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 /* a non-official T10 name for creation GUIDs */
35 static char T10
[] = "Linux-MD";
37 /* DDF timestamps are 1980 based, so we need to add
38 * second-in-decade-of-seventies to convert to linux timestamps.
39 * 10 years with 2 leap years.
41 #define DECADE (3600*24*(365*10+2))
44 const unsigned char *buf
,
47 /* The DDF metadata handling.
48 * DDF metadata lives at the end of the device.
49 * The last 512 byte block provides an 'anchor' which is used to locate
50 * the rest of the metadata which usually lives immediately behind the anchor.
53 * - all multibyte numeric fields are bigendian.
54 * - all strings are space padded.
58 /* Primary Raid Level (PRL) */
59 #define DDF_RAID0 0x00
60 #define DDF_RAID1 0x01
61 #define DDF_RAID3 0x03
62 #define DDF_RAID4 0x04
63 #define DDF_RAID5 0x05
64 #define DDF_RAID1E 0x11
66 #define DDF_CONCAT 0x1f
67 #define DDF_RAID5E 0x15
68 #define DDF_RAID5EE 0x25
69 #define DDF_RAID6 0x06
71 /* Raid Level Qualifier (RLQ) */
72 #define DDF_RAID0_SIMPLE 0x00
73 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
74 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
75 #define DDF_RAID3_0 0x00 /* parity in first extent */
76 #define DDF_RAID3_N 0x01 /* parity in last extent */
77 #define DDF_RAID4_0 0x00 /* parity in first extent */
78 #define DDF_RAID4_N 0x01 /* parity in last extent */
79 /* these apply to raid5e and raid5ee as well */
80 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
81 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
82 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
83 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
85 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
86 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
88 /* Secondary RAID Level (SRL) */
89 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
90 #define DDF_2MIRRORED 0x01
91 #define DDF_2CONCAT 0x02
92 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
95 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
96 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
97 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
98 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
99 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
100 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
101 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
102 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
103 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
104 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
106 #define DDF_GUID_LEN 24
107 #define DDF_REVISION_0 "01.00.00"
108 #define DDF_REVISION_2 "01.02.00"
111 __u32 magic
; /* DDF_HEADER_MAGIC */
113 char guid
[DDF_GUID_LEN
];
114 char revision
[8]; /* 01.02.00 */
115 __u32 seq
; /* starts at '1' */
120 __u8 pad0
; /* 0xff */
121 __u8 pad1
[12]; /* 12 * 0xff */
122 /* 64 bytes so far */
123 __u8 header_ext
[32]; /* reserved: fill with 0xff */
127 __u8 pad2
[3]; /* 0xff */
128 __u32 workspace_len
; /* sectors for vendor space -
129 * at least 32768(sectors) */
131 __u16 max_pd_entries
; /* one of 15, 63, 255, 1023, 4095 */
132 __u16 max_vd_entries
; /* 2^(4,6,8,10,12)-1 : i.e. as above */
133 __u16 max_partitions
; /* i.e. max num of configuration
134 record entries per disk */
135 __u16 config_record_len
; /* 1 +ROUNDUP(max_primary_element_entries
137 __u16 max_primary_element_entries
; /* 16, 64, 256, 1024, or 4096 */
138 __u8 pad3
[54]; /* 0xff */
139 /* 192 bytes so far */
140 __u32 controller_section_offset
;
141 __u32 controller_section_length
;
142 __u32 phys_section_offset
;
143 __u32 phys_section_length
;
144 __u32 virt_section_offset
;
145 __u32 virt_section_length
;
146 __u32 config_section_offset
;
147 __u32 config_section_length
;
148 __u32 data_section_offset
;
149 __u32 data_section_length
;
150 __u32 bbm_section_offset
;
151 __u32 bbm_section_length
;
152 __u32 diag_space_offset
;
153 __u32 diag_space_length
;
156 /* 256 bytes so far */
157 __u8 pad4
[256]; /* 0xff */
161 #define DDF_HEADER_ANCHOR 0x00
162 #define DDF_HEADER_PRIMARY 0x01
163 #define DDF_HEADER_SECONDARY 0x02
165 /* The content of the 'controller section' - global scope */
166 struct ddf_controller_data
{
167 __u32 magic
; /* DDF_CONTROLLER_MAGIC */
169 char guid
[DDF_GUID_LEN
];
170 struct controller_type
{
177 __u8 pad
[8]; /* 0xff */
178 __u8 vendor_data
[448];
181 /* The content of phys_section - global scope */
183 __u32 magic
; /* DDF_PHYS_RECORDS_MAGIC */
188 struct phys_disk_entry
{
189 char guid
[DDF_GUID_LEN
];
193 __u64 config_size
; /* DDF structures must be after here */
194 char path
[18]; /* another horrible structure really */
199 /* phys_disk_entry.type is a bitmap - bigendian remember */
200 #define DDF_Forced_PD_GUID 1
201 #define DDF_Active_in_VD 2
202 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
203 #define DDF_Spare 8 /* overrides Global_spare */
204 #define DDF_Foreign 16
205 #define DDF_Legacy 32 /* no DDF on this device */
207 #define DDF_Interface_mask 0xf00
208 #define DDF_Interface_SCSI 0x100
209 #define DDF_Interface_SAS 0x200
210 #define DDF_Interface_SATA 0x300
211 #define DDF_Interface_FC 0x400
213 /* phys_disk_entry.state is a bigendian bitmap */
215 #define DDF_Failed 2 /* overrides 1,4,8 */
216 #define DDF_Rebuilding 4
217 #define DDF_Transition 8
219 #define DDF_ReadErrors 32
220 #define DDF_Missing 64
222 /* The content of the virt_section global scope */
223 struct virtual_disk
{
224 __u32 magic
; /* DDF_VIRT_RECORDS_MAGIC */
226 __u16 populated_vdes
;
229 struct virtual_entry
{
230 char guid
[DDF_GUID_LEN
];
232 __u16 pad0
; /* 0xffff */
242 /* virtual_entry.type is a bitmap - bigendian */
244 #define DDF_Enforce_Groups 2
245 #define DDF_Unicode 4
246 #define DDF_Owner_Valid 8
248 /* virtual_entry.state is a bigendian bitmap */
249 #define DDF_state_mask 0x7
250 #define DDF_state_optimal 0x0
251 #define DDF_state_degraded 0x1
252 #define DDF_state_deleted 0x2
253 #define DDF_state_missing 0x3
254 #define DDF_state_failed 0x4
255 #define DDF_state_part_optimal 0x5
257 #define DDF_state_morphing 0x8
258 #define DDF_state_inconsistent 0x10
260 /* virtual_entry.init_state is a bigendian bitmap */
261 #define DDF_initstate_mask 0x03
262 #define DDF_init_not 0x00
263 #define DDF_init_quick 0x01 /* initialisation is progress.
264 * i.e. 'state_inconsistent' */
265 #define DDF_init_full 0x02
267 #define DDF_access_mask 0xc0
268 #define DDF_access_rw 0x00
269 #define DDF_access_ro 0x80
270 #define DDF_access_blocked 0xc0
272 /* The content of the config_section - local scope
273 * It has multiple records each config_record_len sectors
274 * They can be vd_config or spare_assign
278 __u32 magic
; /* DDF_VD_CONF_MAGIC */
280 char guid
[DDF_GUID_LEN
];
284 __u16 prim_elmnt_count
;
285 __u8 chunk_shift
; /* 0 == 512, 1==1024 etc */
288 __u8 sec_elmnt_count
;
291 __u64 blocks
; /* blocks per component could be different
292 * on different component devices...(only
293 * for concat I hope) */
294 __u64 array_blocks
; /* blocks in array */
302 __u8 v0
[32]; /* reserved- 0xff */
303 __u8 v1
[32]; /* reserved- 0xff */
304 __u8 v2
[16]; /* reserved- 0xff */
305 __u8 v3
[16]; /* reserved- 0xff */
307 __u32 phys_refnum
[0]; /* refnum of each disk in sequence */
308 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
309 bvd are always the same size */
312 /* vd_config.cache_pol[7] is a bitmap */
313 #define DDF_cache_writeback 1 /* else writethrough */
314 #define DDF_cache_wadaptive 2 /* only applies if writeback */
315 #define DDF_cache_readahead 4
316 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
317 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
318 #define DDF_cache_wallowed 32 /* enable write caching */
319 #define DDF_cache_rallowed 64 /* enable read caching */
321 struct spare_assign
{
322 __u32 magic
; /* DDF_SPARE_ASSIGN_MAGIC */
327 __u16 populated
; /* SAEs used */
328 __u16 max
; /* max SAEs */
330 struct spare_assign_entry
{
331 char guid
[DDF_GUID_LEN
];
332 __u16 secondary_element
;
336 /* spare_assign.type is a bitmap */
337 #define DDF_spare_dedicated 0x1 /* else global */
338 #define DDF_spare_revertible 0x2 /* else committable */
339 #define DDF_spare_active 0x4 /* else not active */
340 #define DDF_spare_affinity 0x8 /* enclosure affinity */
342 /* The data_section contents - local scope */
344 __u32 magic
; /* DDF_PHYS_DATA_MAGIC */
346 char guid
[DDF_GUID_LEN
];
347 __u32 refnum
; /* crc of some magic drive data ... */
348 __u8 forced_ref
; /* set when above was not result of magic */
349 __u8 forced_guid
; /* set if guid was forced rather than magic */
354 /* bbm_section content */
355 struct bad_block_log
{
362 struct mapped_block
{
363 __u64 defective_start
;
364 __u32 replacement_start
;
370 /* Struct for internally holding ddf structures */
371 /* The DDF structure stored on each device is potentially
372 * quite different, as some data is global and some is local.
373 * The global data is:
376 * - Physical disk records
377 * - Virtual disk records
379 * - Configuration records
380 * - Physical Disk data section
381 * ( and Bad block and vendor which I don't care about yet).
383 * The local data is parsed into separate lists as it is read
384 * and reconstructed for writing. This means that we only need
385 * to make config changes once and they are automatically
386 * propagated to all devices.
387 * Note that the ddf_super has space of the conf and disk data
388 * for this disk and also for a list of all such data.
389 * The list is only used for the superblock that is being
390 * built in Create or Assemble to describe the whole array.
