]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super-ddf.c
projects / thirdparty / mdadm.git / blob
? search:


a5470062c995db98e5548f3a64e71312a949ba27
[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 int currentdev;
405 struct vcl {
406 struct vcl *next;
407 __u64 *lba_offset; /* location in 'conf' of
408 * the lba table */
409 int vcnum; /* index into ->virt */
410 __u64 *block_sizes; /* NULL if all the same */
411 struct vd_config conf;
412 } *conflist, *currentconf;
413 struct dl {
414 struct dl *next;
415 struct disk_data disk;
416 int major, minor;
417 char *devname;
418 int fd;
419 unsigned long long size; /* sectors */
420 int pdnum; /* index in ->phys */
421 struct spare_assign *spare;
422 struct vcl *vlist[0]; /* max_part in size */
423 } *dlist;
424 };
425
426 #ifndef offsetof
427 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
428 #endif
429
430 static struct superswitch super_ddf_container, super_ddf_bvd, super_ddf_svd;
431
432 static int calc_crc(void *buf, int len)
433 {
434 /* crcs are always at the same place as in the ddf_header */
435 struct ddf_header *ddf = buf;
436 __u32 oldcrc = ddf->crc;
437 __u32 newcrc;
438 ddf->crc = 0xffffffff;
439
440 newcrc = crc32(0, buf, len);
441 ddf->crc = oldcrc;
442 return newcrc;
443 }
444
445 static int load_ddf_header(int fd, unsigned long long lba,
446 unsigned long long size,
447 int type,
448 struct ddf_header *hdr, struct ddf_header *anchor)
449 {
450 /* read a ddf header (primary or secondary) from fd/lba
451 * and check that it is consistent with anchor
452 * Need to check:
453 * magic, crc, guid, rev, and LBA's header_type, and
454 * everything after header_type must be the same
455 */
456 if (lba >= size-1)
457 return 0;
458
459 if (lseek64(fd, lba<<9, 0) < 0)
460 return 0;
461
462 if (read(fd, hdr, 512) != 512)
463 return 0;
464
465 if (hdr->magic != DDF_HEADER_MAGIC)
466 return 0;
467 if (calc_crc(hdr, 512) != hdr->crc)
468 return 0;
469 if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
470 memcmp(anchor->revision, hdr->revision, 8) != 0 ||
471 anchor->primary_lba != hdr->primary_lba ||
472 anchor->secondary_lba != hdr->secondary_lba ||
473 hdr->type != type ||
474 memcmp(anchor->pad2, hdr->pad2, 512 -
475 offsetof(struct ddf_header, pad2)) != 0)
476 return 0;
477
478 /* Looks good enough to me... */
479 return 1;
480 }
481
482 static void *load_section(int fd, struct ddf_super *super, void *buf,
483 __u32 offset_be, __u32 len_be, int check)
484 {
485 unsigned long long offset = __be32_to_cpu(offset_be);
486 unsigned long long len = __be32_to_cpu(len_be);
487 int dofree = (buf == NULL);
488
489 if (check)
490 if (len != 2 && len != 8 && len != 32
491 && len != 128 && len != 512)
492 return NULL;
493
494 if (len > 1024)
495 return NULL;
496 if (buf) {
497 /* All pre-allocated sections are a single block */
498 if (len != 1)
499 return NULL;
500 } else
501 buf = malloc(len<<9);
502 if (!buf)
503 return NULL;
504
505 if (super->active->type == 1)
506 offset += __be64_to_cpu(super->active->primary_lba);
507 else
508 offset += __be64_to_cpu(super->active->secondary_lba);
509
510 if (lseek64(fd, offset<<9, 0) != (offset<<9)) {
511 if (dofree)
512 free(buf);
513 return NULL;
514 }
515 if (read(fd, buf, len<<9) != (len<<9)) {
516 if (dofree)
517 free(buf);
518 return NULL;
519 }
520 return buf;
521 }
522
523 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
524 {
525 unsigned long long dsize;
526
527 get_dev_size(fd, NULL, &dsize);
528
529 if (lseek64(fd, dsize-512, 0) < 0) {
530 if (devname)
531 fprintf(stderr,
532 Name": Cannot seek to anchor block on %s: %s\n",
533 devname, strerror(errno));
534 return 1;
535 }
536 if (read(fd, &super->anchor, 512) != 512) {
537 if (devname)
538 fprintf(stderr,
539 Name ": Cannot read anchor block on %s: %s\n",
540 devname, strerror(errno));
541 return 1;
542 }
543 if (super->anchor.magic != DDF_HEADER_MAGIC) {
544 if (devname)
545 fprintf(stderr, Name ": no DDF anchor found on %s\n",
546 devname);
547 return 2;
548 }
549 if (calc_crc(&super->anchor, 512) != super->anchor.crc) {
550 if (devname)
551 fprintf(stderr, Name ": bad CRC on anchor on %s\n",
552 devname);
553 return 2;
554 }
555 if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
556 memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
557 if (devname)
558 fprintf(stderr, Name ": can only support super revision"
559 " %.8s and earlier, not %.8s on %s\n",
560 DDF_REVISION_2, super->anchor.revision,devname);
561 return 2;
562 }
563 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba),
564 dsize >> 9, 1,
565 &super->primary, &super->anchor) == 0) {
566 if (devname)
567 fprintf(stderr,
568 Name ": Failed to load primary DDF header "
569 "on %s\n", devname);
570 return 2;
571 }
572 super->active = &super->primary;
573 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.secondary_lba),
574 dsize >> 9, 2,
575 &super->secondary, &super->anchor)) {
576 if ((__be32_to_cpu(super->primary.seq)
577 < __be32_to_cpu(super->secondary.seq) &&
578 !super->secondary.openflag)
579 || (__be32_to_cpu(super->primary.seq)
580 == __be32_to_cpu(super->secondary.seq) &&
581 super->primary.openflag && !super->secondary.openflag)
582 )
583 super->active = &super->secondary;
584 }
585 return 0;
586 }
587
588 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
589 {
590 void *ok;
591 ok = load_section(fd, super, &super->controller,
592 super->active->controller_section_offset,
593 super->active->controller_section_length,
594 0);
595 super->phys = load_section(fd, super, NULL,
596 super->active->phys_section_offset,
597 super->active->phys_section_length,
598 1);
599 super->pdsize = __be32_to_cpu(super->active->phys_section_length) * 512;
600
601 super->virt = load_section(fd, super, NULL,
602 super->active->virt_section_offset,
603 super->active->virt_section_length,
604 1);
605 super->vdsize = __be32_to_cpu(super->active->virt_section_length) * 512;
606 if (!ok ||
607 !super->phys ||
608 !super->virt) {
609 free(super->phys);
610 free(super->virt);
611 super->phys = NULL;
612 super->virt = NULL;
613 return 2;
614 }
615 super->conflist = NULL;
616 super->dlist = NULL;
617
618 super->max_part = __be16_to_cpu(super->active->max_partitions);
619 super->mppe = __be16_to_cpu(super->active->max_primary_element_entries);
620 super->conf_rec_len = __be16_to_cpu(super->active->config_record_len);
621 return 0;
622 }
623
624 static int load_ddf_local(int fd, struct ddf_super *super,
625 char *devname, int keep)
626 {
627 struct dl *dl;
628 struct stat stb;
629 char *conf;
630 int i;
631 int vnum;
632 int max_virt_disks = __be16_to_cpu(super->active->max_vd_entries);
633 unsigned long long dsize;
634
635 /* First the local disk info */
636 dl = malloc(sizeof(*dl) +
637 (super->max_part) * sizeof(dl->vlist[0]));
638
639 load_section(fd, super, &dl->disk,
640 super->active->data_section_offset,
641 super->active->data_section_length,
642 0);
643 dl->devname = devname ? strdup(devname) : NULL;
644
645 fstat(fd, &stb);
646 dl->major = major(stb.st_rdev);
647 dl->minor = minor(stb.st_rdev);
648 dl->next = super->dlist;
649 dl->fd = keep ? fd : -1;
650
651 dl->size = 0;
652 if (get_dev_size(fd, devname, &dsize))
653 dl->size = dsize >> 9;
654 dl->spare = NULL;
655 for (i=0 ; i < super->max_part ; i++)
656 dl->vlist[i] = NULL;
657 super->dlist = dl;
658 dl->pdnum = -1;
659 for (i=0; i < __be16_to_cpu(super->active->max_pd_entries); i++)
660 if (memcmp(super->phys->entries[i].guid,
661 dl->disk.guid, DDF_GUID_LEN) == 0)
662 dl->pdnum = i;
663
664 /* Now the config list. */
665 /* 'conf' is an array of config entries, some of which are
666 * probably invalid. Those which are good need to be copied into
667 * the conflist
668 */
669
670 conf = load_section(fd, super, NULL,
671 super->active->config_section_offset,
672 super->active->config_section_length,
673 0);
674
675 vnum = 0;
676 for (i = 0;
677 i < __be32_to_cpu(super->active->config_section_length);
678 i += super->conf_rec_len) {
679 struct vd_config *vd =
680 (struct vd_config *)((char*)conf + i*512);
681 struct vcl *vcl;
682
683 if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) {
684 if (dl->spare)
685 continue;
686 dl->spare = malloc(super->conf_rec_len*512);
687 memcpy(dl->spare, vd, super->conf_rec_len*512);
688 continue;
689 }
690 if (vd->magic != DDF_VD_CONF_MAGIC)
691 continue;
692 for (vcl = super->conflist; vcl; vcl = vcl->next) {
693 if (memcmp(vcl->conf.guid,
694 vd->guid, DDF_GUID_LEN) == 0)
695 break;
696 }
697
698 if (vcl) {
699 dl->vlist[vnum++] = vcl;
700 if (__be32_to_cpu(vd->seqnum) <=
701 __be32_to_cpu(vcl->conf.seqnum))
702 continue;
703 } else {
704 vcl = malloc(super->conf_rec_len*512 +
705 offsetof(struct vcl, conf));
706 vcl->next = super->conflist;
707 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
708 super->conflist = vcl;
709 dl->vlist[vnum++] = vcl;
710 }
711 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
712 vcl->lba_offset = (__u64*)
713 &vcl->conf.phys_refnum[super->mppe];
714
715 for (i=0; i < max_virt_disks ; i++)
716 if (memcmp(super->virt->entries[i].guid,
717 vcl->conf.guid, DDF_GUID_LEN)==0)
718 break;
719 if (i < max_virt_disks)
720 vcl->vcnum = i;
721 }
722 free(conf);
723
724 return 0;
725 }
726
727 #ifndef MDASSEMBLE
728 static int load_super_ddf_all(struct supertype *st, int fd,
729 void **sbp, char *devname, int keep_fd);
730 #endif
731 static int load_super_ddf(struct supertype *st, int fd,
732 char *devname)
733 {
734 unsigned long long dsize;
735 struct ddf_super *super;
736 int rv;
737
738 #ifndef MDASSEMBLE
739 /* if 'fd' is a container, load metadata from all the devices */
740 if (load_super_ddf_all(st, fd, &st->sb, devname, 1) == 0)
741 return 0;
742 #endif
743 if (st->subarray[0])
744 return 1; /* FIXME Is this correct */
745
746 if (get_dev_size(fd, devname, &dsize) == 0)
747 return 1;
748
749 /* 32M is a lower bound */
750 if (dsize <= 32*1024*1024) {
751 if (devname) {
752 fprintf(stderr,
753 Name ": %s is too small for ddf: "
754 "size is %llu sectors.\n",
755 devname, dsize>>9);
756 return 1;
757 }
758 }
759 if (dsize & 511) {
760 if (devname) {
761 fprintf(stderr,
762 Name ": %s is an odd size for ddf: "
763 "size is %llu bytes.\n",
764 devname, dsize);
765 return 1;
766 }
