]> 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
Remove the multiple super_switchs for 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 0x06
75
76 /* Raid Level Qualifier (RLQ) */
77 #define DDF_RAID0_SIMPLE 0x00
78 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
79 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
80 #define DDF_RAID3_0 0x00 /* parity in first extent */
81 #define DDF_RAID3_N 0x01 /* parity in last extent */
82 #define DDF_RAID4_0 0x00 /* parity in first extent */
83 #define DDF_RAID4_N 0x01 /* parity in last extent */
84 /* these apply to raid5e and raid5ee as well */
85 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
86 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
87 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
88 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
89
90 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
91 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
92
93 /* Secondary RAID Level (SRL) */
94 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
95 #define DDF_2MIRRORED 0x01
96 #define DDF_2CONCAT 0x02
97 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
98
99 /* Magic numbers */
100 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
101 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
102 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
103 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
104 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
105 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
106 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
107 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
108 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
109 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
110
111 #define DDF_GUID_LEN 24
112 #define DDF_REVISION_0 "01.00.00"
113 #define DDF_REVISION_2 "01.02.00"
114
115 struct ddf_header {
116 __u32 magic; /* DDF_HEADER_MAGIC */
117 __u32 crc;
118 char guid[DDF_GUID_LEN];
119 char revision[8]; /* 01.02.00 */
120 __u32 seq; /* starts at '1' */
121 __u32 timestamp;
122 __u8 openflag;
123 __u8 foreignflag;
124 __u8 enforcegroups;
125 __u8 pad0; /* 0xff */
126 __u8 pad1[12]; /* 12 * 0xff */
127 /* 64 bytes so far */
128 __u8 header_ext[32]; /* reserved: fill with 0xff */
129 __u64 primary_lba;
130 __u64 secondary_lba;
131 __u8 type;
132 __u8 pad2[3]; /* 0xff */
133 __u32 workspace_len; /* sectors for vendor space -
134 * at least 32768(sectors) */
135 __u64 workspace_lba;
136 __u16 max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */
137 __u16 max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */
138 __u16 max_partitions; /* i.e. max num of configuration
139 record entries per disk */
140 __u16 config_record_len; /* 1 +ROUNDUP(max_primary_element_entries
141 *12/512) */
142 __u16 max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */
143 __u8 pad3[54]; /* 0xff */
144 /* 192 bytes so far */
145 __u32 controller_section_offset;
146 __u32 controller_section_length;
147 __u32 phys_section_offset;
148 __u32 phys_section_length;
149 __u32 virt_section_offset;
150 __u32 virt_section_length;
151 __u32 config_section_offset;
152 __u32 config_section_length;
153 __u32 data_section_offset;
154 __u32 data_section_length;
155 __u32 bbm_section_offset;
156 __u32 bbm_section_length;
157 __u32 diag_space_offset;
158 __u32 diag_space_length;
159 __u32 vendor_offset;
160 __u32 vendor_length;
161 /* 256 bytes so far */
162 __u8 pad4[256]; /* 0xff */
163 };
164
165 /* type field */
166 #define DDF_HEADER_ANCHOR 0x00
167 #define DDF_HEADER_PRIMARY 0x01
168 #define DDF_HEADER_SECONDARY 0x02
169
170 /* The content of the 'controller section' - global scope */
171 struct ddf_controller_data {
172 __u32 magic; /* DDF_CONTROLLER_MAGIC */
173 __u32 crc;
174 char guid[DDF_GUID_LEN];
175 struct controller_type {
176 __u16 vendor_id;
177 __u16 device_id;
178 __u16 sub_vendor_id;
179 __u16 sub_device_id;
180 } type;
181 char product_id[16];
182 __u8 pad[8]; /* 0xff */
183 __u8 vendor_data[448];
184 };
185
186 /* The content of phys_section - global scope */
187 struct phys_disk {
188 __u32 magic; /* DDF_PHYS_RECORDS_MAGIC */
189 __u32 crc;
190 __u16 used_pdes;
191 __u16 max_pdes;
192 __u8 pad[52];
193 struct phys_disk_entry {
194 char guid[DDF_GUID_LEN];
195 __u32 refnum;
196 __u16 type;
197 __u16 state;
198 __u64 config_size; /* DDF structures must be after here */
199 char path[18]; /* another horrible structure really */
200 __u8 pad[6];
201 } entries[0];
202 };
203
204 /* phys_disk_entry.type is a bitmap - bigendian remember */
205 #define DDF_Forced_PD_GUID 1
206 #define DDF_Active_in_VD 2
207 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
208 #define DDF_Spare 8 /* overrides Global_spare */
209 #define DDF_Foreign 16
210 #define DDF_Legacy 32 /* no DDF on this device */
211
212 #define DDF_Interface_mask 0xf00
213 #define DDF_Interface_SCSI 0x100
214 #define DDF_Interface_SAS 0x200
215 #define DDF_Interface_SATA 0x300
216 #define DDF_Interface_FC 0x400
217
218 /* phys_disk_entry.state is a bigendian bitmap */
219 #define DDF_Online 1
220 #define DDF_Failed 2 /* overrides 1,4,8 */
221 #define DDF_Rebuilding 4
222 #define DDF_Transition 8
223 #define DDF_SMART 16
224 #define DDF_ReadErrors 32
225 #define DDF_Missing 64
226
227 /* The content of the virt_section global scope */
228 struct virtual_disk {
229 __u32 magic; /* DDF_VIRT_RECORDS_MAGIC */
230 __u32 crc;
231 __u16 populated_vdes;
232 __u16 max_vdes;
233 __u8 pad[52];
234 struct virtual_entry {
235 char guid[DDF_GUID_LEN];
236 __u16 unit;
237 __u16 pad0; /* 0xffff */
238 __u16 guid_crc;
239 __u16 type;
240 __u8 state;
241 __u8 init_state;
242 __u8 pad1[14];
243 char name[16];
244 } entries[0];
245 };
246
247 /* virtual_entry.type is a bitmap - bigendian */
248 #define DDF_Shared 1
249 #define DDF_Enforce_Groups 2
250 #define DDF_Unicode 4
251 #define DDF_Owner_Valid 8
252
253 /* virtual_entry.state is a bigendian bitmap */
254 #define DDF_state_mask 0x7
255 #define DDF_state_optimal 0x0
256 #define DDF_state_degraded 0x1
257 #define DDF_state_deleted 0x2
258 #define DDF_state_missing 0x3
259 #define DDF_state_failed 0x4
260 #define DDF_state_part_optimal 0x5
261
262 #define DDF_state_morphing 0x8
263 #define DDF_state_inconsistent 0x10
264
265 /* virtual_entry.init_state is a bigendian bitmap */
266 #define DDF_initstate_mask 0x03
267 #define DDF_init_not 0x00
268 #define DDF_init_quick 0x01 /* initialisation is progress.
269 * i.e. 'state_inconsistent' */
270 #define DDF_init_full 0x02
271
272 #define DDF_access_mask 0xc0
273 #define DDF_access_rw 0x00
274 #define DDF_access_ro 0x80
275 #define DDF_access_blocked 0xc0
276
277 /* The content of the config_section - local scope
278 * It has multiple records each config_record_len sectors
279 * They can be vd_config or spare_assign
280 */
281
282 struct vd_config {
283 __u32 magic; /* DDF_VD_CONF_MAGIC */
284 __u32 crc;
285 char guid[DDF_GUID_LEN];
286 __u32 timestamp;
287 __u32 seqnum;
288 __u8 pad0[24];
289 __u16 prim_elmnt_count;
290 __u8 chunk_shift; /* 0 == 512, 1==1024 etc */
291 __u8 prl;
292 __u8 rlq;
293 __u8 sec_elmnt_count;
294 __u8 sec_elmnt_seq;
295 __u8 srl;
296 __u64 blocks; /* blocks per component could be different
297 * on different component devices...(only
298 * for concat I hope) */
299 __u64 array_blocks; /* blocks in array */
300 __u8 pad1[8];
301 __u32 spare_refs[8];
302 __u8 cache_pol[8];
303 __u8 bg_rate;
304 __u8 pad2[3];
305 __u8 pad3[52];
306 __u8 pad4[192];
307 __u8 v0[32]; /* reserved- 0xff */
308 __u8 v1[32]; /* reserved- 0xff */
309 __u8 v2[16]; /* reserved- 0xff */
310 __u8 v3[16]; /* reserved- 0xff */
311 __u8 vendor[32];
312 __u32 phys_refnum[0]; /* refnum of each disk in sequence */
313 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
314 bvd are always the same size */
315 };
316
317 /* vd_config.cache_pol[7] is a bitmap */
318 #define DDF_cache_writeback 1 /* else writethrough */
319 #define DDF_cache_wadaptive 2 /* only applies if writeback */
320 #define DDF_cache_readahead 4
321 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
322 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
323 #define DDF_cache_wallowed 32 /* enable write caching */
324 #define DDF_cache_rallowed 64 /* enable read caching */
325
326 struct spare_assign {
327 __u32 magic; /* DDF_SPARE_ASSIGN_MAGIC */
328 __u32 crc;
329 __u32 timestamp;
330 __u8 reserved[7];
331 __u8 type;
332 __u16 populated; /* SAEs used */
333 __u16 max; /* max SAEs */
334 __u8 pad[8];
335 struct spare_assign_entry {
336 char guid[DDF_GUID_LEN];
337 __u16 secondary_element;
338 __u8 pad[6];
339 } spare_ents[0];
340 };
341 /* spare_assign.type is a bitmap */
342 #define DDF_spare_dedicated 0x1 /* else global */
343 #define DDF_spare_revertible 0x2 /* else committable */
344 #define DDF_spare_active 0x4 /* else not active */
345 #define DDF_spare_affinity 0x8 /* enclosure affinity */
346
347 /* The data_section contents - local scope */
348 struct disk_data {
349 __u32 magic; /* DDF_PHYS_DATA_MAGIC */
350 __u32 crc;
351 char guid[DDF_GUID_LEN];
352 __u32 refnum; /* crc of some magic drive data ... */
353 __u8 forced_ref; /* set when above was not result of magic */
354 __u8 forced_guid; /* set if guid was forced rather than magic */
355 __u8 vendor[32];
356 __u8 pad[442];
357 };
358
359 /* bbm_section content */
360 struct bad_block_log {
361 __u32 magic;
362 __u32 crc;
363 __u16 entry_count;
364 __u32 spare_count;
365 __u8 pad[10];
366 __u64 first_spare;
367 struct mapped_block {
368 __u64 defective_start;
369 __u32 replacement_start;
370 __u16 remap_count;
371 __u8 pad[2];
372 } entries[0];
373 };
374
375 /* Struct for internally holding ddf structures */
376 /* The DDF structure stored on each device is potentially
377 * quite different, as some data is global and some is local.
378 * The global data is:
379 * - ddf header
380 * - controller_data
381 * - Physical disk records
382 * - Virtual disk records
383 * The local data is:
384 * - Configuration records
385 * - Physical Disk data section
386 * ( and Bad block and vendor which I don't care about yet).
387 *
388 * The local data is parsed into separate lists as it is read
389 * and reconstructed for writing. This means that we only need
390 * to make config changes once and they are automatically
391 * propagated to all devices.
392 * Note that the ddf_super has space of the conf and disk data
393 * for this disk and also for a list of all such data.
394 * The list is only used for the superblock that is being
395 * built in Create or Assemble to describe the whole array.
