]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super-ddf.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Add subarray field to supertype.
[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 static inline int ROUND_UP(int a, int base)
35 {
36 return ((a+base-1)/base)*base;
37 }
38
39 /* a non-official T10 name for creation GUIDs */
40 static char T10[] = "Linux-MD";
41
42 /* DDF timestamps are 1980 based, so we need to add
43 * second-in-decade-of-seventies to convert to linux timestamps.
44 * 10 years with 2 leap years.
45 */
46 #define DECADE (3600*24*(365*10+2))
47 unsigned long crc32(
48 unsigned long crc,
49 const unsigned char *buf,
50 unsigned len);
51
52 /* The DDF metadata handling.
53 * DDF metadata lives at the end of the device.
54 * The last 512 byte block provides an 'anchor' which is used to locate
55 * the rest of the metadata which usually lives immediately behind the anchor.
56 *
57 * Note:
58 * - all multibyte numeric fields are bigendian.
59 * - all strings are space padded.
60 *
61 */
62
63 /* Primary Raid Level (PRL) */
64 #define DDF_RAID0 0x00
65 #define DDF_RAID1 0x01
66 #define DDF_RAID3 0x03
67 #define DDF_RAID4 0x04
68 #define DDF_RAID5 0x05
69 #define DDF_RAID1E 0x11
70 #define DDF_JBOD 0x0f
71 #define DDF_CONCAT 0x1f
72 #define DDF_RAID5E 0x15
73 #define DDF_RAID5EE 0x25
74 #define DDF_RAID6 0x06
75
76 /* Raid Level Qualifier (RLQ) */
77 #define DDF_RAID0_SIMPLE 0x00
78 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
79 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
80 #define DDF_RAID3_0 0x00 /* parity in first extent */
81 #define DDF_RAID3_N 0x01 /* parity in last extent */
82 #define DDF_RAID4_0 0x00 /* parity in first extent */
83 #define DDF_RAID4_N 0x01 /* parity in last extent */
84 /* these apply to raid5e and raid5ee as well */
85 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
86 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
87 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
88 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
89
90 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
91 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
92
93 /* Secondary RAID Level (SRL) */
94 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
95 #define DDF_2MIRRORED 0x01
96 #define DDF_2CONCAT 0x02
97 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
98
99 /* Magic numbers */
100 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
101 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
102 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
103 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
104 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
105 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
106 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
107 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
108 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
109 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
110
111 #define DDF_GUID_LEN 24
112 #define DDF_REVISION_0 "01.00.00"
113 #define DDF_REVISION_2 "01.02.00"
114
115 struct ddf_header {
116 __u32 magic; /* DDF_HEADER_MAGIC */
117 __u32 crc;
118 char guid[DDF_GUID_LEN];
119 char revision[8]; /* 01.02.00 */
120 __u32 seq; /* starts at '1' */
121 __u32 timestamp;
122 __u8 openflag;
123 __u8 foreignflag;
124 __u8 enforcegroups;
125 __u8 pad0; /* 0xff */
126 __u8 pad1[12]; /* 12 * 0xff */
127 /* 64 bytes so far */
128 __u8 header_ext[32]; /* reserved: fill with 0xff */
129 __u64 primary_lba;
130 __u64 secondary_lba;
131 __u8 type;
132 __u8 pad2[3]; /* 0xff */
133 __u32 workspace_len; /* sectors for vendor space -
134 * at least 32768(sectors) */
135 __u64 workspace_lba;
136 __u16 max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */
137 __u16 max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */
138 __u16 max_partitions; /* i.e. max num of configuration
139 record entries per disk */
140 __u16 config_record_len; /* 1 +ROUNDUP(max_primary_element_entries
141 *12/512) */
142 __u16 max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */
143 __u8 pad3[54]; /* 0xff */
144 /* 192 bytes so far */
145 __u32 controller_section_offset;
146 __u32 controller_section_length;
147 __u32 phys_section_offset;
148 __u32 phys_section_length;
149 __u32 virt_section_offset;
150 __u32 virt_section_length;
151 __u32 config_section_offset;
152 __u32 config_section_length;
153 __u32 data_section_offset;
154 __u32 data_section_length;
155 __u32 bbm_section_offset;
156 __u32 bbm_section_length;
157 __u32 diag_space_offset;
158 __u32 diag_space_length;
159 __u32 vendor_offset;
160 __u32 vendor_length;
161 /* 256 bytes so far */
162 __u8 pad4[256]; /* 0xff */
163 };
164
165 /* type field */
166 #define DDF_HEADER_ANCHOR 0x00
167 #define DDF_HEADER_PRIMARY 0x01
168 #define DDF_HEADER_SECONDARY 0x02
169
170 /* The content of the 'controller section' - global scope */
171 struct ddf_controller_data {
172 __u32 magic; /* DDF_CONTROLLER_MAGIC */
173 __u32 crc;
174 char guid[DDF_GUID_LEN];
175 struct controller_type {
176 __u16 vendor_id;
177 __u16 device_id;
178 __u16 sub_vendor_id;
179 __u16 sub_device_id;
180 } type;
181 char product_id[16];
182 __u8 pad[8]; /* 0xff */
183 __u8 vendor_data[448];
184 };
185
186 /* The content of phys_section - global scope */
187 struct phys_disk {
188 __u32 magic; /* DDF_PHYS_RECORDS_MAGIC */
189 __u32 crc;
190 __u16 used_pdes;
191 __u16 max_pdes;
192 __u8 pad[52];
193 struct phys_disk_entry {
194 char guid[DDF_GUID_LEN];
195 __u32 refnum;
196 __u16 type;
197 __u16 state;
198 __u64 config_size; /* DDF structures must be after here */
199 char path[18]; /* another horrible structure really */
200 __u8 pad[6];
201 } entries[0];
202 };
203
204 /* phys_disk_entry.type is a bitmap - bigendian remember */
205 #define DDF_Forced_PD_GUID 1
206 #define DDF_Active_in_VD 2
207 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
208 #define DDF_Spare 8 /* overrides Global_spare */
209 #define DDF_Foreign 16
210 #define DDF_Legacy 32 /* no DDF on this device */
211
212 #define DDF_Interface_mask 0xf00
213 #define DDF_Interface_SCSI 0x100
214 #define DDF_Interface_SAS 0x200
215 #define DDF_Interface_SATA 0x300
216 #define DDF_Interface_FC 0x400
217
218 /* phys_disk_entry.state is a bigendian bitmap */
219 #define DDF_Online 1
220 #define DDF_Failed 2 /* overrides 1,4,8 */
221 #define DDF_Rebuilding 4
222 #define DDF_Transition 8
223 #define DDF_SMART 16
224 #define DDF_ReadErrors 32
225 #define DDF_Missing 64
226
227 /* The content of the virt_section global scope */
228 struct virtual_disk {
229 __u32 magic; /* DDF_VIRT_RECORDS_MAGIC */
230 __u32 crc;
231 __u16 populated_vdes;
232 __u16 max_vdes;
233 __u8 pad[52];
234 struct virtual_entry {
235 char guid[DDF_GUID_LEN];
236 __u16 unit;
237 __u16 pad0; /* 0xffff */
238 __u16 guid_crc;
239 __u16 type;
240 __u8 state;
241 __u8 init_state;
242 __u8 pad1[14];
243 char name[16];
244 } entries[0];
245 };
246
247 /* virtual_entry.type is a bitmap - bigendian */
248 #define DDF_Shared 1
249 #define DDF_Enforce_Groups 2
250 #define DDF_Unicode 4
251 #define DDF_Owner_Valid 8
252
253 /* virtual_entry.state is a bigendian bitmap */
254 #define DDF_state_mask 0x7
255 #define DDF_state_optimal 0x0
256 #define DDF_state_degraded 0x1
257 #define DDF_state_deleted 0x2
258 #define DDF_state_missing 0x3
259 #define DDF_state_failed 0x4
260 #define DDF_state_part_optimal 0x5
261
262 #define DDF_state_morphing 0x8
263 #define DDF_state_inconsistent 0x10
264
265 /* virtual_entry.init_state is a bigendian bitmap */
266 #define DDF_initstate_mask 0x03
267 #define DDF_init_not 0x00
268 #define DDF_init_quick 0x01 /* initialisation is progress.
269 * i.e. 'state_inconsistent' */
270 #define DDF_init_full 0x02
271
272 #define DDF_access_mask 0xc0
273 #define DDF_access_rw 0x00
274 #define DDF_access_ro 0x80
275 #define DDF_access_blocked 0xc0
276
277 /* The content of the config_section - local scope
278 * It has multiple records each config_record_len sectors
279 * They can be vd_config or spare_assign
280 */
281
282 struct vd_config {
283 __u32 magic; /* DDF_VD_CONF_MAGIC */
284 __u32 crc;
285 char guid[DDF_GUID_LEN];
286 __u32 timestamp;
287 __u32 seqnum;
288 __u8 pad0[24];
289 __u16 prim_elmnt_count;
290 __u8 chunk_shift; /* 0 == 512, 1==1024 etc */
291 __u8 prl;
292 __u8 rlq;
293 __u8 sec_elmnt_count;
294 __u8 sec_elmnt_seq;
295 __u8 srl;
296 __u64 blocks; /* blocks per component could be different
297 * on different component devices...(only
298 * for concat I hope) */
299 __u64 array_blocks; /* blocks in array */
300 __u8 pad1[8];
301 __u32 spare_refs[8];
302 __u8 cache_pol[8];
303 __u8 bg_rate;
304 __u8 pad2[3];
305 __u8 pad3[52];
306 __u8 pad4[192];
307 __u8 v0[32]; /* reserved- 0xff */
308 __u8 v1[32]; /* reserved- 0xff */
309 __u8 v2[16]; /* reserved- 0xff */
310 __u8 v3[16]; /* reserved- 0xff */
311 __u8 vendor[32];
312 __u32 phys_refnum[0]; /* refnum of each disk in sequence */
313 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
314 bvd are always the same size */
315 };
316
317 /* vd_config.cache_pol[7] is a bitmap */
318 #define DDF_cache_writeback 1 /* else writethrough */
319 #define DDF_cache_wadaptive 2 /* only applies if writeback */
320 #define DDF_cache_readahead 4
321 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
322 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
323 #define DDF_cache_wallowed 32 /* enable write caching */
324 #define DDF_cache_rallowed 64 /* enable read caching */
325
326 struct spare_assign {
327 __u32 magic; /* DDF_SPARE_ASSIGN_MAGIC */
328 __u32 crc;
329 __u32 timestamp;
330 __u8 reserved[7];
331 __u8 type;
332 __u16 populated; /* SAEs used */
333 __u16 max; /* max SAEs */
334 __u8 pad[8];
335 struct spare_assign_entry {
336 char guid[DDF_GUID_LEN];
337 __u16 secondary_element;
338 __u8 pad[6];
339 } spare_ents[0];
340 };
341 /* spare_assign.type is a bitmap */
342 #define DDF_spare_dedicated 0x1 /* else global */
343 #define DDF_spare_revertible 0x2 /* else committable */
344 #define DDF_spare_active 0x4 /* else not active */
345 #define DDF_spare_affinity 0x8 /* enclosure affinity */
346
347 /* The data_section contents - local scope */
348 struct disk_data {
349 __u32 magic; /* DDF_PHYS_DATA_MAGIC */
350 __u32 crc;
351 char guid[DDF_GUID_LEN];
352 __u32 refnum; /* crc of some magic drive data ... */
353 __u8 forced_ref; /* set when above was not result of magic */
354 __u8 forced_guid; /* set if guid was forced rather than magic */
355 __u8 vendor[32];
356 __u8 pad[442];
357 };
358
359 /* bbm_section content */
360 struct bad_block_log {
361 __u32 magic;
362 __u32 crc;
363 __u16 entry_count;
364 __u32 spare_count;
365 __u8 pad[10];
366 __u64 first_spare;
367 struct mapped_block {
368 __u64 defective_start;
369 __u32 replacement_start;
370 __u16 remap_count;
371 __u8 pad[2];
372 } entries[0];
373 };
374
375 /* Struct for internally holding ddf structures */
376 /* The DDF structure stored on each device is potentially
377 * quite different, as some data is global and some is local.
378 * The global data is:
379 * - ddf header
380 * - controller_data
381 * - Physical disk records
382 * - Virtual disk records
383 * The local data is:
384 * - Configuration records
385 * - Physical Disk data section
386 * ( and Bad block and vendor which I don't care about yet).
387 *
388 * The local data is parsed into separate lists as it is read
389 * and reconstructed for writing. This means that we only need
390 * to make config changes once and they are automatically
391 * propagated to all devices.
392 * Note that the ddf_super has space of the conf and disk data
393 * for this disk and also for a list of all such data.
394 * The list is only used for the superblock that is being
395 * built in Create or Assemble to describe the whole array.
