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
;
401 int max_part
, mppe
, conf_rec_len
;
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 spare_assign
*spare
;
416 struct vcl
*vlist
[0]; /* max_part in size */
421 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
424 extern struct superswitch super_ddf_container
, super_ddf_bvd
, super_ddf
;
426 static int calc_crc(void *buf
, int len
)
428 /* crcs are always at the same place as in the ddf_header */
429 struct ddf_header
*ddf
= buf
;
430 __u32 oldcrc
= ddf
->crc
;
432 ddf
->crc
= 0xffffffff;
434 newcrc
= crc32(0, buf
, len
);
439 static int load_ddf_header(int fd
, unsigned long long lba
,
440 unsigned long long size
,
442 struct ddf_header
*hdr
, struct ddf_header
*anchor
)
444 /* read a ddf header (primary or secondary) from fd/lba
445 * and check that it is consistent with anchor
447 * magic, crc, guid, rev, and LBA's header_type, and
448 * everything after header_type must be the same
453 if (lseek64(fd
, lba
<<9, 0) < 0)
456 if (read(fd
, hdr
, 512) != 512)
459 if (hdr
->magic
!= DDF_HEADER_MAGIC
)
461 if (calc_crc(hdr
, 512) != hdr
->crc
)
463 if (memcmp(anchor
->guid
, hdr
->guid
, DDF_GUID_LEN
) != 0 ||
464 memcmp(anchor
->revision
, hdr
->revision
, 8) != 0 ||
465 anchor
->primary_lba
!= hdr
->primary_lba
||
466 anchor
->secondary_lba
!= hdr
->secondary_lba
||
468 memcmp(anchor
->pad2
, hdr
->pad2
, 512 -
469 offsetof(struct ddf_header
, pad2
)) != 0)
472 /* Looks good enough to me... */
476 static void *load_section(int fd
, struct ddf_super
*super
, void *buf
,
477 __u32 offset_be
, __u32 len_be
, int check
)
479 unsigned long long offset
= __be32_to_cpu(offset_be
);
480 unsigned long long len
= __be32_to_cpu(len_be
);
481 int dofree
= (buf
== NULL
);
484 if (len
!= 2 && len
!= 8 && len
!= 32
485 && len
!= 128 && len
!= 512)
491 /* All pre-allocated sections are a single block */
495 buf
= malloc(len
<<9);
499 if (super
->active
->type
== 1)
500 offset
+= __be64_to_cpu(super
->active
->primary_lba
);
502 offset
+= __be64_to_cpu(super
->active
->secondary_lba
);
504 if (lseek64(fd
, offset
<<9, 0) != (offset
<<9)) {
509 if (read(fd
, buf
, len
<<9) != (len
<<9)) {
517 static int load_ddf_headers(int fd
, struct ddf_super
*super
, char *devname
)
519 unsigned long long dsize
;
521 get_dev_size(fd
, NULL
, &dsize
);
523 if (lseek64(fd
, dsize
-512, 0) < 0) {
526 Name
": Cannot seek to anchor block on %s: %s\n",
527 devname
, strerror(errno
));
530 if (read(fd
, &super
->anchor
, 512) != 512) {
533 Name
": Cannot read anchor block on %s: %s\n",
534 devname
, strerror(errno
));
537 if (super
->anchor
.magic
!= DDF_HEADER_MAGIC
) {
539 fprintf(stderr
, Name
": no DDF anchor found on %s\n",
543 if (calc_crc(&super
->anchor
, 512) != super
->anchor
.crc
) {
545 fprintf(stderr
, Name
": bad CRC on anchor on %s\n",
549 if (memcmp(super
->anchor
.revision
, DDF_REVISION
, 8) != 0) {
551 fprintf(stderr
, Name
": can only support super revision"
552 " %.8s, not %.8s on %s\n",
553 DDF_REVISION
, super
->anchor
.revision
, devname
);
556 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.primary_lba
),
558 &super
->primary
, &super
->anchor
) == 0) {
561 Name
": Failed to load primary DDF header "
565 super
->active
= &super
->primary
;
566 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.secondary_lba
),
568 &super
->secondary
, &super
->anchor
)) {
569 if ((__be32_to_cpu(super
->primary
.seq
)
570 < __be32_to_cpu(super
->secondary
.seq
) &&
571 !super
->secondary
.openflag
)
572 || (__be32_to_cpu(super
->primary
.seq
)
573 == __be32_to_cpu(super
->secondary
.seq
) &&
574 super
->primary
.openflag
&& !super
->secondary
.openflag
)
576 super
->active
= &super
->secondary
;
581 static int load_ddf_global(int fd
, struct ddf_super
*super
, char *devname
)
584 ok
= load_section(fd
, super
, &super
->controller
,
585 super
->active
->controller_section_offset
,
586 super
->active
->controller_section_length
,
588 super
->phys
= load_section(fd
, super
, NULL
,
589 super
->active
->phys_section_offset
,
590 super
->active
->phys_section_length
,
592 super
->pdsize
= __be32_to_cpu(super
->active
->phys_section_length
) * 512;
594 super
->virt
= load_section(fd
, super
, NULL
,
595 super
->active
->virt_section_offset
,
596 super
->active
->virt_section_length
,
598 super
->vdsize
= __be32_to_cpu(super
->active
->virt_section_length
) * 512;
608 super
->conflist
= NULL
;
611 super
->max_part
= __be16_to_cpu(super
->active
->max_partitions
);
612 super
->mppe
= __be16_to_cpu(super
->active
->max_primary_element_entries
);
613 super
->conf_rec_len
= __be16_to_cpu(super
->active
->config_record_len
);
617 static int load_ddf_local(int fd
, struct ddf_super
*super
,
618 char *devname
, int keep
)
626 /* First the local disk info */
627 dl
= malloc(sizeof(*dl
) +
628 (super
->max_part
) * sizeof(dl
->vlist
[0]));
630 load_section(fd
, super
, &dl
->disk
,
631 super
->active
->data_section_offset
,
632 super
->active
->data_section_length
,
634 dl
->devname
= devname
? strdup(devname
) : NULL
;
637 dl
->major
= major(stb
.st_rdev
);
638 dl
->minor
= minor(stb
.st_rdev
);
639 dl
->next
= super
->dlist
;
640 dl
->fd
= keep
? fd
: -1;
642 for (i
=0 ; i
< super
->max_part
; i
++)
646 /* Now the config list. */
647 /* 'conf' is an array of config entries, some of which are
648 * probably invalid. Those which are good need to be copied into
652 conf
= load_section(fd
, super
, NULL
,
653 super
->active
->config_section_offset
,
654 super
->active
->config_section_length
,
659 i
< __be32_to_cpu(super
->active
->config_section_length
);
660 i
+= super
->conf_rec_len
) {
661 struct vd_config
*vd
=
662 (struct vd_config
*)((char*)conf
+ i
*512);
665 if (vd
->magic
== DDF_SPARE_ASSIGN_MAGIC
) {
668 dl
->spare
= malloc(super
->conf_rec_len
*512);
669 memcpy(dl
->spare
, vd
, super
->conf_rec_len
*512);
672 if (vd
->magic
!= DDF_VD_CONF_MAGIC
)
674 for (vcl
= super
->conflist
; vcl
; vcl
= vcl
->next
) {
675 if (memcmp(vcl
->conf
.guid
,
676 vd
->guid
, DDF_GUID_LEN
) == 0)
681 dl
->vlist
[vnum
++] = vcl
;
682 if (__be32_to_cpu(vd
->seqnum
) <=
683 __be32_to_cpu(vcl
->conf
.seqnum
))
686 vcl
= malloc(super
->conf_rec_len
*512 +
687 offsetof(struct vcl
, conf
));
688 vcl
->next
= super
->conflist
;
689 super
->conflist
= vcl
;
690 dl
->vlist
[vnum
++] = vcl
;
692 memcpy(&vcl
->conf
, vd
, super
->conf_rec_len
*512);
693 vcl
->lba_offset
= (__u64
*)
694 &vcl
->conf
.phys_refnum
[super
->mppe
];
702 static int load_super_ddf_all(struct supertype
*st
, int fd
,
703 void **sbp
, char *devname
, int keep_fd
);
705 static int load_super_ddf(struct supertype
*st
, int fd
,
708 unsigned long long dsize
;
709 struct ddf_super
*super
;
713 if (load_super_ddf_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
717 if (get_dev_size(fd
, devname
, &dsize
) == 0)
720 /* 32M is a lower bound */
721 if (dsize
<= 32*1024*1024) {
724 Name
": %s is too small for ddf: "
725 "size is %llu sectors.\n",
733 Name
": %s is an odd size for ddf: "
734 "size is %llu bytes.\n",
740 super
= malloc(sizeof(*super
));
742 fprintf(stderr
, Name
": malloc of %zu failed.\n",
746 memset(super
, 0, sizeof(*super
));
748 rv
= load_ddf_headers(fd
, super
, devname
);
754 /* Have valid headers and have chosen the best. Let's read in the rest*/
756 rv
= load_ddf_global(fd
, super
, devname
);
761 Name
": Failed to load all information "
762 "sections on %s\n", devname
);
767 load_ddf_local(fd
, super
