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 /* Default for safe_mode_delay. Same value as for IMSM.
52 static const int DDF_SAFE_MODE_DELAY
= 4000;
54 /* The DDF metadata handling.
55 * DDF metadata lives at the end of the device.
56 * The last 512 byte block provides an 'anchor' which is used to locate
57 * the rest of the metadata which usually lives immediately behind the anchor.
60 * - all multibyte numeric fields are bigendian.
61 * - all strings are space padded.
65 typedef struct __be16
{
68 #define be16_eq(x, y) ((x)._v16 == (y)._v16)
69 #define be16_and(x, y) ((x)._v16 & (y)._v16)
70 #define be16_or(x, y) ((x)._v16 | (y)._v16)
71 #define be16_clear(x, y) ((x)._v16 &= ~(y)._v16)
72 #define be16_set(x, y) ((x)._v16 |= (y)._v16)
74 typedef struct __be32
{
77 #define be32_eq(x, y) ((x)._v32 == (y)._v32)
79 typedef struct __be64
{
82 #define be64_eq(x, y) ((x)._v64 == (y)._v64)
84 #define be16_to_cpu(be) __be16_to_cpu((be)._v16)
85 static inline be16
cpu_to_be16(__u16 x
)
87 be16 be
= { ._v16
= __cpu_to_be16(x
) };
91 #define be32_to_cpu(be) __be32_to_cpu((be)._v32)
92 static inline be32
cpu_to_be32(__u32 x
)
94 be32 be
= { ._v32
= __cpu_to_be32(x
) };
98 #define be64_to_cpu(be) __be64_to_cpu((be)._v64)
99 static inline be64
cpu_to_be64(__u64 x
)
101 be64 be
= { ._v64
= __cpu_to_be64(x
) };
105 /* Primary Raid Level (PRL) */
106 #define DDF_RAID0 0x00
107 #define DDF_RAID1 0x01
108 #define DDF_RAID3 0x03
109 #define DDF_RAID4 0x04
110 #define DDF_RAID5 0x05
111 #define DDF_RAID1E 0x11
112 #define DDF_JBOD 0x0f
113 #define DDF_CONCAT 0x1f
114 #define DDF_RAID5E 0x15
115 #define DDF_RAID5EE 0x25
116 #define DDF_RAID6 0x06
118 /* Raid Level Qualifier (RLQ) */
119 #define DDF_RAID0_SIMPLE 0x00
120 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
121 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
122 #define DDF_RAID3_0 0x00 /* parity in first extent */
123 #define DDF_RAID3_N 0x01 /* parity in last extent */
124 #define DDF_RAID4_0 0x00 /* parity in first extent */
125 #define DDF_RAID4_N 0x01 /* parity in last extent */
126 /* these apply to raid5e and raid5ee as well */
127 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
128 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
129 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
130 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
132 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
133 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
135 /* Secondary RAID Level (SRL) */
136 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
137 #define DDF_2MIRRORED 0x01
138 #define DDF_2CONCAT 0x02
139 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
142 #define DDF_HEADER_MAGIC cpu_to_be32(0xDE11DE11)
143 #define DDF_CONTROLLER_MAGIC cpu_to_be32(0xAD111111)
144 #define DDF_PHYS_RECORDS_MAGIC cpu_to_be32(0x22222222)
145 #define DDF_PHYS_DATA_MAGIC cpu_to_be32(0x33333333)
146 #define DDF_VIRT_RECORDS_MAGIC cpu_to_be32(0xDDDDDDDD)
147 #define DDF_VD_CONF_MAGIC cpu_to_be32(0xEEEEEEEE)
148 #define DDF_SPARE_ASSIGN_MAGIC cpu_to_be32(0x55555555)
149 #define DDF_VU_CONF_MAGIC cpu_to_be32(0x88888888)
150 #define DDF_VENDOR_LOG_MAGIC cpu_to_be32(0x01dBEEF0)
151 #define DDF_BBM_LOG_MAGIC cpu_to_be32(0xABADB10C)
153 #define DDF_GUID_LEN 24
154 #define DDF_REVISION_0 "01.00.00"
155 #define DDF_REVISION_2 "01.02.00"
158 be32 magic
; /* DDF_HEADER_MAGIC */
160 char guid
[DDF_GUID_LEN
];
161 char revision
[8]; /* 01.02.00 */
162 be32 seq
; /* starts at '1' */
167 __u8 pad0
; /* 0xff */
168 __u8 pad1
[12]; /* 12 * 0xff */
169 /* 64 bytes so far */
170 __u8 header_ext
[32]; /* reserved: fill with 0xff */
174 __u8 pad2
[3]; /* 0xff */
175 be32 workspace_len
; /* sectors for vendor space -
176 * at least 32768(sectors) */
178 be16 max_pd_entries
; /* one of 15, 63, 255, 1023, 4095 */
179 be16 max_vd_entries
; /* 2^(4,6,8,10,12)-1 : i.e. as above */
180 be16 max_partitions
; /* i.e. max num of configuration
181 record entries per disk */
182 be16 config_record_len
; /* 1 +ROUNDUP(max_primary_element_entries
184 be16 max_primary_element_entries
; /* 16, 64, 256, 1024, or 4096 */
185 __u8 pad3
[54]; /* 0xff */
186 /* 192 bytes so far */
187 be32 controller_section_offset
;
188 be32 controller_section_length
;
189 be32 phys_section_offset
;
190 be32 phys_section_length
;
191 be32 virt_section_offset
;
192 be32 virt_section_length
;
193 be32 config_section_offset
;
194 be32 config_section_length
;
195 be32 data_section_offset
;
196 be32 data_section_length
;
197 be32 bbm_section_offset
;
198 be32 bbm_section_length
;
199 be32 diag_space_offset
;
200 be32 diag_space_length
;
203 /* 256 bytes so far */
204 __u8 pad4
[256]; /* 0xff */
208 #define DDF_HEADER_ANCHOR 0x00
209 #define DDF_HEADER_PRIMARY 0x01
210 #define DDF_HEADER_SECONDARY 0x02
212 /* The content of the 'controller section' - global scope */
213 struct ddf_controller_data
{
214 be32 magic
; /* DDF_CONTROLLER_MAGIC */
216 char guid
[DDF_GUID_LEN
];
217 struct controller_type
{
224 __u8 pad
[8]; /* 0xff */
225 __u8 vendor_data
[448];
228 /* The content of phys_section - global scope */
230 be32 magic
; /* DDF_PHYS_RECORDS_MAGIC */
235 struct phys_disk_entry
{
236 char guid
[DDF_GUID_LEN
];
240 be64 config_size
; /* DDF structures must be after here */
241 char path
[18]; /* another horrible structure really */
246 /* phys_disk_entry.type is a bitmap - bigendian remember */
247 #define DDF_Forced_PD_GUID 1
248 #define DDF_Active_in_VD 2
249 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
250 #define DDF_Spare 8 /* overrides Global_spare */
251 #define DDF_Foreign 16
252 #define DDF_Legacy 32 /* no DDF on this device */
254 #define DDF_Interface_mask 0xf00
255 #define DDF_Interface_SCSI 0x100
256 #define DDF_Interface_SAS 0x200
257 #define DDF_Interface_SATA 0x300
258 #define DDF_Interface_FC 0x400
260 /* phys_disk_entry.state is a bigendian bitmap */
262 #define DDF_Failed 2 /* overrides 1,4,8 */
263 #define DDF_Rebuilding 4
264 #define DDF_Transition 8
266 #define DDF_ReadErrors 32
267 #define DDF_Missing 64
269 /* The content of the virt_section global scope */
270 struct virtual_disk
{
271 be32 magic
; /* DDF_VIRT_RECORDS_MAGIC */
276 struct virtual_entry
{
277 char guid
[DDF_GUID_LEN
];
279 __u16 pad0
; /* 0xffff */
289 /* virtual_entry.type is a bitmap - bigendian */
291 #define DDF_Enforce_Groups 2
292 #define DDF_Unicode 4
293 #define DDF_Owner_Valid 8
295 /* virtual_entry.state is a bigendian bitmap */
296 #define DDF_state_mask 0x7
297 #define DDF_state_optimal 0x0
298 #define DDF_state_degraded 0x1
299 #define DDF_state_deleted 0x2
300 #define DDF_state_missing 0x3
301 #define DDF_state_failed 0x4
302 #define DDF_state_part_optimal 0x5
304 #define DDF_state_morphing 0x8
305 #define DDF_state_inconsistent 0x10
307 /* virtual_entry.init_state is a bigendian bitmap */
308 #define DDF_initstate_mask 0x03
309 #define DDF_init_not 0x00
310 #define DDF_init_quick 0x01 /* initialisation is progress.
311 * i.e. 'state_inconsistent' */
312 #define DDF_init_full 0x02
314 #define DDF_access_mask 0xc0
315 #define DDF_access_rw 0x00
316 #define DDF_access_ro 0x80
317 #define DDF_access_blocked 0xc0
319 /* The content of the config_section - local scope
320 * It has multiple records each config_record_len sectors
321 * They can be vd_config or spare_assign
325 be32 magic
; /* DDF_VD_CONF_MAGIC */
327 char guid
[DDF_GUID_LEN
];
331 be16 prim_elmnt_count
;
332 __u8 chunk_shift
; /* 0 == 512, 1==1024 etc */
335 __u8 sec_elmnt_count
;
338 be64 blocks
; /* blocks per component could be different
339 * on different component devices...(only
340 * for concat I hope) */
341 be64 array_blocks
; /* blocks in array */
349 __u8 v0
[32]; /* reserved- 0xff */
350 __u8 v1
[32]; /* reserved- 0xff */
351 __u8 v2
[16]; /* reserved- 0xff */
352 __u8 v3
[16]; /* reserved- 0xff */
354 be32 phys_refnum
[0]; /* refnum of each disk in sequence */
355 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
356 bvd are always the same size */
358 #define LBA_OFFSET(ddf, vd) ((be64 *) &(vd)->phys_refnum[(ddf)->mppe])
360 /* vd_config.cache_pol[7] is a bitmap */
361 #define DDF_cache_writeback 1 /* else writethrough */
362 #define DDF_cache_wadaptive 2 /* only applies if writeback */
363 #define DDF_cache_readahead 4
364 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
365 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
366 #define DDF_cache_wallowed 32 /* enable write caching */
367 #define DDF_cache_rallowed 64 /* enable read caching */
369 struct spare_assign
{
370 be32 magic
; /* DDF_SPARE_ASSIGN_MAGIC */
375 be16 populated
; /* SAEs used */
376 be16 max
; /* max SAEs */
378 struct spare_assign_entry
{
379 char guid
[DDF_GUID_LEN
];
380 be16 secondary_element
;
384 /* spare_assign.type is a bitmap */
385 #define DDF_spare_dedicated 0x1 /* else global */
386 #define DDF_spare_revertible 0x2 /* else committable */
387 #define DDF_spare_active 0x4 /* else not active */
388 #define DDF_spare_affinity 0x8 /* enclosure affinity */
390 /* The data_section contents - local scope */
392 be32 magic
; /* DDF_PHYS_DATA_MAGIC */
394 char guid
[DDF_GUID_LEN
];
395 be32 refnum
; /* crc of some magic drive data ... */
396 __u8 forced_ref
; /* set when above was not result of magic */
397 __u8 forced_guid
; /* set if guid was forced rather than magic */
402 /* bbm_section content */
403 struct bad_block_log
{
410 struct mapped_block
{
411 be64 defective_start
;
412 be32 replacement_start
;
418 /* Struct for internally holding ddf structures */
419 /* The DDF structure stored on each device is potentially
420 * quite different, as some data is global and some is local.
421 * The global data is:
424 * - Physical disk records
425 * - Virtual disk records
427 * - Configuration records
428 * - Physical Disk data section
429 * ( and Bad block and vendor which I don't care about yet).
431 * The local data is parsed into separate lists as it is read
432 * and reconstructed for writing. This means that we only need
433 * to make config changes once and they are automatically
434 * propagated to all devices.
435 * Note that the ddf_super has space of the conf and disk data
436 * for this disk and also for a list of all such data.
437 * The list is only used for the superblock that is being
438 * built in Create or Assemble to describe the whole array.
441 struct ddf_header anchor
, primary
, secondary
;
442 struct ddf_controller_data controller
;
443 struct ddf_header
*active
;
444 struct phys_disk
*phys
;
445 struct virtual_disk
*virt
;
448 unsigned int max_part
, mppe
, conf_rec_len
;
456 unsigned int vcnum
; /* index into ->virt */
457 struct vd_config
**other_bvds
;
458 __u64
*block_sizes
; /* NULL if all the same */
461 struct vd_config conf
;
462 } *conflist
, *currentconf
;
471 unsigned long long size
; /* sectors */
472 be64 primary_lba
; /* sectors */
473 be64 secondary_lba
; /* sectors */
474 be64 workspace_lba
; /* sectors */
475 int pdnum
; /* index in ->phys */
476 struct spare_assign
*spare
;
477 void *mdupdate
; /* hold metadata update */
479 /* These fields used by auto-layout */
480 int raiddisk
; /* slot to fill in autolayout */
484 struct disk_data disk
;
485 struct vcl
*vlist
[0]; /* max_part in size */
490 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
494 static int all_ff(const char *guid
);
495 static void pr_state(struct ddf_super
*ddf
, const char *msg
)
498 dprintf("%s/%s: ", __func__
, msg
);
499 for (i
= 0; i
< be16_to_cpu(ddf
->active
->max_vd_entries
); i
++) {
500 if (all_ff(ddf
->virt
->entries
[i
].guid
))
502 dprintf("%u(s=%02x i=%02x) ", i
,
503 ddf
->virt
->entries
[i
].state
,
504 ddf
->virt
->entries
[i
].init_state
);
509 static void pr_state(const struct ddf_super
*ddf
, const char *msg
) {}
512 static void _ddf_set_updates_pending(struct ddf_super
*ddf
, const char *func
)
514 ddf
->updates_pending
= 1;
515 ddf
->active
->seq
= cpu_to_be32((be32_to_cpu(ddf
->active
->seq
)+1));
519 #define ddf_set_updates_pending(x) _ddf_set_updates_pending((x), __func__)
521 static unsigned int get_pd_index_from_refnum(const struct vcl
*vc
,
522 be32 refnum
, unsigned int nmax
,
523 const struct vd_config
**bvd
,
526 static be32
calc_crc(void *buf
, int len
)
528 /* crcs are always at the same place as in the ddf_header */
529 struct ddf_header
*ddf
= buf
;
530 be32 oldcrc
= ddf
->crc
;
532 ddf
->crc
= cpu_to_be32(0xffffffff);
534 newcrc
= crc32(0, buf
, len
);
536 /* The crc is store (like everything) bigendian, so convert
537 * here for simplicity
539 return cpu_to_be32(newcrc
);
542 #define DDF_INVALID_LEVEL 0xff
543 #define DDF_NO_SECONDARY 0xff
544 static int err_bad_md_layout(const mdu_array_info_t
*array
)
546 pr_err("RAID%d layout %x with %d disks is unsupported for DDF\n",
547 array
->level
, array
->layout
, array
->raid_disks
);
551 static int layout_md2ddf(const mdu_array_info_t
*array
,
552 struct vd_config
*conf
)
554 be16 prim_elmnt_count
= cpu_to_be16(array
->raid_disks
);
555 __u8 prl
= DDF_INVALID_LEVEL
, rlq
= 0;
556 __u8 sec_elmnt_count
= 1;
557 __u8 srl
= DDF_NO_SECONDARY
;
559 switch (array
->level
) {
564 rlq
= DDF_RAID0_SIMPLE
;
568 switch (array
->raid_disks
) {
570 rlq
= DDF_RAID1_SIMPLE
;
573 rlq
= DDF_RAID1_MULTI
;
576 return err_bad_md_layout(array
);
581 if (array
->layout
!= 0)
582 return err_bad_md_layout(array
);
587 switch (array
->layout
) {
588 case ALGORITHM_LEFT_ASYMMETRIC
:
589 rlq
= DDF_RAID5_N_RESTART
;
591 case ALGORITHM_RIGHT_ASYMMETRIC
:
592 rlq
= DDF_RAID5_0_RESTART
;
594 case ALGORITHM_LEFT_SYMMETRIC
:
595 rlq
= DDF_RAID5_N_CONTINUE
;
597 case ALGORITHM_RIGHT_SYMMETRIC
:
598 /* not mentioned in standard */
600 return err_bad_md_layout(array
);
605 switch (array
->layout
) {
606 case ALGORITHM_ROTATING_N_RESTART
:
607 rlq
= DDF_RAID5_N_RESTART
;
609 case ALGORITHM_ROTATING_ZERO_RESTART
:
610 rlq
= DDF_RAID6_0_RESTART
;
612 case ALGORITHM_ROTATING_N_CONTINUE
:
613 rlq
= DDF_RAID5_N_CONTINUE
;
616 return err_bad_md_layout(array
);
621 if (array
->raid_disks
% 2 == 0 && array
->layout
== 0x102) {
622 rlq
= DDF_RAID1_SIMPLE
;
623 prim_elmnt_count
= cpu_to_be16(2);
624 sec_elmnt_count
= array
->raid_disks
/ 2;
625 } else if (array
->raid_disks
% 3 == 0
626 && array
->layout
== 0x103) {
627 rlq
= DDF_RAID1_MULTI
;
628 prim_elmnt_count
= cpu_to_be16(3);
629 sec_elmnt_count
= array
->raid_disks
/ 3;
631 return err_bad_md_layout(array
);
636 return err_bad_md_layout(array
);
639 conf
->prim_elmnt_count
= prim_elmnt_count
;
642 conf
->sec_elmnt_count
