2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2006-2009 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neil@brown.name>
24 * Specifications for DDF takes from Common RAID DDF Specification Revision 1.2
25 * (July 28 2006). Reused by permission of SNIA.
28 #define HAVE_STDINT_H 1
34 /* a non-official T10 name for creation GUIDs */
35 static char T10
[] = "Linux-MD";
37 /* DDF timestamps are 1980 based, so we need to add
38 * second-in-decade-of-seventies to convert to linux timestamps.
39 * 10 years with 2 leap years.
41 #define DECADE (3600*24*(365*10+2))
44 const unsigned char *buf
,
47 #define DDF_NOTFOUND (~0U)
48 #define DDF_CONTAINER (DDF_NOTFOUND-1)
50 /* The DDF metadata handling.
51 * DDF metadata lives at the end of the device.
52 * The last 512 byte block provides an 'anchor' which is used to locate
53 * the rest of the metadata which usually lives immediately behind the anchor.
56 * - all multibyte numeric fields are bigendian.
57 * - all strings are space padded.
61 /* Primary Raid Level (PRL) */
62 #define DDF_RAID0 0x00
63 #define DDF_RAID1 0x01
64 #define DDF_RAID3 0x03
65 #define DDF_RAID4 0x04
66 #define DDF_RAID5 0x05
67 #define DDF_RAID1E 0x11
69 #define DDF_CONCAT 0x1f
70 #define DDF_RAID5E 0x15
71 #define DDF_RAID5EE 0x25
72 #define DDF_RAID6 0x06
74 /* Raid Level Qualifier (RLQ) */
75 #define DDF_RAID0_SIMPLE 0x00
76 #define DDF_RAID1_SIMPLE 0x00 /* just 2 devices in this plex */
77 #define DDF_RAID1_MULTI 0x01 /* exactly 3 devices in this plex */
78 #define DDF_RAID3_0 0x00 /* parity in first extent */
79 #define DDF_RAID3_N 0x01 /* parity in last extent */
80 #define DDF_RAID4_0 0x00 /* parity in first extent */
81 #define DDF_RAID4_N 0x01 /* parity in last extent */
82 /* these apply to raid5e and raid5ee as well */
83 #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */
84 #define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */
85 #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */
86 #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */
88 #define DDF_RAID1E_ADJACENT 0x00 /* raid10 nearcopies==2 */
89 #define DDF_RAID1E_OFFSET 0x01 /* raid10 offsetcopies==2 */
91 /* Secondary RAID Level (SRL) */
92 #define DDF_2STRIPED 0x00 /* This is weirder than RAID0 !! */
93 #define DDF_2MIRRORED 0x01
94 #define DDF_2CONCAT 0x02
95 #define DDF_2SPANNED 0x03 /* This is also weird - be careful */
98 #define DDF_HEADER_MAGIC __cpu_to_be32(0xDE11DE11)
99 #define DDF_CONTROLLER_MAGIC __cpu_to_be32(0xAD111111)
100 #define DDF_PHYS_RECORDS_MAGIC __cpu_to_be32(0x22222222)
101 #define DDF_PHYS_DATA_MAGIC __cpu_to_be32(0x33333333)
102 #define DDF_VIRT_RECORDS_MAGIC __cpu_to_be32(0xDDDDDDDD)
103 #define DDF_VD_CONF_MAGIC __cpu_to_be32(0xEEEEEEEE)
104 #define DDF_SPARE_ASSIGN_MAGIC __cpu_to_be32(0x55555555)
105 #define DDF_VU_CONF_MAGIC __cpu_to_be32(0x88888888)
106 #define DDF_VENDOR_LOG_MAGIC __cpu_to_be32(0x01dBEEF0)
107 #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C)
109 #define DDF_GUID_LEN 24
110 #define DDF_REVISION_0 "01.00.00"
111 #define DDF_REVISION_2 "01.02.00"
114 __u32 magic
; /* DDF_HEADER_MAGIC */
116 char guid
[DDF_GUID_LEN
];
117 char revision
[8]; /* 01.02.00 */
118 __u32 seq
; /* starts at '1' */
123 __u8 pad0
; /* 0xff */
124 __u8 pad1
[12]; /* 12 * 0xff */
125 /* 64 bytes so far */
126 __u8 header_ext
[32]; /* reserved: fill with 0xff */
130 __u8 pad2
[3]; /* 0xff */
131 __u32 workspace_len
; /* sectors for vendor space -
132 * at least 32768(sectors) */
134 __u16 max_pd_entries
; /* one of 15, 63, 255, 1023, 4095 */
135 __u16 max_vd_entries
; /* 2^(4,6,8,10,12)-1 : i.e. as above */
136 __u16 max_partitions
; /* i.e. max num of configuration
137 record entries per disk */
138 __u16 config_record_len
; /* 1 +ROUNDUP(max_primary_element_entries
140 __u16 max_primary_element_entries
; /* 16, 64, 256, 1024, or 4096 */
141 __u8 pad3
[54]; /* 0xff */
142 /* 192 bytes so far */
143 __u32 controller_section_offset
;
144 __u32 controller_section_length
;
145 __u32 phys_section_offset
;
146 __u32 phys_section_length
;
147 __u32 virt_section_offset
;
148 __u32 virt_section_length
;
149 __u32 config_section_offset
;
150 __u32 config_section_length
;
151 __u32 data_section_offset
;
152 __u32 data_section_length
;
153 __u32 bbm_section_offset
;
154 __u32 bbm_section_length
;
155 __u32 diag_space_offset
;
156 __u32 diag_space_length
;
159 /* 256 bytes so far */
160 __u8 pad4
[256]; /* 0xff */
164 #define DDF_HEADER_ANCHOR 0x00
165 #define DDF_HEADER_PRIMARY 0x01
166 #define DDF_HEADER_SECONDARY 0x02
168 /* The content of the 'controller section' - global scope */
169 struct ddf_controller_data
{
170 __u32 magic
; /* DDF_CONTROLLER_MAGIC */
172 char guid
[DDF_GUID_LEN
];
173 struct controller_type
{
180 __u8 pad
[8]; /* 0xff */
181 __u8 vendor_data
[448];
184 /* The content of phys_section - global scope */
186 __u32 magic
; /* DDF_PHYS_RECORDS_MAGIC */
191 struct phys_disk_entry
{
192 char guid
[DDF_GUID_LEN
];
196 __u64 config_size
; /* DDF structures must be after here */
197 char path
[18]; /* another horrible structure really */
202 /* phys_disk_entry.type is a bitmap - bigendian remember */
203 #define DDF_Forced_PD_GUID 1
204 #define DDF_Active_in_VD 2
205 #define DDF_Global_Spare 4 /* VD_CONF records are ignored */
206 #define DDF_Spare 8 /* overrides Global_spare */
207 #define DDF_Foreign 16
208 #define DDF_Legacy 32 /* no DDF on this device */
210 #define DDF_Interface_mask 0xf00
211 #define DDF_Interface_SCSI 0x100
212 #define DDF_Interface_SAS 0x200
213 #define DDF_Interface_SATA 0x300
214 #define DDF_Interface_FC 0x400
216 /* phys_disk_entry.state is a bigendian bitmap */
218 #define DDF_Failed 2 /* overrides 1,4,8 */
219 #define DDF_Rebuilding 4
220 #define DDF_Transition 8
222 #define DDF_ReadErrors 32
223 #define DDF_Missing 64
225 /* The content of the virt_section global scope */
226 struct virtual_disk
{
227 __u32 magic
; /* DDF_VIRT_RECORDS_MAGIC */
229 __u16 populated_vdes
;
232 struct virtual_entry
{
233 char guid
[DDF_GUID_LEN
];
235 __u16 pad0
; /* 0xffff */
245 /* virtual_entry.type is a bitmap - bigendian */
247 #define DDF_Enforce_Groups 2
248 #define DDF_Unicode 4
249 #define DDF_Owner_Valid 8
251 /* virtual_entry.state is a bigendian bitmap */
252 #define DDF_state_mask 0x7
253 #define DDF_state_optimal 0x0
254 #define DDF_state_degraded 0x1
255 #define DDF_state_deleted 0x2
256 #define DDF_state_missing 0x3
257 #define DDF_state_failed 0x4
258 #define DDF_state_part_optimal 0x5
260 #define DDF_state_morphing 0x8
261 #define DDF_state_inconsistent 0x10
263 /* virtual_entry.init_state is a bigendian bitmap */
264 #define DDF_initstate_mask 0x03
265 #define DDF_init_not 0x00
266 #define DDF_init_quick 0x01 /* initialisation is progress.
267 * i.e. 'state_inconsistent' */
268 #define DDF_init_full 0x02
270 #define DDF_access_mask 0xc0
271 #define DDF_access_rw 0x00
272 #define DDF_access_ro 0x80
273 #define DDF_access_blocked 0xc0
275 /* The content of the config_section - local scope
276 * It has multiple records each config_record_len sectors
277 * They can be vd_config or spare_assign
281 __u32 magic
; /* DDF_VD_CONF_MAGIC */
283 char guid
[DDF_GUID_LEN
];
287 __u16 prim_elmnt_count
;
288 __u8 chunk_shift
; /* 0 == 512, 1==1024 etc */
291 __u8 sec_elmnt_count
;
294 __u64 blocks
; /* blocks per component could be different
295 * on different component devices...(only
296 * for concat I hope) */
297 __u64 array_blocks
; /* blocks in array */
305 __u8 v0
[32]; /* reserved- 0xff */
306 __u8 v1
[32]; /* reserved- 0xff */
307 __u8 v2
[16]; /* reserved- 0xff */
308 __u8 v3
[16]; /* reserved- 0xff */
310 __u32 phys_refnum
[0]; /* refnum of each disk in sequence */
311 /*__u64 lba_offset[0]; LBA offset in each phys. Note extents in a
312 bvd are always the same size */
314 #define LBA_OFFSET(ddf, vd) ((__u64 *) &(vd)->phys_refnum[(ddf)->mppe])
316 /* vd_config.cache_pol[7] is a bitmap */
317 #define DDF_cache_writeback 1 /* else writethrough */
318 #define DDF_cache_wadaptive 2 /* only applies if writeback */
319 #define DDF_cache_readahead 4
320 #define DDF_cache_radaptive 8 /* only if doing read-ahead */
321 #define DDF_cache_ifnobatt 16 /* even to write cache if battery is poor */
322 #define DDF_cache_wallowed 32 /* enable write caching */
323 #define DDF_cache_rallowed 64 /* enable read caching */
325 struct spare_assign
{
326 __u32 magic
; /* DDF_SPARE_ASSIGN_MAGIC */
331 __u16 populated
; /* SAEs used */
332 __u16 max
; /* max SAEs */
334 struct spare_assign_entry
{
335 char guid
[DDF_GUID_LEN
];
336 __u16 secondary_element
;
340 /* spare_assign.type is a bitmap */
341 #define DDF_spare_dedicated 0x1 /* else global */
342 #define DDF_spare_revertible 0x2 /* else committable */
343 #define DDF_spare_active 0x4 /* else not active */
344 #define DDF_spare_affinity 0x8 /* enclosure affinity */
346 /* The data_section contents - local scope */
348 __u32 magic
; /* DDF_PHYS_DATA_MAGIC */
350 char guid
[DDF_GUID_LEN
];
351 __u32 refnum
; /* crc of some magic drive data ... */
352 __u8 forced_ref
; /* set when above was not result of magic */
353 __u8 forced_guid
; /* set if guid was forced rather than magic */
358 /* bbm_section content */
359 struct bad_block_log
{
366 struct mapped_block
{
367 __u64 defective_start
;
368 __u32 replacement_start
;
374 /* Struct for internally holding ddf structures */
375 /* The DDF structure stored on each device is potentially
376 * quite different, as some data is global and some is local.
377 * The global data is:
380 * - Physical disk records
381 * - Virtual disk records
383 * - Configuration records
384 * - Physical Disk data section
385 * ( and Bad block and vendor which I don't care about yet).
387 * The local data is parsed into separate lists as it is read
388 * and reconstructed for writing. This means that we only need
389 * to make config changes once and they are automatically
390 * propagated to all devices.
391 * Note that the ddf_super has space of the conf and disk data
392 * for this disk and also for a list of all such data.
393 * The list is only used for the superblock that is being
394 * built in Create or Assemble to describe the whole array.
397 struct ddf_header anchor
, primary
, secondary
;
398 struct ddf_controller_data controller
;
399 struct ddf_header
*active
;
400 struct phys_disk
*phys
;
401 struct virtual_disk
*virt
;
403 unsigned int max_part
, mppe
, conf_rec_len
;
411 unsigned int vcnum
; /* index into ->virt */
412 struct vd_config
**other_bvds
;
413 __u64
*block_sizes
; /* NULL if all the same */
416 struct vd_config conf
;
417 } *conflist
, *currentconf
;
426 unsigned long long size
; /* sectors */
427 unsigned long long primary_lba
; /* sectors */
428 unsigned long long secondary_lba
; /* sectors */
429 unsigned long long workspace_lba
; /* sectors */
430 int pdnum
; /* index in ->phys */
431 struct spare_assign
*spare
;
432 void *mdupdate
; /* hold metadata update */
434 /* These fields used by auto-layout */
435 int raiddisk
; /* slot to fill in autolayout */
439 struct disk_data disk
;
440 struct vcl
*vlist
[0]; /* max_part in size */
445 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
449 static int all_ff(const char *guid
);
450 static void pr_state(struct ddf_super
*ddf
, const char *msg
)
453 dprintf("%s/%s: ", __func__
, msg
);
454 for (i
= 0; i
< __be16_to_cpu(ddf
->active
->max_vd_entries
); i
++) {
455 if (all_ff(ddf
->virt
->entries
[i
].guid
))
457 dprintf("%u(s=%02x i=%02x) ", i
,
458 ddf
->virt
->entries
[i
].state
,
459 ddf
->virt
->entries
[i
].init_state
);
464 static void pr_state(const struct ddf_super
*ddf
, const char *msg
) {}
467 #define ddf_set_updates_pending(x) \
468 do { (x)->updates_pending = 1; pr_state(x, __func__); } while (0)
470 static unsigned int get_pd_index_from_refnum(const struct vcl
*vc
,
471 __u32 refnum
, unsigned int nmax
,
472 const struct vd_config
**bvd
,
475 static unsigned int calc_crc(void *buf
, int len
)
477 /* crcs are always at the same place as in the ddf_header */
478 struct ddf_header
*ddf
= buf
;
479 __u32 oldcrc
= ddf
->crc
;
481 ddf
->crc
= 0xffffffff;
483 newcrc
= crc32(0, buf
, len
);
485 /* The crc is store (like everything) bigendian, so convert
486 * here for simplicity
488 return __cpu_to_be32(newcrc
);
491 #define DDF_INVALID_LEVEL 0xff
492 #define DDF_NO_SECONDARY 0xff
493 static int err_bad_md_layout(const mdu_array_info_t
*array
)
495 pr_err("RAID%d layout %x with %d disks is unsupported for DDF\n",
496 array
->level
, array
->layout
, array
->raid_disks
);
497 return DDF_INVALID_LEVEL
;
500 static int layout_md2ddf(const mdu_array_info_t
*array
,
501 struct vd_config
*conf
)
503 __u16 prim_elmnt_count
= __cpu_to_be16(array
->raid_disks
);
504 __u8 prl
= DDF_INVALID_LEVEL
, rlq
= 0;
505 __u8 sec_elmnt_count
= 1;
506 __u8 srl
= DDF_NO_SECONDARY
;
508 switch (array
->level
) {
513 rlq
= DDF_RAID0_SIMPLE
;
517 switch (array
->raid_disks
) {
519 rlq
= DDF_RAID1_SIMPLE
;
522 rlq
= DDF_RAID1_MULTI
;
525 return err_bad_md_layout(array
);
530 if (array
->layout
!= 0)
531 return err_bad_md_layout(array
);
536 switch (array
->layout
) {
537 case ALGORITHM_LEFT_ASYMMETRIC
:
538 rlq
= DDF_RAID5_N_RESTART
;
540 case ALGORITHM_RIGHT_ASYMMETRIC
:
541 rlq
= DDF_RAID5_0_RESTART
;
543 case ALGORITHM_LEFT_SYMMETRIC
:
544 rlq
= DDF_RAID5_N_CONTINUE
;
546 case ALGORITHM_RIGHT_SYMMETRIC
:
547 /* not mentioned in standard */
549 return err_bad_md_layout(array
);
554 switch (array
->layout
) {
555 case ALGORITHM_ROTATING_N_RESTART
:
556 rlq
= DDF_RAID5_N_RESTART
;
558 case ALGORITHM_ROTATING_ZERO_RESTART
:
559 rlq
= DDF_RAID6_0_RESTART
;
561 case ALGORITHM_ROTATING_N_CONTINUE
:
562 rlq
= DDF_RAID5_N_CONTINUE
;
565 return err_bad_md_layout(array
);
570 if (array
->raid_disks
% 2 == 0 && array
->layout
== 0x102) {
571 rlq
= DDF_RAID1_SIMPLE
;
572 prim_elmnt_count
= __cpu_to_be16(2);
573 sec_elmnt_count
= array
->raid_disks
/ 2;
574 } else if (array
->raid_disks
% 3 == 0
575 && array
->layout
== 0x103) {
576 rlq
= DDF_RAID1_MULTI
;
577 prim_elmnt_count
= __cpu_to_be16(3);
578 sec_elmnt_count
= array
->raid_disks
/ 3;
