5ad89c25a84c427a6da3899f73832afe6a90c029
[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 static int load_super_ddf(struct supertype *st, int fd,
766 char *devname)
767 {
768 unsigned long long dsize;
769 struct ddf_super *super;
770 int rv;
771
772 #ifndef MDASSEMBLE
773 /* if 'fd' is a container, load metadata from all the devices */
774 if (load_super_ddf_all(st, fd, &st->sb, devname, 1) == 0)
775 return 0;
776 #endif
777 if (st->subarray[0])
778 return 1; /* FIXME Is this correct */
779
780 if (get_dev_size(fd, devname, &dsize) == 0)
781 return 1;
782
783 /* 32M is a lower bound */
784 if (dsize <= 32*1024*1024) {
785 if (devname)
786 fprintf(stderr,
787 Name ": %s is too small for ddf: "
788 "size is %llu sectors.\n",
789 devname, dsize>>9);
790 return 1;
791 }
792 if (dsize & 511) {
793 if (devname)
794 fprintf(stderr,
795 Name ": %s is an odd size for ddf: "
796 "size is %llu bytes.\n",
797 devname, dsize);
798 return 1;
799 }
800
801 if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) {
802 fprintf(stderr, Name ": malloc of %zu failed.\n",
803 sizeof(*super));
804 return 1;
805 }
806 memset(super, 0, sizeof(*super));
807
808 rv = load_ddf_headers(fd, super, devname);
809 if (rv) {
810 free(super);
811 return rv;
812 }
813
814 /* Have valid headers and have chosen the best. Let's read in the rest*/
815
816 rv = load_ddf_global(fd, super, devname);
817
818 if (rv) {
819 if (devname)
820 fprintf(stderr,
821 Name ": Failed to load all information "
822 "sections on %s\n", devname);
823 free(super);
824 return rv;
825 }
826
827 rv = load_ddf_local(fd, super, devname, 0);
828
829 if (rv) {
830 if (devname)
831 fprintf(stderr,
832 Name ": Failed to load all information "
833 "sections on %s\n", devname);
834 free(super);
835 return rv;
836 }
837
838 /* Should possibly check the sections .... */
839
840 st->sb = super;
841 if (st->ss == NULL) {
842 st->ss = &super_ddf;
843 st->minor_version = 0;
844 st->max_devs = 512;
845 }
846 st->loaded_container = 0;
847 return 0;
848
849 }
850
851 static void free_super_ddf(struct supertype *st)
852 {
853 struct ddf_super *ddf = st->sb;
854 if (ddf == NULL)
855 return;
856 free(ddf->phys);
857 free(ddf->virt);
858 while (ddf->conflist) {
859 struct vcl *v = ddf->conflist;
860 ddf->conflist = v->next;
861 if (v->block_sizes)
862 free(v->block_sizes);
863 free(v);
864 }
865 while (ddf->dlist) {
866 struct dl *d = ddf->dlist;
867 ddf->dlist = d->next;
868 if (d->fd >= 0)
869 close(d->fd);
870 if (d->spare)
871 free(d->spare);
872 free(d);
873 }
874 free(ddf);
875 st->sb = NULL;
876 }
877
878 static struct supertype *match_metadata_desc_ddf(char *arg)
879 {
880 /* 'ddf' only support containers */
881 struct supertype *st;
882 if (strcmp(arg, "ddf") != 0 &&
883 strcmp(arg, "default") != 0
884 )
885 return NULL;
886
887 st = malloc(sizeof(*st));
888 memset(st, 0, sizeof(*st));
889 st->ss = &super_ddf;
890 st->max_devs = 512;
891 st->minor_version = 0;
892 st->sb = NULL;
893 return st;
894 }
895
896
897 #ifndef MDASSEMBLE
898
899 static mapping_t ddf_state[] = {
900 { "Optimal", 0},
901 { "Degraded", 1},
902 { "Deleted", 2},
903 { "Missing", 3},
904 { "Failed", 4},
905 { "Partially Optimal", 5},
906 { "-reserved-", 6},
907 { "-reserved-", 7},
908 { NULL, 0}
909 };
910
911 static mapping_t ddf_init_state[] = {
912 { "Not Initialised", 0},
913 { "QuickInit in Progress", 1},
914 { "Fully Initialised", 2},
915 { "*UNKNOWN*", 3},
916 { NULL, 0}
917 };
918 static mapping_t ddf_access[] = {
919 { "Read/Write", 0},
920 { "Reserved", 1},
921 { "Read Only", 2},
922 { "Blocked (no access)", 3},
923 { NULL ,0}
924 };
925
926 static mapping_t ddf_level[] = {
927 { "RAID0", DDF_RAID0},
928 { "RAID1", DDF_RAID1},
929 { "RAID3", DDF_RAID3},
930 { "RAID4", DDF_RAID4},
931 { "RAID5", DDF_RAID5},
932 { "RAID1E",DDF_RAID1E},
933 { "JBOD", DDF_JBOD},
934 { "CONCAT",DDF_CONCAT},
935 { "RAID5E",DDF_RAID5E},
936 { "RAID5EE",DDF_RAID5EE},
937 { "RAID6", DDF_RAID6},
938 { NULL, 0}
939 };
940 static mapping_t ddf_sec_level[] = {
941 { "Striped", DDF_2STRIPED},
942 { "Mirrored", DDF_2MIRRORED},
943 { "Concat", DDF_2CONCAT},
944 { "Spanned", DDF_2SPANNED},
945 { NULL, 0}
946 };
947 #endif
948
949 struct num_mapping {
950 int num1, num2;
951 };
952 static struct num_mapping ddf_level_num[] = {
953 { DDF_RAID0, 0 },
954 { DDF_RAID1, 1 },
955 { DDF_RAID3, LEVEL_UNSUPPORTED },
956 { DDF_RAID4, 4 },
957 { DDF_RAID5, 5 },
958 { DDF_RAID1E, LEVEL_UNSUPPORTED },
959 { DDF_JBOD, LEVEL_UNSUPPORTED },
960 { DDF_CONCAT, LEVEL_LINEAR },
961 { DDF_RAID5E, LEVEL_UNSUPPORTED },
962 { DDF_RAID5EE, LEVEL_UNSUPPORTED },
963 { DDF_RAID6, 6},
964 { MAXINT, MAXINT }
965 };
966
967 static int map_num1(struct num_mapping *map, int num)
968 {
969 int i;
970 for (i=0 ; map[i].num1 != MAXINT; i++)
971 if (map[i].num1 == num)
972 break;
973 return map[i].num2;
974 }
975
976 static int all_ff(char *guid)
977 {
978 int i;
979 for (i = 0; i < DDF_GUID_LEN; i++)
980 if (guid[i] != (char)0xff)
981 return 0;
982 return 1;
983 }
984
985 #ifndef MDASSEMBLE
986 static void print_guid(char *guid, int tstamp)
987 {
988 /* A GUIDs are part (or all) ASCII and part binary.
989 * They tend to be space padded.
990 * We print the GUID in HEX, then in parentheses add
991 * any initial ASCII sequence, and a possible
992 * time stamp from bytes 16-19
993 */
994 int l = DDF_GUID_LEN;
995 int i;
996
997 for (i=0 ; i<DDF_GUID_LEN ; i++) {
998 if ((i&3)==0 && i != 0) printf(":");
999 printf("%02X", guid[i]&255);
1000 }
1001
1002 printf("\n (");
1003 while (l && guid[l-1] == ' ')
1004 l--;
1005 for (i=0 ; i<l ; i++) {
1006 if (guid[i] >= 0x20 && guid[i] < 0x7f)
1007 fputc(guid[i], stdout);
1008 else
1009 break;
1010 }
1011 if (tstamp) {
1012 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
1013 char tbuf[100];
1014 struct tm *tm;
1015 tm = localtime(&then);
1016 strftime(tbuf, 100, " %D %T",tm);
1017 fputs(tbuf, stdout);
1018 }
1019 printf(")");
1020 }
1021
1022 static void examine_vd(int n, struct ddf_super *sb, char *guid)
1023 {
1024 int crl = sb->conf_rec_len;
1025 struct vcl *vcl;
1026
1027 for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
1028 int i;
1029 struct vd_config *vc = &vcl->conf;
1030
1031 if (calc_crc(vc, crl*512) != vc->crc)
1032 continue;
1033 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
1034 continue;
1035
1036 /* Ok, we know about this VD, let's give more details */
1037 printf(" Raid Devices[%d] : %d (", n,
1038 __be16_to_cpu(vc->prim_elmnt_count));
1039 for (i=0; i<__be16_to_cpu(vc->prim_elmnt_count); i++) {
1040 int j;
1041 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1042 for (j=0; j<cnt; j++)
1043 if (vc->phys_refnum[i] == sb->phys->entries[j].refnum)
1044 break;
1045 if (i) printf(" ");
1046 if (j < cnt)
1047 printf("%d", j);
1048 else
1049 printf("--");
1050 }
1051 printf(")\n");
1052 if (vc->chunk_shift != 255)
1053 printf(" Chunk Size[%d] : %d sectors\n", n,
1054 1 << vc->chunk_shift);
1055 printf(" Raid Level[%d] : %s\n", n,
1056 map_num(ddf_level, vc->prl)?:"-unknown-");
1057 if (vc->sec_elmnt_count != 1) {
1058 printf(" Secondary Position[%d] : %d of %d\n", n,
1059 vc->sec_elmnt_seq, vc->sec_elmnt_count);
1060 printf(" Secondary Level[%d] : %s\n", n,
1061 map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1062 }
1063 printf(" Device Size[%d] : %llu\n", n,
1064 (unsigned long long)__be64_to_cpu(vc->blocks)/2);
1065 printf(" Array Size[%d] : %llu\n", n,
1066 (unsigned long long)__be64_to_cpu(vc->array_blocks)/2);
1067 }
1068 }
1069
1070 static void examine_vds(struct ddf_super *sb)
1071 {
1072 int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1073 int i;
1074 printf(" Virtual Disks : %d\n", cnt);
1075
1076 for (i=0; i<cnt; i++) {
1077 struct virtual_entry *ve = &sb->virt->entries[i];
1078 printf("\n");
1079 printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1080 printf("\n");
1081 printf(" unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1082 printf(" state[%d] : %s, %s%s\n", i,
1083 map_num(ddf_state, ve->state & 7),
1084 (ve->state & 8) ? "Morphing, ": "",
1085 (ve->state & 16)? "Not Consistent" : "Consistent");
1086 printf(" init state[%d] : %s\n", i,
1087 map_num(ddf_init_state, ve->init_state&3));
1088 printf(" access[%d] : %s\n", i,
1089 map_num(ddf_access, (ve->init_state>>6) & 3));
1090 printf(" Name[%d] : %.16s\n", i, ve->name);
1091 examine_vd(i, sb, ve->guid);
1092 }
1093 if (cnt) printf("\n");
1094 }
1095
1096 static void examine_pds(struct ddf_super *sb)
1097 {
1098 int cnt = __be16_to_cpu(sb->phys->used_pdes);
1099 int i;
1100 struct dl *dl;
1101 printf(" Physical Disks : %d\n", cnt);
1102 printf(" Number RefNo Size Device Type/State\n");
1103
1104 for (i=0 ; i<cnt ; i++) {
1105 struct phys_disk_entry *pd = &sb->phys->entries[i];
1106 int type = __be16_to_cpu(pd->type);
1107 int state = __be16_to_cpu(pd->state);
1108
1109 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1110 //printf("\n");
1111 printf(" %3d %08x ", i,
1112 __be32_to_cpu(pd->refnum));
1113 printf("%8lluK ",
1114 (unsigned long long)__be64_to_cpu(pd->config_size)>>1);
1115 for (dl = sb->dlist; dl ; dl = dl->next) {
1116 if (dl->disk.refnum == pd->refnum) {
1117 char *dv = map_dev(dl->major, dl->minor, 0);
1118 if (dv) {
1119 printf("%-15s", dv);
1120 break;
1121 }
1122 }
1123 }
1124 if (!dl)
1125 printf("%15s","");
1126 printf(" %s%s%s%s%s",
1127 (type&2) ? "active":"",
1128 (type&4) ? "Global-Spare":"",
1129 (type&8) ? "spare" : "",
1130 (type&16)? ", foreign" : "",
1131 (type&32)? "pass-through" : "");
1132 printf("/%s%s%s%s%s%s%s",
1133 (state&1)? "Online": "Offline",
1134 (state&2)? ", Failed": "",
1135 (state&4)? ", Rebuilding": "",
1136 (state&8)? ", in-transition": "",
1137 (state&16)? ", SMART-errors": "",
1138 (state&32)? ", Unrecovered-Read-Errors": "",
1139 (state&64)? ", Missing" : "");
1140 printf("\n");
1141 }
1142 }
1143
1144 static void examine_super_ddf(struct supertype *st, char *homehost)
1145 {
1146 struct ddf_super *sb = st->sb;
1147
1148 printf(" Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1149 printf(" Version : %.8s\n", sb->anchor.revision);
1150 printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1151 printf("\n");
1152 printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1153 printf("\n");
1154 printf(" Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1155 printf(" Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1156 ?"yes" : "no");
1157 examine_vds(sb);
1158 examine_pds(sb);
1159 }
1160
1161 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info);
1162
1163 static void uuid_from_super_ddf(struct supertype *st, int uuid[4]);
1164
1165 static void brief_examine_super_ddf(struct supertype *st, int verbose)
1166 {
1167 /* We just write a generic DDF ARRAY entry
1168 */
1169 struct ddf_super *ddf = st->sb;
1170 struct mdinfo info;
1171 int i;
1172 char nbuf[64];
1173 getinfo_super_ddf(st, &info);
1174 fname_from_uuid(st, &info, nbuf, ':');
1175
1176 for (i=0; i<__be16_to_cpu(ddf->virt->max_vdes); i++) {
1177 struct virtual_entry *ve = &ddf->virt->entries[i];
1178 struct vcl vcl;
1179 char nbuf1[64];
1180 if (all_ff(ve->guid))
1181 continue;
1182 memcpy(vcl.conf.guid, ve->guid, DDF_GUID_LEN);
1183 ddf->currentconf =&vcl;
1184 uuid_from_super_ddf(st, info.uuid);
1185 fname_from_uuid(st, &info, nbuf1, ':');
1186 printf("ARRAY container=%s member=%d UUID=%s\n",
1187 nbuf+5, i, nbuf1+5);
1188 }
1189 printf("ARRAY metadata=ddf UUID=%s\n", nbuf + 5);
1190 }
1191
1192 static void export_examine_super_ddf(struct supertype *st)
1193 {
1194 struct mdinfo info;
1195 char nbuf[64];
1196 getinfo_super_ddf(st, &info);
1197 fname_from_uuid(st, &info, nbuf, ':');
1198 printf("MD_METADATA=ddf\n");
1199 printf("MD_LEVEL=container\n");
1200 printf("MD_UUID=%s\n", nbuf+5);
1201 }
1202
1203
1204 static void detail_super_ddf(struct supertype *st, char *homehost)
1205 {
1206 /* FIXME later
1207 * Could print DDF GUID
1208 * Need to find which array
1209 * If whole, briefly list all arrays
1210 * If one, give name
1211 */
1212 }
1213
1214 static void brief_detail_super_ddf(struct supertype *st)
1215 {
1216 /* FIXME I really need to know which array we are detailing.
