]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super-ddf.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ddf: Set container_member from subarray in getinfo_super.
[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 int updates_pending;
406 struct vcl {
407 struct vcl *next;
408 __u64 *lba_offset; /* location in 'conf' of
409 * the lba table */
410 int vcnum; /* index into ->virt */
411 __u64 *block_sizes; /* NULL if all the same */
412 struct vd_config conf;
413 } *conflist, *currentconf;
414 struct dl {
415 struct dl *next;
416 struct disk_data disk;
417 int major, minor;
418 char *devname;
419 int fd;
420 unsigned long long size; /* sectors */
421 int pdnum; /* index in ->phys */
422 struct spare_assign *spare;
423 struct vcl *vlist[0]; /* max_part in size */
424 } *dlist;
425 };
426
427 #ifndef offsetof
428 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
429 #endif
430
431
432 static int calc_crc(void *buf, int len)
433 {
434 /* crcs are always at the same place as in the ddf_header */
435 struct ddf_header *ddf = buf;
436 __u32 oldcrc = ddf->crc;
437 __u32 newcrc;
438 ddf->crc = 0xffffffff;
439
440 newcrc = crc32(0, buf, len);
441 ddf->crc = oldcrc;
442 return newcrc;
443 }
444
445 static int load_ddf_header(int fd, unsigned long long lba,
446 unsigned long long size,
447 int type,
448 struct ddf_header *hdr, struct ddf_header *anchor)
449 {
450 /* read a ddf header (primary or secondary) from fd/lba
451 * and check that it is consistent with anchor
452 * Need to check:
453 * magic, crc, guid, rev, and LBA's header_type, and
454 * everything after header_type must be the same
455 */
456 if (lba >= size-1)
457 return 0;
458
459 if (lseek64(fd, lba<<9, 0) < 0)
460 return 0;
461
462 if (read(fd, hdr, 512) != 512)
463 return 0;
464
465 if (hdr->magic != DDF_HEADER_MAGIC)
466 return 0;
467 if (calc_crc(hdr, 512) != hdr->crc)
468 return 0;
469 if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
470 memcmp(anchor->revision, hdr->revision, 8) != 0 ||
471 anchor->primary_lba != hdr->primary_lba ||
472 anchor->secondary_lba != hdr->secondary_lba ||
473 hdr->type != type ||
474 memcmp(anchor->pad2, hdr->pad2, 512 -
475 offsetof(struct ddf_header, pad2)) != 0)
476 return 0;
477
478 /* Looks good enough to me... */
479 return 1;
480 }
481
482 static void *load_section(int fd, struct ddf_super *super, void *buf,
483 __u32 offset_be, __u32 len_be, int check)
484 {
485 unsigned long long offset = __be32_to_cpu(offset_be);
486 unsigned long long len = __be32_to_cpu(len_be);
487 int dofree = (buf == NULL);
488
489 if (check)
490 if (len != 2 && len != 8 && len != 32
491 && len != 128 && len != 512)
492 return NULL;
493
494 if (len > 1024)
495 return NULL;
496 if (buf) {
497 /* All pre-allocated sections are a single block */
498 if (len != 1)
499 return NULL;
500 } else
501 buf = malloc(len<<9);
502 if (!buf)
503 return NULL;
504
505 if (super->active->type == 1)
506 offset += __be64_to_cpu(super->active->primary_lba);
507 else
508 offset += __be64_to_cpu(super->active->secondary_lba);
509
510 if (lseek64(fd, offset<<9, 0) != (offset<<9)) {
511 if (dofree)
512 free(buf);
513 return NULL;
514 }
515 if (read(fd, buf, len<<9) != (len<<9)) {
516 if (dofree)
517 free(buf);
518 return NULL;
519 }
520 return buf;
521 }
522
523 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
524 {
525 unsigned long long dsize;
526
527 get_dev_size(fd, NULL, &dsize);
528
529 if (lseek64(fd, dsize-512, 0) < 0) {
530 if (devname)
531 fprintf(stderr,
532 Name": Cannot seek to anchor block on %s: %s\n",
533 devname, strerror(errno));
534 return 1;
535 }
536 if (read(fd, &super->anchor, 512) != 512) {
537 if (devname)
538 fprintf(stderr,
539 Name ": Cannot read anchor block on %s: %s\n",
540 devname, strerror(errno));
541 return 1;
542 }
543 if (super->anchor.magic != DDF_HEADER_MAGIC) {
544 if (devname)
545 fprintf(stderr, Name ": no DDF anchor found on %s\n",
546 devname);
547 return 2;
548 }
549 if (calc_crc(&super->anchor, 512) != super->anchor.crc) {
550 if (devname)
551 fprintf(stderr, Name ": bad CRC on anchor on %s\n",
552 devname);
553 return 2;
554 }
555 if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
556 memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
557 if (devname)
558 fprintf(stderr, Name ": can only support super revision"
559 " %.8s and earlier, not %.8s on %s\n",
560 DDF_REVISION_2, super->anchor.revision,devname);
561 return 2;
562 }
563 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba),
564 dsize >> 9, 1,
565 &super->primary, &super->anchor) == 0) {
566 if (devname)
567 fprintf(stderr,
568 Name ": Failed to load primary DDF header "
569 "on %s\n", devname);
570 return 2;
571 }
572 super->active = &super->primary;
573 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.secondary_lba),
574 dsize >> 9, 2,
575 &super->secondary, &super->anchor)) {
576 if ((__be32_to_cpu(super->primary.seq)
577 < __be32_to_cpu(super->secondary.seq) &&
578 !super->secondary.openflag)
579 || (__be32_to_cpu(super->primary.seq)
580 == __be32_to_cpu(super->secondary.seq) &&
581 super->primary.openflag && !super->secondary.openflag)
582 )
583 super->active = &super->secondary;
584 }
585 return 0;
586 }
587
588 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
589 {
590 void *ok;
591 ok = load_section(fd, super, &super->controller,
592 super->active->controller_section_offset,
593 super->active->controller_section_length,
594 0);
595 super->phys = load_section(fd, super, NULL,
596 super->active->phys_section_offset,
597 super->active->phys_section_length,
598 1);
599 super->pdsize = __be32_to_cpu(super->active->phys_section_length) * 512;
600
601 super->virt = load_section(fd, super, NULL,
602 super->active->virt_section_offset,
603 super->active->virt_section_length,
604 1);
605 super->vdsize = __be32_to_cpu(super->active->virt_section_length) * 512;
606 if (!ok ||
607 !super->phys ||
608 !super->virt) {
609 free(super->phys);
610 free(super->virt);
611 super->phys = NULL;
612 super->virt = NULL;
613 return 2;
614 }
615 super->conflist = NULL;
616 super->dlist = NULL;
617
618 super->max_part = __be16_to_cpu(super->active->max_partitions);
619 super->mppe = __be16_to_cpu(super->active->max_primary_element_entries);
620 super->conf_rec_len = __be16_to_cpu(super->active->config_record_len);
621 return 0;
622 }
623
624 static int load_ddf_local(int fd, struct ddf_super *super,
625 char *devname, int keep)
626 {
627 struct dl *dl;
628 struct stat stb;
629 char *conf;
630 int i;
631 int vnum;
632 int max_virt_disks = __be16_to_cpu(super->active->max_vd_entries);
633 unsigned long long dsize;
634
635 /* First the local disk info */
636 dl = malloc(sizeof(*dl) +
637 (super->max_part) * sizeof(dl->vlist[0]));
638
639 load_section(fd, super, &dl->disk,
640 super->active->data_section_offset,
641 super->active->data_section_length,
642 0);
643 dl->devname = devname ? strdup(devname) : NULL;
644
645 fstat(fd, &stb);
646 dl->major = major(stb.st_rdev);
647 dl->minor = minor(stb.st_rdev);
648 dl->next = super->dlist;
649 dl->fd = keep ? fd : -1;
650
651 dl->size = 0;
652 if (get_dev_size(fd, devname, &dsize))
653 dl->size = dsize >> 9;
654 dl->spare = NULL;
655 for (i=0 ; i < super->max_part ; i++)
656 dl->vlist[i] = NULL;
657 super->dlist = dl;
658 dl->pdnum = -1;
659 for (i=0; i < __be16_to_cpu(super->active->max_pd_entries); i++)
660 if (memcmp(super->phys->entries[i].guid,
661 dl->disk.guid, DDF_GUID_LEN) == 0)
662 dl->pdnum = i;
663
664 /* Now the config list. */
665 /* 'conf' is an array of config entries, some of which are
666 * probably invalid. Those which are good need to be copied into
667 * the conflist
668 */
669
670 conf = load_section(fd, super, NULL,
671 super->active->config_section_offset,
672 super->active->config_section_length,
673 0);
674
675 vnum = 0;
676 for (i = 0;
677 i < __be32_to_cpu(super->active->config_section_length);
678 i += super->conf_rec_len) {
679 struct vd_config *vd =
680 (struct vd_config *)((char*)conf + i*512);
681 struct vcl *vcl;
682
683 if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) {
684 if (dl->spare)
685 continue;
686 dl->spare = malloc(super->conf_rec_len*512);
687 memcpy(dl->spare, vd, super->conf_rec_len*512);
688 continue;
689 }
690 if (vd->magic != DDF_VD_CONF_MAGIC)
691 continue;
692 for (vcl = super->conflist; vcl; vcl = vcl->next) {
693 if (memcmp(vcl->conf.guid,
694 vd->guid, DDF_GUID_LEN) == 0)
695 break;
696 }
697
698 if (vcl) {
699 dl->vlist[vnum++] = vcl;
700 if (__be32_to_cpu(vd->seqnum) <=
701 __be32_to_cpu(vcl->conf.seqnum))
702 continue;
703 } else {
704 vcl = malloc(super->conf_rec_len*512 +
705 offsetof(struct vcl, conf));
706 vcl->next = super->conflist;
707 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
708 super->conflist = vcl;
709 dl->vlist[vnum++] = vcl;
710 }
711 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
712 vcl->lba_offset = (__u64*)
713 &vcl->conf.phys_refnum[super->mppe];
714
715 for (i=0; i < max_virt_disks ; i++)
716 if (memcmp(super->virt->entries[i].guid,
717 vcl->conf.guid, DDF_GUID_LEN)==0)
718 break;
719 if (i < max_virt_disks)
720 vcl->vcnum = i;
721 }
722 free(conf);
723
724 return 0;
725 }
726
727 #ifndef MDASSEMBLE
728 static int load_super_ddf_all(struct supertype *st, int fd,
729 void **sbp, char *devname, int keep_fd);
730 #endif
731 static int load_super_ddf(struct supertype *st, int fd,
732 char *devname)
733 {
734 unsigned long long dsize;
735 struct ddf_super *super;
736 int rv;
737
738 #ifndef MDASSEMBLE
739 /* if 'fd' is a container, load metadata from all the devices */
740 if (load_super_ddf_all(st, fd, &st->sb, devname, 1) == 0)
741 return 0;
742 #endif
743 if (st->subarray[0])
744 return 1; /* FIXME Is this correct */
745
746 if (get_dev_size(fd, devname, &dsize) == 0)
747 return 1;
748
749 /* 32M is a lower bound */