767 }
768
769 super = malloc(sizeof(*super));
770 if (!super) {
771 fprintf(stderr, Name ": malloc of %zu failed.\n",
772 sizeof(*super));
773 return 1;
774 }
775 memset(super, 0, sizeof(*super));
776
777 rv = load_ddf_headers(fd, super, devname);
778 if (rv) {
779 free(super);
780 return rv;
781 }
782
783 /* Have valid headers and have chosen the best. Let's read in the rest*/
784
785 rv = load_ddf_global(fd, super, devname);
786
787 if (rv) {
788 if (devname)
789 fprintf(stderr,
790 Name ": Failed to load all information "
791 "sections on %s\n", devname);
792 free(super);
793 return rv;
794 }
795
796 load_ddf_local(fd, super, devname, 0);
797
798 /* Should possibly check the sections .... */
799
800 st->sb = super;
801 if (st->ss == NULL) {
802 st->ss = &super_ddf;
803 st->minor_version = 0;
804 st->max_devs = 512;
805 }
806 return 0;
807
808 }
809
810 static void free_super_ddf(struct supertype *st)
811 {
812 struct ddf_super *ddf = st->sb;
813 if (ddf == NULL)
814 return;
815 free(ddf->phys);
816 free(ddf->virt);
817 while (ddf->conflist) {
818 struct vcl *v = ddf->conflist;
819 ddf->conflist = v->next;
820 if (v->block_sizes)
821 free(v->block_sizes);
822 free(v);
823 }
824 while (ddf->dlist) {
825 struct dl *d = ddf->dlist;
826 ddf->dlist = d->next;
827 if (d->fd >= 0)
828 close(d->fd);
829 if (d->spare)
830 free(d->spare);
831 free(d);
832 }
833 free(ddf);
834 st->sb = NULL;
835 }
836
837 static struct supertype *match_metadata_desc_ddf(char *arg)
838 {
839 /* 'ddf' only support containers */
840 struct supertype *st;
841 if (strcmp(arg, "ddf") != 0 &&
842 strcmp(arg, "default") != 0
843 )
844 return NULL;
845
846 st = malloc(sizeof(*st));
847 memset(st, 0, sizeof(*st));
848 st->ss = &super_ddf;
849 st->max_devs = 512;
850 st->minor_version = 0;
851 st->sb = NULL;
852 return st;
853 }
854
855 static struct supertype *match_metadata_desc_ddf_bvd(char *arg)
856 {
857 struct supertype *st;
858 if (strcmp(arg, "ddf/bvd") != 0 &&
859 strcmp(arg, "bvd") != 0 &&
860 strcmp(arg, "default") != 0
861 )
862 return NULL;
863
864 st = malloc(sizeof(*st));
865 memset(st, 0, sizeof(*st));
866 st->ss = &super_ddf_bvd;
867 st->max_devs = 512;
868 st->minor_version = 0;
869 st->sb = NULL;
870 return st;
871 }
872 static struct supertype *match_metadata_desc_ddf_svd(char *arg)
873 {
874 struct supertype *st;
875 if (strcmp(arg, "ddf/svd") != 0 &&
876 strcmp(arg, "svd") != 0 &&
877 strcmp(arg, "default") != 0
878 )
879 return NULL;
880
881 st = malloc(sizeof(*st));
882 memset(st, 0, sizeof(*st));
883 st->ss = &super_ddf_svd;
884 st->max_devs = 512;
885 st->minor_version = 0;
886 st->sb = NULL;
887 return st;
888 }
889
890 #ifndef MDASSEMBLE
891
892 static mapping_t ddf_state[] = {
893 { "Optimal", 0},
894 { "Degraded", 1},
895 { "Deleted", 2},
896 { "Missing", 3},
897 { "Failed", 4},
898 { "Partially Optimal", 5},
899 { "-reserved-", 6},
900 { "-reserved-", 7},
901 { NULL, 0}
902 };
903
904 static mapping_t ddf_init_state[] = {
905 { "Not Initialised", 0},
906 { "QuickInit in Progress", 1},
907 { "Fully Initialised", 2},
908 { "*UNKNOWN*", 3},
909 { NULL, 0}
910 };
911 static mapping_t ddf_access[] = {
912 { "Read/Write", 0},
913 { "Reserved", 1},
914 { "Read Only", 2},
915 { "Blocked (no access)", 3},
916 { NULL ,0}
917 };
918
919 static mapping_t ddf_level[] = {
920 { "RAID0", DDF_RAID0},
921 { "RAID1", DDF_RAID1},
922 { "RAID3", DDF_RAID3},
923 { "RAID4", DDF_RAID4},
924 { "RAID5", DDF_RAID5},
925 { "RAID1E",DDF_RAID1E},
926 { "JBOD", DDF_JBOD},
927 { "CONCAT",DDF_CONCAT},
928 { "RAID5E",DDF_RAID5E},
929 { "RAID5EE",DDF_RAID5EE},
930 { "RAID6", DDF_RAID6},
931 { NULL, 0}
932 };
933 static mapping_t ddf_sec_level[] = {
934 { "Striped", DDF_2STRIPED},
935 { "Mirrored", DDF_2MIRRORED},
936 { "Concat", DDF_2CONCAT},
937 { "Spanned", DDF_2SPANNED},
938 { NULL, 0}
939 };
940 #endif
941
942 struct num_mapping {
943 int num1, num2;
944 };
945 static struct num_mapping ddf_level_num[] = {
946 { DDF_RAID0, 0 },
947 { DDF_RAID1, 1 },
948 { DDF_RAID3, LEVEL_UNSUPPORTED },
949 { DDF_RAID4, 4 },
950 { DDF_RAID5, 5 },
951 { DDF_RAID1E, LEVEL_UNSUPPORTED },
952 { DDF_JBOD, LEVEL_UNSUPPORTED },
953 { DDF_CONCAT, LEVEL_LINEAR },
954 { DDF_RAID5E, LEVEL_UNSUPPORTED },
955 { DDF_RAID5EE, LEVEL_UNSUPPORTED },
956 { DDF_RAID6, 6},
957 { MAXINT, MAXINT }
958 };
959
960 static int map_num1(struct num_mapping *map, int num)
961 {
962 int i;
963 for (i=0 ; map[i].num1 != MAXINT; i++)
964 if (map[i].num1 == num)
965 break;
966 return map[i].num2;
967 }
968
969 #ifndef MDASSEMBLE
970 static void print_guid(char *guid, int tstamp)
971 {
972 /* A GUIDs are part (or all) ASCII and part binary.
973 * They tend to be space padded.
974 * We print the GUID in HEX, then in parentheses add
975 * any initial ASCII sequence, and a possible
976 * time stamp from bytes 16-19
977 */
978 int l = DDF_GUID_LEN;
979 int i;
980
981 for (i=0 ; i<DDF_GUID_LEN ; i++) {
982 if ((i&3)==0 && i != 0) printf(":");
983 printf("%02X", guid[i]&255);
984 }
985
986 printf(" (");
987 while (l && guid[l-1] == ' ')
988 l--;
989 for (i=0 ; i<l ; i++) {
990 if (guid[i] >= 0x20 && guid[i] < 0x7f)
991 fputc(guid[i], stdout);
992 else
993 break;
994 }
995 if (tstamp) {
996 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
997 char tbuf[100];
998 struct tm *tm;
999 tm = localtime(&then);
1000 strftime(tbuf, 100, " %D %T",tm);
1001 fputs(tbuf, stdout);
1002 }
1003 printf(")");
1004 }
1005
1006 static void examine_vd(int n, struct ddf_super *sb, char *guid)
1007 {
1008 int crl = sb->conf_rec_len;
1009 struct vcl *vcl;
1010
1011 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
1012 struct vd_config *vc = &vcl->conf;
1013
1014 if (calc_crc(vc, crl*512) != vc->crc)
1015 continue;
1016 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
1017 continue;
1018
1019 /* Ok, we know about this VD, let's give more details */
1020 printf(" Raid Devices[%d] : %d\n", n,
1021 __be16_to_cpu(vc->prim_elmnt_count));
1022 printf(" Chunk Size[%d] : %d sectors\n", n,
1023 1 << vc->chunk_shift);
1024 printf(" Raid Level[%d] : %s\n", n,
1025 map_num(ddf_level, vc->prl)?:"-unknown-");
1026 if (vc->sec_elmnt_count != 1) {
1027 printf(" Secondary Position[%d] : %d of %d\n", n,
1028 vc->sec_elmnt_seq, vc->sec_elmnt_count);
1029 printf(" Secondary Level[%d] : %s\n", n,
1030 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1031 }
1032 printf(" Device Size[%d] : %llu\n", n,
1033 __be64_to_cpu(vc->blocks)/2);
1034 printf(" Array Size[%d] : %llu\n", n,
1035 __be64_to_cpu(vc->array_blocks)/2);
1036 }
1037 }
1038
1039 static void examine_vds(struct ddf_super *sb)
1040 {
1041 int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1042 int i;
1043 printf(" Virtual Disks : %d\n", cnt);
1044
1045 for (i=0; i<cnt; i++) {
1046 struct virtual_entry *ve = &sb->virt->entries[i];
1047 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1048 printf("\n");
1049 printf(" unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1050 printf(" state[%d] : %s, %s%s\n", i,
1051 map_num(ddf_state, ve->state & 7),
1052 (ve->state & 8) ? "Morphing, ": "",
1053 (ve->state & 16)? "Not Consistent" : "Consistent");
1054 printf(" init state[%d] : %s\n", i,
1055 map_num(ddf_init_state, ve->init_state&3));
1056 printf(" access[%d] : %s\n", i,
1057 map_num(ddf_access, (ve->init_state>>6) & 3));
1058 printf(" Name[%d] : %.16s\n", i, ve->name);
1059 examine_vd(i, sb, ve->guid);
1060 }
1061 if (cnt) printf("\n");
1062 }
1063
1064 static void examine_pds(struct ddf_super *sb)
1065 {
1066 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1067 int i;
1068 struct dl *dl;
1069 printf(" Physical Disks : %d\n", cnt);
1070
1071 for (i=0 ; i<cnt ; i++) {
1072 struct phys_disk_entry *pd = &sb->phys->entries[i];
1073 int type = __be16_to_cpu(pd->type);
1074 int state = __be16_to_cpu(pd->state);
1075
1076 printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1077 printf("\n");
1078 printf(" ref[%d] : %08x\n", i,
1079 __be32_to_cpu(pd->refnum));
1080 printf(" mode[%d] : %s%s%s%s%s\n", i,
1081 (type&2) ? "active":"",
1082 (type&4) ? "Global Spare":"",
1083 (type&8) ? "spare" : "",
1084 (type&16)? ", foreign" : "",
1085 (type&32)? "pass-through" : "");
1086 printf(" state[%d] : %s%s%s%s%s%s%s\n", i,
1087 (state&1)? "Online": "Offline",
1088 (state&2)? ", Failed": "",
1089 (state&4)? ", Rebuilding": "",
1090 (state&8)? ", in-transition": "",
1091 (state&16)? ", SMART errors": "",
1092 (state&32)? ", Unrecovered Read Errors": "",
1093 (state&64)? ", Missing" : "");
1094 printf(" Avail Size[%d] : %llu K\n", i,
1095 __be64_to_cpu(pd->config_size)>>1);
1096 for (dl = sb->dlist; dl ; dl = dl->next) {
1097 if (dl->disk.refnum == pd->refnum) {
1098 char *dv = map_dev(dl->major, dl->minor, 0);
1099 if (dv)
1100 printf(" Device[%d] : %s\n",
1101 i, dv);
1102 }
1103 }
1104 printf("\n");
1105 }
1106 }
1107
1108 static void examine_super_ddf(struct supertype *st, char *homehost)
1109 {
1110 struct ddf_super *sb = st->sb;
1111
1112 printf(" Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1113 printf(" Version : %.8s\n", sb->anchor.revision);
1114 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1115 printf("\n");
1116 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1117 printf("\n");
1118 printf(" Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1119 printf(" Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1120 ?"yes" : "no");
1121 examine_vds(sb);
1122 examine_pds(sb);
1123 }
1124
1125 static void brief_examine_super_ddf(struct supertype *st)
1126 {
1127 /* We just write a generic DDF ARRAY entry
1128 * The uuid is all hex, 6 groups of 4 bytes
1129 */
1130 struct ddf_super *ddf = st->sb;
1131 int i;
1132 printf("ARRAY /dev/ddf metadata=ddf UUID=");
1133 for (i = 0; i < DDF_GUID_LEN; i++) {
1134 if ((i&3) == 0 && i != 0)
1135 printf(":");
1136 printf("%02X", 255&ddf->anchor.guid[i]);
1137 }
1138 printf("\n");
1139 }
1140
1141 static void detail_super_ddf(struct supertype *st, char *homehost)
1142 {
1143 /* FIXME later
1144 * Could print DDF GUID
1145 * Need to find which array
1146 * If whole, briefly list all arrays
1147 * If one, give name
1148 */
1149 }
1150
1151 static void brief_detail_super_ddf(struct supertype *st)
1152 {
1153 /* FIXME I really need to know which array we are detailing.