396 */
397 struct ddf_super {
398 struct ddf_header anchor, primary, secondary, *active;
399 struct ddf_controller_data controller;
400 struct phys_disk *phys;
401 struct virtual_disk *virt;
402 int pdsize, vdsize;
403 int max_part, mppe, conf_rec_len;
404 int currentdev;
405 struct vcl {
406 struct vcl *next;
407 __u64 *lba_offset; /* location in 'conf' of
408 * the lba table */
409 int vcnum; /* index into ->virt */
410 __u64 *block_sizes; /* NULL if all the same */
411 struct vd_config conf;
412 } *conflist, *currentconf;
413 struct dl {
414 struct dl *next;
415 struct disk_data disk;
416 int major, minor;
417 char *devname;
418 int fd;
419 unsigned long long size; /* sectors */
420 int pdnum; /* index in ->phys */
421 struct spare_assign *spare;
422 struct vcl *vlist[0]; /* max_part in size */
423 } *dlist;
424 };
425
426 #ifndef offsetof
427 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
428 #endif
429
430
431 static int calc_crc(void *buf, int len)
432 {
433 /* crcs are always at the same place as in the ddf_header */
434 struct ddf_header *ddf = buf;
435 __u32 oldcrc = ddf->crc;
436 __u32 newcrc;
437 ddf->crc = 0xffffffff;
438
439 newcrc = crc32(0, buf, len);
440 ddf->crc = oldcrc;
441 return newcrc;
442 }
443
444 static int load_ddf_header(int fd, unsigned long long lba,
445 unsigned long long size,
446 int type,
447 struct ddf_header *hdr, struct ddf_header *anchor)
448 {
449 /* read a ddf header (primary or secondary) from fd/lba
450 * and check that it is consistent with anchor
451 * Need to check:
452 * magic, crc, guid, rev, and LBA's header_type, and
453 * everything after header_type must be the same
454 */
455 if (lba >= size-1)
456 return 0;
457
458 if (lseek64(fd, lba<<9, 0) < 0)
459 return 0;
460
461 if (read(fd, hdr, 512) != 512)
462 return 0;
463
464 if (hdr->magic != DDF_HEADER_MAGIC)
465 return 0;
466 if (calc_crc(hdr, 512) != hdr->crc)
467 return 0;
468 if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
469 memcmp(anchor->revision, hdr->revision, 8) != 0 ||
470 anchor->primary_lba != hdr->primary_lba ||
471 anchor->secondary_lba != hdr->secondary_lba ||
472 hdr->type != type ||
473 memcmp(anchor->pad2, hdr->pad2, 512 -
474 offsetof(struct ddf_header, pad2)) != 0)
475 return 0;
476
477 /* Looks good enough to me... */
478 return 1;
479 }
480
481 static void *load_section(int fd, struct ddf_super *super, void *buf,
482 __u32 offset_be, __u32 len_be, int check)
483 {
484 unsigned long long offset = __be32_to_cpu(offset_be);
485 unsigned long long len = __be32_to_cpu(len_be);
486 int dofree = (buf == NULL);
487
488 if (check)
489 if (len != 2 && len != 8 && len != 32
490 && len != 128 && len != 512)
491 return NULL;
492
493 if (len > 1024)
494 return NULL;
495 if (buf) {
496 /* All pre-allocated sections are a single block */
497 if (len != 1)
498 return NULL;
499 } else
500 buf = malloc(len<<9);
501 if (!buf)
502 return NULL;
503
504 if (super->active->type == 1)
505 offset += __be64_to_cpu(super->active->primary_lba);
506 else
507 offset += __be64_to_cpu(super->active->secondary_lba);
508
509 if (lseek64(fd, offset<<9, 0) != (offset<<9)) {
510 if (dofree)
511 free(buf);
512 return NULL;
513 }
514 if (read(fd, buf, len<<9) != (len<<9)) {
515 if (dofree)
516 free(buf);
517 return NULL;
518 }
519 return buf;
520 }
521
522 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
523 {
524 unsigned long long dsize;
525
526 get_dev_size(fd, NULL, &dsize);
527
528 if (lseek64(fd, dsize-512, 0) < 0) {
529 if (devname)
530 fprintf(stderr,
531 Name": Cannot seek to anchor block on %s: %s\n",
532 devname, strerror(errno));
533 return 1;
534 }
535 if (read(fd, &super->anchor, 512) != 512) {
536 if (devname)
537 fprintf(stderr,
538 Name ": Cannot read anchor block on %s: %s\n",
539 devname, strerror(errno));
540 return 1;
541 }
542 if (super->anchor.magic != DDF_HEADER_MAGIC) {
543 if (devname)
544 fprintf(stderr, Name ": no DDF anchor found on %s\n",
545 devname);
546 return 2;
547 }
548 if (calc_crc(&super->anchor, 512) != super->anchor.crc) {
549 if (devname)
550 fprintf(stderr, Name ": bad CRC on anchor on %s\n",
551 devname);
552 return 2;
553 }
554 if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
555 memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
556 if (devname)
557 fprintf(stderr, Name ": can only support super revision"
558 " %.8s and earlier, not %.8s on %s\n",
559 DDF_REVISION_2, super->anchor.revision,devname);
560 return 2;
561 }
562 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba),
563 dsize >> 9, 1,
564 &super->primary, &super->anchor) == 0) {
565 if (devname)
566 fprintf(stderr,
567 Name ": Failed to load primary DDF header "
568 "on %s\n", devname);
569 return 2;
570 }
571 super->active = &super->primary;
572 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.secondary_lba),
573 dsize >> 9, 2,
574 &super->secondary, &super->anchor)) {
575 if ((__be32_to_cpu(super->primary.seq)
576 < __be32_to_cpu(super->secondary.seq) &&
577 !super->secondary.openflag)
578 || (__be32_to_cpu(super->primary.seq)
579 == __be32_to_cpu(super->secondary.seq) &&
580 super->primary.openflag && !super->secondary.openflag)
581 )
582 super->active = &super->secondary;
583 }
584 return 0;
585 }
586
587 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
588 {
589 void *ok;
590 ok = load_section(fd, super, &super->controller,
591 super->active->controller_section_offset,
592 super->active->controller_section_length,
593 0);
594 super->phys = load_section(fd, super, NULL,
595 super->active->phys_section_offset,
596 super->active->phys_section_length,
597 1);
598 super->pdsize = __be32_to_cpu(super->active->phys_section_length) * 512;
599
600 super->virt = load_section(fd, super, NULL,
601 super->active->virt_section_offset,
602 super->active->virt_section_length,
603 1);
604 super->vdsize = __be32_to_cpu(super->active->virt_section_length) * 512;
605 if (!ok ||
606 !super->phys ||
607 !super->virt) {
608 free(super->phys);
609 free(super->virt);
610 super->phys = NULL;
611 super->virt = NULL;
612 return 2;
613 }
614 super->conflist = NULL;
615 super->dlist = NULL;
616
617 super->max_part = __be16_to_cpu(super->active->max_partitions);
618 super->mppe = __be16_to_cpu(super->active->max_primary_element_entries);
619 super->conf_rec_len = __be16_to_cpu(super->active->config_record_len);
620 return 0;
621 }
622
623 static int load_ddf_local(int fd, struct ddf_super *super,
624 char *devname, int keep)
625 {
626 struct dl *dl;
627 struct stat stb;
628 char *conf;
629 int i;
630 int vnum;
631 int max_virt_disks = __be16_to_cpu(super->active->max_vd_entries);
632 unsigned long long dsize;
633
634 /* First the local disk info */
635 dl = malloc(sizeof(*dl) +
636 (super->max_part) * sizeof(dl->vlist[0]));
637
638 load_section(fd, super, &dl->disk,
639 super->active->data_section_offset,
640 super->active->data_section_length,
641 0);
642 dl->devname = devname ? strdup(devname) : NULL;
643
644 fstat(fd, &stb);
645 dl->major = major(stb.st_rdev);
646 dl->minor = minor(stb.st_rdev);
647 dl->next = super->dlist;
648 dl->fd = keep ? fd : -1;
649
650 dl->size = 0;
651 if (get_dev_size(fd, devname, &dsize))
652 dl->size = dsize >> 9;
653 dl->spare = NULL;
654 for (i=0 ; i < super->max_part ; i++)
655 dl->vlist[i] = NULL;
656 super->dlist = dl;
657 dl->pdnum = -1;
658 for (i=0; i < __be16_to_cpu(super->active->max_pd_entries); i++)
659 if (memcmp(super->phys->entries[i].guid,
660 dl->disk.guid, DDF_GUID_LEN) == 0)
661 dl->pdnum = i;
662
663 /* Now the config list. */
664 /* 'conf' is an array of config entries, some of which are
665 * probably invalid. Those which are good need to be copied into
666 * the conflist
667 */
668
669 conf = load_section(fd, super, NULL,
670 super->active->config_section_offset,
671 super->active->config_section_length,
672 0);
673
674 vnum = 0;
675 for (i = 0;
676 i < __be32_to_cpu(super->active->config_section_length);
677 i += super->conf_rec_len) {
678 struct vd_config *vd =
679 (struct vd_config *)((char*)conf + i*512);
680 struct vcl *vcl;
681
682 if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) {
683 if (dl->spare)
684 continue;
685 dl->spare = malloc(super->conf_rec_len*512);
686 memcpy(dl->spare, vd, super->conf_rec_len*512);
687 continue;
688 }
689 if (vd->magic != DDF_VD_CONF_MAGIC)
690 continue;
691 for (vcl = super->conflist; vcl; vcl = vcl->next) {
692 if (memcmp(vcl->conf.guid,
693 vd->guid, DDF_GUID_LEN) == 0)
694 break;
695 }
696
697 if (vcl) {
698 dl->vlist[vnum++] = vcl;
699 if (__be32_to_cpu(vd->seqnum) <=
700 __be32_to_cpu(vcl->conf.seqnum))
701 continue;
702 } else {
703 vcl = malloc(super->conf_rec_len*512 +
704 offsetof(struct vcl, conf));
705 vcl->next = super->conflist;
706 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
707 super->conflist = vcl;
708 dl->vlist[vnum++] = vcl;
709 }
710 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
711 vcl->lba_offset = (__u64*)
712 &vcl->conf.phys_refnum[super->mppe];
713
714 for (i=0; i < max_virt_disks ; i++)
715 if (memcmp(super->virt->entries[i].guid,
716 vcl->conf.guid, DDF_GUID_LEN)==0)
717 break;
718 if (i < max_virt_disks)
719 vcl->vcnum = i;
720 }
721 free(conf);
722
723 return 0;
724 }
725
726 #ifndef MDASSEMBLE
727 static int load_super_ddf_all(struct supertype *st, int fd,
728 void **sbp, char *devname, int keep_fd);
729 #endif
730 static int load_super_ddf(struct supertype *st, int fd,
731 char *devname)
732 {
733 unsigned long long dsize;
734 struct ddf_super *super;
735 int rv;
736
737 #ifndef MDASSEMBLE
738 /* if 'fd' is a container, load metadata from all the devices */
739 if (load_super_ddf_all(st, fd, &st->sb, devname, 1) == 0)
740 return 0;
741 #endif
742 if (st->subarray[0])
743 return 1; /* FIXME Is this correct */
744
745 if (get_dev_size(fd, devname, &dsize) == 0)
746 return 1;
747
748 /* 32M is a lower bound */
749 if (dsize <= 32*1024*1024) {
750 if (devname) {
751 fprintf(stderr,
752 Name ": %s is too small for ddf: "
753 "size is %llu sectors.\n",
754 devname, dsize>>9);
755 return 1;
756 }
757 }
758 if (dsize & 511) {
759 if (devname) {
760 fprintf(stderr,
761 Name ": %s is an odd size for ddf: "
762 "size is %llu bytes.\n",
763 devname, dsize);
764 return 1;
765 }
766 }
767
768 super = malloc(sizeof(*super));
769 if (!super) {
770 fprintf(stderr, Name ": malloc of %zu failed.\n",
771 sizeof(*super));
772 return 1;
773 }
774 memset(super, 0, sizeof(*super));
775
776 rv = load_ddf_headers(fd, super, devname);
777 if (rv) {
778 free(super);
779 return rv;
780 }
781
782 /* Have valid headers and have chosen the best. Let's read in the rest*/
783
784 rv = load_ddf_global(fd, super, devname);
785
786 if (rv) {
787 if (devname)
788 fprintf(stderr,
789 Name ": Failed to load all information "
790 "sections on %s\n", devname);
791 free(super);
792 return rv;
793 }
794
795 load_ddf_local(fd, super, devname, 0);
796
797 /* Should possibly check the sections .... */
798
799 st->sb = super;
800 if (st->ss == NULL) {
801 st->ss = &super_ddf;
802 st->minor_version = 0;
803 st->max_devs = 512;
804 }
805 return 0;
806
807 }
808
809 static void free_super_ddf(struct supertype *st)
810 {
811 struct ddf_super *ddf = st->sb;
812 if (ddf == NULL)
813 return;
814 free(ddf->phys);
815 free(ddf->virt);
816 while (ddf->conflist) {
817 struct vcl *v = ddf->conflist;
818 ddf->conflist = v->next;
819 if (v->block_sizes)
820 free(v->block_sizes);
821 free(v);
822 }
823 while (ddf->dlist) {
824 struct dl *d = ddf->dlist;
825 ddf->dlist = d->next;
826 if (d->fd >= 0)
827 close(d->fd);
828 if (d->spare)
829 free(d->spare);
830 free(d);
831 }
832 free(ddf);
833 st->sb = NULL;
834 }
835
836 static struct supertype *match_metadata_desc_ddf(char *arg)
837 {
838 /* 'ddf' only support containers */
839 struct supertype *st;
840 if (strcmp(arg, "ddf") != 0 &&
841 strcmp(arg, "default") != 0
842 )
843 return NULL;
844
845 st = malloc(sizeof(*st));
846 memset(st, 0, sizeof(*st));
847 st->ss = &super_ddf;
848 st->max_devs = 512;
849 st->minor_version = 0;
850 st->sb = NULL;
851 return st;
852 }
853
854
855 #ifndef MDASSEMBLE
856
857 static mapping_t ddf_state[] = {
858 { "Optimal", 0},
859 { "Degraded", 1},
860 { "Deleted", 2},
861 { "Missing", 3},
862 { "Failed", 4},
863 { "Partially Optimal", 5},
864 { "-reserved-", 6},
865 { "-reserved-", 7},
866 { NULL, 0}
867 };
868
869 static mapping_t ddf_init_state[] = {
870 { "Not Initialised", 0},
871 { "QuickInit in Progress", 1},
872 { "Fully Initialised", 2},
873 { "*UNKNOWN*", 3},
874 { NULL, 0}
875 };
876 static mapping_t ddf_access[] = {
877 { "Read/Write", 0},
878 { "Reserved", 1},
879 { "Read Only", 2},
880 { "Blocked (no access)", 3},
881 { NULL ,0}
882 };
883
884 static mapping_t ddf_level[] = {
885 { "RAID0", DDF_RAID0},
886 { "RAID1", DDF_RAID1},
887 { "RAID3", DDF_RAID3},
888 { "RAID4", DDF_RAID4},
889 { "RAID5", DDF_RAID5},
890 { "RAID1E",DDF_RAID1E},
891 { "JBOD", DDF_JBOD},
892 { "CONCAT",DDF_CONCAT},
893 { "RAID5E",DDF_RAID5E},
894 { "RAID5EE",DDF_RAID5EE},
895 { "RAID6", DDF_RAID6},
896 { NULL, 0}
897 };
898 static mapping_t ddf_sec_level[] = {
899 { "Striped", DDF_2STRIPED},
900 { "Mirrored", DDF_2MIRRORED},
901 { "Concat", DDF_2CONCAT},
902 { "Spanned", DDF_2SPANNED},
903 { NULL, 0}
904 };
905 #endif
906
907 struct num_mapping {
908 int num1, num2;
909 };
910 static struct num_mapping ddf_level_num[] = {
911 { DDF_RAID0, 0 },
912 { DDF_RAID1, 1 },
913 { DDF_RAID3, LEVEL_UNSUPPORTED },
914 { DDF_RAID4, 4 },
915 { DDF_RAID5, 5 },
916 { DDF_RAID1E, LEVEL_UNSUPPORTED },
917 { DDF_JBOD, LEVEL_UNSUPPORTED },
918 { DDF_CONCAT, LEVEL_LINEAR },
919 { DDF_RAID5E, LEVEL_UNSUPPORTED },
920 { DDF_RAID5EE, LEVEL_UNSUPPORTED },
921 { DDF_RAID6, 6},
922 { MAXINT, MAXINT }
923 };
924
925 static int map_num1(struct num_mapping *map, int num)
926 {
927 int i;
928 for (i=0 ; map[i].num1 != MAXINT; i++)
929 if (map[i].num1 == num)
930 break;
931 return map[i].num2;
932 }
933
934 #ifndef MDASSEMBLE
935 static void print_guid(char *guid, int tstamp)
936 {
937 /* A GUIDs are part (or all) ASCII and part binary.
