fix examine_brief segfault
[thirdparty/mdadm.git] / super-ddf.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2006-2009 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 /* a non-official T10 name for creation GUIDs */
35 static char T10[] = "Linux-MD";
36
37 /* DDF timestamps are 1980 based, so we need to add
38 * second-in-decade-of-seventies to convert to linux timestamps.
39 * 10 years with 2 leap years.
40 */
41 #define DECADE (3600*24*(365*10+2))
42 unsigned long crc32(
43 unsigned long crc,
44 const unsigned char *buf,
45 unsigned len);
46
47 /* The DDF metadata handling.
48 * DDF metadata lives at the end of the device.
49 * The last 512 byte block provides an 'anchor' which is used to locate
50 * the rest of the metadata which usually lives immediately behind the anchor.
51 *
52 * Note:
53 * - all multibyte numeric fields are bigendian.
54 * - all strings are space padded.
55 *
56 */
57
58 /* Primary Raid Level (PRL) */
59 #define DDF_RAID0 0x00
60 #define DDF_RAID1 0x01
61 #define DDF_RAID3 0x03
62 #define DDF_RAID4 0x04
63 #define DDF_RAID5 0x05
64 #define DDF_RAID1E 0x11
65 #define DDF_JBOD 0x0f
66 #define DDF_CONCAT 0x1f
67 #define DDF_RAID5E 0x15
68 #define DDF_RAID5EE 0x25
69 #define DDF_RAID6 0x06
70
71 /* Raid Level Qualifier (RLQ) */
72 #define DDF_RAID0_SIMPLE 0x00
73 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
74 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
75 #define DDF_RAID3_0 0x00 /* parity in first extent */
76 #define DDF_RAID3_N 0x01 /* parity in last extent */
77 #define DDF_RAID4_0 0x00 /* parity in first extent */
78 #define DDF_RAID4_N 0x01 /* parity in last extent */
79 /* these apply to raid5e and raid5ee as well */
80 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
81 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
82 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
83 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
84
85 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
86 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
87
88 /* Secondary RAID Level (SRL) */
89 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
90 #define DDF_2MIRRORED 0x01
91 #define DDF_2CONCAT 0x02
92 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
93
94 /* Magic numbers */
95 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
96 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
97 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
98 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
99 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
100 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
101 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
102 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
103 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
104 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
105
106 #define DDF_GUID_LEN 24
107 #define DDF_REVISION_0 "01.00.00"
108 #define DDF_REVISION_2 "01.02.00"
109
110 struct ddf_header {
111 __u32 magic; /* DDF_HEADER_MAGIC */
112 __u32 crc;
113 char guid[DDF_GUID_LEN];
114 char revision[8]; /* 01.02.00 */
115 __u32 seq; /* starts at '1' */
116 __u32 timestamp;
117 __u8 openflag;
118 __u8 foreignflag;
119 __u8 enforcegroups;
120 __u8 pad0; /* 0xff */
121 __u8 pad1[12]; /* 12 * 0xff */
122 /* 64 bytes so far */
123 __u8 header_ext[32]; /* reserved: fill with 0xff */
124 __u64 primary_lba;
125 __u64 secondary_lba;
126 __u8 type;
127 __u8 pad2[3]; /* 0xff */
128 __u32 workspace_len; /* sectors for vendor space -
129 * at least 32768(sectors) */
130 __u64 workspace_lba;
131 __u16 max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */
132 __u16 max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */
133 __u16 max_partitions; /* i.e. max num of configuration
134 record entries per disk */
135 __u16 config_record_len; /* 1 +ROUNDUP(max_primary_element_entries
136 *12/512) */
137 __u16 max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */
138 __u8 pad3[54]; /* 0xff */
139 /* 192 bytes so far */
140 __u32 controller_section_offset;
141 __u32 controller_section_length;
142 __u32 phys_section_offset;
143 __u32 phys_section_length;
144 __u32 virt_section_offset;
145 __u32 virt_section_length;
146 __u32 config_section_offset;
147 __u32 config_section_length;
148 __u32 data_section_offset;
149 __u32 data_section_length;
150 __u32 bbm_section_offset;
151 __u32 bbm_section_length;
152 __u32 diag_space_offset;
153 __u32 diag_space_length;
154 __u32 vendor_offset;
155 __u32 vendor_length;
156 /* 256 bytes so far */
157 __u8 pad4[256]; /* 0xff */
158 };
159
160 /* type field */
161 #define DDF_HEADER_ANCHOR 0x00
162 #define DDF_HEADER_PRIMARY 0x01
163 #define DDF_HEADER_SECONDARY 0x02
164
165 /* The content of the 'controller section' - global scope */
166 struct ddf_controller_data {
167 __u32 magic; /* DDF_CONTROLLER_MAGIC */
168 __u32 crc;
169 char guid[DDF_GUID_LEN];
170 struct controller_type {
171 __u16 vendor_id;
172 __u16 device_id;
173 __u16 sub_vendor_id;
174 __u16 sub_device_id;
175 } type;
176 char product_id[16];
177 __u8 pad[8]; /* 0xff */
178 __u8 vendor_data[448];
179 };
180
181 /* The content of phys_section - global scope */
182 struct phys_disk {
183 __u32 magic; /* DDF_PHYS_RECORDS_MAGIC */
184 __u32 crc;
185 __u16 used_pdes;
186 __u16 max_pdes;
187 __u8 pad[52];
188 struct phys_disk_entry {
189 char guid[DDF_GUID_LEN];
190 __u32 refnum;
191 __u16 type;
192 __u16 state;
193 __u64 config_size; /* DDF structures must be after here */
194 char path[18]; /* another horrible structure really */
195 __u8 pad[6];
196 } entries[0];
197 };
198
199 /* phys_disk_entry.type is a bitmap - bigendian remember */
200 #define DDF_Forced_PD_GUID 1
201 #define DDF_Active_in_VD 2
202 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
203 #define DDF_Spare 8 /* overrides Global_spare */
204 #define DDF_Foreign 16
205 #define DDF_Legacy 32 /* no DDF on this device */
206
207 #define DDF_Interface_mask 0xf00
208 #define DDF_Interface_SCSI 0x100
209 #define DDF_Interface_SAS 0x200
210 #define DDF_Interface_SATA 0x300
211 #define DDF_Interface_FC 0x400
212
213 /* phys_disk_entry.state is a bigendian bitmap */
214 #define DDF_Online 1
215 #define DDF_Failed 2 /* overrides 1,4,8 */
216 #define DDF_Rebuilding 4
217 #define DDF_Transition 8
218 #define DDF_SMART 16
219 #define DDF_ReadErrors 32
220 #define DDF_Missing 64
221
222 /* The content of the virt_section global scope */
223 struct virtual_disk {
224 __u32 magic; /* DDF_VIRT_RECORDS_MAGIC */
225 __u32 crc;
226 __u16 populated_vdes;
227 __u16 max_vdes;
228 __u8 pad[52];
229 struct virtual_entry {
230 char guid[DDF_GUID_LEN];
231 __u16 unit;
232 __u16 pad0; /* 0xffff */
233 __u16 guid_crc;
234 __u16 type;
235 __u8 state;
236 __u8 init_state;
237 __u8 pad1[14];
238 char name[16];
239 } entries[0];
240 };
241
242 /* virtual_entry.type is a bitmap - bigendian */
243 #define DDF_Shared 1
244 #define DDF_Enforce_Groups 2
245 #define DDF_Unicode 4
246 #define DDF_Owner_Valid 8
247
248 /* virtual_entry.state is a bigendian bitmap */
249 #define DDF_state_mask 0x7
250 #define DDF_state_optimal 0x0
251 #define DDF_state_degraded 0x1
252 #define DDF_state_deleted 0x2
253 #define DDF_state_missing 0x3
254 #define DDF_state_failed 0x4
255 #define DDF_state_part_optimal 0x5
256
257 #define DDF_state_morphing 0x8
258 #define DDF_state_inconsistent 0x10
259
260 /* virtual_entry.init_state is a bigendian bitmap */
261 #define DDF_initstate_mask 0x03
262 #define DDF_init_not 0x00
263 #define DDF_init_quick 0x01 /* initialisation is progress.
264 * i.e. 'state_inconsistent' */
265 #define DDF_init_full 0x02
266
267 #define DDF_access_mask 0xc0
268 #define DDF_access_rw 0x00
269 #define DDF_access_ro 0x80
270 #define DDF_access_blocked 0xc0
271
272 /* The content of the config_section - local scope
273 * It has multiple records each config_record_len sectors
274 * They can be vd_config or spare_assign
275 */
276
277 struct vd_config {
278 __u32 magic; /* DDF_VD_CONF_MAGIC */
279 __u32 crc;
280 char guid[DDF_GUID_LEN];
281 __u32 timestamp;
282 __u32 seqnum;
283 __u8 pad0[24];
284 __u16 prim_elmnt_count;
285 __u8 chunk_shift; /* 0 == 512, 1==1024 etc */
286 __u8 prl;
287 __u8 rlq;
288 __u8 sec_elmnt_count;
289 __u8 sec_elmnt_seq;
290 __u8 srl;
291 __u64 blocks; /* blocks per component could be different
292 * on different component devices...(only
293 * for concat I hope) */
294 __u64 array_blocks; /* blocks in array */
295 __u8 pad1[8];
296 __u32 spare_refs[8];
297 __u8 cache_pol[8];
298 __u8 bg_rate;
299 __u8 pad2[3];
300 __u8 pad3[52];
301 __u8 pad4[192];
302 __u8 v0[32]; /* reserved- 0xff */
303 __u8 v1[32]; /* reserved- 0xff */
304 __u8 v2[16]; /* reserved- 0xff */
305 __u8 v3[16]; /* reserved- 0xff */
306 __u8 vendor[32];
307 __u32 phys_refnum[0]; /* refnum of each disk in sequence */
308 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
309 bvd are always the same size */
310 };
311
312 /* vd_config.cache_pol[7] is a bitmap */
313 #define DDF_cache_writeback 1 /* else writethrough */
314 #define DDF_cache_wadaptive 2 /* only applies if writeback */
315 #define DDF_cache_readahead 4
316 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
317 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
318 #define DDF_cache_wallowed 32 /* enable write caching */
319 #define DDF_cache_rallowed 64 /* enable read caching */
320
321 struct spare_assign {
322 __u32 magic; /* DDF_SPARE_ASSIGN_MAGIC */
323 __u32 crc;
324 __u32 timestamp;
325 __u8 reserved[7];
326 __u8 type;
327 __u16 populated; /* SAEs used */
328 __u16 max; /* max SAEs */
329 __u8 pad[8];
330 struct spare_assign_entry {
331 char guid[DDF_GUID_LEN];
332 __u16 secondary_element;
333 __u8 pad[6];
334 } spare_ents[0];
335 };
336 /* spare_assign.type is a bitmap */
337 #define DDF_spare_dedicated 0x1 /* else global */
338 #define DDF_spare_revertible 0x2 /* else committable */
339 #define DDF_spare_active 0x4 /* else not active */
340 #define DDF_spare_affinity 0x8 /* enclosure affinity */
341
342 /* The data_section contents - local scope */
343 struct disk_data {
344 __u32 magic; /* DDF_PHYS_DATA_MAGIC */
345 __u32 crc;
346 char guid[DDF_GUID_LEN];
347 __u32 refnum; /* crc of some magic drive data ... */
348 __u8 forced_ref; /* set when above was not result of magic */
349 __u8 forced_guid; /* set if guid was forced rather than magic */
350 __u8 vendor[32];
351 __u8 pad[442];
352 };
353
354 /* bbm_section content */
355 struct bad_block_log {
356 __u32 magic;
357 __u32 crc;
358 __u16 entry_count;
359 __u32 spare_count;
360 __u8 pad[10];
361 __u64 first_spare;
362 struct mapped_block {
363 __u64 defective_start;
364 __u32 replacement_start;
365 __u16 remap_count;
366 __u8 pad[2];
367 } entries[0];
368 };
369
370 /* Struct for internally holding ddf structures */
371 /* The DDF structure stored on each device is potentially
372 * quite different, as some data is global and some is local.
373 * The global data is:
374 * - ddf header
375 * - controller_data
376 * - Physical disk records
377 * - Virtual disk records
378 * The local data is:
379 * - Configuration records
380 * - Physical Disk data section
381 * ( and Bad block and vendor which I don't care about yet).
382 *
383 * The local data is parsed into separate lists as it is read
384 * and reconstructed for writing. This means that we only need
385 * to make config changes once and they are automatically
386 * propagated to all devices.
387 * Note that the ddf_super has space of the conf and disk data
388 * for this disk and also for a list of all such data.
389 * The list is only used for the superblock that is being
390 * built in Create or Assemble to describe the whole array.