393 struct ddf_header anchor
, primary
, secondary
;
394 struct ddf_controller_data controller
;
395 struct ddf_header
*active
;
396 struct phys_disk
*phys
;
397 struct virtual_disk
*virt
;
399 int max_part
, mppe
, conf_rec_len
;
407 __u64
*lba_offset
; /* location in 'conf' of
409 int vcnum
; /* index into ->virt */
410 __u64
*block_sizes
; /* NULL if all the same */
413 struct vd_config conf
;
414 } *conflist
, *currentconf
;
423 unsigned long long size
; /* sectors */
424 int pdnum
; /* index in ->phys */
425 struct spare_assign
*spare
;
428 struct disk_data disk
;
429 struct vcl
*vlist
[0]; /* max_part in size */
434 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
438 static int calc_crc(void *buf
, int len
)
440 /* crcs are always at the same place as in the ddf_header */
441 struct ddf_header
*ddf
= buf
;
442 __u32 oldcrc
= ddf
->crc
;
444 ddf
->crc
= 0xffffffff;
446 newcrc
= crc32(0, buf
, len
);
451 static int load_ddf_header(int fd
, unsigned long long lba
,
452 unsigned long long size
,
454 struct ddf_header
*hdr
, struct ddf_header
*anchor
)
456 /* read a ddf header (primary or secondary) from fd/lba
457 * and check that it is consistent with anchor
459 * magic, crc, guid, rev, and LBA's header_type, and
460 * everything after header_type must be the same
465 if (lseek64(fd
, lba
<<9, 0) < 0)
468 if (read(fd
, hdr
, 512) != 512)
471 if (hdr
->magic
!= DDF_HEADER_MAGIC
)
473 if (calc_crc(hdr
, 512) != hdr
->crc
)
475 if (memcmp(anchor
->guid
, hdr
->guid
, DDF_GUID_LEN
) != 0 ||
476 memcmp(anchor
->revision
, hdr
->revision
, 8) != 0 ||
477 anchor
->primary_lba
!= hdr
->primary_lba
||
478 anchor
->secondary_lba
!= hdr
->secondary_lba
||
480 memcmp(anchor
->pad2
, hdr
->pad2
, 512 -
481 offsetof(struct ddf_header
, pad2
)) != 0)
484 /* Looks good enough to me... */
488 static void *load_section(int fd
, struct ddf_super
*super
, void *buf
,
489 __u32 offset_be
, __u32 len_be
, int check
)
491 unsigned long long offset
= __be32_to_cpu(offset_be
);
492 unsigned long long len
= __be32_to_cpu(len_be
);
493 int dofree
= (buf
== NULL
);
496 if (len
!= 2 && len
!= 8 && len
!= 32
497 && len
!= 128 && len
!= 512)
503 /* All pre-allocated sections are a single block */
507 posix_memalign(&buf
, 512, len
<<9);
513 if (super
->active
->type
== 1)
514 offset
+= __be64_to_cpu(super
->active
->primary_lba
);
516 offset
+= __be64_to_cpu(super
->active
->secondary_lba
);
518 if (lseek64(fd
, offset
<<9, 0) != (offset
<<9)) {
523 if (read(fd
, buf
, len
<<9) != (len
<<9)) {
531 static int load_ddf_headers(int fd
, struct ddf_super
*super
, char *devname
)
533 unsigned long long dsize
;
535 get_dev_size(fd
, NULL
, &dsize
);
537 if (lseek64(fd
, dsize
-512, 0) < 0) {
540 Name
": Cannot seek to anchor block on %s: %s\n",
541 devname
, strerror(errno
));
544 if (read(fd
, &super
->anchor
, 512) != 512) {
547 Name
": Cannot read anchor block on %s: %s\n",
548 devname
, strerror(errno
));
551 if (super
->anchor
.magic
!= DDF_HEADER_MAGIC
) {
553 fprintf(stderr
, Name
": no DDF anchor found on %s\n",
557 if (calc_crc(&super
->anchor
, 512) != super
->anchor
.crc
) {
559 fprintf(stderr
, Name
": bad CRC on anchor on %s\n",
563 if (memcmp(super
->anchor
.revision
, DDF_REVISION_0
, 8) != 0 &&
564 memcmp(super
->anchor
.revision
, DDF_REVISION_2
, 8) != 0) {
566 fprintf(stderr
, Name
": can only support super revision"
567 " %.8s and earlier, not %.8s on %s\n",
568 DDF_REVISION_2
, super
->anchor
.revision
,devname
);
571 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.primary_lba
),
573 &super
->primary
, &super
->anchor
) == 0) {
576 Name
": Failed to load primary DDF header "
580 super
->active
= &super
->primary
;
581 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.secondary_lba
),
583 &super
->secondary
, &super
->anchor
)) {
584 if ((__be32_to_cpu(super
->primary
.seq
)
585 < __be32_to_cpu(super
->secondary
.seq
) &&
586 !super
->secondary
.openflag
)
587 || (__be32_to_cpu(super
->primary
.seq
)
588 == __be32_to_cpu(super
->secondary
.seq
) &&
589 super
->primary
.openflag
&& !super
->secondary
.openflag
)
591 super
->active
= &super
->secondary
;
596 static int load_ddf_global(int fd
, struct ddf_super
*super
, char *devname
)
599 ok
= load_section(fd
, super
, &super
->controller
,
600 super
->active
->controller_section_offset
,
601 super
->active
->controller_section_length
,
603 super
->phys
= load_section(fd
, super
, NULL
,
604 super
->active
->phys_section_offset
,
605 super
->active
->phys_section_length
,
607 super
->pdsize
= __be32_to_cpu(super
->active
->phys_section_length
) * 512;
609 super
->virt
= load_section(fd
, super
, NULL
,
610 super
->active
->virt_section_offset
,
611 super
->active
->virt_section_length
,
613 super
->vdsize
= __be32_to_cpu(super
->active
->virt_section_length
) * 512;
623 super
->conflist
= NULL
;
626 super
->max_part
= __be16_to_cpu(super
->active
->max_partitions
);
627 super
->mppe
= __be16_to_cpu(super
->active
->max_primary_element_entries
);
628 super
->conf_rec_len
= __be16_to_cpu(super
->active
->config_record_len
);
632 static int load_ddf_local(int fd
, struct ddf_super
*super
,
633 char *devname
, int keep
)
640 int max_virt_disks
= __be16_to_cpu(super
->active
->max_vd_entries
);
641 unsigned long long dsize
;
643 /* First the local disk info */
644 posix_memalign((void**)&dl
, 512,
646 (super
->max_part
) * sizeof(dl
->vlist
[0]));
648 load_section(fd
, super
, &dl
->disk
,
649 super
->active
->data_section_offset
,
650 super
->active
->data_section_length
,
652 dl
->devname
= devname
? strdup(devname
) : NULL
;
655 dl
->major
= major(stb
.st_rdev
);
656 dl
->minor
= minor(stb
.st_rdev
);
657 dl
->next
= super
->dlist
;
658 dl
->fd
= keep
? fd
: -1;
661 if (get_dev_size(fd
, devname
, &dsize
))
662 dl
->size
= dsize
>> 9;
664 for (i
=0 ; i
< super
->max_part
; i
++)
668 for (i
=0; i
< __be16_to_cpu(super
->active
->max_pd_entries
); i
++)
669 if (memcmp(super
->phys
->entries
[i
].guid
,
670 dl
->disk
.guid
, DDF_GUID_LEN
) == 0)
673 /* Now the config list. */
674 /* 'conf' is an array of config entries, some of which are
675 * probably invalid. Those which are good need to be copied into
679 conf
= load_section(fd
, super
, NULL
,
680 super
->active
->config_section_offset
,
681 super
->active
->config_section_length
,
686 i
< __be32_to_cpu(super
->active
->config_section_length
);
687 i
+= super
->conf_rec_len
) {
688 struct vd_config
*vd
=
689 (struct vd_config
*)((char*)conf
+ i
*512);
692 if (vd
->magic
== DDF_SPARE_ASSIGN_MAGIC
) {
695 posix_memalign((void**)&dl
->spare
, 512,
696 super
->conf_rec_len
*512);
697 memcpy(dl
->spare
, vd
, super
->conf_rec_len
*512);
700 if (vd
->magic
!= DDF_VD_CONF_MAGIC
)
702 for (vcl
= super
->conflist
; vcl
; vcl
= vcl
->next
) {
703 if (memcmp(vcl
->conf
.guid
,
704 vd
->guid
, DDF_GUID_LEN
) == 0)
709 dl
->vlist
[vnum
++] = vcl
;
710 if (__be32_to_cpu(vd
->seqnum
) <=
711 __be32_to_cpu(vcl
->conf
.seqnum
))
714 posix_memalign((void**)&vcl
, 512,
715 (super
->conf_rec_len
*512 +
716 offsetof(struct vcl
, conf
)));
717 vcl
->next
= super
->conflist
;
718 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
719 super
->conflist
= vcl
;
720 dl
->vlist
[vnum
++] = vcl
;
722 memcpy(&vcl
->conf
, vd
, super
->conf_rec_len
*512);
723 vcl
->lba_offset
= (__u64
*)
724 &vcl
->conf
.phys_refnum
[super
->mppe
];
726 for (i
=0; i
< max_virt_disks
; i
++)
727 if (memcmp(super
->virt
->entries
[i
].guid
,
728 vcl
->conf
.guid
, DDF_GUID_LEN
)==0)
730 if (i
< max_virt_disks
)
739 static int load_super_ddf_all(struct supertype
*st
, int fd
,
740 void **sbp
, char *devname
, int keep_fd
);
742 static int load_super_ddf(struct supertype
*st
, int fd
,
745 unsigned long long dsize
;
746 struct ddf_super
*super
;
750 /* if 'fd' is a container, load metadata from all the devices */
751 if (load_super_ddf_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
755 return 1; /* FIXME Is this correct */
757 if (get_dev_size(fd
, devname
, &dsize
) == 0)
760 /* 32M is a lower bound */
761 if (dsize
<= 32*1024*1024) {
764 Name
": %s is too small for ddf: "
765 "size is %llu sectors.\n",
773 Name
": %s is an odd size for ddf: "
774 "size is %llu bytes.\n",
780 if (posix_memalign((void**)&super
, 512, sizeof(*super
))!= 0) {
781 fprintf(stderr
, Name
": malloc of %zu failed.\n",
785 memset(super
, 0, sizeof(*super
));
787 rv
= load_ddf_headers(fd
, super
, devname
);
793 /* Have valid headers and have chosen the best. Let's read in the rest*/
795 rv
= load_ddf_global(fd
, super
, devname
);
800 Name
": Failed to load all information "
801 "sections on %s\n", devname
);
806 load_ddf_local(fd
, super
, devname
, 0);
808 /* Should possibly check the sections .... */
811 if (st
->ss
== NULL
) {
813 st
->minor_version
= 0;
820 static void free_super_ddf(struct supertype
*st
)
822 struct ddf_super
*ddf
= st
->sb
;
827 while (ddf
->conflist
) {
828 struct vcl
*v
= ddf
->conflist
;
829 ddf
->conflist
= v
->next
;
831 free(v
->block_sizes
);
835 struct dl
*d
= ddf
->dlist
;
836 ddf
->dlist
= d
->next
;
847 static struct supertype
*match_metadata_desc_ddf(char *arg
)
849 /* 'ddf' only support containers */
850 struct supertype
*st
;
851 if (strcmp(arg
, "ddf") != 0 &&
852 strcmp(arg
, "default") != 0
856 st
= malloc(sizeof(*st
));
857 memset(st
, 0, sizeof(*st
));
860 st
->minor_version
= 0;
868 static mapping_t ddf_state
[] = {
874 { "Partially Optimal", 5},
880 static mapping_t ddf_init_state
[] = {
881 { "Not Initialised", 0},
882 { "QuickInit in Progress", 1},
883 { "Fully Initialised", 2},
887 static mapping_t ddf_access
[] = {
891 { "Blocked (no access)", 3},
895 static mapping_t ddf_level
[] = {
896 { "RAID0", DDF_RAID0
},
897 { "RAID1", DDF_RAID1
},
898 { "RAID3", DDF_RAID3
},
899 { "RAID4", DDF_RAID4
},
900 { "RAID5", DDF_RAID5
},
901 { "RAID1E",DDF_RAID1E
},
903 { "CONCAT",DDF_CONCAT
},
904 { "RAID5E",DDF_RAID5E
},
905 { "RAID5EE",DDF_RAID5EE
},
906 { "RAID6", DDF_RAID6
},
909 static mapping_t ddf_sec_level
[] = {
910 { "Striped", DDF_2STRIPED
},
911 { "Mirrored", DDF_2MIRRORED
},
912 { "Concat", DDF_2CONCAT
},
913 { "Spanned", DDF_2SPANNED
},
921 static struct num_mapping ddf_level_num
[] = {
924 { DDF_RAID3
, LEVEL_UNSUPPORTED
},
927 { DDF_RAID1E
, LEVEL_UNSUPPORTED
},
928 { DDF_JBOD
, LEVEL_UNSUPPORTED
},
929 { DDF_CONCAT
, LEVEL_LINEAR
},
930 { DDF_RAID5E
, LEVEL_UNSUPPORTED
},
931 { DDF_RAID5EE
, LEVEL_UNSUPPORTED
},
936 static int map_num1(struct num_mapping
*map
, int num
)
939 for (i
=0 ; map
[i
].num1
!= MAXINT
; i
++)
940 if (map
[i
].num1
== num
)
946 static void print_guid(char *guid
, int tstamp
)
948 /* A GUIDs are part (or all) ASCII and part binary.
949 * They tend to be space padded.
