2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2006-2007 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neil@brown.name>
24 * Specifications for DDF takes from Common RAID DDF Specification Revision 1.2
25 * (July 28 2006). Reused by permission of SNIA.
28 #define HAVE_STDINT_H 1
34 /* a non-official T10 name for creation GUIDs */
35 static char T10
[] = "Linux-MD";
37 /* DDF timestamps are 1980 based, so we need to add
38 * second-in-decade-of-seventies to convert to linux timestamps.
39 * 10 years with 2 leap years.
41 #define DECADE (3600*24*(365*10+2))
44 const unsigned char *buf
,
47 /* The DDF metadata handling.
48 * DDF metadata lives at the end of the device.
49 * The last 512 byte block provides an 'anchor' which is used to locate
50 * the rest of the metadata which usually lives immediately behind the anchor.
53 * - all multibyte numeric fields are bigendian.
54 * - all strings are space padded.
58 /* Primary Raid Level (PRL) */
59 #define DDF_RAID0 0x00
60 #define DDF_RAID1 0x01
61 #define DDF_RAID3 0x03
62 #define DDF_RAID4 0x04
63 #define DDF_RAID5 0x05
64 #define DDF_RAID1E 0x11
66 #define DDF_CONCAT 0x1f
67 #define DDF_RAID5E 0x15
68 #define DDF_RAID5EE 0x25
69 #define DDF_RAID6 0x06
71 /* Raid Level Qualifier (RLQ) */
72 #define DDF_RAID0_SIMPLE 0x00
73 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
74 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
75 #define DDF_RAID3_0 0x00 /* parity in first extent */
76 #define DDF_RAID3_N 0x01 /* parity in last extent */
77 #define DDF_RAID4_0 0x00 /* parity in first extent */
78 #define DDF_RAID4_N 0x01 /* parity in last extent */
79 /* these apply to raid5e and raid5ee as well */
80 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
81 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
82 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
83 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
85 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
86 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
88 /* Secondary RAID Level (SRL) */
89 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
90 #define DDF_2MIRRORED 0x01
91 #define DDF_2CONCAT 0x02
92 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
95 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
96 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
97 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
98 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
99 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
100 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
101 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
102 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
103 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
104 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
106 #define DDF_GUID_LEN 24
107 #define DDF_REVISION_0 "01.00.00"
108 #define DDF_REVISION_2 "01.02.00"
111 __u32 magic
; /* DDF_HEADER_MAGIC */
113 char guid
[DDF_GUID_LEN
];
114 char revision
[8]; /* 01.02.00 */
115 __u32 seq
; /* starts at '1' */
120 __u8 pad0
; /* 0xff */
121 __u8 pad1
[12]; /* 12 * 0xff */
122 /* 64 bytes so far */
123 __u8 header_ext
[32]; /* reserved: fill with 0xff */
127 __u8 pad2
[3]; /* 0xff */
128 __u32 workspace_len
; /* sectors for vendor space -
129 * at least 32768(sectors) */
131 __u16 max_pd_entries
; /* one of 15, 63, 255, 1023, 4095 */
132 __u16 max_vd_entries
; /* 2^(4,6,8,10,12)-1 : i.e. as above */
133 __u16 max_partitions
; /* i.e. max num of configuration
134 record entries per disk */
135 __u16 config_record_len
; /* 1 +ROUNDUP(max_primary_element_entries
137 __u16 max_primary_element_entries
; /* 16, 64, 256, 1024, or 4096 */
138 __u8 pad3
[54]; /* 0xff */
139 /* 192 bytes so far */
140 __u32 controller_section_offset
;
141 __u32 controller_section_length
;
142 __u32 phys_section_offset
;
143 __u32 phys_section_length
;
144 __u32 virt_section_offset
;
145 __u32 virt_section_length
;
146 __u32 config_section_offset
;
147 __u32 config_section_length
;
148 __u32 data_section_offset
;
149 __u32 data_section_length
;
150 __u32 bbm_section_offset
;
151 __u32 bbm_section_length
;
152 __u32 diag_space_offset
;
153 __u32 diag_space_length
;
156 /* 256 bytes so far */
157 __u8 pad4
[256]; /* 0xff */
161 #define DDF_HEADER_ANCHOR 0x00
162 #define DDF_HEADER_PRIMARY 0x01
163 #define DDF_HEADER_SECONDARY 0x02
165 /* The content of the 'controller section' - global scope */
166 struct ddf_controller_data
{
167 __u32 magic
; /* DDF_CONTROLLER_MAGIC */
169 char guid
[DDF_GUID_LEN
];
170 struct controller_type
{
177 __u8 pad
[8]; /* 0xff */
178 __u8 vendor_data
[448];
181 /* The content of phys_section - global scope */
183 __u32 magic
; /* DDF_PHYS_RECORDS_MAGIC */
188 struct phys_disk_entry
{
189 char guid
[DDF_GUID_LEN
];
193 __u64 config_size
; /* DDF structures must be after here */
194 char path
[18]; /* another horrible structure really */
199 /* phys_disk_entry.type is a bitmap - bigendian remember */
200 #define DDF_Forced_PD_GUID 1
201 #define DDF_Active_in_VD 2
202 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
203 #define DDF_Spare 8 /* overrides Global_spare */
204 #define DDF_Foreign 16
205 #define DDF_Legacy 32 /* no DDF on this device */
207 #define DDF_Interface_mask 0xf00
208 #define DDF_Interface_SCSI 0x100
209 #define DDF_Interface_SAS 0x200
210 #define DDF_Interface_SATA 0x300
211 #define DDF_Interface_FC 0x400
213 /* phys_disk_entry.state is a bigendian bitmap */
215 #define DDF_Failed 2 /* overrides 1,4,8 */
216 #define DDF_Rebuilding 4
217 #define DDF_Transition 8
219 #define DDF_ReadErrors 32
220 #define DDF_Missing 64
222 /* The content of the virt_section global scope */
223 struct virtual_disk
{
224 __u32 magic
; /* DDF_VIRT_RECORDS_MAGIC */
226 __u16 populated_vdes
;
229 struct virtual_entry
{
230 char guid
[DDF_GUID_LEN
];
232 __u16 pad0
; /* 0xffff */
242 /* virtual_entry.type is a bitmap - bigendian */
244 #define DDF_Enforce_Groups 2
245 #define DDF_Unicode 4
246 #define DDF_Owner_Valid 8
248 /* virtual_entry.state is a bigendian bitmap */
249 #define DDF_state_mask 0x7
250 #define DDF_state_optimal 0x0
251 #define DDF_state_degraded 0x1
252 #define DDF_state_deleted 0x2
253 #define DDF_state_missing 0x3
254 #define DDF_state_failed 0x4
255 #define DDF_state_part_optimal 0x5
257 #define DDF_state_morphing 0x8
258 #define DDF_state_inconsistent 0x10
260 /* virtual_entry.init_state is a bigendian bitmap */
261 #define DDF_initstate_mask 0x03
262 #define DDF_init_not 0x00
263 #define DDF_init_quick 0x01 /* initialisation is progress.
264 * i.e. 'state_inconsistent' */
265 #define DDF_init_full 0x02
267 #define DDF_access_mask 0xc0
268 #define DDF_access_rw 0x00
269 #define DDF_access_ro 0x80
270 #define DDF_access_blocked 0xc0
272 /* The content of the config_section - local scope
273 * It has multiple records each config_record_len sectors
274 * They can be vd_config or spare_assign
278 __u32 magic
; /* DDF_VD_CONF_MAGIC */
280 char guid
[DDF_GUID_LEN
];
284 __u16 prim_elmnt_count
;
285 __u8 chunk_shift
; /* 0 == 512, 1==1024 etc */
288 __u8 sec_elmnt_count
;
291 __u64 blocks
; /* blocks per component could be different
292 * on different component devices...(only
293 * for concat I hope) */
294 __u64 array_blocks
; /* blocks in array */
302 __u8 v0
[32]; /* reserved- 0xff */
303 __u8 v1
[32]; /* reserved- 0xff */
304 __u8 v2
[16]; /* reserved- 0xff */
305 __u8 v3
[16]; /* reserved- 0xff */
307 __u32 phys_refnum
[0]; /* refnum of each disk in sequence */
308 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
309 bvd are always the same size */
312 /* vd_config.cache_pol[7] is a bitmap */
313 #define DDF_cache_writeback 1 /* else writethrough */
314 #define DDF_cache_wadaptive 2 /* only applies if writeback */
315 #define DDF_cache_readahead 4
316 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
317 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
318 #define DDF_cache_wallowed 32 /* enable write caching */
319 #define DDF_cache_rallowed 64 /* enable read caching */
321 struct spare_assign
{
322 __u32 magic
; /* DDF_SPARE_ASSIGN_MAGIC */
327 __u16 populated
; /* SAEs used */
328 __u16 max
; /* max SAEs */
330 struct spare_assign_entry
{
331 char guid
[DDF_GUID_LEN
];
332 __u16 secondary_element
;
336 /* spare_assign.type is a bitmap */
337 #define DDF_spare_dedicated 0x1 /* else global */
338 #define DDF_spare_revertible 0x2 /* else committable */
339 #define DDF_spare_active 0x4 /* else not active */
340 #define DDF_spare_affinity 0x8 /* enclosure affinity */
342 /* The data_section contents - local scope */
344 __u32 magic
; /* DDF_PHYS_DATA_MAGIC */
346 char guid
[DDF_GUID_LEN
];
347 __u32 refnum
; /* crc of some magic drive data ... */
348 __u8 forced_ref
; /* set when above was not result of magic */
349 __u8 forced_guid
; /* set if guid was forced rather than magic */
354 /* bbm_section content */
355 struct bad_block_log
{
362 struct mapped_block
{
363 __u64 defective_start
;
364 __u32 replacement_start
;
370 /* Struct for internally holding ddf structures */
371 /* The DDF structure stored on each device is potentially
372 * quite different, as some data is global and some is local.
373 * The global data is:
376 * - Physical disk records
377 * - Virtual disk records
379 * - Configuration records
380 * - Physical Disk data section
381 * ( and Bad block and vendor which I don't care about yet).
383 * The local data is parsed into separate lists as it is read
384 * and reconstructed for writing. This means that we only need
385 * to make config changes once and they are automatically
386 * propagated to all devices.
387 * Note that the ddf_super has space of the conf and disk data
388 * for this disk and also for a list of all such data.
389 * The list is only used for the superblock that is being
390 * built in Create or Assemble to describe the whole array.
