2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2006-2009 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neil@brown.name>
24 * Specifications for DDF takes from Common RAID DDF Specification Revision 1.2
25 * (July 28 2006). Reused by permission of SNIA.
28 #define HAVE_STDINT_H 1
34 /* a non-official T10 name for creation GUIDs */
35 static char T10
[] = "Linux-MD";
37 /* DDF timestamps are 1980 based, so we need to add
38 * second-in-decade-of-seventies to convert to linux timestamps.
39 * 10 years with 2 leap years.
41 #define DECADE (3600*24*(365*10+2))
44 const unsigned char *buf
,
47 #define DDF_NOTFOUND (~0U)
48 #define DDF_CONTAINER (DDF_NOTFOUND-1)
50 /* The DDF metadata handling.
51 * DDF metadata lives at the end of the device.
52 * The last 512 byte block provides an 'anchor' which is used to locate
53 * the rest of the metadata which usually lives immediately behind the anchor.
56 * - all multibyte numeric fields are bigendian.
57 * - all strings are space padded.
61 typedef struct __be16
{
64 #define be16_eq(x, y) ((x)._v16 == (y)._v16)
66 typedef struct __be32
{
69 #define be32_eq(x, y) ((x)._v32 == (y)._v32)
71 typedef struct __be64
{
74 #define be64_eq(x, y) ((x)._v64 == (y)._v64)
76 #define be16_to_cpu(be) __be16_to_cpu((be)._v16)
77 static inline be16
cpu_to_be16(__u16 x
)
79 be16 be
= { ._v16
= __cpu_to_be16(x
) };
83 #define be32_to_cpu(be) __be32_to_cpu((be)._v32)
84 static inline be32
cpu_to_be32(__u32 x
)
86 be32 be
= { ._v32
= __cpu_to_be32(x
) };
90 #define be64_to_cpu(be) __be64_to_cpu((be)._v64)
91 static inline be64
cpu_to_be64(__u64 x
)
93 be64 be
= { ._v64
= __cpu_to_be64(x
) };
97 /* Primary Raid Level (PRL) */
98 #define DDF_RAID0 0x00
99 #define DDF_RAID1 0x01
100 #define DDF_RAID3 0x03
101 #define DDF_RAID4 0x04
102 #define DDF_RAID5 0x05
103 #define DDF_RAID1E 0x11
104 #define DDF_JBOD 0x0f
105 #define DDF_CONCAT 0x1f
106 #define DDF_RAID5E 0x15
107 #define DDF_RAID5EE 0x25
108 #define DDF_RAID6 0x06
110 /* Raid Level Qualifier (RLQ) */
111 #define DDF_RAID0_SIMPLE 0x00
112 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
113 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
114 #define DDF_RAID3_0 0x00 /* parity in first extent */
115 #define DDF_RAID3_N 0x01 /* parity in last extent */
116 #define DDF_RAID4_0 0x00 /* parity in first extent */
117 #define DDF_RAID4_N 0x01 /* parity in last extent */
118 /* these apply to raid5e and raid5ee as well */
119 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
120 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
121 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
122 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
124 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
125 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
127 /* Secondary RAID Level (SRL) */
128 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
129 #define DDF_2MIRRORED 0x01
130 #define DDF_2CONCAT 0x02
131 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
134 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
135 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
136 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
137 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
138 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
139 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
140 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
141 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
142 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
143 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
145 #define DDF_GUID_LEN 24
146 #define DDF_REVISION_0 "01.00.00"
147 #define DDF_REVISION_2 "01.02.00"
150 __u32 magic
; /* DDF_HEADER_MAGIC */
152 char guid
[DDF_GUID_LEN
];
153 char revision
[8]; /* 01.02.00 */
154 __u32 seq
; /* starts at '1' */
159 __u8 pad0
; /* 0xff */
160 __u8 pad1
[12]; /* 12 * 0xff */
161 /* 64 bytes so far */
162 __u8 header_ext
[32]; /* reserved: fill with 0xff */
166 __u8 pad2
[3]; /* 0xff */
167 __u32 workspace_len
; /* sectors for vendor space -
168 * at least 32768(sectors) */
170 __u16 max_pd_entries
; /* one of 15, 63, 255, 1023, 4095 */
171 __u16 max_vd_entries
; /* 2^(4,6,8,10,12)-1 : i.e. as above */
172 __u16 max_partitions
; /* i.e. max num of configuration
173 record entries per disk */
174 __u16 config_record_len
; /* 1 +ROUNDUP(max_primary_element_entries
176 __u16 max_primary_element_entries
; /* 16, 64, 256, 1024, or 4096 */
177 __u8 pad3
[54]; /* 0xff */
178 /* 192 bytes so far */
179 __u32 controller_section_offset
;
180 __u32 controller_section_length
;
181 __u32 phys_section_offset
;
182 __u32 phys_section_length
;
183 __u32 virt_section_offset
;
184 __u32 virt_section_length
;
185 __u32 config_section_offset
;
186 __u32 config_section_length
;
187 __u32 data_section_offset
;
188 __u32 data_section_length
;
189 __u32 bbm_section_offset
;
190 __u32 bbm_section_length
;
191 __u32 diag_space_offset
;
192 __u32 diag_space_length
;
195 /* 256 bytes so far */
196 __u8 pad4
[256]; /* 0xff */
200 #define DDF_HEADER_ANCHOR 0x00
201 #define DDF_HEADER_PRIMARY 0x01
202 #define DDF_HEADER_SECONDARY 0x02
204 /* The content of the 'controller section' - global scope */
205 struct ddf_controller_data
{
206 __u32 magic
; /* DDF_CONTROLLER_MAGIC */
208 char guid
[DDF_GUID_LEN
];
209 struct controller_type
{
216 __u8 pad
[8]; /* 0xff */
217 __u8 vendor_data
[448];
220 /* The content of phys_section - global scope */
222 __u32 magic
; /* DDF_PHYS_RECORDS_MAGIC */
227 struct phys_disk_entry
{
228 char guid
[DDF_GUID_LEN
];
232 __u64 config_size
; /* DDF structures must be after here */
233 char path
[18]; /* another horrible structure really */
238 /* phys_disk_entry.type is a bitmap - bigendian remember */
239 #define DDF_Forced_PD_GUID 1
240 #define DDF_Active_in_VD 2
241 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
242 #define DDF_Spare 8 /* overrides Global_spare */
243 #define DDF_Foreign 16
244 #define DDF_Legacy 32 /* no DDF on this device */
246 #define DDF_Interface_mask 0xf00
247 #define DDF_Interface_SCSI 0x100
248 #define DDF_Interface_SAS 0x200
249 #define DDF_Interface_SATA 0x300
250 #define DDF_Interface_FC 0x400
252 /* phys_disk_entry.state is a bigendian bitmap */
254 #define DDF_Failed 2 /* overrides 1,4,8 */
255 #define DDF_Rebuilding 4
256 #define DDF_Transition 8
258 #define DDF_ReadErrors 32
259 #define DDF_Missing 64
261 /* The content of the virt_section global scope */
262 struct virtual_disk
{
263 __u32 magic
; /* DDF_VIRT_RECORDS_MAGIC */
265 __u16 populated_vdes
;
268 struct virtual_entry
{
269 char guid
[DDF_GUID_LEN
];
271 __u16 pad0
; /* 0xffff */
281 /* virtual_entry.type is a bitmap - bigendian */
283 #define DDF_Enforce_Groups 2
284 #define DDF_Unicode 4
285 #define DDF_Owner_Valid 8
287 /* virtual_entry.state is a bigendian bitmap */
288 #define DDF_state_mask 0x7
289 #define DDF_state_optimal 0x0
290 #define DDF_state_degraded 0x1
291 #define DDF_state_deleted 0x2
292 #define DDF_state_missing 0x3
293 #define DDF_state_failed 0x4
294 #define DDF_state_part_optimal 0x5
296 #define DDF_state_morphing 0x8
297 #define DDF_state_inconsistent 0x10
299 /* virtual_entry.init_state is a bigendian bitmap */
300 #define DDF_initstate_mask 0x03
301 #define DDF_init_not 0x00
302 #define DDF_init_quick 0x01 /* initialisation is progress.
303 * i.e. 'state_inconsistent' */
304 #define DDF_init_full 0x02
306 #define DDF_access_mask 0xc0
307 #define DDF_access_rw 0x00
308 #define DDF_access_ro 0x80
309 #define DDF_access_blocked 0xc0
311 /* The content of the config_section - local scope
312 * It has multiple records each config_record_len sectors
313 * They can be vd_config or spare_assign
317 __u32 magic
; /* DDF_VD_CONF_MAGIC */
319 char guid
[DDF_GUID_LEN
];
323 __u16 prim_elmnt_count
;
324 __u8 chunk_shift
; /* 0 == 512, 1==1024 etc */
327 __u8 sec_elmnt_count
;
330 __u64 blocks
; /* blocks per component could be different
331 * on different component devices...(only
332 * for concat I hope) */
333 __u64 array_blocks
; /* blocks in array */
341 __u8 v0
[32]; /* reserved- 0xff */
342 __u8 v1
[32]; /* reserved- 0xff */
343 __u8 v2
[16]; /* reserved- 0xff */
344 __u8 v3
[16]; /* reserved- 0xff */
346 __u32 phys_refnum
[0]; /* refnum of each disk in sequence */
347 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
348 bvd are always the same size */
350 #define LBA_OFFSET(ddf, vd) ((__u64 *) &(vd)->phys_refnum[(ddf)->mppe])
352 /* vd_config.cache_pol[7] is a bitmap */
353 #define DDF_cache_writeback 1 /* else writethrough */
354 #define DDF_cache_wadaptive 2 /* only applies if writeback */
355 #define DDF_cache_readahead 4
356 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
357 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
358 #define DDF_cache_wallowed 32 /* enable write caching */
359 #define DDF_cache_rallowed 64 /* enable read caching */
361 struct spare_assign
{
362 __u32 magic
; /* DDF_SPARE_ASSIGN_MAGIC */
367 __u16 populated
; /* SAEs used */
368 __u16 max
; /* max SAEs */
370 struct spare_assign_entry
{
371 char guid
[DDF_GUID_LEN
];
372 __u16 secondary_element
;
376 /* spare_assign.type is a bitmap */
377 #define DDF_spare_dedicated 0x1 /* else global */
378 #define DDF_spare_revertible 0x2 /* else committable */
379 #define DDF_spare_active 0x4 /* else not active */
380 #define DDF_spare_affinity 0x8 /* enclosure affinity */
382 /* The data_section contents - local scope */
384 __u32 magic
; /* DDF_PHYS_DATA_MAGIC */
386 char guid
[DDF_GUID_LEN
];
387 __u32 refnum
; /* crc of some magic drive data ... */
388 __u8 forced_ref
; /* set when above was not result of magic */
389 __u8 forced_guid
; /* set if guid was forced rather than magic */
394 /* bbm_section content */
395 struct bad_block_log
{
402 struct mapped_block
{
403 __u64 defective_start
;
404 __u32 replacement_start
;
410 /* Struct for internally holding ddf structures */
411 /* The DDF structure stored on each device is potentially
412 * quite different, as some data is global and some is local.
413 * The global data is:
416 * - Physical disk records
417 * - Virtual disk records
419 * - Configuration records
420 * - Physical Disk data section
421 * ( and Bad block and vendor which I don't care about yet).
423 * The local data is parsed into separate lists as it is read
424 * and reconstructed for writing. This means that we only need
425 * to make config changes once and they are automatically
426 * propagated to all devices.
427 * Note that the ddf_super has space of the conf and disk data
428 * for this disk and also for a list of all such data.
429 * The list is only used for the superblock that is being
430 * built in Create or Assemble to describe the whole array.
433 struct ddf_header anchor
, primary
, secondary
;
434 struct ddf_controller_data controller
;
435 struct ddf_header
*active
;
436 struct phys_disk
*phys
;
437 struct virtual_disk
*virt
;
439 unsigned int max_part
, mppe
, conf_rec_len
;
447 unsigned int vcnum
; /* index into ->virt */
448 struct vd_config
**other_bvds
;
449 __u64
*block_sizes
; /* NULL if all the same */
452 struct vd_config conf
;
453 } *conflist
, *currentconf
;
462 unsigned long long size
; /* sectors */
463 unsigned long long primary_lba
; /* sectors */
464 unsigned long long secondary_lba
; /* sectors */
465 unsigned long long workspace_lba
; /* sectors */
466 int pdnum
; /* index in ->phys */
467 struct spare_assign
*spare
;
468 void *mdupdate
; /* hold metadata update */
470 /* These fields used by auto-layout */
471 int raiddisk
; /* slot to fill in autolayout */
475 struct disk_data disk
;
476 struct vcl
*vlist
[0]; /* max_part in size */
481 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
485 static int all_ff(const char *guid
);
486 static void pr_state(struct ddf_super
*ddf
, const char *msg
)
489 dprintf("%s/%s: ", __func__
, msg
);
490 for (i
= 0; i
< __be16_to_cpu(ddf
->active
->max_vd_entries
); i
++) {
491 if (all_ff(ddf
->virt
->entries
[i
].guid
))
493 dprintf("%u(s=%02x i=%02x) ", i
,
494 ddf
->virt
->entries
[i
].state
,
495 ddf
->virt
->entries
[i
].init_state
);
500 static void pr_state(const struct ddf_super
*ddf
, const char *msg
) {}
503 static void _ddf_set_updates_pending(struct ddf_super
*ddf
, const char *func
)
505 ddf
->updates_pending
= 1;
506 ddf
->active
->seq
= __cpu_to_be32((__be32_to_cpu(ddf
->active
->seq
)+1));
510 #define ddf_set_updates_pending(x) _ddf_set_updates_pending((x), __func__)
512 static unsigned int get_pd_index_from_refnum(const struct vcl
*vc
,
513 __u32 refnum
, unsigned int nmax
,
514 const struct vd_config
**bvd
,
517 static unsigned int calc_crc(void *buf
, int len
)
519 /* crcs are always at the same place as in the ddf_header */
520 struct ddf_header
*ddf
= buf
;
521 __u32 oldcrc
= ddf
->crc
;
523 ddf
->crc
= 0xffffffff;
525 newcrc
= crc32(0, buf
, len
);
527 /* The crc is store (like everything) bigendian, so convert
528 * here for simplicity
530 return __cpu_to_be32(newcrc
);
533 #define DDF_INVALID_LEVEL 0xff
534 #define DDF_NO_SECONDARY 0xff
535 static int err_bad_md_layout(const mdu_array_info_t
*array
)
537 pr_err("RAID%d layout %x with %d disks is unsupported for DDF\n",
538 array
->level
, array
->layout
, array
->raid_disks
);
542 static int layout_md2ddf(const mdu_array_info_t
*array
,
543 struct vd_config
*conf
)
545 __u16 prim_elmnt_count
= __cpu_to_be16(array
->raid_disks
);
546 __u8 prl
= DDF_INVALID_LEVEL
, rlq
= 0;
547 __u8 sec_elmnt_count
= 1;
548 __u8 srl
= DDF_NO_SECONDARY
;
550 switch (array
->level
) {
555 rlq
= DDF_RAID0_SIMPLE
;
559 switch (array
->raid_disks
) {
561 rlq
= DDF_RAID1_SIMPLE
;
564 rlq
= DDF_RAID1_MULTI
;
567 return err_bad_md_layout(array
);
572 if (array
->layout
!= 0)
573 return err_bad_md_layout(array
);
578 switch (array
->layout
) {
579 case ALGORITHM_LEFT_ASYMMETRIC
:
580 rlq
= DDF_RAID5_N_RESTART
;
582 case ALGORITHM_RIGHT_ASYMMETRIC
:
583 rlq
= DDF_RAID5_0_RESTART
;
585 case ALGORITHM_LEFT_SYMMETRIC
:
586 rlq
= DDF_RAID5_N_CONTINUE
;
588 case ALGORITHM_RIGHT_SYMMETRIC
:
589 /* not mentioned in standard */
591 return err_bad_md_layout(array
);
596 switch (array
->layout
) {
597 case ALGORITHM_ROTATING_N_RESTART
:
598 rlq
= DDF_RAID5_N_RESTART
;
600 case ALGORITHM_ROTATING_ZERO_RESTART
:
601 rlq
= DDF_RAID6_0_RESTART
;
603 case ALGORITHM_ROTATING_N_CONTINUE
:
604 rlq
= DDF_RAID5_N_CONTINUE
;
607 return err_bad_md_layout(array
);
612 if (array
->raid_disks
% 2 == 0 && array
->layout
== 0x102) {
