2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2006-2009 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neil@brown.name>
24 * Specifications for DDF takes from Common RAID DDF Specification Revision 1.2
25 * (July 28 2006). Reused by permission of SNIA.
28 #define HAVE_STDINT_H 1
34 /* a non-official T10 name for creation GUIDs */
35 static char T10
[] = "Linux-MD";
37 /* DDF timestamps are 1980 based, so we need to add
38 * second-in-decade-of-seventies to convert to linux timestamps.
39 * 10 years with 2 leap years.
41 #define DECADE (3600*24*(365*10+2))
44 const unsigned char *buf
,
47 #define DDF_NOTFOUND (~0U)
48 #define DDF_CONTAINER (DDF_NOTFOUND-1)
50 /* The DDF metadata handling.
51 * DDF metadata lives at the end of the device.
52 * The last 512 byte block provides an 'anchor' which is used to locate
53 * the rest of the metadata which usually lives immediately behind the anchor.
56 * - all multibyte numeric fields are bigendian.
57 * - all strings are space padded.
61 /* Primary Raid Level (PRL) */
62 #define DDF_RAID0 0x00
63 #define DDF_RAID1 0x01
64 #define DDF_RAID3 0x03
65 #define DDF_RAID4 0x04
66 #define DDF_RAID5 0x05
67 #define DDF_RAID1E 0x11
69 #define DDF_CONCAT 0x1f
70 #define DDF_RAID5E 0x15
71 #define DDF_RAID5EE 0x25
72 #define DDF_RAID6 0x06
74 /* Raid Level Qualifier (RLQ) */
75 #define DDF_RAID0_SIMPLE 0x00
76 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
77 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
78 #define DDF_RAID3_0 0x00 /* parity in first extent */
79 #define DDF_RAID3_N 0x01 /* parity in last extent */
80 #define DDF_RAID4_0 0x00 /* parity in first extent */
81 #define DDF_RAID4_N 0x01 /* parity in last extent */
82 /* these apply to raid5e and raid5ee as well */
83 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
84 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
85 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
86 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
88 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
89 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
91 /* Secondary RAID Level (SRL) */
92 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
93 #define DDF_2MIRRORED 0x01
94 #define DDF_2CONCAT 0x02
95 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
98 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
99 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
100 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
101 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
102 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
103 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
104 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
105 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
106 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
107 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
109 #define DDF_GUID_LEN 24
110 #define DDF_REVISION_0 "01.00.00"
111 #define DDF_REVISION_2 "01.02.00"
114 __u32 magic
; /* DDF_HEADER_MAGIC */
116 char guid
[DDF_GUID_LEN
];
117 char revision
[8]; /* 01.02.00 */
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 */
314 #define LBA_OFFSET(ddf, vd) ((__u64 *) &(vd)->phys_refnum[(ddf)->mppe])
316 /* vd_config.cache_pol[7] is a bitmap */
317 #define DDF_cache_writeback 1 /* else writethrough */
318 #define DDF_cache_wadaptive 2 /* only applies if writeback */
319 #define DDF_cache_readahead 4
320 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
321 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
322 #define DDF_cache_wallowed 32 /* enable write caching */
323 #define DDF_cache_rallowed 64 /* enable read caching */
325 struct spare_assign
{
326 __u32 magic
; /* DDF_SPARE_ASSIGN_MAGIC */
331 __u16 populated
; /* SAEs used */
332 __u16 max
; /* max SAEs */
334 struct spare_assign_entry
{
335 char guid
[DDF_GUID_LEN
];
336 __u16 secondary_element
;
340 /* spare_assign.type is a bitmap */
341 #define DDF_spare_dedicated 0x1 /* else global */
342 #define DDF_spare_revertible 0x2 /* else committable */
343 #define DDF_spare_active 0x4 /* else not active */
344 #define DDF_spare_affinity 0x8 /* enclosure affinity */
346 /* The data_section contents - local scope */
348 __u32 magic
; /* DDF_PHYS_DATA_MAGIC */
350 char guid
[DDF_GUID_LEN
];
351 __u32 refnum
; /* crc of some magic drive data ... */
352 __u8 forced_ref
; /* set when above was not result of magic */
353 __u8 forced_guid
; /* set if guid was forced rather than magic */
358 /* bbm_section content */
359 struct bad_block_log
{
366 struct mapped_block
{
367 __u64 defective_start
;
368 __u32 replacement_start
;
374 /* Struct for internally holding ddf structures */
375 /* The DDF structure stored on each device is potentially
376 * quite different, as some data is global and some is local.
377 * The global data is:
380 * - Physical disk records
381 * - Virtual disk records
383 * - Configuration records
384 * - Physical Disk data section
385 * ( and Bad block and vendor which I don't care about yet).
387 * The local data is parsed into separate lists as it is read
388 * and reconstructed for writing. This means that we only need
389 * to make config changes once and they are automatically
390 * propagated to all devices.
391 * Note that the ddf_super has space of the conf and disk data
392 * for this disk and also for a list of all such data.
393 * The list is only used for the superblock that is being
394 * built in Create or Assemble to describe the whole array.
397 struct ddf_header anchor
, primary
, secondary
;
398 struct ddf_controller_data controller
;
399 struct ddf_header
*active
;
400 struct phys_disk
*phys
;
401 struct virtual_disk
*virt
;
403 unsigned int max_part
, mppe
, conf_rec_len
;
411 unsigned int vcnum
; /* index into ->virt */
412 struct vd_config
**other_bvds
;
413 __u64
*block_sizes
; /* NULL if all the same */
416 struct vd_config conf
;
417 } *conflist
, *currentconf
;
426 unsigned long long size
; /* sectors */
427 unsigned long long primary_lba
; /* sectors */
428 unsigned long long secondary_lba
; /* sectors */
429 unsigned long long workspace_lba
; /* sectors */
430 int pdnum
; /* index in ->phys */
431 struct spare_assign
*spare
;
432 void *mdupdate
; /* hold metadata update */
434 /* These fields used by auto-layout */
435 int raiddisk
; /* slot to fill in autolayout */
439 struct disk_data disk
;
440 struct vcl
*vlist
[0]; /* max_part in size */
445 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
449 static int all_ff(const char *guid
);
450 static void pr_state(struct ddf_super
*ddf
, const char *msg
)
453 dprintf("%s/%s: ", __func__
, msg
);
454 for (i
= 0; i
< __be16_to_cpu(ddf
->active
->max_vd_entries
); i
++) {
455 if (all_ff(ddf
->virt
->entries
[i
].guid
))
457 dprintf("%u(s=%02x i=%02x) ", i
,
458 ddf
->virt
->entries
[i
].state
,
459 ddf
->virt
->entries
[i
].init_state
);
464 static void pr_state(const struct ddf_super
*ddf
, const char *msg
) {}
467 #define ddf_set_updates_pending(x) \
468 do { (x)->updates_pending = 1; pr_state(x, __func__); } while (0)
470 static unsigned int get_pd_index_from_refnum(const struct vcl
*vc
,
471 __u32 refnum
, unsigned int nmax
,
472 const struct vd_config
**bvd
,
475 static unsigned int calc_crc(void *buf
, int len
)
477 /* crcs are always at the same place as in the ddf_header */
478 struct ddf_header
*ddf
= buf
;
479 __u32 oldcrc
= ddf
->crc
;
481 ddf
->crc
= 0xffffffff;
483 newcrc
= crc32(0, buf
, len
);
485 /* The crc is store (like everything) bigendian, so convert
486 * here for simplicity
488 return __cpu_to_be32(newcrc
);
491 #define DDF_INVALID_LEVEL 0xff
492 #define DDF_NO_SECONDARY 0xff
493 static int err_bad_md_layout(const mdu_array_info_t
*array
)
495 pr_err("RAID%d layout %x with %d disks is unsupported for DDF\n",
496 array
->level
, array
->layout
, array
->raid_disks
);
497 return DDF_INVALID_LEVEL
;
500 static int layout_md2ddf(const mdu_array_info_t
*array
,
501 struct vd_config
*conf
)
503 __u16 prim_elmnt_count
= __cpu_to_be16(array
->raid_disks
);
504 __u8 prl
= DDF_INVALID_LEVEL
, rlq
= 0;
505 __u8 sec_elmnt_count
= 1;
506 __u8 srl
= DDF_NO_SECONDARY
;
508 switch (array
->level
) {
513 rlq
= DDF_RAID0_SIMPLE
;
517 switch (array
->raid_disks
) {
519 rlq
= DDF_RAID1_SIMPLE
;
522 rlq
= DDF_RAID1_MULTI
;
525 return err_bad_md_layout(array
);
530 if (array
->layout
!= 0)
531 return err_bad_md_layout(array
);
536 switch (array
->layout
) {
537 case ALGORITHM_LEFT_ASYMMETRIC
:
538 rlq
= DDF_RAID5_N_RESTART
;
540 case ALGORITHM_RIGHT_ASYMMETRIC
:
541 rlq
= DDF_RAID5_0_RESTART
;
543 case ALGORITHM_LEFT_SYMMETRIC
:
544 rlq
= DDF_RAID5_N_CONTINUE
;
546 case ALGORITHM_RIGHT_SYMMETRIC
:
547 /* not mentioned in standard */
549 return err_bad_md_layout(array
);
554 switch (array
->layout
) {
555 case ALGORITHM_ROTATING_N_RESTART
:
556 rlq
= DDF_RAID5_N_RESTART
;
558 case ALGORITHM_ROTATING_ZERO_RESTART
:
559 rlq
= DDF_RAID6_0_RESTART
;
561 case ALGORITHM_ROTATING_N_CONTINUE
:
562 rlq
= DDF_RAID5_N_CONTINUE
;
565 return err_bad_md_layout(array
);
570 if (array
->raid_disks
% 2 == 0 && array
->layout
== 0x102) {
571 rlq
= DDF_RAID1_SIMPLE
;
572 prim_elmnt_count
= __cpu_to_be16(2);
573 sec_elmnt_count
= array
->raid_disks
/ 2;
574 } else if (array
->raid_disks
% 3 == 0
575 && array
->layout
== 0x103) {
576 rlq
= DDF_RAID1_MULTI
;
577 prim_elmnt_count
= __cpu_to_be16(3);
578 sec_elmnt_count
= array
->raid_disks
/ 3;
580 return err_bad_md_layout(array
);
585 return err_bad_md_layout(array
);
588 conf
->prim_elmnt_count
= prim_elmnt_count
;
591 conf
->sec_elmnt_count
= sec_elmnt_count
;
595 static int err_bad_ddf_layout(const struct vd_config
*conf
)
597 pr_err("DDF RAID %u qualifier %u with %u disks is unsupported\n",
598 conf
->prl
, conf
->rlq
, __be16_to_cpu(conf
->prim_elmnt_count
));
602 static int layout_ddf2md(const struct vd_config
*conf
,
603 mdu_array_info_t
*array
)
605 int level
= LEVEL_UNSUPPORTED
;
607 int raiddisks
= __be16_to_cpu(conf
->prim_elmnt_count
);
609 if (conf
->sec_elmnt_count
> 1) {
610 /* see also check_secondary() */
611 if (conf
->prl
!= DDF_RAID1
||
612 (conf
->srl
!= DDF_2STRIPED
&& conf
->srl
!= DDF_2SPANNED
)) {
613 pr_err("Unsupported secondary RAID level %u/%u\n",
614 conf
->prl
, conf
->srl
);
617 if (raiddisks
== 2 && conf
->rlq
== DDF_RAID1_SIMPLE
)
619 else if (raiddisks
== 3 && conf
->rlq
== DDF_RAID1_MULTI
)
622 return err_bad_ddf_layout(conf
);
623 raiddisks
*= conf
->sec_elmnt_count
;
630 level
= LEVEL_LINEAR
;
633 if (conf
->rlq
!= DDF_RAID0_SIMPLE
)
634 return err_bad_ddf_layout(conf
);
638 if (!((conf
->rlq
== DDF_RAID1_SIMPLE
&& raiddisks
== 2) ||
639 (conf
->rlq
== DDF_RAID1_MULTI
&& raiddisks
== 3)))
640 return err_bad_ddf_layout(conf
);
644 if (conf
->rlq
!= DDF_RAID4_N
)
645 return err_bad_ddf_layout(conf
);
650 case DDF_RAID5_N_RESTART
:
651 layout
= ALGORITHM_LEFT_ASYMMETRIC
;
653 case DDF_RAID5_0_RESTART
:
654 layout
= ALGORITHM_RIGHT_ASYMMETRIC
;
656 case DDF_RAID5_N_CONTINUE
:
657 layout
= ALGORITHM_LEFT_SYMMETRIC
;
660 return err_bad_ddf_layout(conf
);
666 case DDF_RAID5_N_RESTART
:
667 layout
= ALGORITHM_ROTATING_N_RESTART
;
669 case DDF_RAID6_0_RESTART
:
670 layout
= ALGORITHM_ROTATING_ZERO_RESTART
;
672 case DDF_RAID5_N_CONTINUE
:
673 layout
= ALGORITHM_ROTATING_N_CONTINUE
;
676 return err_bad_ddf_layout(conf
);
681 return err_bad_ddf_layout(conf
);
685 array
->level
= level
;
686 array
->layout
= layout
;
687 array
->raid_disks
= raiddisks
;
691 static int load_ddf_header(int fd
, unsigned long long lba
,
692 unsigned long long size
,
694 struct ddf_header
*hdr
, struct ddf_header
*anchor
)
696 /* read a ddf header (primary or secondary) from fd/lba
697 * and check that it is consistent with anchor
699 * magic, crc, guid, rev, and LBA's header_type, and
700 * everything after header_type must be the same
705 if (lseek64(fd
, lba
<<9, 0) < 0)
708 if (read(fd
, hdr
, 512) != 512)
711 if (hdr
->magic
!= DDF_HEADER_MAGIC
)
713 if (calc_crc(hdr
, 512) != hdr
->crc
)
715 if (memcmp(anchor
->guid
, hdr
->guid
, DDF_GUID_LEN
) != 0 ||
716 memcmp(anchor
->revision
, hdr
->revision
, 8) != 0 ||
717 anchor
->primary_lba
!= hdr
->primary_lba
||
718 anchor
->secondary_lba
!= hdr
->secondary_lba
||
720 memcmp(anchor
->pad2
, hdr
->pad2
, 512 -
721 offsetof(struct ddf_header
, pad2
)) != 0)
724 /* Looks good enough to me... */
728 static void *load_section(int fd
, struct ddf_super
*super
, void *buf
,
729 __u32 offset_be
, __u32 len_be
, int check
)
731 unsigned long long offset
= __be32_to_cpu(offset_be
);
732 unsigned long long len
= __be32_to_cpu(len_be
);
733 int dofree
= (buf
== NULL
);
736 if (len
!= 2 && len
!= 8 && len
!= 32
737 && len
!= 128 && len
!= 512)
743 /* All pre-allocated sections are a single block */
746 } else if (posix_memalign(&buf
, 512, len
<<9) != 0)
752 if (super
->active
->type
== 1)
753 offset
+= __be64_to_cpu(super
->active
->primary_lba
);
755 offset
+= __be64_to_cpu(super
->active
->secondary_lba
);
757 if ((unsigned long long)lseek64(fd
, offset
<<9, 0) != (offset
<<9)) {
762 if ((unsigned long long)read(fd
, buf
, len
<<9) != (len
<<9)) {
770 static int load_ddf_headers(int fd
, struct ddf_super
*super
, char *devname
)
772 unsigned long long dsize
;
774 get_dev_size(fd
, NULL
, &dsize
);
776 if (lseek64(fd
, dsize
-512, 0) < 0) {
778 pr_err("Cannot seek to anchor block on %s: %s\n",
779 devname
, strerror(errno
));
782 if (read(fd
, &super
->anchor
, 512) != 512) {
784 pr_err("Cannot read anchor block on %s: %s\n",
785 devname
, strerror(errno
));
788 if (super
->anchor
.magic
!= DDF_HEADER_MAGIC
) {
790 pr_err("no DDF anchor found on %s\n",
794 if (calc_crc(&super
->anchor
, 512) != super
->anchor
.crc
) {
796 pr_err("bad CRC on anchor on %s\n",
800 if (memcmp(super
->anchor
.revision
, DDF_REVISION_0
, 8) != 0 &&
801 memcmp(super
->anchor
.revision
, DDF_REVISION_2
, 8) != 0) {
803 pr_err("can only support super revision"
804 " %.8s and earlier, not %.8s on %s\n",
805 DDF_REVISION_2
, super
->anchor
.revision
,devname
);
808 super
->active
= NULL
;
809 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.primary_lba
),
811 &super
->primary
, &super
->anchor
) == 0) {
813 pr_err("Failed to load primary DDF header "
816 super
->active
= &super
->primary
;
817 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.secondary_lba
),
819 &super
->secondary
, &super
->anchor
)) {
820 if ((__be32_to_cpu(super
->primary
.seq
)
821 < __be32_to_cpu(super
->secondary
.seq
) &&
822 !super
->secondary
.openflag
)
823 || (__be32_to_cpu(super
->primary
.seq
)
824 == __be32_to_cpu(super
->secondary
.seq
) &&
825 super
->primary
.openflag
&& !super
->secondary
.openflag
)
826 || super
->active
== NULL
828 super
->active
= &super
->secondary
;
830 pr_err("Failed to load secondary DDF header on %s\n",
832 if (super
->active
== NULL
)
837 static int load_ddf_global(int fd
, struct ddf_super
*super
, char *devname
)
840 ok
= load_section(fd
, super
, &super
->controller
,
841 super
->active
->controller_section_offset
,
842 super
->active
->controller_section_length
,
844 super
->phys
= load_section(fd
, super
, NULL
,
845 super
->active
->phys_section_offset
,
846 super
->active
->phys_section_length
,
848 super
->pdsize
= __be32_to_cpu(super
->active
->phys_section_length
) * 512;
850 super
->virt
= load_section(fd
, super
, NULL
,
851 super
->active
->virt_section_offset
,
852 super
->active
->virt_section_length
,
854 super
->vdsize
= __be32_to_cpu(super
->active
->virt_section_length
) * 512;
864 super
->conflist
= NULL
;
867 super
->max_part
= __be16_to_cpu(super
->active
->max_partitions
);
868 super
->mppe
= __be16_to_cpu(super
->active
->max_primary_element_entries
);
869 super
->conf_rec_len
= __be16_to_cpu(super
->active
->config_record_len
);
873 #define DDF_UNUSED_BVD 0xff
874 static int alloc_other_bvds(const struct ddf_super
*ddf
, struct vcl
*vcl
)
876 unsigned int n_vds
= vcl
->conf
.sec_elmnt_count
- 1;
877 unsigned int i
, vdsize
;
880 vcl
->other_bvds
= NULL
;
883 vdsize
= ddf
->conf_rec_len
* 512;
884 if (posix_memalign(&p
, 512, n_vds
*
885 (vdsize
+ sizeof(struct vd_config
*))) != 0)
887 vcl
->other_bvds
= (struct vd_config
**) (p
+ n_vds
* vdsize
);
888 for (i
= 0; i
< n_vds
; i
++) {
889 vcl
->other_bvds
[i
] = p
+ i
* vdsize
;
890 memset(vcl
->other_bvds
[i
], 0, vdsize
);
891 vcl
->other_bvds
[i
]->sec_elmnt_seq
= DDF_UNUSED_BVD
;
896 static void add_other_bvd(struct vcl
*vcl
, struct vd_config
*vd
,
900 for (i
= 0; i
< vcl
->conf
.sec_elmnt_count
-1; i
++)
901 if (vcl
->other_bvds
[i
]->sec_elmnt_seq
== vd
->sec_elmnt_seq
)
904 if (i
< vcl
->conf
.sec_elmnt_count
-1) {
905 if (vd
->seqnum
<= vcl
->other_bvds
[i
]->seqnum
)
908 for (i
= 0; i
< vcl
->conf
.sec_elmnt_count
-1; i
++)
909 if (vcl
->other_bvds
[i
]->sec_elmnt_seq
== DDF_UNUSED_BVD
)
911 if (i
== vcl
->conf
.sec_elmnt_count
-1) {
912 pr_err("no space for sec level config %u, count is %u\n",
913 vd
->sec_elmnt_seq
, vcl
->conf
.sec_elmnt_count
);
917 memcpy(vcl
->other_bvds
[i
], vd
, len
);
920 static int load_ddf_local(int fd
, struct ddf_super
*super
,
921 char *devname
, int keep
)
927 unsigned int confsec
;
929 unsigned int max_virt_disks
= __be16_to_cpu(super
->active
->max_vd_entries
);
930 unsigned long long dsize
;
932 /* First the local disk info */
933 if (posix_memalign((void**)&dl
, 512,
935 (super
->max_part
) * sizeof(dl
->vlist
[0])) != 0) {
936 pr_err("%s could not allocate disk info buffer\n",
941 load_section(fd
, super
, &dl
->disk
,
942 super
->active
->data_section_offset
,
943 super
->active
->data_section_length
,
945 dl
->devname
= devname
? xstrdup(devname
) : NULL
;
948 dl
->major
= major(stb
.st_rdev
);
949 dl
->minor
= minor(stb
.st_rdev
);
950 dl
->next
= super
->dlist
;
951 dl
->fd
= keep
? fd
: -1;
954 if (get_dev_size(fd
, devname
, &dsize
))
955 dl
->size
= dsize
>> 9;
956 /* If the disks have different sizes, the LBAs will differ
957 * between phys disks.