393 struct ddf_header anchor
, primary
, secondary
;
394 struct ddf_controller_data controller
;
395 struct ddf_header
*active
;
396 struct phys_disk
*phys
;
397 struct virtual_disk
*virt
;
399 int max_part
, mppe
, conf_rec_len
;
407 __u64
*lba_offset
; /* location in 'conf' of
409 int vcnum
; /* index into ->virt */
410 __u64
*block_sizes
; /* NULL if all the same */
413 struct vd_config conf
;
414 } *conflist
, *currentconf
;
423 unsigned long long size
; /* sectors */
424 int pdnum
; /* index in ->phys */
425 struct spare_assign
*spare
;
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
)
647 int max_virt_disks
= __be16_to_cpu(super
->active
->max_vd_entries
);
648 unsigned long long dsize
;
650 /* First the local disk info */
651 if (posix_memalign((void**)&dl
, 512,
653 (super
->max_part
) * sizeof(dl
->vlist
[0])) != 0) {
654 fprintf(stderr
, Name
": %s could not allocate disk info buffer\n",
659 load_section(fd
, super
, &dl
->disk
,
660 super
->active
->data_section_offset
,
661 super
->active
->data_section_length
,
663 dl
->devname
= devname
? strdup(devname
) : NULL
;
666 dl
->major
= major(stb
.st_rdev
);
667 dl
->minor
= minor(stb
.st_rdev
);
668 dl
->next
= super
->dlist
;
669 dl
->fd
= keep
? fd
: -1;
672 if (get_dev_size(fd
, devname
, &dsize
))
673 dl
->size
= dsize
>> 9;
675 for (i
=0 ; i
< super
->max_part
; i
++)
679 for (i
=0; i
< __be16_to_cpu(super
->active
->max_pd_entries
); i
++)
680 if (memcmp(super
->phys
->entries
[i
].guid
,
681 dl
->disk
.guid
, DDF_GUID_LEN
) == 0)
684 /* Now the config list. */
685 /* 'conf' is an array of config entries, some of which are
686 * probably invalid. Those which are good need to be copied into
690 conf
= load_section(fd
, super
, NULL
,
691 super
->active
->config_section_offset
,
692 super
->active
->config_section_length
,
697 i
< __be32_to_cpu(super
->active
->config_section_length
);
698 i
+= super
->conf_rec_len
) {
699 struct vd_config
*vd
=
700 (struct vd_config
*)((char*)conf
+ i
*512);
703 if (vd
->magic
== DDF_SPARE_ASSIGN_MAGIC
) {
706 if (posix_memalign((void**)&dl
->spare
, 512,
707 super
->conf_rec_len
*512) != 0) {
709 ": %s could not allocate spare info buf\n",
714 memcpy(dl
->spare
, vd
, super
->conf_rec_len
*512);
717 if (vd
->magic
!= DDF_VD_CONF_MAGIC
)
719 for (vcl
= super
->conflist
; vcl
; vcl
= vcl
->next
) {
720 if (memcmp(vcl
->conf
.guid
,
721 vd
->guid
, DDF_GUID_LEN
) == 0)
726 dl
->vlist
[vnum
++] = vcl
;
727 if (__be32_to_cpu(vd
->seqnum
) <=
728 __be32_to_cpu(vcl
->conf
.seqnum
))
731 if (posix_memalign((void**)&vcl
, 512,
732 (super
->conf_rec_len
*512 +
733 offsetof(struct vcl
, conf
))) != 0) {
735 ": %s could not allocate vcl buf\n",
739 vcl
->next
= super
->conflist
;
740 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
741 super
->conflist
= vcl
;
742 dl
->vlist
[vnum
++] = vcl
;
744 memcpy(&vcl
->conf
, vd
, super
->conf_rec_len
*512);
745 vcl
->lba_offset
= (__u64
*)
746 &vcl
->conf
.phys_refnum
[super
->mppe
];
748 for (i
=0; i
< max_virt_disks
; i
++)
749 if (memcmp(super
->virt
->entries
[i
].guid
,
750 vcl
->conf
.guid
, DDF_GUID_LEN
)==0)
752 if (i
< max_virt_disks
)
761 static int load_super_ddf_all(struct supertype
*st
, int fd
,
762 void **sbp
, char *devname
, int keep_fd
);
764 static int load_super_ddf(struct supertype
*st
, int fd
,
767 unsigned long long dsize
;
768 struct ddf_super
*super
;
772 /* if 'fd' is a container, load metadata from all the devices */
773 if (load_super_ddf_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
777 return 1; /* FIXME Is this correct */
779 if (get_dev_size(fd
, devname
, &dsize
) == 0)
782 /* 32M is a lower bound */
783 if (dsize
<= 32*1024*1024) {
786 Name
": %s is too small for ddf: "
787 "size is %llu sectors.\n",
795 Name
": %s is an odd size for ddf: "
796 "size is %llu bytes.\n",
802 if (posix_memalign((void**)&super
, 512, sizeof(*super
))!= 0) {
803 fprintf(stderr
, Name
": malloc of %zu failed.\n",
807 memset(super
, 0, sizeof(*super
));
809 rv
= load_ddf_headers(fd
, super
, devname
);
815 /* Have valid headers and have chosen the best. Let's read in the rest*/
817 rv
= load_ddf_global(fd
, super
, devname
);
822 Name
": Failed to load all information "
823 "sections on %s\n", devname
);
828 rv
= load_ddf_local(fd
, super
, devname
, 0);
833 Name
": Failed to load all information "
834 "sections on %s\n", devname
);
839 /* Should possibly check the sections .... */
842 if (st
->ss
== NULL
) {
844 st
->minor_version
= 0;
847 st
->loaded_container
= 0;
852 static void free_super_ddf(struct supertype
*st
)
854 struct ddf_super
*ddf
= st
->sb
;
859 while (ddf
->conflist
) {
860 struct vcl
*v
= ddf
->conflist
;
861 ddf
->conflist
= v
->next
;
863 free(v
->block_sizes
);
867 struct dl
*d
= ddf
->dlist
;
868 ddf
->dlist
= d
->next
;
879 static struct supertype
*match_metadata_desc_ddf(char *arg
)
881 /* 'ddf' only support containers */
882 struct supertype
*st
;
883 if (strcmp(arg
, "ddf") != 0 &&
884 strcmp(arg
, "default") != 0
888 st
= malloc(sizeof(*st
));
889 memset(st
, 0, sizeof(*st
));
892 st
->minor_version
= 0;
900 static mapping_t ddf_state
[] = {
906 { "Partially Optimal", 5},
912 static mapping_t ddf_init_state
[] = {
913 { "Not Initialised", 0},
914 { "QuickInit in Progress", 1},
915 { "Fully Initialised", 2},
919 static mapping_t ddf_access
[] = {
923 { "Blocked (no access)", 3},
927 static mapping_t ddf_level
[] = {
928 { "RAID0", DDF_RAID0
},
929 { "RAID1", DDF_RAID1
},
930 { "RAID3", DDF_RAID3
},
931 { "RAID4", DDF_RAID4
},
932 { "RAID5", DDF_RAID5
},
933 { "RAID1E",DDF_RAID1E
},
935 { "CONCAT",DDF_CONCAT
},
936 { "RAID5E",DDF_RAID5E
},
937 { "RAID5EE",DDF_RAID5EE
},
938 { "RAID6", DDF_RAID6
},
941 static mapping_t ddf_sec_level
[] = {
942 { "Striped", DDF_2STRIPED
},
943 { "Mirrored", DDF_2MIRRORED
},
944 { "Concat", DDF_2CONCAT
},
945 { "Spanned", DDF_2SPANNED
},
953 static struct num_mapping ddf_level_num
[] = {
956 { DDF_RAID3
, LEVEL_UNSUPPORTED
},
959 { DDF_RAID1E
, LEVEL_UNSUPPORTED
},
960 { DDF_JBOD
, LEVEL_UNSUPPORTED
},
961 { DDF_CONCAT
, LEVEL_LINEAR
},
962 { DDF_RAID5E
, LEVEL_UNSUPPORTED
},
963 { DDF_RAID5EE
, LEVEL_UNSUPPORTED
},
968 static int map_num1(struct num_mapping
*map
, int num
)
971 for (i
=0 ; map
[i
].num1
!= MAXINT
; i
++)
972 if (map
[i
].num1
== num
)
977 static int all_ff(char *guid
)
980 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
981 if (guid
[i
] != (char)0xff)
987 static void print_guid(char *guid
, int tstamp
)
989 /* A GUIDs are part (or all) ASCII and part binary.
990 * They tend to be space padded.
991 * We print the GUID in HEX, then in parentheses add
992 * any initial ASCII sequence, and a possible
993 * time stamp from bytes 16-19
995 int l
= DDF_GUID_LEN
;
998 for (i
=0 ; i
<DDF_GUID_LEN
; i
++) {
999 if ((i
&3)==0 && i
!= 0) printf(":");
1000 printf("%02X", guid
[i
]&255);
1004 while (l
&& guid
[l
-1] == ' ')
1006 for (i
=0 ; i
<l
; i
++) {
1007 if (guid
[i
] >= 0x20 && guid
[i
] < 0x7f)
1008 fputc(guid
[i
], stdout
);
1013 time_t then
= __be32_to_cpu(*(__u32
*)(guid
+16)) + DECADE
;
1016 tm
= localtime(&then
);
1017 strftime(tbuf
, 100, " %D %T",tm
);
1018 fputs(tbuf
, stdout
);
1023 static void examine_vd(int n
, struct ddf_super
*sb
, char *guid
)
1025 int crl
= sb
->conf_rec_len
;
1028 for (vcl
= sb
->conflist
; vcl
; vcl
= vcl
->next
) {
1030 struct vd_config
*vc
= &vcl
->conf
;
1032 if (calc_crc(vc
, crl
*512) != vc
->crc
)
1034 if (memcmp(vc
->guid
, guid
, DDF_GUID_LEN
) != 0)
1037 /* Ok, we know about this VD, let's give more details */
1038 printf(" Raid Devices[%d] : %d (", n
,
1039 __be16_to_cpu(vc
->prim_elmnt_count
));
1040 for (i
=0; i
<__be16_to_cpu(vc
->prim_elmnt_count
); i
++) {
1042 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1043 for (j
=0; j
<cnt
; j
++)
1044 if (vc
->phys_refnum
[i
] == sb
->phys
->entries
[j
].refnum
)
1053 if (vc
->chunk_shift
!= 255)
1054 printf(" Chunk Size[%d] : %d sectors\n", n
,
1055 1 << vc
->chunk_shift
);
1056 printf(" Raid Level[%d] : %s\n", n
,
1057 map_num(ddf_level
, vc
->prl
)?:"-unknown-");
1058 if (vc
->sec_elmnt_count
!= 1) {
1059 printf(" Secondary Position[%d] : %d of %d\n", n
,
1060 vc
->sec_elmnt_seq
, vc
->sec_elmnt_count
);
1061 printf(" Secondary Level[%d] : %s\n", n
,
1062 map_num(ddf_sec_level
, vc
->srl
) ?: "-unknown-");
1064 printf(" Device Size[%d] : %llu\n", n
,
1065 __be64_to_cpu(vc
->blocks
)/2);
1066 printf(" Array Size[%d] : %llu\n", n
,
1067 __be64_to_cpu(vc
->array_blocks
)/2);
1071 static void examine_vds(struct ddf_super
*sb
)
1073 int cnt
= __be16_to_cpu(sb
->virt
->populated_vdes
);
1075 printf(" Virtual Disks : %d\n", cnt
);
1077 for (i
=0; i
<cnt
; i
++) {
1078 struct virtual_entry
*ve
= &sb
->virt
->entries
[i
];
1080 printf(" VD GUID[%d] : ", i
); print_guid(ve
->guid
, 1);
1082 printf(" unit[%d] : %d\n", i
, __be16_to_cpu(ve
->unit
));
1083 printf(" state[%d] : %s, %s%s\n", i
,
1084 map_num(ddf_state
, ve
->state
& 7),
1085 (ve
->state
& 8) ? "Morphing, ": "",
1086 (ve
->state
& 16)? "Not Consistent" : "Consistent");
1087 printf(" init state[%d] : %s\n", i
,
1088 map_num(ddf_init_state
, ve
->init_state
&3));
1089 printf(" access[%d] : %s\n", i
,
1090 map_num(ddf_access
, (ve
->init_state
>>6) & 3));
1091 printf(" Name[%d] : %.16s\n", i
, ve
->name
);
1092 examine_vd(i
, sb
, ve
->guid
);
1094 if (cnt
) printf("\n");
1097 static void examine_pds(struct ddf_super
*sb
)
1099 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1102 printf(" Physical Disks : %d\n", cnt
);
1103 printf(" Number RefNo Size Device Type/State\n");
1105 for (i
=0 ; i
<cnt
; i
++) {
1106 struct phys_disk_entry
*pd
= &sb
->phys
->entries
[i
];
1107 int type
= __be16_to_cpu(pd
->type
);
1108 int state
= __be16_to_cpu(pd
->state
);
1110 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1112 printf(" %3d %08x ", i
,
1113 __be32_to_cpu(pd
->refnum
));
1114 printf("%lluK ", __be64_to_cpu(pd
->config_size
)>>1);
1115 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1116 if (dl
->disk
.refnum
== pd
->refnum
) {
1117 char *dv
= map_dev(dl
->major
, dl
->minor
, 0);
1119 printf("%-10s", dv
);
1126 printf(" %s%s%s%s%s",
1127 (type
&2) ? "active":"",
1128 (type
&4) ? "Global-Spare":"",
1129 (type
&8) ? "spare" : "",
1130 (type
&16)? ", foreign" : "",
1131 (type
&32)? "pass-through" : "");
1132 printf("/%s%s%s%s%s%s%s",
1133 (state
&1)? "Online": "Offline",
1134 (state
&2)? ", Failed": "",
1135 (state
&4)? ", Rebuilding": "",
1136 (state
&8)? ", in-transition": "",
1137 (state
&16)? ", SMART-errors": "",
1138 (state
&32)? ", Unrecovered-Read-Errors": "",
1139 (state
&64)? ", Missing" : "");
1144 static void examine_super_ddf(struct supertype
*st
, char *homehost
)
1146 struct ddf_super
*sb
= st
->sb
;
1148 printf(" Magic : %08x\n", __be32_to_cpu(sb
->anchor
.magic
));
1149 printf(" Version : %.8s\n", sb
->anchor
.revision
);
1150 printf("Controller GUID : "); print_guid(sb
->controller
.guid
, 0);
1152 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1154 printf(" Seq : %08x\n", __be32_to_cpu(sb
->active
->seq
));
1155 printf(" Redundant hdr : %s\n", sb
->secondary
.magic
== DDF_HEADER_MAGIC
1161 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
);
1163 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4]);
1165 static void brief_examine_super_ddf(struct supertype
*st
)
1167 /* We just write a generic DDF ARRAY entry
1169 struct ddf_super
*ddf
= st
->sb
;
1173 getinfo_super_ddf(st
, &info
);
1174 fname_from_uuid(st
, &info
, nbuf
, ':');
1175 printf("ARRAY metadata=ddf UUID=%s\n", nbuf
+ 5);
1177 for (i
=0; i
<__be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
1178 struct virtual_entry
*ve
= &ddf
->virt
->entries
[i
];
1181 if (all_ff(ve
->guid
))
1183 memcpy(vcl
.conf
.guid
, ve
->guid
, DDF_GUID_LEN
);
1184 ddf
->currentconf
=&vcl
;
1185 uuid_from_super_ddf(st
, info
.uuid
);
1186 fname_from_uuid(st
, &info
, nbuf1
, ':');
1187 printf("ARRAY container=%s member=%d UUID=%s\n",
1188 nbuf
+5, i
, nbuf1
+5);
1192 static void detail_super_ddf(struct supertype
*st
, char *homehost
)
1195 * Could print DDF GUID
1196 * Need to find which array
1197 * If whole, briefly list all arrays
1202 static void brief_detail_super_ddf(struct supertype
*st
)
1204 /* FIXME I really need to know which array we are detailing.