1217 * Can that be stored in ddf_super??
1218 */
1219 // struct ddf_super *ddf = st->sb;
1220 struct mdinfo info;
1221 char nbuf[64];
1222 getinfo_super_ddf(st, &info);
1223 fname_from_uuid(st, &info, nbuf,':');
1224 printf(" UUID=%s", nbuf + 5);
1225 }
1226 #endif
1227
1228 static int match_home_ddf(struct supertype *st, char *homehost)
1229 {
1230 /* It matches 'this' host if the controller is a
1231 * Linux-MD controller with vendor_data matching
1232 * the hostname
1233 */
1234 struct ddf_super *ddf = st->sb;
1235 int len = strlen(homehost);
1236
1237 return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1238 len < sizeof(ddf->controller.vendor_data) &&
1239 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1240 ddf->controller.vendor_data[len] == 0);
1241 }
1242
1243 #ifndef MDASSEMBLE
1244 static struct vd_config *find_vdcr(struct ddf_super *ddf, int inst)
1245 {
1246 struct vcl *v;
1247
1248 for (v = ddf->conflist; v; v = v->next)
1249 if (inst == v->vcnum)
1250 return &v->conf;
1251 return NULL;
1252 }
1253 #endif
1254
1255 static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1256 {
1257 /* Find the entry in phys_disk which has the given refnum
1258 * and return it's index
1259 */
1260 int i;
1261 for (i=0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1262 if (ddf->phys->entries[i].refnum == phys_refnum)
1263 return i;
1264 return -1;
1265 }
1266
1267 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1268 {
1269 /* The uuid returned here is used for:
1270 * uuid to put into bitmap file (Create, Grow)
1271 * uuid for backup header when saving critical section (Grow)
1272 * comparing uuids when re-adding a device into an array
1273 * In these cases the uuid required is that of the data-array,
1274 * not the device-set.
1275 * uuid to recognise same set when adding a missing device back
1276 * to an array. This is a uuid for the device-set.
1277 *
1278 * For each of these we can make do with a truncated
1279 * or hashed uuid rather than the original, as long as
1280 * everyone agrees.
1281 * In the case of SVD we assume the BVD is of interest,
1282 * though that might be the case if a bitmap were made for
1283 * a mirrored SVD - worry about that later.
1284 * So we need to find the VD configuration record for the
1285 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1286 * The first 16 bytes of the sha1 of these is used.
1287 */
1288 struct ddf_super *ddf = st->sb;
1289 struct vcl *vcl = ddf->currentconf;
1290 char *guid;
1291 char buf[20];
1292 struct sha1_ctx ctx;
1293
1294 if (vcl)
1295 guid = vcl->conf.guid;
1296 else
1297 guid = ddf->anchor.guid;
1298
1299 sha1_init_ctx(&ctx);
1300 sha1_process_bytes(guid, DDF_GUID_LEN, &ctx);
1301 sha1_finish_ctx(&ctx, buf);
1302 memcpy(uuid, buf, 4*4);
1303 }
1304
1305 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info);
1306
1307 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info)
1308 {
1309 struct ddf_super *ddf = st->sb;
1310
1311 if (ddf->currentconf) {
1312 getinfo_super_ddf_bvd(st, info);
1313 return;
1314 }
1315
1316 info->array.raid_disks = __be16_to_cpu(ddf->phys->used_pdes);
1317 info->array.level = LEVEL_CONTAINER;
1318 info->array.layout = 0;
1319 info->array.md_minor = -1;
1320 info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)
1321 (ddf->anchor.guid+16));
1322 info->array.utime = 0;
1323 info->array.chunk_size = 0;
1324
1325
1326 info->disk.major = 0;
1327 info->disk.minor = 0;
1328 if (ddf->dlist) {
1329 info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1330 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1331
1332 info->data_offset = __be64_to_cpu(ddf->phys->
1333 entries[info->disk.raid_disk].
1334 config_size);
1335 info->component_size = ddf->dlist->size - info->data_offset;
1336 } else {
1337 info->disk.number = -1;
1338 info->disk.raid_disk = -1;
1339 // info->disk.raid_disk = find refnum in the table and use index;
1340 }
1341 info->disk.state = (1 << MD_DISK_SYNC);
1342
1343
1344 info->reshape_active = 0;
1345 info->name[0] = 0;
1346
1347 info->array.major_version = -1;
1348 info->array.minor_version = -2;
1349 strcpy(info->text_version, "ddf");
1350 info->safe_mode_delay = 0;
1351
1352 uuid_from_super_ddf(st, info->uuid);
1353
1354 }
1355
1356 static int rlq_to_layout(int rlq, int prl, int raiddisks);
1357
1358 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info)
1359 {
1360 struct ddf_super *ddf = st->sb;
1361 struct vcl *vc = ddf->currentconf;
1362 int cd = ddf->currentdev;
1363 int j;
1364 struct dl *dl;
1365
1366 /* FIXME this returns BVD info - what if we want SVD ?? */
1367
1368 info->array.raid_disks = __be16_to_cpu(vc->conf.prim_elmnt_count);
1369 info->array.level = map_num1(ddf_level_num, vc->conf.prl);
1370 info->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
1371 info->array.raid_disks);
1372 info->array.md_minor = -1;
1373 info->array.ctime = DECADE +
1374 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
1375 info->array.utime = DECADE + __be32_to_cpu(vc->conf.timestamp);
1376 info->array.chunk_size = 512 << vc->conf.chunk_shift;
1377 info->custom_array_size = 0;
1378
1379 if (cd >= 0 && cd < ddf->mppe) {
1380 info->data_offset = __be64_to_cpu(vc->lba_offset[cd]);
1381 if (vc->block_sizes)
1382 info->component_size = vc->block_sizes[cd];
1383 else
1384 info->component_size = __be64_to_cpu(vc->conf.blocks);
1385 }
1386
1387 for (dl = ddf->dlist; dl ; dl = dl->next)
1388 if (dl->raiddisk == info->disk.raid_disk)
1389 break;
1390 info->disk.major = 0;
1391 info->disk.minor = 0;
1392 if (dl) {
1393 info->disk.major = dl->major;
1394 info->disk.minor = dl->minor;
1395 }
1396 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1397 // info->disk.raid_disk = find refnum in the table and use index;
1398 // info->disk.state = ???;
1399
1400 info->container_member = ddf->currentconf->vcnum;
1401
1402 info->resync_start = 0;
1403 if (!(ddf->virt->entries[info->container_member].state
1404 & DDF_state_inconsistent) &&
1405 (ddf->virt->entries[info->container_member].init_state
1406 & DDF_initstate_mask)
1407 == DDF_init_full)
1408 info->resync_start = ~0ULL;
1409
1410 uuid_from_super_ddf(st, info->uuid);
1411
1412 info->container_member = atoi(st->subarray);
1413 info->array.major_version = -1;
1414 info->array.minor_version = -2;
1415 sprintf(info->text_version, "/%s/%s",
1416 devnum2devname(st->container_dev),
1417 st->subarray);
1418 info->safe_mode_delay = 200;
1419
1420 memcpy(info->name, ddf->virt->entries[info->container_member].name, 16);
1421 info->name[16]=0;
1422 for(j=0; j<16; j++)
1423 if (info->name[j] == ' ')
1424 info->name[j] = 0;
1425 }
1426
1427
1428 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1429 char *update,
1430 char *devname, int verbose,
1431 int uuid_set, char *homehost)
1432 {
1433 /* For 'assemble' and 'force' we need to return non-zero if any
1434 * change was made. For others, the return value is ignored.
1435 * Update options are:
1436 * force-one : This device looks a bit old but needs to be included,
1437 * update age info appropriately.
1438 * assemble: clear any 'faulty' flag to allow this device to
1439 * be assembled.
1440 * force-array: Array is degraded but being forced, mark it clean
1441 * if that will be needed to assemble it.
1442 *
1443 * newdev: not used ????
1444 * grow: Array has gained a new device - this is currently for
1445 * linear only
1446 * resync: mark as dirty so a resync will happen.
1447 * uuid: Change the uuid of the array to match what is given
1448 * homehost: update the recorded homehost
1449 * name: update the name - preserving the homehost
1450 * _reshape_progress: record new reshape_progress position.
1451 *
1452 * Following are not relevant for this version:
1453 * sparc2.2 : update from old dodgey metadata
1454 * super-minor: change the preferred_minor number
1455 * summaries: update redundant counters.
1456 */
1457 int rv = 0;
1458 // struct ddf_super *ddf = st->sb;
1459 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1460 // struct virtual_entry *ve = find_ve(ddf);
1461
1462 /* we don't need to handle "force-*" or "assemble" as
1463 * there is no need to 'trick' the kernel. We the metadata is
1464 * first updated to activate the array, all the implied modifications
1465 * will just happen.