1154 * Can that be stored in ddf_super??
1155 */
1156 // struct ddf_super *ddf = st->sb;
1157 }
1158 #endif
1159
1160 static int match_home_ddf(struct supertype *st, char *homehost)
1161 {
1162 /* It matches 'this' host if the controller is a
1163 * Linux-MD controller with vendor_data matching
1164 * the hostname
1165 */
1166 struct ddf_super *ddf = st->sb;
1167 int len = strlen(homehost);
1168
1169 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1170 len < sizeof(ddf->controller.vendor_data) &&
1171 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1172 ddf->controller.vendor_data[len] == 0);
1173 }
1174
1175 static struct vd_config *find_vdcr(struct ddf_super *ddf, int inst)
1176 {
1177 struct vcl *v;
1178
1179 for (v = ddf->conflist; v; v = v->next)
1180 if (inst == v->vcnum)
1181 return &v->conf;
1182 return NULL;
1183 }
1184
1185 static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1186 {
1187 /* Find the entry in phys_disk which has the given refnum
1188 * and return it's index
1189 */
1190 int i;
1191 for (i=0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1192 if (ddf->phys->entries[i].refnum == phys_refnum)
1193 return i;
1194 return -1;
1195 }
1196
1197 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1198 {
1199 /* The uuid returned here is used for:
1200 * uuid to put into bitmap file (Create, Grow)
1201 * uuid for backup header when saving critical section (Grow)
1202 * comparing uuids when re-adding a device into an array
1203 * For each of these we can make do with a truncated
1204 * or hashed uuid rather than the original, as long as
1205 * everyone agrees.
1206 * In each case the uuid required is that of the data-array,
1207 * not the device-set.
1208 * In the case of SVD we assume the BVD is of interest,
1209 * though that might be the case if a bitmap were made for
1210 * a mirrored SVD - worry about that later.
1211 * So we need to find the VD configuration record for the
1212 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1213 * The first 16 bytes of the sha1 of these is used.
1214 */
1215 struct ddf_super *ddf = st->sb;
1216 struct vcl *vcl = ddf->currentconf;
1217
1218 if (!vcl)
1219 memset(uuid, 0, sizeof (uuid));
1220 else {
1221 char buf[20];
1222 struct sha1_ctx ctx;
1223 sha1_init_ctx(&ctx);
1224 sha1_process_bytes(&vcl->conf.guid, DDF_GUID_LEN, &ctx);
1225 if (vcl->conf.sec_elmnt_count > 1)
1226 sha1_process_bytes(&vcl->conf.sec_elmnt_seq, 1, &ctx);
1227 sha1_finish_ctx(&ctx, buf);
1228 memcpy(uuid, buf, sizeof(uuid));
1229 }
1230 }
1231
1232 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info)
1233 {
1234 struct ddf_super *ddf = st->sb;
1235
1236 info->array.raid_disks = __be16_to_cpu(ddf->phys->used_pdes);
1237 info->array.level = LEVEL_CONTAINER;
1238 info->array.layout = 0;
1239 info->array.md_minor = -1;
1240 info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)
1241 (ddf->anchor.guid+16));
1242 info->array.utime = 0;
1243 info->array.chunk_size = 0;
1244
1245
1246 info->disk.major = 0;
1247 info->disk.minor = 0;
1248 if (ddf->dlist) {
1249 info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1250 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1251
1252 info->data_offset = __be64_to_cpu(ddf->phys->
1253 entries[info->disk.raid_disk].
1254 config_size);
1255 info->component_size = ddf->dlist->size - info->data_offset;
1256 } else {
1257 info->disk.number = -1;
1258 // info->disk.raid_disk = find refnum in the table and use index;
1259 }
1260 info->disk.state = (1 << MD_DISK_SYNC);
1261
1262
1263 info->reshape_active = 0;
1264
1265 strcpy(info->text_version, "ddf");
1266
1267 // uuid_from_super_ddf(info->uuid, sbv);
1268
1269 // info->name[] ?? ;
1270 }
1271
1272 static int rlq_to_layout(int rlq, int prl, int raiddisks);
1273
1274 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info)
1275 {
1276 struct ddf_super *ddf = st->sb;
1277 struct vcl *vc = ddf->currentconf;
1278 int cd = ddf->currentdev;
1279
1280 /* FIXME this returns BVD info - what if we want SVD ?? */
1281
1282 info->array.raid_disks = __be16_to_cpu(vc->conf.prim_elmnt_count);
1283 info->array.level = map_num1(ddf_level_num, vc->conf.prl);
1284 info->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
1285 info->array.raid_disks);
1286 info->array.md_minor = -1;
1287 info->array.ctime = DECADE +
1288 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
1289 info->array.utime = DECADE + __be32_to_cpu(vc->conf.timestamp);
1290 info->array.chunk_size = 512 << vc->conf.chunk_shift;
1291
1292 if (cd >= 0 && cd < ddf->mppe) {
1293 info->data_offset = __be64_to_cpu(vc->lba_offset[cd]);
1294 if (vc->block_sizes)
1295 info->component_size = vc->block_sizes[cd];
1296 else
1297 info->component_size = __be64_to_cpu(vc->conf.blocks);
1298 }
1299
1300 info->disk.major = 0;
1301 info->disk.minor = 0;
1302 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1303 // info->disk.raid_disk = find refnum in the table and use index;
1304 // info->disk.state = ???;
1305
1306 info->resync_start = 0;
1307 if (!(ddf->virt->entries[info->container_member].state
1308 & DDF_state_inconsistent) &&
1309 (ddf->virt->entries[info->container_member].init_state
1310 & DDF_initstate_mask)
1311 == DDF_init_full)
1312 info->resync_start = ~0ULL;
1313
1314 uuid_from_super_ddf(st, info->uuid);
1315
1316 sprintf(info->text_version, "/%s/%d",
1317 devnum2devname(st->container_dev),
1318 info->container_member);
1319
1320 // info->name[] ?? ;
1321 }
1322
1323
1324 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1325 char *update,
1326 char *devname, int verbose,
1327 int uuid_set, char *homehost)
1328 {
1329 /* For 'assemble' and 'force' we need to return non-zero if any
1330 * change was made. For others, the return value is ignored.
1331 * Update options are:
1332 * force-one : This device looks a bit old but needs to be included,
1333 * update age info appropriately.
1334 * assemble: clear any 'faulty' flag to allow this device to
1335 * be assembled.
1336 * force-array: Array is degraded but being forced, mark it clean
1337 * if that will be needed to assemble it.
1338 *
1339 * newdev: not used ????
1340 * grow: Array has gained a new device - this is currently for
1341 * linear only
1342 * resync: mark as dirty so a resync will happen.
1343 * uuid: Change the uuid of the array to match what is given
1344 * homehost: update the recorded homehost
1345 * name: update the name - preserving the homehost
1346 * _reshape_progress: record new reshape_progress position.
1347 *
1348 * Following are not relevant for this version:
1349 * sparc2.2 : update from old dodgey metadata
1350 * super-minor: change the preferred_minor number
1351 * summaries: update redundant counters.
1352 */
1353 int rv = 0;
1354 // struct ddf_super *ddf = st->sb;
1355 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1356 // struct virtual_entry *ve = find_ve(ddf);
1357
1358 /* we don't need to handle "force-*" or "assemble" as
1359 * there is no need to 'trick' the kernel. We the metadata is
1360 * first updated to activate the array, all the implied modifications
1361 * will just happen.
1362 */
1363
1364 if (strcmp(update, "grow") == 0) {
1365 /* FIXME */
1366 }
1367 if (strcmp(update, "resync") == 0) {
1368 // info->resync_checkpoint = 0;
1369 }
1370 /* We ignore UUID updates as they make even less sense
1371 * with DDF
1372 */
1373 if (strcmp(update, "homehost") == 0) {
1374 /* homehost is stored in controller->vendor_data,
1375 * or it is when we are the vendor
1376 */
1377 // if (info->vendor_is_local)
1378 // strcpy(ddf->controller.vendor_data, homehost);
1379 }
1380 if (strcmp(update, "name") == 0) {
1381 /* name is stored in virtual_entry->name */
1382 // memset(ve->name, ' ', 16);
1383 // strncpy(ve->name, info->name, 16);
1384 }
1385 if (strcmp(update, "_reshape_progress") == 0) {
1386 /* We don't support reshape yet */
1387 }
1388
1389 // update_all_csum(ddf);
1390
1391 return rv;
1392 }
1393
1394 static void make_header_guid(char *guid)
1395 {
1396 __u32 stamp;
1397 int rfd;
1398 /* Create a DDF Header of Virtual Disk GUID */
1399
1400 /* 24 bytes of fiction required.
1401 * first 8 are a 'vendor-id' - "Linux-MD"
1402 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1403 * Remaining 8 random number plus timestamp
1404 */
1405 memcpy(guid, T10, sizeof(T10));
1406 stamp = __cpu_to_be32(0xdeadbeef);
1407 memcpy(guid+8, &stamp, 4);
1408 stamp = __cpu_to_be32(0);
1409 memcpy(guid+12, &stamp, 4);
1410 stamp = __cpu_to_be32(time(0) - DECADE);
1411 memcpy(guid+16, &stamp, 4);
1412 rfd = open("/dev/urandom", O_RDONLY);
1413 if (rfd < 0 || read(rfd, &stamp, 4) != 4)
1414 stamp = random();
1415 memcpy(guid+20, &stamp, 4);
1416 if (rfd >= 0) close(rfd);
1417 }
1418
1419 static int init_super_ddf(struct supertype *st,
1420 mdu_array_info_t *info,
1421 unsigned long long size, char *name, char *homehost,
1422 int *uuid)
1423 {
1424 /* This is primarily called by Create when creating a new array.
1425 * We will then get add_to_super called for each component, and then
1426 * write_init_super called to write it out to each device.
1427 * For DDF, Create can create on fresh devices or on a pre-existing
1428 * array.
1429 * To create on a pre-existing array a different method will be called.
1430 * This one is just for fresh drives.
1431 *
1432 * We need to create the entire 'ddf' structure which includes:
1433 * DDF headers - these are easy.
1434 * Controller data - a Sector describing this controller .. not that
1435 * this is a controller exactly.
1436 * Physical Disk Record - one entry per device, so
1437 * leave plenty of space.
1438 * Virtual Disk Records - again, just leave plenty of space.
1439 * This just lists VDs, doesn't give details
1440 * Config records - describes the VDs that use this disk
1441 * DiskData - describes 'this' device.
1442 * BadBlockManagement - empty
1443 * Diag Space - empty
1444 * Vendor Logs - Could we put bitmaps here?
1445 *
1446 */
1447 struct ddf_super *ddf;
1448 char hostname[17];
1449 int hostlen;
1450 int max_phys_disks, max_virt_disks;
1451 unsigned long long sector;
1452 int clen;
1453 int i;
1454 int pdsize, vdsize;
1455 struct phys_disk *pd;
1456 struct virtual_disk *vd;
1457
1458 ddf = malloc(sizeof(*ddf));
1459 memset(ddf, 0, sizeof(*ddf));
1460 ddf->dlist = NULL; /* no physical disks yet */
1461 ddf->conflist = NULL; /* No virtual disks yet */
1462
1463 /* At least 32MB *must* be reserved for the ddf. So let's just
1464 * start 32MB from the end, and put the primary header there.
1465 * Don't do secondary for now.
