2 * mdadm - Intel(R) Matrix Storage Manager Support
4 * Copyright (C) 2002-2007 Intel Corporation
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
26 /* MPB == Metadata Parameter Block */
27 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
28 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
29 #define MPB_VERSION_RAID0 "1.0.00"
30 #define MPB_VERSION_RAID1 "1.1.00"
31 #define MPB_VERSION_RAID5 "1.2.02"
32 #define MAX_SIGNATURE_LENGTH 32
33 #define MAX_RAID_SERIAL_LEN 16
34 #define MPB_SECTOR_CNT 418
35 #define IMSM_RESERVED_SECTORS 4096
37 /* Disk configuration info. */
38 #define IMSM_MAX_DEVICES 255
40 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
41 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
42 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
43 __u32 status
; /* 0xF0 - 0xF3 */
44 #define SPARE_DISK 0x01 /* Spare */
45 #define CONFIGURED_DISK 0x02 /* Member of some RaidDev */
46 #define FAILED_DISK 0x04 /* Permanent failure */
47 #define USABLE_DISK 0x08 /* Fully usable unless FAILED_DISK is set */
49 #define IMSM_DISK_FILLERS 5
50 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
53 /* RAID map configuration infos. */
55 __u32 pba_of_lba0
; /* start address of partition */
56 __u32 blocks_per_member
;/* blocks per member */
57 __u32 num_data_stripes
; /* number of data stripes */
58 __u16 blocks_per_strip
;
59 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
60 #define IMSM_T_STATE_NORMAL 0
61 #define IMSM_T_STATE_UNINITIALIZED 1
62 #define IMSM_T_STATE_DEGRADED 2 /* FIXME: is this correct? */
63 #define IMSM_T_STATE_FAILED 3 /* FIXME: is this correct? */
65 #define IMSM_T_RAID0 0
66 #define IMSM_T_RAID1 1
67 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
68 __u8 num_members
; /* number of member disks */
70 __u32 filler
[7]; /* expansion area */
71 #define IMSM_ORD_REBUILD (1 << 24)
72 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
73 * top byte contains some flags
75 } __attribute__ ((packed
));
79 __u8 migr_state
; /* Normal or Migrating */
80 __u8 migr_type
; /* Initializing, Rebuilding, ... */
84 struct imsm_map map
[1];
85 /* here comes another one if migr_state */
86 } __attribute__ ((packed
));
89 __u8 volume
[MAX_RAID_SERIAL_LEN
];
92 __u32 status
; /* Persistent RaidDev status */
93 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
94 #define IMSM_DEV_FILLERS 12
95 __u32 filler
[IMSM_DEV_FILLERS
];
97 } __attribute__ ((packed
));
100 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
101 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
102 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
103 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
104 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
105 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
106 __u32 attributes
; /* 0x34 - 0x37 */
107 __u8 num_disks
; /* 0x38 Number of configured disks */
108 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
109 __u8 error_log_pos
; /* 0x3A */
110 __u8 fill
[1]; /* 0x3B */
111 __u32 cache_size
; /* 0x3c - 0x40 in mb */
112 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
113 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
114 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
115 #define IMSM_FILLERS 35
116 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
117 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
118 /* here comes imsm_dev[num_raid_devs] */
119 /* here comes BBM logs */
120 } __attribute__ ((packed
));
122 #define BBM_LOG_MAX_ENTRIES 254
124 struct bbm_log_entry
{
125 __u64 defective_block_start
;
126 #define UNREADABLE 0xFFFFFFFF
127 __u32 spare_block_offset
;
128 __u16 remapped_marked_count
;
130 } __attribute__ ((__packed__
));
133 __u32 signature
; /* 0xABADB10C */
135 __u32 reserved_spare_block_count
; /* 0 */
136 __u32 reserved
; /* 0xFFFF */
137 __u64 first_spare_lba
;
138 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
139 } __attribute__ ((__packed__
));
143 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
146 static unsigned int sector_count(__u32 bytes
)
148 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
151 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
153 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
156 /* internal representation of IMSM metadata */
159 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
160 struct imsm_super
*anchor
; /* immovable parameters */
162 size_t len
; /* size of the 'buf' allocation */
163 void *next_buf
; /* for realloc'ing buf from the manager */
165 int updates_pending
; /* count of pending updates for mdmon */
166 int creating_imsm
; /* flag to indicate container creation */
167 int current_vol
; /* index of raid device undergoing creation */
168 #define IMSM_MAX_RAID_DEVS 2
169 struct imsm_dev
*dev_tbl
[IMSM_MAX_RAID_DEVS
];
173 __u8 serial
[MAX_RAID_SERIAL_LEN
];
176 struct imsm_disk disk
;
179 struct dl
*add
; /* list of disks to add while mdmon active */
180 struct bbm_log
*bbm_log
;
184 unsigned long long start
, size
;
187 /* definition of messages passed to imsm_process_update */
188 enum imsm_update_type
{
189 update_activate_spare
,
194 struct imsm_update_activate_spare
{
195 enum imsm_update_type type
;
199 struct imsm_update_activate_spare
*next
;
202 struct imsm_update_create_array
{
203 enum imsm_update_type type
;
208 struct imsm_update_add_disk
{
209 enum imsm_update_type type
;
212 static int imsm_env_devname_as_serial(void)
214 char *val
= getenv("IMSM_DEVNAME_AS_SERIAL");
216 if (val
&& atoi(val
) == 1)
223 static struct supertype
*match_metadata_desc_imsm(char *arg
)
225 struct supertype
*st
;
227 if (strcmp(arg
, "imsm") != 0 &&
228 strcmp(arg
, "default") != 0
232 st
= malloc(sizeof(*st
));
233 memset(st
, 0, sizeof(*st
));
234 st
->ss
= &super_imsm
;
235 st
->max_devs
= IMSM_MAX_DEVICES
;
236 st
->minor_version
= 0;
241 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
243 return &mpb
->sig
[MPB_SIG_LEN
];
246 /* retrieve a disk directly from the anchor when the anchor is known to be
247 * up-to-date, currently only at load time
249 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
251 if (index
>= mpb
->num_disks
)
253 return &mpb
->disk
[index
];
256 /* retrieve a disk from the parsed metadata */
257 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
261 for (d
= super
->disks
; d
; d
= d
->next
)
262 if (d
->index
== index
)
268 /* generate a checksum directly from the anchor when the anchor is known to be
269 * up-to-date, currently only at load or write_super after coalescing
271 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
273 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
274 __u32
*p
= (__u32
*) mpb
;
278 sum
+= __le32_to_cpu(*p
++);
280 return sum
- __le32_to_cpu(mpb
->check_sum
);
283 static size_t sizeof_imsm_map(struct imsm_map
*map
)
285 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
288 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
290 struct imsm_map
*map
= &dev
->vol
.map
[0];
292 if (second_map
&& !dev
->vol
.migr_state
)
294 else if (second_map
) {
297 return ptr
+ sizeof_imsm_map(map
);
303 /* return the size of the device.
