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


50fd16b4562f0ea25d2f72d3c9e142d01cd8a446
[thirdparty/mdadm.git] / super-ddf.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2006-2007 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neil@brown.name>
23 *
24 * Specifications for DDF takes from Common RAID DDF Specification Revision 1.2
25 * (July 28 2006). Reused by permission of SNIA.
26 */
27
28 #define HAVE_STDINT_H 1
29 #include "mdadm.h"
30 #include "mdmon.h"
31 #include "sha1.h"
32 #include <values.h>
33
34 static inline int ROUND_UP(int a, int base)
35 {
36 return ((a+base-1)/base)*base;
37 }
38
39 /* a non-official T10 name for creation GUIDs */
40 static char T10[] = "Linux-MD";
41
42 /* DDF timestamps are 1980 based, so we need to add
43 * second-in-decade-of-seventies to convert to linux timestamps.
44 * 10 years with 2 leap years.
45 */
46 #define DECADE (3600*24*(365*10+2))
47 unsigned long crc32(
48 unsigned long crc,
49 const unsigned char *buf,
50 unsigned len);
51
52 /* The DDF metadata handling.
53 * DDF metadata lives at the end of the device.
54 * The last 512 byte block provides an 'anchor' which is used to locate
55 * the rest of the metadata which usually lives immediately behind the anchor.
56 *
57 * Note:
58 * - all multibyte numeric fields are bigendian.
59 * - all strings are space padded.
60 *
61 */
62
63 /* Primary Raid Level (PRL) */
64 #define DDF_RAID0 0x00
65 #define DDF_RAID1 0x01
66 #define DDF_RAID3 0x03
67 #define DDF_RAID4 0x04
68 #define DDF_RAID5 0x05
69 #define DDF_RAID1E 0x11
70 #define DDF_JBOD 0x0f
71 #define DDF_CONCAT 0x1f
72 #define DDF_RAID5E 0x15
73 #define DDF_RAID5EE 0x25
74 #define DDF_RAID6 0x06
75
76 /* Raid Level Qualifier (RLQ) */
77 #define DDF_RAID0_SIMPLE 0x00
78 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
79 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
80 #define DDF_RAID3_0 0x00 /* parity in first extent */
81 #define DDF_RAID3_N 0x01 /* parity in last extent */
82 #define DDF_RAID4_0 0x00 /* parity in first extent */
83 #define DDF_RAID4_N 0x01 /* parity in last extent */
84 /* these apply to raid5e and raid5ee as well */
85 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
86 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
87 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
88 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
89
90 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
91 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
92
93 /* Secondary RAID Level (SRL) */
94 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
95 #define DDF_2MIRRORED 0x01
96 #define DDF_2CONCAT 0x02
97 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
98
99 /* Magic numbers */
100 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
101 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
102 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
103 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
104 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
105 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
106 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
107 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
108 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
109 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
110
111 #define DDF_GUID_LEN 24
112 #define DDF_REVISION_0 "01.00.00"
113 #define DDF_REVISION_2 "01.02.00"
114
115 struct ddf_header {
116 __u32 magic; /* DDF_HEADER_MAGIC */
117 __u32 crc;
118 char guid[DDF_GUID_LEN];
119 char revision[8]; /* 01.02.00 */
120 __u32 seq; /* starts at '1' */
121 __u32 timestamp;
122 __u8 openflag;
123 __u8 foreignflag;
124 __u8 enforcegroups;
125 __u8 pad0; /* 0xff */
126 __u8 pad1[12]; /* 12 * 0xff */
127 /* 64 bytes so far */
128 __u8 header_ext[32]; /* reserved: fill with 0xff */
129 __u64 primary_lba;
130 __u64 secondary_lba;
131 __u8 type;
132 __u8 pad2[3]; /* 0xff */
133 __u32 workspace_len; /* sectors for vendor space -
134 * at least 32768(sectors) */
135 __u64 workspace_lba;
136 __u16 max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */
137 __u16 max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */
138 __u16 max_partitions; /* i.e. max num of configuration
139 record entries per disk */
140 __u16 config_record_len; /* 1 +ROUNDUP(max_primary_element_entries
141 *12/512) */
142 __u16 max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */
143 __u8 pad3[54]; /* 0xff */
144 /* 192 bytes so far */
145 __u32 controller_section_offset;
146 __u32 controller_section_length;
147 __u32 phys_section_offset;
148 __u32 phys_section_length;
149 __u32 virt_section_offset;
150 __u32 virt_section_length;
151 __u32 config_section_offset;
152 __u32 config_section_length;
153 __u32 data_section_offset;
154 __u32 data_section_length;
155 __u32 bbm_section_offset;
156 __u32 bbm_section_length;
157 __u32 diag_space_offset;
158 __u32 diag_space_length;
159 __u32 vendor_offset;
160 __u32 vendor_length;
161 /* 256 bytes so far */
162 __u8 pad4[256]; /* 0xff */
163 };
164
165 /* type field */
166 #define DDF_HEADER_ANCHOR 0x00
167 #define DDF_HEADER_PRIMARY 0x01
168 #define DDF_HEADER_SECONDARY 0x02
169
170 /* The content of the 'controller section' - global scope */
171 struct ddf_controller_data {
172 __u32 magic; /* DDF_CONTROLLER_MAGIC */
173 __u32 crc;
174 char guid[DDF_GUID_LEN];
175 struct controller_type {
176 __u16 vendor_id;
177 __u16 device_id;
178 __u16 sub_vendor_id;
179 __u16 sub_device_id;
180 } type;
181 char product_id[16];
182 __u8 pad[8]; /* 0xff */
183 __u8 vendor_data[448];
184 };
185
186 /* The content of phys_section - global scope */
187 struct phys_disk {
188 __u32 magic; /* DDF_PHYS_RECORDS_MAGIC */
189 __u32 crc;
190 __u16 used_pdes;
191 __u16 max_pdes;
192 __u8 pad[52];
193 struct phys_disk_entry {
194 char guid[DDF_GUID_LEN];
195 __u32 refnum;
196 __u16 type;
197 __u16 state;
198 __u64 config_size; /* DDF structures must be after here */
199 char path[18]; /* another horrible structure really */
200 __u8 pad[6];
201 } entries[0];
202 };
203
204 /* phys_disk_entry.type is a bitmap - bigendian remember */
205 #define DDF_Forced_PD_GUID 1
206 #define DDF_Active_in_VD 2
207 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
208 #define DDF_Spare 8 /* overrides Global_spare */
209 #define DDF_Foreign 16
210 #define DDF_Legacy 32 /* no DDF on this device */
211
212 #define DDF_Interface_mask 0xf00
213 #define DDF_Interface_SCSI 0x100
214 #define DDF_Interface_SAS 0x200
215 #define DDF_Interface_SATA 0x300
216 #define DDF_Interface_FC 0x400
217
218 /* phys_disk_entry.state is a bigendian bitmap */
219 #define DDF_Online 1
220 #define DDF_Failed 2 /* overrides 1,4,8 */
221 #define DDF_Rebuilding 4
222 #define DDF_Transition 8
223 #define DDF_SMART 16
224 #define DDF_ReadErrors 32
225 #define DDF_Missing 64
226
227 /* The content of the virt_section global scope */
228 struct virtual_disk {
229 __u32 magic; /* DDF_VIRT_RECORDS_MAGIC */
230 __u32 crc;
231 __u16 populated_vdes;
232 __u16 max_vdes;
233 __u8 pad[52];
234 struct virtual_entry {
235 char guid[DDF_GUID_LEN];
236 __u16 unit;
237 __u16 pad0; /* 0xffff */
238 __u16 guid_crc;
239 __u16 type;
240 __u8 state;
241 __u8 init_state;
242 __u8 pad1[14];
243 char name[16];
244 } entries[0];
245 };
246
247 /* virtual_entry.type is a bitmap - bigendian */
248 #define DDF_Shared 1
249 #define DDF_Enforce_Groups 2
250 #define DDF_Unicode 4
251 #define DDF_Owner_Valid 8
252
253 /* virtual_entry.state is a bigendian bitmap */
254 #define DDF_state_mask 0x7
255 #define DDF_state_optimal 0x0
256 #define DDF_state_degraded 0x1
257 #define DDF_state_deleted 0x2
258 #define DDF_state_missing 0x3
259 #define DDF_state_failed 0x4
260 #define DDF_state_part_optimal 0x5
261
262 #define DDF_state_morphing 0x8
263 #define DDF_state_inconsistent 0x10
264
265 /* virtual_entry.init_state is a bigendian bitmap */
266 #define DDF_initstate_mask 0x03
267 #define DDF_init_not 0x00
268 #define DDF_init_quick 0x01 /* initialisation is progress.
269 * i.e. 'state_inconsistent' */
270 #define DDF_init_full 0x02
271
272 #define DDF_access_mask 0xc0
273 #define DDF_access_rw 0x00
274 #define DDF_access_ro 0x80
275 #define DDF_access_blocked 0xc0
276
277 /* The content of the config_section - local scope
278 * It has multiple records each config_record_len sectors
279 * They can be vd_config or spare_assign
280 */
281
282 struct vd_config {
283 __u32 magic; /* DDF_VD_CONF_MAGIC */
284 __u32 crc;
285 char guid[DDF_GUID_LEN];
286 __u32 timestamp;
287 __u32 seqnum;
288 __u8 pad0[24];
289 __u16 prim_elmnt_count;
290 __u8 chunk_shift; /* 0 == 512, 1==1024 etc */
291 __u8 prl;
292 __u8 rlq;
293 __u8 sec_elmnt_count;
294 __u8 sec_elmnt_seq;
295 __u8 srl;
296 __u64 blocks; /* blocks per component could be different
297 * on different component devices...(only
298 * for concat I hope) */
299 __u64 array_blocks; /* blocks in array */
300 __u8 pad1[8];
301 __u32 spare_refs[8];
302 __u8 cache_pol[8];
303 __u8 bg_rate;
304 __u8 pad2[3];
305 __u8 pad3[52];
306 __u8 pad4[192];
307 __u8 v0[32]; /* reserved- 0xff */
308 __u8 v1[32]; /* reserved- 0xff */
309 __u8 v2[16]; /* reserved- 0xff */
310 __u8 v3[16]; /* reserved- 0xff */
311 __u8 vendor[32];
312 __u32 phys_refnum[0]; /* refnum of each disk in sequence */
313 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
314 bvd are always the same size */
315 };
316
317 /* vd_config.cache_pol[7] is a bitmap */
318 #define DDF_cache_writeback 1 /* else writethrough */
319 #define DDF_cache_wadaptive 2 /* only applies if writeback */
320 #define DDF_cache_readahead 4
321 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
322 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
323 #define DDF_cache_wallowed 32 /* enable write caching */
324 #define DDF_cache_rallowed 64 /* enable read caching */
325
326 struct spare_assign {
327 __u32 magic; /* DDF_SPARE_ASSIGN_MAGIC */
328 __u32 crc;
329 __u32 timestamp;
330 __u8 reserved[7];
331 __u8 type;
332 __u16 populated; /* SAEs used */
333 __u16 max; /* max SAEs */
334 __u8 pad[8];
335 struct spare_assign_entry {
336 char guid[DDF_GUID_LEN];
337 __u16 secondary_element;
338 __u8 pad[6];
339 } spare_ents[0];
340 };
341 /* spare_assign.type is a bitmap */
342 #define DDF_spare_dedicated 0x1 /* else global */
343 #define DDF_spare_revertible 0x2 /* else committable */
344 #define DDF_spare_active 0x4 /* else not active */
345 #define DDF_spare_affinity 0x8 /* enclosure affinity */
346
347 /* The data_section contents - local scope */
348 struct disk_data {
349 __u32 magic; /* DDF_PHYS_DATA_MAGIC */
350 __u32 crc;
351 char guid[DDF_GUID_LEN];
352 __u32 refnum; /* crc of some magic drive data ... */
353 __u8 forced_ref; /* set when above was not result of magic */
354 __u8 forced_guid; /* set if guid was forced rather than magic */
355 __u8 vendor[32];
356 __u8 pad[442];
357 };
358
359 /* bbm_section content */
360 struct bad_block_log {
361 __u32 magic;
362 __u32 crc;
363 __u16 entry_count;
364 __u32 spare_count;
365 __u8 pad[10];
366 __u64 first_spare;
367 struct mapped_block {
368 __u64 defective_start;
369 __u32 replacement_start;
370 __u16 remap_count;
371 __u8 pad[2];
372 } entries[0];
373 };
374
375 /* Struct for internally holding ddf structures */
376 /* The DDF structure stored on each device is potentially
377 * quite different, as some data is global and some is local.
378 * The global data is:
379 * - ddf header
380 * - controller_data
381 * - Physical disk records
382 * - Virtual disk records
383 * The local data is:
384 * - Configuration records
385 * - Physical Disk data section
386 * ( and Bad block and vendor which I don't care about yet).
387 *
388 * The local data is parsed into separate lists as it is read
389 * and reconstructed for writing. This means that we only need
390 * to make config changes once and they are automatically
391 * propagated to all devices.
392 * Note that the ddf_super has space of the conf and disk data
393 * for this disk and also for a list of all such data.
394 * The list is only used for the superblock that is being
395 * built in Create or Assemble to describe the whole array.