= sec_elmnt_count
;
646 static int err_bad_ddf_layout(const struct vd_config
*conf
)
648 pr_err("DDF RAID %u qualifier %u with %u disks is unsupported\n",
649 conf
->prl
, conf
->rlq
, be16_to_cpu(conf
->prim_elmnt_count
));
653 static int layout_ddf2md(const struct vd_config
*conf
,
654 mdu_array_info_t
*array
)
656 int level
= LEVEL_UNSUPPORTED
;
658 int raiddisks
= be16_to_cpu(conf
->prim_elmnt_count
);
660 if (conf
->sec_elmnt_count
> 1) {
661 /* see also check_secondary() */
662 if (conf
->prl
!= DDF_RAID1
||
663 (conf
->srl
!= DDF_2STRIPED
&& conf
->srl
!= DDF_2SPANNED
)) {
664 pr_err("Unsupported secondary RAID level %u/%u\n",
665 conf
->prl
, conf
->srl
);
668 if (raiddisks
== 2 && conf
->rlq
== DDF_RAID1_SIMPLE
)
670 else if (raiddisks
== 3 && conf
->rlq
== DDF_RAID1_MULTI
)
673 return err_bad_ddf_layout(conf
);
674 raiddisks
*= conf
->sec_elmnt_count
;
681 level
= LEVEL_LINEAR
;
684 if (conf
->rlq
!= DDF_RAID0_SIMPLE
)
685 return err_bad_ddf_layout(conf
);
689 if (!((conf
->rlq
== DDF_RAID1_SIMPLE
&& raiddisks
== 2) ||
690 (conf
->rlq
== DDF_RAID1_MULTI
&& raiddisks
== 3)))
691 return err_bad_ddf_layout(conf
);
695 if (conf
->rlq
!= DDF_RAID4_N
)
696 return err_bad_ddf_layout(conf
);
701 case DDF_RAID5_N_RESTART
:
702 layout
= ALGORITHM_LEFT_ASYMMETRIC
;
704 case DDF_RAID5_0_RESTART
:
705 layout
= ALGORITHM_RIGHT_ASYMMETRIC
;
707 case DDF_RAID5_N_CONTINUE
:
708 layout
= ALGORITHM_LEFT_SYMMETRIC
;
711 return err_bad_ddf_layout(conf
);
717 case DDF_RAID5_N_RESTART
:
718 layout
= ALGORITHM_ROTATING_N_RESTART
;
720 case DDF_RAID6_0_RESTART
:
721 layout
= ALGORITHM_ROTATING_ZERO_RESTART
;
723 case DDF_RAID5_N_CONTINUE
:
724 layout
= ALGORITHM_ROTATING_N_CONTINUE
;
727 return err_bad_ddf_layout(conf
);
732 return err_bad_ddf_layout(conf
);
736 array
->level
= level
;
737 array
->layout
= layout
;
738 array
->raid_disks
= raiddisks
;
742 static int load_ddf_header(int fd
, unsigned long long lba
,
743 unsigned long long size
,
745 struct ddf_header
*hdr
, struct ddf_header
*anchor
)
747 /* read a ddf header (primary or secondary) from fd/lba
748 * and check that it is consistent with anchor
750 * magic, crc, guid, rev, and LBA's header_type, and
751 * everything after header_type must be the same
756 if (lseek64(fd
, lba
<<9, 0) < 0)
759 if (read(fd
, hdr
, 512) != 512)
762 if (!be32_eq(hdr
->magic
, DDF_HEADER_MAGIC
)) {
763 pr_err("%s: bad header magic\n", __func__
);
766 if (!be32_eq(calc_crc(hdr
, 512), hdr
->crc
)) {
767 pr_err("%s: bad CRC\n", __func__
);
770 if (memcmp(anchor
->guid
, hdr
->guid
, DDF_GUID_LEN
) != 0 ||
771 memcmp(anchor
->revision
, hdr
->revision
, 8) != 0 ||
772 !be64_eq(anchor
->primary_lba
, hdr
->primary_lba
) ||
773 !be64_eq(anchor
->secondary_lba
, hdr
->secondary_lba
) ||
775 memcmp(anchor
->pad2
, hdr
->pad2
, 512 -
776 offsetof(struct ddf_header
, pad2
)) != 0) {
777 pr_err("%s: header mismatch\n", __func__
);
781 /* Looks good enough to me... */
785 static void *load_section(int fd
, struct ddf_super
*super
, void *buf
,
786 be32 offset_be
, be32 len_be
, int check
)
788 unsigned long long offset
= be32_to_cpu(offset_be
);
789 unsigned long long len
= be32_to_cpu(len_be
);
790 int dofree
= (buf
== NULL
);
793 if (len
!= 2 && len
!= 8 && len
!= 32
794 && len
!= 128 && len
!= 512)
799 if (!buf
&& posix_memalign(&buf
, 512, len
<<9) != 0)
805 if (super
->active
->type
== 1)
806 offset
+= be64_to_cpu(super
->active
->primary_lba
);
808 offset
+= be64_to_cpu(super
->active
->secondary_lba
);
810 if ((unsigned long long)lseek64(fd
, offset
<<9, 0) != (offset
<<9)) {
815 if ((unsigned long long)read(fd
, buf
, len
<<9) != (len
<<9)) {
823 static int load_ddf_headers(int fd
, struct ddf_super
*super
, char *devname
)
825 unsigned long long dsize
;
827 get_dev_size(fd
, NULL
, &dsize
);
829 if (lseek64(fd
, dsize
-512, 0) < 0) {
831 pr_err("Cannot seek to anchor block on %s: %s\n",
832 devname
, strerror(errno
));
835 if (read(fd
, &super
->anchor
, 512) != 512) {
837 pr_err("Cannot read anchor block on %s: %s\n",
838 devname
, strerror(errno
));
841 if (!be32_eq(super
->anchor
.magic
, DDF_HEADER_MAGIC
)) {
843 pr_err("no DDF anchor found on %s\n",
847 if (!be32_eq(calc_crc(&super
->anchor
, 512), super
->anchor
.crc
)) {
849 pr_err("bad CRC on anchor on %s\n",
853 if (memcmp(super
->anchor
.revision
, DDF_REVISION_0
, 8) != 0 &&
854 memcmp(super
->anchor
.revision
, DDF_REVISION_2
, 8) != 0) {
856 pr_err("can only support super revision"
857 " %.8s and earlier, not %.8s on %s\n",
858 DDF_REVISION_2
, super
->anchor
.revision
,devname
);
861 super
->active
= NULL
;
862 if (load_ddf_header(fd
, be64_to_cpu(super
->anchor
.primary_lba
),
864 &super
->primary
, &super
->anchor
) == 0) {
866 pr_err("Failed to load primary DDF header "
869 super
->active
= &super
->primary
;
871 if (load_ddf_header(fd
, be64_to_cpu(super
->anchor
.secondary_lba
),
873 &super
->secondary
, &super
->anchor
)) {
874 if (super
->active
== NULL
875 || (be32_to_cpu(super
->primary
.seq
)
876 < be32_to_cpu(super
->secondary
.seq
) &&
877 !super
->secondary
.openflag
)
878 || (be32_to_cpu(super
->primary
.seq
)
879 == be32_to_cpu(super
->secondary
.seq
) &&
880 super
->primary
.openflag
&& !super
->secondary
.openflag
)
882 super
->active
= &super
->secondary
;
884 pr_err("Failed to load secondary DDF header on %s\n",
886 if (super
->active
== NULL
)
891 static int load_ddf_global(int fd
, struct ddf_super
*super
, char *devname
)
894 ok
= load_section(fd
, super
, &super
->controller
,
895 super
->active
->controller_section_offset
,
896 super
->active
->controller_section_length
,
898 super
->phys
= load_section(fd
, super
, NULL
,
899 super
->active
->phys_section_offset
,
900 super
->active
->phys_section_length
,
902 super
->pdsize
= be32_to_cpu(super
->active
->phys_section_length
) * 512;
904 super
->virt
= load_section(fd
, super
, NULL
,
905 super
->active
->virt_section_offset
,
906 super
->active
->virt_section_length
,
908 super
->vdsize
= be32_to_cpu(super
->active
->virt_section_length
) * 512;
918 super
->conflist
= NULL
;
921 super
->max_part
= be16_to_cpu(super
->active
->max_partitions
);
922 super
->mppe
= be16_to_cpu(super
->active
->max_primary_element_entries
);
923 super
->conf_rec_len
= be16_to_cpu(super
->active
->config_record_len
);
927 #define DDF_UNUSED_BVD 0xff
928 static int alloc_other_bvds(const struct ddf_super
*ddf
, struct vcl
*vcl
)
930 unsigned int n_vds
= vcl
->conf
.sec_elmnt_count
- 1;
931 unsigned int i
, vdsize
;
934 vcl
->other_bvds
= NULL
;
937 vdsize
= ddf
->conf_rec_len
* 512;
938 if (posix_memalign(&p
, 512, n_vds
*
939 (vdsize
+ sizeof(struct vd_config
*))) != 0)
941 vcl
->other_bvds
= (struct vd_config
**) (p
+ n_vds
* vdsize
);
942 for (i
= 0; i
< n_vds
; i
++) {
943 vcl
->other_bvds
[i
] = p
+ i
* vdsize
;
944 memset(vcl
->other_bvds
[i
], 0, vdsize
);
945 vcl
->other_bvds
[i
]->sec_elmnt_seq
= DDF_UNUSED_BVD
;
950 static void add_other_bvd(struct vcl
*vcl
, struct vd_config
*vd
,
954 for (i
= 0; i
< vcl
->conf
.sec_elmnt_count
-1; i
++)
955 if (vcl
->other_bvds
[i
]->sec_elmnt_seq
== vd
->sec_elmnt_seq
)
958 if (i
< vcl
->conf
.sec_elmnt_count
-1) {
959 if (be32_to_cpu(vd
->seqnum
) <=
960 be32_to_cpu(vcl
->other_bvds
[i
]->seqnum
))
963 for (i
= 0; i
< vcl
->conf
.sec_elmnt_count
-1; i
++)
964 if (vcl
->other_bvds
[i
]->sec_elmnt_seq
== DDF_UNUSED_BVD
)
966 if (i
== vcl
->conf
.sec_elmnt_count
-1) {
967 pr_err("no space for sec level config %u, count is %u\n",
968 vd
->sec_elmnt_seq
, vcl
->conf
.sec_elmnt_count
);
972 memcpy(vcl
->other_bvds
[i
], vd
, len
);
975 static int load_ddf_local(int fd
, struct ddf_super
*super
,
976 char *devname
, int keep
)
982 unsigned int confsec
;
984 unsigned int max_virt_disks
= be16_to_cpu
985 (super
->active
->max_vd_entries
);
986 unsigned long long dsize
;
988 /* First the local disk info */
989 if (posix_memalign((void**)&dl
, 512,
991 (super
->max_part
) * sizeof(dl
->vlist
[0])) != 0) {
992 pr_err("%s could not allocate disk info buffer\n",
997 load_section(fd
, super
, &dl
->disk
,
998 super
->active
->data_section_offset
,
999 super
->active
->data_section_length
,
1001 dl
->devname
= devname
? xstrdup(devname
) : NULL
;
1004 dl
->major
= major(stb
.st_rdev
);
1005 dl
->minor
= minor(stb
.st_rdev
);
1006 dl
->next
= super
->dlist
;
1007 dl
->fd
= keep
? fd
: -1;
1010 if (get_dev_size(fd
, devname
, &dsize
))
1011 dl
->size
= dsize
>> 9;
1012 /* If the disks have different sizes, the LBAs will differ
1013 * between phys disks.
1014 * At this point here, the values in super->active must be valid
1015 * for this phys disk. */
1016 dl
->primary_lba
= super
->active
->primary_lba
;
1017 dl
->secondary_lba
= super
->active
->secondary_lba
;
1018 dl
->workspace_lba
= super
->active
->workspace_lba
;
1020 for (i
= 0 ; i
< super
->max_part
; i
++)
1021 dl
->vlist
[i
] = NULL
;
1024 for (i
= 0; i
< be16_to_cpu(super
->active
->max_pd_entries
); i
++)
1025 if (memcmp(super
->phys
->entries
[i
].guid
,
1026 dl
->disk
.guid
, DDF_GUID_LEN
) == 0)
1029 /* Now the config list. */
1030 /* 'conf' is an array of config entries, some of which are
1031 * probably invalid. Those which are good need to be copied into
1035 conf
= load_section(fd
, super
, super
->conf
,
1036 super
->active
->config_section_offset
,
1037 super
->active
->config_section_length
,
1042 confsec
< be32_to_cpu(super
->active
->config_section_length
);
1043 confsec
+= super
->conf_rec_len
) {
1044 struct vd_config
*vd
=
1045 (struct vd_config
*)((char*)conf
+ confsec
*512);
1048 if (be32_eq(vd
->magic
, DDF_SPARE_ASSIGN_MAGIC
)) {
1051 if (posix_memalign((void**)&dl
->spare
, 512,
1052 super
->conf_rec_len
*512) != 0) {
1053 pr_err("%s could not allocate spare info buf\n",
1058 memcpy(dl
->spare
, vd
, super
->conf_rec_len
*512);
1061 if (!be32_eq(vd
->magic
, DDF_VD_CONF_MAGIC
))
1063 for (vcl
= super
->conflist
; vcl
; vcl
= vcl
->next
) {
1064 if (memcmp(vcl
->conf
.guid
,
1065 vd
->guid
, DDF_GUID_LEN
) == 0)
1070 dl
->vlist
[vnum
++] = vcl
;
1071 if (vcl
->other_bvds
!= NULL
&&
1072 vcl
->conf
.sec_elmnt_seq
!= vd
->sec_elmnt_seq
) {
1073 add_other_bvd(vcl
, vd
, super
->conf_rec_len
*512);
1076 if (be32_to_cpu(vd
->seqnum
) <=
1077 be32_to_cpu(vcl
->conf
.seqnum
))
1080 if (posix_memalign((void**)&vcl
, 512,
1081 (super
->conf_rec_len
*512 +
1082 offsetof(struct vcl
, conf
))) != 0) {
1083 pr_err("%s could not allocate vcl buf\n",
1087 vcl
->next
= super
->conflist
;
1088 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
1089 vcl
->conf
.sec_elmnt_count
= vd
->sec_elmnt_count
;
1090 if (alloc_other_bvds(super
, vcl
) != 0) {
1091 pr_err("%s could not allocate other bvds\n",
1096 super
->conflist
= vcl
;
1097 dl
->vlist
[vnum
++] = vcl
;
1099 memcpy(&vcl
->conf
, vd
, super
->conf_rec_len
*512);
1100 for (i
=0; i
< max_virt_disks
; i
++)
1101 if (memcmp(super
->virt
->entries
[i
].guid
,
1102 vcl
->conf
.guid
, DDF_GUID_LEN
)==0)
1104 if (i
< max_virt_disks
)
1112 static int load_super_ddf_all(struct supertype
*st
, int fd
,
1113 void **sbp
, char *devname
);
1116 static void free_super_ddf(struct supertype
*st
);
1118 static int load_super_ddf(struct supertype
*st
, int fd
,
1121 unsigned long long dsize
;
1122 struct ddf_super
*super
;
1125 if (get_dev_size(fd
, devname
, &dsize
) == 0)
1128 if (!st
->ignore_hw_compat
&& test_partition(fd
))
1129 /* DDF is not allowed on partitions */
1132 /* 32M is a lower bound */
1133 if (dsize
<= 32*1024*1024) {
1135 pr_err("%s is too small for ddf: "
1136 "size is %llu sectors.\n",
1142 pr_err("%s is an odd size for ddf: "
1143 "size is %llu bytes.\n",
1150 if (posix_memalign((void**)&super
, 512, sizeof(*super
))!= 0) {
1151 pr_err("malloc of %zu failed.\n",
1155 memset(super
, 0, sizeof(*super
));
1157 rv
= load_ddf_headers(fd
, super
, devname
);
1163 /* Have valid headers and have chosen the best. Let's read in the rest*/
1165 rv
= load_ddf_global(fd
, super
, devname
);
1169 pr_err("Failed to load all information "
1170 "sections on %s\n", devname
);
1175 rv
= load_ddf_local(fd
, super
, devname
, 0);
1179 pr_err("Failed to load all information "
1180 "sections on %s\n", devname
);
1185 /* Should possibly check the sections .... */
1188 if (st
->ss
== NULL
) {
1189 st
->ss
= &super_ddf
;
1190 st
->minor_version
= 0;
1197 static void free_super_ddf(struct supertype
*st
)
1199 struct ddf_super
*ddf
= st
->sb
;
1205 while (ddf
->conflist
) {
1206 struct vcl
*v
= ddf
->conflist
;
1207 ddf
->conflist
= v
->next
;
1209 free(v
->block_sizes
);
1212 v->other_bvds[0] points to beginning of buffer,
1213 see alloc_other_bvds()
1215 free(v
->other_bvds
[0]);
1218 while (ddf
->dlist
) {
1219 struct dl
*d
= ddf
->dlist
;
1220 ddf
->dlist
= d
->next
;
1227 while (ddf
->add_list
) {
1228 struct dl
*d
= ddf
->add_list
;
1229 ddf
->add_list
= d
->next
;
1240 static struct supertype
*match_metadata_desc_ddf(char *arg
)
1242 /* 'ddf' only support containers */
1243 struct supertype
*st
;
1244 if (strcmp(arg
, "ddf") != 0 &&
1245 strcmp(arg
, "default") != 0
1249 st
= xcalloc(1, sizeof(*st
));
1250 st
->ss
= &super_ddf
;
1252 st
->minor_version
= 0;
1259 static mapping_t ddf_state
[] = {
1265 { "Partially Optimal", 5},
1271 static mapping_t ddf_init_state
[] = {
1272 { "Not Initialised", 0},
1273 { "QuickInit in Progress", 1},
1274 { "Fully Initialised", 2},
1278 static mapping_t ddf_access
[] = {
1282 { "Blocked (no access)", 3},
1286 static mapping_t ddf_level
[] = {
1287 { "RAID0", DDF_RAID0
},
1288 { "RAID1", DDF_RAID1
},
1289 { "RAID3", DDF_RAID3
},
1290 { "RAID4", DDF_RAID4
},
1291 { "RAID5", DDF_RAID5
},
1292 { "RAID1E",DDF_RAID1E
},
1293 { "JBOD", DDF_JBOD
},
1294 { "CONCAT",DDF_CONCAT
},
1295 { "RAID5E",DDF_RAID5E
},
1296 { "RAID5EE",DDF_RAID5EE
},
1297 { "RAID6", DDF_RAID6
},
1300 static mapping_t ddf_sec_level
[] = {
1301 { "Striped", DDF_2STRIPED
},
1302 { "Mirrored", DDF_2MIRRORED
},
1303 { "Concat", DDF_2CONCAT
},
1304 { "Spanned", DDF_2SPANNED
},
1309 static int all_ff(const char *guid
)
1312 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
1313 if (guid
[i
] != (char)0xff)
1318 static const char *guid_str(const char *guid
)
1320 static char buf
[DDF_GUID_LEN
*2+1];
1323 for (i
= 0; i
< DDF_GUID_LEN
; i
++) {
1324 unsigned char c
= guid
[i
];
1325 if (c
>= 32 && c
< 127)
1326 p
+= sprintf(p
, "%c", c
);
1328 p
+= sprintf(p
, "%02x", c
);
1331 return (const char *) buf
;
1335 static void print_guid(char *guid
, int tstamp
)
1337 /* A GUIDs are part (or all) ASCII and part binary.
1338 * They tend to be space padded.
