to_cpu(super->active->phys_section_length) * 512;
608
609         super->virt = load_section(fd, super, NULL,
610                                    super->active->virt_section_offset,
611                                    super->active->virt_section_length,
612                                    1);
613         super->vdsize = __be32_to_cpu(super->active->virt_section_length) * 512;
614         if (!ok ||
615             !super->phys ||
616             !super->virt) {
617                 free(super->phys);
618                 free(super->virt);
619                 super->phys = NULL;
620                 super->virt = NULL;
621                 return 2;
622         }
623         super->conflist = NULL;
624         super->dlist = NULL;
625
626         super->max_part = __be16_to_cpu(super->active->max_partitions);
627         super->mppe = __be16_to_cpu(super->active->max_primary_element_entries);
628         super->conf_rec_len = __be16_to_cpu(super->active->config_record_len);
629         return 0;
630 }
631
632 static int load_ddf_local(int fd, struct ddf_super *super,
633                           char *devname, int keep)
634 {
635         struct dl *dl;
636         struct stat stb;
637         char *conf;
638         int i;
639         int vnum;
640         int max_virt_disks = __be16_to_cpu(super->active->max_vd_entries);
641         unsigned long long dsize;
642
643         /* First the local disk info */
644         posix_memalign((void**)&dl, 512,
645                        sizeof(*dl) +
646                        (super->max_part) * sizeof(dl->vlist[0]));
647
648         load_section(fd, super, &dl->disk,
649                      super->active->data_section_offset,
650                      super->active->data_section_length,
651                      0);
652         dl->devname = devname ? strdup(devname) : NULL;
653
654         fstat(fd, &stb);
655         dl->major = major(stb.st_rdev);
656         dl->minor = minor(stb.st_rdev);
657         dl->next = super->dlist;
658         dl->fd = keep ? fd : -1;
659
660         dl->size = 0;
661         if (get_dev_size(fd, devname, &dsize))
662                 dl->size = dsize >> 9;
663         dl->spare = NULL;
664         for (i=0 ; i < super->max_part ; i++)
665                 dl->vlist[i] = NULL;
666         super->dlist = dl;
667         dl->pdnum = -1;
668         for (i=0; i < __be16_to_cpu(super->active->max_pd_entries); i++)
669                 if (memcmp(super->phys->entries[i].guid,
670                            dl->disk.guid, DDF_GUID_LEN) == 0)
671                         dl->pdnum = i;
672
673         /* Now the config list. */
674         /* 'conf' is an array of config entries, some of which are
675          * probably invalid.  Those which are good need to be copied into
676          * the conflist
677          */
678
679         conf = load_section(fd, super, NULL,
680                             super->active->config_section_offset,
681                             super->active->config_section_length,
682                             0);
683
684         vnum = 0;
685         for (i = 0;
686              i < __be32_to_cpu(super->active->config_section_length);
687              i += super->conf_rec_len) {
688                 struct vd_config *vd =
689                         (struct vd_config *)((char*)conf + i*512);
690                 struct vcl *vcl;
691
692                 if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) {
693                         if (dl->spare)
694                                 continue;
695                         posix_memalign((void**)&dl->spare, 512,
696                                        super->conf_rec_len*512);
697                         memcpy(dl->spare, vd, super->conf_rec_len*512);
698                         continue;
699                 }
700                 if (vd->magic != DDF_VD_CONF_MAGIC)
701                         continue;
702                 for (vcl = super->conflist; vcl; vcl = vcl->next) {
703                         if (memcmp(vcl->conf.guid,
704                                    vd->guid, DDF_GUID_LEN) == 0)
705                                 break;
706                 }
707
708                 if (vcl) {
709                         dl->vlist[vnum++] = vcl;
710                         if (__be32_to_cpu(vd->seqnum) <=
711                             __be32_to_cpu(vcl->conf.seqnum))
712                                 continue;
713                 } else {
714                         posix_memalign((void**)&vcl, 512,
715                                        (super->conf_rec_len*512 +
716                                         offsetof(struct vcl, conf)));
717                         vcl->next = super->conflist;
718                         vcl->block_sizes = NULL; /* FIXME not for CONCAT */
719                         super->conflist = vcl;
720                         dl->vlist[vnum++] = vcl;
721                 }
722                 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
723                 vcl->lba_offset = (__u64*)
724                         &vcl->conf.phys_refnum[super->mppe];
725
726                 for (i=0; i < max_virt_disks ; i++)
727                         if (memcmp(super->virt->entries[i].guid,
728                                    vcl->conf.guid, DDF_GUID_LEN)==0)
729                                 break;
730                 if (i < max_virt_disks)
731                         vcl->vcnum = i;
732         }
733         free(conf);
734
735         return 0;
736 }
737
738 #ifndef MDASSEMBLE
739 static int load_super_ddf_all(struct supertype *st, int fd,
740                               void **sbp, char *devname, int keep_fd);
741 #endif
742 static int load_super_ddf(struct supertype *st, int fd,
743                           char *devname)
744 {
745         unsigned long long dsize;
746         struct ddf_super *super;
747         int rv;
748
749 #ifndef MDASSEMBLE
750         /* if 'fd' is a container, load metadata from all the devices */
751         if (load_super_ddf_all(st, fd, &st->sb, devname, 1) == 0)
752                 return 0;
753 #endif
754         if (st->subarray[0])
755                 return 1; /* FIXME Is this correct */
756
757         if (get_dev_size(fd, devname, &dsize) == 0)
758                 return 1;
759
760         /* 32M is a lower bound */
761         if (dsize <= 32*1024*1024) {
762                 if (devname) {
763                         fprintf(stderr,
764                                 Name ": %s is too small for ddf: "
765                                 "size is %llu sectors.\n",
766                                 devname, dsize>>9);
767                         return 1;
768                 }
769         }
770         if (dsize & 511) {
771                 if (devname) {
772                         fprintf(stderr,
773                                 Name ": %s is an odd size for ddf: "
774                                 "size is %llu bytes.\n",
775                                 devname, dsize);
776                         return 1;
777                 }
778         }
779
780         if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) {
781                 fprintf(stderr, Name ": malloc of %zu failed.\n",
782                         sizeof(*super));
783                 return 1;
784         }
785         memset(super, 0, sizeof(*super));
786
787         rv = load_ddf_headers(fd, super, devname);
788         if (rv) {
789                 free(super);
790                 return rv;
791         }
792
793         /* Have valid headers and have chosen the best. Let's read in the rest*/
794
795         rv = load_ddf_global(fd, super, devname);
796
797         if (rv) {
798                 if (devname)
799                         fprintf(stderr,
800                                 Name ": Failed to load all information "
801                                 "sections on %s\n", devname);
802                 free(super);
803                 return rv;
804         }
805
806         load_ddf_local(fd, super, devname, 0);
807
808         /* Should possibly check the sections .... */
809
810         st->sb = super;
811         if (st->ss == NULL) {
812                 st->ss = &super_ddf;
813                 st->minor_version = 0;
814                 st->max_devs = 512;
815         }
816         return 0;
817
818 }
819
820 static void free_super_ddf(struct supertype *st)
821 {
822         struct ddf_super *ddf = st->sb;
823         if (ddf == NULL)
824                 return;
825         free(ddf->phys);
826         free(ddf->virt);
827         while (ddf->conflist) {
828                 struct vcl *v = ddf->conflist;
829                 ddf->conflist = v->next;
830                 if (v->block_sizes)
831                         free(v->block_sizes);
832                 free(v);
833         }
834         while (ddf->dlist) {
835                 struct dl *d = ddf->dlist;
836                 ddf->dlist = d->next;
837                 if (d->fd >= 0)
838                         close(d->fd);
839                 if (d->spare)
840                         free(d->spare);
841                 free(d);
842         }
843         free(ddf);
844         st->sb = NULL;
845 }
846
847 static struct supertype *match_metadata_desc_ddf(char *arg)
848 {
849         /* 'ddf' only support containers */
850         struct supertype *st;
851         if (strcmp(arg, "ddf") != 0 &&
852             strcmp(arg, "default") != 0
853                 )
854                 return NULL;
855
856         st = malloc(sizeof(*st));
857         memset(st, 0, sizeof(*st));
858         st->ss = &super_ddf;
859         st->max_devs = 512;
860         st->minor_version = 0;
861         st->sb = NULL;
862         return st;
863 }
864
865
866 #ifndef MDASSEMBLE
867
868 static mapping_t ddf_state[] = {
869         { "Optimal", 0},
870         { "Degraded", 1},
871         { "Deleted", 2},
872         { "Missing", 3},
873         { "Failed", 4},
874         { "Partially Optimal", 5},
875         { "-reserved-", 6},
876         { "-reserved-", 7},
877         { NULL, 0}
878 };
879
880 static mapping_t ddf_init_state[] = {
881         { "Not Initialised", 0},
882         { "QuickInit in Progress", 1},
883         { "Fully Initialised", 2},
884         { "*UNKNOWN*", 3},
885         { NULL, 0}
886 };
887 static mapping_t ddf_access[] = {
888         { "Read/Write", 0},
889         { "Reserved", 1},
890         { "Read Only", 2},
891         { "Blocked (no access)", 3},
892         { NULL ,0}
893 };
894
895 static mapping_t ddf_level[] = {
896         { "RAID0", DDF_RAID0},
897         { "RAID1", DDF_RAID1},
898         { "RAID3", DDF_RAID3},
899         { "RAID4", DDF_RAID4},
900         { "RAID5", DDF_RAID5},
901         { "RAID1E",DDF_RAID1E},
902         { "JBOD",  DDF_JBOD},
903         { "CONCAT",DDF_CONCAT},
904         { "RAID5E",DDF_RAID5E},
905         { "RAID5EE",DDF_RAID5EE},
906         { "RAID6", DDF_RAID6},
907         { NULL, 0}
908 };
909 static mapping_t ddf_sec_level[] = {
910         { "Striped", DDF_2STRIPED},
911         { "Mirrored", DDF_2MIRRORED},
912         { "Concat", DDF_2CONCAT},
913         { "Spanned", DDF_2SPANNED},
914         { NULL, 0}
915 };
916 #endif
917
918 struct num_mapping {
919         int num1, num2;
920 };
921 static struct num_mapping ddf_level_num[] = {
922         { DDF_RAID0, 0 },
923         { DDF_RAID1, 1 },
924         { DDF_RAID3, LEVEL_UNSUPPORTED },
925         { DDF_RAID4, 4 },
926         { DDF_RAID5, 5 },
927         { DDF_RAID1E, LEVEL_UNSUPPORTED },
928         { DDF_JBOD, LEVEL_UNSUPPORTED },
929         { DDF_CONCAT, LEVEL_LINEAR },
930         { DDF_RAID5E, LEVEL_UNSUPPORTED },
931         { DDF_RAID5EE, LEVEL_UNSUPPORTED },
932         { DDF_RAID6, 6},
933         { MAXINT, MAXINT }
934 };
935
936 static int map_num1(struct num_mapping *map, int num)
937 {
938         int i;
939         for (i=0 ; map[i].num1 != MAXINT; i++)
940                 if (map[i].num1 == num)
941                         break;
942         return map[i].num2;
943 }
944
945 #ifndef MDASSEMBLE
946 static void print_guid(char *guid, int tstamp)
947 {
948         /* A GUIDs are part (or all) ASCII and part binary.
949          * They tend to be space padded.
