2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2006-2014 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 taken from Common RAID DDF Specification Revision 1.2
25 * (July 28 2006). Reused by permission of SNIA.
28 #define HAVE_STDINT_H 1
35 /* a non-official T10 name for creation GUIDs */
36 static char T10
[] = "Linux-MD";
38 /* DDF timestamps are 1980 based, so we need to add
39 * second-in-decade-of-seventies to convert to linux timestamps.
40 * 10 years with 2 leap years.
42 #define DECADE (3600*24*(365*10+2))
45 const unsigned char *buf
,
48 #define DDF_NOTFOUND (~0U)
49 #define DDF_CONTAINER (DDF_NOTFOUND-1)
51 /* Default for safe_mode_delay. Same value as for IMSM.
53 static const int DDF_SAFE_MODE_DELAY
= 4000;
55 /* The DDF metadata handling.
56 * DDF metadata lives at the end of the device.
57 * The last 512 byte block provides an 'anchor' which is used to locate
58 * the rest of the metadata which usually lives immediately behind the anchor.
61 * - all multibyte numeric fields are bigendian.
62 * - all strings are space padded.
66 typedef struct __be16
{
69 #define be16_eq(x, y) ((x)._v16 == (y)._v16)
70 #define be16_and(x, y) ((x)._v16 & (y)._v16)
71 #define be16_or(x, y) ((x)._v16 | (y)._v16)
72 #define be16_clear(x, y) ((x)._v16 &= ~(y)._v16)
73 #define be16_set(x, y) ((x)._v16 |= (y)._v16)
75 typedef struct __be32
{
78 #define be32_eq(x, y) ((x)._v32 == (y)._v32)
80 typedef struct __be64
{
83 #define be64_eq(x, y) ((x)._v64 == (y)._v64)
85 #define be16_to_cpu(be) __be16_to_cpu((be)._v16)
86 static inline be16
cpu_to_be16(__u16 x
)
88 be16 be
= { ._v16
= __cpu_to_be16(x
) };
92 #define be32_to_cpu(be) __be32_to_cpu((be)._v32)
93 static inline be32
cpu_to_be32(__u32 x
)
95 be32 be
= { ._v32
= __cpu_to_be32(x
) };
99 #define be64_to_cpu(be) __be64_to_cpu((be)._v64)
100 static inline be64
cpu_to_be64(__u64 x
)
102 be64 be
= { ._v64
= __cpu_to_be64(x
) };
106 /* Primary Raid Level (PRL) */
107 #define DDF_RAID0 0x00
108 #define DDF_RAID1 0x01
109 #define DDF_RAID3 0x03
110 #define DDF_RAID4 0x04
111 #define DDF_RAID5 0x05
112 #define DDF_RAID1E 0x11
113 #define DDF_JBOD 0x0f
114 #define DDF_CONCAT 0x1f
115 #define DDF_RAID5E 0x15
116 #define DDF_RAID5EE 0x25
117 #define DDF_RAID6 0x06
119 /* Raid Level Qualifier (RLQ) */
120 #define DDF_RAID0_SIMPLE 0x00
121 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
122 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
123 #define DDF_RAID3_0 0x00 /* parity in first extent */
124 #define DDF_RAID3_N 0x01 /* parity in last extent */
125 #define DDF_RAID4_0 0x00 /* parity in first extent */
126 #define DDF_RAID4_N 0x01 /* parity in last extent */
127 /* these apply to raid5e and raid5ee as well */
128 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
129 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
130 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
131 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
133 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
134 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
136 /* Secondary RAID Level (SRL) */
137 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
138 #define DDF_2MIRRORED 0x01
139 #define DDF_2CONCAT 0x02
140 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
143 #define DDF_HEADER_MAGIC cpu_to_be32(0xDE11DE11)
144 #define DDF_CONTROLLER_MAGIC cpu_to_be32(0xAD111111)
145 #define DDF_PHYS_RECORDS_MAGIC cpu_to_be32(0x22222222)
146 #define DDF_PHYS_DATA_MAGIC cpu_to_be32(0x33333333)
147 #define DDF_VIRT_RECORDS_MAGIC cpu_to_be32(0xDDDDDDDD)
148 #define DDF_VD_CONF_MAGIC cpu_to_be32(0xEEEEEEEE)
149 #define DDF_SPARE_ASSIGN_MAGIC cpu_to_be32(0x55555555)
150 #define DDF_VU_CONF_MAGIC cpu_to_be32(0x88888888)
151 #define DDF_VENDOR_LOG_MAGIC cpu_to_be32(0x01dBEEF0)
152 #define DDF_BBM_LOG_MAGIC cpu_to_be32(0xABADB10C)
154 #define DDF_GUID_LEN 24
155 #define DDF_REVISION_0 "01.00.00"
156 #define DDF_REVISION_2 "01.02.00"
159 be32 magic
; /* DDF_HEADER_MAGIC */
161 char guid
[DDF_GUID_LEN
];
162 char revision
[8]; /* 01.02.00 */
163 be32 seq
; /* starts at '1' */
168 __u8 pad0
; /* 0xff */
169 __u8 pad1
[12]; /* 12 * 0xff */
170 /* 64 bytes so far */
171 __u8 header_ext
[32]; /* reserved: fill with 0xff */
175 __u8 pad2
[3]; /* 0xff */
176 be32 workspace_len
; /* sectors for vendor space -
177 * at least 32768(sectors) */
179 be16 max_pd_entries
; /* one of 15, 63, 255, 1023, 4095 */
180 be16 max_vd_entries
; /* 2^(4,6,8,10,12)-1 : i.e. as above */
181 be16 max_partitions
; /* i.e. max num of configuration
182 record entries per disk */
183 be16 config_record_len
; /* 1 +ROUNDUP(max_primary_element_entries
185 be16 max_primary_element_entries
; /* 16, 64, 256, 1024, or 4096 */
186 __u8 pad3
[54]; /* 0xff */
187 /* 192 bytes so far */
188 be32 controller_section_offset
;
189 be32 controller_section_length
;
190 be32 phys_section_offset
;
191 be32 phys_section_length
;
192 be32 virt_section_offset
;
193 be32 virt_section_length
;
194 be32 config_section_offset
;
195 be32 config_section_length
;
196 be32 data_section_offset
;
197 be32 data_section_length
;
198 be32 bbm_section_offset
;
199 be32 bbm_section_length
;
200 be32 diag_space_offset
;
201 be32 diag_space_length
;
204 /* 256 bytes so far */
205 __u8 pad4
[256]; /* 0xff */
209 #define DDF_HEADER_ANCHOR 0x00
210 #define DDF_HEADER_PRIMARY 0x01
211 #define DDF_HEADER_SECONDARY 0x02
213 /* The content of the 'controller section' - global scope */
214 struct ddf_controller_data
{
215 be32 magic
; /* DDF_CONTROLLER_MAGIC */
217 char guid
[DDF_GUID_LEN
];
218 struct controller_type
{
225 __u8 pad
[8]; /* 0xff */
226 __u8 vendor_data
[448];
229 /* The content of phys_section - global scope */
231 be32 magic
; /* DDF_PHYS_RECORDS_MAGIC */
233 be16 used_pdes
; /* This is a counter, not a max - the list
234 * of used entries may not be dense */
237 struct phys_disk_entry
{
238 char guid
[DDF_GUID_LEN
];
242 be64 config_size
; /* DDF structures must be after here */
243 char path
[18]; /* Another horrible structure really
244 * but is "used for information
250 /* phys_disk_entry.type is a bitmap - bigendian remember */
251 #define DDF_Forced_PD_GUID 1
252 #define DDF_Active_in_VD 2
253 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
254 #define DDF_Spare 8 /* overrides Global_spare */
255 #define DDF_Foreign 16
256 #define DDF_Legacy 32 /* no DDF on this device */
258 #define DDF_Interface_mask 0xf00
259 #define DDF_Interface_SCSI 0x100
260 #define DDF_Interface_SAS 0x200
261 #define DDF_Interface_SATA 0x300
262 #define DDF_Interface_FC 0x400
264 /* phys_disk_entry.state is a bigendian bitmap */
266 #define DDF_Failed 2 /* overrides 1,4,8 */
267 #define DDF_Rebuilding 4
268 #define DDF_Transition 8
270 #define DDF_ReadErrors 32
271 #define DDF_Missing 64
273 /* The content of the virt_section global scope */
274 struct virtual_disk
{
275 be32 magic
; /* DDF_VIRT_RECORDS_MAGIC */
280 struct virtual_entry
{
281 char guid
[DDF_GUID_LEN
];
283 __u16 pad0
; /* 0xffff */
293 /* virtual_entry.type is a bitmap - bigendian */
295 #define DDF_Enforce_Groups 2
296 #define DDF_Unicode 4
297 #define DDF_Owner_Valid 8
299 /* virtual_entry.state is a bigendian bitmap */
300 #define DDF_state_mask 0x7
301 #define DDF_state_optimal 0x0
302 #define DDF_state_degraded 0x1
303 #define DDF_state_deleted 0x2
304 #define DDF_state_missing 0x3
305 #define DDF_state_failed 0x4
306 #define DDF_state_part_optimal 0x5
308 #define DDF_state_morphing 0x8
309 #define DDF_state_inconsistent 0x10
311 /* virtual_entry.init_state is a bigendian bitmap */
312 #define DDF_initstate_mask 0x03
313 #define DDF_init_not 0x00
314 #define DDF_init_quick 0x01 /* initialisation is progress.
315 * i.e. 'state_inconsistent' */
316 #define DDF_init_full 0x02
318 #define DDF_access_mask 0xc0
319 #define DDF_access_rw 0x00
320 #define DDF_access_ro 0x80
321 #define DDF_access_blocked 0xc0
323 /* The content of the config_section - local scope
324 * It has multiple records each config_record_len sectors
325 * They can be vd_config or spare_assign
329 be32 magic
; /* DDF_VD_CONF_MAGIC */
331 char guid
[DDF_GUID_LEN
];
335 be16 prim_elmnt_count
;
336 __u8 chunk_shift
; /* 0 == 512, 1==1024 etc */
339 __u8 sec_elmnt_count
;
342 be64 blocks
; /* blocks per component could be different
343 * on different component devices...(only
344 * for concat I hope) */
345 be64 array_blocks
; /* blocks in array */
347 be32 spare_refs
[8]; /* This is used to detect missing spares.
348 * As we don't have an interface for that
349 * the values are ignored.
356 __u8 v0
[32]; /* reserved- 0xff */
357 __u8 v1
[32]; /* reserved- 0xff */
358 __u8 v2
[16]; /* reserved- 0xff */
359 __u8 v3
[16]; /* reserved- 0xff */
361 be32 phys_refnum
[0]; /* refnum of each disk in sequence */
362 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
363 bvd are always the same size */
365 #define LBA_OFFSET(ddf, vd) ((be64 *) &(vd)->phys_refnum[(ddf)->mppe])
367 /* vd_config.cache_pol[7] is a bitmap */
368 #define DDF_cache_writeback 1 /* else writethrough */
369 #define DDF_cache_wadaptive 2 /* only applies if writeback */
370 #define DDF_cache_readahead 4
371 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
372 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
373 #define DDF_cache_wallowed 32 /* enable write caching */
374 #define DDF_cache_rallowed 64 /* enable read caching */
376 struct spare_assign
{
377 be32 magic
; /* DDF_SPARE_ASSIGN_MAGIC */
382 be16 populated
; /* SAEs used */
383 be16 max
; /* max SAEs */
385 struct spare_assign_entry
{
386 char guid
[DDF_GUID_LEN
];
387 be16 secondary_element
;
391 /* spare_assign.type is a bitmap */
392 #define DDF_spare_dedicated 0x1 /* else global */
393 #define DDF_spare_revertible 0x2 /* else committable */
394 #define DDF_spare_active 0x4 /* else not active */
395 #define DDF_spare_affinity 0x8 /* enclosure affinity */
397 /* The data_section contents - local scope */
399 be32 magic
; /* DDF_PHYS_DATA_MAGIC */
401 char guid
[DDF_GUID_LEN
];
402 be32 refnum
; /* crc of some magic drive data ... */
403 __u8 forced_ref
; /* set when above was not result of magic */
404 __u8 forced_guid
; /* set if guid was forced rather than magic */
409 /* bbm_section content */
410 struct bad_block_log
{
417 struct mapped_block
{
418 be64 defective_start
;
419 be32 replacement_start
;
425 /* Struct for internally holding ddf structures */
426 /* The DDF structure stored on each device is potentially
427 * quite different, as some data is global and some is local.
428 * The global data is:
431 * - Physical disk records
432 * - Virtual disk records
434 * - Configuration records
435 * - Physical Disk data section
436 * ( and Bad block and vendor which I don't care about yet).
438 * The local data is parsed into separate lists as it is read
439 * and reconstructed for writing. This means that we only need
440 * to make config changes once and they are automatically
441 * propagated to all devices.
442 * The global (config and disk data) records are each in a list
443 * of separate data structures. When writing we find the entry
444 * or entries applicable to the particular device.
447 struct ddf_header anchor
, primary
, secondary
;
448 struct ddf_controller_data controller
;
449 struct ddf_header
*active
;
450 struct phys_disk
*phys
;
451 struct virtual_disk
*virt
;
454 unsigned int max_part
, mppe
, conf_rec_len
;
462 unsigned int vcnum
; /* index into ->virt */
463 /* For an array with a secondary level there are
464 * multiple vd_config structures, all with the same
465 * guid but with different sec_elmnt_seq.
466 * One of these structures is in 'conf' below.
467 * The others are in other_bvds, not in any
470 struct vd_config
**other_bvds
;
471 __u64
*block_sizes
; /* NULL if all the same */
474 struct vd_config conf
;
475 } *conflist
, *currentconf
;
484 unsigned long long size
; /* sectors */
485 be64 primary_lba
; /* sectors */
486 be64 secondary_lba
; /* sectors */
487 be64 workspace_lba
; /* sectors */
488 int pdnum
; /* index in ->phys */
489 struct spare_assign
*spare
;
490 void *mdupdate
; /* hold metadata update */
492 /* These fields used by auto-layout */
493 int raiddisk
; /* slot to fill in autolayout */
498 struct disk_data disk
;
499 struct vcl
*vlist
[0]; /* max_part in size */
503 static int load_super_ddf_all(struct supertype
*st
, int fd
,
504 void **sbp
, char *devname
);
505 static int get_svd_state(const struct ddf_super
*, const struct vcl
*);
507 validate_geometry_ddf_container(struct supertype
*st
,
508 int level
, int layout
, int raiddisks
,
509 int chunk
, unsigned long long size
,
510 unsigned long long data_offset
,
511 char *dev
, unsigned long long *freesize
,
514 static int validate_geometry_ddf_bvd(struct supertype
*st
,
515 int level
, int layout
, int raiddisks
,
516 int *chunk
, unsigned long long size
,
517 unsigned long long data_offset
,
518 char *dev
, unsigned long long *freesize
,
521 static void free_super_ddf(struct supertype
*st
);
522 static int all_ff(const char *guid
);
523 static unsigned int get_pd_index_from_refnum(const struct vcl
*vc
,
524 be32 refnum
, unsigned int nmax
,
525 const struct vd_config
**bvd
,
527 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
, char *map
);
528 static void uuid_from_ddf_guid(const char *guid
, int uuid
[4]);
529 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4]);
530 static void _ddf_array_name(char *name
, const struct ddf_super
*ddf
, int i
);
531 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
, char *map
);
532 static int init_super_ddf_bvd(struct supertype
*st
,
533 mdu_array_info_t
*info
,
534 unsigned long long size
,
535 char *name
, char *homehost
,
536 int *uuid
, unsigned long long data_offset
);
539 static void pr_state(struct ddf_super
*ddf
, const char *msg
)
542 dprintf("%s: ", msg
);
543 for (i
= 0; i
< be16_to_cpu(ddf
->active
->max_vd_entries
); i
++) {
544 if (all_ff(ddf
->virt
->entries
[i
].guid
))
546 dprintf_cont("%u(s=%02x i=%02x) ", i
,
547 ddf
->virt
->entries
[i
].state
,
548 ddf
->virt
->entries
[i
].init_state
);
553 static void pr_state(const struct ddf_super
*ddf
, const char *msg
) {}
556 static void _ddf_set_updates_pending(struct ddf_super
*ddf
, struct vd_config
*vc
,
560 vc
->timestamp
= cpu_to_be32(time(0)-DECADE
);
561 vc
->seqnum
= cpu_to_be32(be32_to_cpu(vc
->seqnum
) + 1);
563 if (ddf
->updates_pending
)
565 ddf
->updates_pending
= 1;
566 ddf
->active
->seq
= cpu_to_be32((be32_to_cpu(ddf
->active
->seq
)+1));
570 #define ddf_set_updates_pending(x,v) _ddf_set_updates_pending((x), (v), __func__)
572 static be32
calc_crc(void *buf
, int len
)
574 /* crcs are always at the same place as in the ddf_header */
575 struct ddf_header
*ddf
= buf
;
576 be32 oldcrc
= ddf
->crc
;
578 ddf
->crc
= cpu_to_be32(0xffffffff);
580 newcrc
= crc32(0, buf
, len
);
582 /* The crc is stored (like everything) bigendian, so convert
583 * here for simplicity
585 return cpu_to_be32(newcrc
);
588 #define DDF_INVALID_LEVEL 0xff
589 #define DDF_NO_SECONDARY 0xff
590 static int err_bad_md_layout(const mdu_array_info_t
*array
)
592 pr_err("RAID%d layout %x with %d disks is unsupported for DDF\n",
593 array
->level
, array
->layout
, array
->raid_disks
);
597 static int layout_md2ddf(const mdu_array_info_t
*array
,
598 struct vd_config
*conf
)
600 be16 prim_elmnt_count
= cpu_to_be16(array
->raid_disks
);
601 __u8 prl
= DDF_INVALID_LEVEL
, rlq
= 0;
602 __u8 sec_elmnt_count
= 1;
603 __u8 srl
= DDF_NO_SECONDARY
;
605 switch (array
->level
) {
610 rlq
= DDF_RAID0_SIMPLE
;
614 switch (array
->raid_disks
) {
616 rlq
= DDF_RAID1_SIMPLE
;
619 rlq
= DDF_RAID1_MULTI
;
622 return err_bad_md_layout(array
);
627 if (array
->layout
!= 0)
628 return err_bad_md_layout(array
);
633 switch (array
->layout
) {
634 case ALGORITHM_LEFT_ASYMMETRIC
