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_disk_idx(struct imsm_map
*map
, int slot
)
350 __u32
*ord_tbl
= &map
->disk_ord_tbl
[slot
];
352 /* top byte identifies disk under rebuild
353 * why not just use the USABLE bit... oh well.
355 return __le32_to_cpu(*ord_tbl
& ~(0xff << 24));
358 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
360 struct imsm_map
*map
;
362 if (dev
->vol
.migr_state
)
363 map
= get_imsm_map(dev
, 1);
365 map
= get_imsm_map(dev
, 0);
367 return map
->disk_ord_tbl
[slot
];
370 static int get_imsm_raid_level(struct imsm_map
*map
)
372 if (map
->raid_level
== 1) {
373 if (map
->num_members
== 2)
379 return map
->raid_level
;
382 static int cmp_extent(const void *av
, const void *bv
)
384 const struct extent
*a
= av
;
385 const struct extent
*b
= bv
;
386 if (a
->start
< b
->start
)
388 if (a
->start
> b
->start
)
393 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
395 /* find a list of used extents on the given physical device */
396 struct extent
*rv
, *e
;
400 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
401 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
402 struct imsm_map
*map
= get_imsm_map(dev
, 0);
404 for (j
= 0; j
< map
->num_members
; j
++) {
405 __u32 index
= get_imsm_disk_idx(map
, j
);
407 if (index
== dl
->index
)
411 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
416 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
417 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
418 struct imsm_map
*map
= get_imsm_map(dev
, 0);
420 for (j
= 0; j
< map
->num_members
; j
++) {
421 __u32 index
= get_imsm_disk_idx(map
, j
);
423 if (index
== dl
->index
) {
424 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
425 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
430 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
432 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) -
433 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
439 static void print_imsm_dev(struct imsm_dev
*dev
, int index
)
443 struct imsm_map
*map
= get_imsm_map(dev
, 0);
446 printf("[%s]:\n", dev
->volume
);
447 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
448 printf(" Members : %d\n", map
->num_members
);
449 for (slot
= 0; slot
< map
->num_members
; slot
++)
450 if (index
== get_imsm_disk_idx(map
, slot
))
452 if (slot
< map
->num_members
)
453 printf(" This Slot : %d\n", slot
);
455 printf(" This Slot : ?\n");
456 sz
= __le32_to_cpu(dev
->size_high
);
458 sz
+= __le32_to_cpu(dev
->size_low
);
459 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
460 human_size(sz
* 512));
461 sz
= __le32_to_cpu(map
->blocks_per_member
);
462 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
463 human_size(sz
* 512));
464 printf(" Sector Offset : %u\n",
465 __le32_to_cpu(map
->pba_of_lba0
));
466 printf(" Num Stripes : %u\n",
467 __le32_to_cpu(map
->num_data_stripes
));
468 printf(" Chunk Size : %u KiB\n",
469 __le16_to_cpu(map
->blocks_per_strip
) / 2);
470 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
471 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle");
472 if (dev
->vol
.migr_state
)
473 printf(": %s", dev
->vol
.migr_type
? "rebuilding" : "initializing");
475 printf(" Map State : %s", map_state_str
[map
->map_state
]);
476 if (dev
->vol
.migr_state
) {
477 struct imsm_map
*map
= get_imsm_map(dev
, 1);
478 printf(", %s", map_state_str
[map
->map_state
]);
481 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
484 static void print_imsm_disk(struct imsm_super
*mpb
, int index
)
486 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
487 char str
[MAX_RAID_SERIAL_LEN
];
495 snprintf(str
, MAX_RAID_SERIAL_LEN
, "%s", disk
->serial
);
496 printf(" Disk%02d Serial : %s\n", index
, str
);
497 s
= __le32_to_cpu(disk
->status
);
498 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
499 s
&CONFIGURED_DISK
? " active" : "",
500 s
&FAILED_DISK
? " failed" : "",
501 s
&USABLE_DISK
? " usable" : "");
502 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
503 sz
= __le32_to_cpu(disk
->total_blocks
) -
504 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
* mpb
->num_raid_devs
);
505 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
506 human_size(sz
* 512));
509 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
511 struct intel_super
*super
= st
->sb
;
512 struct imsm_super
*mpb
= super
->anchor
;
513 char str
[MAX_SIGNATURE_LENGTH
];
517 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
518 printf(" Magic : %s\n", str
);
519 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
520 printf(" Version : %s\n", get_imsm_version(mpb
));
521 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
522 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
523 sum
= __le32_to_cpu(mpb
->check_sum
);
524 printf(" Checksum : %08x %s\n", sum
,
525 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
526 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
527 printf(" Disks : %d\n", mpb
->num_disks
);
528 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
529 print_imsm_disk(mpb
, super
->disks
->index
);
530 if (super
->bbm_log
) {
531 struct bbm_log
*log
= super
->bbm_log
;
534 printf("Bad Block Management Log:\n");
535 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
536 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
537 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
538 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
539 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
541 for (i
= 0; i
< mpb
->num_raid_devs
; i
++)
542 print_imsm_dev(__get_imsm_dev(mpb
, i
), super
->disks
->index
);
543 for (i
= 0; i
< mpb
->num_disks
; i
++) {
544 if (i
== super
->disks
->index
)
546 print_imsm_disk(mpb
, i
);
550 static void brief_examine_super_imsm(struct supertype
*st
)
552 printf("ARRAY /dev/imsm metadata=imsm\n");
555 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
557 printf("%s\n", __FUNCTION__
);
560 static void brief_detail_super_imsm(struct supertype
*st
)
562 printf("%s\n", __FUNCTION__
);
566 static int match_home_imsm(struct supertype
*st
, char *homehost
)
568 printf("%s\n", __FUNCTION__
);
573 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
575 /* imsm does not track uuid's so just make sure we never return
576 * the same value twice to break uuid matching in Manage_subdevs
577 * FIXME what about the use of uuid's with bitmap's?
579 static int dummy_id
= 0;
581 uuid
[0] = dummy_id
++;
586 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
588 __u8
*v
= get_imsm_version(mpb
);
589 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
590 char major
[] = { 0, 0, 0 };
591 char minor
[] = { 0 ,0, 0 };
592 char patch
[] = { 0, 0, 0 };
593 char *ver_parse
[] = { major
, minor
, patch
};
597 while (*v
!= '\0' && v
< end
) {
598 if (*v
!= '.' && j
< 2)
599 ver_parse
[i
][j
++] = *v
;
607 *m
= strtol(minor
, NULL
, 0);
608 *p
= strtol(patch
, NULL
, 0);
612 static int imsm_level_to_layout(int level
)
620 return ALGORITHM_LEFT_ASYMMETRIC
;
622 return 0x102; //FIXME is this correct?