613 rlq
= DDF_RAID1_SIMPLE
;
614 prim_elmnt_count
= __cpu_to_be16(2);
615 sec_elmnt_count
= array
->raid_disks
/ 2;
616 } else if (array
->raid_disks
% 3 == 0
617 && array
->layout
== 0x103) {
618 rlq
= DDF_RAID1_MULTI
;
619 prim_elmnt_count
= __cpu_to_be16(3);
620 sec_elmnt_count
= array
->raid_disks
/ 3;
622 return err_bad_md_layout(array
);
627 return err_bad_md_layout(array
);
630 conf
->prim_elmnt_count
= prim_elmnt_count
;
633 conf
->sec_elmnt_count
= sec_elmnt_count
;
637 static int err_bad_ddf_layout(const struct vd_config
*conf
)
639 pr_err("DDF RAID %u qualifier %u with %u disks is unsupported\n",
640 conf
->prl
, conf
->rlq
, __be16_to_cpu(conf
->prim_elmnt_count
));
644 static int layout_ddf2md(const struct vd_config
*conf
,
645 mdu_array_info_t
*array
)
647 int level
= LEVEL_UNSUPPORTED
;
649 int raiddisks
= __be16_to_cpu(conf
->prim_elmnt_count
);
651 if (conf
->sec_elmnt_count
> 1) {
652 /* see also check_secondary() */
653 if (conf
->prl
!= DDF_RAID1
||
654 (conf
->srl
!= DDF_2STRIPED
&& conf
->srl
!= DDF_2SPANNED
)) {
655 pr_err("Unsupported secondary RAID level %u/%u\n",
656 conf
->prl
, conf
->srl
);
659 if (raiddisks
== 2 && conf
->rlq
== DDF_RAID1_SIMPLE
)
661 else if (raiddisks
== 3 && conf
->rlq
== DDF_RAID1_MULTI
)
664 return err_bad_ddf_layout(conf
);
665 raiddisks
*= conf
->sec_elmnt_count
;
672 level
= LEVEL_LINEAR
;
675 if (conf
->rlq
!= DDF_RAID0_SIMPLE
)
676 return err_bad_ddf_layout(conf
);
680 if (!((conf
->rlq
== DDF_RAID1_SIMPLE
&& raiddisks
== 2) ||
681 (conf
->rlq
== DDF_RAID1_MULTI
&& raiddisks
== 3)))
682 return err_bad_ddf_layout(conf
);
686 if (conf
->rlq
!= DDF_RAID4_N
)
687 return err_bad_ddf_layout(conf
);
692 case DDF_RAID5_N_RESTART
:
693 layout
= ALGORITHM_LEFT_ASYMMETRIC
;
695 case DDF_RAID5_0_RESTART
:
696 layout
= ALGORITHM_RIGHT_ASYMMETRIC
;
698 case DDF_RAID5_N_CONTINUE
:
699 layout
= ALGORITHM_LEFT_SYMMETRIC
;
702 return err_bad_ddf_layout(conf
);
708 case DDF_RAID5_N_RESTART
:
709 layout
= ALGORITHM_ROTATING_N_RESTART
;
711 case DDF_RAID6_0_RESTART
:
712 layout
= ALGORITHM_ROTATING_ZERO_RESTART
;
714 case DDF_RAID5_N_CONTINUE
:
715 layout
= ALGORITHM_ROTATING_N_CONTINUE
;
718 return err_bad_ddf_layout(conf
);
723 return err_bad_ddf_layout(conf
);
727 array
->level
= level
;
728 array
->layout
= layout
;
729 array
->raid_disks
= raiddisks
;
733 static int load_ddf_header(int fd
, unsigned long long lba
,
734 unsigned long long size
,
736 struct ddf_header
*hdr
, struct ddf_header
*anchor
)
738 /* read a ddf header (primary or secondary) from fd/lba
739 * and check that it is consistent with anchor
741 * magic, crc, guid, rev, and LBA's header_type, and
742 * everything after header_type must be the same
747 if (lseek64(fd
, lba
<<9, 0) < 0)
750 if (read(fd
, hdr
, 512) != 512)
753 if (hdr
->magic
!= DDF_HEADER_MAGIC
)
755 if (calc_crc(hdr
, 512) != hdr
->crc
)
757 if (memcmp(anchor
->guid
, hdr
->guid
, DDF_GUID_LEN
) != 0 ||
758 memcmp(anchor
->revision
, hdr
->revision
, 8) != 0 ||
759 anchor
->primary_lba
!= hdr
->primary_lba
||
760 anchor
->secondary_lba
!= hdr
->secondary_lba
||
762 memcmp(anchor
->pad2
, hdr
->pad2
, 512 -
763 offsetof(struct ddf_header
, pad2
)) != 0)
766 /* Looks good enough to me... */
770 static void *load_section(int fd
, struct ddf_super
*super
, void *buf
,
771 __u32 offset_be
, __u32 len_be
, int check
)
773 unsigned long long offset
= __be32_to_cpu(offset_be
);
774 unsigned long long len
= __be32_to_cpu(len_be
);
775 int dofree
= (buf
== NULL
);
778 if (len
!= 2 && len
!= 8 && len
!= 32
779 && len
!= 128 && len
!= 512)
785 /* All pre-allocated sections are a single block */
788 } else if (posix_memalign(&buf
, 512, len
<<9) != 0)
794 if (super
->active
->type
== 1)
795 offset
+= __be64_to_cpu(super
->active
->primary_lba
);
797 offset
+= __be64_to_cpu(super
->active
->secondary_lba
);
799 if ((unsigned long long)lseek64(fd
, offset
<<9, 0) != (offset
<<9)) {
804 if ((unsigned long long)read(fd
, buf
, len
<<9) != (len
<<9)) {
812 static int load_ddf_headers(int fd
, struct ddf_super
*super
, char *devname
)
814 unsigned long long dsize
;
816 get_dev_size(fd
, NULL
, &dsize
);
818 if (lseek64(fd
, dsize
-512, 0) < 0) {
820 pr_err("Cannot seek to anchor block on %s: %s\n",
821 devname
, strerror(errno
));
824 if (read(fd
, &super
->anchor
, 512) != 512) {
826 pr_err("Cannot read anchor block on %s: %s\n",
827 devname
, strerror(errno
));
830 if (super
->anchor
.magic
!= DDF_HEADER_MAGIC
) {
832 pr_err("no DDF anchor found on %s\n",
836 if (calc_crc(&super
->anchor
, 512) != super
->anchor
.crc
) {
838 pr_err("bad CRC on anchor on %s\n",
842 if (memcmp(super
->anchor
.revision
, DDF_REVISION_0
, 8) != 0 &&
843 memcmp(super
->anchor
.revision
, DDF_REVISION_2
, 8) != 0) {
845 pr_err("can only support super revision"
846 " %.8s and earlier, not %.8s on %s\n",
847 DDF_REVISION_2
, super
->anchor
.revision
,devname
);
850 super
->active
= NULL
;
851 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.primary_lba
),
853 &super
->primary
, &super
->anchor
) == 0) {
855 pr_err("Failed to load primary DDF header "
858 super
->active
= &super
->primary
;
859 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.secondary_lba
),
861 &super
->secondary
, &super
->anchor
)) {
862 if (super
->active
== NULL
863 || (__be32_to_cpu(super
->primary
.seq
)
864 < __be32_to_cpu(super
->secondary
.seq
) &&
865 !super
->secondary
.openflag
)
866 || (__be32_to_cpu(super
->primary
.seq
)
867 == __be32_to_cpu(super
->secondary
.seq
) &&
868 super
->primary
.openflag
&& !super
->secondary
.openflag
)
870 super
->active
= &super
->secondary
;
872 pr_err("Failed to load secondary DDF header on %s\n",
874 if (super
->active
== NULL
)
879 static int load_ddf_global(int fd
, struct ddf_super
*super
, char *devname
)
882 ok
= load_section(fd
, super
, &super
->controller
,
883 super
->active
->controller_section_offset
,
884 super
->active
->controller_section_length
,
886 super
->phys
= load_section(fd
, super
, NULL
,
887 super
->active
->phys_section_offset
,
888 super
->active
->phys_section_length
,
890 super
->pdsize
= __be32_to_cpu(super
->active
->phys_section_length
) * 512;
892 super
->virt
= load_section(fd
, super
, NULL
,
893 super
->active
->virt_section_offset
,
894 super
->active
->virt_section_length
,
896 super
->vdsize
= __be32_to_cpu(super
->active
->virt_section_length
) * 512;
906 super
->conflist
= NULL
;
909 super
->max_part
= __be16_to_cpu(super
->active
->max_partitions
);
910 super
->mppe
= __be16_to_cpu(super
->active
->max_primary_element_entries
);
911 super
->conf_rec_len
= __be16_to_cpu(super
->active
->config_record_len
);
915 #define DDF_UNUSED_BVD 0xff
916 static int alloc_other_bvds(const struct ddf_super
*ddf
, struct vcl
*vcl
)
918 unsigned int n_vds
= vcl
->conf
.sec_elmnt_count
- 1;
919 unsigned int i
, vdsize
;
922 vcl
->other_bvds
= NULL
;
925 vdsize
= ddf
->conf_rec_len
* 512;
926 if (posix_memalign(&p
, 512, n_vds
*
927 (vdsize
+ sizeof(struct vd_config
*))) != 0)
929 vcl
->other_bvds
= (struct vd_config
**) (p
+ n_vds
* vdsize
);
930 for (i
= 0; i
< n_vds
; i
++) {
931 vcl
->other_bvds
[i
] = p
+ i
* vdsize
;
932 memset(vcl
->other_bvds
[i
], 0, vdsize
);
933 vcl
->other_bvds
[i
]->sec_elmnt_seq
= DDF_UNUSED_BVD
;
938 static void add_other_bvd(struct vcl
*vcl
, struct vd_config
*vd
,
942 for (i
= 0; i
< vcl
->conf
.sec_elmnt_count
-1; i
++)
943 if (vcl
->other_bvds
[i
]->sec_elmnt_seq
== vd
->sec_elmnt_seq
)
946 if (i
< vcl
->conf
.sec_elmnt_count
-1) {
947 if (vd
->seqnum
<= vcl
->other_bvds
[i
]->seqnum
)
950 for (i
= 0; i
< vcl
->conf
.sec_elmnt_count
-1; i
++)
951 if (vcl
->other_bvds
[i
]->sec_elmnt_seq
== DDF_UNUSED_BVD
)
953 if (i
== vcl
->conf
.sec_elmnt_count
-1) {
954 pr_err("no space for sec level config %u, count is %u\n",
955 vd
->sec_elmnt_seq
, vcl
->conf
.sec_elmnt_count
);
959 memcpy(vcl
->other_bvds
[i
], vd
, len
);
962 static int load_ddf_local(int fd
, struct ddf_super
*super
,
963 char *devname
, int keep
)
969 unsigned int confsec
;
971 unsigned int max_virt_disks
= __be16_to_cpu(super
->active
->max_vd_entries
);
972 unsigned long long dsize
;
974 /* First the local disk info */
975 if (posix_memalign((void**)&dl
, 512,
977 (super
->max_part
) * sizeof(dl
->vlist
[0])) != 0) {
978 pr_err("%s could not allocate disk info buffer\n",
983 load_section(fd
, super
, &dl
->disk
,
984 super
->active
->data_section_offset
,
985 super
->active
->data_section_length
,
987 dl
->devname
= devname
? xstrdup(devname
) : NULL
;
990 dl
->major
= major(stb
.st_rdev
);
991 dl
->minor
= minor(stb
.st_rdev
);
992 dl
->next
= super
->dlist
;
993 dl
->fd
= keep
? fd
: -1;
996 if (get_dev_size(fd
, devname
, &dsize
))
997 dl
->size
= dsize
>> 9;
998 /* If the disks have different sizes, the LBAs will differ
999 * between phys disks.
1000 * At this point here, the values in super->active must be valid
1001 * for this phys disk. */
1002 dl
->primary_lba
= super
->active
->primary_lba
;
1003 dl
->secondary_lba
= super
->active
->secondary_lba
;
1004 dl
->workspace_lba
= super
->active
->workspace_lba
;
1006 for (i
= 0 ; i
< super
->max_part
; i
++)
1007 dl
->vlist
[i
] = NULL
;
1010 for (i
= 0; i
< __be16_to_cpu(super
->active
->max_pd_entries
); i
++)
1011 if (memcmp(super
->phys
->entries
[i
].guid
,
1012 dl
->disk
.guid
, DDF_GUID_LEN
) == 0)
1015 /* Now the config list. */
1016 /* 'conf' is an array of config entries, some of which are
1017 * probably invalid. Those which are good need to be copied into
1021 conf
= load_section(fd
, super
, NULL
,
1022 super
->active
->config_section_offset
,
1023 super
->active
->config_section_length
,
1028 confsec
< __be32_to_cpu(super
->active
->config_section_length
);
1029 confsec
+= super
->conf_rec_len
) {
1030 struct vd_config
*vd
=
1031 (struct vd_config
*)((char*)conf
+ confsec
*512);
1034 if (vd
->magic
== DDF_SPARE_ASSIGN_MAGIC
) {
1037 if (posix_memalign((void**)&dl
->spare
, 512,
1038 super
->conf_rec_len
*512) != 0) {
1039 pr_err("%s could not allocate spare info buf\n",
1044 memcpy(dl
->spare
, vd
, super
->conf_rec_len
*512);
1047 if (vd
->magic
!= DDF_VD_CONF_MAGIC
)
1049 for (vcl
= super
->conflist
; vcl
; vcl
= vcl
->next
) {
1050 if (memcmp(vcl
->conf
.guid
,
1051 vd
->guid
, DDF_GUID_LEN
) == 0)
1056 dl
->vlist
[vnum
++] = vcl
;
1057 if (vcl
->other_bvds
!= NULL
&&
1058 vcl
->conf
.sec_elmnt_seq
!= vd
->sec_elmnt_seq
) {
1059 add_other_bvd(vcl
, vd
, super
->conf_rec_len
*512);
1062 if (__be32_to_cpu(vd
->seqnum
) <=
1063 __be32_to_cpu(vcl
->conf
.seqnum
))
1066 if (posix_memalign((void**)&vcl
, 512,
1067 (super
->conf_rec_len
*512 +
1068 offsetof(struct vcl
, conf
))) != 0) {
1069 pr_err("%s could not allocate vcl buf\n",
1073 vcl
->next
= super
->conflist
;
1074 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
1075 vcl
->conf
.sec_elmnt_count
= vd
->sec_elmnt_count
;
1076 if (alloc_other_bvds(super
, vcl
) != 0) {
1077 pr_err("%s could not allocate other bvds\n",
1082 super
->conflist
= vcl
;
1083 dl
->vlist
[vnum
++] = vcl
;
1085 memcpy(&vcl
->conf
, vd
, super
->conf_rec_len
*512);
1086 for (i
=0; i
< max_virt_disks
; i
++)
1087 if (memcmp(super
->virt
->entries
[i
].guid
,
1088 vcl
->conf
.guid
, DDF_GUID_LEN
)==0)
1090 if (i
< max_virt_disks
)
1099 static int load_super_ddf_all(struct supertype
*st
, int fd
,
1100 void **sbp
, char *devname
);
1103 static void free_super_ddf(struct supertype
*st
);
1105 static int load_super_ddf(struct supertype
*st
, int fd
,
1108 unsigned long long dsize
;
1109 struct ddf_super
*super
;
1112 if (get_dev_size(fd
, devname
, &dsize
) == 0)
1115 if (!st
->ignore_hw_compat
&& test_partition(fd
))
1116 /* DDF is not allowed on partitions */
1119 /* 32M is a lower bound */
1120 if (dsize
<= 32*1024*1024) {
1122 pr_err("%s is too small for ddf: "
1123 "size is %llu sectors.\n",
1129 pr_err("%s is an odd size for ddf: "
1130 "size is %llu bytes.\n",
1137 if (posix_memalign((void**)&super
, 512, sizeof(*super
))!= 0) {
1138 pr_err("malloc of %zu failed.\n",
1142 memset(super
, 0, sizeof(*super
));
1144 rv
= load_ddf_headers(fd
, super
, devname
);
1150 /* Have valid headers and have chosen the best. Let's read in the rest*/
1152 rv
= load_ddf_global(fd
, super
, devname
);
1156 pr_err("Failed to load all information "
1157 "sections on %s\n", devname
);
1162 rv
= load_ddf_local(fd
, super
, devname
, 0);
1166 pr_err("Failed to load all information "
1167 "sections on %s\n", devname
);
1172 /* Should possibly check the sections .... */
1175 if (st
->ss
== NULL
) {
1176 st
->ss
= &super_ddf
;
1177 st
->minor_version
= 0;
1184 static void free_super_ddf(struct supertype
*st
)
1186 struct ddf_super
*ddf
= st
->sb
;
1191 while (ddf
->conflist
) {
1192 struct vcl
*v
= ddf
->conflist
;
1193 ddf
->conflist
= v
->next
;
1195 free(v
->block_sizes
);
1198 v->other_bvds[0] points to beginning of buffer,
1199 see alloc_other_bvds()
1201 free(v
->other_bvds
[0]);
1204 while (ddf
->dlist
) {
1205 struct dl
*d
= ddf
->dlist
;
1206 ddf
->dlist
= d
->next
;
1213 while (ddf
->add_list
) {
1214 struct dl
*d
= ddf
->add_list
;
1215 ddf
->add_list
= d
->next
;
1226 static struct supertype
*match_metadata_desc_ddf(char *arg
)
1228 /* 'ddf' only support containers */
1229 struct supertype
*st
;
1230 if (strcmp(arg
, "ddf") != 0 &&
1231 strcmp(arg
, "default") != 0
1235 st
= xcalloc(1, sizeof(*st
));
1236 st
->ss
= &super_ddf
;
1238 st
->minor_version
= 0;
1245 static mapping_t ddf_state
[] = {
1251 { "Partially Optimal", 5},
1257 static mapping_t ddf_init_state
[] = {
1258 { "Not Initialised", 0},
1259 { "QuickInit in Progress", 1},
1260 { "Fully Initialised", 2},
1264 static mapping_t ddf_access
[] = {
1268 { "Blocked (no access)", 3},
1272 static mapping_t ddf_level
[] = {
1273 { "RAID0", DDF_RAID0
},
1274 { "RAID1", DDF_RAID1
},
1275 { "RAID3", DDF_RAID3
},
1276 { "RAID4", DDF_RAID4
},
1277 { "RAID5", DDF_RAID5
},
1278 { "RAID1E",DDF_RAID1E
},
1279 { "JBOD", DDF_JBOD
},
1280 { "CONCAT",DDF_CONCAT
},
1281 { "RAID5E",DDF_RAID5E
},
1282 { "RAID5EE",DDF_RAID5EE
},
1283 { "RAID6", DDF_RAID6
},
1286 static mapping_t ddf_sec_level
[] = {
1287 { "Striped", DDF_2STRIPED
},
1288 { "Mirrored", DDF_2MIRRORED
},
1289 { "Concat", DDF_2CONCAT
},
1290 { "Spanned", DDF_2SPANNED
},
1295 static int all_ff(const char *guid
)
1298 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
1299 if (guid
[i
] != (char)0xff)
1305 static void print_guid(char *guid
, int tstamp
)
1307 /* A GUIDs are part (or all) ASCII and part binary.
1308 * They tend to be space padded.