391 */
392 struct ddf_super {
393 struct ddf_header anchor, primary, secondary;
394 struct ddf_controller_data controller;
395 struct ddf_header *active;
396 struct phys_disk *phys;
397 struct virtual_disk *virt;
398 int pdsize, vdsize;
399 int max_part, mppe, conf_rec_len;
400 int currentdev;
401 int updates_pending;
402 struct vcl {
403 union {
404 char space[512];
405 struct {
406 struct vcl *next;
407 __u64 *lba_offset; /* location in 'conf' of
408 * the lba table */
409 int vcnum; /* index into ->virt */
410 __u64 *block_sizes; /* NULL if all the same */
411 };
412 };
413 struct vd_config conf;
414 } *conflist, *currentconf;
415 struct dl {
416 union {
417 char space[512];
418 struct {
419 struct dl *next;
420 int major, minor;
421 char *devname;
422 int fd;
423 unsigned long long size; /* sectors */
424 int pdnum; /* index in ->phys */
425 struct spare_assign *spare;
426 void *mdupdate; /* hold metadata update */
427
428 /* These fields used by auto-layout */
429 int raiddisk; /* slot to fill in autolayout */
430 __u64 esize;
431 };
432 };
433 struct disk_data disk;
434 struct vcl *vlist[0]; /* max_part in size */
435 } *dlist, *add_list;
436 };
437
438 #ifndef offsetof
439 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
440 #endif
441
442
443 static int calc_crc(void *buf, int len)
444 {
445 /* crcs are always at the same place as in the ddf_header */
446 struct ddf_header *ddf = buf;
447 __u32 oldcrc = ddf->crc;
448 __u32 newcrc;
449 ddf->crc = 0xffffffff;
450
451 newcrc = crc32(0, buf, len);
452 ddf->crc = oldcrc;
453 /* The crc is store (like everything) bigendian, so convert
454 * here for simplicity
455 */
456 return __cpu_to_be32(newcrc);
457 }
458
459 static int load_ddf_header(int fd, unsigned long long lba,
460 unsigned long long size,
461 int type,
462 struct ddf_header *hdr, struct ddf_header *anchor)
463 {
464 /* read a ddf header (primary or secondary) from fd/lba
465 * and check that it is consistent with anchor
466 * Need to check:
467 * magic, crc, guid, rev, and LBA's header_type, and
468 * everything after header_type must be the same
469 */
470 if (lba >= size-1)
471 return 0;
472
473 if (lseek64(fd, lba<<9, 0) < 0)
474 return 0;
475
476 if (read(fd, hdr, 512) != 512)
477 return 0;
478
479 if (hdr->magic != DDF_HEADER_MAGIC)
480 return 0;
481 if (calc_crc(hdr, 512) != hdr->crc)
482 return 0;
483 if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
484 memcmp(anchor->revision, hdr->revision, 8) != 0 ||
485 anchor->primary_lba != hdr->primary_lba ||
486 anchor->secondary_lba != hdr->secondary_lba ||
487 hdr->type != type ||
488 memcmp(anchor->pad2, hdr->pad2, 512 -
489 offsetof(struct ddf_header, pad2)) != 0)
490 return 0;
491
492 /* Looks good enough to me... */
493 return 1;
494 }
495
496 static void *load_section(int fd, struct ddf_super *super, void *buf,
497 __u32 offset_be, __u32 len_be, int check)
498 {
499 unsigned long long offset = __be32_to_cpu(offset_be);
500 unsigned long long len = __be32_to_cpu(len_be);
501 int dofree = (buf == NULL);
502
503 if (check)
504 if (len != 2 && len != 8 && len != 32
505 && len != 128 && len != 512)
506 return NULL;
507
508 if (len > 1024)
509 return NULL;
510 if (buf) {
511 /* All pre-allocated sections are a single block */
512 if (len != 1)
513 return NULL;
514 } else if (posix_memalign(&buf, 512, len<<9) != 0)
515 buf = NULL;
516
517 if (!buf)
518 return NULL;
519
520 if (super->active->type == 1)
521 offset += __be64_to_cpu(super->active->primary_lba);
522 else
523 offset += __be64_to_cpu(super->active->secondary_lba);
524
525 if (lseek64(fd, offset<<9, 0) != (offset<<9)) {
526 if (dofree)
527 free(buf);
528 return NULL;
529 }
530 if (read(fd, buf, len<<9) != (len<<9)) {
531 if (dofree)
532 free(buf);
533 return NULL;
534 }
535 return buf;
536 }
537
538 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
539 {
540 unsigned long long dsize;
541
542 get_dev_size(fd, NULL, &dsize);
543
544 if (lseek64(fd, dsize-512, 0) < 0) {
545 if (devname)
546 fprintf(stderr,
547 Name": Cannot seek to anchor block on %s: %s\n",
548 devname, strerror(errno));
549 return 1;
550 }
551 if (read(fd, &super->anchor, 512) != 512) {
552 if (devname)
553 fprintf(stderr,
554 Name ": Cannot read anchor block on %s: %s\n",
555 devname, strerror(errno));
556 return 1;
557 }
558 if (super->anchor.magic != DDF_HEADER_MAGIC) {
559 if (devname)
560 fprintf(stderr, Name ": no DDF anchor found on %s\n",
561 devname);
562 return 2;
563 }
564 if (calc_crc(&super->anchor, 512) != super->anchor.crc) {
565 if (devname)
566 fprintf(stderr, Name ": bad CRC on anchor on %s\n",
567 devname);
568 return 2;
569 }
570 if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
571 memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
572 if (devname)
573 fprintf(stderr, Name ": can only support super revision"
574 " %.8s and earlier, not %.8s on %s\n",
575 DDF_REVISION_2, super->anchor.revision,devname);
576 return 2;
577 }
578 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba),
579 dsize >> 9, 1,
580 &super->primary, &super->anchor) == 0) {
581 if (devname)
582 fprintf(stderr,
583 Name ": Failed to load primary DDF header "
584 "on %s\n", devname);
585 return 2;
586 }
587 super->active = &super->primary;
588 if (load_ddf_header(fd, __be64_to_cpu(super->anchor.secondary_lba),
589 dsize >> 9, 2,
590 &super->secondary, &super->anchor)) {
591 if ((__be32_to_cpu(super->primary.seq)
592 < __be32_to_cpu(super->secondary.seq) &&
593 !super->secondary.openflag)
594 || (__be32_to_cpu(super->primary.seq)
595 == __be32_to_cpu(super->secondary.seq) &&
596 super->primary.openflag && !super->secondary.openflag)
597 )
598 super->active = &super->secondary;
599 }
600 return 0;
601 }
602
603 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
604 {
605 void *ok;
606 ok = load_section(fd, super, &super->controller,
607 super->active->controller_section_offset,
608 super->active->controller_section_length,
609 0);
610 super->phys = load_section(fd, super, NULL,
611 super->active->phys_section_offset,
612 super->active->phys_section_length,
613 1);
614 super->pdsize = __be32_to_cpu(super->active->phys_section_length) * 512;
615
616 super->virt = load_section(fd, super, NULL,
617 super->active->virt_section_offset,
618 super->active->virt_section_length,
619 1);
620 super->vdsize = __be32_to_cpu(super->active->virt_section_length) * 512;
621 if (!ok ||
622 !super->phys ||
623 !super->virt) {
624 free(super->phys);
625 free(super->virt);
626 super->phys = NULL;
627 super->virt = NULL;
628 return 2;
629 }
630 super->conflist = NULL;
631 super->dlist = NULL;
632
633 super->max_part = __be16_to_cpu(super->active->max_partitions);
634 super->mppe = __be16_to_cpu(super->active->max_primary_element_entries);
635 super->conf_rec_len = __be16_to_cpu(super->active->config_record_len);
636 return 0;
637 }
638
639 static int load_ddf_local(int fd, struct ddf_super *super,
640 char *devname, int keep)
641 {
642 struct dl *dl;
643 struct stat stb;
644 char *conf;
645 int i;
646 int confsec;
647 int vnum;
648 int max_virt_disks = __be16_to_cpu(super->active->max_vd_entries);
649 unsigned long long dsize;
650
651 /* First the local disk info */
652 if (posix_memalign((void**)&dl, 512,
653 sizeof(*dl) +
654 (super->max_part) * sizeof(dl->vlist[0])) != 0) {
655 fprintf(stderr, Name ": %s could not allocate disk info buffer\n",
656 __func__);
657 return 1;
658 }
659
660 load_section(fd, super, &dl->disk,
661 super->active->data_section_offset,
662 super->active->data_section_length,
663 0);
664 dl->devname = devname ? strdup(devname) : NULL;
665
666 fstat(fd, &stb);
667 dl->major = major(stb.st_rdev);
668 dl->minor = minor(stb.st_rdev);
669 dl->next = super->dlist;
670 dl->fd = keep ? fd : -1;
671
672 dl->size = 0;
673 if (get_dev_size(fd, devname, &dsize))
674 dl->size = dsize >> 9;
675 dl->spare = NULL;
676 for (i=0 ; i < super->max_part ; i++)
677 dl->vlist[i] = NULL;
678 super->dlist = dl;
679 dl->pdnum = -1;
680 for (i=0; i < __be16_to_cpu(super->active->max_pd_entries); i++)
681 if (memcmp(super->phys->entries[i].guid,
682 dl->disk.guid, DDF_GUID_LEN) == 0)
683 dl->pdnum = i;
684
685 /* Now the config list. */
686 /* 'conf' is an array of config entries, some of which are
687 * probably invalid. Those which are good need to be copied into
688 * the conflist
689 */
690
691 conf = load_section(fd, super, NULL,
692 super->active->config_section_offset,
693 super->active->config_section_length,
694 0);
695
696 vnum = 0;
697 for (confsec = 0;
698 confsec < __be32_to_cpu(super->active->config_section_length);
699 confsec += super->conf_rec_len) {
700 struct vd_config *vd =
701 (struct vd_config *)((char*)conf + confsec*512);
702 struct vcl *vcl;
703
704 if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) {
705 if (dl->spare)
706 continue;
707 if (posix_memalign((void**)&dl->spare, 512,
708 super->conf_rec_len*512) != 0) {
709 fprintf(stderr, Name
710 ": %s could not allocate spare info buf\n",
711 __func__);
712 return 1;
713 }
714
715 memcpy(dl->spare, vd, super->conf_rec_len*512);
716 continue;
717 }
718 if (vd->magic != DDF_VD_CONF_MAGIC)
719 continue;
720 for (vcl = super->conflist; vcl; vcl = vcl->next) {
721 if (memcmp(vcl->conf.guid,
722 vd->guid, DDF_GUID_LEN) == 0)
723 break;
724 }
725
726 if (vcl) {
727 dl->vlist[vnum++] = vcl;
728 if (__be32_to_cpu(vd->seqnum) <=
729 __be32_to_cpu(vcl->conf.seqnum))
730 continue;
731 } else {
732 if (posix_memalign((void**)&vcl, 512,
733 (super->conf_rec_len*512 +
734 offsetof(struct vcl, conf))) != 0) {
735 fprintf(stderr, Name
736 ": %s could not allocate vcl buf\n",
737 __func__);
738 return 1;
739 }
740 vcl->next = super->conflist;
741 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
742 super->conflist = vcl;
743 dl->vlist[vnum++] = vcl;
744 }
745 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
746 vcl->lba_offset = (__u64*)
747 &vcl->conf.phys_refnum[super->mppe];
748
749 for (i=0; i < max_virt_disks ; i++)
750 if (memcmp(super->virt->entries[i].guid,
751 vcl->conf.guid, DDF_GUID_LEN)==0)
752 break;
753 if (i < max_virt_disks)
754 vcl->vcnum = i;
755 }
756 free(conf);
757
758 return 0;
759 }
760
761 #ifndef MDASSEMBLE
762 static int load_super_ddf_all(struct supertype *st, int fd,
763 void **sbp, char *devname, int keep_fd);
764 #endif
765
766 static void free_super_ddf(struct supertype *st);
767
768 static int load_super_ddf(struct supertype *st, int fd,
769 char *devname)
770 {
771 unsigned long long dsize;
772 struct ddf_super *super;
773 int rv;
774
775 #ifndef MDASSEMBLE
776 /* if 'fd' is a container, load metadata from all the devices */
777 if (load_super_ddf_all(st, fd, &st->sb, devname, 1) == 0)
778 return 0;
779 #endif
780 if (st->subarray[0])
781 return 1; /* FIXME Is this correct */
782
783 if (get_dev_size(fd, devname, &dsize) == 0)
784 return 1;
785
786 /* 32M is a lower bound */
787 if (dsize <= 32*1024*1024) {
788 if (devname)
789 fprintf(stderr,
790 Name ": %s is too small for ddf: "
791 "size is %llu sectors.\n",
792 devname, dsize>>9);
793 return 1;
794 }
795 if (dsize & 511) {
796 if (devname)
797 fprintf(stderr,
798 Name ": %s is an odd size for ddf: "
799 "size is %llu bytes.\n",
800 devname, dsize);
801 return 1;
802 }
803
804 free_super_ddf(st);
805
806 if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) {
807 fprintf(stderr, Name ": malloc of %zu failed.\n",