950          * We print the GUID in HEX, then in parentheses add
951          * any initial ASCII sequence, and a possible
952          * time stamp from bytes 16-19
953          */
954         int l = DDF_GUID_LEN;
955         int i;
956
957         for (i=0 ; i<DDF_GUID_LEN ; i++) {
958                 if ((i&3)==0 && i != 0) printf(":");
959                 printf("%02X", guid[i]&255);
960         }
961
962         printf(" (");
963         while (l && guid[l-1] == ' ')
964                 l--;
965         for (i=0 ; i<l ; i++) {
966                 if (guid[i] >= 0x20 && guid[i] < 0x7f)
967                         fputc(guid[i], stdout);
968                 else
969                         break;
970         }
971         if (tstamp) {
972                 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
973                 char tbuf[100];
974                 struct tm *tm;
975                 tm = localtime(&then);
976                 strftime(tbuf, 100, " %D %T",tm);
977                 fputs(tbuf, stdout);
978         }
979         printf(")");
980 }
981
982 static void examine_vd(int n, struct ddf_super *sb, char *guid)
983 {
984         int crl = sb->conf_rec_len;
985         struct vcl *vcl;
986
987         for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
988                 struct vd_config *vc = &vcl->conf;
989
990                 if (calc_crc(vc, crl*512) != vc->crc)
991                         continue;
992                 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
993                         continue;
994
995                 /* Ok, we know about this VD, let's give more details */
996                 printf(" Raid Devices[%d] : %d\n", n,
997                        __be16_to_cpu(vc->prim_elmnt_count));
998                 printf("   Chunk Size[%d] : %d sectors\n", n,
999                        1 << vc->chunk_shift);
1000                 printf("   Raid Level[%d] : %s\n", n,
1001                        map_num(ddf_level, vc->prl)?:"-unknown-");
1002                 if (vc->sec_elmnt_count != 1) {
1003                         printf("  Secondary Position[%d] : %d of %d\n", n,
1004                                vc->sec_elmnt_seq, vc->sec_elmnt_count);
1005                         printf("  Secondary Level[%d] : %s\n", n,
1006                                map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1007                 }
1008                 printf("  Device Size[%d] : %llu\n", n,
1009                        __be64_to_cpu(vc->blocks)/2);
1010                 printf("   Array Size[%d] : %llu\n", n,
1011                        __be64_to_cpu(vc->array_blocks)/2);
1012         }
1013 }
1014
1015 static void examine_vds(struct ddf_super *sb)
1016 {
1017         int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1018         int i;
1019         printf("  Virtual Disks : %d\n", cnt);
1020
1021         for (i=0; i<cnt; i++) {
1022                 struct virtual_entry *ve = &sb->virt->entries[i];
1023                 printf("      VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1024                 printf("\n");
1025                 printf("         unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1026                 printf("        state[%d] : %s, %s%s\n", i,
1027                        map_num(ddf_state, ve->state & 7),
1028                        (ve->state & 8) ? "Morphing, ": "",
1029                        (ve->state & 16)? "Not Consistent" : "Consistent");
1030                 printf("   init state[%d] : %s\n", i,
1031                        map_num(ddf_init_state, ve->init_state&3));
1032                 printf("       access[%d] : %s\n", i,
1033                        map_num(ddf_access, (ve->init_state>>6) & 3));
1034                 printf("         Name[%d] : %.16s\n", i, ve->name);
1035                 examine_vd(i, sb, ve->guid);
1036         }
1037         if (cnt) printf("\n");
1038 }
1039
1040 static void examine_pds(struct ddf_super *sb)
1041 {
1042         int cnt = __be16_to_cpu(sb->phys->used_pdes);
1043         int i;
1044         struct dl *dl;
1045         printf(" Physical Disks : %d\n", cnt);
1046
1047         for (i=0 ; i<cnt ; i++) {
1048                 struct phys_disk_entry *pd = &sb->phys->entries[i];
1049                 int type = __be16_to_cpu(pd->type);
1050                 int state = __be16_to_cpu(pd->state);
1051
1052                 printf("      PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1053                 printf("\n");
1054                 printf("          ref[%d] : %08x\n", i,
1055                        __be32_to_cpu(pd->refnum));
1056                 printf("         mode[%d] : %s%s%s%s%s\n", i,
1057                        (type&2) ? "active":"",
1058                        (type&4) ? "Global Spare":"",
1059                        (type&8) ? "spare" : "",
1060                        (type&16)? ", foreign" : "",
1061                        (type&32)? "pass-through" : "");
1062                 printf("        state[%d] : %s%s%s%s%s%s%s\n", i,
1063                        (state&1)? "Online": "Offline",
1064                        (state&2)? ", Failed": "",
1065                        (state&4)? ", Rebuilding": "",
1066                        (state&8)? ", in-transition": "",
1067                        (state&16)? ", SMART errors": "",
1068                        (state&32)? ", Unrecovered Read Errors": "",
1069                        (state&64)? ", Missing" : "");
1070                 printf("   Avail Size[%d] : %llu K\n", i,
1071                        __be64_to_cpu(pd->config_size)>>1);
1072                 for (dl = sb->dlist; dl ; dl = dl->next) {
1073                         if (dl->disk.refnum == pd->refnum) {
1074                                 char *dv = map_dev(dl->major, dl->minor, 0);
1075                                 if (dv)
1076                                         printf("       Device[%d] : %s\n",
1077                                                i, dv);
1078                         }
1079                 }
1080                 printf("\n");
1081         }
1082 }
1083
1084 static void examine_super_ddf(struct supertype *st, char *homehost)
1085 {
1086         struct ddf_super *sb = st->sb;
1087
1088         printf("          Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1089         printf("        Version : %.8s\n", sb->anchor.revision);
1090         printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1091         printf("\n");
1092         printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1093         printf("\n");
1094         printf("            Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1095         printf("  Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1096                ?"yes" : "no");
1097         examine_vds(sb);
1098         examine_pds(sb);
1099 }
1100
1101 static void brief_examine_super_ddf(struct supertype *st)
1102 {
1103         /* We just write a generic DDF ARRAY entry
1104          * The uuid is all hex, 6 groups of 4 bytes
1105          */
1106         struct ddf_super *ddf = st->sb;
1107         int i;
1108         printf("ARRAY /dev/ddf metadata=ddf UUID=");
1109         for (i = 0; i < DDF_GUID_LEN; i++) {
1110                 if ((i&3) == 0 && i != 0)
1111                         printf(":");
1112                 printf("%02X", 255&ddf->anchor.guid[i]);
1113         }
1114         printf("\n");
1115 }
1116
1117 static void detail_super_ddf(struct supertype *st, char *homehost)
1118 {
1119         /* FIXME later
1120          * Could print DDF GUID
1121          * Need to find which array
1122          *  If whole, briefly list all arrays
1123          *  If one, give name
1124          */
1125 }
1126
1127 static void brief_detail_super_ddf(struct supertype *st)
1128 {
1129         /* FIXME I really need to know which array we are detailing.
1130          * Can that be stored in ddf_super??
1131          */
1132 //      struct ddf_super *ddf = st->sb;
1133 }
1134 #endif
1135
1136 static int match_home_ddf(struct supertype *st, char *homehost)
1137 {
1138         /* It matches 'this' host if the controller is a
1139          * Linux-MD controller with vendor_data matching
1140          * the hostname
1141          */
1142         struct ddf_super *ddf = st->sb;
1143         int len = strlen(homehost);
1144
1145         return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1146                 len < sizeof(ddf->controller.vendor_data) &&
1147                 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1148                 ddf->controller.vendor_data[len] == 0);
1149 }
1150
1151 static struct vd_config *find_vdcr(struct ddf_super *ddf, int inst)
1152 {
1153         struct vcl *v;
1154
1155         for (v = ddf->conflist; v; v = v->next)
1156                 if (inst == v->vcnum)
1157                         return &v->conf;
1158         return NULL;
1159 }
1160
1161 static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1162 {
1163         /* Find the entry in phys_disk which has the given refnum
1164          * and return it's index
1165          */
1166         int i;
1167         for (i=0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1168                 if (ddf->phys->entries[i].refnum == phys_refnum)
1169                         return i;
1170         return -1;
1171 }
1172
1173 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1174 {
1175         /* The uuid returned here is used for:
1176          *  uuid to put into bitmap file (Create, Grow)
1177          *  uuid for backup header when saving critical section (Grow)
1178          *  comparing uuids when re-adding a device into an array
1179          * For each of these we can make do with a truncated
1180          * or hashed uuid rather than the original, as long as
1181          * everyone agrees.
1182          * In each case the uuid required is that of the data-array,
1183          * not the device-set.
1184          * In the case of SVD we assume the BVD is of interest,
1185          * though that might be the case if a bitmap were made for
1186          * a mirrored SVD - worry about that later.
1187          * So we need to find the VD configuration record for the
1188          * relevant BVD and extract the GUID and Secondary_Element_Seq.
1189          * The first 16 bytes of the sha1 of these is used.
1190          */
1191         struct ddf_super *ddf = st->sb;
1192         struct vcl *vcl = ddf->currentconf;
1193
1194         if (!vcl)
1195                 memset(uuid, 0, sizeof (uuid));
1196         else {
1197                 char buf[20];
1198                 struct sha1_ctx ctx;
1199                 sha1_init_ctx(&ctx);
1200                 sha1_process_bytes(&vcl->conf.guid, DDF_GUID_LEN, &ctx);
1201                 if (vcl->conf.sec_elmnt_count > 1)
1202                         sha1_process_bytes(&vcl->conf.sec_elmnt_seq, 1, &ctx);
1203                 sha1_finish_ctx(&ctx, buf);
1204                 memcpy(uuid, buf, sizeof(uuid));
1205         }
1206 }
1207
1208 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info);
1209
1210 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info)
1211 {
1212         struct ddf_super *ddf = st->sb;
1213
1214         if (ddf->currentconf) {
1215                 getinfo_super_ddf_bvd(st, info);
1216                 return;
1217         }
1218
1219         info->array.raid_disks    = __be16_to_cpu(ddf->phys->used_pdes);
1220         info->array.level         = LEVEL_CONTAINER;
1221         info->array.layout        = 0;
1222         info->array.md_minor      = -1;
1223         info->array.ctime         = DECADE + __be32_to_cpu(*(__u32*)
1224                                                          (ddf->anchor.guid+16));
1225         info->array.utime         = 0;
1226         info->array.chunk_size    = 0;
1227
1228
1229         info->disk.major = 0;
1230         info->disk.minor = 0;
1231         if (ddf->dlist) {
1232                 info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1233                 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1234
1235                 info->data_offset = __be64_to_cpu(ddf->phys->
1236                                           entries[info->disk.raid_disk].
1237                                           config_size);
1238                 info->component_size = ddf->dlist->size - info->data_offset;
1239         } else {
1240                 info->disk.number = -1;
1241 //              info->disk.raid_disk = find refnum in the table and use index;
1242         }
1243         info->disk.state = (1 << MD_DISK_SYNC);
1244
1245
1246         info->reshape_active = 0;
1247
1248         strcpy(info->text_version, "ddf");
1249
1250 //      uuid_from_super_ddf(info->uuid, sbv);
1251
1252 //      info->name[] ?? ;
1253 }
1254
1255 static int rlq_to_layout(int rlq, int prl, int raiddisks);
1256
1257 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info)
1258 {
1259         struct ddf_super *ddf = st->sb;
1260         struct vcl *vc = ddf->currentconf;
1261         int cd = ddf->currentdev;
1262
1263         /* FIXME this returns BVD info - what if we want SVD ?? */
1264
1265         info->array.raid_disks    = __be16_to_cpu(vc->conf.prim_elmnt_count);
1266         info->array.level         = map_num1(ddf_level_num, vc->conf.prl);
1267         info->array.layout        = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
1268                                                   info->array.raid_disks);
1269         info->array.md_minor      = -1;
1270         info->array.ctime         = DECADE +
1271                 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
1272         info->array.utime         = DECADE + __be32_to_cpu(vc->conf.timestamp);
1273         info->array.chunk_size    = 512 << vc->conf.chunk_shift;
1274
1275         if (cd >= 0 && cd < ddf->mppe) {
1276                 info->data_offset         = __be64_to_cpu(vc->lba_offset[cd]);
1277                 if (vc->block_sizes)
1278                         info->component_size = vc->block_sizes[cd];
1279                 else
1280                         info->component_size = __be64_to_cpu(vc->conf.blocks);
1281         }
1282
1283         info->disk.major = 0;
1284         info->disk.minor = 0;
1285 //      info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1286 //      info->disk.raid_disk = find refnum in the table and use index;
1287 //      info->disk.state = ???;
1288
1289         info->resync_start = 0;
1290         if (!(ddf->virt->entries[info->container_member].state
1291               & DDF_state_inconsistent)  &&
1292             (ddf->virt->entries[info->container_member].init_state
1293              & DDF_initstate_mask)
1294             == DDF_init_full)
1295                 info->resync_start = ~0ULL;
1296
1297         uuid_from_super_ddf(st, info->uuid);
1298
1299         info->container_member = atoi(st->subarray);
1300         sprintf(info->text_version, "/%s/%s",
1301                 devnum2devname(st->container_dev),
1302                 st->subarray);
1303
1304 //      info->name[] ?? ;
1305 }
1306
1307
1308 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1309                             char *update,
1310                             char *devname, int verbose,
1311                             int uuid_set, char *homehost)
1312 {
1313         /* For 'assemble' and 'force' we need to return non-zero if any
1314          * change was made.  For others, the return value is ignored.