1466 */
1467
1468 if (strcmp(update, "grow") == 0) {
1469 /* FIXME */
1470 }
1471 if (strcmp(update, "resync") == 0) {
1472 // info->resync_checkpoint = 0;
1473 }
1474 /* We ignore UUID updates as they make even less sense
1475 * with DDF
1476 */
1477 if (strcmp(update, "homehost") == 0) {
1478 /* homehost is stored in controller->vendor_data,
1479 * or it is when we are the vendor
1480 */
1481 // if (info->vendor_is_local)
1482 // strcpy(ddf->controller.vendor_data, homehost);
1483 }
1484 if (strcmp(update, "name") == 0) {
1485 /* name is stored in virtual_entry->name */
1486 // memset(ve->name, ' ', 16);
1487 // strncpy(ve->name, info->name, 16);
1488 }
1489 if (strcmp(update, "_reshape_progress") == 0) {
1490 /* We don't support reshape yet */
1491 }
1492
1493 // update_all_csum(ddf);
1494
1495 return rv;
1496 }
1497
1498 __u32 random32(void)
1499 {
1500 __u32 rv;
1501 int rfd = open("/dev/urandom", O_RDONLY);
1502 if (rfd < 0 || read(rfd, &rv, 4) != 4)
1503 rv = random();
1504 if (rfd >= 0)
1505 close(rfd);
1506 return rv;
1507 }
1508
1509 static void make_header_guid(char *guid)
1510 {
1511 __u32 stamp;
1512 /* Create a DDF Header of Virtual Disk GUID */
1513
1514 /* 24 bytes of fiction required.
1515 * first 8 are a 'vendor-id' - "Linux-MD"
1516 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1517 * Remaining 8 random number plus timestamp
1518 */
1519 memcpy(guid, T10, sizeof(T10));
1520 stamp = __cpu_to_be32(0xdeadbeef);
1521 memcpy(guid+8, &stamp, 4);
1522 stamp = __cpu_to_be32(0);
1523 memcpy(guid+12, &stamp, 4);
1524 stamp = __cpu_to_be32(time(0) - DECADE);
1525 memcpy(guid+16, &stamp, 4);
1526 stamp = random32();
1527 memcpy(guid+20, &stamp, 4);
1528 }
1529
1530 static int init_super_ddf_bvd(struct supertype *st,
1531 mdu_array_info_t *info,
1532 unsigned long long size,
1533 char *name, char *homehost,
1534 int *uuid);
1535
1536 static int init_super_ddf(struct supertype *st,
1537 mdu_array_info_t *info,
1538 unsigned long long size, char *name, char *homehost,
1539 int *uuid)
1540 {
1541 /* This is primarily called by Create when creating a new array.
1542 * We will then get add_to_super called for each component, and then
1543 * write_init_super called to write it out to each device.
1544 * For DDF, Create can create on fresh devices or on a pre-existing
1545 * array.
1546 * To create on a pre-existing array a different method will be called.
1547 * This one is just for fresh drives.
1548 *
1549 * We need to create the entire 'ddf' structure which includes:
1550 * DDF headers - these are easy.
1551 * Controller data - a Sector describing this controller .. not that
1552 * this is a controller exactly.
1553 * Physical Disk Record - one entry per device, so
1554 * leave plenty of space.
1555 * Virtual Disk Records - again, just leave plenty of space.
1556 * This just lists VDs, doesn't give details
1557 * Config records - describes the VDs that use this disk
1558 * DiskData - describes 'this' device.
1559 * BadBlockManagement - empty
1560 * Diag Space - empty
1561 * Vendor Logs - Could we put bitmaps here?
1562 *
1563 */
1564 struct ddf_super *ddf;
1565 char hostname[17];
1566 int hostlen;
1567 int max_phys_disks, max_virt_disks;
1568 unsigned long long sector;
1569 int clen;
1570 int i;
1571 int pdsize, vdsize;
1572 struct phys_disk *pd;
1573 struct virtual_disk *vd;
1574
1575 if (!info) {
1576 st->sb = NULL;
1577 return 0;
1578 }
1579 if (st->sb)
1580 return init_super_ddf_bvd(st, info, size, name, homehost,
1581 uuid);
1582
1583 if (posix_memalign((void**)&ddf, 512, sizeof(*ddf)) != 0) {
1584 fprintf(stderr, Name ": %s could not allocate superblock\n", __func__);
1585 return 0;
1586 }
1587 memset(ddf, 0, sizeof(*ddf));
1588 ddf->dlist = NULL; /* no physical disks yet */
1589 ddf->conflist = NULL; /* No virtual disks yet */
1590
1591 /* At least 32MB *must* be reserved for the ddf. So let's just
1592 * start 32MB from the end, and put the primary header there.
1593 * Don't do secondary for now.
1594 * We don't know exactly where that will be yet as it could be
1595 * different on each device. To just set up the lengths.
1596 *
1597 */
1598
1599 ddf->anchor.magic = DDF_HEADER_MAGIC;
1600 make_header_guid(ddf->anchor.guid);
1601
1602 memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
1603 ddf->anchor.seq = __cpu_to_be32(1);
1604 ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1605 ddf->anchor.openflag = 0xFF;
1606 ddf->anchor.foreignflag = 0;
1607 ddf->anchor.enforcegroups = 0; /* Is this best?? */
1608 ddf->anchor.pad0 = 0xff;
1609 memset(ddf->anchor.pad1, 0xff, 12);
1610 memset(ddf->anchor.header_ext, 0xff, 32);
1611 ddf->anchor.primary_lba = ~(__u64)0;
1612 ddf->anchor.secondary_lba = ~(__u64)0;
1613 ddf->anchor.type = DDF_HEADER_ANCHOR;
1614 memset(ddf->anchor.pad2, 0xff, 3);
1615 ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1616 ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1617 of 32M reserved.. */
1618 max_phys_disks = 1023; /* Should be enough */
1619 ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1620 max_virt_disks = 255;
1621 ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1622 ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1623 ddf->max_part = 64;
1624 ddf->mppe = 256;
1625 ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
1626 ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1627 ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe);
1628 memset(ddf->anchor.pad3, 0xff, 54);
1629 /* controller sections is one sector long immediately
1630 * after the ddf header */
1631 sector = 1;
1632 ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1633 ddf->anchor.controller_section_length = __cpu_to_be32(1);
1634 sector += 1;
1635
1636 /* phys is 8 sectors after that */
1637 pdsize = ROUND_UP(sizeof(struct phys_disk) +
1638 sizeof(struct phys_disk_entry)*max_phys_disks,
1639 512);
1640 switch(pdsize/512) {
1641 case 2: case 8: case 32: case 128: case 512: break;
1642 default: abort();
1643 }
1644 ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1645 ddf->anchor.phys_section_length =
1646 __cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1647 sector += pdsize/512;
1648
1649 /* virt is another 32 sectors */
1650 vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1651 sizeof(struct virtual_entry) * max_virt_disks,
1652 512);
1653 switch(vdsize/512) {
1654 case 2: case 8: case 32: case 128: case 512: break;
1655 default: abort();
1656 }
1657 ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1658 ddf->anchor.virt_section_length =
1659 __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1660 sector += vdsize/512;
1661
1662 clen = ddf->conf_rec_len * (ddf->max_part+1);
1663 ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1664 ddf->anchor.config_section_length = __cpu_to_be32(clen);
1665 sector += clen;
1666
1667 ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1668 ddf->anchor.data_section_length = __cpu_to_be32(1);
1669 sector += 1;
1670
1671 ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1672 ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1673 ddf->anchor.diag_space_length = __cpu_to_be32(0);
1674 ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1675 ddf->anchor.vendor_length = __cpu_to_be32(0);
1676 ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1677
1678 memset(ddf->anchor.pad4, 0xff, 256);
1679
1680 memcpy(&ddf->primary, &ddf->anchor, 512);
1681 memcpy(&ddf->secondary, &ddf->anchor, 512);
1682
1683 ddf->primary.openflag = 1; /* I guess.. */
1684 ddf->primary.type = DDF_HEADER_PRIMARY;
1685
1686 ddf->secondary.openflag = 1; /* I guess.. */
1687 ddf->secondary.type = DDF_HEADER_SECONDARY;
1688
1689 ddf->active = &ddf->primary;
1690
1691 ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1692
1693 /* 24 more bytes of fiction required.
1694 * first 8 are a 'vendor-id' - "Linux-MD"
1695 * Remaining 16 are serial number.... maybe a hostname would do?
1696 */
1697 memcpy(ddf->controller.guid, T10, sizeof(T10));
1698 gethostname(hostname, sizeof(hostname));
1699 hostname[sizeof(hostname) - 1] = 0;
1700 hostlen = strlen(hostname);
1701 memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1702 for (i = strlen(T10) ; i+hostlen < 24; i++)
1703 ddf->controller.guid[i] = ' ';
1704
1705 ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1706 ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1707 ddf->controller.type.sub_vendor_id = 0;
1708 ddf->controller.type.sub_device_id = 0;
1709 memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1710 memset(ddf->controller.pad, 0xff, 8);
1711 memset(ddf->controller.vendor_data, 0xff, 448);
1712 if (homehost && strlen(homehost) < 440)
1713 strcpy((char*)ddf->controller.vendor_data, homehost);
1714
1715 if (posix_memalign((void**)&pd, 512, pdsize) != 0) {
1716 fprintf(stderr, Name ": %s could not allocate pd\n", __func__);
1717 return 0;
1718 }
1719 ddf->phys = pd;
1720 ddf->pdsize = pdsize;
1721
1722 memset(pd, 0xff, pdsize);
1723 memset(pd, 0, sizeof(*pd));
1724 pd->magic = DDF_PHYS_DATA_MAGIC;
1725 pd->used_pdes = __cpu_to_be16(0);
1726 pd->max_pdes = __cpu_to_be16(max_phys_disks);
1727 memset(pd->pad, 0xff, 52);
1728
1729 if (posix_memalign((void**)&vd, 512, vdsize) != 0) {
1730 fprintf(stderr, Name ": %s could not allocate vd\n", __func__);
1731 return 0;
1732 }
1733 ddf->virt = vd;
1734 ddf->vdsize = vdsize;
1735 memset(vd, 0, vdsize);
1736 vd->magic = DDF_VIRT_RECORDS_MAGIC;
1737 vd->populated_vdes = __cpu_to_be16(0);
1738 vd->max_vdes = __cpu_to_be16(max_virt_disks);
1739 memset(vd->pad, 0xff, 52);
1740
1741 for (i=0; i<max_virt_disks; i++)
1742 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1743
1744 st->sb = ddf;
1745 ddf->updates_pending = 1;
1746 return 1;
1747 }
1748
1749 static int chunk_to_shift(int chunksize)
1750 {
1751 return ffs(chunksize/512)-1;
1752 }
1753
1754 static int level_to_prl(int level)
1755 {
1756 switch (level) {
1757 case LEVEL_LINEAR: return DDF_CONCAT;
1758 case 0: return DDF_RAID0;
1759 case 1: return DDF_RAID1;
1760 case 4: return DDF_RAID4;
1761 case 5: return DDF_RAID5;
1762 case 6: return DDF_RAID6;
1763 default: return -1;
1764 }
1765 }
1766 static int layout_to_rlq(int level, int layout, int raiddisks)
1767 {
1768 switch(level) {
1769 case 0:
1770 return DDF_RAID0_SIMPLE;
1771 case 1:
1772 switch(raiddisks) {
1773 case 2: return DDF_RAID1_SIMPLE;
1774 case 3: return DDF_RAID1_MULTI;
1775 default: return -1;
1776 }
1777 case 4:
1778 switch(layout) {
1779 case 0: return DDF_RAID4_N;
1780 }
1781 break;
1782 case 5:
1783 switch(layout) {
1784 case ALGORITHM_LEFT_ASYMMETRIC:
1785 return DDF_RAID5_N_RESTART;
1786 case ALGORITHM_RIGHT_ASYMMETRIC:
1787 return DDF_RAID5_0_RESTART;
1788 case ALGORITHM_LEFT_SYMMETRIC:
1789 return DDF_RAID5_N_CONTINUE;
1790 case ALGORITHM_RIGHT_SYMMETRIC:
1791 return -1; /* not mentioned in standard */
1792 }
1793 case 6:
1794 switch(layout) {
1795 case ALGORITHM_ROTATING_N_RESTART:
1796 return DDF_RAID5_N_RESTART;
1797 case ALGORITHM_ROTATING_ZERO_RESTART:
1798 return DDF_RAID6_0_RESTART;
1799 case ALGORITHM_ROTATING_N_CONTINUE:
1800 return DDF_RAID5_N_CONTINUE;
1801 }
1802 }
1803 return -1;
1804 }
1805
1806 static int rlq_to_layout(int rlq, int prl, int raiddisks)
1807 {
1808 switch(prl) {
1809 case DDF_RAID0:
1810 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1811 case DDF_RAID1:
1812 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1813 on raiddisks*/
1814 case DDF_RAID4:
1815 switch(rlq) {
1816 case DDF_RAID4_N:
1817 return 0;
1818 default:
1819 /* not supported */
1820 return -1; /* FIXME this isn't checked */
1821 }
1822 case DDF_RAID5:
1823 switch(rlq) {
1824 case DDF_RAID5_N_RESTART:
1825 return ALGORITHM_LEFT_ASYMMETRIC;
1826 case DDF_RAID5_0_RESTART:
1827 return ALGORITHM_RIGHT_ASYMMETRIC;
1828 case DDF_RAID5_N_CONTINUE:
1829 return ALGORITHM_LEFT_SYMMETRIC;
1830 default:
1831 return -1;
1832 }
1833 case DDF_RAID6:
1834 switch(rlq) {
1835 case DDF_RAID5_N_RESTART:
1836 return ALGORITHM_ROTATING_N_RESTART;
1837 case DDF_RAID6_0_RESTART:
1838 return ALGORITHM_ROTATING_ZERO_RESTART;
1839 case DDF_RAID5_N_CONTINUE:
1840 return ALGORITHM_ROTATING_N_CONTINUE;
1841 default:
1842 return -1;
1843 }
1844 }
1845 return -1;
1846 }
1847
1848 #ifndef MDASSEMBLE
1849 struct extent {
1850 unsigned long long start, size;
1851 };
1852 static int cmp_extent(const void *av, const void *bv)
1853 {
1854 const struct extent *a = av;
1855 const struct extent *b = bv;
1856 if (a->start < b->start)
1857 return -1;
1858 if (a->start > b->start)
1859 return 1;
1860 return 0;
1861 }
1862
1863 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
1864 {
1865 /* find a list of used extents on the give physical device
1866 * (dnum) of the given ddf.