808 sizeof(*super));
809 return 1;
810 }
811 memset(super, 0, sizeof(*super));
812
813 rv = load_ddf_headers(fd, super, devname);
814 if (rv) {
815 free(super);
816 return rv;
817 }
818
819 /* Have valid headers and have chosen the best. Let's read in the rest*/
820
821 rv = load_ddf_global(fd, super, devname);
822
823 if (rv) {
824 if (devname)
825 fprintf(stderr,
826 Name ": Failed to load all information "
827 "sections on %s\n", devname);
828 free(super);
829 return rv;
830 }
831
832 rv = load_ddf_local(fd, super, devname, 0);
833
834 if (rv) {
835 if (devname)
836 fprintf(stderr,
837 Name ": Failed to load all information "
838 "sections on %s\n", devname);
839 free(super);
840 return rv;
841 }
842
843 if (st->subarray[0]) {
844 struct vcl *v;
845
846 for (v = super->conflist; v; v = v->next)
847 if (v->vcnum == atoi(st->subarray))
848 super->currentconf = v;
849 if (!super->currentconf) {
850 free(super);
851 return 1;
852 }
853 }
854
855 /* Should possibly check the sections .... */
856
857 st->sb = super;
858 if (st->ss == NULL) {
859 st->ss = &super_ddf;
860 st->minor_version = 0;
861 st->max_devs = 512;
862 }
863 st->loaded_container = 0;
864 return 0;
865
866 }
867
868 static void free_super_ddf(struct supertype *st)
869 {
870 struct ddf_super *ddf = st->sb;
871 if (ddf == NULL)
872 return;
873 free(ddf->phys);
874 free(ddf->virt);
875 while (ddf->conflist) {
876 struct vcl *v = ddf->conflist;
877 ddf->conflist = v->next;
878 if (v->block_sizes)
879 free(v->block_sizes);
880 free(v);
881 }
882 while (ddf->dlist) {
883 struct dl *d = ddf->dlist;
884 ddf->dlist = d->next;
885 if (d->fd >= 0)
886 close(d->fd);
887 if (d->spare)
888 free(d->spare);
889 free(d);
890 }
891 free(ddf);
892 st->sb = NULL;
893 }
894
895 static struct supertype *match_metadata_desc_ddf(char *arg)
896 {
897 /* 'ddf' only support containers */
898 struct supertype *st;
899 if (strcmp(arg, "ddf") != 0 &&
900 strcmp(arg, "default") != 0
901 )
902 return NULL;
903
904 st = malloc(sizeof(*st));
905 memset(st, 0, sizeof(*st));
906 st->ss = &super_ddf;
907 st->max_devs = 512;
908 st->minor_version = 0;
909 st->sb = NULL;
910 return st;
911 }
912
913
914 #ifndef MDASSEMBLE
915
916 static mapping_t ddf_state[] = {
917 { "Optimal", 0},
918 { "Degraded", 1},
919 { "Deleted", 2},
920 { "Missing", 3},
921 { "Failed", 4},
922 { "Partially Optimal", 5},
923 { "-reserved-", 6},
924 { "-reserved-", 7},
925 { NULL, 0}
926 };
927
928 static mapping_t ddf_init_state[] = {
929 { "Not Initialised", 0},
930 { "QuickInit in Progress", 1},
931 { "Fully Initialised", 2},
932 { "*UNKNOWN*", 3},
933 { NULL, 0}
934 };
935 static mapping_t ddf_access[] = {
936 { "Read/Write", 0},
937 { "Reserved", 1},
938 { "Read Only", 2},
939 { "Blocked (no access)", 3},
940 { NULL ,0}
941 };
942
943 static mapping_t ddf_level[] = {
944 { "RAID0", DDF_RAID0},
945 { "RAID1", DDF_RAID1},
946 { "RAID3", DDF_RAID3},
947 { "RAID4", DDF_RAID4},
948 { "RAID5", DDF_RAID5},
949 { "RAID1E",DDF_RAID1E},
950 { "JBOD", DDF_JBOD},
951 { "CONCAT",DDF_CONCAT},
952 { "RAID5E",DDF_RAID5E},
953 { "RAID5EE",DDF_RAID5EE},
954 { "RAID6", DDF_RAID6},
955 { NULL, 0}
956 };
957 static mapping_t ddf_sec_level[] = {
958 { "Striped", DDF_2STRIPED},
959 { "Mirrored", DDF_2MIRRORED},
960 { "Concat", DDF_2CONCAT},
961 { "Spanned", DDF_2SPANNED},
962 { NULL, 0}
963 };
964 #endif
965
966 struct num_mapping {
967 int num1, num2;
968 };
969 static struct num_mapping ddf_level_num[] = {
970 { DDF_RAID0, 0 },
971 { DDF_RAID1, 1 },
972 { DDF_RAID3, LEVEL_UNSUPPORTED },
973 { DDF_RAID4, 4 },
974 { DDF_RAID5, 5 },
975 { DDF_RAID1E, LEVEL_UNSUPPORTED },
976 { DDF_JBOD, LEVEL_UNSUPPORTED },
977 { DDF_CONCAT, LEVEL_LINEAR },
978 { DDF_RAID5E, LEVEL_UNSUPPORTED },
979 { DDF_RAID5EE, LEVEL_UNSUPPORTED },
980 { DDF_RAID6, 6},
981 { MAXINT, MAXINT }
982 };
983
984 static int map_num1(struct num_mapping *map, int num)
985 {
986 int i;
987 for (i=0 ; map[i].num1 != MAXINT; i++)
988 if (map[i].num1 == num)
989 break;
990 return map[i].num2;
991 }
992
993 static int all_ff(char *guid)
994 {
995 int i;
996 for (i = 0; i < DDF_GUID_LEN; i++)
997 if (guid[i] != (char)0xff)
998 return 0;
999 return 1;
1000 }
1001
1002 #ifndef MDASSEMBLE
1003 static void print_guid(char *guid, int tstamp)
1004 {
1005 /* A GUIDs are part (or all) ASCII and part binary.
1006 * They tend to be space padded.
1007 * We print the GUID in HEX, then in parentheses add
1008 * any initial ASCII sequence, and a possible
1009 * time stamp from bytes 16-19
1010 */
1011 int l = DDF_GUID_LEN;
1012 int i;
1013
1014 for (i=0 ; i<DDF_GUID_LEN ; i++) {
1015 if ((i&3)==0 && i != 0) printf(":");
1016 printf("%02X", guid[i]&255);
1017 }
1018
1019 printf("\n (");
1020 while (l && guid[l-1] == ' ')
1021 l--;
1022 for (i=0 ; i<l ; i++) {
1023 if (guid[i] >= 0x20 && guid[i] < 0x7f)
1024 fputc(guid[i], stdout);
1025 else
1026 break;
1027 }
1028 if (tstamp) {
1029 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
1030 char tbuf[100];
1031 struct tm *tm;
1032 tm = localtime(&then);
1033 strftime(tbuf, 100, " %D %T",tm);
1034 fputs(tbuf, stdout);
1035 }
1036 printf(")");
1037 }
1038
1039 static void examine_vd(int n, struct ddf_super *sb, char *guid)
1040 {
1041 int crl = sb->conf_rec_len;
1042 struct vcl *vcl;
1043
1044 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
1045 int i;
1046 struct vd_config *vc = &vcl->conf;
1047
1048 if (calc_crc(vc, crl*512) != vc->crc)
1049 continue;
1050 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
1051 continue;
1052
1053 /* Ok, we know about this VD, let's give more details */
1054 printf(" Raid Devices[%d] : %d (", n,
1055 __be16_to_cpu(vc->prim_elmnt_count));
1056 for (i=0; i<__be16_to_cpu(vc->prim_elmnt_count); i++) {
1057 int j;
1058 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1059 for (j=0; j<cnt; j++)
1060 if (vc->phys_refnum[i] == sb->phys->entries[j].refnum)
1061 break;
1062 if (i) printf(" ");
1063 if (j < cnt)
1064 printf("%d", j);
1065 else
1066 printf("--");
1067 }
1068 printf(")\n");
1069 if (vc->chunk_shift != 255)
1070 printf(" Chunk Size[%d] : %d sectors\n", n,
1071 1 << vc->chunk_shift);
1072 printf(" Raid Level[%d] : %s\n", n,
1073 map_num(ddf_level, vc->prl)?:"-unknown-");
1074 if (vc->sec_elmnt_count != 1) {
1075 printf(" Secondary Position[%d] : %d of %d\n", n,
1076 vc->sec_elmnt_seq, vc->sec_elmnt_count);
1077 printf(" Secondary Level[%d] : %s\n", n,
1078 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1079 }
1080 printf(" Device Size[%d] : %llu\n", n,
1081 (unsigned long long)__be64_to_cpu(vc->blocks)/2);
1082 printf(" Array Size[%d] : %llu\n", n,
1083 (unsigned long long)__be64_to_cpu(vc->array_blocks)/2);
1084 }
1085 }
1086
1087 static void examine_vds(struct ddf_super *sb)
1088 {
1089 int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1090 int i;
1091 printf(" Virtual Disks : %d\n", cnt);
1092
1093 for (i=0; i<cnt; i++) {
1094 struct virtual_entry *ve = &sb->virt->entries[i];
1095 printf("\n");
1096 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1097 printf("\n");
1098 printf(" unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1099 printf(" state[%d] : %s, %s%s\n", i,
1100 map_num(ddf_state, ve->state & 7),
1101 (ve->state & 8) ? "Morphing, ": "",
1102 (ve->state & 16)? "Not Consistent" : "Consistent");
1103 printf(" init state[%d] : %s\n", i,
1104 map_num(ddf_init_state, ve->init_state&3));
1105 printf(" access[%d] : %s\n", i,
1106 map_num(ddf_access, (ve->init_state>>6) & 3));
1107 printf(" Name[%d] : %.16s\n", i, ve->name);
1108 examine_vd(i, sb, ve->guid);
1109 }
1110 if (cnt) printf("\n");
1111 }
1112
1113 static void examine_pds(struct ddf_super *sb)
1114 {
1115 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1116 int i;
1117 struct dl *dl;
1118 printf(" Physical Disks : %d\n", cnt);
1119 printf(" Number RefNo Size Device Type/State\n");
1120
1121 for (i=0 ; i<cnt ; i++) {
1122 struct phys_disk_entry *pd = &sb->phys->entries[i];
1123 int type = __be16_to_cpu(pd->type);
1124 int state = __be16_to_cpu(pd->state);
1125
1126 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1127 //printf("\n");
1128 printf(" %3d %08x ", i,
1129 __be32_to_cpu(pd->refnum));
1130 printf("%8lluK ",
1131 (unsigned long long)__be64_to_cpu(pd->config_size)>>1);
1132 for (dl = sb->dlist; dl ; dl = dl->next) {
1133 if (dl->disk.refnum == pd->refnum) {
1134 char *dv = map_dev(dl->major, dl->minor, 0);
1135 if (dv) {
1136 printf("%-15s", dv);
1137 break;
1138 }
1139 }
1140 }
1141 if (!dl)
1142 printf("%15s","");
1143 printf(" %s%s%s%s%s",
1144 (type&2) ? "active":"",
1145 (type&4) ? "Global-Spare":"",
1146 (type&8) ? "spare" : "",
1147 (type&16)? ", foreign" : "",
1148 (type&32)? "pass-through" : "");
1149 printf("/%s%s%s%s%s%s%s",
1150 (state&1)? "Online": "Offline",
1151 (state&2)? ", Failed": "",
1152 (state&4)? ", Rebuilding": "",
1153 (state&8)? ", in-transition": "",
1154 (state&16)? ", SMART-errors": "",
1155 (state&32)? ", Unrecovered-Read-Errors": "",
1156 (state&64)? ", Missing" : "");
1157 printf("\n");
1158 }
1159 }
1160
1161 static void examine_super_ddf(struct supertype *st, char *homehost)
1162 {
1163 struct ddf_super *sb = st->sb;
1164
1165 printf(" Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1166 printf(" Version : %.8s\n", sb->anchor.revision);
1167 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1168 printf("\n");
1169 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1170 printf("\n");
1171 printf(" Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1172 printf(" Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1173 ?"yes" : "no");
1174 examine_vds(sb);
1175 examine_pds(sb);
1176 }
1177
1178 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info);
1179
1180 static void uuid_from_super_ddf(struct supertype *st, int uuid[4]);
1181
1182 static void brief_examine_super_ddf(struct supertype *st, int verbose)
1183 {
1184 /* We just write a generic DDF ARRAY entry
1185 */
1186 struct ddf_super *ddf = st->sb;
1187 struct mdinfo info;
1188 int i;
1189 char nbuf[64];
1190 getinfo_super_ddf(st, &info);
1191 fname_from_uuid(st, &info, nbuf, ':');
1192
1193 for (i=0; i<__be16_to_cpu(ddf->virt->max_vdes); i++) {
1194 struct virtual_entry *ve = &ddf->virt->entries[i];
1195 struct vcl vcl;
1196 char nbuf1[64];
1197 if (all_ff(ve->guid))
1198 continue;
1199 memcpy(vcl.conf.guid, ve->guid, DDF_GUID_LEN);
1200 ddf->currentconf =&vcl;
1201 uuid_from_super_ddf(st, info.uuid);
1202 fname_from_uuid(st, &info, nbuf1, ':');
1203 printf("ARRAY container=%s member=%d UUID=%s\n",
1204 nbuf+5, i, nbuf1+5);
1205 }
1206 printf("ARRAY metadata=ddf UUID=%s\n", nbuf + 5);
1207 }
1208
1209 static void export_examine_super_ddf(struct supertype *st)
1210 {
1211 struct mdinfo info;
1212 char nbuf[64];
1213 getinfo_super_ddf(st, &info);
1214 fname_from_uuid(st, &info, nbuf, ':');
1215 printf("MD_METADATA=ddf\n");
1216 printf("MD_LEVEL=container\n");
1217 printf("MD_UUID=%s\n", nbuf+5);
1218 }
1219
1220
1221 static void detail_super_ddf(struct supertype *st, char *homehost)
1222 {
1223 /* FIXME later
1224 * Could print DDF GUID
1225 * Need to find which array
1226 * If whole, briefly list all arrays
1227 * If one, give name
1228 */
1229 }
1230
1231 static void brief_detail_super_ddf(struct supertype *st)
1232 {
1233 /* FIXME I really need to know which array we are detailing.