1315          * Update options are:
1316          *  force-one : This device looks a bit old but needs to be included,
1317          *        update age info appropriately.
1318          *  assemble: clear any 'faulty' flag to allow this device to
1319          *              be assembled.
1320          *  force-array: Array is degraded but being forced, mark it clean
1321          *         if that will be needed to assemble it.
1322          *
1323          *  newdev:  not used ????
1324          *  grow:  Array has gained a new device - this is currently for
1325          *              linear only
1326          *  resync: mark as dirty so a resync will happen.
1327          *  uuid:  Change the uuid of the array to match what is given
1328          *  homehost:  update the recorded homehost
1329          *  name:  update the name - preserving the homehost
1330          *  _reshape_progress: record new reshape_progress position.
1331          *
1332          * Following are not relevant for this version:
1333          *  sparc2.2 : update from old dodgey metadata
1334          *  super-minor: change the preferred_minor number
1335          *  summaries:  update redundant counters.
1336          */
1337         int rv = 0;
1338 //      struct ddf_super *ddf = st->sb;
1339 //      struct vd_config *vd = find_vdcr(ddf, info->container_member);
1340 //      struct virtual_entry *ve = find_ve(ddf);
1341
1342         /* we don't need to handle "force-*" or "assemble" as
1343          * there is no need to 'trick' the kernel.  We the metadata is
1344          * first updated to activate the array, all the implied modifications
1345          * will just happen.
1346          */
1347
1348         if (strcmp(update, "grow") == 0) {
1349                 /* FIXME */
1350         }
1351         if (strcmp(update, "resync") == 0) {
1352 //              info->resync_checkpoint = 0;
1353         }
1354         /* We ignore UUID updates as they make even less sense
1355          * with DDF
1356          */
1357         if (strcmp(update, "homehost") == 0) {
1358                 /* homehost is stored in controller->vendor_data,
1359                  * or it is when we are the vendor
1360                  */
1361 //              if (info->vendor_is_local)
1362 //                      strcpy(ddf->controller.vendor_data, homehost);
1363         }
1364         if (strcmp(update, "name") == 0) {
1365                 /* name is stored in virtual_entry->name */
1366 //              memset(ve->name, ' ', 16);
1367 //              strncpy(ve->name, info->name, 16);
1368         }
1369         if (strcmp(update, "_reshape_progress") == 0) {
1370                 /* We don't support reshape yet */
1371         }
1372
1373 //      update_all_csum(ddf);
1374
1375         return rv;
1376 }
1377
1378 static void make_header_guid(char *guid)
1379 {
1380         __u32 stamp;
1381         int rfd;
1382         /* Create a DDF Header of Virtual Disk GUID */
1383
1384         /* 24 bytes of fiction required.
1385          * first 8 are a 'vendor-id'  - "Linux-MD"
1386          * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1387          * Remaining 8 random number plus timestamp
1388          */
1389         memcpy(guid, T10, sizeof(T10));
1390         stamp = __cpu_to_be32(0xdeadbeef);
1391         memcpy(guid+8, &stamp, 4);
1392         stamp = __cpu_to_be32(0);
1393         memcpy(guid+12, &stamp, 4);
1394         stamp = __cpu_to_be32(time(0) - DECADE);
1395         memcpy(guid+16, &stamp, 4);
1396         rfd = open("/dev/urandom", O_RDONLY);
1397         if (rfd < 0 || read(rfd, &stamp, 4) != 4)
1398                 stamp = random();
1399         memcpy(guid+20, &stamp, 4);
1400         if (rfd >= 0) close(rfd);
1401 }
1402
1403 static int init_super_ddf_bvd(struct supertype *st,
1404                               mdu_array_info_t *info,
1405                               unsigned long long size,
1406                               char *name, char *homehost,
1407                               int *uuid);
1408
1409 static int init_super_ddf(struct supertype *st,
1410                           mdu_array_info_t *info,
1411                           unsigned long long size, char *name, char *homehost,
1412                           int *uuid)
1413 {
1414         /* This is primarily called by Create when creating a new array.
1415          * We will then get add_to_super called for each component, and then
1416          * write_init_super called to write it out to each device.
1417          * For DDF, Create can create on fresh devices or on a pre-existing
1418          * array.
1419          * To create on a pre-existing array a different method will be called.
1420          * This one is just for fresh drives.
1421          *
1422          * We need to create the entire 'ddf' structure which includes:
1423          *  DDF headers - these are easy.
1424          *  Controller data - a Sector describing this controller .. not that
1425          *                  this is a controller exactly.
1426          *  Physical Disk Record - one entry per device, so
1427          *                      leave plenty of space.
1428          *  Virtual Disk Records - again, just leave plenty of space.
1429          *                   This just lists VDs, doesn't give details
1430          *  Config records - describes the VDs that use this disk
1431          *  DiskData  - describes 'this' device.
1432          *  BadBlockManagement - empty
1433          *  Diag Space - empty
1434          *  Vendor Logs - Could we put bitmaps here?
1435          *
1436          */
1437         struct ddf_super *ddf;
1438         char hostname[17];
1439         int hostlen;
1440         int max_phys_disks, max_virt_disks;
1441         unsigned long long sector;
1442         int clen;
1443         int i;
1444         int pdsize, vdsize;
1445         struct phys_disk *pd;
1446         struct virtual_disk *vd;
1447
1448         if (!info) {
1449                 st->sb = NULL;
1450                 return 0;
1451         }
1452         if (st->sb)
1453                 return init_super_ddf_bvd(st, info, size, name, homehost,
1454                                           uuid);
1455
1456         posix_memalign((void**)&ddf, 512, sizeof(*ddf));
1457         memset(ddf, 0, sizeof(*ddf));
1458         ddf->dlist = NULL; /* no physical disks yet */
1459         ddf->conflist = NULL; /* No virtual disks yet */
1460
1461         /* At least 32MB *must* be reserved for the ddf.  So let's just
1462          * start 32MB from the end, and put the primary header there.
1463          * Don't do secondary for now.
1464          * We don't know exactly where that will be yet as it could be
1465          * different on each device.  To just set up the lengths.
1466          *
1467          */
1468
1469         ddf->anchor.magic = DDF_HEADER_MAGIC;
1470         make_header_guid(ddf->anchor.guid);
1471
1472         memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
1473         ddf->anchor.seq = __cpu_to_be32(1);
1474         ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1475         ddf->anchor.openflag = 0xFF;
1476         ddf->anchor.foreignflag = 0;
1477         ddf->anchor.enforcegroups = 0; /* Is this best?? */
1478         ddf->anchor.pad0 = 0xff;
1479         memset(ddf->anchor.pad1, 0xff, 12);
1480         memset(ddf->anchor.header_ext, 0xff, 32);
1481         ddf->anchor.primary_lba = ~(__u64)0;
1482         ddf->anchor.secondary_lba = ~(__u64)0;
1483         ddf->anchor.type = DDF_HEADER_ANCHOR;
1484         memset(ddf->anchor.pad2, 0xff, 3);
1485         ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1486         ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1487                                                   of 32M reserved.. */
1488         max_phys_disks = 1023;   /* Should be enough */
1489         ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1490         max_virt_disks = 255;
1491         ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1492         ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1493         ddf->max_part = 64;
1494         ddf->mppe = 256;
1495         ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
1496         ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1497         ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe);
1498         memset(ddf->anchor.pad3, 0xff, 54);
1499         /* controller sections is one sector long immediately
1500          * after the ddf header */
1501         sector = 1;
1502         ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1503         ddf->anchor.controller_section_length = __cpu_to_be32(1);
1504         sector += 1;
1505
1506         /* phys is 8 sectors after that */
1507         pdsize = ROUND_UP(sizeof(struct phys_disk) +
1508                           sizeof(struct phys_disk_entry)*max_phys_disks,
1509                           512);
1510         switch(pdsize/512) {
1511         case 2: case 8: case 32: case 128: case 512: break;
1512         default: abort();
1513         }
1514         ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1515         ddf->anchor.phys_section_length =
1516                 __cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1517         sector += pdsize/512;
1518
1519         /* virt is another 32 sectors */
1520         vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1521                           sizeof(struct virtual_entry) * max_virt_disks,
1522                           512);
1523         switch(vdsize/512) {
1524         case 2: case 8: case 32: case 128: case 512: break;
1525         default: abort();
1526         }
1527         ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1528         ddf->anchor.virt_section_length =
1529                 __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1530         sector += vdsize/512;
1531
1532         clen = ddf->conf_rec_len * (ddf->max_part+1);
1533         ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1534         ddf->anchor.config_section_length = __cpu_to_be32(clen);
1535         sector += clen;
1536
1537         ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1538         ddf->anchor.data_section_length = __cpu_to_be32(1);
1539         sector += 1;
1540
1541         ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1542         ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1543         ddf->anchor.diag_space_length = __cpu_to_be32(0);
1544         ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1545         ddf->anchor.vendor_length = __cpu_to_be32(0);
1546         ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1547
1548         memset(ddf->anchor.pad4, 0xff, 256);
1549
1550         memcpy(&ddf->primary, &ddf->anchor, 512);
1551         memcpy(&ddf->secondary, &ddf->anchor, 512);
1552
1553         ddf->primary.openflag = 1; /* I guess.. */
1554         ddf->primary.type = DDF_HEADER_PRIMARY;
1555
1556         ddf->secondary.openflag = 1; /* I guess.. */
1557         ddf->secondary.type = DDF_HEADER_SECONDARY;
1558
1559         ddf->active = &ddf->primary;
1560
1561         ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1562
1563         /* 24 more bytes of fiction required.
1564          * first 8 are a 'vendor-id'  - "Linux-MD"
1565          * Remaining 16 are serial number.... maybe a hostname would do?
