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
34 /* a non-official T10 name for creation GUIDs */
35 static char T10
[] = "Linux-MD";
37 /* DDF timestamps are 1980 based, so we need to add
38 * second-in-decade-of-seventies to convert to linux timestamps.
39 * 10 years with 2 leap years.
41 #define DECADE (3600*24*(365*10+2))
44 const unsigned char *buf
,
47 #define DDF_NOTFOUND (~0U)
48 #define DDF_CONTAINER (DDF_NOTFOUND-1)
50 /* Default for safe_mode_delay. Same value as for IMSM.
52 static const int DDF_SAFE_MODE_DELAY
= 4000;
54 /* The DDF metadata handling.
55 * DDF metadata lives at the end of the device.
56 * The last 512 byte block provides an 'anchor' which is used to locate
57 * the rest of the metadata which usually lives immediately behind the anchor.
60 * - all multibyte numeric fields are bigendian.
61 * - all strings are space padded.
65 typedef struct __be16
{
68 #define be16_eq(x, y) ((x)._v16 == (y)._v16)
69 #define be16_and(x, y) ((x)._v16 & (y)._v16)
70 #define be16_or(x, y) ((x)._v16 | (y)._v16)
71 #define be16_clear(x, y) ((x)._v16 &= ~(y)._v16)
72 #define be16_set(x, y) ((x)._v16 |= (y)._v16)
74 typedef struct __be32
{
77 #define be32_eq(x, y) ((x)._v32 == (y)._v32)
79 typedef struct __be64
{
82 #define be64_eq(x, y) ((x)._v64 == (y)._v64)
84 #define be16_to_cpu(be) __be16_to_cpu((be)._v16)
85 static inline be16
cpu_to_be16(__u16 x
)
87 be16 be
= { ._v16
= __cpu_to_be16(x
) };
91 #define be32_to_cpu(be) __be32_to_cpu((be)._v32)
92 static inline be32
cpu_to_be32(__u32 x
)
94 be32 be
= { ._v32
= __cpu_to_be32(x
) };
98 #define be64_to_cpu(be) __be64_to_cpu((be)._v64)
99 static inline be64
cpu_to_be64(__u64 x
)
101 be64 be
= { ._v64
= __cpu_to_be64(x
) };
105 /* Primary Raid Level (PRL) */
106 #define DDF_RAID0 0x00
107 #define DDF_RAID1 0x01
108 #define DDF_RAID3 0x03
109 #define DDF_RAID4 0x04
110 #define DDF_RAID5 0x05
111 #define DDF_RAID1E 0x11
112 #define DDF_JBOD 0x0f
113 #define DDF_CONCAT 0x1f
114 #define DDF_RAID5E 0x15
115 #define DDF_RAID5EE 0x25
116 #define DDF_RAID6 0x06
118 /* Raid Level Qualifier (RLQ) */
119 #define DDF_RAID0_SIMPLE 0x00
120 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
121 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
122 #define DDF_RAID3_0 0x00 /* parity in first extent */
123 #define DDF_RAID3_N 0x01 /* parity in last extent */
124 #define DDF_RAID4_0 0x00 /* parity in first extent */
125 #define DDF_RAID4_N 0x01 /* parity in last extent */
126 /* these apply to raid5e and raid5ee as well */
127 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
128 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
129 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
130 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
132 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
133 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
135 /* Secondary RAID Level (SRL) */
136 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
137 #define DDF_2MIRRORED 0x01
138 #define DDF_2CONCAT 0x02
139 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
142 #define DDF_HEADER_MAGIC cpu_to_be32(0xDE11DE11)
143 #define DDF_CONTROLLER_MAGIC cpu_to_be32(0xAD111111)
144 #define DDF_PHYS_RECORDS_MAGIC cpu_to_be32(0x22222222)
145 #define DDF_PHYS_DATA_MAGIC cpu_to_be32(0x33333333)
146 #define DDF_VIRT_RECORDS_MAGIC cpu_to_be32(0xDDDDDDDD)
147 #define DDF_VD_CONF_MAGIC cpu_to_be32(0xEEEEEEEE)
148 #define DDF_SPARE_ASSIGN_MAGIC cpu_to_be32(0x55555555)
149 #define DDF_VU_CONF_MAGIC cpu_to_be32(0x88888888)
150 #define DDF_VENDOR_LOG_MAGIC cpu_to_be32(0x01dBEEF0)
151 #define DDF_BBM_LOG_MAGIC cpu_to_be32(0xABADB10C)
153 #define DDF_GUID_LEN 24
154 #define DDF_REVISION_0 "01.00.00"
155 #define DDF_REVISION_2 "01.02.00"
158 be32 magic
; /* DDF_HEADER_MAGIC */
160 char guid
[DDF_GUID_LEN
];
161 char revision
[8]; /* 01.02.00 */
162 be32 seq
; /* starts at '1' */
167 __u8 pad0
; /* 0xff */
168 __u8 pad1
[12]; /* 12 * 0xff */
169 /* 64 bytes so far */
170 __u8 header_ext
[32]; /* reserved: fill with 0xff */
174 __u8 pad2
[3]; /* 0xff */
175 be32 workspace_len
; /* sectors for vendor space -
176 * at least 32768(sectors) */
178 be16 max_pd_entries
; /* one of 15, 63, 255, 1023, 4095 */
179 be16 max_vd_entries
; /* 2^(4,6,8,10,12)-1 : i.e. as above */
180 be16 max_partitions
; /* i.e. max num of configuration
181 record entries per disk */
182 be16 config_record_len
; /* 1 +ROUNDUP(max_primary_element_entries
184 be16 max_primary_element_entries
; /* 16, 64, 256, 1024, or 4096 */
185 __u8 pad3
[54]; /* 0xff */
186 /* 192 bytes so far */
187 be32 controller_section_offset
;
188 be32 controller_section_length
;
189 be32 phys_section_offset
;
190 be32 phys_section_length
;
191 be32 virt_section_offset
;
192 be32 virt_section_length
;
193 be32 config_section_offset
;
194 be32 config_section_length
;
195 be32 data_section_offset
;
196 be32 data_section_length
;
197 be32 bbm_section_offset
;
198 be32 bbm_section_length
;
199 be32 diag_space_offset
;
200 be32 diag_space_length
;
203 /* 256 bytes so far */
204 __u8 pad4
[256]; /* 0xff */
208 #define DDF_HEADER_ANCHOR 0x00
209 #define DDF_HEADER_PRIMARY 0x01
210 #define DDF_HEADER_SECONDARY 0x02
212 /* The content of the 'controller section' - global scope */
213 struct ddf_controller_data
{
214 be32 magic
; /* DDF_CONTROLLER_MAGIC */
216 char guid
[DDF_GUID_LEN
];
217 struct controller_type
{
224 __u8 pad
[8]; /* 0xff */
225 __u8 vendor_data
[448];
228 /* The content of phys_section - global scope */
230 be32 magic
; /* DDF_PHYS_RECORDS_MAGIC */
232 be16 used_pdes
; /* This is a counter, not a max - the list
233 * of used entries may not be dense */
236 struct phys_disk_entry
{
237 char guid
[DDF_GUID_LEN
];
241 be64 config_size
; /* DDF structures must be after here */
242 char path
[18]; /* Another horrible structure really
243 * but is "used for information
249 /* phys_disk_entry.type is a bitmap - bigendian remember */
250 #define DDF_Forced_PD_GUID 1
251 #define DDF_Active_in_VD 2
252 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
253 #define DDF_Spare 8 /* overrides Global_spare */
254 #define DDF_Foreign 16
255 #define DDF_Legacy 32 /* no DDF on this device */
257 #define DDF_Interface_mask 0xf00
258 #define DDF_Interface_SCSI 0x100
259 #define DDF_Interface_SAS 0x200
260 #define DDF_Interface_SATA 0x300
261 #define DDF_Interface_FC 0x400
263 /* phys_disk_entry.state is a bigendian bitmap */
265 #define DDF_Failed 2 /* overrides 1,4,8 */
266 #define DDF_Rebuilding 4
267 #define DDF_Transition 8
269 #define DDF_ReadErrors 32
270 #define DDF_Missing 64
272 /* The content of the virt_section global scope */
273 struct virtual_disk
{
274 be32 magic
; /* DDF_VIRT_RECORDS_MAGIC */
279 struct virtual_entry
{
280 char guid
[DDF_GUID_LEN
];
282 __u16 pad0
; /* 0xffff */
292 /* virtual_entry.type is a bitmap - bigendian */
294 #define DDF_Enforce_Groups 2
295 #define DDF_Unicode 4
296 #define DDF_Owner_Valid 8
298 /* virtual_entry.state is a bigendian bitmap */
299 #define DDF_state_mask 0x7
300 #define DDF_state_optimal 0x0
301 #define DDF_state_degraded 0x1
302 #define DDF_state_deleted 0x2
303 #define DDF_state_missing 0x3
304 #define DDF_state_failed 0x4
305 #define DDF_state_part_optimal 0x5
307 #define DDF_state_morphing 0x8
308 #define DDF_state_inconsistent 0x10
310 /* virtual_entry.init_state is a bigendian bitmap */
311 #define DDF_initstate_mask 0x03
312 #define DDF_init_not 0x00
313 #define DDF_init_quick 0x01 /* initialisation is progress.
314 * i.e. 'state_inconsistent' */
315 #define DDF_init_full 0x02
317 #define DDF_access_mask 0xc0
318 #define DDF_access_rw 0x00
319 #define DDF_access_ro 0x80
320 #define DDF_access_blocked 0xc0
322 /* The content of the config_section - local scope
323 * It has multiple records each config_record_len sectors
324 * They can be vd_config or spare_assign
328 be32 magic
; /* DDF_VD_CONF_MAGIC */
330 char guid
[DDF_GUID_LEN
];
334 be16 prim_elmnt_count
;
335 __u8 chunk_shift
; /* 0 == 512, 1==1024 etc */
338 __u8 sec_elmnt_count
;
341 be64 blocks
; /* blocks per component could be different
342 * on different component devices...(only
343 * for concat I hope) */
344 be64 array_blocks
; /* blocks in array */
346 be32 spare_refs
[8]; /* This is used to detect missing spares.
347 * As we don't have an interface for that
348 * the values are ignored.
355 __u8 v0
[32]; /* reserved- 0xff */
356 __u8 v1
[32]; /* reserved- 0xff */
357 __u8 v2
[16]; /* reserved- 0xff */
358 __u8 v3
[16]; /* reserved- 0xff */
360 be32 phys_refnum
[0]; /* refnum of each disk in sequence */
361 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
362 bvd are always the same size */
364 #define LBA_OFFSET(ddf, vd) ((be64 *) &(vd)->phys_refnum[(ddf)->mppe])
366 /* vd_config.cache_pol[7] is a bitmap */
367 #define DDF_cache_writeback 1 /* else writethrough */
368 #define DDF_cache_wadaptive 2 /* only applies if writeback */
369 #define DDF_cache_readahead 4
370 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
371 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
372 #define DDF_cache_wallowed 32 /* enable write caching */
373 #define DDF_cache_rallowed 64 /* enable read caching */
375 struct spare_assign
{
376 be32 magic
; /* DDF_SPARE_ASSIGN_MAGIC */
381 be16 populated
; /* SAEs used */
382 be16 max
; /* max SAEs */
384 struct spare_assign_entry
{
385 char guid
[DDF_GUID_LEN
];
386 be16 secondary_element
;
390 /* spare_assign.type is a bitmap */
391 #define DDF_spare_dedicated 0x1 /* else global */
392 #define DDF_spare_revertible 0x2 /* else committable */
393 #define DDF_spare_active 0x4 /* else not active */
394 #define DDF_spare_affinity 0x8 /* enclosure affinity */
396 /* The data_section contents - local scope */
398 be32 magic
; /* DDF_PHYS_DATA_MAGIC */
400 char guid
[DDF_GUID_LEN
];
401 be32 refnum
; /* crc of some magic drive data ... */
402 __u8 forced_ref
; /* set when above was not result of magic */
403 __u8 forced_guid
; /* set if guid was forced rather than magic */
408 /* bbm_section content */
409 struct bad_block_log
{
416 struct mapped_block
{
417 be64 defective_start
;
418 be32 replacement_start
;
424 /* Struct for internally holding ddf structures */
425 /* The DDF structure stored on each device is potentially
426 * quite different, as some data is global and some is local.
427 * The global data is:
430 * - Physical disk records
431 * - Virtual disk records
433 * - Configuration records
434 * - Physical Disk data section
435 * ( and Bad block and vendor which I don't care about yet).
437 * The local data is parsed into separate lists as it is read
438 * and reconstructed for writing. This means that we only need
439 * to make config changes once and they are automatically
440 * propagated to all devices.
441 * The global (config and disk data) records are each in a list
442 * of separate data structures. When writing we find the entry
443 * or entries applicable to the particular device.
446 struct ddf_header anchor
, primary
, secondary
;
447 struct ddf_controller_data controller
;
448 struct ddf_header
*active
;
449 struct phys_disk
*phys
;
450 struct virtual_disk
*virt
;
453 unsigned int max_part
, mppe
, conf_rec_len
;
461 unsigned int vcnum
; /* index into ->virt */
462 /* For an array with a secondary level there are
463 * multiple vd_config structures, all with the same
464 * guid but with different sec_elmnt_seq.
465 * One of these structures is in 'conf' below.
466 * The others are in other_bvds, not in any
469 struct vd_config
**other_bvds
;
470 __u64
*block_sizes
; /* NULL if all the same */
473 struct vd_config conf
;
474 } *conflist
, *currentconf
;
483 unsigned long long size
; /* sectors */
484 be64 primary_lba
; /* sectors */
485 be64 secondary_lba
; /* sectors */
486 be64 workspace_lba
; /* sectors */
487 int pdnum
; /* index in ->phys */
488 struct spare_assign
*spare
;
489 void *mdupdate
; /* hold metadata update */
491 /* These fields used by auto-layout */
492 int raiddisk
; /* slot to fill in autolayout */
497 struct disk_data disk
;
498 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
,
522 static void free_super_ddf(struct supertype
*st
);
523 static int all_ff(const char *guid
);
524 static unsigned int get_pd_index_from_refnum(const struct vcl
*vc
,
525 be32 refnum
, unsigned int nmax
,
526 const struct vd_config
**bvd
,
528 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
, char *map
);
529 static void uuid_from_ddf_guid(const char *guid
, int uuid
[4]);
530 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4]);
531 static void _ddf_array_name(char *name
, const struct ddf_super
*ddf
, int i
);
532 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
, char *map
);
533 static int init_super_ddf_bvd(struct supertype
*st
,
534 mdu_array_info_t
*info
,
535 unsigned long long size
,
536 char *name
, char *homehost
,
537 int *uuid
, unsigned long long data_offset
);
540 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
544 static void pr_state(struct ddf_super
*ddf
, const char *msg
)
547 dprintf("%s/%s: ", __func__
, msg
);
548 for (i
= 0; i
< be16_to_cpu(ddf
->active
->max_vd_entries
); i
++) {
549 if (all_ff(ddf
->virt
->entries
[i
].guid
))
551 dprintf("%u(s=%02x i=%02x) ", i
,
552 ddf
->virt
->entries
[i
].state
,
553 ddf
->virt
->entries
[i
].init_state
);
558 static void pr_state(const struct ddf_super
*ddf
, const char *msg
) {}
561 static void _ddf_set_updates_pending(struct ddf_super
*ddf
, struct vd_config
*vc
,
565 vc
->timestamp
= cpu_to_be32(time(0)-DECADE
);
566 vc
->seqnum
= cpu_to_be32(be32_to_cpu(vc
->seqnum
) + 1);
568 if (ddf
->updates_pending
)
570 ddf
->updates_pending
= 1;
571 ddf
->active
->seq
= cpu_to_be32((be32_to_cpu(ddf
->active
->seq
)+1));
575 #define ddf_set_updates_pending(x,v) _ddf_set_updates_pending((x), (v), __func__)
577 static be32
calc_crc(void *buf
, int len
)
579 /* crcs are always at the same place as in the ddf_header */
580 struct ddf_header
*ddf
= buf
;
581 be32 oldcrc
= ddf
->crc
;
583 ddf
->crc
= cpu_to_be32(0xffffffff);
585 newcrc
= crc32(0, buf
, len
);
587 /* The crc is stored (like everything) bigendian, so convert
588 * here for simplicity
590 return cpu_to_be32(newcrc
);
593 #define DDF_INVALID_LEVEL 0xff
594 #define DDF_NO_SECONDARY 0xff
595 static int err_bad_md_layout(const mdu_array_info_t
*array
)
597 pr_err("RAID%d layout %x with %d disks is unsupported for DDF\n",
598 array
->level
, array
->layout
, array
->raid_disks
);
602 static int layout_md2ddf(const mdu_array_info_t
*array
,
603 struct vd_config
*conf
)
605 be16 prim_elmnt_count
= cpu_to_be16(array
->raid_disks
);
606 __u8 prl
= DDF_INVALID_LEVEL
, rlq
= 0;
607 __u8 sec_elmnt_count
= 1;
608 __u8 srl
= DDF_NO_SECONDARY
;
610 switch (array
->level
) {
615 rlq
= DDF_RAID0_SIMPLE
;
619 switch (array
->raid_disks
) {
621 rlq
= DDF_RAID1_SIMPLE
;
624 rlq
= DDF_RAID1_MULTI
;
627 return err_bad_md_layout(array
);
632 if (array
->layout
!= 0)
633 return err_bad_md_layout(array
);
638 switch (array
->layout
) {
639 case ALGORITHM_LEFT_ASYMMETRIC
:
640 rlq
= DDF_RAID5_N_RESTART
;
642 case ALGORITHM_RIGHT_ASYMMETRIC
:
643 rlq
= DDF_RAID5_0_RESTART
;
645 case ALGORITHM_LEFT_SYMMETRIC
:
646 rlq
= DDF_RAID5_N_CONTINUE
;
648 case ALGORITHM_RIGHT_SYMMETRIC
:
649 /* not mentioned in standard */
651 return err_bad_md_layout(array
);
656 switch (array
->layout
) {
657 case ALGORITHM_ROTATING_N_RESTART
:
658 rlq
= DDF_RAID5_N_RESTART
;
660 case ALGORITHM_ROTATING_ZERO_RESTART