, devname
, 0);
769 /* Should possibly check the sections .... */
772 if (st
->ss
== NULL
) {
774 st
->minor_version
= 0;
781 static void free_super_ddf(struct supertype
*st
)
783 struct ddf_super
*ddf
= st
->sb
;
788 while (ddf
->conflist
) {
789 struct vcl
*v
= ddf
->conflist
;
790 ddf
->conflist
= v
->next
;
794 struct dl
*d
= ddf
->dlist
;
795 ddf
->dlist
= d
->next
;
806 static struct supertype
*match_metadata_desc_ddf(char *arg
)
808 /* 'ddf' only support containers */
809 struct supertype
*st
;
810 if (strcmp(arg
, "ddf") != 0 &&
811 strcmp(arg
, "default") != 0
815 st
= malloc(sizeof(*st
));
818 st
->minor_version
= 0;
823 static struct supertype
*match_metadata_desc_ddf_bvd(char *arg
)
825 struct supertype
*st
;
826 if (strcmp(arg
, "ddf/bvd") != 0 &&
827 strcmp(arg
, "bvd") != 0 &&
828 strcmp(arg
, "default") != 0
832 st
= malloc(sizeof(*st
));
833 st
->ss
= &super_ddf_bvd
;
835 st
->minor_version
= 0;
839 static struct supertype
*match_metadata_desc_ddf_svd(char *arg
)
841 struct supertype
*st
;
842 if (strcmp(arg
, "ddf/svd") != 0 &&
843 strcmp(arg
, "svd") != 0 &&
844 strcmp(arg
, "default") != 0
848 st
= malloc(sizeof(*st
));
849 st
->ss
= &super_ddf_svd
;
851 st
->minor_version
= 0;
858 static mapping_t ddf_state
[] = {
864 { "Partially Optimal", 5},
870 static mapping_t ddf_init_state
[] = {
871 { "Not Initialised", 0},
872 { "QuickInit in Progress", 1},
873 { "Fully Initialised", 2},
877 static mapping_t ddf_access
[] = {
881 { "Blocked (no access)", 3},
885 static mapping_t ddf_level
[] = {
886 { "RAID0", DDF_RAID0
},
887 { "RAID1", DDF_RAID1
},
888 { "RAID3", DDF_RAID3
},
889 { "RAID4", DDF_RAID4
},
890 { "RAID5", DDF_RAID5
},
891 { "RAID1E",DDF_RAID1E
},
893 { "CONCAT",DDF_CONCAT
},
894 { "RAID5E",DDF_RAID5E
},
895 { "RAID5EE",DDF_RAID5EE
},
896 { "RAID6", DDF_RAID6
},
899 static mapping_t ddf_sec_level
[] = {
900 { "Striped", DDF_2STRIPED
},
901 { "Mirrored", DDF_2MIRRORED
},
902 { "Concat", DDF_2CONCAT
},
903 { "Spanned", DDF_2SPANNED
},
911 static struct num_mapping ddf_level_num
[] = {
914 { DDF_RAID3
, LEVEL_UNSUPPORTED
},
917 { DDF_RAID1E
, LEVEL_UNSUPPORTED
},
918 { DDF_JBOD
, LEVEL_UNSUPPORTED
},
919 { DDF_CONCAT
, LEVEL_LINEAR
},
920 { DDF_RAID5E
, LEVEL_UNSUPPORTED
},
921 { DDF_RAID5EE
, LEVEL_UNSUPPORTED
},
926 static int map_num1(struct num_mapping
*map
, int num
)
929 for (i
=0 ; map
[i
].num1
!= MAXINT
; i
++)
930 if (map
[i
].num1
== num
)
936 static void print_guid(char *guid
, int tstamp
)
938 /* A GUIDs are part (or all) ASCII and part binary.
939 * They tend to be space padded.
940 * We ignore trailing spaces and print numbers
941 * <0x20 and >=0x7f as \xXX
942 * Some GUIDs have a time stamp in bytes 16-19.
943 * We print that if appropriate
945 int l
= DDF_GUID_LEN
;
947 while (l
&& guid
[l
-1] == ' ')
949 for (i
=0 ; i
<l
; i
++) {
950 if (guid
[i
] >= 0x20 && guid
[i
] < 0x7f)
951 fputc(guid
[i
], stdout
);
953 fprintf(stdout
, "\\x%02x", guid
[i
]&255);
956 time_t then
= __be32_to_cpu(*(__u32
*)(guid
+16)) + DECADE
;
959 tm
= localtime(&then
);
960 strftime(tbuf
, 100, " (%D %T)",tm
);
965 static void examine_vd(int n
, struct ddf_super
*sb
, char *guid
)
967 int crl
= sb
->conf_rec_len
;
970 for (vcl
= sb
->conflist
; vcl
; vcl
= vcl
->next
) {
971 struct vd_config
*vc
= &vcl
->conf
;
973 if (calc_crc(vc
, crl
*512) != vc
->crc
)
975 if (memcmp(vc
->guid
, guid
, DDF_GUID_LEN
) != 0)
978 /* Ok, we know about this VD, let's give more details */
979 printf(" Raid Devices[%d] : %d\n", n
,
980 __be16_to_cpu(vc
->prim_elmnt_count
));
981 printf(" Chunk Size[%d] : %d sectors\n", n
,
982 1 << vc
->chunk_shift
);
983 printf(" Raid Level[%d] : %s\n", n
,
984 map_num(ddf_level
, vc
->prl
)?:"-unknown-");
985 if (vc
->sec_elmnt_count
!= 1) {
986 printf(" Secondary Position[%d] : %d of %d\n", n
,
987 vc
->sec_elmnt_seq
, vc
->sec_elmnt_count
);
988 printf(" Secondary Level[%d] : %s\n", n
,
989 map_num(ddf_sec_level
, vc
->srl
) ?: "-unknown-");
991 printf(" Device Size[%d] : %llu\n", n
,
992 __be64_to_cpu(vc
->blocks
)/2);
993 printf(" Array Size[%d] : %llu\n", n
,
994 __be64_to_cpu(vc
->array_blocks
)/2);
998 static void examine_vds(struct ddf_super
*sb
)
1000 int cnt
= __be16_to_cpu(sb
->virt
->populated_vdes
);
1002 printf(" Virtual Disks : %d\n", cnt
);
1004 for (i
=0; i
<cnt
; i
++) {
1005 struct virtual_entry
*ve
= &sb
->virt
->entries
[i
];
1006 printf(" VD GUID[%d] : ", i
); print_guid(ve
->guid
, 1);
1008 printf(" unit[%d] : %d\n", i
, __be16_to_cpu(ve
->unit
));
1009 printf(" state[%d] : %s, %s%s\n", i
,
1010 map_num(ddf_state
, ve
->state
& 7),
1011 (ve
->state
& 8) ? "Morphing, ": "",
1012 (ve
->state
& 16)? "Not Consistent" : "Consistent");
1013 printf(" init state[%d] : %s\n", i
,
1014 map_num(ddf_init_state
, ve
->init_state
&3));
1015 printf(" access[%d] : %s\n", i
,
1016 map_num(ddf_access
, (ve
->init_state
>>6) & 3));
1017 printf(" Name[%d] : %.16s\n", i
, ve
->name
);
1018 examine_vd(i
, sb
, ve
->guid
);
1020 if (cnt
) printf("\n");
1023 static void examine_pds(struct ddf_super
*sb
)
1025 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1028 printf(" Physical Disks : %d\n", cnt
);
1030 for (i
=0 ; i
<cnt
; i
++) {
1031 struct phys_disk_entry
*pd
= &sb
->phys
->entries
[i
];
1032 int type
= __be16_to_cpu(pd
->type
);
1033 int state
= __be16_to_cpu(pd
->state
);
1035 printf(" PD GUID[%d] : ", i
); print_guid(pd
->guid
, 0);
1037 printf(" ref[%d] : %08x\n", i
,
1038 __be32_to_cpu(pd
->refnum
));
1039 printf(" mode[%d] : %s%s%s%s%s\n", i
,
1040 (type
&2) ? "active":"",
1041 (type
&4) ? "Global Spare":"",
1042 (type
&8) ? "spare" : "",
1043 (type
&16)? ", foreign" : "",
1044 (type
&32)? "pass-through" : "");
1045 printf(" state[%d] : %s%s%s%s%s%s%s\n", i
,
1046 (state
&1)? "Online": "Offline",
1047 (state
&2)? ", Failed": "",
1048 (state
&4)? ", Rebuilding": "",
1049 (state
&8)? ", in-transition": "",
1050 (state
&16)? ", SMART errors": "",
1051 (state
&32)? ", Unrecovered Read Errors": "",
1052 (state
&64)? ", Missing" : "");
1053 printf(" Avail Size[%d] : %llu K\n", i
,
1054 __be64_to_cpu(pd
->config_size
)>>1);
1055 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1056 if (dl
->disk
.refnum
== pd
->refnum
) {
1057 char *dv
= map_dev(dl
->major
, dl
->minor
, 0);
1059 printf(" Device[%d] : %s\n",
1067 static void examine_super_ddf(struct supertype
*st
, char *homehost
)
1069 struct ddf_super
*sb
= st
->sb
;
1071 printf(" Magic : %08x\n", __be32_to_cpu(sb
->anchor
.magic
));
1072 printf(" Version : %.8s\n", sb
->anchor
.revision
);
1073 printf("Controller GUID : "); print_guid(sb
->controller
.guid
, 0);
1075 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1077 printf(" Seq : %08x\n", __be32_to_cpu(sb
->active
->seq
));
1078 printf(" Redundant hdr : %s\n", sb
->secondary
.magic
== DDF_HEADER_MAGIC
1084 static void brief_examine_super_ddf(struct supertype
*st
)
1086 /* We just write a generic DDF ARRAY entry
1087 * The uuid is all hex, 6 groups of 4 bytes
1089 struct ddf_super
*ddf
= st
->sb
;
1091 printf("ARRAY /dev/ddf UUID=");
1092 for (i
= 0; i
< DDF_GUID_LEN
; i
++) {
1093 printf("%02x", ddf
->anchor
.guid
[i
]);
1094 if ((i
&3) == 0 && i
!= 0)
1100 static void detail_super_ddf(struct supertype
*st
, char *homehost
)
1103 * Could print DDF GUID
1104 * Need to find which array
1105 * If whole, briefly list all arrays
1110 static void brief_detail_super_ddf(struct supertype
*st
)
1112 /* FIXME I really need to know which array we are detailing.