396 */
397 struct ddf_super {
398 struct ddf_header anchor, primary, secondary, *active;
399 struct ddf_controller_data controller;
400 struct phys_disk *phys;
401 struct virtual_disk *virt;
402 int pdsize, vdsize;
403 int max_part, mppe, conf_rec_len;
404 struct vcl {
405 struct vcl *next;
406 __u64 *lba_offset; /* location in 'conf' of
407 * the lba table */
408 int vcnum; /* index into ->virt */
409 __u64 *block_sizes; /* NULL if all the same */
410 struct vd_config conf;
411 } *conflist, *newconf;
412 struct dl {
413 struct dl *next;
414 struct disk_data disk;
415 int major, minor;
416 char *devname;
417 int fd;
418 int pdnum; /* index in ->phys */
419 struct spare_assign *spare;
420 struct vcl *vlist[0]; /* max_part in size */
421 } *dlist;
422 };
423
424 #ifndef offsetof
425 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
426 #endif
427
428 static struct superswitch super_ddf_container, super_ddf_bvd, super_ddf_svd;
429
430 static int calc_crc(void *buf, int len)
431 {
432 /* crcs are always at the same place as in the ddf_header */
433 struct ddf_header *ddf = buf;
434 __u32 oldcrc = ddf->crc;
435 __u32 newcrc;
436 ddf->crc = 0xffffffff;
437
438 newcrc = crc32(0, buf, len);
439 ddf->crc = oldcrc;
440 return newcrc;
441 }
442
443 static int load_ddf_header(int fd, unsigned long long lba,
444 unsigned long long size,
445 int type,
446 struct ddf_header *hdr, struct ddf_header *anchor)
447 {
448 /* read a ddf header (primary or secondary) from fd/lba
449 * and check that it is consistent with anchor
450 * Need to check:
451 * magic, crc, guid, rev, and LBA's header_type, and
452 * everything after header_type must be the same
453 */
454 if (lba >= size-1)
455 return 0;
456
457 if (lseek64(fd, lba<<9, 0) < 0)
458 return 0;
459
460 if (read(fd, hdr, 512) != 512)
461 return 0;
462
463 if (hdr->magic != DDF_HEADER_MAGIC)
464 return 0;
465 if (calc_crc(hdr, 512) != hdr->crc)
466 return 0;
467 if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
468 memcmp(anchor->revision, hdr->revision, 8) != 0 ||
469 anchor->primary_lba != hdr->primary_lba ||
470 anchor->secondary_lba != hdr->secondary_lba ||
471 hdr->type != type ||
472 memcmp(anchor->pad2, hdr->pad2, 512 -
473 offsetof(struct ddf_header, pad2)) != 0)
474 return 0;
475
476 /* Looks good enough to me... */
477 return 1;
478 }
479
480 static void *load_section(int fd, struct ddf_super *super, void *buf,
481 __u32 offset_be, __u32 len_be, int check)
482 {
483 unsigned long long offset = __be32_to_cpu(offset_be);
484 unsigned long long len = __be32_to_cpu(len_be);
485 int dofree = (buf == NULL);
486
487 if (check)
488 if (len != 2 && len != 8 && len != 32
489 && len != 128 && len != 512)
490 return NULL;
491
492 if (len > 1024)
493 return NULL;
494 if (buf) {
495 /* All pre-allocated sections are a single block */
496 if (len != 1)
497 return NULL;
498 } else
499 buf = malloc(len<<9);
500 if (!buf)
501 return NULL;
502
503 if (super->active->type == 1)
504 offset += __be64_to_cpu(super->active->primary_lba);
505 else
506 offset += __be64_to_cpu(super->active->secondary_lba);
507
508 if (lseek64(fd, offset<<9, 0) != (offset<<9)) {
509 if (dofree)
510 free(buf);
511 return NULL;
512 }
513 if (read(fd, buf, len<<9) != (len<<9)) {
514 if (dofree)
515 free(buf);
516 return NULL;
517 }
518 return buf;
519 }
520
521 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
522 {
523 unsigned long long dsize;
524
525 get_dev_size(fd, NULL, &dsize);
526
527 if (lseek64(fd, dsize-512, 0) < 0) {
528 if (devname)
529 fprintf(stderr,
530 Name": Cannot seek to anchor block on %s: %s\n",
531 devname, strerror(errno));
532 return 1;
533 }
534 if (read(fd, &super->anchor, 512) != 512) {
535 if (devname)
536 fprintf(stderr,
537 Name ": Cannot read anchor block on %s: %s\n",
538 devname, strerror(errno));
539 return 1;
540 }
541 if (super->anchor.magic != DDF_HEADER_MAGIC) {
542 if (devname)
543 fprintf(stderr, Name ": no DDF anchor found on %s\n",
544 devname);
545 return 2;
546 }
547 if (calc_crc(&super->anchor, 512) != super->anchor.crc) {
548 if (devname)
549 fprintf(stderr, Name ": bad CRC on anchor on %s\n",
550 devname);
551 return 2;
552 }
553 if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
554 memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
555 if (devname)
556 fprintf(stderr, Name ": can only support super revision"
557 " %.8s and earlier, not %.8s on %s\n",
558 DDF_REVISION_2, super->anchor.revision,devname);
559 return 2;
560 }
561 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba),
562 dsize >> 9, 1,
563 &super->primary, &super->anchor) == 0) {
564 if (devname)
565 fprintf(stderr,
566 Name ": Failed to load primary DDF header "
567 "on %s\n", devname);
568 return 2;
569 }
570 super->active = &super->primary;
571 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.secondary_lba),
572 dsize >> 9, 2,
573 &super->secondary, &super->anchor)) {
574 if ((__be32_to_cpu(super->primary.seq)
575 < __be32_to_cpu(super->secondary.seq) &&
576 !super->secondary.openflag)
577 || (__be32_to_cpu(super->primary.seq)
578 == __be32_to_cpu(super->secondary.seq) &&
579 super->primary.openflag && !super->secondary.openflag)
580 )
581 super->active = &super->secondary;
582 }
583 return 0;
584 }
585
586 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
587 {
588 void *ok;
589 ok = load_section(fd, super, &super->controller,
590 super->active->controller_section_offset,
591 super->active->controller_section_length,
592 0);
593 super->phys = load_section(fd, super, NULL,
594 super->active->phys_section_offset,
595 super->active->phys_section_length,
596 1);
597 super->pdsize = __be32_to_cpu(super->active->phys_section_length) * 512;
598
599 super->virt = load_section(fd, super, NULL,
600 super->active->virt_section_offset,
601 super->active->virt_section_length,
602 1);
603 super->vdsize = __be32_to_cpu(super->active->virt_section_length) * 512;
604 if (!ok ||
605 !super->phys ||
606 !super->virt) {
607 free(super->phys);
608 free(super->virt);
609 super->phys = NULL;
610 super->virt = NULL;
611 return 2;
612 }
613 super->conflist = NULL;
614 super->dlist = NULL;
615
616 super->max_part = __be16_to_cpu(super->active->max_partitions);
617 super->mppe = __be16_to_cpu(super->active->max_primary_element_entries);
618 super->conf_rec_len = __be16_to_cpu(super->active->config_record_len);
619 return 0;
620 }
621
622 static int load_ddf_local(int fd, struct ddf_super *super,
623 char *devname, int keep)
624 {
625 struct dl *dl;
626 struct stat stb;
627 char *conf;
628 int i;
629 int vnum;
630 int max_virt_disks = __be16_to_cpu(super->active->max_vd_entries);
631
632 /* First the local disk info */
633 dl = malloc(sizeof(*dl) +
634 (super->max_part) * sizeof(dl->vlist[0]));
635
636 load_section(fd, super, &dl->disk,
637 super->active->data_section_offset,
638 super->active->data_section_length,
639 0);
640 dl->devname = devname ? strdup(devname) : NULL;
641
642 fstat(fd, &stb);
643 dl->major = major(stb.st_rdev);
644 dl->minor = minor(stb.st_rdev);
645 dl->next = super->dlist;
646 dl->fd = keep ? fd : -1;
647 dl->spare = NULL;
648 for (i=0 ; i < super->max_part ; i++)
649 dl->vlist[i] = NULL;
650 super->dlist = dl;
651 dl->pdnum = -1;
652 for (i=0; i < __be16_to_cpu(super->active->max_pd_entries); i++)
653 if (memcmp(super->phys->entries[i].guid,
654 dl->disk.guid, DDF_GUID_LEN) == 0)
655 dl->pdnum = i;
656
657 /* Now the config list. */
658 /* 'conf' is an array of config entries, some of which are
659 * probably invalid. Those which are good need to be copied into
660 * the conflist
661 */
662
663 conf = load_section(fd, super, NULL,
664 super->active->config_section_offset,
665 super->active->config_section_length,
666 0);
667
668 vnum = 0;
669 for (i = 0;
670 i < __be32_to_cpu(super->active->config_section_length);
671 i += super->conf_rec_len) {
672 struct vd_config *vd =
673 (struct vd_config *)((char*)conf + i*512);
674 struct vcl *vcl;
675
676 if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) {
677 if (dl->spare)
678 continue;
679 dl->spare = malloc(super->conf_rec_len*512);
680 memcpy(dl->spare, vd, super->conf_rec_len*512);
681 continue;
682 }
683 if (vd->magic != DDF_VD_CONF_MAGIC)
684 continue;
685 for (vcl = super->conflist; vcl; vcl = vcl->next) {
686 if (memcmp(vcl->conf.guid,
687 vd->guid, DDF_GUID_LEN) == 0)
688 break;
689 }
690
691 if (vcl) {
692 dl->vlist[vnum++] = vcl;
693 if (__be32_to_cpu(vd->seqnum) <=
694 __be32_to_cpu(vcl->conf.seqnum))
695 continue;
696 } else {
697 vcl = malloc(super->conf_rec_len*512 +
698 offsetof(struct vcl, conf));
699 vcl->next = super->conflist;
700 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
701 super->conflist = vcl;
702 dl->vlist[vnum++] = vcl;
703 }
704 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
705 vcl->lba_offset = (__u64*)
706 &vcl->conf.phys_refnum[super->mppe];
707
708 for (i=0; i < max_virt_disks ; i++)
709 if (memcmp(super->virt->entries[i].guid,
710 vcl->conf.guid, DDF_GUID_LEN)==0)
711 break;
712 if (i < max_virt_disks)
713 vcl->vcnum = i;
714 }
715 free(conf);
716
717 return 0;
718 }
719
720 #ifndef MDASSEMBLE
721 static int load_super_ddf_all(struct supertype *st, int fd,
722 void **sbp, char *devname, int keep_fd);
723 #endif
724 static int load_super_ddf(struct supertype *st, int fd,
725 char *devname)
726 {
727 unsigned long long dsize;
728 struct ddf_super *super;
729 int rv;
730
731 #ifndef MDASSEMBLE
732 /* if 'fd' is a container, load metadata from all the devices */
733 if (load_super_ddf_all(st, fd, &st->sb, devname, 1) == 0)
734 return 0;
735 #endif
736 if (st->subarray[0])
737 return 1; /* FIXME Is this correct */
738
739 if (get_dev_size(fd, devname, &dsize) == 0)
740 return 1;
741
742 /* 32M is a lower bound */
743 if (dsize <= 32*1024*1024) {
744 if (devname) {
745 fprintf(stderr,
746 Name ": %s is too small for ddf: "
747 "size is %llu sectors.\n",
748 devname, dsize>>9);
749 return 1;
750 }
751 }
752 if (dsize & 511) {
753 if (devname) {
754 fprintf(stderr,
755 Name ": %s is an odd size for ddf: "
756 "size is %llu bytes.\n",
757 devname, dsize);
758 return 1;
759 }
760 }
761
762 super = malloc(sizeof(*super));
763 if (!super) {
764 fprintf(stderr, Name ": malloc of %zu failed.\n",
765 sizeof(*super));
766 return 1;
767 }
768 memset(super, 0, sizeof(*super));
769
770 rv = load_ddf_headers(fd, super, devname);
771 if (rv) {
772 free(super);
773 return rv;
774 }
775
776 /* Have valid headers and have chosen the best. Let's read in the rest*/
777
778 rv = load_ddf_global(fd, super, devname);
779
780 if (rv) {
781 if (devname)
782 fprintf(stderr,
783 Name ": Failed to load all information "
784 "sections on %s\n", devname);
785 free(super);
786 return rv;
787 }
788
789 load_ddf_local(fd, super, devname, 0);
790
791 /* Should possibly check the sections .... */
792
793 st->sb = super;
794 if (st->ss == NULL) {
795 st->ss = &super_ddf;
796 st->minor_version = 0;
797 st->max_devs = 512;
798 }
799 return 0;
800
801 }
802
803 static void free_super_ddf(struct supertype *st)
804 {
805 struct ddf_super *ddf = st->sb;
806 if (ddf == NULL)
807 return;
808 free(ddf->phys);
809 free(ddf->virt);
810 while (ddf->conflist) {
811 struct vcl *v = ddf->conflist;
812 ddf->conflist = v->next;
813 if (v->block_sizes)
814 free(v->block_sizes);
815 free(v);
816 }
817 while (ddf->dlist) {
818 struct dl *d = ddf->dlist;
819 ddf->dlist = d->next;
820 if (d->fd >= 0)
821 close(d->fd);
822 if (d->spare)
823 free(d->spare);
824 free(d);
825 }
826 free(ddf);
827 st->sb = NULL;
828 }
829
830 static struct supertype *match_metadata_desc_ddf(char *arg)
831 {
832 /* 'ddf' only support containers */
833 struct supertype *st;
834 if (strcmp(arg, "ddf") != 0 &&
835 strcmp(arg, "default") != 0
836 )
837 return NULL;
838
839 st = malloc(sizeof(*st));
840 memset(st, 0, sizeof(*st));
841 st->ss = &super_ddf;
842 st->max_devs = 512;
843 st->minor_version = 0;
844 st->sb = NULL;
845 return st;
846 }
847
848 static struct supertype *match_metadata_desc_ddf_bvd(char *arg)
849 {
850 struct supertype *st;
851 if (strcmp(arg, "ddf/bvd") != 0 &&
852 strcmp(arg, "bvd") != 0 &&
853 strcmp(arg, "default") != 0
854 )
855 return NULL;
856
857 st = malloc(sizeof(*st));
858 memset(st, 0, sizeof(*st));
859 st->ss = &super_ddf_bvd;
860 st->max_devs = 512;
861 st->minor_version = 0;
862 st->sb = NULL;
863 return st;
864 }
865 static struct supertype *match_metadata_desc_ddf_svd(char *arg)
866 {
867 struct supertype *st;
868 if (strcmp(arg, "ddf/svd") != 0 &&
869 strcmp(arg, "svd") != 0 &&
870 strcmp(arg, "default") != 0
871 )
872 return NULL;
873
874 st = malloc(sizeof(*st));
875 memset(st, 0, sizeof(*st));
876 st->ss = &super_ddf_svd;
877 st->max_devs = 512;
878 st->minor_version = 0;
879 st->sb = NULL;
880 return st;
881 }
882
883 #ifndef MDASSEMBLE
884
885 static mapping_t ddf_state[] = {
886 { "Optimal", 0},
887 { "Degraded", 1},
888 { "Deleted", 2},
889 { "Missing", 3},
890 { "Failed", 4},
891 { "Partially Optimal", 5},
892 { "-reserved-", 6},
893 { "-reserved-", 7},
894 { NULL, 0}
895 };
896
897 static mapping_t ddf_init_state[] = {
898 { "Not Initialised", 0},
899 { "QuickInit in Progress", 1},
900 { "Fully Initialised", 2},
901 { "*UNKNOWN*", 3},
902 { NULL, 0}
903 };
904 static mapping_t ddf_access[] = {
905 { "Read/Write", 0},
906 { "Reserved", 1},
907 { "Read Only", 2},
908 { "Blocked (no access)", 3},
909 { NULL ,0}
910 };
911
912 static mapping_t ddf_level[] = {
913 { "RAID0", DDF_RAID0},
914 { "RAID1", DDF_RAID1},
915 { "RAID3", DDF_RAID3},
916 { "RAID4", DDF_RAID4},
917 { "RAID5", DDF_RAID5},
918 { "RAID1E",DDF_RAID1E},
919 { "JBOD", DDF_JBOD},
920 { "CONCAT",DDF_CONCAT},
921 { "RAID5E",DDF_RAID5E},
922 { "RAID5EE",DDF_RAID5EE},
923 { "RAID6", DDF_RAID6},
924 { NULL, 0}
925 };
926 static mapping_t ddf_sec_level[] = {
927 { "Striped", DDF_2STRIPED},
928 { "Mirrored", DDF_2MIRRORED},
929 { "Concat", DDF_2CONCAT},
930 { "Spanned", DDF_2SPANNED},
931 { NULL, 0}
932 };
933 #endif
934
935 struct num_mapping {
936 int num1, num2;
937 };
938 static struct num_mapping ddf_level_num[] = {
939 { DDF_RAID0, 0 },
940 { DDF_RAID1, 1 },
941 { DDF_RAID3, LEVEL_UNSUPPORTED },
942 { DDF_RAID4, 4 },
943 { DDF_RAID5, 5 },
944 { DDF_RAID1E, LEVEL_UNSUPPORTED },
945 { DDF_JBOD, LEVEL_UNSUPPORTED },
946 { DDF_CONCAT, LEVEL_LINEAR },
947 { DDF_RAID5E, LEVEL_UNSUPPORTED },
948 { DDF_RAID5EE, LEVEL_UNSUPPORTED },
949 { DDF_RAID6, 6},
950 { MAXINT, MAXINT }
951 };
952
953 static int map_num1(struct num_mapping *map, int num)
954 {
955 int i;
956 for (i=0 ; map[i].num1 != MAXINT; i++)
957 if (map[i].num1 == num)
958 break;
959 return map[i].num2;
960 }
961
962 #ifndef MDASSEMBLE
963 static void print_guid(char *guid, int tstamp)
964 {
965 /* A GUIDs are part (or all) ASCII and part binary.
