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 void *mdupdate
; /* hold metadata update */
430 struct vcl
*vlist
[0]; /* max_part in size */
435 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
439 static int calc_crc(void *buf
, int len
)
441 /* crcs are always at the same place as in the ddf_header */
442 struct ddf_header
*ddf
= buf
;
443 __u32 oldcrc
= ddf
->crc
;
445 ddf
->crc
= 0xffffffff;
447 newcrc
= crc32(0, buf
, len
);
452 static int load_ddf_header(int fd
, unsigned long long lba
,
453 unsigned long long size
,
455 struct ddf_header
*hdr
, struct ddf_header
*anchor
)
457 /* read a ddf header (primary or secondary) from fd/lba
458 * and check that it is consistent with anchor
460 * magic, crc, guid, rev, and LBA's header_type, and
461 * everything after header_type must be the same
466 if (lseek64(fd
, lba
<<9, 0) < 0)
469 if (read(fd
, hdr
, 512) != 512)
472 if (hdr
->magic
!= DDF_HEADER_MAGIC
)
474 if (calc_crc(hdr
, 512) != hdr
->crc
)
476 if (memcmp(anchor
->guid
, hdr
->guid
, DDF_GUID_LEN
) != 0 ||
477 memcmp(anchor
->revision
, hdr
->revision
, 8) != 0 ||
478 anchor
->primary_lba
!= hdr
->primary_lba
||
479 anchor
->secondary_lba
!= hdr
->secondary_lba
||
481 memcmp(anchor
->pad2
, hdr
->pad2
, 512 -
482 offsetof(struct ddf_header
, pad2
)) != 0)
485 /* Looks good enough to me... */
489 static void *load_section(int fd
, struct ddf_super
*super
, void *buf
,
490 __u32 offset_be
, __u32 len_be
, int check
)
492 unsigned long long offset
= __be32_to_cpu(offset_be
);
493 unsigned long long len
= __be32_to_cpu(len_be
);
494 int dofree
= (buf
== NULL
);
497 if (len
!= 2 && len
!= 8 && len
!= 32
498 && len
!= 128 && len
!= 512)
504 /* All pre-allocated sections are a single block */
508 posix_memalign(&buf
, 512, len
<<9);
514 if (super
->active
->type
== 1)
515 offset
+= __be64_to_cpu(super
->active
->primary_lba
);
517 offset
+= __be64_to_cpu(super
->active
->secondary_lba
);
519 if (lseek64(fd
, offset
<<9, 0) != (offset
<<9)) {
524 if (read(fd
, buf
, len
<<9) != (len
<<9)) {
532 static int load_ddf_headers(int fd
, struct ddf_super
*super
, char *devname
)
534 unsigned long long dsize
;
536 get_dev_size(fd
, NULL
, &dsize
);
538 if (lseek64(fd
, dsize
-512, 0) < 0) {
541 Name
": Cannot seek to anchor block on %s: %s\n",
542 devname
, strerror(errno
));
545 if (read(fd
, &super
->anchor
, 512) != 512) {
548 Name
": Cannot read anchor block on %s: %s\n",
549 devname
, strerror(errno
));
552 if (super
->anchor
.magic
!= DDF_HEADER_MAGIC
) {
554 fprintf(stderr
, Name
": no DDF anchor found on %s\n",
558 if (calc_crc(&super
->anchor
, 512) != super
->anchor
.crc
) {
560 fprintf(stderr
, Name
": bad CRC on anchor on %s\n",
564 if (memcmp(super
->anchor
.revision
, DDF_REVISION_0
, 8) != 0 &&
565 memcmp(super
->anchor
.revision
, DDF_REVISION_2
, 8) != 0) {
567 fprintf(stderr
, Name
": can only support super revision"
568 " %.8s and earlier, not %.8s on %s\n",
569 DDF_REVISION_2
, super
->anchor
.revision
,devname
);
572 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.primary_lba
),
574 &super
->primary
, &super
->anchor
) == 0) {
577 Name
": Failed to load primary DDF header "
581 super
->active
= &super
->primary
;
582 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.secondary_lba
),
584 &super
->secondary
, &super
->anchor
)) {
585 if ((__be32_to_cpu(super
->primary
.seq
)
586 < __be32_to_cpu(super
->secondary
.seq
) &&
587 !super
->secondary
.openflag
)
588 || (__be32_to_cpu(super
->primary
.seq
)
589 == __be32_to_cpu(super
->secondary
.seq
) &&
590 super
->primary
.openflag
&& !super
->secondary
.openflag
)
592 super
->active
= &super
->secondary
;
597 static int load_ddf_global(int fd
, struct ddf_super
*super
, char *devname
)
600 ok
= load_section(fd
, super
, &super
->controller
,
601 super
->active
->controller_section_offset
,
602 super
->active
->controller_section_length
,
604 super
->phys
= load_section(fd
, super
, NULL
,
605 super
->active
->phys_section_offset
,
606 super
->active
->phys_section_length
,
608 super
->pdsize
= __be32_to_cpu(super
->active
->phys_section_length
) * 512;
610 super
->virt
= load_section(fd
, super
, NULL
,
611 super
->active
->virt_section_offset
,
612 super
->active
->virt_section_length
,
614 super
->vdsize
= __be32_to_cpu(super
->active
->virt_section_length
) * 512;
624 super
->conflist
= NULL
;
627 super
->max_part
= __be16_to_cpu(super
->active
->max_partitions
);
628 super
->mppe
= __be16_to_cpu(super
->active
->max_primary_element_entries
);
629 super
->conf_rec_len
= __be16_to_cpu(super
->active
->config_record_len
);
633 static int load_ddf_local(int fd
, struct ddf_super
*super
,
634 char *devname
, int keep
)
641 int max_virt_disks
= __be16_to_cpu(super
->active
->max_vd_entries
);
642 unsigned long long dsize
;
644 /* First the local disk info */
645 posix_memalign((void**)&dl
, 512,
647 (super
->max_part
) * sizeof(dl
->vlist
[0]));
649 load_section(fd
, super
, &dl
->disk
,
650 super
->active
->data_section_offset
,
651 super
->active
->data_section_length
,
653 dl
->devname
= devname
? strdup(devname
) : NULL
;
656 dl
->major
= major(stb
.st_rdev
);
657 dl
->minor
= minor(stb
.st_rdev
);
658 dl
->next
= super
->dlist
;
659 dl
->fd
= keep
? fd
: -1;
662 if (get_dev_size(fd
, devname
, &dsize
))
663 dl
->size
= dsize
>> 9;
665 for (i
=0 ; i
< super
->max_part
; i
++)
669 for (i
=0; i
< __be16_to_cpu(super
->active
->max_pd_entries
); i
++)
670 if (memcmp(super
->phys
->entries
[i
].guid
,
671 dl
->disk
.guid
, DDF_GUID_LEN
) == 0)
674 /* Now the config list. */
675 /* 'conf' is an array of config entries, some of which are
676 * probably invalid. Those which are good need to be copied into
680 conf
= load_section(fd
, super
, NULL
,
681 super
->active
->config_section_offset
,
682 super
->active
->config_section_length
,
687 i
< __be32_to_cpu(super
->active
->config_section_length
);
688 i
+= super
->conf_rec_len
) {
689 struct vd_config
*vd
=
690 (struct vd_config
*)((char*)conf
+ i
*512);
693 if (vd
->magic
== DDF_SPARE_ASSIGN_MAGIC
) {
696 posix_memalign((void**)&dl
->spare
, 512,
697 super
->conf_rec_len
*512);
698 memcpy(dl
->spare
, vd
, super
->conf_rec_len
*512);
701 if (vd
->magic
!= DDF_VD_CONF_MAGIC
)
703 for (vcl
= super
->conflist
; vcl
; vcl
= vcl
->next
) {
704 if (memcmp(vcl
->conf
.guid
,
705 vd
->guid
, DDF_GUID_LEN
) == 0)
710 dl
->vlist
[vnum
++] = vcl
;
711 if (__be32_to_cpu(vd
->seqnum
) <=
712 __be32_to_cpu(vcl
->conf
.seqnum
))
715 posix_memalign((void**)&vcl
, 512,
716 (super
->conf_rec_len
*512 +
717 offsetof(struct vcl
, conf
)));
718 vcl
->next
= super
->conflist
;
719 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
720 super
->conflist
= vcl
;
721 dl
->vlist
[vnum
++] = vcl
;
723 memcpy(&vcl
->conf
, vd
, super
->conf_rec_len
*512);
724 vcl
->lba_offset
= (__u64
*)
725 &vcl
->conf
.phys_refnum
[super
->mppe
];
727 for (i
=0; i
< max_virt_disks
; i
++)
728 if (memcmp(super
->virt
->entries
[i
].guid
,
729 vcl
->conf
.guid
, DDF_GUID_LEN
)==0)
731 if (i
< max_virt_disks
)
740 static int load_super_ddf_all(struct supertype
*st
, int fd
,
741 void **sbp
, char *devname
, int keep_fd
);
743 static int load_super_ddf(struct supertype
*st
, int fd
,
746 unsigned long long dsize
;
747 struct ddf_super
*super
;
751 /* if 'fd' is a container, load metadata from all the devices */
752 if (load_super_ddf_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
756 return 1; /* FIXME Is this correct */
758 if (get_dev_size(fd
, devname
, &dsize
) == 0)
761 /* 32M is a lower bound */
762 if (dsize
<= 32*1024*1024) {
765 Name
": %s is too small for ddf: "
766 "size is %llu sectors.\n",
774 Name
": %s is an odd size for ddf: "
775 "size is %llu bytes.\n",
781 if (posix_memalign((void**)&super
, 512, sizeof(*super
))!= 0) {
782 fprintf(stderr
, Name
": malloc of %zu failed.\n",
786 memset(super
, 0, sizeof(*super
));
788 rv
= load_ddf_headers(fd
, super
, devname
);
794 /* Have valid headers and have chosen the best. Let's read in the rest*/
796 rv
= load_ddf_global(fd
, super
, devname
);
801 Name
": Failed to load all information "
802 "sections on %s\n", devname
);
807 load_ddf_local(fd
, super
, devname
, 0);
809 /* Should possibly check the sections .... */
812 if (st
->ss
== NULL
) {
814 st
->minor_version
= 0;
821 static void free_super_ddf(struct supertype
*st
)
823 struct ddf_super
*ddf
= st
->sb
;
828 while (ddf
->conflist
) {
829 struct vcl
*v
= ddf
->conflist
;
830 ddf
->conflist
= v
->next
;
832 free(v
->block_sizes
);
836 struct dl
*d
= ddf
->dlist
;
837 ddf
->dlist
= d
->next
;
848 static struct supertype
*match_metadata_desc_ddf(char *arg
)
850 /* 'ddf' only support containers */
851 struct supertype
*st
;
852 if (strcmp(arg
, "ddf") != 0 &&
853 strcmp(arg
, "default") != 0
857 st
= malloc(sizeof(*st
));
858 memset(st
, 0, sizeof(*st
));
861 st
->minor_version
= 0;
869 static mapping_t ddf_state
[] = {
875 { "Partially Optimal", 5},
881 static mapping_t ddf_init_state
[] = {
882 { "Not Initialised", 0},
883 { "QuickInit in Progress", 1},
884 { "Fully Initialised", 2},
888 static mapping_t ddf_access
[] = {
892 { "Blocked (no access)", 3},
896 static mapping_t ddf_level
[] = {
897 { "RAID0", DDF_RAID0
},
898 { "RAID1", DDF_RAID1
},
899 { "RAID3", DDF_RAID3
},
900 { "RAID4", DDF_RAID4
},
901 { "RAID5", DDF_RAID5
},
902 { "RAID1E",DDF_RAID1E
},
904 { "CONCAT",DDF_CONCAT
},
905 { "RAID5E",DDF_RAID5E
},
906 { "RAID5EE",DDF_RAID5EE
},
907 { "RAID6", DDF_RAID6
},
910 static mapping_t ddf_sec_level
[] = {
911 { "Striped", DDF_2STRIPED
},
912 { "Mirrored", DDF_2MIRRORED
},
913 { "Concat", DDF_2CONCAT
},
914 { "Spanned", DDF_2SPANNED
},
922 static struct num_mapping ddf_level_num
[] = {
925 { DDF_RAID3
, LEVEL_UNSUPPORTED
},
928 { DDF_RAID1E
, LEVEL_UNSUPPORTED
},
929 { DDF_JBOD
, LEVEL_UNSUPPORTED
},
930 { DDF_CONCAT
, LEVEL_LINEAR
},
931 { DDF_RAID5E
, LEVEL_UNSUPPORTED
},
932 { DDF_RAID5EE
, LEVEL_UNSUPPORTED
},
937 static int map_num1(struct num_mapping
*map
, int num
)
940 for (i
=0 ; map
[i
].num1
!= MAXINT
; i
++)
941 if (map
[i
].num1
== num
)
947 static void print_guid(char *guid
, int tstamp
)
949 /* A GUIDs are part (or all) ASCII and part binary.
