2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2006-2007 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neil@brown.name>
24 * Specifications for DDF takes from Common RAID DDF Specification Revision 1.2
25 * (July 28 2006). Reused by permission of SNIA.
28 #define HAVE_STDINT_H 1
34 static inline int ROUND_UP(int a
, int base
)
36 return ((a
+base
-1)/base
)*base
;
39 /* a non-official T10 name for creation GUIDs */
40 static char T10
[] = "Linux-MD";
42 /* DDF timestamps are 1980 based, so we need to add
43 * second-in-decade-of-seventies to convert to linux timestamps.
44 * 10 years with 2 leap years.
46 #define DECADE (3600*24*(365*10+2))
49 const unsigned char *buf
,
52 /* The DDF metadata handling.
53 * DDF metadata lives at the end of the device.
54 * The last 512 byte block provides an 'anchor' which is used to locate
55 * the rest of the metadata which usually lives immediately behind the anchor.
58 * - all multibyte numeric fields are bigendian.
59 * - all strings are space padded.
63 /* Primary Raid Level (PRL) */
64 #define DDF_RAID0 0x00
65 #define DDF_RAID1 0x01
66 #define DDF_RAID3 0x03
67 #define DDF_RAID4 0x04
68 #define DDF_RAID5 0x05
69 #define DDF_RAID1E 0x11
71 #define DDF_CONCAT 0x1f
72 #define DDF_RAID5E 0x15
73 #define DDF_RAID5EE 0x25
74 #define DDF_RAID6 0x16 /* Vendor unique layout */
76 /* Raid Level Qualifier (RLQ) */
77 #define DDF_RAID0_SIMPLE 0x00
78 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
79 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
80 #define DDF_RAID3_0 0x00 /* parity in first extent */
81 #define DDF_RAID3_N 0x01 /* parity in last extent */
82 #define DDF_RAID4_0 0x00 /* parity in first extent */
83 #define DDF_RAID4_N 0x01 /* parity in last extent */
84 /* these apply to raid5e and raid5ee as well */
85 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
86 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
87 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
89 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
90 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
92 /* Secondary RAID Level (SRL) */
93 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
94 #define DDF_2MIRRORED 0x01
95 #define DDF_2CONCAT 0x02
96 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
99 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
100 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
101 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
102 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
103 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
104 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
105 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
106 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
107 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
108 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
110 #define DDF_GUID_LEN 24
111 #define DDF_REVISION "01.00.00"
114 __u32 magic
; /* DDF_HEADER_MAGIC */
116 char guid
[DDF_GUID_LEN
];
117 char revision
[8]; /* 01.00.00 */
118 __u32 seq
; /* starts at '1' */
123 __u8 pad0
; /* 0xff */
124 __u8 pad1
[12]; /* 12 * 0xff */
125 /* 64 bytes so far */
126 __u8 header_ext
[32]; /* reserved: fill with 0xff */
130 __u8 pad2
[3]; /* 0xff */
131 __u32 workspace_len
; /* sectors for vendor space -
132 * at least 32768(sectors) */
134 __u16 max_pd_entries
; /* one of 15, 63, 255, 1023, 4095 */
135 __u16 max_vd_entries
; /* 2^(4,6,8,10,12)-1 : i.e. as above */
136 __u16 max_partitions
; /* i.e. max num of configuration
137 record entries per disk */
138 __u16 config_record_len
; /* 1 +ROUNDUP(max_primary_element_entries
140 __u16 max_primary_element_entries
; /* 16, 64, 256, 1024, or 4096 */
141 __u8 pad3
[54]; /* 0xff */
142 /* 192 bytes so far */
143 __u32 controller_section_offset
;
144 __u32 controller_section_length
;
145 __u32 phys_section_offset
;
146 __u32 phys_section_length
;
147 __u32 virt_section_offset
;
148 __u32 virt_section_length
;
149 __u32 config_section_offset
;
150 __u32 config_section_length
;
151 __u32 data_section_offset
;
152 __u32 data_section_length
;
153 __u32 bbm_section_offset
;
154 __u32 bbm_section_length
;
155 __u32 diag_space_offset
;
156 __u32 diag_space_length
;
159 /* 256 bytes so far */
160 __u8 pad4
[256]; /* 0xff */
164 #define DDF_HEADER_ANCHOR 0x00
165 #define DDF_HEADER_PRIMARY 0x01
166 #define DDF_HEADER_SECONDARY 0x02
168 /* The content of the 'controller section' - global scope */
169 struct ddf_controller_data
{
170 __u32 magic
; /* DDF_CONTROLLER_MAGIC */
172 char guid
[DDF_GUID_LEN
];
173 struct controller_type
{
180 __u8 pad
[8]; /* 0xff */
181 __u8 vendor_data
[448];
184 /* The content of phys_section - global scope */
186 __u32 magic
; /* DDF_PHYS_RECORDS_MAGIC */
191 struct phys_disk_entry
{
192 char guid
[DDF_GUID_LEN
];
196 __u64 config_size
; /* DDF structures must be after here */
197 char path
[18]; /* another horrible structure really */
202 /* phys_disk_entry.type is a bitmap - bigendian remember */
203 #define DDF_Forced_PD_GUID 1
204 #define DDF_Active_in_VD 2
205 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
206 #define DDF_Spare 8 /* overrides Global_spare */
207 #define DDF_Foreign 16
208 #define DDF_Legacy 32 /* no DDF on this device */
210 #define DDF_Interface_mask 0xf00
211 #define DDF_Interface_SCSI 0x100
212 #define DDF_Interface_SAS 0x200
213 #define DDF_Interface_SATA 0x300
214 #define DDF_Interface_FC 0x400
216 /* phys_disk_entry.state is a bigendian bitmap */
218 #define DDF_Failed 2 /* overrides 1,4,8 */
219 #define DDF_Rebuilding 4
220 #define DDF_Transition 8
222 #define DDF_ReadErrors 32
223 #define DDF_Missing 64
225 /* The content of the virt_section global scope */
226 struct virtual_disk
{
227 __u32 magic
; /* DDF_VIRT_RECORDS_MAGIC */
229 __u16 populated_vdes
;
232 struct virtual_entry
{
233 char guid
[DDF_GUID_LEN
];
235 __u16 pad0
; /* 0xffff */
245 /* virtual_entry.type is a bitmap - bigendian */
247 #define DDF_Enforce_Groups 2
248 #define DDF_Unicode 4
249 #define DDF_Owner_Valid 8
251 /* virtual_entry.state is a bigendian bitmap */
252 #define DDF_state_mask 0x7
253 #define DDF_state_optimal 0x0
254 #define DDF_state_degraded 0x1
255 #define DDF_state_deleted 0x2
256 #define DDF_state_missing 0x3
257 #define DDF_state_failed 0x4
258 #define DDF_state_part_optimal 0x5
260 #define DDF_state_morphing 0x8
261 #define DDF_state_inconsistent 0x10
263 /* virtual_entry.init_state is a bigendian bitmap */
264 #define DDF_initstate_mask 0x03
265 #define DDF_init_not 0x00
266 #define DDF_init_quick 0x01 /* initialisation is progress.
267 * i.e. 'state_inconsistent' */
268 #define DDF_init_full 0x02
270 #define DDF_access_mask 0xc0
271 #define DDF_access_rw 0x00
272 #define DDF_access_ro 0x80
273 #define DDF_access_blocked 0xc0
275 /* The content of the config_section - local scope
276 * It has multiple records each config_record_len sectors
277 * They can be vd_config or spare_assign
281 __u32 magic
; /* DDF_VD_CONF_MAGIC */
283 char guid
[DDF_GUID_LEN
];
287 __u16 prim_elmnt_count
;
288 __u8 chunk_shift
; /* 0 == 512, 1==1024 etc */
291 __u8 sec_elmnt_count
;
294 __u64 blocks
; /* blocks per component could be different
295 * on different component devices...(only
296 * for concat I hope) */
297 __u64 array_blocks
; /* blocks in array */
305 __u8 v0
[32]; /* reserved- 0xff */
306 __u8 v1
[32]; /* reserved- 0xff */
307 __u8 v2
[16]; /* reserved- 0xff */
308 __u8 v3
[16]; /* reserved- 0xff */
310 __u32 phys_refnum
[0]; /* refnum of each disk in sequence */
311 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
312 bvd are always the same size */
315 /* vd_config.cache_pol[7] is a bitmap */
316 #define DDF_cache_writeback 1 /* else writethrough */
317 #define DDF_cache_wadaptive 2 /* only applies if writeback */
318 #define DDF_cache_readahead 4
319 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
320 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
321 #define DDF_cache_wallowed 32 /* enable write caching */
322 #define DDF_cache_rallowed 64 /* enable read caching */
324 struct spare_assign
{
325 __u32 magic
; /* DDF_SPARE_ASSIGN_MAGIC */
330 __u16 populated
; /* SAEs used */
331 __u16 max
; /* max SAEs */
333 struct spare_assign_entry
{
334 char guid
[DDF_GUID_LEN
];
335 __u16 secondary_element
;
339 /* spare_assign.type is a bitmap */
340 #define DDF_spare_dedicated 0x1 /* else global */
341 #define DDF_spare_revertible 0x2 /* else committable */
342 #define DDF_spare_active 0x4 /* else not active */
343 #define DDF_spare_affinity 0x8 /* enclosure affinity */
345 /* The data_section contents - local scope */
347 __u32 magic
; /* DDF_PHYS_DATA_MAGIC */
349 char guid
[DDF_GUID_LEN
];
350 __u32 refnum
; /* crc of some magic drive data ... */
351 __u8 forced_ref
; /* set when above was not result of magic */
352 __u8 forced_guid
; /* set if guid was forced rather than magic */
357 /* bbm_section content */
358 struct bad_block_log
{
365 struct mapped_block
{
366 __u64 defective_start
;
367 __u32 replacement_start
;
373 /* Struct for internally holding ddf structures */
374 /* The DDF structure stored on each device is potentially
375 * quite different, as some data is global and some is local.
376 * The global data is:
379 * - Physical disk records
380 * - Virtual disk records
382 * - Configuration records
383 * - Physical Disk data section
384 * ( and Bad block and vendor which I don't care about yet).
386 * The local data is parsed into separate lists as it is read
387 * and reconstructed for writing. This means that we only need
388 * to make config changes once and they are automatically
389 * propagated to all devices.
390 * Note that the ddf_super has space of the conf and disk data
391 * for this disk and also for a list of all such data.
392 * The list is only used for the superblock that is being
393 * built in Create or Assemble to describe the whole array.