1309 * We print the GUID in HEX, then in parentheses add
1310 * any initial ASCII sequence, and a possible
1311 * time stamp from bytes 16-19
1313 int l
= DDF_GUID_LEN
;
1316 for (i
=0 ; i
<DDF_GUID_LEN
; i
++) {
1317 if ((i
&3)==0 && i
!= 0) printf(":");
1318 printf("%02X", guid
[i
]&255);
1322 while (l
&& guid
[l
-1] == ' ')
1324 for (i
=0 ; i
<l
; i
++) {
1325 if (guid
[i
] >= 0x20 && guid
[i
] < 0x7f)
1326 fputc(guid
[i
], stdout
);
1331 time_t then
= __be32_to_cpu(*(__u32
*)(guid
+16)) + DECADE
;
1334 tm
= localtime(&then
);
1335 strftime(tbuf
, 100, " %D %T",tm
);
1336 fputs(tbuf
, stdout
);
1341 static const char *guid_str(const char *guid
)
1343 static char buf
[DDF_GUID_LEN
*2+1];
1346 for (i
= 0; i
< DDF_GUID_LEN
; i
++) {
1347 unsigned char c
= guid
[i
];
1348 if (c
>= 32 && c
< 127)
1349 p
+= sprintf(p
, "%c", c
);
1351 p
+= sprintf(p
, "%02x", c
);
1354 return (const char *) buf
;
1357 static void examine_vd(int n
, struct ddf_super
*sb
, char *guid
)
1359 int crl
= sb
->conf_rec_len
;
1362 for (vcl
= sb
->conflist
; vcl
; vcl
= vcl
->next
) {
1364 struct vd_config
*vc
= &vcl
->conf
;
1366 if (calc_crc(vc
, crl
*512) != vc
->crc
)
1368 if (memcmp(vc
->guid
, guid
, DDF_GUID_LEN
) != 0)
1371 /* Ok, we know about this VD, let's give more details */
1372 printf(" Raid Devices[%d] : %d (", n
,
1373 __be16_to_cpu(vc
->prim_elmnt_count
));
1374 for (i
= 0; i
< __be16_to_cpu(vc
->prim_elmnt_count
); i
++) {
1376 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1377 for (j
=0; j
<cnt
; j
++)
1378 if (vc
->phys_refnum
[i
] == sb
->phys
->entries
[j
].refnum
)
1387 if (vc
->chunk_shift
!= 255)
1388 printf(" Chunk Size[%d] : %d sectors\n", n
,
1389 1 << vc
->chunk_shift
);
1390 printf(" Raid Level[%d] : %s\n", n
,
1391 map_num(ddf_level
, vc
->prl
)?:"-unknown-");
1392 if (vc
->sec_elmnt_count
!= 1) {
1393 printf(" Secondary Position[%d] : %d of %d\n", n
,
1394 vc
->sec_elmnt_seq
, vc
->sec_elmnt_count
);
1395 printf(" Secondary Level[%d] : %s\n", n
,
1396 map_num(ddf_sec_level
, vc
->srl
) ?: "-unknown-");
1398 printf(" Device Size[%d] : %llu\n", n
,
1399 (unsigned long long)__be64_to_cpu(vc
->blocks
)/2);
1400 printf(" Array Size[%d] : %llu\n", n
,
1401 (unsigned long long)__be64_to_cpu(vc
->array_blocks
)/2);
1405 static void examine_vds(struct ddf_super
*sb
)
1407 int cnt
= __be16_to_cpu(sb
->virt
->populated_vdes
);
1409 printf(" Virtual Disks : %d\n", cnt
);
1411 for (i
= 0; i
< __be16_to_cpu(sb
->virt
->max_vdes
); i
++) {
1412 struct virtual_entry
*ve
= &sb
->virt
->entries
[i
];
1413 if (all_ff(ve
->guid
))
1416 printf(" VD GUID[%d] : ", i
); print_guid(ve
->guid
, 1);
1418 printf(" unit[%d] : %d\n", i
, __be16_to_cpu(ve
->unit
));
1419 printf(" state[%d] : %s, %s%s\n", i
,
1420 map_num(ddf_state
, ve
->state
& 7),
1421 (ve
->state
& 8) ? "Morphing, ": "",
1422 (ve
->state
& 16)? "Not Consistent" : "Consistent");
1423 printf(" init state[%d] : %s\n", i
,
1424 map_num(ddf_init_state
, ve
->init_state
&3));
1425 printf(" access[%d] : %s\n", i
,
1426 map_num(ddf_access
, (ve
->init_state
>>6) & 3));
1427 printf(" Name[%d] : %.16s\n", i
, ve
->name
);
1428 examine_vd(i
, sb
, ve
->guid
);
1430 if (cnt
) printf("\n");
1433 static void examine_pds(struct ddf_super
*sb
)
1435 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1438 printf(" Physical Disks : %d\n", cnt
);
1439 printf(" Number RefNo Size Device Type/State\n");
1441 for (i
=0 ; i
<cnt
; i
++) {
1442 struct phys_disk_entry
*pd
= &sb
->phys
->entries
[i
];
1443 int type
= __be16_to_cpu(pd
->type
);
1444 int state
= __be16_to_cpu(pd
->state
);
1446 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1448 printf(" %3d %08x ", i
,
1449 __be32_to_cpu(pd
->refnum
));
1451 (unsigned long long)__be64_to_cpu(pd
->config_size
)>>1);
1452 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1453 if (dl
->disk
.refnum
== pd
->refnum
) {
1454 char *dv
= map_dev(dl
->major
, dl
->minor
, 0);
1456 printf("%-15s", dv
);
1463 printf(" %s%s%s%s%s",
1464 (type
&2) ? "active":"",
1465 (type
&4) ? "Global-Spare":"",
1466 (type
&8) ? "spare" : "",
1467 (type
&16)? ", foreign" : "",
1468 (type
&32)? "pass-through" : "");
1469 if (state
& DDF_Failed
)
1470 /* This over-rides these three */
1471 state
&= ~(DDF_Online
|DDF_Rebuilding
|DDF_Transition
);
1472 printf("/%s%s%s%s%s%s%s",
1473 (state
&1)? "Online": "Offline",
1474 (state
&2)? ", Failed": "",
1475 (state
&4)? ", Rebuilding": "",
1476 (state
&8)? ", in-transition": "",
1477 (state
&16)? ", SMART-errors": "",
1478 (state
&32)? ", Unrecovered-Read-Errors": "",
1479 (state
&64)? ", Missing" : "");
1484 static void examine_super_ddf(struct supertype
*st
, char *homehost
)
1486 struct ddf_super
*sb
= st
->sb
;
1488 printf(" Magic : %08x\n", __be32_to_cpu(sb
->anchor
.magic
));
1489 printf(" Version : %.8s\n", sb
->anchor
.revision
);
1490 printf("Controller GUID : "); print_guid(sb
->controller
.guid
, 0);
1492 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1494 printf(" Seq : %08x\n", __be32_to_cpu(sb
->active
->seq
));
1495 printf(" Redundant hdr : %s\n", sb
->secondary
.magic
== DDF_HEADER_MAGIC
1501 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1503 static void uuid_from_ddf_guid(const char *guid
, int uuid
[4]);
1504 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4]);
1506 static unsigned int get_vd_num_of_subarray(struct supertype
*st
)
1509 * Figure out the VD number for this supertype.
1510 * Returns DDF_CONTAINER for the container itself,
1511 * and DDF_NOTFOUND on error.
1513 struct ddf_super
*ddf
= st
->sb
;
1518 if (*st
->container_devnm
== '\0')
1519 return DDF_CONTAINER
;
1521 sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
1522 if (!sra
|| sra
->array
.major_version
!= -1 ||
1523 sra
->array
.minor_version
!= -2 ||
1524 !is_subarray(sra
->text_version
))
1525 return DDF_NOTFOUND
;
1527 sub
= strchr(sra
->text_version
+ 1, '/');
1529 vcnum
= strtoul(sub
+ 1, &end
, 10);
1530 if (sub
== NULL
|| *sub
== '\0' || *end
!= '\0' ||
1531 vcnum
>= __be16_to_cpu(ddf
->active
->max_vd_entries
))
1532 return DDF_NOTFOUND
;
1537 static void brief_examine_super_ddf(struct supertype
*st
, int verbose
)
1539 /* We just write a generic DDF ARRAY entry
1543 getinfo_super_ddf(st
, &info
, NULL
);
1544 fname_from_uuid(st
, &info
, nbuf
, ':');
1546 printf("ARRAY metadata=ddf UUID=%s\n", nbuf
+ 5);
1549 static void brief_examine_subarrays_ddf(struct supertype
*st
, int verbose
)
1551 /* We just write a generic DDF ARRAY entry
1553 struct ddf_super
*ddf
= st
->sb
;
1557 getinfo_super_ddf(st
, &info
, NULL
);
1558 fname_from_uuid(st
, &info
, nbuf
, ':');
1560 for (i
= 0; i
< __be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
1561 struct virtual_entry
*ve
= &ddf
->virt
->entries
[i
];
1564 if (all_ff(ve
->guid
))
1566 memcpy(vcl
.conf
.guid
, ve
->guid
, DDF_GUID_LEN
);
1567 ddf
->currentconf
=&vcl
;
1568 uuid_from_super_ddf(st
, info
.uuid
);
1569 fname_from_uuid(st
, &info
, nbuf1
, ':');
1570 printf("ARRAY container=%s member=%d UUID=%s\n",
1571 nbuf
+5, i
, nbuf1
+5);
1575 static void export_examine_super_ddf(struct supertype
*st
)
1579 getinfo_super_ddf(st
, &info
, NULL
);
1580 fname_from_uuid(st
, &info
, nbuf
, ':');
1581 printf("MD_METADATA=ddf\n");
1582 printf("MD_LEVEL=container\n");
1583 printf("MD_UUID=%s\n", nbuf
+5);
1586 static int copy_metadata_ddf(struct supertype
*st
, int from
, int to
)
1589 unsigned long long dsize
, offset
;
1591 struct ddf_header
*ddf
;
1594 /* The meta consists of an anchor, a primary, and a secondary.
1595 * This all lives at the end of the device.
1596 * So it is easiest to find the earliest of primary and
1597 * secondary, and copy everything from there.
1599 * Anchor is 512 from end It contains primary_lba and secondary_lba
1600 * we choose one of those
1603 if (posix_memalign(&buf
, 4096, 4096) != 0)
1606 if (!get_dev_size(from
, NULL
, &dsize
))
1609 if (lseek64(from
, dsize
-512, 0) < 0)
1611 if (read(from
, buf
, 512) != 512)
1614 if (ddf
->magic
!= DDF_HEADER_MAGIC
||
1615 calc_crc(ddf
, 512) != ddf
->crc
||
1616 (memcmp(ddf
->revision
, DDF_REVISION_0
, 8) != 0 &&
1617 memcmp(ddf
->revision
, DDF_REVISION_2
, 8) != 0))
1620 offset
= dsize
- 512;
1621 if ((__be64_to_cpu(ddf
->primary_lba
) << 9) < offset
)
1622 offset
= __be64_to_cpu(ddf
->primary_lba
) << 9;
1623 if ((__be64_to_cpu(ddf
->secondary_lba
) << 9) < offset
)
1624 offset
= __be64_to_cpu(ddf
->secondary_lba
) << 9;
1626 bytes
= dsize
- offset
;
1628 if (lseek64(from
, offset
, 0) < 0 ||
1629 lseek64(to
, offset
, 0) < 0)
1631 while (written
< bytes
) {
1632 int n
= bytes
- written
;
1635 if (read(from
, buf
, n
) != n
)
1637 if (write(to
, buf
, n
) != n
)
1648 static void detail_super_ddf(struct supertype
*st
, char *homehost
)
1651 * Could print DDF GUID
1652 * Need to find which array
1653 * If whole, briefly list all arrays
1658 static void brief_detail_super_ddf(struct supertype
*st
)
1662 struct ddf_super
*ddf
= st
->sb
;
1663 unsigned int vcnum
= get_vd_num_of_subarray(st
);
1664 if (vcnum
== DDF_CONTAINER
)
1665 uuid_from_super_ddf(st
, info
.uuid
);
1666 else if (vcnum
== DDF_NOTFOUND
)
1669 uuid_from_ddf_guid(ddf
->virt
->entries
[vcnum
].guid
, info
.uuid
);
1670 fname_from_uuid(st
, &info
, nbuf
,':');
1671 printf(" UUID=%s", nbuf
+ 5);
1675 static int match_home_ddf(struct supertype
*st
, char *homehost
)
1677 /* It matches 'this' host if the controller is a
1678 * Linux-MD controller with vendor_data matching
1681 struct ddf_super
*ddf
= st
->sb
;
1686 len
= strlen(homehost
);
1688 return (memcmp(ddf
->controller
.guid
, T10
, 8) == 0 &&
1689 len
< sizeof(ddf
->controller
.vendor_data
) &&
1690 memcmp(ddf
->controller
.vendor_data
, homehost
,len
) == 0 &&
1691 ddf
->controller
.vendor_data
[len
] == 0);
1695 static int find_index_in_bvd(const struct ddf_super
*ddf
,
1696 const struct vd_config
*conf
, unsigned int n
,
1697 unsigned int *n_bvd
)
1700 * Find the index of the n-th valid physical disk in this BVD
1703 for (i
= 0, j
= 0; i
< ddf
->mppe
&&
1704 j
< __be16_to_cpu(conf
->prim_elmnt_count
); i
++) {
1705 if (conf
->phys_refnum
[i
] != 0xffffffff) {
1713 dprintf("%s: couldn't find BVD member %u (total %u)\n",
1714 __func__
, n
, __be16_to_cpu(conf
->prim_elmnt_count
));
1718 static struct vd_config
*find_vdcr(struct ddf_super
*ddf
, unsigned int inst
,
1720 unsigned int *n_bvd
, struct vcl
**vcl
)
1724 for (v
= ddf
->conflist
; v
; v
= v
->next
) {
1725 unsigned int nsec
, ibvd
;
1726 struct vd_config
*conf
;
1727 if (inst
!= v
->vcnum
)
1730 if (conf
->sec_elmnt_count
== 1) {
1731 if (find_index_in_bvd(ddf
, conf
, n
, n_bvd
)) {
1737 if (v
->other_bvds
== NULL
) {
1738 pr_err("%s: BUG: other_bvds is NULL, nsec=%u\n",
1739 __func__
, conf
->sec_elmnt_count
);
1742 nsec
= n
/ __be16_to_cpu(conf
->prim_elmnt_count
);
1743 if (conf
->sec_elmnt_seq
!= nsec
) {
1744 for (ibvd
= 1; ibvd
< conf
->sec_elmnt_count
; ibvd
++) {
1745 if (v
->other_bvds
[ibvd
-1]->sec_elmnt_seq
1749 if (ibvd
== conf
->sec_elmnt_count
)
1751 conf
= v
->other_bvds
[ibvd
-1];
1753 if (!find_index_in_bvd(ddf
, conf
,
1754 n
- nsec
*conf
->sec_elmnt_count
, n_bvd
))
1756 dprintf("%s: found disk %u as member %u in bvd %d of array %u\n"
1757 , __func__
, n
, *n_bvd
, ibvd
-1, inst
);
1762 pr_err("%s: Could't find disk %d in array %u\n", __func__
, n
, inst
);
1767 static int find_phys(const struct ddf_super
*ddf
, __u32 phys_refnum
)
1769 /* Find the entry in phys_disk which has the given refnum
1770 * and return it's index
1773 for (i
= 0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
); i
++)
1774 if (ddf
->phys
->entries
[i
].refnum
== phys_refnum
)
1779 static void uuid_from_ddf_guid(const char *guid
, int uuid
[4])
1782 struct sha1_ctx ctx
;
1783 sha1_init_ctx(&ctx
);
1784 sha1_process_bytes(guid
, DDF_GUID_LEN
, &ctx
);
1785 sha1_finish_ctx(&ctx
, buf
);
1786 memcpy(uuid
, buf
, 4*4);
1789 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4])
1791 /* The uuid returned here is used for:
1792 * uuid to put into bitmap file (Create, Grow)
1793 * uuid for backup header when saving critical section (Grow)
1794 * comparing uuids when re-adding a device into an array
1795 * In these cases the uuid required is that of the data-array,
1796 * not the device-set.
1797 * uuid to recognise same set when adding a missing device back
1798 * to an array. This is a uuid for the device-set.
1800 * For each of these we can make do with a truncated
1801 * or hashed uuid rather than the original, as long as
1803 * In the case of SVD we assume the BVD is of interest,
1804 * though that might be the case if a bitmap were made for
1805 * a mirrored SVD - worry about that later.
1806 * So we need to find the VD configuration record for the
1807 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1808 * The first 16 bytes of the sha1 of these is used.
1810 struct ddf_super
*ddf
= st
->sb
;
1811 struct vcl
*vcl
= ddf
->currentconf
;
1815 guid
= vcl
->conf
.guid
;
1817 guid
= ddf
->anchor
.guid
;
1818 uuid_from_ddf_guid(guid
, uuid
);
1821 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1823 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1825 struct ddf_super
*ddf
= st
->sb
;
1826 int map_disks
= info
->array
.raid_disks
;
1829 if (ddf
->currentconf
) {
1830 getinfo_super_ddf_bvd(st
, info
, map
);
1833 memset(info
, 0, sizeof(*info
));
1835 info
->array
.raid_disks
= __be16_to_cpu(ddf
->phys
->used_pdes
);
1836 info
->array
.level
= LEVEL_CONTAINER
;
1837 info
->array
.layout
= 0;
1838 info
->array
.md_minor
= -1;
1839 cptr
= (__u32
*)(ddf
->anchor
.guid
+ 16);
1840 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
1842 info
->array
.utime
= 0;
1843 info
->array
.chunk_size
= 0;
1844 info
->container_enough
= 1;
1846 info
->disk
.major
= 0;
1847 info
->disk
.minor
= 0;
1849 info
->disk
.number
= __be32_to_cpu(ddf
->dlist
->disk
.refnum
);
1850 info
->disk
.raid_disk
= find_phys(ddf
, ddf
->dlist
->disk
.refnum
);
1852 info
->data_offset
= __be64_to_cpu(ddf
->phys
->
1853 entries
[info
->disk
.raid_disk
].
1855 info
->component_size
= ddf
->dlist
->size
- info
->data_offset
;
1857 info
->disk
.number
= -1;
1858 info
->disk
.raid_disk
= -1;
1859 // info->disk.raid_disk = find refnum in the table and use index;
1861 info
->disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
1863 info
->recovery_start
= MaxSector
;
1864 info
->reshape_active
= 0;
1865 info
->recovery_blocked
= 0;
1868 info
->array
.major_version
= -1;
1869 info
->array
.minor_version
= -2;
1870 strcpy(info
->text_version
, "ddf");
1871 info
->safe_mode_delay
= 0;
1873 uuid_from_super_ddf(st
, info
->uuid
);
1877 for (i
= 0 ; i
< map_disks
; i
++) {
1878 if (i
< info
->array
.raid_disks
&&
1879 (__be16_to_cpu(ddf
->phys
->entries
[i
].state
) & DDF_Online
) &&
1880 !(__be16_to_cpu(ddf
->phys
->entries
[i
].state
) & DDF_Failed
))
1888 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1890 struct ddf_super
*ddf
= st
->sb
;
1891 struct vcl
*vc
= ddf
->currentconf
;
1892 int cd
= ddf
->currentdev
;
1896 int map_disks
= info
->array
.raid_disks
;
1898 struct vd_config
*conf
;
1900 memset(info
, 0, sizeof(*info
));
1901 if (layout_ddf2md(&vc
->conf
, &info
->array
) == -1)
1903 info
->array
.md_minor
= -1;
1904 cptr
= (__u32
*)(vc
->conf
.guid
+ 16);
1905 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
1906 info
->array
.utime
= DECADE
+ __be32_to_cpu(vc
->conf
.timestamp
);
1907 info
->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
1908 info
->custom_array_size
= 0;
1911 n_prim
= __be16_to_cpu(conf
->prim_elmnt_count
);
1912 if (conf
->sec_elmnt_count
> 1 && cd
>= n_prim
) {
1913 int ibvd
= cd
/ n_prim
- 1;
1915 conf
= vc
->other_bvds
[ibvd
];
1918 if (cd
>= 0 && (unsigned)cd
< ddf
->mppe
) {
1920 __be64_to_cpu(LBA_OFFSET(ddf
, conf
)[cd
]);
1921 if (vc
->block_sizes
)
1922 info
->component_size
= vc
->block_sizes
[cd
];
1924 info
->component_size
= __be64_to_cpu(conf
->blocks
);
1927 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
1928 if (dl
->disk
.refnum
== conf
->phys_refnum
[cd
])
1931 info
->disk
.major
= 0;
1932 info
->disk
.minor
= 0;
1933 info
->disk
.state
= 0;
1935 info
->disk
.major
= dl
->major
;
1936 info
->disk
.minor
= dl
->minor
;
1937 info
->disk
.raid_disk
= cd
+ conf
->sec_elmnt_seq
1938 * __be16_to_cpu(conf
->prim_elmnt_count
);
1939 info
->disk
.number
= dl
->pdnum
;
1940 info
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
1943 info
->container_member
= ddf
->currentconf
->vcnum
;
1945 info
->recovery_start
= MaxSector
;
1946 info
->resync_start
= 0;
1947 info
->reshape_active
= 0;
1948 info
->recovery_blocked
= 0;
1949 if (!(ddf
->virt
->entries
[info
->container_member
].state
1950 & DDF_state_inconsistent
) &&
1951 (ddf
->virt
->entries
[info
->container_member
].init_state
1952 & DDF_initstate_mask
)
1954 info
->resync_start
= MaxSector
;
1956 uuid_from_super_ddf(st
, info
->uuid
);
1958 info
->array
.major_version
= -1;
1959 info
->array
.minor_version
= -2;
1960 sprintf(info
->text_version
, "/%s/%d",
1961 st
->container_devnm
,
1962 info
->container_member
);
1963 info
->safe_mode_delay
= 200;
1965 memcpy(info
->name
, ddf
->virt
->entries
[info
->container_member
].name
, 16);
1968 if (info
->name
[j
] == ' ')
1972 for (j
= 0; j
< map_disks
; j
++) {
1974 if (j
< info
->array
.raid_disks
) {
1975 int i
= find_phys(ddf
, vc
->conf
.phys_refnum
[j
]);
1977 (__be16_to_cpu(ddf
->phys
->entries
[i
].state
) & DDF_Online
) &&
1978 !(__be16_to_cpu(ddf
->phys
->entries
[i
].state
) & DDF_Failed
))
1984 static int update_super_ddf(struct supertype
*st
, struct mdinfo
*info
,
1986 char *devname
, int verbose
,
1987 int uuid_set
, char *homehost
)
1989 /* For 'assemble' and 'force' we need to return non-zero if any
1990 * change was made. For others, the return value is ignored.