580 return err_bad_md_layout(array
);
585 return err_bad_md_layout(array
);
588 conf
->prim_elmnt_count
= prim_elmnt_count
;
591 conf
->sec_elmnt_count
= sec_elmnt_count
;
595 static int err_bad_ddf_layout(const struct vd_config
*conf
)
597 pr_err("DDF RAID %u qualifier %u with %u disks is unsupported\n",
598 conf
->prl
, conf
->rlq
, __be16_to_cpu(conf
->prim_elmnt_count
));
602 static int layout_ddf2md(const struct vd_config
*conf
,
603 mdu_array_info_t
*array
)
605 int level
= LEVEL_UNSUPPORTED
;
607 int raiddisks
= __be16_to_cpu(conf
->prim_elmnt_count
);
609 if (conf
->sec_elmnt_count
> 1) {
610 /* see also check_secondary() */
611 if (conf
->prl
!= DDF_RAID1
||
612 (conf
->srl
!= DDF_2STRIPED
&& conf
->srl
!= DDF_2SPANNED
)) {
613 pr_err("Unsupported secondary RAID level %u/%u\n",
614 conf
->prl
, conf
->srl
);
617 if (raiddisks
== 2 && conf
->rlq
== DDF_RAID1_SIMPLE
)
619 else if (raiddisks
== 3 && conf
->rlq
== DDF_RAID1_MULTI
)
622 return err_bad_ddf_layout(conf
);
623 raiddisks
*= conf
->sec_elmnt_count
;
630 level
= LEVEL_LINEAR
;
633 if (conf
->rlq
!= DDF_RAID0_SIMPLE
)
634 return err_bad_ddf_layout(conf
);
638 if (!((conf
->rlq
== DDF_RAID1_SIMPLE
&& raiddisks
== 2) ||
639 (conf
->rlq
== DDF_RAID1_MULTI
&& raiddisks
== 3)))
640 return err_bad_ddf_layout(conf
);
644 if (conf
->rlq
!= DDF_RAID4_N
)
645 return err_bad_ddf_layout(conf
);
650 case DDF_RAID5_N_RESTART
:
651 layout
= ALGORITHM_LEFT_ASYMMETRIC
;
653 case DDF_RAID5_0_RESTART
:
654 layout
= ALGORITHM_RIGHT_ASYMMETRIC
;
656 case DDF_RAID5_N_CONTINUE
:
657 layout
= ALGORITHM_LEFT_SYMMETRIC
;
660 return err_bad_ddf_layout(conf
);
666 case DDF_RAID5_N_RESTART
:
667 layout
= ALGORITHM_ROTATING_N_RESTART
;
669 case DDF_RAID6_0_RESTART
:
670 layout
= ALGORITHM_ROTATING_ZERO_RESTART
;
672 case DDF_RAID5_N_CONTINUE
:
673 layout
= ALGORITHM_ROTATING_N_CONTINUE
;
676 return err_bad_ddf_layout(conf
);
681 return err_bad_ddf_layout(conf
);
685 array
->level
= level
;
686 array
->layout
= layout
;
687 array
->raid_disks
= raiddisks
;
691 static int load_ddf_header(int fd
, unsigned long long lba
,
692 unsigned long long size
,
694 struct ddf_header
*hdr
, struct ddf_header
*anchor
)
696 /* read a ddf header (primary or secondary) from fd/lba
697 * and check that it is consistent with anchor
699 * magic, crc, guid, rev, and LBA's header_type, and
700 * everything after header_type must be the same
705 if (lseek64(fd
, lba
<<9, 0) < 0)
708 if (read(fd
, hdr
, 512) != 512)
711 if (hdr
->magic
!= DDF_HEADER_MAGIC
)
713 if (calc_crc(hdr
, 512) != hdr
->crc
)
715 if (memcmp(anchor
->guid
, hdr
->guid
, DDF_GUID_LEN
) != 0 ||
716 memcmp(anchor
->revision
, hdr
->revision
, 8) != 0 ||
717 anchor
->primary_lba
!= hdr
->primary_lba
||
718 anchor
->secondary_lba
!= hdr
->secondary_lba
||
720 memcmp(anchor
->pad2
, hdr
->pad2
, 512 -
721 offsetof(struct ddf_header
, pad2
)) != 0)
724 /* Looks good enough to me... */
728 static void *load_section(int fd
, struct ddf_super
*super
, void *buf
,
729 __u32 offset_be
, __u32 len_be
, int check
)
731 unsigned long long offset
= __be32_to_cpu(offset_be
);
732 unsigned long long len
= __be32_to_cpu(len_be
);
733 int dofree
= (buf
== NULL
);
736 if (len
!= 2 && len
!= 8 && len
!= 32
737 && len
!= 128 && len
!= 512)
743 /* All pre-allocated sections are a single block */
746 } else if (posix_memalign(&buf
, 512, len
<<9) != 0)
752 if (super
->active
->type
== 1)
753 offset
+= __be64_to_cpu(super
->active
->primary_lba
);
755 offset
+= __be64_to_cpu(super
->active
->secondary_lba
);
757 if ((unsigned long long)lseek64(fd
, offset
<<9, 0) != (offset
<<9)) {
762 if ((unsigned long long)read(fd
, buf
, len
<<9) != (len
<<9)) {
770 static int load_ddf_headers(int fd
, struct ddf_super
*super
, char *devname
)
772 unsigned long long dsize
;
774 get_dev_size(fd
, NULL
, &dsize
);
776 if (lseek64(fd
, dsize
-512, 0) < 0) {
778 pr_err("Cannot seek to anchor block on %s: %s\n",
779 devname
, strerror(errno
));
782 if (read(fd
, &super
->anchor
, 512) != 512) {
784 pr_err("Cannot read anchor block on %s: %s\n",
785 devname
, strerror(errno
));
788 if (super
->anchor
.magic
!= DDF_HEADER_MAGIC
) {
790 pr_err("no DDF anchor found on %s\n",
794 if (calc_crc(&super
->anchor
, 512) != super
->anchor
.crc
) {
796 pr_err("bad CRC on anchor on %s\n",
800 if (memcmp(super
->anchor
.revision
, DDF_REVISION_0
, 8) != 0 &&
801 memcmp(super
->anchor
.revision
, DDF_REVISION_2
, 8) != 0) {
803 pr_err("can only support super revision"
804 " %.8s and earlier, not %.8s on %s\n",
805 DDF_REVISION_2
, super
->anchor
.revision
,devname
);
808 super
->active
= NULL
;
809 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.primary_lba
),
811 &super
->primary
, &super
->anchor
) == 0) {
813 pr_err("Failed to load primary DDF header "
816 super
->active
= &super
->primary
;
817 if (load_ddf_header(fd
, __be64_to_cpu(super
->anchor
.secondary_lba
),
819 &super
->secondary
, &super
->anchor
)) {
820 if ((__be32_to_cpu(super
->primary
.seq
)
821 < __be32_to_cpu(super
->secondary
.seq
) &&
822 !super
->secondary
.openflag
)
823 || (__be32_to_cpu(super
->primary
.seq
)
824 == __be32_to_cpu(super
->secondary
.seq
) &&
825 super
->primary
.openflag
&& !super
->secondary
.openflag
)
826 || super
->active
== NULL
828 super
->active
= &super
->secondary
;
830 pr_err("Failed to load secondary DDF header on %s\n",
832 if (super
->active
== NULL
)
837 static int load_ddf_global(int fd
, struct ddf_super
*super
, char *devname
)
840 ok
= load_section(fd
, super
, &super
->controller
,
841 super
->active
->controller_section_offset
,
842 super
->active
->controller_section_length
,
844 super
->phys
= load_section(fd
, super
, NULL
,
845 super
->active
->phys_section_offset
,
846 super
->active
->phys_section_length
,
848 super
->pdsize
= __be32_to_cpu(super
->active
->phys_section_length
) * 512;
850 super
->virt
= load_section(fd
, super
, NULL
,
851 super
->active
->virt_section_offset
,
852 super
->active
->virt_section_length
,
854 super
->vdsize
= __be32_to_cpu(super
->active
->virt_section_length
) * 512;
864 super
->conflist
= NULL
;
867 super
->max_part
= __be16_to_cpu(super
->active
->max_partitions
);
868 super
->mppe
= __be16_to_cpu(super
->active
->max_primary_element_entries
);
869 super
->conf_rec_len
= __be16_to_cpu(super
->active
->config_record_len
);
873 #define DDF_UNUSED_BVD 0xff
874 static int alloc_other_bvds(const struct ddf_super
*ddf
, struct vcl
*vcl
)
876 unsigned int n_vds
= vcl
->conf
.sec_elmnt_count
- 1;
877 unsigned int i
, vdsize
;
880 vcl
->other_bvds
= NULL
;
883 vdsize
= ddf
->conf_rec_len
* 512;
884 if (posix_memalign(&p
, 512, n_vds
*
885 (vdsize
+ sizeof(struct vd_config
*))) != 0)
887 vcl
->other_bvds
= (struct vd_config
**) (p
+ n_vds
* vdsize
);
888 for (i
= 0; i
< n_vds
; i
++) {
889 vcl
->other_bvds
[i
] = p
+ i
* vdsize
;
890 memset(vcl
->other_bvds
[i
], 0, vdsize
);
891 vcl
->other_bvds
[i
]->sec_elmnt_seq
= DDF_UNUSED_BVD
;
896 static void add_other_bvd(struct vcl
*vcl
, struct vd_config
*vd
,
900 for (i
= 0; i
< vcl
->conf
.sec_elmnt_count
-1; i
++)
901 if (vcl
->other_bvds
[i
]->sec_elmnt_seq
== vd
->sec_elmnt_seq
)
904 if (i
< vcl
->conf
.sec_elmnt_count
-1) {
905 if (vd
->seqnum
<= vcl
->other_bvds
[i
]->seqnum
)
908 for (i
= 0; i
< vcl
->conf
.sec_elmnt_count
-1; i
++)
909 if (vcl
->other_bvds
[i
]->sec_elmnt_seq
== DDF_UNUSED_BVD
)
911 if (i
== vcl
->conf
.sec_elmnt_count
-1) {
912 pr_err("no space for sec level config %u, count is %u\n",
913 vd
->sec_elmnt_seq
, vcl
->conf
.sec_elmnt_count
);
917 memcpy(vcl
->other_bvds
[i
], vd
, len
);
920 static int load_ddf_local(int fd
, struct ddf_super
*super
,
921 char *devname
, int keep
)
927 unsigned int confsec
;
929 unsigned int max_virt_disks
= __be16_to_cpu(super
->active
->max_vd_entries
);
930 unsigned long long dsize
;
932 /* First the local disk info */
933 if (posix_memalign((void**)&dl
, 512,
935 (super
->max_part
) * sizeof(dl
->vlist
[0])) != 0) {
936 pr_err("%s could not allocate disk info buffer\n",
941 load_section(fd
, super
, &dl
->disk
,
942 super
->active
->data_section_offset
,
943 super
->active
->data_section_length
,
945 dl
->devname
= devname
? xstrdup(devname
) : NULL
;
948 dl
->major
= major(stb
.st_rdev
);
949 dl
->minor
= minor(stb
.st_rdev
);
950 dl
->next
= super
->dlist
;
951 dl
->fd
= keep
? fd
: -1;
954 if (get_dev_size(fd
, devname
, &dsize
))
955 dl
->size
= dsize
>> 9;
956 /* If the disks have different sizes, the LBAs will differ
957 * between phys disks.
958 * At this point here, the values in super->active must be valid
959 * for this phys disk. */
960 dl
->primary_lba
= super
->active
->primary_lba
;
961 dl
->secondary_lba
= super
->active
->secondary_lba
;
962 dl
->workspace_lba
= super
->active
->workspace_lba
;
964 for (i
= 0 ; i
< super
->max_part
; i
++)
968 for (i
= 0; i
< __be16_to_cpu(super
->active
->max_pd_entries
); i
++)
969 if (memcmp(super
->phys
->entries
[i
].guid
,
970 dl
->disk
.guid
, DDF_GUID_LEN
) == 0)
973 /* Now the config list. */
974 /* 'conf' is an array of config entries, some of which are
975 * probably invalid. Those which are good need to be copied into
979 conf
= load_section(fd
, super
, NULL
,
980 super
->active
->config_section_offset
,
981 super
->active
->config_section_length
,
986 confsec
< __be32_to_cpu(super
->active
->config_section_length
);
987 confsec
+= super
->conf_rec_len
) {
988 struct vd_config
*vd
=
989 (struct vd_config
*)((char*)conf
+ confsec
*512);
992 if (vd
->magic
== DDF_SPARE_ASSIGN_MAGIC
) {
995 if (posix_memalign((void**)&dl
->spare
, 512,
996 super
->conf_rec_len
*512) != 0) {
997 pr_err("%s could not allocate spare info buf\n",
1002 memcpy(dl
->spare
, vd
, super
->conf_rec_len
*512);
1005 if (vd
->magic
!= DDF_VD_CONF_MAGIC
)
1007 for (vcl
= super
->conflist
; vcl
; vcl
= vcl
->next
) {
1008 if (memcmp(vcl
->conf
.guid
,
1009 vd
->guid
, DDF_GUID_LEN
) == 0)
1014 dl
->vlist
[vnum
++] = vcl
;
1015 if (vcl
->other_bvds
!= NULL
&&
1016 vcl
->conf
.sec_elmnt_seq
!= vd
->sec_elmnt_seq
) {
1017 add_other_bvd(vcl
, vd
, super
->conf_rec_len
*512);
1020 if (__be32_to_cpu(vd
->seqnum
) <=
1021 __be32_to_cpu(vcl
->conf
.seqnum
))
1024 if (posix_memalign((void**)&vcl
, 512,
1025 (super
->conf_rec_len
*512 +
1026 offsetof(struct vcl
, conf
))) != 0) {
1027 pr_err("%s could not allocate vcl buf\n",
1031 vcl
->next
= super
->conflist
;
1032 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
1033 vcl
->conf
.sec_elmnt_count
= vd
->sec_elmnt_count
;
1034 if (alloc_other_bvds(super
, vcl
) != 0) {
1035 pr_err("%s could not allocate other bvds\n",
1040 super
->conflist
= vcl
;
1041 dl
->vlist
[vnum
++] = vcl
;
1043 memcpy(&vcl
->conf
, vd
, super
->conf_rec_len
*512);
1044 for (i
=0; i
< max_virt_disks
; i
++)
1045 if (memcmp(super
->virt
->entries
[i
].guid
,
1046 vcl
->conf
.guid
, DDF_GUID_LEN
)==0)
1048 if (i
< max_virt_disks
)
1057 static int load_super_ddf_all(struct supertype
*st
, int fd
,
1058 void **sbp
, char *devname
);
1061 static void free_super_ddf(struct supertype
*st
);
1063 static int load_super_ddf(struct supertype
*st
, int fd
,
1066 unsigned long long dsize
;
1067 struct ddf_super
*super
;
1070 if (get_dev_size(fd
, devname
, &dsize
) == 0)
1073 if (!st
->ignore_hw_compat
&& test_partition(fd
))
1074 /* DDF is not allowed on partitions */
1077 /* 32M is a lower bound */
1078 if (dsize
<= 32*1024*1024) {
1080 pr_err("%s is too small for ddf: "
1081 "size is %llu sectors.\n",
1087 pr_err("%s is an odd size for ddf: "
1088 "size is %llu bytes.\n",
1095 if (posix_memalign((void**)&super
, 512, sizeof(*super
))!= 0) {
1096 pr_err("malloc of %zu failed.\n",
1100 memset(super
, 0, sizeof(*super
));
1102 rv
= load_ddf_headers(fd
, super
, devname
);
1108 /* Have valid headers and have chosen the best. Let's read in the rest*/
1110 rv
= load_ddf_global(fd
, super
, devname
);
1114 pr_err("Failed to load all information "
1115 "sections on %s\n", devname
);
1120 rv
= load_ddf_local(fd
, super
, devname
, 0);
1124 pr_err("Failed to load all information "
1125 "sections on %s\n", devname
);
1130 /* Should possibly check the sections .... */
1133 if (st
->ss
== NULL
) {
1134 st
->ss
= &super_ddf
;
1135 st
->minor_version
= 0;
1142 static void free_super_ddf(struct supertype
*st
)
1144 struct ddf_super
*ddf
= st
->sb
;
1149 while (ddf
->conflist
) {
1150 struct vcl
*v
= ddf
->conflist
;
1151 ddf
->conflist
= v
->next
;
1153 free(v
->block_sizes
);
1156 v->other_bvds[0] points to beginning of buffer,
1157 see alloc_other_bvds()
1159 free(v
->other_bvds
[0]);
1162 while (ddf
->dlist
) {
1163 struct dl
*d
= ddf
->dlist
;
1164 ddf
->dlist
= d
->next
;
1171 while (ddf
->add_list
) {
1172 struct dl
*d
= ddf
->add_list
;
1173 ddf
->add_list
= d
->next
;
1184 static struct supertype
*match_metadata_desc_ddf(char *arg
)
1186 /* 'ddf' only support containers */
1187 struct supertype
*st
;
1188 if (strcmp(arg
, "ddf") != 0 &&
1189 strcmp(arg
, "default") != 0
1193 st
= xcalloc(1, sizeof(*st
));
1194 st
->ss
= &super_ddf
;
1196 st
->minor_version
= 0;
1203 static mapping_t ddf_state
[] = {
1209 { "Partially Optimal", 5},
1215 static mapping_t ddf_init_state
[] = {
1216 { "Not Initialised", 0},
1217 { "QuickInit in Progress", 1},
1218 { "Fully Initialised", 2},
1222 static mapping_t ddf_access
[] = {
1226 { "Blocked (no access)", 3},
1230 static mapping_t ddf_level
[] = {
1231 { "RAID0", DDF_RAID0
},
1232 { "RAID1", DDF_RAID1
},
1233 { "RAID3", DDF_RAID3
},
1234 { "RAID4", DDF_RAID4
},
1235 { "RAID5", DDF_RAID5
},
1236 { "RAID1E",DDF_RAID1E
},
1237 { "JBOD", DDF_JBOD
},
1238 { "CONCAT",DDF_CONCAT
},
1239 { "RAID5E",DDF_RAID5E
},
1240 { "RAID5EE",DDF_RAID5EE
},
1241 { "RAID6", DDF_RAID6
},
1244 static mapping_t ddf_sec_level
[] = {
1245 { "Striped", DDF_2STRIPED
},
1246 { "Mirrored", DDF_2MIRRORED
},
1247 { "Concat", DDF_2CONCAT
},
1248 { "Spanned", DDF_2SPANNED
},
1253 static int all_ff(const char *guid
)
1256 for (i
= 0; i
< DDF_GUID_LEN
; i
++)
1257 if (guid
[i
] != (char)0xff)
1263 static void print_guid(char *guid
, int tstamp
)
1265 /* A GUIDs are part (or all) ASCII and part binary.
1266 * They tend to be space padded.