938 * They tend to be space padded.
939 * We print the GUID in HEX, then in parentheses add
940 * any initial ASCII sequence, and a possible
941 * time stamp from bytes 16-19
942 */
943 int l = DDF_GUID_LEN;
944 int i;
945
946 for (i=0 ; i<DDF_GUID_LEN ; i++) {
947 if ((i&3)==0 && i != 0) printf(":");
948 printf("%02X", guid[i]&255);
949 }
950
951 printf(" (");
952 while (l && guid[l-1] == ' ')
953 l--;
954 for (i=0 ; i<l ; i++) {
955 if (guid[i] >= 0x20 && guid[i] < 0x7f)
956 fputc(guid[i], stdout);
957 else
958 break;
959 }
960 if (tstamp) {
961 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
962 char tbuf[100];
963 struct tm *tm;
964 tm = localtime(&then);
965 strftime(tbuf, 100, " %D %T",tm);
966 fputs(tbuf, stdout);
967 }
968 printf(")");
969 }
970
971 static void examine_vd(int n, struct ddf_super *sb, char *guid)
972 {
973 int crl = sb->conf_rec_len;
974 struct vcl *vcl;
975
976 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
977 struct vd_config *vc = &vcl->conf;
978
979 if (calc_crc(vc, crl*512) != vc->crc)
980 continue;
981 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
982 continue;
983
984 /* Ok, we know about this VD, let's give more details */
985 printf(" Raid Devices[%d] : %d\n", n,
986 __be16_to_cpu(vc->prim_elmnt_count));
987 printf(" Chunk Size[%d] : %d sectors\n", n,
988 1 << vc->chunk_shift);
989 printf(" Raid Level[%d] : %s\n", n,
990 map_num(ddf_level, vc->prl)?:"-unknown-");
991 if (vc->sec_elmnt_count != 1) {
992 printf(" Secondary Position[%d] : %d of %d\n", n,
993 vc->sec_elmnt_seq, vc->sec_elmnt_count);
994 printf(" Secondary Level[%d] : %s\n", n,
995 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
996 }
997 printf(" Device Size[%d] : %llu\n", n,
998 __be64_to_cpu(vc->blocks)/2);
999 printf(" Array Size[%d] : %llu\n", n,
1000 __be64_to_cpu(vc->array_blocks)/2);
1001 }
1002 }
1003
1004 static void examine_vds(struct ddf_super *sb)
1005 {
1006 int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1007 int i;
1008 printf(" Virtual Disks : %d\n", cnt);
1009
1010 for (i=0; i<cnt; i++) {
1011 struct virtual_entry *ve = &sb->virt->entries[i];
1012 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1013 printf("\n");
1014 printf(" unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1015 printf(" state[%d] : %s, %s%s\n", i,
1016 map_num(ddf_state, ve->state & 7),
1017 (ve->state & 8) ? "Morphing, ": "",
1018 (ve->state & 16)? "Not Consistent" : "Consistent");
1019 printf(" init state[%d] : %s\n", i,
1020 map_num(ddf_init_state, ve->init_state&3));
1021 printf(" access[%d] : %s\n", i,
1022 map_num(ddf_access, (ve->init_state>>6) & 3));
1023 printf(" Name[%d] : %.16s\n", i, ve->name);
1024 examine_vd(i, sb, ve->guid);
1025 }
1026 if (cnt) printf("\n");
1027 }
1028
1029 static void examine_pds(struct ddf_super *sb)
1030 {
1031 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1032 int i;
1033 struct dl *dl;
1034 printf(" Physical Disks : %d\n", cnt);
1035
1036 for (i=0 ; i<cnt ; i++) {
1037 struct phys_disk_entry *pd = &sb->phys->entries[i];
1038 int type = __be16_to_cpu(pd->type);
1039 int state = __be16_to_cpu(pd->state);
1040
1041 printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1042 printf("\n");
1043 printf(" ref[%d] : %08x\n", i,
1044 __be32_to_cpu(pd->refnum));
1045 printf(" mode[%d] : %s%s%s%s%s\n", i,
1046 (type&2) ? "active":"",
1047 (type&4) ? "Global Spare":"",
1048 (type&8) ? "spare" : "",
1049 (type&16)? ", foreign" : "",
1050 (type&32)? "pass-through" : "");
1051 printf(" state[%d] : %s%s%s%s%s%s%s\n", i,
1052 (state&1)? "Online": "Offline",
1053 (state&2)? ", Failed": "",
1054 (state&4)? ", Rebuilding": "",
1055 (state&8)? ", in-transition": "",
1056 (state&16)? ", SMART errors": "",
1057 (state&32)? ", Unrecovered Read Errors": "",
1058 (state&64)? ", Missing" : "");
1059 printf(" Avail Size[%d] : %llu K\n", i,
1060 __be64_to_cpu(pd->config_size)>>1);
1061 for (dl = sb->dlist; dl ; dl = dl->next) {
1062 if (dl->disk.refnum == pd->refnum) {
1063 char *dv = map_dev(dl->major, dl->minor, 0);
1064 if (dv)
1065 printf(" Device[%d] : %s\n",
1066 i, dv);
1067 }
1068 }
1069 printf("\n");
1070 }
1071 }
1072
1073 static void examine_super_ddf(struct supertype *st, char *homehost)
1074 {
1075 struct ddf_super *sb = st->sb;
1076
1077 printf(" Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1078 printf(" Version : %.8s\n", sb->anchor.revision);
1079 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1080 printf("\n");
1081 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1082 printf("\n");
1083 printf(" Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1084 printf(" Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1085 ?"yes" : "no");
1086 examine_vds(sb);
1087 examine_pds(sb);
1088 }
1089
1090 static void brief_examine_super_ddf(struct supertype *st)
1091 {
1092 /* We just write a generic DDF ARRAY entry
1093 * The uuid is all hex, 6 groups of 4 bytes
1094 */
1095 struct ddf_super *ddf = st->sb;
1096 int i;
1097 printf("ARRAY /dev/ddf metadata=ddf UUID=");
1098 for (i = 0; i < DDF_GUID_LEN; i++) {
1099 if ((i&3) == 0 && i != 0)
1100 printf(":");
1101 printf("%02X", 255&ddf->anchor.guid[i]);
1102 }
1103 printf("\n");
1104 }
1105
1106 static void detail_super_ddf(struct supertype *st, char *homehost)
1107 {
1108 /* FIXME later
1109 * Could print DDF GUID
1110 * Need to find which array
1111 * If whole, briefly list all arrays
1112 * If one, give name
1113 */
1114 }
1115
1116 static void brief_detail_super_ddf(struct supertype *st)
1117 {
1118 /* FIXME I really need to know which array we are detailing.
1119 * Can that be stored in ddf_super??
1120 */
1121 // struct ddf_super *ddf = st->sb;
1122 }
1123 #endif
1124
1125 static int match_home_ddf(struct supertype *st, char *homehost)
1126 {
1127 /* It matches 'this' host if the controller is a
1128 * Linux-MD controller with vendor_data matching
1129 * the hostname
1130 */
1131 struct ddf_super *ddf = st->sb;
1132 int len = strlen(homehost);
1133
1134 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1135 len < sizeof(ddf->controller.vendor_data) &&
1136 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1137 ddf->controller.vendor_data[len] == 0);
1138 }
1139
1140 static struct vd_config *find_vdcr(struct ddf_super *ddf, int inst)
1141 {
1142 struct vcl *v;
1143
1144 for (v = ddf->conflist; v; v = v->next)
1145 if (inst == v->vcnum)
1146 return &v->conf;
1147 return NULL;
1148 }
1149
1150 static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1151 {
1152 /* Find the entry in phys_disk which has the given refnum
1153 * and return it's index
1154 */
1155 int i;
1156 for (i=0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1157 if (ddf->phys->entries[i].refnum == phys_refnum)
1158 return i;
1159 return -1;
1160 }
1161
1162 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1163 {
1164 /* The uuid returned here is used for:
1165 * uuid to put into bitmap file (Create, Grow)
1166 * uuid for backup header when saving critical section (Grow)
1167 * comparing uuids when re-adding a device into an array
1168 * For each of these we can make do with a truncated
1169 * or hashed uuid rather than the original, as long as
1170 * everyone agrees.
1171 * In each case the uuid required is that of the data-array,
1172 * not the device-set.
1173 * In the case of SVD we assume the BVD is of interest,
1174 * though that might be the case if a bitmap were made for
1175 * a mirrored SVD - worry about that later.
1176 * So we need to find the VD configuration record for the
1177 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1178 * The first 16 bytes of the sha1 of these is used.
1179 */
1180 struct ddf_super *ddf = st->sb;
1181 struct vcl *vcl = ddf->currentconf;
1182
1183 if (!vcl)
1184 memset(uuid, 0, sizeof (uuid));
1185 else {
1186 char buf[20];
1187 struct sha1_ctx ctx;
1188 sha1_init_ctx(&ctx);
1189 sha1_process_bytes(&vcl->conf.guid, DDF_GUID_LEN, &ctx);
1190 if (vcl->conf.sec_elmnt_count > 1)
1191 sha1_process_bytes(&vcl->conf.sec_elmnt_seq, 1, &ctx);
1192 sha1_finish_ctx(&ctx, buf);
1193 memcpy(uuid, buf, sizeof(uuid));
1194 }
1195 }
1196
1197 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info);
1198
1199 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info)
1200 {
1201 struct ddf_super *ddf = st->sb;
1202
1203 if (ddf->currentconf) {
1204 getinfo_super_ddf_bvd(st, info);
1205 return;
1206 }
1207
1208 info->array.raid_disks = __be16_to_cpu(ddf->phys->used_pdes);
1209 info->array.level = LEVEL_CONTAINER;
1210 info->array.layout = 0;
1211 info->array.md_minor = -1;
1212 info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)
1213 (ddf->anchor.guid+16));
1214 info->array.utime = 0;
1215 info->array.chunk_size = 0;
1216
1217
1218 info->disk.major = 0;
1219 info->disk.minor = 0;
1220 if (ddf->dlist) {
1221 info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1222 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1223
1224 info->data_offset = __be64_to_cpu(ddf->phys->
1225 entries[info->disk.raid_disk].