1234 * Can that be stored in ddf_super??
1235 */
1236 // struct ddf_super *ddf = st->sb;
1237 struct mdinfo info;
1238 char nbuf[64];
1239 getinfo_super_ddf(st, &info);
1240 fname_from_uuid(st, &info, nbuf,':');
1241 printf(" UUID=%s", nbuf + 5);
1242 }
1243 #endif
1244
1245 static int match_home_ddf(struct supertype *st, char *homehost)
1246 {
1247 /* It matches 'this' host if the controller is a
1248 * Linux-MD controller with vendor_data matching
1249 * the hostname
1250 */
1251 struct ddf_super *ddf = st->sb;
1252 int len = strlen(homehost);
1253
1254 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1255 len < sizeof(ddf->controller.vendor_data) &&
1256 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1257 ddf->controller.vendor_data[len] == 0);
1258 }
1259
1260 #ifndef MDASSEMBLE
1261 static struct vd_config *find_vdcr(struct ddf_super *ddf, int inst)
1262 {
1263 struct vcl *v;
1264
1265 for (v = ddf->conflist; v; v = v->next)
1266 if (inst == v->vcnum)
1267 return &v->conf;
1268 return NULL;
1269 }
1270 #endif
1271
1272 static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1273 {
1274 /* Find the entry in phys_disk which has the given refnum
1275 * and return it's index
1276 */
1277 int i;
1278 for (i=0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1279 if (ddf->phys->entries[i].refnum == phys_refnum)
1280 return i;
1281 return -1;
1282 }
1283
1284 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1285 {
1286 /* The uuid returned here is used for:
1287 * uuid to put into bitmap file (Create, Grow)
1288 * uuid for backup header when saving critical section (Grow)
1289 * comparing uuids when re-adding a device into an array
1290 * In these cases the uuid required is that of the data-array,
1291 * not the device-set.
1292 * uuid to recognise same set when adding a missing device back
1293 * to an array. This is a uuid for the device-set.
1294 *
1295 * For each of these we can make do with a truncated
1296 * or hashed uuid rather than the original, as long as
1297 * everyone agrees.
1298 * In the case of SVD we assume the BVD is of interest,
1299 * though that might be the case if a bitmap were made for
1300 * a mirrored SVD - worry about that later.
1301 * So we need to find the VD configuration record for the
1302 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1303 * The first 16 bytes of the sha1 of these is used.
1304 */
1305 struct ddf_super *ddf = st->sb;
1306 struct vcl *vcl = ddf->currentconf;
1307 char *guid;
1308 char buf[20];
1309 struct sha1_ctx ctx;
1310
1311 if (vcl)
1312 guid = vcl->conf.guid;
1313 else
1314 guid = ddf->anchor.guid;
1315
1316 sha1_init_ctx(&ctx);
1317 sha1_process_bytes(guid, DDF_GUID_LEN, &ctx);
1318 sha1_finish_ctx(&ctx, buf);
1319 memcpy(uuid, buf, 4*4);
1320 }
1321
1322 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info);
1323
1324 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info)
1325 {
1326 struct ddf_super *ddf = st->sb;
1327
1328 if (ddf->currentconf) {
1329 getinfo_super_ddf_bvd(st, info);
1330 return;
1331 }
1332
1333 info->array.raid_disks = __be16_to_cpu(ddf->phys->used_pdes);
1334 info->array.level = LEVEL_CONTAINER;
1335 info->array.layout = 0;
1336 info->array.md_minor = -1;
1337 info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)
1338 (ddf->anchor.guid+16));
1339 info->array.utime = 0;
1340 info->array.chunk_size = 0;
1341
1342
1343 info->disk.major = 0;
1344 info->disk.minor = 0;
1345 if (ddf->dlist) {
1346 info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1347 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1348
1349 info->data_offset = __be64_to_cpu(ddf->phys->
1350 entries[info->disk.raid_disk].
1351 config_size);
1352 info->component_size = ddf->dlist->size - info->data_offset;
1353 } else {
1354 info->disk.number = -1;
1355 info->disk.raid_disk = -1;
1356 // info->disk.raid_disk = find refnum in the table and use index;
1357 }
1358 info->disk.state = (1 << MD_DISK_SYNC);
1359
1360
1361 info->reshape_active = 0;
1362 info->name[0] = 0;
1363
1364 info->array.major_version = -1;
1365 info->array.minor_version = -2;
1366 strcpy(info->text_version, "ddf");
1367 info->safe_mode_delay = 0;
1368
1369 uuid_from_super_ddf(st, info->uuid);
1370
1371 }
1372
1373 static int rlq_to_layout(int rlq, int prl, int raiddisks);
1374
1375 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info)
1376 {
1377 struct ddf_super *ddf = st->sb;
1378 struct vcl *vc = ddf->currentconf;
1379 int cd = ddf->currentdev;
1380 int j;
1381 struct dl *dl;
1382
1383 /* FIXME this returns BVD info - what if we want SVD ?? */
1384
1385 info->array.raid_disks = __be16_to_cpu(vc->conf.prim_elmnt_count);
1386 info->array.level = map_num1(ddf_level_num, vc->conf.prl);
1387 info->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
1388 info->array.raid_disks);
1389 info->array.md_minor = -1;
1390 info->array.ctime = DECADE +
1391 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
1392 info->array.utime = DECADE + __be32_to_cpu(vc->conf.timestamp);
1393 info->array.chunk_size = 512 << vc->conf.chunk_shift;
1394 info->custom_array_size = 0;
1395
1396 if (cd >= 0 && cd < ddf->mppe) {
1397 info->data_offset = __be64_to_cpu(vc->lba_offset[cd]);
1398 if (vc->block_sizes)
1399 info->component_size = vc->block_sizes[cd];
1400 else
1401 info->component_size = __be64_to_cpu(vc->conf.blocks);
1402 }
1403
1404 for (dl = ddf->dlist; dl ; dl = dl->next)
1405 if (dl->raiddisk == info->disk.raid_disk)
1406 break;
1407 info->disk.major = 0;
1408 info->disk.minor = 0;
1409 if (dl) {
1410 info->disk.major = dl->major;
1411 info->disk.minor = dl->minor;
1412 }
1413 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1414 // info->disk.raid_disk = find refnum in the table and use index;
1415 // info->disk.state = ???;
1416
1417 info->container_member = ddf->currentconf->vcnum;
1418
1419 info->resync_start = 0;
1420 if (!(ddf->virt->entries[info->container_member].state
1421 & DDF_state_inconsistent) &&
1422 (ddf->virt->entries[info->container_member].init_state
1423 & DDF_initstate_mask)
1424 == DDF_init_full)
1425 info->resync_start = ~0ULL;
1426
1427 uuid_from_super_ddf(st, info->uuid);
1428
1429 info->container_member = atoi(st->subarray);
1430 info->array.major_version = -1;
1431 info->array.minor_version = -2;
1432 sprintf(info->text_version, "/%s/%s",
1433 devnum2devname(st->container_dev),
1434 st->subarray);
1435 info->safe_mode_delay = 200;
1436
1437 memcpy(info->name, ddf->virt->entries[info->container_member].name, 16);
1438 info->name[16]=0;
1439 for(j=0; j<16; j++)
1440 if (info->name[j] == ' ')
1441 info->name[j] = 0;
1442 }
1443
1444
1445 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1446 char *update,
1447 char *devname, int verbose,
1448 int uuid_set, char *homehost)
1449 {
1450 /* For 'assemble' and 'force' we need to return non-zero if any
1451 * change was made. For others, the return value is ignored.
1452 * Update options are:
1453 * force-one : This device looks a bit old but needs to be included,
1454 * update age info appropriately.
1455 * assemble: clear any 'faulty' flag to allow this device to
1456 * be assembled.
1457 * force-array: Array is degraded but being forced, mark it clean
1458 * if that will be needed to assemble it.
1459 *
1460 * newdev: not used ????
1461 * grow: Array has gained a new device - this is currently for
1462 * linear only
1463 * resync: mark as dirty so a resync will happen.
1464 * uuid: Change the uuid of the array to match what is given
1465 * homehost: update the recorded homehost
1466 * name: update the name - preserving the homehost
1467 * _reshape_progress: record new reshape_progress position.
1468 *
1469 * Following are not relevant for this version:
1470 * sparc2.2 : update from old dodgey metadata
1471 * super-minor: change the preferred_minor number
1472 * summaries: update redundant counters.
1473 */
1474 int rv = 0;
1475 // struct ddf_super *ddf = st->sb;
1476 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1477 // struct virtual_entry *ve = find_ve(ddf);
1478
1479 /* we don't need to handle "force-*" or "assemble" as
1480 * there is no need to 'trick' the kernel. We the metadata is
1481 * first updated to activate the array, all the implied modifications
1482 * will just happen.
1483 */
1484
1485 if (strcmp(update, "grow") == 0) {
1486 /* FIXME */
1487 }
1488 if (strcmp(update, "resync") == 0) {
1489 // info->resync_checkpoint = 0;
1490 }
1491 /* We ignore UUID updates as they make even less sense
1492 * with DDF
1493 */
1494 if (strcmp(update, "homehost") == 0) {
1495 /* homehost is stored in controller->vendor_data,
1496 * or it is when we are the vendor
1497 */
1498 // if (info->vendor_is_local)
1499 // strcpy(ddf->controller.vendor_data, homehost);
1500 }
1501 if (strcmp(update, "name") == 0) {
1502 /* name is stored in virtual_entry->name */
1503 // memset(ve->name, ' ', 16);
1504 // strncpy(ve->name, info->name, 16);
1505 }
1506 if (strcmp(update, "_reshape_progress") == 0) {
1507 /* We don't support reshape yet */
1508 }
1509
1510 // update_all_csum(ddf);
1511
1512 return rv;
1513 }
1514
1515 __u32 random32(void)
1516 {
1517 __u32 rv;
1518 int rfd = open("/dev/urandom", O_RDONLY);
1519 if (rfd < 0 || read(rfd, &rv, 4) != 4)
1520 rv = random();
1521 if (rfd >= 0)
1522 close(rfd);
1523 return rv;
1524 }
1525
1526 static void make_header_guid(char *guid)
1527 {
1528 __u32 stamp;
1529 /* Create a DDF Header of Virtual Disk GUID */
1530
1531 /* 24 bytes of fiction required.
1532 * first 8 are a 'vendor-id' - "Linux-MD"
1533 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1534 * Remaining 8 random number plus timestamp
1535 */
1536 memcpy(guid, T10, sizeof(T10));
1537 stamp = __cpu_to_be32(0xdeadbeef);
1538 memcpy(guid+8, &stamp, 4);
1539 stamp = __cpu_to_be32(0);
1540 memcpy(guid+12, &stamp, 4);
1541 stamp = __cpu_to_be32(time(0) - DECADE);
1542 memcpy(guid+16, &stamp, 4);
1543 stamp = random32();
1544 memcpy(guid+20, &stamp, 4);
1545 }
1546
1547 static int init_super_ddf_bvd(struct supertype *st,
1548 mdu_array_info_t *info,
1549 unsigned long long size,
1550 char *name, char *homehost,
1551 int *uuid);
1552
1553 static int init_super_ddf(struct supertype *st,
1554 mdu_array_info_t *info,
1555 unsigned long long size, char *name, char *homehost,
1556 int *uuid)
1557 {
1558 /* This is primarily called by Create when creating a new array.
1559 * We will then get add_to_super called for each component, and then
1560 * write_init_super called to write it out to each device.
1561 * For DDF, Create can create on fresh devices or on a pre-existing
1562 * array.
1563 * To create on a pre-existing array a different method will be called.
1564 * This one is just for fresh drives.
1565 *
1566 * We need to create the entire 'ddf' structure which includes:
1567 * DDF headers - these are easy.