:
661 rlq
= DDF_RAID6_0_RESTART
;
663 case ALGORITHM_ROTATING_N_CONTINUE
:
664 rlq
= DDF_RAID5_N_CONTINUE
;
667 return err_bad_md_layout(array
);
672 if (array
->raid_disks
% 2 == 0 && array
->layout
== 0x102) {
673 rlq
= DDF_RAID1_SIMPLE
;
674 prim_elmnt_count
= cpu_to_be16(2);
675 sec_elmnt_count
= array
->raid_disks
/ 2;
678 } else if (array
->raid_disks
% 3 == 0
679 && array
->layout
== 0x103) {
680 rlq
= DDF_RAID1_MULTI
;
681 prim_elmnt_count
= cpu_to_be16(3);
682 sec_elmnt_count
= array
->raid_disks
/ 3;
685 } else if (array
->layout
== 0x201) {
687 rlq
= DDF_RAID1E_OFFSET
;
688 } else if (array
->layout
== 0x102) {
690 rlq
= DDF_RAID1E_ADJACENT
;
692 return err_bad_md_layout(array
);
695 return err_bad_md_layout(array
);
698 conf
->prim_elmnt_count
= prim_elmnt_count
;
701 conf
->sec_elmnt_count
= sec_elmnt_count
;
705 static int err_bad_ddf_layout(const struct vd_config
*conf
)
707 pr_err("DDF RAID %u qualifier %u with %u disks is unsupported\n",
708 conf
->prl
, conf
->rlq
, be16_to_cpu(conf
->prim_elmnt_count
));
712 static int layout_ddf2md(const struct vd_config
*conf
,
713 mdu_array_info_t
*array
)
715 int level
= LEVEL_UNSUPPORTED
;
717 int raiddisks
= be16_to_cpu(conf
->prim_elmnt_count
);
719 if (conf
->sec_elmnt_count
> 1) {
720 /* see also check_secondary() */
721 if (conf
->prl
!= DDF_RAID1
||
722 (conf
->srl
!= DDF_2STRIPED
&& conf
->srl
!= DDF_2SPANNED
)) {
723 pr_err("Unsupported secondary RAID level %u/%u\n",
724 conf
->prl
, conf
->srl
);
727 if (raiddisks
== 2 && conf
->rlq
== DDF_RAID1_SIMPLE
)
729 else if (raiddisks
== 3 && conf
->rlq
== DDF_RAID1_MULTI
)
732 return err_bad_ddf_layout(conf
);
733 raiddisks
*= conf
->sec_elmnt_count
;
740 level
= LEVEL_LINEAR
;
743 if (conf
->rlq
!= DDF_RAID0_SIMPLE
)
744 return err_bad_ddf_layout(conf
);
748 if (!((conf
->rlq
== DDF_RAID1_SIMPLE
&& raiddisks
== 2) ||
749 (conf
->rlq
== DDF_RAID1_MULTI
&& raiddisks
== 3)))
750 return err_bad_ddf_layout(conf
);
754 if (conf
->rlq
== DDF_RAID1E_ADJACENT
)
756 else if (conf
->rlq
== DDF_RAID1E_OFFSET
)
759 return err_bad_ddf_layout(conf
);
763 if (conf
->rlq
!= DDF_RAID4_N
)
764 return err_bad_ddf_layout(conf
);
769 case DDF_RAID5_N_RESTART
:
770 layout
= ALGORITHM_LEFT_ASYMMETRIC
;
772 case DDF_RAID5_0_RESTART
:
773 layout
= ALGORITHM_RIGHT_ASYMMETRIC
;
775 case DDF_RAID5_N_CONTINUE
:
776 layout
= ALGORITHM_LEFT_SYMMETRIC
;
779 return err_bad_ddf_layout(conf
);
785 case DDF_RAID5_N_RESTART
:
786 layout
= ALGORITHM_ROTATING_N_RESTART
;
788 case DDF_RAID6_0_RESTART
:
789 layout
= ALGORITHM_ROTATING_ZERO_RESTART
;
791 case DDF_RAID5_N_CONTINUE
:
792 layout
= ALGORITHM_ROTATING_N_CONTINUE
;
795 return err_bad_ddf_layout(conf
);
800 return err_bad_ddf_layout(conf
);
804 array
->level
= level
;
805 array
->layout
= layout
;
806 array
->raid_disks
= raiddisks
;
810 static int load_ddf_header(int fd
, unsigned long long lba
,
811 unsigned long long size
,
813 struct ddf_header
*hdr
, struct ddf_header
*anchor
)
815 /* read a ddf header (primary or secondary) from fd/lba
816 * and check that it is consistent with anchor
818 * magic, crc, guid, rev, and LBA's header_type, and
819 * everything after header_type must be the same
824 if (lseek64(fd
, lba
<<9, 0) < 0)
827 if (read(fd
, hdr
, 512) != 512)
830 if (!be32_eq(hdr
->magic
, DDF_HEADER_MAGIC
)) {
831 pr_err("%s: bad header magic\n", __func__
);
834 if (!be32_eq(calc_crc(hdr
, 512), hdr
->crc
)) {
835 pr_err("%s: bad CRC\n", __func__
);
838 if (memcmp(anchor
->guid
, hdr
->guid
, DDF_GUID_LEN
) != 0 ||
839 memcmp(anchor
->revision
, hdr
->revision
, 8) != 0 ||
840 !be64_eq(anchor
->primary_lba
, hdr
->primary_lba
) ||
841 !be64_eq(anchor
->secondary_lba
, hdr
->secondary_lba
) ||
843 memcmp(anchor
->pad2
, hdr
->pad2
, 512 -
844 offsetof(struct ddf_header
, pad2
)) != 0) {
845 pr_err("%s: header mismatch\n", __func__
);
849 /* Looks good enough to me... */
853 static void *load_section(int fd
, struct ddf_super
*super
, void *buf
,
854 be32 offset_be
, be32 len_be
, int check
)
856 unsigned long long offset
= be32_to_cpu(offset_be
);
857 unsigned long long len
= be32_to_cpu(len_be
);
858 int dofree
= (buf
== NULL
);
861 if (len
!= 2 && len
!= 8 && len
!= 32
862 && len
!= 128 && len
!= 512)
867 if (!buf
&& posix_memalign(&buf
, 512, len
<<9) != 0)
873 if (super
->active
->type
== 1)
874 offset
+= be64_to_cpu(super
->active
->primary_lba
);
876 offset
+= be64_to_cpu(super
->active
->secondary_lba
);
878 if ((unsigned long long)lseek64(fd
, offset
<<9, 0) != (offset
<<9)) {
883 if ((unsigned long long)read(fd
, buf
, len
<<9) != (len
<<9)) {
891 static int load_ddf_headers(int fd
, struct ddf_super
*super
, char *devname
)
893 unsigned long long dsize
;
895 get_dev_size(fd
, NULL
, &dsize
);
897 if (lseek64(fd
, dsize
-512, 0) < 0) {
899 pr_err("Cannot seek to anchor block on %s: %s\n",
900 devname
, strerror(errno
));
903 if (read(fd
, &super
->anchor
, 512) != 512) {
905 pr_err("Cannot read anchor block on %s: %s\n",
906 devname
, strerror(errno
));
909 if (!be32_eq(super
->anchor
.magic
, DDF_HEADER_MAGIC
)) {
911 pr_err("no DDF anchor found on %s\n",
915 if (!be32_eq(calc_crc(&super
->anchor
, 512), super
->anchor
.crc
)) {
917 pr_err("bad CRC on anchor on %s\n",
921 if (memcmp(super
->anchor
.revision
, DDF_REVISION_0
, 8) != 0 &&
922 memcmp(super
->anchor
.revision
, DDF_REVISION_2
, 8) != 0) {
924 pr_err("can only support super revision"
925 " %.8s and earlier, not %.8s on %s\n",
926 DDF_REVISION_2
, super
->anchor
.revision
,devname
);
929 super
->active
= NULL
;
930 if (load_ddf_header(fd
, be64_to_cpu(super
->anchor
.primary_lba
),
932 &super
->primary
, &super
->anchor
) == 0) {
934 pr_err("Failed to load primary DDF header "
937 super
->active
= &super
->primary
;
939 if (load_ddf_header(fd
, be64_to_cpu(super
->anchor
.secondary_lba
),
941 &super
->secondary
, &super
->anchor
)) {
942 if (super
->active
== NULL
943 || (be32_to_cpu(super
->primary
.seq
)
944 < be32_to_cpu(super
->secondary
.seq
) &&
945 !super
->secondary
.openflag
)
946 || (be32_to_cpu(super
->primary
.seq
)
947 == be32_to_cpu(super
->secondary
.seq
) &&
948 super
->primary
.openflag
&& !super
->secondary
.openflag
)
950 super
->active
= &super
->secondary
;
951 } else if (devname
&&
952 be64_to_cpu(super
->anchor
.secondary_lba
) != ~(__u64
)0)
953 pr_err("Failed to load secondary DDF header on %s\n",
955 if (super
->active
== NULL
)
960 static int load_ddf_global(int fd
, struct ddf_super
*super
, char *devname
)
963 ok
= load_section(fd
, super
, &super
->controller
,
964 super
->active
->controller_section_offset
,
965 super
->active
->controller_section_length
,
967 super
->phys
= load_section(fd
, super
, NULL
,
968 super
->active
->phys_section_offset
,
969 super
->active
->phys_section_length
,
971 super
->pdsize
= be32_to_cpu(super
->active
->phys_section_length
) * 512;
973 super
->virt
= load_section(fd
, super
, NULL
,
974 super
->active
->virt_section_offset
,
975 super
->active
->virt_section_length
,
977 super
->vdsize
= be32_to_cpu(super
->active
->virt_section_length
) * 512;
987 super
->conflist
= NULL
;
990 super
->max_part
= be16_to_cpu(super
->active
->max_partitions
);
991 super
->mppe
= be16_to_cpu(super
->active
->max_primary_element_entries
);
992 super
->conf_rec_len
= be16_to_cpu(super
->active
->config_record_len
);
996 #define DDF_UNUSED_BVD 0xff
997 static int alloc_other_bvds(const struct ddf_super
*ddf
, struct vcl
*vcl
)
999 unsigned int n_vds
= vcl
->conf
.sec_elmnt_count
- 1;
1000 unsigned int i
, vdsize
;
1003 vcl
->other_bvds
= NULL
;
1006 vdsize
= ddf
->conf_rec_len
* 512;
1007 if (posix_memalign(&p
, 512, n_vds
*
1008 (vdsize
+ sizeof(struct vd_config
*))) != 0)
1010 vcl
->other_bvds
= (struct vd_config
**) (p
+ n_vds
* vdsize
);
1011 for (i
= 0; i
< n_vds
; i
++) {
1012 vcl
->other_bvds
[i
] = p
+ i
* vdsize
;
1013 memset(vcl
->other_bvds
[i
], 0, vdsize
);
1014 vcl
->other_bvds
[i
]->sec_elmnt_seq
= DDF_UNUSED_BVD
;
1019 static void add_other_bvd(struct vcl
*vcl
, struct vd_config
*vd
,
1023 for (i
= 0; i
< vcl
->conf
.sec_elmnt_count
-1; i
++)
1024 if (vcl
->other_bvds
[i
]->sec_elmnt_seq
== vd
->sec_elmnt_seq
)
1027 if (i
< vcl
->conf
.sec_elmnt_count
-1) {
1028 if (be32_to_cpu(vd
->seqnum
) <=
1029 be32_to_cpu(vcl
->other_bvds
[i
]->seqnum
))
1032 for (i
= 0; i
< vcl
->conf
.sec_elmnt_count
-1; i
++)
1033 if (vcl
->other_bvds
[i
]->sec_elmnt_seq
== DDF_UNUSED_BVD
)
1035 if (i
== vcl
->conf
.sec_elmnt_count
-1) {
1036 pr_err("no space for sec level config %u, count is %u\n",
1037 vd
->sec_elmnt_seq
, vcl
->conf
.sec_elmnt_count
);
1041 memcpy(vcl
->other_bvds
[i
], vd
, len
);
1044 static int load_ddf_local(int fd
, struct ddf_super
*super
,
1045 char *devname
, int keep
)
1051 unsigned int confsec
;
1053 unsigned int max_virt_disks
=
1054 be16_to_cpu(super
->active
->max_vd_entries
);
1055 unsigned long long dsize
;
1057 /* First the local disk info */
1058 if (posix_memalign((void**)&dl
, 512,
1060 (super
->max_part
) * sizeof(dl
->vlist
[0])) != 0) {
1061 pr_err("%s could not allocate disk info buffer\n",
1066 load_section(fd
, super
, &dl
->disk
,
1067 super
->active
->data_section_offset
,
1068 super
->active
->data_section_length
,
1070 dl
->devname
= devname
? xstrdup(devname
) : NULL
;
1073 dl
->major
= major(stb
.st_rdev
);
1074 dl
->minor
= minor(stb
.st_rdev
);
1075 dl
->next
= super
->dlist
;
1076 dl
->fd
= keep
? fd
: -1;
1079 if (get_dev_size(fd
, devname
, &dsize
))
1080 dl
->size
= dsize
>> 9;
1081 /* If the disks have different sizes, the LBAs will differ
1082 * between phys disks.
1083 * At this point here, the values in super->active must be valid
1084 * for this phys disk. */
1085 dl
->primary_lba
= super
->active
->primary_lba
;
1086 dl
->secondary_lba
= super
->active
->secondary_lba
;
1087 dl
->workspace_lba
= super
->active
->workspace_lba
;
1089 for (i
= 0 ; i
< super
->max_part
; i
++)
1090 dl
->vlist
[i
] = NULL
;
1093 for (i
= 0; i
< be16_to_cpu(super
->active
->max_pd_entries
); i
++)
1094 if (memcmp(super
->phys
->entries
[i
].guid
,
1095 dl
->disk
.guid
, DDF_GUID_LEN
) == 0)
1098 /* Now the config list. */
1099 /* 'conf' is an array of config entries, some of which are
1100 * probably invalid. Those which are good need to be copied into
1104 conf
= load_section(fd
, super
, super
->conf
,
1105 super
->active
->config_section_offset
,
1106 super
->active
->config_section_length
,
1111 confsec
< be32_to_cpu(super
->active
->config_section_length
);
1112 confsec
+= super
->conf_rec_len
) {
1113 struct vd_config
*vd
=
1114 (struct vd_config
*)((char*)conf
+ confsec
*512);
1117 if (be32_eq(vd
->magic
, DDF_SPARE_ASSIGN_MAGIC
)) {
1120 if (posix_memalign((void**)&dl
->spare
, 512,
1121 super
->conf_rec_len
*512) != 0) {
1122 pr_err("%s could not allocate spare info buf\n",
1127 memcpy(dl
->spare
, vd
, super
->conf_rec_len
*512);
1130 if (!be32_eq(vd
->magic
, DDF_VD_CONF_MAGIC
))
1131 /* Must be vendor-unique - I cannot handle those */
1134 for (vcl
= super
->conflist
; vcl
; vcl
= vcl
->next
) {
1135 if (memcmp(vcl
->conf
.guid
,
1136 vd
->guid
, DDF_GUID_LEN
) == 0)
1141 dl
->vlist
[vnum
++] = vcl
;
1142 if (vcl
->other_bvds
!= NULL
&&
1143 vcl
->conf
.sec_elmnt_seq
!= vd
->sec_elmnt_seq
) {
1144 add_other_bvd(vcl
, vd
, super
->conf_rec_len
*512);
1147 if (be32_to_cpu(vd
->seqnum
) <=
1148 be32_to_cpu(vcl
->conf
.seqnum
))
1151 if (posix_memalign((void**)&vcl
, 512,
1152 (super
->conf_rec_len
*512 +
1153 offsetof(struct vcl
, conf
))) != 0) {
1154 pr_err("%s could not allocate vcl buf\n",
1158 vcl
->next
= super
->conflist
;
1159 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
1160 vcl
->conf
.sec_elmnt_count
= vd
->sec_elmnt_count
;
1161 if (alloc_other_bvds(super
, vcl
) != 0) {
1162 pr_err("%s could not allocate other bvds\n",
1167 super
->conflist
= vcl
;
1168 dl
->vlist
[vnum
++] = vcl
;
1170 memcpy(&vcl
->conf
, vd
, super
->conf_rec_len
*512);
1171 for (i
=0; i
< max_virt_disks
; i
++)
1172 if (memcmp(super
->virt
->entries
[i
].guid
,
1173 vcl
->conf
.guid
, DDF_GUID_LEN
)==0)
1175 if (i
< max_virt_disks
)
1182 static int load_super_ddf(struct supertype
*st
, int fd
,
1185 unsigned long long dsize
;
1186 struct ddf_super
*super
;
1189 if (get_dev_size(fd
, devname
, &dsize
) == 0)
1192 if (test_partition(fd
))
1193 /* DDF is not allowed on partitions */
1196 /* 32M is a lower bound */
1197 if (dsize
<= 32*1024*1024) {
1199 pr_err("%s is too small for ddf: "
1200 "size is %llu sectors.\n",
1206 pr_err("%s is an odd size for ddf: "
1207 "size is %llu bytes.\n",
1214 if (posix_memalign((void**)&super
, 512, sizeof(*super
))!= 0) {
1215 pr_err("malloc of %zu failed.\n",
1219 memset(super
, 0, sizeof(*super
));
1221 rv
= load_ddf_headers(fd
, super
, devname
);
1227 /* Have valid headers and have chosen the best. Let's read in the rest*/
1229 rv
= load_ddf_global(fd
, super
, devname
);
1233 pr_err("Failed to load all information "
1234 "sections on %s\n", devname
);
1239 rv
= load_ddf_local(fd
, super
, devname
, 0);
1243 pr_err("Failed to load all information "
1244 "sections on %s\n", devname
);
1249 /* Should possibly check the sections .... */
1252 if (st
->ss
== NULL
) {
1253 st
->ss
= &super_ddf
;
1254 st
->minor_version
= 0;
1261 static void free_super_ddf(struct supertype
*st
)
1263 struct ddf_super
*ddf
= st
->sb
;
1269 while (ddf
->conflist
) {
1270 struct vcl
*v
= ddf
->conflist
;
1271 ddf
->conflist
= v
->next
;
1273 free(v
->block_sizes
);
1276 v->other_bvds[0] points to beginning of buffer,
1277 see alloc_other_bvds()
1279 free(v
->other_bvds
[0]);
1282 while (ddf
->dlist
) {
1283 struct dl
*d
= ddf
->dlist
;
1284 ddf
->dlist
= d
->next
;
1291 while (ddf
->add_list
) {
1292 struct dl
*d
= ddf
->add_list
;
1293 ddf
->add_list
= d
->next
;
1304 static struct supertype
*match_metadata_desc_ddf(char *arg
)
1306 /* 'ddf' only supports containers */
1307 struct supertype
*st
;
1308 if (strcmp(arg
, "ddf") != 0 &&
1309 strcmp(arg
, "default") != 0
1313 st
= xcalloc(1, sizeof(*st
));
1314 st
->ss
= &super_ddf
;
1316 st
->minor_version
= 0;
1323 static mapping_t ddf_state
[] = {
1329 { "Partially Optimal", 5},
1335 static mapping_t ddf_init_state
[] = {
1336 { "Not Initialised", 0},
1337 { "QuickInit in Progress", 1},
1338 { "Fully Initialised", 2},
1342 static mapping_t ddf_access
[] = {
1346 { "Blocked (no access)", 3},
1350 static mapping_t ddf_level
[] = {
1351 { "RAID0", DDF_RAID0
},
1352 { "RAID1", DDF_RAID1
},
1353 { "RAID3", DDF_RAID3
},
1354 { "RAID4", DDF_RAID4
},
1355 { "RAID5", DDF_RAID5
},
1356 { "RAID1E",DDF_RAID1E
},
1357 { "JBOD", DDF_JBOD
},
1358 { "CONCAT",DDF_CONCAT
},
1359 { "RAID5E",DDF_RAID5E
},
1360 { "RAID5EE",DDF_RAID5EE
},
1361 { "RAID6", DDF_RAID6
},
1364 static mapping_t ddf_sec_level
[] = {
1365 { "Striped", DDF_2STRIPED
},
1366 { "Mirrored", DDF_2MIRRORED
},
1367 { "Concat", DDF_2CONCAT
},
1368 { "Spanned", DDF_2SPANNED
},
1373 static int all_ff(const char *guid
)
1376 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
1377 if (guid
[i
] != (char)0xff)
1382 static const char *guid_str(const char *guid
)
1384 static char buf
[DDF_GUID_LEN
*2+1];
1387 for (i
= 0; i
< DDF_GUID_LEN
; i
++) {
1388 unsigned char c
= guid
[i
];
1389 if (c
>= 32 && c
< 127)
1390 p
+= sprintf(p
, "%c", c
);
1392 p
+= sprintf(p
, "%02x", c
);
1395 return (const char *) buf
;
1399 static void print_guid(char *guid
, int tstamp
)
1401 /* A GUIDs are part (or all) ASCII and part binary.
1402 * They tend to be space padded.
