2
* mdadm - manage Linux "md" devices aka RAID arrays.
4
* Copyright (C) 2006-2009 Neil Brown <neilb@suse.de>
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.
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.
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
22
* Email: <neil@brown.name>
24
* Specifications for DDF takes from Common RAID DDF Specification Revision 1.2
25
* (July 28 2006). Reused by permission of SNIA.
28
#define HAVE_STDINT_H 1
34
/* a non-official T10 name for creation GUIDs */
35
static char T10[] = "Linux-MD";
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.
41
#define DECADE (3600*24*(365*10+2))
44
const unsigned char *buf,
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.
53
* - all multibyte numeric fields are bigendian.
54
* - all strings are space padded.
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
66
#define DDF_CONCAT 0x1f
67
#define DDF_RAID5E 0x15
68
#define DDF_RAID5EE 0x25
69
#define DDF_RAID6 0x06
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 */
85
#define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
86
#define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
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 */
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)
106
#define DDF_GUID_LEN 24
107
#define DDF_REVISION_0 "01.00.00"
108
#define DDF_REVISION_2 "01.02.00"
111
__u32 magic; /* DDF_HEADER_MAGIC */
113
char guid[DDF_GUID_LEN];
114
char revision[8]; /* 01.02.00 */
115
__u32 seq; /* starts at '1' */
120
__u8 pad0; /* 0xff */
121
__u8 pad1[12]; /* 12 * 0xff */
122
/* 64 bytes so far */
123
__u8 header_ext[32]; /* reserved: fill with 0xff */
127
__u8 pad2[3]; /* 0xff */
128
__u32 workspace_len; /* sectors for vendor space -
129
* at least 32768(sectors) */
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
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;
156
/* 256 bytes so far */
157
__u8 pad4[256]; /* 0xff */
161
#define DDF_HEADER_ANCHOR 0x00
162
#define DDF_HEADER_PRIMARY 0x01
163
#define DDF_HEADER_SECONDARY 0x02
165
/* The content of the 'controller section' - global scope */
166
struct ddf_controller_data {
167
__u32 magic; /* DDF_CONTROLLER_MAGIC */
169
char guid[DDF_GUID_LEN];
170
struct controller_type {
177
__u8 pad[8]; /* 0xff */
178
__u8 vendor_data[448];
181
/* The content of phys_section - global scope */
183
__u32 magic; /* DDF_PHYS_RECORDS_MAGIC */
188
struct phys_disk_entry {
189
char guid[DDF_GUID_LEN];
193
__u64 config_size; /* DDF structures must be after here */
194
char path[18]; /* another horrible structure really */
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 */
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
213
/* phys_disk_entry.state is a bigendian bitmap */
215
#define DDF_Failed 2 /* overrides 1,4,8 */
216
#define DDF_Rebuilding 4
217
#define DDF_Transition 8
219
#define DDF_ReadErrors 32
220
#define DDF_Missing 64
222
/* The content of the virt_section global scope */
223
struct virtual_disk {
224
__u32 magic; /* DDF_VIRT_RECORDS_MAGIC */
226
__u16 populated_vdes;
229
struct virtual_entry {
230
char guid[DDF_GUID_LEN];
232
__u16 pad0; /* 0xffff */
242
/* virtual_entry.type is a bitmap - bigendian */
244
#define DDF_Enforce_Groups 2
245
#define DDF_Unicode 4
246
#define DDF_Owner_Valid 8
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
257
#define DDF_state_morphing 0x8
258
#define DDF_state_inconsistent 0x10
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
267
#define DDF_access_mask 0xc0
268
#define DDF_access_rw 0x00
269
#define DDF_access_ro 0x80
270
#define DDF_access_blocked 0xc0
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
278
__u32 magic; /* DDF_VD_CONF_MAGIC */
280
char guid[DDF_GUID_LEN];
284
__u16 prim_elmnt_count;
285
__u8 chunk_shift; /* 0 == 512, 1==1024 etc */
288
__u8 sec_elmnt_count;
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 */
302
__u8 v0[32]; /* reserved- 0xff */
303
__u8 v1[32]; /* reserved- 0xff */
304
__u8 v2[16]; /* reserved- 0xff */
305
__u8 v3[16]; /* reserved- 0xff */
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 */
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 */
321
struct spare_assign {
322
__u32 magic; /* DDF_SPARE_ASSIGN_MAGIC */
327
__u16 populated; /* SAEs used */
328
__u16 max; /* max SAEs */
330
struct spare_assign_entry {
331
char guid[DDF_GUID_LEN];
332
__u16 secondary_element;
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 */
342
/* The data_section contents - local scope */
344
__u32 magic; /* DDF_PHYS_DATA_MAGIC */
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 */
354
/* bbm_section content */
355
struct bad_block_log {
362
struct mapped_block {
363
__u64 defective_start;
364
__u32 replacement_start;
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:
376
* - Physical disk records
377
* - Virtual disk records
379
* - Configuration records
380
* - Physical Disk data section
381
* ( and Bad block and vendor which I don't care about yet).
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.
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;
399
unsigned int max_part, mppe, conf_rec_len;
407
__u64 *lba_offset; /* location in 'conf' of
409
unsigned int vcnum; /* index into ->virt */
410
__u64 *block_sizes; /* NULL if all the same */
413
struct vd_config conf;
414
} *conflist, *currentconf;
423
unsigned long long size; /* sectors */
424
int pdnum; /* index in ->phys */
425
struct spare_assign *spare;
426
void *mdupdate; /* hold metadata update */
428
/* These fields used by auto-layout */
429
int raiddisk; /* slot to fill in autolayout */
433
struct disk_data disk;
434
struct vcl *vlist[0]; /* max_part in size */
439
#define offsetof(t,f) ((size_t)&(((t*)0)->f))
443
static unsigned int calc_crc(void *buf, int len)
445
/* crcs are always at the same place as in the ddf_header */
446
struct ddf_header *ddf = buf;
447
__u32 oldcrc = ddf->crc;
449
ddf->crc = 0xffffffff;
451
newcrc = crc32(0, buf, len);
453
/* The crc is store (like everything) bigendian, so convert
454
* here for simplicity
456
return __cpu_to_be32(newcrc);
459
static int load_ddf_header(int fd, unsigned long long lba,
460
unsigned long long size,
462
struct ddf_header *hdr, struct ddf_header *anchor)
464
/* read a ddf header (primary or secondary) from fd/lba
465
* and check that it is consistent with anchor
467
* magic, crc, guid, rev, and LBA's header_type, and
468
* everything after header_type must be the same
473
if (lseek64(fd, lba<<9, 0) < 0)
476
if (read(fd, hdr, 512) != 512)
479
if (hdr->magic != DDF_HEADER_MAGIC)
481
if (calc_crc(hdr, 512) != hdr->crc)
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 ||
488
memcmp(anchor->pad2, hdr->pad2, 512 -
489
offsetof(struct ddf_header, pad2)) != 0)
492
/* Looks good enough to me... */
496
static void *load_section(int fd, struct ddf_super *super, void *buf,
497
__u32 offset_be, __u32 len_be, int check)
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);
504
if (len != 2 && len != 8 && len != 32
505
&& len != 128 && len != 512)
511
/* All pre-allocated sections are a single block */
514
} else if (posix_memalign(&buf, 512, len<<9) != 0)
520
if (super->active->type == 1)
521
offset += __be64_to_cpu(super->active->primary_lba);
523
offset += __be64_to_cpu(super->active->secondary_lba);
525
if ((unsigned long long)lseek64(fd, offset<<9, 0) != (offset<<9)) {
530
if ((unsigned long long)read(fd, buf, len<<9) != (len<<9)) {
538
static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
540
unsigned long long dsize;
542
get_dev_size(fd, NULL, &dsize);
544
if (lseek64(fd, dsize-512, 0) < 0) {
547
Name": Cannot seek to anchor block on %s: %s\n",
548
devname, strerror(errno));
551
if (read(fd, &super->anchor, 512) != 512) {
554
Name ": Cannot read anchor block on %s: %s\n",
555
devname, strerror(errno));
558
if (super->anchor.magic != DDF_HEADER_MAGIC) {
560
fprintf(stderr, Name ": no DDF anchor found on %s\n",
564
if (calc_crc(&super->anchor, 512) != super->anchor.crc) {
566
fprintf(stderr, Name ": bad CRC on anchor on %s\n",
570
if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
571
memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
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);
578
if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba),
580
&super->primary, &super->anchor) == 0) {
583
Name ": Failed to load primary DDF header "
587
super->active = &super->primary;
588
if (load_ddf_header(fd, __be64_to_cpu(super->anchor.secondary_lba),
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)
598
super->active = &super->secondary;
603
static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
606
ok = load_section(fd, super, &super->controller,
607
super->active->controller_section_offset,
608
super->active->controller_section_length,
610
super->phys = load_section(fd, super, NULL,
611
super->active->phys_section_offset,
612
super->active->phys_section_length,
614
super->pdsize = __be32_to_cpu(super->active->phys_section_length) * 512;
616
super->virt = load_section(fd, super, NULL,
617
super->active->virt_section_offset,
618
super->active->virt_section_length,
620
super->vdsize = __be32_to_cpu(super->active->virt_section_length) * 512;
630
super->conflist = NULL;
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);
639
static int load_ddf_local(int fd, struct ddf_super *super,
640
char *devname, int keep)
646
unsigned int confsec;
648
unsigned int max_virt_disks = __be16_to_cpu(super->active->max_vd_entries);
649
unsigned long long dsize;
651
/* First the local disk info */
652
if (posix_memalign((void**)&dl, 512,
654
(super->max_part) * sizeof(dl->vlist[0])) != 0) {
655
fprintf(stderr, Name ": %s could not allocate disk info buffer\n",
660
load_section(fd, super, &dl->disk,
661
super->active->data_section_offset,
662
super->active->data_section_length,
664
dl->devname = devname ? strdup(devname) : NULL;
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;
673
if (get_dev_size(fd, devname, &dsize))
674
dl->size = dsize >> 9;
676
for (i = 0 ; i < super->max_part ; i++)
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)
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
691
conf = load_section(fd, super, NULL,
692
super->active->config_section_offset,
693
super->active->config_section_length,
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);
704
if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) {
707
if (posix_memalign((void**)&dl->spare, 512,
708
super->conf_rec_len*512) != 0) {