1568 * Controller data - a Sector describing this controller .. not that
1569 * this is a controller exactly.
1570 * Physical Disk Record - one entry per device, so
1571 * leave plenty of space.
1572 * Virtual Disk Records - again, just leave plenty of space.
1573 * This just lists VDs, doesn't give details
1574 * Config records - describes the VDs that use this disk
1575 * DiskData - describes 'this' device.
1576 * BadBlockManagement - empty
1577 * Diag Space - empty
1578 * Vendor Logs - Could we put bitmaps here?
1579 *
1580 */
1581 struct ddf_super *ddf;
1582 char hostname[17];
1583 int hostlen;
1584 int max_phys_disks, max_virt_disks;
1585 unsigned long long sector;
1586 int clen;
1587 int i;
1588 int pdsize, vdsize;
1589 struct phys_disk *pd;
1590 struct virtual_disk *vd;
1591
1592 if (!info) {
1593 st->sb = NULL;
1594 return 0;
1595 }
1596 if (st->sb)
1597 return init_super_ddf_bvd(st, info, size, name, homehost,
1598 uuid);
1599
1600 if (posix_memalign((void**)&ddf, 512, sizeof(*ddf)) != 0) {
1601 fprintf(stderr, Name ": %s could not allocate superblock\n", __func__);
1602 return 0;
1603 }
1604 memset(ddf, 0, sizeof(*ddf));
1605 ddf->dlist = NULL; /* no physical disks yet */
1606 ddf->conflist = NULL; /* No virtual disks yet */
1607
1608 /* At least 32MB *must* be reserved for the ddf. So let's just
1609 * start 32MB from the end, and put the primary header there.
1610 * Don't do secondary for now.
1611 * We don't know exactly where that will be yet as it could be
1612 * different on each device. To just set up the lengths.
1613 *
1614 */
1615
1616 ddf->anchor.magic = DDF_HEADER_MAGIC;
1617 make_header_guid(ddf->anchor.guid);
1618
1619 memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
1620 ddf->anchor.seq = __cpu_to_be32(1);
1621 ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1622 ddf->anchor.openflag = 0xFF;
1623 ddf->anchor.foreignflag = 0;
1624 ddf->anchor.enforcegroups = 0; /* Is this best?? */
1625 ddf->anchor.pad0 = 0xff;
1626 memset(ddf->anchor.pad1, 0xff, 12);
1627 memset(ddf->anchor.header_ext, 0xff, 32);
1628 ddf->anchor.primary_lba = ~(__u64)0;
1629 ddf->anchor.secondary_lba = ~(__u64)0;
1630 ddf->anchor.type = DDF_HEADER_ANCHOR;
1631 memset(ddf->anchor.pad2, 0xff, 3);
1632 ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1633 ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1634 of 32M reserved.. */
1635 max_phys_disks = 1023; /* Should be enough */
1636 ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1637 max_virt_disks = 255;
1638 ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1639 ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1640 ddf->max_part = 64;
1641 ddf->mppe = 256;
1642 ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
1643 ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1644 ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe);
1645 memset(ddf->anchor.pad3, 0xff, 54);
1646 /* controller sections is one sector long immediately
1647 * after the ddf header */
1648 sector = 1;
1649 ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1650 ddf->anchor.controller_section_length = __cpu_to_be32(1);
1651 sector += 1;
1652
1653 /* phys is 8 sectors after that */
1654 pdsize = ROUND_UP(sizeof(struct phys_disk) +
1655 sizeof(struct phys_disk_entry)*max_phys_disks,
1656 512);
1657 switch(pdsize/512) {
1658 case 2: case 8: case 32: case 128: case 512: break;
1659 default: abort();
1660 }
1661 ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1662 ddf->anchor.phys_section_length =
1663 __cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1664 sector += pdsize/512;
1665
1666 /* virt is another 32 sectors */
1667 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1668 sizeof(struct virtual_entry) * max_virt_disks,
1669 512);
1670 switch(vdsize/512) {
1671 case 2: case 8: case 32: case 128: case 512: break;
1672 default: abort();
1673 }
1674 ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1675 ddf->anchor.virt_section_length =
1676 __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1677 sector += vdsize/512;
1678
1679 clen = ddf->conf_rec_len * (ddf->max_part+1);
1680 ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1681 ddf->anchor.config_section_length = __cpu_to_be32(clen);
1682 sector += clen;
1683
1684 ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1685 ddf->anchor.data_section_length = __cpu_to_be32(1);
1686 sector += 1;
1687
1688 ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1689 ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1690 ddf->anchor.diag_space_length = __cpu_to_be32(0);
1691 ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1692 ddf->anchor.vendor_length = __cpu_to_be32(0);
1693 ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1694
1695 memset(ddf->anchor.pad4, 0xff, 256);
1696
1697 memcpy(&ddf->primary, &ddf->anchor, 512);
1698 memcpy(&ddf->secondary, &ddf->anchor, 512);
1699
1700 ddf->primary.openflag = 1; /* I guess.. */
1701 ddf->primary.type = DDF_HEADER_PRIMARY;
1702
1703 ddf->secondary.openflag = 1; /* I guess.. */
1704 ddf->secondary.type = DDF_HEADER_SECONDARY;
1705
1706 ddf->active = &ddf->primary;
1707
1708 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1709
1710 /* 24 more bytes of fiction required.
1711 * first 8 are a 'vendor-id' - "Linux-MD"
1712 * Remaining 16 are serial number.... maybe a hostname would do?
1713 */
1714 memcpy(ddf->controller.guid, T10, sizeof(T10));
1715 gethostname(hostname, sizeof(hostname));
1716 hostname[sizeof(hostname) - 1] = 0;
1717 hostlen = strlen(hostname);
1718 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1719 for (i = strlen(T10) ; i+hostlen < 24; i++)
1720 ddf->controller.guid[i] = ' ';
1721
1722 ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1723 ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1724 ddf->controller.type.sub_vendor_id = 0;
1725 ddf->controller.type.sub_device_id = 0;
1726 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1727 memset(ddf->controller.pad, 0xff, 8);
1728 memset(ddf->controller.vendor_data, 0xff, 448);
1729 if (homehost && strlen(homehost) < 440)
1730 strcpy((char*)ddf->controller.vendor_data, homehost);
1731
1732 if (posix_memalign((void**)&pd, 512, pdsize) != 0) {
1733 fprintf(stderr, Name ": %s could not allocate pd\n", __func__);
1734 return 0;
1735 }
1736 ddf->phys = pd;
1737 ddf->pdsize = pdsize;
1738
1739 memset(pd, 0xff, pdsize);
1740 memset(pd, 0, sizeof(*pd));
1741 pd->magic = DDF_PHYS_DATA_MAGIC;
1742 pd->used_pdes = __cpu_to_be16(0);
1743 pd->max_pdes = __cpu_to_be16(max_phys_disks);
1744 memset(pd->pad, 0xff, 52);
1745
1746 if (posix_memalign((void**)&vd, 512, vdsize) != 0) {
1747 fprintf(stderr, Name ": %s could not allocate vd\n", __func__);
1748 return 0;
1749 }
1750 ddf->virt = vd;
1751 ddf->vdsize = vdsize;
1752 memset(vd, 0, vdsize);
1753 vd->magic = DDF_VIRT_RECORDS_MAGIC;
1754 vd->populated_vdes = __cpu_to_be16(0);
1755 vd->max_vdes = __cpu_to_be16(max_virt_disks);
1756 memset(vd->pad, 0xff, 52);
1757
1758 for (i=0; i<max_virt_disks; i++)
1759 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1760
1761 st->sb = ddf;
1762 ddf->updates_pending = 1;
1763 return 1;
1764 }
1765
1766 static int chunk_to_shift(int chunksize)
1767 {
1768 return ffs(chunksize/512)-1;
1769 }
1770
1771 static int level_to_prl(int level)
1772 {
1773 switch (level) {
1774 case LEVEL_LINEAR: return DDF_CONCAT;
1775 case 0: return DDF_RAID0;
1776 case 1: return DDF_RAID1;
1777 case 4: return DDF_RAID4;
1778 case 5: return DDF_RAID5;
1779 case 6: return DDF_RAID6;
1780 default: return -1;
1781 }
1782 }
1783 static int layout_to_rlq(int level, int layout, int raiddisks)
1784 {
1785 switch(level) {
1786 case 0:
1787 return DDF_RAID0_SIMPLE;
1788 case 1:
1789 switch(raiddisks) {
1790 case 2: return DDF_RAID1_SIMPLE;
1791 case 3: return DDF_RAID1_MULTI;
1792 default: return -1;
1793 }
1794 case 4:
1795 switch(layout) {
1796 case 0: return DDF_RAID4_N;
1797 }
1798 break;
1799 case 5:
1800 switch(layout) {
1801 case ALGORITHM_LEFT_ASYMMETRIC:
1802 return DDF_RAID5_N_RESTART;
1803 case ALGORITHM_RIGHT_ASYMMETRIC:
1804 return DDF_RAID5_0_RESTART;
1805 case ALGORITHM_LEFT_SYMMETRIC:
1806 return DDF_RAID5_N_CONTINUE;
1807 case ALGORITHM_RIGHT_SYMMETRIC:
1808 return -1; /* not mentioned in standard */
1809 }
1810 case 6:
1811 switch(layout) {
1812 case ALGORITHM_ROTATING_N_RESTART:
1813 return DDF_RAID5_N_RESTART;
1814 case ALGORITHM_ROTATING_ZERO_RESTART:
1815 return DDF_RAID6_0_RESTART;
1816 case ALGORITHM_ROTATING_N_CONTINUE:
1817 return DDF_RAID5_N_CONTINUE;
1818 }
1819 }
1820 return -1;
1821 }
1822
1823 static int rlq_to_layout(int rlq, int prl, int raiddisks)
1824 {
1825 switch(prl) {
1826 case DDF_RAID0:
1827 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1828 case DDF_RAID1:
1829 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1830 on raiddisks*/
1831 case DDF_RAID4:
1832 switch(rlq) {
1833 case DDF_RAID4_N:
1834 return 0;
1835 default:
1836 /* not supported */
1837 return -1; /* FIXME this isn't checked */
1838 }
1839 case DDF_RAID5:
1840 switch(rlq) {
1841 case DDF_RAID5_N_RESTART:
1842 return ALGORITHM_LEFT_ASYMMETRIC;
1843 case DDF_RAID5_0_RESTART:
1844 return ALGORITHM_RIGHT_ASYMMETRIC;
1845 case DDF_RAID5_N_CONTINUE:
1846 return ALGORITHM_LEFT_SYMMETRIC;
1847 default:
1848 return -1;
1849 }
1850 case DDF_RAID6:
1851 switch(rlq) {
1852 case DDF_RAID5_N_RESTART:
1853 return ALGORITHM_ROTATING_N_RESTART;
1854 case DDF_RAID6_0_RESTART:
1855 return ALGORITHM_ROTATING_ZERO_RESTART;
1856 case DDF_RAID5_N_CONTINUE:
1857 return ALGORITHM_ROTATING_N_CONTINUE;
1858 default:
1859 return -1;
1860 }
1861 }
1862 return -1;
1863 }
1864
1865 #ifndef MDASSEMBLE
1866 struct extent {
1867 unsigned long long start, size;
1868 };
1869 static int cmp_extent(const void *av, const void *bv)
1870 {
1871 const struct extent *a = av;
1872 const struct extent *b = bv;
1873 if (a->start < b->start)
1874 return -1;
1875 if (a->start > b->start)
1876 return 1;
1877 return 0;
1878 }
1879
1880 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
1881 {
1882 /* find a list of used extents on the give physical device
1883 * (dnum) of the given ddf.
1884 * Return a malloced array of 'struct extent'
1885
1886 FIXME ignore DDF_Legacy devices?
1887
1888 */
1889 struct extent *rv;
1890 int n = 0;
1891 int i, j;
1892
1893 rv = malloc(sizeof(struct extent) * (ddf->max_part + 2));
1894 if (!rv)
1895 return NULL;
1896
1897 for (i = 0; i < ddf->max_part; i++) {
1898 struct vcl *v = dl->vlist[i];
1899 if (v == NULL)
1900 continue;
1901 for (j=0; j < v->conf.prim_elmnt_count; j++)
1902 if (v->conf.phys_refnum[j] == dl->disk.refnum) {
1903 /* This device plays role 'j' in 'v'. */
1904 rv[n].start = __be64_to_cpu(v->lba_offset[j]);
1905 rv[n].size = __be64_to_cpu(v->conf.blocks);
1906 n++;
1907 break;
1908 }
1909 }
1910 qsort(rv, n, sizeof(*rv), cmp_extent);
1911
1912 rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
1913 rv[n].size = 0;
1914 return rv;
1915 }
1916 #endif
1917
1918 static int init_super_ddf_bvd(struct supertype *st,
1919 mdu_array_info_t *info,
1920 unsigned long long size,
1921 char *name, char *homehost,
1922 int *uuid)
1923 {
1924 /* We are creating a BVD inside a pre-existing container.
1925 * so st->sb is already set.
