2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2006-2009 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neil@brown.name>
24 * Specifications for DDF takes from Common RAID DDF Specification Revision 1.2
25 * (July 28 2006). Reused by permission of SNIA.
28 #define HAVE_STDINT_H 1
34 /* a non-official T10 name for creation GUIDs */
35 static char T10
[] = "Linux-MD";
37 /* DDF timestamps are 1980 based, so we need to add
38 * second-in-decade-of-seventies to convert to linux timestamps.
39 * 10 years with 2 leap years.
41 #define DECADE (3600*24*(365*10+2))
44 const unsigned char *buf
,
47 /* The DDF metadata handling.
48 * DDF metadata lives at the end of the device.
49 * The last 512 byte block provides an 'anchor' which is used to locate
50 * the rest of the metadata which usually lives immediately behind the anchor.
53 * - all multibyte numeric fields are bigendian.
54 * - all strings are space padded.
58 /* Primary Raid Level (PRL) */
59 #define DDF_RAID0 0x00
60 #define DDF_RAID1 0x01
61 #define DDF_RAID3 0x03
62 #define DDF_RAID4 0x04
63 #define DDF_RAID5 0x05
64 #define DDF_RAID1E 0x11
66 #define DDF_CONCAT 0x1f
67 #define DDF_RAID5E 0x15
68 #define DDF_RAID5EE 0x25
69 #define DDF_RAID6 0x06
71 /* Raid Level Qualifier (RLQ) */
72 #define DDF_RAID0_SIMPLE 0x00
73 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
74 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
75 #define DDF_RAID3_0 0x00 /* parity in first extent */
76 #define DDF_RAID3_N 0x01 /* parity in last extent */
77 #define DDF_RAID4_0 0x00 /* parity in first extent */
78 #define DDF_RAID4_N 0x01 /* parity in last extent */
79 /* these apply to raid5e and raid5ee as well */
80 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
81 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
82 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
83 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
85 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
86 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
88 /* Secondary RAID Level (SRL) */
89 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
90 #define DDF_2MIRRORED 0x01
91 #define DDF_2CONCAT 0x02
92 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
95 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
96 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
97 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
98 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
99 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
100 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
101 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
102 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
103 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
104 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
106 #define DDF_GUID_LEN 24
107 #define DDF_REVISION_0 "01.00.00"
108 #define DDF_REVISION_2 "01.02.00"
111 __u32 magic
; /* DDF_HEADER_MAGIC */
113 char guid
[DDF_GUID_LEN
];
114 char revision
[8]; /* 01.02.00 */
115 __u32 seq
; /* starts at '1' */
120 __u8 pad0
; /* 0xff */
121 __u8 pad1
[12]; /* 12 * 0xff */
122 /* 64 bytes so far */
123 __u8 header_ext
[32]; /* reserved: fill with 0xff */
127 __u8 pad2
[3]; /* 0xff */
128 __u32 workspace_len
; /* sectors for vendor space -
129 * at least 32768(sectors) */
131 __u16 max_pd_entries
; /* one of 15, 63, 255, 1023, 4095 */
132 __u16 max_vd_entries
; /* 2^(4,6,8,10,12)-1 : i.e. as above */
133 __u16 max_partitions
; /* i.e. max num of configuration
134 record entries per disk */
135 __u16 config_record_len
; /* 1 +ROUNDUP(max_primary_element_entries
137 __u16 max_primary_element_entries
; /* 16, 64, 256, 1024, or 4096 */
138 __u8 pad3
[54]; /* 0xff */
139 /* 192 bytes so far */
140 __u32 controller_section_offset
;
141 __u32 controller_section_length
;
142 __u32 phys_section_offset
;
143 __u32 phys_section_length
;
144 __u32 virt_section_offset
;
145 __u32 virt_section_length
;
146 __u32 config_section_offset
;
147 __u32 config_section_length
;
148 __u32 data_section_offset
;
149 __u32 data_section_length
;
150 __u32 bbm_section_offset
;
151 __u32 bbm_section_length
;
152 __u32 diag_space_offset
;
153 __u32 diag_space_length
;
156 /* 256 bytes so far */
157 __u8 pad4
[256]; /* 0xff */
161 #define DDF_HEADER_ANCHOR 0x00
162 #define DDF_HEADER_PRIMARY 0x01
163 #define DDF_HEADER_SECONDARY 0x02
165 /* The content of the 'controller section' - global scope */
166 struct ddf_controller_data
{
167 __u32 magic
; /* DDF_CONTROLLER_MAGIC */
169 char guid
[DDF_GUID_LEN
];
170 struct controller_type
{
177 __u8 pad
[8]; /* 0xff */
178 __u8 vendor_data
[448];
181 /* The content of phys_section - global scope */
183 __u32 magic
; /* DDF_PHYS_RECORDS_MAGIC */
188 struct phys_disk_entry
{
189 char guid
[DDF_GUID_LEN
];
193 __u64 config_size
; /* DDF structures must be after here */
194 char path
[18]; /* another horrible structure really */
199 /* phys_disk_entry.type is a bitmap - bigendian remember */
200 #define DDF_Forced_PD_GUID 1
201 #define DDF_Active_in_VD 2
202 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
203 #define DDF_Spare 8 /* overrides Global_spare */
204 #define DDF_Foreign 16
205 #define DDF_Legacy 32 /* no DDF on this device */
207 #define DDF_Interface_mask 0xf00
208 #define DDF_Interface_SCSI 0x100
209 #define DDF_Interface_SAS 0x200
210 #define DDF_Interface_SATA 0x300
211 #define DDF_Interface_FC 0x400
213 /* phys_disk_entry.state is a bigendian bitmap */
215 #define DDF_Failed 2 /* overrides 1,4,8 */
216 #define DDF_Rebuilding 4
217 #define DDF_Transition 8
219 #define DDF_ReadErrors 32
220 #define DDF_Missing 64
222 /* The content of the virt_section global scope */
223 struct virtual_disk
{
224 __u32 magic
; /* DDF_VIRT_RECORDS_MAGIC */
226 __u16 populated_vdes
;
229 struct virtual_entry
{
230 char guid
[DDF_GUID_LEN
];
232 __u16 pad0
; /* 0xffff */
242 /* virtual_entry.type is a bitmap - bigendian */
244 #define DDF_Enforce_Groups 2
245 #define DDF_Unicode 4
246 #define DDF_Owner_Valid 8
248 /* virtual_entry.state is a bigendian bitmap */
249 #define DDF_state_mask 0x7
250 #define DDF_state_optimal 0x0
251 #define DDF_state_degraded 0x1
252 #define DDF_state_deleted 0x2
253 #define DDF_state_missing 0x3
254 #define DDF_state_failed 0x4
255 #define DDF_state_part_optimal 0x5
257 #define DDF_state_morphing 0x8
258 #define DDF_state_inconsistent 0x10
260 /* virtual_entry.init_state is a bigendian bitmap */
261 #define DDF_initstate_mask 0x03
262 #define DDF_init_not 0x00
263 #define DDF_init_quick 0x01 /* initialisation is progress.
264 * i.e. 'state_inconsistent' */
265 #define DDF_init_full 0x02
267 #define DDF_access_mask 0xc0
268 #define DDF_access_rw 0x00
269 #define DDF_access_ro 0x80
270 #define DDF_access_blocked 0xc0
272 /* The content of the config_section - local scope
273 * It has multiple records each config_record_len sectors
274 * They can be vd_config or spare_assign
278 __u32 magic
; /* DDF_VD_CONF_MAGIC */
280 char guid
[DDF_GUID_LEN
];
284 __u16 prim_elmnt_count
;
285 __u8 chunk_shift
; /* 0 == 512, 1==1024 etc */
288 __u8 sec_elmnt_count
;
291 __u64 blocks
; /* blocks per component could be different
292 * on different component devices...(only
293 * for concat I hope) */
294 __u64 array_blocks
; /* blocks in array */
302 __u8 v0
[32]; /* reserved- 0xff */
303 __u8 v1
[32]; /* reserved- 0xff */
304 __u8 v2
[16]; /* reserved- 0xff */
305 __u8 v3
[16]; /* reserved- 0xff */
307 __u32 phys_refnum
[0]; /* refnum of each disk in sequence */
308 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
309 bvd are always the same size */
312 /* vd_config.cache_pol[7] is a bitmap */
313 #define DDF_cache_writeback 1 /* else writethrough */
314 #define DDF_cache_wadaptive 2 /* only applies if writeback */
315 #define DDF_cache_readahead 4
316 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
317 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
318 #define DDF_cache_wallowed 32 /* enable write caching */
319 #define DDF_cache_rallowed 64 /* enable read caching */
321 struct spare_assign
{
322 __u32 magic
; /* DDF_SPARE_ASSIGN_MAGIC */
327 __u16 populated
; /* SAEs used */
328 __u16 max
; /* max SAEs */
330 struct spare_assign_entry
{
331 char guid
[DDF_GUID_LEN
];
332 __u16 secondary_element
;
336 /* spare_assign.type is a bitmap */
337 #define DDF_spare_dedicated 0x1 /* else global */
338 #define DDF_spare_revertible 0x2 /* else committable */
339 #define DDF_spare_active 0x4 /* else not active */
340 #define DDF_spare_affinity 0x8 /* enclosure affinity */
342 /* The data_section contents - local scope */
344 __u32 magic
; /* DDF_PHYS_DATA_MAGIC */
346 char guid
[DDF_GUID_LEN
];
347 __u32 refnum
; /* crc of some magic drive data ... */
348 __u8 forced_ref
; /* set when above was not result of magic */
349 __u8 forced_guid
; /* set if guid was forced rather than magic */
354 /* bbm_section content */
355 struct bad_block_log
{
362 struct mapped_block
{
363 __u64 defective_start
;
364 __u32 replacement_start
;
370 /* Struct for internally holding ddf structures */
371 /* The DDF structure stored on each device is potentially
372 * quite different, as some data is global and some is local.
373 * The global data is:
376 * - Physical disk records
377 * - Virtual disk records
379 * - Configuration records
380 * - Physical Disk data section
381 * ( and Bad block and vendor which I don't care about yet).
383 * The local data is parsed into separate lists as it is read
384 * and reconstructed for writing. This means that we only need
385 * to make config changes once and they are automatically
386 * propagated to all devices.
387 * Note that the ddf_super has space of the conf and disk data
388 * for this disk and also for a list of all such data.
389 * The list is only used for the superblock that is being
390 * built in Create or Assemble to describe the whole array.