710
": %s could not allocate spare info buf\n",
715
memcpy(dl->spare, vd, super->conf_rec_len*512);
718
if (vd->magic != DDF_VD_CONF_MAGIC)
720
for (vcl = super->conflist; vcl; vcl = vcl->next) {
721
if (memcmp(vcl->conf.guid,
722
vd->guid, DDF_GUID_LEN) == 0)
727
dl->vlist[vnum++] = vcl;
728
if (__be32_to_cpu(vd->seqnum) <=
729
__be32_to_cpu(vcl->conf.seqnum))
732
if (posix_memalign((void**)&vcl, 512,
733
(super->conf_rec_len*512 +
734
offsetof(struct vcl, conf))) != 0) {
736
": %s could not allocate vcl buf\n",
740
vcl->next = super->conflist;
741
vcl->block_sizes = NULL; /* FIXME not for CONCAT */
742
super->conflist = vcl;
743
dl->vlist[vnum++] = vcl;
745
memcpy(&vcl->conf, vd, super->conf_rec_len*512);
746
vcl->lba_offset = (__u64*)
747
&vcl->conf.phys_refnum[super->mppe];
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)
753
if (i < max_virt_disks)
762
static int load_super_ddf_all(struct supertype *st, int fd,
763
void **sbp, char *devname, int keep_fd);
766
static void free_super_ddf(struct supertype *st);
768
static int load_super_ddf(struct supertype *st, int fd,
771
unsigned long long dsize;
772
struct ddf_super *super;
776
/* if 'fd' is a container, load metadata from all the devices */
777
if (load_super_ddf_all(st, fd, &st->sb, devname, 1) == 0)
781
return 1; /* FIXME Is this correct */
783
if (get_dev_size(fd, devname, &dsize) == 0)
786
if (test_partition(fd))
787
/* DDF is not allowed on partitions */
790
/* 32M is a lower bound */
791
if (dsize <= 32*1024*1024) {
794
Name ": %s is too small for ddf: "
795
"size is %llu sectors.\n",
802
Name ": %s is an odd size for ddf: "
803
"size is %llu bytes.\n",
810
if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) {
811
fprintf(stderr, Name ": malloc of %zu failed.\n",
815
memset(super, 0, sizeof(*super));
817
rv = load_ddf_headers(fd, super, devname);
823
/* Have valid headers and have chosen the best. Let's read in the rest*/
825
rv = load_ddf_global(fd, super, devname);
830
Name ": Failed to load all information "
831
"sections on %s\n", devname);
836
rv = load_ddf_local(fd, super, devname, 0);
841
Name ": Failed to load all information "
842
"sections on %s\n", devname);
847
if (st->subarray[0]) {
852
val = strtoul(st->subarray, &ep, 10);
858
for (v = super->conflist; v; v = v->next)
860
super->currentconf = v;
861
if (!super->currentconf) {
867
/* Should possibly check the sections .... */
870
if (st->ss == NULL) {
872
st->minor_version = 0;
875
st->loaded_container = 0;
880
static void free_super_ddf(struct supertype *st)
882
struct ddf_super *ddf = st->sb;
887
while (ddf->conflist) {
888
struct vcl *v = ddf->conflist;
889
ddf->conflist = v->next;
891
free(v->block_sizes);
895
struct dl *d = ddf->dlist;
896
ddf->dlist = d->next;
907
static struct supertype *match_metadata_desc_ddf(char *arg)
909
/* 'ddf' only support containers */
910
struct supertype *st;
911
if (strcmp(arg, "ddf") != 0 &&
912
strcmp(arg, "default") != 0
916
st = malloc(sizeof(*st));
917
memset(st, 0, sizeof(*st));
920
st->minor_version = 0;
928
static mapping_t ddf_state[] = {
934
{ "Partially Optimal", 5},
940
static mapping_t ddf_init_state[] = {
941
{ "Not Initialised", 0},
942
{ "QuickInit in Progress", 1},
943
{ "Fully Initialised", 2},
947
static mapping_t ddf_access[] = {
951
{ "Blocked (no access)", 3},
955
static mapping_t ddf_level[] = {
956
{ "RAID0", DDF_RAID0},
957
{ "RAID1", DDF_RAID1},
958
{ "RAID3", DDF_RAID3},
959
{ "RAID4", DDF_RAID4},
960
{ "RAID5", DDF_RAID5},
961
{ "RAID1E",DDF_RAID1E},
963
{ "CONCAT",DDF_CONCAT},
964
{ "RAID5E",DDF_RAID5E},
965
{ "RAID5EE",DDF_RAID5EE},
966
{ "RAID6", DDF_RAID6},
969
static mapping_t ddf_sec_level[] = {
970
{ "Striped", DDF_2STRIPED},
971
{ "Mirrored", DDF_2MIRRORED},
972
{ "Concat", DDF_2CONCAT},
973
{ "Spanned", DDF_2SPANNED},
981
static struct num_mapping ddf_level_num[] = {
984
{ DDF_RAID3, LEVEL_UNSUPPORTED },
987
{ DDF_RAID1E, LEVEL_UNSUPPORTED },
988
{ DDF_JBOD, LEVEL_UNSUPPORTED },
989
{ DDF_CONCAT, LEVEL_LINEAR },
990
{ DDF_RAID5E, LEVEL_UNSUPPORTED },
991
{ DDF_RAID5EE, LEVEL_UNSUPPORTED },
996
static int map_num1(struct num_mapping *map, int num)
999
for (i=0 ; map[i].num1 != MAXINT; i++)
1000
if (map[i].num1 == num)
1005
static int all_ff(char *guid)
1008
for (i = 0; i < DDF_GUID_LEN; i++)
1009
if (guid[i] != (char)0xff)
1015
static void print_guid(char *guid, int tstamp)
1017
/* A GUIDs are part (or all) ASCII and part binary.
1018
* They tend to be space padded.
1019
* We print the GUID in HEX, then in parentheses add
1020
* any initial ASCII sequence, and a possible
1021
* time stamp from bytes 16-19
1023
int l = DDF_GUID_LEN;
1026
for (i=0 ; i<DDF_GUID_LEN ; i++) {
1027
if ((i&3)==0 && i != 0) printf(":");
1028
printf("%02X", guid[i]&255);
1032
while (l && guid[l-1] == ' ')
1034
for (i=0 ; i<l ; i++) {
1035
if (guid[i] >= 0x20 && guid[i] < 0x7f)
1036
fputc(guid[i], stdout);
1041
time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
1044
tm = localtime(&then);
1045
strftime(tbuf, 100, " %D %T",tm);
1046
fputs(tbuf, stdout);
1051
static void examine_vd(int n, struct ddf_super *sb, char *guid)
1053
int crl = sb->conf_rec_len;
1056
for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
1058
struct vd_config *vc = &vcl->conf;
1060
if (calc_crc(vc, crl*512) != vc->crc)
1062
if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
1065
/* Ok, we know about this VD, let's give more details */
1066
printf(" Raid Devices[%d] : %d (", n,
1067
__be16_to_cpu(vc->prim_elmnt_count));
1068
for (i = 0; i < __be16_to_cpu(vc->prim_elmnt_count); i++) {
1070
int cnt = __be16_to_cpu(sb->phys->used_pdes);
1071
for (j=0; j<cnt; j++)
1072
if (vc->phys_refnum[i] == sb->phys->entries[j].refnum)
1081
if (vc->chunk_shift != 255)
1082
printf(" Chunk Size[%d] : %d sectors\n", n,
1083
1 << vc->chunk_shift);
1084
printf(" Raid Level[%d] : %s\n", n,
1085
map_num(ddf_level, vc->prl)?:"-unknown-");
1086
if (vc->sec_elmnt_count != 1) {
1087
printf(" Secondary Position[%d] : %d of %d\n", n,
1088
vc->sec_elmnt_seq, vc->sec_elmnt_count);
1089
printf(" Secondary Level[%d] : %s\n", n,
1090
map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1092
printf(" Device Size[%d] : %llu\n", n,
1093
(unsigned long long)__be64_to_cpu(vc->blocks)/2);
1094
printf(" Array Size[%d] : %llu\n", n,
1095
(unsigned long long)__be64_to_cpu(vc->array_blocks)/2);
1099
static void examine_vds(struct ddf_super *sb)
1101
int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1103
printf(" Virtual Disks : %d\n", cnt);
1105
for (i=0; i<cnt; i++) {
1106
struct virtual_entry *ve = &sb->virt->entries[i];
1108
printf(" VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1110
printf(" unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1111
printf(" state[%d] : %s, %s%s\n", i,
1112
map_num(ddf_state, ve->state & 7),
1113
(ve->state & 8) ? "Morphing, ": "",
1114
(ve->state & 16)? "Not Consistent" : "Consistent");
1115
printf(" init state[%d] : %s\n", i,
1116
map_num(ddf_init_state, ve->init_state&3));
1117
printf(" access[%d] : %s\n", i,
1118
map_num(ddf_access, (ve->init_state>>6) & 3));
1119
printf(" Name[%d] : %.16s\n", i, ve->name);
1120
examine_vd(i, sb, ve->guid);
1122
if (cnt) printf("\n");
1125
static void examine_pds(struct ddf_super *sb)
1127
int cnt = __be16_to_cpu(sb->phys->used_pdes);
1130
printf(" Physical Disks : %d\n", cnt);
1131
printf(" Number RefNo Size Device Type/State\n");
1133
for (i=0 ; i<cnt ; i++) {
1134
struct phys_disk_entry *pd = &sb->phys->entries[i];
1135
int type = __be16_to_cpu(pd->type);
1136
int state = __be16_to_cpu(pd->state);
1138
//printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1140
printf(" %3d %08x ", i,
1141
__be32_to_cpu(pd->refnum));
1143
(unsigned long long)__be64_to_cpu(pd->config_size)>>1);
1144
for (dl = sb->dlist; dl ; dl = dl->next) {
1145
if (dl->disk.refnum == pd->refnum) {
1146
char *dv = map_dev(dl->major, dl->minor, 0);
1148
printf("%-15s", dv);
1155
printf(" %s%s%s%s%s",
1156
(type&2) ? "active":"",
1157
(type&4) ? "Global-Spare":"",
1158
(type&8) ? "spare" : "",
1159
(type&16)? ", foreign" : "",
1160
(type&32)? "pass-through" : "");
1161
printf("/%s%s%s%s%s%s%s",
1162
(state&1)? "Online": "Offline",
1163
(state&2)? ", Failed": "",
1164
(state&4)? ", Rebuilding": "",
1165
(state&8)? ", in-transition": "",
1166
(state&16)? ", SMART-errors": "",
1167
(state&32)? ", Unrecovered-Read-Errors": "",
1168
(state&64)? ", Missing" : "");
1173
static void examine_super_ddf(struct supertype *st, char *homehost)
1175
struct ddf_super *sb = st->sb;
1177
printf(" Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1178
printf(" Version : %.8s\n", sb->anchor.revision);
1179
printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1181
printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1183
printf(" Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1184
printf(" Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1190
static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info);
1192
static void uuid_from_super_ddf(struct supertype *st, int uuid[4]);
1194
static void brief_examine_super_ddf(struct supertype *st, int verbose)
1196
/* We just write a generic DDF ARRAY entry
1200
getinfo_super_ddf(st, &info);
1201
fname_from_uuid(st, &info, nbuf, ':');
1203
printf("ARRAY metadata=ddf UUID=%s\n", nbuf + 5);
1206
static void brief_examine_subarrays_ddf(struct supertype *st, int verbose)
1208
/* We just write a generic DDF ARRAY entry
1210
struct ddf_super *ddf = st->sb;
1214
getinfo_super_ddf(st, &info);
1215
fname_from_uuid(st, &info, nbuf, ':');
1217
for (i = 0; i < __be16_to_cpu(ddf->virt->max_vdes); i++) {
1218
struct virtual_entry *ve = &ddf->virt->entries[i];
1221
if (all_ff(ve->guid))
1223
memcpy(vcl.conf.guid, ve->guid, DDF_GUID_LEN);
1224
ddf->currentconf =&vcl;
1225
uuid_from_super_ddf(st, info.uuid);
1226
fname_from_uuid(st, &info, nbuf1, ':');
1227
printf("ARRAY container=%s member=%d UUID=%s\n",
1228
nbuf+5, i, nbuf1+5);
1232
static void export_examine_super_ddf(struct supertype *st)
1236
getinfo_super_ddf(st, &info);
1237
fname_from_uuid(st, &info, nbuf, ':');
1238
printf("MD_METADATA=ddf\n");
1239
printf("MD_LEVEL=container\n");
1240
printf("MD_UUID=%s\n", nbuf+5);
1244
static void detail_super_ddf(struct supertype *st, char *homehost)
1247
* Could print DDF GUID
1248
* Need to find which array
1249
* If whole, briefly list all arrays
1254
static void brief_detail_super_ddf(struct supertype *st)
1256
/* FIXME I really need to know which array we are detailing.
