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
;
883 } else if (devname
&&
884 be64_to_cpu(super
->anchor
.secondary_lba
) != ~(__u64
)0)
885 pr_err("Failed to load secondary DDF header on %s\n",
887 if (super
->active
== NULL
)
892 static int load_ddf_global(int fd
, struct ddf_super
*super
, char *devname
)
895 ok
= load_section(fd
, super
, &super
->controller
,
896 super
->active
->controller_section_offset
,
897 super
->active
->controller_section_length
,
899 super
->phys
= load_section(fd
, super
, NULL
,
900 super
->active
->phys_section_offset
,
901 super
->active
->phys_section_length
,
903 super
->pdsize
= be32_to_cpu(super
->active
->phys_section_length
) * 512;
905 super
->virt
= load_section(fd
, super
, NULL
,
906 super
->active
->virt_section_offset
,
907 super
->active
->virt_section_length
,
909 super
->vdsize
= be32_to_cpu(super
->active
->virt_section_length
) * 512;
919 super
->conflist
= NULL
;
922 super
->max_part
= be16_to_cpu(super
->active
->max_partitions
);
923 super
->mppe
= be16_to_cpu(super
->active
->max_primary_element_entries
);
924 super
->conf_rec_len
= be16_to_cpu(super
->active
->config_record_len
);
928 #define DDF_UNUSED_BVD 0xff
929 static int alloc_other_bvds(const struct ddf_super
*ddf
, struct vcl
*vcl
)
931 unsigned int n_vds
= vcl
->conf
.sec_elmnt_count
- 1;
932 unsigned int i
, vdsize
;
935 vcl
->other_bvds
= NULL
;
938 vdsize
= ddf
->conf_rec_len
* 512;
939 if (posix_memalign(&p
, 512, n_vds
*
940 (vdsize
+ sizeof(struct vd_config
*))) != 0)
942 vcl
->other_bvds
= (struct vd_config
**) (p
+ n_vds
* vdsize
);
943 for (i
= 0; i
< n_vds
; i
++) {
944 vcl
->other_bvds
[i
] = p
+ i
* vdsize
;
945 memset(vcl
->other_bvds
[i
], 0, vdsize
);
946 vcl
->other_bvds
[i
]->sec_elmnt_seq
= DDF_UNUSED_BVD
;
951 static void add_other_bvd(struct vcl
*vcl
, struct vd_config
*vd
,
955 for (i
= 0; i
< vcl
->conf
.sec_elmnt_count
-1; i
++)
956 if (vcl
->other_bvds
[i
]->sec_elmnt_seq
== vd
->sec_elmnt_seq
)
959 if (i
< vcl
->conf
.sec_elmnt_count
-1) {
960 if (be32_to_cpu(vd
->seqnum
) <=
961 be32_to_cpu(vcl
->other_bvds
[i
]->seqnum
))
964 for (i
= 0; i
< vcl
->conf
.sec_elmnt_count
-1; i
++)
965 if (vcl
->other_bvds
[i
]->sec_elmnt_seq
== DDF_UNUSED_BVD
)
967 if (i
== vcl
->conf
.sec_elmnt_count
-1) {
968 pr_err("no space for sec level config %u, count is %u\n",
969 vd
->sec_elmnt_seq
, vcl
->conf
.sec_elmnt_count
);
973 memcpy(vcl
->other_bvds
[i
], vd
, len
);
976 static int load_ddf_local(int fd
, struct ddf_super
*super
,
977 char *devname
, int keep
)
983 unsigned int confsec
;
985 unsigned int max_virt_disks
= be16_to_cpu
986 (super
->active
->max_vd_entries
);
987 unsigned long long dsize
;
989 /* First the local disk info */
990 if (posix_memalign((void**)&dl
, 512,
992 (super
->max_part
) * sizeof(dl
->vlist
[0])) != 0) {
993 pr_err("%s could not allocate disk info buffer\n",
998 load_section(fd
, super
, &dl
->disk
,
999 super
->active
->data_section_offset
,
1000 super
->active
->data_section_length
,
1002 dl
->devname
= devname
? xstrdup(devname
) : NULL
;
1005 dl
->major
= major(stb
.st_rdev
);
1006 dl
->minor
= minor(stb
.st_rdev
);
1007 dl
->next
= super
->dlist
;
1008 dl
->fd
= keep
? fd
: -1;
1011 if (get_dev_size(fd
, devname
, &dsize
))
1012 dl
->size
= dsize
>> 9;
1013 /* If the disks have different sizes, the LBAs will differ
1014 * between phys disks.
1015 * At this point here, the values in super->active must be valid
1016 * for this phys disk. */
1017 dl
->primary_lba
= super
->active
->primary_lba
;
1018 dl
->secondary_lba
= super
->active
->secondary_lba
;
1019 dl
->workspace_lba
= super
->active
->workspace_lba
;
1021 for (i
= 0 ; i
< super
->max_part
; i
++)
1022 dl
->vlist
[i
] = NULL
;
1025 for (i
= 0; i
< be16_to_cpu(super
->active
->max_pd_entries
); i
++)
1026 if (memcmp(super
->phys
->entries
[i
].guid
,
1027 dl
->disk
.guid
, DDF_GUID_LEN
) == 0)
1030 /* Now the config list. */
1031 /* 'conf' is an array of config entries, some of which are
1032 * probably invalid. Those which are good need to be copied into
1036 conf
= load_section(fd
, super
, super
->conf
,
1037 super
->active
->config_section_offset
,
1038 super
->active
->config_section_length
,
1043 confsec
< be32_to_cpu(super
->active
->config_section_length
);
1044 confsec
+= super
->conf_rec_len
) {
1045 struct vd_config
*vd
=
1046 (struct vd_config
*)((char*)conf
+ confsec
*512);
1049 if (be32_eq(vd
->magic
, DDF_SPARE_ASSIGN_MAGIC
)) {
1052 if (posix_memalign((void**)&dl
->spare
, 512,
1053 super
->conf_rec_len
*512) != 0) {
1054 pr_err("%s could not allocate spare info buf\n",
1059 memcpy(dl
->spare
, vd
, super
->conf_rec_len
*512);
1062 if (!be32_eq(vd
->magic
, DDF_VD_CONF_MAGIC
))
1064 for (vcl
= super
->conflist
; vcl
; vcl
= vcl
->next
) {
1065 if (memcmp(vcl
->conf
.guid
,
1066 vd
->guid
, DDF_GUID_LEN
) == 0)
1071 dl
->vlist
[vnum
++] = vcl
;
1072 if (vcl
->other_bvds
!= NULL
&&
1073 vcl
->conf
.sec_elmnt_seq
!= vd
->sec_elmnt_seq
) {
1074 add_other_bvd(vcl
, vd
, super
->conf_rec_len
*512);
1077 if (be32_to_cpu(vd
->seqnum
) <=
1078 be32_to_cpu(vcl
->conf
.seqnum
))
1081 if (posix_memalign((void**)&vcl
, 512,
1082 (super
->conf_rec_len
*512 +
1083 offsetof(struct vcl
, conf
))) != 0) {
1084 pr_err("%s could not allocate vcl buf\n",
1088 vcl
->next
= super
->conflist
;
1089 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
1090 vcl
->conf
.sec_elmnt_count
= vd
->sec_elmnt_count
;
1091 if (alloc_other_bvds(super
, vcl
) != 0) {
1092 pr_err("%s could not allocate other bvds\n",
1097 super
->conflist
= vcl
;
1098 dl
->vlist
[vnum
++] = vcl
;
1100 memcpy(&vcl
->conf
, vd
, super
->conf_rec_len
*512);
1101 for (i
=0; i
< max_virt_disks
; i
++)
1102 if (memcmp(super
->virt
->entries
[i
].guid
,
1103 vcl
->conf
.guid
, DDF_GUID_LEN
)==0)
1105 if (i
< max_virt_disks
)
1113 static int load_super_ddf_all(struct supertype
*st
, int fd
,
1114 void **sbp
, char *devname
);
1117 static void free_super_ddf(struct supertype
*st
);
1119 static int load_super_ddf(struct supertype
*st
, int fd
,
1122 unsigned long long dsize
;
1123 struct ddf_super
*super
;
1126 if (get_dev_size(fd
, devname
, &dsize
) == 0)
1129 if (!st
->ignore_hw_compat
&& test_partition(fd
))
1130 /* DDF is not allowed on partitions */
1133 /* 32M is a lower bound */
1134 if (dsize
<= 32*1024*1024) {
1136 pr_err("%s is too small for ddf: "
1137 "size is %llu sectors.\n",
1143 pr_err("%s is an odd size for ddf: "
1144 "size is %llu bytes.\n",
1151 if (posix_memalign((void**)&super
, 512, sizeof(*super
))!= 0) {
1152 pr_err("malloc of %zu failed.\n",
1156 memset(super
, 0, sizeof(*super
));
1158 rv
= load_ddf_headers(fd
, super
, devname
);
1164 /* Have valid headers and have chosen the best. Let's read in the rest*/
1166 rv
= load_ddf_global(fd
, super
, devname
);
1170 pr_err("Failed to load all information "
1171 "sections on %s\n", devname
);
1176 rv
= load_ddf_local(fd
, super
, devname
, 0);
1180 pr_err("Failed to load all information "
1181 "sections on %s\n", devname
);
1186 /* Should possibly check the sections .... */
1189 if (st
->ss
== NULL
) {
1190 st
->ss
= &super_ddf
;
1191 st
->minor_version
= 0;
1198 static void free_super_ddf(struct supertype
*st
)
1200 struct ddf_super
*ddf
= st
->sb
;
1206 while (ddf
->conflist
) {
1207 struct vcl
*v
= ddf
->conflist
;
1208 ddf
->conflist
= v
->next
;
1210 free(v
->block_sizes
);
1213 v->other_bvds[0] points to beginning of buffer,
1214 see alloc_other_bvds()
1216 free(v
->other_bvds
[0]);
1219 while (ddf
->dlist
) {
1220 struct dl
*d
= ddf
->dlist
;
1221 ddf
->dlist
= d
->next
;
1228 while (ddf
->add_list
) {
1229 struct dl
*d
= ddf
->add_list
;
1230 ddf
->add_list
= d
->next
;
1241 static struct supertype
*match_metadata_desc_ddf(char *arg
)
1243 /* 'ddf' only support containers */
1244 struct supertype
*st
;
1245 if (strcmp(arg
, "ddf") != 0 &&
1246 strcmp(arg
, "default") != 0
1250 st
= xcalloc(1, sizeof(*st
));
1251 st
->ss
= &super_ddf
;
1253 st
->minor_version
= 0;
1260 static mapping_t ddf_state
[] = {
1266 { "Partially Optimal", 5},
1272 static mapping_t ddf_init_state
[] = {
1273 { "Not Initialised", 0},
1274 { "QuickInit in Progress", 1},
1275 { "Fully Initialised", 2},
1279 static mapping_t ddf_access
[] = {
1283 { "Blocked (no access)", 3},
1287 static mapping_t ddf_level
[] = {
1288 { "RAID0", DDF_RAID0
},
1289 { "RAID1", DDF_RAID1
},
1290 { "RAID3", DDF_RAID3
},
1291 { "RAID4", DDF_RAID4
},
1292 { "RAID5", DDF_RAID5
},
1293 { "RAID1E",DDF_RAID1E
},
1294 { "JBOD", DDF_JBOD
},
1295 { "CONCAT",DDF_CONCAT
},
1296 { "RAID5E",DDF_RAID5E
},
1297 { "RAID5EE",DDF_RAID5EE
},
1298 { "RAID6", DDF_RAID6
},
1301 static mapping_t ddf_sec_level
[] = {
1302 { "Striped", DDF_2STRIPED
},
1303 { "Mirrored", DDF_2MIRRORED
},
1304 { "Concat", DDF_2CONCAT
},
1305 { "Spanned", DDF_2SPANNED
},
1310 static int all_ff(const char *guid
)
1313 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
1314 if (guid
[i
] != (char)0xff)
1319 static const char *guid_str(const char *guid
)
1321 static char buf
[DDF_GUID_LEN
*2+1];
1324 for (i
= 0; i
< DDF_GUID_LEN
; i
++) {
1325 unsigned char c
= guid
[i
];
1326 if (c
>= 32 && c
< 127)
1327 p
+= sprintf(p
, "%c", c
);
1329 p
+= sprintf(p
, "%02x", c
);
1332 return (const char *) buf
;
1336 static void print_guid(char *guid
, int tstamp
)
1338 /* A GUIDs are part (or all) ASCII and part binary.
1339 * They tend to be space padded.
1340 * We print the GUID in HEX, then in parentheses add
1341 * any initial ASCII sequence, and a possible
1342 * time stamp from bytes 16-19
1344 int l
= DDF_GUID_LEN
;
1347 for (i
=0 ; i
<DDF_GUID_LEN
; i
++) {
1348 if ((i
&3)==0 && i
!= 0) printf(":");
1349 printf("%02X", guid
[i
]&255);
1353 while (l
&& guid
[l
-1] == ' ')
1355 for (i
=0 ; i
<l
; i
++) {
1356 if (guid
[i
] >= 0x20 && guid
[i
] < 0x7f)
1357 fputc(guid
[i
], stdout
);
1362 time_t then
= __be32_to_cpu(*(__u32
*)(guid
+16)) + DECADE
;
1365 tm
= localtime(&then
);
1366 strftime(tbuf
, 100, " %D %T",tm
);
1367 fputs(tbuf
, stdout
);
1372 static void examine_vd(int n
, struct ddf_super
*sb
, char *guid
)
1374 int crl
= sb
->conf_rec_len
;
1377 for (vcl
= sb
->conflist
; vcl
; vcl
= vcl
->next
) {
1379 struct vd_config
*vc
= &vcl
->conf
;
1381 if (!be32_eq(calc_crc(vc
, crl
*512), vc
->crc
))
1383 if (memcmp(vc
->guid
, guid
, DDF_GUID_LEN
) != 0)
1386 /* Ok, we know about this VD, let's give more details */
1387 printf(" Raid Devices[%d] : %d (", n
,
1388 be16_to_cpu(vc
->prim_elmnt_count
));
1389 for (i
= 0; i
< be16_to_cpu(vc
->prim_elmnt_count
); i
++) {
1391 int cnt
= be16_to_cpu(sb
->phys
->used_pdes
);
1392 for (j
=0; j
<cnt
; j
++)
1393 if (be32_eq(vc
->phys_refnum
[i
],
1394 sb
->phys
->entries
[j
].refnum
))
1403 if (vc
->chunk_shift
!= 255)
1404 printf(" Chunk Size[%d] : %d sectors\n", n
,
1405 1 << vc
->chunk_shift
);
1406 printf(" Raid Level[%d] : %s\n", n
,
1407 map_num(ddf_level
, vc
->prl
)?:"-unknown-");
1408 if (vc
->sec_elmnt_count
!= 1) {
1409 printf(" Secondary Position[%d] : %d of %d\n", n
,
1410 vc
->sec_elmnt_seq
, vc
->sec_elmnt_count
);
1411 printf(" Secondary Level[%d] : %s\n", n
,
1412 map_num(ddf_sec_level
, vc
->srl
) ?: "-unknown-");
1414 printf(" Device Size[%d] : %llu\n", n
,
1415 be64_to_cpu(vc
->blocks
)/2);
1416 printf(" Array Size[%d] : %llu\n", n
,
1417 be64_to_cpu(vc
->array_blocks
)/2);
1421 static void examine_vds(struct ddf_super
*sb
)
1423 int cnt
= be16_to_cpu(sb
->virt
->populated_vdes
);
1425 printf(" Virtual Disks : %d\n", cnt
);
1427 for (i
= 0; i
< be16_to_cpu(sb
->virt
->max_vdes
); i
++) {
1428 struct virtual_entry
*ve
= &sb
->virt
->entries
[i
];
1429 if (all_ff(ve
->guid
))
1432 printf(" VD GUID[%d] : ", i
); print_guid(ve
->guid
, 1);
1434 printf(" unit[%d] : %d\n", i
, be16_to_cpu(ve
->unit
));
1435 printf(" state[%d] : %s, %s%s\n", i
,
1436 map_num(ddf_state
, ve
->state
& 7),
1437 (ve
->state
& DDF_state_morphing
) ? "Morphing, ": "",
1438 (ve
->state
& DDF_state_inconsistent
)? "Not Consistent" : "Consistent");
1439 printf(" init state[%d] : %s\n", i
,
1440 map_num(ddf_init_state
, ve
->init_state
&DDF_initstate_mask
));
1441 printf(" access[%d] : %s\n", i
,
1442 map_num(ddf_access
, (ve
->init_state
& DDF_access_mask
) >> 6));
1443 printf(" Name[%d] : %.16s\n", i
, ve
->name
);
1444 examine_vd(i
, sb
, ve
->guid
);
1446 if (cnt
) printf("\n");
1449 static void examine_pds(struct ddf_super
*sb
)
1451 int cnt
= be16_to_cpu(sb
->phys
->used_pdes
);
1454 printf(" Physical Disks : %d\n", cnt
);
1455 printf(" Number RefNo Size Device Type/State\n");
1457 for (i
=0 ; i
<cnt
; i
++) {
1458 struct phys_disk_entry
*pd
= &sb
->phys
->entries
[i
];
1459 int type
= be16_to_cpu(pd
->type
);
1460 int state
= be16_to_cpu(pd
->state
);
1462 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1464 printf(" %3d %08x ", i
,
1465 be32_to_cpu(pd
->refnum
));
1467 be64_to_cpu(pd
->config_size
)>>1);
1468 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1469 if (be32_eq(dl
->disk
.refnum
, pd
->refnum
)) {
1470 char *dv
= map_dev(dl
->major
, dl
->minor
, 0);
1472 printf("%-15s", dv
);
1479 printf(" %s%s%s%s%s",
1480 (type
&2) ? "active":"",
1481 (type
&4) ? "Global-Spare":"",
1482 (type
&8) ? "spare" : "",
1483 (type
&16)? ", foreign" : "",
1484 (type
&32)? "pass-through" : "");
1485 if (state
& DDF_Failed
)
1486 /* This over-rides these three */
1487 state
&= ~(DDF_Online
|DDF_Rebuilding
|DDF_Transition
);
1488 printf("/%s%s%s%s%s%s%s",
1489 (state
&1)? "Online": "Offline",
1490 (state
&2)? ", Failed": "",
1491 (state
&4)? ", Rebuilding": "",
1492 (state
&8)? ", in-transition": "",
1493 (state
&16)? ", SMART-errors": "",
1494 (state
&32)? ", Unrecovered-Read-Errors": "",
1495 (state
&64)? ", Missing" : "");
1500 static void examine_super_ddf(struct supertype
*st
, char *homehost
)
1502 struct ddf_super
*sb
= st
->sb
;
1504 printf(" Magic : %08x\n", be32_to_cpu(sb
->anchor
.magic
));
1505 printf(" Version : %.8s\n", sb
->anchor
.revision
);
1506 printf("Controller GUID : "); print_guid(sb
->controller
.guid
, 0);
1508 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1510 printf(" Seq : %08x\n", be32_to_cpu(sb
->active
->seq
));
1511 printf(" Redundant hdr : %s\n", be32_eq(sb
->secondary
.magic
,
1518 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1520 static void uuid_from_ddf_guid(const char *guid
, int uuid
[4]);
1521 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4]);
1522 static void _ddf_array_name(char *name
, const struct ddf_super
*ddf
, int i
);
1524 static unsigned int get_vd_num_of_subarray(struct supertype
*st
)
1527 * Figure out the VD number for this supertype.