1466 * We don't know exactly where that will be yet as it could be
1467 * different on each device. To just set up the lengths.
1468 *
1469 */
1470
1471 ddf->anchor.magic = DDF_HEADER_MAGIC;
1472 make_header_guid(ddf->anchor.guid);
1473
1474 memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
1475 ddf->anchor.seq = __cpu_to_be32(1);
1476 ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1477 ddf->anchor.openflag = 0xFF;
1478 ddf->anchor.foreignflag = 0;
1479 ddf->anchor.enforcegroups = 0; /* Is this best?? */
1480 ddf->anchor.pad0 = 0xff;
1481 memset(ddf->anchor.pad1, 0xff, 12);
1482 memset(ddf->anchor.header_ext, 0xff, 32);
1483 ddf->anchor.primary_lba = ~(__u64)0;
1484 ddf->anchor.secondary_lba = ~(__u64)0;
1485 ddf->anchor.type = DDF_HEADER_ANCHOR;
1486 memset(ddf->anchor.pad2, 0xff, 3);
1487 ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1488 ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1489 of 32M reserved.. */
1490 max_phys_disks = 1023; /* Should be enough */
1491 ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1492 max_virt_disks = 255;
1493 ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1494 ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1495 ddf->max_part = 64;
1496 ddf->mppe = 256;
1497 ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
1498 ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1499 ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe);
1500 memset(ddf->anchor.pad3, 0xff, 54);
1501 /* controller sections is one sector long immediately
1502 * after the ddf header */
1503 sector = 1;
1504 ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1505 ddf->anchor.controller_section_length = __cpu_to_be32(1);
1506 sector += 1;
1507
1508 /* phys is 8 sectors after that */
1509 pdsize = ROUND_UP(sizeof(struct phys_disk) +
1510 sizeof(struct phys_disk_entry)*max_phys_disks,
1511 512);
1512 switch(pdsize/512) {
1513 case 2: case 8: case 32: case 128: case 512: break;
1514 default: abort();
1515 }
1516 ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1517 ddf->anchor.phys_section_length =
1518 __cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1519 sector += pdsize/512;
1520
1521 /* virt is another 32 sectors */
1522 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1523 sizeof(struct virtual_entry) * max_virt_disks,
1524 512);
1525 switch(vdsize/512) {
1526 case 2: case 8: case 32: case 128: case 512: break;
1527 default: abort();
1528 }
1529 ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1530 ddf->anchor.virt_section_length =
1531 __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1532 sector += vdsize/512;
1533
1534 clen = ddf->conf_rec_len * (ddf->max_part+1);
1535 ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1536 ddf->anchor.config_section_length = __cpu_to_be32(clen);
1537 sector += clen;
1538
1539 ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1540 ddf->anchor.data_section_length = __cpu_to_be32(1);
1541 sector += 1;
1542
1543 ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1544 ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1545 ddf->anchor.diag_space_length = __cpu_to_be32(0);
1546 ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1547 ddf->anchor.vendor_length = __cpu_to_be32(0);
1548 ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1549
1550 memset(ddf->anchor.pad4, 0xff, 256);
1551
1552 memcpy(&ddf->primary, &ddf->anchor, 512);
1553 memcpy(&ddf->secondary, &ddf->anchor, 512);
1554
1555 ddf->primary.openflag = 1; /* I guess.. */
1556 ddf->primary.type = DDF_HEADER_PRIMARY;
1557
1558 ddf->secondary.openflag = 1; /* I guess.. */
1559 ddf->secondary.type = DDF_HEADER_SECONDARY;
1560
1561 ddf->active = &ddf->primary;
1562
1563 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1564
1565 /* 24 more bytes of fiction required.
1566 * first 8 are a 'vendor-id' - "Linux-MD"
1567 * Remaining 16 are serial number.... maybe a hostname would do?
1568 */
1569 memcpy(ddf->controller.guid, T10, sizeof(T10));
1570 gethostname(hostname, sizeof(hostname));
1571 hostname[sizeof(hostname) - 1] = 0;
1572 hostlen = strlen(hostname);
1573 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1574 for (i = strlen(T10) ; i+hostlen < 24; i++)
1575 ddf->controller.guid[i] = ' ';
1576
1577 ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1578 ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1579 ddf->controller.type.sub_vendor_id = 0;
1580 ddf->controller.type.sub_device_id = 0;
1581 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1582 memset(ddf->controller.pad, 0xff, 8);
1583 memset(ddf->controller.vendor_data, 0xff, 448);
1584
1585 pd = ddf->phys = malloc(pdsize);
1586 ddf->pdsize = pdsize;
1587
1588 memset(pd, 0xff, pdsize);
1589 memset(pd, 0, sizeof(*pd));
1590 pd->magic = DDF_PHYS_DATA_MAGIC;
1591 pd->used_pdes = __cpu_to_be16(0);
1592 pd->max_pdes = __cpu_to_be16(max_phys_disks);
1593 memset(pd->pad, 0xff, 52);
1594
1595 vd = ddf->virt = malloc(vdsize);
1596 ddf->vdsize = vdsize;
1597 memset(vd, 0, vdsize);
1598 vd->magic = DDF_VIRT_RECORDS_MAGIC;
1599 vd->populated_vdes = __cpu_to_be16(0);
1600 vd->max_vdes = __cpu_to_be16(max_virt_disks);
1601 memset(vd->pad, 0xff, 52);
1602
1603 for (i=0; i<max_virt_disks; i++)
1604 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1605
1606 st->sb = ddf;
1607 return 1;
1608 }
1609
1610 static int all_ff(char *guid)
1611 {
1612 int i;
1613 for (i = 0; i < DDF_GUID_LEN; i++)
1614 if (guid[i] != (char)0xff)
1615 return 0;
1616 return 1;
1617 }
1618 static int chunk_to_shift(int chunksize)
1619 {
1620 return ffs(chunksize/512)-1;
1621 }
1622
1623 static int level_to_prl(int level)
1624 {
1625 switch (level) {
1626 case LEVEL_LINEAR: return DDF_CONCAT;
1627 case 0: return DDF_RAID0;
1628 case 1: return DDF_RAID1;
1629 case 4: return DDF_RAID4;
1630 case 5: return DDF_RAID5;
1631 case 6: return DDF_RAID6;
1632 default: return -1;
1633 }
1634 }
1635 static int layout_to_rlq(int level, int layout, int raiddisks)
1636 {
1637 switch(level) {
1638 case 0:
1639 return DDF_RAID0_SIMPLE;
1640 case 1:
1641 switch(raiddisks) {
1642 case 2: return DDF_RAID1_SIMPLE;
1643 case 3: return DDF_RAID1_MULTI;
1644 default: return -1;
1645 }
1646 case 4:
1647 switch(layout) {
1648 case 0: return DDF_RAID4_N;
1649 }
1650 break;
1651 case 5:
1652 case 6:
1653 switch(layout) {
1654 case ALGORITHM_LEFT_ASYMMETRIC:
1655 return DDF_RAID5_N_RESTART;
1656 case ALGORITHM_RIGHT_ASYMMETRIC:
1657 if (level == 5)
1658 return DDF_RAID5_0_RESTART;
1659 else
1660 return DDF_RAID6_0_RESTART;
1661 case ALGORITHM_LEFT_SYMMETRIC:
1662 return DDF_RAID5_N_CONTINUE;
1663 case ALGORITHM_RIGHT_SYMMETRIC:
1664 return -1; /* not mentioned in standard */
1665 }
1666 }
1667 return -1;
1668 }
1669
1670 static int rlq_to_layout(int rlq, int prl, int raiddisks)
1671 {
1672 switch(prl) {
1673 case DDF_RAID0:
1674 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1675 case DDF_RAID1:
1676 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1677 on raiddisks*/
1678 case DDF_RAID4:
1679 switch(rlq) {
1680 case DDF_RAID4_N:
1681 return 0;
1682 default:
1683 /* not supported */
1684 return -1; /* FIXME this isn't checked */
1685 }
1686 case DDF_RAID5:
1687 switch(rlq) {
1688 case DDF_RAID5_N_RESTART:
1689 return ALGORITHM_LEFT_ASYMMETRIC;
1690 case DDF_RAID5_0_RESTART:
1691 return ALGORITHM_RIGHT_ASYMMETRIC;
1692 case DDF_RAID5_N_CONTINUE:
1693 return ALGORITHM_LEFT_SYMMETRIC;
1694 default:
1695 return -1;
1696 }
1697 case DDF_RAID6:
1698 switch(rlq) {
1699 case DDF_RAID5_N_RESTART:
1700 return ALGORITHM_LEFT_ASYMMETRIC;
1701 case DDF_RAID6_0_RESTART:
1702 return ALGORITHM_RIGHT_ASYMMETRIC;
1703 case DDF_RAID5_N_CONTINUE:
1704 return ALGORITHM_LEFT_SYMMETRIC;
1705 default:
1706 return -1;
1707 }
1708 }
1709 return -1;
1710 }
1711
1712 struct extent {
1713 unsigned long long start, size;
1714 };
1715 int cmp_extent(const void *av, const void *bv)
1716 {
1717 const struct extent *a = av;
1718 const struct extent *b = bv;
1719 if (a->start < b->start)
1720 return -1;
1721 if (a->start > b->start)
1722 return 1;
1723 return 0;
1724 }
1725
1726 struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
1727 {
1728 /* find a list of used extents on the give physical device
1729 * (dnum) of the given ddf.
1730 * Return a malloced array of 'struct extent'
1731
1732 FIXME ignore DDF_Legacy devices?
1733
1734 */
1735 struct extent *rv;
1736 int n = 0;
1737 int i, j;
1738
1739 rv = malloc(sizeof(struct extent) * (ddf->max_part + 2));
1740 if (!rv)
1741 return NULL;
1742
1743 for (i = 0; i < ddf->max_part; i++) {
1744 struct vcl *v = dl->vlist[i];
1745 if (v == NULL)
1746 continue;
1747 for (j=0; j < v->conf.prim_elmnt_count; j++)
1748 if (v->conf.phys_refnum[j] == dl->disk.refnum) {
1749 /* This device plays role 'j' in 'v'. */
1750 rv[n].start = __be64_to_cpu(v->lba_offset[j]);
1751 rv[n].size = __be64_to_cpu(v->conf.blocks);
1752 n++;
1753 break;
1754 }
1755 }
1756 qsort(rv, n, sizeof(*rv), cmp_extent);
1757
1758 rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
1759 rv[n].size = 0;
1760 return rv;
1761 }
1762
1763 static int init_super_ddf_bvd(struct supertype *st,
1764 mdu_array_info_t *info,
1765 unsigned long long size,
1766 char *name, char *homehost,
1767 int *uuid)
1768 {
1769 /* We are creating a BVD inside a pre-existing container.
1770 * so st->sb is already set.
