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.
20 #define HAVE_STDINT_H 1
28 /* MPB == Metadata Parameter Block */
29 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
30 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
31 #define MPB_VERSION_RAID0 "1.0.00"
32 #define MPB_VERSION_RAID1 "1.1.00"
33 #define MPB_VERSION_RAID5 "1.2.02"
34 #define MAX_SIGNATURE_LENGTH 32
35 #define MAX_RAID_SERIAL_LEN 16
36 #define MPB_SECTOR_CNT 418
37 #define IMSM_RESERVED_SECTORS 4096
39 /* Disk configuration info. */
40 #define IMSM_MAX_DEVICES 255
42 __u8 serial
[MAX_RAID_SERIAL_LEN
];/* 0xD8 - 0xE7 ascii serial number */
43 __u32 total_blocks
; /* 0xE8 - 0xEB total blocks */
44 __u32 scsi_id
; /* 0xEC - 0xEF scsi ID */
45 __u32 status
; /* 0xF0 - 0xF3 */
46 #define SPARE_DISK 0x01 /* Spare */
47 #define CONFIGURED_DISK 0x02 /* Member of some RaidDev */
48 #define FAILED_DISK 0x04 /* Permanent failure */
49 #define USABLE_DISK 0x08 /* Fully usable unless FAILED_DISK is set */
51 #define IMSM_DISK_FILLERS 5
52 __u32 filler
[IMSM_DISK_FILLERS
]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
55 /* RAID map configuration infos. */
57 __u32 pba_of_lba0
; /* start address of partition */
58 __u32 blocks_per_member
;/* blocks per member */
59 __u32 num_data_stripes
; /* number of data stripes */
60 __u16 blocks_per_strip
;
61 __u8 map_state
; /* Normal, Uninitialized, Degraded, Failed */
62 #define IMSM_T_STATE_NORMAL 0
63 #define IMSM_T_STATE_UNINITIALIZED 1
64 #define IMSM_T_STATE_DEGRADED 2 /* FIXME: is this correct? */
65 #define IMSM_T_STATE_FAILED 3 /* FIXME: is this correct? */
67 #define IMSM_T_RAID0 0
68 #define IMSM_T_RAID1 1
69 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
70 __u8 num_members
; /* number of member disks */
72 __u32 filler
[7]; /* expansion area */
73 #define IMSM_ORD_REBUILD (1 << 24)
74 __u32 disk_ord_tbl
[1]; /* disk_ord_tbl[num_members],
75 * top byte contains some flags
77 } __attribute__ ((packed
));
81 __u8 migr_state
; /* Normal or Migrating */
82 __u8 migr_type
; /* Initializing, Rebuilding, ... */
86 struct imsm_map map
[1];
87 /* here comes another one if migr_state */
88 } __attribute__ ((packed
));
91 __u8 volume
[MAX_RAID_SERIAL_LEN
];
94 __u32 status
; /* Persistent RaidDev status */
95 __u32 reserved_blocks
; /* Reserved blocks at beginning of volume */
96 #define IMSM_DEV_FILLERS 12
97 __u32 filler
[IMSM_DEV_FILLERS
];
99 } __attribute__ ((packed
));
102 __u8 sig
[MAX_SIGNATURE_LENGTH
]; /* 0x00 - 0x1F */
103 __u32 check_sum
; /* 0x20 - 0x23 MPB Checksum */
104 __u32 mpb_size
; /* 0x24 - 0x27 Size of MPB */
105 __u32 family_num
; /* 0x28 - 0x2B Checksum from first time this config was written */
106 __u32 generation_num
; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
107 __u32 error_log_size
; /* 0x30 - 0x33 in bytes */
108 __u32 attributes
; /* 0x34 - 0x37 */
109 __u8 num_disks
; /* 0x38 Number of configured disks */
110 __u8 num_raid_devs
; /* 0x39 Number of configured volumes */
111 __u8 error_log_pos
; /* 0x3A */
112 __u8 fill
[1]; /* 0x3B */
113 __u32 cache_size
; /* 0x3c - 0x40 in mb */
114 __u32 orig_family_num
; /* 0x40 - 0x43 original family num */
115 __u32 pwr_cycle_count
; /* 0x44 - 0x47 simulated power cycle count for array */
116 __u32 bbm_log_size
; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
117 #define IMSM_FILLERS 35
118 __u32 filler
[IMSM_FILLERS
]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
119 struct imsm_disk disk
[1]; /* 0xD8 diskTbl[numDisks] */
120 /* here comes imsm_dev[num_raid_devs] */
121 /* here comes BBM logs */
122 } __attribute__ ((packed
));
124 #define BBM_LOG_MAX_ENTRIES 254
126 struct bbm_log_entry
{
127 __u64 defective_block_start
;
128 #define UNREADABLE 0xFFFFFFFF
129 __u32 spare_block_offset
;
130 __u16 remapped_marked_count
;
132 } __attribute__ ((__packed__
));
135 __u32 signature
; /* 0xABADB10C */
137 __u32 reserved_spare_block_count
; /* 0 */
138 __u32 reserved
; /* 0xFFFF */
139 __u64 first_spare_lba
;
140 struct bbm_log_entry mapped_block_entries
[BBM_LOG_MAX_ENTRIES
];
141 } __attribute__ ((__packed__
));
145 static char *map_state_str
[] = { "normal", "uninitialized", "degraded", "failed" };
148 static unsigned int sector_count(__u32 bytes
)
150 return ((bytes
+ (512-1)) & (~(512-1))) / 512;
153 static unsigned int mpb_sectors(struct imsm_super
*mpb
)
155 return sector_count(__le32_to_cpu(mpb
->mpb_size
));
158 /* internal representation of IMSM metadata */
161 void *buf
; /* O_DIRECT buffer for reading/writing metadata */
162 struct imsm_super
*anchor
; /* immovable parameters */
164 size_t len
; /* size of the 'buf' allocation */
165 void *next_buf
; /* for realloc'ing buf from the manager */
167 int updates_pending
; /* count of pending updates for mdmon */
168 int creating_imsm
; /* flag to indicate container creation */
169 int current_vol
; /* index of raid device undergoing creation */
170 #define IMSM_MAX_RAID_DEVS 2
171 struct imsm_dev
*dev_tbl
[IMSM_MAX_RAID_DEVS
];
175 __u8 serial
[MAX_RAID_SERIAL_LEN
];
178 struct imsm_disk disk
;
181 struct dl
*add
; /* list of disks to add while mdmon active */
182 struct bbm_log
*bbm_log
;
186 unsigned long long start
, size
;
189 /* definition of messages passed to imsm_process_update */
190 enum imsm_update_type
{
191 update_activate_spare
,
196 struct imsm_update_activate_spare
{
197 enum imsm_update_type type
;
201 struct imsm_update_activate_spare
*next
;
204 struct imsm_update_create_array
{
205 enum imsm_update_type type
;
210 struct imsm_update_add_disk
{
211 enum imsm_update_type type
;
214 static int imsm_env_devname_as_serial(void)
216 char *val
= getenv("IMSM_DEVNAME_AS_SERIAL");
218 if (val
&& atoi(val
) == 1)
225 static struct supertype
*match_metadata_desc_imsm(char *arg
)
227 struct supertype
*st
;
229 if (strcmp(arg
, "imsm") != 0 &&
230 strcmp(arg
, "default") != 0
234 st
= malloc(sizeof(*st
));
235 memset(st
, 0, sizeof(*st
));
236 st
->ss
= &super_imsm
;
237 st
->max_devs
= IMSM_MAX_DEVICES
;
238 st
->minor_version
= 0;
244 static __u8
*get_imsm_version(struct imsm_super
*mpb
)
246 return &mpb
->sig
[MPB_SIG_LEN
];
250 /* retrieve a disk directly from the anchor when the anchor is known to be
251 * up-to-date, currently only at load time
253 static struct imsm_disk
*__get_imsm_disk(struct imsm_super
*mpb
, __u8 index
)
255 if (index
>= mpb
->num_disks
)
257 return &mpb
->disk
[index
];
261 /* retrieve a disk from the parsed metadata */
262 static struct imsm_disk
*get_imsm_disk(struct intel_super
*super
, __u8 index
)
266 for (d
= super
->disks
; d
; d
= d
->next
)
267 if (d
->index
== index
)
274 /* generate a checksum directly from the anchor when the anchor is known to be
275 * up-to-date, currently only at load or write_super after coalescing
277 static __u32
__gen_imsm_checksum(struct imsm_super
*mpb
)
279 __u32 end
= mpb
->mpb_size
/ sizeof(end
);
280 __u32
*p
= (__u32
*) mpb
;
284 sum
+= __le32_to_cpu(*p
++);
286 return sum
- __le32_to_cpu(mpb
->check_sum
);
289 static size_t sizeof_imsm_map(struct imsm_map
*map
)
291 return sizeof(struct imsm_map
) + sizeof(__u32
) * (map
->num_members
- 1);
294 struct imsm_map
*get_imsm_map(struct imsm_dev
*dev
, int second_map
)
296 struct imsm_map
*map
= &dev
->vol
.map
[0];
298 if (second_map
&& !dev
->vol
.migr_state
)
300 else if (second_map
) {
303 return ptr
+ sizeof_imsm_map(map
);
309 /* return the size of the device.