958 * At this point here, the values in super->active must be valid
959 * for this phys disk. */
960 dl
->primary_lba
= super
->active
->primary_lba
;
961 dl
->secondary_lba
= super
->active
->secondary_lba
;
962 dl
->workspace_lba
= super
->active
->workspace_lba
;
964 for (i
= 0 ; i
< super
->max_part
; i
++)
968 for (i
= 0; i
< __be16_to_cpu(super
->active
->max_pd_entries
); i
++)
969 if (memcmp(super
->phys
->entries
[i
].guid
,
970 dl
->disk
.guid
, DDF_GUID_LEN
) == 0)
973 /* Now the config list. */
974 /* 'conf' is an array of config entries, some of which are
975 * probably invalid. Those which are good need to be copied into
979 conf
= load_section(fd
, super
, NULL
,
980 super
->active
->config_section_offset
,
981 super
->active
->config_section_length
,
986 confsec
< __be32_to_cpu(super
->active
->config_section_length
);
987 confsec
+= super
->conf_rec_len
) {
988 struct vd_config
*vd
=
989 (struct vd_config
*)((char*)conf
+ confsec
*512);
992 if (vd
->magic
== DDF_SPARE_ASSIGN_MAGIC
) {
995 if (posix_memalign((void**)&dl
->spare
, 512,
996 super
->conf_rec_len
*512) != 0) {
997 pr_err("%s could not allocate spare info buf\n",
1002 memcpy(dl
->spare
, vd
, super
->conf_rec_len
*512);
1005 if (vd
->magic
!= DDF_VD_CONF_MAGIC
)
1007 for (vcl
= super
->conflist
; vcl
; vcl
= vcl
->next
) {
1008 if (memcmp(vcl
->conf
.guid
,
1009 vd
->guid
, DDF_GUID_LEN
) == 0)
1014 dl
->vlist
[vnum
++] = vcl
;
1015 if (vcl
->other_bvds
!= NULL
&&
1016 vcl
->conf
.sec_elmnt_seq
!= vd
->sec_elmnt_seq
) {
1017 add_other_bvd(vcl
, vd
, super
->conf_rec_len
*512);
1020 if (__be32_to_cpu(vd
->seqnum
) <=
1021 __be32_to_cpu(vcl
->conf
.seqnum
))
1024 if (posix_memalign((void**)&vcl
, 512,
1025 (super
->conf_rec_len
*512 +
1026 offsetof(struct vcl
, conf
))) != 0) {
1027 pr_err("%s could not allocate vcl buf\n",
1031 vcl
->next
= super
->conflist
;
1032 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
1033 vcl
->conf
.sec_elmnt_count
= vd
->sec_elmnt_count
;
1034 if (alloc_other_bvds(super
, vcl
) != 0) {
1035 pr_err("%s could not allocate other bvds\n",
1040 super
->conflist
= vcl
;
1041 dl
->vlist
[vnum
++] = vcl
;
1043 memcpy(&vcl
->conf
, vd
, super
->conf_rec_len
*512);
1044 for (i
=0; i
< max_virt_disks
; i
++)
1045 if (memcmp(super
->virt
->entries
[i
].guid
,
1046 vcl
->conf
.guid
, DDF_GUID_LEN
)==0)
1048 if (i
< max_virt_disks
)
1057 static int load_super_ddf_all(struct supertype
*st
, int fd
,
1058 void **sbp
, char *devname
);
1061 static void free_super_ddf(struct supertype
*st
);
1063 static int load_super_ddf(struct supertype
*st
, int fd
,
1066 unsigned long long dsize
;
1067 struct ddf_super
*super
;
1070 if (get_dev_size(fd
, devname
, &dsize
) == 0)
1073 if (!st
->ignore_hw_compat
&& test_partition(fd
))
1074 /* DDF is not allowed on partitions */
1077 /* 32M is a lower bound */
1078 if (dsize
<= 32*1024*1024) {
1080 pr_err("%s is too small for ddf: "
1081 "size is %llu sectors.\n",
1087 pr_err("%s is an odd size for ddf: "
1088 "size is %llu bytes.\n",
1095 if (posix_memalign((void**)&super
, 512, sizeof(*super
))!= 0) {
1096 pr_err("malloc of %zu failed.\n",
1100 memset(super
, 0, sizeof(*super
));
1102 rv
= load_ddf_headers(fd
, super
, devname
);
1108 /* Have valid headers and have chosen the best. Let's read in the rest*/
1110 rv
= load_ddf_global(fd
, super
, devname
);
1114 pr_err("Failed to load all information "
1115 "sections on %s\n", devname
);
1120 rv
= load_ddf_local(fd
, super
, devname
, 0);
1124 pr_err("Failed to load all information "
1125 "sections on %s\n", devname
);
1130 /* Should possibly check the sections .... */
1133 if (st
->ss
== NULL
) {
1134 st
->ss
= &super_ddf
;
1135 st
->minor_version
= 0;
1142 static void free_super_ddf(struct supertype
*st
)
1144 struct ddf_super
*ddf
= st
->sb
;
1149 while (ddf
->conflist
) {
1150 struct vcl
*v
= ddf
->conflist
;
1151 ddf
->conflist
= v
->next
;
1153 free(v
->block_sizes
);
1156 v->other_bvds[0] points to beginning of buffer,
1157 see alloc_other_bvds()
1159 free(v
->other_bvds
[0]);
1162 while (ddf
->dlist
) {
1163 struct dl
*d
= ddf
->dlist
;
1164 ddf
->dlist
= d
->next
;
1171 while (ddf
->add_list
) {
1172 struct dl
*d
= ddf
->add_list
;
1173 ddf
->add_list
= d
->next
;
1184 static struct supertype
*match_metadata_desc_ddf(char *arg
)
1186 /* 'ddf' only support containers */
1187 struct supertype
*st
;
1188 if (strcmp(arg
, "ddf") != 0 &&
1189 strcmp(arg
, "default") != 0
1193 st
= xcalloc(1, sizeof(*st
));
1194 st
->ss
= &super_ddf
;
1196 st
->minor_version
= 0;
1203 static mapping_t ddf_state
[] = {
1209 { "Partially Optimal", 5},
1215 static mapping_t ddf_init_state
[] = {
1216 { "Not Initialised", 0},
1217 { "QuickInit in Progress", 1},
1218 { "Fully Initialised", 2},
1222 static mapping_t ddf_access
[] = {
1226 { "Blocked (no access)", 3},
1230 static mapping_t ddf_level
[] = {
1231 { "RAID0", DDF_RAID0
},
1232 { "RAID1", DDF_RAID1
},
1233 { "RAID3", DDF_RAID3
},
1234 { "RAID4", DDF_RAID4
},
1235 { "RAID5", DDF_RAID5
},
1236 { "RAID1E",DDF_RAID1E
},
1237 { "JBOD", DDF_JBOD
},
1238 { "CONCAT",DDF_CONCAT
},
1239 { "RAID5E",DDF_RAID5E
},
1240 { "RAID5EE",DDF_RAID5EE
},
1241 { "RAID6", DDF_RAID6
},
1244 static mapping_t ddf_sec_level
[] = {
1245 { "Striped", DDF_2STRIPED
},
1246 { "Mirrored", DDF_2MIRRORED
},
1247 { "Concat", DDF_2CONCAT
},
1248 { "Spanned", DDF_2SPANNED
},
1253 static int all_ff(const char *guid
)
1256 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
1257 if (guid
[i
] != (char)0xff)
1263 static void print_guid(char *guid
, int tstamp
)
1265 /* A GUIDs are part (or all) ASCII and part binary.
1266 * They tend to be space padded.
1267 * We print the GUID in HEX, then in parentheses add
1268 * any initial ASCII sequence, and a possible
1269 * time stamp from bytes 16-19
1271 int l
= DDF_GUID_LEN
;
1274 for (i
=0 ; i
<DDF_GUID_LEN
; i
++) {
1275 if ((i
&3)==0 && i
!= 0) printf(":");
1276 printf("%02X", guid
[i
]&255);
1280 while (l
&& guid
[l
-1] == ' ')
1282 for (i
=0 ; i
<l
; i
++) {
1283 if (guid
[i
] >= 0x20 && guid
[i
] < 0x7f)
1284 fputc(guid
[i
], stdout
);
1289 time_t then
= __be32_to_cpu(*(__u32
*)(guid
+16)) + DECADE
;
1292 tm
= localtime(&then
);
1293 strftime(tbuf
, 100, " %D %T",tm
);
1294 fputs(tbuf
, stdout
);
1299 static const char *guid_str(const char *guid
)
1301 static char buf
[DDF_GUID_LEN
*2+1];
1304 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
1305 p
+= sprintf(p
, "%02x", (unsigned char)guid
[i
]);
1307 return (const char *) buf
;
1310 static void examine_vd(int n
, struct ddf_super
*sb
, char *guid
)
1312 int crl
= sb
->conf_rec_len
;
1315 for (vcl
= sb
->conflist
; vcl
; vcl
= vcl
->next
) {
1317 struct vd_config
*vc
= &vcl
->conf
;
1319 if (calc_crc(vc
, crl
*512) != vc
->crc
)
1321 if (memcmp(vc
->guid
, guid
, DDF_GUID_LEN
) != 0)
1324 /* Ok, we know about this VD, let's give more details */
1325 printf(" Raid Devices[%d] : %d (", n
,
1326 __be16_to_cpu(vc
->prim_elmnt_count
));
1327 for (i
= 0; i
< __be16_to_cpu(vc
->prim_elmnt_count
); i
++) {
1329 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1330 for (j
=0; j
<cnt
; j
++)
1331 if (vc
->phys_refnum
[i
] == sb
->phys
->entries
[j
].refnum
)
1340 if (vc
->chunk_shift
!= 255)
1341 printf(" Chunk Size[%d] : %d sectors\n", n
,
1342 1 << vc
->chunk_shift
);
1343 printf(" Raid Level[%d] : %s\n", n
,
1344 map_num(ddf_level
, vc
->prl
)?:"-unknown-");
1345 if (vc
->sec_elmnt_count
!= 1) {
1346 printf(" Secondary Position[%d] : %d of %d\n", n
,
1347 vc
->sec_elmnt_seq
, vc
->sec_elmnt_count
);
1348 printf(" Secondary Level[%d] : %s\n", n
,
1349 map_num(ddf_sec_level
, vc
->srl
) ?: "-unknown-");
1351 printf(" Device Size[%d] : %llu\n", n
,
1352 (unsigned long long)__be64_to_cpu(vc
->blocks
)/2);
1353 printf(" Array Size[%d] : %llu\n", n
,
1354 (unsigned long long)__be64_to_cpu(vc
->array_blocks
)/2);
1358 static void examine_vds(struct ddf_super
*sb
)
1360 int cnt
= __be16_to_cpu(sb
->virt
->populated_vdes
);
1362 printf(" Virtual Disks : %d\n", cnt
);
1364 for (i
= 0; i
< __be16_to_cpu(sb
->virt
->max_vdes
); i
++) {
1365 struct virtual_entry
*ve
= &sb
->virt
->entries
[i
];
1366 if (all_ff(ve
->guid
))
1369 printf(" VD GUID[%d] : ", i
); print_guid(ve
->guid
, 1);
1371 printf(" unit[%d] : %d\n", i
, __be16_to_cpu(ve
->unit
));
1372 printf(" state[%d] : %s, %s%s\n", i
,
1373 map_num(ddf_state
, ve
->state
& 7),
1374 (ve
->state
& 8) ? "Morphing, ": "",
1375 (ve
->state
& 16)? "Not Consistent" : "Consistent");
1376 printf(" init state[%d] : %s\n", i
,
1377 map_num(ddf_init_state
, ve
->init_state
&3));
1378 printf(" access[%d] : %s\n", i
,
1379 map_num(ddf_access
, (ve
->init_state
>>6) & 3));
1380 printf(" Name[%d] : %.16s\n", i
, ve
->name
);
1381 examine_vd(i
, sb
, ve
->guid
);
1383 if (cnt
) printf("\n");
1386 static void examine_pds(struct ddf_super
*sb
)
1388 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1391 printf(" Physical Disks : %d\n", cnt
);
1392 printf(" Number RefNo Size Device Type/State\n");
1394 for (i
=0 ; i
<cnt
; i
++) {
1395 struct phys_disk_entry
*pd
= &sb
->phys
->entries
[i
];
1396 int type
= __be16_to_cpu(pd
->type
);
1397 int state
= __be16_to_cpu(pd
->state
);
1399 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1401 printf(" %3d %08x ", i
,
1402 __be32_to_cpu(pd
->refnum
));
1404 (unsigned long long)__be64_to_cpu(pd
->config_size
)>>1);
1405 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1406 if (dl
->disk
.refnum
== pd
->refnum
) {
1407 char *dv
= map_dev(dl
->major
, dl
->minor
, 0);
1409 printf("%-15s", dv
);
1416 printf(" %s%s%s%s%s",
1417 (type
&2) ? "active":"",
1418 (type
&4) ? "Global-Spare":"",
1419 (type
&8) ? "spare" : "",
1420 (type
&16)? ", foreign" : "",
1421 (type
&32)? "pass-through" : "");
1422 if (state
& DDF_Failed
)
1423 /* This over-rides these three */
1424 state
&= ~(DDF_Online
|DDF_Rebuilding
|DDF_Transition
);
1425 printf("/%s%s%s%s%s%s%s",
1426 (state
&1)? "Online": "Offline",
1427 (state
&2)? ", Failed": "",
1428 (state
&4)? ", Rebuilding": "",
1429 (state
&8)? ", in-transition": "",
1430 (state
&16)? ", SMART-errors": "",
1431 (state
&32)? ", Unrecovered-Read-Errors": "",
1432 (state
&64)? ", Missing" : "");
1437 static void examine_super_ddf(struct supertype
*st
, char *homehost
)
1439 struct ddf_super
*sb
= st
->sb
;
1441 printf(" Magic : %08x\n", __be32_to_cpu(sb
->anchor
.magic
));
1442 printf(" Version : %.8s\n", sb
->anchor
.revision
);
1443 printf("Controller GUID : "); print_guid(sb
->controller
.guid
, 0);
1445 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1447 printf(" Seq : %08x\n", __be32_to_cpu(sb
->active
->seq
));
1448 printf(" Redundant hdr : %s\n", sb
->secondary
.magic
== DDF_HEADER_MAGIC
1454 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1456 static void uuid_from_ddf_guid(const char *guid
, int uuid
[4]);
1457 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4]);
1459 static unsigned int get_vd_num_of_subarray(struct supertype
*st
)
1462 * Figure out the VD number for this supertype.