304 * migr_state increases the returned size if map[0] were to be duplicated
306 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
308 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
309 sizeof_imsm_map(get_imsm_map(dev
, 0));
311 /* migrating means an additional map */
312 if (dev
->vol
.migr_state
)
313 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
315 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
320 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
326 if (index
>= mpb
->num_raid_devs
)
329 /* devices start after all disks */
330 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
332 for (i
= 0; i
<= index
; i
++)
334 return _mpb
+ offset
;
336 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
341 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
343 if (index
>= super
->anchor
->num_raid_devs
)
345 return super
->dev_tbl
[index
];
348 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
350 struct imsm_map
*map
;
352 if (dev
->vol
.migr_state
)
353 map
= get_imsm_map(dev
, 1);
355 map
= get_imsm_map(dev
, 0);
357 /* top byte identifies disk under rebuild */
358 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
361 #define ord_to_idx(ord) (((ord) << 8) >> 8)
362 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
364 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
366 return ord_to_idx(ord
);
369 static int get_imsm_raid_level(struct imsm_map
*map
)
371 if (map
->raid_level
== 1) {
372 if (map
->num_members
== 2)
378 return map
->raid_level
;
381 static int cmp_extent(const void *av
, const void *bv
)
383 const struct extent
*a
= av
;
384 const struct extent
*b
= bv
;
385 if (a
->start
< b
->start
)
387 if (a
->start
> b
->start
)
392 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
394 /* find a list of used extents on the given physical device */
395 struct extent
*rv
, *e
;
399 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
400 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
401 struct imsm_map
*map
= get_imsm_map(dev
, 0);
403 for (j
= 0; j
< map
->num_members
; j
++) {
404 __u32 index
= get_imsm_disk_idx(dev
, j
);
406 if (index
== dl
->index
)
410 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
415 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
416 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
417 struct imsm_map
*map
= get_imsm_map(dev
, 0);
419 for (j
= 0; j
< map
->num_members
; j
++) {
420 __u32 index
= get_imsm_disk_idx(dev
, j
);
422 if (index
== dl
->index
) {
423 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
424 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
429 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
431 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) -
432 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
438 static void print_imsm_dev(struct imsm_dev
*dev
, int index
)
442 struct imsm_map
*map
= get_imsm_map(dev
, 0);
445 printf("[%s]:\n", dev
->volume
);
446 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
447 printf(" Members : %d\n", map
->num_members
);
448 for (slot
= 0; slot
< map
->num_members
; slot
++)
449 if (index
== get_imsm_disk_idx(dev
, slot
))
451 if (slot
< map
->num_members
)
452 printf(" This Slot : %d\n", slot
);
454 printf(" This Slot : ?\n");
455 sz
= __le32_to_cpu(dev
->size_high
);
457 sz
+= __le32_to_cpu(dev
->size_low
);
458 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
459 human_size(sz
* 512));
460 sz
= __le32_to_cpu(map
->blocks_per_member
);
461 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
462 human_size(sz
* 512));
463 printf(" Sector Offset : %u\n",
464 __le32_to_cpu(map
->pba_of_lba0
));
465 printf(" Num Stripes : %u\n",
466 __le32_to_cpu(map
->num_data_stripes
));
467 printf(" Chunk Size : %u KiB\n",
468 __le16_to_cpu(map
->blocks_per_strip
) / 2);
469 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
470 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle");
471 if (dev
->vol
.migr_state
)
472 printf(": %s", dev
->vol
.migr_type
? "rebuilding" : "initializing");
474 printf(" Map State : %s", map_state_str
[map
->map_state
]);
475 if (dev
->vol
.migr_state
) {
476 struct imsm_map
*map
= get_imsm_map(dev
, 1);
477 printf(", %s", map_state_str
[map
->map_state
]);
480 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
483 static void print_imsm_disk(struct imsm_super
*mpb
, int index
)
485 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
486 char str
[MAX_RAID_SERIAL_LEN
];
494 snprintf(str
, MAX_RAID_SERIAL_LEN
, "%s", disk
->serial
);
495 printf(" Disk%02d Serial : %s\n", index
, str
);
496 s
= __le32_to_cpu(disk
->status
);
497 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
498 s
&CONFIGURED_DISK
? " active" : "",
499 s
&FAILED_DISK
? " failed" : "",
500 s
&USABLE_DISK
? " usable" : "");
501 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
502 sz
= __le32_to_cpu(disk
->total_blocks
) -
503 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
* mpb
->num_raid_devs
);
504 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
505 human_size(sz
* 512));
508 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
510 struct intel_super
*super
= st
->sb
;
511 struct imsm_super
*mpb
= super
->anchor
;
512 char str
[MAX_SIGNATURE_LENGTH
];
516 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
517 printf(" Magic : %s\n", str
);
518 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
519 printf(" Version : %s\n", get_imsm_version(mpb
));
520 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
521 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
522 sum
= __le32_to_cpu(mpb
->check_sum
);
523 printf(" Checksum : %08x %s\n", sum
,
524 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
525 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
526 printf(" Disks : %d\n", mpb
->num_disks
);
527 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
528 print_imsm_disk(mpb
, super
->disks
->index
);
529 if (super
->bbm_log
) {
530 struct bbm_log
*log
= super
->bbm_log
;
533 printf("Bad Block Management Log:\n");
534 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
535 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
536 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
537 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
538 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
540 for (i
= 0; i
< mpb
->num_raid_devs
; i
++)
541 print_imsm_dev(__get_imsm_dev(mpb
, i
), super
->disks
->index
);
542 for (i
= 0; i
< mpb
->num_disks
; i
++) {
543 if (i
== super
->disks
->index
)
545 print_imsm_disk(mpb
, i
);
549 static void brief_examine_super_imsm(struct supertype
*st
)
551 printf("ARRAY /dev/imsm metadata=imsm\n");
554 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
556 printf("%s\n", __FUNCTION__
);
559 static void brief_detail_super_imsm(struct supertype
*st
)
561 printf("%s\n", __FUNCTION__
);
565 static int match_home_imsm(struct supertype
*st
, char *homehost
)
567 printf("%s\n", __FUNCTION__
);
572 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
574 /* imsm does not track uuid's so just make sure we never return
575 * the same value twice to break uuid matching in Manage_subdevs
576 * FIXME what about the use of uuid's with bitmap's?
578 static int dummy_id
= 0;
580 uuid
[0] = dummy_id
++;
585 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
587 __u8
*v
= get_imsm_version(mpb
);
588 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
589 char major
[] = { 0, 0, 0 };
590 char minor
[] = { 0 ,0, 0 };
591 char patch
[] = { 0, 0, 0 };
592 char *ver_parse
[] = { major
, minor
, patch
};
596 while (*v
!= '\0' && v
< end
) {
597 if (*v
!= '.' && j
< 2)
598 ver_parse
[i
][j
++] = *v
;
606 *m
= strtol(minor
, NULL
, 0);
607 *p
= strtol(patch
, NULL
, 0);
611 static int imsm_level_to_layout(int level
)
619 return ALGORITHM_LEFT_ASYMMETRIC
;
621 return 0x102; //FIXME is this correct?
626 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
628 struct intel_super
*super
= st
->sb
;
629 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
630 struct imsm_map
*map
= get_imsm_map(dev
, 0);
632 info
->container_member
= super
->current_vol
;
633 info
->array
.raid_disks
= map
->num_members
;
634 info
->array
.level
= get_imsm_raid_level(map
);
635 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
636 info
->array
.md_minor
= -1;
637 info
->array
.ctime
= 0;
638 info
->array
.utime
= 0;
639 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
* 512);
641 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
642 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
644 info
->disk
.major
= 0;
645 info
->disk
.minor
= 0;
647 sprintf(info
->text_version
, "/%s/%d",
648 devnum2devname(st
->container_dev
),
649 info
->container_member
);
653 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
655 struct intel_super
*super
= st
->sb
;
656 struct imsm_disk
*disk
;
659 if (super
->current_vol
>= 0) {
660 getinfo_super_imsm_volume(st
, info
);
664 /* Set raid_disks to zero so that Assemble will always pull in valid
667 info
->array
.raid_disks
= 0;
668 info
->array
.level
= LEVEL_CONTAINER
;
669 info
->array
.layout
= 0;
670 info
->array
.md_minor
= -1;
671 info
->array
.ctime
= 0; /* N/A for imsm */
672 info
->array
.utime
= 0;
673 info
->array
.chunk_size
= 0;
675 info
->disk
.major
= 0;
676 info
->disk
.minor
= 0;
677 info
->disk
.raid_disk
= -1;
678 info
->reshape_active
= 0;
679 strcpy(info
->text_version
, "imsm");
680 info
->disk
.number
= -1;
681 info
->disk
.state
= 0;
684 disk
= &super
->disks
->disk
;
685 info
->disk
.number
= super
->disks
->index
;
686 info
->disk
.raid_disk
= super
->disks
->index
;
687 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) -
688 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
689 info
->component_size
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
690 s
= __le32_to_cpu(disk
->status
);
691 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
692 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
693 info
->disk
.state
|= s
& USABLE_DISK
? (1 << MD_DISK_SYNC
) : 0;
697 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
698 char *update
, char *devname
, int verbose
,
699 int uuid_set
, char *homehost
)
703 /* For 'assemble' and 'force' we need to return non-zero if any
704 * change was made. For others, the return value is ignored.
705 * Update options are:
706 * force-one : This device looks a bit old but needs to be included,
707 * update age info appropriately.
708 * assemble: clear any 'faulty' flag to allow this device to
710 * force-array: Array is degraded but being forced, mark it clean
711 * if that will be needed to assemble it.
713 * newdev: not used ????
714 * grow: Array has gained a new device - this is currently for
716 * resync: mark as dirty so a resync will happen.
717 * name: update the name - preserving the homehost
719 * Following are not relevant for this imsm:
720 * sparc2.2 : update from old dodgey metadata
721 * super-minor: change the preferred_minor number
722 * summaries: update redundant counters.
723 * uuid: Change the uuid of the array to match watch is given
724 * homehost: update the recorded homehost
725 * _reshape_progress: record new reshape_progress position.