750 if (dsize <= 32*1024*1024) {
751 if (devname) {
752 fprintf(stderr,
753 Name ": %s is too small for ddf: "
754 "size is %llu sectors.\n",
755 devname, dsize>>9);
756 return 1;
757 }
758 }
759 if (dsize & 511) {
760 if (devname) {
761 fprintf(stderr,
762 Name ": %s is an odd size for ddf: "
763 "size is %llu bytes.\n",
764 devname, dsize);
765 return 1;
766 }
767 }
768
769 super = malloc(sizeof(*super));
770 if (!super) {
771 fprintf(stderr, Name ": malloc of %zu failed.\n",
772 sizeof(*super));
773 return 1;
774 }
775 memset(super, 0, sizeof(*super));
776
777 rv = load_ddf_headers(fd, super, devname);
778 if (rv) {
779 free(super);
780 return rv;
781 }
782
783 /* Have valid headers and have chosen the best. Let's read in the rest*/
784
785 rv = load_ddf_global(fd, super, devname);
786
787 if (rv) {
788 if (devname)
789 fprintf(stderr,
790 Name ": Failed to load all information "
791 "sections on %s\n", devname);
792 free(super);
793 return rv;
794 }
795
796 load_ddf_local(fd, super, devname, 0);
797
798 /* Should possibly check the sections .... */
799
800 st->sb = super;
801 if (st->ss == NULL) {
802 st->ss = &super_ddf;
803 st->minor_version = 0;
804 st->max_devs = 512;
805 }
806 return 0;
807
808 }
809
810 static void free_super_ddf(struct supertype *st)
811 {
812 struct ddf_super *ddf = st->sb;
813 if (ddf == NULL)
814 return;
815 free(ddf->phys);
816 free(ddf->virt);
817 while (ddf->conflist) {
818 struct vcl *v = ddf->conflist;
819 ddf->conflist = v->next;
820 if (v->block_sizes)
821 free(v->block_sizes);
822 free(v);
823 }
824 while (ddf->dlist) {
825 struct dl *d = ddf->dlist;
826 ddf->dlist = d->next;
827 if (d->fd >= 0)
828 close(d->fd);
829 if (d->spare)
830 free(d->spare);
831 free(d);
832 }
833 free(ddf);
834 st->sb = NULL;
835 }
836
837 static struct supertype *match_metadata_desc_ddf(char *arg)
838 {
839 /* 'ddf' only support containers */
840 struct supertype *st;
841 if (strcmp(arg, "ddf") != 0 &&
842 strcmp(arg, "default") != 0
843 )
844 return NULL;
845
846 st = malloc(sizeof(*st));
847 memset(st, 0, sizeof(*st));
848 st->ss = &super_ddf;
849 st->max_devs = 512;
850 st->minor_version = 0;
851 st->sb = NULL;
852 return st;
853 }
854
855
856 #ifndef MDASSEMBLE
857
858 static mapping_t ddf_state[] = {
859 { "Optimal", 0},
860 { "Degraded", 1},
861 { "Deleted", 2},
862 { "Missing", 3},
863 { "Failed", 4},
864 { "Partially Optimal", 5},
865 { "-reserved-", 6},
866 { "-reserved-", 7},
867 { NULL, 0}
868 };
869
870 static mapping_t ddf_init_state[] = {
871 { "Not Initialised", 0},
872 { "QuickInit in Progress", 1},
873 { "Fully Initialised", 2},
874 { "*UNKNOWN*", 3},
875 { NULL, 0}
876 };
877 static mapping_t ddf_access[] = {
878 { "Read/Write", 0},
879 { "Reserved", 1},
880 { "Read Only", 2},
881 { "Blocked (no access)", 3},
882 { NULL ,0}
883 };
884
885 static mapping_t ddf_level[] = {
886 { "RAID0", DDF_RAID0},
887 { "RAID1", DDF_RAID1},
888 { "RAID3", DDF_RAID3},
889 { "RAID4", DDF_RAID4},
890 { "RAID5", DDF_RAID5},
891 { "RAID1E",DDF_RAID1E},
892 { "JBOD", DDF_JBOD},
893 { "CONCAT",DDF_CONCAT},
894 { "RAID5E",DDF_RAID5E},
895 { "RAID5EE",DDF_RAID5EE},
896 { "RAID6", DDF_RAID6},
897 { NULL, 0}
898 };
899 static mapping_t ddf_sec_level[] = {
900 { "Striped", DDF_2STRIPED},
901 { "Mirrored", DDF_2MIRRORED},
902 { "Concat", DDF_2CONCAT},
903 { "Spanned", DDF_2SPANNED},
904 { NULL, 0}
905 };
906 #endif
907
908 struct num_mapping {
909 int num1, num2;
910 };
911 static struct num_mapping ddf_level_num[] = {
912 { DDF_RAID0, 0 },
913 { DDF_RAID1, 1 },
914 { DDF_RAID3, LEVEL_UNSUPPORTED },
915 { DDF_RAID4, 4 },
916 { DDF_RAID5, 5 },
917 { DDF_RAID1E, LEVEL_UNSUPPORTED },
918 { DDF_JBOD, LEVEL_UNSUPPORTED },
919 { DDF_CONCAT, LEVEL_LINEAR },
920 { DDF_RAID5E, LEVEL_UNSUPPORTED },
921 { DDF_RAID5EE, LEVEL_UNSUPPORTED },
922 { DDF_RAID6, 6},
923 { MAXINT, MAXINT }
924 };
925
926 static int map_num1(struct num_mapping *map, int num)
927 {
928 int i;
929 for (i=0 ; map[i].num1 != MAXINT; i++)
930 if (map[i].num1 == num)
931 break;
932 return map[i].num2;
933 }
934
935 #ifndef MDASSEMBLE
936 static void print_guid(char *guid, int tstamp)
937 {
938 /* A GUIDs are part (or all) ASCII and part binary.
939 * They tend to be space padded.
940 * We print the GUID in HEX, then in parentheses add
941 * any initial ASCII sequence, and a possible
942 * time stamp from bytes 16-19
943 */
944 int l = DDF_GUID_LEN;
945 int i;
946
947 for (i=0 ; i<DDF_GUID_LEN ; i++) {
948 if ((i&3)==0 && i != 0) printf(":");
949 printf("%02X", guid[i]&255);
950 }
951
952 printf(" (");
953 while (l && guid[l-1] == ' ')
954 l--;
955 for (i=0 ; i<l ; i++) {
956 if (guid[i] >= 0x20 && guid[i] < 0x7f)
957 fputc(guid[i], stdout);
958 else
959 break;
960 }
961 if (tstamp) {
962 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
963 char tbuf[100];
964 struct tm *tm;
965 tm = localtime(&then);
966 strftime(tbuf, 100, " %D %T",tm);
967 fputs(tbuf, stdout);
968 }
969 printf(")");
970 }
971
972 static void examine_vd(int n, struct ddf_super *sb, char *guid)
973 {
974 int crl = sb->conf_rec_len;
975 struct vcl *vcl;
976
977 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
978 struct vd_config *vc = &vcl->conf;
979
980 if (calc_crc(vc, crl*512) != vc->crc)
981 continue;
982 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
983 continue;
984
985 /* Ok, we know about this VD, let's give more details */
986 printf(" Raid Devices[%d] : %d\n", n,
987 __be16_to_cpu(vc->prim_elmnt_count));
988 printf(" Chunk Size[%d] : %d sectors\n", n,
989 1 << vc->chunk_shift);
990 printf(" Raid Level[%d] : %s\n", n,
991 map_num(ddf_level, vc->prl)?:"-unknown-");
992 if (vc->sec_elmnt_count != 1) {
993 printf(" Secondary Position[%d] : %d of %d\n", n,
994 vc->sec_elmnt_seq, vc->sec_elmnt_count);
995 printf(" Secondary Level[%d] : %s\n", n,
996 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
997 }
998 printf(" Device Size[%d] : %llu\n", n,
999 __be64_to_cpu(vc->blocks)/2);
1000 printf(" Array Size[%d] : %llu\n", n,
1001 __be64_to_cpu(vc->array_blocks)/2);
1002 }
1003 }
1004
1005 static void examine_vds(struct ddf_super *sb)
1006 {
1007 int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1008 int i;
1009 printf(" Virtual Disks : %d\n", cnt);
1010
1011 for (i=0; i<cnt; i++) {
1012 struct virtual_entry *ve = &sb->virt->entries[i];
1013 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1014 printf("\n");
1015 printf(" unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1016 printf(" state[%d] : %s, %s%s\n", i,
1017 map_num(ddf_state, ve->state & 7),
1018 (ve->state & 8) ? "Morphing, ": "",
1019 (ve->state & 16)? "Not Consistent" : "Consistent");
1020 printf(" init state[%d] : %s\n", i,
1021 map_num(ddf_init_state, ve->init_state&3));
1022 printf(" access[%d] : %s\n", i,
1023 map_num(ddf_access, (ve->init_state>>6) & 3));
1024 printf(" Name[%d] : %.16s\n", i, ve->name);
1025 examine_vd(i, sb, ve->guid);
1026 }
1027 if (cnt) printf("\n");
1028 }
1029
1030 static void examine_pds(struct ddf_super *sb)
1031 {
1032 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1033 int i;
1034 struct dl *dl;
1035 printf(" Physical Disks : %d\n", cnt);
1036
1037 for (i=0 ; i<cnt ; i++) {
1038 struct phys_disk_entry *pd = &sb->phys->entries[i];
1039 int type = __be16_to_cpu(pd->type);
1040 int state = __be16_to_cpu(pd->state);
1041
1042 printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1043 printf("\n");
1044 printf(" ref[%d] : %08x\n", i,
1045 __be32_to_cpu(pd->refnum));
1046 printf(" mode[%d] : %s%s%s%s%s\n", i,
1047 (type&2) ? "active":"",
1048 (type&4) ? "Global Spare":"",
1049 (type&8) ? "spare" : "",
1050 (type&16)? ", foreign" : "",
1051 (type&32)? "pass-through" : "");
1052 printf(" state[%d] : %s%s%s%s%s%s%s\n", i,
1053 (state&1)? "Online": "Offline",
1054 (state&2)? ", Failed": "",
1055 (state&4)? ", Rebuilding": "",
1056 (state&8)? ", in-transition": "",
1057 (state&16)? ", SMART errors": "",
1058 (state&32)? ", Unrecovered Read Errors": "",
1059 (state&64)? ", Missing" : "");
1060 printf(" Avail Size[%d] : %llu K\n", i,
1061 __be64_to_cpu(pd->config_size)>>1);
1062 for (dl = sb->dlist; dl ; dl = dl->next) {
1063 if (dl->disk.refnum == pd->refnum) {
1064 char *dv = map_dev(dl->major, dl->minor, 0);
1065 if (dv)
1066 printf(" Device[%d] : %s\n",
1067 i, dv);
1068 }
1069 }
1070 printf("\n");
1071 }
1072 }
1073
1074 static void examine_super_ddf(struct supertype *st, char *homehost)
1075 {
1076 struct ddf_super *sb = st->sb;
1077
1078 printf(" Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1079 printf(" Version : %.8s\n", sb->anchor.revision);
1080 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1081 printf("\n");
1082 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1083 printf("\n");
1084 printf(" Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1085 printf(" Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1086 ?"yes" : "no");
1087 examine_vds(sb);
1088 examine_pds(sb);
1089 }
1090
1091 static void brief_examine_super_ddf(struct supertype *st)
1092 {
1093 /* We just write a generic DDF ARRAY entry
1094 * The uuid is all hex, 6 groups of 4 bytes
1095 */
1096 struct ddf_super *ddf = st->sb;
1097 int i;
1098 printf("ARRAY /dev/ddf metadata=ddf UUID=");
1099 for (i = 0; i < DDF_GUID_LEN; i++) {
1100 if ((i&3) == 0 && i != 0)
1101 printf(":");
1102 printf("%02X", 255&ddf->anchor.guid[i]);
1103 }
1104 printf("\n");
1105 }
1106
1107 static void detail_super_ddf(struct supertype *st, char *homehost)
1108 {
1109 /* FIXME later
1110 * Could print DDF GUID
1111 * Need to find which array
1112 * If whole, briefly list all arrays
1113 * If one, give name
1114 */
1115 }
1116
1117 static void brief_detail_super_ddf(struct supertype *st)
1118 {
1119 /* FIXME I really need to know which array we are detailing.
1120 * Can that be stored in ddf_super??