396 struct ddf_header anchor
, primary
, secondary
, *active
;
397 struct ddf_controller_data controller
;
398 struct phys_disk
*phys
;
399 struct virtual_disk
*virt
;
401 int max_part
, mppe
, conf_rec_len
;
404 __u64
*lba_offset
; /* location in 'conf' of
406 struct vd_config conf
;
407 } *conflist
, *newconf
;
408 int conf_num
; /* Index into 'virt' of entry matching 'newconf' */
411 struct disk_data disk
;
415 int pdnum
; /* index in ->phys */
416 struct spare_assign
*spare
;
417 struct vcl
*vlist
[0]; /* max_part in size */
422 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
425 extern struct superswitch super_ddf_container
, super_ddf_bvd
, super_ddf
;
427 static int calc_crc(void *buf
, int len
)
429 /* crcs are always at the same place as in the ddf_header */
430 struct ddf_header
*ddf
= buf
;
431 __u32 oldcrc
= ddf
->crc
;
433 ddf
->crc
= 0xffffffff;
435 newcrc
= crc32(0, buf
, len
);
440 static int load_ddf_header(int fd
, unsigned long long lba
,
441 unsigned long long size
,
443 struct ddf_header
*hdr
, struct ddf_header
*anchor
)
445 /* read a ddf header (primary or secondary) from fd/lba
446 * and check that it is consistent with anchor
448 * magic, crc, guid, rev, and LBA's header_type, and
449 * everything after header_type must be the same
454 if (lseek64(fd
, lba
<<9, 0) < 0)
457 if (read(fd
, hdr
, 512) != 512)
460 if (hdr
->magic
!= DDF_HEADER_MAGIC
)
462 if (calc_crc(hdr
, 512) != hdr
->crc
)
464 if (memcmp(anchor
->guid
, hdr
->guid
, DDF_GUID_LEN
) != 0 ||
465 memcmp(anchor
->revision
, hdr
->revision
, 8) != 0 ||
466 anchor
->primary_lba
!= hdr
->primary_lba
||
467 anchor
->secondary_lba
!= hdr
->secondary_lba
||
469 memcmp(anchor
->pad2
, hdr
->pad2
, 512 -
470 offsetof(struct ddf_header
, pad2
)) != 0)
473 /* Looks good enough to me... */
477 static void *load_section(int fd
, struct ddf_super
*super
, void *buf
,
478 __u32 offset_be
, __u32 len_be
, int check
)
480 unsigned long long offset
= __be32_to_cpu(offset_be
);
481 unsigned long long len
= __be32_to_cpu(len_be
);
482 int dofree
= (buf
== NULL
);
485 if (len
!= 2 && len
!= 8 && len
!= 32
486 && len
!= 128 && len
!= 512)
492 /* All pre-allocated sections are a single block */
496 buf
= malloc(len
<<9);
500 if (super
->active
->type
== 1)
501 offset
+= __be64_to_cpu(super
->active
->primary_lba
);
503 offset
+= __be64_to_cpu(super
->active
->secondary_lba
);
505 if (lseek64(fd
, offset
<<9, 0) != (offset
<<9)) {
510 if (read(fd
, buf
, len
<<9) != (len
<<9)) {
518 static int load_ddf_headers(int fd
, struct ddf_super
*super
, char *devname
)
520 unsigned long long dsize
;
522 get_dev_size(fd
, NULL
, &dsize
);
524 if (lseek64(fd
, dsize
-512, 0) < 0) {
527 Name
": Cannot seek to anchor block on %s: %s\n",
528 devname
, strerror(errno
));
531 if (read(fd
, &super
->anchor
, 512) != 512) {
534 Name
": Cannot read anchor block on %s: %s\n",
535 devname
, strerror(errno
));
538 if (super
->anchor
.magic
!= DDF_HEADER_MAGIC
) {
540 fprintf(stderr
, Name
": no DDF anchor found on %s\n",
544 if (calc_crc(&super
->anchor
, 512) != super
->anchor
.crc
) {
546 fprintf(stderr
, Name
": bad CRC on anchor on %s\n",
550 if (memcmp(super
->anchor
.revision
, DDF_REVISION
, 8) != 0) {
552 fprintf(stderr
, Name
": can only support super revision"
553 " %.8s, not %.8s on %s\n",
554 DDF_REVISION
, super
->anchor
.revision
, devname
);
557 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.primary_lba
),
559 &super
->primary
, &super
->anchor
) == 0) {
562 Name
": Failed to load primary DDF header "
566 super
->active
= &super
->primary
;
567 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.secondary_lba
),
569 &super
->secondary
, &super
->anchor
)) {
570 if ((__be32_to_cpu(super
->primary
.seq
)
571 < __be32_to_cpu(super
->secondary
.seq
) &&
572 !super
->secondary
.openflag
)
573 || (__be32_to_cpu(super
->primary
.seq
)
574 == __be32_to_cpu(super
->secondary
.seq
) &&
575 super
->primary
.openflag
&& !super
->secondary
.openflag
)
577 super
->active
= &super
->secondary
;
582 static int load_ddf_global(int fd
, struct ddf_super
*super
, char *devname
)
585 ok
= load_section(fd
, super
, &super
->controller
,
586 super
->active
->controller_section_offset
,
587 super
->active
->controller_section_length
,
589 super
->phys
= load_section(fd
, super
, NULL
,
590 super
->active
->phys_section_offset
,
591 super
->active
->phys_section_length
,
593 super
->pdsize
= __be32_to_cpu(super
->active
->phys_section_length
) * 512;
595 super
->virt
= load_section(fd
, super
, NULL
,
596 super
->active
->virt_section_offset
,
597 super
->active
->virt_section_length
,
599 super
->vdsize
= __be32_to_cpu(super
->active
->virt_section_length
) * 512;
609 super
->conflist
= NULL
;
612 super
->max_part
= __be16_to_cpu(super
->active
->max_partitions
);
613 super
->mppe
= __be16_to_cpu(super
->active
->max_primary_element_entries
);
614 super
->conf_rec_len
= __be16_to_cpu(super
->active
->config_record_len
);
618 static int load_ddf_local(int fd
, struct ddf_super
*super
,
619 char *devname
, int keep
)
627 /* First the local disk info */
628 dl
= malloc(sizeof(*dl
) +
629 (super
->max_part
) * sizeof(dl
->vlist
[0]));
631 load_section(fd
, super
, &dl
->disk
,
632 super
->active
->data_section_offset
,
633 super
->active
->data_section_length
,
635 dl
->devname
= devname
? strdup(devname
) : NULL
;
638 dl
->major
= major(stb
.st_rdev
);
639 dl
->minor
= minor(stb
.st_rdev
);
640 dl
->next
= super
->dlist
;
641 dl
->fd
= keep
? fd
: -1;
643 for (i
=0 ; i
< super
->max_part
; i
++)
647 for (i
=0; i
< __be16_to_cpu(super
->active
->max_pd_entries
); i
++)
648 if (memcmp(super
->phys
->entries
[i
].guid
,
649 dl
->disk
.guid
, DDF_GUID_LEN
) == 0)
653 /* Now the config list. */
654 /* 'conf' is an array of config entries, some of which are
655 * probably invalid. Those which are good need to be copied into
659 conf
= load_section(fd
, super
, NULL
,
660 super
->active
->config_section_offset
,
661 super
->active
->config_section_length
,
666 i
< __be32_to_cpu(super
->active
->config_section_length
);
667 i
+= super
->conf_rec_len
) {
668 struct vd_config
*vd
=
669 (struct vd_config
*)((char*)conf
+ i
*512);
672 if (vd
->magic
== DDF_SPARE_ASSIGN_MAGIC
) {
675 dl
->spare
= malloc(super
->conf_rec_len
*512);
676 memcpy(dl
->spare
, vd
, super
->conf_rec_len
*512);
679 if (vd
->magic
!= DDF_VD_CONF_MAGIC
)
681 for (vcl
= super
->conflist
; vcl
; vcl
= vcl
->next
) {
682 if (memcmp(vcl
->conf
.guid
,
683 vd
->guid
, DDF_GUID_LEN
) == 0)
688 dl
->vlist
[vnum
++] = vcl
;
689 if (__be32_to_cpu(vd
->seqnum
) <=
690 __be32_to_cpu(vcl
->conf
.seqnum
))
693 vcl
= malloc(super
->conf_rec_len
*512 +
694 offsetof(struct vcl
, conf
));
695 vcl
->next
= super
->conflist
;
696 super
->conflist
= vcl
;
697 dl
->vlist
[vnum
++] = vcl
;
699 memcpy(&vcl
->conf
, vd
, super
->conf_rec_len
*512);
700 vcl
->lba_offset
= (__u64
*)
701 &vcl
->conf
.phys_refnum
[super
->mppe
];
709 static int load_super_ddf_all(struct supertype
*st
, int fd
,
710 void **sbp
, char *devname
, int keep_fd
);
712 static int load_super_ddf(struct supertype
*st
, int fd
,
715 unsigned long long dsize
;
716 struct ddf_super
*super
;
720 if (load_super_ddf_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
724 if (get_dev_size(fd
, devname
, &dsize
) == 0)
727 /* 32M is a lower bound */
728 if (dsize
<= 32*1024*1024) {
731 Name
": %s is too small for ddf: "
732 "size is %llu sectors.\n",
740 Name
": %s is an odd size for ddf: "
741 "size is %llu bytes.\n",
747 super
= malloc(sizeof(*super
));
749 fprintf(stderr
, Name
": malloc of %zu failed.\n",
753 memset(super
, 0, sizeof(*super
));
755 rv
= load_ddf_headers(fd
, super
, devname
);
761 /* Have valid headers and have chosen the best. Let's read in the rest*/
763 rv
= load_ddf_global(fd
, super
, devname
);
768 Name
": Failed to load all information "
769 "sections on %s\n", devname
);
774 load_ddf_local(fd
, super
, devname
, 0);
776 /* Should possibly check the sections .... */
779 if (st
->ss
== NULL
) {
781 st
->minor_version
= 0;
788 static void free_super_ddf(struct supertype
*st
)
790 struct ddf_super
*ddf
= st
->sb
;
795 while (ddf
->conflist
) {
796 struct vcl
*v
= ddf
->conflist
;
797 ddf
->conflist
= v
->next
;
801 struct dl
*d
= ddf
->dlist
;
802 ddf
->dlist
= d
->next
;
813 static struct supertype
*match_metadata_desc_ddf(char *arg
)
815 /* 'ddf' only support containers */
816 struct supertype
*st
;
817 if (strcmp(arg
, "ddf") != 0 &&
818 strcmp(arg
, "default") != 0
822 st
= malloc(sizeof(*st
));
823 memset(st
, 0, sizeof(*st
));
826 st
->minor_version
= 0;
831 static struct supertype
*match_metadata_desc_ddf_bvd(char *arg
)
833 struct supertype
*st
;
834 if (strcmp(arg
, "ddf/bvd") != 0 &&
835 strcmp(arg
, "bvd") != 0 &&
836 strcmp(arg
, "default") != 0
840 st
= malloc(sizeof(*st
));
841 memset(st
, 0, sizeof(*st
));
842 st
->ss
= &super_ddf_bvd
;
844 st
->minor_version
= 0;
848 static struct supertype
*match_metadata_desc_ddf_svd(char *arg
)
850 struct supertype
*st
;
851 if (strcmp(arg
, "ddf/svd") != 0 &&
852 strcmp(arg
, "svd") != 0 &&
853 strcmp(arg
, "default") != 0
857 st
= malloc(sizeof(*st
));
858 memset(st
, 0, sizeof(*st
));
859 st
->ss
= &super_ddf_svd
;
861 st
->minor_version
= 0;
868 static mapping_t ddf_state
[] = {
874 { "Partially Optimal", 5},
880 static mapping_t ddf_init_state
[] = {
881 { "Not Initialised", 0},
882 { "QuickInit in Progress", 1},
883 { "Fully Initialised", 2},
887 static mapping_t ddf_access
[] = {
891 { "Blocked (no access)", 3},
895 static mapping_t ddf_level
[] = {
896 { "RAID0", DDF_RAID0
},
897 { "RAID1", DDF_RAID1
},
898 { "RAID3", DDF_RAID3
},
899 { "RAID4", DDF_RAID4
},
900 { "RAID5", DDF_RAID5
},
901 { "RAID1E",DDF_RAID1E
},
903 { "CONCAT",DDF_CONCAT
},
904 { "RAID5E",DDF_RAID5E
},
905 { "RAID5EE",DDF_RAID5EE
},
906 { "RAID6", DDF_RAID6
},
909 static mapping_t ddf_sec_level
[] = {
910 { "Striped", DDF_2STRIPED
},
911 { "Mirrored", DDF_2MIRRORED
},
912 { "Concat", DDF_2CONCAT
},
913 { "Spanned", DDF_2SPANNED
},
921 static struct num_mapping ddf_level_num
[] = {
924 { DDF_RAID3
, LEVEL_UNSUPPORTED
},
927 { DDF_RAID1E
, LEVEL_UNSUPPORTED
},
928 { DDF_JBOD
, LEVEL_UNSUPPORTED
},
929 { DDF_CONCAT
, LEVEL_LINEAR
},
930 { DDF_RAID5E
, LEVEL_UNSUPPORTED
},
931 { DDF_RAID5EE
, LEVEL_UNSUPPORTED
},
936 static int map_num1(struct num_mapping
*map
, int num
)
939 for (i
=0 ; map
[i
].num1
!= MAXINT
; i
++)
940 if (map
[i
].num1
== num
)
946 static void print_guid(char *guid
, int tstamp
)
948 /* A GUIDs are part (or all) ASCII and part binary.
949 * They tend to be space padded.
950 * We ignore trailing spaces and print numbers
951 * <0x20 and >=0x7f as \xXX
952 * Some GUIDs have a time stamp in bytes 16-19.