310 * migr_state increases the returned size if map[0] were to be duplicated
312 static size_t sizeof_imsm_dev(struct imsm_dev
*dev
, int migr_state
)
314 size_t size
= sizeof(*dev
) - sizeof(struct imsm_map
) +
315 sizeof_imsm_map(get_imsm_map(dev
, 0));
317 /* migrating means an additional map */
318 if (dev
->vol
.migr_state
)
319 size
+= sizeof_imsm_map(get_imsm_map(dev
, 1));
321 size
+= sizeof_imsm_map(get_imsm_map(dev
, 0));
326 static struct imsm_dev
*__get_imsm_dev(struct imsm_super
*mpb
, __u8 index
)
332 if (index
>= mpb
->num_raid_devs
)
335 /* devices start after all disks */
336 offset
= ((void *) &mpb
->disk
[mpb
->num_disks
]) - _mpb
;
338 for (i
= 0; i
<= index
; i
++)
340 return _mpb
+ offset
;
342 offset
+= sizeof_imsm_dev(_mpb
+ offset
, 0);
347 static struct imsm_dev
*get_imsm_dev(struct intel_super
*super
, __u8 index
)
349 if (index
>= super
->anchor
->num_raid_devs
)
351 return super
->dev_tbl
[index
];
354 static __u32
get_imsm_ord_tbl_ent(struct imsm_dev
*dev
, int slot
)
356 struct imsm_map
*map
;
358 if (dev
->vol
.migr_state
)
359 map
= get_imsm_map(dev
, 1);
361 map
= get_imsm_map(dev
, 0);
363 /* top byte identifies disk under rebuild */
364 return __le32_to_cpu(map
->disk_ord_tbl
[slot
]);
367 #define ord_to_idx(ord) (((ord) << 8) >> 8)
368 static __u32
get_imsm_disk_idx(struct imsm_dev
*dev
, int slot
)
370 __u32 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
372 return ord_to_idx(ord
);
375 static void set_imsm_ord_tbl_ent(struct imsm_map
*map
, int slot
, __u32 ord
)
377 map
->disk_ord_tbl
[slot
] = __cpu_to_le32(ord
);
380 static int get_imsm_raid_level(struct imsm_map
*map
)
382 if (map
->raid_level
== 1) {
383 if (map
->num_members
== 2)
389 return map
->raid_level
;
392 static int cmp_extent(const void *av
, const void *bv
)
394 const struct extent
*a
= av
;
395 const struct extent
*b
= bv
;
396 if (a
->start
< b
->start
)
398 if (a
->start
> b
->start
)
403 static struct extent
*get_extents(struct intel_super
*super
, struct dl
*dl
)
405 /* find a list of used extents on the given physical device */
406 struct extent
*rv
, *e
;
409 __u32 reservation
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
411 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
412 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
413 struct imsm_map
*map
= get_imsm_map(dev
, 0);
415 for (j
= 0; j
< map
->num_members
; j
++) {
416 __u32 index
= get_imsm_disk_idx(dev
, j
);
418 if (index
== dl
->index
)
422 rv
= malloc(sizeof(struct extent
) * (memberships
+ 1));
427 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
428 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
429 struct imsm_map
*map
= get_imsm_map(dev
, 0);
431 for (j
= 0; j
< map
->num_members
; j
++) {
432 __u32 index
= get_imsm_disk_idx(dev
, j
);
434 if (index
== dl
->index
) {
435 e
->start
= __le32_to_cpu(map
->pba_of_lba0
);
436 e
->size
= __le32_to_cpu(map
->blocks_per_member
);
441 qsort(rv
, memberships
, sizeof(*rv
), cmp_extent
);
443 /* determine the start of the metadata
444 * when no raid devices are defined use the default
445 * ...otherwise allow the metadata to truncate the value
446 * as is the case with older versions of imsm
449 struct extent
*last
= &rv
[memberships
- 1];
452 remainder
= __le32_to_cpu(dl
->disk
.total_blocks
) -
453 (last
->start
+ last
->size
);
454 if (reservation
> remainder
)
455 reservation
= remainder
;
457 e
->start
= __le32_to_cpu(dl
->disk
.total_blocks
) - reservation
;
462 /* try to determine how much space is reserved for metadata from
463 * the last get_extents() entry, otherwise fallback to the
466 static __u32
imsm_reserved_sectors(struct intel_super
*super
, struct dl
*dl
)
472 /* for spares just return a minimal reservation which will grow
473 * once the spare is picked up by an array
476 return MPB_SECTOR_CNT
;
478 e
= get_extents(super
, dl
);
480 return MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
482 /* scroll to last entry */
483 for (i
= 0; e
[i
].size
; i
++)
486 rv
= __le32_to_cpu(dl
->disk
.total_blocks
) - e
[i
].start
;
494 static void print_imsm_dev(struct imsm_dev
*dev
, int index
)
498 struct imsm_map
*map
= get_imsm_map(dev
, 0);
502 printf("[%s]:\n", dev
->volume
);
503 printf(" RAID Level : %d\n", get_imsm_raid_level(map
));
504 printf(" Members : %d\n", map
->num_members
);
505 for (slot
= 0; slot
< map
->num_members
; slot
++)
506 if (index
== get_imsm_disk_idx(dev
, slot
))
508 if (slot
< map
->num_members
) {
509 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
510 printf(" This Slot : %d%s\n", slot
,
511 ord
& IMSM_ORD_REBUILD
? " (out-of-sync)" : "");
513 printf(" This Slot : ?\n");
514 sz
= __le32_to_cpu(dev
->size_high
);
516 sz
+= __le32_to_cpu(dev
->size_low
);
517 printf(" Array Size : %llu%s\n", (unsigned long long)sz
,
518 human_size(sz
* 512));
519 sz
= __le32_to_cpu(map
->blocks_per_member
);
520 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz
,
521 human_size(sz
* 512));
522 printf(" Sector Offset : %u\n",
523 __le32_to_cpu(map
->pba_of_lba0
));
524 printf(" Num Stripes : %u\n",
525 __le32_to_cpu(map
->num_data_stripes
));
526 printf(" Chunk Size : %u KiB\n",
527 __le16_to_cpu(map
->blocks_per_strip
) / 2);
528 printf(" Reserved : %d\n", __le32_to_cpu(dev
->reserved_blocks
));
529 printf(" Migrate State : %s", dev
->vol
.migr_state
? "migrating" : "idle");
530 if (dev
->vol
.migr_state
)
531 printf(": %s", dev
->vol
.migr_type
? "rebuilding" : "initializing");
533 printf(" Map State : %s", map_state_str
[map
->map_state
]);
534 if (dev
->vol
.migr_state
) {
535 struct imsm_map
*map
= get_imsm_map(dev
, 1);
536 printf(" <-- %s", map_state_str
[map
->map_state
]);
539 printf(" Dirty State : %s\n", dev
->vol
.dirty
? "dirty" : "clean");
542 static void print_imsm_disk(struct imsm_super
*mpb
, int index
, __u32 reserved
)
544 struct imsm_disk
*disk
= __get_imsm_disk(mpb
, index
);
545 char str
[MAX_RAID_SERIAL_LEN
+ 1];
553 snprintf(str
, MAX_RAID_SERIAL_LEN
+ 1, "%s", disk
->serial
);
554 printf(" Disk%02d Serial : %s\n", index
, str
);
555 s
= __le32_to_cpu(disk
->status
);
556 printf(" State :%s%s%s%s\n", s
&SPARE_DISK
? " spare" : "",
557 s
&CONFIGURED_DISK
? " active" : "",
558 s
&FAILED_DISK
? " failed" : "",
559 s
&USABLE_DISK
? " usable" : "");
560 printf(" Id : %08x\n", __le32_to_cpu(disk
->scsi_id
));
561 sz
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
562 printf(" Usable Size : %llu%s\n", (unsigned long long)sz
,
563 human_size(sz
* 512));
566 static void examine_super_imsm(struct supertype
*st
, char *homehost
)
568 struct intel_super
*super
= st
->sb
;
569 struct imsm_super
*mpb
= super
->anchor
;
570 char str
[MAX_SIGNATURE_LENGTH
];
573 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
575 snprintf(str
, MPB_SIG_LEN
, "%s", mpb
->sig
);
576 printf(" Magic : %s\n", str
);
577 snprintf(str
, strlen(MPB_VERSION_RAID0
), "%s", get_imsm_version(mpb
));
578 printf(" Version : %s\n", get_imsm_version(mpb
));
579 printf(" Family : %08x\n", __le32_to_cpu(mpb
->family_num
));
580 printf(" Generation : %08x\n", __le32_to_cpu(mpb
->generation_num
));
581 sum
= __le32_to_cpu(mpb
->check_sum
);
582 printf(" Checksum : %08x %s\n", sum
,
583 __gen_imsm_checksum(mpb
) == sum
? "correct" : "incorrect");
584 printf(" MPB Sectors : %d\n", mpb_sectors(mpb
));
585 printf(" Disks : %d\n", mpb
->num_disks
);
586 printf(" RAID Devices : %d\n", mpb
->num_raid_devs
);
587 print_imsm_disk(mpb
, super
->disks
->index
, reserved
);
588 if (super
->bbm_log
) {
589 struct bbm_log
*log
= super
->bbm_log
;
592 printf("Bad Block Management Log:\n");
593 printf(" Log Size : %d\n", __le32_to_cpu(mpb
->bbm_log_size
));
594 printf(" Signature : %x\n", __le32_to_cpu(log
->signature
));
595 printf(" Entry Count : %d\n", __le32_to_cpu(log
->entry_count
));
596 printf(" Spare Blocks : %d\n", __le32_to_cpu(log
->reserved_spare_block_count
));
597 printf(" First Spare : %llx\n", __le64_to_cpu(log
->first_spare_lba
));
599 for (i
= 0; i
< mpb
->num_raid_devs
; i
++)
600 print_imsm_dev(__get_imsm_dev(mpb
, i
), super
->disks
->index
);
601 for (i
= 0; i
< mpb
->num_disks
; i
++) {
602 if (i
== super
->disks
->index
)
604 print_imsm_disk(mpb
, i
, reserved
);
608 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
);
610 static void brief_examine_super_imsm(struct supertype
*st
)
612 /* We just write a generic DDF ARRAY entry
617 getinfo_super_imsm(st
, &info
);
618 fname_from_uuid(st
, &info
, nbuf
,'-');
619 printf("ARRAY /dev/imsm metadata=imsm UUID=%s\n", nbuf
+ 5);
622 static void detail_super_imsm(struct supertype
*st
, char *homehost
)
624 printf("%s\n", __FUNCTION__
);
627 static void brief_detail_super_imsm(struct supertype
*st
)
631 getinfo_super_imsm(st
, &info
);
632 fname_from_uuid(st
, &info
, nbuf
,'-');
633 printf(" UUID=%s", nbuf
+ 5);
637 static int match_home_imsm(struct supertype
*st
, char *homehost
)
639 printf("%s\n", __FUNCTION__
);
644 static void uuid_from_super_imsm(struct supertype
*st
, int uuid
[4])
646 /* The uuid returned here is used for:
647 * uuid to put into bitmap file (Create, Grow)
648 * uuid for backup header when saving critical section (Grow)
649 * comparing uuids when re-adding a device into an array
650 * In these cases the uuid required is that of the data-array,
651 * not the device-set.
652 * uuid to recognise same set when adding a missing device back
653 * to an array. This is a uuid for the device-set.
655 * For each of these we can make do with a truncated
656 * or hashed uuid rather than the original, as long as
658 * In each case the uuid required is that of the data-array,
659 * not the device-set.
661 /* imsm does not track uuid's so we synthesis one using sha1 on
662 * - The signature (Which is constant for all imsm array, but no matter)
663 * - the family_num of the container
664 * - the index number of the volume
665 * - the 'serial' number of the volume.
666 * Hopefully these are all constant.