728 //struct intel_super *super = st->sb;
729 //struct imsm_super *mpb = super->mpb;
731 if (strcmp(update
, "grow") == 0) {
733 if (strcmp(update
, "resync") == 0) {
734 /* dev->vol.dirty = 1; */
737 /* IMSM has no concept of UUID or homehost */
742 static size_t disks_to_mpb_size(int disks
)
746 size
= sizeof(struct imsm_super
);
747 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
748 size
+= 2 * sizeof(struct imsm_dev
);
749 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
750 size
+= (4 - 2) * sizeof(struct imsm_map
);
751 /* 4 possible disk_ord_tbl's */
752 size
+= 4 * (disks
- 1) * sizeof(__u32
);
757 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
759 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
762 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
765 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
769 * 0 same, or first was empty, and second was copied
770 * 1 second had wrong number
774 struct intel_super
*first
= st
->sb
;
775 struct intel_super
*sec
= tst
->sb
;
783 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
786 /* if an anchor does not have num_raid_devs set then it is a free
789 if (first
->anchor
->num_raid_devs
> 0 &&
790 sec
->anchor
->num_raid_devs
> 0) {
791 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
795 /* if 'first' is a spare promote it to a populated mpb with sec's
798 if (first
->anchor
->num_raid_devs
== 0 &&
799 sec
->anchor
->num_raid_devs
> 0) {
800 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
801 first
->anchor
->family_num
= sec
->anchor
->family_num
;
807 static void fd2devname(int fd
, char *name
)
816 if (fstat(fd
, &st
) != 0)
818 sprintf(path
, "/sys/dev/block/%d:%d",
819 major(st
.st_rdev
), minor(st
.st_rdev
));
821 rv
= readlink(path
, dname
, sizeof(dname
));
826 nm
= strrchr(dname
, '/');
828 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
832 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
834 static int imsm_read_serial(int fd
, char *devname
,
835 __u8 serial
[MAX_RAID_SERIAL_LEN
])
837 unsigned char scsi_serial
[255];
842 memset(scsi_serial
, 0, sizeof(scsi_serial
));
844 if (imsm_env_devname_as_serial()) {
845 char name
[MAX_RAID_SERIAL_LEN
];
847 fd2devname(fd
, name
);
848 strcpy((char *) serial
, name
);
852 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
857 Name
": Failed to retrieve serial for %s\n",
862 rsp_len
= scsi_serial
[3];
863 for (i
= 0, cnt
= 0; i
< rsp_len
; i
++) {
864 if (!isspace(scsi_serial
[4 + i
]))
865 serial
[cnt
++] = scsi_serial
[4 + i
];
866 if (cnt
== MAX_RAID_SERIAL_LEN
)
870 serial
[MAX_RAID_SERIAL_LEN
- 1] = '\0';
876 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
883 __u8 serial
[MAX_RAID_SERIAL_LEN
];
885 rv
= imsm_read_serial(fd
, devname
, serial
);
890 /* check if this is a disk we have seen before. it may be a spare in
891 * super->disks while the current anchor believes it is a raid member,
892 * check if we need to update dl->index
894 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
895 if (memcmp(dl
->serial
, serial
, MAX_RAID_SERIAL_LEN
) == 0)
899 dl
= malloc(sizeof(*dl
));
906 Name
": failed to allocate disk buffer for %s\n",
913 dl
->major
= major(stb
.st_rdev
);
914 dl
->minor
= minor(stb
.st_rdev
);
915 dl
->next
= super
->disks
;
916 dl
->fd
= keep_fd
? fd
: -1;
917 dl
->devname
= devname
? strdup(devname
) : NULL
;
918 strncpy((char *) dl
->serial
, (char *) serial
, MAX_RAID_SERIAL_LEN
);
920 } else if (keep_fd
) {
925 /* look up this disk's index in the current anchor */
926 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
927 struct imsm_disk
*disk_iter
;
929 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
931 if (memcmp(disk_iter
->serial
, dl
->serial
,
932 MAX_RAID_SERIAL_LEN
) == 0) {
935 dl
->disk
= *disk_iter
;
936 status
= __le32_to_cpu(dl
->disk
.status
);
937 /* only set index on disks that are a member of a
938 * populated contianer, i.e. one with raid_devs
940 if (status
& FAILED_DISK
)
942 else if (status
& SPARE_DISK
)
957 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
959 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
962 static void dup_map(struct imsm_dev
*dev
)
964 struct imsm_map
*dest
= get_imsm_map(dev
, 1);
965 struct imsm_map
*src
= get_imsm_map(dev
, 0);
967 memcpy(dest
, src
, sizeof_imsm_map(src
));
970 static int parse_raid_devices(struct intel_super
*super
)
973 struct imsm_dev
*dev_new
;
974 size_t len
, len_migr
;
975 size_t space_needed
= 0;
976 struct imsm_super
*mpb
= super
->anchor
;
978 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
979 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
981 len
= sizeof_imsm_dev(dev_iter
, 0);
982 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
984 space_needed
+= len_migr
- len
;
986 dev_new
= malloc(len_migr
);
989 imsm_copy_dev(dev_new
, dev_iter
);
990 super
->dev_tbl
[i
] = dev_new
;
993 /* ensure that super->buf is large enough when all raid devices
996 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
999 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1000 if (posix_memalign(&buf
, 512, len
) != 0)
1003 memcpy(buf
, super
->buf
, len
);
1012 /* retrieve a pointer to the bbm log which starts after all raid devices */
1013 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1017 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1019 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1025 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1027 /* load_imsm_mpb - read matrix metadata
1028 * allocates super->mpb to be freed by free_super
1030 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1032 unsigned long long dsize
;
1033 unsigned long long sectors
;
1035 struct imsm_super
*anchor
;
1039 get_dev_size(fd
, NULL
, &dsize
);
1041 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1044 Name
": Cannot seek to anchor block on %s: %s\n",
1045 devname
, strerror(errno
));
1049 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1052 Name
": Failed to allocate imsm anchor buffer"
1053 " on %s\n", devname
);
1056 if (read(fd
, anchor
, 512) != 512) {
1059 Name
": Cannot read anchor block on %s: %s\n",
1060 devname
, strerror(errno
));
1065 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1068 Name
": no IMSM anchor on %s\n", devname
);
1073 __free_imsm(super
, 0);
1074 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1075 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1078 Name
": unable to allocate %zu byte mpb buffer\n",
1083 memcpy(super
->buf
, anchor
, 512);
1085 sectors
= mpb_sectors(anchor
) - 1;
1088 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1090 rc
= parse_raid_devices(super
);
1094 /* read the extended mpb */
1095 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1098 Name
": Cannot seek to extended mpb on %s: %s\n",
1099 devname
, strerror(errno
));
1103 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1106 Name
": Cannot read extended mpb on %s: %s\n",
1107 devname
, strerror(errno
));
1111 check_sum
= __gen_imsm_checksum(super
->anchor
);
1112 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1115 Name
": IMSM checksum %x != %x on %s\n",
1116 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1121 /* FIXME the BBM log is disk specific so we cannot use this global
1122 * buffer for all disks. Ok for now since we only look at the global
1123 * bbm_log_size parameter to gate assembly
1125 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1127 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1129 rc
= parse_raid_devices(super
);
1134 static void __free_imsm_disk(struct dl
*d
)
1143 static void free_imsm_disks(struct intel_super
*super
)
1145 while (super
->disks
) {
1146 struct dl
*d
= super
->disks
;
1148 super
->disks
= d
->next
;
1149 __free_imsm_disk(d
);
1153 /* free all the pieces hanging off of a super pointer */
1154 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1163 free_imsm_disks(super
);
1164 for (i
= 0; i
< IMSM_MAX_RAID_DEVS
; i
++)
1165 if (super
->dev_tbl
[i
]) {
1166 free(super
->dev_tbl
[i
]);
1167 super
->dev_tbl
[i
] = NULL
;
1171 static void free_imsm(struct intel_super
*super
)
1173 __free_imsm(super
, 1);
1177 static void free_super_imsm(struct supertype
*st
)
1179 struct intel_super
*super
= st
->sb
;
1188 static struct intel_super
*alloc_super(int creating_imsm
)
1190 struct intel_super
*super
= malloc(sizeof(*super
));
1193 memset(super
, 0, sizeof(*super
));
1194 super
->creating_imsm
= creating_imsm
;
1195 super
->current_vol
= -1;
1202 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
1203 char *devname
, int keep_fd
)
1206 struct intel_super
*super
;
1207 struct mdinfo
*sd
, *best
= NULL
;
1214 /* check if this disk is a member of an active array */
1215 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
1219 if (sra
->array
.major_version
!= -1 ||
1220 sra
->array
.minor_version
!= -2 ||
1221 strcmp(sra
->text_version
, "imsm") != 0)
1224 super
= alloc_super(0);
1228 /* find the most up to date disk in this array, skipping spares */
1229 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1230 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1231 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1236 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1240 if (super
->anchor
->num_raid_devs
== 0)
1243 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
1244 if (!best
|| gen
> bestgen
) {
1259 /* load the most up to date anchor */
1260 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
1261 dfd
= dev_open(nm
, O_RDONLY
);
1266 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1273 /* re-parse the disk list with the current anchor */
1274 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1275 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1276 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1281 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
1286 if (st
->subarray
[0]) {
1287 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
1288 super
->current_vol
= atoi(st
->subarray
);
1294 st
->container_dev
= fd2devnum(fd
);
1295 if (st
->ss
== NULL
) {
1296 st
->ss
= &super_imsm
;
1297 st
->minor_version
= 0;
1298 st
->max_devs
= IMSM_MAX_DEVICES
;
1305 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
1307 struct intel_super
*super
;
1311 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
1314 if (st
->subarray
[0])
1315 return 1; /* FIXME */
1317 super
= alloc_super(0);
1320 Name
": malloc of %zu failed.\n",
1325 rv
= load_imsm_mpb(fd
, super
, devname
);
1330 Name
": Failed to load all information "
1331 "sections on %s\n", devname
);
1337 if (st
->ss
== NULL
) {
1338 st
->ss
= &super_imsm
;
1339 st
->minor_version
= 0;
1340 st
->max_devs
= IMSM_MAX_DEVICES
;
1346 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
1348 if (info
->level
== 1)
1350 return info
->chunk_size
>> 9;
1353 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
1357 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
1358 if (info
->level
== 1)
1364 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
1366 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
1369 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
1370 unsigned long long size
, char *name
,
1371 char *homehost
, int *uuid
)
1373 /* We are creating a volume inside a pre-existing container.