1267 * We print the GUID in HEX, then in parentheses add
1268 * any initial ASCII sequence, and a possible
1269 * time stamp from bytes 16-19
1271 int l
= DDF_GUID_LEN
;
1274 for (i
=0 ; i
<DDF_GUID_LEN
; i
++) {
1275 if ((i
&3)==0 && i
!= 0) printf(":");
1276 printf("%02X", guid
[i
]&255);
1280 while (l
&& guid
[l
-1] == ' ')
1282 for (i
=0 ; i
<l
; i
++) {
1283 if (guid
[i
] >= 0x20 && guid
[i
] < 0x7f)
1284 fputc(guid
[i
], stdout
);
1289 time_t then
= __be32_to_cpu(*(__u32
*)(guid
+16)) + DECADE
;
1292 tm
= localtime(&then
);
1293 strftime(tbuf
, 100, " %D %T",tm
);
1294 fputs(tbuf
, stdout
);
1299 static const char *guid_str(const char *guid
)
1301 static char buf
[DDF_GUID_LEN
*2+1];
1304 for (i
= 0; i
< DDF_GUID_LEN
; i
++) {
1305 unsigned char c
= guid
[i
];
1306 if (c
>= 32 && c
< 127)
1307 p
+= sprintf(p
, "%c", c
);
1309 p
+= sprintf(p
, "%02x", c
);
1312 return (const char *) buf
;
1315 static void examine_vd(int n
, struct ddf_super
*sb
, char *guid
)
1317 int crl
= sb
->conf_rec_len
;
1320 for (vcl
= sb
->conflist
; vcl
; vcl
= vcl
->next
) {
1322 struct vd_config
*vc
= &vcl
->conf
;
1324 if (calc_crc(vc
, crl
*512) != vc
->crc
)
1326 if (memcmp(vc
->guid
, guid
, DDF_GUID_LEN
) != 0)
1329 /* Ok, we know about this VD, let's give more details */
1330 printf(" Raid Devices[%d] : %d (", n
,
1331 __be16_to_cpu(vc
->prim_elmnt_count
));
1332 for (i
= 0; i
< __be16_to_cpu(vc
->prim_elmnt_count
); i
++) {
1334 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1335 for (j
=0; j
<cnt
; j
++)
1336 if (vc
->phys_refnum
[i
] == sb
->phys
->entries
[j
].refnum
)
1345 if (vc
->chunk_shift
!= 255)
1346 printf(" Chunk Size[%d] : %d sectors\n", n
,
1347 1 << vc
->chunk_shift
);
1348 printf(" Raid Level[%d] : %s\n", n
,
1349 map_num(ddf_level
, vc
->prl
)?:"-unknown-");
1350 if (vc
->sec_elmnt_count
!= 1) {
1351 printf(" Secondary Position[%d] : %d of %d\n", n
,
1352 vc
->sec_elmnt_seq
, vc
->sec_elmnt_count
);
1353 printf(" Secondary Level[%d] : %s\n", n
,
1354 map_num(ddf_sec_level
, vc
->srl
) ?: "-unknown-");
1356 printf(" Device Size[%d] : %llu\n", n
,
1357 (unsigned long long)__be64_to_cpu(vc
->blocks
)/2);
1358 printf(" Array Size[%d] : %llu\n", n
,
1359 (unsigned long long)__be64_to_cpu(vc
->array_blocks
)/2);
1363 static void examine_vds(struct ddf_super
*sb
)
1365 int cnt
= __be16_to_cpu(sb
->virt
->populated_vdes
);
1367 printf(" Virtual Disks : %d\n", cnt
);
1369 for (i
= 0; i
< __be16_to_cpu(sb
->virt
->max_vdes
); i
++) {
1370 struct virtual_entry
*ve
= &sb
->virt
->entries
[i
];
1371 if (all_ff(ve
->guid
))
1374 printf(" VD GUID[%d] : ", i
); print_guid(ve
->guid
, 1);
1376 printf(" unit[%d] : %d\n", i
, __be16_to_cpu(ve
->unit
));
1377 printf(" state[%d] : %s, %s%s\n", i
,
1378 map_num(ddf_state
, ve
->state
& 7),
1379 (ve
->state
& 8) ? "Morphing, ": "",
1380 (ve
->state
& 16)? "Not Consistent" : "Consistent");
1381 printf(" init state[%d] : %s\n", i
,
1382 map_num(ddf_init_state
, ve
->init_state
&3));
1383 printf(" access[%d] : %s\n", i
,
1384 map_num(ddf_access
, (ve
->init_state
>>6) & 3));
1385 printf(" Name[%d] : %.16s\n", i
, ve
->name
);
1386 examine_vd(i
, sb
, ve
->guid
);
1388 if (cnt
) printf("\n");
1391 static void examine_pds(struct ddf_super
*sb
)
1393 int cnt
= __be16_to_cpu(sb
->phys
->used_pdes
);
1396 printf(" Physical Disks : %d\n", cnt
);
1397 printf(" Number RefNo Size Device Type/State\n");
1399 for (i
=0 ; i
<cnt
; i
++) {
1400 struct phys_disk_entry
*pd
= &sb
->phys
->entries
[i
];
1401 int type
= __be16_to_cpu(pd
->type
);
1402 int state
= __be16_to_cpu(pd
->state
);
1404 //printf(" PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1406 printf(" %3d %08x ", i
,
1407 __be32_to_cpu(pd
->refnum
));
1409 (unsigned long long)__be64_to_cpu(pd
->config_size
)>>1);
1410 for (dl
= sb
->dlist
; dl
; dl
= dl
->next
) {
1411 if (dl
->disk
.refnum
== pd
->refnum
) {
1412 char *dv
= map_dev(dl
->major
, dl
->minor
, 0);
1414 printf("%-15s", dv
);
1421 printf(" %s%s%s%s%s",
1422 (type
&2) ? "active":"",
1423 (type
&4) ? "Global-Spare":"",
1424 (type
&8) ? "spare" : "",
1425 (type
&16)? ", foreign" : "",
1426 (type
&32)? "pass-through" : "");
1427 if (state
& DDF_Failed
)
1428 /* This over-rides these three */
1429 state
&= ~(DDF_Online
|DDF_Rebuilding
|DDF_Transition
);
1430 printf("/%s%s%s%s%s%s%s",
1431 (state
&1)? "Online": "Offline",
1432 (state
&2)? ", Failed": "",
1433 (state
&4)? ", Rebuilding": "",
1434 (state
&8)? ", in-transition": "",
1435 (state
&16)? ", SMART-errors": "",
1436 (state
&32)? ", Unrecovered-Read-Errors": "",
1437 (state
&64)? ", Missing" : "");
1442 static void examine_super_ddf(struct supertype
*st
, char *homehost
)
1444 struct ddf_super
*sb
= st
->sb
;
1446 printf(" Magic : %08x\n", __be32_to_cpu(sb
->anchor
.magic
));
1447 printf(" Version : %.8s\n", sb
->anchor
.revision
);
1448 printf("Controller GUID : "); print_guid(sb
->controller
.guid
, 0);
1450 printf(" Container GUID : "); print_guid(sb
->anchor
.guid
, 1);
1452 printf(" Seq : %08x\n", __be32_to_cpu(sb
->active
->seq
));
1453 printf(" Redundant hdr : %s\n", sb
->secondary
.magic
== DDF_HEADER_MAGIC
1459 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1461 static void uuid_from_ddf_guid(const char *guid
, int uuid
[4]);
1462 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4]);
1464 static unsigned int get_vd_num_of_subarray(struct supertype
*st
)
1467 * Figure out the VD number for this supertype.
1468 * Returns DDF_CONTAINER for the container itself,
1469 * and DDF_NOTFOUND on error.
1471 struct ddf_super
*ddf
= st
->sb
;
1476 if (*st
->container_devnm
== '\0')
1477 return DDF_CONTAINER
;
1479 sra
= sysfs_read(-1, st
->devnm
, GET_VERSION
);
1480 if (!sra
|| sra
->array
.major_version
!= -1 ||
1481 sra
->array
.minor_version
!= -2 ||
1482 !is_subarray(sra
->text_version
))
1483 return DDF_NOTFOUND
;
1485 sub
= strchr(sra
->text_version
+ 1, '/');
1487 vcnum
= strtoul(sub
+ 1, &end
, 10);
1488 if (sub
== NULL
|| *sub
== '\0' || *end
!= '\0' ||
1489 vcnum
>= __be16_to_cpu(ddf
->active
->max_vd_entries
))
1490 return DDF_NOTFOUND
;
1495 static void brief_examine_super_ddf(struct supertype
*st
, int verbose
)
1497 /* We just write a generic DDF ARRAY entry
1501 getinfo_super_ddf(st
, &info
, NULL
);
1502 fname_from_uuid(st
, &info
, nbuf
, ':');
1504 printf("ARRAY metadata=ddf UUID=%s\n", nbuf
+ 5);
1507 static void brief_examine_subarrays_ddf(struct supertype
*st
, int verbose
)
1509 /* We just write a generic DDF ARRAY entry
1511 struct ddf_super
*ddf
= st
->sb
;
1515 getinfo_super_ddf(st
, &info
, NULL
);
1516 fname_from_uuid(st
, &info
, nbuf
, ':');
1518 for (i
= 0; i
< __be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
1519 struct virtual_entry
*ve
= &ddf
->virt
->entries
[i
];
1522 if (all_ff(ve
->guid
))
1524 memcpy(vcl
.conf
.guid
, ve
->guid
, DDF_GUID_LEN
);
1525 ddf
->currentconf
=&vcl
;
1526 uuid_from_super_ddf(st
, info
.uuid
);
1527 fname_from_uuid(st
, &info
, nbuf1
, ':');
1528 printf("ARRAY container=%s member=%d UUID=%s\n",
1529 nbuf
+5, i
, nbuf1
+5);
1533 static void export_examine_super_ddf(struct supertype
*st
)
1537 getinfo_super_ddf(st
, &info
, NULL
);
1538 fname_from_uuid(st
, &info
, nbuf
, ':');
1539 printf("MD_METADATA=ddf\n");
1540 printf("MD_LEVEL=container\n");
1541 printf("MD_UUID=%s\n", nbuf
+5);
1544 static int copy_metadata_ddf(struct supertype
*st
, int from
, int to
)
1547 unsigned long long dsize
, offset
;
1549 struct ddf_header
*ddf
;
1552 /* The meta consists of an anchor, a primary, and a secondary.
1553 * This all lives at the end of the device.
1554 * So it is easiest to find the earliest of primary and
1555 * secondary, and copy everything from there.
1557 * Anchor is 512 from end It contains primary_lba and secondary_lba
1558 * we choose one of those
1561 if (posix_memalign(&buf
, 4096, 4096) != 0)
1564 if (!get_dev_size(from
, NULL
, &dsize
))
1567 if (lseek64(from
, dsize
-512, 0) < 0)
1569 if (read(from
, buf
, 512) != 512)
1572 if (ddf
->magic
!= DDF_HEADER_MAGIC
||
1573 calc_crc(ddf
, 512) != ddf
->crc
||
1574 (memcmp(ddf
->revision
, DDF_REVISION_0
, 8) != 0 &&
1575 memcmp(ddf
->revision
, DDF_REVISION_2
, 8) != 0))
1578 offset
= dsize
- 512;
1579 if ((__be64_to_cpu(ddf
->primary_lba
) << 9) < offset
)
1580 offset
= __be64_to_cpu(ddf
->primary_lba
) << 9;
1581 if ((__be64_to_cpu(ddf
->secondary_lba
) << 9) < offset
)
1582 offset
= __be64_to_cpu(ddf
->secondary_lba
) << 9;
1584 bytes
= dsize
- offset
;
1586 if (lseek64(from
, offset
, 0) < 0 ||
1587 lseek64(to
, offset
, 0) < 0)
1589 while (written
< bytes
) {
1590 int n
= bytes
- written
;
1593 if (read(from
, buf
, n
) != n
)
1595 if (write(to
, buf
, n
) != n
)
1606 static void detail_super_ddf(struct supertype
*st
, char *homehost
)
1609 * Could print DDF GUID
1610 * Need to find which array
1611 * If whole, briefly list all arrays
1616 static void brief_detail_super_ddf(struct supertype
*st
)
1620 struct ddf_super
*ddf
= st
->sb
;
1621 unsigned int vcnum
= get_vd_num_of_subarray(st
);
1622 if (vcnum
== DDF_CONTAINER
)
1623 uuid_from_super_ddf(st
, info
.uuid
);
1624 else if (vcnum
== DDF_NOTFOUND
)
1627 uuid_from_ddf_guid(ddf
->virt
->entries
[vcnum
].guid
, info
.uuid
);
1628 fname_from_uuid(st
, &info
, nbuf
,':');
1629 printf(" UUID=%s", nbuf
+ 5);
1633 static int match_home_ddf(struct supertype
*st
, char *homehost
)
1635 /* It matches 'this' host if the controller is a
1636 * Linux-MD controller with vendor_data matching
1639 struct ddf_super
*ddf
= st
->sb
;
1644 len
= strlen(homehost
);
1646 return (memcmp(ddf
->controller
.guid
, T10
, 8) == 0 &&
1647 len
< sizeof(ddf
->controller
.vendor_data
) &&
1648 memcmp(ddf
->controller
.vendor_data
, homehost
,len
) == 0 &&
1649 ddf
->controller
.vendor_data
[len
] == 0);
1653 static int find_index_in_bvd(const struct ddf_super
*ddf
,
1654 const struct vd_config
*conf
, unsigned int n
,
1655 unsigned int *n_bvd
)
1658 * Find the index of the n-th valid physical disk in this BVD
1661 for (i
= 0, j
= 0; i
< ddf
->mppe
&&
1662 j
< __be16_to_cpu(conf
->prim_elmnt_count
); i
++) {
1663 if (conf
->phys_refnum
[i
] != 0xffffffff) {
1671 dprintf("%s: couldn't find BVD member %u (total %u)\n",
1672 __func__
, n
, __be16_to_cpu(conf
->prim_elmnt_count
));
1676 static struct vd_config
*find_vdcr(struct ddf_super
*ddf
, unsigned int inst
,
1678 unsigned int *n_bvd
, struct vcl
**vcl
)
1682 for (v
= ddf
->conflist
; v
; v
= v
->next
) {
1683 unsigned int nsec
, ibvd
;
1684 struct vd_config
*conf
;
1685 if (inst
!= v
->vcnum
)
1688 if (conf
->sec_elmnt_count
== 1) {
1689 if (find_index_in_bvd(ddf
, conf
, n
, n_bvd
)) {
1695 if (v
->other_bvds
== NULL
) {
1696 pr_err("%s: BUG: other_bvds is NULL, nsec=%u\n",
1697 __func__
, conf
->sec_elmnt_count
);
1700 nsec
= n
/ __be16_to_cpu(conf
->prim_elmnt_count
);
1701 if (conf
->sec_elmnt_seq
!= nsec
) {
1702 for (ibvd
= 1; ibvd
< conf
->sec_elmnt_count
; ibvd
++) {
1703 if (v
->other_bvds
[ibvd
-1]->sec_elmnt_seq
1707 if (ibvd
== conf
->sec_elmnt_count
)
1709 conf
= v
->other_bvds
[ibvd
-1];
1711 if (!find_index_in_bvd(ddf
, conf
,
1712 n
- nsec
*conf
->sec_elmnt_count
, n_bvd
))
1714 dprintf("%s: found disk %u as member %u in bvd %d of array %u\n"
1715 , __func__
, n
, *n_bvd
, ibvd
-1, inst
);
1720 pr_err("%s: Could't find disk %d in array %u\n", __func__
, n
, inst
);
1725 static int find_phys(const struct ddf_super
*ddf
, __u32 phys_refnum
)
1727 /* Find the entry in phys_disk which has the given refnum
1728 * and return it's index
1731 for (i
= 0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
); i
++)
1732 if (ddf
->phys
->entries
[i
].refnum
== phys_refnum
)
1737 static void uuid_from_ddf_guid(const char *guid
, int uuid
[4])
1740 struct sha1_ctx ctx
;
1741 sha1_init_ctx(&ctx
);
1742 sha1_process_bytes(guid
, DDF_GUID_LEN
, &ctx
);
1743 sha1_finish_ctx(&ctx
, buf
);
1744 memcpy(uuid
, buf
, 4*4);
1747 static void uuid_from_super_ddf(struct supertype
*st
, int uuid
[4])
1749 /* The uuid returned here is used for:
1750 * uuid to put into bitmap file (Create, Grow)
1751 * uuid for backup header when saving critical section (Grow)
1752 * comparing uuids when re-adding a device into an array
1753 * In these cases the uuid required is that of the data-array,
1754 * not the device-set.
1755 * uuid to recognise same set when adding a missing device back
1756 * to an array. This is a uuid for the device-set.
1758 * For each of these we can make do with a truncated
1759 * or hashed uuid rather than the original, as long as
1761 * In the case of SVD we assume the BVD is of interest,
1762 * though that might be the case if a bitmap were made for
1763 * a mirrored SVD - worry about that later.
1764 * So we need to find the VD configuration record for the
1765 * relevant BVD and extract the GUID and Secondary_Element_Seq.
1766 * The first 16 bytes of the sha1 of these is used.
1768 struct ddf_super
*ddf
= st
->sb
;
1769 struct vcl
*vcl
= ddf
->currentconf
;
1773 guid
= vcl
->conf
.guid
;
1775 guid
= ddf
->anchor
.guid
;
1776 uuid_from_ddf_guid(guid
, uuid
);
1779 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
, char *map
);
1781 static void getinfo_super_ddf(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1783 struct ddf_super
*ddf
= st
->sb
;
1784 int map_disks
= info
->array
.raid_disks
;
1787 if (ddf
->currentconf
) {
1788 getinfo_super_ddf_bvd(st
, info
, map
);
1791 memset(info
, 0, sizeof(*info
));
1793 info
->array
.raid_disks
= __be16_to_cpu(ddf
->phys
->used_pdes
);
1794 info
->array
.level
= LEVEL_CONTAINER
;
1795 info
->array
.layout
= 0;
1796 info
->array
.md_minor
= -1;
1797 cptr
= (__u32
*)(ddf
->anchor
.guid
+ 16);
1798 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
1800 info
->array
.utime
= 0;
1801 info
->array
.chunk_size
= 0;
1802 info
->container_enough
= 1;
1804 info
->disk
.major
= 0;
1805 info
->disk
.minor
= 0;
1807 info
->disk
.number
= __be32_to_cpu(ddf
->dlist
->disk
.refnum
);
1808 info
->disk
.raid_disk
= find_phys(ddf
, ddf
->dlist
->disk
.refnum
);
1810 info
->data_offset
= __be64_to_cpu(ddf
->phys
->
1811 entries
[info
->disk
.raid_disk
].
1813 info
->component_size
= ddf
->dlist
->size
- info
->data_offset
;
1815 info
->disk
.number
= -1;
1816 info
->disk
.raid_disk
= -1;
1817 // info->disk.raid_disk = find refnum in the table and use index;
1819 info
->disk
.state
= (1 << MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
);
1821 info
->recovery_start
= MaxSector
;
1822 info
->reshape_active
= 0;
1823 info
->recovery_blocked
= 0;
1826 info
->array
.major_version
= -1;
1827 info
->array
.minor_version
= -2;
1828 strcpy(info
->text_version
, "ddf");
1829 info
->safe_mode_delay
= 0;
1831 uuid_from_super_ddf(st
, info
->uuid
);
1835 for (i
= 0 ; i
< map_disks
; i
++) {
1836 if (i
< info
->array
.raid_disks
&&
1837 (__be16_to_cpu(ddf
->phys
->entries
[i
].state
) & DDF_Online
) &&
1838 !(__be16_to_cpu(ddf
->phys
->entries
[i
].state
) & DDF_Failed
))
1846 static void getinfo_super_ddf_bvd(struct supertype
*st
, struct mdinfo
*info
, char *map
)
1848 struct ddf_super
*ddf
= st
->sb
;
1849 struct vcl
*vc
= ddf
->currentconf
;
1850 int cd
= ddf
->currentdev
;
1854 int map_disks
= info
->array
.raid_disks
;
1856 struct vd_config
*conf
;
1858 memset(info
, 0, sizeof(*info
));
1859 if (layout_ddf2md(&vc
->conf
, &info
->array
) == -1)
1861 info
->array
.md_minor
= -1;
1862 cptr
= (__u32
*)(vc
->conf
.guid
+ 16);
1863 info
->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
1864 info
->array
.utime
= DECADE
+ __be32_to_cpu(vc
->conf
.timestamp
);
1865 info
->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
1866 info
->custom_array_size
= 0;
1869 n_prim
= __be16_to_cpu(conf
->prim_elmnt_count
);
1870 if (conf
->sec_elmnt_count
> 1 && cd
>= n_prim
) {
1871 int ibvd
= cd
/ n_prim
- 1;
1873 conf
= vc
->other_bvds
[ibvd
];
1876 if (cd
>= 0 && (unsigned)cd
< ddf
->mppe
) {
1878 __be64_to_cpu(LBA_OFFSET(ddf
, &vc
->conf
)[cd
]);
1879 if (vc
->block_sizes
)
1880 info
->component_size
= vc
->block_sizes
[cd
];
1882 info
->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
1885 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
1886 if (dl
->disk
.refnum
== conf
->phys_refnum
[cd
])
1889 info
->disk
.major
= 0;
1890 info
->disk
.minor
= 0;
1891 info
->disk
.state
= 0;
1893 info
->disk
.major
= dl
->major
;
1894 info
->disk
.minor
= dl
->minor
;
1895 info
->disk
.raid_disk
= cd
+ conf
->sec_elmnt_seq
1896 * __be16_to_cpu(conf
->prim_elmnt_count
);
1897 info
->disk
.number
= dl
->pdnum
;
1898 info
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
1901 info
->container_member
= ddf
->currentconf
->vcnum
;
1903 info
->recovery_start
= MaxSector
;
1904 info
->resync_start
= 0;
1905 info
->reshape_active
= 0;
1906 info
->recovery_blocked
= 0;
1907 if (!(ddf
->virt
->entries
[info
->container_member
].state
1908 & DDF_state_inconsistent
) &&
1909 (ddf
->virt
->entries
[info
->container_member
].init_state
1910 & DDF_initstate_mask
)
1912 info
->resync_start
= MaxSector
;
1914 uuid_from_super_ddf(st
, info
->uuid
);
1916 info
->array
.major_version
= -1;
1917 info
->array
.minor_version
= -2;
1918 sprintf(info
->text_version
, "/%s/%d",
1919 st
->container_devnm
,
1920 info
->container_member
);
1921 info
->safe_mode_delay
= 200;
1923 memcpy(info
->name
, ddf
->virt
->entries
[info
->container_member
].name
, 16);
1926 if (info
->name
[j
] == ' ')
1930 for (j
= 0; j
< map_disks
; j
++) {
1932 if (j
< info
->array
.raid_disks
) {
1933 int i
= find_phys(ddf
, vc
->conf
.phys_refnum
[j
]);
1935 (__be16_to_cpu(ddf
->phys
->entries
[i
].state
) & DDF_Online
) &&
1936 !(__be16_to_cpu(ddf
->phys
->entries
[i
].state
) & DDF_Failed
))
1942 static int update_super_ddf(struct supertype
*st
, struct mdinfo
*info
,
1944 char *devname
, int verbose
,
1945 int uuid_set
, char *homehost
)
1947 /* For 'assemble' and 'force' we need to return non-zero if any
1948 * change was made. For others, the return value is ignored.
