2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2006-2007 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 static inline int ROUND_UP(int a
, int base
)
36 return ((a
+base
-1)/base
)*base
;
39 /* a non-official T10 name for creation GUIDs */
40 static char T10
[] = "Linux-MD";
42 /* DDF timestamps are 1980 based, so we need to add
43 * second-in-decade-of-seventies to convert to linux timestamps.
44 * 10 years with 2 leap years.
46 #define DECADE (3600*24*(365*10+2))
49 const unsigned char *buf
,
52 /* The DDF metadata handling.
53 * DDF metadata lives at the end of the device.
54 * The last 512 byte block provides an 'anchor' which is used to locate
55 * the rest of the metadata which usually lives immediately behind the anchor.
58 * - all multibyte numeric fields are bigendian.
59 * - all strings are space padded.
63 /* Primary Raid Level (PRL) */
64 #define DDF_RAID0 0x00
65 #define DDF_RAID1 0x01
66 #define DDF_RAID3 0x03
67 #define DDF_RAID4 0x04
68 #define DDF_RAID5 0x05
69 #define DDF_RAID1E 0x11
71 #define DDF_CONCAT 0x1f
72 #define DDF_RAID5E 0x15
73 #define DDF_RAID5EE 0x25
74 #define DDF_RAID6 0x16 /* Vendor unique layout */
76 /* Raid Level Qualifier (RLQ) */
77 #define DDF_RAID0_SIMPLE 0x00
78 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
79 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
80 #define DDF_RAID3_0 0x00 /* parity in first extent */
81 #define DDF_RAID3_N 0x01 /* parity in last extent */
82 #define DDF_RAID4_0 0x00 /* parity in first extent */
83 #define DDF_RAID4_N 0x01 /* parity in last extent */
84 /* these apply to raid5e and raid5ee as well */
85 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
86 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
87 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
89 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
90 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
92 /* Secondary RAID Level (SRL) */
93 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
94 #define DDF_2MIRRORED 0x01
95 #define DDF_2CONCAT 0x02
96 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
99 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
100 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
101 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
102 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
103 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
104 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
105 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
106 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
107 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
108 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
110 #define DDF_GUID_LEN 24
111 #define DDF_REVISION "01.00.00"
116 char guid
[DDF_GUID_LEN
];
117 char revision
[8]; /* 01.00.00 */
118 __u32 seq
; /* starts at '1' */
123 __u8 pad0
; /* 0xff */
124 __u8 pad1
[12]; /* 12 * 0xff */
125 /* 64 bytes so far */
126 __u8 header_ext
[32]; /* reserved: fill with 0xff */
130 __u8 pad2
[3]; /* 0xff */
131 __u32 workspace_len
; /* sectors for vendor space -
132 * at least 32768(sectors) */
134 __u16 max_pd_entries
; /* one of 15, 63, 255, 1023, 4095 */
135 __u16 max_vd_entries
; /* 2^(4,6,8,10,12)-1 : i.e. as above */
136 __u16 max_partitions
; /* i.e. max num of configuration
137 record entries per disk */
138 __u16 config_record_len
; /* 1 +ROUNDUP(max_primary_element_entries
140 __u16 max_primary_element_entries
; /* 16, 64, 256, 1024, or 4096 */
141 __u8 pad3
[54]; /* 0xff */
142 /* 192 bytes so far */
143 __u32 controller_section_offset
;
144 __u32 controller_section_length
;
145 __u32 phys_section_offset
;
146 __u32 phys_section_length
;
147 __u32 virt_section_offset
;
148 __u32 virt_section_length
;
149 __u32 config_section_offset
;
150 __u32 config_section_length
;
151 __u32 data_section_offset
;
152 __u32 data_section_length
;
153 __u32 bbm_section_offset
;
154 __u32 bbm_section_length
;
155 __u32 diag_space_offset
;
156 __u32 diag_space_length
;
159 /* 256 bytes so far */
160 __u8 pad4
[256]; /* 0xff */
164 #define DDF_HEADER_ANCHOR 0x00
165 #define DDF_HEADER_PRIMARY 0x01
166 #define DDF_HEADER_SECONDARY 0x02
168 /* The content of the 'controller section' - global scope */
169 struct ddf_controller_data
{
172 char guid
[DDF_GUID_LEN
];
173 struct controller_type
{
180 __u8 pad
[8]; /* 0xff */
181 __u8 vendor_data
[448];
184 /* The content of phys_section - global scope */
191 struct phys_disk_entry
{
192 char guid
[DDF_GUID_LEN
];
196 __u64 config_size
; /* DDF structures must be after here */
197 char path
[18]; /* another horrible structure really */
202 /* phys_disk_entry.type is a bitmap - bigendian remember */
203 #define DDF_Forced_PD_GUID 1
204 #define DDF_Active_in_VD 2
205 #define DDF_Global_Spare 4
206 #define DDF_Spare 8 /* overrides Global_spare */
207 #define DDF_Foreign 16
208 #define DDF_Legacy 32 /* no DDF on this device */
210 #define DDF_Interface_mask 0xf00
211 #define DDF_Interface_SCSI 0x100
212 #define DDF_Interface_SAS 0x200
213 #define DDF_Interface_SATA 0x300
214 #define DDF_Interface_FC 0x400
216 /* phys_disk_entry.state is a bigendian bitmap */
218 #define DDF_Failed 2 /* overrides 1,4,8 */
219 #define DDF_Rebuilding 4
220 #define DDF_Transition 8
222 #define DDF_ReadErrors 32
223 #define DDF_Missing 64
225 /* The content of the virt_section global scope */
226 struct virtual_disk
{
229 __u16 populated_vdes
;
232 struct virtual_entry
{
233 char guid
[DDF_GUID_LEN
];
235 __u16 pad0
; /* 0xffff */
245 /* virtual_entry.type is a bitmap - bigendian */
247 #define DDF_Enforce_Groups 2
248 #define DDF_Unicode 4
249 #define DDF_Owner_Valid 8
251 /* virtual_entry.state is a bigendian bitmap */
252 #define DDF_state_mask 0x7
253 #define DDF_state_optimal 0x0
254 #define DDF_state_degraded 0x1
255 #define DDF_state_deleted 0x2
256 #define DDF_state_missing 0x3
257 #define DDF_state_failed 0x4
258 #define DDF_state_part_optimal 0x5
260 #define DDF_state_morphing 0x8
261 #define DDF_state_inconsistent 0x10
263 /* virtual_entry.init_state is a bigendian bitmap */
264 #define DDF_initstate_mask 0x03
265 #define DDF_init_not 0x00
266 #define DDF_init_quick 0x01 /* initialisation is progress.
267 * i.e. 'state_inconsistent' */
268 #define DDF_init_full 0x02
270 #define DDF_access_mask 0xc0
271 #define DDF_access_rw 0x00
272 #define DDF_access_ro 0x80
273 #define DDF_access_blocked 0xc0
275 /* The content of the config_section - local scope
276 * It has multiple records each config_record_len sectors
277 * They can be vd_config or spare_assign
283 char guid
[DDF_GUID_LEN
];
287 __u16 prim_elmnt_count
;
288 __u8 chunk_shift
; /* 0 == 512, 1==1024 etc */
291 __u8 sec_elmnt_count
;
294 __u64 blocks
; /* blocks per component could be different
295 * on different component devices...(only
296 * for concat I hope) */
297 __u64 array_blocks
; /* blocks in array */
305 __u8 v0
[32]; /* reserved- 0xff */
306 __u8 v1
[32]; /* reserved- 0xff */
307 __u8 v2
[16]; /* reserved- 0xff */
308 __u8 v3
[16]; /* reserved- 0xff */
310 __u32 phys_refnum
[0]; /* refnum of each disk in sequence */
311 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
312 bvd are always the same size */
315 /* vd_config.cache_pol[7] is a bitmap */
316 #define DDF_cache_writeback 1 /* else writethrough */
317 #define DDF_cache_wadaptive 2 /* only applies if writeback */
318 #define DDF_cache_readahead 4
319 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
320 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
321 #define DDF_cache_wallowed 32 /* enable write caching */
322 #define DDF_cache_rallowed 64 /* enable read caching */
324 struct spare_assign
{
330 __u16 populated
; /* SAEs used */
331 __u16 max
; /* max SAEs */
333 struct spare_assign_entry
{
334 char guid
[DDF_GUID_LEN
];
335 __u16 secondary_element
;
339 /* spare_assign.type is a bitmap */
340 #define DDF_spare_dedicated 0x1 /* else global */
341 #define DDF_spare_revertible 0x2 /* else committable */
342 #define DDF_spare_active 0x4 /* else not active */
343 #define DDF_spare_affinity 0x8 /* enclosure affinity */
345 /* The data_section contents - local scope */
349 char guid
[DDF_GUID_LEN
];
350 __u32 refnum
; /* crc of some magic drive data ... */
351 __u8 forced_ref
; /* set when above was not result of magic */
352 __u8 forced_guid
; /* set if guid was forced rather than magic */
357 /* bbm_section content */
358 struct bad_block_log
{
365 struct mapped_block
{
366 __u64 defective_start
;
367 __u32 replacement_start
;
373 /* Struct for internally holding ddf structures */
374 /* The DDF structure stored on each device is potentially
375 * quite different, as some data is global and some is local.
376 * The global data is:
379 * - Physical disk records
380 * - Virtual disk records
382 * - Configuration records
383 * - Physical Disk data section
384 * ( and Bad block and vendor which I don't care about yet).
386 * The local data is parsed into separate lists as it is read
387 * and reconstructed for writing. This means that we only need
388 * to make config changes once and they are automatically
389 * propagated to all devices.
390 * Note that the ddf_super has space of the conf and disk data
391 * for this disk and also for a list of all such data.
392 * The list is only used for the superblock that is being
393 * built in Create or Assemble to describe the whole array.