1463 * Returns DDF_CONTAINER for the container itself,
1464 * and DDF_NOTFOUND on error.
1466 struct ddf_super
*ddf
= st
->sb
;
1471 if (*st
->container_devnm
== '\0')
1472 return DDF_CONTAINER
;
1474 sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
1475 if (!sra
|| sra
->array
.major_version
!= -1 ||
1476 sra
->array
.minor_version
!= -2 ||
1477 !is_subarray(sra
->text_version
))
1478 return DDF_NOTFOUND
;
1480 sub
= strchr(sra
->text_version
+ 1, '/');
1482 vcnum
= strtoul(sub
+ 1, &end
, 10);
1483 if (sub
== NULL
|| *sub
== '\0' || *end
!= '\0' ||
1484 vcnum
>= __be16_to_cpu(ddf
->active
->max_vd_entries
))
1485 return DDF_NOTFOUND
;
1490 static void brief_examine_super_ddf(struct supertype
*st
, int verbose
)
1492 /* We just write a generic DDF ARRAY entry
1496 getinfo_super_ddf(st
, &info
, NULL
);
1497 fname_from_uuid(st
, &info
, nbuf
, ':');
1499 printf("ARRAY metadata=ddf UUID=%s\n", nbuf
+ 5);
1502 static void brief_examine_subarrays_ddf(struct supertype
*st
, int verbose
)
1504 /* We just write a generic DDF ARRAY entry
1506 struct ddf_super
*ddf
= st
->sb
;
1510 getinfo_super_ddf(st
, &info
, NULL
);
1511 fname_from_uuid(st
, &info
, nbuf
, ':');
1513 for (i
= 0; i
< __be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
1514 struct virtual_entry
*ve
= &ddf
->virt
->entries
[i
];
1517 if (all_ff(ve
->guid
))
1519 memcpy(vcl
.conf
.guid
, ve
->guid
, DDF_GUID_LEN
);
1520 ddf
->currentconf
=&vcl
;
1521 uuid_from_super_ddf(st
, info
.uuid
);
1522 fname_from_uuid(st
, &info
, nbuf1
, ':');
1523 printf("ARRAY container=%s member=%d UUID=%s\n",
1524 nbuf
+5, i
, nbuf1
+5);
1528 static void export_examine_super_ddf(struct supertype
*st
)
1532 getinfo_super_ddf(st
, &info
, NULL
);
1533 fname_from_uuid(st
, &info
, nbuf
, ':');
1534 printf("MD_METADATA=ddf\n");
1535 printf("MD_LEVEL=container\n");
1536 printf("MD_UUID=%s\n", nbuf
+5);
1539 static int copy_metadata_ddf(struct supertype
*st
, int from
, int to
)
1542 unsigned long long dsize
, offset
;
1544 struct ddf_header
*ddf
;
1547 /* The meta consists of an anchor, a primary, and a secondary.
1548 * This all lives at the end of the device.
1549 * So it is easiest to find the earliest of primary and
1550 * secondary, and copy everything from there.
1552 * Anchor is 512 from end It contains primary_lba and secondary_lba
1553 * we choose one of those
1556 if (posix_memalign(&buf
, 4096, 4096) != 0)
1559 if (!get_dev_size(from
, NULL
, &dsize
))
1562 if (lseek64(from
, dsize
-512, 0) < 0)
1564 if (read(from
, buf
, 512) != 512)
1567 if (ddf
->magic
!= DDF_HEADER_MAGIC
||
1568 calc_crc(ddf
, 512) != ddf
->crc
||
1569 (memcmp(ddf
->revision
, DDF_REVISION_0
, 8) != 0 &&
1570 memcmp(ddf
->revision
, DDF_REVISION_2
, 8) != 0))
1573 offset
= dsize
- 512;
1574 if ((__be64_to_cpu(ddf
->primary_lba
) << 9) < offset
)
1575 offset
= __be64_to_cpu(ddf
->primary_lba
) << 9;
1576 if ((__be64_to_cpu(ddf
->secondary_lba
) << 9) < offset
)
1577 offset
= __be64_to_cpu(ddf
->secondary_lba
) << 9;
1579 bytes
= dsize
- offset
;
1581 if (lseek64(from
, offset
, 0) < 0 ||
1582 lseek64(to
, offset
, 0) < 0)
1584 while (written
< bytes
) {
1585 int n
= bytes
- written
;
1588 if (read(from
, buf
, n
) != n
)
1590 if (write(to
, buf
, n
) != n
)
1601 static void detail_super_ddf(struct supertype
*st
, char *homehost
)
1604 * Could print DDF GUID
1605 * Need to find which array
1606 * If whole, briefly list all arrays
1611 static void brief_detail_super_ddf(struct supertype
*st
)
1615 struct ddf_super
*ddf
= st
->sb
;
1616 unsigned int vcnum
= get_vd_num_of_subarray(st
);
1617 if (vcnum
== DDF_CONTAINER
)
1618 uuid_from_super_ddf(st
, info
.uuid
);
1619 else if (vcnum
== DDF_NOTFOUND
)
1622 uuid_from_ddf_guid(ddf
->virt
->entries
[vcnum
].guid
, info
.uuid
);
1623 fname_from_uuid(st
, &info
, nbuf
,':');
1624 printf(" UUID=%s", nbuf
+ 5);
1628 static int match_home_ddf(struct supertype
*st
, char *homehost
)
1630 /* It matches 'this' host if the controller is a
1631 * Linux-MD controller with vendor_data matching
1634 struct ddf_super
*ddf
= st
->sb
;
1639 len
= strlen(homehost
);
1641 return (memcmp(ddf
->controller
.guid
, T10
, 8) == 0 &&
1642 len
< sizeof(ddf
->controller
.vendor_data
) &&
1643 memcmp(ddf
->controller
.vendor_data
, homehost
,len
) == 0 &&
1644 ddf
->controller
.vendor_data
[len
] == 0);
1648 static int find_index_in_bvd(const struct ddf_super
*ddf
,
1649 const struct vd_config
*conf
, unsigned int n
,
1650 unsigned int *n_bvd
)
1653 * Find the index of the n-th valid physical disk in this BVD
1656 for (i
= 0, j
= 0; i
< ddf
->mppe
&&
1657 j
< __be16_to_cpu(conf
->prim_elmnt_count
); i
++) {
1658 if (conf
->phys_refnum
[i
] != 0xffffffff) {
1666 dprintf("%s: couldn't find BVD member %u (total %u)\n",
1667 __func__
, n
, __be16_to_cpu(conf
->prim_elmnt_count
));
1671 static struct vd_config
*find_vdcr(struct ddf_super
*ddf
, unsigned int inst
,
1673 unsigned int *n_bvd
, struct vcl
**vcl
)
1677 for (v
= ddf
->conflist
; v
; v
= v
->next
) {
1678 unsigned int nsec
, ibvd
;
1679 struct vd_config
*conf
;
1680 if (inst
!= v
->vcnum
)
1683 if (conf
->sec_elmnt_count
== 1) {
1684 if (find_index_in_bvd(ddf
, conf
, n
, n_bvd
)) {
1690 if (v
->other_bvds
== NULL
) {
1691 pr_err("%s: BUG: other_bvds is NULL, nsec=%u\n",
1692 __func__
, conf
->sec_elmnt_count
);
1695 nsec
= n
/ __be16_to_cpu(conf
->prim_elmnt_count
);
1696 if (conf
->sec_elmnt_seq
!= nsec
) {
1697 for (ibvd
= 1; ibvd
< conf
->sec_elmnt_count
; ibvd
++) {
1698 if (v
->other_bvds
[ibvd
-1]->sec_elmnt_seq
1702 if (ibvd
== conf
->sec_elmnt_count
)
1704 conf
= v
->other_bvds
[ibvd
-1];
1706 if (!find_index_in_bvd(ddf
, conf
,
1707 n
- nsec
*conf
->sec_elmnt_count
, n_bvd
))
1709 dprintf("%s: found disk %u as member %u in bvd %d of array %u\n"
1710 , __func__
, n
, *n_bvd
, ibvd
-1, inst
);
1715 pr_err("%s: Could't find disk %d in array %u\n", __func__
, n
, inst
);
1720 static int find_phys(const struct ddf_super
*ddf
, __u32 phys_refnum
)
1722 /* Find the entry in phys_disk which has the given refnum
1723 * and return it's index
1726 for (i
= 0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
); i
++)
1727 if (ddf
->phys
->entries
[i
].refnum
== phys_refnum
)
1732 static void uuid_from_ddf_guid(const char *guid
, int uuid
[4])
1735 struct sha1_ctx ctx
;
1736 sha1_init_ctx(&ctx
);
1737 sha1_process_bytes(guid
, DDF_GUID_LEN
, &ctx
);
1738 sha1_finish_ctx(&ctx
, buf
);
1739 memcpy(uuid
, buf
, 4*4);
1742 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4])
1744 /* The uuid returned here is used for:
1745 * uuid to put into bitmap file (Create, Grow)
1746 * uuid for backup header when saving critical section (Grow)
1747 * comparing uuids when re-adding a device into an array
1748 * In these cases the uuid required is that of the data-array,
1749 * not the device-set.
1750 * uuid to recognise same set when adding a missing device back
1751 * to an array. This is a uuid for the device-set.
1753 * For each of these we can make do with a truncated
1754 * or hashed uuid rather than the original, as long as
1756 * In the case of SVD we assume the BVD is of interest,
1757 * though that might be the case if a bitmap were made for
1758 * a mirrored SVD - worry about that later.
1759 * So we need to find the VD configuration record for the
1760 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1761 * The first 16 bytes of the sha1 of these is used.
1763 struct ddf_super
*ddf
= st
->sb
;
1764 struct vcl
*vcl
= ddf
->currentconf
;
1768 guid
= vcl
->conf
.guid
;
1770 guid
= ddf
->anchor
.guid
;
1771 uuid_from_ddf_guid(guid
, uuid
);
1774 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1776 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1778 struct ddf_super
*ddf
= st
->sb
;
1779 int map_disks
= info
->array
.raid_disks
;
1782 if (ddf
->currentconf
) {
1783 getinfo_super_ddf_bvd(st
, info
, map
);
1786 memset(info
, 0, sizeof(*info
));
1788 info
->array
.raid_disks
= __be16_to_cpu(ddf
->phys
->used_pdes
);
1789 info
->array
.level
= LEVEL_CONTAINER
;
1790 info
->array
.layout
= 0;
1791 info
->array
.md_minor
= -1;
1792 cptr
= (__u32
*)(ddf
->anchor
.guid
+ 16);
1793 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
1795 info
->array
.utime
= 0;
1796 info
->array
.chunk_size
= 0;
1797 info
->container_enough
= 1;
1799 info
->disk
.major
= 0;
1800 info
->disk
.minor
= 0;
1802 info
->disk
.number
= __be32_to_cpu(ddf
->dlist
->disk
.refnum
);
1803 info
->disk
.raid_disk
= find_phys(ddf
, ddf
->dlist
->disk
.refnum
);
1805 info
->data_offset
= __be64_to_cpu(ddf
->phys
->
1806 entries
[info
->disk
.raid_disk
].
1808 info
->component_size
= ddf
->dlist
->size
- info
->data_offset
;
1810 info
->disk
.number
= -1;
1811 info
->disk
.raid_disk
= -1;
1812 // info->disk.raid_disk = find refnum in the table and use index;
1814 info
->disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
1816 info
->recovery_start
= MaxSector
;
1817 info
->reshape_active
= 0;
1818 info
->recovery_blocked
= 0;
1821 info
->array
.major_version
= -1;
1822 info
->array
.minor_version
= -2;
1823 strcpy(info
->text_version
, "ddf");
1824 info
->safe_mode_delay
= 0;
1826 uuid_from_super_ddf(st
, info
->uuid
);
1830 for (i
= 0 ; i
< map_disks
; i
++) {
1831 if (i
< info
->array
.raid_disks
&&
1832 (__be16_to_cpu(ddf
->phys
->entries
[i
].state
) & DDF_Online
) &&
1833 !(__be16_to_cpu(ddf
->phys
->entries
[i
].state
) & DDF_Failed
))
1841 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1843 struct ddf_super
*ddf
= st
->sb
;
1844 struct vcl
*vc
= ddf
->currentconf
;
1845 int cd
= ddf
->currentdev
;
1849 int map_disks
= info
->array
.raid_disks
;
1851 struct vd_config
*conf
;
1853 memset(info
, 0, sizeof(*info
));
1854 if (layout_ddf2md(&vc
->conf
, &info
->array
) == -1)
1856 info
->array
.md_minor
= -1;
1857 cptr
= (__u32
*)(vc
->conf
.guid
+ 16);
1858 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
1859 info
->array
.utime
= DECADE
+ __be32_to_cpu(vc
->conf
.timestamp
);
1860 info
->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
1861 info
->custom_array_size
= 0;
1864 n_prim
= __be16_to_cpu(conf
->prim_elmnt_count
);
1865 if (conf
->sec_elmnt_count
> 1 && cd
>= n_prim
) {
1866 int ibvd
= cd
/ n_prim
- 1;
1868 conf
= vc
->other_bvds
[ibvd
];
1871 if (cd
>= 0 && (unsigned)cd
< ddf
->mppe
) {
1873 __be64_to_cpu(LBA_OFFSET(ddf
, &vc
->conf
)[cd
]);
1874 if (vc
->block_sizes
)
1875 info
->component_size
= vc
->block_sizes
[cd
];
1877 info
->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
1880 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
1881 if (dl
->raiddisk
== ddf
->currentdev
)
1884 info
->disk
.major
= 0;
1885 info
->disk
.minor
= 0;
1886 info
->disk
.state
= 0;
1888 info
->disk
.major
= dl
->major
;
1889 info
->disk
.minor
= dl
->minor
;
1890 info
->disk
.raid_disk
= dl
->raiddisk
;
1891 info
->disk
.number
= dl
->pdnum
;
1892 info
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
1895 info
->container_member
= ddf
->currentconf
->vcnum
;
1897 info
->recovery_start
= MaxSector
;
1898 info
->resync_start
= 0;
1899 info
->reshape_active
= 0;
1900 info
->recovery_blocked
= 0;
1901 if (!(ddf
->virt
->entries
[info
->container_member
].state
1902 & DDF_state_inconsistent
) &&
1903 (ddf
->virt
->entries
[info
->container_member
].init_state
1904 & DDF_initstate_mask
)
1906 info
->resync_start
= MaxSector
;
1908 uuid_from_super_ddf(st
, info
->uuid
);
1910 info
->array
.major_version
= -1;
1911 info
->array
.minor_version
= -2;
1912 sprintf(info
->text_version
, "/%s/%d",
1913 st
->container_devnm
,
1914 info
->container_member
);
1915 info
->safe_mode_delay
= 200;
1917 memcpy(info
->name
, ddf
->virt
->entries
[info
->container_member
].name
, 16);
1920 if (info
->name
[j
] == ' ')
1924 for (j
= 0; j
< map_disks
; j
++) {
1926 if (j
< info
->array
.raid_disks
) {
1927 int i
= find_phys(ddf
, vc
->conf
.phys_refnum
[j
]);
1929 (__be16_to_cpu(ddf
->phys
->entries
[i
].state
) & DDF_Online
) &&
1930 !(__be16_to_cpu(ddf
->phys
->entries
[i
].state
) & DDF_Failed
))
1936 static int update_super_ddf(struct supertype
*st
, struct mdinfo
*info
,
1938 char *devname
, int verbose
,
1939 int uuid_set
, char *homehost
)
1941 /* For 'assemble' and 'force' we need to return non-zero if any
1942 * change was made. For others, the return value is ignored.
1943 * Update options are:
1944 * force-one : This device looks a bit old but needs to be included,
1945 * update age info appropriately.
1946 * assemble: clear any 'faulty' flag to allow this device to
1948 * force-array: Array is degraded but being forced, mark it clean
1949 * if that will be needed to assemble it.
1951 * newdev: not used ????
1952 * grow: Array has gained a new device - this is currently for
1954 * resync: mark as dirty so a resync will happen.
1955 * uuid: Change the uuid of the array to match what is given
1956 * homehost: update the recorded homehost
1957 * name: update the name - preserving the homehost
1958 * _reshape_progress: record new reshape_progress position.
1960 * Following are not relevant for this version:
1961 * sparc2.2 : update from old dodgey metadata
1962 * super-minor: change the preferred_minor number
1963 * summaries: update redundant counters.