1867 * Return a malloced array of 'struct extent'
1868
1869 FIXME ignore DDF_Legacy devices?
1870
1871 */
1872 struct extent *rv;
1873 int n = 0;
1874 int i, j;
1875
1876 rv = malloc(sizeof(struct extent) * (ddf->max_part + 2));
1877 if (!rv)
1878 return NULL;
1879
1880 for (i = 0; i < ddf->max_part; i++) {
1881 struct vcl *v = dl->vlist[i];
1882 if (v == NULL)
1883 continue;
1884 for (j=0; j < v->conf.prim_elmnt_count; j++)
1885 if (v->conf.phys_refnum[j] == dl->disk.refnum) {
1886 /* This device plays role 'j' in 'v'. */
1887 rv[n].start = __be64_to_cpu(v->lba_offset[j]);
1888 rv[n].size = __be64_to_cpu(v->conf.blocks);
1889 n++;
1890 break;
1891 }
1892 }
1893 qsort(rv, n, sizeof(*rv), cmp_extent);
1894
1895 rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
1896 rv[n].size = 0;
1897 return rv;
1898 }
1899 #endif
1900
1901 static int init_super_ddf_bvd(struct supertype *st,
1902 mdu_array_info_t *info,
1903 unsigned long long size,
1904 char *name, char *homehost,
1905 int *uuid)
1906 {
1907 /* We are creating a BVD inside a pre-existing container.
1908 * so st->sb is already set.
1909 * We need to create a new vd_config and a new virtual_entry
1910 */
1911 struct ddf_super *ddf = st->sb;
1912 int venum;
1913 struct virtual_entry *ve;
1914 struct vcl *vcl;
1915 struct vd_config *vc;
1916
1917 if (__be16_to_cpu(ddf->virt->populated_vdes)
1918 >= __be16_to_cpu(ddf->virt->max_vdes)) {
1919 fprintf(stderr, Name": This ddf already has the "
1920 "maximum of %d virtual devices\n",
1921 __be16_to_cpu(ddf->virt->max_vdes));
1922 return 0;
1923 }
1924
1925 for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1926 if (all_ff(ddf->virt->entries[venum].guid))
1927 break;
1928 if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1929 fprintf(stderr, Name ": Cannot find spare slot for "
1930 "virtual disk - DDF is corrupt\n");
1931 return 0;
1932 }
1933 ve = &ddf->virt->entries[venum];
1934
1935 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1936 * timestamp, random number
1937 */
1938 make_header_guid(ve->guid);
1939 ve->unit = __cpu_to_be16(info->md_minor);
1940 ve->pad0 = 0xFFFF;
1941 ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1942 ve->type = 0;
1943 ve->state = DDF_state_degraded; /* Will be modified as devices are added */
1944 if (info->state & 1) /* clean */
1945 ve->init_state = DDF_init_full;
1946 else
1947 ve->init_state = DDF_init_not;
1948
1949 memset(ve->pad1, 0xff, 14);
1950 memset(ve->name, ' ', 16);
1951 if (name)
1952 strncpy(ve->name, name, 16);
1953 ddf->virt->populated_vdes =
1954 __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
1955
1956 /* Now create a new vd_config */
1957 if (posix_memalign((void**)&vcl, 512,
1958 (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512)) != 0) {
1959 fprintf(stderr, Name ": %s could not allocate vd_config\n", __func__);
1960 return 0;
1961 }
1962 vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
1963 vcl->vcnum = venum;
1964 sprintf(st->subarray, "%d", venum);
1965 vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1966
1967 vc = &vcl->conf;
1968
1969 vc->magic = DDF_VD_CONF_MAGIC;
1970 memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
1971 vc->timestamp = __cpu_to_be32(time(0)-DECADE);
1972 vc->seqnum = __cpu_to_be32(1);
1973 memset(vc->pad0, 0xff, 24);
1974 vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
1975 vc->chunk_shift = chunk_to_shift(info->chunk_size);
1976 vc->prl = level_to_prl(info->level);
1977 vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
1978 vc->sec_elmnt_count = 1;
1979 vc->sec_elmnt_seq = 0;
1980 vc->srl = 0;
1981 vc->blocks = __cpu_to_be64(info->size * 2);
1982 vc->array_blocks = __cpu_to_be64(
1983 calc_array_size(info->level, info->raid_disks, info->layout,
1984 info->chunk_size, info->size*2));
1985 memset(vc->pad1, 0xff, 8);
1986 vc->spare_refs[0] = 0xffffffff;
1987 vc->spare_refs[1] = 0xffffffff;
1988 vc->spare_refs[2] = 0xffffffff;
1989 vc->spare_refs[3] = 0xffffffff;
1990 vc->spare_refs[4] = 0xffffffff;
1991 vc->spare_refs[5] = 0xffffffff;
1992 vc->spare_refs[6] = 0xffffffff;
1993 vc->spare_refs[7] = 0xffffffff;
1994 memset(vc->cache_pol, 0, 8);
1995 vc->bg_rate = 0x80;
1996 memset(vc->pad2, 0xff, 3);
1997 memset(vc->pad3, 0xff, 52);
1998 memset(vc->pad4, 0xff, 192);
1999 memset(vc->v0, 0xff, 32);
2000 memset(vc->v1, 0xff, 32);
2001 memset(vc->v2, 0xff, 16);
2002 memset(vc->v3, 0xff, 16);
2003 memset(vc->vendor, 0xff, 32);
2004
2005 memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
2006 memset(vc->phys_refnum+ddf->mppe, 0x00, 8*ddf->mppe);
2007
2008 vcl->next = ddf->conflist;
2009 ddf->conflist = vcl;
2010 ddf->currentconf = vcl;
2011 ddf->updates_pending = 1;
2012 return 1;
2013 }
2014
2015 #ifndef MDASSEMBLE
2016 static void add_to_super_ddf_bvd(struct supertype *st,
2017 mdu_disk_info_t *dk, int fd, char *devname)
2018 {
2019 /* fd and devname identify a device with-in the ddf container (st).
2020 * dk identifies a location in the new BVD.
2021 * We need to find suitable free space in that device and update
2022 * the phys_refnum and lba_offset for the newly created vd_config.
2023 * We might also want to update the type in the phys_disk
2024 * section.
2025 *
2026 * Alternately: fd == -1 and we have already chosen which device to
2027 * use and recorded in dlist->raid_disk;
2028 */
2029 struct dl *dl;
2030 struct ddf_super *ddf = st->sb;
2031 struct vd_config *vc;
2032 __u64 *lba_offset;
2033 int working;
2034 int i;
2035 unsigned long long blocks, pos, esize;
2036 struct extent *ex;
2037
2038 if (fd == -1) {
2039 for (dl = ddf->dlist; dl ; dl = dl->next)
2040 if (dl->raiddisk == dk->raid_disk)
2041 break;
2042 } else {
2043 for (dl = ddf->dlist; dl ; dl = dl->next)
2044 if (dl->major == dk->major &&
2045 dl->minor == dk->minor)
2046 break;
2047 }
2048 if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
2049 return;
2050
2051 vc = &ddf->currentconf->conf;
2052 lba_offset = ddf->currentconf->lba_offset;
2053
2054 ex = get_extents(ddf, dl);
2055 if (!ex)
2056 return;
2057
2058 i = 0; pos = 0;
2059 blocks = __be64_to_cpu(vc->blocks);
2060 if (ddf->currentconf->block_sizes)
2061 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
2062
2063 do {
2064 esize = ex[i].start - pos;
2065 if (esize >= blocks)
2066 break;
2067 pos = ex[i].start + ex[i].size;
2068 i++;
2069 } while (ex[i-1].size);
2070
2071 free(ex);
2072 if (esize < blocks)
2073 return;
2074
2075 ddf->currentdev = dk->raid_disk;
2076 vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
2077 lba_offset[dk->raid_disk] = __cpu_to_be64(pos);
2078
2079 for (i=0; i < ddf->max_part ; i++)
2080 if (dl->vlist[i] == NULL)
2081 break;
2082 if (i == ddf->max_part)
2083 return;
2084 dl->vlist[i] = ddf->currentconf;
2085
2086 if (fd >= 0)
2087 dl->fd = fd;
2088 if (devname)
2089 dl->devname = devname;
2090
2091 /* Check how many working raid_disks, and if we can mark
2092 * array as optimal yet
2093 */
2094 working = 0;
2095
2096 for (i=0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
2097 if (vc->phys_refnum[i] != 0xffffffff)
2098 working++;
2099
2100 /* Find which virtual_entry */
2101 i = ddf->currentconf->vcnum;
2102 if (working == __be16_to_cpu(vc->prim_elmnt_count))
2103 ddf->virt->entries[i].state =
2104 (ddf->virt->entries[i].state & ~DDF_state_mask)
2105 | DDF_state_optimal;
2106
2107 if (vc->prl == DDF_RAID6 &&
2108 working+1 == __be16_to_cpu(vc->prim_elmnt_count))
2109 ddf->virt->entries[i].state =
2110 (ddf->virt->entries[i].state & ~DDF_state_mask)
2111 | DDF_state_part_optimal;
2112
2113 ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
2114 ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
2115 ddf->updates_pending = 1;
2116 }
2117
2118 /* add a device to a container, either while creating it or while
2119 * expanding a pre-existing container
2120 */
2121 static int add_to_super_ddf(struct supertype *st,
2122 mdu_disk_info_t *dk, int fd, char *devname)
2123 {
2124 struct ddf_super *ddf = st->sb;
2125 struct dl *dd;
2126 time_t now;
2127 struct tm *tm;
2128 unsigned long long size;
2129 struct phys_disk_entry *pde;
2130 int n, i;
2131 struct stat stb;
2132
2133 if (ddf->currentconf) {
2134 add_to_super_ddf_bvd(st, dk, fd, devname);
2135 return 0;
2136 }
2137
2138 /* This is device numbered dk->number. We need to create
2139 * a phys_disk entry and a more detailed disk_data entry.