396 struct ddf_header anchor
, primary
, secondary
, *active
;
397 struct ddf_controller_data controller
;
398 struct phys_disk
*phys
;
399 struct virtual_disk
*virt
;
404 __u64
*lba_offset
; /* location in 'conf' of
406 struct vd_config conf
;
407 } *conflist
, *newconf
;
408 int conf_num
; /* Index into 'virt' of entry matching 'newconf' */
411 struct disk_data disk
;
415 struct vcl
*vlist
[0]; /* max_part+1 in size */
420 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
423 extern struct superswitch super_ddf_container
, super_ddf_bvd
, super_ddf
;
425 static int calc_crc(void *buf
, int len
)
427 /* crcs are always at the same place as in the ddf_header */
428 struct ddf_header
*ddf
= buf
;
429 __u32 oldcrc
= ddf
->crc
;
431 ddf
->crc
= 0xffffffff;
433 newcrc
= crc32(0, buf
, len
);
438 static int load_ddf_header(int fd
, unsigned long long lba
,
439 unsigned long long size
,
441 struct ddf_header
*hdr
, struct ddf_header
*anchor
)
443 /* read a ddf header (primary or secondary) from fd/lba
444 * and check that it is consistent with anchor
446 * magic, crc, guid, rev, and LBA's header_type, and
447 * everything after header_type must be the same
452 if (lseek64(fd
, lba
<<9, 0) < 0)
455 if (read(fd
, hdr
, 512) != 512)
458 if (hdr
->magic
!= DDF_HEADER_MAGIC
)
460 if (calc_crc(hdr
, 512) != hdr
->crc
)
462 if (memcmp(anchor
->guid
, hdr
->guid
, DDF_GUID_LEN
) != 0 ||
463 memcmp(anchor
->revision
, hdr
->revision
, 8) != 0 ||
464 anchor
->primary_lba
!= hdr
->primary_lba
||
465 anchor
->secondary_lba
!= hdr
->secondary_lba
||
467 memcmp(anchor
->pad2
, hdr
->pad2
, 512 -
468 offsetof(struct ddf_header
, pad2
)) != 0)
471 /* Looks good enough to me... */
475 static void *load_section(int fd
, struct ddf_super
*super
, void *buf
,
476 __u32 offset_be
, __u32 len_be
, int check
)
478 unsigned long long offset
= __be32_to_cpu(offset_be
);
479 unsigned long long len
= __be32_to_cpu(len_be
);
480 int dofree
= (buf
== NULL
);
483 if (len
!= 2 && len
!= 8 && len
!= 32
484 && len
!= 128 && len
!= 512)
490 /* All pre-allocated sections are a single block */
494 buf
= malloc(len
<<9);
498 if (super
->active
->type
== 1)
499 offset
+= __be64_to_cpu(super
->active
->primary_lba
);
501 offset
+= __be64_to_cpu(super
->active
->secondary_lba
);
503 if (lseek64(fd
, offset
<<9, 0) != (offset
<<9)) {
508 if (read(fd
, buf
, len
<<9) != (len
<<9)) {
516 static int load_ddf_headers(int fd
, struct ddf_super
*super
, char *devname
)
518 unsigned long long dsize
;
520 get_dev_size(fd
, NULL
, &dsize
);
522 if (lseek64(fd
, dsize
-512, 0) < 0) {
525 Name
": Cannot seek to anchor block on %s: %s\n",
526 devname
, strerror(errno
));
529 if (read(fd
, &super
->anchor
, 512) != 512) {
532 Name
": Cannot read anchor block on %s: %s\n",
533 devname
, strerror(errno
));
536 if (super
->anchor
.magic
!= DDF_HEADER_MAGIC
) {
538 fprintf(stderr
, Name
": no DDF anchor found on %s\n",
542 if (calc_crc(&super
->anchor
, 512) != super
->anchor
.crc
) {
544 fprintf(stderr
, Name
": bad CRC on anchor on %s\n",
548 if (memcmp(super
->anchor
.revision
, DDF_REVISION
, 8) != 0) {
550 fprintf(stderr
, Name
": can only support super revision"
551 " %.8s, not %.8s on %s\n",
552 DDF_REVISION
, super
->anchor
.revision
, devname
);
555 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.primary_lba
),
557 &super
->primary
, &super
->anchor
) == 0) {
560 Name
": Failed to load primary DDF header "
564 super
->active
= &super
->primary
;
565 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.secondary_lba
),
567 &super
->secondary
, &super
->anchor
)) {
568 if ((__be32_to_cpu(super
->primary
.seq
)
569 < __be32_to_cpu(super
->secondary
.seq
) &&
570 !super
->secondary
.openflag
)
571 || (__be32_to_cpu(super
->primary
.seq
)
572 == __be32_to_cpu(super
->secondary
.seq
) &&
573 super
->primary
.openflag
&& !super
->secondary
.openflag
)
575 super
->active
= &super
->secondary
;
580 static int load_ddf_global(int fd
, struct ddf_super
*super
, char *devname
)
583 ok
= load_section(fd
, super
, &super
->controller
,
584 super
->active
->controller_section_offset
,
585 super
->active
->controller_section_length
,
587 super
->phys
= load_section(fd
, super
, NULL
,
588 super
->active
->phys_section_offset
,
589 super
->active
->phys_section_length
,
591 super
->pdsize
= __be32_to_cpu(super
->active
->phys_section_length
) * 512;
593 super
->virt
= load_section(fd
, super
, NULL
,
594 super
->active
->virt_section_offset
,
595 super
->active
->virt_section_length
,
597 super
->vdsize
= __be32_to_cpu(super
->active
->virt_section_length
) * 512;
607 super
->conflist
= NULL
;
612 static int load_ddf_local(int fd
, struct ddf_super
*super
,
613 char *devname
, int keep
)
622 /* First the local disk info */
623 super
->max_part
= __be16_to_cpu(super
->active
->max_partitions
);
624 dl
= malloc(sizeof(*dl
) +
625 (super
->max_part
+1) * sizeof(dl
->vlist
[0]));
627 load_section(fd
, super
, &dl
->disk
,
628 super
->active
->data_section_offset
,
629 super
->active
->data_section_length
,
631 dl
->devname
= devname
? strdup(devname
) : NULL
;
634 dl
->major
= major(stb
.st_rdev
);
635 dl
->minor
= minor(stb
.st_rdev
);
636 dl
->next
= super
->dlist
;
637 dl
->fd
= keep
? fd
: -1;
638 for (i
=0 ; i
< super
->max_part
+ 1 ; i
++)
642 /* Now the config list. */
643 /* 'conf' is an array of config entries, some of which are
644 * probably invalid. Those which are good need to be copied into
647 conflen
= __be16_to_cpu(super
->active
->config_record_len
);
649 conf
= load_section(fd
, super
, NULL
,
650 super
->active
->config_section_offset
,
651 super
->active
->config_section_length
,
655 i
< __be32_to_cpu(super
->active
->config_section_length
);
657 struct vd_config
*vd
=
658 (struct vd_config
*)((char*)conf
+ i
*512);
661 if (vd
->magic
!= DDF_VD_CONF_MAGIC
)
663 for (vcl
= super
->conflist
; vcl
; vcl
= vcl
->next
) {
664 if (memcmp(vcl
->conf
.guid
,
665 vd
->guid
, DDF_GUID_LEN
) == 0)
670 dl
->vlist
[i
/conflen
] = vcl
;
671 if (__be32_to_cpu(vd
->seqnum
) <=
672 __be32_to_cpu(vcl
->conf
.seqnum
))
675 vcl
= malloc(conflen
*512 + offsetof(struct vcl
, conf
));
676 vcl
->next
= super
->conflist
;
677 super
->conflist
= vcl
;
679 memcpy(&vcl
->conf
, vd
, conflen
*512);
680 mppe
= __be16_to_cpu(super
->anchor
.max_primary_element_entries
);
681 vcl
->lba_offset
= (__u64
*)
682 &vcl
->conf
.phys_refnum
[mppe
];
683 dl
->vlist
[i
/conflen
] = vcl
;
691 static int load_super_ddf_all(struct supertype
*st
, int fd
,
692 void **sbp
, char *devname
, int keep_fd
);
694 static int load_super_ddf(struct supertype
*st
, int fd
,
697 unsigned long long dsize
;
698 struct ddf_super
*super
;
702 if (load_super_ddf_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
706 if (get_dev_size(fd
, devname
, &dsize
) == 0)
709 /* 32M is a lower bound */
710 if (dsize
<= 32*1024*1024) {
713 Name
": %s is too small for ddf: "
714 "size is %llu sectors.\n",
722 Name
": %s is an odd size for ddf: "
723 "size is %llu bytes.\n",
729 super
= malloc(sizeof(*super
));
731 fprintf(stderr
, Name
": malloc of %zu failed.\n",
735 memset(super
, 0, sizeof(*super
));
737 rv
= load_ddf_headers(fd
, super
, devname
);
743 /* Have valid headers and have chosen the best. Let's read in the rest*/
745 rv
= load_ddf_global(fd
, super
, devname
);
750 Name
": Failed to load all information "
751 "sections on %s\n", devname
);
756 load_ddf_local(fd
, super
, devname
, 0);
758 /* Should possibly check the sections .... */
761 if (st
->ss
== NULL
) {
763 st
->minor_version
= 0;
770 static void free_super_ddf(struct supertype
*st
)
772 struct ddf_super
*ddf
= st
->sb
;
777 while (ddf
->conflist
) {
778 struct vcl
*v
= ddf
->conflist
;
779 ddf
->conflist
= v
->next
;
783 struct dl
*d
= ddf
->dlist
;
784 ddf
->dlist
= d
->next
;
793 static struct supertype
*match_metadata_desc_ddf(char *arg
)
795 /* 'ddf' only support containers */
796 struct supertype
*st
;
797 if (strcmp(arg
, "ddf") != 0 &&
798 strcmp(arg
, "default") != 0
802 st
= malloc(sizeof(*st
));
805 st
->minor_version
= 0;
810 static struct supertype
*match_metadata_desc_ddf_bvd(char *arg
)
812 struct supertype
*st
;
813 if (strcmp(arg
, "ddf/bvd") != 0 &&
814 strcmp(arg
, "bvd") != 0 &&
815 strcmp(arg
, "default") != 0
819 st
= malloc(sizeof(*st
));
820 st
->ss
= &super_ddf_bvd
;
822 st
->minor_version
= 0;
826 static struct supertype
*match_metadata_desc_ddf_svd(char *arg
)
828 struct supertype
*st
;
829 if (strcmp(arg
, "ddf/svd") != 0 &&
830 strcmp(arg
, "svd") != 0 &&
831 strcmp(arg
, "default") != 0
835 st
= malloc(sizeof(*st
));
836 st
->ss
= &super_ddf_svd
;
838 st
->minor_version
= 0;
845 static mapping_t ddf_state
[] = {
851 { "Partially Optimal", 5},
857 static mapping_t ddf_init_state
[] = {
858 { "Not Initialised", 0},
859 { "QuickInit in Progress", 1},
860 { "Fully Initialised", 2},
864 static mapping_t ddf_access
[] = {
868 { "Blocked (no access)", 3},
872 static mapping_t ddf_level
[] = {
873 { "RAID0", DDF_RAID0
},
874 { "RAID1", DDF_RAID1
},
875 { "RAID3", DDF_RAID3
},
876 { "RAID4", DDF_RAID4
},
877 { "RAID5", DDF_RAID5
},
878 { "RAID1E",DDF_RAID1E
},
880 { "CONCAT",DDF_CONCAT
},
881 { "RAID5E",DDF_RAID5E
},
882 { "RAID5EE",DDF_RAID5EE
},
883 { "RAID6", DDF_RAID6
},
886 static mapping_t ddf_sec_level
[] = {
887 { "Striped", DDF_2STRIPED
},
888 { "Mirrored", DDF_2MIRRORED
},
889 { "Concat", DDF_2CONCAT
},
890 { "Spanned", DDF_2SPANNED
},
898 static struct num_mapping ddf_level_num
[] = {
901 { DDF_RAID3
, LEVEL_UNSUPPORTED
},
904 { DDF_RAID1E
, LEVEL_UNSUPPORTED
},
905 { DDF_JBOD
, LEVEL_UNSUPPORTED
},
906 { DDF_CONCAT
, LEVEL_LINEAR
},
907 { DDF_RAID5E
, LEVEL_UNSUPPORTED
},
908 { DDF_RAID5EE
, LEVEL_UNSUPPORTED
},
913 static int map_num1(struct num_mapping
*map
, int num
)
916 for (i
=0 ; map
[i
].num1
!= MAXINT
; i
++)
917 if (map
[i
].num1
== num
)
923 static void print_guid(char *guid
, int tstamp
)
925 /* A GUIDs are part (or all) ASCII and part binary.