1528 * Returns DDF_CONTAINER for the container itself,
1529 * and DDF_NOTFOUND on error.
1531 struct ddf_super
*ddf
= st
->sb
;
1536 if (*st
->container_devnm
== '\0')
1537 return DDF_CONTAINER
;
1539 sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
1540 if (!sra
|| sra
->array
.major_version
!= -1 ||
1541 sra
->array
.minor_version
!= -2 ||
1542 !is_subarray(sra
->text_version
))
1543 return DDF_NOTFOUND
;
1545 sub
= strchr(sra
->text_version
+ 1, '/');
1547 vcnum
= strtoul(sub
+ 1, &end
, 10);
1548 if (sub
== NULL
|| *sub
== '\0' || *end
!= '\0' ||
1549 vcnum
>= be16_to_cpu(ddf
->active
->max_vd_entries
))
1550 return DDF_NOTFOUND
;
1555 static void brief_examine_super_ddf(struct supertype
*st
, int verbose
)
1557 /* We just write a generic DDF ARRAY entry
1561 getinfo_super_ddf(st
, &info
, NULL
);
1562 fname_from_uuid(st
, &info
, nbuf
, ':');
1564 printf("ARRAY metadata=ddf UUID=%s\n", nbuf
+ 5);
1567 static void brief_examine_subarrays_ddf(struct supertype
*st
, int verbose
)
1569 /* We just write a generic DDF ARRAY entry
1571 struct ddf_super
*ddf
= st
->sb
;
1575 getinfo_super_ddf(st
, &info
, NULL
);
1576 fname_from_uuid(st
, &info
, nbuf
, ':');
1578 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
1579 struct virtual_entry
*ve
= &ddf
->virt
->entries
[i
];
1583 if (all_ff(ve
->guid
))
1585 memcpy(vcl
.conf
.guid
, ve
->guid
, DDF_GUID_LEN
);
1586 ddf
->currentconf
=&vcl
;
1588 uuid_from_super_ddf(st
, info
.uuid
);
1589 fname_from_uuid(st
, &info
, nbuf1
, ':');
1590 _ddf_array_name(namebuf
, ddf
, i
);
1591 printf("ARRAY%s%s container=%s member=%d UUID=%s\n",
1592 namebuf
[0] == '\0' ? "" : " /dev/md/", namebuf
,
1593 nbuf
+5, i
, nbuf1
+5);
1597 static void export_examine_super_ddf(struct supertype
*st
)
1601 getinfo_super_ddf(st
, &info
, NULL
);
1602 fname_from_uuid(st
, &info
, nbuf
, ':');
1603 printf("MD_METADATA=ddf\n");
1604 printf("MD_LEVEL=container\n");
1605 printf("MD_UUID=%s\n", nbuf
+5);
1606 printf("MD_DEVICES=%u\n",
1607 be16_to_cpu(((struct ddf_super
*)st
->sb
)->phys
->used_pdes
));
1610 static int copy_metadata_ddf(struct supertype
*st
, int from
, int to
)
1613 unsigned long long dsize
, offset
;
1615 struct ddf_header
*ddf
;
1618 /* The meta consists of an anchor, a primary, and a secondary.
1619 * This all lives at the end of the device.
1620 * So it is easiest to find the earliest of primary and
1621 * secondary, and copy everything from there.
1623 * Anchor is 512 from end It contains primary_lba and secondary_lba
1624 * we choose one of those
1627 if (posix_memalign(&buf
, 4096, 4096) != 0)
1630 if (!get_dev_size(from
, NULL
, &dsize
))
1633 if (lseek64(from
, dsize
-512, 0) < 0)
1635 if (read(from
, buf
, 512) != 512)
1638 if (!be32_eq(ddf
->magic
, DDF_HEADER_MAGIC
) ||
1639 !be32_eq(calc_crc(ddf
, 512), ddf
->crc
) ||
1640 (memcmp(ddf
->revision
, DDF_REVISION_0
, 8) != 0 &&
1641 memcmp(ddf
->revision
, DDF_REVISION_2
, 8) != 0))
1644 offset
= dsize
- 512;
1645 if ((be64_to_cpu(ddf
->primary_lba
) << 9) < offset
)
1646 offset
= be64_to_cpu(ddf
->primary_lba
) << 9;
1647 if ((be64_to_cpu(ddf
->secondary_lba
) << 9) < offset
)
1648 offset
= be64_to_cpu(ddf
->secondary_lba
) << 9;
1650 bytes
= dsize
- offset
;
1652 if (lseek64(from
, offset
, 0) < 0 ||
1653 lseek64(to
, offset
, 0) < 0)
1655 while (written
< bytes
) {
1656 int n
= bytes
- written
;
1659 if (read(from
, buf
, n
) != n
)
1661 if (write(to
, buf
, n
) != n
)
1672 static void detail_super_ddf(struct supertype
*st
, char *homehost
)
1675 * Could print DDF GUID
1676 * Need to find which array
1677 * If whole, briefly list all arrays
1682 static const char *vendors_with_variable_volume_UUID
[] = {
1686 static int volume_id_is_reliable(const struct ddf_super
*ddf
)
1688 int n
= ARRAY_SIZE(vendors_with_variable_volume_UUID
);
1690 for (i
= 0; i
< n
; i
++)
1691 if (!memcmp(ddf
->controller
.guid
,
1692 vendors_with_variable_volume_UUID
[i
], 8))
1697 static void uuid_of_ddf_subarray(const struct ddf_super
*ddf
,
1698 unsigned int vcnum
, int uuid
[4])
1700 char buf
[DDF_GUID_LEN
+18], sha
[20], *p
;
1701 struct sha1_ctx ctx
;
1702 if (volume_id_is_reliable(ddf
)) {
1703 uuid_from_ddf_guid(ddf
->virt
->entries
[vcnum
].guid
, uuid
);
1707 * Some fake RAID BIOSes (in particular, LSI ones) change the
1708 * VD GUID at every boot. These GUIDs are not suitable for
1709 * identifying an array. Luckily the header GUID appears to
1711 * We construct a pseudo-UUID from the header GUID and those
1712 * properties of the subarray that we expect to remain constant.
1714 memset(buf
, 0, sizeof(buf
));
1716 memcpy(p
, ddf
->anchor
.guid
, DDF_GUID_LEN
);
1718 memcpy(p
, ddf
->virt
->entries
[vcnum
].name
, 16);
1720 *((__u16
*) p
) = vcnum
;
1721 sha1_init_ctx(&ctx
);
1722 sha1_process_bytes(buf
, sizeof(buf
), &ctx
);
1723 sha1_finish_ctx(&ctx
, sha
);
1724 memcpy(uuid
, sha
, 4*4);
1727 static void brief_detail_super_ddf(struct supertype
*st
)
1731 struct ddf_super
*ddf
= st
->sb
;
1732 unsigned int vcnum
= get_vd_num_of_subarray(st
);
1733 if (vcnum
== DDF_CONTAINER
)
1734 uuid_from_super_ddf(st
, info
.uuid
);
1735 else if (vcnum
== DDF_NOTFOUND
)
1738 uuid_of_ddf_subarray(ddf
, vcnum
, info
.uuid
);
1739 fname_from_uuid(st
, &info
, nbuf
,':');
1740 printf(" UUID=%s", nbuf
+ 5);
1744 static int match_home_ddf(struct supertype
*st
, char *homehost
)
1746 /* It matches 'this' host if the controller is a
1747 * Linux-MD controller with vendor_data matching
1750 struct ddf_super
*ddf
= st
->sb
;
1755 len
= strlen(homehost
);
1757 return (memcmp(ddf
->controller
.guid
, T10
, 8) == 0 &&
1758 len
< sizeof(ddf
->controller
.vendor_data
) &&
1759 memcmp(ddf
->controller
.vendor_data
, homehost
,len
) == 0 &&
1760 ddf
->controller
.vendor_data
[len
] == 0);
1764 static int find_index_in_bvd(const struct ddf_super
*ddf
,
1765 const struct vd_config
*conf
, unsigned int n
,
1766 unsigned int *n_bvd
)
1769 * Find the index of the n-th valid physical disk in this BVD
1772 for (i
= 0, j
= 0; i
< ddf
->mppe
&&
1773 j
< be16_to_cpu(conf
->prim_elmnt_count
); i
++) {
1774 if (be32_to_cpu(conf
->phys_refnum
[i
]) != 0xffffffff) {
1782 dprintf("%s: couldn't find BVD member %u (total %u)\n",
1783 __func__
, n
, be16_to_cpu(conf
->prim_elmnt_count
));
1787 static struct vd_config
*find_vdcr(struct ddf_super
*ddf
, unsigned int inst
,
1789 unsigned int *n_bvd
, struct vcl
**vcl
)
1793 for (v
= ddf
->conflist
; v
; v
= v
->next
) {
1794 unsigned int nsec
, ibvd
= 0;
1795 struct vd_config
*conf
;
1796 if (inst
!= v
->vcnum
)
1799 if (conf
->sec_elmnt_count
== 1) {
1800 if (find_index_in_bvd(ddf
, conf
, n
, n_bvd
)) {
1806 if (v
->other_bvds
== NULL
) {
1807 pr_err("%s: BUG: other_bvds is NULL, nsec=%u\n",
1808 __func__
, conf
->sec_elmnt_count
);
1811 nsec
= n
/ be16_to_cpu(conf
->prim_elmnt_count
);
1812 if (conf
->sec_elmnt_seq
!= nsec
) {
1813 for (ibvd
= 1; ibvd
< conf
->sec_elmnt_count
; ibvd
++) {
1814 if (v
->other_bvds
[ibvd
-1]->sec_elmnt_seq
1818 if (ibvd
== conf
->sec_elmnt_count
)
1820 conf
= v
->other_bvds
[ibvd
-1];
1822 if (!find_index_in_bvd(ddf
, conf
,
1823 n
- nsec
*conf
->sec_elmnt_count
, n_bvd
))
1825 dprintf("%s: found disk %u as member %u in bvd %d of array %u\n"
1826 , __func__
, n
, *n_bvd
, ibvd
, inst
);
1831 pr_err("%s: Could't find disk %d in array %u\n", __func__
, n
, inst
);
1836 static int find_phys(const struct ddf_super
*ddf
, be32 phys_refnum
)
1838 /* Find the entry in phys_disk which has the given refnum
1839 * and return it's index
1842 for (i
= 0; i
< be16_to_cpu(ddf
->phys
->max_pdes
); i
++)
1843 if (be32_eq(ddf
->phys
->entries
[i
].refnum
, phys_refnum
))
1848 static void uuid_from_ddf_guid(const char *guid
, int uuid
[4])
1851 struct sha1_ctx ctx
;
1852 sha1_init_ctx(&ctx
);
1853 sha1_process_bytes(guid
, DDF_GUID_LEN
, &ctx
);
1854 sha1_finish_ctx(&ctx
, buf
);
1855 memcpy(uuid
, buf
, 4*4);
1858 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4])
1860 /* The uuid returned here is used for:
1861 * uuid to put into bitmap file (Create, Grow)
1862 * uuid for backup header when saving critical section (Grow)
1863 * comparing uuids when re-adding a device into an array
1864 * In these cases the uuid required is that of the data-array,
1865 * not the device-set.
1866 * uuid to recognise same set when adding a missing device back
1867 * to an array. This is a uuid for the device-set.
1869 * For each of these we can make do with a truncated
1870 * or hashed uuid rather than the original, as long as
1872 * In the case of SVD we assume the BVD is of interest,
1873 * though that might be the case if a bitmap were made for
1874 * a mirrored SVD - worry about that later.
1875 * So we need to find the VD configuration record for the
1876 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1877 * The first 16 bytes of the sha1 of these is used.
1879 struct ddf_super
*ddf
= st
->sb
;
1880 struct vcl
*vcl
= ddf
->currentconf
;
1883 uuid_of_ddf_subarray(ddf
, vcl
->vcnum
, uuid
);
1885 uuid_from_ddf_guid(ddf
->anchor
.guid
, uuid
);
1888 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1890 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1892 struct ddf_super
*ddf
= st
->sb
;
1893 int map_disks
= info
->array
.raid_disks
;
1896 if (ddf
->currentconf
) {
1897 getinfo_super_ddf_bvd(st
, info
, map
);
1900 memset(info
, 0, sizeof(*info
));
1902 info
->array
.raid_disks
= be16_to_cpu(ddf
->phys
->used_pdes
);
1903 info
->array
.level
= LEVEL_CONTAINER
;
1904 info
->array
.layout
= 0;
1905 info
->array
.md_minor
= -1;
1906 cptr
= (__u32
*)(ddf
->anchor
.guid
+ 16);
1907 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
1909 info
->array
.utime
= 0;
1910 info
->array
.chunk_size
= 0;
1911 info
->container_enough
= 1;
1913 info
->disk
.major
= 0;
1914 info
->disk
.minor
= 0;
1916 info
->disk
.number
= be32_to_cpu(ddf
->dlist
->disk
.refnum
);
1917 info
->disk
.raid_disk
= find_phys(ddf
, ddf
->dlist
->disk
.refnum
);
1919 info
->data_offset
= be64_to_cpu(ddf
->phys
->
1920 entries
[info
->disk
.raid_disk
].