1113 * Can that be stored in ddf_super??
1115 // struct ddf_super *ddf = st->sb;
1121 static int match_home_ddf(struct supertype
*st
, char *homehost
)
1123 /* It matches 'this' host if the controller is a
1124 * Linux-MD controller with vendor_data matching
1127 struct ddf_super
*ddf
= st
->sb
;
1128 int len
= strlen(homehost
);
1130 return (memcmp(ddf
->controller
.guid
, T10
, 8) == 0 &&
1131 len
< sizeof(ddf
->controller
.vendor_data
) &&
1132 memcmp(ddf
->controller
.vendor_data
, homehost
,len
) == 0 &&
1133 ddf
->controller
.vendor_data
[len
] == 0);
1136 static struct vd_config
*find_vdcr(struct ddf_super
*ddf
, int inst
)
1139 if (inst
< 0 || inst
> __be16_to_cpu(ddf
->virt
->populated_vdes
))
1141 for (v
= ddf
->conflist
; v
; v
= v
->next
)
1142 if (memcmp(v
->conf
.guid
,
1143 ddf
->virt
->entries
[inst
].guid
,
1149 static int find_phys(struct ddf_super
*ddf
, __u32 phys_refnum
)
1151 /* Find the entry in phys_disk which has the given refnum
1152 * and return it's index
1155 for (i
=0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
); i
++)
1156 if (ddf
->phys
->entries
[i
].refnum
== phys_refnum
)
1161 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4])
1163 /* The uuid returned here is used for:
1164 * uuid to put into bitmap file (Create, Grow)
1165 * uuid for backup header when saving critical section (Grow)
1166 * comparing uuids when re-adding a device into an array
1167 * For each of these we can make do with a truncated
1168 * or hashed uuid rather than the original, as long as
1170 * In each case the uuid required is that of the data-array,
1171 * not the device-set.
1172 * In the case of SVD we assume the BVD is of interest,
1173 * though that might be the case if a bitmap were made for
1174 * a mirrored SVD - worry about that later.
1175 * So we need to find the VD configuration record for the
1176 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1177 * The first 16 bytes of the sha1 of these is used.
1179 struct ddf_super
*ddf
= st
->sb
;
1180 struct vd_config
*vd
= find_vdcr(ddf
, ddf
->conf_num
);
1183 memset(uuid
, 0, sizeof (uuid
));
1186 struct sha1_ctx ctx
;
1187 sha1_init_ctx(&ctx
);
1188 sha1_process_bytes(&vd
->guid
, DDF_GUID_LEN
, &ctx
);
1189 if (vd
->sec_elmnt_count
> 1)
1190 sha1_process_bytes(&vd
->sec_elmnt_seq
, 1, &ctx
);
1191 sha1_finish_ctx(&ctx
, buf
);
1192 memcpy(uuid
, buf
, sizeof(uuid
));
1196 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
)
1198 struct ddf_super
*ddf
= st
->sb
;
1201 info
->array
.major_version
= 1000;
1202 info
->array
.minor_version
= 0; /* FIXME use ddf->revision somehow */
1203 info
->array
.patch_version
= 0;
1204 info
->array
.raid_disks
= __be16_to_cpu(ddf
->phys
->used_pdes
);
1205 info
->array
.level
= LEVEL_CONTAINER
;
1206 info
->array
.layout
= 0;
1207 info
->array
.md_minor
= -1;
1208 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*(__u32
*)
1209 (ddf
->anchor
.guid
+16));
1210 info
->array
.utime
= 0;
1211 info
->array
.chunk_size
= 0;
1213 // info->data_offset = ???;
1214 // info->component_size = ???;
1216 info
->disk
.major
= 0;
1217 info
->disk
.minor
= 0;
1219 info
->disk
.number
= __be32_to_cpu(ddf
->dlist
->disk
.refnum
);
1220 info
->disk
.raid_disk
= -1;
1221 for (i
= 0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
) ; i
++)
1222 if (ddf
->phys
->entries
[i
].refnum
==
1223 ddf
->dlist
->disk
.refnum
) {
1224 info
->disk
.raid_disk
= i
;
1228 info
->disk
.number
= -1;
1229 // info->disk.raid_disk = find refnum in the table and use index;
1231 info
->disk
.state
= (1 << MD_DISK_SYNC
);
1233 info
->reshape_active
= 0;
1235 strcpy(info
->text_version
, "ddf");
1237 // uuid_from_super_ddf(info->uuid, sbv);
1239 // info->name[] ?? ;
1242 static void getinfo_super_n_container(struct supertype
*st
, struct mdinfo
*info
)
1244 /* just need offset and size */
1245 struct ddf_super
*ddf
= st
->sb
;
1246 int n
= info
->disk
.number
;
1248 info
->data_offset
= __be64_to_cpu(ddf
->phys
->entries
[n
].config_size
);
1249 info
->component_size
= 32*1024*1024 / 512;
1252 static int rlq_to_layout(int rlq
, int prl
, int raiddisks
);
1254 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
)
1256 struct ddf_super
*ddf
= st
->sb
;
1257 struct vd_config
*vd
= find_vdcr(ddf
, info
->container_member
);
1259 /* FIXME this returns BVD info - what if we want SVD ?? */
1261 info
->array
.major_version
= 1000;
1262 info
->array
.minor_version
= 0; /* FIXME use ddf->revision somehow */
1263 info
->array
.patch_version
= 0;
1264 info
->array
.raid_disks
= __be16_to_cpu(vd
->prim_elmnt_count
);
1265 info
->array
.level
= map_num1(ddf_level_num
, vd
->prl
);
1266 info
->array
.layout
= rlq_to_layout(vd
->rlq
, vd
->prl
,
1267 info
->array
.raid_disks
);
1268 info
->array
.md_minor
= -1;
1269 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*(__u32
*)(vd
->guid
+16));
1270 info
->array
.utime
= DECADE
+ __be32_to_cpu(vd
->timestamp
);
1271 info
->array
.chunk_size
= 512 << vd
->chunk_shift
;
1273 // info->data_offset = ???;
1274 // info->component_size = ???;
1276 info
->disk
.major
= 0;
1277 info
->disk
.minor
= 0;
1278 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1279 // info->disk.raid_disk = find refnum in the table and use index;
1280 // info->disk.state = ???;
1282 uuid_from_super_ddf(st
, info
->uuid
);
1284 sprintf(info
->text_version
, "/%s/%d",
1285 devnum2devname(st
->container_dev
),
1286 info
->container_member
);
1288 // info->name[] ?? ;
1291 static void getinfo_super_n_bvd(struct supertype
*st
, struct mdinfo
*info
)
1293 /* Find the particular details for info->disk.raid_disk.
1294 * This includes data_offset, component_size,
1296 struct ddf_super
*ddf
= st
->sb
;
1297 __u64
*lba_offset
= ddf
->newconf
->lba_offset
;
1298 struct vd_config
*conf
= &ddf
->newconf
->conf
;
1299 info
->data_offset
= __be64_to_cpu(lba_offset
[info
->disk
.raid_disk
]);
1300 info
->component_size
= __be64_to_cpu(conf
->blocks
);
1303 static int update_super_ddf(struct supertype
*st
, struct mdinfo
*info
,
1305 char *devname
, int verbose
,
1306 int uuid_set
, char *homehost
)
1308 /* For 'assemble' and 'force' we need to return non-zero if any
1309 * change was made. For others, the return value is ignored.
1310 * Update options are:
1311 * force-one : This device looks a bit old but needs to be included,
1312 * update age info appropriately.
1313 * assemble: clear any 'faulty' flag to allow this device to
1315 * force-array: Array is degraded but being forced, mark it clean
1316 * if that will be needed to assemble it.
1318 * newdev: not used ????
1319 * grow: Array has gained a new device - this is currently for
1321 * resync: mark as dirty so a resync will happen.
1322 * uuid: Change the uuid of the array to match watch is given
1323 * homehost: update the recorded homehost
1324 * name: update the name - preserving the homehost
1325 * _reshape_progress: record new reshape_progress position.
1327 * Following are not relevant for this version:
1328 * sparc2.2 : update from old dodgey metadata
1329 * super-minor: change the preferred_minor number
1330 * summaries: update redundant counters.