926 * They tend to be space padded.
927 * We ignore trailing spaces and print numbers
928 * <0x20 and >=0x7f as \xXX
929 * Some GUIDs have a time stamp in bytes 16-19.
930 * We print that if appropriate
932 int l
= DDF_GUID_LEN
;
934 while (l
&& guid
[l
-1] == ' ')
936 for (i
=0 ; i
<l
; i
++) {
937 if (guid
[i
] >= 0x20 && guid
[i
] < 0x7f)
938 fputc(guid
[i
], stdout
);
940 fprintf(stdout
, "\\x%02x", guid
[i
]&255);
943 time_t then
= __be32_to_cpu(*(__u32
*)(guid
+16)) + DECADE
;
946 tm
= localtime(&then
);
947 strftime(tbuf
, 100, " (%D %T)",tm
);
952 static void examine_vd(int n
, struct ddf_super
*sb
, char *guid
)
954 int crl
= __be16_to_cpu(sb
->anchor
.config_record_len
);
957 for (vcl
= sb
->conflist
; vcl
; vcl
= vcl
->next
) {
958 struct vd_config
*vc
= &vcl
->conf
;
960 if (calc_crc(vc
, crl
*512) != vc
->crc
)
962 if (memcmp(vc
->guid
, guid
, DDF_GUID_LEN
) != 0)
965 /* Ok, we know about this VD, let's give more details */
966 printf(" Raid Devices[%d] : %d\n", n
,
967 __be16_to_cpu(vc
->prim_elmnt_count
));
968 printf(" Chunk Size[%d] : %d sectors\n", n
,
969 1 << vc
->chunk_shift
);
970 printf(" Raid Level[%d] : %s\n", n
,
971 map_num(ddf_level
, vc
->prl
)?:"-unknown-");
972 if (vc
->sec_elmnt_count
!= 1) {
973 printf(" Secondary Position[%d] : %d of %d\n", n
,
974 vc
->sec_elmnt_seq
, vc
->sec_elmnt_count
);
975 printf(" Secondary Level[%d] : %s\n", n
,
976 map_num(ddf_sec_level
, vc
->srl
) ?: "-unknown-");
978 printf(" Device Size[%d] : %llu\n", n
,
979 __be64_to_cpu(vc
->blocks
)/2);
980 printf(" Array Size[%d] : %llu\n", n
,
981 __be64_to_cpu(vc
->array_blocks
)/2);
985 static void examine_vds(struct ddf_super
*sb
)
987 int cnt
= __be16_to_cpu(sb
->virt
->populated_vdes
);
989 printf(" Virtual Disks : %d\n", cnt
);
991 for (i
=0; i
<cnt
; i
++) {
992 struct virtual_entry
*ve
= &sb
->virt
->entries
[i
];
993 printf(" VD GUID[%d] : ", i
); print_guid(ve
->guid
, 1);
995 printf(" unit[%d] : %d\n", i
, __be16_to_cpu(ve
->unit
));
996 printf(" state[%d] : %s, %s%s\n", i
,
997 map_num(ddf_state
, ve
->state
& 7),
998 (ve
->state
& 8) ? "Morphing, ": "",
999 (ve
->state
& 16)? "Not Consistent" : "Consistent");
1000 printf(" init state[%d] : %s\n", i
,
1001 map_num(ddf_init_state
, ve
->init_state
&3));
1002 printf(" access[%d] : %s\n", i
,
1003 map_num(ddf_access
, (ve
->init_state
>>6) & 3));
1004 printf(" Name[%d] : %.16s\n", i
, ve
->name
);
1005 examine_vd(i
, sb
, ve
->guid
);
1007 if (cnt
) printf("\n");
1010 static void examine_pds(struct ddf_super
*sb
)
1012 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1015 printf(" Physical Disks : %d\n", cnt
);
1017 for (i
=0 ; i
<cnt
; i
++) {
1018 struct phys_disk_entry
*pd
= &sb
->phys
->entries
[i
];
1019 int type
= __be16_to_cpu(pd
->type
);
1020 int state
= __be16_to_cpu(pd
->state
);
1022 printf(" PD GUID[%d] : ", i
); print_guid(pd
->guid
, 0);
1024 printf(" ref[%d] : %08x\n", i
,
1025 __be32_to_cpu(pd
->refnum
));
1026 printf(" mode[%d] : %s%s%s%s%s\n", i
,
1027 (type
&2) ? "active":"",
1028 (type
&4) ? "Global Spare":"",
1029 (type
&8) ? "spare" : "",
1030 (type
&16)? ", foreign" : "",
1031 (type
&32)? "pass-through" : "");
1032 printf(" state[%d] : %s%s%s%s%s%s%s\n", i
,
1033 (state
&1)? "Online": "Offline",
1034 (state
&2)? ", Failed": "",
1035 (state
&4)? ", Rebuilding": "",
1036 (state
&8)? ", in-transition": "",
1037 (state
&16)? ", SMART errors": "",
1038 (state
&32)? ", Unrecovered Read Errors": "",
1039 (state
&64)? ", Missing" : "");
1040 printf(" Avail Size[%d] : %llu K\n", i
,
1041 __be64_to_cpu(pd
->config_size
)>>1);
1042 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1043 if (dl
->disk
.refnum
== pd
->refnum
) {
1044 char *dv
= map_dev(dl
->major
, dl
->minor
, 0);
1046 printf(" Device[%d] : %s\n",
1054 static void examine_super_ddf(struct supertype
*st
, char *homehost
)
1056 struct ddf_super
*sb
= st
->sb
;
1058 printf(" Magic : %08x\n", __be32_to_cpu(sb
->anchor
.magic
));
1059 printf(" Version : %.8s\n", sb
->anchor
.revision
);
1060 printf("Controller GUID : "); print_guid(sb
->controller
.guid
, 0);
1062 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1064 printf(" Seq : %08x\n", __be32_to_cpu(sb
->active
->seq
));
1065 printf(" Redundant hdr : %s\n", sb
->secondary
.magic
== DDF_HEADER_MAGIC
1071 static void brief_examine_super_ddf(struct supertype
*st
)
1073 /* We just write a generic DDF ARRAY entry
1074 * The uuid is all hex, 6 groups of 4 bytes
1076 struct ddf_super
*ddf
= st
->sb
;
1078 printf("ARRAY /dev/ddf UUID=");
1079 for (i
= 0; i
< DDF_GUID_LEN
; i
++) {
1080 printf("%02x", ddf
->anchor
.guid
[i
]);
1081 if ((i
&3) == 0 && i
!= 0)
1087 static void detail_super_ddf(struct supertype
*st
, char *homehost
)
1090 * Could print DDF GUID
1091 * Need to find which array
1092 * If whole, briefly list all arrays
1097 static void brief_detail_super_ddf(struct supertype
*st
)
1099 /* FIXME I really need to know which array we are detailing.
1100 * Can that be stored in ddf_super??
1102 // struct ddf_super *ddf = st->sb;
1108 static int match_home_ddf(struct supertype
*st
, char *homehost
)
1110 /* It matches 'this' host if the controller is a
1111 * Linux-MD controller with vendor_data matching
1114 struct ddf_super
*ddf
= st
->sb
;
1115 int len
= strlen(homehost
);
1117 return (memcmp(ddf
->controller
.guid
, T10
, 8) == 0 &&
1118 len
< sizeof(ddf
->controller
.vendor_data
) &&
1119 memcmp(ddf
->controller
.vendor_data
, homehost
,len
) == 0 &&
1120 ddf
->controller
.vendor_data
[len
] == 0);
1123 static struct vd_config
*find_vdcr(struct ddf_super
*ddf
, int inst
)
1126 if (inst
< 0 || inst
> __be16_to_cpu(ddf
->virt
->populated_vdes
))
1128 for (v
= ddf
->conflist
; v
; v
= v
->next
)
1129 if (memcmp(v
->conf
.guid
,
1130 ddf
->virt
->entries
[inst
].guid
,
1136 static int find_phys(struct ddf_super
*ddf
, __u32 phys_refnum
)
1138 /* Find the entry in phys_disk which has the given refnum
1139 * and return it's index
1142 for (i
=0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
); i
++)
1143 if (ddf
->phys
->entries
[i
].refnum
== phys_refnum
)
1148 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4])
1150 /* The uuid returned here is used for:
1151 * uuid to put into bitmap file (Create, Grow)
1152 * uuid for backup header when saving critical section (Grow)
1153 * comparing uuids when re-adding a device into an array
1154 * For each of these we can make do with a truncated
1155 * or hashed uuid rather than the original, as long as
1157 * In each case the uuid required is that of the data-array,
1158 * not the device-set.
1159 * In the case of SVD we assume the BVD is of interest,
1160 * though that might be the case if a bitmap were made for
1161 * a mirrored SVD - worry about that later.
1162 * So we need to find the VD configuration record for the
1163 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1164 * The first 16 bytes of the sha1 of these is used.