1771 * We need to create a new vd_config and a new virtual_entry
1772 */
1773 struct ddf_super *ddf = st->sb;
1774 int venum;
1775 struct virtual_entry *ve;
1776 struct vcl *vcl;
1777 struct vd_config *vc;
1778
1779 if (__be16_to_cpu(ddf->virt->populated_vdes)
1780 >= __be16_to_cpu(ddf->virt->max_vdes)) {
1781 fprintf(stderr, Name": This ddf already has the "
1782 "maximum of %d virtual devices\n",
1783 __be16_to_cpu(ddf->virt->max_vdes));
1784 return 0;
1785 }
1786
1787 for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1788 if (all_ff(ddf->virt->entries[venum].guid))
1789 break;
1790 if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1791 fprintf(stderr, Name ": Cannot find spare slot for "
1792 "virtual disk - DDF is corrupt\n");
1793 return 0;
1794 }
1795 ve = &ddf->virt->entries[venum];
1796
1797 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1798 * timestamp, random number
1799 */
1800 make_header_guid(ve->guid);
1801 ve->unit = __cpu_to_be16(info->md_minor);
1802 ve->pad0 = 0xFFFF;
1803 ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1804 ve->type = 0;
1805 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
1806 if (info->state & 1) /* clean */
1807 ve->init_state = DDF_init_full;
1808 else
1809 ve->init_state = DDF_init_not;
1810
1811 memset(ve->pad1, 0xff, 14);
1812 memset(ve->name, ' ', 16);
1813 if (name)
1814 strncpy(ve->name, name, 16);
1815 ddf->virt->populated_vdes =
1816 __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
1817
1818 /* Now create a new vd_config */
1819 vcl = malloc(offsetof(struct vcl, conf) + ddf->conf_rec_len * 512);
1820 vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
1821 vcl->vcnum = venum;
1822 sprintf(st->subarray, "%d", venum);
1823 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1824
1825 vc = &vcl->conf;
1826
1827 vc->magic = DDF_VD_CONF_MAGIC;
1828 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
1829 vc->timestamp = __cpu_to_be32(time(0)-DECADE);
1830 vc->seqnum = __cpu_to_be32(1);
1831 memset(vc->pad0, 0xff, 24);
1832 vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
1833 vc->chunk_shift = chunk_to_shift(info->chunk_size);
1834 vc->prl = level_to_prl(info->level);
1835 vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
1836 vc->sec_elmnt_count = 1;
1837 vc->sec_elmnt_seq = 0;
1838 vc->srl = 0;
1839 vc->blocks = __cpu_to_be64(info->size * 2);
1840 vc->array_blocks = __cpu_to_be64(
1841 calc_array_size(info->level, info->raid_disks, info->layout,
1842 info->chunk_size, info->size*2));
1843 memset(vc->pad1, 0xff, 8);
1844 vc->spare_refs[0] = 0xffffffff;
1845 vc->spare_refs[1] = 0xffffffff;
1846 vc->spare_refs[2] = 0xffffffff;
1847 vc->spare_refs[3] = 0xffffffff;
1848 vc->spare_refs[4] = 0xffffffff;
1849 vc->spare_refs[5] = 0xffffffff;
1850 vc->spare_refs[6] = 0xffffffff;
1851 vc->spare_refs[7] = 0xffffffff;
1852 memset(vc->cache_pol, 0, 8);
1853 vc->bg_rate = 0x80;
1854 memset(vc->pad2, 0xff, 3);
1855 memset(vc->pad3, 0xff, 52);
1856 memset(vc->pad4, 0xff, 192);
1857 memset(vc->v0, 0xff, 32);
1858 memset(vc->v1, 0xff, 32);
1859 memset(vc->v2, 0xff, 16);
1860 memset(vc->v3, 0xff, 16);
1861 memset(vc->vendor, 0xff, 32);
1862
1863 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
1864 memset(vc->phys_refnum+(ddf->mppe * 4), 0x00, 8*ddf->mppe);
1865
1866 vcl->next = ddf->conflist;
1867 ddf->conflist = vcl;
1868 ddf->currentconf = vcl;
1869 return 1;
1870 }
1871
1872 static void add_to_super_ddf_bvd(struct supertype *st,
1873 mdu_disk_info_t *dk, int fd, char *devname)
1874 {
1875 /* fd and devname identify a device with-in the ddf container (st).
1876 * dk identifies a location in the new BVD.
1877 * We need to find suitable free space in that device and update
1878 * the phys_refnum and lba_offset for the newly created vd_config.
1879 * We might also want to update the type in the phys_disk
1880 * section.
1881 */
1882 struct dl *dl;
1883 struct ddf_super *ddf = st->sb;
1884 struct vd_config *vc;
1885 __u64 *lba_offset;
1886 int working;
1887 int i;
1888 unsigned long long blocks, pos, esize;
1889 struct extent *ex;
1890
1891 for (dl = ddf->dlist; dl ; dl = dl->next)
1892 if (dl->major == dk->major &&
1893 dl->minor == dk->minor)
1894 break;
1895 if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
1896 return;
1897
1898 vc = &ddf->currentconf->conf;
1899 lba_offset = ddf->currentconf->lba_offset;
1900
1901 ex = get_extents(ddf, dl);
1902 if (!ex)
1903 return;
1904
1905 i = 0; pos = 0;
1906 blocks = __be64_to_cpu(vc->blocks);
1907 if (ddf->currentconf->block_sizes)
1908 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
1909
1910 do {
1911 esize = ex[i].start - pos;
1912 if (esize >= blocks)
1913 break;
1914 pos = ex[i].start + ex[i].size;
1915 i++;
1916 } while (ex[i-1].size);
1917
1918 free(ex);
1919 if (esize < blocks)
1920 return;
1921
1922 ddf->currentdev = dk->raid_disk;
1923 vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
1924 lba_offset[dk->raid_disk] = __cpu_to_be64(pos);
1925
1926 for (i=0; i < ddf->max_part ; i++)
1927 if (dl->vlist[i] == NULL)
1928 break;
1929 if (i == ddf->max_part)
1930 return;
1931 dl->vlist[i] = ddf->currentconf;
1932
1933 dl->fd = fd;
1934 dl->devname = devname;
1935
1936 /* Check how many working raid_disks, and if we can mark
1937 * array as optimal yet
1938 */
1939 working = 0;
1940
1941 for (i=0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
1942 if (vc->phys_refnum[i] != 0xffffffff)
1943 working++;
1944
1945 /* Find which virtual_entry */
1946 i = ddf->currentconf->vcnum;
1947 if (working == __be16_to_cpu(vc->prim_elmnt_count))
1948 ddf->virt->entries[i].state =
1949 (ddf->virt->entries[i].state & ~DDF_state_mask)
1950 | DDF_state_optimal;
1951
1952 if (vc->prl == DDF_RAID6 &&
1953 working+1 == __be16_to_cpu(vc->prim_elmnt_count))
1954 ddf->virt->entries[i].state =
1955 (ddf->virt->entries[i].state & ~DDF_state_mask)
1956 | DDF_state_part_optimal;
1957
1958 ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
1959 ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
1960 }
1961
1962 /* add a device to a container, either while creating it or while
1963 * expanding a pre-existing container
1964 */
1965 static void add_to_super_ddf(struct supertype *st,
1966 mdu_disk_info_t *dk, int fd, char *devname)
1967 {
1968 struct ddf_super *ddf = st->sb;
1969 struct dl *dd;
1970 time_t now;
1971 struct tm *tm;
1972 unsigned long long size;
1973 struct phys_disk_entry *pde;
1974 int n, i;
1975 struct stat stb;
1976
1977 /* This is device numbered dk->number. We need to create
1978 * a phys_disk entry and a more detailed disk_data entry.
1979 */
1980 fstat(fd, &stb);
1981 dd = malloc(sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part);
1982 dd->major = major(stb.st_rdev);
1983 dd->minor = minor(stb.st_rdev);
1984 dd->devname = devname;
1985 dd->next = ddf->dlist;
1986 dd->fd = fd;
1987 dd->spare = NULL;
1988
1989 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
1990 now = time(0);
1991 tm = localtime(&now);
1992 sprintf(dd->disk.guid, "%8s%04d%02d%02d",
1993 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
1994 *(__u32*)(dd->disk.guid + 16) = random();
1995 *(__u32*)(dd->disk.guid + 20) = random();
1996
1997 do {
1998 /* Cannot be bothered finding a CRC of some irrelevant details*/
1999 dd->disk.refnum = random();
2000 for (i = __be16_to_cpu(ddf->active->max_pd_entries) - 1;
2001 i >= 0; i--)
2002 if (ddf->phys->entries[i].refnum == dd->disk.refnum)
2003 break;
2004 } while (i >= 0);
2005
2006 dd->disk.forced_ref = 1;
2007 dd->disk.forced_guid = 1;
2008 memset(dd->disk.vendor, ' ', 32);
2009 memcpy(dd->disk.vendor, "Linux", 5);
2010 memset(dd->disk.pad, 0xff, 442);
2011 for (i = 0; i < ddf->max_part ; i++)
2012 dd->vlist[i] = NULL;
2013
2014 n = __be16_to_cpu(ddf->phys->used_pdes);
2015 pde = &ddf->phys->entries[n];
2016 dd->pdnum = n;
2017
2018 n++;
2019 ddf->phys->used_pdes = __cpu_to_be16(n);
2020
2021 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2022 pde->refnum = dd->disk.refnum;
2023 pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2024 pde->state = __cpu_to_be16(DDF_Online);
2025 get_dev_size(fd, NULL, &size);
2026 /* We are required to reserve 32Meg, and record the size in sectors */
2027 pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
2028 sprintf(pde->path, "%17.17s","Information: nil") ;
2029 memset(pde->pad, 0xff, 6);
2030
2031 dd->size = size >> 9;
2032 ddf->dlist = dd;
2033 }
2034
2035 /*
2036 * This is the write_init_super method for a ddf container. It is
2037 * called when creating a container or adding another device to a
2038 * container.
2039 */
2040
2041 #ifndef MDASSEMBLE
2042 static int __write_init_super_ddf(struct supertype *st, int do_close)
2043 {
2044
2045 struct ddf_super *ddf = st->sb;
2046 int i;
2047 struct dl *d;
2048 int n_config;
2049 int conf_size;
2050
2051 unsigned long long size, sector;
2052
2053 for (d = ddf->dlist; d; d=d->next) {
2054 int fd = d->fd;
2055
2056 if (fd < 0)
2057 continue;
2058
2059 /* We need to fill in the primary, (secondary) and workspace
2060 * lba's in the headers, set their checksums,
2061 * Also checksum phys, virt....
2062 *
2063 * Then write everything out, finally the anchor is written.
2064 */
2065 get_dev_size(fd, NULL, &size);
2066 size /= 512;
2067 ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
2068 ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
2069 ddf->anchor.seq = __cpu_to_be32(1);
2070 memcpy(&ddf->primary, &ddf->anchor, 512);
2071 memcpy(&ddf->secondary, &ddf->anchor, 512);
2072
2073 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
2074 ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
2075 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
2076
2077 ddf->primary.openflag = 0;
2078 ddf->primary.type = DDF_HEADER_PRIMARY;
2079
2080 ddf->secondary.openflag = 0;
2081 ddf->secondary.type = DDF_HEADER_SECONDARY;
2082
2083 ddf->primary.crc = calc_crc(&ddf->primary, 512);
2084 ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
2085
2086 sector = size - 16*1024*2;
2087 lseek64(fd, sector<<9, 0);
2088 write(fd, &ddf->primary, 512);
2089
2090 ddf->controller.crc = calc_crc(&ddf->controller, 512);
2091 write(fd, &ddf->controller, 512);
2092
2093 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
2094
2095 write(fd, ddf->phys, ddf->pdsize);
2096
2097 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
2098 write(fd, ddf->virt, ddf->vdsize);
2099
2100 /* Now write lots of config records. */
2101 n_config = ddf->max_part;
2102 conf_size = ddf->conf_rec_len * 512;
2103 for (i = 0 ; i <= n_config ; i++) {
2104 struct vcl *c = d->vlist[i];
2105 if (i == n_config)
2106 c = (struct vcl*)d->spare;
2107
2108 if (c) {
2109 c->conf.crc = calc_crc(&c->conf, conf_size);
2110 write(fd, &c->conf, conf_size);
2111 } else {
2112 __u32 sig = 0xffffffff;
2113 write(fd, &sig, 4);
2114 lseek64(fd, conf_size-4, SEEK_CUR);
2115 }
2116 }
2117 d->disk.crc = calc_crc(&d->disk, 512);
2118 write(fd, &d->disk, 512);
2119
2120 /* Maybe do the same for secondary */
2121
2122 lseek64(fd, (size-1)*512, SEEK_SET);
2123 write(fd, &ddf->anchor, 512);
2124 if (do_close) {
2125 close(fd);
2126 d->fd = -1;
2127 }
2128 }
2129 return 1;
2130 }
2131
2132 static int write_init_super_ddf(struct supertype *st)
2133 {
2134 return __write_init_super_ddf(st, 1);
2135 }
2136
2137 #endif
2138
2139 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize)
2140 {
2141 /* We must reserve the last 32Meg */
2142 if (devsize <= 32*1024*2)
2143 return 0;
2144 return devsize - 32*1024*2;
2145 }
2146
2147 #ifndef MDASSEMBLE
2148 int validate_geometry_ddf(struct supertype *st,
2149 int level, int layout, int raiddisks,
2150 int chunk, unsigned long long size,
2151 char *dev, unsigned long long *freesize)
2152 {
2153 int fd;
2154 struct mdinfo *sra;
2155 int cfd;
2156
2157 /* ddf potentially supports lots of things, but it depends on
2158 * what devices are offered (and maybe kernel version?)