:
635 rlq
= DDF_RAID5_N_RESTART
;
637 case ALGORITHM_RIGHT_ASYMMETRIC
:
638 rlq
= DDF_RAID5_0_RESTART
;
640 case ALGORITHM_LEFT_SYMMETRIC
:
641 rlq
= DDF_RAID5_N_CONTINUE
;
643 case ALGORITHM_RIGHT_SYMMETRIC
:
644 /* not mentioned in standard */
646 return err_bad_md_layout(array
);
651 switch (array
->layout
) {
652 case ALGORITHM_ROTATING_N_RESTART
:
653 rlq
= DDF_RAID5_N_RESTART
;
655 case ALGORITHM_ROTATING_ZERO_RESTART
:
656 rlq
= DDF_RAID6_0_RESTART
;
658 case ALGORITHM_ROTATING_N_CONTINUE
:
659 rlq
= DDF_RAID5_N_CONTINUE
;
662 return err_bad_md_layout(array
);
667 if (array
->raid_disks
% 2 == 0 && array
->layout
== 0x102) {
668 rlq
= DDF_RAID1_SIMPLE
;
669 prim_elmnt_count
= cpu_to_be16(2);
670 sec_elmnt_count
= array
->raid_disks
/ 2;
673 } else if (array
->raid_disks
% 3 == 0 &&
674 array
->layout
== 0x103) {
675 rlq
= DDF_RAID1_MULTI
;
676 prim_elmnt_count
= cpu_to_be16(3);
677 sec_elmnt_count
= array
->raid_disks
/ 3;
680 } else if (array
->layout
== 0x201) {
682 rlq
= DDF_RAID1E_OFFSET
;
683 } else if (array
->layout
== 0x102) {
685 rlq
= DDF_RAID1E_ADJACENT
;
687 return err_bad_md_layout(array
);
690 return err_bad_md_layout(array
);
693 conf
->prim_elmnt_count
= prim_elmnt_count
;
696 conf
->sec_elmnt_count
= sec_elmnt_count
;
700 static int err_bad_ddf_layout(const struct vd_config
*conf
)
702 pr_err("DDF RAID %u qualifier %u with %u disks is unsupported\n",
703 conf
->prl
, conf
->rlq
, be16_to_cpu(conf
->prim_elmnt_count
));
707 static int layout_ddf2md(const struct vd_config
*conf
,
708 mdu_array_info_t
*array
)
710 int level
= LEVEL_UNSUPPORTED
;
712 int raiddisks
= be16_to_cpu(conf
->prim_elmnt_count
);
714 if (conf
->sec_elmnt_count
> 1) {
715 /* see also check_secondary() */
716 if (conf
->prl
!= DDF_RAID1
||
717 (conf
->srl
!= DDF_2STRIPED
&& conf
->srl
!= DDF_2SPANNED
)) {
718 pr_err("Unsupported secondary RAID level %u/%u\n",
719 conf
->prl
, conf
->srl
);
722 if (raiddisks
== 2 && conf
->rlq
== DDF_RAID1_SIMPLE
)
724 else if (raiddisks
== 3 && conf
->rlq
== DDF_RAID1_MULTI
)
727 return err_bad_ddf_layout(conf
);
728 raiddisks
*= conf
->sec_elmnt_count
;
735 level
= LEVEL_LINEAR
;
738 if (conf
->rlq
!= DDF_RAID0_SIMPLE
)
739 return err_bad_ddf_layout(conf
);
743 if (!((conf
->rlq
== DDF_RAID1_SIMPLE
&& raiddisks
== 2) ||
744 (conf
->rlq
== DDF_RAID1_MULTI
&& raiddisks
== 3)))
745 return err_bad_ddf_layout(conf
);
749 if (conf
->rlq
== DDF_RAID1E_ADJACENT
)
751 else if (conf
->rlq
== DDF_RAID1E_OFFSET
)
754 return err_bad_ddf_layout(conf
);
758 if (conf
->rlq
!= DDF_RAID4_N
)
759 return err_bad_ddf_layout(conf
);
764 case DDF_RAID5_N_RESTART
:
765 layout
= ALGORITHM_LEFT_ASYMMETRIC
;
767 case DDF_RAID5_0_RESTART
:
768 layout
= ALGORITHM_RIGHT_ASYMMETRIC
;
770 case DDF_RAID5_N_CONTINUE
:
771 layout
= ALGORITHM_LEFT_SYMMETRIC
;
774 return err_bad_ddf_layout(conf
);
780 case DDF_RAID5_N_RESTART
:
781 layout
= ALGORITHM_ROTATING_N_RESTART
;
783 case DDF_RAID6_0_RESTART
:
784 layout
= ALGORITHM_ROTATING_ZERO_RESTART
;
786 case DDF_RAID5_N_CONTINUE
:
787 layout
= ALGORITHM_ROTATING_N_CONTINUE
;
790 return err_bad_ddf_layout(conf
);
795 return err_bad_ddf_layout(conf
);
799 array
->level
= level
;
800 array
->layout
= layout
;
801 array
->raid_disks
= raiddisks
;
805 static int load_ddf_header(int fd
, unsigned long long lba
,
806 unsigned long long size
,
808 struct ddf_header
*hdr
, struct ddf_header
*anchor
)
810 /* read a ddf header (primary or secondary) from fd/lba
811 * and check that it is consistent with anchor
813 * magic, crc, guid, rev, and LBA's header_type, and
814 * everything after header_type must be the same
819 if (lseek64(fd
, lba
<<9, 0) < 0)
822 if (read(fd
, hdr
, 512) != 512)
825 if (!be32_eq(hdr
->magic
, DDF_HEADER_MAGIC
)) {
826 pr_err("bad header magic\n");
829 if (!be32_eq(calc_crc(hdr
, 512), hdr
->crc
)) {
833 if (memcmp(anchor
->guid
, hdr
->guid
, DDF_GUID_LEN
) != 0 ||
834 memcmp(anchor
->revision
, hdr
->revision
, 8) != 0 ||
835 !be64_eq(anchor
->primary_lba
, hdr
->primary_lba
) ||
836 !be64_eq(anchor
->secondary_lba
, hdr
->secondary_lba
) ||
838 memcmp(anchor
->pad2
, hdr
->pad2
, 512 -
839 offsetof(struct ddf_header
, pad2
)) != 0) {
840 pr_err("header mismatch\n");
844 /* Looks good enough to me... */
848 static void *load_section(int fd
, struct ddf_super
*super
, void *buf
,
849 be32 offset_be
, be32 len_be
, int check
)
851 unsigned long long offset
= be32_to_cpu(offset_be
);
852 unsigned long long len
= be32_to_cpu(len_be
);
853 int dofree
= (buf
== NULL
);
856 if (len
!= 2 && len
!= 8 && len
!= 32 &&
857 len
!= 128 && len
!= 512)
862 if (!buf
&& posix_memalign(&buf
, 512, len
<<9) != 0)
868 if (super
->active
->type
== 1)
869 offset
+= be64_to_cpu(super
->active
->primary_lba
);
871 offset
+= be64_to_cpu(super
->active
->secondary_lba
);
873 if ((unsigned long long)lseek64(fd
, offset
<<9, 0) != (offset
<<9)) {
878 if ((unsigned long long)read(fd
, buf
, len
<<9) != (len
<<9)) {
886 static int load_ddf_headers(int fd
, struct ddf_super
*super
, char *devname
)
888 unsigned long long dsize
;
890 get_dev_size(fd
, NULL
, &dsize
);
892 if (lseek64(fd
, dsize
-512, 0) < 0) {
894 pr_err("Cannot seek to anchor block on %s: %s\n",
895 devname
, strerror(errno
));
898 if (read(fd
, &super
->anchor
, 512) != 512) {
900 pr_err("Cannot read anchor block on %s: %s\n",
901 devname
, strerror(errno
));
904 if (!be32_eq(super
->anchor
.magic
, DDF_HEADER_MAGIC
)) {
906 pr_err("no DDF anchor found on %s\n",
910 if (!be32_eq(calc_crc(&super
->anchor
, 512), super
->anchor
.crc
)) {
912 pr_err("bad CRC on anchor on %s\n",
916 if (memcmp(super
->anchor
.revision
, DDF_REVISION_0
, 8) != 0 &&
917 memcmp(super
->anchor
.revision
, DDF_REVISION_2
, 8) != 0) {
919 pr_err("can only support super revision %.8s and earlier, not %.8s on %s\n",
920 DDF_REVISION_2
, super
->anchor
.revision
,devname
);
923 super
->active
= NULL
;
924 if (load_ddf_header(fd
, be64_to_cpu(super
->anchor
.primary_lba
),
926 &super
->primary
, &super
->anchor
) == 0) {
928 pr_err("Failed to load primary DDF header on %s\n", devname
);
930 super
->active
= &super
->primary
;
932 if (load_ddf_header(fd
, be64_to_cpu(super
->anchor
.secondary_lba
),
934 &super
->secondary
, &super
->anchor
)) {
935 if (super
->active
== NULL
||
936 (be32_to_cpu(super
->primary
.seq
)
937 < be32_to_cpu(super
->secondary
.seq
) &&
938 !super
->secondary
.openflag
) ||
939 (be32_to_cpu(super
->primary
.seq
) ==
940 be32_to_cpu(super
->secondary
.seq
) &&
941 super
->primary
.openflag
&& !super
->secondary
.openflag
))
942 super
->active
= &super
->secondary
;
943 } else if (devname
&&
944 be64_to_cpu(super
->anchor
.secondary_lba
) != ~(__u64
)0)
945 pr_err("Failed to load secondary DDF header on %s\n",
947 if (super
->active
== NULL
)
952 static int load_ddf_global(int fd
, struct ddf_super
*super
, char *devname
)
955 ok
= load_section(fd
, super
, &super
->controller
,
956 super
->active
->controller_section_offset
,
957 super
->active
->controller_section_length
,
959 super
->phys
= load_section(fd
, super
, NULL
,
960 super
->active
->phys_section_offset
,
961 super
->active
->phys_section_length
,
963 super
->pdsize
= be32_to_cpu(super
->active
->phys_section_length
) * 512;
965 super
->virt
= load_section(fd
, super
, NULL
,
966 super
->active
->virt_section_offset
,
967 super
->active
->virt_section_length
,
969 super
->vdsize
= be32_to_cpu(super
->active
->virt_section_length
) * 512;
979 super
->conflist
= NULL
;
982 super
->max_part
= be16_to_cpu(super
->active
->max_partitions
);
983 super
->mppe
= be16_to_cpu(super
->active
->max_primary_element_entries
);
984 super
->conf_rec_len
= be16_to_cpu(super
->active
->config_record_len
);
988 #define DDF_UNUSED_BVD 0xff
989 static int alloc_other_bvds(const struct ddf_super
*ddf
, struct vcl
*vcl
)
991 unsigned int n_vds
= vcl
->conf
.sec_elmnt_count
- 1;
992 unsigned int i
, vdsize
;
995 vcl
->other_bvds
= NULL
;
998 vdsize
= ddf
->conf_rec_len
* 512;
999 if (posix_memalign(&p
, 512, n_vds
*
1000 (vdsize
+ sizeof(struct vd_config
*))) != 0)
1002 vcl
->other_bvds
= (struct vd_config
**) (p
+ n_vds
* vdsize
);
1003 for (i
= 0; i
< n_vds
; i
++) {
1004 vcl
->other_bvds
[i
] = p
+ i
* vdsize
;
1005 memset(vcl
->other_bvds
[i
], 0, vdsize
);
1006 vcl
->other_bvds
[i
]->sec_elmnt_seq
= DDF_UNUSED_BVD
;
1011 static void add_other_bvd(struct vcl
*vcl
, struct vd_config
*vd
,
1015 for (i
= 0; i
< vcl
->conf
.sec_elmnt_count
-1; i
++)
1016 if (vcl
->other_bvds
[i
]->sec_elmnt_seq
== vd
->sec_elmnt_seq
)
1019 if (i
< vcl
->conf
.sec_elmnt_count
-1) {
1020 if (be32_to_cpu(vd
->seqnum
) <=
1021 be32_to_cpu(vcl
->other_bvds
[i
]->seqnum
))
1024 for (i
= 0; i
< vcl
->conf
.sec_elmnt_count
-1; i
++)
1025 if (vcl
->other_bvds
[i
]->sec_elmnt_seq
== DDF_UNUSED_BVD
)
1027 if (i
== vcl
->conf
.sec_elmnt_count
-1) {
1028 pr_err("no space for sec level config %u, count is %u\n",
1029 vd
->sec_elmnt_seq
, vcl
->conf
.sec_elmnt_count
);
1033 memcpy(vcl
->other_bvds
[i
], vd
, len
);
1036 static int load_ddf_local(int fd
, struct ddf_super
*super
,
1037 char *devname
, int keep
)
1043 unsigned int confsec
;
1045 unsigned int max_virt_disks
=
1046 be16_to_cpu(super
->active
->max_vd_entries
);
1047 unsigned long long dsize
;
1049 /* First the local disk info */
1050 if (posix_memalign((void**)&dl
, 512,
1052 (super
->max_part
) * sizeof(dl
->vlist
[0])) != 0) {
1053 pr_err("could not allocate disk info buffer\n");
1057 load_section(fd
, super
, &dl
->disk
,
1058 super
->active
->data_section_offset
,
1059 super
->active
->data_section_length
,
1061 dl
->devname
= devname
? xstrdup(devname
) : NULL
;
1064 dl
->major
= major(stb
.st_rdev
);
1065 dl
->minor
= minor(stb
.st_rdev
);
1066 dl
->next
= super
->dlist
;
1067 dl
->fd
= keep
? fd
: -1;
1070 if (get_dev_size(fd
, devname
, &dsize
))
1071 dl
->size
= dsize
>> 9;
1072 /* If the disks have different sizes, the LBAs will differ
1073 * between phys disks.
1074 * At this point here, the values in super->active must be valid
1075 * for this phys disk. */
1076 dl
->primary_lba
= super
->active
->primary_lba
;
1077 dl
->secondary_lba
= super
->active
->secondary_lba
;
1078 dl
->workspace_lba
= super
->active
->workspace_lba
;
1080 for (i
= 0 ; i
< super
->max_part
; i
++)
1081 dl
->vlist
[i
] = NULL
;
1084 for (i
= 0; i
< be16_to_cpu(super
->active
->max_pd_entries
); i
++)
1085 if (memcmp(super
->phys
->entries
[i
].guid
,
1086 dl
->disk
.guid
, DDF_GUID_LEN
) == 0)
1089 /* Now the config list. */
1090 /* 'conf' is an array of config entries, some of which are
1091 * probably invalid. Those which are good need to be copied into
1095 conf
= load_section(fd
, super
, super
->conf
,
1096 super
->active
->config_section_offset
,
1097 super
->active
->config_section_length
,
1102 confsec
< be32_to_cpu(super
->active
->config_section_length
);
1103 confsec
+= super
->conf_rec_len
) {
1104 struct vd_config
*vd
=
1105 (struct vd_config
*)((char*)conf
+ confsec
*512);
1108 if (be32_eq(vd
->magic
, DDF_SPARE_ASSIGN_MAGIC
)) {
1111 if (posix_memalign((void**)&dl
->spare
, 512,
1112 super
->conf_rec_len
*512) != 0) {
1113 pr_err("could not allocate spare info buf\n");
1117 memcpy(dl
->spare
, vd
, super
->conf_rec_len
*512);
1120 if (!be32_eq(vd
->magic
, DDF_VD_CONF_MAGIC
))
1121 /* Must be vendor-unique - I cannot handle those */
1124 for (vcl
= super
->conflist
; vcl
; vcl
= vcl
->next
) {
1125 if (memcmp(vcl
->conf
.guid
,
1126 vd
->guid
, DDF_GUID_LEN
) == 0)
1131 dl
->vlist
[vnum
++] = vcl
;
1132 if (vcl
->other_bvds
!= NULL
&&
1133 vcl
->conf
.sec_elmnt_seq
!= vd
->sec_elmnt_seq
) {
1134 add_other_bvd(vcl
, vd
, super
->conf_rec_len
*512);
1137 if (be32_to_cpu(vd
->seqnum
) <=
1138 be32_to_cpu(vcl
->conf
.seqnum
))
1141 if (posix_memalign((void**)&vcl
, 512,
1142 (super
->conf_rec_len
*512 +
1143 offsetof(struct vcl
, conf
))) != 0) {
1144 pr_err("could not allocate vcl buf\n");
1147 vcl
->next
= super
->conflist
;
1148 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
1149 vcl
->conf
.sec_elmnt_count
= vd
->sec_elmnt_count
;
1150 if (alloc_other_bvds(super
, vcl
) != 0) {
1151 pr_err("could not allocate other bvds\n");
1155 super
->conflist
= vcl
;
1156 dl
->vlist
[vnum
++] = vcl
;
1158 memcpy(&vcl
->conf
, vd
, super
->conf_rec_len
*512);
1159 for (i
=0; i
< max_virt_disks
; i
++)
1160 if (memcmp(super
->virt
->entries
[i
].guid
,
1161 vcl
->conf
.guid
, DDF_GUID_LEN
)==0)
1163 if (i
< max_virt_disks
)
1170 static int load_super_ddf(struct supertype
*st
, int fd
,
1173 unsigned long long dsize
;
1174 struct ddf_super
*super
;
1177 if (get_dev_size(fd
, devname
, &dsize
) == 0)
1180 if (test_partition(fd
))
1181 /* DDF is not allowed on partitions */
1184 /* 32M is a lower bound */
1185 if (dsize
<= 32*1024*1024) {
1187 pr_err("%s is too small for ddf: size is %llu sectors.\n",
1193 pr_err("%s is an odd size for ddf: size is %llu bytes.\n",
1200 if (posix_memalign((void**)&super
, 512, sizeof(*super
))!= 0) {
1201 pr_err("malloc of %zu failed.\n",
1205 memset(super
, 0, sizeof(*super
));
1207 rv
= load_ddf_headers(fd
, super
, devname
);
1213 /* Have valid headers and have chosen the best. Let's read in the rest*/
1215 rv
= load_ddf_global(fd
, super
, devname
);
1219 pr_err("Failed to load all information sections on %s\n", devname
);
1224 rv
= load_ddf_local(fd
, super
, devname
, 0);
1228 pr_err("Failed to load all information sections on %s\n", devname
);
1233 /* Should possibly check the sections .... */
1236 if (st
->ss
== NULL
) {
1237 st
->ss
= &super_ddf
;
1238 st
->minor_version
= 0;
1245 static void free_super_ddf(struct supertype
*st
)
1247 struct ddf_super
*ddf
= st
->sb
;
1253 while (ddf
->conflist
) {
1254 struct vcl
*v
= ddf
->conflist
;
1255 ddf
->conflist
= v
->next
;
1257 free(v
->block_sizes
);
1260 v->other_bvds[0] points to beginning of buffer,
1261 see alloc_other_bvds()
1263 free(v
->other_bvds
[0]);
1266 while (ddf
->dlist
) {
1267 struct dl
*d
= ddf
->dlist
;
1268 ddf
->dlist
= d
->next
;
1275 while (ddf
->add_list
) {
1276 struct dl
*d
= ddf
->add_list
;
1277 ddf
->add_list
= d
->next
;
1288 static struct supertype
*match_metadata_desc_ddf(char *arg
)
1290 /* 'ddf' only supports containers */
1291 struct supertype
*st
;
1292 if (strcmp(arg
, "ddf") != 0 &&
1293 strcmp(arg
, "default") != 0
1297 st
= xcalloc(1, sizeof(*st
));
1298 st
->ss
= &super_ddf
;
1300 st
->minor_version
= 0;
1305 static mapping_t ddf_state
[] = {
1311 { "Partially Optimal", 5},
1317 static mapping_t ddf_init_state
[] = {
1318 { "Not Initialised", 0},
1319 { "QuickInit in Progress", 1},
1320 { "Fully Initialised", 2},
1324 static mapping_t ddf_access
[] = {
1328 { "Blocked (no access)", 3},
1332 static mapping_t ddf_level
[] = {
1333 { "RAID0", DDF_RAID0
},
1334 { "RAID1", DDF_RAID1
},
1335 { "RAID3", DDF_RAID3
},
1336 { "RAID4", DDF_RAID4
},
1337 { "RAID5", DDF_RAID5
},
1338 { "RAID1E",DDF_RAID1E
},
1339 { "JBOD", DDF_JBOD
},
1340 { "CONCAT",DDF_CONCAT
},
1341 { "RAID5E",DDF_RAID5E
},
1342 { "RAID5EE",DDF_RAID5EE
},
1343 { "RAID6", DDF_RAID6
},
1346 static mapping_t ddf_sec_level
[] = {
1347 { "Striped", DDF_2STRIPED
},
1348 { "Mirrored", DDF_2MIRRORED
},
1349 { "Concat", DDF_2CONCAT
},
1350 { "Spanned", DDF_2SPANNED
},
1354 static int all_ff(const char *guid
)
1357 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
1358 if (guid
[i
] != (char)0xff)
1363 static const char *guid_str(const char *guid
)
1365 static char buf
[DDF_GUID_LEN
*2+1];
1368 for (i
= 0; i
< DDF_GUID_LEN
; i
++) {
1369 unsigned char c
= guid
[i
];
1370 if (c
>= 32 && c
< 127)
1371 p
+= sprintf(p
, "%c", c
);
1373 p
+= sprintf(p
, "%02x", c
);
1376 return (const char *) buf
;
1379 static void print_guid(char *guid
, int tstamp
)
1381 /* A GUIDs are part (or all) ASCII and part binary.
1382 * They tend to be space padded.