1339 * We print the GUID in HEX, then in parentheses add
1340 * any initial ASCII sequence, and a possible
1341 * time stamp from bytes 16-19
1343 int l
= DDF_GUID_LEN
;
1346 for (i
=0 ; i
<DDF_GUID_LEN
; i
++) {
1347 if ((i
&3)==0 && i
!= 0) printf(":");
1348 printf("%02X", guid
[i
]&255);
1352 while (l
&& guid
[l
-1] == ' ')
1354 for (i
=0 ; i
<l
; i
++) {
1355 if (guid
[i
] >= 0x20 && guid
[i
] < 0x7f)
1356 fputc(guid
[i
], stdout
);
1361 time_t then
= __be32_to_cpu(*(__u32
*)(guid
+16)) + DECADE
;
1364 tm
= localtime(&then
);
1365 strftime(tbuf
, 100, " %D %T",tm
);
1366 fputs(tbuf
, stdout
);
1371 static void examine_vd(int n
, struct ddf_super
*sb
, char *guid
)
1373 int crl
= sb
->conf_rec_len
;
1376 for (vcl
= sb
->conflist
; vcl
; vcl
= vcl
->next
) {
1378 struct vd_config
*vc
= &vcl
->conf
;
1380 if (!be32_eq(calc_crc(vc
, crl
*512), vc
->crc
))
1382 if (memcmp(vc
->guid
, guid
, DDF_GUID_LEN
) != 0)
1385 /* Ok, we know about this VD, let's give more details */
1386 printf(" Raid Devices[%d] : %d (", n
,
1387 be16_to_cpu(vc
->prim_elmnt_count
));
1388 for (i
= 0; i
< be16_to_cpu(vc
->prim_elmnt_count
); i
++) {
1390 int cnt
= be16_to_cpu(sb
->phys
->used_pdes
);
1391 for (j
=0; j
<cnt
; j
++)
1392 if (be32_eq(vc
->phys_refnum
[i
],
1393 sb
->phys
->entries
[j
].refnum
))
1402 if (vc
->chunk_shift
!= 255)
1403 printf(" Chunk Size[%d] : %d sectors\n", n
,
1404 1 << vc
->chunk_shift
);
1405 printf(" Raid Level[%d] : %s\n", n
,
1406 map_num(ddf_level
, vc
->prl
)?:"-unknown-");
1407 if (vc
->sec_elmnt_count
!= 1) {
1408 printf(" Secondary Position[%d] : %d of %d\n", n
,
1409 vc
->sec_elmnt_seq
, vc
->sec_elmnt_count
);
1410 printf(" Secondary Level[%d] : %s\n", n
,
1411 map_num(ddf_sec_level
, vc
->srl
) ?: "-unknown-");
1413 printf(" Device Size[%d] : %llu\n", n
,
1414 be64_to_cpu(vc
->blocks
)/2);
1415 printf(" Array Size[%d] : %llu\n", n
,
1416 be64_to_cpu(vc
->array_blocks
)/2);
1420 static void examine_vds(struct ddf_super
*sb
)
1422 int cnt
= be16_to_cpu(sb
->virt
->populated_vdes
);
1424 printf(" Virtual Disks : %d\n", cnt
);
1426 for (i
= 0; i
< be16_to_cpu(sb
->virt
->max_vdes
); i
++) {
1427 struct virtual_entry
*ve
= &sb
->virt
->entries
[i
];
1428 if (all_ff(ve
->guid
))
1431 printf(" VD GUID[%d] : ", i
); print_guid(ve
->guid
, 1);
1433 printf(" unit[%d] : %d\n", i
, be16_to_cpu(ve
->unit
));
1434 printf(" state[%d] : %s, %s%s\n", i
,
1435 map_num(ddf_state
, ve
->state
& 7),
1436 (ve
->state
& DDF_state_morphing
) ? "Morphing, ": "",
1437 (ve
->state
& DDF_state_inconsistent
)? "Not Consistent" : "Consistent");
1438 printf(" init state[%d] : %s\n", i
,
1439 map_num(ddf_init_state
, ve
->init_state
&DDF_initstate_mask
));
1440 printf(" access[%d] : %s\n", i
,
1441 map_num(ddf_access
, (ve
->init_state
& DDF_access_mask
) >> 6));
1442 printf(" Name[%d] : %.16s\n", i
, ve
->name
);
1443 examine_vd(i
, sb
, ve
->guid
);
1445 if (cnt
) printf("\n");
1448 static void examine_pds(struct ddf_super
*sb
)
1450 int cnt
= be16_to_cpu(sb
->phys
->used_pdes
);
1453 printf(" Physical Disks : %d\n", cnt
);
1454 printf(" Number RefNo Size Device Type/State\n");
1456 for (i
=0 ; i
<cnt
; i
++) {
1457 struct phys_disk_entry
*pd
= &sb
->phys
->entries
[i
];
1458 int type
= be16_to_cpu(pd
->type
);
1459 int state
= be16_to_cpu(pd
->state
);
1461 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1463 printf(" %3d %08x ", i
,
1464 be32_to_cpu(pd
->refnum
));
1466 be64_to_cpu(pd
->config_size
)>>1);
1467 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1468 if (be32_eq(dl
->disk
.refnum
, pd
->refnum
)) {
1469 char *dv
= map_dev(dl
->major
, dl
->minor
, 0);
1471 printf("%-15s", dv
);
1478 printf(" %s%s%s%s%s",
1479 (type
&2) ? "active":"",
1480 (type
&4) ? "Global-Spare":"",
1481 (type
&8) ? "spare" : "",
1482 (type
&16)? ", foreign" : "",
1483 (type
&32)? "pass-through" : "");
1484 if (state
& DDF_Failed
)
1485 /* This over-rides these three */
1486 state
&= ~(DDF_Online
|DDF_Rebuilding
|DDF_Transition
);
1487 printf("/%s%s%s%s%s%s%s",
1488 (state
&1)? "Online": "Offline",
1489 (state
&2)? ", Failed": "",
1490 (state
&4)? ", Rebuilding": "",
1491 (state
&8)? ", in-transition": "",
1492 (state
&16)? ", SMART-errors": "",
1493 (state
&32)? ", Unrecovered-Read-Errors": "",
1494 (state
&64)? ", Missing" : "");
1499 static void examine_super_ddf(struct supertype
*st
, char *homehost
)
1501 struct ddf_super
*sb
= st
->sb
;
1503 printf(" Magic : %08x\n", be32_to_cpu(sb
->anchor
.magic
));
1504 printf(" Version : %.8s\n", sb
->anchor
.revision
);
1505 printf("Controller GUID : "); print_guid(sb
->controller
.guid
, 0);
1507 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1509 printf(" Seq : %08x\n", be32_to_cpu(sb
->active
->seq
));
1510 printf(" Redundant hdr : %s\n", be32_eq(sb
->secondary
.magic
,
1517 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1519 static void uuid_from_ddf_guid(const char *guid
, int uuid
[4]);
1520 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4]);
1522 static unsigned int get_vd_num_of_subarray(struct supertype
*st
)
1525 * Figure out the VD number for this supertype.
1526 * Returns DDF_CONTAINER for the container itself,
1527 * and DDF_NOTFOUND on error.
1529 struct ddf_super
*ddf
= st
->sb
;
1534 if (*st
->container_devnm
== '\0')
1535 return DDF_CONTAINER
;
1537 sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
1538 if (!sra
|| sra
->array
.major_version
!= -1 ||
1539 sra
->array
.minor_version
!= -2 ||
1540 !is_subarray(sra
->text_version
))
1541 return DDF_NOTFOUND
;
1543 sub
= strchr(sra
->text_version
+ 1, '/');
1545 vcnum
= strtoul(sub
+ 1, &end
, 10);
1546 if (sub
== NULL
|| *sub
== '\0' || *end
!= '\0' ||
1547 vcnum
>= be16_to_cpu(ddf
->active
->max_vd_entries
))
1548 return DDF_NOTFOUND
;
1553 static void brief_examine_super_ddf(struct supertype
*st
, int verbose
)
1555 /* We just write a generic DDF ARRAY entry
1559 getinfo_super_ddf(st
, &info
, NULL
);
1560 fname_from_uuid(st
, &info
, nbuf
, ':');
1562 printf("ARRAY metadata=ddf UUID=%s\n", nbuf
+ 5);
1565 static void brief_examine_subarrays_ddf(struct supertype
*st
, int verbose
)
1567 /* We just write a generic DDF ARRAY entry
1569 struct ddf_super
*ddf
= st
->sb
;
1573 getinfo_super_ddf(st
, &info
, NULL
);
1574 fname_from_uuid(st
, &info
, nbuf
, ':');
1576 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
1577 struct virtual_entry
*ve
= &ddf
->virt
->entries
[i
];
1580 if (all_ff(ve
->guid
))
1582 memcpy(vcl
.conf
.guid
, ve
->guid
, DDF_GUID_LEN
);
1583 ddf
->currentconf
=&vcl
;
1584 uuid_from_super_ddf(st
, info
.uuid
);
1585 fname_from_uuid(st
, &info
, nbuf1
, ':');
1586 printf("ARRAY container=%s member=%d UUID=%s\n",
1587 nbuf
+5, i
, nbuf1
+5);
1591 static void export_examine_super_ddf(struct supertype
*st
)
1595 getinfo_super_ddf(st
, &info
, NULL
);
1596 fname_from_uuid(st
, &info
, nbuf
, ':');
1597 printf("MD_METADATA=ddf\n");
1598 printf("MD_LEVEL=container\n");
1599 printf("MD_UUID=%s\n", nbuf
+5);
1600 printf("MD_DEVICES=%u\n",
1601 be16_to_cpu(((struct ddf_super
*)st
->sb
)->phys
->used_pdes
));
1604 static int copy_metadata_ddf(struct supertype
*st
, int from
, int to
)
1607 unsigned long long dsize
, offset
;
1609 struct ddf_header
*ddf
;
1612 /* The meta consists of an anchor, a primary, and a secondary.
1613 * This all lives at the end of the device.
1614 * So it is easiest to find the earliest of primary and
1615 * secondary, and copy everything from there.
1617 * Anchor is 512 from end It contains primary_lba and secondary_lba
1618 * we choose one of those
1621 if (posix_memalign(&buf
, 4096, 4096) != 0)
1624 if (!get_dev_size(from
, NULL
, &dsize
))
1627 if (lseek64(from
, dsize
-512, 0) < 0)
1629 if (read(from
, buf
, 512) != 512)
1632 if (!be32_eq(ddf
->magic
, DDF_HEADER_MAGIC
) ||
1633 !be32_eq(calc_crc(ddf
, 512), ddf
->crc
) ||
1634 (memcmp(ddf
->revision
, DDF_REVISION_0
, 8) != 0 &&
1635 memcmp(ddf
->revision
, DDF_REVISION_2
, 8) != 0))
1638 offset
= dsize
- 512;
1639 if ((be64_to_cpu(ddf
->primary_lba
) << 9) < offset
)
1640 offset
= be64_to_cpu(ddf
->primary_lba
) << 9;
1641 if ((be64_to_cpu(ddf
->secondary_lba
) << 9) < offset
)
1642 offset
= be64_to_cpu(ddf
->secondary_lba
) << 9;
1644 bytes
= dsize
- offset
;
1646 if (lseek64(from
, offset
, 0) < 0 ||
1647 lseek64(to
, offset
, 0) < 0)
1649 while (written
< bytes
) {
1650 int n
= bytes
- written
;
1653 if (read(from
, buf
, n
) != n
)
1655 if (write(to
, buf
, n
) != n
)
1666 static void detail_super_ddf(struct supertype
*st
, char *homehost
)
1669 * Could print DDF GUID
1670 * Need to find which array
1671 * If whole, briefly list all arrays
1676 static void brief_detail_super_ddf(struct supertype
*st
)
1680 struct ddf_super
*ddf
= st
->sb
;
1681 unsigned int vcnum
= get_vd_num_of_subarray(st
);
1682 if (vcnum
== DDF_CONTAINER
)
1683 uuid_from_super_ddf(st
, info
.uuid
);
1684 else if (vcnum
== DDF_NOTFOUND
)
1687 uuid_from_ddf_guid(ddf
->virt
->entries
[vcnum
].guid
, info
.uuid
);
1688 fname_from_uuid(st
, &info
, nbuf
,':');
1689 printf(" UUID=%s", nbuf
+ 5);
1693 static int match_home_ddf(struct supertype
*st
, char *homehost
)
1695 /* It matches 'this' host if the controller is a
1696 * Linux-MD controller with vendor_data matching
1699 struct ddf_super
*ddf
= st
->sb
;
1704 len
= strlen(homehost
);
1706 return (memcmp(ddf
->controller
.guid
, T10
, 8) == 0 &&
1707 len
< sizeof(ddf
->controller
.vendor_data
) &&
1708 memcmp(ddf
->controller
.vendor_data
, homehost
,len
) == 0 &&
1709 ddf
->controller
.vendor_data
[len
] == 0);
1713 static int find_index_in_bvd(const struct ddf_super
*ddf
,
1714 const struct vd_config
*conf
, unsigned int n
,
1715 unsigned int *n_bvd
)
1718 * Find the index of the n-th valid physical disk in this BVD
1721 for (i
= 0, j
= 0; i
< ddf
->mppe
&&
1722 j
< be16_to_cpu(conf
->prim_elmnt_count
); i
++) {
1723 if (be32_to_cpu(conf
->phys_refnum
[i
]) != 0xffffffff) {
1731 dprintf("%s: couldn't find BVD member %u (total %u)\n",
1732 __func__
, n
, be16_to_cpu(conf
->prim_elmnt_count
));
1736 static struct vd_config
*find_vdcr(struct ddf_super
*ddf
, unsigned int inst
,
1738 unsigned int *n_bvd
, struct vcl
**vcl
)
1742 for (v
= ddf
->conflist
; v
; v
= v
->next
) {
1743 unsigned int nsec
, ibvd
= 0;
1744 struct vd_config
*conf
;
1745 if (inst
!= v
->vcnum
)
1748 if (conf
->sec_elmnt_count
== 1) {
1749 if (find_index_in_bvd(ddf
, conf
, n
, n_bvd
)) {
1755 if (v
->other_bvds
== NULL
) {
1756 pr_err("%s: BUG: other_bvds is NULL, nsec=%u\n",
1757 __func__
, conf
->sec_elmnt_count
);
1760 nsec
= n
/ be16_to_cpu(conf
->prim_elmnt_count
);
1761 if (conf
->sec_elmnt_seq
!= nsec
) {
1762 for (ibvd
= 1; ibvd
< conf
->sec_elmnt_count
; ibvd
++) {
1763 if (v
->other_bvds
[ibvd
-1]->sec_elmnt_seq
1767 if (ibvd
== conf
->sec_elmnt_count
)
1769 conf
= v
->other_bvds
[ibvd
-1];
1771 if (!find_index_in_bvd(ddf
, conf
,
1772 n
- nsec
*conf
->sec_elmnt_count
, n_bvd
))
1774 dprintf("%s: found disk %u as member %u in bvd %d of array %u\n"
1775 , __func__
, n
, *n_bvd
, ibvd
, inst
);
1780 pr_err("%s: Could't find disk %d in array %u\n", __func__
, n
, inst
);
1785 static int find_phys(const struct ddf_super
*ddf
, be32 phys_refnum
)
1787 /* Find the entry in phys_disk which has the given refnum
1788 * and return it's index
1791 for (i
= 0; i
< be16_to_cpu(ddf
->phys
->max_pdes
); i
++)
1792 if (be32_eq(ddf
->phys
->entries
[i
].refnum
, phys_refnum
))
1797 static void uuid_from_ddf_guid(const char *guid
, int uuid
[4])
1800 struct sha1_ctx ctx
;
1801 sha1_init_ctx(&ctx
);
1802 sha1_process_bytes(guid
, DDF_GUID_LEN
, &ctx
);
1803 sha1_finish_ctx(&ctx
, buf
);
1804 memcpy(uuid
, buf
, 4*4);
1807 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4])
1809 /* The uuid returned here is used for:
1810 * uuid to put into bitmap file (Create, Grow)
1811 * uuid for backup header when saving critical section (Grow)
1812 * comparing uuids when re-adding a device into an array
1813 * In these cases the uuid required is that of the data-array,
1814 * not the device-set.
1815 * uuid to recognise same set when adding a missing device back
1816 * to an array. This is a uuid for the device-set.
1818 * For each of these we can make do with a truncated
1819 * or hashed uuid rather than the original, as long as
1821 * In the case of SVD we assume the BVD is of interest,
1822 * though that might be the case if a bitmap were made for
1823 * a mirrored SVD - worry about that later.
1824 * So we need to find the VD configuration record for the
1825 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1826 * The first 16 bytes of the sha1 of these is used.
1828 struct ddf_super
*ddf
= st
->sb
;
1829 struct vcl
*vcl
= ddf
->currentconf
;
1833 guid
= vcl
->conf
.guid
;
1835 guid
= ddf
->anchor
.guid
;
1836 uuid_from_ddf_guid(guid
, uuid
);
1839 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1841 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1843 struct ddf_super
*ddf
= st
->sb
;
1844 int map_disks
= info
->array
.raid_disks
;
1847 if (ddf
->currentconf
) {
1848 getinfo_super_ddf_bvd(st
, info
, map
);
1851 memset(info
, 0, sizeof(*info
));
1853 info
->array
.raid_disks
= be16_to_cpu(ddf
->phys
->used_pdes
);
1854 info
->array
.level
= LEVEL_CONTAINER
;
1855 info
->array
.layout
= 0;
1856 info
->array
.md_minor
= -1;
1857 cptr
= (__u32
*)(ddf
->anchor
.guid
+ 16);
1858 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
1860 info
->array
.utime
= 0;
1861 info
->array
.chunk_size
= 0;
1862 info
->container_enough
= 1;
1864 info
->disk
.major
= 0;
1865 info
->disk
.minor
= 0;
1867 info
->disk
.number
= be32_to_cpu(ddf
->dlist
->disk
.refnum
);
1868 info
->disk
.raid_disk
= find_phys(ddf
, ddf
->dlist
->disk
.refnum
);
1870 info
->data_offset
= be64_to_cpu(ddf
->phys
->
1871 entries
[info
->disk
.raid_disk
].
1873 info
->component_size
= ddf
->dlist
->size
- info
->data_offset
;
1875 info
->disk
.number
= -1;
1876 info
->disk
.raid_disk
= -1;
1877 // info->disk.raid_disk = find refnum in the table and use index;
1879 info
->disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
1881 info
->recovery_start
= MaxSector
;
1882 info
->reshape_active
= 0;
1883 info
->recovery_blocked
= 0;
1886 info
->array
.major_version
= -1;
1887 info
->array
.minor_version
= -2;
1888 strcpy(info
->text_version
, "ddf");
1889 info
->safe_mode_delay
= 0;
1891 uuid_from_super_ddf(st
, info
->uuid
);
1895 for (i
= 0 ; i
< map_disks
; i
++) {
1896 if (i
< info
->array
.raid_disks
&&
1897 (be16_to_cpu(ddf
->phys
->entries
[i
].state
)
1899 !(be16_to_cpu(ddf
->phys
->entries
[i
].state
)
1908 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1910 struct ddf_super
*ddf
= st
->sb
;
1911 struct vcl
*vc
= ddf
->currentconf
;
1912 int cd
= ddf
->currentdev
;
1916 int map_disks
= info
->array
.raid_disks
;
1918 struct vd_config
*conf
;
1920 memset(info
, 0, sizeof(*info
));
1921 if (layout_ddf2md(&vc
->conf
, &info
->array
) == -1)
1923 info
->array
.md_minor
= -1;
1924 cptr
= (__u32
*)(vc
->conf
.guid
+ 16);
1925 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
1926 info
->array
.utime
= DECADE
+ be32_to_cpu(vc
->conf
.timestamp
);
1927 info
->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
1928 info
->custom_array_size
= 0;
1931 n_prim
= be16_to_cpu(conf
->prim_elmnt_count
);
1932 if (conf
->sec_elmnt_count
> 1 && cd
>= n_prim
) {
1933 int ibvd
= cd
/ n_prim
- 1;
1935 conf
= vc
->other_bvds
[ibvd
];
1938 if (cd
>= 0 && (unsigned)cd
< ddf
->mppe
) {
1940 be64_to_cpu(LBA_OFFSET(ddf
, conf
)[cd
]);
1941 if (vc
->block_sizes
)
1942 info
->component_size
= vc
->block_sizes
[cd
];
1944 info
->component_size
= be64_to_cpu(conf
->blocks
);
1947 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
1948 if (be32_eq(dl
->disk
.refnum
, conf
->phys_refnum
[cd
]))
1951 info
->disk
.major
= 0;
1952 info
->disk
.minor
= 0;
1953 info
->disk
.state
= 0;
1955 info
->disk
.major
= dl
->major
;
1956 info
->disk
.minor
= dl
->minor
;
1957 info
->disk
.raid_disk
= cd
+ conf
->sec_elmnt_seq
1958 * be16_to_cpu(conf
->prim_elmnt_count
);
1959 info
->disk
.number
= dl
->pdnum
;
1960 info
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
1963 info
->container_member
= ddf
->currentconf
->vcnum
;
1965 info
->recovery_start
= MaxSector
;
1966 info
->resync_start
= 0;
1967 info
->reshape_active
= 0;
1968 info
->recovery_blocked
= 0;
1969 if (!(ddf
->virt
->entries
[info
->container_member
].state
1970 & DDF_state_inconsistent
) &&
1971 (ddf
->virt
->entries
[info
->container_member
].init_state
1972 & DDF_initstate_mask
)
1974 info
->resync_start
= MaxSector
;
1976 uuid_from_super_ddf(st
, info
->uuid
);
1978 info
->array
.major_version
= -1;
1979 info
->array
.minor_version
= -2;
1980 sprintf(info
->text_version
, "/%s/%d",
1981 st
->container_devnm
,
1982 info
->container_member
);
1983 info
->safe_mode_delay
= DDF_SAFE_MODE_DELAY
;
1985 memcpy(info
->name
, ddf
->virt
->entries
[info
->container_member
].name
, 16);
1988 if (info
->name
[j
] == ' ')
1992 for (j
= 0; j
< map_disks
; j
++) {
1994 if (j
< info
->array
.raid_disks
) {
1995 int i
= find_phys(ddf
, vc
->conf
.phys_refnum
[j
]);
1997 (be16_to_cpu(ddf
->phys
->entries
[i
].state
)
1999 !(be16_to_cpu(ddf
->phys
->entries
[i
].state
)
2006 static int update_super_ddf(struct supertype
*st
, struct mdinfo
*info
,
2008 char *devname
, int verbose
,
2009 int uuid_set
, char *homehost
)
2011 /* For 'assemble' and 'force' we need to return non-zero if any
2012 * change was made. For others, the return value is ignored.
