]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super-ddf.c
Report uuid in --detail --brief for ddf and intel
[thirdparty/mdadm.git] / super-ddf.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2006-2007 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neil@brown.name>
23 *
24 * Specifications for DDF takes from Common RAID DDF Specification Revision 1.2
25 * (July 28 2006). Reused by permission of SNIA.
26 */
27
28 #define HAVE_STDINT_H 1
29 #include "mdadm.h"
30 #include "mdmon.h"
31 #include "sha1.h"
32 #include <values.h>
33
34 /* a non-official T10 name for creation GUIDs */
35 static char T10[] = "Linux-MD";
36
37 /* DDF timestamps are 1980 based, so we need to add
38 * second-in-decade-of-seventies to convert to linux timestamps.
39 * 10 years with 2 leap years.
40 */
41 #define DECADE (3600*24*(365*10+2))
42 unsigned long crc32(
43 unsigned long crc,
44 const unsigned char *buf,
45 unsigned len);
46
47 /* The DDF metadata handling.
48 * DDF metadata lives at the end of the device.
49 * The last 512 byte block provides an 'anchor' which is used to locate
50 * the rest of the metadata which usually lives immediately behind the anchor.
51 *
52 * Note:
53 * - all multibyte numeric fields are bigendian.
54 * - all strings are space padded.
55 *
56 */
57
58 /* Primary Raid Level (PRL) */
59 #define DDF_RAID0 0x00
60 #define DDF_RAID1 0x01
61 #define DDF_RAID3 0x03
62 #define DDF_RAID4 0x04
63 #define DDF_RAID5 0x05
64 #define DDF_RAID1E 0x11
65 #define DDF_JBOD 0x0f
66 #define DDF_CONCAT 0x1f
67 #define DDF_RAID5E 0x15
68 #define DDF_RAID5EE 0x25
69 #define DDF_RAID6 0x06
70
71 /* Raid Level Qualifier (RLQ) */
72 #define DDF_RAID0_SIMPLE 0x00
73 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
74 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
75 #define DDF_RAID3_0 0x00 /* parity in first extent */
76 #define DDF_RAID3_N 0x01 /* parity in last extent */
77 #define DDF_RAID4_0 0x00 /* parity in first extent */
78 #define DDF_RAID4_N 0x01 /* parity in last extent */
79 /* these apply to raid5e and raid5ee as well */
80 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
81 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
82 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
83 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
84
85 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
86 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
87
88 /* Secondary RAID Level (SRL) */
89 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
90 #define DDF_2MIRRORED 0x01
91 #define DDF_2CONCAT 0x02
92 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
93
94 /* Magic numbers */
95 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
96 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
97 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
98 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
99 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
100 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
101 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
102 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
103 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
104 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
105
106 #define DDF_GUID_LEN 24
107 #define DDF_REVISION_0 "01.00.00"
108 #define DDF_REVISION_2 "01.02.00"
109
110 struct ddf_header {
111 __u32 magic; /* DDF_HEADER_MAGIC */
112 __u32 crc;
113 char guid[DDF_GUID_LEN];
114 char revision[8]; /* 01.02.00 */
115 __u32 seq; /* starts at '1' */
116 __u32 timestamp;
117 __u8 openflag;
118 __u8 foreignflag;
119 __u8 enforcegroups;
120 __u8 pad0; /* 0xff */
121 __u8 pad1[12]; /* 12 * 0xff */
122 /* 64 bytes so far */
123 __u8 header_ext[32]; /* reserved: fill with 0xff */
124 __u64 primary_lba;
125 __u64 secondary_lba;
126 __u8 type;
127 __u8 pad2[3]; /* 0xff */
128 __u32 workspace_len; /* sectors for vendor space -
129 * at least 32768(sectors) */
130 __u64 workspace_lba;
131 __u16 max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */
132 __u16 max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */
133 __u16 max_partitions; /* i.e. max num of configuration
134 record entries per disk */
135 __u16 config_record_len; /* 1 +ROUNDUP(max_primary_element_entries
136 *12/512) */
137 __u16 max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */
138 __u8 pad3[54]; /* 0xff */
139 /* 192 bytes so far */
140 __u32 controller_section_offset;
141 __u32 controller_section_length;
142 __u32 phys_section_offset;
143 __u32 phys_section_length;
144 __u32 virt_section_offset;
145 __u32 virt_section_length;
146 __u32 config_section_offset;
147 __u32 config_section_length;
148 __u32 data_section_offset;
149 __u32 data_section_length;
150 __u32 bbm_section_offset;
151 __u32 bbm_section_length;
152 __u32 diag_space_offset;
153 __u32 diag_space_length;
154 __u32 vendor_offset;
155 __u32 vendor_length;
156 /* 256 bytes so far */
157 __u8 pad4[256]; /* 0xff */
158 };
159
160 /* type field */
161 #define DDF_HEADER_ANCHOR 0x00
162 #define DDF_HEADER_PRIMARY 0x01
163 #define DDF_HEADER_SECONDARY 0x02
164
165 /* The content of the 'controller section' - global scope */
166 struct ddf_controller_data {
167 __u32 magic; /* DDF_CONTROLLER_MAGIC */
168 __u32 crc;
169 char guid[DDF_GUID_LEN];
170 struct controller_type {
171 __u16 vendor_id;
172 __u16 device_id;
173 __u16 sub_vendor_id;
174 __u16 sub_device_id;
175 } type;
176 char product_id[16];
177 __u8 pad[8]; /* 0xff */
178 __u8 vendor_data[448];
179 };
180
181 /* The content of phys_section - global scope */
182 struct phys_disk {
183 __u32 magic; /* DDF_PHYS_RECORDS_MAGIC */
184 __u32 crc;
185 __u16 used_pdes;
186 __u16 max_pdes;
187 __u8 pad[52];
188 struct phys_disk_entry {
189 char guid[DDF_GUID_LEN];
190 __u32 refnum;
191 __u16 type;
192 __u16 state;
193 __u64 config_size; /* DDF structures must be after here */
194 char path[18]; /* another horrible structure really */
195 __u8 pad[6];
196 } entries[0];
197 };
198
199 /* phys_disk_entry.type is a bitmap - bigendian remember */
200 #define DDF_Forced_PD_GUID 1
201 #define DDF_Active_in_VD 2
202 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
203 #define DDF_Spare 8 /* overrides Global_spare */
204 #define DDF_Foreign 16
205 #define DDF_Legacy 32 /* no DDF on this device */
206
207 #define DDF_Interface_mask 0xf00
208 #define DDF_Interface_SCSI 0x100
209 #define DDF_Interface_SAS 0x200
210 #define DDF_Interface_SATA 0x300
211 #define DDF_Interface_FC 0x400
212
213 /* phys_disk_entry.state is a bigendian bitmap */
214 #define DDF_Online 1
215 #define DDF_Failed 2 /* overrides 1,4,8 */
216 #define DDF_Rebuilding 4
217 #define DDF_Transition 8
218 #define DDF_SMART 16
219 #define DDF_ReadErrors 32
220 #define DDF_Missing 64
221
222 /* The content of the virt_section global scope */
223 struct virtual_disk {
224 __u32 magic; /* DDF_VIRT_RECORDS_MAGIC */
225 __u32 crc;
226 __u16 populated_vdes;
227 __u16 max_vdes;
228 __u8 pad[52];
229 struct virtual_entry {
230 char guid[DDF_GUID_LEN];
231 __u16 unit;
232 __u16 pad0; /* 0xffff */
233 __u16 guid_crc;
234 __u16 type;
235 __u8 state;
236 __u8 init_state;
237 __u8 pad1[14];
238 char name[16];
239 } entries[0];
240 };
241
242 /* virtual_entry.type is a bitmap - bigendian */
243 #define DDF_Shared 1
244 #define DDF_Enforce_Groups 2
245 #define DDF_Unicode 4
246 #define DDF_Owner_Valid 8
247
248 /* virtual_entry.state is a bigendian bitmap */
249 #define DDF_state_mask 0x7
250 #define DDF_state_optimal 0x0
251 #define DDF_state_degraded 0x1
252 #define DDF_state_deleted 0x2
253 #define DDF_state_missing 0x3
254 #define DDF_state_failed 0x4
255 #define DDF_state_part_optimal 0x5
256
257 #define DDF_state_morphing 0x8
258 #define DDF_state_inconsistent 0x10
259
260 /* virtual_entry.init_state is a bigendian bitmap */
261 #define DDF_initstate_mask 0x03
262 #define DDF_init_not 0x00
263 #define DDF_init_quick 0x01 /* initialisation is progress.
264 * i.e. 'state_inconsistent' */
265 #define DDF_init_full 0x02
266
267 #define DDF_access_mask 0xc0
268 #define DDF_access_rw 0x00
269 #define DDF_access_ro 0x80
270 #define DDF_access_blocked 0xc0
271
272 /* The content of the config_section - local scope
273 * It has multiple records each config_record_len sectors
274 * They can be vd_config or spare_assign
275 */
276
277 struct vd_config {
278 __u32 magic; /* DDF_VD_CONF_MAGIC */
279 __u32 crc;
280 char guid[DDF_GUID_LEN];
281 __u32 timestamp;
282 __u32 seqnum;
283 __u8 pad0[24];
284 __u16 prim_elmnt_count;
285 __u8 chunk_shift; /* 0 == 512, 1==1024 etc */
286 __u8 prl;
287 __u8 rlq;
288 __u8 sec_elmnt_count;
289 __u8 sec_elmnt_seq;
290 __u8 srl;
291 __u64 blocks; /* blocks per component could be different
292 * on different component devices...(only
293 * for concat I hope) */
294 __u64 array_blocks; /* blocks in array */
295 __u8 pad1[8];
296 __u32 spare_refs[8];
297 __u8 cache_pol[8];
298 __u8 bg_rate;
299 __u8 pad2[3];
300 __u8 pad3[52];
301 __u8 pad4[192];
302 __u8 v0[32]; /* reserved- 0xff */
303 __u8 v1[32]; /* reserved- 0xff */
304 __u8 v2[16]; /* reserved- 0xff */
305 __u8 v3[16]; /* reserved- 0xff */
306 __u8 vendor[32];
307 __u32 phys_refnum[0]; /* refnum of each disk in sequence */
308 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
309 bvd are always the same size */
310 };
311
312 /* vd_config.cache_pol[7] is a bitmap */
313 #define DDF_cache_writeback 1 /* else writethrough */
314 #define DDF_cache_wadaptive 2 /* only applies if writeback */
315 #define DDF_cache_readahead 4
316 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
317 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
318 #define DDF_cache_wallowed 32 /* enable write caching */
319 #define DDF_cache_rallowed 64 /* enable read caching */
320
321 struct spare_assign {
322 __u32 magic; /* DDF_SPARE_ASSIGN_MAGIC */
323 __u32 crc;
324 __u32 timestamp;
325 __u8 reserved[7];
326 __u8 type;
327 __u16 populated; /* SAEs used */
328 __u16 max; /* max SAEs */
329 __u8 pad[8];
330 struct spare_assign_entry {
331 char guid[DDF_GUID_LEN];
332 __u16 secondary_element;
333 __u8 pad[6];
334 } spare_ents[0];
335 };
336 /* spare_assign.type is a bitmap */
337 #define DDF_spare_dedicated 0x1 /* else global */
338 #define DDF_spare_revertible 0x2 /* else committable */
339 #define DDF_spare_active 0x4 /* else not active */
340 #define DDF_spare_affinity 0x8 /* enclosure affinity */
341
342 /* The data_section contents - local scope */
343 struct disk_data {
344 __u32 magic; /* DDF_PHYS_DATA_MAGIC */
345 __u32 crc;
346 char guid[DDF_GUID_LEN];
347 __u32 refnum; /* crc of some magic drive data ... */
348 __u8 forced_ref; /* set when above was not result of magic */
349 __u8 forced_guid; /* set if guid was forced rather than magic */
350 __u8 vendor[32];
351 __u8 pad[442];
352 };
353
354 /* bbm_section content */
355 struct bad_block_log {
356 __u32 magic;
357 __u32 crc;
358 __u16 entry_count;
359 __u32 spare_count;
360 __u8 pad[10];
361 __u64 first_spare;
362 struct mapped_block {
363 __u64 defective_start;
364 __u32 replacement_start;
365 __u16 remap_count;
366 __u8 pad[2];
367 } entries[0];
368 };
369
370 /* Struct for internally holding ddf structures */
371 /* The DDF structure stored on each device is potentially
372 * quite different, as some data is global and some is local.
373 * The global data is:
374 * - ddf header
375 * - controller_data
376 * - Physical disk records
377 * - Virtual disk records
378 * The local data is:
379 * - Configuration records
380 * - Physical Disk data section
381 * ( and Bad block and vendor which I don't care about yet).
382 *
383 * The local data is parsed into separate lists as it is read
384 * and reconstructed for writing. This means that we only need
385 * to make config changes once and they are automatically
386 * propagated to all devices.
387 * Note that the ddf_super has space of the conf and disk data
388 * for this disk and also for a list of all such data.
389 * The list is only used for the superblock that is being
390 * built in Create or Assemble to describe the whole array.