950 * We print the GUID in HEX, then in parentheses add
951 * any initial ASCII sequence, and a possible
952 * time stamp from bytes 16-19
954 int l
= DDF_GUID_LEN
;
957 for (i
=0 ; i
<DDF_GUID_LEN
; i
++) {
958 if ((i
&3)==0 && i
!= 0) printf(":");
959 printf("%02X", guid
[i
]&255);
963 while (l
&& guid
[l
-1] == ' ')
965 for (i
=0 ; i
<l
; i
++) {
966 if (guid
[i
] >= 0x20 && guid
[i
] < 0x7f)
967 fputc(guid
[i
], stdout
);
972 time_t then
= __be32_to_cpu(*(__u32
*)(guid
+16)) + DECADE
;
975 tm
= localtime(&then
);
976 strftime(tbuf
, 100, " %D %T",tm
);
982 static void examine_vd(int n
, struct ddf_super
*sb
, char *guid
)
984 int crl
= sb
->conf_rec_len
;
987 for (vcl
= sb
->conflist
; vcl
; vcl
= vcl
->next
) {
988 struct vd_config
*vc
= &vcl
->conf
;
990 if (calc_crc(vc
, crl
*512) != vc
->crc
)
992 if (memcmp(vc
->guid
, guid
, DDF_GUID_LEN
) != 0)
995 /* Ok, we know about this VD, let's give more details */
996 printf(" Raid Devices[%d] : %d\n", n
,
997 __be16_to_cpu(vc
->prim_elmnt_count
));
998 printf(" Chunk Size[%d] : %d sectors\n", n
,
999 1 << vc
->chunk_shift
);
1000 printf(" Raid Level[%d] : %s\n", n
,
1001 map_num(ddf_level
, vc
->prl
)?:"-unknown-");
1002 if (vc
->sec_elmnt_count
!= 1) {
1003 printf(" Secondary Position[%d] : %d of %d\n", n
,
1004 vc
->sec_elmnt_seq
, vc
->sec_elmnt_count
);
1005 printf(" Secondary Level[%d] : %s\n", n
,
1006 map_num(ddf_sec_level
, vc
->srl
) ?: "-unknown-");
1008 printf(" Device Size[%d] : %llu\n", n
,
1009 __be64_to_cpu(vc
->blocks
)/2);
1010 printf(" Array Size[%d] : %llu\n", n
,
1011 __be64_to_cpu(vc
->array_blocks
)/2);
1015 static void examine_vds(struct ddf_super
*sb
)
1017 int cnt
= __be16_to_cpu(sb
->virt
->populated_vdes
);
1019 printf(" Virtual Disks : %d\n", cnt
);
1021 for (i
=0; i
<cnt
; i
++) {
1022 struct virtual_entry
*ve
= &sb
->virt
->entries
[i
];
1023 printf(" VD GUID[%d] : ", i
); print_guid(ve
->guid
, 1);
1025 printf(" unit[%d] : %d\n", i
, __be16_to_cpu(ve
->unit
));
1026 printf(" state[%d] : %s, %s%s\n", i
,
1027 map_num(ddf_state
, ve
->state
& 7),
1028 (ve
->state
& 8) ? "Morphing, ": "",
1029 (ve
->state
& 16)? "Not Consistent" : "Consistent");
1030 printf(" init state[%d] : %s\n", i
,
1031 map_num(ddf_init_state
, ve
->init_state
&3));
1032 printf(" access[%d] : %s\n", i
,
1033 map_num(ddf_access
, (ve
->init_state
>>6) & 3));
1034 printf(" Name[%d] : %.16s\n", i
, ve
->name
);
1035 examine_vd(i
, sb
, ve
->guid
);
1037 if (cnt
) printf("\n");
1040 static void examine_pds(struct ddf_super
*sb
)
1042 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1045 printf(" Physical Disks : %d\n", cnt
);
1047 for (i
=0 ; i
<cnt
; i
++) {
1048 struct phys_disk_entry
*pd
= &sb
->phys
->entries
[i
];
1049 int type
= __be16_to_cpu(pd
->type
);
1050 int state
= __be16_to_cpu(pd
->state
);
1052 printf(" PD GUID[%d] : ", i
); print_guid(pd
->guid
, 0);
1054 printf(" ref[%d] : %08x\n", i
,
1055 __be32_to_cpu(pd
->refnum
));
1056 printf(" mode[%d] : %s%s%s%s%s\n", i
,
1057 (type
&2) ? "active":"",
1058 (type
&4) ? "Global Spare":"",
1059 (type
&8) ? "spare" : "",
1060 (type
&16)? ", foreign" : "",
1061 (type
&32)? "pass-through" : "");
1062 printf(" state[%d] : %s%s%s%s%s%s%s\n", i
,
1063 (state
&1)? "Online": "Offline",
1064 (state
&2)? ", Failed": "",
1065 (state
&4)? ", Rebuilding": "",
1066 (state
&8)? ", in-transition": "",
1067 (state
&16)? ", SMART errors": "",
1068 (state
&32)? ", Unrecovered Read Errors": "",
1069 (state
&64)? ", Missing" : "");
1070 printf(" Avail Size[%d] : %llu K\n", i
,
1071 __be64_to_cpu(pd
->config_size
)>>1);
1072 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1073 if (dl
->disk
.refnum
== pd
->refnum
) {
1074 char *dv
= map_dev(dl
->major
, dl
->minor
, 0);
1076 printf(" Device[%d] : %s\n",
1084 static void examine_super_ddf(struct supertype
*st
, char *homehost
)
1086 struct ddf_super
*sb
= st
->sb
;
1088 printf(" Magic : %08x\n", __be32_to_cpu(sb
->anchor
.magic
));
1089 printf(" Version : %.8s\n", sb
->anchor
.revision
);
1090 printf("Controller GUID : "); print_guid(sb
->controller
.guid
, 0);
1092 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1094 printf(" Seq : %08x\n", __be32_to_cpu(sb
->active
->seq
));
1095 printf(" Redundant hdr : %s\n", sb
->secondary
.magic
== DDF_HEADER_MAGIC
1101 static void brief_examine_super_ddf(struct supertype
*st
)
1103 /* We just write a generic DDF ARRAY entry
1104 * The uuid is all hex, 6 groups of 4 bytes
1106 struct ddf_super
*ddf
= st
->sb
;
1108 printf("ARRAY /dev/ddf metadata=ddf UUID=");
1109 for (i
= 0; i
< DDF_GUID_LEN
; i
++) {
1110 if ((i
&3) == 0 && i
!= 0)
1112 printf("%02X", 255&ddf
->anchor
.guid
[i
]);
1117 static void detail_super_ddf(struct supertype
*st
, char *homehost
)
1120 * Could print DDF GUID
1121 * Need to find which array
1122 * If whole, briefly list all arrays
1127 static void brief_detail_super_ddf(struct supertype
*st
)
1129 /* FIXME I really need to know which array we are detailing.
1130 * Can that be stored in ddf_super??
1132 // struct ddf_super *ddf = st->sb;
1136 static int match_home_ddf(struct supertype
*st
, char *homehost
)
1138 /* It matches 'this' host if the controller is a
1139 * Linux-MD controller with vendor_data matching
1142 struct ddf_super
*ddf
= st
->sb
;
1143 int len
= strlen(homehost
);
1145 return (memcmp(ddf
->controller
.guid
, T10
, 8) == 0 &&
1146 len
< sizeof(ddf
->controller
.vendor_data
) &&
1147 memcmp(ddf
->controller
.vendor_data
, homehost
,len
) == 0 &&
1148 ddf
->controller
.vendor_data
[len
] == 0);
1151 static struct vd_config
*find_vdcr(struct ddf_super
*ddf
, int inst
)
1155 for (v
= ddf
->conflist
; v
; v
= v
->next
)
1156 if (inst
== v
->vcnum
)
1161 static int find_phys(struct ddf_super
*ddf
, __u32 phys_refnum
)
1163 /* Find the entry in phys_disk which has the given refnum
1164 * and return it's index
1167 for (i
=0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
); i
++)
1168 if (ddf
->phys
->entries
[i
].refnum
== phys_refnum
)
1173 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4])
1175 /* The uuid returned here is used for:
1176 * uuid to put into bitmap file (Create, Grow)
1177 * uuid for backup header when saving critical section (Grow)
1178 * comparing uuids when re-adding a device into an array
1179 * For each of these we can make do with a truncated
1180 * or hashed uuid rather than the original, as long as
1182 * In each case the uuid required is that of the data-array,
1183 * not the device-set.
1184 * In the case of SVD we assume the BVD is of interest,
1185 * though that might be the case if a bitmap were made for
1186 * a mirrored SVD - worry about that later.
1187 * So we need to find the VD configuration record for the
1188 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1189 * The first 16 bytes of the sha1 of these is used.
1191 struct ddf_super
*ddf
= st
->sb
;
1192 struct vcl
*vcl
= ddf
->currentconf
;
1195 memset(uuid
, 0, sizeof (uuid
));
1198 struct sha1_ctx ctx
;
1199 sha1_init_ctx(&ctx
);
1200 sha1_process_bytes(&vcl
->conf
.guid
, DDF_GUID_LEN
, &ctx
);
1201 if (vcl
->conf
.sec_elmnt_count
> 1)
1202 sha1_process_bytes(&vcl
->conf
.sec_elmnt_seq
, 1, &ctx
);
1203 sha1_finish_ctx(&ctx
, buf
);
1204 memcpy(uuid
, buf
, sizeof(uuid
));
1208 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
);
1210 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
)
1212 struct ddf_super
*ddf
= st
->sb
;
1214 if (ddf
->currentconf
) {
1215 getinfo_super_ddf_bvd(st
, info
);
1219 info
->array
.raid_disks
= __be16_to_cpu(ddf
->phys
->used_pdes
);
1220 info
->array
.level
= LEVEL_CONTAINER
;
1221 info
->array
.layout
= 0;
1222 info
->array
.md_minor
= -1;
1223 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*(__u32
*)
1224 (ddf
->anchor
.guid
+16));
1225 info
->array
.utime
= 0;
1226 info
->array
.chunk_size
= 0;
1229 info
->disk
.major
= 0;
1230 info
->disk
.minor
= 0;
1232 info
->disk
.number
= __be32_to_cpu(ddf
->dlist
->disk
.refnum
);
1233 info
->disk
.raid_disk
= find_phys(ddf
, ddf
->dlist
->disk
.refnum
);
1235 info
->data_offset
= __be64_to_cpu(ddf
->phys
->
1236 entries
[info
->disk
.raid_disk
].
1238 info
->component_size
= ddf
->dlist
->size
- info
->data_offset
;
1240 info
->disk
.number
= -1;
1241 // info->disk.raid_disk = find refnum in the table and use index;
1243 info
->disk
.state
= (1 << MD_DISK_SYNC
);
1246 info
->reshape_active
= 0;
1248 strcpy(info
->text_version
, "ddf");
1250 // uuid_from_super_ddf(info->uuid, sbv);
1252 // info->name[] ?? ;
1255 static int rlq_to_layout(int rlq
, int prl
, int raiddisks
);
1257 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
)
1259 struct ddf_super
*ddf
= st
->sb
;
1260 struct vcl
*vc
= ddf
->currentconf
;
1261 int cd
= ddf
->currentdev
;
1263 /* FIXME this returns BVD info - what if we want SVD ?? */
1265 info
->array
.raid_disks
= __be16_to_cpu(vc
->conf
.prim_elmnt_count
);
1266 info
->array
.level
= map_num1(ddf_level_num
, vc
->conf
.prl
);
1267 info
->array
.layout
= rlq_to_layout(vc
->conf
.rlq
, vc
->conf
.prl
,
1268 info
->array
.raid_disks
);
1269 info
->array
.md_minor
= -1;
1270 info
->array
.ctime
= DECADE
+
1271 __be32_to_cpu(*(__u32
*)(vc
->conf
.guid
+16));
1272 info
->array
.utime
= DECADE
+ __be32_to_cpu(vc
->conf
.timestamp
);
1273 info
->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
1275 if (cd
>= 0 && cd
< ddf
->mppe
) {
1276 info
->data_offset
= __be64_to_cpu(vc
->lba_offset
[cd
]);
1277 if (vc
->block_sizes
)
1278 info
->component_size
= vc
->block_sizes
[cd
];
1280 info
->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
1283 info
->disk
.major
= 0;
1284 info
->disk
.minor
= 0;
1285 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1286 // info->disk.raid_disk = find refnum in the table and use index;
1287 // info->disk.state = ???;
1289 info
->container_member
= ddf
->currentconf
->vcnum
;
1291 info
->resync_start
= 0;
1292 if (!(ddf
->virt
->entries
[info
->container_member
].state
1293 & DDF_state_inconsistent
) &&
1294 (ddf
->virt
->entries
[info
->container_member
].init_state
1295 & DDF_initstate_mask
)
1297 info
->resync_start
= ~0ULL;
1299 uuid_from_super_ddf(st
, info
->uuid
);
1301 info
->container_member
= atoi(st
->subarray
);
1302 sprintf(info
->text_version
, "/%s/%s",
1303 devnum2devname(st
->container_dev
),
1306 // info->name[] ?? ;
1310 static int update_super_ddf(struct supertype
*st
, struct mdinfo
*info
,
1312 char *devname
, int verbose
,
1313 int uuid_set
, char *homehost
)
1315 /* For 'assemble' and 'force' we need to return non-zero if any
1316 * change was made. For others, the return value is ignored.
1317 * Update options are:
1318 * force-one : This device looks a bit old but needs to be included,
1319 * update age info appropriately.