2140 */
2141 fstat(fd, &stb);
2142 if (posix_memalign((void**)&dd, 512,
2143 sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part) != 0) {
2144 fprintf(stderr, Name
2145 ": %s could allocate buffer for new disk, aborting\n",
2146 __func__);
2147 return 1;
2148 }
2149 dd->major = major(stb.st_rdev);
2150 dd->minor = minor(stb.st_rdev);
2151 dd->devname = devname;
2152 dd->fd = fd;
2153 dd->spare = NULL;
2154
2155 dd->disk.magic = DDF_PHYS_DATA_MAGIC;
2156 now = time(0);
2157 tm = localtime(&now);
2158 sprintf(dd->disk.guid, "%8s%04d%02d%02d",
2159 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
2160 *(__u32*)(dd->disk.guid + 16) = random32();
2161 *(__u32*)(dd->disk.guid + 20) = random32();
2162
2163 do {
2164 /* Cannot be bothered finding a CRC of some irrelevant details*/
2165 dd->disk.refnum = random32();
2166 for (i = __be16_to_cpu(ddf->active->max_pd_entries) - 1;
2167 i >= 0; i--)
2168 if (ddf->phys->entries[i].refnum == dd->disk.refnum)
2169 break;
2170 } while (i >= 0);
2171
2172 dd->disk.forced_ref = 1;
2173 dd->disk.forced_guid = 1;
2174 memset(dd->disk.vendor, ' ', 32);
2175 memcpy(dd->disk.vendor, "Linux", 5);
2176 memset(dd->disk.pad, 0xff, 442);
2177 for (i = 0; i < ddf->max_part ; i++)
2178 dd->vlist[i] = NULL;
2179
2180 n = __be16_to_cpu(ddf->phys->used_pdes);
2181 pde = &ddf->phys->entries[n];
2182 dd->pdnum = n;
2183
2184 if (st->update_tail) {
2185 int len = (sizeof(struct phys_disk) +
2186 sizeof(struct phys_disk_entry));
2187 struct phys_disk *pd;
2188
2189 pd = malloc(len);
2190 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2191 pd->used_pdes = __cpu_to_be16(n);
2192 pde = &pd->entries[0];
2193 dd->mdupdate = pd;
2194 } else {
2195 n++;
2196 ddf->phys->used_pdes = __cpu_to_be16(n);
2197 }
2198
2199 memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2200 pde->refnum = dd->disk.refnum;
2201 pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2202 pde->state = __cpu_to_be16(DDF_Online);
2203 get_dev_size(fd, NULL, &size);
2204 /* We are required to reserve 32Meg, and record the size in sectors */
2205 pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
2206 sprintf(pde->path, "%17.17s","Information: nil") ;
2207 memset(pde->pad, 0xff, 6);
2208
2209 dd->size = size >> 9;
2210 if (st->update_tail) {
2211 dd->next = ddf->add_list;
2212 ddf->add_list = dd;
2213 } else {
2214 dd->next = ddf->dlist;
2215 ddf->dlist = dd;
2216 ddf->updates_pending = 1;
2217 }
2218
2219 return 0;
2220 }
2221
2222 /*
2223 * This is the write_init_super method for a ddf container. It is
2224 * called when creating a container or adding another device to a
2225 * container.
2226 */
2227
2228 static unsigned char null_conf[4096+512];
2229
2230 static int __write_init_super_ddf(struct supertype *st, int do_close)
2231 {
2232
2233 struct ddf_super *ddf = st->sb;
2234 int i;
2235 struct dl *d;
2236 int n_config;
2237 int conf_size;
2238 int attempts = 0;
2239 int successes = 0;
2240 unsigned long long size, sector;
2241
2242 /* try to write updated metadata,
2243 * if we catch a failure move on to the next disk
2244 */
2245 for (d = ddf->dlist; d; d=d->next) {
2246 int fd = d->fd;
2247
2248 if (fd < 0)
2249 continue;
2250
2251 attempts++;
2252 /* We need to fill in the primary, (secondary) and workspace
2253 * lba's in the headers, set their checksums,
2254 * Also checksum phys, virt....
2255 *
2256 * Then write everything out, finally the anchor is written.
2257 */
2258 get_dev_size(fd, NULL, &size);
2259 size /= 512;
2260 ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
2261 ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
2262 ddf->anchor.seq = __cpu_to_be32(1);
2263 memcpy(&ddf->primary, &ddf->anchor, 512);
2264 memcpy(&ddf->secondary, &ddf->anchor, 512);
2265
2266 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
2267 ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
2268 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
2269
2270 ddf->primary.openflag = 0;
2271 ddf->primary.type = DDF_HEADER_PRIMARY;
2272
2273 ddf->secondary.openflag = 0;
2274 ddf->secondary.type = DDF_HEADER_SECONDARY;
2275
2276 ddf->primary.crc = calc_crc(&ddf->primary, 512);
2277 ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
2278
2279 sector = size - 16*1024*2;
2280 lseek64(fd, sector<<9, 0);
2281 if (write(fd, &ddf->primary, 512) < 0)
2282 continue;
2283
2284 ddf->controller.crc = calc_crc(&ddf->controller, 512);
2285 if (write(fd, &ddf->controller, 512) < 0)
2286 continue;
2287
2288 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
2289
2290 if (write(fd, ddf->phys, ddf->pdsize) < 0)
2291 continue;
2292
2293 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
2294 if (write(fd, ddf->virt, ddf->vdsize) < 0)
2295 continue;
2296
2297 /* Now write lots of config records. */
2298 n_config = ddf->max_part;
2299 conf_size = ddf->conf_rec_len * 512;
2300 for (i = 0 ; i <= n_config ; i++) {
2301 struct vcl *c = d->vlist[i];
2302 if (i == n_config)
2303 c = (struct vcl*)d->spare;
2304
2305 if (c) {
2306 c->conf.crc = calc_crc(&c->conf, conf_size);
2307 if (write(fd, &c->conf, conf_size) < 0)
2308 break;
2309 } else {
2310 char *null_aligned = (char*)((((unsigned long)null_conf)+511)&~511UL);
2311 if (null_conf[0] != 0xff)
2312 memset(null_conf, 0xff, sizeof(null_conf));
2313 int togo = conf_size;
2314 while (togo > sizeof(null_conf)-512) {
2315 if (write(fd, null_aligned, sizeof(null_conf)-512) < 0)
2316 break;
2317 togo -= sizeof(null_conf)-512;
2318 }
2319 if (write(fd, null_aligned, togo) < 0)
2320 break;
2321 }
2322 }
2323 if (i <= n_config)
2324 continue;
2325 d->disk.crc = calc_crc(&d->disk, 512);
2326 if (write(fd, &d->disk, 512) < 0)
2327 continue;
2328
2329 /* Maybe do the same for secondary */
2330
2331 lseek64(fd, (size-1)*512, SEEK_SET);
2332 if (write(fd, &ddf->anchor, 512) < 0)
2333 continue;
2334 successes++;
2335 }
2336
2337 if (do_close)
2338 for (d = ddf->dlist; d; d=d->next) {
2339 close(d->fd);
2340 d->fd = -1;
2341 }
2342
2343 return attempts != successes;
2344 }
2345
2346 static int write_init_super_ddf(struct supertype *st)
2347 {
2348
2349 if (st->update_tail) {
2350 /* queue the virtual_disk and vd_config as metadata updates */
2351 struct virtual_disk *vd;
2352 struct vd_config *vc;
2353 struct ddf_super *ddf = st->sb;
2354 int len;
2355
2356 if (!ddf->currentconf) {
2357 int len = (sizeof(struct phys_disk) +
2358 sizeof(struct phys_disk_entry));
2359
2360 /* adding a disk to the container. */
2361 if (!ddf->add_list)
2362 return 0;
2363
2364 append_metadata_update(st, ddf->add_list->mdupdate, len);
2365 ddf->add_list->mdupdate = NULL;
2366 return 0;
2367 }
2368
2369 /* Newly created VD */
2370
2371 /* First the virtual disk. We have a slightly fake header */
2372 len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry);
2373 vd = malloc(len);
2374 *vd = *ddf->virt;
2375 vd->entries[0] = ddf->virt->entries[ddf->currentconf->vcnum];
2376 vd->populated_vdes = __cpu_to_be16(ddf->currentconf->vcnum);
2377 append_metadata_update(st, vd, len);
2378
2379 /* Then the vd_config */
2380 len = ddf->conf_rec_len * 512;
2381 vc = malloc(len);
2382 memcpy(vc, &ddf->currentconf->conf, len);
2383 append_metadata_update(st, vc, len);
2384
2385 /* FIXME I need to close the fds! */
2386 return 0;
2387 } else
2388 return __write_init_super_ddf(st, 1);
2389 }
2390
2391 #endif
2392
2393 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize)
2394 {
2395 /* We must reserve the last 32Meg */
2396 if (devsize <= 32*1024*2)
2397 return 0;
2398 return devsize - 32*1024*2;
2399 }
2400
2401 #ifndef MDASSEMBLE
2402
2403 static int reserve_space(struct supertype *st, int raiddisks,
2404 unsigned long long size, int chunk,
2405 unsigned long long *freesize)
2406 {
2407 /* Find 'raiddisks' spare extents at least 'size' big (but
2408 * only caring about multiples of 'chunk') and remember
2409 * them.
2410 * If the cannot be found, fail.
2411 */
2412 struct dl *dl;
2413 struct ddf_super *ddf = st->sb;
2414 int cnt = 0;
2415
2416 for (dl = ddf->dlist; dl ; dl=dl->next) {
2417 dl->raiddisk = -1;
2418 dl->esize = 0;
2419 }
2420 /* Now find largest extent on each device */
2421 for (dl = ddf->dlist ; dl ; dl=dl->next) {
2422 struct extent *e = get_extents(ddf, dl);
2423 unsigned long long pos = 0;
2424 int i = 0;
2425 int found = 0;
2426 unsigned long long minsize = size;
2427
2428 if (size == 0)
2429 minsize = chunk;
2430
2431 if (!e)
2432 continue;
2433 do {
2434 unsigned long long esize;
2435 esize = e[i].start - pos;
2436 if (esize >= minsize) {
2437 found = 1;
2438 minsize = esize;
2439 }
2440 pos = e[i].start + e[i].size;
2441 i++;
2442 } while (e[i-1].size);
2443 if (found) {
2444 cnt++;
2445 dl->esize = minsize;
2446 }
2447 free(e);
2448 }
2449 if (cnt < raiddisks) {
2450 fprintf(stderr, Name ": not enough devices with space to create array.\n");
2451 return 0; /* No enough free spaces large enough */
2452 }
2453 if (size == 0) {
2454 /* choose the largest size of which there are at least 'raiddisk' */
2455 for (dl = ddf->dlist ; dl ; dl=dl->next) {
2456 struct dl *dl2;
2457 if (dl->esize <= size)
2458 continue;
2459 /* This is bigger than 'size', see if there are enough */
2460 cnt = 0;
2461 for (dl2 = dl; dl2 ; dl2=dl2->next)
2462 if (dl2->esize >= dl->esize)
2463 cnt++;
2464 if (cnt >= raiddisks)
2465 size = dl->esize;
2466 }
2467 if (chunk) {
2468 size = size / chunk;
2469 size *= chunk;
2470 }
2471 *freesize = size;
2472 if (size < 32) {
2473 fprintf(stderr, Name ": not enough spare devices to create array.\n");
2474 return 0;
2475 }
2476 }
2477 /* We have a 'size' of which there are enough spaces.