966 * They tend to be space padded.
967 * We print the GUID in HEX, then in parentheses add
968 * any initial ASCII sequence, and a possible
969 * time stamp from bytes 16-19
970 */
971 int l = DDF_GUID_LEN;
972 int i;
973
974 for (i=0 ; i<DDF_GUID_LEN ; i++) {
975 if ((i&3)==0 && i != 0) printf(":");
976 printf("%02X", guid[i]&255);
977 }
978
979 printf(" (");
980 while (l && guid[l-1] == ' ')
981 l--;
982 for (i=0 ; i<l ; i++) {
983 if (guid[i] >= 0x20 && guid[i] < 0x7f)
984 fputc(guid[i], stdout);
985 else
986 break;
987 }
988 if (tstamp) {
989 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
990 char tbuf[100];
991 struct tm *tm;
992 tm = localtime(&then);
993 strftime(tbuf, 100, " %D %T",tm);
994 fputs(tbuf, stdout);
995 }
996 printf(")");
997 }
998
999 static void examine_vd(int n, struct ddf_super *sb, char *guid)
1000 {
1001 int crl = sb->conf_rec_len;
1002 struct vcl *vcl;
1003
1004 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
1005 struct vd_config *vc = &vcl->conf;
1006
1007 if (calc_crc(vc, crl*512) != vc->crc)
1008 continue;
1009 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
1010 continue;
1011
1012 /* Ok, we know about this VD, let's give more details */
1013 printf(" Raid Devices[%d] : %d\n", n,
1014 __be16_to_cpu(vc->prim_elmnt_count));
1015 printf(" Chunk Size[%d] : %d sectors\n", n,
1016 1 << vc->chunk_shift);
1017 printf(" Raid Level[%d] : %s\n", n,
1018 map_num(ddf_level, vc->prl)?:"-unknown-");
1019 if (vc->sec_elmnt_count != 1) {
1020 printf(" Secondary Position[%d] : %d of %d\n", n,
1021 vc->sec_elmnt_seq, vc->sec_elmnt_count);
1022 printf(" Secondary Level[%d] : %s\n", n,
1023 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1024 }
1025 printf(" Device Size[%d] : %llu\n", n,
1026 __be64_to_cpu(vc->blocks)/2);
1027 printf(" Array Size[%d] : %llu\n", n,
1028 __be64_to_cpu(vc->array_blocks)/2);
1029 }
1030 }
1031
1032 static void examine_vds(struct ddf_super *sb)
1033 {
1034 int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1035 int i;
1036 printf(" Virtual Disks : %d\n", cnt);
1037
1038 for (i=0; i<cnt; i++) {
1039 struct virtual_entry *ve = &sb->virt->entries[i];
1040 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1041 printf("\n");
1042 printf(" unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1043 printf(" state[%d] : %s, %s%s\n", i,
1044 map_num(ddf_state, ve->state & 7),
1045 (ve->state & 8) ? "Morphing, ": "",
1046 (ve->state & 16)? "Not Consistent" : "Consistent");
1047 printf(" init state[%d] : %s\n", i,
1048 map_num(ddf_init_state, ve->init_state&3));
1049 printf(" access[%d] : %s\n", i,
1050 map_num(ddf_access, (ve->init_state>>6) & 3));
1051 printf(" Name[%d] : %.16s\n", i, ve->name);
1052 examine_vd(i, sb, ve->guid);
1053 }
1054 if (cnt) printf("\n");
1055 }
1056
1057 static void examine_pds(struct ddf_super *sb)
1058 {
1059 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1060 int i;
1061 struct dl *dl;
1062 printf(" Physical Disks : %d\n", cnt);
1063
1064 for (i=0 ; i<cnt ; i++) {
1065 struct phys_disk_entry *pd = &sb->phys->entries[i];
1066 int type = __be16_to_cpu(pd->type);
1067 int state = __be16_to_cpu(pd->state);
1068
1069 printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1070 printf("\n");
1071 printf(" ref[%d] : %08x\n", i,
1072 __be32_to_cpu(pd->refnum));
1073 printf(" mode[%d] : %s%s%s%s%s\n", i,
1074 (type&2) ? "active":"",
1075 (type&4) ? "Global Spare":"",
1076 (type&8) ? "spare" : "",
1077 (type&16)? ", foreign" : "",
1078 (type&32)? "pass-through" : "");
1079 printf(" state[%d] : %s%s%s%s%s%s%s\n", i,
1080 (state&1)? "Online": "Offline",
1081 (state&2)? ", Failed": "",
1082 (state&4)? ", Rebuilding": "",
1083 (state&8)? ", in-transition": "",
1084 (state&16)? ", SMART errors": "",
1085 (state&32)? ", Unrecovered Read Errors": "",
1086 (state&64)? ", Missing" : "");
1087 printf(" Avail Size[%d] : %llu K\n", i,
1088 __be64_to_cpu(pd->config_size)>>1);
1089 for (dl = sb->dlist; dl ; dl = dl->next) {
1090 if (dl->disk.refnum == pd->refnum) {
1091 char *dv = map_dev(dl->major, dl->minor, 0);
1092 if (dv)
1093 printf(" Device[%d] : %s\n",
1094 i, dv);
1095 }
1096 }
1097 printf("\n");
1098 }
1099 }
1100
1101 static void examine_super_ddf(struct supertype *st, char *homehost)
1102 {
1103 struct ddf_super *sb = st->sb;
1104
1105 printf(" Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1106 printf(" Version : %.8s\n", sb->anchor.revision);
1107 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1108 printf("\n");
1109 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1110 printf("\n");
1111 printf(" Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1112 printf(" Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1113 ?"yes" : "no");
1114 examine_vds(sb);
1115 examine_pds(sb);
1116 }
1117
1118 static void brief_examine_super_ddf(struct supertype *st)
1119 {
1120 /* We just write a generic DDF ARRAY entry
1121 * The uuid is all hex, 6 groups of 4 bytes
1122 */
1123 struct ddf_super *ddf = st->sb;
1124 int i;
1125 printf("ARRAY /dev/ddf metadata=ddf UUID=");
1126 for (i = 0; i < DDF_GUID_LEN; i++) {
1127 if ((i&3) == 0 && i != 0)
1128 printf(":");
1129 printf("%02X", 255&ddf->anchor.guid[i]);
1130 }
1131 printf("\n");
1132 }
1133
1134 static void detail_super_ddf(struct supertype *st, char *homehost)
1135 {
1136 /* FIXME later
1137 * Could print DDF GUID
1138 * Need to find which array
1139 * If whole, briefly list all arrays
1140 * If one, give name
1141 */
1142 }
1143
1144 static void brief_detail_super_ddf(struct supertype *st)
1145 {
1146 /* FIXME I really need to know which array we are detailing.
1147 * Can that be stored in ddf_super??
1148 */
1149 // struct ddf_super *ddf = st->sb;
1150 }
1151 #endif
1152
1153 static int match_home_ddf(struct supertype *st, char *homehost)
1154 {
1155 /* It matches 'this' host if the controller is a
1156 * Linux-MD controller with vendor_data matching
1157 * the hostname
1158 */
1159 struct ddf_super *ddf = st->sb;
1160 int len = strlen(homehost);
1161
1162 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1163 len < sizeof(ddf->controller.vendor_data) &&
1164 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1165 ddf->controller.vendor_data[len] == 0);
1166 }
1167
1168 static struct vd_config *find_vdcr(struct ddf_super *ddf, int inst)
1169 {
1170 struct vcl *v;
1171
1172 for (v = ddf->conflist; v; v = v->next)
1173 if (inst == v->vcnum)
1174 return &v->conf;
1175 return NULL;
1176 }
1177
1178 static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1179 {
1180 /* Find the entry in phys_disk which has the given refnum
1181 * and return it's index
1182 */
1183 int i;
1184 for (i=0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1185 if (ddf->phys->entries[i].refnum == phys_refnum)
1186 return i;
1187 return -1;
1188 }
1189
1190 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1191 {
1192 /* The uuid returned here is used for:
1193 * uuid to put into bitmap file (Create, Grow)
1194 * uuid for backup header when saving critical section (Grow)
1195 * comparing uuids when re-adding a device into an array
1196 * For each of these we can make do with a truncated
1197 * or hashed uuid rather than the original, as long as
1198 * everyone agrees.
1199 * In each case the uuid required is that of the data-array,
1200 * not the device-set.
1201 * In the case of SVD we assume the BVD is of interest,
1202 * though that might be the case if a bitmap were made for
1203 * a mirrored SVD - worry about that later.
1204 * So we need to find the VD configuration record for the
1205 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1206 * The first 16 bytes of the sha1 of these is used.