1320 * assemble: clear any 'faulty' flag to allow this device to
1322 * force-array: Array is degraded but being forced, mark it clean
1323 * if that will be needed to assemble it.
1325 * newdev: not used ????
1326 * grow: Array has gained a new device - this is currently for
1328 * resync: mark as dirty so a resync will happen.
1329 * uuid: Change the uuid of the array to match what is given
1330 * homehost: update the recorded homehost
1331 * name: update the name - preserving the homehost
1332 * _reshape_progress: record new reshape_progress position.
1334 * Following are not relevant for this version:
1335 * sparc2.2 : update from old dodgey metadata
1336 * super-minor: change the preferred_minor number
1337 * summaries: update redundant counters.
1340 // struct ddf_super *ddf = st->sb;
1341 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1342 // struct virtual_entry *ve = find_ve(ddf);
1344 /* we don't need to handle "force-*" or "assemble" as
1345 * there is no need to 'trick' the kernel. We the metadata is
1346 * first updated to activate the array, all the implied modifications
1350 if (strcmp(update
, "grow") == 0) {
1353 if (strcmp(update
, "resync") == 0) {
1354 // info->resync_checkpoint = 0;
1356 /* We ignore UUID updates as they make even less sense
1359 if (strcmp(update
, "homehost") == 0) {
1360 /* homehost is stored in controller->vendor_data,
1361 * or it is when we are the vendor
1363 // if (info->vendor_is_local)
1364 // strcpy(ddf->controller.vendor_data, homehost);
1366 if (strcmp(update
, "name") == 0) {
1367 /* name is stored in virtual_entry->name */
1368 // memset(ve->name, ' ', 16);
1369 // strncpy(ve->name, info->name, 16);
1371 if (strcmp(update
, "_reshape_progress") == 0) {
1372 /* We don't support reshape yet */
1375 // update_all_csum(ddf);
1380 static void make_header_guid(char *guid
)
1384 /* Create a DDF Header of Virtual Disk GUID */
1386 /* 24 bytes of fiction required.
1387 * first 8 are a 'vendor-id' - "Linux-MD"
1388 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1389 * Remaining 8 random number plus timestamp
1391 memcpy(guid
, T10
, sizeof(T10
));
1392 stamp
= __cpu_to_be32(0xdeadbeef);
1393 memcpy(guid
+8, &stamp
, 4);
1394 stamp
= __cpu_to_be32(0);
1395 memcpy(guid
+12, &stamp
, 4);
1396 stamp
= __cpu_to_be32(time(0) - DECADE
);
1397 memcpy(guid
+16, &stamp
, 4);
1398 rfd
= open("/dev/urandom", O_RDONLY
);
1399 if (rfd
< 0 || read(rfd
, &stamp
, 4) != 4)
1401 memcpy(guid
+20, &stamp
, 4);
1402 if (rfd
>= 0) close(rfd
);
1405 static int init_super_ddf_bvd(struct supertype
*st
,
1406 mdu_array_info_t
*info
,
1407 unsigned long long size
,
1408 char *name
, char *homehost
,
1411 static int init_super_ddf(struct supertype
*st
,
1412 mdu_array_info_t
*info
,
1413 unsigned long long size
, char *name
, char *homehost
,
1416 /* This is primarily called by Create when creating a new array.
1417 * We will then get add_to_super called for each component, and then
1418 * write_init_super called to write it out to each device.
1419 * For DDF, Create can create on fresh devices or on a pre-existing
1421 * To create on a pre-existing array a different method will be called.
1422 * This one is just for fresh drives.
1424 * We need to create the entire 'ddf' structure which includes:
1425 * DDF headers - these are easy.
1426 * Controller data - a Sector describing this controller .. not that
1427 * this is a controller exactly.
1428 * Physical Disk Record - one entry per device, so
1429 * leave plenty of space.
1430 * Virtual Disk Records - again, just leave plenty of space.
1431 * This just lists VDs, doesn't give details
1432 * Config records - describes the VDs that use this disk
1433 * DiskData - describes 'this' device.
1434 * BadBlockManagement - empty
1435 * Diag Space - empty
1436 * Vendor Logs - Could we put bitmaps here?
1439 struct ddf_super
*ddf
;
1442 int max_phys_disks
, max_virt_disks
;
1443 unsigned long long sector
;
1447 struct phys_disk
*pd
;
1448 struct virtual_disk
*vd
;
1455 return init_super_ddf_bvd(st
, info
, size
, name
, homehost
,
1458 posix_memalign((void**)&ddf
, 512, sizeof(*ddf
));
1459 memset(ddf
, 0, sizeof(*ddf
));
1460 ddf
->dlist
= NULL
; /* no physical disks yet */
1461 ddf
->conflist
= NULL
; /* No virtual disks yet */
1463 /* At least 32MB *must* be reserved for the ddf. So let's just
1464 * start 32MB from the end, and put the primary header there.
1465 * Don't do secondary for now.
1466 * We don't know exactly where that will be yet as it could be
1467 * different on each device. To just set up the lengths.
1471 ddf
->anchor
.magic
= DDF_HEADER_MAGIC
;
1472 make_header_guid(ddf
->anchor
.guid
);
1474 memcpy(ddf
->anchor
.revision
, DDF_REVISION_2
, 8);
1475 ddf
->anchor
.seq
= __cpu_to_be32(1);
1476 ddf
->anchor
.timestamp
= __cpu_to_be32(time(0) - DECADE
);
1477 ddf
->anchor
.openflag
= 0xFF;
1478 ddf
->anchor
.foreignflag
= 0;
1479 ddf
->anchor
.enforcegroups
= 0; /* Is this best?? */
1480 ddf
->anchor
.pad0
= 0xff;
1481 memset(ddf
->anchor
.pad1
, 0xff, 12);
1482 memset(ddf
->anchor
.header_ext
, 0xff, 32);
1483 ddf
->anchor
.primary_lba
= ~(__u64
)0;
1484 ddf
->anchor
.secondary_lba
= ~(__u64
)0;
1485 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
1486 memset(ddf
->anchor
.pad2
, 0xff, 3);
1487 ddf
->anchor
.workspace_len
= __cpu_to_be32(32768); /* Must be reserved */
1488 ddf
->anchor
.workspace_lba
= ~(__u64
)0; /* Put this at bottom
1489 of 32M reserved.. */
1490 max_phys_disks
= 1023; /* Should be enough */
1491 ddf
->anchor
.max_pd_entries
= __cpu_to_be16(max_phys_disks
);
1492 max_virt_disks
= 255;
1493 ddf
->anchor
.max_vd_entries
= __cpu_to_be16(max_virt_disks
); /* ?? */
1494 ddf
->anchor
.max_partitions
= __cpu_to_be16(64); /* ?? */
1497 ddf
->conf_rec_len
= 1 + ROUND_UP(ddf
->mppe
* (4+8), 512)/512;
1498 ddf
->anchor
.config_record_len
= __cpu_to_be16(ddf
->conf_rec_len
);
1499 ddf
->anchor
.max_primary_element_entries
= __cpu_to_be16(ddf
->mppe
);
1500 memset(ddf
->anchor
.pad3
, 0xff, 54);
1501 /* controller sections is one sector long immediately
1502 * after the ddf header */
1504 ddf
->anchor
.controller_section_offset
= __cpu_to_be32(sector
);
1505 ddf
->anchor
.controller_section_length
= __cpu_to_be32(1);
1508 /* phys is 8 sectors after that */
1509 pdsize
= ROUND_UP(sizeof(struct phys_disk
) +
1510 sizeof(struct phys_disk_entry
)*max_phys_disks
,
1512 switch(pdsize
/512) {
1513 case 2: case 8: case 32: case 128: case 512: break;
1516 ddf
->anchor
.phys_section_offset
= __cpu_to_be32(sector
);
1517 ddf
->anchor
.phys_section_length
=
1518 __cpu_to_be32(pdsize
/512); /* max_primary_element_entries/8 */
1519 sector
+= pdsize
/512;
1521 /* virt is another 32 sectors */
1522 vdsize
= ROUND_UP(sizeof(struct virtual_disk
) +
1523 sizeof(struct virtual_entry
) * max_virt_disks
,
1525 switch(vdsize
/512) {
1526 case 2: case 8: case 32: case 128: case 512: break;
1529 ddf
->anchor
.virt_section_offset
= __cpu_to_be32(sector
);
1530 ddf
->anchor
.virt_section_length
=
1531 __cpu_to_be32(vdsize
/512); /* max_vd_entries/8 */
1532 sector
+= vdsize
/512;
1534 clen
= ddf
->conf_rec_len
* (ddf
->max_part
+1);
1535 ddf
->anchor
.config_section_offset
= __cpu_to_be32(sector
);
1536 ddf
->anchor
.config_section_length
= __cpu_to_be32(clen
);
1539 ddf
->anchor
.data_section_offset
= __cpu_to_be32(sector
);
1540 ddf
->anchor
.data_section_length
= __cpu_to_be32(1);
1543 ddf
->anchor
.bbm_section_length
= __cpu_to_be32(0);
1544 ddf
->anchor
.bbm_section_offset
= __cpu_to_be32(0xFFFFFFFF);
1545 ddf
->anchor
.diag_space_length
= __cpu_to_be32(0);
1546 ddf
->anchor
.diag_space_offset
= __cpu_to_be32(0xFFFFFFFF);
1547 ddf
->anchor
.vendor_length
= __cpu_to_be32(0);
1548 ddf
->anchor
.vendor_offset
= __cpu_to_be32(0xFFFFFFFF);
1550 memset(ddf
->anchor
.pad4
, 0xff, 256);
1552 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
1553 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
1555 ddf
->primary
.openflag
= 1; /* I guess.. */
1556 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
1558 ddf
->secondary
.openflag
= 1; /* I guess.. */
1559 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
1561 ddf
->active
= &ddf
->primary
;
1563 ddf
->controller
.magic
= DDF_CONTROLLER_MAGIC
;
1565 /* 24 more bytes of fiction required.
1566 * first 8 are a 'vendor-id' - "Linux-MD"
1567 * Remaining 16 are serial number.... maybe a hostname would do?