1205 * Can that be stored in ddf_super??
1207 // struct ddf_super *ddf = st->sb;
1210 getinfo_super_ddf(st
, &info
);
1211 fname_from_uuid(st
, &info
, nbuf
,':');
1212 printf(" UUID=%s", nbuf
+ 5);
1216 static int match_home_ddf(struct supertype
*st
, char *homehost
)
1218 /* It matches 'this' host if the controller is a
1219 * Linux-MD controller with vendor_data matching
1222 struct ddf_super
*ddf
= st
->sb
;
1223 int len
= strlen(homehost
);
1225 return (memcmp(ddf
->controller
.guid
, T10
, 8) == 0 &&
1226 len
< sizeof(ddf
->controller
.vendor_data
) &&
1227 memcmp(ddf
->controller
.vendor_data
, homehost
,len
) == 0 &&
1228 ddf
->controller
.vendor_data
[len
] == 0);
1232 static struct vd_config
*find_vdcr(struct ddf_super
*ddf
, int inst
)
1236 for (v
= ddf
->conflist
; v
; v
= v
->next
)
1237 if (inst
== v
->vcnum
)
1243 static int find_phys(struct ddf_super
*ddf
, __u32 phys_refnum
)
1245 /* Find the entry in phys_disk which has the given refnum
1246 * and return it's index
1249 for (i
=0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
); i
++)
1250 if (ddf
->phys
->entries
[i
].refnum
== phys_refnum
)
1255 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4])
1257 /* The uuid returned here is used for:
1258 * uuid to put into bitmap file (Create, Grow)
1259 * uuid for backup header when saving critical section (Grow)
1260 * comparing uuids when re-adding a device into an array
1261 * In these cases the uuid required is that of the data-array,
1262 * not the device-set.
1263 * uuid to recognise same set when adding a missing device back
1264 * to an array. This is a uuid for the device-set.
1266 * For each of these we can make do with a truncated
1267 * or hashed uuid rather than the original, as long as
1269 * In the case of SVD we assume the BVD is of interest,
1270 * though that might be the case if a bitmap were made for
1271 * a mirrored SVD - worry about that later.
1272 * So we need to find the VD configuration record for the
1273 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1274 * The first 16 bytes of the sha1 of these is used.
1276 struct ddf_super
*ddf
= st
->sb
;
1277 struct vcl
*vcl
= ddf
->currentconf
;
1280 struct sha1_ctx ctx
;
1283 guid
= vcl
->conf
.guid
;
1285 guid
= ddf
->anchor
.guid
;
1287 sha1_init_ctx(&ctx
);
1288 sha1_process_bytes(guid
, DDF_GUID_LEN
, &ctx
);
1289 sha1_finish_ctx(&ctx
, buf
);
1290 memcpy(uuid
, buf
, 4*4);
1293 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
);
1295 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
)
1297 struct ddf_super
*ddf
= st
->sb
;
1299 if (ddf
->currentconf
) {
1300 getinfo_super_ddf_bvd(st
, info
);
1304 info
->array
.raid_disks
= __be16_to_cpu(ddf
->phys
->used_pdes
);
1305 info
->array
.level
= LEVEL_CONTAINER
;
1306 info
->array
.layout
= 0;
1307 info
->array
.md_minor
= -1;
1308 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*(__u32
*)
1309 (ddf
->anchor
.guid
+16));
1310 info
->array
.utime
= 0;
1311 info
->array
.chunk_size
= 0;
1314 info
->disk
.major
= 0;
1315 info
->disk
.minor
= 0;
1317 info
->disk
.number
= __be32_to_cpu(ddf
->dlist
->disk
.refnum
);
1318 info
->disk
.raid_disk
= find_phys(ddf
, ddf
->dlist
->disk
.refnum
);
1320 info
->data_offset
= __be64_to_cpu(ddf
->phys
->
1321 entries
[info
->disk
.raid_disk
].
1323 info
->component_size
= ddf
->dlist
->size
- info
->data_offset
;
1325 info
->disk
.number
= -1;
1326 // info->disk.raid_disk = find refnum in the table and use index;
1328 info
->disk
.state
= (1 << MD_DISK_SYNC
);
1331 info
->reshape_active
= 0;
1334 info
->array
.major_version
= -1;
1335 info
->array
.minor_version
= -2;
1336 strcpy(info
->text_version
, "ddf");
1337 info
->safe_mode_delay
= 0;
1339 uuid_from_super_ddf(st
, info
->uuid
);
1343 static int rlq_to_layout(int rlq
, int prl
, int raiddisks
);
1345 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
)
1347 struct ddf_super
*ddf
= st
->sb
;
1348 struct vcl
*vc
= ddf
->currentconf
;
1349 int cd
= ddf
->currentdev
;
1353 /* FIXME this returns BVD info - what if we want SVD ?? */
1355 info
->array
.raid_disks
= __be16_to_cpu(vc
->conf
.prim_elmnt_count
);
1356 info
->array
.level
= map_num1(ddf_level_num
, vc
->conf
.prl
);
1357 info
->array
.layout
= rlq_to_layout(vc
->conf
.rlq
, vc
->conf
.prl
,
1358 info
->array
.raid_disks
);
1359 info
->array
.md_minor
= -1;
1360 info
->array
.ctime
= DECADE
+
1361 __be32_to_cpu(*(__u32
*)(vc
->conf
.guid
+16));
1362 info
->array
.utime
= DECADE
+ __be32_to_cpu(vc
->conf
.timestamp
);
1363 info
->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
1365 if (cd
>= 0 && cd
< ddf
->mppe
) {
1366 info
->data_offset
= __be64_to_cpu(vc
->lba_offset
[cd
]);
1367 if (vc
->block_sizes
)
1368 info
->component_size
= vc
->block_sizes
[cd
];
1370 info
->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
1373 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
1374 if (dl
->raiddisk
== info
->disk
.raid_disk
)
1376 info
->disk
.major
= 0;
1377 info
->disk
.minor
= 0;
1379 info
->disk
.major
= dl
->major
;
1380 info
->disk
.minor
= dl
->minor
;
1382 // info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1383 // info->disk.raid_disk = find refnum in the table and use index;
1384 // info->disk.state = ???;
1386 info
->container_member
= ddf
->currentconf
->vcnum
;
1388 info
->resync_start
= 0;
1389 if (!(ddf
->virt
->entries
[info
->container_member
].state
1390 & DDF_state_inconsistent
) &&
1391 (ddf
->virt
->entries
[info
->container_member
].init_state
1392 & DDF_initstate_mask
)
1394 info
->resync_start
= ~0ULL;
1396 uuid_from_super_ddf(st
, info
->uuid
);
1398 info
->container_member
= atoi(st
->subarray
);
1399 info
->array
.major_version
= -1;
1400 info
->array
.minor_version
= -2;
1401 sprintf(info
->text_version
, "/%s/%s",
1402 devnum2devname(st
->container_dev
),
1404 info
->safe_mode_delay
= 200;
1406 memcpy(info
->name
, ddf
->virt
->entries
[info
->container_member
].name
, 16);
1409 if (info
->name
[j
] == ' ')
1414 static int update_super_ddf(struct supertype
*st
, struct mdinfo
*info
,
1416 char *devname
, int verbose
,
1417 int uuid_set
, char *homehost
)
1419 /* For 'assemble' and 'force' we need to return non-zero if any
1420 * change was made. For others, the return value is ignored.
1421 * Update options are:
1422 * force-one : This device looks a bit old but needs to be included,
1423 * update age info appropriately.
1424 * assemble: clear any 'faulty' flag to allow this device to
1426 * force-array: Array is degraded but being forced, mark it clean
1427 * if that will be needed to assemble it.
1429 * newdev: not used ????
1430 * grow: Array has gained a new device - this is currently for
1432 * resync: mark as dirty so a resync will happen.
1433 * uuid: Change the uuid of the array to match what is given
1434 * homehost: update the recorded homehost
1435 * name: update the name - preserving the homehost
1436 * _reshape_progress: record new reshape_progress position.
1438 * Following are not relevant for this version:
1439 * sparc2.2 : update from old dodgey metadata
1440 * super-minor: change the preferred_minor number
1441 * summaries: update redundant counters.
1444 // struct ddf_super *ddf = st->sb;
1445 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1446 // struct virtual_entry *ve = find_ve(ddf);
1448 /* we don't need to handle "force-*" or "assemble" as
1449 * there is no need to 'trick' the kernel. We the metadata is
1450 * first updated to activate the array, all the implied modifications
1454 if (strcmp(update
, "grow") == 0) {
1457 if (strcmp(update
, "resync") == 0) {
1458 // info->resync_checkpoint = 0;
1460 /* We ignore UUID updates as they make even less sense
1463 if (strcmp(update
, "homehost") == 0) {
1464 /* homehost is stored in controller->vendor_data,
1465 * or it is when we are the vendor
1467 // if (info->vendor_is_local)
1468 // strcpy(ddf->controller.vendor_data, homehost);
1470 if (strcmp(update
, "name") == 0) {
1471 /* name is stored in virtual_entry->name */
1472 // memset(ve->name, ' ', 16);
1473 // strncpy(ve->name, info->name, 16);
1475 if (strcmp(update
, "_reshape_progress") == 0) {
1476 /* We don't support reshape yet */
1479 // update_all_csum(ddf);
1484 static void make_header_guid(char *guid
)
1488 /* Create a DDF Header of Virtual Disk GUID */
1490 /* 24 bytes of fiction required.
1491 * first 8 are a 'vendor-id' - "Linux-MD"
1492 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1493 * Remaining 8 random number plus timestamp
1495 memcpy(guid
, T10
, sizeof(T10
));
1496 stamp
= __cpu_to_be32(0xdeadbeef);
1497 memcpy(guid
+8, &stamp
, 4);
1498 stamp
= __cpu_to_be32(0);
1499 memcpy(guid
+12, &stamp
, 4);
1500 stamp
= __cpu_to_be32(time(0) - DECADE
);
1501 memcpy(guid
+16, &stamp
, 4);
1502 rfd
= open("/dev/urandom", O_RDONLY
);
1503 if (rfd
< 0 || read(rfd
, &stamp
, 4) != 4)
1505 memcpy(guid
+20, &stamp
, 4);
1506 if (rfd
>= 0) close(rfd
);
1509 static int init_super_ddf_bvd(struct supertype
*st
,
1510 mdu_array_info_t
*info
,
1511 unsigned long long size
,
1512 char *name
, char *homehost
,
1515 static int init_super_ddf(struct supertype
*st
,
1516 mdu_array_info_t
*info
,
1517 unsigned long long size
, char *name
, char *homehost
,
1520 /* This is primarily called by Create when creating a new array.
1521 * We will then get add_to_super called for each component, and then
1522 * write_init_super called to write it out to each device.
1523 * For DDF, Create can create on fresh devices or on a pre-existing
1525 * To create on a pre-existing array a different method will be called.
1526 * This one is just for fresh drives.
1528 * We need to create the entire 'ddf' structure which includes:
1529 * DDF headers - these are easy.
1530 * Controller data - a Sector describing this controller .. not that
1531 * this is a controller exactly.
1532 * Physical Disk Record - one entry per device, so
1533 * leave plenty of space.
1534 * Virtual Disk Records - again, just leave plenty of space.
1535 * This just lists VDs, doesn't give details
1536 * Config records - describes the VDs that use this disk
1537 * DiskData - describes 'this' device.
1538 * BadBlockManagement - empty
1539 * Diag Space - empty
1540 * Vendor Logs - Could we put bitmaps here?