1207 */
1208 struct ddf_super *ddf = st->sb;
1209 struct vd_config *vd = &ddf->newconf->conf;
1210
1211 if (!vd)
1212 memset(uuid, 0, sizeof (uuid));
1213 else {
1214 char buf[20];
1215 struct sha1_ctx ctx;
1216 sha1_init_ctx(&ctx);
1217 sha1_process_bytes(&vd->guid, DDF_GUID_LEN, &ctx);
1218 if (vd->sec_elmnt_count > 1)
1219 sha1_process_bytes(&vd->sec_elmnt_seq, 1, &ctx);
1220 sha1_finish_ctx(&ctx, buf);
1221 memcpy(uuid, buf, sizeof(uuid));
1222 }
1223 }
1224
1225 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info)
1226 {
1227 struct ddf_super *ddf = st->sb;
1228
1229 info->array.major_version = 1000;
1230 info->array.minor_version = 0; /* FIXME use ddf->revision somehow */
1231 info->array.patch_version = 0;
1232 info->array.raid_disks = __be16_to_cpu(ddf->phys->used_pdes);
1233 info->array.level = LEVEL_CONTAINER;
1234 info->array.layout = 0;
1235 info->array.md_minor = -1;
1236 info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)
1237 (ddf->anchor.guid+16));
1238 info->array.utime = 0;
1239 info->array.chunk_size = 0;
1240
1241 // info->data_offset = ???;
1242 // info->component_size = ???;
1243
1244 info->disk.major = 0;
1245 info->disk.minor = 0;
1246 if (ddf->dlist) {
1247 info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1248 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1249 } else {
1250 info->disk.number = -1;
1251 // info->disk.raid_disk = find refnum in the table and use index;
1252 }
1253 info->disk.state = (1 << MD_DISK_SYNC);
1254
1255 info->reshape_active = 0;
1256
1257 strcpy(info->text_version, "ddf");
1258
1259 // uuid_from_super_ddf(info->uuid, sbv);
1260
1261 // info->name[] ?? ;
1262 }
1263
1264 static void getinfo_super_n_container(struct supertype *st, struct mdinfo *info)
1265 {
1266 /* just need offset and size */
1267 struct ddf_super *ddf = st->sb;
1268 int n = info->disk.number;
1269
1270 info->data_offset = __be64_to_cpu(ddf->phys->entries[n].config_size);
1271 info->component_size = 32*1024*1024 / 512;
1272 }
1273
1274 static int rlq_to_layout(int rlq, int prl, int raiddisks);
1275
1276 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info)
1277 {
1278 struct ddf_super *ddf = st->sb;
1279 struct vd_config *vd = find_vdcr(ddf, info->container_member);
1280
1281 /* FIXME this returns BVD info - what if we want SVD ?? */
1282
1283 info->array.major_version = 1000;
1284 info->array.minor_version = 0; /* FIXME use ddf->revision somehow */
1285 info->array.patch_version = 0;
1286 info->array.raid_disks = __be16_to_cpu(vd->prim_elmnt_count);
1287 info->array.level = map_num1(ddf_level_num, vd->prl);
1288 info->array.layout = rlq_to_layout(vd->rlq, vd->prl,
1289 info->array.raid_disks);
1290 info->array.md_minor = -1;
1291 info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)(vd->guid+16));
1292 info->array.utime = DECADE + __be32_to_cpu(vd->timestamp);
1293 info->array.chunk_size = 512 << vd->chunk_shift;
1294
1295 // info->data_offset = ???;
1296 // info->component_size = ???;
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->resync_start = 0;
1305 if (!(ddf->virt->entries[info->container_member].state
1306 & DDF_state_inconsistent) &&
1307 (ddf->virt->entries[info->container_member].init_state
1308 & DDF_initstate_mask)
1309 == DDF_init_full)
1310 info->resync_start = ~0ULL;
1311
1312 uuid_from_super_ddf(st, info->uuid);
1313
1314 sprintf(info->text_version, "/%s/%d",
1315 devnum2devname(st->container_dev),
1316 info->container_member);
1317
1318 // info->name[] ?? ;
1319 }
1320
1321 static void getinfo_super_n_bvd(struct supertype *st, struct mdinfo *info)
1322 {
1323 /* Find the particular details for info->disk.raid_disk.
1324 * This includes data_offset, component_size,
1325 */
1326 struct ddf_super *ddf = st->sb;
1327 __u64 *lba_offset = ddf->newconf->lba_offset;
1328 struct vd_config *conf = &ddf->newconf->conf;
1329 info->data_offset = __be64_to_cpu(lba_offset[info->disk.raid_disk]);
1330 info->component_size = __be64_to_cpu(conf->blocks);
1331 }
1332
1333 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1334 char *update,
1335 char *devname, int verbose,
1336 int uuid_set, char *homehost)
1337 {
1338 /* For 'assemble' and 'force' we need to return non-zero if any
1339 * change was made. For others, the return value is ignored.
1340 * Update options are:
1341 * force-one : This device looks a bit old but needs to be included,
1342 * update age info appropriately.
1343 * assemble: clear any 'faulty' flag to allow this device to
1344 * be assembled.
1345 * force-array: Array is degraded but being forced, mark it clean
1346 * if that will be needed to assemble it.
1347 *
1348 * newdev: not used ????
1349 * grow: Array has gained a new device - this is currently for
1350 * linear only
1351 * resync: mark as dirty so a resync will happen.
1352 * uuid: Change the uuid of the array to match what is given
1353 * homehost: update the recorded homehost
1354 * name: update the name - preserving the homehost
1355 * _reshape_progress: record new reshape_progress position.
1356 *
1357 * Following are not relevant for this version:
1358 * sparc2.2 : update from old dodgey metadata
1359 * super-minor: change the preferred_minor number
1360 * summaries: update redundant counters.
1361 */
1362 int rv = 0;
1363 // struct ddf_super *ddf = st->sb;
1364 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1365 // struct virtual_entry *ve = find_ve(ddf);
1366
1367 /* we don't need to handle "force-*" or "assemble" as
1368 * there is no need to 'trick' the kernel. We the metadata is
1369 * first updated to activate the array, all the implied modifications
1370 * will just happen.
1371 */
1372
1373 if (strcmp(update, "grow") == 0) {
1374 /* FIXME */
1375 }
1376 if (strcmp(update, "resync") == 0) {
1377 // info->resync_checkpoint = 0;
1378 }
1379 /* We ignore UUID updates as they make even less sense
1380 * with DDF
1381 */
1382 if (strcmp(update, "homehost") == 0) {
1383 /* homehost is stored in controller->vendor_data,
1384 * or it is when we are the vendor
1385 */
1386 // if (info->vendor_is_local)
1387 // strcpy(ddf->controller.vendor_data, homehost);
1388 }
1389 if (strcmp(update, "name") == 0) {
1390 /* name is stored in virtual_entry->name */
1391 // memset(ve->name, ' ', 16);
1392 // strncpy(ve->name, info->name, 16);
1393 }
1394 if (strcmp(update, "_reshape_progress") == 0) {
1395 /* We don't support reshape yet */
1396 }
1397
1398 // update_all_csum(ddf);
1399
1400 return rv;
1401 }
1402
1403 static void make_header_guid(char *guid)
1404 {
1405 __u32 stamp;
1406 int rfd;
1407 /* Create a DDF Header of Virtual Disk GUID */
1408
1409 /* 24 bytes of fiction required.
1410 * first 8 are a 'vendor-id' - "Linux-MD"
1411 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1412 * Remaining 8 random number plus timestamp
1413 */
1414 memcpy(guid, T10, sizeof(T10));
1415 stamp = __cpu_to_be32(0xdeadbeef);
1416 memcpy(guid+8, &stamp, 4);
1417 stamp = __cpu_to_be32(0);
1418 memcpy(guid+12, &stamp, 4);
1419 stamp = __cpu_to_be32(time(0) - DECADE);
1420 memcpy(guid+16, &stamp, 4);
1421 rfd = open("/dev/urandom", O_RDONLY);
1422 if (rfd < 0 || read(rfd, &stamp, 4) != 4)
1423 stamp = random();
1424 memcpy(guid+20, &stamp, 4);
1425 if (rfd >= 0) close(rfd);
1426 }
1427
1428 static int init_super_ddf(struct supertype *st,
1429 mdu_array_info_t *info,
1430 unsigned long long size, char *name, char *homehost,
1431 int *uuid)
1432 {
1433 /* This is primarily called by Create when creating a new array.
1434 * We will then get add_to_super called for each component, and then
1435 * write_init_super called to write it out to each device.
1436 * For DDF, Create can create on fresh devices or on a pre-existing
1437 * array.
1438 * To create on a pre-existing array a different method will be called.
1439 * This one is just for fresh drives.
1440 *
1441 * We need to create the entire 'ddf' structure which includes:
1442 * DDF headers - these are easy.
1443 * Controller data - a Sector describing this controller .. not that
1444 * this is a controller exactly.
1445 * Physical Disk Record - one entry per device, so
1446 * leave plenty of space.
1447 * Virtual Disk Records - again, just leave plenty of space.
1448 * This just lists VDs, doesn't give details
1449 * Config records - describes the VDs that use this disk
1450 * DiskData - describes 'this' device.
1451 * BadBlockManagement - empty
1452 * Diag Space - empty
1453 * Vendor Logs - Could we put bitmaps here?
1454 *
1455 */
1456 struct ddf_super *ddf;
1457 char hostname[17];
1458 int hostlen;
1459 int max_phys_disks, max_virt_disks;
1460 unsigned long long sector;
1461 int clen;
1462 int i;
1463 int pdsize, vdsize;
1464 struct phys_disk *pd;
1465 struct virtual_disk *vd;
1466
1467 ddf = malloc(sizeof(*ddf));
1468 memset(ddf, 0, sizeof(*ddf));
1469 ddf->dlist = NULL; /* no physical disks yet */
1470 ddf->conflist = NULL; /* No virtual disks yet */
1471
1472 /* At least 32MB *must* be reserved for the ddf. So let's just
1473 * start 32MB from the end, and put the primary header there.
1474 * Don't do secondary for now.
1475 * We don't know exactly where that will be yet as it could be
1476 * different on each device. To just set up the lengths.
1477 *
1478 */
1479
1480 ddf->anchor.magic = DDF_HEADER_MAGIC;
1481 make_header_guid(ddf->anchor.guid);
1482
1483 memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
1484 ddf->anchor.seq = __cpu_to_be32(1);
1485 ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1486 ddf->anchor.openflag = 0xFF;
1487 ddf->anchor.foreignflag = 0;
1488 ddf->anchor.enforcegroups = 0; /* Is this best?? */
1489 ddf->anchor.pad0 = 0xff;
1490 memset(ddf->anchor.pad1, 0xff, 12);
1491 memset(ddf->anchor.header_ext, 0xff, 32);
1492 ddf->anchor.primary_lba = ~(__u64)0;
1493 ddf->anchor.secondary_lba = ~(__u64)0;
1494 ddf->anchor.type = DDF_HEADER_ANCHOR;
1495 memset(ddf->anchor.pad2, 0xff, 3);
1496 ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1497 ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1498 of 32M reserved.. */
1499 max_phys_disks = 1023; /* Should be enough */
1500 ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1501 max_virt_disks = 255;
1502 ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1503 ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1504 ddf->max_part = 64;
1505 ddf->mppe = 256;
1506 ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
1507 ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1508 ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe);
1509 memset(ddf->anchor.pad3, 0xff, 54);
1510 /* controller sections is one sector long immediately
1511 * after the ddf header */
1512 sector = 1;
1513 ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1514 ddf->anchor.controller_section_length = __cpu_to_be32(1);
1515 sector += 1;
1516
1517 /* phys is 8 sectors after that */
1518 pdsize = ROUND_UP(sizeof(struct phys_disk) +
1519 sizeof(struct phys_disk_entry)*max_phys_disks,
1520 512);
1521 switch(pdsize/512) {
1522 case 2: case 8: case 32: case 128: case 512: break;
1523 default: abort();
1524 }
1525 ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1526 ddf->anchor.phys_section_length =
1527 __cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1528 sector += pdsize/512;
1529
1530 /* virt is another 32 sectors */
1531 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1532 sizeof(struct virtual_entry) * max_virt_disks,
1533 512);
1534 switch(vdsize/512) {
1535 case 2: case 8: case 32: case 128: case 512: break;
1536 default: abort();
1537 }
1538 ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1539 ddf->anchor.virt_section_length =
1540 __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1541 sector += vdsize/512;
1542
1543 clen = ddf->conf_rec_len * (ddf->max_part+1);
1544 ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1545 ddf->anchor.config_section_length = __cpu_to_be32(clen);
1546 sector += clen;
1547
1548 ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1549 ddf->anchor.data_section_length = __cpu_to_be32(1);
1550 sector += 1;
1551
1552 ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1553 ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1554 ddf->anchor.diag_space_length = __cpu_to_be32(0);
1555 ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1556 ddf->anchor.vendor_length = __cpu_to_be32(0);
1557 ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1558
1559 memset(ddf->anchor.pad4, 0xff, 256);
1560
1561 memcpy(&ddf->primary, &ddf->anchor, 512);
1562 memcpy(&ddf->secondary, &ddf->anchor, 512);
1563
1564 ddf->primary.openflag = 1; /* I guess.. */
1565 ddf->primary.type = DDF_HEADER_PRIMARY;
1566
1567 ddf->secondary.openflag = 1; /* I guess.. */
1568 ddf->secondary.type = DDF_HEADER_SECONDARY;
1569
1570 ddf->active = &ddf->primary;
1571
1572 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1573
1574 /* 24 more bytes of fiction required.
1575 * first 8 are a 'vendor-id' - "Linux-MD"
1576 * Remaining 16 are serial number.... maybe a hostname would do?