1383 * We print the GUID in HEX, then in parentheses add
1384 * any initial ASCII sequence, and a possible
1385 * time stamp from bytes 16-19
1387 int l
= DDF_GUID_LEN
;
1390 for (i
=0 ; i
<DDF_GUID_LEN
; i
++) {
1391 if ((i
&3)==0 && i
!= 0) printf(":");
1392 printf("%02X", guid
[i
]&255);
1396 while (l
&& guid
[l
-1] == ' ')
1398 for (i
=0 ; i
<l
; i
++) {
1399 if (guid
[i
] >= 0x20 && guid
[i
] < 0x7f)
1400 fputc(guid
[i
], stdout
);
1405 time_t then
= __be32_to_cpu(*(__u32
*)(guid
+16)) + DECADE
;
1408 tm
= localtime(&then
);
1409 strftime(tbuf
, 100, " %D %T",tm
);
1410 fputs(tbuf
, stdout
);
1415 static void examine_vd(int n
, struct ddf_super
*sb
, char *guid
)
1417 int crl
= sb
->conf_rec_len
;
1420 for (vcl
= sb
->conflist
; vcl
; vcl
= vcl
->next
) {
1422 struct vd_config
*vc
= &vcl
->conf
;
1424 if (!be32_eq(calc_crc(vc
, crl
*512), vc
->crc
))
1426 if (memcmp(vc
->guid
, guid
, DDF_GUID_LEN
) != 0)
1429 /* Ok, we know about this VD, let's give more details */
1430 printf(" Raid Devices[%d] : %d (", n
,
1431 be16_to_cpu(vc
->prim_elmnt_count
));
1432 for (i
= 0; i
< be16_to_cpu(vc
->prim_elmnt_count
); i
++) {
1434 int cnt
= be16_to_cpu(sb
->phys
->max_pdes
);
1435 for (j
=0; j
<cnt
; j
++)
1436 if (be32_eq(vc
->phys_refnum
[i
],
1437 sb
->phys
->entries
[j
].refnum
))
1444 printf("@%lluK", (unsigned long long) be64_to_cpu(LBA_OFFSET(sb
, vc
)[i
])/2);
1447 if (vc
->chunk_shift
!= 255)
1448 printf(" Chunk Size[%d] : %d sectors\n", n
,
1449 1 << vc
->chunk_shift
);
1450 printf(" Raid Level[%d] : %s\n", n
,
1451 map_num(ddf_level
, vc
->prl
)?:"-unknown-");
1452 if (vc
->sec_elmnt_count
!= 1) {
1453 printf(" Secondary Position[%d] : %d of %d\n", n
,
1454 vc
->sec_elmnt_seq
, vc
->sec_elmnt_count
);
1455 printf(" Secondary Level[%d] : %s\n", n
,
1456 map_num(ddf_sec_level
, vc
->srl
) ?: "-unknown-");
1458 printf(" Device Size[%d] : %llu\n", n
,
1459 be64_to_cpu(vc
->blocks
)/2);
1460 printf(" Array Size[%d] : %llu\n", n
,
1461 be64_to_cpu(vc
->array_blocks
)/2);
1465 static void examine_vds(struct ddf_super
*sb
)
1467 int cnt
= be16_to_cpu(sb
->virt
->populated_vdes
);
1469 printf(" Virtual Disks : %d\n", cnt
);
1471 for (i
= 0; i
< be16_to_cpu(sb
->virt
->max_vdes
); i
++) {
1472 struct virtual_entry
*ve
= &sb
->virt
->entries
[i
];
1473 if (all_ff(ve
->guid
))
1476 printf(" VD GUID[%d] : ", i
); print_guid(ve
->guid
, 1);
1478 printf(" unit[%d] : %d\n", i
, be16_to_cpu(ve
->unit
));
1479 printf(" state[%d] : %s, %s%s\n", i
,
1480 map_num(ddf_state
, ve
->state
& 7),
1481 (ve
->state
& DDF_state_morphing
) ? "Morphing, ": "",
1482 (ve
->state
& DDF_state_inconsistent
)? "Not Consistent" : "Consistent");
1483 printf(" init state[%d] : %s\n", i
,
1484 map_num(ddf_init_state
, ve
->init_state
&DDF_initstate_mask
));
1485 printf(" access[%d] : %s\n", i
,
1486 map_num(ddf_access
, (ve
->init_state
& DDF_access_mask
) >> 6));
1487 printf(" Name[%d] : %.16s\n", i
, ve
->name
);
1488 examine_vd(i
, sb
, ve
->guid
);
1490 if (cnt
) printf("\n");
1493 static void examine_pds(struct ddf_super
*sb
)
1495 int cnt
= be16_to_cpu(sb
->phys
->max_pdes
);
1499 printf(" Physical Disks : %d\n", cnt
);
1500 printf(" Number RefNo Size Device Type/State\n");
1502 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
)
1505 for (i
=0 ; i
<cnt
; i
++) {
1506 struct phys_disk_entry
*pd
= &sb
->phys
->entries
[i
];
1507 int type
= be16_to_cpu(pd
->type
);
1508 int state
= be16_to_cpu(pd
->state
);
1510 if (be32_to_cpu(pd
->refnum
) == 0xffffffff)
1513 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1515 printf(" %3d %08x ", i
,
1516 be32_to_cpu(pd
->refnum
));
1518 be64_to_cpu(pd
->config_size
)>>1);
1519 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1520 if (be32_eq(dl
->disk
.refnum
, pd
->refnum
)) {
1521 char *dv
= map_dev(dl
->major
, dl
->minor
, 0);
1523 printf("%-15s", dv
);
1532 printf(" %s%s%s%s%s",
1533 (type
&2) ? "active":"",
1534 (type
&4) ? "Global-Spare":"",
1535 (type
&8) ? "spare" : "",
1536 (type
&16)? ", foreign" : "",
1537 (type
&32)? "pass-through" : "");
1538 if (state
& DDF_Failed
)
1539 /* This over-rides these three */
1540 state
&= ~(DDF_Online
|DDF_Rebuilding
|DDF_Transition
);
1541 printf("/%s%s%s%s%s%s%s",
1542 (state
&1)? "Online": "Offline",
1543 (state
&2)? ", Failed": "",
1544 (state
&4)? ", Rebuilding": "",
1545 (state
&8)? ", in-transition": "",
1546 (state
&16)? ", SMART-errors": "",
1547 (state
&32)? ", Unrecovered-Read-Errors": "",
1548 (state
&64)? ", Missing" : "");
1551 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1556 printf(" Physical disks not in metadata!:\n");
1558 dv
= map_dev(dl
->major
, dl
->minor
, 0);
1559 printf(" %08x %s\n", be32_to_cpu(dl
->disk
.refnum
),
1560 dv
? dv
: "-unknown-");
1566 static void examine_super_ddf(struct supertype
*st
, char *homehost
)
1568 struct ddf_super
*sb
= st
->sb
;
1570 printf(" Magic : %08x\n", be32_to_cpu(sb
->anchor
.magic
));
1571 printf(" Version : %.8s\n", sb
->anchor
.revision
);
1572 printf("Controller GUID : "); print_guid(sb
->controller
.guid
, 0);
1574 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1576 printf(" Seq : %08x\n", be32_to_cpu(sb
->active
->seq
));
1577 printf(" Redundant hdr : %s\n", (be32_eq(sb
->secondary
.magic
,
1584 static unsigned int get_vd_num_of_subarray(struct supertype
*st
)
1587 * Figure out the VD number for this supertype.
1588 * Returns DDF_CONTAINER for the container itself,
1589 * and DDF_NOTFOUND on error.
1591 struct ddf_super
*ddf
= st
->sb
;
1596 if (*st
->container_devnm
== '\0')
1597 return DDF_CONTAINER
;
1599 sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
1600 if (!sra
|| sra
->array
.major_version
!= -1 ||
1601 sra
->array
.minor_version
!= -2 ||
1602 !is_subarray(sra
->text_version
))
1603 return DDF_NOTFOUND
;
1605 sub
= strchr(sra
->text_version
+ 1, '/');
1607 vcnum
= strtoul(sub
+ 1, &end
, 10);
1608 if (sub
== NULL
|| *sub
== '\0' || *end
!= '\0' ||
1609 vcnum
>= be16_to_cpu(ddf
->active
->max_vd_entries
))
1610 return DDF_NOTFOUND
;
1615 static void brief_examine_super_ddf(struct supertype
*st
, int verbose
)
1617 /* We just write a generic DDF ARRAY entry
1621 getinfo_super_ddf(st
, &info
, NULL
);
1622 fname_from_uuid(st
, &info
, nbuf
, ':');
1624 printf("ARRAY metadata=ddf UUID=%s\n", nbuf
+ 5);
1627 static void brief_examine_subarrays_ddf(struct supertype
*st
, int verbose
)
1629 /* We write a DDF ARRAY member entry for each vd, identifying container
1630 * by uuid and member by unit number and uuid.
1632 struct ddf_super
*ddf
= st
->sb
;
1636 getinfo_super_ddf(st
, &info
, NULL
);
1637 fname_from_uuid(st
, &info
, nbuf
, ':');
1639 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
1640 struct virtual_entry
*ve
= &ddf
->virt
->entries
[i
];
1644 if (all_ff(ve
->guid
))
1646 memcpy(vcl
.conf
.guid
, ve
->guid
, DDF_GUID_LEN
);
1647 ddf
->currentconf
=&vcl
;
1649 uuid_from_super_ddf(st
, info
.uuid
);
1650 fname_from_uuid(st
, &info
, nbuf1
, ':');
1651 _ddf_array_name(namebuf
, ddf
, i
);
1652 printf("ARRAY%s%s container=%s member=%d UUID=%s\n",
1653 namebuf
[0] == '\0' ? "" : " /dev/md/", namebuf
,
1654 nbuf
+5, i
, nbuf1
+5);
1658 static void export_examine_super_ddf(struct supertype
*st
)
1662 getinfo_super_ddf(st
, &info
, NULL
);
1663 fname_from_uuid(st
, &info
, nbuf
, ':');
1664 printf("MD_METADATA=ddf\n");
1665 printf("MD_LEVEL=container\n");
1666 printf("MD_UUID=%s\n", nbuf
+5);
1667 printf("MD_DEVICES=%u\n",
1668 be16_to_cpu(((struct ddf_super
*)st
->sb
)->phys
->used_pdes
));
1671 static int copy_metadata_ddf(struct supertype
*st
, int from
, int to
)
1674 unsigned long long dsize
, offset
;
1676 struct ddf_header
*ddf
;
1679 /* The meta consists of an anchor, a primary, and a secondary.
1680 * This all lives at the end of the device.
1681 * So it is easiest to find the earliest of primary and
1682 * secondary, and copy everything from there.
1684 * Anchor is 512 from end. It contains primary_lba and secondary_lba
1685 * we choose one of those
1688 if (posix_memalign(&buf
, 4096, 4096) != 0)
1691 if (!get_dev_size(from
, NULL
, &dsize
))
1694 if (lseek64(from
, dsize
-512, 0) < 0)
1696 if (read(from
, buf
, 512) != 512)
1699 if (!be32_eq(ddf
->magic
, DDF_HEADER_MAGIC
) ||
1700 !be32_eq(calc_crc(ddf
, 512), ddf
->crc
) ||
1701 (memcmp(ddf
->revision
, DDF_REVISION_0
, 8) != 0 &&
1702 memcmp(ddf
->revision
, DDF_REVISION_2
, 8) != 0))
1705 offset
= dsize
- 512;
1706 if ((be64_to_cpu(ddf
->primary_lba
) << 9) < offset
)
1707 offset
= be64_to_cpu(ddf
->primary_lba
) << 9;
1708 if ((be64_to_cpu(ddf
->secondary_lba
) << 9) < offset
)
1709 offset
= be64_to_cpu(ddf
->secondary_lba
) << 9;
1711 bytes
= dsize
- offset
;
1713 if (lseek64(from
, offset
, 0) < 0 ||
1714 lseek64(to
, offset
, 0) < 0)
1716 while (written
< bytes
) {
1717 int n
= bytes
- written
;
1720 if (read(from
, buf
, n
) != n
)
1722 if (write(to
, buf
, n
) != n
)
1733 static void detail_super_ddf(struct supertype
*st
, char *homehost
,
1736 struct ddf_super
*sb
= st
->sb
;
1737 int cnt
= be16_to_cpu(sb
->virt
->populated_vdes
);
1739 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1741 printf(" Seq : %08x\n", be32_to_cpu(sb
->active
->seq
));
1742 printf(" Virtual Disks : %d\n", cnt
);
1746 static const char *vendors_with_variable_volume_UUID
[] = {
1750 static int volume_id_is_reliable(const struct ddf_super
*ddf
)
1752 int n
= ARRAY_SIZE(vendors_with_variable_volume_UUID
);
1754 for (i
= 0; i
< n
; i
++)
1755 if (!memcmp(ddf
->controller
.guid
,
1756 vendors_with_variable_volume_UUID
[i
], 8))
1761 static void uuid_of_ddf_subarray(const struct ddf_super
*ddf
,
1762 unsigned int vcnum
, int uuid
[4])
1764 char buf
[DDF_GUID_LEN
+18], sha
[20], *p
;
1765 struct sha1_ctx ctx
;
1766 if (volume_id_is_reliable(ddf
)) {
1767 uuid_from_ddf_guid(ddf
->virt
->entries
[vcnum
].guid
, uuid
);
1771 * Some fake RAID BIOSes (in particular, LSI ones) change the
1772 * VD GUID at every boot. These GUIDs are not suitable for
1773 * identifying an array. Luckily the header GUID appears to
1775 * We construct a pseudo-UUID from the header GUID and those
1776 * properties of the subarray that we expect to remain constant.
1778 memset(buf
, 0, sizeof(buf
));
1780 memcpy(p
, ddf
->anchor
.guid
, DDF_GUID_LEN
);
1782 memcpy(p
, ddf
->virt
->entries
[vcnum
].name
, 16);
1784 *((__u16
*) p
) = vcnum
;
1785 sha1_init_ctx(&ctx
);
1786 sha1_process_bytes(buf
, sizeof(buf
), &ctx
);
1787 sha1_finish_ctx(&ctx
, sha
);
1788 memcpy(uuid
, sha
, 4*4);
1791 static void brief_detail_super_ddf(struct supertype
*st
, char *subarray
)
1795 struct ddf_super
*ddf
= st
->sb
;
1796 unsigned int vcnum
= get_vd_num_of_subarray(st
);
1797 if (vcnum
== DDF_CONTAINER
)
1798 uuid_from_super_ddf(st
, info
.uuid
);
1799 else if (vcnum
== DDF_NOTFOUND
)
1802 uuid_of_ddf_subarray(ddf
, vcnum
, info
.uuid
);
1803 fname_from_uuid(st
, &info
, nbuf
,':');
1804 printf(" UUID=%s", nbuf
+ 5);
1807 static int match_home_ddf(struct supertype
*st
, char *homehost
)
1809 /* It matches 'this' host if the controller is a
1810 * Linux-MD controller with vendor_data matching
1811 * the hostname. It would be nice if we could
1812 * test against controller found in /sys or somewhere...
1814 struct ddf_super
*ddf
= st
->sb
;
1819 len
= strlen(homehost
);
1821 return (memcmp(ddf
->controller
.guid
, T10
, 8) == 0 &&
1822 len
< sizeof(ddf
->controller
.vendor_data
) &&
1823 memcmp(ddf
->controller
.vendor_data
, homehost
,len
) == 0 &&
1824 ddf
->controller
.vendor_data
[len
] == 0);
1827 static int find_index_in_bvd(const struct ddf_super
*ddf
,
1828 const struct vd_config
*conf
, unsigned int n
,
1829 unsigned int *n_bvd
)
1832 * Find the index of the n-th valid physical disk in this BVD.
1833 * Unused entries can be sprinkled in with the used entries,
1838 i
< ddf
->mppe
&& j
< be16_to_cpu(conf
->prim_elmnt_count
);
1840 if (be32_to_cpu(conf
->phys_refnum
[i
]) != 0xffffffff) {
1848 dprintf("couldn't find BVD member %u (total %u)\n",
1849 n
, be16_to_cpu(conf
->prim_elmnt_count
));
1853 /* Given a member array instance number, and a raid disk within that instance,
1854 * find the vd_config structure. The offset of the given disk in the phys_refnum
1855 * table is returned in n_bvd.
1856 * For two-level members with a secondary raid level the vd_config for
1857 * the appropriate BVD is returned.
1858 * The return value is always &vlc->conf, where vlc is returned in last pointer.
1860 static struct vd_config
*find_vdcr(struct ddf_super
*ddf
, unsigned int inst
,
1862 unsigned int *n_bvd
, struct vcl
**vcl
)
1866 for (v
= ddf
->conflist
; v
; v
= v
->next
) {
1867 unsigned int nsec
, ibvd
= 0;
1868 struct vd_config
*conf
;
1869 if (inst
!= v
->vcnum
)
1872 if (conf
->sec_elmnt_count
== 1) {
1873 if (find_index_in_bvd(ddf
, conf
, n
, n_bvd
)) {
1879 if (v
->other_bvds
== NULL
) {
1880 pr_err("BUG: other_bvds is NULL, nsec=%u\n",
1881 conf
->sec_elmnt_count
);
1884 nsec
= n
/ be16_to_cpu(conf
->prim_elmnt_count
);
1885 if (conf
->sec_elmnt_seq
!= nsec
) {
1886 for (ibvd
= 1; ibvd
< conf
->sec_elmnt_count
; ibvd
++) {
1887 if (v
->other_bvds
[ibvd
-1]->sec_elmnt_seq
==
1891 if (ibvd
== conf
->sec_elmnt_count
)
1893 conf
= v
->other_bvds
[ibvd
-1];
1895 if (!find_index_in_bvd(ddf
, conf
,
1896 n
- nsec
*conf
->sec_elmnt_count
, n_bvd
))
1898 dprintf("found disk %u as member %u in bvd %d of array %u\n",
1899 n
, *n_bvd
, ibvd
, inst
);
1904 pr_err("Couldn't find disk %d in array %u\n", n
, inst
);
1908 static int find_phys(const struct ddf_super
*ddf
, be32 phys_refnum
)
1910 /* Find the entry in phys_disk which has the given refnum
1911 * and return it's index
1914 for (i
= 0; i
< be16_to_cpu(ddf
->phys
->max_pdes
); i
++)
1915 if (be32_eq(ddf
->phys
->entries
[i
].refnum
, phys_refnum
))
1920 static void uuid_from_ddf_guid(const char *guid
, int uuid
[4])
1923 struct sha1_ctx ctx
;
1924 sha1_init_ctx(&ctx
);
1925 sha1_process_bytes(guid
, DDF_GUID_LEN
, &ctx
);
1926 sha1_finish_ctx(&ctx
, buf
);
1927 memcpy(uuid
, buf
, 4*4);
1930 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4])
1932 /* The uuid returned here is used for:
1933 * uuid to put into bitmap file (Create, Grow)
1934 * uuid for backup header when saving critical section (Grow)
1935 * comparing uuids when re-adding a device into an array
1936 * In these cases the uuid required is that of the data-array,
1937 * not the device-set.
1938 * uuid to recognise same set when adding a missing device back
1939 * to an array. This is a uuid for the device-set.
1941 * For each of these we can make do with a truncated
1942 * or hashed uuid rather than the original, as long as
1944 * In the case of SVD we assume the BVD is of interest,
1945 * though that might be the case if a bitmap were made for
1946 * a mirrored SVD - worry about that later.
1947 * So we need to find the VD configuration record for the
1948 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1949 * The first 16 bytes of the sha1 of these is used.
1951 struct ddf_super
*ddf
= st
->sb
;
1952 struct vcl
*vcl
= ddf
->currentconf
;
1955 uuid_of_ddf_subarray(ddf
, vcl
->vcnum
, uuid
);
1957 uuid_from_ddf_guid(ddf
->anchor
.guid
, uuid
);
1960 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1962 struct ddf_super
*ddf
= st
->sb
;
1963 int map_disks
= info
->array
.raid_disks
;
1966 if (ddf
->currentconf
) {
1967 getinfo_super_ddf_bvd(st
, info
, map
);
1970 memset(info
, 0, sizeof(*info
));
1972 info
->array
.raid_disks
= be16_to_cpu(ddf
->phys
->used_pdes
);
1973 info
->array
.level
= LEVEL_CONTAINER
;
1974 info
->array
.layout
= 0;
1975 info
->array
.md_minor
= -1;
1976 cptr
= (__u32
*)(ddf
->anchor
.guid
+ 16);
1977 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
1979 info
->array
.chunk_size
= 0;
1980 info
->container_enough
= 1;
1982 info
->disk
.major
= 0;
1983 info
->disk
.minor
= 0;
1985 struct phys_disk_entry
*pde
= NULL
;
1986 info
->disk
.number
= be32_to_cpu(ddf
->dlist
->disk
.refnum
);
1987 info
->disk
.raid_disk
= find_phys(ddf
, ddf
->dlist
->disk
.refnum
);
1989 info
->data_offset
= be64_to_cpu(ddf
->phys
->
1990 entries
[info
->disk
.raid_disk
].