393 struct ddf_header anchor
, primary
, secondary
;
394 struct ddf_controller_data controller
;
395 struct ddf_header
*active
;
396 struct phys_disk
*phys
;
397 struct virtual_disk
*virt
;
399 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
;
426 void *mdupdate
; /* hold metadata update */
428 /* These fields used by auto-layout */
429 int raiddisk
; /* slot to fill in autolayout */
433 struct disk_data disk
;
434 struct vcl
*vlist
[0]; /* max_part in size */
439 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
443 static int calc_crc(void *buf
, int len
)
445 /* crcs are always at the same place as in the ddf_header */
446 struct ddf_header
*ddf
= buf
;
447 __u32 oldcrc
= ddf
->crc
;
449 ddf
->crc
= 0xffffffff;
451 newcrc
= crc32(0, buf
, len
);
453 /* The crc is store (like everything) bigendian, so convert
454 * here for simplicity
456 return __cpu_to_be32(newcrc
);
459 static int load_ddf_header(int fd
, unsigned long long lba
,
460 unsigned long long size
,
462 struct ddf_header
*hdr
, struct ddf_header
*anchor
)
464 /* read a ddf header (primary or secondary) from fd/lba
465 * and check that it is consistent with anchor
467 * magic, crc, guid, rev, and LBA's header_type, and
468 * everything after header_type must be the same
473 if (lseek64(fd
, lba
<<9, 0) < 0)
476 if (read(fd
, hdr
, 512) != 512)
479 if (hdr
->magic
!= DDF_HEADER_MAGIC
)
481 if (calc_crc(hdr
, 512) != hdr
->crc
)
483 if (memcmp(anchor
->guid
, hdr
->guid
, DDF_GUID_LEN
) != 0 ||
484 memcmp(anchor
->revision
, hdr
->revision
, 8) != 0 ||
485 anchor
->primary_lba
!= hdr
->primary_lba
||
486 anchor
->secondary_lba
!= hdr
->secondary_lba
||
488 memcmp(anchor
->pad2
, hdr
->pad2
, 512 -
489 offsetof(struct ddf_header
, pad2
)) != 0)
492 /* Looks good enough to me... */
496 static void *load_section(int fd
, struct ddf_super
*super
, void *buf
,
497 __u32 offset_be
, __u32 len_be
, int check
)
499 unsigned long long offset
= __be32_to_cpu(offset_be
);
500 unsigned long long len
= __be32_to_cpu(len_be
);
501 int dofree
= (buf
== NULL
);
504 if (len
!= 2 && len
!= 8 && len
!= 32
505 && len
!= 128 && len
!= 512)
511 /* All pre-allocated sections are a single block */
514 } else if (posix_memalign(&buf
, 512, len
<<9) != 0)
520 if (super
->active
->type
== 1)
521 offset
+= __be64_to_cpu(super
->active
->primary_lba
);
523 offset
+= __be64_to_cpu(super
->active
->secondary_lba
);
525 if (lseek64(fd
, offset
<<9, 0) != (offset
<<9)) {
530 if (read(fd
, buf
, len
<<9) != (len
<<9)) {
538 static int load_ddf_headers(int fd
, struct ddf_super
*super
, char *devname
)
540 unsigned long long dsize
;
542 get_dev_size(fd
, NULL
, &dsize
);
544 if (lseek64(fd
, dsize
-512, 0) < 0) {
547 Name
": Cannot seek to anchor block on %s: %s\n",
548 devname
, strerror(errno
));
551 if (read(fd
, &super
->anchor
, 512) != 512) {
554 Name
": Cannot read anchor block on %s: %s\n",
555 devname
, strerror(errno
));
558 if (super
->anchor
.magic
!= DDF_HEADER_MAGIC
) {
560 fprintf(stderr
, Name
": no DDF anchor found on %s\n",
564 if (calc_crc(&super
->anchor
, 512) != super
->anchor
.crc
) {
566 fprintf(stderr
, Name
": bad CRC on anchor on %s\n",
570 if (memcmp(super
->anchor
.revision
, DDF_REVISION_0
, 8) != 0 &&
571 memcmp(super
->anchor
.revision
, DDF_REVISION_2
, 8) != 0) {
573 fprintf(stderr
, Name
": can only support super revision"
574 " %.8s and earlier, not %.8s on %s\n",
575 DDF_REVISION_2
, super
->anchor
.revision
,devname
);
578 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.primary_lba
),
580 &super
->primary
, &super
->anchor
) == 0) {
583 Name
": Failed to load primary DDF header "
587 super
->active
= &super
->primary
;
588 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.secondary_lba
),
590 &super
->secondary
, &super
->anchor
)) {
591 if ((__be32_to_cpu(super
->primary
.seq
)
592 < __be32_to_cpu(super
->secondary
.seq
) &&
593 !super
->secondary
.openflag
)
594 || (__be32_to_cpu(super
->primary
.seq
)
595 == __be32_to_cpu(super
->secondary
.seq
) &&
596 super
->primary
.openflag
&& !super
->secondary
.openflag
)
598 super
->active
= &super
->secondary
;
603 static int load_ddf_global(int fd
, struct ddf_super
*super
, char *devname
)
606 ok
= load_section(fd
, super
, &super
->controller
,
607 super
->active
->controller_section_offset
,
608 super
->active
->controller_section_length
,
610 super
->phys
= load_section(fd
, super
, NULL
,
611 super
->active
->phys_section_offset
,
612 super
->active
->phys_section_length
,
614 super
->pdsize
= __be32_to_cpu(super
->active
->phys_section_length
) * 512;
616 super
->virt
= load_section(fd
, super
, NULL
,
617 super
->active
->virt_section_offset
,
618 super
->active
->virt_section_length
,
620 super
->vdsize
= __be32_to_cpu(super
->active
->virt_section_length
) * 512;
630 super
->conflist
= NULL
;
633 super
->max_part
= __be16_to_cpu(super
->active
->max_partitions
);
634 super
->mppe
= __be16_to_cpu(super
->active
->max_primary_element_entries
);
635 super
->conf_rec_len
= __be16_to_cpu(super
->active
->config_record_len
);
639 static int load_ddf_local(int fd
, struct ddf_super
*super
,
640 char *devname
, int keep
)
648 int max_virt_disks
= __be16_to_cpu(super
->active
->max_vd_entries
);
649 unsigned long long dsize
;
651 /* First the local disk info */
652 if (posix_memalign((void**)&dl
, 512,
654 (super
->max_part
) * sizeof(dl
->vlist
[0])) != 0) {
655 fprintf(stderr
, Name
": %s could not allocate disk info buffer\n",
660 load_section(fd
, super
, &dl
->disk
,
661 super
->active
->data_section_offset
,
662 super
->active
->data_section_length
,
664 dl
->devname
= devname
? strdup(devname
) : NULL
;
667 dl
->major
= major(stb
.st_rdev
);
668 dl
->minor
= minor(stb
.st_rdev
);
669 dl
->next
= super
->dlist
;
670 dl
->fd
= keep
? fd
: -1;
673 if (get_dev_size(fd
, devname
, &dsize
))
674 dl
->size
= dsize
>> 9;
676 for (i
=0 ; i
< super
->max_part
; i
++)
680 for (i
=0; i
< __be16_to_cpu(super
->active
->max_pd_entries
); i
++)
681 if (memcmp(super
->phys
->entries
[i
].guid
,
682 dl
->disk
.guid
, DDF_GUID_LEN
) == 0)
685 /* Now the config list. */
686 /* 'conf' is an array of config entries, some of which are
687 * probably invalid. Those which are good need to be copied into
691 conf
= load_section(fd
, super
, NULL
,
692 super
->active
->config_section_offset
,
693 super
->active
->config_section_length
,
698 confsec
< __be32_to_cpu(super
->active
->config_section_length
);
699 confsec
+= super
->conf_rec_len
) {
700 struct vd_config
*vd
=
701 (struct vd_config
*)((char*)conf
+ confsec
*512);
704 if (vd
->magic
== DDF_SPARE_ASSIGN_MAGIC
) {
707 if (posix_memalign((void**)&dl
->spare
, 512,
708 super
->conf_rec_len
*512) != 0) {
710 ": %s could not allocate spare info buf\n",
715 memcpy(dl
->spare
, vd
, super
->conf_rec_len
*512);
718 if (vd
->magic
!= DDF_VD_CONF_MAGIC
)
720 for (vcl
= super
->conflist
; vcl
; vcl
= vcl
->next
) {
721 if (memcmp(vcl
->conf
.guid
,
722 vd
->guid
, DDF_GUID_LEN
) == 0)
727 dl
->vlist
[vnum
++] = vcl
;
728 if (__be32_to_cpu(vd
->seqnum
) <=
729 __be32_to_cpu(vcl
->conf
.seqnum
))
732 if (posix_memalign((void**)&vcl
, 512,
733 (super
->conf_rec_len
*512 +
734 offsetof(struct vcl
, conf
))) != 0) {
736 ": %s could not allocate vcl buf\n",
740 vcl
->next
= super
->conflist
;
741 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
742 super
->conflist
= vcl
;
743 dl
->vlist
[vnum
++] = vcl
;
745 memcpy(&vcl
->conf
, vd
, super
->conf_rec_len
*512);
746 vcl
->lba_offset
= (__u64
*)
747 &vcl
->conf
.phys_refnum
[super
->mppe
];
749 for (i
=0; i
< max_virt_disks
; i
++)
750 if (memcmp(super
->virt
->entries
[i
].guid
,
751 vcl
->conf
.guid
, DDF_GUID_LEN
)==0)
753 if (i
< max_virt_disks
)
762 static int load_super_ddf_all(struct supertype
*st
, int fd
,
763 void **sbp
, char *devname
, int keep_fd
);
765 static int load_super_ddf(struct supertype
*st
, int fd
,
768 unsigned long long dsize
;
769 struct ddf_super
*super
;
773 /* if 'fd' is a container, load metadata from all the devices */
774 if (load_super_ddf_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
778 return 1; /* FIXME Is this correct */
780 if (get_dev_size(fd
, devname
, &dsize
) == 0)
783 /* 32M is a lower bound */
784 if (dsize
<= 32*1024*1024) {
787 Name
": %s is too small for ddf: "
788 "size is %llu sectors.\n",
795 Name
": %s is an odd size for ddf: "
796 "size is %llu bytes.\n",
801 if (posix_memalign((void**)&super
, 512, sizeof(*super
))!= 0) {
802 fprintf(stderr
, Name
": malloc of %zu failed.\n",
806 memset(super
, 0, sizeof(*super
));
808 rv
= load_ddf_headers(fd
, super
, devname
);
814 /* Have valid headers and have chosen the best. Let's read in the rest*/
816 rv
= load_ddf_global(fd
, super
, devname
);
821 Name
": Failed to load all information "
822 "sections on %s\n", devname
);
827 rv
= load_ddf_local(fd
, super
, devname
, 0);
832 Name
": Failed to load all information "
833 "sections on %s\n", devname
);
838 if (st
->subarray
[0]) {
841 for (v
= super
->conflist
; v
; v
= v
->next
)
842 if (v
->vcnum
== atoi(st
->subarray
))
843 super
->currentconf
= v
;
844 if (!super
->currentconf
) {
850 /* Should possibly check the sections .... */
853 if (st
->ss
== NULL
) {
855 st
->minor_version
= 0;
858 st
->loaded_container
= 0;
863 static void free_super_ddf(struct supertype
*st
)
865 struct ddf_super
*ddf
= st
->sb
;
870 while (ddf
->conflist
) {
871 struct vcl
*v
= ddf
->conflist
;
872 ddf
->conflist
= v
->next
;
874 free(v
->block_sizes
);
878 struct dl
*d
= ddf
->dlist
;
879 ddf
->dlist
= d
->next
;
890 static struct supertype
*match_metadata_desc_ddf(char *arg
)
892 /* 'ddf' only support containers */
893 struct supertype
*st
;
894 if (strcmp(arg
, "ddf") != 0 &&
895 strcmp(arg
, "default") != 0
899 st
= malloc(sizeof(*st
));
900 memset(st
, 0, sizeof(*st
));
903 st
->minor_version
= 0;
911 static mapping_t ddf_state
[] = {
917 { "Partially Optimal", 5},
923 static mapping_t ddf_init_state
[] = {
924 { "Not Initialised", 0},
925 { "QuickInit in Progress", 1},
926 { "Fully Initialised", 2},
930 static mapping_t ddf_access
[] = {
934 { "Blocked (no access)", 3},
938 static mapping_t ddf_level
[] = {
939 { "RAID0", DDF_RAID0
},
940 { "RAID1", DDF_RAID1
},
941 { "RAID3", DDF_RAID3
},
942 { "RAID4", DDF_RAID4
},
943 { "RAID5", DDF_RAID5
},
944 { "RAID1E",DDF_RAID1E
},
946 { "CONCAT",DDF_CONCAT
},
947 { "RAID5E",DDF_RAID5E
},
948 { "RAID5EE",DDF_RAID5EE
},
949 { "RAID6", DDF_RAID6
},
952 static mapping_t ddf_sec_level
[] = {
953 { "Striped", DDF_2STRIPED
},
954 { "Mirrored", DDF_2MIRRORED
},
955 { "Concat", DDF_2CONCAT
},
956 { "Spanned", DDF_2SPANNED
},
964 static struct num_mapping ddf_level_num
[] = {
967 { DDF_RAID3
, LEVEL_UNSUPPORTED
},
970 { DDF_RAID1E
, LEVEL_UNSUPPORTED
},
971 { DDF_JBOD
, LEVEL_UNSUPPORTED
},
972 { DDF_CONCAT
, LEVEL_LINEAR
},
973 { DDF_RAID5E
, LEVEL_UNSUPPORTED
},
974 { DDF_RAID5EE
, LEVEL_UNSUPPORTED
},
979 static int map_num1(struct num_mapping
*map
, int num
)
982 for (i
=0 ; map
[i
].num1
!= MAXINT
; i
++)
983 if (map
[i
].num1
== num
)
988 static int all_ff(char *guid
)
991 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
992 if (guid
[i
] != (char)0xff)
998 static void print_guid(char *guid
, int tstamp
)
1000 /* A GUIDs are part (or all) ASCII and part binary.
1001 * They tend to be space padded.