1121 */
1122 // struct ddf_super *ddf = st->sb;
1123 }
1124 #endif
1125
1126 static int match_home_ddf(struct supertype *st, char *homehost)
1127 {
1128 /* It matches 'this' host if the controller is a
1129 * Linux-MD controller with vendor_data matching
1130 * the hostname
1131 */
1132 struct ddf_super *ddf = st->sb;
1133 int len = strlen(homehost);
1134
1135 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1136 len < sizeof(ddf->controller.vendor_data) &&
1137 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1138 ddf->controller.vendor_data[len] == 0);
1139 }
1140
1141 static struct vd_config *find_vdcr(struct ddf_super *ddf, int inst)
1142 {
1143 struct vcl *v;
1144
1145 for (v = ddf->conflist; v; v = v->next)
1146 if (inst == v->vcnum)
1147 return &v->conf;
1148 return NULL;
1149 }
1150
1151 static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1152 {
1153 /* Find the entry in phys_disk which has the given refnum
1154 * and return it's index
1155 */
1156 int i;
1157 for (i=0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1158 if (ddf->phys->entries[i].refnum == phys_refnum)
1159 return i;
1160 return -1;
1161 }
1162
1163 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1164 {
1165 /* The uuid returned here is used for:
1166 * uuid to put into bitmap file (Create, Grow)
1167 * uuid for backup header when saving critical section (Grow)
1168 * comparing uuids when re-adding a device into an array
1169 * For each of these we can make do with a truncated
1170 * or hashed uuid rather than the original, as long as
1171 * everyone agrees.
1172 * In each case the uuid required is that of the data-array,
1173 * not the device-set.
1174 * In the case of SVD we assume the BVD is of interest,
1175 * though that might be the case if a bitmap were made for
1176 * a mirrored SVD - worry about that later.
1177 * So we need to find the VD configuration record for the
1178 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1179 * The first 16 bytes of the sha1 of these is used.
1180 */
1181 struct ddf_super *ddf = st->sb;
1182 struct vcl *vcl = ddf->currentconf;
1183
1184 if (!vcl)
1185 memset(uuid, 0, sizeof (uuid));
1186 else {
1187 char buf[20];
1188 struct sha1_ctx ctx;
1189 sha1_init_ctx(&ctx);
1190 sha1_process_bytes(&vcl->conf.guid, DDF_GUID_LEN, &ctx);
1191 if (vcl->conf.sec_elmnt_count > 1)
1192 sha1_process_bytes(&vcl->conf.sec_elmnt_seq, 1, &ctx);
1193 sha1_finish_ctx(&ctx, buf);
1194 memcpy(uuid, buf, sizeof(uuid));
1195 }
1196 }
1197
1198 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info);
1199
1200 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info)
1201 {
1202 struct ddf_super *ddf = st->sb;
1203
1204 if (ddf->currentconf) {
1205 getinfo_super_ddf_bvd(st, info);
1206 return;
1207 }
1208
1209 info->array.raid_disks = __be16_to_cpu(ddf->phys->used_pdes);
1210 info->array.level = LEVEL_CONTAINER;
1211 info->array.layout = 0;
1212 info->array.md_minor = -1;
1213 info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)
1214 (ddf->anchor.guid+16));
1215 info->array.utime = 0;
1216 info->array.chunk_size = 0;
1217
1218
1219 info->disk.major = 0;
1220 info->disk.minor = 0;
1221 if (ddf->dlist) {
1222 info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1223 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1224
1225 info->data_offset = __be64_to_cpu(ddf->phys->
1226 entries[info->disk.raid_disk].
1227 config_size);
1228 info->component_size = ddf->dlist->size - info->data_offset;
1229 } else {
1230 info->disk.number = -1;
1231 // info->disk.raid_disk = find refnum in the table and use index;
1232 }
1233 info->disk.state = (1 << MD_DISK_SYNC);
1234
1235
1236 info->reshape_active = 0;
1237
1238 strcpy(info->text_version, "ddf");
1239
1240 // uuid_from_super_ddf(info->uuid, sbv);
1241
1242 // info->name[] ?? ;
1243 }
1244
1245 static int rlq_to_layout(int rlq, int prl, int raiddisks);
1246
1247 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info)
1248 {
1249 struct ddf_super *ddf = st->sb;
1250 struct vcl *vc = ddf->currentconf;
1251 int cd = ddf->currentdev;
1252
1253 /* FIXME this returns BVD info - what if we want SVD ?? */
1254
1255 info->array.raid_disks = __be16_to_cpu(vc->conf.prim_elmnt_count);
1256 info->array.level = map_num1(ddf_level_num, vc->conf.prl);
1257 info->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
1258 info->array.raid_disks);
1259 info->array.md_minor = -1;
1260 info->array.ctime = DECADE +
1261 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
1262 info->array.utime = DECADE + __be32_to_cpu(vc->conf.timestamp);
1263 info->array.chunk_size = 512 << vc->conf.chunk_shift;
1264
1265 if (cd >= 0 && cd < ddf->mppe) {
1266 info->data_offset = __be64_to_cpu(vc->lba_offset[cd]);
1267 if (vc->block_sizes)
1268 info->component_size = vc->block_sizes[cd];
1269 else
1270 info->component_size = __be64_to_cpu(vc->conf.blocks);
1271 }
1272
1273 info->disk.major = 0;
1274 info->disk.minor = 0;
1275 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1276 // info->disk.raid_disk = find refnum in the table and use index;
1277 // info->disk.state = ???;
1278
1279 info->resync_start = 0;
1280 if (!(ddf->virt->entries[info->container_member].state
1281 & DDF_state_inconsistent) &&
1282 (ddf->virt->entries[info->container_member].init_state
1283 & DDF_initstate_mask)
1284 == DDF_init_full)
1285 info->resync_start = ~0ULL;
1286
1287 uuid_from_super_ddf(st, info->uuid);
1288
1289 info->container_member = atoi(st->subarray);
1290 sprintf(info->text_version, "/%s/%s",
1291 devnum2devname(st->container_dev),
1292 st->subarray);
1293
1294 // info->name[] ?? ;
1295 }
1296
1297
1298 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1299 char *update,
1300 char *devname, int verbose,
1301 int uuid_set, char *homehost)
1302 {
1303 /* For 'assemble' and 'force' we need to return non-zero if any
1304 * change was made. For others, the return value is ignored.
1305 * Update options are:
1306 * force-one : This device looks a bit old but needs to be included,
1307 * update age info appropriately.
1308 * assemble: clear any 'faulty' flag to allow this device to
1309 * be assembled.
1310 * force-array: Array is degraded but being forced, mark it clean
1311 * if that will be needed to assemble it.
1312 *
1313 * newdev: not used ????
1314 * grow: Array has gained a new device - this is currently for
1315 * linear only
1316 * resync: mark as dirty so a resync will happen.
1317 * uuid: Change the uuid of the array to match what is given
1318 * homehost: update the recorded homehost
1319 * name: update the name - preserving the homehost
1320 * _reshape_progress: record new reshape_progress position.
1321 *
1322 * Following are not relevant for this version:
1323 * sparc2.2 : update from old dodgey metadata
1324 * super-minor: change the preferred_minor number
1325 * summaries: update redundant counters.
1326 */
1327 int rv = 0;
1328 // struct ddf_super *ddf = st->sb;
1329 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1330 // struct virtual_entry *ve = find_ve(ddf);
1331
1332 /* we don't need to handle "force-*" or "assemble" as
1333 * there is no need to 'trick' the kernel. We the metadata is
1334 * first updated to activate the array, all the implied modifications
1335 * will just happen.
1336 */
1337
1338 if (strcmp(update, "grow") == 0) {
1339 /* FIXME */
1340 }
1341 if (strcmp(update, "resync") == 0) {
1342 // info->resync_checkpoint = 0;
1343 }
1344 /* We ignore UUID updates as they make even less sense
1345 * with DDF
1346 */
1347 if (strcmp(update, "homehost") == 0) {
1348 /* homehost is stored in controller->vendor_data,
1349 * or it is when we are the vendor
1350 */
1351 // if (info->vendor_is_local)
1352 // strcpy(ddf->controller.vendor_data, homehost);
1353 }
1354 if (strcmp(update, "name") == 0) {
1355 /* name is stored in virtual_entry->name */
1356 // memset(ve->name, ' ', 16);
1357 // strncpy(ve->name, info->name, 16);
1358 }
1359 if (strcmp(update, "_reshape_progress") == 0) {
1360 /* We don't support reshape yet */
1361 }
1362
1363 // update_all_csum(ddf);
1364
1365 return rv;
1366 }
1367
1368 static void make_header_guid(char *guid)
1369 {
1370 __u32 stamp;
1371 int rfd;
1372 /* Create a DDF Header of Virtual Disk GUID */
1373
1374 /* 24 bytes of fiction required.
1375 * first 8 are a 'vendor-id' - "Linux-MD"
1376 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1377 * Remaining 8 random number plus timestamp
1378 */
1379 memcpy(guid, T10, sizeof(T10));
1380 stamp = __cpu_to_be32(0xdeadbeef);
1381 memcpy(guid+8, &stamp, 4);
1382 stamp = __cpu_to_be32(0);
1383 memcpy(guid+12, &stamp, 4);
1384 stamp = __cpu_to_be32(time(0) - DECADE);
1385 memcpy(guid+16, &stamp, 4);
1386 rfd = open("/dev/urandom", O_RDONLY);
1387 if (rfd < 0 || read(rfd, &stamp, 4) != 4)
1388 stamp = random();
1389 memcpy(guid+20, &stamp, 4);
1390 if (rfd >= 0) close(rfd);
1391 }
1392
1393 static int init_super_ddf_bvd(struct supertype *st,
1394 mdu_array_info_t *info,
1395 unsigned long long size,
1396 char *name, char *homehost,
1397 int *uuid);
1398
1399 static int init_super_ddf(struct supertype *st,
1400 mdu_array_info_t *info,
1401 unsigned long long size, char *name, char *homehost,
1402 int *uuid)
1403 {
1404 /* This is primarily called by Create when creating a new array.
1405 * We will then get add_to_super called for each component, and then
1406 * write_init_super called to write it out to each device.
1407 * For DDF, Create can create on fresh devices or on a pre-existing
1408 * array.
1409 * To create on a pre-existing array a different method will be called.
1410 * This one is just for fresh drives.
1411 *
1412 * We need to create the entire 'ddf' structure which includes:
1413 * DDF headers - these are easy.
1414 * Controller data - a Sector describing this controller .. not that
1415 * this is a controller exactly.
1416 * Physical Disk Record - one entry per device, so
1417 * leave plenty of space.
1418 * Virtual Disk Records - again, just leave plenty of space.
1419 * This just lists VDs, doesn't give details
1420 * Config records - describes the VDs that use this disk
1421 * DiskData - describes 'this' device.
1422 * BadBlockManagement - empty
1423 * Diag Space - empty
1424 * Vendor Logs - Could we put bitmaps here?
1425 *
1426 */
1427 struct ddf_super *ddf;
1428 char hostname[17];
1429 int hostlen;
1430 int max_phys_disks, max_virt_disks;
1431 unsigned long long sector;
1432 int clen;
1433 int i;
1434 int pdsize, vdsize;
1435 struct phys_disk *pd;
1436 struct virtual_disk *vd;
1437
1438 if (!info) {
1439 st->sb = NULL;
1440 return 0;
1441 }
1442 if (st->sb)
1443 return init_super_ddf_bvd(st, info, size, name, homehost,
1444 uuid);
1445
1446 ddf = malloc(sizeof(*ddf));
1447 memset(ddf, 0, sizeof(*ddf));
1448 ddf->dlist = NULL; /* no physical disks yet */
1449 ddf->conflist = NULL; /* No virtual disks yet */
1450
1451 /* At least 32MB *must* be reserved for the ddf. So let's just
1452 * start 32MB from the end, and put the primary header there.
1453 * Don't do secondary for now.
1454 * We don't know exactly where that will be yet as it could be
1455 * different on each device. To just set up the lengths.