1922 info
->component_size
= ddf
->dlist
->size
- info
->data_offset
;
1924 info
->disk
.number
= -1;
1925 info
->disk
.raid_disk
= -1;
1926 // info->disk.raid_disk = find refnum in the table and use index;
1928 info
->disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
1930 info
->recovery_start
= MaxSector
;
1931 info
->reshape_active
= 0;
1932 info
->recovery_blocked
= 0;
1935 info
->array
.major_version
= -1;
1936 info
->array
.minor_version
= -2;
1937 strcpy(info
->text_version
, "ddf");
1938 info
->safe_mode_delay
= 0;
1940 uuid_from_super_ddf(st
, info
->uuid
);
1944 for (i
= 0 ; i
< map_disks
; i
++) {
1945 if (i
< info
->array
.raid_disks
&&
1946 (be16_to_cpu(ddf
->phys
->entries
[i
].state
)
1948 !(be16_to_cpu(ddf
->phys
->entries
[i
].state
)
1957 /* size of name must be at least 17 bytes! */
1958 static void _ddf_array_name(char *name
, const struct ddf_super
*ddf
, int i
)
1961 memcpy(name
, ddf
->virt
->entries
[i
].name
, 16);
1963 for(j
= 0; j
< 16; j
++)
1968 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1970 struct ddf_super
*ddf
= st
->sb
;
1971 struct vcl
*vc
= ddf
->currentconf
;
1972 int cd
= ddf
->currentdev
;
1976 int map_disks
= info
->array
.raid_disks
;
1978 struct vd_config
*conf
;
1980 memset(info
, 0, sizeof(*info
));
1981 if (layout_ddf2md(&vc
->conf
, &info
->array
) == -1)
1983 info
->array
.md_minor
= -1;
1984 cptr
= (__u32
*)(vc
->conf
.guid
+ 16);
1985 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
1986 info
->array
.utime
= DECADE
+ be32_to_cpu(vc
->conf
.timestamp
);
1987 info
->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
1988 info
->custom_array_size
= 0;
1991 n_prim
= be16_to_cpu(conf
->prim_elmnt_count
);
1992 if (conf
->sec_elmnt_count
> 1 && cd
>= n_prim
) {
1993 int ibvd
= cd
/ n_prim
- 1;
1995 conf
= vc
->other_bvds
[ibvd
];
1998 if (cd
>= 0 && (unsigned)cd
< ddf
->mppe
) {
2000 be64_to_cpu(LBA_OFFSET(ddf
, conf
)[cd
]);
2001 if (vc
->block_sizes
)
2002 info
->component_size
= vc
->block_sizes
[cd
];
2004 info
->component_size
= be64_to_cpu(conf
->blocks
);
2007 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2008 if (be32_eq(dl
->disk
.refnum
, conf
->phys_refnum
[cd
]))
2011 info
->disk
.major
= 0;
2012 info
->disk
.minor
= 0;
2013 info
->disk
.state
= 0;
2015 info
->disk
.major
= dl
->major
;
2016 info
->disk
.minor
= dl
->minor
;
2017 info
->disk
.raid_disk
= cd
+ conf
->sec_elmnt_seq
2018 * be16_to_cpu(conf
->prim_elmnt_count
);
2019 info
->disk
.number
= dl
->pdnum
;
2020 info
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
2023 info
->container_member
= ddf
->currentconf
->vcnum
;
2025 info
->recovery_start
= MaxSector
;
2026 info
->resync_start
= 0;
2027 info
->reshape_active
= 0;
2028 info
->recovery_blocked
= 0;
2029 if (!(ddf
->virt
->entries
[info
->container_member
].state
2030 & DDF_state_inconsistent
) &&
2031 (ddf
->virt
->entries
[info
->container_member
].init_state
2032 & DDF_initstate_mask
)
2034 info
->resync_start
= MaxSector
;
2036 uuid_from_super_ddf(st
, info
->uuid
);
2038 info
->array
.major_version
= -1;
2039 info
->array
.minor_version
= -2;
2040 sprintf(info
->text_version
, "/%s/%d",
2041 st
->container_devnm
,
2042 info
->container_member
);
2043 info
->safe_mode_delay
= DDF_SAFE_MODE_DELAY
;
2045 _ddf_array_name(info
->name
, ddf
, info
->container_member
);
2048 for (j
= 0; j
< map_disks
; j
++) {
2050 if (j
< info
->array
.raid_disks
) {
2051 int i
= find_phys(ddf
, vc
->conf
.phys_refnum
[j
]);
2053 (be16_to_cpu(ddf
->phys
->entries
[i
].state
)
2055 !(be16_to_cpu(ddf
->phys
->entries
[i
].state
)
2062 static int update_super_ddf(struct supertype
*st
, struct mdinfo
*info
,
2064 char *devname
, int verbose
,
2065 int uuid_set
, char *homehost
)
2067 /* For 'assemble' and 'force' we need to return non-zero if any
2068 * change was made. For others, the return value is ignored.
2069 * Update options are:
2070 * force-one : This device looks a bit old but needs to be included,
2071 * update age info appropriately.
2072 * assemble: clear any 'faulty' flag to allow this device to
2074 * force-array: Array is degraded but being forced, mark it clean
2075 * if that will be needed to assemble it.
2077 * newdev: not used ????
2078 * grow: Array has gained a new device - this is currently for
2080 * resync: mark as dirty so a resync will happen.
2081 * uuid: Change the uuid of the array to match what is given
2082 * homehost: update the recorded homehost
2083 * name: update the name - preserving the homehost
2084 * _reshape_progress: record new reshape_progress position.
2086 * Following are not relevant for this version:
2087 * sparc2.2 : update from old dodgey metadata
2088 * super-minor: change the preferred_minor number
2089 * summaries: update redundant counters.
2092 // struct ddf_super *ddf = st->sb;
2093 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
2094 // struct virtual_entry *ve = find_ve(ddf);
2096 /* we don't need to handle "force-*" or "assemble" as
2097 * there is no need to 'trick' the kernel. We the metadata is
2098 * first updated to activate the array, all the implied modifications
2102 if (strcmp(update
, "grow") == 0) {
2104 } else if (strcmp(update
, "resync") == 0) {
2105 // info->resync_checkpoint = 0;
2106 } else if (strcmp(update
, "homehost") == 0) {
2107 /* homehost is stored in controller->vendor_data,
2108 * or it is when we are the vendor
2110 // if (info->vendor_is_local)
2111 // strcpy(ddf->controller.vendor_data, homehost);
2113 } else if (strcmp(update
, "name") == 0) {
2114 /* name is stored in virtual_entry->name */
2115 // memset(ve->name, ' ', 16);
2116 // strncpy(ve->name, info->name, 16);
2118 } else if (strcmp(update
, "_reshape_progress") == 0) {
2119 /* We don't support reshape yet */
2120 } else if (strcmp(update
, "assemble") == 0 ) {
2121 /* Do nothing, just succeed */
2126 // update_all_csum(ddf);
2131 static void make_header_guid(char *guid
)
2134 /* Create a DDF Header of Virtual Disk GUID */
2136 /* 24 bytes of fiction required.
2137 * first 8 are a 'vendor-id' - "Linux-MD"
2138 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
2139 * Remaining 8 random number plus timestamp
2141 memcpy(guid
, T10
, sizeof(T10
));
2142 stamp
= cpu_to_be32(0xdeadbeef);
2143 memcpy(guid
+8, &stamp
, 4);
2144 stamp
= cpu_to_be32(0);
2145 memcpy(guid
+12, &stamp
, 4);
2146 stamp
= cpu_to_be32(time(0) - DECADE
);
2147 memcpy(guid
+16, &stamp
, 4);
2148 stamp
._v32
= random32();
2149 memcpy(guid
+20, &stamp
, 4);
2152 static unsigned int find_unused_vde(const struct ddf_super
*ddf
)
2155 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
2156 if (all_ff(ddf
->virt
->entries
[i
].guid
))
2159 return DDF_NOTFOUND
;
2162 static unsigned int find_vde_by_name(const struct ddf_super
*ddf
,
2167 return DDF_NOTFOUND
;
2168 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
2169 if (all_ff(ddf
->virt
->entries
[i
].guid
))
2171 if (!strncmp(name
, ddf
->virt
->entries
[i
].name
,
2172 sizeof(ddf
->virt
->entries
[i
].name
)))
2175 return DDF_NOTFOUND
;
2179 static unsigned int find_vde_by_guid(const struct ddf_super
*ddf
,
2183 if (guid
== NULL
|| all_ff(guid
))
2184 return DDF_NOTFOUND
;
2185 for (i
= 0; i
< be16_to_cpu(ddf
->virt
->max_vdes
); i
++)
2186 if (!memcmp(ddf
->virt
->entries
[i
].guid
, guid
, DDF_GUID_LEN
))
2188 return DDF_NOTFOUND
;
2192 static int init_super_ddf_bvd(struct supertype
*st
,
2193 mdu_array_info_t
*info
,
2194 unsigned long long size
,
2195 char *name
, char *homehost
,
2196 int *uuid
, unsigned long long data_offset
);
2198 static int init_super_ddf(struct supertype
*st
,
2199 mdu_array_info_t
*info
,
2200 unsigned long long size
, char *name
, char *homehost
,
2201 int *uuid
, unsigned long long data_offset
)
2203 /* This is primarily called by Create when creating a new array.
2204 * We will then get add_to_super called for each component, and then
2205 * write_init_super called to write it out to each device.
2206 * For DDF, Create can create on fresh devices or on a pre-existing
2208 * To create on a pre-existing array a different method will be called.
2209 * This one is just for fresh drives.
2211 * We need to create the entire 'ddf' structure which includes:
2212 * DDF headers - these are easy.
2213 * Controller data - a Sector describing this controller .. not that
2214 * this is a controller exactly.
2215 * Physical Disk Record - one entry per device, so
2216 * leave plenty of space.
2217 * Virtual Disk Records - again, just leave plenty of space.
2218 * This just lists VDs, doesn't give details
2219 * Config records - describes the VDs that use this disk
2220 * DiskData - describes 'this' device.
2221 * BadBlockManagement - empty
2222 * Diag Space - empty
2223 * Vendor Logs - Could we put bitmaps here?
2226 struct ddf_super
*ddf
;
2229 int max_phys_disks
, max_virt_disks
;
2230 unsigned long long sector
;
2234 struct phys_disk
*pd
;
2235 struct virtual_disk
*vd
;
2237 if (data_offset
!= INVALID_SECTORS
) {
2238 pr_err("data-offset not supported by DDF\n");
2243 return init_super_ddf_bvd(st
, info
, size
, name
, homehost
, uuid
,
2246 if (posix_memalign((void**)&ddf
, 512, sizeof(*ddf
)) != 0) {
2247 pr_err("%s could not allocate superblock\n", __func__
);
2250 memset(ddf
, 0, sizeof(*ddf
));
2251 ddf
->dlist
= NULL
; /* no physical disks yet */
2252 ddf
->conflist
= NULL
; /* No virtual disks yet */
2256 /* zeroing superblock */
2260 /* At least 32MB *must* be reserved for the ddf. So let's just
2261 * start 32MB from the end, and put the primary header there.
2262 * Don't do secondary for now.
2263 * We don't know exactly where that will be yet as it could be
2264 * different on each device. To just set up the lengths.
2268 ddf
->anchor
.magic
= DDF_HEADER_MAGIC
;
2269 make_header_guid(ddf
->anchor
.guid
);
2271 memcpy(ddf
->anchor
.revision
, DDF_REVISION_2
, 8);
2272 ddf
->anchor
.seq
= cpu_to_be32(1);
2273 ddf
->anchor
.timestamp
= cpu_to_be32(time(0) - DECADE
);
2274 ddf
->anchor
.openflag
= 0xFF;
2275 ddf
->anchor
.foreignflag
= 0;
2276 ddf
->anchor
.enforcegroups
= 0; /* Is this best?? */
2277 ddf
->anchor
.pad0
= 0xff;
2278 memset(ddf
->anchor
.pad1
, 0xff, 12);
2279 memset(ddf
->anchor
.header_ext
, 0xff, 32);
2280 ddf
->anchor
.primary_lba
= cpu_to_be64(~(__u64
)0);
2281 ddf
->anchor
.secondary_lba
= cpu_to_be64(~(__u64
)0);
2282 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
2283 memset(ddf
->anchor
.pad2
, 0xff, 3);
2284 ddf
->anchor
.workspace_len
= cpu_to_be32(32768); /* Must be reserved */
2285 /* Put this at bottom of 32M reserved.. */
2286 ddf
->anchor
.workspace_lba
= cpu_to_be64(~(__u64
)0);
2287 max_phys_disks
= 1023; /* Should be enough */
2288 ddf
->anchor
.max_pd_entries
= cpu_to_be16(max_phys_disks
);
2289 max_virt_disks
= 255;
2290 ddf
->anchor
.max_vd_entries
= cpu_to_be16(max_virt_disks
); /* ?? */
2291 ddf
->anchor
.max_partitions
= cpu_to_be16(64); /* ?? */
2294 ddf
->conf_rec_len
= 1 + ROUND_UP(ddf
->mppe
* (4+8), 512)/512;
2295 ddf
->anchor
.config_record_len
= cpu_to_be16(ddf
->conf_rec_len
);
2296 ddf
->anchor
.max_primary_element_entries
= cpu_to_be16(ddf
->mppe
);
2297 memset(ddf
->anchor
.pad3
, 0xff, 54);
2298 /* controller sections is one sector long immediately
2299 * after the ddf header */
2301 ddf
->anchor
.controller_section_offset
= cpu_to_be32(sector
);
2302 ddf
->anchor
.controller_section_length
= cpu_to_be32(1);
2305 /* phys is 8 sectors after that */
2306 pdsize
= ROUND_UP(sizeof(struct phys_disk
) +
2307 sizeof(struct phys_disk_entry
)*max_phys_disks
,
2309 switch(pdsize
/512) {
2310 case 2: case 8: case 32: case 128: case 512: break;
2313 ddf
->anchor
.phys_section_offset
= cpu_to_be32(sector
);
2314 ddf
->anchor
.phys_section_length
=
2315 cpu_to_be32(pdsize
/512); /* max_primary_element_entries/8 */
2316 sector
+= pdsize
/512;
2318 /* virt is another 32 sectors */
2319 vdsize
= ROUND_UP(sizeof(struct virtual_disk
) +
2320 sizeof(struct virtual_entry
) * max_virt_disks
,
2322 switch(vdsize
/512) {
2323 case 2: case 8: case 32: case 128: case 512: break;
2326 ddf
->anchor
.virt_section_offset
= cpu_to_be32(sector
);
2327 ddf
->anchor
.virt_section_length
=
2328 cpu_to_be32(vdsize
/512); /* max_vd_entries/8 */
2329 sector
+= vdsize
/512;
2331 clen
= ddf
->conf_rec_len
* (ddf
->max_part
+1);
2332 ddf
->anchor
.config_section_offset
= cpu_to_be32(sector
);
2333 ddf
->anchor
.config_section_length
= cpu_to_be32(clen
);
2336 ddf
->anchor
.data_section_offset
= cpu_to_be32(sector
);
2337 ddf
->anchor
.data_section_length
= cpu_to_be32(1);
2340 ddf
->anchor
.bbm_section_length
= cpu_to_be32(0);
2341 ddf
->anchor
.bbm_section_offset
= cpu_to_be32(0xFFFFFFFF);
2342 ddf
->anchor
.diag_space_length
= cpu_to_be32(0);
2343 ddf
->anchor
.diag_space_offset
= cpu_to_be32(0xFFFFFFFF);
2344 ddf
->anchor
.vendor_length
= cpu_to_be32(0);
2345 ddf
->anchor
.vendor_offset
= cpu_to_be32(0xFFFFFFFF);
2347 memset(ddf
->anchor
.pad4
, 0xff, 256);
2349 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2350 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2352 ddf
->primary
.openflag
= 1; /* I guess.. */
2353 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
2355 ddf
->secondary
.openflag
= 1; /* I guess.. */
2356 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
2358 ddf
->active
= &ddf
->primary
;
2360 ddf
->controller
.magic
= DDF_CONTROLLER_MAGIC
;
2362 /* 24 more bytes of fiction required.
2363 * first 8 are a 'vendor-id' - "Linux-MD"
2364 * Remaining 16 are serial number.... maybe a hostname would do?
2366 memcpy(ddf
->controller
.guid
, T10
, sizeof(T10
));
2367 gethostname(hostname
, sizeof(hostname
));
2368 hostname
[sizeof(hostname
) - 1] = 0;
2369 hostlen
= strlen(hostname
);
2370 memcpy(ddf
->controller
.guid
+ 24 - hostlen
, hostname
, hostlen
);
2371 for (i
= strlen(T10
) ; i
+hostlen
< 24; i
++)
2372 ddf
->controller
.guid
[i
] = ' ';
2374 ddf
->controller
.type
.vendor_id
= cpu_to_be16(0xDEAD);
2375 ddf
->controller
.type
.device_id
= cpu_to_be16(0xBEEF);
2376 ddf
->controller
.type
.sub_vendor_id
= cpu_to_be16(0);
2377 ddf
->controller
.type
.sub_device_id
= cpu_to_be16(0);
2378 memcpy(ddf
->controller
.product_id
, "What Is My PID??", 16);
2379 memset(ddf
->controller
.pad
, 0xff, 8);
2380 memset(ddf
->controller
.vendor_data
, 0xff, 448);
2381 if (homehost
&& strlen(homehost
) < 440)
2382 strcpy((char*)ddf
->controller
.vendor_data
, homehost
);
2384 if (posix_memalign((void**)&pd
, 512, pdsize
) != 0) {
2385 pr_err("%s could not allocate pd\n", __func__
);
2389 ddf
->pdsize
= pdsize
;
2391 memset(pd
, 0xff, pdsize
);
2392 memset(pd
, 0, sizeof(*pd
));
2393 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2394 pd
->used_pdes
= cpu_to_be16(0);
2395 pd
->max_pdes
= cpu_to_be16(max_phys_disks
);
2396 memset(pd
->pad
, 0xff, 52);
2397 for (i
= 0; i
< max_phys_disks
; i
++)
2398 memset(pd
->entries
[i
].guid
, 0xff, DDF_GUID_LEN
);
2400 if (posix_memalign((void**)&vd
, 512, vdsize
) != 0) {
2401 pr_err("%s could not allocate vd\n", __func__
);
2405 ddf
->vdsize
= vdsize
;
2406 memset(vd
, 0, vdsize
);
2407 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
2408 vd
->populated_vdes
= cpu_to_be16(0);
2409 vd
->max_vdes
= cpu_to_be16(max_virt_disks
);
2410 memset(vd
->pad
, 0xff, 52);
2412 for (i
=0; i
<max_virt_disks
; i
++)
2413 memset(&vd
->entries
[i
], 0xff, sizeof(struct virtual_entry
));
2416 ddf_set_updates_pending(ddf
);
2420 static int chunk_to_shift(int chunksize
)
2422 return ffs(chunksize
/512)-1;
2427 unsigned long long start
, size
;
2429 static int cmp_extent(const void *av
, const void *bv
)
2431 const struct extent
*a
= av
;
2432 const struct extent
*b
= bv
;
2433 if (a
->start
< b
->start
)
2435 if (a
->start
> b
->start
)
2440 static struct extent
*get_extents(struct ddf_super
*ddf
, struct dl
*dl
)
2442 /* find a list of used extents on the give physical device
2443 * (dnum) of the given ddf.