1992 info
->component_size
= ddf
->dlist
->size
- info
->data_offset
;
1993 if (info
->disk
.raid_disk
>= 0)
1994 pde
= ddf
->phys
->entries
+ info
->disk
.raid_disk
;
1996 !(be16_to_cpu(pde
->state
) & DDF_Failed
) &&
1997 !(be16_to_cpu(pde
->state
) & DDF_Missing
))
1998 info
->disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
2000 info
->disk
.state
= 1 << MD_DISK_FAULTY
;
2003 /* There should always be a dlist, but just in case...*/
2004 info
->disk
.number
= -1;
2005 info
->disk
.raid_disk
= -1;
2006 info
->disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
2008 info
->events
= be32_to_cpu(ddf
->active
->seq
);
2009 info
->array
.utime
= DECADE
+ be32_to_cpu(ddf
->active
->timestamp
);
2011 info
->recovery_start
= MaxSector
;
2012 info
->reshape_active
= 0;
2013 info
->recovery_blocked
= 0;
2016 info
->array
.major_version
= -1;
2017 info
->array
.minor_version
= -2;
2018 strcpy(info
->text_version
, "ddf");
2019 info
->safe_mode_delay
= 0;
2021 uuid_from_super_ddf(st
, info
->uuid
);
2025 int max
= be16_to_cpu(ddf
->phys
->max_pdes
);
2026 for (i
= e
= 0 ; i
< map_disks
; i
++, e
++) {
2028 be32_to_cpu(ddf
->phys
->entries
[e
].refnum
) == 0xffffffff)
2030 if (i
< info
->array
.raid_disks
&& e
< max
&&
2031 !(be16_to_cpu(ddf
->phys
->entries
[e
].state
) &
2040 /* size of name must be at least 17 bytes! */
2041 static void _ddf_array_name(char *name
, const struct ddf_super
*ddf
, int i
)
2044 memcpy(name
, ddf
->virt
->entries
[i
].name
, 16);
2046 for(j
= 0; j
< 16; j
++)
2051 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2053 struct ddf_super
*ddf
= st
->sb
;
2054 struct vcl
*vc
= ddf
->currentconf
;
2055 int cd
= ddf
->currentdev
;
2058 struct dl
*dl
= NULL
;
2059 int map_disks
= info
->array
.raid_disks
;
2061 struct vd_config
*conf
;
2063 memset(info
, 0, sizeof(*info
));
2064 if (layout_ddf2md(&vc
->conf
, &info
->array
) == -1)
2066 info
->array
.md_minor
= -1;
2067 cptr
= (__u32
*)(vc
->conf
.guid
+ 16);
2068 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
2069 info
->array
.utime
= DECADE
+ be32_to_cpu(vc
->conf
.timestamp
);
2070 info
->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
2071 info
->custom_array_size
= be64_to_cpu(vc
->conf
.array_blocks
);
2074 n_prim
= be16_to_cpu(conf
->prim_elmnt_count
);
2075 if (conf
->sec_elmnt_count
> 1 && cd
>= n_prim
) {
2076 int ibvd
= cd
/ n_prim
- 1;
2078 conf
= vc
->other_bvds
[ibvd
];
2081 if (cd
>= 0 && (unsigned)cd
< ddf
->mppe
) {
2083 be64_to_cpu(LBA_OFFSET(ddf
, conf
)[cd
]);
2084 if (vc
->block_sizes
)
2085 info
->component_size
= vc
->block_sizes
[cd
];
2087 info
->component_size
= be64_to_cpu(conf
->blocks
);
2089 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2090 if (be32_eq(dl
->disk
.refnum
, conf
->phys_refnum
[cd
]))
2094 info
->disk
.major
= 0;
2095 info
->disk
.minor
= 0;
2096 info
->disk
.state
= 0;
2097 if (dl
&& dl
->pdnum
>= 0) {
2098 info
->disk
.major
= dl
->major
;
2099 info
->disk
.minor
= dl
->minor
;
2100 info
->disk
.raid_disk
= cd
+ conf
->sec_elmnt_seq
2101 * be16_to_cpu(conf
->prim_elmnt_count
);
2102 info
->disk
.number
= dl
->pdnum
;
2103 info
->disk
.state
= 0;
2104 if (info
->disk
.number
>= 0 &&
2105 (be16_to_cpu(ddf
->phys
->entries
[info
->disk
.number
].state
) & DDF_Online
) &&
2106 !(be16_to_cpu(ddf
->phys
->entries
[info
->disk
.number
].state
) & DDF_Failed
))
2107 info
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
2108 info
->events
= be32_to_cpu(ddf
->active
->seq
);
2111 info
->container_member
= ddf
->currentconf
->vcnum
;
2113 info
->recovery_start
= MaxSector
;
2114 info
->resync_start
= 0;
2115 info
->reshape_active
= 0;
2116 info
->recovery_blocked
= 0;
2117 if (!(ddf
->virt
->entries
[info
->container_member
].state
&
2118 DDF_state_inconsistent
) &&
2119 (ddf
->virt
->entries
[info
->container_member
].init_state
&
2120 DDF_initstate_mask
) == DDF_init_full
)
2121 info
->resync_start
= MaxSector
;
2123 uuid_from_super_ddf(st
, info
->uuid
);
2125 info
->array
.major_version
= -1;
2126 info
->array
.minor_version
= -2;
2127 sprintf(info
->text_version
, "/%s/%d",
2128 st
->container_devnm
,
2129 info
->container_member
);
2130 info
->safe_mode_delay
= DDF_SAFE_MODE_DELAY
;
2132 _ddf_array_name(info
->name
, ddf
, info
->container_member
);
2135 for (j
= 0; j
< map_disks
; j
++) {
2137 if (j
< info
->array
.raid_disks
) {
2138 int i
= find_phys(ddf
, vc
->conf
.phys_refnum
[j
]);
2140 (be16_to_cpu(ddf
->phys
->entries
[i
].state
)
2142 !(be16_to_cpu(ddf
->phys
->entries
[i
].state
)
2149 static int update_super_ddf(struct supertype
*st
, struct mdinfo
*info
,
2151 char *devname
, int verbose
,
2152 int uuid_set
, char *homehost
)
2154 /* For 'assemble' and 'force' we need to return non-zero if any
2155 * change was made. For others, the return value is ignored.
2156 * Update options are:
2157 * force-one : This device looks a bit old but needs to be included,
2158 * update age info appropriately.
2159 * assemble: clear any 'faulty' flag to allow this device to
2161 * force-array: Array is degraded but being forced, mark it clean
2162 * if that will be needed to assemble it.
2164 * newdev: not used ????
2165 * grow: Array has gained a new device - this is currently for
2167 * resync: mark as dirty so a resync will happen.
2168 * uuid: Change the uuid of the array to match what is given
2169 * homehost: update the recorded homehost
2170 * name: update the name - preserving the homehost
2171 * _reshape_progress: record new reshape_progress position.
2173 * Following are not relevant for this version:
2174 * sparc2.2 : update from old dodgey metadata
2175 * super-minor: change the preferred_minor number
2176 * summaries: update redundant counters.
2179 // struct ddf_super *ddf = st->sb;
2180 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
2181 // struct virtual_entry *ve = find_ve(ddf);
2183 /* we don't need to handle "force-*" or "assemble" as
2184 * there is no need to 'trick' the kernel. When the metadata is
2185 * first updated to activate the array, all the implied modifications
2189 if (strcmp(update
, "grow") == 0) {
2191 } else if (strcmp(update
, "resync") == 0) {
2192 // info->resync_checkpoint = 0;
2193 } else if (strcmp(update
, "homehost") == 0) {
2194 /* homehost is stored in controller->vendor_data,
2195 * or it is when we are the vendor
2197 // if (info->vendor_is_local)
2198 // strcpy(ddf->controller.vendor_data, homehost);
2200 } else if (strcmp(update
, "name") == 0) {
2201 /* name is stored in virtual_entry->name */
2202 // memset(ve->name, ' ', 16);
2203 // strncpy(ve->name, info->name, 16);
2205 } else if (strcmp(update
, "_reshape_progress") == 0) {
2206 /* We don't support reshape yet */
2207 } else if (strcmp(update
, "assemble") == 0 ) {
2208 /* Do nothing, just succeed */
2213 // update_all_csum(ddf);
2218 static void make_header_guid(char *guid
)
2221 /* Create a DDF Header of Virtual Disk GUID */
2223 /* 24 bytes of fiction required.
2224 * first 8 are a 'vendor-id' - "Linux-MD"
2225 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
2226 * Remaining 8 random number plus timestamp
2228 memcpy(guid
, T10
, sizeof(T10
));
2229 stamp
= cpu_to_be32(0xdeadbeef);
2230 memcpy(guid
+8, &stamp
, 4);
2231 stamp
= cpu_to_be32(0);
2232 memcpy(guid
+12, &stamp
, 4);
2233 stamp
= cpu_to_be32(time(0) - DECADE
);
2234 memcpy(guid
+16, &stamp
, 4);
2235 stamp
._v32
= random32();
2236 memcpy(guid
+20, &stamp
, 4);
2239 static unsigned int find_unused_vde(const struct ddf_super
*ddf
)
2242 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
2243 if (all_ff(ddf
->virt
->entries
[i
].guid
))
2246 return DDF_NOTFOUND
;
2249 static unsigned int find_vde_by_name(const struct ddf_super
*ddf
,
2254 return DDF_NOTFOUND
;
2255 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
2256 if (all_ff(ddf
->virt
->entries
[i
].guid
))
2258 if (!strncmp(name
, ddf
->virt
->entries
[i
].name
,
2259 sizeof(ddf
->virt
->entries
[i
].name
)))
2262 return DDF_NOTFOUND
;
2265 static unsigned int find_vde_by_guid(const struct ddf_super
*ddf
,
2269 if (guid
== NULL
|| all_ff(guid
))
2270 return DDF_NOTFOUND
;
2271 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++)
2272 if (!memcmp(ddf
->virt
->entries
[i
].guid
, guid
, DDF_GUID_LEN
))
2274 return DDF_NOTFOUND
;
2277 static int init_super_ddf(struct supertype
*st
,
2278 mdu_array_info_t
*info
,
2279 struct shape
*s
, char *name
, char *homehost
,
2280 int *uuid
, unsigned long long data_offset
)
2282 /* This is primarily called by Create when creating a new array.
2283 * We will then get add_to_super called for each component, and then
2284 * write_init_super called to write it out to each device.
2285 * For DDF, Create can create on fresh devices or on a pre-existing
2287 * To create on a pre-existing array a different method will be called.
2288 * This one is just for fresh drives.
2290 * We need to create the entire 'ddf' structure which includes:
2291 * DDF headers - these are easy.
2292 * Controller data - a Sector describing this controller .. not that
2293 * this is a controller exactly.
2294 * Physical Disk Record - one entry per device, so
2295 * leave plenty of space.
2296 * Virtual Disk Records - again, just leave plenty of space.
2297 * This just lists VDs, doesn't give details.
2298 * Config records - describe the VDs that use this disk
2299 * DiskData - describes 'this' device.
2300 * BadBlockManagement - empty
2301 * Diag Space - empty
2302 * Vendor Logs - Could we put bitmaps here?
2305 struct ddf_super
*ddf
;
2308 int max_phys_disks
, max_virt_disks
;
2309 unsigned long long sector
;
2313 struct phys_disk
*pd
;
2314 struct virtual_disk
*vd
;
2317 return init_super_ddf_bvd(st
, info
, s
->size
, name
, homehost
, uuid
,
2320 if (posix_memalign((void**)&ddf
, 512, sizeof(*ddf
)) != 0) {
2321 pr_err("could not allocate superblock\n");
2324 memset(ddf
, 0, sizeof(*ddf
));
2328 /* zeroing superblock */
2332 /* At least 32MB *must* be reserved for the ddf. So let's just
2333 * start 32MB from the end, and put the primary header there.
2334 * Don't do secondary for now.
2335 * We don't know exactly where that will be yet as it could be
2336 * different on each device. So just set up the lengths.
2339 ddf
->anchor
.magic
= DDF_HEADER_MAGIC
;
2340 make_header_guid(ddf
->anchor
.guid
);
2342 memcpy(ddf
->anchor
.revision
, DDF_REVISION_2
, 8);
2343 ddf
->anchor
.seq
= cpu_to_be32(1);
2344 ddf
->anchor
.timestamp
= cpu_to_be32(time(0) - DECADE
);
2345 ddf
->anchor
.openflag
= 0xFF;
2346 ddf
->anchor
.foreignflag
= 0;
2347 ddf
->anchor
.enforcegroups
= 0; /* Is this best?? */
2348 ddf
->anchor
.pad0
= 0xff;
2349 memset(ddf
->anchor
.pad1
, 0xff, 12);
2350 memset(ddf
->anchor
.header_ext
, 0xff, 32);
2351 ddf
->anchor
.primary_lba
= cpu_to_be64(~(__u64
)0);
2352 ddf
->anchor
.secondary_lba
= cpu_to_be64(~(__u64
)0);
2353 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
2354 memset(ddf
->anchor
.pad2
, 0xff, 3);
2355 ddf
->anchor
.workspace_len
= cpu_to_be32(32768); /* Must be reserved */
2356 /* Put this at bottom of 32M reserved.. */
2357 ddf
->anchor
.workspace_lba
= cpu_to_be64(~(__u64
)0);
2358 max_phys_disks
= 1023; /* Should be enough, 4095 is also allowed */
2359 ddf
->anchor
.max_pd_entries
= cpu_to_be16(max_phys_disks
);
2360 max_virt_disks
= 255; /* 15, 63, 255, 1024, 4095 are all allowed */
2361 ddf
->anchor
.max_vd_entries
= cpu_to_be16(max_virt_disks
);
2363 ddf
->anchor
.max_partitions
= cpu_to_be16(ddf
->max_part
);
2364 ddf
->mppe
= 256; /* 16, 64, 256, 1024, 4096 are all allowed */
2365 ddf
->conf_rec_len
= 1 + ROUND_UP(ddf
->mppe
* (4+8), 512)/512;
2366 ddf
->anchor
.config_record_len
= cpu_to_be16(ddf
->conf_rec_len
);
2367 ddf
->anchor
.max_primary_element_entries
= cpu_to_be16(ddf
->mppe
);
2368 memset(ddf
->anchor
.pad3
, 0xff, 54);
2369 /* Controller section is one sector long immediately
2370 * after the ddf header */
2372 ddf
->anchor
.controller_section_offset
= cpu_to_be32(sector
);
2373 ddf
->anchor
.controller_section_length
= cpu_to_be32(1);
2376 /* phys is 8 sectors after that */
2377 pdsize
= ROUND_UP(sizeof(struct phys_disk
) +
2378 sizeof(struct phys_disk_entry
)*max_phys_disks
,
2380 switch(pdsize
/512) {
2381 case 2: case 8: case 32: case 128: case 512: break;
2384 ddf
->anchor
.phys_section_offset
= cpu_to_be32(sector
);
2385 ddf
->anchor
.phys_section_length
=
2386 cpu_to_be32(pdsize
/512); /* max_primary_element_entries/8 */
2387 sector
+= pdsize
/512;
2389 /* virt is another 32 sectors */
2390 vdsize
= ROUND_UP(sizeof(struct virtual_disk
) +
2391 sizeof(struct virtual_entry
) * max_virt_disks
,
2393 switch(vdsize
/512) {
2394 case 2: case 8: case 32: case 128: case 512: break;
2397 ddf
->anchor
.virt_section_offset
= cpu_to_be32(sector
);
2398 ddf
->anchor
.virt_section_length
=
2399 cpu_to_be32(vdsize
/512); /* max_vd_entries/8 */
2400 sector
+= vdsize
/512;
2402 clen
= ddf
->conf_rec_len
* (ddf
->max_part
+1);
2403 ddf
->anchor
.config_section_offset
= cpu_to_be32(sector
);
2404 ddf
->anchor
.config_section_length
= cpu_to_be32(clen
);
2407 ddf
->anchor
.data_section_offset
= cpu_to_be32(sector
);
2408 ddf
->anchor
.data_section_length
= cpu_to_be32(1);
2411 ddf
->anchor
.bbm_section_length
= cpu_to_be32(0);
2412 ddf
->anchor
.bbm_section_offset
= cpu_to_be32(0xFFFFFFFF);
2413 ddf
->anchor
.diag_space_length
= cpu_to_be32(0);
2414 ddf
->anchor
.diag_space_offset
= cpu_to_be32(0xFFFFFFFF);
2415 ddf
->anchor
.vendor_length
= cpu_to_be32(0);
2416 ddf
->anchor
.vendor_offset
= cpu_to_be32(0xFFFFFFFF);
2418 memset(ddf
->anchor
.pad4
, 0xff, 256);
2420 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2421 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2423 ddf
->primary
.openflag
= 1; /* I guess.. */
2424 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
2426 ddf
->secondary
.openflag
= 1; /* I guess.. */
2427 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
2429 ddf
->active
= &ddf
->primary
;
2431 ddf
->controller
.magic
= DDF_CONTROLLER_MAGIC
;
2433 /* 24 more bytes of fiction required.
2434 * first 8 are a 'vendor-id' - "Linux-MD"
2435 * Remaining 16 are serial number.... maybe a hostname would do?
2437 memcpy(ddf
->controller
.guid
, T10
, sizeof(T10
));
2438 gethostname(hostname
, sizeof(hostname
));
2439 hostname
[sizeof(hostname
) - 1] = 0;
2440 hostlen
= strlen(hostname
);
2441 memcpy(ddf
->controller
.guid
+ 24 - hostlen
, hostname
, hostlen
);
2442 for (i
= strlen(T10
) ; i
+hostlen
< 24; i
++)
2443 ddf
->controller
.guid
[i
] = ' ';
2445 ddf
->controller
.type
.vendor_id
= cpu_to_be16(0xDEAD);
2446 ddf
->controller
.type
.device_id
= cpu_to_be16(0xBEEF);
2447 ddf
->controller
.type
.sub_vendor_id
= cpu_to_be16(0);
2448 ddf
->controller
.type
.sub_device_id
= cpu_to_be16(0);
2449 memcpy(ddf
->controller
.product_id
, "What Is My PID??", 16);
2450 memset(ddf
->controller
.pad
, 0xff, 8);
2451 memset(ddf
->controller
.vendor_data
, 0xff, 448);
2452 if (homehost
&& strlen(homehost
) < 440)
2453 strcpy((char*)ddf
->controller
.vendor_data
, homehost
);
2455 if (posix_memalign((void**)&pd
, 512, pdsize
) != 0) {
2456 pr_err("could not allocate pd\n");
2460 ddf
->pdsize
= pdsize
;
2462 memset(pd
, 0xff, pdsize
);
2463 memset(pd
, 0, sizeof(*pd
));
2464 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2465 pd
->used_pdes
= cpu_to_be16(0);
2466 pd
->max_pdes
= cpu_to_be16(max_phys_disks
);
2467 memset(pd
->pad
, 0xff, 52);
2468 for (i
= 0; i
< max_phys_disks
; i
++)
2469 memset(pd
->entries
[i
].guid
, 0xff, DDF_GUID_LEN
);
2471 if (posix_memalign((void**)&vd
, 512, vdsize
) != 0) {
2472 pr_err("could not allocate vd\n");
2476 ddf
->vdsize
= vdsize
;
2477 memset(vd
, 0, vdsize
);
2478 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
2479 vd
->populated_vdes
= cpu_to_be16(0);
2480 vd
->max_vdes
= cpu_to_be16(max_virt_disks
);
2481 memset(vd
->pad
, 0xff, 52);
2483 for (i
=0; i
<max_virt_disks
; i
++)
2484 memset(&vd
->entries
[i
], 0xff, sizeof(struct virtual_entry
));
2487 ddf_set_updates_pending(ddf
, NULL
);
2491 static int chunk_to_shift(int chunksize
)
2493 return ffs(chunksize
/512)-1;
2497 unsigned long long start
, size
;
2499 static int cmp_extent(const void *av
, const void *bv
)
2501 const struct extent
*a
= av
;
2502 const struct extent
*b
= bv
;
2503 if (a
->start
< b
->start
)
2505 if (a
->start
> b
->start
)
2510 static struct extent
*get_extents(struct ddf_super
*ddf
, struct dl
*dl
)
2512 /* Find a list of used extents on the given physical device
2513 * (dnum) of the given ddf.
2514 * Return a malloced array of 'struct extent'
2523 state
= be16_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].state
);
2525 if ((state
& (DDF_Online
|DDF_Failed
|DDF_Missing
)) != DDF_Online
)
2528 rv
= xmalloc(sizeof(struct extent
) * (ddf
->max_part
+ 2));
2530 for (i
= 0; i
< ddf
->max_part
; i
++) {
2531 const struct vd_config
*bvd
;
2533 struct vcl
*v
= dl
->vlist
[i
];
2535 get_pd_index_from_refnum(v
, dl
->disk
.refnum
, ddf
->mppe
,
2536 &bvd
, &ibvd
) == DDF_NOTFOUND
)
2538 rv
[n
].start
= be64_to_cpu(LBA_OFFSET(ddf
, bvd
)[ibvd
]);
2539 rv
[n
].size
= be64_to_cpu(bvd
->blocks
);
2542 qsort(rv
, n
, sizeof(*rv
), cmp_extent
);
2544 rv
[n
].start
= be64_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].config_size
);
2549 static unsigned long long find_space(
2550 struct ddf_super
*ddf
, struct dl
*dl
,
2551 unsigned long long data_offset
,
2552 unsigned long long *size
)
2554 /* Find if the requested amount of space is available.
2555 * If it is, return start.
2556 * If not, set *size to largest space.
2557 * If data_offset != INVALID_SECTORS, then the space must start
2560 struct extent
*e
= get_extents(ddf
, dl
);
2562 unsigned long long pos
= 0;
2563 unsigned long long max_size
= 0;
2567 return INVALID_SECTORS
;
2570 unsigned long long esize
= e
[i
].start
- pos
;
2571 if (data_offset
!= INVALID_SECTORS
&&
2572 pos
<= data_offset
&&
2573 e
[i
].start
> data_offset
) {
2575 esize
= e
[i
].start
- pos
;
2577 if (data_offset
!= INVALID_SECTORS
&&
2578 pos
!= data_offset
) {
2582 if (esize
>= *size
) {
2587 if (esize
> max_size
)
2589 pos
= e
[i
].start
+ e
[i
].size
;
2591 } while (e
[i
-1].size
);
2594 return INVALID_SECTORS
;
2597 static int init_super_ddf_bvd(struct supertype
*st
,
2598 mdu_array_info_t
*info
,
2599 unsigned long long size
,
2600 char *name
, char *homehost
,
2601 int *uuid
, unsigned long long data_offset
)
2603 /* We are creating a BVD inside a pre-existing container.
2604 * so st->sb is already set.