2159 * If given unused devices, we will make a container.
2160 * If given devices in a container, we will make a BVD.
2161 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2162 */
2163
2164 if (level == LEVEL_CONTAINER) {
2165 st->ss = &super_ddf_container;
2166 if (dev) {
2167 int rv =st->ss->validate_geometry(st, level, layout,
2168 raiddisks, chunk,
2169 size,
2170 NULL, freesize);
2171 if (rv)
2172 return rv;
2173 }
2174 return st->ss->validate_geometry(st, level, layout, raiddisks,
2175 chunk, size, dev, freesize);
2176 }
2177
2178 if (st->sb) {
2179 /* creating in a given container */
2180 st->ss = &super_ddf_bvd;
2181 if (dev) {
2182 int rv =st->ss->validate_geometry(st, level, layout,
2183 raiddisks, chunk,
2184 size,
2185 NULL, freesize);
2186 if (rv)
2187 return rv;
2188 }
2189 return st->ss->validate_geometry(st, level, layout, raiddisks,
2190 chunk, size, dev, freesize);
2191 }
2192 /* FIXME should exclude MULTIPATH, or more appropriately, allow
2193 * only known levels.
2194 */
2195 if (!dev)
2196 return 1;
2197
2198 /* This device needs to be either a device in a 'ddf' container,
2199 * or it needs to be a 'ddf-bvd' array.
2200 */
2201
2202 fd = open(dev, O_RDONLY|O_EXCL, 0);
2203 if (fd >= 0) {
2204 sra = sysfs_read(fd, 0, GET_VERSION);
2205 close(fd);
2206 if (sra && sra->array.major_version == -1 &&
2207 strcmp(sra->text_version, "ddf-bvd") == 0) {
2208 st->ss = &super_ddf_svd;
2209 return st->ss->validate_geometry(st, level, layout,
2210 raiddisks, chunk, size,
2211 dev, freesize);
2212 }
2213
2214 fprintf(stderr,
2215 Name ": Cannot create this array on device %s\n",
2216 dev);
2217 return 0;
2218 }
2219 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2220 fprintf(stderr, Name ": Cannot open %s: %s\n",
2221 dev, strerror(errno));
2222 return 0;
2223 }
2224 /* Well, it is in use by someone, maybe a 'ddf' container. */
2225 cfd = open_container(fd);
2226 if (cfd < 0) {
2227 close(fd);
2228 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
2229 dev);
2230 return 0;
2231 }
2232 sra = sysfs_read(cfd, 0, GET_VERSION);
2233 close(fd);
2234 if (sra && sra->array.major_version == -1 &&
2235 strcmp(sra->text_version, "ddf") == 0) {
2236 /* This is a member of a ddf container. Load the container
2237 * and try to create a bvd
2238 */
2239 struct ddf_super *ddf;
2240 st->ss = &super_ddf_bvd;
2241 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL, 1) == 0) {
2242 st->sb = ddf;
2243 st->container_dev = fd2devnum(cfd);
2244 close(cfd);
2245 return st->ss->validate_geometry(st, level, layout,
2246 raiddisks, chunk, size,
2247 dev, freesize);
2248 }
2249 close(cfd);
2250 } else /* device may belong to a different container */
2251 return 0;
2252
2253 return 1;
2254 }
2255
2256 int validate_geometry_ddf_container(struct supertype *st,
2257 int level, int layout, int raiddisks,
2258 int chunk, unsigned long long size,
2259 char *dev, unsigned long long *freesize)
2260 {
2261 int fd;
2262 unsigned long long ldsize;
2263
2264 if (level != LEVEL_CONTAINER)
2265 return 0;
2266 if (!dev)
2267 return 1;
2268
2269 fd = open(dev, O_RDONLY|O_EXCL, 0);
2270 if (fd < 0) {
2271 fprintf(stderr, Name ": Cannot open %s: %s\n",
2272 dev, strerror(errno));
2273 return 0;
2274 }
2275 if (!get_dev_size(fd, dev, &ldsize)) {
2276 close(fd);
2277 return 0;
2278 }
2279 close(fd);
2280
2281 *freesize = avail_size_ddf(st, ldsize >> 9);
2282
2283 return 1;
2284 }
2285
2286 int validate_geometry_ddf_bvd(struct supertype *st,
2287 int level, int layout, int raiddisks,
2288 int chunk, unsigned long long size,
2289 char *dev, unsigned long long *freesize)
2290 {
2291 struct stat stb;
2292 struct ddf_super *ddf = st->sb;
2293 struct dl *dl;
2294 unsigned long long pos = 0;
2295 unsigned long long maxsize;
2296 struct extent *e;
2297 int i;
2298 /* ddf/bvd supports lots of things, but not containers */
2299 if (level == LEVEL_CONTAINER)
2300 return 0;
2301 /* We must have the container info already read in. */
2302 if (!ddf)
2303 return 0;
2304
2305 if (!dev) {
2306 /* General test: make sure there is space for
2307 * 'raiddisks' device extents of size 'size'.
2308 */
2309 unsigned long long minsize = size;
2310 int dcnt = 0;
2311 if (minsize == 0)
2312 minsize = 8;
2313 for (dl = ddf->dlist; dl ; dl = dl->next)
2314 {
2315 int found = 0;
2316 pos = 0;
2317
2318 i = 0;
2319 e = get_extents(ddf, dl);
2320 if (!e) continue;
2321 do {
2322 unsigned long long esize;
2323 esize = e[i].start - pos;
2324 if (esize >= minsize)
2325 found = 1;
2326 pos = e[i].start + e[i].size;
2327 i++;
2328 } while (e[i-1].size);
2329 if (found)
2330 dcnt++;
2331 free(e);
2332 }
2333 if (dcnt < raiddisks) {
2334 fprintf(stderr, Name ": Not enough devices with space "
2335 "for this array (%d < %d)\n",
2336 dcnt, raiddisks);
2337 return 0;
2338 }
2339 return 1;
2340 }
2341 /* This device must be a member of the set */
2342 if (stat(dev, &stb) < 0)
2343 return 0;
2344 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2345 return 0;
2346 for (dl = ddf->dlist ; dl ; dl = dl->next) {
2347 if (dl->major == major(stb.st_rdev) &&
2348 dl->minor == minor(stb.st_rdev))
2349 break;
2350 }
2351 if (!dl) {
2352 fprintf(stderr, Name ": %s is not in the same DDF set\n",
2353 dev);
2354 return 0;
2355 }
2356 e = get_extents(ddf, dl);
2357 maxsize = 0;
2358 i = 0;
2359 if (e) do {
2360 unsigned long long esize;
2361 esize = e[i].start - pos;
2362 if (esize >= maxsize)
2363 maxsize = esize;
2364 pos = e[i].start + e[i].size;
2365 i++;
2366 } while (e[i-1].size);
2367 *freesize = maxsize;
2368 // FIXME here I am
2369
2370 return 1;
2371 }
2372
2373 int validate_geometry_ddf_svd(struct supertype *st,
2374 int level, int layout, int raiddisks,
2375 int chunk, unsigned long long size,
2376 char *dev, unsigned long long *freesize)
2377 {
2378 /* dd/svd only supports striped, mirrored, concat, spanned... */
2379 if (level != LEVEL_LINEAR &&
2380 level != 0 &&
2381 level != 1)
2382 return 0;
2383 return 1;
2384 }
2385
2386 static int load_super_ddf_all(struct supertype *st, int fd,
2387 void **sbp, char *devname, int keep_fd)
2388 {
2389 struct mdinfo *sra;
2390 struct ddf_super *super;
2391 struct mdinfo *sd, *best = NULL;
2392 int bestseq = 0;
2393 int seq;
2394 char nm[20];
2395 int dfd;
2396
2397 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
2398 if (!sra)
2399 return 1;
2400 if (sra->array.major_version != -1 ||
2401 sra->array.minor_version != -2 ||
2402 strcmp(sra->text_version, "ddf") != 0)
2403 return 1;
2404
2405 super = malloc(sizeof(*super));
2406 if (!super)
2407 return 1;
2408 memset(super, 0, sizeof(*super));
2409
2410 /* first, try each device, and choose the best ddf */
2411 for (sd = sra->devs ; sd ; sd = sd->next) {
2412 int rv;
2413 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2414 dfd = dev_open(nm, O_RDONLY);
2415 if (dfd < 0)
2416 return 2;
2417 rv = load_ddf_headers(dfd, super, NULL);
2418 close(dfd);
2419 if (rv == 0) {
2420 seq = __be32_to_cpu(super->active->seq);
2421 if (super->active->openflag)
2422 seq--;
2423 if (!best || seq > bestseq) {
2424 bestseq = seq;
2425 best = sd;
2426 }
2427 }
2428 }
2429 if (!best)
2430 return 1;
2431 /* OK, load this ddf */
2432 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2433 dfd = dev_open(nm, O_RDONLY);
2434 if (dfd < 0)
2435 return 1;
2436 load_ddf_headers(dfd, super, NULL);
2437 load_ddf_global(dfd, super, NULL);
2438 close(dfd);
2439 /* Now we need the device-local bits */
2440 for (sd = sra->devs ; sd ; sd = sd->next) {
2441 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2442 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2443 if (dfd < 0)
2444 return 2;
2445 seq = load_ddf_local(dfd, super, NULL, keep_fd);
2446 if (!keep_fd) close(dfd);
2447 }
2448 if (st->subarray[0]) {
2449 struct vcl *v;
2450
2451 for (v = super->conflist; v; v = v->next)
2452 if (v->vcnum == atoi(st->subarray))
2453 super->currentconf = v;
2454 if (!super->currentconf)
2455 return 1;
2456 }
2457 *sbp = super;
2458 if (st->ss == NULL) {
2459 st->ss = &super_ddf_container;
2460 st->minor_version = 0;
2461 st->max_devs = 512;
2462 st->container_dev = fd2devnum(fd);
2463 }
2464 return 0;
2465 }
2466 #endif
2467
2468 static struct mdinfo *container_content_ddf(struct supertype *st)
2469 {
2470 /* Given a container loaded by load_super_ddf_all,
2471 * extract information about all the arrays into
2472 * an mdinfo tree.
2473 *
2474 * For each vcl in conflist: create an mdinfo, fill it in,
2475 * then look for matching devices (phys_refnum) in dlist
2476 * and create appropriate device mdinfo.