1991 * Update options are:
1992 * force-one : This device looks a bit old but needs to be included,
1993 * update age info appropriately.
1994 * assemble: clear any 'faulty' flag to allow this device to
1996 * force-array: Array is degraded but being forced, mark it clean
1997 * if that will be needed to assemble it.
1999 * newdev: not used ????
2000 * grow: Array has gained a new device - this is currently for
2002 * resync: mark as dirty so a resync will happen.
2003 * uuid: Change the uuid of the array to match what is given
2004 * homehost: update the recorded homehost
2005 * name: update the name - preserving the homehost
2006 * _reshape_progress: record new reshape_progress position.
2008 * Following are not relevant for this version:
2009 * sparc2.2 : update from old dodgey metadata
2010 * super-minor: change the preferred_minor number
2011 * summaries: update redundant counters.
2014 // struct ddf_super *ddf = st->sb;
2015 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
2016 // struct virtual_entry *ve = find_ve(ddf);
2018 /* we don't need to handle "force-*" or "assemble" as
2019 * there is no need to 'trick' the kernel. We the metadata is
2020 * first updated to activate the array, all the implied modifications
2024 if (strcmp(update
, "grow") == 0) {
2026 } else if (strcmp(update
, "resync") == 0) {
2027 // info->resync_checkpoint = 0;
2028 } else if (strcmp(update
, "homehost") == 0) {
2029 /* homehost is stored in controller->vendor_data,
2030 * or it is when we are the vendor
2032 // if (info->vendor_is_local)
2033 // strcpy(ddf->controller.vendor_data, homehost);
2035 } else if (strcmp(update
, "name") == 0) {
2036 /* name is stored in virtual_entry->name */
2037 // memset(ve->name, ' ', 16);
2038 // strncpy(ve->name, info->name, 16);
2040 } else if (strcmp(update
, "_reshape_progress") == 0) {
2041 /* We don't support reshape yet */
2042 } else if (strcmp(update
, "assemble") == 0 ) {
2043 /* Do nothing, just succeed */
2048 // update_all_csum(ddf);
2053 static void make_header_guid(char *guid
)
2056 /* Create a DDF Header of Virtual Disk GUID */
2058 /* 24 bytes of fiction required.
2059 * first 8 are a 'vendor-id' - "Linux-MD"
2060 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
2061 * Remaining 8 random number plus timestamp
2063 memcpy(guid
, T10
, sizeof(T10
));
2064 stamp
= __cpu_to_be32(0xdeadbeef);
2065 memcpy(guid
+8, &stamp
, 4);
2066 stamp
= __cpu_to_be32(0);
2067 memcpy(guid
+12, &stamp
, 4);
2068 stamp
= __cpu_to_be32(time(0) - DECADE
);
2069 memcpy(guid
+16, &stamp
, 4);
2071 memcpy(guid
+20, &stamp
, 4);
2074 static unsigned int find_unused_vde(const struct ddf_super
*ddf
)
2077 for (i
= 0; i
< __be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
2078 if (all_ff(ddf
->virt
->entries
[i
].guid
))
2081 return DDF_NOTFOUND
;
2084 static unsigned int find_vde_by_name(const struct ddf_super
*ddf
,
2089 return DDF_NOTFOUND
;
2090 for (i
= 0; i
< __be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
2091 if (all_ff(ddf
->virt
->entries
[i
].guid
))
2093 if (!strncmp(name
, ddf
->virt
->entries
[i
].name
,
2094 sizeof(ddf
->virt
->entries
[i
].name
)))
2097 return DDF_NOTFOUND
;
2100 static unsigned int find_vde_by_guid(const struct ddf_super
*ddf
,
2104 if (guid
== NULL
|| all_ff(guid
))
2105 return DDF_NOTFOUND
;
2106 for (i
= 0; i
< __be16_to_cpu(ddf
->virt
->max_vdes
); i
++)
2107 if (!memcmp(ddf
->virt
->entries
[i
].guid
, guid
, DDF_GUID_LEN
))
2109 return DDF_NOTFOUND
;
2112 static int init_super_ddf_bvd(struct supertype
*st
,
2113 mdu_array_info_t
*info
,
2114 unsigned long long size
,
2115 char *name
, char *homehost
,
2116 int *uuid
, unsigned long long data_offset
);
2118 static int init_super_ddf(struct supertype
*st
,
2119 mdu_array_info_t
*info
,
2120 unsigned long long size
, char *name
, char *homehost
,
2121 int *uuid
, unsigned long long data_offset
)
2123 /* This is primarily called by Create when creating a new array.
2124 * We will then get add_to_super called for each component, and then
2125 * write_init_super called to write it out to each device.
2126 * For DDF, Create can create on fresh devices or on a pre-existing
2128 * To create on a pre-existing array a different method will be called.
2129 * This one is just for fresh drives.
2131 * We need to create the entire 'ddf' structure which includes:
2132 * DDF headers - these are easy.
2133 * Controller data - a Sector describing this controller .. not that
2134 * this is a controller exactly.
2135 * Physical Disk Record - one entry per device, so
2136 * leave plenty of space.
2137 * Virtual Disk Records - again, just leave plenty of space.
2138 * This just lists VDs, doesn't give details
2139 * Config records - describes the VDs that use this disk
2140 * DiskData - describes 'this' device.
2141 * BadBlockManagement - empty
2142 * Diag Space - empty
2143 * Vendor Logs - Could we put bitmaps here?
2146 struct ddf_super
*ddf
;
2149 int max_phys_disks
, max_virt_disks
;
2150 unsigned long long sector
;
2154 struct phys_disk
*pd
;
2155 struct virtual_disk
*vd
;
2157 if (data_offset
!= INVALID_SECTORS
) {
2158 pr_err("data-offset not supported by DDF\n");
2163 return init_super_ddf_bvd(st
, info
, size
, name
, homehost
, uuid
,
2166 if (posix_memalign((void**)&ddf
, 512, sizeof(*ddf
)) != 0) {
2167 pr_err("%s could not allocate superblock\n", __func__
);
2170 memset(ddf
, 0, sizeof(*ddf
));
2171 ddf
->dlist
= NULL
; /* no physical disks yet */
2172 ddf
->conflist
= NULL
; /* No virtual disks yet */
2176 /* zeroing superblock */
2180 /* At least 32MB *must* be reserved for the ddf. So let's just
2181 * start 32MB from the end, and put the primary header there.
2182 * Don't do secondary for now.
2183 * We don't know exactly where that will be yet as it could be
2184 * different on each device. To just set up the lengths.
2188 ddf
->anchor
.magic
= DDF_HEADER_MAGIC
;
2189 make_header_guid(ddf
->anchor
.guid
);
2191 memcpy(ddf
->anchor
.revision
, DDF_REVISION_2
, 8);
2192 ddf
->anchor
.seq
= __cpu_to_be32(1);
2193 ddf
->anchor
.timestamp
= __cpu_to_be32(time(0) - DECADE
);
2194 ddf
->anchor
.openflag
= 0xFF;
2195 ddf
->anchor
.foreignflag
= 0;
2196 ddf
->anchor
.enforcegroups
= 0; /* Is this best?? */
2197 ddf
->anchor
.pad0
= 0xff;
2198 memset(ddf
->anchor
.pad1
, 0xff, 12);
2199 memset(ddf
->anchor
.header_ext
, 0xff, 32);
2200 ddf
->anchor
.primary_lba
= ~(__u64
)0;
2201 ddf
->anchor
.secondary_lba
= ~(__u64
)0;
2202 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
2203 memset(ddf
->anchor
.pad2
, 0xff, 3);
2204 ddf
->anchor
.workspace_len
= __cpu_to_be32(32768); /* Must be reserved */
2205 ddf
->anchor
.workspace_lba
= ~(__u64
)0; /* Put this at bottom
2206 of 32M reserved.. */
2207 max_phys_disks
= 1023; /* Should be enough */
2208 ddf
->anchor
.max_pd_entries
= __cpu_to_be16(max_phys_disks
);
2209 max_virt_disks
= 255;
2210 ddf
->anchor
.max_vd_entries
= __cpu_to_be16(max_virt_disks
); /* ?? */
2211 ddf
->anchor
.max_partitions
= __cpu_to_be16(64); /* ?? */
2214 ddf
->conf_rec_len
= 1 + ROUND_UP(ddf
->mppe
* (4+8), 512)/512;
2215 ddf
->anchor
.config_record_len
= __cpu_to_be16(ddf
->conf_rec_len
);
2216 ddf
->anchor
.max_primary_element_entries
= __cpu_to_be16(ddf
->mppe
);
2217 memset(ddf
->anchor
.pad3
, 0xff, 54);
2218 /* controller sections is one sector long immediately
2219 * after the ddf header */
2221 ddf
->anchor
.controller_section_offset
= __cpu_to_be32(sector
);
2222 ddf
->anchor
.controller_section_length
= __cpu_to_be32(1);
2225 /* phys is 8 sectors after that */
2226 pdsize
= ROUND_UP(sizeof(struct phys_disk
) +
2227 sizeof(struct phys_disk_entry
)*max_phys_disks
,
2229 switch(pdsize
/512) {
2230 case 2: case 8: case 32: case 128: case 512: break;
2233 ddf
->anchor
.phys_section_offset
= __cpu_to_be32(sector
);
2234 ddf
->anchor
.phys_section_length
=
2235 __cpu_to_be32(pdsize
/512); /* max_primary_element_entries/8 */
2236 sector
+= pdsize
/512;
2238 /* virt is another 32 sectors */
2239 vdsize
= ROUND_UP(sizeof(struct virtual_disk
) +
2240 sizeof(struct virtual_entry
) * max_virt_disks
,
2242 switch(vdsize
/512) {
2243 case 2: case 8: case 32: case 128: case 512: break;
2246 ddf
->anchor
.virt_section_offset
= __cpu_to_be32(sector
);
2247 ddf
->anchor
.virt_section_length
=
2248 __cpu_to_be32(vdsize
/512); /* max_vd_entries/8 */
2249 sector
+= vdsize
/512;
2251 clen
= ddf
->conf_rec_len
* (ddf
->max_part
+1);
2252 ddf
->anchor
.config_section_offset
= __cpu_to_be32(sector
);
2253 ddf
->anchor
.config_section_length
= __cpu_to_be32(clen
);
2256 ddf
->anchor
.data_section_offset
= __cpu_to_be32(sector
);
2257 ddf
->anchor
.data_section_length
= __cpu_to_be32(1);
2260 ddf
->anchor
.bbm_section_length
= __cpu_to_be32(0);
2261 ddf
->anchor
.bbm_section_offset
= __cpu_to_be32(0xFFFFFFFF);
2262 ddf
->anchor
.diag_space_length
= __cpu_to_be32(0);
2263 ddf
->anchor
.diag_space_offset
= __cpu_to_be32(0xFFFFFFFF);
2264 ddf
->anchor
.vendor_length
= __cpu_to_be32(0);
2265 ddf
->anchor
.vendor_offset
= __cpu_to_be32(0xFFFFFFFF);
2267 memset(ddf
->anchor
.pad4
, 0xff, 256);
2269 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2270 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2272 ddf
->primary
.openflag
= 1; /* I guess.. */
2273 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
2275 ddf
->secondary
.openflag
= 1; /* I guess.. */
2276 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
2278 ddf
->active
= &ddf
->primary
;
2280 ddf
->controller
.magic
= DDF_CONTROLLER_MAGIC
;
2282 /* 24 more bytes of fiction required.
2283 * first 8 are a 'vendor-id' - "Linux-MD"
2284 * Remaining 16 are serial number.... maybe a hostname would do?
2286 memcpy(ddf
->controller
.guid
, T10
, sizeof(T10
));
2287 gethostname(hostname
, sizeof(hostname
));
2288 hostname
[sizeof(hostname
) - 1] = 0;
2289 hostlen
= strlen(hostname
);
2290 memcpy(ddf
->controller
.guid
+ 24 - hostlen
, hostname
, hostlen
);
2291 for (i
= strlen(T10
) ; i
+hostlen
< 24; i
++)
2292 ddf
->controller
.guid
[i
] = ' ';
2294 ddf
->controller
.type
.vendor_id
= __cpu_to_be16(0xDEAD);
2295 ddf
->controller
.type
.device_id
= __cpu_to_be16(0xBEEF);
2296 ddf
->controller
.type
.sub_vendor_id
= 0;
2297 ddf
->controller
.type
.sub_device_id
= 0;
2298 memcpy(ddf
->controller
.product_id
, "What Is My PID??", 16);
2299 memset(ddf
->controller
.pad
, 0xff, 8);
2300 memset(ddf
->controller
.vendor_data
, 0xff, 448);
2301 if (homehost
&& strlen(homehost
) < 440)
2302 strcpy((char*)ddf
->controller
.vendor_data
, homehost
);
2304 if (posix_memalign((void**)&pd
, 512, pdsize
) != 0) {
2305 pr_err("%s could not allocate pd\n", __func__
);
2309 ddf
->pdsize
= pdsize
;
2311 memset(pd
, 0xff, pdsize
);
2312 memset(pd
, 0, sizeof(*pd
));
2313 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2314 pd
->used_pdes
= __cpu_to_be16(0);
2315 pd
->max_pdes
= __cpu_to_be16(max_phys_disks
);
2316 memset(pd
->pad
, 0xff, 52);
2317 for (i
= 0; i
< max_phys_disks
; i
++)
2318 memset(pd
->entries
[i
].guid
, 0xff, DDF_GUID_LEN
);
2320 if (posix_memalign((void**)&vd
, 512, vdsize
) != 0) {
2321 pr_err("%s could not allocate vd\n", __func__
);
2325 ddf
->vdsize
= vdsize
;
2326 memset(vd
, 0, vdsize
);
2327 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
2328 vd
->populated_vdes
= __cpu_to_be16(0);
2329 vd
->max_vdes
= __cpu_to_be16(max_virt_disks
);
2330 memset(vd
->pad
, 0xff, 52);
2332 for (i
=0; i
<max_virt_disks
; i
++)
2333 memset(&vd
->entries
[i
], 0xff, sizeof(struct virtual_entry
));
2336 ddf_set_updates_pending(ddf
);
2340 static int chunk_to_shift(int chunksize
)
2342 return ffs(chunksize
/512)-1;
2347 unsigned long long start
, size
;
2349 static int cmp_extent(const void *av
, const void *bv
)
2351 const struct extent
*a
= av
;
2352 const struct extent
*b
= bv
;
2353 if (a
->start
< b
->start
)
2355 if (a
->start
> b
->start
)
2360 static struct extent
*get_extents(struct ddf_super
*ddf
, struct dl
*dl
)
2362 /* find a list of used extents on the give physical device
2363 * (dnum) of the given ddf.
2364 * Return a malloced array of 'struct extent'
2366 * FIXME ignore DDF_Legacy devices?
2373 rv
= xmalloc(sizeof(struct extent
) * (ddf
->max_part
+ 2));
2375 for (i
= 0; i
< ddf
->max_part
; i
++) {
2376 const struct vd_config
*bvd
;
2378 struct vcl
*v
= dl
->vlist
[i
];
2380 get_pd_index_from_refnum(v
, dl
->disk
.refnum
, ddf
->mppe
,
2381 &bvd
, &ibvd
) == DDF_NOTFOUND
)
2383 rv
[n
].start
= __be64_to_cpu(LBA_OFFSET(ddf
, bvd
)[ibvd
]);
2384 rv
[n
].size
= __be64_to_cpu(bvd
->blocks
);
2387 qsort(rv
, n
, sizeof(*rv
), cmp_extent
);
2389 rv
[n
].start
= __be64_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].config_size
);
2395 static int init_super_ddf_bvd(struct supertype
*st
,
2396 mdu_array_info_t
*info
,
2397 unsigned long long size
,
2398 char *name
, char *homehost
,
2399 int *uuid
, unsigned long long data_offset
)
2401 /* We are creating a BVD inside a pre-existing container.
2402 * so st->sb is already set.