668 struct intel_super
*super
= st
->sb
;
672 struct imsm_dev
*dev
= NULL
;
675 sha1_process_bytes(super
->anchor
->sig
, MAX_SIGNATURE_LENGTH
, &ctx
);
676 sha1_process_bytes(&super
->anchor
->family_num
, sizeof(__u32
), &ctx
);
677 if (super
->current_vol
>= 0)
678 dev
= get_imsm_dev(super
, super
->current_vol
);
680 __u32 vol
= super
->current_vol
;
681 sha1_process_bytes(&vol
, sizeof(vol
), &ctx
);
682 sha1_process_bytes(dev
->volume
, MAX_RAID_SERIAL_LEN
, &ctx
);
684 sha1_finish_ctx(&ctx
, buf
);
685 memcpy(uuid
, buf
, 4*4);
690 get_imsm_numerical_version(struct imsm_super
*mpb
, int *m
, int *p
)
692 __u8
*v
= get_imsm_version(mpb
);
693 __u8
*end
= mpb
->sig
+ MAX_SIGNATURE_LENGTH
;
694 char major
[] = { 0, 0, 0 };
695 char minor
[] = { 0 ,0, 0 };
696 char patch
[] = { 0, 0, 0 };
697 char *ver_parse
[] = { major
, minor
, patch
};
701 while (*v
!= '\0' && v
< end
) {
702 if (*v
!= '.' && j
< 2)
703 ver_parse
[i
][j
++] = *v
;
711 *m
= strtol(minor
, NULL
, 0);
712 *p
= strtol(patch
, NULL
, 0);
716 static int imsm_level_to_layout(int level
)
724 return ALGORITHM_LEFT_ASYMMETRIC
;
731 static void getinfo_super_imsm_volume(struct supertype
*st
, struct mdinfo
*info
)
733 struct intel_super
*super
= st
->sb
;
734 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
735 struct imsm_map
*map
= get_imsm_map(dev
, 0);
737 info
->container_member
= super
->current_vol
;
738 info
->array
.raid_disks
= map
->num_members
;
739 info
->array
.level
= get_imsm_raid_level(map
);
740 info
->array
.layout
= imsm_level_to_layout(info
->array
.level
);
741 info
->array
.md_minor
= -1;
742 info
->array
.ctime
= 0;
743 info
->array
.utime
= 0;
744 info
->array
.chunk_size
= __le16_to_cpu(map
->blocks_per_strip
) << 9;
745 info
->array
.state
= !dev
->vol
.dirty
;
747 info
->disk
.major
= 0;
748 info
->disk
.minor
= 0;
750 info
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
751 info
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
752 memset(info
->uuid
, 0, sizeof(info
->uuid
));
754 if (map
->map_state
== IMSM_T_STATE_UNINITIALIZED
||
755 dev
->vol
.dirty
|| dev
->vol
.migr_state
)
756 info
->resync_start
= 0;
758 info
->resync_start
= ~0ULL;
760 strncpy(info
->name
, (char *) dev
->volume
, MAX_RAID_SERIAL_LEN
);
761 info
->name
[MAX_RAID_SERIAL_LEN
] = 0;
763 info
->array
.major_version
= -1;
764 info
->array
.minor_version
= -2;
765 sprintf(info
->text_version
, "/%s/%d",
766 devnum2devname(st
->container_dev
),
767 info
->container_member
);
768 info
->safe_mode_delay
= 4000; /* 4 secs like the Matrix driver */
769 uuid_from_super_imsm(st
, info
->uuid
);
773 static void getinfo_super_imsm(struct supertype
*st
, struct mdinfo
*info
)
775 struct intel_super
*super
= st
->sb
;
776 struct imsm_disk
*disk
;
779 if (super
->current_vol
>= 0) {
780 getinfo_super_imsm_volume(st
, info
);
784 /* Set raid_disks to zero so that Assemble will always pull in valid
787 info
->array
.raid_disks
= 0;
788 info
->array
.level
= LEVEL_CONTAINER
;
789 info
->array
.layout
= 0;
790 info
->array
.md_minor
= -1;
791 info
->array
.ctime
= 0; /* N/A for imsm */
792 info
->array
.utime
= 0;
793 info
->array
.chunk_size
= 0;
795 info
->disk
.major
= 0;
796 info
->disk
.minor
= 0;
797 info
->disk
.raid_disk
= -1;
798 info
->reshape_active
= 0;
799 info
->array
.major_version
= -1;
800 info
->array
.minor_version
= -2;
801 strcpy(info
->text_version
, "imsm");
802 info
->safe_mode_delay
= 0;
803 info
->disk
.number
= -1;
804 info
->disk
.state
= 0;
808 __u32 reserved
= imsm_reserved_sectors(super
, super
->disks
);
810 disk
= &super
->disks
->disk
;
811 info
->disk
.number
= super
->disks
->index
;
812 info
->disk
.raid_disk
= super
->disks
->index
;
813 info
->data_offset
= __le32_to_cpu(disk
->total_blocks
) - reserved
;
814 info
->component_size
= reserved
;
815 s
= __le32_to_cpu(disk
->status
);
816 info
->disk
.state
= s
& CONFIGURED_DISK
? (1 << MD_DISK_ACTIVE
) : 0;
817 info
->disk
.state
|= s
& FAILED_DISK
? (1 << MD_DISK_FAULTY
) : 0;
818 info
->disk
.state
|= s
& USABLE_DISK
? (1 << MD_DISK_SYNC
) : 0;
820 uuid_from_super_imsm(st
, info
->uuid
);
823 static int update_super_imsm(struct supertype
*st
, struct mdinfo
*info
,
824 char *update
, char *devname
, int verbose
,
825 int uuid_set
, char *homehost
)
829 /* For 'assemble' and 'force' we need to return non-zero if any
830 * change was made. For others, the return value is ignored.
831 * Update options are:
832 * force-one : This device looks a bit old but needs to be included,
833 * update age info appropriately.
834 * assemble: clear any 'faulty' flag to allow this device to
836 * force-array: Array is degraded but being forced, mark it clean
837 * if that will be needed to assemble it.
839 * newdev: not used ????
840 * grow: Array has gained a new device - this is currently for
842 * resync: mark as dirty so a resync will happen.
843 * name: update the name - preserving the homehost
845 * Following are not relevant for this imsm:
846 * sparc2.2 : update from old dodgey metadata
847 * super-minor: change the preferred_minor number
848 * summaries: update redundant counters.
849 * uuid: Change the uuid of the array to match watch is given
850 * homehost: update the recorded homehost
851 * _reshape_progress: record new reshape_progress position.
854 //struct intel_super *super = st->sb;
855 //struct imsm_super *mpb = super->mpb;
857 if (strcmp(update
, "grow") == 0) {
859 if (strcmp(update
, "resync") == 0) {
860 /* dev->vol.dirty = 1; */
863 /* IMSM has no concept of UUID or homehost */
868 static size_t disks_to_mpb_size(int disks
)
872 size
= sizeof(struct imsm_super
);
873 size
+= (disks
- 1) * sizeof(struct imsm_disk
);
874 size
+= 2 * sizeof(struct imsm_dev
);
875 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
876 size
+= (4 - 2) * sizeof(struct imsm_map
);
877 /* 4 possible disk_ord_tbl's */
878 size
+= 4 * (disks
- 1) * sizeof(__u32
);
883 static __u64
avail_size_imsm(struct supertype
*st
, __u64 devsize
)
885 if (devsize
< (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
))
888 return devsize
- (MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
);
891 static int compare_super_imsm(struct supertype
*st
, struct supertype
*tst
)
895 * 0 same, or first was empty, and second was copied
896 * 1 second had wrong number
900 struct intel_super
*first
= st
->sb
;
901 struct intel_super
*sec
= tst
->sb
;
909 if (memcmp(first
->anchor
->sig
, sec
->anchor
->sig
, MAX_SIGNATURE_LENGTH
) != 0)
912 /* if an anchor does not have num_raid_devs set then it is a free
915 if (first
->anchor
->num_raid_devs
> 0 &&
916 sec
->anchor
->num_raid_devs
> 0) {
917 if (first
->anchor
->family_num
!= sec
->anchor
->family_num
)
921 /* if 'first' is a spare promote it to a populated mpb with sec's
924 if (first
->anchor
->num_raid_devs
== 0 &&
925 sec
->anchor
->num_raid_devs
> 0) {
926 first
->anchor
->num_raid_devs
= sec
->anchor
->num_raid_devs
;
927 first
->anchor
->family_num
= sec
->anchor
->family_num
;
933 static void fd2devname(int fd
, char *name
)
942 if (fstat(fd
, &st
) != 0)
944 sprintf(path
, "/sys/dev/block/%d:%d",
945 major(st
.st_rdev
), minor(st
.st_rdev
));
947 rv
= readlink(path
, dname
, sizeof(dname
));
952 nm
= strrchr(dname
, '/');
954 snprintf(name
, MAX_RAID_SERIAL_LEN
, "/dev/%s", nm
);
958 extern int scsi_get_serial(int fd
, void *buf
, size_t buf_len
);
960 static int imsm_read_serial(int fd
, char *devname
,
961 __u8 serial
[MAX_RAID_SERIAL_LEN
])
963 unsigned char scsi_serial
[255];
969 memset(scsi_serial
, 0, sizeof(scsi_serial
));
971 rv
= scsi_get_serial(fd
, scsi_serial
, sizeof(scsi_serial
));
973 if (rv
&& imsm_env_devname_as_serial()) {
974 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
975 fd2devname(fd
, (char *) serial
);
982 Name
": Failed to retrieve serial for %s\n",
987 /* trim leading whitespace */
988 rsp_len
= scsi_serial
[3];
989 rsp_buf
= (char *) &scsi_serial
[4];
994 /* truncate len to the end of rsp_buf if necessary */
995 if (c
+ MAX_RAID_SERIAL_LEN
> rsp_buf
+ rsp_len
)
996 len
= rsp_len
- (c
- rsp_buf
);
998 len
= MAX_RAID_SERIAL_LEN
;
1000 /* initialize the buffer and copy rsp_buf characters */
1001 memset(serial
, 0, MAX_RAID_SERIAL_LEN
);
1002 memcpy(serial
, c
, len
);
1004 /* trim trailing whitespace starting with the last character copied */
1005 c
= (char *) &serial
[len
- 1];
1006 while (isspace(*c
) || *c
== '\0')
1012 static int serialcmp(__u8
*s1
, __u8
*s2
)
1014 return strncmp((char *) s1
, (char *) s2
, MAX_RAID_SERIAL_LEN
);
1017 static void serialcpy(__u8
*dest
, __u8
*src
)
1019 strncpy((char *) dest
, (char *) src
, MAX_RAID_SERIAL_LEN
);
1023 load_imsm_disk(int fd
, struct intel_super
*super
, char *devname
, int keep_fd
)
1030 __u8 serial
[MAX_RAID_SERIAL_LEN
];
1032 rv
= imsm_read_serial(fd
, devname
, serial
);
1037 /* check if this is a disk we have seen before. it may be a spare in
1038 * super->disks while the current anchor believes it is a raid member,
1039 * check if we need to update dl->index
1041 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1042 if (serialcmp(dl
->serial
, serial
) == 0)
1046 dl
= malloc(sizeof(*dl
));
1053 Name
": failed to allocate disk buffer for %s\n",
1060 dl
->major
= major(stb
.st_rdev
);
1061 dl
->minor
= minor(stb
.st_rdev
);
1062 dl
->next
= super
->disks
;
1063 dl
->fd
= keep_fd
? fd
: -1;
1064 dl
->devname
= devname
? strdup(devname
) : NULL
;
1065 serialcpy(dl
->serial
, serial
);
1067 } else if (keep_fd
) {
1072 /* look up this disk's index in the current anchor */
1073 for (i
= 0; i
< super
->anchor
->num_disks
; i
++) {
1074 struct imsm_disk
*disk_iter
;
1076 disk_iter
= __get_imsm_disk(super
->anchor
, i
);
1078 if (serialcmp(disk_iter
->serial
, dl
->serial
) == 0) {
1081 dl
->disk
= *disk_iter
;
1082 status
= __le32_to_cpu(dl
->disk
.status
);
1083 /* only set index on disks that are a member of a
1084 * populated contianer, i.e. one with raid_devs
1086 if (status
& FAILED_DISK
)
1088 else if (status
& SPARE_DISK
)
1103 static void imsm_copy_dev(struct imsm_dev
*dest
, struct imsm_dev
*src
)
1105 memcpy(dest
, src
, sizeof_imsm_dev(src
, 0));
1109 /* When migrating map0 contains the 'destination' state while map1
1110 * contains the current state. When not migrating map0 contains the
1111 * current state. This routine assumes that map[0].map_state is set to
1112 * the current array state before being called.