396 */
397 struct ddf_super {
398 struct ddf_header anchor, primary, secondary, *active;
399 struct ddf_controller_data controller;
400 struct phys_disk *phys;
401 struct virtual_disk *virt;
402 int pdsize, vdsize;
403 int max_part, mppe, conf_rec_len;
404 struct vcl {
405 struct vcl *next;
406 __u64 *lba_offset; /* location in 'conf' of
407 * the lba table */
408 int vcnum; /* index into ->virt */
409 __u64 *block_sizes; /* NULL if all the same */
410 struct vd_config conf;
411 } *conflist, *newconf;
412 struct dl {
413 struct dl *next;
414 struct disk_data disk;
415 int major, minor;
416 char *devname;
417 int fd;
418 int pdnum; /* index in ->phys */
419 struct spare_assign *spare;
420 struct vcl *vlist[0]; /* max_part in size */
421 } *dlist;
422 };
423
424 #ifndef offsetof
425 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
426 #endif
427
428 static struct superswitch super_ddf_container, super_ddf_bvd, super_ddf_svd;
429
430 static int calc_crc(void *buf, int len)
431 {
432 /* crcs are always at the same place as in the ddf_header */
433 struct ddf_header *ddf = buf;
434 __u32 oldcrc = ddf->crc;
435 __u32 newcrc;
436 ddf->crc = 0xffffffff;
437
438 newcrc = crc32(0, buf, len);
439 ddf->crc = oldcrc;
440 return newcrc;
441 }
442
443 static int load_ddf_header(int fd, unsigned long long lba,
444 unsigned long long size,
445 int type,
446 struct ddf_header *hdr, struct ddf_header *anchor)
447 {
448 /* read a ddf header (primary or secondary) from fd/lba
449 * and check that it is consistent with anchor
450 * Need to check:
451 * magic, crc, guid, rev, and LBA's header_type, and
452 * everything after header_type must be the same
453 */
454 if (lba >= size-1)
455 return 0;
456
457 if (lseek64(fd, lba<<9, 0) < 0)
458 return 0;
459
460 if (read(fd, hdr, 512) != 512)
461 return 0;
462
463 if (hdr->magic != DDF_HEADER_MAGIC)
464 return 0;
465 if (calc_crc(hdr, 512) != hdr->crc)
466 return 0;
467 if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
468 memcmp(anchor->revision, hdr->revision, 8) != 0 ||
469 anchor->primary_lba != hdr->primary_lba ||
470 anchor->secondary_lba != hdr->secondary_lba ||
471 hdr->type != type ||
472 memcmp(anchor->pad2, hdr->pad2, 512 -
473 offsetof(struct ddf_header, pad2)) != 0)
474 return 0;
475
476 /* Looks good enough to me... */
477 return 1;
478 }
479
480 static void *load_section(int fd, struct ddf_super *super, void *buf,
481 __u32 offset_be, __u32 len_be, int check)
482 {
483 unsigned long long offset = __be32_to_cpu(offset_be);
484 unsigned long long len = __be32_to_cpu(len_be);
485 int dofree = (buf == NULL);
486
487 if (check)
488 if (len != 2 && len != 8 && len != 32
489 && len != 128 && len != 512)
490 return NULL;
491
492 if (len > 1024)
493 return NULL;
494 if (buf) {
495 /* All pre-allocated sections are a single block */
496 if (len != 1)
497 return NULL;
498 } else
499 buf = malloc(len<<9);
500 if (!buf)
501 return NULL;
502
503 if (super->active->type == 1)
504 offset += __be64_to_cpu(super->active->primary_lba);
505 else
506 offset += __be64_to_cpu(super->active->secondary_lba);
507
508 if (lseek64(fd, offset<<9, 0) != (offset<<9)) {
509 if (dofree)
510 free(buf);
511 return NULL;
512 }
513 if (read(fd, buf, len<<9) != (len<<9)) {
514 if (dofree)
515 free(buf);
516 return NULL;
517 }
518 return buf;
519 }
520
521 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
522 {
523 unsigned long long dsize;
524
525 get_dev_size(fd, NULL, &dsize);
526
527 if (lseek64(fd, dsize-512, 0) < 0) {
528 if (devname)
529 fprintf(stderr,
530 Name": Cannot seek to anchor block on %s: %s\n",
531 devname, strerror(errno));
532 return 1;
533 }
534 if (read(fd, &super->anchor, 512) != 512) {
535 if (devname)
536 fprintf(stderr,
537 Name ": Cannot read anchor block on %s: %s\n",
538 devname, strerror(errno));
539 return 1;
540 }
541 if (super->anchor.magic != DDF_HEADER_MAGIC) {
542 if (devname)
543 fprintf(stderr, Name ": no DDF anchor found on %s\n",
544 devname);
545 return 2;
546 }
547 if (calc_crc(&super->anchor, 512) != super->anchor.crc) {
548 if (devname)
549 fprintf(stderr, Name ": bad CRC on anchor on %s\n",
550 devname);
551 return 2;
552 }
553 if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
554 memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
555 if (devname)
556 fprintf(stderr, Name ": can only support super revision"
557 " %.8s and earlier, not %.8s on %s\n",
558 DDF_REVISION_2, super->anchor.revision,devname);
559 return 2;
560 }
561 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba),
562 dsize >> 9, 1,
563 &super->primary, &super->anchor) == 0) {
564 if (devname)
565 fprintf(stderr,
566 Name ": Failed to load primary DDF header "
567 "on %s\n", devname);
568 return 2;
569 }
570 super->active = &super->primary;
571 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.secondary_lba),
572 dsize >> 9, 2,
573 &super->secondary, &super->anchor)) {
574 if ((__be32_to_cpu(super->primary.seq)
575 < __be32_to_cpu(super->secondary.seq) &&
576 !super->secondary.openflag)
577 || (__be32_to_cpu(super->primary.seq)
578 == __be32_to_cpu(super->secondary.seq) &&
579 super->primary.openflag && !super->secondary.openflag)
580 )
581 super->active = &super->secondary;
582 }
583 return 0;
584 }
585
586 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
587 {
588 void *ok;
589 ok = load_section(fd, super, &super->controller,
590 super->active->controller_section_offset,
591 super->active->controller_section_length,
592 0);
593 super->phys = load_section(fd, super, NULL,
594 super->active->phys_section_offset,
595 super->active->phys_section_length,
596 1);
597 super->pdsize = __be32_to_cpu(super->active->phys_section_length) * 512;
598
599 super->virt = load_section(fd, super, NULL,
600 super->active->virt_section_offset,
601 super->active->virt_section_length,
602 1);
603 super->vdsize = __be32_to_cpu(super->active->virt_section_length) * 512;
604 if (!ok ||
605 !super->phys ||
606 !super->virt) {
607 free(super->phys);
608 free(super->virt);
609 super->phys = NULL;
610 super->virt = NULL;
611 return 2;
612 }
613 super->conflist = NULL;
614 super->dlist = NULL;
615
616 super->max_part = __be16_to_cpu(super->active->max_partitions);
617 super->mppe = __be16_to_cpu(super->active->max_primary_element_entries);
618 super->conf_rec_len = __be16_to_cpu(super->active->config_record_len);
619 return 0;
620 }
621
622 static int load_ddf_local(int fd, struct ddf_super *super,
623 char *devname, int keep)
624 {
625 struct dl *dl;
626 struct stat stb;
627 char *conf;
628 int i;
629 int vnum;
630 int max_virt_disks = __be16_to_cpu(super->active->max_vd_entries);
631
632 /* First the local disk info */
633 dl = malloc(sizeof(*dl) +
634 (super->max_part) * sizeof(dl->vlist[0]));
635
636 load_section(fd, super, &dl->disk,
637 super->active->data_section_offset,
638 super->active->data_section_length,
639 0);
640 dl->devname = devname ? strdup(devname) : NULL;
641
642 fstat(fd, &stb);
643 dl->major = major(stb.st_rdev);
644 dl->minor = minor(stb.st_rdev);
645 dl->next = super->dlist;
646 dl->fd = keep ? fd : -1;
647 dl->spare = NULL;
648 for (i=0 ; i < super->max_part ; i++)
649 dl->vlist[i] = NULL;
650 super->dlist = dl;
651 dl->pdnum = -1;
652 for (i=0; i < __be16_to_cpu(super->active->max_pd_entries); i++)
653 if (memcmp(super->phys->entries[i].guid,
654 dl->disk.guid, DDF_GUID_LEN) == 0)
655 dl->pdnum = i;
656
657 /* Now the config list. */
658 /* 'conf' is an array of config entries, some of which are
659 * probably invalid. Those which are good need to be copied into
660 * the conflist
661 */
662
663 conf = load_section(fd, super, NULL,
664 super->active->config_section_offset,
665 super->active->config_section_length,
666 0);
667
668 vnum = 0;
669 for (i = 0;
670 i < __be32_to_cpu(super->active->config_section_length);
671 i += super->conf_rec_len) {
672 struct vd_config *vd =
673 (struct vd_config *)((char*)conf + i*512);
674 struct vcl *vcl;
675
676 if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) {
677 if (dl->spare)
678 continue;
679 dl->spare = malloc(super->conf_rec_len*512);
680 memcpy(dl->spare, vd, super->conf_rec_len*512);
681 continue;
682 }
683 if (vd->magic != DDF_VD_CONF_MAGIC)
684 continue;
685 for (vcl = super->conflist; vcl; vcl = vcl->next) {
686 if (memcmp(vcl->conf.guid,
687 vd->guid, DDF_GUID_LEN) == 0)
688 break;
689 }
690
691 if (vcl) {
692 dl->vlist[vnum++] = vcl;
693 if (__be32_to_cpu(vd->seqnum) <=
694 __be32_to_cpu(vcl->conf.seqnum))
695 continue;
696 } else {
697 vcl = malloc(super->conf_rec_len*512 +
698 offsetof(struct vcl, conf));
699 vcl->next = super->conflist;
700 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
701 super->conflist = vcl;
702 dl->vlist[vnum++] = vcl;
703 }
704 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
705 vcl->lba_offset = (__u64*)
706 &vcl->conf.phys_refnum[super->mppe];
707
708 for (i=0; i < max_virt_disks ; i++)
709 if (memcmp(super->virt->entries[i].guid,
710 vcl->conf.guid, DDF_GUID_LEN)==0)
711 break;
712 if (i < max_virt_disks)
713 vcl->vcnum = i;
714 }
715 free(conf);
716
717 return 0;
718 }
719
720 #ifndef MDASSEMBLE
721 static int load_super_ddf_all(struct supertype *st, int fd,
722 void **sbp, char *devname, int keep_fd);
723 #endif
724 static int load_super_ddf(struct supertype *st, int fd,
725 char *devname)
726 {
727 unsigned long long dsize;
728 struct ddf_super *super;
729 int rv;
730
731 #ifndef MDASSEMBLE
732 /* if 'fd' is a container, load metadata from all the devices */
733 if (load_super_ddf_all(st, fd, &st->sb, devname, 1) == 0)
734 return 0;
735 #endif
736 if (st->subarray[0])
737 return 1; /* FIXME Is this correct */
738
739 if (get_dev_size(fd, devname, &dsize) == 0)
740 return 1;
741
742 /* 32M is a lower bound */
743 if (dsize <= 32*1024*1024) {
744 if (devname) {
745 fprintf(stderr,
746 Name ": %s is too small for ddf: "
747 "size is %llu sectors.\n",
748 devname, dsize>>9);
749 return 1;
750 }
751 }
752 if (dsize & 511) {
753 if (devname) {
754 fprintf(stderr,
755 Name ": %s is an odd size for ddf: "
756 "size is %llu bytes.\n",
757 devname, dsize);
758 return 1;
759 }
760 }
761
762 super = malloc(sizeof(*super));
763 if (!super) {
764 fprintf(stderr, Name ": malloc of %zu failed.\n",
765 sizeof(*super));
766 return 1;
767 }
768 memset(super, 0, sizeof(*super));
769
770 rv = load_ddf_headers(fd, super, devname);
771 if (rv) {
772 free(super);
773 return rv;
774 }
775
776 /* Have valid headers and have chosen the best. Let's read in the rest*/
777
778 rv = load_ddf_global(fd, super, devname);
779
780 if (rv) {
781 if (devname)
782 fprintf(stderr,
783 Name ": Failed to load all information "
784 "sections on %s\n", devname);
785 free(super);
786 return rv;
787 }
788
789 load_ddf_local(fd, super, devname, 0);
790
791 /* Should possibly check the sections .... */
792
793 st->sb = super;
794 if (st->ss == NULL) {
795 st->ss = &super_ddf;
796 st->minor_version = 0;
797 st->max_devs = 512;
798 }
799 return 0;
800
801 }
802
803 static void free_super_ddf(struct supertype *st)
804 {
805 struct ddf_super *ddf = st->sb;
806 if (ddf == NULL)
807 return;
808 free(ddf->phys);
809 free(ddf->virt);
810 while (ddf->conflist) {
811 struct vcl *v = ddf->conflist;
812 ddf->conflist = v->next;
813 if (v->block_sizes)
814 free(v->block_sizes);
815 free(v);
816 }
817 while (ddf->dlist) {
818 struct dl *d = ddf->dlist;
819 ddf->dlist = d->next;
820 if (d->fd >= 0)
821 close(d->fd);
822 if (d->spare)
823 free(d->spare);
824 free(d);
825 }
826 free(ddf);
827 st->sb = NULL;
828 }
829
830 static struct supertype *match_metadata_desc_ddf(char *arg)
831 {
832 /* 'ddf' only support containers */
833 struct supertype *st;
834 if (strcmp(arg, "ddf") != 0 &&
835 strcmp(arg, "default") != 0
836 )
837 return NULL;
838
839 st = malloc(sizeof(*st));
840 memset(st, 0, sizeof(*st));
841 st->ss = &super_ddf;
842 st->max_devs = 512;
843 st->minor_version = 0;
844 st->sb = NULL;
845 return st;
846 }
847
848 static struct supertype *match_metadata_desc_ddf_bvd(char *arg)
849 {
850 struct supertype *st;
851 if (strcmp(arg, "ddf/bvd") != 0 &&
852 strcmp(arg, "bvd") != 0 &&
853 strcmp(arg, "default") != 0
854 )
855 return NULL;
856
857 st = malloc(sizeof(*st));
858 memset(st, 0, sizeof(*st));
859 st->ss = &super_ddf_bvd;
860 st->max_devs = 512;
861 st->minor_version = 0;
862 st->sb = NULL;
863 return st;
864 }
865 static struct supertype *match_metadata_desc_ddf_svd(char *arg)
866 {
867 struct supertype *st;
868 if (strcmp(arg, "ddf/svd") != 0 &&
869 strcmp(arg, "svd") != 0 &&
870 strcmp(arg, "default") != 0
871 )
872 return NULL;
873
874 st = malloc(sizeof(*st));
875 memset(st, 0, sizeof(*st));
876 st->ss = &super_ddf_svd;
877 st->max_devs = 512;
878 st->minor_version = 0;
879 st->sb = NULL;
880 return st;
881 }
882
883 #ifndef MDASSEMBLE
884
885 static mapping_t ddf_state[] = {
886 { "Optimal", 0},
887 { "Degraded", 1},
888 { "Deleted", 2},
889 { "Missing", 3},
890 { "Failed", 4},
891 { "Partially Optimal", 5},
892 { "-reserved-", 6},
893 { "-reserved-", 7},
894 { NULL, 0}
895 };
896
897 static mapping_t ddf_init_state[] = {
898 { "Not Initialised", 0},
899 { "QuickInit in Progress", 1},
900 { "Fully Initialised", 2},
901 { "*UNKNOWN*", 3},
902 { NULL, 0}
903 };
904 static mapping_t ddf_access[] = {
905 { "Read/Write", 0},
906 { "Reserved", 1},
907 { "Read Only", 2},
908 { "Blocked (no access)", 3},
909 { NULL ,0}
910 };
911
912 static mapping_t ddf_level[] = {
913 { "RAID0", DDF_RAID0},
914 { "RAID1", DDF_RAID1},
915 { "RAID3", DDF_RAID3},
916 { "RAID4", DDF_RAID4},
917 { "RAID5", DDF_RAID5},
918 { "RAID1E",DDF_RAID1E},
919 { "JBOD", DDF_JBOD},
920 { "CONCAT",DDF_CONCAT},
921 { "RAID5E",DDF_RAID5E},
922 { "RAID5EE",DDF_RAID5EE},
923 { "RAID6", DDF_RAID6},
924 { NULL, 0}
925 };
926 static mapping_t ddf_sec_level[] = {
927 { "Striped", DDF_2STRIPED},
928 { "Mirrored", DDF_2MIRRORED},
929 { "Concat", DDF_2CONCAT},
930 { "Spanned", DDF_2SPANNED},
931 { NULL, 0}
932 };
933 #endif
934
935 struct num_mapping {
936 int num1, num2;
937 };
938 static struct num_mapping ddf_level_num[] = {
939 { DDF_RAID0, 0 },
940 { DDF_RAID1, 1 },
941 { DDF_RAID3, LEVEL_UNSUPPORTED },
942 { DDF_RAID4, 4 },
943 { DDF_RAID5, 5 },
944 { DDF_RAID1E, LEVEL_UNSUPPORTED },
945 { DDF_JBOD, LEVEL_UNSUPPORTED },
946 { DDF_CONCAT, LEVEL_LINEAR },
947 { DDF_RAID5E, LEVEL_UNSUPPORTED },
948 { DDF_RAID5EE, LEVEL_UNSUPPORTED },
949 { DDF_RAID6, 6},
950 { MAXINT, MAXINT }
951 };
952
953 static int map_num1(struct num_mapping *map, int num)
954 {
955 int i;
956 for (i=0 ; map[i].num1 != MAXINT; i++)
957 if (map[i].num1 == num)
958 break;
959 return map[i].num2;
960 }
961
962 #ifndef MDASSEMBLE
963 static void print_guid(char *guid, int tstamp)
964 {
965 /* A GUIDs are part (or all) ASCII and part binary.