391 */
392 struct ddf_super {
393 struct ddf_header anchor, primary, secondary;
394 struct ddf_controller_data controller;
395 struct ddf_header *active;
396 struct phys_disk *phys;
397 struct virtual_disk *virt;
398 int pdsize, vdsize;
399 int max_part, mppe, conf_rec_len;
400 int currentdev;
401 int updates_pending;
402 struct vcl {
403 union {
404 char space[512];
405 struct {
406 struct vcl *next;
407 __u64 *lba_offset; /* location in 'conf' of
408 * the lba table */
409 int vcnum; /* index into ->virt */
410 __u64 *block_sizes; /* NULL if all the same */
411 };
412 };
413 struct vd_config conf;
414 } *conflist, *currentconf;
415 struct dl {
416 union {
417 char space[512];
418 struct {
419 struct dl *next;
420 int major, minor;
421 char *devname;
422 int fd;
423 unsigned long long size; /* sectors */
424 int pdnum; /* index in ->phys */
425 struct spare_assign *spare;
426 };
427 };
428 struct disk_data disk;
429 void *mdupdate; /* hold metadata update */
430 struct vcl *vlist[0]; /* max_part in size */
431 } *dlist, *add_list;
432 };
433
434 #ifndef offsetof
435 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
436 #endif
437
438
439 static int calc_crc(void *buf, int len)
440 {
441 /* crcs are always at the same place as in the ddf_header */
442 struct ddf_header *ddf = buf;
443 __u32 oldcrc = ddf->crc;
444 __u32 newcrc;
445 ddf->crc = 0xffffffff;
446
447 newcrc = crc32(0, buf, len);
448 ddf->crc = oldcrc;
449 return newcrc;
450 }
451
452 static int load_ddf_header(int fd, unsigned long long lba,
453 unsigned long long size,
454 int type,
455 struct ddf_header *hdr, struct ddf_header *anchor)
456 {
457 /* read a ddf header (primary or secondary) from fd/lba
458 * and check that it is consistent with anchor
459 * Need to check:
460 * magic, crc, guid, rev, and LBA's header_type, and
461 * everything after header_type must be the same
462 */
463 if (lba >= size-1)
464 return 0;
465
466 if (lseek64(fd, lba<<9, 0) < 0)
467 return 0;
468
469 if (read(fd, hdr, 512) != 512)
470 return 0;
471
472 if (hdr->magic != DDF_HEADER_MAGIC)
473 return 0;
474 if (calc_crc(hdr, 512) != hdr->crc)
475 return 0;
476 if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
477 memcmp(anchor->revision, hdr->revision, 8) != 0 ||
478 anchor->primary_lba != hdr->primary_lba ||
479 anchor->secondary_lba != hdr->secondary_lba ||
480 hdr->type != type ||
481 memcmp(anchor->pad2, hdr->pad2, 512 -
482 offsetof(struct ddf_header, pad2)) != 0)
483 return 0;
484
485 /* Looks good enough to me... */
486 return 1;
487 }
488
489 static void *load_section(int fd, struct ddf_super *super, void *buf,
490 __u32 offset_be, __u32 len_be, int check)
491 {
492 unsigned long long offset = __be32_to_cpu(offset_be);
493 unsigned long long len = __be32_to_cpu(len_be);
494 int dofree = (buf == NULL);
495
496 if (check)
497 if (len != 2 && len != 8 && len != 32
498 && len != 128 && len != 512)
499 return NULL;
500
501 if (len > 1024)
502 return NULL;
503 if (buf) {
504 /* All pre-allocated sections are a single block */
505 if (len != 1)
506 return NULL;
507 } else {
508 posix_memalign(&buf, 512, len<<9);
509 }
510
511 if (!buf)
512 return NULL;
513
514 if (super->active->type == 1)
515 offset += __be64_to_cpu(super->active->primary_lba);
516 else
517 offset += __be64_to_cpu(super->active->secondary_lba);
518
519 if (lseek64(fd, offset<<9, 0) != (offset<<9)) {
520 if (dofree)
521 free(buf);
522 return NULL;
523 }
524 if (read(fd, buf, len<<9) != (len<<9)) {
525 if (dofree)
526 free(buf);
527 return NULL;
528 }
529 return buf;
530 }
531
532 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
533 {
534 unsigned long long dsize;
535
536 get_dev_size(fd, NULL, &dsize);
537
538 if (lseek64(fd, dsize-512, 0) < 0) {
539 if (devname)
540 fprintf(stderr,
541 Name": Cannot seek to anchor block on %s: %s\n",
542 devname, strerror(errno));
543 return 1;
544 }
545 if (read(fd, &super->anchor, 512) != 512) {
546 if (devname)
547 fprintf(stderr,
548 Name ": Cannot read anchor block on %s: %s\n",
549 devname, strerror(errno));
550 return 1;
551 }
552 if (super->anchor.magic != DDF_HEADER_MAGIC) {
553 if (devname)
554 fprintf(stderr, Name ": no DDF anchor found on %s\n",
555 devname);
556 return 2;
557 }
558 if (calc_crc(&super->anchor, 512) != super->anchor.crc) {
559 if (devname)
560 fprintf(stderr, Name ": bad CRC on anchor on %s\n",
561 devname);
562 return 2;
563 }
564 if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
565 memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
566 if (devname)
567 fprintf(stderr, Name ": can only support super revision"
568 " %.8s and earlier, not %.8s on %s\n",
569 DDF_REVISION_2, super->anchor.revision,devname);
570 return 2;
571 }
572 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba),
573 dsize >> 9, 1,
574 &super->primary, &super->anchor) == 0) {
575 if (devname)
576 fprintf(stderr,
577 Name ": Failed to load primary DDF header "
578 "on %s\n", devname);
579 return 2;
580 }
581 super->active = &super->primary;
582 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.secondary_lba),
583 dsize >> 9, 2,
584 &super->secondary, &super->anchor)) {
585 if ((__be32_to_cpu(super->primary.seq)
586 < __be32_to_cpu(super->secondary.seq) &&
587 !super->secondary.openflag)
588 || (__be32_to_cpu(super->primary.seq)
589 == __be32_to_cpu(super->secondary.seq) &&
590 super->primary.openflag && !super->secondary.openflag)
591 )
592 super->active = &super->secondary;
593 }
594 return 0;
595 }
596
597 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
598 {
599 void *ok;
600 ok = load_section(fd, super, &super->controller,
601 super->active->controller_section_offset,
602 super->active->controller_section_length,
603 0);
604 super->phys = load_section(fd, super, NULL,
605 super->active->phys_section_offset,
606 super->active->phys_section_length,
607 1);
608 super->pdsize = __be32_to_cpu(super->active->phys_section_length) * 512;
609
610 super->virt = load_section(fd, super, NULL,
611 super->active->virt_section_offset,
612 super->active->virt_section_length,
613 1);
614 super->vdsize = __be32_to_cpu(super->active->virt_section_length) * 512;
615 if (!ok ||
616 !super->phys ||
617 !super->virt) {
618 free(super->phys);
619 free(super->virt);
620 super->phys = NULL;
621 super->virt = NULL;
622 return 2;
623 }
624 super->conflist = NULL;
625 super->dlist = NULL;
626
627 super->max_part = __be16_to_cpu(super->active->max_partitions);
628 super->mppe = __be16_to_cpu(super->active->max_primary_element_entries);
629 super->conf_rec_len = __be16_to_cpu(super->active->config_record_len);
630 return 0;
631 }
632
633 static int load_ddf_local(int fd, struct ddf_super *super,
634 char *devname, int keep)
635 {
636 struct dl *dl;
637 struct stat stb;
638 char *conf;
639 int i;
640 int vnum;
641 int max_virt_disks = __be16_to_cpu(super->active->max_vd_entries);
642 unsigned long long dsize;
643
644 /* First the local disk info */
645 posix_memalign((void**)&dl, 512,
646 sizeof(*dl) +
647 (super->max_part) * sizeof(dl->vlist[0]));
648
649 load_section(fd, super, &dl->disk,
650 super->active->data_section_offset,
651 super->active->data_section_length,
652 0);
653 dl->devname = devname ? strdup(devname) : NULL;
654
655 fstat(fd, &stb);
656 dl->major = major(stb.st_rdev);
657 dl->minor = minor(stb.st_rdev);
658 dl->next = super->dlist;
659 dl->fd = keep ? fd : -1;
660
661 dl->size = 0;
662 if (get_dev_size(fd, devname, &dsize))
663 dl->size = dsize >> 9;
664 dl->spare = NULL;
665 for (i=0 ; i < super->max_part ; i++)
666 dl->vlist[i] = NULL;
667 super->dlist = dl;
668 dl->pdnum = -1;
669 for (i=0; i < __be16_to_cpu(super->active->max_pd_entries); i++)
670 if (memcmp(super->phys->entries[i].guid,
671 dl->disk.guid, DDF_GUID_LEN) == 0)
672 dl->pdnum = i;
673
674 /* Now the config list. */
675 /* 'conf' is an array of config entries, some of which are
676 * probably invalid. Those which are good need to be copied into
677 * the conflist
678 */
679
680 conf = load_section(fd, super, NULL,
681 super->active->config_section_offset,
682 super->active->config_section_length,
683 0);
684
685 vnum = 0;
686 for (i = 0;
687 i < __be32_to_cpu(super->active->config_section_length);
688 i += super->conf_rec_len) {
689 struct vd_config *vd =
690 (struct vd_config *)((char*)conf + i*512);
691 struct vcl *vcl;
692
693 if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) {
694 if (dl->spare)
695 continue;
696 posix_memalign((void**)&dl->spare, 512,
697 super->conf_rec_len*512);
698 memcpy(dl->spare, vd, super->conf_rec_len*512);
699 continue;
700 }
701 if (vd->magic != DDF_VD_CONF_MAGIC)
702 continue;
703 for (vcl = super->conflist; vcl; vcl = vcl->next) {
704 if (memcmp(vcl->conf.guid,
705 vd->guid, DDF_GUID_LEN) == 0)
706 break;
707 }
708
709 if (vcl) {
710 dl->vlist[vnum++] = vcl;
711 if (__be32_to_cpu(vd->seqnum) <=
712 __be32_to_cpu(vcl->conf.seqnum))
713 continue;
714 } else {
715 posix_memalign((void**)&vcl, 512,
716 (super->conf_rec_len*512 +
717 offsetof(struct vcl, conf)));
718 vcl->next = super->conflist;
719 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
720 super->conflist = vcl;
721 dl->vlist[vnum++] = vcl;
722 }
723 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
724 vcl->lba_offset = (__u64*)
725 &vcl->conf.phys_refnum[super->mppe];
726
727 for (i=0; i < max_virt_disks ; i++)
728 if (memcmp(super->virt->entries[i].guid,
729 vcl->conf.guid, DDF_GUID_LEN)==0)
730 break;
731 if (i < max_virt_disks)
732 vcl->vcnum = i;
733 }
734 free(conf);
735
736 return 0;
737 }
738
739 #ifndef MDASSEMBLE
740 static int load_super_ddf_all(struct supertype *st, int fd,
741 void **sbp, char *devname, int keep_fd);
742 #endif
743 static int load_super_ddf(struct supertype *st, int fd,
744 char *devname)
745 {
746 unsigned long long dsize;
747 struct ddf_super *super;
748 int rv;
749
750 #ifndef MDASSEMBLE
751 /* if 'fd' is a container, load metadata from all the devices */
752 if (load_super_ddf_all(st, fd, &st->sb, devname, 1) == 0)
753 return 0;
754 #endif
755 if (st->subarray[0])
756 return 1; /* FIXME Is this correct */
757
758 if (get_dev_size(fd, devname, &dsize) == 0)
759 return 1;
760
761 /* 32M is a lower bound */
762 if (dsize <= 32*1024*1024) {
763 if (devname) {
764 fprintf(stderr,
765 Name ": %s is too small for ddf: "
766 "size is %llu sectors.\n",
767 devname, dsize>>9);
768 return 1;
769 }
770 }
771 if (dsize & 511) {
772 if (devname) {
773 fprintf(stderr,
774 Name ": %s is an odd size for ddf: "
775 "size is %llu bytes.\n",
776 devname, dsize);
777 return 1;
778 }
779 }
780
781 if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) {
782 fprintf(stderr, Name ": malloc of %zu failed.\n",
783 sizeof(*super));
784 return 1;
785 }
786 memset(super, 0, sizeof(*super));
787
788 rv = load_ddf_headers(fd, super, devname);
789 if (rv) {
790 free(super);
791 return rv;
792 }
793
794 /* Have valid headers and have chosen the best. Let's read in the rest*/
795
796 rv = load_ddf_global(fd, super, devname);
797
798 if (rv) {
799 if (devname)
800 fprintf(stderr,
801 Name ": Failed to load all information "
802 "sections on %s\n", devname);
803 free(super);
804 return rv;
805 }
806
807 load_ddf_local(fd, super, devname, 0);
808
809 /* Should possibly check the sections .... */
810
811 st->sb = super;
812 if (st->ss == NULL) {
813 st->ss = &super_ddf;
814 st->minor_version = 0;
815 st->max_devs = 512;
816 }
817 st->loaded_container = 0;
818 return 0;
819
820 }
821
822 static void free_super_ddf(struct supertype *st)
823 {
824 struct ddf_super *ddf = st->sb;
825 if (ddf == NULL)
826 return;
827 free(ddf->phys);
828 free(ddf->virt);
829 while (ddf->conflist) {
830 struct vcl *v = ddf->conflist;
831 ddf->conflist = v->next;
832 if (v->block_sizes)
833 free(v->block_sizes);
834 free(v);
835 }
836 while (ddf->dlist) {
837 struct dl *d = ddf->dlist;
838 ddf->dlist = d->next;
839 if (d->fd >= 0)
840 close(d->fd);
841 if (d->spare)
842 free(d->spare);
843 free(d);
844 }
845 free(ddf);
846 st->sb = NULL;
847 }
848
849 static struct supertype *match_metadata_desc_ddf(char *arg)
850 {
851 /* 'ddf' only support containers */
852 struct supertype *st;
853 if (strcmp(arg, "ddf") != 0 &&
854 strcmp(arg, "default") != 0
855 )
856 return NULL;
857
858 st = malloc(sizeof(*st));
859 memset(st, 0, sizeof(*st));
860 st->ss = &super_ddf;
861 st->max_devs = 512;
862 st->minor_version = 0;
863 st->sb = NULL;
864 return st;
865 }
866
867
868 #ifndef MDASSEMBLE
869
870 static mapping_t ddf_state[] = {
871 { "Optimal", 0},
872 { "Degraded", 1},
873 { "Deleted", 2},
874 { "Missing", 3},
875 { "Failed", 4},
876 { "Partially Optimal", 5},
877 { "-reserved-", 6},
878 { "-reserved-", 7},
879 { NULL, 0}
880 };
881
882 static mapping_t ddf_init_state[] = {
883 { "Not Initialised", 0},
884 { "QuickInit in Progress", 1},
885 { "Fully Initialised", 2},
886 { "*UNKNOWN*", 3},
887 { NULL, 0}
888 };
889 static mapping_t ddf_access[] = {
890 { "Read/Write", 0},
891 { "Reserved", 1},
892 { "Read Only", 2},
893 { "Blocked (no access)", 3},
894 { NULL ,0}
895 };
896
897 static mapping_t ddf_level[] = {
898 { "RAID0", DDF_RAID0},
899 { "RAID1", DDF_RAID1},
900 { "RAID3", DDF_RAID3},
901 { "RAID4", DDF_RAID4},
902 { "RAID5", DDF_RAID5},
903 { "RAID1E",DDF_RAID1E},
904 { "JBOD", DDF_JBOD},
905 { "CONCAT",DDF_CONCAT},
906 { "RAID5E",DDF_RAID5E},
907 { "RAID5EE",DDF_RAID5EE},
908 { "RAID6", DDF_RAID6},
909 { NULL, 0}
910 };
911 static mapping_t ddf_sec_level[] = {
912 { "Striped", DDF_2STRIPED},
913 { "Mirrored", DDF_2MIRRORED},
914 { "Concat", DDF_2CONCAT},
915 { "Spanned", DDF_2SPANNED},
916 { NULL, 0}
917 };
918 #endif
919
920 struct num_mapping {
921 int num1, num2;
922 };
923 static struct num_mapping ddf_level_num[] = {
924 { DDF_RAID0, 0 },
925 { DDF_RAID1, 1 },
926 { DDF_RAID3, LEVEL_UNSUPPORTED },
927 { DDF_RAID4, 4 },
928 { DDF_RAID5, 5 },
929 { DDF_RAID1E, LEVEL_UNSUPPORTED },
930 { DDF_JBOD, LEVEL_UNSUPPORTED },
931 { DDF_CONCAT, LEVEL_LINEAR },
932 { DDF_RAID5E, LEVEL_UNSUPPORTED },
933 { DDF_RAID5EE, LEVEL_UNSUPPORTED },
934 { DDF_RAID6, 6},
935 { MAXINT, MAXINT }
936 };
937
938 static int map_num1(struct num_mapping *map, int num)
939 {
940 int i;
941 for (i=0 ; map[i].num1 != MAXINT; i++)
942 if (map[i].num1 == num)
943 break;
944 return map[i].num2;
945 }
946
947 #ifndef MDASSEMBLE
948 static void print_guid(char *guid, int tstamp)
949 {
950 /* A GUIDs are part (or all) ASCII and part binary.