1543 struct ddf_super
*ddf
;
1546 int max_phys_disks
, max_virt_disks
;
1547 unsigned long long sector
;
1551 struct phys_disk
*pd
;
1552 struct virtual_disk
*vd
;
1559 return init_super_ddf_bvd(st
, info
, size
, name
, homehost
,
1562 if (posix_memalign((void**)&ddf
, 512, sizeof(*ddf
)) != 0) {
1563 fprintf(stderr
, Name
": %s could not allocate superblock\n", __func__
);
1566 memset(ddf
, 0, sizeof(*ddf
));
1567 ddf
->dlist
= NULL
; /* no physical disks yet */
1568 ddf
->conflist
= NULL
; /* No virtual disks yet */
1570 /* At least 32MB *must* be reserved for the ddf. So let's just
1571 * start 32MB from the end, and put the primary header there.
1572 * Don't do secondary for now.
1573 * We don't know exactly where that will be yet as it could be
1574 * different on each device. To just set up the lengths.
1578 ddf
->anchor
.magic
= DDF_HEADER_MAGIC
;
1579 make_header_guid(ddf
->anchor
.guid
);
1581 memcpy(ddf
->anchor
.revision
, DDF_REVISION_2
, 8);
1582 ddf
->anchor
.seq
= __cpu_to_be32(1);
1583 ddf
->anchor
.timestamp
= __cpu_to_be32(time(0) - DECADE
);
1584 ddf
->anchor
.openflag
= 0xFF;
1585 ddf
->anchor
.foreignflag
= 0;
1586 ddf
->anchor
.enforcegroups
= 0; /* Is this best?? */
1587 ddf
->anchor
.pad0
= 0xff;
1588 memset(ddf
->anchor
.pad1
, 0xff, 12);
1589 memset(ddf
->anchor
.header_ext
, 0xff, 32);
1590 ddf
->anchor
.primary_lba
= ~(__u64
)0;
1591 ddf
->anchor
.secondary_lba
= ~(__u64
)0;
1592 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
1593 memset(ddf
->anchor
.pad2
, 0xff, 3);
1594 ddf
->anchor
.workspace_len
= __cpu_to_be32(32768); /* Must be reserved */
1595 ddf
->anchor
.workspace_lba
= ~(__u64
)0; /* Put this at bottom
1596 of 32M reserved.. */
1597 max_phys_disks
= 1023; /* Should be enough */
1598 ddf
->anchor
.max_pd_entries
= __cpu_to_be16(max_phys_disks
);
1599 max_virt_disks
= 255;
1600 ddf
->anchor
.max_vd_entries
= __cpu_to_be16(max_virt_disks
); /* ?? */
1601 ddf
->anchor
.max_partitions
= __cpu_to_be16(64); /* ?? */
1604 ddf
->conf_rec_len
= 1 + ROUND_UP(ddf
->mppe
* (4+8), 512)/512;
1605 ddf
->anchor
.config_record_len
= __cpu_to_be16(ddf
->conf_rec_len
);
1606 ddf
->anchor
.max_primary_element_entries
= __cpu_to_be16(ddf
->mppe
);
1607 memset(ddf
->anchor
.pad3
, 0xff, 54);
1608 /* controller sections is one sector long immediately
1609 * after the ddf header */
1611 ddf
->anchor
.controller_section_offset
= __cpu_to_be32(sector
);
1612 ddf
->anchor
.controller_section_length
= __cpu_to_be32(1);
1615 /* phys is 8 sectors after that */
1616 pdsize
= ROUND_UP(sizeof(struct phys_disk
) +
1617 sizeof(struct phys_disk_entry
)*max_phys_disks
,
1619 switch(pdsize
/512) {
1620 case 2: case 8: case 32: case 128: case 512: break;
1623 ddf
->anchor
.phys_section_offset
= __cpu_to_be32(sector
);
1624 ddf
->anchor
.phys_section_length
=
1625 __cpu_to_be32(pdsize
/512); /* max_primary_element_entries/8 */
1626 sector
+= pdsize
/512;
1628 /* virt is another 32 sectors */
1629 vdsize
= ROUND_UP(sizeof(struct virtual_disk
) +
1630 sizeof(struct virtual_entry
) * max_virt_disks
,
1632 switch(vdsize
/512) {
1633 case 2: case 8: case 32: case 128: case 512: break;
1636 ddf
->anchor
.virt_section_offset
= __cpu_to_be32(sector
);
1637 ddf
->anchor
.virt_section_length
=
1638 __cpu_to_be32(vdsize
/512); /* max_vd_entries/8 */
1639 sector
+= vdsize
/512;
1641 clen
= ddf
->conf_rec_len
* (ddf
->max_part
+1);
1642 ddf
->anchor
.config_section_offset
= __cpu_to_be32(sector
);
1643 ddf
->anchor
.config_section_length
= __cpu_to_be32(clen
);
1646 ddf
->anchor
.data_section_offset
= __cpu_to_be32(sector
);
1647 ddf
->anchor
.data_section_length
= __cpu_to_be32(1);
1650 ddf
->anchor
.bbm_section_length
= __cpu_to_be32(0);
1651 ddf
->anchor
.bbm_section_offset
= __cpu_to_be32(0xFFFFFFFF);
1652 ddf
->anchor
.diag_space_length
= __cpu_to_be32(0);
1653 ddf
->anchor
.diag_space_offset
= __cpu_to_be32(0xFFFFFFFF);
1654 ddf
->anchor
.vendor_length
= __cpu_to_be32(0);
1655 ddf
->anchor
.vendor_offset
= __cpu_to_be32(0xFFFFFFFF);
1657 memset(ddf
->anchor
.pad4
, 0xff, 256);
1659 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
1660 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
1662 ddf
->primary
.openflag
= 1; /* I guess.. */
1663 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
1665 ddf
->secondary
.openflag
= 1; /* I guess.. */
1666 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
1668 ddf
->active
= &ddf
->primary
;
1670 ddf
->controller
.magic
= DDF_CONTROLLER_MAGIC
;
1672 /* 24 more bytes of fiction required.
1673 * first 8 are a 'vendor-id' - "Linux-MD"
1674 * Remaining 16 are serial number.... maybe a hostname would do?
1676 memcpy(ddf
->controller
.guid
, T10
, sizeof(T10
));
1677 gethostname(hostname
, sizeof(hostname
));
1678 hostname
[sizeof(hostname
) - 1] = 0;
1679 hostlen
= strlen(hostname
);
1680 memcpy(ddf
->controller
.guid
+ 24 - hostlen
, hostname
, hostlen
);
1681 for (i
= strlen(T10
) ; i
+hostlen
< 24; i
++)
1682 ddf
->controller
.guid
[i
] = ' ';
1684 ddf
->controller
.type
.vendor_id
= __cpu_to_be16(0xDEAD);
1685 ddf
->controller
.type
.device_id
= __cpu_to_be16(0xBEEF);
1686 ddf
->controller
.type
.sub_vendor_id
= 0;
1687 ddf
->controller
.type
.sub_device_id
= 0;
1688 memcpy(ddf
->controller
.product_id
, "What Is My PID??", 16);
1689 memset(ddf
->controller
.pad
, 0xff, 8);
1690 memset(ddf
->controller
.vendor_data
, 0xff, 448);
1691 if (homehost
&& strlen(homehost
) < 440)
1692 strcpy((char*)ddf
->controller
.vendor_data
, homehost
);
1694 if (posix_memalign((void**)&pd
, 512, pdsize
) != 0) {
1695 fprintf(stderr
, Name
": %s could not allocate pd\n", __func__
);
1699 ddf
->pdsize
= pdsize
;
1701 memset(pd
, 0xff, pdsize
);
1702 memset(pd
, 0, sizeof(*pd
));
1703 pd
->magic
= DDF_PHYS_DATA_MAGIC
;
1704 pd
->used_pdes
= __cpu_to_be16(0);
1705 pd
->max_pdes
= __cpu_to_be16(max_phys_disks
);
1706 memset(pd
->pad
, 0xff, 52);
1708 if (posix_memalign((void**)&vd
, 512, vdsize
) != 0) {
1709 fprintf(stderr
, Name
": %s could not allocate vd\n", __func__
);
1713 ddf
->vdsize
= vdsize
;
1714 memset(vd
, 0, vdsize
);
1715 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
1716 vd
->populated_vdes
= __cpu_to_be16(0);
1717 vd
->max_vdes
= __cpu_to_be16(max_virt_disks
);
1718 memset(vd
->pad
, 0xff, 52);
1720 for (i
=0; i
<max_virt_disks
; i
++)
1721 memset(&vd
->entries
[i
], 0xff, sizeof(struct virtual_entry
));
1724 ddf
->updates_pending
= 1;
1728 static int chunk_to_shift(int chunksize
)
1730 return ffs(chunksize
/512)-1;
1733 static int level_to_prl(int level
)
1736 case LEVEL_LINEAR
: return DDF_CONCAT
;
1737 case 0: return DDF_RAID0
;
1738 case 1: return DDF_RAID1
;
1739 case 4: return DDF_RAID4
;
1740 case 5: return DDF_RAID5
;
1741 case 6: return DDF_RAID6
;
1745 static int layout_to_rlq(int level
, int layout
, int raiddisks
)
1749 return DDF_RAID0_SIMPLE
;
1752 case 2: return DDF_RAID1_SIMPLE
;
1753 case 3: return DDF_RAID1_MULTI
;
1758 case 0: return DDF_RAID4_N
;
1764 case ALGORITHM_LEFT_ASYMMETRIC
:
1765 return DDF_RAID5_N_RESTART
;
1766 case ALGORITHM_RIGHT_ASYMMETRIC
:
1768 return DDF_RAID5_0_RESTART
;
1770 return DDF_RAID6_0_RESTART
;
1771 case ALGORITHM_LEFT_SYMMETRIC
:
1772 return DDF_RAID5_N_CONTINUE
;
1773 case ALGORITHM_RIGHT_SYMMETRIC
:
1774 return -1; /* not mentioned in standard */
1780 static int rlq_to_layout(int rlq
, int prl
, int raiddisks
)
1784 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1786 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1794 return -1; /* FIXME this isn't checked */
1798 case DDF_RAID5_N_RESTART
:
1799 return ALGORITHM_LEFT_ASYMMETRIC
;
1800 case DDF_RAID5_0_RESTART
:
1801 return ALGORITHM_RIGHT_ASYMMETRIC
;
1802 case DDF_RAID5_N_CONTINUE
:
1803 return ALGORITHM_LEFT_SYMMETRIC
;
1809 case DDF_RAID5_N_RESTART
:
1810 return ALGORITHM_LEFT_ASYMMETRIC
;
1811 case DDF_RAID6_0_RESTART
:
1812 return ALGORITHM_RIGHT_ASYMMETRIC
;
1813 case DDF_RAID5_N_CONTINUE
:
1814 return ALGORITHM_LEFT_SYMMETRIC
;
1824 unsigned long long start
, size
;
1826 static int cmp_extent(const void *av
, const void *bv
)
1828 const struct extent
*a
= av
;
1829 const struct extent
*b
= bv
;
1830 if (a
->start
< b
->start
)
1832 if (a
->start
> b
->start
)
1837 static struct extent
*get_extents(struct ddf_super
*ddf
, struct dl
*dl
)
1839 /* find a list of used extents on the give physical device
1840 * (dnum) of the given ddf.
1841 * Return a malloced array of 'struct extent'
1843 FIXME ignore DDF_Legacy devices?
1850 rv
= malloc(sizeof(struct extent
) * (ddf
->max_part
+ 2));
1854 for (i
= 0; i
< ddf
->max_part
; i
++) {
1855 struct vcl
*v
= dl
->vlist
[i
];
1858 for (j
=0; j
< v
->conf
.prim_elmnt_count
; j
++)
1859 if (v
->conf
.phys_refnum
[j
] == dl
->disk
.refnum
) {
1860 /* This device plays role 'j' in 'v'. */
1861 rv
[n
].start
= __be64_to_cpu(v
->lba_offset
[j
]);
1862 rv
[n
].size
= __be64_to_cpu(v
->conf
.blocks
);
1867 qsort(rv
, n
, sizeof(*rv
), cmp_extent
);
1869 rv
[n
].start
= __be64_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].config_size
);
1875 static int init_super_ddf_bvd(struct supertype
*st
,
1876 mdu_array_info_t
*info
,
1877 unsigned long long size
,
1878 char *name
, char *homehost
,
1881 /* We are creating a BVD inside a pre-existing container.
1882 * so st->sb is already set.