627 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
629 struct intel_super
*super
= st
->sb
;
630 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
631 struct imsm_map
*map
= get_imsm_map(dev
, 0);
633 info
->container_member
= super
->current_vol
;
634 info
->array
.raid_disks
= map
->num_members
;
635 info
->array
.level
= get_imsm_raid_level(map
);
636 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
637 info
->array
.md_minor
= -1;
638 info
->array
.ctime
= 0;
639 info
->array
.utime
= 0;
640 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
* 512);
642 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
643 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
645 info
->disk
.major
= 0;
646 info
->disk
.minor
= 0;
648 sprintf(info
->text_version
, "/%s/%d",
649 devnum2devname(st
->container_dev
),
650 info
->container_member
);
654 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
656 struct intel_super
*super
= st
->sb
;
657 struct imsm_disk
*disk
;
660 if (super
->current_vol
>= 0) {
661 getinfo_super_imsm_volume(st
, info
);
665 /* Set raid_disks to zero so that Assemble will always pull in valid
668 info
->array
.raid_disks
= 0;
669 info
->array
.level
= LEVEL_CONTAINER
;
670 info
->array
.layout
= 0;
671 info
->array
.md_minor
= -1;
672 info
->array
.ctime
= 0; /* N/A for imsm */
673 info
->array
.utime
= 0;
674 info
->array
.chunk_size
= 0;
676 info
->disk
.major
= 0;
677 info
->disk
.minor
= 0;
678 info
->disk
.raid_disk
= -1;
679 info
->reshape_active
= 0;
680 strcpy(info
->text_version
, "imsm");
681 info
->disk
.number
= -1;
682 info
->disk
.state
= 0;
685 disk
= &super
->disks
->disk
;
686 info
->disk
.number
= super
->disks
->index
;
687 info
->disk
.raid_disk
= super
->disks
->index
;
688 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) -
689 (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
690 info
->component_size
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
691 s
= __le32_to_cpu(disk
->status
);
692 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
693 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
694 info
->disk
.state
|= s
& USABLE_DISK
? (1 << MD_DISK_SYNC
) : 0;
698 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
699 char *update
, char *devname
, int verbose
,
700 int uuid_set
, char *homehost
)
704 /* For 'assemble' and 'force' we need to return non-zero if any
705 * change was made. For others, the return value is ignored.
706 * Update options are:
707 * force-one : This device looks a bit old but needs to be included,
708 * update age info appropriately.
709 * assemble: clear any 'faulty' flag to allow this device to
711 * force-array: Array is degraded but being forced, mark it clean
712 * if that will be needed to assemble it.
714 * newdev: not used ????
715 * grow: Array has gained a new device - this is currently for
717 * resync: mark as dirty so a resync will happen.
718 * name: update the name - preserving the homehost
720 * Following are not relevant for this imsm:
721 * sparc2.2 : update from old dodgey metadata
722 * super-minor: change the preferred_minor number
723 * summaries: update redundant counters.
724 * uuid: Change the uuid of the array to match watch is given
725 * homehost: update the recorded homehost
726 * _reshape_progress: record new reshape_progress position.
729 //struct intel_super *super = st->sb;
730 //struct imsm_super *mpb = super->mpb;
732 if (strcmp(update
, "grow") == 0) {
734 if (strcmp(update
, "resync") == 0) {
735 /* dev->vol.dirty = 1; */
738 /* IMSM has no concept of UUID or homehost */
743 static size_t disks_to_mpb_size(int disks
)
747 size
= sizeof(struct imsm_super
);
748 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
749 size
+= 2 * sizeof(struct imsm_dev
);
750 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
751 size
+= (4 - 2) * sizeof(struct imsm_map
);
752 /* 4 possible disk_ord_tbl's */
753 size
+= 4 * (disks
- 1) * sizeof(__u32
);
758 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
760 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
763 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
766 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
770 * 0 same, or first was empty, and second was copied
771 * 1 second had wrong number
775 struct intel_super
*first
= st
->sb
;
776 struct intel_super
*sec
= tst
->sb
;
784 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
787 /* if an anchor does not have num_raid_devs set then it is a free
790 if (first
->anchor
->num_raid_devs
> 0 &&
791 sec
->anchor
->num_raid_devs
> 0) {
792 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
799 static void fd2devname(int fd
, char *name
)
808 if (fstat(fd
, &st
) != 0)
810 sprintf(path
, "/sys/dev/block/%d:%d",
811 major(st
.st_rdev
), minor(st
.st_rdev
));
813 rv
= readlink(path
, dname
, sizeof(dname
));
818 nm
= strrchr(dname
, '/');
820 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
824 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
826 static int imsm_read_serial(int fd
, char *devname
,
827 __u8 serial
[MAX_RAID_SERIAL_LEN
])
829 unsigned char scsi_serial
[255];
834 memset(scsi_serial
, 0, sizeof(scsi_serial
));
836 if (imsm_env_devname_as_serial()) {
837 char name
[MAX_RAID_SERIAL_LEN
];
839 fd2devname(fd
, name
);
840 strcpy((char *) serial
, name
);
844 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
849 Name
": Failed to retrieve serial for %s\n",
854 rsp_len
= scsi_serial
[3];
855 for (i
= 0, cnt
= 0; i
< rsp_len
; i
++) {
856 if (!isspace(scsi_serial
[4 + i
]))
857 serial
[cnt
++] = scsi_serial
[4 + i
];
858 if (cnt
== MAX_RAID_SERIAL_LEN
)
862 serial
[MAX_RAID_SERIAL_LEN
- 1] = '\0';
868 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
875 __u8 serial
[MAX_RAID_SERIAL_LEN
];
877 rv
= imsm_read_serial(fd
, devname
, serial
);
882 /* check if this is a disk we have seen before. it may be a spare in
883 * super->disks while the current anchor believes it is a raid member,
884 * check if we need to update dl->index
886 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
887 if (memcmp(dl
->serial
, serial
, MAX_RAID_SERIAL_LEN
) == 0)
891 dl
= malloc(sizeof(*dl
));
898 Name
": failed to allocate disk buffer for %s\n",
905 dl
->major
= major(stb
.st_rdev
);
906 dl
->minor
= minor(stb
.st_rdev
);
907 dl
->next
= super
->disks
;
908 dl
->fd
= keep_fd
? fd
: -1;
909 dl
->devname
= devname
? strdup(devname
) : NULL
;
910 strncpy((char *) dl
->serial
, (char *) serial
, MAX_RAID_SERIAL_LEN
);
912 } else if (keep_fd
) {
917 /* look up this disk's index in the current anchor */
918 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
919 struct imsm_disk
*disk_iter
;
921 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
923 if (memcmp(disk_iter
->serial
, dl
->serial
,
924 MAX_RAID_SERIAL_LEN
) == 0) {
927 dl
->disk
= *disk_iter
;
928 status
= __le32_to_cpu(dl
->disk
.status
);
929 /* only set index on disks that are a member of a
930 * populated contianer, i.e. one with raid_devs
932 if (status
& FAILED_DISK
)
934 else if (status
& SPARE_DISK
)
949 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
951 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
954 static void dup_map(struct imsm_dev
*dev
)
956 struct imsm_map
*dest
= get_imsm_map(dev
, 1);
957 struct imsm_map
*src
= get_imsm_map(dev
, 0);
959 memcpy(dest
, src
, sizeof_imsm_map(src
));
962 static int parse_raid_devices(struct intel_super
*super
)
965 struct imsm_dev
*dev_new
;
966 size_t len
, len_migr
;
967 size_t space_needed
= 0;
968 struct imsm_super
*mpb
= super
->anchor
;
970 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
971 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
973 len
= sizeof_imsm_dev(dev_iter
, 0);
974 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
976 space_needed
+= len_migr
- len
;
978 dev_new
= malloc(len_migr
);
981 imsm_copy_dev(dev_new
, dev_iter
);
982 super
->dev_tbl
[i
] = dev_new
;
985 /* ensure that super->buf is large enough when all raid devices
988 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
991 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
992 if (posix_memalign(&buf
, 512, len
) != 0)
995 memcpy(buf
, super
->buf
, len
);
1004 /* retrieve a pointer to the bbm log which starts after all raid devices */
1005 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1009 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1011 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1017 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1019 /* load_imsm_mpb - read matrix metadata
1020 * allocates super->mpb to be freed by free_super
1022 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1024 unsigned long long dsize
;
1025 unsigned long long sectors