1002 * We print the GUID in HEX, then in parentheses add
1003 * any initial ASCII sequence, and a possible
1004 * time stamp from bytes 16-19
1006 int l
= DDF_GUID_LEN
;
1009 for (i
=0 ; i
<DDF_GUID_LEN
; i
++) {
1010 if ((i
&3)==0 && i
!= 0) printf(":");
1011 printf("%02X", guid
[i
]&255);
1015 while (l
&& guid
[l
-1] == ' ')
1017 for (i
=0 ; i
<l
; i
++) {
1018 if (guid
[i
] >= 0x20 && guid
[i
] < 0x7f)
1019 fputc(guid
[i
], stdout
);
1024 time_t then
= __be32_to_cpu(*(__u32
*)(guid
+16)) + DECADE
;
1027 tm
= localtime(&then
);
1028 strftime(tbuf
, 100, " %D %T",tm
);
1029 fputs(tbuf
, stdout
);
1034 static void examine_vd(int n
, struct ddf_super
*sb
, char *guid
)
1036 int crl
= sb
->conf_rec_len
;
1039 for (vcl
= sb
->conflist
; vcl
; vcl
= vcl
->next
) {
1041 struct vd_config
*vc
= &vcl
->conf
;
1043 if (calc_crc(vc
, crl
*512) != vc
->crc
)
1045 if (memcmp(vc
->guid
, guid
, DDF_GUID_LEN
) != 0)
1048 /* Ok, we know about this VD, let's give more details */
1049 printf(" Raid Devices[%d] : %d (", n
,
1050 __be16_to_cpu(vc
->prim_elmnt_count
));
1051 for (i
=0; i
<__be16_to_cpu(vc
->prim_elmnt_count
); i
++) {
1053 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1054 for (j
=0; j
<cnt
; j
++)
1055 if (vc
->phys_refnum
[i
] == sb
->phys
->entries
[j
].refnum
)
1064 if (vc
->chunk_shift
!= 255)
1065 printf(" Chunk Size[%d] : %d sectors\n", n
,
1066 1 << vc
->chunk_shift
);
1067 printf(" Raid Level[%d] : %s\n", n
,
1068 map_num(ddf_level
, vc
->prl
)?:"-unknown-");
1069 if (vc
->sec_elmnt_count
!= 1) {
1070 printf(" Secondary Position[%d] : %d of %d\n", n
,
1071 vc
->sec_elmnt_seq
, vc
->sec_elmnt_count
);
1072 printf(" Secondary Level[%d] : %s\n", n
,
1073 map_num(ddf_sec_level
, vc
->srl
) ?: "-unknown-");
1075 printf(" Device Size[%d] : %llu\n", n
,
1076 (unsigned long long)__be64_to_cpu(vc
->blocks
)/2);
1077 printf(" Array Size[%d] : %llu\n", n
,
1078 (unsigned long long)__be64_to_cpu(vc
->array_blocks
)/2);
1082 static void examine_vds(struct ddf_super
*sb
)
1084 int cnt
= __be16_to_cpu(sb
->virt
->populated_vdes
);
1086 printf(" Virtual Disks : %d\n", cnt
);
1088 for (i
=0; i
<cnt
; i
++) {
1089 struct virtual_entry
*ve
= &sb
->virt
->entries
[i
];
1091 printf(" VD GUID[%d] : ", i
); print_guid(ve
->guid
, 1);
1093 printf(" unit[%d] : %d\n", i
, __be16_to_cpu(ve
->unit
));
1094 printf(" state[%d] : %s, %s%s\n", i
,
1095 map_num(ddf_state
, ve
->state
& 7),
1096 (ve
->state
& 8) ? "Morphing, ": "",
1097 (ve
->state
& 16)? "Not Consistent" : "Consistent");
1098 printf(" init state[%d] : %s\n", i
,
1099 map_num(ddf_init_state
, ve
->init_state
&3));
1100 printf(" access[%d] : %s\n", i
,
1101 map_num(ddf_access
, (ve
->init_state
>>6) & 3));
1102 printf(" Name[%d] : %.16s\n", i
, ve
->name
);
1103 examine_vd(i
, sb
, ve
->guid
);
1105 if (cnt
) printf("\n");
1108 static void examine_pds(struct ddf_super
*sb
)
1110 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1113 printf(" Physical Disks : %d\n", cnt
);
1114 printf(" Number RefNo Size Device Type/State\n");
1116 for (i
=0 ; i
<cnt
; i
++) {
1117 struct phys_disk_entry
*pd
= &sb
->phys
->entries
[i
];
1118 int type
= __be16_to_cpu(pd
->type
);
1119 int state
= __be16_to_cpu(pd
->state
);
1121 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1123 printf(" %3d %08x ", i
,
1124 __be32_to_cpu(pd
->refnum
));
1126 (unsigned long long)__be64_to_cpu(pd
->config_size
)>>1);
1127 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1128 if (dl
->disk
.refnum
== pd
->refnum
) {
1129 char *dv
= map_dev(dl
->major
, dl
->minor
, 0);
1131 printf("%-15s", dv
);
1138 printf(" %s%s%s%s%s",
1139 (type
&2) ? "active":"",
1140 (type
&4) ? "Global-Spare":"",
1141 (type
&8) ? "spare" : "",
1142 (type
&16)? ", foreign" : "",
1143 (type
&32)? "pass-through" : "");
1144 printf("/%s%s%s%s%s%s%s",
1145 (state
&1)? "Online": "Offline",
1146 (state
&2)? ", Failed": "",
1147 (state
&4)? ", Rebuilding": "",
1148 (state
&8)? ", in-transition": "",
1149 (state
&16)? ", SMART-errors": "",
1150 (state
&32)? ", Unrecovered-Read-Errors": "",
1151 (state
&64)? ", Missing" : "");
1156 static void examine_super_ddf(struct supertype
*st
, char *homehost
)
1158 struct ddf_super
*sb
= st
->sb
;
1160 printf(" Magic : %08x\n", __be32_to_cpu(sb
->anchor
.magic
));
1161 printf(" Version : %.8s\n", sb
->anchor
.revision
);
1162 printf("Controller GUID : "); print_guid(sb
->controller
.guid
, 0);
1164 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1166 printf(" Seq : %08x\n", __be32_to_cpu(sb
->active
->seq
));
1167 printf(" Redundant hdr : %s\n", sb
->secondary
.magic
== DDF_HEADER_MAGIC
1173 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
);
1175 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4]);
1177 static void brief_examine_super_ddf(struct supertype
*st
, int verbose
)
1179 /* We just write a generic DDF ARRAY entry
1181 struct ddf_super
*ddf
= st
->sb
;
1185 getinfo_super_ddf(st
, &info
);
1186 fname_from_uuid(st
, &info
, nbuf
, ':');
1188 for (i
=0; i
<__be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
1189 struct virtual_entry
*ve
= &ddf
->virt
->entries
[i
];
1192 if (all_ff(ve
->guid
))
1194 memcpy(vcl
.conf
.guid
, ve
->guid
, DDF_GUID_LEN
);
1195 ddf
->currentconf
=&vcl
;
1196 uuid_from_super_ddf(st
, info
.uuid
);
1197 fname_from_uuid(st
, &info
, nbuf1
, ':');
1198 printf("ARRAY container=%s member=%d UUID=%s\n",
1199 nbuf
+5, i
, nbuf1
+5);
1201 printf("ARRAY metadata=ddf UUID=%s\n", nbuf
+ 5);
1204 static void export_examine_super_ddf(struct supertype
*st
)
1208 getinfo_super_ddf(st
, &info
);
1209 fname_from_uuid(st
, &info
, nbuf
, ':');
1210 printf("MD_METADATA=ddf\n");
1211 printf("MD_LEVEL=container\n");
1212 printf("MD_UUID=%s\n", nbuf
+5);
1216 static void detail_super_ddf(struct supertype
*st
, char *homehost
)
1219 * Could print DDF GUID
1220 * Need to find which array
1221 * If whole, briefly list all arrays
1226 static void brief_detail_super_ddf(struct supertype
*st
)
1228 /* FIXME I really need to know which array we are detailing.
1229 * Can that be stored in ddf_super??
1231 // struct ddf_super *ddf = st->sb;
1234 getinfo_super_ddf(st
, &info
);
1235 fname_from_uuid(st
, &info
, nbuf
,':');
1236 printf(" UUID=%s", nbuf
+ 5);
1240 static int match_home_ddf(struct supertype
*st
, char *homehost
)
1242 /* It matches 'this' host if the controller is a
1243 * Linux-MD controller with vendor_data matching
1246 struct ddf_super
*ddf
= st
->sb
;
1247 int len
= strlen(homehost
);
1249 return (memcmp(ddf
->controller
.guid
, T10
, 8) == 0 &&
1250 len
< sizeof(ddf
->controller
.vendor_data
) &&
1251 memcmp(ddf
->controller
.vendor_data
, homehost
,len
) == 0 &&
1252 ddf
->controller
.vendor_data
[len
] == 0);
1256 static struct vd_config
*find_vdcr(struct ddf_super
*ddf
, int inst
)
1260 for (v
= ddf
->conflist
; v
; v
= v
->next
)
1261 if (inst
== v
->vcnum
)
1267 static int find_phys(struct ddf_super
*ddf
, __u32 phys_refnum
)
1269 /* Find the entry in phys_disk which has the given refnum
1270 * and return it's index
1273 for (i
=0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
); i
++)
1274 if (ddf
->phys
->entries
[i
].refnum
== phys_refnum
)
1279 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4])
1281 /* The uuid returned here is used for:
1282 * uuid to put into bitmap file (Create, Grow)
1283 * uuid for backup header when saving critical section (Grow)
1284 * comparing uuids when re-adding a device into an array
1285 * In these cases the uuid required is that of the data-array,
1286 * not the device-set.
1287 * uuid to recognise same set when adding a missing device back
1288 * to an array. This is a uuid for the device-set.
1290 * For each of these we can make do with a truncated
1291 * or hashed uuid rather than the original, as long as
1293 * In the case of SVD we assume the BVD is of interest,
1294 * though that might be the case if a bitmap were made for
1295 * a mirrored SVD - worry about that later.
1296 * So we need to find the VD configuration record for the
1297 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1298 * The first 16 bytes of the sha1 of these is used.
1300 struct ddf_super
*ddf
= st
->sb
;
1301 struct vcl
*vcl
= ddf
->currentconf
;
1304 struct sha1_ctx ctx
;
1307 guid
= vcl
->conf
.guid
;
1309 guid
= ddf
->anchor
.guid
;
1311 sha1_init_ctx(&ctx
);
1312 sha1_process_bytes(guid
, DDF_GUID_LEN
, &ctx
);
1313 sha1_finish_ctx(&ctx
, buf
);
1314 memcpy(uuid
, buf
, 4*4);
1317 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
);
1319 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
)
1321 struct ddf_super
*ddf
= st
->sb
;
1323 if (ddf
->currentconf
) {
1324 getinfo_super_ddf_bvd(st
, info
);
1328 info
->array
.raid_disks
= __be16_to_cpu(ddf
->phys
->used_pdes
);
1329 info
->array
.level
= LEVEL_CONTAINER
;
1330 info
->array
.layout
= 0;
1331 info
->array
.md_minor
= -1;
1332 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*(__u32
*)
1333 (ddf
->anchor
.guid
+16));
1334 info
->array
.utime
= 0;
1335 info
->array
.chunk_size
= 0;
1338 info
->disk
.major
= 0;
1339 info
->disk
.minor
= 0;
1341 info
->disk
.number
= __be32_to_cpu(ddf
->dlist
->disk
.refnum
);
1342 info
->disk
.raid_disk
= find_phys(ddf
, ddf
->dlist
->disk
.refnum
);
1344 info
->data_offset
= __be64_to_cpu(ddf
->phys
->
1345 entries
[info
->disk
.raid_disk
].
1347 info
->component_size
= ddf
->dlist
->size
- info
->data_offset
;
1349 info
->disk
.number
= -1;
1350 info
->disk
.raid_disk
= -1;
1351 // info->disk.raid_disk = find refnum in the table and use index;
1353 info
->disk
.state
= (1 << MD_DISK_SYNC
);
1356 info
->reshape_active
= 0;
1359 info
->array
.major_version
= -1;
1360 info
->array
.minor_version
= -2;
1361 strcpy(info
->text_version
, "ddf");
1362 info
->safe_mode_delay
= 0;
1364 uuid_from_super_ddf(st
, info
->uuid
);
1368 static int rlq_to_layout(int rlq
, int prl
, int raiddisks
);
1370 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
)
1372 struct ddf_super
*ddf
= st
->sb
;
1373 struct vcl
*vc
= ddf
->currentconf
;
1374 int cd
= ddf
->currentdev
;
1378 /* FIXME this returns BVD info - what if we want SVD ?? */
1380 info
->array
.raid_disks
= __be16_to_cpu(vc
->conf
.prim_elmnt_count
);
1381 info
->array
.level
= map_num1(ddf_level_num
, vc
->conf
.prl
);
1382 info
->array
.layout
= rlq_to_layout(vc
->conf
.rlq
, vc
->conf
.prl
,
1383 info
->array
.raid_disks
);
1384 info
->array
.md_minor
= -1;
1385 info
->array
.ctime
= DECADE
+
1386 __be32_to_cpu(*(__u32
*)(vc
->conf
.guid
+16));
1387 info
->array
.utime
= DECADE
+ __be32_to_cpu(vc
->conf
.timestamp
);
1388 info
->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
1389 info
->custom_array_size
= 0;
1391 if (cd
>= 0 && cd
< ddf
->mppe
) {
1392 info
->data_offset
= __be64_to_cpu(vc
->lba_offset
[cd
]);
1393 if (vc
->block_sizes
)
1394 info
->component_size
= vc
->block_sizes
[cd
];
1396 info
->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
1399 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
1400 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1402 info
->disk
.major
= 0;
1403 info
->disk
.minor
= 0;
1405 info
->disk
.major
= dl
->major
;
1406 info
->disk
.minor
= dl
->minor
;
1408 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1409 // info->disk.raid_disk = find refnum in the table and use index;
1410 // info->disk.state = ???;
1412 info
->container_member
= ddf
->currentconf
->vcnum
;
1414 info
->resync_start
= 0;
1415 if (!(ddf
->virt
->entries
[info
->container_member
].state
1416 & DDF_state_inconsistent
) &&
1417 (ddf
->virt
->entries
[info
->container_member
].init_state
1418 & DDF_initstate_mask
)
1420 info
->resync_start
= ~0ULL;
1422 uuid_from_super_ddf(st
, info
->uuid
);
1424 info
->container_member
= atoi(st
->subarray
);
1425 info
->array
.major_version
= -1;
1426 info
->array
.minor_version
= -2;
1427 sprintf(info
->text_version
, "/%s/%s",
1428 devnum2devname(st
->container_dev
),
1430 info
->safe_mode_delay
= 200;
1432 memcpy(info
->name
, ddf
->virt
->entries
[info
->container_member
].name
, 16);
1435 if (info
->name
[j
] == ' ')
1440 static int update_super_ddf(struct supertype
*st
, struct mdinfo
*info
,
1442 char *devname
, int verbose
,
1443 int uuid_set
, char *homehost
)
1445 /* For 'assemble' and 'force' we need to return non-zero if any
1446 * change was made. For others, the return value is ignored.
1447 * Update options are:
1448 * force-one : This device looks a bit old but needs to be included,
1449 * update age info appropriately.
1450 * assemble: clear any 'faulty' flag to allow this device to
1452 * force-array: Array is degraded but being forced, mark it clean
1453 * if that will be needed to assemble it.
1455 * newdev: not used ????
1456 * grow: Array has gained a new device - this is currently for
1458 * resync: mark as dirty so a resync will happen.
1459 * uuid: Change the uuid of the array to match what is given
1460 * homehost: update the recorded homehost
1461 * name: update the name - preserving the homehost
1462 * _reshape_progress: record new reshape_progress position.