950 * They tend to be space padded.
951 * We print the GUID in HEX, then in parentheses add
952 * any initial ASCII sequence, and a possible
953 * time stamp from bytes 16-19
955 int l
= DDF_GUID_LEN
;
958 for (i
=0 ; i
<DDF_GUID_LEN
; i
++) {
959 if ((i
&3)==0 && i
!= 0) printf(":");
960 printf("%02X", guid
[i
]&255);
964 while (l
&& guid
[l
-1] == ' ')
966 for (i
=0 ; i
<l
; i
++) {
967 if (guid
[i
] >= 0x20 && guid
[i
] < 0x7f)
968 fputc(guid
[i
], stdout
);
973 time_t then
= __be32_to_cpu(*(__u32
*)(guid
+16)) + DECADE
;
976 tm
= localtime(&then
);
977 strftime(tbuf
, 100, " %D %T",tm
);
983 static void examine_vd(int n
, struct ddf_super
*sb
, char *guid
)
985 int crl
= sb
->conf_rec_len
;
988 for (vcl
= sb
->conflist
; vcl
; vcl
= vcl
->next
) {
989 struct vd_config
*vc
= &vcl
->conf
;
991 if (calc_crc(vc
, crl
*512) != vc
->crc
)
993 if (memcmp(vc
->guid
, guid
, DDF_GUID_LEN
) != 0)
996 /* Ok, we know about this VD, let's give more details */
997 printf(" Raid Devices[%d] : %d\n", n
,
998 __be16_to_cpu(vc
->prim_elmnt_count
));
999 printf(" Chunk Size[%d] : %d sectors\n", n
,
1000 1 << vc
->chunk_shift
);
1001 printf(" Raid Level[%d] : %s\n", n
,
1002 map_num(ddf_level
, vc
->prl
)?:"-unknown-");
1003 if (vc
->sec_elmnt_count
!= 1) {
1004 printf(" Secondary Position[%d] : %d of %d\n", n
,
1005 vc
->sec_elmnt_seq
, vc
->sec_elmnt_count
);
1006 printf(" Secondary Level[%d] : %s\n", n
,
1007 map_num(ddf_sec_level
, vc
->srl
) ?: "-unknown-");
1009 printf(" Device Size[%d] : %llu\n", n
,
1010 __be64_to_cpu(vc
->blocks
)/2);
1011 printf(" Array Size[%d] : %llu\n", n
,
1012 __be64_to_cpu(vc
->array_blocks
)/2);
1016 static void examine_vds(struct ddf_super
*sb
)
1018 int cnt
= __be16_to_cpu(sb
->virt
->populated_vdes
);
1020 printf(" Virtual Disks : %d\n", cnt
);
1022 for (i
=0; i
<cnt
; i
++) {
1023 struct virtual_entry
*ve
= &sb
->virt
->entries
[i
];
1024 printf(" VD GUID[%d] : ", i
); print_guid(ve
->guid
, 1);
1026 printf(" unit[%d] : %d\n", i
, __be16_to_cpu(ve
->unit
));
1027 printf(" state[%d] : %s, %s%s\n", i
,
1028 map_num(ddf_state
, ve
->state
& 7),
1029 (ve
->state
& 8) ? "Morphing, ": "",
1030 (ve
->state
& 16)? "Not Consistent" : "Consistent");
1031 printf(" init state[%d] : %s\n", i
,
1032 map_num(ddf_init_state
, ve
->init_state
&3));
1033 printf(" access[%d] : %s\n", i
,
1034 map_num(ddf_access
, (ve
->init_state
>>6) & 3));
1035 printf(" Name[%d] : %.16s\n", i
, ve
->name
);
1036 examine_vd(i
, sb
, ve
->guid
);
1038 if (cnt
) printf("\n");
1041 static void examine_pds(struct ddf_super
*sb
)
1043 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1046 printf(" Physical Disks : %d\n", cnt
);
1048 for (i
=0 ; i
<cnt
; i
++) {
1049 struct phys_disk_entry
*pd
= &sb
->phys
->entries
[i
];
1050 int type
= __be16_to_cpu(pd
->type
);
1051 int state
= __be16_to_cpu(pd
->state
);
1053 printf(" PD GUID[%d] : ", i
); print_guid(pd
->guid
, 0);
1055 printf(" ref[%d] : %08x\n", i
,
1056 __be32_to_cpu(pd
->refnum
));
1057 printf(" mode[%d] : %s%s%s%s%s\n", i
,
1058 (type
&2) ? "active":"",
1059 (type
&4) ? "Global Spare":"",
1060 (type
&8) ? "spare" : "",
1061 (type
&16)? ", foreign" : "",
1062 (type
&32)? "pass-through" : "");
1063 printf(" state[%d] : %s%s%s%s%s%s%s\n", i
,
1064 (state
&1)? "Online": "Offline",
1065 (state
&2)? ", Failed": "",
1066 (state
&4)? ", Rebuilding": "",
1067 (state
&8)? ", in-transition": "",
1068 (state
&16)? ", SMART errors": "",
1069 (state
&32)? ", Unrecovered Read Errors": "",
1070 (state
&64)? ", Missing" : "");
1071 printf(" Avail Size[%d] : %llu K\n", i
,
1072 __be64_to_cpu(pd
->config_size
)>>1);
1073 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1074 if (dl
->disk
.refnum
== pd
->refnum
) {
1075 char *dv
= map_dev(dl
->major
, dl
->minor
, 0);
1077 printf(" Device[%d] : %s\n",
1085 static void examine_super_ddf(struct supertype
*st
, char *homehost
)
1087 struct ddf_super
*sb
= st
->sb
;
1089 printf(" Magic : %08x\n", __be32_to_cpu(sb
->anchor
.magic
));
1090 printf(" Version : %.8s\n", sb
->anchor
.revision
);
1091 printf("Controller GUID : "); print_guid(sb
->controller
.guid
, 0);
1093 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1095 printf(" Seq : %08x\n", __be32_to_cpu(sb
->active
->seq
));
1096 printf(" Redundant hdr : %s\n", sb
->secondary
.magic
== DDF_HEADER_MAGIC
1102 static void brief_examine_super_ddf(struct supertype
*st
)
1104 /* We just write a generic DDF ARRAY entry
1105 * The uuid is all hex, 6 groups of 4 bytes
1107 struct ddf_super
*ddf
= st
->sb
;
1109 printf("ARRAY /dev/ddf metadata=ddf UUID=");
1110 for (i
= 0; i
< DDF_GUID_LEN
; i
++) {
1111 if ((i
&3) == 0 && i
!= 0)
1113 printf("%02X", 255&ddf
->anchor
.guid
[i
]);
1118 static void detail_super_ddf(struct supertype
*st
, char *homehost
)
1121 * Could print DDF GUID
1122 * Need to find which array
1123 * If whole, briefly list all arrays
1128 static void brief_detail_super_ddf(struct supertype
*st
)
1130 /* FIXME I really need to know which array we are detailing.
1131 * Can that be stored in ddf_super??
1133 // struct ddf_super *ddf = st->sb;
1137 static int match_home_ddf(struct supertype
*st
, char *homehost
)
1139 /* It matches 'this' host if the controller is a
1140 * Linux-MD controller with vendor_data matching
1143 struct ddf_super
*ddf
= st
->sb
;
1144 int len
= strlen(homehost
);
1146 return (memcmp(ddf
->controller
.guid
, T10
, 8) == 0 &&
1147 len
< sizeof(ddf
->controller
.vendor_data
) &&
1148 memcmp(ddf
->controller
.vendor_data
, homehost
,len
) == 0 &&
1149 ddf
->controller
.vendor_data
[len
] == 0);
1152 static struct vd_config
*find_vdcr(struct ddf_super
*ddf
, int inst
)
1156 for (v
= ddf
->conflist
; v
; v
= v
->next
)
1157 if (inst
== v
->vcnum
)
1162 static int find_phys(struct ddf_super
*ddf
, __u32 phys_refnum
)
1164 /* Find the entry in phys_disk which has the given refnum
1165 * and return it's index
1168 for (i
=0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
); i
++)
1169 if (ddf
->phys
->entries
[i
].refnum
== phys_refnum
)
1174 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4])
1176 /* The uuid returned here is used for:
1177 * uuid to put into bitmap file (Create, Grow)
1178 * uuid for backup header when saving critical section (Grow)
1179 * comparing uuids when re-adding a device into an array
1180 * For each of these we can make do with a truncated
1181 * or hashed uuid rather than the original, as long as
1183 * In each case the uuid required is that of the data-array,
1184 * not the device-set.
1185 * In the case of SVD we assume the BVD is of interest,
1186 * though that might be the case if a bitmap were made for
1187 * a mirrored SVD - worry about that later.
1188 * So we need to find the VD configuration record for the
1189 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1190 * The first 16 bytes of the sha1 of these is used.
1192 struct ddf_super
*ddf
= st
->sb
;
1193 struct vcl
*vcl
= ddf
->currentconf
;
1196 memset(uuid
, 0, sizeof (uuid
));
1199 struct sha1_ctx ctx
;
1200 sha1_init_ctx(&ctx
);
1201 sha1_process_bytes(&vcl
->conf
.guid
, DDF_GUID_LEN
, &ctx
);
1202 if (vcl
->conf
.sec_elmnt_count
> 1)
1203 sha1_process_bytes(&vcl
->conf
.sec_elmnt_seq
, 1, &ctx
);
1204 sha1_finish_ctx(&ctx
, buf
);
1205 memcpy(uuid
, buf
, sizeof(uuid
));
1209 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
);
1211 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
)
1213 struct ddf_super
*ddf
= st
->sb
;
1215 if (ddf
->currentconf
) {
1216 getinfo_super_ddf_bvd(st
, info
);
1220 info
->array
.raid_disks
= __be16_to_cpu(ddf
->phys
->used_pdes
);
1221 info
->array
.level
= LEVEL_CONTAINER
;
1222 info
->array
.layout
= 0;
1223 info
->array
.md_minor
= -1;
1224 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*(__u32
*)
1225 (ddf
->anchor
.guid
+16));
1226 info
->array
.utime
= 0;
1227 info
->array
.chunk_size
= 0;
1230 info
->disk
.major
= 0;
1231 info
->disk
.minor
= 0;
1233 info
->disk
.number
= __be32_to_cpu(ddf
->dlist
->disk
.refnum
);
1234 info
->disk
.raid_disk
= find_phys(ddf
, ddf
->dlist
->disk
.refnum
);
1236 info
->data_offset
= __be64_to_cpu(ddf
->phys
->
1237 entries
[info
->disk
.raid_disk
].