1569 memcpy(ddf
->controller
.guid
, T10
, sizeof(T10
));
1570 gethostname(hostname
, sizeof(hostname
));
1571 hostname
[sizeof(hostname
) - 1] = 0;
1572 hostlen
= strlen(hostname
);
1573 memcpy(ddf
->controller
.guid
+ 24 - hostlen
, hostname
, hostlen
);
1574 for (i
= strlen(T10
) ; i
+hostlen
< 24; i
++)
1575 ddf
->controller
.guid
[i
] = ' ';
1577 ddf
->controller
.type
.vendor_id
= __cpu_to_be16(0xDEAD);
1578 ddf
->controller
.type
.device_id
= __cpu_to_be16(0xBEEF);
1579 ddf
->controller
.type
.sub_vendor_id
= 0;
1580 ddf
->controller
.type
.sub_device_id
= 0;
1581 memcpy(ddf
->controller
.product_id
, "What Is My PID??", 16);
1582 memset(ddf
->controller
.pad
, 0xff, 8);
1583 memset(ddf
->controller
.vendor_data
, 0xff, 448);
1585 posix_memalign((void**)&pd
, 512, pdsize
);
1587 ddf
->pdsize
= pdsize
;
1589 memset(pd
, 0xff, pdsize
);
1590 memset(pd
, 0, sizeof(*pd
));
1591 pd
->magic
= DDF_PHYS_DATA_MAGIC
;
1592 pd
->used_pdes
= __cpu_to_be16(0);
1593 pd
->max_pdes
= __cpu_to_be16(max_phys_disks
);
1594 memset(pd
->pad
, 0xff, 52);
1596 posix_memalign((void**)&vd
, 512, vdsize
);
1598 ddf
->vdsize
= vdsize
;
1599 memset(vd
, 0, vdsize
);
1600 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
1601 vd
->populated_vdes
= __cpu_to_be16(0);
1602 vd
->max_vdes
= __cpu_to_be16(max_virt_disks
);
1603 memset(vd
->pad
, 0xff, 52);
1605 for (i
=0; i
<max_virt_disks
; i
++)
1606 memset(&vd
->entries
[i
], 0xff, sizeof(struct virtual_entry
));
1609 ddf
->updates_pending
= 1;
1613 static int all_ff(char *guid
)
1616 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
1617 if (guid
[i
] != (char)0xff)
1621 static int chunk_to_shift(int chunksize
)
1623 return ffs(chunksize
/512)-1;
1626 static int level_to_prl(int level
)
1629 case LEVEL_LINEAR
: return DDF_CONCAT
;
1630 case 0: return DDF_RAID0
;
1631 case 1: return DDF_RAID1
;
1632 case 4: return DDF_RAID4
;
1633 case 5: return DDF_RAID5
;
1634 case 6: return DDF_RAID6
;
1638 static int layout_to_rlq(int level
, int layout
, int raiddisks
)
1642 return DDF_RAID0_SIMPLE
;
1645 case 2: return DDF_RAID1_SIMPLE
;
1646 case 3: return DDF_RAID1_MULTI
;
1651 case 0: return DDF_RAID4_N
;
1657 case ALGORITHM_LEFT_ASYMMETRIC
:
1658 return DDF_RAID5_N_RESTART
;
1659 case ALGORITHM_RIGHT_ASYMMETRIC
:
1661 return DDF_RAID5_0_RESTART
;
1663 return DDF_RAID6_0_RESTART
;
1664 case ALGORITHM_LEFT_SYMMETRIC
:
1665 return DDF_RAID5_N_CONTINUE
;
1666 case ALGORITHM_RIGHT_SYMMETRIC
:
1667 return -1; /* not mentioned in standard */
1673 static int rlq_to_layout(int rlq
, int prl
, int raiddisks
)
1677 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1679 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1687 return -1; /* FIXME this isn't checked */
1691 case DDF_RAID5_N_RESTART
:
1692 return ALGORITHM_LEFT_ASYMMETRIC
;
1693 case DDF_RAID5_0_RESTART
:
1694 return ALGORITHM_RIGHT_ASYMMETRIC
;
1695 case DDF_RAID5_N_CONTINUE
:
1696 return ALGORITHM_LEFT_SYMMETRIC
;
1702 case DDF_RAID5_N_RESTART
:
1703 return ALGORITHM_LEFT_ASYMMETRIC
;
1704 case DDF_RAID6_0_RESTART
:
1705 return ALGORITHM_RIGHT_ASYMMETRIC
;
1706 case DDF_RAID5_N_CONTINUE
:
1707 return ALGORITHM_LEFT_SYMMETRIC
;
1716 unsigned long long start
, size
;
1718 static int cmp_extent(const void *av
, const void *bv
)
1720 const struct extent
*a
= av
;
1721 const struct extent
*b
= bv
;
1722 if (a
->start
< b
->start
)
1724 if (a
->start
> b
->start
)
1729 static struct extent
*get_extents(struct ddf_super
*ddf
, struct dl
*dl
)
1731 /* find a list of used extents on the give physical device
1732 * (dnum) of the given ddf.
1733 * Return a malloced array of 'struct extent'
1735 FIXME ignore DDF_Legacy devices?
1742 rv
= malloc(sizeof(struct extent
) * (ddf
->max_part
+ 2));
1746 for (i
= 0; i
< ddf
->max_part
; i
++) {
1747 struct vcl
*v
= dl
->vlist
[i
];
1750 for (j
=0; j
< v
->conf
.prim_elmnt_count
; j
++)
1751 if (v
->conf
.phys_refnum
[j
] == dl
->disk
.refnum
) {
1752 /* This device plays role 'j' in 'v'. */
1753 rv
[n
].start
= __be64_to_cpu(v
->lba_offset
[j
]);
1754 rv
[n
].size
= __be64_to_cpu(v
->conf
.blocks
);
1759 qsort(rv
, n
, sizeof(*rv
), cmp_extent
);
1761 rv
[n
].start
= __be64_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].config_size
);
1766 static int init_super_ddf_bvd(struct supertype
*st
,
1767 mdu_array_info_t
*info
,
1768 unsigned long long size
,
1769 char *name
, char *homehost
,
1772 /* We are creating a BVD inside a pre-existing container.
1773 * so st->sb is already set.
1774 * We need to create a new vd_config and a new virtual_entry
1776 struct ddf_super
*ddf
= st
->sb
;
1778 struct virtual_entry
*ve
;
1780 struct vd_config
*vc
;
1782 if (__be16_to_cpu(ddf
->virt
->populated_vdes
)
1783 >= __be16_to_cpu(ddf
->virt
->max_vdes
)) {
1784 fprintf(stderr
, Name
": This ddf already has the "
1785 "maximum of %d virtual devices\n",
1786 __be16_to_cpu(ddf
->virt
->max_vdes
));
1790 for (venum
= 0; venum
< __be16_to_cpu(ddf
->virt
->max_vdes
); venum
++)
1791 if (all_ff(ddf
->virt
->entries
[venum
].guid
))
1793 if (venum
== __be16_to_cpu(ddf
->virt
->max_vdes
)) {
1794 fprintf(stderr
, Name
": Cannot find spare slot for "
1795 "virtual disk - DDF is corrupt\n");
1798 ve
= &ddf
->virt
->entries
[venum
];
1800 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1801 * timestamp, random number
1803 make_header_guid(ve
->guid
);
1804 ve
->unit
= __cpu_to_be16(info
->md_minor
);
1806 ve
->guid_crc
= crc32(0, (unsigned char*)ddf
->anchor
.guid
, DDF_GUID_LEN
);
1808 ve
->state
= DDF_state_degraded
; /* Will be modified as devices are added */
1809 if (info
->state
& 1) /* clean */
1810 ve
->init_state
= DDF_init_full
;
1812 ve
->init_state
= DDF_init_not
;
1814 memset(ve
->pad1
, 0xff, 14);
1815 memset(ve
->name
, ' ', 16);
1817 strncpy(ve
->name
, name
, 16);
1818 ddf
->virt
->populated_vdes
=
1819 __cpu_to_be16(__be16_to_cpu(ddf
->virt
->populated_vdes
)+1);
1821 /* Now create a new vd_config */
1822 posix_memalign((void**)&vcl
, 512,
1823 (offsetof(struct vcl
, conf
) + ddf
->conf_rec_len
* 512));
1824 vcl
->lba_offset
= (__u64
*) &vcl
->conf
.phys_refnum
[ddf
->mppe
];
1826 sprintf(st
->subarray
, "%d", venum
);
1827 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
1831 vc
->magic
= DDF_VD_CONF_MAGIC
;
1832 memcpy(vc
->guid
, ve
->guid
, DDF_GUID_LEN
);
1833 vc
->timestamp
= __cpu_to_be32(time(0)-DECADE
);
1834 vc
->seqnum
= __cpu_to_be32(1);
1835 memset(vc
->pad0
, 0xff, 24);
1836 vc
->prim_elmnt_count
= __cpu_to_be16(info
->raid_disks
);
1837 vc
->chunk_shift
= chunk_to_shift(info
->chunk_size
);
1838 vc
->prl
= level_to_prl(info
->level
);
1839 vc
->rlq
= layout_to_rlq(info
->level
, info
->layout
, info
->raid_disks
);
1840 vc
->sec_elmnt_count
= 1;
1841 vc
->sec_elmnt_seq
= 0;
1843 vc
->blocks
= __cpu_to_be64(info
->size
* 2);
1844 vc
->array_blocks
= __cpu_to_be64(
1845 calc_array_size(info
->level
, info
->raid_disks
, info
->layout
,
1846 info
->chunk_size
, info
->size
*2));
1847 memset(vc
->pad1
, 0xff, 8);
1848 vc
->spare_refs
[0] = 0xffffffff;
1849 vc
->spare_refs
[1] = 0xffffffff;
1850 vc
->spare_refs
[2] = 0xffffffff;
1851 vc
->spare_refs
[3] = 0xffffffff;
1852 vc
->spare_refs
[4] = 0xffffffff;
1853 vc
->spare_refs
[5] = 0xffffffff;
1854 vc
->spare_refs
[6] = 0xffffffff;
1855 vc
->spare_refs
[7] = 0xffffffff;
1856 memset(vc
->cache_pol
, 0, 8);
1858 memset(vc
->pad2
, 0xff, 3);
1859 memset(vc
->pad3
, 0xff, 52);
1860 memset(vc
->pad4
, 0xff, 192);
1861 memset(vc
->v0
, 0xff, 32);
1862 memset(vc
->v1
, 0xff, 32);
1863 memset(vc
->v2
, 0xff, 16);
1864 memset(vc
->v3
, 0xff, 16);
1865 memset(vc
->vendor
, 0xff, 32);
1867 memset(vc
->phys_refnum
, 0xff, 4*ddf
->mppe
);
1868 memset(vc
->phys_refnum
+(ddf
->mppe
* 4), 0x00, 8*ddf
->mppe
);
1870 vcl
->next
= ddf
->conflist
;
1871 ddf
->conflist
= vcl
;
1872 ddf
->currentconf
= vcl
;
1873 ddf
->updates_pending
= 1;
1877 static void add_to_super_ddf_bvd(struct supertype
*st
,
1878 mdu_disk_info_t
*dk
, int fd
, char *devname
)
1880 /* fd and devname identify a device with-in the ddf container (st).
1881 * dk identifies a location in the new BVD.
1882 * We need to find suitable free space in that device and update
1883 * the phys_refnum and lba_offset for the newly created vd_config.
1884 * We might also want to update the type in the phys_disk
1888 struct ddf_super
*ddf
= st
->sb
;
1889 struct vd_config
*vc
;
1893 unsigned long long blocks
, pos
, esize
;
1896 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
1897 if (dl
->major
== dk
->major
&&
1898 dl
->minor
== dk
->minor
)
1900 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1903 vc
= &ddf
->currentconf
->conf
;
1904 lba_offset
= ddf
->currentconf
->lba_offset
;
1906 ex
= get_extents(ddf
, dl
);
1911 blocks
= __be64_to_cpu(vc
->blocks
);
1912 if (ddf
->currentconf
->block_sizes
)
1913 blocks
= ddf
->currentconf
->block_sizes
[dk
->raid_disk
];
1916 esize
= ex
[i
].start
- pos
;
1917 if (esize
>= blocks
)
1919 pos
= ex
[i
].start
+ ex
[i
].size
;
1921 } while (ex
[i
-1].size
);
1927 ddf
->currentdev
= dk
->raid_disk
;
1928 vc
->phys_refnum
[dk
->raid_disk
] = dl
->disk
.refnum
;
1929 lba_offset
[dk
->raid_disk
] = __cpu_to_be64(pos
);
1931 for (i
=0; i
< ddf
->max_part
; i
++)
1932 if (dl
->vlist
[i
] == NULL
)
1934 if (i
== ddf
->max_part
)
1936 dl
->vlist
[i
] = ddf
->currentconf
;
1939 dl
->devname
= devname
;
1941 /* Check how many working raid_disks, and if we can mark
1942 * array as optimal yet
1946 for (i
=0; i
< __be16_to_cpu(vc
->prim_elmnt_count
); i
++)
1947 if (vc
->phys_refnum
[i
] != 0xffffffff)
1950 /* Find which virtual_entry */
1951 i
= ddf
->currentconf
->vcnum
;
1952 if (working
== __be16_to_cpu(vc
->prim_elmnt_count
))
1953 ddf
->virt
->entries
[i
].state
=
1954 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
1955 | DDF_state_optimal
;
1957 if (vc
->prl
== DDF_RAID6
&&
1958 working
+1 == __be16_to_cpu(vc
->prim_elmnt_count
))
1959 ddf
->virt
->entries
[i
].state
=
1960 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
1961 | DDF_state_part_optimal
;
1963 ddf
->phys
->entries
[dl
->pdnum
].type
&= ~__cpu_to_be16(DDF_Global_Spare
);
1964 ddf
->phys
->entries
[dl
->pdnum
].type
|= __cpu_to_be16(DDF_Active_in_VD
);
1965 ddf
->updates_pending
= 1;
1968 /* add a device to a container, either while creating it or while
1969 * expanding a pre-existing container
1971 static void add_to_super_ddf(struct supertype
*st
,
1972 mdu_disk_info_t
*dk
, int fd
, char *devname
)
1974 struct ddf_super
*ddf
= st
->sb
;
1978 unsigned long long size
;
1979 struct phys_disk_entry
*pde
;
1983 if (ddf
->currentconf
) {
1984 add_to_super_ddf_bvd(st
, dk
, fd
, devname
);
1988 /* This is device numbered dk->number. We need to create
1989 * a phys_disk entry and a more detailed disk_data entry.