1566          */
1567         memcpy(ddf->controller.guid, T10, sizeof(T10));
1568         gethostname(hostname, sizeof(hostname));
1569         hostname[sizeof(hostname) - 1] = 0;
1570         hostlen = strlen(hostname);
1571         memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1572         for (i = strlen(T10) ; i+hostlen < 24; i++)
1573                 ddf->controller.guid[i] = ' ';
1574
1575         ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1576         ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1577         ddf->controller.type.sub_vendor_id = 0;
1578         ddf->controller.type.sub_device_id = 0;
1579         memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1580         memset(ddf->controller.pad, 0xff, 8);
1581         memset(ddf->controller.vendor_data, 0xff, 448);
1582
1583         posix_memalign((void**)&pd, 512, pdsize);
1584         ddf->phys = pd;
1585         ddf->pdsize = pdsize;
1586
1587         memset(pd, 0xff, pdsize);
1588         memset(pd, 0, sizeof(*pd));
1589         pd->magic = DDF_PHYS_DATA_MAGIC;
1590         pd->used_pdes = __cpu_to_be16(0);
1591         pd->max_pdes = __cpu_to_be16(max_phys_disks);
1592         memset(pd->pad, 0xff, 52);
1593
1594         posix_memalign((void**)&vd, 512, vdsize);
1595         ddf->virt = vd;
1596         ddf->vdsize = vdsize;
1597         memset(vd, 0, vdsize);
1598         vd->magic = DDF_VIRT_RECORDS_MAGIC;
1599         vd->populated_vdes = __cpu_to_be16(0);
1600         vd->max_vdes = __cpu_to_be16(max_virt_disks);
1601         memset(vd->pad, 0xff, 52);
1602
1603         for (i=0; i<max_virt_disks; i++)
1604                 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1605
1606         st->sb = ddf;
1607         ddf->updates_pending = 1;
1608         return 1;
1609 }
1610
1611 static int all_ff(char *guid)
1612 {
1613         int i;
1614         for (i = 0; i < DDF_GUID_LEN; i++)
1615                 if (guid[i] != (char)0xff)
1616                         return 0;
1617         return 1;
1618 }
1619 static int chunk_to_shift(int chunksize)
1620 {
1621         return ffs(chunksize/512)-1;
1622 }
1623
1624 static int level_to_prl(int level)
1625 {
1626         switch (level) {
1627         case LEVEL_LINEAR: return DDF_CONCAT;
1628         case 0: return DDF_RAID0;
1629         case 1: return DDF_RAID1;
1630         case 4: return DDF_RAID4;
1631         case 5: return DDF_RAID5;
1632         case 6: return DDF_RAID6;
1633         default: return -1;
1634         }
1635 }
1636 static int layout_to_rlq(int level, int layout, int raiddisks)
1637 {
1638         switch(level) {
1639         case 0:
1640                 return DDF_RAID0_SIMPLE;
1641         case 1:
1642                 switch(raiddisks) {
1643                 case 2: return DDF_RAID1_SIMPLE;
1644                 case 3: return DDF_RAID1_MULTI;
1645                 default: return -1;
1646                 }
1647         case 4:
1648                 switch(layout) {
1649                 case 0: return DDF_RAID4_N;
1650                 }
1651                 break;
1652         case 5:
1653         case 6:
1654                 switch(layout) {
1655                 case ALGORITHM_LEFT_ASYMMETRIC:
1656                         return DDF_RAID5_N_RESTART;
1657                 case ALGORITHM_RIGHT_ASYMMETRIC:
1658                         if (level == 5)
1659                                 return DDF_RAID5_0_RESTART;
1660                         else
1661                                 return DDF_RAID6_0_RESTART;
1662                 case ALGORITHM_LEFT_SYMMETRIC:
1663                         return DDF_RAID5_N_CONTINUE;
1664                 case ALGORITHM_RIGHT_SYMMETRIC:
1665                         return -1; /* not mentioned in standard */
1666                 }
1667         }
1668         return -1;
1669 }
1670
1671 static int rlq_to_layout(int rlq, int prl, int raiddisks)
1672 {
1673         switch(prl) {
1674         case DDF_RAID0:
1675                 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1676         case DDF_RAID1:
1677                 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1678                              on raiddisks*/
1679         case DDF_RAID4:
1680                 switch(rlq) {
1681                 case DDF_RAID4_N:
1682                         return 0;
1683                 default:
1684                         /* not supported */
1685                         return -1; /* FIXME this isn't checked */
1686                 }
1687         case DDF_RAID5:
1688                 switch(rlq) {
1689                 case DDF_RAID5_N_RESTART:
1690                         return ALGORITHM_LEFT_ASYMMETRIC;
1691                 case DDF_RAID5_0_RESTART:
1692                         return ALGORITHM_RIGHT_ASYMMETRIC;
1693                 case DDF_RAID5_N_CONTINUE:
1694                         return ALGORITHM_LEFT_SYMMETRIC;
1695                 default:
1696                         return -1;
1697                 }
1698         case DDF_RAID6:
1699                 switch(rlq) {
1700                 case DDF_RAID5_N_RESTART:
1701                         return ALGORITHM_LEFT_ASYMMETRIC;
1702                 case DDF_RAID6_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         }
1710         return -1;
1711 }
1712
1713 struct extent {
1714         unsigned long long start, size;
1715 };
1716 static int cmp_extent(const void *av, const void *bv)
1717 {
1718         const struct extent *a = av;
1719         const struct extent *b = bv;
1720         if (a->start < b->start)
1721                 return -1;
1722         if (a->start > b->start)
1723                 return 1;
1724         return 0;
1725 }
1726
1727 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
1728 {
1729         /* find a list of used extents on the give physical device
1730          * (dnum) of the given ddf.
1731          * Return a malloced array of 'struct extent'
1732
1733 FIXME ignore DDF_Legacy devices?
1734
1735          */
1736         struct extent *rv;
1737         int n = 0;
1738         int i, j;
1739
1740         rv = malloc(sizeof(struct extent) * (ddf->max_part + 2));
1741         if (!rv)
1742                 return NULL;
1743
1744         for (i = 0; i < ddf->max_part; i++) {
1745                 struct vcl *v = dl->vlist[i];
1746                 if (v == NULL)
1747                         continue;
1748                 for (j=0; j < v->conf.prim_elmnt_count; j++)
1749                         if (v->conf.phys_refnum[j] == dl->disk.refnum) {
1750                                 /* This device plays role 'j' in  'v'. */
1751                                 rv[n].start = __be64_to_cpu(v->lba_offset[j]);
1752                                 rv[n].size = __be64_to_cpu(v->conf.blocks);
1753                                 n++;
1754                                 break;
1755                         }
1756         }
1757         qsort(rv, n, sizeof(*rv), cmp_extent);
1758
1759         rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
1760         rv[n].size = 0;
1761         return rv;
1762 }
1763
1764 static int init_super_ddf_bvd(struct supertype *st,
1765                               mdu_array_info_t *info,
1766                               unsigned long long size,
1767                               char *name, char *homehost,
1768                               int *uuid)
1769 {
1770         /* We are creating a BVD inside a pre-existing container.
1771          * so st->sb is already set.
1772          * We need to create a new vd_config and a new virtual_entry
1773          */
1774         struct ddf_super *ddf = st->sb;
1775         int venum;
1776         struct virtual_entry *ve;
1777         struct vcl *vcl;
1778         struct vd_config *vc;
1779
1780         if (__be16_to_cpu(ddf->virt->populated_vdes)
1781             >= __be16_to_cpu(ddf->virt->max_vdes)) {
1782                 fprintf(stderr, Name": This ddf already has the "
1783                         "maximum of %d virtual devices\n",
1784                         __be16_to_cpu(ddf->virt->max_vdes));
1785                 return 0;
1786         }
1787
1788         for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1789                 if (all_ff(ddf->virt->entries[venum].guid))
1790                         break;
1791         if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1792                 fprintf(stderr, Name ": Cannot find spare slot for "
1793                         "virtual disk - DDF is corrupt\n");
1794                 return 0;
1795         }
1796         ve = &ddf->virt->entries[venum];
1797
1798         /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1799          * timestamp, random number
1800          */
1801         make_header_guid(ve->guid);
1802         ve->unit = __cpu_to_be16(info->md_minor);
1803         ve->pad0 = 0xFFFF;
1804         ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1805         ve->type = 0;
1806         ve->state = DDF_state_degraded; /* Will be modified as devices are added */
1807         if (info->state & 1) /* clean */
1808                 ve->init_state = DDF_init_full;
1809         else
1810                 ve->init_state = DDF_init_not;
1811
1812         memset(ve->pad1, 0xff, 14);
1813         memset(ve->name, ' ', 16);
1814         if (name)
1815                 strncpy(ve->name, name, 16);
1816         ddf->virt->populated_vdes =
1817                 __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
1818
1819         /* Now create a new vd_config */
1820         posix_memalign((void**)&vcl, 512,
1821                        (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512));
1822         vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
1823         vcl->vcnum = venum;
1824         sprintf(st->subarray, "%d", venum);
1825         vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1826
1827         vc = &vcl->conf;
1828
1829         vc->magic = DDF_VD_CONF_MAGIC;
1830         memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
1831         vc->timestamp = __cpu_to_be32(time(0)-DECADE);
1832         vc->seqnum = __cpu_to_be32(1);
1833         memset(vc->pad0, 0xff, 24);
1834         vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
1835         vc->chunk_shift = chunk_to_shift(info->chunk_size);
1836         vc->prl = level_to_prl(info->level);
1837         vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
1838         vc->sec_elmnt_count = 1;
1839         vc->sec_elmnt_seq = 0;
1840         vc->srl = 0;
1841         vc->blocks = __cpu_to_be64(info->size * 2);
1842         vc->array_blocks = __cpu_to_be64(
1843                 calc_array_size(info->level, info->raid_disks, info->layout,
1844                                 info->chunk_size, info->size*2));
1845         memset(vc->pad1, 0xff, 8);
1846         vc->spare_refs[0] = 0xffffffff;
1847         vc->spare_refs[1] = 0xffffffff;
1848         vc->spare_refs[2] = 0xffffffff;
1849         vc->spare_refs[3] = 0xffffffff;
1850         vc->spare_refs[4] = 0xffffffff;
1851         vc->spare_refs[5] = 0xffffffff;
1852         vc->spare_refs[6] = 0xffffffff;
1853         vc->spare_refs[7] = 0xffffffff;
1854         memset(vc->cache_pol, 0, 8);
1855         vc->bg_rate = 0x80;
1856         memset(vc->pad2, 0xff, 3);
1857         memset(vc->pad3, 0xff, 52);
1858         memset(vc->pad4, 0xff, 192);
1859         memset(vc->v0, 0xff, 32);
1860         memset(vc->v1, 0xff, 32);
1861         memset(vc->v2, 0xff, 16);
1862         memset(vc->v3, 0xff, 16);
1863         memset(vc->vendor, 0xff, 32);
1864
1865         memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
1866         memset(vc->phys_refnum+(ddf->mppe * 4), 0x00, 8*ddf->mppe);
1867
1868         vcl->next = ddf->conflist;
1869         ddf->conflist = vcl;
1870         ddf->currentconf = vcl;
1871         ddf->updates_pending = 1;
1872         return 1;
1873 }
1874
1875 static void add_to_super_ddf_bvd(struct supertype *st,
1876                                  mdu_disk_info_t *dk, int fd, char *devname)
1877 {
1878         /* fd and devname identify a device with-in the ddf container (st).
1879          * dk identifies a location in the new BVD.
1880          * We need to find suitable free space in that device and update
1881          * the phys_refnum and lba_offset for the newly created vd_config.
1882          * We might also want to update the type in the phys_disk
1883          * section.
