]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super-ddf.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ddf: convert size paramter to sectors when calling avail_size_ddf
[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 0x16 /* Vendor unique layout */
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_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
87 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
88
89 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
90 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
91
92 /* Secondary RAID Level (SRL) */
93 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
94 #define DDF_2MIRRORED 0x01
95 #define DDF_2CONCAT 0x02
96 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
97
98 /* Magic numbers */
99 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
100 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
101 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
102 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
103 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
104 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
105 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
106 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
107 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
108 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
109
110 #define DDF_GUID_LEN 24
111 #define DDF_REVISION "01.00.00"
112
113 struct ddf_header {
114 __u32 magic; /* DDF_HEADER_MAGIC */
115 __u32 crc;
116 char guid[DDF_GUID_LEN];
117 char revision[8]; /* 01.00.00 */
118 __u32 seq; /* starts at '1' */
119 __u32 timestamp;
120 __u8 openflag;
121 __u8 foreignflag;
122 __u8 enforcegroups;
123 __u8 pad0; /* 0xff */
124 __u8 pad1[12]; /* 12 * 0xff */
125 /* 64 bytes so far */
126 __u8 header_ext[32]; /* reserved: fill with 0xff */
127 __u64 primary_lba;
128 __u64 secondary_lba;
129 __u8 type;
130 __u8 pad2[3]; /* 0xff */
131 __u32 workspace_len; /* sectors for vendor space -
132 * at least 32768(sectors) */
133 __u64 workspace_lba;
134 __u16 max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */
135 __u16 max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */
136 __u16 max_partitions; /* i.e. max num of configuration
137 record entries per disk */
138 __u16 config_record_len; /* 1 +ROUNDUP(max_primary_element_entries
139 *12/512) */
140 __u16 max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */
141 __u8 pad3[54]; /* 0xff */
142 /* 192 bytes so far */
143 __u32 controller_section_offset;
144 __u32 controller_section_length;
145 __u32 phys_section_offset;
146 __u32 phys_section_length;
147 __u32 virt_section_offset;
148 __u32 virt_section_length;
149 __u32 config_section_offset;
150 __u32 config_section_length;
151 __u32 data_section_offset;
152 __u32 data_section_length;
153 __u32 bbm_section_offset;
154 __u32 bbm_section_length;
155 __u32 diag_space_offset;
156 __u32 diag_space_length;
157 __u32 vendor_offset;
158 __u32 vendor_length;
159 /* 256 bytes so far */
160 __u8 pad4[256]; /* 0xff */
161 };
162
163 /* type field */
164 #define DDF_HEADER_ANCHOR 0x00
165 #define DDF_HEADER_PRIMARY 0x01
166 #define DDF_HEADER_SECONDARY 0x02
167
168 /* The content of the 'controller section' - global scope */
169 struct ddf_controller_data {
170 __u32 magic; /* DDF_CONTROLLER_MAGIC */
171 __u32 crc;
172 char guid[DDF_GUID_LEN];
173 struct controller_type {
174 __u16 vendor_id;
175 __u16 device_id;
176 __u16 sub_vendor_id;
177 __u16 sub_device_id;
178 } type;
179 char product_id[16];
180 __u8 pad[8]; /* 0xff */
181 __u8 vendor_data[448];
182 };
183
184 /* The content of phys_section - global scope */
185 struct phys_disk {
186 __u32 magic; /* DDF_PHYS_RECORDS_MAGIC */
187 __u32 crc;
188 __u16 used_pdes;
189 __u16 max_pdes;
190 __u8 pad[52];
191 struct phys_disk_entry {
192 char guid[DDF_GUID_LEN];
193 __u32 refnum;
194 __u16 type;
195 __u16 state;
196 __u64 config_size; /* DDF structures must be after here */
197 char path[18]; /* another horrible structure really */
198 __u8 pad[6];
199 } entries[0];
200 };
201
202 /* phys_disk_entry.type is a bitmap - bigendian remember */
203 #define DDF_Forced_PD_GUID 1
204 #define DDF_Active_in_VD 2
205 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
206 #define DDF_Spare 8 /* overrides Global_spare */
207 #define DDF_Foreign 16
208 #define DDF_Legacy 32 /* no DDF on this device */
209
210 #define DDF_Interface_mask 0xf00
211 #define DDF_Interface_SCSI 0x100
212 #define DDF_Interface_SAS 0x200
213 #define DDF_Interface_SATA 0x300
214 #define DDF_Interface_FC 0x400
215
216 /* phys_disk_entry.state is a bigendian bitmap */
217 #define DDF_Online 1
218 #define DDF_Failed 2 /* overrides 1,4,8 */
219 #define DDF_Rebuilding 4
220 #define DDF_Transition 8
221 #define DDF_SMART 16
222 #define DDF_ReadErrors 32
223 #define DDF_Missing 64
224
225 /* The content of the virt_section global scope */
226 struct virtual_disk {
227 __u32 magic; /* DDF_VIRT_RECORDS_MAGIC */
228 __u32 crc;
229 __u16 populated_vdes;
230 __u16 max_vdes;
231 __u8 pad[52];
232 struct virtual_entry {
233 char guid[DDF_GUID_LEN];
234 __u16 unit;
235 __u16 pad0; /* 0xffff */
236 __u16 guid_crc;
237 __u16 type;
238 __u8 state;
239 __u8 init_state;
240 __u8 pad1[14];
241 char name[16];
242 } entries[0];
243 };
244
245 /* virtual_entry.type is a bitmap - bigendian */
246 #define DDF_Shared 1
247 #define DDF_Enforce_Groups 2
248 #define DDF_Unicode 4
249 #define DDF_Owner_Valid 8
250
251 /* virtual_entry.state is a bigendian bitmap */
252 #define DDF_state_mask 0x7
253 #define DDF_state_optimal 0x0
254 #define DDF_state_degraded 0x1
255 #define DDF_state_deleted 0x2
256 #define DDF_state_missing 0x3
257 #define DDF_state_failed 0x4
258 #define DDF_state_part_optimal 0x5
259
260 #define DDF_state_morphing 0x8
261 #define DDF_state_inconsistent 0x10
262
263 /* virtual_entry.init_state is a bigendian bitmap */
264 #define DDF_initstate_mask 0x03
265 #define DDF_init_not 0x00
266 #define DDF_init_quick 0x01 /* initialisation is progress.
267 * i.e. 'state_inconsistent' */
268 #define DDF_init_full 0x02
269
270 #define DDF_access_mask 0xc0
271 #define DDF_access_rw 0x00
272 #define DDF_access_ro 0x80
273 #define DDF_access_blocked 0xc0
274
275 /* The content of the config_section - local scope
276 * It has multiple records each config_record_len sectors
277 * They can be vd_config or spare_assign
278 */
279
280 struct vd_config {
281 __u32 magic; /* DDF_VD_CONF_MAGIC */
282 __u32 crc;
283 char guid[DDF_GUID_LEN];
284 __u32 timestamp;
285 __u32 seqnum;
286 __u8 pad0[24];
287 __u16 prim_elmnt_count;
288 __u8 chunk_shift; /* 0 == 512, 1==1024 etc */
289 __u8 prl;
290 __u8 rlq;
291 __u8 sec_elmnt_count;
292 __u8 sec_elmnt_seq;
293 __u8 srl;
294 __u64 blocks; /* blocks per component could be different
295 * on different component devices...(only
296 * for concat I hope) */
297 __u64 array_blocks; /* blocks in array */
298 __u8 pad1[8];
299 __u32 spare_refs[8];
300 __u8 cache_pol[8];
301 __u8 bg_rate;
302 __u8 pad2[3];
303 __u8 pad3[52];
304 __u8 pad4[192];
305 __u8 v0[32]; /* reserved- 0xff */
306 __u8 v1[32]; /* reserved- 0xff */
307 __u8 v2[16]; /* reserved- 0xff */
308 __u8 v3[16]; /* reserved- 0xff */
309 __u8 vendor[32];
310 __u32 phys_refnum[0]; /* refnum of each disk in sequence */
311 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
312 bvd are always the same size */
313 };
314
315 /* vd_config.cache_pol[7] is a bitmap */
316 #define DDF_cache_writeback 1 /* else writethrough */
317 #define DDF_cache_wadaptive 2 /* only applies if writeback */
318 #define DDF_cache_readahead 4
319 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
320 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
321 #define DDF_cache_wallowed 32 /* enable write caching */
322 #define DDF_cache_rallowed 64 /* enable read caching */
323
324 struct spare_assign {
325 __u32 magic; /* DDF_SPARE_ASSIGN_MAGIC */
326 __u32 crc;
327 __u32 timestamp;
328 __u8 reserved[7];
329 __u8 type;
330 __u16 populated; /* SAEs used */
331 __u16 max; /* max SAEs */
332 __u8 pad[8];
333 struct spare_assign_entry {
334 char guid[DDF_GUID_LEN];
335 __u16 secondary_element;
336 __u8 pad[6];
337 } spare_ents[0];
338 };
339 /* spare_assign.type is a bitmap */
340 #define DDF_spare_dedicated 0x1 /* else global */
341 #define DDF_spare_revertible 0x2 /* else committable */
342 #define DDF_spare_active 0x4 /* else not active */
343 #define DDF_spare_affinity 0x8 /* enclosure affinity */
344
345 /* The data_section contents - local scope */
346 struct disk_data {
347 __u32 magic; /* DDF_PHYS_DATA_MAGIC */
348 __u32 crc;
349 char guid[DDF_GUID_LEN];
350 __u32 refnum; /* crc of some magic drive data ... */
351 __u8 forced_ref; /* set when above was not result of magic */
352 __u8 forced_guid; /* set if guid was forced rather than magic */
353 __u8 vendor[32];
354 __u8 pad[442];
355 };
356
357 /* bbm_section content */
358 struct bad_block_log {
359 __u32 magic;
360 __u32 crc;
361 __u16 entry_count;
362 __u32 spare_count;
363 __u8 pad[10];
364 __u64 first_spare;
365 struct mapped_block {
366 __u64 defective_start;
367 __u32 replacement_start;
368 __u16 remap_count;
369 __u8 pad[2];
370 } entries[0];
371 };
372
373 /* Struct for internally holding ddf structures */
374 /* The DDF structure stored on each device is potentially
375 * quite different, as some data is global and some is local.
376 * The global data is:
377 * - ddf header
378 * - controller_data
379 * - Physical disk records
380 * - Virtual disk records
381 * The local data is:
382 * - Configuration records
383 * - Physical Disk data section
384 * ( and Bad block and vendor which I don't care about yet).
385 *
386 * The local data is parsed into separate lists as it is read
387 * and reconstructed for writing. This means that we only need
388 * to make config changes once and they are automatically
389 * propagated to all devices.
390 * Note that the ddf_super has space of the conf and disk data
391 * for this disk and also for a list of all such data.
392 * The list is only used for the superblock that is being
393 * built in Create or Assemble to describe the whole array.