1992 posix_memalign((void**)&dd
, 512,
1993 sizeof(*dd
) + sizeof(dd
->vlist
[0]) * ddf
->max_part
);
1994 dd
->major
= major(stb
.st_rdev
);
1995 dd
->minor
= minor(stb
.st_rdev
);
1996 dd
->devname
= devname
;
1997 dd
->next
= ddf
->dlist
;
2001 dd
->disk
.magic
= DDF_PHYS_DATA_MAGIC
;
2003 tm
= localtime(&now
);
2004 sprintf(dd
->disk
.guid
, "%8s%04d%02d%02d",
2005 T10
, tm
->tm_year
+1900, tm
->tm_mon
+1, tm
->tm_mday
);
2006 *(__u32
*)(dd
->disk
.guid
+ 16) = random();
2007 *(__u32
*)(dd
->disk
.guid
+ 20) = random();
2010 /* Cannot be bothered finding a CRC of some irrelevant details*/
2011 dd
->disk
.refnum
= random();
2012 for (i
= __be16_to_cpu(ddf
->active
->max_pd_entries
) - 1;
2014 if (ddf
->phys
->entries
[i
].refnum
== dd
->disk
.refnum
)
2018 dd
->disk
.forced_ref
= 1;
2019 dd
->disk
.forced_guid
= 1;
2020 memset(dd
->disk
.vendor
, ' ', 32);
2021 memcpy(dd
->disk
.vendor
, "Linux", 5);
2022 memset(dd
->disk
.pad
, 0xff, 442);
2023 for (i
= 0; i
< ddf
->max_part
; i
++)
2024 dd
->vlist
[i
] = NULL
;
2026 n
= __be16_to_cpu(ddf
->phys
->used_pdes
);
2027 pde
= &ddf
->phys
->entries
[n
];
2031 ddf
->phys
->used_pdes
= __cpu_to_be16(n
);
2033 memcpy(pde
->guid
, dd
->disk
.guid
, DDF_GUID_LEN
);
2034 pde
->refnum
= dd
->disk
.refnum
;
2035 pde
->type
= __cpu_to_be16(DDF_Forced_PD_GUID
| DDF_Global_Spare
);
2036 pde
->state
= __cpu_to_be16(DDF_Online
);
2037 get_dev_size(fd
, NULL
, &size
);
2038 /* We are required to reserve 32Meg, and record the size in sectors */
2039 pde
->config_size
= __cpu_to_be64( (size
- 32*1024*1024) / 512);
2040 sprintf(pde
->path
, "%17.17s","Information: nil") ;
2041 memset(pde
->pad
, 0xff, 6);
2043 dd
->size
= size
>> 9;
2045 ddf
->updates_pending
= 1;
2049 * This is the write_init_super method for a ddf container. It is
2050 * called when creating a container or adding another device to a
2056 static unsigned char null_conf
[4096+512];
2058 static int __write_init_super_ddf(struct supertype
*st
, int do_close
)
2061 struct ddf_super
*ddf
= st
->sb
;
2067 unsigned long long size
, sector
;
2069 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
2075 /* We need to fill in the primary, (secondary) and workspace
2076 * lba's in the headers, set their checksums,
2077 * Also checksum phys, virt....
2079 * Then write everything out, finally the anchor is written.
2081 get_dev_size(fd
, NULL
, &size
);
2083 ddf
->anchor
.workspace_lba
= __cpu_to_be64(size
- 32*1024*2);
2084 ddf
->anchor
.primary_lba
= __cpu_to_be64(size
- 16*1024*2);
2085 ddf
->anchor
.seq
= __cpu_to_be32(1);
2086 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2087 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2089 ddf
->anchor
.openflag
= 0xFF; /* 'open' means nothing */
2090 ddf
->anchor
.seq
= 0xFFFFFFFF; /* no sequencing in anchor */
2091 ddf
->anchor
.crc
= calc_crc(&ddf
->anchor
, 512);
2093 ddf
->primary
.openflag
= 0;
2094 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
2096 ddf
->secondary
.openflag
= 0;
2097 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
2099 ddf
->primary
.crc
= calc_crc(&ddf
->primary
, 512);
2100 ddf
->secondary
.crc
= calc_crc(&ddf
->secondary
, 512);
2102 sector
= size
- 16*1024*2;
2103 lseek64(fd
, sector
<<9, 0);
2104 write(fd
, &ddf
->primary
, 512);
2106 ddf
->controller
.crc
= calc_crc(&ddf
->controller
, 512);
2107 write(fd
, &ddf
->controller
, 512);
2109 ddf
->phys
->crc
= calc_crc(ddf
->phys
, ddf
->pdsize
);
2111 write(fd
, ddf
->phys
, ddf
->pdsize
);
2113 ddf
->virt
->crc
= calc_crc(ddf
->virt
, ddf
->vdsize
);
2114 write(fd
, ddf
->virt
, ddf
->vdsize
);
2116 /* Now write lots of config records. */
2117 n_config
= ddf
->max_part
;
2118 conf_size
= ddf
->conf_rec_len
* 512;
2119 for (i
= 0 ; i
<= n_config
; i
++) {
2120 struct vcl
*c
= d
->vlist
[i
];
2122 c
= (struct vcl
*)d
->spare
;
2125 c
->conf
.crc
= calc_crc(&c
->conf
, conf_size
);
2126 write(fd
, &c
->conf
, conf_size
);
2128 char *null_aligned
= (char*)((((unsigned long)null_conf
)+511)&~511UL);
2129 if (null_conf
[0] != 0xff)
2130 memset(null_conf
, 0xff, sizeof(null_conf
));
2131 int togo
= conf_size
;
2132 while (togo
> sizeof(null_conf
)-512) {
2133 write(fd
, null_aligned
, sizeof(null_conf
)-512);
2134 togo
-= sizeof(null_conf
)-512;
2136 write(fd
, null_aligned
, togo
);
2139 d
->disk
.crc
= calc_crc(&d
->disk
, 512);
2140 write(fd
, &d
->disk
, 512);
2142 /* Maybe do the same for secondary */
2144 lseek64(fd
, (size
-1)*512, SEEK_SET
);
2145 write(fd
, &ddf
->anchor
, 512);
2154 static int write_init_super_ddf(struct supertype
*st
)
2157 if (st
->update_tail
) {
2158 /* queue the virtual_disk and vd_config as metadata updates */
2159 struct virtual_disk
*vd
;
2160 struct vd_config
*vc
;
2161 struct ddf_super
*ddf
= st
->sb
;
2164 /* First the virtual disk. We have a slightly fake header */
2165 len
= sizeof(struct virtual_disk
) + sizeof(struct virtual_entry
);
2168 vd
->entries
[0] = ddf
->virt
->entries
[ddf
->currentconf
->vcnum
];
2169 vd
->populated_vdes
= __cpu_to_be16(ddf
->currentconf
->vcnum
);
2170 append_metadata_update(st
, vd
, len
);
2172 /* Then the vd_config */
2173 len
= ddf
->conf_rec_len
* 512;
2175 memcpy(vc
, &ddf
->currentconf
->conf
, len
);
2176 append_metadata_update(st
, vc
, len
);
2178 /* FIXME I need to close the fds! */
2181 return __write_init_super_ddf(st
, 1);
2186 static __u64
avail_size_ddf(struct supertype
*st
, __u64 devsize
)
2188 /* We must reserve the last 32Meg */
2189 if (devsize
<= 32*1024*2)
2191 return devsize
- 32*1024*2;
2196 validate_geometry_ddf_container(struct supertype
*st
,
2197 int level
, int layout
, int raiddisks
,
2198 int chunk
, unsigned long long size
,
2199 char *dev
, unsigned long long *freesize
,
2202 static int validate_geometry_ddf_bvd(struct supertype
*st
,
2203 int level
, int layout
, int raiddisks
,
2204 int chunk
, unsigned long long size
,
2205 char *dev
, unsigned long long *freesize
,
2208 static int validate_geometry_ddf(struct supertype
*st
,
2209 int level
, int layout
, int raiddisks
,
2210 int chunk
, unsigned long long size
,
2211 char *dev
, unsigned long long *freesize
,
2218 /* ddf potentially supports lots of things, but it depends on
2219 * what devices are offered (and maybe kernel version?)
2220 * If given unused devices, we will make a container.
2221 * If given devices in a container, we will make a BVD.
2222 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2225 if (level
== LEVEL_CONTAINER
) {
2226 /* Must be a fresh device to add to a container */
2227 return validate_geometry_ddf_container(st
, level
, layout
,
2229 size
, dev
, freesize
,
2234 /* A container has already been opened, so we are
2235 * creating in there. Maybe a BVD, maybe an SVD.
2236 * Should make a distinction one day.
2238 return validate_geometry_ddf_bvd(st
, level
, layout
, raiddisks
,
2239 chunk
, size
, dev
, freesize
,
2243 /* Initial sanity check. Exclude illegal levels. */
2245 for (i
=0; ddf_level_num
[i
].num1
!= MAXINT
; i
++)
2246 if (ddf_level_num
[i
].num2
== level
)
2248 if (ddf_level_num
[i
].num1
== MAXINT
)
2250 /* Should check layout? etc */
2254 /* This is the first device for the array.
2255 * If it is a container, we read it in and do automagic allocations,
2256 * no other devices should be given.
2257 * Otherwise it must be a member device of a container, and we
2258 * do manual allocation.
2259 * Later we should check for a BVD and make an SVD.
2261 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2263 sra
= sysfs_read(fd
, 0, GET_VERSION
);
2265 if (sra
&& sra
->array
.major_version
== -1 &&
2266 strcmp(sra
->text_version
, "ddf") == 0) {
2269 /* find space for 'n' devices. */
2270 /* remember the devices */
2271 /* Somehow return the fact that we have enough */
2276 Name
": ddf: Cannot create this array "
2281 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2283 fprintf(stderr
, Name
": ddf: Cannot open %s: %s\n",
2284 dev
, strerror(errno
));
2287 /* Well, it is in use by someone, maybe a 'ddf' container. */
2288 cfd
= open_container(fd
);
2292 fprintf(stderr
, Name
": ddf: Cannot use %s: %s\n",
2293 dev
, strerror(EBUSY
));
2296 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2298 if (sra
&& sra
->array
.major_version
== -1 &&
2299 strcmp(sra
->text_version
, "ddf") == 0) {
2300 /* This is a member of a ddf container. Load the container
2301 * and try to create a bvd
2303 struct ddf_super
*ddf
;
2304 if (load_super_ddf_all(st
, cfd
, (void **)&ddf
, NULL
, 1) == 0) {
2306 st
->container_dev
= fd2devnum(cfd
);
2308 return validate_geometry_ddf_bvd(st
, level
, layout
,
2309 raiddisks
, chunk
, size
,
2314 } else /* device may belong to a different container */
2321 validate_geometry_ddf_container(struct supertype
*st
,
2322 int level
, int layout
, int raiddisks
,
2323 int chunk
, unsigned long long size
,
2324 char *dev
, unsigned long long *freesize
,
2328 unsigned long long ldsize
;
2330 if (level
!= LEVEL_CONTAINER
)
2335 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2338 fprintf(stderr
, Name
": ddf: Cannot open %s: %s\n",
2339 dev
, strerror(errno
));
2342 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2348 *freesize
= avail_size_ddf(st
, ldsize
>> 9);
2353 static int validate_geometry_ddf_bvd(struct supertype
*st
,
2354 int level
, int layout
, int raiddisks
,
2355 int chunk
, unsigned long long size
,
2356 char *dev
, unsigned long long *freesize
,
2360 struct ddf_super
*ddf
= st
->sb
;
2362 unsigned long long pos
= 0;
2363 unsigned long long maxsize
;
2366 /* ddf/bvd supports lots of things, but not containers */
2367 if (level
== LEVEL_CONTAINER
)
2369 /* We must have the container info already read in. */
2374 /* General test: make sure there is space for
2375 * 'raiddisks' device extents of size 'size'.