1374 * so st->sb is already set.
1376 struct intel_super
*super
= st
->sb
;
1377 struct imsm_super
*mpb
= super
->anchor
;
1378 struct imsm_dev
*dev
;
1379 struct imsm_vol
*vol
;
1380 struct imsm_map
*map
;
1381 int idx
= mpb
->num_raid_devs
;
1383 unsigned long long array_blocks
;
1385 size_t size_old
, size_new
;
1387 if (mpb
->num_raid_devs
>= 2) {
1388 fprintf(stderr
, Name
": This imsm-container already has the "
1389 "maximum of 2 volumes\n");
1393 /* ensure the mpb is large enough for the new data */
1394 size_old
= __le32_to_cpu(mpb
->mpb_size
);
1395 size_new
= disks_to_mpb_size(info
->nr_disks
);
1396 if (size_new
> size_old
) {
1398 size_t size_round
= ROUND_UP(size_new
, 512);
1400 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
1401 fprintf(stderr
, Name
": could not allocate new mpb\n");
1404 memcpy(mpb_new
, mpb
, size_old
);
1407 super
->anchor
= mpb_new
;
1408 mpb
->mpb_size
= __cpu_to_le32(size_new
);
1409 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
1411 super
->current_vol
= idx
;
1412 /* when creating the first raid device in this container set num_disks
1413 * to zero, i.e. delete this spare and add raid member devices in
1414 * add_to_super_imsm_volume()
1416 if (super
->current_vol
== 0)
1418 sprintf(st
->subarray
, "%d", idx
);
1419 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
1421 fprintf(stderr
, Name
": could not allocate raid device\n");
1424 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
1425 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
1426 info
->layout
, info
->chunk_size
,
1428 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
1429 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
1430 dev
->status
= __cpu_to_le32(0);
1431 dev
->reserved_blocks
= __cpu_to_le32(0);
1433 vol
->migr_state
= 0;
1436 for (i
= 0; i
< idx
; i
++) {
1437 struct imsm_dev
*prev
= get_imsm_dev(super
, i
);
1438 struct imsm_map
*pmap
= get_imsm_map(prev
, 0);
1440 offset
+= __le32_to_cpu(pmap
->blocks_per_member
);
1441 offset
+= IMSM_RESERVED_SECTORS
;
1443 map
= get_imsm_map(dev
, 0);
1444 map
->pba_of_lba0
= __cpu_to_le32(offset
);
1445 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
1446 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
1447 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
1448 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
1449 IMSM_T_STATE_NORMAL
;
1451 if (info
->level
== 1 && info
->raid_disks
> 2) {
1452 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
1453 "in a raid1 volume\n");
1456 if (info
->level
== 10)
1457 map
->raid_level
= 1;
1459 map
->raid_level
= info
->level
;
1461 map
->num_members
= info
->raid_disks
;
1462 for (i
= 0; i
< map
->num_members
; i
++) {
1463 /* initialized in add_to_super */
1464 map
->disk_ord_tbl
[i
] = __cpu_to_le32(0);
1466 mpb
->num_raid_devs
++;
1467 super
->dev_tbl
[super
->current_vol
] = dev
;
1472 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
1473 unsigned long long size
, char *name
,
1474 char *homehost
, int *uuid
)
1476 /* This is primarily called by Create when creating a new array.
1477 * We will then get add_to_super called for each component, and then
1478 * write_init_super called to write it out to each device.
1479 * For IMSM, Create can create on fresh devices or on a pre-existing
1481 * To create on a pre-existing array a different method will be called.
1482 * This one is just for fresh drives.
1484 struct intel_super
*super
;
1485 struct imsm_super
*mpb
;
1493 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
1496 super
= alloc_super(1);
1499 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
1500 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
1505 memset(mpb
, 0, mpb_size
);
1507 memcpy(mpb
->sig
, MPB_SIGNATURE
, strlen(MPB_SIGNATURE
));
1508 memcpy(mpb
->sig
+ strlen(MPB_SIGNATURE
), MPB_VERSION_RAID5
,
1509 strlen(MPB_VERSION_RAID5
));
1510 mpb
->mpb_size
= mpb_size
;
1516 static void add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
1517 int fd
, char *devname
)
1519 struct intel_super
*super
= st
->sb
;
1520 struct imsm_super
*mpb
= super
->anchor
;
1522 struct imsm_dev
*dev
;
1523 struct imsm_map
*map
;
1526 dev
= get_imsm_dev(super
, super
->current_vol
);
1527 map
= get_imsm_map(dev
, 0);
1529 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1530 if (dl
->major
== dk
->major
&&
1531 dl
->minor
== dk
->minor
)
1534 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1537 /* add a pristine spare to the metadata */
1538 if (dl
->index
< 0) {
1539 dl
->index
= super
->anchor
->num_disks
;
1540 super
->anchor
->num_disks
++;
1542 map
->disk_ord_tbl
[dk
->number
] = __cpu_to_le32(dl
->index
);
1543 status
= CONFIGURED_DISK
| USABLE_DISK
;
1544 dl
->disk
.status
= __cpu_to_le32(status
);
1546 /* if we are creating the first raid device update the family number */
1547 if (super
->current_vol
== 0) {
1549 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
1550 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
1554 sum
= __gen_imsm_checksum(mpb
);
1555 mpb
->family_num
= __cpu_to_le32(sum
);
1559 static void add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
1560 int fd
, char *devname
)
1562 struct intel_super
*super
= st
->sb
;
1564 unsigned long long size
;
1569 if (super
->current_vol
>= 0) {
1570 add_to_super_imsm_volume(st
, dk
, fd
, devname
);
1575 dd
= malloc(sizeof(*dd
));
1578 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
1581 memset(dd
, 0, sizeof(*dd
));
1582 dd
->major
= major(stb
.st_rdev
);
1583 dd
->minor
= minor(stb
.st_rdev
);
1585 dd
->devname
= devname
? strdup(devname
) : NULL
;
1587 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
1590 Name
": failed to retrieve scsi serial, aborting\n");
1595 get_dev_size(fd
, NULL
, &size
);
1597 status
= USABLE_DISK
| SPARE_DISK
;
1598 strcpy((char *) dd
->disk
.serial
, (char *) dd
->serial
);
1599 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
1600 dd
->disk
.status
= __cpu_to_le32(status
);
1601 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
1602 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
1604 dd
->disk
.scsi_id
= __cpu_to_le32(0);
1606 if (st
->update_tail
) {
1607 dd
->next
= super
->add
;
1610 dd
->next
= super
->disks
;
1615 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
1617 /* spare records have their own family number and do not have any defined raid
1620 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
1622 struct imsm_super mpb_save
;
1623 struct imsm_super
*mpb
= super
->anchor
;
1628 mpb
->num_raid_devs
= 0;
1630 mpb
->mpb_size
= sizeof(struct imsm_super
);
1631 mpb
->generation_num
= __cpu_to_le32(1UL);
1633 for (d
= super
->disks
; d
; d
= d
->next
) {
1637 mpb
->disk
[0] = d
->disk
;
1638 sum
= __gen_imsm_checksum(mpb
);
1639 mpb
->family_num
= __cpu_to_le32(sum
);
1640 sum
= __gen_imsm_checksum(mpb
);
1641 mpb
->check_sum
= __cpu_to_le32(sum
);
1643 if (store_imsm_mpb(d
->fd
, super
)) {
1644 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1645 __func__
, d
->major
, d
->minor
, strerror(errno
));
1659 static int write_super_imsm(struct intel_super
*super
, int doclose
)
1661 struct imsm_super
*mpb
= super
->anchor
;
1667 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
1669 /* 'generation' is incremented everytime the metadata is written */