951 * They tend to be space padded.
952 * We print the GUID in HEX, then in parentheses add
953 * any initial ASCII sequence, and a possible
954 * time stamp from bytes 16-19
955 */
956 int l = DDF_GUID_LEN;
957 int i;
958
959 for (i=0 ; i<DDF_GUID_LEN ; i++) {
960 if ((i&3)==0 && i != 0) printf(":");
961 printf("%02X", guid[i]&255);
962 }
963
964 printf(" (");
965 while (l && guid[l-1] == ' ')
966 l--;
967 for (i=0 ; i<l ; i++) {
968 if (guid[i] >= 0x20 && guid[i] < 0x7f)
969 fputc(guid[i], stdout);
970 else
971 break;
972 }
973 if (tstamp) {
974 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
975 char tbuf[100];
976 struct tm *tm;
977 tm = localtime(&then);
978 strftime(tbuf, 100, " %D %T",tm);
979 fputs(tbuf, stdout);
980 }
981 printf(")");
982 }
983
984 static void examine_vd(int n, struct ddf_super *sb, char *guid)
985 {
986 int crl = sb->conf_rec_len;
987 struct vcl *vcl;
988
989 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
990 struct vd_config *vc = &vcl->conf;
991
992 if (calc_crc(vc, crl*512) != vc->crc)
993 continue;
994 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
995 continue;
996
997 /* Ok, we know about this VD, let's give more details */
998 printf(" Raid Devices[%d] : %d\n", n,
999 __be16_to_cpu(vc->prim_elmnt_count));
1000 printf(" Chunk Size[%d] : %d sectors\n", n,
1001 1 << vc->chunk_shift);
1002 printf(" Raid Level[%d] : %s\n", n,
1003 map_num(ddf_level, vc->prl)?:"-unknown-");
1004 if (vc->sec_elmnt_count != 1) {
1005 printf(" Secondary Position[%d] : %d of %d\n", n,
1006 vc->sec_elmnt_seq, vc->sec_elmnt_count);
1007 printf(" Secondary Level[%d] : %s\n", n,
1008 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1009 }
1010 printf(" Device Size[%d] : %llu\n", n,
1011 __be64_to_cpu(vc->blocks)/2);
1012 printf(" Array Size[%d] : %llu\n", n,
1013 __be64_to_cpu(vc->array_blocks)/2);
1014 }
1015 }
1016
1017 static void examine_vds(struct ddf_super *sb)
1018 {
1019 int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1020 int i;
1021 printf(" Virtual Disks : %d\n", cnt);
1022
1023 for (i=0; i<cnt; i++) {
1024 struct virtual_entry *ve = &sb->virt->entries[i];
1025 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1026 printf("\n");
1027 printf(" unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1028 printf(" state[%d] : %s, %s%s\n", i,
1029 map_num(ddf_state, ve->state & 7),
1030 (ve->state & 8) ? "Morphing, ": "",
1031 (ve->state & 16)? "Not Consistent" : "Consistent");
1032 printf(" init state[%d] : %s\n", i,
1033 map_num(ddf_init_state, ve->init_state&3));
1034 printf(" access[%d] : %s\n", i,
1035 map_num(ddf_access, (ve->init_state>>6) & 3));
1036 printf(" Name[%d] : %.16s\n", i, ve->name);
1037 examine_vd(i, sb, ve->guid);
1038 }
1039 if (cnt) printf("\n");
1040 }
1041
1042 static void examine_pds(struct ddf_super *sb)
1043 {
1044 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1045 int i;
1046 struct dl *dl;
1047 printf(" Physical Disks : %d\n", cnt);
1048
1049 for (i=0 ; i<cnt ; i++) {
1050 struct phys_disk_entry *pd = &sb->phys->entries[i];
1051 int type = __be16_to_cpu(pd->type);
1052 int state = __be16_to_cpu(pd->state);
1053
1054 printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1055 printf("\n");
1056 printf(" ref[%d] : %08x\n", i,
1057 __be32_to_cpu(pd->refnum));
1058 printf(" mode[%d] : %s%s%s%s%s\n", i,
1059 (type&2) ? "active":"",
1060 (type&4) ? "Global Spare":"",
1061 (type&8) ? "spare" : "",
1062 (type&16)? ", foreign" : "",
1063 (type&32)? "pass-through" : "");
1064 printf(" state[%d] : %s%s%s%s%s%s%s\n", i,
1065 (state&1)? "Online": "Offline",
1066 (state&2)? ", Failed": "",
1067 (state&4)? ", Rebuilding": "",
1068 (state&8)? ", in-transition": "",
1069 (state&16)? ", SMART errors": "",
1070 (state&32)? ", Unrecovered Read Errors": "",
1071 (state&64)? ", Missing" : "");
1072 printf(" Avail Size[%d] : %llu K\n", i,
1073 __be64_to_cpu(pd->config_size)>>1);
1074 for (dl = sb->dlist; dl ; dl = dl->next) {
1075 if (dl->disk.refnum == pd->refnum) {
1076 char *dv = map_dev(dl->major, dl->minor, 0);
1077 if (dv)
1078 printf(" Device[%d] : %s\n",
1079 i, dv);
1080 }
1081 }
1082 printf("\n");
1083 }
1084 }
1085
1086 static void examine_super_ddf(struct supertype *st, char *homehost)
1087 {
1088 struct ddf_super *sb = st->sb;
1089
1090 printf(" Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1091 printf(" Version : %.8s\n", sb->anchor.revision);
1092 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1093 printf("\n");
1094 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1095 printf("\n");
1096 printf(" Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1097 printf(" Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1098 ?"yes" : "no");
1099 examine_vds(sb);
1100 examine_pds(sb);
1101 }
1102
1103 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info);
1104
1105
1106 static void brief_examine_super_ddf(struct supertype *st)
1107 {
1108 /* We just write a generic DDF ARRAY entry
1109 */
1110 struct mdinfo info;
1111 char nbuf[64];
1112 getinfo_super_ddf(st, &info);
1113 fname_from_uuid(st, &info, nbuf, ':');
1114 printf("ARRAY /dev/ddf metadata=ddf UUID=%s\n", nbuf + 5);
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 struct mdinfo info;
1134 char nbuf[64];
1135 getinfo_super_ddf(st, &info);
1136 fname_from_uuid(st, &info, nbuf,':');
1137 printf(" UUID=%s", nbuf + 5);
1138 }
1139 #endif
1140
1141 static int match_home_ddf(struct supertype *st, char *homehost)
1142 {
1143 /* It matches 'this' host if the controller is a
1144 * Linux-MD controller with vendor_data matching
1145 * the hostname
1146 */
1147 struct ddf_super *ddf = st->sb;
1148 int len = strlen(homehost);
1149
1150 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1151 len < sizeof(ddf->controller.vendor_data) &&
1152 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1153 ddf->controller.vendor_data[len] == 0);
1154 }
1155
1156 #ifndef MDASSEMBLE
1157 static struct vd_config *find_vdcr(struct ddf_super *ddf, int inst)
1158 {
1159 struct vcl *v;
1160
1161 for (v = ddf->conflist; v; v = v->next)
1162 if (inst == v->vcnum)
1163 return &v->conf;
1164 return NULL;
1165 }
1166 #endif
1167
1168 static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1169 {
1170 /* Find the entry in phys_disk which has the given refnum
1171 * and return it's index
1172 */
1173 int i;
1174 for (i=0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1175 if (ddf->phys->entries[i].refnum == phys_refnum)
1176 return i;
1177 return -1;
1178 }
1179
1180 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1181 {
1182 /* The uuid returned here is used for:
1183 * uuid to put into bitmap file (Create, Grow)
1184 * uuid for backup header when saving critical section (Grow)
1185 * comparing uuids when re-adding a device into an array
1186 * In these cases the uuid required is that of the data-array,
1187 * not the device-set.
1188 * uuid to recognise same set when adding a missing device back
1189 * to an array. This is a uuid for the device-set.
1190 *
1191 * For each of these we can make do with a truncated
1192 * or hashed uuid rather than the original, as long as
1193 * everyone agrees.
1194 * In the case of SVD we assume the BVD is of interest,
1195 * though that might be the case if a bitmap were made for
1196 * a mirrored SVD - worry about that later.
1197 * So we need to find the VD configuration record for the
1198 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1199 * The first 16 bytes of the sha1 of these is used.
1200 */
1201 struct ddf_super *ddf = st->sb;
1202 struct vcl *vcl = ddf->currentconf;
1203 char *guid;
1204 char buf[20];
1205 struct sha1_ctx ctx;
1206
1207 if (vcl)
1208 guid = vcl->conf.guid;
1209 else
1210 guid = ddf->anchor.guid;
1211
1212 sha1_init_ctx(&ctx);
1213 sha1_process_bytes(guid, DDF_GUID_LEN, &ctx);
1214 if (vcl && vcl->conf.sec_elmnt_count > 1)
1215 sha1_process_bytes(&vcl->conf.sec_elmnt_seq, 1, &ctx);
1216 sha1_finish_ctx(&ctx, buf);
1217 memcpy(uuid, buf, 4*4);
1218 }
1219
1220 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info);
1221
1222 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info)
1223 {
1224 struct ddf_super *ddf = st->sb;
1225
1226 if (ddf->currentconf) {
1227 getinfo_super_ddf_bvd(st, info);
1228 return;
1229 }
1230
1231 info->array.raid_disks = __be16_to_cpu(ddf->phys->used_pdes);
1232 info->array.level = LEVEL_CONTAINER;
1233 info->array.layout = 0;
1234 info->array.md_minor = -1;
1235 info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)
1236 (ddf->anchor.guid+16));
1237 info->array.utime = 0;
1238 info->array.chunk_size = 0;
1239
1240
1241 info->disk.major = 0;
1242 info->disk.minor = 0;
1243 if (ddf->dlist) {
1244 info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1245 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1246
1247 info->data_offset = __be64_to_cpu(ddf->phys->
1248 entries[info->disk.raid_disk].
1249 config_size);
1250 info->component_size = ddf->dlist->size - info->data_offset;
1251 } else {
1252 info->disk.number = -1;
1253 // info->disk.raid_disk = find refnum in the table and use index;
1254 }
1255 info->disk.state = (1 << MD_DISK_SYNC);
1256
1257
1258 info->reshape_active = 0;
1259 info->name[0] = 0;
1260
1261 info->array.major_version = -1;
1262 info->array.minor_version = -2;
1263 strcpy(info->text_version, "ddf");
1264 info->safe_mode_delay = 0;
1265
1266 uuid_from_super_ddf(st, info->uuid);
1267
1268 }
1269
1270 static int rlq_to_layout(int rlq, int prl, int raiddisks);
1271
1272 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info)
1273 {
1274 struct ddf_super *ddf = st->sb;
1275 struct vcl *vc = ddf->currentconf;
1276 int cd = ddf->currentdev;
1277
1278 /* FIXME this returns BVD info - what if we want SVD ?? */
1279
1280 info->array.raid_disks = __be16_to_cpu(vc->conf.prim_elmnt_count);
1281 info->array.level = map_num1(ddf_level_num, vc->conf.prl);
1282 info->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
1283 info->array.raid_disks);
1284 info->array.md_minor = -1;
1285 info->array.ctime = DECADE +
1286 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
1287 info->array.utime = DECADE + __be32_to_cpu(vc->conf.timestamp);
1288 info->array.chunk_size = 512 << vc->conf.chunk_shift;
1289
1290 if (cd >= 0 && cd < ddf->mppe) {
1291 info->data_offset = __be64_to_cpu(vc->lba_offset[cd]);
1292 if (vc->block_sizes)
1293 info->component_size = vc->block_sizes[cd];
1294 else
1295 info->component_size = __be64_to_cpu(vc->conf.blocks);
1296 }
1297
1298 info->disk.major = 0;
1299 info->disk.minor = 0;
1300 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1301 // info->disk.raid_disk = find refnum in the table and use index;
1302 // info->disk.state = ???;
1303
1304 info->container_member = ddf->currentconf->vcnum;
1305
1306 info->resync_start = 0;
1307 if (!(ddf->virt->entries[info->container_member].state
1308 & DDF_state_inconsistent) &&
1309 (ddf->virt->entries[info->container_member].init_state
1310 & DDF_initstate_mask)
1311 == DDF_init_full)
1312 info->resync_start = ~0ULL;
1313
1314 uuid_from_super_ddf(st, info->uuid);
1315
1316 info->container_member = atoi(st->subarray);
1317 info->array.major_version = -1;
1318 info->array.minor_version = -2;
1319 sprintf(info->text_version, "/%s/%s",
1320 devnum2devname(st->container_dev),
1321 st->subarray);
1322 info->safe_mode_delay = 200;
1323
1324 info->name[0] = 0;
1325 }
1326
1327
1328 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1329 char *update,
1330 char *devname, int verbose,
1331 int uuid_set, char *homehost)
1332 {
1333 /* For 'assemble' and 'force' we need to return non-zero if any
1334 * change was made. For others, the return value is ignored.