2444 * Return a malloced array of 'struct extent'
2446 * FIXME ignore DDF_Legacy devices?
2452 __u16 state
= be16_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].state
);
2454 if ((state
& (DDF_Online
|DDF_Failed
|DDF_Missing
)) != DDF_Online
)
2457 rv
= xmalloc(sizeof(struct extent
) * (ddf
->max_part
+ 2));
2459 for (i
= 0; i
< ddf
->max_part
; i
++) {
2460 const struct vd_config
*bvd
;
2462 struct vcl
*v
= dl
->vlist
[i
];
2464 get_pd_index_from_refnum(v
, dl
->disk
.refnum
, ddf
->mppe
,
2465 &bvd
, &ibvd
) == DDF_NOTFOUND
)
2467 rv
[n
].start
= be64_to_cpu(LBA_OFFSET(ddf
, bvd
)[ibvd
]);
2468 rv
[n
].size
= be64_to_cpu(bvd
->blocks
);
2471 qsort(rv
, n
, sizeof(*rv
), cmp_extent
);
2473 rv
[n
].start
= be64_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].config_size
);
2479 static int init_super_ddf_bvd(struct supertype
*st
,
2480 mdu_array_info_t
*info
,
2481 unsigned long long size
,
2482 char *name
, char *homehost
,
2483 int *uuid
, unsigned long long data_offset
)
2485 /* We are creating a BVD inside a pre-existing container.
2486 * so st->sb is already set.
2487 * We need to create a new vd_config and a new virtual_entry
2489 struct ddf_super
*ddf
= st
->sb
;
2490 unsigned int venum
, i
;
2491 struct virtual_entry
*ve
;
2493 struct vd_config
*vc
;
2495 if (find_vde_by_name(ddf
, name
) != DDF_NOTFOUND
) {
2496 pr_err("This ddf already has an array called %s\n", name
);
2499 venum
= find_unused_vde(ddf
);
2500 if (venum
== DDF_NOTFOUND
) {
2501 pr_err("Cannot find spare slot for virtual disk\n");
2504 ve
= &ddf
->virt
->entries
[venum
];
2506 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
2507 * timestamp, random number
2509 make_header_guid(ve
->guid
);
2510 ve
->unit
= cpu_to_be16(info
->md_minor
);
2512 ve
->guid_crc
._v16
= crc32(0, (unsigned char *)ddf
->anchor
.guid
,
2514 ve
->type
= cpu_to_be16(0);
2515 ve
->state
= DDF_state_degraded
; /* Will be modified as devices are added */
2516 if (info
->state
& 1) /* clean */
2517 ve
->init_state
= DDF_init_full
;
2519 ve
->init_state
= DDF_init_not
;
2521 memset(ve
->pad1
, 0xff, 14);
2522 memset(ve
->name
, ' ', 16);
2524 strncpy(ve
->name
, name
, 16);
2525 ddf
->virt
->populated_vdes
=
2526 cpu_to_be16(be16_to_cpu(ddf
->virt
->populated_vdes
)+1);
2528 /* Now create a new vd_config */
2529 if (posix_memalign((void**)&vcl
, 512,
2530 (offsetof(struct vcl
, conf
) + ddf
->conf_rec_len
* 512)) != 0) {
2531 pr_err("%s could not allocate vd_config\n", __func__
);
2535 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
2538 vc
->magic
= DDF_VD_CONF_MAGIC
;
2539 memcpy(vc
->guid
, ve
->guid
, DDF_GUID_LEN
);
2540 vc
->timestamp
= cpu_to_be32(time(0)-DECADE
);
2541 vc
->seqnum
= cpu_to_be32(1);
2542 memset(vc
->pad0
, 0xff, 24);
2543 vc
->chunk_shift
= chunk_to_shift(info
->chunk_size
);
2544 if (layout_md2ddf(info
, vc
) == -1 ||
2545 be16_to_cpu(vc
->prim_elmnt_count
) > ddf
->mppe
) {
2546 pr_err("%s: unsupported RAID level/layout %d/%d with %d disks\n",
2547 __func__
, info
->level
, info
->layout
, info
->raid_disks
);
2551 vc
->sec_elmnt_seq
= 0;
2552 if (alloc_other_bvds(ddf
, vcl
) != 0) {
2553 pr_err("%s could not allocate other bvds\n",
2558 vc
->blocks
= cpu_to_be64(info
->size
* 2);
2559 vc
->array_blocks
= cpu_to_be64(
2560 calc_array_size(info
->level
, info
->raid_disks
, info
->layout
,
2561 info
->chunk_size
, info
->size
*2));
2562 memset(vc
->pad1
, 0xff, 8);
2563 vc
->spare_refs
[0] = cpu_to_be32(0xffffffff);
2564 vc
->spare_refs
[1] = cpu_to_be32(0xffffffff);
2565 vc
->spare_refs
[2] = cpu_to_be32(0xffffffff);
2566 vc
->spare_refs
[3] = cpu_to_be32(0xffffffff);
2567 vc
->spare_refs
[4] = cpu_to_be32(0xffffffff);
2568 vc
->spare_refs
[5] = cpu_to_be32(0xffffffff);
2569 vc
->spare_refs
[6] = cpu_to_be32(0xffffffff);
2570 vc
->spare_refs
[7] = cpu_to_be32(0xffffffff);
2571 memset(vc
->cache_pol
, 0, 8);
2573 memset(vc
->pad2
, 0xff, 3);
2574 memset(vc
->pad3
, 0xff, 52);
2575 memset(vc
->pad4
, 0xff, 192);
2576 memset(vc
->v0
, 0xff, 32);
2577 memset(vc
->v1
, 0xff, 32);
2578 memset(vc
->v2
, 0xff, 16);
2579 memset(vc
->v3
, 0xff, 16);
2580 memset(vc
->vendor
, 0xff, 32);
2582 memset(vc
->phys_refnum
, 0xff, 4*ddf
->mppe
);
2583 memset(vc
->phys_refnum
+ddf
->mppe
, 0x00, 8*ddf
->mppe
);
2585 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++) {
2586 memcpy(vcl
->other_bvds
[i
-1], vc
, ddf
->conf_rec_len
* 512);
2587 vcl
->other_bvds
[i
-1]->sec_elmnt_seq
= i
;
2590 vcl
->next
= ddf
->conflist
;
2591 ddf
->conflist
= vcl
;
2592 ddf
->currentconf
= vcl
;
2593 ddf_set_updates_pending(ddf
);
2599 static int get_svd_state(const struct ddf_super
*, const struct vcl
*);
2601 static void add_to_super_ddf_bvd(struct supertype
*st
,
2602 mdu_disk_info_t
*dk
, int fd
, char *devname
)
2604 /* fd and devname identify a device with-in the ddf container (st).
2605 * dk identifies a location in the new BVD.
2606 * We need to find suitable free space in that device and update
2607 * the phys_refnum and lba_offset for the newly created vd_config.
2608 * We might also want to update the type in the phys_disk
2611 * Alternately: fd == -1 and we have already chosen which device to
2612 * use and recorded in dlist->raid_disk;
2615 struct ddf_super
*ddf
= st
->sb
;
2616 struct vd_config
*vc
;
2618 unsigned long long blocks
, pos
, esize
;
2620 unsigned int raid_disk
= dk
->raid_disk
;
2623 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2624 if (dl
->raiddisk
== dk
->raid_disk
)
2627 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2628 if (dl
->major
== dk
->major
&&
2629 dl
->minor
== dk
->minor
)
2632 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
2635 vc
= &ddf
->currentconf
->conf
;
2636 if (vc
->sec_elmnt_count
> 1) {
2637 unsigned int n
= be16_to_cpu(vc
->prim_elmnt_count
);
2639 vc
= ddf
->currentconf
->other_bvds
[raid_disk
/ n
- 1];
2643 ex
= get_extents(ddf
, dl
);
2648 blocks
= be64_to_cpu(vc
->blocks
);
2649 if (ddf
->currentconf
->block_sizes
)
2650 blocks
= ddf
->currentconf
->block_sizes
[dk
->raid_disk
];
2653 esize
= ex
[i
].start
- pos
;
2654 if (esize
>= blocks
)
2656 pos
= ex
[i
].start
+ ex
[i
].size
;
2658 } while (ex
[i
-1].size
);
2664 ddf
->currentdev
= dk
->raid_disk
;
2665 vc
->phys_refnum
[raid_disk
] = dl
->disk
.refnum
;
2666 LBA_OFFSET(ddf
, vc
)[raid_disk
] = cpu_to_be64(pos
);
2668 for (i
= 0; i
< ddf
->max_part
; i
++)
2669 if (dl
->vlist
[i
] == NULL
)
2671 if (i
== ddf
->max_part
)
2673 dl
->vlist
[i
] = ddf
->currentconf
;
2678 dl
->devname
= devname
;
2680 /* Check if we can mark array as optimal yet */
2681 i
= ddf
->currentconf
->vcnum
;
2682 ddf
->virt
->entries
[i
].state
=
2683 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
2684 | get_svd_state(ddf
, ddf
->currentconf
);
2685 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
2686 cpu_to_be16(DDF_Global_Spare
));
2687 be16_set(ddf
->phys
->entries
[dl
->pdnum
].type
,
2688 cpu_to_be16(DDF_Active_in_VD
));
2689 dprintf("%s: added disk %d/%08x to VD %d/%s as disk %d\n",
2690 __func__
, dl
->pdnum
, be32_to_cpu(dl
->disk
.refnum
),
2691 ddf
->currentconf
->vcnum
, guid_str(vc
->guid
),
2693 ddf_set_updates_pending(ddf
);
2696 static unsigned int find_unused_pde(const struct ddf_super
*ddf
)
2699 for (i
= 0; i
< be16_to_cpu(ddf
->phys
->max_pdes
); i
++) {
2700 if (all_ff(ddf
->phys
->entries
[i
].guid
))
2703 return DDF_NOTFOUND
;
2706 static void _set_config_size(struct phys_disk_entry
*pde
, const struct dl
*dl
)
2709 cfs
= min(dl
->size
- 32*1024*2ULL, be64_to_cpu(dl
->primary_lba
));
2710 t
= be64_to_cpu(dl
->secondary_lba
);
2714 * Some vendor DDF structures interpret workspace_lba
2715 * very differently then us. Make a sanity check on the value.
2717 t
= be64_to_cpu(dl
->workspace_lba
);
2719 __u64 wsp
= cfs
- t
;
2720 if (wsp
> 1024*1024*2ULL && wsp
> dl
->size
/ 16) {
2721 pr_err("%s: %x:%x: workspace size 0x%llx too big, ignoring\n",
2722 __func__
, dl
->major
, dl
->minor
, wsp
);
2726 pde
->config_size
= cpu_to_be64(cfs
);
2727 dprintf("%s: %x:%x config_size %llx, DDF structure is %llx blocks\n",
2728 __func__
, dl
->major
, dl
->minor
, cfs
, dl
->size
-cfs
);
2731 /* add a device to a container, either while creating it or while
2732 * expanding a pre-existing container
2734 static int add_to_super_ddf(struct supertype
*st
,
2735 mdu_disk_info_t
*dk
, int fd
, char *devname
,
2736 unsigned long long data_offset
)
2738 struct ddf_super
*ddf
= st
->sb
;
2742 unsigned long long size
;
2743 struct phys_disk_entry
*pde
;
2748 if (ddf
->currentconf
) {
2749 add_to_super_ddf_bvd(st
, dk
, fd
, devname
);
2753 /* This is device numbered dk->number. We need to create
2754 * a phys_disk entry and a more detailed disk_data entry.
2757 n
= find_unused_pde(ddf
);
2758 if (n
== DDF_NOTFOUND
) {
2759 pr_err("%s: No free slot in array, cannot add disk\n",
2763 pde
= &ddf
->phys
->entries
[n
];
2764 get_dev_size(fd
, NULL
, &size
);
2765 if (size
<= 32*1024*1024) {
2766 pr_err("%s: device size must be at least 32MB\n",
2772 if (posix_memalign((void**)&dd
, 512,
2773 sizeof(*dd
) + sizeof(dd
->vlist
[0]) * ddf
->max_part
) != 0) {
2774 pr_err("%s could allocate buffer for new disk, aborting\n",
2778 dd
->major
= major(stb
.st_rdev
);
2779 dd
->minor
= minor(stb
.st_rdev
);
2780 dd
->devname
= devname
;
2784 dd
->disk
.magic
= DDF_PHYS_DATA_MAGIC
;
2786 tm
= localtime(&now
);
2787 sprintf(dd
->disk
.guid
, "%8s%04d%02d%02d",
2788 T10
, tm
->tm_year
+1900, tm
->tm_mon
+1, tm
->tm_mday
);
2789 tptr
= (__u32
*)(dd
->disk
.guid
+ 16);
2790 *tptr
++ = random32();
2794 /* Cannot be bothered finding a CRC of some irrelevant details*/
2795 dd
->disk
.refnum
._v32
= random32();
2796 for (i
= be16_to_cpu(ddf
->active
->max_pd_entries
);
2798 if (be32_eq(ddf
->phys
->entries
[i
-1].refnum
,
2803 dd
->disk
.forced_ref
= 1;
2804 dd
->disk
.forced_guid
= 1;
2805 memset(dd
->disk
.vendor
, ' ', 32);
2806 memcpy(dd
->disk
.vendor
, "Linux", 5);
2807 memset(dd
->disk
.pad
, 0xff, 442);
2808 for (i
= 0; i
< ddf
->max_part
; i
++)
2809 dd
->vlist
[i
] = NULL
;
2813 if (st
->update_tail
) {
2814 int len
= (sizeof(struct phys_disk
) +
2815 sizeof(struct phys_disk_entry
));
2816 struct phys_disk
*pd
;
2819 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2820 pd
->used_pdes
= cpu_to_be16(n
);
2821 pde
= &pd
->entries
[0];
2824 ddf
->phys
->used_pdes
= cpu_to_be16(
2825 1 + be16_to_cpu(ddf
->phys
->used_pdes
));
2827 memcpy(pde
->guid
, dd
->disk
.guid
, DDF_GUID_LEN
);
2828 pde
->refnum
= dd
->disk
.refnum
;
2829 pde
->type
= cpu_to_be16(DDF_Forced_PD_GUID
| DDF_Global_Spare
);
2830 pde
->state
= cpu_to_be16(DDF_Online
);
2833 * If there is already a device in dlist, try to reserve the same
2834 * amount of workspace. Otherwise, use 32MB.
2835 * We checked disk size above already.
2837 #define __calc_lba(new, old, lba, mb) do { \
2838 unsigned long long dif; \
2839 if ((old) != NULL) \
2840 dif = (old)->size - be64_to_cpu((old)->lba); \
2842 dif = (new)->size; \
2843 if ((new)->size > dif) \
2844 (new)->lba = cpu_to_be64((new)->size - dif); \
2846 (new)->lba = cpu_to_be64((new)->size - (mb*1024*2)); \
2848 __calc_lba(dd
, ddf
->dlist
, workspace_lba
, 32);
2849 __calc_lba(dd
, ddf
->dlist
, primary_lba
, 16);
2850 if (ddf
->dlist
== NULL
||
2851 be64_to_cpu(ddf
->dlist
->secondary_lba
) != ~(__u64
)0)
2852 __calc_lba(dd
, ddf
->dlist
, secondary_lba
, 32);
2853 _set_config_size(pde
, dd
);
2855 sprintf(pde
->path
, "%17.17s","Information: nil") ;
2856 memset(pde
->pad
, 0xff, 6);
2858 if (st
->update_tail
) {
2859 dd
->next
= ddf
->add_list
;
2862 dd
->next
= ddf
->dlist
;
2864 ddf_set_updates_pending(ddf
);
2870 static int remove_from_super_ddf(struct supertype
*st
, mdu_disk_info_t
*dk
)
2872 struct ddf_super
*ddf
= st
->sb
;
2875 /* mdmon has noticed that this disk (dk->major/dk->minor) has
2876 * disappeared from the container.
2877 * We need to arrange that it disappears from the metadata and
2878 * internal data structures too.
2879 * Most of the work is done by ddf_process_update which edits
2880 * the metadata and closes the file handle and attaches the memory
2881 * where free_updates will free it.