1226 config_size);
1227 info->component_size = ddf->dlist->size - info->data_offset;
1228 } else {
1229 info->disk.number = -1;
1230 // info->disk.raid_disk = find refnum in the table and use index;
1231 }
1232 info->disk.state = (1 << MD_DISK_SYNC);
1233
1234
1235 info->reshape_active = 0;
1236
1237 strcpy(info->text_version, "ddf");
1238
1239 // uuid_from_super_ddf(info->uuid, sbv);
1240
1241 // info->name[] ?? ;
1242 }
1243
1244 static int rlq_to_layout(int rlq, int prl, int raiddisks);
1245
1246 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info)
1247 {
1248 struct ddf_super *ddf = st->sb;
1249 struct vcl *vc = ddf->currentconf;
1250 int cd = ddf->currentdev;
1251
1252 /* FIXME this returns BVD info - what if we want SVD ?? */
1253
1254 info->array.raid_disks = __be16_to_cpu(vc->conf.prim_elmnt_count);
1255 info->array.level = map_num1(ddf_level_num, vc->conf.prl);
1256 info->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
1257 info->array.raid_disks);
1258 info->array.md_minor = -1;
1259 info->array.ctime = DECADE +
1260 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
1261 info->array.utime = DECADE + __be32_to_cpu(vc->conf.timestamp);
1262 info->array.chunk_size = 512 << vc->conf.chunk_shift;
1263
1264 if (cd >= 0 && cd < ddf->mppe) {
1265 info->data_offset = __be64_to_cpu(vc->lba_offset[cd]);
1266 if (vc->block_sizes)
1267 info->component_size = vc->block_sizes[cd];
1268 else
1269 info->component_size = __be64_to_cpu(vc->conf.blocks);
1270 }
1271
1272 info->disk.major = 0;
1273 info->disk.minor = 0;
1274 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1275 // info->disk.raid_disk = find refnum in the table and use index;
1276 // info->disk.state = ???;
1277
1278 info->resync_start = 0;
1279 if (!(ddf->virt->entries[info->container_member].state
1280 & DDF_state_inconsistent) &&
1281 (ddf->virt->entries[info->container_member].init_state
1282 & DDF_initstate_mask)
1283 == DDF_init_full)
1284 info->resync_start = ~0ULL;
1285
1286 uuid_from_super_ddf(st, info->uuid);
1287
1288 sprintf(info->text_version, "/%s/%d",
1289 devnum2devname(st->container_dev),
1290 info->container_member);
1291
1292 // info->name[] ?? ;
1293 }
1294
1295
1296 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1297 char *update,
1298 char *devname, int verbose,
1299 int uuid_set, char *homehost)
1300 {
1301 /* For 'assemble' and 'force' we need to return non-zero if any
1302 * change was made. For others, the return value is ignored.
1303 * Update options are:
1304 * force-one : This device looks a bit old but needs to be included,
1305 * update age info appropriately.
1306 * assemble: clear any 'faulty' flag to allow this device to
1307 * be assembled.
1308 * force-array: Array is degraded but being forced, mark it clean
1309 * if that will be needed to assemble it.
1310 *
1311 * newdev: not used ????
1312 * grow: Array has gained a new device - this is currently for
1313 * linear only
1314 * resync: mark as dirty so a resync will happen.
1315 * uuid: Change the uuid of the array to match what is given
1316 * homehost: update the recorded homehost
1317 * name: update the name - preserving the homehost
1318 * _reshape_progress: record new reshape_progress position.
1319 *
1320 * Following are not relevant for this version:
1321 * sparc2.2 : update from old dodgey metadata
1322 * super-minor: change the preferred_minor number
1323 * summaries: update redundant counters.
1324 */
1325 int rv = 0;
1326 // struct ddf_super *ddf = st->sb;
1327 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1328 // struct virtual_entry *ve = find_ve(ddf);
1329
1330 /* we don't need to handle "force-*" or "assemble" as
1331 * there is no need to 'trick' the kernel. We the metadata is
1332 * first updated to activate the array, all the implied modifications
1333 * will just happen.
1334 */
1335
1336 if (strcmp(update, "grow") == 0) {
1337 /* FIXME */
1338 }
1339 if (strcmp(update, "resync") == 0) {
1340 // info->resync_checkpoint = 0;
1341 }
1342 /* We ignore UUID updates as they make even less sense
1343 * with DDF
1344 */
1345 if (strcmp(update, "homehost") == 0) {
1346 /* homehost is stored in controller->vendor_data,
1347 * or it is when we are the vendor
1348 */
1349 // if (info->vendor_is_local)
1350 // strcpy(ddf->controller.vendor_data, homehost);
1351 }
1352 if (strcmp(update, "name") == 0) {
1353 /* name is stored in virtual_entry->name */
1354 // memset(ve->name, ' ', 16);
1355 // strncpy(ve->name, info->name, 16);
1356 }
1357 if (strcmp(update, "_reshape_progress") == 0) {
1358 /* We don't support reshape yet */
1359 }
1360
1361 // update_all_csum(ddf);
1362
1363 return rv;
1364 }
1365
1366 static void make_header_guid(char *guid)
1367 {
1368 __u32 stamp;
1369 int rfd;
1370 /* Create a DDF Header of Virtual Disk GUID */
1371
1372 /* 24 bytes of fiction required.
1373 * first 8 are a 'vendor-id' - "Linux-MD"
1374 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1375 * Remaining 8 random number plus timestamp
1376 */
1377 memcpy(guid, T10, sizeof(T10));
1378 stamp = __cpu_to_be32(0xdeadbeef);
1379 memcpy(guid+8, &stamp, 4);
1380 stamp = __cpu_to_be32(0);
1381 memcpy(guid+12, &stamp, 4);
1382 stamp = __cpu_to_be32(time(0) - DECADE);
1383 memcpy(guid+16, &stamp, 4);
1384 rfd = open("/dev/urandom", O_RDONLY);
1385 if (rfd < 0 || read(rfd, &stamp, 4) != 4)
1386 stamp = random();
1387 memcpy(guid+20, &stamp, 4);
1388 if (rfd >= 0) close(rfd);
1389 }
1390
1391 static int init_super_ddf_bvd(struct supertype *st,
1392 mdu_array_info_t *info,
1393 unsigned long long size,
1394 char *name, char *homehost,
1395 int *uuid);
1396
1397 static int init_super_ddf(struct supertype *st,
1398 mdu_array_info_t *info,
1399 unsigned long long size, char *name, char *homehost,
1400 int *uuid)
1401 {
1402 /* This is primarily called by Create when creating a new array.
1403 * We will then get add_to_super called for each component, and then
1404 * write_init_super called to write it out to each device.
1405 * For DDF, Create can create on fresh devices or on a pre-existing
1406 * array.
1407 * To create on a pre-existing array a different method will be called.
1408 * This one is just for fresh drives.
1409 *
1410 * We need to create the entire 'ddf' structure which includes:
1411 * DDF headers - these are easy.
1412 * Controller data - a Sector describing this controller .. not that
1413 * this is a controller exactly.
1414 * Physical Disk Record - one entry per device, so
1415 * leave plenty of space.
1416 * Virtual Disk Records - again, just leave plenty of space.
1417 * This just lists VDs, doesn't give details
1418 * Config records - describes the VDs that use this disk
1419 * DiskData - describes 'this' device.
1420 * BadBlockManagement - empty
1421 * Diag Space - empty
1422 * Vendor Logs - Could we put bitmaps here?
1423 *
1424 */
1425 struct ddf_super *ddf;
1426 char hostname[17];
1427 int hostlen;
1428 int max_phys_disks, max_virt_disks;
1429 unsigned long long sector;
1430 int clen;
1431 int i;
1432 int pdsize, vdsize;
1433 struct phys_disk *pd;
1434 struct virtual_disk *vd;
1435
1436 if (!info) {
1437 st->sb = NULL;
1438 return 0;
1439 }
1440 if (st->sb)
1441 return init_super_ddf_bvd(st, info, size, name, homehost,
1442 uuid);
1443
1444 ddf = malloc(sizeof(*ddf));
1445 memset(ddf, 0, sizeof(*ddf));
1446 ddf->dlist = NULL; /* no physical disks yet */
1447 ddf->conflist = NULL; /* No virtual disks yet */
1448
1449 /* At least 32MB *must* be reserved for the ddf. So let's just
1450 * start 32MB from the end, and put the primary header there.
1451 * Don't do secondary for now.
1452 * We don't know exactly where that will be yet as it could be
1453 * different on each device. To just set up the lengths.
1454 *
1455 */
1456
1457 ddf->anchor.magic = DDF_HEADER_MAGIC;
1458 make_header_guid(ddf->anchor.guid);
1459
1460 memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
1461 ddf->anchor.seq = __cpu_to_be32(1);
1462 ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1463 ddf->anchor.openflag = 0xFF;
1464 ddf->anchor.foreignflag = 0;
1465 ddf->anchor.enforcegroups = 0; /* Is this best?? */
1466 ddf->anchor.pad0 = 0xff;
1467 memset(ddf->anchor.pad1, 0xff, 12);
1468 memset(ddf->anchor.header_ext, 0xff, 32);
1469 ddf->anchor.primary_lba = ~(__u64)0;
1470 ddf->anchor.secondary_lba = ~(__u64)0;
1471 ddf->anchor.type = DDF_HEADER_ANCHOR;
1472 memset(ddf->anchor.pad2, 0xff, 3);
1473 ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1474 ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1475 of 32M reserved.. */
1476 max_phys_disks = 1023; /* Should be enough */
1477 ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1478 max_virt_disks = 255;
1479 ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1480 ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1481 ddf->max_part = 64;
1482 ddf->mppe = 256;
1483 ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
1484 ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1485 ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe);
1486 memset(ddf->anchor.pad3, 0xff, 54);
1487 /* controller sections is one sector long immediately
1488 * after the ddf header */
1489 sector = 1;
1490 ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1491 ddf->anchor.controller_section_length = __cpu_to_be32(1);
1492 sector += 1;
1493
1494 /* phys is 8 sectors after that */
1495 pdsize = ROUND_UP(sizeof(struct phys_disk) +
1496 sizeof(struct phys_disk_entry)*max_phys_disks,
1497 512);
1498 switch(pdsize/512) {
1499 case 2: case 8: case 32: case 128: case 512: break;
1500 default: abort();
1501 }
1502 ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1503 ddf->anchor.phys_section_length =
1504 __cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1505 sector += pdsize/512;
1506
1507 /* virt is another 32 sectors */
1508 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1509 sizeof(struct virtual_entry) * max_virt_disks,
1510 512);
1511 switch(vdsize/512) {
1512 case 2: case 8: case 32: case 128: case 512: break;
1513 default: abort();
1514 }
1515 ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1516 ddf->anchor.virt_section_length =
1517 __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1518 sector += vdsize/512;
1519
1520 clen = ddf->conf_rec_len * (ddf->max_part+1);
1521 ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1522 ddf->anchor.config_section_length = __cpu_to_be32(clen);
1523 sector += clen;
1524
1525 ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1526 ddf->anchor.data_section_length = __cpu_to_be32(1);
1527 sector += 1;
1528
1529 ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1530 ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1531 ddf->anchor.diag_space_length = __cpu_to_be32(0);
1532 ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1533 ddf->anchor.vendor_length = __cpu_to_be32(0);
1534 ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1535
1536 memset(ddf->anchor.pad4, 0xff, 256);
1537
1538 memcpy(&ddf->primary, &ddf->anchor, 512);
1539 memcpy(&ddf->secondary, &ddf->anchor, 512);
1540
1541 ddf->primary.openflag = 1; /* I guess.. */
1542 ddf->primary.type = DDF_HEADER_PRIMARY;
1543
1544 ddf->secondary.openflag = 1; /* I guess.. */
1545 ddf->secondary.type = DDF_HEADER_SECONDARY;
1546
1547 ddf->active = &ddf->primary;
1548
1549 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1550
1551 /* 24 more bytes of fiction required.
1552 * first 8 are a 'vendor-id' - "Linux-MD"
1553 * Remaining 16 are serial number.... maybe a hostname would do?