1670 generation
= __le32_to_cpu(mpb
->generation_num
);
1672 mpb
->generation_num
= __cpu_to_le32(generation
);
1674 for (d
= super
->disks
; d
; d
= d
->next
) {
1678 mpb
->disk
[d
->index
] = d
->disk
;
1679 mpb_size
+= sizeof(struct imsm_disk
);
1683 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1684 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1686 imsm_copy_dev(dev
, super
->dev_tbl
[i
]);
1687 mpb_size
+= sizeof_imsm_dev(dev
, 0);
1689 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
1690 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
1692 /* recalculate checksum */
1693 sum
= __gen_imsm_checksum(mpb
);
1694 mpb
->check_sum
= __cpu_to_le32(sum
);
1696 /* write the mpb for disks that compose raid devices */
1697 for (d
= super
->disks
; d
; d
= d
->next
) {
1700 if (store_imsm_mpb(d
->fd
, super
))
1701 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1702 __func__
, d
->major
, d
->minor
, strerror(errno
));
1710 return write_super_imsm_spares(super
, doclose
);
1715 static int create_array(struct supertype
*st
)
1718 struct imsm_update_create_array
*u
;
1719 struct intel_super
*super
= st
->sb
;
1720 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1722 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0);
1725 fprintf(stderr
, "%s: failed to allocate update buffer\n",
1730 u
->type
= update_create_array
;
1731 u
->dev_idx
= super
->current_vol
;
1732 imsm_copy_dev(&u
->dev
, dev
);
1733 append_metadata_update(st
, u
, len
);
1738 static int add_disk(struct supertype
*st
)
1740 struct intel_super
*super
= st
->sb
;
1742 struct imsm_update_add_disk
*u
;
1750 fprintf(stderr
, "%s: failed to allocate update buffer\n",
1755 u
->type
= update_add_disk
;
1756 append_metadata_update(st
, u
, len
);
1761 static int write_init_super_imsm(struct supertype
*st
)
1763 if (st
->update_tail
) {
1764 /* queue the recently created array / added disk
1765 * as a metadata update */
1766 struct intel_super
*super
= st
->sb
;
1770 /* determine if we are creating a volume or adding a disk */
1771 if (super
->current_vol
< 0) {
1772 /* in the add disk case we are running in mdmon
1773 * context, so don't close fd's
1775 return add_disk(st
);
1777 rv
= create_array(st
);
1779 for (d
= super
->disks
; d
; d
= d
->next
) {
1786 return write_super_imsm(st
->sb
, 1);
1789 static int store_zero_imsm(struct supertype
*st
, int fd
)
1791 unsigned long long dsize
;
1794 get_dev_size(fd
, NULL
, &dsize
);
1796 /* first block is stored on second to last sector of the disk */
1797 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
1800 if (posix_memalign(&buf
, 512, 512) != 0)
1803 memset(buf
, 0, 512);
1804 if (write(fd
, buf
, 512) != 512)
1809 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
1810 int layout
, int raiddisks
, int chunk
,
1811 unsigned long long size
, char *dev
,
1812 unsigned long long *freesize
,
1816 unsigned long long ldsize
;
1818 if (level
!= LEVEL_CONTAINER
)
1823 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
1826 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
1827 dev
, strerror(errno
));
1830 if (!get_dev_size(fd
, dev
, &ldsize
)) {
1836 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
1841 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
1842 * FIX ME add ahci details
1844 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
1845 int layout
, int raiddisks
, int chunk
,
1846 unsigned long long size
, char *dev
,
1847 unsigned long long *freesize
,
1851 struct intel_super
*super
= st
->sb
;
1853 unsigned long long pos
= 0;
1854 unsigned long long maxsize
;
1858 if (level
== LEVEL_CONTAINER
)
1861 if (level
== 1 && raiddisks
> 2) {
1863 fprintf(stderr
, Name
": imsm does not support more "
1864 "than 2 in a raid1 configuration\n");
1868 /* We must have the container info already read in. */
1873 /* General test: make sure there is space for
1874 * 'raiddisks' device extents of size 'size' at a given
1877 unsigned long long minsize
= size
*2 /* convert to blocks */;
1878 unsigned long long start_offset
= ~0ULL;
1881 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1882 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1887 e
= get_extents(super
, dl
);
1890 unsigned long long esize
;
1891 esize
= e
[i
].start
- pos
;
1892 if (esize
>= minsize
)
1894 if (found
&& start_offset
== ~0ULL) {
1897 } else if (found
&& pos
!= start_offset
) {
1901 pos
= e
[i
].start
+ e
[i
].size
;
1903 } while (e
[i
-1].size
);
1908 if (dcnt
< raiddisks
) {
1910 fprintf(stderr
, Name
": imsm: Not enough "
1911 "devices with space for this array "
1918 /* This device must be a member of the set */
1919 if (stat(dev
, &stb
) < 0)
1921 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
1923 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1924 if (dl
->major
== major(stb
.st_rdev
) &&
1925 dl
->minor
== minor(stb
.st_rdev
))
1930 fprintf(stderr
, Name
": %s is not in the "
1931 "same imsm set\n", dev
);
1934 e
= get_extents(super
, dl
);
1938 unsigned long long esize
;
1939 esize
= e
[i
].start
- pos
;
1940 if (esize
>= maxsize
)
1942 pos
= e
[i
].start
+ e
[i
].size
;
1944 } while (e
[i
-1].size
);
1945 *freesize
= maxsize
;
1950 int imsm_bbm_log_size(struct imsm_super
*mpb
)
1952 return __le32_to_cpu(mpb
->bbm_log_size
);
1955 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
1956 int raiddisks
, int chunk
, unsigned long long size
,
1957 char *dev
, unsigned long long *freesize
,
1963 /* if given unused devices create a container
1964 * if given given devices in a container create a member volume
1966 if (level
== LEVEL_CONTAINER
) {
1967 /* Must be a fresh device to add to a container */
1968 return validate_geometry_imsm_container(st
, level
, layout
,
1969 raiddisks
, chunk
, size
,
1975 /* creating in a given container */
1976 return validate_geometry_imsm_volume(st
, level
, layout
,
1977 raiddisks
, chunk
, size
,
1978 dev
, freesize
, verbose
);
1981 /* limit creation to the following levels */
1993 /* This device needs to be a device in an 'imsm' container */
1994 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
1998 Name
": Cannot create this array on device %s\n",
2003 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2005 fprintf(stderr
, Name
": Cannot open %s: %s\n",
2006 dev
, strerror(errno
));
2009 /* Well, it is in use by someone, maybe an 'imsm' container. */
2010 cfd
= open_container(fd
);
2014 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
2018 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2020 if (sra
&& sra
->array
.major_version
== -1 &&
2021 strcmp(sra
->text_version
, "imsm") == 0) {
2022 /* This is a member of a imsm container. Load the container
2023 * and try to create a volume
2025 struct intel_super
*super
;
2027 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
2029 st
->container_dev
= fd2devnum(cfd
);
2031 return validate_geometry_imsm_volume(st
, level
, layout
,
2037 } else /* may belong to another container */
2043 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
2045 /* Given a container loaded by load_super_imsm_all,
2046 * extract information about all the arrays into
2049 * For each imsm_dev create an mdinfo, fill it in,
2050 * then look for matching devices in super->disks
2051 * and create appropriate device mdinfo.