1335 * Update options are:
1336 * force-one : This device looks a bit old but needs to be included,
1337 * update age info appropriately.
1338 * assemble: clear any 'faulty' flag to allow this device to
1339 * be assembled.
1340 * force-array: Array is degraded but being forced, mark it clean
1341 * if that will be needed to assemble it.
1342 *
1343 * newdev: not used ????
1344 * grow: Array has gained a new device - this is currently for
1345 * linear only
1346 * resync: mark as dirty so a resync will happen.
1347 * uuid: Change the uuid of the array to match what is given
1348 * homehost: update the recorded homehost
1349 * name: update the name - preserving the homehost
1350 * _reshape_progress: record new reshape_progress position.
1351 *
1352 * Following are not relevant for this version:
1353 * sparc2.2 : update from old dodgey metadata
1354 * super-minor: change the preferred_minor number
1355 * summaries: update redundant counters.
1356 */
1357 int rv = 0;
1358 // struct ddf_super *ddf = st->sb;
1359 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1360 // struct virtual_entry *ve = find_ve(ddf);
1361
1362 /* we don't need to handle "force-*" or "assemble" as
1363 * there is no need to 'trick' the kernel. We the metadata is
1364 * first updated to activate the array, all the implied modifications
1365 * will just happen.
1366 */
1367
1368 if (strcmp(update, "grow") == 0) {
1369 /* FIXME */
1370 }
1371 if (strcmp(update, "resync") == 0) {
1372 // info->resync_checkpoint = 0;
1373 }
1374 /* We ignore UUID updates as they make even less sense
1375 * with DDF
1376 */
1377 if (strcmp(update, "homehost") == 0) {
1378 /* homehost is stored in controller->vendor_data,
1379 * or it is when we are the vendor
1380 */
1381 // if (info->vendor_is_local)
1382 // strcpy(ddf->controller.vendor_data, homehost);
1383 }
1384 if (strcmp(update, "name") == 0) {
1385 /* name is stored in virtual_entry->name */
1386 // memset(ve->name, ' ', 16);
1387 // strncpy(ve->name, info->name, 16);
1388 }
1389 if (strcmp(update, "_reshape_progress") == 0) {
1390 /* We don't support reshape yet */
1391 }
1392
1393 // update_all_csum(ddf);
1394
1395 return rv;
1396 }
1397
1398 static void make_header_guid(char *guid)
1399 {
1400 __u32 stamp;
1401 int rfd;
1402 /* Create a DDF Header of Virtual Disk GUID */
1403
1404 /* 24 bytes of fiction required.
1405 * first 8 are a 'vendor-id' - "Linux-MD"
1406 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1407 * Remaining 8 random number plus timestamp
1408 */
1409 memcpy(guid, T10, sizeof(T10));
1410 stamp = __cpu_to_be32(0xdeadbeef);
1411 memcpy(guid+8, &stamp, 4);
1412 stamp = __cpu_to_be32(0);
1413 memcpy(guid+12, &stamp, 4);
1414 stamp = __cpu_to_be32(time(0) - DECADE);
1415 memcpy(guid+16, &stamp, 4);
1416 rfd = open("/dev/urandom", O_RDONLY);
1417 if (rfd < 0 || read(rfd, &stamp, 4) != 4)
1418 stamp = random();
1419 memcpy(guid+20, &stamp, 4);
1420 if (rfd >= 0) close(rfd);
1421 }
1422
1423 static int init_super_ddf_bvd(struct supertype *st,
1424 mdu_array_info_t *info,
1425 unsigned long long size,
1426 char *name, char *homehost,
1427 int *uuid);
1428
1429 static int init_super_ddf(struct supertype *st,
1430 mdu_array_info_t *info,
1431 unsigned long long size, char *name, char *homehost,
1432 int *uuid)
1433 {
1434 /* This is primarily called by Create when creating a new array.
1435 * We will then get add_to_super called for each component, and then
1436 * write_init_super called to write it out to each device.
1437 * For DDF, Create can create on fresh devices or on a pre-existing
1438 * array.
1439 * To create on a pre-existing array a different method will be called.
1440 * This one is just for fresh drives.
1441 *
1442 * We need to create the entire 'ddf' structure which includes:
1443 * DDF headers - these are easy.
1444 * Controller data - a Sector describing this controller .. not that
1445 * this is a controller exactly.
1446 * Physical Disk Record - one entry per device, so
1447 * leave plenty of space.
1448 * Virtual Disk Records - again, just leave plenty of space.
1449 * This just lists VDs, doesn't give details
1450 * Config records - describes the VDs that use this disk
1451 * DiskData - describes 'this' device.
1452 * BadBlockManagement - empty
1453 * Diag Space - empty
1454 * Vendor Logs - Could we put bitmaps here?
1455 *
1456 */
1457 struct ddf_super *ddf;
1458 char hostname[17];
1459 int hostlen;
1460 int max_phys_disks, max_virt_disks;
1461 unsigned long long sector;
1462 int clen;
1463 int i;
1464 int pdsize, vdsize;
1465 struct phys_disk *pd;
1466 struct virtual_disk *vd;
1467
1468 if (!info) {
1469 st->sb = NULL;
1470 return 0;
1471 }
1472 if (st->sb)
1473 return init_super_ddf_bvd(st, info, size, name, homehost,
1474 uuid);
1475
1476 posix_memalign((void**)&ddf, 512, sizeof(*ddf));
1477 memset(ddf, 0, sizeof(*ddf));
1478 ddf->dlist = NULL; /* no physical disks yet */
1479 ddf->conflist = NULL; /* No virtual disks yet */
1480
1481 /* At least 32MB *must* be reserved for the ddf. So let's just
1482 * start 32MB from the end, and put the primary header there.
1483 * Don't do secondary for now.
1484 * We don't know exactly where that will be yet as it could be
1485 * different on each device. To just set up the lengths.
1486 *
1487 */
1488
1489 ddf->anchor.magic = DDF_HEADER_MAGIC;
1490 make_header_guid(ddf->anchor.guid);
1491
1492 memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
1493 ddf->anchor.seq = __cpu_to_be32(1);
1494 ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1495 ddf->anchor.openflag = 0xFF;
1496 ddf->anchor.foreignflag = 0;
1497 ddf->anchor.enforcegroups = 0; /* Is this best?? */
1498 ddf->anchor.pad0 = 0xff;
1499 memset(ddf->anchor.pad1, 0xff, 12);
1500 memset(ddf->anchor.header_ext, 0xff, 32);
1501 ddf->anchor.primary_lba = ~(__u64)0;
1502 ddf->anchor.secondary_lba = ~(__u64)0;
1503 ddf->anchor.type = DDF_HEADER_ANCHOR;
1504 memset(ddf->anchor.pad2, 0xff, 3);
1505 ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1506 ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1507 of 32M reserved.. */
1508 max_phys_disks = 1023; /* Should be enough */
1509 ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1510 max_virt_disks = 255;
1511 ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1512 ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1513 ddf->max_part = 64;
1514 ddf->mppe = 256;
1515 ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
1516 ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1517 ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe);
1518 memset(ddf->anchor.pad3, 0xff, 54);
1519 /* controller sections is one sector long immediately
1520 * after the ddf header */
1521 sector = 1;
1522 ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1523 ddf->anchor.controller_section_length = __cpu_to_be32(1);
1524 sector += 1;
1525
1526 /* phys is 8 sectors after that */
1527 pdsize = ROUND_UP(sizeof(struct phys_disk) +
1528 sizeof(struct phys_disk_entry)*max_phys_disks,
1529 512);
1530 switch(pdsize/512) {
1531 case 2: case 8: case 32: case 128: case 512: break;
1532 default: abort();
1533 }
1534 ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1535 ddf->anchor.phys_section_length =
1536 __cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1537 sector += pdsize/512;
1538
1539 /* virt is another 32 sectors */
1540 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1541 sizeof(struct virtual_entry) * max_virt_disks,
1542 512);
1543 switch(vdsize/512) {
1544 case 2: case 8: case 32: case 128: case 512: break;
1545 default: abort();
1546 }
1547 ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1548 ddf->anchor.virt_section_length =
1549 __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1550 sector += vdsize/512;
1551
1552 clen = ddf->conf_rec_len * (ddf->max_part+1);
1553 ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1554 ddf->anchor.config_section_length = __cpu_to_be32(clen);
1555 sector += clen;
1556
1557 ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1558 ddf->anchor.data_section_length = __cpu_to_be32(1);
1559 sector += 1;
1560
1561 ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1562 ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1563 ddf->anchor.diag_space_length = __cpu_to_be32(0);
1564 ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1565 ddf->anchor.vendor_length = __cpu_to_be32(0);
1566 ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1567
1568 memset(ddf->anchor.pad4, 0xff, 256);
1569
1570 memcpy(&ddf->primary, &ddf->anchor, 512);
1571 memcpy(&ddf->secondary, &ddf->anchor, 512);
1572
1573 ddf->primary.openflag = 1; /* I guess.. */
1574 ddf->primary.type = DDF_HEADER_PRIMARY;
1575
1576 ddf->secondary.openflag = 1; /* I guess.. */
1577 ddf->secondary.type = DDF_HEADER_SECONDARY;
1578
1579 ddf->active = &ddf->primary;
1580
1581 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1582
1583 /* 24 more bytes of fiction required.
1584 * first 8 are a 'vendor-id' - "Linux-MD"
1585 * Remaining 16 are serial number.... maybe a hostname would do?
1586 */
1587 memcpy(ddf->controller.guid, T10, sizeof(T10));
1588 gethostname(hostname, sizeof(hostname));
1589 hostname[sizeof(hostname) - 1] = 0;
1590 hostlen = strlen(hostname);
1591 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1592 for (i = strlen(T10) ; i+hostlen < 24; i++)
1593 ddf->controller.guid[i] = ' ';
1594
1595 ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1596 ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1597 ddf->controller.type.sub_vendor_id = 0;
1598 ddf->controller.type.sub_device_id = 0;
1599 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1600 memset(ddf->controller.pad, 0xff, 8);
1601 memset(ddf->controller.vendor_data, 0xff, 448);
1602
1603 posix_memalign((void**)&pd, 512, pdsize);
1604 ddf->phys = pd;
1605 ddf->pdsize = pdsize;
1606
1607 memset(pd, 0xff, pdsize);
1608 memset(pd, 0, sizeof(*pd));
1609 pd->magic = DDF_PHYS_DATA_MAGIC;
1610 pd->used_pdes = __cpu_to_be16(0);
1611 pd->max_pdes = __cpu_to_be16(max_phys_disks);
1612 memset(pd->pad, 0xff, 52);
1613
1614 posix_memalign((void**)&vd, 512, vdsize);
1615 ddf->virt = vd;
1616 ddf->vdsize = vdsize;
1617 memset(vd, 0, vdsize);
1618 vd->magic = DDF_VIRT_RECORDS_MAGIC;
1619 vd->populated_vdes = __cpu_to_be16(0);
1620 vd->max_vdes = __cpu_to_be16(max_virt_disks);
1621 memset(vd->pad, 0xff, 52);
1622
1623 for (i=0; i<max_virt_disks; i++)
1624 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1625
1626 st->sb = ddf;
1627 ddf->updates_pending = 1;
1628 return 1;
1629 }
1630
1631 static int all_ff(char *guid)
1632 {
1633 int i;
1634 for (i = 0; i < DDF_GUID_LEN; i++)
1635 if (guid[i] != (char)0xff)
1636 return 0;
1637 return 1;
1638 }
1639 static int chunk_to_shift(int chunksize)
1640 {
1641 return ffs(chunksize/512)-1;
1642 }
1643
1644 static int level_to_prl(int level)
1645 {
1646 switch (level) {
1647 case LEVEL_LINEAR: return DDF_CONCAT;
1648 case 0: return DDF_RAID0;
1649 case 1: return DDF_RAID1;
1650 case 4: return DDF_RAID4;
1651 case 5: return DDF_RAID5;
1652 case 6: return DDF_RAID6;
1653 default: return -1;
1654 }
1655 }
1656 static int layout_to_rlq(int level, int layout, int raiddisks)
1657 {
1658 switch(level) {
1659 case 0:
1660 return DDF_RAID0_SIMPLE;
1661 case 1:
1662 switch(raiddisks) {
1663 case 2: return DDF_RAID1_SIMPLE;
1664 case 3: return DDF_RAID1_MULTI;
1665 default: return -1;
1666 }
1667 case 4:
1668 switch(layout) {
1669 case 0: return DDF_RAID4_N;
1670 }
1671 break;
1672 case 5:
1673 case 6:
1674 switch(layout) {
1675 case ALGORITHM_LEFT_ASYMMETRIC:
1676 return DDF_RAID5_N_RESTART;
1677 case ALGORITHM_RIGHT_ASYMMETRIC:
1678 if (level == 5)
1679 return DDF_RAID5_0_RESTART;
1680 else
1681 return DDF_RAID6_0_RESTART;
1682 case ALGORITHM_LEFT_SYMMETRIC:
1683 return DDF_RAID5_N_CONTINUE;
1684 case ALGORITHM_RIGHT_SYMMETRIC:
1685 return -1; /* not mentioned in standard */
1686 }
1687 }
1688 return -1;
1689 }
1690
1691 static int rlq_to_layout(int rlq, int prl, int raiddisks)
1692 {
1693 switch(prl) {
1694 case DDF_RAID0:
1695 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1696 case DDF_RAID1:
1697 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1698 on raiddisks*/
1699 case DDF_RAID4:
1700 switch(rlq) {
1701 case DDF_RAID4_N:
1702 return 0;
1703 default:
1704 /* not supported */
1705 return -1; /* FIXME this isn't checked */
1706 }
1707 case DDF_RAID5:
1708 switch(rlq) {
1709 case DDF_RAID5_N_RESTART:
1710 return ALGORITHM_LEFT_ASYMMETRIC;
1711 case DDF_RAID5_0_RESTART:
1712 return ALGORITHM_RIGHT_ASYMMETRIC;
1713 case DDF_RAID5_N_CONTINUE:
1714 return ALGORITHM_LEFT_SYMMETRIC;
1715 default:
1716 return -1;
1717 }
1718 case DDF_RAID6:
1719 switch(rlq) {
1720 case DDF_RAID5_N_RESTART:
1721 return ALGORITHM_LEFT_ASYMMETRIC;
1722 case DDF_RAID6_0_RESTART:
1723 return ALGORITHM_RIGHT_ASYMMETRIC;
1724 case DDF_RAID5_N_CONTINUE:
1725 return ALGORITHM_LEFT_SYMMETRIC;
1726 default:
1727 return -1;
1728 }
1729 }
1730 return -1;
1731 }
1732
1733 #ifndef MDASSEMBLE
1734 struct extent {
1735 unsigned long long start, size;
1736 };
1737 static int cmp_extent(const void *av, const void *bv)
1738 {
1739 const struct extent *a = av;
1740 const struct extent *b = bv;
1741 if (a->start < b->start)
1742 return -1;
1743 if (a->start > b->start)
1744 return 1;
1745 return 0;
1746 }
1747
1748 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
1749 {
1750 /* find a list of used extents on the give physical device
1751 * (dnum) of the given ddf.