1949 * Update options are:
1950 * force-one : This device looks a bit old but needs to be included,
1951 * update age info appropriately.
1952 * assemble: clear any 'faulty' flag to allow this device to
1954 * force-array: Array is degraded but being forced, mark it clean
1955 * if that will be needed to assemble it.
1957 * newdev: not used ????
1958 * grow: Array has gained a new device - this is currently for
1960 * resync: mark as dirty so a resync will happen.
1961 * uuid: Change the uuid of the array to match what is given
1962 * homehost: update the recorded homehost
1963 * name: update the name - preserving the homehost
1964 * _reshape_progress: record new reshape_progress position.
1966 * Following are not relevant for this version:
1967 * sparc2.2 : update from old dodgey metadata
1968 * super-minor: change the preferred_minor number
1969 * summaries: update redundant counters.
1972 // struct ddf_super *ddf = st->sb;
1973 // struct vd_config *vd = find_vdcr(ddf, info->container_member);
1974 // struct virtual_entry *ve = find_ve(ddf);
1976 /* we don't need to handle "force-*" or "assemble" as
1977 * there is no need to 'trick' the kernel. We the metadata is
1978 * first updated to activate the array, all the implied modifications
1982 if (strcmp(update
, "grow") == 0) {
1984 } else if (strcmp(update
, "resync") == 0) {
1985 // info->resync_checkpoint = 0;
1986 } else if (strcmp(update
, "homehost") == 0) {
1987 /* homehost is stored in controller->vendor_data,
1988 * or it is when we are the vendor
1990 // if (info->vendor_is_local)
1991 // strcpy(ddf->controller.vendor_data, homehost);
1993 } else if (strcmp(update
, "name") == 0) {
1994 /* name is stored in virtual_entry->name */
1995 // memset(ve->name, ' ', 16);
1996 // strncpy(ve->name, info->name, 16);
1998 } else if (strcmp(update
, "_reshape_progress") == 0) {
1999 /* We don't support reshape yet */
2000 } else if (strcmp(update
, "assemble") == 0 ) {
2001 /* Do nothing, just succeed */
2006 // update_all_csum(ddf);
2011 static void make_header_guid(char *guid
)
2014 /* Create a DDF Header of Virtual Disk GUID */
2016 /* 24 bytes of fiction required.
2017 * first 8 are a 'vendor-id' - "Linux-MD"
2018 * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
2019 * Remaining 8 random number plus timestamp
2021 memcpy(guid
, T10
, sizeof(T10
));
2022 stamp
= __cpu_to_be32(0xdeadbeef);
2023 memcpy(guid
+8, &stamp
, 4);
2024 stamp
= __cpu_to_be32(0);
2025 memcpy(guid
+12, &stamp
, 4);
2026 stamp
= __cpu_to_be32(time(0) - DECADE
);
2027 memcpy(guid
+16, &stamp
, 4);
2029 memcpy(guid
+20, &stamp
, 4);
2032 static unsigned int find_unused_vde(const struct ddf_super
*ddf
)
2035 for (i
= 0; i
< __be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
2036 if (all_ff(ddf
->virt
->entries
[i
].guid
))
2039 return DDF_NOTFOUND
;
2042 static unsigned int find_vde_by_name(const struct ddf_super
*ddf
,
2047 return DDF_NOTFOUND
;
2048 for (i
= 0; i
< __be16_to_cpu(ddf
->virt
->max_vdes
); i
++) {
2049 if (all_ff(ddf
->virt
->entries
[i
].guid
))
2051 if (!strncmp(name
, ddf
->virt
->entries
[i
].name
,
2052 sizeof(ddf
->virt
->entries
[i
].name
)))
2055 return DDF_NOTFOUND
;
2058 static unsigned int find_vde_by_guid(const struct ddf_super
*ddf
,
2062 if (guid
== NULL
|| all_ff(guid
))
2063 return DDF_NOTFOUND
;
2064 for (i
= 0; i
< __be16_to_cpu(ddf
->virt
->max_vdes
); i
++)
2065 if (!memcmp(ddf
->virt
->entries
[i
].guid
, guid
, DDF_GUID_LEN
))
2067 return DDF_NOTFOUND
;
2070 static int init_super_ddf_bvd(struct supertype
*st
,
2071 mdu_array_info_t
*info
,
2072 unsigned long long size
,
2073 char *name
, char *homehost
,
2074 int *uuid
, unsigned long long data_offset
);
2076 static int init_super_ddf(struct supertype
*st
,
2077 mdu_array_info_t
*info
,
2078 unsigned long long size
, char *name
, char *homehost
,
2079 int *uuid
, unsigned long long data_offset
)
2081 /* This is primarily called by Create when creating a new array.
2082 * We will then get add_to_super called for each component, and then
2083 * write_init_super called to write it out to each device.
2084 * For DDF, Create can create on fresh devices or on a pre-existing
2086 * To create on a pre-existing array a different method will be called.
2087 * This one is just for fresh drives.
2089 * We need to create the entire 'ddf' structure which includes:
2090 * DDF headers - these are easy.
2091 * Controller data - a Sector describing this controller .. not that
2092 * this is a controller exactly.
2093 * Physical Disk Record - one entry per device, so
2094 * leave plenty of space.
2095 * Virtual Disk Records - again, just leave plenty of space.
2096 * This just lists VDs, doesn't give details
2097 * Config records - describes the VDs that use this disk
2098 * DiskData - describes 'this' device.
2099 * BadBlockManagement - empty
2100 * Diag Space - empty
2101 * Vendor Logs - Could we put bitmaps here?
2104 struct ddf_super
*ddf
;
2107 int max_phys_disks
, max_virt_disks
;
2108 unsigned long long sector
;
2112 struct phys_disk
*pd
;
2113 struct virtual_disk
*vd
;
2115 if (data_offset
!= INVALID_SECTORS
) {
2116 pr_err("data-offset not supported by DDF\n");
2121 return init_super_ddf_bvd(st
, info
, size
, name
, homehost
, uuid
,
2124 if (posix_memalign((void**)&ddf
, 512, sizeof(*ddf
)) != 0) {
2125 pr_err("%s could not allocate superblock\n", __func__
);
2128 memset(ddf
, 0, sizeof(*ddf
));
2129 ddf
->dlist
= NULL
; /* no physical disks yet */
2130 ddf
->conflist
= NULL
; /* No virtual disks yet */
2134 /* zeroing superblock */
2138 /* At least 32MB *must* be reserved for the ddf. So let's just
2139 * start 32MB from the end, and put the primary header there.
2140 * Don't do secondary for now.
2141 * We don't know exactly where that will be yet as it could be
2142 * different on each device. To just set up the lengths.
2146 ddf
->anchor
.magic
= DDF_HEADER_MAGIC
;
2147 make_header_guid(ddf
->anchor
.guid
);
2149 memcpy(ddf
->anchor
.revision
, DDF_REVISION_2
, 8);
2150 ddf
->anchor
.seq
= __cpu_to_be32(1);
2151 ddf
->anchor
.timestamp
= __cpu_to_be32(time(0) - DECADE
);
2152 ddf
->anchor
.openflag
= 0xFF;
2153 ddf
->anchor
.foreignflag
= 0;
2154 ddf
->anchor
.enforcegroups
= 0; /* Is this best?? */
2155 ddf
->anchor
.pad0
= 0xff;
2156 memset(ddf
->anchor
.pad1
, 0xff, 12);
2157 memset(ddf
->anchor
.header_ext
, 0xff, 32);
2158 ddf
->anchor
.primary_lba
= ~(__u64
)0;
2159 ddf
->anchor
.secondary_lba
= ~(__u64
)0;
2160 ddf
->anchor
.type
= DDF_HEADER_ANCHOR
;
2161 memset(ddf
->anchor
.pad2
, 0xff, 3);
2162 ddf
->anchor
.workspace_len
= __cpu_to_be32(32768); /* Must be reserved */
2163 ddf
->anchor
.workspace_lba
= ~(__u64
)0; /* Put this at bottom
2164 of 32M reserved.. */
2165 max_phys_disks
= 1023; /* Should be enough */
2166 ddf
->anchor
.max_pd_entries
= __cpu_to_be16(max_phys_disks
);
2167 max_virt_disks
= 255;
2168 ddf
->anchor
.max_vd_entries
= __cpu_to_be16(max_virt_disks
); /* ?? */
2169 ddf
->anchor
.max_partitions
= __cpu_to_be16(64); /* ?? */
2172 ddf
->conf_rec_len
= 1 + ROUND_UP(ddf
->mppe
* (4+8), 512)/512;
2173 ddf
->anchor
.config_record_len
= __cpu_to_be16(ddf
->conf_rec_len
);
2174 ddf
->anchor
.max_primary_element_entries
= __cpu_to_be16(ddf
->mppe
);
2175 memset(ddf
->anchor
.pad3
, 0xff, 54);
2176 /* controller sections is one sector long immediately
2177 * after the ddf header */
2179 ddf
->anchor
.controller_section_offset
= __cpu_to_be32(sector
);
2180 ddf
->anchor
.controller_section_length
= __cpu_to_be32(1);
2183 /* phys is 8 sectors after that */
2184 pdsize
= ROUND_UP(sizeof(struct phys_disk
) +
2185 sizeof(struct phys_disk_entry
)*max_phys_disks
,
2187 switch(pdsize
/512) {
2188 case 2: case 8: case 32: case 128: case 512: break;
2191 ddf
->anchor
.phys_section_offset
= __cpu_to_be32(sector
);
2192 ddf
->anchor
.phys_section_length
=
2193 __cpu_to_be32(pdsize
/512); /* max_primary_element_entries/8 */
2194 sector
+= pdsize
/512;
2196 /* virt is another 32 sectors */
2197 vdsize
= ROUND_UP(sizeof(struct virtual_disk
) +
2198 sizeof(struct virtual_entry
) * max_virt_disks
,
2200 switch(vdsize
/512) {
2201 case 2: case 8: case 32: case 128: case 512: break;
2204 ddf
->anchor
.virt_section_offset
= __cpu_to_be32(sector
);
2205 ddf
->anchor
.virt_section_length
=
2206 __cpu_to_be32(vdsize
/512); /* max_vd_entries/8 */
2207 sector
+= vdsize
/512;
2209 clen
= ddf
->conf_rec_len
* (ddf
->max_part
+1);
2210 ddf
->anchor
.config_section_offset
= __cpu_to_be32(sector
);
2211 ddf
->anchor
.config_section_length
= __cpu_to_be32(clen
);
2214 ddf
->anchor
.data_section_offset
= __cpu_to_be32(sector
);
2215 ddf
->anchor
.data_section_length
= __cpu_to_be32(1);
2218 ddf
->anchor
.bbm_section_length
= __cpu_to_be32(0);
2219 ddf
->anchor
.bbm_section_offset
= __cpu_to_be32(0xFFFFFFFF);
2220 ddf
->anchor
.diag_space_length
= __cpu_to_be32(0);
2221 ddf
->anchor
.diag_space_offset
= __cpu_to_be32(0xFFFFFFFF);
2222 ddf
->anchor
.vendor_length
= __cpu_to_be32(0);
2223 ddf
->anchor
.vendor_offset
= __cpu_to_be32(0xFFFFFFFF);
2225 memset(ddf
->anchor
.pad4
, 0xff, 256);
2227 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2228 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2230 ddf
->primary
.openflag
= 1; /* I guess.. */
2231 ddf
->primary
.type
= DDF_HEADER_PRIMARY
;
2233 ddf
->secondary
.openflag
= 1; /* I guess.. */
2234 ddf
->secondary
.type
= DDF_HEADER_SECONDARY
;
2236 ddf
->active
= &ddf
->primary
;
2238 ddf
->controller
.magic
= DDF_CONTROLLER_MAGIC
;
2240 /* 24 more bytes of fiction required.
2241 * first 8 are a 'vendor-id' - "Linux-MD"
2242 * Remaining 16 are serial number.... maybe a hostname would do?
2244 memcpy(ddf
->controller
.guid
, T10
, sizeof(T10
));
2245 gethostname(hostname
, sizeof(hostname
));
2246 hostname
[sizeof(hostname
) - 1] = 0;
2247 hostlen
= strlen(hostname
);
2248 memcpy(ddf
->controller
.guid
+ 24 - hostlen
, hostname
, hostlen
);
2249 for (i
= strlen(T10
) ; i
+hostlen
< 24; i
++)
2250 ddf
->controller
.guid
[i
] = ' ';
2252 ddf
->controller
.type
.vendor_id
= __cpu_to_be16(0xDEAD);
2253 ddf
->controller
.type
.device_id
= __cpu_to_be16(0xBEEF);
2254 ddf
->controller
.type
.sub_vendor_id
= 0;
2255 ddf
->controller
.type
.sub_device_id
= 0;
2256 memcpy(ddf
->controller
.product_id
, "What Is My PID??", 16);
2257 memset(ddf
->controller
.pad
, 0xff, 8);
2258 memset(ddf
->controller
.vendor_data
, 0xff, 448);
2259 if (homehost
&& strlen(homehost
) < 440)
2260 strcpy((char*)ddf
->controller
.vendor_data
, homehost
);
2262 if (posix_memalign((void**)&pd
, 512, pdsize
) != 0) {
2263 pr_err("%s could not allocate pd\n", __func__
);
2267 ddf
->pdsize
= pdsize
;
2269 memset(pd
, 0xff, pdsize
);
2270 memset(pd
, 0, sizeof(*pd
));
2271 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2272 pd
->used_pdes
= __cpu_to_be16(0);
2273 pd
->max_pdes
= __cpu_to_be16(max_phys_disks
);
2274 memset(pd
->pad
, 0xff, 52);
2275 for (i
= 0; i
< max_phys_disks
; i
++)
2276 memset(pd
->entries
[i
].guid
, 0xff, DDF_GUID_LEN
);
2278 if (posix_memalign((void**)&vd
, 512, vdsize
) != 0) {
2279 pr_err("%s could not allocate vd\n", __func__
);
2283 ddf
->vdsize
= vdsize
;
2284 memset(vd
, 0, vdsize
);
2285 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
2286 vd
->populated_vdes
= __cpu_to_be16(0);
2287 vd
->max_vdes
= __cpu_to_be16(max_virt_disks
);
2288 memset(vd
->pad
, 0xff, 52);
2290 for (i
=0; i
<max_virt_disks
; i
++)
2291 memset(&vd
->entries
[i
], 0xff, sizeof(struct virtual_entry
));
2294 ddf_set_updates_pending(ddf
);
2298 static int chunk_to_shift(int chunksize
)
2300 return ffs(chunksize
/512)-1;
2305 unsigned long long start
, size
;
2307 static int cmp_extent(const void *av
, const void *bv
)
2309 const struct extent
*a
= av
;
2310 const struct extent
*b
= bv
;
2311 if (a
->start
< b
->start
)
2313 if (a
->start
> b
->start
)
2318 static struct extent
*get_extents(struct ddf_super
*ddf
, struct dl
*dl
)
2320 /* find a list of used extents on the give physical device
2321 * (dnum) of the given ddf.
2322 * Return a malloced array of 'struct extent'
2324 * FIXME ignore DDF_Legacy devices?
2331 rv
= xmalloc(sizeof(struct extent
) * (ddf
->max_part
+ 2));
2333 for (i
= 0; i
< ddf
->max_part
; i
++) {
2334 const struct vd_config
*bvd
;
2336 struct vcl
*v
= dl
->vlist
[i
];
2338 get_pd_index_from_refnum(v
, dl
->disk
.refnum
, ddf
->mppe
,
2339 &bvd
, &ibvd
) == DDF_NOTFOUND
)
2341 rv
[n
].start
= __be64_to_cpu(LBA_OFFSET(ddf
, bvd
)[ibvd
]);
2342 rv
[n
].size
= __be64_to_cpu(bvd
->blocks
);
2345 qsort(rv
, n
, sizeof(*rv
), cmp_extent
);
2347 rv
[n
].start
= __be64_to_cpu(ddf
->phys
->entries
[dl
->pdnum
].config_size
);
2353 static int init_super_ddf_bvd(struct supertype
*st
,
2354 mdu_array_info_t
*info
,
2355 unsigned long long size
,
2356 char *name
, char *homehost
,
2357 int *uuid
, unsigned long long data_offset
)
2359 /* We are creating a BVD inside a pre-existing container.
2360 * so st->sb is already set.