1257
* Can that be stored in ddf_super??
1259
// struct ddf_super *ddf = st->sb;
1262
getinfo_super_ddf(st, &info);
1263
fname_from_uuid(st, &info, nbuf,':');
1264
printf(" UUID=%s", nbuf + 5);
1268
static int match_home_ddf(struct supertype *st, char *homehost)
1270
/* It matches 'this' host if the controller is a
1271
* Linux-MD controller with vendor_data matching
1274
struct ddf_super *ddf = st->sb;
1279
len = strlen(homehost);
1281
return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1282
len < sizeof(ddf->controller.vendor_data) &&
1283
memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1284
ddf->controller.vendor_data[len] == 0);
1288
static struct vd_config *find_vdcr(struct ddf_super *ddf, unsigned int inst)
1292
for (v = ddf->conflist; v; v = v->next)
1293
if (inst == v->vcnum)
1299
static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1301
/* Find the entry in phys_disk which has the given refnum
1302
* and return it's index
1305
for (i = 0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1306
if (ddf->phys->entries[i].refnum == phys_refnum)
1311
static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1313
/* The uuid returned here is used for:
1314
* uuid to put into bitmap file (Create, Grow)
1315
* uuid for backup header when saving critical section (Grow)
1316
* comparing uuids when re-adding a device into an array
1317
* In these cases the uuid required is that of the data-array,
1318
* not the device-set.
1319
* uuid to recognise same set when adding a missing device back
1320
* to an array. This is a uuid for the device-set.
1322
* For each of these we can make do with a truncated
1323
* or hashed uuid rather than the original, as long as
1325
* In the case of SVD we assume the BVD is of interest,
1326
* though that might be the case if a bitmap were made for
1327
* a mirrored SVD - worry about that later.
1328
* So we need to find the VD configuration record for the
1329
* relevant BVD and extract the GUID and Secondary_Element_Seq.
1330
* The first 16 bytes of the sha1 of these is used.
1332
struct ddf_super *ddf = st->sb;
1333
struct vcl *vcl = ddf->currentconf;
1336
struct sha1_ctx ctx;
1339
guid = vcl->conf.guid;
1341
guid = ddf->anchor.guid;
1343
sha1_init_ctx(&ctx);
1344
sha1_process_bytes(guid, DDF_GUID_LEN, &ctx);
1345
sha1_finish_ctx(&ctx, buf);
1346
memcpy(uuid, buf, 4*4);
1349
static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info);
1351
static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info)
1353
struct ddf_super *ddf = st->sb;
1355
if (ddf->currentconf) {
1356
getinfo_super_ddf_bvd(st, info);
1360
info->array.raid_disks = __be16_to_cpu(ddf->phys->used_pdes);
1361
info->array.level = LEVEL_CONTAINER;
1362
info->array.layout = 0;
1363
info->array.md_minor = -1;
1364
info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)
1365
(ddf->anchor.guid+16));
1366
info->array.utime = 0;
1367
info->array.chunk_size = 0;
1368
info->container_enough = 1;
1371
info->disk.major = 0;
1372
info->disk.minor = 0;
1374
info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1375
info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1377
info->data_offset = __be64_to_cpu(ddf->phys->
1378
entries[info->disk.raid_disk].
1380
info->component_size = ddf->dlist->size - info->data_offset;
1382
info->disk.number = -1;
1383
info->disk.raid_disk = -1;
1384
// info->disk.raid_disk = find refnum in the table and use index;
1386
info->disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
1389
info->recovery_start = MaxSector;
1390
info->reshape_active = 0;
1393
info->array.major_version = -1;
1394
info->array.minor_version = -2;
1395
strcpy(info->text_version, "ddf");
1396
info->safe_mode_delay = 0;
1398
uuid_from_super_ddf(st, info->uuid);
1402
static int rlq_to_layout(int rlq, int prl, int raiddisks);
1404
static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info)
1406
struct ddf_super *ddf = st->sb;
1407
struct vcl *vc = ddf->currentconf;
1408
int cd = ddf->currentdev;
1412
/* FIXME this returns BVD info - what if we want SVD ?? */
1414
info->array.raid_disks = __be16_to_cpu(vc->conf.prim_elmnt_count);
1415
info->array.level = map_num1(ddf_level_num, vc->conf.prl);
1416
info->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
1417
info->array.raid_disks);
1418
info->array.md_minor = -1;
1419
info->array.ctime = DECADE +
1420
__be32_to_cpu(*(__u32*)(vc->conf.guid+16));
1421
info->array.utime = DECADE + __be32_to_cpu(vc->conf.timestamp);
1422
info->array.chunk_size = 512 << vc->conf.chunk_shift;
1423
info->custom_array_size = 0;
1425
if (cd >= 0 && (unsigned)cd < ddf->mppe) {
1426
info->data_offset = __be64_to_cpu(vc->lba_offset[cd]);
1427
if (vc->block_sizes)
1428
info->component_size = vc->block_sizes[cd];
1430
info->component_size = __be64_to_cpu(vc->conf.blocks);
1433
for (dl = ddf->dlist; dl ; dl = dl->next)
1434
if (dl->raiddisk == info->disk.raid_disk)
1436
info->disk.major = 0;
1437
info->disk.minor = 0;
1439
info->disk.major = dl->major;
1440
info->disk.minor = dl->minor;
1442
// info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1443
// info->disk.raid_disk = find refnum in the table and use index;
1444
// info->disk.state = ???;
1446
info->container_member = ddf->currentconf->vcnum;
1448
info->recovery_start = MaxSector;
1449
info->resync_start = 0;
1450
info->reshape_active = 0;
1451
if (!(ddf->virt->entries[info->container_member].state
1452
& DDF_state_inconsistent) &&
1453
(ddf->virt->entries[info->container_member].init_state
1454
& DDF_initstate_mask)
1456
info->resync_start = MaxSector;
1458
uuid_from_super_ddf(st, info->uuid);
1460
info->container_member = atoi(st->subarray);
1461
info->array.major_version = -1;
1462
info->array.minor_version = -2;
1463
sprintf(info->text_version, "/%s/%s",
1464
devnum2devname(st->container_dev),
1466
info->safe_mode_delay = 200;
1468
memcpy(info->name, ddf->virt->entries[info->container_member].name, 16);
1471
if (info->name[j] == ' ')
1476
static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1478
char *devname, int verbose,
1479
int uuid_set, char *homehost)
1481
/* For 'assemble' and 'force' we need to return non-zero if any
1482
* change was made. For others, the return value is ignored.
1483
* Update options are:
1484
* force-one : This device looks a bit old but needs to be included,
1485
* update age info appropriately.
1486
* assemble: clear any 'faulty' flag to allow this device to
1488
* force-array: Array is degraded but being forced, mark it clean
1489
* if that will be needed to assemble it.
1491
* newdev: not used ????
1492
* grow: Array has gained a new device - this is currently for
1494
* resync: mark as dirty so a resync will happen.
1495
* uuid: Change the uuid of the array to match what is given
1496
* homehost: update the recorded homehost
1497
* name: update the name - preserving the homehost
1498
* _reshape_progress: record new reshape_progress position.
1500
* Following are not relevant for this version:
1501
* sparc2.2 : update from old dodgey metadata
1502
* super-minor: change the preferred_minor number
1503
* summaries: update redundant counters.
1506
// struct ddf_super *ddf = st->sb;
1507
// struct vd_config *vd = find_vdcr(ddf, info->container_member);
1508
// struct virtual_entry *ve = find_ve(ddf);
1510
/* we don't need to handle "force-*" or "assemble" as
1511
* there is no need to 'trick' the kernel. We the metadata is
1512
* first updated to activate the array, all the implied modifications
1516
if (strcmp(update, "grow") == 0) {
1519
if (strcmp(update, "resync") == 0) {
1520
// info->resync_checkpoint = 0;
1522
/* We ignore UUID updates as they make even less sense
1525
if (strcmp(update, "homehost") == 0) {
1526
/* homehost is stored in controller->vendor_data,
1527
* or it is when we are the vendor
1529
// if (info->vendor_is_local)
1530
// strcpy(ddf->controller.vendor_data, homehost);
1532
if (strcmp(update, "name") == 0) {
1533
/* name is stored in virtual_entry->name */
1534
// memset(ve->name, ' ', 16);
1535
// strncpy(ve->name, info->name, 16);
1537
if (strcmp(update, "_reshape_progress") == 0) {
1538
/* We don't support reshape yet */
1541
// update_all_csum(ddf);
1546
static void make_header_guid(char *guid)
1549
/* Create a DDF Header of Virtual Disk GUID */
1551
/* 24 bytes of fiction required.
1552
* first 8 are a 'vendor-id' - "Linux-MD"
1553
* next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1554
* Remaining 8 random number plus timestamp
1556
memcpy(guid, T10, sizeof(T10));
1557
stamp = __cpu_to_be32(0xdeadbeef);
1558
memcpy(guid+8, &stamp, 4);
1559
stamp = __cpu_to_be32(0);
1560
memcpy(guid+12, &stamp, 4);
1561
stamp = __cpu_to_be32(time(0) - DECADE);
1562
memcpy(guid+16, &stamp, 4);
1564
memcpy(guid+20, &stamp, 4);
1567
static int init_super_ddf_bvd(struct supertype *st,
1568
mdu_array_info_t *info,
1569
unsigned long long size,
1570
char *name, char *homehost,
1573
static int init_super_ddf(struct supertype *st,
1574
mdu_array_info_t *info,
1575
unsigned long long size, char *name, char *homehost,
1578
/* This is primarily called by Create when creating a new array.
1579
* We will then get add_to_super called for each component, and then
1580
* write_init_super called to write it out to each device.
1581
* For DDF, Create can create on fresh devices or on a pre-existing
1583
* To create on a pre-existing array a different method will be called.
1584
* This one is just for fresh drives.
1586
* We need to create the entire 'ddf' structure which includes:
1587
* DDF headers - these are easy.
1588
* Controller data - a Sector describing this controller .. not that
1589
* this is a controller exactly.
1590
* Physical Disk Record - one entry per device, so
1591
* leave plenty of space.
1592
* Virtual Disk Records - again, just leave plenty of space.
1593
* This just lists VDs, doesn't give details
1594
* Config records - describes the VDs that use this disk
1595
* DiskData - describes 'this' device.
1596
* BadBlockManagement - empty
1597
* Diag Space - empty
1598
* Vendor Logs - Could we put bitmaps here?