2477 */
2478 struct ddf_super *ddf = st->sb;
2479 struct mdinfo *rest = NULL;
2480 struct vcl *vc;
2481
2482 for (vc = ddf->conflist ; vc ; vc=vc->next)
2483 {
2484 int i;
2485 struct mdinfo *this;
2486 this = malloc(sizeof(*this));
2487 memset(this, 0, sizeof(*this));
2488 this->next = rest;
2489 rest = this;
2490
2491 this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2492 this->array.raid_disks =
2493 __be16_to_cpu(vc->conf.prim_elmnt_count);
2494 this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2495 this->array.raid_disks);
2496 this->array.md_minor = -1;
2497 this->array.ctime = DECADE +
2498 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
2499 this->array.utime = DECADE +
2500 __be32_to_cpu(vc->conf.timestamp);
2501 this->array.chunk_size = 512 << vc->conf.chunk_shift;
2502
2503 i = vc->vcnum;
2504 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2505 (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
2506 DDF_init_full) {
2507 this->array.state = 0;
2508 this->resync_start = 0;
2509 } else {
2510 this->array.state = 1;
2511 this->resync_start = ~0ULL;
2512 }
2513 memcpy(this->name, ddf->virt->entries[i].name, 32);
2514 this->name[33]=0;
2515
2516 memset(this->uuid, 0, sizeof(this->uuid));
2517 this->component_size = __be64_to_cpu(vc->conf.blocks);
2518 this->array.size = this->component_size / 2;
2519 this->container_member = i;
2520
2521 sprintf(this->text_version, "/%s/%d",
2522 devnum2devname(st->container_dev),
2523 this->container_member);
2524
2525 for (i=0 ; i < ddf->mppe ; i++) {
2526 struct mdinfo *dev;
2527 struct dl *d;
2528
2529 if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
2530 continue;
2531
2532 this->array.working_disks++;
2533
2534 for (d = ddf->dlist; d ; d=d->next)
2535 if (d->disk.refnum == vc->conf.phys_refnum[i])
2536 break;
2537 if (d == NULL)
2538 break;
2539
2540 dev = malloc(sizeof(*dev));
2541 memset(dev, 0, sizeof(*dev));
2542 dev->next = this->devs;
2543 this->devs = dev;
2544
2545 dev->disk.number = __be32_to_cpu(d->disk.refnum);
2546 dev->disk.major = d->major;
2547 dev->disk.minor = d->minor;
2548 dev->disk.raid_disk = i;
2549 dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2550
2551 dev->events = __le32_to_cpu(ddf->primary.seq);
2552 dev->data_offset = vc->lba_offset[i];
2553 dev->component_size = __be64_to_cpu(vc->conf.blocks);
2554 if (d->devname)
2555 strcpy(dev->name, d->devname);
2556 }
2557 }
2558 return rest;
2559 }
2560
2561 static int init_zero_ddf(struct supertype *st,
2562 mdu_array_info_t *info,
2563 unsigned long long size, char *name,
2564 char *homehost, int *uuid)
2565 {
2566 st->sb = NULL;
2567 return 0;
2568 }
2569
2570 static int store_zero_ddf(struct supertype *st, int fd)
2571 {
2572 unsigned long long dsize;
2573 char buf[512];
2574 memset(buf, 0, 512);
2575
2576 if (!get_dev_size(fd, NULL, &dsize))
2577 return 1;
2578
2579 lseek64(fd, dsize-512, 0);
2580 write(fd, buf, 512);
2581 return 0;
2582 }
2583
2584 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
2585 {
2586 /*
2587 * return:
2588 * 0 same, or first was empty, and second was copied
2589 * 1 second had wrong number
2590 * 2 wrong uuid
2591 * 3 wrong other info
2592 */
2593 struct ddf_super *first = st->sb;
2594 struct ddf_super *second = tst->sb;
2595
2596 if (!first) {
2597 st->sb = tst->sb;
2598 tst->sb = NULL;
2599 return 0;
2600 }
2601
2602 if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
2603 return 2;
2604
2605 /* FIXME should I look at anything else? */
2606 return 0;
2607 }
2608
2609 /*
2610 * A new array 'a' has been started which claims to be instance 'inst'
2611 * within container 'c'.
2612 * We need to confirm that the array matches the metadata in 'c' so
2613 * that we don't corrupt any metadata.
2614 */
2615 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
2616 {
2617 fprintf(stderr, "ddf: open_new %s\n", inst);
2618 a->info.container_member = atoi(inst);
2619 return 0;
2620 }
2621
2622 /*
2623 * The array 'a' is to be marked clean in the metadata.
2624 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2625 * clean up to the point (in sectors). If that cannot be recorded in the
2626 * metadata, then leave it as dirty.
2627 *
2628 * For DDF, we need to clear the DDF_state_inconsistent bit in the
2629 * !global! virtual_disk.virtual_entry structure.
2630 */
2631 static void ddf_set_array_state(struct active_array *a, int consistent)
2632 {
2633 struct ddf_super *ddf = a->container->sb;
2634 int inst = a->info.container_member;
2635 if (consistent)
2636 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
2637 else
2638 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
2639 ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
2640 if (a->resync_start == ~0ULL)
2641 ddf->virt->entries[inst].init_state |= DDF_init_full;
2642 else if (a->resync_start == 0)
2643 ddf->virt->entries[inst].init_state |= DDF_init_not;
2644 else
2645 ddf->virt->entries[inst].init_state |= DDF_init_quick;
2646
2647 printf("ddf mark %s %llu\n", consistent?"clean":"dirty",
2648 a->resync_start);
2649 }
2650
2651 /*
2652 * The state of each disk is stored in the global phys_disk structure
2653 * in phys_disk.entries[n].state.
2654 * This makes various combinations awkward.
2655 * - When a device fails in any array, it must be failed in all arrays
2656 * that include a part of this device.
2657 * - When a component is rebuilding, we cannot include it officially in the
2658 * array unless this is the only array that uses the device.
2659 *
2660 * So: when transitioning:
2661 * Online -> failed, just set failed flag. monitor will propagate
2662 * spare -> online, the device might need to be added to the array.
2663 * spare -> failed, just set failed. Don't worry if in array or not.
2664 */
2665 static void ddf_set_disk(struct active_array *a, int n, int state)
2666 {
2667 struct ddf_super *ddf = a->container->sb;
2668 int inst = a->info.container_member;
2669 struct vd_config *vc = find_vdcr(ddf, inst);
2670 int pd = find_phys(ddf, vc->phys_refnum[n]);
2671 int i, st, working;
2672
2673 if (vc == NULL) {
2674 fprintf(stderr, "ddf: cannot find instance %d!!\n", inst);
2675 return;
2676 }
2677 if (pd < 0) {
2678 /* disk doesn't currently exist. If it is now in_sync,
2679 * insert it. */
2680 if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
2681 /* Find dev 'n' in a->info->devs, determine the
2682 * ddf refnum, and set vc->phys_refnum and update
2683 * phys->entries[]
2684 */
2685 /* FIXME */
2686 }
2687 } else {
2688 if (state & DS_FAULTY)
2689 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Failed);
2690 if (state & DS_INSYNC) {
2691 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Online);
2692 ddf->phys->entries[pd].state &= __cpu_to_be16(~DDF_Rebuilding);
2693 }
2694 }
2695
2696 fprintf(stderr, "ddf: set_disk %d to %x\n", n, state);
2697
2698 /* Now we need to check the state of the array and update
2699 * virtual_disk.entries[n].state.
2700 * It needs to be one of "optimal", "degraded", "failed".
2701 * I don't understand 'deleted' or 'missing'.
2702 */
2703 working = 0;
2704 for (i=0; i < a->info.array.raid_disks; i++) {
2705 pd = find_phys(ddf, vc->phys_refnum[i]);
2706 if (pd < 0)
2707 continue;
2708 st = __be16_to_cpu(ddf->phys->entries[pd].state);
2709 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
2710 == DDF_Online)
2711 working++;
2712 }
2713 state = DDF_state_degraded;
2714 if (working == a->info.array.raid_disks)
2715 state = DDF_state_optimal;
2716 else switch(vc->prl) {
2717 case DDF_RAID0:
2718 case DDF_CONCAT:
2719 case DDF_JBOD:
2720 state = DDF_state_failed;
2721 break;
2722 case DDF_RAID1:
2723 if (working == 0)
2724 state = DDF_state_failed;
2725 break;
2726 case DDF_RAID4:
2727 case DDF_RAID5:
2728 if (working < a->info.array.raid_disks-1)
2729 state = DDF_state_failed;
2730 break;
2731 case DDF_RAID6:
2732 if (working < a->info.array.raid_disks-2)
2733 state = DDF_state_failed;
2734 else if (working == a->info.array.raid_disks-1)
2735 state = DDF_state_part_optimal;
2736 break;
2737 }
2738
2739 ddf->virt->entries[inst].state =
2740 (ddf->virt->entries[inst].state & ~DDF_state_mask)
2741 | state;
2742
2743 }
2744
2745 static void ddf_sync_metadata(struct supertype *st)
2746 {
2747
2748 /*
2749 * Write all data to all devices.
2750 * Later, we might be able to track whether only local changes
2751 * have been made, or whether any global data has been changed,
2752 * but ddf is sufficiently weird that it probably always
2753 * changes global data ....
2754 */
2755 __write_init_super_ddf(st, 0);
2756 fprintf(stderr, "ddf: sync_metadata\n");
2757 }
2758
2759 static void ddf_process_update(struct supertype *st,
2760 struct metadata_update *update)
2761 {
2762 /* Apply this update to the metadata.
2763 * The first 4 bytes are a DDF_*_MAGIC which guides
2764 * our actions.
2765 * Possible update are:
2766 * DDF_PHYS_RECORDS_MAGIC
2767 * Add a new physical device. Changes to this record
2768 * only happen implicitly.
2769 * used_pdes is the device number.
2770 * DDF_VIRT_RECORDS_MAGIC
2771 * Add a new VD. Possibly also change the 'access' bits.
2772 * populated_vdes is the entry number.
2773 * DDF_VD_CONF_MAGIC
2774 * New or updated VD. the VIRT_RECORD must already
2775 * exist. For an update, phys_refnum and lba_offset
2776 * (at least) are updated, and the VD_CONF must
2777 * be written to precisely those devices listed with
2778 * a phys_refnum.
2779 * DDF_SPARE_ASSIGN_MAGIC
2780 * replacement Spare Assignment Record... but for which device?
2781 *
2782 * So, e.g.:
2783 * - to create a new array, we send a VIRT_RECORD and
2784 * a VD_CONF. Then assemble and start the array.
2785 * - to activate a spare we send a VD_CONF to add the phys_refnum
2786 * and offset. This will also mark the spare as active with
2787 * a spare-assignment record.
2788 */
2789 struct ddf_super *ddf = st->sb;
2790 __u32 *magic = (__u32*)update->buf;
2791 struct phys_disk *pd;
2792 struct virtual_disk *vd;
2793 struct vd_config *vc;
2794 struct vcl *vcl;
2795 struct dl *dl;
2796 int mppe;
2797 int ent;
2798
2799 printf("Process update %x\n", *magic);
2800
2801 switch (*magic) {
2802 case DDF_PHYS_RECORDS_MAGIC:
2803
2804 if (update->len != (sizeof(struct phys_disk) +
2805 sizeof(struct phys_disk_entry)))
2806 return;
2807 pd = (struct phys_disk*)update->buf;
2808
2809 ent = __be16_to_cpu(pd->used_pdes);
2810 if (ent >= __be16_to_cpu(ddf->phys->max_pdes))
2811 return;
2812 if (!all_ff(ddf->phys->entries[ent].guid))
2813 return;
2814 ddf->phys->entries[ent] = pd->entries[0];
2815 ddf->phys->used_pdes = __cpu_to_be16(1 +
2816 __be16_to_cpu(ddf->phys->used_pdes));
2817 break;
2818
2819 case DDF_VIRT_RECORDS_MAGIC:
2820
2821 if (update->len != (sizeof(struct virtual_disk) +
2822 sizeof(struct virtual_entry)))
2823 return;
2824 vd = (struct virtual_disk*)update->buf;
2825
2826 ent = __be16_to_cpu(vd->populated_vdes);
2827 if (ent >= __be16_to_cpu(ddf->virt->max_vdes))
2828 return;
2829 if (!all_ff(ddf->virt->entries[ent].guid))
2830 return;
2831 ddf->virt->entries[ent] = vd->entries[0];
2832 ddf->virt->populated_vdes = __cpu_to_be16(1 +
2833 __be16_to_cpu(ddf->virt->populated_vdes));
2834 break;
2835
2836 case DDF_VD_CONF_MAGIC:
2837 printf("len %d %d\n", update->len, ddf->conf_rec_len);
2838
2839 mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries);
2840 if (update->len != ddf->conf_rec_len)
2841 return;
2842 vc = (struct vd_config*)update->buf;
2843 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2844 if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
2845 break;
2846 printf("vcl = %p\n", vcl);
2847 if (vcl) {
2848 /* An update, just copy the phys_refnum and lba_offset
2849 * fields
2850 */
2851 memcpy(vcl->conf.phys_refnum, vc->phys_refnum,
2852 mppe * (sizeof(__u32) + sizeof(__u64)));
2853 } else {
2854 /* A new VD_CONF */
2855 vcl = update->space;
2856 update->space = NULL;
2857 vcl->next = ddf->conflist;
2858 vcl->conf = *vc;
2859 vcl->lba_offset = (__u64*)
2860 &vcl->conf.phys_refnum[mppe];
2861 ddf->conflist = vcl;
2862 }
2863 /* Now make sure vlist is correct for each dl. */
2864 for (dl = ddf->dlist; dl; dl = dl->next) {
2865 int dn;
2866 int vn = 0;
2867 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2868 for (dn=0; dn < ddf->mppe ; dn++)
2869 if (vcl->conf.phys_refnum[dn] ==
2870 dl->disk.refnum) {
2871 printf("dev %d has %p at %d\n",
2872 dl->pdnum, vcl, vn);
2873 dl->vlist[vn++] = vcl;
2874 break;
2875 }
2876 while (vn < ddf->max_part)
2877 dl->vlist[vn++] = NULL;
2878 if (dl->vlist[0]) {
2879 ddf->phys->entries[dl->pdnum].type &=
2880 ~__cpu_to_be16(DDF_Global_Spare);
2881 ddf->phys->entries[dl->pdnum].type |=
2882 __cpu_to_be16(DDF_Active_in_VD);
2883 }
2884 if (dl->spare) {
2885 ddf->phys->entries[dl->pdnum].type &=
2886 ~__cpu_to_be16(DDF_Global_Spare);
2887 ddf->phys->entries[dl->pdnum].type |=
2888 __cpu_to_be16(DDF_Spare);
2889 }
2890 if (!dl->vlist[0] && !dl->spare) {
2891 ddf->phys->entries[dl->pdnum].type |=
2892 __cpu_to_be16(DDF_Global_Spare);
2893 ddf->phys->entries[dl->pdnum].type &=
2894 ~__cpu_to_be16(DDF_Spare |
2895 DDF_Active_in_VD);
2896 }
2897 }
2898 break;
2899 case DDF_SPARE_ASSIGN_MAGIC:
2900 default: break;
2901 }
2902 }
2903
2904 /*
2905 * Check if the array 'a' is degraded but not failed.