1554 */
1555 memcpy(ddf->controller.guid, T10, sizeof(T10));
1556 gethostname(hostname, sizeof(hostname));
1557 hostname[sizeof(hostname) - 1] = 0;
1558 hostlen = strlen(hostname);
1559 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1560 for (i = strlen(T10) ; i+hostlen < 24; i++)
1561 ddf->controller.guid[i] = ' ';
1562
1563 ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1564 ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1565 ddf->controller.type.sub_vendor_id = 0;
1566 ddf->controller.type.sub_device_id = 0;
1567 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1568 memset(ddf->controller.pad, 0xff, 8);
1569 memset(ddf->controller.vendor_data, 0xff, 448);
1570
1571 pd = ddf->phys = malloc(pdsize);
1572 ddf->pdsize = pdsize;
1573
1574 memset(pd, 0xff, pdsize);
1575 memset(pd, 0, sizeof(*pd));
1576 pd->magic = DDF_PHYS_DATA_MAGIC;
1577 pd->used_pdes = __cpu_to_be16(0);
1578 pd->max_pdes = __cpu_to_be16(max_phys_disks);
1579 memset(pd->pad, 0xff, 52);
1580
1581 vd = ddf->virt = malloc(vdsize);
1582 ddf->vdsize = vdsize;
1583 memset(vd, 0, vdsize);
1584 vd->magic = DDF_VIRT_RECORDS_MAGIC;
1585 vd->populated_vdes = __cpu_to_be16(0);
1586 vd->max_vdes = __cpu_to_be16(max_virt_disks);
1587 memset(vd->pad, 0xff, 52);
1588
1589 for (i=0; i<max_virt_disks; i++)
1590 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1591
1592 st->sb = ddf;
1593 return 1;
1594 }
1595
1596 static int all_ff(char *guid)
1597 {
1598 int i;
1599 for (i = 0; i < DDF_GUID_LEN; i++)
1600 if (guid[i] != (char)0xff)
1601 return 0;
1602 return 1;
1603 }
1604 static int chunk_to_shift(int chunksize)
1605 {
1606 return ffs(chunksize/512)-1;
1607 }
1608
1609 static int level_to_prl(int level)
1610 {
1611 switch (level) {
1612 case LEVEL_LINEAR: return DDF_CONCAT;
1613 case 0: return DDF_RAID0;
1614 case 1: return DDF_RAID1;
1615 case 4: return DDF_RAID4;
1616 case 5: return DDF_RAID5;
1617 case 6: return DDF_RAID6;
1618 default: return -1;
1619 }
1620 }
1621 static int layout_to_rlq(int level, int layout, int raiddisks)
1622 {
1623 switch(level) {
1624 case 0:
1625 return DDF_RAID0_SIMPLE;
1626 case 1:
1627 switch(raiddisks) {
1628 case 2: return DDF_RAID1_SIMPLE;
1629 case 3: return DDF_RAID1_MULTI;
1630 default: return -1;
1631 }
1632 case 4:
1633 switch(layout) {
1634 case 0: return DDF_RAID4_N;
1635 }
1636 break;
1637 case 5:
1638 case 6:
1639 switch(layout) {
1640 case ALGORITHM_LEFT_ASYMMETRIC:
1641 return DDF_RAID5_N_RESTART;
1642 case ALGORITHM_RIGHT_ASYMMETRIC:
1643 if (level == 5)
1644 return DDF_RAID5_0_RESTART;
1645 else
1646 return DDF_RAID6_0_RESTART;
1647 case ALGORITHM_LEFT_SYMMETRIC:
1648 return DDF_RAID5_N_CONTINUE;
1649 case ALGORITHM_RIGHT_SYMMETRIC:
1650 return -1; /* not mentioned in standard */
1651 }
1652 }
1653 return -1;
1654 }
1655
1656 static int rlq_to_layout(int rlq, int prl, int raiddisks)
1657 {
1658 switch(prl) {
1659 case DDF_RAID0:
1660 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1661 case DDF_RAID1:
1662 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1663 on raiddisks*/
1664 case DDF_RAID4:
1665 switch(rlq) {
1666 case DDF_RAID4_N:
1667 return 0;
1668 default:
1669 /* not supported */
1670 return -1; /* FIXME this isn't checked */
1671 }
1672 case DDF_RAID5:
1673 switch(rlq) {
1674 case DDF_RAID5_N_RESTART:
1675 return ALGORITHM_LEFT_ASYMMETRIC;
1676 case DDF_RAID5_0_RESTART:
1677 return ALGORITHM_RIGHT_ASYMMETRIC;
1678 case DDF_RAID5_N_CONTINUE:
1679 return ALGORITHM_LEFT_SYMMETRIC;
1680 default:
1681 return -1;
1682 }
1683 case DDF_RAID6:
1684 switch(rlq) {
1685 case DDF_RAID5_N_RESTART:
1686 return ALGORITHM_LEFT_ASYMMETRIC;
1687 case DDF_RAID6_0_RESTART:
1688 return ALGORITHM_RIGHT_ASYMMETRIC;
1689 case DDF_RAID5_N_CONTINUE:
1690 return ALGORITHM_LEFT_SYMMETRIC;
1691 default:
1692 return -1;
1693 }
1694 }
1695 return -1;
1696 }
1697
1698 struct extent {
1699 unsigned long long start, size;
1700 };
1701 static int cmp_extent(const void *av, const void *bv)
1702 {
1703 const struct extent *a = av;
1704 const struct extent *b = bv;
1705 if (a->start < b->start)
1706 return -1;
1707 if (a->start > b->start)
1708 return 1;
1709 return 0;
1710 }
1711
1712 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
1713 {
1714 /* find a list of used extents on the give physical device
1715 * (dnum) of the given ddf.
1716 * Return a malloced array of 'struct extent'
1717
1718 FIXME ignore DDF_Legacy devices?
1719
1720 */
1721 struct extent *rv;
1722 int n = 0;
1723 int i, j;
1724
1725 rv = malloc(sizeof(struct extent) * (ddf->max_part + 2));
1726 if (!rv)
1727 return NULL;
1728
1729 for (i = 0; i < ddf->max_part; i++) {
1730 struct vcl *v = dl->vlist[i];
1731 if (v == NULL)
1732 continue;
1733 for (j=0; j < v->conf.prim_elmnt_count; j++)
1734 if (v->conf.phys_refnum[j] == dl->disk.refnum) {
1735 /* This device plays role 'j' in 'v'. */
1736 rv[n].start = __be64_to_cpu(v->lba_offset[j]);
1737 rv[n].size = __be64_to_cpu(v->conf.blocks);
1738 n++;
1739 break;
1740 }
1741 }
1742 qsort(rv, n, sizeof(*rv), cmp_extent);
1743
1744 rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
1745 rv[n].size = 0;
1746 return rv;
1747 }
1748
1749 static int init_super_ddf_bvd(struct supertype *st,
1750 mdu_array_info_t *info,
1751 unsigned long long size,
1752 char *name, char *homehost,
1753 int *uuid)
1754 {
1755 /* We are creating a BVD inside a pre-existing container.
1756 * so st->sb is already set.
1757 * We need to create a new vd_config and a new virtual_entry
1758 */
1759 struct ddf_super *ddf = st->sb;
1760 int venum;
1761 struct virtual_entry *ve;
1762 struct vcl *vcl;
1763 struct vd_config *vc;
1764
1765 if (__be16_to_cpu(ddf->virt->populated_vdes)
1766 >= __be16_to_cpu(ddf->virt->max_vdes)) {
1767 fprintf(stderr, Name": This ddf already has the "
1768 "maximum of %d virtual devices\n",
1769 __be16_to_cpu(ddf->virt->max_vdes));
1770 return 0;
1771 }
1772
1773 for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1774 if (all_ff(ddf->virt->entries[venum].guid))
1775 break;
1776 if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1777 fprintf(stderr, Name ": Cannot find spare slot for "
1778 "virtual disk - DDF is corrupt\n");
1779 return 0;
1780 }
1781 ve = &ddf->virt->entries[venum];
1782
1783 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1784 * timestamp, random number
1785 */
1786 make_header_guid(ve->guid);
1787 ve->unit = __cpu_to_be16(info->md_minor);
1788 ve->pad0 = 0xFFFF;
1789 ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1790 ve->type = 0;
1791 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
1792 if (info->state & 1) /* clean */
1793 ve->init_state = DDF_init_full;
1794 else
1795 ve->init_state = DDF_init_not;
1796
1797 memset(ve->pad1, 0xff, 14);
1798 memset(ve->name, ' ', 16);
1799 if (name)
1800 strncpy(ve->name, name, 16);
1801 ddf->virt->populated_vdes =
1802 __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
1803
1804 /* Now create a new vd_config */
1805 vcl = malloc(offsetof(struct vcl, conf) + ddf->conf_rec_len * 512);
1806 vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
1807 vcl->vcnum = venum;
1808 sprintf(st->subarray, "%d", venum);
1809 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1810
1811 vc = &vcl->conf;
1812
1813 vc->magic = DDF_VD_CONF_MAGIC;
1814 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
1815 vc->timestamp = __cpu_to_be32(time(0)-DECADE);
1816 vc->seqnum = __cpu_to_be32(1);
1817 memset(vc->pad0, 0xff, 24);
1818 vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
1819 vc->chunk_shift = chunk_to_shift(info->chunk_size);
1820 vc->prl = level_to_prl(info->level);
1821 vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
1822 vc->sec_elmnt_count = 1;
1823 vc->sec_elmnt_seq = 0;
1824 vc->srl = 0;
1825 vc->blocks = __cpu_to_be64(info->size * 2);
1826 vc->array_blocks = __cpu_to_be64(
1827 calc_array_size(info->level, info->raid_disks, info->layout,
1828 info->chunk_size, info->size*2));
1829 memset(vc->pad1, 0xff, 8);
1830 vc->spare_refs[0] = 0xffffffff;
1831 vc->spare_refs[1] = 0xffffffff;
1832 vc->spare_refs[2] = 0xffffffff;
1833 vc->spare_refs[3] = 0xffffffff;
1834 vc->spare_refs[4] = 0xffffffff;
1835 vc->spare_refs[5] = 0xffffffff;
1836 vc->spare_refs[6] = 0xffffffff;
1837 vc->spare_refs[7] = 0xffffffff;
1838 memset(vc->cache_pol, 0, 8);
1839 vc->bg_rate = 0x80;
1840 memset(vc->pad2, 0xff, 3);
1841 memset(vc->pad3, 0xff, 52);
1842 memset(vc->pad4, 0xff, 192);
1843 memset(vc->v0, 0xff, 32);
1844 memset(vc->v1, 0xff, 32);
1845 memset(vc->v2, 0xff, 16);
1846 memset(vc->v3, 0xff, 16);
1847 memset(vc->vendor, 0xff, 32);
1848
1849 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
1850 memset(vc->phys_refnum+(ddf->mppe * 4), 0x00, 8*ddf->mppe);
1851
1852 vcl->next = ddf->conflist;
1853 ddf->conflist = vcl;
1854 ddf->currentconf = vcl;
1855 return 1;
1856 }
1857
1858 static void add_to_super_ddf_bvd(struct supertype *st,
1859 mdu_disk_info_t *dk, int fd, char *devname)
1860 {
1861 /* fd and devname identify a device with-in the ddf container (st).
1862 * dk identifies a location in the new BVD.
1863 * We need to find suitable free space in that device and update
1864 * the phys_refnum and lba_offset for the newly created vd_config.
1865 * We might also want to update the type in the phys_disk
1866 * section.
1867 */
1868 struct dl *dl;
1869 struct ddf_super *ddf = st->sb;
1870 struct vd_config *vc;
1871 __u64 *lba_offset;
1872 int working;
1873 int i;
1874 unsigned long long blocks, pos, esize;
1875 struct extent *ex;
1876
1877 for (dl = ddf->dlist; dl ; dl = dl->next)
1878 if (dl->major == dk->major &&
1879 dl->minor == dk->minor)
1880 break;
1881 if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
1882 return;
1883
1884 vc = &ddf->currentconf->conf;
1885 lba_offset = ddf->currentconf->lba_offset;
1886
1887 ex = get_extents(ddf, dl);
1888 if (!ex)
1889 return;
1890
1891 i = 0; pos = 0;
1892 blocks = __be64_to_cpu(vc->blocks);
1893 if (ddf->currentconf->block_sizes)
1894 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
1895
1896 do {
1897 esize = ex[i].start - pos;
1898 if (esize >= blocks)
1899 break;
1900 pos = ex[i].start + ex[i].size;
1901 i++;
1902 } while (ex[i-1].size);
1903
1904 free(ex);
1905 if (esize < blocks)
1906 return;
1907
1908 ddf->currentdev = dk->raid_disk;
1909 vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
1910 lba_offset[dk->raid_disk] = __cpu_to_be64(pos);
1911
1912 for (i=0; i < ddf->max_part ; i++)
1913 if (dl->vlist[i] == NULL)
1914 break;
1915 if (i == ddf->max_part)
1916 return;
1917 dl->vlist[i] = ddf->currentconf;
1918
1919 dl->fd = fd;
1920 dl->devname = devname;
1921
1922 /* Check how many working raid_disks, and if we can mark
1923 * array as optimal yet
1924 */
1925 working = 0;
1926
1927 for (i=0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
1928 if (vc->phys_refnum[i] != 0xffffffff)
1929 working++;
1930
1931 /* Find which virtual_entry */
1932 i = ddf->currentconf->vcnum;
1933 if (working == __be16_to_cpu(vc->prim_elmnt_count))
1934 ddf->virt->entries[i].state =
1935 (ddf->virt->entries[i].state & ~DDF_state_mask)
1936 | DDF_state_optimal;
1937
1938 if (vc->prl == DDF_RAID6 &&
1939 working+1 == __be16_to_cpu(vc->prim_elmnt_count))
1940 ddf->virt->entries[i].state =
1941 (ddf->virt->entries[i].state & ~DDF_state_mask)
1942 | DDF_state_part_optimal;
1943
1944 ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
1945 ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
1946 }
1947
1948 /* add a device to a container, either while creating it or while
1949 * expanding a pre-existing container
1950 */
1951 static void add_to_super_ddf(struct supertype *st,
1952 mdu_disk_info_t *dk, int fd, char *devname)
1953 {
1954 struct ddf_super *ddf = st->sb;
1955 struct dl *dd;
1956 time_t now;
1957 struct tm *tm;
1958 unsigned long long size;
1959 struct phys_disk_entry *pde;
1960 int n, i;
1961 struct stat stb;
1962
1963 if (ddf->currentconf) {
1964 add_to_super_ddf_bvd(st, dk, fd, devname);
1965 return;
1966 }
1967
1968 /* This is device numbered dk->number. We need to create
1969 * a phys_disk entry and a more detailed disk_data entry.