;
1027 struct imsm_super
*anchor
;
1031 get_dev_size(fd
, NULL
, &dsize
);
1033 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1036 Name
": Cannot seek to anchor block on %s: %s\n",
1037 devname
, strerror(errno
));
1041 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1044 Name
": Failed to allocate imsm anchor buffer"
1045 " on %s\n", devname
);
1048 if (read(fd
, anchor
, 512) != 512) {
1051 Name
": Cannot read anchor block on %s: %s\n",
1052 devname
, strerror(errno
));
1057 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1060 Name
": no IMSM anchor on %s\n", devname
);
1065 __free_imsm(super
, 0);
1066 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1067 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1070 Name
": unable to allocate %zu byte mpb buffer\n",
1075 memcpy(super
->buf
, anchor
, 512);
1077 sectors
= mpb_sectors(anchor
) - 1;
1080 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1082 rc
= parse_raid_devices(super
);
1086 /* read the extended mpb */
1087 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1090 Name
": Cannot seek to extended mpb on %s: %s\n",
1091 devname
, strerror(errno
));
1095 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1098 Name
": Cannot read extended mpb on %s: %s\n",
1099 devname
, strerror(errno
));
1103 check_sum
= __gen_imsm_checksum(super
->anchor
);
1104 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1107 Name
": IMSM checksum %x != %x on %s\n",
1108 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1113 /* FIXME the BBM log is disk specific so we cannot use this global
1114 * buffer for all disks. Ok for now since we only look at the global
1115 * bbm_log_size parameter to gate assembly
1117 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1119 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1121 rc
= parse_raid_devices(super
);
1126 static void __free_imsm_disk(struct dl
*d
)
1135 static void free_imsm_disks(struct intel_super
*super
)
1137 while (super
->disks
) {
1138 struct dl
*d
= super
->disks
;
1140 super
->disks
= d
->next
;
1141 __free_imsm_disk(d
);
1145 /* free all the pieces hanging off of a super pointer */
1146 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1155 free_imsm_disks(super
);
1156 for (i
= 0; i
< IMSM_MAX_RAID_DEVS
; i
++)
1157 if (super
->dev_tbl
[i
]) {
1158 free(super
->dev_tbl
[i
]);
1159 super
->dev_tbl
[i
] = NULL
;
1163 static void free_imsm(struct intel_super
*super
)
1165 __free_imsm(super
, 1);
1169 static void free_super_imsm(struct supertype
*st
)
1171 struct intel_super
*super
= st
->sb
;
1180 static struct intel_super
*alloc_super(int creating_imsm
)
1182 struct intel_super
*super
= malloc(sizeof(*super
));
1185 memset(super
, 0, sizeof(*super
));
1186 super
->creating_imsm
= creating_imsm
;
1187 super
->current_vol
= -1;
1194 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
1195 char *devname
, int keep_fd
)
1198 struct intel_super
*super
;
1199 struct mdinfo
*sd
, *best
= NULL
;
1206 /* check if this disk is a member of an active array */
1207 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
1211 if (sra
->array
.major_version
!= -1 ||
1212 sra
->array
.minor_version
!= -2 ||
1213 strcmp(sra
->text_version
, "imsm") != 0)
1216 super
= alloc_super(0);
1220 /* find the most up to date disk in this array, skipping spares */
1221 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1222 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1223 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1228 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1232 if (super
->anchor
->num_raid_devs
== 0)
1235 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
1236 if (!best
|| gen
> bestgen
) {
1251 /* load the most up to date anchor */
1252 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
1253 dfd
= dev_open(nm
, O_RDONLY
);
1258 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1265 /* re-parse the disk list with the current anchor */
1266 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1267 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1268 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1273 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
1278 if (st
->subarray
[0]) {
1279 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
1280 super
->current_vol
= atoi(st
->subarray
);
1286 st
->container_dev
= fd2devnum(fd
);
1287 if (st
->ss
== NULL
) {
1288 st
->ss
= &super_imsm
;
1289 st
->minor_version
= 0;
1290 st
->max_devs
= IMSM_MAX_DEVICES
;
1297 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
1299 struct intel_super
*super
;
1303 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
1306 if (st
->subarray
[0])
1307 return 1; /* FIXME */
1309 super
= alloc_super(0);
1312 Name
": malloc of %zu failed.\n",
1317 rv
= load_imsm_mpb(fd
, super
, devname
);
1322 Name
": Failed to load all information "
1323 "sections on %s\n", devname
);
1329 if (st
->ss
== NULL
) {
1330 st
->ss
= &super_imsm
;
1331 st
->minor_version
= 0;
1332 st
->max_devs
= IMSM_MAX_DEVICES
;
1338 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
1340 if (info
->level
== 1)
1342 return info
->chunk_size
>> 9;
1345 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
1349 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
1350 if (info
->level
== 1)
1356 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
1358 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
1361 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
1362 unsigned long long size
, char *name
,
1363 char *homehost
, int *uuid
)
1365 /* We are creating a volume inside a pre-existing container.
1366 * so st->sb is already set.
1368 struct intel_super
*super
= st
->sb
;
1369 struct imsm_super
*mpb
= super
->anchor
;
1370 struct imsm_dev
*dev
;
1371 struct imsm_vol
*vol
;
1372 struct imsm_map
*map
;
1373 int idx
= mpb
->num_raid_devs
;
1375 unsigned long long array_blocks
;
1377 size_t size_old
, size_new
;
1379 if (mpb
->num_raid_devs
>= 2) {
1380 fprintf(stderr
, Name
": This imsm-container already has the "
1381 "maximum of 2 volumes\n");
1385 /* ensure the mpb is large enough for the new data */
1386 size_old
= __le32_to_cpu(mpb
->mpb_size
);
1387 size_new
= disks_to_mpb_size(info
->nr_disks
);
1388 if (size_new
> size_old
) {
1390 size_t size_round
= ROUND_UP(size_new
, 512);
1392 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
1393 fprintf(stderr
, Name
": could not allocate new mpb\n");
1396 memcpy(mpb_new
, mpb
, size_old
);
1399 super
->anchor
= mpb_new
;
1400 mpb
->mpb_size
= __cpu_to_le32(size_new
);
1401 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
1403 super
->current_vol
= idx
;
1404 /* when creating the first raid device in this container set num_disks
1405 * to zero, i.e. delete this spare and add raid member devices in
1406 * add_to_super_imsm_volume()
1408 if (super
->current_vol
== 0)
1410 sprintf(st
->subarray
, "%d", idx
);
1411 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
1413 fprintf(stderr
, Name
": could not allocate raid device\n");
1416 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
1417 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
1418 info
->layout
, info
->chunk_size
,
1420 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
1421 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
1422 dev
->status
= __cpu_to_le32(0);
1423 dev
->reserved_blocks
= __cpu_to_le32(0);
1425 vol
->migr_state
= 0;
1428 for (i
= 0; i
< idx
; i
++) {
1429 struct imsm_dev
*prev
= get_imsm_dev(super
, i
);
1430 struct imsm_map
*pmap
= get_imsm_map(prev
, 0);
1432 offset
+= __le32_to_cpu(pmap
->blocks_per_member
);
1433 offset
+= IMSM_RESERVED_SECTORS
;
1435 map
= get_imsm_map(dev
, 0);
1436 map
->pba_of_lba0
= __cpu_to_le32(offset
);
1437 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
1438 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
1439 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
1440 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
1441 IMSM_T_STATE_NORMAL
;
1443 if (info
->level
== 1 && info
->raid_disks
> 2) {
1444 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
1445 "in a raid1 volume\n");
1448 if (info
->level
== 10)
1449 map
->raid_level
= 1;
1451 map
->raid_level
= info
->level
;
1453 map
->num_members
= info
->raid_disks
;
1454 for (i
= 0; i
< map
->num_members
; i
++) {
1455 /* initialized in add_to_super */
1456 map
->disk_ord_tbl
[i
] = __cpu_to_le32(0);
1458 mpb
->num_raid_devs
++;
1459 super
->dev_tbl
[super
->current_vol
] = dev
;
1464 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
1465 unsigned long long size
, char *name
,
1466 char *homehost
, int *uuid
)
1468 /* This is primarily called by Create when creating a new array.