1601
struct ddf_super *ddf;
1604
int max_phys_disks, max_virt_disks;
1605
unsigned long long sector;
1609
struct phys_disk *pd;
1610
struct virtual_disk *vd;
1613
return init_super_ddf_bvd(st, info, size, name, homehost, uuid);
1615
if (posix_memalign((void**)&ddf, 512, sizeof(*ddf)) != 0) {
1616
fprintf(stderr, Name ": %s could not allocate superblock\n", __func__);
1619
memset(ddf, 0, sizeof(*ddf));
1620
ddf->dlist = NULL; /* no physical disks yet */
1621
ddf->conflist = NULL; /* No virtual disks yet */
1625
/* zeroing superblock */
1629
/* At least 32MB *must* be reserved for the ddf. So let's just
1630
* start 32MB from the end, and put the primary header there.
1631
* Don't do secondary for now.
1632
* We don't know exactly where that will be yet as it could be
1633
* different on each device. To just set up the lengths.
1637
ddf->anchor.magic = DDF_HEADER_MAGIC;
1638
make_header_guid(ddf->anchor.guid);
1640
memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
1641
ddf->anchor.seq = __cpu_to_be32(1);
1642
ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1643
ddf->anchor.openflag = 0xFF;
1644
ddf->anchor.foreignflag = 0;
1645
ddf->anchor.enforcegroups = 0; /* Is this best?? */
1646
ddf->anchor.pad0 = 0xff;
1647
memset(ddf->anchor.pad1, 0xff, 12);
1648
memset(ddf->anchor.header_ext, 0xff, 32);
1649
ddf->anchor.primary_lba = ~(__u64)0;
1650
ddf->anchor.secondary_lba = ~(__u64)0;
1651
ddf->anchor.type = DDF_HEADER_ANCHOR;
1652
memset(ddf->anchor.pad2, 0xff, 3);
1653
ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1654
ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1655
of 32M reserved.. */
1656
max_phys_disks = 1023; /* Should be enough */
1657
ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1658
max_virt_disks = 255;
1659
ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1660
ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1663
ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
1664
ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1665
ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe);
1666
memset(ddf->anchor.pad3, 0xff, 54);
1667
/* controller sections is one sector long immediately
1668
* after the ddf header */
1670
ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1671
ddf->anchor.controller_section_length = __cpu_to_be32(1);
1674
/* phys is 8 sectors after that */
1675
pdsize = ROUND_UP(sizeof(struct phys_disk) +
1676
sizeof(struct phys_disk_entry)*max_phys_disks,
1678
switch(pdsize/512) {
1679
case 2: case 8: case 32: case 128: case 512: break;
1682
ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1683
ddf->anchor.phys_section_length =
1684
__cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1685
sector += pdsize/512;
1687
/* virt is another 32 sectors */
1688
vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1689
sizeof(struct virtual_entry) * max_virt_disks,
1691
switch(vdsize/512) {
1692
case 2: case 8: case 32: case 128: case 512: break;
1695
ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1696
ddf->anchor.virt_section_length =
1697
__cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1698
sector += vdsize/512;
1700
clen = ddf->conf_rec_len * (ddf->max_part+1);
1701
ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1702
ddf->anchor.config_section_length = __cpu_to_be32(clen);
1705
ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1706
ddf->anchor.data_section_length = __cpu_to_be32(1);
1709
ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1710
ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1711
ddf->anchor.diag_space_length = __cpu_to_be32(0);
1712
ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1713
ddf->anchor.vendor_length = __cpu_to_be32(0);
1714
ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1716
memset(ddf->anchor.pad4, 0xff, 256);
1718
memcpy(&ddf->primary, &ddf->anchor, 512);
1719
memcpy(&ddf->secondary, &ddf->anchor, 512);
1721
ddf->primary.openflag = 1; /* I guess.. */
1722
ddf->primary.type = DDF_HEADER_PRIMARY;
1724
ddf->secondary.openflag = 1; /* I guess.. */
1725
ddf->secondary.type = DDF_HEADER_SECONDARY;
1727
ddf->active = &ddf->primary;
1729
ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1731
/* 24 more bytes of fiction required.
1732
* first 8 are a 'vendor-id' - "Linux-MD"
1733
* Remaining 16 are serial number.... maybe a hostname would do?
1735
memcpy(ddf->controller.guid, T10, sizeof(T10));
1736
gethostname(hostname, sizeof(hostname));
1737
hostname[sizeof(hostname) - 1] = 0;
1738
hostlen = strlen(hostname);
1739
memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1740
for (i = strlen(T10) ; i+hostlen < 24; i++)
1741
ddf->controller.guid[i] = ' ';
1743
ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1744
ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1745
ddf->controller.type.sub_vendor_id = 0;
1746
ddf->controller.type.sub_device_id = 0;
1747
memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1748
memset(ddf->controller.pad, 0xff, 8);
1749
memset(ddf->controller.vendor_data, 0xff, 448);
1750
if (homehost && strlen(homehost) < 440)
1751
strcpy((char*)ddf->controller.vendor_data, homehost);
1753
if (posix_memalign((void**)&pd, 512, pdsize) != 0) {
1754
fprintf(stderr, Name ": %s could not allocate pd\n", __func__);
1758
ddf->pdsize = pdsize;
1760
memset(pd, 0xff, pdsize);
1761
memset(pd, 0, sizeof(*pd));
1762
pd->magic = DDF_PHYS_RECORDS_MAGIC;
1763
pd->used_pdes = __cpu_to_be16(0);
1764
pd->max_pdes = __cpu_to_be16(max_phys_disks);
1765
memset(pd->pad, 0xff, 52);
1767
if (posix_memalign((void**)&vd, 512, vdsize) != 0) {
1768
fprintf(stderr, Name ": %s could not allocate vd\n", __func__);
1772
ddf->vdsize = vdsize;
1773
memset(vd, 0, vdsize);
1774
vd->magic = DDF_VIRT_RECORDS_MAGIC;
1775
vd->populated_vdes = __cpu_to_be16(0);
1776
vd->max_vdes = __cpu_to_be16(max_virt_disks);
1777
memset(vd->pad, 0xff, 52);
1779
for (i=0; i<max_virt_disks; i++)
1780
memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1783
ddf->updates_pending = 1;
1787
static int chunk_to_shift(int chunksize)
1789
return ffs(chunksize/512)-1;
1792
static int level_to_prl(int level)
1795
case LEVEL_LINEAR: return DDF_CONCAT;
1796
case 0: return DDF_RAID0;
1797
case 1: return DDF_RAID1;
1798
case 4: return DDF_RAID4;
1799
case 5: return DDF_RAID5;
1800
case 6: return DDF_RAID6;
1804
static int layout_to_rlq(int level, int layout, int raiddisks)
1808
return DDF_RAID0_SIMPLE;
1811
case 2: return DDF_RAID1_SIMPLE;
1812
case 3: return DDF_RAID1_MULTI;
1817
case 0: return DDF_RAID4_N;
1822
case ALGORITHM_LEFT_ASYMMETRIC:
1823
return DDF_RAID5_N_RESTART;
1824
case ALGORITHM_RIGHT_ASYMMETRIC:
1825
return DDF_RAID5_0_RESTART;
1826
case ALGORITHM_LEFT_SYMMETRIC:
1827
return DDF_RAID5_N_CONTINUE;
1828
case ALGORITHM_RIGHT_SYMMETRIC:
1829
return -1; /* not mentioned in standard */
1833
case ALGORITHM_ROTATING_N_RESTART:
1834
return DDF_RAID5_N_RESTART;
1835
case ALGORITHM_ROTATING_ZERO_RESTART:
1836
return DDF_RAID6_0_RESTART;
1837
case ALGORITHM_ROTATING_N_CONTINUE:
1838
return DDF_RAID5_N_CONTINUE;
1844
static int rlq_to_layout(int rlq, int prl, int raiddisks)
1848
return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1850
return 0; /* hopefully rlq == SIMPLE or MULTI depending
1858
return -1; /* FIXME this isn't checked */
1862
case DDF_RAID5_N_RESTART:
1863
return ALGORITHM_LEFT_ASYMMETRIC;
1864
case DDF_RAID5_0_RESTART:
1865
return ALGORITHM_RIGHT_ASYMMETRIC;
1866
case DDF_RAID5_N_CONTINUE:
1867
return ALGORITHM_LEFT_SYMMETRIC;
1873
case DDF_RAID5_N_RESTART:
1874
return ALGORITHM_ROTATING_N_RESTART;
1875
case DDF_RAID6_0_RESTART:
1876
return ALGORITHM_ROTATING_ZERO_RESTART;
1877
case DDF_RAID5_N_CONTINUE:
1878
return ALGORITHM_ROTATING_N_CONTINUE;
1888
unsigned long long start, size;
1890
static int cmp_extent(const void *av, const void *bv)
1892
const struct extent *a = av;
1893
const struct extent *b = bv;
1894
if (a->start < b->start)
1896
if (a->start > b->start)
1901
static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
1903
/* find a list of used extents on the give physical device
1904
* (dnum) of the given ddf.
1905
* Return a malloced array of 'struct extent'
1907
FIXME ignore DDF_Legacy devices?
1914
rv = malloc(sizeof(struct extent) * (ddf->max_part + 2));
1918
for (i = 0; i < ddf->max_part; i++) {
1919
struct vcl *v = dl->vlist[i];
1922
for (j = 0; j < v->conf.prim_elmnt_count; j++)
1923
if (v->conf.phys_refnum[j] == dl->disk.refnum) {
1924
/* This device plays role 'j' in 'v'. */
1925
rv[n].start = __be64_to_cpu(v->lba_offset[j]);
1926
rv[n].size = __be64_to_cpu(v->conf.blocks);
1931
qsort(rv, n, sizeof(*rv), cmp_extent);
1933
rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
1939
static int init_super_ddf_bvd(struct supertype *st,
1940
mdu_array_info_t *info,
1941
unsigned long long size,
1942
char *name, char *homehost,
1945
/* We are creating a BVD inside a pre-existing container.
1946
* so st->sb is already set.