394 */
395 struct ddf_super {
396 struct ddf_header anchor, primary, secondary, *active;
397 struct ddf_controller_data controller;
398 struct phys_disk *phys;
399 struct virtual_disk *virt;
400 int pdsize, vdsize;
401 int max_part, mppe, conf_rec_len;
402 struct vcl {
403 struct vcl *next;
404 __u64 *lba_offset; /* location in 'conf' of
405 * the lba table */
406 struct vd_config conf;
407 } *conflist, *newconf;
408 int conf_num; /* Index into 'virt' of entry matching 'newconf' */
409 struct dl {
410 struct dl *next;
411 struct disk_data disk;
412 int major, minor;
413 char *devname;
414 int fd;
415 int pdnum; /* index in ->phys */
416 struct spare_assign *spare;
417 struct vcl *vlist[0]; /* max_part in size */
418 } *dlist;
419 };
420
421 #ifndef offsetof
422 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
423 #endif
424
425 extern struct superswitch super_ddf_container, super_ddf_bvd, super_ddf;
426
427 static int calc_crc(void *buf, int len)
428 {
429 /* crcs are always at the same place as in the ddf_header */
430 struct ddf_header *ddf = buf;
431 __u32 oldcrc = ddf->crc;
432 __u32 newcrc;
433 ddf->crc = 0xffffffff;
434
435 newcrc = crc32(0, buf, len);
436 ddf->crc = oldcrc;
437 return newcrc;
438 }
439
440 static int load_ddf_header(int fd, unsigned long long lba,
441 unsigned long long size,
442 int type,
443 struct ddf_header *hdr, struct ddf_header *anchor)
444 {
445 /* read a ddf header (primary or secondary) from fd/lba
446 * and check that it is consistent with anchor
447 * Need to check:
448 * magic, crc, guid, rev, and LBA's header_type, and
449 * everything after header_type must be the same
450 */
451 if (lba >= size-1)
452 return 0;
453
454 if (lseek64(fd, lba<<9, 0) < 0)
455 return 0;
456
457 if (read(fd, hdr, 512) != 512)
458 return 0;
459
460 if (hdr->magic != DDF_HEADER_MAGIC)
461 return 0;
462 if (calc_crc(hdr, 512) != hdr->crc)
463 return 0;
464 if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
465 memcmp(anchor->revision, hdr->revision, 8) != 0 ||
466 anchor->primary_lba != hdr->primary_lba ||
467 anchor->secondary_lba != hdr->secondary_lba ||
468 hdr->type != type ||
469 memcmp(anchor->pad2, hdr->pad2, 512 -
470 offsetof(struct ddf_header, pad2)) != 0)
471 return 0;
472
473 /* Looks good enough to me... */
474 return 1;
475 }
476
477 static void *load_section(int fd, struct ddf_super *super, void *buf,
478 __u32 offset_be, __u32 len_be, int check)
479 {
480 unsigned long long offset = __be32_to_cpu(offset_be);
481 unsigned long long len = __be32_to_cpu(len_be);
482 int dofree = (buf == NULL);
483
484 if (check)
485 if (len != 2 && len != 8 && len != 32
486 && len != 128 && len != 512)
487 return NULL;
488
489 if (len > 1024)
490 return NULL;
491 if (buf) {
492 /* All pre-allocated sections are a single block */
493 if (len != 1)
494 return NULL;
495 } else
496 buf = malloc(len<<9);
497 if (!buf)
498 return NULL;
499
500 if (super->active->type == 1)
501 offset += __be64_to_cpu(super->active->primary_lba);
502 else
503 offset += __be64_to_cpu(super->active->secondary_lba);
504
505 if (lseek64(fd, offset<<9, 0) != (offset<<9)) {
506 if (dofree)
507 free(buf);
508 return NULL;
509 }
510 if (read(fd, buf, len<<9) != (len<<9)) {
511 if (dofree)
512 free(buf);
513 return NULL;
514 }
515 return buf;
516 }
517
518 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
519 {
520 unsigned long long dsize;
521
522 get_dev_size(fd, NULL, &dsize);
523
524 if (lseek64(fd, dsize-512, 0) < 0) {
525 if (devname)
526 fprintf(stderr,
527 Name": Cannot seek to anchor block on %s: %s\n",
528 devname, strerror(errno));
529 return 1;
530 }
531 if (read(fd, &super->anchor, 512) != 512) {
532 if (devname)
533 fprintf(stderr,
534 Name ": Cannot read anchor block on %s: %s\n",
535 devname, strerror(errno));
536 return 1;
537 }
538 if (super->anchor.magic != DDF_HEADER_MAGIC) {
539 if (devname)
540 fprintf(stderr, Name ": no DDF anchor found on %s\n",
541 devname);
542 return 2;
543 }
544 if (calc_crc(&super->anchor, 512) != super->anchor.crc) {
545 if (devname)
546 fprintf(stderr, Name ": bad CRC on anchor on %s\n",
547 devname);
548 return 2;
549 }
550 if (memcmp(super->anchor.revision, DDF_REVISION, 8) != 0) {
551 if (devname)
552 fprintf(stderr, Name ": can only support super revision"
553 " %.8s, not %.8s on %s\n",
554 DDF_REVISION, super->anchor.revision, devname);
555 return 2;
556 }
557 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba),
558 dsize >> 9, 1,
559 &super->primary, &super->anchor) == 0) {
560 if (devname)
561 fprintf(stderr,
562 Name ": Failed to load primary DDF header "
563 "on %s\n", devname);
564 return 2;
565 }
566 super->active = &super->primary;
567 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.secondary_lba),
568 dsize >> 9, 2,
569 &super->secondary, &super->anchor)) {
570 if ((__be32_to_cpu(super->primary.seq)
571 < __be32_to_cpu(super->secondary.seq) &&
572 !super->secondary.openflag)
573 || (__be32_to_cpu(super->primary.seq)
574 == __be32_to_cpu(super->secondary.seq) &&
575 super->primary.openflag && !super->secondary.openflag)
576 )
577 super->active = &super->secondary;
578 }
579 return 0;
580 }
581
582 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
583 {
584 void *ok;
585 ok = load_section(fd, super, &super->controller,
586 super->active->controller_section_offset,
587 super->active->controller_section_length,
588 0);
589 super->phys = load_section(fd, super, NULL,
590 super->active->phys_section_offset,
591 super->active->phys_section_length,
592 1);
593 super->pdsize = __be32_to_cpu(super->active->phys_section_length) * 512;
594
595 super->virt = load_section(fd, super, NULL,
596 super->active->virt_section_offset,
597 super->active->virt_section_length,
598 1);
599 super->vdsize = __be32_to_cpu(super->active->virt_section_length) * 512;
600 if (!ok ||
601 !super->phys ||
602 !super->virt) {
603 free(super->phys);
604 free(super->virt);
605 super->phys = NULL;
606 super->virt = NULL;
607 return 2;
608 }
609 super->conflist = NULL;
610 super->dlist = NULL;
611
612 super->max_part = __be16_to_cpu(super->active->max_partitions);
613 super->mppe = __be16_to_cpu(super->active->max_primary_element_entries);
614 super->conf_rec_len = __be16_to_cpu(super->active->config_record_len);
615 return 0;
616 }
617
618 static int load_ddf_local(int fd, struct ddf_super *super,
619 char *devname, int keep)
620 {
621 struct dl *dl;
622 struct stat stb;
623 char *conf;
624 int i;
625 int vnum;
626
627 /* First the local disk info */
628 dl = malloc(sizeof(*dl) +
629 (super->max_part) * sizeof(dl->vlist[0]));
630
631 load_section(fd, super, &dl->disk,
632 super->active->data_section_offset,
633 super->active->data_section_length,
634 0);
635 dl->devname = devname ? strdup(devname) : NULL;
636
637 fstat(fd, &stb);
638 dl->major = major(stb.st_rdev);
639 dl->minor = minor(stb.st_rdev);
640 dl->next = super->dlist;
641 dl->fd = keep ? fd : -1;
642 dl->spare = NULL;
643 for (i=0 ; i < super->max_part ; i++)
644 dl->vlist[i] = NULL;
645 super->dlist = dl;
646 dl->pdnum = 0;
647 for (i=0; i < __be16_to_cpu(super->active->max_pd_entries); i++)
648 if (memcmp(super->phys->entries[i].guid,
649 dl->disk.guid, DDF_GUID_LEN) == 0)
650 dl->pdnum = i;
651
652
653 /* Now the config list. */
654 /* 'conf' is an array of config entries, some of which are
655 * probably invalid. Those which are good need to be copied into
656 * the conflist
657 */
658
659 conf = load_section(fd, super, NULL,
660 super->active->config_section_offset,
661 super->active->config_section_length,
662 0);
663
664 vnum = 0;
665 for (i = 0;
666 i < __be32_to_cpu(super->active->config_section_length);
667 i += super->conf_rec_len) {
668 struct vd_config *vd =
669 (struct vd_config *)((char*)conf + i*512);
670 struct vcl *vcl;
671
672 if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) {
673 if (dl->spare)
674 continue;
675 dl->spare = malloc(super->conf_rec_len*512);
676 memcpy(dl->spare, vd, super->conf_rec_len*512);
677 continue;
678 }
679 if (vd->magic != DDF_VD_CONF_MAGIC)
680 continue;
681 for (vcl = super->conflist; vcl; vcl = vcl->next) {
682 if (memcmp(vcl->conf.guid,
683 vd->guid, DDF_GUID_LEN) == 0)
684 break;
685 }
686
687 if (vcl) {
688 dl->vlist[vnum++] = vcl;
689 if (__be32_to_cpu(vd->seqnum) <=
690 __be32_to_cpu(vcl->conf.seqnum))
691 continue;
692 } else {
693 vcl = malloc(super->conf_rec_len*512 +
694 offsetof(struct vcl, conf));
695 vcl->next = super->conflist;
696 super->conflist = vcl;
697 dl->vlist[vnum++] = vcl;
698 }
699 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
700 vcl->lba_offset = (__u64*)
701 &vcl->conf.phys_refnum[super->mppe];
702 }
703 free(conf);
704
705 return 0;
706 }
707
708 #ifndef MDASSEMBLE
709 static int load_super_ddf_all(struct supertype *st, int fd,
710 void **sbp, char *devname, int keep_fd);
711 #endif
712 static int load_super_ddf(struct supertype *st, int fd,
713 char *devname)
714 {
715 unsigned long long dsize;
716 struct ddf_super *super;
717 int rv;
718
719 #ifndef MDASSEMBLE
720 if (load_super_ddf_all(st, fd, &st->sb, devname, 1) == 0)
721 return 0;
722 #endif
723
724 if (get_dev_size(fd, devname, &dsize) == 0)
725 return 1;
726
727 /* 32M is a lower bound */
728 if (dsize <= 32*1024*1024) {
729 if (devname) {
730 fprintf(stderr,
731 Name ": %s is too small for ddf: "
732 "size is %llu sectors.\n",
733 devname, dsize>>9);
734 return 1;
735 }
736 }
737 if (dsize & 511) {
738 if (devname) {
739 fprintf(stderr,
740 Name ": %s is an odd size for ddf: "
741 "size is %llu bytes.\n",
742 devname, dsize);
743 return 1;
744 }
745 }
746
747 super = malloc(sizeof(*super));
748 if (!super) {
749 fprintf(stderr, Name ": malloc of %zu failed.\n",
750 sizeof(*super));
751 return 1;
752 }
753 memset(super, 0, sizeof(*super));
754
755 rv = load_ddf_headers(fd, super, devname);
756 if (rv) {
757 free(super);
758 return rv;
759 }
760
761 /* Have valid headers and have chosen the best. Let's read in the rest*/
762
763 rv = load_ddf_global(fd, super, devname);
764
765 if (rv) {
766 if (devname)
767 fprintf(stderr,
768 Name ": Failed to load all information "
769 "sections on %s\n", devname);
770 free(super);
771 return rv;
772 }
773
774 load_ddf_local(fd, super, devname, 0);
775
776 /* Should possibly check the sections .... */
777
778 st->sb = super;
779 if (st->ss == NULL) {
780 st->ss = &super_ddf;
781 st->minor_version = 0;
782 st->max_devs = 512;
783 }
784 return 0;
785
786 }
787
788 static void free_super_ddf(struct supertype *st)
789 {
790 struct ddf_super *ddf = st->sb;
791 if (ddf == NULL)
792 return;
793 free(ddf->phys);
794 free(ddf->virt);
795 while (ddf->conflist) {
796 struct vcl *v = ddf->conflist;
797 ddf->conflist = v->next;
798 free(v);
799 }
800 while (ddf->dlist) {
801 struct dl *d = ddf->dlist;
802 ddf->dlist = d->next;
803 if (d->fd >= 0)
804 close(d->fd);
805 if (d->spare)
806 free(d->spare);
807 free(d);
808 }
809 free(ddf);
810 st->sb = NULL;
811 }
812
813 static struct supertype *match_metadata_desc_ddf(char *arg)
814 {
815 /* 'ddf' only support containers */
816 struct supertype *st;
817 if (strcmp(arg, "ddf") != 0 &&
818 strcmp(arg, "default") != 0
819 )
820 return NULL;
821
822 st = malloc(sizeof(*st));
823 st->ss = &super_ddf;
824 st->max_devs = 512;
825 st->minor_version = 0;
826 st->sb = NULL;
827 return st;
828 }
829
830 static struct supertype *match_metadata_desc_ddf_bvd(char *arg)
831 {
832 struct supertype *st;
833 if (strcmp(arg, "ddf/bvd") != 0 &&
834 strcmp(arg, "bvd") != 0 &&
835 strcmp(arg, "default") != 0
836 )
837 return NULL;
838
839 st = malloc(sizeof(*st));
840 st->ss = &super_ddf_bvd;
841 st->max_devs = 512;
842 st->minor_version = 0;
843 st->sb = NULL;
844 return st;
845 }
846 static struct supertype *match_metadata_desc_ddf_svd(char *arg)
847 {
848 struct supertype *st;
849 if (strcmp(arg, "ddf/svd") != 0 &&
850 strcmp(arg, "svd") != 0 &&
851 strcmp(arg, "default") != 0
852 )
853 return NULL;
854
855 st = malloc(sizeof(*st));
856 st->ss = &super_ddf_svd;
857 st->max_devs = 512;
858 st->minor_version = 0;
859 st->sb = NULL;
860 return st;
861 }
862
863 #ifndef MDASSEMBLE
864
865 static mapping_t ddf_state[] = {
866 { "Optimal", 0},
867 { "Degraded", 1},
868 { "Deleted", 2},
869 { "Missing", 3},
870 { "Failed", 4},
871 { "Partially Optimal", 5},
872 { "-reserved-", 6},
873 { "-reserved-", 7},
874 { NULL, 0}
875 };
876
877 static mapping_t ddf_init_state[] = {
878 { "Not Initialised", 0},
879 { "QuickInit in Progress", 1},
880 { "Fully Initialised", 2},
881 { "*UNKNOWN*", 3},
882 { NULL, 0}
883 };
884 static mapping_t ddf_access[] = {
885 { "Read/Write", 0},
886 { "Reserved", 1},
887 { "Read Only", 2},
888 { "Blocked (no access)", 3},
889 { NULL ,0}
890 };
891
892 static mapping_t ddf_level[] = {
893 { "RAID0", DDF_RAID0},
894 { "RAID1", DDF_RAID1},
895 { "RAID3", DDF_RAID3},
896 { "RAID4", DDF_RAID4},
897 { "RAID5", DDF_RAID5},
898 { "RAID1E",DDF_RAID1E},
899 { "JBOD", DDF_JBOD},
900 { "CONCAT",DDF_CONCAT},
901 { "RAID5E",DDF_RAID5E},
902 { "RAID5EE",DDF_RAID5EE},
903 { "RAID6", DDF_RAID6},
904 { NULL, 0}
905 };
906 static mapping_t ddf_sec_level[] = {
907 { "Striped", DDF_2STRIPED},
908 { "Mirrored", DDF_2MIRRORED},
909 { "Concat", DDF_2CONCAT},
910 { "Spanned", DDF_2SPANNED},
911 { NULL, 0}
912 };
913 #endif
914
915 struct num_mapping {
916 int num1, num2;
917 };
918 static struct num_mapping ddf_level_num[] = {
919 { DDF_RAID0, 0 },
920 { DDF_RAID1, 1 },
921 { DDF_RAID3, LEVEL_UNSUPPORTED },
922 { DDF_RAID4, 4 },
923 { DDF_RAID5, 5 },
924 { DDF_RAID1E, LEVEL_UNSUPPORTED },
925 { DDF_JBOD, LEVEL_UNSUPPORTED },
926 { DDF_CONCAT, LEVEL_LINEAR },
927 { DDF_RAID5E, LEVEL_UNSUPPORTED },
928 { DDF_RAID5EE, LEVEL_UNSUPPORTED },
929 { DDF_RAID6, 6},
930 { MAXINT, MAXINT }
931 };
932
933 static int map_num1(struct num_mapping *map, int num)
934 {
935 int i;
936 for (i=0 ; map[i].num1 != MAXINT; i++)
937 if (map[i].num1 == num)
938 break;
939 return map[i].num2;
940 }
941
942 #ifndef MDASSEMBLE
943 static void print_guid(char *guid, int tstamp)
944 {
945 /* A GUIDs are part (or all) ASCII and part binary.