2403 * We need to create a new vd_config and a new virtual_entry
2405 struct ddf_super
*ddf
= st
->sb
;
2406 unsigned int venum
, i
;
2407 struct virtual_entry
*ve
;
2409 struct vd_config
*vc
;
2411 if (find_vde_by_name(ddf
, name
) != DDF_NOTFOUND
) {
2412 pr_err("This ddf already has an array called %s\n", name
);
2415 venum
= find_unused_vde(ddf
);
2416 if (venum
== DDF_NOTFOUND
) {
2417 pr_err("Cannot find spare slot for virtual disk\n");
2420 ve
= &ddf
->virt
->entries
[venum
];
2422 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
2423 * timestamp, random number
2425 make_header_guid(ve
->guid
);
2426 ve
->unit
= __cpu_to_be16(info
->md_minor
);
2428 ve
->guid_crc
= crc32(0, (unsigned char*)ddf
->anchor
.guid
, DDF_GUID_LEN
);
2430 ve
->state
= DDF_state_degraded
; /* Will be modified as devices are added */
2431 if (info
->state
& 1) /* clean */
2432 ve
->init_state
= DDF_init_full
;
2434 ve
->init_state
= DDF_init_not
;
2436 memset(ve
->pad1
, 0xff, 14);
2437 memset(ve
->name
, ' ', 16);
2439 strncpy(ve
->name
, name
, 16);
2440 ddf
->virt
->populated_vdes
=
2441 __cpu_to_be16(__be16_to_cpu(ddf
->virt
->populated_vdes
)+1);
2443 /* Now create a new vd_config */
2444 if (posix_memalign((void**)&vcl
, 512,
2445 (offsetof(struct vcl
, conf
) + ddf
->conf_rec_len
* 512)) != 0) {
2446 pr_err("%s could not allocate vd_config\n", __func__
);
2450 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
2453 vc
->magic
= DDF_VD_CONF_MAGIC
;
2454 memcpy(vc
->guid
, ve
->guid
, DDF_GUID_LEN
);
2455 vc
->timestamp
= __cpu_to_be32(time(0)-DECADE
);
2456 vc
->seqnum
= __cpu_to_be32(1);
2457 memset(vc
->pad0
, 0xff, 24);
2458 vc
->chunk_shift
= chunk_to_shift(info
->chunk_size
);
2459 if (layout_md2ddf(info
, vc
) == -1 ||
2460 __be16_to_cpu(vc
->prim_elmnt_count
) > ddf
->mppe
) {
2461 pr_err("%s: unsupported RAID level/layout %d/%d with %d disks\n",
2462 __func__
, info
->level
, info
->layout
, info
->raid_disks
);
2466 vc
->sec_elmnt_seq
= 0;
2467 if (alloc_other_bvds(ddf
, vcl
) != 0) {
2468 pr_err("%s could not allocate other bvds\n",
2473 vc
->blocks
= __cpu_to_be64(info
->size
* 2);
2474 vc
->array_blocks
= __cpu_to_be64(
2475 calc_array_size(info
->level
, info
->raid_disks
, info
->layout
,
2476 info
->chunk_size
, info
->size
*2));
2477 memset(vc
->pad1
, 0xff, 8);
2478 vc
->spare_refs
[0] = 0xffffffff;
2479 vc
->spare_refs
[1] = 0xffffffff;
2480 vc
->spare_refs
[2] = 0xffffffff;
2481 vc
->spare_refs
[3] = 0xffffffff;
2482 vc
->spare_refs
[4] = 0xffffffff;
2483 vc
->spare_refs
[5] = 0xffffffff;
2484 vc
->spare_refs
[6] = 0xffffffff;
2485 vc
->spare_refs
[7] = 0xffffffff;
2486 memset(vc
->cache_pol
, 0, 8);
2488 memset(vc
->pad2
, 0xff, 3);
2489 memset(vc
->pad3
, 0xff, 52);
2490 memset(vc
->pad4
, 0xff, 192);
2491 memset(vc
->v0
, 0xff, 32);
2492 memset(vc
->v1
, 0xff, 32);
2493 memset(vc
->v2
, 0xff, 16);
2494 memset(vc
->v3
, 0xff, 16);
2495 memset(vc
->vendor
, 0xff, 32);
2497 memset(vc
->phys_refnum
, 0xff, 4*ddf
->mppe
);
2498 memset(vc
->phys_refnum
+ddf
->mppe
, 0x00, 8*ddf
->mppe
);
2500 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++) {
2501 memcpy(vcl
->other_bvds
[i
-1], vc
, ddf
->conf_rec_len
* 512);
2502 vcl
->other_bvds
[i
-1]->sec_elmnt_seq
= i
;
2505 vcl
->next
= ddf
->conflist
;
2506 ddf
->conflist
= vcl
;
2507 ddf
->currentconf
= vcl
;
2508 ddf_set_updates_pending(ddf
);
2512 static int get_svd_state(const struct ddf_super
*, const struct vcl
*);
2515 static void add_to_super_ddf_bvd(struct supertype
*st
,
2516 mdu_disk_info_t
*dk
, int fd
, char *devname
)
2518 /* fd and devname identify a device with-in the ddf container (st).
2519 * dk identifies a location in the new BVD.
2520 * We need to find suitable free space in that device and update
2521 * the phys_refnum and lba_offset for the newly created vd_config.
2522 * We might also want to update the type in the phys_disk
2525 * Alternately: fd == -1 and we have already chosen which device to
2526 * use and recorded in dlist->raid_disk;
2529 struct ddf_super
*ddf
= st
->sb
;
2530 struct vd_config
*vc
;
2532 unsigned long long blocks
, pos
, esize
;
2534 unsigned int raid_disk
= dk
->raid_disk
;
2537 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2538 if (dl
->raiddisk
== dk
->raid_disk
)
2541 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2542 if (dl
->major
== dk
->major
&&
2543 dl
->minor
== dk
->minor
)
2546 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
2549 vc
= &ddf
->currentconf
->conf
;
2550 if (vc
->sec_elmnt_count
> 1) {
2551 unsigned int n
= __be16_to_cpu(vc
->prim_elmnt_count
);
2553 vc
= ddf
->currentconf
->other_bvds
[raid_disk
/ n
- 1];
2557 ex
= get_extents(ddf
, dl
);
2562 blocks
= __be64_to_cpu(vc
->blocks
);
2563 if (ddf
->currentconf
->block_sizes
)
2564 blocks
= ddf
->currentconf
->block_sizes
[dk
->raid_disk
];
2567 esize
= ex
[i
].start
- pos
;
2568 if (esize
>= blocks
)
2570 pos
= ex
[i
].start
+ ex
[i
].size
;
2572 } while (ex
[i
-1].size
);
2578 ddf
->currentdev
= dk
->raid_disk
;
2579 vc
->phys_refnum
[raid_disk
] = dl
->disk
.refnum
;
2580 LBA_OFFSET(ddf
, vc
)[raid_disk
] = __cpu_to_be64(pos
);
2582 for (i
= 0; i
< ddf
->max_part
; i
++)
2583 if (dl
->vlist
[i
] == NULL
)
2585 if (i
== ddf
->max_part
)
2587 dl
->vlist
[i
] = ddf
->currentconf
;
2592 dl
->devname
= devname
;
2594 /* Check if we can mark array as optimal yet */
2595 i
= ddf
->currentconf
->vcnum
;
2596 ddf
->virt
->entries
[i
].state
=
2597 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
2598 | get_svd_state(ddf
, ddf
->currentconf
);
2599 ddf
->phys
->entries
[dl
->pdnum
].type
&= ~__cpu_to_be16(DDF_Global_Spare
);
2600 ddf
->phys
->entries
[dl
->pdnum
].type
|= __cpu_to_be16(DDF_Active_in_VD
);
2601 dprintf("%s: added disk %d/%08x to VD %d/%s as disk %d\n",
2602 __func__
, dl
->pdnum
, __be32_to_cpu(dl
->disk
.refnum
),
2603 ddf
->currentconf
->vcnum
, guid_str(vc
->guid
),
2605 ddf_set_updates_pending(ddf
);
2608 static unsigned int find_unused_pde(const struct ddf_super
*ddf
)
2611 for (i
= 0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
); i
++) {
2612 if (all_ff(ddf
->phys
->entries
[i
].guid
))
2615 return DDF_NOTFOUND
;
2618 /* add a device to a container, either while creating it or while
2619 * expanding a pre-existing container
2621 static int add_to_super_ddf(struct supertype
*st
,
2622 mdu_disk_info_t
*dk
, int fd
, char *devname
,
2623 unsigned long long data_offset
)
2625 struct ddf_super
*ddf
= st
->sb
;
2629 unsigned long long size
;
2630 struct phys_disk_entry
*pde
;
2635 if (ddf
->currentconf
) {
2636 add_to_super_ddf_bvd(st
, dk
, fd
, devname
);
2640 /* This is device numbered dk->number. We need to create
2641 * a phys_disk entry and a more detailed disk_data entry.
2644 n
= find_unused_pde(ddf
);
2645 if (n
== DDF_NOTFOUND
) {
2646 pr_err("%s: No free slot in array, cannot add disk\n",
2650 pde
= &ddf
->phys
->entries
[n
];
2651 get_dev_size(fd
, NULL
, &size
);
2652 if (size
<= 32*1024*1024) {
2653 pr_err("%s: device size must be at least 32MB\n",
2659 if (posix_memalign((void**)&dd
, 512,
2660 sizeof(*dd
) + sizeof(dd
->vlist
[0]) * ddf
->max_part
) != 0) {
2661 pr_err("%s could allocate buffer for new disk, aborting\n",
2665 dd
->major
= major(stb
.st_rdev
);
2666 dd
->minor
= minor(stb
.st_rdev
);
2667 dd
->devname
= devname
;
2671 dd
->disk
.magic
= DDF_PHYS_DATA_MAGIC
;
2673 tm
= localtime(&now
);
2674 sprintf(dd
->disk
.guid
, "%8s%04d%02d%02d",
2675 T10
, tm
->tm_year
+1900, tm
->tm_mon
+1, tm
->tm_mday
);
2676 tptr
= (__u32
*)(dd
->disk
.guid
+ 16);
2677 *tptr
++ = random32();
2681 /* Cannot be bothered finding a CRC of some irrelevant details*/
2682 dd
->disk
.refnum
= random32();
2683 for (i
= __be16_to_cpu(ddf
->active
->max_pd_entries
);
2685 if (ddf
->phys
->entries
[i
-1].refnum
== dd
->disk
.refnum
)
2689 dd
->disk
.forced_ref
= 1;
2690 dd
->disk
.forced_guid
= 1;
2691 memset(dd
->disk
.vendor
, ' ', 32);
2692 memcpy(dd
->disk
.vendor
, "Linux", 5);
2693 memset(dd
->disk
.pad
, 0xff, 442);
2694 for (i
= 0; i
< ddf
->max_part
; i
++)
2695 dd
->vlist
[i
] = NULL
;
2699 if (st
->update_tail
) {
2700 int len
= (sizeof(struct phys_disk
) +
2701 sizeof(struct phys_disk_entry
));
2702 struct phys_disk
*pd
;
2705 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2706 pd
->used_pdes
= __cpu_to_be16(n
);
2707 pde
= &pd
->entries
[0];
2710 ddf
->phys
->used_pdes
= __cpu_to_be16(
2711 1 + __be16_to_cpu(ddf
->phys
->used_pdes
));
2713 memcpy(pde
->guid
, dd
->disk
.guid
, DDF_GUID_LEN
);
2714 pde
->refnum
= dd
->disk
.refnum
;
2715 pde
->type
= __cpu_to_be16(DDF_Forced_PD_GUID
| DDF_Global_Spare
);
2716 pde
->state
= __cpu_to_be16(DDF_Online
);
2719 * If there is already a device in dlist, try to reserve the same
2720 * amount of workspace. Otherwise, use 32MB.
2721 * We checked disk size above already.
2723 #define __calc_lba(new, old, lba, mb) do { \
2724 unsigned long long dif; \
2725 if ((old) != NULL) \
2726 dif = (old)->size - __be64_to_cpu((old)->lba); \
2728 dif = (new)->size; \
2729 if ((new)->size > dif) \
2730 (new)->lba = __cpu_to_be64((new)->size - dif); \
2732 (new)->lba = __cpu_to_be64((new)->size - (mb*1024*2)); \
2734 __calc_lba(dd
, ddf
->dlist
, workspace_lba
, 32);
2735 __calc_lba(dd
, ddf
->dlist
, primary_lba
, 16);
2736 __calc_lba(dd
, ddf
->dlist
, secondary_lba
, 32);
2737 pde
->config_size
= dd
->workspace_lba
;
2739 sprintf(pde
->path
, "%17.17s","Information: nil") ;
2740 memset(pde
->pad
, 0xff, 6);
2742 if (st
->update_tail
) {
2743 dd
->next
= ddf
->add_list
;
2746 dd
->next
= ddf
->dlist
;
2748 ddf_set_updates_pending(ddf
);
2754 static int remove_from_super_ddf(struct supertype
*st
, mdu_disk_info_t
*dk
)
2756 struct ddf_super
*ddf
= st
->sb
;
2759 /* mdmon has noticed that this disk (dk->major/dk->minor) has
2760 * disappeared from the container.
2761 * We need to arrange that it disappears from the metadata and
2762 * internal data structures too.
2763 * Most of the work is done by ddf_process_update which edits
2764 * the metadata and closes the file handle and attaches the memory
2765 * where free_updates will free it.
2767 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2768 if (dl
->major
== dk
->major
&&
2769 dl
->minor
== dk
->minor
)
2774 if (st
->update_tail
) {
2775 int len
= (sizeof(struct phys_disk
) +
2776 sizeof(struct phys_disk_entry
));
2777 struct phys_disk
*pd
;
2780 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2781 pd
->used_pdes
= __cpu_to_be16(dl
->pdnum
);
2782 pd
->entries
[0].state
= __cpu_to_be16(DDF_Missing
);
2783 append_metadata_update(st
, pd
, len
);
2789 * This is the write_init_super method for a ddf container. It is
2790 * called when creating a container or adding another device to a
2793 #define NULL_CONF_SZ 4096
2795 static char *null_aligned
;
2796 static int __write_ddf_structure(struct dl
*d
, struct ddf_super
*ddf
, __u8 type
)
2798 unsigned long long sector
;
2799 struct ddf_header
*header
;
2800 int fd
, i
, n_config
, conf_size
;
2803 if (null_aligned
== NULL
) {
2804 if (posix_memalign((void **)&null_aligned
, 4096, NULL_CONF_SZ
)
2807 memset(null_aligned
, 0xff, NULL_CONF_SZ
);
2813 case DDF_HEADER_PRIMARY
:
2814 header
= &ddf
->primary
;
2815 sector
= __be64_to_cpu(header
->primary_lba
);
2817 case DDF_HEADER_SECONDARY
:
2818 header
= &ddf
->secondary
;
2819 sector
= __be64_to_cpu(header
->secondary_lba
);
2825 header
->type
= type
;
2826 header
->openflag
= 1;
2827 header
->crc
= calc_crc(header
, 512);
2829 lseek64(fd
, sector
<<9, 0);
2830 if (write(fd
, header
, 512) < 0)
2833 ddf
->controller
.crc
= calc_crc(&ddf
->controller
, 512);
2834 if (write(fd
, &ddf
->controller
, 512) < 0)
2837 ddf
->phys
->crc
= calc_crc(ddf
->phys
, ddf
->pdsize
);
2838 if (write(fd
, ddf
->phys
, ddf
->pdsize
) < 0)
2840 ddf
->virt
->crc
= calc_crc(ddf
->virt
, ddf
->vdsize
);
2841 if (write(fd
, ddf
->virt
, ddf
->vdsize
) < 0)
2844 /* Now write lots of config records. */
2845 n_config
= ddf
->max_part
;
2846 conf_size
= ddf
->conf_rec_len
* 512;
2847 for (i
= 0 ; i
<= n_config
; i
++) {
2849 struct vd_config
*vdc
= NULL
;
2850 if (i
== n_config
) {
2851 c
= (struct vcl
*)d
->spare
;
2858 get_pd_index_from_refnum(
2861 (const struct vd_config
**)&vdc
,
2865 dprintf("writing conf record %i on disk %08x for %s/%u\n",
2866 i
, __be32_to_cpu(d
->disk
.refnum
),
2867 guid_str(vdc
->guid
),
2868 vdc
->sec_elmnt_seq
);
2869 vdc
->seqnum
= header
->seq
;
2870 vdc
->crc
= calc_crc(vdc
, conf_size
);
2871 if (write(fd
, vdc
, conf_size
) < 0)
2874 unsigned int togo
= conf_size
;
2875 while (togo
> NULL_CONF_SZ
) {
2876 if (write(fd
, null_aligned
, NULL_CONF_SZ
) < 0)
2878 togo
-= NULL_CONF_SZ
;
2880 if (write(fd
, null_aligned
, togo
) < 0)
2887 d
->disk
.crc
= calc_crc(&d
->disk
, 512);
2888 if (write(fd
, &d
->disk
, 512) < 0)
2893 header
->openflag
= 0;
2894 header
->crc
= calc_crc(header
, 512);
2896 lseek64(fd
, sector
<<9, 0);
2897 if (write(fd
, header
, 512) < 0)
2903 static int _write_super_to_disk(struct ddf_super
*ddf
, struct dl
*d
)
2905 unsigned long long size
;
2910 /* We need to fill in the primary, (secondary) and workspace
2911 * lba's in the headers, set their checksums,
2912 * Also checksum phys, virt....
2914 * Then write everything out, finally the anchor is written.
2916 get_dev_size(fd
, NULL
, &size
);
2918 if (d
->workspace_lba
!= 0)
2919 ddf
->anchor
.workspace_lba
= d
->workspace_lba
;
2921 ddf
->anchor
.workspace_lba
=
2922 __cpu_to_be64(size
- 32*1024*2);
2923 if (d
->primary_lba
!= 0)
2924 ddf
->anchor
.primary_lba
= d
->primary_lba
;
2926 ddf
->anchor
.primary_lba
=
2927 __cpu_to_be64(size
- 16*1024*2);
2928 if (d
->secondary_lba
!= 0)
2929 ddf
->anchor
.secondary_lba
= d
->secondary_lba
;
2931 ddf
->anchor
.secondary_lba
=
2932 __cpu_to_be64(size
- 32*1024*2);
2933 ddf
->anchor
.seq
= ddf
->active
->seq
;
2934 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2935 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2937 ddf
->anchor
.openflag
= 0xFF; /* 'open' means nothing */
2938 ddf
->anchor
.seq
= 0xFFFFFFFF; /* no sequencing in anchor */
2939 ddf
->anchor
.crc
= calc_crc(&ddf
->anchor
, 512);
2941 if (!__write_ddf_structure(d
, ddf
, DDF_HEADER_PRIMARY
))
2944 if (!__write_ddf_structure(d
, ddf
, DDF_HEADER_SECONDARY
))
2947 lseek64(fd
, (size
-1)*512, SEEK_SET
);
2948 if (write(fd
, &ddf
->anchor
, 512) < 0)
2954 static int __write_init_super_ddf(struct supertype
*st
)
2956 struct ddf_super
*ddf
= st
->sb
;
2961 pr_state(ddf
, __func__
);
2963 /* try to write updated metadata,
2964 * if we catch a failure move on to the next disk
2966 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
2968 successes
+= _write_super_to_disk(ddf
, d
);
2971 return attempts
!= successes
;
2974 static int write_init_super_ddf(struct supertype
*st
)
2976 struct ddf_super
*ddf
= st
->sb
;
2977 struct vcl
*currentconf
= ddf
->currentconf
;
2979 /* we are done with currentconf reset it to point st at the container */
2980 ddf
->currentconf
= NULL
;
2982 if (st
->update_tail
) {
2983 /* queue the virtual_disk and vd_config as metadata updates */
2984 struct virtual_disk
*vd
;
2985 struct vd_config
*vc
;
2990 int len
= (sizeof(struct phys_disk
) +
2991 sizeof(struct phys_disk_entry
));
2993 /* adding a disk to the container. */
2997 append_metadata_update(st
, ddf
->add_list
->mdupdate
, len
);
2998 ddf
->add_list
->mdupdate
= NULL
;
3002 /* Newly created VD */
3004 /* First the virtual disk. We have a slightly fake header */
3005 len
= sizeof(struct virtual_disk
) + sizeof(struct virtual_entry
);
3008 vd
->entries
[0] = ddf
->virt
->entries
[currentconf
->vcnum
];
3009 vd
->populated_vdes
= __cpu_to_be16(currentconf
->vcnum
);
3010 append_metadata_update(st
, vd
, len
);
3012 /* Then the vd_config */
3013 len
= ddf
->conf_rec_len
* 512;
3014 tlen
= len
* currentconf
->conf
.sec_elmnt_count
;
3016 memcpy(vc
, ¤tconf
->conf
, len
);
3017 for (i
= 1; i
< currentconf
->conf
.sec_elmnt_count
; i
++)
3018 memcpy((char *)vc
+ i
*len
, currentconf
->other_bvds
[i
-1],
3020 append_metadata_update(st
, vc
, tlen
);
3022 /* FIXME I need to close the fds! */
3027 for (d
= ddf
->dlist
; d
; d
=d
->next
)
3028 while (Kill(d
->devname
, NULL
, 0, -1, 1) == 0);
3029 return __write_init_super_ddf(st
);
3035 static __u64
avail_size_ddf(struct supertype
*st
, __u64 devsize
,
3036 unsigned long long data_offset
)
3038 /* We must reserve the last 32Meg */
3039 if (devsize
<= 32*1024*2)
3041 return devsize
- 32*1024*2;
3046 static int reserve_space(struct supertype
*st
, int raiddisks
,
3047 unsigned long long size
, int chunk
,
3048 unsigned long long *freesize
)
3050 /* Find 'raiddisks' spare extents at least 'size' big (but
3051 * only caring about multiples of 'chunk') and remember
3053 * If the cannot be found, fail.