1469 * We will then get add_to_super called for each component, and then
1470 * write_init_super called to write it out to each device.
1471 * For IMSM, Create can create on fresh devices or on a pre-existing
1473 * To create on a pre-existing array a different method will be called.
1474 * This one is just for fresh drives.
1476 struct intel_super
*super
;
1477 struct imsm_super
*mpb
;
1485 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
1488 super
= alloc_super(1);
1491 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
1492 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
1497 memset(mpb
, 0, mpb_size
);
1499 memcpy(mpb
->sig
, MPB_SIGNATURE
, strlen(MPB_SIGNATURE
));
1500 memcpy(mpb
->sig
+ strlen(MPB_SIGNATURE
), MPB_VERSION_RAID5
,
1501 strlen(MPB_VERSION_RAID5
));
1502 mpb
->mpb_size
= mpb_size
;
1508 static void add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
1509 int fd
, char *devname
)
1511 struct intel_super
*super
= st
->sb
;
1512 struct imsm_super
*mpb
= super
->anchor
;
1514 struct imsm_dev
*dev
;
1515 struct imsm_map
*map
;
1518 dev
= get_imsm_dev(super
, super
->current_vol
);
1519 map
= get_imsm_map(dev
, 0);
1521 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1522 if (dl
->major
== dk
->major
&&
1523 dl
->minor
== dk
->minor
)
1526 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1529 /* add a pristine spare to the metadata */
1530 if (dl
->index
< 0) {
1531 dl
->index
= super
->anchor
->num_disks
;
1532 super
->anchor
->num_disks
++;
1534 map
->disk_ord_tbl
[dk
->number
] = __cpu_to_le32(dl
->index
);
1535 status
= CONFIGURED_DISK
| USABLE_DISK
;
1536 dl
->disk
.status
= __cpu_to_le32(status
);
1538 /* if we are creating the first raid device update the family number */
1539 if (super
->current_vol
== 0) {
1541 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
1542 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
1546 sum
= __gen_imsm_checksum(mpb
);
1547 mpb
->family_num
= __cpu_to_le32(sum
);
1551 static void add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
1552 int fd
, char *devname
)
1554 struct intel_super
*super
= st
->sb
;
1556 unsigned long long size
;
1561 if (super
->current_vol
>= 0) {
1562 add_to_super_imsm_volume(st
, dk
, fd
, devname
);
1567 dd
= malloc(sizeof(*dd
));
1570 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
1573 memset(dd
, 0, sizeof(*dd
));
1574 dd
->major
= major(stb
.st_rdev
);
1575 dd
->minor
= minor(stb
.st_rdev
);
1577 dd
->devname
= devname
? strdup(devname
) : NULL
;
1579 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
1582 Name
": failed to retrieve scsi serial, aborting\n");
1587 get_dev_size(fd
, NULL
, &size
);
1589 status
= USABLE_DISK
| SPARE_DISK
;
1590 strcpy((char *) dd
->disk
.serial
, (char *) dd
->serial
);
1591 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
1592 dd
->disk
.status
= __cpu_to_le32(status
);
1593 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
1594 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
1596 dd
->disk
.scsi_id
= __cpu_to_le32(0);
1598 if (st
->update_tail
) {
1599 dd
->next
= super
->add
;
1602 dd
->next
= super
->disks
;
1607 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
1609 /* spare records have their own family number and do not have any defined raid
1612 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
1614 struct imsm_super mpb_save
;
1615 struct imsm_super
*mpb
= super
->anchor
;
1620 mpb
->num_raid_devs
= 0;
1622 mpb
->mpb_size
= sizeof(struct imsm_super
);
1623 mpb
->generation_num
= __cpu_to_le32(1UL);
1625 for (d
= super
->disks
; d
; d
= d
->next
) {
1629 mpb
->disk
[0] = d
->disk
;
1630 sum
= __gen_imsm_checksum(mpb
);
1631 mpb
->family_num
= __cpu_to_le32(sum
);
1632 sum
= __gen_imsm_checksum(mpb
);
1633 mpb
->check_sum
= __cpu_to_le32(sum
);
1635 if (store_imsm_mpb(d
->fd
, super
)) {
1636 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1637 __func__
, d
->major
, d
->minor
, strerror(errno
));
1651 static int write_super_imsm(struct intel_super
*super
, int doclose
)
1653 struct imsm_super
*mpb
= super
->anchor
;
1659 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
1661 /* 'generation' is incremented everytime the metadata is written */
1662 generation
= __le32_to_cpu(mpb
->generation_num
);
1664 mpb
->generation_num
= __cpu_to_le32(generation
);
1666 for (d
= super
->disks
; d
; d
= d
->next
) {
1670 mpb
->disk
[d
->index
] = d
->disk
;
1671 mpb_size
+= sizeof(struct imsm_disk
);
1675 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1676 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1678 imsm_copy_dev(dev
, super
->dev_tbl
[i
]);
1679 mpb_size
+= sizeof_imsm_dev(dev
, 0);
1681 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
1682 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
1684 /* recalculate checksum */
1685 sum
= __gen_imsm_checksum(mpb
);
1686 mpb
->check_sum
= __cpu_to_le32(sum
);
1688 /* write the mpb for disks that compose raid devices */
1689 for (d
= super
->disks
; d
; d
= d
->next
) {
1692 if (store_imsm_mpb(d
->fd
, super
))
1693 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1694 __func__
, d
->major
, d
->minor
, strerror(errno
));
1702 return write_super_imsm_spares(super
, doclose
);
1707 static int create_array(struct supertype
*st
)
1710 struct imsm_update_create_array
*u
;
1711 struct intel_super
*super
= st
->sb
;
1712 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1714 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0);
1717 fprintf(stderr
, "%s: failed to allocate update buffer\n",
1722 u
->type
= update_create_array
;
1723 u
->dev_idx
= super
->current_vol
;
1724 imsm_copy_dev(&u
->dev
, dev
);
1725 append_metadata_update(st
, u
, len
);
1730 static int add_disk(struct supertype
*st
)
1732 struct intel_super
*super
= st
->sb
;
1734 struct imsm_update_add_disk
*u
;
1742 fprintf(stderr
, "%s: failed to allocate update buffer\n",
1747 u
->type
= update_add_disk
;
1748 append_metadata_update(st
, u
, len
);
1753 static int write_init_super_imsm(struct supertype
*st
)
1755 if (st
->update_tail
) {
1756 /* queue the recently created array / added disk
1757 * as a metadata update */
1758 struct intel_super
*super
= st
->sb
;
1762 /* determine if we are creating a volume or adding a disk */
1763 if (super
->current_vol
< 0) {
1764 /* in the add disk case we are running in mdmon
1765 * context, so don't close fd's
1767 return add_disk(st
);
1769 rv
= create_array(st
);
1771 for (d
= super
->disks
; d
; d
= d
->next
) {
1778 return write_super_imsm(st
->sb
, 1);
1781 static int store_zero_imsm(struct supertype
*st
, int fd
)
1783 unsigned long long dsize
;
1786 get_dev_size(fd
, NULL
, &dsize
);
1788 /* first block is stored on second to last sector of the disk */
1789 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
1792 if (posix_memalign(&buf
, 512, 512) != 0)
1795 memset(buf
, 0, 512);
1796 if (write(fd
, buf
, 512) != 512)
1801 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
1802 int layout
, int raiddisks
, int chunk
,
1803 unsigned long long size
, char *dev
,
1804 unsigned long long *freesize
,
1808 unsigned long long ldsize
;
1810 if (level
!= LEVEL_CONTAINER
)
1815 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
1818 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
1819 dev
, strerror(errno
));
1822 if (!get_dev_size(fd
, dev
, &ldsize
)) {
1828 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
1833 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
1834 * FIX ME add ahci details
1836 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
1837 int layout
, int raiddisks
, int chunk
,
1838 unsigned long long size
, char *dev
,
1839 unsigned long long *freesize