1403 * We print the GUID in HEX, then in parentheses add
1404 * any initial ASCII sequence, and a possible
1405 * time stamp from bytes 16-19
1407 int l
= DDF_GUID_LEN
;
1410 for (i
=0 ; i
<DDF_GUID_LEN
; i
++) {
1411 if ((i
&3)==0 && i
!= 0) printf(":");
1412 printf("%02X", guid
[i
]&255);
1416 while (l
&& guid
[l
-1] == ' ')
1418 for (i
=0 ; i
<l
; i
++) {
1419 if (guid
[i
] >= 0x20 && guid
[i
] < 0x7f)
1420 fputc(guid
[i
], stdout
);
1425 time_t then
= __be32_to_cpu(*(__u32
*)(guid
+16)) + DECADE
;
1428 tm
= localtime(&then
);
1429 strftime(tbuf
, 100, " %D %T",tm
);
1430 fputs(tbuf
, stdout
);
1435 static void examine_vd(int n
, struct ddf_super
*sb
, char *guid
)
1437 int crl
= sb
->conf_rec_len
;
1440 for (vcl
= sb
->conflist
; vcl
; vcl
= vcl
->next
) {
1442 struct vd_config
*vc
= &vcl
->conf
;
1444 if (!be32_eq(calc_crc(vc
, crl
*512), vc
->crc
))
1446 if (memcmp(vc
->guid
, guid
, DDF_GUID_LEN
) != 0)
1449 /* Ok, we know about this VD, let's give more details */
1450 printf(" Raid Devices[%d] : %d (", n
,
1451 be16_to_cpu(vc
->prim_elmnt_count
));
1452 for (i
= 0; i
< be16_to_cpu(vc
->prim_elmnt_count
); i
++) {
1454 int cnt
= be16_to_cpu(sb
->phys
->max_pdes
);
1455 for (j
=0; j
<cnt
; j
++)
1456 if (be32_eq(vc
->phys_refnum
[i
],
1457 sb
->phys
->entries
[j
].refnum
))
1464 printf("@%lluK", (unsigned long long) be64_to_cpu(LBA_OFFSET(sb
, vc
)[i
])/2);
1467 if (vc
->chunk_shift
!= 255)
1468 printf(" Chunk Size[%d] : %d sectors\n", n
,
1469 1 << vc
->chunk_shift
);
1470 printf(" Raid Level[%d] : %s\n", n
,
1471 map_num(ddf_level
, vc
->prl
)?:"-unknown-");
1472 if (vc
->sec_elmnt_count
!= 1) {
1473 printf(" Secondary Position[%d] : %d of %d\n", n
,
1474 vc
->sec_elmnt_seq
, vc
->sec_elmnt_count
);
1475 printf(" Secondary Level[%d] : %s\n", n
,
1476 map_num(ddf_sec_level
, vc
->srl
) ?: "-unknown-");
1478 printf(" Device Size[%d] : %llu\n", n
,
1479 be64_to_cpu(vc
->blocks
)/2);
1480 printf(" Array Size[%d] : %llu\n", n
,
1481 be64_to_cpu(vc
->array_blocks
)/2);
1485 static void examine_vds(struct ddf_super
*sb
)
1487 int cnt
= be16_to_cpu(sb
->virt
->populated_vdes
);
1489 printf(" Virtual Disks : %d\n", cnt
);
1491 for (i
= 0; i
< be16_to_cpu(sb
->virt
->max_vdes
); i
++) {
1492 struct virtual_entry
*ve
= &sb
->virt
->entries
[i
];
1493 if (all_ff(ve
->guid
))
1496 printf(" VD GUID[%d] : ", i
); print_guid(ve
->guid
, 1);
1498 printf(" unit[%d] : %d\n", i
, be16_to_cpu(ve
->unit
));
1499 printf(" state[%d] : %s, %s%s\n", i
,
1500 map_num(ddf_state
, ve
->state
& 7),
1501 (ve
->state
& DDF_state_morphing
) ? "Morphing, ": "",
1502 (ve
->state
& DDF_state_inconsistent
)? "Not Consistent" : "Consistent");
1503 printf(" init state[%d] : %s\n", i
,
1504 map_num(ddf_init_state
, ve
->init_state
&DDF_initstate_mask
));
1505 printf(" access[%d] : %s\n", i
,
1506 map_num(ddf_access
, (ve
->init_state
& DDF_access_mask
) >> 6));
1507 printf(" Name[%d] : %.16s\n", i
, ve
->name
);
1508 examine_vd(i
, sb
, ve
->guid
);
1510 if (cnt
) printf("\n");
1513 static void examine_pds(struct ddf_super
*sb
)
1515 int cnt
= be16_to_cpu(sb
->phys
->max_pdes
);
1519 printf(" Physical Disks : %d\n", cnt
);
1520 printf(" Number RefNo Size Device Type/State\n");
1522 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
)
1525 for (i
=0 ; i
<cnt
; i
++) {
1526 struct phys_disk_entry
*pd
= &sb
->phys
->entries
[i
];
1527 int type
= be16_to_cpu(pd
->type
);
1528 int state
= be16_to_cpu(pd
->state
);
1530 if (be32_to_cpu(pd
->refnum
) == 0xffffffff)
1533 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1535 printf(" %3d %08x ", i
,
1536 be32_to_cpu(pd
->refnum
));
1538 be64_to_cpu(pd
->config_size
)>>1);
1539 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1540 if (be32_eq(dl
->disk
.refnum
, pd
->refnum
)) {
1541 char *dv
= map_dev(dl
->major
, dl
->minor
, 0);
1543 printf("%-15s", dv
);
1552 printf(" %s%s%s%s%s",
1553 (type
&2) ? "active":"",
1554 (type
&4) ? "Global-Spare":"",
1555 (type
&8) ? "spare" : "",
1556 (type
&16)? ", foreign" : "",
1557 (type
&32)? "pass-through" : "");
1558 if (state
& DDF_Failed
)
1559 /* This over-rides these three */
1560 state
&= ~(DDF_Online
|DDF_Rebuilding
|DDF_Transition
);
1561 printf("/%s%s%s%s%s%s%s",
1562 (state
&1)? "Online": "Offline",
1563 (state
&2)? ", Failed": "",
1564 (state
&4)? ", Rebuilding": "",
1565 (state
&8)? ", in-transition": "",
1566 (state
&16)? ", SMART-errors": "",
1567 (state
&32)? ", Unrecovered-Read-Errors": "",
1568 (state
&64)? ", Missing" : "");
1571 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1576 printf(" Physical disks not in metadata!:\n");
1578 dv
= map_dev(dl
->major
, dl
->minor
, 0);
1579 printf(" %08x %s\n", be32_to_cpu(dl
->disk
.refnum
),
1580 dv
? dv
: "-unknown-");
1586 static void examine_super_ddf(struct supertype
*st
, char *homehost
)
1588 struct ddf_super
*sb
= st
->sb
;
1590 printf(" Magic : %08x\n", be32_to_cpu(sb
->anchor
.magic
));
1591 printf(" Version : %.8s\n", sb
->anchor
.revision
);
1592 printf("Controller GUID : "); print_guid(sb
->controller
.guid
, 0);
1594 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1596 printf(" Seq : %08x\n", be32_to_cpu(sb
->active
->seq
));
1597 printf(" Redundant hdr : %s\n", (be32_eq(sb
->secondary
.magic
,
1604 static unsigned int get_vd_num_of_subarray(struct supertype
*st
)
1607 * Figure out the VD number for this supertype.
1608 * Returns DDF_CONTAINER for the container itself,
1609 * and DDF_NOTFOUND on error.
1611 struct ddf_super
*ddf
= st
->sb
;
1616 if (*st
->container_devnm
== '\0')
1617 return DDF_CONTAINER
;
1619 sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
1620 if (!sra
|| sra
->array
.major_version
!= -1 ||
1621 sra
->array
.minor_version
!= -2 ||
1622 !is_subarray(sra
->text_version
))
1623 return DDF_NOTFOUND
;
1625 sub
= strchr(sra
->text_version
+ 1, '/');
1627 vcnum
= strtoul(sub
+ 1, &end
, 10);
1628 if (sub
== NULL
|| *sub
== '\0' || *end
!= '\0' ||
1629 vcnum
>= be16_to_cpu(ddf
->active
->max_vd_entries
))
1630 return DDF_NOTFOUND
;
1635 static void brief_examine_super_ddf(struct supertype
*st
, int verbose
)
1637 /* We just write a generic DDF ARRAY entry
1641 getinfo_super_ddf(st
, &info
, NULL
);
1642 fname_from_uuid(st
, &info
, nbuf
, ':');
1644 printf("ARRAY metadata=ddf UUID=%s\n", nbuf
+ 5);
1647 static void brief_examine_subarrays_ddf(struct supertype
*st
, int verbose
)
1649 /* We write a DDF ARRAY member entry for each vd, identifying container
1650 * by uuid and member by unit number and uuid.
1652 struct ddf_super
*ddf
= st
->sb
;
1656 getinfo_super_ddf(st
, &info
, NULL
);
1657 fname_from_uuid(st
, &info
, nbuf
, ':');
1659 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
1660 struct virtual_entry
*ve
= &ddf
->virt
->entries
[i
];
1664 if (all_ff(ve
->guid
))
1666 memcpy(vcl
.conf
.guid
, ve
->guid
, DDF_GUID_LEN
);
1667 ddf
->currentconf
=&vcl
;
1669 uuid_from_super_ddf(st
, info
.uuid
);
1670 fname_from_uuid(st
, &info
, nbuf1
, ':');
1671 _ddf_array_name(namebuf
, ddf
, i
);
1672 printf("ARRAY%s%s container=%s member=%d UUID=%s\n",
1673 namebuf
[0] == '\0' ? "" : " /dev/md/", namebuf
,
1674 nbuf
+5, i
, nbuf1
+5);
1678 static void export_examine_super_ddf(struct supertype
*st
)
1682 getinfo_super_ddf(st
, &info
, NULL
);
1683 fname_from_uuid(st
, &info
, nbuf
, ':');
1684 printf("MD_METADATA=ddf\n");
1685 printf("MD_LEVEL=container\n");
1686 printf("MD_UUID=%s\n", nbuf
+5);
1687 printf("MD_DEVICES=%u\n",
1688 be16_to_cpu(((struct ddf_super
*)st
->sb
)->phys
->used_pdes
));
1691 static int copy_metadata_ddf(struct supertype
*st
, int from
, int to
)
1694 unsigned long long dsize
, offset
;
1696 struct ddf_header
*ddf
;
1699 /* The meta consists of an anchor, a primary, and a secondary.
1700 * This all lives at the end of the device.
1701 * So it is easiest to find the earliest of primary and
1702 * secondary, and copy everything from there.
1704 * Anchor is 512 from end. It contains primary_lba and secondary_lba
1705 * we choose one of those
1708 if (posix_memalign(&buf
, 4096, 4096) != 0)
1711 if (!get_dev_size(from
, NULL
, &dsize
))
1714 if (lseek64(from
, dsize
-512, 0) < 0)
1716 if (read(from
, buf
, 512) != 512)
1719 if (!be32_eq(ddf
->magic
, DDF_HEADER_MAGIC
) ||
1720 !be32_eq(calc_crc(ddf
, 512), ddf
->crc
) ||
1721 (memcmp(ddf
->revision
, DDF_REVISION_0
, 8) != 0 &&
1722 memcmp(ddf
->revision
, DDF_REVISION_2
, 8) != 0))
1725 offset
= dsize
- 512;
1726 if ((be64_to_cpu(ddf
->primary_lba
) << 9) < offset
)
1727 offset
= be64_to_cpu(ddf
->primary_lba
) << 9;
1728 if ((be64_to_cpu(ddf
->secondary_lba
) << 9) < offset
)
1729 offset
= be64_to_cpu(ddf
->secondary_lba
) << 9;
1731 bytes
= dsize
- offset
;
1733 if (lseek64(from
, offset
, 0) < 0 ||
1734 lseek64(to
, offset
, 0) < 0)
1736 while (written
< bytes
) {
1737 int n
= bytes
- written
;
1740 if (read(from
, buf
, n
) != n
)
1742 if (write(to
, buf
, n
) != n
)
1753 static void detail_super_ddf(struct supertype
*st
, char *homehost
)
1755 struct ddf_super
*sb
= st
->sb
;
1756 int cnt
= be16_to_cpu(sb
->virt
->populated_vdes
);
1758 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1760 printf(" Seq : %08x\n", be32_to_cpu(sb
->active
->seq
));
1761 printf(" Virtual Disks : %d\n", cnt
);
1765 static const char *vendors_with_variable_volume_UUID
[] = {
1769 static int volume_id_is_reliable(const struct ddf_super
*ddf
)
1771 int n
= ARRAY_SIZE(vendors_with_variable_volume_UUID
);
1773 for (i
= 0; i
< n
; i
++)
1774 if (!memcmp(ddf
->controller
.guid
,
1775 vendors_with_variable_volume_UUID
[i
], 8))
1780 static void uuid_of_ddf_subarray(const struct ddf_super
*ddf
,
1781 unsigned int vcnum
, int uuid
[4])
1783 char buf
[DDF_GUID_LEN
+18], sha
[20], *p
;
1784 struct sha1_ctx ctx
;
1785 if (volume_id_is_reliable(ddf
)) {
1786 uuid_from_ddf_guid(ddf
->virt
->entries
[vcnum
].guid
, uuid
);
1790 * Some fake RAID BIOSes (in particular, LSI ones) change the
1791 * VD GUID at every boot. These GUIDs are not suitable for
1792 * identifying an array. Luckily the header GUID appears to
1794 * We construct a pseudo-UUID from the header GUID and those
1795 * properties of the subarray that we expect to remain constant.
1797 memset(buf
, 0, sizeof(buf
));
1799 memcpy(p
, ddf
->anchor
.guid
, DDF_GUID_LEN
);
1801 memcpy(p
, ddf
->virt
->entries
[vcnum
].name
, 16);
1803 *((__u16
*) p
) = vcnum
;
1804 sha1_init_ctx(&ctx
);
1805 sha1_process_bytes(buf
, sizeof(buf
), &ctx
);
1806 sha1_finish_ctx(&ctx
, sha
);
1807 memcpy(uuid
, sha
, 4*4);
1810 static void brief_detail_super_ddf(struct supertype
*st
)
1814 struct ddf_super
*ddf
= st
->sb
;
1815 unsigned int vcnum
= get_vd_num_of_subarray(st
);
1816 if (vcnum
== DDF_CONTAINER
)
1817 uuid_from_super_ddf(st
, info
.uuid
);
1818 else if (vcnum
== DDF_NOTFOUND
)
1821 uuid_of_ddf_subarray(ddf
, vcnum
, info
.uuid
);
1822 fname_from_uuid(st
, &info
, nbuf
,':');
1823 printf(" UUID=%s", nbuf
+ 5);
1827 static int match_home_ddf(struct supertype
*st
, char *homehost
)
1829 /* It matches 'this' host if the controller is a
1830 * Linux-MD controller with vendor_data matching
1831 * the hostname. It would be nice if we could
1832 * test against controller found in /sys or somewhere...
1834 struct ddf_super
*ddf
= st
->sb
;
1839 len
= strlen(homehost
);
1841 return (memcmp(ddf
->controller
.guid
, T10
, 8) == 0 &&
1842 len
< sizeof(ddf
->controller
.vendor_data
) &&
1843 memcmp(ddf
->controller
.vendor_data
, homehost
,len
) == 0 &&
1844 ddf
->controller
.vendor_data
[len
] == 0);
1848 static int find_index_in_bvd(const struct ddf_super
*ddf
,
1849 const struct vd_config
*conf
, unsigned int n
,
1850 unsigned int *n_bvd
)
1853 * Find the index of the n-th valid physical disk in this BVD.
1854 * Unused entries can be sprinkled in with the used entries,
1859 i
< ddf
->mppe
&& j
< be16_to_cpu(conf
->prim_elmnt_count
);
1861 if (be32_to_cpu(conf
->phys_refnum
[i
]) != 0xffffffff) {
1869 dprintf("%s: couldn't find BVD member %u (total %u)\n",
1870 __func__
, n
, be16_to_cpu(conf
->prim_elmnt_count
));
1874 /* Given a member array instance number, and a raid disk within that instance,
1875 * find the vd_config structure. The offset of the given disk in the phys_refnum
1876 * table is returned in n_bvd.
1877 * For two-level members with a secondary raid level the vd_config for
1878 * the appropriate BVD is returned.
1879 * The return value is always &vlc->conf, where vlc is returned in last pointer.
1881 static struct vd_config
*find_vdcr(struct ddf_super
*ddf
, unsigned int inst
,
1883 unsigned int *n_bvd
, struct vcl
**vcl
)
1887 for (v
= ddf
->conflist
; v
; v
= v
->next
) {
1888 unsigned int nsec
, ibvd
= 0;
1889 struct vd_config
*conf
;
1890 if (inst
!= v
->vcnum
)
1893 if (conf
->sec_elmnt_count
== 1) {
1894 if (find_index_in_bvd(ddf
, conf
, n
, n_bvd
)) {
1900 if (v
->other_bvds
== NULL
) {
1901 pr_err("%s: BUG: other_bvds is NULL, nsec=%u\n",
1902 __func__
, conf
->sec_elmnt_count
);
1905 nsec
= n
/ be16_to_cpu(conf
->prim_elmnt_count
);
1906 if (conf
->sec_elmnt_seq
!= nsec
) {
1907 for (ibvd
= 1; ibvd
< conf
->sec_elmnt_count
; ibvd
++) {
1908 if (v
->other_bvds
[ibvd
-1]->sec_elmnt_seq
1912 if (ibvd
== conf
->sec_elmnt_count
)
1914 conf
= v
->other_bvds
[ibvd
-1];
1916 if (!find_index_in_bvd(ddf
, conf
,
1917 n
- nsec
*conf
->sec_elmnt_count
, n_bvd
))
1919 dprintf("%s: found disk %u as member %u in bvd %d of array %u\n"
1920 , __func__
, n
, *n_bvd
, ibvd
, inst
);
1925 pr_err("%s: Could't find disk %d in array %u\n", __func__
, n
, inst
);
1930 static int find_phys(const struct ddf_super
*ddf
, be32 phys_refnum
)
1932 /* Find the entry in phys_disk which has the given refnum
1933 * and return it's index
1936 for (i
= 0; i
< be16_to_cpu(ddf
->phys
->max_pdes
); i
++)
1937 if (be32_eq(ddf
->phys
->entries
[i
].refnum
, phys_refnum
))
1942 static void uuid_from_ddf_guid(const char *guid
, int uuid
[4])
1945 struct sha1_ctx ctx
;
1946 sha1_init_ctx(&ctx
);
1947 sha1_process_bytes(guid
, DDF_GUID_LEN
, &ctx
);
1948 sha1_finish_ctx(&ctx
, buf
);
1949 memcpy(uuid
, buf
, 4*4);
1952 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4])
1954 /* The uuid returned here is used for:
1955 * uuid to put into bitmap file (Create, Grow)
1956 * uuid for backup header when saving critical section (Grow)
1957 * comparing uuids when re-adding a device into an array
1958 * In these cases the uuid required is that of the data-array,
1959 * not the device-set.
1960 * uuid to recognise same set when adding a missing device back
1961 * to an array. This is a uuid for the device-set.
1963 * For each of these we can make do with a truncated
1964 * or hashed uuid rather than the original, as long as
1966 * In the case of SVD we assume the BVD is of interest,
1967 * though that might be the case if a bitmap were made for
1968 * a mirrored SVD - worry about that later.
1969 * So we need to find the VD configuration record for the
1970 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1971 * The first 16 bytes of the sha1 of these is used.
1973 struct ddf_super
*ddf
= st
->sb
;
1974 struct vcl
*vcl
= ddf
->currentconf
;
1977 uuid_of_ddf_subarray(ddf
, vcl
->vcnum
, uuid
);
1979 uuid_from_ddf_guid(ddf
->anchor
.guid
, uuid
);
1982 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1984 struct ddf_super
*ddf
= st
->sb
;
1985 int map_disks
= info
->array
.raid_disks
;
1988 if (ddf
->currentconf
) {
1989 getinfo_super_ddf_bvd(st
, info
, map
);
1992 memset(info
, 0, sizeof(*info
));
1994 info
->array
.raid_disks
= be16_to_cpu(ddf
->phys
->used_pdes
);
1995 info
->array
.level
= LEVEL_CONTAINER
;
1996 info
->array
.layout
= 0;
1997 info
->array
.md_minor
= -1;
1998 cptr
= (__u32
*)(ddf
->anchor
.guid
+ 16);
1999 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
2001 info
->array
.chunk_size
= 0;
2002 info
->container_enough
= 1;
2004 info
->disk
.major
= 0;
2005 info
->disk
.minor
= 0;
2007 struct phys_disk_entry
*pde
= NULL
;
2008 info
->disk
.number
= be32_to_cpu(ddf
->dlist
->disk
.refnum
);
2009 info
->disk
.raid_disk
= find_phys(ddf
, ddf
->dlist
->disk
.refnum
);
2011 info
->data_offset
= be64_to_cpu(ddf
->phys
->
2012 entries
[info
->disk
.raid_disk
].