1926 * We need to create a new vd_config and a new virtual_entry
1927 */
1928 struct ddf_super *ddf = st->sb;
1929 int venum;
1930 struct virtual_entry *ve;
1931 struct vcl *vcl;
1932 struct vd_config *vc;
1933
1934 if (__be16_to_cpu(ddf->virt->populated_vdes)
1935 >= __be16_to_cpu(ddf->virt->max_vdes)) {
1936 fprintf(stderr, Name": This ddf already has the "
1937 "maximum of %d virtual devices\n",
1938 __be16_to_cpu(ddf->virt->max_vdes));
1939 return 0;
1940 }
1941
1942 for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1943 if (all_ff(ddf->virt->entries[venum].guid))
1944 break;
1945 if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1946 fprintf(stderr, Name ": Cannot find spare slot for "
1947 "virtual disk - DDF is corrupt\n");
1948 return 0;
1949 }
1950 ve = &ddf->virt->entries[venum];
1951
1952 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1953 * timestamp, random number
1954 */
1955 make_header_guid(ve->guid);
1956 ve->unit = __cpu_to_be16(info->md_minor);
1957 ve->pad0 = 0xFFFF;
1958 ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1959 ve->type = 0;
1960 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
1961 if (info->state & 1) /* clean */
1962 ve->init_state = DDF_init_full;
1963 else
1964 ve->init_state = DDF_init_not;
1965
1966 memset(ve->pad1, 0xff, 14);
1967 memset(ve->name, ' ', 16);
1968 if (name)
1969 strncpy(ve->name, name, 16);
1970 ddf->virt->populated_vdes =
1971 __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
1972
1973 /* Now create a new vd_config */
1974 if (posix_memalign((void**)&vcl, 512,
1975 (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512)) != 0) {
1976 fprintf(stderr, Name ": %s could not allocate vd_config\n", __func__);
1977 return 0;
1978 }
1979 vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
1980 vcl->vcnum = venum;
1981 sprintf(st->subarray, "%d", venum);
1982 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1983
1984 vc = &vcl->conf;
1985
1986 vc->magic = DDF_VD_CONF_MAGIC;
1987 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
1988 vc->timestamp = __cpu_to_be32(time(0)-DECADE);
1989 vc->seqnum = __cpu_to_be32(1);
1990 memset(vc->pad0, 0xff, 24);
1991 vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
1992 vc->chunk_shift = chunk_to_shift(info->chunk_size);
1993 vc->prl = level_to_prl(info->level);
1994 vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
1995 vc->sec_elmnt_count = 1;
1996 vc->sec_elmnt_seq = 0;
1997 vc->srl = 0;
1998 vc->blocks = __cpu_to_be64(info->size * 2);
1999 vc->array_blocks = __cpu_to_be64(
2000 calc_array_size(info->level, info->raid_disks, info->layout,
2001 info->chunk_size, info->size*2));
2002 memset(vc->pad1, 0xff, 8);
2003 vc->spare_refs[0] = 0xffffffff;
2004 vc->spare_refs[1] = 0xffffffff;
2005 vc->spare_refs[2] = 0xffffffff;
2006 vc->spare_refs[3] = 0xffffffff;
2007 vc->spare_refs[4] = 0xffffffff;
2008 vc->spare_refs[5] = 0xffffffff;
2009 vc->spare_refs[6] = 0xffffffff;
2010 vc->spare_refs[7] = 0xffffffff;
2011 memset(vc->cache_pol, 0, 8);
2012 vc->bg_rate = 0x80;
2013 memset(vc->pad2, 0xff, 3);
2014 memset(vc->pad3, 0xff, 52);
2015 memset(vc->pad4, 0xff, 192);
2016 memset(vc->v0, 0xff, 32);
2017 memset(vc->v1, 0xff, 32);
2018 memset(vc->v2, 0xff, 16);
2019 memset(vc->v3, 0xff, 16);
2020 memset(vc->vendor, 0xff, 32);
2021
2022 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
2023 memset(vc->phys_refnum+ddf->mppe, 0x00, 8*ddf->mppe);
2024
2025 vcl->next = ddf->conflist;
2026 ddf->conflist = vcl;
2027 ddf->currentconf = vcl;
2028 ddf->updates_pending = 1;
2029 return 1;
2030 }
2031
2032 #ifndef MDASSEMBLE
2033 static void add_to_super_ddf_bvd(struct supertype *st,
2034 mdu_disk_info_t *dk, int fd, char *devname)
2035 {
2036 /* fd and devname identify a device with-in the ddf container (st).
2037 * dk identifies a location in the new BVD.
2038 * We need to find suitable free space in that device and update
2039 * the phys_refnum and lba_offset for the newly created vd_config.
2040 * We might also want to update the type in the phys_disk
2041 * section.
2042 *
2043 * Alternately: fd == -1 and we have already chosen which device to
2044 * use and recorded in dlist->raid_disk;
2045 */
2046 struct dl *dl;
2047 struct ddf_super *ddf = st->sb;
2048 struct vd_config *vc;
2049 __u64 *lba_offset;
2050 int working;
2051 int i;
2052 unsigned long long blocks, pos, esize;
2053 struct extent *ex;
2054
2055 if (fd == -1) {
2056 for (dl = ddf->dlist; dl ; dl = dl->next)
2057 if (dl->raiddisk == dk->raid_disk)
2058 break;
2059 } else {
2060 for (dl = ddf->dlist; dl ; dl = dl->next)
2061 if (dl->major == dk->major &&
2062 dl->minor == dk->minor)
2063 break;
2064 }
2065 if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
2066 return;
2067
2068 vc = &ddf->currentconf->conf;
2069 lba_offset = ddf->currentconf->lba_offset;
2070
2071 ex = get_extents(ddf, dl);
2072 if (!ex)
2073 return;
2074
2075 i = 0; pos = 0;
2076 blocks = __be64_to_cpu(vc->blocks);
2077 if (ddf->currentconf->block_sizes)
2078 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
2079
2080 do {
2081 esize = ex[i].start - pos;
2082 if (esize >= blocks)
2083 break;
2084 pos = ex[i].start + ex[i].size;
2085 i++;
2086 } while (ex[i-1].size);
2087
2088 free(ex);
2089 if (esize < blocks)
2090 return;
2091
2092 ddf->currentdev = dk->raid_disk;
2093 vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
2094 lba_offset[dk->raid_disk] = __cpu_to_be64(pos);
2095
2096 for (i=0; i < ddf->max_part ; i++)
2097 if (dl->vlist[i] == NULL)
2098 break;
2099 if (i == ddf->max_part)
2100 return;
2101 dl->vlist[i] = ddf->currentconf;
2102
2103 if (fd >= 0)
2104 dl->fd = fd;
2105 if (devname)
2106 dl->devname = devname;
2107
2108 /* Check how many working raid_disks, and if we can mark
2109 * array as optimal yet
2110 */
2111 working = 0;
2112
2113 for (i=0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
2114 if (vc->phys_refnum[i] != 0xffffffff)
2115 working++;
2116
2117 /* Find which virtual_entry */
2118 i = ddf->currentconf->vcnum;
2119 if (working == __be16_to_cpu(vc->prim_elmnt_count))
2120 ddf->virt->entries[i].state =
2121 (ddf->virt->entries[i].state & ~DDF_state_mask)
2122 | DDF_state_optimal;
2123
2124 if (vc->prl == DDF_RAID6 &&
2125 working+1 == __be16_to_cpu(vc->prim_elmnt_count))
2126 ddf->virt->entries[i].state =
2127 (ddf->virt->entries[i].state & ~DDF_state_mask)
2128 | DDF_state_part_optimal;
2129
2130 ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
2131 ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
2132 ddf->updates_pending = 1;
2133 }
2134
2135 /* add a device to a container, either while creating it or while
2136 * expanding a pre-existing container
2137 */
2138 static int add_to_super_ddf(struct supertype *st,
2139 mdu_disk_info_t *dk, int fd, char *devname)
2140 {
2141 struct ddf_super *ddf = st->sb;
2142 struct dl *dd;
2143 time_t now;
2144 struct tm *tm;
2145 unsigned long long size;
2146 struct phys_disk_entry *pde;
2147 int n, i;
2148 struct stat stb;
2149
2150 if (ddf->currentconf) {
2151 add_to_super_ddf_bvd(st, dk, fd, devname);
2152 return 0;
2153 }
2154
2155 /* This is device numbered dk->number. We need to create
2156 * a phys_disk entry and a more detailed disk_data entry.
2157 */
2158 fstat(fd, &stb);
2159 if (posix_memalign((void**)&dd, 512,
2160 sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part) != 0) {
2161 fprintf(stderr, Name
2162 ": %s could allocate buffer for new disk, aborting\n",
2163 __func__);
2164 return 1;
2165 }
2166 dd->major = major(stb.st_rdev);
2167 dd->minor = minor(stb.st_rdev);
2168 dd->devname = devname;
2169 dd->fd = fd;
2170 dd->spare = NULL;
2171
2172 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
2173 now = time(0);
2174 tm = localtime(&now);
2175 sprintf(dd->disk.guid, "%8s%04d%02d%02d",
2176 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
2177 *(__u32*)(dd->disk.guid + 16) = random32();
2178 *(__u32*)(dd->disk.guid + 20) = random32();
2179
2180 do {
2181 /* Cannot be bothered finding a CRC of some irrelevant details*/
2182 dd->disk.refnum = random32();
2183 for (i = __be16_to_cpu(ddf->active->max_pd_entries) - 1;
2184 i >= 0; i--)
2185 if (ddf->phys->entries[i].refnum == dd->disk.refnum)
2186 break;
2187 } while (i >= 0);
2188
2189 dd->disk.forced_ref = 1;
2190 dd->disk.forced_guid = 1;
2191 memset(dd->disk.vendor, ' ', 32);
2192 memcpy(dd->disk.vendor, "Linux", 5);
2193 memset(dd->disk.pad, 0xff, 442);
2194 for (i = 0; i < ddf->max_part ; i++)
2195 dd->vlist[i] = NULL;
2196
2197 n = __be16_to_cpu(ddf->phys->used_pdes);
2198 pde = &ddf->phys->entries[n];
2199 dd->pdnum = n;
2200
2201 if (st->update_tail) {
2202 int len = (sizeof(struct phys_disk) +
2203 sizeof(struct phys_disk_entry));
2204 struct phys_disk *pd;
2205
2206 pd = malloc(len);
2207 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2208 pd->used_pdes = __cpu_to_be16(n);
2209 pde = &pd->entries[0];
2210 dd->mdupdate = pd;
2211 } else {
2212 n++;
2213 ddf->phys->used_pdes = __cpu_to_be16(n);
2214 }
2215
2216 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2217 pde->refnum = dd->disk.refnum;
2218 pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2219 pde->state = __cpu_to_be16(DDF_Online);
2220 get_dev_size(fd, NULL, &size);
2221 /* We are required to reserve 32Meg, and record the size in sectors */
2222 pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
2223 sprintf(pde->path, "%17.17s","Information: nil") ;
2224 memset(pde->pad, 0xff, 6);
2225
2226 dd->size = size >> 9;
2227 if (st->update_tail) {
2228 dd->next = ddf->add_list;
2229 ddf->add_list = dd;
2230 } else {
2231 dd->next = ddf->dlist;
2232 ddf->dlist = dd;
2233 ddf->updates_pending = 1;
2234 }
2235
2236 return 0;
2237 }
2238
2239 /*
2240 * This is the write_init_super method for a ddf container. It is
2241 * called when creating a container or adding another device to a
2242 * container.
2243 */
2244
2245 static unsigned char null_conf[4096+512];
2246
2247 static int __write_init_super_ddf(struct supertype *st, int do_close)
2248 {
2249
2250 struct ddf_super *ddf = st->sb;
2251 int i;
2252 struct dl *d;
2253 int n_config;
2254 int conf_size;
2255 int attempts = 0;
2256 int successes = 0;
2257 unsigned long long size, sector;
2258
2259 /* try to write updated metadata,
2260 * if we catch a failure move on to the next disk
2261 */
2262 for (d = ddf->dlist; d; d=d->next) {
2263 int fd = d->fd;
2264
2265 if (fd < 0)
2266 continue;
2267
2268 attempts++;
2269 /* We need to fill in the primary, (secondary) and workspace
2270 * lba's in the headers, set their checksums,
2271 * Also checksum phys, virt....
2272 *
2273 * Then write everything out, finally the anchor is written.