2883 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2884 if (dl
->major
== dk
->major
&&
2885 dl
->minor
== dk
->minor
)
2890 if (st
->update_tail
) {
2891 int len
= (sizeof(struct phys_disk
) +
2892 sizeof(struct phys_disk_entry
));
2893 struct phys_disk
*pd
;
2896 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2897 pd
->used_pdes
= cpu_to_be16(dl
->pdnum
);
2898 pd
->entries
[0].state
= cpu_to_be16(DDF_Missing
);
2899 append_metadata_update(st
, pd
, len
);
2906 * This is the write_init_super method for a ddf container. It is
2907 * called when creating a container or adding another device to a
2910 #define NULL_CONF_SZ 4096
2912 static int __write_ddf_structure(struct dl
*d
, struct ddf_super
*ddf
, __u8 type
)
2914 unsigned long long sector
;
2915 struct ddf_header
*header
;
2916 int fd
, i
, n_config
, conf_size
, buf_size
;
2923 case DDF_HEADER_PRIMARY
:
2924 header
= &ddf
->primary
;
2925 sector
= be64_to_cpu(header
->primary_lba
);
2927 case DDF_HEADER_SECONDARY
:
2928 header
= &ddf
->secondary
;
2929 sector
= be64_to_cpu(header
->secondary_lba
);
2934 if (sector
== ~(__u64
)0)
2937 header
->type
= type
;
2938 header
->openflag
= 1;
2939 header
->crc
= calc_crc(header
, 512);
2941 lseek64(fd
, sector
<<9, 0);
2942 if (write(fd
, header
, 512) < 0)
2945 ddf
->controller
.crc
= calc_crc(&ddf
->controller
, 512);
2946 if (write(fd
, &ddf
->controller
, 512) < 0)
2949 ddf
->phys
->crc
= calc_crc(ddf
->phys
, ddf
->pdsize
);
2950 if (write(fd
, ddf
->phys
, ddf
->pdsize
) < 0)
2952 ddf
->virt
->crc
= calc_crc(ddf
->virt
, ddf
->vdsize
);
2953 if (write(fd
, ddf
->virt
, ddf
->vdsize
) < 0)
2956 /* Now write lots of config records. */
2957 n_config
= ddf
->max_part
;
2958 conf_size
= ddf
->conf_rec_len
* 512;
2960 buf_size
= conf_size
* (n_config
+ 1);
2962 if (posix_memalign((void**)&conf
, 512, buf_size
) != 0)
2966 for (i
= 0 ; i
<= n_config
; i
++) {
2968 struct vd_config
*vdc
= NULL
;
2969 if (i
== n_config
) {
2970 c
= (struct vcl
*)d
->spare
;
2977 get_pd_index_from_refnum(
2980 (const struct vd_config
**)&vdc
,
2984 dprintf("writing conf record %i on disk %08x for %s/%u\n",
2985 i
, be32_to_cpu(d
->disk
.refnum
),
2986 guid_str(vdc
->guid
),
2987 vdc
->sec_elmnt_seq
);
2988 vdc
->seqnum
= header
->seq
;
2989 vdc
->crc
= calc_crc(vdc
, conf_size
);
2990 memcpy(conf
+ i
*conf_size
, vdc
, conf_size
);
2992 memset(conf
+ i
*conf_size
, 0xff, conf_size
);
2994 if (write(fd
, conf
, buf_size
) != buf_size
)
2997 d
->disk
.crc
= calc_crc(&d
->disk
, 512);
2998 if (write(fd
, &d
->disk
, 512) < 0)
3003 header
->openflag
= 0;
3004 header
->crc
= calc_crc(header
, 512);
3006 lseek64(fd
, sector
<<9, 0);
3007 if (write(fd
, header
, 512) < 0)
3013 static int _write_super_to_disk(struct ddf_super
*ddf
, struct dl
*d
)
3015 unsigned long long size
;
3020 /* We need to fill in the primary, (secondary) and workspace
3021 * lba's in the headers, set their checksums,
3022 * Also checksum phys, virt....
3024 * Then write everything out, finally the anchor is written.
3026 get_dev_size(fd
, NULL
, &size
);
3028 if (be64_to_cpu(d
->workspace_lba
) != 0ULL)
3029 ddf
->anchor
.workspace_lba
= d
->workspace_lba
;
3031 ddf
->anchor
.workspace_lba
=
3032 cpu_to_be64(size
- 32*1024*2);
3033 if (be64_to_cpu(d
->primary_lba
) != 0ULL)
3034 ddf
->anchor
.primary_lba
= d
->primary_lba
;
3036 ddf
->anchor
.primary_lba
=
3037 cpu_to_be64(size
- 16*1024*2);
3038 if (be64_to_cpu(d
->secondary_lba
) != 0ULL)
3039 ddf
->anchor
.secondary_lba
= d
->secondary_lba
;
3041 ddf
->anchor
.secondary_lba
=
3042 cpu_to_be64(size
- 32*1024*2);
3043 ddf
->anchor
.seq
= ddf
->active
->seq
;
3044 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
3045 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
3047 ddf
->anchor
.openflag
= 0xFF; /* 'open' means nothing */
3048 ddf
->anchor
.seq
= cpu_to_be32(0xFFFFFFFF); /* no sequencing in anchor */
3049 ddf
->anchor
.crc
= calc_crc(&ddf
->anchor
, 512);
3051 if (!__write_ddf_structure(d
, ddf
, DDF_HEADER_PRIMARY
))
3054 if (!__write_ddf_structure(d
, ddf
, DDF_HEADER_SECONDARY
))
3057 lseek64(fd
, (size
-1)*512, SEEK_SET
);
3058 if (write(fd
, &ddf
->anchor
, 512) < 0)
3065 static int __write_init_super_ddf(struct supertype
*st
)
3067 struct ddf_super
*ddf
= st
->sb
;
3072 pr_state(ddf
, __func__
);
3074 /* try to write updated metadata,
3075 * if we catch a failure move on to the next disk
3077 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
3079 successes
+= _write_super_to_disk(ddf
, d
);
3082 return attempts
!= successes
;
3085 static int write_init_super_ddf(struct supertype
*st
)
3087 struct ddf_super
*ddf
= st
->sb
;
3088 struct vcl
*currentconf
= ddf
->currentconf
;
3090 /* we are done with currentconf reset it to point st at the container */
3091 ddf
->currentconf
= NULL
;
3093 if (st
->update_tail
) {
3094 /* queue the virtual_disk and vd_config as metadata updates */
3095 struct virtual_disk
*vd
;
3096 struct vd_config
*vc
;
3101 int len
= (sizeof(struct phys_disk
) +
3102 sizeof(struct phys_disk_entry
));
3104 /* adding a disk to the container. */
3108 append_metadata_update(st
, ddf
->add_list
->mdupdate
, len
);
3109 ddf
->add_list
->mdupdate
= NULL
;
3113 /* Newly created VD */
3115 /* First the virtual disk. We have a slightly fake header */
3116 len
= sizeof(struct virtual_disk
) + sizeof(struct virtual_entry
);
3119 vd
->entries
[0] = ddf
->virt
->entries
[currentconf
->vcnum
];
3120 vd
->populated_vdes
= cpu_to_be16(currentconf
->vcnum
);
3121 append_metadata_update(st
, vd
, len
);
3123 /* Then the vd_config */
3124 len
= ddf
->conf_rec_len
* 512;
3125 tlen
= len
* currentconf
->conf
.sec_elmnt_count
;
3127 memcpy(vc
, ¤tconf
->conf
, len
);
3128 for (i
= 1; i
< currentconf
->conf
.sec_elmnt_count
; i
++)
3129 memcpy((char *)vc
+ i
*len
, currentconf
->other_bvds
[i
-1],
3131 append_metadata_update(st
, vc
, tlen
);
3133 /* FIXME I need to close the fds! */
3138 for (d
= ddf
->dlist
; d
; d
=d
->next
)
3139 while (Kill(d
->devname
, NULL
, 0, -1, 1) == 0);
3140 return __write_init_super_ddf(st
);
3146 static __u64
avail_size_ddf(struct supertype
*st
, __u64 devsize
,
3147 unsigned long long data_offset
)
3149 /* We must reserve the last 32Meg */
3150 if (devsize
<= 32*1024*2)
3152 return devsize
- 32*1024*2;
3157 static int reserve_space(struct supertype
*st
, int raiddisks
,
3158 unsigned long long size
, int chunk
,
3159 unsigned long long *freesize
)
3161 /* Find 'raiddisks' spare extents at least 'size' big (but
3162 * only caring about multiples of 'chunk') and remember
3164 * If the cannot be found, fail.
3167 struct ddf_super
*ddf
= st
->sb
;
3170 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3174 /* Now find largest extent on each device */
3175 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3176 struct extent
*e
= get_extents(ddf
, dl
);
3177 unsigned long long pos
= 0;
3180 unsigned long long minsize
= size
;
3188 unsigned long long esize
;
3189 esize
= e
[i
].start
- pos
;
3190 if (esize
>= minsize
) {
3194 pos
= e
[i
].start
+ e
[i
].size
;
3196 } while (e
[i
-1].size
);
3199 dl
->esize
= minsize
;
3203 if (cnt
< raiddisks
) {
3204 pr_err("not enough devices with space to create array.\n");
3205 return 0; /* No enough free spaces large enough */
3208 /* choose the largest size of which there are at least 'raiddisk' */
3209 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3211 if (dl
->esize
<= size
)
3213 /* This is bigger than 'size', see if there are enough */
3215 for (dl2
= ddf
->dlist
; dl2
; dl2
=dl2
->next
)
3216 if (dl2
->esize
>= dl
->esize
)
3218 if (cnt
>= raiddisks
)
3222 size
= size
/ chunk
;
3227 pr_err("not enough spare devices to create array.\n");
3231 /* We have a 'size' of which there are enough spaces.
3232 * We simply do a first-fit */
3234 for (dl
= ddf
->dlist
; dl
&& cnt
< raiddisks
; dl
=dl
->next
) {
3235 if (dl
->esize
< size
)
3245 validate_geometry_ddf_container(struct supertype
*st
,
3246 int level
, int layout
, int raiddisks
,
3247 int chunk
, unsigned long long size
,
3248 unsigned long long data_offset
,
3249 char *dev
, unsigned long long *freesize
,
3252 static int validate_geometry_ddf_bvd(struct supertype
*st
,
3253 int level
, int layout
, int raiddisks
,
3254 int *chunk
, unsigned long long size
,
3255 unsigned long long data_offset
,
3256 char *dev
, unsigned long long *freesize
,
3259 static int validate_geometry_ddf(struct supertype
*st
,
3260 int level
, int layout
, int raiddisks
,
3261 int *chunk
, unsigned long long size
,
3262 unsigned long long data_offset
,
3263 char *dev
, unsigned long long *freesize
,
3270 /* ddf potentially supports lots of things, but it depends on
3271 * what devices are offered (and maybe kernel version?)
3272 * If given unused devices, we will make a container.
3273 * If given devices in a container, we will make a BVD.
3274 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
3277 if (*chunk
== UnSet
)
3278 *chunk
= DEFAULT_CHUNK
;
3280 if (level
== -1000000) level
= LEVEL_CONTAINER
;
3281 if (level
== LEVEL_CONTAINER
) {
3282 /* Must be a fresh device to add to a container */
3283 return validate_geometry_ddf_container(st
, level
, layout
,
3285 size
, data_offset
, dev
,
3291 mdu_array_info_t array
= {
3292 .level
= level
, .layout
= layout
,
3293 .raid_disks
= raiddisks
3295 struct vd_config conf
;
3296 if (layout_md2ddf(&array
, &conf
) == -1) {
3298 pr_err("DDF does not support level %d /layout %d arrays with %d disks\n",
3299 level
, layout
, raiddisks
);
3302 /* Should check layout? etc */
3304 if (st
->sb
&& freesize
) {
3305 /* --create was given a container to create in.
3306 * So we need to check that there are enough
3307 * free spaces and return the amount of space.
3308 * We may as well remember which drives were
3309 * chosen so that add_to_super/getinfo_super
3312 return reserve_space(st
, raiddisks
, size
, *chunk
, freesize
);
3318 /* A container has already been opened, so we are
3319 * creating in there. Maybe a BVD, maybe an SVD.
3320 * Should make a distinction one day.
3322 return validate_geometry_ddf_bvd(st
, level
, layout
, raiddisks
,
3323 chunk
, size
, data_offset
, dev
,
3327 /* This is the first device for the array.
3328 * If it is a container, we read it in and do automagic allocations,
3329 * no other devices should be given.
3330 * Otherwise it must be a member device of a container, and we
3331 * do manual allocation.
3332 * Later we should check for a BVD and make an SVD.
3334 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3336 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
3338 if (sra
&& sra
->array
.major_version
== -1 &&
3339 strcmp(sra
->text_version
, "ddf") == 0) {
3342 /* find space for 'n' devices. */
3343 /* remember the devices */
3344 /* Somehow return the fact that we have enough */
3348 pr_err("ddf: Cannot create this array "
3349 "on device %s - a container is required.\n",
3353 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
3355 pr_err("ddf: Cannot open %s: %s\n",
3356 dev
, strerror(errno
));
3359 /* Well, it is in use by someone, maybe a 'ddf' container. */
3360 cfd
= open_container(fd
);
3364 pr_err("ddf: Cannot use %s: %s\n",
3365 dev
, strerror(EBUSY
));
3368 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
3370 if (sra
&& sra
->array
.major_version
== -1 &&
3371 strcmp(sra
->text_version
, "ddf") == 0) {
3372 /* This is a member of a ddf container. Load the container
3373 * and try to create a bvd
3375 struct ddf_super
*ddf
;
3376 if (load_super_ddf_all(st
, cfd
, (void **)&ddf
, NULL
) == 0) {
3378 strcpy(st
->container_devnm
, fd2devnm(cfd
));
3380 return validate_geometry_ddf_bvd(st
, level
, layout
,
3381 raiddisks
, chunk
, size
,
3387 } else /* device may belong to a different container */
3394 validate_geometry_ddf_container(struct supertype
*st
,
3395 int level
, int layout
, int raiddisks
,
3396 int chunk
, unsigned long long size
,
3397 unsigned long long data_offset
,
3398 char *dev
, unsigned long long *freesize
,
3402 unsigned long long ldsize
;
3404 if (level
!= LEVEL_CONTAINER
)
3409 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3412 pr_err("ddf: Cannot open %s: %s\n",
3413 dev
, strerror(errno
));
3416 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3422 *freesize
= avail_size_ddf(st
, ldsize
>> 9, INVALID_SECTORS
);
3429 static int validate_geometry_ddf_bvd(struct supertype
*st
,
3430 int level
, int layout
, int raiddisks
,
3431 int *chunk
, unsigned long long size
,
3432 unsigned long long data_offset
,
3433 char *dev
, unsigned long long *freesize
,
3437 struct ddf_super
*ddf
= st
->sb
;
3439 unsigned long long pos
= 0;
3440 unsigned long long maxsize
;
3443 /* ddf/bvd supports lots of things, but not containers */
3444 if (level
== LEVEL_CONTAINER
) {
3446 pr_err("DDF cannot create a container within an container\n");
3449 /* We must have the container info already read in. */
3454 /* General test: make sure there is space for
3455 * 'raiddisks' device extents of size 'size'.
3457 unsigned long long minsize
= size
;
3461 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
3467 e
= get_extents(ddf
, dl
);
3470 unsigned long long esize
;
3471 esize
= e
[i
].start
- pos
;
3472 if (esize
>= minsize
)
3474 pos
= e
[i
].start
+ e
[i
].size
;
3476 } while (e
[i
-1].size
);
3481 if (dcnt
< raiddisks
) {
3483 pr_err("ddf: Not enough devices with "
3484 "space for this array (%d < %d)\n",
3490 /* This device must be a member of the set */
3491 if (stat(dev
, &stb
) < 0)
3493 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3495 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
3496 if (dl
->major
== (int)major(stb
.st_rdev
) &&
3497 dl
->minor
== (int)minor(stb
.st_rdev
))
3502 pr_err("ddf: %s is not in the "
3507 e
= get_extents(ddf
, dl
);
3511 unsigned long long esize
;
3512 esize
= e
[i
].start
- pos
;
3513 if (esize
>= maxsize
)
3515 pos
= e
[i
].start
+ e
[i
].size
;
3517 } while (e
[i
-1].size
);
3518 *freesize
= maxsize
;
3524 static int load_super_ddf_all(struct supertype
*st
, int fd
,
3525 void **sbp
, char *devname
)
3528 struct ddf_super
*super
;
3529 struct mdinfo
*sd
, *best
= NULL
;
3535 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3538 if (sra
->array
.major_version
!= -1 ||
3539 sra
->array
.minor_version
!= -2 ||
3540 strcmp(sra
->text_version
, "ddf") != 0)
3543 if (posix_memalign((void**)&super
, 512, sizeof(*super
)) != 0)
3545 memset(super
, 0, sizeof(*super
));
3547 /* first, try each device, and choose the best ddf */
3548 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3550 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3551 dfd
= dev_open(nm
, O_RDONLY
);
3554 rv
= load_ddf_headers(dfd
, super
, NULL
);
3557 seq
= be32_to_cpu(super
->active
->seq
);
3558 if (super
->active
->openflag
)
3560 if (!best
|| seq
> bestseq
) {
3568 /* OK, load this ddf */
3569 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
3570 dfd
= dev_open(nm
, O_RDONLY
);
3573 load_ddf_headers(dfd
, super
, NULL
);
3574 load_ddf_global(dfd
, super
, NULL
);
3576 /* Now we need the device-local bits */
3577 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3580 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3581 dfd
= dev_open(nm
, O_RDWR
);
3584 rv
= load_ddf_headers(dfd
, super
, NULL
);
3586 rv
= load_ddf_local(dfd
, super
, NULL
, 1);
3592 if (st
->ss
== NULL
) {
3593 st
->ss
= &super_ddf
;
3594 st
->minor_version
= 0;
3597 strcpy(st
->container_devnm
, fd2devnm(fd
));
3601 static int load_container_ddf(struct supertype
*st
, int fd
,
3604 return load_super_ddf_all(st
, fd
, &st
->sb
, devname
);
3607 #endif /* MDASSEMBLE */
3609 static int check_secondary(const struct vcl
*vc
)
3611 const struct vd_config
*conf
= &vc
->conf
;
3614 /* The only DDF secondary RAID level md can support is
3615 * RAID 10, if the stripe sizes and Basic volume sizes
3617 * Other configurations could in theory be supported by exposing
3618 * the BVDs to user space and using device mapper for the secondary
3619 * mapping. So far we don't support that.