1333 // struct ddf_super *ddf = st->sb;
1334 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1335 // struct virtual_entry *ve = find_ve(ddf);
1338 /* we don't need to handle "force-*" or "assemble" as
1339 * there is no need to 'trick' the kernel. We the metadata is
1340 * first updated to activate the array, all the implied modifications
1344 if (strcmp(update
, "grow") == 0) {
1347 if (strcmp(update
, "resync") == 0) {
1348 // info->resync_checkpoint = 0;
1350 /* We ignore UUID updates as they make even less sense
1353 if (strcmp(update
, "homehost") == 0) {
1354 /* homehost is stored in controller->vendor_data,
1355 * or it is when we are the vendor
1357 // if (info->vendor_is_local)
1358 // strcpy(ddf->controller.vendor_data, homehost);
1360 if (strcmp(update
, "name") == 0) {
1361 /* name is stored in virtual_entry->name */
1362 // memset(ve->name, ' ', 16);
1363 // strncpy(ve->name, info->name, 16);
1365 if (strcmp(update
, "_reshape_progress") == 0) {
1366 /* We don't support reshape yet */
1369 // update_all_csum(ddf);
1374 static void make_header_guid(char *guid
)
1378 /* Create a DDF Header of Virtual Disk GUID */
1380 /* 24 bytes of fiction required.
1381 * first 8 are a 'vendor-id' - "Linux-MD"
1382 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1383 * Remaining 8 random number plus timestamp
1385 memcpy(guid
, T10
, sizeof(T10
));
1386 stamp
= __cpu_to_be32(0xdeadbeef);
1387 memcpy(guid
+8, &stamp
, 4);
1388 stamp
= __cpu_to_be32(0);
1389 memcpy(guid
+12, &stamp
, 4);
1390 stamp
= __cpu_to_be32(time(0) - DECADE
);
1391 memcpy(guid
+16, &stamp
, 4);
1392 rfd
= open("/dev/urandom", O_RDONLY
);
1393 if (rfd
< 0 || read(rfd
, &stamp
, 4) != 4)
1395 memcpy(guid
+20, &stamp
, 4);
1396 if (rfd
>= 0) close(rfd
);
1398 static int init_super_ddf(struct supertype
*st
,
1399 mdu_array_info_t
*info
,
1400 unsigned long long size
, char *name
, char *homehost
,
1403 /* This is primarily called by Create when creating a new array.
1404 * We will then get add_to_super called for each component, and then
1405 * write_init_super called to write it out to each device.
1406 * For DDF, Create can create on fresh devices or on a pre-existing
1408 * To create on a pre-existing array a different method will be called.
1409 * This one is just for fresh drives.
1411 * We need to create the entire 'ddf' structure which includes:
1412 * DDF headers - these are easy.
1413 * Controller data - a Sector describing this controller .. not that
1414 * this is a controller exactly.
1415 * Physical Disk Record - one entry per device, so
1416 * leave plenty of space.
1417 * Virtual Disk Records - again, just leave plenty of space.
1418 * This just lists VDs, doesn't give details
1419 * Config records - describes the VDs that use this disk
1420 * DiskData - describes 'this' device.
1421 * BadBlockManagement - empty
1422 * Diag Space - empty
1423 * Vendor Logs - Could we put bitmaps here?
1426 struct ddf_super
*ddf
;
1429 int max_phys_disks
, max_virt_disks
;
1430 unsigned long long sector
;
1434 struct phys_disk
*pd
;
1435 struct virtual_disk
*vd
;
1437 ddf
= malloc(sizeof(*ddf
));
1438 ddf
->dlist
= NULL
; /* no physical disks yet */
1439 ddf
->conflist
= NULL
; /* No virtual disks yet */
1441 /* At least 32MB *must* be reserved for the ddf. So let's just
1442 * start 32MB from the end, and put the primary header there.
1443 * Don't do secondary for now.
1444 * We don't know exactly where that will be yet as it could be
1445 * different on each device. To just set up the lengths.
1449 ddf
->anchor
.magic
= DDF_HEADER_MAGIC
;
1450 make_header_guid(ddf
->anchor
.guid
);
1452 memcpy(ddf
->anchor
.revision
, DDF_REVISION
, 8);
1453 ddf
->anchor
.seq
= __cpu_to_be32(1);
1454 ddf
->anchor
.timestamp
= __cpu_to_be32(time(0) - DECADE
);
1455 ddf
->anchor
.openflag
= 0xFF;
1456 ddf
->anchor
.foreignflag
= 0;
1457 ddf
->anchor
.enforcegroups
= 0; /* Is this best?? */
1458 ddf
->anchor
.pad0
= 0xff;
1459 memset(ddf
->anchor
.pad1
, 0xff, 12);
1460 memset(ddf
->anchor
.header_ext
, 0xff, 32);
1461 ddf
->anchor
.primary_lba
= ~(__u64
)0;
1462 ddf
->anchor
.secondary_lba
= ~(__u64
)0;
1463 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
1464 memset(ddf
->anchor
.pad2
, 0xff, 3);
1465 ddf
->anchor
.workspace_len
= __cpu_to_be32(32768); /* Must be reserved */
1466 ddf
->anchor
.workspace_lba
= ~(__u64
)0; /* Put this at bottom
1467 of 32M reserved.. */
1468 max_phys_disks
= 1023; /* Should be enough */
1469 ddf
->anchor
.max_pd_entries
= __cpu_to_be16(max_phys_disks
);
1470 max_virt_disks
= 255;
1471 ddf
->anchor
.max_vd_entries
= __cpu_to_be16(max_virt_disks
); /* ?? */
1472 ddf
->anchor
.max_partitions
= __cpu_to_be16(64); /* ?? */
1474 ddf
->conf_rec_len
= 1 + 256 * 12 / 512;
1475 ddf
->anchor
.config_record_len
= __cpu_to_be16(ddf
->conf_rec_len
);
1476 ddf
->anchor
.max_primary_element_entries
= __cpu_to_be16(256);
1478 memset(ddf
->anchor
.pad3
, 0xff, 54);
1480 /* controller sections is one sector long immediately
1481 * after the ddf header */
1483 ddf
->anchor
.controller_section_offset
= __cpu_to_be32(sector
);
1484 ddf
->anchor
.controller_section_length
= __cpu_to_be32(1);
1487 /* phys is 8 sectors after that */
1488 pdsize
= ROUND_UP(sizeof(struct phys_disk
) +
1489 sizeof(struct phys_disk_entry
)*max_phys_disks
,
1491 switch(pdsize
/512) {
1492 case 2: case 8: case 32: case 128: case 512: break;
1495 ddf
->anchor
.phys_section_offset
= __cpu_to_be32(sector
);
1496 ddf
->anchor
.phys_section_length
=
1497 __cpu_to_be32(pdsize
/512); /* max_primary_element_entries/8 */
1498 sector
+= pdsize
/512;
1500 /* virt is another 32 sectors */
1501 vdsize
= ROUND_UP(sizeof(struct virtual_disk
) +
1502 sizeof(struct virtual_entry
) * max_virt_disks
,
1504 switch(vdsize
/512) {
1505 case 2: case 8: case 32: case 128: case 512: break;
1508 ddf
->anchor
.virt_section_offset
= __cpu_to_be32(sector
);
1509 ddf
->anchor
.virt_section_length
=
1510 __cpu_to_be32(vdsize
/512); /* max_vd_entries/8 */
1511 sector
+= vdsize
/512;
1513 clen
= (1 + 256*12/512) * (64+1);
1514 ddf
->anchor
.config_section_offset
= __cpu_to_be32(sector
);
1515 ddf
->anchor
.config_section_length
= __cpu_to_be32(clen
);
1518 ddf
->anchor
.data_section_offset
= __cpu_to_be32(sector
);
1519 ddf
->anchor
.data_section_length
= __cpu_to_be32(1);
1522 ddf
->anchor
.bbm_section_length
= __cpu_to_be32(0);
1523 ddf
->anchor
.bbm_section_offset
= __cpu_to_be32(0xFFFFFFFF);
1524 ddf
->anchor
.diag_space_length
= __cpu_to_be32(0);
1525 ddf
->anchor
.diag_space_offset
= __cpu_to_be32(0xFFFFFFFF);
1526 ddf
->anchor
.vendor_length
= __cpu_to_be32(0);
1527 ddf
->anchor
.vendor_offset
= __cpu_to_be32(0xFFFFFFFF);
1529 memset(ddf
->anchor
.pad4
, 0xff, 256);
1531 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
1532 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
1534 ddf
->primary
.openflag
= 1; /* I guess.. */
1535 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
1537 ddf
->secondary
.openflag
= 1; /* I guess.. */
1538 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
1540 ddf
->active
= &ddf
->primary
;
1542 ddf
->controller
.magic
= DDF_CONTROLLER_MAGIC
;
1544 /* 24 more bytes of fiction required.
1545 * first 8 are a 'vendor-id' - "Linux-MD"
1546 * Remaining 16 are serial number.... maybe a hostname would do?