946 * They tend to be space padded.
947 * We ignore trailing spaces and print numbers
948 * <0x20 and >=0x7f as \xXX
949 * Some GUIDs have a time stamp in bytes 16-19.
950 * We print that if appropriate
951 */
952 int l = DDF_GUID_LEN;
953 int i;
954 while (l && guid[l-1] == ' ')
955 l--;
956 for (i=0 ; i<l ; i++) {
957 if (guid[i] >= 0x20 && guid[i] < 0x7f)
958 fputc(guid[i], stdout);
959 else
960 fprintf(stdout, "\\x%02x", guid[i]&255);
961 }
962 if (tstamp) {
963 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
964 char tbuf[100];
965 struct tm *tm;
966 tm = localtime(&then);
967 strftime(tbuf, 100, " (%D %T)",tm);
968 fputs(tbuf, stdout);
969 }
970 }
971
972 static void examine_vd(int n, struct ddf_super *sb, char *guid)
973 {
974 int crl = sb->conf_rec_len;
975 struct vcl *vcl;
976
977 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
978 struct vd_config *vc = &vcl->conf;
979
980 if (calc_crc(vc, crl*512) != vc->crc)
981 continue;
982 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
983 continue;
984
985 /* Ok, we know about this VD, let's give more details */
986 printf(" Raid Devices[%d] : %d\n", n,
987 __be16_to_cpu(vc->prim_elmnt_count));
988 printf(" Chunk Size[%d] : %d sectors\n", n,
989 1 << vc->chunk_shift);
990 printf(" Raid Level[%d] : %s\n", n,
991 map_num(ddf_level, vc->prl)?:"-unknown-");
992 if (vc->sec_elmnt_count != 1) {
993 printf(" Secondary Position[%d] : %d of %d\n", n,
994 vc->sec_elmnt_seq, vc->sec_elmnt_count);
995 printf(" Secondary Level[%d] : %s\n", n,
996 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
997 }
998 printf(" Device Size[%d] : %llu\n", n,
999 __be64_to_cpu(vc->blocks)/2);
1000 printf(" Array Size[%d] : %llu\n", n,
1001 __be64_to_cpu(vc->array_blocks)/2);
1002 }
1003 }
1004
1005 static void examine_vds(struct ddf_super *sb)
1006 {
1007 int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1008 int i;
1009 printf(" Virtual Disks : %d\n", cnt);
1010
1011 for (i=0; i<cnt; i++) {
1012 struct virtual_entry *ve = &sb->virt->entries[i];
1013 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1014 printf("\n");
1015 printf(" unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1016 printf(" state[%d] : %s, %s%s\n", i,
1017 map_num(ddf_state, ve->state & 7),
1018 (ve->state & 8) ? "Morphing, ": "",
1019 (ve->state & 16)? "Not Consistent" : "Consistent");
1020 printf(" init state[%d] : %s\n", i,
1021 map_num(ddf_init_state, ve->init_state&3));
1022 printf(" access[%d] : %s\n", i,
1023 map_num(ddf_access, (ve->init_state>>6) & 3));
1024 printf(" Name[%d] : %.16s\n", i, ve->name);
1025 examine_vd(i, sb, ve->guid);
1026 }
1027 if (cnt) printf("\n");
1028 }
1029
1030 static void examine_pds(struct ddf_super *sb)
1031 {
1032 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1033 int i;
1034 struct dl *dl;
1035 printf(" Physical Disks : %d\n", cnt);
1036
1037 for (i=0 ; i<cnt ; i++) {
1038 struct phys_disk_entry *pd = &sb->phys->entries[i];
1039 int type = __be16_to_cpu(pd->type);
1040 int state = __be16_to_cpu(pd->state);
1041
1042 printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1043 printf("\n");
1044 printf(" ref[%d] : %08x\n", i,
1045 __be32_to_cpu(pd->refnum));
1046 printf(" mode[%d] : %s%s%s%s%s\n", i,
1047 (type&2) ? "active":"",
1048 (type&4) ? "Global Spare":"",
1049 (type&8) ? "spare" : "",
1050 (type&16)? ", foreign" : "",
1051 (type&32)? "pass-through" : "");
1052 printf(" state[%d] : %s%s%s%s%s%s%s\n", i,
1053 (state&1)? "Online": "Offline",
1054 (state&2)? ", Failed": "",
1055 (state&4)? ", Rebuilding": "",
1056 (state&8)? ", in-transition": "",
1057 (state&16)? ", SMART errors": "",
1058 (state&32)? ", Unrecovered Read Errors": "",
1059 (state&64)? ", Missing" : "");
1060 printf(" Avail Size[%d] : %llu K\n", i,
1061 __be64_to_cpu(pd->config_size)>>1);
1062 for (dl = sb->dlist; dl ; dl = dl->next) {
1063 if (dl->disk.refnum == pd->refnum) {
1064 char *dv = map_dev(dl->major, dl->minor, 0);
1065 if (dv)
1066 printf(" Device[%d] : %s\n",
1067 i, dv);
1068 }
1069 }
1070 printf("\n");
1071 }
1072 }
1073
1074 static void examine_super_ddf(struct supertype *st, char *homehost)
1075 {
1076 struct ddf_super *sb = st->sb;
1077
1078 printf(" Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1079 printf(" Version : %.8s\n", sb->anchor.revision);
1080 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1081 printf("\n");
1082 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1083 printf("\n");
1084 printf(" Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1085 printf(" Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1086 ?"yes" : "no");
1087 examine_vds(sb);
1088 examine_pds(sb);
1089 }
1090
1091 static void brief_examine_super_ddf(struct supertype *st)
1092 {
1093 /* We just write a generic DDF ARRAY entry
1094 * The uuid is all hex, 6 groups of 4 bytes
1095 */
1096 struct ddf_super *ddf = st->sb;
1097 int i;
1098 printf("ARRAY /dev/ddf UUID=");
1099 for (i = 0; i < DDF_GUID_LEN; i++) {
1100 printf("%02x", ddf->anchor.guid[i]);
1101 if ((i&3) == 0 && i != 0)
1102 printf(":");
1103 }
1104 printf("\n");
1105 }
1106
1107 static void detail_super_ddf(struct supertype *st, char *homehost)
1108 {
1109 /* FIXME later
1110 * Could print DDF GUID
1111 * Need to find which array
1112 * If whole, briefly list all arrays
1113 * If one, give name
1114 */
1115 }
1116
1117 static void brief_detail_super_ddf(struct supertype *st)
1118 {
1119 /* FIXME I really need to know which array we are detailing.
1120 * Can that be stored in ddf_super??
1121 */
1122 // struct ddf_super *ddf = st->sb;
1123 }
1124
1125
1126 #endif
1127
1128 static int match_home_ddf(struct supertype *st, char *homehost)
1129 {
1130 /* It matches 'this' host if the controller is a
1131 * Linux-MD controller with vendor_data matching
1132 * the hostname
1133 */
1134 struct ddf_super *ddf = st->sb;
1135 int len = strlen(homehost);
1136
1137 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1138 len < sizeof(ddf->controller.vendor_data) &&
1139 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1140 ddf->controller.vendor_data[len] == 0);
1141 }
1142
1143 static struct vd_config *find_vdcr(struct ddf_super *ddf, int inst)
1144 {
1145 struct vcl *v;
1146 if (inst < 0 || inst > __be16_to_cpu(ddf->virt->populated_vdes))
1147 return NULL;
1148 for (v = ddf->conflist; v; v = v->next)
1149 if (memcmp(v->conf.guid,
1150 ddf->virt->entries[inst].guid,
1151 DDF_GUID_LEN) == 0)
1152 return &v->conf;
1153 return NULL;
1154 }
1155
1156 static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1157 {
1158 /* Find the entry in phys_disk which has the given refnum
1159 * and return it's index
1160 */
1161 int i;
1162 for (i=0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1163 if (ddf->phys->entries[i].refnum == phys_refnum)
1164 return i;
1165 return -1;
1166 }
1167
1168 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1169 {
1170 /* The uuid returned here is used for:
1171 * uuid to put into bitmap file (Create, Grow)
1172 * uuid for backup header when saving critical section (Grow)
1173 * comparing uuids when re-adding a device into an array
1174 * For each of these we can make do with a truncated
1175 * or hashed uuid rather than the original, as long as
1176 * everyone agrees.
1177 * In each case the uuid required is that of the data-array,
1178 * not the device-set.
1179 * In the case of SVD we assume the BVD is of interest,
1180 * though that might be the case if a bitmap were made for
1181 * a mirrored SVD - worry about that later.
1182 * So we need to find the VD configuration record for the
1183 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1184 * The first 16 bytes of the sha1 of these is used.
1185 */
1186 struct ddf_super *ddf = st->sb;
1187 struct vd_config *vd = find_vdcr(ddf, ddf->conf_num);
1188
1189 if (!vd)
1190 memset(uuid, 0, sizeof (uuid));
1191 else {
1192 char buf[20];
1193 struct sha1_ctx ctx;
1194 sha1_init_ctx(&ctx);
1195 sha1_process_bytes(&vd->guid, DDF_GUID_LEN, &ctx);
1196 if (vd->sec_elmnt_count > 1)
1197 sha1_process_bytes(&vd->sec_elmnt_seq, 1, &ctx);
1198 sha1_finish_ctx(&ctx, buf);
1199 memcpy(uuid, buf, sizeof(uuid));
1200 }
1201 }
1202
1203 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info)
1204 {
1205 struct ddf_super *ddf = st->sb;
1206 int i;
1207
1208 info->array.major_version = 1000;
1209 info->array.minor_version = 0; /* FIXME use ddf->revision somehow */
1210 info->array.patch_version = 0;
1211 info->array.raid_disks = __be16_to_cpu(ddf->phys->used_pdes);
1212 info->array.level = LEVEL_CONTAINER;
1213 info->array.layout = 0;
1214 info->array.md_minor = -1;
1215 info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)
1216 (ddf->anchor.guid+16));
1217 info->array.utime = 0;
1218 info->array.chunk_size = 0;
1219
1220 // info->data_offset = ???;
1221 // info->component_size = ???;
1222
1223 info->disk.major = 0;
1224 info->disk.minor = 0;
1225 if (ddf->dlist) {
1226 info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1227 info->disk.raid_disk = -1;
1228 for (i = 0; i < __be16_to_cpu(ddf->phys->max_pdes) ; i++)
1229 if (ddf->phys->entries[i].refnum ==
1230 ddf->dlist->disk.refnum) {
1231 info->disk.raid_disk = i;
1232 break;
1233 }
1234 } else {
1235 info->disk.number = -1;
1236 // info->disk.raid_disk = find refnum in the table and use index;
1237 }
1238 info->disk.state = (1 << MD_DISK_SYNC);
1239
1240 info->reshape_active = 0;
1241
1242 strcpy(info->text_version, "ddf");
1243
1244 // uuid_from_super_ddf(info->uuid, sbv);
1245
1246 // info->name[] ?? ;
1247 }
1248
1249 static void getinfo_super_n_container(struct supertype *st, struct mdinfo *info)
1250 {
1251 /* just need offset and size */
1252 struct ddf_super *ddf = st->sb;
1253 int n = info->disk.number;
1254
1255 info->data_offset = __be64_to_cpu(ddf->phys->entries[n].config_size);
1256 info->component_size = 32*1024*1024 / 512;
1257 }
1258
1259 static int rlq_to_layout(int rlq, int prl, int raiddisks);
1260
1261 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info)
1262 {
1263 struct ddf_super *ddf = st->sb;
1264 struct vd_config *vd = find_vdcr(ddf, info->container_member);
1265
1266 /* FIXME this returns BVD info - what if we want SVD ?? */
1267
1268 info->array.major_version = 1000;
1269 info->array.minor_version = 0; /* FIXME use ddf->revision somehow */
1270 info->array.patch_version = 0;
1271 info->array.raid_disks = __be16_to_cpu(vd->prim_elmnt_count);
1272 info->array.level = map_num1(ddf_level_num, vd->prl);
1273 info->array.layout = rlq_to_layout(vd->rlq, vd->prl,
1274 info->array.raid_disks);
1275 info->array.md_minor = -1;
1276 info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)(vd->guid+16));
1277 info->array.utime = DECADE + __be32_to_cpu(vd->timestamp);
1278 info->array.chunk_size = 512 << vd->chunk_shift;
1279
1280 // info->data_offset = ???;
1281 // info->component_size = ???;
1282
1283 info->disk.major = 0;
1284 info->disk.minor = 0;
1285 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1286 // info->disk.raid_disk = find refnum in the table and use index;
1287 // info->disk.state = ???;
1288
1289 info->resync_start = 0;
1290 if (!(ddf->virt->entries[info->container_member].state
1291 & DDF_state_inconsistent) &&
1292 (ddf->virt->entries[info->container_member].init_state
1293 & DDF_initstate_mask)
1294 == DDF_init_full)
1295 info->resync_start = ~0ULL;
1296
1297 uuid_from_super_ddf(st, info->uuid);
1298
1299 sprintf(info->text_version, "/%s/%d",
1300 devnum2devname(st->container_dev),
1301 info->container_member);
1302
1303 // info->name[] ?? ;
1304 }
1305
1306 static void getinfo_super_n_bvd(struct supertype *st, struct mdinfo *info)
1307 {
1308 /* Find the particular details for info->disk.raid_disk.