953 * We print that if appropriate
955 int l
= DDF_GUID_LEN
;
957 while (l
&& guid
[l
-1] == ' ')
959 for (i
=0 ; i
<l
; i
++) {
960 if (guid
[i
] >= 0x20 && guid
[i
] < 0x7f)
961 fputc(guid
[i
], stdout
);
963 fprintf(stdout
, "\\x%02x", guid
[i
]&255);
966 time_t then
= __be32_to_cpu(*(__u32
*)(guid
+16)) + DECADE
;
969 tm
= localtime(&then
);
970 strftime(tbuf
, 100, " (%D %T)",tm
);
975 static void examine_vd(int n
, struct ddf_super
*sb
, char *guid
)
977 int crl
= sb
->conf_rec_len
;
980 for (vcl
= sb
->conflist
; vcl
; vcl
= vcl
->next
) {
981 struct vd_config
*vc
= &vcl
->conf
;
983 if (calc_crc(vc
, crl
*512) != vc
->crc
)
985 if (memcmp(vc
->guid
, guid
, DDF_GUID_LEN
) != 0)
988 /* Ok, we know about this VD, let's give more details */
989 printf(" Raid Devices[%d] : %d\n", n
,
990 __be16_to_cpu(vc
->prim_elmnt_count
));
991 printf(" Chunk Size[%d] : %d sectors\n", n
,
992 1 << vc
->chunk_shift
);
993 printf(" Raid Level[%d] : %s\n", n
,
994 map_num(ddf_level
, vc
->prl
)?:"-unknown-");
995 if (vc
->sec_elmnt_count
!= 1) {
996 printf(" Secondary Position[%d] : %d of %d\n", n
,
997 vc
->sec_elmnt_seq
, vc
->sec_elmnt_count
);
998 printf(" Secondary Level[%d] : %s\n", n
,
999 map_num(ddf_sec_level
, vc
->srl
) ?: "-unknown-");
1001 printf(" Device Size[%d] : %llu\n", n
,
1002 __be64_to_cpu(vc
->blocks
)/2);
1003 printf(" Array Size[%d] : %llu\n", n
,
1004 __be64_to_cpu(vc
->array_blocks
)/2);
1008 static void examine_vds(struct ddf_super
*sb
)
1010 int cnt
= __be16_to_cpu(sb
->virt
->populated_vdes
);
1012 printf(" Virtual Disks : %d\n", cnt
);
1014 for (i
=0; i
<cnt
; i
++) {
1015 struct virtual_entry
*ve
= &sb
->virt
->entries
[i
];
1016 printf(" VD GUID[%d] : ", i
); print_guid(ve
->guid
, 1);
1018 printf(" unit[%d] : %d\n", i
, __be16_to_cpu(ve
->unit
));
1019 printf(" state[%d] : %s, %s%s\n", i
,
1020 map_num(ddf_state
, ve
->state
& 7),
1021 (ve
->state
& 8) ? "Morphing, ": "",
1022 (ve
->state
& 16)? "Not Consistent" : "Consistent");
1023 printf(" init state[%d] : %s\n", i
,
1024 map_num(ddf_init_state
, ve
->init_state
&3));
1025 printf(" access[%d] : %s\n", i
,
1026 map_num(ddf_access
, (ve
->init_state
>>6) & 3));
1027 printf(" Name[%d] : %.16s\n", i
, ve
->name
);
1028 examine_vd(i
, sb
, ve
->guid
);
1030 if (cnt
) printf("\n");
1033 static void examine_pds(struct ddf_super
*sb
)
1035 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1038 printf(" Physical Disks : %d\n", cnt
);
1040 for (i
=0 ; i
<cnt
; i
++) {
1041 struct phys_disk_entry
*pd
= &sb
->phys
->entries
[i
];
1042 int type
= __be16_to_cpu(pd
->type
);
1043 int state
= __be16_to_cpu(pd
->state
);
1045 printf(" PD GUID[%d] : ", i
); print_guid(pd
->guid
, 0);
1047 printf(" ref[%d] : %08x\n", i
,
1048 __be32_to_cpu(pd
->refnum
));
1049 printf(" mode[%d] : %s%s%s%s%s\n", i
,
1050 (type
&2) ? "active":"",
1051 (type
&4) ? "Global Spare":"",
1052 (type
&8) ? "spare" : "",
1053 (type
&16)? ", foreign" : "",
1054 (type
&32)? "pass-through" : "");
1055 printf(" state[%d] : %s%s%s%s%s%s%s\n", i
,
1056 (state
&1)? "Online": "Offline",
1057 (state
&2)? ", Failed": "",
1058 (state
&4)? ", Rebuilding": "",
1059 (state
&8)? ", in-transition": "",
1060 (state
&16)? ", SMART errors": "",
1061 (state
&32)? ", Unrecovered Read Errors": "",
1062 (state
&64)? ", Missing" : "");
1063 printf(" Avail Size[%d] : %llu K\n", i
,
1064 __be64_to_cpu(pd
->config_size
)>>1);
1065 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1066 if (dl
->disk
.refnum
== pd
->refnum
) {
1067 char *dv
= map_dev(dl
->major
, dl
->minor
, 0);
1069 printf(" Device[%d] : %s\n",
1077 static void examine_super_ddf(struct supertype
*st
, char *homehost
)
1079 struct ddf_super
*sb
= st
->sb
;
1081 printf(" Magic : %08x\n", __be32_to_cpu(sb
->anchor
.magic
));
1082 printf(" Version : %.8s\n", sb
->anchor
.revision
);
1083 printf("Controller GUID : "); print_guid(sb
->controller
.guid
, 0);
1085 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1087 printf(" Seq : %08x\n", __be32_to_cpu(sb
->active
->seq
));
1088 printf(" Redundant hdr : %s\n", sb
->secondary
.magic
== DDF_HEADER_MAGIC
1094 static void brief_examine_super_ddf(struct supertype
*st
)
1096 /* We just write a generic DDF ARRAY entry
1097 * The uuid is all hex, 6 groups of 4 bytes
1099 struct ddf_super
*ddf
= st
->sb
;
1101 printf("ARRAY /dev/ddf UUID=");
1102 for (i
= 0; i
< DDF_GUID_LEN
; i
++) {
1103 printf("%02x", ddf
->anchor
.guid
[i
]);
1104 if ((i
&3) == 0 && i
!= 0)
1110 static void detail_super_ddf(struct supertype
*st
, char *homehost
)
1113 * Could print DDF GUID
1114 * Need to find which array
1115 * If whole, briefly list all arrays
1120 static void brief_detail_super_ddf(struct supertype
*st
)
1122 /* FIXME I really need to know which array we are detailing.
1123 * Can that be stored in ddf_super??
1125 // struct ddf_super *ddf = st->sb;
1131 static int match_home_ddf(struct supertype
*st
, char *homehost
)
1133 /* It matches 'this' host if the controller is a
1134 * Linux-MD controller with vendor_data matching
1137 struct ddf_super
*ddf
= st
->sb
;
1138 int len
= strlen(homehost
);
1140 return (memcmp(ddf
->controller
.guid
, T10
, 8) == 0 &&
1141 len
< sizeof(ddf
->controller
.vendor_data
) &&
1142 memcmp(ddf
->controller
.vendor_data
, homehost
,len
) == 0 &&
1143 ddf
->controller
.vendor_data
[len
] == 0);
1146 static struct vd_config
*find_vdcr(struct ddf_super
*ddf
, int inst
)
1149 if (inst
< 0 || inst
> __be16_to_cpu(ddf
->virt
->populated_vdes
))
1151 for (v
= ddf
->conflist
; v
; v
= v
->next
)
1152 if (memcmp(v
->conf
.guid
,
1153 ddf
->virt
->entries
[inst
].guid
,
1159 static int find_phys(struct ddf_super
*ddf
, __u32 phys_refnum
)
1161 /* Find the entry in phys_disk which has the given refnum
1162 * and return it's index
1165 for (i
=0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
); i
++)
1166 if (ddf
->phys
->entries
[i
].refnum
== phys_refnum
)
1171 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4])
1173 /* The uuid returned here is used for:
1174 * uuid to put into bitmap file (Create, Grow)
1175 * uuid for backup header when saving critical section (Grow)
1176 * comparing uuids when re-adding a device into an array
1177 * For each of these we can make do with a truncated
1178 * or hashed uuid rather than the original, as long as
1180 * In each case the uuid required is that of the data-array,
1181 * not the device-set.
1182 * In the case of SVD we assume the BVD is of interest,
1183 * though that might be the case if a bitmap were made for
1184 * a mirrored SVD - worry about that later.
1185 * So we need to find the VD configuration record for the
1186 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1187 * The first 16 bytes of the sha1 of these is used.
1189 struct ddf_super
*ddf
= st
->sb
;
1190 struct vd_config
*vd
= find_vdcr(ddf
, ddf
->conf_num
);
1193 memset(uuid
, 0, sizeof (uuid
));
1196 struct sha1_ctx ctx
;
1197 sha1_init_ctx(&ctx
);
1198 sha1_process_bytes(&vd
->guid
, DDF_GUID_LEN
, &ctx
);
1199 if (vd
->sec_elmnt_count
> 1)
1200 sha1_process_bytes(&vd
->sec_elmnt_seq
, 1, &ctx
);
1201 sha1_finish_ctx(&ctx
, buf
);
1202 memcpy(uuid
, buf
, sizeof(uuid
));
1206 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
)
1208 struct ddf_super
*ddf
= st
->sb
;
1211 info
->array
.major_version
= 1000;
1212 info
->array
.minor_version
= 0; /* FIXME use ddf->revision somehow */
1213 info
->array
.patch_version
= 0;
1214 info
->array
.raid_disks
= __be16_to_cpu(ddf
->phys
->used_pdes
);
1215 info
->array
.level
= LEVEL_CONTAINER
;
1216 info
->array
.layout
= 0;
1217 info
->array
.md_minor
= -1;
1218 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*(__u32
*)
1219 (ddf
->anchor
.guid
+16));
1220 info
->array
.utime
= 0;
1221 info
->array
.chunk_size
= 0;
1223 // info->data_offset = ???;
1224 // info->component_size = ???;
1226 info
->disk
.major
= 0;
1227 info
->disk
.minor
= 0;
1229 info
->disk
.number
= __be32_to_cpu(ddf
->dlist
->disk
.refnum
);
1230 info
->disk
.raid_disk
= -1;
1231 for (i
= 0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
) ; i
++)
1232 if (ddf
->phys
->entries
[i
].refnum
==
1233 ddf
->dlist
->disk
.refnum
) {
1234 info
->disk
.raid_disk
= i
;
1238 info
->disk
.number
= -1;
1239 // info->disk.raid_disk = find refnum in the table and use index;
1241 info
->disk
.state
= (1 << MD_DISK_SYNC
);
1243 info
->reshape_active
= 0;
1245 strcpy(info
->text_version
, "ddf");
1247 // uuid_from_super_ddf(info->uuid, sbv);
1249 // info->name[] ?? ;
1252 static void getinfo_super_n_container(struct supertype
*st
, struct mdinfo
*info
)
1254 /* just need offset and size */
1255 struct ddf_super
*ddf
= st
->sb
;
1256 int n
= info
->disk
.number
;
1258 info
->data_offset
= __be64_to_cpu(ddf
->phys
->entries
[n
].config_size
);
1259 info
->component_size
= 32*1024*1024 / 512;
1262 static int rlq_to_layout(int rlq
, int prl
, int raiddisks
);
1264 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
)
1266 struct ddf_super
*ddf
= st
->sb
;
1267 struct vd_config
*vd
= find_vdcr(ddf
, info
->container_member
);
1269 /* FIXME this returns BVD info - what if we want SVD ?? */
1271 info
->array
.major_version
= 1000;
1272 info
->array
.minor_version
= 0; /* FIXME use ddf->revision somehow */
1273 info
->array
.patch_version
= 0;
1274 info
->array
.raid_disks
= __be16_to_cpu(vd
->prim_elmnt_count
);
1275 info
->array
.level
= map_num1(ddf_level_num
, vd
->prl
);
1276 info
->array
.layout
= rlq_to_layout(vd
->rlq
, vd
->prl
,
1277 info
->array
.raid_disks
);
1278 info
->array
.md_minor
= -1;
1279 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*(__u32
*)(vd
->guid
+16));
1280 info
->array
.utime
= DECADE
+ __be32_to_cpu(vd
->timestamp
);
1281 info
->array
.chunk_size
= 512 << vd
->chunk_shift
;
1283 // info->data_offset = ???;
1284 // info->component_size = ???;
1286 info
->disk
.major
= 0;
1287 info
->disk
.minor
= 0;
1288 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1289 // info->disk.raid_disk = find refnum in the table and use index;
1290 // info->disk.state = ???;
1292 info
->resync_start
= 0;
1293 if (!(ddf
->virt
->entries
[info
->container_member
].state
1294 & DDF_state_inconsistent
) &&
1295 (ddf
->virt
->entries
[info
->container_member
].init_state
1296 & DDF_initstate_mask
)
1298 info
->resync_start
= ~0ULL;
1300 uuid_from_super_ddf(st
, info
->uuid
);
1302 sprintf(info
->text_version
, "/%s/%d",
1303 devnum2devname(st
->container_dev
),
1304 info
->container_member
);
1306 // info->name[] ?? ;
1309 static void getinfo_super_n_bvd(struct supertype
*st
, struct mdinfo
*info
)
1311 /* Find the particular details for info->disk.raid_disk.