2377 unsigned long long minsize
= size
;
2381 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2387 e
= get_extents(ddf
, dl
);
2390 unsigned long long esize
;
2391 esize
= e
[i
].start
- pos
;
2392 if (esize
>= minsize
)
2394 pos
= e
[i
].start
+ e
[i
].size
;
2396 } while (e
[i
-1].size
);
2401 if (dcnt
< raiddisks
) {
2404 Name
": ddf: Not enough devices with "
2405 "space for this array (%d < %d)\n",
2411 /* This device must be a member of the set */
2412 if (stat(dev
, &stb
) < 0)
2414 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2416 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
2417 if (dl
->major
== major(stb
.st_rdev
) &&
2418 dl
->minor
== minor(stb
.st_rdev
))
2423 fprintf(stderr
, Name
": ddf: %s is not in the "
2428 e
= get_extents(ddf
, dl
);
2432 unsigned long long esize
;
2433 esize
= e
[i
].start
- pos
;
2434 if (esize
>= maxsize
)
2436 pos
= e
[i
].start
+ e
[i
].size
;
2438 } while (e
[i
-1].size
);
2439 *freesize
= maxsize
;
2445 static int load_super_ddf_all(struct supertype
*st
, int fd
,
2446 void **sbp
, char *devname
, int keep_fd
)
2449 struct ddf_super
*super
;
2450 struct mdinfo
*sd
, *best
= NULL
;
2456 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
2459 if (sra
->array
.major_version
!= -1 ||
2460 sra
->array
.minor_version
!= -2 ||
2461 strcmp(sra
->text_version
, "ddf") != 0)
2464 if (posix_memalign((void**)&super
, 512, sizeof(*super
)) != 0)
2466 memset(super
, 0, sizeof(*super
));
2468 /* first, try each device, and choose the best ddf */
2469 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2471 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2472 dfd
= dev_open(nm
, O_RDONLY
);
2475 rv
= load_ddf_headers(dfd
, super
, NULL
);
2478 seq
= __be32_to_cpu(super
->active
->seq
);
2479 if (super
->active
->openflag
)
2481 if (!best
|| seq
> bestseq
) {
2489 /* OK, load this ddf */
2490 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
2491 dfd
= dev_open(nm
, O_RDONLY
);
2494 load_ddf_headers(dfd
, super
, NULL
);
2495 load_ddf_global(dfd
, super
, NULL
);
2497 /* Now we need the device-local bits */
2498 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2499 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2500 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2503 seq
= load_ddf_local(dfd
, super
, NULL
, keep_fd
);
2504 if (!keep_fd
) close(dfd
);
2506 if (st
->subarray
[0]) {
2509 for (v
= super
->conflist
; v
; v
= v
->next
)
2510 if (v
->vcnum
== atoi(st
->subarray
))
2511 super
->currentconf
= v
;
2512 if (!super
->currentconf
)
2516 if (st
->ss
== NULL
) {
2517 st
->ss
= &super_ddf
;
2518 st
->minor_version
= 0;
2520 st
->container_dev
= fd2devnum(fd
);
2526 static struct mdinfo
*container_content_ddf(struct supertype
*st
)
2528 /* Given a container loaded by load_super_ddf_all,
2529 * extract information about all the arrays into
2532 * For each vcl in conflist: create an mdinfo, fill it in,
2533 * then look for matching devices (phys_refnum) in dlist
2534 * and create appropriate device mdinfo.
2536 struct ddf_super
*ddf
= st
->sb
;
2537 struct mdinfo
*rest
= NULL
;
2540 for (vc
= ddf
->conflist
; vc
; vc
=vc
->next
)
2543 struct mdinfo
*this;
2544 this = malloc(sizeof(*this));
2545 memset(this, 0, sizeof(*this));
2549 this->array
.level
= map_num1(ddf_level_num
, vc
->conf
.prl
);
2550 this->array
.raid_disks
=
2551 __be16_to_cpu(vc
->conf
.prim_elmnt_count
);
2552 this->array
.layout
= rlq_to_layout(vc
->conf
.rlq
, vc
->conf
.prl
,
2553 this->array
.raid_disks
);
2554 this->array
.md_minor
= -1;
2555 this->array
.ctime
= DECADE
+
2556 __be32_to_cpu(*(__u32
*)(vc
->conf
.guid
+16));
2557 this->array
.utime
= DECADE
+
2558 __be32_to_cpu(vc
->conf
.timestamp
);
2559 this->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
2562 if ((ddf
->virt
->entries
[i
].state
& DDF_state_inconsistent
) ||
2563 (ddf
->virt
->entries
[i
].init_state
& DDF_initstate_mask
) !=
2565 this->array
.state
= 0;
2566 this->resync_start
= 0;
2568 this->array
.state
= 1;
2569 this->resync_start
= ~0ULL;
2571 memcpy(this->name
, ddf
->virt
->entries
[i
].name
, 32);
2574 memset(this->uuid
, 0, sizeof(this->uuid
));
2575 this->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
2576 this->array
.size
= this->component_size
/ 2;
2577 this->container_member
= i
;
2579 sprintf(this->text_version
, "/%s/%d",
2580 devnum2devname(st
->container_dev
),
2581 this->container_member
);
2583 for (i
=0 ; i
< ddf
->mppe
; i
++) {
2587 if (vc
->conf
.phys_refnum
[i
] == 0xFFFFFFFF)
2590 this->array
.working_disks
++;
2592 for (d
= ddf
->dlist
; d
; d
=d
->next
)
2593 if (d
->disk
.refnum
== vc
->conf
.phys_refnum
[i
])
2598 dev
= malloc(sizeof(*dev
));
2599 memset(dev
, 0, sizeof(*dev
));
2600 dev
->next
= this->devs
;
2603 dev
->disk
.number
= __be32_to_cpu(d
->disk
.refnum
);
2604 dev
->disk
.major
= d
->major
;
2605 dev
->disk
.minor
= d
->minor
;
2606 dev
->disk
.raid_disk
= i
;
2607 dev
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
2609 dev
->events
= __be32_to_cpu(ddf
->primary
.seq
);
2610 dev
->data_offset
= __be64_to_cpu(vc
->lba_offset
[i
]);
2611 dev
->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
2613 strcpy(dev
->name
, d
->devname
);
2619 static int store_zero_ddf(struct supertype
*st
, int fd
)
2621 unsigned long long dsize
;
2624 if (!get_dev_size(fd
, NULL
, &dsize
))
2627 posix_memalign(&buf
, 512, 512);
2628 memset(buf
, 0, 512);
2630 lseek64(fd
, dsize
-512, 0);
2631 write(fd
, buf
, 512);
2636 static int compare_super_ddf(struct supertype
*st
, struct supertype
*tst
)
2640 * 0 same, or first was empty, and second was copied
2641 * 1 second had wrong number
2643 * 3 wrong other info
2645 struct ddf_super
*first
= st
->sb
;
2646 struct ddf_super
*second
= tst
->sb
;
2654 if (memcmp(first
->anchor
.guid
, second
->anchor
.guid
, DDF_GUID_LEN
) != 0)
2657 /* FIXME should I look at anything else? */
2662 * A new array 'a' has been started which claims to be instance 'inst'
2663 * within container 'c'.
2664 * We need to confirm that the array matches the metadata in 'c' so
2665 * that we don't corrupt any metadata.
2667 static int ddf_open_new(struct supertype
*c
, struct active_array
*a
, char *inst
)
2669 dprintf("ddf: open_new %s\n", inst
);
2670 a
->info
.container_member
= atoi(inst
);
2675 * The array 'a' is to be marked clean in the metadata.
2676 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2677 * clean up to the point (in sectors). If that cannot be recorded in the
2678 * metadata, then leave it as dirty.
2680 * For DDF, we need to clear the DDF_state_inconsistent bit in the
2681 * !global! virtual_disk.virtual_entry structure.
2683 static void ddf_set_array_state(struct active_array
*a
, int consistent
)
2685 struct ddf_super
*ddf
= a
->container
->sb
;
2686 int inst
= a
->info
.container_member
;
2687 int old
= ddf
->virt
->entries
[inst
].state
;
2689 ddf
->virt
->entries
[inst
].state
&= ~DDF_state_inconsistent
;
2691 ddf
->virt
->entries
[inst
].state
|= DDF_state_inconsistent
;
2692 if (old
!= ddf
->virt
->entries
[inst
].state
)
2693 ddf
->updates_pending
= 1;
2695 old
= ddf
->virt
->entries
[inst
].init_state
;
2696 ddf
->virt
->entries
[inst
].init_state
&= ~DDF_initstate_mask
;
2697 if (a
->resync_start
== ~0ULL)
2698 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_full
;
2699 else if (a
->resync_start
== 0)
2700 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_not
;
2702 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_quick
;
2703 if (old
!= ddf
->virt
->entries
[inst
].init_state
)
2704 ddf
->updates_pending
= 1;
2706 dprintf("ddf mark %d %s %llu\n", inst
, consistent
?"clean":"dirty",
2711 * The state of each disk is stored in the global phys_disk structure
2712 * in phys_disk.entries[n].state.
2713 * This makes various combinations awkward.
2714 * - When a device fails in any array, it must be failed in all arrays
2715 * that include a part of this device.
2716 * - When a component is rebuilding, we cannot include it officially in the
2717 * array unless this is the only array that uses the device.
2719 * So: when transitioning:
2720 * Online -> failed, just set failed flag. monitor will propagate
2721 * spare -> online, the device might need to be added to the array.
2722 * spare -> failed, just set failed. Don't worry if in array or not.
2724 static void ddf_set_disk(struct active_array
*a
, int n
, int state
)
2726 struct ddf_super
*ddf
= a
->container
->sb
;
2727 int inst
= a
->info
.container_member
;
2728 struct vd_config
*vc
= find_vdcr(ddf
, inst
);
2729 int pd
= find_phys(ddf
, vc
->phys_refnum
[n
]);
2733 dprintf("ddf: cannot find instance %d!!\n", inst
);
2737 /* disk doesn't currently exist. If it is now in_sync,
2739 if ((state
& DS_INSYNC
) && ! (state
& DS_FAULTY
)) {
2740 /* Find dev 'n' in a->info->devs, determine the
2741 * ddf refnum, and set vc->phys_refnum and update
2747 int old
= ddf
->phys
->entries
[pd
].state
;
2748 if (state
& DS_FAULTY
)
2749 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Failed
);
2750 if (state
& DS_INSYNC
) {
2751 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Online
);
2752 ddf
->phys
->entries
[pd
].state
&= __cpu_to_be16(~DDF_Rebuilding
);
2754 if (old
!= ddf
->phys
->entries
[pd
].state
)
2755 ddf
->updates_pending
= 1;
2758 dprintf("ddf: set_disk %d to %x\n", n
, state
);
2760 /* Now we need to check the state of the array and update
2761 * virtual_disk.entries[n].state.
2762 * It needs to be one of "optimal", "degraded", "failed".
2763 * I don't understand 'deleted' or 'missing'.
2766 for (i
=0; i
< a
->info
.array
.raid_disks
; i
++) {
2767 pd
= find_phys(ddf
, vc
->phys_refnum
[i
]);
2770 st
= __be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
2771 if ((st
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
2775 state
= DDF_state_degraded
;
2776 if (working
== a
->info
.array
.raid_disks
)
2777 state
= DDF_state_optimal
;
2778 else switch(vc
->prl
) {
2782 state
= DDF_state_failed
;
2786 state
= DDF_state_failed
;
2790 if (working
< a
->info
.array
.raid_disks
-1)
2791 state
= DDF_state_failed
;
2794 if (working
< a
->info
.array
.raid_disks
-2)
2795 state
= DDF_state_failed
;
2796 else if (working
== a
->info
.array
.raid_disks
-1)
2797 state
= DDF_state_part_optimal
;
2801 if (ddf
->virt
->entries
[inst
].state
!=
2802 ((ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
2805 ddf
->virt
->entries
[inst
].state
=
2806 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
2808 ddf
->updates_pending
= 1;
2813 static void ddf_sync_metadata(struct supertype
*st
)
2817 * Write all data to all devices.
2818 * Later, we might be able to track whether only local changes
2819 * have been made, or whether any global data has been changed,
2820 * but ddf is sufficiently weird that it probably always
2821 * changes global data ....
2823 struct ddf_super
*ddf
= st
->sb
;
2824 if (!ddf
->updates_pending
)
2826 ddf
->updates_pending
= 0;
2827 __write_init_super_ddf(st
, 0);
2828 dprintf("ddf: sync_metadata\n");
2831 static void ddf_process_update(struct supertype
*st
,
2832 struct metadata_update
*update
)
2834 /* Apply this update to the metadata.
2835 * The first 4 bytes are a DDF_*_MAGIC which guides
2837 * Possible update are:
2838 * DDF_PHYS_RECORDS_MAGIC
2839 * Add a new physical device. Changes to this record
2840 * only happen implicitly.
2841 * used_pdes is the device number.
2842 * DDF_VIRT_RECORDS_MAGIC
2843 * Add a new VD. Possibly also change the 'access' bits.
2844 * populated_vdes is the entry number.
2846 * New or updated VD. the VIRT_RECORD must already
2847 * exist. For an update, phys_refnum and lba_offset
2848 * (at least) are updated, and the VD_CONF must
2849 * be written to precisely those devices listed with
2851 * DDF_SPARE_ASSIGN_MAGIC
2852 * replacement Spare Assignment Record... but for which device?