1239 info
->component_size
= ddf
->dlist
->size
- info
->data_offset
;
1241 info
->disk
.number
= -1;
1242 // info->disk.raid_disk = find refnum in the table and use index;
1244 info
->disk
.state
= (1 << MD_DISK_SYNC
);
1247 info
->reshape_active
= 0;
1249 strcpy(info
->text_version
, "ddf");
1250 info
->safe_mode_delay
= 0;
1252 // uuid_from_super_ddf(info->uuid, sbv);
1254 // info->name[] ?? ;
1257 static int rlq_to_layout(int rlq
, int prl
, int raiddisks
);
1259 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
)
1261 struct ddf_super
*ddf
= st
->sb
;
1262 struct vcl
*vc
= ddf
->currentconf
;
1263 int cd
= ddf
->currentdev
;
1265 /* FIXME this returns BVD info - what if we want SVD ?? */
1267 info
->array
.raid_disks
= __be16_to_cpu(vc
->conf
.prim_elmnt_count
);
1268 info
->array
.level
= map_num1(ddf_level_num
, vc
->conf
.prl
);
1269 info
->array
.layout
= rlq_to_layout(vc
->conf
.rlq
, vc
->conf
.prl
,
1270 info
->array
.raid_disks
);
1271 info
->array
.md_minor
= -1;
1272 info
->array
.ctime
= DECADE
+
1273 __be32_to_cpu(*(__u32
*)(vc
->conf
.guid
+16));
1274 info
->array
.utime
= DECADE
+ __be32_to_cpu(vc
->conf
.timestamp
);
1275 info
->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
1277 if (cd
>= 0 && cd
< ddf
->mppe
) {
1278 info
->data_offset
= __be64_to_cpu(vc
->lba_offset
[cd
]);
1279 if (vc
->block_sizes
)
1280 info
->component_size
= vc
->block_sizes
[cd
];
1282 info
->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
1285 info
->disk
.major
= 0;
1286 info
->disk
.minor
= 0;
1287 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1288 // info->disk.raid_disk = find refnum in the table and use index;
1289 // info->disk.state = ???;
1291 info
->container_member
= ddf
->currentconf
->vcnum
;
1293 info
->resync_start
= 0;
1294 if (!(ddf
->virt
->entries
[info
->container_member
].state
1295 & DDF_state_inconsistent
) &&
1296 (ddf
->virt
->entries
[info
->container_member
].init_state
1297 & DDF_initstate_mask
)
1299 info
->resync_start
= ~0ULL;
1301 uuid_from_super_ddf(st
, info
->uuid
);
1303 info
->container_member
= atoi(st
->subarray
);
1304 sprintf(info
->text_version
, "/%s/%s",
1305 devnum2devname(st
->container_dev
),
1307 info
->safe_mode_delay
= 200;
1309 // info->name[] ?? ;
1313 static int update_super_ddf(struct supertype
*st
, struct mdinfo
*info
,
1315 char *devname
, int verbose
,
1316 int uuid_set
, char *homehost
)
1318 /* For 'assemble' and 'force' we need to return non-zero if any
1319 * change was made. For others, the return value is ignored.
1320 * Update options are:
1321 * force-one : This device looks a bit old but needs to be included,
1322 * update age info appropriately.
1323 * assemble: clear any 'faulty' flag to allow this device to
1325 * force-array: Array is degraded but being forced, mark it clean
1326 * if that will be needed to assemble it.
1328 * newdev: not used ????
1329 * grow: Array has gained a new device - this is currently for
1331 * resync: mark as dirty so a resync will happen.
1332 * uuid: Change the uuid of the array to match what is given
1333 * homehost: update the recorded homehost
1334 * name: update the name - preserving the homehost
1335 * _reshape_progress: record new reshape_progress position.
1337 * Following are not relevant for this version:
1338 * sparc2.2 : update from old dodgey metadata
1339 * super-minor: change the preferred_minor number
1340 * summaries: update redundant counters.
1343 // struct ddf_super *ddf = st->sb;
1344 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1345 // struct virtual_entry *ve = find_ve(ddf);
1347 /* we don't need to handle "force-*" or "assemble" as
1348 * there is no need to 'trick' the kernel. We the metadata is
1349 * first updated to activate the array, all the implied modifications
1353 if (strcmp(update
, "grow") == 0) {
1356 if (strcmp(update
, "resync") == 0) {
1357 // info->resync_checkpoint = 0;
1359 /* We ignore UUID updates as they make even less sense
1362 if (strcmp(update
, "homehost") == 0) {
1363 /* homehost is stored in controller->vendor_data,
1364 * or it is when we are the vendor
1366 // if (info->vendor_is_local)
1367 // strcpy(ddf->controller.vendor_data, homehost);
1369 if (strcmp(update
, "name") == 0) {
1370 /* name is stored in virtual_entry->name */
1371 // memset(ve->name, ' ', 16);
1372 // strncpy(ve->name, info->name, 16);
1374 if (strcmp(update
, "_reshape_progress") == 0) {
1375 /* We don't support reshape yet */
1378 // update_all_csum(ddf);
1383 static void make_header_guid(char *guid
)
1387 /* Create a DDF Header of Virtual Disk GUID */
1389 /* 24 bytes of fiction required.
1390 * first 8 are a 'vendor-id' - "Linux-MD"
1391 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1392 * Remaining 8 random number plus timestamp
1394 memcpy(guid
, T10
, sizeof(T10
));
1395 stamp
= __cpu_to_be32(0xdeadbeef);
1396 memcpy(guid
+8, &stamp
, 4);
1397 stamp
= __cpu_to_be32(0);
1398 memcpy(guid
+12, &stamp
, 4);
1399 stamp
= __cpu_to_be32(time(0) - DECADE
);
1400 memcpy(guid
+16, &stamp
, 4);
1401 rfd
= open("/dev/urandom", O_RDONLY
);
1402 if (rfd
< 0 || read(rfd
, &stamp
, 4) != 4)
1404 memcpy(guid
+20, &stamp
, 4);
1405 if (rfd
>= 0) close(rfd
);
1408 static int init_super_ddf_bvd(struct supertype
*st
,
1409 mdu_array_info_t
*info
,
1410 unsigned long long size
,
1411 char *name
, char *homehost
,
1414 static int init_super_ddf(struct supertype
*st
,
1415 mdu_array_info_t
*info
,
1416 unsigned long long size
, char *name
, char *homehost
,
1419 /* This is primarily called by Create when creating a new array.
1420 * We will then get add_to_super called for each component, and then
1421 * write_init_super called to write it out to each device.
1422 * For DDF, Create can create on fresh devices or on a pre-existing
1424 * To create on a pre-existing array a different method will be called.
1425 * This one is just for fresh drives.
1427 * We need to create the entire 'ddf' structure which includes:
1428 * DDF headers - these are easy.
1429 * Controller data - a Sector describing this controller .. not that
1430 * this is a controller exactly.
1431 * Physical Disk Record - one entry per device, so
1432 * leave plenty of space.
1433 * Virtual Disk Records - again, just leave plenty of space.
1434 * This just lists VDs, doesn't give details
1435 * Config records - describes the VDs that use this disk
1436 * DiskData - describes 'this' device.
1437 * BadBlockManagement - empty
1438 * Diag Space - empty
1439 * Vendor Logs - Could we put bitmaps here?
1442 struct ddf_super
*ddf
;
1445 int max_phys_disks
, max_virt_disks
;
1446 unsigned long long sector
;
1450 struct phys_disk
*pd
;
1451 struct virtual_disk
*vd
;
1458 return init_super_ddf_bvd(st
, info
, size
, name
, homehost
,
1461 posix_memalign((void**)&ddf
, 512, sizeof(*ddf
));
1462 memset(ddf
, 0, sizeof(*ddf
));
1463 ddf
->dlist
= NULL
; /* no physical disks yet */
1464 ddf
->conflist
= NULL
; /* No virtual disks yet */
1466 /* At least 32MB *must* be reserved for the ddf. So let's just
1467 * start 32MB from the end, and put the primary header there.
1468 * Don't do secondary for now.
1469 * We don't know exactly where that will be yet as it could be
1470 * different on each device. To just set up the lengths.