1456 *
1457 */
1458
1459 ddf->anchor.magic = DDF_HEADER_MAGIC;
1460 make_header_guid(ddf->anchor.guid);
1461
1462 memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
1463 ddf->anchor.seq = __cpu_to_be32(1);
1464 ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1465 ddf->anchor.openflag = 0xFF;
1466 ddf->anchor.foreignflag = 0;
1467 ddf->anchor.enforcegroups = 0; /* Is this best?? */
1468 ddf->anchor.pad0 = 0xff;
1469 memset(ddf->anchor.pad1, 0xff, 12);
1470 memset(ddf->anchor.header_ext, 0xff, 32);
1471 ddf->anchor.primary_lba = ~(__u64)0;
1472 ddf->anchor.secondary_lba = ~(__u64)0;
1473 ddf->anchor.type = DDF_HEADER_ANCHOR;
1474 memset(ddf->anchor.pad2, 0xff, 3);
1475 ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1476 ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1477 of 32M reserved.. */
1478 max_phys_disks = 1023; /* Should be enough */
1479 ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1480 max_virt_disks = 255;
1481 ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1482 ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1483 ddf->max_part = 64;
1484 ddf->mppe = 256;
1485 ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
1486 ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1487 ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe);
1488 memset(ddf->anchor.pad3, 0xff, 54);
1489 /* controller sections is one sector long immediately
1490 * after the ddf header */
1491 sector = 1;
1492 ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1493 ddf->anchor.controller_section_length = __cpu_to_be32(1);
1494 sector += 1;
1495
1496 /* phys is 8 sectors after that */
1497 pdsize = ROUND_UP(sizeof(struct phys_disk) +
1498 sizeof(struct phys_disk_entry)*max_phys_disks,
1499 512);
1500 switch(pdsize/512) {
1501 case 2: case 8: case 32: case 128: case 512: break;
1502 default: abort();
1503 }
1504 ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1505 ddf->anchor.phys_section_length =
1506 __cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1507 sector += pdsize/512;
1508
1509 /* virt is another 32 sectors */
1510 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1511 sizeof(struct virtual_entry) * max_virt_disks,
1512 512);
1513 switch(vdsize/512) {
1514 case 2: case 8: case 32: case 128: case 512: break;
1515 default: abort();
1516 }
1517 ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1518 ddf->anchor.virt_section_length =
1519 __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1520 sector += vdsize/512;
1521
1522 clen = ddf->conf_rec_len * (ddf->max_part+1);
1523 ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1524 ddf->anchor.config_section_length = __cpu_to_be32(clen);
1525 sector += clen;
1526
1527 ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1528 ddf->anchor.data_section_length = __cpu_to_be32(1);
1529 sector += 1;
1530
1531 ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1532 ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1533 ddf->anchor.diag_space_length = __cpu_to_be32(0);
1534 ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1535 ddf->anchor.vendor_length = __cpu_to_be32(0);
1536 ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1537
1538 memset(ddf->anchor.pad4, 0xff, 256);
1539
1540 memcpy(&ddf->primary, &ddf->anchor, 512);
1541 memcpy(&ddf->secondary, &ddf->anchor, 512);
1542
1543 ddf->primary.openflag = 1; /* I guess.. */
1544 ddf->primary.type = DDF_HEADER_PRIMARY;
1545
1546 ddf->secondary.openflag = 1; /* I guess.. */
1547 ddf->secondary.type = DDF_HEADER_SECONDARY;
1548
1549 ddf->active = &ddf->primary;
1550
1551 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1552
1553 /* 24 more bytes of fiction required.
1554 * first 8 are a 'vendor-id' - "Linux-MD"
1555 * Remaining 16 are serial number.... maybe a hostname would do?
1556 */
1557 memcpy(ddf->controller.guid, T10, sizeof(T10));
1558 gethostname(hostname, sizeof(hostname));
1559 hostname[sizeof(hostname) - 1] = 0;
1560 hostlen = strlen(hostname);
1561 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1562 for (i = strlen(T10) ; i+hostlen < 24; i++)
1563 ddf->controller.guid[i] = ' ';
1564
1565 ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1566 ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1567 ddf->controller.type.sub_vendor_id = 0;
1568 ddf->controller.type.sub_device_id = 0;
1569 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1570 memset(ddf->controller.pad, 0xff, 8);
1571 memset(ddf->controller.vendor_data, 0xff, 448);
1572
1573 pd = ddf->phys = malloc(pdsize);
1574 ddf->pdsize = pdsize;
1575
1576 memset(pd, 0xff, pdsize);
1577 memset(pd, 0, sizeof(*pd));
1578 pd->magic = DDF_PHYS_DATA_MAGIC;
1579 pd->used_pdes = __cpu_to_be16(0);
1580 pd->max_pdes = __cpu_to_be16(max_phys_disks);
1581 memset(pd->pad, 0xff, 52);
1582
1583 vd = ddf->virt = malloc(vdsize);
1584 ddf->vdsize = vdsize;
1585 memset(vd, 0, vdsize);
1586 vd->magic = DDF_VIRT_RECORDS_MAGIC;
1587 vd->populated_vdes = __cpu_to_be16(0);
1588 vd->max_vdes = __cpu_to_be16(max_virt_disks);
1589 memset(vd->pad, 0xff, 52);
1590
1591 for (i=0; i<max_virt_disks; i++)
1592 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1593
1594 st->sb = ddf;
1595 ddf->updates_pending = 1;
1596 return 1;
1597 }
1598
1599 static int all_ff(char *guid)
1600 {
1601 int i;
1602 for (i = 0; i < DDF_GUID_LEN; i++)
1603 if (guid[i] != (char)0xff)
1604 return 0;
1605 return 1;
1606 }
1607 static int chunk_to_shift(int chunksize)
1608 {
1609 return ffs(chunksize/512)-1;
1610 }
1611
1612 static int level_to_prl(int level)
1613 {
1614 switch (level) {
1615 case LEVEL_LINEAR: return DDF_CONCAT;
1616 case 0: return DDF_RAID0;
1617 case 1: return DDF_RAID1;
1618 case 4: return DDF_RAID4;
1619 case 5: return DDF_RAID5;
1620 case 6: return DDF_RAID6;
1621 default: return -1;
1622 }
1623 }
1624 static int layout_to_rlq(int level, int layout, int raiddisks)
1625 {
1626 switch(level) {
1627 case 0:
1628 return DDF_RAID0_SIMPLE;
1629 case 1:
1630 switch(raiddisks) {
1631 case 2: return DDF_RAID1_SIMPLE;
1632 case 3: return DDF_RAID1_MULTI;
1633 default: return -1;
1634 }
1635 case 4:
1636 switch(layout) {
1637 case 0: return DDF_RAID4_N;
1638 }
1639 break;
1640 case 5:
1641 case 6:
1642 switch(layout) {
1643 case ALGORITHM_LEFT_ASYMMETRIC:
1644 return DDF_RAID5_N_RESTART;
1645 case ALGORITHM_RIGHT_ASYMMETRIC:
1646 if (level == 5)
1647 return DDF_RAID5_0_RESTART;
1648 else
1649 return DDF_RAID6_0_RESTART;
1650 case ALGORITHM_LEFT_SYMMETRIC:
1651 return DDF_RAID5_N_CONTINUE;
1652 case ALGORITHM_RIGHT_SYMMETRIC:
1653 return -1; /* not mentioned in standard */
1654 }
1655 }
1656 return -1;
1657 }
1658
1659 static int rlq_to_layout(int rlq, int prl, int raiddisks)
1660 {
1661 switch(prl) {
1662 case DDF_RAID0:
1663 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1664 case DDF_RAID1:
1665 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1666 on raiddisks*/
1667 case DDF_RAID4:
1668 switch(rlq) {
1669 case DDF_RAID4_N:
1670 return 0;
1671 default:
1672 /* not supported */
1673 return -1; /* FIXME this isn't checked */
1674 }
1675 case DDF_RAID5:
1676 switch(rlq) {
1677 case DDF_RAID5_N_RESTART:
1678 return ALGORITHM_LEFT_ASYMMETRIC;
1679 case DDF_RAID5_0_RESTART:
1680 return ALGORITHM_RIGHT_ASYMMETRIC;
1681 case DDF_RAID5_N_CONTINUE:
1682 return ALGORITHM_LEFT_SYMMETRIC;
1683 default:
1684 return -1;
1685 }
1686 case DDF_RAID6:
1687 switch(rlq) {
1688 case DDF_RAID5_N_RESTART:
1689 return ALGORITHM_LEFT_ASYMMETRIC;
1690 case DDF_RAID6_0_RESTART:
1691 return ALGORITHM_RIGHT_ASYMMETRIC;
1692 case DDF_RAID5_N_CONTINUE:
1693 return ALGORITHM_LEFT_SYMMETRIC;
1694 default:
1695 return -1;
1696 }
1697 }
1698 return -1;
1699 }
1700
1701 struct extent {
1702 unsigned long long start, size;
1703 };
1704 static int cmp_extent(const void *av, const void *bv)
1705 {
1706 const struct extent *a = av;
1707 const struct extent *b = bv;
1708 if (a->start < b->start)
1709 return -1;
1710 if (a->start > b->start)
1711 return 1;
1712 return 0;
1713 }
1714
1715 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
1716 {
1717 /* find a list of used extents on the give physical device
1718 * (dnum) of the given ddf.
1719 * Return a malloced array of 'struct extent'
1720
1721 FIXME ignore DDF_Legacy devices?
1722
1723 */
1724 struct extent *rv;
1725 int n = 0;
1726 int i, j;
1727
1728 rv = malloc(sizeof(struct extent) * (ddf->max_part + 2));
1729 if (!rv)
1730 return NULL;
1731
1732 for (i = 0; i < ddf->max_part; i++) {
1733 struct vcl *v = dl->vlist[i];
1734 if (v == NULL)
1735 continue;
1736 for (j=0; j < v->conf.prim_elmnt_count; j++)
1737 if (v->conf.phys_refnum[j] == dl->disk.refnum) {
1738 /* This device plays role 'j' in 'v'. */
1739 rv[n].start = __be64_to_cpu(v->lba_offset[j]);
1740 rv[n].size = __be64_to_cpu(v->conf.blocks);
1741 n++;
1742 break;
1743 }
1744 }
1745 qsort(rv, n, sizeof(*rv), cmp_extent);
1746
1747 rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
1748 rv[n].size = 0;
1749 return rv;
1750 }
1751
1752 static int init_super_ddf_bvd(struct supertype *st,
1753 mdu_array_info_t *info,
1754 unsigned long long size,
1755 char *name, char *homehost,
1756 int *uuid)
1757 {
1758 /* We are creating a BVD inside a pre-existing container.
1759 * so st->sb is already set.
1760 * We need to create a new vd_config and a new virtual_entry
1761 */
1762 struct ddf_super *ddf = st->sb;
1763 int venum;
1764 struct virtual_entry *ve;
1765 struct vcl *vcl;
1766 struct vd_config *vc;
1767
1768 if (__be16_to_cpu(ddf->virt->populated_vdes)
1769 >= __be16_to_cpu(ddf->virt->max_vdes)) {
1770 fprintf(stderr, Name": This ddf already has the "
1771 "maximum of %d virtual devices\n",
1772 __be16_to_cpu(ddf->virt->max_vdes));
1773 return 0;
1774 }
1775
1776 for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1777 if (all_ff(ddf->virt->entries[venum].guid))
1778 break;
1779 if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1780 fprintf(stderr, Name ": Cannot find spare slot for "
1781 "virtual disk - DDF is corrupt\n");
1782 return 0;
1783 }
1784 ve = &ddf->virt->entries[venum];
1785
1786 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1787 * timestamp, random number
1788 */
1789 make_header_guid(ve->guid);
1790 ve->unit = __cpu_to_be16(info->md_minor);
1791 ve->pad0 = 0xFFFF;
1792 ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1793 ve->type = 0;
1794 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
1795 if (info->state & 1) /* clean */
1796 ve->init_state = DDF_init_full;
1797 else
1798 ve->init_state = DDF_init_not;
1799
1800 memset(ve->pad1, 0xff, 14);
1801 memset(ve->name, ' ', 16);
1802 if (name)
1803 strncpy(ve->name, name, 16);
1804 ddf->virt->populated_vdes =
1805 __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
1806
1807 /* Now create a new vd_config */
1808 vcl = malloc(offsetof(struct vcl, conf) + ddf->conf_rec_len * 512);
1809 vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
1810 vcl->vcnum = venum;
1811 sprintf(st->subarray, "%d", venum);
1812 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1813
1814 vc = &vcl->conf;
1815
1816 vc->magic = DDF_VD_CONF_MAGIC;
1817 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
1818 vc->timestamp = __cpu_to_be32(time(0)-DECADE);
1819 vc->seqnum = __cpu_to_be32(1);
1820 memset(vc->pad0, 0xff, 24);
1821 vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
1822 vc->chunk_shift = chunk_to_shift(info->chunk_size);
1823 vc->prl = level_to_prl(info->level);
1824 vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
1825 vc->sec_elmnt_count = 1;
1826 vc->sec_elmnt_seq = 0;
1827 vc->srl = 0;
1828 vc->blocks = __cpu_to_be64(info->size * 2);
1829 vc->array_blocks = __cpu_to_be64(
1830 calc_array_size(info->level, info->raid_disks, info->layout,
1831 info->chunk_size, info->size*2));
1832 memset(vc->pad1, 0xff, 8);
1833 vc->spare_refs[0] = 0xffffffff;
1834 vc->spare_refs[1] = 0xffffffff;
1835 vc->spare_refs[2] = 0xffffffff;
1836 vc->spare_refs[3] = 0xffffffff;
1837 vc->spare_refs[4] = 0xffffffff;
1838 vc->spare_refs[5] = 0xffffffff;
1839 vc->spare_refs[6] = 0xffffffff;
1840 vc->spare_refs[7] = 0xffffffff;
1841 memset(vc->cache_pol, 0, 8);
1842 vc->bg_rate = 0x80;
1843 memset(vc->pad2, 0xff, 3);
1844 memset(vc->pad3, 0xff, 52);
1845 memset(vc->pad4, 0xff, 192);
1846 memset(vc->v0, 0xff, 32);
1847 memset(vc->v1, 0xff, 32);
1848 memset(vc->v2, 0xff, 16);
1849 memset(vc->v3, 0xff, 16);
1850 memset(vc->vendor, 0xff, 32);
1851
1852 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
1853 memset(vc->phys_refnum+(ddf->mppe * 4), 0x00, 8*ddf->mppe);
1854
1855 vcl->next = ddf->conflist;
1856 ddf->conflist = vcl;
1857 ddf->currentconf = vcl;
1858 ddf->updates_pending = 1;
1859 return 1;
1860 }
1861
1862 static void add_to_super_ddf_bvd(struct supertype *st,
1863 mdu_disk_info_t *dk, int fd, char *devname)
1864 {
1865 /* fd and devname identify a device with-in the ddf container (st).