1312 * This includes data_offset, component_size,
1314 struct ddf_super
*ddf
= st
->sb
;
1315 __u64
*lba_offset
= ddf
->newconf
->lba_offset
;
1316 struct vd_config
*conf
= &ddf
->newconf
->conf
;
1317 info
->data_offset
= __be64_to_cpu(lba_offset
[info
->disk
.raid_disk
]);
1318 info
->component_size
= __be64_to_cpu(conf
->blocks
);
1321 static int update_super_ddf(struct supertype
*st
, struct mdinfo
*info
,
1323 char *devname
, int verbose
,
1324 int uuid_set
, char *homehost
)
1326 /* For 'assemble' and 'force' we need to return non-zero if any
1327 * change was made. For others, the return value is ignored.
1328 * Update options are:
1329 * force-one : This device looks a bit old but needs to be included,
1330 * update age info appropriately.
1331 * assemble: clear any 'faulty' flag to allow this device to
1333 * force-array: Array is degraded but being forced, mark it clean
1334 * if that will be needed to assemble it.
1336 * newdev: not used ????
1337 * grow: Array has gained a new device - this is currently for
1339 * resync: mark as dirty so a resync will happen.
1340 * uuid: Change the uuid of the array to match watch is given
1341 * homehost: update the recorded homehost
1342 * name: update the name - preserving the homehost
1343 * _reshape_progress: record new reshape_progress position.
1345 * Following are not relevant for this version:
1346 * sparc2.2 : update from old dodgey metadata
1347 * super-minor: change the preferred_minor number
1348 * summaries: update redundant counters.
1351 // struct ddf_super *ddf = st->sb;
1352 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1353 // struct virtual_entry *ve = find_ve(ddf);
1356 /* we don't need to handle "force-*" or "assemble" as
1357 * there is no need to 'trick' the kernel. We the metadata is
1358 * first updated to activate the array, all the implied modifications
1362 if (strcmp(update
, "grow") == 0) {
1365 if (strcmp(update
, "resync") == 0) {
1366 // info->resync_checkpoint = 0;
1368 /* We ignore UUID updates as they make even less sense
1371 if (strcmp(update
, "homehost") == 0) {
1372 /* homehost is stored in controller->vendor_data,
1373 * or it is when we are the vendor
1375 // if (info->vendor_is_local)
1376 // strcpy(ddf->controller.vendor_data, homehost);
1378 if (strcmp(update
, "name") == 0) {
1379 /* name is stored in virtual_entry->name */
1380 // memset(ve->name, ' ', 16);
1381 // strncpy(ve->name, info->name, 16);
1383 if (strcmp(update
, "_reshape_progress") == 0) {
1384 /* We don't support reshape yet */
1387 // update_all_csum(ddf);
1392 static void make_header_guid(char *guid
)
1396 /* Create a DDF Header of Virtual Disk GUID */
1398 /* 24 bytes of fiction required.
1399 * first 8 are a 'vendor-id' - "Linux-MD"
1400 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1401 * Remaining 8 random number plus timestamp
1403 memcpy(guid
, T10
, sizeof(T10
));
1404 stamp
= __cpu_to_be32(0xdeadbeef);
1405 memcpy(guid
+8, &stamp
, 4);
1406 stamp
= __cpu_to_be32(0);
1407 memcpy(guid
+12, &stamp
, 4);
1408 stamp
= __cpu_to_be32(time(0) - DECADE
);
1409 memcpy(guid
+16, &stamp
, 4);
1410 rfd
= open("/dev/urandom", O_RDONLY
);
1411 if (rfd
< 0 || read(rfd
, &stamp
, 4) != 4)
1413 memcpy(guid
+20, &stamp
, 4);
1414 if (rfd
>= 0) close(rfd
);
1416 static int init_super_ddf(struct supertype
*st
,
1417 mdu_array_info_t
*info
,
1418 unsigned long long size
, char *name
, char *homehost
,
1421 /* This is primarily called by Create when creating a new array.
1422 * We will then get add_to_super called for each component, and then
1423 * write_init_super called to write it out to each device.
1424 * For DDF, Create can create on fresh devices or on a pre-existing
1426 * To create on a pre-existing array a different method will be called.
1427 * This one is just for fresh drives.
1429 * We need to create the entire 'ddf' structure which includes:
1430 * DDF headers - these are easy.
1431 * Controller data - a Sector describing this controller .. not that
1432 * this is a controller exactly.
1433 * Physical Disk Record - one entry per device, so
1434 * leave plenty of space.
1435 * Virtual Disk Records - again, just leave plenty of space.
1436 * This just lists VDs, doesn't give details
1437 * Config records - describes the VDs that use this disk
1438 * DiskData - describes 'this' device.
1439 * BadBlockManagement - empty
1440 * Diag Space - empty
1441 * Vendor Logs - Could we put bitmaps here?
1444 struct ddf_super
*ddf
;
1447 int max_phys_disks
, max_virt_disks
;
1448 unsigned long long sector
;
1452 struct phys_disk
*pd
;
1453 struct virtual_disk
*vd
;
1455 ddf
= malloc(sizeof(*ddf
));
1456 ddf
->dlist
= NULL
; /* no physical disks yet */
1457 ddf
->conflist
= NULL
; /* No virtual disks yet */
1459 /* At least 32MB *must* be reserved for the ddf. So let's just
1460 * start 32MB from the end, and put the primary header there.
1461 * Don't do secondary for now.
1462 * We don't know exactly where that will be yet as it could be
1463 * different on each device. To just set up the lengths.
1467 ddf
->anchor
.magic
= DDF_HEADER_MAGIC
;
1468 make_header_guid(ddf
->anchor
.guid
);
1470 memcpy(ddf
->anchor
.revision
, DDF_REVISION
, 8);
1471 ddf
->anchor
.seq
= __cpu_to_be32(1);
1472 ddf
->anchor
.timestamp
= __cpu_to_be32(time(0) - DECADE
);
1473 ddf
->anchor
.openflag
= 0xFF;
1474 ddf
->anchor
.foreignflag
= 0;
1475 ddf
->anchor
.enforcegroups
= 0; /* Is this best?? */
1476 ddf
->anchor
.pad0
= 0xff;
1477 memset(ddf
->anchor
.pad1
, 0xff, 12);
1478 memset(ddf
->anchor
.header_ext
, 0xff, 32);
1479 ddf
->anchor
.primary_lba
= ~(__u64
)0;
1480 ddf
->anchor
.secondary_lba
= ~(__u64
)0;
1481 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
1482 memset(ddf
->anchor
.pad2
, 0xff, 3);
1483 ddf
->anchor
.workspace_len
= __cpu_to_be32(32768); /* Must be reserved */
1484 ddf
->anchor
.workspace_lba
= ~(__u64
)0; /* Put this at bottom
1485 of 32M reserved.. */
1486 max_phys_disks
= 1023; /* Should be enough */
1487 ddf
->anchor
.max_pd_entries
= __cpu_to_be16(max_phys_disks
);
1488 max_virt_disks
= 255;
1489 ddf
->anchor
.max_vd_entries
= __cpu_to_be16(max_virt_disks
); /* ?? */
1490 ddf
->anchor
.max_partitions
= __cpu_to_be16(64); /* ?? */
1492 ddf
->conf_rec_len
= 1 + 256 * 12 / 512;
1493 ddf
->anchor
.config_record_len
= __cpu_to_be16(ddf
->conf_rec_len
);
1494 ddf
->anchor
.max_primary_element_entries
= __cpu_to_be16(256);
1496 memset(ddf
->anchor
.pad3
, 0xff, 54);
1498 /* controller sections is one sector long immediately
1499 * after the ddf header */
1501 ddf
->anchor
.controller_section_offset
= __cpu_to_be32(sector
);
1502 ddf
->anchor
.controller_section_length
= __cpu_to_be32(1);
1505 /* phys is 8 sectors after that */
1506 pdsize
= ROUND_UP(sizeof(struct phys_disk
) +
1507 sizeof(struct phys_disk_entry
)*max_phys_disks
,
1509 switch(pdsize
/512) {
1510 case 2: case 8: case 32: case 128: case 512: break;
1513 ddf
->anchor
.phys_section_offset
= __cpu_to_be32(sector
);
1514 ddf
->anchor
.phys_section_length
=
1515 __cpu_to_be32(pdsize
/512); /* max_primary_element_entries/8 */
1516 sector
+= pdsize
/512;
1518 /* virt is another 32 sectors */
1519 vdsize
= ROUND_UP(sizeof(struct virtual_disk
) +
1520 sizeof(struct virtual_entry
) * max_virt_disks
,
1522 switch(vdsize
/512) {
1523 case 2: case 8: case 32: case 128: case 512: break;
1526 ddf
->anchor
.virt_section_offset
= __cpu_to_be32(sector
);
1527 ddf
->anchor
.virt_section_length
=
1528 __cpu_to_be32(vdsize
/512); /* max_vd_entries/8 */
1529 sector
+= vdsize
/512;
1531 clen
= (1 + 256*12/512) * (64+1);
1532 ddf
->anchor
.config_section_offset
= __cpu_to_be32(sector
);
1533 ddf
->anchor
.config_section_length
= __cpu_to_be32(clen
);
1536 ddf
->anchor
.data_section_offset
= __cpu_to_be32(sector
);
1537 ddf
->anchor
.data_section_length
= __cpu_to_be32(1);
1540 ddf
->anchor
.bbm_section_length
= __cpu_to_be32(0);
1541 ddf
->anchor
.bbm_section_offset
= __cpu_to_be32(0xFFFFFFFF);
1542 ddf
->anchor
.diag_space_length
= __cpu_to_be32(0);
1543 ddf
->anchor
.diag_space_offset
= __cpu_to_be32(0xFFFFFFFF);
1544 ddf
->anchor
.vendor_length
= __cpu_to_be32(0);
1545 ddf
->anchor
.vendor_offset
= __cpu_to_be32(0xFFFFFFFF);
1547 memset(ddf
->anchor
.pad4
, 0xff, 256);
1549 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
1550 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
1552 ddf
->primary
.openflag
= 1; /* I guess.. */
1553 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
1555 ddf
->secondary
.openflag
= 1; /* I guess.. */
1556 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
1558 ddf
->active
= &ddf
->primary
;
1560 ddf
->controller
.magic
= DDF_CONTROLLER_MAGIC
;
1562 /* 24 more bytes of fiction required.
1563 * first 8 are a 'vendor-id' - "Linux-MD"
1564 * Remaining 16 are serial number.... maybe a hostname would do?