3056 struct ddf_super
*ddf
= st
->sb
;
3059 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3063 /* Now find largest extent on each device */
3064 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3065 struct extent
*e
= get_extents(ddf
, dl
);
3066 unsigned long long pos
= 0;
3069 unsigned long long minsize
= size
;
3077 unsigned long long esize
;
3078 esize
= e
[i
].start
- pos
;
3079 if (esize
>= minsize
) {
3083 pos
= e
[i
].start
+ e
[i
].size
;
3085 } while (e
[i
-1].size
);
3088 dl
->esize
= minsize
;
3092 if (cnt
< raiddisks
) {
3093 pr_err("not enough devices with space to create array.\n");
3094 return 0; /* No enough free spaces large enough */
3097 /* choose the largest size of which there are at least 'raiddisk' */
3098 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3100 if (dl
->esize
<= size
)
3102 /* This is bigger than 'size', see if there are enough */
3104 for (dl2
= ddf
->dlist
; dl2
; dl2
=dl2
->next
)
3105 if (dl2
->esize
>= dl
->esize
)
3107 if (cnt
>= raiddisks
)
3111 size
= size
/ chunk
;
3116 pr_err("not enough spare devices to create array.\n");
3120 /* We have a 'size' of which there are enough spaces.
3121 * We simply do a first-fit */
3123 for (dl
= ddf
->dlist
; dl
&& cnt
< raiddisks
; dl
=dl
->next
) {
3124 if (dl
->esize
< size
)
3134 validate_geometry_ddf_container(struct supertype
*st
,
3135 int level
, int layout
, int raiddisks
,
3136 int chunk
, unsigned long long size
,
3137 unsigned long long data_offset
,
3138 char *dev
, unsigned long long *freesize
,
3141 static int validate_geometry_ddf_bvd(struct supertype
*st
,
3142 int level
, int layout
, int raiddisks
,
3143 int *chunk
, unsigned long long size
,
3144 unsigned long long data_offset
,
3145 char *dev
, unsigned long long *freesize
,
3148 static int validate_geometry_ddf(struct supertype
*st
,
3149 int level
, int layout
, int raiddisks
,
3150 int *chunk
, unsigned long long size
,
3151 unsigned long long data_offset
,
3152 char *dev
, unsigned long long *freesize
,
3159 /* ddf potentially supports lots of things, but it depends on
3160 * what devices are offered (and maybe kernel version?)
3161 * If given unused devices, we will make a container.
3162 * If given devices in a container, we will make a BVD.
3163 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
3166 if (chunk
&& *chunk
== UnSet
)
3167 *chunk
= DEFAULT_CHUNK
;
3169 if (level
== -1000000) level
= LEVEL_CONTAINER
;
3170 if (level
== LEVEL_CONTAINER
) {
3171 /* Must be a fresh device to add to a container */
3172 return validate_geometry_ddf_container(st
, level
, layout
,
3173 raiddisks
, chunk
?*chunk
:0,
3174 size
, data_offset
, dev
,
3180 mdu_array_info_t array
= {
3181 .level
= level
, .layout
= layout
,
3182 .raid_disks
= raiddisks
3184 struct vd_config conf
;
3185 if (layout_md2ddf(&array
, &conf
) == -1) {
3187 pr_err("DDF does not support level %d /layout %d arrays with %d disks\n",
3188 level
, layout
, raiddisks
);
3191 /* Should check layout? etc */
3193 if (st
->sb
&& freesize
) {
3194 /* --create was given a container to create in.
3195 * So we need to check that there are enough
3196 * free spaces and return the amount of space.
3197 * We may as well remember which drives were
3198 * chosen so that add_to_super/getinfo_super
3201 return reserve_space(st
, raiddisks
, size
, chunk
?*chunk
:0, freesize
);
3207 /* A container has already been opened, so we are
3208 * creating in there. Maybe a BVD, maybe an SVD.
3209 * Should make a distinction one day.
3211 return validate_geometry_ddf_bvd(st
, level
, layout
, raiddisks
,
3212 chunk
, size
, data_offset
, dev
,
3216 /* This is the first device for the array.
3217 * If it is a container, we read it in and do automagic allocations,
3218 * no other devices should be given.
3219 * Otherwise it must be a member device of a container, and we
3220 * do manual allocation.
3221 * Later we should check for a BVD and make an SVD.
3223 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3225 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
3227 if (sra
&& sra
->array
.major_version
== -1 &&
3228 strcmp(sra
->text_version
, "ddf") == 0) {
3231 /* find space for 'n' devices. */
3232 /* remember the devices */
3233 /* Somehow return the fact that we have enough */
3237 pr_err("ddf: Cannot create this array "
3238 "on device %s - a container is required.\n",
3242 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
3244 pr_err("ddf: Cannot open %s: %s\n",
3245 dev
, strerror(errno
));
3248 /* Well, it is in use by someone, maybe a 'ddf' container. */
3249 cfd
= open_container(fd
);
3253 pr_err("ddf: Cannot use %s: %s\n",
3254 dev
, strerror(EBUSY
));
3257 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
3259 if (sra
&& sra
->array
.major_version
== -1 &&
3260 strcmp(sra
->text_version
, "ddf") == 0) {
3261 /* This is a member of a ddf container. Load the container
3262 * and try to create a bvd
3264 struct ddf_super
*ddf
;
3265 if (load_super_ddf_all(st
, cfd
, (void **)&ddf
, NULL
) == 0) {
3267 strcpy(st
->container_devnm
, fd2devnm(cfd
));
3269 return validate_geometry_ddf_bvd(st
, level
, layout
,
3270 raiddisks
, chunk
, size
,
3276 } else /* device may belong to a different container */
3283 validate_geometry_ddf_container(struct supertype
*st
,
3284 int level
, int layout
, int raiddisks
,
3285 int chunk
, unsigned long long size
,
3286 unsigned long long data_offset
,
3287 char *dev
, unsigned long long *freesize
,
3291 unsigned long long ldsize
;
3293 if (level
!= LEVEL_CONTAINER
)
3298 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3301 pr_err("ddf: Cannot open %s: %s\n",
3302 dev
, strerror(errno
));
3305 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3311 *freesize
= avail_size_ddf(st
, ldsize
>> 9, INVALID_SECTORS
);
3318 static int validate_geometry_ddf_bvd(struct supertype
*st
,
3319 int level
, int layout
, int raiddisks
,
3320 int *chunk
, unsigned long long size
,
3321 unsigned long long data_offset
,
3322 char *dev
, unsigned long long *freesize
,
3326 struct ddf_super
*ddf
= st
->sb
;
3328 unsigned long long pos
= 0;
3329 unsigned long long maxsize
;
3332 /* ddf/bvd supports lots of things, but not containers */
3333 if (level
== LEVEL_CONTAINER
) {
3335 pr_err("DDF cannot create a container within an container\n");
3338 /* We must have the container info already read in. */
3343 /* General test: make sure there is space for
3344 * 'raiddisks' device extents of size 'size'.
3346 unsigned long long minsize
= size
;
3350 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
3356 e
= get_extents(ddf
, dl
);
3359 unsigned long long esize
;
3360 esize
= e
[i
].start
- pos
;
3361 if (esize
>= minsize
)
3363 pos
= e
[i
].start
+ e
[i
].size
;
3365 } while (e
[i
-1].size
);
3370 if (dcnt
< raiddisks
) {
3372 pr_err("ddf: Not enough devices with "
3373 "space for this array (%d < %d)\n",
3379 /* This device must be a member of the set */
3380 if (stat(dev
, &stb
) < 0)
3382 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3384 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
3385 if (dl
->major
== (int)major(stb
.st_rdev
) &&
3386 dl
->minor
== (int)minor(stb
.st_rdev
))
3391 pr_err("ddf: %s is not in the "
3396 e
= get_extents(ddf
, dl
);
3400 unsigned long long esize
;
3401 esize
= e
[i
].start
- pos
;
3402 if (esize
>= maxsize
)
3404 pos
= e
[i
].start
+ e
[i
].size
;
3406 } while (e
[i
-1].size
);
3407 *freesize
= maxsize
;
3413 static int load_super_ddf_all(struct supertype
*st
, int fd
,
3414 void **sbp
, char *devname
)
3417 struct ddf_super
*super
;
3418 struct mdinfo
*sd
, *best
= NULL
;
3424 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3427 if (sra
->array
.major_version
!= -1 ||
3428 sra
->array
.minor_version
!= -2 ||
3429 strcmp(sra
->text_version
, "ddf") != 0)
3432 if (posix_memalign((void**)&super
, 512, sizeof(*super
)) != 0)
3434 memset(super
, 0, sizeof(*super
));
3436 /* first, try each device, and choose the best ddf */
3437 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3439 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3440 dfd
= dev_open(nm
, O_RDONLY
);
3443 rv
= load_ddf_headers(dfd
, super
, NULL
);
3446 seq
= __be32_to_cpu(super
->active
->seq
);
3447 if (super
->active
->openflag
)
3449 if (!best
|| seq
> bestseq
) {
3457 /* OK, load this ddf */
3458 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
3459 dfd
= dev_open(nm
, O_RDONLY
);
3462 load_ddf_headers(dfd
, super
, NULL
);
3463 load_ddf_global(dfd
, super
, NULL
);
3465 /* Now we need the device-local bits */
3466 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3469 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3470 dfd
= dev_open(nm
, O_RDWR
);
3473 rv
= load_ddf_headers(dfd
, super
, NULL
);
3475 rv
= load_ddf_local(dfd
, super
, NULL
, 1);
3481 if (st
->ss
== NULL
) {
3482 st
->ss
= &super_ddf
;
3483 st
->minor_version
= 0;
3486 strcpy(st
->container_devnm
, fd2devnm(fd
));
3490 static int load_container_ddf(struct supertype
*st
, int fd
,
3493 return load_super_ddf_all(st
, fd
, &st
->sb
, devname
);
3496 #endif /* MDASSEMBLE */
3498 static int check_secondary(const struct vcl
*vc
)
3500 const struct vd_config
*conf
= &vc
->conf
;
3503 /* The only DDF secondary RAID level md can support is
3504 * RAID 10, if the stripe sizes and Basic volume sizes
3506 * Other configurations could in theory be supported by exposing
3507 * the BVDs to user space and using device mapper for the secondary
3508 * mapping. So far we don't support that.
3511 __u64 sec_elements
[4] = {0, 0, 0, 0};
3512 #define __set_sec_seen(n) (sec_elements[(n)>>6] |= (1<<((n)&63)))
3513 #define __was_sec_seen(n) ((sec_elements[(n)>>6] & (1<<((n)&63))) != 0)
3515 if (vc
->other_bvds
== NULL
) {
3516 pr_err("No BVDs for secondary RAID found\n");
3519 if (conf
->prl
!= DDF_RAID1
) {
3520 pr_err("Secondary RAID level only supported for mirrored BVD\n");
3523 if (conf
->srl
!= DDF_2STRIPED
&& conf
->srl
!= DDF_2SPANNED
) {
3524 pr_err("Secondary RAID level %d is unsupported\n",
3528 __set_sec_seen(conf
->sec_elmnt_seq
);
3529 for (i
= 0; i
< conf
->sec_elmnt_count
-1; i
++) {
3530 const struct vd_config
*bvd
= vc
->other_bvds
[i
];
3531 if (bvd
->sec_elmnt_seq
== DDF_UNUSED_BVD
)
3533 if (bvd
->srl
!= conf
->srl
) {
3534 pr_err("Inconsistent secondary RAID level across BVDs\n");
3537 if (bvd
->prl
!= conf
->prl
) {
3538 pr_err("Different RAID levels for BVDs are unsupported\n");
3541 if (bvd
->prim_elmnt_count
!= conf
->prim_elmnt_count
) {
3542 pr_err("All BVDs must have the same number of primary elements\n");
3545 if (bvd
->chunk_shift
!= conf
->chunk_shift
) {
3546 pr_err("Different strip sizes for BVDs are unsupported\n");
3549 if (bvd
->array_blocks
!= conf
->array_blocks
) {
3550 pr_err("Different BVD sizes are unsupported\n");
3553 __set_sec_seen(bvd
->sec_elmnt_seq
);
3555 for (i
= 0; i
< conf
->sec_elmnt_count
; i
++) {
3556 if (!__was_sec_seen(i
)) {
3557 pr_err("BVD %d is missing\n", i
);
3564 static unsigned int get_pd_index_from_refnum(const struct vcl
*vc
,
3565 __u32 refnum
, unsigned int nmax
,
3566 const struct vd_config
**bvd
,
3569 unsigned int i
, j
, n
, sec
, cnt
;
3571 cnt
= __be16_to_cpu(vc
->conf
.prim_elmnt_count
);
3572 sec
= (vc
->conf
.sec_elmnt_count
== 1 ? 0 : vc
->conf
.sec_elmnt_seq
);
3574 for (i
= 0, j
= 0 ; i
< nmax
; i
++) {
3575 /* j counts valid entries for this BVD */
3576 if (vc
->conf
.phys_refnum
[i
] != 0xffffffff)
3578 if (vc
->conf
.phys_refnum
[i
] == refnum
) {
3581 return sec
* cnt
+ j
- 1;
3584 if (vc
->other_bvds
== NULL
)
3587 for (n
= 1; n
< vc
->conf
.sec_elmnt_count
; n
++) {
3588 struct vd_config
*vd
= vc
->other_bvds
[n
-1];
3589 sec
= vd
->sec_elmnt_seq
;
3590 if (sec
== DDF_UNUSED_BVD
)
3592 for (i
= 0, j
= 0 ; i
< nmax
; i
++) {
3593 if (vd
->phys_refnum
[i
] != 0xffffffff)
3595 if (vd
->phys_refnum
[i
] == refnum
) {
3598 return sec
* cnt
+ j
- 1;
3604 return DDF_NOTFOUND
;
3607 static struct mdinfo
*container_content_ddf(struct supertype
*st
, char *subarray
)
3609 /* Given a container loaded by load_super_ddf_all,
3610 * extract information about all the arrays into
3613 * For each vcl in conflist: create an mdinfo, fill it in,
3614 * then look for matching devices (phys_refnum) in dlist
3615 * and create appropriate device mdinfo.
3617 struct ddf_super
*ddf
= st
->sb
;
3618 struct mdinfo
*rest
= NULL
;
3621 for (vc
= ddf
->conflist
; vc
; vc
=vc
->next
)
3625 struct mdinfo
*this;
3631 (strtoul(subarray
, &ep
, 10) != vc
->vcnum
||
3635 if (vc
->conf
.sec_elmnt_count
> 1) {
3636 if (check_secondary(vc
) != 0)
3640 this = xcalloc(1, sizeof(*this));
3644 if (layout_ddf2md(&vc
->conf
, &this->array
))
3646 this->array
.md_minor
= -1;
3647 this->array
.major_version
= -1;
3648 this->array
.minor_version
= -2;
3649 cptr
= (__u32
*)(vc
->conf
.guid
+ 16);
3650 this->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
3651 this->array
.utime
= DECADE
+
3652 __be32_to_cpu(vc
->conf
.timestamp
);
3653 this->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
3656 if ((ddf
->virt
->entries
[i
].state
& DDF_state_inconsistent
) ||
3657 (ddf
->virt
->entries
[i
].init_state
& DDF_initstate_mask
) !=
3659 this->array
.state
= 0;
3660 this->resync_start
= 0;
3662 this->array
.state
= 1;
3663 this->resync_start
= MaxSector
;
3665 memcpy(this->name
, ddf
->virt
->entries
[i
].name
, 16);
3668 if (this->name
[j
] == ' ')
3671 memset(this->uuid
, 0, sizeof(this->uuid
));
3672 this->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
3673 this->array
.size
= this->component_size
/ 2;
3674 this->container_member
= i
;
3676 ddf
->currentconf
= vc
;
3677 uuid_from_super_ddf(st
, this->uuid
);
3679 ddf
->currentconf
= NULL
;
3681 sprintf(this->text_version
, "/%s/%d",
3682 st
->container_devnm
, this->container_member
);
3684 for (pd
= 0; pd
< __be16_to_cpu(ddf
->phys
->used_pdes
); pd
++) {
3687 const struct vd_config
*bvd
;
3691 if (ddf
->phys
->entries
[pd
].refnum
== 0xFFFFFFFF)
3694 stt
= __be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
3695 if ((stt
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
3699 i
= get_pd_index_from_refnum(
3700 vc
, ddf
->phys
->entries
[pd
].refnum
,
3701 ddf
->mppe
, &bvd
, &iphys
);
3702 if (i
== DDF_NOTFOUND
)
3705 this->array
.working_disks
++;
3707 for (d
= ddf
->dlist
; d
; d
=d
->next
)
3708 if (d
->disk
.refnum
==
3709 ddf
->phys
->entries
[pd
].refnum
)
3712 /* Haven't found that one yet, maybe there are others */
3715 dev
= xcalloc(1, sizeof(*dev
));
3716 dev
->next
= this->devs
;
3719 dev
->disk
.number
= __be32_to_cpu(d
->disk
.refnum
);
3720 dev
->disk
.major
= d
->major
;
3721 dev
->disk
.minor
= d
->minor
;
3722 dev
->disk
.raid_disk
= i
;
3723 dev
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
3724 dev
->recovery_start
= MaxSector
;
3726 dev
->events
= __be32_to_cpu(ddf
->primary
.seq
);
3728 __be64_to_cpu(LBA_OFFSET(ddf
, bvd
)[iphys
]);
3729 dev
->component_size
= __be64_to_cpu(bvd
->blocks
);
3731 strcpy(dev
->name
, d
->devname
);
3737 static int store_super_ddf(struct supertype
*st
, int fd
)
3739 struct ddf_super
*ddf
= st
->sb
;
3740 unsigned long long dsize
;
3747 if (!get_dev_size(fd
, NULL
, &dsize
))
3750 if (ddf
->dlist
|| ddf
->conflist
) {
3755 if (fstat(fd
, &sta
) == -1 || !S_ISBLK(sta
.st_mode
)) {
3756 pr_err("%s: file descriptor for invalid device\n",
3760 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
3761 if (dl
->major
== (int)major(sta
.st_rdev
) &&
3762 dl
->minor
== (int)minor(sta
.st_rdev
))
3765 pr_err("%s: couldn't find disk %d/%d\n", __func__
,
3766 (int)major(sta
.st_rdev
),
3767 (int)minor(sta
.st_rdev
));
3772 ret
= (_write_super_to_disk(ddf
, dl
) != 1);
3777 if (posix_memalign(&buf
, 512, 512) != 0)
3779 memset(buf
, 0, 512);
3781 lseek64(fd
, dsize
-512, 0);
3782 rc
= write(fd
, buf
, 512);
3789 static int compare_super_ddf(struct supertype
*st
, struct supertype
*tst
)
3793 * 0 same, or first was empty, and second was copied
3794 * 1 second had wrong number
3796 * 3 wrong other info
3798 struct ddf_super
*first
= st
->sb
;
3799 struct ddf_super
*second
= tst
->sb
;
3800 struct dl
*dl1
, *dl2
;
3801 struct vcl
*vl1
, *vl2
;
3802 unsigned int max_vds
, max_pds
, pd
, vd
;
3810 if (memcmp(first
->anchor
.guid
, second
->anchor
.guid
, DDF_GUID_LEN
) != 0)
3813 if (first
->anchor
.seq
!= second
->anchor
.seq
) {
3814 dprintf("%s: sequence number mismatch %u/%u\n", __func__
,
3815 __be32_to_cpu(first
->anchor
.seq
),
3816 __be32_to_cpu(second
->anchor
.seq
));
3819 if (first
->max_part
!= second
->max_part
||
3820 first
->phys
->used_pdes
!= second
->phys
->used_pdes
||
3821 first
->virt
->populated_vdes
!= second
->virt
->populated_vdes
) {
3822 dprintf("%s: PD/VD number mismatch\n", __func__
);
3826 max_pds
= __be16_to_cpu(first
->phys
->used_pdes
);
3827 for (dl2
= second
->dlist
; dl2
; dl2
= dl2
->next
) {
3828 for (pd
= 0; pd
< max_pds
; pd
++)
3829 if (first
->phys
->entries
[pd
].refnum
== dl2
->disk
.refnum
)
3831 if (pd
== max_pds
) {
3832 dprintf("%s: no match for disk %08x\n", __func__
,
3833 __be32_to_cpu(dl2
->disk
.refnum
));
3838 max_vds
= __be16_to_cpu(first
->active
->max_vd_entries
);
3839 for (vl2
= second
->conflist
; vl2
; vl2
= vl2
->next
) {
3840 if (vl2
->conf
.magic
!= DDF_VD_CONF_MAGIC
)
3842 for (vd
= 0; vd
< max_vds
; vd
++)
3843 if (!memcmp(first
->virt
->entries
[vd
].guid
,
3844 vl2
->conf
.guid
, DDF_GUID_LEN
))
3846 if (vd
== max_vds
) {
3847 dprintf("%s: no match for VD config\n", __func__
);