2014 info
->component_size
= ddf
->dlist
->size
- info
->data_offset
;
2015 if (info
->disk
.raid_disk
>= 0)
2016 pde
= ddf
->phys
->entries
+ info
->disk
.raid_disk
;
2018 !(be16_to_cpu(pde
->state
) & DDF_Failed
) &&
2019 !(be16_to_cpu(pde
->state
) & DDF_Missing
))
2020 info
->disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
2022 info
->disk
.state
= 1 << MD_DISK_FAULTY
;
2025 /* There should always be a dlist, but just in case...*/
2026 info
->disk
.number
= -1;
2027 info
->disk
.raid_disk
= -1;
2028 info
->disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
2030 info
->events
= be32_to_cpu(ddf
->active
->seq
);
2031 info
->array
.utime
= DECADE
+ be32_to_cpu(ddf
->active
->timestamp
);
2033 info
->recovery_start
= MaxSector
;
2034 info
->reshape_active
= 0;
2035 info
->recovery_blocked
= 0;
2038 info
->array
.major_version
= -1;
2039 info
->array
.minor_version
= -2;
2040 strcpy(info
->text_version
, "ddf");
2041 info
->safe_mode_delay
= 0;
2043 uuid_from_super_ddf(st
, info
->uuid
);
2047 int max
= be16_to_cpu(ddf
->phys
->max_pdes
);
2048 for (i
= e
= 0 ; i
< map_disks
; i
++, e
++) {
2050 be32_to_cpu(ddf
->phys
->entries
[e
].refnum
) == 0xffffffff)
2052 if (i
< info
->array
.raid_disks
&& e
< max
&&
2053 !(be16_to_cpu(ddf
->phys
->entries
[e
].state
)
2062 /* size of name must be at least 17 bytes! */
2063 static void _ddf_array_name(char *name
, const struct ddf_super
*ddf
, int i
)
2066 memcpy(name
, ddf
->virt
->entries
[i
].name
, 16);
2068 for(j
= 0; j
< 16; j
++)
2073 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
, char *map
)
2075 struct ddf_super
*ddf
= st
->sb
;
2076 struct vcl
*vc
= ddf
->currentconf
;
2077 int cd
= ddf
->currentdev
;
2081 int map_disks
= info
->array
.raid_disks
;
2083 struct vd_config
*conf
;
2085 memset(info
, 0, sizeof(*info
));
2086 if (layout_ddf2md(&vc
->conf
, &info
->array
) == -1)
2088 info
->array
.md_minor
= -1;
2089 cptr
= (__u32
*)(vc
->conf
.guid
+ 16);
2090 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
2091 info
->array
.utime
= DECADE
+ be32_to_cpu(vc
->conf
.timestamp
);
2092 info
->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
2093 info
->custom_array_size
= be64_to_cpu(vc
->conf
.array_blocks
);
2096 n_prim
= be16_to_cpu(conf
->prim_elmnt_count
);
2097 if (conf
->sec_elmnt_count
> 1 && cd
>= n_prim
) {
2098 int ibvd
= cd
/ n_prim
- 1;
2100 conf
= vc
->other_bvds
[ibvd
];
2103 if (cd
>= 0 && (unsigned)cd
< ddf
->mppe
) {
2105 be64_to_cpu(LBA_OFFSET(ddf
, conf
)[cd
]);
2106 if (vc
->block_sizes
)
2107 info
->component_size
= vc
->block_sizes
[cd
];
2109 info
->component_size
= be64_to_cpu(conf
->blocks
);
2111 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2112 if (be32_eq(dl
->disk
.refnum
, conf
->phys_refnum
[cd
]))
2116 info
->disk
.major
= 0;
2117 info
->disk
.minor
= 0;
2118 info
->disk
.state
= 0;
2119 if (dl
&& dl
->pdnum
>= 0) {
2120 info
->disk
.major
= dl
->major
;
2121 info
->disk
.minor
= dl
->minor
;
2122 info
->disk
.raid_disk
= cd
+ conf
->sec_elmnt_seq
2123 * be16_to_cpu(conf
->prim_elmnt_count
);
2124 info
->disk
.number
= dl
->pdnum
;
2125 info
->disk
.state
= 0;
2126 if (info
->disk
.number
>= 0 &&
2127 (be16_to_cpu(ddf
->phys
->entries
[info
->disk
.number
].state
) & DDF_Online
) &&
2128 !(be16_to_cpu(ddf
->phys
->entries
[info
->disk
.number
].state
) & DDF_Failed
))
2129 info
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
2130 info
->events
= be32_to_cpu(ddf
->active
->seq
);
2133 info
->container_member
= ddf
->currentconf
->vcnum
;
2135 info
->recovery_start
= MaxSector
;
2136 info
->resync_start
= 0;
2137 info
->reshape_active
= 0;
2138 info
->recovery_blocked
= 0;
2139 if (!(ddf
->virt
->entries
[info
->container_member
].state
2140 & DDF_state_inconsistent
) &&
2141 (ddf
->virt
->entries
[info
->container_member
].init_state
2142 & DDF_initstate_mask
)
2144 info
->resync_start
= MaxSector
;
2146 uuid_from_super_ddf(st
, info
->uuid
);
2148 info
->array
.major_version
= -1;
2149 info
->array
.minor_version
= -2;
2150 sprintf(info
->text_version
, "/%s/%d",
2151 st
->container_devnm
,
2152 info
->container_member
);
2153 info
->safe_mode_delay
= DDF_SAFE_MODE_DELAY
;
2155 _ddf_array_name(info
->name
, ddf
, info
->container_member
);
2158 for (j
= 0; j
< map_disks
; j
++) {
2160 if (j
< info
->array
.raid_disks
) {
2161 int i
= find_phys(ddf
, vc
->conf
.phys_refnum
[j
]);
2163 (be16_to_cpu(ddf
->phys
->entries
[i
].state
)
2165 !(be16_to_cpu(ddf
->phys
->entries
[i
].state
)
2172 static int update_super_ddf(struct supertype
*st
, struct mdinfo
*info
,
2174 char *devname
, int verbose
,
2175 int uuid_set
, char *homehost
)
2177 /* For 'assemble' and 'force' we need to return non-zero if any
2178 * change was made. For others, the return value is ignored.
2179 * Update options are:
2180 * force-one : This device looks a bit old but needs to be included,
2181 * update age info appropriately.
2182 * assemble: clear any 'faulty' flag to allow this device to
2184 * force-array: Array is degraded but being forced, mark it clean
2185 * if that will be needed to assemble it.
2187 * newdev: not used ????
2188 * grow: Array has gained a new device - this is currently for
2190 * resync: mark as dirty so a resync will happen.
2191 * uuid: Change the uuid of the array to match what is given
2192 * homehost: update the recorded homehost
2193 * name: update the name - preserving the homehost
2194 * _reshape_progress: record new reshape_progress position.
2196 * Following are not relevant for this version:
2197 * sparc2.2 : update from old dodgey metadata
2198 * super-minor: change the preferred_minor number
2199 * summaries: update redundant counters.
2202 // struct ddf_super *ddf = st->sb;
2203 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
2204 // struct virtual_entry *ve = find_ve(ddf);
2206 /* we don't need to handle "force-*" or "assemble" as
2207 * there is no need to 'trick' the kernel. When the metadata is
2208 * first updated to activate the array, all the implied modifications
2212 if (strcmp(update
, "grow") == 0) {
2214 } else if (strcmp(update
, "resync") == 0) {
2215 // info->resync_checkpoint = 0;
2216 } else if (strcmp(update
, "homehost") == 0) {
2217 /* homehost is stored in controller->vendor_data,
2218 * or it is when we are the vendor
2220 // if (info->vendor_is_local)
2221 // strcpy(ddf->controller.vendor_data, homehost);
2223 } else if (strcmp(update
, "name") == 0) {
2224 /* name is stored in virtual_entry->name */
2225 // memset(ve->name, ' ', 16);
2226 // strncpy(ve->name, info->name, 16);
2228 } else if (strcmp(update
, "_reshape_progress") == 0) {
2229 /* We don't support reshape yet */
2230 } else if (strcmp(update
, "assemble") == 0 ) {
2231 /* Do nothing, just succeed */
2236 // update_all_csum(ddf);
2241 static void make_header_guid(char *guid
)
2244 /* Create a DDF Header of Virtual Disk GUID */
2246 /* 24 bytes of fiction required.
2247 * first 8 are a 'vendor-id' - "Linux-MD"
2248 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
2249 * Remaining 8 random number plus timestamp
2251 memcpy(guid
, T10
, sizeof(T10
));
2252 stamp
= cpu_to_be32(0xdeadbeef);
2253 memcpy(guid
+8, &stamp
, 4);
2254 stamp
= cpu_to_be32(0);
2255 memcpy(guid
+12, &stamp
, 4);
2256 stamp
= cpu_to_be32(time(0) - DECADE
);
2257 memcpy(guid
+16, &stamp
, 4);
2258 stamp
._v32
= random32();
2259 memcpy(guid
+20, &stamp
, 4);
2262 static unsigned int find_unused_vde(const struct ddf_super
*ddf
)
2265 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
2266 if (all_ff(ddf
->virt
->entries
[i
].guid
))
2269 return DDF_NOTFOUND
;
2272 static unsigned int find_vde_by_name(const struct ddf_super
*ddf
,
2277 return DDF_NOTFOUND
;
2278 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
2279 if (all_ff(ddf
->virt
->entries
[i
].guid
))
2281 if (!strncmp(name
, ddf
->virt
->entries
[i
].name
,
2282 sizeof(ddf
->virt
->entries
[i
].name
)))
2285 return DDF_NOTFOUND
;
2289 static unsigned int find_vde_by_guid(const struct ddf_super
*ddf
,
2293 if (guid
== NULL
|| all_ff(guid
))
2294 return DDF_NOTFOUND
;
2295 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++)
2296 if (!memcmp(ddf
->virt
->entries
[i
].guid
, guid
, DDF_GUID_LEN
))
2298 return DDF_NOTFOUND
;
2302 static int init_super_ddf(struct supertype
*st
,
2303 mdu_array_info_t
*info
,
2304 unsigned long long size
, char *name
, char *homehost
,
2305 int *uuid
, unsigned long long data_offset
)
2307 /* This is primarily called by Create when creating a new array.
2308 * We will then get add_to_super called for each component, and then
2309 * write_init_super called to write it out to each device.
2310 * For DDF, Create can create on fresh devices or on a pre-existing
2312 * To create on a pre-existing array a different method will be called.
2313 * This one is just for fresh drives.
2315 * We need to create the entire 'ddf' structure which includes:
2316 * DDF headers - these are easy.
2317 * Controller data - a Sector describing this controller .. not that
2318 * this is a controller exactly.
2319 * Physical Disk Record - one entry per device, so
2320 * leave plenty of space.
2321 * Virtual Disk Records - again, just leave plenty of space.
2322 * This just lists VDs, doesn't give details.
2323 * Config records - describe the VDs that use this disk
2324 * DiskData - describes 'this' device.
2325 * BadBlockManagement - empty
2326 * Diag Space - empty
2327 * Vendor Logs - Could we put bitmaps here?
2330 struct ddf_super
*ddf
;
2333 int max_phys_disks
, max_virt_disks
;
2334 unsigned long long sector
;
2338 struct phys_disk
*pd
;
2339 struct virtual_disk
*vd
;
2342 return init_super_ddf_bvd(st
, info
, size
, name
, homehost
, uuid
,
2345 if (posix_memalign((void**)&ddf
, 512, sizeof(*ddf
)) != 0) {
2346 pr_err("%s could not allocate superblock\n", __func__
);
2349 memset(ddf
, 0, sizeof(*ddf
));
2353 /* zeroing superblock */
2357 /* At least 32MB *must* be reserved for the ddf. So let's just
2358 * start 32MB from the end, and put the primary header there.
2359 * Don't do secondary for now.
2360 * We don't know exactly where that will be yet as it could be
2361 * different on each device. So just set up the lengths.
2364 ddf
->anchor
.magic
= DDF_HEADER_MAGIC
;
2365 make_header_guid(ddf
->anchor
.guid
);
2367 memcpy(ddf
->anchor
.revision
, DDF_REVISION_2
, 8);
2368 ddf
->anchor
.seq
= cpu_to_be32(1);
2369 ddf
->anchor
.timestamp
= cpu_to_be32(time(0) - DECADE
);
2370 ddf
->anchor
.openflag
= 0xFF;
2371 ddf
->anchor
.foreignflag
= 0;
2372 ddf
->anchor
.enforcegroups
= 0; /* Is this best?? */
2373 ddf
->anchor
.pad0
= 0xff;
2374 memset(ddf
->anchor
.pad1
, 0xff, 12);
2375 memset(ddf
->anchor
.header_ext
, 0xff, 32);
2376 ddf
->anchor
.primary_lba
= cpu_to_be64(~(__u64
)0);
2377 ddf
->anchor
.secondary_lba
= cpu_to_be64(~(__u64
)0);
2378 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
2379 memset(ddf
->anchor
.pad2
, 0xff, 3);
2380 ddf
->anchor
.workspace_len
= cpu_to_be32(32768); /* Must be reserved */
2381 /* Put this at bottom of 32M reserved.. */
2382 ddf
->anchor
.workspace_lba
= cpu_to_be64(~(__u64
)0);
2383 max_phys_disks
= 1023; /* Should be enough, 4095 is also allowed */
2384 ddf
->anchor
.max_pd_entries
= cpu_to_be16(max_phys_disks
);
2385 max_virt_disks
= 255; /* 15, 63, 255, 1024, 4095 are all allowed */
2386 ddf
->anchor
.max_vd_entries
= cpu_to_be16(max_virt_disks
);
2388 ddf
->anchor
.max_partitions
= cpu_to_be16(ddf
->max_part
);
2389 ddf
->mppe
= 256; /* 16, 64, 256, 1024, 4096 are all allowed */
2390 ddf
->conf_rec_len
= 1 + ROUND_UP(ddf
->mppe
* (4+8), 512)/512;
2391 ddf
->anchor
.config_record_len
= cpu_to_be16(ddf
->conf_rec_len
);
2392 ddf
->anchor
.max_primary_element_entries
= cpu_to_be16(ddf
->mppe
);
2393 memset(ddf
->anchor
.pad3
, 0xff, 54);
2394 /* Controller section is one sector long immediately
2395 * after the ddf header */
2397 ddf
->anchor
.controller_section_offset
= cpu_to_be32(sector
);
2398 ddf
->anchor
.controller_section_length
= cpu_to_be32(1);
2401 /* phys is 8 sectors after that */
2402 pdsize
= ROUND_UP(sizeof(struct phys_disk
) +
2403 sizeof(struct phys_disk_entry
)*max_phys_disks
,
2405 switch(pdsize
/512) {
2406 case 2: case 8: case 32: case 128: case 512: break;
2409 ddf
->anchor
.phys_section_offset
= cpu_to_be32(sector
);
2410 ddf
->anchor
.phys_section_length
=
2411 cpu_to_be32(pdsize
/512); /* max_primary_element_entries/8 */
2412 sector
+= pdsize
/512;
2414 /* virt is another 32 sectors */
2415 vdsize
= ROUND_UP(sizeof(struct virtual_disk
) +
2416 sizeof(struct virtual_entry
) * max_virt_disks
,
2418 switch(vdsize
/512) {
2419 case 2: case 8: case 32: case 128: case 512: break;
2422 ddf
->anchor
.virt_section_offset
= cpu_to_be32(sector
);
2423 ddf
->anchor
.virt_section_length
=
2424 cpu_to_be32(vdsize
/512); /* max_vd_entries/8 */
2425 sector
+= vdsize
/512;
2427 clen
= ddf
->conf_rec_len
* (ddf
->max_part
+1);
2428 ddf
->anchor
.config_section_offset
= cpu_to_be32(sector
);
2429 ddf
->anchor
.config_section_length
= cpu_to_be32(clen
);
2432 ddf
->anchor
.data_section_offset
= cpu_to_be32(sector
);
2433 ddf
->anchor
.data_section_length
= cpu_to_be32(1);
2436 ddf
->anchor
.bbm_section_length
= cpu_to_be32(0);
2437 ddf
->anchor
.bbm_section_offset
= cpu_to_be32(0xFFFFFFFF);
2438 ddf
->anchor
.diag_space_length
= cpu_to_be32(0);
2439 ddf
->anchor
.diag_space_offset
= cpu_to_be32(0xFFFFFFFF);
2440 ddf
->anchor
.vendor_length
= cpu_to_be32(0);
2441 ddf
->anchor
.vendor_offset
= cpu_to_be32(0xFFFFFFFF);
2443 memset(ddf
->anchor
.pad4
, 0xff, 256);
2445 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2446 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2448 ddf
->primary
.openflag
= 1; /* I guess.. */
2449 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
2451 ddf
->secondary
.openflag
= 1; /* I guess.. */
2452 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
2454 ddf
->active
= &ddf
->primary
;
2456 ddf
->controller
.magic
= DDF_CONTROLLER_MAGIC
;
2458 /* 24 more bytes of fiction required.
2459 * first 8 are a 'vendor-id' - "Linux-MD"
2460 * Remaining 16 are serial number.... maybe a hostname would do?
2462 memcpy(ddf
->controller
.guid
, T10
, sizeof(T10
));
2463 gethostname(hostname
, sizeof(hostname
));
2464 hostname
[sizeof(hostname
) - 1] = 0;
2465 hostlen
= strlen(hostname
);
2466 memcpy(ddf
->controller
.guid
+ 24 - hostlen
, hostname
, hostlen
);
2467 for (i
= strlen(T10
) ; i
+hostlen
< 24; i
++)
2468 ddf
->controller
.guid
[i
] = ' ';
2470 ddf
->controller
.type
.vendor_id
= cpu_to_be16(0xDEAD);
2471 ddf
->controller
.type
.device_id
= cpu_to_be16(0xBEEF);
2472 ddf
->controller
.type
.sub_vendor_id
= cpu_to_be16(0);
2473 ddf
->controller
.type
.sub_device_id
= cpu_to_be16(0);
2474 memcpy(ddf
->controller
.product_id
, "What Is My PID??", 16);
2475 memset(ddf
->controller
.pad
, 0xff, 8);
2476 memset(ddf
->controller
.vendor_data
, 0xff, 448);
2477 if (homehost
&& strlen(homehost
) < 440)
2478 strcpy((char*)ddf
->controller
.vendor_data
, homehost
);
2480 if (posix_memalign((void**)&pd
, 512, pdsize
) != 0) {
2481 pr_err("%s could not allocate pd\n", __func__
);
2485 ddf
->pdsize
= pdsize
;
2487 memset(pd
, 0xff, pdsize
);
2488 memset(pd
, 0, sizeof(*pd
));
2489 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2490 pd
->used_pdes
= cpu_to_be16(0);
2491 pd
->max_pdes
= cpu_to_be16(max_phys_disks
);
2492 memset(pd
->pad
, 0xff, 52);
2493 for (i
= 0; i
< max_phys_disks
; i
++)
2494 memset(pd
->entries
[i
].guid
, 0xff, DDF_GUID_LEN
);
2496 if (posix_memalign((void**)&vd
, 512, vdsize
) != 0) {
2497 pr_err("%s could not allocate vd\n", __func__
);
2501 ddf
->vdsize
= vdsize
;
2502 memset(vd
, 0, vdsize
);
2503 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
2504 vd
->populated_vdes
= cpu_to_be16(0);
2505 vd
->max_vdes
= cpu_to_be16(max_virt_disks
);
2506 memset(vd
->pad
, 0xff, 52);
2508 for (i
=0; i
<max_virt_disks
; i
++)
2509 memset(&vd
->entries
[i
], 0xff, sizeof(struct virtual_entry
));
2512 ddf_set_updates_pending(ddf
, NULL
);
2516 static int chunk_to_shift(int chunksize
)
2518 return ffs(chunksize
/512)-1;
2523 unsigned long long start
, size
;
2525 static int cmp_extent(const void *av
, const void *bv
)
2527 const struct extent
*a
= av
;
2528 const struct extent
*b
= bv
;
2529 if (a
->start
< b
->start
)
2531 if (a
->start
> b
->start
)
2536 static struct extent
*get_extents(struct ddf_super
*ddf
, struct dl
*dl
)
2538 /* Find a list of used extents on the given physical device
2539 * (dnum) of the given ddf.
2540 * Return a malloced array of 'struct extent'
2549 state
= be16_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].state
);
2551 if ((state
& (DDF_Online
|DDF_Failed
|DDF_Missing
)) != DDF_Online
)
2554 rv
= xmalloc(sizeof(struct extent
) * (ddf
->max_part
+ 2));
2556 for (i
= 0; i
< ddf
->max_part
; i
++) {
2557 const struct vd_config
*bvd
;
2559 struct vcl
*v
= dl
->vlist
[i
];
2561 get_pd_index_from_refnum(v
, dl
->disk
.refnum
, ddf
->mppe
,
2562 &bvd
, &ibvd
) == DDF_NOTFOUND
)
2564 rv
[n
].start
= be64_to_cpu(LBA_OFFSET(ddf
, bvd
)[ibvd
]);
2565 rv
[n
].size
= be64_to_cpu(bvd
->blocks
);
2568 qsort(rv
, n
, sizeof(*rv
), cmp_extent
);
2570 rv
[n
].start
= be64_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].config_size
);
2575 static unsigned long long find_space(
2576 struct ddf_super
*ddf
, struct dl
*dl
,
2577 unsigned long long data_offset
,
2578 unsigned long long *size
)
2580 /* Find if the requested amount of space is available.
2581 * If it is, return start.
2582 * If not, set *size to largest space.
2583 * If data_offset != INVALID_SECTORS, then the space must start
2586 struct extent
*e
= get_extents(ddf
, dl
);
2588 unsigned long long pos
= 0;
2589 unsigned long long max_size
= 0;
2593 return INVALID_SECTORS
;
2596 unsigned long long esize
= e
[i
].start
- pos
;
2597 if (data_offset
!= INVALID_SECTORS
&&
2598 pos
<= data_offset
&&
2599 e
[i
].start
> data_offset
) {
2601 esize
= e
[i
].start
- pos
;
2603 if (data_offset
!= INVALID_SECTORS
&&
2604 pos
!= data_offset
) {
2608 if (esize
>= *size
) {
2613 if (esize
> max_size
)
2615 pos
= e
[i
].start
+ e
[i
].size
;
2617 } while (e
[i
-1].size
);
2620 return INVALID_SECTORS
;
2624 static int init_super_ddf_bvd(struct supertype
*st
,
2625 mdu_array_info_t
*info
,
2626 unsigned long long size
,
2627 char *name
, char *homehost
,
2628 int *uuid
, unsigned long long data_offset
)
2630 /* We are creating a BVD inside a pre-existing container.
2631 * so st->sb is already set.