1309 * This includes data_offset, component_size,
1310 */
1311 struct ddf_super *ddf = st->sb;
1312 __u64 *lba_offset = ddf->newconf->lba_offset;
1313 struct vd_config *conf = &ddf->newconf->conf;
1314 info->data_offset = __be64_to_cpu(lba_offset[info->disk.raid_disk]);
1315 info->component_size = __be64_to_cpu(conf->blocks);
1316 }
1317
1318 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1319 char *update,
1320 char *devname, int verbose,
1321 int uuid_set, char *homehost)
1322 {
1323 /* For 'assemble' and 'force' we need to return non-zero if any
1324 * change was made. For others, the return value is ignored.
1325 * Update options are:
1326 * force-one : This device looks a bit old but needs to be included,
1327 * update age info appropriately.
1328 * assemble: clear any 'faulty' flag to allow this device to
1329 * be assembled.
1330 * force-array: Array is degraded but being forced, mark it clean
1331 * if that will be needed to assemble it.
1332 *
1333 * newdev: not used ????
1334 * grow: Array has gained a new device - this is currently for
1335 * linear only
1336 * resync: mark as dirty so a resync will happen.
1337 * uuid: Change the uuid of the array to match watch is given
1338 * homehost: update the recorded homehost
1339 * name: update the name - preserving the homehost
1340 * _reshape_progress: record new reshape_progress position.
1341 *
1342 * Following are not relevant for this version:
1343 * sparc2.2 : update from old dodgey metadata
1344 * super-minor: change the preferred_minor number
1345 * summaries: update redundant counters.
1346 */
1347 int rv = 0;
1348 // struct ddf_super *ddf = st->sb;
1349 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1350 // struct virtual_entry *ve = find_ve(ddf);
1351
1352
1353 /* we don't need to handle "force-*" or "assemble" as
1354 * there is no need to 'trick' the kernel. We the metadata is
1355 * first updated to activate the array, all the implied modifications
1356 * will just happen.
1357 */
1358
1359 if (strcmp(update, "grow") == 0) {
1360 /* FIXME */
1361 }
1362 if (strcmp(update, "resync") == 0) {
1363 // info->resync_checkpoint = 0;
1364 }
1365 /* We ignore UUID updates as they make even less sense
1366 * with DDF
1367 */
1368 if (strcmp(update, "homehost") == 0) {
1369 /* homehost is stored in controller->vendor_data,
1370 * or it is when we are the vendor
1371 */
1372 // if (info->vendor_is_local)
1373 // strcpy(ddf->controller.vendor_data, homehost);
1374 }
1375 if (strcmp(update, "name") == 0) {
1376 /* name is stored in virtual_entry->name */
1377 // memset(ve->name, ' ', 16);
1378 // strncpy(ve->name, info->name, 16);
1379 }
1380 if (strcmp(update, "_reshape_progress") == 0) {
1381 /* We don't support reshape yet */
1382 }
1383
1384 // update_all_csum(ddf);
1385
1386 return rv;
1387 }
1388
1389 static void make_header_guid(char *guid)
1390 {
1391 __u32 stamp;
1392 int rfd;
1393 /* Create a DDF Header of Virtual Disk GUID */
1394
1395 /* 24 bytes of fiction required.
1396 * first 8 are a 'vendor-id' - "Linux-MD"
1397 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1398 * Remaining 8 random number plus timestamp
1399 */
1400 memcpy(guid, T10, sizeof(T10));
1401 stamp = __cpu_to_be32(0xdeadbeef);
1402 memcpy(guid+8, &stamp, 4);
1403 stamp = __cpu_to_be32(0);
1404 memcpy(guid+12, &stamp, 4);
1405 stamp = __cpu_to_be32(time(0) - DECADE);
1406 memcpy(guid+16, &stamp, 4);
1407 rfd = open("/dev/urandom", O_RDONLY);
1408 if (rfd < 0 || read(rfd, &stamp, 4) != 4)
1409 stamp = random();
1410 memcpy(guid+20, &stamp, 4);
1411 if (rfd >= 0) close(rfd);
1412 }
1413 static int init_super_ddf(struct supertype *st,
1414 mdu_array_info_t *info,
1415 unsigned long long size, char *name, char *homehost,
1416 int *uuid)
1417 {
1418 /* This is primarily called by Create when creating a new array.
1419 * We will then get add_to_super called for each component, and then
1420 * write_init_super called to write it out to each device.
1421 * For DDF, Create can create on fresh devices or on a pre-existing
1422 * array.
1423 * To create on a pre-existing array a different method will be called.
1424 * This one is just for fresh drives.
1425 *
1426 * We need to create the entire 'ddf' structure which includes:
1427 * DDF headers - these are easy.
1428 * Controller data - a Sector describing this controller .. not that
1429 * this is a controller exactly.
1430 * Physical Disk Record - one entry per device, so
1431 * leave plenty of space.
1432 * Virtual Disk Records - again, just leave plenty of space.
1433 * This just lists VDs, doesn't give details
1434 * Config records - describes the VDs that use this disk
1435 * DiskData - describes 'this' device.
1436 * BadBlockManagement - empty
1437 * Diag Space - empty
1438 * Vendor Logs - Could we put bitmaps here?
1439 *
1440 */
1441 struct ddf_super *ddf;
1442 char hostname[17];
1443 int hostlen;
1444 int max_phys_disks, max_virt_disks;
1445 unsigned long long sector;
1446 int clen;
1447 int i;
1448 int pdsize, vdsize;
1449 struct phys_disk *pd;
1450 struct virtual_disk *vd;
1451
1452 ddf = malloc(sizeof(*ddf));
1453 ddf->dlist = NULL; /* no physical disks yet */
1454 ddf->conflist = NULL; /* No virtual disks yet */
1455
1456 /* At least 32MB *must* be reserved for the ddf. So let's just
1457 * start 32MB from the end, and put the primary header there.
1458 * Don't do secondary for now.
1459 * We don't know exactly where that will be yet as it could be
1460 * different on each device. To just set up the lengths.
1461 *
1462 */
1463
1464 ddf->anchor.magic = DDF_HEADER_MAGIC;
1465 make_header_guid(ddf->anchor.guid);
1466
1467 memcpy(ddf->anchor.revision, DDF_REVISION, 8);
1468 ddf->anchor.seq = __cpu_to_be32(1);
1469 ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1470 ddf->anchor.openflag = 0xFF;
1471 ddf->anchor.foreignflag = 0;
1472 ddf->anchor.enforcegroups = 0; /* Is this best?? */
1473 ddf->anchor.pad0 = 0xff;
1474 memset(ddf->anchor.pad1, 0xff, 12);
1475 memset(ddf->anchor.header_ext, 0xff, 32);
1476 ddf->anchor.primary_lba = ~(__u64)0;
1477 ddf->anchor.secondary_lba = ~(__u64)0;
1478 ddf->anchor.type = DDF_HEADER_ANCHOR;
1479 memset(ddf->anchor.pad2, 0xff, 3);
1480 ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1481 ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1482 of 32M reserved.. */
1483 max_phys_disks = 1023; /* Should be enough */
1484 ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1485 max_virt_disks = 255;
1486 ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1487 ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1488 ddf->max_part = 64;
1489 ddf->conf_rec_len = 1 + 256 * 12 / 512;
1490 ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1491 ddf->anchor.max_primary_element_entries = __cpu_to_be16(256);
1492 ddf->mppe = 256;
1493 memset(ddf->anchor.pad3, 0xff, 54);
1494
1495 /* controller sections is one sector long immediately
1496 * after the ddf header */
1497 sector = 1;
1498 ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1499 ddf->anchor.controller_section_length = __cpu_to_be32(1);
1500 sector += 1;
1501
1502 /* phys is 8 sectors after that */
1503 pdsize = ROUND_UP(sizeof(struct phys_disk) +
1504 sizeof(struct phys_disk_entry)*max_phys_disks,
1505 512);
1506 switch(pdsize/512) {
1507 case 2: case 8: case 32: case 128: case 512: break;
1508 default: abort();
1509 }
1510 ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1511 ddf->anchor.phys_section_length =
1512 __cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1513 sector += pdsize/512;
1514
1515 /* virt is another 32 sectors */
1516 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1517 sizeof(struct virtual_entry) * max_virt_disks,
1518 512);
1519 switch(vdsize/512) {
1520 case 2: case 8: case 32: case 128: case 512: break;
1521 default: abort();
1522 }
1523 ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1524 ddf->anchor.virt_section_length =
1525 __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1526 sector += vdsize/512;
1527
1528 clen = (1 + 256*12/512) * (64+1);
1529 ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1530 ddf->anchor.config_section_length = __cpu_to_be32(clen);
1531 sector += clen;
1532
1533 ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1534 ddf->anchor.data_section_length = __cpu_to_be32(1);
1535 sector += 1;
1536
1537 ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1538 ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1539 ddf->anchor.diag_space_length = __cpu_to_be32(0);
1540 ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1541 ddf->anchor.vendor_length = __cpu_to_be32(0);
1542 ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1543
1544 memset(ddf->anchor.pad4, 0xff, 256);
1545
1546 memcpy(&ddf->primary, &ddf->anchor, 512);
1547 memcpy(&ddf->secondary, &ddf->anchor, 512);
1548
1549 ddf->primary.openflag = 1; /* I guess.. */
1550 ddf->primary.type = DDF_HEADER_PRIMARY;
1551
1552 ddf->secondary.openflag = 1; /* I guess.. */
1553 ddf->secondary.type = DDF_HEADER_SECONDARY;
1554
1555 ddf->active = &ddf->primary;
1556
1557 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1558
1559 /* 24 more bytes of fiction required.
1560 * first 8 are a 'vendor-id' - "Linux-MD"
1561 * Remaining 16 are serial number.... maybe a hostname would do?
1562 */
1563 memcpy(ddf->controller.guid, T10, sizeof(T10));
1564 gethostname(hostname, 17);
1565 hostname[17] = 0;
1566 hostlen = strlen(hostname);
1567 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1568 for (i = strlen(T10) ; i+hostlen < 24; i++)
1569 ddf->controller.guid[i] = ' ';
1570
1571 ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1572 ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1573 ddf->controller.type.sub_vendor_id = 0;
1574 ddf->controller.type.sub_device_id = 0;
1575 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1576 memset(ddf->controller.pad, 0xff, 8);
1577 memset(ddf->controller.vendor_data, 0xff, 448);
1578
1579 pd = ddf->phys = malloc(pdsize);
1580 ddf->pdsize = pdsize;
1581
1582 memset(pd, 0xff, pdsize);
1583 memset(pd, 0, sizeof(*pd));
1584 pd->magic = DDF_PHYS_DATA_MAGIC;
1585 pd->used_pdes = __cpu_to_be16(0);
1586 pd->max_pdes = __cpu_to_be16(max_phys_disks);
1587 memset(pd->pad, 0xff, 52);
1588
1589 vd = ddf->virt = malloc(vdsize);
1590 ddf->vdsize = vdsize;
1591 memset(vd, 0, vdsize);
1592 vd->magic = DDF_VIRT_RECORDS_MAGIC;
1593 vd->populated_vdes = __cpu_to_be16(0);
1594 vd->max_vdes = __cpu_to_be16(max_virt_disks);
1595 memset(vd->pad, 0xff, 52);
1596
1597 for (i=0; i<max_virt_disks; i++)
1598 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1599
1600 st->sb = ddf;
1601 return 1;
1602 }
1603
1604 static int all_ff(char *guid)
1605 {
1606 int i;
1607 for (i = 0; i < DDF_GUID_LEN; i++)
1608 if (guid[i] != (char)0xff)
1609 return 0;
1610 return 1;
1611 }
1612 static int chunk_to_shift(int chunksize)
1613 {
1614 return ffs(chunksize/512)-1;
1615 }
1616
1617 static int level_to_prl(int level)
1618 {
1619 switch (level) {
1620 case LEVEL_LINEAR: return DDF_CONCAT;
1621 case 0: return DDF_RAID0;
1622 case 1: return DDF_RAID1;
1623 case 4: return DDF_RAID4;
1624 case 5: return DDF_RAID5;
1625 case 6: return DDF_RAID6;
1626 default: return -1;
1627 }
1628 }
1629 static int layout_to_rlq(int level, int layout, int raiddisks)
1630 {
1631 switch(level) {
1632 case 0:
1633 return DDF_RAID0_SIMPLE;
1634 case 1:
1635 switch(raiddisks) {
1636 case 2: return DDF_RAID1_SIMPLE;
1637 case 3: return DDF_RAID1_MULTI;
1638 default: return -1;
1639 }
1640 case 4:
1641 switch(layout) {
1642 case 0: return DDF_RAID4_N;
1643 }
1644 break;
1645 case 5:
1646 case 6:
1647 switch(layout) {
1648 case ALGORITHM_LEFT_ASYMMETRIC:
1649 return DDF_RAID5_N_RESTART;
1650 case ALGORITHM_RIGHT_ASYMMETRIC:
1651 return DDF_RAID5_0_RESTART;
1652 case ALGORITHM_LEFT_SYMMETRIC:
1653 return DDF_RAID5_N_CONTINUE;
1654 case ALGORITHM_RIGHT_SYMMETRIC:
1655 return -1; /* not mentioned in standard */
1656 }
1657 }
1658 return -1;
1659 }
1660
1661 static int rlq_to_layout(int rlq, int prl, int raiddisks)
1662 {
1663 switch(prl) {
1664 case DDF_RAID0:
1665 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1666 case DDF_RAID1:
1667 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1668 on raiddisks*/
1669 case DDF_RAID4:
1670 switch(rlq) {
1671 case DDF_RAID4_N:
1672 return 0;
1673 default:
1674 /* not supported */
1675 return -1; /* FIXME this isn't checked */
1676 }
1677 case DDF_RAID5:
1678 case DDF_RAID6:
1679 switch(rlq) {
1680 case DDF_RAID5_N_RESTART:
1681 return ALGORITHM_LEFT_ASYMMETRIC;
1682 case DDF_RAID5_0_RESTART:
1683 return ALGORITHM_RIGHT_ASYMMETRIC;
1684 case DDF_RAID5_N_CONTINUE:
1685 return ALGORITHM_LEFT_SYMMETRIC;
1686 default:
1687 return -1;
1688 }
1689 }
1690 return -1;
1691 }
1692
1693 static int init_super_ddf_bvd(struct supertype *st,
1694 mdu_array_info_t *info,
1695 unsigned long long size,
1696 char *name, char *homehost,
1697 int *uuid)
1698 {
1699 /* We are creating a BVD inside a pre-existing container.