,
1843 struct intel_super
*super
= st
->sb
;
1845 unsigned long long pos
= 0;
1846 unsigned long long maxsize
;
1850 if (level
== LEVEL_CONTAINER
)
1853 if (level
== 1 && raiddisks
> 2) {
1855 fprintf(stderr
, Name
": imsm does not support more "
1856 "than 2 in a raid1 configuration\n");
1860 /* We must have the container info already read in. */
1865 /* General test: make sure there is space for
1866 * 'raiddisks' device extents of size 'size' at a given
1869 unsigned long long minsize
= size
*2 /* convert to blocks */;
1870 unsigned long long start_offset
= ~0ULL;
1873 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
1874 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1879 e
= get_extents(super
, dl
);
1882 unsigned long long esize
;
1883 esize
= e
[i
].start
- pos
;
1884 if (esize
>= minsize
)
1886 if (found
&& start_offset
== ~0ULL) {
1889 } else if (found
&& pos
!= start_offset
) {
1893 pos
= e
[i
].start
+ e
[i
].size
;
1895 } while (e
[i
-1].size
);
1900 if (dcnt
< raiddisks
) {
1902 fprintf(stderr
, Name
": imsm: Not enough "
1903 "devices with space for this array "
1910 /* This device must be a member of the set */
1911 if (stat(dev
, &stb
) < 0)
1913 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
1915 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
1916 if (dl
->major
== major(stb
.st_rdev
) &&
1917 dl
->minor
== minor(stb
.st_rdev
))
1922 fprintf(stderr
, Name
": %s is not in the "
1923 "same imsm set\n", dev
);
1926 e
= get_extents(super
, dl
);
1930 unsigned long long esize
;
1931 esize
= e
[i
].start
- pos
;
1932 if (esize
>= maxsize
)
1934 pos
= e
[i
].start
+ e
[i
].size
;
1936 } while (e
[i
-1].size
);
1937 *freesize
= maxsize
;
1942 int imsm_bbm_log_size(struct imsm_super
*mpb
)
1944 return __le32_to_cpu(mpb
->bbm_log_size
);
1947 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
1948 int raiddisks
, int chunk
, unsigned long long size
,
1949 char *dev
, unsigned long long *freesize
,
1955 /* if given unused devices create a container
1956 * if given given devices in a container create a member volume
1958 if (level
== LEVEL_CONTAINER
) {
1959 /* Must be a fresh device to add to a container */
1960 return validate_geometry_imsm_container(st
, level
, layout
,
1961 raiddisks
, chunk
, size
,
1967 /* creating in a given container */
1968 return validate_geometry_imsm_volume(st
, level
, layout
,
1969 raiddisks
, chunk
, size
,
1970 dev
, freesize
, verbose
);
1973 /* limit creation to the following levels */
1985 /* This device needs to be a device in an 'imsm' container */
1986 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
1990 Name
": Cannot create this array on device %s\n",
1995 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
1997 fprintf(stderr
, Name
": Cannot open %s: %s\n",
1998 dev
, strerror(errno
));
2001 /* Well, it is in use by someone, maybe an 'imsm' container. */
2002 cfd
= open_container(fd
);
2006 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
2010 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2012 if (sra
&& sra
->array
.major_version
== -1 &&
2013 strcmp(sra
->text_version
, "imsm") == 0) {
2014 /* This is a member of a imsm container. Load the container
2015 * and try to create a volume
2017 struct intel_super
*super
;
2019 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
2021 st
->container_dev
= fd2devnum(cfd
);
2023 return validate_geometry_imsm_volume(st
, level
, layout
,
2029 } else /* may belong to another container */
2035 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
2037 /* Given a container loaded by load_super_imsm_all,
2038 * extract information about all the arrays into
2041 * For each imsm_dev create an mdinfo, fill it in,
2042 * then look for matching devices in super->disks
2043 * and create appropriate device mdinfo.
2045 struct intel_super
*super
= st
->sb
;
2046 struct imsm_super
*mpb
= super
->anchor
;
2047 struct mdinfo
*rest
= NULL
;
2050 /* do not assemble arrays that might have bad blocks */
2051 if (imsm_bbm_log_size(super
->anchor
)) {
2052 fprintf(stderr
, Name
": BBM log found in metadata. "
2053 "Cannot activate array(s).\n");
2057 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2058 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2059 struct imsm_vol
*vol
= &dev
->vol
;
2060 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2061 struct mdinfo
*this;
2064 this = malloc(sizeof(*this));
2065 memset(this, 0, sizeof(*this));
2068 this->array
.level
= get_imsm_raid_level(map
);
2069 this->array
.raid_disks
= map
->num_members
;
2070 this->array
.layout
= imsm_level_to_layout(this->array
.level
);
2071 this->array
.md_minor
= -1;
2072 this->array
.ctime
= 0;
2073 this->array
.utime
= 0;
2074 this->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
2075 this->array
.state
= !vol
->dirty
;
2076 this->container_member
= i
;
2077 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
2078 dev
->vol
.dirty
|| dev
->vol
.migr_state
)
2079 this->resync_start
= 0;
2081 this->resync_start
= ~0ULL;
2083 strncpy(this->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
2084 this->name
[MAX_RAID_SERIAL_LEN
] = 0;
2086 sprintf(this->text_version
, "/%s/%d",
2087 devnum2devname(st
->container_dev
),
2088 this->container_member
);
2090 memset(this->uuid
, 0, sizeof(this->uuid
));
2092 this->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2094 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
2095 struct mdinfo
*info_d
;
2103 idx
= get_imsm_disk_idx(map
, slot
);
2104 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
2105 for (d
= super
->disks
; d
; d
= d
->next
)
2106 if (d
->index
== idx
)
2112 s
= d
? __le32_to_cpu(d
->disk
.status
) : 0;
2113 if (s
& FAILED_DISK
)
2115 if (!(s
& USABLE_DISK
))
2117 if (ord
& IMSM_ORD_REBUILD
)
2121 * if we skip some disks the array will be assmebled degraded;
2122 * reset resync start to avoid a dirty-degraded situation
2124 * FIXME handle dirty degraded
2126 if (skip
&& !dev
->vol
.dirty
)
2127 this->resync_start
= ~0ULL;
2131 info_d
= malloc(sizeof(*info_d
));
2133 fprintf(stderr
, Name
": failed to allocate disk"
2134 " for volume %s\n", (char *) dev
->volume
);
2139 memset(info_d
, 0, sizeof(*info_d
));
2140 info_d
->next
= this->devs
;
2141 this->devs
= info_d
;
2143 info_d
->disk
.number
= d
->index
;
2144 info_d
->disk
.major
= d
->major
;
2145 info_d
->disk
.minor
= d
->minor
;
2146 info_d
->disk
.raid_disk
= slot
;
2148 this->array
.working_disks
++;
2150 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
2151 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2152 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2154 strcpy(info_d
->name
, d
->devname
);
2163 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
2166 struct intel_super
*super
= c
->sb
;
2167 struct imsm_super
*mpb
= super
->anchor
;
2169 if (atoi(inst
) >= mpb
->num_raid_devs
) {
2170 fprintf(stderr
, "%s: subarry index %d, out of range\n",
2171 __func__
, atoi(inst
));
2175 dprintf("imsm: open_new %s\n", inst
);
2176 a
->info
.container_member
= atoi(inst
);
2180 static __u8
imsm_check_degraded(struct intel_super
*super
, int n
, int failed
)
2182 struct imsm_dev
*dev
= get_imsm_dev(super
, n
);
2183 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2186 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
2187 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
2189 switch (get_imsm_raid_level(map
)) {
2191 return IMSM_T_STATE_FAILED
;
2194 if (failed
< map
->num_members
)
2195 return IMSM_T_STATE_DEGRADED
;
2197 return IMSM_T_STATE_FAILED
;
2202 * check to see if any mirrors have failed,
2203 * otherwise we are degraded
2205 int device_per_mirror
= 2; /* FIXME is this always the case?