1884          */
1885         struct dl *dl;
1886         struct ddf_super *ddf = st->sb;
1887         struct vd_config *vc;
1888         __u64 *lba_offset;
1889         int working;
1890         int i;
1891         unsigned long long blocks, pos, esize;
1892         struct extent *ex;
1893
1894         for (dl = ddf->dlist; dl ; dl = dl->next)
1895                 if (dl->major == dk->major &&
1896                     dl->minor == dk->minor)
1897                         break;
1898         if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
1899                 return;
1900
1901         vc = &ddf->currentconf->conf;
1902         lba_offset = ddf->currentconf->lba_offset;
1903
1904         ex = get_extents(ddf, dl);
1905         if (!ex)
1906                 return;
1907
1908         i = 0; pos = 0;
1909         blocks = __be64_to_cpu(vc->blocks);
1910         if (ddf->currentconf->block_sizes)
1911                 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
1912
1913         do {
1914                 esize = ex[i].start - pos;
1915                 if (esize >= blocks)
1916                         break;
1917                 pos = ex[i].start + ex[i].size;
1918                 i++;
1919         } while (ex[i-1].size);
1920
1921         free(ex);
1922         if (esize < blocks)
1923                 return;
1924
1925         ddf->currentdev = dk->raid_disk;
1926         vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
1927         lba_offset[dk->raid_disk] = __cpu_to_be64(pos);
1928
1929         for (i=0; i < ddf->max_part ; i++)
1930                 if (dl->vlist[i] == NULL)
1931                         break;
1932         if (i == ddf->max_part)
1933                 return;
1934         dl->vlist[i] = ddf->currentconf;
1935
1936         dl->fd = fd;
1937         dl->devname = devname;
1938
1939         /* Check how many working raid_disks, and if we can mark
1940          * array as optimal yet
1941          */
1942         working = 0;
1943
1944         for (i=0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
1945                 if (vc->phys_refnum[i] != 0xffffffff)
1946                         working++;
1947
1948         /* Find which virtual_entry */
1949         i = ddf->currentconf->vcnum;
1950         if (working == __be16_to_cpu(vc->prim_elmnt_count))
1951                 ddf->virt->entries[i].state =
1952                         (ddf->virt->entries[i].state & ~DDF_state_mask)
1953                         | DDF_state_optimal;
1954
1955         if (vc->prl == DDF_RAID6 &&
1956             working+1 == __be16_to_cpu(vc->prim_elmnt_count))
1957                 ddf->virt->entries[i].state =
1958                         (ddf->virt->entries[i].state & ~DDF_state_mask)
1959                         | DDF_state_part_optimal;
1960
1961         ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
1962         ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
1963         ddf->updates_pending = 1;
1964 }
1965
1966 /* add a device to a container, either while creating it or while
1967  * expanding a pre-existing container
1968  */
1969 static void add_to_super_ddf(struct supertype *st,
1970                              mdu_disk_info_t *dk, int fd, char *devname)
1971 {
1972         struct ddf_super *ddf = st->sb;
1973         struct dl *dd;
1974         time_t now;
1975         struct tm *tm;
1976         unsigned long long size;
1977         struct phys_disk_entry *pde;
1978         int n, i;
1979         struct stat stb;
1980
1981         if (ddf->currentconf) {
1982                 add_to_super_ddf_bvd(st, dk, fd, devname);
1983                 return;
1984         }
1985
1986         /* This is device numbered dk->number.  We need to create
1987          * a phys_disk entry and a more detailed disk_data entry.
1988          */
1989         fstat(fd, &stb);
1990         posix_memalign((void**)&dd, 512,
1991                        sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part);
1992         dd->major = major(stb.st_rdev);
1993         dd->minor = minor(stb.st_rdev);
1994         dd->devname = devname;
1995         dd->next = ddf->dlist;
1996         dd->fd = fd;
1997         dd->spare = NULL;
1998
1999         dd->disk.magic = DDF_PHYS_DATA_MAGIC;
2000         now = time(0);
2001         tm = localtime(&now);
2002         sprintf(dd->disk.guid, "%8s%04d%02d%02d",
2003                 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
2004         *(__u32*)(dd->disk.guid + 16) = random();
2005         *(__u32*)(dd->disk.guid + 20) = random();
2006
2007         do {
2008                 /* Cannot be bothered finding a CRC of some irrelevant details*/
2009                 dd->disk.refnum = random();
2010                 for (i = __be16_to_cpu(ddf->active->max_pd_entries) - 1;
2011                      i >= 0; i--)
2012                         if (ddf->phys->entries[i].refnum == dd->disk.refnum)
2013                                 break;
2014         } while (i >= 0);
2015
2016         dd->disk.forced_ref = 1;
2017         dd->disk.forced_guid = 1;
2018         memset(dd->disk.vendor, ' ', 32);
2019         memcpy(dd->disk.vendor, "Linux", 5);
2020         memset(dd->disk.pad, 0xff, 442);
2021         for (i = 0; i < ddf->max_part ; i++)
2022                 dd->vlist[i] = NULL;
2023
2024         n = __be16_to_cpu(ddf->phys->used_pdes);
2025         pde = &ddf->phys->entries[n];
2026         dd->pdnum = n;
2027
2028         n++;
2029         ddf->phys->used_pdes = __cpu_to_be16(n);
2030
2031         memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2032         pde->refnum = dd->disk.refnum;
2033         pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2034         pde->state = __cpu_to_be16(DDF_Online);
2035         get_dev_size(fd, NULL, &size);
2036         /* We are required to reserve 32Meg, and record the size in sectors */
2037         pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
2038         sprintf(pde->path, "%17.17s","Information: nil") ;
2039         memset(pde->pad, 0xff, 6);
2040
2041         dd->size = size >> 9;
2042         ddf->dlist = dd;
2043         ddf->updates_pending = 1;
2044 }
2045
2046 /*
2047  * This is the write_init_super method for a ddf container.  It is
2048  * called when creating a container or adding another device to a
2049  * container.
2050  */
2051
2052 #ifndef MDASSEMBLE
2053
2054 static unsigned char null_conf[4096+512];
2055
2056 static int __write_init_super_ddf(struct supertype *st, int do_close)
2057 {
2058
2059         struct ddf_super *ddf = st->sb;
2060         int i;
2061         struct dl *d;
2062         int n_config;
2063         int conf_size;
2064
2065         unsigned long long size, sector;
2066
2067         for (d = ddf->dlist; d; d=d->next) {
2068                 int fd = d->fd;
2069
2070                 if (fd < 0)
2071                         continue;
2072
2073                 /* We need to fill in the primary, (secondary) and workspace
2074                  * lba's in the headers, set their checksums,
2075                  * Also checksum phys, virt....
2076                  *
2077                  * Then write everything out, finally the anchor is written.
2078                  */
2079                 get_dev_size(fd, NULL, &size);
2080                 size /= 512;
2081                 ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
2082                 ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
2083                 ddf->anchor.seq = __cpu_to_be32(1);
2084                 memcpy(&ddf->primary, &ddf->anchor, 512);
2085                 memcpy(&ddf->secondary, &ddf->anchor, 512);
2086
2087                 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
2088                 ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
2089                 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
2090
2091                 ddf->primary.openflag = 0;
2092                 ddf->primary.type = DDF_HEADER_PRIMARY;
2093
2094                 ddf->secondary.openflag = 0;
2095                 ddf->secondary.type = DDF_HEADER_SECONDARY;
2096
2097                 ddf->primary.crc = calc_crc(&ddf->primary, 512);
2098                 ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
2099
2100                 sector = size - 16*1024*2;
2101                 lseek64(fd, sector<<9, 0);
2102                 write(fd, &ddf->primary, 512);
2103
2104                 ddf->controller.crc = calc_crc(&ddf->controller, 512);
2105                 write(fd, &ddf->controller, 512);
2106
2107                 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
2108
2109                 write(fd, ddf->phys, ddf->pdsize);
2110
2111                 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
2112                 write(fd, ddf->virt, ddf->vdsize);
2113
2114                 /* Now write lots of config records. */
2115                 n_config = ddf->max_part;
2116                 conf_size = ddf->conf_rec_len * 512;
2117                 for (i = 0 ; i <= n_config ; i++) {
2118                         struct vcl *c = d->vlist[i];
2119                         if (i == n_config)
2120                                 c = (struct vcl*)d->spare;
2121
2122                         if (c) {
2123                                 c->conf.crc = calc_crc(&c->conf, conf_size);
2124                                 write(fd, &c->conf, conf_size);
2125                         } else {
2126                                 char *null_aligned = (char*)((((unsigned long)null_conf)+511)&~511UL);
2127                                 if (null_conf[0] != 0xff)
2128                                         memset(null_conf, 0xff, sizeof(null_conf));
2129                                 int togo = conf_size;
2130                                 while (togo > sizeof(null_conf)-512) {
2131                                         write(fd, null_aligned, sizeof(null_conf)-512);
2132                                         togo -= sizeof(null_conf)-512;
2133                                 }
2134                                 write(fd, null_aligned, togo);
2135                         }
2136                 }
2137                 d->disk.crc = calc_crc(&d->disk, 512);
2138                 write(fd, &d->disk, 512);
2139
2140                 /* Maybe do the same for secondary */
2141
2142                 lseek64(fd, (size-1)*512, SEEK_SET);
2143                 write(fd, &ddf->anchor, 512);
2144                 if (do_close) {
2145                         close(fd);
2146                         d->fd = -1;
2147                 }
2148         }
2149         return 1;
2150 }
2151
2152 static int write_init_super_ddf(struct supertype *st)
2153 {
2154
2155         if (st->update_tail) {
2156                 /* queue the virtual_disk and vd_config as metadata updates */
2157                 struct virtual_disk *vd;
2158                 struct vd_config *vc;
2159                 struct ddf_super *ddf = st->sb;
2160                 int len;
2161
2162                 /* First the virtual disk.  We have a slightly fake header */
2163                 len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry);
2164                 vd = malloc(len);
2165                 *vd = *ddf->virt;
2166                 vd->entries[0] = ddf->virt->entries[ddf->currentconf->vcnum];
2167                 vd->populated_vdes = __cpu_to_be16(ddf->currentconf->vcnum);
2168                 append_metadata_update(st, vd, len);
2169
2170                 /* Then the vd_config */
2171                 len = ddf->conf_rec_len * 512;
2172                 vc = malloc(len);
2173                 memcpy(vc, &ddf->currentconf->conf, len);
2174                 append_metadata_update(st, vc, len);
2175
2176                 /* FIXME I need to close the fds! */
2177                 return 0;
2178         } else 
2179                 return __write_init_super_ddf(st, 1);
2180 }
2181
2182 #endif
2183
2184 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize)
2185 {
2186         /* We must reserve the last 32Meg */
2187         if (devsize <= 32*1024*2)
2188                 return 0;
2189         return devsize - 32*1024*2;
2190 }
2191
2192 #ifndef MDASSEMBLE
2193 static int
2194 validate_geometry_ddf_container(struct supertype *st,
2195                                 int level, int layout, int raiddisks,
2196                                 int chunk, unsigned long long size,
2197                                 char *dev, unsigned long long *freesize,
2198                                 int verbose);
2199
2200 static int validate_geometry_ddf_bvd(struct supertype *st,
2201                                      int level, int layout, int raiddisks,
2202                                      int chunk, unsigned long long size,
2203                                      char *dev, unsigned long long *freesize,
2204                                      int verbose);
2205
2206 static int validate_geometry_ddf(struct supertype *st,
2207                                  int level, int layout, int raiddisks,
2208                                  int chunk, unsigned long long size,
2209                                  char *dev, unsigned long long *freesize,
2210                                  int verbose)
2211 {
2212         int fd;
2213         struct mdinfo *sra;
2214         int cfd;
2215
2216         /* ddf potentially supports lots of things, but it depends on
2217          * what devices are offered (and maybe kernel version?)
2218          * If given unused devices, we will make a container.
2219          * If given devices in a container, we will make a BVD.
2220          * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2221          */
2222
2223         if (level == LEVEL_CONTAINER) {
2224                 /* Must be a fresh device to add to a container */
2225                 return validate_geometry_ddf_container(st, level, layout,
2226                                                        raiddisks, chunk,
2227                                                        size, dev, freesize,
2228                                                        verbose);
2229         }
2230
2231         if (st->sb) {
2232                 /* A container has already been opened, so we are
2233                  * creating in there.  Maybe a BVD, maybe an SVD.
2234                  * Should make a distinction one day.
2235                  */
2236                 return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
2237                                                  chunk, size, dev, freesize,
2238                                                  verbose);
2239         }
2240         if (!dev) {
2241                 /* Initial sanity check.  Exclude illegal levels. */
2242                 int i;
2243                 for (i=0; ddf_level_num[i].num1 != MAXINT; i++)
2244                         if (ddf_level_num[i].num2 == level)
2245                                 break;
2246                 if (ddf_level_num[i].num1 == MAXINT)
2247                         return 0;
2248                 /* Should check layout? etc */
2249                 return 1;
2250         }
2251
2252         /* This is the first device for the array.
2253          * If it is a container, we read it in and do automagic allocations,
2254          * no other devices should be given.
2255          * Otherwise it must be a member device of a container, and we
2256          * do manual allocation.
2257          * Later we should check for a BVD and make an SVD.