3851 /* FIXME should I look at anything else? */
3854 At this point we are fairly sure that the meta data matches.
3855 But the new disk may contain additional local data.
3856 Add it to the super block.
3858 for (vl2
= second
->conflist
; vl2
; vl2
= vl2
->next
) {
3859 for (vl1
= first
->conflist
; vl1
; vl1
= vl1
->next
)
3860 if (!memcmp(vl1
->conf
.guid
, vl2
->conf
.guid
,
3864 if (vl1
->other_bvds
!= NULL
&&
3865 vl1
->conf
.sec_elmnt_seq
!=
3866 vl2
->conf
.sec_elmnt_seq
) {
3867 dprintf("%s: adding BVD %u\n", __func__
,
3868 vl2
->conf
.sec_elmnt_seq
);
3869 add_other_bvd(vl1
, &vl2
->conf
,
3870 first
->conf_rec_len
*512);
3875 if (posix_memalign((void **)&vl1
, 512,
3876 (first
->conf_rec_len
*512 +
3877 offsetof(struct vcl
, conf
))) != 0) {
3878 pr_err("%s could not allocate vcl buf\n",
3883 vl1
->next
= first
->conflist
;
3884 vl1
->block_sizes
= NULL
;
3885 memcpy(&vl1
->conf
, &vl2
->conf
, first
->conf_rec_len
*512);
3886 if (alloc_other_bvds(first
, vl1
) != 0) {
3887 pr_err("%s could not allocate other bvds\n",
3892 for (vd
= 0; vd
< max_vds
; vd
++)
3893 if (!memcmp(first
->virt
->entries
[vd
].guid
,
3894 vl1
->conf
.guid
, DDF_GUID_LEN
))
3897 dprintf("%s: added config for VD %u\n", __func__
, vl1
->vcnum
);
3898 first
->conflist
= vl1
;
3901 for (dl2
= second
->dlist
; dl2
; dl2
= dl2
->next
) {
3902 for (dl1
= first
->dlist
; dl1
; dl1
= dl1
->next
)
3903 if (dl1
->disk
.refnum
== dl2
->disk
.refnum
)
3908 if (posix_memalign((void **)&dl1
, 512,
3909 sizeof(*dl1
) + (first
->max_part
) * sizeof(dl1
->vlist
[0]))
3911 pr_err("%s could not allocate disk info buffer\n",
3915 memcpy(dl1
, dl2
, sizeof(*dl1
));
3916 dl1
->mdupdate
= NULL
;
3917 dl1
->next
= first
->dlist
;
3919 for (pd
= 0; pd
< max_pds
; pd
++)
3920 if (first
->phys
->entries
[pd
].refnum
== dl1
->disk
.refnum
)
3924 if (posix_memalign((void **)&dl1
->spare
, 512,
3925 first
->conf_rec_len
*512) != 0) {
3926 pr_err("%s could not allocate spare info buf\n",
3930 memcpy(dl1
->spare
, dl2
->spare
, first
->conf_rec_len
*512);
3932 for (vd
= 0 ; vd
< first
->max_part
; vd
++) {
3933 if (!dl2
->vlist
[vd
]) {
3934 dl1
->vlist
[vd
] = NULL
;
3937 for (vl1
= first
->conflist
; vl1
; vl1
= vl1
->next
) {
3938 if (!memcmp(vl1
->conf
.guid
,
3939 dl2
->vlist
[vd
]->conf
.guid
,
3942 dl1
->vlist
[vd
] = vl1
;
3946 dprintf("%s: added disk %d: %08x\n", __func__
, dl1
->pdnum
,
3947 __be32_to_cpu(dl1
->disk
.refnum
));
3955 * A new array 'a' has been started which claims to be instance 'inst'
3956 * within container 'c'.
3957 * We need to confirm that the array matches the metadata in 'c' so
3958 * that we don't corrupt any metadata.
3960 static int ddf_open_new(struct supertype
*c
, struct active_array
*a
, char *inst
)
3962 struct ddf_super
*ddf
= c
->sb
;
3964 if (all_ff(ddf
->virt
->entries
[n
].guid
)) {
3965 pr_err("%s: subarray %d doesn't exist\n", __func__
, n
);
3968 dprintf("ddf: open_new %d\n", n
);
3969 a
->info
.container_member
= n
;
3974 * The array 'a' is to be marked clean in the metadata.
3975 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
3976 * clean up to the point (in sectors). If that cannot be recorded in the
3977 * metadata, then leave it as dirty.
3979 * For DDF, we need to clear the DDF_state_inconsistent bit in the
3980 * !global! virtual_disk.virtual_entry structure.
3982 static int ddf_set_array_state(struct active_array
*a
, int consistent
)
3984 struct ddf_super
*ddf
= a
->container
->sb
;
3985 int inst
= a
->info
.container_member
;
3986 int old
= ddf
->virt
->entries
[inst
].state
;
3987 if (consistent
== 2) {
3988 /* Should check if a recovery should be started FIXME */
3990 if (!is_resync_complete(&a
->info
))
3994 ddf
->virt
->entries
[inst
].state
&= ~DDF_state_inconsistent
;
3996 ddf
->virt
->entries
[inst
].state
|= DDF_state_inconsistent
;
3997 if (old
!= ddf
->virt
->entries
[inst
].state
)
3998 ddf_set_updates_pending(ddf
);
4000 old
= ddf
->virt
->entries
[inst
].init_state
;
4001 ddf
->virt
->entries
[inst
].init_state
&= ~DDF_initstate_mask
;
4002 if (is_resync_complete(&a
->info
))
4003 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_full
;
4004 else if (a
->info
.resync_start
== 0)
4005 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_not
;
4007 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_quick
;
4008 if (old
!= ddf
->virt
->entries
[inst
].init_state
)
4009 ddf_set_updates_pending(ddf
);
4011 dprintf("ddf mark %d/%s (%d) %s %llu\n", inst
,
4012 guid_str(ddf
->virt
->entries
[inst
].guid
), a
->curr_state
,
4013 consistent
?"clean":"dirty",
4014 a
->info
.resync_start
);
4018 static int get_bvd_state(const struct ddf_super
*ddf
,
4019 const struct vd_config
*vc
)
4021 unsigned int i
, n_bvd
, working
= 0;
4022 unsigned int n_prim
= __be16_to_cpu(vc
->prim_elmnt_count
);
4024 for (i
= 0; i
< n_prim
; i
++) {
4025 if (!find_index_in_bvd(ddf
, vc
, i
, &n_bvd
))
4027 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
4030 st
= __be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
4031 if ((st
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
4036 state
= DDF_state_degraded
;
4037 if (working
== n_prim
)
4038 state
= DDF_state_optimal
;
4044 state
= DDF_state_failed
;
4048 state
= DDF_state_failed
;
4049 else if (working
>= 2)
4050 state
= DDF_state_part_optimal
;
4054 if (working
< n_prim
- 1)
4055 state
= DDF_state_failed
;
4058 if (working
< n_prim
- 2)
4059 state
= DDF_state_failed
;
4060 else if (working
== n_prim
- 1)
4061 state
= DDF_state_part_optimal
;
4067 static int secondary_state(int state
, int other
, int seclevel
)
4069 if (state
== DDF_state_optimal
&& other
== DDF_state_optimal
)
4070 return DDF_state_optimal
;
4071 if (seclevel
== DDF_2MIRRORED
) {
4072 if (state
== DDF_state_optimal
|| other
== DDF_state_optimal
)
4073 return DDF_state_part_optimal
;
4074 if (state
== DDF_state_failed
&& other
== DDF_state_failed
)
4075 return DDF_state_failed
;
4076 return DDF_state_degraded
;
4078 if (state
== DDF_state_failed
|| other
== DDF_state_failed
)
4079 return DDF_state_failed
;
4080 if (state
== DDF_state_degraded
|| other
== DDF_state_degraded
)
4081 return DDF_state_degraded
;
4082 return DDF_state_part_optimal
;
4086 static int get_svd_state(const struct ddf_super
*ddf
, const struct vcl
*vcl
)
4088 int state
= get_bvd_state(ddf
, &vcl
->conf
);
4090 for (i
= 1; i
< vcl
->conf
.sec_elmnt_count
; i
++) {
4091 state
= secondary_state(
4093 get_bvd_state(ddf
, vcl
->other_bvds
[i
-1]),
4100 * The state of each disk is stored in the global phys_disk structure
4101 * in phys_disk.entries[n].state.
4102 * This makes various combinations awkward.
4103 * - When a device fails in any array, it must be failed in all arrays
4104 * that include a part of this device.
4105 * - When a component is rebuilding, we cannot include it officially in the
4106 * array unless this is the only array that uses the device.
4108 * So: when transitioning:
4109 * Online -> failed, just set failed flag. monitor will propagate
4110 * spare -> online, the device might need to be added to the array.
4111 * spare -> failed, just set failed. Don't worry if in array or not.
4113 static void ddf_set_disk(struct active_array
*a
, int n
, int state
)
4115 struct ddf_super
*ddf
= a
->container
->sb
;
4116 unsigned int inst
= a
->info
.container_member
, n_bvd
;
4118 struct vd_config
*vc
= find_vdcr(ddf
, inst
, (unsigned int)n
,
4125 dprintf("ddf: cannot find instance %d!!\n", inst
);
4128 /* Find the matching slot in 'info'. */
4129 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
4130 if (mdi
->disk
.raid_disk
== n
)
4135 /* and find the 'dl' entry corresponding to that. */
4136 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4137 if (mdi
->state_fd
>= 0 &&
4138 mdi
->disk
.major
== dl
->major
&&
4139 mdi
->disk
.minor
== dl
->minor
)
4144 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
4145 if (pd
< 0 || pd
!= dl
->pdnum
) {
4146 /* disk doesn't currently exist or has changed.
4147 * If it is now in_sync, insert it. */
4148 dprintf("%s: phys disk not found for %d: %d/%d ref %08x\n",
4149 __func__
, dl
->pdnum
, dl
->major
, dl
->minor
,
4150 __be32_to_cpu(dl
->disk
.refnum
));
4151 dprintf("%s: array %u disk %u ref %08x pd %d\n",
4152 __func__
, inst
, n_bvd
, vc
->phys_refnum
[n_bvd
], pd
);
4153 if ((state
& DS_INSYNC
) && ! (state
& DS_FAULTY
)) {
4154 pd
= dl
->pdnum
; /* FIXME: is this really correct ? */
4155 vc
->phys_refnum
[n_bvd
] = dl
->disk
.refnum
;
4156 LBA_OFFSET(ddf
, vc
)[n_bvd
] =
4157 __cpu_to_be64(mdi
->data_offset
);
4158 ddf
->phys
->entries
[pd
].type
&=
4159 ~__cpu_to_be16(DDF_Global_Spare
);
4160 ddf
->phys
->entries
[pd
].type
|=
4161 __cpu_to_be16(DDF_Active_in_VD
);
4162 ddf_set_updates_pending(ddf
);
4165 int old
= ddf
->phys
->entries
[pd
].state
;
4166 if (state
& DS_FAULTY
)
4167 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Failed
);
4168 if (state
& DS_INSYNC
) {
4169 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Online
);
4170 ddf
->phys
->entries
[pd
].state
&= __cpu_to_be16(~DDF_Rebuilding
);
4172 if (old
!= ddf
->phys
->entries
[pd
].state
)
4173 ddf_set_updates_pending(ddf
);
4176 dprintf("ddf: set_disk %d to %x\n", n
, state
);
4178 /* Now we need to check the state of the array and update
4179 * virtual_disk.entries[n].state.
4180 * It needs to be one of "optimal", "degraded", "failed".
4181 * I don't understand 'deleted' or 'missing'.
4183 state
= get_svd_state(ddf
, vcl
);
4185 if (ddf
->virt
->entries
[inst
].state
!=
4186 ((ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4189 ddf
->virt
->entries
[inst
].state
=
4190 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4192 ddf_set_updates_pending(ddf
);
4197 static void ddf_sync_metadata(struct supertype
*st
)
4201 * Write all data to all devices.
4202 * Later, we might be able to track whether only local changes
4203 * have been made, or whether any global data has been changed,
4204 * but ddf is sufficiently weird that it probably always
4205 * changes global data ....
4207 struct ddf_super
*ddf
= st
->sb
;
4208 if (!ddf
->updates_pending
)
4210 ddf
->updates_pending
= 0;
4211 __write_init_super_ddf(st
);
4212 dprintf("ddf: sync_metadata\n");
4215 static int del_from_conflist(struct vcl
**list
, const char *guid
)
4219 for (p
= list
; p
&& *p
; p
= &((*p
)->next
))
4220 if (!memcmp((*p
)->conf
.guid
, guid
, DDF_GUID_LEN
)) {
4227 static int _kill_subarray_ddf(struct ddf_super
*ddf
, const char *guid
)
4230 unsigned int vdnum
, i
;
4231 vdnum
= find_vde_by_guid(ddf
, guid
);
4232 if (vdnum
== DDF_NOTFOUND
) {
4233 pr_err("%s: could not find VD %s\n", __func__
,
4237 if (del_from_conflist(&ddf
->conflist
, guid
) == 0) {
4238 pr_err("%s: could not find conf %s\n", __func__
,
4242 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4243 for (i
= 0; i
< ddf
->max_part
; i
++)
4244 if (dl
->vlist
[i
] != NULL
&&
4245 !memcmp(dl
->vlist
[i
]->conf
.guid
, guid
,
4247 dl
->vlist
[i
] = NULL
;
4248 memset(ddf
->virt
->entries
[vdnum
].guid
, 0xff, DDF_GUID_LEN
);
4249 dprintf("%s: deleted %s\n", __func__
, guid_str(guid
));
4253 static int kill_subarray_ddf(struct supertype
*st
)
4255 struct ddf_super
*ddf
= st
->sb
;
4257 * currentconf is set in container_content_ddf,
4258 * called with subarray arg
4260 struct vcl
*victim
= ddf
->currentconf
;
4261 struct vd_config
*conf
;
4262 ddf
->currentconf
= NULL
;
4265 pr_err("%s: nothing to kill\n", __func__
);
4268 conf
= &victim
->conf
;
4269 vdnum
= find_vde_by_guid(ddf
, conf
->guid
);
4270 if (vdnum
== DDF_NOTFOUND
) {
4271 pr_err("%s: could not find VD %s\n", __func__
,
4272 guid_str(conf
->guid
));
4275 if (st
->update_tail
) {
4276 struct virtual_disk
*vd
;
4277 int len
= sizeof(struct virtual_disk
)
4278 + sizeof(struct virtual_entry
);
4281 pr_err("%s: failed to allocate %d bytes\n", __func__
,
4285 memset(vd
, 0 , len
);
4286 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
4287 vd
->populated_vdes
= 0;
4288 memcpy(vd
->entries
[0].guid
, conf
->guid
, DDF_GUID_LEN
);
4289 /* we use DDF_state_deleted as marker */
4290 vd
->entries
[0].state
= DDF_state_deleted
;
4291 append_metadata_update(st
, vd
, len
);
4293 _kill_subarray_ddf(ddf
, conf
->guid
);
4294 ddf_set_updates_pending(ddf
);
4295 ddf_sync_metadata(st
);
4300 static void copy_matching_bvd(struct ddf_super
*ddf
,
4301 struct vd_config
*conf
,
4302 const struct metadata_update
*update
)
4305 __be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
4306 unsigned int len
= ddf
->conf_rec_len
* 512;
4308 struct vd_config
*vc
;
4309 for (p
= update
->buf
; p
< update
->buf
+ update
->len
; p
+= len
) {
4310 vc
= (struct vd_config
*) p
;
4311 if (vc
->sec_elmnt_seq
== conf
->sec_elmnt_seq
) {
4312 memcpy(conf
->phys_refnum
, vc
->phys_refnum
,
4313 mppe
* (sizeof(__u32
) + sizeof(__u64
)));
4317 pr_err("%s: no match for BVD %d of %s in update\n", __func__
,
4318 conf
->sec_elmnt_seq
, guid_str(conf
->guid
));
4321 static void ddf_process_update(struct supertype
*st
,
4322 struct metadata_update
*update
)
4324 /* Apply this update to the metadata.
4325 * The first 4 bytes are a DDF_*_MAGIC which guides
4327 * Possible update are:
4328 * DDF_PHYS_RECORDS_MAGIC
4329 * Add a new physical device or remove an old one.
4330 * Changes to this record only happen implicitly.
4331 * used_pdes is the device number.