2478 * We simply do a first-fit */
2479 cnt = 0;
2480 for (dl = ddf->dlist ; dl && cnt < raiddisks ; dl=dl->next) {
2481 if (dl->esize < size)
2482 continue;
2483
2484 dl->raiddisk = cnt;
2485 cnt++;
2486 }
2487 return 1;
2488 }
2489
2490
2491
2492 static int
2493 validate_geometry_ddf_container(struct supertype *st,
2494 int level, int layout, int raiddisks,
2495 int chunk, unsigned long long size,
2496 char *dev, unsigned long long *freesize,
2497 int verbose);
2498
2499 static int validate_geometry_ddf_bvd(struct supertype *st,
2500 int level, int layout, int raiddisks,
2501 int chunk, unsigned long long size,
2502 char *dev, unsigned long long *freesize,
2503 int verbose);
2504
2505 static int validate_geometry_ddf(struct supertype *st,
2506 int level, int layout, int raiddisks,
2507 int chunk, unsigned long long size,
2508 char *dev, unsigned long long *freesize,
2509 int verbose)
2510 {
2511 int fd;
2512 struct mdinfo *sra;
2513 int cfd;
2514
2515 /* ddf potentially supports lots of things, but it depends on
2516 * what devices are offered (and maybe kernel version?)
2517 * If given unused devices, we will make a container.
2518 * If given devices in a container, we will make a BVD.
2519 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2520 */
2521
2522 if (level == LEVEL_CONTAINER) {
2523 /* Must be a fresh device to add to a container */
2524 return validate_geometry_ddf_container(st, level, layout,
2525 raiddisks, chunk,
2526 size, dev, freesize,
2527 verbose);
2528 }
2529
2530 if (!dev) {
2531 /* Initial sanity check. Exclude illegal levels. */
2532 int i;
2533 for (i=0; ddf_level_num[i].num1 != MAXINT; i++)
2534 if (ddf_level_num[i].num2 == level)
2535 break;
2536 if (ddf_level_num[i].num1 == MAXINT)
2537 return 0;
2538 /* Should check layout? etc */
2539
2540 if (st->sb && freesize) {
2541 /* --create was given a container to create in.
2542 * So we need to check that there are enough
2543 * free spaces and return the amount of space.
2544 * We may as well remember which drives were
2545 * chosen so that add_to_super/getinfo_super
2546 * can return them.
2547 */
2548 return reserve_space(st, raiddisks, size, chunk, freesize);
2549 }
2550 return 1;
2551 }
2552
2553 if (st->sb) {
2554 /* A container has already been opened, so we are
2555 * creating in there. Maybe a BVD, maybe an SVD.
2556 * Should make a distinction one day.
2557 */
2558 return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
2559 chunk, size, dev, freesize,
2560 verbose);
2561 }
2562 /* This is the first device for the array.
2563 * If it is a container, we read it in and do automagic allocations,
2564 * no other devices should be given.
2565 * Otherwise it must be a member device of a container, and we
2566 * do manual allocation.
2567 * Later we should check for a BVD and make an SVD.
2568 */
2569 fd = open(dev, O_RDONLY|O_EXCL, 0);
2570 if (fd >= 0) {
2571 sra = sysfs_read(fd, 0, GET_VERSION);
2572 close(fd);
2573 if (sra && sra->array.major_version == -1 &&
2574 strcmp(sra->text_version, "ddf") == 0) {
2575
2576 /* load super */
2577 /* find space for 'n' devices. */
2578 /* remember the devices */
2579 /* Somehow return the fact that we have enough */
2580 }
2581
2582 if (verbose)
2583 fprintf(stderr,
2584 Name ": ddf: Cannot create this array "
2585 "on device %s\n",
2586 dev);
2587 return 0;
2588 }
2589 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2590 if (verbose)
2591 fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2592 dev, strerror(errno));
2593 return 0;
2594 }
2595 /* Well, it is in use by someone, maybe a 'ddf' container. */
2596 cfd = open_container(fd);
2597 if (cfd < 0) {
2598 close(fd);
2599 if (verbose)
2600 fprintf(stderr, Name ": ddf: Cannot use %s: %s\n",
2601 dev, strerror(EBUSY));
2602 return 0;
2603 }
2604 sra = sysfs_read(cfd, 0, GET_VERSION);
2605 close(fd);
2606 if (sra && sra->array.major_version == -1 &&
2607 strcmp(sra->text_version, "ddf") == 0) {
2608 /* This is a member of a ddf container. Load the container
2609 * and try to create a bvd
2610 */
2611 struct ddf_super *ddf;
2612 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL, 1) == 0) {
2613 st->sb = ddf;
2614 st->container_dev = fd2devnum(cfd);
2615 close(cfd);
2616 return validate_geometry_ddf_bvd(st, level, layout,
2617 raiddisks, chunk, size,
2618 dev, freesize,
2619 verbose);
2620 }
2621 close(cfd);
2622 } else /* device may belong to a different container */
2623 return 0;
2624
2625 return 1;
2626 }
2627
2628 static int
2629 validate_geometry_ddf_container(struct supertype *st,
2630 int level, int layout, int raiddisks,
2631 int chunk, unsigned long long size,
2632 char *dev, unsigned long long *freesize,
2633 int verbose)
2634 {
2635 int fd;
2636 unsigned long long ldsize;
2637
2638 if (level != LEVEL_CONTAINER)
2639 return 0;
2640 if (!dev)
2641 return 1;
2642
2643 fd = open(dev, O_RDONLY|O_EXCL, 0);
2644 if (fd < 0) {
2645 if (verbose)
2646 fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2647 dev, strerror(errno));
2648 return 0;
2649 }
2650 if (!get_dev_size(fd, dev, &ldsize)) {
2651 close(fd);
2652 return 0;
2653 }
2654 close(fd);
2655
2656 *freesize = avail_size_ddf(st, ldsize >> 9);
2657 if (*freesize == 0)
2658 return 0;
2659
2660 return 1;
2661 }
2662
2663 static int validate_geometry_ddf_bvd(struct supertype *st,
2664 int level, int layout, int raiddisks,
2665 int chunk, unsigned long long size,
2666 char *dev, unsigned long long *freesize,
2667 int verbose)
2668 {
2669 struct stat stb;
2670 struct ddf_super *ddf = st->sb;
2671 struct dl *dl;
2672 unsigned long long pos = 0;
2673 unsigned long long maxsize;
2674 struct extent *e;
2675 int i;
2676 /* ddf/bvd supports lots of things, but not containers */
2677 if (level == LEVEL_CONTAINER)
2678 return 0;
2679 /* We must have the container info already read in. */
2680 if (!ddf)
2681 return 0;
2682
2683 if (!dev) {
2684 /* General test: make sure there is space for
2685 * 'raiddisks' device extents of size 'size'.
2686 */
2687 unsigned long long minsize = size;
2688 int dcnt = 0;
2689 if (minsize == 0)
2690 minsize = 8;
2691 for (dl = ddf->dlist; dl ; dl = dl->next)
2692 {
2693 int found = 0;
2694 pos = 0;
2695
2696 i = 0;
2697 e = get_extents(ddf, dl);
2698 if (!e) continue;
2699 do {
2700 unsigned long long esize;
2701 esize = e[i].start - pos;
2702 if (esize >= minsize)
2703 found = 1;
2704 pos = e[i].start + e[i].size;
2705 i++;
2706 } while (e[i-1].size);
2707 if (found)
2708 dcnt++;
2709 free(e);
2710 }
2711 if (dcnt < raiddisks) {
2712 if (verbose)
2713 fprintf(stderr,
2714 Name ": ddf: Not enough devices with "
2715 "space for this array (%d < %d)\n",
2716 dcnt, raiddisks);
2717 return 0;
2718 }
2719 return 1;
2720 }
2721 /* This device must be a member of the set */
2722 if (stat(dev, &stb) < 0)
2723 return 0;
2724 if ((S_IFMT & stb.st_mode) != S_IFBLK)
2725 return 0;
2726 for (dl = ddf->dlist ; dl ; dl = dl->next) {
2727 if (dl->major == major(stb.st_rdev) &&
2728 dl->minor == minor(stb.st_rdev))
2729 break;
2730 }
2731 if (!dl) {
2732 if (verbose)
2733 fprintf(stderr, Name ": ddf: %s is not in the "
2734 "same DDF set\n",
2735 dev);
2736 return 0;
2737 }
2738 e = get_extents(ddf, dl);
2739 maxsize = 0;
2740 i = 0;
2741 if (e) do {
2742 unsigned long long esize;
2743 esize = e[i].start - pos;
2744 if (esize >= maxsize)
2745 maxsize = esize;
2746 pos = e[i].start + e[i].size;
2747 i++;
2748 } while (e[i-1].size);
2749 *freesize = maxsize;
2750 // FIXME here I am
2751
2752 return 1;
2753 }
2754
2755 static int load_super_ddf_all(struct supertype *st, int fd,
2756 void **sbp, char *devname, int keep_fd)
2757 {
2758 struct mdinfo *sra;
2759 struct ddf_super *super;
2760 struct mdinfo *sd, *best = NULL;
2761 int bestseq = 0;
2762 int seq;
2763 char nm[20];
2764 int dfd;
2765 int devnum = fd2devnum(fd);
2766 enum sysfs_read_flags flags;
2767
2768 flags = GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE;
2769 if (mdmon_running(devnum))
2770 flags |= SKIP_GONE_DEVS;
2771
2772 sra = sysfs_read(fd, 0, flags);
2773 if (!sra)
2774 return 1;
2775 if (sra->array.major_version != -1 ||
2776 sra->array.minor_version != -2 ||
2777 strcmp(sra->text_version, "ddf") != 0)
2778 return 1;
2779
2780 if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0)
2781 return 1;
2782 memset(super, 0, sizeof(*super));
2783
2784 /* first, try each device, and choose the best ddf */
2785 for (sd = sra->devs ; sd ; sd = sd->next) {
2786 int rv;
2787 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2788 dfd = dev_open(nm, O_RDONLY);
2789 if (dfd < 0)
2790 return 2;
2791 rv = load_ddf_headers(dfd, super, NULL);
2792 close(dfd);
2793 if (rv == 0) {
2794 seq = __be32_to_cpu(super->active->seq);
2795 if (super->active->openflag)
2796 seq--;
2797 if (!best || seq > bestseq) {
2798 bestseq = seq;
2799 best = sd;
2800 }
2801 }
2802 }
2803 if (!best)
2804 return 1;
2805 /* OK, load this ddf */
2806 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2807 dfd = dev_open(nm, O_RDONLY);
2808 if (dfd < 0)
2809 return 1;
2810 load_ddf_headers(dfd, super, NULL);
2811 load_ddf_global(dfd, super, NULL);
2812 close(dfd);
2813 /* Now we need the device-local bits */
2814 for (sd = sra->devs ; sd ; sd = sd->next) {
2815 int rv;
2816
2817 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2818 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2819 if (dfd < 0)
2820 return 2;
2821 rv = load_ddf_headers(dfd, super, NULL);
2822 if (rv == 0)
2823 rv = load_ddf_local(dfd, super, NULL, keep_fd);
2824 if (!keep_fd) close(dfd);
2825 if (rv)
2826 return 1;
2827 }
2828 if (st->subarray[0]) {
2829 struct vcl *v;
2830
2831 for (v = super->conflist; v; v = v->next)
2832 if (v->vcnum == atoi(st->subarray))
2833 super->currentconf = v;
2834 if (!super->currentconf)
2835 return 1;
2836 }
2837 *sbp = super;
2838 if (st->ss == NULL) {
2839 st->ss = &super_ddf;
2840 st->minor_version = 0;
2841 st->max_devs = 512;
2842 st->container_dev = fd2devnum(fd);
2843 }
2844 st->loaded_container = 1;
2845 return 0;
2846 }
2847 #endif /* MDASSEMBLE */
2848
2849 static struct mdinfo *container_content_ddf(struct supertype *st)
2850 {
2851 /* Given a container loaded by load_super_ddf_all,
2852 * extract information about all the arrays into
2853 * an mdinfo tree.
2854 *
2855 * For each vcl in conflist: create an mdinfo, fill it in,
2856 * then look for matching devices (phys_refnum) in dlist
2857 * and create appropriate device mdinfo.