2206 * and are they always adjacent?
2211 for (i
= 0; i
< map
->num_members
; i
++) {
2212 int idx
= get_imsm_disk_idx(map
, i
);
2213 struct imsm_disk
*disk
= get_imsm_disk(super
, idx
);
2217 else if (__le32_to_cpu(disk
->status
) & FAILED_DISK
)
2220 if (r10fail
>= device_per_mirror
)
2221 return IMSM_T_STATE_FAILED
;
2223 /* reset 'r10fail' for next mirror set */
2224 if (!((i
+ 1) % device_per_mirror
))
2228 return IMSM_T_STATE_DEGRADED
;
2232 return IMSM_T_STATE_DEGRADED
;
2234 return IMSM_T_STATE_FAILED
;
2240 return map
->map_state
;
2243 static int imsm_count_failed(struct intel_super
*super
, struct imsm_map
*map
)
2247 struct imsm_disk
*disk
;
2249 for (i
= 0; i
< map
->num_members
; i
++) {
2250 int idx
= get_imsm_disk_idx(map
, i
);
2252 disk
= get_imsm_disk(super
, idx
);
2255 else if (__le32_to_cpu(disk
->status
) & FAILED_DISK
)
2257 else if (!(__le32_to_cpu(disk
->status
) & USABLE_DISK
))
2264 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
2266 int inst
= a
->info
.container_member
;
2267 struct intel_super
*super
= a
->container
->sb
;
2268 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2269 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2270 int dirty
= !consistent
;
2274 failed
= imsm_count_failed(super
, map
);
2275 map_state
= imsm_check_degraded(super
, inst
, failed
);
2277 if (consistent
&& !dev
->vol
.dirty
&&
2278 (dev
->vol
.migr_state
|| map_state
!= IMSM_T_STATE_NORMAL
))
2279 a
->resync_start
= 0ULL;
2280 if (consistent
== 2 && a
->resync_start
!= ~0ULL)
2283 if (a
->resync_start
== ~0ULL) {
2284 /* complete recovery or initial resync */
2285 if (map
->map_state
!= map_state
) {
2286 dprintf("imsm: map_state %d: %d\n",
2288 map
->map_state
= map_state
;
2289 super
->updates_pending
++;
2291 if (dev
->vol
.migr_state
) {
2292 dprintf("imsm: mark resync complete\n");
2293 dev
->vol
.migr_state
= 0;
2294 dev
->vol
.migr_type
= 0;
2295 super
->updates_pending
++;
2297 } else if (!dev
->vol
.migr_state
) {
2298 dprintf("imsm: mark '%s' (%llu)\n",
2299 failed
? "rebuild" : "initializing", a
->resync_start
);
2300 /* mark that we are rebuilding */
2301 map
->map_state
= failed
? map_state
: IMSM_T_STATE_NORMAL
;
2302 dev
->vol
.migr_state
= 1;
2303 dev
->vol
.migr_type
= failed
? 1 : 0;
2305 a
->check_degraded
= 1;
2306 super
->updates_pending
++;
2309 /* mark dirty / clean */
2310 if (dirty
!= dev
->vol
.dirty
) {
2311 dprintf("imsm: mark '%s' (%llu)\n",
2312 dirty
? "dirty" : "clean", a
->resync_start
);
2313 dev
->vol
.dirty
= dirty
;
2314 super
->updates_pending
++;
2319 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
2321 int inst
= a
->info
.container_member
;
2322 struct intel_super
*super
= a
->container
->sb
;
2323 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2324 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2325 struct imsm_disk
*disk
;
2328 int new_failure
= 0;
2330 if (n
> map
->num_members
)
2331 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
2332 n
, map
->num_members
- 1);
2337 dprintf("imsm: set_disk %d:%x\n", n
, state
);
2339 disk
= get_imsm_disk(super
, get_imsm_disk_idx(map
, n
));
2341 /* check for new failures */
2342 status
= __le32_to_cpu(disk
->status
);
2343 if ((state
& DS_FAULTY
) && !(status
& FAILED_DISK
)) {
2344 status
|= FAILED_DISK
;
2345 disk
->status
= __cpu_to_le32(status
);
2346 disk
->scsi_id
= __cpu_to_le32(~0UL);
2347 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
2349 super
->updates_pending
++;
2351 /* check if in_sync */
2352 if ((state
& DS_INSYNC
) && !(status
& USABLE_DISK
)) {
2353 status
|= USABLE_DISK
;
2354 disk
->status
= __cpu_to_le32(status
);
2355 super
->updates_pending
++;
2358 /* the number of failures have changed, count up 'failed' to determine
2359 * degraded / failed status
2361 if (new_failure
&& map
->map_state
!= IMSM_T_STATE_FAILED
)
2362 failed
= imsm_count_failed(super
, map
);
2364 /* determine map_state based on failed or in_sync count */
2366 map
->map_state
= imsm_check_degraded(super
, inst
, failed
);
2367 else if (map
->map_state
== IMSM_T_STATE_DEGRADED
) {
2371 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2372 if (d
->curr_state
& DS_INSYNC
)
2375 if (working
== a
->info
.array
.raid_disks
) {
2376 map
->map_state
= IMSM_T_STATE_NORMAL
;
2377 dev
->vol
.migr_state
= 0;
2378 dev
->vol
.migr_type
= 0;
2379 super
->updates_pending
++;
2384 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
2386 struct imsm_super
*mpb
= super
->anchor
;
2387 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
2388 unsigned long long dsize
;
2389 unsigned long long sectors
;
2391 get_dev_size(fd
, NULL
, &dsize
);
2393 if (mpb_size
> 512) {
2394 /* -1 to account for anchor */
2395 sectors
= mpb_sectors(mpb
) - 1;
2397 /* write the extended mpb to the sectors preceeding the anchor */
2398 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
2401 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
2405 /* first block is stored on second to last sector of the disk */
2406 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2409 if (write(fd
, super
->buf
, 512) != 512)
2415 static void imsm_sync_metadata(struct supertype
*container
)
2417 struct intel_super
*super
= container
->sb
;
2419 if (!super
->updates_pending
)
2422 write_super_imsm(super
, 0);
2424 super
->updates_pending
= 0;
2427 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
2429 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2430 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2431 int i
= get_imsm_disk_idx(map
, idx
);
2434 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2438 if (dl
&& __le32_to_cpu(dl
->disk
.status
) & FAILED_DISK
)
2442 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
2447 static struct dl
*imsm_add_spare(struct intel_super
*super
, int idx
, struct active_array
*a
)
2449 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2450 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2451 unsigned long long esize
;
2452 unsigned long long pos
;
2461 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2462 /* If in this array, skip */
2463 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2464 if (d
->disk
.major
== dl
->major
&&
2465 d
->disk
.minor
== dl
->minor
) {
2466 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
2472 /* skip marked in use or failed drives */
2473 status
= __le32_to_cpu(dl
->disk
.status
);
2474 if (status
& FAILED_DISK
|| status
& CONFIGURED_DISK
) {
2475 dprintf("%x:%x status ( %s%s)\n",
2476 dl
->major
, dl
->minor
,
2477 status
& FAILED_DISK
? "failed " : "",
2478 status
& CONFIGURED_DISK
? "configured " : "");
2482 /* Does this unused device have the requisite free space?