3622 __u64 sec_elements
[4] = {0, 0, 0, 0};
3623 #define __set_sec_seen(n) (sec_elements[(n)>>6] |= (1<<((n)&63)))
3624 #define __was_sec_seen(n) ((sec_elements[(n)>>6] & (1<<((n)&63))) != 0)
3626 if (vc
->other_bvds
== NULL
) {
3627 pr_err("No BVDs for secondary RAID found\n");
3630 if (conf
->prl
!= DDF_RAID1
) {
3631 pr_err("Secondary RAID level only supported for mirrored BVD\n");
3634 if (conf
->srl
!= DDF_2STRIPED
&& conf
->srl
!= DDF_2SPANNED
) {
3635 pr_err("Secondary RAID level %d is unsupported\n",
3639 __set_sec_seen(conf
->sec_elmnt_seq
);
3640 for (i
= 0; i
< conf
->sec_elmnt_count
-1; i
++) {
3641 const struct vd_config
*bvd
= vc
->other_bvds
[i
];
3642 if (bvd
->sec_elmnt_seq
== DDF_UNUSED_BVD
)
3644 if (bvd
->srl
!= conf
->srl
) {
3645 pr_err("Inconsistent secondary RAID level across BVDs\n");
3648 if (bvd
->prl
!= conf
->prl
) {
3649 pr_err("Different RAID levels for BVDs are unsupported\n");
3652 if (!be16_eq(bvd
->prim_elmnt_count
, conf
->prim_elmnt_count
)) {
3653 pr_err("All BVDs must have the same number of primary elements\n");
3656 if (bvd
->chunk_shift
!= conf
->chunk_shift
) {
3657 pr_err("Different strip sizes for BVDs are unsupported\n");
3660 if (!be64_eq(bvd
->array_blocks
, conf
->array_blocks
)) {
3661 pr_err("Different BVD sizes are unsupported\n");
3664 __set_sec_seen(bvd
->sec_elmnt_seq
);
3666 for (i
= 0; i
< conf
->sec_elmnt_count
; i
++) {
3667 if (!__was_sec_seen(i
)) {
3668 pr_err("BVD %d is missing\n", i
);
3675 static unsigned int get_pd_index_from_refnum(const struct vcl
*vc
,
3676 be32 refnum
, unsigned int nmax
,
3677 const struct vd_config
**bvd
,
3680 unsigned int i
, j
, n
, sec
, cnt
;
3682 cnt
= be16_to_cpu(vc
->conf
.prim_elmnt_count
);
3683 sec
= (vc
->conf
.sec_elmnt_count
== 1 ? 0 : vc
->conf
.sec_elmnt_seq
);
3685 for (i
= 0, j
= 0 ; i
< nmax
; i
++) {
3686 /* j counts valid entries for this BVD */
3687 if (be32_to_cpu(vc
->conf
.phys_refnum
[i
]) != 0xffffffff)
3689 if (be32_eq(vc
->conf
.phys_refnum
[i
], refnum
)) {
3692 return sec
* cnt
+ j
- 1;
3695 if (vc
->other_bvds
== NULL
)
3698 for (n
= 1; n
< vc
->conf
.sec_elmnt_count
; n
++) {
3699 struct vd_config
*vd
= vc
->other_bvds
[n
-1];
3700 sec
= vd
->sec_elmnt_seq
;
3701 if (sec
== DDF_UNUSED_BVD
)
3703 for (i
= 0, j
= 0 ; i
< nmax
; i
++) {
3704 if (be32_to_cpu(vd
->phys_refnum
[i
]) != 0xffffffff)
3706 if (be32_eq(vd
->phys_refnum
[i
], refnum
)) {
3709 return sec
* cnt
+ j
- 1;
3715 return DDF_NOTFOUND
;
3718 static struct mdinfo
*container_content_ddf(struct supertype
*st
, char *subarray
)
3720 /* Given a container loaded by load_super_ddf_all,
3721 * extract information about all the arrays into
3724 * For each vcl in conflist: create an mdinfo, fill it in,
3725 * then look for matching devices (phys_refnum) in dlist
3726 * and create appropriate device mdinfo.
3728 struct ddf_super
*ddf
= st
->sb
;
3729 struct mdinfo
*rest
= NULL
;
3732 for (vc
= ddf
->conflist
; vc
; vc
=vc
->next
)
3735 struct mdinfo
*this;
3741 (strtoul(subarray
, &ep
, 10) != vc
->vcnum
||
3745 if (vc
->conf
.sec_elmnt_count
> 1) {
3746 if (check_secondary(vc
) != 0)
3750 this = xcalloc(1, sizeof(*this));
3754 if (layout_ddf2md(&vc
->conf
, &this->array
))
3756 this->array
.md_minor
= -1;
3757 this->array
.major_version
= -1;
3758 this->array
.minor_version
= -2;
3759 this->safe_mode_delay
= DDF_SAFE_MODE_DELAY
;
3760 cptr
= (__u32
*)(vc
->conf
.guid
+ 16);
3761 this->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
3762 this->array
.utime
= DECADE
+
3763 be32_to_cpu(vc
->conf
.timestamp
);
3764 this->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
3767 if ((ddf
->virt
->entries
[i
].state
& DDF_state_inconsistent
) ||
3768 (ddf
->virt
->entries
[i
].init_state
& DDF_initstate_mask
) !=
3770 this->array
.state
= 0;
3771 this->resync_start
= 0;
3773 this->array
.state
= 1;
3774 this->resync_start
= MaxSector
;
3776 _ddf_array_name(this->name
, ddf
, i
);
3777 memset(this->uuid
, 0, sizeof(this->uuid
));
3778 this->component_size
= be64_to_cpu(vc
->conf
.blocks
);
3779 this->array
.size
= this->component_size
/ 2;
3780 this->container_member
= i
;
3782 ddf
->currentconf
= vc
;
3783 uuid_from_super_ddf(st
, this->uuid
);
3785 ddf
->currentconf
= NULL
;
3787 sprintf(this->text_version
, "/%s/%d",
3788 st
->container_devnm
, this->container_member
);
3790 for (pd
= 0; pd
< be16_to_cpu(ddf
->phys
->used_pdes
); pd
++) {
3793 const struct vd_config
*bvd
;
3797 if (be32_to_cpu(ddf
->phys
->entries
[pd
].refnum
)
3801 stt
= be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
3802 if ((stt
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
3806 i
= get_pd_index_from_refnum(
3807 vc
, ddf
->phys
->entries
[pd
].refnum
,
3808 ddf
->mppe
, &bvd
, &iphys
);
3809 if (i
== DDF_NOTFOUND
)
3812 this->array
.working_disks
++;
3814 for (d
= ddf
->dlist
; d
; d
=d
->next
)
3815 if (be32_eq(d
->disk
.refnum
,
3816 ddf
->phys
->entries
[pd
].refnum
))
3819 /* Haven't found that one yet, maybe there are others */
3822 dev
= xcalloc(1, sizeof(*dev
));
3823 dev
->next
= this->devs
;
3826 dev
->disk
.number
= be32_to_cpu(d
->disk
.refnum
);
3827 dev
->disk
.major
= d
->major
;
3828 dev
->disk
.minor
= d
->minor
;
3829 dev
->disk
.raid_disk
= i
;
3830 dev
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
3831 dev
->recovery_start
= MaxSector
;
3833 dev
->events
= be32_to_cpu(ddf
->primary
.seq
);
3835 be64_to_cpu(LBA_OFFSET(ddf
, bvd
)[iphys
]);
3836 dev
->component_size
= be64_to_cpu(bvd
->blocks
);
3838 strcpy(dev
->name
, d
->devname
);
3844 static int store_super_ddf(struct supertype
*st
, int fd
)
3846 struct ddf_super
*ddf
= st
->sb
;
3847 unsigned long long dsize
;
3854 if (!get_dev_size(fd
, NULL
, &dsize
))
3857 if (ddf
->dlist
|| ddf
->conflist
) {
3862 if (fstat(fd
, &sta
) == -1 || !S_ISBLK(sta
.st_mode
)) {
3863 pr_err("%s: file descriptor for invalid device\n",
3867 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
3868 if (dl
->major
== (int)major(sta
.st_rdev
) &&
3869 dl
->minor
== (int)minor(sta
.st_rdev
))
3872 pr_err("%s: couldn't find disk %d/%d\n", __func__
,
3873 (int)major(sta
.st_rdev
),
3874 (int)minor(sta
.st_rdev
));
3879 ret
= (_write_super_to_disk(ddf
, dl
) != 1);
3884 if (posix_memalign(&buf
, 512, 512) != 0)
3886 memset(buf
, 0, 512);
3888 lseek64(fd
, dsize
-512, 0);
3889 rc
= write(fd
, buf
, 512);
3896 static int compare_super_ddf(struct supertype
*st
, struct supertype
*tst
)
3900 * 0 same, or first was empty, and second was copied
3901 * 1 second had wrong number
3903 * 3 wrong other info
3905 struct ddf_super
*first
= st
->sb
;
3906 struct ddf_super
*second
= tst
->sb
;
3907 struct dl
*dl1
, *dl2
;
3908 struct vcl
*vl1
, *vl2
;
3909 unsigned int max_vds
, max_pds
, pd
, vd
;
3917 if (memcmp(first
->anchor
.guid
, second
->anchor
.guid
, DDF_GUID_LEN
) != 0)
3920 if (!be32_eq(first
->active
->seq
, second
->active
->seq
)) {
3921 dprintf("%s: sequence number mismatch %u<->%u\n", __func__
,
3922 be32_to_cpu(first
->active
->seq
),
3923 be32_to_cpu(second
->active
->seq
));
3926 if (first
->max_part
!= second
->max_part
||
3927 !be16_eq(first
->phys
->used_pdes
, second
->phys
->used_pdes
) ||
3928 !be16_eq(first
->virt
->populated_vdes
,
3929 second
->virt
->populated_vdes
)) {
3930 dprintf("%s: PD/VD number mismatch\n", __func__
);
3934 max_pds
= be16_to_cpu(first
->phys
->used_pdes
);
3935 for (dl2
= second
->dlist
; dl2
; dl2
= dl2
->next
) {
3936 for (pd
= 0; pd
< max_pds
; pd
++)
3937 if (be32_eq(first
->phys
->entries
[pd
].refnum
,
3940 if (pd
== max_pds
) {
3941 dprintf("%s: no match for disk %08x\n", __func__
,
3942 be32_to_cpu(dl2
->disk
.refnum
));
3947 max_vds
= be16_to_cpu(first
->active
->max_vd_entries
);
3948 for (vl2
= second
->conflist
; vl2
; vl2
= vl2
->next
) {
3949 if (!be32_eq(vl2
->conf
.magic
, DDF_VD_CONF_MAGIC
))
3951 for (vd
= 0; vd
< max_vds
; vd
++)
3952 if (!memcmp(first
->virt
->entries
[vd
].guid
,
3953 vl2
->conf
.guid
, DDF_GUID_LEN
))
3955 if (vd
== max_vds
) {
3956 dprintf("%s: no match for VD config\n", __func__
);
3960 /* FIXME should I look at anything else? */
3963 At this point we are fairly sure that the meta data matches.
3964 But the new disk may contain additional local data.
3965 Add it to the super block.
3967 for (vl2
= second
->conflist
; vl2
; vl2
= vl2
->next
) {
3968 for (vl1
= first
->conflist
; vl1
; vl1
= vl1
->next
)
3969 if (!memcmp(vl1
->conf
.guid
, vl2
->conf
.guid
,
3973 if (vl1
->other_bvds
!= NULL
&&
3974 vl1
->conf
.sec_elmnt_seq
!=
3975 vl2
->conf
.sec_elmnt_seq
) {
3976 dprintf("%s: adding BVD %u\n", __func__
,
3977 vl2
->conf
.sec_elmnt_seq
);
3978 add_other_bvd(vl1
, &vl2
->conf
,
3979 first
->conf_rec_len
*512);
3984 if (posix_memalign((void **)&vl1
, 512,
3985 (first
->conf_rec_len
*512 +
3986 offsetof(struct vcl
, conf
))) != 0) {
3987 pr_err("%s could not allocate vcl buf\n",
3992 vl1
->next
= first
->conflist
;
3993 vl1
->block_sizes
= NULL
;
3994 memcpy(&vl1
->conf
, &vl2
->conf
, first
->conf_rec_len
*512);
3995 if (alloc_other_bvds(first
, vl1
) != 0) {
3996 pr_err("%s could not allocate other bvds\n",
4001 for (vd
= 0; vd
< max_vds
; vd
++)
4002 if (!memcmp(first
->virt
->entries
[vd
].guid
,
4003 vl1
->conf
.guid
, DDF_GUID_LEN
))
4006 dprintf("%s: added config for VD %u\n", __func__
, vl1
->vcnum
);
4007 first
->conflist
= vl1
;
4010 for (dl2
= second
->dlist
; dl2
; dl2
= dl2
->next
) {
4011 for (dl1
= first
->dlist
; dl1
; dl1
= dl1
->next
)
4012 if (be32_eq(dl1
->disk
.refnum
, dl2
->disk
.refnum
))
4017 if (posix_memalign((void **)&dl1
, 512,
4018 sizeof(*dl1
) + (first
->max_part
) * sizeof(dl1
->vlist
[0]))
4020 pr_err("%s could not allocate disk info buffer\n",
4024 memcpy(dl1
, dl2
, sizeof(*dl1
));
4025 dl1
->mdupdate
= NULL
;
4026 dl1
->next
= first
->dlist
;
4028 for (pd
= 0; pd
< max_pds
; pd
++)
4029 if (be32_eq(first
->phys
->entries
[pd
].refnum
,
4034 if (posix_memalign((void **)&dl1
->spare
, 512,
4035 first
->conf_rec_len
*512) != 0) {
4036 pr_err("%s could not allocate spare info buf\n",
4040 memcpy(dl1
->spare
, dl2
->spare
, first
->conf_rec_len
*512);
4042 for (vd
= 0 ; vd
< first
->max_part
; vd
++) {
4043 if (!dl2
->vlist
[vd
]) {
4044 dl1
->vlist
[vd
] = NULL
;
4047 for (vl1
= first
->conflist
; vl1
; vl1
= vl1
->next
) {
4048 if (!memcmp(vl1
->conf
.guid
,
4049 dl2
->vlist
[vd
]->conf
.guid
,
4052 dl1
->vlist
[vd
] = vl1
;
4056 dprintf("%s: added disk %d: %08x\n", __func__
, dl1
->pdnum
,
4057 be32_to_cpu(dl1
->disk
.refnum
));
4065 * A new array 'a' has been started which claims to be instance 'inst'
4066 * within container 'c'.
4067 * We need to confirm that the array matches the metadata in 'c' so
4068 * that we don't corrupt any metadata.
4070 static int ddf_open_new(struct supertype
*c
, struct active_array
*a
, char *inst
)
4072 struct ddf_super
*ddf
= c
->sb
;
4076 static const char faulty
[] = "faulty";
4078 if (all_ff(ddf
->virt
->entries
[n
].guid
)) {
4079 pr_err("%s: subarray %d doesn't exist\n", __func__
, n
);
4082 dprintf("%s: new subarray %d, GUID: %s\n", __func__
, n
,
4083 guid_str(ddf
->virt
->entries
[n
].guid
));
4084 for (dev
= a
->info
.devs
; dev
; dev
= dev
->next
) {
4085 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4086 if (dl
->major
== dev
->disk
.major
&&
4087 dl
->minor
== dev
->disk
.minor
)
4090 pr_err("%s: device %d/%d of subarray %d not found in meta data\n",
4091 __func__
, dev
->disk
.major
, dev
->disk
.minor
, n
);
4094 if ((be16_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].state
) &
4095 (DDF_Online
|DDF_Missing
|DDF_Failed
)) != DDF_Online
) {
4096 pr_err("%s: new subarray %d contains broken device %d/%d (%02x)\n",
4097 __func__
, n
, dl
->major
, dl
->minor
,
4099 ddf
->phys
->entries
[dl
->pdnum
].state
));
4100 if (write(dev
->state_fd
, faulty
, sizeof(faulty
)-1) !=
4102 pr_err("Write to state_fd failed\n");
4103 dev
->curr_state
= DS_FAULTY
;
4106 a
->info
.container_member
= n
;
4111 * The array 'a' is to be marked clean in the metadata.
4112 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
4113 * clean up to the point (in sectors). If that cannot be recorded in the
4114 * metadata, then leave it as dirty.
4116 * For DDF, we need to clear the DDF_state_inconsistent bit in the
4117 * !global! virtual_disk.virtual_entry structure.