1114 * Migration is indicated by one of the following states
1115 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1116 * 2/ Initialize (migr_state=1 migr_type=0 map0state=normal
1117 * map1state=unitialized)
1118 * 3/ Verify (Resync) (migr_state=1 migr_type=1 map0state=normal
1120 * 4/ Rebuild (migr_state=1 migr_type=1 map0state=normal
1121 * map1state=degraded)
1123 static void migrate(struct imsm_dev
*dev
, __u8 to_state
, int rebuild_resync
)
1125 struct imsm_map
*dest
;
1126 struct imsm_map
*src
= get_imsm_map(dev
, 0);
1128 dev
->vol
.migr_state
= 1;
1129 dev
->vol
.migr_type
= rebuild_resync
;
1130 dest
= get_imsm_map(dev
, 1);
1132 memcpy(dest
, src
, sizeof_imsm_map(src
));
1133 src
->map_state
= to_state
;
1137 static int parse_raid_devices(struct intel_super
*super
)
1140 struct imsm_dev
*dev_new
;
1141 size_t len
, len_migr
;
1142 size_t space_needed
= 0;
1143 struct imsm_super
*mpb
= super
->anchor
;
1145 for (i
= 0; i
< super
->anchor
->num_raid_devs
; i
++) {
1146 struct imsm_dev
*dev_iter
= __get_imsm_dev(super
->anchor
, i
);
1148 len
= sizeof_imsm_dev(dev_iter
, 0);
1149 len_migr
= sizeof_imsm_dev(dev_iter
, 1);
1151 space_needed
+= len_migr
- len
;
1153 dev_new
= malloc(len_migr
);
1156 imsm_copy_dev(dev_new
, dev_iter
);
1157 super
->dev_tbl
[i
] = dev_new
;
1160 /* ensure that super->buf is large enough when all raid devices
1163 if (__le32_to_cpu(mpb
->mpb_size
) + space_needed
> super
->len
) {
1166 len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + space_needed
, 512);
1167 if (posix_memalign(&buf
, 512, len
) != 0)
1170 memcpy(buf
, super
->buf
, len
);
1179 /* retrieve a pointer to the bbm log which starts after all raid devices */
1180 struct bbm_log
*__get_imsm_bbm_log(struct imsm_super
*mpb
)
1184 if (__le32_to_cpu(mpb
->bbm_log_size
)) {
1186 ptr
+= mpb
->mpb_size
- __le32_to_cpu(mpb
->bbm_log_size
);
1192 static void __free_imsm(struct intel_super
*super
, int free_disks
);
1194 /* load_imsm_mpb - read matrix metadata
1195 * allocates super->mpb to be freed by free_super
1197 static int load_imsm_mpb(int fd
, struct intel_super
*super
, char *devname
)
1199 unsigned long long dsize
;
1200 unsigned long long sectors
;
1202 struct imsm_super
*anchor
;
1206 get_dev_size(fd
, NULL
, &dsize
);
1208 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0) {
1211 Name
": Cannot seek to anchor block on %s: %s\n",
1212 devname
, strerror(errno
));
1216 if (posix_memalign((void**)&anchor
, 512, 512) != 0) {
1219 Name
": Failed to allocate imsm anchor buffer"
1220 " on %s\n", devname
);
1223 if (read(fd
, anchor
, 512) != 512) {
1226 Name
": Cannot read anchor block on %s: %s\n",
1227 devname
, strerror(errno
));
1232 if (strncmp((char *) anchor
->sig
, MPB_SIGNATURE
, MPB_SIG_LEN
) != 0) {
1235 Name
": no IMSM anchor on %s\n", devname
);
1240 __free_imsm(super
, 0);
1241 super
->len
= ROUND_UP(anchor
->mpb_size
, 512);
1242 if (posix_memalign(&super
->buf
, 512, super
->len
) != 0) {
1245 Name
": unable to allocate %zu byte mpb buffer\n",
1250 memcpy(super
->buf
, anchor
, 512);
1252 sectors
= mpb_sectors(anchor
) - 1;
1255 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1257 rc
= parse_raid_devices(super
);
1261 /* read the extended mpb */
1262 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0) {
1265 Name
": Cannot seek to extended mpb on %s: %s\n",
1266 devname
, strerror(errno
));
1270 if (read(fd
, super
->buf
+ 512, super
->len
- 512) != super
->len
- 512) {
1273 Name
": Cannot read extended mpb on %s: %s\n",
1274 devname
, strerror(errno
));
1278 check_sum
= __gen_imsm_checksum(super
->anchor
);
1279 if (check_sum
!= __le32_to_cpu(super
->anchor
->check_sum
)) {
1282 Name
": IMSM checksum %x != %x on %s\n",
1283 check_sum
, __le32_to_cpu(super
->anchor
->check_sum
),
1288 /* FIXME the BBM log is disk specific so we cannot use this global
1289 * buffer for all disks. Ok for now since we only look at the global
1290 * bbm_log_size parameter to gate assembly
1292 super
->bbm_log
= __get_imsm_bbm_log(super
->anchor
);
1294 rc
= load_imsm_disk(fd
, super
, devname
, 0);
1296 rc
= parse_raid_devices(super
);
1301 static void __free_imsm_disk(struct dl
*d
)
1310 static void free_imsm_disks(struct intel_super
*super
)
1312 while (super
->disks
) {
1313 struct dl
*d
= super
->disks
;
1315 super
->disks
= d
->next
;
1316 __free_imsm_disk(d
);
1320 /* free all the pieces hanging off of a super pointer */
1321 static void __free_imsm(struct intel_super
*super
, int free_disks
)
1330 free_imsm_disks(super
);
1331 for (i
= 0; i
< IMSM_MAX_RAID_DEVS
; i
++)
1332 if (super
->dev_tbl
[i
]) {
1333 free(super
->dev_tbl
[i
]);
1334 super
->dev_tbl
[i
] = NULL
;
1338 static void free_imsm(struct intel_super
*super
)
1340 __free_imsm(super
, 1);
1344 static void free_super_imsm(struct supertype
*st
)
1346 struct intel_super
*super
= st
->sb
;
1355 static struct intel_super
*alloc_super(int creating_imsm
)
1357 struct intel_super
*super
= malloc(sizeof(*super
));
1360 memset(super
, 0, sizeof(*super
));
1361 super
->creating_imsm
= creating_imsm
;
1362 super
->current_vol
= -1;
1369 static int load_super_imsm_all(struct supertype
*st
, int fd
, void **sbp
,
1370 char *devname
, int keep_fd
)
1373 struct intel_super
*super
;
1374 struct mdinfo
*sd
, *best
= NULL
;
1381 /* check if this disk is a member of an active array */
1382 sra
= sysfs_read(fd
, 0, GET_LEVEL
|GET_VERSION
|GET_DEVS
|GET_STATE
);
1386 if (sra
->array
.major_version
!= -1 ||
1387 sra
->array
.minor_version
!= -2 ||
1388 strcmp(sra
->text_version
, "imsm") != 0)
1391 super
= alloc_super(0);
1395 /* find the most up to date disk in this array, skipping spares */
1396 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1397 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1398 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1403 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1407 if (super
->anchor
->num_raid_devs
== 0)
1410 gen
= __le32_to_cpu(super
->anchor
->generation_num
);
1411 if (!best
|| gen
> bestgen
) {
1426 /* load the most up to date anchor */
1427 sprintf(nm
, "%d:%d", best
->disk
.major
, best
->disk
.minor
);
1428 dfd
= dev_open(nm
, O_RDONLY
);
1433 rv
= load_imsm_mpb(dfd
, super
, NULL
);
1440 /* re-parse the disk list with the current anchor */
1441 for (sd
= sra
->devs
; sd
; sd
= sd
->next
) {
1442 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
1443 dfd
= dev_open(nm
, keep_fd
? O_RDWR
: O_RDONLY
);
1448 load_imsm_disk(dfd
, super
, NULL
, keep_fd
);
1453 if (st
->subarray
[0]) {
1454 if (atoi(st
->subarray
) <= super
->anchor
->num_raid_devs
)
1455 super
->current_vol
= atoi(st
->subarray
);
1461 st
->container_dev
= fd2devnum(fd
);
1462 if (st
->ss
== NULL
) {
1463 st
->ss
= &super_imsm
;
1464 st
->minor_version
= 0;
1465 st
->max_devs
= IMSM_MAX_DEVICES
;
1467 st
->loaded_container
= 1;
1473 static int load_super_imsm(struct supertype
*st
, int fd
, char *devname
)
1475 struct intel_super
*super
;
1479 if (load_super_imsm_all(st
, fd
, &st
->sb
, devname
, 1) == 0)
1482 if (st
->subarray
[0])
1483 return 1; /* FIXME */
1485 super
= alloc_super(0);
1488 Name
": malloc of %zu failed.\n",
1493 rv
= load_imsm_mpb(fd
, super
, devname
);
1498 Name
": Failed to load all information "
1499 "sections on %s\n", devname
);
1505 if (st
->ss
== NULL
) {
1506 st
->ss
= &super_imsm
;
1507 st
->minor_version
= 0;
1508 st
->max_devs
= IMSM_MAX_DEVICES
;
1510 st
->loaded_container
= 0;
1515 static __u16
info_to_blocks_per_strip(mdu_array_info_t
*info
)
1517 if (info
->level
== 1)
1519 return info
->chunk_size
>> 9;
1522 static __u32
info_to_num_data_stripes(mdu_array_info_t
*info
)
1526 num_stripes
= (info
->size
* 2) / info_to_blocks_per_strip(info
);
1527 if (info
->level
== 1)
1533 static __u32
info_to_blocks_per_member(mdu_array_info_t
*info
)
1535 return (info
->size
* 2) & ~(info_to_blocks_per_strip(info
) - 1);
1538 static int init_super_imsm_volume(struct supertype
*st
, mdu_array_info_t
*info
,
1539 unsigned long long size
, char *name
,
1540 char *homehost
, int *uuid
)
1542 /* We are creating a volume inside a pre-existing container.
1543 * so st->sb is already set.
1545 struct intel_super
*super
= st
->sb
;
1546 struct imsm_super
*mpb
= super
->anchor
;
1547 struct imsm_dev
*dev
;
1548 struct imsm_vol
*vol
;
1549 struct imsm_map
*map
;
1550 int idx
= mpb
->num_raid_devs
;
1552 unsigned long long array_blocks
;
1554 size_t size_old
, size_new
;
1556 if (mpb
->num_raid_devs
>= 2) {
1557 fprintf(stderr
, Name
": This imsm-container already has the "
1558 "maximum of 2 volumes\n");
1562 /* ensure the mpb is large enough for the new data */
1563 size_old
= __le32_to_cpu(mpb
->mpb_size
);
1564 size_new
= disks_to_mpb_size(info
->nr_disks
);
1565 if (size_new
> size_old
) {
1567 size_t size_round
= ROUND_UP(size_new
, 512);
1569 if (posix_memalign(&mpb_new
, 512, size_round
) != 0) {
1570 fprintf(stderr
, Name
": could not allocate new mpb\n");
1573 memcpy(mpb_new
, mpb
, size_old
);
1576 super
->anchor
= mpb_new
;
1577 mpb
->mpb_size
= __cpu_to_le32(size_new
);
1578 memset(mpb_new
+ size_old
, 0, size_round
- size_old
);
1580 super
->current_vol
= idx
;
1581 /* when creating the first raid device in this container set num_disks
1582 * to zero, i.e. delete this spare and add raid member devices in
1583 * add_to_super_imsm_volume()
1585 if (super
->current_vol
== 0)
1587 sprintf(st
->subarray
, "%d", idx
);
1588 dev
= malloc(sizeof(*dev
) + sizeof(__u32
) * (info
->raid_disks
- 1));
1590 fprintf(stderr
, Name
": could not allocate raid device\n");
1593 strncpy((char *) dev
->volume
, name
, MAX_RAID_SERIAL_LEN
);
1594 array_blocks
= calc_array_size(info
->level
, info
->raid_disks
,
1595 info
->layout
, info
->chunk_size
,
1597 dev
->size_low
= __cpu_to_le32((__u32
) array_blocks
);
1598 dev
->size_high
= __cpu_to_le32((__u32
) (array_blocks
>> 32));
1599 dev
->status
= __cpu_to_le32(0);
1600 dev
->reserved_blocks
= __cpu_to_le32(0);
1602 vol
->migr_state
= 0;
1605 for (i
= 0; i
< idx
; i
++) {
1606 struct imsm_dev
*prev
= get_imsm_dev(super
, i
);
1607 struct imsm_map
*pmap
= get_imsm_map(prev
, 0);
1609 offset
+= __le32_to_cpu(pmap
->blocks_per_member
);
1610 offset
+= IMSM_RESERVED_SECTORS
;
1612 map
= get_imsm_map(dev
, 0);
1613 map
->pba_of_lba0
= __cpu_to_le32(offset
);
1614 map
->blocks_per_member
= __cpu_to_le32(info_to_blocks_per_member(info
));
1615 map
->blocks_per_strip
= __cpu_to_le16(info_to_blocks_per_strip(info
));
1616 map
->num_data_stripes
= __cpu_to_le32(info_to_num_data_stripes(info
));
1617 map
->map_state
= info
->level
? IMSM_T_STATE_UNINITIALIZED
:
1618 IMSM_T_STATE_NORMAL
;
1620 if (info
->level
== 1 && info
->raid_disks
> 2) {
1621 fprintf(stderr
, Name
": imsm does not support more than 2 disks"
1622 "in a raid1 volume\n");
1625 if (info
->level
== 10)
1626 map
->raid_level
= 1;
1628 map
->raid_level
= info
->level
;
1630 map
->num_members
= info
->raid_disks
;
1631 for (i
= 0; i
< map
->num_members
; i
++) {
1632 /* initialized in add_to_super */
1633 set_imsm_ord_tbl_ent(map
, i
, 0);
1635 mpb
->num_raid_devs
++;
1636 super
->dev_tbl
[super
->current_vol
] = dev
;
1641 static int init_super_imsm(struct supertype
*st
, mdu_array_info_t
*info
,
1642 unsigned long long size
, char *name
,
1643 char *homehost
, int *uuid
)
1645 /* This is primarily called by Create when creating a new array.