1474 ddf
->anchor
.magic
= DDF_HEADER_MAGIC
;
1475 make_header_guid(ddf
->anchor
.guid
);
1477 memcpy(ddf
->anchor
.revision
, DDF_REVISION_2
, 8);
1478 ddf
->anchor
.seq
= __cpu_to_be32(1);
1479 ddf
->anchor
.timestamp
= __cpu_to_be32(time(0) - DECADE
);
1480 ddf
->anchor
.openflag
= 0xFF;
1481 ddf
->anchor
.foreignflag
= 0;
1482 ddf
->anchor
.enforcegroups
= 0; /* Is this best?? */
1483 ddf
->anchor
.pad0
= 0xff;
1484 memset(ddf
->anchor
.pad1
, 0xff, 12);
1485 memset(ddf
->anchor
.header_ext
, 0xff, 32);
1486 ddf
->anchor
.primary_lba
= ~(__u64
)0;
1487 ddf
->anchor
.secondary_lba
= ~(__u64
)0;
1488 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
1489 memset(ddf
->anchor
.pad2
, 0xff, 3);
1490 ddf
->anchor
.workspace_len
= __cpu_to_be32(32768); /* Must be reserved */
1491 ddf
->anchor
.workspace_lba
= ~(__u64
)0; /* Put this at bottom
1492 of 32M reserved.. */
1493 max_phys_disks
= 1023; /* Should be enough */
1494 ddf
->anchor
.max_pd_entries
= __cpu_to_be16(max_phys_disks
);
1495 max_virt_disks
= 255;
1496 ddf
->anchor
.max_vd_entries
= __cpu_to_be16(max_virt_disks
); /* ?? */
1497 ddf
->anchor
.max_partitions
= __cpu_to_be16(64); /* ?? */
1500 ddf
->conf_rec_len
= 1 + ROUND_UP(ddf
->mppe
* (4+8), 512)/512;
1501 ddf
->anchor
.config_record_len
= __cpu_to_be16(ddf
->conf_rec_len
);
1502 ddf
->anchor
.max_primary_element_entries
= __cpu_to_be16(ddf
->mppe
);
1503 memset(ddf
->anchor
.pad3
, 0xff, 54);
1504 /* controller sections is one sector long immediately
1505 * after the ddf header */
1507 ddf
->anchor
.controller_section_offset
= __cpu_to_be32(sector
);
1508 ddf
->anchor
.controller_section_length
= __cpu_to_be32(1);
1511 /* phys is 8 sectors after that */
1512 pdsize
= ROUND_UP(sizeof(struct phys_disk
) +
1513 sizeof(struct phys_disk_entry
)*max_phys_disks
,
1515 switch(pdsize
/512) {
1516 case 2: case 8: case 32: case 128: case 512: break;
1519 ddf
->anchor
.phys_section_offset
= __cpu_to_be32(sector
);
1520 ddf
->anchor
.phys_section_length
=
1521 __cpu_to_be32(pdsize
/512); /* max_primary_element_entries/8 */
1522 sector
+= pdsize
/512;
1524 /* virt is another 32 sectors */
1525 vdsize
= ROUND_UP(sizeof(struct virtual_disk
) +
1526 sizeof(struct virtual_entry
) * max_virt_disks
,
1528 switch(vdsize
/512) {
1529 case 2: case 8: case 32: case 128: case 512: break;
1532 ddf
->anchor
.virt_section_offset
= __cpu_to_be32(sector
);
1533 ddf
->anchor
.virt_section_length
=
1534 __cpu_to_be32(vdsize
/512); /* max_vd_entries/8 */
1535 sector
+= vdsize
/512;
1537 clen
= ddf
->conf_rec_len
* (ddf
->max_part
+1);
1538 ddf
->anchor
.config_section_offset
= __cpu_to_be32(sector
);
1539 ddf
->anchor
.config_section_length
= __cpu_to_be32(clen
);
1542 ddf
->anchor
.data_section_offset
= __cpu_to_be32(sector
);
1543 ddf
->anchor
.data_section_length
= __cpu_to_be32(1);
1546 ddf
->anchor
.bbm_section_length
= __cpu_to_be32(0);
1547 ddf
->anchor
.bbm_section_offset
= __cpu_to_be32(0xFFFFFFFF);
1548 ddf
->anchor
.diag_space_length
= __cpu_to_be32(0);
1549 ddf
->anchor
.diag_space_offset
= __cpu_to_be32(0xFFFFFFFF);
1550 ddf
->anchor
.vendor_length
= __cpu_to_be32(0);
1551 ddf
->anchor
.vendor_offset
= __cpu_to_be32(0xFFFFFFFF);
1553 memset(ddf
->anchor
.pad4
, 0xff, 256);
1555 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
1556 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
1558 ddf
->primary
.openflag
= 1; /* I guess.. */
1559 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
1561 ddf
->secondary
.openflag
= 1; /* I guess.. */
1562 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
1564 ddf
->active
= &ddf
->primary
;
1566 ddf
->controller
.magic
= DDF_CONTROLLER_MAGIC
;
1568 /* 24 more bytes of fiction required.
1569 * first 8 are a 'vendor-id' - "Linux-MD"
1570 * Remaining 16 are serial number.... maybe a hostname would do?
1572 memcpy(ddf
->controller
.guid
, T10
, sizeof(T10
));
1573 gethostname(hostname
, sizeof(hostname
));
1574 hostname
[sizeof(hostname
) - 1] = 0;
1575 hostlen
= strlen(hostname
);
1576 memcpy(ddf
->controller
.guid
+ 24 - hostlen
, hostname
, hostlen
);
1577 for (i
= strlen(T10
) ; i
+hostlen
< 24; i
++)
1578 ddf
->controller
.guid
[i
] = ' ';
1580 ddf
->controller
.type
.vendor_id
= __cpu_to_be16(0xDEAD);
1581 ddf
->controller
.type
.device_id
= __cpu_to_be16(0xBEEF);
1582 ddf
->controller
.type
.sub_vendor_id
= 0;
1583 ddf
->controller
.type
.sub_device_id
= 0;
1584 memcpy(ddf
->controller
.product_id
, "What Is My PID??", 16);
1585 memset(ddf
->controller
.pad
, 0xff, 8);
1586 memset(ddf
->controller
.vendor_data
, 0xff, 448);
1588 posix_memalign((void**)&pd
, 512, pdsize
);
1590 ddf
->pdsize
= pdsize
;
1592 memset(pd
, 0xff, pdsize
);
1593 memset(pd
, 0, sizeof(*pd
));
1594 pd
->magic
= DDF_PHYS_DATA_MAGIC
;
1595 pd
->used_pdes
= __cpu_to_be16(0);
1596 pd
->max_pdes
= __cpu_to_be16(max_phys_disks
);
1597 memset(pd
->pad
, 0xff, 52);
1599 posix_memalign((void**)&vd
, 512, vdsize
);
1601 ddf
->vdsize
= vdsize
;
1602 memset(vd
, 0, vdsize
);
1603 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
1604 vd
->populated_vdes
= __cpu_to_be16(0);
1605 vd
->max_vdes
= __cpu_to_be16(max_virt_disks
);
1606 memset(vd
->pad
, 0xff, 52);
1608 for (i
=0; i
<max_virt_disks
; i
++)
1609 memset(&vd
->entries
[i
], 0xff, sizeof(struct virtual_entry
));
1612 ddf
->updates_pending
= 1;
1616 static int all_ff(char *guid
)
1619 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
1620 if (guid
[i
] != (char)0xff)
1624 static int chunk_to_shift(int chunksize
)
1626 return ffs(chunksize
/512)-1;
1629 static int level_to_prl(int level
)
1632 case LEVEL_LINEAR
: return DDF_CONCAT
;
1633 case 0: return DDF_RAID0
;
1634 case 1: return DDF_RAID1
;
1635 case 4: return DDF_RAID4
;
1636 case 5: return DDF_RAID5
;
1637 case 6: return DDF_RAID6
;
1641 static int layout_to_rlq(int level
, int layout
, int raiddisks
)
1645 return DDF_RAID0_SIMPLE
;
1648 case 2: return DDF_RAID1_SIMPLE
;
1649 case 3: return DDF_RAID1_MULTI
;
1654 case 0: return DDF_RAID4_N
;
1660 case ALGORITHM_LEFT_ASYMMETRIC
:
1661 return DDF_RAID5_N_RESTART
;
1662 case ALGORITHM_RIGHT_ASYMMETRIC
:
1664 return DDF_RAID5_0_RESTART
;
1666 return DDF_RAID6_0_RESTART
;
1667 case ALGORITHM_LEFT_SYMMETRIC
:
1668 return DDF_RAID5_N_CONTINUE
;
1669 case ALGORITHM_RIGHT_SYMMETRIC
:
1670 return -1; /* not mentioned in standard */
1676 static int rlq_to_layout(int rlq
, int prl
, int raiddisks
)
1680 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1682 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1690 return -1; /* FIXME this isn't checked */
1694 case DDF_RAID5_N_RESTART
:
1695 return ALGORITHM_LEFT_ASYMMETRIC
;
1696 case DDF_RAID5_0_RESTART
:
1697 return ALGORITHM_RIGHT_ASYMMETRIC
;
1698 case DDF_RAID5_N_CONTINUE
:
1699 return ALGORITHM_LEFT_SYMMETRIC
;
1705 case DDF_RAID5_N_RESTART
:
1706 return ALGORITHM_LEFT_ASYMMETRIC
;
1707 case DDF_RAID6_0_RESTART
:
1708 return ALGORITHM_RIGHT_ASYMMETRIC
;
1709 case DDF_RAID5_N_CONTINUE
:
1710 return ALGORITHM_LEFT_SYMMETRIC
;
1719 unsigned long long start
, size
;
1721 static int cmp_extent(const void *av
, const void *bv
)
1723 const struct extent
*a
= av
;
1724 const struct extent
*b
= bv
;
1725 if (a
->start
< b
->start
)
1727 if (a
->start
> b
->start
)
1732 static struct extent
*get_extents(struct ddf_super
*ddf
, struct dl
*dl
)
1734 /* find a list of used extents on the give physical device
1735 * (dnum) of the given ddf.
1736 * Return a malloced array of 'struct extent'
1738 FIXME ignore DDF_Legacy devices?
1745 rv
= malloc(sizeof(struct extent
) * (ddf
->max_part
+ 2));
1749 for (i
= 0; i
< ddf
->max_part
; i
++) {
1750 struct vcl
*v
= dl
->vlist
[i
];
1753 for (j
=0; j
< v
->conf
.prim_elmnt_count
; j
++)
1754 if (v
->conf
.phys_refnum
[j
] == dl
->disk
.refnum
) {
1755 /* This device plays role 'j' in 'v'. */
1756 rv
[n
].start
= __be64_to_cpu(v
->lba_offset
[j
]);
1757 rv
[n
].size
= __be64_to_cpu(v
->conf
.blocks
);
1762 qsort(rv
, n
, sizeof(*rv
), cmp_extent
);
1764 rv
[n
].start
= __be64_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].config_size
);
1769 static int init_super_ddf_bvd(struct supertype
*st
,
1770 mdu_array_info_t
*info
,
1771 unsigned long long size
,
1772 char *name
, char *homehost
,
1775 /* We are creating a BVD inside a pre-existing container.
1776 * so st->sb is already set.