1577 */
1578 memcpy(ddf->controller.guid, T10, sizeof(T10));
1579 gethostname(hostname, sizeof(hostname));
1580 hostname[sizeof(hostname) - 1] = 0;
1581 hostlen = strlen(hostname);
1582 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1583 for (i = strlen(T10) ; i+hostlen < 24; i++)
1584 ddf->controller.guid[i] = ' ';
1585
1586 ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1587 ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1588 ddf->controller.type.sub_vendor_id = 0;
1589 ddf->controller.type.sub_device_id = 0;
1590 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1591 memset(ddf->controller.pad, 0xff, 8);
1592 memset(ddf->controller.vendor_data, 0xff, 448);
1593
1594 pd = ddf->phys = malloc(pdsize);
1595 ddf->pdsize = pdsize;
1596
1597 memset(pd, 0xff, pdsize);
1598 memset(pd, 0, sizeof(*pd));
1599 pd->magic = DDF_PHYS_DATA_MAGIC;
1600 pd->used_pdes = __cpu_to_be16(0);
1601 pd->max_pdes = __cpu_to_be16(max_phys_disks);
1602 memset(pd->pad, 0xff, 52);
1603
1604 vd = ddf->virt = malloc(vdsize);
1605 ddf->vdsize = vdsize;
1606 memset(vd, 0, vdsize);
1607 vd->magic = DDF_VIRT_RECORDS_MAGIC;
1608 vd->populated_vdes = __cpu_to_be16(0);
1609 vd->max_vdes = __cpu_to_be16(max_virt_disks);
1610 memset(vd->pad, 0xff, 52);
1611
1612 for (i=0; i<max_virt_disks; i++)
1613 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1614
1615 st->sb = ddf;
1616 return 1;
1617 }
1618
1619 static int all_ff(char *guid)
1620 {
1621 int i;
1622 for (i = 0; i < DDF_GUID_LEN; i++)
1623 if (guid[i] != (char)0xff)
1624 return 0;
1625 return 1;
1626 }
1627 static int chunk_to_shift(int chunksize)
1628 {
1629 return ffs(chunksize/512)-1;
1630 }
1631
1632 static int level_to_prl(int level)
1633 {
1634 switch (level) {
1635 case LEVEL_LINEAR: return DDF_CONCAT;
1636 case 0: return DDF_RAID0;
1637 case 1: return DDF_RAID1;
1638 case 4: return DDF_RAID4;
1639 case 5: return DDF_RAID5;
1640 case 6: return DDF_RAID6;
1641 default: return -1;
1642 }
1643 }
1644 static int layout_to_rlq(int level, int layout, int raiddisks)
1645 {
1646 switch(level) {
1647 case 0:
1648 return DDF_RAID0_SIMPLE;
1649 case 1:
1650 switch(raiddisks) {
1651 case 2: return DDF_RAID1_SIMPLE;
1652 case 3: return DDF_RAID1_MULTI;
1653 default: return -1;
1654 }
1655 case 4:
1656 switch(layout) {
1657 case 0: return DDF_RAID4_N;
1658 }
1659 break;
1660 case 5:
1661 case 6:
1662 switch(layout) {
1663 case ALGORITHM_LEFT_ASYMMETRIC:
1664 return DDF_RAID5_N_RESTART;
1665 case ALGORITHM_RIGHT_ASYMMETRIC:
1666 if (level == 5)
1667 return DDF_RAID5_0_RESTART;
1668 else
1669 return DDF_RAID6_0_RESTART;
1670 case ALGORITHM_LEFT_SYMMETRIC:
1671 return DDF_RAID5_N_CONTINUE;
1672 case ALGORITHM_RIGHT_SYMMETRIC:
1673 return -1; /* not mentioned in standard */
1674 }
1675 }
1676 return -1;
1677 }
1678
1679 static int rlq_to_layout(int rlq, int prl, int raiddisks)
1680 {
1681 switch(prl) {
1682 case DDF_RAID0:
1683 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1684 case DDF_RAID1:
1685 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1686 on raiddisks*/
1687 case DDF_RAID4:
1688 switch(rlq) {
1689 case DDF_RAID4_N:
1690 return 0;
1691 default:
1692 /* not supported */
1693 return -1; /* FIXME this isn't checked */
1694 }
1695 case DDF_RAID5:
1696 switch(rlq) {
1697 case DDF_RAID5_N_RESTART:
1698 return ALGORITHM_LEFT_ASYMMETRIC;
1699 case DDF_RAID5_0_RESTART:
1700 return ALGORITHM_RIGHT_ASYMMETRIC;
1701 case DDF_RAID5_N_CONTINUE:
1702 return ALGORITHM_LEFT_SYMMETRIC;
1703 default:
1704 return -1;
1705 }
1706 case DDF_RAID6:
1707 switch(rlq) {
1708 case DDF_RAID5_N_RESTART:
1709 return ALGORITHM_LEFT_ASYMMETRIC;
1710 case DDF_RAID6_0_RESTART:
1711 return ALGORITHM_RIGHT_ASYMMETRIC;
1712 case DDF_RAID5_N_CONTINUE:
1713 return ALGORITHM_LEFT_SYMMETRIC;
1714 default:
1715 return -1;
1716 }
1717 }
1718 return -1;
1719 }
1720
1721 struct extent {
1722 unsigned long long start, size;
1723 };
1724 int cmp_extent(const void *av, const void *bv)
1725 {
1726 const struct extent *a = av;
1727 const struct extent *b = bv;
1728 if (a->start < b->start)
1729 return -1;
1730 if (a->start > b->start)
1731 return 1;
1732 return 0;
1733 }
1734
1735 struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
1736 {
1737 /* find a list of used extents on the give physical device
1738 * (dnum) of the given ddf.
1739 * Return a malloced array of 'struct extent'
1740
1741 FIXME ignore DDF_Legacy devices?
1742
1743 */
1744 struct extent *rv;
1745 int n = 0;
1746 int i, j;
1747
1748 rv = malloc(sizeof(struct extent) * (ddf->max_part + 2));
1749 if (!rv)
1750 return NULL;
1751
1752 for (i = 0; i < ddf->max_part; i++) {
1753 struct vcl *v = dl->vlist[i];
1754 if (v == NULL)
1755 continue;
1756 for (j=0; j < v->conf.prim_elmnt_count; j++)
1757 if (v->conf.phys_refnum[j] == dl->disk.refnum) {
1758 /* This device plays role 'j' in 'v'. */
1759 rv[n].start = __be64_to_cpu(v->lba_offset[j]);
1760 rv[n].size = __be64_to_cpu(v->conf.blocks);
1761 n++;
1762 break;
1763 }
1764 }
1765 qsort(rv, n, sizeof(*rv), cmp_extent);
1766
1767 rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
1768 rv[n].size = 0;
1769 return rv;
1770 }
1771
1772 static int init_super_ddf_bvd(struct supertype *st,
1773 mdu_array_info_t *info,
1774 unsigned long long size,
1775 char *name, char *homehost,
1776 int *uuid)
1777 {
1778 /* We are creating a BVD inside a pre-existing container.
1779 * so st->sb is already set.
1780 * We need to create a new vd_config and a new virtual_entry
1781 */
1782 struct ddf_super *ddf = st->sb;
1783 int venum;
1784 struct virtual_entry *ve;
1785 struct vcl *vcl;
1786 struct vd_config *vc;
1787
1788 if (__be16_to_cpu(ddf->virt->populated_vdes)
1789 >= __be16_to_cpu(ddf->virt->max_vdes)) {
1790 fprintf(stderr, Name": This ddf already has the "
1791 "maximum of %d virtual devices\n",
1792 __be16_to_cpu(ddf->virt->max_vdes));
1793 return 0;
1794 }
1795
1796 for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1797 if (all_ff(ddf->virt->entries[venum].guid))
1798 break;
1799 if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1800 fprintf(stderr, Name ": Cannot find spare slot for "
1801 "virtual disk - DDF is corrupt\n");
1802 return 0;
1803 }
1804 ve = &ddf->virt->entries[venum];
1805
1806 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1807 * timestamp, random number
1808 */
1809 make_header_guid(ve->guid);
1810 ve->unit = __cpu_to_be16(info->md_minor);
1811 ve->pad0 = 0xFFFF;
1812 ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1813 ve->type = 0;
1814 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
1815 if (info->state & 1) /* clean */
1816 ve->init_state = DDF_init_full;
1817 else
1818 ve->init_state = DDF_init_not;
1819
1820 memset(ve->pad1, 0xff, 14);
1821 memset(ve->name, ' ', 16);
1822 if (name)
1823 strncpy(ve->name, name, 16);
1824 ddf->virt->populated_vdes =
1825 __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
1826
1827 /* Now create a new vd_config */
1828 vcl = malloc(offsetof(struct vcl, conf) + ddf->conf_rec_len * 512);
1829 vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
1830 vcl->vcnum = venum;
1831 sprintf(st->subarray, "%d", venum);
1832 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1833
1834 vc = &vcl->conf;
1835
1836 vc->magic = DDF_VD_CONF_MAGIC;
1837 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
1838 vc->timestamp = __cpu_to_be32(time(0)-DECADE);
1839 vc->seqnum = __cpu_to_be32(1);
1840 memset(vc->pad0, 0xff, 24);
1841 vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
1842 vc->chunk_shift = chunk_to_shift(info->chunk_size);
1843 vc->prl = level_to_prl(info->level);
1844 vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
1845 vc->sec_elmnt_count = 1;
1846 vc->sec_elmnt_seq = 0;
1847 vc->srl = 0;
1848 vc->blocks = __cpu_to_be64(info->size * 2);
1849 vc->array_blocks = __cpu_to_be64(
1850 calc_array_size(info->level, info->raid_disks, info->layout,
1851 info->chunk_size, info->size*2));
1852 memset(vc->pad1, 0xff, 8);
1853 vc->spare_refs[0] = 0xffffffff;
1854 vc->spare_refs[1] = 0xffffffff;
1855 vc->spare_refs[2] = 0xffffffff;
1856 vc->spare_refs[3] = 0xffffffff;
1857 vc->spare_refs[4] = 0xffffffff;
1858 vc->spare_refs[5] = 0xffffffff;
1859 vc->spare_refs[6] = 0xffffffff;
1860 vc->spare_refs[7] = 0xffffffff;
1861 memset(vc->cache_pol, 0, 8);
1862 vc->bg_rate = 0x80;
1863 memset(vc->pad2, 0xff, 3);
1864 memset(vc->pad3, 0xff, 52);
1865 memset(vc->pad4, 0xff, 192);
1866 memset(vc->v0, 0xff, 32);
1867 memset(vc->v1, 0xff, 32);
1868 memset(vc->v2, 0xff, 16);
1869 memset(vc->v3, 0xff, 16);
1870 memset(vc->vendor, 0xff, 32);
1871
1872 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
1873 memset(vc->phys_refnum+(ddf->mppe * 4), 0x00, 8*ddf->mppe);
1874
1875 vcl->next = ddf->conflist;
1876 ddf->conflist = vcl;
1877 ddf->newconf = vcl;
1878 return 1;
1879 }
1880
1881 static void add_to_super_ddf_bvd(struct supertype *st,
1882 mdu_disk_info_t *dk, int fd, char *devname)
1883 {
1884 /* fd and devname identify a device with-in the ddf container (st).
1885 * dk identifies a location in the new BVD.
1886 * We need to find suitable free space in that device and update
1887 * the phys_refnum and lba_offset for the newly created vd_config.
1888 * We might also want to update the type in the phys_disk
1889 * section.
1890 */
1891 struct dl *dl;
1892 struct ddf_super *ddf = st->sb;
1893 struct vd_config *vc;
1894 __u64 *lba_offset;
1895 int working;
1896 int i;
1897 unsigned long long blocks, pos, esize;
1898 struct extent *ex;
1899
1900 for (dl = ddf->dlist; dl ; dl = dl->next)
1901 if (dl->major == dk->major &&
1902 dl->minor == dk->minor)
1903 break;
1904 if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
1905 return;
1906
1907 vc = &ddf->newconf->conf;
1908 lba_offset = ddf->newconf->lba_offset;
1909
1910 ex = get_extents(ddf, dl);
1911 if (!ex)
1912 return;
1913
1914 i = 0; pos = 0;
1915 blocks = __be64_to_cpu(vc->blocks);
1916 if (ddf->newconf->block_sizes)
1917 blocks = ddf->newconf->block_sizes[dk->raid_disk];
1918
1919 do {
1920 esize = ex[i].start - pos;
1921 if (esize >= blocks)
1922 break;
1923 pos = ex[i].start + ex[i].size;
1924 i++;
1925 } while (ex[i-1].size);
1926
1927 free(ex);
1928 if (esize < blocks)
1929 return;
1930
1931 vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
1932 lba_offset[dk->raid_disk] = __cpu_to_be64(pos);
1933
1934 for (i=0; i < ddf->max_part ; i++)
1935 if (dl->vlist[i] == NULL)
1936 break;
1937 if (i == ddf->max_part)
1938 return;
1939 dl->vlist[i] = ddf->newconf;
1940
1941 dl->fd = fd;
1942 dl->devname = devname;
1943
1944 /* Check how many working raid_disks, and if we can mark
1945 * array as optimal yet
1946 */
1947 working = 0;
1948
1949 for (i=0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
1950 if (vc->phys_refnum[i] != 0xffffffff)
1951 working++;
1952
1953 /* Find which virtual_entry */
1954 i = ddf->newconf->vcnum;
1955 if (working == __be16_to_cpu(vc->prim_elmnt_count))
1956 ddf->virt->entries[i].state =
1957 (ddf->virt->entries[i].state & ~DDF_state_mask)
1958 | DDF_state_optimal;
1959
1960 if (vc->prl == DDF_RAID6 &&
1961 working+1 == __be16_to_cpu(vc->prim_elmnt_count))
1962 ddf->virt->entries[i].state =
1963 (ddf->virt->entries[i].state & ~DDF_state_mask)
1964 | DDF_state_part_optimal;
1965
1966 ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
1967 ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
1968 }
1969
1970 /* add a device to a container, either while creating it or while
1971 * expanding a pre-existing container
1972 */
1973 static void add_to_super_ddf(struct supertype *st,
1974 mdu_disk_info_t *dk, int fd, char *devname)
1975 {
1976 struct ddf_super *ddf = st->sb;
1977 struct dl *dd;
1978 time_t now;
1979 struct tm *tm;
1980 unsigned long long size;
1981 struct phys_disk_entry *pde;
1982 int n, i;
1983 struct stat stb;
1984
1985 /* This is device numbered dk->number. We need to create
1986 * a phys_disk entry and a more detailed disk_data entry.