1700 * so st->sb is already set.
1701 * We need to create a new vd_config and a new virtual_entry
1702 */
1703 struct ddf_super *ddf = st->sb;
1704 int venum;
1705 struct virtual_entry *ve;
1706 struct vcl *vcl;
1707 struct vd_config *vc;
1708
1709 if (__be16_to_cpu(ddf->virt->populated_vdes)
1710 >= __be16_to_cpu(ddf->virt->max_vdes)) {
1711 fprintf(stderr, Name": This ddf already has the "
1712 "maximum of %d virtual devices\n",
1713 __be16_to_cpu(ddf->virt->max_vdes));
1714 return 0;
1715 }
1716
1717 for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1718 if (all_ff(ddf->virt->entries[venum].guid))
1719 break;
1720 if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1721 fprintf(stderr, Name ": Cannot find spare slot for "
1722 "virtual disk - DDF is corrupt\n");
1723 return 0;
1724 }
1725 ve = &ddf->virt->entries[venum];
1726 ddf->conf_num = venum;
1727
1728 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1729 * timestamp, random number
1730 */
1731 make_header_guid(ve->guid);
1732 ve->unit = __cpu_to_be16(info->md_minor);
1733 ve->pad0 = 0xFFFF;
1734 ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1735 ve->type = 0;
1736 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
1737 if (info->state & 1) /* clean */
1738 ve->init_state = DDF_init_full;
1739 else
1740 ve->init_state = DDF_init_not;
1741
1742 memset(ve->pad1, 0xff, 14);
1743 memset(ve->name, ' ', 16);
1744 if (name)
1745 strncpy(ve->name, name, 16);
1746 ddf->virt->populated_vdes =
1747 __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
1748
1749 /* Now create a new vd_config */
1750 vcl = malloc(offsetof(struct vcl, conf) + ddf->conf_rec_len * 512);
1751 vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
1752
1753 vc = &vcl->conf;
1754
1755 vc->magic = DDF_VD_CONF_MAGIC;
1756 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
1757 vc->timestamp = __cpu_to_be32(time(0)-DECADE);
1758 vc->seqnum = __cpu_to_be32(1);
1759 memset(vc->pad0, 0xff, 24);
1760 vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
1761 vc->chunk_shift = chunk_to_shift(info->chunk_size);
1762 vc->prl = level_to_prl(info->level);
1763 vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
1764 vc->sec_elmnt_count = 1;
1765 vc->sec_elmnt_seq = 0;
1766 vc->srl = 0;
1767 vc->blocks = __cpu_to_be64(info->size * 2);
1768 vc->array_blocks = __cpu_to_be64(
1769 calc_array_size(info->level, info->raid_disks, info->layout,
1770 info->chunk_size, info->size*2));
1771 memset(vc->pad1, 0xff, 8);
1772 vc->spare_refs[0] = 0xffffffff;
1773 vc->spare_refs[1] = 0xffffffff;
1774 vc->spare_refs[2] = 0xffffffff;
1775 vc->spare_refs[3] = 0xffffffff;
1776 vc->spare_refs[4] = 0xffffffff;
1777 vc->spare_refs[5] = 0xffffffff;
1778 vc->spare_refs[6] = 0xffffffff;
1779 vc->spare_refs[7] = 0xffffffff;
1780 memset(vc->cache_pol, 0, 8);
1781 vc->bg_rate = 0x80;
1782 memset(vc->pad2, 0xff, 3);
1783 memset(vc->pad3, 0xff, 52);
1784 memset(vc->pad4, 0xff, 192);
1785 memset(vc->v0, 0xff, 32);
1786 memset(vc->v1, 0xff, 32);
1787 memset(vc->v2, 0xff, 16);
1788 memset(vc->v3, 0xff, 16);
1789 memset(vc->vendor, 0xff, 32);
1790
1791 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
1792 memset(vc->phys_refnum+(ddf->mppe * 4), 0x00, 8*ddf->mppe);
1793
1794 vcl->next = ddf->conflist;
1795 ddf->conflist = vcl;
1796 ddf->newconf = vcl;
1797 return 1;
1798 }
1799
1800 static void add_to_super_ddf_bvd(struct supertype *st,
1801 mdu_disk_info_t *dk, int fd, char *devname)
1802 {
1803 /* fd and devname identify a device with-in the ddf container (st).
1804 * dk identifies a location in the new BVD.
1805 * We need to find suitable free space in that device and update
1806 * the phys_refnum and lba_offset for the newly created vd_config.
1807 * We might also want to update the type in the phys_disk
1808 * section.
1809 */
1810 struct dl *dl;
1811 struct ddf_super *ddf = st->sb;
1812 struct vd_config *vc;
1813 __u64 *lba_offset;
1814 int working;
1815 int i;
1816 int max_virt_disks;
1817
1818 for (dl = ddf->dlist; dl ; dl = dl->next)
1819 if (dl->major == dk->major &&
1820 dl->minor == dk->minor)
1821 break;
1822 if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
1823 return;
1824
1825 vc = &ddf->newconf->conf;
1826 lba_offset = ddf->newconf->lba_offset;
1827 vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
1828 lba_offset[dk->raid_disk] = 0; /* FIXME */
1829
1830 for (i=0; i < ddf->max_part ; i++)
1831 if (dl->vlist[i] == NULL)
1832 break;
1833 if (i == ddf->max_part)
1834 return;
1835 dl->vlist[i] = ddf->newconf;
1836
1837 dl->fd = fd;
1838 dl->devname = devname;
1839
1840 /* Check how many working raid_disks, and if we can mark
1841 * array as optimal yet
1842 */
1843 working = 0;
1844
1845 for (i=0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
1846 if (vc->phys_refnum[i] != 0xffffffff)
1847 working++;
1848 /* Find which virtual_entry */
1849 max_virt_disks = __be16_to_cpu(ddf->active->max_vd_entries);
1850 for (i=0; i < max_virt_disks ; i++)
1851 if (memcmp(ddf->virt->entries[i].guid,
1852 vc->guid, DDF_GUID_LEN)==0)
1853 break;
1854 if (i == max_virt_disks)
1855 return;
1856 if (working == __be16_to_cpu(vc->prim_elmnt_count))
1857 ddf->virt->entries[i].state =
1858 (ddf->virt->entries[i].state & ~DDF_state_mask)
1859 | DDF_state_optimal;
1860
1861 if (vc->prl == DDF_RAID6 &&
1862 working+1 == __be16_to_cpu(vc->prim_elmnt_count))
1863 ddf->virt->entries[i].state =
1864 (ddf->virt->entries[i].state & ~DDF_state_mask)
1865 | DDF_state_part_optimal;
1866
1867 ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
1868 ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
1869 }
1870
1871 /* add a device to a container, either while creating it or while
1872 * expanding a pre-existing container
1873 */
1874 static void add_to_super_ddf(struct supertype *st,
1875 mdu_disk_info_t *dk, int fd, char *devname)
1876 {
1877 struct ddf_super *ddf = st->sb;
1878 struct dl *dd;
1879 time_t now;
1880 struct tm *tm;
1881 unsigned long long size;
1882 struct phys_disk_entry *pde;
1883 int n, i;
1884 struct stat stb;
1885
1886 /* This is device numbered dk->number. We need to create
1887 * a phys_disk entry and a more detailed disk_data entry.
1888 */
1889 fstat(fd, &stb);
1890 dd = malloc(sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part);
1891 dd->major = major(stb.st_rdev);
1892 dd->minor = minor(stb.st_rdev);
1893 dd->devname = devname;
1894 dd->next = ddf->dlist;
1895 dd->fd = fd;
1896 dd->spare = NULL;
1897
1898 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
1899 now = time(0);
1900 tm = localtime(&now);
1901 sprintf(dd->disk.guid, "%8s%04d%02d%02d",
1902 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
1903 *(__u32*)(dd->disk.guid + 16) = random();
1904 *(__u32*)(dd->disk.guid + 20) = random();
1905
1906 dd->disk.refnum = random(); /* and hope for the best FIXME check this is unique!!*/
1907 dd->disk.forced_ref = 1;
1908 dd->disk.forced_guid = 1;
1909 memset(dd->disk.vendor, ' ', 32);
1910 memcpy(dd->disk.vendor, "Linux", 5);
1911 memset(dd->disk.pad, 0xff, 442);
1912 for (i = 0; i < ddf->max_part ; i++)
1913 dd->vlist[i] = NULL;
1914
1915 n = __be16_to_cpu(ddf->phys->used_pdes);
1916 pde = &ddf->phys->entries[n];
1917 dd->pdnum = n;
1918
1919 n++;
1920 ddf->phys->used_pdes = __cpu_to_be16(n);
1921
1922 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
1923 pde->refnum = dd->disk.refnum;
1924 pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
1925 pde->state = __cpu_to_be16(DDF_Online);
1926 get_dev_size(fd, NULL, &size);
1927 /* We are required to reserve 32Meg, and record the size in sectors */
1928 pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
1929 sprintf(pde->path, "%17.17s","Information: nil") ;
1930 memset(pde->pad, 0xff, 6);
1931
1932 ddf->dlist = dd;
1933 }
1934
1935 /*
1936 * This is the write_init_super method for a ddf container. It is
1937 * called when creating a container or adding another device to a
1938 * container.
1939 */
1940
1941 #ifndef MDASSEMBLE
1942 static int __write_init_super_ddf(struct supertype *st, int do_close)
1943 {
1944
1945 struct ddf_super *ddf = st->sb;
1946 int i;
1947 struct dl *d;
1948 int n_config;
1949 int conf_size;
1950
1951 unsigned long long size, sector;
1952
1953 for (d = ddf->dlist; d; d=d->next) {
1954 int fd = d->fd;
1955
1956 if (fd < 0)
1957 continue;
1958
1959 /* We need to fill in the primary, (secondary) and workspace
1960 * lba's in the headers, set their checksums,
1961 * Also checksum phys, virt....
1962 *
1963 * Then write everything out, finally the anchor is written.