1464 * Following are not relevant for this version:
1465 * sparc2.2 : update from old dodgey metadata
1466 * super-minor: change the preferred_minor number
1467 * summaries: update redundant counters.
1470 // struct ddf_super *ddf = st->sb;
1471 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1472 // struct virtual_entry *ve = find_ve(ddf);
1474 /* we don't need to handle "force-*" or "assemble" as
1475 * there is no need to 'trick' the kernel. We the metadata is
1476 * first updated to activate the array, all the implied modifications
1480 if (strcmp(update
, "grow") == 0) {
1483 if (strcmp(update
, "resync") == 0) {
1484 // info->resync_checkpoint = 0;
1486 /* We ignore UUID updates as they make even less sense
1489 if (strcmp(update
, "homehost") == 0) {
1490 /* homehost is stored in controller->vendor_data,
1491 * or it is when we are the vendor
1493 // if (info->vendor_is_local)
1494 // strcpy(ddf->controller.vendor_data, homehost);
1496 if (strcmp(update
, "name") == 0) {
1497 /* name is stored in virtual_entry->name */
1498 // memset(ve->name, ' ', 16);
1499 // strncpy(ve->name, info->name, 16);
1501 if (strcmp(update
, "_reshape_progress") == 0) {
1502 /* We don't support reshape yet */
1505 // update_all_csum(ddf);
1510 __u32
random32(void)
1513 int rfd
= open("/dev/urandom", O_RDONLY
);
1514 if (rfd
< 0 || read(rfd
, &rv
, 4) != 4)
1521 static void make_header_guid(char *guid
)
1524 /* Create a DDF Header of Virtual Disk GUID */
1526 /* 24 bytes of fiction required.
1527 * first 8 are a 'vendor-id' - "Linux-MD"
1528 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1529 * Remaining 8 random number plus timestamp
1531 memcpy(guid
, T10
, sizeof(T10
));
1532 stamp
= __cpu_to_be32(0xdeadbeef);
1533 memcpy(guid
+8, &stamp
, 4);
1534 stamp
= __cpu_to_be32(0);
1535 memcpy(guid
+12, &stamp
, 4);
1536 stamp
= __cpu_to_be32(time(0) - DECADE
);
1537 memcpy(guid
+16, &stamp
, 4);
1539 memcpy(guid
+20, &stamp
, 4);
1542 static int init_super_ddf_bvd(struct supertype
*st
,
1543 mdu_array_info_t
*info
,
1544 unsigned long long size
,
1545 char *name
, char *homehost
,
1548 static int init_super_ddf(struct supertype
*st
,
1549 mdu_array_info_t
*info
,
1550 unsigned long long size
, char *name
, char *homehost
,
1553 /* This is primarily called by Create when creating a new array.
1554 * We will then get add_to_super called for each component, and then
1555 * write_init_super called to write it out to each device.
1556 * For DDF, Create can create on fresh devices or on a pre-existing
1558 * To create on a pre-existing array a different method will be called.
1559 * This one is just for fresh drives.
1561 * We need to create the entire 'ddf' structure which includes:
1562 * DDF headers - these are easy.
1563 * Controller data - a Sector describing this controller .. not that
1564 * this is a controller exactly.
1565 * Physical Disk Record - one entry per device, so
1566 * leave plenty of space.
1567 * Virtual Disk Records - again, just leave plenty of space.
1568 * This just lists VDs, doesn't give details
1569 * Config records - describes the VDs that use this disk
1570 * DiskData - describes 'this' device.
1571 * BadBlockManagement - empty
1572 * Diag Space - empty
1573 * Vendor Logs - Could we put bitmaps here?
1576 struct ddf_super
*ddf
;
1579 int max_phys_disks
, max_virt_disks
;
1580 unsigned long long sector
;
1584 struct phys_disk
*pd
;
1585 struct virtual_disk
*vd
;
1592 return init_super_ddf_bvd(st
, info
, size
, name
, homehost
,
1595 if (posix_memalign((void**)&ddf
, 512, sizeof(*ddf
)) != 0) {
1596 fprintf(stderr
, Name
": %s could not allocate superblock\n", __func__
);
1599 memset(ddf
, 0, sizeof(*ddf
));
1600 ddf
->dlist
= NULL
; /* no physical disks yet */
1601 ddf
->conflist
= NULL
; /* No virtual disks yet */
1603 /* At least 32MB *must* be reserved for the ddf. So let's just
1604 * start 32MB from the end, and put the primary header there.
1605 * Don't do secondary for now.
1606 * We don't know exactly where that will be yet as it could be
1607 * different on each device. To just set up the lengths.
1611 ddf
->anchor
.magic
= DDF_HEADER_MAGIC
;
1612 make_header_guid(ddf
->anchor
.guid
);
1614 memcpy(ddf
->anchor
.revision
, DDF_REVISION_2
, 8);
1615 ddf
->anchor
.seq
= __cpu_to_be32(1);
1616 ddf
->anchor
.timestamp
= __cpu_to_be32(time(0) - DECADE
);
1617 ddf
->anchor
.openflag
= 0xFF;
1618 ddf
->anchor
.foreignflag
= 0;
1619 ddf
->anchor
.enforcegroups
= 0; /* Is this best?? */
1620 ddf
->anchor
.pad0
= 0xff;
1621 memset(ddf
->anchor
.pad1
, 0xff, 12);
1622 memset(ddf
->anchor
.header_ext
, 0xff, 32);
1623 ddf
->anchor
.primary_lba
= ~(__u64
)0;
1624 ddf
->anchor
.secondary_lba
= ~(__u64
)0;
1625 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
1626 memset(ddf
->anchor
.pad2
, 0xff, 3);
1627 ddf
->anchor
.workspace_len
= __cpu_to_be32(32768); /* Must be reserved */
1628 ddf
->anchor
.workspace_lba
= ~(__u64
)0; /* Put this at bottom
1629 of 32M reserved.. */
1630 max_phys_disks
= 1023; /* Should be enough */
1631 ddf
->anchor
.max_pd_entries
= __cpu_to_be16(max_phys_disks
);
1632 max_virt_disks
= 255;
1633 ddf
->anchor
.max_vd_entries
= __cpu_to_be16(max_virt_disks
); /* ?? */
1634 ddf
->anchor
.max_partitions
= __cpu_to_be16(64); /* ?? */
1637 ddf
->conf_rec_len
= 1 + ROUND_UP(ddf
->mppe
* (4+8), 512)/512;
1638 ddf
->anchor
.config_record_len
= __cpu_to_be16(ddf
->conf_rec_len
);
1639 ddf
->anchor
.max_primary_element_entries
= __cpu_to_be16(ddf
->mppe
);
1640 memset(ddf
->anchor
.pad3
, 0xff, 54);
1641 /* controller sections is one sector long immediately
1642 * after the ddf header */
1644 ddf
->anchor
.controller_section_offset
= __cpu_to_be32(sector
);
1645 ddf
->anchor
.controller_section_length
= __cpu_to_be32(1);
1648 /* phys is 8 sectors after that */
1649 pdsize
= ROUND_UP(sizeof(struct phys_disk
) +
1650 sizeof(struct phys_disk_entry
)*max_phys_disks
,
1652 switch(pdsize
/512) {
1653 case 2: case 8: case 32: case 128: case 512: break;
1656 ddf
->anchor
.phys_section_offset
= __cpu_to_be32(sector
);
1657 ddf
->anchor
.phys_section_length
=
1658 __cpu_to_be32(pdsize
/512); /* max_primary_element_entries/8 */
1659 sector
+= pdsize
/512;
1661 /* virt is another 32 sectors */
1662 vdsize
= ROUND_UP(sizeof(struct virtual_disk
) +
1663 sizeof(struct virtual_entry
) * max_virt_disks
,
1665 switch(vdsize
/512) {
1666 case 2: case 8: case 32: case 128: case 512: break;
1669 ddf
->anchor
.virt_section_offset
= __cpu_to_be32(sector
);
1670 ddf
->anchor
.virt_section_length
=
1671 __cpu_to_be32(vdsize
/512); /* max_vd_entries/8 */
1672 sector
+= vdsize
/512;
1674 clen
= ddf
->conf_rec_len
* (ddf
->max_part
+1);
1675 ddf
->anchor
.config_section_offset
= __cpu_to_be32(sector
);
1676 ddf
->anchor
.config_section_length
= __cpu_to_be32(clen
);
1679 ddf
->anchor
.data_section_offset
= __cpu_to_be32(sector
);
1680 ddf
->anchor
.data_section_length
= __cpu_to_be32(1);
1683 ddf
->anchor
.bbm_section_length
= __cpu_to_be32(0);
1684 ddf
->anchor
.bbm_section_offset
= __cpu_to_be32(0xFFFFFFFF);
1685 ddf
->anchor
.diag_space_length
= __cpu_to_be32(0);
1686 ddf
->anchor
.diag_space_offset
= __cpu_to_be32(0xFFFFFFFF);
1687 ddf
->anchor
.vendor_length
= __cpu_to_be32(0);
1688 ddf
->anchor
.vendor_offset
= __cpu_to_be32(0xFFFFFFFF);
1690 memset(ddf
->anchor
.pad4
, 0xff, 256);
1692 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
1693 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
1695 ddf
->primary
.openflag
= 1; /* I guess.. */
1696 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
1698 ddf
->secondary
.openflag
= 1; /* I guess.. */
1699 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
1701 ddf
->active
= &ddf
->primary
;
1703 ddf
->controller
.magic
= DDF_CONTROLLER_MAGIC
;
1705 /* 24 more bytes of fiction required.
1706 * first 8 are a 'vendor-id' - "Linux-MD"
1707 * Remaining 16 are serial number.... maybe a hostname would do?
1709 memcpy(ddf
->controller
.guid
, T10
, sizeof(T10
));
1710 gethostname(hostname
, sizeof(hostname
));
1711 hostname
[sizeof(hostname
) - 1] = 0;
1712 hostlen
= strlen(hostname
);
1713 memcpy(ddf
->controller
.guid
+ 24 - hostlen
, hostname
, hostlen
);
1714 for (i
= strlen(T10
) ; i
+hostlen
< 24; i
++)
1715 ddf
->controller
.guid
[i
] = ' ';
1717 ddf
->controller
.type
.vendor_id
= __cpu_to_be16(0xDEAD);
1718 ddf
->controller
.type
.device_id
= __cpu_to_be16(0xBEEF);
1719 ddf
->controller
.type
.sub_vendor_id
= 0;
1720 ddf
->controller
.type
.sub_device_id
= 0;
1721 memcpy(ddf
->controller
.product_id
, "What Is My PID??", 16);
1722 memset(ddf
->controller
.pad
, 0xff, 8);
1723 memset(ddf
->controller
.vendor_data
, 0xff, 448);
1724 if (homehost
&& strlen(homehost
) < 440)
1725 strcpy((char*)ddf
->controller
.vendor_data
, homehost
);
1727 if (posix_memalign((void**)&pd
, 512, pdsize
) != 0) {
1728 fprintf(stderr
, Name
": %s could not allocate pd\n", __func__
);
1732 ddf
->pdsize
= pdsize
;
1734 memset(pd
, 0xff, pdsize
);
1735 memset(pd
, 0, sizeof(*pd
));
1736 pd
->magic
= DDF_PHYS_DATA_MAGIC
;
1737 pd
->used_pdes
= __cpu_to_be16(0);
1738 pd
->max_pdes
= __cpu_to_be16(max_phys_disks
);
1739 memset(pd
->pad
, 0xff, 52);
1741 if (posix_memalign((void**)&vd
, 512, vdsize
) != 0) {
1742 fprintf(stderr
, Name
": %s could not allocate vd\n", __func__
);
1746 ddf
->vdsize
= vdsize
;
1747 memset(vd
, 0, vdsize
);
1748 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
1749 vd
->populated_vdes
= __cpu_to_be16(0);
1750 vd
->max_vdes
= __cpu_to_be16(max_virt_disks
);
1751 memset(vd
->pad
, 0xff, 52);
1753 for (i
=0; i
<max_virt_disks
; i
++)
1754 memset(&vd
->entries
[i
], 0xff, sizeof(struct virtual_entry
));
1757 ddf
->updates_pending
= 1;
1761 static int chunk_to_shift(int chunksize
)
1763 return ffs(chunksize
/512)-1;
1766 static int level_to_prl(int level
)
1769 case LEVEL_LINEAR
: return DDF_CONCAT
;
1770 case 0: return DDF_RAID0
;
1771 case 1: return DDF_RAID1
;
1772 case 4: return DDF_RAID4
;
1773 case 5: return DDF_RAID5
;
1774 case 6: return DDF_RAID6
;
1778 static int layout_to_rlq(int level
, int layout
, int raiddisks
)
1782 return DDF_RAID0_SIMPLE
;
1785 case 2: return DDF_RAID1_SIMPLE
;
1786 case 3: return DDF_RAID1_MULTI
;
1791 case 0: return DDF_RAID4_N
;
1796 case ALGORITHM_LEFT_ASYMMETRIC
:
1797 return DDF_RAID5_N_RESTART
;
1798 case ALGORITHM_RIGHT_ASYMMETRIC
:
1799 return DDF_RAID5_0_RESTART
;
1800 case ALGORITHM_LEFT_SYMMETRIC
:
1801 return DDF_RAID5_N_CONTINUE
;
1802 case ALGORITHM_RIGHT_SYMMETRIC
:
1803 return -1; /* not mentioned in standard */
1807 case ALGORITHM_ROTATING_N_RESTART
:
1808 return DDF_RAID5_N_RESTART
;
1809 case ALGORITHM_ROTATING_ZERO_RESTART
:
1810 return DDF_RAID6_0_RESTART
;
1811 case ALGORITHM_ROTATING_N_CONTINUE
:
1812 return DDF_RAID5_N_CONTINUE
;
1818 static int rlq_to_layout(int rlq
, int prl
, int raiddisks
)
1822 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1824 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1832 return -1; /* FIXME this isn't checked */
1836 case DDF_RAID5_N_RESTART
:
1837 return ALGORITHM_LEFT_ASYMMETRIC
;
1838 case DDF_RAID5_0_RESTART
:
1839 return ALGORITHM_RIGHT_ASYMMETRIC
;
1840 case DDF_RAID5_N_CONTINUE
:
1841 return ALGORITHM_LEFT_SYMMETRIC
;
1847 case DDF_RAID5_N_RESTART
:
1848 return ALGORITHM_ROTATING_N_RESTART
;
1849 case DDF_RAID6_0_RESTART
:
1850 return ALGORITHM_ROTATING_ZERO_RESTART
;
1851 case DDF_RAID5_N_CONTINUE
:
1852 return ALGORITHM_ROTATING_N_CONTINUE
;
1862 unsigned long long start
, size
;
1864 static int cmp_extent(const void *av
, const void *bv
)
1866 const struct extent
*a
= av
;
1867 const struct extent
*b
= bv
;
1868 if (a
->start
< b
->start
)
1870 if (a
->start
> b
->start
)
1875 static struct extent
*get_extents(struct ddf_super
*ddf
, struct dl
*dl
)
1877 /* find a list of used extents on the give physical device
1878 * (dnum) of the given ddf.