1566 memcpy(ddf
->controller
.guid
, T10
, sizeof(T10
));
1567 gethostname(hostname
, sizeof(hostname
));
1568 hostname
[sizeof(hostname
) - 1] = 0;
1569 hostlen
= strlen(hostname
);
1570 memcpy(ddf
->controller
.guid
+ 24 - hostlen
, hostname
, hostlen
);
1571 for (i
= strlen(T10
) ; i
+hostlen
< 24; i
++)
1572 ddf
->controller
.guid
[i
] = ' ';
1574 ddf
->controller
.type
.vendor_id
= __cpu_to_be16(0xDEAD);
1575 ddf
->controller
.type
.device_id
= __cpu_to_be16(0xBEEF);
1576 ddf
->controller
.type
.sub_vendor_id
= 0;
1577 ddf
->controller
.type
.sub_device_id
= 0;
1578 memcpy(ddf
->controller
.product_id
, "What Is My PID??", 16);
1579 memset(ddf
->controller
.pad
, 0xff, 8);
1580 memset(ddf
->controller
.vendor_data
, 0xff, 448);
1582 pd
= ddf
->phys
= malloc(pdsize
);
1583 ddf
->pdsize
= pdsize
;
1585 memset(pd
, 0xff, pdsize
);
1586 memset(pd
, 0, sizeof(*pd
));
1587 pd
->magic
= DDF_PHYS_DATA_MAGIC
;
1588 pd
->used_pdes
= __cpu_to_be16(0);
1589 pd
->max_pdes
= __cpu_to_be16(max_phys_disks
);
1590 memset(pd
->pad
, 0xff, 52);
1592 vd
= ddf
->virt
= malloc(vdsize
);
1593 ddf
->vdsize
= vdsize
;
1594 memset(vd
, 0, vdsize
);
1595 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
1596 vd
->populated_vdes
= __cpu_to_be16(0);
1597 vd
->max_vdes
= __cpu_to_be16(max_virt_disks
);
1598 memset(vd
->pad
, 0xff, 52);
1600 for (i
=0; i
<max_virt_disks
; i
++)
1601 memset(&vd
->entries
[i
], 0xff, sizeof(struct virtual_entry
));
1607 static int all_ff(char *guid
)
1610 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
1611 if (guid
[i
] != (char)0xff)
1615 static int chunk_to_shift(int chunksize
)
1617 return ffs(chunksize
/512)-1;
1620 static int level_to_prl(int level
)
1623 case LEVEL_LINEAR
: return DDF_CONCAT
;
1624 case 0: return DDF_RAID0
;
1625 case 1: return DDF_RAID1
;
1626 case 4: return DDF_RAID4
;
1627 case 5: return DDF_RAID5
;
1628 case 6: return DDF_RAID6
;
1632 static int layout_to_rlq(int level
, int layout
, int raiddisks
)
1636 return DDF_RAID0_SIMPLE
;
1639 case 2: return DDF_RAID1_SIMPLE
;
1640 case 3: return DDF_RAID1_MULTI
;
1645 case 0: return DDF_RAID4_N
;
1651 case ALGORITHM_LEFT_ASYMMETRIC
:
1652 return DDF_RAID5_N_RESTART
;
1653 case ALGORITHM_RIGHT_ASYMMETRIC
:
1654 return DDF_RAID5_0_RESTART
;
1655 case ALGORITHM_LEFT_SYMMETRIC
:
1656 return DDF_RAID5_N_CONTINUE
;
1657 case ALGORITHM_RIGHT_SYMMETRIC
:
1658 return -1; /* not mentioned in standard */
1664 static int rlq_to_layout(int rlq
, int prl
, int raiddisks
)
1668 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1670 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1678 return -1; /* FIXME this isn't checked */
1683 case DDF_RAID5_N_RESTART
:
1684 return ALGORITHM_LEFT_ASYMMETRIC
;
1685 case DDF_RAID5_0_RESTART
:
1686 return ALGORITHM_RIGHT_ASYMMETRIC
;
1687 case DDF_RAID5_N_CONTINUE
:
1688 return ALGORITHM_LEFT_SYMMETRIC
;
1696 static int init_super_ddf_bvd(struct supertype
*st
,
1697 mdu_array_info_t
*info
,
1698 unsigned long long size
,
1699 char *name
, char *homehost
,
1702 /* We are creating a BVD inside a pre-existing container.
1703 * so st->sb is already set.
1704 * We need to create a new vd_config and a new virtual_entry
1706 struct ddf_super
*ddf
= st
->sb
;
1708 struct virtual_entry
*ve
;
1710 struct vd_config
*vc
;
1712 if (__be16_to_cpu(ddf
->virt
->populated_vdes
)
1713 >= __be16_to_cpu(ddf
->virt
->max_vdes
)) {
1714 fprintf(stderr
, Name
": This ddf already has the "
1715 "maximum of %d virtual devices\n",
1716 __be16_to_cpu(ddf
->virt
->max_vdes
));
1720 for (venum
= 0; venum
< __be16_to_cpu(ddf
->virt
->max_vdes
); venum
++)
1721 if (all_ff(ddf
->virt
->entries
[venum
].guid
))
1723 if (venum
== __be16_to_cpu(ddf
->virt
->max_vdes
)) {
1724 fprintf(stderr
, Name
": Cannot find spare slot for "
1725 "virtual disk - DDF is corrupt\n");
1728 ve
= &ddf
->virt
->entries
[venum
];
1729 ddf
->conf_num
= venum
;
1731 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1732 * timestamp, random number
1734 make_header_guid(ve
->guid
);
1735 ve
->unit
= __cpu_to_be16(info
->md_minor
);
1737 ve
->guid_crc
= crc32(0, (unsigned char*)ddf
->anchor
.guid
, DDF_GUID_LEN
);
1739 ve
->state
= DDF_state_degraded
; /* Will be modified as devices are added */
1740 if (info
->state
& 1) /* clean */
1741 ve
->init_state
= DDF_init_full
;
1743 ve
->init_state
= DDF_init_not
;
1745 memset(ve
->pad1
, 0xff, 14);
1746 memset(ve
->name
, ' ', 16);
1748 strncpy(ve
->name
, name
, 16);
1749 ddf
->virt
->populated_vdes
=
1750 __cpu_to_be16(__be16_to_cpu(ddf
->virt
->populated_vdes
)+1);
1752 /* Now create a new vd_config */
1753 vcl
= malloc(offsetof(struct vcl
, conf
) + ddf
->conf_rec_len
* 512);
1754 vcl
->lba_offset
= (__u64
*) &vcl
->conf
.phys_refnum
[ddf
->mppe
];
1758 vc
->magic
= DDF_VD_CONF_MAGIC
;
1759 memcpy(vc
->guid
, ve
->guid
, DDF_GUID_LEN
);
1760 vc
->timestamp
= __cpu_to_be32(time(0)-DECADE
);
1761 vc
->seqnum
= __cpu_to_be32(1);
1762 memset(vc
->pad0
, 0xff, 24);
1763 vc
->prim_elmnt_count
= __cpu_to_be16(info
->raid_disks
);
1764 vc
->chunk_shift
= chunk_to_shift(info
->chunk_size
);
1765 vc
->prl
= level_to_prl(info
->level
);
1766 vc
->rlq
= layout_to_rlq(info
->level
, info
->layout
, info
->raid_disks
);
1767 vc
->sec_elmnt_count
= 1;
1768 vc
->sec_elmnt_seq
= 0;
1770 vc
->blocks
= __cpu_to_be64(info
->size
* 2);
1771 vc
->array_blocks
= __cpu_to_be64(
1772 calc_array_size(info
->level
, info
->raid_disks
, info
->layout
,
1773 info
->chunk_size
, info
->size
*2));
1774 memset(vc
->pad1
, 0xff, 8);
1775 vc
->spare_refs
[0] = 0xffffffff;
1776 vc
->spare_refs
[1] = 0xffffffff;
1777 vc
->spare_refs
[2] = 0xffffffff;
1778 vc
->spare_refs
[3] = 0xffffffff;
1779 vc
->spare_refs
[4] = 0xffffffff;
1780 vc
->spare_refs
[5] = 0xffffffff;
1781 vc
->spare_refs
[6] = 0xffffffff;
1782 vc
->spare_refs
[7] = 0xffffffff;
1783 memset(vc
->cache_pol
, 0, 8);
1785 memset(vc
->pad2
, 0xff, 3);
1786 memset(vc
->pad3
, 0xff, 52);
1787 memset(vc
->pad4
, 0xff, 192);
1788 memset(vc
->v0
, 0xff, 32);
1789 memset(vc
->v1
, 0xff, 32);
1790 memset(vc
->v2
, 0xff, 16);
1791 memset(vc
->v3
, 0xff, 16);
1792 memset(vc
->vendor
, 0xff, 32);
1794 memset(vc
->phys_refnum
, 0xff, 4*ddf
->mppe
);
1795 memset(vc
->phys_refnum
+(ddf
->mppe
* 4), 0x00, 8*ddf
->mppe
);
1797 vcl
->next
= ddf
->conflist
;
1798 ddf
->conflist
= vcl
;
1803 static void add_to_super_ddf_bvd(struct supertype
*st
,
1804 mdu_disk_info_t
*dk
, int fd
, char *devname
)
1806 /* fd and devname identify a device with-in the ddf container (st).
1807 * dk identifies a location in the new BVD.
1808 * We need to find suitable free space in that device and update
1809 * the phys_refnum and lba_offset for the newly created vd_config.
1810 * We might also want to update the type in the phys_disk
1814 struct ddf_super
*ddf
= st
->sb
;
1815 struct vd_config
*vc
;
1821 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
1822 if (dl
->major
== dk
->major
&&
1823 dl
->minor
== dk
->minor
)
1825 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1828 vc
= &ddf
->newconf
->conf
;
1829 lba_offset
= ddf
->newconf
->lba_offset
;
1830 vc
->phys_refnum
[dk
->raid_disk
] = dl
->disk
.refnum
;
1831 lba_offset
[dk
->raid_disk
] = 0; /* FIXME */
1833 for (i
=0; i
< ddf
->max_part
; i
++)
1834 if (dl
->vlist
[i
] == NULL
)
1836 if (i
== ddf
->max_part
)
1838 dl
->vlist
[i
] = ddf
->newconf
;
1841 dl
->devname
= devname
;
1843 /* Check how many working raid_disks, and if we can mark
1844 * array as optimal yet
1848 for (i
=0; i
< __be16_to_cpu(vc
->prim_elmnt_count
); i
++)
1849 if (vc
->phys_refnum
[i
] != 0xffffffff)
1851 /* Find which virtual_entry */
1852 max_virt_disks
= __be16_to_cpu(ddf
->active
->max_vd_entries
);
1853 for (i
=0; i
< max_virt_disks
; i
++)
1854 if (memcmp(ddf
->virt
->entries
[i
].guid
,
1855 vc
->guid
, DDF_GUID_LEN
)==0)
1857 if (i
== max_virt_disks
)
1859 if (working
== __be16_to_cpu(vc
->prim_elmnt_count
))
1860 ddf
->virt
->entries
[i
].state
=
1861 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
1862 | DDF_state_optimal
;
1864 if (vc
->prl
== DDF_RAID6
&&
1865 working
+1 == __be16_to_cpu(vc
->prim_elmnt_count
))
1866 ddf
->virt
->entries
[i
].state
=
1867 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
1868 | DDF_state_part_optimal
;
1870 ddf
->phys
->entries
[dl
->pdnum
].type
&= ~__cpu_to_be16(DDF_Global_Spare
);
1871 ddf
->phys
->entries
[dl
->pdnum
].type
|= __cpu_to_be16(DDF_Active_in_VD
);
1874 /* add a device to a container, either while creating it or while
1875 * expanding a pre-existing container
1877 static void add_to_super_ddf(struct supertype
*st
,
1878 mdu_disk_info_t
*dk
, int fd
, char *devname
)
1880 struct ddf_super
*ddf
= st
->sb
;
1884 unsigned long long size
;
1885 struct phys_disk_entry
*pde
;
1889 /* This is device numbered dk->number. We need to create
1890 * a phys_disk entry and a more detailed disk_data entry.