1987 */
1988 fstat(fd, &stb);
1989 dd = malloc(sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part);
1990 dd->major = major(stb.st_rdev);
1991 dd->minor = minor(stb.st_rdev);
1992 dd->devname = devname;
1993 dd->next = ddf->dlist;
1994 dd->fd = fd;
1995 dd->spare = NULL;
1996
1997 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
1998 now = time(0);
1999 tm = localtime(&now);
2000 sprintf(dd->disk.guid, "%8s%04d%02d%02d",
2001 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
2002 *(__u32*)(dd->disk.guid + 16) = random();
2003 *(__u32*)(dd->disk.guid + 20) = random();
2004
2005 do {
2006 /* Cannot be bothered finding a CRC of some irrelevant details*/
2007 dd->disk.refnum = random();
2008 for (i = __be16_to_cpu(ddf->active->max_pd_entries) - 1;
2009 i >= 0; i--)
2010 if (ddf->phys->entries[i].refnum == dd->disk.refnum)
2011 break;
2012 } while (i >= 0);
2013
2014 dd->disk.forced_ref = 1;
2015 dd->disk.forced_guid = 1;
2016 memset(dd->disk.vendor, ' ', 32);
2017 memcpy(dd->disk.vendor, "Linux", 5);
2018 memset(dd->disk.pad, 0xff, 442);
2019 for (i = 0; i < ddf->max_part ; i++)
2020 dd->vlist[i] = NULL;
2021
2022 n = __be16_to_cpu(ddf->phys->used_pdes);
2023 pde = &ddf->phys->entries[n];
2024 dd->pdnum = n;
2025
2026 n++;
2027 ddf->phys->used_pdes = __cpu_to_be16(n);
2028
2029 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2030 pde->refnum = dd->disk.refnum;
2031 pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2032 pde->state = __cpu_to_be16(DDF_Online);
2033 get_dev_size(fd, NULL, &size);
2034 /* We are required to reserve 32Meg, and record the size in sectors */
2035 pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
2036 sprintf(pde->path, "%17.17s","Information: nil") ;
2037 memset(pde->pad, 0xff, 6);
2038
2039 ddf->dlist = dd;
2040 }
2041
2042 /*
2043 * This is the write_init_super method for a ddf container. It is
2044 * called when creating a container or adding another device to a
2045 * container.
2046 */
2047
2048 #ifndef MDASSEMBLE
2049 static int __write_init_super_ddf(struct supertype *st, int do_close)
2050 {
2051
2052 struct ddf_super *ddf = st->sb;
2053 int i;
2054 struct dl *d;
2055 int n_config;
2056 int conf_size;
2057
2058 unsigned long long size, sector;
2059
2060 for (d = ddf->dlist; d; d=d->next) {
2061 int fd = d->fd;
2062
2063 if (fd < 0)
2064 continue;
2065
2066 /* We need to fill in the primary, (secondary) and workspace
2067 * lba's in the headers, set their checksums,
2068 * Also checksum phys, virt....
2069 *
2070 * Then write everything out, finally the anchor is written.
2071 */
2072 get_dev_size(fd, NULL, &size);
2073 size /= 512;
2074 ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
2075 ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
2076 ddf->anchor.seq = __cpu_to_be32(1);
2077 memcpy(&ddf->primary, &ddf->anchor, 512);
2078 memcpy(&ddf->secondary, &ddf->anchor, 512);
2079
2080 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
2081 ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
2082 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
2083
2084 ddf->primary.openflag = 0;
2085 ddf->primary.type = DDF_HEADER_PRIMARY;
2086
2087 ddf->secondary.openflag = 0;
2088 ddf->secondary.type = DDF_HEADER_SECONDARY;
2089
2090 ddf->primary.crc = calc_crc(&ddf->primary, 512);
2091 ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
2092
2093 sector = size - 16*1024*2;
2094 lseek64(fd, sector<<9, 0);
2095 write(fd, &ddf->primary, 512);
2096
2097 ddf->controller.crc = calc_crc(&ddf->controller, 512);
2098 write(fd, &ddf->controller, 512);
2099
2100 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
2101
2102 write(fd, ddf->phys, ddf->pdsize);
2103
2104 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
2105 write(fd, ddf->virt, ddf->vdsize);
2106
2107 /* Now write lots of config records. */
2108 n_config = ddf->max_part;
2109 conf_size = ddf->conf_rec_len * 512;
2110 for (i = 0 ; i <= n_config ; i++) {
2111 struct vcl *c = d->vlist[i];
2112 if (i == n_config)
2113 c = (struct vcl*)d->spare;
2114
2115 if (c) {
2116 c->conf.crc = calc_crc(&c->conf, conf_size);
2117 write(fd, &c->conf, conf_size);
2118 } else {
2119 __u32 sig = 0xffffffff;
2120 write(fd, &sig, 4);
2121 lseek64(fd, conf_size-4, SEEK_CUR);
2122 }
2123 }
2124 d->disk.crc = calc_crc(&d->disk, 512);
2125 write(fd, &d->disk, 512);
2126
2127 /* Maybe do the same for secondary */
2128
2129 lseek64(fd, (size-1)*512, SEEK_SET);
2130 write(fd, &ddf->anchor, 512);
2131 if (do_close) {
2132 close(fd);
2133 d->fd = -1;
2134 }
2135 }
2136 return 1;
2137 }
2138
2139 static int write_init_super_ddf(struct supertype *st)
2140 {
2141 return __write_init_super_ddf(st, 1);
2142 }
2143
2144 #endif
2145
2146 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize)
2147 {
2148 /* We must reserve the last 32Meg */
2149 if (devsize <= 32*1024*2)
2150 return 0;
2151 return devsize - 32*1024*2;
2152 }
2153
2154 #ifndef MDASSEMBLE
2155 int validate_geometry_ddf(struct supertype *st,
2156 int level, int layout, int raiddisks,
2157 int chunk, unsigned long long size,
2158 char *dev, unsigned long long *freesize)
2159 {
2160 int fd;
2161 struct mdinfo *sra;
2162 int cfd;
2163
2164 /* ddf potentially supports lots of things, but it depends on
2165 * what devices are offered (and maybe kernel version?)
2166 * If given unused devices, we will make a container.
2167 * If given devices in a container, we will make a BVD.
2168 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2169 */
2170
2171 if (level == LEVEL_CONTAINER) {
2172 st->ss = &super_ddf_container;
2173 if (dev) {
2174 int rv =st->ss->validate_geometry(st, level, layout,
2175 raiddisks, chunk,
2176 size,
2177 NULL, freesize);
2178 if (rv)
2179 return rv;
2180 }
2181 return st->ss->validate_geometry(st, level, layout, raiddisks,
2182 chunk, size, dev, freesize);
2183 }
2184
2185 if (st->sb) {
2186 /* creating in a given container */
2187 st->ss = &super_ddf_bvd;
2188 if (dev) {
2189 int rv =st->ss->validate_geometry(st, level, layout,
2190 raiddisks, chunk,
2191 size,
2192 NULL, freesize);
2193 if (rv)
2194 return rv;
2195 }
2196 return st->ss->validate_geometry(st, level, layout, raiddisks,
2197 chunk, size, dev, freesize);
2198 }
2199 /* FIXME should exclude MULTIPATH, or more appropriately, allow
2200 * only known levels.
2201 */
2202 if (!dev)
2203 return 1;
2204
2205 /* This device needs to be either a device in a 'ddf' container,
2206 * or it needs to be a 'ddf-bvd' array.
2207 */
2208
2209 fd = open(dev, O_RDONLY|O_EXCL, 0);
2210 if (fd >= 0) {
2211 sra = sysfs_read(fd, 0, GET_VERSION);
2212 close(fd);
2213 if (sra && sra->array.major_version == -1 &&
2214 strcmp(sra->text_version, "ddf-bvd") == 0) {
2215 st->ss = &super_ddf_svd;
2216 return st->ss->validate_geometry(st, level, layout,
2217 raiddisks, chunk, size,
2218 dev, freesize);
2219 }
2220
2221 fprintf(stderr,
2222 Name ": Cannot create this array on device %s\n",
2223 dev);
2224 return 0;
2225 }
2226 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2227 fprintf(stderr, Name ": Cannot open %s: %s\n",
2228 dev, strerror(errno));
2229 return 0;
2230 }
2231 /* Well, it is in use by someone, maybe a 'ddf' container. */
2232 cfd = open_container(fd);
2233 if (cfd < 0) {
2234 close(fd);
2235 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
2236 dev);
2237 return 0;
2238 }
2239 sra = sysfs_read(cfd, 0, GET_VERSION);
2240 close(fd);
2241 if (sra && sra->array.major_version == -1 &&
2242 strcmp(sra->text_version, "ddf") == 0) {
2243 /* This is a member of a ddf container. Load the container
2244 * and try to create a bvd
2245 */
2246 struct ddf_super *ddf;
2247 st->ss = &super_ddf_bvd;
2248 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL, 1) == 0) {
2249 st->sb = ddf;
2250 st->container_dev = fd2devnum(cfd);
2251 close(cfd);
2252 return st->ss->validate_geometry(st, level, layout,
2253 raiddisks, chunk, size,
2254 dev, freesize);
2255 }
2256 close(cfd);
2257 } else /* device may belong to a different container */
2258 return 0;
2259
2260 return 1;
2261 }
2262
2263 int validate_geometry_ddf_container(struct supertype *st,
2264 int level, int layout, int raiddisks,
2265 int chunk, unsigned long long size,
2266 char *dev, unsigned long long *freesize)
2267 {
2268 int fd;
2269 unsigned long long ldsize;
2270
2271 if (level != LEVEL_CONTAINER)
2272 return 0;
2273 if (!dev)
2274 return 1;
2275
2276 fd = open(dev, O_RDONLY|O_EXCL, 0);
2277 if (fd < 0) {
2278 fprintf(stderr, Name ": Cannot open %s: %s\n",
2279 dev, strerror(errno));
2280 return 0;
2281 }
2282 if (!get_dev_size(fd, dev, &ldsize)) {
2283 close(fd);
2284 return 0;
2285 }
2286 close(fd);
2287
2288 *freesize = avail_size_ddf(st, ldsize >> 9);
2289
2290 return 1;
2291 }
2292
2293 int validate_geometry_ddf_bvd(struct supertype *st,
2294 int level, int layout, int raiddisks,
2295 int chunk, unsigned long long size,
2296 char *dev, unsigned long long *freesize)
2297 {
2298 struct stat stb;
2299 struct ddf_super *ddf = st->sb;
2300 struct dl *dl;
2301 unsigned long long pos = 0;
2302 unsigned long long maxsize;
2303 struct extent *e;
2304 int i;
2305 /* ddf/bvd supports lots of things, but not containers */
2306 if (level == LEVEL_CONTAINER)
2307 return 0;
2308 /* We must have the container info already read in. */
2309 if (!ddf)
2310 return 0;
2311
2312 if (!dev) {
2313 /* General test: make sure there is space for
2314 * 'raiddisks' device extents of size 'size'.
2315 */
2316 unsigned long long minsize = size;
2317 int dcnt = 0;
2318 if (minsize == 0)
2319 minsize = 8;
2320 for (dl = ddf->dlist; dl ; dl = dl->next)
2321 {
2322 int found = 0;
2323 pos = 0;
2324
2325 i = 0;
2326 e = get_extents(ddf, dl);
2327 if (!e) continue;
2328 do {
2329 unsigned long long esize;
2330 esize = e[i].start - pos;
2331 if (esize >= minsize)
2332 found = 1;
2333 pos = e[i].start + e[i].size;
2334 i++;
2335 } while (e[i-1].size);
2336 if (found)
2337 dcnt++;
2338 free(e);
2339 }
2340 if (dcnt < raiddisks) {
2341 fprintf(stderr, Name ": Not enough devices with space "
2342 "for this array (%d < %d)\n",
2343 dcnt, raiddisks);
2344 return 0;
2345 }
2346 return 1;
2347 }
2348 /* This device must be a member of the set */
2349 if (stat(dev, &stb) < 0)
2350 return 0;
2351 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2352 return 0;
2353 for (dl = ddf->dlist ; dl ; dl = dl->next) {
2354 if (dl->major == major(stb.st_rdev) &&
2355 dl->minor == minor(stb.st_rdev))
2356 break;
2357 }
2358 if (!dl) {
2359 fprintf(stderr, Name ": %s is not in the same DDF set\n",
2360 dev);
2361 return 0;
2362 }
2363 e = get_extents(ddf, dl);
2364 maxsize = 0;
2365 i = 0;
2366 if (e) do {
2367 unsigned long long esize;
2368 esize = e[i].start - pos;
2369 if (esize >= maxsize)
2370 maxsize = esize;
2371 pos = e[i].start + e[i].size;
2372 i++;
2373 } while (e[i-1].size);
2374 *freesize = maxsize;
2375 // FIXME here I am
2376
2377 return 1;
2378 }
2379
2380 int validate_geometry_ddf_svd(struct supertype *st,
2381 int level, int layout, int raiddisks,
2382 int chunk, unsigned long long size,
2383 char *dev, unsigned long long *freesize)
2384 {
2385 /* dd/svd only supports striped, mirrored, concat, spanned... */
2386 if (level != LEVEL_LINEAR &&
2387 level != 0 &&
2388 level != 1)
2389 return 0;
2390 return 1;
2391 }
2392
2393 static int load_super_ddf_all(struct supertype *st, int fd,
2394 void **sbp, char *devname, int keep_fd)
2395 {
2396 struct mdinfo *sra;
2397 struct ddf_super *super;
2398 struct mdinfo *sd, *best = NULL;
2399 int bestseq = 0;
2400 int seq;
2401 char nm[20];
2402 int dfd;
2403
2404 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
2405 if (!sra)
2406 return 1;
2407 if (sra->array.major_version != -1 ||
2408 sra->array.minor_version != -2 ||
2409 strcmp(sra->text_version, "ddf") != 0)
2410 return 1;
2411
2412 super = malloc(sizeof(*super));
2413 if (!super)
2414 return 1;
2415 memset(super, 0, sizeof(*super));
2416
2417 /* first, try each device, and choose the best ddf */
2418 for (sd = sra->devs ; sd ; sd = sd->next) {
2419 int rv;
2420 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2421 dfd = dev_open(nm, O_RDONLY);
2422 if (dfd < 0)
2423 return 2;
2424 rv = load_ddf_headers(dfd, super, NULL);
2425 close(dfd);
2426 if (rv == 0) {
2427 seq = __be32_to_cpu(super->active->seq);
2428 if (super->active->openflag)
2429 seq--;
2430 if (!best || seq > bestseq) {
2431 bestseq = seq;
2432 best = sd;
2433 }
2434 }
2435 }
2436 if (!best)
2437 return 1;
2438 /* OK, load this ddf */
2439 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2440 dfd = dev_open(nm, O_RDONLY);
2441 if (dfd < 0)
2442 return 1;
2443 load_ddf_headers(dfd, super, NULL);
2444 load_ddf_global(dfd, super, NULL);
2445 close(dfd);
2446 /* Now we need the device-local bits */
2447 for (sd = sra->devs ; sd ; sd = sd->next) {
2448 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2449 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2450 if (dfd < 0)
2451 return 2;
2452 seq = load_ddf_local(dfd, super, NULL, keep_fd);
2453 if (!keep_fd) close(dfd);
2454 }
2455 if (st->subarray[0]) {
2456 struct vcl *v;
2457
2458 for (v = super->conflist; v; v = v->next)
2459 if (v->vcnum == atoi(st->subarray))
2460 super->newconf = v;
2461 if (!super->newconf)
2462 return 1;
2463 }
2464 *sbp = super;
2465 if (st->ss == NULL) {
2466 st->ss = &super_ddf_container;
2467 st->minor_version = 0;
2468 st->max_devs = 512;
2469 st->container_dev = fd2devnum(fd);
2470 }
2471 return 0;
2472 }
2473 #endif
2474
2475 static struct mdinfo *container_content_ddf(struct supertype *st)
2476 {
2477 /* Given a container loaded by load_super_ddf_all,
2478 * extract information about all the arrays into
2479 * an mdinfo tree.