1947
* We need to create a new vd_config and a new virtual_entry
1949
struct ddf_super *ddf = st->sb;
1951
struct virtual_entry *ve;
1953
struct vd_config *vc;
1955
if (__be16_to_cpu(ddf->virt->populated_vdes)
1956
>= __be16_to_cpu(ddf->virt->max_vdes)) {
1957
fprintf(stderr, Name": This ddf already has the "
1958
"maximum of %d virtual devices\n",
1959
__be16_to_cpu(ddf->virt->max_vdes));
1963
for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1964
if (all_ff(ddf->virt->entries[venum].guid))
1966
if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1967
fprintf(stderr, Name ": Cannot find spare slot for "
1968
"virtual disk - DDF is corrupt\n");
1971
ve = &ddf->virt->entries[venum];
1973
/* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1974
* timestamp, random number
1976
make_header_guid(ve->guid);
1977
ve->unit = __cpu_to_be16(info->md_minor);
1979
ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1981
ve->state = DDF_state_degraded; /* Will be modified as devices are added */
1982
if (info->state & 1) /* clean */
1983
ve->init_state = DDF_init_full;
1985
ve->init_state = DDF_init_not;
1987
memset(ve->pad1, 0xff, 14);
1988
memset(ve->name, ' ', 16);
1990
strncpy(ve->name, name, 16);
1991
ddf->virt->populated_vdes =
1992
__cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
1994
/* Now create a new vd_config */
1995
if (posix_memalign((void**)&vcl, 512,
1996
(offsetof(struct vcl, conf) + ddf->conf_rec_len * 512)) != 0) {
1997
fprintf(stderr, Name ": %s could not allocate vd_config\n", __func__);
2000
vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
2002
sprintf(st->subarray, "%d", venum);
2003
vcl->block_sizes = NULL; /* FIXME not for CONCAT */
2007
vc->magic = DDF_VD_CONF_MAGIC;
2008
memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
2009
vc->timestamp = __cpu_to_be32(time(0)-DECADE);
2010
vc->seqnum = __cpu_to_be32(1);
2011
memset(vc->pad0, 0xff, 24);
2012
vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
2013
vc->chunk_shift = chunk_to_shift(info->chunk_size);
2014
vc->prl = level_to_prl(info->level);
2015
vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
2016
vc->sec_elmnt_count = 1;
2017
vc->sec_elmnt_seq = 0;
2019
vc->blocks = __cpu_to_be64(info->size * 2);
2020
vc->array_blocks = __cpu_to_be64(
2021
calc_array_size(info->level, info->raid_disks, info->layout,
2022
info->chunk_size, info->size*2));
2023
memset(vc->pad1, 0xff, 8);
2024
vc->spare_refs[0] = 0xffffffff;
2025
vc->spare_refs[1] = 0xffffffff;
2026
vc->spare_refs[2] = 0xffffffff;
2027
vc->spare_refs[3] = 0xffffffff;
2028
vc->spare_refs[4] = 0xffffffff;
2029
vc->spare_refs[5] = 0xffffffff;
2030
vc->spare_refs[6] = 0xffffffff;
2031
vc->spare_refs[7] = 0xffffffff;
2032
memset(vc->cache_pol, 0, 8);
2034
memset(vc->pad2, 0xff, 3);
2035
memset(vc->pad3, 0xff, 52);
2036
memset(vc->pad4, 0xff, 192);
2037
memset(vc->v0, 0xff, 32);
2038
memset(vc->v1, 0xff, 32);
2039
memset(vc->v2, 0xff, 16);
2040
memset(vc->v3, 0xff, 16);
2041
memset(vc->vendor, 0xff, 32);
2043
memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
2044
memset(vc->phys_refnum+ddf->mppe, 0x00, 8*ddf->mppe);
2046
vcl->next = ddf->conflist;
2047
ddf->conflist = vcl;
2048
ddf->currentconf = vcl;
2049
ddf->updates_pending = 1;
2054
static void add_to_super_ddf_bvd(struct supertype *st,
2055
mdu_disk_info_t *dk, int fd, char *devname)
2057
/* fd and devname identify a device with-in the ddf container (st).
2058
* dk identifies a location in the new BVD.
2059
* We need to find suitable free space in that device and update
2060
* the phys_refnum and lba_offset for the newly created vd_config.
2061
* We might also want to update the type in the phys_disk
2064
* Alternately: fd == -1 and we have already chosen which device to
2065
* use and recorded in dlist->raid_disk;
2068
struct ddf_super *ddf = st->sb;
2069
struct vd_config *vc;
2071
unsigned int working;
2073
unsigned long long blocks, pos, esize;
2077
for (dl = ddf->dlist; dl ; dl = dl->next)
2078
if (dl->raiddisk == dk->raid_disk)
2081
for (dl = ddf->dlist; dl ; dl = dl->next)
2082
if (dl->major == dk->major &&
2083
dl->minor == dk->minor)
2086
if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
2089
vc = &ddf->currentconf->conf;
2090
lba_offset = ddf->currentconf->lba_offset;
2092
ex = get_extents(ddf, dl);
2097
blocks = __be64_to_cpu(vc->blocks);
2098
if (ddf->currentconf->block_sizes)
2099
blocks = ddf->currentconf->block_sizes[dk->raid_disk];
2102
esize = ex[i].start - pos;
2103
if (esize >= blocks)
2105
pos = ex[i].start + ex[i].size;
2107
} while (ex[i-1].size);
2113
ddf->currentdev = dk->raid_disk;
2114
vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
2115
lba_offset[dk->raid_disk] = __cpu_to_be64(pos);
2117
for (i = 0; i < ddf->max_part ; i++)
2118
if (dl->vlist[i] == NULL)
2120
if (i == ddf->max_part)
2122
dl->vlist[i] = ddf->currentconf;
2127
dl->devname = devname;
2129
/* Check how many working raid_disks, and if we can mark
2130
* array as optimal yet
2134
for (i = 0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
2135
if (vc->phys_refnum[i] != 0xffffffff)
2138
/* Find which virtual_entry */
2139
i = ddf->currentconf->vcnum;
2140
if (working == __be16_to_cpu(vc->prim_elmnt_count))
2141
ddf->virt->entries[i].state =
2142
(ddf->virt->entries[i].state & ~DDF_state_mask)
2143
| DDF_state_optimal;
2145
if (vc->prl == DDF_RAID6 &&
2146
working+1 == __be16_to_cpu(vc->prim_elmnt_count))
2147
ddf->virt->entries[i].state =
2148
(ddf->virt->entries[i].state & ~DDF_state_mask)
2149
| DDF_state_part_optimal;
2151
ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
2152
ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
2153
ddf->updates_pending = 1;
2156
/* add a device to a container, either while creating it or while
2157
* expanding a pre-existing container
2159
static int add_to_super_ddf(struct supertype *st,
2160
mdu_disk_info_t *dk, int fd, char *devname)
2162
struct ddf_super *ddf = st->sb;
2166
unsigned long long size;
2167
struct phys_disk_entry *pde;
2171
if (ddf->currentconf) {
2172
add_to_super_ddf_bvd(st, dk, fd, devname);
2176
/* This is device numbered dk->number. We need to create
2177
* a phys_disk entry and a more detailed disk_data entry.
2180
if (posix_memalign((void**)&dd, 512,
2181
sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part) != 0) {
2182
fprintf(stderr, Name
2183
": %s could allocate buffer for new disk, aborting\n",
2187
dd->major = major(stb.st_rdev);
2188
dd->minor = minor(stb.st_rdev);
2189
dd->devname = devname;
2193
dd->disk.magic = DDF_PHYS_DATA_MAGIC;
2195
tm = localtime(&now);
2196
sprintf(dd->disk.guid, "%8s%04d%02d%02d",
2197
T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
2198
*(__u32*)(dd->disk.guid + 16) = random32();
2199
*(__u32*)(dd->disk.guid + 20) = random32();
2202
/* Cannot be bothered finding a CRC of some irrelevant details*/
2203
dd->disk.refnum = random32();
2204
for (i = __be16_to_cpu(ddf->active->max_pd_entries);
2206
if (ddf->phys->entries[i-1].refnum == dd->disk.refnum)
2210
dd->disk.forced_ref = 1;
2211
dd->disk.forced_guid = 1;
2212
memset(dd->disk.vendor, ' ', 32);
2213
memcpy(dd->disk.vendor, "Linux", 5);
2214
memset(dd->disk.pad, 0xff, 442);
2215
for (i = 0; i < ddf->max_part ; i++)
2216
dd->vlist[i] = NULL;
2218
n = __be16_to_cpu(ddf->phys->used_pdes);
2219
pde = &ddf->phys->entries[n];
2222
if (st->update_tail) {
2223
int len = (sizeof(struct phys_disk) +
2224
sizeof(struct phys_disk_entry));
2225
struct phys_disk *pd;
2228
pd->magic = DDF_PHYS_RECORDS_MAGIC;
2229
pd->used_pdes = __cpu_to_be16(n);
2230
pde = &pd->entries[0];
2234
ddf->phys->used_pdes = __cpu_to_be16(n);
2237
memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2238
pde->refnum = dd->disk.refnum;
2239
pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2240
pde->state = __cpu_to_be16(DDF_Online);
2241
get_dev_size(fd, NULL, &size);
2242
/* We are required to reserve 32Meg, and record the size in sectors */
2243
pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
2244
sprintf(pde->path, "%17.17s","Information: nil") ;
2245
memset(pde->pad, 0xff, 6);
2247
dd->size = size >> 9;
2248
if (st->update_tail) {
2249
dd->next = ddf->add_list;
2252
dd->next = ddf->dlist;
2254
ddf->updates_pending = 1;
2261
* This is the write_init_super method for a ddf container. It is
2262
* called when creating a container or adding another device to a
2266
static unsigned char null_conf[4096+512];
2268
static int __write_init_super_ddf(struct supertype *st, int do_close)
2271
struct ddf_super *ddf = st->sb;
2278
unsigned long long size, sector;
2280
/* try to write updated metadata,
2281
* if we catch a failure move on to the next disk
2283
for (d = ddf->dlist; d; d=d->next) {
2290
/* We need to fill in the primary, (secondary) and workspace
2291
* lba's in the headers, set their checksums,
2292
* Also checksum phys, virt....
2294
* Then write everything out, finally the anchor is written.
2296
get_dev_size(fd, NULL, &size);
2298
ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
2299
ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
2300
ddf->anchor.seq = __cpu_to_be32(1);
2301
memcpy(&ddf->primary, &ddf->anchor, 512);
2302
memcpy(&ddf->secondary, &ddf->anchor, 512);
2304
ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
2305
ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
2306
ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
2308
ddf->primary.openflag = 0;
2309
ddf->primary.type = DDF_HEADER_PRIMARY;
2311
ddf->secondary.openflag = 0;
2312
ddf->secondary.type = DDF_HEADER_SECONDARY;
2314
ddf->primary.crc = calc_crc(&ddf->primary, 512);
2315
ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
2317
sector = size - 16*1024*2;
2318
lseek64(fd, sector<<9, 0);
2319
if (write(fd, &ddf->primary, 512) < 0)
2322
ddf->controller.crc = calc_crc(&ddf->controller, 512);
2323
if (write(fd, &ddf->controller, 512) < 0)
2326
ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
2328
if (write(fd, ddf->phys, ddf->pdsize) < 0)
2331
ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
2332
if (write(fd, ddf->virt, ddf->vdsize) < 0)
2335
/* Now write lots of config records. */
2336
n_config = ddf->max_part;
2337
conf_size = ddf->conf_rec_len * 512;
2338
for (i = 0 ; i <= n_config ; i++) {
2339
struct vcl *c = d->vlist[i];
2341
c = (struct vcl*)d->spare;
2344
c->conf.crc = calc_crc(&c->conf, conf_size);
2345
if (write(fd, &c->conf, conf_size) < 0)
2348
char *null_aligned = (char*)((((unsigned long)null_conf)+511)&~511UL);
2349
if (null_conf[0] != 0xff)
2350
memset(null_conf, 0xff, sizeof(null_conf));
2351
unsigned int togo = conf_size;
2352
while (togo > sizeof(null_conf)-512) {
2353
if (write(fd, null_aligned, sizeof(null_conf)-512) < 0)
2355
togo -= sizeof(null_conf)-512;
2357
if (write(fd, null_aligned, togo) < 0)
2363
d->disk.crc = calc_crc(&d->disk, 512);
2364
if (write(fd, &d->disk, 512) < 0)
2367
/* Maybe do the same for secondary */
2369
lseek64(fd, (size-1)*512, SEEK_SET);
2370
if (write(fd, &ddf->anchor, 512) < 0)
2376
for (d = ddf->dlist; d; d=d->next) {
2381
return attempts != successes;
2384
static int write_init_super_ddf(struct supertype *st)
2386
struct ddf_super *ddf = st->sb;
2387
struct vcl *currentconf = ddf->currentconf;
2389
/* we are done with currentconf reset it to point st at the container */
2390
ddf->currentconf = NULL;
2392
if (st->update_tail) {
2393
/* queue the virtual_disk and vd_config as metadata updates */
2394
struct virtual_disk *vd;
2395
struct vd_config *vc;
2399
int len = (sizeof(struct phys_disk) +
2400
sizeof(struct phys_disk_entry));
2402
/* adding a disk to the container. */
2406
append_metadata_update(st, ddf->add_list->mdupdate, len);
2407
ddf->add_list->mdupdate = NULL;
2411
/* Newly created VD */
2413
/* First the virtual disk. We have a slightly fake header */
2414
len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry);
2417
vd->entries[0] = ddf->virt->entries[currentconf->vcnum];
2418
vd->populated_vdes = __cpu_to_be16(currentconf->vcnum);
2419
append_metadata_update(st, vd, len);
2421
/* Then the vd_config */
2422
len = ddf->conf_rec_len * 512;
2424
memcpy(vc, ¤tconf->conf, len);
2425
append_metadata_update(st, vc, len);
2427
/* FIXME I need to close the fds! */
2431
for (d = ddf->dlist; d; d=d->next)
2432
while (Kill(d->devname, NULL, 0, 1, 1) == 0);
2433
return __write_init_super_ddf(st, 1);
2439
static __u64 avail_size_ddf(struct supertype *st, __u64 devsize)
2441
/* We must reserve the last 32Meg */
2442
if (devsize <= 32*1024*2)
2444
return devsize - 32*1024*2;
2449
static int reserve_space(struct supertype *st, int raiddisks,
2450
unsigned long long size, int chunk,
2451
unsigned long long *freesize)
2453
/* Find 'raiddisks' spare extents at least 'size' big (but
2454
* only caring about multiples of 'chunk') and remember
2456
* If the cannot be found, fail.