1893 dd
= malloc(sizeof(*dd
) + sizeof(dd
->vlist
[0]) * ddf
->max_part
);
1894 dd
->major
= major(stb
.st_rdev
);
1895 dd
->minor
= minor(stb
.st_rdev
);
1896 dd
->devname
= devname
;
1897 dd
->next
= ddf
->dlist
;
1901 dd
->disk
.magic
= DDF_PHYS_DATA_MAGIC
;
1903 tm
= localtime(&now
);
1904 sprintf(dd
->disk
.guid
, "%8s%04d%02d%02d",
1905 T10
, tm
->tm_year
+1900, tm
->tm_mon
+1, tm
->tm_mday
);
1906 *(__u32
*)(dd
->disk
.guid
+ 16) = random();
1907 *(__u32
*)(dd
->disk
.guid
+ 20) = random();
1909 dd
->disk
.refnum
= random(); /* and hope for the best FIXME check this is unique!!*/
1910 dd
->disk
.forced_ref
= 1;
1911 dd
->disk
.forced_guid
= 1;
1912 memset(dd
->disk
.vendor
, ' ', 32);
1913 memcpy(dd
->disk
.vendor
, "Linux", 5);
1914 memset(dd
->disk
.pad
, 0xff, 442);
1915 for (i
= 0; i
< ddf
->max_part
; i
++)
1916 dd
->vlist
[i
] = NULL
;
1918 n
= __be16_to_cpu(ddf
->phys
->used_pdes
);
1919 pde
= &ddf
->phys
->entries
[n
];
1923 ddf
->phys
->used_pdes
= __cpu_to_be16(n
);
1925 memcpy(pde
->guid
, dd
->disk
.guid
, DDF_GUID_LEN
);
1926 pde
->refnum
= dd
->disk
.refnum
;
1927 pde
->type
= __cpu_to_be16(DDF_Forced_PD_GUID
| DDF_Global_Spare
);
1928 pde
->state
= __cpu_to_be16(DDF_Online
);
1929 get_dev_size(fd
, NULL
, &size
);
1930 /* We are required to reserve 32Meg, and record the size in sectors */
1931 pde
->config_size
= __cpu_to_be64( (size
- 32*1024*1024) / 512);
1932 sprintf(pde
->path
, "%17.17s","Information: nil") ;
1933 memset(pde
->pad
, 0xff, 6);
1939 * This is the write_init_super method for a ddf container. It is
1940 * called when creating a container or adding another device to a
1945 static int __write_init_super_ddf(struct supertype
*st
, int do_close
)
1948 struct ddf_super
*ddf
= st
->sb
;
1954 unsigned long long size
, sector
;
1956 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
1962 /* We need to fill in the primary, (secondary) and workspace
1963 * lba's in the headers, set their checksums,
1964 * Also checksum phys, virt....
1966 * Then write everything out, finally the anchor is written.
1968 get_dev_size(fd
, NULL
, &size
);
1970 ddf
->anchor
.workspace_lba
= __cpu_to_be64(size
- 32*1024*2);
1971 ddf
->anchor
.primary_lba
= __cpu_to_be64(size
- 16*1024*2);
1972 ddf
->anchor
.seq
= __cpu_to_be32(1);
1973 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
1974 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
1976 ddf
->anchor
.openflag
= 0xFF; /* 'open' means nothing */
1977 ddf
->anchor
.seq
= 0xFFFFFFFF; /* no sequencing in anchor */
1978 ddf
->anchor
.crc
= calc_crc(&ddf
->anchor
, 512);
1980 ddf
->primary
.openflag
= 0;
1981 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
1983 ddf
->secondary
.openflag
= 0;
1984 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
1986 ddf
->primary
.crc
= calc_crc(&ddf
->primary
, 512);
1987 ddf
->secondary
.crc
= calc_crc(&ddf
->secondary
, 512);
1989 sector
= size
- 16*1024*2;
1990 lseek64(fd
, sector
<<9, 0);
1991 write(fd
, &ddf
->primary
, 512);
1993 ddf
->controller
.crc
= calc_crc(&ddf
->controller
, 512);
1994 write(fd
, &ddf
->controller
, 512);
1996 ddf
->phys
->crc
= calc_crc(ddf
->phys
, ddf
->pdsize
);
1998 write(fd
, ddf
->phys
, ddf
->pdsize
);
2000 ddf
->virt
->crc
= calc_crc(ddf
->virt
, ddf
->vdsize
);
2001 write(fd
, ddf
->virt
, ddf
->vdsize
);
2003 /* Now write lots of config records. */
2004 n_config
= ddf
->max_part
;
2005 conf_size
= ddf
->conf_rec_len
* 512;
2006 for (i
= 0 ; i
<= n_config
; i
++) {
2007 struct vcl
*c
= d
->vlist
[i
];
2009 c
= (struct vcl
*)d
->spare
;
2012 c
->conf
.crc
= calc_crc(&c
->conf
, conf_size
);
2013 write(fd
, &c
->conf
, conf_size
);
2015 __u32 sig
= 0xffffffff;
2017 lseek64(fd
, conf_size
-4, SEEK_CUR
);
2020 d
->disk
.crc
= calc_crc(&d
->disk
, 512);
2021 write(fd
, &d
->disk
, 512);
2023 /* Maybe do the same for secondary */
2025 lseek64(fd
, (size
-1)*512, SEEK_SET
);
2026 write(fd
, &ddf
->anchor
, 512);
2035 static int write_init_super_ddf(struct supertype
*st
)
2037 return __write_init_super_ddf(st
, 1);
2042 static __u64
avail_size_ddf(struct supertype
*st
, __u64 devsize
)
2044 /* We must reserve the last 32Meg */
2045 if (devsize
<= 32*1024*2)
2047 return devsize
- 32*1024*2;
2051 int validate_geometry_ddf(struct supertype
*st
,
2052 int level
, int layout
, int raiddisks
,
2053 int chunk
, unsigned long long size
,
2054 char *dev
, unsigned long long *freesize
)
2060 /* ddf potentially supports lots of things, but it depends on
2061 * what devices are offered (and maybe kernel version?)
2062 * If given unused devices, we will make a container.
2063 * If given devices in a container, we will make a BVD.
2064 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2067 if (level
== LEVEL_CONTAINER
) {
2068 st
->ss
= &super_ddf_container
;
2070 int rv
=st
->ss
->validate_geometry(st
, level
, layout
,
2077 return st
->ss
->validate_geometry(st
, level
, layout
, raiddisks
,
2078 chunk
, size
, dev
, freesize
);
2082 /* creating in a given container */
2083 st
->ss
= &super_ddf_bvd
;
2085 int rv
=st
->ss
->validate_geometry(st
, level
, layout
,
2092 return st
->ss
->validate_geometry(st
, level
, layout
, raiddisks
,
2093 chunk
, size
, dev
, freesize
);
2095 /* FIXME should exclude MULTIPATH, or more appropriately, allow
2096 * only known levels.
2101 /* This device needs to be either a device in a 'ddf' container,
2102 * or it needs to be a 'ddf-bvd' array.
2105 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2107 sra
= sysfs_read(fd
, 0, GET_VERSION
);
2109 if (sra
&& sra
->array
.major_version
== -1 &&
2110 strcmp(sra
->text_version
, "ddf-bvd") == 0) {
2111 st
->ss
= &super_ddf_svd
;
2112 return st
->ss
->validate_geometry(st
, level
, layout
,
2113 raiddisks
, chunk
, size
,
2118 Name
": Cannot create this array on device %s\n",
2122 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2123 fprintf(stderr
, Name
": Cannot open %s: %s\n",
2124 dev
, strerror(errno
));
2127 /* Well, it is in use by someone, maybe a 'ddf' container. */
2128 cfd
= open_container(fd
);
2131 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
2135 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2137 if (sra
&& sra
->array
.major_version
== -1 &&
2138 strcmp(sra
->text_version
, "ddf") == 0) {
2139 /* This is a member of a ddf container. Load the container
2140 * and try to create a bvd
2142 struct ddf_super
*ddf
;
2143 st
->ss
= &super_ddf_bvd
;
2144 if (load_super_ddf_all(st
, cfd
, (void **)&ddf
, NULL
, 1) == 0) {
2146 st
->container_dev
= fd2devnum(cfd
);
2148 return st
->ss
->validate_geometry(st
, level
, layout
,
2149 raiddisks
, chunk
, size
,
2153 } else /* device may belong to a different container */
2159 int validate_geometry_ddf_container(struct supertype
*st
,
2160 int level
, int layout
, int raiddisks
,
2161 int chunk
, unsigned long long size
,
2162 char *dev
, unsigned long long *freesize
)
2165 unsigned long long ldsize
;
2167 if (level
!= LEVEL_CONTAINER
)
2172 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2174 fprintf(stderr
, Name
": Cannot open %s: %s\n",
2175 dev
, strerror(errno
));
2178 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2184 *freesize
= avail_size_ddf(st
, ldsize
>> 9);
2190 unsigned long long start
, size
;
2192 int cmp_extent(const void *av
, const void *bv
)
2194 const struct extent
*a
= av
;
2195 const struct extent
*b
= bv
;
2196 if (a
->start
< b
->start
)
2198 if (a
->start
> b
->start
)
2203 struct extent
*get_extents(struct ddf_super
*ddf
, struct dl
*dl
)
2205 /* find a list of used extents on the give physical device
2206 * (dnum) of the given ddf.
2207 * Return a malloced array of 'struct extent'
2209 FIXME ignore DDF_Legacy devices?
2217 /* FIXME this is dl->pdnum */
2218 for (dnum
= 0; dnum
< ddf
->phys
->used_pdes
; dnum
++)
2219 if (memcmp(dl
->disk
.guid
,
2220 ddf
->phys
->entries
[dnum
].guid
,
2224 if (dnum
== ddf
->phys
->used_pdes
)
2227 rv
= malloc(sizeof(struct extent
) * (ddf
->max_part
+ 2));
2231 for (i
= 0; i
< ddf
->max_part
; i
++) {
2232 struct vcl
*v
= dl
->vlist
[i
];
2235 for (j
=0; j
< v
->conf
.prim_elmnt_count
; j
++)
2236 if (v
->conf
.phys_refnum
[j
] == dl
->disk
.refnum
) {
2237 /* This device plays role 'j' in 'v'. */
2238 rv
[n
].start
= __be64_to_cpu(v
->lba_offset
[j
]);
2239 rv
[n
].size
= __be64_to_cpu(v
->conf
.blocks
);
2244 qsort(rv
, n
, sizeof(*rv
), cmp_extent
);
2246 rv
[n
].start
= __be64_to_cpu(ddf
->phys
->entries
[dnum
].config_size
);
2251 int validate_geometry_ddf_bvd(struct supertype
*st
,
2252 int level
, int layout
, int raiddisks
,
2253 int chunk
, unsigned long long size
,
2254 char *dev
, unsigned long long *freesize
)
2257 struct ddf_super
*ddf
= st
->sb
;
2259 unsigned long long pos
= 0;
2260 unsigned long long maxsize
;
2263 /* ddf/bvd supports lots of things, but not containers */
2264 if (level
== LEVEL_CONTAINER
)
2266 /* We must have the container info already read in. */
2271 /* General test: make sure there is space for
2272 * 'raiddisks' device extents of size 'size'.
2274 unsigned long long minsize
= size
;
2278 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2284 e
= get_extents(ddf
, dl
);
2287 unsigned long long esize
;
2288 esize
= e
[i
].start
- pos
;
2289 if (esize
>= minsize
)
2291 pos
= e
[i
].start
+ e
[i
].size
;
2293 } while (e
[i
-1].size
);
2298 if (dcnt
< raiddisks
) {
2299 fprintf(stderr
, Name
": Not enough devices with space "
2300 "for this array (%d < %d)\n",
2306 /* This device must be a member of the set */
2307 if (stat(dev
, &stb
) < 0)
2309 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2311 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
2312 if (dl
->major
== major(stb
.st_rdev
) &&
2313 dl
->minor
== minor(stb
.st_rdev
))
2317 fprintf(stderr
, Name
": %s is not in the same DDF set\n",
2321 e
= get_extents(ddf
, dl
);
2325 unsigned long long esize
;
2326 esize
= e
[i
].start
- pos
;
2327 if (esize
>= maxsize
)
2329 pos
= e
[i
].start
+ e
[i
].size
;
2331 } while (e
[i
-1].size
);
2332 *freesize
= maxsize
;
2337 int validate_geometry_ddf_svd(struct supertype
*st
,
2338 int level
, int layout
, int raiddisks
,
2339 int chunk
, unsigned long long size
,
2340 char *dev
, unsigned long long *freesize
)
2342 /* dd/svd only supports striped, mirrored, concat, spanned... */
2343 if (level
!= LEVEL_LINEAR
&&
2351 static int load_super_ddf_all(struct supertype
*st
, int fd
,
2352 void **sbp
, char *devname
, int keep_fd
)
2355 struct ddf_super
*super
;
2356 struct mdinfo
*sd
, *best
= NULL
;
2362 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
2365 if (sra
->array
.major_version
!= -1 ||
2366 sra
->array
.minor_version
!= -2 ||
2367 strcmp(sra
->text_version
, "ddf") != 0)
2370 super
= malloc(sizeof(*super
));
2373 memset(super
, 0, sizeof(*super
));
2375 /* first, try each device, and choose the best ddf */
2376 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2378 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2379 dfd
= dev_open(nm
, O_RDONLY
);
2382 rv
= load_ddf_headers(dfd
, super
, NULL
);
2385 seq
= __be32_to_cpu(super
->active
->seq
);
2386 if (super
->active
->openflag
)
2388 if (!best
|| seq
> bestseq
) {
2396 /* OK, load this ddf */
2397 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
2398 dfd
= dev_open(nm
, O_RDONLY
);
2401 load_ddf_headers(dfd
, super
, NULL
);
2402 load_ddf_global(dfd
, super
, NULL
);
2404 /* Now we need the device-local bits */
2405 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2406 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2407 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2410 seq
= load_ddf_local(dfd
, super
, NULL
, keep_fd
);
2411 if (!keep_fd
) close(dfd
);
2414 if (st
->ss
== NULL
) {
2415 st
->ss
= &super_ddf_container
;
2416 st
->minor_version
= 0;
2418 st
->container_dev
= fd2devnum(fd
);
2426 static struct mdinfo
*container_content_ddf(struct supertype
*st
)
2428 /* Given a container loaded by load_super_ddf_all,
2429 * extract information about all the arrays into
2432 * For each vcl in conflist: create an mdinfo, fill it in,
2433 * then look for matching devices (phys_refnum) in dlist
2434 * and create appropriate device mdinfo.