1879 * Return a malloced array of 'struct extent'
1881 FIXME ignore DDF_Legacy devices?
1888 rv
= malloc(sizeof(struct extent
) * (ddf
->max_part
+ 2));
1892 for (i
= 0; i
< ddf
->max_part
; i
++) {
1893 struct vcl
*v
= dl
->vlist
[i
];
1896 for (j
=0; j
< v
->conf
.prim_elmnt_count
; j
++)
1897 if (v
->conf
.phys_refnum
[j
] == dl
->disk
.refnum
) {
1898 /* This device plays role 'j' in 'v'. */
1899 rv
[n
].start
= __be64_to_cpu(v
->lba_offset
[j
]);
1900 rv
[n
].size
= __be64_to_cpu(v
->conf
.blocks
);
1905 qsort(rv
, n
, sizeof(*rv
), cmp_extent
);
1907 rv
[n
].start
= __be64_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].config_size
);
1913 static int init_super_ddf_bvd(struct supertype
*st
,
1914 mdu_array_info_t
*info
,
1915 unsigned long long size
,
1916 char *name
, char *homehost
,
1919 /* We are creating a BVD inside a pre-existing container.
1920 * so st->sb is already set.
1921 * We need to create a new vd_config and a new virtual_entry
1923 struct ddf_super
*ddf
= st
->sb
;
1925 struct virtual_entry
*ve
;
1927 struct vd_config
*vc
;
1929 if (__be16_to_cpu(ddf
->virt
->populated_vdes
)
1930 >= __be16_to_cpu(ddf
->virt
->max_vdes
)) {
1931 fprintf(stderr
, Name
": This ddf already has the "
1932 "maximum of %d virtual devices\n",
1933 __be16_to_cpu(ddf
->virt
->max_vdes
));
1937 for (venum
= 0; venum
< __be16_to_cpu(ddf
->virt
->max_vdes
); venum
++)
1938 if (all_ff(ddf
->virt
->entries
[venum
].guid
))
1940 if (venum
== __be16_to_cpu(ddf
->virt
->max_vdes
)) {
1941 fprintf(stderr
, Name
": Cannot find spare slot for "
1942 "virtual disk - DDF is corrupt\n");
1945 ve
= &ddf
->virt
->entries
[venum
];
1947 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1948 * timestamp, random number
1950 make_header_guid(ve
->guid
);
1951 ve
->unit
= __cpu_to_be16(info
->md_minor
);
1953 ve
->guid_crc
= crc32(0, (unsigned char*)ddf
->anchor
.guid
, DDF_GUID_LEN
);
1955 ve
->state
= DDF_state_degraded
; /* Will be modified as devices are added */
1956 if (info
->state
& 1) /* clean */
1957 ve
->init_state
= DDF_init_full
;
1959 ve
->init_state
= DDF_init_not
;
1961 memset(ve
->pad1
, 0xff, 14);
1962 memset(ve
->name
, ' ', 16);
1964 strncpy(ve
->name
, name
, 16);
1965 ddf
->virt
->populated_vdes
=
1966 __cpu_to_be16(__be16_to_cpu(ddf
->virt
->populated_vdes
)+1);
1968 /* Now create a new vd_config */
1969 if (posix_memalign((void**)&vcl
, 512,
1970 (offsetof(struct vcl
, conf
) + ddf
->conf_rec_len
* 512)) != 0) {
1971 fprintf(stderr
, Name
": %s could not allocate vd_config\n", __func__
);
1974 vcl
->lba_offset
= (__u64
*) &vcl
->conf
.phys_refnum
[ddf
->mppe
];
1976 sprintf(st
->subarray
, "%d", venum
);
1977 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
1981 vc
->magic
= DDF_VD_CONF_MAGIC
;
1982 memcpy(vc
->guid
, ve
->guid
, DDF_GUID_LEN
);
1983 vc
->timestamp
= __cpu_to_be32(time(0)-DECADE
);
1984 vc
->seqnum
= __cpu_to_be32(1);
1985 memset(vc
->pad0
, 0xff, 24);
1986 vc
->prim_elmnt_count
= __cpu_to_be16(info
->raid_disks
);
1987 vc
->chunk_shift
= chunk_to_shift(info
->chunk_size
);
1988 vc
->prl
= level_to_prl(info
->level
);
1989 vc
->rlq
= layout_to_rlq(info
->level
, info
->layout
, info
->raid_disks
);
1990 vc
->sec_elmnt_count
= 1;
1991 vc
->sec_elmnt_seq
= 0;
1993 vc
->blocks
= __cpu_to_be64(info
->size
* 2);
1994 vc
->array_blocks
= __cpu_to_be64(
1995 calc_array_size(info
->level
, info
->raid_disks
, info
->layout
,
1996 info
->chunk_size
, info
->size
*2));
1997 memset(vc
->pad1
, 0xff, 8);
1998 vc
->spare_refs
[0] = 0xffffffff;
1999 vc
->spare_refs
[1] = 0xffffffff;
2000 vc
->spare_refs
[2] = 0xffffffff;
2001 vc
->spare_refs
[3] = 0xffffffff;
2002 vc
->spare_refs
[4] = 0xffffffff;
2003 vc
->spare_refs
[5] = 0xffffffff;
2004 vc
->spare_refs
[6] = 0xffffffff;
2005 vc
->spare_refs
[7] = 0xffffffff;
2006 memset(vc
->cache_pol
, 0, 8);
2008 memset(vc
->pad2
, 0xff, 3);
2009 memset(vc
->pad3
, 0xff, 52);
2010 memset(vc
->pad4
, 0xff, 192);
2011 memset(vc
->v0
, 0xff, 32);
2012 memset(vc
->v1
, 0xff, 32);
2013 memset(vc
->v2
, 0xff, 16);
2014 memset(vc
->v3
, 0xff, 16);
2015 memset(vc
->vendor
, 0xff, 32);
2017 memset(vc
->phys_refnum
, 0xff, 4*ddf
->mppe
);
2018 memset(vc
->phys_refnum
+ddf
->mppe
, 0x00, 8*ddf
->mppe
);
2020 vcl
->next
= ddf
->conflist
;
2021 ddf
->conflist
= vcl
;
2022 ddf
->currentconf
= vcl
;
2023 ddf
->updates_pending
= 1;
2028 static void add_to_super_ddf_bvd(struct supertype
*st
,
2029 mdu_disk_info_t
*dk
, int fd
, char *devname
)
2031 /* fd and devname identify a device with-in the ddf container (st).
2032 * dk identifies a location in the new BVD.
2033 * We need to find suitable free space in that device and update
2034 * the phys_refnum and lba_offset for the newly created vd_config.
2035 * We might also want to update the type in the phys_disk
2038 * Alternately: fd == -1 and we have already chosen which device to
2039 * use and recorded in dlist->raid_disk;
2042 struct ddf_super
*ddf
= st
->sb
;
2043 struct vd_config
*vc
;
2047 unsigned long long blocks
, pos
, esize
;
2051 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2052 if (dl
->raiddisk
== dk
->raid_disk
)
2055 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2056 if (dl
->major
== dk
->major
&&
2057 dl
->minor
== dk
->minor
)
2060 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
2063 vc
= &ddf
->currentconf
->conf
;
2064 lba_offset
= ddf
->currentconf
->lba_offset
;
2066 ex
= get_extents(ddf
, dl
);
2071 blocks
= __be64_to_cpu(vc
->blocks
);
2072 if (ddf
->currentconf
->block_sizes
)
2073 blocks
= ddf
->currentconf
->block_sizes
[dk
->raid_disk
];
2076 esize
= ex
[i
].start
- pos
;
2077 if (esize
>= blocks
)
2079 pos
= ex
[i
].start
+ ex
[i
].size
;
2081 } while (ex
[i
-1].size
);
2087 ddf
->currentdev
= dk
->raid_disk
;
2088 vc
->phys_refnum
[dk
->raid_disk
] = dl
->disk
.refnum
;
2089 lba_offset
[dk
->raid_disk
] = __cpu_to_be64(pos
);
2091 for (i
=0; i
< ddf
->max_part
; i
++)
2092 if (dl
->vlist
[i
] == NULL
)
2094 if (i
== ddf
->max_part
)
2096 dl
->vlist
[i
] = ddf
->currentconf
;
2101 dl
->devname
= devname
;
2103 /* Check how many working raid_disks, and if we can mark
2104 * array as optimal yet
2108 for (i
=0; i
< __be16_to_cpu(vc
->prim_elmnt_count
); i
++)
2109 if (vc
->phys_refnum
[i
] != 0xffffffff)
2112 /* Find which virtual_entry */
2113 i
= ddf
->currentconf
->vcnum
;
2114 if (working
== __be16_to_cpu(vc
->prim_elmnt_count
))
2115 ddf
->virt
->entries
[i
].state
=
2116 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
2117 | DDF_state_optimal
;
2119 if (vc
->prl
== DDF_RAID6
&&
2120 working
+1 == __be16_to_cpu(vc
->prim_elmnt_count
))
2121 ddf
->virt
->entries
[i
].state
=
2122 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
2123 | DDF_state_part_optimal
;
2125 ddf
->phys
->entries
[dl
->pdnum
].type
&= ~__cpu_to_be16(DDF_Global_Spare
);
2126 ddf
->phys
->entries
[dl
->pdnum
].type
|= __cpu_to_be16(DDF_Active_in_VD
);
2127 ddf
->updates_pending
= 1;
2130 /* add a device to a container, either while creating it or while
2131 * expanding a pre-existing container
2133 static int add_to_super_ddf(struct supertype
*st
,
2134 mdu_disk_info_t
*dk
, int fd
, char *devname
)
2136 struct ddf_super
*ddf
= st
->sb
;
2140 unsigned long long size
;
2141 struct phys_disk_entry
*pde
;
2145 if (ddf
->currentconf
) {
2146 add_to_super_ddf_bvd(st
, dk
, fd
, devname
);
2150 /* This is device numbered dk->number. We need to create
2151 * a phys_disk entry and a more detailed disk_data entry.
2154 if (posix_memalign((void**)&dd
, 512,
2155 sizeof(*dd
) + sizeof(dd
->vlist
[0]) * ddf
->max_part
) != 0) {
2156 fprintf(stderr
, Name
2157 ": %s could allocate buffer for new disk, aborting\n",
2161 dd
->major
= major(stb
.st_rdev
);
2162 dd
->minor
= minor(stb
.st_rdev
);
2163 dd
->devname
= devname
;
2167 dd
->disk
.magic
= DDF_PHYS_DATA_MAGIC
;
2169 tm
= localtime(&now
);
2170 sprintf(dd
->disk
.guid
, "%8s%04d%02d%02d",
2171 T10
, tm
->tm_year
+1900, tm
->tm_mon
+1, tm
->tm_mday
);
2172 *(__u32
*)(dd
->disk
.guid
+ 16) = random32();
2173 *(__u32
*)(dd
->disk
.guid
+ 20) = random32();
2176 /* Cannot be bothered finding a CRC of some irrelevant details*/
2177 dd
->disk
.refnum
= random32();
2178 for (i
= __be16_to_cpu(ddf
->active
->max_pd_entries
) - 1;
2180 if (ddf
->phys
->entries
[i
].refnum
== dd
->disk
.refnum
)
2184 dd
->disk
.forced_ref
= 1;
2185 dd
->disk
.forced_guid
= 1;
2186 memset(dd
->disk
.vendor
, ' ', 32);
2187 memcpy(dd
->disk
.vendor
, "Linux", 5);
2188 memset(dd
->disk
.pad
, 0xff, 442);
2189 for (i
= 0; i
< ddf
->max_part
; i
++)
2190 dd
->vlist
[i
] = NULL
;
2192 n
= __be16_to_cpu(ddf
->phys
->used_pdes
);
2193 pde
= &ddf
->phys
->entries
[n
];
2196 if (st
->update_tail
) {
2197 int len
= (sizeof(struct phys_disk
) +
2198 sizeof(struct phys_disk_entry
));
2199 struct phys_disk
*pd
;
2202 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2203 pd
->used_pdes
= __cpu_to_be16(n
);
2204 pde
= &pd
->entries
[0];
2208 ddf
->phys
->used_pdes
= __cpu_to_be16(n
);
2211 memcpy(pde
->guid
, dd
->disk
.guid
, DDF_GUID_LEN
);
2212 pde
->refnum
= dd
->disk
.refnum
;
2213 pde
->type
= __cpu_to_be16(DDF_Forced_PD_GUID
| DDF_Global_Spare
);
2214 pde
->state
= __cpu_to_be16(DDF_Online
);
2215 get_dev_size(fd
, NULL
, &size
);
2216 /* We are required to reserve 32Meg, and record the size in sectors */
2217 pde
->config_size
= __cpu_to_be64( (size
- 32*1024*1024) / 512);
2218 sprintf(pde
->path
, "%17.17s","Information: nil") ;
2219 memset(pde
->pad
, 0xff, 6);
2221 dd
->size
= size
>> 9;
2222 if (st
->update_tail
) {
2223 dd
->next
= ddf
->add_list
;
2226 dd
->next
= ddf
->dlist
;
2228 ddf
->updates_pending
= 1;
2235 * This is the write_init_super method for a ddf container. It is
2236 * called when creating a container or adding another device to a
2240 static unsigned char null_conf
[4096+512];
2242 static int __write_init_super_ddf(struct supertype
*st
, int do_close
)
2245 struct ddf_super
*ddf
= st
->sb
;
2252 unsigned long long size
, sector
;
2254 /* try to write updated metadata,
2255 * if we catch a failure move on to the next disk
2257 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
2264 /* We need to fill in the primary, (secondary) and workspace
2265 * lba's in the headers, set their checksums,
2266 * Also checksum phys, virt....