966 * They tend to be space padded.
967 * We print the GUID in HEX, then in parentheses add
968 * any initial ASCII sequence, and a possible
969 * time stamp from bytes 16-19
970 */
971 int l = DDF_GUID_LEN;
972 int i;
973
974 for (i=0 ; i<DDF_GUID_LEN ; i++) {
975 if ((i&3)==0 && i != 0) printf(":");
976 printf("%02X", guid[i]&255);
977 }
978
979 printf(" (");
980 while (l && guid[l-1] == ' ')
981 l--;
982 for (i=0 ; i<l ; i++) {
983 if (guid[i] >= 0x20 && guid[i] < 0x7f)
984 fputc(guid[i], stdout);
985 else
986 break;
987 }
988 if (tstamp) {
989 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
990 char tbuf[100];
991 struct tm *tm;
992 tm = localtime(&then);
993 strftime(tbuf, 100, " %D %T",tm);
994 fputs(tbuf, stdout);
995 }
996 printf(")");
997 }
998
999 static void examine_vd(int n, struct ddf_super *sb, char *guid)
1000 {
1001 int crl = sb->conf_rec_len;
1002 struct vcl *vcl;
1003
1004 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
1005 struct vd_config *vc = &vcl->conf;
1006
1007 if (calc_crc(vc, crl*512) != vc->crc)
1008 continue;
1009 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
1010 continue;
1011
1012 /* Ok, we know about this VD, let's give more details */
1013 printf(" Raid Devices[%d] : %d\n", n,
1014 __be16_to_cpu(vc->prim_elmnt_count));
1015 printf(" Chunk Size[%d] : %d sectors\n", n,
1016 1 << vc->chunk_shift);
1017 printf(" Raid Level[%d] : %s\n", n,
1018 map_num(ddf_level, vc->prl)?:"-unknown-");
1019 if (vc->sec_elmnt_count != 1) {
1020 printf(" Secondary Position[%d] : %d of %d\n", n,
1021 vc->sec_elmnt_seq, vc->sec_elmnt_count);
1022 printf(" Secondary Level[%d] : %s\n", n,
1023 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1024 }
1025 printf(" Device Size[%d] : %llu\n", n,
1026 __be64_to_cpu(vc->blocks)/2);
1027 printf(" Array Size[%d] : %llu\n", n,
1028 __be64_to_cpu(vc->array_blocks)/2);
1029 }
1030 }
1031
1032 static void examine_vds(struct ddf_super *sb)
1033 {
1034 int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1035 int i;
1036 printf(" Virtual Disks : %d\n", cnt);
1037
1038 for (i=0; i<cnt; i++) {
1039 struct virtual_entry *ve = &sb->virt->entries[i];
1040 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1041 printf("\n");
1042 printf(" unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1043 printf(" state[%d] : %s, %s%s\n", i,
1044 map_num(ddf_state, ve->state & 7),
1045 (ve->state & 8) ? "Morphing, ": "",
1046 (ve->state & 16)? "Not Consistent" : "Consistent");
1047 printf(" init state[%d] : %s\n", i,
1048 map_num(ddf_init_state, ve->init_state&3));
1049 printf(" access[%d] : %s\n", i,
1050 map_num(ddf_access, (ve->init_state>>6) & 3));
1051 printf(" Name[%d] : %.16s\n", i, ve->name);
1052 examine_vd(i, sb, ve->guid);
1053 }
1054 if (cnt) printf("\n");
1055 }
1056
1057 static void examine_pds(struct ddf_super *sb)
1058 {
1059 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1060 int i;
1061 struct dl *dl;
1062 printf(" Physical Disks : %d\n", cnt);
1063
1064 for (i=0 ; i<cnt ; i++) {
1065 struct phys_disk_entry *pd = &sb->phys->entries[i];
1066 int type = __be16_to_cpu(pd->type);
1067 int state = __be16_to_cpu(pd->state);
1068
1069 printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1070 printf("\n");
1071 printf(" ref[%d] : %08x\n", i,
1072 __be32_to_cpu(pd->refnum));
1073 printf(" mode[%d] : %s%s%s%s%s\n", i,
1074 (type&2) ? "active":"",
1075 (type&4) ? "Global Spare":"",
1076 (type&8) ? "spare" : "",
1077 (type&16)? ", foreign" : "",
1078 (type&32)? "pass-through" : "");
1079 printf(" state[%d] : %s%s%s%s%s%s%s\n", i,
1080 (state&1)? "Online": "Offline",
1081 (state&2)? ", Failed": "",
1082 (state&4)? ", Rebuilding": "",
1083 (state&8)? ", in-transition": "",
1084 (state&16)? ", SMART errors": "",
1085 (state&32)? ", Unrecovered Read Errors": "",
1086 (state&64)? ", Missing" : "");
1087 printf(" Avail Size[%d] : %llu K\n", i,
1088 __be64_to_cpu(pd->config_size)>>1);
1089 for (dl = sb->dlist; dl ; dl = dl->next) {
1090 if (dl->disk.refnum == pd->refnum) {
1091 char *dv = map_dev(dl->major, dl->minor, 0);
1092 if (dv)
1093 printf(" Device[%d] : %s\n",
1094 i, dv);
1095 }
1096 }
1097 printf("\n");
1098 }
1099 }
1100
1101 static void examine_super_ddf(struct supertype *st, char *homehost)
1102 {
1103 struct ddf_super *sb = st->sb;
1104
1105 printf(" Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1106 printf(" Version : %.8s\n", sb->anchor.revision);
1107 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1108 printf("\n");
1109 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1110 printf("\n");
1111 printf(" Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1112 printf(" Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1113 ?"yes" : "no");
1114 examine_vds(sb);
1115 examine_pds(sb);
1116 }
1117
1118 static void brief_examine_super_ddf(struct supertype *st)
1119 {
1120 /* We just write a generic DDF ARRAY entry
1121 * The uuid is all hex, 6 groups of 4 bytes
1122 */
1123 struct ddf_super *ddf = st->sb;
1124 int i;
1125 printf("ARRAY /dev/ddf metadata=ddf UUID=");
1126 for (i = 0; i < DDF_GUID_LEN; i++) {
1127 if ((i&3) == 0 && i != 0)
1128 printf(":");
1129 printf("%02X", 255&ddf->anchor.guid[i]);
1130 }
1131 printf("\n");
1132 }
1133
1134 static void detail_super_ddf(struct supertype *st, char *homehost)
1135 {
1136 /* FIXME later
1137 * Could print DDF GUID
1138 * Need to find which array
1139 * If whole, briefly list all arrays
1140 * If one, give name
1141 */
1142 }
1143
1144 static void brief_detail_super_ddf(struct supertype *st)
1145 {
1146 /* FIXME I really need to know which array we are detailing.
1147 * Can that be stored in ddf_super??
1148 */
1149 // struct ddf_super *ddf = st->sb;
1150 }
1151 #endif
1152
1153 static int match_home_ddf(struct supertype *st, char *homehost)
1154 {
1155 /* It matches 'this' host if the controller is a
1156 * Linux-MD controller with vendor_data matching
1157 * the hostname
1158 */
1159 struct ddf_super *ddf = st->sb;
1160 int len = strlen(homehost);
1161
1162 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1163 len < sizeof(ddf->controller.vendor_data) &&
1164 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1165 ddf->controller.vendor_data[len] == 0);
1166 }
1167
1168 static struct vd_config *find_vdcr(struct ddf_super *ddf, int inst)
1169 {
1170 struct vcl *v;
1171
1172 for (v = ddf->conflist; v; v = v->next)
1173 if (inst == v->vcnum)
1174 return &v->conf;
1175 return NULL;
1176 }
1177
1178 static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1179 {
1180 /* Find the entry in phys_disk which has the given refnum
1181 * and return it's index
1182 */
1183 int i;
1184 for (i=0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1185 if (ddf->phys->entries[i].refnum == phys_refnum)
1186 return i;
1187 return -1;
1188 }
1189
1190 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1191 {
1192 /* The uuid returned here is used for:
1193 * uuid to put into bitmap file (Create, Grow)
1194 * uuid for backup header when saving critical section (Grow)
1195 * comparing uuids when re-adding a device into an array
1196 * For each of these we can make do with a truncated
1197 * or hashed uuid rather than the original, as long as
1198 * everyone agrees.
1199 * In each case the uuid required is that of the data-array,
1200 * not the device-set.
1201 * In the case of SVD we assume the BVD is of interest,
1202 * though that might be the case if a bitmap were made for
1203 * a mirrored SVD - worry about that later.
1204 * So we need to find the VD configuration record for the
1205 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1206 * The first 16 bytes of the sha1 of these is used.