2436 struct ddf_super
*ddf
= st
->sb
;
2437 struct mdinfo
*rest
= NULL
;
2440 for (vc
= ddf
->conflist
; vc
; vc
=vc
->next
)
2443 struct mdinfo
*this;
2444 this = malloc(sizeof(*this));
2445 memset(this, 0, sizeof(*this));
2449 this->array
.major_version
= 1000;
2450 this->array
.minor_version
= 0;
2451 this->array
.patch_version
= 0;
2452 this->array
.level
= map_num1(ddf_level_num
, vc
->conf
.prl
);
2453 this->array
.raid_disks
=
2454 __be16_to_cpu(vc
->conf
.prim_elmnt_count
);
2455 this->array
.layout
= rlq_to_layout(vc
->conf
.rlq
, vc
->conf
.prl
,
2456 this->array
.raid_disks
);
2457 this->array
.md_minor
= -1;
2458 this->array
.ctime
= DECADE
+
2459 __be32_to_cpu(*(__u32
*)(vc
->conf
.guid
+16));
2460 this->array
.utime
= DECADE
+
2461 __be32_to_cpu(vc
->conf
.timestamp
);
2462 this->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
2464 for (i
=0; i
< __be16_to_cpu(ddf
->virt
->populated_vdes
); i
++)
2465 if (memcmp(ddf
->virt
->entries
[i
].guid
,
2466 vc
->conf
.guid
, DDF_GUID_LEN
) == 0)
2468 if ((ddf
->virt
->entries
[i
].state
& DDF_state_inconsistent
) ||
2469 (ddf
->virt
->entries
[i
].init_state
& DDF_initstate_mask
) !=
2471 this->array
.state
= 0;
2472 this->resync_start
= 0;
2474 this->array
.state
= 1;
2475 this->resync_start
= ~0ULL;
2477 memcpy(this->name
, ddf
->virt
->entries
[i
].name
, 32);
2480 memset(this->uuid
, 0, sizeof(this->uuid
));
2481 this->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
2482 this->array
.size
= this->component_size
/ 2;
2483 this->container_member
= i
;
2485 sprintf(this->text_version
, "/%s/%d",
2486 devnum2devname(st
->container_dev
),
2487 this->container_member
);
2490 for (i
=0 ; i
< ddf
->mppe
; i
++) {
2494 if (vc
->conf
.phys_refnum
[i
] == 0xFFFFFFFF)
2497 this->array
.working_disks
++;
2499 for (d
= ddf
->dlist
; d
; d
=d
->next
)
2500 if (d
->disk
.refnum
== vc
->conf
.phys_refnum
[i
])
2505 dev
= malloc(sizeof(*dev
));
2506 memset(dev
, 0, sizeof(*dev
));
2507 dev
->next
= this->devs
;
2510 dev
->disk
.number
= __be32_to_cpu(d
->disk
.refnum
);
2511 dev
->disk
.major
= d
->major
;
2512 dev
->disk
.minor
= d
->minor
;
2513 dev
->disk
.raid_disk
= i
;
2514 dev
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
2516 dev
->events
= __le32_to_cpu(ddf
->primary
.seq
);
2517 dev
->data_offset
= vc
->lba_offset
[i
];
2518 dev
->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
2520 strcpy(dev
->name
, d
->devname
);
2526 static int init_zero_ddf(struct supertype
*st
,
2527 mdu_array_info_t
*info
,
2528 unsigned long long size
, char *name
,
2529 char *homehost
, int *uuid
)
2535 static int store_zero_ddf(struct supertype
*st
, int fd
)
2537 unsigned long long dsize
;
2539 memset(buf
, 0, 512);
2542 if (!get_dev_size(fd
, NULL
, &dsize
))
2545 lseek64(fd
, dsize
-512, 0);
2546 write(fd
, buf
, 512);
2550 static int compare_super_ddf(struct supertype
*st
, struct supertype
*tst
)
2554 * 0 same, or first was empty, and second was copied
2555 * 1 second had wrong number
2557 * 3 wrong other info
2559 struct ddf_super
*first
= st
->sb
;
2560 struct ddf_super
*second
= tst
->sb
;
2568 if (memcmp(first
->anchor
.guid
, second
->anchor
.guid
, DDF_GUID_LEN
) != 0)
2571 /* FIXME should I look at anything else? */
2576 * A new array 'a' has been started which claims to be instance 'inst'
2577 * within container 'c'.
2578 * We need to confirm that the array matches the metadata in 'c' so
2579 * that we don't corrupt any metadata.
2581 static int ddf_open_new(struct supertype
*c
, struct active_array
*a
, char *inst
)
2583 fprintf(stderr
, "ddf: open_new %s\n", inst
);
2584 a
->info
.container_member
= atoi(inst
);
2589 * The array 'a' is to be marked clean in the metadata.
2590 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2591 * clean up to the point (in sectors). If that cannot be recorded in the
2592 * metadata, then leave it as dirty.
2594 * For DDF, we need to clear the DDF_state_inconsistent bit in the
2595 * !global! virtual_disk.virtual_entry structure.
2597 static void ddf_set_array_state(struct active_array
*a
, int consistent
)
2599 struct ddf_super
*ddf
= a
->container
->sb
;
2600 int inst
= a
->info
.container_member
;
2602 ddf
->virt
->entries
[inst
].state
&= ~DDF_state_inconsistent
;
2604 ddf
->virt
->entries
[inst
].state
|= DDF_state_inconsistent
;
2605 ddf
->virt
->entries
[inst
].init_state
&= ~DDF_initstate_mask
;
2606 if (a
->resync_start
== ~0ULL)
2607 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_full
;
2608 else if (a
->resync_start
== 0)
2609 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_not
;
2611 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_quick
;
2613 printf("ddf mark %s %llu\n", consistent
?"clean":"dirty",
2618 * The state of each disk is stored in the global phys_disk structure
2619 * in phys_disk.entries[n].state.
2620 * This makes various combinations awkward.
2621 * - When a device fails in any array, it must be failed in all arrays
2622 * that include a part of this device.
2623 * - When a component is rebuilding, we cannot include it officially in the
2624 * array unless this is the only array that uses the device.
2626 * So: when transitioning:
2627 * Online -> failed, just set failed flag. monitor will propagate
2628 * spare -> online, the device might need to be added to the array.
2629 * spare -> failed, just set failed. Don't worry if in array or not.
2631 static void ddf_set_disk(struct active_array
*a
, int n
, int state
)
2633 struct ddf_super
*ddf
= a
->container
->sb
;
2634 int inst
= a
->info
.container_member
;
2635 struct vd_config
*vc
= find_vdcr(ddf
, inst
);
2636 int pd
= find_phys(ddf
, vc
->phys_refnum
[n
]);
2640 fprintf(stderr
, "ddf: cannot find instance %d!!\n", inst
);
2644 /* disk doesn't currently exist. If it is now in_sync,
2646 if ((state
& DS_INSYNC
) && ! (state
& DS_FAULTY
)) {
2647 /* Find dev 'n' in a->info->devs, determine the
2648 * ddf refnum, and set vc->phys_refnum and update
2654 if (state
& DS_FAULTY
)
2655 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Failed
);
2656 if (state
& DS_INSYNC
) {
2657 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Online
);
2658 ddf
->phys
->entries
[pd
].state
&= __cpu_to_be16(~DDF_Rebuilding
);
2662 fprintf(stderr
, "ddf: set_disk %d to %x\n", n
, state
);
2664 /* Now we need to check the state of the array and update
2665 * virtual_disk.entries[n].state.
2666 * It needs to be one of "optimal", "degraded", "failed".
2667 * I don't understand 'deleted' or 'missing'.
2670 for (i
=0; i
< a
->info
.array
.raid_disks
; i
++) {
2671 pd
= find_phys(ddf
, vc
->phys_refnum
[i
]);
2674 st
= __be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
2675 if ((st
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
2679 state
= DDF_state_degraded
;
2680 if (working
== a
->info
.array
.raid_disks
)
2681 state
= DDF_state_optimal
;
2682 else switch(vc
->prl
) {
2686 state
= DDF_state_failed
;
2690 state
= DDF_state_failed
;
2694 if (working
< a
->info
.array
.raid_disks
-1)
2695 state
= DDF_state_failed
;
2698 if (working
< a
->info
.array
.raid_disks
-2)
2699 state
= DDF_state_failed
;
2700 else if (working
== a
->info
.array
.raid_disks
-1)
2701 state
= DDF_state_part_optimal
;
2705 ddf
->virt
->entries
[inst
].state
=
2706 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
2711 static void ddf_sync_metadata(struct supertype
*st
)
2715 * Write all data to all devices.
2716 * Later, we might be able to track whether only local changes
2717 * have been made, or whether any global data has been changed,
2718 * but ddf is sufficiently weird that it probably always
2719 * changes global data ....
2721 __write_init_super_ddf(st
, 0);
2722 fprintf(stderr
, "ddf: sync_metadata\n");
2725 static void ddf_process_update(struct supertype
*st
,
2726 struct metadata_update
*update
)
2728 /* Apply this update to the metadata.
2729 * The first 4 bytes are a DDF_*_MAGIC which guides
2731 * Possible update are:
2732 * DDF_PHYS_RECORDS_MAGIC
2733 * Add a new physical device. Changes to this record
2734 * only happen implicitly.
2735 * used_pdes is the device number.
2736 * DDF_VIRT_RECORDS_MAGIC
2737 * Add a new VD. Possibly also change the 'access' bits.
2738 * populated_vdes is the entry number.
2740 * New or updated VD. the VIRT_RECORD must already
2741 * exist. For an update, phys_refnum and lba_offset
2742 * (at least) are updated, and the VD_CONF must
2743 * be written to precisely those devices listed with
2745 * DDF_SPARE_ASSIGN_MAGIC
2746 * replacement Spare Assignment Record... but for which device?
2749 * - to create a new array, we send a VIRT_RECORD and
2750 * a VD_CONF. Then assemble and start the array.
2751 * - to activate a spare we send a VD_CONF to add the phys_refnum
2752 * and offset. This will also mark the spare as active with
2753 * a spare-assignment record.