1883 * We need to create a new vd_config and a new virtual_entry
1885 struct ddf_super
*ddf
= st
->sb
;
1887 struct virtual_entry
*ve
;
1889 struct vd_config
*vc
;
1891 if (__be16_to_cpu(ddf
->virt
->populated_vdes
)
1892 >= __be16_to_cpu(ddf
->virt
->max_vdes
)) {
1893 fprintf(stderr
, Name
": This ddf already has the "
1894 "maximum of %d virtual devices\n",
1895 __be16_to_cpu(ddf
->virt
->max_vdes
));
1899 for (venum
= 0; venum
< __be16_to_cpu(ddf
->virt
->max_vdes
); venum
++)
1900 if (all_ff(ddf
->virt
->entries
[venum
].guid
))
1902 if (venum
== __be16_to_cpu(ddf
->virt
->max_vdes
)) {
1903 fprintf(stderr
, Name
": Cannot find spare slot for "
1904 "virtual disk - DDF is corrupt\n");
1907 ve
= &ddf
->virt
->entries
[venum
];
1909 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1910 * timestamp, random number
1912 make_header_guid(ve
->guid
);
1913 ve
->unit
= __cpu_to_be16(info
->md_minor
);
1915 ve
->guid_crc
= crc32(0, (unsigned char*)ddf
->anchor
.guid
, DDF_GUID_LEN
);
1917 ve
->state
= DDF_state_degraded
; /* Will be modified as devices are added */
1918 if (info
->state
& 1) /* clean */
1919 ve
->init_state
= DDF_init_full
;
1921 ve
->init_state
= DDF_init_not
;
1923 memset(ve
->pad1
, 0xff, 14);
1924 memset(ve
->name
, ' ', 16);
1926 strncpy(ve
->name
, name
, 16);
1927 ddf
->virt
->populated_vdes
=
1928 __cpu_to_be16(__be16_to_cpu(ddf
->virt
->populated_vdes
)+1);
1930 /* Now create a new vd_config */
1931 if (posix_memalign((void**)&vcl
, 512,
1932 (offsetof(struct vcl
, conf
) + ddf
->conf_rec_len
* 512)) != 0) {
1933 fprintf(stderr
, Name
": %s could not allocate vd_config\n", __func__
);
1936 vcl
->lba_offset
= (__u64
*) &vcl
->conf
.phys_refnum
[ddf
->mppe
];
1938 sprintf(st
->subarray
, "%d", venum
);
1939 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
1943 vc
->magic
= DDF_VD_CONF_MAGIC
;
1944 memcpy(vc
->guid
, ve
->guid
, DDF_GUID_LEN
);
1945 vc
->timestamp
= __cpu_to_be32(time(0)-DECADE
);
1946 vc
->seqnum
= __cpu_to_be32(1);
1947 memset(vc
->pad0
, 0xff, 24);
1948 vc
->prim_elmnt_count
= __cpu_to_be16(info
->raid_disks
);
1949 vc
->chunk_shift
= chunk_to_shift(info
->chunk_size
);
1950 vc
->prl
= level_to_prl(info
->level
);
1951 vc
->rlq
= layout_to_rlq(info
->level
, info
->layout
, info
->raid_disks
);
1952 vc
->sec_elmnt_count
= 1;
1953 vc
->sec_elmnt_seq
= 0;
1955 vc
->blocks
= __cpu_to_be64(info
->size
* 2);
1956 vc
->array_blocks
= __cpu_to_be64(
1957 calc_array_size(info
->level
, info
->raid_disks
, info
->layout
,
1958 info
->chunk_size
, info
->size
*2));
1959 memset(vc
->pad1
, 0xff, 8);
1960 vc
->spare_refs
[0] = 0xffffffff;
1961 vc
->spare_refs
[1] = 0xffffffff;
1962 vc
->spare_refs
[2] = 0xffffffff;
1963 vc
->spare_refs
[3] = 0xffffffff;
1964 vc
->spare_refs
[4] = 0xffffffff;
1965 vc
->spare_refs
[5] = 0xffffffff;
1966 vc
->spare_refs
[6] = 0xffffffff;
1967 vc
->spare_refs
[7] = 0xffffffff;
1968 memset(vc
->cache_pol
, 0, 8);
1970 memset(vc
->pad2
, 0xff, 3);
1971 memset(vc
->pad3
, 0xff, 52);
1972 memset(vc
->pad4
, 0xff, 192);
1973 memset(vc
->v0
, 0xff, 32);
1974 memset(vc
->v1
, 0xff, 32);
1975 memset(vc
->v2
, 0xff, 16);
1976 memset(vc
->v3
, 0xff, 16);
1977 memset(vc
->vendor
, 0xff, 32);
1979 memset(vc
->phys_refnum
, 0xff, 4*ddf
->mppe
);
1980 memset(vc
->phys_refnum
+(ddf
->mppe
* 4), 0x00, 8*ddf
->mppe
);
1982 vcl
->next
= ddf
->conflist
;
1983 ddf
->conflist
= vcl
;
1984 ddf
->currentconf
= vcl
;
1985 ddf
->updates_pending
= 1;
1990 static void add_to_super_ddf_bvd(struct supertype
*st
,
1991 mdu_disk_info_t
*dk
, int fd
, char *devname
)
1993 /* fd and devname identify a device with-in the ddf container (st).
1994 * dk identifies a location in the new BVD.
1995 * We need to find suitable free space in that device and update
1996 * the phys_refnum and lba_offset for the newly created vd_config.
1997 * We might also want to update the type in the phys_disk
2000 * Alternately: fd == -1 and we have already chosen which device to
2001 * use and recorded in dlist->raid_disk;
2004 struct ddf_super
*ddf
= st
->sb
;
2005 struct vd_config
*vc
;
2009 unsigned long long blocks
, pos
, esize
;
2013 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2014 if (dl
->raiddisk
== dk
->raid_disk
)
2017 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2018 if (dl
->major
== dk
->major
&&
2019 dl
->minor
== dk
->minor
)
2022 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
2025 vc
= &ddf
->currentconf
->conf
;
2026 lba_offset
= ddf
->currentconf
->lba_offset
;
2028 ex
= get_extents(ddf
, dl
);
2033 blocks
= __be64_to_cpu(vc
->blocks
);
2034 if (ddf
->currentconf
->block_sizes
)
2035 blocks
= ddf
->currentconf
->block_sizes
[dk
->raid_disk
];
2038 esize
= ex
[i
].start
- pos
;
2039 if (esize
>= blocks
)
2041 pos
= ex
[i
].start
+ ex
[i
].size
;
2043 } while (ex
[i
-1].size
);
2049 ddf
->currentdev
= dk
->raid_disk
;
2050 vc
->phys_refnum
[dk
->raid_disk
] = dl
->disk
.refnum
;
2051 lba_offset
[dk
->raid_disk
] = __cpu_to_be64(pos
);
2053 for (i
=0; i
< ddf
->max_part
; i
++)
2054 if (dl
->vlist
[i
] == NULL
)
2056 if (i
== ddf
->max_part
)
2058 dl
->vlist
[i
] = ddf
->currentconf
;
2063 dl
->devname
= devname
;
2065 /* Check how many working raid_disks, and if we can mark
2066 * array as optimal yet
2070 for (i
=0; i
< __be16_to_cpu(vc
->prim_elmnt_count
); i
++)
2071 if (vc
->phys_refnum
[i
] != 0xffffffff)
2074 /* Find which virtual_entry */
2075 i
= ddf
->currentconf
->vcnum
;
2076 if (working
== __be16_to_cpu(vc
->prim_elmnt_count
))
2077 ddf
->virt
->entries
[i
].state
=
2078 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
2079 | DDF_state_optimal
;
2081 if (vc
->prl
== DDF_RAID6
&&
2082 working
+1 == __be16_to_cpu(vc
->prim_elmnt_count
))
2083 ddf
->virt
->entries
[i
].state
=
2084 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
2085 | DDF_state_part_optimal
;
2087 ddf
->phys
->entries
[dl
->pdnum
].type
&= ~__cpu_to_be16(DDF_Global_Spare
);
2088 ddf
->phys
->entries
[dl
->pdnum
].type
|= __cpu_to_be16(DDF_Active_in_VD
);
2089 ddf
->updates_pending
= 1;
2092 /* add a device to a container, either while creating it or while
2093 * expanding a pre-existing container
2095 static int add_to_super_ddf(struct supertype
*st
,
2096 mdu_disk_info_t
*dk
, int fd
, char *devname
)
2098 struct ddf_super
*ddf
= st
->sb
;
2102 unsigned long long size
;
2103 struct phys_disk_entry
*pde
;
2107 if (ddf
->currentconf
) {
2108 add_to_super_ddf_bvd(st
, dk
, fd
, devname
);
2112 /* This is device numbered dk->number. We need to create
2113 * a phys_disk entry and a more detailed disk_data entry.
2116 if (posix_memalign((void**)&dd
, 512,
2117 sizeof(*dd
) + sizeof(dd
->vlist
[0]) * ddf
->max_part
) != 0) {
2118 fprintf(stderr
, Name
2119 ": %s could allocate buffer for new disk, aborting\n",
2123 dd
->major
= major(stb
.st_rdev
);
2124 dd
->minor
= minor(stb
.st_rdev
);
2125 dd
->devname
= devname
;
2129 dd
->disk
.magic
= DDF_PHYS_DATA_MAGIC
;
2131 tm
= localtime(&now
);
2132 sprintf(dd
->disk
.guid
, "%8s%04d%02d%02d",
2133 T10
, tm
->tm_year
+1900, tm
->tm_mon
+1, tm
->tm_mday
);
2134 *(__u32
*)(dd
->disk
.guid
+ 16) = random();
2135 *(__u32
*)(dd
->disk
.guid
+ 20) = random();
2138 /* Cannot be bothered finding a CRC of some irrelevant details*/
2139 dd
->disk
.refnum
= random();
2140 for (i
= __be16_to_cpu(ddf
->active
->max_pd_entries
) - 1;
2142 if (ddf
->phys
->entries
[i
].refnum
== dd
->disk
.refnum
)
2146 dd
->disk
.forced_ref
= 1;
2147 dd
->disk
.forced_guid
= 1;
2148 memset(dd
->disk
.vendor
, ' ', 32);
2149 memcpy(dd
->disk
.vendor
, "Linux", 5);
2150 memset(dd
->disk
.pad
, 0xff, 442);
2151 for (i
= 0; i
< ddf
->max_part
; i
++)
2152 dd
->vlist
[i
] = NULL
;
2154 n
= __be16_to_cpu(ddf
->phys
->used_pdes
);
2155 pde
= &ddf
->phys
->entries
[n
];
2158 if (st
->update_tail
) {
2159 int len
= (sizeof(struct phys_disk
) +
2160 sizeof(struct phys_disk_entry
));
2161 struct phys_disk
*pd
;
2164 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2165 pd
->used_pdes
= __cpu_to_be16(n
);
2166 pde
= &pd
->entries
[0];
2170 ddf
->phys
->used_pdes
= __cpu_to_be16(n
);
2173 memcpy(pde
->guid
, dd
->disk
.guid
, DDF_GUID_LEN
);
2174 pde
->refnum
= dd
->disk
.refnum
;
2175 pde
->type
= __cpu_to_be16(DDF_Forced_PD_GUID
| DDF_Global_Spare
);
2176 pde
->state
= __cpu_to_be16(DDF_Online
);
2177 get_dev_size(fd
, NULL
, &size
);
2178 /* We are required to reserve 32Meg, and record the size in sectors */
2179 pde
->config_size
= __cpu_to_be64( (size
- 32*1024*1024) / 512);
2180 sprintf(pde
->path
, "%17.17s","Information: nil") ;
2181 memset(pde
->pad
, 0xff, 6);
2183 dd
->size
= size
>> 9;
2184 if (st
->update_tail
) {
2185 dd
->next
= ddf
->add_list
;
2188 dd
->next
= ddf
->dlist
;
2190 ddf
->updates_pending
= 1;
2197 * This is the write_init_super method for a ddf container. It is
2198 * called when creating a container or adding another device to a
2202 static unsigned char null_conf
[4096+512];
2204 static int __write_init_super_ddf(struct supertype
*st
, int do_close
)
2207 struct ddf_super
*ddf
= st
->sb
;
2214 unsigned long long size
, sector
;
2216 /* try to write updated metadata,
2217 * if we catch a failure move on to the next disk
2219 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
2226 /* We need to fill in the primary, (secondary) and workspace
2227 * lba's in the headers, set their checksums,
2228 * Also checksum phys, virt....
2230 * Then write everything out, finally the anchor is written.