1207 */
1208 struct ddf_super *ddf = st->sb;
1209 struct vd_config *vd = &ddf->newconf->conf;
1210
1211 if (!vd)
1212 memset(uuid, 0, sizeof (uuid));
1213 else {
1214 char buf[20];
1215 struct sha1_ctx ctx;
1216 sha1_init_ctx(&ctx);
1217 sha1_process_bytes(&vd->guid, DDF_GUID_LEN, &ctx);
1218 if (vd->sec_elmnt_count > 1)
1219 sha1_process_bytes(&vd->sec_elmnt_seq, 1, &ctx);
1220 sha1_finish_ctx(&ctx, buf);
1221 memcpy(uuid, buf, sizeof(uuid));
1222 }
1223 }
1224
1225 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info)
1226 {
1227 struct ddf_super *ddf = st->sb;
1228
1229 info->array.raid_disks = __be16_to_cpu(ddf->phys->used_pdes);
1230 info->array.level = LEVEL_CONTAINER;
1231 info->array.layout = 0;
1232 info->array.md_minor = -1;
1233 info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)
1234 (ddf->anchor.guid+16));
1235 info->array.utime = 0;
1236 info->array.chunk_size = 0;
1237
1238 // info->data_offset = ???;
1239 // info->component_size = ???;
1240
1241 info->disk.major = 0;
1242 info->disk.minor = 0;
1243 if (ddf->dlist) {
1244 info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1245 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1246 } else {
1247 info->disk.number = -1;
1248 // info->disk.raid_disk = find refnum in the table and use index;
1249 }
1250 info->disk.state = (1 << MD_DISK_SYNC);
1251
1252 info->reshape_active = 0;
1253
1254 strcpy(info->text_version, "ddf");
1255
1256 // uuid_from_super_ddf(info->uuid, sbv);
1257
1258 // info->name[] ?? ;
1259 }
1260
1261 static void getinfo_super_n_container(struct supertype *st, struct mdinfo *info)
1262 {
1263 /* just need offset and size */
1264 struct ddf_super *ddf = st->sb;
1265 int n = info->disk.number;
1266
1267 info->data_offset = __be64_to_cpu(ddf->phys->entries[n].config_size);
1268 info->component_size = 32*1024*1024 / 512;
1269 }
1270
1271 static int rlq_to_layout(int rlq, int prl, int raiddisks);
1272
1273 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info)
1274 {
1275 struct ddf_super *ddf = st->sb;
1276 struct vd_config *vd = find_vdcr(ddf, info->container_member);
1277
1278 /* FIXME this returns BVD info - what if we want SVD ?? */
1279
1280 info->array.raid_disks = __be16_to_cpu(vd->prim_elmnt_count);
1281 info->array.level = map_num1(ddf_level_num, vd->prl);
1282 info->array.layout = rlq_to_layout(vd->rlq, vd->prl,
1283 info->array.raid_disks);
1284 info->array.md_minor = -1;
1285 info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)(vd->guid+16));
1286 info->array.utime = DECADE + __be32_to_cpu(vd->timestamp);
1287 info->array.chunk_size = 512 << vd->chunk_shift;
1288
1289 // info->data_offset = ???;
1290 // info->component_size = ???;
1291
1292 info->disk.major = 0;
1293 info->disk.minor = 0;
1294 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1295 // info->disk.raid_disk = find refnum in the table and use index;
1296 // info->disk.state = ???;
1297
1298 info->resync_start = 0;
1299 if (!(ddf->virt->entries[info->container_member].state
1300 & DDF_state_inconsistent) &&
1301 (ddf->virt->entries[info->container_member].init_state
1302 & DDF_initstate_mask)
1303 == DDF_init_full)
1304 info->resync_start = ~0ULL;
1305
1306 uuid_from_super_ddf(st, info->uuid);
1307
1308 sprintf(info->text_version, "/%s/%d",
1309 devnum2devname(st->container_dev),
1310 info->container_member);
1311
1312 // info->name[] ?? ;
1313 }
1314
1315 static void getinfo_super_n_bvd(struct supertype *st, struct mdinfo *info)
1316 {
1317 /* Find the particular details for info->disk.raid_disk.
1318 * This includes data_offset, component_size,
1319 */
1320 struct ddf_super *ddf = st->sb;
1321 __u64 *lba_offset = ddf->newconf->lba_offset;
1322 struct vd_config *conf = &ddf->newconf->conf;
1323 info->data_offset = __be64_to_cpu(lba_offset[info->disk.raid_disk]);
1324 info->component_size = __be64_to_cpu(conf->blocks);
1325 }
1326
1327 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1328 char *update,
1329 char *devname, int verbose,
1330 int uuid_set, char *homehost)
1331 {
1332 /* For 'assemble' and 'force' we need to return non-zero if any
1333 * change was made. For others, the return value is ignored.
1334 * Update options are:
1335 * force-one : This device looks a bit old but needs to be included,
1336 * update age info appropriately.
1337 * assemble: clear any 'faulty' flag to allow this device to
1338 * be assembled.
1339 * force-array: Array is degraded but being forced, mark it clean
1340 * if that will be needed to assemble it.
1341 *
1342 * newdev: not used ????
1343 * grow: Array has gained a new device - this is currently for
1344 * linear only
1345 * resync: mark as dirty so a resync will happen.
1346 * uuid: Change the uuid of the array to match what is given
1347 * homehost: update the recorded homehost
1348 * name: update the name - preserving the homehost
1349 * _reshape_progress: record new reshape_progress position.
1350 *
1351 * Following are not relevant for this version:
1352 * sparc2.2 : update from old dodgey metadata
1353 * super-minor: change the preferred_minor number
1354 * summaries: update redundant counters.
1355 */
1356 int rv = 0;
1357 // struct ddf_super *ddf = st->sb;
1358 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1359 // struct virtual_entry *ve = find_ve(ddf);
1360
1361 /* we don't need to handle "force-*" or "assemble" as
1362 * there is no need to 'trick' the kernel. We the metadata is
1363 * first updated to activate the array, all the implied modifications
1364 * will just happen.
1365 */
1366
1367 if (strcmp(update, "grow") == 0) {
1368 /* FIXME */
1369 }
1370 if (strcmp(update, "resync") == 0) {
1371 // info->resync_checkpoint = 0;
1372 }
1373 /* We ignore UUID updates as they make even less sense
1374 * with DDF
1375 */
1376 if (strcmp(update, "homehost") == 0) {
1377 /* homehost is stored in controller->vendor_data,
1378 * or it is when we are the vendor
1379 */
1380 // if (info->vendor_is_local)
1381 // strcpy(ddf->controller.vendor_data, homehost);
1382 }
1383 if (strcmp(update, "name") == 0) {
1384 /* name is stored in virtual_entry->name */
1385 // memset(ve->name, ' ', 16);
1386 // strncpy(ve->name, info->name, 16);
1387 }
1388 if (strcmp(update, "_reshape_progress") == 0) {
1389 /* We don't support reshape yet */
1390 }
1391
1392 // update_all_csum(ddf);
1393
1394 return rv;
1395 }
1396
1397 static void make_header_guid(char *guid)
1398 {
1399 __u32 stamp;
1400 int rfd;
1401 /* Create a DDF Header of Virtual Disk GUID */
1402
1403 /* 24 bytes of fiction required.
1404 * first 8 are a 'vendor-id' - "Linux-MD"
1405 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1406 * Remaining 8 random number plus timestamp
1407 */
1408 memcpy(guid, T10, sizeof(T10));
1409 stamp = __cpu_to_be32(0xdeadbeef);
1410 memcpy(guid+8, &stamp, 4);
1411 stamp = __cpu_to_be32(0);
1412 memcpy(guid+12, &stamp, 4);
1413 stamp = __cpu_to_be32(time(0) - DECADE);
1414 memcpy(guid+16, &stamp, 4);
1415 rfd = open("/dev/urandom", O_RDONLY);
1416 if (rfd < 0 || read(rfd, &stamp, 4) != 4)
1417 stamp = random();
1418 memcpy(guid+20, &stamp, 4);
1419 if (rfd >= 0) close(rfd);
1420 }
1421
1422 static int init_super_ddf(struct supertype *st,
1423 mdu_array_info_t *info,
1424 unsigned long long size, char *name, char *homehost,
1425 int *uuid)
1426 {
1427 /* This is primarily called by Create when creating a new array.
1428 * We will then get add_to_super called for each component, and then
1429 * write_init_super called to write it out to each device.
1430 * For DDF, Create can create on fresh devices or on a pre-existing
1431 * array.
1432 * To create on a pre-existing array a different method will be called.
1433 * This one is just for fresh drives.
1434 *
1435 * We need to create the entire 'ddf' structure which includes:
1436 * DDF headers - these are easy.
1437 * Controller data - a Sector describing this controller .. not that
1438 * this is a controller exactly.
1439 * Physical Disk Record - one entry per device, so
1440 * leave plenty of space.
1441 * Virtual Disk Records - again, just leave plenty of space.
1442 * This just lists VDs, doesn't give details
1443 * Config records - describes the VDs that use this disk
1444 * DiskData - describes 'this' device.
1445 * BadBlockManagement - empty
1446 * Diag Space - empty
1447 * Vendor Logs - Could we put bitmaps here?
1448 *
1449 */
1450 struct ddf_super *ddf;
1451 char hostname[17];
1452 int hostlen;
1453 int max_phys_disks, max_virt_disks;
1454 unsigned long long sector;
1455 int clen;
1456 int i;
1457 int pdsize, vdsize;
1458 struct phys_disk *pd;
1459 struct virtual_disk *vd;
1460
1461 ddf = malloc(sizeof(*ddf));
1462 memset(ddf, 0, sizeof(*ddf));
1463 ddf->dlist = NULL; /* no physical disks yet */
1464 ddf->conflist = NULL; /* No virtual disks yet */
1465
1466 /* At least 32MB *must* be reserved for the ddf. So let's just
1467 * start 32MB from the end, and put the primary header there.
1468 * Don't do secondary for now.
1469 * We don't know exactly where that will be yet as it could be
1470 * different on each device. To just set up the lengths.
1471 *
1472 */
1473
1474 ddf->anchor.magic = DDF_HEADER_MAGIC;
1475 make_header_guid(ddf->anchor.guid);
1476
1477 memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
1478 ddf->anchor.seq = __cpu_to_be32(1);
1479 ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1480 ddf->anchor.openflag = 0xFF;
1481 ddf->anchor.foreignflag = 0;
1482 ddf->anchor.enforcegroups = 0; /* Is this best?? */
1483 ddf->anchor.pad0 = 0xff;
1484 memset(ddf->anchor.pad1, 0xff, 12);
1485 memset(ddf->anchor.header_ext, 0xff, 32);
1486 ddf->anchor.primary_lba = ~(__u64)0;
1487 ddf->anchor.secondary_lba = ~(__u64)0;
1488 ddf->anchor.type = DDF_HEADER_ANCHOR;
1489 memset(ddf->anchor.pad2, 0xff, 3);
1490 ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1491 ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1492 of 32M reserved.. */
1493 max_phys_disks = 1023; /* Should be enough */
1494 ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1495 max_virt_disks = 255;
1496 ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1497 ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1498 ddf->max_part = 64;
1499 ddf->mppe = 256;
1500 ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
1501 ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1502 ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe);
1503 memset(ddf->anchor.pad3, 0xff, 54);
1504 /* controller sections is one sector long immediately
1505 * after the ddf header */
1506 sector = 1;
1507 ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1508 ddf->anchor.controller_section_length = __cpu_to_be32(1);
1509 sector += 1;
1510
1511 /* phys is 8 sectors after that */
1512 pdsize = ROUND_UP(sizeof(struct phys_disk) +
1513 sizeof(struct phys_disk_entry)*max_phys_disks,
1514 512);
1515 switch(pdsize/512) {
1516 case 2: case 8: case 32: case 128: case 512: break;
1517 default: abort();
1518 }
1519 ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1520 ddf->anchor.phys_section_length =
1521 __cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1522 sector += pdsize/512;
1523
1524 /* virt is another 32 sectors */
1525 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1526 sizeof(struct virtual_entry) * max_virt_disks,
1527 512);
1528 switch(vdsize/512) {
1529 case 2: case 8: case 32: case 128: case 512: break;
1530 default: abort();
1531 }
1532 ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1533 ddf->anchor.virt_section_length =
1534 __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1535 sector += vdsize/512;
1536
1537 clen = ddf->conf_rec_len * (ddf->max_part+1);
1538 ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1539 ddf->anchor.config_section_length = __cpu_to_be32(clen);
1540 sector += clen;
1541
1542 ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1543 ddf->anchor.data_section_length = __cpu_to_be32(1);
1544 sector += 1;
1545
1546 ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1547 ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1548 ddf->anchor.diag_space_length = __cpu_to_be32(0);
1549 ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1550 ddf->anchor.vendor_length = __cpu_to_be32(0);
1551 ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1552
1553 memset(ddf->anchor.pad4, 0xff, 256);
1554
1555 memcpy(&ddf->primary, &ddf->anchor, 512);
1556 memcpy(&ddf->secondary, &ddf->anchor, 512);
1557
1558 ddf->primary.openflag = 1; /* I guess.. */
1559 ddf->primary.type = DDF_HEADER_PRIMARY;
1560
1561 ddf->secondary.openflag = 1; /* I guess.. */
1562 ddf->secondary.type = DDF_HEADER_SECONDARY;
1563
1564 ddf->active = &ddf->primary;
1565
1566 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1567
1568 /* 24 more bytes of fiction required.
1569 * first 8 are a 'vendor-id' - "Linux-MD"
1570 * Remaining 16 are serial number.... maybe a hostname would do?