1777 * We need to create a new vd_config and a new virtual_entry
1779 struct ddf_super
*ddf
= st
->sb
;
1781 struct virtual_entry
*ve
;
1783 struct vd_config
*vc
;
1785 if (__be16_to_cpu(ddf
->virt
->populated_vdes
)
1786 >= __be16_to_cpu(ddf
->virt
->max_vdes
)) {
1787 fprintf(stderr
, Name
": This ddf already has the "
1788 "maximum of %d virtual devices\n",
1789 __be16_to_cpu(ddf
->virt
->max_vdes
));
1793 for (venum
= 0; venum
< __be16_to_cpu(ddf
->virt
->max_vdes
); venum
++)
1794 if (all_ff(ddf
->virt
->entries
[venum
].guid
))
1796 if (venum
== __be16_to_cpu(ddf
->virt
->max_vdes
)) {
1797 fprintf(stderr
, Name
": Cannot find spare slot for "
1798 "virtual disk - DDF is corrupt\n");
1801 ve
= &ddf
->virt
->entries
[venum
];
1803 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1804 * timestamp, random number
1806 make_header_guid(ve
->guid
);
1807 ve
->unit
= __cpu_to_be16(info
->md_minor
);
1809 ve
->guid_crc
= crc32(0, (unsigned char*)ddf
->anchor
.guid
, DDF_GUID_LEN
);
1811 ve
->state
= DDF_state_degraded
; /* Will be modified as devices are added */
1812 if (info
->state
& 1) /* clean */
1813 ve
->init_state
= DDF_init_full
;
1815 ve
->init_state
= DDF_init_not
;
1817 memset(ve
->pad1
, 0xff, 14);
1818 memset(ve
->name
, ' ', 16);
1820 strncpy(ve
->name
, name
, 16);
1821 ddf
->virt
->populated_vdes
=
1822 __cpu_to_be16(__be16_to_cpu(ddf
->virt
->populated_vdes
)+1);
1824 /* Now create a new vd_config */
1825 posix_memalign((void**)&vcl
, 512,
1826 (offsetof(struct vcl
, conf
) + ddf
->conf_rec_len
* 512));
1827 vcl
->lba_offset
= (__u64
*) &vcl
->conf
.phys_refnum
[ddf
->mppe
];
1829 sprintf(st
->subarray
, "%d", venum
);
1830 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
1834 vc
->magic
= DDF_VD_CONF_MAGIC
;
1835 memcpy(vc
->guid
, ve
->guid
, DDF_GUID_LEN
);
1836 vc
->timestamp
= __cpu_to_be32(time(0)-DECADE
);
1837 vc
->seqnum
= __cpu_to_be32(1);
1838 memset(vc
->pad0
, 0xff, 24);
1839 vc
->prim_elmnt_count
= __cpu_to_be16(info
->raid_disks
);
1840 vc
->chunk_shift
= chunk_to_shift(info
->chunk_size
);
1841 vc
->prl
= level_to_prl(info
->level
);
1842 vc
->rlq
= layout_to_rlq(info
->level
, info
->layout
, info
->raid_disks
);
1843 vc
->sec_elmnt_count
= 1;
1844 vc
->sec_elmnt_seq
= 0;
1846 vc
->blocks
= __cpu_to_be64(info
->size
* 2);
1847 vc
->array_blocks
= __cpu_to_be64(
1848 calc_array_size(info
->level
, info
->raid_disks
, info
->layout
,
1849 info
->chunk_size
, info
->size
*2));
1850 memset(vc
->pad1
, 0xff, 8);
1851 vc
->spare_refs
[0] = 0xffffffff;
1852 vc
->spare_refs
[1] = 0xffffffff;
1853 vc
->spare_refs
[2] = 0xffffffff;
1854 vc
->spare_refs
[3] = 0xffffffff;
1855 vc
->spare_refs
[4] = 0xffffffff;
1856 vc
->spare_refs
[5] = 0xffffffff;
1857 vc
->spare_refs
[6] = 0xffffffff;
1858 vc
->spare_refs
[7] = 0xffffffff;
1859 memset(vc
->cache_pol
, 0, 8);
1861 memset(vc
->pad2
, 0xff, 3);
1862 memset(vc
->pad3
, 0xff, 52);
1863 memset(vc
->pad4
, 0xff, 192);
1864 memset(vc
->v0
, 0xff, 32);
1865 memset(vc
->v1
, 0xff, 32);
1866 memset(vc
->v2
, 0xff, 16);
1867 memset(vc
->v3
, 0xff, 16);
1868 memset(vc
->vendor
, 0xff, 32);
1870 memset(vc
->phys_refnum
, 0xff, 4*ddf
->mppe
);
1871 memset(vc
->phys_refnum
+(ddf
->mppe
* 4), 0x00, 8*ddf
->mppe
);
1873 vcl
->next
= ddf
->conflist
;
1874 ddf
->conflist
= vcl
;
1875 ddf
->currentconf
= vcl
;
1876 ddf
->updates_pending
= 1;
1880 static void add_to_super_ddf_bvd(struct supertype
*st
,
1881 mdu_disk_info_t
*dk
, int fd
, char *devname
)
1883 /* fd and devname identify a device with-in the ddf container (st).
1884 * dk identifies a location in the new BVD.
1885 * We need to find suitable free space in that device and update
1886 * the phys_refnum and lba_offset for the newly created vd_config.
1887 * We might also want to update the type in the phys_disk
1891 struct ddf_super
*ddf
= st
->sb
;
1892 struct vd_config
*vc
;
1896 unsigned long long blocks
, pos
, esize
;
1899 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
1900 if (dl
->major
== dk
->major
&&
1901 dl
->minor
== dk
->minor
)
1903 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1906 vc
= &ddf
->currentconf
->conf
;
1907 lba_offset
= ddf
->currentconf
->lba_offset
;
1909 ex
= get_extents(ddf
, dl
);
1914 blocks
= __be64_to_cpu(vc
->blocks
);
1915 if (ddf
->currentconf
->block_sizes
)
1916 blocks
= ddf
->currentconf
->block_sizes
[dk
->raid_disk
];
1919 esize
= ex
[i
].start
- pos
;
1920 if (esize
>= blocks
)
1922 pos
= ex
[i
].start
+ ex
[i
].size
;
1924 } while (ex
[i
-1].size
);
1930 ddf
->currentdev
= dk
->raid_disk
;
1931 vc
->phys_refnum
[dk
->raid_disk
] = dl
->disk
.refnum
;
1932 lba_offset
[dk
->raid_disk
] = __cpu_to_be64(pos
);
1934 for (i
=0; i
< ddf
->max_part
; i
++)
1935 if (dl
->vlist
[i
] == NULL
)
1937 if (i
== ddf
->max_part
)
1939 dl
->vlist
[i
] = ddf
->currentconf
;
1942 dl
->devname
= devname
;
1944 /* Check how many working raid_disks, and if we can mark
1945 * array as optimal yet
1949 for (i
=0; i
< __be16_to_cpu(vc
->prim_elmnt_count
); i
++)
1950 if (vc
->phys_refnum
[i
] != 0xffffffff)
1953 /* Find which virtual_entry */
1954 i
= ddf
->currentconf
->vcnum
;
1955 if (working
== __be16_to_cpu(vc
->prim_elmnt_count
))
1956 ddf
->virt
->entries
[i
].state
=
1957 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
1958 | DDF_state_optimal
;
1960 if (vc
->prl
== DDF_RAID6
&&
1961 working
+1 == __be16_to_cpu(vc
->prim_elmnt_count
))
1962 ddf
->virt
->entries
[i
].state
=
1963 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
1964 | DDF_state_part_optimal
;
1966 ddf
->phys
->entries
[dl
->pdnum
].type
&= ~__cpu_to_be16(DDF_Global_Spare
);
1967 ddf
->phys
->entries
[dl
->pdnum
].type
|= __cpu_to_be16(DDF_Active_in_VD
);
1968 ddf
->updates_pending
= 1;
1971 /* add a device to a container, either while creating it or while
1972 * expanding a pre-existing container
1974 static void add_to_super_ddf(struct supertype
*st
,
1975 mdu_disk_info_t
*dk
, int fd
, char *devname
)
1977 struct ddf_super
*ddf
= st
->sb
;
1981 unsigned long long size
;
1982 struct phys_disk_entry
*pde
;
1986 if (ddf
->currentconf
) {
1987 add_to_super_ddf_bvd(st
, dk
, fd
, devname
);
1991 /* This is device numbered dk->number. We need to create
1992 * a phys_disk entry and a more detailed disk_data entry.
1995 posix_memalign((void**)&dd
, 512,
1996 sizeof(*dd
) + sizeof(dd
->vlist
[0]) * ddf
->max_part
);
1997 dd
->major
= major(stb
.st_rdev
);
1998 dd
->minor
= minor(stb
.st_rdev
);
1999 dd
->devname
= devname
;
2003 dd
->disk
.magic
= DDF_PHYS_DATA_MAGIC
;
2005 tm
= localtime(&now
);
2006 sprintf(dd
->disk
.guid
, "%8s%04d%02d%02d",
2007 T10
, tm
->tm_year
+1900, tm
->tm_mon
+1, tm
->tm_mday
);
2008 *(__u32
*)(dd
->disk
.guid
+ 16) = random();
2009 *(__u32
*)(dd
->disk
.guid
+ 20) = random();
2012 /* Cannot be bothered finding a CRC of some irrelevant details*/
2013 dd
->disk
.refnum
= random();
2014 for (i
= __be16_to_cpu(ddf
->active
->max_pd_entries
) - 1;
2016 if (ddf
->phys
->entries
[i
].refnum
== dd
->disk
.refnum
)
2020 dd
->disk
.forced_ref
= 1;
2021 dd
->disk
.forced_guid
= 1;
2022 memset(dd
->disk
.vendor
, ' ', 32);
2023 memcpy(dd
->disk
.vendor
, "Linux", 5);
2024 memset(dd
->disk
.pad
, 0xff, 442);
2025 for (i
= 0; i
< ddf
->max_part
; i
++)
2026 dd
->vlist
[i
] = NULL
;
2028 n
= __be16_to_cpu(ddf
->phys
->used_pdes
);
2029 pde
= &ddf
->phys
->entries
[n
];
2032 if (st
->update_tail
) {
2033 int len
= (sizeof(struct phys_disk
) +
2034 sizeof(struct phys_disk_entry
));
2035 struct phys_disk
*pd
;
2038 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2039 pd
->used_pdes
= __cpu_to_be16(n
);
2040 pde
= &pd
->entries
[0];
2044 ddf
->phys
->used_pdes
= __cpu_to_be16(n
);
2047 memcpy(pde
->guid
, dd
->disk
.guid
, DDF_GUID_LEN
);
2048 pde
->refnum
= dd
->disk
.refnum
;
2049 pde
->type
= __cpu_to_be16(DDF_Forced_PD_GUID
| DDF_Global_Spare
);
2050 pde
->state
= __cpu_to_be16(DDF_Online
);
2051 get_dev_size(fd
, NULL
, &size
);
2052 /* We are required to reserve 32Meg, and record the size in sectors */
2053 pde
->config_size
= __cpu_to_be64( (size
- 32*1024*1024) / 512);
2054 sprintf(pde
->path
, "%17.17s","Information: nil") ;
2055 memset(pde
->pad
, 0xff, 6);
2057 dd
->size
= size
>> 9;
2058 if (st
->update_tail
) {
2059 dd
->next
= ddf
->add_list
;
2062 dd
->next
= ddf
->dlist
;
2064 ddf
->updates_pending
= 1;
2069 * This is the write_init_super method for a ddf container. It is
2070 * called when creating a container or adding another device to a
2076 static unsigned char null_conf
[4096+512];
2078 static int __write_init_super_ddf(struct supertype
*st
, int do_close
)
2081 struct ddf_super
*ddf
= st
->sb
;
2087 unsigned long long size
, sector
;
2089 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
2095 /* We need to fill in the primary, (secondary) and workspace
2096 * lba's in the headers, set their checksums,
2097 * Also checksum phys, virt....
2099 * Then write everything out, finally the anchor is written.