2361 * We need to create a new vd_config and a new virtual_entry
2363 struct ddf_super
*ddf
= st
->sb
;
2364 unsigned int venum
, i
;
2365 struct virtual_entry
*ve
;
2367 struct vd_config
*vc
;
2369 if (find_vde_by_name(ddf
, name
) != DDF_NOTFOUND
) {
2370 pr_err("This ddf already has an array called %s\n", name
);
2373 venum
= find_unused_vde(ddf
);
2374 if (venum
== DDF_NOTFOUND
) {
2375 pr_err("Cannot find spare slot for virtual disk\n");
2378 ve
= &ddf
->virt
->entries
[venum
];
2380 /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
2381 * timestamp, random number
2383 make_header_guid(ve
->guid
);
2384 ve
->unit
= __cpu_to_be16(info
->md_minor
);
2386 ve
->guid_crc
= crc32(0, (unsigned char*)ddf
->anchor
.guid
, DDF_GUID_LEN
);
2388 ve
->state
= DDF_state_degraded
; /* Will be modified as devices are added */
2389 if (info
->state
& 1) /* clean */
2390 ve
->init_state
= DDF_init_full
;
2392 ve
->init_state
= DDF_init_not
;
2394 memset(ve
->pad1
, 0xff, 14);
2395 memset(ve
->name
, ' ', 16);
2397 strncpy(ve
->name
, name
, 16);
2398 ddf
->virt
->populated_vdes
=
2399 __cpu_to_be16(__be16_to_cpu(ddf
->virt
->populated_vdes
)+1);
2401 /* Now create a new vd_config */
2402 if (posix_memalign((void**)&vcl
, 512,
2403 (offsetof(struct vcl
, conf
) + ddf
->conf_rec_len
* 512)) != 0) {
2404 pr_err("%s could not allocate vd_config\n", __func__
);
2408 vcl
->block_sizes
= NULL
; /* FIXME not for CONCAT */
2411 vc
->magic
= DDF_VD_CONF_MAGIC
;
2412 memcpy(vc
->guid
, ve
->guid
, DDF_GUID_LEN
);
2413 vc
->timestamp
= __cpu_to_be32(time(0)-DECADE
);
2414 vc
->seqnum
= __cpu_to_be32(1);
2415 memset(vc
->pad0
, 0xff, 24);
2416 vc
->chunk_shift
= chunk_to_shift(info
->chunk_size
);
2417 if (layout_md2ddf(info
, vc
) == -1 ||
2418 __be16_to_cpu(vc
->prim_elmnt_count
) > ddf
->mppe
) {
2419 pr_err("%s: unsupported RAID level/layout %d/%d with %d disks\n",
2420 __func__
, info
->level
, info
->layout
, info
->raid_disks
);
2424 vc
->sec_elmnt_seq
= 0;
2425 if (alloc_other_bvds(ddf
, vcl
) != 0) {
2426 pr_err("%s could not allocate other bvds\n",
2431 vc
->blocks
= __cpu_to_be64(info
->size
* 2);
2432 vc
->array_blocks
= __cpu_to_be64(
2433 calc_array_size(info
->level
, info
->raid_disks
, info
->layout
,
2434 info
->chunk_size
, info
->size
*2));
2435 memset(vc
->pad1
, 0xff, 8);
2436 vc
->spare_refs
[0] = 0xffffffff;
2437 vc
->spare_refs
[1] = 0xffffffff;
2438 vc
->spare_refs
[2] = 0xffffffff;
2439 vc
->spare_refs
[3] = 0xffffffff;
2440 vc
->spare_refs
[4] = 0xffffffff;
2441 vc
->spare_refs
[5] = 0xffffffff;
2442 vc
->spare_refs
[6] = 0xffffffff;
2443 vc
->spare_refs
[7] = 0xffffffff;
2444 memset(vc
->cache_pol
, 0, 8);
2446 memset(vc
->pad2
, 0xff, 3);
2447 memset(vc
->pad3
, 0xff, 52);
2448 memset(vc
->pad4
, 0xff, 192);
2449 memset(vc
->v0
, 0xff, 32);
2450 memset(vc
->v1
, 0xff, 32);
2451 memset(vc
->v2
, 0xff, 16);
2452 memset(vc
->v3
, 0xff, 16);
2453 memset(vc
->vendor
, 0xff, 32);
2455 memset(vc
->phys_refnum
, 0xff, 4*ddf
->mppe
);
2456 memset(vc
->phys_refnum
+ddf
->mppe
, 0x00, 8*ddf
->mppe
);
2458 for (i
= 1; i
< vc
->sec_elmnt_count
; i
++) {
2459 memcpy(vcl
->other_bvds
[i
-1], vc
, ddf
->conf_rec_len
* 512);
2460 vcl
->other_bvds
[i
-1]->sec_elmnt_seq
= i
;
2463 vcl
->next
= ddf
->conflist
;
2464 ddf
->conflist
= vcl
;
2465 ddf
->currentconf
= vcl
;
2466 ddf_set_updates_pending(ddf
);
2470 static int get_svd_state(const struct ddf_super
*, const struct vcl
*);
2473 static void add_to_super_ddf_bvd(struct supertype
*st
,
2474 mdu_disk_info_t
*dk
, int fd
, char *devname
)
2476 /* fd and devname identify a device with-in the ddf container (st).
2477 * dk identifies a location in the new BVD.
2478 * We need to find suitable free space in that device and update
2479 * the phys_refnum and lba_offset for the newly created vd_config.
2480 * We might also want to update the type in the phys_disk
2483 * Alternately: fd == -1 and we have already chosen which device to
2484 * use and recorded in dlist->raid_disk;
2487 struct ddf_super
*ddf
= st
->sb
;
2488 struct vd_config
*vc
;
2490 unsigned long long blocks
, pos
, esize
;
2492 unsigned int raid_disk
= dk
->raid_disk
;
2495 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2496 if (dl
->raiddisk
== dk
->raid_disk
)
2499 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2500 if (dl
->major
== dk
->major
&&
2501 dl
->minor
== dk
->minor
)
2504 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
2507 vc
= &ddf
->currentconf
->conf
;
2508 if (vc
->sec_elmnt_count
> 1) {
2509 unsigned int n
= __be16_to_cpu(vc
->prim_elmnt_count
);
2511 vc
= ddf
->currentconf
->other_bvds
[raid_disk
/ n
- 1];
2515 ex
= get_extents(ddf
, dl
);
2520 blocks
= __be64_to_cpu(vc
->blocks
);
2521 if (ddf
->currentconf
->block_sizes
)
2522 blocks
= ddf
->currentconf
->block_sizes
[dk
->raid_disk
];
2525 esize
= ex
[i
].start
- pos
;
2526 if (esize
>= blocks
)
2528 pos
= ex
[i
].start
+ ex
[i
].size
;
2530 } while (ex
[i
-1].size
);
2536 ddf
->currentdev
= dk
->raid_disk
;
2537 vc
->phys_refnum
[raid_disk
] = dl
->disk
.refnum
;
2538 LBA_OFFSET(ddf
, vc
)[raid_disk
] = __cpu_to_be64(pos
);
2540 for (i
= 0; i
< ddf
->max_part
; i
++)
2541 if (dl
->vlist
[i
] == NULL
)
2543 if (i
== ddf
->max_part
)
2545 dl
->vlist
[i
] = ddf
->currentconf
;
2550 dl
->devname
= devname
;
2552 /* Check if we can mark array as optimal yet */
2553 i
= ddf
->currentconf
->vcnum
;
2554 ddf
->virt
->entries
[i
].state
=
2555 (ddf
->virt
->entries
[i
].state
& ~DDF_state_mask
)
2556 | get_svd_state(ddf
, ddf
->currentconf
);
2557 ddf
->phys
->entries
[dl
->pdnum
].type
&= ~__cpu_to_be16(DDF_Global_Spare
);
2558 ddf
->phys
->entries
[dl
->pdnum
].type
|= __cpu_to_be16(DDF_Active_in_VD
);
2559 dprintf("%s: added disk %d/%08x to VD %d/%s as disk %d\n",
2560 __func__
, dl
->pdnum
, __be32_to_cpu(dl
->disk
.refnum
),
2561 ddf
->currentconf
->vcnum
, guid_str(vc
->guid
),
2563 ddf_set_updates_pending(ddf
);
2566 static unsigned int find_unused_pde(const struct ddf_super
*ddf
)
2569 for (i
= 0; i
< __be16_to_cpu(ddf
->phys
->max_pdes
); i
++) {
2570 if (all_ff(ddf
->phys
->entries
[i
].guid
))
2573 return DDF_NOTFOUND
;
2576 /* add a device to a container, either while creating it or while
2577 * expanding a pre-existing container
2579 static int add_to_super_ddf(struct supertype
*st
,
2580 mdu_disk_info_t
*dk
, int fd
, char *devname
,
2581 unsigned long long data_offset
)
2583 struct ddf_super
*ddf
= st
->sb
;
2587 unsigned long long size
;
2588 struct phys_disk_entry
*pde
;
2593 if (ddf
->currentconf
) {
2594 add_to_super_ddf_bvd(st
, dk
, fd
, devname
);
2598 /* This is device numbered dk->number. We need to create
2599 * a phys_disk entry and a more detailed disk_data entry.
2602 n
= find_unused_pde(ddf
);
2603 if (n
== DDF_NOTFOUND
) {
2604 pr_err("%s: No free slot in array, cannot add disk\n",
2608 pde
= &ddf
->phys
->entries
[n
];
2609 get_dev_size(fd
, NULL
, &size
);
2610 if (size
<= 32*1024*1024) {
2611 pr_err("%s: device size must be at least 32MB\n",
2617 if (posix_memalign((void**)&dd
, 512,
2618 sizeof(*dd
) + sizeof(dd
->vlist
[0]) * ddf
->max_part
) != 0) {
2619 pr_err("%s could allocate buffer for new disk, aborting\n",
2623 dd
->major
= major(stb
.st_rdev
);
2624 dd
->minor
= minor(stb
.st_rdev
);
2625 dd
->devname
= devname
;
2629 dd
->disk
.magic
= DDF_PHYS_DATA_MAGIC
;
2631 tm
= localtime(&now
);
2632 sprintf(dd
->disk
.guid
, "%8s%04d%02d%02d",
2633 T10
, tm
->tm_year
+1900, tm
->tm_mon
+1, tm
->tm_mday
);
2634 tptr
= (__u32
*)(dd
->disk
.guid
+ 16);
2635 *tptr
++ = random32();
2639 /* Cannot be bothered finding a CRC of some irrelevant details*/
2640 dd
->disk
.refnum
= random32();
2641 for (i
= __be16_to_cpu(ddf
->active
->max_pd_entries
);
2643 if (ddf
->phys
->entries
[i
-1].refnum
== dd
->disk
.refnum
)
2647 dd
->disk
.forced_ref
= 1;
2648 dd
->disk
.forced_guid
= 1;
2649 memset(dd
->disk
.vendor
, ' ', 32);
2650 memcpy(dd
->disk
.vendor
, "Linux", 5);
2651 memset(dd
->disk
.pad
, 0xff, 442);
2652 for (i
= 0; i
< ddf
->max_part
; i
++)
2653 dd
->vlist
[i
] = NULL
;
2657 if (st
->update_tail
) {
2658 int len
= (sizeof(struct phys_disk
) +
2659 sizeof(struct phys_disk_entry
));
2660 struct phys_disk
*pd
;
2663 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2664 pd
->used_pdes
= __cpu_to_be16(n
);
2665 pde
= &pd
->entries
[0];
2668 ddf
->phys
->used_pdes
= __cpu_to_be16(
2669 1 + __be16_to_cpu(ddf
->phys
->used_pdes
));
2671 memcpy(pde
->guid
, dd
->disk
.guid
, DDF_GUID_LEN
);
2672 pde
->refnum
= dd
->disk
.refnum
;
2673 pde
->type
= __cpu_to_be16(DDF_Forced_PD_GUID
| DDF_Global_Spare
);
2674 pde
->state
= __cpu_to_be16(DDF_Online
);
2677 * If there is already a device in dlist, try to reserve the same
2678 * amount of workspace. Otherwise, use 32MB.
2679 * We checked disk size above already.
2681 #define __calc_lba(new, old, lba, mb) do { \
2682 unsigned long long dif; \
2683 if ((old) != NULL) \
2684 dif = (old)->size - __be64_to_cpu((old)->lba); \
2686 dif = (new)->size; \
2687 if ((new)->size > dif) \
2688 (new)->lba = __cpu_to_be64((new)->size - dif); \
2690 (new)->lba = __cpu_to_be64((new)->size - (mb*1024*2)); \
2692 __calc_lba(dd
, ddf
->dlist
, workspace_lba
, 32);
2693 __calc_lba(dd
, ddf
->dlist
, primary_lba
, 16);
2694 __calc_lba(dd
, ddf
->dlist
, secondary_lba
, 32);
2695 pde
->config_size
= dd
->workspace_lba
;
2697 sprintf(pde
->path
, "%17.17s","Information: nil") ;
2698 memset(pde
->pad
, 0xff, 6);
2700 if (st
->update_tail
) {
2701 dd
->next
= ddf
->add_list
;
2704 dd
->next
= ddf
->dlist
;
2706 ddf_set_updates_pending(ddf
);
2712 static int remove_from_super_ddf(struct supertype
*st
, mdu_disk_info_t
*dk
)
2714 struct ddf_super
*ddf
= st
->sb
;
2717 /* mdmon has noticed that this disk (dk->major/dk->minor) has
2718 * disappeared from the container.
2719 * We need to arrange that it disappears from the metadata and
2720 * internal data structures too.
2721 * Most of the work is done by ddf_process_update which edits
2722 * the metadata and closes the file handle and attaches the memory
2723 * where free_updates will free it.
2725 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
2726 if (dl
->major
== dk
->major
&&
2727 dl
->minor
== dk
->minor
)
2732 if (st
->update_tail
) {
2733 int len
= (sizeof(struct phys_disk
) +
2734 sizeof(struct phys_disk_entry
));
2735 struct phys_disk
*pd
;
2738 pd
->magic
= DDF_PHYS_RECORDS_MAGIC
;
2739 pd
->used_pdes
= __cpu_to_be16(dl
->pdnum
);
2740 pd
->entries
[0].state
= __cpu_to_be16(DDF_Missing
);
2741 append_metadata_update(st
, pd
, len
);
2747 * This is the write_init_super method for a ddf container. It is
2748 * called when creating a container or adding another device to a
2751 #define NULL_CONF_SZ 4096
2753 static int __write_ddf_structure(struct dl
*d
, struct ddf_super
*ddf
, __u8 type
,
2756 unsigned long long sector
;
2757 struct ddf_header
*header
;
2758 int fd
, i
, n_config
, conf_size
;
2764 case DDF_HEADER_PRIMARY
:
2765 header
= &ddf
->primary
;
2766 sector
= __be64_to_cpu(header
->primary_lba
);
2768 case DDF_HEADER_SECONDARY
:
2769 header
= &ddf
->secondary
;
2770 sector
= __be64_to_cpu(header
->secondary_lba
);
2776 header
->type
= type
;
2777 header
->openflag
= 1;
2778 header
->crc
= calc_crc(header
, 512);
2780 lseek64(fd
, sector
<<9, 0);
2781 if (write(fd
, header
, 512) < 0)
2784 ddf
->controller
.crc
= calc_crc(&ddf
->controller
, 512);
2785 if (write(fd
, &ddf
->controller
, 512) < 0)
2788 ddf
->phys
->crc
= calc_crc(ddf
->phys
, ddf
->pdsize
);
2789 if (write(fd
, ddf
->phys
, ddf
->pdsize
) < 0)
2791 ddf
->virt
->crc
= calc_crc(ddf
->virt
, ddf
->vdsize
);
2792 if (write(fd
, ddf
->virt
, ddf
->vdsize
) < 0)
2795 /* Now write lots of config records. */
2796 n_config
= ddf
->max_part
;
2797 conf_size
= ddf
->conf_rec_len
* 512;
2798 for (i
= 0 ; i
<= n_config
; i
++) {
2800 struct vd_config
*vdc
= NULL
;
2801 if (i
== n_config
) {
2802 c
= (struct vcl
*)d
->spare
;
2809 get_pd_index_from_refnum(
2812 (const struct vd_config
**)&vdc
,
2816 dprintf("writing conf record %i on disk %08x for %s/%u\n",
2817 i
, __be32_to_cpu(d
->disk
.refnum
),
2818 guid_str(vdc
->guid
),
2819 vdc
->sec_elmnt_seq
);
2820 vdc
->seqnum
= header
->seq
;
2821 vdc
->crc
= calc_crc(vdc
, conf_size
);
2822 if (write(fd
, vdc
, conf_size
) < 0)
2825 unsigned int togo
= conf_size
;
2826 while (togo
> NULL_CONF_SZ
) {
2827 if (write(fd
, null_aligned
, NULL_CONF_SZ
) < 0)
2829 togo
-= NULL_CONF_SZ
;
2831 if (write(fd
, null_aligned
, togo
) < 0)
2838 d
->disk
.crc
= calc_crc(&d
->disk
, 512);
2839 if (write(fd
, &d
->disk
, 512) < 0)
2844 header
->openflag
= 0;
2845 header
->crc
= calc_crc(header
, 512);
2847 lseek64(fd
, sector
<<9, 0);
2848 if (write(fd
, header
, 512) < 0)
2854 static int __write_init_super_ddf(struct supertype
*st
)
2856 struct ddf_super
*ddf
= st
->sb
;
2860 unsigned long long size
;
2864 pr_state(ddf
, __func__
);
2865 if (posix_memalign((void**)&null_aligned
, 4096, NULL_CONF_SZ
) != 0) {
2868 memset(null_aligned
, 0xff, NULL_CONF_SZ
);
2870 seq
= ddf
->active
->seq
+ 1;
2872 /* try to write updated metadata,
2873 * if we catch a failure move on to the next disk
2875 for (d
= ddf
->dlist
; d
; d
=d
->next
) {
2882 /* We need to fill in the primary, (secondary) and workspace
2883 * lba's in the headers, set their checksums,
2884 * Also checksum phys, virt....
2886 * Then write everything out, finally the anchor is written.
2888 get_dev_size(fd
, NULL
, &size
);
2890 if (d
->workspace_lba
!= 0)
2891 ddf
->anchor
.workspace_lba
= d
->workspace_lba
;
2893 ddf
->anchor
.workspace_lba
=
2894 __cpu_to_be64(size
- 32*1024*2);
2895 if (d
->primary_lba
!= 0)
2896 ddf
->anchor
.primary_lba
= d
->primary_lba
;
2898 ddf
->anchor
.primary_lba
=
2899 __cpu_to_be64(size
- 16*1024*2);
2900 if (d
->secondary_lba
!= 0)
2901 ddf
->anchor
.secondary_lba
= d
->secondary_lba
;
2903 ddf
->anchor
.secondary_lba
=
2904 __cpu_to_be64(size
- 32*1024*2);
2905 ddf
->anchor
.seq
= seq
;
2906 memcpy(&ddf
->primary
, &ddf
->anchor
, 512);
2907 memcpy(&ddf
->secondary
, &ddf
->anchor
, 512);
2909 ddf
->anchor
.openflag
= 0xFF; /* 'open' means nothing */
2910 ddf
->anchor
.seq
= 0xFFFFFFFF; /* no sequencing in anchor */
2911 ddf
->anchor
.crc
= calc_crc(&ddf
->anchor
, 512);
2913 if (!__write_ddf_structure(d
, ddf
, DDF_HEADER_PRIMARY
,
2917 if (!__write_ddf_structure(d
, ddf
, DDF_HEADER_SECONDARY
,
2921 lseek64(fd
, (size
-1)*512, SEEK_SET
);
2922 if (write(fd
, &ddf
->anchor
, 512) < 0)
2928 return attempts
!= successes
;
2931 static int write_init_super_ddf(struct supertype
*st
)
2933 struct ddf_super
*ddf
= st
->sb
;
2934 struct vcl
*currentconf
= ddf
->currentconf
;
2936 /* we are done with currentconf reset it to point st at the container */
2937 ddf
->currentconf
= NULL
;
2939 if (st
->update_tail
) {
2940 /* queue the virtual_disk and vd_config as metadata updates */
2941 struct virtual_disk
*vd
;
2942 struct vd_config
*vc
;
2946 int len
= (sizeof(struct phys_disk
) +
2947 sizeof(struct phys_disk_entry
));
2949 /* adding a disk to the container. */
2953 append_metadata_update(st
, ddf
->add_list
->mdupdate
, len
);
2954 ddf
->add_list
->mdupdate
= NULL
;
2958 /* Newly created VD */
2960 /* First the virtual disk. We have a slightly fake header */
2961 len
= sizeof(struct virtual_disk
) + sizeof(struct virtual_entry
);
2964 vd
->entries
[0] = ddf
->virt
->entries
[currentconf
->vcnum
];
2965 vd
->populated_vdes
= __cpu_to_be16(currentconf
->vcnum
);
2966 append_metadata_update(st
, vd
, len
);
2968 /* Then the vd_config */
2969 len
= ddf
->conf_rec_len
* 512;
2971 memcpy(vc
, ¤tconf
->conf
, len
);
2972 append_metadata_update(st
, vc
, len
);
2974 /* FIXME I need to close the fds! */
2979 for (d
= ddf
->dlist
; d
; d
=d
->next
)
2980 while (Kill(d
->devname
, NULL
, 0, -1, 1) == 0);
2981 return __write_init_super_ddf(st
);
2987 static __u64
avail_size_ddf(struct supertype
*st
, __u64 devsize
,
2988 unsigned long long data_offset
)
2990 /* We must reserve the last 32Meg */
2991 if (devsize
<= 32*1024*2)
2993 return devsize
- 32*1024*2;
2998 static int reserve_space(struct supertype
*st
, int raiddisks
,
2999 unsigned long long size
, int chunk
,
3000 unsigned long long *freesize
)
3002 /* Find 'raiddisks' spare extents at least 'size' big (but
3003 * only caring about multiples of 'chunk') and remember
3005 * If the cannot be found, fail.