1548 memcpy(ddf
->controller
.guid
, T10
, sizeof(T10
));
1549 gethostname(hostname
, 17);
1551 hostlen
= strlen(hostname
);
1552 memcpy(ddf
->controller
.guid
+ 24 - hostlen
, hostname
, hostlen
);
1553 for (i
= strlen(T10
) ; i
+hostlen
< 24; i
++)
1554 ddf
->controller
.guid
[i
] = ' ';
1556 ddf
->controller
.type
.vendor_id
= __cpu_to_be16(0xDEAD);
1557 ddf
->controller
.type
.device_id
= __cpu_to_be16(0xBEEF);
1558 ddf
->controller
.type
.sub_vendor_id
= 0;
1559 ddf
->controller
.type
.sub_device_id
= 0;
1560 memcpy(ddf
->controller
.product_id
, "What Is My PID??", 16);
1561 memset(ddf
->controller
.pad
, 0xff, 8);
1562 memset(ddf
->controller
.vendor_data
, 0xff, 448);
1564 pd
= ddf
->phys
= malloc(pdsize
);
1565 ddf
->pdsize
= pdsize
;
1567 memset(pd
, 0xff, pdsize
);
1568 memset(pd
, 0, sizeof(*pd
));
1569 pd
->magic
= DDF_PHYS_DATA_MAGIC
;
1570 pd
->used_pdes
= __cpu_to_be16(0);
1571 pd
->max_pdes
= __cpu_to_be16(max_phys_disks
);
1572 memset(pd
->pad
, 0xff, 52);
1574 vd
= ddf
->virt
= malloc(vdsize
);
1575 ddf
->vdsize
= vdsize
;
1576 memset(vd
, 0, vdsize
);
1577 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
1578 vd
->populated_vdes
= __cpu_to_be16(0);
1579 vd
->max_vdes
= __cpu_to_be16(max_virt_disks
);
1580 memset(vd
->pad
, 0xff, 52);
1582 for (i
=0; i
<max_virt_disks
; i
++)
1583 memset(&vd
->entries
[i
], 0xff, sizeof(struct virtual_entry
));
1589 static int all_ff(char *guid
)
1592 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
1593 if (guid
[i
] != (char)0xff)
1597 static int chunk_to_shift(int chunksize
)
1599 return ffs(chunksize
/512)-1;
1602 static int level_to_prl(int level
)
1605 case LEVEL_LINEAR
: return DDF_CONCAT
;
1606 case 0: return DDF_RAID0
;
1607 case 1: return DDF_RAID1
;
1608 case 4: return DDF_RAID4
;
1609 case 5: return DDF_RAID5
;
1610 case 6: return DDF_RAID6
;
1614 static int layout_to_rlq(int level
, int layout
, int raiddisks
)
1618 return DDF_RAID0_SIMPLE
;
1621 case 2: return DDF_RAID1_SIMPLE
;
1622 case 3: return DDF_RAID1_MULTI
;
1627 case 0: return DDF_RAID4_N
;
1633 case ALGORITHM_LEFT_ASYMMETRIC
:
1634 return DDF_RAID5_N_RESTART
;
1635 case ALGORITHM_RIGHT_ASYMMETRIC
:
1636 return DDF_RAID5_0_RESTART
;
1637 case ALGORITHM_LEFT_SYMMETRIC
:
1638 return DDF_RAID5_N_CONTINUE
;
1639 case ALGORITHM_RIGHT_SYMMETRIC
:
1640 return -1; /* not mentioned in standard */
1646 static int rlq_to_layout(int rlq
, int prl
, int raiddisks
)
1650 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1652 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1660 return -1; /* FIXME this isn't checked */
1665 case DDF_RAID5_N_RESTART
:
1666 return ALGORITHM_LEFT_ASYMMETRIC
;
1667 case DDF_RAID5_0_RESTART
:
1668 return ALGORITHM_RIGHT_ASYMMETRIC
;
1669 case DDF_RAID5_N_CONTINUE
:
1670 return ALGORITHM_LEFT_SYMMETRIC
;
1678 static int init_super_ddf_bvd(struct supertype
*st
,
1679 mdu_array_info_t
*info
,
1680 unsigned long long size
,
1681 char *name
, char *homehost
,
1684 /* We are creating a BVD inside a pre-existing container.
1685 * so st->sb is already set.
1686 * We need to create a new vd_config and a new virtual_entry
1688 struct ddf_super
*ddf
= st
->sb
;
1690 struct virtual_entry
*ve
;
1692 struct vd_config
*vc
;
1694 if (__be16_to_cpu(ddf
->virt
->populated_vdes
)
1695 >= __be16_to_cpu(ddf
->virt
->max_vdes
)) {
1696 fprintf(stderr
, Name
": This ddf already has the "
1697 "maximum of %d virtual devices\n",
1698 __be16_to_cpu(ddf
->virt
->max_vdes
));
1702 for (venum
= 0; venum
< __be16_to_cpu(ddf
->virt
->max_vdes
); venum
++)
1703 if (all_ff(ddf
->virt
->entries
[venum
].guid
))
1705 if (venum
== __be16_to_cpu(ddf
->virt
->max_vdes
)) {
1706 fprintf(stderr
, Name
": Cannot find spare slot for "
1707 "virtual disk - DDF is corrupt\n");
1710 ve
= &ddf
->virt
->entries
[venum
];
1711 ddf
->conf_num
= venum
;
1713 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1714 * timestamp, random number
1716 make_header_guid(ve
->guid
);
1717 ve
->unit
= __cpu_to_be16(info
->md_minor
);
1719 ve
->guid_crc
= crc32(0, (unsigned char*)ddf
->anchor
.guid
, DDF_GUID_LEN
);
1721 ve
->state
= DDF_state_degraded
; /* Will be modified as devices are added */
1722 if (info
->state
& 1) /* clean */
1723 ve
->init_state
= DDF_init_full
;
1725 ve
->init_state
= DDF_init_not
;
1727 memset(ve
->pad1
, 0xff, 14);
1728 memset(ve
->name
, ' ', 16);
1730 strncpy(ve
->name
, name
, 16);
1731 ddf
->virt
->populated_vdes
=
1732 __cpu_to_be16(__be16_to_cpu(ddf
->virt
->populated_vdes
)+1);
1734 /* Now create a new vd_config */
1735 vcl
= malloc(offsetof(struct vcl
, conf
) + ddf
->conf_rec_len
* 512);
1736 vcl
->lba_offset
= (__u64
*) &vcl
->conf
.phys_refnum
[ddf
->mppe
];
1740 vc
->magic
= DDF_VD_CONF_MAGIC
;
1741 memcpy(vc
->guid
, ve
->guid
, DDF_GUID_LEN
);
1742 vc
->timestamp
= __cpu_to_be32(time(0)-DECADE
);
1743 vc
->seqnum
= __cpu_to_be32(1);
1744 memset(vc
->pad0
, 0xff, 24);
1745 vc
->prim_elmnt_count
= __cpu_to_be16(info
->raid_disks
);
1746 vc
->chunk_shift
= chunk_to_shift(info
->chunk_size
);
1747 vc
->prl
= level_to_prl(info
->level
);
1748 vc
->rlq
= layout_to_rlq(info
->level
, info
->layout
, info
->raid_disks
);
1749 vc
->sec_elmnt_count
= 1;
1750 vc
->sec_elmnt_seq
= 0;
1752 vc
->blocks
= __cpu_to_be64(info
->size
* 2);
1753 vc
->array_blocks
= __cpu_to_be64(
1754 calc_array_size(info
->level
, info
->raid_disks
, info
->layout
,
1755 info
->chunk_size
, info
->size
*2));
1756 memset(vc
->pad1
, 0xff, 8);
1757 vc
->spare_refs
[0] = 0xffffffff;
1758 vc
->spare_refs
[1] = 0xffffffff;
1759 vc
->spare_refs
[2] = 0xffffffff;
1760 vc
->spare_refs
[3] = 0xffffffff;
1761 vc
->spare_refs
[4] = 0xffffffff;
1762 vc
->spare_refs
[5] = 0xffffffff;
1763 vc
->spare_refs
[6] = 0xffffffff;
1764 vc
->spare_refs
[7] = 0xffffffff;
1765 memset(vc
->cache_pol
, 0, 8);
1767 memset(vc
->pad2
, 0xff, 3);
1768 memset(vc
->pad3
, 0xff, 52);
1769 memset(vc
->pad4
, 0xff, 192);
1770 memset(vc
->v0
, 0xff, 32);
1771 memset(vc
->v1
, 0xff, 32);
1772 memset(vc
->v2
, 0xff, 16);
1773 memset(vc
->v3
, 0xff, 16);
1774 memset(vc
->vendor
, 0xff, 32);
1776 memset(vc
->phys_refnum
, 0xff, 4*ddf
->mppe
);
1777 memset(vc
->phys_refnum
+(ddf
->mppe
* 4), 0x00, 8*ddf
->mppe
);
1779 vcl
->next
= ddf
->conflist
;
1780 ddf
->conflist
= vcl
;
1785 static void add_to_super_ddf_bvd(struct supertype
*st
,
1786 mdu_disk_info_t
*dk
, int fd
, char *devname
)
1788 /* fd and devname identify a device with-in the ddf container (st).
1789 * dk identifies a location in the new BVD.
1790 * We need to find suitable free space in that device and update
1791 * the phys_refnum and lba_offset for the newly created vd_config.