1966 // struct ddf_super *ddf = st->sb;
1967 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1968 // struct virtual_entry *ve = find_ve(ddf);
1970 /* we don't need to handle "force-*" or "assemble" as
1971 * there is no need to 'trick' the kernel. We the metadata is
1972 * first updated to activate the array, all the implied modifications
1976 if (strcmp(update
, "grow") == 0) {
1978 } else if (strcmp(update
, "resync") == 0) {
1979 // info->resync_checkpoint = 0;
1980 } else if (strcmp(update
, "homehost") == 0) {
1981 /* homehost is stored in controller->vendor_data,
1982 * or it is when we are the vendor
1984 // if (info->vendor_is_local)
1985 // strcpy(ddf->controller.vendor_data, homehost);
1987 } else if (strcmp(update
, "name") == 0) {
1988 /* name is stored in virtual_entry->name */
1989 // memset(ve->name, ' ', 16);
1990 // strncpy(ve->name, info->name, 16);
1992 } else if (strcmp(update
, "_reshape_progress") == 0) {
1993 /* We don't support reshape yet */
1994 } else if (strcmp(update
, "assemble") == 0 ) {
1995 /* Do nothing, just succeed */
2000 // update_all_csum(ddf);
2005 static void make_header_guid(char *guid
)
2008 /* Create a DDF Header of Virtual Disk GUID */
2010 /* 24 bytes of fiction required.
2011 * first 8 are a 'vendor-id' - "Linux-MD"
2012 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
2013 * Remaining 8 random number plus timestamp
2015 memcpy(guid
, T10
, sizeof(T10
));
2016 stamp
= __cpu_to_be32(0xdeadbeef);
2017 memcpy(guid
+8, &stamp
, 4);
2018 stamp
= __cpu_to_be32(0);
2019 memcpy(guid
+12, &stamp
, 4);
2020 stamp
= __cpu_to_be32(time(0) - DECADE
);
2021 memcpy(guid
+16, &stamp
, 4);
2023 memcpy(guid
+20, &stamp
, 4);
2026 static unsigned int find_unused_vde(const struct ddf_super
*ddf
)
2029 for (i
= 0; i
< __be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
2030 if (all_ff(ddf
->virt
->entries
[i
].guid
))
2033 return DDF_NOTFOUND
;
2036 static unsigned int find_vde_by_name(const struct ddf_super
*ddf
,
2041 return DDF_NOTFOUND
;
2042 for (i
= 0; i
< __be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
2043 if (all_ff(ddf
->virt
->entries
[i
].guid
))
2045 if (!strncmp(name
, ddf
->virt
->entries
[i
].name
,
2046 sizeof(ddf
->virt
->entries
[i
].name
)))
2049 return DDF_NOTFOUND
;
2052 static unsigned int find_vde_by_guid(const struct ddf_super
*ddf
,
2056 if (guid
== NULL
|| all_ff(guid
))
2057 return DDF_NOTFOUND
;
2058 for (i
= 0; i
< __be16_to_cpu(ddf
->virt
->max_vdes
); i
++)
2059 if (!memcmp(ddf
->virt
->entries
[i
].guid
, guid
, DDF_GUID_LEN
))
2061 return DDF_NOTFOUND
;
2064 static int init_super_ddf_bvd(struct supertype
*st
,
2065 mdu_array_info_t
*info
,
2066 unsigned long long size
,
2067 char *name
, char *homehost
,
2068 int *uuid
, unsigned long long data_offset
);
2070 static int init_super_ddf(struct supertype
*st
,
2071 mdu_array_info_t
*info
,
2072 unsigned long long size
, char *name
, char *homehost
,
2073 int *uuid
, unsigned long long data_offset
)
2075 /* This is primarily called by Create when creating a new array.
2076 * We will then get add_to_super called for each component, and then
2077 * write_init_super called to write it out to each device.
2078 * For DDF, Create can create on fresh devices or on a pre-existing
2080 * To create on a pre-existing array a different method will be called.
2081 * This one is just for fresh drives.
2083 * We need to create the entire 'ddf' structure which includes:
2084 * DDF headers - these are easy.
2085 * Controller data - a Sector describing this controller .. not that
2086 * this is a controller exactly.
2087 * Physical Disk Record - one entry per device, so
2088 * leave plenty of space.
2089 * Virtual Disk Records - again, just leave plenty of space.
2090 * This just lists VDs, doesn't give details
2091 * Config records - describes the VDs that use this disk
2092 * DiskData - describes 'this' device.
2093 * BadBlockManagement - empty
2094 * Diag Space - empty
2095 * Vendor Logs - Could we put bitmaps here?
2098 struct ddf_super
*ddf
;
2101 int max_phys_disks
, max_virt_disks
;
2102 unsigned long long sector
;
2106 struct phys_disk
*pd
;
2107 struct virtual_disk
*vd
;
2109 if (data_offset
!= INVALID_SECTORS
) {
2110 pr_err("data-offset not supported by DDF\n");
2115 return init_super_ddf_bvd(st
, info
, size
, name
, homehost
, uuid
,
2118 if (posix_memalign((void**)&ddf
, 512, sizeof(*ddf
)) != 0) {
2119 pr_err("%s could not allocate superblock\n", __func__
);
2122 memset(ddf
, 0, sizeof(*ddf
));
2123 ddf
->dlist
= NULL
; /* no physical disks yet */
2124 ddf
->conflist
= NULL
; /* No virtual disks yet */
2128 /* zeroing superblock */
2132 /* At least 32MB *must* be reserved for the ddf. So let's just
2133 * start 32MB from the end, and put the primary header there.
2134 * Don't do secondary for now.
2135 * We don't know exactly where that will be yet as it could be
2136 * different on each device. To just set up the lengths.
2140 ddf
->anchor
.magic
= DDF_HEADER_MAGIC
;
2141 make_header_guid(ddf
->anchor
.guid
);
2143 memcpy(ddf
->anchor
.revision
, DDF_REVISION_2
, 8);
2144 ddf
->anchor
.seq
= __cpu_to_be32(1);
2145 ddf
->anchor
.timestamp
= __cpu_to_be32(time(0) - DECADE
);
2146 ddf
->anchor
.openflag
= 0xFF;
2147 ddf
->anchor
.foreignflag
= 0;
2148 ddf
->anchor
.enforcegroups
= 0; /* Is this best?? */
2149 ddf
->anchor
.pad0
= 0xff;
2150 memset(ddf
->anchor
.pad1
, 0xff, 12);
2151 memset(ddf
->anchor
.header_ext
, 0xff, 32);
2152 ddf
->anchor
.primary_lba
= ~(__u64
)0;
2153 ddf
->anchor
.secondary_lba
= ~(__u64
)0;
2154 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
2155 memset(ddf
->anchor
.pad2
, 0xff, 3);
2156 ddf
->anchor
.workspace_len
= __cpu_to_be32(32768); /* Must be reserved */
2157 ddf
->anchor
.workspace_lba
= ~(__u64
)0; /* Put this at bottom
2158 of 32M reserved.. */
2159 max_phys_disks
= 1023; /* Should be enough */
2160 ddf
->anchor
.max_pd_entries
= __cpu_to_be16(max_phys_disks
);
2161 max_virt_disks
= 255;
2162 ddf
->anchor
.max_vd_entries
= __cpu_to_be16(max_virt_disks
); /* ?? */
2163 ddf
->anchor
.max_partitions
= __cpu_to_be16(64); /* ?? */
2166 ddf
->conf_rec_len
= 1 + ROUND_UP(ddf
->mppe
* (4+8), 512)/512;
2167 ddf
->anchor
.config_record_len
= __cpu_to_be16(ddf
->conf_rec_len
);
2168 ddf
->anchor
.max_primary_element_entries
= __cpu_to_be16(ddf
->mppe
);
2169 memset(ddf
->anchor
.pad3
, 0xff, 54);
2170 /* controller sections is one sector long immediately
2171 * after the ddf header */
2173 ddf
->anchor
.controller_section_offset
= __cpu_to_be32(sector
);
2174 ddf
->anchor
.controller_section_length
= __cpu_to_be32(1);
2177 /* phys is 8 sectors after that */
2178 pdsize
= ROUND_UP(sizeof(struct phys_disk
) +
2179 sizeof(struct phys_disk_entry
)*max_phys_disks
,
2181 switch(pdsize
/512) {
2182 case 2: case 8: case 32: case 128: case 512: break;
2185 ddf
->anchor
.phys_section_offset
= __cpu_to_be32(sector
);
2186 ddf
->anchor
.phys_section_length
=
2187 __cpu_to_be32(pdsize
/512); /* max_primary_element_entries/8 */
2188 sector
+= pdsize
/512;
2190 /* virt is another 32 sectors */
2191 vdsize
= ROUND_UP(sizeof(struct virtual_disk
) +
2192 sizeof(struct virtual_entry
) * max_virt_disks
,
2194 switch(vdsize
/512) {
2195 case 2: case 8: case 32: case 128: case 512: break;
2198 ddf
->anchor
.virt_section_offset
= __cpu_to_be32(sector
);
2199 ddf
->anchor
.virt_section_length
=
2200 __cpu_to_be32(vdsize
/512); /* max_vd_entries/8 */
2201 sector
+= vdsize
/512;
2203 clen
= ddf
->conf_rec_len
* (ddf
->max_part
+1);
2204 ddf
->anchor
.config_section_offset
= __cpu_to_be32(sector
);
2205 ddf
->anchor
.config_section_length
= __cpu_to_be32(clen
);
2208 ddf
->anchor
.data_section_offset
= __cpu_to_be32(sector
);
2209 ddf
->anchor
.data_section_length
= __cpu_to_be32(1);
2212 ddf
->anchor
.bbm_section_length
= __cpu_to_be32(0);
2213 ddf
->anchor
.bbm_section_offset
= __cpu_to_be32(0xFFFFFFFF);
2214 ddf
->anchor
.diag_space_length
= __cpu_to_be32(0);
2215 ddf
->anchor
.diag_space_offset
= __cpu_to_be32(0xFFFFFFFF);
2216 ddf
->anchor
.vendor_length
= __cpu_to_be32(0);
2217 ddf
->anchor
.vendor_offset
= __cpu_to_be32(0xFFFFFFFF);
2219 memset(ddf
->anchor
.pad4
, 0xff, 256);
2221 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2222 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2224 ddf
->primary
.openflag
= 1; /* I guess.. */
2225 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
2227 ddf
->secondary
.openflag
= 1; /* I guess.. */
2228 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
2230 ddf
->active
= &ddf
->primary
;
2232 ddf
->controller
.magic
= DDF_CONTROLLER_MAGIC
;
2234 /* 24 more bytes of fiction required.
2235 * first 8 are a 'vendor-id' - "Linux-MD"
2236 * Remaining 16 are serial number.... maybe a hostname would do?
2238 memcpy(ddf
->controller
.guid
, T10
, sizeof(T10
));
2239 gethostname(hostname
, sizeof(hostname
));
2240 hostname
[sizeof(hostname
) - 1] = 0;
2241 hostlen
= strlen(hostname
);
2242 memcpy(ddf
->controller
.guid
+ 24 - hostlen
, hostname
, hostlen
);
2243 for (i
= strlen(T10
) ; i
+hostlen
< 24; i
++)
2244 ddf
->controller
.guid
[i
] = ' ';
2246 ddf
->controller
.type
.vendor_id
= __cpu_to_be16(0xDEAD);
2247 ddf
->controller
.type
.device_id
= __cpu_to_be16(0xBEEF);
2248 ddf
->controller
.type
.sub_vendor_id
= 0;
2249 ddf
->controller
.type
.sub_device_id
= 0;
2250 memcpy(ddf
->controller
.product_id
, "What Is My PID??", 16);
2251 memset(ddf
->controller
.pad
, 0xff, 8);
2252 memset(ddf
->controller
.vendor_data
, 0xff, 448);
2253 if (homehost
&& strlen(homehost
) < 440)
2254 strcpy((char*)ddf
->controller
.vendor_data
, homehost
);
2256 if (posix_memalign((void**)&pd
, 512, pdsize
) != 0) {
2257 pr_err("%s could not allocate pd\n", __func__
);
2261 ddf
->pdsize
= pdsize
;
2263 memset(pd
, 0xff, pdsize
);
2264 memset(pd
, 0, sizeof(*pd
));
2265 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2266 pd
->used_pdes
= __cpu_to_be16(0);
2267 pd
->max_pdes
= __cpu_to_be16(max_phys_disks
);
2268 memset(pd
->pad
, 0xff, 52);
2269 for (i
= 0; i
< max_phys_disks
; i
++)
2270 memset(pd
->entries
[i
].guid
, 0xff, DDF_GUID_LEN
);
2272 if (posix_memalign((void**)&vd
, 512, vdsize
) != 0) {
2273 pr_err("%s could not allocate vd\n", __func__
);
2277 ddf
->vdsize
= vdsize
;
2278 memset(vd
, 0, vdsize
);
2279 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
2280 vd
->populated_vdes
= __cpu_to_be16(0);
2281 vd
->max_vdes
= __cpu_to_be16(max_virt_disks
);
2282 memset(vd
->pad
, 0xff, 52);
2284 for (i
=0; i
<max_virt_disks
; i
++)
2285 memset(&vd
->entries
[i
], 0xff, sizeof(struct virtual_entry
));
2288 ddf_set_updates_pending(ddf
);
2292 static int chunk_to_shift(int chunksize
)
2294 return ffs(chunksize
/512)-1;
2299 unsigned long long start
, size
;
2301 static int cmp_extent(const void *av
, const void *bv
)
2303 const struct extent
*a
= av
;
2304 const struct extent
*b
= bv
;
2305 if (a
->start
< b
->start
)
2307 if (a
->start
> b
->start
)
2312 static struct extent
*get_extents(struct ddf_super
*ddf
, struct dl
*dl
)
2314 /* find a list of used extents on the give physical device
2315 * (dnum) of the given ddf.
2316 * Return a malloced array of 'struct extent'
2318 * FIXME ignore DDF_Legacy devices?
2325 rv
= xmalloc(sizeof(struct extent
) * (ddf
->max_part
+ 2));
2327 for (i
= 0; i
< ddf
->max_part
; i
++) {
2328 const struct vd_config
*bvd
;
2330 struct vcl
*v
= dl
->vlist
[i
];
2332 get_pd_index_from_refnum(v
, dl
->disk
.refnum
, ddf
->mppe
,
2333 &bvd
, &ibvd
) == DDF_NOTFOUND
)
2335 rv
[n
].start
= __be64_to_cpu(LBA_OFFSET(ddf
, bvd
)[ibvd
]);
2336 rv
[n
].size
= __be64_to_cpu(bvd
->blocks
);
2339 qsort(rv
, n
, sizeof(*rv
), cmp_extent
);
2341 rv
[n
].start
= __be64_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].config_size
);
2347 static int init_super_ddf_bvd(struct supertype
*st
,
2348 mdu_array_info_t
*info
,
2349 unsigned long long size
,
2350 char *name
, char *homehost
,
2351 int *uuid
, unsigned long long data_offset
)
2353 /* We are creating a BVD inside a pre-existing container.
2354 * so st->sb is already set.
2355 * We need to create a new vd_config and a new virtual_entry
2357 struct ddf_super
*ddf
= st
->sb
;
2358 unsigned int venum
, i
;
2359 struct virtual_entry
*ve
;
2361 struct vd_config
*vc
;
2363 if (find_vde_by_name(ddf
, name
) != DDF_NOTFOUND
) {
2364 pr_err("This ddf already has an array called %s\n", name
);
2367 venum
= find_unused_vde(ddf
);
2368 if (venum
== DDF_NOTFOUND
) {
2369 pr_err("Cannot find spare slot for virtual disk\n");
2372 ve
= &ddf
->virt
->entries
[venum
];
2374 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
2375 * timestamp, random number
2377 make_header_guid(ve
->guid
);
2378 ve
->unit
= __cpu_to_be16(info
->md_minor
);
2380 ve
->guid_crc
= crc32(0, (unsigned char*)ddf
->anchor
.guid
, DDF_GUID_LEN
);
2382 ve
->state
= DDF_state_degraded
; /* Will be modified as devices are added */
2383 if (info
->state
& 1) /* clean */
2384 ve
->init_state
= DDF_init_full
;
2386 ve
->init_state
= DDF_init_not
;
2388 memset(ve
->pad1
, 0xff, 14);
2389 memset(ve
->name
, ' ', 16);
2391 strncpy(ve
->name
, name
, 16);
2392 ddf
->virt
->populated_vdes
=
2393 __cpu_to_be16(__be16_to_cpu(ddf
->virt
->populated_vdes
)+1);
2395 /* Now create a new vd_config */
2396 if (posix_memalign((void**)&vcl
, 512,
2397 (offsetof(struct vcl
, conf
) + ddf
->conf_rec_len
* 512)) != 0) {
2398 pr_err("%s could not allocate vd_config\n", __func__
);
2402 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
2405 vc
->magic
= DDF_VD_CONF_MAGIC
;
2406 memcpy(vc
->guid
, ve
->guid
, DDF_GUID_LEN
);
2407 vc
->timestamp
= __cpu_to_be32(time(0)-DECADE
);
2408 vc
->seqnum
= __cpu_to_be32(1);
2409 memset(vc
->pad0
, 0xff, 24);
2410 vc
->chunk_shift
= chunk_to_shift(info
->chunk_size
);
2411 if (layout_md2ddf(info
, vc
) == -1 ||
2412 __be16_to_cpu(vc
->prim_elmnt_count
) > ddf
->mppe
) {
2413 pr_err("%s: unsupported RAID level/layout %d/%d with %d disks\n",
2414 __func__
, info
->level
, info
->layout
, info
->raid_disks
);
2418 vc
->sec_elmnt_seq
= 0;
2419 if (alloc_other_bvds(ddf
, vcl
) != 0) {
2420 pr_err("%s could not allocate other bvds\n",
2425 vc
->blocks
= __cpu_to_be64(info
->size
* 2);
2426 vc
->array_blocks
= __cpu_to_be64(
2427 calc_array_size(info
->level
, info
->raid_disks
, info
->layout
,
2428 info
->chunk_size
, info
->size
*2));
2429 memset(vc
->pad1
, 0xff, 8);
2430 vc
->spare_refs
[0] = 0xffffffff;
2431 vc
->spare_refs
[1] = 0xffffffff;
2432 vc
->spare_refs
[2] = 0xffffffff;
2433 vc
->spare_refs
[3] = 0xffffffff;
2434 vc
->spare_refs
[4] = 0xffffffff;
2435 vc
->spare_refs
[5] = 0xffffffff;
2436 vc
->spare_refs
[6] = 0xffffffff;
2437 vc
->spare_refs
[7] = 0xffffffff;
2438 memset(vc
->cache_pol
, 0, 8);
2440 memset(vc
->pad2
, 0xff, 3);
2441 memset(vc
->pad3
, 0xff, 52);
2442 memset(vc
->pad4
, 0xff, 192);
2443 memset(vc
->v0
, 0xff, 32);
2444 memset(vc
->v1
, 0xff, 32);
2445 memset(vc
->v2
, 0xff, 16);
2446 memset(vc
->v3
, 0xff, 16);
2447 memset(vc
->vendor
, 0xff, 32);
2449 memset(vc
->phys_refnum
, 0xff, 4*ddf
->mppe
);
2450 memset(vc
->phys_refnum
+ddf
->mppe
, 0x00, 8*ddf
->mppe
);
2452 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++) {
2453 memcpy(vcl
->other_bvds
[i
-1], vc
, ddf
->conf_rec_len
* 512);
2454 vcl
->other_bvds
[i
-1]->sec_elmnt_seq
= i
;
2457 vcl
->next
= ddf
->conflist
;
2458 ddf
->conflist
= vcl
;
2459 ddf
->currentconf
= vcl
;
2460 ddf_set_updates_pending(ddf
);
2464 static int get_svd_state(const struct ddf_super
*, const struct vcl
*);
2467 static void add_to_super_ddf_bvd(struct supertype
*st
,
2468 mdu_disk_info_t
*dk
, int fd
, char *devname
)
2470 /* fd and devname identify a device with-in the ddf container (st).