2013 * Update options are:
2014 * force-one : This device looks a bit old but needs to be included,
2015 * update age info appropriately.
2016 * assemble: clear any 'faulty' flag to allow this device to
2018 * force-array: Array is degraded but being forced, mark it clean
2019 * if that will be needed to assemble it.
2021 * newdev: not used ????
2022 * grow: Array has gained a new device - this is currently for
2024 * resync: mark as dirty so a resync will happen.
2025 * uuid: Change the uuid of the array to match what is given
2026 * homehost: update the recorded homehost
2027 * name: update the name - preserving the homehost
2028 * _reshape_progress: record new reshape_progress position.
2030 * Following are not relevant for this version:
2031 * sparc2.2 : update from old dodgey metadata
2032 * super-minor: change the preferred_minor number
2033 * summaries: update redundant counters.
2036 // struct ddf_super *ddf = st->sb;
2037 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
2038 // struct virtual_entry *ve = find_ve(ddf);
2040 /* we don't need to handle "force-*" or "assemble" as
2041 * there is no need to 'trick' the kernel. We the metadata is
2042 * first updated to activate the array, all the implied modifications
2046 if (strcmp(update
, "grow") == 0) {
2048 } else if (strcmp(update
, "resync") == 0) {
2049 // info->resync_checkpoint = 0;
2050 } else if (strcmp(update
, "homehost") == 0) {
2051 /* homehost is stored in controller->vendor_data,
2052 * or it is when we are the vendor
2054 // if (info->vendor_is_local)
2055 // strcpy(ddf->controller.vendor_data, homehost);
2057 } else if (strcmp(update
, "name") == 0) {
2058 /* name is stored in virtual_entry->name */
2059 // memset(ve->name, ' ', 16);
2060 // strncpy(ve->name, info->name, 16);
2062 } else if (strcmp(update
, "_reshape_progress") == 0) {
2063 /* We don't support reshape yet */
2064 } else if (strcmp(update
, "assemble") == 0 ) {
2065 /* Do nothing, just succeed */
2070 // update_all_csum(ddf);
2075 static void make_header_guid(char *guid
)
2078 /* Create a DDF Header of Virtual Disk GUID */
2080 /* 24 bytes of fiction required.
2081 * first 8 are a 'vendor-id' - "Linux-MD"
2082 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
2083 * Remaining 8 random number plus timestamp
2085 memcpy(guid
, T10
, sizeof(T10
));
2086 stamp
= cpu_to_be32(0xdeadbeef);
2087 memcpy(guid
+8, &stamp
, 4);
2088 stamp
= cpu_to_be32(0);
2089 memcpy(guid
+12, &stamp
, 4);
2090 stamp
= cpu_to_be32(time(0) - DECADE
);
2091 memcpy(guid
+16, &stamp
, 4);
2092 stamp
._v32
= random32();
2093 memcpy(guid
+20, &stamp
, 4);
2096 static unsigned int find_unused_vde(const struct ddf_super
*ddf
)
2099 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
2100 if (all_ff(ddf
->virt
->entries
[i
].guid
))
2103 return DDF_NOTFOUND
;
2106 static unsigned int find_vde_by_name(const struct ddf_super
*ddf
,
2111 return DDF_NOTFOUND
;
2112 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
2113 if (all_ff(ddf
->virt
->entries
[i
].guid
))
2115 if (!strncmp(name
, ddf
->virt
->entries
[i
].name
,
2116 sizeof(ddf
->virt
->entries
[i
].name
)))
2119 return DDF_NOTFOUND
;
2123 static unsigned int find_vde_by_guid(const struct ddf_super
*ddf
,
2127 if (guid
== NULL
|| all_ff(guid
))
2128 return DDF_NOTFOUND
;
2129 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++)
2130 if (!memcmp(ddf
->virt
->entries
[i
].guid
, guid
, DDF_GUID_LEN
))
2132 return DDF_NOTFOUND
;
2136 static int init_super_ddf_bvd(struct supertype
*st
,
2137 mdu_array_info_t
*info
,
2138 unsigned long long size
,
2139 char *name
, char *homehost
,
2140 int *uuid
, unsigned long long data_offset
);
2142 static int init_super_ddf(struct supertype
*st
,
2143 mdu_array_info_t
*info
,
2144 unsigned long long size
, char *name
, char *homehost
,
2145 int *uuid
, unsigned long long data_offset
)
2147 /* This is primarily called by Create when creating a new array.
2148 * We will then get add_to_super called for each component, and then
2149 * write_init_super called to write it out to each device.
2150 * For DDF, Create can create on fresh devices or on a pre-existing
2152 * To create on a pre-existing array a different method will be called.
2153 * This one is just for fresh drives.
2155 * We need to create the entire 'ddf' structure which includes:
2156 * DDF headers - these are easy.
2157 * Controller data - a Sector describing this controller .. not that
2158 * this is a controller exactly.
2159 * Physical Disk Record - one entry per device, so
2160 * leave plenty of space.
2161 * Virtual Disk Records - again, just leave plenty of space.
2162 * This just lists VDs, doesn't give details
2163 * Config records - describes the VDs that use this disk
2164 * DiskData - describes 'this' device.
2165 * BadBlockManagement - empty
2166 * Diag Space - empty
2167 * Vendor Logs - Could we put bitmaps here?
2170 struct ddf_super
*ddf
;
2173 int max_phys_disks
, max_virt_disks
;
2174 unsigned long long sector
;
2178 struct phys_disk
*pd
;
2179 struct virtual_disk
*vd
;
2181 if (data_offset
!= INVALID_SECTORS
) {
2182 pr_err("data-offset not supported by DDF\n");
2187 return init_super_ddf_bvd(st
, info
, size
, name
, homehost
, uuid
,
2190 if (posix_memalign((void**)&ddf
, 512, sizeof(*ddf
)) != 0) {
2191 pr_err("%s could not allocate superblock\n", __func__
);
2194 memset(ddf
, 0, sizeof(*ddf
));
2195 ddf
->dlist
= NULL
; /* no physical disks yet */
2196 ddf
->conflist
= NULL
; /* No virtual disks yet */
2200 /* zeroing superblock */
2204 /* At least 32MB *must* be reserved for the ddf. So let's just
2205 * start 32MB from the end, and put the primary header there.
2206 * Don't do secondary for now.
2207 * We don't know exactly where that will be yet as it could be
2208 * different on each device. To just set up the lengths.
2212 ddf
->anchor
.magic
= DDF_HEADER_MAGIC
;
2213 make_header_guid(ddf
->anchor
.guid
);
2215 memcpy(ddf
->anchor
.revision
, DDF_REVISION_2
, 8);
2216 ddf
->anchor
.seq
= cpu_to_be32(1);
2217 ddf
->anchor
.timestamp
= cpu_to_be32(time(0) - DECADE
);
2218 ddf
->anchor
.openflag
= 0xFF;
2219 ddf
->anchor
.foreignflag
= 0;
2220 ddf
->anchor
.enforcegroups
= 0; /* Is this best?? */
2221 ddf
->anchor
.pad0
= 0xff;
2222 memset(ddf
->anchor
.pad1
, 0xff, 12);
2223 memset(ddf
->anchor
.header_ext
, 0xff, 32);
2224 ddf
->anchor
.primary_lba
= cpu_to_be64(~(__u64
)0);
2225 ddf
->anchor
.secondary_lba
= cpu_to_be64(~(__u64
)0);
2226 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
2227 memset(ddf
->anchor
.pad2
, 0xff, 3);
2228 ddf
->anchor
.workspace_len
= cpu_to_be32(32768); /* Must be reserved */
2229 /* Put this at bottom of 32M reserved.. */
2230 ddf
->anchor
.workspace_lba
= cpu_to_be64(~(__u64
)0);
2231 max_phys_disks
= 1023; /* Should be enough */
2232 ddf
->anchor
.max_pd_entries
= cpu_to_be16(max_phys_disks
);
2233 max_virt_disks
= 255;
2234 ddf
->anchor
.max_vd_entries
= cpu_to_be16(max_virt_disks
); /* ?? */
2235 ddf
->anchor
.max_partitions
= cpu_to_be16(64); /* ?? */
2238 ddf
->conf_rec_len
= 1 + ROUND_UP(ddf
->mppe
* (4+8), 512)/512;
2239 ddf
->anchor
.config_record_len
= cpu_to_be16(ddf
->conf_rec_len
);
2240 ddf
->anchor
.max_primary_element_entries
= cpu_to_be16(ddf
->mppe
);
2241 memset(ddf
->anchor
.pad3
, 0xff, 54);
2242 /* controller sections is one sector long immediately
2243 * after the ddf header */
2245 ddf
->anchor
.controller_section_offset
= cpu_to_be32(sector
);
2246 ddf
->anchor
.controller_section_length
= cpu_to_be32(1);
2249 /* phys is 8 sectors after that */
2250 pdsize
= ROUND_UP(sizeof(struct phys_disk
) +
2251 sizeof(struct phys_disk_entry
)*max_phys_disks
,
2253 switch(pdsize
/512) {
2254 case 2: case 8: case 32: case 128: case 512: break;
2257 ddf
->anchor
.phys_section_offset
= cpu_to_be32(sector
);
2258 ddf
->anchor
.phys_section_length
=
2259 cpu_to_be32(pdsize
/512); /* max_primary_element_entries/8 */
2260 sector
+= pdsize
/512;
2262 /* virt is another 32 sectors */
2263 vdsize
= ROUND_UP(sizeof(struct virtual_disk
) +
2264 sizeof(struct virtual_entry
) * max_virt_disks
,
2266 switch(vdsize
/512) {
2267 case 2: case 8: case 32: case 128: case 512: break;
2270 ddf
->anchor
.virt_section_offset
= cpu_to_be32(sector
);
2271 ddf
->anchor
.virt_section_length
=
2272 cpu_to_be32(vdsize
/512); /* max_vd_entries/8 */
2273 sector
+= vdsize
/512;
2275 clen
= ddf
->conf_rec_len
* (ddf
->max_part
+1);
2276 ddf
->anchor
.config_section_offset
= cpu_to_be32(sector
);
2277 ddf
->anchor
.config_section_length
= cpu_to_be32(clen
);
2280 ddf
->anchor
.data_section_offset
= cpu_to_be32(sector
);
2281 ddf
->anchor
.data_section_length
= cpu_to_be32(1);
2284 ddf
->anchor
.bbm_section_length
= cpu_to_be32(0);
2285 ddf
->anchor
.bbm_section_offset
= cpu_to_be32(0xFFFFFFFF);
2286 ddf
->anchor
.diag_space_length
= cpu_to_be32(0);
2287 ddf
->anchor
.diag_space_offset
= cpu_to_be32(0xFFFFFFFF);
2288 ddf
->anchor
.vendor_length
= cpu_to_be32(0);
2289 ddf
->anchor
.vendor_offset
= cpu_to_be32(0xFFFFFFFF);
2291 memset(ddf
->anchor
.pad4
, 0xff, 256);
2293 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2294 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2296 ddf
->primary
.openflag
= 1; /* I guess.. */
2297 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
2299 ddf
->secondary
.openflag
= 1; /* I guess.. */
2300 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
2302 ddf
->active
= &ddf
->primary
;
2304 ddf
->controller
.magic
= DDF_CONTROLLER_MAGIC
;
2306 /* 24 more bytes of fiction required.
2307 * first 8 are a 'vendor-id' - "Linux-MD"
2308 * Remaining 16 are serial number.... maybe a hostname would do?
2310 memcpy(ddf
->controller
.guid
, T10
, sizeof(T10
));
2311 gethostname(hostname
, sizeof(hostname
));
2312 hostname
[sizeof(hostname
) - 1] = 0;
2313 hostlen
= strlen(hostname
);
2314 memcpy(ddf
->controller
.guid
+ 24 - hostlen
, hostname
, hostlen
);
2315 for (i
= strlen(T10
) ; i
+hostlen
< 24; i
++)
2316 ddf
->controller
.guid
[i
] = ' ';
2318 ddf
->controller
.type
.vendor_id
= cpu_to_be16(0xDEAD);
2319 ddf
->controller
.type
.device_id
= cpu_to_be16(0xBEEF);
2320 ddf
->controller
.type
.sub_vendor_id
= cpu_to_be16(0);
2321 ddf
->controller
.type
.sub_device_id
= cpu_to_be16(0);
2322 memcpy(ddf
->controller
.product_id
, "What Is My PID??", 16);
2323 memset(ddf
->controller
.pad
, 0xff, 8);
2324 memset(ddf
->controller
.vendor_data
, 0xff, 448);
2325 if (homehost
&& strlen(homehost
) < 440)
2326 strcpy((char*)ddf
->controller
.vendor_data
, homehost
);
2328 if (posix_memalign((void**)&pd
, 512, pdsize
) != 0) {
2329 pr_err("%s could not allocate pd\n", __func__
);
2333 ddf
->pdsize
= pdsize
;
2335 memset(pd
, 0xff, pdsize
);
2336 memset(pd
, 0, sizeof(*pd
));
2337 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2338 pd
->used_pdes
= cpu_to_be16(0);
2339 pd
->max_pdes
= cpu_to_be16(max_phys_disks
);
2340 memset(pd
->pad
, 0xff, 52);
2341 for (i
= 0; i
< max_phys_disks
; i
++)
2342 memset(pd
->entries
[i
].guid
, 0xff, DDF_GUID_LEN
);
2344 if (posix_memalign((void**)&vd
, 512, vdsize
) != 0) {
2345 pr_err("%s could not allocate vd\n", __func__
);
2349 ddf
->vdsize
= vdsize
;
2350 memset(vd
, 0, vdsize
);
2351 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
2352 vd
->populated_vdes
= cpu_to_be16(0);
2353 vd
->max_vdes
= cpu_to_be16(max_virt_disks
);
2354 memset(vd
->pad
, 0xff, 52);
2356 for (i
=0; i
<max_virt_disks
; i
++)
2357 memset(&vd
->entries
[i
], 0xff, sizeof(struct virtual_entry
));
2360 ddf_set_updates_pending(ddf
);
2364 static int chunk_to_shift(int chunksize
)
2366 return ffs(chunksize
/512)-1;
2371 unsigned long long start
, size
;
2373 static int cmp_extent(const void *av
, const void *bv
)
2375 const struct extent
*a
= av
;
2376 const struct extent
*b
= bv
;
2377 if (a
->start
< b
->start
)
2379 if (a
->start
> b
->start
)
2384 static struct extent
*get_extents(struct ddf_super
*ddf
, struct dl
*dl
)
2386 /* find a list of used extents on the give physical device
2387 * (dnum) of the given ddf.
2388 * Return a malloced array of 'struct extent'
2390 * FIXME ignore DDF_Legacy devices?
2396 __u16 state
= be16_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].state
);
2398 if ((state
& (DDF_Online
|DDF_Failed
|DDF_Missing
)) != DDF_Online
)
2401 rv
= xmalloc(sizeof(struct extent
) * (ddf
->max_part
+ 2));
2403 for (i
= 0; i
< ddf
->max_part
; i
++) {
2404 const struct vd_config
*bvd
;
2406 struct vcl
*v
= dl
->vlist
[i
];
2408 get_pd_index_from_refnum(v
, dl
->disk
.refnum
, ddf
->mppe
,
2409 &bvd
, &ibvd
) == DDF_NOTFOUND
)
2411 rv
[n
].start
= be64_to_cpu(LBA_OFFSET(ddf
, bvd
)[ibvd
]);
2412 rv
[n
].size
= be64_to_cpu(bvd
->blocks
);
2415 qsort(rv
, n
, sizeof(*rv
), cmp_extent
);
2417 rv
[n
].start
= be64_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].config_size
);
2423 static int init_super_ddf_bvd(struct supertype
*st
,
2424 mdu_array_info_t
*info
,
2425 unsigned long long size
,
2426 char *name
, char *homehost
,
2427 int *uuid
, unsigned long long data_offset
)
2429 /* We are creating a BVD inside a pre-existing container.
2430 * so st->sb is already set.
2431 * We need to create a new vd_config and a new virtual_entry
2433 struct ddf_super
*ddf
= st
->sb
;
2434 unsigned int venum
, i
;
2435 struct virtual_entry
*ve
;
2437 struct vd_config
*vc
;
2439 if (find_vde_by_name(ddf
, name
) != DDF_NOTFOUND
) {
2440 pr_err("This ddf already has an array called %s\n", name
);
2443 venum
= find_unused_vde(ddf
);
2444 if (venum
== DDF_NOTFOUND
) {
2445 pr_err("Cannot find spare slot for virtual disk\n");
2448 ve
= &ddf
->virt
->entries
[venum
];
2450 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
2451 * timestamp, random number
2453 make_header_guid(ve
->guid
);
2454 ve
->unit
= cpu_to_be16(info
->md_minor
);
2456 ve
->guid_crc
._v16
= crc32(0, (unsigned char *)ddf
->anchor
.guid
,
2458 ve
->type
= cpu_to_be16(0);
2459 ve
->state
= DDF_state_degraded
; /* Will be modified as devices are added */
2460 if (info
->state
& 1) /* clean */
2461 ve
->init_state
= DDF_init_full
;
2463 ve
->init_state
= DDF_init_not
;
2465 memset(ve
->pad1
, 0xff, 14);
2466 memset(ve
->name
, ' ', 16);
2468 strncpy(ve
->name
, name
, 16);
2469 ddf
->virt
->populated_vdes
=
2470 cpu_to_be16(be16_to_cpu(ddf
->virt
->populated_vdes
)+1);
2472 /* Now create a new vd_config */
2473 if (posix_memalign((void**)&vcl
, 512,
2474 (offsetof(struct vcl
, conf
) + ddf
->conf_rec_len
* 512)) != 0) {
2475 pr_err("%s could not allocate vd_config\n", __func__
);
2479 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
2482 vc
->magic
= DDF_VD_CONF_MAGIC
;
2483 memcpy(vc
->guid
, ve
->guid
, DDF_GUID_LEN
);
2484 vc
->timestamp
= cpu_to_be32(time(0)-DECADE
);
2485 vc
->seqnum
= cpu_to_be32(1);
2486 memset(vc
->pad0
, 0xff, 24);
2487 vc
->chunk_shift
= chunk_to_shift(info
->chunk_size
);
2488 if (layout_md2ddf(info
, vc
) == -1 ||
2489 be16_to_cpu(vc
->prim_elmnt_count
) > ddf
->mppe
) {
2490 pr_err("%s: unsupported RAID level/layout %d/%d with %d disks\n",
2491 __func__
, info
->level
, info
->layout
, info
->raid_disks
);
2495 vc
->sec_elmnt_seq
= 0;
2496 if (alloc_other_bvds(ddf
, vcl
) != 0) {
2497 pr_err("%s could not allocate other bvds\n",
2502 vc
->blocks
= cpu_to_be64(info
->size
* 2);
2503 vc
->array_blocks
= cpu_to_be64(
2504 calc_array_size(info
->level
, info
->raid_disks
, info
->layout
,
2505 info
->chunk_size
, info
->size
*2));
2506 memset(vc
->pad1
, 0xff, 8);
2507 vc
->spare_refs
[0] = cpu_to_be32(0xffffffff);
2508 vc
->spare_refs
[1] = cpu_to_be32(0xffffffff);
2509 vc
->spare_refs
[2] = cpu_to_be32(0xffffffff);
2510 vc
->spare_refs
[3] = cpu_to_be32(0xffffffff);
2511 vc
->spare_refs
[4] = cpu_to_be32(0xffffffff);
2512 vc
->spare_refs
[5] = cpu_to_be32(0xffffffff);
2513 vc
->spare_refs
[6] = cpu_to_be32(0xffffffff);
2514 vc
->spare_refs
[7] = cpu_to_be32(0xffffffff);
2515 memset(vc
->cache_pol
, 0, 8);
2517 memset(vc
->pad2
, 0xff, 3);
2518 memset(vc
->pad3
, 0xff, 52);
2519 memset(vc
->pad4
, 0xff, 192);
2520 memset(vc
->v0
, 0xff, 32);
2521 memset(vc
->v1
, 0xff, 32);
2522 memset(vc
->v2
, 0xff, 16);
2523 memset(vc
->v3
, 0xff, 16);
2524 memset(vc
->vendor
, 0xff, 32);
2526 memset(vc
->phys_refnum
, 0xff, 4*ddf
->mppe
);
2527 memset(vc
->phys_refnum
+ddf
->mppe
, 0x00, 8*ddf
->mppe
);
2529 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++) {
2530 memcpy(vcl
->other_bvds
[i
-1], vc
, ddf
->conf_rec_len
* 512);
2531 vcl
->other_bvds
[i
-1]->sec_elmnt_seq
= i
;
2534 vcl
->next
= ddf
->conflist
;
2535 ddf
->conflist
= vcl
;
2536 ddf
->currentconf
= vcl
;
2537 ddf_set_updates_pending(ddf
);
2543 static int get_svd_state(const struct ddf_super
*, const struct vcl
*);
2545 static void add_to_super_ddf_bvd(struct supertype
*st
,
2546 mdu_disk_info_t
*dk
, int fd
, char *devname
)
2548 /* fd and devname identify a device with-in the ddf container (st).