2459
struct ddf_super *ddf = st->sb;
2462
for (dl = ddf->dlist; dl ; dl=dl->next) {
2466
/* Now find largest extent on each device */
2467
for (dl = ddf->dlist ; dl ; dl=dl->next) {
2468
struct extent *e = get_extents(ddf, dl);
2469
unsigned long long pos = 0;
2472
unsigned long long minsize = size;
2480
unsigned long long esize;
2481
esize = e[i].start - pos;
2482
if (esize >= minsize) {
2486
pos = e[i].start + e[i].size;
2488
} while (e[i-1].size);
2491
dl->esize = minsize;
2495
if (cnt < raiddisks) {
2496
fprintf(stderr, Name ": not enough devices with space to create array.\n");
2497
return 0; /* No enough free spaces large enough */
2500
/* choose the largest size of which there are at least 'raiddisk' */
2501
for (dl = ddf->dlist ; dl ; dl=dl->next) {
2503
if (dl->esize <= size)
2505
/* This is bigger than 'size', see if there are enough */
2507
for (dl2 = dl; dl2 ; dl2=dl2->next)
2508
if (dl2->esize >= dl->esize)
2510
if (cnt >= raiddisks)
2514
size = size / chunk;
2519
fprintf(stderr, Name ": not enough spare devices to create array.\n");
2523
/* We have a 'size' of which there are enough spaces.
2524
* We simply do a first-fit */
2526
for (dl = ddf->dlist ; dl && cnt < raiddisks ; dl=dl->next) {
2527
if (dl->esize < size)
2539
validate_geometry_ddf_container(struct supertype *st,
2540
int level, int layout, int raiddisks,
2541
int chunk, unsigned long long size,
2542
char *dev, unsigned long long *freesize,
2545
static int validate_geometry_ddf_bvd(struct supertype *st,
2546
int level, int layout, int raiddisks,
2547
int chunk, unsigned long long size,
2548
char *dev, unsigned long long *freesize,
2551
static int validate_geometry_ddf(struct supertype *st,
2552
int level, int layout, int raiddisks,
2553
int chunk, unsigned long long size,
2554
char *dev, unsigned long long *freesize,
2561
/* ddf potentially supports lots of things, but it depends on
2562
* what devices are offered (and maybe kernel version?)
2563
* If given unused devices, we will make a container.
2564
* If given devices in a container, we will make a BVD.
2565
* If given BVDs, we make an SVD, changing all the GUIDs in the process.
2568
if (level == LEVEL_CONTAINER) {
2569
/* Must be a fresh device to add to a container */
2570
return validate_geometry_ddf_container(st, level, layout,
2572
size, dev, freesize,
2577
/* Initial sanity check. Exclude illegal levels. */
2579
for (i=0; ddf_level_num[i].num1 != MAXINT; i++)
2580
if (ddf_level_num[i].num2 == level)
2582
if (ddf_level_num[i].num1 == MAXINT) {
2584
fprintf(stderr, Name ": DDF does not support level %d arrays\n",
2588
/* Should check layout? etc */
2590
if (st->sb && freesize) {
2591
/* --create was given a container to create in.
2592
* So we need to check that there are enough
2593
* free spaces and return the amount of space.
2594
* We may as well remember which drives were
2595
* chosen so that add_to_super/getinfo_super
2598
return reserve_space(st, raiddisks, size, chunk, freesize);
2604
/* A container has already been opened, so we are
2605
* creating in there. Maybe a BVD, maybe an SVD.
2606
* Should make a distinction one day.
2608
return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
2609
chunk, size, dev, freesize,
2612
/* This is the first device for the array.
2613
* If it is a container, we read it in and do automagic allocations,
2614
* no other devices should be given.
2615
* Otherwise it must be a member device of a container, and we
2616
* do manual allocation.
2617
* Later we should check for a BVD and make an SVD.
2619
fd = open(dev, O_RDONLY|O_EXCL, 0);
2621
sra = sysfs_read(fd, 0, GET_VERSION);
2623
if (sra && sra->array.major_version == -1 &&
2624
strcmp(sra->text_version, "ddf") == 0) {
2627
/* find space for 'n' devices. */
2628
/* remember the devices */
2629
/* Somehow return the fact that we have enough */
2634
Name ": ddf: Cannot create this array "
2635
"on device %s - a container is required.\n",
2639
if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2641
fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2642
dev, strerror(errno));
2645
/* Well, it is in use by someone, maybe a 'ddf' container. */
2646
cfd = open_container(fd);
2650
fprintf(stderr, Name ": ddf: Cannot use %s: %s\n",
2651
dev, strerror(EBUSY));
2654
sra = sysfs_read(cfd, 0, GET_VERSION);
2656
if (sra && sra->array.major_version == -1 &&
2657
strcmp(sra->text_version, "ddf") == 0) {
2658
/* This is a member of a ddf container. Load the container
2659
* and try to create a bvd
2661
struct ddf_super *ddf;
2662
if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL, 1) == 0) {
2664
st->container_dev = fd2devnum(cfd);
2666
return validate_geometry_ddf_bvd(st, level, layout,
2667
raiddisks, chunk, size,
2672
} else /* device may belong to a different container */
2679
validate_geometry_ddf_container(struct supertype *st,
2680
int level, int layout, int raiddisks,
2681
int chunk, unsigned long long size,
2682
char *dev, unsigned long long *freesize,
2686
unsigned long long ldsize;
2688
if (level != LEVEL_CONTAINER)
2693
fd = open(dev, O_RDONLY|O_EXCL, 0);
2696
fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2697
dev, strerror(errno));
2700
if (!get_dev_size(fd, dev, &ldsize)) {
2706
*freesize = avail_size_ddf(st, ldsize >> 9);
2713
static int validate_geometry_ddf_bvd(struct supertype *st,
2714
int level, int layout, int raiddisks,
2715
int chunk, unsigned long long size,
2716
char *dev, unsigned long long *freesize,
2720
struct ddf_super *ddf = st->sb;
2722
unsigned long long pos = 0;
2723
unsigned long long maxsize;
2726
/* ddf/bvd supports lots of things, but not containers */
2727
if (level == LEVEL_CONTAINER) {
2729
fprintf(stderr, Name ": DDF cannot create a container within an container\n");
2732
/* We must have the container info already read in. */
2737
/* General test: make sure there is space for
2738
* 'raiddisks' device extents of size 'size'.
2740
unsigned long long minsize = size;
2744
for (dl = ddf->dlist; dl ; dl = dl->next)
2750
e = get_extents(ddf, dl);
2753
unsigned long long esize;
2754
esize = e[i].start - pos;
2755
if (esize >= minsize)
2757
pos = e[i].start + e[i].size;
2759
} while (e[i-1].size);
2764
if (dcnt < raiddisks) {
2767
Name ": ddf: Not enough devices with "
2768
"space for this array (%d < %d)\n",
2774
/* This device must be a member of the set */
2775
if (stat(dev, &stb) < 0)
2777
if ((S_IFMT & stb.st_mode) != S_IFBLK)
2779
for (dl = ddf->dlist ; dl ; dl = dl->next) {
2780
if (dl->major == (int)major(stb.st_rdev) &&
2781
dl->minor == (int)minor(stb.st_rdev))
2786
fprintf(stderr, Name ": ddf: %s is not in the "
2791
e = get_extents(ddf, dl);
2795
unsigned long long esize;
2796
esize = e[i].start - pos;
2797
if (esize >= maxsize)
2799
pos = e[i].start + e[i].size;
2801
} while (e[i-1].size);
2802
*freesize = maxsize;
2808
static int load_super_ddf_all(struct supertype *st, int fd,
2809
void **sbp, char *devname, int keep_fd)
2812
struct ddf_super *super;
2813
struct mdinfo *sd, *best = NULL;
2819
sra = sysfs_read(fd, 0, GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE);
2822
if (sra->array.major_version != -1 ||
2823
sra->array.minor_version != -2 ||
2824
strcmp(sra->text_version, "ddf") != 0)
2827
if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0)
2829
memset(super, 0, sizeof(*super));
2831
/* first, try each device, and choose the best ddf */
2832
for (sd = sra->devs ; sd ; sd = sd->next) {
2834
sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2835
dfd = dev_open(nm, O_RDONLY);
2838
rv = load_ddf_headers(dfd, super, NULL);
2841
seq = __be32_to_cpu(super->active->seq);
2842
if (super->active->openflag)
2844
if (!best || seq > bestseq) {
2852
/* OK, load this ddf */
2853
sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2854
dfd = dev_open(nm, O_RDONLY);
2857
load_ddf_headers(dfd, super, NULL);
2858
load_ddf_global(dfd, super, NULL);
2860
/* Now we need the device-local bits */
2861
for (sd = sra->devs ; sd ; sd = sd->next) {
2864
sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2865
dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2868
rv = load_ddf_headers(dfd, super, NULL);
2870
rv = load_ddf_local(dfd, super, NULL, keep_fd);
2871
if (!keep_fd) close(dfd);
2875
if (st->subarray[0]) {
2880
val = strtoul(st->subarray, &ep, 10);
2886
for (v = super->conflist; v; v = v->next)
2887
if (v->vcnum == val)
2888
super->currentconf = v;
2889
if (!super->currentconf) {
2896
if (st->ss == NULL) {
2897
st->ss = &super_ddf;
2898
st->minor_version = 0;
2900
st->container_dev = fd2devnum(fd);
2902
st->loaded_container = 1;
2905
#endif /* MDASSEMBLE */
2907
static struct mdinfo *container_content_ddf(struct supertype *st)
2909
/* Given a container loaded by load_super_ddf_all,
2910
* extract information about all the arrays into
2913
* For each vcl in conflist: create an mdinfo, fill it in,
2914
* then look for matching devices (phys_refnum) in dlist
2915
* and create appropriate device mdinfo.