1646 * We will then get add_to_super called for each component, and then
1647 * write_init_super called to write it out to each device.
1648 * For IMSM, Create can create on fresh devices or on a pre-existing
1650 * To create on a pre-existing array a different method will be called.
1651 * This one is just for fresh drives.
1653 struct intel_super
*super
;
1654 struct imsm_super
*mpb
;
1662 return init_super_imsm_volume(st
, info
, size
, name
, homehost
,
1665 super
= alloc_super(1);
1668 mpb_size
= disks_to_mpb_size(info
->nr_disks
);
1669 if (posix_memalign(&super
->buf
, 512, mpb_size
) != 0) {
1674 memset(mpb
, 0, mpb_size
);
1676 memcpy(mpb
->sig
, MPB_SIGNATURE
, strlen(MPB_SIGNATURE
));
1677 memcpy(mpb
->sig
+ strlen(MPB_SIGNATURE
), MPB_VERSION_RAID5
,
1678 strlen(MPB_VERSION_RAID5
));
1679 mpb
->mpb_size
= mpb_size
;
1686 static void add_to_super_imsm_volume(struct supertype
*st
, mdu_disk_info_t
*dk
,
1687 int fd
, char *devname
)
1689 struct intel_super
*super
= st
->sb
;
1690 struct imsm_super
*mpb
= super
->anchor
;
1692 struct imsm_dev
*dev
;
1693 struct imsm_map
*map
;
1696 dev
= get_imsm_dev(super
, super
->current_vol
);
1697 map
= get_imsm_map(dev
, 0);
1699 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
1700 if (dl
->major
== dk
->major
&&
1701 dl
->minor
== dk
->minor
)
1704 if (!dl
|| ! (dk
->state
& (1<<MD_DISK_SYNC
)))
1707 /* add a pristine spare to the metadata */
1708 if (dl
->index
< 0) {
1709 dl
->index
= super
->anchor
->num_disks
;
1710 super
->anchor
->num_disks
++;
1712 set_imsm_ord_tbl_ent(map
, dk
->number
, dl
->index
);
1713 status
= CONFIGURED_DISK
| USABLE_DISK
;
1714 dl
->disk
.status
= __cpu_to_le32(status
);
1716 /* if we are creating the first raid device update the family number */
1717 if (super
->current_vol
== 0) {
1719 struct imsm_dev
*_dev
= __get_imsm_dev(mpb
, 0);
1720 struct imsm_disk
*_disk
= __get_imsm_disk(mpb
, dl
->index
);
1724 sum
= __gen_imsm_checksum(mpb
);
1725 mpb
->family_num
= __cpu_to_le32(sum
);
1729 static void add_to_super_imsm(struct supertype
*st
, mdu_disk_info_t
*dk
,
1730 int fd
, char *devname
)
1732 struct intel_super
*super
= st
->sb
;
1734 unsigned long long size
;
1739 if (super
->current_vol
>= 0) {
1740 add_to_super_imsm_volume(st
, dk
, fd
, devname
);
1745 dd
= malloc(sizeof(*dd
));
1748 Name
": malloc failed %s:%d.\n", __func__
, __LINE__
);
1751 memset(dd
, 0, sizeof(*dd
));
1752 dd
->major
= major(stb
.st_rdev
);
1753 dd
->minor
= minor(stb
.st_rdev
);
1755 dd
->devname
= devname
? strdup(devname
) : NULL
;
1757 rv
= imsm_read_serial(fd
, devname
, dd
->serial
);
1760 Name
": failed to retrieve scsi serial, aborting\n");
1765 get_dev_size(fd
, NULL
, &size
);
1767 status
= USABLE_DISK
| SPARE_DISK
;
1768 serialcpy(dd
->disk
.serial
, dd
->serial
);
1769 dd
->disk
.total_blocks
= __cpu_to_le32(size
);
1770 dd
->disk
.status
= __cpu_to_le32(status
);
1771 if (sysfs_disk_to_scsi_id(fd
, &id
) == 0)
1772 dd
->disk
.scsi_id
= __cpu_to_le32(id
);
1774 dd
->disk
.scsi_id
= __cpu_to_le32(0);
1776 if (st
->update_tail
) {
1777 dd
->next
= super
->add
;
1780 dd
->next
= super
->disks
;
1785 static int store_imsm_mpb(int fd
, struct intel_super
*super
);
1787 /* spare records have their own family number and do not have any defined raid
1790 static int write_super_imsm_spares(struct intel_super
*super
, int doclose
)
1792 struct imsm_super mpb_save
;
1793 struct imsm_super
*mpb
= super
->anchor
;
1798 mpb
->num_raid_devs
= 0;
1800 mpb
->mpb_size
= sizeof(struct imsm_super
);
1801 mpb
->generation_num
= __cpu_to_le32(1UL);
1803 for (d
= super
->disks
; d
; d
= d
->next
) {
1807 mpb
->disk
[0] = d
->disk
;
1808 sum
= __gen_imsm_checksum(mpb
);
1809 mpb
->family_num
= __cpu_to_le32(sum
);
1810 sum
= __gen_imsm_checksum(mpb
);
1811 mpb
->check_sum
= __cpu_to_le32(sum
);
1813 if (store_imsm_mpb(d
->fd
, super
)) {
1814 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1815 __func__
, d
->major
, d
->minor
, strerror(errno
));
1829 static int write_super_imsm(struct intel_super
*super
, int doclose
)
1831 struct imsm_super
*mpb
= super
->anchor
;
1837 __u32 mpb_size
= sizeof(struct imsm_super
) - sizeof(struct imsm_disk
);
1839 /* 'generation' is incremented everytime the metadata is written */
1840 generation
= __le32_to_cpu(mpb
->generation_num
);
1842 mpb
->generation_num
= __cpu_to_le32(generation
);
1844 for (d
= super
->disks
; d
; d
= d
->next
) {
1848 mpb
->disk
[d
->index
] = d
->disk
;
1849 mpb_size
+= sizeof(struct imsm_disk
);
1853 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
1854 struct imsm_dev
*dev
= __get_imsm_dev(mpb
, i
);
1856 imsm_copy_dev(dev
, super
->dev_tbl
[i
]);
1857 mpb_size
+= sizeof_imsm_dev(dev
, 0);
1859 mpb_size
+= __le32_to_cpu(mpb
->bbm_log_size
);
1860 mpb
->mpb_size
= __cpu_to_le32(mpb_size
);
1862 /* recalculate checksum */
1863 sum
= __gen_imsm_checksum(mpb
);
1864 mpb
->check_sum
= __cpu_to_le32(sum
);
1866 /* write the mpb for disks that compose raid devices */
1867 for (d
= super
->disks
; d
; d
= d
->next
) {
1870 if (store_imsm_mpb(d
->fd
, super
))
1871 fprintf(stderr
, "%s: failed for device %d:%d %s\n",
1872 __func__
, d
->major
, d
->minor
, strerror(errno
));
1880 return write_super_imsm_spares(super
, doclose
);
1886 static int create_array(struct supertype
*st
)
1889 struct imsm_update_create_array
*u
;
1890 struct intel_super
*super
= st
->sb
;
1891 struct imsm_dev
*dev
= get_imsm_dev(super
, super
->current_vol
);
1893 len
= sizeof(*u
) - sizeof(*dev
) + sizeof_imsm_dev(dev
, 0);
1896 fprintf(stderr
, "%s: failed to allocate update buffer\n",
1901 u
->type
= update_create_array
;
1902 u
->dev_idx
= super
->current_vol
;
1903 imsm_copy_dev(&u
->dev
, dev
);
1904 append_metadata_update(st
, u
, len
);
1909 static int _add_disk(struct supertype
*st
)
1911 struct intel_super
*super
= st
->sb
;
1913 struct imsm_update_add_disk
*u
;
1921 fprintf(stderr
, "%s: failed to allocate update buffer\n",
1926 u
->type
= update_add_disk
;
1927 append_metadata_update(st
, u
, len
);
1932 static int write_init_super_imsm(struct supertype
*st
)
1934 if (st
->update_tail
) {
1935 /* queue the recently created array / added disk
1936 * as a metadata update */
1937 struct intel_super
*super
= st
->sb
;
1941 /* determine if we are creating a volume or adding a disk */
1942 if (super
->current_vol
< 0) {
1943 /* in the add disk case we are running in mdmon
1944 * context, so don't close fd's
1946 return _add_disk(st
);
1948 rv
= create_array(st
);
1950 for (d
= super
->disks
; d
; d
= d
->next
) {
1957 return write_super_imsm(st
->sb
, 1);
1961 static int store_zero_imsm(struct supertype
*st
, int fd
)
1963 unsigned long long dsize
;
1966 get_dev_size(fd
, NULL
, &dsize
);
1968 /* first block is stored on second to last sector of the disk */
1969 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
1972 if (posix_memalign(&buf
, 512, 512) != 0)
1975 memset(buf
, 0, 512);
1976 if (write(fd
, buf
, 512) != 512)
1981 static int imsm_bbm_log_size(struct imsm_super
*mpb
)
1983 return __le32_to_cpu(mpb
->bbm_log_size
);
1987 static int validate_geometry_imsm_container(struct supertype
*st
, int level
,
1988 int layout
, int raiddisks
, int chunk
,
1989 unsigned long long size
, char *dev
,
1990 unsigned long long *freesize
,
1994 unsigned long long ldsize
;
1996 if (level
!= LEVEL_CONTAINER
)
2001 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2004 fprintf(stderr
, Name
": imsm: Cannot open %s: %s\n",
2005 dev
, strerror(errno
));
2008 if (!get_dev_size(fd
, dev
, &ldsize
)) {
2014 *freesize
= avail_size_imsm(st
, ldsize
>> 9);
2019 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2020 * FIX ME add ahci details
2022 static int validate_geometry_imsm_volume(struct supertype
*st
, int level
,
2023 int layout
, int raiddisks
, int chunk
,
2024 unsigned long long size
, char *dev
,
2025 unsigned long long *freesize
,
2029 struct intel_super
*super
= st
->sb
;
2031 unsigned long long pos
= 0;
2032 unsigned long long maxsize
;
2036 if (level
== LEVEL_CONTAINER
)
2039 if (level
== 1 && raiddisks
> 2) {
2041 fprintf(stderr
, Name
": imsm does not support more "
2042 "than 2 in a raid1 configuration\n");