1166 struct ddf_super
*ddf
= st
->sb
;
1167 struct vd_config
*vd
= find_vdcr(ddf
, ddf
->conf_num
);
1170 memset(uuid
, 0, sizeof (uuid
));
1173 struct sha1_ctx ctx
;
1174 sha1_init_ctx(&ctx
);
1175 sha1_process_bytes(&vd
->guid
, DDF_GUID_LEN
, &ctx
);
1176 if (vd
->sec_elmnt_count
> 1)
1177 sha1_process_bytes(&vd
->sec_elmnt_seq
, 1, &ctx
);
1178 sha1_finish_ctx(&ctx
, buf
);
1179 memcpy(uuid
, buf
, sizeof(uuid
));
1183 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
)
1185 struct ddf_super
*ddf
= st
->sb
;
1188 info
->array
.major_version
= 1000;
1189 info
->array
.minor_version
= 0; /* FIXME use ddf->revision somehow */
1190 info
->array
.patch_version
= 0;
1191 info
->array
.raid_disks
= __be16_to_cpu(ddf
->phys
->used_pdes
);
1192 info
->array
.level
= LEVEL_CONTAINER
;
1193 info
->array
.layout
= 0;
1194 info
->array
.md_minor
= -1;
1195 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*(__u32
*)
1196 (ddf
->anchor
.guid
+16));
1197 info
->array
.utime
= 0;
1198 info
->array
.chunk_size
= 0;
1200 // info->data_offset = ???;
1201 // info->component_size = ???;
1203 info
->disk
.major
= 0;
1204 info
->disk
.minor
= 0;
1206 info
->disk
.number
= __be32_to_cpu(ddf
->dlist
->disk
.refnum
);
1207 info
->disk
.raid_disk
= -1;
1208 for (i
= 0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
) ; i
++)
1209 if (ddf
->phys
->entries
[i
].refnum
==
1210 ddf
->dlist
->disk
.refnum
) {
1211 info
->disk
.raid_disk
= i
;
1215 info
->disk
.number
= -1;
1216 // info->disk.raid_disk = find refnum in the table and use index;
1218 info
->disk
.state
= (1 << MD_DISK_SYNC
);
1220 info
->reshape_active
= 0;
1222 strcpy(info
->text_version
, "ddf");
1224 // uuid_from_super_ddf(info->uuid, sbv);
1226 // info->name[] ?? ;
1229 static void getinfo_super_n_container(struct supertype
*st
, struct mdinfo
*info
)
1231 /* just need offset and size */
1232 struct ddf_super
*ddf
= st
->sb
;
1233 int n
= info
->disk
.number
;
1235 info
->data_offset
= __be64_to_cpu(ddf
->phys
->entries
[n
].config_size
);
1236 info
->component_size
= 32*1024*1024 / 512;
1239 static int rlq_to_layout(int rlq
, int prl
, int raiddisks
);
1241 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
)
1243 struct ddf_super
*ddf
= st
->sb
;
1244 struct vd_config
*vd
= find_vdcr(ddf
, info
->container_member
);
1246 /* FIXME this returns BVD info - what if we want SVD ?? */
1248 info
->array
.major_version
= 1000;
1249 info
->array
.minor_version
= 0; /* FIXME use ddf->revision somehow */
1250 info
->array
.patch_version
= 0;
1251 info
->array
.raid_disks
= __be16_to_cpu(vd
->prim_elmnt_count
);
1252 info
->array
.level
= map_num1(ddf_level_num
, vd
->prl
);
1253 info
->array
.layout
= rlq_to_layout(vd
->rlq
, vd
->prl
,
1254 info
->array
.raid_disks
);
1255 info
->array
.md_minor
= -1;
1256 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*(__u32
*)(vd
->guid
+16));
1257 info
->array
.utime
= DECADE
+ __be32_to_cpu(vd
->timestamp
);
1258 info
->array
.chunk_size
= 512 << vd
->chunk_shift
;
1260 // info->data_offset = ???;
1261 // info->component_size = ???;
1263 info
->disk
.major
= 0;
1264 info
->disk
.minor
= 0;
1265 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1266 // info->disk.raid_disk = find refnum in the table and use index;
1267 // info->disk.state = ???;
1269 uuid_from_super_ddf(st
, info
->uuid
);
1271 sprintf(info
->text_version
, "/%s/%d",
1272 devnum2devname(st
->container_dev
),
1273 info
->container_member
);
1275 // info->name[] ?? ;
1278 static void getinfo_super_n_bvd(struct supertype
*st
, struct mdinfo
*info
)
1280 /* Find the particular details for info->disk.raid_disk.
1281 * This includes data_offset, component_size,
1283 struct ddf_super
*ddf
= st
->sb
;
1284 __u64
*lba_offset
= ddf
->newconf
->lba_offset
;
1285 struct vd_config
*conf
= &ddf
->newconf
->conf
;
1286 info
->data_offset
= __be64_to_cpu(lba_offset
[info
->disk
.raid_disk
]);
1287 info
->component_size
= __be64_to_cpu(conf
->blocks
);
1290 static int update_super_ddf(struct supertype
*st
, struct mdinfo
*info
,
1292 char *devname
, int verbose
,
1293 int uuid_set
, char *homehost
)
1295 /* For 'assemble' and 'force' we need to return non-zero if any
1296 * change was made. For others, the return value is ignored.
1297 * Update options are:
1298 * force-one : This device looks a bit old but needs to be included,
1299 * update age info appropriately.
1300 * assemble: clear any 'faulty' flag to allow this device to
1302 * force-array: Array is degraded but being forced, mark it clean
1303 * if that will be needed to assemble it.
1305 * newdev: not used ????
1306 * grow: Array has gained a new device - this is currently for
1308 * resync: mark as dirty so a resync will happen.
1309 * uuid: Change the uuid of the array to match watch is given
1310 * homehost: update the recorded homehost
1311 * name: update the name - preserving the homehost
1312 * _reshape_progress: record new reshape_progress position.
1314 * Following are not relevant for this version:
1315 * sparc2.2 : update from old dodgey metadata
1316 * super-minor: change the preferred_minor number
1317 * summaries: update redundant counters.
1320 // struct ddf_super *ddf = st->sb;
1321 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1322 // struct virtual_entry *ve = find_ve(ddf);
1325 /* we don't need to handle "force-*" or "assemble" as
1326 * there is no need to 'trick' the kernel. We the metadata is
1327 * first updated to activate the array, all the implied modifications
1331 if (strcmp(update
, "grow") == 0) {
1334 if (strcmp(update
, "resync") == 0) {
1335 // info->resync_checkpoint = 0;
1337 /* We ignore UUID updates as they make even less sense
1340 if (strcmp(update
, "homehost") == 0) {
1341 /* homehost is stored in controller->vendor_data,
1342 * or it is when we are the vendor
1344 // if (info->vendor_is_local)
1345 // strcpy(ddf->controller.vendor_data, homehost);
1347 if (strcmp(update
, "name") == 0) {
1348 /* name is stored in virtual_entry->name */
1349 // memset(ve->name, ' ', 16);
1350 // strncpy(ve->name, info->name, 16);
1352 if (strcmp(update
, "_reshape_progress") == 0) {
1353 /* We don't support reshape yet */
1356 // update_all_csum(ddf);
1361 static void make_header_guid(char *guid
)
1365 /* Create a DDF Header of Virtual Disk GUID */
1367 /* 24 bytes of fiction required.
1368 * first 8 are a 'vendor-id' - "Linux-MD"
1369 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1370 * Remaining 8 random number plus timestamp
1372 memcpy(guid
, T10
, sizeof(T10
));
1373 stamp
= __cpu_to_be32(0xdeadbeef);
1374 memcpy(guid
+8, &stamp
, 4);
1375 stamp
= __cpu_to_be32(0);
1376 memcpy(guid
+12, &stamp
, 4);
1377 stamp
= __cpu_to_be32(time(0) - DECADE
);
1378 memcpy(guid
+16, &stamp
, 4);
1379 rfd
= open("/dev/urandom", O_RDONLY
);
1380 if (rfd
< 0 || read(rfd
, &stamp
, 4) != 4)
1382 memcpy(guid
+20, &stamp
, 4);
1383 if (rfd
>= 0) close(rfd
);
1385 static int init_super_ddf(struct supertype
*st
,
1386 mdu_array_info_t
*info
,
1387 unsigned long long size
, char *name
, char *homehost
,
1390 /* This is primarily called by Create when creating a new array.
1391 * We will then get add_to_super called for each component, and then
1392 * write_init_super called to write it out to each device.
1393 * For DDF, Create can create on fresh devices or on a pre-existing
1395 * To create on a pre-existing array a different method will be called.
1396 * This one is just for fresh drives.
1398 * We need to create the entire 'ddf' structure which includes:
1399 * DDF headers - these are easy.
1400 * Controller data - a Sector describing this controller .. not that
1401 * this is a controller exactly.
1402 * Physical Disk Record - one entry per device, so
1403 * leave plenty of space.
1404 * Virtual Disk Records - again, just leave plenty of space.
1405 * This just lists VDs, doesn't give details
1406 * Config records - describes the VDs that use this disk
1407 * DiskData - describes 'this' device.
1408 * BadBlockManagement - empty
1409 * Diag Space - empty
1410 * Vendor Logs - Could we put bitmaps here?
1413 struct ddf_super
*ddf
;
1416 int max_phys_disks
, max_virt_disks
;
1417 unsigned long long sector
;
1421 struct phys_disk
*pd
;
1422 struct virtual_disk
*vd
;
1424 ddf
= malloc(sizeof(*ddf
));
1425 ddf
->dlist
= NULL
; /* no physical disks yet */
1426 ddf
->conflist
= NULL
; /* No virtual disks yet */
1428 /* At least 32MB *must* be reserved for the ddf. So let's just
1429 * start 32MB from the end, and put the primary header there.
1430 * Don't do secondary for now.
1431 * We don't know exactly where that will be yet as it could be
1432 * different on each device. To just set up the lengths.