2632 * We need to create a new vd_config and a new virtual_entry
2634 struct ddf_super
*ddf
= st
->sb
;
2635 unsigned int venum
, i
;
2636 struct virtual_entry
*ve
;
2638 struct vd_config
*vc
;
2640 if (find_vde_by_name(ddf
, name
) != DDF_NOTFOUND
) {
2641 pr_err("This ddf already has an array called %s\n", name
);
2644 venum
= find_unused_vde(ddf
);
2645 if (venum
== DDF_NOTFOUND
) {
2646 pr_err("Cannot find spare slot for virtual disk\n");
2649 ve
= &ddf
->virt
->entries
[venum
];
2651 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
2652 * timestamp, random number
2654 make_header_guid(ve
->guid
);
2655 ve
->unit
= cpu_to_be16(info
->md_minor
);
2657 ve
->guid_crc
._v16
= crc32(0, (unsigned char *)ddf
->anchor
.guid
,
2659 ve
->type
= cpu_to_be16(0);
2660 ve
->state
= DDF_state_degraded
; /* Will be modified as devices are added */
2661 if (info
->state
& 1) /* clean */
2662 ve
->init_state
= DDF_init_full
;
2664 ve
->init_state
= DDF_init_not
;
2666 memset(ve
->pad1
, 0xff, 14);
2667 memset(ve
->name
, ' ', 16);
2669 strncpy(ve
->name
, name
, 16);
2670 ddf
->virt
->populated_vdes
=
2671 cpu_to_be16(be16_to_cpu(ddf
->virt
->populated_vdes
)+1);
2673 /* Now create a new vd_config */
2674 if (posix_memalign((void**)&vcl
, 512,
2675 (offsetof(struct vcl
, conf
) + ddf
->conf_rec_len
* 512)) != 0) {
2676 pr_err("%s could not allocate vd_config\n", __func__
);
2680 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
2683 vc
->magic
= DDF_VD_CONF_MAGIC
;
2684 memcpy(vc
->guid
, ve
->guid
, DDF_GUID_LEN
);
2685 vc
->timestamp
= cpu_to_be32(time(0)-DECADE
);
2686 vc
->seqnum
= cpu_to_be32(1);
2687 memset(vc
->pad0
, 0xff, 24);
2688 vc
->chunk_shift
= chunk_to_shift(info
->chunk_size
);
2689 if (layout_md2ddf(info
, vc
) == -1 ||
2690 be16_to_cpu(vc
->prim_elmnt_count
) > ddf
->mppe
) {
2691 pr_err("%s: unsupported RAID level/layout %d/%d with %d disks\n",
2692 __func__
, info
->level
, info
->layout
, info
->raid_disks
);
2696 vc
->sec_elmnt_seq
= 0;
2697 if (alloc_other_bvds(ddf
, vcl
) != 0) {
2698 pr_err("%s could not allocate other bvds\n",
2703 vc
->blocks
= cpu_to_be64(info
->size
* 2);
2704 vc
->array_blocks
= cpu_to_be64(
2705 calc_array_size(info
->level
, info
->raid_disks
, info
->layout
,
2706 info
->chunk_size
, info
->size
*2));
2707 memset(vc
->pad1
, 0xff, 8);
2708 vc
->spare_refs
[0] = cpu_to_be32(0xffffffff);
2709 vc
->spare_refs
[1] = cpu_to_be32(0xffffffff);
2710 vc
->spare_refs
[2] = cpu_to_be32(0xffffffff);
2711 vc
->spare_refs
[3] = cpu_to_be32(0xffffffff);
2712 vc
->spare_refs
[4] = cpu_to_be32(0xffffffff);
2713 vc
->spare_refs
[5] = cpu_to_be32(0xffffffff);
2714 vc
->spare_refs
[6] = cpu_to_be32(0xffffffff);
2715 vc
->spare_refs
[7] = cpu_to_be32(0xffffffff);
2716 memset(vc
->cache_pol
, 0, 8);
2718 memset(vc
->pad2
, 0xff, 3);
2719 memset(vc
->pad3
, 0xff, 52);
2720 memset(vc
->pad4
, 0xff, 192);
2721 memset(vc
->v0
, 0xff, 32);
2722 memset(vc
->v1
, 0xff, 32);
2723 memset(vc
->v2
, 0xff, 16);
2724 memset(vc
->v3
, 0xff, 16);
2725 memset(vc
->vendor
, 0xff, 32);
2727 memset(vc
->phys_refnum
, 0xff, 4*ddf
->mppe
);
2728 memset(vc
->phys_refnum
+ddf
->mppe
, 0x00, 8*ddf
->mppe
);
2730 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++) {
2731 memcpy(vcl
->other_bvds
[i
-1], vc
, ddf
->conf_rec_len
* 512);
2732 vcl
->other_bvds
[i
-1]->sec_elmnt_seq
= i
;
2735 vcl
->next
= ddf
->conflist
;
2736 ddf
->conflist
= vcl
;
2737 ddf
->currentconf
= vcl
;
2738 ddf_set_updates_pending(ddf
, NULL
);
2743 static void add_to_super_ddf_bvd(struct supertype
*st
,
2744 mdu_disk_info_t
*dk
, int fd
, char *devname
,
2745 unsigned long long data_offset
)
2747 /* fd and devname identify a device within the ddf container (st).
2748 * dk identifies a location in the new BVD.
2749 * We need to find suitable free space in that device and update
2750 * the phys_refnum and lba_offset for the newly created vd_config.
2751 * We might also want to update the type in the phys_disk
2754 * Alternately: fd == -1 and we have already chosen which device to
2755 * use and recorded in dlist->raid_disk;
2758 struct ddf_super
*ddf
= st
->sb
;
2759 struct vd_config
*vc
;
2761 unsigned long long blocks
, pos
;
2762 unsigned int raid_disk
= dk
->raid_disk
;
2765 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2766 if (dl
->raiddisk
== dk
->raid_disk
)
2769 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2770 if (dl
->major
== dk
->major
&&
2771 dl
->minor
== dk
->minor
)
2774 if (!dl
|| dl
->pdnum
< 0 || ! (dk
->state
& (1<<MD_DISK_SYNC
)))
2777 vc
= &ddf
->currentconf
->conf
;
2778 if (vc
->sec_elmnt_count
> 1) {
2779 unsigned int n
= be16_to_cpu(vc
->prim_elmnt_count
);
2781 vc
= ddf
->currentconf
->other_bvds
[raid_disk
/ n
- 1];
2785 blocks
= be64_to_cpu(vc
->blocks
);
2786 if (ddf
->currentconf
->block_sizes
)
2787 blocks
= ddf
->currentconf
->block_sizes
[dk
->raid_disk
];
2789 pos
= find_space(ddf
, dl
, data_offset
, &blocks
);
2790 if (pos
== INVALID_SECTORS
)
2793 ddf
->currentdev
= dk
->raid_disk
;
2794 vc
->phys_refnum
[raid_disk
] = dl
->disk
.refnum
;
2795 LBA_OFFSET(ddf
, vc
)[raid_disk
] = cpu_to_be64(pos
);
2797 for (i
= 0; i
< ddf
->max_part
; i
++)
2798 if (dl
->vlist
[i
] == NULL
)
2800 if (i
== ddf
->max_part
)
2802 dl
->vlist
[i
] = ddf
->currentconf
;
2807 dl
->devname
= devname
;
2809 /* Check if we can mark array as optimal yet */
2810 i
= ddf
->currentconf
->vcnum
;
2811 ddf
->virt
->entries
[i
].state
=
2812 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
2813 | get_svd_state(ddf
, ddf
->currentconf
);
2814 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
2815 cpu_to_be16(DDF_Global_Spare
));
2816 be16_set(ddf
->phys
->entries
[dl
->pdnum
].type
,
2817 cpu_to_be16(DDF_Active_in_VD
));
2818 dprintf("%s: added disk %d/%08x to VD %d/%s as disk %d\n",
2819 __func__
, dl
->pdnum
, be32_to_cpu(dl
->disk
.refnum
),
2820 ddf
->currentconf
->vcnum
, guid_str(vc
->guid
),
2822 ddf_set_updates_pending(ddf
, vc
);
2825 static unsigned int find_unused_pde(const struct ddf_super
*ddf
)
2828 for (i
= 0; i
< be16_to_cpu(ddf
->phys
->max_pdes
); i
++) {
2829 if (all_ff(ddf
->phys
->entries
[i
].guid
))
2832 return DDF_NOTFOUND
;
2835 static void _set_config_size(struct phys_disk_entry
*pde
, const struct dl
*dl
)
2838 cfs
= min(dl
->size
- 32*1024*2ULL, be64_to_cpu(dl
->primary_lba
));
2839 t
= be64_to_cpu(dl
->secondary_lba
);
2843 * Some vendor DDF structures interpret workspace_lba
2844 * very differently than we do: Make a sanity check on the value.
2846 t
= be64_to_cpu(dl
->workspace_lba
);
2848 __u64 wsp
= cfs
- t
;
2849 if (wsp
> 1024*1024*2ULL && wsp
> dl
->size
/ 16) {
2850 pr_err("%s: %x:%x: workspace size 0x%llx too big, ignoring\n",
2851 __func__
, dl
->major
, dl
->minor
, wsp
);
2855 pde
->config_size
= cpu_to_be64(cfs
);
2856 dprintf("%s: %x:%x config_size %llx, DDF structure is %llx blocks\n",
2857 __func__
, dl
->major
, dl
->minor
, cfs
, dl
->size
-cfs
);
2860 /* Add a device to a container, either while creating it or while
2861 * expanding a pre-existing container
2863 static int add_to_super_ddf(struct supertype
*st
,
2864 mdu_disk_info_t
*dk
, int fd
, char *devname
,
2865 unsigned long long data_offset
)
2867 struct ddf_super
*ddf
= st
->sb
;
2871 unsigned long long size
;
2872 struct phys_disk_entry
*pde
;
2877 if (ddf
->currentconf
) {
2878 add_to_super_ddf_bvd(st
, dk
, fd
, devname
, data_offset
);
2882 /* This is device numbered dk->number. We need to create
2883 * a phys_disk entry and a more detailed disk_data entry.
2886 n
= find_unused_pde(ddf
);
2887 if (n
== DDF_NOTFOUND
) {
2888 pr_err("%s: No free slot in array, cannot add disk\n",
2892 pde
= &ddf
->phys
->entries
[n
];
2893 get_dev_size(fd
, NULL
, &size
);
2894 if (size
<= 32*1024*1024) {
2895 pr_err("%s: device size must be at least 32MB\n",
2901 if (posix_memalign((void**)&dd
, 512,
2902 sizeof(*dd
) + sizeof(dd
->vlist
[0]) * ddf
->max_part
) != 0) {
2903 pr_err("%s could allocate buffer for new disk, aborting\n",
2907 dd
->major
= major(stb
.st_rdev
);
2908 dd
->minor
= minor(stb
.st_rdev
);
2909 dd
->devname
= devname
;
2913 dd
->disk
.magic
= DDF_PHYS_DATA_MAGIC
;
2915 tm
= localtime(&now
);
2916 sprintf(dd
->disk
.guid
, "%8s%04d%02d%02d",
2917 T10
, tm
->tm_year
+1900, tm
->tm_mon
+1, tm
->tm_mday
);
2918 tptr
= (__u32
*)(dd
->disk
.guid
+ 16);
2919 *tptr
++ = random32();
2923 /* Cannot be bothered finding a CRC of some irrelevant details*/
2924 dd
->disk
.refnum
._v32
= random32();
2925 for (i
= be16_to_cpu(ddf
->active
->max_pd_entries
);
2927 if (be32_eq(ddf
->phys
->entries
[i
-1].refnum
,
2932 dd
->disk
.forced_ref
= 1;
2933 dd
->disk
.forced_guid
= 1;
2934 memset(dd
->disk
.vendor
, ' ', 32);
2935 memcpy(dd
->disk
.vendor
, "Linux", 5);
2936 memset(dd
->disk
.pad
, 0xff, 442);
2937 for (i
= 0; i
< ddf
->max_part
; i
++)
2938 dd
->vlist
[i
] = NULL
;
2942 if (st
->update_tail
) {
2943 int len
= (sizeof(struct phys_disk
) +
2944 sizeof(struct phys_disk_entry
));
2945 struct phys_disk
*pd
;
2948 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2949 pd
->used_pdes
= cpu_to_be16(n
);
2950 pde
= &pd
->entries
[0];
2953 ddf
->phys
->used_pdes
= cpu_to_be16(
2954 1 + be16_to_cpu(ddf
->phys
->used_pdes
));
2956 memcpy(pde
->guid
, dd
->disk
.guid
, DDF_GUID_LEN
);
2957 pde
->refnum
= dd
->disk
.refnum
;
2958 pde
->type
= cpu_to_be16(DDF_Forced_PD_GUID
| DDF_Global_Spare
);
2959 pde
->state
= cpu_to_be16(DDF_Online
);
2962 * If there is already a device in dlist, try to reserve the same
2963 * amount of workspace. Otherwise, use 32MB.
2964 * We checked disk size above already.
2966 #define __calc_lba(new, old, lba, mb) do { \
2967 unsigned long long dif; \
2968 if ((old) != NULL) \
2969 dif = (old)->size - be64_to_cpu((old)->lba); \
2971 dif = (new)->size; \
2972 if ((new)->size > dif) \
2973 (new)->lba = cpu_to_be64((new)->size - dif); \
2975 (new)->lba = cpu_to_be64((new)->size - (mb*1024*2)); \
2977 __calc_lba(dd
, ddf
->dlist
, workspace_lba
, 32);
2978 __calc_lba(dd
, ddf
->dlist
, primary_lba
, 16);
2979 if (ddf
->dlist
== NULL
||
2980 be64_to_cpu(ddf
->dlist
->secondary_lba
) != ~(__u64
)0)
2981 __calc_lba(dd
, ddf
->dlist
, secondary_lba
, 32);
2982 _set_config_size(pde
, dd
);
2984 sprintf(pde
->path
, "%17.17s","Information: nil") ;
2985 memset(pde
->pad
, 0xff, 6);
2987 if (st
->update_tail
) {
2988 dd
->next
= ddf
->add_list
;
2991 dd
->next
= ddf
->dlist
;
2993 ddf_set_updates_pending(ddf
, NULL
);
2999 static int remove_from_super_ddf(struct supertype
*st
, mdu_disk_info_t
*dk
)
3001 struct ddf_super
*ddf
= st
->sb
;
3004 /* mdmon has noticed that this disk (dk->major/dk->minor) has
3005 * disappeared from the container.
3006 * We need to arrange that it disappears from the metadata and
3007 * internal data structures too.
3008 * Most of the work is done by ddf_process_update which edits
3009 * the metadata and closes the file handle and attaches the memory
3010 * where free_updates will free it.
3012 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
3013 if (dl
->major
== dk
->major
&&
3014 dl
->minor
== dk
->minor
)
3016 if (!dl
|| dl
->pdnum
< 0)
3019 if (st
->update_tail
) {
3020 int len
= (sizeof(struct phys_disk
) +
3021 sizeof(struct phys_disk_entry
));
3022 struct phys_disk
*pd
;
3025 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
3026 pd
->used_pdes
= cpu_to_be16(dl
->pdnum
);
3027 pd
->entries
[0].state
= cpu_to_be16(DDF_Missing
);
3028 append_metadata_update(st
, pd
, len
);
3035 * This is the write_init_super method for a ddf container. It is
3036 * called when creating a container or adding another device to a
3040 static int __write_ddf_structure(struct dl
*d
, struct ddf_super
*ddf
, __u8 type
)
3042 unsigned long long sector
;
3043 struct ddf_header
*header
;
3044 int fd
, i
, n_config
, conf_size
, buf_size
;
3051 case DDF_HEADER_PRIMARY
:
3052 header
= &ddf
->primary
;
3053 sector
= be64_to_cpu(header
->primary_lba
);
3055 case DDF_HEADER_SECONDARY
:
3056 header
= &ddf
->secondary
;
3057 sector
= be64_to_cpu(header
->secondary_lba
);
3062 if (sector
== ~(__u64
)0)
3065 header
->type
= type
;
3066 header
->openflag
= 1;
3067 header
->crc
= calc_crc(header
, 512);
3069 lseek64(fd
, sector
<<9, 0);
3070 if (write(fd
, header
, 512) < 0)
3073 ddf
->controller
.crc
= calc_crc(&ddf
->controller
, 512);
3074 if (write(fd
, &ddf
->controller
, 512) < 0)
3077 ddf
->phys
->crc
= calc_crc(ddf
->phys
, ddf
->pdsize
);
3078 if (write(fd
, ddf
->phys
, ddf
->pdsize
) < 0)
3080 ddf
->virt
->crc
= calc_crc(ddf
->virt
, ddf
->vdsize
);
3081 if (write(fd
, ddf
->virt
, ddf
->vdsize
) < 0)
3084 /* Now write lots of config records. */
3085 n_config
= ddf
->max_part
;
3086 conf_size
= ddf
->conf_rec_len
* 512;
3088 buf_size
= conf_size
* (n_config
+ 1);
3090 if (posix_memalign((void**)&conf
, 512, buf_size
) != 0)
3094 for (i
= 0 ; i
<= n_config
; i
++) {
3096 struct vd_config
*vdc
= NULL
;
3097 if (i
== n_config
) {
3098 c
= (struct vcl
*)d
->spare
;
3105 get_pd_index_from_refnum(
3108 (const struct vd_config
**)&vdc
,
3112 dprintf("writing conf record %i on disk %08x for %s/%u\n",
3113 i
, be32_to_cpu(d
->disk
.refnum
),
3114 guid_str(vdc
->guid
),
3115 vdc
->sec_elmnt_seq
);
3116 vdc
->crc
= calc_crc(vdc
, conf_size
);
3117 memcpy(conf
+ i
*conf_size
, vdc
, conf_size
);
3119 memset(conf
+ i
*conf_size
, 0xff, conf_size
);
3121 if (write(fd
, conf
, buf_size
) != buf_size
)
3124 d
->disk
.crc
= calc_crc(&d
->disk
, 512);
3125 if (write(fd
, &d
->disk
, 512) < 0)
3130 header
->openflag
= 0;
3131 header
->crc
= calc_crc(header
, 512);
3133 lseek64(fd
, sector
<<9, 0);
3134 if (write(fd
, header
, 512) < 0)
3140 static int _write_super_to_disk(struct ddf_super
*ddf
, struct dl
*d
)
3142 unsigned long long size
;
3147 /* We need to fill in the primary, (secondary) and workspace
3148 * lba's in the headers, set their checksums,
3149 * Also checksum phys, virt....
3151 * Then write everything out, finally the anchor is written.
3153 get_dev_size(fd
, NULL
, &size
);
3155 memcpy(&ddf
->anchor
, ddf
->active
, 512);
3156 if (be64_to_cpu(d
->workspace_lba
) != 0ULL)
3157 ddf
->anchor
.workspace_lba
= d
->workspace_lba
;
3159 ddf
->anchor
.workspace_lba
=
3160 cpu_to_be64(size
- 32*1024*2);
3161 if (be64_to_cpu(d
->primary_lba
) != 0ULL)
3162 ddf
->anchor
.primary_lba
= d
->primary_lba
;
3164 ddf
->anchor
.primary_lba
=
3165 cpu_to_be64(size
- 16*1024*2);
3166 if (be64_to_cpu(d
->secondary_lba
) != 0ULL)
3167 ddf
->anchor
.secondary_lba
= d
->secondary_lba
;
3169 ddf
->anchor
.secondary_lba
=
3170 cpu_to_be64(size
- 32*1024*2);
3171 ddf
->anchor
.timestamp
= cpu_to_be32(time(0) - DECADE
);
3172 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
3173 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
3175 ddf
->anchor
.openflag
= 0xFF; /* 'open' means nothing */
3176 ddf
->anchor
.seq
= cpu_to_be32(0xFFFFFFFF); /* no sequencing in anchor */
3177 ddf
->anchor
.crc
= calc_crc(&ddf
->anchor
, 512);
3179 if (!__write_ddf_structure(d
, ddf
, DDF_HEADER_PRIMARY
))
3182 if (!__write_ddf_structure(d
, ddf
, DDF_HEADER_SECONDARY
))
3185 lseek64(fd
, (size
-1)*512, SEEK_SET
);
3186 if (write(fd
, &ddf
->anchor
, 512) < 0)
3193 static int __write_init_super_ddf(struct supertype
*st
)
3195 struct ddf_super
*ddf
= st
->sb
;
3200 pr_state(ddf
, __func__
);
3202 /* try to write updated metadata,
3203 * if we catch a failure move on to the next disk
3205 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
3207 successes
+= _write_super_to_disk(ddf
, d
);
3210 return attempts
!= successes
;
3213 static int write_init_super_ddf(struct supertype
*st
)
3215 struct ddf_super
*ddf
= st
->sb
;
3216 struct vcl
*currentconf
= ddf
->currentconf
;
3218 /* We are done with currentconf - reset it so st refers to the container */
3219 ddf
->currentconf
= NULL
;
3221 if (st
->update_tail
) {
3222 /* queue the virtual_disk and vd_config as metadata updates */
3223 struct virtual_disk
*vd
;
3224 struct vd_config
*vc
;
3229 /* Must be adding a physical disk to the container */
3230 int len
= (sizeof(struct phys_disk
) +
3231 sizeof(struct phys_disk_entry
));
3233 /* adding a disk to the container. */
3237 append_metadata_update(st
, ddf
->add_list
->mdupdate
, len
);
3238 ddf
->add_list
->mdupdate
= NULL
;
3242 /* Newly created VD */
3244 /* First the virtual disk. We have a slightly fake header */
3245 len
= sizeof(struct virtual_disk
) + sizeof(struct virtual_entry
);
3248 vd
->entries
[0] = ddf
->virt
->entries
[currentconf
->vcnum
];
3249 vd
->populated_vdes
= cpu_to_be16(currentconf
->vcnum
);
3250 append_metadata_update(st
, vd
, len
);
3252 /* Then the vd_config */
3253 len
= ddf
->conf_rec_len
* 512;
3254 tlen
= len
* currentconf
->conf
.sec_elmnt_count
;
3256 memcpy(vc
, ¤tconf
->conf
, len
);
3257 for (i
= 1; i
< currentconf
->conf
.sec_elmnt_count
; i
++)
3258 memcpy((char *)vc
+ i
*len
, currentconf
->other_bvds
[i
-1],
3260 append_metadata_update(st
, vc
, tlen
);
3262 /* FIXME I need to close the fds! */
3267 for (d
= ddf
->dlist
; d
; d
=d
->next
)
3268 while (Kill(d
->devname
, NULL
, 0, -1, 1) == 0);
3269 return __write_init_super_ddf(st
);
3275 static __u64
avail_size_ddf(struct supertype
*st
, __u64 devsize
,
3276 unsigned long long data_offset
)
3278 /* We must reserve the last 32Meg */
3279 if (devsize
<= 32*1024*2)
3281 return devsize
- 32*1024*2;
3286 static int reserve_space(struct supertype
*st
, int raiddisks
,
3287 unsigned long long size
, int chunk
,
3288 unsigned long long data_offset
,
3289 unsigned long long *freesize
)
3291 /* Find 'raiddisks' spare extents at least 'size' big (but
3292 * only caring about multiples of 'chunk') and remember
3293 * them. If size==0, find the largest size possible.