1752 * Return a malloced array of 'struct extent'
1753
1754 FIXME ignore DDF_Legacy devices?
1755
1756 */
1757 struct extent *rv;
1758 int n = 0;
1759 int i, j;
1760
1761 rv = malloc(sizeof(struct extent) * (ddf->max_part + 2));
1762 if (!rv)
1763 return NULL;
1764
1765 for (i = 0; i < ddf->max_part; i++) {
1766 struct vcl *v = dl->vlist[i];
1767 if (v == NULL)
1768 continue;
1769 for (j=0; j < v->conf.prim_elmnt_count; j++)
1770 if (v->conf.phys_refnum[j] == dl->disk.refnum) {
1771 /* This device plays role 'j' in 'v'. */
1772 rv[n].start = __be64_to_cpu(v->lba_offset[j]);
1773 rv[n].size = __be64_to_cpu(v->conf.blocks);
1774 n++;
1775 break;
1776 }
1777 }
1778 qsort(rv, n, sizeof(*rv), cmp_extent);
1779
1780 rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
1781 rv[n].size = 0;
1782 return rv;
1783 }
1784 #endif
1785
1786 static int init_super_ddf_bvd(struct supertype *st,
1787 mdu_array_info_t *info,
1788 unsigned long long size,
1789 char *name, char *homehost,
1790 int *uuid)
1791 {
1792 /* We are creating a BVD inside a pre-existing container.
1793 * so st->sb is already set.
1794 * We need to create a new vd_config and a new virtual_entry
1795 */
1796 struct ddf_super *ddf = st->sb;
1797 int venum;
1798 struct virtual_entry *ve;
1799 struct vcl *vcl;
1800 struct vd_config *vc;
1801
1802 if (__be16_to_cpu(ddf->virt->populated_vdes)
1803 >= __be16_to_cpu(ddf->virt->max_vdes)) {
1804 fprintf(stderr, Name": This ddf already has the "
1805 "maximum of %d virtual devices\n",
1806 __be16_to_cpu(ddf->virt->max_vdes));
1807 return 0;
1808 }
1809
1810 for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1811 if (all_ff(ddf->virt->entries[venum].guid))
1812 break;
1813 if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1814 fprintf(stderr, Name ": Cannot find spare slot for "
1815 "virtual disk - DDF is corrupt\n");
1816 return 0;
1817 }
1818 ve = &ddf->virt->entries[venum];
1819
1820 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1821 * timestamp, random number
1822 */
1823 make_header_guid(ve->guid);
1824 ve->unit = __cpu_to_be16(info->md_minor);
1825 ve->pad0 = 0xFFFF;
1826 ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1827 ve->type = 0;
1828 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
1829 if (info->state & 1) /* clean */
1830 ve->init_state = DDF_init_full;
1831 else
1832 ve->init_state = DDF_init_not;
1833
1834 memset(ve->pad1, 0xff, 14);
1835 memset(ve->name, ' ', 16);
1836 if (name)
1837 strncpy(ve->name, name, 16);
1838 ddf->virt->populated_vdes =
1839 __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
1840
1841 /* Now create a new vd_config */
1842 posix_memalign((void**)&vcl, 512,
1843 (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512));
1844 vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
1845 vcl->vcnum = venum;
1846 sprintf(st->subarray, "%d", venum);
1847 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1848
1849 vc = &vcl->conf;
1850
1851 vc->magic = DDF_VD_CONF_MAGIC;
1852 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
1853 vc->timestamp = __cpu_to_be32(time(0)-DECADE);
1854 vc->seqnum = __cpu_to_be32(1);
1855 memset(vc->pad0, 0xff, 24);
1856 vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
1857 vc->chunk_shift = chunk_to_shift(info->chunk_size);
1858 vc->prl = level_to_prl(info->level);
1859 vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
1860 vc->sec_elmnt_count = 1;
1861 vc->sec_elmnt_seq = 0;
1862 vc->srl = 0;
1863 vc->blocks = __cpu_to_be64(info->size * 2);
1864 vc->array_blocks = __cpu_to_be64(
1865 calc_array_size(info->level, info->raid_disks, info->layout,
1866 info->chunk_size, info->size*2));
1867 memset(vc->pad1, 0xff, 8);
1868 vc->spare_refs[0] = 0xffffffff;
1869 vc->spare_refs[1] = 0xffffffff;
1870 vc->spare_refs[2] = 0xffffffff;
1871 vc->spare_refs[3] = 0xffffffff;
1872 vc->spare_refs[4] = 0xffffffff;
1873 vc->spare_refs[5] = 0xffffffff;
1874 vc->spare_refs[6] = 0xffffffff;
1875 vc->spare_refs[7] = 0xffffffff;
1876 memset(vc->cache_pol, 0, 8);
1877 vc->bg_rate = 0x80;
1878 memset(vc->pad2, 0xff, 3);
1879 memset(vc->pad3, 0xff, 52);
1880 memset(vc->pad4, 0xff, 192);
1881 memset(vc->v0, 0xff, 32);
1882 memset(vc->v1, 0xff, 32);
1883 memset(vc->v2, 0xff, 16);
1884 memset(vc->v3, 0xff, 16);
1885 memset(vc->vendor, 0xff, 32);
1886
1887 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
1888 memset(vc->phys_refnum+(ddf->mppe * 4), 0x00, 8*ddf->mppe);
1889
1890 vcl->next = ddf->conflist;
1891 ddf->conflist = vcl;
1892 ddf->currentconf = vcl;
1893 ddf->updates_pending = 1;
1894 return 1;
1895 }
1896
1897 #ifndef MDASSEMBLE
1898 static void add_to_super_ddf_bvd(struct supertype *st,
1899 mdu_disk_info_t *dk, int fd, char *devname)
1900 {
1901 /* fd and devname identify a device with-in the ddf container (st).
1902 * dk identifies a location in the new BVD.
1903 * We need to find suitable free space in that device and update
1904 * the phys_refnum and lba_offset for the newly created vd_config.
1905 * We might also want to update the type in the phys_disk
1906 * section.
1907 */
1908 struct dl *dl;
1909 struct ddf_super *ddf = st->sb;
1910 struct vd_config *vc;
1911 __u64 *lba_offset;
1912 int working;
1913 int i;
1914 unsigned long long blocks, pos, esize;
1915 struct extent *ex;
1916
1917 for (dl = ddf->dlist; dl ; dl = dl->next)
1918 if (dl->major == dk->major &&
1919 dl->minor == dk->minor)
1920 break;
1921 if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
1922 return;
1923
1924 vc = &ddf->currentconf->conf;
1925 lba_offset = ddf->currentconf->lba_offset;
1926
1927 ex = get_extents(ddf, dl);
1928 if (!ex)
1929 return;
1930
1931 i = 0; pos = 0;
1932 blocks = __be64_to_cpu(vc->blocks);
1933 if (ddf->currentconf->block_sizes)
1934 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
1935
1936 do {
1937 esize = ex[i].start - pos;
1938 if (esize >= blocks)
1939 break;
1940 pos = ex[i].start + ex[i].size;
1941 i++;
1942 } while (ex[i-1].size);
1943
1944 free(ex);
1945 if (esize < blocks)
1946 return;
1947
1948 ddf->currentdev = dk->raid_disk;
1949 vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
1950 lba_offset[dk->raid_disk] = __cpu_to_be64(pos);
1951
1952 for (i=0; i < ddf->max_part ; i++)
1953 if (dl->vlist[i] == NULL)
1954 break;
1955 if (i == ddf->max_part)
1956 return;
1957 dl->vlist[i] = ddf->currentconf;
1958
1959 dl->fd = fd;
1960 dl->devname = devname;
1961
1962 /* Check how many working raid_disks, and if we can mark
1963 * array as optimal yet
1964 */
1965 working = 0;
1966
1967 for (i=0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
1968 if (vc->phys_refnum[i] != 0xffffffff)
1969 working++;
1970
1971 /* Find which virtual_entry */
1972 i = ddf->currentconf->vcnum;
1973 if (working == __be16_to_cpu(vc->prim_elmnt_count))
1974 ddf->virt->entries[i].state =
1975 (ddf->virt->entries[i].state & ~DDF_state_mask)
1976 | DDF_state_optimal;
1977
1978 if (vc->prl == DDF_RAID6 &&
1979 working+1 == __be16_to_cpu(vc->prim_elmnt_count))
1980 ddf->virt->entries[i].state =
1981 (ddf->virt->entries[i].state & ~DDF_state_mask)
1982 | DDF_state_part_optimal;
1983
1984 ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
1985 ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
1986 ddf->updates_pending = 1;
1987 }
1988
1989 /* add a device to a container, either while creating it or while
1990 * expanding a pre-existing container
1991 */
1992 static void add_to_super_ddf(struct supertype *st,
1993 mdu_disk_info_t *dk, int fd, char *devname)
1994 {
1995 struct ddf_super *ddf = st->sb;
1996 struct dl *dd;
1997 time_t now;
1998 struct tm *tm;
1999 unsigned long long size;
2000 struct phys_disk_entry *pde;
2001 int n, i;
2002 struct stat stb;
2003
2004 if (ddf->currentconf) {
2005 add_to_super_ddf_bvd(st, dk, fd, devname);
2006 return;
2007 }
2008
2009 /* This is device numbered dk->number. We need to create
2010 * a phys_disk entry and a more detailed disk_data entry.
2011 */
2012 fstat(fd, &stb);
2013 posix_memalign((void**)&dd, 512,
2014 sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part);
2015 dd->major = major(stb.st_rdev);
2016 dd->minor = minor(stb.st_rdev);
2017 dd->devname = devname;
2018 dd->fd = fd;
2019 dd->spare = NULL;
2020
2021 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
2022 now = time(0);
2023 tm = localtime(&now);
2024 sprintf(dd->disk.guid, "%8s%04d%02d%02d",
2025 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
2026 *(__u32*)(dd->disk.guid + 16) = random();
2027 *(__u32*)(dd->disk.guid + 20) = random();
2028
2029 do {
2030 /* Cannot be bothered finding a CRC of some irrelevant details*/
2031 dd->disk.refnum = random();
2032 for (i = __be16_to_cpu(ddf->active->max_pd_entries) - 1;
2033 i >= 0; i--)
2034 if (ddf->phys->entries[i].refnum == dd->disk.refnum)
2035 break;
2036 } while (i >= 0);
2037
2038 dd->disk.forced_ref = 1;
2039 dd->disk.forced_guid = 1;
2040 memset(dd->disk.vendor, ' ', 32);
2041 memcpy(dd->disk.vendor, "Linux", 5);
2042 memset(dd->disk.pad, 0xff, 442);
2043 for (i = 0; i < ddf->max_part ; i++)
2044 dd->vlist[i] = NULL;
2045
2046 n = __be16_to_cpu(ddf->phys->used_pdes);
2047 pde = &ddf->phys->entries[n];
2048 dd->pdnum = n;
2049
2050 if (st->update_tail) {
2051 int len = (sizeof(struct phys_disk) +
2052 sizeof(struct phys_disk_entry));
2053 struct phys_disk *pd;
2054
2055 pd = malloc(len);
2056 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2057 pd->used_pdes = __cpu_to_be16(n);
2058 pde = &pd->entries[0];
2059 dd->mdupdate = pd;
2060 } else {
2061 n++;
2062 ddf->phys->used_pdes = __cpu_to_be16(n);
2063 }
2064
2065 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2066 pde->refnum = dd->disk.refnum;
2067 pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2068 pde->state = __cpu_to_be16(DDF_Online);
2069 get_dev_size(fd, NULL, &size);
2070 /* We are required to reserve 32Meg, and record the size in sectors */
2071 pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
2072 sprintf(pde->path, "%17.17s","Information: nil") ;
2073 memset(pde->pad, 0xff, 6);
2074
2075 dd->size = size >> 9;
2076 if (st->update_tail) {
2077 dd->next = ddf->add_list;
2078 ddf->add_list = dd;
2079 } else {
2080 dd->next = ddf->dlist;
2081 ddf->dlist = dd;
2082 ddf->updates_pending = 1;
2083 }
2084 }
2085
2086 /*
2087 * This is the write_init_super method for a ddf container. It is
2088 * called when creating a container or adding another device to a
2089 * container.