1964 */
1965 get_dev_size(fd, NULL, &size);
1966 size /= 512;
1967 ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
1968 ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
1969 ddf->anchor.seq = __cpu_to_be32(1);
1970 memcpy(&ddf->primary, &ddf->anchor, 512);
1971 memcpy(&ddf->secondary, &ddf->anchor, 512);
1972
1973 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
1974 ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
1975 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
1976
1977 ddf->primary.openflag = 0;
1978 ddf->primary.type = DDF_HEADER_PRIMARY;
1979
1980 ddf->secondary.openflag = 0;
1981 ddf->secondary.type = DDF_HEADER_SECONDARY;
1982
1983 ddf->primary.crc = calc_crc(&ddf->primary, 512);
1984 ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
1985
1986 sector = size - 16*1024*2;
1987 lseek64(fd, sector<<9, 0);
1988 write(fd, &ddf->primary, 512);
1989
1990 ddf->controller.crc = calc_crc(&ddf->controller, 512);
1991 write(fd, &ddf->controller, 512);
1992
1993 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
1994
1995 write(fd, ddf->phys, ddf->pdsize);
1996
1997 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
1998 write(fd, ddf->virt, ddf->vdsize);
1999
2000 /* Now write lots of config records. */
2001 n_config = ddf->max_part;
2002 conf_size = ddf->conf_rec_len * 512;
2003 for (i = 0 ; i <= n_config ; i++) {
2004 struct vcl *c = d->vlist[i];
2005 if (i == n_config)
2006 c = (struct vcl*)d->spare;
2007
2008 if (c) {
2009 c->conf.crc = calc_crc(&c->conf, conf_size);
2010 write(fd, &c->conf, conf_size);
2011 } else {
2012 __u32 sig = 0xffffffff;
2013 write(fd, &sig, 4);
2014 lseek64(fd, conf_size-4, SEEK_CUR);
2015 }
2016 }
2017 d->disk.crc = calc_crc(&d->disk, 512);
2018 write(fd, &d->disk, 512);
2019
2020 /* Maybe do the same for secondary */
2021
2022 lseek64(fd, (size-1)*512, SEEK_SET);
2023 write(fd, &ddf->anchor, 512);
2024 if (do_close) {
2025 close(fd);
2026 d->fd = -1;
2027 }
2028 }
2029 return 1;
2030 }
2031
2032 static int write_init_super_ddf(struct supertype *st)
2033 {
2034 return __write_init_super_ddf(st, 1);
2035 }
2036
2037 #endif
2038
2039 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize)
2040 {
2041 /* We must reserve the last 32Meg */
2042 if (devsize <= 32*1024*2)
2043 return 0;
2044 return devsize - 32*1024*2;
2045 }
2046
2047 #ifndef MDASSEMBLE
2048 int validate_geometry_ddf(struct supertype *st,
2049 int level, int layout, int raiddisks,
2050 int chunk, unsigned long long size,
2051 char *dev, unsigned long long *freesize)
2052 {
2053 int fd;
2054 struct mdinfo *sra;
2055 int cfd;
2056
2057 /* ddf potentially supports lots of things, but it depends on
2058 * what devices are offered (and maybe kernel version?)
2059 * If given unused devices, we will make a container.
2060 * If given devices in a container, we will make a BVD.
2061 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2062 */
2063
2064 if (level == LEVEL_CONTAINER) {
2065 st->ss = &super_ddf_container;
2066 if (dev) {
2067 int rv =st->ss->validate_geometry(st, level, layout,
2068 raiddisks, chunk,
2069 size,
2070 NULL, freesize);
2071 if (rv)
2072 return rv;
2073 }
2074 return st->ss->validate_geometry(st, level, layout, raiddisks,
2075 chunk, size, dev, freesize);
2076 }
2077
2078 if (st->sb) {
2079 /* creating in a given container */
2080 st->ss = &super_ddf_bvd;
2081 if (dev) {
2082 int rv =st->ss->validate_geometry(st, level, layout,
2083 raiddisks, chunk,
2084 size,
2085 NULL, freesize);
2086 if (rv)
2087 return rv;
2088 }
2089 return st->ss->validate_geometry(st, level, layout, raiddisks,
2090 chunk, size, dev, freesize);
2091 }
2092 /* FIXME should exclude MULTIPATH, or more appropriately, allow
2093 * only known levels.
2094 */
2095 if (!dev)
2096 return 1;
2097
2098 /* This device needs to be either a device in a 'ddf' container,
2099 * or it needs to be a 'ddf-bvd' array.
2100 */
2101
2102 fd = open(dev, O_RDONLY|O_EXCL, 0);
2103 if (fd >= 0) {
2104 sra = sysfs_read(fd, 0, GET_VERSION);
2105 close(fd);
2106 if (sra && sra->array.major_version == -1 &&
2107 strcmp(sra->text_version, "ddf-bvd") == 0) {
2108 st->ss = &super_ddf_svd;
2109 return st->ss->validate_geometry(st, level, layout,
2110 raiddisks, chunk, size,
2111 dev, freesize);
2112 }
2113
2114 fprintf(stderr,
2115 Name ": Cannot create this array on device %s\n",
2116 dev);
2117 return 0;
2118 }
2119 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2120 fprintf(stderr, Name ": Cannot open %s: %s\n",
2121 dev, strerror(errno));
2122 return 0;
2123 }
2124 /* Well, it is in use by someone, maybe a 'ddf' container. */
2125 cfd = open_container(fd);
2126 if (cfd < 0) {
2127 close(fd);
2128 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
2129 dev);
2130 return 0;
2131 }
2132 sra = sysfs_read(cfd, 0, GET_VERSION);
2133 close(fd);
2134 if (sra && sra->array.major_version == -1 &&
2135 strcmp(sra->text_version, "ddf") == 0) {
2136 /* This is a member of a ddf container. Load the container
2137 * and try to create a bvd
2138 */
2139 struct ddf_super *ddf;
2140 st->ss = &super_ddf_bvd;
2141 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL, 1) == 0) {
2142 st->sb = ddf;
2143 st->container_dev = fd2devnum(cfd);
2144 close(cfd);
2145 return st->ss->validate_geometry(st, level, layout,
2146 raiddisks, chunk, size,
2147 dev, freesize);
2148 }
2149 close(cfd);
2150 }
2151 fprintf(stderr, Name ": Cannot use %s: Already in use\n",
2152 dev);
2153 return 1;
2154 }
2155
2156 int validate_geometry_ddf_container(struct supertype *st,
2157 int level, int layout, int raiddisks,
2158 int chunk, unsigned long long size,
2159 char *dev, unsigned long long *freesize)
2160 {
2161 int fd;
2162 unsigned long long ldsize;
2163
2164 if (level != LEVEL_CONTAINER)
2165 return 0;
2166 if (!dev)
2167 return 1;
2168
2169 fd = open(dev, O_RDONLY|O_EXCL, 0);
2170 if (fd < 0) {
2171 fprintf(stderr, Name ": Cannot open %s: %s\n",
2172 dev, strerror(errno));
2173 return 0;
2174 }
2175 if (!get_dev_size(fd, dev, &ldsize)) {
2176 close(fd);
2177 return 0;
2178 }
2179 close(fd);
2180
2181 *freesize = avail_size_ddf(st, ldsize >> 9);
2182
2183 return 1;
2184 }
2185
2186 struct extent {
2187 unsigned long long start, size;
2188 };
2189 int cmp_extent(const void *av, const void *bv)
2190 {
2191 const struct extent *a = av;
2192 const struct extent *b = bv;
2193 if (a->start < b->start)
2194 return -1;
2195 if (a->start > b->start)
2196 return 1;
2197 return 0;
2198 }
2199
2200 struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
2201 {
2202 /* find a list of used extents on the give physical device
2203 * (dnum) of the given ddf.
2204 * Return a malloced array of 'struct extent'
2205
2206 FIXME ignore DDF_Legacy devices?
2207
2208 */
2209 struct extent *rv;
2210 int n = 0;
2211 int dnum;
2212 int i, j;
2213
2214 /* FIXME this is dl->pdnum */
2215 for (dnum = 0; dnum < ddf->phys->used_pdes; dnum++)
2216 if (memcmp(dl->disk.guid,
2217 ddf->phys->entries[dnum].guid,
2218 DDF_GUID_LEN) == 0)
2219 break;
2220
2221 if (dnum == ddf->phys->used_pdes)
2222 return NULL;
2223
2224 rv = malloc(sizeof(struct extent) * (ddf->max_part + 2));
2225 if (!rv)
2226 return NULL;
2227
2228 for (i = 0; i < ddf->max_part; i++) {
2229 struct vcl *v = dl->vlist[i];
2230 if (v == NULL)
2231 continue;
2232 for (j=0; j < v->conf.prim_elmnt_count; j++)
2233 if (v->conf.phys_refnum[j] == dl->disk.refnum) {
2234 /* This device plays role 'j' in 'v'. */
2235 rv[n].start = __be64_to_cpu(v->lba_offset[j]);
2236 rv[n].size = __be64_to_cpu(v->conf.blocks);
2237 n++;
2238 break;
2239 }
2240 }
2241 qsort(rv, n, sizeof(*rv), cmp_extent);
2242
2243 rv[n].start = __be64_to_cpu(ddf->phys->entries[dnum].config_size);
2244 rv[n].size = 0;
2245 return rv;
2246 }
2247
2248 int validate_geometry_ddf_bvd(struct supertype *st,
2249 int level, int layout, int raiddisks,
2250 int chunk, unsigned long long size,
2251 char *dev, unsigned long long *freesize)
2252 {
2253 struct stat stb;
2254 struct ddf_super *ddf = st->sb;
2255 struct dl *dl;
2256 unsigned long long pos = 0;
2257 unsigned long long maxsize;
2258 struct extent *e;
2259 int i;
2260 /* ddf/bvd supports lots of things, but not containers */
2261 if (level == LEVEL_CONTAINER)
2262 return 0;
2263 /* We must have the container info already read in. */
2264 if (!ddf)
2265 return 0;
2266
2267 if (!dev) {
2268 /* General test: make sure there is space for
2269 * 'raiddisks' device extents of size 'size'.
2270 */
2271 unsigned long long minsize = size;
2272 int dcnt = 0;
2273 if (minsize == 0)
2274 minsize = 8;
2275 for (dl = ddf->dlist; dl ; dl = dl->next)
2276 {
2277 int found = 0;
2278 pos = 0;
2279
2280 i = 0;
2281 e = get_extents(ddf, dl);
2282 if (!e) continue;
2283 do {
2284 unsigned long long esize;
2285 esize = e[i].start - pos;
2286 if (esize >= minsize)
2287 found = 1;
2288 pos = e[i].start + e[i].size;
2289 i++;
2290 } while (e[i-1].size);
2291 if (found)
2292 dcnt++;
2293 free(e);
2294 }
2295 if (dcnt < raiddisks) {
2296 fprintf(stderr, Name ": Not enough devices with space "
2297 "for this array (%d < %d)\n",
2298 dcnt, raiddisks);
2299 return 0;
2300 }
2301 return 1;
2302 }
2303 /* This device must be a member of the set */
2304 if (stat(dev, &stb) < 0)
2305 return 0;
2306 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2307 return 0;
2308 for (dl = ddf->dlist ; dl ; dl = dl->next) {
2309 if (dl->major == major(stb.st_rdev) &&
2310 dl->minor == minor(stb.st_rdev))
2311 break;
2312 }
2313 if (!dl) {
2314 fprintf(stderr, Name ": %s is not in the same DDF set\n",
2315 dev);
2316 return 0;
2317 }
2318 e = get_extents(ddf, dl);
2319 maxsize = 0;
2320 i = 0;
2321 if (e) do {
2322 unsigned long long esize;
2323 esize = e[i].start - pos;
2324 if (esize >= maxsize)
2325 maxsize = esize;
2326 pos = e[i].start + e[i].size;
2327 i++;
2328 } while (e[i-1].size);
2329 *freesize = maxsize;
2330 // FIXME here I am
2331
2332 return 1;
2333 }
2334 int validate_geometry_ddf_svd(struct supertype *st,
2335 int level, int layout, int raiddisks,
2336 int chunk, unsigned long long size,
2337 char *dev, unsigned long long *freesize)
2338 {
2339 /* dd/svd only supports striped, mirrored, concat, spanned... */
2340 if (level != LEVEL_LINEAR &&
2341 level != 0 &&
2342 level != 1)
2343 return 0;
2344 return 1;
2345 }
2346
2347
2348 static int load_super_ddf_all(struct supertype *st, int fd,
2349 void **sbp, char *devname, int keep_fd)
2350 {
2351 struct mdinfo *sra;
2352 struct ddf_super *super;
2353 struct mdinfo *sd, *best = NULL;
2354 int bestseq = 0;
2355 int seq;
2356 char nm[20];
2357 int dfd;
2358
2359 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
2360 if (!sra)
2361 return 1;
2362 if (sra->array.major_version != -1 ||
2363 sra->array.minor_version != -2 ||
2364 strcmp(sra->text_version, "ddf") != 0)
2365 return 1;
2366
2367 super = malloc(sizeof(*super));
2368 if (!super)
2369 return 1;
2370 memset(super, 0, sizeof(*super));
2371
2372 /* first, try each device, and choose the best ddf */
2373 for (sd = sra->devs ; sd ; sd = sd->next) {
2374 int rv;
2375 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2376 dfd = dev_open(nm, O_RDONLY);
2377 if (dfd < 0)
2378 return 2;
2379 rv = load_ddf_headers(dfd, super, NULL);
2380 close(dfd);
2381 if (rv == 0) {
2382 seq = __be32_to_cpu(super->active->seq);
2383 if (super->active->openflag)
2384 seq--;
2385 if (!best || seq > bestseq) {
2386 bestseq = seq;
2387 best = sd;
2388 }
2389 }
2390 }
2391 if (!best)
2392 return 1;
2393 /* OK, load this ddf */
2394 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2395 dfd = dev_open(nm, O_RDONLY);
2396 if (dfd < 0)
2397 return 1;
2398 load_ddf_headers(dfd, super, NULL);
2399 load_ddf_global(dfd, super, NULL);
2400 close(dfd);
2401 /* Now we need the device-local bits */
2402 for (sd = sra->devs ; sd ; sd = sd->next) {
2403 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2404 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2405 if (dfd < 0)
2406 return 2;
2407 seq = load_ddf_local(dfd, super, NULL, keep_fd);
2408 if (!keep_fd) close(dfd);
2409 }
2410 *sbp = super;
2411 if (st->ss == NULL) {
2412 st->ss = &super_ddf_container;
2413 st->minor_version = 0;
2414 st->max_devs = 512;
2415 st->container_dev = fd2devnum(fd);
2416 }
2417 return 0;
2418 }
2419 #endif
2420
2421
2422
2423 static struct mdinfo *container_content_ddf(struct supertype *st)
2424 {
2425 /* Given a container loaded by load_super_ddf_all,
2426 * extract information about all the arrays into
2427 * an mdinfo tree.