2605 * We need to create a new vd_config and a new virtual_entry
2607 struct ddf_super
*ddf
= st
->sb
;
2608 unsigned int venum
, i
;
2609 struct virtual_entry
*ve
;
2611 struct vd_config
*vc
;
2613 if (find_vde_by_name(ddf
, name
) != DDF_NOTFOUND
) {
2614 pr_err("This ddf already has an array called %s\n", name
);
2617 venum
= find_unused_vde(ddf
);
2618 if (venum
== DDF_NOTFOUND
) {
2619 pr_err("Cannot find spare slot for virtual disk\n");
2622 ve
= &ddf
->virt
->entries
[venum
];
2624 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
2625 * timestamp, random number
2627 make_header_guid(ve
->guid
);
2628 ve
->unit
= cpu_to_be16(info
->md_minor
);
2630 ve
->guid_crc
._v16
= crc32(0, (unsigned char *)ddf
->anchor
.guid
,
2632 ve
->type
= cpu_to_be16(0);
2633 ve
->state
= DDF_state_degraded
; /* Will be modified as devices are added */
2634 if (info
->state
& 1) /* clean */
2635 ve
->init_state
= DDF_init_full
;
2637 ve
->init_state
= DDF_init_not
;
2639 memset(ve
->pad1
, 0xff, 14);
2640 memset(ve
->name
, ' ', 16);
2642 strncpy(ve
->name
, name
, 16);
2643 ddf
->virt
->populated_vdes
=
2644 cpu_to_be16(be16_to_cpu(ddf
->virt
->populated_vdes
)+1);
2646 /* Now create a new vd_config */
2647 if (posix_memalign((void**)&vcl
, 512,
2648 (offsetof(struct vcl
, conf
) + ddf
->conf_rec_len
* 512)) != 0) {
2649 pr_err("could not allocate vd_config\n");
2653 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
2656 vc
->magic
= DDF_VD_CONF_MAGIC
;
2657 memcpy(vc
->guid
, ve
->guid
, DDF_GUID_LEN
);
2658 vc
->timestamp
= cpu_to_be32(time(0)-DECADE
);
2659 vc
->seqnum
= cpu_to_be32(1);
2660 memset(vc
->pad0
, 0xff, 24);
2661 vc
->chunk_shift
= chunk_to_shift(info
->chunk_size
);
2662 if (layout_md2ddf(info
, vc
) == -1 ||
2663 be16_to_cpu(vc
->prim_elmnt_count
) > ddf
->mppe
) {
2664 pr_err("unsupported RAID level/layout %d/%d with %d disks\n",
2665 info
->level
, info
->layout
, info
->raid_disks
);
2669 vc
->sec_elmnt_seq
= 0;
2670 if (alloc_other_bvds(ddf
, vcl
) != 0) {
2671 pr_err("could not allocate other bvds\n");
2675 vc
->blocks
= cpu_to_be64(size
* 2);
2676 vc
->array_blocks
= cpu_to_be64(
2677 calc_array_size(info
->level
, info
->raid_disks
, info
->layout
,
2678 info
->chunk_size
, size
* 2));
2679 memset(vc
->pad1
, 0xff, 8);
2680 vc
->spare_refs
[0] = cpu_to_be32(0xffffffff);
2681 vc
->spare_refs
[1] = cpu_to_be32(0xffffffff);
2682 vc
->spare_refs
[2] = cpu_to_be32(0xffffffff);
2683 vc
->spare_refs
[3] = cpu_to_be32(0xffffffff);
2684 vc
->spare_refs
[4] = cpu_to_be32(0xffffffff);
2685 vc
->spare_refs
[5] = cpu_to_be32(0xffffffff);
2686 vc
->spare_refs
[6] = cpu_to_be32(0xffffffff);
2687 vc
->spare_refs
[7] = cpu_to_be32(0xffffffff);
2688 memset(vc
->cache_pol
, 0, 8);
2690 memset(vc
->pad2
, 0xff, 3);
2691 memset(vc
->pad3
, 0xff, 52);
2692 memset(vc
->pad4
, 0xff, 192);
2693 memset(vc
->v0
, 0xff, 32);
2694 memset(vc
->v1
, 0xff, 32);
2695 memset(vc
->v2
, 0xff, 16);
2696 memset(vc
->v3
, 0xff, 16);
2697 memset(vc
->vendor
, 0xff, 32);
2699 memset(vc
->phys_refnum
, 0xff, 4*ddf
->mppe
);
2700 memset(vc
->phys_refnum
+ddf
->mppe
, 0x00, 8*ddf
->mppe
);
2702 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++) {
2703 memcpy(vcl
->other_bvds
[i
-1], vc
, ddf
->conf_rec_len
* 512);
2704 vcl
->other_bvds
[i
-1]->sec_elmnt_seq
= i
;
2707 vcl
->next
= ddf
->conflist
;
2708 ddf
->conflist
= vcl
;
2709 ddf
->currentconf
= vcl
;
2710 ddf_set_updates_pending(ddf
, NULL
);
2714 static void add_to_super_ddf_bvd(struct supertype
*st
,
2715 mdu_disk_info_t
*dk
, int fd
, char *devname
,
2716 unsigned long long data_offset
)
2718 /* fd and devname identify a device within the ddf container (st).
2719 * dk identifies a location in the new BVD.
2720 * We need to find suitable free space in that device and update
2721 * the phys_refnum and lba_offset for the newly created vd_config.
2722 * We might also want to update the type in the phys_disk
2725 * Alternately: fd == -1 and we have already chosen which device to
2726 * use and recorded in dlist->raid_disk;
2729 struct ddf_super
*ddf
= st
->sb
;
2730 struct vd_config
*vc
;
2732 unsigned long long blocks
, pos
;
2733 unsigned int raid_disk
= dk
->raid_disk
;
2736 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2737 if (dl
->raiddisk
== dk
->raid_disk
)
2740 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2741 if (dl
->major
== dk
->major
&&
2742 dl
->minor
== dk
->minor
)
2745 if (!dl
|| dl
->pdnum
< 0 || ! (dk
->state
& (1<<MD_DISK_SYNC
)))
2748 vc
= &ddf
->currentconf
->conf
;
2749 if (vc
->sec_elmnt_count
> 1) {
2750 unsigned int n
= be16_to_cpu(vc
->prim_elmnt_count
);
2752 vc
= ddf
->currentconf
->other_bvds
[raid_disk
/ n
- 1];
2756 blocks
= be64_to_cpu(vc
->blocks
);
2757 if (ddf
->currentconf
->block_sizes
)
2758 blocks
= ddf
->currentconf
->block_sizes
[dk
->raid_disk
];
2760 pos
= find_space(ddf
, dl
, data_offset
, &blocks
);
2761 if (pos
== INVALID_SECTORS
)
2764 ddf
->currentdev
= dk
->raid_disk
;
2765 vc
->phys_refnum
[raid_disk
] = dl
->disk
.refnum
;
2766 LBA_OFFSET(ddf
, vc
)[raid_disk
] = cpu_to_be64(pos
);
2768 for (i
= 0; i
< ddf
->max_part
; i
++)
2769 if (dl
->vlist
[i
] == NULL
)
2771 if (i
== ddf
->max_part
)
2773 dl
->vlist
[i
] = ddf
->currentconf
;
2778 dl
->devname
= devname
;
2780 /* Check if we can mark array as optimal yet */
2781 i
= ddf
->currentconf
->vcnum
;
2782 ddf
->virt
->entries
[i
].state
=
2783 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
2784 | get_svd_state(ddf
, ddf
->currentconf
);
2785 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
2786 cpu_to_be16(DDF_Global_Spare
));
2787 be16_set(ddf
->phys
->entries
[dl
->pdnum
].type
,
2788 cpu_to_be16(DDF_Active_in_VD
));
2789 dprintf("added disk %d/%08x to VD %d/%s as disk %d\n",
2790 dl
->pdnum
, be32_to_cpu(dl
->disk
.refnum
),
2791 ddf
->currentconf
->vcnum
, guid_str(vc
->guid
),
2793 ddf_set_updates_pending(ddf
, vc
);
2796 static unsigned int find_unused_pde(const struct ddf_super
*ddf
)
2799 for (i
= 0; i
< be16_to_cpu(ddf
->phys
->max_pdes
); i
++) {
2800 if (all_ff(ddf
->phys
->entries
[i
].guid
))
2803 return DDF_NOTFOUND
;
2806 static void _set_config_size(struct phys_disk_entry
*pde
, const struct dl
*dl
)
2809 cfs
= min(dl
->size
- 32*1024*2ULL, be64_to_cpu(dl
->primary_lba
));
2810 t
= be64_to_cpu(dl
->secondary_lba
);
2814 * Some vendor DDF structures interpret workspace_lba
2815 * very differently than we do: Make a sanity check on the value.
2817 t
= be64_to_cpu(dl
->workspace_lba
);
2819 __u64 wsp
= cfs
- t
;
2820 if (wsp
> 1024*1024*2ULL && wsp
> dl
->size
/ 16) {
2821 pr_err("%x:%x: workspace size 0x%llx too big, ignoring\n",
2822 dl
->major
, dl
->minor
, (unsigned long long)wsp
);
2826 pde
->config_size
= cpu_to_be64(cfs
);
2827 dprintf("%x:%x config_size %llx, DDF structure is %llx blocks\n",
2828 dl
->major
, dl
->minor
,
2829 (unsigned long long)cfs
, (unsigned long long)(dl
->size
-cfs
));
2832 /* Add a device to a container, either while creating it or while
2833 * expanding a pre-existing container
2835 static int add_to_super_ddf(struct supertype
*st
,
2836 mdu_disk_info_t
*dk
, int fd
, char *devname
,
2837 unsigned long long data_offset
)
2839 struct ddf_super
*ddf
= st
->sb
;
2843 unsigned long long size
;
2844 struct phys_disk_entry
*pde
;
2849 if (ddf
->currentconf
) {
2850 add_to_super_ddf_bvd(st
, dk
, fd
, devname
, data_offset
);
2854 /* This is device numbered dk->number. We need to create
2855 * a phys_disk entry and a more detailed disk_data entry.
2858 n
= find_unused_pde(ddf
);
2859 if (n
== DDF_NOTFOUND
) {
2860 pr_err("No free slot in array, cannot add disk\n");
2863 pde
= &ddf
->phys
->entries
[n
];
2864 get_dev_size(fd
, NULL
, &size
);
2865 if (size
<= 32*1024*1024) {
2866 pr_err("device size must be at least 32MB\n");
2871 if (posix_memalign((void**)&dd
, 512,
2872 sizeof(*dd
) + sizeof(dd
->vlist
[0]) * ddf
->max_part
) != 0) {
2873 pr_err("could allocate buffer for new disk, aborting\n");
2876 dd
->major
= major(stb
.st_rdev
);
2877 dd
->minor
= minor(stb
.st_rdev
);
2878 dd
->devname
= devname
;
2882 dd
->disk
.magic
= DDF_PHYS_DATA_MAGIC
;
2884 tm
= localtime(&now
);
2885 sprintf(dd
->disk
.guid
, "%8s%04d%02d%02d", T10
,
2886 (__u16
)tm
->tm_year
+1900,
2887 (__u8
)tm
->tm_mon
+1, (__u8
)tm
->tm_mday
);
2888 tptr
= (__u32
*)(dd
->disk
.guid
+ 16);
2889 *tptr
++ = random32();
2893 /* Cannot be bothered finding a CRC of some irrelevant details*/
2894 dd
->disk
.refnum
._v32
= random32();
2895 for (i
= be16_to_cpu(ddf
->active
->max_pd_entries
);
2897 if (be32_eq(ddf
->phys
->entries
[i
-1].refnum
,
2902 dd
->disk
.forced_ref
= 1;
2903 dd
->disk
.forced_guid
= 1;
2904 memset(dd
->disk
.vendor
, ' ', 32);
2905 memcpy(dd
->disk
.vendor
, "Linux", 5);
2906 memset(dd
->disk
.pad
, 0xff, 442);
2907 for (i
= 0; i
< ddf
->max_part
; i
++)
2908 dd
->vlist
[i
] = NULL
;
2912 if (st
->update_tail
) {
2913 int len
= (sizeof(struct phys_disk
) +
2914 sizeof(struct phys_disk_entry
));
2915 struct phys_disk
*pd
;
2918 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2919 pd
->used_pdes
= cpu_to_be16(n
);
2920 pde
= &pd
->entries
[0];
2923 ddf
->phys
->used_pdes
= cpu_to_be16(
2924 1 + be16_to_cpu(ddf
->phys
->used_pdes
));
2926 memcpy(pde
->guid
, dd
->disk
.guid
, DDF_GUID_LEN
);
2927 pde
->refnum
= dd
->disk
.refnum
;
2928 pde
->type
= cpu_to_be16(DDF_Forced_PD_GUID
| DDF_Global_Spare
);
2929 pde
->state
= cpu_to_be16(DDF_Online
);
2932 * If there is already a device in dlist, try to reserve the same
2933 * amount of workspace. Otherwise, use 32MB.
2934 * We checked disk size above already.
2936 #define __calc_lba(new, old, lba, mb) do { \
2937 unsigned long long dif; \
2938 if ((old) != NULL) \
2939 dif = (old)->size - be64_to_cpu((old)->lba); \
2941 dif = (new)->size; \
2942 if ((new)->size > dif) \
2943 (new)->lba = cpu_to_be64((new)->size - dif); \
2945 (new)->lba = cpu_to_be64((new)->size - (mb*1024*2)); \
2947 __calc_lba(dd
, ddf
->dlist
, workspace_lba
, 32);
2948 __calc_lba(dd
, ddf
->dlist
, primary_lba
, 16);
2949 if (ddf
->dlist
== NULL
||
2950 be64_to_cpu(ddf
->dlist
->secondary_lba
) != ~(__u64
)0)
2951 __calc_lba(dd
, ddf
->dlist
, secondary_lba
, 32);
2952 _set_config_size(pde
, dd
);
2954 sprintf(pde
->path
, "%17.17s","Information: nil") ;
2955 memset(pde
->pad
, 0xff, 6);
2957 if (st
->update_tail
) {
2958 dd
->next
= ddf
->add_list
;
2961 dd
->next
= ddf
->dlist
;
2963 ddf_set_updates_pending(ddf
, NULL
);
2969 static int remove_from_super_ddf(struct supertype
*st
, mdu_disk_info_t
*dk
)
2971 struct ddf_super
*ddf
= st
->sb
;
2974 /* mdmon has noticed that this disk (dk->major/dk->minor) has
2975 * disappeared from the container.
2976 * We need to arrange that it disappears from the metadata and
2977 * internal data structures too.
2978 * Most of the work is done by ddf_process_update which edits
2979 * the metadata and closes the file handle and attaches the memory
2980 * where free_updates will free it.
2982 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2983 if (dl
->major
== dk
->major
&&
2984 dl
->minor
== dk
->minor
)
2986 if (!dl
|| dl
->pdnum
< 0)
2989 if (st
->update_tail
) {
2990 int len
= (sizeof(struct phys_disk
) +
2991 sizeof(struct phys_disk_entry
));
2992 struct phys_disk
*pd
;
2995 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2996 pd
->used_pdes
= cpu_to_be16(dl
->pdnum
);
2997 pd
->entries
[0].state
= cpu_to_be16(DDF_Missing
);
2998 append_metadata_update(st
, pd
, len
);
3004 * This is the write_init_super method for a ddf container. It is
3005 * called when creating a container or adding another device to a
3009 static int __write_ddf_structure(struct dl
*d
, struct ddf_super
*ddf
, __u8 type
)
3011 unsigned long long sector
;
3012 struct ddf_header
*header
;
3013 int fd
, i
, n_config
, conf_size
, buf_size
;
3020 case DDF_HEADER_PRIMARY
:
3021 header
= &ddf
->primary
;
3022 sector
= be64_to_cpu(header
->primary_lba
);
3024 case DDF_HEADER_SECONDARY
:
3025 header
= &ddf
->secondary
;
3026 sector
= be64_to_cpu(header
->secondary_lba
);
3031 if (sector
== ~(__u64
)0)
3034 header
->type
= type
;
3035 header
->openflag
= 1;
3036 header
->crc
= calc_crc(header
, 512);
3038 lseek64(fd
, sector
<<9, 0);
3039 if (write(fd
, header
, 512) < 0)
3042 ddf
->controller
.crc
= calc_crc(&ddf
->controller
, 512);
3043 if (write(fd
, &ddf
->controller
, 512) < 0)
3046 ddf
->phys
->crc
= calc_crc(ddf
->phys
, ddf
->pdsize
);
3047 if (write(fd
, ddf
->phys
, ddf
->pdsize
) < 0)
3049 ddf
->virt
->crc
= calc_crc(ddf
->virt
, ddf
->vdsize
);
3050 if (write(fd
, ddf
->virt
, ddf
->vdsize
) < 0)
3053 /* Now write lots of config records. */
3054 n_config
= ddf
->max_part
;
3055 conf_size
= ddf
->conf_rec_len
* 512;
3057 buf_size
= conf_size
* (n_config
+ 1);
3059 if (posix_memalign((void**)&conf
, 512, buf_size
) != 0)
3063 for (i
= 0 ; i
<= n_config
; i
++) {
3065 struct vd_config
*vdc
= NULL
;
3066 if (i
== n_config
) {
3067 c
= (struct vcl
*)d
->spare
;
3074 get_pd_index_from_refnum(
3077 (const struct vd_config
**)&vdc
,
3081 dprintf("writing conf record %i on disk %08x for %s/%u\n",
3082 i
, be32_to_cpu(d
->disk
.refnum
),
3083 guid_str(vdc
->guid
),
3084 vdc
->sec_elmnt_seq
);
3085 vdc
->crc
= calc_crc(vdc
, conf_size
);
3086 memcpy(conf
+ i
*conf_size
, vdc
, conf_size
);
3088 memset(conf
+ i
*conf_size
, 0xff, conf_size
);
3090 if (write(fd
, conf
, buf_size
) != buf_size
)
3093 d
->disk
.crc
= calc_crc(&d
->disk
, 512);
3094 if (write(fd
, &d
->disk
, 512) < 0)
3099 header
->openflag
= 0;
3100 header
->crc
= calc_crc(header
, 512);
3102 lseek64(fd
, sector
<<9, 0);
3103 if (write(fd
, header
, 512) < 0)
3109 static int _write_super_to_disk(struct ddf_super
*ddf
, struct dl
*d
)
3111 unsigned long long size
;
3116 /* We need to fill in the primary, (secondary) and workspace
3117 * lba's in the headers, set their checksums,
3118 * Also checksum phys, virt....
3120 * Then write everything out, finally the anchor is written.
3122 get_dev_size(fd
, NULL
, &size
);
3124 memcpy(&ddf
->anchor
, ddf
->active
, 512);
3125 if (be64_to_cpu(d
->workspace_lba
) != 0ULL)
3126 ddf
->anchor
.workspace_lba
= d
->workspace_lba
;
3128 ddf
->anchor
.workspace_lba
=
3129 cpu_to_be64(size
- 32*1024*2);
3130 if (be64_to_cpu(d
->primary_lba
) != 0ULL)
3131 ddf
->anchor
.primary_lba
= d
->primary_lba
;
3133 ddf
->anchor
.primary_lba
=
3134 cpu_to_be64(size
- 16*1024*2);
3135 if (be64_to_cpu(d
->secondary_lba
) != 0ULL)
3136 ddf
->anchor
.secondary_lba
= d
->secondary_lba
;
3138 ddf
->anchor
.secondary_lba
=
3139 cpu_to_be64(size
- 32*1024*2);
3140 ddf
->anchor
.timestamp
= cpu_to_be32(time(0) - DECADE
);
3141 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
3142 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
3144 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
3145 ddf
->anchor
.openflag
= 0xFF; /* 'open' means nothing */
3146 ddf
->anchor
.seq
= cpu_to_be32(0xFFFFFFFF); /* no sequencing in anchor */
3147 ddf
->anchor
.crc
= calc_crc(&ddf
->anchor
, 512);
3149 if (!__write_ddf_structure(d
, ddf
, DDF_HEADER_PRIMARY
))
3152 if (!__write_ddf_structure(d
, ddf
, DDF_HEADER_SECONDARY
))
3155 lseek64(fd
, (size
-1)*512, SEEK_SET
);
3156 if (write(fd
, &ddf
->anchor
, 512) < 0)
3162 static int __write_init_super_ddf(struct supertype
*st
)
3164 struct ddf_super
*ddf
= st
->sb
;
3169 pr_state(ddf
, __func__
);
3171 /* try to write updated metadata,
3172 * if we catch a failure move on to the next disk
3174 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
3176 successes
+= _write_super_to_disk(ddf
, d
);
3179 return attempts
!= successes
;
3182 static int write_init_super_ddf(struct supertype
*st
)
3184 struct ddf_super
*ddf
= st
->sb
;
3185 struct vcl
*currentconf
= ddf
->currentconf
;
3187 /* We are done with currentconf - reset it so st refers to the container */
3188 ddf
->currentconf
= NULL
;
3190 if (st
->update_tail
) {
3191 /* queue the virtual_disk and vd_config as metadata updates */
3192 struct virtual_disk
*vd
;
3193 struct vd_config
*vc
;
3198 /* Must be adding a physical disk to the container */
3199 int len
= (sizeof(struct phys_disk
) +
3200 sizeof(struct phys_disk_entry
));
3202 /* adding a disk to the container. */
3206 append_metadata_update(st
, ddf
->add_list
->mdupdate
, len
);
3207 ddf
->add_list
->mdupdate
= NULL
;
3211 /* Newly created VD */
3213 /* First the virtual disk. We have a slightly fake header */
3214 len
= sizeof(struct virtual_disk
) + sizeof(struct virtual_entry
);
3217 vd
->entries
[0] = ddf
->virt
->entries
[currentconf
->vcnum
];
3218 vd
->populated_vdes
= cpu_to_be16(currentconf
->vcnum
);
3219 append_metadata_update(st
, vd
, len
);
3221 /* Then the vd_config */
3222 len
= ddf
->conf_rec_len
* 512;
3223 tlen
= len
* currentconf
->conf
.sec_elmnt_count
;
3225 memcpy(vc
, ¤tconf
->conf
, len
);
3226 for (i
= 1; i
< currentconf
->conf
.sec_elmnt_count
; i
++)
3227 memcpy((char *)vc
+ i
*len
, currentconf
->other_bvds
[i
-1],
3229 append_metadata_update(st
, vc
, tlen
);
3235 for (d
= ddf
->dlist
; d
; d
=d
->next
)
3236 while (Kill(d
->devname
, NULL
, 0, -1, 1) == 0);
3237 /* Note: we don't close the fd's now, but a subsequent
3238 * ->free_super() will
3240 return __write_init_super_ddf(st
);
3244 static __u64
avail_size_ddf(struct supertype
*st
, __u64 devsize
,
3245 unsigned long long data_offset
)
3247 /* We must reserve the last 32Meg */
3248 if (devsize
<= 32*1024*2)
3250 return devsize
- 32*1024*2;
3253 static int reserve_space(struct supertype
*st
, int raiddisks
,
3254 unsigned long long size
, int chunk
,
3255 unsigned long long data_offset
,
3256 unsigned long long *freesize
)
3258 /* Find 'raiddisks' spare extents at least 'size' big (but
3259 * only caring about multiples of 'chunk') and remember
3260 * them. If size==0, find the largest size possible.