2268 * Then write everything out, finally the anchor is written.
2270 get_dev_size(fd
, NULL
, &size
);
2272 ddf
->anchor
.workspace_lba
= __cpu_to_be64(size
- 32*1024*2);
2273 ddf
->anchor
.primary_lba
= __cpu_to_be64(size
- 16*1024*2);
2274 ddf
->anchor
.seq
= __cpu_to_be32(1);
2275 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2276 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2278 ddf
->anchor
.openflag
= 0xFF; /* 'open' means nothing */
2279 ddf
->anchor
.seq
= 0xFFFFFFFF; /* no sequencing in anchor */
2280 ddf
->anchor
.crc
= calc_crc(&ddf
->anchor
, 512);
2282 ddf
->primary
.openflag
= 0;
2283 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
2285 ddf
->secondary
.openflag
= 0;
2286 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
2288 ddf
->primary
.crc
= calc_crc(&ddf
->primary
, 512);
2289 ddf
->secondary
.crc
= calc_crc(&ddf
->secondary
, 512);
2291 sector
= size
- 16*1024*2;
2292 lseek64(fd
, sector
<<9, 0);
2293 if (write(fd
, &ddf
->primary
, 512) < 0)
2296 ddf
->controller
.crc
= calc_crc(&ddf
->controller
, 512);
2297 if (write(fd
, &ddf
->controller
, 512) < 0)
2300 ddf
->phys
->crc
= calc_crc(ddf
->phys
, ddf
->pdsize
);
2302 if (write(fd
, ddf
->phys
, ddf
->pdsize
) < 0)
2305 ddf
->virt
->crc
= calc_crc(ddf
->virt
, ddf
->vdsize
);
2306 if (write(fd
, ddf
->virt
, ddf
->vdsize
) < 0)
2309 /* Now write lots of config records. */
2310 n_config
= ddf
->max_part
;
2311 conf_size
= ddf
->conf_rec_len
* 512;
2312 for (i
= 0 ; i
<= n_config
; i
++) {
2313 struct vcl
*c
= d
->vlist
[i
];
2315 c
= (struct vcl
*)d
->spare
;
2318 c
->conf
.crc
= calc_crc(&c
->conf
, conf_size
);
2319 if (write(fd
, &c
->conf
, conf_size
) < 0)
2322 char *null_aligned
= (char*)((((unsigned long)null_conf
)+511)&~511UL);
2323 if (null_conf
[0] != 0xff)
2324 memset(null_conf
, 0xff, sizeof(null_conf
));
2325 int togo
= conf_size
;
2326 while (togo
> sizeof(null_conf
)-512) {
2327 if (write(fd
, null_aligned
, sizeof(null_conf
)-512) < 0)
2329 togo
-= sizeof(null_conf
)-512;
2331 if (write(fd
, null_aligned
, togo
) < 0)
2337 d
->disk
.crc
= calc_crc(&d
->disk
, 512);
2338 if (write(fd
, &d
->disk
, 512) < 0)
2341 /* Maybe do the same for secondary */
2343 lseek64(fd
, (size
-1)*512, SEEK_SET
);
2344 if (write(fd
, &ddf
->anchor
, 512) < 0)
2350 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
2355 return attempts
!= successes
;
2358 static int write_init_super_ddf(struct supertype
*st
)
2361 if (st
->update_tail
) {
2362 /* queue the virtual_disk and vd_config as metadata updates */
2363 struct virtual_disk
*vd
;
2364 struct vd_config
*vc
;
2365 struct ddf_super
*ddf
= st
->sb
;
2368 if (!ddf
->currentconf
) {
2369 int len
= (sizeof(struct phys_disk
) +
2370 sizeof(struct phys_disk_entry
));
2372 /* adding a disk to the container. */
2376 append_metadata_update(st
, ddf
->add_list
->mdupdate
, len
);
2377 ddf
->add_list
->mdupdate
= NULL
;
2381 /* Newly created VD */
2383 /* First the virtual disk. We have a slightly fake header */
2384 len
= sizeof(struct virtual_disk
) + sizeof(struct virtual_entry
);
2387 vd
->entries
[0] = ddf
->virt
->entries
[ddf
->currentconf
->vcnum
];
2388 vd
->populated_vdes
= __cpu_to_be16(ddf
->currentconf
->vcnum
);
2389 append_metadata_update(st
, vd
, len
);
2391 /* Then the vd_config */
2392 len
= ddf
->conf_rec_len
* 512;
2394 memcpy(vc
, &ddf
->currentconf
->conf
, len
);
2395 append_metadata_update(st
, vc
, len
);
2397 /* FIXME I need to close the fds! */
2400 return __write_init_super_ddf(st
, 1);
2405 static __u64
avail_size_ddf(struct supertype
*st
, __u64 devsize
)
2407 /* We must reserve the last 32Meg */
2408 if (devsize
<= 32*1024*2)
2410 return devsize
- 32*1024*2;
2415 static int reserve_space(struct supertype
*st
, int raiddisks
,
2416 unsigned long long size
, int chunk
,
2417 unsigned long long *freesize
)
2419 /* Find 'raiddisks' spare extents at least 'size' big (but
2420 * only caring about multiples of 'chunk') and remember
2422 * If the cannot be found, fail.
2425 struct ddf_super
*ddf
= st
->sb
;
2428 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
2432 /* Now find largest extent on each device */
2433 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
2434 struct extent
*e
= get_extents(ddf
, dl
);
2435 unsigned long long pos
= 0;
2438 unsigned long long minsize
= size
;
2446 unsigned long long esize
;
2447 esize
= e
[i
].start
- pos
;
2448 if (esize
>= minsize
) {
2452 pos
= e
[i
].start
+ e
[i
].size
;
2454 } while (e
[i
-1].size
);
2457 dl
->esize
= minsize
;
2461 if (cnt
< raiddisks
) {
2462 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
2463 return 0; /* No enough free spaces large enough */
2466 /* choose the largest size of which there are at least 'raiddisk' */
2467 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
2469 if (dl
->esize
<= size
)
2471 /* This is bigger than 'size', see if there are enough */
2473 for (dl2
= dl
; dl2
; dl2
=dl2
->next
)
2474 if (dl2
->esize
>= dl
->esize
)
2476 if (cnt
>= raiddisks
)
2480 size
= size
/ chunk
;
2485 fprintf(stderr
, Name
": not enough spare devices to create array.\n");
2489 /* We have a 'size' of which there are enough spaces.
2490 * We simply do a first-fit */
2492 for (dl
= ddf
->dlist
; dl
&& cnt
< raiddisks
; dl
=dl
->next
) {
2493 if (dl
->esize
< size
)
2505 validate_geometry_ddf_container(struct supertype
*st
,
2506 int level
, int layout
, int raiddisks
,
2507 int chunk
, unsigned long long size
,
2508 char *dev
, unsigned long long *freesize
,
2511 static int validate_geometry_ddf_bvd(struct supertype
*st
,
2512 int level
, int layout
, int raiddisks
,
2513 int chunk
, unsigned long long size
,
2514 char *dev
, unsigned long long *freesize
,
2517 static int validate_geometry_ddf(struct supertype
*st
,
2518 int level
, int layout
, int raiddisks
,
2519 int chunk
, unsigned long long size
,
2520 char *dev
, unsigned long long *freesize
,
2527 /* ddf potentially supports lots of things, but it depends on
2528 * what devices are offered (and maybe kernel version?)
2529 * If given unused devices, we will make a container.
2530 * If given devices in a container, we will make a BVD.
2531 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2534 if (level
== LEVEL_CONTAINER
) {
2535 /* Must be a fresh device to add to a container */
2536 return validate_geometry_ddf_container(st
, level
, layout
,
2538 size
, dev
, freesize
,
2543 /* Initial sanity check. Exclude illegal levels. */
2545 for (i
=0; ddf_level_num
[i
].num1
!= MAXINT
; i
++)
2546 if (ddf_level_num
[i
].num2
== level
)
2548 if (ddf_level_num
[i
].num1
== MAXINT
)
2550 /* Should check layout? etc */
2552 if (st
->sb
&& freesize
) {
2553 /* --create was given a container to create in.
2554 * So we need to check that there are enough
2555 * free spaces and return the amount of space.
2556 * We may as well remember which drives were
2557 * chosen so that add_to_super/getinfo_super
2560 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
2566 /* A container has already been opened, so we are
2567 * creating in there. Maybe a BVD, maybe an SVD.
2568 * Should make a distinction one day.
2570 return validate_geometry_ddf_bvd(st
, level
, layout
, raiddisks
,
2571 chunk
, size
, dev
, freesize
,
2574 /* This is the first device for the array.
2575 * If it is a container, we read it in and do automagic allocations,
2576 * no other devices should be given.
2577 * Otherwise it must be a member device of a container, and we
2578 * do manual allocation.
2579 * Later we should check for a BVD and make an SVD.
2581 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2583 sra
= sysfs_read(fd
, 0, GET_VERSION
);
2585 if (sra
&& sra
->array
.major_version
== -1 &&
2586 strcmp(sra
->text_version
, "ddf") == 0) {
2589 /* find space for 'n' devices. */
2590 /* remember the devices */
2591 /* Somehow return the fact that we have enough */
2596 Name
": ddf: Cannot create this array "
2601 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2603 fprintf(stderr
, Name
": ddf: Cannot open %s: %s\n",
2604 dev
, strerror(errno
));
2607 /* Well, it is in use by someone, maybe a 'ddf' container. */
2608 cfd
= open_container(fd
);
2612 fprintf(stderr
, Name
": ddf: Cannot use %s: %s\n",
2613 dev
, strerror(EBUSY
));
2616 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2618 if (sra
&& sra
->array
.major_version
== -1 &&
2619 strcmp(sra
->text_version
, "ddf") == 0) {
2620 /* This is a member of a ddf container. Load the container
2621 * and try to create a bvd
2623 struct ddf_super
*ddf
;
2624 if (load_super_ddf_all(st
, cfd
, (void **)&ddf
, NULL
, 1) == 0) {
2626 st
->container_dev
= fd2devnum(cfd
);
2628 return validate_geometry_ddf_bvd(st
, level
, layout
,
2629 raiddisks
, chunk
, size
,
2634 } else /* device may belong to a different container */
2641 validate_geometry_ddf_container(struct supertype
*st
,
2642 int level
, int layout
, int raiddisks
,
2643 int chunk
, unsigned long long size
,
2644 char *dev
, unsigned long long *freesize
,
2648 unsigned long long ldsize
;
2650 if (level
!= LEVEL_CONTAINER
)
2655 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2658 fprintf(stderr
, Name
": ddf: Cannot open %s: %s\n",
2659 dev
, strerror(errno
));
2662 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2668 *freesize
= avail_size_ddf(st
, ldsize
>> 9);
2675 static int validate_geometry_ddf_bvd(struct supertype
*st
,
2676 int level
, int layout
, int raiddisks
,
2677 int chunk
, unsigned long long size
,
2678 char *dev
, unsigned long long *freesize
,
2682 struct ddf_super
*ddf
= st
->sb
;
2684 unsigned long long pos
= 0;
2685 unsigned long long maxsize
;
2688 /* ddf/bvd supports lots of things, but not containers */
2689 if (level
== LEVEL_CONTAINER
)
2691 /* We must have the container info already read in. */
2696 /* General test: make sure there is space for
2697 * 'raiddisks' device extents of size 'size'.