3294 * Report available size in *freesize
3295 * If space cannot be found, fail.
3298 struct ddf_super
*ddf
= st
->sb
;
3301 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3305 /* Now find largest extent on each device */
3306 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3307 unsigned long long minsize
= ULLONG_MAX
;
3309 find_space(ddf
, dl
, data_offset
, &minsize
);
3310 if (minsize
>= size
&& minsize
>= (unsigned)chunk
) {
3312 dl
->esize
= minsize
;
3315 if (cnt
< raiddisks
) {
3316 pr_err("not enough devices with space to create array.\n");
3317 return 0; /* No enough free spaces large enough */
3320 /* choose the largest size of which there are at least 'raiddisk' */
3321 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3323 if (dl
->esize
<= size
)
3325 /* This is bigger than 'size', see if there are enough */
3327 for (dl2
= ddf
->dlist
; dl2
; dl2
=dl2
->next
)
3328 if (dl2
->esize
>= dl
->esize
)
3330 if (cnt
>= raiddisks
)
3334 size
= size
/ chunk
;
3339 pr_err("not enough spare devices to create array.\n");
3343 /* We have a 'size' of which there are enough spaces.
3344 * We simply do a first-fit */
3346 for (dl
= ddf
->dlist
; dl
&& cnt
< raiddisks
; dl
=dl
->next
) {
3347 if (dl
->esize
< size
)
3356 static int validate_geometry_ddf(struct supertype
*st
,
3357 int level
, int layout
, int raiddisks
,
3358 int *chunk
, unsigned long long size
,
3359 unsigned long long data_offset
,
3360 char *dev
, unsigned long long *freesize
,
3367 /* ddf potentially supports lots of things, but it depends on
3368 * what devices are offered (and maybe kernel version?)
3369 * If given unused devices, we will make a container.
3370 * If given devices in a container, we will make a BVD.
3371 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
3374 if (*chunk
== UnSet
)
3375 *chunk
= DEFAULT_CHUNK
;
3377 if (level
== LEVEL_NONE
)
3378 level
= LEVEL_CONTAINER
;
3379 if (level
== LEVEL_CONTAINER
) {
3380 /* Must be a fresh device to add to a container */
3381 return validate_geometry_ddf_container(st
, level
, layout
,
3383 size
, data_offset
, dev
,
3389 mdu_array_info_t array
= {
3392 .raid_disks
= raiddisks
3394 struct vd_config conf
;
3395 if (layout_md2ddf(&array
, &conf
) == -1) {
3397 pr_err("DDF does not support level %d /layout %d arrays with %d disks\n",
3398 level
, layout
, raiddisks
);
3401 /* Should check layout? etc */
3403 if (st
->sb
&& freesize
) {
3404 /* --create was given a container to create in.
3405 * So we need to check that there are enough
3406 * free spaces and return the amount of space.
3407 * We may as well remember which drives were
3408 * chosen so that add_to_super/getinfo_super
3411 return reserve_space(st
, raiddisks
, size
, *chunk
,
3412 data_offset
, freesize
);
3418 /* A container has already been opened, so we are
3419 * creating in there. Maybe a BVD, maybe an SVD.
3420 * Should make a distinction one day.
3422 return validate_geometry_ddf_bvd(st
, level
, layout
, raiddisks
,
3423 chunk
, size
, data_offset
, dev
,
3427 /* This is the first device for the array.
3428 * If it is a container, we read it in and do automagic allocations,
3429 * no other devices should be given.
3430 * Otherwise it must be a member device of a container, and we
3431 * do manual allocation.
3432 * Later we should check for a BVD and make an SVD.
3434 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3436 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
3438 if (sra
&& sra
->array
.major_version
== -1 &&
3439 strcmp(sra
->text_version
, "ddf") == 0) {
3441 /* find space for 'n' devices. */
3442 /* remember the devices */
3443 /* Somehow return the fact that we have enough */
3447 pr_err("ddf: Cannot create this array "
3448 "on device %s - a container is required.\n",
3452 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
3454 pr_err("ddf: Cannot open %s: %s\n",
3455 dev
, strerror(errno
));
3458 /* Well, it is in use by someone, maybe a 'ddf' container. */
3459 cfd
= open_container(fd
);
3463 pr_err("ddf: Cannot use %s: %s\n",
3464 dev
, strerror(EBUSY
));
3467 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
3469 if (sra
&& sra
->array
.major_version
== -1 &&
3470 strcmp(sra
->text_version
, "ddf") == 0) {
3471 /* This is a member of a ddf container. Load the container
3472 * and try to create a bvd
3474 struct ddf_super
*ddf
;
3475 if (load_super_ddf_all(st
, cfd
, (void **)&ddf
, NULL
) == 0) {
3477 strcpy(st
->container_devnm
, fd2devnm(cfd
));
3479 return validate_geometry_ddf_bvd(st
, level
, layout
,
3480 raiddisks
, chunk
, size
,
3486 } else /* device may belong to a different container */
3493 validate_geometry_ddf_container(struct supertype
*st
,
3494 int level
, int layout
, int raiddisks
,
3495 int chunk
, unsigned long long size
,
3496 unsigned long long data_offset
,
3497 char *dev
, unsigned long long *freesize
,
3501 unsigned long long ldsize
;
3503 if (level
!= LEVEL_CONTAINER
)
3508 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3511 pr_err("ddf: Cannot open %s: %s\n",
3512 dev
, strerror(errno
));
3515 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3521 *freesize
= avail_size_ddf(st
, ldsize
>> 9, INVALID_SECTORS
);
3528 static int validate_geometry_ddf_bvd(struct supertype
*st
,
3529 int level
, int layout
, int raiddisks
,
3530 int *chunk
, unsigned long long size
,
3531 unsigned long long data_offset
,
3532 char *dev
, unsigned long long *freesize
,
3536 struct ddf_super
*ddf
= st
->sb
;
3538 unsigned long long maxsize
;
3539 /* ddf/bvd supports lots of things, but not containers */
3540 if (level
== LEVEL_CONTAINER
) {
3542 pr_err("DDF cannot create a container within an container\n");
3545 /* We must have the container info already read in. */
3550 /* General test: make sure there is space for
3551 * 'raiddisks' device extents of size 'size'.
3553 unsigned long long minsize
= size
;
3557 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
3558 if (find_space(ddf
, dl
, data_offset
, &minsize
)
3562 if (dcnt
< raiddisks
) {
3564 pr_err("ddf: Not enough devices with "
3565 "space for this array (%d < %d)\n",
3571 /* This device must be a member of the set */
3572 if (stat(dev
, &stb
) < 0)
3574 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3576 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
3577 if (dl
->major
== (int)major(stb
.st_rdev
) &&
3578 dl
->minor
== (int)minor(stb
.st_rdev
))
3583 pr_err("ddf: %s is not in the "
3588 maxsize
= ULLONG_MAX
;
3589 find_space(ddf
, dl
, data_offset
, &maxsize
);
3590 *freesize
= maxsize
;
3596 static int load_super_ddf_all(struct supertype
*st
, int fd
,
3597 void **sbp
, char *devname
)
3600 struct ddf_super
*super
;
3601 struct mdinfo
*sd
, *best
= NULL
;
3607 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3610 if (sra
->array
.major_version
!= -1 ||
3611 sra
->array
.minor_version
!= -2 ||
3612 strcmp(sra
->text_version
, "ddf") != 0)
3615 if (posix_memalign((void**)&super
, 512, sizeof(*super
)) != 0)
3617 memset(super
, 0, sizeof(*super
));
3619 /* first, try each device, and choose the best ddf */
3620 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3622 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3623 dfd
= dev_open(nm
, O_RDONLY
);
3626 rv
= load_ddf_headers(dfd
, super
, NULL
);
3629 seq
= be32_to_cpu(super
->active
->seq
);
3630 if (super
->active
->openflag
)
3632 if (!best
|| seq
> bestseq
) {
3640 /* OK, load this ddf */
3641 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
3642 dfd
= dev_open(nm
, O_RDONLY
);
3645 load_ddf_headers(dfd
, super
, NULL
);
3646 load_ddf_global(dfd
, super
, NULL
);
3648 /* Now we need the device-local bits */
3649 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3652 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3653 dfd
= dev_open(nm
, O_RDWR
);
3656 rv
= load_ddf_headers(dfd
, super
, NULL
);
3658 rv
= load_ddf_local(dfd
, super
, NULL
, 1);
3664 if (st
->ss
== NULL
) {
3665 st
->ss
= &super_ddf
;
3666 st
->minor_version
= 0;
3669 strcpy(st
->container_devnm
, fd2devnm(fd
));
3673 static int load_container_ddf(struct supertype
*st
, int fd
,
3676 return load_super_ddf_all(st
, fd
, &st
->sb
, devname
);
3679 #endif /* MDASSEMBLE */
3681 static int check_secondary(const struct vcl
*vc
)
3683 const struct vd_config
*conf
= &vc
->conf
;
3686 /* The only DDF secondary RAID level md can support is
3687 * RAID 10, if the stripe sizes and Basic volume sizes
3689 * Other configurations could in theory be supported by exposing
3690 * the BVDs to user space and using device mapper for the secondary
3691 * mapping. So far we don't support that.
3694 __u64 sec_elements
[4] = {0, 0, 0, 0};
3695 #define __set_sec_seen(n) (sec_elements[(n)>>6] |= (1<<((n)&63)))
3696 #define __was_sec_seen(n) ((sec_elements[(n)>>6] & (1<<((n)&63))) != 0)
3698 if (vc
->other_bvds
== NULL
) {
3699 pr_err("No BVDs for secondary RAID found\n");
3702 if (conf
->prl
!= DDF_RAID1
) {
3703 pr_err("Secondary RAID level only supported for mirrored BVD\n");
3706 if (conf
->srl
!= DDF_2STRIPED
&& conf
->srl
!= DDF_2SPANNED
) {
3707 pr_err("Secondary RAID level %d is unsupported\n",
3711 __set_sec_seen(conf
->sec_elmnt_seq
);
3712 for (i
= 0; i
< conf
->sec_elmnt_count
-1; i
++) {
3713 const struct vd_config
*bvd
= vc
->other_bvds
[i
];
3714 if (bvd
->sec_elmnt_seq
== DDF_UNUSED_BVD
)
3716 if (bvd
->srl
!= conf
->srl
) {
3717 pr_err("Inconsistent secondary RAID level across BVDs\n");
3720 if (bvd
->prl
!= conf
->prl
) {
3721 pr_err("Different RAID levels for BVDs are unsupported\n");
3724 if (!be16_eq(bvd
->prim_elmnt_count
, conf
->prim_elmnt_count
)) {
3725 pr_err("All BVDs must have the same number of primary elements\n");
3728 if (bvd
->chunk_shift
!= conf
->chunk_shift
) {
3729 pr_err("Different strip sizes for BVDs are unsupported\n");
3732 if (!be64_eq(bvd
->array_blocks
, conf
->array_blocks
)) {
3733 pr_err("Different BVD sizes are unsupported\n");
3736 __set_sec_seen(bvd
->sec_elmnt_seq
);
3738 for (i
= 0; i
< conf
->sec_elmnt_count
; i
++) {
3739 if (!__was_sec_seen(i
)) {
3740 pr_err("BVD %d is missing\n", i
);
3747 static unsigned int get_pd_index_from_refnum(const struct vcl
*vc
,
3748 be32 refnum
, unsigned int nmax
,
3749 const struct vd_config
**bvd
,
3752 unsigned int i
, j
, n
, sec
, cnt
;
3754 cnt
= be16_to_cpu(vc
->conf
.prim_elmnt_count
);
3755 sec
= (vc
->conf
.sec_elmnt_count
== 1 ? 0 : vc
->conf
.sec_elmnt_seq
);
3757 for (i
= 0, j
= 0 ; i
< nmax
; i
++) {
3758 /* j counts valid entries for this BVD */
3759 if (be32_eq(vc
->conf
.phys_refnum
[i
], refnum
)) {
3762 return sec
* cnt
+ j
;
3764 if (be32_to_cpu(vc
->conf
.phys_refnum
[i
]) != 0xffffffff)
3767 if (vc
->other_bvds
== NULL
)
3770 for (n
= 1; n
< vc
->conf
.sec_elmnt_count
; n
++) {
3771 struct vd_config
*vd
= vc
->other_bvds
[n
-1];
3772 sec
= vd
->sec_elmnt_seq
;
3773 if (sec
== DDF_UNUSED_BVD
)
3775 for (i
= 0, j
= 0 ; i
< nmax
; i
++) {
3776 if (be32_eq(vd
->phys_refnum
[i
], refnum
)) {
3779 return sec
* cnt
+ j
;
3781 if (be32_to_cpu(vd
->phys_refnum
[i
]) != 0xffffffff)
3787 return DDF_NOTFOUND
;
3790 static struct mdinfo
*container_content_ddf(struct supertype
*st
, char *subarray
)
3792 /* Given a container loaded by load_super_ddf_all,
3793 * extract information about all the arrays into
3796 * For each vcl in conflist: create an mdinfo, fill it in,
3797 * then look for matching devices (phys_refnum) in dlist
3798 * and create appropriate device mdinfo.
3800 struct ddf_super
*ddf
= st
->sb
;
3801 struct mdinfo
*rest
= NULL
;
3804 for (vc
= ddf
->conflist
; vc
; vc
=vc
->next
) {
3806 struct mdinfo
*this;
3812 (strtoul(subarray
, &ep
, 10) != vc
->vcnum
||
3816 if (vc
->conf
.sec_elmnt_count
> 1) {
3817 if (check_secondary(vc
) != 0)
3821 this = xcalloc(1, sizeof(*this));
3825 if (layout_ddf2md(&vc
->conf
, &this->array
))
3827 this->array
.md_minor
= -1;
3828 this->array
.major_version
= -1;
3829 this->array
.minor_version
= -2;
3830 this->safe_mode_delay
= DDF_SAFE_MODE_DELAY
;
3831 cptr
= (__u32
*)(vc
->conf
.guid
+ 16);
3832 this->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
3833 this->array
.utime
= DECADE
+
3834 be32_to_cpu(vc
->conf
.timestamp
);
3835 this->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
3838 if ((ddf
->virt
->entries
[i
].state
& DDF_state_inconsistent
) ||
3839 (ddf
->virt
->entries
[i
].init_state
& DDF_initstate_mask
) !=
3841 this->array
.state
= 0;
3842 this->resync_start
= 0;
3844 this->array
.state
= 1;
3845 this->resync_start
= MaxSector
;
3847 _ddf_array_name(this->name
, ddf
, i
);
3848 memset(this->uuid
, 0, sizeof(this->uuid
));
3849 this->component_size
= be64_to_cpu(vc
->conf
.blocks
);
3850 this->array
.size
= this->component_size
/ 2;
3851 this->container_member
= i
;
3853 ddf
->currentconf
= vc
;
3854 uuid_from_super_ddf(st
, this->uuid
);
3856 ddf
->currentconf
= NULL
;
3858 sprintf(this->text_version
, "/%s/%d",
3859 st
->container_devnm
, this->container_member
);
3861 for (pd
= 0; pd
< be16_to_cpu(ddf
->phys
->max_pdes
); pd
++) {
3864 const struct vd_config
*bvd
;
3868 if (be32_to_cpu(ddf
->phys
->entries
[pd
].refnum
)
3872 stt
= be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
3873 if ((stt
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
3877 i
= get_pd_index_from_refnum(
3878 vc
, ddf
->phys
->entries
[pd
].refnum
,
3879 ddf
->mppe
, &bvd
, &iphys
);
3880 if (i
== DDF_NOTFOUND
)
3883 this->array
.working_disks
++;
3885 for (d
= ddf
->dlist
; d
; d
=d
->next
)
3886 if (be32_eq(d
->disk
.refnum
,
3887 ddf
->phys
->entries
[pd
].refnum
))
3890 /* Haven't found that one yet, maybe there are others */
3893 dev
= xcalloc(1, sizeof(*dev
));
3894 dev
->next
= this->devs
;
3897 dev
->disk
.number
= be32_to_cpu(d
->disk
.refnum
);
3898 dev
->disk
.major
= d
->major
;
3899 dev
->disk
.minor
= d
->minor
;
3900 dev
->disk
.raid_disk
= i
;
3901 dev
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
3902 dev
->recovery_start
= MaxSector
;
3904 dev
->events
= be32_to_cpu(ddf
->active
->seq
);
3906 be64_to_cpu(LBA_OFFSET(ddf
, bvd
)[iphys
]);
3907 dev
->component_size
= be64_to_cpu(bvd
->blocks
);
3909 strcpy(dev
->name
, d
->devname
);
3915 static int store_super_ddf(struct supertype
*st
, int fd
)
3917 struct ddf_super
*ddf
= st
->sb
;
3918 unsigned long long dsize
;
3925 if (!get_dev_size(fd
, NULL
, &dsize
))
3928 if (ddf
->dlist
|| ddf
->conflist
) {
3933 if (fstat(fd
, &sta
) == -1 || !S_ISBLK(sta
.st_mode
)) {
3934 pr_err("%s: file descriptor for invalid device\n",
3938 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
3939 if (dl
->major
== (int)major(sta
.st_rdev
) &&
3940 dl
->minor
== (int)minor(sta
.st_rdev
))
3943 pr_err("%s: couldn't find disk %d/%d\n", __func__
,
3944 (int)major(sta
.st_rdev
),
3945 (int)minor(sta
.st_rdev
));
3950 ret
= (_write_super_to_disk(ddf
, dl
) != 1);
3955 if (posix_memalign(&buf
, 512, 512) != 0)
3957 memset(buf
, 0, 512);
3959 lseek64(fd
, dsize
-512, 0);
3960 rc
= write(fd
, buf
, 512);
3967 static int compare_super_ddf(struct supertype
*st
, struct supertype
*tst
)
3971 * 0 same, or first was empty, and second was copied
3972 * 1 second had wrong magic number - but that isn't possible
3974 * 3 wrong other info
3976 struct ddf_super
*first
= st
->sb
;
3977 struct ddf_super
*second
= tst
->sb
;
3978 struct dl
*dl1
, *dl2
;
3979 struct vcl
*vl1
, *vl2
;
3980 unsigned int max_vds
, max_pds
, pd
, vd
;
3988 if (memcmp(first
->anchor
.guid
, second
->anchor
.guid
, DDF_GUID_LEN
) != 0)
3991 /* It is only OK to compare info in the anchor. Anything else
3992 * could be changing due to a reconfig so must be ignored.
3993 * guid really should be enough anyway.
3996 if (!be32_eq(first
->active
->seq
, second
->active
->seq
)) {
3997 dprintf("%s: sequence number mismatch %u<->%u\n", __func__
,
3998 be32_to_cpu(first
->active
->seq
),
3999 be32_to_cpu(second
->active
->seq
));
4004 * At this point we are fairly sure that the meta data matches.
4005 * But the new disk may contain additional local data.
4006 * Add it to the super block.