2101 get_dev_size(fd
, NULL
, &size
);
2103 ddf
->anchor
.workspace_lba
= __cpu_to_be64(size
- 32*1024*2);
2104 ddf
->anchor
.primary_lba
= __cpu_to_be64(size
- 16*1024*2);
2105 ddf
->anchor
.seq
= __cpu_to_be32(1);
2106 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2107 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2109 ddf
->anchor
.openflag
= 0xFF; /* 'open' means nothing */
2110 ddf
->anchor
.seq
= 0xFFFFFFFF; /* no sequencing in anchor */
2111 ddf
->anchor
.crc
= calc_crc(&ddf
->anchor
, 512);
2113 ddf
->primary
.openflag
= 0;
2114 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
2116 ddf
->secondary
.openflag
= 0;
2117 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
2119 ddf
->primary
.crc
= calc_crc(&ddf
->primary
, 512);
2120 ddf
->secondary
.crc
= calc_crc(&ddf
->secondary
, 512);
2122 sector
= size
- 16*1024*2;
2123 lseek64(fd
, sector
<<9, 0);
2124 write(fd
, &ddf
->primary
, 512);
2126 ddf
->controller
.crc
= calc_crc(&ddf
->controller
, 512);
2127 write(fd
, &ddf
->controller
, 512);
2129 ddf
->phys
->crc
= calc_crc(ddf
->phys
, ddf
->pdsize
);
2131 write(fd
, ddf
->phys
, ddf
->pdsize
);
2133 ddf
->virt
->crc
= calc_crc(ddf
->virt
, ddf
->vdsize
);
2134 write(fd
, ddf
->virt
, ddf
->vdsize
);
2136 /* Now write lots of config records. */
2137 n_config
= ddf
->max_part
;
2138 conf_size
= ddf
->conf_rec_len
* 512;
2139 for (i
= 0 ; i
<= n_config
; i
++) {
2140 struct vcl
*c
= d
->vlist
[i
];
2142 c
= (struct vcl
*)d
->spare
;
2145 c
->conf
.crc
= calc_crc(&c
->conf
, conf_size
);
2146 write(fd
, &c
->conf
, conf_size
);
2148 char *null_aligned
= (char*)((((unsigned long)null_conf
)+511)&~511UL);
2149 if (null_conf
[0] != 0xff)
2150 memset(null_conf
, 0xff, sizeof(null_conf
));
2151 int togo
= conf_size
;
2152 while (togo
> sizeof(null_conf
)-512) {
2153 write(fd
, null_aligned
, sizeof(null_conf
)-512);
2154 togo
-= sizeof(null_conf
)-512;
2156 write(fd
, null_aligned
, togo
);
2159 d
->disk
.crc
= calc_crc(&d
->disk
, 512);
2160 write(fd
, &d
->disk
, 512);
2162 /* Maybe do the same for secondary */
2164 lseek64(fd
, (size
-1)*512, SEEK_SET
);
2165 write(fd
, &ddf
->anchor
, 512);
2174 static int write_init_super_ddf(struct supertype
*st
)
2177 if (st
->update_tail
) {
2178 /* queue the virtual_disk and vd_config as metadata updates */
2179 struct virtual_disk
*vd
;
2180 struct vd_config
*vc
;
2181 struct ddf_super
*ddf
= st
->sb
;
2184 if (!ddf
->currentconf
) {
2185 int len
= (sizeof(struct phys_disk
) +
2186 sizeof(struct phys_disk_entry
));
2188 /* adding a disk to the container. */
2192 append_metadata_update(st
, ddf
->add_list
->mdupdate
, len
);
2193 ddf
->add_list
->mdupdate
= NULL
;
2197 /* Newly created VD */
2199 /* First the virtual disk. We have a slightly fake header */
2200 len
= sizeof(struct virtual_disk
) + sizeof(struct virtual_entry
);
2203 vd
->entries
[0] = ddf
->virt
->entries
[ddf
->currentconf
->vcnum
];
2204 vd
->populated_vdes
= __cpu_to_be16(ddf
->currentconf
->vcnum
);
2205 append_metadata_update(st
, vd
, len
);
2207 /* Then the vd_config */
2208 len
= ddf
->conf_rec_len
* 512;
2210 memcpy(vc
, &ddf
->currentconf
->conf
, len
);
2211 append_metadata_update(st
, vc
, len
);
2213 /* FIXME I need to close the fds! */
2216 return __write_init_super_ddf(st
, 1);
2221 static __u64
avail_size_ddf(struct supertype
*st
, __u64 devsize
)
2223 /* We must reserve the last 32Meg */
2224 if (devsize
<= 32*1024*2)
2226 return devsize
- 32*1024*2;
2231 validate_geometry_ddf_container(struct supertype
*st
,
2232 int level
, int layout
, int raiddisks
,
2233 int chunk
, unsigned long long size
,
2234 char *dev
, unsigned long long *freesize
,
2237 static int validate_geometry_ddf_bvd(struct supertype
*st
,
2238 int level
, int layout
, int raiddisks
,
2239 int chunk
, unsigned long long size
,
2240 char *dev
, unsigned long long *freesize
,
2243 static int validate_geometry_ddf(struct supertype
*st
,
2244 int level
, int layout
, int raiddisks
,
2245 int chunk
, unsigned long long size
,
2246 char *dev
, unsigned long long *freesize
,
2253 /* ddf potentially supports lots of things, but it depends on
2254 * what devices are offered (and maybe kernel version?)
2255 * If given unused devices, we will make a container.
2256 * If given devices in a container, we will make a BVD.
2257 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2260 if (level
== LEVEL_CONTAINER
) {
2261 /* Must be a fresh device to add to a container */
2262 return validate_geometry_ddf_container(st
, level
, layout
,
2264 size
, dev
, freesize
,
2269 /* A container has already been opened, so we are
2270 * creating in there. Maybe a BVD, maybe an SVD.
2271 * Should make a distinction one day.
2273 return validate_geometry_ddf_bvd(st
, level
, layout
, raiddisks
,
2274 chunk
, size
, dev
, freesize
,
2278 /* Initial sanity check. Exclude illegal levels. */
2280 for (i
=0; ddf_level_num
[i
].num1
!= MAXINT
; i
++)
2281 if (ddf_level_num
[i
].num2
== level
)
2283 if (ddf_level_num
[i
].num1
== MAXINT
)
2285 /* Should check layout? etc */
2289 /* This is the first device for the array.
2290 * If it is a container, we read it in and do automagic allocations,
2291 * no other devices should be given.
2292 * Otherwise it must be a member device of a container, and we
2293 * do manual allocation.
2294 * Later we should check for a BVD and make an SVD.
2296 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2298 sra
= sysfs_read(fd
, 0, GET_VERSION
);
2300 if (sra
&& sra
->array
.major_version
== -1 &&
2301 strcmp(sra
->text_version
, "ddf") == 0) {
2304 /* find space for 'n' devices. */
2305 /* remember the devices */
2306 /* Somehow return the fact that we have enough */
2311 Name
": ddf: Cannot create this array "
2316 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2318 fprintf(stderr
, Name
": ddf: Cannot open %s: %s\n",
2319 dev
, strerror(errno
));
2322 /* Well, it is in use by someone, maybe a 'ddf' container. */
2323 cfd
= open_container(fd
);
2327 fprintf(stderr
, Name
": ddf: Cannot use %s: %s\n",
2328 dev
, strerror(EBUSY
));
2331 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2333 if (sra
&& sra
->array
.major_version
== -1 &&
2334 strcmp(sra
->text_version
, "ddf") == 0) {
2335 /* This is a member of a ddf container. Load the container
2336 * and try to create a bvd
2338 struct ddf_super
*ddf
;
2339 if (load_super_ddf_all(st
, cfd
, (void **)&ddf
, NULL
, 1) == 0) {
2341 st
->container_dev
= fd2devnum(cfd
);
2343 return validate_geometry_ddf_bvd(st
, level
, layout
,
2344 raiddisks
, chunk
, size
,
2349 } else /* device may belong to a different container */
2356 validate_geometry_ddf_container(struct supertype
*st
,
2357 int level
, int layout
, int raiddisks
,
2358 int chunk
, unsigned long long size
,
2359 char *dev
, unsigned long long *freesize
,
2363 unsigned long long ldsize
;
2365 if (level
!= LEVEL_CONTAINER
)
2370 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2373 fprintf(stderr
, Name
": ddf: Cannot open %s: %s\n",
2374 dev
, strerror(errno
));
2377 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2383 *freesize
= avail_size_ddf(st
, ldsize
>> 9);
2388 static int validate_geometry_ddf_bvd(struct supertype
*st
,
2389 int level
, int layout
, int raiddisks
,
2390 int chunk
, unsigned long long size
,
2391 char *dev
, unsigned long long *freesize
,
2395 struct ddf_super
*ddf
= st
->sb
;
2397 unsigned long long pos
= 0;
2398 unsigned long long maxsize
;
2401 /* ddf/bvd supports lots of things, but not containers */
2402 if (level
== LEVEL_CONTAINER
)
2404 /* We must have the container info already read in. */
2409 /* General test: make sure there is space for
2410 * 'raiddisks' device extents of size 'size'.
2412 unsigned long long minsize
= size
;
2416 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2422 e
= get_extents(ddf
, dl
);
2425 unsigned long long esize
;
2426 esize
= e
[i
].start
- pos
;
2427 if (esize
>= minsize
)
2429 pos
= e
[i
].start
+ e
[i
].size
;
2431 } while (e
[i
-1].size
);
2436 if (dcnt
< raiddisks
) {
2439 Name
": ddf: Not enough devices with "
2440 "space for this array (%d < %d)\n",
2446 /* This device must be a member of the set */
2447 if (stat(dev
, &stb
) < 0)
2449 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2451 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
2452 if (dl
->major
== major(stb
.st_rdev
) &&
2453 dl
->minor
== minor(stb
.st_rdev
))
2458 fprintf(stderr
, Name
": ddf: %s is not in the "
2463 e
= get_extents(ddf
, dl
);
2467 unsigned long long esize
;
2468 esize
= e
[i
].start
- pos
;
2469 if (esize
>= maxsize
)
2471 pos
= e
[i
].start
+ e
[i
].size
;
2473 } while (e
[i
-1].size
);
2474 *freesize
= maxsize
;
2480 static int load_super_ddf_all(struct supertype
*st
, int fd
,
2481 void **sbp
, char *devname
, int keep_fd
)
2484 struct ddf_super
*super
;
2485 struct mdinfo
*sd
, *best
= NULL
;
2491 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
2494 if (sra
->array
.major_version
!= -1 ||
2495 sra
->array
.minor_version
!= -2 ||
2496 strcmp(sra
->text_version
, "ddf") != 0)
2499 if (posix_memalign((void**)&super
, 512, sizeof(*super
)) != 0)
2501 memset(super
, 0, sizeof(*super
));
2503 /* first, try each device, and choose the best ddf */
2504 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2506 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2507 dfd
= dev_open(nm
, O_RDONLY
);
2510 rv
= load_ddf_headers(dfd
, super
, NULL
);
2513 seq
= __be32_to_cpu(super
->active
->seq
);
2514 if (super
->active
->openflag
)
2516 if (!best
|| seq
> bestseq
) {
2524 /* OK, load this ddf */
2525 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
2526 dfd
= dev_open(nm
, O_RDONLY
);
2529 load_ddf_headers(dfd
, super
, NULL
);
2530 load_ddf_global(dfd
, super
, NULL
);
2532 /* Now we need the device-local bits */
2533 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2534 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2535 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2538 load_ddf_headers(dfd
, super
, NULL
);
2539 seq
= load_ddf_local(dfd
, super
, NULL
, keep_fd
);
2540 if (!keep_fd
) close(dfd
);
2542 if (st
->subarray
[0]) {
2545 for (v
= super
->conflist
; v
; v
= v
->next
)
2546 if (v
->vcnum
== atoi(st
->subarray
))
2547 super
->currentconf
= v
;
2548 if (!super
->currentconf
)
2552 if (st
->ss
== NULL
) {
2553 st
->ss
= &super_ddf
;
2554 st
->minor_version
= 0;
2556 st
->container_dev
= fd2devnum(fd
);
2562 static struct mdinfo
*container_content_ddf(struct supertype
*st
)
2564 /* Given a container loaded by load_super_ddf_all,
2565 * extract information about all the arrays into
2568 * For each vcl in conflist: create an mdinfo, fill it in,
2569 * then look for matching devices (phys_refnum) in dlist
2570 * and create appropriate device mdinfo.