2232 get_dev_size(fd
, NULL
, &size
);
2234 ddf
->anchor
.workspace_lba
= __cpu_to_be64(size
- 32*1024*2);
2235 ddf
->anchor
.primary_lba
= __cpu_to_be64(size
- 16*1024*2);
2236 ddf
->anchor
.seq
= __cpu_to_be32(1);
2237 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2238 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2240 ddf
->anchor
.openflag
= 0xFF; /* 'open' means nothing */
2241 ddf
->anchor
.seq
= 0xFFFFFFFF; /* no sequencing in anchor */
2242 ddf
->anchor
.crc
= calc_crc(&ddf
->anchor
, 512);
2244 ddf
->primary
.openflag
= 0;
2245 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
2247 ddf
->secondary
.openflag
= 0;
2248 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
2250 ddf
->primary
.crc
= calc_crc(&ddf
->primary
, 512);
2251 ddf
->secondary
.crc
= calc_crc(&ddf
->secondary
, 512);
2253 sector
= size
- 16*1024*2;
2254 lseek64(fd
, sector
<<9, 0);
2255 if (write(fd
, &ddf
->primary
, 512) < 0)
2258 ddf
->controller
.crc
= calc_crc(&ddf
->controller
, 512);
2259 if (write(fd
, &ddf
->controller
, 512) < 0)
2262 ddf
->phys
->crc
= calc_crc(ddf
->phys
, ddf
->pdsize
);
2264 if (write(fd
, ddf
->phys
, ddf
->pdsize
) < 0)
2267 ddf
->virt
->crc
= calc_crc(ddf
->virt
, ddf
->vdsize
);
2268 if (write(fd
, ddf
->virt
, ddf
->vdsize
) < 0)
2271 /* Now write lots of config records. */
2272 n_config
= ddf
->max_part
;
2273 conf_size
= ddf
->conf_rec_len
* 512;
2274 for (i
= 0 ; i
<= n_config
; i
++) {
2275 struct vcl
*c
= d
->vlist
[i
];
2277 c
= (struct vcl
*)d
->spare
;
2280 c
->conf
.crc
= calc_crc(&c
->conf
, conf_size
);
2281 if (write(fd
, &c
->conf
, conf_size
) < 0)
2284 char *null_aligned
= (char*)((((unsigned long)null_conf
)+511)&~511UL);
2285 if (null_conf
[0] != 0xff)
2286 memset(null_conf
, 0xff, sizeof(null_conf
));
2287 int togo
= conf_size
;
2288 while (togo
> sizeof(null_conf
)-512) {
2289 if (write(fd
, null_aligned
, sizeof(null_conf
)-512) < 0)
2291 togo
-= sizeof(null_conf
)-512;
2293 if (write(fd
, null_aligned
, togo
) < 0)
2299 d
->disk
.crc
= calc_crc(&d
->disk
, 512);
2300 if (write(fd
, &d
->disk
, 512) < 0)
2303 /* Maybe do the same for secondary */
2305 lseek64(fd
, (size
-1)*512, SEEK_SET
);
2306 if (write(fd
, &ddf
->anchor
, 512) < 0)
2312 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
2317 return attempts
!= successes
;
2320 static int write_init_super_ddf(struct supertype
*st
)
2323 if (st
->update_tail
) {
2324 /* queue the virtual_disk and vd_config as metadata updates */
2325 struct virtual_disk
*vd
;
2326 struct vd_config
*vc
;
2327 struct ddf_super
*ddf
= st
->sb
;
2330 if (!ddf
->currentconf
) {
2331 int len
= (sizeof(struct phys_disk
) +
2332 sizeof(struct phys_disk_entry
));
2334 /* adding a disk to the container. */
2338 append_metadata_update(st
, ddf
->add_list
->mdupdate
, len
);
2339 ddf
->add_list
->mdupdate
= NULL
;
2343 /* Newly created VD */
2345 /* First the virtual disk. We have a slightly fake header */
2346 len
= sizeof(struct virtual_disk
) + sizeof(struct virtual_entry
);
2349 vd
->entries
[0] = ddf
->virt
->entries
[ddf
->currentconf
->vcnum
];
2350 vd
->populated_vdes
= __cpu_to_be16(ddf
->currentconf
->vcnum
);
2351 append_metadata_update(st
, vd
, len
);
2353 /* Then the vd_config */
2354 len
= ddf
->conf_rec_len
* 512;
2356 memcpy(vc
, &ddf
->currentconf
->conf
, len
);
2357 append_metadata_update(st
, vc
, len
);
2359 /* FIXME I need to close the fds! */
2362 return __write_init_super_ddf(st
, 1);
2367 static __u64
avail_size_ddf(struct supertype
*st
, __u64 devsize
)
2369 /* We must reserve the last 32Meg */
2370 if (devsize
<= 32*1024*2)
2372 return devsize
- 32*1024*2;
2377 static int reserve_space(struct supertype
*st
, int raiddisks
,
2378 unsigned long long size
, int chunk
,
2379 unsigned long long *freesize
)
2381 /* Find 'raiddisks' spare extents at least 'size' big (but
2382 * only caring about multiples of 'chunk') and remember
2384 * If the cannot be found, fail.
2387 struct ddf_super
*ddf
= st
->sb
;
2390 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
2394 /* Now find largest extent on each device */
2395 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
2396 struct extent
*e
= get_extents(ddf
, dl
);
2397 unsigned long long pos
= 0;
2400 unsigned long long minsize
= size
;
2408 unsigned long long esize
;
2409 esize
= e
[i
].start
- pos
;
2410 if (esize
>= minsize
) {
2414 pos
= e
[i
].start
+ e
[i
].size
;
2416 } while (e
[i
-1].size
);
2419 dl
->esize
= minsize
;
2423 if (cnt
< raiddisks
) {
2424 fprintf(stderr
, Name
": not enough devices with space to create array.\n");
2425 return 0; /* No enough free spaces large enough */
2428 /* choose the largest size of which there are at least 'raiddisk' */
2429 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
2431 if (dl
->esize
<= size
)
2433 /* This is bigger than 'size', see if there are enough */
2435 for (dl2
= dl
; dl2
; dl2
=dl2
->next
)
2436 if (dl2
->esize
>= dl
->esize
)
2438 if (cnt
>= raiddisks
)
2442 size
= size
/ chunk
;
2447 fprintf(stderr
, Name
": not enough spare devices to create array.\n");
2451 /* We have a 'size' of which there are enough spaces.
2452 * We simply do a first-fit */
2454 for (dl
= ddf
->dlist
; dl
&& cnt
< raiddisks
; dl
=dl
->next
) {
2455 if (dl
->esize
< size
)
2467 validate_geometry_ddf_container(struct supertype
*st
,
2468 int level
, int layout
, int raiddisks
,
2469 int chunk
, unsigned long long size
,
2470 char *dev
, unsigned long long *freesize
,
2473 static int validate_geometry_ddf_bvd(struct supertype
*st
,
2474 int level
, int layout
, int raiddisks
,
2475 int chunk
, unsigned long long size
,
2476 char *dev
, unsigned long long *freesize
,
2479 static int validate_geometry_ddf(struct supertype
*st
,
2480 int level
, int layout
, int raiddisks
,
2481 int chunk
, unsigned long long size
,
2482 char *dev
, unsigned long long *freesize
,
2489 /* ddf potentially supports lots of things, but it depends on
2490 * what devices are offered (and maybe kernel version?)
2491 * If given unused devices, we will make a container.
2492 * If given devices in a container, we will make a BVD.
2493 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2496 if (level
== LEVEL_CONTAINER
) {
2497 /* Must be a fresh device to add to a container */
2498 return validate_geometry_ddf_container(st
, level
, layout
,
2500 size
, dev
, freesize
,
2505 /* Initial sanity check. Exclude illegal levels. */
2507 for (i
=0; ddf_level_num
[i
].num1
!= MAXINT
; i
++)
2508 if (ddf_level_num
[i
].num2
== level
)
2510 if (ddf_level_num
[i
].num1
== MAXINT
)
2512 /* Should check layout? etc */
2514 if (st
->sb
&& freesize
) {
2515 /* --create was given a container to create in.
2516 * So we need to check that there are enough
2517 * free spaces and return the amount of space.
2518 * We may as well remember which drives were
2519 * chosen so that add_to_super/getinfo_super
2522 return reserve_space(st
, raiddisks
, size
, chunk
, freesize
);
2528 /* A container has already been opened, so we are
2529 * creating in there. Maybe a BVD, maybe an SVD.
2530 * Should make a distinction one day.
2532 return validate_geometry_ddf_bvd(st
, level
, layout
, raiddisks
,
2533 chunk
, size
, dev
, freesize
,
2536 /* This is the first device for the array.
2537 * If it is a container, we read it in and do automagic allocations,
2538 * no other devices should be given.
2539 * Otherwise it must be a member device of a container, and we
2540 * do manual allocation.
2541 * Later we should check for a BVD and make an SVD.
2543 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2545 sra
= sysfs_read(fd
, 0, GET_VERSION
);
2547 if (sra
&& sra
->array
.major_version
== -1 &&
2548 strcmp(sra
->text_version
, "ddf") == 0) {
2551 /* find space for 'n' devices. */
2552 /* remember the devices */
2553 /* Somehow return the fact that we have enough */
2558 Name
": ddf: Cannot create this array "
2563 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2565 fprintf(stderr
, Name
": ddf: Cannot open %s: %s\n",
2566 dev
, strerror(errno
));
2569 /* Well, it is in use by someone, maybe a 'ddf' container. */
2570 cfd
= open_container(fd
);
2574 fprintf(stderr
, Name
": ddf: Cannot use %s: %s\n",
2575 dev
, strerror(EBUSY
));
2578 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2580 if (sra
&& sra
->array
.major_version
== -1 &&
2581 strcmp(sra
->text_version
, "ddf") == 0) {
2582 /* This is a member of a ddf container. Load the container
2583 * and try to create a bvd
2585 struct ddf_super
*ddf
;
2586 if (load_super_ddf_all(st
, cfd
, (void **)&ddf
, NULL
, 1) == 0) {
2588 st
->container_dev
= fd2devnum(cfd
);
2590 return validate_geometry_ddf_bvd(st
, level
, layout
,
2591 raiddisks
, chunk
, size
,
2596 } else /* device may belong to a different container */
2603 validate_geometry_ddf_container(struct supertype
*st
,
2604 int level
, int layout
, int raiddisks
,
2605 int chunk
, unsigned long long size
,
2606 char *dev
, unsigned long long *freesize
,
2610 unsigned long long ldsize
;
2612 if (level
!= LEVEL_CONTAINER
)
2617 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2620 fprintf(stderr
, Name
": ddf: Cannot open %s: %s\n",
2621 dev
, strerror(errno
));
2624 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2630 *freesize
= avail_size_ddf(st
, ldsize
>> 9);
2635 static int validate_geometry_ddf_bvd(struct supertype
*st
,
2636 int level
, int layout
, int raiddisks
,
2637 int chunk
, unsigned long long size
,
2638 char *dev
, unsigned long long *freesize
,
2642 struct ddf_super
*ddf
= st
->sb
;
2644 unsigned long long pos
= 0;
2645 unsigned long long maxsize
;
2648 /* ddf/bvd supports lots of things, but not containers */
2649 if (level
== LEVEL_CONTAINER
)
2651 /* We must have the container info already read in. */
2656 /* General test: make sure there is space for
2657 * 'raiddisks' device extents of size 'size'.