4119 static int ddf_set_array_state(struct active_array
*a
, int consistent
)
4121 struct ddf_super
*ddf
= a
->container
->sb
;
4122 int inst
= a
->info
.container_member
;
4123 int old
= ddf
->virt
->entries
[inst
].state
;
4124 if (consistent
== 2) {
4125 /* Should check if a recovery should be started FIXME */
4127 if (!is_resync_complete(&a
->info
))
4131 ddf
->virt
->entries
[inst
].state
&= ~DDF_state_inconsistent
;
4133 ddf
->virt
->entries
[inst
].state
|= DDF_state_inconsistent
;
4134 if (old
!= ddf
->virt
->entries
[inst
].state
)
4135 ddf_set_updates_pending(ddf
);
4137 old
= ddf
->virt
->entries
[inst
].init_state
;
4138 ddf
->virt
->entries
[inst
].init_state
&= ~DDF_initstate_mask
;
4139 if (is_resync_complete(&a
->info
))
4140 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_full
;
4141 else if (a
->info
.resync_start
== 0)
4142 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_not
;
4144 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_quick
;
4145 if (old
!= ddf
->virt
->entries
[inst
].init_state
)
4146 ddf_set_updates_pending(ddf
);
4148 dprintf("ddf mark %d/%s (%d) %s %llu\n", inst
,
4149 guid_str(ddf
->virt
->entries
[inst
].guid
), a
->curr_state
,
4150 consistent
?"clean":"dirty",
4151 a
->info
.resync_start
);
4155 static int get_bvd_state(const struct ddf_super
*ddf
,
4156 const struct vd_config
*vc
)
4158 unsigned int i
, n_bvd
, working
= 0;
4159 unsigned int n_prim
= be16_to_cpu(vc
->prim_elmnt_count
);
4161 for (i
= 0; i
< n_prim
; i
++) {
4162 if (!find_index_in_bvd(ddf
, vc
, i
, &n_bvd
))
4164 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
4167 st
= be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
4168 if ((st
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
4173 state
= DDF_state_degraded
;
4174 if (working
== n_prim
)
4175 state
= DDF_state_optimal
;
4181 state
= DDF_state_failed
;
4185 state
= DDF_state_failed
;
4186 else if (working
>= 2)
4187 state
= DDF_state_part_optimal
;
4191 if (working
< n_prim
- 1)
4192 state
= DDF_state_failed
;
4195 if (working
< n_prim
- 2)
4196 state
= DDF_state_failed
;
4197 else if (working
== n_prim
- 1)
4198 state
= DDF_state_part_optimal
;
4204 static int secondary_state(int state
, int other
, int seclevel
)
4206 if (state
== DDF_state_optimal
&& other
== DDF_state_optimal
)
4207 return DDF_state_optimal
;
4208 if (seclevel
== DDF_2MIRRORED
) {
4209 if (state
== DDF_state_optimal
|| other
== DDF_state_optimal
)
4210 return DDF_state_part_optimal
;
4211 if (state
== DDF_state_failed
&& other
== DDF_state_failed
)
4212 return DDF_state_failed
;
4213 return DDF_state_degraded
;
4215 if (state
== DDF_state_failed
|| other
== DDF_state_failed
)
4216 return DDF_state_failed
;
4217 if (state
== DDF_state_degraded
|| other
== DDF_state_degraded
)
4218 return DDF_state_degraded
;
4219 return DDF_state_part_optimal
;
4223 static int get_svd_state(const struct ddf_super
*ddf
, const struct vcl
*vcl
)
4225 int state
= get_bvd_state(ddf
, &vcl
->conf
);
4227 for (i
= 1; i
< vcl
->conf
.sec_elmnt_count
; i
++) {
4228 state
= secondary_state(
4230 get_bvd_state(ddf
, vcl
->other_bvds
[i
-1]),
4237 * The state of each disk is stored in the global phys_disk structure
4238 * in phys_disk.entries[n].state.
4239 * This makes various combinations awkward.
4240 * - When a device fails in any array, it must be failed in all arrays
4241 * that include a part of this device.
4242 * - When a component is rebuilding, we cannot include it officially in the
4243 * array unless this is the only array that uses the device.
4245 * So: when transitioning:
4246 * Online -> failed, just set failed flag. monitor will propagate
4247 * spare -> online, the device might need to be added to the array.
4248 * spare -> failed, just set failed. Don't worry if in array or not.
4250 static void ddf_set_disk(struct active_array
*a
, int n
, int state
)
4252 struct ddf_super
*ddf
= a
->container
->sb
;
4253 unsigned int inst
= a
->info
.container_member
, n_bvd
;
4255 struct vd_config
*vc
= find_vdcr(ddf
, inst
, (unsigned int)n
,
4261 dprintf("%s: %d to %x\n", __func__
, n
, state
);
4263 dprintf("ddf: cannot find instance %d!!\n", inst
);
4266 /* Find the matching slot in 'info'. */
4267 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
4268 if (mdi
->disk
.raid_disk
== n
)
4271 pr_err("%s: cannot find raid disk %d\n",
4276 /* and find the 'dl' entry corresponding to that. */
4277 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4278 if (mdi
->state_fd
>= 0 &&
4279 mdi
->disk
.major
== dl
->major
&&
4280 mdi
->disk
.minor
== dl
->minor
)
4283 pr_err("%s: cannot find raid disk %d (%d/%d)\n",
4285 mdi
->disk
.major
, mdi
->disk
.minor
);
4289 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
4290 if (pd
< 0 || pd
!= dl
->pdnum
) {
4291 /* disk doesn't currently exist or has changed.
4292 * If it is now in_sync, insert it. */
4293 dprintf("%s: phys disk not found for %d: %d/%d ref %08x\n",
4294 __func__
, dl
->pdnum
, dl
->major
, dl
->minor
,
4295 be32_to_cpu(dl
->disk
.refnum
));
4296 dprintf("%s: array %u disk %u ref %08x pd %d\n",
4297 __func__
, inst
, n_bvd
,
4298 be32_to_cpu(vc
->phys_refnum
[n_bvd
]), pd
);
4299 if ((state
& DS_INSYNC
) && ! (state
& DS_FAULTY
)) {
4300 pd
= dl
->pdnum
; /* FIXME: is this really correct ? */
4301 vc
->phys_refnum
[n_bvd
] = dl
->disk
.refnum
;
4302 LBA_OFFSET(ddf
, vc
)[n_bvd
] =
4303 cpu_to_be64(mdi
->data_offset
);
4304 be16_clear(ddf
->phys
->entries
[pd
].type
,
4305 cpu_to_be16(DDF_Global_Spare
));
4306 be16_set(ddf
->phys
->entries
[pd
].type
,
4307 cpu_to_be16(DDF_Active_in_VD
));
4308 ddf_set_updates_pending(ddf
);
4311 be16 old
= ddf
->phys
->entries
[pd
].state
;
4312 if (state
& DS_FAULTY
)
4313 be16_set(ddf
->phys
->entries
[pd
].state
,
4314 cpu_to_be16(DDF_Failed
));
4315 if (state
& DS_INSYNC
) {
4316 be16_set(ddf
->phys
->entries
[pd
].state
,
4317 cpu_to_be16(DDF_Online
));
4318 be16_clear(ddf
->phys
->entries
[pd
].state
,
4319 cpu_to_be16(DDF_Rebuilding
));
4321 if (!be16_eq(old
, ddf
->phys
->entries
[pd
].state
))
4322 ddf_set_updates_pending(ddf
);
4325 dprintf("ddf: set_disk %d (%08x) to %x->%02x\n", n
,
4326 be32_to_cpu(dl
->disk
.refnum
), state
,
4327 be16_to_cpu(ddf
->phys
->entries
[pd
].state
));
4329 /* Now we need to check the state of the array and update
4330 * virtual_disk.entries[n].state.
4331 * It needs to be one of "optimal", "degraded", "failed".
4332 * I don't understand 'deleted' or 'missing'.
4334 state
= get_svd_state(ddf
, vcl
);
4336 if (ddf
->virt
->entries
[inst
].state
!=
4337 ((ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4340 ddf
->virt
->entries
[inst
].state
=
4341 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4343 ddf_set_updates_pending(ddf
);
4348 static void ddf_sync_metadata(struct supertype
*st
)
4352 * Write all data to all devices.
4353 * Later, we might be able to track whether only local changes
4354 * have been made, or whether any global data has been changed,
4355 * but ddf is sufficiently weird that it probably always
4356 * changes global data ....
4358 struct ddf_super
*ddf
= st
->sb
;
4359 if (!ddf
->updates_pending
)
4361 ddf
->updates_pending
= 0;
4362 __write_init_super_ddf(st
);
4363 dprintf("ddf: sync_metadata\n");
4366 static int del_from_conflist(struct vcl
**list
, const char *guid
)
4370 for (p
= list
; p
&& *p
; p
= &((*p
)->next
))
4371 if (!memcmp((*p
)->conf
.guid
, guid
, DDF_GUID_LEN
)) {
4378 static int _kill_subarray_ddf(struct ddf_super
*ddf
, const char *guid
)
4381 unsigned int vdnum
, i
;
4382 vdnum
= find_vde_by_guid(ddf
, guid
);
4383 if (vdnum
== DDF_NOTFOUND
) {
4384 pr_err("%s: could not find VD %s\n", __func__
,
4388 if (del_from_conflist(&ddf
->conflist
, guid
) == 0) {
4389 pr_err("%s: could not find conf %s\n", __func__
,
4393 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4394 for (i
= 0; i
< ddf
->max_part
; i
++)
4395 if (dl
->vlist
[i
] != NULL
&&
4396 !memcmp(dl
->vlist
[i
]->conf
.guid
, guid
,
4398 dl
->vlist
[i
] = NULL
;
4399 memset(ddf
->virt
->entries
[vdnum
].guid
, 0xff, DDF_GUID_LEN
);
4400 dprintf("%s: deleted %s\n", __func__
, guid_str(guid
));
4404 static int kill_subarray_ddf(struct supertype
*st
)
4406 struct ddf_super
*ddf
= st
->sb
;
4408 * currentconf is set in container_content_ddf,
4409 * called with subarray arg
4411 struct vcl
*victim
= ddf
->currentconf
;
4412 struct vd_config
*conf
;
4413 ddf
->currentconf
= NULL
;
4416 pr_err("%s: nothing to kill\n", __func__
);
4419 conf
= &victim
->conf
;
4420 vdnum
= find_vde_by_guid(ddf
, conf
->guid
);
4421 if (vdnum
== DDF_NOTFOUND
) {
4422 pr_err("%s: could not find VD %s\n", __func__
,
4423 guid_str(conf
->guid
));
4426 if (st
->update_tail
) {
4427 struct virtual_disk
*vd
;
4428 int len
= sizeof(struct virtual_disk
)
4429 + sizeof(struct virtual_entry
);
4432 pr_err("%s: failed to allocate %d bytes\n", __func__
,
4436 memset(vd
, 0 , len
);
4437 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
4438 vd
->populated_vdes
= cpu_to_be16(0);
4439 memcpy(vd
->entries
[0].guid
, conf
->guid
, DDF_GUID_LEN
);
4440 /* we use DDF_state_deleted as marker */
4441 vd
->entries
[0].state
= DDF_state_deleted
;
4442 append_metadata_update(st
, vd
, len
);
4444 _kill_subarray_ddf(ddf
, conf
->guid
);
4445 ddf_set_updates_pending(ddf
);
4446 ddf_sync_metadata(st
);
4451 static void copy_matching_bvd(struct ddf_super
*ddf
,
4452 struct vd_config
*conf
,
4453 const struct metadata_update
*update
)
4456 be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
4457 unsigned int len
= ddf
->conf_rec_len
* 512;
4459 struct vd_config
*vc
;
4460 for (p
= update
->buf
; p
< update
->buf
+ update
->len
; p
+= len
) {
4461 vc
= (struct vd_config
*) p
;
4462 if (vc
->sec_elmnt_seq
== conf
->sec_elmnt_seq
) {
4463 memcpy(conf
->phys_refnum
, vc
->phys_refnum
,
4464 mppe
* (sizeof(__u32
) + sizeof(__u64
)));
4468 pr_err("%s: no match for BVD %d of %s in update\n", __func__
,
4469 conf
->sec_elmnt_seq
, guid_str(conf
->guid
));
4472 static void ddf_process_update(struct supertype
*st
,
4473 struct metadata_update
*update
)
4475 /* Apply this update to the metadata.
4476 * The first 4 bytes are a DDF_*_MAGIC which guides
4478 * Possible update are:
4479 * DDF_PHYS_RECORDS_MAGIC
4480 * Add a new physical device or remove an old one.
4481 * Changes to this record only happen implicitly.
4482 * used_pdes is the device number.
4483 * DDF_VIRT_RECORDS_MAGIC
4484 * Add a new VD. Possibly also change the 'access' bits.
4485 * populated_vdes is the entry number.
4487 * New or updated VD. the VIRT_RECORD must already
4488 * exist. For an update, phys_refnum and lba_offset
4489 * (at least) are updated, and the VD_CONF must
4490 * be written to precisely those devices listed with
4492 * DDF_SPARE_ASSIGN_MAGIC
4493 * replacement Spare Assignment Record... but for which device?
4496 * - to create a new array, we send a VIRT_RECORD and
4497 * a VD_CONF. Then assemble and start the array.
4498 * - to activate a spare we send a VD_CONF to add the phys_refnum
4499 * and offset. This will also mark the spare as active with
4500 * a spare-assignment record.
4502 struct ddf_super
*ddf
= st
->sb
;
4503 be32
*magic
= (be32
*)update
->buf
;
4504 struct phys_disk
*pd
;
4505 struct virtual_disk
*vd
;
4506 struct vd_config
*vc
;
4510 unsigned int pdnum
, pd2
, len
;
4512 dprintf("Process update %x\n", be32_to_cpu(*magic
));
4514 if (be32_eq(*magic
, DDF_PHYS_RECORDS_MAGIC
)) {
4516 if (update
->len
!= (sizeof(struct phys_disk
) +
4517 sizeof(struct phys_disk_entry
)))
4519 pd
= (struct phys_disk
*)update
->buf
;
4521 ent
= be16_to_cpu(pd
->used_pdes
);
4522 if (ent
>= be16_to_cpu(ddf
->phys
->max_pdes
))
4524 if (be16_and(pd
->entries
[0].state
, cpu_to_be16(DDF_Missing
))) {
4526 /* removing this disk. */
4527 be16_set(ddf
->phys
->entries
[ent
].state
,
4528 cpu_to_be16(DDF_Missing
));
4529 for (dlp
= &ddf
->dlist
; *dlp
; dlp
= &(*dlp
)->next
) {
4530 struct dl
*dl
= *dlp
;
4531 if (dl
->pdnum
== (signed)ent
) {
4534 /* FIXME this doesn't free
4541 ddf_set_updates_pending(ddf
);
4544 if (!all_ff(ddf
->phys
->entries
[ent
].guid
))
4546 ddf
->phys
->entries
[ent
] = pd
->entries
[0];
4547 ddf
->phys
->used_pdes
= cpu_to_be16
4548 (1 + be16_to_cpu(ddf
->phys
->used_pdes
));
4549 ddf_set_updates_pending(ddf
);
4550 if (ddf
->add_list
) {
4551 struct active_array
*a
;
4552 struct dl
*al
= ddf
->add_list
;
4553 ddf
->add_list
= al
->next
;
4555 al
->next
= ddf
->dlist
;
4558 /* As a device has been added, we should check
4559 * for any degraded devices that might make
4560 * use of this spare */
4561 for (a
= st
->arrays
; a
; a
=a
->next
)
4562 a
->check_degraded
= 1;
4564 } else if (be32_eq(*magic
, DDF_VIRT_RECORDS_MAGIC
)) {
4566 if (update
->len
!= (sizeof(struct virtual_disk
) +
4567 sizeof(struct virtual_entry
)))
4569 vd
= (struct virtual_disk
*)update
->buf
;
4571 if (vd
->entries
[0].state
== DDF_state_deleted
) {
4572 if (_kill_subarray_ddf(ddf
, vd
->entries
[0].guid
))
4576 ent
= find_vde_by_guid(ddf
, vd
->entries
[0].guid
);
4577 if (ent
!= DDF_NOTFOUND
) {
4578 dprintf("%s: VD %s exists already in slot %d\n",
4579 __func__
, guid_str(vd
->entries
[0].guid
),
4583 ent
= find_unused_vde(ddf
);
4584 if (ent
== DDF_NOTFOUND
)
4586 ddf
->virt
->entries
[ent
] = vd
->entries
[0];
4587 ddf
->virt
->populated_vdes
=
4590 ddf
->virt
->populated_vdes
));
4591 dprintf("%s: added VD %s in slot %d(s=%02x i=%02x)\n",
4592 __func__
, guid_str(vd
->entries
[0].guid
), ent
,
4593 ddf
->virt
->entries
[ent
].state
,
4594 ddf
->virt
->entries
[ent
].init_state
);
4596 ddf_set_updates_pending(ddf
);
4599 else if (be32_eq(*magic
, DDF_VD_CONF_MAGIC
)) {
4600 vc
= (struct vd_config
*)update
->buf
;
4601 len
= ddf
->conf_rec_len
* 512;
4602 if ((unsigned int)update
->len
!= len
* vc
->sec_elmnt_count
) {
4603 pr_err("%s: %s: insufficient data (%d) for %u BVDs\n",
4604 __func__
, guid_str(vc
->guid
), update
->len
,
4605 vc
->sec_elmnt_count
);
4608 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
4609 if (memcmp(vcl
->conf
.guid
, vc
->guid
, DDF_GUID_LEN
) == 0)
4611 dprintf("%s: conf update for %s (%s)\n", __func__
,
4612 guid_str(vc
->guid
), (vcl
? "old" : "new"));
4614 /* An update, just copy the phys_refnum and lba_offset
4619 copy_matching_bvd(ddf
, &vcl
->conf
, update
);
4620 for (k
= 0; k
< be16_to_cpu(vc
->prim_elmnt_count
); k
++)
4621 dprintf("BVD %u has %08x at %llu\n", 0,
4622 be32_to_cpu(vcl
->conf
.phys_refnum
[k
]),
4623 be64_to_cpu(LBA_OFFSET(ddf
,
4625 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++) {
4626 copy_matching_bvd(ddf
, vcl
->other_bvds
[i
-1],
4628 for (k
= 0; k
< be16_to_cpu(
4629 vc
->prim_elmnt_count
); k
++)
4630 dprintf("BVD %u has %08x at %llu\n", i
,
4632 (vcl
->other_bvds
[i
-1]->
4637 vcl
->other_bvds
[i
-1])[k
]));
4644 vcl
= update
->space
;
4645 update
->space
= NULL
;
4646 vcl
->next
= ddf
->conflist
;
4647 memcpy(&vcl
->conf
, vc
, len
);
4648 ent
= find_vde_by_guid(ddf
, vc
->guid
);
4649 if (ent
== DDF_NOTFOUND
)
4652 ddf
->conflist
= vcl
;
4653 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++)
4654 memcpy(vcl
->other_bvds
[i
-1],
4655 update
->buf
+ len
* i
, len
);
4657 /* Set DDF_Transition on all Failed devices - to help
4658 * us detect those that are no longer in use
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_set(ddf
->phys
->entries
[pdnum
].state
,
4665 cpu_to_be16(DDF_Transition
));
4666 /* Now make sure vlist is correct for each dl. */
4667 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
4668 unsigned int vn
= 0;
4669 int in_degraded
= 0;
4670 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
) {
4671 unsigned int dn
, ibvd
;
4672 const struct vd_config
*conf
;
4674 dn
= get_pd_index_from_refnum(vcl
,
4678 if (dn
== DDF_NOTFOUND
)
4680 dprintf("dev %d/%08x has %s (sec=%u) at %d\n",
4682 be32_to_cpu(dl
->disk
.refnum
),
4683 guid_str(conf
->guid
),
4684 conf
->sec_elmnt_seq
, vn
);
4685 /* Clear the Transition flag */
4687 (ddf
->phys
->entries
[dl
->pdnum
].state
,
4688 cpu_to_be16(DDF_Failed
)))
4689 be16_clear(ddf
->phys
4690 ->entries
[dl
->pdnum
].state
,
4691 cpu_to_be16(DDF_Transition
));
4692 dl
->vlist
[vn
++] = vcl
;
4693 vstate
= ddf
->virt
->entries
[vcl
->vcnum
].state
4695 if (vstate
== DDF_state_degraded
||
4696 vstate
== DDF_state_part_optimal
)
4699 while (vn
< ddf
->max_part
)
4700 dl
->vlist
[vn
++] = NULL
;
4702 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4703 cpu_to_be16(DDF_Global_Spare
));
4704 if (!be16_and(ddf
->phys
4705 ->entries
[dl
->pdnum
].type
,
4706 cpu_to_be16(DDF_Active_in_VD
))) {
4708 ->entries
[dl
->pdnum
].type
,
4709 cpu_to_be16(DDF_Active_in_VD
));
4712 ->entries
[dl
->pdnum
]
4719 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4720 cpu_to_be16(DDF_Global_Spare
));
4721 be16_set(ddf
->phys
->entries
[dl
->pdnum
].type
,
4722 cpu_to_be16(DDF_Spare
));
4724 if (!dl
->vlist
[0] && !dl
->spare
) {
4725 be16_set(ddf
->phys
->entries
[dl
->pdnum
].type
,
4726 cpu_to_be16(DDF_Global_Spare
));
4727 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4728 cpu_to_be16(DDF_Spare
));
4729 be16_clear(ddf
->phys
->entries
[dl
->pdnum
].type
,
4730 cpu_to_be16(DDF_Active_in_VD
));
4734 /* Now remove any 'Failed' devices that are not part
4735 * of any VD. They will have the Transition flag set.