2549 * dk identifies a location in the new BVD.
2550 * We need to find suitable free space in that device and update
2551 * the phys_refnum and lba_offset for the newly created vd_config.
2552 * We might also want to update the type in the phys_disk
2555 * Alternately: fd == -1 and we have already chosen which device to
2556 * use and recorded in dlist->raid_disk;
2559 struct ddf_super
*ddf
= st
->sb
;
2560 struct vd_config
*vc
;
2562 unsigned long long blocks
, pos
, esize
;
2564 unsigned int raid_disk
= dk
->raid_disk
;
2567 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2568 if (dl
->raiddisk
== dk
->raid_disk
)
2571 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2572 if (dl
->major
== dk
->major
&&
2573 dl
->minor
== dk
->minor
)
2576 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
2579 vc
= &ddf
->currentconf
->conf
;
2580 if (vc
->sec_elmnt_count
> 1) {
2581 unsigned int n
= be16_to_cpu(vc
->prim_elmnt_count
);
2583 vc
= ddf
->currentconf
->other_bvds
[raid_disk
/ n
- 1];
2587 ex
= get_extents(ddf
, dl
);
2592 blocks
= be64_to_cpu(vc
->blocks
);
2593 if (ddf
->currentconf
->block_sizes
)
2594 blocks
= ddf
->currentconf
->block_sizes
[dk
->raid_disk
];
2597 esize
= ex
[i
].start
- pos
;
2598 if (esize
>= blocks
)
2600 pos
= ex
[i
].start
+ ex
[i
].size
;
2602 } while (ex
[i
-1].size
);
2608 ddf
->currentdev
= dk
->raid_disk
;
2609 vc
->phys_refnum
[raid_disk
] = dl
->disk
.refnum
;
2610 LBA_OFFSET(ddf
, vc
)[raid_disk
] = cpu_to_be64(pos
);
2612 for (i
= 0; i
< ddf
->max_part
; i
++)
2613 if (dl
->vlist
[i
] == NULL
)
2615 if (i
== ddf
->max_part
)
2617 dl
->vlist
[i
] = ddf
->currentconf
;
2622 dl
->devname
= devname
;
2624 /* Check if we can mark array as optimal yet */
2625 i
= ddf
->currentconf
->vcnum
;
2626 ddf
->virt
->entries
[i
].state
=
2627 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
2628 | get_svd_state(ddf
, ddf
->currentconf
);
2629 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
2630 cpu_to_be16(DDF_Global_Spare
));
2631 be16_set(ddf
->phys
->entries
[dl
->pdnum
].type
,
2632 cpu_to_be16(DDF_Active_in_VD
));
2633 dprintf("%s: added disk %d/%08x to VD %d/%s as disk %d\n",
2634 __func__
, dl
->pdnum
, be32_to_cpu(dl
->disk
.refnum
),
2635 ddf
->currentconf
->vcnum
, guid_str(vc
->guid
),
2637 ddf_set_updates_pending(ddf
);
2640 static unsigned int find_unused_pde(const struct ddf_super
*ddf
)
2643 for (i
= 0; i
< be16_to_cpu(ddf
->phys
->max_pdes
); i
++) {
2644 if (all_ff(ddf
->phys
->entries
[i
].guid
))
2647 return DDF_NOTFOUND
;
2650 /* add a device to a container, either while creating it or while
2651 * expanding a pre-existing container
2653 static int add_to_super_ddf(struct supertype
*st
,
2654 mdu_disk_info_t
*dk
, int fd
, char *devname
,
2655 unsigned long long data_offset
)
2657 struct ddf_super
*ddf
= st
->sb
;
2661 unsigned long long size
;
2662 struct phys_disk_entry
*pde
;
2667 if (ddf
->currentconf
) {
2668 add_to_super_ddf_bvd(st
, dk
, fd
, devname
);
2672 /* This is device numbered dk->number. We need to create
2673 * a phys_disk entry and a more detailed disk_data entry.
2676 n
= find_unused_pde(ddf
);
2677 if (n
== DDF_NOTFOUND
) {
2678 pr_err("%s: No free slot in array, cannot add disk\n",
2682 pde
= &ddf
->phys
->entries
[n
];
2683 get_dev_size(fd
, NULL
, &size
);
2684 if (size
<= 32*1024*1024) {
2685 pr_err("%s: device size must be at least 32MB\n",
2691 if (posix_memalign((void**)&dd
, 512,
2692 sizeof(*dd
) + sizeof(dd
->vlist
[0]) * ddf
->max_part
) != 0) {
2693 pr_err("%s could allocate buffer for new disk, aborting\n",
2697 dd
->major
= major(stb
.st_rdev
);
2698 dd
->minor
= minor(stb
.st_rdev
);
2699 dd
->devname
= devname
;
2703 dd
->disk
.magic
= DDF_PHYS_DATA_MAGIC
;
2705 tm
= localtime(&now
);
2706 sprintf(dd
->disk
.guid
, "%8s%04d%02d%02d",
2707 T10
, tm
->tm_year
+1900, tm
->tm_mon
+1, tm
->tm_mday
);
2708 tptr
= (__u32
*)(dd
->disk
.guid
+ 16);
2709 *tptr
++ = random32();
2713 /* Cannot be bothered finding a CRC of some irrelevant details*/
2714 dd
->disk
.refnum
._v32
= random32();
2715 for (i
= be16_to_cpu(ddf
->active
->max_pd_entries
);
2717 if (be32_eq(ddf
->phys
->entries
[i
-1].refnum
,
2722 dd
->disk
.forced_ref
= 1;
2723 dd
->disk
.forced_guid
= 1;
2724 memset(dd
->disk
.vendor
, ' ', 32);
2725 memcpy(dd
->disk
.vendor
, "Linux", 5);
2726 memset(dd
->disk
.pad
, 0xff, 442);
2727 for (i
= 0; i
< ddf
->max_part
; i
++)
2728 dd
->vlist
[i
] = NULL
;
2732 if (st
->update_tail
) {
2733 int len
= (sizeof(struct phys_disk
) +
2734 sizeof(struct phys_disk_entry
));
2735 struct phys_disk
*pd
;
2738 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2739 pd
->used_pdes
= cpu_to_be16(n
);
2740 pde
= &pd
->entries
[0];
2743 ddf
->phys
->used_pdes
= cpu_to_be16(
2744 1 + be16_to_cpu(ddf
->phys
->used_pdes
));
2746 memcpy(pde
->guid
, dd
->disk
.guid
, DDF_GUID_LEN
);
2747 pde
->refnum
= dd
->disk
.refnum
;
2748 pde
->type
= cpu_to_be16(DDF_Forced_PD_GUID
| DDF_Global_Spare
);
2749 pde
->state
= cpu_to_be16(DDF_Online
);
2752 * If there is already a device in dlist, try to reserve the same
2753 * amount of workspace. Otherwise, use 32MB.
2754 * We checked disk size above already.
2756 #define __calc_lba(new, old, lba, mb) do { \
2757 unsigned long long dif; \
2758 if ((old) != NULL) \
2759 dif = (old)->size - be64_to_cpu((old)->lba); \
2761 dif = (new)->size; \
2762 if ((new)->size > dif) \
2763 (new)->lba = cpu_to_be64((new)->size - dif); \
2765 (new)->lba = cpu_to_be64((new)->size - (mb*1024*2)); \
2767 __calc_lba(dd
, ddf
->dlist
, workspace_lba
, 32);
2768 __calc_lba(dd
, ddf
->dlist
, primary_lba
, 16);
2769 __calc_lba(dd
, ddf
->dlist
, secondary_lba
, 32);
2770 pde
->config_size
= dd
->workspace_lba
;
2772 sprintf(pde
->path
, "%17.17s","Information: nil") ;
2773 memset(pde
->pad
, 0xff, 6);
2775 if (st
->update_tail
) {
2776 dd
->next
= ddf
->add_list
;
2779 dd
->next
= ddf
->dlist
;
2781 ddf_set_updates_pending(ddf
);
2787 static int remove_from_super_ddf(struct supertype
*st
, mdu_disk_info_t
*dk
)
2789 struct ddf_super
*ddf
= st
->sb
;
2792 /* mdmon has noticed that this disk (dk->major/dk->minor) has
2793 * disappeared from the container.
2794 * We need to arrange that it disappears from the metadata and
2795 * internal data structures too.
2796 * Most of the work is done by ddf_process_update which edits
2797 * the metadata and closes the file handle and attaches the memory
2798 * where free_updates will free it.
2800 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2801 if (dl
->major
== dk
->major
&&
2802 dl
->minor
== dk
->minor
)
2807 if (st
->update_tail
) {
2808 int len
= (sizeof(struct phys_disk
) +
2809 sizeof(struct phys_disk_entry
));
2810 struct phys_disk
*pd
;
2813 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2814 pd
->used_pdes
= cpu_to_be16(dl
->pdnum
);
2815 pd
->entries
[0].state
= cpu_to_be16(DDF_Missing
);
2816 append_metadata_update(st
, pd
, len
);
2823 * This is the write_init_super method for a ddf container. It is
2824 * called when creating a container or adding another device to a
2827 #define NULL_CONF_SZ 4096
2829 static int __write_ddf_structure(struct dl
*d
, struct ddf_super
*ddf
, __u8 type
)
2831 unsigned long long sector
;
2832 struct ddf_header
*header
;
2833 int fd
, i
, n_config
, conf_size
, buf_size
;
2840 case DDF_HEADER_PRIMARY
:
2841 header
= &ddf
->primary
;
2842 sector
= be64_to_cpu(header
->primary_lba
);
2844 case DDF_HEADER_SECONDARY
:
2845 header
= &ddf
->secondary
;
2846 sector
= be64_to_cpu(header
->secondary_lba
);
2852 header
->type
= type
;
2853 header
->openflag
= 1;
2854 header
->crc
= calc_crc(header
, 512);
2856 lseek64(fd
, sector
<<9, 0);
2857 if (write(fd
, header
, 512) < 0)
2860 ddf
->controller
.crc
= calc_crc(&ddf
->controller
, 512);
2861 if (write(fd
, &ddf
->controller
, 512) < 0)
2864 ddf
->phys
->crc
= calc_crc(ddf
->phys
, ddf
->pdsize
);
2865 if (write(fd
, ddf
->phys
, ddf
->pdsize
) < 0)
2867 ddf
->virt
->crc
= calc_crc(ddf
->virt
, ddf
->vdsize
);
2868 if (write(fd
, ddf
->virt
, ddf
->vdsize
) < 0)
2871 /* Now write lots of config records. */
2872 n_config
= ddf
->max_part
;
2873 conf_size
= ddf
->conf_rec_len
* 512;
2875 buf_size
= conf_size
* (n_config
+ 1);
2877 if (posix_memalign((void**)&conf
, 512, buf_size
) != 0)
2881 for (i
= 0 ; i
<= n_config
; i
++) {
2883 struct vd_config
*vdc
= NULL
;
2884 if (i
== n_config
) {
2885 c
= (struct vcl
*)d
->spare
;
2892 get_pd_index_from_refnum(
2895 (const struct vd_config
**)&vdc
,
2899 dprintf("writing conf record %i on disk %08x for %s/%u\n",
2900 i
, be32_to_cpu(d
->disk
.refnum
),
2901 guid_str(vdc
->guid
),
2902 vdc
->sec_elmnt_seq
);
2903 vdc
->seqnum
= header
->seq
;
2904 vdc
->crc
= calc_crc(vdc
, conf_size
);
2905 memcpy(conf
+ i
*conf_size
, vdc
, conf_size
);
2907 memset(conf
+ i
*conf_size
, 0xff, conf_size
);
2909 if (write(fd
, conf
, buf_size
) != buf_size
)
2912 d
->disk
.crc
= calc_crc(&d
->disk
, 512);
2913 if (write(fd
, &d
->disk
, 512) < 0)
2918 header
->openflag
= 0;
2919 header
->crc
= calc_crc(header
, 512);
2921 lseek64(fd
, sector
<<9, 0);
2922 if (write(fd
, header
, 512) < 0)
2928 static int _write_super_to_disk(struct ddf_super
*ddf
, struct dl
*d
)
2930 unsigned long long size
;
2935 /* We need to fill in the primary, (secondary) and workspace
2936 * lba's in the headers, set their checksums,
2937 * Also checksum phys, virt....
2939 * Then write everything out, finally the anchor is written.
2941 get_dev_size(fd
, NULL
, &size
);
2943 if (be64_to_cpu(d
->workspace_lba
) != 0ULL)
2944 ddf
->anchor
.workspace_lba
= d
->workspace_lba
;
2946 ddf
->anchor
.workspace_lba
=
2947 cpu_to_be64(size
- 32*1024*2);
2948 if (be64_to_cpu(d
->primary_lba
) != 0ULL)
2949 ddf
->anchor
.primary_lba
= d
->primary_lba
;
2951 ddf
->anchor
.primary_lba
=
2952 cpu_to_be64(size
- 16*1024*2);
2953 if (be64_to_cpu(d
->secondary_lba
) != 0ULL)
2954 ddf
->anchor
.secondary_lba
= d
->secondary_lba
;
2956 ddf
->anchor
.secondary_lba
=
2957 cpu_to_be64(size
- 32*1024*2);
2958 ddf
->anchor
.seq
= ddf
->active
->seq
;
2959 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2960 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2962 ddf
->anchor
.openflag
= 0xFF; /* 'open' means nothing */
2963 ddf
->anchor
.seq
= cpu_to_be32(0xFFFFFFFF); /* no sequencing in anchor */
2964 ddf
->anchor
.crc
= calc_crc(&ddf
->anchor
, 512);
2966 if (!__write_ddf_structure(d
, ddf
, DDF_HEADER_PRIMARY
))
2969 if (!__write_ddf_structure(d
, ddf
, DDF_HEADER_SECONDARY
))
2972 lseek64(fd
, (size
-1)*512, SEEK_SET
);
2973 if (write(fd
, &ddf
->anchor
, 512) < 0)
2980 static int __write_init_super_ddf(struct supertype
*st
)
2982 struct ddf_super
*ddf
= st
->sb
;
2987 pr_state(ddf
, __func__
);
2989 /* try to write updated metadata,
2990 * if we catch a failure move on to the next disk
2992 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
2994 successes
+= _write_super_to_disk(ddf
, d
);
2997 return attempts
!= successes
;
3000 static int write_init_super_ddf(struct supertype
*st
)
3002 struct ddf_super
*ddf
= st
->sb
;
3003 struct vcl
*currentconf
= ddf
->currentconf
;
3005 /* we are done with currentconf reset it to point st at the container */
3006 ddf
->currentconf
= NULL
;
3008 if (st
->update_tail
) {
3009 /* queue the virtual_disk and vd_config as metadata updates */
3010 struct virtual_disk
*vd
;
3011 struct vd_config
*vc
;
3016 int len
= (sizeof(struct phys_disk
) +
3017 sizeof(struct phys_disk_entry
));
3019 /* adding a disk to the container. */
3023 append_metadata_update(st
, ddf
->add_list
->mdupdate
, len
);
3024 ddf
->add_list
->mdupdate
= NULL
;
3028 /* Newly created VD */
3030 /* First the virtual disk. We have a slightly fake header */
3031 len
= sizeof(struct virtual_disk
) + sizeof(struct virtual_entry
);
3034 vd
->entries
[0] = ddf
->virt
->entries
[currentconf
->vcnum
];
3035 vd
->populated_vdes
= cpu_to_be16(currentconf
->vcnum
);
3036 append_metadata_update(st
, vd
, len
);
3038 /* Then the vd_config */
3039 len
= ddf
->conf_rec_len
* 512;
3040 tlen
= len
* currentconf
->conf
.sec_elmnt_count
;
3042 memcpy(vc
, ¤tconf
->conf
, len
);
3043 for (i
= 1; i
< currentconf
->conf
.sec_elmnt_count
; i
++)
3044 memcpy((char *)vc
+ i
*len
, currentconf
->other_bvds
[i
-1],
3046 append_metadata_update(st
, vc
, tlen
);
3048 /* FIXME I need to close the fds! */
3053 for (d
= ddf
->dlist
; d
; d
=d
->next
)
3054 while (Kill(d
->devname
, NULL
, 0, -1, 1) == 0);
3055 return __write_init_super_ddf(st
);
3061 static __u64
avail_size_ddf(struct supertype
*st
, __u64 devsize
,
3062 unsigned long long data_offset
)
3064 /* We must reserve the last 32Meg */
3065 if (devsize
<= 32*1024*2)
3067 return devsize
- 32*1024*2;
3072 static int reserve_space(struct supertype
*st
, int raiddisks
,
3073 unsigned long long size
, int chunk
,
3074 unsigned long long *freesize
)
3076 /* Find 'raiddisks' spare extents at least 'size' big (but
3077 * only caring about multiples of 'chunk') and remember
3079 * If the cannot be found, fail.
3082 struct ddf_super
*ddf
= st
->sb
;
3085 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3089 /* Now find largest extent on each device */
3090 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3091 struct extent
*e
= get_extents(ddf
, dl
);
3092 unsigned long long pos
= 0;
3095 unsigned long long minsize
= size
;
3103 unsigned long long esize
;
3104 esize
= e
[i
].start
- pos
;
3105 if (esize
>= minsize
) {
3109 pos
= e
[i
].start
+ e
[i
].size
;
3111 } while (e
[i
-1].size
);
3114 dl
->esize
= minsize
;
3118 if (cnt
< raiddisks
) {
3119 pr_err("not enough devices with space to create array.\n");
3120 return 0; /* No enough free spaces large enough */
3123 /* choose the largest size of which there are at least 'raiddisk' */
3124 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3126 if (dl
->esize
<= size
)
3128 /* This is bigger than 'size', see if there are enough */
3130 for (dl2
= ddf
->dlist
; dl2
; dl2
=dl2
->next
)
3131 if (dl2
->esize
>= dl
->esize
)
3133 if (cnt
>= raiddisks
)
3137 size
= size
/ chunk
;
3142 pr_err("not enough spare devices to create array.\n");
3146 /* We have a 'size' of which there are enough spaces.
3147 * We simply do a first-fit */
3149 for (dl
= ddf
->dlist
; dl
&& cnt
< raiddisks
; dl
=dl
->next
) {
3150 if (dl
->esize
< size
)
3160 validate_geometry_ddf_container(struct supertype
*st
,
3161 int level
, int layout
, int raiddisks
,
3162 int chunk
, unsigned long long size
,
3163 unsigned long long data_offset
,
3164 char *dev
, unsigned long long *freesize
,
3167 static int validate_geometry_ddf_bvd(struct supertype
*st
,
3168 int level
, int layout
, int raiddisks
,
3169 int *chunk
, unsigned long long size
,
3170 unsigned long long data_offset
,
3171 char *dev
, unsigned long long *freesize
,
3174 static int validate_geometry_ddf(struct supertype
*st
,
3175 int level
, int layout
, int raiddisks
,
3176 int *chunk
, unsigned long long size
,
3177 unsigned long long data_offset
,
3178 char *dev
, unsigned long long *freesize
,
3185 /* ddf potentially supports lots of things, but it depends on
3186 * what devices are offered (and maybe kernel version?)