2858 */
2859 struct ddf_super *ddf = st->sb;
2860 struct mdinfo *rest = NULL;
2861 struct vcl *vc;
2862
2863 for (vc = ddf->conflist ; vc ; vc=vc->next)
2864 {
2865 int i;
2866 int j;
2867 struct mdinfo *this;
2868 this = malloc(sizeof(*this));
2869 memset(this, 0, sizeof(*this));
2870 this->next = rest;
2871 rest = this;
2872
2873 this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2874 this->array.raid_disks =
2875 __be16_to_cpu(vc->conf.prim_elmnt_count);
2876 this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2877 this->array.raid_disks);
2878 this->array.md_minor = -1;
2879 this->array.major_version = -1;
2880 this->array.minor_version = -2;
2881 this->array.ctime = DECADE +
2882 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
2883 this->array.utime = DECADE +
2884 __be32_to_cpu(vc->conf.timestamp);
2885 this->array.chunk_size = 512 << vc->conf.chunk_shift;
2886
2887 i = vc->vcnum;
2888 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2889 (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
2890 DDF_init_full) {
2891 this->array.state = 0;
2892 this->resync_start = 0;
2893 } else {
2894 this->array.state = 1;
2895 this->resync_start = ~0ULL;
2896 }
2897 memcpy(this->name, ddf->virt->entries[i].name, 16);
2898 this->name[16]=0;
2899 for(j=0; j<16; j++)
2900 if (this->name[j] == ' ')
2901 this->name[j] = 0;
2902
2903 memset(this->uuid, 0, sizeof(this->uuid));
2904 this->component_size = __be64_to_cpu(vc->conf.blocks);
2905 this->array.size = this->component_size / 2;
2906 this->container_member = i;
2907
2908 ddf->currentconf = vc;
2909 uuid_from_super_ddf(st, this->uuid);
2910 ddf->currentconf = NULL;
2911
2912 sprintf(this->text_version, "/%s/%d",
2913 devnum2devname(st->container_dev),
2914 this->container_member);
2915
2916 for (i=0 ; i < ddf->mppe ; i++) {
2917 struct mdinfo *dev;
2918 struct dl *d;
2919
2920 if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
2921 continue;
2922
2923 this->array.working_disks++;
2924
2925 for (d = ddf->dlist; d ; d=d->next)
2926 if (d->disk.refnum == vc->conf.phys_refnum[i])
2927 break;
2928 if (d == NULL)
2929 /* Haven't found that one yet, maybe there are others */
2930 continue;
2931
2932 dev = malloc(sizeof(*dev));
2933 memset(dev, 0, sizeof(*dev));
2934 dev->next = this->devs;
2935 this->devs = dev;
2936
2937 dev->disk.number = __be32_to_cpu(d->disk.refnum);
2938 dev->disk.major = d->major;
2939 dev->disk.minor = d->minor;
2940 dev->disk.raid_disk = i;
2941 dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2942
2943 dev->events = __be32_to_cpu(ddf->primary.seq);
2944 dev->data_offset = __be64_to_cpu(vc->lba_offset[i]);
2945 dev->component_size = __be64_to_cpu(vc->conf.blocks);
2946 if (d->devname)
2947 strcpy(dev->name, d->devname);
2948 }
2949 }
2950 return rest;
2951 }
2952
2953 static int store_zero_ddf(struct supertype *st, int fd)
2954 {
2955 unsigned long long dsize;
2956 void *buf;
2957 int rc;
2958
2959 if (!get_dev_size(fd, NULL, &dsize))
2960 return 1;
2961
2962 if (posix_memalign(&buf, 512, 512) != 0)
2963 return 1;
2964 memset(buf, 0, 512);
2965
2966 lseek64(fd, dsize-512, 0);
2967 rc = write(fd, buf, 512);
2968 free(buf);
2969 if (rc < 0)
2970 return 1;
2971 return 0;
2972 }
2973
2974 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
2975 {
2976 /*
2977 * return:
2978 * 0 same, or first was empty, and second was copied
2979 * 1 second had wrong number
2980 * 2 wrong uuid
2981 * 3 wrong other info
2982 */
2983 struct ddf_super *first = st->sb;
2984 struct ddf_super *second = tst->sb;
2985
2986 if (!first) {
2987 st->sb = tst->sb;
2988 tst->sb = NULL;
2989 return 0;
2990 }
2991
2992 if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
2993 return 2;
2994
2995 /* FIXME should I look at anything else? */
2996 return 0;
2997 }
2998
2999 #ifndef MDASSEMBLE
3000 /*
3001 * A new array 'a' has been started which claims to be instance 'inst'
3002 * within container 'c'.
3003 * We need to confirm that the array matches the metadata in 'c' so
3004 * that we don't corrupt any metadata.
3005 */
3006 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
3007 {
3008 dprintf("ddf: open_new %s\n", inst);
3009 a->info.container_member = atoi(inst);
3010 return 0;
3011 }
3012
3013 /*
3014 * The array 'a' is to be marked clean in the metadata.
3015 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
3016 * clean up to the point (in sectors). If that cannot be recorded in the
3017 * metadata, then leave it as dirty.
3018 *
3019 * For DDF, we need to clear the DDF_state_inconsistent bit in the
3020 * !global! virtual_disk.virtual_entry structure.
3021 */
3022 static int ddf_set_array_state(struct active_array *a, int consistent)
3023 {
3024 struct ddf_super *ddf = a->container->sb;
3025 int inst = a->info.container_member;
3026 int old = ddf->virt->entries[inst].state;
3027 if (consistent == 2) {
3028 /* Should check if a recovery should be started FIXME */
3029 consistent = 1;
3030 if (!is_resync_complete(a))
3031 consistent = 0;
3032 }
3033 if (consistent)
3034 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
3035 else
3036 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
3037 if (old != ddf->virt->entries[inst].state)
3038 ddf->updates_pending = 1;
3039
3040 old = ddf->virt->entries[inst].init_state;
3041 ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
3042 if (is_resync_complete(a))
3043 ddf->virt->entries[inst].init_state |= DDF_init_full;
3044 else if (a->resync_start == 0)
3045 ddf->virt->entries[inst].init_state |= DDF_init_not;
3046 else
3047 ddf->virt->entries[inst].init_state |= DDF_init_quick;
3048 if (old != ddf->virt->entries[inst].init_state)
3049 ddf->updates_pending = 1;
3050
3051 dprintf("ddf mark %d %s %llu\n", inst, consistent?"clean":"dirty",
3052 a->resync_start);
3053 return consistent;
3054 }
3055
3056 /*
3057 * The state of each disk is stored in the global phys_disk structure
3058 * in phys_disk.entries[n].state.
3059 * This makes various combinations awkward.
3060 * - When a device fails in any array, it must be failed in all arrays
3061 * that include a part of this device.
3062 * - When a component is rebuilding, we cannot include it officially in the
3063 * array unless this is the only array that uses the device.
3064 *
3065 * So: when transitioning:
3066 * Online -> failed, just set failed flag. monitor will propagate
3067 * spare -> online, the device might need to be added to the array.
3068 * spare -> failed, just set failed. Don't worry if in array or not.
3069 */
3070 static void ddf_set_disk(struct active_array *a, int n, int state)
3071 {
3072 struct ddf_super *ddf = a->container->sb;
3073 int inst = a->info.container_member;
3074 struct vd_config *vc = find_vdcr(ddf, inst);
3075 int pd = find_phys(ddf, vc->phys_refnum[n]);
3076 int i, st, working;
3077
3078 if (vc == NULL) {
3079 dprintf("ddf: cannot find instance %d!!\n", inst);
3080 return;
3081 }
3082 if (pd < 0) {
3083 /* disk doesn't currently exist. If it is now in_sync,
3084 * insert it. */
3085 if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
3086 /* Find dev 'n' in a->info->devs, determine the
3087 * ddf refnum, and set vc->phys_refnum and update
3088 * phys->entries[]
3089 */
3090 /* FIXME */
3091 }
3092 } else {
3093 int old = ddf->phys->entries[pd].state;
3094 if (state & DS_FAULTY)
3095 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Failed);
3096 if (state & DS_INSYNC) {
3097 ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Online);
3098 ddf->phys->entries[pd].state &= __cpu_to_be16(~DDF_Rebuilding);
3099 }
3100 if (old != ddf->phys->entries[pd].state)
3101 ddf->updates_pending = 1;
3102 }
3103
3104 dprintf("ddf: set_disk %d to %x\n", n, state);
3105
3106 /* Now we need to check the state of the array and update
3107 * virtual_disk.entries[n].state.
3108 * It needs to be one of "optimal", "degraded", "failed".
3109 * I don't understand 'deleted' or 'missing'.
3110 */
3111 working = 0;
3112 for (i=0; i < a->info.array.raid_disks; i++) {
3113 pd = find_phys(ddf, vc->phys_refnum[i]);
3114 if (pd < 0)
3115 continue;
3116 st = __be16_to_cpu(ddf->phys->entries[pd].state);
3117 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
3118 == DDF_Online)
3119 working++;
3120 }
3121 state = DDF_state_degraded;
3122 if (working == a->info.array.raid_disks)
3123 state = DDF_state_optimal;
3124 else switch(vc->prl) {
3125 case DDF_RAID0:
3126 case DDF_CONCAT:
3127 case DDF_JBOD:
3128 state = DDF_state_failed;
3129 break;
3130 case DDF_RAID1:
3131 if (working == 0)
3132 state = DDF_state_failed;
3133 break;
3134 case DDF_RAID4:
3135 case DDF_RAID5:
3136 if (working < a->info.array.raid_disks-1)
3137 state = DDF_state_failed;
3138 break;
3139 case DDF_RAID6:
3140 if (working < a->info.array.raid_disks-2)
3141 state = DDF_state_failed;
3142 else if (working == a->info.array.raid_disks-1)
3143 state = DDF_state_part_optimal;
3144 break;
3145 }
3146
3147 if (ddf->virt->entries[inst].state !=
3148 ((ddf->virt->entries[inst].state & ~DDF_state_mask)
3149 | state)) {
3150
3151 ddf->virt->entries[inst].state =
3152 (ddf->virt->entries[inst].state & ~DDF_state_mask)
3153 | state;
3154 ddf->updates_pending = 1;
3155 }
3156
3157 }
3158
3159 static void ddf_sync_metadata(struct supertype *st)
3160 {
3161
3162 /*
3163 * Write all data to all devices.
3164 * Later, we might be able to track whether only local changes
3165 * have been made, or whether any global data has been changed,
3166 * but ddf is sufficiently weird that it probably always
3167 * changes global data ....
3168 */
3169 struct ddf_super *ddf = st->sb;
3170 if (!ddf->updates_pending)
3171 return;
3172 ddf->updates_pending = 0;
3173 __write_init_super_ddf(st, 0);
3174 dprintf("ddf: sync_metadata\n");
3175 }
3176
3177 static void ddf_process_update(struct supertype *st,
3178 struct metadata_update *update)
3179 {
3180 /* Apply this update to the metadata.
3181 * The first 4 bytes are a DDF_*_MAGIC which guides
3182 * our actions.
3183 * Possible update are:
3184 * DDF_PHYS_RECORDS_MAGIC
3185 * Add a new physical device. Changes to this record
3186 * only happen implicitly.
3187 * used_pdes is the device number.
3188 * DDF_VIRT_RECORDS_MAGIC
3189 * Add a new VD. Possibly also change the 'access' bits.
3190 * populated_vdes is the entry number.
3191 * DDF_VD_CONF_MAGIC
3192 * New or updated VD. the VIRT_RECORD must already
3193 * exist. For an update, phys_refnum and lba_offset
3194 * (at least) are updated, and the VD_CONF must
3195 * be written to precisely those devices listed with
3196 * a phys_refnum.
3197 * DDF_SPARE_ASSIGN_MAGIC
3198 * replacement Spare Assignment Record... but for which device?
3199 *
3200 * So, e.g.:
3201 * - to create a new array, we send a VIRT_RECORD and
3202 * a VD_CONF. Then assemble and start the array.
3203 * - to activate a spare we send a VD_CONF to add the phys_refnum
3204 * and offset. This will also mark the spare as active with
3205 * a spare-assignment record.