1792 * We might also want to update the type in the phys_disk
1796 struct ddf_super
*ddf
= st
->sb
;
1797 struct vd_config
*vc
;
1801 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
1802 if (dl
->major
== dk
->major
&&
1803 dl
->minor
== dk
->minor
)
1805 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1808 vc
= &ddf
->newconf
->conf
;
1809 vc
->phys_refnum
[dk
->raid_disk
] = dl
->disk
.refnum
;
1810 lba_offset
= (__u64
*)(vc
->phys_refnum
+ ddf
->mppe
);
1811 lba_offset
[dk
->raid_disk
] = 0; /* FIXME */
1813 dl
->vlist
[0] = ddf
->newconf
; /* FIXME */
1816 dl
->devname
= devname
;
1818 /* Check how many working raid_disks, and if we can mark
1819 * array as optimal yet
1823 for (i
=0; i
< __be16_to_cpu(vc
->prim_elmnt_count
); i
++)
1824 if (vc
->phys_refnum
[i
] != 0xffffffff)
1826 if (working
== __be16_to_cpu(vc
->prim_elmnt_count
))
1827 ->entries
[xx
].state
= (->entries
[xx
].state
& ~DDF_state_mask
)
1828 | DDF_state_optimal
;
1830 if (vc
->prl
== DDF_RAID6
&&
1831 working
+1 == __be16_to_cpu(vc
->prim_elmnt_count
))
1832 ->entries
[xx
].state
= (->entries
[xx
].state
& ~DDF_state_mask
)
1833 | DDF_state_part_optimal
;
1837 /* add a device to a container, either while creating it or while
1838 * expanding a pre-existing container
1840 static void add_to_super_ddf(struct supertype
*st
,
1841 mdu_disk_info_t
*dk
, int fd
, char *devname
)
1843 struct ddf_super
*ddf
= st
->sb
;
1847 unsigned long long size
;
1848 struct phys_disk_entry
*pde
;
1852 /* This is device numbered dk->number. We need to create
1853 * a phys_disk entry and a more detailed disk_data entry.
1856 dd
= malloc(sizeof(*dd
) + sizeof(dd
->vlist
[0]) * ddf
->max_part
);
1857 dd
->major
= major(stb
.st_rdev
);
1858 dd
->minor
= minor(stb
.st_rdev
);
1859 dd
->devname
= devname
;
1860 dd
->next
= ddf
->dlist
;
1864 dd
->disk
.magic
= DDF_PHYS_DATA_MAGIC
;
1866 tm
= localtime(&now
);
1867 sprintf(dd
->disk
.guid
, "%8s%04d%02d%02d",
1868 T10
, tm
->tm_year
+1900, tm
->tm_mon
+1, tm
->tm_mday
);
1869 *(__u32
*)(dd
->disk
.guid
+ 16) = random();
1870 *(__u32
*)(dd
->disk
.guid
+ 20) = random();
1872 dd
->disk
.refnum
= random(); /* and hope for the best FIXME check this is unique!!*/
1873 dd
->disk
.forced_ref
= 1;
1874 dd
->disk
.forced_guid
= 1;
1875 memset(dd
->disk
.vendor
, ' ', 32);
1876 memcpy(dd
->disk
.vendor
, "Linux", 5);
1877 memset(dd
->disk
.pad
, 0xff, 442);
1878 for (i
= 0; i
< ddf
->max_part
; i
++)
1879 dd
->vlist
[i
] = NULL
;
1881 n
= __be16_to_cpu(ddf
->phys
->used_pdes
);
1882 pde
= &ddf
->phys
->entries
[n
];
1884 ddf
->phys
->used_pdes
= __cpu_to_be16(n
);
1886 memcpy(pde
->guid
, dd
->disk
.guid
, DDF_GUID_LEN
);
1887 pde
->refnum
= dd
->disk
.refnum
;
1888 pde
->type
= __cpu_to_be16(DDF_Forced_PD_GUID
|DDF_Global_Spare
);
1889 pde
->state
= __cpu_to_be16(DDF_Online
);
1890 get_dev_size(fd
, NULL
, &size
);
1891 /* We are required to reserve 32Meg, and record the size in sectors */
1892 pde
->config_size
= __cpu_to_be64( (size
- 32*1024*1024) / 512);
1893 sprintf(pde
->path
, "%17.17s","Information: nil") ;
1894 memset(pde
->pad
, 0xff, 6);
1900 * This is the write_init_super method for a ddf container. It is
1901 * called when creating a container or adding another device to a
1906 static int __write_init_super_ddf(struct supertype
*st
, int do_close
)
1909 struct ddf_super
*ddf
= st
->sb
;
1915 unsigned long long size
, sector
;
1917 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
1923 /* We need to fill in the primary, (secondary) and workspace
1924 * lba's in the headers, set their checksums,
1925 * Also checksum phys, virt....
1927 * Then write everything out, finally the anchor is written.
1929 get_dev_size(fd
, NULL
, &size
);
1931 ddf
->anchor
.workspace_lba
= __cpu_to_be64(size
- 32*1024*2);
1932 ddf
->anchor
.primary_lba
= __cpu_to_be64(size
- 16*1024*2);
1933 ddf
->anchor
.seq
= __cpu_to_be32(1);
1934 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
1935 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
1937 ddf
->anchor
.openflag
= 0xFF; /* 'open' means nothing */
1938 ddf
->anchor
.seq
= 0xFFFFFFFF; /* no sequencing in anchor */
1939 ddf
->anchor
.crc
= calc_crc(&ddf
->anchor
, 512);
1941 ddf
->primary
.openflag
= 0;
1942 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
1944 ddf
->secondary
.openflag
= 0;
1945 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
1947 ddf
->primary
.crc
= calc_crc(&ddf
->primary
, 512);
1948 ddf
->secondary
.crc
= calc_crc(&ddf
->secondary
, 512);
1950 sector
= size
- 16*1024*2;
1951 lseek64(fd
, sector
<<9, 0);
1952 write(fd
, &ddf
->primary
, 512);
1954 ddf
->controller
.crc
= calc_crc(&ddf
->controller
, 512);
1955 write(fd
, &ddf
->controller
, 512);
1957 ddf
->phys
->crc
= calc_crc(ddf
->phys
, ddf
->pdsize
);
1959 write(fd
, ddf
->phys
, ddf
->pdsize
);
1961 ddf
->virt
->crc
= calc_crc(ddf
->virt
, ddf
->vdsize
);
1962 write(fd
, ddf
->virt
, ddf
->vdsize
);
1964 /* Now write lots of config records. */
1965 n_config
= ddf
->max_part
;
1966 conf_size
= ddf
->conf_rec_len
* 512;
1967 for (i
= 0 ; i
<= n_config
; i
++) {
1968 struct vcl
*c
= d
->vlist
[i
];
1970 c
= (struct vcl
*)d
->spare
;
1973 c
->conf
.crc
= calc_crc(&c
->conf
, conf_size
);
1974 write(fd
, &c
->conf
, conf_size
);
1976 __u32 sig
= 0xffffffff;
1978 lseek64(fd
, conf_size
-4, SEEK_CUR
);
1981 d
->disk
.crc
= calc_crc(&d
->disk
, 512);
1982 write(fd
, &d
->disk
, 512);
1984 /* Maybe do the same for secondary */
1986 lseek64(fd
, (size
-1)*512, SEEK_SET
);
1987 write(fd
, &ddf
->anchor
, 512);
1996 static int write_init_super_ddf(struct supertype
*st
)
1998 return __write_init_super_ddf(st
, 1);
2003 static __u64
avail_size_ddf(struct supertype
*st
, __u64 devsize
)
2005 /* We must reserve the last 32Meg */
2006 if (devsize
<= 32*1024*2)
2008 return devsize
- 32*1024*2;
2012 int validate_geometry_ddf(struct supertype
*st
,
2013 int level
, int layout
, int raiddisks
,
2014 int chunk
, unsigned long long size
,
2015 char *dev
, unsigned long long *freesize
)
2021 /* ddf potentially supports lots of things, but it depends on
2022 * what devices are offered (and maybe kernel version?)
2023 * If given unused devices, we will make a container.
2024 * If given devices in a container, we will make a BVD.
2025 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2028 if (level
== LEVEL_CONTAINER
) {
2029 st
->ss
= &super_ddf_container
;
2031 int rv
=st
->ss
->validate_geometry(st
, level
, layout
,
2038 return st
->ss
->validate_geometry(st
, level
, layout
, raiddisks
,
2039 chunk
, size
, dev
, freesize
);
2043 /* creating in a given container */
2044 st
->ss
= &super_ddf_bvd
;
2046 int rv
=st
->ss
->validate_geometry(st
, level
, layout
,
2053 return st
->ss
->validate_geometry(st
, level
, layout
, raiddisks
,
2054 chunk
, size
, dev
, freesize
);
2056 /* FIXME should exclude MULTIPATH, or more appropriately, allow
2057 * only known levels.
2062 /* This device needs to be either a device in a 'ddf' container,
2063 * or it needs to be a 'ddf-bvd' array.