1866 * dk identifies a location in the new BVD.
1867 * We need to find suitable free space in that device and update
1868 * the phys_refnum and lba_offset for the newly created vd_config.
1869 * We might also want to update the type in the phys_disk
1870 * section.
1871 */
1872 struct dl *dl;
1873 struct ddf_super *ddf = st->sb;
1874 struct vd_config *vc;
1875 __u64 *lba_offset;
1876 int working;
1877 int i;
1878 unsigned long long blocks, pos, esize;
1879 struct extent *ex;
1880
1881 for (dl = ddf->dlist; dl ; dl = dl->next)
1882 if (dl->major == dk->major &&
1883 dl->minor == dk->minor)
1884 break;
1885 if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
1886 return;
1887
1888 vc = &ddf->currentconf->conf;
1889 lba_offset = ddf->currentconf->lba_offset;
1890
1891 ex = get_extents(ddf, dl);
1892 if (!ex)
1893 return;
1894
1895 i = 0; pos = 0;
1896 blocks = __be64_to_cpu(vc->blocks);
1897 if (ddf->currentconf->block_sizes)
1898 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
1899
1900 do {
1901 esize = ex[i].start - pos;
1902 if (esize >= blocks)
1903 break;
1904 pos = ex[i].start + ex[i].size;
1905 i++;
1906 } while (ex[i-1].size);
1907
1908 free(ex);
1909 if (esize < blocks)
1910 return;
1911
1912 ddf->currentdev = dk->raid_disk;
1913 vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
1914 lba_offset[dk->raid_disk] = __cpu_to_be64(pos);
1915
1916 for (i=0; i < ddf->max_part ; i++)
1917 if (dl->vlist[i] == NULL)
1918 break;
1919 if (i == ddf->max_part)
1920 return;
1921 dl->vlist[i] = ddf->currentconf;
1922
1923 dl->fd = fd;
1924 dl->devname = devname;
1925
1926 /* Check how many working raid_disks, and if we can mark
1927 * array as optimal yet
1928 */
1929 working = 0;
1930
1931 for (i=0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
1932 if (vc->phys_refnum[i] != 0xffffffff)
1933 working++;
1934
1935 /* Find which virtual_entry */
1936 i = ddf->currentconf->vcnum;
1937 if (working == __be16_to_cpu(vc->prim_elmnt_count))
1938 ddf->virt->entries[i].state =
1939 (ddf->virt->entries[i].state & ~DDF_state_mask)
1940 | DDF_state_optimal;
1941
1942 if (vc->prl == DDF_RAID6 &&
1943 working+1 == __be16_to_cpu(vc->prim_elmnt_count))
1944 ddf->virt->entries[i].state =
1945 (ddf->virt->entries[i].state & ~DDF_state_mask)
1946 | DDF_state_part_optimal;
1947
1948 ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
1949 ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
1950 ddf->updates_pending = 1;
1951 }
1952
1953 /* add a device to a container, either while creating it or while
1954 * expanding a pre-existing container
1955 */
1956 static void add_to_super_ddf(struct supertype *st,
1957 mdu_disk_info_t *dk, int fd, char *devname)
1958 {
1959 struct ddf_super *ddf = st->sb;
1960 struct dl *dd;
1961 time_t now;
1962 struct tm *tm;
1963 unsigned long long size;
1964 struct phys_disk_entry *pde;
1965 int n, i;
1966 struct stat stb;
1967
1968 if (ddf->currentconf) {
1969 add_to_super_ddf_bvd(st, dk, fd, devname);
1970 return;
1971 }
1972
1973 /* This is device numbered dk->number. We need to create
1974 * a phys_disk entry and a more detailed disk_data entry.
1975 */
1976 fstat(fd, &stb);
1977 dd = malloc(sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part);
1978 dd->major = major(stb.st_rdev);
1979 dd->minor = minor(stb.st_rdev);
1980 dd->devname = devname;
1981 dd->next = ddf->dlist;
1982 dd->fd = fd;
1983 dd->spare = NULL;
1984
1985 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
1986 now = time(0);
1987 tm = localtime(&now);
1988 sprintf(dd->disk.guid, "%8s%04d%02d%02d",
1989 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
1990 *(__u32*)(dd->disk.guid + 16) = random();
1991 *(__u32*)(dd->disk.guid + 20) = random();
1992
1993 do {
1994 /* Cannot be bothered finding a CRC of some irrelevant details*/
1995 dd->disk.refnum = random();
1996 for (i = __be16_to_cpu(ddf->active->max_pd_entries) - 1;
1997 i >= 0; i--)
1998 if (ddf->phys->entries[i].refnum == dd->disk.refnum)
1999 break;
2000 } while (i >= 0);
2001
2002 dd->disk.forced_ref = 1;
2003 dd->disk.forced_guid = 1;
2004 memset(dd->disk.vendor, ' ', 32);
2005 memcpy(dd->disk.vendor, "Linux", 5);
2006 memset(dd->disk.pad, 0xff, 442);
2007 for (i = 0; i < ddf->max_part ; i++)
2008 dd->vlist[i] = NULL;
2009
2010 n = __be16_to_cpu(ddf->phys->used_pdes);
2011 pde = &ddf->phys->entries[n];
2012 dd->pdnum = n;
2013
2014 n++;
2015 ddf->phys->used_pdes = __cpu_to_be16(n);
2016
2017 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2018 pde->refnum = dd->disk.refnum;
2019 pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2020 pde->state = __cpu_to_be16(DDF_Online);
2021 get_dev_size(fd, NULL, &size);
2022 /* We are required to reserve 32Meg, and record the size in sectors */
2023 pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
2024 sprintf(pde->path, "%17.17s","Information: nil") ;
2025 memset(pde->pad, 0xff, 6);
2026
2027 dd->size = size >> 9;
2028 ddf->dlist = dd;
2029 ddf->updates_pending = 1;
2030 }
2031
2032 /*
2033 * This is the write_init_super method for a ddf container. It is
2034 * called when creating a container or adding another device to a
2035 * container.
2036 */
2037
2038 #ifndef MDASSEMBLE
2039
2040 static unsigned char null_conf[4096];
2041
2042 static int __write_init_super_ddf(struct supertype *st, int do_close)
2043 {
2044
2045 struct ddf_super *ddf = st->sb;
2046 int i;
2047 struct dl *d;
2048 int n_config;
2049 int conf_size;
2050
2051 unsigned long long size, sector;
2052
2053 for (d = ddf->dlist; d; d=d->next) {
2054 int fd = d->fd;
2055
2056 if (fd < 0)
2057 continue;
2058
2059 /* We need to fill in the primary, (secondary) and workspace
2060 * lba's in the headers, set their checksums,
2061 * Also checksum phys, virt....
2062 *
2063 * Then write everything out, finally the anchor is written.
2064 */
2065 get_dev_size(fd, NULL, &size);
2066 size /= 512;
2067 ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
2068 ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
2069 ddf->anchor.seq = __cpu_to_be32(1);
2070 memcpy(&ddf->primary, &ddf->anchor, 512);
2071 memcpy(&ddf->secondary, &ddf->anchor, 512);
2072
2073 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
2074 ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
2075 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
2076
2077 ddf->primary.openflag = 0;
2078 ddf->primary.type = DDF_HEADER_PRIMARY;
2079
2080 ddf->secondary.openflag = 0;
2081 ddf->secondary.type = DDF_HEADER_SECONDARY;
2082
2083 ddf->primary.crc = calc_crc(&ddf->primary, 512);
2084 ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
2085
2086 sector = size - 16*1024*2;
2087 lseek64(fd, sector<<9, 0);
2088 write(fd, &ddf->primary, 512);
2089
2090 ddf->controller.crc = calc_crc(&ddf->controller, 512);
2091 write(fd, &ddf->controller, 512);
2092
2093 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
2094
2095 write(fd, ddf->phys, ddf->pdsize);
2096
2097 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
2098 write(fd, ddf->virt, ddf->vdsize);
2099
2100 /* Now write lots of config records. */
2101 n_config = ddf->max_part;
2102 conf_size = ddf->conf_rec_len * 512;
2103 for (i = 0 ; i <= n_config ; i++) {
2104 struct vcl *c = d->vlist[i];
2105 if (i == n_config)
2106 c = (struct vcl*)d->spare;
2107
2108 if (c) {
2109 c->conf.crc = calc_crc(&c->conf, conf_size);
2110 write(fd, &c->conf, conf_size);
2111 } else {
2112 if (null_conf[0] != 0xff)
2113 memset(null_conf, 0xff, sizeof(null_conf));
2114 int togo = conf_size;
2115 while (togo > sizeof(null_conf)) {
2116 write(fd, null_conf, sizeof(null_conf));
2117 togo -= sizeof(null_conf);
2118 }
2119 write(fd, null_conf, togo);
2120 }
2121 }
2122 d->disk.crc = calc_crc(&d->disk, 512);
2123 write(fd, &d->disk, 512);
2124
2125 /* Maybe do the same for secondary */
2126
2127 lseek64(fd, (size-1)*512, SEEK_SET);
2128 write(fd, &ddf->anchor, 512);
2129 if (do_close) {
2130 close(fd);
2131 d->fd = -1;
2132 }
2133 }
2134 return 1;
2135 }
2136
2137 static int write_init_super_ddf(struct supertype *st)
2138 {
2139
2140 if (st->update_tail) {
2141 /* queue the virtual_disk and vd_config as metadata updates */
2142 struct virtual_disk *vd;
2143 struct vd_config *vc;
2144 struct ddf_super *ddf = st->sb;
2145 int len;
2146
2147 /* First the virtual disk. We have a slightly fake header */
2148 len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry);
2149 vd = malloc(len);
2150 *vd = *ddf->virt;
2151 vd->entries[0] = ddf->virt->entries[ddf->currentconf->vcnum];
2152 vd->populated_vdes = __cpu_to_be16(ddf->currentconf->vcnum);
2153 append_metadata_update(st, vd, len);
2154
2155 /* Then the vd_config */
2156 len = ddf->conf_rec_len * 512;
2157 vc = malloc(len);
2158 memcpy(vc, &ddf->currentconf->conf, len);
2159 append_metadata_update(st, vc, len);
2160
2161 /* FIXME I need to close the fds! */
2162 return 0;
2163 } else
2164 return __write_init_super_ddf(st, 1);
2165 }
2166
2167 #endif
2168
2169 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize)
2170 {
2171 /* We must reserve the last 32Meg */
2172 if (devsize <= 32*1024*2)
2173 return 0;
2174 return devsize - 32*1024*2;
2175 }
2176
2177 #ifndef MDASSEMBLE
2178 static int validate_geometry_ddf_container(struct supertype *st,
2179 int level, int layout, int raiddisks,
2180 int chunk, unsigned long long size,
2181 char *dev, unsigned long long *freesize);
2182
2183 static int validate_geometry_ddf_bvd(struct supertype *st,
2184 int level, int layout, int raiddisks,
2185 int chunk, unsigned long long size,
2186 char *dev, unsigned long long *freesize);
2187
2188 static int validate_geometry_ddf(struct supertype *st,
2189 int level, int layout, int raiddisks,
2190 int chunk, unsigned long long size,
2191 char *dev, unsigned long long *freesize)
2192 {
2193 int fd;
2194 struct mdinfo *sra;
2195 int cfd;
2196
2197 /* ddf potentially supports lots of things, but it depends on
2198 * what devices are offered (and maybe kernel version?)