1571 */
1572 memcpy(ddf->controller.guid, T10, sizeof(T10));
1573 gethostname(hostname, sizeof(hostname));
1574 hostname[sizeof(hostname) - 1] = 0;
1575 hostlen = strlen(hostname);
1576 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1577 for (i = strlen(T10) ; i+hostlen < 24; i++)
1578 ddf->controller.guid[i] = ' ';
1579
1580 ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1581 ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1582 ddf->controller.type.sub_vendor_id = 0;
1583 ddf->controller.type.sub_device_id = 0;
1584 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1585 memset(ddf->controller.pad, 0xff, 8);
1586 memset(ddf->controller.vendor_data, 0xff, 448);
1587
1588 pd = ddf->phys = malloc(pdsize);
1589 ddf->pdsize = pdsize;
1590
1591 memset(pd, 0xff, pdsize);
1592 memset(pd, 0, sizeof(*pd));
1593 pd->magic = DDF_PHYS_DATA_MAGIC;
1594 pd->used_pdes = __cpu_to_be16(0);
1595 pd->max_pdes = __cpu_to_be16(max_phys_disks);
1596 memset(pd->pad, 0xff, 52);
1597
1598 vd = ddf->virt = malloc(vdsize);
1599 ddf->vdsize = vdsize;
1600 memset(vd, 0, vdsize);
1601 vd->magic = DDF_VIRT_RECORDS_MAGIC;
1602 vd->populated_vdes = __cpu_to_be16(0);
1603 vd->max_vdes = __cpu_to_be16(max_virt_disks);
1604 memset(vd->pad, 0xff, 52);
1605
1606 for (i=0; i<max_virt_disks; i++)
1607 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1608
1609 st->sb = ddf;
1610 return 1;
1611 }
1612
1613 static int all_ff(char *guid)
1614 {
1615 int i;
1616 for (i = 0; i < DDF_GUID_LEN; i++)
1617 if (guid[i] != (char)0xff)
1618 return 0;
1619 return 1;
1620 }
1621 static int chunk_to_shift(int chunksize)
1622 {
1623 return ffs(chunksize/512)-1;
1624 }
1625
1626 static int level_to_prl(int level)
1627 {
1628 switch (level) {
1629 case LEVEL_LINEAR: return DDF_CONCAT;
1630 case 0: return DDF_RAID0;
1631 case 1: return DDF_RAID1;
1632 case 4: return DDF_RAID4;
1633 case 5: return DDF_RAID5;
1634 case 6: return DDF_RAID6;
1635 default: return -1;
1636 }
1637 }
1638 static int layout_to_rlq(int level, int layout, int raiddisks)
1639 {
1640 switch(level) {
1641 case 0:
1642 return DDF_RAID0_SIMPLE;
1643 case 1:
1644 switch(raiddisks) {
1645 case 2: return DDF_RAID1_SIMPLE;
1646 case 3: return DDF_RAID1_MULTI;
1647 default: return -1;
1648 }
1649 case 4:
1650 switch(layout) {
1651 case 0: return DDF_RAID4_N;
1652 }
1653 break;
1654 case 5:
1655 case 6:
1656 switch(layout) {
1657 case ALGORITHM_LEFT_ASYMMETRIC:
1658 return DDF_RAID5_N_RESTART;
1659 case ALGORITHM_RIGHT_ASYMMETRIC:
1660 if (level == 5)
1661 return DDF_RAID5_0_RESTART;
1662 else
1663 return DDF_RAID6_0_RESTART;
1664 case ALGORITHM_LEFT_SYMMETRIC:
1665 return DDF_RAID5_N_CONTINUE;
1666 case ALGORITHM_RIGHT_SYMMETRIC:
1667 return -1; /* not mentioned in standard */
1668 }
1669 }
1670 return -1;
1671 }
1672
1673 static int rlq_to_layout(int rlq, int prl, int raiddisks)
1674 {
1675 switch(prl) {
1676 case DDF_RAID0:
1677 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1678 case DDF_RAID1:
1679 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1680 on raiddisks*/
1681 case DDF_RAID4:
1682 switch(rlq) {
1683 case DDF_RAID4_N:
1684 return 0;
1685 default:
1686 /* not supported */
1687 return -1; /* FIXME this isn't checked */
1688 }
1689 case DDF_RAID5:
1690 switch(rlq) {
1691 case DDF_RAID5_N_RESTART:
1692 return ALGORITHM_LEFT_ASYMMETRIC;
1693 case DDF_RAID5_0_RESTART:
1694 return ALGORITHM_RIGHT_ASYMMETRIC;
1695 case DDF_RAID5_N_CONTINUE:
1696 return ALGORITHM_LEFT_SYMMETRIC;
1697 default:
1698 return -1;
1699 }
1700 case DDF_RAID6:
1701 switch(rlq) {
1702 case DDF_RAID5_N_RESTART:
1703 return ALGORITHM_LEFT_ASYMMETRIC;
1704 case DDF_RAID6_0_RESTART:
1705 return ALGORITHM_RIGHT_ASYMMETRIC;
1706 case DDF_RAID5_N_CONTINUE:
1707 return ALGORITHM_LEFT_SYMMETRIC;
1708 default:
1709 return -1;
1710 }
1711 }
1712 return -1;
1713 }
1714
1715 struct extent {
1716 unsigned long long start, size;
1717 };
1718 int cmp_extent(const void *av, const void *bv)
1719 {
1720 const struct extent *a = av;
1721 const struct extent *b = bv;
1722 if (a->start < b->start)
1723 return -1;
1724 if (a->start > b->start)
1725 return 1;
1726 return 0;
1727 }
1728
1729 struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
1730 {
1731 /* find a list of used extents on the give physical device
1732 * (dnum) of the given ddf.
1733 * Return a malloced array of 'struct extent'
1734
1735 FIXME ignore DDF_Legacy devices?
1736
1737 */
1738 struct extent *rv;
1739 int n = 0;
1740 int i, j;
1741
1742 rv = malloc(sizeof(struct extent) * (ddf->max_part + 2));
1743 if (!rv)
1744 return NULL;
1745
1746 for (i = 0; i < ddf->max_part; i++) {
1747 struct vcl *v = dl->vlist[i];
1748 if (v == NULL)
1749 continue;
1750 for (j=0; j < v->conf.prim_elmnt_count; j++)
1751 if (v->conf.phys_refnum[j] == dl->disk.refnum) {
1752 /* This device plays role 'j' in 'v'. */
1753 rv[n].start = __be64_to_cpu(v->lba_offset[j]);
1754 rv[n].size = __be64_to_cpu(v->conf.blocks);
1755 n++;
1756 break;
1757 }
1758 }
1759 qsort(rv, n, sizeof(*rv), cmp_extent);
1760
1761 rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
1762 rv[n].size = 0;
1763 return rv;
1764 }
1765
1766 static int init_super_ddf_bvd(struct supertype *st,
1767 mdu_array_info_t *info,
1768 unsigned long long size,
1769 char *name, char *homehost,
1770 int *uuid)
1771 {
1772 /* We are creating a BVD inside a pre-existing container.
1773 * so st->sb is already set.
1774 * We need to create a new vd_config and a new virtual_entry
1775 */
1776 struct ddf_super *ddf = st->sb;
1777 int venum;
1778 struct virtual_entry *ve;
1779 struct vcl *vcl;
1780 struct vd_config *vc;
1781
1782 if (__be16_to_cpu(ddf->virt->populated_vdes)
1783 >= __be16_to_cpu(ddf->virt->max_vdes)) {
1784 fprintf(stderr, Name": This ddf already has the "
1785 "maximum of %d virtual devices\n",
1786 __be16_to_cpu(ddf->virt->max_vdes));
1787 return 0;
1788 }
1789
1790 for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1791 if (all_ff(ddf->virt->entries[venum].guid))
1792 break;
1793 if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1794 fprintf(stderr, Name ": Cannot find spare slot for "
1795 "virtual disk - DDF is corrupt\n");
1796 return 0;
1797 }
1798 ve = &ddf->virt->entries[venum];
1799
1800 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1801 * timestamp, random number
1802 */
1803 make_header_guid(ve->guid);
1804 ve->unit = __cpu_to_be16(info->md_minor);
1805 ve->pad0 = 0xFFFF;
1806 ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1807 ve->type = 0;
1808 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
1809 if (info->state & 1) /* clean */
1810 ve->init_state = DDF_init_full;
1811 else
1812 ve->init_state = DDF_init_not;
1813
1814 memset(ve->pad1, 0xff, 14);
1815 memset(ve->name, ' ', 16);
1816 if (name)
1817 strncpy(ve->name, name, 16);
1818 ddf->virt->populated_vdes =
1819 __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
1820
1821 /* Now create a new vd_config */
1822 vcl = malloc(offsetof(struct vcl, conf) + ddf->conf_rec_len * 512);
1823 vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
1824 vcl->vcnum = venum;
1825 sprintf(st->subarray, "%d", venum);
1826 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1827
1828 vc = &vcl->conf;
1829
1830 vc->magic = DDF_VD_CONF_MAGIC;
1831 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
1832 vc->timestamp = __cpu_to_be32(time(0)-DECADE);
1833 vc->seqnum = __cpu_to_be32(1);
1834 memset(vc->pad0, 0xff, 24);
1835 vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
1836 vc->chunk_shift = chunk_to_shift(info->chunk_size);
1837 vc->prl = level_to_prl(info->level);
1838 vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
1839 vc->sec_elmnt_count = 1;
1840 vc->sec_elmnt_seq = 0;
1841 vc->srl = 0;
1842 vc->blocks = __cpu_to_be64(info->size * 2);
1843 vc->array_blocks = __cpu_to_be64(
1844 calc_array_size(info->level, info->raid_disks, info->layout,
1845 info->chunk_size, info->size*2));
1846 memset(vc->pad1, 0xff, 8);
1847 vc->spare_refs[0] = 0xffffffff;
1848 vc->spare_refs[1] = 0xffffffff;
1849 vc->spare_refs[2] = 0xffffffff;
1850 vc->spare_refs[3] = 0xffffffff;
1851 vc->spare_refs[4] = 0xffffffff;
1852 vc->spare_refs[5] = 0xffffffff;
1853 vc->spare_refs[6] = 0xffffffff;
1854 vc->spare_refs[7] = 0xffffffff;
1855 memset(vc->cache_pol, 0, 8);
1856 vc->bg_rate = 0x80;
1857 memset(vc->pad2, 0xff, 3);
1858 memset(vc->pad3, 0xff, 52);
1859 memset(vc->pad4, 0xff, 192);
1860 memset(vc->v0, 0xff, 32);
1861 memset(vc->v1, 0xff, 32);
1862 memset(vc->v2, 0xff, 16);
1863 memset(vc->v3, 0xff, 16);
1864 memset(vc->vendor, 0xff, 32);
1865
1866 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
1867 memset(vc->phys_refnum+(ddf->mppe * 4), 0x00, 8*ddf->mppe);
1868
1869 vcl->next = ddf->conflist;
1870 ddf->conflist = vcl;
1871 ddf->newconf = vcl;
1872 return 1;
1873 }
1874
1875 static void add_to_super_ddf_bvd(struct supertype *st,
1876 mdu_disk_info_t *dk, int fd, char *devname)
1877 {
1878 /* fd and devname identify a device with-in the ddf container (st).
1879 * dk identifies a location in the new BVD.
1880 * We need to find suitable free space in that device and update
1881 * the phys_refnum and lba_offset for the newly created vd_config.
1882 * We might also want to update the type in the phys_disk
1883 * section.
1884 */
1885 struct dl *dl;
1886 struct ddf_super *ddf = st->sb;
1887 struct vd_config *vc;
1888 __u64 *lba_offset;
1889 int working;
1890 int i;
1891 unsigned long long blocks, pos, esize;
1892 struct extent *ex;
1893
1894 for (dl = ddf->dlist; dl ; dl = dl->next)
1895 if (dl->major == dk->major &&
1896 dl->minor == dk->minor)
1897 break;
1898 if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
1899 return;
1900
1901 vc = &ddf->newconf->conf;
1902 lba_offset = ddf->newconf->lba_offset;
1903
1904 ex = get_extents(ddf, dl);
1905 if (!ex)
1906 return;
1907
1908 i = 0; pos = 0;
1909 blocks = __be64_to_cpu(vc->blocks);
1910 if (ddf->newconf->block_sizes)
1911 blocks = ddf->newconf->block_sizes[dk->raid_disk];
1912
1913 do {
1914 esize = ex[i].start - pos;
1915 if (esize >= blocks)
1916 break;
1917 pos = ex[i].start + ex[i].size;
1918 i++;
1919 } while (ex[i-1].size);
1920
1921 free(ex);
1922 if (esize < blocks)
1923 return;
1924
1925 vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
1926 lba_offset[dk->raid_disk] = __cpu_to_be64(pos);
1927
1928 for (i=0; i < ddf->max_part ; i++)
1929 if (dl->vlist[i] == NULL)
1930 break;
1931 if (i == ddf->max_part)
1932 return;
1933 dl->vlist[i] = ddf->newconf;
1934
1935 dl->fd = fd;
1936 dl->devname = devname;
1937
1938 /* Check how many working raid_disks, and if we can mark
1939 * array as optimal yet
1940 */
1941 working = 0;
1942
1943 for (i=0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
1944 if (vc->phys_refnum[i] != 0xffffffff)
1945 working++;
1946
1947 /* Find which virtual_entry */
1948 i = ddf->newconf->vcnum;
1949 if (working == __be16_to_cpu(vc->prim_elmnt_count))
1950 ddf->virt->entries[i].state =
1951 (ddf->virt->entries[i].state & ~DDF_state_mask)
1952 | DDF_state_optimal;
1953
1954 if (vc->prl == DDF_RAID6 &&
1955 working+1 == __be16_to_cpu(vc->prim_elmnt_count))
1956 ddf->virt->entries[i].state =
1957 (ddf->virt->entries[i].state & ~DDF_state_mask)
1958 | DDF_state_part_optimal;
1959
1960 ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
1961 ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
1962 }
1963
1964 /* add a device to a container, either while creating it or while
1965 * expanding a pre-existing container
1966 */
1967 static void add_to_super_ddf(struct supertype *st,
1968 mdu_disk_info_t *dk, int fd, char *devname)
1969 {
1970 struct ddf_super *ddf = st->sb;
1971 struct dl *dd;
1972 time_t now;
1973 struct tm *tm;
1974 unsigned long long size;
1975 struct phys_disk_entry *pde;
1976 int n, i;
1977 struct stat stb;
1978
1979 /* This is device numbered dk->number. We need to create
1980 * a phys_disk entry and a more detailed disk_data entry.