1970 */
1971 fstat(fd, &stb);
1972 dd = malloc(sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part);
1973 dd->major = major(stb.st_rdev);
1974 dd->minor = minor(stb.st_rdev);
1975 dd->devname = devname;
1976 dd->next = ddf->dlist;
1977 dd->fd = fd;
1978 dd->spare = NULL;
1979
1980 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
1981 now = time(0);
1982 tm = localtime(&now);
1983 sprintf(dd->disk.guid, "%8s%04d%02d%02d",
1984 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
1985 *(__u32*)(dd->disk.guid + 16) = random();
1986 *(__u32*)(dd->disk.guid + 20) = random();
1987
1988 do {
1989 /* Cannot be bothered finding a CRC of some irrelevant details*/
1990 dd->disk.refnum = random();
1991 for (i = __be16_to_cpu(ddf->active->max_pd_entries) - 1;
1992 i >= 0; i--)
1993 if (ddf->phys->entries[i].refnum == dd->disk.refnum)
1994 break;
1995 } while (i >= 0);
1996
1997 dd->disk.forced_ref = 1;
1998 dd->disk.forced_guid = 1;
1999 memset(dd->disk.vendor, ' ', 32);
2000 memcpy(dd->disk.vendor, "Linux", 5);
2001 memset(dd->disk.pad, 0xff, 442);
2002 for (i = 0; i < ddf->max_part ; i++)
2003 dd->vlist[i] = NULL;
2004
2005 n = __be16_to_cpu(ddf->phys->used_pdes);
2006 pde = &ddf->phys->entries[n];
2007 dd->pdnum = n;
2008
2009 n++;
2010 ddf->phys->used_pdes = __cpu_to_be16(n);
2011
2012 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2013 pde->refnum = dd->disk.refnum;
2014 pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2015 pde->state = __cpu_to_be16(DDF_Online);
2016 get_dev_size(fd, NULL, &size);
2017 /* We are required to reserve 32Meg, and record the size in sectors */
2018 pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
2019 sprintf(pde->path, "%17.17s","Information: nil") ;
2020 memset(pde->pad, 0xff, 6);
2021
2022 dd->size = size >> 9;
2023 ddf->dlist = dd;
2024 }
2025
2026 /*
2027 * This is the write_init_super method for a ddf container. It is
2028 * called when creating a container or adding another device to a
2029 * container.
2030 */
2031
2032 #ifndef MDASSEMBLE
2033 static int __write_init_super_ddf(struct supertype *st, int do_close)
2034 {
2035
2036 struct ddf_super *ddf = st->sb;
2037 int i;
2038 struct dl *d;
2039 int n_config;
2040 int conf_size;
2041
2042 unsigned long long size, sector;
2043
2044 for (d = ddf->dlist; d; d=d->next) {
2045 int fd = d->fd;
2046
2047 if (fd < 0)
2048 continue;
2049
2050 /* We need to fill in the primary, (secondary) and workspace
2051 * lba's in the headers, set their checksums,
2052 * Also checksum phys, virt....
2053 *
2054 * Then write everything out, finally the anchor is written.
2055 */
2056 get_dev_size(fd, NULL, &size);
2057 size /= 512;
2058 ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
2059 ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
2060 ddf->anchor.seq = __cpu_to_be32(1);
2061 memcpy(&ddf->primary, &ddf->anchor, 512);
2062 memcpy(&ddf->secondary, &ddf->anchor, 512);
2063
2064 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
2065 ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
2066 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
2067
2068 ddf->primary.openflag = 0;
2069 ddf->primary.type = DDF_HEADER_PRIMARY;
2070
2071 ddf->secondary.openflag = 0;
2072 ddf->secondary.type = DDF_HEADER_SECONDARY;
2073
2074 ddf->primary.crc = calc_crc(&ddf->primary, 512);
2075 ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
2076
2077 sector = size - 16*1024*2;
2078 lseek64(fd, sector<<9, 0);
2079 write(fd, &ddf->primary, 512);
2080
2081 ddf->controller.crc = calc_crc(&ddf->controller, 512);
2082 write(fd, &ddf->controller, 512);
2083
2084 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
2085
2086 write(fd, ddf->phys, ddf->pdsize);
2087
2088 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
2089 write(fd, ddf->virt, ddf->vdsize);
2090
2091 /* Now write lots of config records. */
2092 n_config = ddf->max_part;
2093 conf_size = ddf->conf_rec_len * 512;
2094 for (i = 0 ; i <= n_config ; i++) {
2095 struct vcl *c = d->vlist[i];
2096 if (i == n_config)
2097 c = (struct vcl*)d->spare;
2098
2099 if (c) {
2100 c->conf.crc = calc_crc(&c->conf, conf_size);
2101 write(fd, &c->conf, conf_size);
2102 } else {
2103 __u32 sig = 0xffffffff;
2104 write(fd, &sig, 4);
2105 lseek64(fd, conf_size-4, SEEK_CUR);
2106 }
2107 }
2108 d->disk.crc = calc_crc(&d->disk, 512);
2109 write(fd, &d->disk, 512);
2110
2111 /* Maybe do the same for secondary */
2112
2113 lseek64(fd, (size-1)*512, SEEK_SET);
2114 write(fd, &ddf->anchor, 512);
2115 if (do_close) {
2116 close(fd);
2117 d->fd = -1;
2118 }
2119 }
2120 return 1;
2121 }
2122
2123 static int write_init_super_ddf(struct supertype *st)
2124 {
2125 return __write_init_super_ddf(st, 1);
2126 }
2127
2128 #endif
2129
2130 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize)
2131 {
2132 /* We must reserve the last 32Meg */
2133 if (devsize <= 32*1024*2)
2134 return 0;
2135 return devsize - 32*1024*2;
2136 }
2137
2138 #ifndef MDASSEMBLE
2139 static int validate_geometry_ddf_container(struct supertype *st,
2140 int level, int layout, int raiddisks,
2141 int chunk, unsigned long long size,
2142 char *dev, unsigned long long *freesize);
2143
2144 static int validate_geometry_ddf_bvd(struct supertype *st,
2145 int level, int layout, int raiddisks,
2146 int chunk, unsigned long long size,
2147 char *dev, unsigned long long *freesize);
2148
2149 static int validate_geometry_ddf(struct supertype *st,
2150 int level, int layout, int raiddisks,
2151 int chunk, unsigned long long size,
2152 char *dev, unsigned long long *freesize)
2153 {
2154 int fd;
2155 struct mdinfo *sra;
2156 int cfd;
2157
2158 /* ddf potentially supports lots of things, but it depends on
2159 * what devices are offered (and maybe kernel version?)
2160 * If given unused devices, we will make a container.
2161 * If given devices in a container, we will make a BVD.
2162 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2163 */
2164
2165 if (level == LEVEL_CONTAINER) {
2166 /* Must be a fresh device to add to a container */
2167 return validate_geometry_ddf_container(st, level, layout,
2168 raiddisks,
2169 chunk, size, dev, freesize);
2170 }
2171
2172 if (st->sb) {
2173 /* A container has already been opened, so we are
2174 * creating in there. Maybe a BVD, maybe an SVD.
2175 * Should make a distinction one day.
2176 */
2177 return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
2178 chunk, size, dev, freesize);
2179 }
2180 if (!dev) {
2181 /* Initial sanity check. Exclude illegal levels. */
2182 int i;
2183 for (i=0; ddf_level_num[i].num1 != MAXINT; i++)
2184 if (ddf_level_num[i].num2 == level)
2185 break;
2186 if (ddf_level_num[i].num1 == MAXINT)
2187 return 0;
2188 /* Should check layout? etc */
2189 return 1;
2190 }
2191
2192 /* This is the first device for the array.
2193 * If it is a container, we read it in and do automagic allocations,
2194 * no other devices should be given.
2195 * Otherwise it must be a member device of a container, and we
2196 * do manual allocation.
2197 * Later we should check for a BVD and make an SVD.
2198 */
2199 fd = open(dev, O_RDONLY|O_EXCL, 0);
2200 if (fd >= 0) {
2201 sra = sysfs_read(fd, 0, GET_VERSION);
2202 close(fd);
2203 if (sra && sra->array.major_version == -1 &&
2204 strcmp(sra->text_version, "ddf") == 0) {
2205
2206 /* load super */
2207 /* find space for 'n' devices. */
2208 /* remember the devices */
2209 /* Somehow return the fact that we have enough */
2210 }
2211
2212 fprintf(stderr,
2213 Name ": Cannot create this array on device %s\n",
2214 dev);
2215 return 0;
2216 }
2217 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2218 fprintf(stderr, Name ": Cannot open %s: %s\n",
2219 dev, strerror(errno));
2220 return 0;
2221 }
2222 /* Well, it is in use by someone, maybe a 'ddf' container. */
2223 cfd = open_container(fd);
2224 if (cfd < 0) {
2225 close(fd);
2226 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
2227 dev);
2228 return 0;
2229 }
2230 sra = sysfs_read(cfd, 0, GET_VERSION);
2231 close(fd);
2232 if (sra && sra->array.major_version == -1 &&
2233 strcmp(sra->text_version, "ddf") == 0) {
2234 /* This is a member of a ddf container. Load the container
2235 * and try to create a bvd
2236 */
2237 struct ddf_super *ddf;
2238 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL, 1) == 0) {
2239 st->sb = ddf;
2240 st->container_dev = fd2devnum(cfd);
2241 close(cfd);
2242 return validate_geometry_ddf_bvd(st, level, layout,
2243 raiddisks, chunk, size,
2244 dev, freesize);
2245 }
2246 close(cfd);
2247 } else /* device may belong to a different container */
2248 return 0;
2249
2250 return 1;
2251 }
2252
2253 static int validate_geometry_ddf_container(struct supertype *st,
2254 int level, int layout, int raiddisks,
2255 int chunk, unsigned long long size,
2256 char *dev, unsigned long long *freesize)
2257 {
2258 int fd;
2259 unsigned long long ldsize;
2260
2261 if (level != LEVEL_CONTAINER)
2262 return 0;
2263 if (!dev)
2264 return 1;
2265
2266 fd = open(dev, O_RDONLY|O_EXCL, 0);
2267 if (fd < 0) {
2268 fprintf(stderr, Name ": Cannot open %s: %s\n",
2269 dev, strerror(errno));
2270 return 0;
2271 }
2272 if (!get_dev_size(fd, dev, &ldsize)) {
2273 close(fd);
2274 return 0;
2275 }
2276 close(fd);
2277
2278 *freesize = avail_size_ddf(st, ldsize >> 9);
2279
2280 return 1;
2281 }
2282
2283 static int validate_geometry_ddf_bvd(struct supertype *st,
2284 int level, int layout, int raiddisks,
2285 int chunk, unsigned long long size,
2286 char *dev, unsigned long long *freesize)
2287 {
2288 struct stat stb;
2289 struct ddf_super *ddf = st->sb;
2290 struct dl *dl;
2291 unsigned long long pos = 0;
2292 unsigned long long maxsize;
2293 struct extent *e;
2294 int i;
2295 /* ddf/bvd supports lots of things, but not containers */
2296 if (level == LEVEL_CONTAINER)
2297 return 0;
2298 /* We must have the container info already read in. */
2299 if (!ddf)
2300 return 0;
2301
2302 if (!dev) {
2303 /* General test: make sure there is space for
2304 * 'raiddisks' device extents of size 'size'.