3008 struct ddf_super
*ddf
= st
->sb
;
3011 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3015 /* Now find largest extent on each device */
3016 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3017 struct extent
*e
= get_extents(ddf
, dl
);
3018 unsigned long long pos
= 0;
3021 unsigned long long minsize
= size
;
3029 unsigned long long esize
;
3030 esize
= e
[i
].start
- pos
;
3031 if (esize
>= minsize
) {
3035 pos
= e
[i
].start
+ e
[i
].size
;
3037 } while (e
[i
-1].size
);
3040 dl
->esize
= minsize
;
3044 if (cnt
< raiddisks
) {
3045 pr_err("not enough devices with space to create array.\n");
3046 return 0; /* No enough free spaces large enough */
3049 /* choose the largest size of which there are at least 'raiddisk' */
3050 for (dl
= ddf
->dlist
; dl
; dl
=dl
->next
) {
3052 if (dl
->esize
<= size
)
3054 /* This is bigger than 'size', see if there are enough */
3056 for (dl2
= ddf
->dlist
; dl2
; dl2
=dl2
->next
)
3057 if (dl2
->esize
>= dl
->esize
)
3059 if (cnt
>= raiddisks
)
3063 size
= size
/ chunk
;
3068 pr_err("not enough spare devices to create array.\n");
3072 /* We have a 'size' of which there are enough spaces.
3073 * We simply do a first-fit */
3075 for (dl
= ddf
->dlist
; dl
&& cnt
< raiddisks
; dl
=dl
->next
) {
3076 if (dl
->esize
< size
)
3086 validate_geometry_ddf_container(struct supertype
*st
,
3087 int level
, int layout
, int raiddisks
,
3088 int chunk
, unsigned long long size
,
3089 unsigned long long data_offset
,
3090 char *dev
, unsigned long long *freesize
,
3093 static int validate_geometry_ddf_bvd(struct supertype
*st
,
3094 int level
, int layout
, int raiddisks
,
3095 int *chunk
, unsigned long long size
,
3096 unsigned long long data_offset
,
3097 char *dev
, unsigned long long *freesize
,
3100 static int validate_geometry_ddf(struct supertype
*st
,
3101 int level
, int layout
, int raiddisks
,
3102 int *chunk
, unsigned long long size
,
3103 unsigned long long data_offset
,
3104 char *dev
, unsigned long long *freesize
,
3111 /* ddf potentially supports lots of things, but it depends on
3112 * what devices are offered (and maybe kernel version?)
3113 * If given unused devices, we will make a container.
3114 * If given devices in a container, we will make a BVD.
3115 * If given BVDs, we make an SVD, changing all the GUIDs in the process.
3118 if (chunk
&& *chunk
== UnSet
)
3119 *chunk
= DEFAULT_CHUNK
;
3121 if (level
== -1000000) level
= LEVEL_CONTAINER
;
3122 if (level
== LEVEL_CONTAINER
) {
3123 /* Must be a fresh device to add to a container */
3124 return validate_geometry_ddf_container(st
, level
, layout
,
3125 raiddisks
, chunk
?*chunk
:0,
3126 size
, data_offset
, dev
,
3132 mdu_array_info_t array
= {
3133 .level
= level
, .layout
= layout
,
3134 .raid_disks
= raiddisks
3136 struct vd_config conf
;
3137 if (layout_md2ddf(&array
, &conf
) == -1) {
3139 pr_err("DDF does not support level %d /layout %d arrays with %d disks\n",
3140 level
, layout
, raiddisks
);
3143 /* Should check layout? etc */
3145 if (st
->sb
&& freesize
) {
3146 /* --create was given a container to create in.
3147 * So we need to check that there are enough
3148 * free spaces and return the amount of space.
3149 * We may as well remember which drives were
3150 * chosen so that add_to_super/getinfo_super
3153 return reserve_space(st
, raiddisks
, size
, chunk
?*chunk
:0, freesize
);
3159 /* A container has already been opened, so we are
3160 * creating in there. Maybe a BVD, maybe an SVD.
3161 * Should make a distinction one day.
3163 return validate_geometry_ddf_bvd(st
, level
, layout
, raiddisks
,
3164 chunk
, size
, data_offset
, dev
,
3168 /* This is the first device for the array.
3169 * If it is a container, we read it in and do automagic allocations,
3170 * no other devices should be given.
3171 * Otherwise it must be a member device of a container, and we
3172 * do manual allocation.
3173 * Later we should check for a BVD and make an SVD.
3175 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3177 sra
= sysfs_read(fd
, NULL
, GET_VERSION
);
3179 if (sra
&& sra
->array
.major_version
== -1 &&
3180 strcmp(sra
->text_version
, "ddf") == 0) {
3183 /* find space for 'n' devices. */
3184 /* remember the devices */
3185 /* Somehow return the fact that we have enough */
3189 pr_err("ddf: Cannot create this array "
3190 "on device %s - a container is required.\n",
3194 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
3196 pr_err("ddf: Cannot open %s: %s\n",
3197 dev
, strerror(errno
));
3200 /* Well, it is in use by someone, maybe a 'ddf' container. */
3201 cfd
= open_container(fd
);
3205 pr_err("ddf: Cannot use %s: %s\n",
3206 dev
, strerror(EBUSY
));
3209 sra
= sysfs_read(cfd
, NULL
, GET_VERSION
);
3211 if (sra
&& sra
->array
.major_version
== -1 &&
3212 strcmp(sra
->text_version
, "ddf") == 0) {
3213 /* This is a member of a ddf container. Load the container
3214 * and try to create a bvd
3216 struct ddf_super
*ddf
;
3217 if (load_super_ddf_all(st
, cfd
, (void **)&ddf
, NULL
) == 0) {
3219 strcpy(st
->container_devnm
, fd2devnm(cfd
));
3221 return validate_geometry_ddf_bvd(st
, level
, layout
,
3222 raiddisks
, chunk
, size
,
3228 } else /* device may belong to a different container */
3235 validate_geometry_ddf_container(struct supertype
*st
,
3236 int level
, int layout
, int raiddisks
,
3237 int chunk
, unsigned long long size
,
3238 unsigned long long data_offset
,
3239 char *dev
, unsigned long long *freesize
,
3243 unsigned long long ldsize
;
3245 if (level
!= LEVEL_CONTAINER
)
3250 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
3253 pr_err("ddf: Cannot open %s: %s\n",
3254 dev
, strerror(errno
));
3257 if (!get_dev_size(fd
, dev
, &ldsize
)) {
3263 *freesize
= avail_size_ddf(st
, ldsize
>> 9, INVALID_SECTORS
);
3270 static int validate_geometry_ddf_bvd(struct supertype
*st
,
3271 int level
, int layout
, int raiddisks
,
3272 int *chunk
, unsigned long long size
,
3273 unsigned long long data_offset
,
3274 char *dev
, unsigned long long *freesize
,
3278 struct ddf_super
*ddf
= st
->sb
;
3280 unsigned long long pos
= 0;
3281 unsigned long long maxsize
;
3284 /* ddf/bvd supports lots of things, but not containers */
3285 if (level
== LEVEL_CONTAINER
) {
3287 pr_err("DDF cannot create a container within an container\n");
3290 /* We must have the container info already read in. */
3295 /* General test: make sure there is space for
3296 * 'raiddisks' device extents of size 'size'.
3298 unsigned long long minsize
= size
;
3302 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
3308 e
= get_extents(ddf
, dl
);
3311 unsigned long long esize
;
3312 esize
= e
[i
].start
- pos
;
3313 if (esize
>= minsize
)
3315 pos
= e
[i
].start
+ e
[i
].size
;
3317 } while (e
[i
-1].size
);
3322 if (dcnt
< raiddisks
) {
3324 pr_err("ddf: Not enough devices with "
3325 "space for this array (%d < %d)\n",
3331 /* This device must be a member of the set */
3332 if (stat(dev
, &stb
) < 0)
3334 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
3336 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
3337 if (dl
->major
== (int)major(stb
.st_rdev
) &&
3338 dl
->minor
== (int)minor(stb
.st_rdev
))
3343 pr_err("ddf: %s is not in the "
3348 e
= get_extents(ddf
, dl
);
3352 unsigned long long esize
;
3353 esize
= e
[i
].start
- pos
;
3354 if (esize
>= maxsize
)
3356 pos
= e
[i
].start
+ e
[i
].size
;
3358 } while (e
[i
-1].size
);
3359 *freesize
= maxsize
;
3365 static int load_super_ddf_all(struct supertype
*st
, int fd
,
3366 void **sbp
, char *devname
)
3369 struct ddf_super
*super
;
3370 struct mdinfo
*sd
, *best
= NULL
;
3376 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
3379 if (sra
->array
.major_version
!= -1 ||
3380 sra
->array
.minor_version
!= -2 ||
3381 strcmp(sra
->text_version
, "ddf") != 0)
3384 if (posix_memalign((void**)&super
, 512, sizeof(*super
)) != 0)
3386 memset(super
, 0, sizeof(*super
));
3388 /* first, try each device, and choose the best ddf */
3389 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3391 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3392 dfd
= dev_open(nm
, O_RDONLY
);
3395 rv
= load_ddf_headers(dfd
, super
, NULL
);
3398 seq
= __be32_to_cpu(super
->active
->seq
);
3399 if (super
->active
->openflag
)
3401 if (!best
|| seq
> bestseq
) {
3409 /* OK, load this ddf */
3410 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
3411 dfd
= dev_open(nm
, O_RDONLY
);
3414 load_ddf_headers(dfd
, super
, NULL
);
3415 load_ddf_global(dfd
, super
, NULL
);
3417 /* Now we need the device-local bits */
3418 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
3421 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
3422 dfd
= dev_open(nm
, O_RDWR
);
3425 rv
= load_ddf_headers(dfd
, super
, NULL
);
3427 rv
= load_ddf_local(dfd
, super
, NULL
, 1);
3433 if (st
->ss
== NULL
) {
3434 st
->ss
= &super_ddf
;
3435 st
->minor_version
= 0;
3438 strcpy(st
->container_devnm
, fd2devnm(fd
));
3442 static int load_container_ddf(struct supertype
*st
, int fd
,
3445 return load_super_ddf_all(st
, fd
, &st
->sb
, devname
);
3448 #endif /* MDASSEMBLE */
3450 static int check_secondary(const struct vcl
*vc
)
3452 const struct vd_config
*conf
= &vc
->conf
;
3455 /* The only DDF secondary RAID level md can support is
3456 * RAID 10, if the stripe sizes and Basic volume sizes
3458 * Other configurations could in theory be supported by exposing
3459 * the BVDs to user space and using device mapper for the secondary
3460 * mapping. So far we don't support that.
3463 __u64 sec_elements
[4] = {0, 0, 0, 0};
3464 #define __set_sec_seen(n) (sec_elements[(n)>>6] |= (1<<((n)&63)))
3465 #define __was_sec_seen(n) ((sec_elements[(n)>>6] & (1<<((n)&63))) != 0)
3467 if (vc
->other_bvds
== NULL
) {
3468 pr_err("No BVDs for secondary RAID found\n");
3471 if (conf
->prl
!= DDF_RAID1
) {
3472 pr_err("Secondary RAID level only supported for mirrored BVD\n");
3475 if (conf
->srl
!= DDF_2STRIPED
&& conf
->srl
!= DDF_2SPANNED
) {
3476 pr_err("Secondary RAID level %d is unsupported\n",
3480 __set_sec_seen(conf
->sec_elmnt_seq
);
3481 for (i
= 0; i
< conf
->sec_elmnt_count
-1; i
++) {
3482 const struct vd_config
*bvd
= vc
->other_bvds
[i
];
3483 if (bvd
->sec_elmnt_seq
== DDF_UNUSED_BVD
)
3485 if (bvd
->srl
!= conf
->srl
) {
3486 pr_err("Inconsistent secondary RAID level across BVDs\n");
3489 if (bvd
->prl
!= conf
->prl
) {
3490 pr_err("Different RAID levels for BVDs are unsupported\n");
3493 if (bvd
->prim_elmnt_count
!= conf
->prim_elmnt_count
) {
3494 pr_err("All BVDs must have the same number of primary elements\n");
3497 if (bvd
->chunk_shift
!= conf
->chunk_shift
) {
3498 pr_err("Different strip sizes for BVDs are unsupported\n");
3501 if (bvd
->array_blocks
!= conf
->array_blocks
) {
3502 pr_err("Different BVD sizes are unsupported\n");
3505 __set_sec_seen(bvd
->sec_elmnt_seq
);
3507 for (i
= 0; i
< conf
->sec_elmnt_count
; i
++) {
3508 if (!__was_sec_seen(i
)) {
3509 pr_err("BVD %d is missing\n", i
);
3516 static unsigned int get_pd_index_from_refnum(const struct vcl
*vc
,
3517 __u32 refnum
, unsigned int nmax
,
3518 const struct vd_config
**bvd
,
3521 unsigned int i
, j
, n
, sec
, cnt
;
3523 cnt
= __be16_to_cpu(vc
->conf
.prim_elmnt_count
);
3524 sec
= (vc
->conf
.sec_elmnt_count
== 1 ? 0 : vc
->conf
.sec_elmnt_seq
);
3526 for (i
= 0, j
= 0 ; i
< nmax
; i
++) {
3527 /* j counts valid entries for this BVD */
3528 if (vc
->conf
.phys_refnum
[i
] != 0xffffffff)
3530 if (vc
->conf
.phys_refnum
[i
] == refnum
) {
3533 return sec
* cnt
+ j
- 1;
3536 if (vc
->other_bvds
== NULL
)
3539 for (n
= 1; n
< vc
->conf
.sec_elmnt_count
; n
++) {
3540 struct vd_config
*vd
= vc
->other_bvds
[n
-1];
3541 sec
= vd
->sec_elmnt_seq
;
3542 if (sec
== DDF_UNUSED_BVD
)
3544 for (i
= 0, j
= 0 ; i
< nmax
; i
++) {
3545 if (vd
->phys_refnum
[i
] != 0xffffffff)
3547 if (vd
->phys_refnum
[i
] == refnum
) {
3550 return sec
* cnt
+ j
- 1;
3556 return DDF_NOTFOUND
;
3559 static struct mdinfo
*container_content_ddf(struct supertype
*st
, char *subarray
)
3561 /* Given a container loaded by load_super_ddf_all,
3562 * extract information about all the arrays into
3565 * For each vcl in conflist: create an mdinfo, fill it in,
3566 * then look for matching devices (phys_refnum) in dlist
3567 * and create appropriate device mdinfo.
3569 struct ddf_super
*ddf
= st
->sb
;
3570 struct mdinfo
*rest
= NULL
;
3573 for (vc
= ddf
->conflist
; vc
; vc
=vc
->next
)
3577 struct mdinfo
*this;
3583 (strtoul(subarray
, &ep
, 10) != vc
->vcnum
||
3587 if (vc
->conf
.sec_elmnt_count
> 1) {
3588 if (check_secondary(vc
) != 0)
3592 this = xcalloc(1, sizeof(*this));
3596 if (layout_ddf2md(&vc
->conf
, &this->array
))
3598 this->array
.md_minor
= -1;
3599 this->array
.major_version
= -1;
3600 this->array
.minor_version
= -2;
3601 cptr
= (__u32
*)(vc
->conf
.guid
+ 16);
3602 this->array
.ctime
= DECADE
+ __be32_to_cpu(*cptr
);
3603 this->array
.utime
= DECADE
+
3604 __be32_to_cpu(vc
->conf
.timestamp
);
3605 this->array
.chunk_size
= 512 << vc
->conf
.chunk_shift
;
3608 if ((ddf
->virt
->entries
[i
].state
& DDF_state_inconsistent
) ||
3609 (ddf
->virt
->entries
[i
].init_state
& DDF_initstate_mask
) !=
3611 this->array
.state
= 0;
3612 this->resync_start
= 0;
3614 this->array
.state
= 1;
3615 this->resync_start
= MaxSector
;
3617 memcpy(this->name
, ddf
->virt
->entries
[i
].name
, 16);
3620 if (this->name
[j
] == ' ')
3623 memset(this->uuid
, 0, sizeof(this->uuid
));
3624 this->component_size
= __be64_to_cpu(vc
->conf
.blocks
);
3625 this->array
.size
= this->component_size
/ 2;
3626 this->container_member
= i
;
3628 ddf
->currentconf
= vc
;
3629 uuid_from_super_ddf(st
, this->uuid
);
3631 ddf
->currentconf
= NULL
;
3633 sprintf(this->text_version
, "/%s/%d",
3634 st
->container_devnm
, this->container_member
);
3636 for (pd
= 0; pd
< __be16_to_cpu(ddf
->phys
->used_pdes
); pd
++) {
3639 const struct vd_config
*bvd
;
3643 if (ddf
->phys
->entries
[pd
].refnum
== 0xFFFFFFFF)
3646 stt
= __be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
3647 if ((stt
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
3651 i
= get_pd_index_from_refnum(
3652 vc
, ddf
->phys
->entries
[pd
].refnum
,
3653 ddf
->mppe
, &bvd
, &iphys
);
3654 if (i
== DDF_NOTFOUND
)
3657 this->array
.working_disks
++;
3659 for (d
= ddf
->dlist
; d
; d
=d
->next
)
3660 if (d
->disk
.refnum
==
3661 ddf
->phys
->entries
[pd
].refnum
)
3664 /* Haven't found that one yet, maybe there are others */
3667 dev
= xcalloc(1, sizeof(*dev
));
3668 dev
->next
= this->devs
;
3671 dev
->disk
.number
= __be32_to_cpu(d
->disk
.refnum
);
3672 dev
->disk
.major
= d
->major
;
3673 dev
->disk
.minor
= d
->minor
;
3674 dev
->disk
.raid_disk
= i
;
3675 dev
->disk
.state
= (1<<MD_DISK_SYNC
)|(1<<MD_DISK_ACTIVE
);
3676 dev
->recovery_start
= MaxSector
;
3678 dev
->events
= __be32_to_cpu(ddf
->primary
.seq
);
3680 __be64_to_cpu(LBA_OFFSET(ddf
, bvd
)[iphys
]);
3681 dev
->component_size
= __be64_to_cpu(bvd
->blocks
);
3683 strcpy(dev
->name
, d
->devname
);
3689 static int store_super_ddf(struct supertype
*st
, int fd
)
3691 struct ddf_super
*ddf
= st
->sb
;
3692 unsigned long long dsize
;
3699 if (!get_dev_size(fd
, NULL
, &dsize
))
3702 if (ddf
->dlist
|| ddf
->conflist
) {
3707 if (fstat(fd
, &sta
) == -1 || !S_ISBLK(sta
.st_mode
)) {
3708 pr_err("%s: file descriptor for invalid device\n",
3712 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
3713 if (dl
->major
== (int)major(sta
.st_rdev
) &&
3714 dl
->minor
== (int)minor(sta
.st_rdev
))
3717 pr_err("%s: couldn't find disk %d/%d\n", __func__
,
3718 (int)major(sta
.st_rdev
),
3719 (int)minor(sta
.st_rdev
));
3723 For DDF, writing to just one disk makes no sense.
3724 We would run the risk of writing inconsistent meta data
3725 to the devices. So just call __write_init_super_ddf and
3726 write to all devices, including this one.