2699 unsigned long long minsize
= size
;
2703 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2709 e
= get_extents(ddf
, dl
);
2712 unsigned long long esize
;
2713 esize
= e
[i
].start
- pos
;
2714 if (esize
>= minsize
)
2716 pos
= e
[i
].start
+ e
[i
].size
;
2718 } while (e
[i
-1].size
);
2723 if (dcnt
< raiddisks
) {
2726 Name
": ddf: Not enough devices with "
2727 "space for this array (%d < %d)\n",
2733 /* This device must be a member of the set */
2734 if (stat(dev
, &stb
) < 0)
2736 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2738 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
2739 if (dl
->major
== major(stb
.st_rdev
) &&
2740 dl
->minor
== minor(stb
.st_rdev
))
2745 fprintf(stderr
, Name
": ddf: %s is not in the "
2750 e
= get_extents(ddf
, dl
);
2754 unsigned long long esize
;
2755 esize
= e
[i
].start
- pos
;
2756 if (esize
>= maxsize
)
2758 pos
= e
[i
].start
+ e
[i
].size
;
2760 } while (e
[i
-1].size
);
2761 *freesize
= maxsize
;
2767 static int load_super_ddf_all(struct supertype
*st
, int fd
,
2768 void **sbp
, char *devname
, int keep_fd
)
2771 struct ddf_super
*super
;
2772 struct mdinfo
*sd
, *best
= NULL
;
2777 int devnum
= fd2devnum(fd
);
2778 enum sysfs_read_flags flags
;
2780 flags
= GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
;
2781 if (mdmon_running(devnum
))
2782 flags
|= SKIP_GONE_DEVS
;
2784 sra
= sysfs_read(fd
, 0, flags
);
2787 if (sra
->array
.major_version
!= -1 ||
2788 sra
->array
.minor_version
!= -2 ||
2789 strcmp(sra
->text_version
, "ddf") != 0)
2792 if (posix_memalign((void**)&super
, 512, sizeof(*super
)) != 0)
2794 memset(super
, 0, sizeof(*super
));
2796 /* first, try each device, and choose the best ddf */
2797 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2799 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2800 dfd
= dev_open(nm
, O_RDONLY
);
2803 rv
= load_ddf_headers(dfd
, super
, NULL
);
2806 seq
= __be32_to_cpu(super
->active
->seq
);
2807 if (super
->active
->openflag
)
2809 if (!best
|| seq
> bestseq
) {
2817 /* OK, load this ddf */
2818 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
2819 dfd
= dev_open(nm
, O_RDONLY
);
2822 load_ddf_headers(dfd
, super
, NULL
);
2823 load_ddf_global(dfd
, super
, NULL
);
2825 /* Now we need the device-local bits */
2826 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2829 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2830 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2833 rv
= load_ddf_headers(dfd
, super
, NULL
);
2835 rv
= load_ddf_local(dfd
, super
, NULL
, keep_fd
);
2836 if (!keep_fd
) close(dfd
);
2840 if (st
->subarray
[0]) {
2843 for (v
= super
->conflist
; v
; v
= v
->next
)
2844 if (v
->vcnum
== atoi(st
->subarray
))
2845 super
->currentconf
= v
;
2846 if (!super
->currentconf
)
2850 if (st
->ss
== NULL
) {
2851 st
->ss
= &super_ddf
;
2852 st
->minor_version
= 0;
2854 st
->container_dev
= fd2devnum(fd
);
2856 st
->loaded_container
= 1;
2859 #endif /* MDASSEMBLE */
2861 static struct mdinfo
*container_content_ddf(struct supertype
*st
)
2863 /* Given a container loaded by load_super_ddf_all,
2864 * extract information about all the arrays into
2867 * For each vcl in conflist: create an mdinfo, fill it in,
2868 * then look for matching devices (phys_refnum) in dlist
2869 * and create appropriate device mdinfo.
2871 struct ddf_super
*ddf
= st
->sb
;
2872 struct mdinfo
*rest
= NULL
;
2875 for (vc
= ddf
->conflist
; vc
; vc
=vc
->next
)
2879 struct mdinfo
*this;
2880 this = malloc(sizeof(*this));
2881 memset(this, 0, sizeof(*this));
2885 this->array
.level
= map_num1(ddf_level_num
, vc
->conf
.prl
);
2886 this->array
.raid_disks
=
2887 __be16_to_cpu(vc
->conf
.prim_elmnt_count
);
2888 this->array
.layout
= rlq_to_layout(vc
->conf
.rlq
, vc
->conf
.prl
,
2889 this->array
.raid_disks
);
2890 this->array
.md_minor
= -1;
2891 this->array
.major_version
= -1;
2892 this->array
.minor_version
= -2;
2893 this->array
.ctime
= DECADE
+
2894 __be32_to_cpu(*(__u32
*)(vc
->conf
.guid
+16));
2895 this->array
.utime
= DECADE
+
2896 __be32_to_cpu(vc
->conf
.timestamp
);
2897 this->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
2900 if ((ddf
->virt
->entries
[i
].state
& DDF_state_inconsistent
) ||
2901 (ddf
->virt
->entries
[i
].init_state
& DDF_initstate_mask
) !=
2903 this->array
.state
= 0;
2904 this->resync_start
= 0;
2906 this->array
.state
= 1;
2907 this->resync_start
= ~0ULL;
2909 memcpy(this->name
, ddf
->virt
->entries
[i
].name
, 16);
2912 if (this->name
[j
] == ' ')
2915 memset(this->uuid
, 0, sizeof(this->uuid
));
2916 this->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
2917 this->array
.size
= this->component_size
/ 2;
2918 this->container_member
= i
;
2920 ddf
->currentconf
= vc
;
2921 uuid_from_super_ddf(st
, this->uuid
);
2922 ddf
->currentconf
= NULL
;
2924 sprintf(this->text_version
, "/%s/%d",
2925 devnum2devname(st
->container_dev
),
2926 this->container_member
);
2928 for (i
=0 ; i
< ddf
->mppe
; i
++) {
2932 if (vc
->conf
.phys_refnum
[i
] == 0xFFFFFFFF)
2935 this->array
.working_disks
++;
2937 for (d
= ddf
->dlist
; d
; d
=d
->next
)
2938 if (d
->disk
.refnum
== vc
->conf
.phys_refnum
[i
])
2941 /* Haven't found that one yet, maybe there are others */
2944 dev
= malloc(sizeof(*dev
));
2945 memset(dev
, 0, sizeof(*dev
));
2946 dev
->next
= this->devs
;
2949 dev
->disk
.number
= __be32_to_cpu(d
->disk
.refnum
);
2950 dev
->disk
.major
= d
->major
;
2951 dev
->disk
.minor
= d
->minor
;
2952 dev
->disk
.raid_disk
= i
;
2953 dev
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
2955 dev
->events
= __be32_to_cpu(ddf
->primary
.seq
);
2956 dev
->data_offset
= __be64_to_cpu(vc
->lba_offset
[i
]);
2957 dev
->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
2959 strcpy(dev
->name
, d
->devname
);
2965 static int store_zero_ddf(struct supertype
*st
, int fd
)
2967 unsigned long long dsize
;
2971 if (!get_dev_size(fd
, NULL
, &dsize
))
2974 if (posix_memalign(&buf
, 512, 512) != 0)
2976 memset(buf
, 0, 512);
2978 lseek64(fd
, dsize
-512, 0);
2979 rc
= write(fd
, buf
, 512);
2986 static int compare_super_ddf(struct supertype
*st
, struct supertype
*tst
)
2990 * 0 same, or first was empty, and second was copied
2991 * 1 second had wrong number
2993 * 3 wrong other info
2995 struct ddf_super
*first
= st
->sb
;
2996 struct ddf_super
*second
= tst
->sb
;
3004 if (memcmp(first
->anchor
.guid
, second
->anchor
.guid
, DDF_GUID_LEN
) != 0)
3007 /* FIXME should I look at anything else? */
3013 * A new array 'a' has been started which claims to be instance 'inst'
3014 * within container 'c'.
3015 * We need to confirm that the array matches the metadata in 'c' so
3016 * that we don't corrupt any metadata.
3018 static int ddf_open_new(struct supertype
*c
, struct active_array
*a
, char *inst
)
3020 dprintf("ddf: open_new %s\n", inst
);
3021 a
->info
.container_member
= atoi(inst
);
3026 * The array 'a' is to be marked clean in the metadata.
3027 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
3028 * clean up to the point (in sectors). If that cannot be recorded in the
3029 * metadata, then leave it as dirty.
3031 * For DDF, we need to clear the DDF_state_inconsistent bit in the
3032 * !global! virtual_disk.virtual_entry structure.
3034 static int ddf_set_array_state(struct active_array
*a
, int consistent
)
3036 struct ddf_super
*ddf
= a
->container
->sb
;
3037 int inst
= a
->info
.container_member
;
3038 int old
= ddf
->virt
->entries
[inst
].state
;
3039 if (consistent
== 2) {
3040 /* Should check if a recovery should be started FIXME */
3042 if (!is_resync_complete(a
))
3046 ddf
->virt
->entries
[inst
].state
&= ~DDF_state_inconsistent
;
3048 ddf
->virt
->entries
[inst
].state
|= DDF_state_inconsistent
;
3049 if (old
!= ddf
->virt
->entries
[inst
].state
)
3050 ddf
->updates_pending
= 1;
3052 old
= ddf
->virt
->entries
[inst
].init_state
;
3053 ddf
->virt
->entries
[inst
].init_state
&= ~DDF_initstate_mask
;
3054 if (is_resync_complete(a
))
3055 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_full
;
3056 else if (a
->resync_start
== 0)
3057 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_not
;
3059 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_quick
;
3060 if (old
!= ddf
->virt
->entries
[inst
].init_state
)
3061 ddf
->updates_pending
= 1;
3063 dprintf("ddf mark %d %s %llu\n", inst
, consistent
?"clean":"dirty",
3069 * The state of each disk is stored in the global phys_disk structure
3070 * in phys_disk.entries[n].state.
3071 * This makes various combinations awkward.
3072 * - When a device fails in any array, it must be failed in all arrays
3073 * that include a part of this device.
3074 * - When a component is rebuilding, we cannot include it officially in the
3075 * array unless this is the only array that uses the device.
3077 * So: when transitioning:
3078 * Online -> failed, just set failed flag. monitor will propagate
3079 * spare -> online, the device might need to be added to the array.
3080 * spare -> failed, just set failed. Don't worry if in array or not.
3082 static void ddf_set_disk(struct active_array
*a
, int n
, int state
)
3084 struct ddf_super
*ddf
= a
->container
->sb
;
3085 int inst
= a
->info
.container_member
;
3086 struct vd_config
*vc
= find_vdcr(ddf
, inst
);
3087 int pd
= find_phys(ddf
, vc
->phys_refnum
[n
]);
3091 dprintf("ddf: cannot find instance %d!!\n", inst
);
3095 /* disk doesn't currently exist. If it is now in_sync,
3097 if ((state
& DS_INSYNC
) && ! (state
& DS_FAULTY
)) {
3098 /* Find dev 'n' in a->info->devs, determine the
3099 * ddf refnum, and set vc->phys_refnum and update
3105 int old
= ddf
->phys
->entries
[pd
].state
;
3106 if (state
& DS_FAULTY
)
3107 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Failed
);
3108 if (state
& DS_INSYNC
) {
3109 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Online
);
3110 ddf
->phys
->entries
[pd
].state
&= __cpu_to_be16(~DDF_Rebuilding
);
3112 if (old
!= ddf
->phys
->entries
[pd
].state
)
3113 ddf
->updates_pending
= 1;
3116 dprintf("ddf: set_disk %d to %x\n", n
, state
);
3118 /* Now we need to check the state of the array and update
3119 * virtual_disk.entries[n].state.
3120 * It needs to be one of "optimal", "degraded", "failed".
3121 * I don't understand 'deleted' or 'missing'.
3124 for (i
=0; i
< a
->info
.array
.raid_disks
; i
++) {
3125 pd
= find_phys(ddf
, vc
->phys_refnum
[i
]);
3128 st
= __be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
3129 if ((st
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
3133 state
= DDF_state_degraded
;
3134 if (working
== a
->info
.array
.raid_disks
)
3135 state
= DDF_state_optimal
;
3136 else switch(vc
->prl
) {
3140 state
= DDF_state_failed
;
3144 state
= DDF_state_failed
;
3148 if (working
< a
->info
.array
.raid_disks
-1)
3149 state
= DDF_state_failed
;
3152 if (working
< a
->info
.array
.raid_disks
-2)
3153 state
= DDF_state_failed
;
3154 else if (working
== a
->info
.array
.raid_disks
-1)
3155 state
= DDF_state_part_optimal
;
3159 if (ddf
->virt
->entries
[inst
].state
!=
3160 ((ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
3163 ddf
->virt
->entries
[inst
].state
=
3164 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
3166 ddf
->updates_pending
= 1;
3171 static void ddf_sync_metadata(struct supertype
*st
)
3175 * Write all data to all devices.
3176 * Later, we might be able to track whether only local changes
3177 * have been made, or whether any global data has been changed,
3178 * but ddf is sufficiently weird that it probably always
3179 * changes global data ....
3181 struct ddf_super
*ddf
= st
->sb
;
3182 if (!ddf
->updates_pending
)
3184 ddf
->updates_pending
= 0;
3185 __write_init_super_ddf(st
, 0);
3186 dprintf("ddf: sync_metadata\n");
3189 static void ddf_process_update(struct supertype
*st
,
3190 struct metadata_update
*update
)
3192 /* Apply this update to the metadata.
3193 * The first 4 bytes are a DDF_*_MAGIC which guides
3195 * Possible update are:
3196 * DDF_PHYS_RECORDS_MAGIC
3197 * Add a new physical device. Changes to this record
3198 * only happen implicitly.
3199 * used_pdes is the device number.
3200 * DDF_VIRT_RECORDS_MAGIC
3201 * Add a new VD. Possibly also change the 'access' bits.
3202 * populated_vdes is the entry number.