2305 */
2306 unsigned long long minsize = size;
2307 int dcnt = 0;
2308 if (minsize == 0)
2309 minsize = 8;
2310 for (dl = ddf->dlist; dl ; dl = dl->next)
2311 {
2312 int found = 0;
2313 pos = 0;
2314
2315 i = 0;
2316 e = get_extents(ddf, dl);
2317 if (!e) continue;
2318 do {
2319 unsigned long long esize;
2320 esize = e[i].start - pos;
2321 if (esize >= minsize)
2322 found = 1;
2323 pos = e[i].start + e[i].size;
2324 i++;
2325 } while (e[i-1].size);
2326 if (found)
2327 dcnt++;
2328 free(e);
2329 }
2330 if (dcnt < raiddisks) {
2331 fprintf(stderr, Name ": Not enough devices with space "
2332 "for this array (%d < %d)\n",
2333 dcnt, raiddisks);
2334 return 0;
2335 }
2336 return 1;
2337 }
2338 /* This device must be a member of the set */
2339 if (stat(dev, &stb) < 0)
2340 return 0;
2341 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2342 return 0;
2343 for (dl = ddf->dlist ; dl ; dl = dl->next) {
2344 if (dl->major == major(stb.st_rdev) &&
2345 dl->minor == minor(stb.st_rdev))
2346 break;
2347 }
2348 if (!dl) {
2349 fprintf(stderr, Name ": %s is not in the same DDF set\n",
2350 dev);
2351 return 0;
2352 }
2353 e = get_extents(ddf, dl);
2354 maxsize = 0;
2355 i = 0;
2356 if (e) do {
2357 unsigned long long esize;
2358 esize = e[i].start - pos;
2359 if (esize >= maxsize)
2360 maxsize = esize;
2361 pos = e[i].start + e[i].size;
2362 i++;
2363 } while (e[i-1].size);
2364 *freesize = maxsize;
2365 // FIXME here I am
2366
2367 return 1;
2368 }
2369
2370 static int load_super_ddf_all(struct supertype *st, int fd,
2371 void **sbp, char *devname, int keep_fd)
2372 {
2373 struct mdinfo *sra;
2374 struct ddf_super *super;
2375 struct mdinfo *sd, *best = NULL;
2376 int bestseq = 0;
2377 int seq;
2378 char nm[20];
2379 int dfd;
2380
2381 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
2382 if (!sra)
2383 return 1;
2384 if (sra->array.major_version != -1 ||
2385 sra->array.minor_version != -2 ||
2386 strcmp(sra->text_version, "ddf") != 0)
2387 return 1;
2388
2389 super = malloc(sizeof(*super));
2390 if (!super)
2391 return 1;
2392 memset(super, 0, sizeof(*super));
2393
2394 /* first, try each device, and choose the best ddf */
2395 for (sd = sra->devs ; sd ; sd = sd->next) {
2396 int rv;
2397 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2398 dfd = dev_open(nm, O_RDONLY);
2399 if (dfd < 0)
2400 return 2;
2401 rv = load_ddf_headers(dfd, super, NULL);
2402 close(dfd);
2403 if (rv == 0) {
2404 seq = __be32_to_cpu(super->active->seq);
2405 if (super->active->openflag)
2406 seq--;
2407 if (!best || seq > bestseq) {
2408 bestseq = seq;
2409 best = sd;
2410 }
2411 }
2412 }
2413 if (!best)
2414 return 1;
2415 /* OK, load this ddf */
2416 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2417 dfd = dev_open(nm, O_RDONLY);
2418 if (dfd < 0)
2419 return 1;
2420 load_ddf_headers(dfd, super, NULL);
2421 load_ddf_global(dfd, super, NULL);
2422 close(dfd);
2423 /* Now we need the device-local bits */
2424 for (sd = sra->devs ; sd ; sd = sd->next) {
2425 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2426 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2427 if (dfd < 0)
2428 return 2;
2429 seq = load_ddf_local(dfd, super, NULL, keep_fd);
2430 if (!keep_fd) close(dfd);
2431 }
2432 if (st->subarray[0]) {
2433 struct vcl *v;
2434
2435 for (v = super->conflist; v; v = v->next)
2436 if (v->vcnum == atoi(st->subarray))
2437 super->currentconf = v;
2438 if (!super->currentconf)
2439 return 1;
2440 }
2441 *sbp = super;
2442 if (st->ss == NULL) {
2443 st->ss = &super_ddf;
2444 st->minor_version = 0;
2445 st->max_devs = 512;
2446 st->container_dev = fd2devnum(fd);
2447 }
2448 return 0;
2449 }
2450 #endif
2451
2452 static struct mdinfo *container_content_ddf(struct supertype *st)
2453 {
2454 /* Given a container loaded by load_super_ddf_all,
2455 * extract information about all the arrays into
2456 * an mdinfo tree.
2457 *
2458 * For each vcl in conflist: create an mdinfo, fill it in,
2459 * then look for matching devices (phys_refnum) in dlist
2460 * and create appropriate device mdinfo.
2461 */
2462 struct ddf_super *ddf = st->sb;
2463 struct mdinfo *rest = NULL;
2464 struct vcl *vc;
2465
2466 for (vc = ddf->conflist ; vc ; vc=vc->next)
2467 {
2468 int i;
2469 struct mdinfo *this;
2470 this = malloc(sizeof(*this));
2471 memset(this, 0, sizeof(*this));
2472 this->next = rest;
2473 rest = this;
2474
2475 this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2476 this->array.raid_disks =
2477 __be16_to_cpu(vc->conf.prim_elmnt_count);
2478 this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2479 this->array.raid_disks);
2480 this->array.md_minor = -1;
2481 this->array.ctime = DECADE +
2482 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
2483 this->array.utime = DECADE +
2484 __be32_to_cpu(vc->conf.timestamp);
2485 this->array.chunk_size = 512 << vc->conf.chunk_shift;
2486
2487 i = vc->vcnum;
2488 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2489 (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
2490 DDF_init_full) {
2491 this->array.state = 0;
2492 this->resync_start = 0;
2493 } else {
2494 this->array.state = 1;
2495 this->resync_start = ~0ULL;
2496 }
2497 memcpy(this->name, ddf->virt->entries[i].name, 32);
2498 this->name[33]=0;
2499
2500 memset(this->uuid, 0, sizeof(this->uuid));
2501 this->component_size = __be64_to_cpu(vc->conf.blocks);
2502 this->array.size = this->component_size / 2;
2503 this->container_member = i;
2504
2505 sprintf(this->text_version, "/%s/%d",
2506 devnum2devname(st->container_dev),
2507 this->container_member);
2508
2509 for (i=0 ; i < ddf->mppe ; i++) {
2510 struct mdinfo *dev;
2511 struct dl *d;
2512
2513 if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
2514 continue;
2515
2516 this->array.working_disks++;
2517
2518 for (d = ddf->dlist; d ; d=d->next)
2519 if (d->disk.refnum == vc->conf.phys_refnum[i])
2520 break;
2521 if (d == NULL)
2522 break;
2523
2524 dev = malloc(sizeof(*dev));
2525 memset(dev, 0, sizeof(*dev));
2526 dev->next = this->devs;
2527 this->devs = dev;
2528
2529 dev->disk.number = __be32_to_cpu(d->disk.refnum);
2530 dev->disk.major = d->major;
2531 dev->disk.minor = d->minor;
2532 dev->disk.raid_disk = i;
2533 dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2534
2535 dev->events = __le32_to_cpu(ddf->primary.seq);
2536 dev->data_offset = vc->lba_offset[i];
2537 dev->component_size = __be64_to_cpu(vc->conf.blocks);
2538 if (d->devname)
2539 strcpy(dev->name, d->devname);
2540 }
2541 }
2542 return rest;
2543 }
2544
2545 static int store_zero_ddf(struct supertype *st, int fd)
2546 {
2547 unsigned long long dsize;
2548 char buf[512];
2549 memset(buf, 0, 512);
2550
2551 if (!get_dev_size(fd, NULL, &dsize))
2552 return 1;
2553
2554 lseek64(fd, dsize-512, 0);
2555 write(fd, buf, 512);
2556 return 0;
2557 }
2558
2559 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
2560 {
2561 /*
2562 * return:
2563 * 0 same, or first was empty, and second was copied
2564 * 1 second had wrong number
2565 * 2 wrong uuid
2566 * 3 wrong other info
2567 */
2568 struct ddf_super *first = st->sb;
2569 struct ddf_super *second = tst->sb;
2570
2571 if (!first) {
2572 st->sb = tst->sb;
2573 tst->sb = NULL;
2574 return 0;
2575 }
2576
2577 if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
2578 return 2;
2579
2580 /* FIXME should I look at anything else? */
2581 return 0;
2582 }
2583
2584 /*
2585 * A new array 'a' has been started which claims to be instance 'inst'
2586 * within container 'c'.
2587 * We need to confirm that the array matches the metadata in 'c' so
2588 * that we don't corrupt any metadata.
2589 */
2590 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
2591 {
2592 fprintf(stderr, "ddf: open_new %s\n", inst);
2593 a->info.container_member = atoi(inst);
2594 return 0;
2595 }
2596
2597 /*
2598 * The array 'a' is to be marked clean in the metadata.
2599 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2600 * clean up to the point (in sectors). If that cannot be recorded in the
2601 * metadata, then leave it as dirty.
2602 *
2603 * For DDF, we need to clear the DDF_state_inconsistent bit in the
2604 * !global! virtual_disk.virtual_entry structure.
2605 */
2606 static void ddf_set_array_state(struct active_array *a, int consistent)
2607 {
2608 struct ddf_super *ddf = a->container->sb;
2609 int inst = a->info.container_member;
2610 if (consistent)
2611 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
2612 else
2613 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
2614 ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
2615 if (a->resync_start == ~0ULL)
2616 ddf->virt->entries[inst].init_state |= DDF_init_full;
2617 else if (a->resync_start == 0)
2618 ddf->virt->entries[inst].init_state |= DDF_init_not;
2619 else
2620 ddf->virt->entries[inst].init_state |= DDF_init_quick;
2621
2622 printf("ddf mark %s %llu\n", consistent?"clean":"dirty",
2623 a->resync_start);
2624 }
2625
2626 /*
2627 * The state of each disk is stored in the global phys_disk structure
2628 * in phys_disk.entries[n].state.
2629 * This makes various combinations awkward.
2630 * - When a device fails in any array, it must be failed in all arrays
2631 * that include a part of this device.
2632 * - When a component is rebuilding, we cannot include it officially in the
2633 * array unless this is the only array that uses the device.
2634 *
2635 * So: when transitioning:
2636 * Online -> failed, just set failed flag. monitor will propagate
2637 * spare -> online, the device might need to be added to the array.
2638 * spare -> failed, just set failed. Don't worry if in array or not.
2639 */
2640 static void ddf_set_disk(struct active_array *a, int n, int state)
2641 {
2642 struct ddf_super *ddf = a->container->sb;
2643 int inst = a->info.container_member;
2644 struct vd_config *vc = find_vdcr(ddf, inst);
2645 int pd = find_phys(ddf, vc->phys_refnum[n]);
2646 int i, st, working;
2647
2648 if (vc == NULL) {
2649 fprintf(stderr, "ddf: cannot find instance %d!!\n", inst);
2650 return;
2651 }
2652 if (pd < 0) {
2653 /* disk doesn't currently exist. If it is now in_sync,
2654 * insert it. */
2655 if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
2656 /* Find dev 'n' in a->info->devs, determine the
2657 * ddf refnum, and set vc->phys_refnum and update
2658 * phys->entries[]
2659 */
2660 /* FIXME */
2661 }
2662 } else {
2663 if (state & DS_FAULTY)
2664 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Failed);
2665 if (state & DS_INSYNC) {
2666 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Online);
2667 ddf->phys->entries[pd].state &= __cpu_to_be16(~DDF_Rebuilding);
2668 }
2669 }
2670
2671 fprintf(stderr, "ddf: set_disk %d to %x\n", n, state);
2672
2673 /* Now we need to check the state of the array and update
2674 * virtual_disk.entries[n].state.
2675 * It needs to be one of "optimal", "degraded", "failed".
2676 * I don't understand 'deleted' or 'missing'.
2677 */
2678 working = 0;
2679 for (i=0; i < a->info.array.raid_disks; i++) {
2680 pd = find_phys(ddf, vc->phys_refnum[i]);
2681 if (pd < 0)
2682 continue;
2683 st = __be16_to_cpu(ddf->phys->entries[pd].state);
2684 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
2685 == DDF_Online)
2686 working++;
2687 }
2688 state = DDF_state_degraded;
2689 if (working == a->info.array.raid_disks)
2690 state = DDF_state_optimal;
2691 else switch(vc->prl) {
2692 case DDF_RAID0:
2693 case DDF_CONCAT:
2694 case DDF_JBOD:
2695 state = DDF_state_failed;
2696 break;
2697 case DDF_RAID1:
2698 if (working == 0)
2699 state = DDF_state_failed;
2700 break;
2701 case DDF_RAID4:
2702 case DDF_RAID5:
2703 if (working < a->info.array.raid_disks-1)
2704 state = DDF_state_failed;
2705 break;
2706 case DDF_RAID6:
2707 if (working < a->info.array.raid_disks-2)
2708 state = DDF_state_failed;
2709 else if (working == a->info.array.raid_disks-1)
2710 state = DDF_state_part_optimal;
2711 break;
2712 }
2713
2714 ddf->virt->entries[inst].state =
2715 (ddf->virt->entries[inst].state & ~DDF_state_mask)
2716 | state;
2717
2718 }
2719
2720 static void ddf_sync_metadata(struct supertype *st)
2721 {
2722
2723 /*
2724 * Write all data to all devices.
2725 * Later, we might be able to track whether only local changes
2726 * have been made, or whether any global data has been changed,
2727 * but ddf is sufficiently weird that it probably always
2728 * changes global data ....
2729 */
2730 __write_init_super_ddf(st, 0);
2731 fprintf(stderr, "ddf: sync_metadata\n");
2732 }
2733
2734 static void ddf_process_update(struct supertype *st,
2735 struct metadata_update *update)
2736 {
2737 /* Apply this update to the metadata.
2738 * The first 4 bytes are a DDF_*_MAGIC which guides
2739 * our actions.
2740 * Possible update are:
2741 * DDF_PHYS_RECORDS_MAGIC
2742 * Add a new physical device. Changes to this record
2743 * only happen implicitly.
2744 * used_pdes is the device number.
2745 * DDF_VIRT_RECORDS_MAGIC
2746 * Add a new VD. Possibly also change the 'access' bits.
2747 * populated_vdes is the entry number.
2748 * DDF_VD_CONF_MAGIC
2749 * New or updated VD. the VIRT_RECORD must already
2750 * exist. For an update, phys_refnum and lba_offset
2751 * (at least) are updated, and the VD_CONF must
2752 * be written to precisely those devices listed with
2753 * a phys_refnum.
2754 * DDF_SPARE_ASSIGN_MAGIC
2755 * replacement Spare Assignment Record... but for which device?