2274 */
2275 get_dev_size(fd, NULL, &size);
2276 size /= 512;
2277 ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
2278 ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
2279 ddf->anchor.seq = __cpu_to_be32(1);
2280 memcpy(&ddf->primary, &ddf->anchor, 512);
2281 memcpy(&ddf->secondary, &ddf->anchor, 512);
2282
2283 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
2284 ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
2285 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
2286
2287 ddf->primary.openflag = 0;
2288 ddf->primary.type = DDF_HEADER_PRIMARY;
2289
2290 ddf->secondary.openflag = 0;
2291 ddf->secondary.type = DDF_HEADER_SECONDARY;
2292
2293 ddf->primary.crc = calc_crc(&ddf->primary, 512);
2294 ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
2295
2296 sector = size - 16*1024*2;
2297 lseek64(fd, sector<<9, 0);
2298 if (write(fd, &ddf->primary, 512) < 0)
2299 continue;
2300
2301 ddf->controller.crc = calc_crc(&ddf->controller, 512);
2302 if (write(fd, &ddf->controller, 512) < 0)
2303 continue;
2304
2305 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
2306
2307 if (write(fd, ddf->phys, ddf->pdsize) < 0)
2308 continue;
2309
2310 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
2311 if (write(fd, ddf->virt, ddf->vdsize) < 0)
2312 continue;
2313
2314 /* Now write lots of config records. */
2315 n_config = ddf->max_part;
2316 conf_size = ddf->conf_rec_len * 512;
2317 for (i = 0 ; i <= n_config ; i++) {
2318 struct vcl *c = d->vlist[i];
2319 if (i == n_config)
2320 c = (struct vcl*)d->spare;
2321
2322 if (c) {
2323 c->conf.crc = calc_crc(&c->conf, conf_size);
2324 if (write(fd, &c->conf, conf_size) < 0)
2325 break;
2326 } else {
2327 char *null_aligned = (char*)((((unsigned long)null_conf)+511)&~511UL);
2328 if (null_conf[0] != 0xff)
2329 memset(null_conf, 0xff, sizeof(null_conf));
2330 int togo = conf_size;
2331 while (togo > sizeof(null_conf)-512) {
2332 if (write(fd, null_aligned, sizeof(null_conf)-512) < 0)
2333 break;
2334 togo -= sizeof(null_conf)-512;
2335 }
2336 if (write(fd, null_aligned, togo) < 0)
2337 break;
2338 }
2339 }
2340 if (i <= n_config)
2341 continue;
2342 d->disk.crc = calc_crc(&d->disk, 512);
2343 if (write(fd, &d->disk, 512) < 0)
2344 continue;
2345
2346 /* Maybe do the same for secondary */
2347
2348 lseek64(fd, (size-1)*512, SEEK_SET);
2349 if (write(fd, &ddf->anchor, 512) < 0)
2350 continue;
2351 successes++;
2352 }
2353
2354 if (do_close)
2355 for (d = ddf->dlist; d; d=d->next) {
2356 close(d->fd);
2357 d->fd = -1;
2358 }
2359
2360 return attempts != successes;
2361 }
2362
2363 static int write_init_super_ddf(struct supertype *st)
2364 {
2365
2366 if (st->update_tail) {
2367 /* queue the virtual_disk and vd_config as metadata updates */
2368 struct virtual_disk *vd;
2369 struct vd_config *vc;
2370 struct ddf_super *ddf = st->sb;
2371 int len;
2372
2373 if (!ddf->currentconf) {
2374 int len = (sizeof(struct phys_disk) +
2375 sizeof(struct phys_disk_entry));
2376
2377 /* adding a disk to the container. */
2378 if (!ddf->add_list)
2379 return 0;
2380
2381 append_metadata_update(st, ddf->add_list->mdupdate, len);
2382 ddf->add_list->mdupdate = NULL;
2383 return 0;
2384 }
2385
2386 /* Newly created VD */
2387
2388 /* First the virtual disk. We have a slightly fake header */
2389 len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry);
2390 vd = malloc(len);
2391 *vd = *ddf->virt;
2392 vd->entries[0] = ddf->virt->entries[ddf->currentconf->vcnum];
2393 vd->populated_vdes = __cpu_to_be16(ddf->currentconf->vcnum);
2394 append_metadata_update(st, vd, len);
2395
2396 /* Then the vd_config */
2397 len = ddf->conf_rec_len * 512;
2398 vc = malloc(len);
2399 memcpy(vc, &ddf->currentconf->conf, len);
2400 append_metadata_update(st, vc, len);
2401
2402 /* FIXME I need to close the fds! */
2403 return 0;
2404 } else
2405 return __write_init_super_ddf(st, 1);
2406 }
2407
2408 #endif
2409
2410 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize)
2411 {
2412 /* We must reserve the last 32Meg */
2413 if (devsize <= 32*1024*2)
2414 return 0;
2415 return devsize - 32*1024*2;
2416 }
2417
2418 #ifndef MDASSEMBLE
2419
2420 static int reserve_space(struct supertype *st, int raiddisks,
2421 unsigned long long size, int chunk,
2422 unsigned long long *freesize)
2423 {
2424 /* Find 'raiddisks' spare extents at least 'size' big (but
2425 * only caring about multiples of 'chunk') and remember
2426 * them.
2427 * If the cannot be found, fail.
2428 */
2429 struct dl *dl;
2430 struct ddf_super *ddf = st->sb;
2431 int cnt = 0;
2432
2433 for (dl = ddf->dlist; dl ; dl=dl->next) {
2434 dl->raiddisk = -1;
2435 dl->esize = 0;
2436 }
2437 /* Now find largest extent on each device */
2438 for (dl = ddf->dlist ; dl ; dl=dl->next) {
2439 struct extent *e = get_extents(ddf, dl);
2440 unsigned long long pos = 0;
2441 int i = 0;
2442 int found = 0;
2443 unsigned long long minsize = size;
2444
2445 if (size == 0)
2446 minsize = chunk;
2447
2448 if (!e)
2449 continue;
2450 do {
2451 unsigned long long esize;
2452 esize = e[i].start - pos;
2453 if (esize >= minsize) {
2454 found = 1;
2455 minsize = esize;
2456 }
2457 pos = e[i].start + e[i].size;
2458 i++;
2459 } while (e[i-1].size);
2460 if (found) {
2461 cnt++;
2462 dl->esize = minsize;
2463 }
2464 free(e);
2465 }
2466 if (cnt < raiddisks) {
2467 fprintf(stderr, Name ": not enough devices with space to create array.\n");
2468 return 0; /* No enough free spaces large enough */
2469 }
2470 if (size == 0) {
2471 /* choose the largest size of which there are at least 'raiddisk' */
2472 for (dl = ddf->dlist ; dl ; dl=dl->next) {
2473 struct dl *dl2;
2474 if (dl->esize <= size)
2475 continue;
2476 /* This is bigger than 'size', see if there are enough */
2477 cnt = 0;
2478 for (dl2 = dl; dl2 ; dl2=dl2->next)
2479 if (dl2->esize >= dl->esize)
2480 cnt++;
2481 if (cnt >= raiddisks)
2482 size = dl->esize;
2483 }
2484 if (chunk) {
2485 size = size / chunk;
2486 size *= chunk;
2487 }
2488 *freesize = size;
2489 if (size < 32) {
2490 fprintf(stderr, Name ": not enough spare devices to create array.\n");
2491 return 0;
2492 }
2493 }
2494 /* We have a 'size' of which there are enough spaces.
2495 * We simply do a first-fit */
2496 cnt = 0;
2497 for (dl = ddf->dlist ; dl && cnt < raiddisks ; dl=dl->next) {
2498 if (dl->esize < size)
2499 continue;
2500
2501 dl->raiddisk = cnt;
2502 cnt++;
2503 }
2504 return 1;
2505 }
2506
2507
2508
2509 static int
2510 validate_geometry_ddf_container(struct supertype *st,
2511 int level, int layout, int raiddisks,
2512 int chunk, unsigned long long size,
2513 char *dev, unsigned long long *freesize,
2514 int verbose);
2515
2516 static int validate_geometry_ddf_bvd(struct supertype *st,
2517 int level, int layout, int raiddisks,
2518 int chunk, unsigned long long size,
2519 char *dev, unsigned long long *freesize,
2520 int verbose);
2521
2522 static int validate_geometry_ddf(struct supertype *st,
2523 int level, int layout, int raiddisks,
2524 int chunk, unsigned long long size,
2525 char *dev, unsigned long long *freesize,
2526 int verbose)
2527 {
2528 int fd;
2529 struct mdinfo *sra;
2530 int cfd;
2531
2532 /* ddf potentially supports lots of things, but it depends on
2533 * what devices are offered (and maybe kernel version?)
2534 * If given unused devices, we will make a container.
2535 * If given devices in a container, we will make a BVD.
2536 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2537 */
2538
2539 if (level == LEVEL_CONTAINER) {
2540 /* Must be a fresh device to add to a container */
2541 return validate_geometry_ddf_container(st, level, layout,
2542 raiddisks, chunk,
2543 size, dev, freesize,
2544 verbose);
2545 }
2546
2547 if (!dev) {
2548 /* Initial sanity check. Exclude illegal levels. */
2549 int i;
2550 for (i=0; ddf_level_num[i].num1 != MAXINT; i++)
2551 if (ddf_level_num[i].num2 == level)
2552 break;
2553 if (ddf_level_num[i].num1 == MAXINT)
2554 return 0;
2555 /* Should check layout? etc */
2556
2557 if (st->sb && freesize) {
2558 /* --create was given a container to create in.
2559 * So we need to check that there are enough
2560 * free spaces and return the amount of space.
2561 * We may as well remember which drives were
2562 * chosen so that add_to_super/getinfo_super
2563 * can return them.
2564 */
2565 return reserve_space(st, raiddisks, size, chunk, freesize);
2566 }
2567 return 1;
2568 }
2569
2570 if (st->sb) {
2571 /* A container has already been opened, so we are
2572 * creating in there. Maybe a BVD, maybe an SVD.
2573 * Should make a distinction one day.
2574 */
2575 return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
2576 chunk, size, dev, freesize,
2577 verbose);
2578 }
2579 /* This is the first device for the array.
2580 * If it is a container, we read it in and do automagic allocations,
2581 * no other devices should be given.
2582 * Otherwise it must be a member device of a container, and we
2583 * do manual allocation.
2584 * Later we should check for a BVD and make an SVD.
2585 */
2586 fd = open(dev, O_RDONLY|O_EXCL, 0);
2587 if (fd >= 0) {
2588 sra = sysfs_read(fd, 0, GET_VERSION);
2589 close(fd);
2590 if (sra && sra->array.major_version == -1 &&
2591 strcmp(sra->text_version, "ddf") == 0) {
2592
2593 /* load super */
2594 /* find space for 'n' devices. */
2595 /* remember the devices */
2596 /* Somehow return the fact that we have enough */
2597 }
2598
2599 if (verbose)
2600 fprintf(stderr,
2601 Name ": ddf: Cannot create this array "
2602 "on device %s\n",
2603 dev);
2604 return 0;
2605 }
2606 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2607 if (verbose)
2608 fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2609 dev, strerror(errno));
2610 return 0;
2611 }
2612 /* Well, it is in use by someone, maybe a 'ddf' container. */
2613 cfd = open_container(fd);
2614 if (cfd < 0) {
2615 close(fd);
2616 if (verbose)
2617 fprintf(stderr, Name ": ddf: Cannot use %s: %s\n",
2618 dev, strerror(EBUSY));
2619 return 0;
2620 }
2621 sra = sysfs_read(cfd, 0, GET_VERSION);
2622 close(fd);
2623 if (sra && sra->array.major_version == -1 &&
2624 strcmp(sra->text_version, "ddf") == 0) {
2625 /* This is a member of a ddf container. Load the container
2626 * and try to create a bvd
2627 */
2628 struct ddf_super *ddf;
2629 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL, 1) == 0) {
2630 st->sb = ddf;
2631 st->container_dev = fd2devnum(cfd);
2632 close(cfd);
2633 return validate_geometry_ddf_bvd(st, level, layout,
2634 raiddisks, chunk, size,
2635 dev, freesize,
2636 verbose);
2637 }
2638 close(cfd);
2639 } else /* device may belong to a different container */
2640 return 0;
2641
2642 return 1;
2643 }
2644
2645 static int
2646 validate_geometry_ddf_container(struct supertype *st,
2647 int level, int layout, int raiddisks,
2648 int chunk, unsigned long long size,
2649 char *dev, unsigned long long *freesize,
2650 int verbose)
2651 {
2652 int fd;
2653 unsigned long long ldsize;
2654
2655 if (level != LEVEL_CONTAINER)
2656 return 0;
2657 if (!dev)
2658 return 1;
2659
2660 fd = open(dev, O_RDONLY|O_EXCL, 0);
2661 if (fd < 0) {
2662 if (verbose)
2663 fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2664 dev, strerror(errno));
2665 return 0;
2666 }
2667 if (!get_dev_size(fd, dev, &ldsize)) {
2668 close(fd);
2669 return 0;
2670 }
2671 close(fd);
2672
2673 *freesize = avail_size_ddf(st, ldsize >> 9);
2674 if (*freesize == 0)
2675 return 0;
2676
2677 return 1;
2678 }
2679
2680 static int validate_geometry_ddf_bvd(struct supertype *st,
2681 int level, int layout, int raiddisks,
2682 int chunk, unsigned long long size,
2683 char *dev, unsigned long long *freesize,
2684 int verbose)
2685 {
2686 struct stat stb;
2687 struct ddf_super *ddf = st->sb;
2688 struct dl *dl;
2689 unsigned long long pos = 0;
2690 unsigned long long maxsize;
2691 struct extent *e;
2692 int i;
2693 /* ddf/bvd supports lots of things, but not containers */
2694 if (level == LEVEL_CONTAINER)
2695 return 0;
2696 /* We must have the container info already read in. */
2697 if (!ddf)
2698 return 0;
2699
2700 if (!dev) {
2701 /* General test: make sure there is space for
2702 * 'raiddisks' device extents of size 'size'.