2471 * dk identifies a location in the new BVD.
2472 * We need to find suitable free space in that device and update
2473 * the phys_refnum and lba_offset for the newly created vd_config.
2474 * We might also want to update the type in the phys_disk
2477 * Alternately: fd == -1 and we have already chosen which device to
2478 * use and recorded in dlist->raid_disk;
2481 struct ddf_super
*ddf
= st
->sb
;
2482 struct vd_config
*vc
;
2484 unsigned long long blocks
, pos
, esize
;
2486 unsigned int raid_disk
= dk
->raid_disk
;
2489 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2490 if (dl
->raiddisk
== dk
->raid_disk
)
2493 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2494 if (dl
->major
== dk
->major
&&
2495 dl
->minor
== dk
->minor
)
2498 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
2501 vc
= &ddf
->currentconf
->conf
;
2502 if (vc
->sec_elmnt_count
> 1) {
2503 unsigned int n
= __be16_to_cpu(vc
->prim_elmnt_count
);
2505 vc
= ddf
->currentconf
->other_bvds
[raid_disk
/ n
- 1];
2509 ex
= get_extents(ddf
, dl
);
2514 blocks
= __be64_to_cpu(vc
->blocks
);
2515 if (ddf
->currentconf
->block_sizes
)
2516 blocks
= ddf
->currentconf
->block_sizes
[dk
->raid_disk
];
2519 esize
= ex
[i
].start
- pos
;
2520 if (esize
>= blocks
)
2522 pos
= ex
[i
].start
+ ex
[i
].size
;
2524 } while (ex
[i
-1].size
);
2530 ddf
->currentdev
= dk
->raid_disk
;
2531 vc
->phys_refnum
[raid_disk
] = dl
->disk
.refnum
;
2532 LBA_OFFSET(ddf
, vc
)[raid_disk
] = __cpu_to_be64(pos
);
2534 for (i
= 0; i
< ddf
->max_part
; i
++)
2535 if (dl
->vlist
[i
] == NULL
)
2537 if (i
== ddf
->max_part
)
2539 dl
->vlist
[i
] = ddf
->currentconf
;
2544 dl
->devname
= devname
;
2546 /* Check if we can mark array as optimal yet */
2547 i
= ddf
->currentconf
->vcnum
;
2548 ddf
->virt
->entries
[i
].state
=
2549 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
2550 | get_svd_state(ddf
, ddf
->currentconf
);
2551 ddf
->phys
->entries
[dl
->pdnum
].type
&= ~__cpu_to_be16(DDF_Global_Spare
);
2552 ddf
->phys
->entries
[dl
->pdnum
].type
|= __cpu_to_be16(DDF_Active_in_VD
);
2553 ddf_set_updates_pending(ddf
);
2556 static unsigned int find_unused_pde(const struct ddf_super
*ddf
)
2559 for (i
= 0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
); i
++) {
2560 if (all_ff(ddf
->phys
->entries
[i
].guid
))
2563 return DDF_NOTFOUND
;
2566 /* add a device to a container, either while creating it or while
2567 * expanding a pre-existing container
2569 static int add_to_super_ddf(struct supertype
*st
,
2570 mdu_disk_info_t
*dk
, int fd
, char *devname
,
2571 unsigned long long data_offset
)
2573 struct ddf_super
*ddf
= st
->sb
;
2577 unsigned long long size
;
2578 struct phys_disk_entry
*pde
;
2583 if (ddf
->currentconf
) {
2584 add_to_super_ddf_bvd(st
, dk
, fd
, devname
);
2588 /* This is device numbered dk->number. We need to create
2589 * a phys_disk entry and a more detailed disk_data entry.
2592 n
= find_unused_pde(ddf
);
2593 if (n
== DDF_NOTFOUND
) {
2594 pr_err("%s: No free slot in array, cannot add disk\n",
2598 pde
= &ddf
->phys
->entries
[n
];
2599 get_dev_size(fd
, NULL
, &size
);
2600 if (size
<= 32*1024*1024) {
2601 pr_err("%s: device size must be at least 32MB\n",
2607 if (posix_memalign((void**)&dd
, 512,
2608 sizeof(*dd
) + sizeof(dd
->vlist
[0]) * ddf
->max_part
) != 0) {
2609 pr_err("%s could allocate buffer for new disk, aborting\n",
2613 dd
->major
= major(stb
.st_rdev
);
2614 dd
->minor
= minor(stb
.st_rdev
);
2615 dd
->devname
= devname
;
2619 dd
->disk
.magic
= DDF_PHYS_DATA_MAGIC
;
2621 tm
= localtime(&now
);
2622 sprintf(dd
->disk
.guid
, "%8s%04d%02d%02d",
2623 T10
, tm
->tm_year
+1900, tm
->tm_mon
+1, tm
->tm_mday
);
2624 tptr
= (__u32
*)(dd
->disk
.guid
+ 16);
2625 *tptr
++ = random32();
2629 /* Cannot be bothered finding a CRC of some irrelevant details*/
2630 dd
->disk
.refnum
= random32();
2631 for (i
= __be16_to_cpu(ddf
->active
->max_pd_entries
);
2633 if (ddf
->phys
->entries
[i
-1].refnum
== dd
->disk
.refnum
)
2637 dd
->disk
.forced_ref
= 1;
2638 dd
->disk
.forced_guid
= 1;
2639 memset(dd
->disk
.vendor
, ' ', 32);
2640 memcpy(dd
->disk
.vendor
, "Linux", 5);
2641 memset(dd
->disk
.pad
, 0xff, 442);
2642 for (i
= 0; i
< ddf
->max_part
; i
++)
2643 dd
->vlist
[i
] = NULL
;
2647 if (st
->update_tail
) {
2648 int len
= (sizeof(struct phys_disk
) +
2649 sizeof(struct phys_disk_entry
));
2650 struct phys_disk
*pd
;
2653 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2654 pd
->used_pdes
= __cpu_to_be16(n
);
2655 pde
= &pd
->entries
[0];
2658 ddf
->phys
->used_pdes
= __cpu_to_be16(
2659 1 + __be16_to_cpu(ddf
->phys
->used_pdes
));
2661 memcpy(pde
->guid
, dd
->disk
.guid
, DDF_GUID_LEN
);
2662 pde
->refnum
= dd
->disk
.refnum
;
2663 pde
->type
= __cpu_to_be16(DDF_Forced_PD_GUID
| DDF_Global_Spare
);
2664 pde
->state
= __cpu_to_be16(DDF_Online
);
2667 * If there is already a device in dlist, try to reserve the same
2668 * amount of workspace. Otherwise, use 32MB.
2669 * We checked disk size above already.
2671 #define __calc_lba(new, old, lba, mb) do { \
2672 unsigned long long dif; \
2673 if ((old) != NULL) \
2674 dif = (old)->size - __be64_to_cpu((old)->lba); \
2676 dif = (new)->size; \
2677 if ((new)->size > dif) \
2678 (new)->lba = __cpu_to_be64((new)->size - dif); \
2680 (new)->lba = __cpu_to_be64((new)->size - (mb*1024*2)); \
2682 __calc_lba(dd
, ddf
->dlist
, workspace_lba
, 32);
2683 __calc_lba(dd
, ddf
->dlist
, primary_lba
, 16);
2684 __calc_lba(dd
, ddf
->dlist
, secondary_lba
, 32);
2685 pde
->config_size
= dd
->workspace_lba
;
2687 sprintf(pde
->path
, "%17.17s","Information: nil") ;
2688 memset(pde
->pad
, 0xff, 6);
2690 if (st
->update_tail
) {
2691 dd
->next
= ddf
->add_list
;
2694 dd
->next
= ddf
->dlist
;
2696 ddf_set_updates_pending(ddf
);
2702 static int remove_from_super_ddf(struct supertype
*st
, mdu_disk_info_t
*dk
)
2704 struct ddf_super
*ddf
= st
->sb
;
2707 /* mdmon has noticed that this disk (dk->major/dk->minor) has
2708 * disappeared from the container.
2709 * We need to arrange that it disappears from the metadata and
2710 * internal data structures too.
2711 * Most of the work is done by ddf_process_update which edits
2712 * the metadata and closes the file handle and attaches the memory
2713 * where free_updates will free it.
2715 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2716 if (dl
->major
== dk
->major
&&
2717 dl
->minor
== dk
->minor
)
2722 if (st
->update_tail
) {
2723 int len
= (sizeof(struct phys_disk
) +
2724 sizeof(struct phys_disk_entry
));
2725 struct phys_disk
*pd
;
2728 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2729 pd
->used_pdes
= __cpu_to_be16(dl
->pdnum
);
2730 pd
->entries
[0].state
= __cpu_to_be16(DDF_Missing
);
2731 append_metadata_update(st
, pd
, len
);
2737 * This is the write_init_super method for a ddf container. It is
2738 * called when creating a container or adding another device to a
2741 #define NULL_CONF_SZ 4096
2743 static int __write_ddf_structure(struct dl
*d
, struct ddf_super
*ddf
, __u8 type
,
2746 unsigned long long sector
;
2747 struct ddf_header
*header
;
2748 int fd
, i
, n_config
, conf_size
;
2754 case DDF_HEADER_PRIMARY
:
2755 header
= &ddf
->primary
;
2756 sector
= __be64_to_cpu(header
->primary_lba
);
2758 case DDF_HEADER_SECONDARY
:
2759 header
= &ddf
->secondary
;
2760 sector
= __be64_to_cpu(header
->secondary_lba
);
2766 header
->type
= type
;
2767 header
->openflag
= 1;
2768 header
->crc
= calc_crc(header
, 512);
2770 lseek64(fd
, sector
<<9, 0);
2771 if (write(fd
, header
, 512) < 0)
2774 ddf
->controller
.crc
= calc_crc(&ddf
->controller
, 512);
2775 if (write(fd
, &ddf
->controller
, 512) < 0)
2778 ddf
->phys
->crc
= calc_crc(ddf
->phys
, ddf
->pdsize
);
2779 if (write(fd
, ddf
->phys
, ddf
->pdsize
) < 0)
2781 ddf
->virt
->crc
= calc_crc(ddf
->virt
, ddf
->vdsize
);
2782 if (write(fd
, ddf
->virt
, ddf
->vdsize
) < 0)
2785 /* Now write lots of config records. */
2786 n_config
= ddf
->max_part
;
2787 conf_size
= ddf
->conf_rec_len
* 512;
2788 for (i
= 0 ; i
<= n_config
; i
++) {
2790 struct vd_config
*vdc
= NULL
;
2791 if (i
== n_config
) {
2792 c
= (struct vcl
*)d
->spare
;
2799 get_pd_index_from_refnum(
2802 (const struct vd_config
**)&vdc
,
2806 dprintf("writing conf record %i on disk %08x for %s/%u\n",
2807 i
, d
->disk
.refnum
, guid_str(vdc
->guid
),
2808 vdc
->sec_elmnt_seq
);
2809 vdc
->seqnum
= header
->seq
;
2810 vdc
->crc
= calc_crc(vdc
, conf_size
);
2811 if (write(fd
, vdc
, conf_size
) < 0)
2814 unsigned int togo
= conf_size
;
2815 while (togo
> NULL_CONF_SZ
) {
2816 if (write(fd
, null_aligned
, NULL_CONF_SZ
) < 0)
2818 togo
-= NULL_CONF_SZ
;
2820 if (write(fd
, null_aligned
, togo
) < 0)
2827 d
->disk
.crc
= calc_crc(&d
->disk
, 512);
2828 if (write(fd
, &d
->disk
, 512) < 0)
2833 header
->openflag
= 0;
2834 header
->crc
= calc_crc(header
, 512);
2836 lseek64(fd
, sector
<<9, 0);
2837 if (write(fd
, header
, 512) < 0)
2843 static int __write_init_super_ddf(struct supertype
*st
)
2845 struct ddf_super
*ddf
= st
->sb
;
2849 unsigned long long size
;
2853 pr_state(ddf
, __func__
);
2854 if (posix_memalign((void**)&null_aligned
, 4096, NULL_CONF_SZ
) != 0) {
2857 memset(null_aligned
, 0xff, NULL_CONF_SZ
);
2859 seq
= ddf
->active
->seq
+ 1;
2861 /* try to write updated metadata,
2862 * if we catch a failure move on to the next disk
2864 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
2871 /* We need to fill in the primary, (secondary) and workspace
2872 * lba's in the headers, set their checksums,
2873 * Also checksum phys, virt....
2875 * Then write everything out, finally the anchor is written.
2877 get_dev_size(fd
, NULL
, &size
);
2879 if (d
->workspace_lba
!= 0)
2880 ddf
->anchor
.workspace_lba
= d
->workspace_lba
;
2882 ddf
->anchor
.workspace_lba
=
2883 __cpu_to_be64(size
- 32*1024*2);
2884 if (d
->primary_lba
!= 0)
2885 ddf
->anchor
.primary_lba
= d
->primary_lba
;
2887 ddf
->anchor
.primary_lba
=
2888 __cpu_to_be64(size
- 16*1024*2);
2889 if (d
->secondary_lba
!= 0)
2890 ddf
->anchor
.secondary_lba
= d
->secondary_lba
;
2892 ddf
->anchor
.secondary_lba
=
2893 __cpu_to_be64(size
- 32*1024*2);
2894 ddf
->anchor
.seq
= seq
;
2895 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2896 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2898 ddf
->anchor
.openflag
= 0xFF; /* 'open' means nothing */
2899 ddf
->anchor
.seq
= 0xFFFFFFFF; /* no sequencing in anchor */
2900 ddf
->anchor
.crc
= calc_crc(&ddf
->anchor
, 512);
2902 if (!__write_ddf_structure(d
, ddf
, DDF_HEADER_PRIMARY
,
2906 if (!__write_ddf_structure(d
, ddf
, DDF_HEADER_SECONDARY
,
2910 lseek64(fd
, (size
-1)*512, SEEK_SET
);
2911 if (write(fd
, &ddf
->anchor
, 512) < 0)
2917 return attempts
!= successes
;
2920 static int write_init_super_ddf(struct supertype
*st
)
2922 struct ddf_super
*ddf
= st
->sb
;
2923 struct vcl
*currentconf
= ddf
->currentconf
;
2925 /* we are done with currentconf reset it to point st at the container */
2926 ddf
->currentconf
= NULL
;
2928 if (st
->update_tail
) {
2929 /* queue the virtual_disk and vd_config as metadata updates */
2930 struct virtual_disk
*vd
;
2931 struct vd_config
*vc
;
2935 int len
= (sizeof(struct phys_disk
) +
2936 sizeof(struct phys_disk_entry
));
2938 /* adding a disk to the container. */
2942 append_metadata_update(st
, ddf
->add_list
->mdupdate
, len
);
2943 ddf
->add_list
->mdupdate
= NULL
;
2947 /* Newly created VD */
2949 /* First the virtual disk. We have a slightly fake header */
2950 len
= sizeof(struct virtual_disk
) + sizeof(struct virtual_entry
);
2953 vd
->entries
[0] = ddf
->virt
->entries
[currentconf
->vcnum
];
2954 vd
->populated_vdes
= __cpu_to_be16(currentconf
->vcnum
);
2955 append_metadata_update(st
, vd
, len
);
2957 /* Then the vd_config */
2958 len
= ddf
->conf_rec_len
* 512;
2960 memcpy(vc
, ¤tconf
->conf
, len
);
2961 append_metadata_update(st
, vc
, len
);
2963 /* FIXME I need to close the fds! */
2967 for (d
= ddf
->dlist
; d
; d
=d
->next
)
2968 while (Kill(d
->devname
, NULL
, 0, -1, 1) == 0);
2969 return __write_init_super_ddf(st
);
2975 static __u64
avail_size_ddf(struct supertype
*st
, __u64 devsize
,
2976 unsigned long long data_offset
)
2978 /* We must reserve the last 32Meg */
2979 if (devsize
<= 32*1024*2)
2981 return devsize
- 32*1024*2;
2986 static int reserve_space(struct supertype
*st
, int raiddisks
,
2987 unsigned long long size
, int chunk
,
2988 unsigned long long *freesize
)
2990 /* Find 'raiddisks' spare extents at least 'size' big (but
2991 * only caring about multiples of 'chunk') and remember
2993 * If the cannot be found, fail.