2483 * We need a->info.component_size sectors
2485 ex
= get_extents(super
, dl
);
2487 dprintf("cannot get extents\n");
2493 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
2496 /* check that we can start at pba_of_lba0 with
2497 * a->info.component_size of space
2499 esize
= ex
[j
].start
- pos
;
2500 if (array_start
>= pos
&&
2501 array_start
+ a
->info
.component_size
< ex
[j
].start
) {
2505 pos
= ex
[j
].start
+ ex
[j
].size
;
2508 } while (ex
[j
-1].size
);
2512 dprintf("%x:%x does not have %llu at %d\n",
2513 dl
->major
, dl
->minor
,
2514 a
->info
.component_size
,
2515 __le32_to_cpu(map
->pba_of_lba0
));
2525 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
2526 struct metadata_update
**updates
)
2529 * Find a device with unused free space and use it to replace a
2530 * failed/vacant region in an array. We replace failed regions one a
2531 * array at a time. The result is that a new spare disk will be added
2532 * to the first failed array and after the monitor has finished
2533 * propagating failures the remainder will be consumed.
2535 * FIXME add a capability for mdmon to request spares from another
2539 struct intel_super
*super
= a
->container
->sb
;
2540 int inst
= a
->info
.container_member
;
2541 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2542 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2543 int failed
= a
->info
.array
.raid_disks
;
2544 struct mdinfo
*rv
= NULL
;
2547 struct metadata_update
*mu
;
2549 struct imsm_update_activate_spare
*u
;
2553 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
2554 if ((d
->curr_state
& DS_FAULTY
) &&
2556 /* wait for Removal to happen */
2558 if (d
->state_fd
>= 0)
2562 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
2563 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
2564 if (imsm_check_degraded(super
, inst
, failed
) != IMSM_T_STATE_DEGRADED
)
2567 /* For each slot, if it is not working, find a spare */
2568 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
2569 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2570 if (d
->disk
.raid_disk
== i
)
2572 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
2573 if (d
&& (d
->state_fd
>= 0))
2577 * OK, this device needs recovery. Try to re-add the previous
2578 * occupant of this slot, if this fails add a new spare
2580 dl
= imsm_readd(super
, i
, a
);
2582 dl
= imsm_add_spare(super
, i
, a
);
2586 /* found a usable disk with enough space */
2587 di
= malloc(sizeof(*di
));
2588 memset(di
, 0, sizeof(*di
));
2590 /* dl->index will be -1 in the case we are activating a
2591 * pristine spare. imsm_process_update() will create a
2592 * new index in this case. Once a disk is found to be
2593 * failed in all member arrays it is kicked from the
2596 di
->disk
.number
= dl
->index
;
2598 /* (ab)use di->devs to store a pointer to the device
2601 di
->devs
= (struct mdinfo
*) dl
;
2603 di
->disk
.raid_disk
= i
;
2604 di
->disk
.major
= dl
->major
;
2605 di
->disk
.minor
= dl
->minor
;
2607 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2608 di
->component_size
= a
->info
.component_size
;
2609 di
->container_member
= inst
;
2613 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
2614 i
, di
->data_offset
);
2620 /* No spares found */
2622 /* Now 'rv' has a list of devices to return.
2623 * Create a metadata_update record to update the
2624 * disk_ord_tbl for the array
2626 mu
= malloc(sizeof(*mu
));
2627 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
2629 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
2630 mu
->next
= *updates
;
2631 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
2633 for (di
= rv
; di
; di
= di
->next
) {
2634 u
->type
= update_activate_spare
;
2635 u
->dl
= (struct dl
*) di
->devs
;
2637 u
->slot
= di
->disk
.raid_disk
;
2648 static int disks_overlap(struct imsm_map
*m1
, struct imsm_map
*m2
)
2654 for (i
= 0; i
< m1
->num_members
; i
++) {
2655 idx
= get_imsm_disk_idx(m1
, i
);
2656 for (j
= 0; j
< m2
->num_members
; j
++)
2657 if (idx
== get_imsm_disk_idx(m2
, j
))
2664 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
);
2666 static void imsm_process_update(struct supertype
*st
,
2667 struct metadata_update
*update
)
2670 * crack open the metadata_update envelope to find the update record
2671 * update can be one of:
2672 * update_activate_spare - a spare device has replaced a failed
2673 * device in an array, update the disk_ord_tbl. If this disk is
2674 * present in all member arrays then also clear the SPARE_DISK
2677 struct intel_super
*super
= st
->sb
;
2678 struct imsm_super
*mpb
;
2679 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
2681 /* update requires a larger buf but the allocation failed */
2682 if (super
->next_len
&& !super
->next_buf
) {
2683 super
->next_len
= 0;
2687 if (super
->next_buf
) {
2688 memcpy(super
->next_buf
, super
->buf
, super
->len
);
2690 super
->len
= super
->next_len
;
2691 super
->buf
= super
->next_buf
;
2693 super
->next_len
= 0;
2694 super
->next_buf
= NULL
;
2697 mpb
= super
->anchor
;
2700 case update_activate_spare
: {
2701 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
2702 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
2703 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2704 struct active_array
*a
;
2705 struct imsm_disk
*disk
;
2712 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2717 fprintf(stderr
, "error: imsm_activate_spare passed "
2718 "an unknown disk (index: %d serial: %s)\n",
2719 u
->dl
->index
, u
->dl
->serial
);
2723 super
->updates_pending
++;
2725 /* adding a pristine spare, assign a new index */
2726 if (dl
->index
< 0) {
2727 dl
->index
= super
->anchor
->num_disks
;
2728 super
->anchor
->num_disks
++;
2730 victim
= get_imsm_disk_idx(map
, u
->slot
);
2731 map
->disk_ord_tbl
[u
->slot
] = __cpu_to_le32(dl
->index
);
2733 status
= __le32_to_cpu(disk
->status
);
2734 status
|= CONFIGURED_DISK
;
2735 status
&= ~(SPARE_DISK
| USABLE_DISK
);
2736 disk
->status
= __cpu_to_le32(status
);
2738 /* count arrays using the victim in the metadata */
2740 for (a
= st
->arrays
; a
; a
= a
->next
) {
2741 dev
= get_imsm_dev(super
, a
->info
.container_member
);
2742 map
= get_imsm_map(dev
, 0);
2743 for (i
= 0; i
< map
->num_members
; i
++)
2744 if (victim
== get_imsm_disk_idx(map
, i
))
2748 /* clear some flags if the victim is no longer being
2753 for (dlp
= &super
->disks
; *dlp
; )
2754 if ((*dlp
)->index
== victim
)
2756 disk
= &(*dlp
)->disk
;
2757 status
= __le32_to_cpu(disk
->status
);
2758 status
&= ~(CONFIGURED_DISK
| USABLE_DISK
);
2759 disk
->status
= __cpu_to_le32(status
);
2760 /* We know that 'manager' isn't touching anything,
2763 imsm_delete(super
, dlp
);
2767 case update_create_array
: {
2768 /* someone wants to create a new array, we need to be aware of
2769 * a few races/collisions:
2770 * 1/ 'Create' called by two separate instances of mdadm
2771 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
2772 * devices that have since been assimilated via
2774 * In the event this update can not be carried out mdadm will
2775 * (FIX ME) notice that its update did not take hold.