2199 * If given unused devices, we will make a container.
2200 * If given devices in a container, we will make a BVD.
2201 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2202 */
2203
2204 if (level == LEVEL_CONTAINER) {
2205 /* Must be a fresh device to add to a container */
2206 return validate_geometry_ddf_container(st, level, layout,
2207 raiddisks,
2208 chunk, size, dev, freesize);
2209 }
2210
2211 if (st->sb) {
2212 /* A container has already been opened, so we are
2213 * creating in there. Maybe a BVD, maybe an SVD.
2214 * Should make a distinction one day.
2215 */
2216 return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
2217 chunk, size, dev, freesize);
2218 }
2219 if (!dev) {
2220 /* Initial sanity check. Exclude illegal levels. */
2221 int i;
2222 for (i=0; ddf_level_num[i].num1 != MAXINT; i++)
2223 if (ddf_level_num[i].num2 == level)
2224 break;
2225 if (ddf_level_num[i].num1 == MAXINT)
2226 return 0;
2227 /* Should check layout? etc */
2228 return 1;
2229 }
2230
2231 /* This is the first device for the array.
2232 * If it is a container, we read it in and do automagic allocations,
2233 * no other devices should be given.
2234 * Otherwise it must be a member device of a container, and we
2235 * do manual allocation.
2236 * Later we should check for a BVD and make an SVD.
2237 */
2238 fd = open(dev, O_RDONLY|O_EXCL, 0);
2239 if (fd >= 0) {
2240 sra = sysfs_read(fd, 0, GET_VERSION);
2241 close(fd);
2242 if (sra && sra->array.major_version == -1 &&
2243 strcmp(sra->text_version, "ddf") == 0) {
2244
2245 /* load super */
2246 /* find space for 'n' devices. */
2247 /* remember the devices */
2248 /* Somehow return the fact that we have enough */
2249 }
2250
2251 fprintf(stderr,
2252 Name ": Cannot create this array on device %s\n",
2253 dev);
2254 return 0;
2255 }
2256 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2257 fprintf(stderr, Name ": Cannot open %s: %s\n",
2258 dev, strerror(errno));
2259 return 0;
2260 }
2261 /* Well, it is in use by someone, maybe a 'ddf' container. */
2262 cfd = open_container(fd);
2263 if (cfd < 0) {
2264 close(fd);
2265 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
2266 dev);
2267 return 0;
2268 }
2269 sra = sysfs_read(cfd, 0, GET_VERSION);
2270 close(fd);
2271 if (sra && sra->array.major_version == -1 &&
2272 strcmp(sra->text_version, "ddf") == 0) {
2273 /* This is a member of a ddf container. Load the container
2274 * and try to create a bvd
2275 */
2276 struct ddf_super *ddf;
2277 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL, 1) == 0) {
2278 st->sb = ddf;
2279 st->container_dev = fd2devnum(cfd);
2280 close(cfd);
2281 return validate_geometry_ddf_bvd(st, level, layout,
2282 raiddisks, chunk, size,
2283 dev, freesize);
2284 }
2285 close(cfd);
2286 } else /* device may belong to a different container */
2287 return 0;
2288
2289 return 1;
2290 }
2291
2292 static int validate_geometry_ddf_container(struct supertype *st,
2293 int level, int layout, int raiddisks,
2294 int chunk, unsigned long long size,
2295 char *dev, unsigned long long *freesize)
2296 {
2297 int fd;
2298 unsigned long long ldsize;
2299
2300 if (level != LEVEL_CONTAINER)
2301 return 0;
2302 if (!dev)
2303 return 1;
2304
2305 fd = open(dev, O_RDONLY|O_EXCL, 0);
2306 if (fd < 0) {
2307 fprintf(stderr, Name ": Cannot open %s: %s\n",
2308 dev, strerror(errno));
2309 return 0;
2310 }
2311 if (!get_dev_size(fd, dev, &ldsize)) {
2312 close(fd);
2313 return 0;
2314 }
2315 close(fd);
2316
2317 *freesize = avail_size_ddf(st, ldsize >> 9);
2318
2319 return 1;
2320 }
2321
2322 static int validate_geometry_ddf_bvd(struct supertype *st,
2323 int level, int layout, int raiddisks,
2324 int chunk, unsigned long long size,
2325 char *dev, unsigned long long *freesize)
2326 {
2327 struct stat stb;
2328 struct ddf_super *ddf = st->sb;
2329 struct dl *dl;
2330 unsigned long long pos = 0;
2331 unsigned long long maxsize;
2332 struct extent *e;
2333 int i;
2334 /* ddf/bvd supports lots of things, but not containers */
2335 if (level == LEVEL_CONTAINER)
2336 return 0;
2337 /* We must have the container info already read in. */
2338 if (!ddf)
2339 return 0;
2340
2341 if (!dev) {
2342 /* General test: make sure there is space for
2343 * 'raiddisks' device extents of size 'size'.
2344 */
2345 unsigned long long minsize = size;
2346 int dcnt = 0;
2347 if (minsize == 0)
2348 minsize = 8;
2349 for (dl = ddf->dlist; dl ; dl = dl->next)
2350 {
2351 int found = 0;
2352 pos = 0;
2353
2354 i = 0;
2355 e = get_extents(ddf, dl);
2356 if (!e) continue;
2357 do {
2358 unsigned long long esize;
2359 esize = e[i].start - pos;
2360 if (esize >= minsize)
2361 found = 1;
2362 pos = e[i].start + e[i].size;
2363 i++;
2364 } while (e[i-1].size);
2365 if (found)
2366 dcnt++;
2367 free(e);
2368 }
2369 if (dcnt < raiddisks) {
2370 fprintf(stderr, Name ": Not enough devices with space "
2371 "for this array (%d < %d)\n",
2372 dcnt, raiddisks);
2373 return 0;
2374 }
2375 return 1;
2376 }
2377 /* This device must be a member of the set */
2378 if (stat(dev, &stb) < 0)
2379 return 0;
2380 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2381 return 0;
2382 for (dl = ddf->dlist ; dl ; dl = dl->next) {
2383 if (dl->major == major(stb.st_rdev) &&
2384 dl->minor == minor(stb.st_rdev))
2385 break;
2386 }
2387 if (!dl) {
2388 fprintf(stderr, Name ": %s is not in the same DDF set\n",
2389 dev);
2390 return 0;
2391 }
2392 e = get_extents(ddf, dl);
2393 maxsize = 0;
2394 i = 0;
2395 if (e) do {
2396 unsigned long long esize;
2397 esize = e[i].start - pos;
2398 if (esize >= maxsize)
2399 maxsize = esize;
2400 pos = e[i].start + e[i].size;
2401 i++;
2402 } while (e[i-1].size);
2403 *freesize = maxsize;
2404 // FIXME here I am
2405
2406 return 1;
2407 }
2408
2409 static int load_super_ddf_all(struct supertype *st, int fd,
2410 void **sbp, char *devname, int keep_fd)
2411 {
2412 struct mdinfo *sra;
2413 struct ddf_super *super;
2414 struct mdinfo *sd, *best = NULL;
2415 int bestseq = 0;
2416 int seq;
2417 char nm[20];
2418 int dfd;
2419
2420 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
2421 if (!sra)
2422 return 1;
2423 if (sra->array.major_version != -1 ||
2424 sra->array.minor_version != -2 ||
2425 strcmp(sra->text_version, "ddf") != 0)
2426 return 1;
2427
2428 super = malloc(sizeof(*super));
2429 if (!super)
2430 return 1;
2431 memset(super, 0, sizeof(*super));
2432
2433 /* first, try each device, and choose the best ddf */
2434 for (sd = sra->devs ; sd ; sd = sd->next) {
2435 int rv;
2436 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2437 dfd = dev_open(nm, O_RDONLY);
2438 if (dfd < 0)
2439 return 2;
2440 rv = load_ddf_headers(dfd, super, NULL);
2441 close(dfd);
2442 if (rv == 0) {
2443 seq = __be32_to_cpu(super->active->seq);
2444 if (super->active->openflag)
2445 seq--;
2446 if (!best || seq > bestseq) {
2447 bestseq = seq;
2448 best = sd;
2449 }
2450 }
2451 }
2452 if (!best)
2453 return 1;
2454 /* OK, load this ddf */
2455 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2456 dfd = dev_open(nm, O_RDONLY);
2457 if (dfd < 0)
2458 return 1;
2459 load_ddf_headers(dfd, super, NULL);
2460 load_ddf_global(dfd, super, NULL);
2461 close(dfd);
2462 /* Now we need the device-local bits */
2463 for (sd = sra->devs ; sd ; sd = sd->next) {
2464 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2465 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2466 if (dfd < 0)
2467 return 2;
2468 seq = load_ddf_local(dfd, super, NULL, keep_fd);
2469 if (!keep_fd) close(dfd);
2470 }
2471 if (st->subarray[0]) {
2472 struct vcl *v;
2473
2474 for (v = super->conflist; v; v = v->next)
2475 if (v->vcnum == atoi(st->subarray))
2476 super->currentconf = v;
2477 if (!super->currentconf)
2478 return 1;
2479 }
2480 *sbp = super;
2481 if (st->ss == NULL) {
2482 st->ss = &super_ddf;
2483 st->minor_version = 0;
2484 st->max_devs = 512;
2485 st->container_dev = fd2devnum(fd);
2486 }
2487 return 0;
2488 }
2489 #endif
2490
2491 static struct mdinfo *container_content_ddf(struct supertype *st)
2492 {
2493 /* Given a container loaded by load_super_ddf_all,
2494 * extract information about all the arrays into
2495 * an mdinfo tree.
2496 *
2497 * For each vcl in conflist: create an mdinfo, fill it in,
2498 * then look for matching devices (phys_refnum) in dlist
2499 * and create appropriate device mdinfo.