2480 *
2481 * For each vcl in conflist: create an mdinfo, fill it in,
2482 * then look for matching devices (phys_refnum) in dlist
2483 * and create appropriate device mdinfo.
2484 */
2485 struct ddf_super *ddf = st->sb;
2486 struct mdinfo *rest = NULL;
2487 struct vcl *vc;
2488
2489 for (vc = ddf->conflist ; vc ; vc=vc->next)
2490 {
2491 int i;
2492 struct mdinfo *this;
2493 this = malloc(sizeof(*this));
2494 memset(this, 0, sizeof(*this));
2495 this->next = rest;
2496 rest = this;
2497
2498 this->array.major_version = 1000;
2499 this->array.minor_version = 0;
2500 this->array.patch_version = 0;
2501 this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2502 this->array.raid_disks =
2503 __be16_to_cpu(vc->conf.prim_elmnt_count);
2504 this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2505 this->array.raid_disks);
2506 this->array.md_minor = -1;
2507 this->array.ctime = DECADE +
2508 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
2509 this->array.utime = DECADE +
2510 __be32_to_cpu(vc->conf.timestamp);
2511 this->array.chunk_size = 512 << vc->conf.chunk_shift;
2512
2513 i = vc->vcnum;
2514 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2515 (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
2516 DDF_init_full) {
2517 this->array.state = 0;
2518 this->resync_start = 0;
2519 } else {
2520 this->array.state = 1;
2521 this->resync_start = ~0ULL;
2522 }
2523 memcpy(this->name, ddf->virt->entries[i].name, 32);
2524 this->name[33]=0;
2525
2526 memset(this->uuid, 0, sizeof(this->uuid));
2527 this->component_size = __be64_to_cpu(vc->conf.blocks);
2528 this->array.size = this->component_size / 2;
2529 this->container_member = i;
2530
2531 sprintf(this->text_version, "/%s/%d",
2532 devnum2devname(st->container_dev),
2533 this->container_member);
2534
2535 for (i=0 ; i < ddf->mppe ; i++) {
2536 struct mdinfo *dev;
2537 struct dl *d;
2538
2539 if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
2540 continue;
2541
2542 this->array.working_disks++;
2543
2544 for (d = ddf->dlist; d ; d=d->next)
2545 if (d->disk.refnum == vc->conf.phys_refnum[i])
2546 break;
2547 if (d == NULL)
2548 break;
2549
2550 dev = malloc(sizeof(*dev));
2551 memset(dev, 0, sizeof(*dev));
2552 dev->next = this->devs;
2553 this->devs = dev;
2554
2555 dev->disk.number = __be32_to_cpu(d->disk.refnum);
2556 dev->disk.major = d->major;
2557 dev->disk.minor = d->minor;
2558 dev->disk.raid_disk = i;
2559 dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2560
2561 dev->events = __le32_to_cpu(ddf->primary.seq);
2562 dev->data_offset = vc->lba_offset[i];
2563 dev->component_size = __be64_to_cpu(vc->conf.blocks);
2564 if (d->devname)
2565 strcpy(dev->name, d->devname);
2566 }
2567 }
2568 return rest;
2569 }
2570
2571 static int init_zero_ddf(struct supertype *st,
2572 mdu_array_info_t *info,
2573 unsigned long long size, char *name,
2574 char *homehost, int *uuid)
2575 {
2576 st->sb = NULL;
2577 return 0;
2578 }
2579
2580 static int store_zero_ddf(struct supertype *st, int fd)
2581 {
2582 unsigned long long dsize;
2583 char buf[512];
2584 memset(buf, 0, 512);
2585
2586 if (!get_dev_size(fd, NULL, &dsize))
2587 return 1;
2588
2589 lseek64(fd, dsize-512, 0);
2590 write(fd, buf, 512);
2591 return 0;
2592 }
2593
2594 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
2595 {
2596 /*
2597 * return:
2598 * 0 same, or first was empty, and second was copied
2599 * 1 second had wrong number
2600 * 2 wrong uuid
2601 * 3 wrong other info
2602 */
2603 struct ddf_super *first = st->sb;
2604 struct ddf_super *second = tst->sb;
2605
2606 if (!first) {
2607 st->sb = tst->sb;
2608 tst->sb = NULL;
2609 return 0;
2610 }
2611
2612 if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
2613 return 2;
2614
2615 /* FIXME should I look at anything else? */
2616 return 0;
2617 }
2618
2619 /*
2620 * A new array 'a' has been started which claims to be instance 'inst'
2621 * within container 'c'.
2622 * We need to confirm that the array matches the metadata in 'c' so
2623 * that we don't corrupt any metadata.
2624 */
2625 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
2626 {
2627 fprintf(stderr, "ddf: open_new %s\n", inst);
2628 a->info.container_member = atoi(inst);
2629 return 0;
2630 }
2631
2632 /*
2633 * The array 'a' is to be marked clean in the metadata.
2634 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2635 * clean up to the point (in sectors). If that cannot be recorded in the
2636 * metadata, then leave it as dirty.
2637 *
2638 * For DDF, we need to clear the DDF_state_inconsistent bit in the
2639 * !global! virtual_disk.virtual_entry structure.
2640 */
2641 static void ddf_set_array_state(struct active_array *a, int consistent)
2642 {
2643 struct ddf_super *ddf = a->container->sb;
2644 int inst = a->info.container_member;
2645 if (consistent)
2646 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
2647 else
2648 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
2649 ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
2650 if (a->resync_start == ~0ULL)
2651 ddf->virt->entries[inst].init_state |= DDF_init_full;
2652 else if (a->resync_start == 0)
2653 ddf->virt->entries[inst].init_state |= DDF_init_not;
2654 else
2655 ddf->virt->entries[inst].init_state |= DDF_init_quick;
2656
2657 printf("ddf mark %s %llu\n", consistent?"clean":"dirty",
2658 a->resync_start);
2659 }
2660
2661 /*
2662 * The state of each disk is stored in the global phys_disk structure
2663 * in phys_disk.entries[n].state.
2664 * This makes various combinations awkward.
2665 * - When a device fails in any array, it must be failed in all arrays
2666 * that include a part of this device.
2667 * - When a component is rebuilding, we cannot include it officially in the
2668 * array unless this is the only array that uses the device.
2669 *
2670 * So: when transitioning:
2671 * Online -> failed, just set failed flag. monitor will propagate
2672 * spare -> online, the device might need to be added to the array.
2673 * spare -> failed, just set failed. Don't worry if in array or not.
2674 */
2675 static void ddf_set_disk(struct active_array *a, int n, int state)
2676 {
2677 struct ddf_super *ddf = a->container->sb;
2678 int inst = a->info.container_member;
2679 struct vd_config *vc = find_vdcr(ddf, inst);
2680 int pd = find_phys(ddf, vc->phys_refnum[n]);
2681 int i, st, working;
2682
2683 if (vc == NULL) {
2684 fprintf(stderr, "ddf: cannot find instance %d!!\n", inst);
2685 return;
2686 }
2687 if (pd < 0) {
2688 /* disk doesn't currently exist. If it is now in_sync,
2689 * insert it. */
2690 if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
2691 /* Find dev 'n' in a->info->devs, determine the
2692 * ddf refnum, and set vc->phys_refnum and update
2693 * phys->entries[]
2694 */
2695 /* FIXME */
2696 }
2697 } else {
2698 if (state & DS_FAULTY)
2699 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Failed);
2700 if (state & DS_INSYNC) {
2701 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Online);
2702 ddf->phys->entries[pd].state &= __cpu_to_be16(~DDF_Rebuilding);
2703 }
2704 }
2705
2706 fprintf(stderr, "ddf: set_disk %d to %x\n", n, state);
2707
2708 /* Now we need to check the state of the array and update
2709 * virtual_disk.entries[n].state.
2710 * It needs to be one of "optimal", "degraded", "failed".
2711 * I don't understand 'deleted' or 'missing'.
2712 */
2713 working = 0;
2714 for (i=0; i < a->info.array.raid_disks; i++) {
2715 pd = find_phys(ddf, vc->phys_refnum[i]);
2716 if (pd < 0)
2717 continue;
2718 st = __be16_to_cpu(ddf->phys->entries[pd].state);
2719 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
2720 == DDF_Online)
2721 working++;
2722 }
2723 state = DDF_state_degraded;
2724 if (working == a->info.array.raid_disks)
2725 state = DDF_state_optimal;
2726 else switch(vc->prl) {
2727 case DDF_RAID0:
2728 case DDF_CONCAT:
2729 case DDF_JBOD:
2730 state = DDF_state_failed;
2731 break;
2732 case DDF_RAID1:
2733 if (working == 0)
2734 state = DDF_state_failed;
2735 break;
2736 case DDF_RAID4:
2737 case DDF_RAID5:
2738 if (working < a->info.array.raid_disks-1)
2739 state = DDF_state_failed;
2740 break;
2741 case DDF_RAID6:
2742 if (working < a->info.array.raid_disks-2)
2743 state = DDF_state_failed;
2744 else if (working == a->info.array.raid_disks-1)
2745 state = DDF_state_part_optimal;
2746 break;
2747 }
2748
2749 ddf->virt->entries[inst].state =
2750 (ddf->virt->entries[inst].state & ~DDF_state_mask)
2751 | state;
2752
2753 }
2754
2755 static void ddf_sync_metadata(struct supertype *st)
2756 {
2757
2758 /*
2759 * Write all data to all devices.
2760 * Later, we might be able to track whether only local changes
2761 * have been made, or whether any global data has been changed,
2762 * but ddf is sufficiently weird that it probably always
2763 * changes global data ....
2764 */
2765 __write_init_super_ddf(st, 0);
2766 fprintf(stderr, "ddf: sync_metadata\n");
2767 }
2768
2769 static void ddf_process_update(struct supertype *st,
2770 struct metadata_update *update)
2771 {
2772 /* Apply this update to the metadata.
2773 * The first 4 bytes are a DDF_*_MAGIC which guides
2774 * our actions.
2775 * Possible update are:
2776 * DDF_PHYS_RECORDS_MAGIC
2777 * Add a new physical device. Changes to this record
2778 * only happen implicitly.
2779 * used_pdes is the device number.
2780 * DDF_VIRT_RECORDS_MAGIC
2781 * Add a new VD. Possibly also change the 'access' bits.
2782 * populated_vdes is the entry number.
2783 * DDF_VD_CONF_MAGIC
2784 * New or updated VD. the VIRT_RECORD must already
2785 * exist. For an update, phys_refnum and lba_offset
2786 * (at least) are updated, and the VD_CONF must
2787 * be written to precisely those devices listed with
2788 * a phys_refnum.
2789 * DDF_SPARE_ASSIGN_MAGIC
2790 * replacement Spare Assignment Record... but for which device?
2791 *
2792 * So, e.g.:
2793 * - to create a new array, we send a VIRT_RECORD and
2794 * a VD_CONF. Then assemble and start the array.
2795 * - to activate a spare we send a VD_CONF to add the phys_refnum
2796 * and offset. This will also mark the spare as active with
2797 * a spare-assignment record.