1436 ddf
->anchor
.magic
= DDF_HEADER_MAGIC
;
1437 make_header_guid(ddf
->anchor
.guid
);
1439 memcpy(ddf
->anchor
.revision
, DDF_REVISION
, 8);
1440 ddf
->anchor
.seq
= __cpu_to_be32(1);
1441 ddf
->anchor
.timestamp
= __cpu_to_be32(time(0) - DECADE
);
1442 ddf
->anchor
.openflag
= 0xFF;
1443 ddf
->anchor
.foreignflag
= 0;
1444 ddf
->anchor
.enforcegroups
= 0; /* Is this best?? */
1445 ddf
->anchor
.pad0
= 0xff;
1446 memset(ddf
->anchor
.pad1
, 0xff, 12);
1447 memset(ddf
->anchor
.header_ext
, 0xff, 32);
1448 ddf
->anchor
.primary_lba
= ~(__u64
)0;
1449 ddf
->anchor
.secondary_lba
= ~(__u64
)0;
1450 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
1451 memset(ddf
->anchor
.pad2
, 0xff, 3);
1452 ddf
->anchor
.workspace_len
= __cpu_to_be32(32768); /* Must be reserved */
1453 ddf
->anchor
.workspace_lba
= ~(__u64
)0; /* Put this at bottom
1454 of 32M reserved.. */
1455 max_phys_disks
= 1023; /* Should be enough */
1456 ddf
->anchor
.max_pd_entries
= __cpu_to_be16(max_phys_disks
);
1457 max_virt_disks
= 255;
1458 ddf
->anchor
.max_vd_entries
= __cpu_to_be16(max_virt_disks
); /* ?? */
1459 ddf
->anchor
.max_partitions
= __cpu_to_be16(64); /* ?? */
1461 ddf
->anchor
.config_record_len
= __cpu_to_be16(1 + 256*12/512);
1462 ddf
->anchor
.max_primary_element_entries
= __cpu_to_be16(256);
1463 memset(ddf
->anchor
.pad3
, 0xff, 54);
1465 /* controller sections is one sector long immediately
1466 * after the ddf header */
1468 ddf
->anchor
.controller_section_offset
= __cpu_to_be32(sector
);
1469 ddf
->anchor
.controller_section_length
= __cpu_to_be32(1);
1472 /* phys is 8 sectors after that */
1473 pdsize
= ROUND_UP(sizeof(struct phys_disk
) +
1474 sizeof(struct phys_disk_entry
)*max_phys_disks
,
1476 switch(pdsize
/512) {
1477 case 2: case 8: case 32: case 128: case 512: break;
1480 ddf
->anchor
.phys_section_offset
= __cpu_to_be32(sector
);
1481 ddf
->anchor
.phys_section_length
=
1482 __cpu_to_be32(pdsize
/512); /* max_primary_element_entries/8 */
1483 sector
+= pdsize
/512;
1485 /* virt is another 32 sectors */
1486 vdsize
= ROUND_UP(sizeof(struct virtual_disk
) +
1487 sizeof(struct virtual_entry
) * max_virt_disks
,
1489 switch(vdsize
/512) {
1490 case 2: case 8: case 32: case 128: case 512: break;
1493 ddf
->anchor
.virt_section_offset
= __cpu_to_be32(sector
);
1494 ddf
->anchor
.virt_section_length
=
1495 __cpu_to_be32(vdsize
/512); /* max_vd_entries/8 */
1496 sector
+= vdsize
/512;
1498 clen
= (1 + 256*12/512) * (64+1);
1499 ddf
->anchor
.config_section_offset
= __cpu_to_be32(sector
);
1500 ddf
->anchor
.config_section_length
= __cpu_to_be32(clen
);
1503 ddf
->anchor
.data_section_offset
= __cpu_to_be32(sector
);
1504 ddf
->anchor
.data_section_length
= __cpu_to_be32(1);
1507 ddf
->anchor
.bbm_section_length
= __cpu_to_be32(0);
1508 ddf
->anchor
.bbm_section_offset
= __cpu_to_be32(0xFFFFFFFF);
1509 ddf
->anchor
.diag_space_length
= __cpu_to_be32(0);
1510 ddf
->anchor
.diag_space_offset
= __cpu_to_be32(0xFFFFFFFF);
1511 ddf
->anchor
.vendor_length
= __cpu_to_be32(0);
1512 ddf
->anchor
.vendor_offset
= __cpu_to_be32(0xFFFFFFFF);
1514 memset(ddf
->anchor
.pad4
, 0xff, 256);
1516 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
1517 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
1519 ddf
->primary
.openflag
= 1; /* I guess.. */
1520 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
1522 ddf
->secondary
.openflag
= 1; /* I guess.. */
1523 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
1525 ddf
->active
= &ddf
->primary
;
1527 ddf
->controller
.magic
= DDF_CONTROLLER_MAGIC
;
1529 /* 24 more bytes of fiction required.
1530 * first 8 are a 'vendor-id' - "Linux-MD"
1531 * Remaining 16 are serial number.... maybe a hostname would do?
1533 memcpy(ddf
->controller
.guid
, T10
, sizeof(T10
));
1534 gethostname(hostname
, 17);
1536 hostlen
= strlen(hostname
);
1537 memcpy(ddf
->controller
.guid
+ 24 - hostlen
, hostname
, hostlen
);
1538 for (i
= strlen(T10
) ; i
+hostlen
< 24; i
++)
1539 ddf
->controller
.guid
[i
] = ' ';
1541 ddf
->controller
.type
.vendor_id
= __cpu_to_be16(0xDEAD);
1542 ddf
->controller
.type
.device_id
= __cpu_to_be16(0xBEEF);
1543 ddf
->controller
.type
.sub_vendor_id
= 0;
1544 ddf
->controller
.type
.sub_device_id
= 0;
1545 memcpy(ddf
->controller
.product_id
, "What Is My PID??", 16);
1546 memset(ddf
->controller
.pad
, 0xff, 8);
1547 memset(ddf
->controller
.vendor_data
, 0xff, 448);
1549 pd
= ddf
->phys
= malloc(pdsize
);
1550 ddf
->pdsize
= pdsize
;
1552 memset(pd
, 0xff, pdsize
);
1553 memset(pd
, 0, sizeof(*pd
));
1554 pd
->magic
= DDF_PHYS_DATA_MAGIC
;
1555 pd
->used_pdes
= __cpu_to_be16(0);
1556 pd
->max_pdes
= __cpu_to_be16(max_phys_disks
);
1557 memset(pd
->pad
, 0xff, 52);
1559 vd
= ddf
->virt
= malloc(vdsize
);
1560 ddf
->vdsize
= vdsize
;
1561 memset(vd
, 0, vdsize
);
1562 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
1563 vd
->populated_vdes
= __cpu_to_be16(0);
1564 vd
->max_vdes
= __cpu_to_be16(max_virt_disks
);
1565 memset(vd
->pad
, 0xff, 52);
1567 for (i
=0; i
<max_virt_disks
; i
++)
1568 memset(&vd
->entries
[i
], 0xff, sizeof(struct virtual_entry
));
1574 static int all_ff(char *guid
)
1577 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
1578 if (guid
[i
] != (char)0xff)
1582 static int chunk_to_shift(int chunksize
)
1584 return ffs(chunksize
/512)-1;
1587 static int level_to_prl(int level
)
1590 case LEVEL_LINEAR
: return DDF_CONCAT
;
1591 case 0: return DDF_RAID0
;
1592 case 1: return DDF_RAID1
;
1593 case 4: return DDF_RAID4
;
1594 case 5: return DDF_RAID5
;
1595 case 6: return DDF_RAID6
;
1599 static int layout_to_rlq(int level
, int layout
, int raiddisks
)
1603 return DDF_RAID0_SIMPLE
;
1606 case 2: return DDF_RAID1_SIMPLE
;
1607 case 3: return DDF_RAID1_MULTI
;
1612 case 0: return DDF_RAID4_N
;
1618 case ALGORITHM_LEFT_ASYMMETRIC
:
1619 return DDF_RAID5_N_RESTART
;
1620 case ALGORITHM_RIGHT_ASYMMETRIC
:
1621 return DDF_RAID5_0_RESTART
;
1622 case ALGORITHM_LEFT_SYMMETRIC
:
1623 return DDF_RAID5_N_CONTINUE
;
1624 case ALGORITHM_RIGHT_SYMMETRIC
:
1625 return -1; /* not mentioned in standard */
1631 static int rlq_to_layout(int rlq
, int prl
, int raiddisks
)
1635 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1637 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1645 return -1; /* FIXME this isn't checked */
1650 case DDF_RAID5_N_RESTART
:
1651 return ALGORITHM_LEFT_ASYMMETRIC
;
1652 case DDF_RAID5_0_RESTART
:
1653 return ALGORITHM_RIGHT_ASYMMETRIC
;
1654 case DDF_RAID5_N_CONTINUE
:
1655 return ALGORITHM_LEFT_SYMMETRIC
;
1663 static int init_super_ddf_bvd(struct supertype
*st
,
1664 mdu_array_info_t
*info
,
1665 unsigned long long size
,
1666 char *name
, char *homehost
,
1669 /* We are creating a BVD inside a pre-existing container.
1670 * so st->sb is already set.