2258          */
2259         fd = open(dev, O_RDONLY|O_EXCL, 0);
2260         if (fd >= 0) {
2261                 sra = sysfs_read(fd, 0, GET_VERSION);
2262                 close(fd);
2263                 if (sra && sra->array.major_version == -1 &&
2264                     strcmp(sra->text_version, "ddf") == 0) {
2265
2266                         /* load super */
2267                         /* find space for 'n' devices. */
2268                         /* remember the devices */
2269                         /* Somehow return the fact that we have enough */
2270                 }
2271
2272                 if (verbose)
2273                         fprintf(stderr,
2274                                 Name ": ddf: Cannot create this array "
2275                                 "on device %s\n",
2276                                 dev);
2277                 return 0;
2278         }
2279         if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2280                 if (verbose)
2281                         fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2282                                 dev, strerror(errno));
2283                 return 0;
2284         }
2285         /* Well, it is in use by someone, maybe a 'ddf' container. */
2286         cfd = open_container(fd);
2287         if (cfd < 0) {
2288                 close(fd);
2289                 if (verbose)
2290                         fprintf(stderr, Name ": ddf: Cannot use %s: %s\n",
2291                                 dev, strerror(EBUSY));
2292                 return 0;
2293         }
2294         sra = sysfs_read(cfd, 0, GET_VERSION);
2295         close(fd);
2296         if (sra && sra->array.major_version == -1 &&
2297             strcmp(sra->text_version, "ddf") == 0) {
2298                 /* This is a member of a ddf container.  Load the container
2299                  * and try to create a bvd
2300                  */
2301                 struct ddf_super *ddf;
2302                 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL, 1) == 0) {
2303                         st->sb = ddf;
2304                         st->container_dev = fd2devnum(cfd);
2305                         close(cfd);
2306                         return validate_geometry_ddf_bvd(st, level, layout,
2307                                                          raiddisks, chunk, size,
2308                                                          dev, freesize,
2309                                                          verbose);
2310                 }
2311                 close(cfd);
2312         } else /* device may belong to a different container */
2313                 return 0;
2314
2315         return 1;
2316 }
2317
2318 static int
2319 validate_geometry_ddf_container(struct supertype *st,
2320                                 int level, int layout, int raiddisks,
2321                                 int chunk, unsigned long long size,
2322                                 char *dev, unsigned long long *freesize,
2323                                 int verbose)
2324 {
2325         int fd;
2326         unsigned long long ldsize;
2327
2328         if (level != LEVEL_CONTAINER)
2329                 return 0;
2330         if (!dev)
2331                 return 1;
2332
2333         fd = open(dev, O_RDONLY|O_EXCL, 0);
2334         if (fd < 0) {
2335                 if (verbose)
2336                         fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2337                                 dev, strerror(errno));
2338                 return 0;
2339         }
2340         if (!get_dev_size(fd, dev, &ldsize)) {
2341                 close(fd);
2342                 return 0;
2343         }
2344         close(fd);
2345
2346         *freesize = avail_size_ddf(st, ldsize >> 9);
2347
2348         return 1;
2349 }
2350
2351 static int validate_geometry_ddf_bvd(struct supertype *st,
2352                                      int level, int layout, int raiddisks,
2353                                      int chunk, unsigned long long size,
2354                                      char *dev, unsigned long long *freesize,
2355                                      int verbose)
2356 {
2357         struct stat stb;
2358         struct ddf_super *ddf = st->sb;
2359         struct dl *dl;
2360         unsigned long long pos = 0;
2361         unsigned long long maxsize;
2362         struct extent *e;
2363         int i;
2364         /* ddf/bvd supports lots of things, but not containers */
2365         if (level == LEVEL_CONTAINER)
2366                 return 0;
2367         /* We must have the container info already read in. */
2368         if (!ddf)
2369                 return 0;
2370
2371         if (!dev) {
2372                 /* General test:  make sure there is space for
2373                  * 'raiddisks' device extents of size 'size'.
2374                  */
2375                 unsigned long long minsize = size;
2376                 int dcnt = 0;
2377                 if (minsize == 0)
2378                         minsize = 8;
2379                 for (dl = ddf->dlist; dl ; dl = dl->next)
2380                 {
2381                         int found = 0;
2382                         pos = 0;
2383
2384                         i = 0;
2385                         e = get_extents(ddf, dl);
2386                         if (!e) continue;
2387                         do {
2388                                 unsigned long long esize;
2389                                 esize = e[i].start - pos;
2390                                 if (esize >= minsize)
2391                                         found = 1;
2392                                 pos = e[i].start + e[i].size;
2393                                 i++;
2394                         } while (e[i-1].size);
2395                         if (found)
2396                                 dcnt++;
2397                         free(e);
2398                 }
2399                 if (dcnt < raiddisks) {
2400                         if (verbose)
2401                                 fprintf(stderr,
2402                                         Name ": ddf: Not enough devices with "
2403                                         "space for this array (%d < %d)\n",
2404                                         dcnt, raiddisks);
2405                         return 0;
2406                 }
2407                 return 1;
2408         }
2409         /* This device must be a member of the set */
2410         if (stat(dev, &stb) < 0)
2411                 return 0;
2412         if ((S_IFMT & stb.st_mode) != S_IFBLK)
2413                 return 0;
2414         for (dl = ddf->dlist ; dl ; dl = dl->next) {
2415                 if (dl->major == major(stb.st_rdev) &&
2416                     dl->minor == minor(stb.st_rdev))
2417                         break;
2418         }
2419         if (!dl) {
2420                 if (verbose)
2421                         fprintf(stderr, Name ": ddf: %s is not in the "
2422                                 "same DDF set\n",
2423                                 dev);
2424                 return 0;
2425         }
2426         e = get_extents(ddf, dl);
2427         maxsize = 0;
2428         i = 0;
2429         if (e) do {
2430                 unsigned long long esize;
2431                 esize = e[i].start - pos;
2432                 if (esize >= maxsize)
2433                         maxsize = esize;
2434                 pos = e[i].start + e[i].size;
2435                 i++;
2436         } while (e[i-1].size);
2437         *freesize = maxsize;
2438         // FIXME here I am
2439
2440         return 1;
2441 }
2442
2443 static int load_super_ddf_all(struct supertype *st, int fd,
2444                               void **sbp, char *devname, int keep_fd)
2445 {
2446         struct mdinfo *sra;
2447         struct ddf_super *super;
2448         struct mdinfo *sd, *best = NULL;
2449         int bestseq = 0;
2450         int seq;
2451         char nm[20];
2452         int dfd;
2453
2454         sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
2455         if (!sra)
2456                 return 1;
2457         if (sra->array.major_version != -1 ||
2458             sra->array.minor_version != -2 ||
2459             strcmp(sra->text_version, "ddf") != 0)
2460                 return 1;
2461
2462         if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0)
2463                 return 1;
2464         memset(super, 0, sizeof(*super));
2465
2466         /* first, try each device, and choose the best ddf */
2467         for (sd = sra->devs ; sd ; sd = sd->next) {
2468                 int rv;
2469                 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2470                 dfd = dev_open(nm, O_RDONLY);
2471                 if (dfd < 0)
2472                         return 2;
2473                 rv = load_ddf_headers(dfd, super, NULL);
2474                 close(dfd);
2475                 if (rv == 0) {
2476                         seq = __be32_to_cpu(super->active->seq);
2477                         if (super->active->openflag)
2478                                 seq--;
2479                         if (!best || seq > bestseq) {
2480                                 bestseq = seq;
2481                                 best = sd;
2482                         }
2483                 }
2484         }
2485         if (!best)
2486                 return 1;
2487         /* OK, load this ddf */
2488         sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2489         dfd = dev_open(nm, O_RDONLY);
2490         if (dfd < 0)
2491                 return 1;
2492         load_ddf_headers(dfd, super, NULL);
2493         load_ddf_global(dfd, super, NULL);
2494         close(dfd);
2495         /* Now we need the device-local bits */
2496         for (sd = sra->devs ; sd ; sd = sd->next) {
2497                 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2498                 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2499                 if (dfd < 0)
2500                         return 2;
2501                 seq = load_ddf_local(dfd, super, NULL, keep_fd);
2502                 if (!keep_fd) close(dfd);
2503         }
2504         if (st->subarray[0]) {
2505                 struct vcl *v;
2506
2507                 for (v = super->conflist; v; v = v->next)
2508                         if (v->vcnum == atoi(st->subarray))
2509                                 super->currentconf = v;
2510                 if (!super->currentconf)
2511                         return 1;
2512         }
2513         *sbp = super;
2514         if (st->ss == NULL) {
2515                 st->ss = &super_ddf;
2516                 st->minor_version = 0;
2517                 st->max_devs = 512;
2518                 st->container_dev = fd2devnum(fd);
2519         }
2520         return 0;
2521 }
2522 #endif
2523
2524 static struct mdinfo *container_content_ddf(struct supertype *st)
2525 {
2526         /* Given a container loaded by load_super_ddf_all,
2527          * extract information about all the arrays into
2528          * an mdinfo tree.
2529          *
2530          * For each vcl in conflist: create an mdinfo, fill it in,
2531          *  then look for matching devices (phys_refnum) in dlist
2532          *  and create appropriate device mdinfo.
2533          */
2534         struct ddf_super *ddf = st->sb;
2535         struct mdinfo *rest = NULL;
2536         struct vcl *vc;
2537
2538         for (vc = ddf->conflist ; vc ; vc=vc->next)
2539         {
2540                 int i;
2541                 struct mdinfo *this;
2542                 this = malloc(sizeof(*this));
2543                 memset(this, 0, sizeof(*this));
2544                 this->next = rest;
2545                 rest = this;
2546
2547                 this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2548                 this->array.raid_disks =
2549                         __be16_to_cpu(vc->conf.prim_elmnt_count);
2550                 this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2551                                                    this->array.raid_disks);
2552                 this->array.md_minor      = -1;
2553                 this->array.ctime         = DECADE +
2554                         __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
2555                 this->array.utime         = DECADE +
2556                         __be32_to_cpu(vc->conf.timestamp);
2557                 this->array.chunk_size    = 512 << vc->conf.chunk_shift;
2558
2559                 i = vc->vcnum;
2560                 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2561                     (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
2562                     DDF_init_full) {
2563                         this->array.state = 0;
2564                         this->resync_start = 0;
2565                 } else {
2566                         this->array.state = 1;
2567                         this->resync_start = ~0ULL;
2568                 }
2569                 memcpy(this->name, ddf->virt->entries[i].name, 32);
2570                 this->name[33]=0;
2571
2572                 memset(this->uuid, 0, sizeof(this->uuid));
2573                 this->component_size = __be64_to_cpu(vc->conf.blocks);
2574                 this->array.size = this->component_size / 2;
2575                 this->container_member = i;
2576
2577                 sprintf(this->text_version, "/%s/%d",
2578                         devnum2devname(st->container_dev),
2579                         this->container_member);
2580
2581                 for (i=0 ; i < ddf->mppe ; i++) {
2582                         struct mdinfo *dev;
2583                         struct dl *d;
2584
2585                         if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
2586                                 continue;
2587
2588                         this->array.working_disks++;
2589
2590                         for (d = ddf->dlist; d ; d=d->next)
2591                                 if (d->disk.refnum == vc->conf.phys_refnum[i])
2592                                         break;
2593                         if (d == NULL)
2594                                 break;
2595
2596                         dev = malloc(sizeof(*dev));
2597                         memset(dev, 0, sizeof(*dev));
2598                         dev->next = this->devs;
2599                         this->devs = dev;
2600
2601                         dev->disk.number = __be32_to_cpu(d->disk.refnum);
2602                         dev->disk.major = d->major;
2603                         dev->disk.minor = d->minor;
2604                         dev->disk.raid_disk = i;
2605                         dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2606
2607                         dev->events = __be32_to_cpu(ddf->primary.seq);
2608                         dev->data_offset = __be64_to_cpu(vc->lba_offset[i]);
2609                         dev->component_size = __be64_to_cpu(vc->conf.blocks);
2610                         if (d->devname)
2611                                 strcpy(dev->name, d->devname);
2612                 }
2613         }
2614         return rest;
2615 }
2616
2617 static int store_zero_ddf(struct supertype *st, int fd)
2618 {
2619         unsigned long long dsize;
2620         void *buf;
2621
2622         if (!get_dev_size(fd, NULL, &dsize))
2623                 return 1;
2624
2625         posix_memalign(&buf, 512, 512);
2626         memset(buf, 0, 512);
2627
2628         lseek64(fd, dsize-512, 0);
2629         write(fd, buf, 512);
2630         free(buf);
2631         return 0;
2632 }
2633
2634 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
2635 {
2636         /*
2637          * return:
2638          *  0 same, or first was empty, and second was copied
2639          *  1 second had wrong number
2640          *  2 wrong uuid
2641          *  3 wrong other info
2642          */
2643         struct ddf_super *first = st->sb;
2644         struct ddf_super *second = tst->sb;
2645
2646         if (!first) {
2647                 st->sb = tst->sb;
2648                 tst->sb = NULL;
2649                 return 0;
2650         }
2651
2652         if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
2653                 return 2;
2654
2655         /* FIXME should I look at anything else? */
2656         return 0;
2657 }
2658
2659 /*
2660  * A new array 'a' has been started which claims to be instance 'inst'
2661  * within container 'c'.