2046 /* We must have the container info already read in. */
2051 /* General test: make sure there is space for
2052 * 'raiddisks' device extents of size 'size' at a given
2055 unsigned long long minsize
= size
*2 /* convert to blocks */;
2056 unsigned long long start_offset
= ~0ULL;
2059 minsize
= MPB_SECTOR_CNT
+ IMSM_RESERVED_SECTORS
;
2060 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2065 e
= get_extents(super
, dl
);
2068 unsigned long long esize
;
2069 esize
= e
[i
].start
- pos
;
2070 if (esize
>= minsize
)
2072 if (found
&& start_offset
== ~0ULL) {
2075 } else if (found
&& pos
!= start_offset
) {
2079 pos
= e
[i
].start
+ e
[i
].size
;
2081 } while (e
[i
-1].size
);
2086 if (dcnt
< raiddisks
) {
2088 fprintf(stderr
, Name
": imsm: Not enough "
2089 "devices with space for this array "
2096 /* This device must be a member of the set */
2097 if (stat(dev
, &stb
) < 0)
2099 if ((S_IFMT
& stb
.st_mode
) != S_IFBLK
)
2101 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2102 if (dl
->major
== major(stb
.st_rdev
) &&
2103 dl
->minor
== minor(stb
.st_rdev
))
2108 fprintf(stderr
, Name
": %s is not in the "
2109 "same imsm set\n", dev
);
2112 e
= get_extents(super
, dl
);
2116 unsigned long long esize
;
2117 esize
= e
[i
].start
- pos
;
2118 if (esize
>= maxsize
)
2120 pos
= e
[i
].start
+ e
[i
].size
;
2122 } while (e
[i
-1].size
);
2123 *freesize
= maxsize
;
2128 static int validate_geometry_imsm(struct supertype
*st
, int level
, int layout
,
2129 int raiddisks
, int chunk
, unsigned long long size
,
2130 char *dev
, unsigned long long *freesize
,
2136 /* if given unused devices create a container
2137 * if given given devices in a container create a member volume
2139 if (level
== LEVEL_CONTAINER
) {
2140 /* Must be a fresh device to add to a container */
2141 return validate_geometry_imsm_container(st
, level
, layout
,
2142 raiddisks
, chunk
, size
,
2148 /* creating in a given container */
2149 return validate_geometry_imsm_volume(st
, level
, layout
,
2150 raiddisks
, chunk
, size
,
2151 dev
, freesize
, verbose
);
2154 /* limit creation to the following levels */
2166 /* This device needs to be a device in an 'imsm' container */
2167 fd
= open(dev
, O_RDONLY
|O_EXCL
, 0);
2171 Name
": Cannot create this array on device %s\n",
2176 if (errno
!= EBUSY
|| (fd
= open(dev
, O_RDONLY
, 0)) < 0) {
2178 fprintf(stderr
, Name
": Cannot open %s: %s\n",
2179 dev
, strerror(errno
));
2182 /* Well, it is in use by someone, maybe an 'imsm' container. */
2183 cfd
= open_container(fd
);
2187 fprintf(stderr
, Name
": Cannot use %s: It is busy\n",
2191 sra
= sysfs_read(cfd
, 0, GET_VERSION
);
2193 if (sra
&& sra
->array
.major_version
== -1 &&
2194 strcmp(sra
->text_version
, "imsm") == 0) {
2195 /* This is a member of a imsm container. Load the container
2196 * and try to create a volume
2198 struct intel_super
*super
;
2200 if (load_super_imsm_all(st
, cfd
, (void **) &super
, NULL
, 1) == 0) {
2202 st
->container_dev
= fd2devnum(cfd
);
2204 return validate_geometry_imsm_volume(st
, level
, layout
,
2210 } else /* may belong to another container */
2215 #endif /* MDASSEMBLE */
2217 static struct mdinfo
*container_content_imsm(struct supertype
*st
)
2219 /* Given a container loaded by load_super_imsm_all,
2220 * extract information about all the arrays into
2223 * For each imsm_dev create an mdinfo, fill it in,
2224 * then look for matching devices in super->disks
2225 * and create appropriate device mdinfo.
2227 struct intel_super
*super
= st
->sb
;
2228 struct imsm_super
*mpb
= super
->anchor
;
2229 struct mdinfo
*rest
= NULL
;
2232 /* do not assemble arrays that might have bad blocks */
2233 if (imsm_bbm_log_size(super
->anchor
)) {
2234 fprintf(stderr
, Name
": BBM log found in metadata. "
2235 "Cannot activate array(s).\n");
2239 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
2240 struct imsm_dev
*dev
= get_imsm_dev(super
, i
);
2241 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2242 struct mdinfo
*this;
2245 this = malloc(sizeof(*this));
2246 memset(this, 0, sizeof(*this));
2249 super
->current_vol
= i
;
2250 getinfo_super_imsm_volume(st
, this);
2251 for (slot
= 0 ; slot
< map
->num_members
; slot
++) {
2252 struct mdinfo
*info_d
;
2260 idx
= get_imsm_disk_idx(dev
, slot
);
2261 ord
= get_imsm_ord_tbl_ent(dev
, slot
);
2262 for (d
= super
->disks
; d
; d
= d
->next
)
2263 if (d
->index
== idx
)
2269 s
= d
? __le32_to_cpu(d
->disk
.status
) : 0;
2270 if (s
& FAILED_DISK
)
2272 if (!(s
& USABLE_DISK
))
2274 if (ord
& IMSM_ORD_REBUILD
)
2278 * if we skip some disks the array will be assmebled degraded;
2279 * reset resync start to avoid a dirty-degraded situation
2281 * FIXME handle dirty degraded
2283 if (skip
&& !dev
->vol
.dirty
)
2284 this->resync_start
= ~0ULL;
2288 info_d
= malloc(sizeof(*info_d
));
2290 fprintf(stderr
, Name
": failed to allocate disk"
2291 " for volume %s\n", (char *) dev
->volume
);
2296 memset(info_d
, 0, sizeof(*info_d
));
2297 info_d
->next
= this->devs
;
2298 this->devs
= info_d
;
2300 info_d
->disk
.number
= d
->index
;
2301 info_d
->disk
.major
= d
->major
;
2302 info_d
->disk
.minor
= d
->minor
;
2303 info_d
->disk
.raid_disk
= slot
;
2305 this->array
.working_disks
++;
2307 info_d
->events
= __le32_to_cpu(mpb
->generation_num
);
2308 info_d
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2309 info_d
->component_size
= __le32_to_cpu(map
->blocks_per_member
);
2311 strcpy(info_d
->name
, d
->devname
);
2321 static int imsm_open_new(struct supertype
*c
, struct active_array
*a
,
2324 struct intel_super
*super
= c
->sb
;
2325 struct imsm_super
*mpb
= super
->anchor
;
2327 if (atoi(inst
) >= mpb
->num_raid_devs
) {
2328 fprintf(stderr
, "%s: subarry index %d, out of range\n",
2329 __func__
, atoi(inst
));
2333 dprintf("imsm: open_new %s\n", inst
);
2334 a
->info
.container_member
= atoi(inst
);
2338 static __u8
imsm_check_degraded(struct intel_super
*super
, struct imsm_dev
*dev
, int failed
)
2340 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2343 return map
->map_state
== IMSM_T_STATE_UNINITIALIZED
?
2344 IMSM_T_STATE_UNINITIALIZED
: IMSM_T_STATE_NORMAL
;
2346 switch (get_imsm_raid_level(map
)) {
2348 return IMSM_T_STATE_FAILED
;
2351 if (failed
< map
->num_members
)
2352 return IMSM_T_STATE_DEGRADED
;
2354 return IMSM_T_STATE_FAILED
;
2359 * check to see if any mirrors have failed, otherwise we
2360 * are degraded. Even numbered slots are mirrored on
2366 for (i
= 0; i
< map
->num_members
; i
++) {
2367 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
2368 int idx
= ord_to_idx(ord
);
2369 struct imsm_disk
*disk
;
2371 /* reset the potential in-sync count on even-numbered
2372 * slots. num_copies is always 2 for imsm raid10
2377 disk
= get_imsm_disk(super
, idx
);
2379 __le32_to_cpu(disk
->status
) & FAILED_DISK
||
2380 ord
& IMSM_ORD_REBUILD
)
2383 /* no in-sync disks left in this mirror the
2387 return IMSM_T_STATE_FAILED
;
2390 return IMSM_T_STATE_DEGRADED
;
2394 return IMSM_T_STATE_DEGRADED
;
2396 return IMSM_T_STATE_FAILED
;
2402 return map
->map_state
;
2405 static int imsm_count_failed(struct intel_super
*super
, struct imsm_dev
*dev
)
2409 struct imsm_disk
*disk
;
2410 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2412 for (i
= 0; i
< map
->num_members
; i
++) {
2413 __u32 ord
= get_imsm_ord_tbl_ent(dev
, i
);
2414 int idx
= ord_to_idx(ord
);
2416 disk
= get_imsm_disk(super
, idx
);
2418 __le32_to_cpu(disk
->status
) & FAILED_DISK
||
2419 ord
& IMSM_ORD_REBUILD
)
2426 static int is_resyncing(struct imsm_dev
*dev
)
2428 struct imsm_map
*migr_map
;
2430 if (!dev
->vol
.migr_state
)
2433 if (dev
->vol
.migr_type
== 0)
2436 migr_map
= get_imsm_map(dev
, 1);
2438 if (migr_map
->map_state
== IMSM_T_STATE_NORMAL
)
2444 static int is_rebuilding(struct imsm_dev
*dev
)
2446 struct imsm_map
*migr_map
;
2448 if (!dev
->vol
.migr_state
)
2451 if (dev
->vol
.migr_type
== 0)
2454 migr_map
= get_imsm_map(dev
, 1);
2456 if (migr_map
->map_state
== IMSM_T_STATE_DEGRADED
)
2462 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
2463 * states are handled in imsm_set_disk() with one exception, when a
2464 * resync is stopped due to a new failure this routine will set the
2465 * 'degraded' state for the array.