2703 */
2704 unsigned long long minsize = size;
2705 int dcnt = 0;
2706 if (minsize == 0)
2707 minsize = 8;
2708 for (dl = ddf->dlist; dl ; dl = dl->next)
2709 {
2710 int found = 0;
2711 pos = 0;
2712
2713 i = 0;
2714 e = get_extents(ddf, dl);
2715 if (!e) continue;
2716 do {
2717 unsigned long long esize;
2718 esize = e[i].start - pos;
2719 if (esize >= minsize)
2720 found = 1;
2721 pos = e[i].start + e[i].size;
2722 i++;
2723 } while (e[i-1].size);
2724 if (found)
2725 dcnt++;
2726 free(e);
2727 }
2728 if (dcnt < raiddisks) {
2729 if (verbose)
2730 fprintf(stderr,
2731 Name ": ddf: Not enough devices with "
2732 "space for this array (%d < %d)\n",
2733 dcnt, raiddisks);
2734 return 0;
2735 }
2736 return 1;
2737 }
2738 /* This device must be a member of the set */
2739 if (stat(dev, &stb) < 0)
2740 return 0;
2741 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2742 return 0;
2743 for (dl = ddf->dlist ; dl ; dl = dl->next) {
2744 if (dl->major == major(stb.st_rdev) &&
2745 dl->minor == minor(stb.st_rdev))
2746 break;
2747 }
2748 if (!dl) {
2749 if (verbose)
2750 fprintf(stderr, Name ": ddf: %s is not in the "
2751 "same DDF set\n",
2752 dev);
2753 return 0;
2754 }
2755 e = get_extents(ddf, dl);
2756 maxsize = 0;
2757 i = 0;
2758 if (e) do {
2759 unsigned long long esize;
2760 esize = e[i].start - pos;
2761 if (esize >= maxsize)
2762 maxsize = esize;
2763 pos = e[i].start + e[i].size;
2764 i++;
2765 } while (e[i-1].size);
2766 *freesize = maxsize;
2767 // FIXME here I am
2768
2769 return 1;
2770 }
2771
2772 static int load_super_ddf_all(struct supertype *st, int fd,
2773 void **sbp, char *devname, int keep_fd)
2774 {
2775 struct mdinfo *sra;
2776 struct ddf_super *super;
2777 struct mdinfo *sd, *best = NULL;
2778 int bestseq = 0;
2779 int seq;
2780 char nm[20];
2781 int dfd;
2782 int devnum = fd2devnum(fd);
2783 enum sysfs_read_flags flags;
2784
2785 flags = GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE;
2786 if (mdmon_running(devnum))
2787 flags |= SKIP_GONE_DEVS;
2788
2789 sra = sysfs_read(fd, 0, flags);
2790 if (!sra)
2791 return 1;
2792 if (sra->array.major_version != -1 ||
2793 sra->array.minor_version != -2 ||
2794 strcmp(sra->text_version, "ddf") != 0)
2795 return 1;
2796
2797 if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0)
2798 return 1;
2799 memset(super, 0, sizeof(*super));
2800
2801 /* first, try each device, and choose the best ddf */
2802 for (sd = sra->devs ; sd ; sd = sd->next) {
2803 int rv;
2804 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2805 dfd = dev_open(nm, O_RDONLY);
2806 if (dfd < 0)
2807 return 2;
2808 rv = load_ddf_headers(dfd, super, NULL);
2809 close(dfd);
2810 if (rv == 0) {
2811 seq = __be32_to_cpu(super->active->seq);
2812 if (super->active->openflag)
2813 seq--;
2814 if (!best || seq > bestseq) {
2815 bestseq = seq;
2816 best = sd;
2817 }
2818 }
2819 }
2820 if (!best)
2821 return 1;
2822 /* OK, load this ddf */
2823 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2824 dfd = dev_open(nm, O_RDONLY);
2825 if (dfd < 0)
2826 return 1;
2827 load_ddf_headers(dfd, super, NULL);
2828 load_ddf_global(dfd, super, NULL);
2829 close(dfd);
2830 /* Now we need the device-local bits */
2831 for (sd = sra->devs ; sd ; sd = sd->next) {
2832 int rv;
2833
2834 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2835 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2836 if (dfd < 0)
2837 return 2;
2838 rv = load_ddf_headers(dfd, super, NULL);
2839 if (rv == 0)
2840 rv = load_ddf_local(dfd, super, NULL, keep_fd);
2841 if (!keep_fd) close(dfd);
2842 if (rv)
2843 return 1;
2844 }
2845 if (st->subarray[0]) {
2846 struct vcl *v;
2847
2848 for (v = super->conflist; v; v = v->next)
2849 if (v->vcnum == atoi(st->subarray))
2850 super->currentconf = v;
2851 if (!super->currentconf)
2852 return 1;
2853 }
2854 *sbp = super;
2855 if (st->ss == NULL) {
2856 st->ss = &super_ddf;
2857 st->minor_version = 0;
2858 st->max_devs = 512;
2859 st->container_dev = fd2devnum(fd);
2860 }
2861 st->loaded_container = 1;
2862 return 0;
2863 }
2864 #endif /* MDASSEMBLE */
2865
2866 static struct mdinfo *container_content_ddf(struct supertype *st)
2867 {
2868 /* Given a container loaded by load_super_ddf_all,
2869 * extract information about all the arrays into
2870 * an mdinfo tree.
2871 *
2872 * For each vcl in conflist: create an mdinfo, fill it in,
2873 * then look for matching devices (phys_refnum) in dlist
2874 * and create appropriate device mdinfo.
2875 */
2876 struct ddf_super *ddf = st->sb;
2877 struct mdinfo *rest = NULL;
2878 struct vcl *vc;
2879
2880 for (vc = ddf->conflist ; vc ; vc=vc->next)
2881 {
2882 int i;
2883 int j;
2884 struct mdinfo *this;
2885 this = malloc(sizeof(*this));
2886 memset(this, 0, sizeof(*this));
2887 this->next = rest;
2888 rest = this;
2889
2890 this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2891 this->array.raid_disks =
2892 __be16_to_cpu(vc->conf.prim_elmnt_count);
2893 this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2894 this->array.raid_disks);
2895 this->array.md_minor = -1;
2896 this->array.major_version = -1;
2897 this->array.minor_version = -2;
2898 this->array.ctime = DECADE +
2899 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
2900 this->array.utime = DECADE +
2901 __be32_to_cpu(vc->conf.timestamp);
2902 this->array.chunk_size = 512 << vc->conf.chunk_shift;
2903
2904 i = vc->vcnum;
2905 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2906 (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
2907 DDF_init_full) {
2908 this->array.state = 0;
2909 this->resync_start = 0;
2910 } else {
2911 this->array.state = 1;
2912 this->resync_start = ~0ULL;
2913 }
2914 memcpy(this->name, ddf->virt->entries[i].name, 16);
2915 this->name[16]=0;
2916 for(j=0; j<16; j++)
2917 if (this->name[j] == ' ')
2918 this->name[j] = 0;
2919
2920 memset(this->uuid, 0, sizeof(this->uuid));
2921 this->component_size = __be64_to_cpu(vc->conf.blocks);
2922 this->array.size = this->component_size / 2;
2923 this->container_member = i;
2924
2925 ddf->currentconf = vc;
2926 uuid_from_super_ddf(st, this->uuid);
2927 ddf->currentconf = NULL;
2928
2929 sprintf(this->text_version, "/%s/%d",
2930 devnum2devname(st->container_dev),
2931 this->container_member);
2932
2933 for (i=0 ; i < ddf->mppe ; i++) {
2934 struct mdinfo *dev;
2935 struct dl *d;
2936
2937 if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
2938 continue;
2939
2940 this->array.working_disks++;
2941
2942 for (d = ddf->dlist; d ; d=d->next)
2943 if (d->disk.refnum == vc->conf.phys_refnum[i])
2944 break;
2945 if (d == NULL)
2946 /* Haven't found that one yet, maybe there are others */
2947 continue;
2948
2949 dev = malloc(sizeof(*dev));
2950 memset(dev, 0, sizeof(*dev));
2951 dev->next = this->devs;
2952 this->devs = dev;
2953
2954 dev->disk.number = __be32_to_cpu(d->disk.refnum);
2955 dev->disk.major = d->major;
2956 dev->disk.minor = d->minor;
2957 dev->disk.raid_disk = i;
2958 dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2959
2960 dev->events = __be32_to_cpu(ddf->primary.seq);
2961 dev->data_offset = __be64_to_cpu(vc->lba_offset[i]);
2962 dev->component_size = __be64_to_cpu(vc->conf.blocks);
2963 if (d->devname)
2964 strcpy(dev->name, d->devname);
2965 }
2966 }
2967 return rest;
2968 }
2969
2970 static int store_zero_ddf(struct supertype *st, int fd)
2971 {
2972 unsigned long long dsize;
2973 void *buf;
2974 int rc;
2975
2976 if (!get_dev_size(fd, NULL, &dsize))
2977 return 1;
2978
2979 if (posix_memalign(&buf, 512, 512) != 0)
2980 return 1;
2981 memset(buf, 0, 512);
2982
2983 lseek64(fd, dsize-512, 0);
2984 rc = write(fd, buf, 512);
2985 free(buf);
2986 if (rc < 0)
2987 return 1;
2988 return 0;
2989 }
2990
2991 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
2992 {
2993 /*
2994 * return:
2995 * 0 same, or first was empty, and second was copied
2996 * 1 second had wrong number
2997 * 2 wrong uuid
2998 * 3 wrong other info
2999 */
3000 struct ddf_super *first = st->sb;
3001 struct ddf_super *second = tst->sb;
3002
3003 if (!first) {
3004 st->sb = tst->sb;
3005 tst->sb = NULL;
3006 return 0;
3007 }
3008
3009 if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
3010 return 2;
3011
3012 /* FIXME should I look at anything else? */
3013 return 0;
3014 }
3015
3016 #ifndef MDASSEMBLE
3017 /*
3018 * A new array 'a' has been started which claims to be instance 'inst'
3019 * within container 'c'.
3020 * We need to confirm that the array matches the metadata in 'c' so
3021 * that we don't corrupt any metadata.
3022 */
3023 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
3024 {
3025 dprintf("ddf: open_new %s\n", inst);
3026 a->info.container_member = atoi(inst);
3027 return 0;
3028 }
3029
3030 /*
3031 * The array 'a' is to be marked clean in the metadata.
3032 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
3033 * clean up to the point (in sectors). If that cannot be recorded in the
3034 * metadata, then leave it as dirty.
3035 *
3036 * For DDF, we need to clear the DDF_state_inconsistent bit in the
3037 * !global! virtual_disk.virtual_entry structure.
3038 */
3039 static int ddf_set_array_state(struct active_array *a, int consistent)
3040 {
3041 struct ddf_super *ddf = a->container->sb;
3042 int inst = a->info.container_member;
3043 int old = ddf->virt->entries[inst].state;
3044 if (consistent == 2) {
3045 /* Should check if a recovery should be started FIXME */
3046 consistent = 1;
3047 if (!is_resync_complete(a))
3048 consistent = 0;
3049 }
3050 if (consistent)
3051 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
3052 else
3053 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
3054 if (old != ddf->virt->entries[inst].state)
3055 ddf->updates_pending = 1;
3056
3057 old = ddf->virt->entries[inst].init_state;
3058 ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
3059 if (is_resync_complete(a))
3060 ddf->virt->entries[inst].init_state |= DDF_init_full;
3061 else if (a->resync_start == 0)
3062 ddf->virt->entries[inst].init_state |= DDF_init_not;
3063 else
3064 ddf->virt->entries[inst].init_state |= DDF_init_quick;
3065 if (old != ddf->virt->entries[inst].init_state)
3066 ddf->updates_pending = 1;
3067
3068 dprintf("ddf mark %d %s %llu\n", inst, consistent?"clean":"dirty",
3069 a->resync_start);
3070 return consistent;
3071 }
3072
3073 /*
3074 * The state of each disk is stored in the global phys_disk structure
3075 * in phys_disk.entries[n].state.
3076 * This makes various combinations awkward.
3077 * - When a device fails in any array, it must be failed in all arrays
3078 * that include a part of this device.
3079 * - When a component is rebuilding, we cannot include it officially in the
3080 * array unless this is the only array that uses the device.
3081 *
3082 * So: when transitioning:
3083 * Online -> failed, just set failed flag. monitor will propagate
3084 * spare -> online, the device might need to be added to the array.
3085 * spare -> failed, just set failed. Don't worry if in array or not.
3086 */
3087 static void ddf_set_disk(struct active_array *