3206 */
3207 struct ddf_super *ddf = st->sb;
3208 __u32 *magic = (__u32*)update->buf;
3209 struct phys_disk *pd;
3210 struct virtual_disk *vd;
3211 struct vd_config *vc;
3212 struct vcl *vcl;
3213 struct dl *dl;
3214 int mppe;
3215 int ent;
3216
3217 dprintf("Process update %x\n", *magic);
3218
3219 switch (*magic) {
3220 case DDF_PHYS_RECORDS_MAGIC:
3221
3222 if (update->len != (sizeof(struct phys_disk) +
3223 sizeof(struct phys_disk_entry)))
3224 return;
3225 pd = (struct phys_disk*)update->buf;
3226
3227 ent = __be16_to_cpu(pd->used_pdes);
3228 if (ent >= __be16_to_cpu(ddf->phys->max_pdes))
3229 return;
3230 if (!all_ff(ddf->phys->entries[ent].guid))
3231 return;
3232 ddf->phys->entries[ent] = pd->entries[0];
3233 ddf->phys->used_pdes = __cpu_to_be16(1 +
3234 __be16_to_cpu(ddf->phys->used_pdes));
3235 ddf->updates_pending = 1;
3236 if (ddf->add_list) {
3237 struct active_array *a;
3238 struct dl *al = ddf->add_list;
3239 ddf->add_list = al->next;
3240
3241 al->next = ddf->dlist;
3242 ddf->dlist = al;
3243
3244 /* As a device has been added, we should check
3245 * for any degraded devices that might make
3246 * use of this spare */
3247 for (a = st->arrays ; a; a=a->next)
3248 a->check_degraded = 1;
3249 }
3250 break;
3251
3252 case DDF_VIRT_RECORDS_MAGIC:
3253
3254 if (update->len != (sizeof(struct virtual_disk) +
3255 sizeof(struct virtual_entry)))
3256 return;
3257 vd = (struct virtual_disk*)update->buf;
3258
3259 ent = __be16_to_cpu(vd->populated_vdes);
3260 if (ent >= __be16_to_cpu(ddf->virt->max_vdes))
3261 return;
3262 if (!all_ff(ddf->virt->entries[ent].guid))
3263 return;
3264 ddf->virt->entries[ent] = vd->entries[0];
3265 ddf->virt->populated_vdes = __cpu_to_be16(1 +
3266 __be16_to_cpu(ddf->virt->populated_vdes));
3267 ddf->updates_pending = 1;
3268 break;
3269
3270 case DDF_VD_CONF_MAGIC:
3271 dprintf("len %d %d\n", update->len, ddf->conf_rec_len);
3272
3273 mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries);
3274 if (update->len != ddf->conf_rec_len * 512)
3275 return;
3276 vc = (struct vd_config*)update->buf;
3277 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
3278 if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
3279 break;
3280 dprintf("vcl = %p\n", vcl);
3281 if (vcl) {
3282 /* An update, just copy the phys_refnum and lba_offset
3283 * fields
3284 */
3285 memcpy(vcl->conf.phys_refnum, vc->phys_refnum,
3286 mppe * (sizeof(__u32) + sizeof(__u64)));
3287 } else {
3288 /* A new VD_CONF */
3289 if (!update->space)
3290 return;
3291 vcl = update->space;
3292 update->space = NULL;
3293 vcl->next = ddf->conflist;
3294 memcpy(&vcl->conf, vc, update->len);
3295 vcl->lba_offset = (__u64*)
3296 &vcl->conf.phys_refnum[mppe];
3297 ddf->conflist = vcl;
3298 }
3299 /* Now make sure vlist is correct for each dl. */
3300 for (dl = ddf->dlist; dl; dl = dl->next) {
3301 int dn;
3302 int vn = 0;
3303 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
3304 for (dn=0; dn < ddf->mppe ; dn++)
3305 if (vcl->conf.phys_refnum[dn] ==
3306 dl->disk.refnum) {
3307 dprintf("dev %d has %p at %d\n",
3308 dl->pdnum, vcl, vn);
3309 dl->vlist[vn++] = vcl;
3310 break;
3311 }
3312 while (vn < ddf->max_part)
3313 dl->vlist[vn++] = NULL;
3314 if (dl->vlist[0]) {
3315 ddf->phys->entries[dl->pdnum].type &=
3316 ~__cpu_to_be16(DDF_Global_Spare);
3317 ddf->phys->entries[dl->pdnum].type |=
3318 __cpu_to_be16(DDF_Active_in_VD);
3319 }
3320 if (dl->spare) {
3321 ddf->phys->entries[dl->pdnum].type &=
3322 ~__cpu_to_be16(DDF_Global_Spare);
3323 ddf->phys->entries[dl->pdnum].type |=
3324 __cpu_to_be16(DDF_Spare);
3325 }
3326 if (!dl->vlist[0] && !dl->spare) {
3327 ddf->phys->entries[dl->pdnum].type |=
3328 __cpu_to_be16(DDF_Global_Spare);
3329 ddf->phys->entries[dl->pdnum].type &=
3330 ~__cpu_to_be16(DDF_Spare |
3331 DDF_Active_in_VD);
3332 }
3333 }
3334 ddf->updates_pending = 1;
3335 break;
3336 case DDF_SPARE_ASSIGN_MAGIC:
3337 default: break;
3338 }
3339 }
3340
3341 static void ddf_prepare_update(struct supertype *st,
3342 struct metadata_update *update)
3343 {
3344 /* This update arrived at managemon.
3345 * We are about to pass it to monitor.
3346 * If a malloc is needed, do it here.
3347 */
3348 struct ddf_super *ddf = st->sb;
3349 __u32 *magic = (__u32*)update->buf;
3350 if (*magic == DDF_VD_CONF_MAGIC)
3351 if (posix_memalign(&update->space, 512,
3352 offsetof(struct vcl, conf)
3353 + ddf->conf_rec_len * 512) != 0)
3354 update->space = NULL;
3355 }
3356
3357 /*
3358 * Check if the array 'a' is degraded but not failed.
3359 * If it is, find as many spares as are available and needed and
3360 * arrange for their inclusion.
3361 * We only choose devices which are not already in the array,
3362 * and prefer those with a spare-assignment to this array.
3363 * otherwise we choose global spares - assuming always that
3364 * there is enough room.
3365 * For each spare that we assign, we return an 'mdinfo' which
3366 * describes the position for the device in the array.
3367 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
3368 * the new phys_refnum and lba_offset values.
3369 *
3370 * Only worry about BVDs at the moment.
3371 */
3372 static struct mdinfo *ddf_activate_spare(struct active_array *a,
3373 struct metadata_update **updates)
3374 {
3375 int working = 0;
3376 struct mdinfo *d;
3377 struct ddf_super *ddf = a->container->sb;
3378 int global_ok = 0;
3379 struct mdinfo *rv = NULL;
3380 struct mdinfo *di;
3381 struct metadata_update *mu;
3382 struct dl *dl;
3383 int i;
3384 struct vd_config *vc;
3385 __u64 *lba;
3386
3387 for (d = a->info.devs ; d ; d = d->next) {
3388 if ((d->curr_state & DS_FAULTY) &&
3389 d->state_fd >= 0)
3390 /* wait for Removal to happen */
3391 return NULL;
3392 if (d->state_fd >= 0)
3393 working ++;
3394 }
3395
3396 dprintf("ddf_activate: working=%d (%d) level=%d\n", working, a->info.array.raid_disks,
3397 a->info.array.level);
3398 if (working == a->info.array.raid_disks)
3399 return NULL; /* array not degraded */
3400 switch (a->info.array.level) {
3401 case 1:
3402 if (working == 0)
3403 return NULL; /* failed */
3404 break;
3405 case 4:
3406 case 5:
3407 if (working < a->info.array.raid_disks - 1)
3408 return NULL; /* failed */
3409 break;
3410 case 6:
3411 if (working < a->info.array.raid_disks - 2)
3412 return NULL; /* failed */
3413 break;
3414 default: /* concat or stripe */
3415 return NULL; /* failed */
3416 }
3417
3418 /* For each slot, if it is not working, find a spare */
3419 dl = ddf->dlist;
3420 for (i = 0; i < a->info.array.raid_disks; i++) {
3421 for (d = a->info.devs ; d ; d = d->next)
3422 if (d->disk.raid_disk == i)
3423 break;
3424 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
3425 if (d && (d->state_fd >= 0))
3426 continue;
3427
3428 /* OK, this device needs recovery. Find a spare */
3429 again:
3430 for ( ; dl ; dl = dl->next) {
3431 unsigned long long esize;
3432 unsigned long long pos;
3433 struct mdinfo *d2;
3434 int is_global = 0;
3435 int is_dedicated = 0;
3436 struct extent *ex;
3437 int j;
3438 /* If in this array, skip */
3439 for (d2 = a->info.devs ; d2 ; d2 = d2->next)
3440 if (d2->disk.major == dl->major &&
3441 d2->disk.minor == dl->minor) {
3442 dprintf("%x:%x already in array\n", dl->major, dl->minor);
3443 break;
3444 }
3445 if (d2)
3446 continue;
3447 if (ddf->phys->entries[dl->pdnum].type &
3448 __cpu_to_be16(DDF_Spare)) {
3449 /* Check spare assign record */
3450 if (dl->spare) {
3451 if (dl->spare->type & DDF_spare_dedicated) {
3452 /* check spare_ents for guid */
3453 for (j = 0 ;
3454 j < __be16_to_cpu(dl->spare->populated);
3455 j++) {
3456 if (memcmp(dl->spare->spare_ents[j].guid,
3457 ddf->virt->entries[a->info.container_member].guid,
3458 DDF_GUID_LEN) == 0)
3459 is_dedicated = 1;
3460 }
3461 } else
3462 is_global = 1;
3463 }
3464 } else if (ddf->phys->entries[dl->pdnum].type &
3465 __cpu_to_be16(DDF_Global_Spare)) {
3466 is_global = 1;
3467 }
3468 if ( ! (is_dedicated ||
3469 (is_global && global_ok))) {
3470 dprintf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
3471 is_dedicated, is_global);
3472 continue;
3473 }
3474
3475 /* We are allowed to use this device - is there space?
3476 * We need a->info.component_size sectors */
3477 ex = get_extents(ddf, dl);
3478 if (!ex) {
3479 dprintf("cannot get extents\n");
3480 continue;
3481 }
3482 j = 0; pos = 0;
3483 esize = 0;
3484
3485 do {
3486 esize = ex[j].start - pos;
3487 if (esize >= a->info.component_size)
3488 break;
3489 pos = ex[i].start + ex[i].size;
3490 i++;
3491 } while (ex[i-1].size);
3492
3493 free(ex);
3494 if (esize < a->info.component_size) {
3495 dprintf("%x:%x has no room: %llu %llu\n", dl->major, dl->minor,
3496 esize, a->info.component_size);
3497 /* No room */
3498 continue;
3499 }
3500
3501 /* Cool, we have a device with some space at pos */
3502 di = malloc(sizeof(*di));
3503 if (!di)
3504 continue;
3505 memset(di, 0, sizeof(*di));
3506 di->disk.number = i;
3507 di->disk.raid_disk = i;
3508 di->disk.major = dl->major;
3509 di->disk.minor = dl->minor;
3510 di->disk.state = 0;
3511 di->data_offset = pos;
3512 di->component_size = a->info.component_size;
3513 di->container_member = dl->pdnum;
3514 di->next = rv;
3515 rv = di;
3516 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3517 i, pos);
3518
3519 break;
3520 }
3521 if (!dl && ! global_ok) {
3522 /* not enough dedicated spares, try global */
3523 global_ok = 1;
3524 dl = ddf->dlist;
3525 goto again;
3526 }
3527 }
3528
3529 if (!rv)
3530 /* No spares found */
3531 return rv;
3532 /* Now 'rv' has a list of devices to return.
3533 * Create a metadata_update record to update the
3534 * phys_refnum and lba_offset values
3535 */
3536 mu = malloc(sizeof(*mu));
3537 if (mu && posix_memalign(&mu->space, 512, sizeof(struct vcl)) != 0) {
3538 free(mu);
3539 mu = NULL;
3540 }
3541 if (!mu) {