2053 struct intel_super
*super
= st
->sb
;
2054 struct imsm_super
*mpb
= super
->anchor
;
2055 struct mdinfo
*rest
= NULL
;
2058 /* do not assemble arrays that might have bad blocks */
2059 if (imsm_bbm_log_size(super
->anchor
)) {
2060 fprintf(stderr
, Name
": BBM log found in metadata. "
2061 "Cannot activate array(s).\n");
2065 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2066 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2067 struct imsm_vol
*vol
= &dev
->vol
;
2068 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2069 struct mdinfo
*this;
2072 this = malloc(sizeof(*this));
2073 memset(this, 0, sizeof(*this));
2076 this->array
.level
= get_imsm_raid_level(map
);
2077 this->array
.raid_disks
= map
->num_members
;
2078 this->array
.layout
= imsm_level_to_layout(this->array
.level
);
2079 this->array
.md_minor
= -1;
2080 this->array
.ctime
= 0;
2081 this->array
.utime
= 0;
2082 this->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2083 this->array
.state
= !vol
->dirty
;
2084 this->container_member
= i
;
2085 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2086 dev
->vol
.dirty
|| dev
->vol
.migr_state
)
2087 this->resync_start
= 0;
2089 this->resync_start
= ~0ULL;
2091 strncpy(this->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2092 this->name
[MAX_RAID_SERIAL_LEN
] = 0;
2094 sprintf(this->text_version
, "/%s/%d",
2095 devnum2devname(st
->container_dev
),
2096 this->container_member
);
2098 memset(this->uuid
, 0, sizeof(this->uuid
));
2100 this->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2102 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
2103 struct mdinfo
*info_d
;
2111 idx
= get_imsm_disk_idx(dev
, slot
);
2112 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
2113 for (d
= super
->disks
; d
; d
= d
->next
)
2114 if (d
->index
== idx
)
2120 s
= d
? __le32_to_cpu(d
->disk
.status
) : 0;
2121 if (s
& FAILED_DISK
)
2123 if (!(s
& USABLE_DISK
))
2125 if (ord
& IMSM_ORD_REBUILD
)
2129 * if we skip some disks the array will be assmebled degraded;
2130 * reset resync start to avoid a dirty-degraded situation
2132 * FIXME handle dirty degraded
2134 if (skip
&& !dev
->vol
.dirty
)
2135 this->resync_start
= ~0ULL;
2139 info_d
= malloc(sizeof(*info_d
));
2141 fprintf(stderr
, Name
": failed to allocate disk"
2142 " for volume %s\n", (char *) dev
->volume
);
2147 memset(info_d
, 0, sizeof(*info_d
));
2148 info_d
->next
= this->devs
;
2149 this->devs
= info_d
;
2151 info_d
->disk
.number
= d
->index
;
2152 info_d
->disk
.major
= d
->major
;
2153 info_d
->disk
.minor
= d
->minor
;
2154 info_d
->disk
.raid_disk
= slot
;
2156 this->array
.working_disks
++;
2158 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
2159 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2160 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2162 strcpy(info_d
->name
, d
->devname
);
2171 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
2174 struct intel_super
*super
= c
->sb
;
2175 struct imsm_super
*mpb
= super
->anchor
;
2177 if (atoi(inst
) >= mpb
->num_raid_devs
) {
2178 fprintf(stderr
, "%s: subarry index %d, out of range\n",
2179 __func__
, atoi(inst
));
2183 dprintf("imsm: open_new %s\n", inst
);
2184 a
->info
.container_member
= atoi(inst
);
2188 static __u8
imsm_check_degraded(struct intel_super
*super
, int n
, int failed
)
2190 struct imsm_dev
*dev
= get_imsm_dev(super
, n
);
2191 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2194 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
2195 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
2197 switch (get_imsm_raid_level(map
)) {
2199 return IMSM_T_STATE_FAILED
;
2202 if (failed
< map
->num_members
)
2203 return IMSM_T_STATE_DEGRADED
;
2205 return IMSM_T_STATE_FAILED
;
2210 * check to see if any mirrors have failed,
2211 * otherwise we are degraded
2213 int device_per_mirror
= 2; /* FIXME is this always the case?
2214 * and are they always adjacent?
2219 for (i
= 0; i
< map
->num_members
; i
++) {
2220 int idx
= get_imsm_disk_idx(dev
, i
);
2221 struct imsm_disk
*disk
= get_imsm_disk(super
, idx
);
2225 else if (__le32_to_cpu(disk
->status
) & FAILED_DISK
)
2228 if (r10fail
>= device_per_mirror
)
2229 return IMSM_T_STATE_FAILED
;
2231 /* reset 'r10fail' for next mirror set */
2232 if (!((i
+ 1) % device_per_mirror
))
2236 return IMSM_T_STATE_DEGRADED
;
2240 return IMSM_T_STATE_DEGRADED
;
2242 return IMSM_T_STATE_FAILED
;
2248 return map
->map_state
;
2251 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
2255 struct imsm_disk
*disk
;
2256 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2258 for (i
= 0; i
< map
->num_members
; i
++) {
2259 int idx
= get_imsm_disk_idx(dev
, i
);
2261 disk
= get_imsm_disk(super
, idx
);
2264 else if (__le32_to_cpu(disk
->status
) & FAILED_DISK
)
2266 else if (!(__le32_to_cpu(disk
->status
) & USABLE_DISK
))
2273 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
2275 int inst
= a
->info
.container_member
;
2276 struct intel_super
*super
= a
->container
->sb
;
2277 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2278 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2279 int dirty
= !consistent
;
2283 failed
= imsm_count_failed(super
, dev
);
2284 map_state
= imsm_check_degraded(super
, inst
, failed
);
2286 if (consistent
&& !dev
->vol
.dirty
&&
2287 (dev
->vol
.migr_state
|| map_state
!= IMSM_T_STATE_NORMAL
))
2288 a
->resync_start
= 0ULL;
2289 if (consistent
== 2 && a
->resync_start
!= ~0ULL)
2292 if (a
->resync_start
== ~0ULL) {
2293 /* complete recovery or initial resync */
2294 if (map
->map_state
!= map_state
) {
2295 dprintf("imsm: map_state %d: %d\n",
2297 map
->map_state
= map_state
;
2298 super
->updates_pending
++;
2300 if (dev
->vol
.migr_state
) {
2301 dprintf("imsm: mark resync complete\n");
2302 dev
->vol
.migr_state
= 0;
2303 dev
->vol
.migr_type
= 0;
2304 super
->updates_pending
++;
2306 } else if (!dev
->vol
.migr_state
) {
2307 dprintf("imsm: mark '%s' (%llu)\n",
2308 failed
? "rebuild" : "initializing", a
->resync_start
);
2309 /* mark that we are rebuilding */
2310 map
->map_state
= failed
? map_state
: IMSM_T_STATE_NORMAL
;
2311 dev
->vol
.migr_state
= 1;
2312 dev
->vol
.migr_type
= failed
? 1 : 0;
2314 a
->check_degraded
= 1;
2315 super
->updates_pending
++;
2318 /* mark dirty / clean */
2319 if (dirty
!= dev
->vol
.dirty
) {
2320 dprintf("imsm: mark '%s' (%llu)\n",
2321 dirty
? "dirty" : "clean", a
->resync_start
);
2322 dev
->vol
.dirty
= dirty
;
2323 super
->updates_pending
++;
2328 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
2330 int inst
= a
->info
.container_member
;
2331 struct intel_super
*super
= a
->container
->sb
;
2332 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2333 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2334 struct imsm_disk
*disk
;
2337 int new_failure
= 0;
2339 if (n
> map
->num_members
)
2340 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
2341 n
, map
->num_members
- 1);
2346 dprintf("imsm: set_disk %d:%x\n", n
, state
);
2348 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, n
));
2350 /* check for new failures */
2351 status
= __le32_to_cpu(disk
->status
);
2352 if ((state
& DS_FAULTY
) && !(status
& FAILED_DISK
)) {
2353 status
|= FAILED_DISK
;
2354 disk
->status
= __cpu_to_le32(status
);
2355 disk
->scsi_id
= __cpu_to_le32(~0UL);
2356 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
2358 super
->updates_pending
++;
2360 /* check if in_sync */
2361 if ((state
& DS_INSYNC
) && !(status
& USABLE_DISK
)) {
2362 status
|= USABLE_DISK
;
2363 disk
->status
= __cpu_to_le32(status
);
2364 super
->updates_pending
++;
2367 /* the number of failures have changed, count up 'failed' to determine
2368 * degraded / failed status
2370 if (new_failure
&& map
->map_state
!= IMSM_T_STATE_FAILED
)
2371 failed
= imsm_count_failed(super
, dev
);
2373 /* determine map_state based on failed or in_sync count */
2375 map
->map_state
= imsm_check_degraded(super
, inst
, failed
);
2376 else if (map
->map_state
== IMSM_T_STATE_DEGRADED
) {
2380 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2381 if (d
->curr_state
& DS_INSYNC
)
2384 if (working
== a
->info
.array
.raid_disks
) {
2385 map
->map_state
= IMSM_T_STATE_NORMAL
;
2386 dev
->vol
.migr_state
= 0;
2387 dev
->vol
.migr_type
= 0;
2388 super
->updates_pending
++;
2393 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
2395 struct imsm_super
*mpb
= super
->anchor
;
2396 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
2397 unsigned long long dsize
;
2398 unsigned long long sectors
;
2400 get_dev_size(fd
, NULL
, &dsize
);
2402 if (mpb_size
> 512) {
2403 /* -1 to account for anchor */
2404 sectors
= mpb_sectors(mpb
) - 1;
2406 /* write the extended mpb to the sectors preceeding the anchor */
2407 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
2410 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