2090 */
2091
2092 static unsigned char null_conf[4096+512];
2093
2094 static int __write_init_super_ddf(struct supertype *st, int do_close)
2095 {
2096
2097 struct ddf_super *ddf = st->sb;
2098 int i;
2099 struct dl *d;
2100 int n_config;
2101 int conf_size;
2102
2103 unsigned long long size, sector;
2104
2105 for (d = ddf->dlist; d; d=d->next) {
2106 int fd = d->fd;
2107
2108 if (fd < 0)
2109 continue;
2110
2111 /* We need to fill in the primary, (secondary) and workspace
2112 * lba's in the headers, set their checksums,
2113 * Also checksum phys, virt....
2114 *
2115 * Then write everything out, finally the anchor is written.
2116 */
2117 get_dev_size(fd, NULL, &size);
2118 size /= 512;
2119 ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
2120 ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
2121 ddf->anchor.seq = __cpu_to_be32(1);
2122 memcpy(&ddf->primary, &ddf->anchor, 512);
2123 memcpy(&ddf->secondary, &ddf->anchor, 512);
2124
2125 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
2126 ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
2127 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
2128
2129 ddf->primary.openflag = 0;
2130 ddf->primary.type = DDF_HEADER_PRIMARY;
2131
2132 ddf->secondary.openflag = 0;
2133 ddf->secondary.type = DDF_HEADER_SECONDARY;
2134
2135 ddf->primary.crc = calc_crc(&ddf->primary, 512);
2136 ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
2137
2138 sector = size - 16*1024*2;
2139 lseek64(fd, sector<<9, 0);
2140 write(fd, &ddf->primary, 512);
2141
2142 ddf->controller.crc = calc_crc(&ddf->controller, 512);
2143 write(fd, &ddf->controller, 512);
2144
2145 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
2146
2147 write(fd, ddf->phys, ddf->pdsize);
2148
2149 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
2150 write(fd, ddf->virt, ddf->vdsize);
2151
2152 /* Now write lots of config records. */
2153 n_config = ddf->max_part;
2154 conf_size = ddf->conf_rec_len * 512;
2155 for (i = 0 ; i <= n_config ; i++) {
2156 struct vcl *c = d->vlist[i];
2157 if (i == n_config)
2158 c = (struct vcl*)d->spare;
2159
2160 if (c) {
2161 c->conf.crc = calc_crc(&c->conf, conf_size);
2162 write(fd, &c->conf, conf_size);
2163 } else {
2164 char *null_aligned = (char*)((((unsigned long)null_conf)+511)&~511UL);
2165 if (null_conf[0] != 0xff)
2166 memset(null_conf, 0xff, sizeof(null_conf));
2167 int togo = conf_size;
2168 while (togo > sizeof(null_conf)-512) {
2169 write(fd, null_aligned, sizeof(null_conf)-512);
2170 togo -= sizeof(null_conf)-512;
2171 }
2172 write(fd, null_aligned, togo);
2173 }
2174 }
2175 d->disk.crc = calc_crc(&d->disk, 512);
2176 write(fd, &d->disk, 512);
2177
2178 /* Maybe do the same for secondary */
2179
2180 lseek64(fd, (size-1)*512, SEEK_SET);
2181 write(fd, &ddf->anchor, 512);
2182 if (do_close) {
2183 close(fd);
2184 d->fd = -1;
2185 }
2186 }
2187 return 1;
2188 }
2189
2190 static int write_init_super_ddf(struct supertype *st)
2191 {
2192
2193 if (st->update_tail) {
2194 /* queue the virtual_disk and vd_config as metadata updates */
2195 struct virtual_disk *vd;
2196 struct vd_config *vc;
2197 struct ddf_super *ddf = st->sb;
2198 int len;
2199
2200 if (!ddf->currentconf) {
2201 int len = (sizeof(struct phys_disk) +
2202 sizeof(struct phys_disk_entry));
2203
2204 /* adding a disk to the container. */
2205 if (!ddf->add_list)
2206 return 0;
2207
2208 append_metadata_update(st, ddf->add_list->mdupdate, len);
2209 ddf->add_list->mdupdate = NULL;
2210 return 0;
2211 }
2212
2213 /* Newly created VD */
2214
2215 /* First the virtual disk. We have a slightly fake header */
2216 len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry);
2217 vd = malloc(len);
2218 *vd = *ddf->virt;
2219 vd->entries[0] = ddf->virt->entries[ddf->currentconf->vcnum];
2220 vd->populated_vdes = __cpu_to_be16(ddf->currentconf->vcnum);
2221 append_metadata_update(st, vd, len);
2222
2223 /* Then the vd_config */
2224 len = ddf->conf_rec_len * 512;
2225 vc = malloc(len);
2226 memcpy(vc, &ddf->currentconf->conf, len);
2227 append_metadata_update(st, vc, len);
2228
2229 /* FIXME I need to close the fds! */
2230 return 0;
2231 } else
2232 return __write_init_super_ddf(st, 1);
2233 }
2234
2235 #endif
2236
2237 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize)
2238 {
2239 /* We must reserve the last 32Meg */
2240 if (devsize <= 32*1024*2)
2241 return 0;
2242 return devsize - 32*1024*2;
2243 }
2244
2245 #ifndef MDASSEMBLE
2246 static int
2247 validate_geometry_ddf_container(struct supertype *st,
2248 int level, int layout, int raiddisks,
2249 int chunk, unsigned long long size,
2250 char *dev, unsigned long long *freesize,
2251 int verbose);
2252
2253 static int validate_geometry_ddf_bvd(struct supertype *st,
2254 int level, int layout, int raiddisks,
2255 int chunk, unsigned long long size,
2256 char *dev, unsigned long long *freesize,
2257 int verbose);
2258
2259 static int validate_geometry_ddf(struct supertype *st,
2260 int level, int layout, int raiddisks,
2261 int chunk, unsigned long long size,
2262 char *dev, unsigned long long *freesize,
2263 int verbose)
2264 {
2265 int fd;
2266 struct mdinfo *sra;
2267 int cfd;
2268
2269 /* ddf potentially supports lots of things, but it depends on
2270 * what devices are offered (and maybe kernel version?)
2271 * If given unused devices, we will make a container.
2272 * If given devices in a container, we will make a BVD.
2273 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2274 */
2275
2276 if (level == LEVEL_CONTAINER) {
2277 /* Must be a fresh device to add to a container */
2278 return validate_geometry_ddf_container(st, level, layout,
2279 raiddisks, chunk,
2280 size, dev, freesize,
2281 verbose);
2282 }
2283
2284 if (st->sb) {
2285 /* A container has already been opened, so we are
2286 * creating in there. Maybe a BVD, maybe an SVD.
2287 * Should make a distinction one day.
2288 */
2289 return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
2290 chunk, size, dev, freesize,
2291 verbose);
2292 }
2293 if (!dev) {
2294 /* Initial sanity check. Exclude illegal levels. */
2295 int i;
2296 for (i=0; ddf_level_num[i].num1 != MAXINT; i++)
2297 if (ddf_level_num[i].num2 == level)
2298 break;
2299 if (ddf_level_num[i].num1 == MAXINT)
2300 return 0;
2301 /* Should check layout? etc */
2302 return 1;
2303 }
2304
2305 /* This is the first device for the array.
2306 * If it is a container, we read it in and do automagic allocations,
2307 * no other devices should be given.
2308 * Otherwise it must be a member device of a container, and we
2309 * do manual allocation.
2310 * Later we should check for a BVD and make an SVD.
2311 */
2312 fd = open(dev, O_RDONLY|O_EXCL, 0);
2313 if (fd >= 0) {
2314 sra = sysfs_read(fd, 0, GET_VERSION);
2315 close(fd);
2316 if (sra && sra->array.major_version == -1 &&
2317 strcmp(sra->text_version, "ddf") == 0) {
2318
2319 /* load super */
2320 /* find space for 'n' devices. */
2321 /* remember the devices */
2322 /* Somehow return the fact that we have enough */
2323 }
2324
2325 if (verbose)
2326 fprintf(stderr,
2327 Name ": ddf: Cannot create this array "
2328 "on device %s\n",
2329 dev);
2330 return 0;
2331 }
2332 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2333 if (verbose)
2334 fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2335 dev, strerror(errno));
2336 return 0;
2337 }
2338 /* Well, it is in use by someone, maybe a 'ddf' container. */
2339 cfd = open_container(fd);
2340 if (cfd < 0) {
2341 close(fd);
2342 if (verbose)
2343 fprintf(stderr, Name ": ddf: Cannot use %s: %s\n",
2344 dev, strerror(EBUSY));
2345 return 0;
2346 }
2347 sra = sysfs_read(cfd, 0, GET_VERSION);
2348 close(fd);
2349 if (sra && sra->array.major_version == -1 &&
2350 strcmp(sra->text_version, "ddf") == 0) {
2351 /* This is a member of a ddf container. Load the container
2352 * and try to create a bvd
2353 */
2354 struct ddf_super *ddf;
2355 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL, 1) == 0) {
2356 st->sb = ddf;
2357 st->container_dev = fd2devnum(cfd);
2358 close(cfd);
2359 return validate_geometry_ddf_bvd(st, level, layout,
2360 raiddisks, chunk, size,
2361 dev, freesize,
2362 verbose);
2363 }
2364 close(cfd);
2365 } else /* device may belong to a different container */
2366 return 0;
2367
2368 return 1;
2369 }
2370
2371 static int
2372 validate_geometry_ddf_container(struct supertype *st,
2373 int level, int layout, int raiddisks,
2374 int chunk, unsigned long long size,
2375 char *dev, unsigned long long *freesize,
2376 int verbose)
2377 {
2378 int fd;
2379 unsigned long long ldsize;
2380
2381 if (level != LEVEL_CONTAINER)
2382 return 0;
2383 if (!dev)
2384 return 1;
2385
2386 fd = open(dev, O_RDONLY|O_EXCL, 0);
2387 if (fd < 0) {
2388 if (verbose)
2389 fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2390 dev, strerror(errno));
2391 return 0;
2392 }
2393 if (!get_dev_size(fd, dev, &ldsize)) {
2394 close(fd);
2395 return 0;
2396 }
2397 close(fd);
2398
2399 *freesize = avail_size_ddf(st, ldsize >> 9);
2400
2401 return 1;
2402 }
2403
2404 static int validate_geometry_ddf_bvd(struct supertype *st,
2405 int level, int layout, int raiddisks,
2406 int chunk, unsigned long long size,
2407 char *dev, unsigned long long *freesize,
2408 int verbose)
2409 {
2410 struct stat stb;
2411 struct ddf_super *ddf = st->sb;
2412 struct dl *dl;
2413 unsigned long long pos = 0;
2414 unsigned long long maxsize;
2415 struct extent *e;
2416 int i;
2417 /* ddf/bvd supports lots of things, but not containers */
2418 if (level == LEVEL_CONTAINER)
2419 return 0;
2420 /* We must have the container info already read in. */
2421 if (!ddf)
2422 return 0;
2423
2424 if (!dev) {
2425 /* General test: make sure there is space for
2426 * 'raiddisks' device extents of size 'size'.
2427 */
2428 unsigned long long minsize = size;
2429 int dcnt = 0;
2430 if (minsize == 0)
2431 minsize = 8;
2432 for (dl = ddf->dlist; dl ; dl = dl->next)
2433 {
2434 int found = 0;
2435 pos = 0;
2436
2437 i = 0;
2438 e = get_extents(ddf, dl);
2439 if (!e) continue;
2440 do {
2441 unsigned long long esize;
2442 esize = e[i].start - pos;
2443 if (esize >= minsize)
2444 found = 1;
2445 pos = e[i].start + e[i].size;
2446 i++;
2447 } while (e[i-1].size);
2448 if (found)
2449 dcnt++;
2450 free(e);
2451 }
2452 if (dcnt < raiddisks) {
2453 if (verbose)
2454 fprintf(stderr,
2455 Name ": ddf: Not enough devices with "
2456 "space for this array (%d < %d)\n",
2457 dcnt, raiddisks);
2458 return 0;
2459 }
2460 return 1;
2461 }
2462 /* This device must be a member of the set */
2463 if (stat(dev, &stb) < 0)
2464 return 0;
2465 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2466 return 0;
2467 for (dl = ddf->dlist ; dl ; dl = dl->next) {
2468 if (dl->major == major(stb.st_rdev) &&
2469 dl->minor == minor(stb.st_rdev))
2470 break;
2471 }
2472 if (!dl) {
2473 if (verbose)
2474 fprintf(stderr, Name ": ddf: %s is not in the "
2475 "same DDF set\n",
2476 dev);
2477 return 0;
2478 }
2479 e = get_extents(ddf, dl);
2480 maxsize = 0;
2481 i = 0;
2482 if (e) do {
2483 unsigned long long esize;
2484 esize = e[i].start - pos;
2485 if (esize >= maxsize)
2486 maxsize = esize;
2487 pos = e[i].start + e[i].size;
2488 i++;
2489 } while (e[i-1].size);
2490 *freesize = maxsize;
2491 // FIXME here I am
2492
2493 return 1;
2494 }
2495
2496 static int load_super_ddf_all(struct supertype *st, int fd,
2497 void **sbp, char *devname, int keep_fd)
2498 {
2499 struct mdinfo *sra;
2500 struct ddf_super *super;
2501 struct mdinfo *sd, *best = NULL;
2502 int bestseq = 0;
2503 int seq;
2504 char nm[20];
2505 int dfd;
2506
2507 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
2508 if (!sra)
2509 return 1;
2510 if (sra->array.major_version != -1 ||
2511 sra->array.minor_version != -2 ||
2512 strcmp(sra->text_version, "ddf") != 0)
2513 return 1;
2514
2515 if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0)
2516 return 1;
2517 memset(super, 0, sizeof(*super));
2518
2519 /* first, try each device, and choose the best ddf */
2520 for (sd = sra->devs ; sd ; sd = sd->next) {
2521 int rv;
2522 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2523 dfd = dev_open(nm, O_RDONLY);
2524 if (dfd < 0)
2525 return 2;
2526 rv = load_ddf_headers(dfd, super, NULL);
2527 close(dfd);
2528 if (rv == 0) {
2529 seq = __be32_to_cpu(super->active->seq);
2530 if (super->active->openflag)
2531 seq--;
2532 if (!best || seq > bestseq) {
2533 bestseq = seq;
2534 best = sd;
2535 }
2536 }
2537 }
2538 if (!best)
2539 return 1;
2540 /* OK, load this ddf */
2541 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2542 dfd = dev_open(nm, O_RDONLY);
2543 if (dfd < 0)
2544 return 1;
2545 load_ddf_headers(dfd, super, NULL);
2546 load_ddf_global(dfd, super, NULL);
2547 close(dfd);
2548 /* Now we need the device-local bits */
2549 for (sd = sra->devs ; sd ; sd = sd->next) {
2550 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2551 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2552 if (dfd < 0)
2553 return 2;
2554 load_ddf_headers(dfd, super, NULL);
2555 seq = load_ddf_local(dfd, super, NULL, keep_fd);
2556 if (!keep_fd) close(dfd);
2557 }
2558 if (st->subarray[0]) {
2559 struct vcl *v;
2560
2561 for (v = super->conflist; v; v = v->next)
2562 if (v->vcnum == atoi(st->subarray))
2563 super->currentconf = v;
2564 if (!super->currentconf)
2565 return 1;
2566 }
2567 *sbp = super;
2568 if (st->ss == NULL) {
2569 st->ss = &super_ddf;
2570 st->minor_version = 0;
2571 st->max_devs = 512;
2572 st->container_dev = fd2devnum(fd);
2573 }
2574 st->loaded_container = 1;
2575 return 0;
2576 }
2577 #endif /* MDASSEMBLE */
2578
2579 static struct mdinfo *container_content_ddf(struct supertype *st)
2580 {
2581 /* Given a container loaded by load_super_ddf_all,
2582 * extract information about all the arrays into
2583 * an mdinfo tree.