1671 * We need to create a new vd_config and a new virtual_entry
1673 struct ddf_super
*ddf
= st
->sb
;
1675 struct virtual_entry
*ve
;
1677 struct vd_config
*vc
;
1681 if (__be16_to_cpu(ddf
->virt
->populated_vdes
)
1682 >= __be16_to_cpu(ddf
->virt
->max_vdes
)) {
1683 fprintf(stderr
, Name
": This ddf already has the "
1684 "maximum of %d virtual devices\n",
1685 __be16_to_cpu(ddf
->virt
->max_vdes
));
1689 for (venum
= 0; venum
< __be16_to_cpu(ddf
->virt
->max_vdes
); venum
++)
1690 if (all_ff(ddf
->virt
->entries
[venum
].guid
))
1692 if (venum
== __be16_to_cpu(ddf
->virt
->max_vdes
)) {
1693 fprintf(stderr
, Name
": Cannot find spare slot for "
1694 "virtual disk - DDF is corrupt\n");
1697 ve
= &ddf
->virt
->entries
[venum
];
1698 ddf
->conf_num
= venum
;
1700 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1701 * timestamp, random number
1703 make_header_guid(ve
->guid
);
1704 ve
->unit
= __cpu_to_be16(info
->md_minor
);
1706 ve
->guid_crc
= crc32(0, (unsigned char*)ddf
->anchor
.guid
, DDF_GUID_LEN
);
1708 ve
->state
= DDF_state_degraded
; /* Will be modified as devices are added */
1709 if (info
->state
& 1) /* clean */
1710 ve
->init_state
= DDF_init_full
;
1712 ve
->init_state
= DDF_init_not
;
1714 memset(ve
->pad1
, 0xff, 14);
1715 memset(ve
->name
, ' ', 16);
1717 strncpy(ve
->name
, name
, 16);
1718 ddf
->virt
->populated_vdes
=
1719 __cpu_to_be16(__be16_to_cpu(ddf
->virt
->populated_vdes
)+1);
1721 /* Now create a new vd_config */
1722 conflen
= __be16_to_cpu(ddf
->active
->config_record_len
);
1723 vcl
= malloc(offsetof(struct vcl
, conf
) + conflen
* 512);
1724 mppe
= __be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
1725 vcl
->lba_offset
= (__u64
*) &vcl
->conf
.phys_refnum
[mppe
];
1729 vc
->magic
= DDF_VD_CONF_MAGIC
;
1730 memcpy(vc
->guid
, ve
->guid
, DDF_GUID_LEN
);
1731 vc
->timestamp
= __cpu_to_be32(time(0)-DECADE
);
1732 vc
->seqnum
= __cpu_to_be32(1);
1733 memset(vc
->pad0
, 0xff, 24);
1734 vc
->prim_elmnt_count
= __cpu_to_be16(info
->raid_disks
);
1735 vc
->chunk_shift
= chunk_to_shift(info
->chunk_size
);
1736 vc
->prl
= level_to_prl(info
->level
);
1737 vc
->rlq
= layout_to_rlq(info
->level
, info
->layout
, info
->raid_disks
);
1738 vc
->sec_elmnt_count
= 1;
1739 vc
->sec_elmnt_seq
= 0;
1741 vc
->blocks
= __cpu_to_be64(info
->size
* 2);
1742 vc
->array_blocks
= __cpu_to_be64(
1743 calc_array_size(info
->level
, info
->raid_disks
, info
->layout
,
1744 info
->chunk_size
, info
->size
*2));
1745 memset(vc
->pad1
, 0xff, 8);
1746 vc
->spare_refs
[0] = 0xffffffff;
1747 vc
->spare_refs
[1] = 0xffffffff;
1748 vc
->spare_refs
[2] = 0xffffffff;
1749 vc
->spare_refs
[3] = 0xffffffff;
1750 vc
->spare_refs
[4] = 0xffffffff;
1751 vc
->spare_refs
[5] = 0xffffffff;
1752 vc
->spare_refs
[6] = 0xffffffff;
1753 vc
->spare_refs
[7] = 0xffffffff;
1754 memset(vc
->cache_pol
, 0, 8);
1756 memset(vc
->pad2
, 0xff, 3);
1757 memset(vc
->pad3
, 0xff, 52);
1758 memset(vc
->pad4
, 0xff, 192);
1759 memset(vc
->v0
, 0xff, 32);
1760 memset(vc
->v1
, 0xff, 32);
1761 memset(vc
->v2
, 0xff, 16);
1762 memset(vc
->v3
, 0xff, 16);
1763 memset(vc
->vendor
, 0xff, 32);
1765 memset(vc
->phys_refnum
, 0xff, 4*mppe
);
1766 memset(vc
->phys_refnum
+mppe
, 0x00, 8*mppe
);
1768 vcl
->next
= ddf
->conflist
;
1769 ddf
->conflist
= vcl
;
1774 static void add_to_super_ddf_bvd(struct supertype
*st
,
1775 mdu_disk_info_t
*dk
, int fd
, char *devname
)
1777 /* fd and devname identify a device with-in the ddf container (st).
1778 * dk identifies a location in the new BVD.
1779 * We need to find suitable free space in that device and update
1780 * the phys_refnum and lba_offset for the newly created vd_config.
1781 * We might also want to update the type in the phys_disk
1785 struct ddf_super
*ddf
= st
->sb
;
1786 struct vd_config
*vc
;
1791 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
1792 if (dl
->major
== dk
->major
&&
1793 dl
->minor
== dk
->minor
)
1795 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1798 vc
= &ddf
->newconf
->conf
;
1799 vc
->phys_refnum
[dk
->raid_disk
] = dl
->disk
.refnum
;
1800 mppe
= __be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
1801 lba_offset
= (__u64
*)(vc
->phys_refnum
+ mppe
);
1802 lba_offset
[dk
->raid_disk
] = 0; /* FIXME */
1804 dl
->vlist
[0] = ddf
->newconf
; /* FIXME */
1807 dl
->devname
= devname
;
1809 /* Check how many working raid_disks, and if we can mark
1810 * array as optimal yet
1814 for (i
=0; i
< __be16_to_cpu(vc
->prim_elmnt_count
); i
++)
1815 if (vc
->phys_refnum
[i
] != 0xffffffff)
1817 if (working
== __be16_to_cpu(vc
->prim_elmnt_count
))
1818 ->entries
[xx
].state
= (->entries
[xx
].state
& ~DDF_state_mask
)
1819 | DDF_state_optimal
;
1821 if (vc
->prl
== DDF_RAID6
&&
1822 working
+1 == __be16_to_cpu(vc
->prim_elmnt_count
))
1823 ->entries
[xx
].state
= (->entries
[xx
].state
& ~DDF_state_mask
)
1824 | DDF_state_part_optimal
;
1828 /* add a device to a container, either while creating it or while
1829 * expanding a pre-existing container
1831 static void add_to_super_ddf(struct supertype
*st
,
1832 mdu_disk_info_t
*dk
, int fd
, char *devname
)
1834 struct ddf_super
*ddf
= st
->sb
;
1838 unsigned long long size
;
1839 struct phys_disk_entry
*pde
;
1843 /* This is device numbered dk->number. We need to create
1844 * a phys_disk entry and a more detailed disk_data entry.
1847 dd
= malloc(sizeof(*dd
) + sizeof(dd
->vlist
[0]) * (ddf
->max_part
+1));
1848 dd
->major
= major(stb
.st_rdev
);
1849 dd
->minor
= minor(stb
.st_rdev
);
1850 dd
->devname
= devname
;
1851 dd
->next
= ddf
->dlist
;
1854 dd
->disk
.magic
= DDF_PHYS_DATA_MAGIC
;
1856 tm
= localtime(&now
);
1857 sprintf(dd
->disk
.guid
, "%8s%04d%02d%02d",
1858 T10
, tm
->tm_year
+1900, tm
->tm_mon
+1, tm
->tm_mday
);
1859 *(__u32
*)(dd
->disk
.guid
+ 16) = random();
1860 *(__u32
*)(dd
->disk
.guid
+ 20) = random();
1862 dd
->disk
.refnum
= random(); /* and hope for the best FIXME check this is unique!!*/
1863 dd
->disk
.forced_ref
= 1;
1864 dd
->disk
.forced_guid
= 1;
1865 memset(dd
->disk
.vendor
, ' ', 32);
1866 memcpy(dd
->disk
.vendor
, "Linux", 5);
1867 memset(dd
->disk
.pad
, 0xff, 442);
1868 for (i
= 0; i
< ddf
->max_part
+1 ; i
++)
1869 dd
->vlist
[i
] = NULL
;
1871 n
= __be16_to_cpu(ddf
->phys
->used_pdes
);
1872 pde
= &ddf
->phys
->entries
[n
];
1874 ddf
->phys
->used_pdes
= __cpu_to_be16(n
);
1876 memcpy(pde
->guid
, dd
->disk
.guid
, DDF_GUID_LEN
);
1877 pde
->refnum
= dd
->disk
.refnum
;
1878 pde
->type
= __cpu_to_be16(DDF_Forced_PD_GUID
|DDF_Global_Spare
);
1879 pde
->state
= __cpu_to_be16(DDF_Online
);
1880 get_dev_size(fd
, NULL
, &size
);
1881 /* We are required to reserve 32Meg, and record the size in sectors */
1882 pde
->config_size
= __cpu_to_be64( (size
- 32*1024*1024) / 512);
1883 sprintf(pde
->path
, "%17.17s","Information: nil") ;
1884 memset(pde
->pad
, 0xff, 6);
1890 * This is the write_init_super method for a ddf container. It is
1891 * called when creating a container or adding another device to a
1896 static int __write_init_super_ddf(struct supertype
*st
, int do_close
)
1899 struct ddf_super
*ddf
= st
->sb
;
1905 unsigned long long size
, sector
;
1907 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
1913 /* We need to fill in the primary, (secondary) and workspace
1914 * lba's in the headers, set their checksums,
1915 * Also checksum phys, virt....
1917 * Then write everything out, finally the anchor is written.
1919 get_dev_size(fd
, NULL
, &size
);
1921 ddf
->anchor
.workspace_lba
= __cpu_to_be64(size
- 32*1024*2);
1922 ddf
->anchor
.primary_lba
= __cpu_to_be64(size
- 16*1024*2);
1923 ddf
->anchor
.seq
= __cpu_to_be32(1);
1924 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
1925 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
1927 ddf
->anchor
.openflag
= 0xFF; /* 'open' means nothing */
1928 ddf
->anchor
.seq
= 0xFFFFFFFF; /* no sequencing in anchor */
1929 ddf
->anchor
.crc
= calc_crc(&ddf
->anchor
, 512);
1931 ddf
->primary
.openflag
= 0;
1932 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
1934 ddf
->secondary
.openflag
= 0;
1935 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
1937 ddf
->primary
.crc
= calc_crc(&ddf
->primary
, 512);
1938 ddf
->secondary
.crc
= calc_crc(&ddf
->secondary
, 512);
1940 sector
= size
- 16*1024*2;
1941 lseek64(fd
, sector
<<9, 0);
1942 write(fd
, &ddf
->primary
, 512);
1944 ddf
->controller
.crc
= calc_crc(&ddf
->controller
, 512);
1945 write(fd
, &ddf
->controller
, 512);
1947 ddf
->phys
->crc
= calc_crc(ddf
->phys
, ddf
->pdsize
);
1949 write(fd
, ddf
->phys
, ddf
->pdsize
);
1951 ddf
->virt
->crc
= calc_crc(ddf
->virt
, ddf
->vdsize
);
1952 write(fd
, ddf
->virt
, ddf
->vdsize
);
1954 /* Now write lots of config records. */
1955 n_config
= __be16_to_cpu(ddf
->active
->max_partitions
);
1956 conf_size
= __be16_to_cpu(ddf
->active
->config_record_len
) * 512;
1957 for (i
= 0 ; i
<= n_config
; i
++) {
1958 struct vcl
*c
= d
->vlist
[i
];
1961 c
->conf
.crc
= calc_crc(&c
->conf
, conf_size
);
1962 write(fd
, &c
->conf
, conf_size
);
1964 __u32 sig
= 0xffffffff;
1966 lseek64(fd
, conf_size
-4, SEEK_CUR
);
1969 d
->disk
.crc
= calc_crc(&d
->disk
, 512);
1970 write(fd
, &d
->disk
, 512);
1972 /* Maybe do the same for secondary */
1974 lseek64(fd
, (size
-1)*512, SEEK_SET
);
1975 write(fd
, &ddf
->anchor
, 512);
1984 static int write_init_super_ddf(struct supertype
*st
)
1986 return __write_init_super_ddf(st
, 1);
1991 static __u64
avail_size_ddf(struct supertype
*st
, __u64 devsize
)
1993 /* We must reserve the last 32Meg */
1994 if (devsize
<= 32*1024*2)
1996 return devsize
- 32*1024*2;
2000 int validate_geometry_ddf(struct supertype
*st
,
2001 int level
, int layout
, int raiddisks
,
2002 int chunk
, unsigned long long size
,
2003 char *dev
, unsigned long long *freesize
)
2009 /* ddf potentially supports lots of things, but it depends on
2010 * what devices are offered (and maybe kernel version?)