3204 * New or updated VD. the VIRT_RECORD must already
3205 * exist. For an update, phys_refnum and lba_offset
3206 * (at least) are updated, and the VD_CONF must
3207 * be written to precisely those devices listed with
3209 * DDF_SPARE_ASSIGN_MAGIC
3210 * replacement Spare Assignment Record... but for which device?
3213 * - to create a new array, we send a VIRT_RECORD and
3214 * a VD_CONF. Then assemble and start the array.
3215 * - to activate a spare we send a VD_CONF to add the phys_refnum
3216 * and offset. This will also mark the spare as active with
3217 * a spare-assignment record.
3219 struct ddf_super
*ddf
= st
->sb
;
3220 __u32
*magic
= (__u32
*)update
->buf
;
3221 struct phys_disk
*pd
;
3222 struct virtual_disk
*vd
;
3223 struct vd_config
*vc
;
3229 dprintf("Process update %x\n", *magic
);
3232 case DDF_PHYS_RECORDS_MAGIC
:
3234 if (update
->len
!= (sizeof(struct phys_disk
) +
3235 sizeof(struct phys_disk_entry
)))
3237 pd
= (struct phys_disk
*)update
->buf
;
3239 ent
= __be16_to_cpu(pd
->used_pdes
);
3240 if (ent
>= __be16_to_cpu(ddf
->phys
->max_pdes
))
3242 if (!all_ff(ddf
->phys
->entries
[ent
].guid
))
3244 ddf
->phys
->entries
[ent
] = pd
->entries
[0];
3245 ddf
->phys
->used_pdes
= __cpu_to_be16(1 +
3246 __be16_to_cpu(ddf
->phys
->used_pdes
));
3247 ddf
->updates_pending
= 1;
3248 if (ddf
->add_list
) {
3249 struct active_array
*a
;
3250 struct dl
*al
= ddf
->add_list
;
3251 ddf
->add_list
= al
->next
;
3253 al
->next
= ddf
->dlist
;
3256 /* As a device has been added, we should check
3257 * for any degraded devices that might make
3258 * use of this spare */
3259 for (a
= st
->arrays
; a
; a
=a
->next
)
3260 a
->check_degraded
= 1;
3264 case DDF_VIRT_RECORDS_MAGIC
:
3266 if (update
->len
!= (sizeof(struct virtual_disk
) +
3267 sizeof(struct virtual_entry
)))
3269 vd
= (struct virtual_disk
*)update
->buf
;
3271 ent
= __be16_to_cpu(vd
->populated_vdes
);
3272 if (ent
>= __be16_to_cpu(ddf
->virt
->max_vdes
))
3274 if (!all_ff(ddf
->virt
->entries
[ent
].guid
))
3276 ddf
->virt
->entries
[ent
] = vd
->entries
[0];
3277 ddf
->virt
->populated_vdes
= __cpu_to_be16(1 +
3278 __be16_to_cpu(ddf
->virt
->populated_vdes
));
3279 ddf
->updates_pending
= 1;
3282 case DDF_VD_CONF_MAGIC
:
3283 dprintf("len %d %d\n", update
->len
, ddf
->conf_rec_len
);
3285 mppe
= __be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
3286 if (update
->len
!= ddf
->conf_rec_len
* 512)
3288 vc
= (struct vd_config
*)update
->buf
;
3289 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
3290 if (memcmp(vcl
->conf
.guid
, vc
->guid
, DDF_GUID_LEN
) == 0)
3292 dprintf("vcl = %p\n", vcl
);
3294 /* An update, just copy the phys_refnum and lba_offset
3297 memcpy(vcl
->conf
.phys_refnum
, vc
->phys_refnum
,
3298 mppe
* (sizeof(__u32
) + sizeof(__u64
)));
3303 vcl
= update
->space
;
3304 update
->space
= NULL
;
3305 vcl
->next
= ddf
->conflist
;
3306 memcpy(&vcl
->conf
, vc
, update
->len
);
3307 vcl
->lba_offset
= (__u64
*)
3308 &vcl
->conf
.phys_refnum
[mppe
];
3309 ddf
->conflist
= vcl
;
3311 /* Now make sure vlist is correct for each dl. */
3312 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
3315 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
3316 for (dn
=0; dn
< ddf
->mppe
; dn
++)
3317 if (vcl
->conf
.phys_refnum
[dn
] ==
3319 dprintf("dev %d has %p at %d\n",
3320 dl
->pdnum
, vcl
, vn
);
3321 dl
->vlist
[vn
++] = vcl
;
3324 while (vn
< ddf
->max_part
)
3325 dl
->vlist
[vn
++] = NULL
;
3327 ddf
->phys
->entries
[dl
->pdnum
].type
&=
3328 ~__cpu_to_be16(DDF_Global_Spare
);
3329 ddf
->phys
->entries
[dl
->pdnum
].type
|=
3330 __cpu_to_be16(DDF_Active_in_VD
);
3333 ddf
->phys
->entries
[dl
->pdnum
].type
&=
3334 ~__cpu_to_be16(DDF_Global_Spare
);
3335 ddf
->phys
->entries
[dl
->pdnum
].type
|=
3336 __cpu_to_be16(DDF_Spare
);
3338 if (!dl
->vlist
[0] && !dl
->spare
) {
3339 ddf
->phys
->entries
[dl
->pdnum
].type
|=
3340 __cpu_to_be16(DDF_Global_Spare
);
3341 ddf
->phys
->entries
[dl
->pdnum
].type
&=
3342 ~__cpu_to_be16(DDF_Spare
|
3346 ddf
->updates_pending
= 1;
3348 case DDF_SPARE_ASSIGN_MAGIC
:
3353 static void ddf_prepare_update(struct supertype
*st
,
3354 struct metadata_update
*update
)
3356 /* This update arrived at managemon.
3357 * We are about to pass it to monitor.
3358 * If a malloc is needed, do it here.
3360 struct ddf_super
*ddf
= st
->sb
;
3361 __u32
*magic
= (__u32
*)update
->buf
;
3362 if (*magic
== DDF_VD_CONF_MAGIC
)
3363 if (posix_memalign(&update
->space
, 512,
3364 offsetof(struct vcl
, conf
)
3365 + ddf
->conf_rec_len
* 512) != 0)
3366 update
->space
= NULL
;
3370 * Check if the array 'a' is degraded but not failed.
3371 * If it is, find as many spares as are available and needed and
3372 * arrange for their inclusion.
3373 * We only choose devices which are not already in the array,
3374 * and prefer those with a spare-assignment to this array.
3375 * otherwise we choose global spares - assuming always that
3376 * there is enough room.
3377 * For each spare that we assign, we return an 'mdinfo' which
3378 * describes the position for the device in the array.
3379 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
3380 * the new phys_refnum and lba_offset values.
3382 * Only worry about BVDs at the moment.
3384 static struct mdinfo
*ddf_activate_spare(struct active_array
*a
,
3385 struct metadata_update
**updates
)
3389 struct ddf_super
*ddf
= a
->container
->sb
;
3391 struct mdinfo
*rv
= NULL
;
3393 struct metadata_update
*mu
;
3396 struct vd_config
*vc
;
3399 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
3400 if ((d
->curr_state
& DS_FAULTY
) &&
3402 /* wait for Removal to happen */
3404 if (d
->state_fd
>= 0)
3408 dprintf("ddf_activate: working=%d (%d) level=%d\n", working
, a
->info
.array
.raid_disks
,
3409 a
->info
.array
.level
);
3410 if (working
== a
->info
.array
.raid_disks
)
3411 return NULL
; /* array not degraded */
3412 switch (a
->info
.array
.level
) {
3415 return NULL
; /* failed */
3419 if (working
< a
->info
.array
.raid_disks
- 1)
3420 return NULL
; /* failed */
3423 if (working
< a
->info
.array
.raid_disks
- 2)
3424 return NULL
; /* failed */
3426 default: /* concat or stripe */
3427 return NULL
; /* failed */
3430 /* For each slot, if it is not working, find a spare */
3432 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
3433 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3434 if (d
->disk
.raid_disk
== i
)
3436 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
3437 if (d
&& (d
->state_fd
>= 0))
3440 /* OK, this device needs recovery. Find a spare */
3442 for ( ; dl
; dl
= dl
->next
) {
3443 unsigned long long esize
;
3444 unsigned long long pos
;
3447 int is_dedicated
= 0;
3450 /* If in this array, skip */
3451 for (d2
= a
->info
.devs
; d2
; d2
= d2
->next
)
3452 if (d2
->disk
.major
== dl
->major
&&
3453 d2
->disk
.minor
== dl
->minor
) {
3454 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
3459 if (ddf
->phys
->entries
[dl
->pdnum
].type
&
3460 __cpu_to_be16(DDF_Spare
)) {
3461 /* Check spare assign record */
3463 if (dl
->spare
->type
& DDF_spare_dedicated
) {
3464 /* check spare_ents for guid */
3466 j
< __be16_to_cpu(dl
->spare
->populated
);
3468 if (memcmp(dl
->spare
->spare_ents
[j
].guid
,
3469 ddf
->virt
->entries
[a
->info
.container_member
].guid
,
3476 } else if (ddf
->phys
->entries
[dl
->pdnum
].type
&
3477 __cpu_to_be16(DDF_Global_Spare
)) {
3480 if ( ! (is_dedicated
||
3481 (is_global
&& global_ok
))) {
3482 dprintf("%x:%x not suitable: %d %d\n", dl
->major
, dl
->minor
,
3483 is_dedicated
, is_global
);
3487 /* We are allowed to use this device - is there space?
3488 * We need a->info.component_size sectors */
3489 ex
= get_extents(ddf
, dl
);
3491 dprintf("cannot get extents\n");
3498 esize
= ex
[j
].start
- pos
;
3499 if (esize
>= a
->info
.component_size
)
3501 pos
= ex
[i
].start
+ ex
[i
].size
;
3503 } while (ex
[i
-1].size
);
3506 if (esize
< a
->info
.component_size
) {
3507 dprintf("%x:%x has no room: %llu %llu\n", dl
->major
, dl
->minor
,
3508 esize
, a
->info
.component_size
);
3513 /* Cool, we have a device with some space at pos */
3514 di
= malloc(sizeof(*di
));
3517 memset(di
, 0, sizeof(*di
));
3518 di
->disk
.number
= i
;
3519 di
->disk
.raid_disk
= i
;
3520 di
->disk
.major
= dl
->major
;
3521 di
->disk
.minor
= dl
->minor
;
3523 di
->data_offset
= pos
;
3524 di
->component_size
= a
->info
.component_size
;
3525 di
->container_member
= dl
->pdnum
;
3528 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
3533 if (!dl
&& ! global_ok
) {
3534 /* not enough dedicated spares, try global */
3542 /* No spares found */
3544 /* Now 'rv' has a list of devices to return.
3545 * Create a metadata_update record to update the
3546 * phys_refnum and lba_offset values
3548 mu
= malloc(sizeof(*mu
));
3549 if (mu
&& posix_memalign(&mu
->space
, 512, sizeof(struct vcl
)) != 0) {
3555 struct mdinfo
*n
= rv
->next
;
3563 mu
->buf
= malloc(ddf
->conf_rec_len
* 512);
3564 mu
->len
= ddf
->conf_rec_len
;
3565 mu
->next
= *updates
;
3566 vc
= find_vdcr(ddf
, a
->info
.container_member
);
3567 memcpy(mu
->buf
, vc
, ddf
->conf_rec_len
* 512);
3569 vc
= (struct vd_config
*)mu
->buf
;
3570 lba
= (__u64
*)&vc
->phys_refnum
[ddf
->mppe
];
3571 for (di
= rv
; di
; di
= di
->next
) {
3572 vc
->phys_refnum
[di
->disk
.raid_disk
] =
3573 ddf
->phys
->entries
[dl
->pdnum
].refnum
;
3574 lba
[di
->disk
.raid_disk
] = di
->data_offset
;
3579 #endif /* MDASSEMBLE */
3581 static int ddf_level_to_layout(int level
)
3588 return ALGORITHM_LEFT_SYMMETRIC
;
3590 return ALGORITHM_ROTATING_N_CONTINUE
;
3598 struct superswitch super_ddf
= {
3600 .examine_super
= examine_super_ddf
,
3601 .brief_examine_super
= brief_examine_super_ddf
,
3602 .export_examine_super
= export_examine_super_ddf
,
3603 .detail_super
= detail_super_ddf
,
3604 .brief_detail_super
= brief_detail_super_ddf
,
3605 .validate_geometry
= validate_geometry_ddf
,
3606 .write_init_super
= write_init_super_ddf
,
3607 .add_to_super
= add_to_super_ddf
,
3609 .match_home
= match_home_ddf
,
3610 .uuid_from_super
= uuid_from_super_ddf
,
3611 .getinfo_super
= getinfo_super_ddf
,
3612 .update_super
= update_super_ddf
,
3614 .avail_size
= avail_size_ddf
,
3616 .compare_super
= compare_super_ddf
,
3618 .load_super
= load_super_ddf
,
3619 .init_super
= init_super_ddf
,
3620 .store_super
= store_zero_ddf
,
3621 .free_super
= free_super_ddf
,
3622 .match_metadata_desc
= match_metadata_desc_ddf
,
3623 .container_content
= container_content_ddf
,
3624 .default_layout
= ddf_level_to_layout
,
3630 .open_new
= ddf_open_new
,
3631 .set_array_state
= ddf_set_array_state
,
3632 .set_disk
= ddf_set_disk
,
3633 .sync_metadata
= ddf_sync_metadata
,
3634 .process_update
= ddf_process_update
,
3635 .prepare_update
= ddf_prepare_update
,
3636 .activate_spare
= ddf_activate_spare
,