3727 Use the fd passed to this function, just in case dl->fd
3732 ret
= __write_init_super_ddf(st
);
3737 if (posix_memalign(&buf
, 512, 512) != 0)
3739 memset(buf
, 0, 512);
3741 lseek64(fd
, dsize
-512, 0);
3742 rc
= write(fd
, buf
, 512);
3749 static int compare_super_ddf(struct supertype
*st
, struct supertype
*tst
)
3753 * 0 same, or first was empty, and second was copied
3754 * 1 second had wrong number
3756 * 3 wrong other info
3758 struct ddf_super
*first
= st
->sb
;
3759 struct ddf_super
*second
= tst
->sb
;
3760 struct dl
*dl1
, *dl2
;
3761 struct vcl
*vl1
, *vl2
;
3762 unsigned int max_vds
, max_pds
, pd
, vd
;
3770 if (memcmp(first
->anchor
.guid
, second
->anchor
.guid
, DDF_GUID_LEN
) != 0)
3773 if (first
->anchor
.seq
!= second
->anchor
.seq
) {
3774 dprintf("%s: sequence number mismatch %u/%u\n", __func__
,
3775 __be32_to_cpu(first
->anchor
.seq
),
3776 __be32_to_cpu(second
->anchor
.seq
));
3779 if (first
->max_part
!= second
->max_part
||
3780 first
->phys
->used_pdes
!= second
->phys
->used_pdes
||
3781 first
->virt
->populated_vdes
!= second
->virt
->populated_vdes
) {
3782 dprintf("%s: PD/VD number mismatch\n", __func__
);
3786 max_pds
= __be16_to_cpu(first
->phys
->used_pdes
);
3787 for (dl2
= second
->dlist
; dl2
; dl2
= dl2
->next
) {
3788 for (pd
= 0; pd
< max_pds
; pd
++)
3789 if (first
->phys
->entries
[pd
].refnum
== dl2
->disk
.refnum
)
3791 if (pd
== max_pds
) {
3792 dprintf("%s: no match for disk %08x\n", __func__
,
3793 __be32_to_cpu(dl2
->disk
.refnum
));
3798 max_vds
= __be16_to_cpu(first
->active
->max_vd_entries
);
3799 for (vl2
= second
->conflist
; vl2
; vl2
= vl2
->next
) {
3800 if (vl2
->conf
.magic
!= DDF_VD_CONF_MAGIC
)
3802 for (vd
= 0; vd
< max_vds
; vd
++)
3803 if (!memcmp(first
->virt
->entries
[vd
].guid
,
3804 vl2
->conf
.guid
, DDF_GUID_LEN
))
3806 if (vd
== max_vds
) {
3807 dprintf("%s: no match for VD config\n", __func__
);
3811 /* FIXME should I look at anything else? */
3814 At this point we are fairly sure that the meta data matches.
3815 But the new disk may contain additional local data.
3816 Add it to the super block.
3818 for (vl2
= second
->conflist
; vl2
; vl2
= vl2
->next
) {
3819 for (vl1
= first
->conflist
; vl1
; vl1
= vl1
->next
)
3820 if (!memcmp(vl1
->conf
.guid
, vl2
->conf
.guid
,
3824 if (vl1
->other_bvds
!= NULL
&&
3825 vl1
->conf
.sec_elmnt_seq
!=
3826 vl2
->conf
.sec_elmnt_seq
) {
3827 dprintf("%s: adding BVD %u\n", __func__
,
3828 vl2
->conf
.sec_elmnt_seq
);
3829 add_other_bvd(vl1
, &vl2
->conf
,
3830 first
->conf_rec_len
*512);
3835 if (posix_memalign((void **)&vl1
, 512,
3836 (first
->conf_rec_len
*512 +
3837 offsetof(struct vcl
, conf
))) != 0) {
3838 pr_err("%s could not allocate vcl buf\n",
3843 vl1
->next
= first
->conflist
;
3844 vl1
->block_sizes
= NULL
;
3845 memcpy(&vl1
->conf
, &vl2
->conf
, first
->conf_rec_len
*512);
3846 if (alloc_other_bvds(first
, vl1
) != 0) {
3847 pr_err("%s could not allocate other bvds\n",
3852 for (vd
= 0; vd
< max_vds
; vd
++)
3853 if (!memcmp(first
->virt
->entries
[vd
].guid
,
3854 vl1
->conf
.guid
, DDF_GUID_LEN
))
3857 dprintf("%s: added config for VD %u\n", __func__
, vl1
->vcnum
);
3858 first
->conflist
= vl1
;
3861 for (dl2
= second
->dlist
; dl2
; dl2
= dl2
->next
) {
3862 for (dl1
= first
->dlist
; dl1
; dl1
= dl1
->next
)
3863 if (dl1
->disk
.refnum
== dl2
->disk
.refnum
)
3868 if (posix_memalign((void **)&dl1
, 512,
3869 sizeof(*dl1
) + (first
->max_part
) * sizeof(dl1
->vlist
[0]))
3871 pr_err("%s could not allocate disk info buffer\n",
3875 memcpy(dl1
, dl2
, sizeof(*dl1
));
3876 dl1
->mdupdate
= NULL
;
3877 dl1
->next
= first
->dlist
;
3879 for (pd
= 0; pd
< max_pds
; pd
++)
3880 if (first
->phys
->entries
[pd
].refnum
== dl1
->disk
.refnum
)
3884 if (posix_memalign((void **)&dl1
->spare
, 512,
3885 first
->conf_rec_len
*512) != 0) {
3886 pr_err("%s could not allocate spare info buf\n",
3890 memcpy(dl1
->spare
, dl2
->spare
, first
->conf_rec_len
*512);
3892 for (vd
= 0 ; vd
< first
->max_part
; vd
++) {
3893 if (!dl2
->vlist
[vd
]) {
3894 dl1
->vlist
[vd
] = NULL
;
3897 for (vl1
= first
->conflist
; vl1
; vl1
= vl1
->next
) {
3898 if (!memcmp(vl1
->conf
.guid
,
3899 dl2
->vlist
[vd
]->conf
.guid
,
3902 dl1
->vlist
[vd
] = vl1
;
3906 dprintf("%s: added disk %d: %08x\n", __func__
, dl1
->pdnum
,
3907 __be32_to_cpu(dl1
->disk
.refnum
));
3915 * A new array 'a' has been started which claims to be instance 'inst'
3916 * within container 'c'.
3917 * We need to confirm that the array matches the metadata in 'c' so
3918 * that we don't corrupt any metadata.
3920 static int ddf_open_new(struct supertype
*c
, struct active_array
*a
, char *inst
)
3922 struct ddf_super
*ddf
= c
->sb
;
3924 if (all_ff(ddf
->virt
->entries
[n
].guid
)) {
3925 pr_err("%s: subarray %d doesn't exist\n", __func__
, n
);
3928 dprintf("ddf: open_new %d\n", n
);
3929 a
->info
.container_member
= n
;
3934 * The array 'a' is to be marked clean in the metadata.
3935 * If '->resync_start' is not ~(unsigned long long)0, then the array is only
3936 * clean up to the point (in sectors). If that cannot be recorded in the
3937 * metadata, then leave it as dirty.
3939 * For DDF, we need to clear the DDF_state_inconsistent bit in the
3940 * !global! virtual_disk.virtual_entry structure.
3942 static int ddf_set_array_state(struct active_array
*a
, int consistent
)
3944 struct ddf_super
*ddf
= a
->container
->sb
;
3945 int inst
= a
->info
.container_member
;
3946 int old
= ddf
->virt
->entries
[inst
].state
;
3947 if (consistent
== 2) {
3948 /* Should check if a recovery should be started FIXME */
3950 if (!is_resync_complete(&a
->info
))
3954 ddf
->virt
->entries
[inst
].state
&= ~DDF_state_inconsistent
;
3956 ddf
->virt
->entries
[inst
].state
|= DDF_state_inconsistent
;
3957 if (old
!= ddf
->virt
->entries
[inst
].state
)
3958 ddf_set_updates_pending(ddf
);
3960 old
= ddf
->virt
->entries
[inst
].init_state
;
3961 ddf
->virt
->entries
[inst
].init_state
&= ~DDF_initstate_mask
;
3962 if (is_resync_complete(&a
->info
))
3963 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_full
;
3964 else if (a
->info
.resync_start
== 0)
3965 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_not
;
3967 ddf
->virt
->entries
[inst
].init_state
|= DDF_init_quick
;
3968 if (old
!= ddf
->virt
->entries
[inst
].init_state
)
3969 ddf_set_updates_pending(ddf
);
3971 dprintf("ddf mark %d/%s (%d) %s %llu\n", inst
,
3972 guid_str(ddf
->virt
->entries
[inst
].guid
), a
->curr_state
,
3973 consistent
?"clean":"dirty",
3974 a
->info
.resync_start
);
3978 static int get_bvd_state(const struct ddf_super
*ddf
,
3979 const struct vd_config
*vc
)
3981 unsigned int i
, n_bvd
, working
= 0;
3982 unsigned int n_prim
= __be16_to_cpu(vc
->prim_elmnt_count
);
3984 for (i
= 0; i
< n_prim
; i
++) {
3985 if (!find_index_in_bvd(ddf
, vc
, i
, &n_bvd
))
3987 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
3990 st
= __be16_to_cpu(ddf
->phys
->entries
[pd
].state
);
3991 if ((st
& (DDF_Online
|DDF_Failed
|DDF_Rebuilding
))
3996 state
= DDF_state_degraded
;
3997 if (working
== n_prim
)
3998 state
= DDF_state_optimal
;
4004 state
= DDF_state_failed
;
4008 state
= DDF_state_failed
;
4009 else if (working
>= 2)
4010 state
= DDF_state_part_optimal
;
4014 if (working
< n_prim
- 1)
4015 state
= DDF_state_failed
;
4018 if (working
< n_prim
- 2)
4019 state
= DDF_state_failed
;
4020 else if (working
== n_prim
- 1)
4021 state
= DDF_state_part_optimal
;
4027 static int secondary_state(int state
, int other
, int seclevel
)
4029 if (state
== DDF_state_optimal
&& other
== DDF_state_optimal
)
4030 return DDF_state_optimal
;
4031 if (seclevel
== DDF_2MIRRORED
) {
4032 if (state
== DDF_state_optimal
|| other
== DDF_state_optimal
)
4033 return DDF_state_part_optimal
;
4034 if (state
== DDF_state_failed
&& other
== DDF_state_failed
)
4035 return DDF_state_failed
;
4036 return DDF_state_degraded
;
4038 if (state
== DDF_state_failed
|| other
== DDF_state_failed
)
4039 return DDF_state_failed
;
4040 if (state
== DDF_state_degraded
|| other
== DDF_state_degraded
)
4041 return DDF_state_degraded
;
4042 return DDF_state_part_optimal
;
4046 static int get_svd_state(const struct ddf_super
*ddf
, const struct vcl
*vcl
)
4048 int state
= get_bvd_state(ddf
, &vcl
->conf
);
4050 for (i
= 1; i
< vcl
->conf
.sec_elmnt_count
; i
++) {
4051 state
= secondary_state(
4053 get_bvd_state(ddf
, vcl
->other_bvds
[i
-1]),
4060 * The state of each disk is stored in the global phys_disk structure
4061 * in phys_disk.entries[n].state.
4062 * This makes various combinations awkward.
4063 * - When a device fails in any array, it must be failed in all arrays
4064 * that include a part of this device.
4065 * - When a component is rebuilding, we cannot include it officially in the
4066 * array unless this is the only array that uses the device.
4068 * So: when transitioning:
4069 * Online -> failed, just set failed flag. monitor will propagate
4070 * spare -> online, the device might need to be added to the array.
4071 * spare -> failed, just set failed. Don't worry if in array or not.
4073 static void ddf_set_disk(struct active_array
*a
, int n
, int state
)
4075 struct ddf_super
*ddf
= a
->container
->sb
;
4076 unsigned int inst
= a
->info
.container_member
, n_bvd
;
4078 struct vd_config
*vc
= find_vdcr(ddf
, inst
, (unsigned int)n
,
4085 dprintf("ddf: cannot find instance %d!!\n", inst
);
4088 /* Find the matching slot in 'info'. */
4089 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
4090 if (mdi
->disk
.raid_disk
== n
)
4095 /* and find the 'dl' entry corresponding to that. */
4096 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4097 if (mdi
->state_fd
>= 0 &&
4098 mdi
->disk
.major
== dl
->major
&&
4099 mdi
->disk
.minor
== dl
->minor
)
4104 pd
= find_phys(ddf
, vc
->phys_refnum
[n_bvd
]);
4105 if (pd
< 0 || pd
!= dl
->pdnum
) {
4106 /* disk doesn't currently exist or has changed.
4107 * If it is now in_sync, insert it. */
4108 dprintf("%s: phys disk not found for %d: %d/%d ref %08x\n",
4109 __func__
, dl
->pdnum
, dl
->major
, dl
->minor
,
4110 __be32_to_cpu(dl
->disk
.refnum
));
4111 dprintf("%s: array %u disk %u ref %08x pd %d\n",
4112 __func__
, inst
, n_bvd
, vc
->phys_refnum
[n_bvd
], pd
);
4113 if ((state
& DS_INSYNC
) && ! (state
& DS_FAULTY
)) {
4114 pd
= dl
->pdnum
; /* FIXME: is this really correct ? */
4115 vc
->phys_refnum
[n_bvd
] = dl
->disk
.refnum
;
4116 LBA_OFFSET(ddf
, vc
)[n_bvd
] =
4117 __cpu_to_be64(mdi
->data_offset
);
4118 ddf
->phys
->entries
[pd
].type
&=
4119 ~__cpu_to_be16(DDF_Global_Spare
);
4120 ddf
->phys
->entries
[pd
].type
|=
4121 __cpu_to_be16(DDF_Active_in_VD
);
4122 ddf_set_updates_pending(ddf
);
4125 int old
= ddf
->phys
->entries
[pd
].state
;
4126 if (state
& DS_FAULTY
)
4127 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Failed
);
4128 if (state
& DS_INSYNC
) {
4129 ddf
->phys
->entries
[pd
].state
|= __cpu_to_be16(DDF_Online
);
4130 ddf
->phys
->entries
[pd
].state
&= __cpu_to_be16(~DDF_Rebuilding
);
4132 if (old
!= ddf
->phys
->entries
[pd
].state
)
4133 ddf_set_updates_pending(ddf
);
4136 dprintf("ddf: set_disk %d to %x\n", n
, state
);
4138 /* Now we need to check the state of the array and update
4139 * virtual_disk.entries[n].state.
4140 * It needs to be one of "optimal", "degraded", "failed".
4141 * I don't understand 'deleted' or 'missing'.
4143 state
= get_svd_state(ddf
, vcl
);
4145 if (ddf
->virt
->entries
[inst
].state
!=
4146 ((ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4149 ddf
->virt
->entries
[inst
].state
=
4150 (ddf
->virt
->entries
[inst
].state
& ~DDF_state_mask
)
4152 ddf_set_updates_pending(ddf
);
4157 static void ddf_sync_metadata(struct supertype
*st
)
4161 * Write all data to all devices.
4162 * Later, we might be able to track whether only local changes
4163 * have been made, or whether any global data has been changed,
4164 * but ddf is sufficiently weird that it probably always
4165 * changes global data ....
4167 struct ddf_super
*ddf
= st
->sb
;
4168 if (!ddf
->updates_pending
)
4170 ddf
->updates_pending
= 0;
4171 __write_init_super_ddf(st
);
4172 dprintf("ddf: sync_metadata\n");
4175 static int del_from_conflist(struct vcl
**list
, const char *guid
)
4179 for (p
= list
; p
&& *p
; p
= &((*p
)->next
))
4180 if (!memcmp((*p
)->conf
.guid
, guid
, DDF_GUID_LEN
)) {
4187 static int _kill_subarray_ddf(struct ddf_super
*ddf
, const char *guid
)
4190 unsigned int vdnum
, i
;
4191 vdnum
= find_vde_by_guid(ddf
, guid
);
4192 if (vdnum
== DDF_NOTFOUND
) {
4193 pr_err("%s: could not find VD %s\n", __func__
,
4197 if (del_from_conflist(&ddf
->conflist
, guid
) == 0) {
4198 pr_err("%s: could not find conf %s\n", __func__
,
4202 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4203 for (i
= 0; i
< ddf
->max_part
; i
++)
4204 if (dl
->vlist
[i
] != NULL
&&
4205 !memcmp(dl
->vlist
[i
]->conf
.guid
, guid
,
4207 dl
->vlist
[i
] = NULL
;
4208 memset(ddf
->virt
->entries
[vdnum
].guid
, 0xff, DDF_GUID_LEN
);
4209 dprintf("%s: deleted %s\n", __func__
, guid_str(guid
));
4213 static int kill_subarray_ddf(struct supertype
*st
)
4215 struct ddf_super
*ddf
= st
->sb
;
4217 * currentconf is set in container_content_ddf,
4218 * called with subarray arg
4220 struct vcl
*victim
= ddf
->currentconf
;
4221 struct vd_config
*conf
;
4222 ddf
->currentconf
= NULL
;
4225 pr_err("%s: nothing to kill\n", __func__
);
4228 conf
= &victim
->conf
;
4229 vdnum
= find_vde_by_guid(ddf
, conf
->guid
);
4230 if (vdnum
== DDF_NOTFOUND
) {
4231 pr_err("%s: could not find VD %s\n", __func__
,
4232 guid_str(conf
->guid
));
4235 if (st
->update_tail
) {
4236 struct virtual_disk
*vd
;
4237 int len
= sizeof(struct virtual_disk
)
4238 + sizeof(struct virtual_entry
);
4241 pr_err("%s: failed to allocate %d bytes\n", __func__
,
4245 memset(vd
, 0 , len
);
4246 vd
->magic
= DDF_VIRT_RECORDS_MAGIC
;
4247 vd
->populated_vdes
= 0;
4248 memcpy(vd
->entries
[0].guid
, conf
->guid
, DDF_GUID_LEN
);
4249 /* we use DDF_state_deleted as marker */
4250 vd
->entries
[0].state
= DDF_state_deleted
;
4251 append_metadata_update(st
, vd
, len
);
4253 _kill_subarray_ddf(ddf
, conf
->guid
);
4257 static void ddf_process_update(struct supertype
*st
,
4258 struct metadata_update
*update
)
4260 /* Apply this update to the metadata.
4261 * The first 4 bytes are a DDF_*_MAGIC which guides
4263 * Possible update are:
4264 * DDF_PHYS_RECORDS_MAGIC
4265 * Add a new physical device or remove an old one.