2662  * We need to confirm that the array matches the metadata in 'c' so
2663  * that we don't corrupt any metadata.
2664  */
2665 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
2666 {
2667         dprintf("ddf: open_new %s\n", inst);
2668         a->info.container_member = atoi(inst);
2669         return 0;
2670 }
2671
2672 /*
2673  * The array 'a' is to be marked clean in the metadata.
2674  * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2675  * clean up to the point (in sectors).  If that cannot be recorded in the
2676  * metadata, then leave it as dirty.
2677  *
2678  * For DDF, we need to clear the DDF_state_inconsistent bit in the
2679  * !global! virtual_disk.virtual_entry structure.
2680  */
2681 static void ddf_set_array_state(struct active_array *a, int consistent)
2682 {
2683         struct ddf_super *ddf = a->container->sb;
2684         int inst = a->info.container_member;
2685         int old = ddf->virt->entries[inst].state;
2686         if (consistent)
2687                 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
2688         else
2689                 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
2690         if (old != ddf->virt->entries[inst].state)
2691                 ddf->updates_pending = 1;
2692
2693         old = ddf->virt->entries[inst].init_state;
2694         ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
2695         if (a->resync_start == ~0ULL)
2696                 ddf->virt->entries[inst].init_state |= DDF_init_full;
2697         else if (a->resync_start == 0)
2698                 ddf->virt->entries[inst].init_state |= DDF_init_not;
2699         else
2700                 ddf->virt->entries[inst].init_state |= DDF_init_quick;
2701         if (old != ddf->virt->entries[inst].init_state)
2702                 ddf->updates_pending = 1;
2703
2704         dprintf("ddf mark %d %s %llu\n", inst, consistent?"clean":"dirty",
2705                 a->resync_start);
2706 }
2707
2708 /*
2709  * The state of each disk is stored in the global phys_disk structure
2710  * in phys_disk.entries[n].state.
2711  * This makes various combinations awkward.
2712  * - When a device fails in any array, it must be failed in all arrays
2713  *   that include a part of this device.
2714  * - When a component is rebuilding, we cannot include it officially in the
2715  *   array unless this is the only array that uses the device.
2716  *
2717  * So: when transitioning:
2718  *   Online -> failed,  just set failed flag.  monitor will propagate
2719  *   spare -> online,   the device might need to be added to the array.
2720  *   spare -> failed,   just set failed.  Don't worry if in array or not.
2721  */
2722 static void ddf_set_disk(struct active_array *a, int n, int state)
2723 {
2724         struct ddf_super *ddf = a->container->sb;
2725         int inst = a->info.container_member;
2726         struct vd_config *vc = find_vdcr(ddf, inst);
2727         int pd = find_phys(ddf, vc->phys_refnum[n]);
2728         int i, st, working;
2729
2730         if (vc == NULL) {
2731                 dprintf("ddf: cannot find instance %d!!\n", inst);
2732                 return;
2733         }
2734         if (pd < 0) {
2735                 /* disk doesn't currently exist. If it is now in_sync,
2736                  * insert it. */
2737                 if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
2738                         /* Find dev 'n' in a->info->devs, determine the
2739                          * ddf refnum, and set vc->phys_refnum and update
2740                          * phys->entries[]
2741                          */
2742                         /* FIXME */
2743                 }
2744         } else {
2745                 int old = ddf->phys->entries[pd].state;
2746                 if (state & DS_FAULTY)
2747                         ddf->phys->entries[pd].state  |= __cpu_to_be16(DDF_Failed);
2748                 if (state & DS_INSYNC) {
2749                         ddf->phys->entries[pd].state  |= __cpu_to_be16(DDF_Online);
2750                         ddf->phys->entries[pd].state  &= __cpu_to_be16(~DDF_Rebuilding);
2751                 }
2752                 if (old != ddf->phys->entries[pd].state)
2753                         ddf->updates_pending = 1;
2754         }
2755
2756         dprintf("ddf: set_disk %d to %x\n", n, state);
2757
2758         /* Now we need to check the state of the array and update
2759          * virtual_disk.entries[n].state.
2760          * It needs to be one of "optimal", "degraded", "failed".
2761          * I don't understand 'deleted' or 'missing'.
2762          */
2763         working = 0;
2764         for (i=0; i < a->info.array.raid_disks; i++) {
2765                 pd = find_phys(ddf, vc->phys_refnum[i]);
2766                 if (pd < 0)
2767                         continue;
2768                 st = __be16_to_cpu(ddf->phys->entries[pd].state);
2769                 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
2770                     == DDF_Online)
2771                         working++;
2772         }
2773         state = DDF_state_degraded;
2774         if (working == a->info.array.raid_disks)
2775                 state = DDF_state_optimal;
2776         else switch(vc->prl) {
2777         case DDF_RAID0:
2778         case DDF_CONCAT:
2779         case DDF_JBOD:
2780                 state = DDF_state_failed;
2781                 break;
2782         case DDF_RAID1:
2783                 if (working == 0)
2784                         state = DDF_state_failed;
2785                 break;
2786         case DDF_RAID4:
2787         case DDF_RAID5:
2788                 if (working < a->info.array.raid_disks-1)
2789                         state = DDF_state_failed;
2790                 break;
2791         case DDF_RAID6:
2792                 if (working < a->info.array.raid_disks-2)
2793                         state = DDF_state_failed;
2794                 else if (working == a->info.array.raid_disks-1)
2795                         state = DDF_state_part_optimal;
2796                 break;
2797         }
2798
2799         if (ddf->virt->entries[inst].state !=
2800             ((ddf->virt->entries[inst].state & ~DDF_state_mask)
2801              | state)) {
2802
2803                 ddf->virt->entries[inst].state =
2804                         (ddf->virt->entries[inst].state & ~DDF_state_mask)
2805                         | state;
2806                 ddf->updates_pending = 1;
2807         }
2808
2809 }
2810
2811 static void ddf_sync_metadata(struct supertype *st)
2812 {
2813
2814         /*
2815          * Write all data to all devices.
2816          * Later, we might be able to track whether only local changes
2817          * have been made, or whether any global data has been changed,
2818          * but ddf is sufficiently weird that it probably always
2819          * changes global data ....
2820          */
2821         struct ddf_super *ddf = st->sb;
2822         if (!ddf->updates_pending)
2823                 return;
2824         ddf->updates_pending = 0;
2825         __write_init_super_ddf(st, 0);
2826         dprintf("ddf: sync_metadata\n");
2827 }
2828
2829 static void ddf_process_update(struct supertype *st,
2830                                struct metadata_update *update)
2831 {
2832         /* Apply this update to the metadata.
2833          * The first 4 bytes are a DDF_*_MAGIC which guides
2834          * our actions.
2835          * Possible update are:
2836          *  DDF_PHYS_RECORDS_MAGIC
2837          *    Add a new physical device.  Changes to this record
2838          *    only happen implicitly.
2839          *    used_pdes is the device number.
2840          *  DDF_VIRT_RECORDS_MAGIC
2841          *    Add a new VD.  Possibly also change the 'access' bits.
2842          *    populated_vdes is the entry number.
2843          *  DDF_VD_CONF_MAGIC
2844          *    New or updated VD.  the VIRT_RECORD must already
2845          *    exist.  For an update, phys_refnum and lba_offset
2846          *    (at least) are updated, and the VD_CONF must
2847          *    be written to precisely those devices listed with
2848          *    a phys_refnum.
2849          *  DDF_SPARE_ASSIGN_MAGIC
2850          *    replacement Spare Assignment Record... but for which device?
2851          *
2852          * So, e.g.:
2853          *  - to create a new array, we send a VIRT_RECORD and
2854          *    a VD_CONF.  Then assemble and start the array.
2855          *  - to activate a spare we send a VD_CONF to add the phys_refnum
2856          *    and offset.  This will also mark the spare as active with
2857          *    a spare-assignment record.
2858          */
2859         struct ddf_super *ddf = st->sb;
2860         __u32 *magic = (__u32*)update->buf;
2861         struct phys_disk *pd;
2862         struct virtual_disk *vd;
2863         struct vd_config *vc;
2864         struct vcl *vcl;
2865         struct dl *dl;
2866         int mppe;
2867         int ent;
2868
2869         dprintf("Process update %x\n", *magic);
2870
2871         switch (*magic) {
2872         case DDF_PHYS_RECORDS_MAGIC:
2873
2874                 if (update->len != (sizeof(struct phys_disk) +
2875                                     sizeof(struct phys_disk_entry)))
2876                         return;
2877                 pd = (struct phys_disk*)update->buf;
2878
2879                 ent = __be16_to_cpu(pd->used_pdes);
2880                 if (ent >= __be16_to_cpu(ddf->phys->max_pdes))
2881                         return;
2882                 if (!all_ff(ddf->phys->entries[ent].guid))
2883                         return;
2884                 ddf->phys->entries[ent] = pd->entries[0];
2885                 ddf->phys->used_pdes = __cpu_to_be16(1 +
2886                                            __be16_to_cpu(ddf->phys->used_pdes));
2887                 ddf->updates_pending = 1;
2888                 break;
2889
2890         case DDF_VIRT_RECORDS_MAGIC:
2891
2892                 if (update->len != (sizeof(struct virtual_disk) +
2893                                     sizeof(struct virtual_entry)))
2894                         return;
2895                 vd = (struct virtual_disk*)update->buf;
2896
2897                 ent = __be16_to_cpu(vd->populated_vdes);
2898                 if (ent >= __be16_to_cpu(ddf->virt->max_vdes))
2899                         return;
2900                 if (!all_ff(ddf->virt->entries[ent].guid))
2901                         return;
2902                 ddf->virt->entries[ent] = vd->entries[0];
2903                 ddf->virt->populated_vdes = __cpu_to_be16(1 +
2904                               __be16_to_cpu(ddf->virt->populated_vdes));
2905                 ddf->updates_pending = 1;
2906                 break;
2907
2908         case DDF_VD_CONF_MAGIC:
2909                 dprintf("len %d %d\n", update->len, ddf->conf_rec_len);
2910
2911                 mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries);
2912                 if (update->len != ddf->conf_rec_len * 512)
2913                         return;
2914                 vc = (struct vd_config*)update->buf;
2915                 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2916                         if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
2917                                 break;
2918                 dprintf("vcl = %p\n", vcl);
2919                 if (vcl) {
2920                         /* An update, just copy the phys_refnum and lba_offset
2921                          * fields
2922                          */
2923                         memcpy(vcl->conf.phys_refnum, vc->phys_refnum,
2924                                mppe * (sizeof(__u32) + sizeof(__u64)));
2925                 } else {
2926                         /* A new VD_CONF */
2927                         vcl = update->space;
2928                         update->space = NULL;
2929                         vcl->next = ddf->conflist;
2930                         memcpy(&vcl->conf, vc, update->len);
2931                         vcl->lba_offset = (__u64*)
2932                                 &vcl->conf.phys_refnum[mppe];
2933                         ddf->conflist = vcl;
2934                 }
2935                 /* Now make sure vlist is correct for each dl. */
2936                 for (dl = ddf->dlist; dl; dl = dl->next) {
2937                         int dn;
2938                         int vn = 0;
2939                         for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2940                                 for (dn=0; dn < ddf->mppe ; dn++)
2941                                         if (vcl->conf.phys_refnum[dn] ==
2942                                             dl->disk.refnum) {
2943                                                 dprintf("dev %d has %p at %d\n",
2944                                                         dl->pdnum, vcl, vn);
2945                                                 dl->vlist[vn++] = vcl;
2946                                                 break;
2947                                         }
2948                         while (vn < ddf->max_part)
2949                                 dl->vlist[vn++] = NULL;
2950                         if (dl->vlist[0]) {
2951                                 ddf->phys->entries[dl->pdnum].type &=
2952                                         ~__cpu_to_be16(DDF_Global_Spare);
2953                                 ddf->phys->entries[dl->pdnum].type |=
2954                                         __cpu_to_be16(DDF_Active_in_VD);
2955                         }
2956                         if (dl->spare) {
2957                                 ddf->phys->entries[dl->pdnum].type &=
2958                                         ~__cpu_to_be16(DDF_Global_Spare);
2959                                 ddf->phys->entries[dl->pdnum].type |=
2960                                         __cpu_to_be16(DDF_Spare);
2961                         }
2962                         if (!dl->vlist[0] && !dl->spare) {
2963                                 ddf->phys->entries[dl->pdnum].type |=
2964                                         __cpu_to_be16(DDF_Global_Spare);
2965                                 ddf->phys->entries[dl->pdnum].type &=
2966                                         ~__cpu_to_be16(DDF_Spare |
2967                                                        DDF_Active_in_VD);
2968                         }
2969                 }
2970                 ddf->updates_pending = 1;
2971                 break;
2972         case DDF_SPARE_ASSIGN_MAGIC:
2973         default: break;
2974         }
2975 }
2976
2977 static void ddf_prepare_update(struct supertype *st,
2978                                struct metadata_update *update)
2979 {
2980         /* This update arrived at managemon.