1981 */
1982 fstat(fd, &stb);
1983 dd = malloc(sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part);
1984 dd->major = major(stb.st_rdev);
1985 dd->minor = minor(stb.st_rdev);
1986 dd->devname = devname;
1987 dd->next = ddf->dlist;
1988 dd->fd = fd;
1989 dd->spare = NULL;
1990
1991 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
1992 now = time(0);
1993 tm = localtime(&now);
1994 sprintf(dd->disk.guid, "%8s%04d%02d%02d",
1995 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
1996 *(__u32*)(dd->disk.guid + 16) = random();
1997 *(__u32*)(dd->disk.guid + 20) = random();
1998
1999 do {
2000 /* Cannot be bothered finding a CRC of some irrelevant details*/
2001 dd->disk.refnum = random();
2002 for (i = __be16_to_cpu(ddf->active->max_pd_entries) - 1;
2003 i >= 0; i--)
2004 if (ddf->phys->entries[i].refnum == dd->disk.refnum)
2005 break;
2006 } while (i >= 0);
2007
2008 dd->disk.forced_ref = 1;
2009 dd->disk.forced_guid = 1;
2010 memset(dd->disk.vendor, ' ', 32);
2011 memcpy(dd->disk.vendor, "Linux", 5);
2012 memset(dd->disk.pad, 0xff, 442);
2013 for (i = 0; i < ddf->max_part ; i++)
2014 dd->vlist[i] = NULL;
2015
2016 n = __be16_to_cpu(ddf->phys->used_pdes);
2017 pde = &ddf->phys->entries[n];
2018 dd->pdnum = n;
2019
2020 n++;
2021 ddf->phys->used_pdes = __cpu_to_be16(n);
2022
2023 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2024 pde->refnum = dd->disk.refnum;
2025 pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2026 pde->state = __cpu_to_be16(DDF_Online);
2027 get_dev_size(fd, NULL, &size);
2028 /* We are required to reserve 32Meg, and record the size in sectors */
2029 pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
2030 sprintf(pde->path, "%17.17s","Information: nil") ;
2031 memset(pde->pad, 0xff, 6);
2032
2033 ddf->dlist = dd;
2034 }
2035
2036 /*
2037 * This is the write_init_super method for a ddf container. It is
2038 * called when creating a container or adding another device to a
2039 * container.
2040 */
2041
2042 #ifndef MDASSEMBLE
2043 static int __write_init_super_ddf(struct supertype *st, int do_close)
2044 {
2045
2046 struct ddf_super *ddf = st->sb;
2047 int i;
2048 struct dl *d;
2049 int n_config;
2050 int conf_size;
2051
2052 unsigned long long size, sector;
2053
2054 for (d = ddf->dlist; d; d=d->next) {
2055 int fd = d->fd;
2056
2057 if (fd < 0)
2058 continue;
2059
2060 /* We need to fill in the primary, (secondary) and workspace
2061 * lba's in the headers, set their checksums,
2062 * Also checksum phys, virt....
2063 *
2064 * Then write everything out, finally the anchor is written.
2065 */
2066 get_dev_size(fd, NULL, &size);
2067 size /= 512;
2068 ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
2069 ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
2070 ddf->anchor.seq = __cpu_to_be32(1);
2071 memcpy(&ddf->primary, &ddf->anchor, 512);
2072 memcpy(&ddf->secondary, &ddf->anchor, 512);
2073
2074 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
2075 ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
2076 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
2077
2078 ddf->primary.openflag = 0;
2079 ddf->primary.type = DDF_HEADER_PRIMARY;
2080
2081 ddf->secondary.openflag = 0;
2082 ddf->secondary.type = DDF_HEADER_SECONDARY;
2083
2084 ddf->primary.crc = calc_crc(&ddf->primary, 512);
2085 ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
2086
2087 sector = size - 16*1024*2;
2088 lseek64(fd, sector<<9, 0);
2089 write(fd, &ddf->primary, 512);
2090
2091 ddf->controller.crc = calc_crc(&ddf->controller, 512);
2092 write(fd, &ddf->controller, 512);
2093
2094 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
2095
2096 write(fd, ddf->phys, ddf->pdsize);
2097
2098 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
2099 write(fd, ddf->virt, ddf->vdsize);
2100
2101 /* Now write lots of config records. */
2102 n_config = ddf->max_part;
2103 conf_size = ddf->conf_rec_len * 512;
2104 for (i = 0 ; i <= n_config ; i++) {
2105 struct vcl *c = d->vlist[i];
2106 if (i == n_config)
2107 c = (struct vcl*)d->spare;
2108
2109 if (c) {
2110 c->conf.crc = calc_crc(&c->conf, conf_size);
2111 write(fd, &c->conf, conf_size);
2112 } else {
2113 __u32 sig = 0xffffffff;
2114 write(fd, &sig, 4);
2115 lseek64(fd, conf_size-4, SEEK_CUR);
2116 }
2117 }
2118 d->disk.crc = calc_crc(&d->disk, 512);
2119 write(fd, &d->disk, 512);
2120
2121 /* Maybe do the same for secondary */
2122
2123 lseek64(fd, (size-1)*512, SEEK_SET);
2124 write(fd, &ddf->anchor, 512);
2125 if (do_close) {
2126 close(fd);
2127 d->fd = -1;
2128 }
2129 }
2130 return 1;
2131 }
2132
2133 static int write_init_super_ddf(struct supertype *st)
2134 {
2135 return __write_init_super_ddf(st, 1);
2136 }
2137
2138 #endif
2139
2140 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize)
2141 {
2142 /* We must reserve the last 32Meg */
2143 if (devsize <= 32*1024*2)
2144 return 0;
2145 return devsize - 32*1024*2;
2146 }
2147
2148 #ifndef MDASSEMBLE
2149 int validate_geometry_ddf(struct supertype *st,
2150 int level, int layout, int raiddisks,
2151 int chunk, unsigned long long size,
2152 char *dev, unsigned long long *freesize)
2153 {
2154 int fd;
2155 struct mdinfo *sra;
2156 int cfd;
2157
2158 /* ddf potentially supports lots of things, but it depends on
2159 * what devices are offered (and maybe kernel version?)
2160 * If given unused devices, we will make a container.
2161 * If given devices in a container, we will make a BVD.
2162 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2163 */
2164
2165 if (level == LEVEL_CONTAINER) {
2166 st->ss = &super_ddf_container;
2167 if (dev) {
2168 int rv =st->ss->validate_geometry(st, level, layout,
2169 raiddisks, chunk,
2170 size,
2171 NULL, freesize);
2172 if (rv)
2173 return rv;
2174 }
2175 return st->ss->validate_geometry(st, level, layout, raiddisks,
2176 chunk, size, dev, freesize);
2177 }
2178
2179 if (st->sb) {
2180 /* creating in a given container */
2181 st->ss = &super_ddf_bvd;
2182 if (dev) {
2183 int rv =st->ss->validate_geometry(st, level, layout,
2184 raiddisks, chunk,
2185 size,
2186 NULL, freesize);
2187 if (rv)
2188 return rv;
2189 }
2190 return st->ss->validate_geometry(st, level, layout, raiddisks,
2191 chunk, size, dev, freesize);
2192 }
2193 /* FIXME should exclude MULTIPATH, or more appropriately, allow
2194 * only known levels.
2195 */
2196 if (!dev)
2197 return 1;
2198
2199 /* This device needs to be either a device in a 'ddf' container,
2200 * or it needs to be a 'ddf-bvd' array.
2201 */
2202
2203 fd = open(dev, O_RDONLY|O_EXCL, 0);
2204 if (fd >= 0) {
2205 sra = sysfs_read(fd, 0, GET_VERSION);
2206 close(fd);
2207 if (sra && sra->array.major_version == -1 &&
2208 strcmp(sra->text_version, "ddf-bvd") == 0) {
2209 st->ss = &super_ddf_svd;
2210 return st->ss->validate_geometry(st, level, layout,
2211 raiddisks, chunk, size,
2212 dev, freesize);
2213 }
2214
2215 fprintf(stderr,
2216 Name ": Cannot create this array on device %s\n",
2217 dev);
2218 return 0;
2219 }
2220 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2221 fprintf(stderr, Name ": Cannot open %s: %s\n",
2222 dev, strerror(errno));
2223 return 0;
2224 }
2225 /* Well, it is in use by someone, maybe a 'ddf' container. */
2226 cfd = open_container(fd);
2227 if (cfd < 0) {
2228 close(fd);
2229 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
2230 dev);
2231 return 0;
2232 }
2233 sra = sysfs_read(cfd, 0, GET_VERSION);
2234 close(fd);
2235 if (sra && sra->array.major_version == -1 &&
2236 strcmp(sra->text_version, "ddf") == 0) {
2237 /* This is a member of a ddf container. Load the container
2238 * and try to create a bvd
2239 */
2240 struct ddf_super *ddf;
2241 st->ss = &super_ddf_bvd;
2242 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL, 1) == 0) {
2243 st->sb = ddf;
2244 st->container_dev = fd2devnum(cfd);
2245 close(cfd);
2246 return st->ss->validate_geometry(st, level, layout,
2247 raiddisks, chunk, size,
2248 dev, freesize);
2249 }
2250 close(cfd);
2251 } else /* device may belong to a different container */
2252 return 0;
2253
2254 return 1;
2255 }
2256
2257 int validate_geometry_ddf_container(struct supertype *st,
2258 int level, int layout, int raiddisks,
2259 int chunk, unsigned long long size,
2260 char *dev, unsigned long long *freesize)
2261 {
2262 int fd;
2263 unsigned long long ldsize;
2264
2265 if (level != LEVEL_CONTAINER)
2266 return 0;
2267 if (!dev)
2268 return 1;
2269
2270 fd = open(dev, O_RDONLY|O_EXCL, 0);
2271 if (fd < 0) {
2272 fprintf(stderr, Name ": Cannot open %s: %s\n",
2273 dev, strerror(errno));
2274 return 0;
2275 }
2276 if (!get_dev_size(fd, dev, &ldsize)) {
2277 close(fd);
2278 return 0;
2279 }
2280 close(fd);
2281
2282 *freesize = avail_size_ddf(st, ldsize >> 9);
2283
2284 return 1;
2285 }
2286
2287 int validate_geometry_ddf_bvd(struct supertype *st,
2288 int level, int layout, int raiddisks,
2289 int chunk, unsigned long long size,
2290 char *dev, unsigned long long *freesize)
2291 {
2292 struct stat stb;
2293 struct ddf_super *ddf = st->sb;
2294 struct dl *dl;
2295 unsigned long long pos = 0;
2296 unsigned long long maxsize;
2297 struct extent *e;
2298 int i;
2299 /* ddf/bvd supports lots of things, but not containers */
2300 if (level == LEVEL_CONTAINER)
2301 return 0;
2302 /* We must have the container info already read in. */
2303 if (!ddf)
2304 return 0;
2305
2306 if (!dev) {
2307 /* General test: make sure there is space for
2308 * 'raiddisks' device extents of size 'size'.
2309 */
2310 unsigned long long minsize = size;
2311 int dcnt = 0;
2312 if (minsize == 0)
2313 minsize = 8;
2314 for (dl = ddf->dlist; dl ; dl = dl->next)
2315 {
2316 int found = 0;
2317 pos = 0;
2318
2319 i = 0;
2320 e = get_extents(ddf, dl);
2321 if (!e) continue;
2322 do {
2323 unsigned long long esize;
2324 esize = e[i].start - pos;
2325 if (esize >= minsize)
2326 found = 1;
2327 pos = e[i].start + e[i].size;
2328 i++;
2329 } while (e[i-1].size);
2330 if (found)
2331 dcnt++;
2332 free(e);
2333 }
2334 if (dcnt < raiddisks) {
2335 fprintf(stderr, Name ": Not enough devices with space "
2336 "for this array (%d < %d)\n",
2337 dcnt, raiddisks);
2338 return 0;
2339 }
2340 return 1;
2341 }
2342 /* This device must be a member of the set */
2343 if (stat(dev, &stb) < 0)
2344 return 0;
2345 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2346 return 0;
2347 for (dl = ddf->dlist ; dl ; dl = dl->next) {
2348 if (dl->major == major(stb.st_rdev) &&
2349 dl->minor == minor(stb.st_rdev))
2350 break;
2351 }
2352 if (!dl) {
2353 fprintf(stderr, Name ": %s is not in the same DDF set\n",
2354 dev);
2355 return 0;
2356 }
2357 e = get_extents(ddf, dl);
2358 maxsize = 0;
2359 i = 0;
2360 if (e) do {
2361 unsigned long long esize;
2362 esize = e[i].start - pos;
2363 if (esize >= maxsize)
2364 maxsize = esize;
2365 pos = e[i].start + e[i].size;
2366 i++;
2367 } while (e[i-1].size);
2368 *freesize = maxsize;
2369 // FIXME here I am
2370
2371 return 1;
2372 }
2373
2374 int validate_geometry_ddf_svd(struct supertype *st,
2375 int level, int layout, int raiddisks,
2376 int chunk, unsigned long long size,
2377 char *dev, unsigned long long *freesize)
2378 {
2379 /* dd/svd only supports striped, mirrored, concat, spanned... */
2380 if (level != LEVEL_LINEAR &&
2381 level != 0 &&
2382 level != 1)
2383 return 0;
2384 return 1;
2385 }
2386
2387 static int load_super_ddf_all(struct supertype *st, int fd,
2388 void **sbp, char *devname, int keep_fd)
2389 {
2390 struct mdinfo *sra;
2391 struct ddf_super *super;
2392 struct mdinfo *sd, *best = NULL;
2393 int bestseq = 0;
2394 int seq;
2395 char nm[20];
2396 int dfd;
2397
2398 sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
2399 if (!sra)
2400 return 1;
2401 if (sra->array.major_version != -1 ||
2402 sra->array.minor_version != -2 ||
2403 strcmp(sra->text_version, "ddf") != 0)
2404 return 1;
2405
2406 super = malloc(sizeof(*super));
2407 if (!super)
2408 return 1;
2409 memset(super, 0, sizeof(*super));
2410
2411 /* first, try each device, and choose the best ddf */
2412 for (sd = sra->devs ; sd ; sd = sd->next) {
2413 int rv;
2414 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2415 dfd = dev_open(nm, O_RDONLY);
2416 if (dfd < 0)
2417 return 2;
2418 rv = load_ddf_headers(dfd, super, NULL);
2419 close(dfd);
2420 if (rv == 0) {
2421 seq = __be32_to_cpu(super->active->seq);
2422 if (super->active->openflag)
2423 seq--;
2424 if (!best || seq > bestseq) {
2425 bestseq = seq;
2426 best = sd;
2427 }
2428 }
2429 }
2430 if (!best)
2431 return 1;
2432 /* OK, load this ddf */
2433 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2434 dfd = dev_open(nm, O_RDONLY);
2435 if (dfd < 0)
2436 return 1;
2437 load_ddf_headers(dfd, super, NULL);
2438 load_ddf_global(dfd, super, NULL);
2439 close(dfd);
2440 /* Now we need the device-local bits */
2441 for (sd = sra->devs ; sd ; sd = sd->next) {
2442 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2443 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2444 if (dfd < 0)
2445 return 2;
2446 seq = load_ddf_local(dfd, super, NULL, keep_fd);
2447 if (!keep_fd) close(dfd);
2448 }
2449 if (st->subarray[0]) {
2450 struct vcl *v;
2451
2452 for (v = super->conflist; v; v = v->next)
2453 if (v->vcnum == atoi(st->subarray))
2454 super->newconf = v;
2455 if (!super->newconf)
2456 return 1;
2457 }
2458 *sbp = super;
2459 if (st->ss == NULL) {
2460 st->ss = &super_ddf_container;
2461 st->minor_version = 0;
2462 st->max_devs = 512;
2463 st->container_dev = fd2devnum(fd);
2464 }
2465 return 0;
2466 }
2467 #endif
2468
2469 static struct mdinfo *container_content_ddf(struct supertype *st)
2470 {
2471 /* Given a container loaded by load_super_ddf_all,
2472 * extract information about all the arrays into
2473 * an mdinfo tree.