2572 struct ddf_super
*ddf
= st
->sb
;
2573 struct mdinfo
*rest
= NULL
;
2576 for (vc
= ddf
->conflist
; vc
; vc
=vc
->next
)
2579 struct mdinfo
*this;
2580 this = malloc(sizeof(*this));
2581 memset(this, 0, sizeof(*this));
2585 this->array
.level
= map_num1(ddf_level_num
, vc
->conf
.prl
);
2586 this->array
.raid_disks
=
2587 __be16_to_cpu(vc
->conf
.prim_elmnt_count
);
2588 this->array
.layout
= rlq_to_layout(vc
->conf
.rlq
, vc
->conf
.prl
,
2589 this->array
.raid_disks
);
2590 this->array
.md_minor
= -1;
2591 this->array
.ctime
= DECADE
+
2592 __be32_to_cpu(*(__u32
*)(vc
->conf
.guid
+16));
2593 this->array
.utime
= DECADE
+
2594 __be32_to_cpu(vc
->conf
.timestamp
);
2595 this->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
2598 if ((ddf
->virt
->entries
[i
].state
& DDF_state_inconsistent
) ||
2599 (ddf
->virt
->entries
[i
].init_state
& DDF_initstate_mask
) !=
2601 this->array
.state
= 0;
2602 this->resync_start
= 0;
2604 this->array
.state
= 1;
2605 this->resync_start
= ~0ULL;
2607 memcpy(this->name
, ddf
->virt
->entries
[i
].name
, 32);
2610 memset(this->uuid
, 0, sizeof(this->uuid
));
2611 this->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
2612 this->array
.size
= this->component_size
/ 2;
2613 this->container_member
= i
;
2615 sprintf(this->text_version
, "/%s/%d",
2616 devnum2devname(st
->container_dev
),
2617 this->container_member
);
2619 for (i
=0 ; i
< ddf
->mppe
; i
++) {
2623 if (vc
->conf
.phys_refnum
[i
] == 0xFFFFFFFF)
2626 this->array
.working_disks
++;
2628 for (d
= ddf
->dlist
; d
; d
=d
->next
)
2629 if (d
->disk
.refnum
== vc
->conf
.phys_refnum
[i
])
2634 dev
= malloc(sizeof(*dev
));
2635 memset(dev
, 0, sizeof(*dev
));
2636 dev
->next
= this->devs
;
2639 dev
->disk
.number
= __be32_to_cpu(d
->disk
.refnum
);
2640 dev
->disk
.major
= d
->major
;
2641 dev
->disk
.minor
= d
->minor
;
2642 dev
->disk
.raid_disk
= i
;
2643 dev
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
2645 dev
->events
= __be32_to_cpu(ddf
->primary
.seq
);
2646 dev
->data_offset
= __be64_to_cpu(vc
->lba_offset
[i
]);
2647 dev
->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
2649 strcpy(dev
->name
, d
->devname
);
2655 static int store_zero_ddf(struct supertype
*st
, int fd
)
2657 unsigned long long dsize
;
2660 if (!get_dev_size(fd
, NULL
, &dsize
))
2663 posix_memalign(&buf
, 512, 512);
2664 memset(buf
, 0, 512);
2666 lseek64(fd
, dsize
-512, 0);
2667 write(fd
, buf
, 512);
2672 static int compare_super_ddf(struct supertype
*st
, struct supertype
*tst
)
2676 * 0 same, or first was empty, and second was copied
2677 * 1 second had wrong number
2679 * 3 wrong other info
2681 struct ddf_super
*first
= st
->sb
;
2682 struct ddf_super
*second
= tst
->sb
;
2690 if (memcmp(first
->anchor
.guid
, second
->anchor
.guid
, DDF_GUID_LEN
) != 0)
2693 /* FIXME should I look at anything else? */
2698 * A new array 'a' has been started which claims to be instance 'inst'
2699 * within container 'c'.
2700 * We need to confirm that the array matches the metadata in 'c' so
2701 * that we don't corrupt any metadata.
2703 static int ddf_open_new(struct supertype
*c
, struct active_array
*a
, char *inst
)
2705 dprintf("ddf: open_new %s\n", inst
);
2706 a
->info
.container_member
= atoi(inst
);
2711 * The array 'a' is to be marked clean in the metadata.
2712 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2713 * clean up to the point (in sectors). If that cannot be recorded in the
2714 * metadata, then leave it as dirty.
2716 * For DDF, we need to clear the DDF_state_inconsistent bit in the
2717 * !global! virtual_disk.virtual_entry structure.
2719 static int ddf_set_array_state(struct active_array
*a
, int consistent
)
2721 struct ddf_super
*ddf
= a
->container
->sb
;
2722 int inst
= a
->info
.container_member
;
2723 int old
= ddf
->virt
->entries
[inst
].state
;
2724 if (consistent
== 2) {
2725 /* Should check if a recovery should be started FIXME */
2727 if (a
->resync_start
!= ~0ULL)
2731 ddf
->virt
->entries
[inst
].state
&= ~DDF_state_inconsistent
;
2733 ddf
->virt
->entries
[inst
].state
|= DDF_state_inconsistent
;
2734 if (old
!= ddf
->virt
->entries
[inst
].state
)
2735 ddf
->updates_pending
= 1;
2737 old
= ddf
->virt
->entries
[inst
].init_state
;
2738 ddf
->virt
->entries
[inst
].init_state
&= ~DDF_initstate_mask
;
2739 if (a
->resync_start
== ~0ULL)
2740 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_full
;
2741 else if (a
->resync_start
== 0)
2742 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_not
;
2744 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_quick
;
2745 if (old
!= ddf
->virt
->entries
[inst
].init_state
)
2746 ddf
->updates_pending
= 1;
2748 dprintf("ddf mark %d %s %llu\n", inst
, consistent
?"clean":"dirty",
2754 * The state of each disk is stored in the global phys_disk structure
2755 * in phys_disk.entries[n].state.
2756 * This makes various combinations awkward.
2757 * - When a device fails in any array, it must be failed in all arrays
2758 * that include a part of this device.
2759 * - When a component is rebuilding, we cannot include it officially in the
2760 * array unless this is the only array that uses the device.
2762 * So: when transitioning:
2763 * Online -> failed, just set failed flag. monitor will propagate
2764 * spare -> online, the device might need to be added to the array.
2765 * spare -> failed, just set failed. Don't worry if in array or not.
2767 static void ddf_set_disk(struct active_array
*a
, int n
, int state
)
2769 struct ddf_super
*ddf
= a
->container
->sb
;
2770 int inst
= a
->info
.container_member
;
2771 struct vd_config
*vc
= find_vdcr(ddf
, inst
);
2772 int pd
= find_phys(ddf
, vc
->phys_refnum
[n
]);
2776 dprintf("ddf: cannot find instance %d!!\n", inst
);
2780 /* disk doesn't currently exist. If it is now in_sync,
2782 if ((state
& DS_INSYNC
) && ! (state
& DS_FAULTY
)) {
2783 /* Find dev 'n' in a->info->devs, determine the
2784 * ddf refnum, and set vc->phys_refnum and update
2790 int old
= ddf
->phys
->entries
[pd
].state
;
2791 if (state
& DS_FAULTY
)
2792 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Failed
);
2793 if (state
& DS_INSYNC
) {
2794 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Online
);
2795 ddf
->phys
->entries
[pd
].state
&= __cpu_to_be16(~DDF_Rebuilding
);
2797 if (old
!= ddf
->phys
->entries
[pd
].state
)
2798 ddf
->updates_pending
= 1;
2801 dprintf("ddf: set_disk %d to %x\n", n
, state
);
2803 /* Now we need to check the state of the array and update
2804 * virtual_disk.entries[n].state.
2805 * It needs to be one of "optimal", "degraded", "failed".
2806 * I don't understand 'deleted' or 'missing'.
2809 for (i
=0; i
< a
->info
.array
.raid_disks
; i
++) {
2810 pd
= find_phys(ddf
, vc
->phys_refnum
[i
]);
2813 st
= __be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
2814 if ((st
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
2818 state
= DDF_state_degraded
;
2819 if (working
== a
->info
.array
.raid_disks
)
2820 state
= DDF_state_optimal
;
2821 else switch(vc
->prl
) {
2825 state
= DDF_state_failed
;
2829 state
= DDF_state_failed
;
2833 if (working
< a
->info
.array
.raid_disks
-1)
2834 state
= DDF_state_failed
;
2837 if (working
< a
->info
.array
.raid_disks
-2)
2838 state
= DDF_state_failed
;
2839 else if (working
== a
->info
.array
.raid_disks
-1)
2840 state
= DDF_state_part_optimal
;
2844 if (ddf
->virt
->entries
[inst
].state
!=
2845 ((ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
2848 ddf
->virt
->entries
[inst
].state
=
2849 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
2851 ddf
->updates_pending
= 1;
2856 static void ddf_sync_metadata(struct supertype
*st
)
2860 * Write all data to all devices.
2861 * Later, we might be able to track whether only local changes
2862 * have been made, or whether any global data has been changed,
2863 * but ddf is sufficiently weird that it probably always
2864 * changes global data ....
2866 struct ddf_super
*ddf
= st
->sb
;
2867 if (!ddf
->updates_pending
)
2869 ddf
->updates_pending
= 0;
2870 __write_init_super_ddf(st
, 0);
2871 dprintf("ddf: sync_metadata\n");
2874 static void ddf_process_update(struct supertype
*st
,
2875 struct metadata_update
*update
)
2877 /* Apply this update to the metadata.
2878 * The first 4 bytes are a DDF_*_MAGIC which guides
2880 * Possible update are:
2881 * DDF_PHYS_RECORDS_MAGIC
2882 * Add a new physical device. Changes to this record
2883 * only happen implicitly.
2884 * used_pdes is the device number.
2885 * DDF_VIRT_RECORDS_MAGIC
2886 * Add a new VD. Possibly also change the 'access' bits.
2887 * populated_vdes is the entry number.
2889 * New or updated VD. the VIRT_RECORD must already
2890 * exist. For an update, phys_refnum and lba_offset
2891 * (at least) are updated, and the VD_CONF must
2892 * be written to precisely those devices listed with
2894 * DDF_SPARE_ASSIGN_MAGIC
2895 * replacement Spare Assignment Record... but for which device?
2898 * - to create a new array, we send a VIRT_RECORD and
2899 * a VD_CONF. Then assemble and start the array.
2900 * - to activate a spare we send a VD_CONF to add the phys_refnum
2901 * and offset. This will also mark the spare as active with
2902 * a spare-assignment record.