2584 *
2585 * For each vcl in conflist: create an mdinfo, fill it in,
2586 * then look for matching devices (phys_refnum) in dlist
2587 * and create appropriate device mdinfo.
2588 */
2589 struct ddf_super *ddf = st->sb;
2590 struct mdinfo *rest = NULL;
2591 struct vcl *vc;
2592
2593 for (vc = ddf->conflist ; vc ; vc=vc->next)
2594 {
2595 int i;
2596 struct mdinfo *this;
2597 this = malloc(sizeof(*this));
2598 memset(this, 0, sizeof(*this));
2599 this->next = rest;
2600 rest = this;
2601
2602 this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2603 this->array.raid_disks =
2604 __be16_to_cpu(vc->conf.prim_elmnt_count);
2605 this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2606 this->array.raid_disks);
2607 this->array.md_minor = -1;
2608 this->array.major_version = -1;
2609 this->array.minor_version = -2;
2610 this->array.ctime = DECADE +
2611 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
2612 this->array.utime = DECADE +
2613 __be32_to_cpu(vc->conf.timestamp);
2614 this->array.chunk_size = 512 << vc->conf.chunk_shift;
2615
2616 i = vc->vcnum;
2617 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2618 (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
2619 DDF_init_full) {
2620 this->array.state = 0;
2621 this->resync_start = 0;
2622 } else {
2623 this->array.state = 1;
2624 this->resync_start = ~0ULL;
2625 }
2626 memcpy(this->name, ddf->virt->entries[i].name, 32);
2627 this->name[32]=0;
2628
2629 memset(this->uuid, 0, sizeof(this->uuid));
2630 this->component_size = __be64_to_cpu(vc->conf.blocks);
2631 this->array.size = this->component_size / 2;
2632 this->container_member = i;
2633
2634 ddf->currentconf = vc;
2635 uuid_from_super_ddf(st, this->uuid);
2636 ddf->currentconf = NULL;
2637
2638 sprintf(this->text_version, "/%s/%d",
2639 devnum2devname(st->container_dev),
2640 this->container_member);
2641
2642 for (i=0 ; i < ddf->mppe ; i++) {
2643 struct mdinfo *dev;
2644 struct dl *d;
2645
2646 if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
2647 continue;
2648
2649 this->array.working_disks++;
2650
2651 for (d = ddf->dlist; d ; d=d->next)
2652 if (d->disk.refnum == vc->conf.phys_refnum[i])
2653 break;
2654 if (d == NULL)
2655 break;
2656
2657 dev = malloc(sizeof(*dev));
2658 memset(dev, 0, sizeof(*dev));
2659 dev->next = this->devs;
2660 this->devs = dev;
2661
2662 dev->disk.number = __be32_to_cpu(d->disk.refnum);
2663 dev->disk.major = d->major;
2664 dev->disk.minor = d->minor;
2665 dev->disk.raid_disk = i;
2666 dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2667
2668 dev->events = __be32_to_cpu(ddf->primary.seq);
2669 dev->data_offset = __be64_to_cpu(vc->lba_offset[i]);
2670 dev->component_size = __be64_to_cpu(vc->conf.blocks);
2671 if (d->devname)
2672 strcpy(dev->name, d->devname);
2673 }
2674 }
2675 return rest;
2676 }
2677
2678 static int store_zero_ddf(struct supertype *st, int fd)
2679 {
2680 unsigned long long dsize;
2681 void *buf;
2682
2683 if (!get_dev_size(fd, NULL, &dsize))
2684 return 1;
2685
2686 posix_memalign(&buf, 512, 512);
2687 memset(buf, 0, 512);
2688
2689 lseek64(fd, dsize-512, 0);
2690 write(fd, buf, 512);
2691 free(buf);
2692 return 0;
2693 }
2694
2695 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
2696 {
2697 /*
2698 * return:
2699 * 0 same, or first was empty, and second was copied
2700 * 1 second had wrong number
2701 * 2 wrong uuid
2702 * 3 wrong other info
2703 */
2704 struct ddf_super *first = st->sb;
2705 struct ddf_super *second = tst->sb;
2706
2707 if (!first) {
2708 st->sb = tst->sb;
2709 tst->sb = NULL;
2710 return 0;
2711 }
2712
2713 if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
2714 return 2;
2715
2716 /* FIXME should I look at anything else? */
2717 return 0;
2718 }
2719
2720 #ifndef MDASSEMBLE
2721 /*
2722 * A new array 'a' has been started which claims to be instance 'inst'
2723 * within container 'c'.
2724 * We need to confirm that the array matches the metadata in 'c' so
2725 * that we don't corrupt any metadata.
2726 */
2727 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
2728 {
2729 dprintf("ddf: open_new %s\n", inst);
2730 a->info.container_member = atoi(inst);
2731 return 0;
2732 }
2733
2734 /*
2735 * The array 'a' is to be marked clean in the metadata.
2736 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2737 * clean up to the point (in sectors). If that cannot be recorded in the
2738 * metadata, then leave it as dirty.
2739 *
2740 * For DDF, we need to clear the DDF_state_inconsistent bit in the
2741 * !global! virtual_disk.virtual_entry structure.
2742 */
2743 static int ddf_set_array_state(struct active_array *a, int consistent)
2744 {
2745 struct ddf_super *ddf = a->container->sb;
2746 int inst = a->info.container_member;
2747 int old = ddf->virt->entries[inst].state;
2748 if (consistent == 2) {
2749 /* Should check if a recovery should be started FIXME */
2750 consistent = 1;
2751 if (a->resync_start != ~0ULL)
2752 consistent = 0;
2753 }
2754 if (consistent)
2755 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
2756 else
2757 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
2758 if (old != ddf->virt->entries[inst].state)
2759 ddf->updates_pending = 1;
2760
2761 old = ddf->virt->entries[inst].init_state;
2762 ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
2763 if (a->resync_start == ~0ULL)
2764 ddf->virt->entries[inst].init_state |= DDF_init_full;
2765 else if (a->resync_start == 0)
2766 ddf->virt->entries[inst].init_state |= DDF_init_not;
2767 else
2768 ddf->virt->entries[inst].init_state |= DDF_init_quick;
2769 if (old != ddf->virt->entries[inst].init_state)
2770 ddf->updates_pending = 1;
2771
2772 dprintf("ddf mark %d %s %llu\n", inst, consistent?"clean":"dirty",
2773 a->resync_start);
2774 return consistent;
2775 }
2776
2777 /*
2778 * The state of each disk is stored in the global phys_disk structure
2779 * in phys_disk.entries[n].state.
2780 * This makes various combinations awkward.
2781 * - When a device fails in any array, it must be failed in all arrays
2782 * that include a part of this device.
2783 * - When a component is rebuilding, we cannot include it officially in the
2784 * array unless this is the only array that uses the device.
2785 *
2786 * So: when transitioning:
2787 * Online -> failed, just set failed flag. monitor will propagate
2788 * spare -> online, the device might need to be added to the array.
2789 * spare -> failed, just set failed. Don't worry if in array or not.
2790 */
2791 static void ddf_set_disk(struct active_array *a, int n, int state)
2792 {
2793 struct ddf_super *ddf = a->container->sb;
2794 int inst = a->info.container_member;
2795 struct vd_config *vc = find_vdcr(ddf, inst);
2796 int pd = find_phys(ddf, vc->phys_refnum[n]);
2797 int i, st, working;
2798
2799 if (vc == NULL) {
2800 dprintf("ddf: cannot find instance %d!!\n", inst);
2801 return;
2802 }
2803 if (pd < 0) {
2804 /* disk doesn't currently exist. If it is now in_sync,
2805 * insert it. */
2806 if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
2807 /* Find dev 'n' in a->info->devs, determine the
2808 * ddf refnum, and set vc->phys_refnum and update
2809 * phys->entries[]
2810 */
2811 /* FIXME */
2812 }
2813 } else {
2814 int old = ddf->phys->entries[pd].state;
2815 if (state & DS_FAULTY)
2816 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Failed);
2817 if (state & DS_INSYNC) {
2818 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Online);
2819 ddf->phys->entries[pd].state &= __cpu_to_be16(~DDF_Rebuilding);
2820 }
2821 if (old != ddf->phys->entries[pd].state)
2822 ddf->updates_pending = 1;
2823 }
2824
2825 dprintf("ddf: set_disk %d to %x\n", n, state);
2826
2827 /* Now we need to check the state of the array and update
2828 * virtual_disk.entries[n].state.
2829 * It needs to be one of "optimal", "degraded", "failed".
2830 * I don't understand 'deleted' or 'missing'.
2831 */
2832 working = 0;
2833 for (i=0; i < a->info.array.raid_disks; i++) {
2834 pd = find_phys(ddf, vc->phys_refnum[i]);
2835 if (pd < 0)
2836 continue;
2837 st = __be16_to_cpu(ddf->phys->entries[pd].state);
2838 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
2839 == DDF_Online)
2840 working++;
2841 }
2842 state = DDF_state_degraded;
2843 if (working == a->info.array.raid_disks)
2844 state = DDF_state_optimal;
2845 else switch(vc->prl) {
2846 case DDF_RAID0:
2847 case DDF_CONCAT:
2848 case DDF_JBOD:
2849 state = DDF_state_failed;
2850 break;
2851 case DDF_RAID1:
2852 if (working == 0)
2853 state = DDF_state_failed;
2854 break;
2855 case DDF_RAID4:
2856 case DDF_RAID5:
2857 if (working < a->info.array.raid_disks-1)
2858 state = DDF_state_failed;
2859 break;
2860 case DDF_RAID6:
2861 if (working < a->info.array.raid_disks-2)
2862 state = DDF_state_failed;
2863 else if (working == a->info.array.raid_disks-1)
2864 state = DDF_state_part_optimal;
2865 break;
2866 }
2867
2868 if (ddf->virt->entries[inst].state !=
2869 ((ddf->virt->entries[inst].state & ~DDF_state_mask)
2870 | state)) {
2871
2872 ddf->virt->entries[inst].state =
2873 (ddf->virt->entries[inst].state & ~DDF_state_mask)
2874 | state;
2875 ddf->updates_pending = 1;
2876 }
2877
2878 }
2879
2880 static void ddf_sync_metadata(struct supertype *st)
2881 {
2882
2883 /*
2884 * Write all data to all devices.
2885 * Later, we might be able to track whether only local changes
2886 * have been made, or whether any global data has been changed,
2887 * but ddf is sufficiently weird that it probably always
2888 * changes global data ....
2889 */
2890 struct ddf_super *ddf = st->sb;
2891 if (!ddf->updates_pending)
2892 return;
2893 ddf->updates_pending = 0;
2894 __write_init_super_ddf(st, 0);
2895 dprintf("ddf: sync_metadata\n");
2896 }
2897
2898 static void ddf_process_update(struct supertype *st,
2899 struct metadata_update *update)
2900 {
2901 /* Apply this update to the metadata.
2902 * The first 4 bytes are a DDF_*_MAGIC which guides
2903 * our actions.
2904 * Possible update are:
2905 * DDF_PHYS_RECORDS_MAGIC
2906 * Add a new physical device. Changes to this record
2907 * only happen implicitly.
2908 * used_pdes is the device number.
2909 * DDF_VIRT_RECORDS_MAGIC
2910 * Add a new VD. Possibly also change the 'access' bits.
2911 * populated_vdes is the entry number.
2912 * DDF_VD_CONF_MAGIC
2913 * New or updated VD. the VIRT_RECORD must already
2914 * exist. For an update, phys_refnum and lba_offset
2915 * (at least) are updated, and the VD_CONF must
2916 * be written to precisely those devices listed with
2917 * a phys_refnum.
2918 * DDF_SPARE_ASSIGN_MAGIC
2919 * replacement Spare Assignment Record... but for which device?
2920 *
2921 * So, e.g.:
2922 * - to create a new array, we send a VIRT_RECORD and
2923 * a VD_CONF. Then assemble and start the array.
2924 * - to activate a spare we send a VD_CONF to add the phys_refnum
2925 * and offset. This will also mark the spare as active with
2926 * a spare-assignment record.