2500 */
2501 struct ddf_super *ddf = st->sb;
2502 struct mdinfo *rest = NULL;
2503 struct vcl *vc;
2504
2505 for (vc = ddf->conflist ; vc ; vc=vc->next)
2506 {
2507 int i;
2508 struct mdinfo *this;
2509 this = malloc(sizeof(*this));
2510 memset(this, 0, sizeof(*this));
2511 this->next = rest;
2512 rest = this;
2513
2514 this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2515 this->array.raid_disks =
2516 __be16_to_cpu(vc->conf.prim_elmnt_count);
2517 this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2518 this->array.raid_disks);
2519 this->array.md_minor = -1;
2520 this->array.ctime = DECADE +
2521 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
2522 this->array.utime = DECADE +
2523 __be32_to_cpu(vc->conf.timestamp);
2524 this->array.chunk_size = 512 << vc->conf.chunk_shift;
2525
2526 i = vc->vcnum;
2527 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2528 (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
2529 DDF_init_full) {
2530 this->array.state = 0;
2531 this->resync_start = 0;
2532 } else {
2533 this->array.state = 1;
2534 this->resync_start = ~0ULL;
2535 }
2536 memcpy(this->name, ddf->virt->entries[i].name, 32);
2537 this->name[33]=0;
2538
2539 memset(this->uuid, 0, sizeof(this->uuid));
2540 this->component_size = __be64_to_cpu(vc->conf.blocks);
2541 this->array.size = this->component_size / 2;
2542 this->container_member = i;
2543
2544 sprintf(this->text_version, "/%s/%d",
2545 devnum2devname(st->container_dev),
2546 this->container_member);
2547
2548 for (i=0 ; i < ddf->mppe ; i++) {
2549 struct mdinfo *dev;
2550 struct dl *d;
2551
2552 if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
2553 continue;
2554
2555 this->array.working_disks++;
2556
2557 for (d = ddf->dlist; d ; d=d->next)
2558 if (d->disk.refnum == vc->conf.phys_refnum[i])
2559 break;
2560 if (d == NULL)
2561 break;
2562
2563 dev = malloc(sizeof(*dev));
2564 memset(dev, 0, sizeof(*dev));
2565 dev->next = this->devs;
2566 this->devs = dev;
2567
2568 dev->disk.number = __be32_to_cpu(d->disk.refnum);
2569 dev->disk.major = d->major;
2570 dev->disk.minor = d->minor;
2571 dev->disk.raid_disk = i;
2572 dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2573
2574 dev->events = __le32_to_cpu(ddf->primary.seq);
2575 dev->data_offset = vc->lba_offset[i];
2576 dev->component_size = __be64_to_cpu(vc->conf.blocks);
2577 if (d->devname)
2578 strcpy(dev->name, d->devname);
2579 }
2580 }
2581 return rest;
2582 }
2583
2584 static int store_zero_ddf(struct supertype *st, int fd)
2585 {
2586 unsigned long long dsize;
2587 char buf[512];
2588 memset(buf, 0, 512);
2589
2590 if (!get_dev_size(fd, NULL, &dsize))
2591 return 1;
2592
2593 lseek64(fd, dsize-512, 0);
2594 write(fd, buf, 512);
2595 return 0;
2596 }
2597
2598 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
2599 {
2600 /*
2601 * return:
2602 * 0 same, or first was empty, and second was copied
2603 * 1 second had wrong number
2604 * 2 wrong uuid
2605 * 3 wrong other info
2606 */
2607 struct ddf_super *first = st->sb;
2608 struct ddf_super *second = tst->sb;
2609
2610 if (!first) {
2611 st->sb = tst->sb;
2612 tst->sb = NULL;
2613 return 0;
2614 }
2615
2616 if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
2617 return 2;
2618
2619 /* FIXME should I look at anything else? */
2620 return 0;
2621 }
2622
2623 /*
2624 * A new array 'a' has been started which claims to be instance 'inst'
2625 * within container 'c'.
2626 * We need to confirm that the array matches the metadata in 'c' so
2627 * that we don't corrupt any metadata.
2628 */
2629 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
2630 {
2631 fprintf(stderr, "ddf: open_new %s\n", inst);
2632 a->info.container_member = atoi(inst);
2633 return 0;
2634 }
2635
2636 /*
2637 * The array 'a' is to be marked clean in the metadata.
2638 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2639 * clean up to the point (in sectors). If that cannot be recorded in the
2640 * metadata, then leave it as dirty.
2641 *
2642 * For DDF, we need to clear the DDF_state_inconsistent bit in the
2643 * !global! virtual_disk.virtual_entry structure.
2644 */
2645 static void ddf_set_array_state(struct active_array *a, int consistent)
2646 {
2647 struct ddf_super *ddf = a->container->sb;
2648 int inst = a->info.container_member;
2649 int old = ddf->virt->entries[inst].state;
2650 if (consistent)
2651 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
2652 else
2653 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
2654 if (old != ddf->virt->entries[inst].state)
2655 ddf->updates_pending = 1;
2656
2657 old = ddf->virt->entries[inst].init_state;
2658 ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
2659 if (a->resync_start == ~0ULL)
2660 ddf->virt->entries[inst].init_state |= DDF_init_full;
2661 else if (a->resync_start == 0)
2662 ddf->virt->entries[inst].init_state |= DDF_init_not;
2663 else
2664 ddf->virt->entries[inst].init_state |= DDF_init_quick;
2665 if (old != ddf->virt->entries[inst].init_state)
2666 ddf->updates_pending = 1;
2667
2668 printf("ddf mark %d %s %llu\n", inst, consistent?"clean":"dirty",
2669 a->resync_start);
2670 }
2671
2672 /*
2673 * The state of each disk is stored in the global phys_disk structure
2674 * in phys_disk.entries[n].state.
2675 * This makes various combinations awkward.
2676 * - When a device fails in any array, it must be failed in all arrays
2677 * that include a part of this device.
2678 * - When a component is rebuilding, we cannot include it officially in the
2679 * array unless this is the only array that uses the device.
2680 *
2681 * So: when transitioning:
2682 * Online -> failed, just set failed flag. monitor will propagate
2683 * spare -> online, the device might need to be added to the array.
2684 * spare -> failed, just set failed. Don't worry if in array or not.
2685 */
2686 static void ddf_set_disk(struct active_array *a, int n, int state)
2687 {
2688 struct ddf_super *ddf = a->container->sb;
2689 int inst = a->info.container_member;
2690 struct vd_config *vc = find_vdcr(ddf, inst);
2691 int pd = find_phys(ddf, vc->phys_refnum[n]);
2692 int i, st, working;
2693
2694 if (vc == NULL) {
2695 fprintf(stderr, "ddf: cannot find instance %d!!\n", inst);
2696 return;
2697 }
2698 if (pd < 0) {
2699 /* disk doesn't currently exist. If it is now in_sync,
2700 * insert it. */
2701 if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
2702 /* Find dev 'n' in a->info->devs, determine the
2703 * ddf refnum, and set vc->phys_refnum and update
2704 * phys->entries[]
2705 */
2706 /* FIXME */
2707 }
2708 } else {
2709 int old = ddf->phys->entries[pd].state;
2710 if (state & DS_FAULTY)
2711 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Failed);
2712 if (state & DS_INSYNC) {
2713 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Online);
2714 ddf->phys->entries[pd].state &= __cpu_to_be16(~DDF_Rebuilding);
2715 }
2716 if (old != ddf->phys->entries[pd].state)
2717 ddf->updates_pending = 1;
2718 }
2719
2720 fprintf(stderr, "ddf: set_disk %d to %x\n", n, state);
2721
2722 /* Now we need to check the state of the array and update
2723 * virtual_disk.entries[n].state.
2724 * It needs to be one of "optimal", "degraded", "failed".
2725 * I don't understand 'deleted' or 'missing'.
2726 */
2727 working = 0;
2728 for (i=0; i < a->info.array.raid_disks; i++) {
2729 pd = find_phys(ddf, vc->phys_refnum[i]);
2730 if (pd < 0)
2731 continue;
2732 st = __be16_to_cpu(ddf->phys->entries[pd].state);
2733 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
2734 == DDF_Online)
2735 working++;
2736 }
2737 state = DDF_state_degraded;
2738 if (working == a->info.array.raid_disks)
2739 state = DDF_state_optimal;
2740 else switch(vc->prl) {
2741 case DDF_RAID0:
2742 case DDF_CONCAT:
2743 case DDF_JBOD:
2744 state = DDF_state_failed;
2745 break;
2746 case DDF_RAID1:
2747 if (working == 0)
2748 state = DDF_state_failed;
2749 break;
2750 case DDF_RAID4:
2751 case DDF_RAID5:
2752 if (working < a->info.array.raid_disks-1)
2753 state = DDF_state_failed;
2754 break;
2755 case DDF_RAID6:
2756 if (working < a->info.array.raid_disks-2)
2757 state = DDF_state_failed;
2758 else if (working == a->info.array.raid_disks-1)
2759 state = DDF_state_part_optimal;
2760 break;
2761 }
2762
2763 if (ddf->virt->entries[inst].state !=
2764 ((ddf->virt->entries[inst].state & ~DDF_state_mask)
2765 | state)) {
2766
2767 ddf->virt->entries[inst].state =
2768 (ddf->virt->entries[inst].state & ~DDF_state_mask)
2769 | state;
2770 ddf->updates_pending = 1;
2771 }
2772
2773 }
2774
2775 static void ddf_sync_metadata(struct supertype *st)
2776 {
2777
2778 /*
2779 * Write all data to all devices.
2780 * Later, we might be able to track whether only local changes
2781 * have been made, or whether any global data has been changed,
2782 * but ddf is sufficiently weird that it probably always
2783 * changes global data ....
2784 */
2785 struct ddf_super *ddf = st->sb;
2786 if (!ddf->updates_pending)
2787 return;
2788 ddf->updates_pending = 0;
2789 __write_init_super_ddf(st, 0);
2790 fprintf(stderr, "ddf: sync_metadata\n");
2791 }
2792
2793 static void ddf_process_update(struct supertype *st,
2794 struct metadata_update *update)
2795 {
2796 /* Apply this update to the metadata.
2797 * The first 4 bytes are a DDF_*_MAGIC which guides
2798 * our actions.
2799 * Possible update are:
2800 * DDF_PHYS_RECORDS_MAGIC
2801 * Add a new physical device. Changes to this record
2802 * only happen implicitly.
2803 * used_pdes is the device number.
2804 * DDF_VIRT_RECORDS_MAGIC
2805 * Add a new VD. Possibly also change the 'access' bits.
2806 * populated_vdes is the entry number.
2807 * DDF_VD_CONF_MAGIC
2808 * New or updated VD. the VIRT_RECORD must already
2809 * exist. For an update, phys_refnum and lba_offset
2810 * (at least) are updated, and the VD_CONF must
2811 * be written to precisely those devices listed with
2812 * a phys_refnum.
2813 * DDF_SPARE_ASSIGN_MAGIC
2814 * replacement Spare Assignment Record... but for which device?
2815 *
2816 * So, e.g.:
2817 * - to create a new array, we send a VIRT_RECORD and
2818 * a VD_CONF. Then assemble and start the array.
2819 * - to activate a spare we send a VD_CONF to add the phys_refnum
2820 * and offset. This will also mark the spare as active with
2821 * a spare-assignment record.