2428 *
2429 * For each vcl in conflist: create an mdinfo, fill it in,
2430 * then look for matching devices (phys_refnum) in dlist
2431 * and create appropriate device mdinfo.
2432 */
2433 struct ddf_super *ddf = st->sb;
2434 struct mdinfo *rest = NULL;
2435 struct vcl *vc;
2436
2437 for (vc = ddf->conflist ; vc ; vc=vc->next)
2438 {
2439 int i;
2440 struct mdinfo *this;
2441 this = malloc(sizeof(*this));
2442 memset(this, 0, sizeof(*this));
2443 this->next = rest;
2444 rest = this;
2445
2446 this->array.major_version = 1000;
2447 this->array.minor_version = 0;
2448 this->array.patch_version = 0;
2449 this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2450 this->array.raid_disks =
2451 __be16_to_cpu(vc->conf.prim_elmnt_count);
2452 this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2453 this->array.raid_disks);
2454 this->array.md_minor = -1;
2455 this->array.ctime = DECADE +
2456 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
2457 this->array.utime = DECADE +
2458 __be32_to_cpu(vc->conf.timestamp);
2459 this->array.chunk_size = 512 << vc->conf.chunk_shift;
2460
2461 for (i=0; i < __be16_to_cpu(ddf->virt->populated_vdes); i++)
2462 if (memcmp(ddf->virt->entries[i].guid,
2463 vc->conf.guid, DDF_GUID_LEN) == 0)
2464 break;
2465 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2466 (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
2467 DDF_init_full) {
2468 this->array.state = 0;
2469 this->resync_start = 0;
2470 } else {
2471 this->array.state = 1;
2472 this->resync_start = ~0ULL;
2473 }
2474 memcpy(this->name, ddf->virt->entries[i].name, 32);
2475 this->name[33]=0;
2476
2477 memset(this->uuid, 0, sizeof(this->uuid));
2478 this->component_size = __be64_to_cpu(vc->conf.blocks);
2479 this->array.size = this->component_size / 2;
2480 this->container_member = i;
2481
2482 sprintf(this->text_version, "/%s/%d",
2483 devnum2devname(st->container_dev),
2484 this->container_member);
2485
2486
2487 for (i=0 ; i < ddf->mppe ; i++) {
2488 struct mdinfo *dev;
2489 struct dl *d;
2490
2491 if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
2492 continue;
2493
2494 this->array.working_disks++;
2495
2496 for (d = ddf->dlist; d ; d=d->next)
2497 if (d->disk.refnum == vc->conf.phys_refnum[i])
2498 break;
2499 if (d == NULL)
2500 break;
2501
2502 dev = malloc(sizeof(*dev));
2503 memset(dev, 0, sizeof(*dev));
2504 dev->next = this->devs;
2505 this->devs = dev;
2506
2507 dev->disk.number = __be32_to_cpu(d->disk.refnum);
2508 dev->disk.major = d->major;
2509 dev->disk.minor = d->minor;
2510 dev->disk.raid_disk = i;
2511 dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2512
2513 dev->events = __le32_to_cpu(ddf->primary.seq);
2514 dev->data_offset = vc->lba_offset[i];
2515 dev->component_size = __be64_to_cpu(vc->conf.blocks);
2516 if (d->devname)
2517 strcpy(dev->name, d->devname);
2518 }
2519 }
2520 return rest;
2521 }
2522
2523 static int init_zero_ddf(struct supertype *st,
2524 mdu_array_info_t *info,
2525 unsigned long long size, char *name,
2526 char *homehost, int *uuid)
2527 {
2528 st->sb = NULL;
2529 return 0;
2530 }
2531
2532 static int store_zero_ddf(struct supertype *st, int fd)
2533 {
2534 unsigned long long dsize;
2535 char buf[512];
2536 memset(buf, 0, 512);
2537
2538
2539 if (!get_dev_size(fd, NULL, &dsize))
2540 return 1;
2541
2542 lseek64(fd, dsize-512, 0);
2543 write(fd, buf, 512);
2544 return 0;
2545 }
2546
2547 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
2548 {
2549 /*
2550 * return:
2551 * 0 same, or first was empty, and second was copied
2552 * 1 second had wrong number
2553 * 2 wrong uuid
2554 * 3 wrong other info
2555 */
2556 struct ddf_super *first = st->sb;
2557 struct ddf_super *second = tst->sb;
2558
2559 if (!first) {
2560 st->sb = tst->sb;
2561 tst->sb = NULL;
2562 return 0;
2563 }
2564
2565 if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
2566 return 2;
2567
2568 /* FIXME should I look at anything else? */
2569 return 0;
2570 }
2571
2572 /*
2573 * A new array 'a' has been started which claims to be instance 'inst'
2574 * within container 'c'.
2575 * We need to confirm that the array matches the metadata in 'c' so
2576 * that we don't corrupt any metadata.
2577 */
2578 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
2579 {
2580 fprintf(stderr, "ddf: open_new %s\n", inst);
2581 a->info.container_member = atoi(inst);
2582 return 0;
2583 }
2584
2585 /*
2586 * The array 'a' is to be marked clean in the metadata.
2587 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2588 * clean up to the point (in sectors). If that cannot be recorded in the
2589 * metadata, then leave it as dirty.
2590 *
2591 * For DDF, we need to clear the DDF_state_inconsistent bit in the
2592 * !global! virtual_disk.virtual_entry structure.
2593 */
2594 static void ddf_set_array_state(struct active_array *a, int consistent)
2595 {
2596 struct ddf_super *ddf = a->container->sb;
2597 int inst = a->info.container_member;
2598 if (consistent)
2599 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
2600 else
2601 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
2602 ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
2603 if (a->resync_start == ~0ULL)
2604 ddf->virt->entries[inst].init_state |= DDF_init_full;
2605 else if (a->resync_start == 0)
2606 ddf->virt->entries[inst].init_state |= DDF_init_not;
2607 else
2608 ddf->virt->entries[inst].init_state |= DDF_init_quick;
2609
2610 printf("ddf mark %s %llu\n", consistent?"clean":"dirty",
2611 a->resync_start);
2612 }
2613
2614 /*
2615 * The state of each disk is stored in the global phys_disk structure
2616 * in phys_disk.entries[n].state.
2617 * This makes various combinations awkward.
2618 * - When a device fails in any array, it must be failed in all arrays
2619 * that include a part of this device.
2620 * - When a component is rebuilding, we cannot include it officially in the
2621 * array unless this is the only array that uses the device.
2622 *
2623 * So: when transitioning:
2624 * Online -> failed, just set failed flag. monitor will propagate
2625 * spare -> online, the device might need to be added to the array.
2626 * spare -> failed, just set failed. Don't worry if in array or not.
2627 */
2628 static void ddf_set_disk(struct active_array *a, int n, int state)
2629 {
2630 struct ddf_super *ddf = a->container->sb;
2631 int inst = a->info.container_member;
2632 struct vd_config *vc = find_vdcr(ddf, inst);
2633 int pd = find_phys(ddf, vc->phys_refnum[n]);
2634 int i, st, working;
2635
2636 if (vc == NULL) {
2637 fprintf(stderr, "ddf: cannot find instance %d!!\n", inst);
2638 return;
2639 }
2640 if (pd < 0) {
2641 /* disk doesn't currently exist. If it is now in_sync,
2642 * insert it. */
2643 if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
2644 /* Find dev 'n' in a->info->devs, determine the
2645 * ddf refnum, and set vc->phys_refnum and update
2646 * phys->entries[]
2647 */
2648 /* FIXME */
2649 }
2650 } else {
2651 if (state & DS_FAULTY)
2652 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Failed);
2653 if (state & DS_INSYNC) {
2654 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Online);
2655 ddf->phys->entries[pd].state &= __cpu_to_be16(~DDF_Rebuilding);
2656 }
2657 }
2658
2659 fprintf(stderr, "ddf: set_disk %d to %x\n", n, state);
2660
2661 /* Now we need to check the state of the array and update
2662 * virtual_disk.entries[n].state.
2663 * It needs to be one of "optimal", "degraded", "failed".
2664 * I don't understand 'deleted' or 'missing'.
2665 */
2666 working = 0;
2667 for (i=0; i < a->info.array.raid_disks; i++) {
2668 pd = find_phys(ddf, vc->phys_refnum[i]);
2669 if (pd < 0)
2670 continue;
2671 st = __be16_to_cpu(ddf->phys->entries[pd].state);
2672 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
2673 == DDF_Online)
2674 working++;
2675 }
2676 state = DDF_state_degraded;
2677 if (working == a->info.array.raid_disks)
2678 state = DDF_state_optimal;
2679 else switch(vc->prl) {
2680 case DDF_RAID0:
2681 case DDF_CONCAT:
2682 case DDF_JBOD:
2683 state = DDF_state_failed;
2684 break;
2685 case DDF_RAID1:
2686 if (working == 0)
2687 state = DDF_state_failed;
2688 break;
2689 case DDF_RAID4:
2690 case DDF_RAID5:
2691 if (working < a->info.array.raid_disks-1)
2692 state = DDF_state_failed;
2693 break;
2694 case DDF_RAID6:
2695 if (working < a->info.array.raid_disks-2)
2696 state = DDF_state_failed;
2697 else if (working == a->info.array.raid_disks-1)
2698 state = DDF_state_part_optimal;
2699 break;
2700 }
2701
2702 ddf->virt->entries[inst].state =
2703 (ddf->virt->entries[inst].state & ~DDF_state_mask)
2704 | state;
2705
2706 }
2707
2708 static void ddf_sync_metadata(struct supertype *st)
2709 {
2710
2711 /*
2712 * Write all data to all devices.
2713 * Later, we might be able to track whether only local changes
2714 * have been made, or whether any global data has been changed,
2715 * but ddf is sufficiently weird that it probably always
2716 * changes global data ....
2717 */
2718 __write_init_super_ddf(st, 0);
2719 fprintf(stderr, "ddf: sync_metadata\n");
2720 }
2721
2722 static void ddf_process_update(struct supertype *st,
2723 struct metadata_update *update)
2724 {
2725 /* Apply this update to the metadata.
2726 * The first 4 bytes are a DDF_*_MAGIC which guides
2727 * our actions.
2728 * Possible update are:
2729 * DDF_PHYS_RECORDS_MAGIC
2730 * Add a new physical device. Changes to this record
2731 * only happen implicitly.
2732 * used_pdes is the device number.
2733 * DDF_VIRT_RECORDS_MAGIC
2734 * Add a new VD. Possibly also change the 'access' bits.
2735 * populated_vdes is the entry number.
2736 * DDF_VD_CONF_MAGIC
2737 * New or updated VD. the VIRT_RECORD must already
2738 * exist. For an update, phys_refnum and lba_offset
2739 * (at least) are updated, and the VD_CONF must
2740 * be written to precisely those devices listed with
2741 * a phys_refnum.
2742 * DDF_SPARE_ASSIGN_MAGIC
2743 * replacement Spare Assignment Record... but for which device?
2744 *
2745 * So, e.g.:
2746 * - to create a new array, we send a VIRT_RECORD and
2747 * a VD_CONF. Then assemble and start the array.
2748 * - to activate a spare we send a VD_CONF to add the phys_refnum
2749 * and offset. This will also mark the spare as active with
2750 * a spare-assignment record.