3261 * Report available size in *freesize
3262 * If space cannot be found, fail.
3265 struct ddf_super
*ddf
= st
->sb
;
3268 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3272 /* Now find largest extent on each device */
3273 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3274 unsigned long long minsize
= ULLONG_MAX
;
3276 find_space(ddf
, dl
, data_offset
, &minsize
);
3277 if (minsize
>= size
&& minsize
>= (unsigned)chunk
) {
3279 dl
->esize
= minsize
;
3282 if (cnt
< raiddisks
) {
3283 pr_err("not enough devices with space to create array.\n");
3284 return 0; /* No enough free spaces large enough */
3287 /* choose the largest size of which there are at least 'raiddisk' */
3288 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3290 if (dl
->esize
<= size
)
3292 /* This is bigger than 'size', see if there are enough */
3294 for (dl2
= ddf
->dlist
; dl2
; dl2
=dl2
->next
)
3295 if (dl2
->esize
>= dl
->esize
)
3297 if (cnt
>= raiddisks
)
3301 size
= size
/ chunk
;
3306 pr_err("not enough spare devices to create array.\n");
3310 /* We have a 'size' of which there are enough spaces.
3311 * We simply do a first-fit */
3313 for (dl
= ddf
->dlist
; dl
&& cnt
< raiddisks
; dl
=dl
->next
) {
3314 if (dl
->esize
< size
)
3323 static int validate_geometry_ddf(struct supertype
*st
,
3324 int level
, int layout
, int raiddisks
,
3325 int *chunk
, unsigned long long size
,
3326 unsigned long long data_offset
,
3327 char *dev
, unsigned long long *freesize
,
3328 int consistency_policy
, int verbose
)
3334 /* ddf potentially supports lots of things, but it depends on
3335 * what devices are offered (and maybe kernel version?)
3336 * If given unused devices, we will make a container.
3337 * If given devices in a container, we will make a BVD.
3338 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
3341 if (*chunk
== UnSet
)
3342 *chunk
= DEFAULT_CHUNK
;
3344 if (level
== LEVEL_NONE
)
3345 level
= LEVEL_CONTAINER
;
3346 if (level
== LEVEL_CONTAINER
) {
3347 /* Must be a fresh device to add to a container */
3348 return validate_geometry_ddf_container(st
, level
, layout
,
3350 size
, data_offset
, dev
,
3356 mdu_array_info_t array
= {
3359 .raid_disks
= raiddisks
3361 struct vd_config conf
;
3362 if (layout_md2ddf(&array
, &conf
) == -1) {
3364 pr_err("DDF does not support level %d /layout %d arrays with %d disks\n",
3365 level
, layout
, raiddisks
);
3368 /* Should check layout? etc */
3370 if (st
->sb
&& freesize
) {
3371 /* --create was given a container to create in.
3372 * So we need to check that there are enough
3373 * free spaces and return the amount of space.
3374 * We may as well remember which drives were
3375 * chosen so that add_to_super/getinfo_super
3378 return reserve_space(st
, raiddisks
, size
, *chunk
,
3379 data_offset
, freesize
);
3385 /* A container has already been opened, so we are
3386 * creating in there. Maybe a BVD, maybe an SVD.
3387 * Should make a distinction one day.
3389 return validate_geometry_ddf_bvd(st
, level
, layout
, raiddisks
,
3390 chunk
, size
, data_offset
, dev
,
3394 /* This is the first device for the array.
3395 * If it is a container, we read it in and do automagic allocations,
3396 * no other devices should be given.
3397 * Otherwise it must be a member device of a container, and we
3398 * do manual allocation.
3399 * Later we should check for a BVD and make an SVD.
3401 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3404 /* Just a bare device, no good to us */
3406 pr_err("ddf: Cannot create this array on device %s - a container is required.\n",
3410 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
3412 pr_err("ddf: Cannot open %s: %s\n",
3413 dev
, strerror(errno
));
3416 /* Well, it is in use by someone, maybe a 'ddf' container. */
3417 cfd
= open_container(fd
);
3421 pr_err("ddf: Cannot use %s: %s\n",
3422 dev
, strerror(EBUSY
));
3425 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
3427 if (sra
&& sra
->array
.major_version
== -1 &&
3428 strcmp(sra
->text_version
, "ddf") == 0) {
3429 /* This is a member of a ddf container. Load the container
3430 * and try to create a bvd
3432 struct ddf_super
*ddf
;
3433 if (load_super_ddf_all(st
, cfd
, (void **)&ddf
, NULL
) == 0) {
3435 strcpy(st
->container_devnm
, fd2devnm(cfd
));
3437 return validate_geometry_ddf_bvd(st
, level
, layout
,
3438 raiddisks
, chunk
, size
,
3444 } else /* device may belong to a different container */
3451 validate_geometry_ddf_container(struct supertype
*st
,
3452 int level
, int layout
, int raiddisks
,
3453 int chunk
, unsigned long long size
,
3454 unsigned long long data_offset
,
3455 char *dev
, unsigned long long *freesize
,
3459 unsigned long long ldsize
;
3461 if (level
!= LEVEL_CONTAINER
)
3466 fd
= dev_open(dev
, O_RDONLY
|O_EXCL
);
3469 pr_err("ddf: Cannot open %s: %s\n",
3470 dev
, strerror(errno
));
3473 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3479 *freesize
= avail_size_ddf(st
, ldsize
>> 9, INVALID_SECTORS
);
3487 static int validate_geometry_ddf_bvd(struct supertype
*st
,
3488 int level
, int layout
, int raiddisks
,
3489 int *chunk
, unsigned long long size
,
3490 unsigned long long data_offset
,
3491 char *dev
, unsigned long long *freesize
,
3495 struct ddf_super
*ddf
= st
->sb
;
3497 unsigned long long maxsize
;
3498 /* ddf/bvd supports lots of things, but not containers */
3499 if (level
== LEVEL_CONTAINER
) {
3501 pr_err("DDF cannot create a container within an container\n");
3504 /* We must have the container info already read in. */
3509 /* General test: make sure there is space for
3510 * 'raiddisks' device extents of size 'size'.
3512 unsigned long long minsize
= size
;
3516 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
3517 if (find_space(ddf
, dl
, data_offset
, &minsize
) !=
3521 if (dcnt
< raiddisks
) {
3523 pr_err("ddf: Not enough devices with space for this array (%d < %d)\n",
3529 /* This device must be a member of the set */
3530 if (!stat_is_blkdev(dev
, &rdev
))
3532 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
3533 if (dl
->major
== (int)major(rdev
) &&
3534 dl
->minor
== (int)minor(rdev
))
3539 pr_err("ddf: %s is not in the same DDF set\n",
3543 maxsize
= ULLONG_MAX
;
3544 find_space(ddf
, dl
, data_offset
, &maxsize
);
3545 *freesize
= maxsize
;
3550 static int load_super_ddf_all(struct supertype
*st
, int fd
,
3551 void **sbp
, char *devname
)
3554 struct ddf_super
*super
;
3555 struct mdinfo
*sd
, *best
= NULL
;
3561 sra
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3564 if (sra
->array
.major_version
!= -1 ||
3565 sra
->array
.minor_version
!= -2 ||
3566 strcmp(sra
->text_version
, "ddf") != 0)
3569 if (posix_memalign((void**)&super
, 512, sizeof(*super
)) != 0)
3571 memset(super
, 0, sizeof(*super
));
3573 /* first, try each device, and choose the best ddf */
3574 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3576 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3577 dfd
= dev_open(nm
, O_RDONLY
);
3580 rv
= load_ddf_headers(dfd
, super
, NULL
);
3583 seq
= be32_to_cpu(super
->active
->seq
);
3584 if (super
->active
->openflag
)
3586 if (!best
|| seq
> bestseq
) {
3594 /* OK, load this ddf */
3595 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
3596 dfd
= dev_open(nm
, O_RDONLY
);
3599 load_ddf_headers(dfd
, super
, NULL
);
3600 load_ddf_global(dfd
, super
, NULL
);
3602 /* Now we need the device-local bits */
3603 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3606 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3607 dfd
= dev_open(nm
, O_RDWR
);
3610 rv
= load_ddf_headers(dfd
, super
, NULL
);
3612 rv
= load_ddf_local(dfd
, super
, NULL
, 1);
3618 if (st
->ss
== NULL
) {
3619 st
->ss
= &super_ddf
;
3620 st
->minor_version
= 0;
3623 strcpy(st
->container_devnm
, fd2devnm(fd
));
3627 static int load_container_ddf(struct supertype
*st
, int fd
,
3630 return load_super_ddf_all(st
, fd
, &st
->sb
, devname
);
3633 static int check_secondary(const struct vcl
*vc
)
3635 const struct vd_config
*conf
= &vc
->conf
;
3638 /* The only DDF secondary RAID level md can support is
3639 * RAID 10, if the stripe sizes and Basic volume sizes
3641 * Other configurations could in theory be supported by exposing
3642 * the BVDs to user space and using device mapper for the secondary
3643 * mapping. So far we don't support that.
3646 __u64 sec_elements
[4] = {0, 0, 0, 0};
3647 #define __set_sec_seen(n) (sec_elements[(n)>>6] |= (1<<((n)&63)))
3648 #define __was_sec_seen(n) ((sec_elements[(n)>>6] & (1<<((n)&63))) != 0)
3650 if (vc
->other_bvds
== NULL
) {
3651 pr_err("No BVDs for secondary RAID found\n");
3654 if (conf
->prl
!= DDF_RAID1
) {
3655 pr_err("Secondary RAID level only supported for mirrored BVD\n");
3658 if (conf
->srl
!= DDF_2STRIPED
&& conf
->srl
!= DDF_2SPANNED
) {
3659 pr_err("Secondary RAID level %d is unsupported\n",
3663 __set_sec_seen(conf
->sec_elmnt_seq
);
3664 for (i
= 0; i
< conf
->sec_elmnt_count
-1; i
++) {
3665 const struct vd_config
*bvd
= vc
->other_bvds
[i
];
3666 if (bvd
->sec_elmnt_seq
== DDF_UNUSED_BVD
)
3668 if (bvd
->srl
!= conf
->srl
) {
3669 pr_err("Inconsistent secondary RAID level across BVDs\n");
3672 if (bvd
->prl
!= conf
->prl
) {
3673 pr_err("Different RAID levels for BVDs are unsupported\n");
3676 if (!be16_eq(bvd
->prim_elmnt_count
, conf
->prim_elmnt_count
)) {
3677 pr_err("All BVDs must have the same number of primary elements\n");
3680 if (bvd
->chunk_shift
!= conf
->chunk_shift
) {
3681 pr_err("Different strip sizes for BVDs are unsupported\n");
3684 if (!be64_eq(bvd
->array_blocks
, conf
->array_blocks
)) {
3685 pr_err("Different BVD sizes are unsupported\n");
3688 __set_sec_seen(bvd
->sec_elmnt_seq
);
3690 for (i
= 0; i
< conf
->sec_elmnt_count
; i
++) {
3691 if (!__was_sec_seen(i
)) {
3692 /* pr_err("BVD %d is missing\n", i); */
3699 static unsigned int get_pd_index_from_refnum(const struct vcl
*vc
,
3700 be32 refnum
, unsigned int nmax
,
3701 const struct vd_config
**bvd
,
3704 unsigned int i
, j
, n
, sec
, cnt
;
3706 cnt
= be16_to_cpu(vc
->conf
.prim_elmnt_count
);
3707 sec
= (vc
->conf
.sec_elmnt_count
== 1 ? 0 : vc
->conf
.sec_elmnt_seq
);
3709 for (i
= 0, j
= 0 ; i
< nmax
; i
++) {
3710 /* j counts valid entries for this BVD */
3711 if (be32_eq(vc
->conf
.phys_refnum
[i
], refnum
)) {
3714 return sec
* cnt
+ j
;
3716 if (be32_to_cpu(vc
->conf
.phys_refnum
[i
]) != 0xffffffff)
3719 if (vc
->other_bvds
== NULL
)
3722 for (n
= 1; n
< vc
->conf
.sec_elmnt_count
; n
++) {
3723 struct vd_config
*vd
= vc
->other_bvds
[n
-1];
3724 sec
= vd
->sec_elmnt_seq
;
3725 if (sec
== DDF_UNUSED_BVD
)
3727 for (i
= 0, j
= 0 ; i
< nmax
; i
++) {
3728 if (be32_eq(vd
->phys_refnum
[i
], refnum
)) {
3731 return sec
* cnt
+ j
;
3733 if (be32_to_cpu(vd
->phys_refnum
[i
]) != 0xffffffff)
3739 return DDF_NOTFOUND
;
3742 static struct mdinfo
*container_content_ddf(struct supertype
*st
, char *subarray
)
3744 /* Given a container loaded by load_super_ddf_all,
3745 * extract information about all the arrays into
3748 * For each vcl in conflist: create an mdinfo, fill it in,
3749 * then look for matching devices (phys_refnum) in dlist
3750 * and create appropriate device mdinfo.
3752 struct ddf_super
*ddf
= st
->sb
;
3753 struct mdinfo
*rest
= NULL
;
3756 for (vc
= ddf
->conflist
; vc
; vc
=vc
->next
) {
3758 struct mdinfo
*this;
3764 (strtoul(subarray
, &ep
, 10) != vc
->vcnum
||
3768 if (vc
->conf
.sec_elmnt_count
> 1) {
3769 if (check_secondary(vc
) != 0)
3773 this = xcalloc(1, sizeof(*this));
3777 if (layout_ddf2md(&vc
->conf
, &this->array
))
3779 this->array
.md_minor
= -1;
3780 this->array
.major_version
= -1;
3781 this->array
.minor_version
= -2;
3782 this->safe_mode_delay
= DDF_SAFE_MODE_DELAY
;
3783 cptr
= (__u32
*)(vc
->conf
.guid
+ 16);
3784 this->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
3785 this->array
.utime
= DECADE
+
3786 be32_to_cpu(vc
->conf
.timestamp
);
3787 this->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
3790 if ((ddf
->virt
->entries
[i
].state
& DDF_state_inconsistent
) ||
3791 (ddf
->virt
->entries
[i
].init_state
& DDF_initstate_mask
) !=
3793 this->array
.state
= 0;
3794 this->resync_start
= 0;
3796 this->array
.state
= 1;
3797 this->resync_start
= MaxSector
;
3799 _ddf_array_name(this->name
, ddf
, i
);
3800 memset(this->uuid
, 0, sizeof(this->uuid
));
3801 this->component_size
= be64_to_cpu(vc
->conf
.blocks
);
3802 this->array
.size
= this->component_size
/ 2;
3803 this->container_member
= i
;
3805 ddf
->currentconf
= vc
;
3806 uuid_from_super_ddf(st
, this->uuid
);
3808 ddf
->currentconf
= NULL
;
3810 sprintf(this->text_version
, "/%s/%d",
3811 st
->container_devnm
, this->container_member
);
3813 for (pd
= 0; pd
< be16_to_cpu(ddf
->phys
->max_pdes
); pd
++) {
3816 const struct vd_config
*bvd
;
3820 if (be32_to_cpu(ddf
->phys
->entries
[pd
].refnum
) ==
3824 stt
= be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
3825 if ((stt
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
)) !=
3829 i
= get_pd_index_from_refnum(
3830 vc
, ddf
->phys
->entries
[pd
].refnum
,
3831 ddf
->mppe
, &bvd
, &iphys
);
3832 if (i
== DDF_NOTFOUND
)
3835 this->array
.working_disks
++;
3837 for (d
= ddf
->dlist
; d
; d
=d
->next
)
3838 if (be32_eq(d
->disk
.refnum
,
3839 ddf
->phys
->entries
[pd
].refnum
))
3842 /* Haven't found that one yet, maybe there are others */
3845 dev
= xcalloc(1, sizeof(*dev
));
3846 dev
->next
= this->devs
;
3849 dev
->disk
.number
= be32_to_cpu(d
->disk
.refnum
);
3850 dev
->disk
.major
= d
->major
;
3851 dev
->disk
.minor
= d
->minor
;
3852 dev
->disk
.raid_disk
= i
;
3853 dev
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
3854 dev
->recovery_start
= MaxSector
;
3856 dev
->events
= be32_to_cpu(ddf
->active
->seq
);
3858 be64_to_cpu(LBA_OFFSET(ddf
, bvd
)[iphys
]);
3859 dev
->component_size
= be64_to_cpu(bvd
->blocks
);
3861 strcpy(dev
->name
, d
->devname
);
3867 static int store_super_ddf(struct supertype
*st
, int fd
)
3869 struct ddf_super
*ddf
= st
->sb
;
3870 unsigned long long dsize
;
3877 if (!get_dev_size(fd
, NULL
, &dsize
))
3880 if (ddf
->dlist
|| ddf
->conflist
) {
3885 if (fstat(fd
, &sta
) == -1 || !S_ISBLK(sta
.st_mode
)) {
3886 pr_err("file descriptor for invalid device\n");
3889 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
3890 if (dl
->major
== (int)major(sta
.st_rdev
) &&
3891 dl
->minor
== (int)minor(sta
.st_rdev
))
3894 pr_err("couldn't find disk %d/%d\n",
3895 (int)major(sta
.st_rdev
),
3896 (int)minor(sta
.st_rdev
));
3901 ret
= (_write_super_to_disk(ddf
, dl
) != 1);
3906 if (posix_memalign(&buf
, 512, 512) != 0)
3908 memset(buf
, 0, 512);
3910 lseek64(fd
, dsize
-512, 0);
3911 rc
= write(fd
, buf
, 512);
3918 static int compare_super_ddf(struct supertype
*st
, struct supertype
*tst
,
3923 * 0 same, or first was empty, and second was copied
3924 * 1 second had wrong magic number - but that isn't possible
3926 * 3 wrong other info
3928 struct ddf_super
*first
= st
->sb
;
3929 struct ddf_super
*second
= tst
->sb
;
3930 struct dl
*dl1
, *dl2
;
3931 struct vcl
*vl1
, *vl2
;
3932 unsigned int max_vds
, max_pds
, pd
, vd
;
3940 if (memcmp(first
->anchor
.guid
, second
->anchor
.guid
, DDF_GUID_LEN
) != 0)
3943 /* It is only OK to compare info in the anchor. Anything else
3944 * could be changing due to a reconfig so must be ignored.
3945 * guid really should be enough anyway.
3948 if (!be32_eq(first
->active
->seq
, second
->active
->seq
)) {
3949 dprintf("sequence number mismatch %u<->%u\n",
3950 be32_to_cpu(first
->active
->seq
),
3951 be32_to_cpu(second
->active
->seq
));
3956 * At this point we are fairly sure that the meta data matches.
3957 * But the new disk may contain additional local data.
3958 * Add it to the super block.