2659 unsigned long long minsize
= size
;
2663 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2669 e
= get_extents(ddf
, dl
);
2672 unsigned long long esize
;
2673 esize
= e
[i
].start
- pos
;
2674 if (esize
>= minsize
)
2676 pos
= e
[i
].start
+ e
[i
].size
;
2678 } while (e
[i
-1].size
);
2683 if (dcnt
< raiddisks
) {
2686 Name
": ddf: Not enough devices with "
2687 "space for this array (%d < %d)\n",
2693 /* This device must be a member of the set */
2694 if (stat(dev
, &stb
) < 0)
2696 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2698 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
2699 if (dl
->major
== major(stb
.st_rdev
) &&
2700 dl
->minor
== minor(stb
.st_rdev
))
2705 fprintf(stderr
, Name
": ddf: %s is not in the "
2710 e
= get_extents(ddf
, dl
);
2714 unsigned long long esize
;
2715 esize
= e
[i
].start
- pos
;
2716 if (esize
>= maxsize
)
2718 pos
= e
[i
].start
+ e
[i
].size
;
2720 } while (e
[i
-1].size
);
2721 *freesize
= maxsize
;
2727 static int load_super_ddf_all(struct supertype
*st
, int fd
,
2728 void **sbp
, char *devname
, int keep_fd
)
2731 struct ddf_super
*super
;
2732 struct mdinfo
*sd
, *best
= NULL
;
2738 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
2741 if (sra
->array
.major_version
!= -1 ||
2742 sra
->array
.minor_version
!= -2 ||
2743 strcmp(sra
->text_version
, "ddf") != 0)
2746 if (posix_memalign((void**)&super
, 512, sizeof(*super
)) != 0)
2748 memset(super
, 0, sizeof(*super
));
2750 /* first, try each device, and choose the best ddf */
2751 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2753 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2754 dfd
= dev_open(nm
, O_RDONLY
);
2757 rv
= load_ddf_headers(dfd
, super
, NULL
);
2760 seq
= __be32_to_cpu(super
->active
->seq
);
2761 if (super
->active
->openflag
)
2763 if (!best
|| seq
> bestseq
) {
2771 /* OK, load this ddf */
2772 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
2773 dfd
= dev_open(nm
, O_RDONLY
);
2776 load_ddf_headers(dfd
, super
, NULL
);
2777 load_ddf_global(dfd
, super
, NULL
);
2779 /* Now we need the device-local bits */
2780 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
2783 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
2784 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
2787 rv
= load_ddf_headers(dfd
, super
, NULL
);
2789 rv
= load_ddf_local(dfd
, super
, NULL
, keep_fd
);
2790 if (!keep_fd
) close(dfd
);
2794 if (st
->subarray
[0]) {
2797 for (v
= super
->conflist
; v
; v
= v
->next
)
2798 if (v
->vcnum
== atoi(st
->subarray
))
2799 super
->currentconf
= v
;
2800 if (!super
->currentconf
)
2804 if (st
->ss
== NULL
) {
2805 st
->ss
= &super_ddf
;
2806 st
->minor_version
= 0;
2808 st
->container_dev
= fd2devnum(fd
);
2810 st
->loaded_container
= 1;
2813 #endif /* MDASSEMBLE */
2815 static struct mdinfo
*container_content_ddf(struct supertype
*st
)
2817 /* Given a container loaded by load_super_ddf_all,
2818 * extract information about all the arrays into
2821 * For each vcl in conflist: create an mdinfo, fill it in,
2822 * then look for matching devices (phys_refnum) in dlist
2823 * and create appropriate device mdinfo.
2825 struct ddf_super
*ddf
= st
->sb
;
2826 struct mdinfo
*rest
= NULL
;
2829 for (vc
= ddf
->conflist
; vc
; vc
=vc
->next
)
2833 struct mdinfo
*this;
2834 this = malloc(sizeof(*this));
2835 memset(this, 0, sizeof(*this));
2839 this->array
.level
= map_num1(ddf_level_num
, vc
->conf
.prl
);
2840 this->array
.raid_disks
=
2841 __be16_to_cpu(vc
->conf
.prim_elmnt_count
);
2842 this->array
.layout
= rlq_to_layout(vc
->conf
.rlq
, vc
->conf
.prl
,
2843 this->array
.raid_disks
);
2844 this->array
.md_minor
= -1;
2845 this->array
.major_version
= -1;
2846 this->array
.minor_version
= -2;
2847 this->array
.ctime
= DECADE
+
2848 __be32_to_cpu(*(__u32
*)(vc
->conf
.guid
+16));
2849 this->array
.utime
= DECADE
+
2850 __be32_to_cpu(vc
->conf
.timestamp
);
2851 this->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
2854 if ((ddf
->virt
->entries
[i
].state
& DDF_state_inconsistent
) ||
2855 (ddf
->virt
->entries
[i
].init_state
& DDF_initstate_mask
) !=
2857 this->array
.state
= 0;
2858 this->resync_start
= 0;
2860 this->array
.state
= 1;
2861 this->resync_start
= ~0ULL;
2863 memcpy(this->name
, ddf
->virt
->entries
[i
].name
, 16);
2866 if (this->name
[j
] == ' ')
2869 memset(this->uuid
, 0, sizeof(this->uuid
));
2870 this->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
2871 this->array
.size
= this->component_size
/ 2;
2872 this->container_member
= i
;
2874 ddf
->currentconf
= vc
;
2875 uuid_from_super_ddf(st
, this->uuid
);
2876 ddf
->currentconf
= NULL
;
2878 sprintf(this->text_version
, "/%s/%d",
2879 devnum2devname(st
->container_dev
),
2880 this->container_member
);
2882 for (i
=0 ; i
< ddf
->mppe
; i
++) {
2886 if (vc
->conf
.phys_refnum
[i
] == 0xFFFFFFFF)
2889 this->array
.working_disks
++;
2891 for (d
= ddf
->dlist
; d
; d
=d
->next
)
2892 if (d
->disk
.refnum
== vc
->conf
.phys_refnum
[i
])
2897 dev
= malloc(sizeof(*dev
));
2898 memset(dev
, 0, sizeof(*dev
));
2899 dev
->next
= this->devs
;
2902 dev
->disk
.number
= __be32_to_cpu(d
->disk
.refnum
);
2903 dev
->disk
.major
= d
->major
;
2904 dev
->disk
.minor
= d
->minor
;
2905 dev
->disk
.raid_disk
= i
;
2906 dev
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
2908 dev
->events
= __be32_to_cpu(ddf
->primary
.seq
);
2909 dev
->data_offset
= __be64_to_cpu(vc
->lba_offset
[i
]);
2910 dev
->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
2912 strcpy(dev
->name
, d
->devname
);
2918 static int store_zero_ddf(struct supertype
*st
, int fd
)
2920 unsigned long long dsize
;
2924 if (!get_dev_size(fd
, NULL
, &dsize
))
2927 if (posix_memalign(&buf
, 512, 512) != 0)
2929 memset(buf
, 0, 512);
2931 lseek64(fd
, dsize
-512, 0);
2932 rc
= write(fd
, buf
, 512);
2939 static int compare_super_ddf(struct supertype
*st
, struct supertype
*tst
)
2943 * 0 same, or first was empty, and second was copied
2944 * 1 second had wrong number
2946 * 3 wrong other info
2948 struct ddf_super
*first
= st
->sb
;
2949 struct ddf_super
*second
= tst
->sb
;
2957 if (memcmp(first
->anchor
.guid
, second
->anchor
.guid
, DDF_GUID_LEN
) != 0)
2960 /* FIXME should I look at anything else? */
2966 * A new array 'a' has been started which claims to be instance 'inst'
2967 * within container 'c'.
2968 * We need to confirm that the array matches the metadata in 'c' so
2969 * that we don't corrupt any metadata.
2971 static int ddf_open_new(struct supertype
*c
, struct active_array
*a
, char *inst
)
2973 dprintf("ddf: open_new %s\n", inst
);
2974 a
->info
.container_member
= atoi(inst
);
2979 * The array 'a' is to be marked clean in the metadata.
2980 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
2981 * clean up to the point (in sectors). If that cannot be recorded in the
2982 * metadata, then leave it as dirty.
2984 * For DDF, we need to clear the DDF_state_inconsistent bit in the
2985 * !global! virtual_disk.virtual_entry structure.
2987 static int ddf_set_array_state(struct active_array
*a
, int consistent
)
2989 struct ddf_super
*ddf
= a
->container
->sb
;
2990 int inst
= a
->info
.container_member
;
2991 int old
= ddf
->virt
->entries
[inst
].state
;
2992 if (consistent
== 2) {
2993 /* Should check if a recovery should be started FIXME */
2995 if (!is_resync_complete(a
))
2999 ddf
->virt
->entries
[inst
].state
&= ~DDF_state_inconsistent
;
3001 ddf
->virt
->entries
[inst
].state
|= DDF_state_inconsistent
;
3002 if (old
!= ddf
->virt
->entries
[inst
].state
)
3003 ddf
->updates_pending
= 1;
3005 old
= ddf
->virt
->entries
[inst
].init_state
;
3006 ddf
->virt
->entries
[inst
].init_state
&= ~DDF_initstate_mask
;
3007 if (is_resync_complete(a
))
3008 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_full
;
3009 else if (a
->resync_start
== 0)
3010 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_not
;
3012 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_quick
;
3013 if (old
!= ddf
->virt
->entries
[inst
].init_state
)
3014 ddf
->updates_pending
= 1;
3016 dprintf("ddf mark %d %s %llu\n", inst
, consistent
?"clean":"dirty",
3022 * The state of each disk is stored in the global phys_disk structure
3023 * in phys_disk.entries[n].state.
3024 * This makes various combinations awkward.
3025 * - When a device fails in any array, it must be failed in all arrays
3026 * that include a part of this device.
3027 * - When a component is rebuilding, we cannot include it officially in the
3028 * array unless this is the only array that uses the device.
3030 * So: when transitioning:
3031 * Online -> failed, just set failed flag. monitor will propagate
3032 * spare -> online, the device might need to be added to the array.
3033 * spare -> failed, just set failed. Don't worry if in array or not.
3035 static void ddf_set_disk(struct active_array
*a
, int n
, int state
)
3037 struct ddf_super
*ddf
= a
->container
->sb
;
3038 int inst
= a
->info
.container_member
;
3039 struct vd_config
*vc
= find_vdcr(ddf
, inst
);
3040 int pd
= find_phys(ddf
, vc
->phys_refnum
[n
]);
3044 dprintf("ddf: cannot find instance %d!!\n", inst
);
3048 /* disk doesn't currently exist. If it is now in_sync,
3050 if ((state
& DS_INSYNC
) && ! (state
& DS_FAULTY
)) {
3051 /* Find dev 'n' in a->info->devs, determine the
3052 * ddf refnum, and set vc->phys_refnum and update
3058 int old
= ddf
->phys
->entries
[pd
].state
;
3059 if (state
& DS_FAULTY
)
3060 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Failed
);
3061 if (state
& DS_INSYNC
) {
3062 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Online
);
3063 ddf
->phys
->entries
[pd
].state
&= __cpu_to_be16(~DDF_Rebuilding
);
3065 if (old
!= ddf
->phys
->entries
[pd
].state
)
3066 ddf
->updates_pending
= 1;
3069 dprintf("ddf: set_disk %d to %x\n", n
, state
);
3071 /* Now we need to check the state of the array and update
3072 * virtual_disk.entries[n].state.
3073 * It needs to be one of "optimal", "degraded", "failed".
3074 * I don't understand 'deleted' or 'missing'.
3077 for (i
=0; i
< a
->info
.array
.raid_disks
; i
++) {
3078 pd
= find_phys(ddf
, vc
->phys_refnum
[i
]);
3081 st
= __be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
3082 if ((st
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
3086 state
= DDF_state_degraded
;
3087 if (working
== a
->info
.array
.raid_disks
)
3088 state
= DDF_state_optimal
;
3089 else switch(vc
->prl
) {
3093 state
= DDF_state_failed
;
3097 state
= DDF_state_failed
;
3101 if (working
< a
->info
.array
.raid_disks
-1)
3102 state
= DDF_state_failed
;
3105 if (working
< a
->info
.array
.raid_disks
-2)
3106 state
= DDF_state_failed
;
3107 else if (working
== a
->info
.array
.raid_disks
-1)
3108 state
= DDF_state_part_optimal
;
3112 if (ddf
->virt
->entries
[inst
].state
!=
3113 ((ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
3116 ddf
->virt
->entries
[inst
].state
=
3117 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
3119 ddf
->updates_pending
= 1;
3124 static void ddf_sync_metadata(struct supertype
*st
)
3128 * Write all data to all devices.
3129 * Later, we might be able to track whether only local changes
3130 * have been made, or whether any global data has been changed,
3131 * but ddf is sufficiently weird that it probably always
3132 * changes global data ....
3134 struct ddf_super
*ddf
= st
->sb
;
3135 if (!ddf
->updates_pending
)
3137 ddf
->updates_pending
= 0;
3138 __write_init_super_ddf(st
, 0);
3139 dprintf("ddf: sync_metadata\n");
3142 static void ddf_process_update(struct supertype
*st
,
3143 struct metadata_update
*update
)
3145 /* Apply this update to the metadata.
3146 * The first 4 bytes are a DDF_*_MAGIC which guides
3148 * Possible update are:
3149 * DDF_PHYS_RECORDS_MAGIC
3150 * Add a new physical device. Changes to this record
3151 * only happen implicitly.
3152 * used_pdes is the device number.
3153 * DDF_VIRT_RECORDS_MAGIC
3154 * Add a new VD. Possibly also change the 'access' bits.
3155 * populated_vdes is the entry number.
3157 * New or updated VD. the VIRT_RECORD must already
3158 * exist. For an update, phys_refnum and lba_offset
3159 * (at least) are updated, and the VD_CONF must
3160 * be written to precisely those devices listed with
3162 * DDF_SPARE_ASSIGN_MAGIC
3163 * replacement Spare Assignment Record... but for which device?
3166 * - to create a new array, we send a VIRT_RECORD and
3167 * a VD_CONF. Then assemble and start the array.
3168 * - to activate a spare we send a VD_CONF to add the phys_refnum
3169 * and offset. This will also mark the spare as active with
3170 * a spare-assignment record.