2414 /* first block is stored on second to last sector of the disk */
2415 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2418 if (write(fd
, super
->buf
, 512) != 512)
2424 static void imsm_sync_metadata(struct supertype
*container
)
2426 struct intel_super
*super
= container
->sb
;
2428 if (!super
->updates_pending
)
2431 write_super_imsm(super
, 0);
2433 super
->updates_pending
= 0;
2436 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
2438 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2439 int i
= get_imsm_disk_idx(dev
, idx
);
2442 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2446 if (dl
&& __le32_to_cpu(dl
->disk
.status
) & FAILED_DISK
)
2450 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
2455 static struct dl
*imsm_add_spare(struct intel_super
*super
, int idx
, struct active_array
*a
)
2457 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2458 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2459 unsigned long long esize
;
2460 unsigned long long pos
;
2469 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2470 /* If in this array, skip */
2471 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2472 if (d
->disk
.major
== dl
->major
&&
2473 d
->disk
.minor
== dl
->minor
) {
2474 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
2480 /* skip marked in use or failed drives */
2481 status
= __le32_to_cpu(dl
->disk
.status
);
2482 if (status
& FAILED_DISK
|| status
& CONFIGURED_DISK
) {
2483 dprintf("%x:%x status ( %s%s)\n",
2484 dl
->major
, dl
->minor
,
2485 status
& FAILED_DISK
? "failed " : "",
2486 status
& CONFIGURED_DISK
? "configured " : "");
2490 /* Does this unused device have the requisite free space?
2491 * We need a->info.component_size sectors
2493 ex
= get_extents(super
, dl
);
2495 dprintf("cannot get extents\n");
2501 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
2504 /* check that we can start at pba_of_lba0 with
2505 * a->info.component_size of space
2507 esize
= ex
[j
].start
- pos
;
2508 if (array_start
>= pos
&&
2509 array_start
+ a
->info
.component_size
< ex
[j
].start
) {
2513 pos
= ex
[j
].start
+ ex
[j
].size
;
2516 } while (ex
[j
-1].size
);
2520 dprintf("%x:%x does not have %llu at %d\n",
2521 dl
->major
, dl
->minor
,
2522 a
->info
.component_size
,
2523 __le32_to_cpu(map
->pba_of_lba0
));
2533 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
2534 struct metadata_update
**updates
)
2537 * Find a device with unused free space and use it to replace a
2538 * failed/vacant region in an array. We replace failed regions one a
2539 * array at a time. The result is that a new spare disk will be added
2540 * to the first failed array and after the monitor has finished
2541 * propagating failures the remainder will be consumed.
2543 * FIXME add a capability for mdmon to request spares from another
2547 struct intel_super
*super
= a
->container
->sb
;
2548 int inst
= a
->info
.container_member
;
2549 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2550 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2551 int failed
= a
->info
.array
.raid_disks
;
2552 struct mdinfo
*rv
= NULL
;
2555 struct metadata_update
*mu
;
2557 struct imsm_update_activate_spare
*u
;
2561 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
2562 if ((d
->curr_state
& DS_FAULTY
) &&
2564 /* wait for Removal to happen */
2566 if (d
->state_fd
>= 0)
2570 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
2571 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
2572 if (imsm_check_degraded(super
, inst
, failed
) != IMSM_T_STATE_DEGRADED
)
2575 /* For each slot, if it is not working, find a spare */
2576 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
2577 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2578 if (d
->disk
.raid_disk
== i
)
2580 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
2581 if (d
&& (d
->state_fd
>= 0))
2585 * OK, this device needs recovery. Try to re-add the previous
2586 * occupant of this slot, if this fails add a new spare
2588 dl
= imsm_readd(super
, i
, a
);
2590 dl
= imsm_add_spare(super
, i
, a
);
2594 /* found a usable disk with enough space */
2595 di
= malloc(sizeof(*di
));
2596 memset(di
, 0, sizeof(*di
));
2598 /* dl->index will be -1 in the case we are activating a
2599 * pristine spare. imsm_process_update() will create a
2600 * new index in this case. Once a disk is found to be
2601 * failed in all member arrays it is kicked from the
2604 di
->disk
.number
= dl
->index
;
2606 /* (ab)use di->devs to store a pointer to the device
2609 di
->devs
= (struct mdinfo
*) dl
;
2611 di
->disk
.raid_disk
= i
;
2612 di
->disk
.major
= dl
->major
;
2613 di
->disk
.minor
= dl
->minor
;
2615 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2616 di
->component_size
= a
->info
.component_size
;
2617 di
->container_member
= inst
;
2621 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
2622 i
, di
->data_offset
);
2628 /* No spares found */
2630 /* Now 'rv' has a list of devices to return.
2631 * Create a metadata_update record to update the
2632 * disk_ord_tbl for the array
2634 mu
= malloc(sizeof(*mu
));
2635 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
2637 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
2638 mu
->next
= *updates
;
2639 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
2641 for (di
= rv
; di
; di
= di
->next
) {
2642 u
->type
= update_activate_spare
;
2643 u
->dl
= (struct dl
*) di
->devs
;
2645 u
->slot
= di
->disk
.raid_disk
;
2656 static int disks_overlap(struct imsm_dev
*d1
, struct imsm_dev
*d2
)
2658 struct imsm_map
*m1
= get_imsm_map(d1
, 0);
2659 struct imsm_map
*m2
= get_imsm_map(d2
, 0);
2664 for (i
= 0; i
< m1
->num_members
; i
++) {
2665 idx
= get_imsm_disk_idx(d1
, i
);
2666 for (j
= 0; j
< m2
->num_members
; j
++)
2667 if (idx
== get_imsm_disk_idx(d2
, j
))
2674 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
);
2676 static void imsm_process_update(struct supertype
*st
,
2677 struct metadata_update
*update
)
2680 * crack open the metadata_update envelope to find the update record
2681 * update can be one of:
2682 * update_activate_spare - a spare device has replaced a failed
2683 * device in an array, update the disk_ord_tbl. If this disk is
2684 * present in all member arrays then also clear the SPARE_DISK
2687 struct intel_super
*super
= st
->sb
;
2688 struct imsm_super
*mpb
;
2689 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
2691 /* update requires a larger buf but the allocation failed */
2692 if (super
->next_len
&& !super
->next_buf
) {
2693 super
->next_len
= 0;
2697 if (super
->next_buf
) {
2698 memcpy(super
->next_buf
, super
->buf
, super
->len
);
2700 super
->len
= super
->next_len
;
2701 super
->buf
= super
->next_buf
;
2703 super
->next_len
= 0;
2704 super
->next_buf
= NULL
;
2707 mpb
= super
->anchor
;
2710 case update_activate_spare
: {
2711 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
2712 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
2713 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2714 struct active_array
*a
;
2715 struct imsm_disk
*disk
;
2722 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2727 fprintf(stderr
, "error: imsm_activate_spare passed "
2728 "an unknown disk (index: %d serial: %s)\n",
2729 u
->dl
->index
, u
->dl
->serial
);
2733 super
->updates_pending
++;
2735 /* adding a pristine spare, assign a new index */
2736 if (dl
->index
< 0) {
2737 dl
->index
= super
->anchor
->num_disks
;
2738 super
->anchor
->num_disks
++;
2740 victim
= get_imsm_disk_idx(dev
, u
->slot
);
2741 map
->disk_ord_tbl
[u
->slot
] = __cpu_to_le32(dl
->index
);
2743 status
= __le32_to_cpu(disk
->status
);
2744 status
|= CONFIGURED_DISK
;
2745 status
&= ~(SPARE_DISK
| USABLE_DISK
);
2746 disk
->status
= __cpu_to_le32(status
);
2748 /* count arrays using the victim in the metadata */
2750 for (a
= st
->arrays
; a
; a
= a
->next
) {
2751 dev
= get_imsm_dev(super
, a
->info
.container_member
);
2752 for (i
= 0; i
< map
->num_members
; i
++)
2753 if (victim
== get_imsm_disk_idx(dev
, i
))
2757 /* clear some flags if the victim is no longer being
2762 for (dlp
= &super
->disks
; *dlp
; )
2763 if ((*dlp
)->index
== victim
)
2765 disk
= &(*dlp
)->disk
;
2766 status
= __le32_to_cpu(disk
->status
);
2767 status
&= ~(CONFIGURED_DISK
| USABLE_DISK
);
2768 disk
->status
= __cpu_to_le32(status
);
2769 /* We know that 'manager' isn't touching anything,
2772 imsm_delete(super
, dlp
);
2776 case update_create_array
: {
2777 /* someone wants to create a new array, we need to be aware of
2778 * a few races/collisions:
2779 * 1/ 'Create' called by two separate instances of mdadm
2780 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
2781 * devices that have since been assimilated via
2783 * In the event this update can not be carried out mdadm will
2784 * (FIX ME) notice that its update did not take hold.