2011 * If given unused devices, we will make a container.
2012 * If given devices in a container, we will make a BVD.
2013 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2016 if (level
== LEVEL_CONTAINER
) {
2017 st
->ss
= &super_ddf_container
;
2019 int rv
=st
->ss
->validate_geometry(st
, level
, layout
,
2026 return st
->ss
->validate_geometry(st
, level
, layout
, raiddisks
,
2027 chunk
, size
, dev
, freesize
);
2031 /* creating in a given container */
2032 st
->ss
= &super_ddf_bvd
;
2034 int rv
=st
->ss
->validate_geometry(st
, level
, layout
,
2041 return st
->ss
->validate_geometry(st
, level
, layout
, raiddisks
,
2042 chunk
, size
, dev
, freesize
);
2044 /* FIXME should exclude MULTIPATH, or more appropriately, allow
2045 * only known levels.
2050 /* This device needs to be either a device in a 'ddf' container,
2051 * or it needs to be a 'ddf-bvd' array.
2054 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2056 sra
= sysfs_read(fd
, 0, GET_VERSION
);
2058 if (sra
&& sra
->array
.major_version
== -1 &&
2059 strcmp(sra
->text_version
, "ddf-bvd") == 0) {
2060 st
->ss
= &super_ddf_svd
;
2061 return st
->ss
->validate_geometry(st
, level
, layout
,
2062 raiddisks
, chunk
, size
,
2067 Name
": Cannot create this array on device %s\n",
2071 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2072 fprintf(stderr
, Name
": Cannot open %s: %s\n",
2073 dev
, strerror(errno
));
2076 /* Well, it is in use by someone, maybe a 'ddf' container. */
2077 cfd
= open_container(fd
);
2080 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
2084 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2086 if (sra
&& sra
->array
.major_version
== -1 &&
2087 strcmp(sra
->text_version
, "ddf") == 0) {
2088 /* This is a member of a ddf container. Load the container
2089 * and try to create a bvd
2091 struct ddf_super
*ddf
;
2092 st
->ss
= &super_ddf_bvd
;
2093 if (load_super_ddf_all(st
, cfd
, (void **)&ddf
, NULL
, 1) == 0) {
2095 st
->container_dev
= fd2devnum(cfd
);
2097 return st
->ss
->validate_geometry(st
, level
, layout
,
2098 raiddisks
, chunk
, size
,
2103 fprintf(stderr
, Name
": Cannot use %s: Already in use\n",
2108 int validate_geometry_ddf_container(struct supertype
*st
,
2109 int level
, int layout
, int raiddisks
,
2110 int chunk
, unsigned long long size
,
2111 char *dev
, unsigned long long *freesize
)
2114 unsigned long long ldsize
;
2116 if (level
!= LEVEL_CONTAINER
)
2121 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2123 fprintf(stderr
, Name
": Cannot open %s: %s\n",
2124 dev
, strerror(errno
));
2127 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2133 *freesize
= avail_size_ddf(st
, ldsize
);
2139 unsigned long long start
, size
;
2141 int cmp_extent(const void *av
, const void *bv
)
2143 const struct extent
*a
= av
;
2144 const struct extent
*b
= bv
;
2145 if (a
->start
< b
->start
)
2147 if (a
->start
> b
->start
)
2152 struct extent
*get_extents(struct ddf_super
*ddf
, struct dl
*dl
)
2154 /* find a list of used extents on the give physical device
2155 * (dnum) or the given ddf.
2156 * Return a malloced array of 'struct extent'
2158 FIXME ignore DDF_Legacy devices?
2166 for (dnum
= 0; dnum
< ddf
->phys
->used_pdes
; dnum
++)
2167 if (memcmp(dl
->disk
.guid
,
2168 ddf
->phys
->entries
[dnum
].guid
,
2172 if (dnum
== ddf
->phys
->used_pdes
)
2175 rv
= malloc(sizeof(struct extent
) * (ddf
->max_part
+ 2));
2179 for (i
= 0; i
< ddf
->max_part
+1; i
++) {
2180 struct vcl
*v
= dl
->vlist
[i
];
2183 for (j
=0; j
< v
->conf
.prim_elmnt_count
; j
++)
2184 if (v
->conf
.phys_refnum
[j
] == dl
->disk
.refnum
) {
2185 /* This device plays role 'j' in 'v'. */
2186 rv
[n
].start
= __be64_to_cpu(v
->lba_offset
[j
]);
2187 rv
[n
].size
= __be64_to_cpu(v
->conf
.blocks
);
2192 qsort(rv
, n
, sizeof(*rv
), cmp_extent
);
2194 rv
[n
].start
= __be64_to_cpu(ddf
->phys
->entries
[dnum
].config_size
);
2199 int validate_geometry_ddf_bvd(struct supertype
*st
,
2200 int level
, int layout
, int raiddisks
,
2201 int chunk
, unsigned long long size
,
2202 char *dev
, unsigned long long *freesize
)
2205 struct ddf_super
*ddf
= st
->sb
;
2207 unsigned long long pos
= 0;
2208 unsigned long long maxsize
;
2211 /* ddf/bvd supports lots of things, but not containers */
2212 if (level
== LEVEL_CONTAINER
)
2214 /* We must have the container info already read in. */
2219 /* General test: make sure there is space for
2220 * 'raiddisks' device extents of size 'size'.
2222 unsigned long long minsize
= size
;
2226 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2231 e
= get_extents(ddf
, dl
);
2234 unsigned long long esize
;
2235 esize
= e
[i
].start
- pos
;
2236 if (esize
>= minsize
)
2238 pos
= e
[i
].start
+ e
[i
].size
;
2240 } while (e
[i
-1].size
);
2245 if (dcnt
< raiddisks
) {
2246 fprintf(stderr
, Name
": Not enough devices with space "
2247 "for this array (%d < %d)\n",
2253 /* This device must be a member of the set */
2254 if (stat(dev
, &stb
) < 0)
2256 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2258 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
2259 if (dl
->major
== major(stb
.st_rdev
) &&
2260 dl
->minor
== minor(stb
.st_rdev
))
2264 fprintf(stderr
, Name
": %s is not in the same DDF set\n",
2268 e
= get_extents(ddf
, dl
);
2272 unsigned long long esize
;
2273 esize
= e
[i
].start
- pos
;
2274 if (esize
>= maxsize
)
2276 pos
= e
[i
].start
+ e
[i
].size
;
2278 } while (e
[i
-1].size
);
2279 *freesize
= maxsize
;
2284 int validate_geometry_ddf_svd(struct supertype
*st
,
2285 int level
, int layout
, int raiddisks
,
2286 int chunk
, unsigned long long size
,
2287 char *dev
, unsigned long long *freesize
)
2289 /* dd/svd only supports striped, mirrored, concat, spanned... */
2290 if (level
!= LEVEL_LINEAR
&&
2298 static int load_super_ddf_all(struct supertype
*st
, int fd
,
2299 void **sbp
, char *devname
, int keep_fd
)
2302 struct ddf_super
*super
;
2303 struct mdinfo
*sd
, *best
= NULL
;
2309 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
2312 if (sra
->array
.major_version
!= -1 ||
2313 sra
->array
.minor_version
!= -2 ||
2314 strcmp(sra
->text_version
, "ddf") != 0)
2317 super
= malloc(sizeof(*super
));
2320 memset(super
, 0, sizeof(*super
));
2322 /* first, try each device, and choose the best ddf */
2323 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2325 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2326 dfd
= dev_open(nm
, O_RDONLY
);
2329 rv
= load_ddf_headers(dfd
, super
, NULL
);
2332 seq
= __be32_to_cpu(super
->active
->seq
);
2333 if (super
->active
->openflag
)
2335 if (!best
|| seq
> bestseq
) {
2343 /* OK, load this ddf */
2344 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
2345 dfd
= dev_open(nm
, O_RDONLY
);
2348 load_ddf_headers(dfd
, super
, NULL
);
2349 load_ddf_global(dfd
, super
, NULL
);
2351 /* Now we need the device-local bits */
2352 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2353 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2354 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2357 seq
= load_ddf_local(dfd
, super
, NULL
, keep_fd
);
2358 if (!keep_fd
) close(dfd
);
2361 if (st
->ss
== NULL
) {
2362 st
->ss
= &super_ddf_container
;
2363 st
->minor_version
= 0;
2365 st
->container_dev
= fd2devnum(fd
);
2373 static struct mdinfo
*container_content_ddf(struct supertype
*st
)
2375 /* Given a container loaded by load_super_ddf_all,
2376 * extract information about all the arrays into
2379 * For each vcl in conflist: create an mdinfo, fill it in,
2380 * then look for matching devices (phys_refnum) in dlist
2381 * and create appropriate device mdinfo.