2906 * If it is, find as many spares as are available and needed and
2907 * arrange for their inclusion.
2908 * We only choose devices which are not already in the array,
2909 * and prefer those with a spare-assignment to this array.
2910 * otherwise we choose global spares - assuming always that
2911 * there is enough room.
2912 * For each spare that we assign, we return an 'mdinfo' which
2913 * describes the position for the device in the array.
2914 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
2915 * the new phys_refnum and lba_offset values.
2916 *
2917 * Only worry about BVDs at the moment.
2918 */
2919 static struct mdinfo *ddf_activate_spare(struct active_array *a,
2920 struct metadata_update **updates)
2921 {
2922 int working = 0;
2923 struct mdinfo *d;
2924 struct ddf_super *ddf = a->container->sb;
2925 int global_ok = 0;
2926 struct mdinfo *rv = NULL;
2927 struct mdinfo *di;
2928 struct metadata_update *mu;
2929 struct dl *dl;
2930 int i;
2931 struct vd_config *vc;
2932 __u64 *lba;
2933
2934 for (d = a->info.devs ; d ; d = d->next) {
2935 if ((d->curr_state & DS_FAULTY) &&
2936 d->state_fd >= 0)
2937 /* wait for Removal to happen */
2938 return NULL;
2939 if (d->state_fd >= 0)
2940 working ++;
2941 }
2942
2943 printf("ddf_activate: working=%d (%d) level=%d\n", working, a->info.array.raid_disks,
2944 a->info.array.level);
2945 if (working == a->info.array.raid_disks)
2946 return NULL; /* array not degraded */
2947 switch (a->info.array.level) {
2948 case 1:
2949 if (working == 0)
2950 return NULL; /* failed */
2951 break;
2952 case 4:
2953 case 5:
2954 if (working < a->info.array.raid_disks - 1)
2955 return NULL; /* failed */
2956 break;
2957 case 6:
2958 if (working < a->info.array.raid_disks - 2)
2959 return NULL; /* failed */
2960 break;
2961 default: /* concat or stripe */
2962 return NULL; /* failed */
2963 }
2964
2965 /* For each slot, if it is not working, find a spare */
2966 dl = ddf->dlist;
2967 for (i = 0; i < a->info.array.raid_disks; i++) {
2968 for (d = a->info.devs ; d ; d = d->next)
2969 if (d->disk.raid_disk == i)
2970 break;
2971 printf("found %d: %p %x\n", i, d, d?d->curr_state:0);
2972 if (d && (d->state_fd >= 0))
2973 continue;
2974
2975 /* OK, this device needs recovery. Find a spare */
2976 again:
2977 for ( ; dl ; dl = dl->next) {
2978 unsigned long long esize;
2979 unsigned long long pos;
2980 struct mdinfo *d2;
2981 int is_global = 0;
2982 int is_dedicated = 0;
2983 struct extent *ex;
2984 int j;
2985 /* If in this array, skip */
2986 for (d2 = a->info.devs ; d2 ; d2 = d2->next)
2987 if (d2->disk.major == dl->major &&
2988 d2->disk.minor == dl->minor) {
2989 printf("%x:%x already in array\n", dl->major, dl->minor);
2990 break;
2991 }
2992 if (d2)
2993 continue;
2994 if (ddf->phys->entries[dl->pdnum].type &
2995 __cpu_to_be16(DDF_Spare)) {
2996 /* Check spare assign record */
2997 if (dl->spare) {
2998 if (dl->spare->type & DDF_spare_dedicated) {
2999 /* check spare_ents for guid */
3000 for (j = 0 ;
3001 j < __be16_to_cpu(dl->spare->populated);
3002 j++) {
3003 if (memcmp(dl->spare->spare_ents[j].guid,
3004 ddf->virt->entries[a->info.container_member].guid,
3005 DDF_GUID_LEN) == 0)
3006 is_dedicated = 1;
3007 }
3008 } else
3009 is_global = 1;
3010 }
3011 } else if (ddf->phys->entries[dl->pdnum].type &
3012 __cpu_to_be16(DDF_Global_Spare)) {
3013 is_global = 1;
3014 }
3015 if ( ! (is_dedicated ||
3016 (is_global && global_ok))) {
3017 printf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
3018 is_dedicated, is_global);
3019 continue;
3020 }
3021
3022 /* We are allowed to use this device - is there space?
3023 * We need a->info.component_size sectors */
3024 ex = get_extents(ddf, dl);
3025 if (!ex) {
3026 printf("cannot get extents\n");
3027 continue;
3028 }
3029 j = 0; pos = 0;
3030 esize = 0;
3031
3032 do {
3033 esize = ex[j].start - pos;
3034 if (esize >= a->info.component_size)
3035 break;
3036 pos = ex[i].start + ex[i].size;
3037 i++;
3038 } while (ex[i-1].size);
3039
3040 free(ex);
3041 if (esize < a->info.component_size) {
3042 printf("%x:%x has no room: %llu %llu\n", dl->major, dl->minor,
3043 esize, a->info.component_size);
3044 /* No room */
3045 continue;
3046 }
3047
3048 /* Cool, we have a device with some space at pos */
3049 di = malloc(sizeof(*di));
3050 memset(di, 0, sizeof(*di));
3051 di->disk.number = i;
3052 di->disk.raid_disk = i;
3053 di->disk.major = dl->major;
3054 di->disk.minor = dl->minor;
3055 di->disk.state = 0;
3056 di->data_offset = pos;
3057 di->component_size = a->info.component_size;
3058 di->container_member = dl->pdnum;
3059 di->next = rv;
3060 rv = di;
3061 printf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3062 i, pos);
3063
3064 break;
3065 }
3066 if (!dl && ! global_ok) {
3067 /* not enough dedicated spares, try global */
3068 global_ok = 1;
3069 dl = ddf->dlist;
3070 goto again;
3071 }
3072 }
3073
3074 if (!rv)
3075 /* No spares found */
3076 return rv;
3077 /* Now 'rv' has a list of devices to return.
3078 * Create a metadata_update record to update the
3079 * phys_refnum and lba_offset values
3080 */
3081 mu = malloc(sizeof(*mu));
3082 mu->buf = malloc(ddf->conf_rec_len * 512);
3083 mu->space = malloc(sizeof(struct vcl));
3084 mu->len = ddf->conf_rec_len;
3085 mu->next = *updates;
3086 vc = find_vdcr(ddf, a->info.container_member);
3087 memcpy(mu->buf, vc, ddf->conf_rec_len * 512);
3088
3089 vc = (struct vd_config*)mu->buf;
3090 lba = (__u64*)&vc->phys_refnum[ddf->mppe];
3091 for (di = rv ; di ; di = di->next) {
3092 vc->phys_refnum[di->disk.raid_disk] =
3093 ddf->phys->entries[dl->pdnum].refnum;
3094 lba[di->disk.raid_disk] = di->data_offset;
3095 }
3096 *updates = mu;
3097 return rv;
3098 }
3099
3100 struct superswitch super_ddf = {
3101 #ifndef MDASSEMBLE
3102 .examine_super = examine_super_ddf,
3103 .brief_examine_super = brief_examine_super_ddf,
3104 .detail_super = detail_super_ddf,
3105 .brief_detail_super = brief_detail_super_ddf,
3106 .validate_geometry = validate_geometry_ddf,
3107 #endif
3108 .match_home = match_home_ddf,
3109 .uuid_from_super= uuid_from_super_ddf,
3110 .getinfo_super = getinfo_super_ddf,
3111 .update_super = update_super_ddf,
3112
3113 .avail_size = avail_size_ddf,
3114
3115 .compare_super = compare_super_ddf,
3116
3117 .load_super = load_super_ddf,
3118 .init_super = init_zero_ddf,
3119 .store_super = store_zero_ddf,
3120 .free_super = free_super_ddf,
3121 .match_metadata_desc = match_metadata_desc_ddf,
3122
3123 .external = 1,
3124
3125 /* for mdmon */
3126 .open_new = ddf_open_new,
3127 .set_array_state= ddf_set_array_state,
3128 .set_disk = ddf_set_disk,
3129 .sync_metadata = ddf_sync_metadata,
3130 .process_update = ddf_process_update,
3131 .activate_spare = ddf_activate_spare,
3132
3133 };
3134
3135 /* Super_ddf_container is set by validate_geometry_ddf when given a
3136 * device that is not part of any array
3137 */
3138 static struct superswitch super_ddf_container = {
3139 #ifndef MDASSEMBLE
3140 .validate_geometry = validate_geometry_ddf_container,
3141 .write_init_super = write_init_super_ddf,
3142 #endif
3143
3144 .load_super = load_super_ddf,
3145 .init_super = init_super_ddf,
3146 .add_to_super = add_to_super_ddf,
3147 .getinfo_super = getinfo_super_ddf,
3148
3149 .free_super = free_super_ddf,
3150
3151 .container_content = container_content_ddf,
3152
3153 .external = 1,
3154 };
3155
3156 static struct superswitch super_ddf_bvd = {
3157 #ifndef MDASSEMBLE
3158 // .detail_super = detail_super_ddf_bvd,
3159 // .brief_detail_super = brief_detail_super_ddf_bvd,
3160 .validate_geometry = validate_geometry_ddf_bvd,
3161 .write_init_super = write_init_super_ddf,
3162 #endif
3163 .update_super = update_super_ddf,
3164 .init_super = init_super_ddf_bvd,
3165 .add_to_super = add_to_super_ddf_bvd,
3166 .getinfo_super = getinfo_super_ddf_bvd,
3167
3168 .load_super = load_super_ddf,
3169 .free_super = free_super_ddf,
3170 .match_metadata_desc = match_metadata_desc_ddf_bvd,
3171
3172 .external = 2,
3173 };
3174
3175 static struct superswitch super_ddf_svd = {
3176 #ifndef MDASSEMBLE
3177 // .detail_super = detail_super_ddf_svd,
3178 // .brief_detail_super = brief_detail_super_ddf_svd,
3179 .validate_geometry = validate_geometry_ddf_svd,
3180 #endif
3181 .update_super = update_super_ddf,
3182 .init_super = init_super_ddf,
3183
3184 .load_super = load_super_ddf,
3185 .free_super = free_super_ddf,
3186 .match_metadata_desc = match_metadata_desc_ddf_svd,
3187
3188 .external = 2,
3189 };
Unnamed repository; edit this file 'description' to name the repository.