3187 * If given unused devices, we will make a container.
3188 * If given devices in a container, we will make a BVD.
3189 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
3192 if (*chunk
== UnSet
)
3193 *chunk
= DEFAULT_CHUNK
;
3195 if (level
== -1000000) level
= LEVEL_CONTAINER
;
3196 if (level
== LEVEL_CONTAINER
) {
3197 /* Must be a fresh device to add to a container */
3198 return validate_geometry_ddf_container(st
, level
, layout
,
3200 size
, data_offset
, dev
,
3206 mdu_array_info_t array
= {
3207 .level
= level
, .layout
= layout
,
3208 .raid_disks
= raiddisks
3210 struct vd_config conf
;
3211 if (layout_md2ddf(&array
, &conf
) == -1) {
3213 pr_err("DDF does not support level %d /layout %d arrays with %d disks\n",
3214 level
, layout
, raiddisks
);
3217 /* Should check layout? etc */
3219 if (st
->sb
&& freesize
) {
3220 /* --create was given a container to create in.
3221 * So we need to check that there are enough
3222 * free spaces and return the amount of space.
3223 * We may as well remember which drives were
3224 * chosen so that add_to_super/getinfo_super
3227 return reserve_space(st
, raiddisks
, size
, *chunk
, freesize
);
3233 /* A container has already been opened, so we are
3234 * creating in there. Maybe a BVD, maybe an SVD.
3235 * Should make a distinction one day.
3237 return validate_geometry_ddf_bvd(st
, level
, layout
, raiddisks
,
3238 chunk
, size
, data_offset
, dev
,
3242 /* This is the first device for the array.
3243 * If it is a container, we read it in and do automagic allocations,
3244 * no other devices should be given.
3245 * Otherwise it must be a member device of a container, and we
3246 * do manual allocation.
3247 * Later we should check for a BVD and make an SVD.
3249 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3251 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
3253 if (sra
&& sra
->array
.major_version
== -1 &&
3254 strcmp(sra
->text_version
, "ddf") == 0) {
3257 /* find space for 'n' devices. */
3258 /* remember the devices */
3259 /* Somehow return the fact that we have enough */
3263 pr_err("ddf: Cannot create this array "
3264 "on device %s - a container is required.\n",
3268 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
3270 pr_err("ddf: Cannot open %s: %s\n",
3271 dev
, strerror(errno
));
3274 /* Well, it is in use by someone, maybe a 'ddf' container. */
3275 cfd
= open_container(fd
);
3279 pr_err("ddf: Cannot use %s: %s\n",
3280 dev
, strerror(EBUSY
));
3283 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
3285 if (sra
&& sra
->array
.major_version
== -1 &&
3286 strcmp(sra
->text_version
, "ddf") == 0) {
3287 /* This is a member of a ddf container. Load the container
3288 * and try to create a bvd
3290 struct ddf_super
*ddf
;
3291 if (load_super_ddf_all(st
, cfd
, (void **)&ddf
, NULL
) == 0) {
3293 strcpy(st
->container_devnm
, fd2devnm(cfd
));
3295 return validate_geometry_ddf_bvd(st
, level
, layout
,
3296 raiddisks
, chunk
, size
,
3302 } else /* device may belong to a different container */
3309 validate_geometry_ddf_container(struct supertype
*st
,
3310 int level
, int layout
, int raiddisks
,
3311 int chunk
, unsigned long long size
,
3312 unsigned long long data_offset
,
3313 char *dev
, unsigned long long *freesize
,
3317 unsigned long long ldsize
;
3319 if (level
!= LEVEL_CONTAINER
)
3324 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3327 pr_err("ddf: Cannot open %s: %s\n",
3328 dev
, strerror(errno
));
3331 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3337 *freesize
= avail_size_ddf(st
, ldsize
>> 9, INVALID_SECTORS
);
3344 static int validate_geometry_ddf_bvd(struct supertype
*st
,
3345 int level
, int layout
, int raiddisks
,
3346 int *chunk
, unsigned long long size
,
3347 unsigned long long data_offset
,
3348 char *dev
, unsigned long long *freesize
,
3352 struct ddf_super
*ddf
= st
->sb
;
3354 unsigned long long pos
= 0;
3355 unsigned long long maxsize
;
3358 /* ddf/bvd supports lots of things, but not containers */
3359 if (level
== LEVEL_CONTAINER
) {
3361 pr_err("DDF cannot create a container within an container\n");
3364 /* We must have the container info already read in. */
3369 /* General test: make sure there is space for
3370 * 'raiddisks' device extents of size 'size'.
3372 unsigned long long minsize
= size
;
3376 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
3382 e
= get_extents(ddf
, dl
);
3385 unsigned long long esize
;
3386 esize
= e
[i
].start
- pos
;
3387 if (esize
>= minsize
)
3389 pos
= e
[i
].start
+ e
[i
].size
;
3391 } while (e
[i
-1].size
);
3396 if (dcnt
< raiddisks
) {
3398 pr_err("ddf: Not enough devices with "
3399 "space for this array (%d < %d)\n",
3405 /* This device must be a member of the set */
3406 if (stat(dev
, &stb
) < 0)
3408 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3410 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
3411 if (dl
->major
== (int)major(stb
.st_rdev
) &&
3412 dl
->minor
== (int)minor(stb
.st_rdev
))
3417 pr_err("ddf: %s is not in the "
3422 e
= get_extents(ddf
, dl
);
3426 unsigned long long esize
;
3427 esize
= e
[i
].start
- pos
;
3428 if (esize
>= maxsize
)
3430 pos
= e
[i
].start
+ e
[i
].size
;
3432 } while (e
[i
-1].size
);
3433 *freesize
= maxsize
;
3439 static int load_super_ddf_all(struct supertype
*st
, int fd
,
3440 void **sbp
, char *devname
)
3443 struct ddf_super
*super
;
3444 struct mdinfo
*sd
, *best
= NULL
;
3450 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3453 if (sra
->array
.major_version
!= -1 ||
3454 sra
->array
.minor_version
!= -2 ||
3455 strcmp(sra
->text_version
, "ddf") != 0)
3458 if (posix_memalign((void**)&super
, 512, sizeof(*super
)) != 0)
3460 memset(super
, 0, sizeof(*super
));
3462 /* first, try each device, and choose the best ddf */
3463 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3465 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3466 dfd
= dev_open(nm
, O_RDONLY
);
3469 rv
= load_ddf_headers(dfd
, super
, NULL
);
3472 seq
= be32_to_cpu(super
->active
->seq
);
3473 if (super
->active
->openflag
)
3475 if (!best
|| seq
> bestseq
) {
3483 /* OK, load this ddf */
3484 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
3485 dfd
= dev_open(nm
, O_RDONLY
);
3488 load_ddf_headers(dfd
, super
, NULL
);
3489 load_ddf_global(dfd
, super
, NULL
);
3491 /* Now we need the device-local bits */
3492 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3495 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3496 dfd
= dev_open(nm
, O_RDWR
);
3499 rv
= load_ddf_headers(dfd
, super
, NULL
);
3501 rv
= load_ddf_local(dfd
, super
, NULL
, 1);
3507 if (st
->ss
== NULL
) {
3508 st
->ss
= &super_ddf
;
3509 st
->minor_version
= 0;
3512 strcpy(st
->container_devnm
, fd2devnm(fd
));
3516 static int load_container_ddf(struct supertype
*st
, int fd
,
3519 return load_super_ddf_all(st
, fd
, &st
->sb
, devname
);
3522 #endif /* MDASSEMBLE */
3524 static int check_secondary(const struct vcl
*vc
)
3526 const struct vd_config
*conf
= &vc
->conf
;
3529 /* The only DDF secondary RAID level md can support is
3530 * RAID 10, if the stripe sizes and Basic volume sizes
3532 * Other configurations could in theory be supported by exposing
3533 * the BVDs to user space and using device mapper for the secondary
3534 * mapping. So far we don't support that.
3537 __u64 sec_elements
[4] = {0, 0, 0, 0};
3538 #define __set_sec_seen(n) (sec_elements[(n)>>6] |= (1<<((n)&63)))
3539 #define __was_sec_seen(n) ((sec_elements[(n)>>6] & (1<<((n)&63))) != 0)
3541 if (vc
->other_bvds
== NULL
) {
3542 pr_err("No BVDs for secondary RAID found\n");
3545 if (conf
->prl
!= DDF_RAID1
) {
3546 pr_err("Secondary RAID level only supported for mirrored BVD\n");
3549 if (conf
->srl
!= DDF_2STRIPED
&& conf
->srl
!= DDF_2SPANNED
) {
3550 pr_err("Secondary RAID level %d is unsupported\n",
3554 __set_sec_seen(conf
->sec_elmnt_seq
);
3555 for (i
= 0; i
< conf
->sec_elmnt_count
-1; i
++) {
3556 const struct vd_config
*bvd
= vc
->other_bvds
[i
];
3557 if (bvd
->sec_elmnt_seq
== DDF_UNUSED_BVD
)
3559 if (bvd
->srl
!= conf
->srl
) {
3560 pr_err("Inconsistent secondary RAID level across BVDs\n");
3563 if (bvd
->prl
!= conf
->prl
) {
3564 pr_err("Different RAID levels for BVDs are unsupported\n");
3567 if (!be16_eq(bvd
->prim_elmnt_count
, conf
->prim_elmnt_count
)) {
3568 pr_err("All BVDs must have the same number of primary elements\n");
3571 if (bvd
->chunk_shift
!= conf
->chunk_shift
) {
3572 pr_err("Different strip sizes for BVDs are unsupported\n");
3575 if (!be64_eq(bvd
->array_blocks
, conf
->array_blocks
)) {
3576 pr_err("Different BVD sizes are unsupported\n");
3579 __set_sec_seen(bvd
->sec_elmnt_seq
);
3581 for (i
= 0; i
< conf
->sec_elmnt_count
; i
++) {
3582 if (!__was_sec_seen(i
)) {
3583 pr_err("BVD %d is missing\n", i
);
3590 static unsigned int get_pd_index_from_refnum(const struct vcl
*vc
,
3591 be32 refnum
, unsigned int nmax
,
3592 const struct vd_config
**bvd
,
3595 unsigned int i
, j
, n
, sec
, cnt
;
3597 cnt
= be16_to_cpu(vc
->conf
.prim_elmnt_count
);
3598 sec
= (vc
->conf
.sec_elmnt_count
== 1 ? 0 : vc
->conf
.sec_elmnt_seq
);
3600 for (i
= 0, j
= 0 ; i
< nmax
; i
++) {
3601 /* j counts valid entries for this BVD */
3602 if (be32_to_cpu(vc
->conf
.phys_refnum
[i
]) != 0xffffffff)
3604 if (be32_eq(vc
->conf
.phys_refnum
[i
], refnum
)) {
3607 return sec
* cnt
+ j
- 1;
3610 if (vc
->other_bvds
== NULL
)
3613 for (n
= 1; n
< vc
->conf
.sec_elmnt_count
; n
++) {
3614 struct vd_config
*vd
= vc
->other_bvds
[n
-1];
3615 sec
= vd
->sec_elmnt_seq
;
3616 if (sec
== DDF_UNUSED_BVD
)
3618 for (i
= 0, j
= 0 ; i
< nmax
; i
++) {
3619 if (be32_to_cpu(vd
->phys_refnum
[i
]) != 0xffffffff)
3621 if (be32_eq(vd
->phys_refnum
[i
], refnum
)) {
3624 return sec
* cnt
+ j
- 1;
3630 return DDF_NOTFOUND
;
3633 static struct mdinfo
*container_content_ddf(struct supertype
*st
, char *subarray
)
3635 /* Given a container loaded by load_super_ddf_all,
3636 * extract information about all the arrays into
3639 * For each vcl in conflist: create an mdinfo, fill it in,
3640 * then look for matching devices (phys_refnum) in dlist
3641 * and create appropriate device mdinfo.
3643 struct ddf_super
*ddf
= st
->sb
;
3644 struct mdinfo
*rest
= NULL
;
3647 for (vc
= ddf
->conflist
; vc
; vc
=vc
->next
)
3651 struct mdinfo
*this;
3657 (strtoul(subarray
, &ep
, 10) != vc
->vcnum
||
3661 if (vc
->conf
.sec_elmnt_count
> 1) {
3662 if (check_secondary(vc
) != 0)
3666 this = xcalloc(1, sizeof(*this));
3670 if (layout_ddf2md(&vc
->conf
, &this->array
))
3672 this->array
.md_minor
= -1;
3673 this->array
.major_version
= -1;
3674 this->array
.minor_version
= -2;
3675 this->safe_mode_delay
= DDF_SAFE_MODE_DELAY
;
3676 cptr
= (__u32
*)(vc
->conf
.guid
+ 16);
3677 this->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
3678 this->array
.utime
= DECADE
+
3679 be32_to_cpu(vc
->conf
.timestamp
);
3680 this->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
3683 if ((ddf
->virt
->entries
[i
].state
& DDF_state_inconsistent
) ||
3684 (ddf
->virt
->entries
[i
].init_state
& DDF_initstate_mask
) !=
3686 this->array
.state
= 0;
3687 this->resync_start
= 0;
3689 this->array
.state
= 1;
3690 this->resync_start
= MaxSector
;
3692 memcpy(this->name
, ddf
->virt
->entries
[i
].name
, 16);
3695 if (this->name
[j
] == ' ')
3698 memset(this->uuid
, 0, sizeof(this->uuid
));
3699 this->component_size
= be64_to_cpu(vc
->conf
.blocks
);
3700 this->array
.size
= this->component_size
/ 2;
3701 this->container_member
= i
;
3703 ddf
->currentconf
= vc
;
3704 uuid_from_super_ddf(st
, this->uuid
);
3706 ddf
->currentconf
= NULL
;
3708 sprintf(this->text_version
, "/%s/%d",
3709 st
->container_devnm
, this->container_member
);
3711 for (pd
= 0; pd
< be16_to_cpu(ddf
->phys
->used_pdes
); pd
++) {
3714 const struct vd_config
*bvd
;
3718 if (be32_to_cpu(ddf
->phys
->entries
[pd
].refnum
)
3722 stt
= be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
3723 if ((stt
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
3727 i
= get_pd_index_from_refnum(
3728 vc
, ddf
->phys
->entries
[pd
].refnum
,
3729 ddf
->mppe
, &bvd
, &iphys
);
3730 if (i
== DDF_NOTFOUND
)
3733 this->array
.working_disks
++;
3735 for (d
= ddf
->dlist
; d
; d
=d
->next
)
3736 if (be32_eq(d
->disk
.refnum
,
3737 ddf
->phys
->entries
[pd
].refnum
))
3740 /* Haven't found that one yet, maybe there are others */
3743 dev
= xcalloc(1, sizeof(*dev
));
3744 dev
->next
= this->devs
;
3747 dev
->disk
.number
= be32_to_cpu(d
->disk
.refnum
);
3748 dev
->disk
.major
= d
->major
;
3749 dev
->disk
.minor
= d
->minor
;
3750 dev
->disk
.raid_disk
= i
;
3751 dev
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
3752 dev
->recovery_start
= MaxSector
;
3754 dev
->events
= be32_to_cpu(ddf
->primary
.seq
);
3756 be64_to_cpu(LBA_OFFSET(ddf
, bvd
)[iphys
]);
3757 dev
->component_size
= be64_to_cpu(bvd
->blocks
);
3759 strcpy(dev
->name
, d
->devname
);
3765 static int store_super_ddf(struct supertype
*st
, int fd
)
3767 struct ddf_super
*ddf
= st
->sb
;
3768 unsigned long long dsize
;
3775 if (!get_dev_size(fd
, NULL
, &dsize
))
3778 if (ddf
->dlist
|| ddf
->conflist
) {
3783 if (fstat(fd
, &sta
) == -1 || !S_ISBLK(sta
.st_mode
)) {
3784 pr_err("%s: file descriptor for invalid device\n",
3788 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
3789 if (dl
->major
== (int)major(sta
.st_rdev
) &&
3790 dl
->minor
== (int)minor(sta
.st_rdev
))
3793 pr_err("%s: couldn't find disk %d/%d\n", __func__
,
3794 (int)major(sta
.st_rdev
),
3795 (int)minor(sta
.st_rdev
));
3800 ret
= (_write_super_to_disk(ddf
, dl
) != 1);
3805 if (posix_memalign(&buf
, 512, 512) != 0)
3807 memset(buf
, 0, 512);
3809 lseek64(fd
, dsize
-512, 0);
3810 rc
= write(fd
, buf
, 512);
3817 static int compare_super_ddf(struct supertype
*st
, struct supertype
*tst
)
3821 * 0 same, or first was empty, and second was copied
3822 * 1 second had wrong number
3824 * 3 wrong other info
3826 struct ddf_super
*first
= st
->sb
;
3827 struct ddf_super
*second
= tst
->sb
;
3828 struct dl
*dl1
, *dl2
;
3829 struct vcl
*vl1
, *vl2
;
3830 unsigned int max_vds
, max_pds
, pd
, vd
;
3838 if (memcmp(first
->anchor
.guid
, second
->anchor
.guid
, DDF_GUID_LEN
) != 0)
3841 if (!be32_eq(first
->anchor
.seq
, second
->anchor
.seq
)) {
3842 dprintf("%s: sequence number mismatch %u/%u\n", __func__
,
3843 be32_to_cpu(first
->anchor
.seq
),
3844 be32_to_cpu(second
->anchor
.seq
));
3847 if (first
->max_part
!= second
->max_part
||
3848 !be16_eq(first
->phys
->used_pdes
, second
->phys
->used_pdes
) ||
3849 !be16_eq(first
->virt
->populated_vdes
,
3850 second
->virt
->populated_vdes
)) {
3851 dprintf("%s: PD/VD number mismatch\n", __func__
);
3855 max_pds
= be16_to_cpu(first
->phys
->used_pdes
);
3856 for (dl2
= second
->dlist
; dl2
; dl2
= dl2
->next
) {
3857 for (pd
= 0; pd
< max_pds
; pd
++)
3858 if (be32_eq(first
->phys
->entries
[pd
].refnum
,
3861 if (pd
== max_pds
) {
3862 dprintf("%s: no match for disk %08x\n", __func__
,
3863 be32_to_cpu(dl2
->disk
.refnum
));
3868 max_vds
= be16_to_cpu(first
->active
->max_vd_entries
);
3869 for (vl2
= second
->conflist
; vl2
; vl2
= vl2
->next
) {
3870 if (!be32_eq(vl2
->conf
.magic
, DDF_VD_CONF_MAGIC
))
3872 for (vd
= 0; vd
< max_vds
; vd
++)
3873 if (!memcmp(first
->virt
->entries
[vd
].guid
,
3874 vl2
->conf
.guid
, DDF_GUID_LEN
))
3876 if (vd
== max_vds
) {
3877 dprintf("%s: no match for VD config\n", __func__
);
3881 /* FIXME should I look at anything else? */
3884 At this point we are fairly sure that the meta data matches.
3885 But the new disk may contain additional local data.
3886 Add it to the super block.
3888 for (vl2
= second
->conflist
; vl2
; vl2
= vl2
->next
) {
3889 for (vl1
= first
->conflist
; vl1
; vl1
= vl1
->next
)
3890 if (!memcmp(vl1
->conf
.guid
, vl2
->conf
.guid
,
3894 if (vl1
->other_bvds
!= NULL
&&
3895 vl1
->conf
.sec_elmnt_seq
!=
3896 vl2
->conf
.sec_elmnt_seq
) {
3897 dprintf("%s: adding BVD %u\n", __func__
,
3898 vl2
->conf
.sec_elmnt_seq
);
3899 add_other_bvd(vl1
, &vl2
->conf
,
3900 first
->conf_rec_len
*512);
3905 if (posix_memalign((void **)&vl1
, 512,
3906 (first
->conf_rec_len
*512 +
3907 offsetof(struct vcl
, conf
))) != 0) {
3908 pr_err("%s could not allocate vcl buf\n",
3913 vl1
->next
= first
->conflist
;
3914 vl1
->block_sizes
= NULL
;
3915 memcpy(&vl1
->conf
, &vl2
->conf
, first
->conf_rec_len
*512);
3916 if (alloc_other_bvds(first
, vl1
) != 0) {
3917 pr_err("%s could not allocate other bvds\n",
3922 for (vd
= 0; vd
< max_vds
; vd
++)
3923 if (!memcmp(first
->virt
->entries
[vd
].guid
,
3924 vl1
->conf
.guid
, DDF_GUID_LEN
))
3927 dprintf("%s: added config for VD %u\n", __func__
, vl1
->vcnum
);
3928 first
->conflist
= vl1
;
3931 for (dl2
= second
->dlist
; dl2
; dl2
= dl2
->next
) {
3932 for (dl1
= first
->dlist
; dl1
; dl1
= dl1
->next
)
3933 if (be32_eq(dl1
->disk
.refnum
, dl2
->disk
.refnum
))
3938 if (posix_memalign((void **)&dl1
, 512,
3939 sizeof(*dl1
) + (first
->max_part
) * sizeof(dl1
->vlist
[0]))
3941 pr_err("%s could not allocate disk info buffer\n",
3945 memcpy(dl1
, dl2
, sizeof(*dl1
));
3946 dl1
->mdupdate
= NULL
;
3947 dl1
->next
= first
->dlist
;
3949 for (pd
= 0; pd
< max_pds
; pd
++)
3950 if (be32_eq(first
->phys
->entries
[pd
].refnum
,
3955 if (posix_memalign((void **)&dl1
->spare
, 512,
3956 first
->conf_rec_len
*512) != 0) {
3957 pr_err("%s could not allocate spare info buf\n",
3961 memcpy(dl1
->spare
, dl2
->spare
, first
->conf_rec_len
*512);
3963 for (vd
= 0 ; vd
< first
->max_part
; vd
++) {
3964 if (!dl2
->vlist
[vd
]) {
3965 dl1
->vlist
[vd
] = NULL
;
3968 for (vl1
= first
->conflist
; vl1
; vl1
= vl1
->next
) {
3969 if (!memcmp(vl1
->conf
.guid
,
3970 dl2
->vlist
[vd
]->conf
.guid
,
3973 dl1
->vlist
[vd
] = vl1
;
3977 dprintf("%s: added disk %d: %08x\n", __func__
, dl1
->pdnum
,
3978 be32_to_cpu(dl1
->disk
.refnum
));
3986 * A new array 'a' has been started which claims to be instance 'inst'
3987 * within container 'c'.