2066 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2068 sra
= sysfs_read(fd
, 0, GET_VERSION
);
2070 if (sra
&& sra
->array
.major_version
== -1 &&
2071 strcmp(sra
->text_version
, "ddf-bvd") == 0) {
2072 st
->ss
= &super_ddf_svd
;
2073 return st
->ss
->validate_geometry(st
, level
, layout
,
2074 raiddisks
, chunk
, size
,
2079 Name
": Cannot create this array on device %s\n",
2083 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2084 fprintf(stderr
, Name
": Cannot open %s: %s\n",
2085 dev
, strerror(errno
));
2088 /* Well, it is in use by someone, maybe a 'ddf' container. */
2089 cfd
= open_container(fd
);
2092 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
2096 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2098 if (sra
&& sra
->array
.major_version
== -1 &&
2099 strcmp(sra
->text_version
, "ddf") == 0) {
2100 /* This is a member of a ddf container. Load the container
2101 * and try to create a bvd
2103 struct ddf_super
*ddf
;
2104 st
->ss
= &super_ddf_bvd
;
2105 if (load_super_ddf_all(st
, cfd
, (void **)&ddf
, NULL
, 1) == 0) {
2107 st
->container_dev
= fd2devnum(cfd
);
2109 return st
->ss
->validate_geometry(st
, level
, layout
,
2110 raiddisks
, chunk
, size
,
2115 fprintf(stderr
, Name
": Cannot use %s: Already in use\n",
2120 int validate_geometry_ddf_container(struct supertype
*st
,
2121 int level
, int layout
, int raiddisks
,
2122 int chunk
, unsigned long long size
,
2123 char *dev
, unsigned long long *freesize
)
2126 unsigned long long ldsize
;
2128 if (level
!= LEVEL_CONTAINER
)
2133 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2135 fprintf(stderr
, Name
": Cannot open %s: %s\n",
2136 dev
, strerror(errno
));
2139 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2145 *freesize
= avail_size_ddf(st
, ldsize
);
2151 unsigned long long start
, size
;
2153 int cmp_extent(const void *av
, const void *bv
)
2155 const struct extent
*a
= av
;
2156 const struct extent
*b
= bv
;
2157 if (a
->start
< b
->start
)
2159 if (a
->start
> b
->start
)
2164 struct extent
*get_extents(struct ddf_super
*ddf
, struct dl
*dl
)
2166 /* find a list of used extents on the give physical device
2167 * (dnum) or the given ddf.
2168 * Return a malloced array of 'struct extent'
2170 FIXME ignore DDF_Legacy devices?
2178 for (dnum
= 0; dnum
< ddf
->phys
->used_pdes
; dnum
++)
2179 if (memcmp(dl
->disk
.guid
,
2180 ddf
->phys
->entries
[dnum
].guid
,
2184 if (dnum
== ddf
->phys
->used_pdes
)
2187 rv
= malloc(sizeof(struct extent
) * (ddf
->max_part
+ 2));
2191 for (i
= 0; i
< ddf
->max_part
; i
++) {
2192 struct vcl
*v
= dl
->vlist
[i
];
2195 for (j
=0; j
< v
->conf
.prim_elmnt_count
; j
++)
2196 if (v
->conf
.phys_refnum
[j
] == dl
->disk
.refnum
) {
2197 /* This device plays role 'j' in 'v'. */
2198 rv
[n
].start
= __be64_to_cpu(v
->lba_offset
[j
]);
2199 rv
[n
].size
= __be64_to_cpu(v
->conf
.blocks
);
2204 qsort(rv
, n
, sizeof(*rv
), cmp_extent
);
2206 rv
[n
].start
= __be64_to_cpu(ddf
->phys
->entries
[dnum
].config_size
);
2211 int validate_geometry_ddf_bvd(struct supertype
*st
,
2212 int level
, int layout
, int raiddisks
,
2213 int chunk
, unsigned long long size
,
2214 char *dev
, unsigned long long *freesize
)
2217 struct ddf_super
*ddf
= st
->sb
;
2219 unsigned long long pos
= 0;
2220 unsigned long long maxsize
;
2223 /* ddf/bvd supports lots of things, but not containers */
2224 if (level
== LEVEL_CONTAINER
)
2226 /* We must have the container info already read in. */
2231 /* General test: make sure there is space for
2232 * 'raiddisks' device extents of size 'size'.
2234 unsigned long long minsize
= size
;
2238 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2243 e
= get_extents(ddf
, dl
);
2246 unsigned long long esize
;
2247 esize
= e
[i
].start
- pos
;
2248 if (esize
>= minsize
)
2250 pos
= e
[i
].start
+ e
[i
].size
;
2252 } while (e
[i
-1].size
);
2257 if (dcnt
< raiddisks
) {
2258 fprintf(stderr
, Name
": Not enough devices with space "
2259 "for this array (%d < %d)\n",
2265 /* This device must be a member of the set */
2266 if (stat(dev
, &stb
) < 0)
2268 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2270 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
2271 if (dl
->major
== major(stb
.st_rdev
) &&
2272 dl
->minor
== minor(stb
.st_rdev
))
2276 fprintf(stderr
, Name
": %s is not in the same DDF set\n",
2280 e
= get_extents(ddf
, dl
);
2284 unsigned long long esize
;
2285 esize
= e
[i
].start
- pos
;
2286 if (esize
>= maxsize
)
2288 pos
= e
[i
].start
+ e
[i
].size
;
2290 } while (e
[i
-1].size
);
2291 *freesize
= maxsize
;
2296 int validate_geometry_ddf_svd(struct supertype
*st
,
2297 int level
, int layout
, int raiddisks
,
2298 int chunk
, unsigned long long size
,
2299 char *dev
, unsigned long long *freesize
)
2301 /* dd/svd only supports striped, mirrored, concat, spanned... */
2302 if (level
!= LEVEL_LINEAR
&&
2310 static int load_super_ddf_all(struct supertype
*st
, int fd
,
2311 void **sbp
, char *devname
, int keep_fd
)
2314 struct ddf_super
*super
;
2315 struct mdinfo
*sd
, *best
= NULL
;
2321 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
2324 if (sra
->array
.major_version
!= -1 ||
2325 sra
->array
.minor_version
!= -2 ||
2326 strcmp(sra
->text_version
, "ddf") != 0)
2329 super
= malloc(sizeof(*super
));
2332 memset(super
, 0, sizeof(*super
));
2334 /* first, try each device, and choose the best ddf */
2335 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2337 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2338 dfd
= dev_open(nm
, O_RDONLY
);
2341 rv
= load_ddf_headers(dfd
, super
, NULL
);
2344 seq
= __be32_to_cpu(super
->active
->seq
);
2345 if (super
->active
->openflag
)
2347 if (!best
|| seq
> bestseq
) {
2355 /* OK, load this ddf */
2356 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
2357 dfd
= dev_open(nm
, O_RDONLY
);
2360 load_ddf_headers(dfd
, super
, NULL
);
2361 load_ddf_global(dfd
, super
, NULL
);
2363 /* Now we need the device-local bits */
2364 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2365 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2366 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2369 seq
= load_ddf_local(dfd
, super
, NULL
, keep_fd
);
2370 if (!keep_fd
) close(dfd
);
2373 if (st
->ss
== NULL
) {
2374 st
->ss
= &super_ddf_container
;
2375 st
->minor_version
= 0;
2377 st
->container_dev
= fd2devnum(fd
);
2385 static struct mdinfo
*container_content_ddf(struct supertype
*st
)
2387 /* Given a container loaded by load_super_ddf_all,
2388 * extract information about all the arrays into
2391 * For each vcl in conflist: create an mdinfo, fill it in,
2392 * then look for matching devices (phys_refnum) in dlist
2393 * and create appropriate device mdinfo.
2395 struct ddf_super
*ddf
= st
->sb
;
2396 struct mdinfo
*rest
= NULL
;
2399 for (vc
= ddf
->conflist
; vc
; vc
=vc
->next
)
2402 struct mdinfo
*this;
2403 this = malloc(sizeof(*this));
2404 memset(this, 0, sizeof(*this));
2408 this->array
.major_version
= 1000;
2409 this->array
.minor_version
= 0;
2410 this->array
.patch_version
= 0;
2411 this->array
.level
= map_num1(ddf_level_num
, vc
->conf
.prl
);
2412 this->array
.raid_disks
=
2413 __be16_to_cpu(vc
->conf
.prim_elmnt_count
);
2414 this->array
.layout
= rlq_to_layout(vc
->conf
.rlq
, vc
->conf
.prl
,
2415 this->array
.raid_disks
);
2416 this->array
.md_minor
= -1;
2417 this->array
.ctime
= DECADE
+
2418 __be32_to_cpu(*(__u32
*)(vc
->conf
.guid
+16));
2419 this->array
.utime
= DECADE
+
2420 __be32_to_cpu(vc
->conf
.timestamp
);
2421 this->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
2423 for (i
=0; i
< __be16_to_cpu(ddf
->virt
->populated_vdes
); i
++)
2424 if (memcmp(ddf
->virt
->entries
[i
].guid
,
2425 vc
->conf
.guid
, DDF_GUID_LEN
) == 0)
2427 if ((ddf
->virt
->entries
[i
].state
& DDF_state_inconsistent
) ||
2428 (ddf
->virt
->entries
[i
].init_state
& DDF_initstate_mask
) !=
2430 this->array
.state
= 0;
2431 this->resync_start
= 0;
2433 this->array
.state
= 1;
2434 this->resync_start
= ~0ULL;
2436 memcpy(this->name
, ddf
->virt
->entries
[i
].name
, 32);
2439 memset(this->uuid
, 0, sizeof(this->uuid
));
2440 this->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
2441 this->array
.size
= this->component_size
/ 2;
2442 this->container_member
= i
;
2444 sprintf(this->text_version
, "/%s/%d",
2445 devnum2devname(st
->container_dev
),
2446 this->container_member
);
2449 for (i
=0 ; i
< ddf
->mppe
; i
++) {
2453 if (vc
->conf
.phys_refnum
[i
] == 0xFFFFFFFF)
2456 this->array
.working_disks
++;
2458 for (d
= ddf
->dlist
; d
; d
=d
->next
)
2459 if (d
->disk
.refnum
== vc
->conf
.phys_refnum
[i
])
2464 dev
= malloc(sizeof(*dev
));
2465 memset(dev
, 0, sizeof(*dev
));
2466 dev
->next
= this->devs
;
2469 dev
->disk
.number
= __be32_to_cpu(d
->disk
.refnum
);
2470 dev
->disk
.major
= d
->major
;
2471 dev
->disk
.minor
= d
->minor
;
2472 dev
->disk
.raid_disk
= i
;
2473 dev
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
2475 dev
->events
= __le32_to_cpu(ddf
->primary
.seq
);
2476 dev
->data_offset
= vc
->lba_offset
[i
];
2477 dev
->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
2479 strcpy(dev
->name
, d
->devname
);
2485 static int init_zero_ddf(struct supertype
*st
,
2486 mdu_array_info_t
*info
,
2487 unsigned long long size
, char *name
,
2488 char *homehost
, int *uuid
)
2494 static int store_zero_ddf(struct supertype
*st
, int fd
)
2496 unsigned long long dsize
;
2498 memset(buf
, 0, 512);
2501 if (!get_dev_size(fd
, NULL
, &dsize
))
2504 lseek64(fd
, dsize
-512, 0);
2505 write(fd
, buf
, 512);
2509 static int compare_super_ddf(struct supertype
*st
, struct supertype
*tst
)
2513 * 0 same, or first was empty, and second was copied
2514 * 1 second had wrong number
2516 * 3 wrong other info
2518 struct ddf_super
*first
= st
->sb
;
2519 struct ddf_super
*second
= tst
->sb
;
2527 if (memcmp(first
->anchor
.guid
, second
->anchor
.guid
, DDF_GUID_LEN
) != 0)
2530 /* FIXME should I look at anything else? */
2535 * A new array 'a' has been started which claims to be instance 'inst'
2536 * within container 'c'.