3960 max_vds
= be16_to_cpu(first
->active
->max_vd_entries
);
3961 max_pds
= be16_to_cpu(first
->phys
->max_pdes
);
3962 for (vl2
= second
->conflist
; vl2
; vl2
= vl2
->next
) {
3963 for (vl1
= first
->conflist
; vl1
; vl1
= vl1
->next
)
3964 if (!memcmp(vl1
->conf
.guid
, vl2
->conf
.guid
,
3968 if (vl1
->other_bvds
!= NULL
&&
3969 vl1
->conf
.sec_elmnt_seq
!=
3970 vl2
->conf
.sec_elmnt_seq
) {
3971 dprintf("adding BVD %u\n",
3972 vl2
->conf
.sec_elmnt_seq
);
3973 add_other_bvd(vl1
, &vl2
->conf
,
3974 first
->conf_rec_len
*512);
3979 if (posix_memalign((void **)&vl1
, 512,
3980 (first
->conf_rec_len
*512 +
3981 offsetof(struct vcl
, conf
))) != 0) {
3982 pr_err("could not allocate vcl buf\n");
3986 vl1
->next
= first
->conflist
;
3987 vl1
->block_sizes
= NULL
;
3988 memcpy(&vl1
->conf
, &vl2
->conf
, first
->conf_rec_len
*512);
3989 if (alloc_other_bvds(first
, vl1
) != 0) {
3990 pr_err("could not allocate other bvds\n");
3994 for (vd
= 0; vd
< max_vds
; vd
++)
3995 if (!memcmp(first
->virt
->entries
[vd
].guid
,
3996 vl1
->conf
.guid
, DDF_GUID_LEN
))
3999 dprintf("added config for VD %u\n", vl1
->vcnum
);
4000 first
->conflist
= vl1
;
4003 for (dl2
= second
->dlist
; dl2
; dl2
= dl2
->next
) {
4004 for (dl1
= first
->dlist
; dl1
; dl1
= dl1
->next
)
4005 if (be32_eq(dl1
->disk
.refnum
, dl2
->disk
.refnum
))
4010 if (posix_memalign((void **)&dl1
, 512,
4011 sizeof(*dl1
) + (first
->max_part
) *
4012 sizeof(dl1
->vlist
[0])) != 0) {
4013 pr_err("could not allocate disk info buffer\n");
4016 memcpy(dl1
, dl2
, sizeof(*dl1
));
4017 dl1
->mdupdate
= NULL
;
4018 dl1
->next
= first
->dlist
;
4020 for (pd
= 0; pd
< max_pds
; pd
++)
4021 if (be32_eq(first
->phys
->entries
[pd
].refnum
,
4024 dl1
->pdnum
= pd
< max_pds
? (int)pd
: -1;
4026 if (posix_memalign((void **)&dl1
->spare
, 512,
4027 first
->conf_rec_len
*512) != 0) {
4028 pr_err("could not allocate spare info buf\n");
4031 memcpy(dl1
->spare
, dl2
->spare
, first
->conf_rec_len
*512);
4033 for (vd
= 0 ; vd
< first
->max_part
; vd
++) {
4034 if (!dl2
->vlist
[vd
]) {
4035 dl1
->vlist
[vd
] = NULL
;
4038 for (vl1
= first
->conflist
; vl1
; vl1
= vl1
->next
) {
4039 if (!memcmp(vl1
->conf
.guid
,
4040 dl2
->vlist
[vd
]->conf
.guid
,
4043 dl1
->vlist
[vd
] = vl1
;
4047 dprintf("added disk %d: %08x\n", dl1
->pdnum
,
4048 be32_to_cpu(dl1
->disk
.refnum
));
4055 * A new array 'a' has been started which claims to be instance 'inst'
4056 * within container 'c'.
4057 * We need to confirm that the array matches the metadata in 'c' so
4058 * that we don't corrupt any metadata.
4060 static int ddf_open_new(struct supertype
*c
, struct active_array
*a
, char *inst
)
4062 struct ddf_super
*ddf
= c
->sb
;
4066 static const char faulty
[] = "faulty";
4068 if (all_ff(ddf
->virt
->entries
[n
].guid
)) {
4069 pr_err("subarray %d doesn't exist\n", n
);
4072 dprintf("new subarray %d, GUID: %s\n", n
,
4073 guid_str(ddf
->virt
->entries
[n
].guid
));
4074 for (dev
= a
->info
.devs
; dev
; dev
= dev
->next
) {
4075 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4076 if (dl
->major
== dev
->disk
.major
&&
4077 dl
->minor
== dev
->disk
.minor
)
4079 if (!dl
|| dl
->pdnum
< 0) {
4080 pr_err("device %d/%d of subarray %d not found in meta data\n",
4081 dev
->disk
.major
, dev
->disk
.minor
, n
);
4084 if ((be16_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].state
) &
4085 (DDF_Online
|DDF_Missing
|DDF_Failed
)) != DDF_Online
) {
4086 pr_err("new subarray %d contains broken device %d/%d (%02x)\n",
4087 n
, dl
->major
, dl
->minor
,
4088 be16_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].state
));
4089 if (write(dev
->state_fd
, faulty
, sizeof(faulty
)-1) !=
4091 pr_err("Write to state_fd failed\n");
4092 dev
->curr_state
= DS_FAULTY
;
4095 a
->info
.container_member
= n
;
4099 static void handle_missing(struct ddf_super
*ddf
, struct active_array
*a
, int inst
)
4101 /* This member array is being activated. If any devices
4102 * are missing they must now be marked as failed.
4104 struct vd_config
*vc
;
4112 for (n
= 0; ; n
++) {
4113 vc
= find_vdcr(ddf
, inst
, n
, &n_bvd
, &vcl
);
4116 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4117 if (be32_eq(dl
->disk
.refnum
, vc
->phys_refnum
[n_bvd
]))
4120 /* Found this disk, so not missing */
4123 /* Mark the device as failed/missing. */
4124 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
4125 if (pd
>= 0 && be16_and(ddf
->phys
->entries
[pd
].state
,
4126 cpu_to_be16(DDF_Online
))) {
4127 be16_clear(ddf
->phys
->entries
[pd
].state
,
4128 cpu_to_be16(DDF_Online
));
4129 be16_set(ddf
->phys
->entries
[pd
].state
,
4130 cpu_to_be16(DDF_Failed
|DDF_Missing
));
4131 vc
->phys_refnum
[n_bvd
] = cpu_to_be32(0);
4132 ddf_set_updates_pending(ddf
, vc
);
4135 /* Mark the array as Degraded */
4136 state
= get_svd_state(ddf
, vcl
);
4137 if (ddf
->virt
->entries
[inst
].state
!=
4138 ((ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4140 ddf
->virt
->entries
[inst
].state
=
4141 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4143 a
->check_degraded
= 1;
4144 ddf_set_updates_pending(ddf
, vc
);
4150 * The array 'a' is to be marked clean in the metadata.
4151 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
4152 * clean up to the point (in sectors). If that cannot be recorded in the
4153 * metadata, then leave it as dirty.
4155 * For DDF, we need to clear the DDF_state_inconsistent bit in the
4156 * !global! virtual_disk.virtual_entry structure.
4158 static int ddf_set_array_state(struct active_array
*a
, int consistent
)
4160 struct ddf_super
*ddf
= a
->container
->sb
;
4161 int inst
= a
->info
.container_member
;
4162 int old
= ddf
->virt
->entries
[inst
].state
;
4163 if (consistent
== 2) {
4164 handle_missing(ddf
, a
, inst
);
4166 if (!is_resync_complete(&a
->info
))
4170 ddf
->virt
->entries
[inst
].state
&= ~DDF_state_inconsistent
;
4172 ddf
->virt
->entries
[inst
].state
|= DDF_state_inconsistent
;
4173 if (old
!= ddf
->virt
->entries
[inst
].state
)
4174 ddf_set_updates_pending(ddf
, NULL
);
4176 old
= ddf
->virt
->entries
[inst
].init_state
;
4177 ddf
->virt
->entries
[inst
].init_state
&= ~DDF_initstate_mask
;
4178 if (is_resync_complete(&a
->info
))
4179 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_full
;
4180 else if (a
->info
.resync_start
== 0)
4181 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_not
;
4183 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_quick
;
4184 if (old
!= ddf
->virt
->entries
[inst
].init_state
)
4185 ddf_set_updates_pending(ddf
, NULL
);
4187 dprintf("ddf mark %d/%s (%d) %s %llu\n", inst
,
4188 guid_str(ddf
->virt
->entries
[inst
].guid
), a
->curr_state
,
4189 consistent
?"clean":"dirty",
4190 a
->info
.resync_start
);
4194 static int get_bvd_state(const struct ddf_super
*ddf
,
4195 const struct vd_config
*vc
)
4197 unsigned int i
, n_bvd
, working
= 0;
4198 unsigned int n_prim
= be16_to_cpu(vc
->prim_elmnt_count
);
4200 char *avail
= xcalloc(1, n_prim
);
4201 mdu_array_info_t array
;
4203 layout_ddf2md(vc
, &array
);
4205 for (i
= 0; i
< n_prim
; i
++) {
4206 if (!find_index_in_bvd(ddf
, vc
, i
, &n_bvd
))
4208 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
4211 st
= be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
4212 if ((st
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
)) ==
4219 state
= DDF_state_degraded
;
4220 if (working
== n_prim
)
4221 state
= DDF_state_optimal
;
4227 state
= DDF_state_failed
;
4231 state
= DDF_state_failed
;
4232 else if (working
>= 2)
4233 state
= DDF_state_part_optimal
;
4236 if (!enough(10, n_prim
, array
.layout
, 1, avail
))
4237 state
= DDF_state_failed
;
4241 if (working
< n_prim
- 1)
4242 state
= DDF_state_failed
;
4245 if (working
< n_prim
- 2)
4246 state
= DDF_state_failed
;
4247 else if (working
== n_prim
- 1)
4248 state
= DDF_state_part_optimal
;
4254 static int secondary_state(int state
, int other
, int seclevel
)
4256 if (state
== DDF_state_optimal
&& other
== DDF_state_optimal
)
4257 return DDF_state_optimal
;
4258 if (seclevel
== DDF_2MIRRORED
) {
4259 if (state
== DDF_state_optimal
|| other
== DDF_state_optimal
)
4260 return DDF_state_part_optimal
;
4261 if (state
== DDF_state_failed
&& other
== DDF_state_failed
)
4262 return DDF_state_failed
;
4263 return DDF_state_degraded
;
4265 if (state
== DDF_state_failed
|| other
== DDF_state_failed
)
4266 return DDF_state_failed
;
4267 if (state
== DDF_state_degraded
|| other
== DDF_state_degraded
)
4268 return DDF_state_degraded
;
4269 return DDF_state_part_optimal
;
4273 static int get_svd_state(const struct ddf_super
*ddf
, const struct vcl
*vcl
)
4275 int state
= get_bvd_state(ddf
, &vcl
->conf
);
4277 for (i
= 1; i
< vcl
->conf
.sec_elmnt_count
; i
++) {
4278 state
= secondary_state(
4280 get_bvd_state(ddf
, vcl
->other_bvds
[i
-1]),
4287 * The state of each disk is stored in the global phys_disk structure
4288 * in phys_disk.entries[n].state.
4289 * This makes various combinations awkward.
4290 * - When a device fails in any array, it must be failed in all arrays
4291 * that include a part of this device.
4292 * - When a component is rebuilding, we cannot include it officially in the
4293 * array unless this is the only array that uses the device.
4295 * So: when transitioning:
4296 * Online -> failed, just set failed flag. monitor will propagate
4297 * spare -> online, the device might need to be added to the array.
4298 * spare -> failed, just set failed. Don't worry if in array or not.
4300 static void ddf_set_disk(struct active_array
*a
, int n
, int state
)
4302 struct ddf_super
*ddf
= a
->container
->sb
;
4303 unsigned int inst
= a
->info
.container_member
, n_bvd
;
4305 struct vd_config
*vc
= find_vdcr(ddf
, inst
, (unsigned int)n
,
4312 dprintf("%d to %x\n", n
, state
);
4314 dprintf("ddf: cannot find instance %d!!\n", inst
);
4317 /* Find the matching slot in 'info'. */
4318 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
4319 if (mdi
->disk
.raid_disk
== n
)
4322 pr_err("cannot find raid disk %d\n", n
);
4326 /* and find the 'dl' entry corresponding to that. */
4327 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4328 if (mdi
->state_fd
>= 0 &&
4329 mdi
->disk
.major
== dl
->major
&&
4330 mdi
->disk
.minor
== dl
->minor
)
4333 pr_err("cannot find raid disk %d (%d/%d)\n",
4334 n
, mdi
->disk
.major
, mdi
->disk
.minor
);
4338 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
4339 if (pd
< 0 || pd
!= dl
->pdnum
) {
4340 /* disk doesn't currently exist or has changed.
4341 * If it is now in_sync, insert it. */
4342 dprintf("phys disk not found for %d: %d/%d ref %08x\n",
4343 dl
->pdnum
, dl
->major
, dl
->minor
,
4344 be32_to_cpu(dl
->disk
.refnum
));
4345 dprintf("array %u disk %u ref %08x pd %d\n",
4347 be32_to_cpu(vc
->phys_refnum
[n_bvd
]), pd
);
4348 if ((state
& DS_INSYNC
) && ! (state
& DS_FAULTY
) &&
4351 vc
->phys_refnum
[n_bvd
] = dl
->disk
.refnum
;
4352 LBA_OFFSET(ddf
, vc
)[n_bvd
] =
4353 cpu_to_be64(mdi
->data_offset
);
4354 be16_clear(ddf
->phys
->entries
[pd
].type
,
4355 cpu_to_be16(DDF_Global_Spare
));
4356 be16_set(ddf
->phys
->entries
[pd
].type
,
4357 cpu_to_be16(DDF_Active_in_VD
));
4361 be16 old
= ddf
->phys
->entries
[pd
].state
;
4362 if (state
& DS_FAULTY
)
4363 be16_set(ddf
->phys
->entries
[pd
].state
,
4364 cpu_to_be16(DDF_Failed
));
4365 if (state
& DS_INSYNC
) {
4366 be16_set(ddf
->phys
->entries
[pd
].state
,
4367 cpu_to_be16(DDF_Online
));
4368 be16_clear(ddf
->phys
->entries
[pd
].state
,
4369 cpu_to_be16(DDF_Rebuilding
));
4371 if (!be16_eq(old
, ddf
->phys
->entries
[pd
].state
))
4375 dprintf("ddf: set_disk %d (%08x) to %x->%02x\n", n
,
4376 be32_to_cpu(dl
->disk
.refnum
), state
,
4377 be16_to_cpu(ddf
->phys
->entries
[pd
].state
));
4379 /* Now we need to check the state of the array and update
4380 * virtual_disk.entries[n].state.
4381 * It needs to be one of "optimal", "degraded", "failed".
4382 * I don't understand 'deleted' or 'missing'.
4384 state
= get_svd_state(ddf
, vcl
);
4386 if (ddf
->virt
->entries
[inst
].state
!=
4387 ((ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4389 ddf
->virt
->entries
[inst
].state
=
4390 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4395 ddf_set_updates_pending(ddf
, vc
);
4398 static void ddf_sync_metadata(struct supertype
*st
)
4401 * Write all data to all devices.
4402 * Later, we might be able to track whether only local changes
4403 * have been made, or whether any global data has been changed,
4404 * but ddf is sufficiently weird that it probably always
4405 * changes global data ....
4407 struct ddf_super
*ddf
= st
->sb
;
4408 if (!ddf
->updates_pending
)
4410 ddf
->updates_pending
= 0;
4411 __write_init_super_ddf(st
);
4412 dprintf("ddf: sync_metadata\n");
4415 static int del_from_conflist(struct vcl
**list
, const char *guid
)
4419 for (p
= list
; p
&& *p
; p
= &((*p
)->next
))
4420 if (!memcmp((*p
)->conf
.guid
, guid
, DDF_GUID_LEN
)) {
4427 static int _kill_subarray_ddf(struct ddf_super
*ddf
, const char *guid
)
4430 unsigned int vdnum
, i
;
4431 vdnum
= find_vde_by_guid(ddf
, guid
);
4432 if (vdnum
== DDF_NOTFOUND
) {
4433 pr_err("could not find VD %s\n", guid_str(guid
));
4436 if (del_from_conflist(&ddf
->conflist
, guid
) == 0) {
4437 pr_err("could not find conf %s\n", guid_str(guid
));
4440 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4441 for (i
= 0; i
< ddf
->max_part
; i
++)
4442 if (dl
->vlist
[i
] != NULL
&&
4443 !memcmp(dl
->vlist
[i
]->conf
.guid
, guid
,
4445 dl
->vlist
[i
] = NULL
;
4446 memset(ddf
->virt
->entries
[vdnum
].guid
, 0xff, DDF_GUID_LEN
);
4447 dprintf("deleted %s\n", guid_str(guid
));
4451 static int kill_subarray_ddf(struct supertype
*st
, char *subarray_id
)
4453 struct ddf_super
*ddf
= st
->sb
;
4455 * currentconf is set in container_content_ddf,
4456 * called with subarray arg
4458 struct vcl
*victim
= ddf
->currentconf
;
4459 struct vd_config
*conf
;
4462 ddf
->currentconf
= NULL
;
4464 pr_err("nothing to kill\n");
4467 conf
= &victim
->conf
;
4468 vdnum
= find_vde_by_guid(ddf
, conf
->guid
);
4469 if (vdnum
== DDF_NOTFOUND
) {
4470 pr_err("could not find VD %s\n", guid_str(conf
->guid
));
4473 if (st
->update_tail
) {
4474 struct virtual_disk
*vd
;
4475 int len
= sizeof(struct virtual_disk
)
4476 + sizeof(struct virtual_entry
);
4479 pr_err("failed to allocate %d bytes\n", len
);
4482 memset(vd
, 0 , len
);
4483 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
4484 vd
->populated_vdes
= cpu_to_be16(0);
4485 memcpy(vd
->entries
[0].guid
, conf
->guid
, DDF_GUID_LEN
);
4486 /* we use DDF_state_deleted as marker */
4487 vd
->entries
[0].state
= DDF_state_deleted
;
4488 append_metadata_update(st
, vd
, len
);
4490 _kill_subarray_ddf(ddf
, conf
->guid
);
4491 ddf_set_updates_pending(ddf
, NULL
);
4492 ddf_sync_metadata(st
);
4497 static void copy_matching_bvd(struct ddf_super
*ddf
,
4498 struct vd_config
*conf
,
4499 const struct metadata_update
*update
)
4502 be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
4503 unsigned int len
= ddf
->conf_rec_len
* 512;
4505 struct vd_config
*vc
;
4506 for (p
= update
->buf
; p
< update
->buf
+ update
->len
; p
+= len
) {
4507 vc
= (struct vd_config
*) p
;
4508 if (vc
->sec_elmnt_seq
== conf
->sec_elmnt_seq
) {
4509 memcpy(conf
->phys_refnum
, vc
->phys_refnum
,
4510 mppe
* (sizeof(__u32
) + sizeof(__u64
)));
4514 pr_err("no match for BVD %d of %s in update\n",
4515 conf
->sec_elmnt_seq
, guid_str(conf
->guid
));
4518 static void ddf_process_phys_update(struct supertype
*st
,
4519 struct metadata_update
*update
)
4521 struct ddf_super
*ddf
= st
->sb
;
4522 struct phys_disk
*pd
;
4525 pd
= (struct phys_disk
*)update
->buf
;
4526 ent
= be16_to_cpu(pd
->used_pdes
);
4527 if (ent
>= be16_to_cpu(ddf
->phys
->max_pdes
))
4529 if (be16_and(pd
->entries
[0].state
, cpu_to_be16(DDF_Missing
))) {
4531 /* removing this disk. */
4532 be16_set(ddf
->phys
->entries
[ent
].state
,
4533 cpu_to_be16(DDF_Missing
));
4534 for (dlp
= &ddf
->dlist
; *dlp
; dlp
= &(*dlp
)->next
) {
4535 struct dl
*dl
= *dlp
;
4536 if (dl
->pdnum
== (signed)ent
) {
4540 update
->space
= dl
->devname
;
4541 *(void**)dl
= update
->space_list
;
4542 update
->space_list
= (void**)dl
;
4546 ddf_set_updates_pending(ddf
, NULL
);
4549 if (!all_ff(ddf
->phys
->entries
[ent
].guid
))
4551 ddf
->phys
->entries
[ent
] = pd
->entries
[0];
4552 ddf
->phys
->used_pdes
= cpu_to_be16
4553 (1 + be16_to_cpu(ddf
->phys
->used_pdes
));
4554 ddf_set_updates_pending(ddf
, NULL
);
4555 if (ddf
->add_list
) {
4556 struct active_array
*a
;
4557 struct dl
*al
= ddf
->add_list
;
4558 ddf
->add_list
= al
->next
;
4560 al
->next
= ddf
->dlist
;
4563 /* As a device has been added, we should check
4564 * for any degraded devices that might make
4565 * use of this spare */
4566 for (a
= st
->arrays
; a
; a
=a
->next
)
4567 a
->check_degraded
= 1;
4571 static void ddf_process_virt_update(struct supertype
*st
,
4572 struct metadata_update
*update
)
4574 struct ddf_super
*ddf
= st
->sb
;
4575 struct virtual_disk
*vd
;
4578 vd
= (struct virtual_disk
*)update
->buf
;
4580 if (vd
->entries
[0].state
== DDF_state_deleted
) {
4581 if (_kill_subarray_ddf(ddf
, vd
->entries
[0].guid
))
4584 ent
= find_vde_by_guid(ddf
, vd
->entries
[0].guid
);
4585 if (ent
!= DDF_NOTFOUND
) {
4586 dprintf("VD %s exists already in slot %d\n",
4587 guid_str(vd
->entries
[0].guid
),
4591 ent
= find_unused_vde(ddf
);
4592 if (ent
== DDF_NOTFOUND
)
4594 ddf
->virt
->entries
[ent
] = vd
->entries
[0];
4595 ddf
->virt
->populated_vdes
=
4598 ddf
->virt
->populated_vdes
));
4599 dprintf("added VD %s in slot %d(s=%02x i=%02x)\n",
4600 guid_str(vd
->entries
[0].guid
), ent
,
4601 ddf
->virt
->entries
[ent
].state
,
4602 ddf
->virt
->entries
[ent
].init_state
);
4604 ddf_set_updates_pending(ddf
, NULL
);
4607 static void ddf_remove_failed(struct ddf_super
*ddf
)
4609 /* Now remove any 'Failed' devices that are not part
4610 * of any VD. They will have the Transition flag set.