2822 */
2823 struct ddf_super *ddf = st->sb;
2824 __u32 *magic = (__u32*)update->buf;
2825 struct phys_disk *pd;
2826 struct virtual_disk *vd;
2827 struct vd_config *vc;
2828 struct vcl *vcl;
2829 struct dl *dl;
2830 int mppe;
2831 int ent;
2832
2833 // printf("Process update %x\n", *magic);
2834
2835 switch (*magic) {
2836 case DDF_PHYS_RECORDS_MAGIC:
2837
2838 if (update->len != (sizeof(struct phys_disk) +
2839 sizeof(struct phys_disk_entry)))
2840 return;
2841 pd = (struct phys_disk*)update->buf;
2842
2843 ent = __be16_to_cpu(pd->used_pdes);
2844 if (ent >= __be16_to_cpu(ddf->phys->max_pdes))
2845 return;
2846 if (!all_ff(ddf->phys->entries[ent].guid))
2847 return;
2848 ddf->phys->entries[ent] = pd->entries[0];
2849 ddf->phys->used_pdes = __cpu_to_be16(1 +
2850 __be16_to_cpu(ddf->phys->used_pdes));
2851 ddf->updates_pending = 1;
2852 break;
2853
2854 case DDF_VIRT_RECORDS_MAGIC:
2855
2856 if (update->len != (sizeof(struct virtual_disk) +
2857 sizeof(struct virtual_entry)))
2858 return;
2859 vd = (struct virtual_disk*)update->buf;
2860
2861 ent = __be16_to_cpu(vd->populated_vdes);
2862 if (ent >= __be16_to_cpu(ddf->virt->max_vdes))
2863 return;
2864 if (!all_ff(ddf->virt->entries[ent].guid))
2865 return;
2866 ddf->virt->entries[ent] = vd->entries[0];
2867 ddf->virt->populated_vdes = __cpu_to_be16(1 +
2868 __be16_to_cpu(ddf->virt->populated_vdes));
2869 ddf->updates_pending = 1;
2870 break;
2871
2872 case DDF_VD_CONF_MAGIC:
2873 // printf("len %d %d\n", update->len, ddf->conf_rec_len);
2874
2875 mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries);
2876 if (update->len != ddf->conf_rec_len * 512)
2877 return;
2878 vc = (struct vd_config*)update->buf;
2879 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2880 if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
2881 break;
2882 // printf("vcl = %p\n", vcl);
2883 if (vcl) {
2884 /* An update, just copy the phys_refnum and lba_offset
2885 * fields
2886 */
2887 memcpy(vcl->conf.phys_refnum, vc->phys_refnum,
2888 mppe * (sizeof(__u32) + sizeof(__u64)));
2889 } else {
2890 /* A new VD_CONF */
2891 vcl = update->space;
2892 update->space = NULL;
2893 vcl->next = ddf->conflist;
2894 memcpy(&vcl->conf, vc, update->len);
2895 vcl->lba_offset = (__u64*)
2896 &vcl->conf.phys_refnum[mppe];
2897 ddf->conflist = vcl;
2898 }
2899 /* Now make sure vlist is correct for each dl. */
2900 for (dl = ddf->dlist; dl; dl = dl->next) {
2901 int dn;
2902 int vn = 0;
2903 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2904 for (dn=0; dn < ddf->mppe ; dn++)
2905 if (vcl->conf.phys_refnum[dn] ==
2906 dl->disk.refnum) {
2907 // printf("dev %d has %p at %d\n",
2908 // dl->pdnum, vcl, vn);
2909 dl->vlist[vn++] = vcl;
2910 break;
2911 }
2912 while (vn < ddf->max_part)
2913 dl->vlist[vn++] = NULL;
2914 if (dl->vlist[0]) {
2915 ddf->phys->entries[dl->pdnum].type &=
2916 ~__cpu_to_be16(DDF_Global_Spare);
2917 ddf->phys->entries[dl->pdnum].type |=
2918 __cpu_to_be16(DDF_Active_in_VD);
2919 }
2920 if (dl->spare) {
2921 ddf->phys->entries[dl->pdnum].type &=
2922 ~__cpu_to_be16(DDF_Global_Spare);
2923 ddf->phys->entries[dl->pdnum].type |=
2924 __cpu_to_be16(DDF_Spare);
2925 }
2926 if (!dl->vlist[0] && !dl->spare) {
2927 ddf->phys->entries[dl->pdnum].type |=
2928 __cpu_to_be16(DDF_Global_Spare);
2929 ddf->phys->entries[dl->pdnum].type &=
2930 ~__cpu_to_be16(DDF_Spare |
2931 DDF_Active_in_VD);
2932 }
2933 }
2934 ddf->updates_pending = 1;
2935 break;
2936 case DDF_SPARE_ASSIGN_MAGIC:
2937 default: break;
2938 }
2939 }
2940
2941 static void ddf_prepare_update(struct supertype *st,
2942 struct metadata_update *update)
2943 {
2944 /* This update arrived at managemon.
2945 * We are about to pass it to monitor.
2946 * If a malloc is needed, do it here.
2947 */
2948 struct ddf_super *ddf = st->sb;
2949 __u32 *magic = (__u32*)update->buf;
2950 if (*magic == DDF_VD_CONF_MAGIC)
2951 update->space = malloc(offsetof(struct vcl, conf)
2952 + ddf->conf_rec_len * 512);
2953 }
2954
2955 /*
2956 * Check if the array 'a' is degraded but not failed.
2957 * If it is, find as many spares as are available and needed and
2958 * arrange for their inclusion.
2959 * We only choose devices which are not already in the array,
2960 * and prefer those with a spare-assignment to this array.
2961 * otherwise we choose global spares - assuming always that
2962 * there is enough room.
2963 * For each spare that we assign, we return an 'mdinfo' which
2964 * describes the position for the device in the array.
2965 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
2966 * the new phys_refnum and lba_offset values.
2967 *
2968 * Only worry about BVDs at the moment.
2969 */
2970 static struct mdinfo *ddf_activate_spare(struct active_array *a,
2971 struct metadata_update **updates)
2972 {
2973 int working = 0;
2974 struct mdinfo *d;
2975 struct ddf_super *ddf = a->container->sb;
2976 int global_ok = 0;
2977 struct mdinfo *rv = NULL;
2978 struct mdinfo *di;
2979 struct metadata_update *mu;
2980 struct dl *dl;
2981 int i;
2982 struct vd_config *vc;
2983 __u64 *lba;
2984
2985 for (d = a->info.devs ; d ; d = d->next) {
2986 if ((d->curr_state & DS_FAULTY) &&
2987 d->state_fd >= 0)
2988 /* wait for Removal to happen */
2989 return NULL;
2990 if (d->state_fd >= 0)
2991 working ++;
2992 }
2993
2994 // printf("ddf_activate: working=%d (%d) level=%d\n", working, a->info.array.raid_disks,
2995 // a->info.array.level);
2996 if (working == a->info.array.raid_disks)
2997 return NULL; /* array not degraded */
2998 switch (a->info.array.level) {
2999 case 1:
3000 if (working == 0)
3001 return NULL; /* failed */
3002 break;
3003 case 4:
3004 case 5:
3005 if (working < a->info.array.raid_disks - 1)
3006 return NULL; /* failed */
3007 break;
3008 case 6:
3009 if (working < a->info.array.raid_disks - 2)
3010 return NULL; /* failed */
3011 break;
3012 default: /* concat or stripe */
3013 return NULL; /* failed */
3014 }
3015
3016 /* For each slot, if it is not working, find a spare */
3017 dl = ddf->dlist;
3018 for (i = 0; i < a->info.array.raid_disks; i++) {
3019 for (d = a->info.devs ; d ; d = d->next)
3020 if (d->disk.raid_disk == i)
3021 break;
3022 printf("found %d: %p %x\n", i, d, d?d->curr_state:0);
3023 if (d && (d->state_fd >= 0))
3024 continue;
3025
3026 /* OK, this device needs recovery. Find a spare */
3027 again:
3028 for ( ; dl ; dl = dl->next) {
3029 unsigned long long esize;
3030 unsigned long long pos;
3031 struct mdinfo *d2;
3032 int is_global = 0;
3033 int is_dedicated = 0;
3034 struct extent *ex;
3035 int j;
3036 /* If in this array, skip */
3037 for (d2 = a->info.devs ; d2 ; d2 = d2->next)
3038 if (d2->disk.major == dl->major &&
3039 d2->disk.minor == dl->minor) {
3040 printf("%x:%x already in array\n", dl->major, dl->minor);
3041 break;
3042 }
3043 if (d2)
3044 continue;
3045 if (ddf->phys->entries[dl->pdnum].type &
3046 __cpu_to_be16(DDF_Spare)) {
3047 /* Check spare assign record */
3048 if (dl->spare) {
3049 if (dl->spare->type & DDF_spare_dedicated) {
3050 /* check spare_ents for guid */
3051 for (j = 0 ;
3052 j < __be16_to_cpu(dl->spare->populated);
3053 j++) {
3054 if (memcmp(dl->spare->spare_ents[j].guid,
3055 ddf->virt->entries[a->info.container_member].guid,
3056 DDF_GUID_LEN) == 0)
3057 is_dedicated = 1;
3058 }
3059 } else
3060 is_global = 1;
3061 }
3062 } else if (ddf->phys->entries[dl->pdnum].type &
3063 __cpu_to_be16(DDF_Global_Spare)) {
3064 is_global = 1;
3065 }
3066 if ( ! (is_dedicated ||
3067 (is_global && global_ok))) {
3068 printf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
3069 is_dedicated, is_global);
3070 continue;
3071 }
3072
3073 /* We are allowed to use this device - is there space?
3074 * We need a->info.component_size sectors */
3075 ex = get_extents(ddf, dl);
3076 if (!ex) {
3077 printf("cannot get extents\n");
3078 continue;
3079 }
3080 j = 0; pos = 0;
3081 esize = 0;
3082
3083 do {
3084 esize = ex[j].start - pos;
3085 if (esize >= a->info.component_size)
3086 break;
3087 pos = ex[i].start + ex[i].size;
3088 i++;
3089 } while (ex[i-1].size);
3090
3091 free(ex);
3092 if (esize < a->info.component_size) {
3093 printf("%x:%x has no room: %llu %llu\n", dl->major, dl->minor,
3094 esize, a->info.component_size);
3095 /* No room */
3096 continue;
3097 }
3098
3099 /* Cool, we have a device with some space at pos */
3100 di = malloc(sizeof(*di));
3101 memset(di, 0, sizeof(*di));
3102 di->disk.number = i;
3103 di->disk.raid_disk = i;
3104 di->disk.major = dl->major;
3105 di->disk.minor = dl->minor;
3106 di->disk.state = 0;
3107 di->data_offset = pos;
3108 di->component_size = a->info.component_size;
3109 di->container_member = dl->pdnum;
3110 di->next = rv;
3111 rv = di;
3112 printf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3113 i, pos);
3114
3115 break;
3116 }
3117 if (!dl && ! global_ok) {
3118 /* not enough dedicated spares, try global */
3119 global_ok = 1;
3120 dl = ddf->dlist;
3121 goto again;
3122 }
3123 }
3124
3125 if (!rv)
3126 /* No spares found */
3127 return rv;
3128 /* Now 'rv' has a list of devices to return.
3129 * Create a metadata_update record to update the
3130 * phys_refnum and lba_offset values
3131 */
3132 mu = malloc(sizeof(*mu));
3133 mu->buf = malloc(ddf->conf_rec_len * 512);
3134 mu->space = malloc(sizeof(struct vcl));
3135 mu->len = ddf->conf_rec_len;
3136 mu->next = *updates;
3137 vc = find_vdcr(ddf, a->info.container_member);
3138 memcpy(mu->buf, vc, ddf->conf_rec_len * 512);
3139
3140 vc = (struct vd_config*)mu->buf;
3141 lba = (__u64*)&vc->phys_refnum[ddf->mppe];
3142 for (di = rv ; di ; di = di->next) {
3143 vc->phys_refnum[di->disk.raid_disk] =
3144 ddf->phys->entries[dl->pdnum].refnum;
3145 lba[di->disk.raid_disk] = di->data_offset;
3146 }
3147 *updates = mu;
3148 return rv;
3149 }
3150
3151 struct superswitch super_ddf = {
3152 #ifndef MDASSEMBLE
3153 .examine_super = examine_super_ddf,
3154 .brief_examine_super = brief_examine_super_ddf,
3155 .detail_super = detail_super_ddf,
3156 .brief_detail_super = brief_detail_super_ddf,
3157 .validate_geometry = validate_geometry_ddf,
3158 .write_init_super = write_init_super_ddf,
3159 #endif
3160 .match_home = match_home_ddf,
3161 .uuid_from_super= uuid_from_super_ddf,
3162 .getinfo_super = getinfo_super_ddf,
3163 .update_super = update_super_ddf,
3164
3165 .avail_size = avail_size_ddf,
3166
3167 .compare_super = compare_super_ddf,
3168
3169 .load_super = load_super_ddf,
3170 .init_super = init_super_ddf,
3171 .store_super = store_zero_ddf,
3172 .free_super = free_super_ddf,
3173 .match_metadata_desc = match_metadata_desc_ddf,
3174 .add_to_super = add_to_super_ddf,
3175 .container_content = container_content_ddf,
3176
3177 .external = 1,
3178
3179 /* for mdmon */
3180 .open_new = ddf_open_new,
3181 .set_array_state= ddf_set_array_state,
3182 .set_disk = ddf_set_disk,
3183 .sync_metadata = ddf_sync_metadata,
3184 .process_update = ddf_process_update,
3185 .prepare_update = ddf_prepare_update,
3186 .activate_spare = ddf_activate_spare,
3187
3188 };
Unnamed repository; edit this file 'description' to name the repository.