4008 max_vds
= be16_to_cpu(first
->active
->max_vd_entries
);
4009 max_pds
= be16_to_cpu(first
->phys
->max_pdes
);
4010 for (vl2
= second
->conflist
; vl2
; vl2
= vl2
->next
) {
4011 for (vl1
= first
->conflist
; vl1
; vl1
= vl1
->next
)
4012 if (!memcmp(vl1
->conf
.guid
, vl2
->conf
.guid
,
4016 if (vl1
->other_bvds
!= NULL
&&
4017 vl1
->conf
.sec_elmnt_seq
!=
4018 vl2
->conf
.sec_elmnt_seq
) {
4019 dprintf("%s: adding BVD %u\n", __func__
,
4020 vl2
->conf
.sec_elmnt_seq
);
4021 add_other_bvd(vl1
, &vl2
->conf
,
4022 first
->conf_rec_len
*512);
4027 if (posix_memalign((void **)&vl1
, 512,
4028 (first
->conf_rec_len
*512 +
4029 offsetof(struct vcl
, conf
))) != 0) {
4030 pr_err("%s could not allocate vcl buf\n",
4035 vl1
->next
= first
->conflist
;
4036 vl1
->block_sizes
= NULL
;
4037 memcpy(&vl1
->conf
, &vl2
->conf
, first
->conf_rec_len
*512);
4038 if (alloc_other_bvds(first
, vl1
) != 0) {
4039 pr_err("%s could not allocate other bvds\n",
4044 for (vd
= 0; vd
< max_vds
; vd
++)
4045 if (!memcmp(first
->virt
->entries
[vd
].guid
,
4046 vl1
->conf
.guid
, DDF_GUID_LEN
))
4049 dprintf("%s: added config for VD %u\n", __func__
, vl1
->vcnum
);
4050 first
->conflist
= vl1
;
4053 for (dl2
= second
->dlist
; dl2
; dl2
= dl2
->next
) {
4054 for (dl1
= first
->dlist
; dl1
; dl1
= dl1
->next
)
4055 if (be32_eq(dl1
->disk
.refnum
, dl2
->disk
.refnum
))
4060 if (posix_memalign((void **)&dl1
, 512,
4061 sizeof(*dl1
) + (first
->max_part
) * sizeof(dl1
->vlist
[0]))
4063 pr_err("%s could not allocate disk info buffer\n",
4067 memcpy(dl1
, dl2
, sizeof(*dl1
));
4068 dl1
->mdupdate
= NULL
;
4069 dl1
->next
= first
->dlist
;
4071 for (pd
= 0; pd
< max_pds
; pd
++)
4072 if (be32_eq(first
->phys
->entries
[pd
].refnum
,
4075 dl1
->pdnum
= pd
< max_pds
? (int)pd
: -1;
4077 if (posix_memalign((void **)&dl1
->spare
, 512,
4078 first
->conf_rec_len
*512) != 0) {
4079 pr_err("%s could not allocate spare info buf\n",
4083 memcpy(dl1
->spare
, dl2
->spare
, first
->conf_rec_len
*512);
4085 for (vd
= 0 ; vd
< first
->max_part
; vd
++) {
4086 if (!dl2
->vlist
[vd
]) {
4087 dl1
->vlist
[vd
] = NULL
;
4090 for (vl1
= first
->conflist
; vl1
; vl1
= vl1
->next
) {
4091 if (!memcmp(vl1
->conf
.guid
,
4092 dl2
->vlist
[vd
]->conf
.guid
,
4095 dl1
->vlist
[vd
] = vl1
;
4099 dprintf("%s: added disk %d: %08x\n", __func__
, dl1
->pdnum
,
4100 be32_to_cpu(dl1
->disk
.refnum
));
4108 * A new array 'a' has been started which claims to be instance 'inst'
4109 * within container 'c'.
4110 * We need to confirm that the array matches the metadata in 'c' so
4111 * that we don't corrupt any metadata.
4113 static int ddf_open_new(struct supertype
*c
, struct active_array
*a
, char *inst
)
4115 struct ddf_super
*ddf
= c
->sb
;
4119 static const char faulty
[] = "faulty";
4121 if (all_ff(ddf
->virt
->entries
[n
].guid
)) {
4122 pr_err("%s: subarray %d doesn't exist\n", __func__
, n
);
4125 dprintf("%s: new subarray %d, GUID: %s\n", __func__
, n
,
4126 guid_str(ddf
->virt
->entries
[n
].guid
));
4127 for (dev
= a
->info
.devs
; dev
; dev
= dev
->next
) {
4128 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4129 if (dl
->major
== dev
->disk
.major
&&
4130 dl
->minor
== dev
->disk
.minor
)
4132 if (!dl
|| dl
->pdnum
< 0) {
4133 pr_err("%s: device %d/%d of subarray %d not found in meta data\n",
4134 __func__
, dev
->disk
.major
, dev
->disk
.minor
, n
);
4137 if ((be16_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].state
) &
4138 (DDF_Online
|DDF_Missing
|DDF_Failed
)) != DDF_Online
) {
4139 pr_err("%s: new subarray %d contains broken device %d/%d (%02x)\n",
4140 __func__
, n
, dl
->major
, dl
->minor
,
4142 ddf
->phys
->entries
[dl
->pdnum
].state
));
4143 if (write(dev
->state_fd
, faulty
, sizeof(faulty
)-1) !=
4145 pr_err("Write to state_fd failed\n");
4146 dev
->curr_state
= DS_FAULTY
;
4149 a
->info
.container_member
= n
;
4153 static void handle_missing(struct ddf_super
*ddf
, struct active_array
*a
, int inst
)
4155 /* This member array is being activated. If any devices
4156 * are missing they must now be marked as failed.
4158 struct vd_config
*vc
;
4166 for (n
= 0; ; n
++) {
4167 vc
= find_vdcr(ddf
, inst
, n
, &n_bvd
, &vcl
);
4170 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4171 if (be32_eq(dl
->disk
.refnum
, vc
->phys_refnum
[n_bvd
]))
4174 /* Found this disk, so not missing */
4177 /* Mark the device as failed/missing. */
4178 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
4179 if (pd
>= 0 && be16_and(ddf
->phys
->entries
[pd
].state
,
4180 cpu_to_be16(DDF_Online
))) {
4181 be16_clear(ddf
->phys
->entries
[pd
].state
,
4182 cpu_to_be16(DDF_Online
));
4183 be16_set(ddf
->phys
->entries
[pd
].state
,
4184 cpu_to_be16(DDF_Failed
|DDF_Missing
));
4185 vc
->phys_refnum
[n_bvd
] = cpu_to_be32(0);
4186 ddf_set_updates_pending(ddf
, vc
);
4189 /* Mark the array as Degraded */
4190 state
= get_svd_state(ddf
, vcl
);
4191 if (ddf
->virt
->entries
[inst
].state
!=
4192 ((ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4194 ddf
->virt
->entries
[inst
].state
=
4195 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4197 a
->check_degraded
= 1;
4198 ddf_set_updates_pending(ddf
, vc
);
4204 * The array 'a' is to be marked clean in the metadata.
4205 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
4206 * clean up to the point (in sectors). If that cannot be recorded in the
4207 * metadata, then leave it as dirty.
4209 * For DDF, we need to clear the DDF_state_inconsistent bit in the
4210 * !global! virtual_disk.virtual_entry structure.
4212 static int ddf_set_array_state(struct active_array
*a
, int consistent
)
4214 struct ddf_super
*ddf
= a
->container
->sb
;
4215 int inst
= a
->info
.container_member
;
4216 int old
= ddf
->virt
->entries
[inst
].state
;
4217 if (consistent
== 2) {
4218 handle_missing(ddf
, a
, inst
);
4219 /* Should check if a recovery should be started FIXME */
4221 if (!is_resync_complete(&a
->info
))
4225 ddf
->virt
->entries
[inst
].state
&= ~DDF_state_inconsistent
;
4227 ddf
->virt
->entries
[inst
].state
|= DDF_state_inconsistent
;
4228 if (old
!= ddf
->virt
->entries
[inst
].state
)
4229 ddf_set_updates_pending(ddf
, NULL
);
4231 old
= ddf
->virt
->entries
[inst
].init_state
;
4232 ddf
->virt
->entries
[inst
].init_state
&= ~DDF_initstate_mask
;
4233 if (is_resync_complete(&a
->info
))
4234 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_full
;
4235 else if (a
->info
.resync_start
== 0)
4236 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_not
;
4238 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_quick
;
4239 if (old
!= ddf
->virt
->entries
[inst
].init_state
)
4240 ddf_set_updates_pending(ddf
, NULL
);
4242 dprintf("ddf mark %d/%s (%d) %s %llu\n", inst
,
4243 guid_str(ddf
->virt
->entries
[inst
].guid
), a
->curr_state
,
4244 consistent
?"clean":"dirty",
4245 a
->info
.resync_start
);
4249 static int get_bvd_state(const struct ddf_super
*ddf
,
4250 const struct vd_config
*vc
)
4252 unsigned int i
, n_bvd
, working
= 0;
4253 unsigned int n_prim
= be16_to_cpu(vc
->prim_elmnt_count
);
4255 char *avail
= xcalloc(1, n_prim
);
4256 mdu_array_info_t array
;
4258 layout_ddf2md(vc
, &array
);
4260 for (i
= 0; i
< n_prim
; i
++) {
4261 if (!find_index_in_bvd(ddf
, vc
, i
, &n_bvd
))
4263 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
4266 st
= be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
4267 if ((st
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
4274 state
= DDF_state_degraded
;
4275 if (working
== n_prim
)
4276 state
= DDF_state_optimal
;
4282 state
= DDF_state_failed
;
4286 state
= DDF_state_failed
;
4287 else if (working
>= 2)
4288 state
= DDF_state_part_optimal
;
4291 if (!enough(10, n_prim
, array
.layout
, 1, avail
))
4292 state
= DDF_state_failed
;
4296 if (working
< n_prim
- 1)
4297 state
= DDF_state_failed
;
4300 if (working
< n_prim
- 2)
4301 state
= DDF_state_failed
;
4302 else if (working
== n_prim
- 1)
4303 state
= DDF_state_part_optimal
;
4309 static int secondary_state(int state
, int other
, int seclevel
)
4311 if (state
== DDF_state_optimal
&& other
== DDF_state_optimal
)
4312 return DDF_state_optimal
;
4313 if (seclevel
== DDF_2MIRRORED
) {
4314 if (state
== DDF_state_optimal
|| other
== DDF_state_optimal
)
4315 return DDF_state_part_optimal
;
4316 if (state
== DDF_state_failed
&& other
== DDF_state_failed
)
4317 return DDF_state_failed
;
4318 return DDF_state_degraded
;
4320 if (state
== DDF_state_failed
|| other
== DDF_state_failed
)
4321 return DDF_state_failed
;
4322 if (state
== DDF_state_degraded
|| other
== DDF_state_degraded
)
4323 return DDF_state_degraded
;
4324 return DDF_state_part_optimal
;
4328 static int get_svd_state(const struct ddf_super
*ddf
, const struct vcl
*vcl
)
4330 int state
= get_bvd_state(ddf
, &vcl
->conf
);
4332 for (i
= 1; i
< vcl
->conf
.sec_elmnt_count
; i
++) {
4333 state
= secondary_state(
4335 get_bvd_state(ddf
, vcl
->other_bvds
[i
-1]),
4342 * The state of each disk is stored in the global phys_disk structure
4343 * in phys_disk.entries[n].state.
4344 * This makes various combinations awkward.
4345 * - When a device fails in any array, it must be failed in all arrays
4346 * that include a part of this device.
4347 * - When a component is rebuilding, we cannot include it officially in the
4348 * array unless this is the only array that uses the device.
4350 * So: when transitioning:
4351 * Online -> failed, just set failed flag. monitor will propagate
4352 * spare -> online, the device might need to be added to the array.
4353 * spare -> failed, just set failed. Don't worry if in array or not.
4355 static void ddf_set_disk(struct active_array
*a
, int n
, int state
)
4357 struct ddf_super
*ddf
= a
->container
->sb
;
4358 unsigned int inst
= a
->info
.container_member
, n_bvd
;
4360 struct vd_config
*vc
= find_vdcr(ddf
, inst
, (unsigned int)n
,
4367 dprintf("%s: %d to %x\n", __func__
, n
, state
);
4369 dprintf("ddf: cannot find instance %d!!\n", inst
);
4372 /* Find the matching slot in 'info'. */
4373 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
4374 if (mdi
->disk
.raid_disk
== n
)
4377 pr_err("%s: cannot find raid disk %d\n",
4382 /* and find the 'dl' entry corresponding to that. */
4383 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4384 if (mdi
->state_fd
>= 0 &&
4385 mdi
->disk
.major
== dl
->major
&&
4386 mdi
->disk
.minor
== dl
->minor
)
4389 pr_err("%s: cannot find raid disk %d (%d/%d)\n",
4391 mdi
->disk
.major
, mdi
->disk
.minor
);
4395 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
4396 if (pd
< 0 || pd
!= dl
->pdnum
) {
4397 /* disk doesn't currently exist or has changed.
4398 * If it is now in_sync, insert it. */
4399 dprintf("%s: phys disk not found for %d: %d/%d ref %08x\n",
4400 __func__
, dl
->pdnum
, dl
->major
, dl
->minor
,
4401 be32_to_cpu(dl
->disk
.refnum
));
4402 dprintf("%s: array %u disk %u ref %08x pd %d\n",
4403 __func__
, inst
, n_bvd
,
4404 be32_to_cpu(vc
->phys_refnum
[n_bvd
]), pd
);
4405 if ((state
& DS_INSYNC
) && ! (state
& DS_FAULTY
)) {
4406 pd
= dl
->pdnum
; /* FIXME: is this really correct ? */
4407 vc
->phys_refnum
[n_bvd
] = dl
->disk
.refnum
;
4408 LBA_OFFSET(ddf
, vc
)[n_bvd
] =
4409 cpu_to_be64(mdi
->data_offset
);
4410 be16_clear(ddf
->phys
->entries
[pd
].type
,
4411 cpu_to_be16(DDF_Global_Spare
));
4412 be16_set(ddf
->phys
->entries
[pd
].type
,
4413 cpu_to_be16(DDF_Active_in_VD
));
4417 be16 old
= ddf
->phys
->entries
[pd
].state
;
4418 if (state
& DS_FAULTY
)
4419 be16_set(ddf
->phys
->entries
[pd
].state
,
4420 cpu_to_be16(DDF_Failed
));
4421 if (state
& DS_INSYNC
) {
4422 be16_set(ddf
->phys
->entries
[pd
].state
,
4423 cpu_to_be16(DDF_Online
));
4424 be16_clear(ddf
->phys
->entries
[pd
].state
,
4425 cpu_to_be16(DDF_Rebuilding
));
4427 if (!be16_eq(old
, ddf
->phys
->entries
[pd
].state
))
4431 dprintf("ddf: set_disk %d (%08x) to %x->%02x\n", n
,
4432 be32_to_cpu(dl
->disk
.refnum
), state
,
4433 be16_to_cpu(ddf
->phys
->entries
[pd
].state
));
4435 /* Now we need to check the state of the array and update
4436 * virtual_disk.entries[n].state.
4437 * It needs to be one of "optimal", "degraded", "failed".
4438 * I don't understand 'deleted' or 'missing'.
4440 state
= get_svd_state(ddf
, vcl
);
4442 if (ddf
->virt
->entries
[inst
].state
!=
4443 ((ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4445 ddf
->virt
->entries
[inst
].state
=
4446 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4451 ddf_set_updates_pending(ddf
, vc
);
4454 static void ddf_sync_metadata(struct supertype
*st
)
4457 * Write all data to all devices.
4458 * Later, we might be able to track whether only local changes
4459 * have been made, or whether any global data has been changed,
4460 * but ddf is sufficiently weird that it probably always
4461 * changes global data ....
4463 struct ddf_super
*ddf
= st
->sb
;
4464 if (!ddf
->updates_pending
)
4466 ddf
->updates_pending
= 0;
4467 __write_init_super_ddf(st
);
4468 dprintf("ddf: sync_metadata\n");
4471 static int del_from_conflist(struct vcl
**list
, const char *guid
)
4475 for (p
= list
; p
&& *p
; p
= &((*p
)->next
))
4476 if (!memcmp((*p
)->conf
.guid
, guid
, DDF_GUID_LEN
)) {
4483 static int _kill_subarray_ddf(struct ddf_super
*ddf
, const char *guid
)
4486 unsigned int vdnum
, i
;
4487 vdnum
= find_vde_by_guid(ddf
, guid
);
4488 if (vdnum
== DDF_NOTFOUND
) {
4489 pr_err("%s: could not find VD %s\n", __func__
,
4493 if (del_from_conflist(&ddf
->conflist
, guid
) == 0) {
4494 pr_err("%s: could not find conf %s\n", __func__
,
4498 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4499 for (i
= 0; i
< ddf
->max_part
; i
++)
4500 if (dl
->vlist
[i
] != NULL
&&
4501 !memcmp(dl
->vlist
[i
]->conf
.guid
, guid
,
4503 dl
->vlist
[i
] = NULL
;
4504 memset(ddf
->virt
->entries
[vdnum
].guid
, 0xff, DDF_GUID_LEN
);
4505 dprintf("%s: deleted %s\n", __func__
, guid_str(guid
));
4509 static int kill_subarray_ddf(struct supertype
*st
)
4511 struct ddf_super
*ddf
= st
->sb
;
4513 * currentconf is set in container_content_ddf,
4514 * called with subarray arg
4516 struct vcl
*victim
= ddf
->currentconf
;
4517 struct vd_config
*conf
;
4520 ddf
->currentconf
= NULL
;
4522 pr_err("%s: nothing to kill\n", __func__
);
4525 conf
= &victim
->conf
;
4526 vdnum
= find_vde_by_guid(ddf
, conf
->guid
);
4527 if (vdnum
== DDF_NOTFOUND
) {
4528 pr_err("%s: could not find VD %s\n", __func__
,
4529 guid_str(conf
->guid
));
4532 if (st
->update_tail
) {
4533 struct virtual_disk
*vd
;
4534 int len
= sizeof(struct virtual_disk
)
4535 + sizeof(struct virtual_entry
);
4538 pr_err("%s: failed to allocate %d bytes\n", __func__
,
4542 memset(vd
, 0 , len
);
4543 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
4544 vd
->populated_vdes
= cpu_to_be16(0);
4545 memcpy(vd
->entries
[0].guid
, conf
->guid
, DDF_GUID_LEN
);
4546 /* we use DDF_state_deleted as marker */
4547 vd
->entries
[0].state
= DDF_state_deleted
;
4548 append_metadata_update(st
, vd
, len
);
4550 _kill_subarray_ddf(ddf
, conf
->guid
);
4551 ddf_set_updates_pending(ddf
, NULL
);
4552 ddf_sync_metadata(st
);
4557 static void copy_matching_bvd(struct ddf_super
*ddf
,
4558 struct vd_config
*conf
,
4559 const struct metadata_update
*update
)
4562 be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
4563 unsigned int len
= ddf
->conf_rec_len
* 512;
4565 struct vd_config
*vc
;
4566 for (p
= update
->buf
; p
< update
->buf
+ update
->len
; p
+= len
) {
4567 vc
= (struct vd_config
*) p
;
4568 if (vc
->sec_elmnt_seq
== conf
->sec_elmnt_seq
) {
4569 memcpy(conf
->phys_refnum
, vc
->phys_refnum
,
4570 mppe
* (sizeof(__u32
) + sizeof(__u64
)));
4574 pr_err("%s: no match for BVD %d of %s in update\n", __func__
,
4575 conf
->sec_elmnt_seq
, guid_str(conf
->guid
));
4578 static void ddf_process_phys_update(struct supertype
*st
,
4579 struct metadata_update
*update
)
4581 struct ddf_super
*ddf
= st
->sb
;
4582 struct phys_disk
*pd
;
4585 pd
= (struct phys_disk
*)update
->buf
;
4586 ent
= be16_to_cpu(pd
->used_pdes
);
4587 if (ent
>= be16_to_cpu(ddf
->phys
->max_pdes
))
4589 if (be16_and(pd
->entries
[0].state
, cpu_to_be16(DDF_Missing
))) {
4591 /* removing this disk. */
4592 be16_set(ddf
->phys
->entries
[ent
].state
,
4593 cpu_to_be16(DDF_Missing
));
4594 for (dlp
= &ddf
->dlist
; *dlp
; dlp
= &(*dlp
)->next
) {
4595 struct dl
*dl
= *dlp
;
4596 if (dl
->pdnum
== (signed)ent
) {
4599 /* FIXME this doesn't free
4606 ddf_set_updates_pending(ddf
, NULL
);
4609 if (!all_ff(ddf
->phys
->entries
[ent
].guid
))
4611 ddf
->phys
->entries
[ent
] = pd
->entries
[0];
4612 ddf
->phys
->used_pdes
= cpu_to_be16
4613 (1 + be16_to_cpu(ddf
->phys
->used_pdes
));
4614 ddf_set_updates_pending(ddf
, NULL
);
4615 if (ddf
->add_list
) {
4616 struct active_array
*a
;
4617 struct dl
*al
= ddf
->add_list
;
4618 ddf
->add_list
= al
->next
;
4620 al
->next
= ddf
->dlist
;
4623 /* As a device has been added, we should check
4624 * for any degraded devices that might make
4625 * use of this spare */
4626 for (a
= st
->arrays
; a
; a
=a
->next
)
4627 a
->check_degraded
= 1;
4631 static void ddf_process_virt_update(struct supertype
*st
,
4632 struct metadata_update
*update
)
4634 struct ddf_super
*ddf
= st
->sb
;
4635 struct virtual_disk
*vd
;
4638 vd
= (struct virtual_disk
*)update
->buf
;
4640 if (vd
->entries
[0].state
== DDF_state_deleted
) {
4641 if (_kill_subarray_ddf(ddf
, vd
->entries
[0].guid
))
4644 ent
= find_vde_by_guid(ddf
, vd
->entries
[0].guid
);
4645 if (ent
!= DDF_NOTFOUND
) {
4646 dprintf("%s: VD %s exists already in slot %d\n",
4647 __func__
, guid_str(vd
->entries
[0].guid
),
4651 ent
= find_unused_vde(ddf
);
4652 if (ent
== DDF_NOTFOUND
)
4654 ddf
->virt
->entries
[ent
] = vd
->entries
[0];
4655 ddf
->virt
->populated_vdes
=
4658 ddf
->virt
->populated_vdes
));
4659 dprintf("%s: added VD %s in slot %d(s=%02x i=%02x)\n",
4660 __func__
, guid_str(vd
->entries
[0].guid
), ent
,
4661 ddf
->virt
->entries
[ent
].state
,
4662 ddf
->virt
->entries
[ent
].init_state
);
4664 ddf_set_updates_pending(ddf
, NULL
);
4667 static void ddf_remove_failed(struct ddf_super
*ddf
)
4669 /* Now remove any 'Failed' devices that are not part
4670 * of any VD. They will have the Transition flag set.