4266 * Changes to this record only happen implicitly.
4267 * used_pdes is the device number.
4268 * DDF_VIRT_RECORDS_MAGIC
4269 * Add a new VD. Possibly also change the 'access' bits.
4270 * populated_vdes is the entry number.
4272 * New or updated VD. the VIRT_RECORD must already
4273 * exist. For an update, phys_refnum and lba_offset
4274 * (at least) are updated, and the VD_CONF must
4275 * be written to precisely those devices listed with
4277 * DDF_SPARE_ASSIGN_MAGIC
4278 * replacement Spare Assignment Record... but for which device?
4281 * - to create a new array, we send a VIRT_RECORD and
4282 * a VD_CONF. Then assemble and start the array.
4283 * - to activate a spare we send a VD_CONF to add the phys_refnum
4284 * and offset. This will also mark the spare as active with
4285 * a spare-assignment record.
4287 struct ddf_super
*ddf
= st
->sb
;
4288 __u32
*magic
= (__u32
*)update
->buf
;
4289 struct phys_disk
*pd
;
4290 struct virtual_disk
*vd
;
4291 struct vd_config
*vc
;
4296 unsigned int pdnum
, pd2
;
4298 dprintf("Process update %x\n", *magic
);
4301 case DDF_PHYS_RECORDS_MAGIC
:
4303 if (update
->len
!= (sizeof(struct phys_disk
) +
4304 sizeof(struct phys_disk_entry
)))
4306 pd
= (struct phys_disk
*)update
->buf
;
4308 ent
= __be16_to_cpu(pd
->used_pdes
);
4309 if (ent
>= __be16_to_cpu(ddf
->phys
->max_pdes
))
4311 if (pd
->entries
[0].state
& __cpu_to_be16(DDF_Missing
)) {
4313 /* removing this disk. */
4314 ddf
->phys
->entries
[ent
].state
|= __cpu_to_be16(DDF_Missing
);
4315 for (dlp
= &ddf
->dlist
; *dlp
; dlp
= &(*dlp
)->next
) {
4316 struct dl
*dl
= *dlp
;
4317 if (dl
->pdnum
== (signed)ent
) {
4320 /* FIXME this doesn't free
4327 ddf_set_updates_pending(ddf
);
4330 if (!all_ff(ddf
->phys
->entries
[ent
].guid
))
4332 ddf
->phys
->entries
[ent
] = pd
->entries
[0];
4333 ddf
->phys
->used_pdes
= __cpu_to_be16(1 +
4334 __be16_to_cpu(ddf
->phys
->used_pdes
));
4335 ddf_set_updates_pending(ddf
);
4336 if (ddf
->add_list
) {
4337 struct active_array
*a
;
4338 struct dl
*al
= ddf
->add_list
;
4339 ddf
->add_list
= al
->next
;
4341 al
->next
= ddf
->dlist
;
4344 /* As a device has been added, we should check
4345 * for any degraded devices that might make
4346 * use of this spare */
4347 for (a
= st
->arrays
; a
; a
=a
->next
)
4348 a
->check_degraded
= 1;
4352 case DDF_VIRT_RECORDS_MAGIC
:
4354 if (update
->len
!= (sizeof(struct virtual_disk
) +
4355 sizeof(struct virtual_entry
)))
4357 vd
= (struct virtual_disk
*)update
->buf
;
4359 if (vd
->entries
[0].state
== DDF_state_deleted
) {
4360 if (_kill_subarray_ddf(ddf
, vd
->entries
[0].guid
))
4364 ent
= find_vde_by_guid(ddf
, vd
->entries
[0].guid
);
4365 if (ent
!= DDF_NOTFOUND
) {
4366 dprintf("%s: VD %s exists already in slot %d\n",
4367 __func__
, guid_str(vd
->entries
[0].guid
),
4371 ent
= find_unused_vde(ddf
);
4372 if (ent
== DDF_NOTFOUND
)
4374 ddf
->virt
->entries
[ent
] = vd
->entries
[0];
4375 ddf
->virt
->populated_vdes
=
4378 ddf
->virt
->populated_vdes
));
4379 dprintf("%s: added VD %s in slot %d\n",
4380 __func__
, guid_str(vd
->entries
[0].guid
), ent
);
4382 ddf_set_updates_pending(ddf
);
4385 case DDF_VD_CONF_MAGIC
:
4386 dprintf("len %d %d\n", update
->len
, ddf
->conf_rec_len
);
4388 mppe
= __be16_to_cpu(ddf
->anchor
.max_primary_element_entries
);
4389 if ((unsigned)update
->len
!= ddf
->conf_rec_len
* 512)
4391 vc
= (struct vd_config
*)update
->buf
;
4392 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
)
4393 if (memcmp(vcl
->conf
.guid
, vc
->guid
, DDF_GUID_LEN
) == 0)
4395 dprintf("vcl = %p\n", vcl
);
4397 /* An update, just copy the phys_refnum and lba_offset
4400 struct vd_config
*conf
= &vcl
->conf
;
4401 if (vcl
->other_bvds
!= NULL
&&
4402 conf
->sec_elmnt_seq
!= vc
->sec_elmnt_seq
) {
4404 for (i
= 1; i
< conf
->sec_elmnt_count
; i
++)
4405 if (vcl
->other_bvds
[i
-1]->sec_elmnt_seq
4406 == vc
->sec_elmnt_seq
)
4408 if (i
== conf
->sec_elmnt_count
) {
4409 pr_err("%s/DDF_VD_CONF_MAGIC: BVD %u not found\n",
4410 __func__
, vc
->sec_elmnt_seq
);
4413 conf
= vcl
->other_bvds
[i
-1];
4415 memcpy(conf
->phys_refnum
, vc
->phys_refnum
,
4416 mppe
* (sizeof(__u32
) + sizeof(__u64
)));
4421 vcl
= update
->space
;
4422 update
->space
= NULL
;
4423 vcl
->next
= ddf
->conflist
;
4424 memcpy(&vcl
->conf
, vc
, update
->len
);
4425 ent
= find_vde_by_guid(ddf
, vc
->guid
);
4426 if (ent
== DDF_NOTFOUND
)
4429 ddf
->conflist
= vcl
;
4431 /* Set DDF_Transition on all Failed devices - to help
4432 * us detect those that are no longer in use
4434 for (pdnum
= 0; pdnum
< __be16_to_cpu(ddf
->phys
->used_pdes
); pdnum
++)
4435 if (ddf
->phys
->entries
[pdnum
].state
4436 & __be16_to_cpu(DDF_Failed
))
4437 ddf
->phys
->entries
[pdnum
].state
4438 |= __be16_to_cpu(DDF_Transition
);
4439 /* Now make sure vlist is correct for each dl. */
4440 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
) {
4441 unsigned int vn
= 0;
4442 int in_degraded
= 0;
4443 for (vcl
= ddf
->conflist
; vcl
; vcl
= vcl
->next
) {
4444 unsigned int dn
, ibvd
;
4445 const struct vd_config
*conf
;
4447 dn
= get_pd_index_from_refnum(vcl
,
4451 if (dn
== DDF_NOTFOUND
)
4453 dprintf("dev %d/%08x has %s (sec=%u) at %d\n",
4455 __be32_to_cpu(dl
->disk
.refnum
),
4456 guid_str(conf
->guid
),
4457 conf
->sec_elmnt_seq
, vn
);
4458 /* Clear the Transition flag */
4459 if (ddf
->phys
->entries
[dl
->pdnum
].state
4460 & __be16_to_cpu(DDF_Failed
))
4461 ddf
->phys
->entries
[dl
->pdnum
].state
&=
4462 ~__be16_to_cpu(DDF_Transition
);
4463 dl
->vlist
[vn
++] = vcl
;
4464 vstate
= ddf
->virt
->entries
[vcl
->vcnum
].state
4466 if (vstate
== DDF_state_degraded
||
4467 vstate
== DDF_state_part_optimal
)
4470 while (vn
< ddf
->max_part
)
4471 dl
->vlist
[vn
++] = NULL
;
4473 ddf
->phys
->entries
[dl
->pdnum
].type
&=
4474 ~__cpu_to_be16(DDF_Global_Spare
);
4475 if (!(ddf
->phys
->entries
[dl
->pdnum
].type
&
4476 __cpu_to_be16(DDF_Active_in_VD
))) {
4477 ddf
->phys
->entries
[dl
->pdnum
].type
|=
4478 __cpu_to_be16(DDF_Active_in_VD
);
4480 ddf
->phys
->entries
[dl
->pdnum
].state
|=
4481 __cpu_to_be16(DDF_Rebuilding
);
4485 ddf
->phys
->entries
[dl
->pdnum
].type
&=
4486 ~__cpu_to_be16(DDF_Global_Spare
);
4487 ddf
->phys
->entries
[dl
->pdnum
].type
|=
4488 __cpu_to_be16(DDF_Spare
);
4490 if (!dl
->vlist
[0] && !dl
->spare
) {
4491 ddf
->phys
->entries
[dl
->pdnum
].type
|=
4492 __cpu_to_be16(DDF_Global_Spare
);
4493 ddf
->phys
->entries
[dl
->pdnum
].type
&=
4494 ~__cpu_to_be16(DDF_Spare
|
4499 /* Now remove any 'Failed' devices that are not part
4500 * of any VD. They will have the Transition flag set.
4501 * Once done, we need to update all dl->pdnum numbers.
4504 for (pdnum
= 0; pdnum
< __be16_to_cpu(ddf
->phys
->used_pdes
); pdnum
++)
4505 if ((ddf
->phys
->entries
[pdnum
].state
4506 & __be16_to_cpu(DDF_Failed
))
4507 && (ddf
->phys
->entries
[pdnum
].state
4508 & __be16_to_cpu(DDF_Transition
)))
4509 /* skip this one */;
4510 else if (pdnum
== pd2
)
4513 ddf
->phys
->entries
[pd2
] = ddf
->phys
->entries
[pdnum
];
4514 for (dl
= ddf
->dlist
; dl
; dl
= dl
->next
)
4515 if (dl
->pdnum
== (int)pdnum
)
4519 ddf
->phys
->used_pdes
= __cpu_to_be16(pd2
);
4520 while (pd2
< pdnum
) {
4521 memset(ddf
->phys
->entries
[pd2
].guid
, 0xff, DDF_GUID_LEN
);
4525 ddf_set_updates_pending(ddf
);
4527 case DDF_SPARE_ASSIGN_MAGIC
:
4532 static void ddf_prepare_update(struct supertype
*st
,
4533 struct metadata_update
*update
)
4535 /* This update arrived at managemon.
4536 * We are about to pass it to monitor.
4537 * If a malloc is needed, do it here.
4539 struct ddf_super
*ddf
= st
->sb
;
4540 __u32
*magic
= (__u32
*)update
->buf
;
4541 if (*magic
== DDF_VD_CONF_MAGIC
)
4542 if (posix_memalign(&update
->space
, 512,
4543 offsetof(struct vcl
, conf
)
4544 + ddf
->conf_rec_len
* 512) != 0)
4545 update
->space
= NULL
;
4549 * Check if the array 'a' is degraded but not failed.
4550 * If it is, find as many spares as are available and needed and
4551 * arrange for their inclusion.
4552 * We only choose devices which are not already in the array,
4553 * and prefer those with a spare-assignment to this array.
4554 * otherwise we choose global spares - assuming always that
4555 * there is enough room.
4556 * For each spare that we assign, we return an 'mdinfo' which
4557 * describes the position for the device in the array.
4558 * We also add to 'updates' a DDF_VD_CONF_MAGIC update with
4559 * the new phys_refnum and lba_offset values.
4561 * Only worry about BVDs at the moment.
4563 static struct mdinfo
*ddf_activate_spare(struct active_array
*a
,
4564 struct metadata_update
**updates
)
4568 struct ddf_super
*ddf
= a
->container
->sb
;
4570 struct mdinfo
*rv
= NULL
;
4572 struct metadata_update
*mu
;
4576 struct vd_config
*vc
;
4579 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
4580 if ((d
->curr_state
& DS_FAULTY
) &&
4582 /* wait for Removal to happen */
4584 if (d
->state_fd
>= 0)
4588 dprintf("ddf_activate: working=%d (%d) level=%d\n", working
, a
->info
.array
.raid_disks
,
4589 a
->info
.array
.level
);
4590 if (working
== a
->info
.array
.raid_disks
)
4591 return NULL
; /* array not degraded */
4592 switch (a
->info
.array
.level
) {
4595 return NULL
; /* failed */
4599 if (working
< a
->info
.array
.raid_disks
- 1)
4600 return NULL
; /* failed */
4603 if (working
< a
->info
.array
.raid_disks
- 2)
4604 return NULL
; /* failed */
4606 default: /* concat or stripe */
4607 return NULL
; /* failed */
4610 /* For each slot, if it is not working, find a spare */
4612 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
4613 for (d
= a
->info
.devs
; d
; d
= d
->next
)
4614 if (d
->disk
.raid_disk
== i
)
4616 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
4617 if (d
&& (d
->state_fd
>= 0))
4620 /* OK, this device needs recovery. Find a spare */
4622 for ( ; dl
; dl
= dl
->next
) {
4623 unsigned long long esize
;
4624 unsigned long long pos
;
4627 int is_dedicated
= 0;
4630 /* If in this array, skip */
4631 for (d2
= a
->info
.devs
; d2
; d2
= d2
->next
)
4632 if (d2
->state_fd
>= 0 &&
4633 d2
->disk
.major
== dl
->major
&&
4634 d2
->disk
.minor
== dl
->minor
) {
4635 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
4640 if (ddf
->phys
->entries
[dl
->pdnum
].type
&
4641 __cpu_to_be16(DDF_Spare
)) {
4642 /* Check spare assign record */
4644 if (dl
->spare
->type
& DDF_spare_dedicated
) {
4645 /* check spare_ents for guid */
4647 j
< __be16_to_cpu(dl
->spare
->populated
);
4649 if (memcmp(dl
->spare
->spare_ents
[j
].guid
,
4650 ddf
->virt
->entries
[a
->info
.container_member
].guid
,
4657 } else if (ddf
->phys
->entries
[dl
->pdnum
].type
&
4658 __cpu_to_be16(DDF_Global_Spare
)) {
4660 } else if (!(ddf
->phys
->entries
[dl
->pdnum
].state
&
4661 __cpu_to_be16(DDF_Failed
))) {
4662 /* we can possibly use some of this */
4665 if ( ! (is_dedicated
||
4666 (is_global
&& global_ok
))) {
4667 dprintf("%x:%x not suitable: %d %d\n", dl
->major
, dl
->minor
,
4668 is_dedicated
, is_global
);
4672 /* We are allowed to use this device - is there space?
4673 * We need a->info.component_size sectors */
4674 ex
= get_extents(ddf
, dl
);
4676 dprintf("cannot get extents\n");
4683 esize
= ex
[j
].start
- pos
;
4684 if (esize
>= a
->info
.component_size
)
4686 pos
= ex
[j
].start
+ ex
[j
].size
;
4688 } while (ex
[j
-1].size
);
4691 if (esize
< a
->info
.component_size
) {
4692 dprintf("%x:%x has no room: %llu %llu\n",
4693 dl
->major
, dl
->minor
,
4694 esize
, a
->info
.component_size
);
4699 /* Cool, we have a device with some space at pos */
4700 di
= xcalloc(1, sizeof(*di
));
4701 di
->disk
.number
= i
;
4702 di
->disk
.raid_disk
= i
;
4703 di
->disk
.major
= dl
->major
;
4704 di
->disk
.minor
= dl
->minor
;
4706 di
->recovery_start
= 0;
4707 di
->data_offset
= pos
;
4708 di
->component_size
= a
->info
.component_size
;
4709 di
->container_member
= dl
->pdnum
;
4712 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
4717 if (!dl
&& ! global_ok
) {
4718 /* not enough dedicated spares, try global */
4726 /* No spares found */
4728 /* Now 'rv' has a list of devices to return.
4729 * Create a metadata_update record to update the
4730 * phys_refnum and lba_offset values
4732 mu
= xmalloc(sizeof(*mu
));
4733 if (posix_memalign(&mu
->space
, 512, sizeof(struct vcl
)) != 0) {
4737 mu
->buf
= xmalloc(ddf
->conf_rec_len
* 512);
4738 mu
->len
= ddf
->conf_rec_len
* 512;
4740 mu
->space_list
= NULL
;
4741 mu
->next
= *updates
;
4742 vc
= find_vdcr(ddf
, a
->info
.container_member
, di
->disk
.raid_disk
,
4744 memcpy(mu
->buf
, vc
, ddf
->conf_rec_len
* 512);
4746 vc
= (struct vd_config
*)mu
->buf
;
4747 for (di
= rv
; di
; di
= di
->next
) {
4748 vc
->phys_refnum
[di
->disk
.raid_disk
] =
4749 ddf
->phys
->entries
[dl
->pdnum
].refnum
;
4750 LBA_OFFSET(ddf
, vc
)[di
->disk
.raid_disk
]
4751 = __cpu_to_be64(di
->data_offset
);
4756 #endif /* MDASSEMBLE */
4758 static int ddf_level_to_layout(int level
)
4765 return ALGORITHM_LEFT_SYMMETRIC
;
4767 return ALGORITHM_ROTATING_N_CONTINUE
;
4775 static void default_geometry_ddf(struct supertype
*st
, int *level
, int *layout
, int *chunk
)
4777 if (level
&& *level
== UnSet
)
4778 *level
= LEVEL_CONTAINER
;
4780 if (level
&& layout
&& *layout
== UnSet
)
4781 *layout
= ddf_level_to_layout(*level
);
4784 struct superswitch super_ddf
= {
4786 .examine_super
= examine_super_ddf
,
4787 .brief_examine_super
= brief_examine_super_ddf
,
4788 .brief_examine_subarrays
= brief_examine_subarrays_ddf
,
4789 .export_examine_super
= export_examine_super_ddf
,
4790 .detail_super
= detail_super_ddf
,
4791 .brief_detail_super
= brief_detail_super_ddf
,
4792 .validate_geometry
= validate_geometry_ddf
,
4793 .write_init_super
= write_init_super_ddf
,
4794 .add_to_super
= add_to_super_ddf
,
4795 .remove_from_super
= remove_from_super_ddf
,
4796 .load_container
= load_container_ddf
,
4797 .copy_metadata
= copy_metadata_ddf
,
4799 .match_home
= match_home_ddf
,
4800 .uuid_from_super
= uuid_from_super_ddf
,
4801 .getinfo_super
= getinfo_super_ddf
,
4802 .update_super
= update_super_ddf
,
4804 .avail_size
= avail_size_ddf
,
4806 .compare_super
= compare_super_ddf
,
4808 .load_super
= load_super_ddf
,
4809 .init_super
= init_super_ddf
,
4810 .store_super
= store_super_ddf
,
4811 .free_super
= free_super_ddf
,
4812 .match_metadata_desc
= match_metadata_desc_ddf
,
4813 .container_content
= container_content_ddf
,
4814 .default_geometry
= default_geometry_ddf
,
4815 .kill_subarray
= kill_subarray_ddf
,
4821 .open_new
= ddf_open_new
,
4822 .set_array_state
= ddf_set_array_state
,
4823 .set_disk
= ddf_set_disk
,
4824 .sync_metadata
= ddf_sync_metadata
,
4825 .process_update
= ddf_process_update
,
4826 .prepare_update
= ddf_prepare_update
,
4827 .activate_spare
= ddf_activate_spare
,