3988 * We need to confirm that the array matches the metadata in 'c' so
3989 * that we don't corrupt any metadata.
3991 static int ddf_open_new(struct supertype
*c
, struct active_array
*a
, char *inst
)
3993 struct ddf_super
*ddf
= c
->sb
;
3997 static const char faulty
[] = "faulty";
3999 if (all_ff(ddf
->virt
->entries
[n
].guid
)) {
4000 pr_err("%s: subarray %d doesn't exist\n", __func__
, n
);
4003 dprintf("%s: new subarray %d, GUID: %s\n", __func__
, n
,
4004 guid_str(ddf
->virt
->entries
[n
].guid
));
4005 for (dev
= a
->info
.devs
; dev
; dev
= dev
->next
) {
4006 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4007 if (dl
->major
== dev
->disk
.major
&&
4008 dl
->minor
== dev
->disk
.minor
)
4011 pr_err("%s: device %d/%d of subarray %d not found in meta data\n",
4012 __func__
, dev
->disk
.major
, dev
->disk
.minor
, n
);
4015 if ((be16_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].state
) &
4016 (DDF_Online
|DDF_Missing
|DDF_Failed
)) != DDF_Online
) {
4017 pr_err("%s: new subarray %d contains broken device %d/%d (%02x)\n",
4018 __func__
, n
, dl
->major
, dl
->minor
,
4020 ddf
->phys
->entries
[dl
->pdnum
].state
));
4021 if (write(dev
->state_fd
, faulty
, sizeof(faulty
)-1) !=
4023 pr_err("Write to state_fd failed\n");
4024 dev
->curr_state
= DS_FAULTY
;
4027 a
->info
.container_member
= n
;
4032 * The array 'a' is to be marked clean in the metadata.
4033 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
4034 * clean up to the point (in sectors). If that cannot be recorded in the
4035 * metadata, then leave it as dirty.
4037 * For DDF, we need to clear the DDF_state_inconsistent bit in the
4038 * !global! virtual_disk.virtual_entry structure.
4040 static int ddf_set_array_state(struct active_array
*a
, int consistent
)
4042 struct ddf_super
*ddf
= a
->container
->sb
;
4043 int inst
= a
->info
.container_member
;
4044 int old
= ddf
->virt
->entries
[inst
].state
;
4045 if (consistent
== 2) {
4046 /* Should check if a recovery should be started FIXME */
4048 if (!is_resync_complete(&a
->info
))
4052 ddf
->virt
->entries
[inst
].state
&= ~DDF_state_inconsistent
;
4054 ddf
->virt
->entries
[inst
].state
|= DDF_state_inconsistent
;
4055 if (old
!= ddf
->virt
->entries
[inst
].state
)
4056 ddf_set_updates_pending(ddf
);
4058 old
= ddf
->virt
->entries
[inst
].init_state
;
4059 ddf
->virt
->entries
[inst
].init_state
&= ~DDF_initstate_mask
;
4060 if (is_resync_complete(&a
->info
))
4061 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_full
;
4062 else if (a
->info
.resync_start
== 0)
4063 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_not
;
4065 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_quick
;
4066 if (old
!= ddf
->virt
->entries
[inst
].init_state
)
4067 ddf_set_updates_pending(ddf
);
4069 dprintf("ddf mark %d/%s (%d) %s %llu\n", inst
,
4070 guid_str(ddf
->virt
->entries
[inst
].guid
), a
->curr_state
,
4071 consistent
?"clean":"dirty",
4072 a
->info
.resync_start
);
4076 static int get_bvd_state(const struct ddf_super
*ddf
,
4077 const struct vd_config
*vc
)
4079 unsigned int i
, n_bvd
, working
= 0;
4080 unsigned int n_prim
= be16_to_cpu(vc
->prim_elmnt_count
);
4082 for (i
= 0; i
< n_prim
; i
++) {
4083 if (!find_index_in_bvd(ddf
, vc
, i
, &n_bvd
))
4085 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
4088 st
= be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
4089 if ((st
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
4094 state
= DDF_state_degraded
;
4095 if (working
== n_prim
)
4096 state
= DDF_state_optimal
;
4102 state
= DDF_state_failed
;
4106 state
= DDF_state_failed
;
4107 else if (working
>= 2)
4108 state
= DDF_state_part_optimal
;
4112 if (working
< n_prim
- 1)
4113 state
= DDF_state_failed
;
4116 if (working
< n_prim
- 2)
4117 state
= DDF_state_failed
;
4118 else if (working
== n_prim
- 1)
4119 state
= DDF_state_part_optimal
;
4125 static int secondary_state(int state
, int other
, int seclevel
)
4127 if (state
== DDF_state_optimal
&& other
== DDF_state_optimal
)
4128 return DDF_state_optimal
;
4129 if (seclevel
== DDF_2MIRRORED
) {
4130 if (state
== DDF_state_optimal
|| other
== DDF_state_optimal
)
4131 return DDF_state_part_optimal
;
4132 if (state
== DDF_state_failed
&& other
== DDF_state_failed
)
4133 return DDF_state_failed
;
4134 return DDF_state_degraded
;
4136 if (state
== DDF_state_failed
|| other
== DDF_state_failed
)
4137 return DDF_state_failed
;
4138 if (state
== DDF_state_degraded
|| other
== DDF_state_degraded
)
4139 return DDF_state_degraded
;
4140 return DDF_state_part_optimal
;
4144 static int get_svd_state(const struct ddf_super
*ddf
, const struct vcl
*vcl
)
4146 int state
= get_bvd_state(ddf
, &vcl
->conf
);
4148 for (i
= 1; i
< vcl
->conf
.sec_elmnt_count
; i
++) {
4149 state
= secondary_state(
4151 get_bvd_state(ddf
, vcl
->other_bvds
[i
-1]),
4158 * The state of each disk is stored in the global phys_disk structure
4159 * in phys_disk.entries[n].state.
4160 * This makes various combinations awkward.
4161 * - When a device fails in any array, it must be failed in all arrays
4162 * that include a part of this device.
4163 * - When a component is rebuilding, we cannot include it officially in the
4164 * array unless this is the only array that uses the device.
4166 * So: when transitioning:
4167 * Online -> failed, just set failed flag. monitor will propagate
4168 * spare -> online, the device might need to be added to the array.
4169 * spare -> failed, just set failed. Don't worry if in array or not.
4171 static void ddf_set_disk(struct active_array
*a
, int n
, int state
)
4173 struct ddf_super
*ddf
= a
->container
->sb
;
4174 unsigned int inst
= a
->info
.container_member
, n_bvd
;
4176 struct vd_config
*vc
= find_vdcr(ddf
, inst
, (unsigned int)n
,
4182 dprintf("%s: %d to %x\n", __func__
, n
, state
);
4184 dprintf("ddf: cannot find instance %d!!\n", inst
);
4187 /* Find the matching slot in 'info'. */
4188 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
4189 if (mdi
->disk
.raid_disk
== n
)
4192 pr_err("%s: cannot find raid disk %d\n",
4197 /* and find the 'dl' entry corresponding to that. */
4198 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4199 if (mdi
->state_fd
>= 0 &&
4200 mdi
->disk
.major
== dl
->major
&&
4201 mdi
->disk
.minor
== dl
->minor
)
4204 pr_err("%s: cannot find raid disk %d (%d/%d)\n",
4206 mdi
->disk
.major
, mdi
->disk
.minor
);
4210 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
4211 if (pd
< 0 || pd
!= dl
->pdnum
) {
4212 /* disk doesn't currently exist or has changed.
4213 * If it is now in_sync, insert it. */
4214 dprintf("%s: phys disk not found for %d: %d/%d ref %08x\n",
4215 __func__
, dl
->pdnum
, dl
->major
, dl
->minor
,
4216 be32_to_cpu(dl
->disk
.refnum
));
4217 dprintf("%s: array %u disk %u ref %08x pd %d\n",
4218 __func__
, inst
, n_bvd
,
4219 be32_to_cpu(vc
->phys_refnum
[n_bvd
]), pd
);
4220 if ((state
& DS_INSYNC
) && ! (state
& DS_FAULTY
)) {
4221 pd
= dl
->pdnum
; /* FIXME: is this really correct ? */
4222 vc
->phys_refnum
[n_bvd
] = dl
->disk
.refnum
;
4223 LBA_OFFSET(ddf
, vc
)[n_bvd
] =
4224 cpu_to_be64(mdi
->data_offset
);
4225 be16_clear(ddf
->phys
->entries
[pd
].type
,
4226 cpu_to_be16(DDF_Global_Spare
));
4227 be16_set(ddf
->phys
->entries
[pd
].type
,
4228 cpu_to_be16(DDF_Active_in_VD
));
4229 ddf_set_updates_pending(ddf
);
4232 be16 old
= ddf
->phys
->entries
[pd
].state
;
4233 if (state
& DS_FAULTY
)
4234 be16_set(ddf
->phys
->entries
[pd
].state
,
4235 cpu_to_be16(DDF_Failed
));
4236 if (state
& DS_INSYNC
) {
4237 be16_set(ddf
->phys
->entries
[pd
].state
,
4238 cpu_to_be16(DDF_Online
));
4239 be16_clear(ddf
->phys
->entries
[pd
].state
,
4240 cpu_to_be16(DDF_Rebuilding
));
4242 if (!be16_eq(old
, ddf
->phys
->entries
[pd
].state
))
4243 ddf_set_updates_pending(ddf
);
4246 dprintf("ddf: set_disk %d (%08x) to %x->%02x\n", n
,
4247 be32_to_cpu(dl
->disk
.refnum
), state
,
4248 be16_to_cpu(ddf
->phys
->entries
[pd
].state
));
4250 /* Now we need to check the state of the array and update
4251 * virtual_disk.entries[n].state.
4252 * It needs to be one of "optimal", "degraded", "failed".
4253 * I don't understand 'deleted' or 'missing'.
4255 state
= get_svd_state(ddf
, vcl
);
4257 if (ddf
->virt
->entries
[inst
].state
!=
4258 ((ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4261 ddf
->virt
->entries
[inst
].state
=
4262 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4264 ddf_set_updates_pending(ddf
);
4269 static void ddf_sync_metadata(struct supertype
*st
)
4273 * Write all data to all devices.
4274 * Later, we might be able to track whether only local changes
4275 * have been made, or whether any global data has been changed,
4276 * but ddf is sufficiently weird that it probably always
4277 * changes global data ....
4279 struct ddf_super
*ddf
= st
->sb
;
4280 if (!ddf
->updates_pending
)
4282 ddf
->updates_pending
= 0;
4283 __write_init_super_ddf(st
);
4284 dprintf("ddf: sync_metadata\n");
4287 static int del_from_conflist(struct vcl
**list
, const char *guid
)
4291 for (p
= list
; p
&& *p
; p
= &((*p
)->next
))
4292 if (!memcmp((*p
)->conf
.guid
, guid
, DDF_GUID_LEN
)) {
4299 static int _kill_subarray_ddf(struct ddf_super
*ddf
, const char *guid
)
4302 unsigned int vdnum
, i
;
4303 vdnum
= find_vde_by_guid(ddf
, guid
);
4304 if (vdnum
== DDF_NOTFOUND
) {
4305 pr_err("%s: could not find VD %s\n", __func__
,
4309 if (del_from_conflist(&ddf
->conflist
, guid
) == 0) {
4310 pr_err("%s: could not find conf %s\n", __func__
,
4314 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4315 for (i
= 0; i
< ddf
->max_part
; i
++)
4316 if (dl
->vlist
[i
] != NULL
&&
4317 !memcmp(dl
->vlist
[i
]->conf
.guid
, guid
,
4319 dl
->vlist
[i
] = NULL
;
4320 memset(ddf
->virt
->entries
[vdnum
].guid
, 0xff, DDF_GUID_LEN
);
4321 dprintf("%s: deleted %s\n", __func__
, guid_str(guid
));
4325 static int kill_subarray_ddf(struct supertype
*st
)
4327 struct ddf_super
*ddf
= st
->sb
;
4329 * currentconf is set in container_content_ddf,
4330 * called with subarray arg
4332 struct vcl
*victim
= ddf
->currentconf
;
4333 struct vd_config
*conf
;
4334 ddf
->currentconf
= NULL
;
4337 pr_err("%s: nothing to kill\n", __func__
);
4340 conf
= &victim
->conf
;
4341 vdnum
= find_vde_by_guid(ddf
, conf
->guid
);
4342 if (vdnum
== DDF_NOTFOUND
) {
4343 pr_err("%s: could not find VD %s\n", __func__
,
4344 guid_str(conf
->guid
));
4347 if (st
->update_tail
) {
4348 struct virtual_disk
*vd
;
4349 int len
= sizeof(struct virtual_disk
)
4350 + sizeof(struct virtual_entry
);
4353 pr_err("%s: failed to allocate %d bytes\n", __func__
,
4357 memset(vd
, 0 , len
);
4358 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
4359 vd
->populated_vdes
= cpu_to_be16(0);
4360 memcpy(vd
->entries
[0].guid
, conf
->guid
, DDF_GUID_LEN
);
4361 /* we use DDF_state_deleted as marker */
4362 vd
->entries
[0].state
= DDF_state_deleted
;
4363 append_metadata_update(st
, vd
, len
);
4365 _kill_subarray_ddf(ddf
, conf
->guid
);
4366 ddf_set_updates_pending(ddf
);
4367 ddf_sync_metadata(st
);
4372 static void copy_matching_bvd(struct ddf_super
*ddf
,
4373 struct vd_config
*conf
,
4374 const struct metadata_update
*update
)
4377 be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
4378 unsigned int len
= ddf
->conf_rec_len
* 512;
4380 struct vd_config
*vc
;
4381 for (p
= update
->buf
; p
< update
->buf
+ update
->len
; p
+= len
) {
4382 vc
= (struct vd_config
*) p
;
4383 if (vc
->sec_elmnt_seq
== conf
->sec_elmnt_seq
) {
4384 memcpy(conf
->phys_refnum
, vc
->phys_refnum
,
4385 mppe
* (sizeof(__u32
) + sizeof(__u64
)));
4389 pr_err("%s: no match for BVD %d of %s in update\n", __func__
,
4390 conf
->sec_elmnt_seq
, guid_str(conf
->guid
));
4393 static void ddf_process_update(struct supertype
*st
,
4394 struct metadata_update
*update
)
4396 /* Apply this update to the metadata.
4397 * The first 4 bytes are a DDF_*_MAGIC which guides
4399 * Possible update are:
4400 * DDF_PHYS_RECORDS_MAGIC
4401 * Add a new physical device or remove an old one.
4402 * Changes to this record only happen implicitly.
4403 * used_pdes is the device number.
4404 * DDF_VIRT_RECORDS_MAGIC
4405 * Add a new VD. Possibly also change the 'access' bits.
4406 * populated_vdes is the entry number.
4408 * New or updated VD. the VIRT_RECORD must already
4409 * exist. For an update, phys_refnum and lba_offset
4410 * (at least) are updated, and the VD_CONF must
4411 * be written to precisely those devices listed with
4413 * DDF_SPARE_ASSIGN_MAGIC
4414 * replacement Spare Assignment Record... but for which device?
4417 * - to create a new array, we send a VIRT_RECORD and
4418 * a VD_CONF. Then assemble and start the array.
4419 * - to activate a spare we send a VD_CONF to add the phys_refnum
4420 * and offset. This will also mark the spare as active with
4421 * a spare-assignment record.