2798 */
2799 struct ddf_super *ddf = st->sb;
2800 __u32 *magic = (__u32*)update->buf;
2801 struct phys_disk *pd;
2802 struct virtual_disk *vd;
2803 struct vd_config *vc;
2804 struct vcl *vcl;
2805 struct dl *dl;
2806 int mppe;
2807 int ent;
2808
2809 printf("Process update %x\n", *magic);
2810
2811 switch (*magic) {
2812 case DDF_PHYS_RECORDS_MAGIC:
2813
2814 if (update->len != (sizeof(struct phys_disk) +
2815 sizeof(struct phys_disk_entry)))
2816 return;
2817 pd = (struct phys_disk*)update->buf;
2818
2819 ent = __be16_to_cpu(pd->used_pdes);
2820 if (ent >= __be16_to_cpu(ddf->phys->max_pdes))
2821 return;
2822 if (!all_ff(ddf->phys->entries[ent].guid))
2823 return;
2824 ddf->phys->entries[ent] = pd->entries[0];
2825 ddf->phys->used_pdes = __cpu_to_be16(1 +
2826 __be16_to_cpu(ddf->phys->used_pdes));
2827 break;
2828
2829 case DDF_VIRT_RECORDS_MAGIC:
2830
2831 if (update->len != (sizeof(struct virtual_disk) +
2832 sizeof(struct virtual_entry)))
2833 return;
2834 vd = (struct virtual_disk*)update->buf;
2835
2836 ent = __be16_to_cpu(vd->populated_vdes);
2837 if (ent >= __be16_to_cpu(ddf->virt->max_vdes))
2838 return;
2839 if (!all_ff(ddf->virt->entries[ent].guid))
2840 return;
2841 ddf->virt->entries[ent] = vd->entries[0];
2842 ddf->virt->populated_vdes = __cpu_to_be16(1 +
2843 __be16_to_cpu(ddf->virt->populated_vdes));
2844 break;
2845
2846 case DDF_VD_CONF_MAGIC:
2847 printf("len %d %d\n", update->len, ddf->conf_rec_len);
2848
2849 mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries);
2850 if (update->len != ddf->conf_rec_len)
2851 return;
2852 vc = (struct vd_config*)update->buf;
2853 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2854 if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
2855 break;
2856 printf("vcl = %p\n", vcl);
2857 if (vcl) {
2858 /* An update, just copy the phys_refnum and lba_offset
2859 * fields
2860 */
2861 memcpy(vcl->conf.phys_refnum, vc->phys_refnum,
2862 mppe * (sizeof(__u32) + sizeof(__u64)));
2863 } else {
2864 /* A new VD_CONF */
2865 vcl = update->space;
2866 update->space = NULL;
2867 vcl->next = ddf->conflist;
2868 vcl->conf = *vc;
2869 vcl->lba_offset = (__u64*)
2870 &vcl->conf.phys_refnum[mppe];
2871 ddf->conflist = vcl;
2872 }
2873 /* Now make sure vlist is correct for each dl. */
2874 for (dl = ddf->dlist; dl; dl = dl->next) {
2875 int dn;
2876 int vn = 0;
2877 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2878 for (dn=0; dn < ddf->mppe ; dn++)
2879 if (vcl->conf.phys_refnum[dn] ==
2880 dl->disk.refnum) {
2881 printf("dev %d has %p at %d\n",
2882 dl->pdnum, vcl, vn);
2883 dl->vlist[vn++] = vcl;
2884 break;
2885 }
2886 while (vn < ddf->max_part)
2887 dl->vlist[vn++] = NULL;
2888 if (dl->vlist[0]) {
2889 ddf->phys->entries[dl->pdnum].type &=
2890 ~__cpu_to_be16(DDF_Global_Spare);
2891 ddf->phys->entries[dl->pdnum].type |=
2892 __cpu_to_be16(DDF_Active_in_VD);
2893 }
2894 if (dl->spare) {
2895 ddf->phys->entries[dl->pdnum].type &=
2896 ~__cpu_to_be16(DDF_Global_Spare);
2897 ddf->phys->entries[dl->pdnum].type |=
2898 __cpu_to_be16(DDF_Spare);
2899 }
2900 if (!dl->vlist[0] && !dl->spare) {
2901 ddf->phys->entries[dl->pdnum].type |=
2902 __cpu_to_be16(DDF_Global_Spare);
2903 ddf->phys->entries[dl->pdnum].type &=
2904 ~__cpu_to_be16(DDF_Spare |
2905 DDF_Active_in_VD);
2906 }
2907 }
2908 break;
2909 case DDF_SPARE_ASSIGN_MAGIC:
2910 default: break;
2911 }
2912 }
2913
2914 /*
2915 * Check if the array 'a' is degraded but not failed.
2916 * If it is, find as many spares as are available and needed and
2917 * arrange for their inclusion.
2918 * We only choose devices which are not already in the array,
2919 * and prefer those with a spare-assignment to this array.
2920 * otherwise we choose global spares - assuming always that
2921 * there is enough room.
2922 * For each spare that we assign, we return an 'mdinfo' which
2923 * describes the position for the device in the array.
2924 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
2925 * the new phys_refnum and lba_offset values.
2926 *
2927 * Only worry about BVDs at the moment.
2928 */
2929 static struct mdinfo *ddf_activate_spare(struct active_array *a,
2930 struct metadata_update **updates)
2931 {
2932 int working = 0;
2933 struct mdinfo *d;
2934 struct ddf_super *ddf = a->container->sb;
2935 int global_ok = 0;
2936 struct mdinfo *rv = NULL;
2937 struct mdinfo *di;
2938 struct metadata_update *mu;
2939 struct dl *dl;
2940 int i;
2941 struct vd_config *vc;
2942 __u64 *lba;
2943
2944 for (d = a->info.devs ; d ; d = d->next) {
2945 if ((d->curr_state & DS_FAULTY) &&
2946 d->state_fd >= 0)
2947 /* wait for Removal to happen */
2948 return NULL;
2949 if (d->state_fd >= 0)
2950 working ++;
2951 }
2952
2953 printf("ddf_activate: working=%d (%d) level=%d\n", working, a->info.array.raid_disks,
2954 a->info.array.level);
2955 if (working == a->info.array.raid_disks)
2956 return NULL; /* array not degraded */
2957 switch (a->info.array.level) {
2958 case 1:
2959 if (working == 0)
2960 return NULL; /* failed */
2961 break;
2962 case 4:
2963 case 5:
2964 if (working < a->info.array.raid_disks - 1)
2965 return NULL; /* failed */
2966 break;
2967 case 6:
2968 if (working < a->info.array.raid_disks - 2)
2969 return NULL; /* failed */
2970 break;
2971 default: /* concat or stripe */
2972 return NULL; /* failed */
2973 }
2974
2975 /* For each slot, if it is not working, find a spare */
2976 dl = ddf->dlist;
2977 for (i = 0; i < a->info.array.raid_disks; i++) {
2978 for (d = a->info.devs ; d ; d = d->next)
2979 if (d->disk.raid_disk == i)
2980 break;
2981 printf("found %d: %p %x\n", i, d, d?d->curr_state:0);
2982 if (d && (d->state_fd >= 0))
2983 continue;
2984
2985 /* OK, this device needs recovery. Find a spare */
2986 again:
2987 for ( ; dl ; dl = dl->next) {
2988 unsigned long long esize;
2989 unsigned long long pos;
2990 struct mdinfo *d2;
2991 int is_global = 0;
2992 int is_dedicated = 0;
2993 struct extent *ex;
2994 int j;
2995 /* If in this array, skip */
2996 for (d2 = a->info.devs ; d2 ; d2 = d2->next)
2997 if (d2->disk.major == dl->major &&
2998 d2->disk.minor == dl->minor) {
2999 printf("%x:%x already in array\n", dl->major, dl->minor);
3000 break;
3001 }
3002 if (d2)
3003 continue;
3004 if (ddf->phys->entries[dl->pdnum].type &
3005 __cpu_to_be16(DDF_Spare)) {
3006 /* Check spare assign record */
3007 if (dl->spare) {
3008 if (dl->spare->type & DDF_spare_dedicated) {
3009 /* check spare_ents for guid */
3010 for (j = 0 ;
3011 j < __be16_to_cpu(dl->spare->populated);
3012 j++) {
3013 if (memcmp(dl->spare->spare_ents[j].guid,
3014 ddf->virt->entries[a->info.container_member].guid,
3015 DDF_GUID_LEN) == 0)
3016 is_dedicated = 1;
3017 }
3018 } else
3019 is_global = 1;
3020 }
3021 } else if (ddf->phys->entries[dl->pdnum].type &
3022 __cpu_to_be16(DDF_Global_Spare)) {
3023 is_global = 1;
3024 }
3025 if ( ! (is_dedicated ||
3026 (is_global && global_ok))) {
3027 printf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
3028 is_dedicated, is_global);
3029 continue;
3030 }
3031
3032 /* We are allowed to use this device - is there space?
3033 * We need a->info.component_size sectors */
3034 ex = get_extents(ddf, dl);
3035 if (!ex) {
3036 printf("cannot get extents\n");
3037 continue;
3038 }
3039 j = 0; pos = 0;
3040 esize = 0;
3041
3042 do {
3043 esize = ex[j].start - pos;
3044 if (esize >= a->info.component_size)
3045 break;
3046 pos = ex[i].start + ex[i].size;
3047 i++;
3048 } while (ex[i-1].size);
3049
3050 free(ex);
3051 if (esize < a->info.component_size) {
3052 printf("%x:%x has no room: %llu %llu\n", dl->major, dl->minor,
3053 esize, a->info.component_size);
3054 /* No room */
3055 continue;
3056 }
3057
3058 /* Cool, we have a device with some space at pos */
3059 di = malloc(sizeof(*di));
3060 memset(di, 0, sizeof(*di));
3061 di->disk.number = i;
3062 di->disk.raid_disk = i;
3063 di->disk.major = dl->major;
3064 di->disk.minor = dl->minor;
3065 di->disk.state = 0;
3066 di->data_offset = pos;
3067 di->component_size = a->info.component_size;
3068 di->container_member = dl->pdnum;
3069 di->next = rv;
3070 rv = di;
3071 printf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3072 i, pos);
3073
3074 break;
3075 }
3076 if (!dl && ! global_ok) {
3077 /* not enough dedicated spares, try global */
3078 global_ok = 1;
3079 dl = ddf->dlist;
3080 goto again;
3081 }
3082 }
3083
3084 if (!rv)
3085 /* No spares found */
3086 return rv;
3087 /* Now 'rv' has a list of devices to return.
3088 * Create a metadata_update record to update the
3089 * phys_refnum and lba_offset values
3090 */
3091 mu = malloc(sizeof(*mu));
3092 mu->buf = malloc(ddf->conf_rec_len * 512);
3093 mu->space = malloc(sizeof(struct vcl));
3094 mu->len = ddf->conf_rec_len;
3095 mu->next = *updates;
3096 vc = find_vdcr(ddf, a->info.container_member);
3097 memcpy(mu->buf, vc, ddf->conf_rec_len * 512);
3098
3099 vc = (struct vd_config*)mu->buf;
3100 lba = (__u64*)&vc->phys_refnum[ddf->mppe];
3101 for (di = rv ; di ; di = di->next) {
3102 vc->phys_refnum[di->disk.raid_disk] =
3103 ddf->phys->entries[dl->pdnum].refnum;
3104 lba[di->disk.raid_disk] = di->data_offset;
3105 }
3106 *updates = mu;
3107 return rv;
3108 }
3109
3110 struct superswitch super_ddf = {
3111 #ifndef MDASSEMBLE
3112 .examine_super = examine_super_ddf,
3113 .brief_examine_super = brief_examine_super_ddf,
3114 .detail_super = detail_super_ddf,
3115 .brief_detail_super = brief_detail_super_ddf,
3116 .validate_geometry = validate_geometry_ddf,
3117 #endif
3118 .match_home = match_home_ddf,
3119 .uuid_from_super= uuid_from_super_ddf,
3120 .getinfo_super = getinfo_super_ddf,
3121 .update_super = update_super_ddf,
3122
3123 .avail_size = avail_size_ddf,
3124
3125 .compare_super = compare_super_ddf,
3126
3127 .load_super = load_super_ddf,
3128 .init_super = init_zero_ddf,
3129 .store_super = store_zero_ddf,
3130 .free_super = free_super_ddf,
3131 .match_metadata_desc = match_metadata_desc_ddf,
3132 .getinfo_super_n = getinfo_super_n_container,
3133
3134 .swapuuid = 0,
3135 .external = 1,
3136
3137 /* for mdmon */
3138 .open_new = ddf_open_new,
3139 .set_array_state= ddf_set_array_state,
3140 .set_disk = ddf_set_disk,
3141 .sync_metadata = ddf_sync_metadata,
3142 .process_update = ddf_process_update,
3143 .activate_spare = ddf_activate_spare,
3144
3145 };
3146
3147 /* Super_ddf_container is set by validate_geometry_ddf when given a
3148 * device that is not part of any array
3149 */
3150 static struct superswitch super_ddf_container = {
3151 #ifndef MDASSEMBLE
3152 .validate_geometry = validate_geometry_ddf_container,
3153 .write_init_super = write_init_super_ddf,
3154 #endif
3155
3156 .load_super = load_super_ddf,
3157 .init_super = init_super_ddf,
3158 .add_to_super = add_to_super_ddf,
3159 .getinfo_super = getinfo_super_ddf,
3160
3161 .free_super = free_super_ddf,
3162
3163 .container_content = container_content_ddf,
3164 .getinfo_super_n = getinfo_super_n_container,
3165
3166 .swapuuid = 0,
3167 .external = 1,
3168 };
3169
3170 static struct superswitch super_ddf_bvd = {
3171 #ifndef MDASSEMBLE
3172 // .detail_super = detail_super_ddf_bvd,
3173 // .brief_detail_super = brief_detail_super_ddf_bvd,
3174 .validate_geometry = validate_geometry_ddf_bvd,
3175 .write_init_super = write_init_super_ddf,
3176 #endif
3177 .update_super = update_super_ddf,
3178 .init_super = init_super_ddf_bvd,
3179 .add_to_super = add_to_super_ddf_bvd,
3180 .getinfo_super = getinfo_super_ddf_bvd,
3181 .getinfo_super_n = getinfo_super_n_bvd,
3182
3183 .load_super = load_super_ddf,
3184 .free_super = free_super_ddf,
3185 .match_metadata_desc = match_metadata_desc_ddf_bvd,
3186
3187 .swapuuid = 0,
3188 .external = 2,
3189 };
3190
3191 static struct superswitch super_ddf_svd = {
3192 #ifndef MDASSEMBLE
3193 // .detail_super = detail_super_ddf_svd,
3194 // .brief_detail_super = brief_detail_super_ddf_svd,
3195 .validate_geometry = validate_geometry_ddf_svd,
3196 #endif
3197 .update_super = update_super_ddf,
3198 .init_super = init_super_ddf,
3199
3200 .load_super = load_super_ddf,
3201 .free_super = free_super_ddf,
3202 .match_metadata_desc = match_metadata_desc_ddf_svd,
3203
3204 .swapuuid = 0,
3205 .external = 2,
3206 };
Unnamed repository; edit this file 'description' to name the repository.