2996 struct ddf_super
*ddf
= st
->sb
;
2999 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3003 /* Now find largest extent on each device */
3004 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3005 struct extent
*e
= get_extents(ddf
, dl
);
3006 unsigned long long pos
= 0;
3009 unsigned long long minsize
= size
;
3017 unsigned long long esize
;
3018 esize
= e
[i
].start
- pos
;
3019 if (esize
>= minsize
) {
3023 pos
= e
[i
].start
+ e
[i
].size
;
3025 } while (e
[i
-1].size
);
3028 dl
->esize
= minsize
;
3032 if (cnt
< raiddisks
) {
3033 pr_err("not enough devices with space to create array.\n");
3034 return 0; /* No enough free spaces large enough */
3037 /* choose the largest size of which there are at least 'raiddisk' */
3038 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3040 if (dl
->esize
<= size
)
3042 /* This is bigger than 'size', see if there are enough */
3044 for (dl2
= ddf
->dlist
; dl2
; dl2
=dl2
->next
)
3045 if (dl2
->esize
>= dl
->esize
)
3047 if (cnt
>= raiddisks
)
3051 size
= size
/ chunk
;
3056 pr_err("not enough spare devices to create array.\n");
3060 /* We have a 'size' of which there are enough spaces.
3061 * We simply do a first-fit */
3063 for (dl
= ddf
->dlist
; dl
&& cnt
< raiddisks
; dl
=dl
->next
) {
3064 if (dl
->esize
< size
)
3074 validate_geometry_ddf_container(struct supertype
*st
,
3075 int level
, int layout
, int raiddisks
,
3076 int chunk
, unsigned long long size
,
3077 unsigned long long data_offset
,
3078 char *dev
, unsigned long long *freesize
,
3081 static int validate_geometry_ddf_bvd(struct supertype
*st
,
3082 int level
, int layout
, int raiddisks
,
3083 int *chunk
, unsigned long long size
,
3084 unsigned long long data_offset
,
3085 char *dev
, unsigned long long *freesize
,
3088 static int validate_geometry_ddf(struct supertype
*st
,
3089 int level
, int layout
, int raiddisks
,
3090 int *chunk
, unsigned long long size
,
3091 unsigned long long data_offset
,
3092 char *dev
, unsigned long long *freesize
,
3099 /* ddf potentially supports lots of things, but it depends on
3100 * what devices are offered (and maybe kernel version?)
3101 * If given unused devices, we will make a container.
3102 * If given devices in a container, we will make a BVD.
3103 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
3106 if (chunk
&& *chunk
== UnSet
)
3107 *chunk
= DEFAULT_CHUNK
;
3109 if (level
== -1000000) level
= LEVEL_CONTAINER
;
3110 if (level
== LEVEL_CONTAINER
) {
3111 /* Must be a fresh device to add to a container */
3112 return validate_geometry_ddf_container(st
, level
, layout
,
3113 raiddisks
, chunk
?*chunk
:0,
3114 size
, data_offset
, dev
,
3120 mdu_array_info_t array
= {
3121 .level
= level
, .layout
= layout
,
3122 .raid_disks
= raiddisks
3124 struct vd_config conf
;
3125 if (layout_md2ddf(&array
, &conf
) == -1) {
3127 pr_err("DDF does not support level %d /layout %d arrays with %d disks\n",
3128 level
, layout
, raiddisks
);
3131 /* Should check layout? etc */
3133 if (st
->sb
&& freesize
) {
3134 /* --create was given a container to create in.
3135 * So we need to check that there are enough
3136 * free spaces and return the amount of space.
3137 * We may as well remember which drives were
3138 * chosen so that add_to_super/getinfo_super
3141 return reserve_space(st
, raiddisks
, size
, chunk
?*chunk
:0, freesize
);
3147 /* A container has already been opened, so we are
3148 * creating in there. Maybe a BVD, maybe an SVD.
3149 * Should make a distinction one day.
3151 return validate_geometry_ddf_bvd(st
, level
, layout
, raiddisks
,
3152 chunk
, size
, data_offset
, dev
,
3156 /* This is the first device for the array.
3157 * If it is a container, we read it in and do automagic allocations,
3158 * no other devices should be given.
3159 * Otherwise it must be a member device of a container, and we
3160 * do manual allocation.
3161 * Later we should check for a BVD and make an SVD.
3163 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3165 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
3167 if (sra
&& sra
->array
.major_version
== -1 &&
3168 strcmp(sra
->text_version
, "ddf") == 0) {
3171 /* find space for 'n' devices. */
3172 /* remember the devices */
3173 /* Somehow return the fact that we have enough */
3177 pr_err("ddf: Cannot create this array "
3178 "on device %s - a container is required.\n",
3182 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
3184 pr_err("ddf: Cannot open %s: %s\n",
3185 dev
, strerror(errno
));
3188 /* Well, it is in use by someone, maybe a 'ddf' container. */
3189 cfd
= open_container(fd
);
3193 pr_err("ddf: Cannot use %s: %s\n",
3194 dev
, strerror(EBUSY
));
3197 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
3199 if (sra
&& sra
->array
.major_version
== -1 &&
3200 strcmp(sra
->text_version
, "ddf") == 0) {
3201 /* This is a member of a ddf container. Load the container
3202 * and try to create a bvd
3204 struct ddf_super
*ddf
;
3205 if (load_super_ddf_all(st
, cfd
, (void **)&ddf
, NULL
) == 0) {
3207 strcpy(st
->container_devnm
, fd2devnm(cfd
));
3209 return validate_geometry_ddf_bvd(st
, level
, layout
,
3210 raiddisks
, chunk
, size
,
3216 } else /* device may belong to a different container */
3223 validate_geometry_ddf_container(struct supertype
*st
,
3224 int level
, int layout
, int raiddisks
,
3225 int chunk
, unsigned long long size
,
3226 unsigned long long data_offset
,
3227 char *dev
, unsigned long long *freesize
,
3231 unsigned long long ldsize
;
3233 if (level
!= LEVEL_CONTAINER
)
3238 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3241 pr_err("ddf: Cannot open %s: %s\n",
3242 dev
, strerror(errno
));
3245 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3251 *freesize
= avail_size_ddf(st
, ldsize
>> 9, INVALID_SECTORS
);
3258 static int validate_geometry_ddf_bvd(struct supertype
*st
,
3259 int level
, int layout
, int raiddisks
,
3260 int *chunk
, unsigned long long size
,
3261 unsigned long long data_offset
,
3262 char *dev
, unsigned long long *freesize
,
3266 struct ddf_super
*ddf
= st
->sb
;
3268 unsigned long long pos
= 0;
3269 unsigned long long maxsize
;
3272 /* ddf/bvd supports lots of things, but not containers */
3273 if (level
== LEVEL_CONTAINER
) {
3275 pr_err("DDF cannot create a container within an container\n");
3278 /* We must have the container info already read in. */
3283 /* General test: make sure there is space for
3284 * 'raiddisks' device extents of size 'size'.
3286 unsigned long long minsize
= size
;
3290 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
3296 e
= get_extents(ddf
, dl
);
3299 unsigned long long esize
;
3300 esize
= e
[i
].start
- pos
;
3301 if (esize
>= minsize
)
3303 pos
= e
[i
].start
+ e
[i
].size
;
3305 } while (e
[i
-1].size
);
3310 if (dcnt
< raiddisks
) {
3312 pr_err("ddf: Not enough devices with "
3313 "space for this array (%d < %d)\n",
3319 /* This device must be a member of the set */
3320 if (stat(dev
, &stb
) < 0)
3322 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3324 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
3325 if (dl
->major
== (int)major(stb
.st_rdev
) &&
3326 dl
->minor
== (int)minor(stb
.st_rdev
))
3331 pr_err("ddf: %s is not in the "
3336 e
= get_extents(ddf
, dl
);
3340 unsigned long long esize
;
3341 esize
= e
[i
].start
- pos
;
3342 if (esize
>= maxsize
)
3344 pos
= e
[i
].start
+ e
[i
].size
;
3346 } while (e
[i
-1].size
);
3347 *freesize
= maxsize
;
3353 static int load_super_ddf_all(struct supertype
*st
, int fd
,
3354 void **sbp
, char *devname
)
3357 struct ddf_super
*super
;
3358 struct mdinfo
*sd
, *best
= NULL
;
3364 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3367 if (sra
->array
.major_version
!= -1 ||
3368 sra
->array
.minor_version
!= -2 ||
3369 strcmp(sra
->text_version
, "ddf") != 0)
3372 if (posix_memalign((void**)&super
, 512, sizeof(*super
)) != 0)
3374 memset(super
, 0, sizeof(*super
));
3376 /* first, try each device, and choose the best ddf */
3377 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3379 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3380 dfd
= dev_open(nm
, O_RDONLY
);
3383 rv
= load_ddf_headers(dfd
, super
, NULL
);
3386 seq
= __be32_to_cpu(super
->active
->seq
);
3387 if (super
->active
->openflag
)
3389 if (!best
|| seq
> bestseq
) {
3397 /* OK, load this ddf */
3398 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
3399 dfd
= dev_open(nm
, O_RDONLY
);
3402 load_ddf_headers(dfd
, super
, NULL
);
3403 load_ddf_global(dfd
, super
, NULL
);
3405 /* Now we need the device-local bits */
3406 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3409 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3410 dfd
= dev_open(nm
, O_RDWR
);
3413 rv
= load_ddf_headers(dfd
, super
, NULL
);
3415 rv
= load_ddf_local(dfd
, super
, NULL
, 1);
3421 if (st
->ss
== NULL
) {
3422 st
->ss
= &super_ddf
;
3423 st
->minor_version
= 0;
3426 strcpy(st
->container_devnm
, fd2devnm(fd
));
3430 static int load_container_ddf(struct supertype
*st
, int fd
,
3433 return load_super_ddf_all(st
, fd
, &st
->sb
, devname
);
3436 #endif /* MDASSEMBLE */
3438 static int check_secondary(const struct vcl
*vc
)
3440 const struct vd_config
*conf
= &vc
->conf
;
3443 /* The only DDF secondary RAID level md can support is
3444 * RAID 10, if the stripe sizes and Basic volume sizes
3446 * Other configurations could in theory be supported by exposing
3447 * the BVDs to user space and using device mapper for the secondary
3448 * mapping. So far we don't support that.
3451 __u64 sec_elements
[4] = {0, 0, 0, 0};
3452 #define __set_sec_seen(n) (sec_elements[(n)>>6] |= (1<<((n)&63)))
3453 #define __was_sec_seen(n) ((sec_elements[(n)>>6] & (1<<((n)&63))) != 0)
3455 if (vc
->other_bvds
== NULL
) {
3456 pr_err("No BVDs for secondary RAID found\n");
3459 if (conf
->prl
!= DDF_RAID1
) {
3460 pr_err("Secondary RAID level only supported for mirrored BVD\n");
3463 if (conf
->srl
!= DDF_2STRIPED
&& conf
->srl
!= DDF_2SPANNED
) {
3464 pr_err("Secondary RAID level %d is unsupported\n",
3468 __set_sec_seen(conf
->sec_elmnt_seq
);
3469 for (i
= 0; i
< conf
->sec_elmnt_count
-1; i
++) {
3470 const struct vd_config
*bvd
= vc
->other_bvds
[i
];
3471 if (bvd
->sec_elmnt_seq
== DDF_UNUSED_BVD
)
3473 if (bvd
->srl
!= conf
->srl
) {
3474 pr_err("Inconsistent secondary RAID level across BVDs\n");
3477 if (bvd
->prl
!= conf
->prl
) {
3478 pr_err("Different RAID levels for BVDs are unsupported\n");
3481 if (bvd
->prim_elmnt_count
!= conf
->prim_elmnt_count
) {
3482 pr_err("All BVDs must have the same number of primary elements\n");
3485 if (bvd
->chunk_shift
!= conf
->chunk_shift
) {
3486 pr_err("Different strip sizes for BVDs are unsupported\n");
3489 if (bvd
->array_blocks
!= conf
->array_blocks
) {
3490 pr_err("Different BVD sizes are unsupported\n");
3493 __set_sec_seen(bvd
->sec_elmnt_seq
);
3495 for (i
= 0; i
< conf
->sec_elmnt_count
; i
++) {
3496 if (!__was_sec_seen(i
)) {
3497 pr_err("BVD %d is missing\n", i
);
3504 static unsigned int get_pd_index_from_refnum(const struct vcl
*vc
,
3505 __u32 refnum
, unsigned int nmax
,
3506 const struct vd_config
**bvd
,
3509 unsigned int i
, j
, n
, sec
, cnt
;
3511 cnt
= __be16_to_cpu(vc
->conf
.prim_elmnt_count
);
3512 sec
= (vc
->conf
.sec_elmnt_count
== 1 ? 0 : vc
->conf
.sec_elmnt_seq
);
3514 for (i
= 0, j
= 0 ; i
< nmax
; i
++) {
3515 /* j counts valid entries for this BVD */
3516 if (vc
->conf
.phys_refnum
[i
] != 0xffffffff)
3518 if (vc
->conf
.phys_refnum
[i
] == refnum
) {
3521 return sec
* cnt
+ j
- 1;
3524 if (vc
->other_bvds
== NULL
)
3527 for (n
= 1; n
< vc
->conf
.sec_elmnt_count
; n
++) {
3528 struct vd_config
*vd
= vc
->other_bvds
[n
-1];
3529 sec
= vd
->sec_elmnt_seq
;
3530 if (sec
== DDF_UNUSED_BVD
)
3532 for (i
= 0, j
= 0 ; i
< nmax
; i
++) {
3533 if (vd
->phys_refnum
[i
] != 0xffffffff)
3535 if (vd
->phys_refnum
[i
] == refnum
) {
3538 return sec
* cnt
+ j
- 1;
3544 return DDF_NOTFOUND
;
3547 static struct mdinfo
*container_content_ddf(struct supertype
*st
, char *subarray
)
3549 /* Given a container loaded by load_super_ddf_all,
3550 * extract information about all the arrays into
3553 * For each vcl in conflist: create an mdinfo, fill it in,
3554 * then look for matching devices (phys_refnum) in dlist
3555 * and create appropriate device mdinfo.