2917
struct ddf_super *ddf = st->sb;
2918
struct mdinfo *rest = NULL;
2921
for (vc = ddf->conflist ; vc ; vc=vc->next)
2925
struct mdinfo *this;
2926
this = malloc(sizeof(*this));
2927
memset(this, 0, sizeof(*this));
2931
this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2932
this->array.raid_disks =
2933
__be16_to_cpu(vc->conf.prim_elmnt_count);
2934
this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2935
this->array.raid_disks);
2936
this->array.md_minor = -1;
2937
this->array.major_version = -1;
2938
this->array.minor_version = -2;
2939
this->array.ctime = DECADE +
2940
__be32_to_cpu(*(__u32*)(vc->conf.guid+16));
2941
this->array.utime = DECADE +
2942
__be32_to_cpu(vc->conf.timestamp);
2943
this->array.chunk_size = 512 << vc->conf.chunk_shift;
2946
if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2947
(ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
2949
this->array.state = 0;
2950
this->resync_start = 0;
2952
this->array.state = 1;
2953
this->resync_start = MaxSector;
2955
memcpy(this->name, ddf->virt->entries[i].name, 16);
2958
if (this->name[j] == ' ')
2961
memset(this->uuid, 0, sizeof(this->uuid));
2962
this->component_size = __be64_to_cpu(vc->conf.blocks);
2963
this->array.size = this->component_size / 2;
2964
this->container_member = i;
2966
ddf->currentconf = vc;
2967
uuid_from_super_ddf(st, this->uuid);
2968
ddf->currentconf = NULL;
2970
sprintf(this->text_version, "/%s/%d",
2971
devnum2devname(st->container_dev),
2972
this->container_member);
2974
for (i = 0 ; i < ddf->mppe ; i++) {
2978
if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
2981
this->array.working_disks++;
2983
for (d = ddf->dlist; d ; d=d->next)
2984
if (d->disk.refnum == vc->conf.phys_refnum[i])
2987
/* Haven't found that one yet, maybe there are others */
2990
dev = malloc(sizeof(*dev));
2991
memset(dev, 0, sizeof(*dev));
2992
dev->next = this->devs;
2995
dev->disk.number = __be32_to_cpu(d->disk.refnum);
2996
dev->disk.major = d->major;
2997
dev->disk.minor = d->minor;
2998
dev->disk.raid_disk = i;
2999
dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
3000
dev->recovery_start = MaxSector;
3002
dev->events = __be32_to_cpu(ddf->primary.seq);
3003
dev->data_offset = __be64_to_cpu(vc->lba_offset[i]);
3004
dev->component_size = __be64_to_cpu(vc->conf.blocks);
3006
strcpy(dev->name, d->devname);
3012
static int store_super_ddf(struct supertype *st, int fd)
3014
struct ddf_super *ddf = st->sb;
3015
unsigned long long dsize;
3022
/* ->dlist and ->conflist will be set for updates, currently not
3025
if (ddf->dlist || ddf->conflist)
3028
if (!get_dev_size(fd, NULL, &dsize))
3031
if (posix_memalign(&buf, 512, 512) != 0)
3033
memset(buf, 0, 512);
3035
lseek64(fd, dsize-512, 0);
3036
rc = write(fd, buf, 512);
3043
static int compare_super_ddf(struct supertype *st, struct supertype *tst)
3047
* 0 same, or first was empty, and second was copied
3048
* 1 second had wrong number
3050
* 3 wrong other info
3052
struct ddf_super *first = st->sb;
3053
struct ddf_super *second = tst->sb;
3061
if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
3064
/* FIXME should I look at anything else? */
3070
* A new array 'a' has been started which claims to be instance 'inst'
3071
* within container 'c'.
3072
* We need to confirm that the array matches the metadata in 'c' so
3073
* that we don't corrupt any metadata.
3075
static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
3077
dprintf("ddf: open_new %s\n", inst);
3078
a->info.container_member = atoi(inst);
3083
* The array 'a' is to be marked clean in the metadata.
3084
* If '->resync_start' is not ~(unsigned long long)0, then the array is only
3085
* clean up to the point (in sectors). If that cannot be recorded in the
3086
* metadata, then leave it as dirty.
3088
* For DDF, we need to clear the DDF_state_inconsistent bit in the
3089
* !global! virtual_disk.virtual_entry structure.
3091
static int ddf_set_array_state(struct active_array *a, int consistent)
3093
struct ddf_super *ddf = a->container->sb;
3094
int inst = a->info.container_member;
3095
int old = ddf->virt->entries[inst].state;
3096
if (consistent == 2) {
3097
/* Should check if a recovery should be started FIXME */
3099
if (!is_resync_complete(&a->info))
3103
ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
3105
ddf->virt->entries[inst].state |= DDF_state_inconsistent;
3106
if (old != ddf->virt->entries[inst].state)
3107
ddf->updates_pending = 1;
3109
old = ddf->virt->entries[inst].init_state;
3110
ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
3111
if (is_resync_complete(&a->info))
3112
ddf->virt->entries[inst].init_state |= DDF_init_full;
3113
else if (a->info.resync_start == 0)
3114
ddf->virt->entries[inst].init_state |= DDF_init_not;
3116
ddf->virt->entries[inst].init_state |= DDF_init_quick;
3117
if (old != ddf->virt->entries[inst].init_state)
3118
ddf->updates_pending = 1;
3120
dprintf("ddf mark %d %s %llu\n", inst, consistent?"clean":"dirty",
3121
a->info.resync_start);
3126
* The state of each disk is stored in the global phys_disk structure
3127
* in phys_disk.entries[n].state.
3128
* This makes various combinations awkward.
3129
* - When a device fails in any array, it must be failed in all arrays
3130
* that include a part of this device.
3131
* - When a component is rebuilding, we cannot include it officially in the
3132
* array unless this is the only array that uses the device.
3134
* So: when transitioning:
3135
* Online -> failed, just set failed flag. monitor will propagate
3136
* spare -> online, the device might need to be added to the array.
3137
* spare -> failed, just set failed. Don't worry if in array or not.
3139
static void ddf_set_disk(struct active_array *a, int n, int state)
3141
struct ddf_super *ddf = a->container->sb;
3142
unsigned int inst = a->info.container_member;
3143
struct vd_config *vc = find_vdcr(ddf, inst);
3144
int pd = find_phys(ddf, vc->phys_refnum[n]);
3148
dprintf("ddf: cannot find instance %d!!\n", inst);
3152
/* disk doesn't currently exist. If it is now in_sync,
3154
if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
3155
/* Find dev 'n' in a->info->devs, determine the
3156
* ddf refnum, and set vc->phys_refnum and update
3162
int old = ddf->phys->entries[pd].state;
3163
if (state & DS_FAULTY)
3164
ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Failed);
3165
if (state & DS_INSYNC) {
3166
ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Online);
3167
ddf->phys->entries[pd].state &= __cpu_to_be16(~DDF_Rebuilding);
3169
if (old != ddf->phys->entries[pd].state)
3170
ddf->updates_pending = 1;
3173
dprintf("ddf: set_disk %d to %x\n", n, state);
3175
/* Now we need to check the state of the array and update
3176
* virtual_disk.entries[n].state.
3177
* It needs to be one of "optimal", "degraded", "failed".
3178
* I don't understand 'deleted' or 'missing'.
3181
for (i=0; i < a->info.array.raid_disks; i++) {
3182
pd = find_phys(ddf, vc->phys_refnum[i]);
3185
st = __be16_to_cpu(ddf->phys->entries[pd].state);
3186
if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
3190
state = DDF_state_degraded;
3191
if (working == a->info.array.raid_disks)
3192
state = DDF_state_optimal;
3193
else switch(vc->prl) {
3197
state = DDF_state_failed;
3201
state = DDF_state_failed;
3205
if (working < a->info.array.raid_disks-1)
3206
state = DDF_state_failed;
3209
if (working < a->info.array.raid_disks-2)
3210
state = DDF_state_failed;
3211
else if (working == a->info.array.raid_disks-1)
3212
state = DDF_state_part_optimal;
3216
if (ddf->virt->entries[inst].state !=
3217
((ddf->virt->entries[inst].state & ~DDF_state_mask)
3220
ddf->virt->entries[inst].state =
3221
(ddf->virt->entries[inst].state & ~DDF_state_mask)
3223
ddf->updates_pending = 1;
3228
static void ddf_sync_metadata(struct supertype *st)
3232
* Write all data to all devices.
3233
* Later, we might be able to track whether only local changes
3234
* have been made, or whether any global data has been changed,
3235
* but ddf is sufficiently weird that it probably always
3236
* changes global data ....
3238
struct ddf_super *ddf = st->sb;
3239
if (!ddf->updates_pending)
3241
ddf->updates_pending = 0;
3242
__write_init_super_ddf(st, 0);
3243
dprintf("ddf: sync_metadata\n");
3246
static void ddf_process_update(struct supertype *st,
3247
struct metadata_update *update)
3249
/* Apply this update to the metadata.
3250
* The first 4 bytes are a DDF_*_MAGIC which guides
3252
* Possible update are:
3253
* DDF_PHYS_RECORDS_MAGIC
3254
* Add a new physical device. Changes to this record
3255
* only happen implicitly.
3256
* used_pdes is the device number.
3257
* DDF_VIRT_RECORDS_MAGIC
3258
* Add a new VD. Possibly also change the 'access' bits.
3259
* populated_vdes is the entry number.
3261
* New or updated VD. the VIRT_RECORD must already
3262
* exist. For an update, phys_refnum and lba_offset
3263
* (at least) are updated, and the VD_CONF must
3264
* be written to precisely those devices listed with
3266
* DDF_SPARE_ASSIGN_MAGIC
3267
* replacement Spare Assignment Record... but for which device?
3270
* - to create a new array, we send a VIRT_RECORD and
3271
* a VD_CONF. Then assemble and start the array.
3272
* - to activate a spare we send a VD_CONF to add the phys_refnum
3273
* and offset. This will also mark the spare as active with
3274
* a spare-assignment record.