3172 struct ddf_super
*ddf
= st
->sb
;
3173 __u32
*magic
= (__u32
*)update
->buf
;
3174 struct phys_disk
*pd
;
3175 struct virtual_disk
*vd
;
3176 struct vd_config
*vc
;
3182 dprintf("Process update %x\n", *magic
);
3185 case DDF_PHYS_RECORDS_MAGIC
:
3187 if (update
->len
!= (sizeof(struct phys_disk
) +
3188 sizeof(struct phys_disk_entry
)))
3190 pd
= (struct phys_disk
*)update
->buf
;
3192 ent
= __be16_to_cpu(pd
->used_pdes
);
3193 if (ent
>= __be16_to_cpu(ddf
->phys
->max_pdes
))
3195 if (!all_ff(ddf
->phys
->entries
[ent
].guid
))
3197 ddf
->phys
->entries
[ent
] = pd
->entries
[0];
3198 ddf
->phys
->used_pdes
= __cpu_to_be16(1 +
3199 __be16_to_cpu(ddf
->phys
->used_pdes
));
3200 ddf
->updates_pending
= 1;
3201 if (ddf
->add_list
) {
3202 struct active_array
*a
;
3203 struct dl
*al
= ddf
->add_list
;
3204 ddf
->add_list
= al
->next
;
3206 al
->next
= ddf
->dlist
;
3209 /* As a device has been added, we should check
3210 * for any degraded devices that might make
3211 * use of this spare */
3212 for (a
= st
->arrays
; a
; a
=a
->next
)
3213 a
->check_degraded
= 1;
3217 case DDF_VIRT_RECORDS_MAGIC
:
3219 if (update
->len
!= (sizeof(struct virtual_disk
) +
3220 sizeof(struct virtual_entry
)))
3222 vd
= (struct virtual_disk
*)update
->buf
;
3224 ent
= __be16_to_cpu(vd
->populated_vdes
);
3225 if (ent
>= __be16_to_cpu(ddf
->virt
->max_vdes
))
3227 if (!all_ff(ddf
->virt
->entries
[ent
].guid
))
3229 ddf
->virt
->entries
[ent
] = vd
->entries
[0];
3230 ddf
->virt
->populated_vdes
= __cpu_to_be16(1 +
3231 __be16_to_cpu(ddf
->virt
->populated_vdes
));
3232 ddf
->updates_pending
= 1;
3235 case DDF_VD_CONF_MAGIC
:
3236 dprintf("len %d %d\n", update
->len
, ddf
->conf_rec_len
);
3238 mppe
= __be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
3239 if (update
->len
!= ddf
->conf_rec_len
* 512)
3241 vc
= (struct vd_config
*)update
->buf
;
3242 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
3243 if (memcmp(vcl
->conf
.guid
, vc
->guid
, DDF_GUID_LEN
) == 0)
3245 dprintf("vcl = %p\n", vcl
);
3247 /* An update, just copy the phys_refnum and lba_offset
3250 memcpy(vcl
->conf
.phys_refnum
, vc
->phys_refnum
,
3251 mppe
* (sizeof(__u32
) + sizeof(__u64
)));
3256 vcl
= update
->space
;
3257 update
->space
= NULL
;
3258 vcl
->next
= ddf
->conflist
;
3259 memcpy(&vcl
->conf
, vc
, update
->len
);
3260 vcl
->lba_offset
= (__u64
*)
3261 &vcl
->conf
.phys_refnum
[mppe
];
3262 ddf
->conflist
= vcl
;
3264 /* Now make sure vlist is correct for each dl. */
3265 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
3268 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
3269 for (dn
=0; dn
< ddf
->mppe
; dn
++)
3270 if (vcl
->conf
.phys_refnum
[dn
] ==
3272 dprintf("dev %d has %p at %d\n",
3273 dl
->pdnum
, vcl
, vn
);
3274 dl
->vlist
[vn
++] = vcl
;
3277 while (vn
< ddf
->max_part
)
3278 dl
->vlist
[vn
++] = NULL
;
3280 ddf
->phys
->entries
[dl
->pdnum
].type
&=
3281 ~__cpu_to_be16(DDF_Global_Spare
);
3282 ddf
->phys
->entries
[dl
->pdnum
].type
|=
3283 __cpu_to_be16(DDF_Active_in_VD
);
3286 ddf
->phys
->entries
[dl
->pdnum
].type
&=
3287 ~__cpu_to_be16(DDF_Global_Spare
);
3288 ddf
->phys
->entries
[dl
->pdnum
].type
|=
3289 __cpu_to_be16(DDF_Spare
);
3291 if (!dl
->vlist
[0] && !dl
->spare
) {
3292 ddf
->phys
->entries
[dl
->pdnum
].type
|=
3293 __cpu_to_be16(DDF_Global_Spare
);
3294 ddf
->phys
->entries
[dl
->pdnum
].type
&=
3295 ~__cpu_to_be16(DDF_Spare
|
3299 ddf
->updates_pending
= 1;
3301 case DDF_SPARE_ASSIGN_MAGIC
:
3306 static void ddf_prepare_update(struct supertype
*st
,
3307 struct metadata_update
*update
)
3309 /* This update arrived at managemon.
3310 * We are about to pass it to monitor.
3311 * If a malloc is needed, do it here.
3313 struct ddf_super
*ddf
= st
->sb
;
3314 __u32
*magic
= (__u32
*)update
->buf
;
3315 if (*magic
== DDF_VD_CONF_MAGIC
)
3316 if (posix_memalign(&update
->space
, 512,
3317 offsetof(struct vcl
, conf
)
3318 + ddf
->conf_rec_len
* 512) != 0)
3319 update
->space
= NULL
;
3323 * Check if the array 'a' is degraded but not failed.
3324 * If it is, find as many spares as are available and needed and
3325 * arrange for their inclusion.
3326 * We only choose devices which are not already in the array,
3327 * and prefer those with a spare-assignment to this array.
3328 * otherwise we choose global spares - assuming always that
3329 * there is enough room.
3330 * For each spare that we assign, we return an 'mdinfo' which
3331 * describes the position for the device in the array.
3332 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
3333 * the new phys_refnum and lba_offset values.
3335 * Only worry about BVDs at the moment.
3337 static struct mdinfo
*ddf_activate_spare(struct active_array
*a
,
3338 struct metadata_update
**updates
)
3342 struct ddf_super
*ddf
= a
->container
->sb
;
3344 struct mdinfo
*rv
= NULL
;
3346 struct metadata_update
*mu
;
3349 struct vd_config
*vc
;
3352 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
3353 if ((d
->curr_state
& DS_FAULTY
) &&
3355 /* wait for Removal to happen */
3357 if (d
->state_fd
>= 0)
3361 dprintf("ddf_activate: working=%d (%d) level=%d\n", working
, a
->info
.array
.raid_disks
,
3362 a
->info
.array
.level
);
3363 if (working
== a
->info
.array
.raid_disks
)
3364 return NULL
; /* array not degraded */
3365 switch (a
->info
.array
.level
) {
3368 return NULL
; /* failed */
3372 if (working
< a
->info
.array
.raid_disks
- 1)
3373 return NULL
; /* failed */
3376 if (working
< a
->info
.array
.raid_disks
- 2)
3377 return NULL
; /* failed */
3379 default: /* concat or stripe */
3380 return NULL
; /* failed */
3383 /* For each slot, if it is not working, find a spare */
3385 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
3386 for (d
= a
->info
.devs
; d
; d
= d
->next
)
3387 if (d
->disk
.raid_disk
== i
)
3389 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
3390 if (d
&& (d
->state_fd
>= 0))
3393 /* OK, this device needs recovery. Find a spare */
3395 for ( ; dl
; dl
= dl
->next
) {
3396 unsigned long long esize
;
3397 unsigned long long pos
;
3400 int is_dedicated
= 0;
3403 /* If in this array, skip */
3404 for (d2
= a
->info
.devs
; d2
; d2
= d2
->next
)
3405 if (d2
->disk
.major
== dl
->major
&&
3406 d2
->disk
.minor
== dl
->minor
) {
3407 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
3412 if (ddf
->phys
->entries
[dl
->pdnum
].type
&
3413 __cpu_to_be16(DDF_Spare
)) {
3414 /* Check spare assign record */
3416 if (dl
->spare
->type
& DDF_spare_dedicated
) {
3417 /* check spare_ents for guid */
3419 j
< __be16_to_cpu(dl
->spare
->populated
);
3421 if (memcmp(dl
->spare
->spare_ents
[j
].guid
,
3422 ddf
->virt
->entries
[a
->info
.container_member
].guid
,
3429 } else if (ddf
->phys
->entries
[dl
->pdnum
].type
&
3430 __cpu_to_be16(DDF_Global_Spare
)) {
3433 if ( ! (is_dedicated
||
3434 (is_global
&& global_ok
))) {
3435 dprintf("%x:%x not suitable: %d %d\n", dl
->major
, dl
->minor
,
3436 is_dedicated
, is_global
);
3440 /* We are allowed to use this device - is there space?
3441 * We need a->info.component_size sectors */
3442 ex
= get_extents(ddf
, dl
);
3444 dprintf("cannot get extents\n");
3451 esize
= ex
[j
].start
- pos
;
3452 if (esize
>= a
->info
.component_size
)
3454 pos
= ex
[i
].start
+ ex
[i
].size
;
3456 } while (ex
[i
-1].size
);
3459 if (esize
< a
->info
.component_size
) {
3460 dprintf("%x:%x has no room: %llu %llu\n", dl
->major
, dl
->minor
,
3461 esize
, a
->info
.component_size
);
3466 /* Cool, we have a device with some space at pos */
3467 di
= malloc(sizeof(*di
));
3470 memset(di
, 0, sizeof(*di
));
3471 di
->disk
.number
= i
;
3472 di
->disk
.raid_disk
= i
;
3473 di
->disk
.major
= dl
->major
;
3474 di
->disk
.minor
= dl
->minor
;
3476 di
->data_offset
= pos
;
3477 di
->component_size
= a
->info
.component_size
;
3478 di
->container_member
= dl
->pdnum
;
3481 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
3486 if (!dl
&& ! global_ok
) {
3487 /* not enough dedicated spares, try global */
3495 /* No spares found */
3497 /* Now 'rv' has a list of devices to return.
3498 * Create a metadata_update record to update the
3499 * phys_refnum and lba_offset values
3501 mu
= malloc(sizeof(*mu
));
3502 if (mu
&& posix_memalign(&mu
->space
, 512, sizeof(struct vcl
)) != 0) {
3508 struct mdinfo
*n
= rv
->next
;
3516 mu
->buf
= malloc(ddf
->conf_rec_len
* 512);
3517 mu
->len
= ddf
->conf_rec_len
;
3518 mu
->next
= *updates
;
3519 vc
= find_vdcr(ddf
, a
->info
.container_member
);
3520 memcpy(mu
->buf
, vc
, ddf
->conf_rec_len
* 512);
3522 vc
= (struct vd_config
*)mu
->buf
;
3523 lba
= (__u64
*)&vc
->phys_refnum
[ddf
->mppe
];
3524 for (di
= rv
; di
; di
= di
->next
) {
3525 vc
->phys_refnum
[di
->disk
.raid_disk
] =
3526 ddf
->phys
->entries
[dl
->pdnum
].refnum
;
3527 lba
[di
->disk
.raid_disk
] = di
->data_offset
;
3532 #endif /* MDASSEMBLE */
3534 struct superswitch super_ddf
= {
3536 .examine_super
= examine_super_ddf
,
3537 .brief_examine_super
= brief_examine_super_ddf
,
3538 .detail_super
= detail_super_ddf
,
3539 .brief_detail_super
= brief_detail_super_ddf
,
3540 .validate_geometry
= validate_geometry_ddf
,
3541 .write_init_super
= write_init_super_ddf
,
3542 .add_to_super
= add_to_super_ddf
,
3544 .match_home
= match_home_ddf
,
3545 .uuid_from_super
= uuid_from_super_ddf
,
3546 .getinfo_super
= getinfo_super_ddf
,
3547 .update_super
= update_super_ddf
,
3549 .avail_size
= avail_size_ddf
,
3551 .compare_super
= compare_super_ddf
,
3553 .load_super
= load_super_ddf
,
3554 .init_super
= init_super_ddf
,
3555 .store_super
= store_zero_ddf
,
3556 .free_super
= free_super_ddf
,
3557 .match_metadata_desc
= match_metadata_desc_ddf
,
3558 .container_content
= container_content_ddf
,
3564 .open_new
= ddf_open_new
,
3565 .set_array_state
= ddf_set_array_state
,
3566 .set_disk
= ddf_set_disk
,
3567 .sync_metadata
= ddf_sync_metadata
,
3568 .process_update
= ddf_process_update
,
3569 .prepare_update
= ddf_prepare_update
,
3570 .activate_spare
= ddf_activate_spare
,