3557 struct ddf_super
*ddf
= st
->sb
;
3558 struct mdinfo
*rest
= NULL
;
3561 for (vc
= ddf
->conflist
; vc
; vc
=vc
->next
)
3565 struct mdinfo
*this;
3571 (strtoul(subarray
, &ep
, 10) != vc
->vcnum
||
3575 if (vc
->conf
.sec_elmnt_count
> 1) {
3576 if (check_secondary(vc
) != 0)
3580 this = xcalloc(1, sizeof(*this));
3584 if (layout_ddf2md(&vc
->conf
, &this->array
))
3586 this->array
.md_minor
= -1;
3587 this->array
.major_version
= -1;
3588 this->array
.minor_version
= -2;
3589 cptr
= (__u32
*)(vc
->conf
.guid
+ 16);
3590 this->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
3591 this->array
.utime
= DECADE
+
3592 __be32_to_cpu(vc
->conf
.timestamp
);
3593 this->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
3596 if ((ddf
->virt
->entries
[i
].state
& DDF_state_inconsistent
) ||
3597 (ddf
->virt
->entries
[i
].init_state
& DDF_initstate_mask
) !=
3599 this->array
.state
= 0;
3600 this->resync_start
= 0;
3602 this->array
.state
= 1;
3603 this->resync_start
= MaxSector
;
3605 memcpy(this->name
, ddf
->virt
->entries
[i
].name
, 16);
3608 if (this->name
[j
] == ' ')
3611 memset(this->uuid
, 0, sizeof(this->uuid
));
3612 this->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
3613 this->array
.size
= this->component_size
/ 2;
3614 this->container_member
= i
;
3616 ddf
->currentconf
= vc
;
3617 uuid_from_super_ddf(st
, this->uuid
);
3618 ddf
->currentconf
= NULL
;
3620 sprintf(this->text_version
, "/%s/%d",
3621 st
->container_devnm
, this->container_member
);
3623 for (pd
= 0; pd
< __be16_to_cpu(ddf
->phys
->used_pdes
); pd
++) {
3626 const struct vd_config
*bvd
;
3630 if (ddf
->phys
->entries
[pd
].refnum
== 0xFFFFFFFF)
3633 stt
= __be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
3634 if ((stt
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
3638 i
= get_pd_index_from_refnum(
3639 vc
, ddf
->phys
->entries
[pd
].refnum
,
3640 ddf
->mppe
, &bvd
, &iphys
);
3641 if (i
== DDF_NOTFOUND
)
3644 this->array
.working_disks
++;
3646 for (d
= ddf
->dlist
; d
; d
=d
->next
)
3647 if (d
->disk
.refnum
==
3648 ddf
->phys
->entries
[pd
].refnum
)
3651 /* Haven't found that one yet, maybe there are others */
3654 dev
= xcalloc(1, sizeof(*dev
));
3655 dev
->next
= this->devs
;
3658 dev
->disk
.number
= __be32_to_cpu(d
->disk
.refnum
);
3659 dev
->disk
.major
= d
->major
;
3660 dev
->disk
.minor
= d
->minor
;
3661 dev
->disk
.raid_disk
= i
;
3662 dev
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
3663 dev
->recovery_start
= MaxSector
;
3665 dev
->events
= __be32_to_cpu(ddf
->primary
.seq
);
3667 __be64_to_cpu(LBA_OFFSET(ddf
, bvd
)[iphys
]);
3668 dev
->component_size
= __be64_to_cpu(bvd
->blocks
);
3670 strcpy(dev
->name
, d
->devname
);
3676 static int store_super_ddf(struct supertype
*st
, int fd
)
3678 struct ddf_super
*ddf
= st
->sb
;
3679 unsigned long long dsize
;
3686 if (!get_dev_size(fd
, NULL
, &dsize
))
3689 if (ddf
->dlist
|| ddf
->conflist
) {
3694 if (fstat(fd
, &sta
) == -1 || !S_ISBLK(sta
.st_mode
)) {
3695 pr_err("%s: file descriptor for invalid device\n",
3699 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
3700 if (dl
->major
== (int)major(sta
.st_rdev
) &&
3701 dl
->minor
== (int)minor(sta
.st_rdev
))
3704 pr_err("%s: couldn't find disk %d/%d\n", __func__
,
3705 (int)major(sta
.st_rdev
),
3706 (int)minor(sta
.st_rdev
));
3710 For DDF, writing to just one disk makes no sense.
3711 We would run the risk of writing inconsistent meta data
3712 to the devices. So just call __write_init_super_ddf and
3713 write to all devices, including this one.
3714 Use the fd passed to this function, just in case dl->fd
3719 ret
= __write_init_super_ddf(st
);
3724 if (posix_memalign(&buf
, 512, 512) != 0)
3726 memset(buf
, 0, 512);
3728 lseek64(fd
, dsize
-512, 0);
3729 rc
= write(fd
, buf
, 512);
3736 static int compare_super_ddf(struct supertype
*st
, struct supertype
*tst
)
3740 * 0 same, or first was empty, and second was copied
3741 * 1 second had wrong number
3743 * 3 wrong other info
3745 struct ddf_super
*first
= st
->sb
;
3746 struct ddf_super
*second
= tst
->sb
;
3747 struct dl
*dl1
, *dl2
;
3748 struct vcl
*vl1
, *vl2
;
3749 unsigned int max_vds
, max_pds
, pd
, vd
;
3757 if (memcmp(first
->anchor
.guid
, second
->anchor
.guid
, DDF_GUID_LEN
) != 0)
3760 if (first
->anchor
.seq
!= second
->anchor
.seq
) {
3761 dprintf("%s: sequence number mismatch %u/%u\n", __func__
,
3762 __be32_to_cpu(first
->anchor
.seq
),
3763 __be32_to_cpu(second
->anchor
.seq
));
3766 if (first
->max_part
!= second
->max_part
||
3767 first
->phys
->used_pdes
!= second
->phys
->used_pdes
||
3768 first
->virt
->populated_vdes
!= second
->virt
->populated_vdes
) {
3769 dprintf("%s: PD/VD number mismatch\n", __func__
);
3773 max_pds
= __be16_to_cpu(first
->phys
->used_pdes
);
3774 for (dl2
= second
->dlist
; dl2
; dl2
= dl2
->next
) {
3775 for (pd
= 0; pd
< max_pds
; pd
++)
3776 if (first
->phys
->entries
[pd
].refnum
== dl2
->disk
.refnum
)
3778 if (pd
== max_pds
) {
3779 dprintf("%s: no match for disk %08x\n", __func__
,
3780 __be32_to_cpu(dl2
->disk
.refnum
));
3785 max_vds
= __be16_to_cpu(first
->active
->max_vd_entries
);
3786 for (vl2
= second
->conflist
; vl2
; vl2
= vl2
->next
) {
3787 if (vl2
->conf
.magic
!= DDF_VD_CONF_MAGIC
)
3789 for (vd
= 0; vd
< max_vds
; vd
++)
3790 if (!memcmp(first
->virt
->entries
[vd
].guid
,
3791 vl2
->conf
.guid
, DDF_GUID_LEN
))
3793 if (vd
== max_vds
) {
3794 dprintf("%s: no match for VD config\n", __func__
);
3798 /* FIXME should I look at anything else? */
3801 At this point we are fairly sure that the meta data matches.
3802 But the new disk may contain additional local data.
3803 Add it to the super block.
3805 for (vl2
= second
->conflist
; vl2
; vl2
= vl2
->next
) {
3806 for (vl1
= first
->conflist
; vl1
; vl1
= vl1
->next
)
3807 if (!memcmp(vl1
->conf
.guid
, vl2
->conf
.guid
,
3811 if (vl1
->other_bvds
!= NULL
&&
3812 vl1
->conf
.sec_elmnt_seq
!=
3813 vl2
->conf
.sec_elmnt_seq
) {
3814 dprintf("%s: adding BVD %u\n", __func__
,
3815 vl2
->conf
.sec_elmnt_seq
);
3816 add_other_bvd(vl1
, &vl2
->conf
,
3817 first
->conf_rec_len
*512);
3822 if (posix_memalign((void **)&vl1
, 512,
3823 (first
->conf_rec_len
*512 +
3824 offsetof(struct vcl
, conf
))) != 0) {
3825 pr_err("%s could not allocate vcl buf\n",
3830 vl1
->next
= first
->conflist
;
3831 vl1
->block_sizes
= NULL
;
3832 memcpy(&vl1
->conf
, &vl2
->conf
, first
->conf_rec_len
*512);
3833 if (alloc_other_bvds(first
, vl1
) != 0) {
3834 pr_err("%s could not allocate other bvds\n",
3839 for (vd
= 0; vd
< max_vds
; vd
++)
3840 if (!memcmp(first
->virt
->entries
[vd
].guid
,
3841 vl1
->conf
.guid
, DDF_GUID_LEN
))
3844 dprintf("%s: added config for VD %u\n", __func__
, vl1
->vcnum
);
3845 first
->conflist
= vl1
;
3848 for (dl2
= second
->dlist
; dl2
; dl2
= dl2
->next
) {
3849 for (dl1
= first
->dlist
; dl1
; dl1
= dl1
->next
)
3850 if (dl1
->disk
.refnum
== dl2
->disk
.refnum
)
3855 if (posix_memalign((void **)&dl1
, 512,
3856 sizeof(*dl1
) + (first
->max_part
) * sizeof(dl1
->vlist
[0]))
3858 pr_err("%s could not allocate disk info buffer\n",
3862 memcpy(dl1
, dl2
, sizeof(*dl1
));
3863 dl1
->mdupdate
= NULL
;
3864 dl1
->next
= first
->dlist
;
3866 for (pd
= 0; pd
< max_pds
; pd
++)
3867 if (first
->phys
->entries
[pd
].refnum
== dl1
->disk
.refnum
)
3871 if (posix_memalign((void **)&dl1
->spare
, 512,
3872 first
->conf_rec_len
*512) != 0) {
3873 pr_err("%s could not allocate spare info buf\n",
3877 memcpy(dl1
->spare
, dl2
->spare
, first
->conf_rec_len
*512);
3879 for (vd
= 0 ; vd
< first
->max_part
; vd
++) {
3880 if (!dl2
->vlist
[vd
]) {
3881 dl1
->vlist
[vd
] = NULL
;
3884 for (vl1
= first
->conflist
; vl1
; vl1
= vl1
->next
) {
3885 if (!memcmp(vl1
->conf
.guid
,
3886 dl2
->vlist
[vd
]->conf
.guid
,
3889 dl1
->vlist
[vd
] = vl1
;
3893 dprintf("%s: added disk %d: %08x\n", __func__
, dl1
->pdnum
,
3902 * A new array 'a' has been started which claims to be instance 'inst'
3903 * within container 'c'.
3904 * We need to confirm that the array matches the metadata in 'c' so
3905 * that we don't corrupt any metadata.
3907 static int ddf_open_new(struct supertype
*c
, struct active_array
*a
, char *inst
)
3909 struct ddf_super
*ddf
= c
->sb
;
3911 if (all_ff(ddf
->virt
->entries
[n
].guid
)) {
3912 pr_err("%s: subarray %d doesn't exist\n", __func__
, n
);
3915 dprintf("ddf: open_new %d\n", n
);
3916 a
->info
.container_member
= n
;
3921 * The array 'a' is to be marked clean in the metadata.
3922 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
3923 * clean up to the point (in sectors). If that cannot be recorded in the
3924 * metadata, then leave it as dirty.
3926 * For DDF, we need to clear the DDF_state_inconsistent bit in the
3927 * !global! virtual_disk.virtual_entry structure.
3929 static int ddf_set_array_state(struct active_array
*a
, int consistent
)
3931 struct ddf_super
*ddf
= a
->container
->sb
;
3932 int inst
= a
->info
.container_member
;
3933 int old
= ddf
->virt
->entries
[inst
].state
;
3934 if (consistent
== 2) {
3935 /* Should check if a recovery should be started FIXME */
3937 if (!is_resync_complete(&a
->info
))
3941 ddf
->virt
->entries
[inst
].state
&= ~DDF_state_inconsistent
;
3943 ddf
->virt
->entries
[inst
].state
|= DDF_state_inconsistent
;
3944 if (old
!= ddf
->virt
->entries
[inst
].state
)
3945 ddf_set_updates_pending(ddf
);
3947 old
= ddf
->virt
->entries
[inst
].init_state
;
3948 ddf
->virt
->entries
[inst
].init_state
&= ~DDF_initstate_mask
;
3949 if (is_resync_complete(&a
->info
))
3950 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_full
;
3951 else if (a
->info
.resync_start
== 0)
3952 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_not
;
3954 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_quick
;
3955 if (old
!= ddf
->virt
->entries
[inst
].init_state
)
3956 ddf_set_updates_pending(ddf
);
3958 dprintf("ddf mark %d/%s (%d) %s %llu\n", inst
,
3959 guid_str(ddf
->virt
->entries
[inst
].guid
), a
->curr_state
,
3960 consistent
?"clean":"dirty",
3961 a
->info
.resync_start
);
3965 static int get_bvd_state(const struct ddf_super
*ddf
,
3966 const struct vd_config
*vc
)
3968 unsigned int i
, n_bvd
, working
= 0;
3969 unsigned int n_prim
= __be16_to_cpu(vc
->prim_elmnt_count
);
3971 for (i
= 0; i
< n_prim
; i
++) {
3972 if (!find_index_in_bvd(ddf
, vc
, i
, &n_bvd
))
3974 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
3977 st
= __be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
3978 if ((st
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
3983 state
= DDF_state_degraded
;
3984 if (working
== n_prim
)
3985 state
= DDF_state_optimal
;
3991 state
= DDF_state_failed
;
3995 state
= DDF_state_failed
;
3996 else if (working
>= 2)
3997 state
= DDF_state_part_optimal
;
4001 if (working
< n_prim
- 1)
4002 state
= DDF_state_failed
;
4005 if (working
< n_prim
- 2)
4006 state
= DDF_state_failed
;
4007 else if (working
== n_prim
- 1)
4008 state
= DDF_state_part_optimal
;
4014 static int secondary_state(int state
, int other
, int seclevel
)
4016 if (state
== DDF_state_optimal
&& other
== DDF_state_optimal
)
4017 return DDF_state_optimal
;
4018 if (seclevel
== DDF_2MIRRORED
) {
4019 if (state
== DDF_state_optimal
|| other
== DDF_state_optimal
)
4020 return DDF_state_part_optimal
;
4021 if (state
== DDF_state_failed
&& other
== DDF_state_failed
)
4022 return DDF_state_failed
;
4023 return DDF_state_degraded
;
4025 if (state
== DDF_state_failed
|| other
== DDF_state_failed
)
4026 return DDF_state_failed
;
4027 if (state
== DDF_state_degraded
|| other
== DDF_state_degraded
)
4028 return DDF_state_degraded
;
4029 return DDF_state_part_optimal
;
4033 static int get_svd_state(const struct ddf_super
*ddf
, const struct vcl
*vcl
)
4035 int state
= get_bvd_state(ddf
, &vcl
->conf
);
4037 for (i
= 1; i
< vcl
->conf
.sec_elmnt_count
; i
++) {
4038 state
= secondary_state(
4040 get_bvd_state(ddf
, vcl
->other_bvds
[i
-1]),
4047 * The state of each disk is stored in the global phys_disk structure
4048 * in phys_disk.entries[n].state.
4049 * This makes various combinations awkward.
4050 * - When a device fails in any array, it must be failed in all arrays
4051 * that include a part of this device.
4052 * - When a component is rebuilding, we cannot include it officially in the
4053 * array unless this is the only array that uses the device.
4055 * So: when transitioning:
4056 * Online -> failed, just set failed flag. monitor will propagate
4057 * spare -> online, the device might need to be added to the array.
4058 * spare -> failed, just set failed. Don't worry if in array or not.
4060 static void ddf_set_disk(struct active_array
*a
, int n
, int state
)
4062 struct ddf_super
*ddf
= a
->container
->sb
;
4063 unsigned int inst
= a
->info
.container_member
, n_bvd
;
4065 struct vd_config
*vc
= find_vdcr(ddf
, inst
, (unsigned int)n
,
4072 dprintf("ddf: cannot find instance %d!!\n", inst
);
4075 /* Find the matching slot in 'info'. */
4076 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
4077 if (mdi
->disk
.raid_disk
== n
)
4082 /* and find the 'dl' entry corresponding to that. */
4083 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4084 if (mdi
->state_fd
>= 0 &&
4085 mdi
->disk
.major
== dl
->major
&&
4086 mdi
->disk
.minor
== dl
->minor
)
4091 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
4092 if (pd
< 0 || pd
!= dl
->pdnum
) {
4093 /* disk doesn't currently exist or has changed.
4094 * If it is now in_sync, insert it. */
4095 dprintf("%s: phys disk not found for %d: %d/%d ref %08x\n",
4096 __func__
, dl
->pdnum
, dl
->major
, dl
->minor
,
4098 dprintf("%s: array %u disk %u ref %08x pd %d\n",
4099 __func__
, inst
, n_bvd
, vc
->phys_refnum
[n_bvd
], pd
);
4100 if ((state
& DS_INSYNC
) && ! (state
& DS_FAULTY
)) {
4101 pd
= dl
->pdnum
; /* FIXME: is this really correct ? */
4102 vc
->phys_refnum
[n_bvd
] = dl
->disk
.refnum
;
4103 LBA_OFFSET(ddf
, vc
)[n_bvd
] =
4104 __cpu_to_be64(mdi
->data_offset
);
4105 ddf
->phys
->entries
[pd
].type
&=
4106 ~__cpu_to_be16(DDF_Global_Spare
);
4107 ddf
->phys
->entries
[pd
].type
|=
4108 __cpu_to_be16(DDF_Active_in_VD
);
4109 ddf_set_updates_pending(ddf
);
4112 int old
= ddf
->phys
->entries
[pd
].state
;
4113 if (state
& DS_FAULTY
)
4114 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Failed
);
4115 if (state
& DS_INSYNC
) {
4116 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Online
);
4117 ddf
->phys
->entries
[pd
].state
&= __cpu_to_be16(~DDF_Rebuilding
);
4119 if (old
!= ddf
->phys
->entries
[pd
].state
)
4120 ddf_set_updates_pending(ddf
);
4123 dprintf("ddf: set_disk %d to %x\n", n
, state
);
4125 /* Now we need to check the state of the array and update
4126 * virtual_disk.entries[n].state.
4127 * It needs to be one of "optimal", "degraded", "failed".
4128 * I don't understand 'deleted' or 'missing'.
4130 state
= get_svd_state(ddf
, vcl
);
4132 if (ddf
->virt
->entries
[inst
].state
!=
4133 ((ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4136 ddf
->virt
->entries
[inst
].state
=
4137 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4139 ddf_set_updates_pending(ddf
);
4144 static void ddf_sync_metadata(struct supertype
*st
)
4148 * Write all data to all devices.
4149 * Later, we might be able to track whether only local changes
4150 * have been made, or whether any global data has been changed,
4151 * but ddf is sufficiently weird that it probably always
4152 * changes global data ....
4154 struct ddf_super
*ddf
= st
->sb
;
4155 if (!ddf
->updates_pending
)
4157 ddf
->updates_pending
= 0;
4158 __write_init_super_ddf(st
);
4159 dprintf("ddf: sync_metadata\n");
4162 static void ddf_process_update(struct supertype
*st
,
4163 struct metadata_update
*update
)
4165 /* Apply this update to the metadata.
4166 * The first 4 bytes are a DDF_*_MAGIC which guides
4168 * Possible update are:
4169 * DDF_PHYS_RECORDS_MAGIC
4170 * Add a new physical device or remove an old one.
4171 * Changes to this record only happen implicitly.