2981          * We are about to pass it to monitor.
2982          * If a malloc is needed, do it here.
2983          */
2984         struct ddf_super *ddf = st->sb;
2985         __u32 *magic = (__u32*)update->buf;
2986         if (*magic == DDF_VD_CONF_MAGIC)
2987                 posix_memalign(&update->space, 512,
2988                                offsetof(struct vcl, conf)
2989                                + ddf->conf_rec_len * 512);
2990 }
2991
2992 /*
2993  * Check if the array 'a' is degraded but not failed.
2994  * If it is, find as many spares as are available and needed and
2995  * arrange for their inclusion.
2996  * We only choose devices which are not already in the array,
2997  * and prefer those with a spare-assignment to this array.
2998  * otherwise we choose global spares - assuming always that
2999  * there is enough room.
3000  * For each spare that we assign, we return an 'mdinfo' which
3001  * describes the position for the device in the array.
3002  * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
3003  * the new phys_refnum and lba_offset values.
3004  *
3005  * Only worry about BVDs at the moment.
3006  */
3007 static struct mdinfo *ddf_activate_spare(struct active_array *a,
3008                                          struct metadata_update **updates)
3009 {
3010         int working = 0;
3011         struct mdinfo *d;
3012         struct ddf_super *ddf = a->container->sb;
3013         int global_ok = 0;
3014         struct mdinfo *rv = NULL;
3015         struct mdinfo *di;
3016         struct metadata_update *mu;
3017         struct dl *dl;
3018         int i;
3019         struct vd_config *vc;
3020         __u64 *lba;
3021
3022         for (d = a->info.devs ; d ; d = d->next) {
3023                 if ((d->curr_state & DS_FAULTY) &&
3024                         d->state_fd >= 0)
3025                         /* wait for Removal to happen */
3026                         return NULL;
3027                 if (d->state_fd >= 0)
3028                         working ++;
3029         }
3030
3031         dprintf("ddf_activate: working=%d (%d) level=%d\n", working, a->info.array.raid_disks,
3032                 a->info.array.level);
3033         if (working == a->info.array.raid_disks)
3034                 return NULL; /* array not degraded */
3035         switch (a->info.array.level) {
3036         case 1:
3037                 if (working == 0)
3038                         return NULL; /* failed */
3039                 break;
3040         case 4:
3041         case 5:
3042                 if (working < a->info.array.raid_disks - 1)
3043                         return NULL; /* failed */
3044                 break;
3045         case 6:
3046                 if (working < a->info.array.raid_disks - 2)
3047                         return NULL; /* failed */
3048                 break;
3049         default: /* concat or stripe */
3050                 return NULL; /* failed */
3051         }
3052
3053         /* For each slot, if it is not working, find a spare */
3054         dl = ddf->dlist;
3055         for (i = 0; i < a->info.array.raid_disks; i++) {
3056                 for (d = a->info.devs ; d ; d = d->next)
3057                         if (d->disk.raid_disk == i)
3058                                 break;
3059                 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
3060                 if (d && (d->state_fd >= 0))
3061                         continue;
3062
3063                 /* OK, this device needs recovery.  Find a spare */
3064         again:
3065                 for ( ; dl ; dl = dl->next) {
3066                         unsigned long long esize;
3067                         unsigned long long pos;
3068                         struct mdinfo *d2;
3069                         int is_global = 0;
3070                         int is_dedicated = 0;
3071                         struct extent *ex;
3072                         int j;
3073                         /* If in this array, skip */
3074                         for (d2 = a->info.devs ; d2 ; d2 = d2->next)
3075                                 if (d2->disk.major == dl->major &&
3076                                     d2->disk.minor == dl->minor) {
3077                                         dprintf("%x:%x already in array\n", dl->major, dl->minor);
3078                                         break;
3079                                 }
3080                         if (d2)
3081                                 continue;
3082                         if (ddf->phys->entries[dl->pdnum].type &
3083                             __cpu_to_be16(DDF_Spare)) {
3084                                 /* Check spare assign record */
3085                                 if (dl->spare) {
3086                                         if (dl->spare->type & DDF_spare_dedicated) {
3087                                                 /* check spare_ents for guid */
3088                                                 for (j = 0 ;
3089                                                      j < __be16_to_cpu(dl->spare->populated);
3090                                                      j++) {
3091                                                         if (memcmp(dl->spare->spare_ents[j].guid,
3092                                                                    ddf->virt->entries[a->info.container_member].guid,
3093                                                                    DDF_GUID_LEN) == 0)
3094                                                                 is_dedicated = 1;
3095                                                 }
3096                                         } else
3097                                                 is_global = 1;
3098                                 }
3099                         } else if (ddf->phys->entries[dl->pdnum].type &
3100                                    __cpu_to_be16(DDF_Global_Spare)) {
3101                                 is_global = 1;
3102                         }
3103                         if ( ! (is_dedicated ||
3104                                 (is_global && global_ok))) {
3105                                 dprintf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
3106                                        is_dedicated, is_global);
3107                                 continue;
3108                         }
3109
3110                         /* We are allowed to use this device - is there space?
3111                          * We need a->info.component_size sectors */
3112                         ex = get_extents(ddf, dl);
3113                         if (!ex) {
3114                                 dprintf("cannot get extents\n");
3115                                 continue;
3116                         }
3117                         j = 0; pos = 0;
3118                         esize = 0;
3119
3120                         do {
3121                                 esize = ex[j].start - pos;
3122                                 if (esize >= a->info.component_size)
3123                                         break;
3124                                 pos = ex[i].start + ex[i].size;
3125                                 i++;
3126                         } while (ex[i-1].size);
3127
3128                         free(ex);
3129                         if (esize < a->info.component_size) {
3130                                 dprintf("%x:%x has no room: %llu %llu\n", dl->major, dl->minor,
3131                                         esize, a->info.component_size);
3132                                 /* No room */
3133                                 continue;
3134                         }
3135
3136                         /* Cool, we have a device with some space at pos */
3137                         di = malloc(sizeof(*di));
3138                         memset(di, 0, sizeof(*di));
3139                         di->disk.number = i;
3140                         di->disk.raid_disk = i;
3141                         di->disk.major = dl->major;
3142                         di->disk.minor = dl->minor;
3143                         di->disk.state = 0;
3144                         di->data_offset = pos;
3145                         di->component_size = a->info.component_size;
3146                         di->container_member = dl->pdnum;
3147                         di->next = rv;
3148                         rv = di;
3149                         dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3150                                 i, pos);
3151
3152                         break;
3153                 }
3154                 if (!dl && ! global_ok) {
3155                         /* not enough dedicated spares, try global */
3156                         global_ok = 1;
3157                         dl = ddf->dlist;
3158                         goto again;
3159                 }
3160         }
3161
3162         if (!rv)
3163                 /* No spares found */
3164                 return rv;
3165         /* Now 'rv' has a list of devices to return.
3166          * Create a metadata_update record to update the
3167          * phys_refnum and lba_offset values
3168          */
3169         mu = malloc(sizeof(*mu));
3170         mu->buf = malloc(ddf->conf_rec_len * 512);
3171         posix_memalign(&mu->space, 512, sizeof(struct vcl));
3172         mu->len = ddf->conf_rec_len;
3173         mu->next = *updates;
3174         vc = find_vdcr(ddf, a->info.container_member);
3175         memcpy(mu->buf, vc, ddf->conf_rec_len * 512);
3176
3177         vc = (struct vd_config*)mu->buf;
3178         lba = (__u64*)&vc->phys_refnum[ddf->mppe];
3179         for (di = rv ; di ; di = di->next) {
3180                 vc->phys_refnum[di->disk.raid_disk] =
3181                         ddf->phys->entries[dl->pdnum].refnum;
3182                 lba[di->disk.raid_disk] = di->data_offset;
3183         }
3184         *updates = mu;
3185         return rv;
3186 }
3187
3188 struct superswitch super_ddf = {
3189 #ifndef MDASSEMBLE
3190         .examine_super  = examine_super_ddf,
3191         .brief_examine_super = brief_examine_super_ddf,
3192         .detail_super   = detail_super_ddf,
3193         .brief_detail_super = brief_detail_super_ddf,
3194         .validate_geometry = validate_geometry_ddf,
3195         .write_init_super = write_init_super_ddf,
3196 #endif
3197         .match_home     = match_home_ddf,
3198         .uuid_from_super= uuid_from_super_ddf,
3199         .getinfo_super  = getinfo_super_ddf,
3200         .update_super   = update_super_ddf,
3201
3202         .avail_size     = avail_size_ddf,
3203
3204         .compare_super  = compare_super_ddf,
3205
3206         .load_super     = load_super_ddf,
3207         .init_super     = init_super_ddf,
3208         .store_super    = store_zero_ddf,
3209         .free_super     = free_super_ddf,
3210         .match_metadata_desc = match_metadata_desc_ddf,
3211         .add_to_super   = add_to_super_ddf,
3212         .container_content = container_content_ddf,
3213
3214         .external       = 1,
3215
3216 /* for mdmon */
3217         .open_new       = ddf_open_new,
3218         .set_array_state= ddf_set_array_state,
3219         .set_disk       = ddf_set_disk,
3220         .sync_metadata  = ddf_sync_metadata,
3221         .process_update = ddf_process_update,
3222         .prepare_update = ddf_prepare_update,
3223         .activate_spare = ddf_activate_spare,
3224
3225 };