2467 static int imsm_set_array_state(struct active_array
*a
, int consistent
)
2469 int inst
= a
->info
.container_member
;
2470 struct intel_super
*super
= a
->container
->sb
;
2471 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2472 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2473 int failed
= imsm_count_failed(super
, dev
);
2474 __u8 map_state
= imsm_check_degraded(super
, dev
, failed
);
2476 if (consistent
== 2 &&
2477 (!is_resync_complete(a
) ||
2478 map_state
!= IMSM_T_STATE_NORMAL
||
2479 dev
->vol
.migr_state
))
2482 if (is_resync_complete(a
)) {
2483 /* complete intialization / resync,
2484 * recovery is completed in ->set_disk
2486 if (is_resyncing(dev
)) {
2487 dprintf("imsm: mark resync done\n");
2488 dev
->vol
.migr_state
= 0;
2489 map
->map_state
= map_state
;
2490 super
->updates_pending
++;
2492 } else if (!is_resyncing(dev
) && !failed
) {
2493 /* mark the start of the init process if nothing is failed */
2494 dprintf("imsm: mark resync start (%llu)\n", a
->resync_start
);
2495 map
->map_state
= map_state
;
2496 migrate(dev
, IMSM_T_STATE_NORMAL
,
2497 map
->map_state
== IMSM_T_STATE_NORMAL
);
2498 super
->updates_pending
++;
2501 /* mark dirty / clean */
2502 if (dev
->vol
.dirty
!= !consistent
) {
2503 dprintf("imsm: mark '%s' (%llu)\n",
2504 consistent
? "clean" : "dirty", a
->resync_start
);
2509 super
->updates_pending
++;
2514 static void imsm_set_disk(struct active_array
*a
, int n
, int state
)
2516 int inst
= a
->info
.container_member
;
2517 struct intel_super
*super
= a
->container
->sb
;
2518 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2519 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2520 struct imsm_disk
*disk
;
2526 if (n
> map
->num_members
)
2527 fprintf(stderr
, "imsm: set_disk %d out of range 0..%d\n",
2528 n
, map
->num_members
- 1);
2533 dprintf("imsm: set_disk %d:%x\n", n
, state
);
2535 ord
= get_imsm_ord_tbl_ent(dev
, n
);
2536 disk
= get_imsm_disk(super
, ord_to_idx(ord
));
2538 /* check for new failures */
2539 status
= __le32_to_cpu(disk
->status
);
2540 if ((state
& DS_FAULTY
) && !(status
& FAILED_DISK
)) {
2541 status
|= FAILED_DISK
;
2542 disk
->status
= __cpu_to_le32(status
);
2543 disk
->scsi_id
= __cpu_to_le32(~(__u32
)0);
2544 memmove(&disk
->serial
[0], &disk
->serial
[1], MAX_RAID_SERIAL_LEN
- 1);
2545 super
->updates_pending
++;
2547 /* check if in_sync */
2548 if (state
& DS_INSYNC
&& ord
& IMSM_ORD_REBUILD
) {
2549 struct imsm_map
*migr_map
= get_imsm_map(dev
, 1);
2551 set_imsm_ord_tbl_ent(migr_map
, n
, ord_to_idx(ord
));
2552 super
->updates_pending
++;
2555 failed
= imsm_count_failed(super
, dev
);
2556 map_state
= imsm_check_degraded(super
, dev
, failed
);
2558 /* check if recovery complete, newly degraded, or failed */
2559 if (map_state
== IMSM_T_STATE_NORMAL
&& is_rebuilding(dev
)) {
2560 map
->map_state
= map_state
;
2561 dev
->vol
.migr_state
= 0;
2562 super
->updates_pending
++;
2563 } else if (map_state
== IMSM_T_STATE_DEGRADED
&&
2564 map
->map_state
!= map_state
&&
2565 !dev
->vol
.migr_state
) {
2566 dprintf("imsm: mark degraded\n");
2567 map
->map_state
= map_state
;
2568 super
->updates_pending
++;
2569 } else if (map_state
== IMSM_T_STATE_FAILED
&&
2570 map
->map_state
!= map_state
) {
2571 dprintf("imsm: mark failed\n");
2572 dev
->vol
.migr_state
= 0;
2573 map
->map_state
= map_state
;
2574 super
->updates_pending
++;
2578 static int store_imsm_mpb(int fd
, struct intel_super
*super
)
2580 struct imsm_super
*mpb
= super
->anchor
;
2581 __u32 mpb_size
= __le32_to_cpu(mpb
->mpb_size
);
2582 unsigned long long dsize
;
2583 unsigned long long sectors
;
2585 get_dev_size(fd
, NULL
, &dsize
);
2587 if (mpb_size
> 512) {
2588 /* -1 to account for anchor */
2589 sectors
= mpb_sectors(mpb
) - 1;
2591 /* write the extended mpb to the sectors preceeding the anchor */
2592 if (lseek64(fd
, dsize
- (512 * (2 + sectors
)), SEEK_SET
) < 0)
2595 if (write(fd
, super
->buf
+ 512, 512 * sectors
) != 512 * sectors
)
2599 /* first block is stored on second to last sector of the disk */
2600 if (lseek64(fd
, dsize
- (512 * 2), SEEK_SET
) < 0)
2603 if (write(fd
, super
->buf
, 512) != 512)
2609 static void imsm_sync_metadata(struct supertype
*container
)
2611 struct intel_super
*super
= container
->sb
;
2613 if (!super
->updates_pending
)
2616 write_super_imsm(super
, 0);
2618 super
->updates_pending
= 0;
2621 static struct dl
*imsm_readd(struct intel_super
*super
, int idx
, struct active_array
*a
)
2623 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2624 int i
= get_imsm_disk_idx(dev
, idx
);
2627 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2631 if (dl
&& __le32_to_cpu(dl
->disk
.status
) & FAILED_DISK
)
2635 dprintf("%s: found %x:%x\n", __func__
, dl
->major
, dl
->minor
);
2640 static struct dl
*imsm_add_spare(struct intel_super
*super
, int slot
, struct active_array
*a
)
2642 struct imsm_dev
*dev
= get_imsm_dev(super
, a
->info
.container_member
);
2643 int idx
= get_imsm_disk_idx(dev
, slot
);
2644 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2645 unsigned long long esize
;
2646 unsigned long long pos
;
2655 for (dl
= super
->disks
; dl
; dl
= dl
->next
) {
2656 /* If in this array, skip */
2657 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2658 if (d
->state_fd
>= 0 &&
2659 d
->disk
.major
== dl
->major
&&
2660 d
->disk
.minor
== dl
->minor
) {
2661 dprintf("%x:%x already in array\n", dl
->major
, dl
->minor
);
2667 /* skip in use or failed drives */
2668 status
= __le32_to_cpu(dl
->disk
.status
);
2669 if (status
& FAILED_DISK
|| idx
== dl
->index
) {
2670 dprintf("%x:%x status ( %s%s)\n",
2671 dl
->major
, dl
->minor
,
2672 status
& FAILED_DISK
? "failed " : "",
2673 idx
== dl
->index
? "in use " : "");
2677 /* Does this unused device have the requisite free space?
2678 * We need a->info.component_size sectors
2680 ex
= get_extents(super
, dl
);
2682 dprintf("cannot get extents\n");
2688 array_start
= __le32_to_cpu(map
->pba_of_lba0
);
2691 /* check that we can start at pba_of_lba0 with
2692 * a->info.component_size of space
2694 esize
= ex
[j
].start
- pos
;
2695 if (array_start
>= pos
&&
2696 array_start
+ a
->info
.component_size
< ex
[j
].start
) {
2700 pos
= ex
[j
].start
+ ex
[j
].size
;
2703 } while (ex
[j
-1].size
);
2707 dprintf("%x:%x does not have %llu at %d\n",
2708 dl
->major
, dl
->minor
,
2709 a
->info
.component_size
,
2710 __le32_to_cpu(map
->pba_of_lba0
));
2720 static struct mdinfo
*imsm_activate_spare(struct active_array
*a
,
2721 struct metadata_update
**updates
)
2724 * Find a device with unused free space and use it to replace a
2725 * failed/vacant region in an array. We replace failed regions one a
2726 * array at a time. The result is that a new spare disk will be added
2727 * to the first failed array and after the monitor has finished
2728 * propagating failures the remainder will be consumed.
2730 * FIXME add a capability for mdmon to request spares from another
2734 struct intel_super
*super
= a
->container
->sb
;
2735 int inst
= a
->info
.container_member
;
2736 struct imsm_dev
*dev
= get_imsm_dev(super
, inst
);
2737 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2738 int failed
= a
->info
.array
.raid_disks
;
2739 struct mdinfo
*rv
= NULL
;
2742 struct metadata_update
*mu
;
2744 struct imsm_update_activate_spare
*u
;
2748 for (d
= a
->info
.devs
; d
; d
= d
->next
) {
2749 if ((d
->curr_state
& DS_FAULTY
) &&
2751 /* wait for Removal to happen */
2753 if (d
->state_fd
>= 0)
2757 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
2758 inst
, failed
, a
->info
.array
.raid_disks
, a
->info
.array
.level
);
2759 if (imsm_check_degraded(super
, dev
, failed
) != IMSM_T_STATE_DEGRADED
)
2762 /* For each slot, if it is not working, find a spare */
2763 for (i
= 0; i
< a
->info
.array
.raid_disks
; i
++) {
2764 for (d
= a
->info
.devs
; d
; d
= d
->next
)
2765 if (d
->disk
.raid_disk
== i
)
2767 dprintf("found %d: %p %x\n", i
, d
, d
?d
->curr_state
:0);
2768 if (d
&& (d
->state_fd
>= 0))
2772 * OK, this device needs recovery. Try to re-add the previous
2773 * occupant of this slot, if this fails add a new spare
2775 dl
= imsm_readd(super
, i
, a
);
2777 dl
= imsm_add_spare(super
, i
, a
);
2781 /* found a usable disk with enough space */
2782 di
= malloc(sizeof(*di
));
2783 memset(di
, 0, sizeof(*di
));
2785 /* dl->index will be -1 in the case we are activating a
2786 * pristine spare. imsm_process_update() will create a
2787 * new index in this case. Once a disk is found to be
2788 * failed in all member arrays it is kicked from the
2791 di
->disk
.number
= dl
->index
;
2793 /* (ab)use di->devs to store a pointer to the device
2796 di
->devs
= (struct mdinfo
*) dl
;
2798 di
->disk
.raid_disk
= i
;
2799 di
->disk
.major
= dl
->major
;
2800 di
->disk
.minor
= dl
->minor
;
2802 di
->data_offset
= __le32_to_cpu(map
->pba_of_lba0
);
2803 di
->component_size
= a
->info
.component_size
;
2804 di
->container_member
= inst
;
2808 dprintf("%x:%x to be %d at %llu\n", dl
->major
, dl
->minor
,
2809 i
, di
->data_offset
);
2815 /* No spares found */
2817 /* Now 'rv' has a list of devices to return.