2786 struct imsm_update_create_array
*u
= (void *) update
->buf
;
2787 struct imsm_dev
*dev
;
2788 struct imsm_map
*map
, *new_map
;
2789 unsigned long long start
, end
;
2790 unsigned long long new_start
, new_end
;
2794 /* handle racing creates: first come first serve */
2795 if (u
->dev_idx
< mpb
->num_raid_devs
) {
2796 dprintf("%s: subarray %d already defined\n",
2797 __func__
, u
->dev_idx
);
2801 /* check update is next in sequence */
2802 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
2803 dprintf("%s: can not create array %d expected index %d\n",
2804 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
2808 new_map
= get_imsm_map(&u
->dev
, 0);
2809 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
2810 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
2812 /* handle activate_spare versus create race:
2813 * check to make sure that overlapping arrays do not include
2816 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2817 dev
= get_imsm_dev(super
, i
);
2818 map
= get_imsm_map(dev
, 0);
2819 start
= __le32_to_cpu(map
->pba_of_lba0
);
2820 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
2821 if ((new_start
>= start
&& new_start
<= end
) ||
2822 (start
>= new_start
&& start
<= new_end
))
2824 if (overlap
&& disks_overlap(dev
, &u
->dev
)) {
2825 dprintf("%s: arrays overlap\n", __func__
);
2829 /* check num_members sanity */
2830 if (new_map
->num_members
> mpb
->num_disks
) {
2831 dprintf("%s: num_disks out of range\n", __func__
);
2835 /* check that prepare update was successful */
2836 if (!update
->space
) {
2837 dprintf("%s: prepare update failed\n", __func__
);
2841 super
->updates_pending
++;
2842 dev
= update
->space
;
2843 map
= get_imsm_map(dev
, 0);
2844 update
->space
= NULL
;
2845 imsm_copy_dev(dev
, &u
->dev
);
2846 map
= get_imsm_map(dev
, 0);
2847 super
->dev_tbl
[u
->dev_idx
] = dev
;
2848 mpb
->num_raid_devs
++;
2851 for (i
= 0; i
< map
->num_members
; i
++) {
2852 struct imsm_disk
*disk
;
2855 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
2856 status
= __le32_to_cpu(disk
->status
);
2857 status
|= CONFIGURED_DISK
;
2858 status
&= ~SPARE_DISK
;
2859 disk
->status
= __cpu_to_le32(status
);
2863 case update_add_disk
:
2865 /* we may be able to repair some arrays if disks are
2868 struct active_array
*a
;
2869 for (a
= st
->arrays
; a
; a
= a
->next
)
2870 a
->check_degraded
= 1;
2872 /* check if we can add / replace some disks in the
2874 while (super
->add
) {
2875 struct dl
**dlp
, *dl
, *al
;
2877 super
->add
= al
->next
;
2878 for (dlp
= &super
->disks
; *dlp
; ) {
2879 if (memcmp(al
->serial
, (*dlp
)->serial
,
2880 MAX_RAID_SERIAL_LEN
) == 0) {
2882 *dlp
= (*dlp
)->next
;
2883 __free_imsm_disk(dl
);
2886 dlp
= &(*dlp
)->next
;
2888 al
->next
= super
->disks
;
2896 static void imsm_prepare_update(struct supertype
*st
,
2897 struct metadata_update
*update
)
2900 * Allocate space to hold new disk entries, raid-device entries or a new
2901 * mpb if necessary. The manager synchronously waits for updates to
2902 * complete in the monitor, so new mpb buffers allocated here can be
2903 * integrated by the monitor thread without worrying about live pointers
2904 * in the manager thread.
2906 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
2907 struct intel_super
*super
= st
->sb
;
2908 struct imsm_super
*mpb
= super
->anchor
;
2913 case update_create_array
: {
2914 struct imsm_update_create_array
*u
= (void *) update
->buf
;
2916 len
= sizeof_imsm_dev(&u
->dev
, 1);
2917 update
->space
= malloc(len
);
2924 /* check if we need a larger metadata buffer */
2925 if (super
->next_buf
)
2926 buf_len
= super
->next_len
;
2928 buf_len
= super
->len
;
2930 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
2931 /* ok we need a larger buf than what is currently allocated
2932 * if this allocation fails process_update will notice that
2933 * ->next_len is set and ->next_buf is NULL
2935 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
2936 if (super
->next_buf
)
2937 free(super
->next_buf
);
2939 super
->next_len
= buf_len
;
2940 if (posix_memalign(&super
->next_buf
, buf_len
, 512) != 0)
2941 super
->next_buf
= NULL
;
2945 /* must be called while manager is quiesced */
2946 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
)
2948 struct imsm_super
*mpb
= super
->anchor
;
2949 struct dl
*dl
= *dlp
;
2951 struct imsm_dev
*dev
;
2952 struct imsm_map
*map
;
2955 dprintf("%s: deleting device %x:%x from imsm_super\n",
2956 __func__
, dl
->major
, dl
->minor
);
2958 /* shift all indexes down one */
2959 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
2960 if (iter
->index
> dl
->index
)
2963 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2964 dev
= get_imsm_dev(super
, i
);
2965 map
= get_imsm_map(dev
, 0);
2967 for (j
= 0; j
< map
->num_members
; j
++) {
2968 int idx
= get_imsm_disk_idx(dev
, j
);
2970 if (idx
> dl
->index
)
2971 map
->disk_ord_tbl
[j
] = __cpu_to_le32(idx
- 1);
2976 super
->updates_pending
++;
2977 *dlp
= (*dlp
)->next
;
2978 __free_imsm_disk(dl
);
2981 struct superswitch super_imsm
= {
2983 .examine_super
= examine_super_imsm
,
2984 .brief_examine_super
= brief_examine_super_imsm
,
2985 .detail_super
= detail_super_imsm
,
2986 .brief_detail_super
= brief_detail_super_imsm
,
2987 .write_init_super
= write_init_super_imsm
,
2989 .match_home
= match_home_imsm
,
2990 .uuid_from_super
= uuid_from_super_imsm
,
2991 .getinfo_super
= getinfo_super_imsm
,
2992 .update_super
= update_super_imsm
,
2994 .avail_size
= avail_size_imsm
,
2996 .compare_super
= compare_super_imsm
,
2998 .load_super
= load_super_imsm
,
2999 .init_super
= init_super_imsm
,
3000 .add_to_super
= add_to_super_imsm
,
3001 .store_super
= store_zero_imsm
,
3002 .free_super
= free_super_imsm
,
3003 .match_metadata_desc
= match_metadata_desc_imsm
,
3004 .container_content
= container_content_imsm
,
3006 .validate_geometry
= validate_geometry_imsm
,
3010 .open_new
= imsm_open_new
,
3011 .load_super
= load_super_imsm
,
3012 .set_array_state
= imsm_set_array_state
,
3013 .set_disk
= imsm_set_disk
,
3014 .sync_metadata
= imsm_sync_metadata
,
3015 .activate_spare
= imsm_activate_spare
,
3016 .process_update
= imsm_process_update
,
3017 .prepare_update
= imsm_prepare_update
,