2755 struct ddf_super
*ddf
= st
->sb
;
2756 __u32
*magic
= (__u32
*)update
->buf
;
2757 struct phys_disk
*pd
;
2758 struct virtual_disk
*vd
;
2759 struct vd_config
*vc
;
2765 printf("Process update %x\n", *magic
);
2768 case DDF_PHYS_RECORDS_MAGIC
:
2770 if (update
->len
!= (sizeof(struct phys_disk
) +
2771 sizeof(struct phys_disk_entry
)))
2773 pd
= (struct phys_disk
*)update
->buf
;
2775 ent
= __be16_to_cpu(pd
->used_pdes
);
2776 if (ent
>= __be16_to_cpu(ddf
->phys
->max_pdes
))
2778 if (!all_ff(ddf
->phys
->entries
[ent
].guid
))
2780 ddf
->phys
->entries
[ent
] = pd
->entries
[0];
2781 ddf
->phys
->used_pdes
= __cpu_to_be16(1 +
2782 __be16_to_cpu(ddf
->phys
->used_pdes
));
2785 case DDF_VIRT_RECORDS_MAGIC
:
2787 if (update
->len
!= (sizeof(struct virtual_disk
) +
2788 sizeof(struct virtual_entry
)))
2790 vd
= (struct virtual_disk
*)update
->buf
;
2792 ent
= __be16_to_cpu(vd
->populated_vdes
);
2793 if (ent
>= __be16_to_cpu(ddf
->virt
->max_vdes
))
2795 if (!all_ff(ddf
->virt
->entries
[ent
].guid
))
2797 ddf
->virt
->entries
[ent
] = vd
->entries
[0];
2798 ddf
->virt
->populated_vdes
= __cpu_to_be16(1 +
2799 __be16_to_cpu(ddf
->virt
->populated_vdes
));
2802 case DDF_VD_CONF_MAGIC
:
2803 printf("len %d %d\n", update
->len
, ddf
->conf_rec_len
);
2805 mppe
= __be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
2806 if (update
->len
!= ddf
->conf_rec_len
)
2808 vc
= (struct vd_config
*)update
->buf
;
2809 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
2810 if (memcmp(vcl
->conf
.guid
, vc
->guid
, DDF_GUID_LEN
) == 0)
2812 printf("vcl = %p\n", vcl
);
2814 /* An update, just copy the phys_refnum and lba_offset
2817 memcpy(vcl
->conf
.phys_refnum
, vc
->phys_refnum
,
2818 mppe
* (sizeof(__u32
) + sizeof(__u64
)));
2821 vcl
= update
->space
;
2822 update
->space
= NULL
;
2823 vcl
->next
= ddf
->conflist
;
2825 vcl
->lba_offset
= (__u64
*)
2826 &vcl
->conf
.phys_refnum
[mppe
];
2827 ddf
->conflist
= vcl
;
2829 /* Now make sure vlist is correct for each dl. */
2830 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
2833 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
2834 for (dn
=0; dn
< ddf
->mppe
; dn
++)
2835 if (vcl
->conf
.phys_refnum
[dn
] ==
2837 printf("dev %d has %p at %d\n",
2838 dl
->pdnum
, vcl
, vn
);
2839 dl
->vlist
[vn
++] = vcl
;
2842 while (vn
< ddf
->max_part
)
2843 dl
->vlist
[vn
++] = NULL
;
2845 ddf
->phys
->entries
[dl
->pdnum
].type
&=
2846 ~__cpu_to_be16(DDF_Global_Spare
);
2847 ddf
->phys
->entries
[dl
->pdnum
].type
|=
2848 __cpu_to_be16(DDF_Active_in_VD
);
2851 ddf
->phys
->entries
[dl
->pdnum
].type
&=
2852 ~__cpu_to_be16(DDF_Global_Spare
);
2853 ddf
->phys
->entries
[dl
->pdnum
].type
|=
2854 __cpu_to_be16(DDF_Spare
);
2856 if (!dl
->vlist
[0] && !dl
->spare
) {
2857 ddf
->phys
->entries
[dl
->pdnum
].type
|=
2858 __cpu_to_be16(DDF_Global_Spare
);
2859 ddf
->phys
->entries
[dl
->pdnum
].type
&=
2860 ~__cpu_to_be16(DDF_Spare
|
2865 case DDF_SPARE_ASSIGN_MAGIC
:
2871 * Check if the array 'a' is degraded but not failed.
2872 * If it is, find as many spares as are available and needed and
2873 * arrange for their inclusion.
2874 * We only choose devices which are not already in the array,
2875 * and prefer those with a spare-assignment to this array.
2876 * otherwise we choose global spares - assuming always that
2877 * there is enough room.
2878 * For each spare that we assign, we return an 'mdinfo' which
2879 * describes the position for the device in the array.
2880 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
2881 * the new phys_refnum and lba_offset values.
2883 * Only worry about BVDs at the moment.
2885 static struct mdinfo
*ddf_activate_spare(struct active_array
*a
,
2886 struct metadata_update
**updates
)
2890 struct ddf_super
*ddf
= a
->container
->sb
;
2892 struct mdinfo
*rv
= NULL
;
2894 struct metadata_update
*mu
;
2897 struct vd_config
*vc
;
2900 /* FIXME, If there is a DS_FAULTY, we want to wait for it to be
2901 * removed. Then only look at DS_REMOVE devices.
2902 * What about !DS_INSYNC - how can that happen?
2904 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
2905 if ((d
->curr_state
& DS_FAULTY
) &&
2907 /* wait for Removal to happen */
2909 if (d
->state_fd
>= 0)
2913 printf("ddf_activate: working=%d (%d) level=%d\n", working
, a
->info
.array
.raid_disks
,
2914 a
->info
.array
.level
);
2915 if (working
== a
->info
.array
.raid_disks
)
2916 return NULL
; /* array not degraded */
2917 switch (a
->info
.array
.level
) {
2920 return NULL
; /* failed */
2924 if (working
< a
->info
.array
.raid_disks
- 1)
2925 return NULL
; /* failed */
2928 if (working
< a
->info
.array
.raid_disks
- 2)
2929 return NULL
; /* failed */
2931 default: /* concat or stripe */
2932 return NULL
; /* failed */
2935 /* For each slot, if it is not working, find a spare */
2937 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
2938 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2939 if (d
->disk
.raid_disk
== i
)
2941 printf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
2942 if (d
&& (d
->state_fd
>= 0))
2945 /* OK, this device needs recovery. Find a spare */
2947 for ( ; dl
; dl
= dl
->next
) {
2948 unsigned long long esize
;
2949 unsigned long long pos
;
2952 int is_dedicated
= 0;
2955 /* If in this array, skip */
2956 for (d2
= a
->info
.devs
; d2
; d2
= d2
->next
)
2957 if (d2
->disk
.major
== dl
->major
&&
2958 d2
->disk
.minor
== dl
->minor
) {
2959 printf("%x:%x already in array\n", dl
->major
, dl
->minor
);
2964 if (ddf
->phys
->entries
[dl
->pdnum
].type
&
2965 __cpu_to_be16(DDF_Spare
)) {
2966 /* Check spare assign record */
2968 if (dl
->spare
->type
& DDF_spare_dedicated
) {
2969 /* check spare_ents for guid */
2971 j
< __be16_to_cpu(dl
->spare
->populated
);
2973 if (memcmp(dl
->spare
->spare_ents
[j
].guid
,
2974 ddf
->virt
->entries
[a
->info
.container_member
].guid
,
2981 } else if (ddf
->phys
->entries
[dl
->pdnum
].type
&
2982 __cpu_to_be16(DDF_Global_Spare
)) {
2985 if ( ! (is_dedicated
||
2986 (is_global
&& global_ok
))) {
2987 printf("%x:%x not suitable: %d %d\n", dl
->major
, dl
->minor
,
2988 is_dedicated
, is_global
);
2992 /* We are allowed to use this device - is there space?
2993 * We need a->info.component_size sectors */
2994 ex
= get_extents(ddf
, dl
);
2996 printf("cannot get extents\n");
3003 esize
= ex
[j
].start
- pos
;
3004 if (esize
>= a
->info
.component_size
)
3006 pos
= ex
[i
].start
+ ex
[i
].size
;
3008 } while (ex
[i
-1].size
);
3011 if (esize
< a
->info
.component_size
) {
3012 printf("%x:%x has no room: %llu %llu\n", dl
->major
, dl
->minor
,
3013 esize
, a
->info
.component_size
);
3018 /* Cool, we have a device with some space at pos */
3019 di
= malloc(sizeof(*di
));
3020 memset(di
, 0, sizeof(*di
));
3021 di
->disk
.number
= i
;
3022 di
->disk
.raid_disk
= i
;
3023 di
->disk
.major
= dl
->major
;
3024 di
->disk
.minor
= dl
->minor
;
3026 di
->data_offset
= pos
;
3027 di
->component_size
= a
->info
.component_size
;
3028 di
->container_member
= dl
->pdnum
;
3031 printf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
3036 if (!dl
&& ! global_ok
) {
3037 /* not enough dedicated spares, try global */
3045 /* No spares found */
3047 /* Now 'rv' has a list of devices to return.
3048 * Create a metadata_update record to update the
3049 * phys_refnum and lba_offset values
3051 mu
= malloc(sizeof(*mu
));
3052 mu
->buf
= malloc(ddf
->conf_rec_len
* 512);
3053 mu
->space
= malloc(sizeof(struct vcl
));
3054 mu
->len
= ddf
->conf_rec_len
;
3055 mu
->next
= *updates
;
3056 vc
= find_vdcr(ddf
, a
->info
.container_member
);
3057 memcpy(mu
->buf
, vc
, ddf
->conf_rec_len
* 512);
3059 vc
= (struct vd_config
*)mu
->buf
;
3060 lba
= (__u64
*)&vc
->phys_refnum
[ddf
->mppe
];
3061 for (di
= rv
; di
; di
= di
->next
) {
3062 vc
->phys_refnum
[di
->disk
.raid_disk
] =
3063 ddf
->phys
->entries
[dl
->pdnum
].refnum
;
3064 lba
[di
->disk
.raid_disk
] = di
->data_offset
;
3070 struct superswitch super_ddf
= {
3072 .examine_super
= examine_super_ddf
,
3073 .brief_examine_super
= brief_examine_super_ddf
,
3074 .detail_super
= detail_super_ddf
,
3075 .brief_detail_super
= brief_detail_super_ddf
,
3076 .validate_geometry
= validate_geometry_ddf
,
3078 .match_home
= match_home_ddf
,
3079 .uuid_from_super
= uuid_from_super_ddf
,
3080 .getinfo_super
= getinfo_super_ddf
,
3081 .update_super
= update_super_ddf
,
3083 .avail_size
= avail_size_ddf
,
3085 .compare_super
= compare_super_ddf
,
3087 .load_super
= load_super_ddf
,
3088 .init_super
= init_zero_ddf
,
3089 .store_super
= store_zero_ddf
,
3090 .free_super
= free_super_ddf
,
3091 .match_metadata_desc
= match_metadata_desc_ddf
,
3092 .getinfo_super_n
= getinfo_super_n_container
,
3100 .open_new
= ddf_open_new
,
3101 .set_array_state
= ddf_set_array_state
,
3102 .set_disk
= ddf_set_disk
,
3103 .sync_metadata
= ddf_sync_metadata
,
3104 .process_update
= ddf_process_update
,
3105 .activate_spare
= ddf_activate_spare
,
3109 /* Super_ddf_container is set by validate_geometry_ddf when given a
3110 * device that is not part of any array
3112 struct superswitch super_ddf_container
= {
3114 .validate_geometry
= validate_geometry_ddf_container
,
3115 .write_init_super
= write_init_super_ddf
,
3118 .load_super
= load_super_ddf
,
3119 .init_super
= init_super_ddf
,
3120 .add_to_super
= add_to_super_ddf
,
3121 .getinfo_super
= getinfo_super_ddf
,
3123 .free_super
= free_super_ddf
,
3125 .container_content
= container_content_ddf
,
3126 .getinfo_super_n
= getinfo_super_n_container
,
3133 struct superswitch super_ddf_bvd
= {
3135 // .detail_super = detail_super_ddf_bvd,
3136 // .brief_detail_super = brief_detail_super_ddf_bvd,
3137 .validate_geometry
= validate_geometry_ddf_bvd
,
3138 .write_init_super
= write_init_super_ddf
,
3140 .update_super
= update_super_ddf
,
3141 .init_super
= init_super_ddf_bvd
,
3142 .add_to_super
= add_to_super_ddf_bvd
,
3143 .getinfo_super
= getinfo_super_ddf_bvd
,
3144 .getinfo_super_n
= getinfo_super_n_bvd
,
3146 .load_super
= load_super_ddf
,
3147 .free_super
= free_super_ddf
,
3148 .match_metadata_desc
= match_metadata_desc_ddf_bvd
,
3156 struct superswitch super_ddf_svd
= {
3158 // .detail_super = detail_super_ddf_svd,
3159 // .brief_detail_super = brief_detail_super_ddf_svd,
3160 .validate_geometry
= validate_geometry_ddf_svd
,
3162 .update_super
= update_super_ddf
,
3163 .init_super
= init_super_ddf
,
3165 .load_super
= load_super_ddf
,
3166 .free_super
= free_super_ddf
,
3167 .match_metadata_desc
= match_metadata_desc_ddf_svd
,