2474 *
2475 * For each vcl in conflist: create an mdinfo, fill it in,
2476 * then look for matching devices (phys_refnum) in dlist
2477 * and create appropriate device mdinfo.
2478 */
2479 struct ddf_super *ddf = st->sb;
2480 struct mdinfo *rest = NULL;
2481 struct vcl *vc;
2482
2483 for (vc = ddf->conflist ; vc ; vc=vc->next)
2484 {
2485 int i;
2486 struct mdinfo *this;
2487 this = malloc(sizeof(*this));
2488 memset(this, 0, sizeof(*this));
2489 this->next = rest;
2490 rest = this;
2491
2492 this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2493 this->array.raid_disks =
2494 __be16_to_cpu(vc->conf.prim_elmnt_count);
2495 this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2496 this->array.raid_disks);
2497 this->array.md_minor = -1;
2498 this->array.ctime = DECADE +
2499 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
2500 this->array.utime = DECADE +
2501 __be32_to_cpu(vc->conf.timestamp);
2502 this->array.chunk_size = 512 << vc->conf.chunk_shift;
2503
2504 i = vc->vcnum;
2505 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2506 (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
2507 DDF_init_full) {
2508 this->array.state = 0;
2509 this->resync_start = 0;
2510 } else {
2511 this->array.state = 1;
2512 this->resync_start = ~0ULL;
2513 }
2514 memcpy(this->name, ddf->virt->entries[i].name, 32);
2515 this->name[33]=0;
2516
2517 memset(this->uuid, 0, sizeof(this->uuid));
2518 this->component_size = __be64_to_cpu(vc->conf.blocks);
2519 this->array.size = this->component_size / 2;
2520 this->container_member = i;
2521
2522 sprintf(this->text_version, "/%s/%d",
2523 devnum2devname(st->container_dev),
2524 this->container_member);
2525
2526 for (i=0 ; i < ddf->mppe ; i++) {
2527 struct mdinfo *dev;
2528 struct dl *d;
2529
2530 if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
2531 continue;
2532
2533 this->array.working_disks++;
2534
2535 for (d = ddf->dlist; d ; d=d->next)
2536 if (d->disk.refnum == vc->conf.phys_refnum[i])
2537 break;
2538 if (d == NULL)
2539 break;
2540
2541 dev = malloc(sizeof(*dev));
2542 memset(dev, 0, sizeof(*dev));
2543 dev->next = this->devs;
2544 this->devs = dev;
2545
2546 dev->disk.number = __be32_to_cpu(d->disk.refnum);
2547 dev->disk.major = d->major;
2548 dev->disk.minor = d->minor;
2549 dev->disk.raid_disk = i;
2550 dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2551
2552 dev->events = __le32_to_cpu(ddf->primary.seq);
2553 dev->data_offset = vc->lba_offset[i];
2554 dev->component_size = __be64_to_cpu(vc->conf.blocks);
2555 if (d->devname)
2556 strcpy(dev->name, d->devname);
2557 }
2558 }
2559 return rest;
2560 }
2561
2562 static int init_zero_ddf(struct supertype *st,
2563 mdu_array_info_t *info,
2564 unsigned long long size, char *name,
2565 char *homehost, int *uuid)
2566 {
2567 st->sb = NULL;
2568 return 0;
2569 }
2570
2571 static int store_zero_ddf(struct supertype *st, int fd)
2572 {
2573 unsigned long long dsize;
2574 char buf[512];
2575 memset(buf, 0, 512);
2576
2577 if (!get_dev_size(fd, NULL, &dsize))
2578 return 1;
2579
2580 lseek64(fd, dsize-512, 0);
2581 write(fd, buf, 512);
2582 return 0;
2583 }
2584
2585 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
2586 {
2587 /*
2588 * return:
2589 * 0 same, or first was empty, and second was copied
2590 * 1 second had wrong number
2591 * 2 wrong uuid
2592 * 3 wrong other info
2593 */
2594 struct ddf_super *first = st->sb;
2595 struct ddf_super *second = tst->sb;
2596
2597 if (!first) {
2598 st->sb = tst->sb;
2599 tst->sb = NULL;
2600 return 0;
2601 }
2602
2603 if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
2604 return 2;
2605
2606 /* FIXME should I look at anything else? */
2607 return 0;
2608 }
2609
2610 /*
2611 * A new array 'a' has been started which claims to be instance 'inst'
2612 * within container 'c'.
2613 * We need to confirm that the array matches the metadata in 'c' so
2614 * that we don't corrupt any metadata.
2615 */
2616 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
2617 {
2618 fprintf(stderr, "ddf: open_new %s\n", inst);
2619 a->info.container_member = atoi(inst);
2620 return 0;
2621 }
2622
2623 /*
2624 * The array 'a' is to be marked clean in the metadata.
2625 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2626 * clean up to the point (in sectors). If that cannot be recorded in the
2627 * metadata, then leave it as dirty.
2628 *
2629 * For DDF, we need to clear the DDF_state_inconsistent bit in the
2630 * !global! virtual_disk.virtual_entry structure.
2631 */
2632 static void ddf_set_array_state(struct active_array *a, int consistent)
2633 {
2634 struct ddf_super *ddf = a->container->sb;
2635 int inst = a->info.container_member;
2636 if (consistent)
2637 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
2638 else
2639 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
2640 ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
2641 if (a->resync_start == ~0ULL)
2642 ddf->virt->entries[inst].init_state |= DDF_init_full;
2643 else if (a->resync_start == 0)
2644 ddf->virt->entries[inst].init_state |= DDF_init_not;
2645 else
2646 ddf->virt->entries[inst].init_state |= DDF_init_quick;
2647
2648 printf("ddf mark %s %llu\n", consistent?"clean":"dirty",
2649 a->resync_start);
2650 }
2651
2652 /*
2653 * The state of each disk is stored in the global phys_disk structure
2654 * in phys_disk.entries[n].state.
2655 * This makes various combinations awkward.
2656 * - When a device fails in any array, it must be failed in all arrays
2657 * that include a part of this device.
2658 * - When a component is rebuilding, we cannot include it officially in the
2659 * array unless this is the only array that uses the device.
2660 *
2661 * So: when transitioning:
2662 * Online -> failed, just set failed flag. monitor will propagate
2663 * spare -> online, the device might need to be added to the array.
2664 * spare -> failed, just set failed. Don't worry if in array or not.
2665 */
2666 static void ddf_set_disk(struct active_array *a, int n, int state)
2667 {
2668 struct ddf_super *ddf = a->container->sb;
2669 int inst = a->info.container_member;
2670 struct vd_config *vc = find_vdcr(ddf, inst);
2671 int pd = find_phys(ddf, vc->phys_refnum[n]);
2672 int i, st, working;
2673
2674 if (vc == NULL) {
2675 fprintf(stderr, "ddf: cannot find instance %d!!\n", inst);
2676 return;
2677 }
2678 if (pd < 0) {
2679 /* disk doesn't currently exist. If it is now in_sync,
2680 * insert it. */
2681 if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
2682 /* Find dev 'n' in a->info->devs, determine the
2683 * ddf refnum, and set vc->phys_refnum and update
2684 * phys->entries[]
2685 */
2686 /* FIXME */
2687 }
2688 } else {
2689 if (state & DS_FAULTY)
2690 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Failed);
2691 if (state & DS_INSYNC) {
2692 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Online);
2693 ddf->phys->entries[pd].state &= __cpu_to_be16(~DDF_Rebuilding);
2694 }
2695 }
2696
2697 fprintf(stderr, "ddf: set_disk %d to %x\n", n, state);
2698
2699 /* Now we need to check the state of the array and update
2700 * virtual_disk.entries[n].state.
2701 * It needs to be one of "optimal", "degraded", "failed".
2702 * I don't understand 'deleted' or 'missing'.
2703 */
2704 working = 0;
2705 for (i=0; i < a->info.array.raid_disks; i++) {
2706 pd = find_phys(ddf, vc->phys_refnum[i]);
2707 if (pd < 0)
2708 continue;
2709 st = __be16_to_cpu(ddf->phys->entries[pd].state);
2710 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
2711 == DDF_Online)
2712 working++;
2713 }
2714 state = DDF_state_degraded;
2715 if (working == a->info.array.raid_disks)
2716 state = DDF_state_optimal;
2717 else switch(vc->prl) {
2718 case DDF_RAID0:
2719 case DDF_CONCAT:
2720 case DDF_JBOD:
2721 state = DDF_state_failed;
2722 break;
2723 case DDF_RAID1:
2724 if (working == 0)
2725 state = DDF_state_failed;
2726 break;
2727 case DDF_RAID4:
2728 case DDF_RAID5:
2729 if (working < a->info.array.raid_disks-1)
2730 state = DDF_state_failed;
2731 break;
2732 case DDF_RAID6:
2733 if (working < a->info.array.raid_disks-2)
2734 state = DDF_state_failed;
2735 else if (working == a->info.array.raid_disks-1)
2736 state = DDF_state_part_optimal;
2737 break;
2738 }
2739
2740 ddf->virt->entries[inst].state =
2741 (ddf->virt->entries[inst].state & ~DDF_state_mask)
2742 | state;
2743
2744 }
2745
2746 static void ddf_sync_metadata(struct supertype *st)
2747 {
2748
2749 /*
2750 * Write all data to all devices.
2751 * Later, we might be able to track whether only local changes
2752 * have been made, or whether any global data has been changed,
2753 * but ddf is sufficiently weird that it probably always
2754 * changes global data ....
2755 */
2756 __write_init_super_ddf(st, 0);
2757 fprintf(stderr, "ddf: sync_metadata\n");
2758 }
2759
2760 static void ddf_process_update(struct supertype *st,
2761 struct metadata_update *update)
2762 {
2763 /* Apply this update to the metadata.
2764 * The first 4 bytes are a DDF_*_MAGIC which guides
2765 * our actions.
2766 * Possible update are:
2767 * DDF_PHYS_RECORDS_MAGIC
2768 * Add a new physical device. Changes to this record
2769 * only happen implicitly.
2770 * used_pdes is the device number.
2771 * DDF_VIRT_RECORDS_MAGIC
2772 * Add a new VD. Possibly also change the 'access' bits.
2773 * populated_vdes is the entry number.
2774 * DDF_VD_CONF_MAGIC
2775 * New or updated VD. the VIRT_RECORD must already
2776 * exist. For an update, phys_refnum and lba_offset
2777 * (at least) are updated, and the VD_CONF must
2778 * be written to precisely those devices listed with
2779 * a phys_refnum.
2780 * DDF_SPARE_ASSIGN_MAGIC
2781 * replacement Spare Assignment Record... but for which device?
2782 *
2783 * So, e.g.:
2784 * - to create a new array, we send a VIRT_RECORD and
2785 * a VD_CONF. Then assemble and start the array.
2786 * - to activate a spare we send a VD_CONF to add the phys_refnum
2787 * and offset. This will also mark the spare as active with
2788 * a spare-assignment record.