4172 * used_pdes is the device number.
4173 * DDF_VIRT_RECORDS_MAGIC
4174 * Add a new VD. Possibly also change the 'access' bits.
4175 * populated_vdes is the entry number.
4177 * New or updated VD. the VIRT_RECORD must already
4178 * exist. For an update, phys_refnum and lba_offset
4179 * (at least) are updated, and the VD_CONF must
4180 * be written to precisely those devices listed with
4182 * DDF_SPARE_ASSIGN_MAGIC
4183 * replacement Spare Assignment Record... but for which device?
4186 * - to create a new array, we send a VIRT_RECORD and
4187 * a VD_CONF. Then assemble and start the array.
4188 * - to activate a spare we send a VD_CONF to add the phys_refnum
4189 * and offset. This will also mark the spare as active with
4190 * a spare-assignment record.
4192 struct ddf_super
*ddf
= st
->sb
;
4193 __u32
*magic
= (__u32
*)update
->buf
;
4194 struct phys_disk
*pd
;
4195 struct virtual_disk
*vd
;
4196 struct vd_config
*vc
;
4201 unsigned int pdnum
, pd2
;
4203 dprintf("Process update %x\n", *magic
);
4206 case DDF_PHYS_RECORDS_MAGIC
:
4208 if (update
->len
!= (sizeof(struct phys_disk
) +
4209 sizeof(struct phys_disk_entry
)))
4211 pd
= (struct phys_disk
*)update
->buf
;
4213 ent
= __be16_to_cpu(pd
->used_pdes
);
4214 if (ent
>= __be16_to_cpu(ddf
->phys
->max_pdes
))
4216 if (pd
->entries
[0].state
& __cpu_to_be16(DDF_Missing
)) {
4218 /* removing this disk. */
4219 ddf
->phys
->entries
[ent
].state
|= __cpu_to_be16(DDF_Missing
);
4220 for (dlp
= &ddf
->dlist
; *dlp
; dlp
= &(*dlp
)->next
) {
4221 struct dl
*dl
= *dlp
;
4222 if (dl
->pdnum
== (signed)ent
) {
4225 /* FIXME this doesn't free
4232 ddf_set_updates_pending(ddf
);
4235 if (!all_ff(ddf
->phys
->entries
[ent
].guid
))
4237 ddf
->phys
->entries
[ent
] = pd
->entries
[0];
4238 ddf
->phys
->used_pdes
= __cpu_to_be16(1 +
4239 __be16_to_cpu(ddf
->phys
->used_pdes
));
4240 ddf_set_updates_pending(ddf
);
4241 if (ddf
->add_list
) {
4242 struct active_array
*a
;
4243 struct dl
*al
= ddf
->add_list
;
4244 ddf
->add_list
= al
->next
;
4246 al
->next
= ddf
->dlist
;
4249 /* As a device has been added, we should check
4250 * for any degraded devices that might make
4251 * use of this spare */
4252 for (a
= st
->arrays
; a
; a
=a
->next
)
4253 a
->check_degraded
= 1;
4257 case DDF_VIRT_RECORDS_MAGIC
:
4259 if (update
->len
!= (sizeof(struct virtual_disk
) +
4260 sizeof(struct virtual_entry
)))
4262 vd
= (struct virtual_disk
*)update
->buf
;
4264 ent
= find_unused_vde(ddf
);
4265 if (ent
== DDF_NOTFOUND
)
4267 ddf
->virt
->entries
[ent
] = vd
->entries
[0];
4268 ddf
->virt
->populated_vdes
= __cpu_to_be16(1 +
4269 __be16_to_cpu(ddf
->virt
->populated_vdes
));
4270 ddf_set_updates_pending(ddf
);
4273 case DDF_VD_CONF_MAGIC
:
4274 dprintf("len %d %d\n", update
->len
, ddf
->conf_rec_len
);
4276 mppe
= __be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
4277 if ((unsigned)update
->len
!= ddf
->conf_rec_len
* 512)
4279 vc
= (struct vd_config
*)update
->buf
;
4280 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
4281 if (memcmp(vcl
->conf
.guid
, vc
->guid
, DDF_GUID_LEN
) == 0)
4283 dprintf("vcl = %p\n", vcl
);
4285 /* An update, just copy the phys_refnum and lba_offset
4288 memcpy(vcl
->conf
.phys_refnum
, vc
->phys_refnum
,
4289 mppe
* (sizeof(__u32
) + sizeof(__u64
)));
4294 vcl
= update
->space
;
4295 update
->space
= NULL
;
4296 vcl
->next
= ddf
->conflist
;
4297 memcpy(&vcl
->conf
, vc
, update
->len
);
4298 ent
= find_vde_by_guid(ddf
, vc
->guid
);
4299 if (ent
== DDF_NOTFOUND
)
4302 ddf
->conflist
= vcl
;
4304 /* Set DDF_Transition on all Failed devices - to help
4305 * us detect those that are no longer in use
4307 for (pdnum
= 0; pdnum
< __be16_to_cpu(ddf
->phys
->used_pdes
); pdnum
++)
4308 if (ddf
->phys
->entries
[pdnum
].state
4309 & __be16_to_cpu(DDF_Failed
))
4310 ddf
->phys
->entries
[pdnum
].state
4311 |= __be16_to_cpu(DDF_Transition
);
4312 /* Now make sure vlist is correct for each dl. */
4313 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
4315 unsigned int vn
= 0;
4316 int in_degraded
= 0;
4317 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
4318 for (dn
=0; dn
< ddf
->mppe
; dn
++)
4319 if (vcl
->conf
.phys_refnum
[dn
] ==
4322 dprintf("dev %d has %p at %d\n",
4323 dl
->pdnum
, vcl
, vn
);
4324 /* Clear the Transition flag */
4325 if (ddf
->phys
->entries
[dl
->pdnum
].state
4326 & __be16_to_cpu(DDF_Failed
))
4327 ddf
->phys
->entries
[dl
->pdnum
].state
&=
4328 ~__be16_to_cpu(DDF_Transition
);
4330 dl
->vlist
[vn
++] = vcl
;
4331 vstate
= ddf
->virt
->entries
[vcl
->vcnum
].state
4333 if (vstate
== DDF_state_degraded
||
4334 vstate
== DDF_state_part_optimal
)
4338 while (vn
< ddf
->max_part
)
4339 dl
->vlist
[vn
++] = NULL
;
4341 ddf
->phys
->entries
[dl
->pdnum
].type
&=
4342 ~__cpu_to_be16(DDF_Global_Spare
);
4343 if (!(ddf
->phys
->entries
[dl
->pdnum
].type
&
4344 __cpu_to_be16(DDF_Active_in_VD
))) {
4345 ddf
->phys
->entries
[dl
->pdnum
].type
|=
4346 __cpu_to_be16(DDF_Active_in_VD
);
4348 ddf
->phys
->entries
[dl
->pdnum
].state
|=
4349 __cpu_to_be16(DDF_Rebuilding
);
4353 ddf
->phys
->entries
[dl
->pdnum
].type
&=
4354 ~__cpu_to_be16(DDF_Global_Spare
);
4355 ddf
->phys
->entries
[dl
->pdnum
].type
|=
4356 __cpu_to_be16(DDF_Spare
);
4358 if (!dl
->vlist
[0] && !dl
->spare
) {
4359 ddf
->phys
->entries
[dl
->pdnum
].type
|=
4360 __cpu_to_be16(DDF_Global_Spare
);
4361 ddf
->phys
->entries
[dl
->pdnum
].type
&=
4362 ~__cpu_to_be16(DDF_Spare
|
4367 /* Now remove any 'Failed' devices that are not part
4368 * of any VD. They will have the Transition flag set.
4369 * Once done, we need to update all dl->pdnum numbers.
4372 for (pdnum
= 0; pdnum
< __be16_to_cpu(ddf
->phys
->used_pdes
); pdnum
++)
4373 if ((ddf
->phys
->entries
[pdnum
].state
4374 & __be16_to_cpu(DDF_Failed
))
4375 && (ddf
->phys
->entries
[pdnum
].state
4376 & __be16_to_cpu(DDF_Transition
)))
4377 /* skip this one */;
4378 else if (pdnum
== pd2
)
4381 ddf
->phys
->entries
[pd2
] = ddf
->phys
->entries
[pdnum
];
4382 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4383 if (dl
->pdnum
== (int)pdnum
)
4387 ddf
->phys
->used_pdes
= __cpu_to_be16(pd2
);
4388 while (pd2
< pdnum
) {
4389 memset(ddf
->phys
->entries
[pd2
].guid
, 0xff, DDF_GUID_LEN
);
4393 ddf_set_updates_pending(ddf
);
4395 case DDF_SPARE_ASSIGN_MAGIC
:
4400 static void ddf_prepare_update(struct supertype
*st
,
4401 struct metadata_update
*update
)
4403 /* This update arrived at managemon.
4404 * We are about to pass it to monitor.
4405 * If a malloc is needed, do it here.
4407 struct ddf_super
*ddf
= st
->sb
;
4408 __u32
*magic
= (__u32
*)update
->buf
;
4409 if (*magic
== DDF_VD_CONF_MAGIC
)
4410 if (posix_memalign(&update
->space
, 512,
4411 offsetof(struct vcl
, conf
)
4412 + ddf
->conf_rec_len
* 512) != 0)
4413 update
->space
= NULL
;
4417 * Check if the array 'a' is degraded but not failed.
4418 * If it is, find as many spares as are available and needed and
4419 * arrange for their inclusion.
4420 * We only choose devices which are not already in the array,
4421 * and prefer those with a spare-assignment to this array.
4422 * otherwise we choose global spares - assuming always that
4423 * there is enough room.
4424 * For each spare that we assign, we return an 'mdinfo' which
4425 * describes the position for the device in the array.
4426 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
4427 * the new phys_refnum and lba_offset values.
4429 * Only worry about BVDs at the moment.
4431 static struct mdinfo
*ddf_activate_spare(struct active_array
*a
,
4432 struct metadata_update
**updates
)
4436 struct ddf_super
*ddf
= a
->container
->sb
;
4438 struct mdinfo
*rv
= NULL
;
4440 struct metadata_update
*mu
;
4444 struct vd_config
*vc
;
4447 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
4448 if ((d
->curr_state
& DS_FAULTY
) &&
4450 /* wait for Removal to happen */
4452 if (d
->state_fd
>= 0)
4456 dprintf("ddf_activate: working=%d (%d) level=%d\n", working
, a
->info
.array
.raid_disks
,
4457 a
->info
.array
.level
);
4458 if (working
== a
->info
.array
.raid_disks
)
4459 return NULL
; /* array not degraded */
4460 switch (a
->info
.array
.level
) {
4463 return NULL
; /* failed */
4467 if (working
< a
->info
.array
.raid_disks
- 1)
4468 return NULL
; /* failed */
4471 if (working
< a
->info
.array
.raid_disks
- 2)
4472 return NULL
; /* failed */
4474 default: /* concat or stripe */
4475 return NULL
; /* failed */
4478 /* For each slot, if it is not working, find a spare */
4480 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
4481 for (d
= a
->info
.devs
; d
; d
= d
->next
)
4482 if (d
->disk
.raid_disk
== i
)
4484 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
4485 if (d
&& (d
->state_fd
>= 0))
4488 /* OK, this device needs recovery. Find a spare */
4490 for ( ; dl
; dl
= dl
->next
) {
4491 unsigned long long esize
;
4492 unsigned long long pos
;
4495 int is_dedicated
= 0;
4498 /* If in this array, skip */
4499 for (d2
= a
->info
.devs
; d2
; d2
= d2
->next
)
4500 if (d2
->state_fd
>= 0 &&
4501 d2
->disk
.major
== dl
->major
&&
4502 d2
->disk
.minor
== dl
->minor
) {
4503 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
4508 if (ddf
->phys
->entries
[dl
->pdnum
].type
&
4509 __cpu_to_be16(DDF_Spare
)) {
4510 /* Check spare assign record */
4512 if (dl
->spare
->type
& DDF_spare_dedicated
) {
4513 /* check spare_ents for guid */
4515 j
< __be16_to_cpu(dl
->spare
->populated
);
4517 if (memcmp(dl
->spare
->spare_ents
[j
].guid
,
4518 ddf
->virt
->entries
[a
->info
.container_member
].guid
,
4525 } else if (ddf
->phys
->entries
[dl
->pdnum
].type
&
4526 __cpu_to_be16(DDF_Global_Spare
)) {
4528 } else if (!(ddf
->phys
->entries
[dl
->pdnum
].state
&
4529 __cpu_to_be16(DDF_Failed
))) {
4530 /* we can possibly use some of this */
4533 if ( ! (is_dedicated
||
4534 (is_global
&& global_ok
))) {
4535 dprintf("%x:%x not suitable: %d %d\n", dl
->major
, dl
->minor
,
4536 is_dedicated
, is_global
);
4540 /* We are allowed to use this device - is there space?
4541 * We need a->info.component_size sectors */
4542 ex
= get_extents(ddf
, dl
);
4544 dprintf("cannot get extents\n");
4551 esize
= ex
[j
].start
- pos
;
4552 if (esize
>= a
->info
.component_size
)
4554 pos
= ex
[j
].start
+ ex
[j
].size
;
4556 } while (ex
[j
-1].size
);
4559 if (esize
< a
->info
.component_size
) {
4560 dprintf("%x:%x has no room: %llu %llu\n",
4561 dl
->major
, dl
->minor
,
4562 esize
, a
->info
.component_size
);
4567 /* Cool, we have a device with some space at pos */
4568 di
= xcalloc(1, sizeof(*di
));
4569 di
->disk
.number
= i
;
4570 di
->disk
.raid_disk
= i
;
4571 di
->disk
.major
= dl
->major
;
4572 di
->disk
.minor
= dl
->minor
;
4574 di
->recovery_start
= 0;
4575 di
->data_offset
= pos
;
4576 di
->component_size
= a
->info
.component_size
;
4577 di
->container_member
= dl
->pdnum
;
4580 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
4585 if (!dl
&& ! global_ok
) {
4586 /* not enough dedicated spares, try global */
4594 /* No spares found */
4596 /* Now 'rv' has a list of devices to return.
4597 * Create a metadata_update record to update the
4598 * phys_refnum and lba_offset values
4600 mu
= xmalloc(sizeof(*mu
));
4601 if (posix_memalign(&mu
->space
, 512, sizeof(struct vcl
)) != 0) {
4605 mu
->buf
= xmalloc(ddf
->conf_rec_len
* 512);
4606 mu
->len
= ddf
->conf_rec_len
* 512;
4608 mu
->space_list
= NULL
;
4609 mu
->next
= *updates
;
4610 vc
= find_vdcr(ddf
, a
->info
.container_member
, di
->disk
.raid_disk
,
4612 memcpy(mu
->buf
, vc
, ddf
->conf_rec_len
* 512);
4614 vc
= (struct vd_config
*)mu
->buf
;
4615 for (di
= rv
; di
; di
= di
->next
) {
4616 vc
->phys_refnum
[di
->disk
.raid_disk
] =
4617 ddf
->phys
->entries
[dl
->pdnum
].refnum
;
4618 LBA_OFFSET(ddf
, vc
)[di
->disk
.raid_disk
]
4619 = __cpu_to_be64(di
->data_offset
);
4624 #endif /* MDASSEMBLE */
4626 static int ddf_level_to_layout(int level
)
4633 return ALGORITHM_LEFT_SYMMETRIC
;
4635 return ALGORITHM_ROTATING_N_CONTINUE
;
4643 static void default_geometry_ddf(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
4645 if (level
&& *level
== UnSet
)
4646 *level
= LEVEL_CONTAINER
;
4648 if (level
&& layout
&& *layout
== UnSet
)
4649 *layout
= ddf_level_to_layout(*level
);
4652 struct superswitch super_ddf
= {
4654 .examine_super
= examine_super_ddf
,
4655 .brief_examine_super
= brief_examine_super_ddf
,
4656 .brief_examine_subarrays
= brief_examine_subarrays_ddf
,
4657 .export_examine_super
= export_examine_super_ddf
,
4658 .detail_super
= detail_super_ddf
,
4659 .brief_detail_super
= brief_detail_super_ddf
,
4660 .validate_geometry
= validate_geometry_ddf
,
4661 .write_init_super
= write_init_super_ddf
,
4662 .add_to_super
= add_to_super_ddf
,
4663 .remove_from_super
= remove_from_super_ddf
,
4664 .load_container
= load_container_ddf
,
4665 .copy_metadata
= copy_metadata_ddf
,
4667 .match_home
= match_home_ddf
,
4668 .uuid_from_super
= uuid_from_super_ddf
,
4669 .getinfo_super
= getinfo_super_ddf
,
4670 .update_super
= update_super_ddf
,
4672 .avail_size
= avail_size_ddf
,
4674 .compare_super
= compare_super_ddf
,
4676 .load_super
= load_super_ddf
,
4677 .init_super
= init_super_ddf
,
4678 .store_super
= store_super_ddf
,
4679 .free_super
= free_super_ddf
,
4680 .match_metadata_desc
= match_metadata_desc_ddf
,
4681 .container_content
= container_content_ddf
,
4682 .default_geometry
= default_geometry_ddf
,
4688 .open_new
= ddf_open_new
,
4689 .set_array_state
= ddf_set_array_state
,
4690 .set_disk
= ddf_set_disk
,
4691 .sync_metadata
= ddf_sync_metadata
,
4692 .process_update
= ddf_process_update
,
4693 .prepare_update
= ddf_prepare_update
,
4694 .activate_spare
= ddf_activate_spare
,