4332 * DDF_VIRT_RECORDS_MAGIC
4333 * Add a new VD. Possibly also change the 'access' bits.
4334 * populated_vdes is the entry number.
4336 * New or updated VD. the VIRT_RECORD must already
4337 * exist. For an update, phys_refnum and lba_offset
4338 * (at least) are updated, and the VD_CONF must
4339 * be written to precisely those devices listed with
4341 * DDF_SPARE_ASSIGN_MAGIC
4342 * replacement Spare Assignment Record... but for which device?
4345 * - to create a new array, we send a VIRT_RECORD and
4346 * a VD_CONF. Then assemble and start the array.
4347 * - to activate a spare we send a VD_CONF to add the phys_refnum
4348 * and offset. This will also mark the spare as active with
4349 * a spare-assignment record.
4351 struct ddf_super
*ddf
= st
->sb
;
4352 __u32
*magic
= (__u32
*)update
->buf
;
4353 struct phys_disk
*pd
;
4354 struct virtual_disk
*vd
;
4355 struct vd_config
*vc
;
4359 unsigned int pdnum
, pd2
, len
;
4361 dprintf("Process update %x\n", *magic
);
4364 case DDF_PHYS_RECORDS_MAGIC
:
4366 if (update
->len
!= (sizeof(struct phys_disk
) +
4367 sizeof(struct phys_disk_entry
)))
4369 pd
= (struct phys_disk
*)update
->buf
;
4371 ent
= __be16_to_cpu(pd
->used_pdes
);
4372 if (ent
>= __be16_to_cpu(ddf
->phys
->max_pdes
))
4374 if (pd
->entries
[0].state
& __cpu_to_be16(DDF_Missing
)) {
4376 /* removing this disk. */
4377 ddf
->phys
->entries
[ent
].state
|= __cpu_to_be16(DDF_Missing
);
4378 for (dlp
= &ddf
->dlist
; *dlp
; dlp
= &(*dlp
)->next
) {
4379 struct dl
*dl
= *dlp
;
4380 if (dl
->pdnum
== (signed)ent
) {
4383 /* FIXME this doesn't free
4390 ddf_set_updates_pending(ddf
);
4393 if (!all_ff(ddf
->phys
->entries
[ent
].guid
))
4395 ddf
->phys
->entries
[ent
] = pd
->entries
[0];
4396 ddf
->phys
->used_pdes
= __cpu_to_be16(1 +
4397 __be16_to_cpu(ddf
->phys
->used_pdes
));
4398 ddf_set_updates_pending(ddf
);
4399 if (ddf
->add_list
) {
4400 struct active_array
*a
;
4401 struct dl
*al
= ddf
->add_list
;
4402 ddf
->add_list
= al
->next
;
4404 al
->next
= ddf
->dlist
;
4407 /* As a device has been added, we should check
4408 * for any degraded devices that might make
4409 * use of this spare */
4410 for (a
= st
->arrays
; a
; a
=a
->next
)
4411 a
->check_degraded
= 1;
4415 case DDF_VIRT_RECORDS_MAGIC
:
4417 if (update
->len
!= (sizeof(struct virtual_disk
) +
4418 sizeof(struct virtual_entry
)))
4420 vd
= (struct virtual_disk
*)update
->buf
;
4422 if (vd
->entries
[0].state
== DDF_state_deleted
) {
4423 if (_kill_subarray_ddf(ddf
, vd
->entries
[0].guid
))
4427 ent
= find_vde_by_guid(ddf
, vd
->entries
[0].guid
);
4428 if (ent
!= DDF_NOTFOUND
) {
4429 dprintf("%s: VD %s exists already in slot %d\n",
4430 __func__
, guid_str(vd
->entries
[0].guid
),
4434 ent
= find_unused_vde(ddf
);
4435 if (ent
== DDF_NOTFOUND
)
4437 ddf
->virt
->entries
[ent
] = vd
->entries
[0];
4438 ddf
->virt
->populated_vdes
=
4441 ddf
->virt
->populated_vdes
));
4442 dprintf("%s: added VD %s in slot %d(s=%02x i=%02x)\n",
4443 __func__
, guid_str(vd
->entries
[0].guid
), ent
,
4444 ddf
->virt
->entries
[ent
].state
,
4445 ddf
->virt
->entries
[ent
].init_state
);
4447 ddf_set_updates_pending(ddf
);
4450 case DDF_VD_CONF_MAGIC
:
4451 vc
= (struct vd_config
*)update
->buf
;
4452 len
= ddf
->conf_rec_len
* 512;
4453 if ((unsigned int)update
->len
!= len
* vc
->sec_elmnt_count
) {
4454 pr_err("%s: %s: insufficient data (%d) for %u BVDs\n",
4455 __func__
, guid_str(vc
->guid
), update
->len
,
4456 vc
->sec_elmnt_count
);
4459 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
4460 if (memcmp(vcl
->conf
.guid
, vc
->guid
, DDF_GUID_LEN
) == 0)
4462 dprintf("%s: conf update for %s (%s)\n", __func__
,
4463 guid_str(vc
->guid
), (vcl
? "old" : "new"));
4465 /* An update, just copy the phys_refnum and lba_offset
4469 copy_matching_bvd(ddf
, &vcl
->conf
, update
);
4470 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++)
4471 copy_matching_bvd(ddf
, vcl
->other_bvds
[i
-1],
4478 vcl
= update
->space
;
4479 update
->space
= NULL
;
4480 vcl
->next
= ddf
->conflist
;
4481 memcpy(&vcl
->conf
, vc
, len
);
4482 ent
= find_vde_by_guid(ddf
, vc
->guid
);
4483 if (ent
== DDF_NOTFOUND
)
4486 ddf
->conflist
= vcl
;
4487 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++)
4488 memcpy(vcl
->other_bvds
[i
-1],
4489 update
->buf
+ len
* i
, len
);
4491 /* Set DDF_Transition on all Failed devices - to help
4492 * us detect those that are no longer in use
4494 for (pdnum
= 0; pdnum
< __be16_to_cpu(ddf
->phys
->used_pdes
); pdnum
++)
4495 if (ddf
->phys
->entries
[pdnum
].state
4496 & __be16_to_cpu(DDF_Failed
))
4497 ddf
->phys
->entries
[pdnum
].state
4498 |= __be16_to_cpu(DDF_Transition
);
4499 /* Now make sure vlist is correct for each dl. */
4500 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
4501 unsigned int vn
= 0;
4502 int in_degraded
= 0;
4503 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
) {
4504 unsigned int dn
, ibvd
;
4505 const struct vd_config
*conf
;
4507 dn
= get_pd_index_from_refnum(vcl
,
4511 if (dn
== DDF_NOTFOUND
)
4513 dprintf("dev %d/%08x has %s (sec=%u) at %d\n",
4515 __be32_to_cpu(dl
->disk
.refnum
),
4516 guid_str(conf
->guid
),
4517 conf
->sec_elmnt_seq
, vn
);
4518 /* Clear the Transition flag */
4519 if (ddf
->phys
->entries
[dl
->pdnum
].state
4520 & __be16_to_cpu(DDF_Failed
))
4521 ddf
->phys
->entries
[dl
->pdnum
].state
&=
4522 ~__be16_to_cpu(DDF_Transition
);
4523 dl
->vlist
[vn
++] = vcl
;
4524 vstate
= ddf
->virt
->entries
[vcl
->vcnum
].state
4526 if (vstate
== DDF_state_degraded
||
4527 vstate
== DDF_state_part_optimal
)
4530 while (vn
< ddf
->max_part
)
4531 dl
->vlist
[vn
++] = NULL
;
4533 ddf
->phys
->entries
[dl
->pdnum
].type
&=
4534 ~__cpu_to_be16(DDF_Global_Spare
);
4535 if (!(ddf
->phys
->entries
[dl
->pdnum
].type
&
4536 __cpu_to_be16(DDF_Active_in_VD
))) {
4537 ddf
->phys
->entries
[dl
->pdnum
].type
|=
4538 __cpu_to_be16(DDF_Active_in_VD
);
4540 ddf
->phys
->entries
[dl
->pdnum
].state
|=
4541 __cpu_to_be16(DDF_Rebuilding
);
4545 ddf
->phys
->entries
[dl
->pdnum
].type
&=
4546 ~__cpu_to_be16(DDF_Global_Spare
);
4547 ddf
->phys
->entries
[dl
->pdnum
].type
|=
4548 __cpu_to_be16(DDF_Spare
);
4550 if (!dl
->vlist
[0] && !dl
->spare
) {
4551 ddf
->phys
->entries
[dl
->pdnum
].type
|=
4552 __cpu_to_be16(DDF_Global_Spare
);
4553 ddf
->phys
->entries
[dl
->pdnum
].type
&=
4554 ~__cpu_to_be16(DDF_Spare
|
4559 /* Now remove any 'Failed' devices that are not part
4560 * of any VD. They will have the Transition flag set.
4561 * Once done, we need to update all dl->pdnum numbers.
4564 for (pdnum
= 0; pdnum
< __be16_to_cpu(ddf
->phys
->used_pdes
); pdnum
++)
4565 if ((ddf
->phys
->entries
[pdnum
].state
4566 & __be16_to_cpu(DDF_Failed
))
4567 && (ddf
->phys
->entries
[pdnum
].state
4568 & __be16_to_cpu(DDF_Transition
)))
4569 /* skip this one */;
4570 else if (pdnum
== pd2
)
4573 ddf
->phys
->entries
[pd2
] = ddf
->phys
->entries
[pdnum
];
4574 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4575 if (dl
->pdnum
== (int)pdnum
)
4579 ddf
->phys
->used_pdes
= __cpu_to_be16(pd2
);
4580 while (pd2
< pdnum
) {
4581 memset(ddf
->phys
->entries
[pd2
].guid
, 0xff, DDF_GUID_LEN
);
4585 ddf_set_updates_pending(ddf
);
4587 case DDF_SPARE_ASSIGN_MAGIC
:
4592 static void ddf_prepare_update(struct supertype
*st
,
4593 struct metadata_update
*update
)
4595 /* This update arrived at managemon.
4596 * We are about to pass it to monitor.
4597 * If a malloc is needed, do it here.
4599 struct ddf_super
*ddf
= st
->sb
;
4600 __u32
*magic
= (__u32
*)update
->buf
;
4601 if (*magic
== DDF_VD_CONF_MAGIC
) {
4603 struct vd_config
*conf
= (struct vd_config
*) update
->buf
;
4604 if (posix_memalign(&update
->space
, 512,
4605 offsetof(struct vcl
, conf
)
4606 + ddf
->conf_rec_len
* 512) != 0) {
4607 update
->space
= NULL
;
4610 vcl
= update
->space
;
4611 vcl
->conf
.sec_elmnt_count
= conf
->sec_elmnt_count
;
4612 if (alloc_other_bvds(ddf
, vcl
) != 0) {
4613 free(update
->space
);
4614 update
->space
= NULL
;
4620 * Check if the array 'a' is degraded but not failed.
4621 * If it is, find as many spares as are available and needed and
4622 * arrange for their inclusion.
4623 * We only choose devices which are not already in the array,
4624 * and prefer those with a spare-assignment to this array.
4625 * otherwise we choose global spares - assuming always that
4626 * there is enough room.
4627 * For each spare that we assign, we return an 'mdinfo' which
4628 * describes the position for the device in the array.
4629 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
4630 * the new phys_refnum and lba_offset values.
4632 * Only worry about BVDs at the moment.
4634 static struct mdinfo
*ddf_activate_spare(struct active_array
*a
,
4635 struct metadata_update
**updates
)
4639 struct ddf_super
*ddf
= a
->container
->sb
;
4641 struct mdinfo
*rv
= NULL
;
4643 struct metadata_update
*mu
;
4647 struct vd_config
*vc
;
4650 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
4651 if ((d
->curr_state
& DS_FAULTY
) &&
4653 /* wait for Removal to happen */
4655 if (d
->state_fd
>= 0)
4659 dprintf("ddf_activate: working=%d (%d) level=%d\n", working
, a
->info
.array
.raid_disks
,
4660 a
->info
.array
.level
);
4661 if (working
== a
->info
.array
.raid_disks
)
4662 return NULL
; /* array not degraded */
4663 switch (a
->info
.array
.level
) {
4666 return NULL
; /* failed */
4670 if (working
< a
->info
.array
.raid_disks
- 1)
4671 return NULL
; /* failed */
4674 if (working
< a
->info
.array
.raid_disks
- 2)
4675 return NULL
; /* failed */
4677 default: /* concat or stripe */
4678 return NULL
; /* failed */
4681 /* For each slot, if it is not working, find a spare */
4683 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
4684 for (d
= a
->info
.devs
; d
; d
= d
->next
)
4685 if (d
->disk
.raid_disk
== i
)
4687 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
4688 if (d
&& (d
->state_fd
>= 0))
4691 /* OK, this device needs recovery. Find a spare */
4693 for ( ; dl
; dl
= dl
->next
) {
4694 unsigned long long esize
;
4695 unsigned long long pos
;
4698 int is_dedicated
= 0;
4701 /* If in this array, skip */
4702 for (d2
= a
->info
.devs
; d2
; d2
= d2
->next
)
4703 if (d2
->state_fd
>= 0 &&
4704 d2
->disk
.major
== dl
->major
&&
4705 d2
->disk
.minor
== dl
->minor
) {
4706 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
4711 if (ddf
->phys
->entries
[dl
->pdnum
].type
&
4712 __cpu_to_be16(DDF_Spare
)) {
4713 /* Check spare assign record */
4715 if (dl
->spare
->type
& DDF_spare_dedicated
) {
4716 /* check spare_ents for guid */
4718 j
< __be16_to_cpu(dl
->spare
->populated
);
4720 if (memcmp(dl
->spare
->spare_ents
[j
].guid
,
4721 ddf
->virt
->entries
[a
->info
.container_member
].guid
,
4728 } else if (ddf
->phys
->entries
[dl
->pdnum
].type
&
4729 __cpu_to_be16(DDF_Global_Spare
)) {
4731 } else if (!(ddf
->phys
->entries
[dl
->pdnum
].state
&
4732 __cpu_to_be16(DDF_Failed
))) {
4733 /* we can possibly use some of this */
4736 if ( ! (is_dedicated
||
4737 (is_global
&& global_ok
))) {
4738 dprintf("%x:%x not suitable: %d %d\n", dl
->major
, dl
->minor
,
4739 is_dedicated
, is_global
);
4743 /* We are allowed to use this device - is there space?
4744 * We need a->info.component_size sectors */
4745 ex
= get_extents(ddf
, dl
);
4747 dprintf("cannot get extents\n");
4754 esize
= ex
[j
].start
- pos
;
4755 if (esize
>= a
->info
.component_size
)
4757 pos
= ex
[j
].start
+ ex
[j
].size
;
4759 } while (ex
[j
-1].size
);
4762 if (esize
< a
->info
.component_size
) {
4763 dprintf("%x:%x has no room: %llu %llu\n",
4764 dl
->major
, dl
->minor
,
4765 esize
, a
->info
.component_size
);
4770 /* Cool, we have a device with some space at pos */
4771 di
= xcalloc(1, sizeof(*di
));
4772 di
->disk
.number
= i
;
4773 di
->disk
.raid_disk
= i
;
4774 di
->disk
.major
= dl
->major
;
4775 di
->disk
.minor
= dl
->minor
;
4777 di
->recovery_start
= 0;
4778 di
->data_offset
= pos
;
4779 di
->component_size
= a
->info
.component_size
;
4780 di
->container_member
= dl
->pdnum
;
4783 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
4788 if (!dl
&& ! global_ok
) {
4789 /* not enough dedicated spares, try global */
4797 /* No spares found */
4799 /* Now 'rv' has a list of devices to return.
4800 * Create a metadata_update record to update the
4801 * phys_refnum and lba_offset values
4803 mu
= xmalloc(sizeof(*mu
));
4804 if (posix_memalign(&mu
->space
, 512, sizeof(struct vcl
)) != 0) {
4808 mu
->buf
= xmalloc(ddf
->conf_rec_len
* 512);
4809 mu
->len
= ddf
->conf_rec_len
* 512;
4811 mu
->space_list
= NULL
;
4812 mu
->next
= *updates
;
4813 vc
= find_vdcr(ddf
, a
->info
.container_member
, di
->disk
.raid_disk
,
4815 memcpy(mu
->buf
, vc
, ddf
->conf_rec_len
* 512);
4817 vc
= (struct vd_config
*)mu
->buf
;
4818 for (di
= rv
; di
; di
= di
->next
) {
4819 vc
->phys_refnum
[di
->disk
.raid_disk
] =
4820 ddf
->phys
->entries
[dl
->pdnum
].refnum
;
4821 LBA_OFFSET(ddf
, vc
)[di
->disk
.raid_disk
]
4822 = __cpu_to_be64(di
->data_offset
);
4827 #endif /* MDASSEMBLE */
4829 static int ddf_level_to_layout(int level
)
4836 return ALGORITHM_LEFT_SYMMETRIC
;
4838 return ALGORITHM_ROTATING_N_CONTINUE
;
4846 static void default_geometry_ddf(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
4848 if (level
&& *level
== UnSet
)
4849 *level
= LEVEL_CONTAINER
;
4851 if (level
&& layout
&& *layout
== UnSet
)
4852 *layout
= ddf_level_to_layout(*level
);
4855 struct superswitch super_ddf
= {
4857 .examine_super
= examine_super_ddf
,
4858 .brief_examine_super
= brief_examine_super_ddf
,
4859 .brief_examine_subarrays
= brief_examine_subarrays_ddf
,
4860 .export_examine_super
= export_examine_super_ddf
,
4861 .detail_super
= detail_super_ddf
,
4862 .brief_detail_super
= brief_detail_super_ddf
,
4863 .validate_geometry
= validate_geometry_ddf
,
4864 .write_init_super
= write_init_super_ddf
,
4865 .add_to_super
= add_to_super_ddf
,
4866 .remove_from_super
= remove_from_super_ddf
,
4867 .load_container
= load_container_ddf
,
4868 .copy_metadata
= copy_metadata_ddf
,
4870 .match_home
= match_home_ddf
,
4871 .uuid_from_super
= uuid_from_super_ddf
,
4872 .getinfo_super
= getinfo_super_ddf
,
4873 .update_super
= update_super_ddf
,
4875 .avail_size
= avail_size_ddf
,
4877 .compare_super
= compare_super_ddf
,
4879 .load_super
= load_super_ddf
,
4880 .init_super
= init_super_ddf
,
4881 .store_super
= store_super_ddf
,
4882 .free_super
= free_super_ddf
,
4883 .match_metadata_desc
= match_metadata_desc_ddf
,
4884 .container_content
= container_content_ddf
,
4885 .default_geometry
= default_geometry_ddf
,
4886 .kill_subarray
= kill_subarray_ddf
,
4892 .open_new
= ddf_open_new
,
4893 .set_array_state
= ddf_set_array_state
,
4894 .set_disk
= ddf_set_disk
,
4895 .sync_metadata
= ddf_sync_metadata
,
4896 .process_update
= ddf_process_update
,
4897 .prepare_update
= ddf_prepare_update
,
4898 .activate_spare
= ddf_activate_spare
,