2383 struct ddf_super
*ddf
= st
->sb
;
2384 struct mdinfo
*rest
= NULL
;
2387 for (vc
= ddf
->conflist
; vc
; vc
=vc
->next
)
2391 struct mdinfo
*this;
2392 this = malloc(sizeof(*this));
2393 memset(this, 0, sizeof(*this));
2397 this->array
.major_version
= 1000;
2398 this->array
.minor_version
= 0;
2399 this->array
.patch_version
= 0;
2400 this->array
.level
= map_num1(ddf_level_num
, vc
->conf
.prl
);
2401 this->array
.raid_disks
=
2402 __be16_to_cpu(vc
->conf
.prim_elmnt_count
);
2403 this->array
.layout
= rlq_to_layout(vc
->conf
.rlq
, vc
->conf
.prl
,
2404 this->array
.raid_disks
);
2405 this->array
.md_minor
= -1;
2406 this->array
.ctime
= DECADE
+
2407 __be32_to_cpu(*(__u32
*)(vc
->conf
.guid
+16));
2408 this->array
.utime
= DECADE
+
2409 __be32_to_cpu(vc
->conf
.timestamp
);
2410 this->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
2412 for (i
=0; i
< __be16_to_cpu(ddf
->virt
->populated_vdes
); i
++)
2413 if (memcmp(ddf
->virt
->entries
[i
].guid
,
2414 vc
->conf
.guid
, DDF_GUID_LEN
) == 0)
2416 if ((ddf
->virt
->entries
[i
].state
& DDF_state_inconsistent
) ||
2417 (ddf
->virt
->entries
[i
].init_state
& DDF_initstate_mask
) !=
2419 this->array
.state
= 0;
2420 this->resync_start
= 0;
2422 this->array
.state
= 1;
2423 this->resync_start
= ~0ULL;
2425 memcpy(this->name
, ddf
->virt
->entries
[i
].name
, 32);
2428 memset(this->uuid
, 0, sizeof(this->uuid
));
2429 this->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
2430 this->array
.size
= this->component_size
/ 2;
2431 this->container_member
= i
;
2433 sprintf(this->text_version
, "/%s/%d",
2434 devnum2devname(st
->container_dev
),
2435 this->container_member
);
2438 mppe
= __be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
2439 for (i
=0 ; i
< mppe
; i
++) {
2443 if (vc
->conf
.phys_refnum
[i
] == 0xFFFFFFFF)
2446 this->array
.working_disks
++;
2448 for (d
= ddf
->dlist
; d
; d
=d
->next
)
2449 if (d
->disk
.refnum
== vc
->conf
.phys_refnum
[i
])
2454 dev
= malloc(sizeof(*dev
));
2455 memset(dev
, 0, sizeof(*dev
));
2456 dev
->next
= this->devs
;
2459 dev
->disk
.number
= __be32_to_cpu(d
->disk
.refnum
);
2460 dev
->disk
.major
= d
->major
;
2461 dev
->disk
.minor
= d
->minor
;
2462 dev
->disk
.raid_disk
= i
;
2463 dev
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
2465 dev
->events
= __le32_to_cpu(ddf
->primary
.seq
);
2466 dev
->data_offset
= vc
->lba_offset
[i
];
2467 dev
->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
2469 strcpy(dev
->name
, d
->devname
);
2475 static int init_zero_ddf(struct supertype
*st
,
2476 mdu_array_info_t
*info
,
2477 unsigned long long size
, char *name
,
2478 char *homehost
, int *uuid
)
2484 static int store_zero_ddf(struct supertype
*st
, int fd
)
2486 unsigned long long dsize
;
2488 memset(buf
, 0, 512);
2491 if (!get_dev_size(fd
, NULL
, &dsize
))
2494 lseek64(fd
, dsize
-512, 0);
2495 write(fd
, buf
, 512);
2499 static int compare_super_ddf(struct supertype
*st
, struct supertype
*tst
)
2503 * 0 same, or first was empty, and second was copied
2504 * 1 second had wrong number
2506 * 3 wrong other info
2508 struct ddf_super
*first
= st
->sb
;
2509 struct ddf_super
*second
= tst
->sb
;
2517 if (memcmp(first
->anchor
.guid
, second
->anchor
.guid
, DDF_GUID_LEN
) != 0)
2520 /* FIXME should I look at anything else? */
2525 * A new array 'a' has been started which claims to be instance 'inst'
2526 * within container 'c'.
2527 * We need to confirm that the array matches the metadata in 'c' so
2528 * that we don't corrupt any metadata.
2530 static int ddf_open_new(struct supertype
*c
, struct active_array
*a
, char *inst
)
2532 fprintf(stderr
, "ddf: open_new %s\n", inst
);
2533 a
->info
.container_member
= atoi(inst
);
2538 * The array 'a' is to be marked clean in the metadata.
2539 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2540 * clean up to the point (in sectors). If that cannot be recorded in the
2541 * metadata, then leave it as dirty.
2543 * For DDF, we need to clear the DDF_state_inconsistent bit in the
2544 * !global! virtual_disk.virtual_entry structure.
2546 static void ddf_set_array_state(struct active_array
*a
, int consistent
)
2548 struct ddf_super
*ddf
= a
->container
->sb
;
2549 int inst
= a
->info
.container_member
;
2551 ddf
->virt
->entries
[inst
].state
&= ~DDF_state_inconsistent
;
2553 ddf
->virt
->entries
[inst
].state
|= DDF_state_inconsistent
;
2554 ddf
->virt
->entries
[inst
].init_state
&= ~DDF_initstate_mask
;
2555 if (a
->resync_start
== ~0ULL)
2556 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_full
;
2557 else if (a
->resync_start
== 0)
2558 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_not
;
2560 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_quick
;
2562 printf("ddf mark %s %llu\n", consistent
?"clean":"dirty",
2567 * The state of each disk is stored in the global phys_disk structure
2568 * in phys_disk.entries[n].state.
2569 * This makes various combinations awkward.
2570 * - When a device fails in any array, it must be failed in all arrays
2571 * that include a part of this device.
2572 * - When a component is rebuilding, we cannot include it officially in the
2573 * array unless this is the only array that uses the device.
2575 * So: when transitioning:
2576 * Online -> failed, just set failed flag. monitor will propagate
2577 * spare -> online, the device might need to be added to the array.
2578 * spare -> failed, just set failed. Don't worry if in array or not.
2580 static void ddf_set_disk(struct active_array
*a
, int n
, int state
)
2582 struct ddf_super
*ddf
= a
->container
->sb
;
2583 int inst
= a
->info
.container_member
;
2584 struct vd_config
*vc
= find_vdcr(ddf
, inst
);
2585 int pd
= find_phys(ddf
, vc
->phys_refnum
[n
]);
2589 fprintf(stderr
, "ddf: cannot find instance %d!!\n", inst
);
2593 /* disk doesn't currently exist. If it is now in_sync,
2595 if ((state
& DS_INSYNC
) && ! (state
& DS_FAULTY
)) {
2596 /* Find dev 'n' in a->info->devs, determine the
2597 * ddf refnum, and set vc->phys_refnum and update
2603 if (state
& DS_FAULTY
)
2604 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Failed
);
2605 if (state
& DS_INSYNC
) {
2606 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Online
);
2607 ddf
->phys
->entries
[pd
].state
&= __cpu_to_be16(~DDF_Rebuilding
);
2611 /* Now we need to check the state of the array and update
2612 * virtual_disk.entries[n].state.
2613 * It needs to be one of "optimal", "degraded", "failed".
2614 * I don't understand 'deleted' or 'missing'.
2617 for (i
=0; i
< a
->info
.array
.raid_disks
; i
++) {
2618 pd
= find_phys(ddf
, vc
->phys_refnum
[i
]);
2621 st
= ddf
->phys
->entries
[pd
].state
;
2622 if ((state
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
2626 state
= DDF_state_degraded
;
2627 if (working
== a
->info
.array
.raid_disks
)
2628 state
= DDF_state_optimal
;
2629 else switch(vc
->prl
) {
2633 state
= DDF_state_failed
;
2637 state
= DDF_state_failed
;
2641 if (working
< a
->info
.array
.raid_disks
-1)
2642 state
= DDF_state_failed
;
2645 if (working
< a
->info
.array
.raid_disks
-2)
2646 state
= DDF_state_failed
;
2647 else if (working
== a
->info
.array
.raid_disks
-1)
2648 state
= DDF_state_part_optimal
;
2652 ddf
->virt
->entries
[inst
].state
=
2653 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
2656 fprintf(stderr
, "ddf: set_disk %d\n", n
);
2659 static void ddf_sync_metadata(struct active_array
*a
)
2663 * Write all data to all devices.
2664 * Later, we might be able to track whether only local changes
2665 * have been made, or whether any global data has been changed,
2666 * but ddf is sufficiently weird that it probably always
2667 * changes global data ....
2669 __write_init_super_ddf(a
->container
, 0);
2670 fprintf(stderr
, "ddf: sync_metadata\n");
2673 struct superswitch super_ddf
= {
2675 .examine_super
= examine_super_ddf
,
2676 .brief_examine_super
= brief_examine_super_ddf
,
2677 .detail_super
= detail_super_ddf
,
2678 .brief_detail_super
= brief_detail_super_ddf
,
2679 .validate_geometry
= validate_geometry_ddf
,
2681 .match_home
= match_home_ddf
,
2682 .uuid_from_super
= uuid_from_super_ddf
,
2683 .getinfo_super
= getinfo_super_ddf
,
2684 .update_super
= update_super_ddf
,
2686 .avail_size
= avail_size_ddf
,
2688 .compare_super
= compare_super_ddf
,
2690 .load_super
= load_super_ddf
,
2691 .init_super
= init_zero_ddf
,
2692 .store_super
= store_zero_ddf
,
2693 .free_super
= free_super_ddf
,
2694 .match_metadata_desc
= match_metadata_desc_ddf
,
2695 .getinfo_super_n
= getinfo_super_n_container
,
2703 .open_new
= ddf_open_new
,
2704 .set_array_state
= ddf_set_array_state
,
2705 .set_disk
= ddf_set_disk
,
2706 .sync_metadata
= ddf_sync_metadata
,
2711 /* Super_ddf_container is set by validate_geometry_ddf when given a
2712 * device that is not part of any array
2714 struct superswitch super_ddf_container
= {
2716 .validate_geometry
= validate_geometry_ddf_container
,
2717 .write_init_super
= write_init_super_ddf
,
2720 .load_super
= load_super_ddf
,
2721 .init_super
= init_super_ddf
,
2722 .add_to_super
= add_to_super_ddf
,
2723 .getinfo_super
= getinfo_super_ddf
,
2725 .free_super
= free_super_ddf
,
2727 .container_content
= container_content_ddf
,
2728 .getinfo_super_n
= getinfo_super_n_container
,
2735 struct superswitch super_ddf_bvd
= {
2737 // .detail_super = detail_super_ddf_bvd,
2738 // .brief_detail_super = brief_detail_super_ddf_bvd,
2739 .validate_geometry
= validate_geometry_ddf_bvd
,
2740 .write_init_super
= write_init_super_ddf
,
2742 .update_super
= update_super_ddf
,
2743 .init_super
= init_super_ddf_bvd
,
2744 .add_to_super
= add_to_super_ddf_bvd
,
2745 .getinfo_super
= getinfo_super_ddf_bvd
,
2746 .getinfo_super_n
= getinfo_super_n_bvd
,
2748 .load_super
= load_super_ddf
,
2749 .free_super
= free_super_ddf
,
2750 .match_metadata_desc
= match_metadata_desc_ddf_bvd
,
2758 struct superswitch super_ddf_svd
= {
2760 // .detail_super = detail_super_ddf_svd,
2761 // .brief_detail_super = brief_detail_super_ddf_svd,
2762 .validate_geometry
= validate_geometry_ddf_svd
,
2764 .update_super
= update_super_ddf
,
2765 .init_super
= init_super_ddf
,
2767 .load_super
= load_super_ddf
,
2768 .free_super
= free_super_ddf
,
2769 .match_metadata_desc
= match_metadata_desc_ddf_svd
,