4423 struct ddf_super
*ddf
= st
->sb
;
4424 be32
*magic
= (be32
*)update
->buf
;
4425 struct phys_disk
*pd
;
4426 struct virtual_disk
*vd
;
4427 struct vd_config
*vc
;
4431 unsigned int pdnum
, pd2
, len
;
4433 dprintf("Process update %x\n", be32_to_cpu(*magic
));
4435 if (be32_eq(*magic
, DDF_PHYS_RECORDS_MAGIC
)) {
4437 if (update
->len
!= (sizeof(struct phys_disk
) +
4438 sizeof(struct phys_disk_entry
)))
4440 pd
= (struct phys_disk
*)update
->buf
;
4442 ent
= be16_to_cpu(pd
->used_pdes
);
4443 if (ent
>= be16_to_cpu(ddf
->phys
->max_pdes
))
4445 if (be16_and(pd
->entries
[0].state
, cpu_to_be16(DDF_Missing
))) {
4447 /* removing this disk. */
4448 be16_set(ddf
->phys
->entries
[ent
].state
,
4449 cpu_to_be16(DDF_Missing
));
4450 for (dlp
= &ddf
->dlist
; *dlp
; dlp
= &(*dlp
)->next
) {
4451 struct dl
*dl
= *dlp
;
4452 if (dl
->pdnum
== (signed)ent
) {
4455 /* FIXME this doesn't free
4462 ddf_set_updates_pending(ddf
);
4465 if (!all_ff(ddf
->phys
->entries
[ent
].guid
))
4467 ddf
->phys
->entries
[ent
] = pd
->entries
[0];
4468 ddf
->phys
->used_pdes
= cpu_to_be16
4469 (1 + be16_to_cpu(ddf
->phys
->used_pdes
));
4470 ddf_set_updates_pending(ddf
);
4471 if (ddf
->add_list
) {
4472 struct active_array
*a
;
4473 struct dl
*al
= ddf
->add_list
;
4474 ddf
->add_list
= al
->next
;
4476 al
->next
= ddf
->dlist
;
4479 /* As a device has been added, we should check
4480 * for any degraded devices that might make
4481 * use of this spare */
4482 for (a
= st
->arrays
; a
; a
=a
->next
)
4483 a
->check_degraded
= 1;
4485 } else if (be32_eq(*magic
, DDF_VIRT_RECORDS_MAGIC
)) {
4487 if (update
->len
!= (sizeof(struct virtual_disk
) +
4488 sizeof(struct virtual_entry
)))
4490 vd
= (struct virtual_disk
*)update
->buf
;
4492 if (vd
->entries
[0].state
== DDF_state_deleted
) {
4493 if (_kill_subarray_ddf(ddf
, vd
->entries
[0].guid
))
4497 ent
= find_vde_by_guid(ddf
, vd
->entries
[0].guid
);
4498 if (ent
!= DDF_NOTFOUND
) {
4499 dprintf("%s: VD %s exists already in slot %d\n",
4500 __func__
, guid_str(vd
->entries
[0].guid
),
4504 ent
= find_unused_vde(ddf
);
4505 if (ent
== DDF_NOTFOUND
)
4507 ddf
->virt
->entries
[ent
] = vd
->entries
[0];
4508 ddf
->virt
->populated_vdes
=
4511 ddf
->virt
->populated_vdes
));
4512 dprintf("%s: added VD %s in slot %d(s=%02x i=%02x)\n",
4513 __func__
, guid_str(vd
->entries
[0].guid
), ent
,
4514 ddf
->virt
->entries
[ent
].state
,
4515 ddf
->virt
->entries
[ent
].init_state
);
4517 ddf_set_updates_pending(ddf
);
4520 else if (be32_eq(*magic
, DDF_VD_CONF_MAGIC
)) {
4521 vc
= (struct vd_config
*)update
->buf
;
4522 len
= ddf
->conf_rec_len
* 512;
4523 if ((unsigned int)update
->len
!= len
* vc
->sec_elmnt_count
) {
4524 pr_err("%s: %s: insufficient data (%d) for %u BVDs\n",
4525 __func__
, guid_str(vc
->guid
), update
->len
,
4526 vc
->sec_elmnt_count
);
4529 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
4530 if (memcmp(vcl
->conf
.guid
, vc
->guid
, DDF_GUID_LEN
) == 0)
4532 dprintf("%s: conf update for %s (%s)\n", __func__
,
4533 guid_str(vc
->guid
), (vcl
? "old" : "new"));
4535 /* An update, just copy the phys_refnum and lba_offset
4540 copy_matching_bvd(ddf
, &vcl
->conf
, update
);
4541 for (k
= 0; k
< be16_to_cpu(vc
->prim_elmnt_count
); k
++)
4542 dprintf("BVD %u has %08x at %llu\n", 0,
4543 be32_to_cpu(vcl
->conf
.phys_refnum
[k
]),
4544 be64_to_cpu(LBA_OFFSET(ddf
,
4546 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++) {
4547 copy_matching_bvd(ddf
, vcl
->other_bvds
[i
-1],
4549 for (k
= 0; k
< be16_to_cpu(
4550 vc
->prim_elmnt_count
); k
++)
4551 dprintf("BVD %u has %08x at %llu\n", i
,
4553 (vcl
->other_bvds
[i
-1]->
4558 vcl
->other_bvds
[i
-1])[k
]));
4565 vcl
= update
->space
;
4566 update
->space
= NULL
;
4567 vcl
->next
= ddf
->conflist
;
4568 memcpy(&vcl
->conf
, vc
, len
);
4569 ent
= find_vde_by_guid(ddf
, vc
->guid
);
4570 if (ent
== DDF_NOTFOUND
)
4573 ddf
->conflist
= vcl
;
4574 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++)
4575 memcpy(vcl
->other_bvds
[i
-1],
4576 update
->buf
+ len
* i
, len
);
4578 /* Set DDF_Transition on all Failed devices - to help
4579 * us detect those that are no longer in use
4581 for (pdnum
= 0; pdnum
< be16_to_cpu(ddf
->phys
->used_pdes
);
4583 if (be16_and(ddf
->phys
->entries
[pdnum
].state
,
4584 cpu_to_be16(DDF_Failed
)))
4585 be16_set(ddf
->phys
->entries
[pdnum
].state
,
4586 cpu_to_be16(DDF_Transition
));
4587 /* Now make sure vlist is correct for each dl. */
4588 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
4589 unsigned int vn
= 0;
4590 int in_degraded
= 0;
4591 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
) {
4592 unsigned int dn
, ibvd
;
4593 const struct vd_config
*conf
;
4595 dn
= get_pd_index_from_refnum(vcl
,
4599 if (dn
== DDF_NOTFOUND
)
4601 dprintf("dev %d/%08x has %s (sec=%u) at %d\n",
4603 be32_to_cpu(dl
->disk
.refnum
),
4604 guid_str(conf
->guid
),
4605 conf
->sec_elmnt_seq
, vn
);
4606 /* Clear the Transition flag */
4608 (ddf
->phys
->entries
[dl
->pdnum
].state
,
4609 cpu_to_be16(DDF_Failed
)))
4610 be16_clear(ddf
->phys
4611 ->entries
[dl
->pdnum
].state
,
4612 cpu_to_be16(DDF_Transition
));
4613 dl
->vlist
[vn
++] = vcl
;
4614 vstate
= ddf
->virt
->entries
[vcl
->vcnum
].state
4616 if (vstate
== DDF_state_degraded
||
4617 vstate
== DDF_state_part_optimal
)
4620 while (vn
< ddf
->max_part
)
4621 dl
->vlist
[vn
++] = NULL
;
4623 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4624 cpu_to_be16(DDF_Global_Spare
));
4625 if (!be16_and(ddf
->phys
4626 ->entries
[dl
->pdnum
].type
,
4627 cpu_to_be16(DDF_Active_in_VD
))) {
4629 ->entries
[dl
->pdnum
].type
,
4630 cpu_to_be16(DDF_Active_in_VD
));
4633 ->entries
[dl
->pdnum
]
4640 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4641 cpu_to_be16(DDF_Global_Spare
));
4642 be16_set(ddf
->phys
->entries
[dl
->pdnum
].type
,
4643 cpu_to_be16(DDF_Spare
));
4645 if (!dl
->vlist
[0] && !dl
->spare
) {
4646 be16_set(ddf
->phys
->entries
[dl
->pdnum
].type
,
4647 cpu_to_be16(DDF_Global_Spare
));
4648 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4649 cpu_to_be16(DDF_Spare
));
4650 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4651 cpu_to_be16(DDF_Active_in_VD
));
4655 /* Now remove any 'Failed' devices that are not part
4656 * of any VD. They will have the Transition flag set.
4657 * Once done, we need to update all dl->pdnum numbers.
4660 for (pdnum
= 0; pdnum
< be16_to_cpu(ddf
->phys
->used_pdes
);
4662 if (be16_and(ddf
->phys
->entries
[pdnum
].state
,
4663 cpu_to_be16(DDF_Failed
))
4664 && be16_and(ddf
->phys
->entries
[pdnum
].state
,
4665 cpu_to_be16(DDF_Transition
))) {
4666 /* skip this one unless in dlist*/
4667 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4668 if (dl
->pdnum
== (int)pdnum
)
4676 ddf
->phys
->entries
[pd2
] =
4677 ddf
->phys
->entries
[pdnum
];
4678 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4679 if (dl
->pdnum
== (int)pdnum
)
4684 ddf
->phys
->used_pdes
= cpu_to_be16(pd2
);
4685 while (pd2
< pdnum
) {
4686 memset(ddf
->phys
->entries
[pd2
].guid
, 0xff,
4691 ddf_set_updates_pending(ddf
);
4693 /* case DDF_SPARE_ASSIGN_MAGIC */
4696 static void ddf_prepare_update(struct supertype
*st
,
4697 struct metadata_update
*update
)
4699 /* This update arrived at managemon.
4700 * We are about to pass it to monitor.
4701 * If a malloc is needed, do it here.
4703 struct ddf_super
*ddf
= st
->sb
;
4704 be32
*magic
= (be32
*)update
->buf
;
4705 if (be32_eq(*magic
, DDF_VD_CONF_MAGIC
)) {
4707 struct vd_config
*conf
= (struct vd_config
*) update
->buf
;
4708 if (posix_memalign(&update
->space
, 512,
4709 offsetof(struct vcl
, conf
)
4710 + ddf
->conf_rec_len
* 512) != 0) {
4711 update
->space
= NULL
;
4714 vcl
= update
->space
;
4715 vcl
->conf
.sec_elmnt_count
= conf
->sec_elmnt_count
;
4716 if (alloc_other_bvds(ddf
, vcl
) != 0) {
4717 free(update
->space
);
4718 update
->space
= NULL
;
4724 * Check degraded state of a RAID10.
4725 * returns 2 for good, 1 for degraded, 0 for failed, and -1 for error
4727 static int raid10_degraded(struct mdinfo
*info
)
4735 n_prim
= info
->array
.layout
& ~0x100;
4736 n_bvds
= info
->array
.raid_disks
/ n_prim
;
4737 found
= xmalloc(n_bvds
);
4740 memset(found
, 0, n_bvds
);
4741 for (d
= info
->devs
; d
; d
= d
->next
) {
4742 i
= d
->disk
.raid_disk
/ n_prim
;
4744 pr_err("%s: BUG: invalid raid disk\n", __func__
);
4747 if (d
->state_fd
> 0)
4751 for (i
= 0; i
< n_bvds
; i
++)
4753 dprintf("%s: BVD %d/%d failed\n", __func__
, i
, n_bvds
);
4756 } else if (found
[i
] < n_prim
) {
4757 dprintf("%s: BVD %d/%d degraded\n", __func__
, i
,
4767 * Check if the array 'a' is degraded but not failed.
4768 * If it is, find as many spares as are available and needed and
4769 * arrange for their inclusion.
4770 * We only choose devices which are not already in the array,
4771 * and prefer those with a spare-assignment to this array.
4772 * otherwise we choose global spares - assuming always that
4773 * there is enough room.
4774 * For each spare that we assign, we return an 'mdinfo' which
4775 * describes the position for the device in the array.
4776 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
4777 * the new phys_refnum and lba_offset values.
4779 * Only worry about BVDs at the moment.
4781 static struct mdinfo
*ddf_activate_spare(struct active_array
*a
,
4782 struct metadata_update
**updates
)
4786 struct ddf_super
*ddf
= a
->container
->sb
;
4788 struct mdinfo
*rv
= NULL
;
4790 struct metadata_update
*mu
;
4795 struct vd_config
*vc
;
4798 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
4799 if ((d
->curr_state
& DS_FAULTY
) &&
4801 /* wait for Removal to happen */
4803 if (d
->state_fd
>= 0)
4807 dprintf("%s: working=%d (%d) level=%d\n", __func__
, working
,
4808 a
->info
.array
.raid_disks
,
4809 a
->info
.array
.level
);
4810 if (working
== a
->info
.array
.raid_disks
)
4811 return NULL
; /* array not degraded */
4812 switch (a
->info
.array
.level
) {
4815 return NULL
; /* failed */
4819 if (working
< a
->info
.array
.raid_disks
- 1)
4820 return NULL
; /* failed */
4823 if (working
< a
->info
.array
.raid_disks
- 2)
4824 return NULL
; /* failed */
4827 if (raid10_degraded(&a
->info
) < 1)
4830 default: /* concat or stripe */
4831 return NULL
; /* failed */
4834 /* For each slot, if it is not working, find a spare */
4836 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
4837 for (d
= a
->info
.devs
; d
; d
= d
->next
)
4838 if (d
->disk
.raid_disk
== i
)
4840 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
4841 if (d
&& (d
->state_fd
>= 0))
4844 /* OK, this device needs recovery. Find a spare */
4846 for ( ; dl
; dl
= dl
->next
) {
4847 unsigned long long esize
;
4848 unsigned long long pos
;
4851 int is_dedicated
= 0;
4854 be16 state
= ddf
->phys
->entries
[dl
->pdnum
].state
;
4856 cpu_to_be16(DDF_Failed
|DDF_Missing
)) ||
4858 cpu_to_be16(DDF_Online
)))
4861 /* If in this array, skip */
4862 for (d2
= a
->info
.devs
; d2
; d2
= d2
->next
)
4863 if (d2
->state_fd
>= 0 &&
4864 d2
->disk
.major
== dl
->major
&&
4865 d2
->disk
.minor
== dl
->minor
) {
4866 dprintf("%x:%x (%08x) already in array\n",
4867 dl
->major
, dl
->minor
,
4868 be32_to_cpu(dl
->disk
.refnum
));
4873 if (be16_and(ddf
->phys
->entries
[dl
->pdnum
].type
,
4874 cpu_to_be16(DDF_Spare
))) {
4875 /* Check spare assign record */
4877 if (dl
->spare
->type
& DDF_spare_dedicated
) {
4878 /* check spare_ents for guid */
4884 if (memcmp(dl
->spare
->spare_ents
[j
].guid
,
4885 ddf
->virt
->entries
[a
->info
.container_member
].guid
,
4892 } else if (be16_and(ddf
->phys
->entries
[dl
->pdnum
].type
,
4893 cpu_to_be16(DDF_Global_Spare
))) {
4895 } else if (!be16_and(ddf
->phys
4896 ->entries
[dl
->pdnum
].state
,
4897 cpu_to_be16(DDF_Failed
))) {
4898 /* we can possibly use some of this */
4901 if ( ! (is_dedicated
||
4902 (is_global
&& global_ok
))) {
4903 dprintf("%x:%x not suitable: %d %d\n", dl
->major
, dl
->minor
,
4904 is_dedicated
, is_global
);
4908 /* We are allowed to use this device - is there space?
4909 * We need a->info.component_size sectors */
4910 ex
= get_extents(ddf
, dl
);
4912 dprintf("cannot get extents\n");
4919 esize
= ex
[j
].start
- pos
;
4920 if (esize
>= a
->info
.component_size
)
4922 pos
= ex
[j
].start
+ ex
[j
].size
;
4924 } while (ex
[j
-1].size
);
4927 if (esize
< a
->info
.component_size
) {
4928 dprintf("%x:%x has no room: %llu %llu\n",
4929 dl
->major
, dl
->minor
,
4930 esize
, a
->info
.component_size
);
4935 /* Cool, we have a device with some space at pos */
4936 di
= xcalloc(1, sizeof(*di
));
4937 di
->disk
.number
= i
;
4938 di
->disk
.raid_disk
= i
;
4939 di
->disk
.major
= dl
->major
;
4940 di
->disk
.minor
= dl
->minor
;
4942 di
->recovery_start
= 0;
4943 di
->data_offset
= pos
;
4944 di
->component_size
= a
->info
.component_size
;
4945 di
->container_member
= dl
->pdnum
;
4948 dprintf("%x:%x (%08x) to be %d at %llu\n",
4949 dl
->major
, dl
->minor
,
4950 be32_to_cpu(dl
->disk
.refnum
), i
, pos
);
4954 if (!dl
&& ! global_ok
) {
4955 /* not enough dedicated spares, try global */
4963 /* No spares found */
4965 /* Now 'rv' has a list of devices to return.
4966 * Create a metadata_update record to update the
4967 * phys_refnum and lba_offset values
4969 vc
= find_vdcr(ddf
, a
->info
.container_member
, rv
->disk
.raid_disk
,
4974 mu
= xmalloc(sizeof(*mu
));
4975 if (posix_memalign(&mu
->space
, 512, sizeof(struct vcl
)) != 0) {
4980 mu
->len
= ddf
->conf_rec_len
* 512 * vcl
->conf
.sec_elmnt_count
;
4981 mu
->buf
= xmalloc(mu
->len
);
4983 mu
->space_list
= NULL
;
4984 mu
->next
= *updates
;
4985 memcpy(mu
->buf
, &vcl
->conf
, ddf
->conf_rec_len
* 512);
4986 for (j
= 1; j
< vcl
->conf
.sec_elmnt_count
; j
++)
4987 memcpy(mu
->buf
+ j
* ddf
->conf_rec_len
* 512,
4988 vcl
->other_bvds
[j
-1], ddf
->conf_rec_len
* 512);
4990 vc
= (struct vd_config
*)mu
->buf
;
4991 for (di
= rv
; di
; di
= di
->next
) {
4992 unsigned int i_sec
, i_prim
;
4993 i_sec
= di
->disk
.raid_disk
4994 / be16_to_cpu(vcl
->conf
.prim_elmnt_count
);
4995 i_prim
= di
->disk
.raid_disk
4996 % be16_to_cpu(vcl
->conf
.prim_elmnt_count
);
4997 vc
= (struct vd_config
*)(mu
->buf
4998 + i_sec
* ddf
->conf_rec_len
* 512);
4999 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
5000 if (dl
->major
== di
->disk
.major
5001 && dl
->minor
== di
->disk
.minor
)
5004 pr_err("%s: BUG: can't find disk %d (%d/%d)\n",
5005 __func__
, di
->disk
.raid_disk
,
5006 di
->disk
.major
, di
->disk
.minor
);
5009 vc
->phys_refnum
[i_prim
] = ddf
->phys
->entries
[dl
->pdnum
].refnum
;
5010 LBA_OFFSET(ddf
, vc
)[i_prim
] = cpu_to_be64(di
->data_offset
);
5011 dprintf("BVD %u gets %u: %08x at %llu\n", i_sec
, i_prim
,
5012 be32_to_cpu(vc
->phys_refnum
[i_prim
]),
5013 be64_to_cpu(LBA_OFFSET(ddf
, vc
)[i_prim
]));
5018 #endif /* MDASSEMBLE */
5020 static int ddf_level_to_layout(int level
)
5027 return ALGORITHM_LEFT_SYMMETRIC
;
5029 return ALGORITHM_ROTATING_N_CONTINUE
;
5037 static void default_geometry_ddf(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5039 if (level
&& *level
== UnSet
)
5040 *level
= LEVEL_CONTAINER
;
5042 if (level
&& layout
&& *layout
== UnSet
)
5043 *layout
= ddf_level_to_layout(*level
);
5046 struct superswitch super_ddf
= {
5048 .examine_super
= examine_super_ddf
,
5049 .brief_examine_super
= brief_examine_super_ddf
,
5050 .brief_examine_subarrays
= brief_examine_subarrays_ddf
,
5051 .export_examine_super
= export_examine_super_ddf
,
5052 .detail_super
= detail_super_ddf
,
5053 .brief_detail_super
= brief_detail_super_ddf
,
5054 .validate_geometry
= validate_geometry_ddf
,
5055 .write_init_super
= write_init_super_ddf
,
5056 .add_to_super
= add_to_super_ddf
,
5057 .remove_from_super
= remove_from_super_ddf
,
5058 .load_container
= load_container_ddf
,
5059 .copy_metadata
= copy_metadata_ddf
,
5060 .kill_subarray
= kill_subarray_ddf
,
5062 .match_home
= match_home_ddf
,
5063 .uuid_from_super
= uuid_from_super_ddf
,
5064 .getinfo_super
= getinfo_super_ddf
,
5065 .update_super
= update_super_ddf
,
5067 .avail_size
= avail_size_ddf
,
5069 .compare_super
= compare_super_ddf
,
5071 .load_super
= load_super_ddf
,
5072 .init_super
= init_super_ddf
,
5073 .store_super
= store_super_ddf
,
5074 .free_super
= free_super_ddf
,
5075 .match_metadata_desc
= match_metadata_desc_ddf
,
5076 .container_content
= container_content_ddf
,
5077 .default_geometry
= default_geometry_ddf
,
5083 .open_new
= ddf_open_new
,
5084 .set_array_state
= ddf_set_array_state
,
5085 .set_disk
= ddf_set_disk
,
5086 .sync_metadata
= ddf_sync_metadata
,
5087 .process_update
= ddf_process_update
,
5088 .prepare_update
= ddf_prepare_update
,
5089 .activate_spare
= ddf_activate_spare
,