2756 *
2757 * So, e.g.:
2758 * - to create a new array, we send a VIRT_RECORD and
2759 * a VD_CONF. Then assemble and start the array.
2760 * - to activate a spare we send a VD_CONF to add the phys_refnum
2761 * and offset. This will also mark the spare as active with
2762 * a spare-assignment record.
2763 */
2764 struct ddf_super *ddf = st->sb;
2765 __u32 *magic = (__u32*)update->buf;
2766 struct phys_disk *pd;
2767 struct virtual_disk *vd;
2768 struct vd_config *vc;
2769 struct vcl *vcl;
2770 struct dl *dl;
2771 int mppe;
2772 int ent;
2773
2774 printf("Process update %x\n", *magic);
2775
2776 switch (*magic) {
2777 case DDF_PHYS_RECORDS_MAGIC:
2778
2779 if (update->len != (sizeof(struct phys_disk) +
2780 sizeof(struct phys_disk_entry)))
2781 return;
2782 pd = (struct phys_disk*)update->buf;
2783
2784 ent = __be16_to_cpu(pd->used_pdes);
2785 if (ent >= __be16_to_cpu(ddf->phys->max_pdes))
2786 return;
2787 if (!all_ff(ddf->phys->entries[ent].guid))
2788 return;
2789 ddf->phys->entries[ent] = pd->entries[0];
2790 ddf->phys->used_pdes = __cpu_to_be16(1 +
2791 __be16_to_cpu(ddf->phys->used_pdes));
2792 break;
2793
2794 case DDF_VIRT_RECORDS_MAGIC:
2795
2796 if (update->len != (sizeof(struct virtual_disk) +
2797 sizeof(struct virtual_entry)))
2798 return;
2799 vd = (struct virtual_disk*)update->buf;
2800
2801 ent = __be16_to_cpu(vd->populated_vdes);
2802 if (ent >= __be16_to_cpu(ddf->virt->max_vdes))
2803 return;
2804 if (!all_ff(ddf->virt->entries[ent].guid))
2805 return;
2806 ddf->virt->entries[ent] = vd->entries[0];
2807 ddf->virt->populated_vdes = __cpu_to_be16(1 +
2808 __be16_to_cpu(ddf->virt->populated_vdes));
2809 break;
2810
2811 case DDF_VD_CONF_MAGIC:
2812 printf("len %d %d\n", update->len, ddf->conf_rec_len);
2813
2814 mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries);
2815 if (update->len != ddf->conf_rec_len)
2816 return;
2817 vc = (struct vd_config*)update->buf;
2818 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2819 if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
2820 break;
2821 printf("vcl = %p\n", vcl);
2822 if (vcl) {
2823 /* An update, just copy the phys_refnum and lba_offset
2824 * fields
2825 */
2826 memcpy(vcl->conf.phys_refnum, vc->phys_refnum,
2827 mppe * (sizeof(__u32) + sizeof(__u64)));
2828 } else {
2829 /* A new VD_CONF */
2830 vcl = update->space;
2831 update->space = NULL;
2832 vcl->next = ddf->conflist;
2833 vcl->conf = *vc;
2834 vcl->lba_offset = (__u64*)
2835 &vcl->conf.phys_refnum[mppe];
2836 ddf->conflist = vcl;
2837 }
2838 /* Now make sure vlist is correct for each dl. */
2839 for (dl = ddf->dlist; dl; dl = dl->next) {
2840 int dn;
2841 int vn = 0;
2842 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2843 for (dn=0; dn < ddf->mppe ; dn++)
2844 if (vcl->conf.phys_refnum[dn] ==
2845 dl->disk.refnum) {
2846 printf("dev %d has %p at %d\n",
2847 dl->pdnum, vcl, vn);
2848 dl->vlist[vn++] = vcl;
2849 break;
2850 }
2851 while (vn < ddf->max_part)
2852 dl->vlist[vn++] = NULL;
2853 if (dl->vlist[0]) {
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_Active_in_VD);
2858 }
2859 if (dl->spare) {
2860 ddf->phys->entries[dl->pdnum].type &=
2861 ~__cpu_to_be16(DDF_Global_Spare);
2862 ddf->phys->entries[dl->pdnum].type |=
2863 __cpu_to_be16(DDF_Spare);
2864 }
2865 if (!dl->vlist[0] && !dl->spare) {
2866 ddf->phys->entries[dl->pdnum].type |=
2867 __cpu_to_be16(DDF_Global_Spare);
2868 ddf->phys->entries[dl->pdnum].type &=
2869 ~__cpu_to_be16(DDF_Spare |
2870 DDF_Active_in_VD);
2871 }
2872 }
2873 break;
2874 case DDF_SPARE_ASSIGN_MAGIC:
2875 default: break;
2876 }
2877 }
2878
2879 /*
2880 * Check if the array 'a' is degraded but not failed.
2881 * If it is, find as many spares as are available and needed and
2882 * arrange for their inclusion.
2883 * We only choose devices which are not already in the array,
2884 * and prefer those with a spare-assignment to this array.
2885 * otherwise we choose global spares - assuming always that
2886 * there is enough room.
2887 * For each spare that we assign, we return an 'mdinfo' which
2888 * describes the position for the device in the array.
2889 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
2890 * the new phys_refnum and lba_offset values.
2891 *
2892 * Only worry about BVDs at the moment.
2893 */
2894 static struct mdinfo *ddf_activate_spare(struct active_array *a,
2895 struct metadata_update **updates)
2896 {
2897 int working = 0;
2898 struct mdinfo *d;
2899 struct ddf_super *ddf = a->container->sb;
2900 int global_ok = 0;
2901 struct mdinfo *rv = NULL;
2902 struct mdinfo *di;
2903 struct metadata_update *mu;
2904 struct dl *dl;
2905 int i;
2906 struct vd_config *vc;
2907 __u64 *lba;
2908
2909 for (d = a->info.devs ; d ; d = d->next) {
2910 if ((d->curr_state & DS_FAULTY) &&
2911 d->state_fd >= 0)
2912 /* wait for Removal to happen */
2913 return NULL;
2914 if (d->state_fd >= 0)
2915 working ++;
2916 }
2917
2918 printf("ddf_activate: working=%d (%d) level=%d\n", working, a->info.array.raid_disks,
2919 a->info.array.level);
2920 if (working == a->info.array.raid_disks)
2921 return NULL; /* array not degraded */
2922 switch (a->info.array.level) {
2923 case 1:
2924 if (working == 0)
2925 return NULL; /* failed */
2926 break;
2927 case 4:
2928 case 5:
2929 if (working < a->info.array.raid_disks - 1)
2930 return NULL; /* failed */
2931 break;
2932 case 6:
2933 if (working < a->info.array.raid_disks - 2)
2934 return NULL; /* failed */
2935 break;
2936 default: /* concat or stripe */
2937 return NULL; /* failed */
2938 }
2939
2940 /* For each slot, if it is not working, find a spare */
2941 dl = ddf->dlist;
2942 for (i = 0; i < a->info.array.raid_disks; i++) {
2943 for (d = a->info.devs ; d ; d = d->next)
2944 if (d->disk.raid_disk == i)
2945 break;
2946 printf("found %d: %p %x\n", i, d, d?d->curr_state:0);
2947 if (d && (d->state_fd >= 0))
2948 continue;
2949
2950 /* OK, this device needs recovery. Find a spare */
2951 again:
2952 for ( ; dl ; dl = dl->next) {
2953 unsigned long long esize;
2954 unsigned long long pos;
2955 struct mdinfo *d2;
2956 int is_global = 0;
2957 int is_dedicated = 0;
2958 struct extent *ex;
2959 int j;
2960 /* If in this array, skip */
2961 for (d2 = a->info.devs ; d2 ; d2 = d2->next)
2962 if (d2->disk.major == dl->major &&
2963 d2->disk.minor == dl->minor) {
2964 printf("%x:%x already in array\n", dl->major, dl->minor);
2965 break;
2966 }
2967 if (d2)
2968 continue;
2969 if (ddf->phys->entries[dl->pdnum].type &
2970 __cpu_to_be16(DDF_Spare)) {
2971 /* Check spare assign record */
2972 if (dl->spare) {
2973 if (dl->spare->type & DDF_spare_dedicated) {
2974 /* check spare_ents for guid */
2975 for (j = 0 ;
2976 j < __be16_to_cpu(dl->spare->populated);
2977 j++) {
2978 if (memcmp(dl->spare->spare_ents[j].guid,
2979 ddf->virt->entries[a->info.container_member].guid,
2980 DDF_GUID_LEN) == 0)
2981 is_dedicated = 1;
2982 }
2983 } else
2984 is_global = 1;
2985 }
2986 } else if (ddf->phys->entries[dl->pdnum].type &
2987 __cpu_to_be16(DDF_Global_Spare)) {
2988 is_global = 1;
2989 }
2990 if ( ! (is_dedicated ||
2991 (is_global && global_ok))) {
2992 printf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
2993 is_dedicated, is_global);
2994 continue;
2995 }
2996
2997 /* We are allowed to use this device - is there space?
2998 * We need a->info.component_size sectors */
2999 ex = get_extents(ddf, dl);
3000 if (!ex) {
3001 printf("cannot get extents\n");
3002 continue;
3003 }
3004 j = 0; pos = 0;
3005 esize = 0;
3006
3007 do {
3008 esize = ex[j].start - pos;
3009 if (esize >= a->info.component_size)
3010 break;
3011 pos = ex[i].start + ex[i].size;
3012 i++;
3013 } while (ex[i-1].size);
3014
3015 free(ex);
3016 if (esize < a->info.component_size) {
3017 printf("%x:%x has no room: %llu %llu\n", dl->major, dl->minor,
3018 esize, a->info.component_size);
3019 /* No room */
3020 continue;
3021 }
3022
3023 /* Cool, we have a device with some space at pos */
3024 di = malloc(sizeof(*di));
3025 memset(di, 0, sizeof(*di));
3026 di->disk.number = i;
3027 di->disk.raid_disk = i;
3028 di->disk.major = dl->major;
3029 di->disk.minor = dl->minor;
3030 di->disk.state = 0;
3031 di->data_offset = pos;
3032 di->component_size = a->info.component_size;
3033 di->container_member = dl->pdnum;
3034 di->next = rv;
3035 rv = di;
3036 printf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3037 i, pos);
3038
3039 break;
3040 }
3041 if (!dl && ! global_ok) {
3042 /* not enough dedicated spares, try global */
3043 global_ok = 1;
3044 dl = ddf->dlist;
3045 goto again;
3046 }
3047 }
3048
3049 if (!rv)
3050 /* No spares found */
3051 return rv;
3052 /* Now 'rv' has a list of devices to return.
3053 * Create a metadata_update record to update the
3054 * phys_refnum and lba_offset values
3055 */
3056 mu = malloc(sizeof(*mu));
3057 mu->buf = malloc(ddf->conf_rec_len * 512);
3058 mu->space = malloc(sizeof(struct vcl));
3059 mu->len = ddf->conf_rec_len;
3060 mu->next = *updates;
3061 vc = find_vdcr(ddf, a->info.container_member);
3062 memcpy(mu->buf, vc, ddf->conf_rec_len * 512);
3063
3064 vc = (struct vd_config*)mu->buf;
3065 lba = (__u64*)&vc->phys_refnum[ddf->mppe];
3066 for (di = rv ; di ; di = di->next) {
3067 vc->phys_refnum[di->disk.raid_disk] =
3068 ddf->phys->entries[dl->pdnum].refnum;
3069 lba[di->disk.raid_disk] = di->data_offset;
3070 }
3071 *updates = mu;
3072 return rv;
3073 }
3074
3075 struct superswitch super_ddf = {
3076 #ifndef MDASSEMBLE
3077 .examine_super = examine_super_ddf,
3078 .brief_examine_super = brief_examine_super_ddf,
3079 .detail_super = detail_super_ddf,
3080 .brief_detail_super = brief_detail_super_ddf,
3081 .validate_geometry = validate_geometry_ddf,
3082 .write_init_super = write_init_super_ddf,
3083 #endif
3084 .match_home = match_home_ddf,
3085 .uuid_from_super= uuid_from_super_ddf,
3086 .getinfo_super = getinfo_super_ddf,
3087 .update_super = update_super_ddf,
3088
3089 .avail_size = avail_size_ddf,
3090
3091 .compare_super = compare_super_ddf,
3092
3093 .load_super = load_super_ddf,
3094 .init_super = init_super_ddf,
3095 .store_super = store_zero_ddf,
3096 .free_super = free_super_ddf,
3097 .match_metadata_desc = match_metadata_desc_ddf,
3098 .add_to_super = add_to_super_ddf,
3099 .container_content = container_content_ddf,
3100
3101 .external = 1,
3102
3103 /* for mdmon */
3104 .open_new = ddf_open_new,
3105 .set_array_state= ddf_set_array_state,
3106 .set_disk = ddf_set_disk,
3107 .sync_metadata = ddf_sync_metadata,
3108 .process_update = ddf_process_update,
3109 .activate_spare = ddf_activate_spare,
3110
3111 };
Unnamed repository; edit this file 'description' to name the repository.