2904 struct ddf_super
*ddf
= st
->sb
;
2905 __u32
*magic
= (__u32
*)update
->buf
;
2906 struct phys_disk
*pd
;
2907 struct virtual_disk
*vd
;
2908 struct vd_config
*vc
;
2914 dprintf("Process update %x\n", *magic
);
2917 case DDF_PHYS_RECORDS_MAGIC
:
2919 if (update
->len
!= (sizeof(struct phys_disk
) +
2920 sizeof(struct phys_disk_entry
)))
2922 pd
= (struct phys_disk
*)update
->buf
;
2924 ent
= __be16_to_cpu(pd
->used_pdes
);
2925 if (ent
>= __be16_to_cpu(ddf
->phys
->max_pdes
))
2927 if (!all_ff(ddf
->phys
->entries
[ent
].guid
))
2929 ddf
->phys
->entries
[ent
] = pd
->entries
[0];
2930 ddf
->phys
->used_pdes
= __cpu_to_be16(1 +
2931 __be16_to_cpu(ddf
->phys
->used_pdes
));
2932 ddf
->updates_pending
= 1;
2933 if (ddf
->add_list
) {
2934 struct active_array
*a
;
2935 struct dl
*al
= ddf
->add_list
;
2936 ddf
->add_list
= al
->next
;
2938 al
->next
= ddf
->dlist
;
2941 /* As a device has been added, we should check
2942 * for any degraded devices that might make
2943 * use of this spare */
2944 for (a
= st
->arrays
; a
; a
=a
->next
)
2945 a
->check_degraded
= 1;
2949 case DDF_VIRT_RECORDS_MAGIC
:
2951 if (update
->len
!= (sizeof(struct virtual_disk
) +
2952 sizeof(struct virtual_entry
)))
2954 vd
= (struct virtual_disk
*)update
->buf
;
2956 ent
= __be16_to_cpu(vd
->populated_vdes
);
2957 if (ent
>= __be16_to_cpu(ddf
->virt
->max_vdes
))
2959 if (!all_ff(ddf
->virt
->entries
[ent
].guid
))
2961 ddf
->virt
->entries
[ent
] = vd
->entries
[0];
2962 ddf
->virt
->populated_vdes
= __cpu_to_be16(1 +
2963 __be16_to_cpu(ddf
->virt
->populated_vdes
));
2964 ddf
->updates_pending
= 1;
2967 case DDF_VD_CONF_MAGIC
:
2968 dprintf("len %d %d\n", update
->len
, ddf
->conf_rec_len
);
2970 mppe
= __be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
2971 if (update
->len
!= ddf
->conf_rec_len
* 512)
2973 vc
= (struct vd_config
*)update
->buf
;
2974 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
2975 if (memcmp(vcl
->conf
.guid
, vc
->guid
, DDF_GUID_LEN
) == 0)
2977 dprintf("vcl = %p\n", vcl
);
2979 /* An update, just copy the phys_refnum and lba_offset
2982 memcpy(vcl
->conf
.phys_refnum
, vc
->phys_refnum
,
2983 mppe
* (sizeof(__u32
) + sizeof(__u64
)));
2986 vcl
= update
->space
;
2987 update
->space
= NULL
;
2988 vcl
->next
= ddf
->conflist
;
2989 memcpy(&vcl
->conf
, vc
, update
->len
);
2990 vcl
->lba_offset
= (__u64
*)
2991 &vcl
->conf
.phys_refnum
[mppe
];
2992 ddf
->conflist
= vcl
;
2994 /* Now make sure vlist is correct for each dl. */
2995 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
2998 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
2999 for (dn
=0; dn
< ddf
->mppe
; dn
++)
3000 if (vcl
->conf
.phys_refnum
[dn
] ==
3002 dprintf("dev %d has %p at %d\n",
3003 dl
->pdnum
, vcl
, vn
);
3004 dl
->vlist
[vn
++] = vcl
;
3007 while (vn
< ddf
->max_part
)
3008 dl
->vlist
[vn
++] = NULL
;
3010 ddf
->phys
->entries
[dl
->pdnum
].type
&=
3011 ~__cpu_to_be16(DDF_Global_Spare
);
3012 ddf
->phys
->entries
[dl
->pdnum
].type
|=
3013 __cpu_to_be16(DDF_Active_in_VD
);
3016 ddf
->phys
->entries
[dl
->pdnum
].type
&=
3017 ~__cpu_to_be16(DDF_Global_Spare
);
3018 ddf
->phys
->entries
[dl
->pdnum
].type
|=
3019 __cpu_to_be16(DDF_Spare
);
3021 if (!dl
->vlist
[0] && !dl
->spare
) {
3022 ddf
->phys
->entries
[dl
->pdnum
].type
|=
3023 __cpu_to_be16(DDF_Global_Spare
);
3024 ddf
->phys
->entries
[dl
->pdnum
].type
&=
3025 ~__cpu_to_be16(DDF_Spare
|
3029 ddf
->updates_pending
= 1;
3031 case DDF_SPARE_ASSIGN_MAGIC
:
3036 static void ddf_prepare_update(struct supertype
*st
,
3037 struct metadata_update
*update
)
3039 /* This update arrived at managemon.
3040 * We are about to pass it to monitor.
3041 * If a malloc is needed, do it here.
3043 struct ddf_super
*ddf
= st
->sb
;
3044 __u32
*magic
= (__u32
*)update
->buf
;
3045 if (*magic
== DDF_VD_CONF_MAGIC
)
3046 posix_memalign(&update
->space
, 512,
3047 offsetof(struct vcl
, conf
)
3048 + ddf
->conf_rec_len
* 512);
3052 * Check if the array 'a' is degraded but not failed.
3053 * If it is, find as many spares as are available and needed and
3054 * arrange for their inclusion.
3055 * We only choose devices which are not already in the array,
3056 * and prefer those with a spare-assignment to this array.
3057 * otherwise we choose global spares - assuming always that
3058 * there is enough room.
3059 * For each spare that we assign, we return an 'mdinfo' which
3060 * describes the position for the device in the array.
3061 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
3062 * the new phys_refnum and lba_offset values.
3064 * Only worry about BVDs at the moment.
3066 static struct mdinfo
*ddf_activate_spare(struct active_array
*a
,
3067 struct metadata_update
**updates
)
3071 struct ddf_super
*ddf
= a
->container
->sb
;
3073 struct mdinfo
*rv
= NULL
;
3075 struct metadata_update
*mu
;
3078 struct vd_config
*vc
;
3081 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
3082 if ((d
->curr_state
& DS_FAULTY
) &&
3084 /* wait for Removal to happen */
3086 if (d
->state_fd
>= 0)
3090 dprintf("ddf_activate: working=%d (%d) level=%d\n", working
, a
->info
.array
.raid_disks
,
3091 a
->info
.array
.level
);
3092 if (working
== a
->info
.array
.raid_disks
)
3093 return NULL
; /* array not degraded */
3094 switch (a
->info
.array
.level
) {
3097 return NULL
; /* failed */
3101 if (working
< a
->info
.array
.raid_disks
- 1)
3102 return NULL
; /* failed */
3105 if (working
< a
->info
.array
.raid_disks
- 2)
3106 return NULL
; /* failed */
3108 default: /* concat or stripe */
3109 return NULL
; /* failed */
3112 /* For each slot, if it is not working, find a spare */
3114 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
3115 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3116 if (d
->disk
.raid_disk
== i
)
3118 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
3119 if (d
&& (d
->state_fd
>= 0))
3122 /* OK, this device needs recovery. Find a spare */
3124 for ( ; dl
; dl
= dl
->next
) {
3125 unsigned long long esize
;
3126 unsigned long long pos
;
3129 int is_dedicated
= 0;
3132 /* If in this array, skip */
3133 for (d2
= a
->info
.devs
; d2
; d2
= d2
->next
)
3134 if (d2
->disk
.major
== dl
->major
&&
3135 d2
->disk
.minor
== dl
->minor
) {
3136 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
3141 if (ddf
->phys
->entries
[dl
->pdnum
].type
&
3142 __cpu_to_be16(DDF_Spare
)) {
3143 /* Check spare assign record */
3145 if (dl
->spare
->type
& DDF_spare_dedicated
) {
3146 /* check spare_ents for guid */
3148 j
< __be16_to_cpu(dl
->spare
->populated
);
3150 if (memcmp(dl
->spare
->spare_ents
[j
].guid
,
3151 ddf
->virt
->entries
[a
->info
.container_member
].guid
,
3158 } else if (ddf
->phys
->entries
[dl
->pdnum
].type
&
3159 __cpu_to_be16(DDF_Global_Spare
)) {
3162 if ( ! (is_dedicated
||
3163 (is_global
&& global_ok
))) {
3164 dprintf("%x:%x not suitable: %d %d\n", dl
->major
, dl
->minor
,
3165 is_dedicated
, is_global
);
3169 /* We are allowed to use this device - is there space?
3170 * We need a->info.component_size sectors */
3171 ex
= get_extents(ddf
, dl
);
3173 dprintf("cannot get extents\n");
3180 esize
= ex
[j
].start
- pos
;
3181 if (esize
>= a
->info
.component_size
)
3183 pos
= ex
[i
].start
+ ex
[i
].size
;
3185 } while (ex
[i
-1].size
);
3188 if (esize
< a
->info
.component_size
) {
3189 dprintf("%x:%x has no room: %llu %llu\n", dl
->major
, dl
->minor
,
3190 esize
, a
->info
.component_size
);
3195 /* Cool, we have a device with some space at pos */
3196 di
= malloc(sizeof(*di
));
3197 memset(di
, 0, sizeof(*di
));
3198 di
->disk
.number
= i
;
3199 di
->disk
.raid_disk
= i
;
3200 di
->disk
.major
= dl
->major
;
3201 di
->disk
.minor
= dl
->minor
;
3203 di
->data_offset
= pos
;
3204 di
->component_size
= a
->info
.component_size
;
3205 di
->container_member
= dl
->pdnum
;
3208 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
3213 if (!dl
&& ! global_ok
) {
3214 /* not enough dedicated spares, try global */
3222 /* No spares found */
3224 /* Now 'rv' has a list of devices to return.
3225 * Create a metadata_update record to update the
3226 * phys_refnum and lba_offset values
3228 mu
= malloc(sizeof(*mu
));
3229 mu
->buf
= malloc(ddf
->conf_rec_len
* 512);
3230 posix_memalign(&mu
->space
, 512, sizeof(struct vcl
));
3231 mu
->len
= ddf
->conf_rec_len
;
3232 mu
->next
= *updates
;
3233 vc
= find_vdcr(ddf
, a
->info
.container_member
);
3234 memcpy(mu
->buf
, vc
, ddf
->conf_rec_len
* 512);
3236 vc
= (struct vd_config
*)mu
->buf
;
3237 lba
= (__u64
*)&vc
->phys_refnum
[ddf
->mppe
];
3238 for (di
= rv
; di
; di
= di
->next
) {
3239 vc
->phys_refnum
[di
->disk
.raid_disk
] =
3240 ddf
->phys
->entries
[dl
->pdnum
].refnum
;
3241 lba
[di
->disk
.raid_disk
] = di
->data_offset
;
3247 struct superswitch super_ddf
= {
3249 .examine_super
= examine_super_ddf
,
3250 .brief_examine_super
= brief_examine_super_ddf
,
3251 .detail_super
= detail_super_ddf
,
3252 .brief_detail_super
= brief_detail_super_ddf
,
3253 .validate_geometry
= validate_geometry_ddf
,
3254 .write_init_super
= write_init_super_ddf
,
3256 .match_home
= match_home_ddf
,
3257 .uuid_from_super
= uuid_from_super_ddf
,
3258 .getinfo_super
= getinfo_super_ddf
,
3259 .update_super
= update_super_ddf
,
3261 .avail_size
= avail_size_ddf
,
3263 .compare_super
= compare_super_ddf
,
3265 .load_super
= load_super_ddf
,
3266 .init_super
= init_super_ddf
,
3267 .store_super
= store_zero_ddf
,
3268 .free_super
= free_super_ddf
,
3269 .match_metadata_desc
= match_metadata_desc_ddf
,
3270 .add_to_super
= add_to_super_ddf
,
3271 .container_content
= container_content_ddf
,
3276 .open_new
= ddf_open_new
,
3277 .set_array_state
= ddf_set_array_state
,
3278 .set_disk
= ddf_set_disk
,
3279 .sync_metadata
= ddf_sync_metadata
,
3280 .process_update
= ddf_process_update
,
3281 .prepare_update
= ddf_prepare_update
,
3282 .activate_spare
= ddf_activate_spare
,