2789 */
2790 struct ddf_super *ddf = st->sb;
2791 __u32 *magic = (__u32*)update->buf;
2792 struct phys_disk *pd;
2793 struct virtual_disk *vd;
2794 struct vd_config *vc;
2795 struct vcl *vcl;
2796 struct dl *dl;
2797 int mppe;
2798 int ent;
2799
2800 printf("Process update %x\n", *magic);
2801
2802 switch (*magic) {
2803 case DDF_PHYS_RECORDS_MAGIC:
2804
2805 if (update->len != (sizeof(struct phys_disk) +
2806 sizeof(struct phys_disk_entry)))
2807 return;
2808 pd = (struct phys_disk*)update->buf;
2809
2810 ent = __be16_to_cpu(pd->used_pdes);
2811 if (ent >= __be16_to_cpu(ddf->phys->max_pdes))
2812 return;
2813 if (!all_ff(ddf->phys->entries[ent].guid))
2814 return;
2815 ddf->phys->entries[ent] = pd->entries[0];
2816 ddf->phys->used_pdes = __cpu_to_be16(1 +
2817 __be16_to_cpu(ddf->phys->used_pdes));
2818 break;
2819
2820 case DDF_VIRT_RECORDS_MAGIC:
2821
2822 if (update->len != (sizeof(struct virtual_disk) +
2823 sizeof(struct virtual_entry)))
2824 return;
2825 vd = (struct virtual_disk*)update->buf;
2826
2827 ent = __be16_to_cpu(vd->populated_vdes);
2828 if (ent >= __be16_to_cpu(ddf->virt->max_vdes))
2829 return;
2830 if (!all_ff(ddf->virt->entries[ent].guid))
2831 return;
2832 ddf->virt->entries[ent] = vd->entries[0];
2833 ddf->virt->populated_vdes = __cpu_to_be16(1 +
2834 __be16_to_cpu(ddf->virt->populated_vdes));
2835 break;
2836
2837 case DDF_VD_CONF_MAGIC:
2838 printf("len %d %d\n", update->len, ddf->conf_rec_len);
2839
2840 mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries);
2841 if (update->len != ddf->conf_rec_len)
2842 return;
2843 vc = (struct vd_config*)update->buf;
2844 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2845 if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
2846 break;
2847 printf("vcl = %p\n", vcl);
2848 if (vcl) {
2849 /* An update, just copy the phys_refnum and lba_offset
2850 * fields
2851 */
2852 memcpy(vcl->conf.phys_refnum, vc->phys_refnum,
2853 mppe * (sizeof(__u32) + sizeof(__u64)));
2854 } else {
2855 /* A new VD_CONF */
2856 vcl = update->space;
2857 update->space = NULL;
2858 vcl->next = ddf->conflist;
2859 vcl->conf = *vc;
2860 vcl->lba_offset = (__u64*)
2861 &vcl->conf.phys_refnum[mppe];
2862 ddf->conflist = vcl;
2863 }
2864 /* Now make sure vlist is correct for each dl. */
2865 for (dl = ddf->dlist; dl; dl = dl->next) {
2866 int dn;
2867 int vn = 0;
2868 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
2869 for (dn=0; dn < ddf->mppe ; dn++)
2870 if (vcl->conf.phys_refnum[dn] ==
2871 dl->disk.refnum) {
2872 printf("dev %d has %p at %d\n",
2873 dl->pdnum, vcl, vn);
2874 dl->vlist[vn++] = vcl;
2875 break;
2876 }
2877 while (vn < ddf->max_part)
2878 dl->vlist[vn++] = NULL;
2879 if (dl->vlist[0]) {
2880 ddf->phys->entries[dl->pdnum].type &=
2881 ~__cpu_to_be16(DDF_Global_Spare);
2882 ddf->phys->entries[dl->pdnum].type |=
2883 __cpu_to_be16(DDF_Active_in_VD);
2884 }
2885 if (dl->spare) {
2886 ddf->phys->entries[dl->pdnum].type &=
2887 ~__cpu_to_be16(DDF_Global_Spare);
2888 ddf->phys->entries[dl->pdnum].type |=
2889 __cpu_to_be16(DDF_Spare);
2890 }
2891 if (!dl->vlist[0] && !dl->spare) {
2892 ddf->phys->entries[dl->pdnum].type |=
2893 __cpu_to_be16(DDF_Global_Spare);
2894 ddf->phys->entries[dl->pdnum].type &=
2895 ~__cpu_to_be16(DDF_Spare |
2896 DDF_Active_in_VD);
2897 }
2898 }
2899 break;
2900 case DDF_SPARE_ASSIGN_MAGIC:
2901 default: break;
2902 }
2903 }
2904
2905 /*
2906 * Check if the array 'a' is degraded but not failed.
2907 * If it is, find as many spares as are available and needed and
2908 * arrange for their inclusion.
2909 * We only choose devices which are not already in the array,
2910 * and prefer those with a spare-assignment to this array.
2911 * otherwise we choose global spares - assuming always that
2912 * there is enough room.
2913 * For each spare that we assign, we return an 'mdinfo' which
2914 * describes the position for the device in the array.
2915 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
2916 * the new phys_refnum and lba_offset values.
2917 *
2918 * Only worry about BVDs at the moment.
2919 */
2920 static struct mdinfo *ddf_activate_spare(struct active_array *a,
2921 struct metadata_update **updates)
2922 {
2923 int working = 0;
2924 struct mdinfo *d;
2925 struct ddf_super *ddf = a->container->sb;
2926 int global_ok = 0;
2927 struct mdinfo *rv = NULL;
2928 struct mdinfo *di;
2929 struct metadata_update *mu;
2930 struct dl *dl;
2931 int i;
2932 struct vd_config *vc;
2933 __u64 *lba;
2934
2935 for (d = a->info.devs ; d ; d = d->next) {
2936 if ((d->curr_state & DS_FAULTY) &&
2937 d->state_fd >= 0)
2938 /* wait for Removal to happen */
2939 return NULL;
2940 if (d->state_fd >= 0)
2941 working ++;
2942 }
2943
2944 printf("ddf_activate: working=%d (%d) level=%d\n", working, a->info.array.raid_disks,
2945 a->info.array.level);
2946 if (working == a->info.array.raid_disks)
2947 return NULL; /* array not degraded */
2948 switch (a->info.array.level) {
2949 case 1:
2950 if (working == 0)
2951 return NULL; /* failed */
2952 break;
2953 case 4:
2954 case 5:
2955 if (working < a->info.array.raid_disks - 1)
2956 return NULL; /* failed */
2957 break;
2958 case 6:
2959 if (working < a->info.array.raid_disks - 2)
2960 return NULL; /* failed */
2961 break;
2962 default: /* concat or stripe */
2963 return NULL; /* failed */
2964 }
2965
2966 /* For each slot, if it is not working, find a spare */
2967 dl = ddf->dlist;
2968 for (i = 0; i < a->info.array.raid_disks; i++) {
2969 for (d = a->info.devs ; d ; d = d->next)
2970 if (d->disk.raid_disk == i)
2971 break;
2972 printf("found %d: %p %x\n", i, d, d?d->curr_state:0);
2973 if (d && (d->state_fd >= 0))
2974 continue;
2975
2976 /* OK, this device needs recovery. Find a spare */
2977 again:
2978 for ( ; dl ; dl = dl->next) {
2979 unsigned long long esize;
2980 unsigned long long pos;
2981 struct mdinfo *d2;
2982 int is_global = 0;
2983 int is_dedicated = 0;
2984 struct extent *ex;
2985 int j;
2986 /* If in this array, skip */
2987 for (d2 = a->info.devs ; d2 ; d2 = d2->next)
2988 if (d2->disk.major == dl->major &&
2989 d2->disk.minor == dl->minor) {
2990 printf("%x:%x already in array\n", dl->major, dl->minor);
2991 break;
2992 }
2993 if (d2)
2994 continue;
2995 if (ddf->phys->entries[dl->pdnum].type &
2996 __cpu_to_be16(DDF_Spare)) {
2997 /* Check spare assign record */
2998 if (dl->spare) {
2999 if (dl->spare->type & DDF_spare_dedicated) {
3000 /* check spare_ents for guid */
3001 for (j = 0 ;
3002 j < __be16_to_cpu(dl->spare->populated);
3003 j++) {
3004 if (memcmp(dl->spare->spare_ents[j].guid,
3005 ddf->virt->entries[a->info.container_member].guid,
3006 DDF_GUID_LEN) == 0)
3007 is_dedicated = 1;
3008 }
3009 } else
3010 is_global = 1;
3011 }
3012 } else if (ddf->phys->entries[dl->pdnum].type &
3013 __cpu_to_be16(DDF_Global_Spare)) {
3014 is_global = 1;
3015 }
3016 if ( ! (is_dedicated ||
3017 (is_global && global_ok))) {
3018 printf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
3019 is_dedicated, is_global);
3020 continue;
3021 }
3022
3023 /* We are allowed to use this device - is there space?
3024 * We need a->info.component_size sectors */
3025 ex = get_extents(ddf, dl);
3026 if (!ex) {
3027 printf("cannot get extents\n");
3028 continue;
3029 }
3030 j = 0; pos = 0;
3031 esize = 0;
3032
3033 do {
3034 esize = ex[j].start - pos;
3035 if (esize >= a->info.component_size)
3036 break;
3037 pos = ex[i].start + ex[i].size;
3038 i++;
3039 } while (ex[i-1].size);
3040
3041 free(ex);
3042 if (esize < a->info.component_size) {
3043 printf("%x:%x has no room: %llu %llu\n", dl->major, dl->minor,
3044 esize, a->info.component_size);
3045 /* No room */
3046 continue;
3047 }
3048
3049 /* Cool, we have a device with some space at pos */
3050 di = malloc(sizeof(*di));
3051 memset(di, 0, sizeof(*di));
3052 di->disk.number = i;
3053 di->disk.raid_disk = i;
3054 di->disk.major = dl->major;
3055 di->disk.minor = dl->minor;
3056 di->disk.state = 0;
3057 di->data_offset = pos;
3058 di->component_size = a->info.component_size;
3059 di->container_member = dl->pdnum;
3060 di->next = rv;
3061 rv = di;
3062 printf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3063 i, pos);
3064
3065 break;
3066 }
3067 if (!dl && ! global_ok) {
3068 /* not enough dedicated spares, try global */
3069 global_ok = 1;
3070 dl = ddf->dlist;
3071 goto again;
3072 }
3073 }
3074
3075 if (!rv)
3076 /* No spares found */
3077 return rv;
3078 /* Now 'rv' has a list of devices to return.
3079 * Create a metadata_update record to update the
3080 * phys_refnum and lba_offset values
3081 */
3082 mu = malloc(sizeof(*mu));
3083 mu->buf = malloc(ddf->conf_rec_len * 512);
3084 mu->space = malloc(sizeof(struct vcl));
3085 mu->len = ddf->conf_rec_len;
3086 mu->next = *updates;
3087 vc = find_vdcr(ddf, a->info.container_member);
3088 memcpy(mu->buf, vc, ddf->conf_rec_len * 512);
3089
3090 vc = (struct vd_config*)mu->buf;
3091 lba = (__u64*)&vc->phys_refnum[ddf->mppe];
3092 for (di = rv ; di ; di = di->next) {
3093 vc->phys_refnum[di->disk.raid_disk] =
3094 ddf->phys->entries[dl->pdnum].refnum;
3095 lba[di->disk.raid_disk] = di->data_offset;
3096 }
3097 *updates = mu;
3098 return rv;
3099 }
3100
3101 struct superswitch super_ddf = {
3102 #ifndef MDASSEMBLE
3103 .examine_super = examine_super_ddf,
3104 .brief_examine_super = brief_examine_super_ddf,
3105 .detail_super = detail_super_ddf,
3106 .brief_detail_super = brief_detail_super_ddf,
3107 .validate_geometry = validate_geometry_ddf,
3108 #endif
3109 .match_home = match_home_ddf,
3110 .uuid_from_super= uuid_from_super_ddf,
3111 .getinfo_super = getinfo_super_ddf,
3112 .update_super = update_super_ddf,
3113
3114 .avail_size = avail_size_ddf,
3115
3116 .compare_super = compare_super_ddf,
3117
3118 .load_super = load_super_ddf,
3119 .init_super = init_zero_ddf,
3120 .store_super = store_zero_ddf,
3121 .free_super = free_super_ddf,
3122 .match_metadata_desc = match_metadata_desc_ddf,
3123 .getinfo_super_n = getinfo_super_n_container,
3124
3125 .swapuuid = 0,
3126 .external = 1,
3127
3128 /* for mdmon */
3129 .open_new = ddf_open_new,
3130 .set_array_state= ddf_set_array_state,
3131 .set_disk = ddf_set_disk,
3132 .sync_metadata = ddf_sync_metadata,
3133 .process_update = ddf_process_update,
3134 .activate_spare = ddf_activate_spare,
3135
3136 };
3137
3138 /* Super_ddf_container is set by validate_geometry_ddf when given a
3139 * device that is not part of any array
3140 */
3141 static struct superswitch super_ddf_container = {
3142 #ifndef MDASSEMBLE
3143 .validate_geometry = validate_geometry_ddf_container,
3144 .write_init_super = write_init_super_ddf,
3145 #endif
3146
3147 .load_super = load_super_ddf,
3148 .init_super = init_super_ddf,
3149 .add_to_super = add_to_super_ddf,
3150 .getinfo_super = getinfo_super_ddf,
3151
3152 .free_super = free_super_ddf,
3153
3154 .container_content = container_content_ddf,
3155 .getinfo_super_n = getinfo_super_n_container,
3156
3157 .swapuuid = 0,
3158 .external = 1,
3159 };
3160
3161 static struct superswitch super_ddf_bvd = {
3162 #ifndef MDASSEMBLE
3163 // .detail_super = detail_super_ddf_bvd,
3164 // .brief_detail_super = brief_detail_super_ddf_bvd,
3165 .validate_geometry = validate_geometry_ddf_bvd,
3166 .write_init_super = write_init_super_ddf,
3167 #endif
3168 .update_super = update_super_ddf,
3169 .init_super = init_super_ddf_bvd,
3170 .add_to_super = add_to_super_ddf_bvd,
3171 .getinfo_super = getinfo_super_ddf_bvd,
3172 .getinfo_super_n = getinfo_super_n_bvd,
3173
3174 .load_super = load_super_ddf,
3175 .free_super = free_super_ddf,
3176 .match_metadata_desc = match_metadata_desc_ddf_bvd,
3177
3178 .swapuuid = 0,
3179 .external = 2,
3180 };
3181
3182 static struct superswitch super_ddf_svd = {
3183 #ifndef MDASSEMBLE
3184 // .detail_super = detail_super_ddf_svd,
3185 // .brief_detail_super = brief_detail_super_ddf_svd,
3186 .validate_geometry = validate_geometry_ddf_svd,
3187 #endif
3188 .update_super = update_super_ddf,
3189 .init_super = init_super_ddf,
3190
3191 .load_super = load_super_ddf,
3192 .free_super = free_super_ddf,
3193 .match_metadata_desc = match_metadata_desc_ddf_svd,
3194
3195 .swapuuid = 0,
3196 .external = 2,
3197 };
Unnamed repository; edit this file 'description' to name the repository.