3276
struct ddf_super *ddf = st->sb;
3277
__u32 *magic = (__u32*)update->buf;
3278
struct phys_disk *pd;
3279
struct virtual_disk *vd;
3280
struct vd_config *vc;
3286
dprintf("Process update %x\n", *magic);
3289
case DDF_PHYS_RECORDS_MAGIC:
3291
if (update->len != (sizeof(struct phys_disk) +
3292
sizeof(struct phys_disk_entry)))
3294
pd = (struct phys_disk*)update->buf;
3296
ent = __be16_to_cpu(pd->used_pdes);
3297
if (ent >= __be16_to_cpu(ddf->phys->max_pdes))
3299
if (!all_ff(ddf->phys->entries[ent].guid))
3301
ddf->phys->entries[ent] = pd->entries[0];
3302
ddf->phys->used_pdes = __cpu_to_be16(1 +
3303
__be16_to_cpu(ddf->phys->used_pdes));
3304
ddf->updates_pending = 1;
3305
if (ddf->add_list) {
3306
struct active_array *a;
3307
struct dl *al = ddf->add_list;
3308
ddf->add_list = al->next;
3310
al->next = ddf->dlist;
3313
/* As a device has been added, we should check
3314
* for any degraded devices that might make
3315
* use of this spare */
3316
for (a = st->arrays ; a; a=a->next)
3317
a->check_degraded = 1;
3321
case DDF_VIRT_RECORDS_MAGIC:
3323
if (update->len != (sizeof(struct virtual_disk) +
3324
sizeof(struct virtual_entry)))
3326
vd = (struct virtual_disk*)update->buf;
3328
ent = __be16_to_cpu(vd->populated_vdes);
3329
if (ent >= __be16_to_cpu(ddf->virt->max_vdes))
3331
if (!all_ff(ddf->virt->entries[ent].guid))
3333
ddf->virt->entries[ent] = vd->entries[0];
3334
ddf->virt->populated_vdes = __cpu_to_be16(1 +
3335
__be16_to_cpu(ddf->virt->populated_vdes));
3336
ddf->updates_pending = 1;
3339
case DDF_VD_CONF_MAGIC:
3340
dprintf("len %d %d\n", update->len, ddf->conf_rec_len);
3342
mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries);
3343
if ((unsigned)update->len != ddf->conf_rec_len * 512)
3345
vc = (struct vd_config*)update->buf;
3346
for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
3347
if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
3349
dprintf("vcl = %p\n", vcl);
3351
/* An update, just copy the phys_refnum and lba_offset
3354
memcpy(vcl->conf.phys_refnum, vc->phys_refnum,
3355
mppe * (sizeof(__u32) + sizeof(__u64)));
3360
vcl = update->space;
3361
update->space = NULL;
3362
vcl->next = ddf->conflist;
3363
memcpy(&vcl->conf, vc, update->len);
3364
vcl->lba_offset = (__u64*)
3365
&vcl->conf.phys_refnum[mppe];
3366
ddf->conflist = vcl;
3368
/* Now make sure vlist is correct for each dl. */
3369
for (dl = ddf->dlist; dl; dl = dl->next) {
3371
unsigned int vn = 0;
3372
for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
3373
for (dn=0; dn < ddf->mppe ; dn++)
3374
if (vcl->conf.phys_refnum[dn] ==
3376
dprintf("dev %d has %p at %d\n",
3377
dl->pdnum, vcl, vn);
3378
dl->vlist[vn++] = vcl;
3381
while (vn < ddf->max_part)
3382
dl->vlist[vn++] = NULL;
3384
ddf->phys->entries[dl->pdnum].type &=
3385
~__cpu_to_be16(DDF_Global_Spare);
3386
ddf->phys->entries[dl->pdnum].type |=
3387
__cpu_to_be16(DDF_Active_in_VD);
3390
ddf->phys->entries[dl->pdnum].type &=
3391
~__cpu_to_be16(DDF_Global_Spare);
3392
ddf->phys->entries[dl->pdnum].type |=
3393
__cpu_to_be16(DDF_Spare);
3395
if (!dl->vlist[0] && !dl->spare) {
3396
ddf->phys->entries[dl->pdnum].type |=
3397
__cpu_to_be16(DDF_Global_Spare);
3398
ddf->phys->entries[dl->pdnum].type &=
3399
~__cpu_to_be16(DDF_Spare |
3403
ddf->updates_pending = 1;
3405
case DDF_SPARE_ASSIGN_MAGIC:
3410
static void ddf_prepare_update(struct supertype *st,
3411
struct metadata_update *update)
3413
/* This update arrived at managemon.
3414
* We are about to pass it to monitor.
3415
* If a malloc is needed, do it here.
3417
struct ddf_super *ddf = st->sb;
3418
__u32 *magic = (__u32*)update->buf;
3419
if (*magic == DDF_VD_CONF_MAGIC)
3420
if (posix_memalign(&update->space, 512,
3421
offsetof(struct vcl, conf)
3422
+ ddf->conf_rec_len * 512) != 0)
3423
update->space = NULL;
3427
* Check if the array 'a' is degraded but not failed.
3428
* If it is, find as many spares as are available and needed and
3429
* arrange for their inclusion.
3430
* We only choose devices which are not already in the array,
3431
* and prefer those with a spare-assignment to this array.
3432
* otherwise we choose global spares - assuming always that
3433
* there is enough room.
3434
* For each spare that we assign, we return an 'mdinfo' which
3435
* describes the position for the device in the array.
3436
* We also add to 'updates' a DDF_VD_CONF_MAGIC update with
3437
* the new phys_refnum and lba_offset values.
3439
* Only worry about BVDs at the moment.
3441
static struct mdinfo *ddf_activate_spare(struct active_array *a,
3442
struct metadata_update **updates)
3446
struct ddf_super *ddf = a->container->sb;
3448
struct mdinfo *rv = NULL;
3450
struct metadata_update *mu;
3453
struct vd_config *vc;
3456
for (d = a->info.devs ; d ; d = d->next) {
3457
if ((d->curr_state & DS_FAULTY) &&
3459
/* wait for Removal to happen */
3461
if (d->state_fd >= 0)
3465
dprintf("ddf_activate: working=%d (%d) level=%d\n", working, a->info.array.raid_disks,
3466
a->info.array.level);
3467
if (working == a->info.array.raid_disks)
3468
return NULL; /* array not degraded */
3469
switch (a->info.array.level) {
3472
return NULL; /* failed */
3476
if (working < a->info.array.raid_disks - 1)
3477
return NULL; /* failed */
3480
if (working < a->info.array.raid_disks - 2)
3481
return NULL; /* failed */
3483
default: /* concat or stripe */
3484
return NULL; /* failed */
3487
/* For each slot, if it is not working, find a spare */
3489
for (i = 0; i < a->info.array.raid_disks; i++) {
3490
for (d = a->info.devs ; d ; d = d->next)
3491
if (d->disk.raid_disk == i)
3493
dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
3494
if (d && (d->state_fd >= 0))
3497
/* OK, this device needs recovery. Find a spare */
3499
for ( ; dl ; dl = dl->next) {
3500
unsigned long long esize;
3501
unsigned long long pos;
3504
int is_dedicated = 0;
3507
/* If in this array, skip */
3508
for (d2 = a->info.devs ; d2 ; d2 = d2->next)
3509
if (d2->disk.major == dl->major &&
3510
d2->disk.minor == dl->minor) {
3511
dprintf("%x:%x already in array\n", dl->major, dl->minor);
3516
if (ddf->phys->entries[dl->pdnum].type &
3517
__cpu_to_be16(DDF_Spare)) {
3518
/* Check spare assign record */
3520
if (dl->spare->type & DDF_spare_dedicated) {
3521
/* check spare_ents for guid */
3523
j < __be16_to_cpu(dl->spare->populated);
3525
if (memcmp(dl->spare->spare_ents[j].guid,
3526
ddf->virt->entries[a->info.container_member].guid,
3533
} else if (ddf->phys->entries[dl->pdnum].type &
3534
__cpu_to_be16(DDF_Global_Spare)) {
3537
if ( ! (is_dedicated ||
3538
(is_global && global_ok))) {
3539
dprintf("%x:%x not suitable: %d %d\n", dl->major, dl->minor,
3540
is_dedicated, is_global);
3544
/* We are allowed to use this device - is there space?
3545
* We need a->info.component_size sectors */
3546
ex = get_extents(ddf, dl);
3548
dprintf("cannot get extents\n");
3555
esize = ex[j].start - pos;
3556
if (esize >= a->info.component_size)
3558
pos = ex[i].start + ex[i].size;
3560
} while (ex[i-1].size);
3563
if (esize < a->info.component_size) {
3564
dprintf("%x:%x has no room: %llu %llu\n", dl->major, dl->minor,
3565
esize, a->info.component_size);
3570
/* Cool, we have a device with some space at pos */
3571
di = malloc(sizeof(*di));
3574
memset(di, 0, sizeof(*di));
3575
di->disk.number = i;
3576
di->disk.raid_disk = i;
3577
di->disk.major = dl->major;
3578
di->disk.minor = dl->minor;
3580
di->recovery_start = 0;
3581
di->data_offset = pos;
3582
di->component_size = a->info.component_size;
3583
di->container_member = dl->pdnum;
3586
dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3591
if (!dl && ! global_ok) {
3592
/* not enough dedicated spares, try global */
3600
/* No spares found */
3602
/* Now 'rv' has a list of devices to return.
3603
* Create a metadata_update record to update the
3604
* phys_refnum and lba_offset values
3606
mu = malloc(sizeof(*mu));
3607
if (mu && posix_memalign(&mu->space, 512, sizeof(struct vcl)) != 0) {
3613
struct mdinfo *n = rv->next;
3621
mu->buf = malloc(ddf->conf_rec_len * 512);
3622
mu->len = ddf->conf_rec_len;
3623
mu->next = *updates;
3624
vc = find_vdcr(ddf, a->info.container_member);
3625
memcpy(mu->buf, vc, ddf->conf_rec_len * 512);
3627
vc = (struct vd_config*)mu->buf;
3628
lba = (__u64*)&vc->phys_refnum[ddf->mppe];
3629
for (di = rv ; di ; di = di->next) {
3630
vc->phys_refnum[di->disk.raid_disk] =
3631
ddf->phys->entries[dl->pdnum].refnum;
3632
lba[di->disk.raid_disk] = di->data_offset;
3637
#endif /* MDASSEMBLE */
3639
static int ddf_level_to_layout(int level)
3646
return ALGORITHM_LEFT_SYMMETRIC;
3648
return ALGORITHM_ROTATING_N_CONTINUE;
3656
struct superswitch super_ddf = {
3658
.examine_super = examine_super_ddf,
3659
.brief_examine_super = brief_examine_super_ddf,
3660
.brief_examine_subarrays = brief_examine_subarrays_ddf,
3661
.export_examine_super = export_examine_super_ddf,
3662
.detail_super = detail_super_ddf,
3663
.brief_detail_super = brief_detail_super_ddf,
3664
.validate_geometry = validate_geometry_ddf,
3665
.write_init_super = write_init_super_ddf,
3666
.add_to_super = add_to_super_ddf,
3668
.match_home = match_home_ddf,
3669
.uuid_from_super= uuid_from_super_ddf,
3670
.getinfo_super = getinfo_super_ddf,
3671
.update_super = update_super_ddf,
3673
.avail_size = avail_size_ddf,
3675
.compare_super = compare_super_ddf,
3677
.load_super = load_super_ddf,
3678
.init_super = init_super_ddf,
3679
.store_super = store_super_ddf,
3680
.free_super = free_super_ddf,
3681
.match_metadata_desc = match_metadata_desc_ddf,
3682
.container_content = container_content_ddf,
3683
.default_layout = ddf_level_to_layout,
3689
.open_new = ddf_open_new,
3690
.set_array_state= ddf_set_array_state,
3691
.set_disk = ddf_set_disk,
3692
.sync_metadata = ddf_sync_metadata,
3693
.process_update = ddf_process_update,
3694
.prepare_update = ddf_prepare_update,
3695
.activate_spare = ddf_activate_spare,