2855 * - to create a new array, we send a VIRT_RECORD and
2856 * a VD_CONF. Then assemble and start the array.
2857 * - to activate a spare we send a VD_CONF to add the phys_refnum
2858 * and offset. This will also mark the spare as active with
2859 * a spare-assignment record.
2861 struct ddf_super
*ddf
= st
->sb
;
2862 __u32
*magic
= (__u32
*)update
->buf
;
2863 struct phys_disk
*pd
;
2864 struct virtual_disk
*vd
;
2865 struct vd_config
*vc
;
2871 dprintf("Process update %x\n", *magic
);
2874 case DDF_PHYS_RECORDS_MAGIC
:
2876 if (update
->len
!= (sizeof(struct phys_disk
) +
2877 sizeof(struct phys_disk_entry
)))
2879 pd
= (struct phys_disk
*)update
->buf
;
2881 ent
= __be16_to_cpu(pd
->used_pdes
);
2882 if (ent
>= __be16_to_cpu(ddf
->phys
->max_pdes
))
2884 if (!all_ff(ddf
->phys
->entries
[ent
].guid
))
2886 ddf
->phys
->entries
[ent
] = pd
->entries
[0];
2887 ddf
->phys
->used_pdes
= __cpu_to_be16(1 +
2888 __be16_to_cpu(ddf
->phys
->used_pdes
));
2889 ddf
->updates_pending
= 1;
2892 case DDF_VIRT_RECORDS_MAGIC
:
2894 if (update
->len
!= (sizeof(struct virtual_disk
) +
2895 sizeof(struct virtual_entry
)))
2897 vd
= (struct virtual_disk
*)update
->buf
;
2899 ent
= __be16_to_cpu(vd
->populated_vdes
);
2900 if (ent
>= __be16_to_cpu(ddf
->virt
->max_vdes
))
2902 if (!all_ff(ddf
->virt
->entries
[ent
].guid
))
2904 ddf
->virt
->entries
[ent
] = vd
->entries
[0];
2905 ddf
->virt
->populated_vdes
= __cpu_to_be16(1 +
2906 __be16_to_cpu(ddf
->virt
->populated_vdes
));
2907 ddf
->updates_pending
= 1;
2910 case DDF_VD_CONF_MAGIC
:
2911 dprintf("len %d %d\n", update
->len
, ddf
->conf_rec_len
);
2913 mppe
= __be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
2914 if (update
->len
!= ddf
->conf_rec_len
* 512)
2916 vc
= (struct vd_config
*)update
->buf
;
2917 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
2918 if (memcmp(vcl
->conf
.guid
, vc
->guid
, DDF_GUID_LEN
) == 0)
2920 dprintf("vcl = %p\n", vcl
);
2922 /* An update, just copy the phys_refnum and lba_offset
2925 memcpy(vcl
->conf
.phys_refnum
, vc
->phys_refnum
,
2926 mppe
* (sizeof(__u32
) + sizeof(__u64
)));
2929 vcl
= update
->space
;
2930 update
->space
= NULL
;
2931 vcl
->next
= ddf
->conflist
;
2932 memcpy(&vcl
->conf
, vc
, update
->len
);
2933 vcl
->lba_offset
= (__u64
*)
2934 &vcl
->conf
.phys_refnum
[mppe
];
2935 ddf
->conflist
= vcl
;
2937 /* Now make sure vlist is correct for each dl. */
2938 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
2941 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
2942 for (dn
=0; dn
< ddf
->mppe
; dn
++)
2943 if (vcl
->conf
.phys_refnum
[dn
] ==
2945 dprintf("dev %d has %p at %d\n",
2946 dl
->pdnum
, vcl
, vn
);
2947 dl
->vlist
[vn
++] = vcl
;
2950 while (vn
< ddf
->max_part
)
2951 dl
->vlist
[vn
++] = NULL
;
2953 ddf
->phys
->entries
[dl
->pdnum
].type
&=
2954 ~__cpu_to_be16(DDF_Global_Spare
);
2955 ddf
->phys
->entries
[dl
->pdnum
].type
|=
2956 __cpu_to_be16(DDF_Active_in_VD
);
2959 ddf
->phys
->entries
[dl
->pdnum
].type
&=
2960 ~__cpu_to_be16(DDF_Global_Spare
);
2961 ddf
->phys
->entries
[dl
->pdnum
].type
|=
2962 __cpu_to_be16(DDF_Spare
);
2964 if (!dl
->vlist
[0] && !dl
->spare
) {
2965 ddf
->phys
->entries
[dl
->pdnum
].type
|=
2966 __cpu_to_be16(DDF_Global_Spare
);
2967 ddf
->phys
->entries
[dl
->pdnum
].type
&=
2968 ~__cpu_to_be16(DDF_Spare
|
2972 ddf
->updates_pending
= 1;
2974 case DDF_SPARE_ASSIGN_MAGIC
:
2979 static void ddf_prepare_update(struct supertype
*st
,
2980 struct metadata_update
*update
)
2982 /* This update arrived at managemon.
2983 * We are about to pass it to monitor.
2984 * If a malloc is needed, do it here.
2986 struct ddf_super
*ddf
= st
->sb
;
2987 __u32
*magic
= (__u32
*)update
->buf
;
2988 if (*magic
== DDF_VD_CONF_MAGIC
)
2989 posix_memalign(&update
->space
, 512,
2990 offsetof(struct vcl
, conf
)
2991 + ddf
->conf_rec_len
* 512);
2995 * Check if the array 'a' is degraded but not failed.
2996 * If it is, find as many spares as are available and needed and
2997 * arrange for their inclusion.
2998 * We only choose devices which are not already in the array,
2999 * and prefer those with a spare-assignment to this array.
3000 * otherwise we choose global spares - assuming always that
3001 * there is enough room.
3002 * For each spare that we assign, we return an 'mdinfo' which
3003 * describes the position for the device in the array.
3004 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
3005 * the new phys_refnum and lba_offset values.
3007 * Only worry about BVDs at the moment.
3009 static struct mdinfo
*ddf_activate_spare(struct active_array
*a
,
3010 struct metadata_update
**updates
)
3014 struct ddf_super
*ddf
= a
->container
->sb
;
3016 struct mdinfo
*rv
= NULL
;
3018 struct metadata_update
*mu
;
3021 struct vd_config
*vc
;
3024 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
3025 if ((d
->curr_state
& DS_FAULTY
) &&
3027 /* wait for Removal to happen */
3029 if (d
->state_fd
>= 0)
3033 dprintf("ddf_activate: working=%d (%d) level=%d\n", working
, a
->info
.array
.raid_disks
,
3034 a
->info
.array
.level
);
3035 if (working
== a
->info
.array
.raid_disks
)
3036 return NULL
; /* array not degraded */
3037 switch (a
->info
.array
.level
) {
3040 return NULL
; /* failed */
3044 if (working
< a
->info
.array
.raid_disks
- 1)
3045 return NULL
; /* failed */
3048 if (working
< a
->info
.array
.raid_disks
- 2)
3049 return NULL
; /* failed */
3051 default: /* concat or stripe */
3052 return NULL
; /* failed */
3055 /* For each slot, if it is not working, find a spare */
3057 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
3058 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3059 if (d
->disk
.raid_disk
== i
)
3061 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
3062 if (d
&& (d
->state_fd
>= 0))
3065 /* OK, this device needs recovery. Find a spare */
3067 for ( ; dl
; dl
= dl
->next
) {
3068 unsigned long long esize
;
3069 unsigned long long pos
;
3072 int is_dedicated
= 0;
3075 /* If in this array, skip */
3076 for (d2
= a
->info
.devs
; d2
; d2
= d2
->next
)
3077 if (d2
->disk
.major
== dl
->major
&&
3078 d2
->disk
.minor
== dl
->minor
) {
3079 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
3084 if (ddf
->phys
->entries
[dl
->pdnum
].type
&
3085 __cpu_to_be16(DDF_Spare
)) {
3086 /* Check spare assign record */
3088 if (dl
->spare
->type
& DDF_spare_dedicated
) {
3089 /* check spare_ents for guid */
3091 j
< __be16_to_cpu(dl
->spare
->populated
);
3093 if (memcmp(dl
->spare
->spare_ents
[j
].guid
,
3094 ddf
->virt
->entries
[a
->info
.container_member
].guid
,
3101 } else if (ddf
->phys
->entries
[dl
->pdnum
].type
&
3102 __cpu_to_be16(DDF_Global_Spare
)) {
3105 if ( ! (is_dedicated
||
3106 (is_global
&& global_ok
))) {
3107 dprintf("%x:%x not suitable: %d %d\n", dl
->major
, dl
->minor
,
3108 is_dedicated
, is_global
);
3112 /* We are allowed to use this device - is there space?
3113 * We need a->info.component_size sectors */
3114 ex
= get_extents(ddf
, dl
);
3116 dprintf("cannot get extents\n");
3123 esize
= ex
[j
].start
- pos
;
3124 if (esize
>= a
->info
.component_size
)
3126 pos
= ex
[i
].start
+ ex
[i
].size
;
3128 } while (ex
[i
-1].size
);
3131 if (esize
< a
->info
.component_size
) {
3132 dprintf("%x:%x has no room: %llu %llu\n", dl
->major
, dl
->minor
,
3133 esize
, a
->info
.component_size
);
3138 /* Cool, we have a device with some space at pos */
3139 di
= malloc(sizeof(*di
));
3140 memset(di
, 0, sizeof(*di
));
3141 di
->disk
.number
= i
;
3142 di
->disk
.raid_disk
= i
;
3143 di
->disk
.major
= dl
->major
;
3144 di
->disk
.minor
= dl
->minor
;
3146 di
->data_offset
= pos
;
3147 di
->component_size
= a
->info
.component_size
;
3148 di
->container_member
= dl
->pdnum
;
3151 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
3156 if (!dl
&& ! global_ok
) {
3157 /* not enough dedicated spares, try global */
3165 /* No spares found */
3167 /* Now 'rv' has a list of devices to return.
3168 * Create a metadata_update record to update the
3169 * phys_refnum and lba_offset values
3171 mu
= malloc(sizeof(*mu
));
3172 mu
->buf
= malloc(ddf
->conf_rec_len
* 512);
3173 posix_memalign(&mu
->space
, 512, sizeof(struct vcl
));
3174 mu
->len
= ddf
->conf_rec_len
;
3175 mu
->next
= *updates
;
3176 vc
= find_vdcr(ddf
, a
->info
.container_member
);
3177 memcpy(mu
->buf
, vc
, ddf
->conf_rec_len
* 512);
3179 vc
= (struct vd_config
*)mu
->buf
;
3180 lba
= (__u64
*)&vc
->phys_refnum
[ddf
->mppe
];
3181 for (di
= rv
; di
; di
= di
->next
) {
3182 vc
->phys_refnum
[di
->disk
.raid_disk
] =
3183 ddf
->phys
->entries
[dl
->pdnum
].refnum
;
3184 lba
[di
->disk
.raid_disk
] = di
->data_offset
;
3190 struct superswitch super_ddf
= {
3192 .examine_super
= examine_super_ddf
,
3193 .brief_examine_super
= brief_examine_super_ddf
,
3194 .detail_super
= detail_super_ddf
,
3195 .brief_detail_super
= brief_detail_super_ddf
,
3196 .validate_geometry
= validate_geometry_ddf
,
3197 .write_init_super
= write_init_super_ddf
,
3199 .match_home
= match_home_ddf
,
3200 .uuid_from_super
= uuid_from_super_ddf
,
3201 .getinfo_super
= getinfo_super_ddf
,
3202 .update_super
= update_super_ddf
,
3204 .avail_size
= avail_size_ddf
,
3206 .compare_super
= compare_super_ddf
,
3208 .load_super
= load_super_ddf
,
3209 .init_super
= init_super_ddf
,
3210 .store_super
= store_zero_ddf
,
3211 .free_super
= free_super_ddf
,
3212 .match_metadata_desc
= match_metadata_desc_ddf
,
3213 .add_to_super
= add_to_super_ddf
,
3214 .container_content
= container_content_ddf
,
3219 .open_new
= ddf_open_new
,
3220 .set_array_state
= ddf_set_array_state
,
3221 .set_disk
= ddf_set_disk
,
3222 .sync_metadata
= ddf_sync_metadata
,
3223 .process_update
= ddf_process_update
,
3224 .prepare_update
= ddf_prepare_update
,
3225 .activate_spare
= ddf_activate_spare
,