2818 * Create a metadata_update record to update the
2819 * disk_ord_tbl for the array
2821 mu
= malloc(sizeof(*mu
));
2822 mu
->buf
= malloc(sizeof(struct imsm_update_activate_spare
) * num_spares
);
2824 mu
->len
= sizeof(struct imsm_update_activate_spare
) * num_spares
;
2825 mu
->next
= *updates
;
2826 u
= (struct imsm_update_activate_spare
*) mu
->buf
;
2828 for (di
= rv
; di
; di
= di
->next
) {
2829 u
->type
= update_activate_spare
;
2830 u
->dl
= (struct dl
*) di
->devs
;
2832 u
->slot
= di
->disk
.raid_disk
;
2843 static int disks_overlap(struct imsm_dev
*d1
, struct imsm_dev
*d2
)
2845 struct imsm_map
*m1
= get_imsm_map(d1
, 0);
2846 struct imsm_map
*m2
= get_imsm_map(d2
, 0);
2851 for (i
= 0; i
< m1
->num_members
; i
++) {
2852 idx
= get_imsm_disk_idx(d1
, i
);
2853 for (j
= 0; j
< m2
->num_members
; j
++)
2854 if (idx
== get_imsm_disk_idx(d2
, j
))
2861 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
);
2863 static void imsm_process_update(struct supertype
*st
,
2864 struct metadata_update
*update
)
2867 * crack open the metadata_update envelope to find the update record
2868 * update can be one of:
2869 * update_activate_spare - a spare device has replaced a failed
2870 * device in an array, update the disk_ord_tbl. If this disk is
2871 * present in all member arrays then also clear the SPARE_DISK
2874 struct intel_super
*super
= st
->sb
;
2875 struct imsm_super
*mpb
;
2876 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
2878 /* update requires a larger buf but the allocation failed */
2879 if (super
->next_len
&& !super
->next_buf
) {
2880 super
->next_len
= 0;
2884 if (super
->next_buf
) {
2885 memcpy(super
->next_buf
, super
->buf
, super
->len
);
2887 super
->len
= super
->next_len
;
2888 super
->buf
= super
->next_buf
;
2890 super
->next_len
= 0;
2891 super
->next_buf
= NULL
;
2894 mpb
= super
->anchor
;
2897 case update_activate_spare
: {
2898 struct imsm_update_activate_spare
*u
= (void *) update
->buf
;
2899 struct imsm_dev
*dev
= get_imsm_dev(super
, u
->array
);
2900 struct imsm_map
*map
= get_imsm_map(dev
, 0);
2901 struct imsm_map
*migr_map
;
2902 struct active_array
*a
;
2903 struct imsm_disk
*disk
;
2909 int victim
= get_imsm_disk_idx(dev
, u
->slot
);
2912 for (dl
= super
->disks
; dl
; dl
= dl
->next
)
2917 fprintf(stderr
, "error: imsm_activate_spare passed "
2918 "an unknown disk (index: %d)\n",
2923 super
->updates_pending
++;
2925 /* count failures (excluding rebuilds and the victim)
2926 * to determine map[0] state
2929 for (i
= 0; i
< map
->num_members
; i
++) {
2932 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
2934 __le32_to_cpu(disk
->status
) & FAILED_DISK
)
2938 /* adding a pristine spare, assign a new index */
2939 if (dl
->index
< 0) {
2940 dl
->index
= super
->anchor
->num_disks
;
2941 super
->anchor
->num_disks
++;
2944 status
= __le32_to_cpu(disk
->status
);
2945 status
|= CONFIGURED_DISK
;
2946 status
&= ~SPARE_DISK
;
2947 disk
->status
= __cpu_to_le32(status
);
2950 to_state
= imsm_check_degraded(super
, dev
, failed
);
2951 map
->map_state
= IMSM_T_STATE_DEGRADED
;
2952 migrate(dev
, to_state
, 1);
2953 migr_map
= get_imsm_map(dev
, 1);
2954 set_imsm_ord_tbl_ent(map
, u
->slot
, dl
->index
);
2955 set_imsm_ord_tbl_ent(migr_map
, u
->slot
, dl
->index
| IMSM_ORD_REBUILD
);
2957 /* count arrays using the victim in the metadata */
2959 for (a
= st
->arrays
; a
; a
= a
->next
) {
2960 dev
= get_imsm_dev(super
, a
->info
.container_member
);
2961 for (i
= 0; i
< map
->num_members
; i
++)
2962 if (victim
== get_imsm_disk_idx(dev
, i
))
2966 /* delete the victim if it is no longer being
2972 for (dlp
= &super
->disks
; *dlp
; dlp
= &(*dlp
)->next
)
2973 if ((*dlp
)->index
== victim
)
2975 /* We know that 'manager' isn't touching anything,
2978 imsm_delete(super
, dlp
, victim
);
2982 case update_create_array
: {
2983 /* someone wants to create a new array, we need to be aware of
2984 * a few races/collisions:
2985 * 1/ 'Create' called by two separate instances of mdadm
2986 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
2987 * devices that have since been assimilated via
2989 * In the event this update can not be carried out mdadm will
2990 * (FIX ME) notice that its update did not take hold.
2992 struct imsm_update_create_array
*u
= (void *) update
->buf
;
2993 struct imsm_dev
*dev
;
2994 struct imsm_map
*map
, *new_map
;
2995 unsigned long long start
, end
;
2996 unsigned long long new_start
, new_end
;
3000 /* handle racing creates: first come first serve */
3001 if (u
->dev_idx
< mpb
->num_raid_devs
) {
3002 dprintf("%s: subarray %d already defined\n",
3003 __func__
, u
->dev_idx
);
3007 /* check update is next in sequence */
3008 if (u
->dev_idx
!= mpb
->num_raid_devs
) {
3009 dprintf("%s: can not create array %d expected index %d\n",
3010 __func__
, u
->dev_idx
, mpb
->num_raid_devs
);
3014 new_map
= get_imsm_map(&u
->dev
, 0);
3015 new_start
= __le32_to_cpu(new_map
->pba_of_lba0
);
3016 new_end
= new_start
+ __le32_to_cpu(new_map
->blocks_per_member
);
3018 /* handle activate_spare versus create race:
3019 * check to make sure that overlapping arrays do not include
3022 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3023 dev
= get_imsm_dev(super
, i
);
3024 map
= get_imsm_map(dev
, 0);
3025 start
= __le32_to_cpu(map
->pba_of_lba0
);
3026 end
= start
+ __le32_to_cpu(map
->blocks_per_member
);
3027 if ((new_start
>= start
&& new_start
<= end
) ||
3028 (start
>= new_start
&& start
<= new_end
))
3030 if (overlap
&& disks_overlap(dev
, &u
->dev
)) {
3031 dprintf("%s: arrays overlap\n", __func__
);
3035 /* check num_members sanity */
3036 if (new_map
->num_members
> mpb
->num_disks
) {
3037 dprintf("%s: num_disks out of range\n", __func__
);
3041 /* check that prepare update was successful */
3042 if (!update
->space
) {
3043 dprintf("%s: prepare update failed\n", __func__
);
3047 super
->updates_pending
++;
3048 dev
= update
->space
;
3049 map
= get_imsm_map(dev
, 0);
3050 update
->space
= NULL
;
3051 imsm_copy_dev(dev
, &u
->dev
);
3052 map
= get_imsm_map(dev
, 0);
3053 super
->dev_tbl
[u
->dev_idx
] = dev
;
3054 mpb
->num_raid_devs
++;
3057 for (i
= 0; i
< map
->num_members
; i
++) {
3058 struct imsm_disk
*disk
;
3061 disk
= get_imsm_disk(super
, get_imsm_disk_idx(dev
, i
));
3062 status
= __le32_to_cpu(disk
->status
);
3063 status
|= CONFIGURED_DISK
;
3064 status
&= ~SPARE_DISK
;
3065 disk
->status
= __cpu_to_le32(status
);
3069 case update_add_disk
:
3071 /* we may be able to repair some arrays if disks are
3074 struct active_array
*a
;
3075 for (a
= st
->arrays
; a
; a
= a
->next
)
3076 a
->check_degraded
= 1;
3078 /* add some spares to the metadata */
3079 while (super
->add
) {
3083 super
->add
= al
->next
;
3084 al
->next
= super
->disks
;
3086 dprintf("%s: added %x:%x\n",
3087 __func__
, al
->major
, al
->minor
);
3094 static void imsm_prepare_update(struct supertype
*st
,
3095 struct metadata_update
*update
)
3098 * Allocate space to hold new disk entries, raid-device entries or a new
3099 * mpb if necessary. The manager synchronously waits for updates to
3100 * complete in the monitor, so new mpb buffers allocated here can be
3101 * integrated by the monitor thread without worrying about live pointers
3102 * in the manager thread.
3104 enum imsm_update_type type
= *(enum imsm_update_type
*) update
->buf
;
3105 struct intel_super
*super
= st
->sb
;
3106 struct imsm_super
*mpb
= super
->anchor
;
3111 case update_create_array
: {
3112 struct imsm_update_create_array
*u
= (void *) update
->buf
;
3114 len
= sizeof_imsm_dev(&u
->dev
, 1);
3115 update
->space
= malloc(len
);
3122 /* check if we need a larger metadata buffer */
3123 if (super
->next_buf
)
3124 buf_len
= super
->next_len
;
3126 buf_len
= super
->len
;
3128 if (__le32_to_cpu(mpb
->mpb_size
) + len
> buf_len
) {
3129 /* ok we need a larger buf than what is currently allocated
3130 * if this allocation fails process_update will notice that
3131 * ->next_len is set and ->next_buf is NULL
3133 buf_len
= ROUND_UP(__le32_to_cpu(mpb
->mpb_size
) + len
, 512);
3134 if (super
->next_buf
)
3135 free(super
->next_buf
);
3137 super
->next_len
= buf_len
;
3138 if (posix_memalign(&super
->next_buf
, buf_len
, 512) != 0)
3139 super
->next_buf
= NULL
;
3143 /* must be called while manager is quiesced */
3144 static void imsm_delete(struct intel_super
*super
, struct dl
**dlp
, int index
)
3146 struct imsm_super
*mpb
= super
->anchor
;
3148 struct imsm_dev
*dev
;
3149 struct imsm_map
*map
;
3150 int i
, j
, num_members
;
3153 dprintf("%s: deleting device[%d] from imsm_super\n",
3156 /* shift all indexes down one */
3157 for (iter
= super
->disks
; iter
; iter
= iter
->next
)
3158 if (iter
->index
> index
)
3161 for (i
= 0; i
< mpb
->num_raid_devs
; i
++) {
3162 dev
= get_imsm_dev(super
, i
);
3163 map
= get_imsm_map(dev
, 0);
3164 num_members
= map
->num_members
;
3165 for (j
= 0; j
< num_members
; j
++) {
3166 /* update ord entries being careful not to propagate
3167 * ord-flags to the first map
3169 ord
= get_imsm_ord_tbl_ent(dev
, j
);
3171 if (ord_to_idx(ord
) <= index
)
3174 map
= get_imsm_map(dev
, 0);
3175 set_imsm_ord_tbl_ent(map
, j
, ord_to_idx(ord
- 1));
3176 map
= get_imsm_map(dev
, 1);
3178 set_imsm_ord_tbl_ent(map
, j
, ord
- 1);
3183 super
->updates_pending
++;
3185 struct dl
*dl
= *dlp
;
3187 *dlp
= (*dlp
)->next
;
3188 __free_imsm_disk(dl
);
3191 #endif /* MDASSEMBLE */
3193 struct superswitch super_imsm
= {
3195 .examine_super
= examine_super_imsm
,
3196 .brief_examine_super
= brief_examine_super_imsm
,
3197 .detail_super
= detail_super_imsm
,
3198 .brief_detail_super
= brief_detail_super_imsm
,
3199 .write_init_super
= write_init_super_imsm
,
3200 .validate_geometry
= validate_geometry_imsm
,
3201 .add_to_super
= add_to_super_imsm
,
3203 .match_home
= match_home_imsm
,
3204 .uuid_from_super
= uuid_from_super_imsm
,
3205 .getinfo_super
= getinfo_super_imsm
,
3206 .update_super
= update_super_imsm
,
3208 .avail_size
= avail_size_imsm
,
3210 .compare_super
= compare_super_imsm
,
3212 .load_super
= load_super_imsm
,
3213 .init_super
= init_super_imsm
,
3214 .store_super
= store_zero_imsm
,
3215 .free_super
= free_super_imsm
,
3216 .match_metadata_desc
= match_metadata_desc_imsm
,
3217 .container_content
= container_content_imsm
,
3223 .open_new
= imsm_open_new
,
3224 .load_super
= load_super_imsm
,
3225 .set_array_state
= imsm_set_array_state
,
3226 .set_disk
= imsm_set_disk
,
3227 .sync_metadata
= imsm_sync_metadata
,
3228 .activate_spare
= imsm_activate_spare
,
3229 .process_update
= imsm_process_update
,
3230 .prepare_update
= imsm_prepare_update
,
3231 #endif /* MDASSEMBLE */