2537 * We need to confirm that the array matches the metadata in 'c' so
2538 * that we don't corrupt any metadata.
2540 static int ddf_open_new(struct supertype
*c
, struct active_array
*a
, char *inst
)
2542 fprintf(stderr
, "ddf: open_new %s\n", inst
);
2543 a
->info
.container_member
= atoi(inst
);
2548 * The array 'a' is to be marked clean in the metadata.
2549 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2550 * clean up to the point (in sectors). If that cannot be recorded in the
2551 * metadata, then leave it as dirty.
2553 * For DDF, we need to clear the DDF_state_inconsistent bit in the
2554 * !global! virtual_disk.virtual_entry structure.
2556 static void ddf_set_array_state(struct active_array
*a
, int consistent
)
2558 struct ddf_super
*ddf
= a
->container
->sb
;
2559 int inst
= a
->info
.container_member
;
2561 ddf
->virt
->entries
[inst
].state
&= ~DDF_state_inconsistent
;
2563 ddf
->virt
->entries
[inst
].state
|= DDF_state_inconsistent
;
2564 ddf
->virt
->entries
[inst
].init_state
&= ~DDF_initstate_mask
;
2565 if (a
->resync_start
== ~0ULL)
2566 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_full
;
2567 else if (a
->resync_start
== 0)
2568 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_not
;
2570 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_quick
;
2572 printf("ddf mark %s %llu\n", consistent
?"clean":"dirty",
2577 * The state of each disk is stored in the global phys_disk structure
2578 * in phys_disk.entries[n].state.
2579 * This makes various combinations awkward.
2580 * - When a device fails in any array, it must be failed in all arrays
2581 * that include a part of this device.
2582 * - When a component is rebuilding, we cannot include it officially in the
2583 * array unless this is the only array that uses the device.
2585 * So: when transitioning:
2586 * Online -> failed, just set failed flag. monitor will propagate
2587 * spare -> online, the device might need to be added to the array.
2588 * spare -> failed, just set failed. Don't worry if in array or not.
2590 static void ddf_set_disk(struct active_array
*a
, int n
, int state
)
2592 struct ddf_super
*ddf
= a
->container
->sb
;
2593 int inst
= a
->info
.container_member
;
2594 struct vd_config
*vc
= find_vdcr(ddf
, inst
);
2595 int pd
= find_phys(ddf
, vc
->phys_refnum
[n
]);
2599 fprintf(stderr
, "ddf: cannot find instance %d!!\n", inst
);
2603 /* disk doesn't currently exist. If it is now in_sync,
2605 if ((state
& DS_INSYNC
) && ! (state
& DS_FAULTY
)) {
2606 /* Find dev 'n' in a->info->devs, determine the
2607 * ddf refnum, and set vc->phys_refnum and update
2613 if (state
& DS_FAULTY
)
2614 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Failed
);
2615 if (state
& DS_INSYNC
) {
2616 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Online
);
2617 ddf
->phys
->entries
[pd
].state
&= __cpu_to_be16(~DDF_Rebuilding
);
2621 /* Now we need to check the state of the array and update
2622 * virtual_disk.entries[n].state.
2623 * It needs to be one of "optimal", "degraded", "failed".
2624 * I don't understand 'deleted' or 'missing'.
2627 for (i
=0; i
< a
->info
.array
.raid_disks
; i
++) {
2628 pd
= find_phys(ddf
, vc
->phys_refnum
[i
]);
2631 st
= ddf
->phys
->entries
[pd
].state
;
2632 if ((state
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
2636 state
= DDF_state_degraded
;
2637 if (working
== a
->info
.array
.raid_disks
)
2638 state
= DDF_state_optimal
;
2639 else switch(vc
->prl
) {
2643 state
= DDF_state_failed
;
2647 state
= DDF_state_failed
;
2651 if (working
< a
->info
.array
.raid_disks
-1)
2652 state
= DDF_state_failed
;
2655 if (working
< a
->info
.array
.raid_disks
-2)
2656 state
= DDF_state_failed
;
2657 else if (working
== a
->info
.array
.raid_disks
-1)
2658 state
= DDF_state_part_optimal
;
2662 ddf
->virt
->entries
[inst
].state
=
2663 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
2666 fprintf(stderr
, "ddf: set_disk %d\n", n
);
2669 static void ddf_sync_metadata(struct supertype
*st
)
2673 * Write all data to all devices.
2674 * Later, we might be able to track whether only local changes
2675 * have been made, or whether any global data has been changed,
2676 * but ddf is sufficiently weird that it probably always
2677 * changes global data ....
2679 __write_init_super_ddf(st
, 0);
2680 fprintf(stderr
, "ddf: sync_metadata\n");
2683 struct superswitch super_ddf
= {
2685 .examine_super
= examine_super_ddf
,
2686 .brief_examine_super
= brief_examine_super_ddf
,
2687 .detail_super
= detail_super_ddf
,
2688 .brief_detail_super
= brief_detail_super_ddf
,
2689 .validate_geometry
= validate_geometry_ddf
,
2691 .match_home
= match_home_ddf
,
2692 .uuid_from_super
= uuid_from_super_ddf
,
2693 .getinfo_super
= getinfo_super_ddf
,
2694 .update_super
= update_super_ddf
,
2696 .avail_size
= avail_size_ddf
,
2698 .compare_super
= compare_super_ddf
,
2700 .load_super
= load_super_ddf
,
2701 .init_super
= init_zero_ddf
,
2702 .store_super
= store_zero_ddf
,
2703 .free_super
= free_super_ddf
,
2704 .match_metadata_desc
= match_metadata_desc_ddf
,
2705 .getinfo_super_n
= getinfo_super_n_container
,
2713 .open_new
= ddf_open_new
,
2714 .set_array_state
= ddf_set_array_state
,
2715 .set_disk
= ddf_set_disk
,
2716 .sync_metadata
= ddf_sync_metadata
,
2721 /* Super_ddf_container is set by validate_geometry_ddf when given a
2722 * device that is not part of any array
2724 struct superswitch super_ddf_container
= {
2726 .validate_geometry
= validate_geometry_ddf_container
,
2727 .write_init_super
= write_init_super_ddf
,
2730 .load_super
= load_super_ddf
,
2731 .init_super
= init_super_ddf
,
2732 .add_to_super
= add_to_super_ddf
,
2733 .getinfo_super
= getinfo_super_ddf
,
2735 .free_super
= free_super_ddf
,
2737 .container_content
= container_content_ddf
,
2738 .getinfo_super_n
= getinfo_super_n_container
,
2745 struct superswitch super_ddf_bvd
= {
2747 // .detail_super = detail_super_ddf_bvd,
2748 // .brief_detail_super = brief_detail_super_ddf_bvd,
2749 .validate_geometry
= validate_geometry_ddf_bvd
,
2750 .write_init_super
= write_init_super_ddf
,
2752 .update_super
= update_super_ddf
,
2753 .init_super
= init_super_ddf_bvd
,
2754 .add_to_super
= add_to_super_ddf_bvd
,
2755 .getinfo_super
= getinfo_super_ddf_bvd
,
2756 .getinfo_super_n
= getinfo_super_n_bvd
,
2758 .load_super
= load_super_ddf
,
2759 .free_super
= free_super_ddf
,
2760 .match_metadata_desc
= match_metadata_desc_ddf_bvd
,
2768 struct superswitch super_ddf_svd
= {
2770 // .detail_super = detail_super_ddf_svd,
2771 // .brief_detail_super = brief_detail_super_ddf_svd,
2772 .validate_geometry
= validate_geometry_ddf_svd
,
2774 .update_super
= update_super_ddf
,
2775 .init_super
= init_super_ddf
,
2777 .load_super
= load_super_ddf
,
2778 .free_super
= free_super_ddf
,
2779 .match_metadata_desc
= match_metadata_desc_ddf_svd
,