4736 * Once done, we need to update all dl->pdnum numbers.
4739 for (pdnum
= 0; pdnum
< be16_to_cpu(ddf
->phys
->used_pdes
);
4741 if (be16_and(ddf
->phys
->entries
[pdnum
].state
,
4742 cpu_to_be16(DDF_Failed
))
4743 && be16_and(ddf
->phys
->entries
[pdnum
].state
,
4744 cpu_to_be16(DDF_Transition
))) {
4745 /* skip this one unless in dlist*/
4746 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4747 if (dl
->pdnum
== (int)pdnum
)
4755 ddf
->phys
->entries
[pd2
] =
4756 ddf
->phys
->entries
[pdnum
];
4757 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4758 if (dl
->pdnum
== (int)pdnum
)
4763 ddf
->phys
->used_pdes
= cpu_to_be16(pd2
);
4764 while (pd2
< pdnum
) {
4765 memset(ddf
->phys
->entries
[pd2
].guid
, 0xff,
4770 ddf_set_updates_pending(ddf
);
4772 /* case DDF_SPARE_ASSIGN_MAGIC */
4775 static void ddf_prepare_update(struct supertype
*st
,
4776 struct metadata_update
*update
)
4778 /* This update arrived at managemon.
4779 * We are about to pass it to monitor.
4780 * If a malloc is needed, do it here.
4782 struct ddf_super
*ddf
= st
->sb
;
4783 be32
*magic
= (be32
*)update
->buf
;
4784 if (be32_eq(*magic
, DDF_VD_CONF_MAGIC
)) {
4786 struct vd_config
*conf
= (struct vd_config
*) update
->buf
;
4787 if (posix_memalign(&update
->space
, 512,
4788 offsetof(struct vcl
, conf
)
4789 + ddf
->conf_rec_len
* 512) != 0) {
4790 update
->space
= NULL
;
4793 vcl
= update
->space
;
4794 vcl
->conf
.sec_elmnt_count
= conf
->sec_elmnt_count
;
4795 if (alloc_other_bvds(ddf
, vcl
) != 0) {
4796 free(update
->space
);
4797 update
->space
= NULL
;
4803 * Check degraded state of a RAID10.
4804 * returns 2 for good, 1 for degraded, 0 for failed, and -1 for error
4806 static int raid10_degraded(struct mdinfo
*info
)
4814 n_prim
= info
->array
.layout
& ~0x100;
4815 n_bvds
= info
->array
.raid_disks
/ n_prim
;
4816 found
= xmalloc(n_bvds
);
4819 memset(found
, 0, n_bvds
);
4820 for (d
= info
->devs
; d
; d
= d
->next
) {
4821 i
= d
->disk
.raid_disk
/ n_prim
;
4823 pr_err("%s: BUG: invalid raid disk\n", __func__
);
4826 if (d
->state_fd
> 0)
4830 for (i
= 0; i
< n_bvds
; i
++)
4832 dprintf("%s: BVD %d/%d failed\n", __func__
, i
, n_bvds
);
4835 } else if (found
[i
] < n_prim
) {
4836 dprintf("%s: BVD %d/%d degraded\n", __func__
, i
,
4846 * Check if the array 'a' is degraded but not failed.
4847 * If it is, find as many spares as are available and needed and
4848 * arrange for their inclusion.
4849 * We only choose devices which are not already in the array,
4850 * and prefer those with a spare-assignment to this array.
4851 * otherwise we choose global spares - assuming always that
4852 * there is enough room.
4853 * For each spare that we assign, we return an 'mdinfo' which
4854 * describes the position for the device in the array.
4855 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
4856 * the new phys_refnum and lba_offset values.
4858 * Only worry about BVDs at the moment.
4860 static struct mdinfo
*ddf_activate_spare(struct active_array
*a
,
4861 struct metadata_update
**updates
)
4865 struct ddf_super
*ddf
= a
->container
->sb
;
4867 struct mdinfo
*rv
= NULL
;
4869 struct metadata_update
*mu
;
4874 struct vd_config
*vc
;
4877 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
4878 if ((d
->curr_state
& DS_FAULTY
) &&
4880 /* wait for Removal to happen */
4882 if (d
->state_fd
>= 0)
4886 dprintf("%s: working=%d (%d) level=%d\n", __func__
, working
,
4887 a
->info
.array
.raid_disks
,
4888 a
->info
.array
.level
);
4889 if (working
== a
->info
.array
.raid_disks
)
4890 return NULL
; /* array not degraded */
4891 switch (a
->info
.array
.level
) {
4894 return NULL
; /* failed */
4898 if (working
< a
->info
.array
.raid_disks
- 1)
4899 return NULL
; /* failed */
4902 if (working
< a
->info
.array
.raid_disks
- 2)
4903 return NULL
; /* failed */
4906 if (raid10_degraded(&a
->info
) < 1)
4909 default: /* concat or stripe */
4910 return NULL
; /* failed */
4913 /* For each slot, if it is not working, find a spare */
4915 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
4916 for (d
= a
->info
.devs
; d
; d
= d
->next
)
4917 if (d
->disk
.raid_disk
== i
)
4919 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
4920 if (d
&& (d
->state_fd
>= 0))
4923 /* OK, this device needs recovery. Find a spare */
4925 for ( ; dl
; dl
= dl
->next
) {
4926 unsigned long long esize
;
4927 unsigned long long pos
;
4930 int is_dedicated
= 0;
4933 be16 state
= ddf
->phys
->entries
[dl
->pdnum
].state
;
4935 cpu_to_be16(DDF_Failed
|DDF_Missing
)) ||
4937 cpu_to_be16(DDF_Online
)))
4940 /* If in this array, skip */
4941 for (d2
= a
->info
.devs
; d2
; d2
= d2
->next
)
4942 if (d2
->state_fd
>= 0 &&
4943 d2
->disk
.major
== dl
->major
&&
4944 d2
->disk
.minor
== dl
->minor
) {
4945 dprintf("%x:%x (%08x) already in array\n",
4946 dl
->major
, dl
->minor
,
4947 be32_to_cpu(dl
->disk
.refnum
));
4952 if (be16_and(ddf
->phys
->entries
[dl
->pdnum
].type
,
4953 cpu_to_be16(DDF_Spare
))) {
4954 /* Check spare assign record */
4956 if (dl
->spare
->type
& DDF_spare_dedicated
) {
4957 /* check spare_ents for guid */
4963 if (memcmp(dl
->spare
->spare_ents
[j
].guid
,
4964 ddf
->virt
->entries
[a
->info
.container_member
].guid
,
4971 } else if (be16_and(ddf
->phys
->entries
[dl
->pdnum
].type
,
4972 cpu_to_be16(DDF_Global_Spare
))) {
4974 } else if (!be16_and(ddf
->phys
4975 ->entries
[dl
->pdnum
].state
,
4976 cpu_to_be16(DDF_Failed
))) {
4977 /* we can possibly use some of this */
4980 if ( ! (is_dedicated
||
4981 (is_global
&& global_ok
))) {
4982 dprintf("%x:%x not suitable: %d %d\n", dl
->major
, dl
->minor
,
4983 is_dedicated
, is_global
);
4987 /* We are allowed to use this device - is there space?
4988 * We need a->info.component_size sectors */
4989 ex
= get_extents(ddf
, dl
);
4991 dprintf("cannot get extents\n");
4998 esize
= ex
[j
].start
- pos
;
4999 if (esize
>= a
->info
.component_size
)
5001 pos
= ex
[j
].start
+ ex
[j
].size
;
5003 } while (ex
[j
-1].size
);
5006 if (esize
< a
->info
.component_size
) {
5007 dprintf("%x:%x has no room: %llu %llu\n",
5008 dl
->major
, dl
->minor
,
5009 esize
, a
->info
.component_size
);
5014 /* Cool, we have a device with some space at pos */
5015 di
= xcalloc(1, sizeof(*di
));
5016 di
->disk
.number
= i
;
5017 di
->disk
.raid_disk
= i
;
5018 di
->disk
.major
= dl
->major
;
5019 di
->disk
.minor
= dl
->minor
;
5021 di
->recovery_start
= 0;
5022 di
->data_offset
= pos
;
5023 di
->component_size
= a
->info
.component_size
;
5024 di
->container_member
= dl
->pdnum
;
5027 dprintf("%x:%x (%08x) to be %d at %llu\n",
5028 dl
->major
, dl
->minor
,
5029 be32_to_cpu(dl
->disk
.refnum
), i
, pos
);
5033 if (!dl
&& ! global_ok
) {
5034 /* not enough dedicated spares, try global */
5042 /* No spares found */
5044 /* Now 'rv' has a list of devices to return.
5045 * Create a metadata_update record to update the
5046 * phys_refnum and lba_offset values
5048 vc
= find_vdcr(ddf
, a
->info
.container_member
, rv
->disk
.raid_disk
,
5053 mu
= xmalloc(sizeof(*mu
));
5054 if (posix_memalign(&mu
->space
, 512, sizeof(struct vcl
)) != 0) {
5059 mu
->len
= ddf
->conf_rec_len
* 512 * vcl
->conf
.sec_elmnt_count
;
5060 mu
->buf
= xmalloc(mu
->len
);
5062 mu
->space_list
= NULL
;
5063 mu
->next
= *updates
;
5064 memcpy(mu
->buf
, &vcl
->conf
, ddf
->conf_rec_len
* 512);
5065 for (j
= 1; j
< vcl
->conf
.sec_elmnt_count
; j
++)
5066 memcpy(mu
->buf
+ j
* ddf
->conf_rec_len
* 512,
5067 vcl
->other_bvds
[j
-1], ddf
->conf_rec_len
* 512);
5069 vc
= (struct vd_config
*)mu
->buf
;
5070 for (di
= rv
; di
; di
= di
->next
) {
5071 unsigned int i_sec
, i_prim
;
5072 i_sec
= di
->disk
.raid_disk
5073 / be16_to_cpu(vcl
->conf
.prim_elmnt_count
);
5074 i_prim
= di
->disk
.raid_disk
5075 % be16_to_cpu(vcl
->conf
.prim_elmnt_count
);
5076 vc
= (struct vd_config
*)(mu
->buf
5077 + i_sec
* ddf
->conf_rec_len
* 512);
5078 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
5079 if (dl
->major
== di
->disk
.major
5080 && dl
->minor
== di
->disk
.minor
)
5083 pr_err("%s: BUG: can't find disk %d (%d/%d)\n",
5084 __func__
, di
->disk
.raid_disk
,
5085 di
->disk
.major
, di
->disk
.minor
);
5088 vc
->phys_refnum
[i_prim
] = ddf
->phys
->entries
[dl
->pdnum
].refnum
;
5089 LBA_OFFSET(ddf
, vc
)[i_prim
] = cpu_to_be64(di
->data_offset
);
5090 dprintf("BVD %u gets %u: %08x at %llu\n", i_sec
, i_prim
,
5091 be32_to_cpu(vc
->phys_refnum
[i_prim
]),
5092 be64_to_cpu(LBA_OFFSET(ddf
, vc
)[i_prim
]));
5097 #endif /* MDASSEMBLE */
5099 static int ddf_level_to_layout(int level
)
5106 return ALGORITHM_LEFT_SYMMETRIC
;
5108 return ALGORITHM_ROTATING_N_CONTINUE
;
5116 static void default_geometry_ddf(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
5118 if (level
&& *level
== UnSet
)
5119 *level
= LEVEL_CONTAINER
;
5121 if (level
&& layout
&& *layout
== UnSet
)
5122 *layout
= ddf_level_to_layout(*level
);
5125 struct superswitch super_ddf
= {
5127 .examine_super
= examine_super_ddf
,
5128 .brief_examine_super
= brief_examine_super_ddf
,
5129 .brief_examine_subarrays
= brief_examine_subarrays_ddf
,
5130 .export_examine_super
= export_examine_super_ddf
,
5131 .detail_super
= detail_super_ddf
,
5132 .brief_detail_super
= brief_detail_super_ddf
,
5133 .validate_geometry
= validate_geometry_ddf
,
5134 .write_init_super
= write_init_super_ddf
,
5135 .add_to_super
= add_to_super_ddf
,
5136 .remove_from_super
= remove_from_super_ddf
,
5137 .load_container
= load_container_ddf
,
5138 .copy_metadata
= copy_metadata_ddf
,
5139 .kill_subarray
= kill_subarray_ddf
,
5141 .match_home
= match_home_ddf
,
5142 .uuid_from_super
= uuid_from_super_ddf
,
5143 .getinfo_super
= getinfo_super_ddf
,
5144 .update_super
= update_super_ddf
,
5146 .avail_size
= avail_size_ddf
,
5148 .compare_super
= compare_super_ddf
,
5150 .load_super
= load_super_ddf
,
5151 .init_super
= init_super_ddf
,
5152 .store_super
= store_super_ddf
,
5153 .free_super
= free_super_ddf
,
5154 .match_metadata_desc
= match_metadata_desc_ddf
,
5155 .container_content
= container_content_ddf
,
5156 .default_geometry
= default_geometry_ddf
,
5162 .open_new
= ddf_open_new
,
5163 .set_array_state
= ddf_set_array_state
,
5164 .set_disk
= ddf_set_disk
,
5165 .sync_metadata
= ddf_sync_metadata
,
5166 .process_update
= ddf_process_update
,
5167 .prepare_update
= ddf_prepare_update
,
5168 .activate_spare
= ddf_activate_spare
,