2751 */
2752 struct ddf_super *ddf = st->sb;
2753 __u32 *magic = (__u32*)update->buf;
2754 struct phys_disk *pd;
2755 struct virtual_disk *vd;
2756 struct vd_config *vc;
2757 struct vcl *vcl;
2758 struct dl *dl;
2759 int mppe;
2760 int ent;
2761
2762 printf("Process update %x\n", *magic);
2763
2764 switch (*magic) {
2765 case DDF_PHYS_RECORDS_MAGIC:
2766
2767 if (update->len != (sizeof(struct phys_disk) +
2768 sizeof(struct phys_disk_entry)))
2769 return;
2770 pd = (struct phys_disk*)update->buf;
2771
2772 ent = __be16_to_cpu(pd->used_pdes);
2773 if (ent >= __be16_to_cpu(ddf->phys->max_pdes))
2774 return;
2775 if (!all_ff(ddf->phys->entries[ent].guid))
2776 return;
2777 ddf->phys->entries[ent] = pd->entries[0];
2778 ddf->phys->used_pdes = __cpu_to_be16(1 +
2779 __be16_to_cpu(ddf->phys->used_pdes));
2780 break;
2781
2782 case DDF_VIRT_RECORDS_MAGIC:
2783
2784 if (update->len != (sizeof(struct virtual_disk) +
2785 sizeof(struct virtual_entry)))
2786 return;
2787 vd = (struct virtual_disk*)update->buf;
2788
2789 ent = __be16_to_cpu(vd->populated_vdes);
2790 if (ent >= __be16_to_cpu(ddf->virt->max_vdes))
2791 return;
2792 if (!all_ff(ddf->virt->entries[ent].guid))
2793 return;
2794 ddf->virt->entries[ent] = vd->entries[0];
2795 ddf->virt->populated_vdes = __cpu_to_be16(1 +
2796 __be16_to_cpu(ddf->virt->populated_vdes));
2797 break;
2798
2799 case DDF_VD_CONF_MAGIC:
2800 printf("len %d %d\n", update->len, ddf->conf_rec_len);
2801
2802 mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries);
2803 if (update->len != ddf->conf_rec_len)
2804 return;
2805 vc = (struct vd_config*)update->buf;
2806 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2807 if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
2808 break;
2809 printf("vcl = %p\n", vcl);
2810 if (vcl) {
2811 /* An update, just copy the phys_refnum and lba_offset
2812 * fields
2813 */
2814 memcpy(vcl->conf.phys_refnum, vc->phys_refnum,
2815 mppe * (sizeof(__u32) + sizeof(__u64)));
2816 } else {
2817 /* A new VD_CONF */
2818 vcl = update->space;
2819 update->space = NULL;
2820 vcl->next = ddf->conflist;
2821 vcl->conf = *vc;
2822 vcl->lba_offset = (__u64*)
2823 &vcl->conf.phys_refnum[mppe];
2824 ddf->conflist = vcl;
2825 }
2826 /* Now make sure vlist is correct for each dl. */
2827 for (dl = ddf->dlist; dl; dl = dl->next) {
2828 int dn;
2829 int vn = 0;
2830 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2831 for (dn=0; dn < ddf->mppe ; dn++)
2832 if (vcl->conf.phys_refnum[dn] ==
2833 dl->disk.refnum) {
2834 printf("dev %d has %p at %d\n",
2835 dl->pdnum, vcl, vn);
2836 dl->vlist[vn++] = vcl;
2837 break;
2838 }
2839 while (vn < ddf->max_part)
2840 dl->vlist[vn++] = NULL;
2841 if (dl->vlist[0]) {
2842 ddf->phys->entries[dl->pdnum].type &=
2843 ~__cpu_to_be16(DDF_Global_Spare);
2844 ddf->phys->entries[dl->pdnum].type |=
2845 __cpu_to_be16(DDF_Active_in_VD);
2846 }
2847 if (dl->spare) {
2848 ddf->phys->entries[dl->pdnum].type &=
2849 ~__cpu_to_be16(DDF_Global_Spare);
2850 ddf->phys->entries[dl->pdnum].type |=
2851 __cpu_to_be16(DDF_Spare);
2852 }
2853 if (!dl->vlist[0] && !dl->spare) {
2854 ddf->phys->entries[dl->pdnum].type |=
2855 __cpu_to_be16(DDF_Global_Spare);
2856 ddf->phys->entries[dl->pdnum].type &=
2857 ~__cpu_to_be16(DDF_Spare |
2858 DDF_Active_in_VD);
2859 }
2860 }
2861 break;
2862 case DDF_SPARE_ASSIGN_MAGIC:
2863 default: break;
2864 }
2865 }
2866
2867 /*
2868 * Check if the array 'a' is degraded but not failed.
2869 * If it is, find as many spares as are available and needed and
2870 * arrange for their inclusion.
2871 * We only choose devices which are not already in the array,
2872 * and prefer those with a spare-assignment to this array.
2873 * otherwise we choose global spares - assuming always that
2874 * there is enough room.
2875 * For each spare that we assign, we return an 'mdinfo' which
2876 * describes the position for the device in the array.
2877 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
2878 * the new phys_refnum and lba_offset values.
2879 *
2880 * Only worry about BVDs at the moment.
2881 */
2882 static struct mdinfo *ddf_activate_spare(struct active_array *a,
2883 struct metadata_update **updates)
2884 {
2885 int working = 0;
2886 struct mdinfo *d;
2887 struct ddf_super *ddf = a->container->sb;
2888 int global_ok = 0;
2889 struct mdinfo *rv = NULL;
2890 struct mdinfo *di;
2891 struct metadata_update *mu;
2892 struct dl *dl;
2893 int i;
2894 struct vd_config *vc;
2895 __u64 *lba;
2896
2897 /* FIXME, If there is a DS_FAULTY, we want to wait for it to be
2898 * removed. Then only look at DS_REMOVE devices.
2899 * What about !DS_INSYNC - how can that happen?
2900 */
2901 for (d = a->info.devs ; d ; d = d->next) {
2902 if ((d->curr_state & DS_FAULTY) &&
2903 d->state_fd >= 0)
2904 /* wait for Removal to happen */
2905 return NULL;
2906 if (d->state_fd >= 0)
2907 working ++;
2908 }
2909
2910 printf("ddf_activate: working=%d (%d) level=%d\n", working, a->info.array.raid_disks,
2911 a->info.array.level);
2912 if (working == a->info.array.raid_disks)
2913 return NULL; /* array not degraded */
2914 switch (a->info.array.level) {
2915 case 1:
2916 if (working == 0)
2917 return NULL; /* failed */
2918 break;
2919 case 4:
2920 case 5:
2921 if (working < a->info.array.raid_disks - 1)
2922 return NULL; /* failed */
2923 break;
2924 case 6:
2925 if (working < a->info.array.raid_disks - 2)
2926 return NULL; /* failed */
2927 break;
2928 default: /* concat or stripe */
2929 return NULL; /* failed */
2930 }
2931
2932 /* For each slot, if it is not working, find a spare */
2933 dl = ddf->dlist;
2934 for (i = 0; i < a->info.array.raid_disks; i++) {
2935 for (d = a->info.devs ; d ; d = d->next)
2936 if (d->disk.raid_disk == i)
2937 break;
2938 printf("found %d: %p %x\n", i, d, d?d->curr_state:0);
2939 if (d && (d->state_fd >= 0))
2940 continue;
2941
2942 /* OK, this device needs recovery. Find a spare */
2943 again:
2944 for ( ; dl ; dl = dl->next) {
2945 unsigned long long esize;
2946 unsigned long long pos;
2947 struct mdinfo *d2;
2948 int is_global = 0;
2949 int is_dedicated = 0;
2950 struct extent *ex;
2951 int j;
2952 /* If in this array, skip */
2953 for (d2 = a->info.devs ; d2 ; d2 = d2->next)
2954 if (d2->disk.major == dl->major &&
2955 d2->disk.minor == dl->minor) {
2956 printf("%x:%x already in array\n", dl->major, dl->minor);
2957 break;
2958 }
2959 if (d2)
2960 continue;
2961 if (ddf->phys->entries[dl->pdnum].type &
2962 __cpu_to_be16(DDF_Spare)) {
2963 /* Check spare assign record */
2964 if (dl->spare) {
2965 if (dl->spare->type & DDF_spare_dedicated) {
2966 /* check spare_ents for guid */
2967 for (j = 0 ;
2968 j < __be16_to_cpu(dl->spare->populated);
2969 j++) {
2970 if (memcmp(dl->spare->spare_ents[j].guid,
2971 ddf->virt->entries[a->info.container_member].guid,
2972 DDF_GUID_LEN) == 0)
2973 is_dedicated = 1;
2974 }
2975 } else
2976 is_global = 1;
2977 }
2978 } else if (ddf->phys->entries[dl->pdnum].type &
2979 __cpu_to_be16(DDF_Global_Spare)) {
2980 is_global = 1;
2981 }
2982 if ( ! (is_dedicated ||
2983 (is_global && global_ok))) {
2984 printf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
2985 is_dedicated, is_global);
2986 continue;
2987 }
2988
2989 /* We are allowed to use this device - is there space?
2990 * We need a->info.component_size sectors */
2991 ex = get_extents(ddf, dl);
2992 if (!ex) {
2993 printf("cannot get extents\n");
2994 continue;
2995 }
2996 j = 0; pos = 0;
2997 esize = 0;
2998
2999 do {
3000 esize = ex[j].start - pos;
3001 if (esize >= a->info.component_size)
3002 break;
3003 pos = ex[i].start + ex[i].size;
3004 i++;
3005 } while (ex[i-1].size);
3006
3007 free(ex);
3008 if (esize < a->info.component_size) {
3009 printf("%x:%x has no room: %llu %llu\n", dl->major, dl->minor,
3010 esize, a->info.component_size);
3011 /* No room */
3012 continue;
3013 }
3014
3015 /* Cool, we have a device with some space at pos */
3016 di = malloc(sizeof(*di));
3017 memset(di, 0, sizeof(*di));
3018 di->disk.number = i;
3019 di->disk.raid_disk = i;
3020 di->disk.major = dl->major;
3021 di->disk.minor = dl->minor;
3022 di->disk.state = 0;
3023 di->data_offset = pos;
3024 di->component_size = a->info.component_size;
3025 di->container_member = dl->pdnum;
3026 di->next = rv;
3027 rv = di;
3028 printf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3029 i, pos);
3030
3031 break;
3032 }
3033 if (!dl && ! global_ok) {
3034 /* not enough dedicated spares, try global */
3035 global_ok = 1;
3036 dl = ddf->dlist;
3037 goto again;
3038 }
3039 }
3040
3041 if (!rv)
3042 /* No spares found */
3043 return rv;
3044 /* Now 'rv' has a list of devices to return.
3045 * Create a metadata_update record to update the
3046 * phys_refnum and lba_offset values
3047 */
3048 mu = malloc(sizeof(*mu) + ddf->conf_rec_len * 512);
3049 mu->buf = (char*)(mu+1);
3050 mu->space = malloc(sizeof(struct vcl));
3051 mu->len = ddf->conf_rec_len;
3052 mu->next = *updates;
3053 vc = find_vdcr(ddf, a->info.container_member);
3054 memcpy(mu->buf, vc, ddf->conf_rec_len * 512);
3055
3056 vc = (struct vd_config*)mu->buf;
3057 lba = (__u64*)&vc->phys_refnum[ddf->mppe];
3058 for (di = rv ; di ; di = di->next) {
3059 vc->phys_refnum[di->disk.raid_disk] =
3060 ddf->phys->entries[dl->pdnum].refnum;
3061 lba[di->disk.raid_disk] = di->data_offset;
3062 }
3063 *updates = mu;
3064 return rv;
3065 }
3066
3067 struct superswitch super_ddf = {
3068 #ifndef MDASSEMBLE
3069 .examine_super = examine_super_ddf,
3070 .brief_examine_super = brief_examine_super_ddf,
3071 .detail_super = detail_super_ddf,
3072 .brief_detail_super = brief_detail_super_ddf,
3073 .validate_geometry = validate_geometry_ddf,
3074 #endif
3075 .match_home = match_home_ddf,
3076 .uuid_from_super= uuid_from_super_ddf,
3077 .getinfo_super = getinfo_super_ddf,
3078 .update_super = update_super_ddf,
3079
3080 .avail_size = avail_size_ddf,
3081
3082 .compare_super = compare_super_ddf,
3083
3084 .load_super = load_super_ddf,
3085 .init_super = init_zero_ddf,
3086 .store_super = store_zero_ddf,
3087 .free_super = free_super_ddf,
3088 .match_metadata_desc = match_metadata_desc_ddf,
3089 .getinfo_super_n = getinfo_super_n_container,
3090
3091
3092 .major = 1000,
3093 .swapuuid = 0,
3094 .external = 1,
3095
3096 /* for mdmon */
3097 .open_new = ddf_open_new,
3098 .set_array_state= ddf_set_array_state,
3099 .set_disk = ddf_set_disk,
3100 .sync_metadata = ddf_sync_metadata,
3101 .process_update = ddf_process_update,
3102 .activate_spare = ddf_activate_spare,
3103
3104 };
3105
3106 /* Super_ddf_container is set by validate_geometry_ddf when given a
3107 * device that is not part of any array
3108 */
3109 struct superswitch super_ddf_container = {
3110 #ifndef MDASSEMBLE
3111 .validate_geometry = validate_geometry_ddf_container,
3112 .write_init_super = write_init_super_ddf,
3113 #endif
3114
3115 .load_super = load_super_ddf,
3116 .init_super = init_super_ddf,
3117 .add_to_super = add_to_super_ddf,
3118 .getinfo_super = getinfo_super_ddf,
3119
3120 .free_super = free_super_ddf,
3121
3122 .container_content = container_content_ddf,
3123 .getinfo_super_n = getinfo_super_n_container,
3124
3125 .major = 1000,
3126 .swapuuid = 0,
3127 .external = 1,
3128 };
3129
3130 struct superswitch super_ddf_bvd = {
3131 #ifndef MDASSEMBLE
3132 // .detail_super = detail_super_ddf_bvd,
3133 // .brief_detail_super = brief_detail_super_ddf_bvd,
3134 .validate_geometry = validate_geometry_ddf_bvd,
3135 .write_init_super = write_init_super_ddf,
3136 #endif
3137 .update_super = update_super_ddf,
3138 .init_super = init_super_ddf_bvd,
3139 .add_to_super = add_to_super_ddf_bvd,
3140 .getinfo_super = getinfo_super_ddf_bvd,
3141 .getinfo_super_n = getinfo_super_n_bvd,
3142
3143 .load_super = load_super_ddf,
3144 .free_super = free_super_ddf,
3145 .match_metadata_desc = match_metadata_desc_ddf_bvd,
3146
3147
3148 .major = 1001,
3149 .swapuuid = 0,
3150 .external = 2,
3151 };
3152
3153 struct superswitch super_ddf_svd = {
3154 #ifndef MDASSEMBLE
3155 // .detail_super = detail_super_ddf_svd,
3156 // .brief_detail_super = brief_detail_super_ddf_svd,
3157 .validate_geometry = validate_geometry_ddf_svd,
3158 #endif
3159 .update_super = update_super_ddf,
3160 .init_super = init_super_ddf,
3161
3162 .load_super = load_super_ddf,
3163 .free_super = free_super_ddf,
3164 .match_metadata_desc = match_metadata_desc_ddf_svd,
3165
3166 .major = 1002,
3167 .swapuuid = 0,
3168 .external = 2,
3169 };
Unnamed repository; edit this file 'description' to name the repository.