4671 * Once done, we need to update all dl->pdnum numbers.
4674 unsigned int pd2
= 0;
4677 for (pdnum
= 0; pdnum
< be16_to_cpu(ddf
->phys
->max_pdes
);
4679 if (be32_to_cpu(ddf
->phys
->entries
[pdnum
].refnum
) ==
4682 if (be16_and(ddf
->phys
->entries
[pdnum
].state
,
4683 cpu_to_be16(DDF_Failed
))
4684 && be16_and(ddf
->phys
->entries
[pdnum
].state
,
4685 cpu_to_be16(DDF_Transition
))) {
4686 /* skip this one unless in dlist*/
4687 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4688 if (dl
->pdnum
== (int)pdnum
)
4696 ddf
->phys
->entries
[pd2
] =
4697 ddf
->phys
->entries
[pdnum
];
4698 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4699 if (dl
->pdnum
== (int)pdnum
)
4704 ddf
->phys
->used_pdes
= cpu_to_be16(pd2
);
4705 while (pd2
< pdnum
) {
4706 memset(ddf
->phys
->entries
[pd2
].guid
, 0xff,
4712 static void ddf_update_vlist(struct ddf_super
*ddf
, struct dl
*dl
)
4715 unsigned int vn
= 0;
4716 int in_degraded
= 0;
4720 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
) {
4721 unsigned int dn
, ibvd
;
4722 const struct vd_config
*conf
;
4724 dn
= get_pd_index_from_refnum(vcl
,
4728 if (dn
== DDF_NOTFOUND
)
4730 dprintf("dev %d/%08x has %s (sec=%u) at %d\n",
4732 be32_to_cpu(dl
->disk
.refnum
),
4733 guid_str(conf
->guid
),
4734 conf
->sec_elmnt_seq
, vn
);
4735 /* Clear the Transition flag */
4737 (ddf
->phys
->entries
[dl
->pdnum
].state
,
4738 cpu_to_be16(DDF_Failed
)))
4739 be16_clear(ddf
->phys
4740 ->entries
[dl
->pdnum
].state
,
4741 cpu_to_be16(DDF_Transition
));
4742 dl
->vlist
[vn
++] = vcl
;
4743 vstate
= ddf
->virt
->entries
[vcl
->vcnum
].state
4745 if (vstate
== DDF_state_degraded
||
4746 vstate
== DDF_state_part_optimal
)
4749 while (vn
< ddf
->max_part
)
4750 dl
->vlist
[vn
++] = NULL
;
4752 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4753 cpu_to_be16(DDF_Global_Spare
));
4754 if (!be16_and(ddf
->phys
4755 ->entries
[dl
->pdnum
].type
,
4756 cpu_to_be16(DDF_Active_in_VD
))) {
4758 ->entries
[dl
->pdnum
].type
,
4759 cpu_to_be16(DDF_Active_in_VD
));
4762 ->entries
[dl
->pdnum
]
4769 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4770 cpu_to_be16(DDF_Global_Spare
));
4771 be16_set(ddf
->phys
->entries
[dl
->pdnum
].type
,
4772 cpu_to_be16(DDF_Spare
));
4774 if (!dl
->vlist
[0] && !dl
->spare
) {
4775 be16_set(ddf
->phys
->entries
[dl
->pdnum
].type
,
4776 cpu_to_be16(DDF_Global_Spare
));
4777 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4778 cpu_to_be16(DDF_Spare
));
4779 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4780 cpu_to_be16(DDF_Active_in_VD
));
4784 static void ddf_process_conf_update(struct supertype
*st
,
4785 struct metadata_update
*update
)
4787 struct ddf_super
*ddf
= st
->sb
;
4788 struct vd_config
*vc
;
4792 unsigned int pdnum
, len
;
4794 vc
= (struct vd_config
*)update
->buf
;
4795 len
= ddf
->conf_rec_len
* 512;
4796 if ((unsigned int)update
->len
!= len
* vc
->sec_elmnt_count
) {
4797 pr_err("%s: %s: insufficient data (%d) for %u BVDs\n",
4798 __func__
, guid_str(vc
->guid
), update
->len
,
4799 vc
->sec_elmnt_count
);
4802 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
4803 if (memcmp(vcl
->conf
.guid
, vc
->guid
, DDF_GUID_LEN
) == 0)
4805 dprintf("%s: conf update for %s (%s)\n", __func__
,
4806 guid_str(vc
->guid
), (vcl
? "old" : "new"));
4808 /* An update, just copy the phys_refnum and lba_offset
4813 copy_matching_bvd(ddf
, &vcl
->conf
, update
);
4814 for (k
= 0; k
< be16_to_cpu(vc
->prim_elmnt_count
); k
++)
4815 dprintf("BVD %u has %08x at %llu\n", 0,
4816 be32_to_cpu(vcl
->conf
.phys_refnum
[k
]),
4817 be64_to_cpu(LBA_OFFSET(ddf
,
4819 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++) {
4820 copy_matching_bvd(ddf
, vcl
->other_bvds
[i
-1],
4822 for (k
= 0; k
< be16_to_cpu(
4823 vc
->prim_elmnt_count
); k
++)
4824 dprintf("BVD %u has %08x at %llu\n", i
,
4826 (vcl
->other_bvds
[i
-1]->
4831 vcl
->other_bvds
[i
-1])[k
]));
4838 vcl
= update
->space
;
4839 update
->space
= NULL
;
4840 vcl
->next
= ddf
->conflist
;
4841 memcpy(&vcl
->conf
, vc
, len
);
4842 ent
= find_vde_by_guid(ddf
, vc
->guid
);
4843 if (ent
== DDF_NOTFOUND
)
4846 ddf
->conflist
= vcl
;
4847 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++)
4848 memcpy(vcl
->other_bvds
[i
-1],
4849 update
->buf
+ len
* i
, len
);
4851 /* Set DDF_Transition on all Failed devices - to help
4852 * us detect those that are no longer in use
4854 for (pdnum
= 0; pdnum
< be16_to_cpu(ddf
->phys
->max_pdes
);
4856 if (be16_and(ddf
->phys
->entries
[pdnum
].state
,
4857 cpu_to_be16(DDF_Failed
)))
4858 be16_set(ddf
->phys
->entries
[pdnum
].state
,
4859 cpu_to_be16(DDF_Transition
));
4861 /* Now make sure vlist is correct for each dl. */
4862 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4863 ddf_update_vlist(ddf
, dl
);
4864 ddf_remove_failed(ddf
);
4866 ddf_set_updates_pending(ddf
, vc
);
4869 static void ddf_process_update(struct supertype
*st
,
4870 struct metadata_update
*update
)
4872 /* Apply this update to the metadata.
4873 * The first 4 bytes are a DDF_*_MAGIC which guides
4875 * Possible update are:
4876 * DDF_PHYS_RECORDS_MAGIC
4877 * Add a new physical device or remove an old one.
4878 * Changes to this record only happen implicitly.
4879 * used_pdes is the device number.
4880 * DDF_VIRT_RECORDS_MAGIC
4881 * Add a new VD. Possibly also change the 'access' bits.
4882 * populated_vdes is the entry number.
4884 * New or updated VD. the VIRT_RECORD must already
4885 * exist. For an update, phys_refnum and lba_offset
4886 * (at least) are updated, and the VD_CONF must
4887 * be written to precisely those devices listed with
4889 * DDF_SPARE_ASSIGN_MAGIC
4890 * replacement Spare Assignment Record... but for which device?
4893 * - to create a new array, we send a VIRT_RECORD and
4894 * a VD_CONF. Then assemble and start the array.
4895 * - to activate a spare we send a VD_CONF to add the phys_refnum
4896 * and offset. This will also mark the spare as active with
4897 * a spare-assignment record.
4899 be32
*magic
= (be32
*)update
->buf
;
4901 dprintf("Process update %x\n", be32_to_cpu(*magic
));
4903 if (be32_eq(*magic
, DDF_PHYS_RECORDS_MAGIC
)) {
4904 if (update
->len
== (sizeof(struct phys_disk
) +
4905 sizeof(struct phys_disk_entry
)))
4906 ddf_process_phys_update(st
, update
);
4907 } else if (be32_eq(*magic
, DDF_VIRT_RECORDS_MAGIC
)) {
4908 if (update
->len
== (sizeof(struct virtual_disk
) +
4909 sizeof(struct virtual_entry
)))
4910 ddf_process_virt_update(st
, update
);
4911 } else if (be32_eq(*magic
, DDF_VD_CONF_MAGIC
)) {
4912 ddf_process_conf_update(st
, update
);
4914 /* case DDF_SPARE_ASSIGN_MAGIC */
4917 static void ddf_prepare_update(struct supertype
*st
,
4918 struct metadata_update
*update
)
4920 /* This update arrived at managemon.
4921 * We are about to pass it to monitor.
4922 * If a malloc is needed, do it here.
4924 struct ddf_super
*ddf
= st
->sb
;
4925 be32
*magic
= (be32
*)update
->buf
;
4926 if (be32_eq(*magic
, DDF_VD_CONF_MAGIC
)) {
4928 struct vd_config
*conf
= (struct vd_config
*) update
->buf
;
4929 if (posix_memalign(&update
->space
, 512,
4930 offsetof(struct vcl
, conf
)
4931 + ddf
->conf_rec_len
* 512) != 0) {
4932 update
->space
= NULL
;
4935 vcl
= update
->space
;
4936 vcl
->conf
.sec_elmnt_count
= conf
->sec_elmnt_count
;
4937 if (alloc_other_bvds(ddf
, vcl
) != 0) {
4938 free(update
->space
);
4939 update
->space
= NULL
;
4945 * Check degraded state of a RAID10.
4946 * returns 2 for good, 1 for degraded, 0 for failed, and -1 for error
4948 static int raid10_degraded(struct mdinfo
*info
)
4956 n_prim
= info
->array
.layout
& ~0x100;
4957 n_bvds
= info
->array
.raid_disks
/ n_prim
;
4958 found
= xmalloc(n_bvds
);
4961 memset(found
, 0, n_bvds
);
4962 for (d
= info
->devs
; d
; d
= d
->next
) {
4963 i
= d
->disk
.raid_disk
/ n_prim
;
4965 pr_err("%s: BUG: invalid raid disk\n", __func__
);
4968 if (d
->state_fd
> 0)
4972 for (i
= 0; i
< n_bvds
; i
++)
4974 dprintf("%s: BVD %d/%d failed\n", __func__
, i
, n_bvds
);
4977 } else if (found
[i
] < n_prim
) {
4978 dprintf("%s: BVD %d/%d degraded\n", __func__
, i
,
4988 * Check if the array 'a' is degraded but not failed.
4989 * If it is, find as many spares as are available and needed and
4990 * arrange for their inclusion.
4991 * We only choose devices which are not already in the array,
4992 * and prefer those with a spare-assignment to this array.
4993 * Otherwise we choose global spares - assuming always that
4994 * there is enough room.
4995 * For each spare that we assign, we return an 'mdinfo' which
4996 * describes the position for the device in the array.
4997 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
4998 * the new phys_refnum and lba_offset values.
5000 * Only worry about BVDs at the moment.
5002 static struct mdinfo
*ddf_activate_spare(struct active_array
*a
,
5003 struct metadata_update
**updates
)
5007 struct ddf_super
*ddf
= a
->container
->sb
;
5009 struct mdinfo
*rv
= NULL
;
5011 struct metadata_update
*mu
;
5016 struct vd_config
*vc
;
5019 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
5020 if ((d
->curr_state
& DS_FAULTY
) &&
5022 /* wait for Removal to happen */
5024 if (d
->state_fd
>= 0)
5028 dprintf("%s: working=%d (%d) level=%d\n", __func__
, working
,
5029 a
->info
.array
.raid_disks
,
5030 a
->info
.array
.level
);
5031 if (working
== a
->info
.array
.raid_disks
)
5032 return NULL
; /* array not degraded */
5033 switch (a
->info
.array
.level
) {
5036 return NULL
; /* failed */
5040 if (working
< a
->info
.array
.raid_disks
- 1)
5041 return NULL
; /* failed */
5044 if (working
< a
->info
.array
.raid_disks
- 2)
5045 return NULL
; /* failed */
5048 if (raid10_degraded(&a
->info
) < 1)
5051 default: /* concat or stripe */
5052 return NULL
; /* failed */
5055 /* For each slot, if it is not working, find a spare */
5057 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
5058 for (d
= a
->info
.devs
; d
; d
= d
->next
)
5059 if (d
->disk
.raid_disk
== i
)
5061 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
5062 if (d
&& (d
->state_fd
>= 0))
5065 /* OK, this device needs recovery. Find a spare */
5067 for ( ; dl
; dl
= dl
->next
) {
5068 unsigned long long esize
;
5069 unsigned long long pos
;
5072 int is_dedicated
= 0;
5077 state
= ddf
->phys
->entries
[dl
->pdnum
].state
;
5079 cpu_to_be16(DDF_Failed
|DDF_Missing
)) ||
5081 cpu_to_be16(DDF_Online
)))
5084 /* If in this array, skip */
5085 for (d2
= a
->info
.devs
; d2
; d2
= d2
->next
)
5086 if (d2
->state_fd
>= 0 &&
5087 d2
->disk
.major
== dl
->major
&&
5088 d2
->disk
.minor
== dl
->minor
) {
5089 dprintf("%x:%x (%08x) already in array\n",
5090 dl
->major
, dl
->minor
,
5091 be32_to_cpu(dl
->disk
.refnum
));
5096 if (be16_and(ddf
->phys
->entries
[dl
->pdnum
].type
,
5097 cpu_to_be16(DDF_Spare
))) {
5098 /* Check spare assign record */
5100 if (dl
->spare
->type
& DDF_spare_dedicated
) {
5101 /* check spare_ents for guid */
5108 if (memcmp(dl
->spare
->spare_ents
[j
].guid
,
5109 ddf
->virt
->entries
[a
->info
.container_member
].guid
,
5116 } else if (be16_and(ddf
->phys
->entries
[dl
->pdnum
].type
,
5117 cpu_to_be16(DDF_Global_Spare
))) {
5119 } else if (!be16_and(ddf
->phys
5120 ->entries
[dl
->pdnum
].state
,
5121 cpu_to_be16(DDF_Failed
))) {
5122 /* we can possibly use some of this */
5125 if ( ! (is_dedicated
||
5126 (is_global
&& global_ok
))) {
5127 dprintf("%x:%x not suitable: %d %d\n", dl
->major
, dl
->minor
,
5128 is_dedicated
, is_global
);
5132 /* We are allowed to use this device - is there space?
5133 * We need a->info.component_size sectors */
5134 esize
= a
->info
.component_size
;
5135 pos
= find_space(ddf
, dl
, INVALID_SECTORS
, &esize
);
5137 if (esize
< a
->info
.component_size
) {
5138 dprintf("%x:%x has no room: %llu %llu\n",
5139 dl
->major
, dl
->minor
,
5140 esize
, a
->info
.component_size
);
5145 /* Cool, we have a device with some space at pos */
5146 di
= xcalloc(1, sizeof(*di
));
5147 di
->disk
.number
= i
;
5148 di
->disk
.raid_disk
= i
;
5149 di
->disk
.major
= dl
->major
;
5150 di
->disk
.minor
= dl
->minor
;
5152 di
->recovery_start
= 0;
5153 di
->data_offset
= pos
;
5154 di
->component_size
= a
->info
.component_size
;
5157 dprintf("%x:%x (%08x) to be %d at %llu\n",
5158 dl
->major
, dl
->minor
,
5159 be32_to_cpu(dl
->disk
.refnum
), i
, pos
);
5163 if (!dl
&& ! global_ok
) {
5164 /* not enough dedicated spares, try global */
5172 /* No spares found */
5174 /* Now 'rv' has a list of devices to return.
5175 * Create a metadata_update record to update the
5176 * phys_refnum and lba_offset values
5178 vc
= find_vdcr(ddf
, a
->info
.container_member
, rv
->disk
.raid_disk
,
5183 mu
= xmalloc(sizeof(*mu
));
5184 if (posix_memalign(&mu
->space
, 512, sizeof(struct vcl
)) != 0) {
5189 mu
->len
= ddf
->conf_rec_len
* 512 * vcl
->conf
.sec_elmnt_count
;
5190 mu
->buf
= xmalloc(mu
->len
);
5192 mu
->space_list
= NULL
;
5193 mu
->next
= *updates
;
5194 memcpy(mu
->buf
, &vcl
->conf
, ddf
->conf_rec_len
* 512);
5195 for (j
= 1; j
< vcl
->conf
.sec_elmnt_count
; j
++)
5196 memcpy(mu
->buf
+ j
* ddf
->conf_rec_len
* 512,
5197 vcl
->other_bvds
[j
-1], ddf
->conf_rec_len
* 512);
5199 vc
= (struct vd_config
*)mu
->buf
;
5200 for (di
= rv
; di
; di
= di
->next
) {
5201 unsigned int i_sec
, i_prim
;
5202 i_sec
= di
->disk
.raid_disk
5203 / be16_to_cpu(vcl
->conf
.prim_elmnt_count
);
5204 i_prim
= di
->disk
.raid_disk
5205 % be16_to_cpu(vcl
->conf
.prim_elmnt_count
);
5206 vc
= (struct vd_config
*)(mu
->buf
5207 + i_sec
* ddf
->conf_rec_len
* 512);
5208 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
5209 if (dl
->major
== di
->disk
.major
5210 && dl
->minor
== di
->disk
.minor
)
5212 if (!dl
|| dl
->pdnum
< 0) {
5213 pr_err("%s: BUG: can't find disk %d (%d/%d)\n",
5214 __func__
, di
->disk
.raid_disk
,
5215 di
->disk
.major
, di
->disk
.minor
);
5218 vc
->phys_refnum
[i_prim
] = ddf
->phys
->entries
[dl
->pdnum
].refnum
;
5219 LBA_OFFSET(ddf
, vc
)[i_prim
] = cpu_to_be64(di
->data_offset
);
5220 dprintf("BVD %u gets %u: %08x at %llu\n", i_sec
, i_prim
,
5221 be32_to_cpu(vc
->phys_refnum
[i_prim
]),
5222 be64_to_cpu(LBA_OFFSET(ddf
, vc
)[i_prim
]));
5227 #endif /* MDASSEMBLE */
5229 static int ddf_level_to_layout(int level
)
5236 return ALGORITHM_LEFT_SYMMETRIC
;
5238 return ALGORITHM_ROTATING_N_CONTINUE
;
5246 static void default_geometry_ddf(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5248 if (level
&& *level
== UnSet
)
5249 *level
= LEVEL_CONTAINER
;
5251 if (level
&& layout
&& *layout
== UnSet
)
5252 *layout
= ddf_level_to_layout(*level
);
5255 struct superswitch super_ddf
= {
5257 .examine_super
= examine_super_ddf
,
5258 .brief_examine_super
= brief_examine_super_ddf
,
5259 .brief_examine_subarrays
= brief_examine_subarrays_ddf
,
5260 .export_examine_super
= export_examine_super_ddf
,
5261 .detail_super
= detail_super_ddf
,
5262 .brief_detail_super
= brief_detail_super_ddf
,
5263 .validate_geometry
= validate_geometry_ddf
,
5264 .write_init_super
= write_init_super_ddf
,
5265 .add_to_super
= add_to_super_ddf
,
5266 .remove_from_super
= remove_from_super_ddf
,
5267 .load_container
= load_container_ddf
,
5268 .copy_metadata
= copy_metadata_ddf
,
5269 .kill_subarray
= kill_subarray_ddf
,
5271 .match_home
= match_home_ddf
,
5272 .uuid_from_super
= uuid_from_super_ddf
,
5273 .getinfo_super
= getinfo_super_ddf
,
5274 .update_super
= update_super_ddf
,
5276 .avail_size
= avail_size_ddf
,
5278 .compare_super
= compare_super_ddf
,
5280 .load_super
= load_super_ddf
,
5281 .init_super
= init_super_ddf
,
5282 .store_super
= store_super_ddf
,
5283 .free_super
= free_super_ddf
,
5284 .match_metadata_desc
= match_metadata_desc_ddf
,
5285 .container_content
= container_content_ddf
,
5286 .default_geometry
= default_geometry_ddf
,
5292 .open_new
= ddf_open_new
,
5293 .set_array_state
= ddf_set_array_state
,
5294 .set_disk
= ddf_set_disk
,
5295 .sync_metadata
= ddf_sync_metadata
,
5296 .process_update
= ddf_process_update
,
5297 .prepare_update
= ddf_prepare_update
,
5298 .activate_spare
= ddf_activate_spare
,