4611 * Once done, we need to update all dl->pdnum numbers.
4614 unsigned int pd2
= 0;
4617 for (pdnum
= 0; pdnum
< be16_to_cpu(ddf
->phys
->max_pdes
);
4619 if (be32_to_cpu(ddf
->phys
->entries
[pdnum
].refnum
) ==
4622 if (be16_and(ddf
->phys
->entries
[pdnum
].state
,
4623 cpu_to_be16(DDF_Failed
)) &&
4624 be16_and(ddf
->phys
->entries
[pdnum
].state
,
4625 cpu_to_be16(DDF_Transition
))) {
4626 /* skip this one unless in dlist*/
4627 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4628 if (dl
->pdnum
== (int)pdnum
)
4636 ddf
->phys
->entries
[pd2
] =
4637 ddf
->phys
->entries
[pdnum
];
4638 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4639 if (dl
->pdnum
== (int)pdnum
)
4644 ddf
->phys
->used_pdes
= cpu_to_be16(pd2
);
4645 while (pd2
< pdnum
) {
4646 memset(ddf
->phys
->entries
[pd2
].guid
, 0xff,
4652 static void ddf_update_vlist(struct ddf_super
*ddf
, struct dl
*dl
)
4655 unsigned int vn
= 0;
4656 int in_degraded
= 0;
4660 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
) {
4661 unsigned int dn
, ibvd
;
4662 const struct vd_config
*conf
;
4664 dn
= get_pd_index_from_refnum(vcl
,
4668 if (dn
== DDF_NOTFOUND
)
4670 dprintf("dev %d/%08x has %s (sec=%u) at %d\n",
4672 be32_to_cpu(dl
->disk
.refnum
),
4673 guid_str(conf
->guid
),
4674 conf
->sec_elmnt_seq
, vn
);
4675 /* Clear the Transition flag */
4677 (ddf
->phys
->entries
[dl
->pdnum
].state
,
4678 cpu_to_be16(DDF_Failed
)))
4679 be16_clear(ddf
->phys
4680 ->entries
[dl
->pdnum
].state
,
4681 cpu_to_be16(DDF_Transition
));
4682 dl
->vlist
[vn
++] = vcl
;
4683 vstate
= ddf
->virt
->entries
[vcl
->vcnum
].state
4685 if (vstate
== DDF_state_degraded
||
4686 vstate
== DDF_state_part_optimal
)
4689 while (vn
< ddf
->max_part
)
4690 dl
->vlist
[vn
++] = NULL
;
4692 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4693 cpu_to_be16(DDF_Global_Spare
));
4694 if (!be16_and(ddf
->phys
4695 ->entries
[dl
->pdnum
].type
,
4696 cpu_to_be16(DDF_Active_in_VD
))) {
4698 ->entries
[dl
->pdnum
].type
,
4699 cpu_to_be16(DDF_Active_in_VD
));
4702 ->entries
[dl
->pdnum
]
4709 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4710 cpu_to_be16(DDF_Global_Spare
));
4711 be16_set(ddf
->phys
->entries
[dl
->pdnum
].type
,
4712 cpu_to_be16(DDF_Spare
));
4714 if (!dl
->vlist
[0] && !dl
->spare
) {
4715 be16_set(ddf
->phys
->entries
[dl
->pdnum
].type
,
4716 cpu_to_be16(DDF_Global_Spare
));
4717 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4718 cpu_to_be16(DDF_Spare
));
4719 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4720 cpu_to_be16(DDF_Active_in_VD
));
4724 static void ddf_process_conf_update(struct supertype
*st
,
4725 struct metadata_update
*update
)
4727 struct ddf_super
*ddf
= st
->sb
;
4728 struct vd_config
*vc
;
4732 unsigned int pdnum
, len
;
4734 vc
= (struct vd_config
*)update
->buf
;
4735 len
= ddf
->conf_rec_len
* 512;
4736 if ((unsigned int)update
->len
!= len
* vc
->sec_elmnt_count
) {
4737 pr_err("%s: insufficient data (%d) for %u BVDs\n",
4738 guid_str(vc
->guid
), update
->len
,
4739 vc
->sec_elmnt_count
);
4742 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
4743 if (memcmp(vcl
->conf
.guid
, vc
->guid
, DDF_GUID_LEN
) == 0)
4745 dprintf("conf update for %s (%s)\n",
4746 guid_str(vc
->guid
), (vcl
? "old" : "new"));
4748 /* An update, just copy the phys_refnum and lba_offset
4753 copy_matching_bvd(ddf
, &vcl
->conf
, update
);
4754 for (k
= 0; k
< be16_to_cpu(vc
->prim_elmnt_count
); k
++)
4755 dprintf("BVD %u has %08x at %llu\n", 0,
4756 be32_to_cpu(vcl
->conf
.phys_refnum
[k
]),
4757 be64_to_cpu(LBA_OFFSET(ddf
,
4759 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++) {
4760 copy_matching_bvd(ddf
, vcl
->other_bvds
[i
-1],
4762 for (k
= 0; k
< be16_to_cpu(
4763 vc
->prim_elmnt_count
); k
++)
4764 dprintf("BVD %u has %08x at %llu\n", i
,
4766 (vcl
->other_bvds
[i
-1]->
4771 vcl
->other_bvds
[i
-1])[k
]));
4778 vcl
= update
->space
;
4779 update
->space
= NULL
;
4780 vcl
->next
= ddf
->conflist
;
4781 memcpy(&vcl
->conf
, vc
, len
);
4782 ent
= find_vde_by_guid(ddf
, vc
->guid
);
4783 if (ent
== DDF_NOTFOUND
)
4786 ddf
->conflist
= vcl
;
4787 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++)
4788 memcpy(vcl
->other_bvds
[i
-1],
4789 update
->buf
+ len
* i
, len
);
4791 /* Set DDF_Transition on all Failed devices - to help
4792 * us detect those that are no longer in use
4794 for (pdnum
= 0; pdnum
< be16_to_cpu(ddf
->phys
->max_pdes
);
4796 if (be16_and(ddf
->phys
->entries
[pdnum
].state
,
4797 cpu_to_be16(DDF_Failed
)))
4798 be16_set(ddf
->phys
->entries
[pdnum
].state
,
4799 cpu_to_be16(DDF_Transition
));
4801 /* Now make sure vlist is correct for each dl. */
4802 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4803 ddf_update_vlist(ddf
, dl
);
4804 ddf_remove_failed(ddf
);
4806 ddf_set_updates_pending(ddf
, vc
);
4809 static void ddf_process_update(struct supertype
*st
,
4810 struct metadata_update
*update
)
4812 /* Apply this update to the metadata.
4813 * The first 4 bytes are a DDF_*_MAGIC which guides
4815 * Possible update are:
4816 * DDF_PHYS_RECORDS_MAGIC
4817 * Add a new physical device or remove an old one.
4818 * Changes to this record only happen implicitly.
4819 * used_pdes is the device number.
4820 * DDF_VIRT_RECORDS_MAGIC
4821 * Add a new VD. Possibly also change the 'access' bits.
4822 * populated_vdes is the entry number.
4824 * New or updated VD. the VIRT_RECORD must already
4825 * exist. For an update, phys_refnum and lba_offset
4826 * (at least) are updated, and the VD_CONF must
4827 * be written to precisely those devices listed with
4829 * DDF_SPARE_ASSIGN_MAGIC
4830 * replacement Spare Assignment Record... but for which device?
4833 * - to create a new array, we send a VIRT_RECORD and
4834 * a VD_CONF. Then assemble and start the array.
4835 * - to activate a spare we send a VD_CONF to add the phys_refnum
4836 * and offset. This will also mark the spare as active with
4837 * a spare-assignment record.
4839 be32
*magic
= (be32
*)update
->buf
;
4841 dprintf("Process update %x\n", be32_to_cpu(*magic
));
4843 if (be32_eq(*magic
, DDF_PHYS_RECORDS_MAGIC
)) {
4844 if (update
->len
== (sizeof(struct phys_disk
) +
4845 sizeof(struct phys_disk_entry
)))
4846 ddf_process_phys_update(st
, update
);
4847 } else if (be32_eq(*magic
, DDF_VIRT_RECORDS_MAGIC
)) {
4848 if (update
->len
== (sizeof(struct virtual_disk
) +
4849 sizeof(struct virtual_entry
)))
4850 ddf_process_virt_update(st
, update
);
4851 } else if (be32_eq(*magic
, DDF_VD_CONF_MAGIC
)) {
4852 ddf_process_conf_update(st
, update
);
4854 /* case DDF_SPARE_ASSIGN_MAGIC */
4857 static int ddf_prepare_update(struct supertype
*st
,
4858 struct metadata_update
*update
)
4860 /* This update arrived at managemon.
4861 * We are about to pass it to monitor.
4862 * If a malloc is needed, do it here.
4864 struct ddf_super
*ddf
= st
->sb
;
4866 if (update
->len
< 4)
4868 magic
= (be32
*)update
->buf
;
4869 if (be32_eq(*magic
, DDF_VD_CONF_MAGIC
)) {
4871 struct vd_config
*conf
;
4872 if (update
->len
< (int)sizeof(*conf
))
4874 conf
= (struct vd_config
*) update
->buf
;
4875 if (posix_memalign(&update
->space
, 512,
4876 offsetof(struct vcl
, conf
)
4877 + ddf
->conf_rec_len
* 512) != 0) {
4878 update
->space
= NULL
;
4881 vcl
= update
->space
;
4882 vcl
->conf
.sec_elmnt_count
= conf
->sec_elmnt_count
;
4883 if (alloc_other_bvds(ddf
, vcl
) != 0) {
4884 free(update
->space
);
4885 update
->space
= NULL
;
4893 * Check degraded state of a RAID10.
4894 * returns 2 for good, 1 for degraded, 0 for failed, and -1 for error
4896 static int raid10_degraded(struct mdinfo
*info
)
4904 n_prim
= info
->array
.layout
& ~0x100;
4905 n_bvds
= info
->array
.raid_disks
/ n_prim
;
4906 found
= xmalloc(n_bvds
);
4909 memset(found
, 0, n_bvds
);
4910 for (d
= info
->devs
; d
; d
= d
->next
) {
4911 i
= d
->disk
.raid_disk
/ n_prim
;
4913 pr_err("BUG: invalid raid disk\n");
4916 if (d
->state_fd
> 0)
4920 for (i
= 0; i
< n_bvds
; i
++)
4922 dprintf("BVD %d/%d failed\n", i
, n_bvds
);
4925 } else if (found
[i
] < n_prim
) {
4926 dprintf("BVD %d/%d degraded\n", i
, n_bvds
);
4935 * Check if the array 'a' is degraded but not failed.
4936 * If it is, find as many spares as are available and needed and
4937 * arrange for their inclusion.
4938 * We only choose devices which are not already in the array,
4939 * and prefer those with a spare-assignment to this array.
4940 * Otherwise we choose global spares - assuming always that
4941 * there is enough room.
4942 * For each spare that we assign, we return an 'mdinfo' which
4943 * describes the position for the device in the array.
4944 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
4945 * the new phys_refnum and lba_offset values.
4947 * Only worry about BVDs at the moment.
4949 static struct mdinfo
*ddf_activate_spare(struct active_array
*a
,
4950 struct metadata_update
**updates
)
4954 struct ddf_super
*ddf
= a
->container
->sb
;
4956 struct mdinfo
*rv
= NULL
;
4958 struct metadata_update
*mu
;
4963 struct vd_config
*vc
;
4966 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
4967 if ((d
->curr_state
& DS_FAULTY
) &&
4969 /* wait for Removal to happen */
4971 if (d
->state_fd
>= 0)
4975 dprintf("working=%d (%d) level=%d\n", working
,
4976 a
->info
.array
.raid_disks
,
4977 a
->info
.array
.level
);
4978 if (working
== a
->info
.array
.raid_disks
)
4979 return NULL
; /* array not degraded */
4980 switch (a
->info
.array
.level
) {
4983 return NULL
; /* failed */
4987 if (working
< a
->info
.array
.raid_disks
- 1)
4988 return NULL
; /* failed */
4991 if (working
< a
->info
.array
.raid_disks
- 2)
4992 return NULL
; /* failed */
4995 if (raid10_degraded(&a
->info
) < 1)
4998 default: /* concat or stripe */
4999 return NULL
; /* failed */
5002 /* For each slot, if it is not working, find a spare */
5004 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
5005 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5006 if (d
->disk
.raid_disk
== i
)
5008 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
5009 if (d
&& (d
->state_fd
>= 0))
5012 /* OK, this device needs recovery. Find a spare */
5014 for ( ; dl
; dl
= dl
->next
) {
5015 unsigned long long esize
;
5016 unsigned long long pos
;
5019 int is_dedicated
= 0;
5024 state
= ddf
->phys
->entries
[dl
->pdnum
].state
;
5026 cpu_to_be16(DDF_Failed
|DDF_Missing
)) ||
5028 cpu_to_be16(DDF_Online
)))
5031 /* If in this array, skip */
5032 for (d2
= a
->info
.devs
; d2
; d2
= d2
->next
)
5033 if (d2
->state_fd
>= 0 &&
5034 d2
->disk
.major
== dl
->major
&&
5035 d2
->disk
.minor
== dl
->minor
) {
5036 dprintf("%x:%x (%08x) already in array\n",
5037 dl
->major
, dl
->minor
,
5038 be32_to_cpu(dl
->disk
.refnum
));
5043 if (be16_and(ddf
->phys
->entries
[dl
->pdnum
].type
,
5044 cpu_to_be16(DDF_Spare
))) {
5045 /* Check spare assign record */
5047 if (dl
->spare
->type
& DDF_spare_dedicated
) {
5048 /* check spare_ents for guid */
5055 if (memcmp(dl
->spare
->spare_ents
[j
].guid
,
5056 ddf
->virt
->entries
[a
->info
.container_member
].guid
,
5063 } else if (be16_and(ddf
->phys
->entries
[dl
->pdnum
].type
,
5064 cpu_to_be16(DDF_Global_Spare
))) {
5066 } else if (!be16_and(ddf
->phys
5067 ->entries
[dl
->pdnum
].state
,
5068 cpu_to_be16(DDF_Failed
))) {
5069 /* we can possibly use some of this */
5072 if ( ! (is_dedicated
||
5073 (is_global
&& global_ok
))) {
5074 dprintf("%x:%x not suitable: %d %d\n", dl
->major
, dl
->minor
,
5075 is_dedicated
, is_global
);
5079 /* We are allowed to use this device - is there space?
5080 * We need a->info.component_size sectors */
5081 esize
= a
->info
.component_size
;
5082 pos
= find_space(ddf
, dl
, INVALID_SECTORS
, &esize
);
5084 if (esize
< a
->info
.component_size
) {
5085 dprintf("%x:%x has no room: %llu %llu\n",
5086 dl
->major
, dl
->minor
,
5087 esize
, a
->info
.component_size
);
5092 /* Cool, we have a device with some space at pos */
5093 di
= xcalloc(1, sizeof(*di
));
5094 di
->disk
.number
= i
;
5095 di
->disk
.raid_disk
= i
;
5096 di
->disk
.major
= dl
->major
;
5097 di
->disk
.minor
= dl
->minor
;
5099 di
->recovery_start
= 0;
5100 di
->data_offset
= pos
;
5101 di
->component_size
= a
->info
.component_size
;
5104 dprintf("%x:%x (%08x) to be %d at %llu\n",
5105 dl
->major
, dl
->minor
,
5106 be32_to_cpu(dl
->disk
.refnum
), i
, pos
);
5110 if (!dl
&& ! global_ok
) {
5111 /* not enough dedicated spares, try global */
5119 /* No spares found */
5121 /* Now 'rv' has a list of devices to return.
5122 * Create a metadata_update record to update the
5123 * phys_refnum and lba_offset values
5125 vc
= find_vdcr(ddf
, a
->info
.container_member
, rv
->disk
.raid_disk
,
5130 mu
= xmalloc(sizeof(*mu
));
5131 if (posix_memalign(&mu
->space
, 512, sizeof(struct vcl
)) != 0) {
5136 mu
->len
= ddf
->conf_rec_len
* 512 * vcl
->conf
.sec_elmnt_count
;
5137 mu
->buf
= xmalloc(mu
->len
);
5139 mu
->space_list
= NULL
;
5140 mu
->next
= *updates
;
5141 memcpy(mu
->buf
, &vcl
->conf
, ddf
->conf_rec_len
* 512);
5142 for (j
= 1; j
< vcl
->conf
.sec_elmnt_count
; j
++)
5143 memcpy(mu
->buf
+ j
* ddf
->conf_rec_len
* 512,
5144 vcl
->other_bvds
[j
-1], ddf
->conf_rec_len
* 512);
5146 vc
= (struct vd_config
*)mu
->buf
;
5147 for (di
= rv
; di
; di
= di
->next
) {
5148 unsigned int i_sec
, i_prim
;
5149 i_sec
= di
->disk
.raid_disk
5150 / be16_to_cpu(vcl
->conf
.prim_elmnt_count
);
5151 i_prim
= di
->disk
.raid_disk
5152 % be16_to_cpu(vcl
->conf
.prim_elmnt_count
);
5153 vc
= (struct vd_config
*)(mu
->buf
5154 + i_sec
* ddf
->conf_rec_len
* 512);
5155 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
5156 if (dl
->major
== di
->disk
.major
&&
5157 dl
->minor
== di
->disk
.minor
)
5159 if (!dl
|| dl
->pdnum
< 0) {
5160 pr_err("BUG: can't find disk %d (%d/%d)\n",
5162 di
->disk
.major
, di
->disk
.minor
);
5165 vc
->phys_refnum
[i_prim
] = ddf
->phys
->entries
[dl
->pdnum
].refnum
;
5166 LBA_OFFSET(ddf
, vc
)[i_prim
] = cpu_to_be64(di
->data_offset
);
5167 dprintf("BVD %u gets %u: %08x at %llu\n", i_sec
, i_prim
,
5168 be32_to_cpu(vc
->phys_refnum
[i_prim
]),
5169 be64_to_cpu(LBA_OFFSET(ddf
, vc
)[i_prim
]));
5175 static int ddf_level_to_layout(int level
)
5182 return ALGORITHM_LEFT_SYMMETRIC
;
5184 return ALGORITHM_ROTATING_N_CONTINUE
;
5192 static void default_geometry_ddf(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5194 if (level
&& *level
== UnSet
)
5195 *level
= LEVEL_CONTAINER
;
5197 if (level
&& layout
&& *layout
== UnSet
)
5198 *layout
= ddf_level_to_layout(*level
);
5201 struct superswitch super_ddf
= {
5202 .examine_super
= examine_super_ddf
,
5203 .brief_examine_super
= brief_examine_super_ddf
,
5204 .brief_examine_subarrays
= brief_examine_subarrays_ddf
,
5205 .export_examine_super
= export_examine_super_ddf
,
5206 .detail_super
= detail_super_ddf
,
5207 .brief_detail_super
= brief_detail_super_ddf
,
5208 .validate_geometry
= validate_geometry_ddf
,
5209 .write_init_super
= write_init_super_ddf
,
5210 .add_to_super
= add_to_super_ddf
,
5211 .remove_from_super
= remove_from_super_ddf
,
5212 .load_container
= load_container_ddf
,
5213 .copy_metadata
= copy_metadata_ddf
,
5214 .kill_subarray
= kill_subarray_ddf
,
5215 .match_home
= match_home_ddf
,
5216 .uuid_from_super
= uuid_from_super_ddf
,
5217 .getinfo_super
= getinfo_super_ddf
,
5218 .update_super
= update_super_ddf
,
5220 .avail_size
= avail_size_ddf
,
5222 .compare_super
= compare_super_ddf
,
5224 .load_super
= load_super_ddf
,
5225 .init_super
= init_super_ddf
,
5226 .store_super
= store_super_ddf
,
5227 .free_super
= free_super_ddf
,
5228 .match_metadata_desc
= match_metadata_desc_ddf
,
5229 .container_content
= container_content_ddf
,
5230 .default_geometry
= default_geometry_ddf
,
5235 .open_new
= ddf_open_new
,
5236 .set_array_state
= ddf_set_array_state
,
5237 .set_disk
= ddf_set_disk
,
5238 .sync_metadata
= ddf_sync_metadata
,
5239 .process_update
= ddf_process_update
,
5240 .prepare_update
= ddf_prepare_update
,
5241 .activate_spare
= ddf_activate_spare
,