2777 struct imsm_update_create_array
*u
= (void *) update
->buf
;
2778 struct imsm_dev
*dev
;
2779 struct imsm_map
*map
, *new_map
;
2780 unsigned long long start
, end
;
2781 unsigned long long new_start
, new_end
;
2785 /* handle racing creates: first come first serve */
2786 if (u
->dev_idx
< mpb
->num_raid_devs
) {
2787 dprintf("%s: subarray %d already defined\n",
2788 __func__
, u
->dev_idx
);
2792 /* check update is next in sequence */
2793 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
2794 dprintf("%s: can not create array %d expected index %d\n",
2795 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
2799 new_map
= get_imsm_map(&u
->dev
, 0);
2800 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
2801 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
2803 /* handle activate_spare versus create race:
2804 * check to make sure that overlapping arrays do not include
2807 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2808 dev
= get_imsm_dev(super
, i
);
2809 map
= get_imsm_map(dev
, 0);
2810 start
= __le32_to_cpu(map
->pba_of_lba0
);
2811 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
2812 if ((new_start
>= start
&& new_start
<= end
) ||
2813 (start
>= new_start
&& start
<= new_end
))
2815 if (overlap
&& disks_overlap(map
, new_map
)) {
2816 dprintf("%s: arrays overlap\n", __func__
);
2820 /* check num_members sanity */
2821 if (new_map
->num_members
> mpb
->num_disks
) {
2822 dprintf("%s: num_disks out of range\n", __func__
);
2826 /* check that prepare update was successful */
2827 if (!update
->space
) {
2828 dprintf("%s: prepare update failed\n", __func__
);
2832 super
->updates_pending
++;
2833 dev
= update
->space
;
2834 update
->space
= NULL
;
2835 imsm_copy_dev(dev
, &u
->dev
);
2836 map
= get_imsm_map(dev
, 0);
2837 super
->dev_tbl
[u
->dev_idx
] = dev
;
2838 mpb
->num_raid_devs
++;
2841 for (i
= 0; i
< map
->num_members
; i
++) {
2842 struct imsm_disk
*disk
;
2845 disk
= get_imsm_disk(super
, get_imsm_disk_idx(map
, i
));
2846 status
= __le32_to_cpu(disk
->status
);
2847 status
|= CONFIGURED_DISK
;
2848 status
&= ~SPARE_DISK
;
2849 disk
->status
= __cpu_to_le32(status
);
2853 case update_add_disk
:
2855 /* we may be able to repair some arrays if disks are
2858 struct active_array
*a
;
2859 for (a
= st
->arrays
; a
; a
= a
->next
)
2860 a
->check_degraded
= 1;
2862 /* check if we can add / replace some disks in the
2864 while (super
->add
) {
2865 struct dl
**dlp
, *dl
, *al
;
2867 super
->add
= al
->next
;
2868 for (dlp
= &super
->disks
; *dlp
; ) {
2869 if (memcmp(al
->serial
, (*dlp
)->serial
,
2870 MAX_RAID_SERIAL_LEN
) == 0) {
2872 *dlp
= (*dlp
)->next
;
2873 __free_imsm_disk(dl
);
2876 dlp
= &(*dlp
)->next
;
2878 al
->next
= super
->disks
;
2886 static void imsm_prepare_update(struct supertype
*st
,
2887 struct metadata_update
*update
)
2890 * Allocate space to hold new disk entries, raid-device entries or a new
2891 * mpb if necessary. The manager synchronously waits for updates to
2892 * complete in the monitor, so new mpb buffers allocated here can be
2893 * integrated by the monitor thread without worrying about live pointers
2894 * in the manager thread.
2896 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
2897 struct intel_super
*super
= st
->sb
;
2898 struct imsm_super
*mpb
= super
->anchor
;
2903 case update_create_array
: {
2904 struct imsm_update_create_array
*u
= (void *) update
->buf
;
2906 len
= sizeof_imsm_dev(&u
->dev
, 1);
2907 update
->space
= malloc(len
);
2914 /* check if we need a larger metadata buffer */
2915 if (super
->next_buf
)
2916 buf_len
= super
->next_len
;
2918 buf_len
= super
->len
;
2920 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
2921 /* ok we need a larger buf than what is currently allocated
2922 * if this allocation fails process_update will notice that
2923 * ->next_len is set and ->next_buf is NULL
2925 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
2926 if (super
->next_buf
)
2927 free(super
->next_buf
);
2929 super
->next_len
= buf_len
;
2930 if (posix_memalign(&super
->next_buf
, buf_len
, 512) != 0)
2931 super
->next_buf
= NULL
;
2935 /* must be called while manager is quiesced */
2936 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
)
2938 struct imsm_super
*mpb
= super
->anchor
;
2939 struct dl
*dl
= *dlp
;
2941 struct imsm_dev
*dev
;
2942 struct imsm_map
*map
;
2945 dprintf("%s: deleting device %x:%x from imsm_super\n",
2946 __func__
, dl
->major
, dl
->minor
);
2948 /* shift all indexes down one */
2949 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
2950 if (iter
->index
> dl
->index
)
2953 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2954 dev
= get_imsm_dev(super
, i
);
2955 map
= get_imsm_map(dev
, 0);
2957 for (j
= 0; j
< map
->num_members
; j
++) {
2958 int idx
= get_imsm_disk_idx(map
, j
);
2960 if (idx
> dl
->index
)
2961 map
->disk_ord_tbl
[j
] = __cpu_to_le32(idx
- 1);
2966 super
->updates_pending
++;
2967 *dlp
= (*dlp
)->next
;
2968 __free_imsm_disk(dl
);
2971 struct superswitch super_imsm
= {
2973 .examine_super
= examine_super_imsm
,
2974 .brief_examine_super
= brief_examine_super_imsm
,
2975 .detail_super
= detail_super_imsm
,
2976 .brief_detail_super
= brief_detail_super_imsm
,
2977 .write_init_super
= write_init_super_imsm
,
2979 .match_home
= match_home_imsm
,
2980 .uuid_from_super
= uuid_from_super_imsm
,
2981 .getinfo_super
= getinfo_super_imsm
,
2982 .update_super
= update_super_imsm
,
2984 .avail_size
= avail_size_imsm
,
2986 .compare_super
= compare_super_imsm
,
2988 .load_super
= load_super_imsm
,
2989 .init_super
= init_super_imsm
,
2990 .add_to_super
= add_to_super_imsm
,
2991 .store_super
= store_zero_imsm
,
2992 .free_super
= free_super_imsm
,
2993 .match_metadata_desc
= match_metadata_desc_imsm
,
2994 .container_content
= container_content_imsm
,
2996 .validate_geometry
= validate_geometry_imsm
,
3000 .open_new
= imsm_open_new
,
3001 .load_super
= load_super_imsm
,
3002 .set_array_state
= imsm_set_array_state
,
3003 .set_disk
= imsm_set_disk
,
3004 .sync_metadata
= imsm_sync_metadata
,
3005 .activate_spare
= imsm_activate_spare
,
3006 .process_update
= imsm_process_update
,
3007 .prepare_update
= imsm_prepare_update
,