2927 */
2928 struct ddf_super *ddf = st->sb;
2929 __u32 *magic = (__u32*)update->buf;
2930 struct phys_disk *pd;
2931 struct virtual_disk *vd;
2932 struct vd_config *vc;
2933 struct vcl *vcl;
2934 struct dl *dl;
2935 int mppe;
2936 int ent;
2937
2938 dprintf("Process update %x\n", *magic);
2939
2940 switch (*magic) {
2941 case DDF_PHYS_RECORDS_MAGIC:
2942
2943 if (update->len != (sizeof(struct phys_disk) +
2944 sizeof(struct phys_disk_entry)))
2945 return;
2946 pd = (struct phys_disk*)update->buf;
2947
2948 ent = __be16_to_cpu(pd->used_pdes);
2949 if (ent >= __be16_to_cpu(ddf->phys->max_pdes))
2950 return;
2951 if (!all_ff(ddf->phys->entries[ent].guid))
2952 return;
2953 ddf->phys->entries[ent] = pd->entries[0];
2954 ddf->phys->used_pdes = __cpu_to_be16(1 +
2955 __be16_to_cpu(ddf->phys->used_pdes));
2956 ddf->updates_pending = 1;
2957 if (ddf->add_list) {
2958 struct active_array *a;
2959 struct dl *al = ddf->add_list;
2960 ddf->add_list = al->next;
2961
2962 al->next = ddf->dlist;
2963 ddf->dlist = al;
2964
2965 /* As a device has been added, we should check
2966 * for any degraded devices that might make
2967 * use of this spare */
2968 for (a = st->arrays ; a; a=a->next)
2969 a->check_degraded = 1;
2970 }
2971 break;
2972
2973 case DDF_VIRT_RECORDS_MAGIC:
2974
2975 if (update->len != (sizeof(struct virtual_disk) +
2976 sizeof(struct virtual_entry)))
2977 return;
2978 vd = (struct virtual_disk*)update->buf;
2979
2980 ent = __be16_to_cpu(vd->populated_vdes);
2981 if (ent >= __be16_to_cpu(ddf->virt->max_vdes))
2982 return;
2983 if (!all_ff(ddf->virt->entries[ent].guid))
2984 return;
2985 ddf->virt->entries[ent] = vd->entries[0];
2986 ddf->virt->populated_vdes = __cpu_to_be16(1 +
2987 __be16_to_cpu(ddf->virt->populated_vdes));
2988 ddf->updates_pending = 1;
2989 break;
2990
2991 case DDF_VD_CONF_MAGIC:
2992 dprintf("len %d %d\n", update->len, ddf->conf_rec_len);
2993
2994 mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries);
2995 if (update->len != ddf->conf_rec_len * 512)
2996 return;
2997 vc = (struct vd_config*)update->buf;
2998 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2999 if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
3000 break;
3001 dprintf("vcl = %p\n", vcl);
3002 if (vcl) {
3003 /* An update, just copy the phys_refnum and lba_offset
3004 * fields
3005 */
3006 memcpy(vcl->conf.phys_refnum, vc->phys_refnum,
3007 mppe * (sizeof(__u32) + sizeof(__u64)));
3008 } else {
3009 /* A new VD_CONF */
3010 vcl = update->space;
3011 update->space = NULL;
3012 vcl->next = ddf->conflist;
3013 memcpy(&vcl->conf, vc, update->len);
3014 vcl->lba_offset = (__u64*)
3015 &vcl->conf.phys_refnum[mppe];
3016 ddf->conflist = vcl;
3017 }
3018 /* Now make sure vlist is correct for each dl. */
3019 for (dl = ddf->dlist; dl; dl = dl->next) {
3020 int dn;
3021 int vn = 0;
3022 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
3023 for (dn=0; dn < ddf->mppe ; dn++)
3024 if (vcl->conf.phys_refnum[dn] ==
3025 dl->disk.refnum) {
3026 dprintf("dev %d has %p at %d\n",
3027 dl->pdnum, vcl, vn);
3028 dl->vlist[vn++] = vcl;
3029 break;
3030 }
3031 while (vn < ddf->max_part)
3032 dl->vlist[vn++] = NULL;
3033 if (dl->vlist[0]) {
3034 ddf->phys->entries[dl->pdnum].type &=
3035 ~__cpu_to_be16(DDF_Global_Spare);
3036 ddf->phys->entries[dl->pdnum].type |=
3037 __cpu_to_be16(DDF_Active_in_VD);
3038 }
3039 if (dl->spare) {
3040 ddf->phys->entries[dl->pdnum].type &=
3041 ~__cpu_to_be16(DDF_Global_Spare);
3042 ddf->phys->entries[dl->pdnum].type |=
3043 __cpu_to_be16(DDF_Spare);
3044 }
3045 if (!dl->vlist[0] && !dl->spare) {
3046 ddf->phys->entries[dl->pdnum].type |=
3047 __cpu_to_be16(DDF_Global_Spare);
3048 ddf->phys->entries[dl->pdnum].type &=
3049 ~__cpu_to_be16(DDF_Spare |
3050 DDF_Active_in_VD);
3051 }
3052 }
3053 ddf->updates_pending = 1;
3054 break;
3055 case DDF_SPARE_ASSIGN_MAGIC:
3056 default: break;
3057 }
3058 }
3059
3060 static void ddf_prepare_update(struct supertype *st,
3061 struct metadata_update *update)
3062 {
3063 /* This update arrived at managemon.
3064 * We are about to pass it to monitor.
3065 * If a malloc is needed, do it here.
3066 */
3067 struct ddf_super *ddf = st->sb;
3068 __u32 *magic = (__u32*)update->buf;
3069 if (*magic == DDF_VD_CONF_MAGIC)
3070 posix_memalign(&update->space, 512,
3071 offsetof(struct vcl, conf)
3072 + ddf->conf_rec_len * 512);
3073 }
3074
3075 /*
3076 * Check if the array 'a' is degraded but not failed.
3077 * If it is, find as many spares as are available and needed and
3078 * arrange for their inclusion.
3079 * We only choose devices which are not already in the array,
3080 * and prefer those with a spare-assignment to this array.
3081 * otherwise we choose global spares - assuming always that
3082 * there is enough room.
3083 * For each spare that we assign, we return an 'mdinfo' which
3084 * describes the position for the device in the array.
3085 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
3086 * the new phys_refnum and lba_offset values.
3087 *
3088 * Only worry about BVDs at the moment.
3089 */
3090 static struct mdinfo *ddf_activate_spare(struct active_array *a,
3091 struct metadata_update **updates)
3092 {
3093 int working = 0;
3094 struct mdinfo *d;
3095 struct ddf_super *ddf = a->container->sb;
3096 int global_ok = 0;
3097 struct mdinfo *rv = NULL;
3098 struct mdinfo *di;
3099 struct metadata_update *mu;
3100 struct dl *dl;
3101 int i;
3102 struct vd_config *vc;
3103 __u64 *lba;
3104
3105 for (d = a->info.devs ; d ; d = d->next) {
3106 if ((d->curr_state & DS_FAULTY) &&
3107 d->state_fd >= 0)
3108 /* wait for Removal to happen */
3109 return NULL;
3110 if (d->state_fd >= 0)
3111 working ++;
3112 }
3113
3114 dprintf("ddf_activate: working=%d (%d) level=%d\n", working, a->info.array.raid_disks,
3115 a->info.array.level);
3116 if (working == a->info.array.raid_disks)
3117 return NULL; /* array not degraded */
3118 switch (a->info.array.level) {
3119 case 1:
3120 if (working == 0)
3121 return NULL; /* failed */
3122 break;
3123 case 4:
3124 case 5:
3125 if (working < a->info.array.raid_disks - 1)
3126 return NULL; /* failed */
3127 break;
3128 case 6:
3129 if (working < a->info.array.raid_disks - 2)
3130 return NULL; /* failed */
3131 break;
3132 default: /* concat or stripe */
3133 return NULL; /* failed */
3134 }
3135
3136 /* For each slot, if it is not working, find a spare */
3137 dl = ddf->dlist;
3138 for (i = 0; i < a->info.array.raid_disks; i++) {
3139 for (d = a->info.devs ; d ; d = d->next)
3140 if (d->disk.raid_disk == i)
3141 break;
3142 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
3143 if (d && (d->state_fd >= 0))
3144 continue;
3145
3146 /* OK, this device needs recovery. Find a spare */
3147 again:
3148 for ( ; dl ; dl = dl->next) {
3149 unsigned long long esize;
3150 unsigned long long pos;
3151 struct mdinfo *d2;
3152 int is_global = 0;
3153 int is_dedicated = 0;
3154 struct extent *ex;
3155 int j;
3156 /* If in this array, skip */
3157 for (d2 = a->info.devs ; d2 ; d2 = d2->next)
3158 if (d2->disk.major == dl->major &&
3159 d2->disk.minor == dl->minor) {
3160 dprintf("%x:%x already in array\n", dl->major, dl->minor);
3161 break;
3162 }
3163 if (d2)
3164 continue;
3165 if (ddf->phys->entries[dl->pdnum].type &
3166 __cpu_to_be16(DDF_Spare)) {
3167 /* Check spare assign record */
3168 if (dl->spare) {
3169 if (dl->spare->type & DDF_spare_dedicated) {
3170 /* check spare_ents for guid */
3171 for (j = 0 ;
3172 j < __be16_to_cpu(dl->spare->populated);
3173 j++) {
3174 if (memcmp(dl->spare->spare_ents[j].guid,
3175 ddf->virt->entries[a->info.container_member].guid,
3176 DDF_GUID_LEN) == 0)
3177 is_dedicated = 1;
3178 }
3179 } else
3180 is_global = 1;
3181 }
3182 } else if (ddf->phys->entries[dl->pdnum].type &
3183 __cpu_to_be16(DDF_Global_Spare)) {
3184 is_global = 1;
3185 }
3186 if ( ! (is_dedicated ||
3187 (is_global && global_ok))) {
3188 dprintf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
3189 is_dedicated, is_global);
3190 continue;
3191 }
3192
3193 /* We are allowed to use this device - is there space?
3194 * We need a->info.component_size sectors */
3195 ex = get_extents(ddf, dl);
3196 if (!ex) {
3197 dprintf("cannot get extents\n");
3198 continue;
3199 }
3200 j = 0; pos = 0;
3201 esize = 0;
3202
3203 do {
3204 esize = ex[j].start - pos;
3205 if (esize >= a->info.component_size)
3206 break;
3207 pos = ex[i].start + ex[i].size;
3208 i++;
3209 } while (ex[i-1].size);
3210
3211 free(ex);
3212 if (esize < a->info.component_size) {
3213 dprintf("%x:%x has no room: %llu %llu\n", dl->major, dl->minor,
3214 esize, a->info.component_size);
3215 /* No room */
3216 continue;
3217 }
3218
3219 /* Cool, we have a device with some space at pos */
3220 di = malloc(sizeof(*di));
3221 memset(di, 0, sizeof(*di));
3222 di->disk.number = i;
3223 di->disk.raid_disk = i;
3224 di->disk.major = dl->major;
3225 di->disk.minor = dl->minor;
3226 di->disk.state = 0;
3227 di->data_offset = pos;
3228 di->component_size = a->info.component_size;
3229 di->container_member = dl->pdnum;
3230 di->next = rv;
3231 rv = di;
3232 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3233 i, pos);
3234
3235 break;
3236 }
3237 if (!dl && ! global_ok) {
3238 /* not enough dedicated spares, try global */
3239 global_ok = 1;
3240 dl = ddf->dlist;
3241 goto again;
3242 }
3243 }
3244
3245 if (!rv)
3246 /* No spares found */
3247 return rv;
3248 /* Now 'rv' has a list of devices to return.
3249 * Create a metadata_update record to update the
3250 * phys_refnum and lba_offset values
3251 */
3252 mu = malloc(sizeof(*mu));
3253 mu->buf = malloc(ddf->conf_rec_len * 512);
3254 posix_memalign(&mu->space, 512, sizeof(struct vcl));
3255 mu->len = ddf->conf_rec_len;
3256 mu->next = *updates;
3257 vc = find_vdcr(ddf, a->info.container_member);
3258 memcpy(mu->buf, vc, ddf->conf_rec_len * 512);
3259
3260 vc = (struct vd_config*)mu->buf;
3261 lba = (__u64*)&vc->phys_refnum[ddf->mppe];
3262 for (di = rv ; di ; di = di->next) {
3263 vc->phys_refnum[di->disk.raid_disk] =
3264 ddf->phys->entries[dl->pdnum].refnum;
3265 lba[di->disk.raid_disk] = di->data_offset;
3266 }
3267 *updates = mu;
3268 return rv;
3269 }
3270 #endif /* MDASSEMBLE */
3271
3272 struct superswitch super_ddf = {
3273 #ifndef MDASSEMBLE
3274 .examine_super = examine_super_ddf,
3275 .brief_examine_super = brief_examine_super_ddf,
3276 .detail_super = detail_super_ddf,
3277 .brief_detail_super = brief_detail_super_ddf,
3278 .validate_geometry = validate_geometry_ddf,
3279 .write_init_super = write_init_super_ddf,
3280 .add_to_super = add_to_super_ddf,
3281 #endif
3282 .match_home = match_home_ddf,
3283 .uuid_from_super= uuid_from_super_ddf,
3284 .getinfo_super = getinfo_super_ddf,
3285 .update_super = update_super_ddf,
3286
3287 .avail_size = avail_size_ddf,
3288
3289 .compare_super = compare_super_ddf,
3290
3291 .load_super = load_super_ddf,
3292 .init_super = init_super_ddf,
3293 .store_super = store_zero_ddf,
3294 .free_super = free_super_ddf,
3295 .match_metadata_desc = match_metadata_desc_ddf,
3296 .container_content = container_content_ddf,
3297
3298 .external = 1,
3299
3300 #ifndef MDASSEMBLE
3301 /* for mdmon */
3302 .open_new = ddf_open_new,
3303 .set_array_state= ddf_set_array_state,
3304 .set_disk = ddf_set_disk,
3305 .sync_metadata = ddf_sync_metadata,
3306 .process_update = ddf_process_update,
3307 .prepare_update = ddf_prepare_update,
3308 .activate_spare = ddf_activate_spare,
3309 #endif
3310 };